diff --git a/parse/train/3FK30d5BZdu/3FK30d5BZdu.md b/parse/train/3FK30d5BZdu/3FK30d5BZdu.md
new file mode 100644
index 0000000000000000000000000000000000000000..6b658c4eb551dcf3e04f9000e9f97e457e75c048
--- /dev/null
+++ b/parse/train/3FK30d5BZdu/3FK30d5BZdu.md
@@ -0,0 +1,445 @@
+# HIDDEN INCENTIVES FOR AUTO-INDUCED DISTRIBUTIONAL SHIFT
+
+Anonymous authors Paper under double-blind review
+
+# ABSTRACT
+
+Decisions made by machine learning systems have increasing influence on the world, yet it is common for machine learning algorithms to assume that no such influence exists. An example is the use of the i.i.d. assumption in content recommendation. In fact, the (choice of) content displayed can change users’ perceptions and preferences, or even drive them away, causing a shift in the distribution of users. We introduce the term auto-induced distributional shift (ADS) to describe the phenomenon of an algorithm causing a change in the distribution of its own inputs. Our goal is to ensure that machine learning systems do not leverage ADS to increase performance when doing so could be undesirable. We demonstrate that changes to the learning algorithm, such as the introduction of meta-learning, can cause hidden incentives for auto-induced distributional shift (HI-ADS) to be revealed. To address this issue, we introduce ‘unit tests’ and a mitigation strategy for HI-ADS, as well as a toy environment for modelling real-world issues with HIADS in content recommendation, where we demonstrate that strong meta-learners achieve gains in performance via ADS. We show meta-learning and Q-learning both sometimes fail unit tests, but pass when using our mitigation strategy.
+
+# 1 INTRODUCTION
+
+Consider a content recommendation system whose performance is measured by accuracy of predicting what users will click. This system can achieve better performance by either 1) making better predictions, or 2) changing the distribution of users such that predictions are easier to make. We propose the term auto-induced distributional shift (ADS) to describe this latter kind of distributional shift, caused by the algorithm’s own predictions or behaviour (Figure 1).
+
+![](images/833d9ac6619b77e4b4c37b9014af227c1cffc0cdcc8236556ae4c01b6b89cd90.jpg)  
+Figure 1: Distributions of users over time. Left: A distribution which remains constant over time, following the i.i.d assumption. Right: Auto-induced Distributional Shift (ADS) results in a change in the distribution of users in our content recommendation environment. (see Section 5.2 for details).
+
+ADS are not inherently bad, and are sometimes even desirable. But they can cause problems if they occur unexpectedly. It is typical in machine learning (ML) to assume (e.g. via the i.i.d. assumption) that (2) will not happen. However, given the increasing real-world use of ML algorithms, we believe it is important to model and experimentally observe what happens when assumptions like this are violated. This is the motivation of our work.
+
+In many cases, including news recommendation, we would consider (2) a form of cheating—the algorithm changed the task rather than solving it as intended. We care which means the algorithm used to solve the problem (e.g. (1) and/or (2)), but we only told it about the ends, so it didn’t know not to ’cheat’. This is an example of a specification problem (Leike et al., 2017; Ortega et al., 2018): a problem which arises from a discrepancy between the performance metric (maximize accuracy) and “what we really meant”: in this case, to maximize accuracy via (1).
+
+Ideally, we’d like to quantify the desirability of all possible means, e.g. assign appropriate rewards to all potential strategies and “side-effects”, but this is intractable for real-world settings. Using human feedback to learn reward functions which account for such impacts is a promising approach to specifying desired behavior (Leike et al., 2018; Christiano et al., 2017). But the same issue can arise whenever human feedback is used in training: one means of improving performance could be to alter human preferences, making them easier to satisfy. Thus in this work, we pursue a complementary approach: managing learners’ incentives.
+
+A learner has an incentive to behave in a certain way when doing so can increase performance (e.g. accuracy or reward). Informally, we say an incentive is hidden when the learner behaves as if it were not present. But we note that changes to the learning algorithm or training regime could cause previously hidden incentives to be revealed, resulting in unexpected and potentially undesirable behaviour. Managing incentives (e.g. controlling which incentives are hidden/ revealed) can allow algorithm designers to disincentivize broad classes of strategies (such as any that rely on manipulating human preferences) without knowing their exact instantiation.1
+
+The goal of our work is to provide insights and practical tools for understanding and managing incentives, specifically hidden incentives for auto-induced distributional shift: HI-ADS. To study which conditions cause HI-ADS to be revealed, we present unit tests for detecting HI-ADS in supervised learning (SL) and reinforcement learning (RL). We also create an environment that models ADS in news recommendation, illustrating possible effects of revealing HI-ADS in this setting.
+
+The unit tests both have two means by which the learner can improve performance: one which creates ADS and one which does not. The intended method of improving performance is one that does not induce ADS; the other is ’hidden’ and we want it to remain hidden. A learner "fails" the unit test if it nonetheless pursues the incentive to increase performance via ADS. In both the RL and SL unit tests, we find that introducing an outer-loop of meta-learning (e.g. Population-Based Training (PBT) Jaderberg et al. (2017)) can lead to high levels of failure. Similarly, recommender systems trained with PBT induce larger drifts in user base and user interests. These results suggest that failure of our unit tests indicates that an algorithm is prone to revealing HI-ADS in other settings. Finally, we propose and test a mitigation strategy we call context swapping. The strategy consists of rotating learners through different environments throughout learning, so that they can’t see the results or correlations of their actions in one environment over longer time horizons. This effectively mitigates HI-ADS in our unit test environments, but did not work well in content recommendation experiments.
+
+# 2 BACKGROUND
+
+# 2.1 META-LEARNING AND POPULATION BASED TRAINING
+
+Meta-learning is the use of machine learning techniques to learn machine learning algorithms. This involves instantiating multiple learning scenarios which run in an inner loop $\mathbf { \Pi } ( \mathbf { I I L } )$ , while an outer loop (OL) uses the outcomes of the inner loop(s) as data-points from which to learn which learning algorithms are most effective (Metz et al., 2019). The number of IL steps per OL step is called the interval. Many recent works focus on multi-task meta-learning, where the OL seeks to find learning rules that generalize to unseen tasks by training the IL on a distribution of tasks (Finn et al., 2017). Single-task meta-learning includes learning an optimizer for a single task (Gong et al., 2018), and adaptive methods for selecting models (Kalousis, 2000) or setting hyperparameters (Snoek et al., 2012). For simplicity in this initial study we focus on single-task meta-learning.
+
+Population-based training (PBT; Jaderberg et al., 2017) is a meta-learning algorithm that trains multiple learners $L _ { 1 } , . . . , L _ { n }$ in parallel, after each interval ( $T$ steps of IL) applying an evolutionary OL step which consists of: (1) Evaluate the performance of each learner, (2) Replace both parameters and hyperparameters of $20 \%$ lowest-performing learners with copies of those from the $20 \%$ highperforming learners (EXPLOIT). (3) Randomly perturb the hyperparameters (but not the parameters) of all learners (EXPLORE).
+
+# 2.2 DISTRIBUTIONAL SHIFT AND CONTENT RECOMMENDATION
+
+In general, distributional shift refers to change of the data distribution over time. In supervised learning with data $\mathbf { x }$ and labels $y$ , this can be more specifically described as dataset shift: change in the joint distribution of $P ( \mathbf { x } , y )$ between the training and test sets (Moreno-Torres et al., 2012; Quionero-Candela et al., 2009). As identified by Moreno-Torres et al. (2012), two common kinds of shift are: (1) Covariate shift: changing $P ( \mathbf { x } )$ . In the example of content recommendation, this corresponds to changing the user base of the recommendation system. For instance, a media outlet which publishes inflammatory content may appeal to users with extreme views while alienating more moderate users. This self-selection effect (Kayhan, 2015) may appear to a recommendation system as an increase in performance, leading to a feedback effect, as previously noted by Shah et al. (2018). This type of feedback effect has been identified as contributing to filter bubbles and radicalization (Pariser, 2011; Kayhan, 2015). (2) Concept shift: changing $P ( \boldsymbol { y } | \mathbf { x } )$ . In the example of content recommendation, this corresponds to changing a given user’s interest in different kinds of content. For example, exposure to a fake news story has been shown to increase the perceived accuracy of (and thus presumably future interest in) the content, an example of the illusory truth effect (Pennycook et al., 2019). For further details on such effects in content recommendation, see Appendix 8.
+
+# 3 AUTO-INDUCED DISTRIBUTION SHIFT (ADS)
+
+Auto-induced distribution shift (ADS) is distributional shift caused by an algorithm’s behaviour. This is in contrast to distributional shift which would happen even if the learner were not present - e.g. for a crash prediction algorithm trained on data from the summer, encountering snowy roads is an example of distributional shift, but not auto-induced distributional shift (ADS).
+
+![](images/e073c2dfa79ed13d098885e0aecd106a567f60dc458af17cb1a90e8412bd2143.jpg)
+
+Figure 2: The widely studied problems of reinforcement learning (RL) with state $s$ , action $a$ , reward $r$ tuples, and i.i.d. supervised learning (SL) with inputs $x$ , predictions $\hat { y }$ and loss $l$ (a,d) are free from incentive problems. We focus on cases where there are incentives present which the learner is not meant to pursue $^ { ( \mathbf { b } , \mathbf { c } ) }$ . Lines show paths of influence. The learner may have incentives to influence any nodes descending from its action, $A$ , or prediction, $\hat { y }$ . Which incentives are undesirable (orange) or desirable (cyan) for the learner to pursue is context-dependent.
+
+We emphasize that ADS are not inherently bad or good; often ADS can even be desirable: consider an algorithm meant to alert drivers of imminent collisions. If it works well, such a system will help drivers avoid crashing, thus making self-refuting predictions which result in ADS. What separates desirable and undesirable ADS? The collision-alert system alters its data distribution in a way that is aligned with the goal of fewer collisions, whereas the news manipulation results in changes that are misaligned with the goal of better predicting existing users’ interests (Leike et al., 2018).
+
+In reinforcement learning (RL), ADS are typically encouraged as a means to increase performance. On the other hand, in supervised learning (SL), the i.i.d. assumption precludes ADS in theory. In practice, however, the possibility of using ADS to increase performance (and thus an incentive to do so) often remains. For instance, this occurs in online learning. In our experiments, we explicitly model such situations where i.i.d. assumptions are violated: We study the behavior of SL and myopic RL algorithms, in environments designed to include incentives for ADS, in order to understand when incentives are effectively hidden. Fig. 2 contrasts these settings with typical RL and SL.
+
+# 4 INCENTIVES
+
+For our study of incentives, we use the following terminology: an incentive for a behavior (e.g. an action, a classification, etc.) is present (not absent) to the extent that the behaviour will increase performance (e.g. reward, accuracy, etc.) (Everitt & Hutter, 2019). This incentive is revealed to (not hidden from) a learner if it would, at higher than chance levels, learn to perform the behavior given sufficient capacity and training experience. The incentive is pursued (not eschewed) by a learner if it actually performs the incentivized behaviour. Note even when an incentive is revealed, it may not be pursued, e.g. due to limited capacity and/or data, or simply chance. See Fig 3.
+
+For example, in content recommendation, the incentive to drive users away is present if some user types are easier to predict than others. But this incentive may be hidden from the learner by using a myopic algorithm, e.g. one that does not see the effects of its actions on the distribution of users. The incentive might instead be revealed to the outer loop of a meta-learning algorithm like PBT, which does see the effects of learner’s actions.
+
+![](images/2ade91d4be27c22a6f418d3a986aa9448815d5a0ddb40aa939127276f08bc9bc.jpg)  
+Figure 3: Types of incentives, and their relationship to ADS.
+
+Even when this incentive is revealed, however, it might not end up being pursued. For example, this could happen if predicting which recommendations will drive away users is too difficult a learning problem, or if the incentive to do so is dominated by other incentives (e.g. change individual users’ interests, or improve accuracy of predictions). In general, it may be difficult to determine empirically which incentives are revealed, because failure to pursue an incentive can be due to limited capacity, insufficient training, and/or random chance. To address this challenge, we devise extremely simple environments (“unit tests”), where we can be confident that revealed incentives will be pursued.
+
+# 4.1 HIDDEN INCENTIVES FOR AUTO-INDUCED DISTRIBUTIONAL SHIFT (HI-ADS)
+
+Following from the definitions in Sections 3 and 4, HI-ADS are incentives for behaviors that cause Auto-induced Distributional Shift that are hidden from the learner, i.e. the learner would not learn to perform the incentivized behaviors at higher than chance levels, even given infinite capacity and training experience.
+
+Like ADS, HI-ADS are not necessarily problematic. Indeed, hiding incentives can be an effective method of influencing learner behavior. For example, hiding the incentive to manipulate users from a content recommendation algorithm could prevent it from influencing users in a way they would not endorse. However, if machine learning practitioners are not aware that incentives are present, or that properties of the learning algorithm are hiding them, then seemingly innocuous changes to the learning algorithm may reveal HI-ADS, and lead to significant unexpected changes in behavior.
+
+Hiding incentives for ADS may seem counter-intuitive and counter-productive in the context of reinforcement learning (RL), where moving towards high-reward states is typically desirable. However, for real-world applications of RL, the ultimate goal is not a system that achieves high reward, but rather one that behaves according to the designer’s intentions. And as we discussed in the introduction, it can be intractable to design reward functions that perfectly specify intended behavior. Thus managing (e.g. hiding) some incentives can provide a useful tool for specification, even in RL.
+
+We have several reasons for focusing on HI-ADS: (1) The issue of HI-ADS has not yet been identified, and thus is liable to be neglected in practice. Indeed, our “unit tests” are the first published empirical methodology for assessing whether incentives are hidden or revealed by different learning algorithms. (2) Machine learning algorithms are commonly deployed in settings where ADS are present, violating assumptions used to analyze their properties theoretically. This means learners could exploit ADS in unexpected and undesirable ways if incentives for ADS are revealed. Hiding these incentives heuristically (e.g. via off-line training) is a common approach, but potentially brittle (if practitioners don’t understand how HI-ADS could become revealed). In particular, meta-learning can reveal HI-ADS in online learning settings. (3) Substantial real-world issues could result from improper management of learner’s incentives. Examples include tampering with human-generated reward signals (Everitt & Hutter, 2018) (e.g. selecting news articles which manipulate user interests), and creating “self-fulfilling prophecies” (e.g. driving up the value of an asset by publicly predicting its value will increase (Armstrong & O’Rorke, 2017)).
+
+# 4.2 REMOVING HI-ADS VIA CONTEXT SWAPPING
+
+We propose a technique called context swapping for removing incentives for ADS. The technique trains $N$ learners in parallel, and shuffles the learners through $N$ different copies of the same (or similar) environments.We use a deterministic permutation of learners in environment copies, so that the $i$ -th learner inhabits the $j$ -th environment on time-steps $t$ where $~ j ~ = ~ ( i + t )$ mod $N$ , makes an observation, takes an action, and receives a reward before moving to the next environment.
+
+When $N$ is larger than the interval of the OL optimizer, each learner inhabits each copy for at most a single time-step before an OL step is applied. Under the assumption that different copies of the environment do not influence each other, this technique can address HI-ADS in practice, as we show in Sec. 5.1.1.
+
+![](images/06fd14fc3c3117d01ce92cc59924cf8d1fe7f705cf8d340291c9e7534c747cd1.jpg)  
+Figure 4: Context swapping (right).
+
+# 5 EXPERIMENTS
+
+In Section 5.1, we introduce ‘unit tests’ for HI-ADS. Our primary goal with these experiments is to convey a crisp understanding of potential issues caused by revealing HI-ADS. Put simply, our experiments show that you can have a learner which behaves as intended, and just by using metalearning (e.g. PBT), without changing the performance metric (e.g. loss or rewards), the learner’s behavior can change completely. We also show that context swapping is an effective mitigation technique in these environments. On the practical side, the unit tests can be used to compare learning algorithms and diagnose their propensity to reveal incentives.
+
+In Section 5.2, we model a content recommendation system. The goal of these experiments is to demonstrate how HI-ADS could create issues for real-world content recommendation systems such as news feeds, search results, or automated suggestions. They also validate the usefulness of the unit tests: algorithms that failed the unit tests also reveal HI-ADS in this setting. We emphasize that ADS takes place in this environment by construction. The point of our experiments is that meta-learning can increase the rate and/or extent of ADS, by revealing this incentive. We find that context swapping is not effective in this environment, highlighting the need for alternative mitigation strategies.
+
+# 5.1 HI-ADS UNIT TESTS
+
+Unit test 1: Supervised Learning. This unit test consists of a simple prediction problem. There are no inputs, only an underlying state $s \in \{ 0 , 1 \}$ , and targets $\boldsymbol { y } \in \mathbb { R } ^ { 2 }$ with $\chi _ { 1 } , y _ { 2 } \sim \mathcal { N } ( 0 , s * \sigma ^ { 2 } ) , \mathcal { N } ( 0 , 1 )$ , with corresponding predictions $\hat { y } _ { 1 } , \hat { y } _ { 2 }$ . Additionally, $s _ { t + 1 } = 0$ iff $\hat { y } _ { 2 } >$ .5. We use Mean Squared Error as the loss function, so the optimal predictor is $\hat { y _ { 1 } } , \hat { y _ { 2 } } = ( 0 , 0 )$ . However, predicting $\hat { y } _ { 2 } > . 5$ reduces the variance of $\hat { y _ { 1 } }$ , i.e. reduces future loss. The baseline/IL predictor learns $\hat { y _ { 1 } } , \hat { y _ { 2 } }$ as parameters using SGD with a learning rate of 0.001. For experiments with meta-learning, PBT is the OL (with default settings, see Section 2.2), used to tune the learning rate, with negative loss on the final time-step of the interval as the performance measure for PBT.
+
+Unit test 2: Myopic RL. This unit test is based on a modification of the prisoner’s dilemma (Prisner, 2014) where an agent plays each round against its past self. The reward function is presented in Table 1. An agent in this environment has a long-term, non-myopic, incentive for cooperation (with its future self), but a current-time-step, myopic, incentive for defection (from its future self).
+
+The unit test evaluates whether a agent reveals the non-myopic incentive even when the agent is meant to optimize for the present reward only (i.e. uses discount rate $\gamma = 0$ ). Naively, we’d expect the nonmyopic incentive to be hidden from the agent in this case, and for the agent to consistently defect; learning algorithms that do so pass the test. But some learning algorithms also fail the unit test, revealing the incentive for the agent to cooperate with its future self. While aiming for myopic behavior
+
+Table 1: Rewards for the RL unit test. Note that the myopic (defect (D)) action always increases reward at the current time-step, but decreases reward at the next time-step - the incentive to (cooperate (C)) with one’s future self is hidden from the point of view of a myopic learner.
+
+<table><tr><td rowspan=1 colspan=1></td><td rowspan=1 colspan=1>at=D</td><td rowspan=1 colspan=1>at=C</td></tr><tr><td rowspan=1 colspan=1>St=at-1=D</td><td rowspan=1 colspan=1>-1/2</td><td rowspan=1 colspan=1>-1</td></tr><tr><td rowspan=1 colspan=1>St=at-1=C</td><td rowspan=1 colspan=1>1/2</td><td rowspan=1 colspan=1>0</td></tr></table>
+
+may seem odd, myopic learners have no incentives to cause distributional shift, since it can only improve future performance. And while making learners myopic may seem like a ’brute-force’ guaranteed way to manage HI-ADS, we show it is in fact non-trivial to implement. See Appendix 9.1 for details and experiments varying the reward structure.
+
+# 5.1.1 HI-ADS UNIT TESTS EXPERIMENTAL RESULTS AND DISCUSSION
+
+We first show that agents trained with PBT fail the unit tests more often than “vanilla” algorithms which do not use meta-learning. We initialize the learning rate log-uniformly between 0.01 and 1.0 for all experiments (whether using PBT or not). We expect and confirm that the following two factors lead to higher rates of unit test failure: (1) Shorter intervals: These give the OL more opportunities to influence the population. (2) Larger populations: These make outliers with exceptional non-myopic performance more likely, and OL makes them likely to survive and propagate.
+
+The baseline (no meta-learning) algorithms all pass the unit tests: hidden incentives are almost never revealed - see blue curves in Fig. 5. However, agents trained with meta-learning and large populations often fail the unit tests: see orange curves in top rows of Fig. 5.
+
+Furthermore, we verify that context swapping significantly mitigates the effect of HI-ADS in both unit tests, decreasing undesirable behaviour to near-baseline levels - see bottom rows of Fig. 5. This effect can be explained as follows: Because context swapping transfers the benefits of one learner’s action to the next learner to inhabit that environment, it increases the second learner’s fitness, and thereby reduces the relative fitness (as evaluated by PBT’s EXPLOIT step) of the non-myopic cooperate behaviour. We observe some interesting exceptions with the combination of small populations and short PBT intervals: Although context swapping still significantly decreases the effect of HIADS, non-myopic cooperate behaviour is observed as much as $20 \%$ of the time (for #learner $_ { = 1 0 }$ , $T = 1$ ; see bottom-left plot).
+
+We also observe that PBT reveals HI-ADS even when $T = 1$ , where the explanation that PBT operates on a longer time horizon than the inner loop does not apply. We provide a detailed explanation for how this might happen in Appendix 9.1.2, but in summary, we hypothesize that there are at least 2 mechanisms by which PBT is revealing HI-ADS: (1) optimizing over a longer time-scale, and (2) picking up on the correlation between an agent’s current policy and the underlying state. Mechanism (2) can be explained informally as reasoning as: “If I’m cooperating, then I was probably cooperating on the last time-step as well, so my reward should be higher”. As support for these hypotheses, we run control experiments identifying two algorithms (each sharing only one of these properties) that can fail the unit test. Context swapping remains effective.
+
+# (1) Optimizing over a longer time-scale: replacing PBT with REINFORCE as an
+
+![](images/fb963b752d78b641eafae0011c4187a77f39ee589e94374f2b91ae0ecc03c850.jpg)  
+Figure 5: (A) Values of $\hat { y _ { 2 } }$ in the supervised learning (SL) unit test. Larger values mean sacrificing present performance for future performance (i.e. non-myopic exploitation of ADS). (B) Average level of non-myopic cooperate behavior observed in the RL unit test for HI-ADS, with two meta-learning algorithms (B1) PBT and (B2) REINFORCE. Lower is better, since the goal is for non-myopic incentives to remain hidden. Despite the inner loop being fully myopic (simple MLP in the SL test, $\gamma = 0$ in RL test), in all cases outer-loop (OL) optimizers reveal HI-ADS (top rows). Context swapping significantly mitigates HI-ADS (bottom rows).
+
+outer-loop optimizer. The outer-loop optimizes the parameters to maximize the summed reward of the last $T$ time-steps. As with PBT, we observe non-myopic behavior, but now only when $T > 1$ This supports our hypothesis that exploitation of HI-ADS is due not to PBT in particular, but just to the introduction of sufficiently powerful meta-learning. See Fig. 5 B2.
+
+(2) Exploiting correlation: Q-learning with $\gamma = 0$ an $\epsilon = 0 . 1$ -greedy behavior policy and no meta-learning. If either state was equally likely, the $\mathbf { Q }$ -values would be the average of the values in each column in Table 1, so the estimated $Q ( { \mathrm { d e f e c t } } )$ would be larger. But the $\epsilon$ -greedy policy correlates the previous action (i.e. the current state) and current action (so long as the policy did not just change), so the top-left and bottomright entries carry more weight in the estimates, sometimes causing $Q ( { \mathrm { d e f e c t } } ) \approx Q ($ (cooperate) and persistent nonmyopic behavior. See Fig. 6 for results, Appendix 9.1.4 for more results, and Appendix 9.1.3 for experimental details.
+
+# 5.2 CONTENT RECOMMENDATION
+
+![](images/c27ce5d1176c26223b5a51441be7d7fafa717c25c484532ba211b635ec5f8ba5.jpg)  
+Figure 6: Q-learning can fail the unit test, playing ${ \sim } 8 0 { \cdot } 9 0 \%$ cooperate in 3 of 5 experiments (bottom row). Each column represents an independent experiment. Q-values for the cooperate and defect actions stay tightly coupled in the failure cases (col. 1,2,5), while in the cases passing the unit test (col. 3,4) the Q-value of cooperate decreases over time.
+
+We now present a toy environment for modeling content recommendation of news articles, which includes the potential for ADS by incorporating the mechanisms mentioned in Sec. 2.2, discussed as contributing factors to the problems of fake news and filter bubbles. Specifically, the environment assumes that presenting an article to a user can influence (1) their interest in similar articles, and (2) their propensity to use the recommendation service. These correspond to modeling auto-induced concept shift of users, and auto-induced covariate shift of the user base, respectively (see Sec. 2.2).
+
+This environment includes the following components, which change over (discrete) time: User type: $x ^ { t }$ , Article type: $y ^ { t }$ , User interests: $\bar { \mathbf { W } } ^ { t }$ (propensity for users of each type to click on articles of each type), and User loyalty: $\mathbf { g } ^ { t }$ (propensity for users of each type to use the platform). At each time step $t$ , a user $x ^ { t }$ is sampled from a categorical distribution, based on the loyalty of the different user types. The recommendation system (a classifier) selects which type of article to present in the top position, and finally the user ‘clicks’ an article $y ^ { t }$ , according to their interests. User loyalty for user type $x ^ { t }$ undergoes covariate shift: in accordance with the self-selection effect, $g ^ { t }$ increases or decreases proportionally to that user type’s interest in the top article. The interests of user type $x ^ { t }$ (represented by a column of $\mathbf { W } ^ { t }$ ) undergoing concept shift; in accordance with the illusory truth effect, interest in the topic of the top article chosen by the recommender system always increases.
+
+# 5.2.1 CONTENT RECOMMENDATION EXPERIMENTAL RESULTS AND DISCUSSION
+
+We run 20 trials using an MLP trained with SGD for the recommender system. We find that PBT yields significant improvements in training time and accuracy, but also greater distributional shift (Fig. 7). User base and user interests both change faster with PBT, and user interests change more overall. We measure concept/covariate shift using the cosine distance and KL-divergence, respectively. We observe that the distributions over user types typically saturate (to a single user type) after a few hundred time-steps (Fig 1 and Fig. 7, Right). We run long enough to reach such states, to demonstrate that the increase in ADS from PBT is not transitory. The environment has a number of free parameters, and our results are qualitatively consistent so long as the covariate shift rate $( \alpha _ { 1 } )$ is faster than the concept shift rate $\left( \alpha _ { 2 } \right)$ . See Appendix 9.2.1 for details.
+
+![](images/95efffd0c2948471a88671bd2e264c0a9ccf142775dc7e578a618fd609eb6f42.jpg)  
+Figure 7: Content recommendation experiments. Left: using Population Based Training (PBT) increases accuracy of predictions faster, leads to a faster and larger drift in users’ interests, $P ( \boldsymbol { y } | \mathbf { x } )$ , (Center); as well as the distribution of users, $P ( \mathbf { x } )$ , (Right). Shading shows std error over 20 runs.
+
+# 6 RELATED WORK
+
+ADS in practice: We introduce the term ADS, but we are far from the first to study it. Caruana et al. (2015) provide an example of asthmatic patients having lower predicted risk of pneumonia. Treating asthmatics with pneumonia less aggressively on this basis would be an example of harmful ADS; the reason they had lower pneumonia risk was because they had received more aggressive care already. Schulam & Saria (2017) note that such predictive models are commonly used to inform decision-making, and propose modeling counterfactuals (e.g. “how would this patient fare with less aggressive treatment”) to avoid making such self-refuting predictions. While their goal is to make accurate predictions in the presence of ADS, our goal is to identify and manage incentives for ADS.
+
+Non-i.i.d bandits: Contextual bandits (Wang et al., 2005) are a common approach to content recommendation (Li et al., 2010). While bandit algorithms typically make the i.i.d. assumption, counter-examples exist (Gheshlaghi Azar et al., 2014; Auer et al., 1995). Closest to our work is Shah et al. (2018), who consider covariate shift caused by recommender systems’ recommendations. But while they seek to exploit ADS, our aim is to avoid hidden incentives for exploiting ADS.
+
+Safety and incentives: Emergent incentives to influence the world (such as HI-ADS) are at the heart of many concerns about the safety of advanced AI systems (Omohundro, 2008; Bostrom, 2014). Understanding and managing the incentives of learners is also a focus of Armstrong & O’Rourke (2017); Everitt (2018); Everitt et al. (2019); Cohen et al.. While Everitt et al. (2019) focus on identifying which incentives are present, we note that incentives may be present and yet not be revealed or pursued - for example, in supervised learning, there is an incentive to make predictions that are over-fit to the test set, but we typically hide the test set from the learner, which effectively hides this incentive. While Carey et al. (2020); Everitt et al. (2019); Armstrong & O’Rourke (2017) discuss methods of removing problematic incentives, we note in practice incentives are often hidden rather than removed. Our work addresses the efficacy of this approach and ways in which it can fail.
+
+HI-ADS and meta-learning: We believe our work is the first to consider HI-ADS and their relation to meta-learning. A few previous works have some relevance. Rabinowitz (2019) documents qualitative differences in learning behavior when meta-learning is applied. MacKay et al. (2019) and Lorraine & Duvenaud (2018) view meta-learning as a bilevel optimization problem, with the inner loop playing a best-response to the outer loop. In our work, the outer loop has a greater influence, and the inner loop often fails to play best-response. Sutton et al. (2007) noted that meta-learning can improve performance by preventing convergence of the inner loop to best response.
+
+# 7 DISCUSSION AND CONCLUSION
+
+We identify the phenomenon of auto-induced distributional shift (ADS) and problems that can arise when there are hidden incentives for learners to induce distributional shift (HI-ADS). We show that meta-learning can reveal HI-ADS and lead learners to use ADS as a means of increasing performance.
+
+Our work highlights the interdisciplinary nature of issues with real-world deployment of ML systems - we show how HI-ADS could play a role in important technosocial issues like filter bubbles and the propagation of fake news. There are a number of potential implications for our work: (1) When HI-ADS are a concern, our methodology and environments can be used to help diagnose whether and to what extent the final performance/behavior of a learner is due to ADS and/or incentives for ADS, i.e. to quantify their influence on that learner. (2) Comparing this quantitative analysis for different algorithms could help us understand which features of algorithms affect their propensity to reveal HI-ADS, and aid in the development of safer and more robust algorithms. (3) Characterizing and identifying HI-ADS in these tests is a first step to analyzing and mitigating other (problematic) hidden incentives, as well as to developing theoretical understanding of hidden incentives.
+
+Broadly speaking, our work emphasizes that the choice of machine learning algorithm plays an important role in specification, independently of the choice of performance metric. A learner can use ADS to increase performance according to the intended performance metric, and yet still behave in an undesirable way, if we did not intend the learner to improve performance by that method. In other words, performance metrics are typically incomplete specifications: they only specify our goals or ends, while our choice of learning algorithm plays a role in specifying the means by which we intend an learner to achieve those ends. With increasing deployment of ML algorithms in daily life, we believe that (1) understanding incentives and (2) specifying desired/allowed means of improving performance are important avenues of future work to ensure fair, robust, and safe outcomes.
+
+# REFERENCES
+
+Hunt Allcott and Matthew Gentzkow. Social media and fake news in the 2016 election. Journal of Economic Perspectives, 31(2):211–36, May 2017.
+
+Michelle A. Amazeen and Bartosz W. Wojdynski. Reducing native advertising deception: Revisiting the antecedents and consequences of persuasion knowledge in digital news contexts. Mass Communication and Society, 0(0):1–26, 2018.
+
+Stuart Armstrong and Xavier O’Rorke. Good and safe uses of ai oracles. arXiv preprint arXiv:1711.05541, 2017.
+
+Stuart Armstrong and Xavier O’Rourke. Indifference methods for managing agent rewards. Technical report, Future of Humanity Institute, 2017.
+
+K. J. Åström. Optimal control of Markov Processes with incomplete state information. Journal of Mathematical Analysis and Applications, 10:174–205, January 1965.
+
+P. Auer, N. Cesa-Bianchi, Y. Freund, and R. E. Schapire. Gambling in a rigged casino: The adversarial multi-armed bandit problem. In Foundations of Computer Science, 1995.
+
+Eytan Bakshy, Solomon Messing, and Lada A. Adamic. Exposure to ideologically diverse news and opinion on Facebook. Science, 348(6239):1130–1132, 2015. ISSN 0036-8075. doi: 10.1126/science.aaa1160. URL http://science.sciencemag.org/content/348/ 6239/1130.
+
+Nick Bostrom. Superintelligence: Paths, Dangers, Strategies. Oxford University Press, Inc., New York, NY, USA, 1st edition, 2014.
+
+Ryan Carey, Eric Langlois, Tom Everitt, and Shane Legg. The incentives that shape behaviour. arXiv preprint arXiv:2001.07118, 2020.
+
+Rich Caruana, Yin Lou, Johannes Gehrke, Paul Koch, Marc Sturm, and Noemie Elhadad. Intelligible models for healthcare: Predicting pneumonia risk and hospital 30-day readmission. In International Conference on Knowledge Discovery and Data Mining, pp. 1721–1730, 2015.
+
+Paul Christiano, Jan Leike, Tom B. Brown, Miljan Martic, Shane Legg, and Dario Amodei. Deep reinforcement learning from human preferences, 2017.
+
+Michael K Cohen, Badri N Vellambi, and Marcus Hutter. Asymptotically unambitious artificial general intelligence.
+
+Dominic DiFranzo and Kristine Gloria-Garcia. Filter bubbles and fake news. XRDS, 23(3):32–35, April 2017. ISSN 1528-4972. doi: 10.1145/3055153. URL http://doi.acm.org/10. 1145/3055153.
+
+Mostafa M. El-Bermawy. Your echo chamber is destroying democracy, 2016. URL https: //www.wired.com/2016/11/filter-bubble-destroying-democracy/.
+
+Tom Everitt. Towards Safe Artificial General Intelligence. PhD thesis, Australian National University, 2018.
+
+Tom Everitt and Marcus Hutter. The alignment problem for bayesian history-based reinforcement learners. 2018.
+
+Tom Everitt and Marcus Hutter. Reward tampering problems and solutions in reinforcement learning: A causal influence diagram perspective. arXiv preprint arXiv:1908.04734, 2019.
+
+Tom Everitt, Pedro A. Ortega, Elizabeth Barnes, and Shane Legg. Understanding agent incentives using causal influence diagrams. part i: Single action settings, 2019.
+
+Lisa K. Fazio, Nadia M. Brashier, B. Keith Payne, and Elizabeth J. Marsh. Knowledge does not protect against illusory truth. Journal of Experimental Psychology: General, 144(5):993–1002, 10 2015. ISSN 0096-3445. doi: 10.1037/xge0000098.
+
+Chelsea Finn, Pieter Abbeel, and Sergey Levine. Model-agnostic meta-learning for fast adaptation of deep networks. In International Conference on Machine Learning, 2017.
+
+Seth Flaxman and Sharad Goel. Filter bubbles, echo chambers, and online news consumption. Public Opinion Quarterly, 2015.
+
+Mohammad Gheshlaghi Azar, Alessandro Lazaric, and Emma Brunskill. Online stochastic optimization under correlated bandit feedback. ArXiv preprint, 2014.
+
+Dong Gong, Zhen Zhang, Qinfeng Shi, Anton van den Hengel, Chunhua Shen, and Yanning Zhang. Learning an optimizer for image deconvolution. ArXiv preprint, 2018.
+
+Jacob Groshek and Karolina Koc-Michalska. Helping populism win? Social media use, filter bubbles, and support for populist presidential candidates in the 2016 us election campaign. Information, Communication & Society, 20(9):1389–1407, 2017. doi: 10.1080/1369118X.2017.1329334. URL https://doi.org/10.1080/1369118X.2017.1329334.
+
+Dylan Hadfield-Menell, Smitha Milli, Pieter Abbeel, Stuart Russell, and Anca Dragan. Inverse reward design. In Neural Information Processing Systems, 2017.
+
+M. Jaderberg, V. Dalibard, S. Osindero, W. M. Czarnecki, J. Donahue, A. Razavi, O. Vinyals, T. Green, I. Dunning, K. Simonyan, C. Fernando, and K. Kavukcuoglu. Population Based Training of Neural Networks. ArXiv preprint, 2017.
+
+Leslie Pack Kaelbling, Michael L. Littman, and Anthony R. Cassandra. Planning and acting in partially observable stochastic domains. Artificial Intelligence, 101(1-2):99–134, May 1998. ISSN 0004-3702.
+
+A. Kalousis. Model selection via meta-learning: A comparative study. In IEEE International Conference on Tools with Artificial Intelligence, 2000.
+
+Varol Kayhan. Confirmation bias: Roles of search engines and search contexts. In International Conference on Information Systems, 2015.
+
+Ed. Lee Howell. Digital wildfires in a hyperconnected world. In Lee Howell (ed.), Global Risks 2013. World Economic Forum, 2013. URL http://reports.weforum.org/global-risks-2013/risk-case-1/ digital-wildfires-in-a-hyperconnected-world/.
+
+Jan Leike, Miljan Martic, Victoria Krakovna, Pedro A. Ortega, Tom Everitt, Andrew Lefrancq, Laurent Orseau, and Shane Legg. AI safety gridworlds. Technical report, DeepMind, 2017.
+
+Jan Leike, David Krueger, Tom Everitt, Miljan Martic, Vishal Maini, and Shane Legg. Scalable agent alignment via reward modeling: a research direction. Technical report, DeepMind Safety Research, 2018.
+
+Lihong Li, Wei Chu, John Langford, and Robert E. Schapire. A contextual-bandit approach to personalized news article recommendation. In International Conference on World Wide Web, 2010.
+
+Jonathan Lorraine and David Duvenaud. Stochastic hyperparameter optimization through hypernetworks. ArXiv preprint, 2018.
+
+D. D. Luxton, J. D. June, and J. M. Fairall. Social media and suicide: A public health perspective. American journal of public health, 102(2):195–200, 2012.
+
+Matthew MacKay, Paul Vicol, Jonathan Lorraine, David Duvenaud, and Roger Grosse. Self-tuning networks: Bilevel optimization of hyperparameters using structured best-response functions. ArXiv preprint, 2019.
+
+Merriam-Webster. The real story of fake news, 2017. URL https://www.merriam-webster. com/words-at-play/the-real-story-of-fake-news.
+
+Luke Metz, Niru Maheswaranathan, Brian Cheung, and Jascha Sohl-Dickstein. Learning unsupervised learning rules. In International Conference on Learning Representations, 2019.
+
+Paul Mihailidis and Samantha Viotty. Spreadable spectacle in digital culture: Civic expression, fake news, and the role of media literacies in "post-fact" society. American Behavioural Scientist, 2017.
+
+Jose G. Moreno-Torres, Troy Raeder, RocíO Alaiz-RodríGuez, Nitesh V. Chawla, and Francisco Herrera. A unifying view on dataset shift in classification. Pattern Recognition, 45(1):521–530, January 2012.
+
+Tien T. Nguyen, Pik-Mai Hui, F. Maxwell Harper, Loren Terveen, and Joseph A. Konstan. Exploring the filter bubble: The effect of using recommender systems on content diversity. In Proceedings of the 23rd International Conference on World Wide Web, WWW ’14, pp. 677–686, New York, NY, USA, 2014. ACM. ISBN 978-1-4503-2744-2. doi: 10.1145/2566486.2568012. URL http://doi.acm.org/10.1145/2566486.2568012.
+
+Safiya Umoja Noble. Algorithms of Oppression: How Search Engines Reinforce Racism. NYC Press, 2018.
+
+Stephen M. Omohundro. The basic AI drives. In Conference on Artificial General Intelligence, 2008.
+
+Pedro A. Ortega, Vishal Maini, et al. Building safe artificial intelligence: specification, robustness, and assurance, 2018.
+
+Eli Pariser. The Filter Bubble: What the Internet Is Hiding from You. The Penguin Group, 2011.
+
+Gordon Pennycook, Tyrone D Cannon, and David G. Rand. Prior exposure increases perceived accuracy of fake news. Journal of Experimental Psychology (forthcoming), 2019.
+
+Erich Prisner. Game Theory Through Examples. Mathematical Association of America, 2014.
+
+Joaquin Quionero-Candela, Masashi Sugiyama, Anton Schwaighofer, and Neil D. Lawrence. Dataset Shift in Machine Learning. The MIT Press, 2009.
+
+Neil C. Rabinowitz. Meta-learners’ learning dynamics are unlike learners’. ArXiv preprint, 2019.
+
+David Robson. The myth of the online echo chamber, 2018. URL http://www.bbc.com/ future/story/20180416-the-myth-of-the-online-echo-chamber.
+
+Peter Schulam and Suchi Saria. Reliable decision support using counterfactual models. In Neural Information Processing Systems, 2017.
+
+Virag Shah, Jose Blanchet, and Ramesh Johari. Bandit learning with positive externalities. Neural Information Processing Systems, 2018.
+
+Chengcheng Shao, Giovanni Luca Ciampaglia, Onur Varol, Kai-Cheng Yang, Alessandro Flammini, and Filippo Menczer. The spread of low-credibility content by social bots. Nature Communications, 9(4787), 2018.
+
+Jasper Snoek, Hugo Larochelle, and Ryan P Adams. Practical Bayesian optimization of machine learning algorithms. In Neural Information Processing Systems, 2012.
+
+Richard S Sutton and Andrew G Barto. Introduction to Reinforcement Learning. MIT Press, 1998.
+
+Richard S Sutton, Anna Koop, and David Silver. On the role of tracking in stationary environments. In International conference on Machine learning, 2007.
+
+Techopedia. Filter bubble, 2018. URL https://www.techopedia.com/definition/ 28556/filter-bubble.
+
+Chih-Chun Wang, Sanjeev R Kulkarni, and H Vincent Poor. Bandit problems with side observations. IEEE Transactions on Automatic Control, 50(3):338–355, 2005.
+
+Wikipedia contributors. Confirmation bias — Wikipedia, the free encyclopedia, 2018. URL https://en.wikipedia.org/w/index.php?title=Confirmation_bias& oldid=875026726. [Online; accessed 20-January-2019].
+
+Ronald J. Williams. Simple statistical gradient-following algorithms for connectionist reinforcement learning. In Machine Learning, pp. 229–256, 1992.
+
+# APPENDICES
+
+# 8 CONTENT RECOMMENDATION IN THE WILD
+
+Filter bubbles, the spread of fake news, and other techno-social issues are widely reported to be responsible for the rise of populism (Groshek & Koc-Michalska, 2017), increase in racism and prejudice against immigrants and refugees (Noble, 2018), increase in social isolation and suicide (Luxton et al., 2012), and, particularly with reference to the 2016 US elections, are decried as threatening the foundations of democracy (El-Bermawy, 2016). Even in 2013, well before the 2016 American elections, a World Economic Forum report identified these problems as a global crisis (Lee Howell, 2013).
+
+We focus on two related issues in which content recommendation algorithms play a role: fake news and filter bubbles.
+
+# 8.1 FAKE NEWS
+
+Fake news (also called false news or junk news) is an extreme version of yellow journalism, propaganda, or clickbait, in which media that is ostensibly providing information focuses on being eye-catching or appealing, at the expense of the quality of research and exposition of factual information. Fake news is distinguished by being specifically and deliberately created to spread falsehoods or misinformation (Merriam-Webster, 2017; Mihailidis & Viotty, 2017).
+
+Why does fake news spread? It may at first seem the solution is simply to educate people about the truth, but research tells us the problem is more multifaceted and insidious, due to a combination of related biases and cognitive effects including confirmation bias (people are more likely to believe things that fit with their existing beliefs), priming (exposure to information unconsciously influences the processing of subsequent information, i.e. seeing something in a credible context makes things seem more credible) and the illusory truth effect (i.e. people are more likely to believe something simply if they are told it is true).
+
+Allcott & Gentzkow (2017) track about 150 fake news stories during the 2016 US election, and find the average American adult saw 1-2 fake news stories, just over half believed the story was true, and likelihood of believing fake news increased with ideological segregation (polarization) of their social media. Shao et al. (2018) examine the role of social bots in spreading fake news by analyzing 14 million Twitter messages. They find that bots are far more likely than humans to spread misinformation, and that success of a fake news story (in terms of human retweets) was heavily dependent on whether bots had shared the story.
+
+Pennycook et al. (2019) examine the role of the illusory truth effect in fake news. They find that even a single exposure to a news story makes people more likely to believe that it is true, and repeat viewings increase this likelihood. They find that this is not true for extremely implausible statements (e.g. “the world is a perfect cube”), but that “only a small degree of potential plausibility is sufficient for repetition to increase perceived accuracy” of the story. The situation is further complicated by peoples’ inability to distinguish promoted content from real news - Amazeen & Wojdynski (2018) find that fewer than 1/10 people were able to tell when content was an advertisement, even when it was explicitly labelled as such. Similarly, Fazio et al. (2015) find that repeated exposure to incorrect trivia make people more likely to believe it, even when they are later able to identify the trivia as incorrect.
+
+# 8.2 FILTER BUBBLES
+
+Filter bubbles, a term coined and popularized by Pariser (2011) are created by positive or negative feedback loops which encourage users or groups of users towards increasing within-group similarity, while driving up between-group dissimilarity. The curation of this echo chamber is called selfselection (people are more likely to look for or select things that fit their existing preferences), and favours what Techopedia (2018) calls intellectual isolation. In the context of social and political opinions, this is often called the polarization effect (Wikipedia contributors, 2018).
+
+Filter bubbles can be encouraged by algorithms in two main ways. The first is the most commonly described: simply by showing content that is similar to what a user has already searched for, search or recommender systems create a positive feedback loop of increasingly-similar content (Pariser, 2011; Kayhan, 2015). The second way is similar but opposite - if the predictions of an algorithm are good for a certain group of people, but bad for others, the algorithm can do better on its metrics by driving hard-to-predict users away. Then new users to the site will either be turned off entirely, or see an artificially homogenous community of like-minded peers, a phenomena Shah et al. (2018) call positive externalities.
+
+In a study of 50,000 US-based internet users, Flaxman & Goel (2015) find that two things increase with social media and search engine use: (1) exposure of an individual to opposing or different viewpoints, and (2) mean ideological distance between users. Many studies cite the first result as evidence of the benefits of internet and social media (Robson, 2018; Bakshy et al., 2015), but the correlation of exposure with ideological distances demonstrates that exposure is not enough, and might even be counterproductive.
+
+Facebook’s own study on filter bubbles results show that the impact of the news feed algorithm on filter bubble “size” (a measure of homogeneity of posts relative to a baseline) is almost as large as the impact of friend group composition (Bakshy et al., 2015). Kayhan (2015) specifically study the role of search engines in confirmation bias, and find that search context and the similarity of results in search engine results both reinforce existing biases and increase the likelihood of future biased searches. Nguyen et al. (2014) similarly study the effect of recommender systems on individual users’ content diversity, and find that the set of options recommended narrows over time.
+
+Filter bubbles create an ideal environment for the spread of fake news: they increase the likelihood of repeat viewings of similar content, and because of the illusory truth effect, that content is more likely to be believed and shared (Pennycook et al., 2019; DiFranzo & Gloria-Garcia, 2017; Pariser, 2011). We are not claiming that HI-ADS are entirely or even mostly responsible for these problems, but we do note that they can play a role that is worth addressing.
+
+![](images/5c7a8c62362379a4ec118b2d7cacc1bfea380d1d1281d25198b5cb5100eaef60.jpg)  
+incentive-compatible $\beta = 0 . 5$ )   
+Figure 8: Average level of non-myopic (i.e. cooperate) behavior learned by agents in the unit test for HI-ADS. Despite making the inner loop fully myopic $( \gamma = 0 )$ ), population-based training (PBT) can cause HI-ADS, leading agents to choose the cooperate action (top row). context swapping successfully prevents this (bottom row). Columns (from left to right) show results for populations of 10, 100, and 1000 learners. In the legend, “interval” refers to the interval $( T )$ of PBT (see Sec. 2.2). Sufficiently large populations and short intervals are necessary for PBT to induce nonmyopic behavior.
+
+# 9 EXTRA EXPERIMENTS AND REPRODUCIBILITY DETAILS
+
+# 9.1 HI-ADS UNIT TEST
+
+We used REINFORCE (Williams, 1992) with discount factor $\gamma = 0$ as the baseline/IL optimizer. PBT (with default settings, see Section 2.2) is used to tune the learning rate, with reward on the final time-step of the interval as the performance measure for PBT.
+
+Formally, the environment is not a $2 \mathbf { x } 2$ game (as the original prisoner’s dilemma); it’s a partially observable Markov Decision Process (Åström, 1965; Kaelbling et al., 1998):
+
+$$
+\begin{array} { r l r } & { } & { s _ { t } , o _ { t } = a _ { t - 1 } , \{ \} } \\ & { } & { a _ { t } \in \{ \mathsf { d e f e c t } , ~ \mathsf { c o o p e r a t e } \} } \\ & { } & { P ( s _ { t } , a _ { t } ) = \delta ( a _ { t } ) } \\ & { } & { R ( s _ { t } , a _ { t } ) = I ( s _ { t } = \mathsf { c o o p e r a t e } ) + \beta ~ I ( a _ { t } = \mathsf { c o o p e r a t e } ) - 1 / 2 } \end{array}
+$$
+
+where $I$ is an indicator function, and $\beta = - 1 / 2$ is a parameter controlling the alignment of incentives. The initial state is sampled as $s _ { 0 } \sim U ( { \tt d e f e c t }$ , cooperate). Policies are represented by a single real-valued parameter $\theta$ (initialized as $\theta \sim \mathcal { N } ( 0 , 1 ) \big )$ ) passed through a sigmoid whose output represents $P ( a _ { t } = \mathrm { \bar { d e f e c t } } )$ .
+
+# 9.1.1 ALIGNMENT OF INCENTIVES EXPLORATION
+
+This section presents an exploration of the parameter $\beta$ , which controls the alignment of incentives in the HI-ADS unit tests (see Table 2).
+
+To clarify the interpretation of experiments, we distinguish between environments in which myopic (defect) vs. nonmyopic (cooperate) incentives are opposed, orthogonal, or compatible. Note that in this unit test myopic behaviour (defection) is what we want to see.
+
+1. Incentive-opposed: Optimal myopic behavior is incompatible with optimal nonmyopic behavior (classic prisoner’s dilemma; these experiments are in the main paper).   
+2. Incentive-orthogonal: Optimal myopic behavior may or may not be optimal nonmyopic behavior.   
+3. Incentive-compatible: Optimal myopic behavior is necessarily also optimal nonmyopic behavior.
+
+We focused on incentive-opposed environment $( \beta = - 1 / 2 )$ ) in the main paper in order to demonstrate that HI-ADS can be powerful enough to change the behavior of the system in an undesirable way. Here we also explore incentive-compatible and incentive-orthogonal environments because they provide useful baselines, helping us distinguish a systematic bias towards nonmyopic behavior from other reasons (such as randomness or optimization issues) for behavior that does not follow a myopically optimal policy.
+
+# 9.1.2 WORKING THROUGH A DETAILED EXAMPLE FOR PBT WITH $T = 1$
+
+To help provide intuition on how (mechanistically) PBT could lead to persistent levels of cooperation, we walk through a simple example (with no inner loop). Consider PBT with $T = 1$ and a population of 5 deterministic agents $A _ { 1 } , . . . , A _ { 5 }$ playing cooperate and receiving reward of $r ( A _ { i } ) = 0$ . Now suppose $A _ { 1 }$ suddenly switches to play defect. Then $r ( A _ { 1 } ) = 1 / 2$ on the next time-step (while the other agents’ reward is still 0), and so PBT’s EXPLOIT step will copy $A _ { 1 }$ (without loss of generality to $A _ { 2 }$ ). On the following time-step, $r ( A _ { 2 } ) = 1 / 2$ , and $r ( A _ { 1 } ) = - 1 / 2$ , so PBT will clone $A _ { 2 }$ to $A _ { 1 }$ , and the cycle repeats. Similar reasoning applies for larger populations, and $T > 1$ .
+
+# 9.1.3 Q-LEARNING EXPERIMENT DETAILS
+
+We show that, under certain conditions, Q-learning can learn to (primarily) cooperate, and thus fails the HI-ADS unit test. We estimate Q-values using the sample-average method, which is guaranteed to converge in the fully observed, tabular case (Sutton & Barto, 1998). The agent follows the $\epsilon$ -greedy policy with $\epsilon = 0 . 1$ . In order to achieve this result, we additionally start the agent off with one synthetic memory where both state and action are defect and therefor $R ( \mathrm { d e f e c t } ) = - . 5$ , and we hard-code the starting state to be cooperate (which normally only happens $50 \%$ of the time). Without this kind of an initialization, the agent always learns to defect. However, under these conditions, we find that 10/30 agents learned to play cooperate most of the time, with $Q$ (cooperate) and $Q ( { \mathrm { d e f e c t } } )$ both hovering around $- 0 . 0 7$ , while others learn to always defect, with $Q ( { \mathrm { c o o p e r a t e } } ) \approx - 0 . 9 2$ and $Q ( { \mathrm { d e f e c t } } ) \approx - 0 . 4 5$ . context swapping, however, prevents majority-cooperate behavior from ever emerging, see Figure 11.
+
+Table 2: $\beta$ controls the extent to which myopic and nonmyopic incentives are aligned.   
+
+<table><tr><td>β</td><td>Environment</td><td>Cooperating</td></tr><tr><td>&lt;0</td><td>incentive-opposed</td><td>yields less reward on the current time-step (myopically detrimental)</td></tr><tr><td>=0</td><td>incentive-orthogonal</td><td>does not affect the current reward (myopically indifferent)</td></tr><tr><td>v0</td><td>incentive-compatible</td><td>yields more reward on the current time-step (myopically beneficial)</td></tr></table>
+
+# 9.1.4 Q-LEARNING: FURTHER RESULTS
+
+To give a more representative picture of how often Q-learning fails the unit test, we run a larger set of experiments with Q-learning, results are in Figure 10. It’s possible that the failure of Q-learning is not persistent, since we have not proved otherwise, but we did run much longer experiments and still observe persistent failure, see Figure 9.
+
+![](images/93bc9858be4e122e4aa54709b97526f0b712f8b3de8f38002ec30b69c8b8d7ee.jpg)  
+Figure 9: The same experiments as Figures 6, 10, run for 50,000 time-steps instead of 3000, to illustrate the persistence of non-myopic behavior.
+
+![](images/3e860d515c7ba817cb56c44c846a248ed6df5909c2c2c53c73553c58f29bb252.jpg)  
+Figure 10: More independent experiments with Q-learning, exactly following Figure 6. Q-learning fails the unit test in a total of 10/30 experiments (including those from Figure 6).
+
+![](images/18011e0877d74bf4ea8b3877bc9a66d2f427d421200bb03d1cb2dd7167b3d022.jpg)  
+Figure 11: More independent experiments with Q-learning, exactly following Figure 6, except also using context swapping. This leads to a $100 \%$ success rate on the unit test.
+
+# 9.2 CONTENT RECOMMENDATION
+
+# 9.2.1 ENVIRONMENT DETAILS
+
+The evironment has the following components:
+
+1. User type, $x ^ { t }$ : categorical variable representing different types of users. The content recommender conditions its predictions on the type of the current user.   
+2. User loyalty, $\mathbf { g } ^ { t }$ : the propensity for users of each type to use the platform. User $x ^ { t }$ is sampled from a categorical distribution with parameters given by softmax $( \mathbf { g } ^ { t } )$ .   
+3. Article type, $y ^ { t }$ : a categorical variable (one-hot encoding) representing the type of article selected by the user.   
+4. User interests, $\mathbf { W } ^ { t }$ : a matrix whose entries $W _ { x , y } ^ { t }$ represent the average interest user of type $x$ have in articles of type $y$ .
+
+At each time step $t$ , a user $x ^ { t }$ is sampled from a categorical distribution (based on the loyalty of the different user types), then the recommendation system selects which type of article to present in the top position, and finally, the user selects an article. The goal of the recommendation system is to predict the likelihood that the user would click on each of the available articles, in order to select the one which is most interesting to the user.
+
+User loyalty for $x ^ { t }$ then changes in accordance with the self-selection effect, increasing or decreasing proportionally to their interest in the top article. The interests of user type $x ^ { t }$ (represented by a column of $\mathbf { W } ^ { t }$ ) also change; in accordance with the illusory truth effect, their interest in the topic of the top article (as chosen by the recommender system) always increases. Overall, this environment is an extremely crude representation of reality, but it allows us to incorporate both the effects of self-selection (via covariate shift), and the illusory truth effect (via concept shift).
+
+Formally, this environment is similar to a POMDP\R, i.e. a POMDP with no reward function, also known as a world model (Armstrong & O’Rourke, 2017; Hadfield-Menell et al., 2017); the difference is that the learner observes the input before acting and only observes the target after acting. The states, observations, and actions given below.
+
+$$
+\begin{array} { c } { { \boldsymbol { s } ^ { t } = ( \mathbf { g } ^ { t } , \mathbf { W } ^ { t } , \boldsymbol { x } ^ { t } , \boldsymbol { y } ^ { t } ) } } \\ { { \boldsymbol { o } _ { \mathrm { p r e } } ^ { t } , ~ \boldsymbol { a } ^ { t } , ~ \boldsymbol { o } _ { \mathrm { p o s t } } ^ { t } = ( \boldsymbol { x } ^ { t } , \hat { \boldsymbol { y } } ^ { t } , \boldsymbol { y } ^ { t } ) } } \end{array}
+$$
+
+The state transition function is defined by:
+
+$$
+\begin{array} { r l r } { \mathbf { g } _ { x ^ { t } } ^ { t + 1 } = \mathbf { g } _ { x ^ { t } } ^ { t } + \alpha _ { 1 } W _ { x ^ { t } , \hat { y } ^ { t } } ^ { t } } & { } & \\ { \mathbf { W } _ { x ^ { t } , \hat { y } ^ { t } } ^ { t + 1 / 2 } = W _ { x ^ { t } , \hat { y } ^ { t } } ^ { t } + \alpha _ { 2 } ; \quad \mathbf { W } _ { x ^ { t } } ^ { t + 1 } = \frac { \mathbf { W } _ { x ^ { t } } ^ { t + 1 / 2 } } { \Vert \mathbf { W } _ { x ^ { t } } ^ { t + 1 / 2 } \Vert _ { 2 } } } & { } & \\ { x ^ { t + 1 } \sim \mathrm { s o f t m a x } ( \mathbf { g } ^ { t + 1 } ) } & { } & \\ { y ^ { t + 1 } \sim \mathrm { s o f t m a x } ( \mathbf { W } _ { x ^ { t + 1 } } ^ { t + 1 } ) } & { } & \end{array}
+$$
+
+Where $\hat { y } ^ { t }$ is the top article as chosen by the recommender, and $\alpha _ { 1 }$ , $\alpha _ { 2 }$ represent the rate of covariate and concept shift (respectively). The update for $\mathbf { W } ^ { t + 1 }$ merely increases the interest of user type $x ^ { t }$ in article type $\hat { y } ^ { t }$ , then normalizes the interests for that user type.
+
+# 9.2.2 REPRODUCIBILITY DETAILS
+
+For these experiments, the recommendation system is a ReLU-MLP with 1 hidden layer of 100 units, trained via supervised learning with SGD (learning rate $= 0 . 0 1 $ ) to predict which article a user will select. Actions are sampled from the MLP’s predictive distribution. We apply PBT without any hyperparameter selection (this amounts to just doing the EXPLOIT step), and an interval of 10, selecting on accuracy. We use a population of 20 learners (whether applying PBT or not), and match random seeds for the trials with and without PBT. We initialize $\mathbf { g } ^ { 1 }$ and $\dot { \mathbf { W } } ^ { 1 }$ to be the same across the 20 copies of the environment (i.e. the learners start with the same user population), but these values diverge throughout learning. For the environment, we set the number of user and article types both to 10. Initial user loyalties are randomly sampled from $\mathcal { N } ( 0 , 0 . 0 3 )$ , $\alpha _ { 1 } = 0 . 0 3$ , and $\alpha _ { 2 } = 0 . 0 0 3$ .
+
+# 9.2.3 DETAILS OF EVALUATION
+
+We measure concept shift (change in $P ( \boldsymbol { y } | \mathbf { x } ) .$ ) as the cosine distance between each user types’ initial and current interest vectors. And we measure covariate shift (change in $P ( \mathbf { x } ) )$ ) as the KL-divergence between the current and initial user distributions, parametrized by $\mathbf { g } ^ { 1 }$ and $\mathbf { g } ^ { t }$ , respectively. Results are presented in 7 (main text). In Figure 12, we additionally plot concept shift and covariate shift as a function of accuracy. We observe that for both types of ADS, at low levels of accuracy PBT actually causes less shift than occur in baseline agents; HI-ADS are only observed for accuracies above $60 \%$ . This suggests that only relatively strong performers are able to pick up on the HI-ADS revealed by PBT (Fig. 12).
+
+![](images/4ee83ea25b84d00e9bd2c93c72bcfe37100fc4c406703659c71aeb8e4ba6b65a.jpg)  
+Figure 12: Amount of auto-induced covariate shift (left) and auto-induced concept shift (right) as a function of performance (accuracy) averaged over all trials, learners, and time-steps. Only relatively strong learners (those which achieve accuracy $> 6 0 \%$ ) exhibit HI-ADS.
+
+# 9.2.4 CONTEXT SWAPPING IN CONTENT RECOMMENDATION
+
+We believe context swapping is not appropriate for the content recommendation environment, since when the environments diverge, optimal behavior may differ across environments. Nevertheless, we ran experiments with it for completeness. The main effect appears to be to hamper learning when PBT is not used, see Figure 13. Notably, it does not appear to significantly influence the rate or extent of ADS when combined with PBT.
+
+# 9.2.5 EXPLORATION OF ENVIRONMENT PARAMETERS
+
+In Figure 14, we examine the effect of the rate-of-change parameters $( \alpha _ { 1 } , \alpha _ { 2 } )$ of the content recommendation environment on the results provided in the paper. As noted there, our results are qualitatively consistent so long as (1) the initial user distribution is approximately uniform, and (2) the covariate shift rate $( \alpha _ { 1 } )$ is faster than the concept shift rate $\left( \alpha _ { 2 } \right)$ . These distributions are updated by different mechanisms, and are not directly comparable. Concept shift changes the task more radically, requiring a learner to change its predictions, rather than just become accurate on a wider range of inputs. We conjecture that changes in $P ( \boldsymbol { y } | \boldsymbol { x } )$ must therefore be kept smooth enough for the outer loop to have pressure to capitalize on HI-ADS.
+
+![](images/32bff10367ed93273e67c56861072a16c782eb88536e394129f2eaa0da72bbf8.jpg)  
+Figure 13: Context swapping doesn’t have the desired effect in the content recommendation environment.
+
+![](images/a89f27f3d6a426966da5d92bf05f995060ebcc886850b49f7d5327641103133d.jpg)  
+Figure 14: Content recommendation results for different values of $\alpha _ { 1 } , \alpha _ { 2 }$
+
+# 10 OFFLINE Q-LEARNING CAN REVEAL INCENTIVES FOR ADS
+
+First, recall that this unit test is a POMDP, and the state is not observed. Since there are only 2 possible actions, a policy is defined by a single parameter $\theta = p ( { \mathrm { c o o p e r a t e } } )$ . Now, the state distribution is $P ( s = { \mathrm { c o o p e r a t e } } ) = \theta$ (ignoring the first state, which is appropriate in the limit of infinite data). More specifically, the probability of each state-action combination are as follows:
+
+Suppose we have a dataset of $N$ examples generated by following a fixed policy.
+
+$$
+\begin{array} { r l } & { Q ( C ) = \frac { | s = C , a = C | R ( s = C , a = C ) + | s = D , a = C | R ( s = D , a = C ) } { | a = C | } } \\ & { \quad \quad = \frac { N P ( s = C , a = C ) R ( s = C , a = C ) + N P ( s = D , a = C ) R ( s = D , a = C ) } { N P ( a = C ) } } \\ & { \quad \quad = \frac { P ( s = C , a = C ) R ( s = C , a = C ) + P ( s = D , a = C ) R ( s = D , a = C ) } { P ( a = C ) } } \\ & { \quad \quad = \frac { \theta ^ { 2 } R ( s = C , a = C ) + \theta ( 1 - \theta ) R ( s = D , a = C ) } { \theta } } \\ & { \quad \quad = \theta R ( s = C , a = C ) + ( 1 - \theta ) R ( s = D , a = C ) } \\ & { \quad \quad = \theta ( 0 ) + ( 1 - \theta ) ( - 1 ) } \\ & { \quad \quad = \theta - 1 } \end{array}
+$$
+
+$$
+\begin{array} { l } { Q ( D ) = \frac { \left| s = C , a = D \right| R \left( s = C , a = D \right) + \left| s = D , a = D \right| R \left( s = D , a = D \right) } { \left| a = D \right| } } \\ { \ = \frac { P \left( s = C , a = D \right) R \left( s = C , a = D \right) + P \left( s = D , a = D \right) R \left( s = D , a = D \right) } { P \left( a = D \right) } } \\ { \ = P ( s = C ) R ( s = C , a = D ) + P ( s = D ) R ( s = D , a = D ) } \\ { \ = \theta ( 1 / 2 ) + ( 1 - \theta ) ( - 1 / 2 ) } \\ { \ = 1 / 2 ( 2 \theta - 1 ) } \\ { \ = \theta - 1 / 2 } \end{array}
+$$
+
+So we see that $Q ( D ) > Q ( C )$ , regardless of $\theta$ .
+
+Now, suppose instead that we have $N$ examples from each of 2 different policies (given by parameters $\theta _ { 1 }$ and $\theta _ { 2 }$ ) operating in different environments. Intuitively, this sort of data might arise in practice from “A/B testing”, where 2 different users have been assigned to 2 different policies in order to compare the policies’ performance. We now use $_ { D C }$ to represent $s = D , a = C$ , etc.
+
+$$
+\begin{array} { r l } & { Q ^ { \theta _ { 1 } , \theta _ { 2 } } ( C ) = \frac { | C C | R ( C C ) + | D C | | R ( U C ) } { | C | } } \\ & { \quad \quad \quad = \frac { N \left( P ^ { \theta _ { 1 } } ( C C ) + P ^ { \theta _ { 2 } } ( C C ) \right) R ( C C ) + N ( P ^ { \theta _ { 1 } } ( D C ) + P ^ { \theta _ { 2 } } ( D C ) ) R ( D C ) } { N \left( P ^ { \theta _ { 1 } } ( C ) + P ^ { \theta _ { 2 } } ( C ) \right) } } \\ & { \quad \quad \quad = \frac { ( P ^ { \theta _ { 1 } } ( C C ) + P ^ { \theta _ { 2 } } ( C C ) ) I ( C C ) C + ( P ^ { \theta _ { 1 } } ( D C ) + P ^ { \theta _ { 2 } } ( D C ) ) R ( D C ) } { ( F ^ { \theta _ { 1 } } ( C ) + P ^ { \theta _ { 2 } } ( C ) ) } } \\ & { \quad \quad \quad = \frac { ( \theta _ { 1 } ^ { 2 } + \theta _ { 2 } ^ { 2 } ) R ( C C ) + ( \theta _ { 1 } ( 1 - \theta _ { 1 } ) + \theta _ { 2 } ( 1 - \theta _ { 2 } ) ) R ( D C ) } { \theta _ { 1 } + \theta _ { 2 } } } \\ & { \quad \quad \quad = - \frac { \theta _ { 1 } ( 1 - \theta _ { 1 } ) + \theta _ { 2 } ( 1 - \theta _ { 2 } ) } { \theta _ { 1 } + \theta _ { 2 } } } \\ & { \quad \quad \quad = \frac { \theta _ { 1 } ^ { 2 } - \theta _ { 1 } + \theta _ { 2 } ^ { 2 } - \theta _ { 2 } } { \theta _ { 1 } + \theta _ { 2 } } } \\ & { \quad \quad \quad = \frac { \theta _ { 1 } ^ { 2 } - \theta _ { 1 } + \theta _ { 2 } ^ { 2 } - \theta _ { 2 } } { \theta _ { 1 } + \theta _ { 2 } } } \\ & { \quad \quad \quad = \frac { \theta _ { 1 } ^ { 2 } - \theta _ { 1 } + \theta _ { 2 } ^ { 2 } - \theta _ { 2 } } { \theta _ { 1 } + \theta _ { 2 } } } \end{array}
+$$
+
+$$
+\begin{array} { l } { { Q ^ { \theta _ { 1 } , \theta _ { 2 } } ( D ) = \frac { \left| C D \right| R \left( C D \right) + \left| D D \right| R \left( D D \right) } { \left| D \right| } } } \\ { { \ = \frac { \left( P ^ { \theta _ { 1 } } \left( C D \right) + P ^ { \theta _ { 2 } } \left( C D \right) \right) R \left( C D \right) + \left( P ^ { \theta _ { 1 } } \left( D D \right) + P ^ { \theta _ { 2 } } \left( D D \right) \right) R \left( D D \right) } { \left( P ^ { \theta _ { 1 } } \left( D \right) + P ^ { \theta _ { 2 } } \left( D \right) \right) } } } \\ { { \ = \frac { 1 / 2 \left( \theta _ { 1 } \left( 1 - \theta _ { 1 } \right) + \theta _ { 2 } \left( 1 - \theta _ { 2 } \right) \right) - 1 / 2 \left( ( 1 - \theta _ { 1 } ) ^ { 2 } + ( 1 - \theta _ { 2 } ) ^ { 2 } \right) } { 2 - \theta _ { 1 } - \theta _ { 2 } } } } \\ { { \ = \frac { \left( 2 \theta _ { 1 } - 1 \right) \left( 1 - \theta _ { 1 } \right) + \left( 2 \theta _ { 2 } - 1 \right) \left( 1 - \theta _ { 2 } \right) } { 4 - 2 \theta _ { 1 } - 2 \theta _ { 2 } } } } \end{array}
+$$
+
+Now, in Figure 15 we see that $Q ( C ) > Q ( D )$ when one of the policies cooperates with high probability, and the other defects with high probability. Intuitively, the result of pooling data from 2 such policies is very similar to collecting data from an $\epsilon$ -greedy policy trained online (as in Figure 6).
+
+![](images/f8cf957c6b390dfb762d079853f1b8f334d627fe895c972607d8fe3c71e3938d.jpg)  
+Figure 15: Offline Q-learning can also reveal HI-ADS, when pooling data from different (policy, environment) pairs. Yellow regions represent policy pairs for which $\mathbf { \bar { Q } } ( C ) > Q ( D )$ , resulting in non-myopic behavior.
\ No newline at end of file
diff --git a/parse/train/3FK30d5BZdu/3FK30d5BZdu_content_list.json b/parse/train/3FK30d5BZdu/3FK30d5BZdu_content_list.json
new file mode 100644
index 0000000000000000000000000000000000000000..d0d33526d4606521d660fdd77b6c3b6a8711b44e
--- /dev/null
+++ b/parse/train/3FK30d5BZdu/3FK30d5BZdu_content_list.json
@@ -0,0 +1,2407 @@
+[
+    {
+        "type": "text",
+        "text": "HIDDEN INCENTIVES FOR AUTO-INDUCED DISTRIBUTIONAL SHIFT ",
+        "text_level": 1,
+        "bbox": [
+            174,
+            99,
+            828,
+            145
+        ],
+        "page_idx": 0
+    },
+    {
+        "type": "text",
+        "text": "Anonymous authors Paper under double-blind review ",
+        "bbox": [
+            183,
+            170,
+            398,
+            198
+        ],
+        "page_idx": 0
+    },
+    {
+        "type": "text",
+        "text": "ABSTRACT ",
+        "text_level": 1,
+        "bbox": [
+            454,
+            234,
+            544,
+            251
+        ],
+        "page_idx": 0
+    },
+    {
+        "type": "text",
+        "text": "Decisions made by machine learning systems have increasing influence on the world, yet it is common for machine learning algorithms to assume that no such influence exists. An example is the use of the i.i.d. assumption in content recommendation. In fact, the (choice of) content displayed can change users’ perceptions and preferences, or even drive them away, causing a shift in the distribution of users. We introduce the term auto-induced distributional shift (ADS) to describe the phenomenon of an algorithm causing a change in the distribution of its own inputs. Our goal is to ensure that machine learning systems do not leverage ADS to increase performance when doing so could be undesirable. We demonstrate that changes to the learning algorithm, such as the introduction of meta-learning, can cause hidden incentives for auto-induced distributional shift (HI-ADS) to be revealed. To address this issue, we introduce ‘unit tests’ and a mitigation strategy for HI-ADS, as well as a toy environment for modelling real-world issues with HIADS in content recommendation, where we demonstrate that strong meta-learners achieve gains in performance via ADS. We show meta-learning and Q-learning both sometimes fail unit tests, but pass when using our mitigation strategy. ",
+        "bbox": [
+            233,
+            266,
+            766,
+            489
+        ],
+        "page_idx": 0
+    },
+    {
+        "type": "text",
+        "text": "1 INTRODUCTION ",
+        "text_level": 1,
+        "bbox": [
+            176,
+            513,
+            336,
+            529
+        ],
+        "page_idx": 0
+    },
+    {
+        "type": "text",
+        "text": "Consider a content recommendation system whose performance is measured by accuracy of predicting what users will click. This system can achieve better performance by either 1) making better predictions, or 2) changing the distribution of users such that predictions are easier to make. We propose the term auto-induced distributional shift (ADS) to describe this latter kind of distributional shift, caused by the algorithm’s own predictions or behaviour (Figure 1). ",
+        "bbox": [
+            174,
+            537,
+            826,
+            608
+        ],
+        "page_idx": 0
+    },
+    {
+        "type": "image",
+        "img_path": "images/833d9ac6619b77e4b4c37b9014af227c1cffc0cdcc8236556ae4c01b6b89cd90.jpg",
+        "image_caption": [
+            "Figure 1: Distributions of users over time. Left: A distribution which remains constant over time, following the i.i.d assumption. Right: Auto-induced Distributional Shift (ADS) results in a change in the distribution of users in our content recommendation environment. (see Section 5.2 for details). "
+        ],
+        "image_footnote": [],
+        "bbox": [
+            282,
+            626,
+            732,
+            734
+        ],
+        "page_idx": 0
+    },
+    {
+        "type": "text",
+        "text": "ADS are not inherently bad, and are sometimes even desirable. But they can cause problems if they occur unexpectedly. It is typical in machine learning (ML) to assume (e.g. via the i.i.d. assumption) that (2) will not happen. However, given the increasing real-world use of ML algorithms, we believe it is important to model and experimentally observe what happens when assumptions like this are violated. This is the motivation of our work. ",
+        "bbox": [
+            174,
+            804,
+            825,
+            875
+        ],
+        "page_idx": 0
+    },
+    {
+        "type": "text",
+        "text": "In many cases, including news recommendation, we would consider (2) a form of cheating—the algorithm changed the task rather than solving it as intended. We care which means the algorithm used to solve the problem (e.g. (1) and/or (2)), but we only told it about the ends, so it didn’t know not to ’cheat’. This is an example of a specification problem (Leike et al., 2017; Ortega et al., 2018): a problem which arises from a discrepancy between the performance metric (maximize accuracy) and “what we really meant”: in this case, to maximize accuracy via (1). ",
+        "bbox": [
+            176,
+            882,
+            823,
+            924
+        ],
+        "page_idx": 0
+    },
+    {
+        "type": "text",
+        "text": "",
+        "bbox": [
+            174,
+            103,
+            823,
+            146
+        ],
+        "page_idx": 1
+    },
+    {
+        "type": "text",
+        "text": "Ideally, we’d like to quantify the desirability of all possible means, e.g. assign appropriate rewards to all potential strategies and “side-effects”, but this is intractable for real-world settings. Using human feedback to learn reward functions which account for such impacts is a promising approach to specifying desired behavior (Leike et al., 2018; Christiano et al., 2017). But the same issue can arise whenever human feedback is used in training: one means of improving performance could be to alter human preferences, making them easier to satisfy. Thus in this work, we pursue a complementary approach: managing learners’ incentives. ",
+        "bbox": [
+            174,
+            152,
+            825,
+            250
+        ],
+        "page_idx": 1
+    },
+    {
+        "type": "text",
+        "text": "A learner has an incentive to behave in a certain way when doing so can increase performance (e.g. accuracy or reward). Informally, we say an incentive is hidden when the learner behaves as if it were not present. But we note that changes to the learning algorithm or training regime could cause previously hidden incentives to be revealed, resulting in unexpected and potentially undesirable behaviour. Managing incentives (e.g. controlling which incentives are hidden/ revealed) can allow algorithm designers to disincentivize broad classes of strategies (such as any that rely on manipulating human preferences) without knowing their exact instantiation.1 ",
+        "bbox": [
+            174,
+            257,
+            825,
+            354
+        ],
+        "page_idx": 1
+    },
+    {
+        "type": "text",
+        "text": "The goal of our work is to provide insights and practical tools for understanding and managing incentives, specifically hidden incentives for auto-induced distributional shift: HI-ADS. To study which conditions cause HI-ADS to be revealed, we present unit tests for detecting HI-ADS in supervised learning (SL) and reinforcement learning (RL). We also create an environment that models ADS in news recommendation, illustrating possible effects of revealing HI-ADS in this setting. ",
+        "bbox": [
+            174,
+            361,
+            825,
+            431
+        ],
+        "page_idx": 1
+    },
+    {
+        "type": "text",
+        "text": "The unit tests both have two means by which the learner can improve performance: one which creates ADS and one which does not. The intended method of improving performance is one that does not induce ADS; the other is ’hidden’ and we want it to remain hidden. A learner \"fails\" the unit test if it nonetheless pursues the incentive to increase performance via ADS. In both the RL and SL unit tests, we find that introducing an outer-loop of meta-learning (e.g. Population-Based Training (PBT) Jaderberg et al. (2017)) can lead to high levels of failure. Similarly, recommender systems trained with PBT induce larger drifts in user base and user interests. These results suggest that failure of our unit tests indicates that an algorithm is prone to revealing HI-ADS in other settings. Finally, we propose and test a mitigation strategy we call context swapping. The strategy consists of rotating learners through different environments throughout learning, so that they can’t see the results or correlations of their actions in one environment over longer time horizons. This effectively mitigates HI-ADS in our unit test environments, but did not work well in content recommendation experiments. ",
+        "bbox": [
+            174,
+            438,
+            825,
+            604
+        ],
+        "page_idx": 1
+    },
+    {
+        "type": "text",
+        "text": "2 BACKGROUND ",
+        "text_level": 1,
+        "bbox": [
+            174,
+            618,
+            326,
+            633
+        ],
+        "page_idx": 1
+    },
+    {
+        "type": "text",
+        "text": "2.1 META-LEARNING AND POPULATION BASED TRAINING ",
+        "text_level": 1,
+        "bbox": [
+            174,
+            643,
+            586,
+            656
+        ],
+        "page_idx": 1
+    },
+    {
+        "type": "text",
+        "text": "Meta-learning is the use of machine learning techniques to learn machine learning algorithms. This involves instantiating multiple learning scenarios which run in an inner loop $\\mathbf { \\Pi } ( \\mathbf { I I L } )$ , while an outer loop (OL) uses the outcomes of the inner loop(s) as data-points from which to learn which learning algorithms are most effective (Metz et al., 2019). The number of IL steps per OL step is called the interval. Many recent works focus on multi-task meta-learning, where the OL seeks to find learning rules that generalize to unseen tasks by training the IL on a distribution of tasks (Finn et al., 2017). Single-task meta-learning includes learning an optimizer for a single task (Gong et al., 2018), and adaptive methods for selecting models (Kalousis, 2000) or setting hyperparameters (Snoek et al., 2012). For simplicity in this initial study we focus on single-task meta-learning. ",
+        "bbox": [
+            173,
+            661,
+            825,
+            786
+        ],
+        "page_idx": 1
+    },
+    {
+        "type": "text",
+        "text": "Population-based training (PBT; Jaderberg et al., 2017) is a meta-learning algorithm that trains multiple learners $L _ { 1 } , . . . , L _ { n }$ in parallel, after each interval ( $T$ steps of IL) applying an evolutionary OL step which consists of: (1) Evaluate the performance of each learner, (2) Replace both parameters and hyperparameters of $20 \\%$ lowest-performing learners with copies of those from the $20 \\%$ highperforming learners (EXPLOIT). (3) Randomly perturb the hyperparameters (but not the parameters) of all learners (EXPLORE). ",
+        "bbox": [
+            174,
+            794,
+            825,
+            849
+        ],
+        "page_idx": 1
+    },
+    {
+        "type": "text",
+        "text": "",
+        "bbox": [
+            171,
+            103,
+            825,
+            132
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "text",
+        "text": "2.2 DISTRIBUTIONAL SHIFT AND CONTENT RECOMMENDATION ",
+        "text_level": 1,
+        "bbox": [
+            174,
+            147,
+            624,
+            160
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "text",
+        "text": "In general, distributional shift refers to change of the data distribution over time. In supervised learning with data $\\mathbf { x }$ and labels $y$ , this can be more specifically described as dataset shift: change in the joint distribution of $P ( \\mathbf { x } , y )$ between the training and test sets (Moreno-Torres et al., 2012; Quionero-Candela et al., 2009). As identified by Moreno-Torres et al. (2012), two common kinds of shift are: (1) Covariate shift: changing $P ( \\mathbf { x } )$ . In the example of content recommendation, this corresponds to changing the user base of the recommendation system. For instance, a media outlet which publishes inflammatory content may appeal to users with extreme views while alienating more moderate users. This self-selection effect (Kayhan, 2015) may appear to a recommendation system as an increase in performance, leading to a feedback effect, as previously noted by Shah et al. (2018). This type of feedback effect has been identified as contributing to filter bubbles and radicalization (Pariser, 2011; Kayhan, 2015). (2) Concept shift: changing $P ( \\boldsymbol { y } | \\mathbf { x } )$ . In the example of content recommendation, this corresponds to changing a given user’s interest in different kinds of content. For example, exposure to a fake news story has been shown to increase the perceived accuracy of (and thus presumably future interest in) the content, an example of the illusory truth effect (Pennycook et al., 2019). For further details on such effects in content recommendation, see Appendix 8. ",
+        "bbox": [
+            173,
+            165,
+            826,
+            375
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "text",
+        "text": "3 AUTO-INDUCED DISTRIBUTION SHIFT (ADS) ",
+        "text_level": 1,
+        "bbox": [
+            174,
+            391,
+            584,
+            407
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "text",
+        "text": "Auto-induced distribution shift (ADS) is distributional shift caused by an algorithm’s behaviour. This is in contrast to distributional shift which would happen even if the learner were not present - e.g. for a crash prediction algorithm trained on data from the summer, encountering snowy roads is an example of distributional shift, but not auto-induced distributional shift (ADS). ",
+        "bbox": [
+            174,
+            417,
+            826,
+            473
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "image",
+        "img_path": "images/e073c2dfa79ed13d098885e0aecd106a567f60dc458af17cb1a90e8412bd2143.jpg",
+        "image_caption": [],
+        "image_footnote": [],
+        "bbox": [
+            173,
+            493,
+            826,
+            626
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "text",
+        "text": "Figure 2: The widely studied problems of reinforcement learning (RL) with state $s$ , action $a$ , reward $r$ tuples, and i.i.d. supervised learning (SL) with inputs $x$ , predictions $\\hat { y }$ and loss $l$ (a,d) are free from incentive problems. We focus on cases where there are incentives present which the learner is not meant to pursue $^ { ( \\mathbf { b } , \\mathbf { c } ) }$ . Lines show paths of influence. The learner may have incentives to influence any nodes descending from its action, $A$ , or prediction, $\\hat { y }$ . Which incentives are undesirable (orange) or desirable (cyan) for the learner to pursue is context-dependent. ",
+        "bbox": [
+            173,
+            638,
+            825,
+            722
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "text",
+        "text": "We emphasize that ADS are not inherently bad or good; often ADS can even be desirable: consider an algorithm meant to alert drivers of imminent collisions. If it works well, such a system will help drivers avoid crashing, thus making self-refuting predictions which result in ADS. What separates desirable and undesirable ADS? The collision-alert system alters its data distribution in a way that is aligned with the goal of fewer collisions, whereas the news manipulation results in changes that are misaligned with the goal of better predicting existing users’ interests (Leike et al., 2018). ",
+        "bbox": [
+            174,
+            739,
+            825,
+            824
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "text",
+        "text": "In reinforcement learning (RL), ADS are typically encouraged as a means to increase performance. On the other hand, in supervised learning (SL), the i.i.d. assumption precludes ADS in theory. In practice, however, the possibility of using ADS to increase performance (and thus an incentive to do so) often remains. For instance, this occurs in online learning. In our experiments, we explicitly model such situations where i.i.d. assumptions are violated: We study the behavior of SL and myopic RL algorithms, in environments designed to include incentives for ADS, in order to understand when incentives are effectively hidden. Fig. 2 contrasts these settings with typical RL and SL. ",
+        "bbox": [
+            174,
+            830,
+            825,
+            928
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "text",
+        "text": "4 INCENTIVES ",
+        "text_level": 1,
+        "bbox": [
+            174,
+            102,
+            308,
+            117
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "For our study of incentives, we use the following terminology: an incentive for a behavior (e.g. an action, a classification, etc.) is present (not absent) to the extent that the behaviour will increase performance (e.g. reward, accuracy, etc.) (Everitt & Hutter, 2019). This incentive is revealed to (not hidden from) a learner if it would, at higher than chance levels, learn to perform the behavior given sufficient capacity and training experience. The incentive is pursued (not eschewed) by a learner if it actually performs the incentivized behaviour. Note even when an incentive is revealed, it may not be pursued, e.g. due to limited capacity and/or data, or simply chance. See Fig 3. ",
+        "bbox": [
+            173,
+            128,
+            825,
+            227
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "For example, in content recommendation, the incentive to drive users away is present if some user types are easier to predict than others. But this incentive may be hidden from the learner by using a myopic algorithm, e.g. one that does not see the effects of its actions on the distribution of users. The incentive might instead be revealed to the outer loop of a meta-learning algorithm like PBT, which does see the effects of learner’s actions. ",
+        "bbox": [
+            174,
+            234,
+            517,
+            358
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "image",
+        "img_path": "images/2ade91d4be27c22a6f418d3a986aa9448815d5a0ddb40aa939127276f08bc9bc.jpg",
+        "image_caption": [
+            "Figure 3: Types of incentives, and their relationship to ADS. "
+        ],
+        "image_footnote": [],
+        "bbox": [
+            532,
+            234,
+            821,
+            314
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "Even when this incentive is revealed, however, it might not end up being pursued. For example, this could happen if predicting which recommendations will drive away users is too difficult a learning problem, or if the incentive to do so is dominated by other incentives (e.g. change individual users’ interests, or improve accuracy of predictions). In general, it may be difficult to determine empirically which incentives are revealed, because failure to pursue an incentive can be due to limited capacity, insufficient training, and/or random chance. To address this challenge, we devise extremely simple environments (“unit tests”), where we can be confident that revealed incentives will be pursued. ",
+        "bbox": [
+            174,
+            366,
+            825,
+            463
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "4.1 HIDDEN INCENTIVES FOR AUTO-INDUCED DISTRIBUTIONAL SHIFT (HI-ADS) ",
+        "text_level": 1,
+        "bbox": [
+            173,
+            478,
+            754,
+            492
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "Following from the definitions in Sections 3 and 4, HI-ADS are incentives for behaviors that cause Auto-induced Distributional Shift that are hidden from the learner, i.e. the learner would not learn to perform the incentivized behaviors at higher than chance levels, even given infinite capacity and training experience. ",
+        "bbox": [
+            173,
+            498,
+            825,
+            555
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "Like ADS, HI-ADS are not necessarily problematic. Indeed, hiding incentives can be an effective method of influencing learner behavior. For example, hiding the incentive to manipulate users from a content recommendation algorithm could prevent it from influencing users in a way they would not endorse. However, if machine learning practitioners are not aware that incentives are present, or that properties of the learning algorithm are hiding them, then seemingly innocuous changes to the learning algorithm may reveal HI-ADS, and lead to significant unexpected changes in behavior. ",
+        "bbox": [
+            174,
+            563,
+            825,
+            646
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "Hiding incentives for ADS may seem counter-intuitive and counter-productive in the context of reinforcement learning (RL), where moving towards high-reward states is typically desirable. However, for real-world applications of RL, the ultimate goal is not a system that achieves high reward, but rather one that behaves according to the designer’s intentions. And as we discussed in the introduction, it can be intractable to design reward functions that perfectly specify intended behavior. Thus managing (e.g. hiding) some incentives can provide a useful tool for specification, even in RL. ",
+        "bbox": [
+            174,
+            652,
+            825,
+            736
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "We have several reasons for focusing on HI-ADS: (1) The issue of HI-ADS has not yet been identified, and thus is liable to be neglected in practice. Indeed, our “unit tests” are the first published empirical methodology for assessing whether incentives are hidden or revealed by different learning algorithms. (2) Machine learning algorithms are commonly deployed in settings where ADS are present, violating assumptions used to analyze their properties theoretically. This means learners could exploit ADS in unexpected and undesirable ways if incentives for ADS are revealed. Hiding these incentives heuristically (e.g. via off-line training) is a common approach, but potentially brittle (if practitioners don’t understand how HI-ADS could become revealed). In particular, meta-learning can reveal HI-ADS in online learning settings. (3) Substantial real-world issues could result from improper management of learner’s incentives. Examples include tampering with human-generated reward signals (Everitt & Hutter, 2018) (e.g. selecting news articles which manipulate user interests), and creating “self-fulfilling prophecies” (e.g. driving up the value of an asset by publicly predicting its value will increase (Armstrong & O’Rorke, 2017)). ",
+        "bbox": [
+            174,
+            742,
+            825,
+            924
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "4.2 REMOVING HI-ADS VIA CONTEXT SWAPPING ",
+        "text_level": 1,
+        "bbox": [
+            173,
+            103,
+            537,
+            117
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "text",
+        "text": "We propose a technique called context swapping for removing incentives for ADS. The technique trains $N$ learners in parallel, and shuffles the learners through $N$ different copies of the same (or similar) environments.We use a deterministic permutation of learners in environment copies, so that the $i$ -th learner inhabits the $j$ -th environment on time-steps $t$ where $~ j ~ = ~ ( i + t )$ mod $N$ , makes an observation, takes an action, and receives a reward before moving to the next environment. ",
+        "bbox": [
+            173,
+            122,
+            825,
+            191
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "text",
+        "text": "When $N$ is larger than the interval of the OL optimizer, each learner inhabits each copy for at most a single time-step before an OL step is applied. Under the assumption that different copies of the environment do not influence each other, this technique can address HI-ADS in practice, as we show in Sec. 5.1.1. ",
+        "bbox": [
+            174,
+            193,
+            517,
+            275
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "image",
+        "img_path": "images/06fd14fc3c3117d01ce92cc59924cf8d1fe7f705cf8d340291c9e7534c747cd1.jpg",
+        "image_caption": [
+            "Figure 4: Context swapping (right). "
+        ],
+        "image_footnote": [],
+        "bbox": [
+            534,
+            204,
+            820,
+            252
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "text",
+        "text": "5 EXPERIMENTS ",
+        "text_level": 1,
+        "bbox": [
+            174,
+            289,
+            328,
+            304
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "text",
+        "text": "In Section 5.1, we introduce ‘unit tests’ for HI-ADS. Our primary goal with these experiments is to convey a crisp understanding of potential issues caused by revealing HI-ADS. Put simply, our experiments show that you can have a learner which behaves as intended, and just by using metalearning (e.g. PBT), without changing the performance metric (e.g. loss or rewards), the learner’s behavior can change completely. We also show that context swapping is an effective mitigation technique in these environments. On the practical side, the unit tests can be used to compare learning algorithms and diagnose their propensity to reveal incentives. ",
+        "bbox": [
+            174,
+            309,
+            825,
+            406
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "text",
+        "text": "In Section 5.2, we model a content recommendation system. The goal of these experiments is to demonstrate how HI-ADS could create issues for real-world content recommendation systems such as news feeds, search results, or automated suggestions. They also validate the usefulness of the unit tests: algorithms that failed the unit tests also reveal HI-ADS in this setting. We emphasize that ADS takes place in this environment by construction. The point of our experiments is that meta-learning can increase the rate and/or extent of ADS, by revealing this incentive. We find that context swapping is not effective in this environment, highlighting the need for alternative mitigation strategies. ",
+        "bbox": [
+            174,
+            414,
+            825,
+            512
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "text",
+        "text": "5.1 HI-ADS UNIT TESTS ",
+        "text_level": 1,
+        "bbox": [
+            174,
+            521,
+            364,
+            535
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "text",
+        "text": "Unit test 1: Supervised Learning. This unit test consists of a simple prediction problem. There are no inputs, only an underlying state $s \\in \\{ 0 , 1 \\}$ , and targets $\\boldsymbol { y } \\in \\mathbb { R } ^ { 2 }$ with $\\chi _ { 1 } , y _ { 2 } \\sim \\mathcal { N } ( 0 , s * \\sigma ^ { 2 } ) , \\mathcal { N } ( 0 , 1 )$ , with corresponding predictions $\\hat { y } _ { 1 } , \\hat { y } _ { 2 }$ . Additionally, $s _ { t + 1 } = 0$ iff $\\hat { y } _ { 2 } >$ .5. We use Mean Squared Error as the loss function, so the optimal predictor is $\\hat { y _ { 1 } } , \\hat { y _ { 2 } } = ( 0 , 0 )$ . However, predicting $\\hat { y } _ { 2 } > . 5$ reduces the variance of $\\hat { y _ { 1 } }$ , i.e. reduces future loss. The baseline/IL predictor learns $\\hat { y _ { 1 } } , \\hat { y _ { 2 } }$ as parameters using SGD with a learning rate of 0.001. For experiments with meta-learning, PBT is the OL (with default settings, see Section 2.2), used to tune the learning rate, with negative loss on the final time-step of the interval as the performance measure for PBT. ",
+        "bbox": [
+            173,
+            540,
+            825,
+            651
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "text",
+        "text": "Unit test 2: Myopic RL. This unit test is based on a modification of the prisoner’s dilemma (Prisner, 2014) where an agent plays each round against its past self. The reward function is presented in Table 1. An agent in this environment has a long-term, non-myopic, incentive for cooperation (with its future self), but a current-time-step, myopic, incentive for defection (from its future self). ",
+        "bbox": [
+            174,
+            662,
+            825,
+            719
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "text",
+        "text": "The unit test evaluates whether a agent reveals the non-myopic incentive even when the agent is meant to optimize for the present reward only (i.e. uses discount rate $\\gamma = 0$ ). Naively, we’d expect the nonmyopic incentive to be hidden from the agent in this case, and for the agent to consistently defect; learning algorithms that do so pass the test. But some learning algorithms also fail the unit test, revealing the incentive for the agent to cooperate with its future self. While aiming for myopic behavior ",
+        "bbox": [
+            174,
+            719,
+            504,
+            857
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "text",
+        "text": "Table 1: Rewards for the RL unit test. Note that the myopic (defect (D)) action always increases reward at the current time-step, but decreases reward at the next time-step - the incentive to (cooperate (C)) with one’s future self is hidden from the point of view of a myopic learner. ",
+        "bbox": [
+            516,
+            728,
+            825,
+            811
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "table",
+        "img_path": "images/370e974762175107c7aed2e60f9f736ca0b327ddca422ff17241cb1b39d1c0b0.jpg",
+        "table_caption": [],
+        "table_footnote": [],
+        "table_body": "<table><tr><td rowspan=1 colspan=1></td><td rowspan=1 colspan=1>at=D</td><td rowspan=1 colspan=1>at=C</td></tr><tr><td rowspan=1 colspan=1>St=at-1=D</td><td rowspan=1 colspan=1>-1/2</td><td rowspan=1 colspan=1>-1</td></tr><tr><td rowspan=1 colspan=1>St=at-1=C</td><td rowspan=1 colspan=1>1/2</td><td rowspan=1 colspan=1>0</td></tr></table>",
+        "bbox": [
+            542,
+            813,
+            797,
+            857
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "text",
+        "text": "may seem odd, myopic learners have no incentives to cause distributional shift, since it can only improve future performance. And while making learners myopic may seem like a ’brute-force’ guaranteed way to manage HI-ADS, we show it is in fact non-trivial to implement. See Appendix 9.1 for details and experiments varying the reward structure. ",
+        "bbox": [
+            174,
+            858,
+            826,
+            912
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "text",
+        "text": "5.1.1 HI-ADS UNIT TESTS EXPERIMENTAL RESULTS AND DISCUSSION ",
+        "text_level": 1,
+        "bbox": [
+            173,
+            103,
+            679,
+            117
+        ],
+        "page_idx": 5
+    },
+    {
+        "type": "text",
+        "text": "We first show that agents trained with PBT fail the unit tests more often than “vanilla” algorithms which do not use meta-learning. We initialize the learning rate log-uniformly between 0.01 and 1.0 for all experiments (whether using PBT or not). We expect and confirm that the following two factors lead to higher rates of unit test failure: (1) Shorter intervals: These give the OL more opportunities to influence the population. (2) Larger populations: These make outliers with exceptional non-myopic performance more likely, and OL makes them likely to survive and propagate. ",
+        "bbox": [
+            173,
+            109,
+            825,
+            202
+        ],
+        "page_idx": 5
+    },
+    {
+        "type": "text",
+        "text": "The baseline (no meta-learning) algorithms all pass the unit tests: hidden incentives are almost never revealed - see blue curves in Fig. 5. However, agents trained with meta-learning and large populations often fail the unit tests: see orange curves in top rows of Fig. 5. ",
+        "bbox": [
+            174,
+            208,
+            825,
+            251
+        ],
+        "page_idx": 5
+    },
+    {
+        "type": "text",
+        "text": "Furthermore, we verify that context swapping significantly mitigates the effect of HI-ADS in both unit tests, decreasing undesirable behaviour to near-baseline levels - see bottom rows of Fig. 5. This effect can be explained as follows: Because context swapping transfers the benefits of one learner’s action to the next learner to inhabit that environment, it increases the second learner’s fitness, and thereby reduces the relative fitness (as evaluated by PBT’s EXPLOIT step) of the non-myopic cooperate behaviour. We observe some interesting exceptions with the combination of small populations and short PBT intervals: Although context swapping still significantly decreases the effect of HIADS, non-myopic cooperate behaviour is observed as much as $20 \\%$ of the time (for #learner $_ { = 1 0 }$ , $T = 1$ ; see bottom-left plot). ",
+        "bbox": [
+            174,
+            258,
+            452,
+            535
+        ],
+        "page_idx": 5
+    },
+    {
+        "type": "text",
+        "text": "We also observe that PBT reveals HI-ADS even when $T = 1$ , where the explanation that PBT operates on a longer time horizon than the inner loop does not apply. We provide a detailed explanation for how this might happen in Appendix 9.1.2, but in summary, we hypothesize that there are at least 2 mechanisms by which PBT is revealing HI-ADS: (1) optimizing over a longer time-scale, and (2) picking up on the correlation between an agent’s current policy and the underlying state. Mechanism (2) can be explained informally as reasoning as: “If I’m cooperating, then I was probably cooperating on the last time-step as well, so my reward should be higher”. As support for these hypotheses, we run control experiments identifying two algorithms (each sharing only one of these properties) that can fail the unit test. Context swapping remains effective. ",
+        "bbox": [
+            174,
+            541,
+            452,
+            833
+        ],
+        "page_idx": 5
+    },
+    {
+        "type": "text",
+        "text": "(1) Optimizing over a longer time-scale: replacing PBT with REINFORCE as an ",
+        "text_level": 1,
+        "bbox": [
+            173,
+            840,
+            452,
+            868
+        ],
+        "page_idx": 5
+    },
+    {
+        "type": "image",
+        "img_path": "images/fb963b752d78b641eafae0011c4187a77f39ee589e94374f2b91ae0ecc03c850.jpg",
+        "image_caption": [
+            "Figure 5: (A) Values of $\\hat { y _ { 2 } }$ in the supervised learning (SL) unit test. Larger values mean sacrificing present performance for future performance (i.e. non-myopic exploitation of ADS). (B) Average level of non-myopic cooperate behavior observed in the RL unit test for HI-ADS, with two meta-learning algorithms (B1) PBT and (B2) REINFORCE. Lower is better, since the goal is for non-myopic incentives to remain hidden. Despite the inner loop being fully myopic (simple MLP in the SL test, $\\gamma = 0$ in RL test), in all cases outer-loop (OL) optimizers reveal HI-ADS (top rows). Context swapping significantly mitigates HI-ADS (bottom rows). "
+        ],
+        "image_footnote": [],
+        "bbox": [
+            468,
+            261,
+            820,
+            670
+        ],
+        "page_idx": 5
+    },
+    {
+        "type": "text",
+        "text": "outer-loop optimizer. The outer-loop optimizes the parameters to maximize the summed reward of the last $T$ time-steps. As with PBT, we observe non-myopic behavior, but now only when $T > 1$ This supports our hypothesis that exploitation of HI-ADS is due not to PBT in particular, but just to the introduction of sufficiently powerful meta-learning. See Fig. 5 B2. ",
+        "bbox": [
+            174,
+            868,
+            825,
+            924
+        ],
+        "page_idx": 5
+    },
+    {
+        "type": "text",
+        "text": "(2) Exploiting correlation: Q-learning with $\\gamma = 0$ an $\\epsilon = 0 . 1$ -greedy behavior policy and no meta-learning. If either state was equally likely, the $\\mathbf { Q }$ -values would be the average of the values in each column in Table 1, so the estimated $Q ( { \\mathrm { d e f e c t } } )$ would be larger. But the $\\epsilon$ -greedy policy correlates the previous action (i.e. the current state) and current action (so long as the policy did not just change), so the top-left and bottomright entries carry more weight in the estimates, sometimes causing $Q ( { \\mathrm { d e f e c t } } ) \\approx Q ($ (cooperate) and persistent nonmyopic behavior. See Fig. 6 for results, Appendix 9.1.4 for more results, and Appendix 9.1.3 for experimental details. ",
+        "bbox": [
+            174,
+            104,
+            485,
+            297
+        ],
+        "page_idx": 6
+    },
+    {
+        "type": "text",
+        "text": "5.2 CONTENT RECOMMENDATION ",
+        "text_level": 1,
+        "bbox": [
+            178,
+            315,
+            421,
+            329
+        ],
+        "page_idx": 6
+    },
+    {
+        "type": "image",
+        "img_path": "images/c27ce5d1176c26223b5a51441be7d7fafa717c25c484532ba211b635ec5f8ba5.jpg",
+        "image_caption": [
+            "Figure 6: Q-learning can fail the unit test, playing ${ \\sim } 8 0 { \\cdot } 9 0 \\%$ cooperate in 3 of 5 experiments (bottom row). Each column represents an independent experiment. Q-values for the cooperate and defect actions stay tightly coupled in the failure cases (col. 1,2,5), while in the cases passing the unit test (col. 3,4) the Q-value of cooperate decreases over time. "
+        ],
+        "image_footnote": [],
+        "bbox": [
+            496,
+            104,
+            821,
+            209
+        ],
+        "page_idx": 6
+    },
+    {
+        "type": "text",
+        "text": "We now present a toy environment for modeling content recommendation of news articles, which includes the potential for ADS by incorporating the mechanisms mentioned in Sec. 2.2, discussed as contributing factors to the problems of fake news and filter bubbles. Specifically, the environment assumes that presenting an article to a user can influence (1) their interest in similar articles, and (2) their propensity to use the recommendation service. These correspond to modeling auto-induced concept shift of users, and auto-induced covariate shift of the user base, respectively (see Sec. 2.2). ",
+        "bbox": [
+            174,
+            340,
+            825,
+            424
+        ],
+        "page_idx": 6
+    },
+    {
+        "type": "text",
+        "text": "This environment includes the following components, which change over (discrete) time: User type: $x ^ { t }$ , Article type: $y ^ { t }$ , User interests: $\\bar { \\mathbf { W } } ^ { t }$ (propensity for users of each type to click on articles of each type), and User loyalty: $\\mathbf { g } ^ { t }$ (propensity for users of each type to use the platform). At each time step $t$ , a user $x ^ { t }$ is sampled from a categorical distribution, based on the loyalty of the different user types. The recommendation system (a classifier) selects which type of article to present in the top position, and finally the user ‘clicks’ an article $y ^ { t }$ , according to their interests. User loyalty for user type $x ^ { t }$ undergoes covariate shift: in accordance with the self-selection effect, $g ^ { t }$ increases or decreases proportionally to that user type’s interest in the top article. The interests of user type $x ^ { t }$ (represented by a column of $\\mathbf { W } ^ { t }$ ) undergoing concept shift; in accordance with the illusory truth effect, interest in the topic of the top article chosen by the recommender system always increases. ",
+        "bbox": [
+            173,
+            431,
+            825,
+            570
+        ],
+        "page_idx": 6
+    },
+    {
+        "type": "text",
+        "text": "5.2.1 CONTENT RECOMMENDATION EXPERIMENTAL RESULTS AND DISCUSSION ",
+        "text_level": 1,
+        "bbox": [
+            176,
+            587,
+            736,
+            599
+        ],
+        "page_idx": 6
+    },
+    {
+        "type": "text",
+        "text": "We run 20 trials using an MLP trained with SGD for the recommender system. We find that PBT yields significant improvements in training time and accuracy, but also greater distributional shift (Fig. 7). User base and user interests both change faster with PBT, and user interests change more overall. We measure concept/covariate shift using the cosine distance and KL-divergence, respectively. We observe that the distributions over user types typically saturate (to a single user type) after a few hundred time-steps (Fig 1 and Fig. 7, Right). We run long enough to reach such states, to demonstrate that the increase in ADS from PBT is not transitory. The environment has a number of free parameters, and our results are qualitatively consistent so long as the covariate shift rate $( \\alpha _ { 1 } )$ is faster than the concept shift rate $\\left( \\alpha _ { 2 } \\right)$ . See Appendix 9.2.1 for details. ",
+        "bbox": [
+            173,
+            609,
+            826,
+            736
+        ],
+        "page_idx": 6
+    },
+    {
+        "type": "image",
+        "img_path": "images/95efffd0c2948471a88671bd2e264c0a9ccf142775dc7e578a618fd609eb6f42.jpg",
+        "image_caption": [
+            "Figure 7: Content recommendation experiments. Left: using Population Based Training (PBT) increases accuracy of predictions faster, leads to a faster and larger drift in users’ interests, $P ( \\boldsymbol { y } | \\mathbf { x } )$ , (Center); as well as the distribution of users, $P ( \\mathbf { x } )$ , (Right). Shading shows std error over 20 runs. "
+        ],
+        "image_footnote": [],
+        "bbox": [
+            308,
+            747,
+            687,
+            859
+        ],
+        "page_idx": 6
+    },
+    {
+        "type": "text",
+        "text": "6 RELATED WORK ",
+        "text_level": 1,
+        "bbox": [
+            176,
+            102,
+            341,
+            117
+        ],
+        "page_idx": 7
+    },
+    {
+        "type": "text",
+        "text": "ADS in practice: We introduce the term ADS, but we are far from the first to study it. Caruana et al. (2015) provide an example of asthmatic patients having lower predicted risk of pneumonia. Treating asthmatics with pneumonia less aggressively on this basis would be an example of harmful ADS; the reason they had lower pneumonia risk was because they had received more aggressive care already. Schulam & Saria (2017) note that such predictive models are commonly used to inform decision-making, and propose modeling counterfactuals (e.g. “how would this patient fare with less aggressive treatment”) to avoid making such self-refuting predictions. While their goal is to make accurate predictions in the presence of ADS, our goal is to identify and manage incentives for ADS. ",
+        "bbox": [
+            174,
+            125,
+            825,
+            236
+        ],
+        "page_idx": 7
+    },
+    {
+        "type": "text",
+        "text": "Non-i.i.d bandits: Contextual bandits (Wang et al., 2005) are a common approach to content recommendation (Li et al., 2010). While bandit algorithms typically make the i.i.d. assumption, counter-examples exist (Gheshlaghi Azar et al., 2014; Auer et al., 1995). Closest to our work is Shah et al. (2018), who consider covariate shift caused by recommender systems’ recommendations. But while they seek to exploit ADS, our aim is to avoid hidden incentives for exploiting ADS. ",
+        "bbox": [
+            174,
+            243,
+            825,
+            313
+        ],
+        "page_idx": 7
+    },
+    {
+        "type": "text",
+        "text": "Safety and incentives: Emergent incentives to influence the world (such as HI-ADS) are at the heart of many concerns about the safety of advanced AI systems (Omohundro, 2008; Bostrom, 2014). Understanding and managing the incentives of learners is also a focus of Armstrong & O’Rourke (2017); Everitt (2018); Everitt et al. (2019); Cohen et al.. While Everitt et al. (2019) focus on identifying which incentives are present, we note that incentives may be present and yet not be revealed or pursued - for example, in supervised learning, there is an incentive to make predictions that are over-fit to the test set, but we typically hide the test set from the learner, which effectively hides this incentive. While Carey et al. (2020); Everitt et al. (2019); Armstrong & O’Rourke (2017) discuss methods of removing problematic incentives, we note in practice incentives are often hidden rather than removed. Our work addresses the efficacy of this approach and ways in which it can fail. ",
+        "bbox": [
+            174,
+            319,
+            825,
+            459
+        ],
+        "page_idx": 7
+    },
+    {
+        "type": "text",
+        "text": "HI-ADS and meta-learning: We believe our work is the first to consider HI-ADS and their relation to meta-learning. A few previous works have some relevance. Rabinowitz (2019) documents qualitative differences in learning behavior when meta-learning is applied. MacKay et al. (2019) and Lorraine & Duvenaud (2018) view meta-learning as a bilevel optimization problem, with the inner loop playing a best-response to the outer loop. In our work, the outer loop has a greater influence, and the inner loop often fails to play best-response. Sutton et al. (2007) noted that meta-learning can improve performance by preventing convergence of the inner loop to best response. ",
+        "bbox": [
+            174,
+            465,
+            825,
+            564
+        ],
+        "page_idx": 7
+    },
+    {
+        "type": "text",
+        "text": "7 DISCUSSION AND CONCLUSION ",
+        "text_level": 1,
+        "bbox": [
+            178,
+            580,
+            468,
+            597
+        ],
+        "page_idx": 7
+    },
+    {
+        "type": "text",
+        "text": "We identify the phenomenon of auto-induced distributional shift (ADS) and problems that can arise when there are hidden incentives for learners to induce distributional shift (HI-ADS). We show that meta-learning can reveal HI-ADS and lead learners to use ADS as a means of increasing performance. ",
+        "bbox": [
+            176,
+            603,
+            823,
+            646
+        ],
+        "page_idx": 7
+    },
+    {
+        "type": "text",
+        "text": "Our work highlights the interdisciplinary nature of issues with real-world deployment of ML systems - we show how HI-ADS could play a role in important technosocial issues like filter bubbles and the propagation of fake news. There are a number of potential implications for our work: (1) When HI-ADS are a concern, our methodology and environments can be used to help diagnose whether and to what extent the final performance/behavior of a learner is due to ADS and/or incentives for ADS, i.e. to quantify their influence on that learner. (2) Comparing this quantitative analysis for different algorithms could help us understand which features of algorithms affect their propensity to reveal HI-ADS, and aid in the development of safer and more robust algorithms. (3) Characterizing and identifying HI-ADS in these tests is a first step to analyzing and mitigating other (problematic) hidden incentives, as well as to developing theoretical understanding of hidden incentives. ",
+        "bbox": [
+            173,
+            652,
+            825,
+            791
+        ],
+        "page_idx": 7
+    },
+    {
+        "type": "text",
+        "text": "Broadly speaking, our work emphasizes that the choice of machine learning algorithm plays an important role in specification, independently of the choice of performance metric. A learner can use ADS to increase performance according to the intended performance metric, and yet still behave in an undesirable way, if we did not intend the learner to improve performance by that method. In other words, performance metrics are typically incomplete specifications: they only specify our goals or ends, while our choice of learning algorithm plays a role in specifying the means by which we intend an learner to achieve those ends. With increasing deployment of ML algorithms in daily life, we believe that (1) understanding incentives and (2) specifying desired/allowed means of improving performance are important avenues of future work to ensure fair, robust, and safe outcomes. ",
+        "bbox": [
+            174,
+            797,
+            825,
+            924
+        ],
+        "page_idx": 7
+    },
+    {
+        "type": "text",
+        "text": "REFERENCES ",
+        "text_level": 1,
+        "bbox": [
+            174,
+            102,
+            287,
+            118
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "Hunt Allcott and Matthew Gentzkow. Social media and fake news in the 2016 election. Journal of Economic Perspectives, 31(2):211–36, May 2017. ",
+        "bbox": [
+            174,
+            126,
+            825,
+            155
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "Michelle A. Amazeen and Bartosz W. Wojdynski. Reducing native advertising deception: Revisiting the antecedents and consequences of persuasion knowledge in digital news contexts. Mass Communication and Society, 0(0):1–26, 2018. ",
+        "bbox": [
+            174,
+            162,
+            823,
+            207
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "Stuart Armstrong and Xavier O’Rorke. Good and safe uses of ai oracles. arXiv preprint arXiv:1711.05541, 2017. ",
+        "bbox": [
+            171,
+            215,
+            825,
+            244
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "Stuart Armstrong and Xavier O’Rourke. Indifference methods for managing agent rewards. Technical report, Future of Humanity Institute, 2017. ",
+        "bbox": [
+            171,
+            253,
+            825,
+            282
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "K. J. Åström. Optimal control of Markov Processes with incomplete state information. Journal of Mathematical Analysis and Applications, 10:174–205, January 1965. ",
+        "bbox": [
+            171,
+            292,
+            826,
+            321
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "P. Auer, N. Cesa-Bianchi, Y. Freund, and R. E. Schapire. Gambling in a rigged casino: The adversarial multi-armed bandit problem. In Foundations of Computer Science, 1995. ",
+        "bbox": [
+            176,
+            330,
+            823,
+            361
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "Eytan Bakshy, Solomon Messing, and Lada A. Adamic. Exposure to ideologically diverse news and opinion on Facebook. Science, 348(6239):1130–1132, 2015. ISSN 0036-8075. doi: 10.1126/science.aaa1160. URL http://science.sciencemag.org/content/348/ 6239/1130. ",
+        "bbox": [
+            173,
+            368,
+            826,
+            426
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "Nick Bostrom. Superintelligence: Paths, Dangers, Strategies. Oxford University Press, Inc., New York, NY, USA, 1st edition, 2014. ",
+        "bbox": [
+            173,
+            434,
+            823,
+            464
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "Ryan Carey, Eric Langlois, Tom Everitt, and Shane Legg. The incentives that shape behaviour. arXiv preprint arXiv:2001.07118, 2020. ",
+        "bbox": [
+            173,
+            473,
+            823,
+            502
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "Rich Caruana, Yin Lou, Johannes Gehrke, Paul Koch, Marc Sturm, and Noemie Elhadad. Intelligible models for healthcare: Predicting pneumonia risk and hospital 30-day readmission. In International Conference on Knowledge Discovery and Data Mining, pp. 1721–1730, 2015. ",
+        "bbox": [
+            173,
+            510,
+            825,
+            554
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "Paul Christiano, Jan Leike, Tom B. Brown, Miljan Martic, Shane Legg, and Dario Amodei. Deep reinforcement learning from human preferences, 2017. ",
+        "bbox": [
+            173,
+            563,
+            823,
+            593
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "Michael K Cohen, Badri N Vellambi, and Marcus Hutter. Asymptotically unambitious artificial general intelligence. ",
+        "bbox": [
+            171,
+            601,
+            825,
+            631
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "Dominic DiFranzo and Kristine Gloria-Garcia. Filter bubbles and fake news. XRDS, 23(3):32–35, April 2017. ISSN 1528-4972. doi: 10.1145/3055153. URL http://doi.acm.org/10. 1145/3055153. ",
+        "bbox": [
+            174,
+            638,
+            826,
+            683
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "Mostafa M. El-Bermawy. Your echo chamber is destroying democracy, 2016. URL https: //www.wired.com/2016/11/filter-bubble-destroying-democracy/. ",
+        "bbox": [
+            171,
+            690,
+            825,
+            720
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "Tom Everitt. Towards Safe Artificial General Intelligence. PhD thesis, Australian National University, 2018. ",
+        "bbox": [
+            173,
+            728,
+            825,
+            758
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "Tom Everitt and Marcus Hutter. The alignment problem for bayesian history-based reinforcement learners. 2018. ",
+        "bbox": [
+            173,
+            766,
+            825,
+            796
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "Tom Everitt and Marcus Hutter. Reward tampering problems and solutions in reinforcement learning: A causal influence diagram perspective. arXiv preprint arXiv:1908.04734, 2019. ",
+        "bbox": [
+            171,
+            805,
+            823,
+            835
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "Tom Everitt, Pedro A. Ortega, Elizabeth Barnes, and Shane Legg. Understanding agent incentives using causal influence diagrams. part i: Single action settings, 2019. ",
+        "bbox": [
+            171,
+            843,
+            823,
+            872
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "Lisa K. Fazio, Nadia M. Brashier, B. Keith Payne, and Elizabeth J. Marsh. Knowledge does not protect against illusory truth. Journal of Experimental Psychology: General, 144(5):993–1002, 10 2015. ISSN 0096-3445. doi: 10.1037/xge0000098. ",
+        "bbox": [
+            174,
+            881,
+            825,
+            924
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "Chelsea Finn, Pieter Abbeel, and Sergey Levine. Model-agnostic meta-learning for fast adaptation of deep networks. In International Conference on Machine Learning, 2017. ",
+        "bbox": [
+            171,
+            103,
+            825,
+            133
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "Seth Flaxman and Sharad Goel. Filter bubbles, echo chambers, and online news consumption. Public Opinion Quarterly, 2015. ",
+        "bbox": [
+            173,
+            141,
+            823,
+            170
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "Mohammad Gheshlaghi Azar, Alessandro Lazaric, and Emma Brunskill. Online stochastic optimization under correlated bandit feedback. ArXiv preprint, 2014. ",
+        "bbox": [
+            173,
+            179,
+            825,
+            208
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "Dong Gong, Zhen Zhang, Qinfeng Shi, Anton van den Hengel, Chunhua Shen, and Yanning Zhang. Learning an optimizer for image deconvolution. ArXiv preprint, 2018. ",
+        "bbox": [
+            171,
+            217,
+            825,
+            246
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "Jacob Groshek and Karolina Koc-Michalska. Helping populism win? Social media use, filter bubbles, and support for populist presidential candidates in the 2016 us election campaign. Information, Communication & Society, 20(9):1389–1407, 2017. doi: 10.1080/1369118X.2017.1329334. URL https://doi.org/10.1080/1369118X.2017.1329334. ",
+        "bbox": [
+            173,
+            255,
+            826,
+            311
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "Dylan Hadfield-Menell, Smitha Milli, Pieter Abbeel, Stuart Russell, and Anca Dragan. Inverse reward design. In Neural Information Processing Systems, 2017. ",
+        "bbox": [
+            173,
+            319,
+            821,
+            349
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "M. Jaderberg, V. Dalibard, S. Osindero, W. M. Czarnecki, J. Donahue, A. Razavi, O. Vinyals, T. Green, I. Dunning, K. Simonyan, C. Fernando, and K. Kavukcuoglu. Population Based Training of Neural Networks. ArXiv preprint, 2017. ",
+        "bbox": [
+            176,
+            357,
+            825,
+            401
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "Leslie Pack Kaelbling, Michael L. Littman, and Anthony R. Cassandra. Planning and acting in partially observable stochastic domains. Artificial Intelligence, 101(1-2):99–134, May 1998. ISSN 0004-3702. ",
+        "bbox": [
+            173,
+            409,
+            826,
+            452
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "A. Kalousis. Model selection via meta-learning: A comparative study. In IEEE International Conference on Tools with Artificial Intelligence, 2000. ",
+        "bbox": [
+            173,
+            462,
+            821,
+            491
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "Varol Kayhan. Confirmation bias: Roles of search engines and search contexts. In International Conference on Information Systems, 2015. ",
+        "bbox": [
+            174,
+            500,
+            823,
+            529
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "Ed. Lee Howell. Digital wildfires in a hyperconnected world. In Lee Howell (ed.), Global Risks 2013. World Economic Forum, 2013. URL http://reports.weforum.org/global-risks-2013/risk-case-1/ digital-wildfires-in-a-hyperconnected-world/. ",
+        "bbox": [
+            174,
+            536,
+            825,
+            594
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "Jan Leike, Miljan Martic, Victoria Krakovna, Pedro A. Ortega, Tom Everitt, Andrew Lefrancq, Laurent Orseau, and Shane Legg. AI safety gridworlds. Technical report, DeepMind, 2017. ",
+        "bbox": [
+            168,
+            602,
+            825,
+            632
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "Jan Leike, David Krueger, Tom Everitt, Miljan Martic, Vishal Maini, and Shane Legg. Scalable agent alignment via reward modeling: a research direction. Technical report, DeepMind Safety Research, 2018. ",
+        "bbox": [
+            173,
+            640,
+            825,
+            683
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "Lihong Li, Wei Chu, John Langford, and Robert E. Schapire. A contextual-bandit approach to personalized news article recommendation. In International Conference on World Wide Web, 2010. ",
+        "bbox": [
+            171,
+            691,
+            825,
+            722
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "Jonathan Lorraine and David Duvenaud. Stochastic hyperparameter optimization through hypernetworks. ArXiv preprint, 2018. ",
+        "bbox": [
+            171,
+            729,
+            825,
+            758
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "D. D. Luxton, J. D. June, and J. M. Fairall. Social media and suicide: A public health perspective. American journal of public health, 102(2):195–200, 2012. ",
+        "bbox": [
+            169,
+            767,
+            825,
+            797
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "Matthew MacKay, Paul Vicol, Jonathan Lorraine, David Duvenaud, and Roger Grosse. Self-tuning networks: Bilevel optimization of hyperparameters using structured best-response functions. ArXiv preprint, 2019. ",
+        "bbox": [
+            174,
+            805,
+            823,
+            848
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "Merriam-Webster. The real story of fake news, 2017. URL https://www.merriam-webster. com/words-at-play/the-real-story-of-fake-news. ",
+        "bbox": [
+            173,
+            857,
+            821,
+            887
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "Luke Metz, Niru Maheswaranathan, Brian Cheung, and Jascha Sohl-Dickstein. Learning unsupervised learning rules. In International Conference on Learning Representations, 2019. ",
+        "bbox": [
+            173,
+            895,
+            820,
+            924
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "Paul Mihailidis and Samantha Viotty. Spreadable spectacle in digital culture: Civic expression, fake news, and the role of media literacies in \"post-fact\" society. American Behavioural Scientist, 2017. ",
+        "bbox": [
+            171,
+            103,
+            825,
+            132
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Jose G. Moreno-Torres, Troy Raeder, RocíO Alaiz-RodríGuez, Nitesh V. Chawla, and Francisco Herrera. A unifying view on dataset shift in classification. Pattern Recognition, 45(1):521–530, January 2012. ",
+        "bbox": [
+            173,
+            140,
+            825,
+            183
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Tien T. Nguyen, Pik-Mai Hui, F. Maxwell Harper, Loren Terveen, and Joseph A. Konstan. Exploring the filter bubble: The effect of using recommender systems on content diversity. In Proceedings of the 23rd International Conference on World Wide Web, WWW ’14, pp. 677–686, New York, NY, USA, 2014. ACM. ISBN 978-1-4503-2744-2. doi: 10.1145/2566486.2568012. URL http://doi.acm.org/10.1145/2566486.2568012. ",
+        "bbox": [
+            173,
+            191,
+            825,
+            262
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Safiya Umoja Noble. Algorithms of Oppression: How Search Engines Reinforce Racism. NYC Press, 2018. ",
+        "bbox": [
+            174,
+            270,
+            825,
+            299
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Stephen M. Omohundro. The basic AI drives. In Conference on Artificial General Intelligence, 2008. ",
+        "bbox": [
+            171,
+            306,
+            825,
+            323
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Pedro A. Ortega, Vishal Maini, et al. Building safe artificial intelligence: specification, robustness, and assurance, 2018. ",
+        "bbox": [
+            171,
+            329,
+            826,
+            359
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Eli Pariser. The Filter Bubble: What the Internet Is Hiding from You. The Penguin Group, 2011. ",
+        "bbox": [
+            176,
+            366,
+            805,
+            383
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Gordon Pennycook, Tyrone D Cannon, and David G. Rand. Prior exposure increases perceived accuracy of fake news. Journal of Experimental Psychology (forthcoming), 2019. ",
+        "bbox": [
+            173,
+            388,
+            825,
+            420
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Erich Prisner. Game Theory Through Examples. Mathematical Association of America, 2014. ",
+        "bbox": [
+            176,
+            426,
+            792,
+            443
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Joaquin Quionero-Candela, Masashi Sugiyama, Anton Schwaighofer, and Neil D. Lawrence. Dataset Shift in Machine Learning. The MIT Press, 2009. ",
+        "bbox": [
+            171,
+            450,
+            825,
+            479
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Neil C. Rabinowitz. Meta-learners’ learning dynamics are unlike learners’. ArXiv preprint, 2019. ",
+        "bbox": [
+            173,
+            487,
+            810,
+            503
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "David Robson. The myth of the online echo chamber, 2018. URL http://www.bbc.com/ future/story/20180416-the-myth-of-the-online-echo-chamber. ",
+        "bbox": [
+            173,
+            511,
+            821,
+            540
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Peter Schulam and Suchi Saria. Reliable decision support using counterfactual models. In Neural Information Processing Systems, 2017. ",
+        "bbox": [
+            173,
+            547,
+            825,
+            577
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Virag Shah, Jose Blanchet, and Ramesh Johari. Bandit learning with positive externalities. Neural Information Processing Systems, 2018. ",
+        "bbox": [
+            173,
+            584,
+            825,
+            614
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Chengcheng Shao, Giovanni Luca Ciampaglia, Onur Varol, Kai-Cheng Yang, Alessandro Flammini, and Filippo Menczer. The spread of low-credibility content by social bots. Nature Communications, 9(4787), 2018. ",
+        "bbox": [
+            173,
+            621,
+            826,
+            665
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Jasper Snoek, Hugo Larochelle, and Ryan P Adams. Practical Bayesian optimization of machine learning algorithms. In Neural Information Processing Systems, 2012. ",
+        "bbox": [
+            173,
+            672,
+            823,
+            702
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Richard S Sutton and Andrew G Barto. Introduction to Reinforcement Learning. MIT Press, 1998. ",
+        "bbox": [
+            171,
+            709,
+            820,
+            726
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Richard S Sutton, Anna Koop, and David Silver. On the role of tracking in stationary environments. In International conference on Machine learning, 2007. ",
+        "bbox": [
+            173,
+            733,
+            823,
+            762
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Techopedia. Filter bubble, 2018. URL https://www.techopedia.com/definition/ 28556/filter-bubble. ",
+        "bbox": [
+            173,
+            770,
+            823,
+            799
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Chih-Chun Wang, Sanjeev R Kulkarni, and H Vincent Poor. Bandit problems with side observations. IEEE Transactions on Automatic Control, 50(3):338–355, 2005. ",
+        "bbox": [
+            173,
+            806,
+            823,
+            837
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Wikipedia contributors. Confirmation bias — Wikipedia, the free encyclopedia, 2018. URL https://en.wikipedia.org/w/index.php?title=Confirmation_bias& oldid=875026726. [Online; accessed 20-January-2019]. ",
+        "bbox": [
+            174,
+            844,
+            825,
+            887
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Ronald J. Williams. Simple statistical gradient-following algorithms for connectionist reinforcement learning. In Machine Learning, pp. 229–256, 1992. ",
+        "bbox": [
+            173,
+            895,
+            823,
+            924
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "APPENDICES ",
+        "text_level": 1,
+        "bbox": [
+            446,
+            103,
+            552,
+            117
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "8 CONTENT RECOMMENDATION IN THE WILD ",
+        "text_level": 1,
+        "bbox": [
+            174,
+            140,
+            565,
+            156
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "Filter bubbles, the spread of fake news, and other techno-social issues are widely reported to be responsible for the rise of populism (Groshek & Koc-Michalska, 2017), increase in racism and prejudice against immigrants and refugees (Noble, 2018), increase in social isolation and suicide (Luxton et al., 2012), and, particularly with reference to the 2016 US elections, are decried as threatening the foundations of democracy (El-Bermawy, 2016). Even in 2013, well before the 2016 American elections, a World Economic Forum report identified these problems as a global crisis (Lee Howell, 2013). ",
+        "bbox": [
+            174,
+            175,
+            825,
+            273
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "We focus on two related issues in which content recommendation algorithms play a role: fake news and filter bubbles. ",
+        "bbox": [
+            173,
+            280,
+            823,
+            308
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "8.1 FAKE NEWS ",
+        "text_level": 1,
+        "bbox": [
+            174,
+            333,
+            297,
+            347
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "Fake news (also called false news or junk news) is an extreme version of yellow journalism, propaganda, or clickbait, in which media that is ostensibly providing information focuses on being eye-catching or appealing, at the expense of the quality of research and exposition of factual information. Fake news is distinguished by being specifically and deliberately created to spread falsehoods or misinformation (Merriam-Webster, 2017; Mihailidis & Viotty, 2017). ",
+        "bbox": [
+            174,
+            362,
+            825,
+            431
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "Why does fake news spread? It may at first seem the solution is simply to educate people about the truth, but research tells us the problem is more multifaceted and insidious, due to a combination of related biases and cognitive effects including confirmation bias (people are more likely to believe things that fit with their existing beliefs), priming (exposure to information unconsciously influences the processing of subsequent information, i.e. seeing something in a credible context makes things seem more credible) and the illusory truth effect (i.e. people are more likely to believe something simply if they are told it is true). ",
+        "bbox": [
+            174,
+            439,
+            825,
+            536
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "Allcott & Gentzkow (2017) track about 150 fake news stories during the 2016 US election, and find the average American adult saw 1-2 fake news stories, just over half believed the story was true, and likelihood of believing fake news increased with ideological segregation (polarization) of their social media. Shao et al. (2018) examine the role of social bots in spreading fake news by analyzing 14 million Twitter messages. They find that bots are far more likely than humans to spread misinformation, and that success of a fake news story (in terms of human retweets) was heavily dependent on whether bots had shared the story. ",
+        "bbox": [
+            174,
+            542,
+            825,
+            641
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "Pennycook et al. (2019) examine the role of the illusory truth effect in fake news. They find that even a single exposure to a news story makes people more likely to believe that it is true, and repeat viewings increase this likelihood. They find that this is not true for extremely implausible statements (e.g. “the world is a perfect cube”), but that “only a small degree of potential plausibility is sufficient for repetition to increase perceived accuracy” of the story. The situation is further complicated by peoples’ inability to distinguish promoted content from real news - Amazeen & Wojdynski (2018) find that fewer than 1/10 people were able to tell when content was an advertisement, even when it was explicitly labelled as such. Similarly, Fazio et al. (2015) find that repeated exposure to incorrect trivia make people more likely to believe it, even when they are later able to identify the trivia as incorrect. ",
+        "bbox": [
+            174,
+            647,
+            825,
+            786
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "8.2 FILTER BUBBLES ",
+        "text_level": 1,
+        "bbox": [
+            176,
+            810,
+            331,
+            825
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "Filter bubbles, a term coined and popularized by Pariser (2011) are created by positive or negative feedback loops which encourage users or groups of users towards increasing within-group similarity, while driving up between-group dissimilarity. The curation of this echo chamber is called selfselection (people are more likely to look for or select things that fit their existing preferences), and favours what Techopedia (2018) calls intellectual isolation. In the context of social and political opinions, this is often called the polarization effect (Wikipedia contributors, 2018). ",
+        "bbox": [
+            174,
+            839,
+            825,
+            924
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "Filter bubbles can be encouraged by algorithms in two main ways. The first is the most commonly described: simply by showing content that is similar to what a user has already searched for, search or recommender systems create a positive feedback loop of increasingly-similar content (Pariser, 2011; Kayhan, 2015). The second way is similar but opposite - if the predictions of an algorithm are good for a certain group of people, but bad for others, the algorithm can do better on its metrics by driving hard-to-predict users away. Then new users to the site will either be turned off entirely, or see an artificially homogenous community of like-minded peers, a phenomena Shah et al. (2018) call positive externalities. ",
+        "bbox": [
+            174,
+            103,
+            825,
+            214
+        ],
+        "page_idx": 12
+    },
+    {
+        "type": "text",
+        "text": "In a study of 50,000 US-based internet users, Flaxman & Goel (2015) find that two things increase with social media and search engine use: (1) exposure of an individual to opposing or different viewpoints, and (2) mean ideological distance between users. Many studies cite the first result as evidence of the benefits of internet and social media (Robson, 2018; Bakshy et al., 2015), but the correlation of exposure with ideological distances demonstrates that exposure is not enough, and might even be counterproductive. ",
+        "bbox": [
+            174,
+            222,
+            825,
+            305
+        ],
+        "page_idx": 12
+    },
+    {
+        "type": "text",
+        "text": "Facebook’s own study on filter bubbles results show that the impact of the news feed algorithm on filter bubble “size” (a measure of homogeneity of posts relative to a baseline) is almost as large as the impact of friend group composition (Bakshy et al., 2015). Kayhan (2015) specifically study the role of search engines in confirmation bias, and find that search context and the similarity of results in search engine results both reinforce existing biases and increase the likelihood of future biased searches. Nguyen et al. (2014) similarly study the effect of recommender systems on individual users’ content diversity, and find that the set of options recommended narrows over time. ",
+        "bbox": [
+            174,
+            313,
+            825,
+            410
+        ],
+        "page_idx": 12
+    },
+    {
+        "type": "text",
+        "text": "Filter bubbles create an ideal environment for the spread of fake news: they increase the likelihood of repeat viewings of similar content, and because of the illusory truth effect, that content is more likely to be believed and shared (Pennycook et al., 2019; DiFranzo & Gloria-Garcia, 2017; Pariser, 2011). We are not claiming that HI-ADS are entirely or even mostly responsible for these problems, but we do note that they can play a role that is worth addressing. ",
+        "bbox": [
+            174,
+            416,
+            825,
+            487
+        ],
+        "page_idx": 12
+    },
+    {
+        "type": "image",
+        "img_path": "images/5c7a8c62362379a4ec118b2d7cacc1bfea380d1d1281d25198b5cb5100eaef60.jpg",
+        "image_caption": [
+            "incentive-compatible $\\beta = 0 . 5$ ) ",
+            "Figure 8: Average level of non-myopic (i.e. cooperate) behavior learned by agents in the unit test for HI-ADS. Despite making the inner loop fully myopic $( \\gamma = 0 )$ ), population-based training (PBT) can cause HI-ADS, leading agents to choose the cooperate action (top row). context swapping successfully prevents this (bottom row). Columns (from left to right) show results for populations of 10, 100, and 1000 learners. In the legend, “interval” refers to the interval $( T )$ of PBT (see Sec. 2.2). Sufficiently large populations and short intervals are necessary for PBT to induce nonmyopic behavior. "
+        ],
+        "image_footnote": [],
+        "bbox": [
+            232,
+            145,
+            764,
+            779
+        ],
+        "page_idx": 13
+    },
+    {
+        "type": "text",
+        "text": "9 EXTRA EXPERIMENTS AND REPRODUCIBILITY DETAILS ",
+        "text_level": 1,
+        "bbox": [
+            173,
+            103,
+            665,
+            118
+        ],
+        "page_idx": 14
+    },
+    {
+        "type": "text",
+        "text": "9.1 HI-ADS UNIT TEST ",
+        "text_level": 1,
+        "bbox": [
+            174,
+            133,
+            352,
+            148
+        ],
+        "page_idx": 14
+    },
+    {
+        "type": "text",
+        "text": "We used REINFORCE (Williams, 1992) with discount factor $\\gamma = 0$ as the baseline/IL optimizer. PBT (with default settings, see Section 2.2) is used to tune the learning rate, with reward on the final time-step of the interval as the performance measure for PBT. ",
+        "bbox": [
+            174,
+            159,
+            826,
+            202
+        ],
+        "page_idx": 14
+    },
+    {
+        "type": "text",
+        "text": "Formally, the environment is not a $2 \\mathbf { x } 2$ game (as the original prisoner’s dilemma); it’s a partially observable Markov Decision Process (Åström, 1965; Kaelbling et al., 1998): ",
+        "bbox": [
+            173,
+            208,
+            823,
+            239
+        ],
+        "page_idx": 14
+    },
+    {
+        "type": "equation",
+        "img_path": "images/909ce0b5c22c4e5607063381a463b92ae72bf2dc7c3edecf48a8704f92fad06f.jpg",
+        "text": "$$\n\\begin{array} { r l r } & { } & { s _ { t } , o _ { t } = a _ { t - 1 } , \\{ \\} } \\\\ & { } & { a _ { t } \\in \\{ \\mathsf { d e f e c t } , ~ \\mathsf { c o o p e r a t e } \\} } \\\\ & { } & { P ( s _ { t } , a _ { t } ) = \\delta ( a _ { t } ) } \\\\ & { } & { R ( s _ { t } , a _ { t } ) = I ( s _ { t } = \\mathsf { c o o p e r a t e } ) + \\beta ~ I ( a _ { t } = \\mathsf { c o o p e r a t e } ) - 1 / 2 } \\end{array}\n$$",
+        "text_format": "latex",
+        "bbox": [
+            199,
+            243,
+            661,
+            304
+        ],
+        "page_idx": 14
+    },
+    {
+        "type": "text",
+        "text": "where $I$ is an indicator function, and $\\beta = - 1 / 2$ is a parameter controlling the alignment of incentives. The initial state is sampled as $s _ { 0 } \\sim U ( { \\tt d e f e c t }$ , cooperate). Policies are represented by a single real-valued parameter $\\theta$ (initialized as $\\theta \\sim \\mathcal { N } ( 0 , 1 ) \\big )$ ) passed through a sigmoid whose output represents $P ( a _ { t } = \\mathrm { \\bar { d e f e c t } } )$ . ",
+        "bbox": [
+            176,
+            308,
+            820,
+            364
+        ],
+        "page_idx": 14
+    },
+    {
+        "type": "text",
+        "text": "9.1.1 ALIGNMENT OF INCENTIVES EXPLORATION ",
+        "text_level": 1,
+        "bbox": [
+            176,
+            380,
+            527,
+            393
+        ],
+        "page_idx": 14
+    },
+    {
+        "type": "text",
+        "text": "This section presents an exploration of the parameter $\\beta$ , which controls the alignment of incentives in the HI-ADS unit tests (see Table 2). ",
+        "bbox": [
+            173,
+            404,
+            823,
+            433
+        ],
+        "page_idx": 14
+    },
+    {
+        "type": "text",
+        "text": "To clarify the interpretation of experiments, we distinguish between environments in which myopic (defect) vs. nonmyopic (cooperate) incentives are opposed, orthogonal, or compatible. Note that in this unit test myopic behaviour (defection) is what we want to see. ",
+        "bbox": [
+            176,
+            439,
+            823,
+            481
+        ],
+        "page_idx": 14
+    },
+    {
+        "type": "text",
+        "text": "1. Incentive-opposed: Optimal myopic behavior is incompatible with optimal nonmyopic behavior (classic prisoner’s dilemma; these experiments are in the main paper).   \n2. Incentive-orthogonal: Optimal myopic behavior may or may not be optimal nonmyopic behavior.   \n3. Incentive-compatible: Optimal myopic behavior is necessarily also optimal nonmyopic behavior. ",
+        "bbox": [
+            187,
+            492,
+            826,
+            587
+        ],
+        "page_idx": 14
+    },
+    {
+        "type": "text",
+        "text": "We focused on incentive-opposed environment $( \\beta = - 1 / 2 )$ ) in the main paper in order to demonstrate that HI-ADS can be powerful enough to change the behavior of the system in an undesirable way. Here we also explore incentive-compatible and incentive-orthogonal environments because they provide useful baselines, helping us distinguish a systematic bias towards nonmyopic behavior from other reasons (such as randomness or optimization issues) for behavior that does not follow a myopically optimal policy. ",
+        "bbox": [
+            173,
+            598,
+            825,
+            683
+        ],
+        "page_idx": 14
+    },
+    {
+        "type": "text",
+        "text": "9.1.2 WORKING THROUGH A DETAILED EXAMPLE FOR PBT WITH $T = 1$ ",
+        "text_level": 1,
+        "bbox": [
+            174,
+            698,
+            687,
+            713
+        ],
+        "page_idx": 14
+    },
+    {
+        "type": "text",
+        "text": "To help provide intuition on how (mechanistically) PBT could lead to persistent levels of cooperation, we walk through a simple example (with no inner loop). Consider PBT with $T = 1$ and a population of 5 deterministic agents $A _ { 1 } , . . . , A _ { 5 }$ playing cooperate and receiving reward of $r ( A _ { i } ) = 0$ . Now suppose $A _ { 1 }$ suddenly switches to play defect. Then $r ( A _ { 1 } ) = 1 / 2$ on the next time-step (while the other agents’ reward is still 0), and so PBT’s EXPLOIT step will copy $A _ { 1 }$ (without loss of generality to $A _ { 2 }$ ). On the following time-step, $r ( A _ { 2 } ) = 1 / 2$ , and $r ( A _ { 1 } ) = - 1 / 2$ , so PBT will clone $A _ { 2 }$ to $A _ { 1 }$ , and the cycle repeats. Similar reasoning applies for larger populations, and $T > 1$ . ",
+        "bbox": [
+            173,
+            722,
+            826,
+            820
+        ],
+        "page_idx": 14
+    },
+    {
+        "type": "text",
+        "text": "9.1.3 Q-LEARNING EXPERIMENT DETAILS ",
+        "text_level": 1,
+        "bbox": [
+            174,
+            871,
+            478,
+            886
+        ],
+        "page_idx": 14
+    },
+    {
+        "type": "text",
+        "text": "We show that, under certain conditions, Q-learning can learn to (primarily) cooperate, and thus fails the HI-ADS unit test. We estimate Q-values using the sample-average method, which is guaranteed to converge in the fully observed, tabular case (Sutton & Barto, 1998). The agent follows the $\\epsilon$ -greedy policy with $\\epsilon = 0 . 1$ . In order to achieve this result, we additionally start the agent off with one synthetic memory where both state and action are defect and therefor $R ( \\mathrm { d e f e c t } ) = - . 5$ , and we hard-code the starting state to be cooperate (which normally only happens $50 \\%$ of the time). Without this kind of an initialization, the agent always learns to defect. However, under these conditions, we find that 10/30 agents learned to play cooperate most of the time, with $Q$ (cooperate) and $Q ( { \\mathrm { d e f e c t } } )$ both hovering around $- 0 . 0 7$ , while others learn to always defect, with $Q ( { \\mathrm { c o o p e r a t e } } ) \\approx - 0 . 9 2$ and $Q ( { \\mathrm { d e f e c t } } ) \\approx - 0 . 4 5$ . context swapping, however, prevents majority-cooperate behavior from ever emerging, see Figure 11. ",
+        "bbox": [
+            173,
+            895,
+            823,
+            924
+        ],
+        "page_idx": 14
+    },
+    {
+        "type": "table",
+        "img_path": "images/1175f254f167db3c316ece81017afe9fac641b6f9b455be985267608b1241e01.jpg",
+        "table_caption": [
+            "Table 2: $\\beta$ controls the extent to which myopic and nonmyopic incentives are aligned. "
+        ],
+        "table_footnote": [],
+        "table_body": "<table><tr><td>β</td><td>Environment</td><td>Cooperating</td></tr><tr><td>&lt;0</td><td>incentive-opposed</td><td>yields less reward on the current time-step (myopically detrimental)</td></tr><tr><td>=0</td><td>incentive-orthogonal</td><td>does not affect the current reward (myopically indifferent)</td></tr><tr><td>v0</td><td>incentive-compatible</td><td>yields more reward on the current time-step (myopically beneficial)</td></tr></table>",
+        "bbox": [
+            176,
+            127,
+            833,
+            189
+        ],
+        "page_idx": 15
+    },
+    {
+        "type": "text",
+        "text": "",
+        "bbox": [
+            173,
+            213,
+            826,
+            339
+        ],
+        "page_idx": 15
+    },
+    {
+        "type": "text",
+        "text": "9.1.4 Q-LEARNING: FURTHER RESULTS ",
+        "text_level": 1,
+        "bbox": [
+            176,
+            354,
+            460,
+            369
+        ],
+        "page_idx": 15
+    },
+    {
+        "type": "text",
+        "text": "To give a more representative picture of how often Q-learning fails the unit test, we run a larger set of experiments with Q-learning, results are in Figure 10. It’s possible that the failure of Q-learning is not persistent, since we have not proved otherwise, but we did run much longer experiments and still observe persistent failure, see Figure 9. ",
+        "bbox": [
+            176,
+            378,
+            825,
+            435
+        ],
+        "page_idx": 15
+    },
+    {
+        "type": "image",
+        "img_path": "images/93bc9858be4e122e4aa54709b97526f0b712f8b3de8f38002ec30b69c8b8d7ee.jpg",
+        "image_caption": [
+            "Figure 9: The same experiments as Figures 6, 10, run for 50,000 time-steps instead of 3000, to illustrate the persistence of non-myopic behavior. "
+        ],
+        "image_footnote": [],
+        "bbox": [
+            261,
+            446,
+            741,
+            590
+        ],
+        "page_idx": 15
+    },
+    {
+        "type": "image",
+        "img_path": "images/3e860d515c7ba817cb56c44c846a248ed6df5909c2c2c53c73553c58f29bb252.jpg",
+        "image_caption": [
+            "Figure 10: More independent experiments with Q-learning, exactly following Figure 6. Q-learning fails the unit test in a total of 10/30 experiments (including those from Figure 6). "
+        ],
+        "image_footnote": [],
+        "bbox": [
+            263,
+            131,
+            730,
+            848
+        ],
+        "page_idx": 16
+    },
+    {
+        "type": "image",
+        "img_path": "images/18011e0877d74bf4ea8b3877bc9a66d2f427d421200bb03d1cb2dd7167b3d022.jpg",
+        "image_caption": [
+            "Figure 11: More independent experiments with Q-learning, exactly following Figure 6, except also using context swapping. This leads to a $100 \\%$ success rate on the unit test. "
+        ],
+        "image_footnote": [],
+        "bbox": [
+            259,
+            114,
+            730,
+            863
+        ],
+        "page_idx": 17
+    },
+    {
+        "type": "text",
+        "text": "9.2 CONTENT RECOMMENDATION ",
+        "text_level": 1,
+        "bbox": [
+            176,
+            103,
+            423,
+            118
+        ],
+        "page_idx": 18
+    },
+    {
+        "type": "text",
+        "text": "9.2.1 ENVIRONMENT DETAILS ",
+        "text_level": 1,
+        "bbox": [
+            176,
+            128,
+            398,
+            143
+        ],
+        "page_idx": 18
+    },
+    {
+        "type": "text",
+        "text": "The evironment has the following components: ",
+        "bbox": [
+            174,
+            154,
+            482,
+            169
+        ],
+        "page_idx": 18
+    },
+    {
+        "type": "text",
+        "text": "1. User type, $x ^ { t }$ : categorical variable representing different types of users. The content recommender conditions its predictions on the type of the current user.   \n2. User loyalty, $\\mathbf { g } ^ { t }$ : the propensity for users of each type to use the platform. User $x ^ { t }$ is sampled from a categorical distribution with parameters given by softmax $( \\mathbf { g } ^ { t } )$ .   \n3. Article type, $y ^ { t }$ : a categorical variable (one-hot encoding) representing the type of article selected by the user.   \n4. User interests, $\\mathbf { W } ^ { t }$ : a matrix whose entries $W _ { x , y } ^ { t }$ represent the average interest user of type $x$ have in articles of type $y$ . ",
+        "bbox": [
+            207,
+            179,
+            825,
+            306
+        ],
+        "page_idx": 18
+    },
+    {
+        "type": "text",
+        "text": "At each time step $t$ , a user $x ^ { t }$ is sampled from a categorical distribution (based on the loyalty of the different user types), then the recommendation system selects which type of article to present in the top position, and finally, the user selects an article. The goal of the recommendation system is to predict the likelihood that the user would click on each of the available articles, in order to select the one which is most interesting to the user. ",
+        "bbox": [
+            174,
+            316,
+            825,
+            387
+        ],
+        "page_idx": 18
+    },
+    {
+        "type": "text",
+        "text": "User loyalty for $x ^ { t }$ then changes in accordance with the self-selection effect, increasing or decreasing proportionally to their interest in the top article. The interests of user type $x ^ { t }$ (represented by a column of $\\mathbf { W } ^ { t }$ ) also change; in accordance with the illusory truth effect, their interest in the topic of the top article (as chosen by the recommender system) always increases. Overall, this environment is an extremely crude representation of reality, but it allows us to incorporate both the effects of self-selection (via covariate shift), and the illusory truth effect (via concept shift). ",
+        "bbox": [
+            173,
+            393,
+            825,
+            477
+        ],
+        "page_idx": 18
+    },
+    {
+        "type": "text",
+        "text": "Formally, this environment is similar to a POMDP\\R, i.e. a POMDP with no reward function, also known as a world model (Armstrong & O’Rourke, 2017; Hadfield-Menell et al., 2017); the difference is that the learner observes the input before acting and only observes the target after acting. The states, observations, and actions given below. ",
+        "bbox": [
+            174,
+            483,
+            826,
+            540
+        ],
+        "page_idx": 18
+    },
+    {
+        "type": "equation",
+        "img_path": "images/db0b38546afafe534e88febd9a237072bd24f328ca4cb11931aed443be558f9d.jpg",
+        "text": "$$\n\\begin{array} { c } { { \\boldsymbol { s } ^ { t } = ( \\mathbf { g } ^ { t } , \\mathbf { W } ^ { t } , \\boldsymbol { x } ^ { t } , \\boldsymbol { y } ^ { t } ) } } \\\\ { { \\boldsymbol { o } _ { \\mathrm { p r e } } ^ { t } , ~ \\boldsymbol { a } ^ { t } , ~ \\boldsymbol { o } _ { \\mathrm { p o s t } } ^ { t } = ( \\boldsymbol { x } ^ { t } , \\hat { \\boldsymbol { y } } ^ { t } , \\boldsymbol { y } ^ { t } ) } } \\end{array}\n$$",
+        "text_format": "latex",
+        "bbox": [
+            383,
+            544,
+            614,
+            585
+        ],
+        "page_idx": 18
+    },
+    {
+        "type": "text",
+        "text": "The state transition function is defined by: ",
+        "bbox": [
+            176,
+            588,
+            450,
+            603
+        ],
+        "page_idx": 18
+    },
+    {
+        "type": "equation",
+        "img_path": "images/18371f2cadc36f7f8a1bbad2aa2e9888bab73b266fa7d957e9964ff1779ada1c.jpg",
+        "text": "$$\n\\begin{array} { r l r } { \\mathbf { g } _ { x ^ { t } } ^ { t + 1 } = \\mathbf { g } _ { x ^ { t } } ^ { t } + \\alpha _ { 1 } W _ { x ^ { t } , \\hat { y } ^ { t } } ^ { t } } & { } & \\\\ { \\mathbf { W } _ { x ^ { t } , \\hat { y } ^ { t } } ^ { t + 1 / 2 } = W _ { x ^ { t } , \\hat { y } ^ { t } } ^ { t } + \\alpha _ { 2 } ; \\quad \\mathbf { W } _ { x ^ { t } } ^ { t + 1 } = \\frac { \\mathbf { W } _ { x ^ { t } } ^ { t + 1 / 2 } } { \\Vert \\mathbf { W } _ { x ^ { t } } ^ { t + 1 / 2 } \\Vert _ { 2 } } } & { } & \\\\ { x ^ { t + 1 } \\sim \\mathrm { s o f t m a x } ( \\mathbf { g } ^ { t + 1 } ) } & { } & \\\\ { y ^ { t + 1 } \\sim \\mathrm { s o f t m a x } ( \\mathbf { W } _ { x ^ { t + 1 } } ^ { t + 1 } ) } & { } & \\end{array}\n$$",
+        "text_format": "latex",
+        "bbox": [
+            328,
+            606,
+            668,
+            712
+        ],
+        "page_idx": 18
+    },
+    {
+        "type": "text",
+        "text": "Where $\\hat { y } ^ { t }$ is the top article as chosen by the recommender, and $\\alpha _ { 1 }$ , $\\alpha _ { 2 }$ represent the rate of covariate and concept shift (respectively). The update for $\\mathbf { W } ^ { t + 1 }$ merely increases the interest of user type $x ^ { t }$ in article type $\\hat { y } ^ { t }$ , then normalizes the interests for that user type. ",
+        "bbox": [
+            174,
+            717,
+            823,
+            760
+        ],
+        "page_idx": 18
+    },
+    {
+        "type": "text",
+        "text": "9.2.2 REPRODUCIBILITY DETAILS ",
+        "text_level": 1,
+        "bbox": [
+            176,
+            773,
+            421,
+            789
+        ],
+        "page_idx": 18
+    },
+    {
+        "type": "text",
+        "text": "For these experiments, the recommendation system is a ReLU-MLP with 1 hidden layer of 100 units, trained via supervised learning with SGD (learning rate $= 0 . 0 1 $ ) to predict which article a user will select. Actions are sampled from the MLP’s predictive distribution. We apply PBT without any hyperparameter selection (this amounts to just doing the EXPLOIT step), and an interval of 10, selecting on accuracy. We use a population of 20 learners (whether applying PBT or not), and match random seeds for the trials with and without PBT. We initialize $\\mathbf { g } ^ { 1 }$ and $\\dot { \\mathbf { W } } ^ { 1 }$ to be the same across the 20 copies of the environment (i.e. the learners start with the same user population), but these values diverge throughout learning. For the environment, we set the number of user and article types both to 10. Initial user loyalties are randomly sampled from $\\mathcal { N } ( 0 , 0 . 0 3 )$ , $\\alpha _ { 1 } = 0 . 0 3$ , and $\\alpha _ { 2 } = 0 . 0 0 3$ . ",
+        "bbox": [
+            173,
+            797,
+            826,
+            924
+        ],
+        "page_idx": 18
+    },
+    {
+        "type": "text",
+        "text": "9.2.3 DETAILS OF EVALUATION ",
+        "text_level": 1,
+        "bbox": [
+            176,
+            103,
+            405,
+            117
+        ],
+        "page_idx": 19
+    },
+    {
+        "type": "text",
+        "text": "We measure concept shift (change in $P ( \\boldsymbol { y } | \\mathbf { x } ) .$ ) as the cosine distance between each user types’ initial and current interest vectors. And we measure covariate shift (change in $P ( \\mathbf { x } ) )$ ) as the KL-divergence between the current and initial user distributions, parametrized by $\\mathbf { g } ^ { 1 }$ and $\\mathbf { g } ^ { t }$ , respectively. Results are presented in 7 (main text). In Figure 12, we additionally plot concept shift and covariate shift as a function of accuracy. We observe that for both types of ADS, at low levels of accuracy PBT actually causes less shift than occur in baseline agents; HI-ADS are only observed for accuracies above $60 \\%$ . This suggests that only relatively strong performers are able to pick up on the HI-ADS revealed by PBT (Fig. 12). ",
+        "bbox": [
+            173,
+            127,
+            825,
+            238
+        ],
+        "page_idx": 19
+    },
+    {
+        "type": "image",
+        "img_path": "images/4ee83ea25b84d00e9bd2c93c72bcfe37100fc4c406703659c71aeb8e4ba6b65a.jpg",
+        "image_caption": [
+            "Figure 12: Amount of auto-induced covariate shift (left) and auto-induced concept shift (right) as a function of performance (accuracy) averaged over all trials, learners, and time-steps. Only relatively strong learners (those which achieve accuracy $> 6 0 \\%$ ) exhibit HI-ADS. "
+        ],
+        "image_footnote": [],
+        "bbox": [
+            236,
+            256,
+            740,
+            429
+        ],
+        "page_idx": 19
+    },
+    {
+        "type": "text",
+        "text": "9.2.4 CONTEXT SWAPPING IN CONTENT RECOMMENDATION ",
+        "text_level": 1,
+        "bbox": [
+            174,
+            511,
+            599,
+            523
+        ],
+        "page_idx": 19
+    },
+    {
+        "type": "text",
+        "text": "We believe context swapping is not appropriate for the content recommendation environment, since when the environments diverge, optimal behavior may differ across environments. Nevertheless, we ran experiments with it for completeness. The main effect appears to be to hamper learning when PBT is not used, see Figure 13. Notably, it does not appear to significantly influence the rate or extent of ADS when combined with PBT. ",
+        "bbox": [
+            174,
+            534,
+            825,
+            603
+        ],
+        "page_idx": 19
+    },
+    {
+        "type": "text",
+        "text": "9.2.5 EXPLORATION OF ENVIRONMENT PARAMETERS ",
+        "text_level": 1,
+        "bbox": [
+            173,
+            619,
+            557,
+            633
+        ],
+        "page_idx": 19
+    },
+    {
+        "type": "text",
+        "text": "In Figure 14, we examine the effect of the rate-of-change parameters $( \\alpha _ { 1 } , \\alpha _ { 2 } )$ of the content recommendation environment on the results provided in the paper. As noted there, our results are qualitatively consistent so long as (1) the initial user distribution is approximately uniform, and (2) the covariate shift rate $( \\alpha _ { 1 } )$ is faster than the concept shift rate $\\left( \\alpha _ { 2 } \\right)$ . These distributions are updated by different mechanisms, and are not directly comparable. Concept shift changes the task more radically, requiring a learner to change its predictions, rather than just become accurate on a wider range of inputs. We conjecture that changes in $P ( \\boldsymbol { y } | \\boldsymbol { x } )$ must therefore be kept smooth enough for the outer loop to have pressure to capitalize on HI-ADS. ",
+        "bbox": [
+            173,
+            643,
+            825,
+            756
+        ],
+        "page_idx": 19
+    },
+    {
+        "type": "image",
+        "img_path": "images/32bff10367ed93273e67c56861072a16c782eb88536e394129f2eaa0da72bbf8.jpg",
+        "image_caption": [
+            "Figure 13: Context swapping doesn’t have the desired effect in the content recommendation environment. "
+        ],
+        "image_footnote": [],
+        "bbox": [
+            200,
+            324,
+            797,
+            660
+        ],
+        "page_idx": 20
+    },
+    {
+        "type": "image",
+        "img_path": "images/a89f27f3d6a426966da5d92bf05f995060ebcc886850b49f7d5327641103133d.jpg",
+        "image_caption": [
+            "Figure 14: Content recommendation results for different values of $\\alpha _ { 1 } , \\alpha _ { 2 }$ "
+        ],
+        "image_footnote": [],
+        "bbox": [
+            236,
+            70,
+            761,
+            896
+        ],
+        "page_idx": 21
+    },
+    {
+        "type": "text",
+        "text": "10 OFFLINE Q-LEARNING CAN REVEAL INCENTIVES FOR ADS ",
+        "text_level": 1,
+        "bbox": [
+            174,
+            102,
+            709,
+            118
+        ],
+        "page_idx": 22
+    },
+    {
+        "type": "text",
+        "text": "First, recall that this unit test is a POMDP, and the state is not observed. Since there are only 2 possible actions, a policy is defined by a single parameter $\\theta = p ( { \\mathrm { c o o p e r a t e } } )$ . Now, the state distribution is $P ( s = { \\mathrm { c o o p e r a t e } } ) = \\theta$ (ignoring the first state, which is appropriate in the limit of infinite data). More specifically, the probability of each state-action combination are as follows: ",
+        "bbox": [
+            174,
+            132,
+            825,
+            190
+        ],
+        "page_idx": 22
+    },
+    {
+        "type": "text",
+        "text": "Suppose we have a dataset of $N$ examples generated by following a fixed policy. ",
+        "bbox": [
+            176,
+            195,
+            700,
+            212
+        ],
+        "page_idx": 22
+    },
+    {
+        "type": "equation",
+        "img_path": "images/7fb69a1c5ab1fc2b242a18d9902a3514228eef0bb301a5ba80a763a06f220cf4.jpg",
+        "text": "$$\n\\begin{array} { r l } & { Q ( C ) = \\frac { | s = C , a = C | R ( s = C , a = C ) + | s = D , a = C | R ( s = D , a = C ) } { | a = C | } } \\\\ & { \\quad \\quad = \\frac { N P ( s = C , a = C ) R ( s = C , a = C ) + N P ( s = D , a = C ) R ( s = D , a = C ) } { N P ( a = C ) } } \\\\ & { \\quad \\quad = \\frac { P ( s = C , a = C ) R ( s = C , a = C ) + P ( s = D , a = C ) R ( s = D , a = C ) } { P ( a = C ) } } \\\\ & { \\quad \\quad = \\frac { \\theta ^ { 2 } R ( s = C , a = C ) + \\theta ( 1 - \\theta ) R ( s = D , a = C ) } { \\theta } } \\\\ & { \\quad \\quad = \\theta R ( s = C , a = C ) + ( 1 - \\theta ) R ( s = D , a = C ) } \\\\ & { \\quad \\quad = \\theta ( 0 ) + ( 1 - \\theta ) ( - 1 ) } \\\\ & { \\quad \\quad = \\theta - 1 } \\end{array}\n$$",
+        "text_format": "latex",
+        "bbox": [
+            192,
+            217,
+            784,
+            407
+        ],
+        "page_idx": 22
+    },
+    {
+        "type": "equation",
+        "img_path": "images/ed2ca0ff8e4e61ab4623d117e82a4bde4865f66e2921751fbf7508dd654aaa90.jpg",
+        "text": "$$\n\\begin{array} { l } { Q ( D ) = \\frac { \\left| s = C , a = D \\right| R \\left( s = C , a = D \\right) + \\left| s = D , a = D \\right| R \\left( s = D , a = D \\right) } { \\left| a = D \\right| } } \\\\ { \\ = \\frac { P \\left( s = C , a = D \\right) R \\left( s = C , a = D \\right) + P \\left( s = D , a = D \\right) R \\left( s = D , a = D \\right) } { P \\left( a = D \\right) } } \\\\ { \\ = P ( s = C ) R ( s = C , a = D ) + P ( s = D ) R ( s = D , a = D ) } \\\\ { \\ = \\theta ( 1 / 2 ) + ( 1 - \\theta ) ( - 1 / 2 ) } \\\\ { \\ = 1 / 2 ( 2 \\theta - 1 ) } \\\\ { \\ = \\theta - 1 / 2 } \\end{array}\n$$",
+        "text_format": "latex",
+        "bbox": [
+            215,
+            438,
+            779,
+            580
+        ],
+        "page_idx": 22
+    },
+    {
+        "type": "text",
+        "text": "So we see that $Q ( D ) > Q ( C )$ , regardless of $\\theta$ . ",
+        "bbox": [
+            174,
+            592,
+            480,
+            608
+        ],
+        "page_idx": 22
+    },
+    {
+        "type": "text",
+        "text": "Now, suppose instead that we have $N$ examples from each of 2 different policies (given by parameters $\\theta _ { 1 }$ and $\\theta _ { 2 }$ ) operating in different environments. Intuitively, this sort of data might arise in practice from “A/B testing”, where 2 different users have been assigned to 2 different policies in order to compare the policies’ performance. We now use $_ { D C }$ to represent $s = D , a = C$ , etc. ",
+        "bbox": [
+            174,
+            613,
+            823,
+            670
+        ],
+        "page_idx": 22
+    },
+    {
+        "type": "equation",
+        "img_path": "images/d4bb4fc9c5a13bc7c7368888d2836a3dc036de75040b79dfca768e03e860184f.jpg",
+        "text": "$$\n\\begin{array} { r l } & { Q ^ { \\theta _ { 1 } , \\theta _ { 2 } } ( C ) = \\frac { | C C | R ( C C ) + | D C | | R ( U C ) } { | C | } } \\\\ & { \\quad \\quad \\quad = \\frac { N \\left( P ^ { \\theta _ { 1 } } ( C C ) + P ^ { \\theta _ { 2 } } ( C C ) \\right) R ( C C ) + N ( P ^ { \\theta _ { 1 } } ( D C ) + P ^ { \\theta _ { 2 } } ( D C ) ) R ( D C ) } { N \\left( P ^ { \\theta _ { 1 } } ( C ) + P ^ { \\theta _ { 2 } } ( C ) \\right) } } \\\\ & { \\quad \\quad \\quad = \\frac { ( P ^ { \\theta _ { 1 } } ( C C ) + P ^ { \\theta _ { 2 } } ( C C ) ) I ( C C ) C + ( P ^ { \\theta _ { 1 } } ( D C ) + P ^ { \\theta _ { 2 } } ( D C ) ) R ( D C ) } { ( F ^ { \\theta _ { 1 } } ( C ) + P ^ { \\theta _ { 2 } } ( C ) ) } } \\\\ & { \\quad \\quad \\quad = \\frac { ( \\theta _ { 1 } ^ { 2 } + \\theta _ { 2 } ^ { 2 } ) R ( C C ) + ( \\theta _ { 1 } ( 1 - \\theta _ { 1 } ) + \\theta _ { 2 } ( 1 - \\theta _ { 2 } ) ) R ( D C ) } { \\theta _ { 1 } + \\theta _ { 2 } } } \\\\ & { \\quad \\quad \\quad = - \\frac { \\theta _ { 1 } ( 1 - \\theta _ { 1 } ) + \\theta _ { 2 } ( 1 - \\theta _ { 2 } ) } { \\theta _ { 1 } + \\theta _ { 2 } } } \\\\ & { \\quad \\quad \\quad = \\frac { \\theta _ { 1 } ^ { 2 } - \\theta _ { 1 } + \\theta _ { 2 } ^ { 2 } - \\theta _ { 2 } } { \\theta _ { 1 } + \\theta _ { 2 } } } \\\\ & { \\quad \\quad \\quad = \\frac { \\theta _ { 1 } ^ { 2 } - \\theta _ { 1 } + \\theta _ { 2 } ^ { 2 } - \\theta _ { 2 } } { \\theta _ { 1 } + \\theta _ { 2 } } } \\\\ & { \\quad \\quad \\quad = \\frac { \\theta _ { 1 } ^ { 2 } - \\theta _ { 1 } + \\theta _ { 2 } ^ { 2 } - \\theta _ { 2 } } { \\theta _ { 1 } + \\theta _ { 2 } } } \\end{array}\n$$",
+        "text_format": "latex",
+        "bbox": [
+            194,
+            694,
+            772,
+            909
+        ],
+        "page_idx": 22
+    },
+    {
+        "type": "equation",
+        "img_path": "images/b154449f50ac688679c16d1b39473d4ce271210e3cd6613c45b3e3f3d0e9126b.jpg",
+        "text": "$$\n\\begin{array} { l } { { Q ^ { \\theta _ { 1 } , \\theta _ { 2 } } ( D ) = \\frac { \\left| C D \\right| R \\left( C D \\right) + \\left| D D \\right| R \\left( D D \\right) } { \\left| D \\right| } } } \\\\ { { \\ = \\frac { \\left( P ^ { \\theta _ { 1 } } \\left( C D \\right) + P ^ { \\theta _ { 2 } } \\left( C D \\right) \\right) R \\left( C D \\right) + \\left( P ^ { \\theta _ { 1 } } \\left( D D \\right) + P ^ { \\theta _ { 2 } } \\left( D D \\right) \\right) R \\left( D D \\right) } { \\left( P ^ { \\theta _ { 1 } } \\left( D \\right) + P ^ { \\theta _ { 2 } } \\left( D \\right) \\right) } } } \\\\ { { \\ = \\frac { 1 / 2 \\left( \\theta _ { 1 } \\left( 1 - \\theta _ { 1 } \\right) + \\theta _ { 2 } \\left( 1 - \\theta _ { 2 } \\right) \\right) - 1 / 2 \\left( ( 1 - \\theta _ { 1 } ) ^ { 2 } + ( 1 - \\theta _ { 2 } ) ^ { 2 } \\right) } { 2 - \\theta _ { 1 } - \\theta _ { 2 } } } } \\\\ { { \\ = \\frac { \\left( 2 \\theta _ { 1 } - 1 \\right) \\left( 1 - \\theta _ { 1 } \\right) + \\left( 2 \\theta _ { 2 } - 1 \\right) \\left( 1 - \\theta _ { 2 } \\right) } { 4 - 2 \\theta _ { 1 } - 2 \\theta _ { 2 } } } } \\end{array}\n$$",
+        "text_format": "latex",
+        "bbox": [
+            209,
+            121,
+            761,
+            263
+        ],
+        "page_idx": 23
+    },
+    {
+        "type": "text",
+        "text": "Now, in Figure 15 we see that $Q ( C ) > Q ( D )$ when one of the policies cooperates with high probability, and the other defects with high probability. Intuitively, the result of pooling data from 2 such policies is very similar to collecting data from an $\\epsilon$ -greedy policy trained online (as in Figure 6). ",
+        "bbox": [
+            173,
+            273,
+            823,
+            316
+        ],
+        "page_idx": 23
+    },
+    {
+        "type": "image",
+        "img_path": "images/f8cf957c6b390dfb762d079853f1b8f334d627fe895c972607d8fe3c71e3938d.jpg",
+        "image_caption": [
+            "Figure 15: Offline Q-learning can also reveal HI-ADS, when pooling data from different (policy, environment) pairs. Yellow regions represent policy pairs for which $\\mathbf { \\bar { Q } } ( C ) > Q ( D )$ , resulting in non-myopic behavior. "
+        ],
+        "image_footnote": [],
+        "bbox": [
+            267,
+            337,
+            728,
+            678
+        ],
+        "page_idx": 23
+    }
+]
\ No newline at end of file
diff --git a/parse/train/3FK30d5BZdu/3FK30d5BZdu_middle.json b/parse/train/3FK30d5BZdu/3FK30d5BZdu_middle.json
new file mode 100644
index 0000000000000000000000000000000000000000..dc271c005a6179eb400ba2b0c6f93199d016c893
--- /dev/null
+++ b/parse/train/3FK30d5BZdu/3FK30d5BZdu_middle.json
@@ -0,0 +1,54960 @@
+{
+    "pdf_info": [
+        {
+            "preproc_blocks": [
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        79,
+                        507,
+                        115
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                78,
+                                507,
+                                98
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        78,
+                                        507,
+                                        98
+                                    ],
+                                    "score": 1.0,
+                                    "content": "HIDDEN INCENTIVES FOR AUTO-INDUCED DISTRIBU-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                99,
+                                213,
+                                117
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        99,
+                                        213,
+                                        117
+                                    ],
+                                    "score": 1.0,
+                                    "content": "TIONAL SHIFT",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        }
+                    ],
+                    "index": 0.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        112,
+                        135,
+                        244,
+                        157
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                113,
+                                135,
+                                201,
+                                147
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        113,
+                                        135,
+                                        201,
+                                        147
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Anonymous authors",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                111,
+                                146,
+                                245,
+                                158
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        111,
+                                        146,
+                                        245,
+                                        158
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Paper under double-blind review",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        }
+                    ],
+                    "index": 2.5
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        278,
+                        186,
+                        333,
+                        199
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                277,
+                                186,
+                                335,
+                                200
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        277,
+                                        186,
+                                        335,
+                                        200
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ABSTRACT",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        }
+                    ],
+                    "index": 4
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        143,
+                        211,
+                        469,
+                        388
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                141,
+                                211,
+                                469,
+                                223
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        211,
+                                        469,
+                                        223
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Decisions made by machine learning systems have increasing influence on the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                142,
+                                223,
+                                469,
+                                235
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        142,
+                                        223,
+                                        469,
+                                        235
+                                    ],
+                                    "score": 1.0,
+                                    "content": "world, yet it is common for machine learning algorithms to assume that no such",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                233,
+                                470,
+                                246
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        233,
+                                        470,
+                                        246
+                                    ],
+                                    "score": 1.0,
+                                    "content": "influence exists. An example is the use of the i.i.d. assumption in content recom-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                244,
+                                469,
+                                256
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        244,
+                                        469,
+                                        256
+                                    ],
+                                    "score": 1.0,
+                                    "content": "mendation. In fact, the (choice of) content displayed can change users’ perceptions",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                255,
+                                470,
+                                268
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        255,
+                                        470,
+                                        268
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and preferences, or even drive them away, causing a shift in the distribution of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                266,
+                                469,
+                                278
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        266,
+                                        469,
+                                        278
+                                    ],
+                                    "score": 1.0,
+                                    "content": "users. We introduce the term auto-induced distributional shift (ADS) to describe",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                277,
+                                470,
+                                290
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        277,
+                                        470,
+                                        290
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the phenomenon of an algorithm causing a change in the distribution of its own",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                288,
+                                470,
+                                300
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        288,
+                                        470,
+                                        300
+                                    ],
+                                    "score": 1.0,
+                                    "content": "inputs. Our goal is to ensure that machine learning systems do not leverage ADS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                300,
+                                470,
+                                311
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        300,
+                                        470,
+                                        311
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to increase performance when doing so could be undesirable. We demonstrate that",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                309,
+                                470,
+                                323
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        309,
+                                        470,
+                                        323
+                                    ],
+                                    "score": 1.0,
+                                    "content": "changes to the learning algorithm, such as the introduction of meta-learning, can",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                321,
+                                469,
+                                333
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        321,
+                                        469,
+                                        333
+                                    ],
+                                    "score": 1.0,
+                                    "content": "cause hidden incentives for auto-induced distributional shift (HI-ADS) to be",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                331,
+                                470,
+                                345
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        331,
+                                        470,
+                                        345
+                                    ],
+                                    "score": 1.0,
+                                    "content": "revealed. To address this issue, we introduce ‘unit tests’ and a mitigation strategy",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                343,
+                                470,
+                                355
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        343,
+                                        470,
+                                        355
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for HI-ADS, as well as a toy environment for modelling real-world issues with HI-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                354,
+                                469,
+                                366
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        354,
+                                        469,
+                                        366
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ADS in content recommendation, where we demonstrate that strong meta-learners",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                363,
+                                470,
+                                378
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        363,
+                                        470,
+                                        378
+                                    ],
+                                    "score": 1.0,
+                                    "content": "achieve gains in performance via ADS. We show meta-learning and Q-learning",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                374,
+                                442,
+                                390
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        374,
+                                        442,
+                                        390
+                                    ],
+                                    "score": 1.0,
+                                    "content": "both sometimes fail unit tests, but pass when using our mitigation strategy.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        }
+                    ],
+                    "index": 12.5
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        407,
+                        206,
+                        419
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                406,
+                                208,
+                                423
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        406,
+                                        208,
+                                        423
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1 INTRODUCTION",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        }
+                    ],
+                    "index": 21
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        426,
+                        506,
+                        482
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                425,
+                                505,
+                                439
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        425,
+                                        505,
+                                        439
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Consider a content recommendation system whose performance is measured by accuracy of predicting",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                436,
+                                506,
+                                450
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        436,
+                                        506,
+                                        450
+                                    ],
+                                    "score": 1.0,
+                                    "content": "what users will click. This system can achieve better performance by either 1) making better predic-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                448,
+                                506,
+                                461
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        448,
+                                        506,
+                                        461
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tions, or 2) changing the distribution of users such that predictions are easier to make. We propose",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                459,
+                                506,
+                                471
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        459,
+                                        506,
+                                        471
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the term auto-induced distributional shift (ADS) to describe this latter kind of distributional shift,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                470,
+                                375,
+                                483
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        470,
+                                        375,
+                                        483
+                                    ],
+                                    "score": 1.0,
+                                    "content": "caused by the algorithm’s own predictions or behaviour (Figure 1).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        }
+                    ],
+                    "index": 24
+                },
+                {
+                    "type": "image",
+                    "bbox": [
+                        173,
+                        496,
+                        448,
+                        582
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                173,
+                                496,
+                                448,
+                                582
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        173,
+                                        496,
+                                        448,
+                                        582
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                173,
+                                                496,
+                                                448,
+                                                582
+                                            ],
+                                            "score": 0.966,
+                                            "type": "image",
+                                            "image_path": "833d9ac6619b77e4b4c37b9014af227c1cffc0cdcc8236556ae4c01b6b89cd90.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 28,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        173,
+                                        496,
+                                        448,
+                                        524.6666666666666
+                                    ],
+                                    "spans": [],
+                                    "index": 27
+                                },
+                                {
+                                    "bbox": [
+                                        173,
+                                        524.6666666666666,
+                                        448,
+                                        553.3333333333333
+                                    ],
+                                    "spans": [],
+                                    "index": 28
+                                },
+                                {
+                                    "bbox": [
+                                        173,
+                                        553.3333333333333,
+                                        448,
+                                        581.9999999999999
+                                    ],
+                                    "spans": [],
+                                    "index": 29
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                108,
+                                593,
+                                505,
+                                627
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        592,
+                                        507,
+                                        606
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                592,
+                                                507,
+                                                606
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 1: Distributions of users over time. Left: A distribution which remains constant over time,",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 30
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        604,
+                                        505,
+                                        617
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                604,
+                                                505,
+                                                617
+                                            ],
+                                            "score": 1.0,
+                                            "content": "following the i.i.d assumption. Right: Auto-induced Distributional Shift (ADS) results in a change",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 31
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        616,
+                                        506,
+                                        627
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                616,
+                                                506,
+                                                627
+                                            ],
+                                            "score": 1.0,
+                                            "content": "in the distribution of users in our content recommendation environment. (see Section 5.2 for details).",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 32
+                                }
+                            ],
+                            "index": 31
+                        }
+                    ],
+                    "index": 29.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        637,
+                        505,
+                        693
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                637,
+                                505,
+                                651
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        637,
+                                        505,
+                                        651
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ADS are not inherently bad, and are sometimes even desirable. But they can cause problems if they",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                648,
+                                505,
+                                661
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        648,
+                                        505,
+                                        661
+                                    ],
+                                    "score": 1.0,
+                                    "content": "occur unexpectedly. It is typical in machine learning (ML) to assume (e.g. via the i.i.d. assumption)",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                660,
+                                505,
+                                673
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        660,
+                                        505,
+                                        673
+                                    ],
+                                    "score": 1.0,
+                                    "content": "that (2) will not happen. However, given the increasing real-world use of ML algorithms, we believe",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                671,
+                                505,
+                                683
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        671,
+                                        505,
+                                        683
+                                    ],
+                                    "score": 1.0,
+                                    "content": "it is important to model and experimentally observe what happens when assumptions like this are",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                682,
+                                284,
+                                692
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        682,
+                                        284,
+                                        692
+                                    ],
+                                    "score": 1.0,
+                                    "content": "violated. This is the motivation of our work.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        }
+                    ],
+                    "index": 35
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        699,
+                        504,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                698,
+                                506,
+                                712
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        698,
+                                        506,
+                                        712
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In many cases, including news recommendation, we would consider (2) a form of cheating—the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                710,
+                                505,
+                                722
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        710,
+                                        505,
+                                        722
+                                    ],
+                                    "score": 1.0,
+                                    "content": "algorithm changed the task rather than solving it as intended. We care which means the algorithm",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                720,
+                                506,
+                                733
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        720,
+                                        506,
+                                        733
+                                    ],
+                                    "score": 1.0,
+                                    "content": "used to solve the problem (e.g. (1) and/or (2)), but we only told it about the ends, so it didn’t know",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        }
+                    ],
+                    "index": 39
+                }
+            ],
+            "page_idx": 0,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        107,
+                        27,
+                        307,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                26,
+                                308,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        26,
+                                        308,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Under review as a conference paper at ICLR 2021",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        302,
+                        751,
+                        308,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                302,
+                                751,
+                                308,
+                                761
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        302,
+                                        751,
+                                        308,
+                                        761
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        79,
+                        507,
+                        115
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                78,
+                                507,
+                                98
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        78,
+                                        507,
+                                        98
+                                    ],
+                                    "score": 1.0,
+                                    "content": "HIDDEN INCENTIVES FOR AUTO-INDUCED DISTRIBU-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                99,
+                                213,
+                                117
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        99,
+                                        213,
+                                        117
+                                    ],
+                                    "score": 1.0,
+                                    "content": "TIONAL SHIFT",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        }
+                    ],
+                    "index": 0.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        112,
+                        135,
+                        244,
+                        157
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                113,
+                                135,
+                                201,
+                                147
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        113,
+                                        135,
+                                        201,
+                                        147
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Anonymous authors",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                111,
+                                146,
+                                245,
+                                158
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        111,
+                                        146,
+                                        245,
+                                        158
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Paper under double-blind review",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        }
+                    ],
+                    "index": 2.5,
+                    "bbox_fs": [
+                        111,
+                        135,
+                        245,
+                        158
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        278,
+                        186,
+                        333,
+                        199
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                277,
+                                186,
+                                335,
+                                200
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        277,
+                                        186,
+                                        335,
+                                        200
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ABSTRACT",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        }
+                    ],
+                    "index": 4
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        143,
+                        211,
+                        469,
+                        388
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                141,
+                                211,
+                                469,
+                                223
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        211,
+                                        469,
+                                        223
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Decisions made by machine learning systems have increasing influence on the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                142,
+                                223,
+                                469,
+                                235
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        142,
+                                        223,
+                                        469,
+                                        235
+                                    ],
+                                    "score": 1.0,
+                                    "content": "world, yet it is common for machine learning algorithms to assume that no such",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                233,
+                                470,
+                                246
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        233,
+                                        470,
+                                        246
+                                    ],
+                                    "score": 1.0,
+                                    "content": "influence exists. An example is the use of the i.i.d. assumption in content recom-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                244,
+                                469,
+                                256
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        244,
+                                        469,
+                                        256
+                                    ],
+                                    "score": 1.0,
+                                    "content": "mendation. In fact, the (choice of) content displayed can change users’ perceptions",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                255,
+                                470,
+                                268
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        255,
+                                        470,
+                                        268
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and preferences, or even drive them away, causing a shift in the distribution of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                266,
+                                469,
+                                278
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        266,
+                                        469,
+                                        278
+                                    ],
+                                    "score": 1.0,
+                                    "content": "users. We introduce the term auto-induced distributional shift (ADS) to describe",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                277,
+                                470,
+                                290
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        277,
+                                        470,
+                                        290
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the phenomenon of an algorithm causing a change in the distribution of its own",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                288,
+                                470,
+                                300
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        288,
+                                        470,
+                                        300
+                                    ],
+                                    "score": 1.0,
+                                    "content": "inputs. Our goal is to ensure that machine learning systems do not leverage ADS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                300,
+                                470,
+                                311
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        300,
+                                        470,
+                                        311
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to increase performance when doing so could be undesirable. We demonstrate that",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                309,
+                                470,
+                                323
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        309,
+                                        470,
+                                        323
+                                    ],
+                                    "score": 1.0,
+                                    "content": "changes to the learning algorithm, such as the introduction of meta-learning, can",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                321,
+                                469,
+                                333
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        321,
+                                        469,
+                                        333
+                                    ],
+                                    "score": 1.0,
+                                    "content": "cause hidden incentives for auto-induced distributional shift (HI-ADS) to be",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                331,
+                                470,
+                                345
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        331,
+                                        470,
+                                        345
+                                    ],
+                                    "score": 1.0,
+                                    "content": "revealed. To address this issue, we introduce ‘unit tests’ and a mitigation strategy",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                343,
+                                470,
+                                355
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        343,
+                                        470,
+                                        355
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for HI-ADS, as well as a toy environment for modelling real-world issues with HI-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                354,
+                                469,
+                                366
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        354,
+                                        469,
+                                        366
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ADS in content recommendation, where we demonstrate that strong meta-learners",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                363,
+                                470,
+                                378
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        363,
+                                        470,
+                                        378
+                                    ],
+                                    "score": 1.0,
+                                    "content": "achieve gains in performance via ADS. We show meta-learning and Q-learning",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                374,
+                                442,
+                                390
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        374,
+                                        442,
+                                        390
+                                    ],
+                                    "score": 1.0,
+                                    "content": "both sometimes fail unit tests, but pass when using our mitigation strategy.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        }
+                    ],
+                    "index": 12.5,
+                    "bbox_fs": [
+                        141,
+                        211,
+                        470,
+                        390
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        407,
+                        206,
+                        419
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                406,
+                                208,
+                                423
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        406,
+                                        208,
+                                        423
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1 INTRODUCTION",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        }
+                    ],
+                    "index": 21
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        426,
+                        506,
+                        482
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                425,
+                                505,
+                                439
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        425,
+                                        505,
+                                        439
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Consider a content recommendation system whose performance is measured by accuracy of predicting",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                436,
+                                506,
+                                450
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        436,
+                                        506,
+                                        450
+                                    ],
+                                    "score": 1.0,
+                                    "content": "what users will click. This system can achieve better performance by either 1) making better predic-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                448,
+                                506,
+                                461
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        448,
+                                        506,
+                                        461
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tions, or 2) changing the distribution of users such that predictions are easier to make. We propose",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                459,
+                                506,
+                                471
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        459,
+                                        506,
+                                        471
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the term auto-induced distributional shift (ADS) to describe this latter kind of distributional shift,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                470,
+                                375,
+                                483
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        470,
+                                        375,
+                                        483
+                                    ],
+                                    "score": 1.0,
+                                    "content": "caused by the algorithm’s own predictions or behaviour (Figure 1).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        }
+                    ],
+                    "index": 24,
+                    "bbox_fs": [
+                        105,
+                        425,
+                        506,
+                        483
+                    ]
+                },
+                {
+                    "type": "image",
+                    "bbox": [
+                        173,
+                        496,
+                        448,
+                        582
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                173,
+                                496,
+                                448,
+                                582
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        173,
+                                        496,
+                                        448,
+                                        582
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                173,
+                                                496,
+                                                448,
+                                                582
+                                            ],
+                                            "score": 0.966,
+                                            "type": "image",
+                                            "image_path": "833d9ac6619b77e4b4c37b9014af227c1cffc0cdcc8236556ae4c01b6b89cd90.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 28,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        173,
+                                        496,
+                                        448,
+                                        524.6666666666666
+                                    ],
+                                    "spans": [],
+                                    "index": 27
+                                },
+                                {
+                                    "bbox": [
+                                        173,
+                                        524.6666666666666,
+                                        448,
+                                        553.3333333333333
+                                    ],
+                                    "spans": [],
+                                    "index": 28
+                                },
+                                {
+                                    "bbox": [
+                                        173,
+                                        553.3333333333333,
+                                        448,
+                                        581.9999999999999
+                                    ],
+                                    "spans": [],
+                                    "index": 29
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                108,
+                                593,
+                                505,
+                                627
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        592,
+                                        507,
+                                        606
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                592,
+                                                507,
+                                                606
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 1: Distributions of users over time. Left: A distribution which remains constant over time,",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 30
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        604,
+                                        505,
+                                        617
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                604,
+                                                505,
+                                                617
+                                            ],
+                                            "score": 1.0,
+                                            "content": "following the i.i.d assumption. Right: Auto-induced Distributional Shift (ADS) results in a change",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 31
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        616,
+                                        506,
+                                        627
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                616,
+                                                506,
+                                                627
+                                            ],
+                                            "score": 1.0,
+                                            "content": "in the distribution of users in our content recommendation environment. (see Section 5.2 for details).",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 32
+                                }
+                            ],
+                            "index": 31
+                        }
+                    ],
+                    "index": 29.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        637,
+                        505,
+                        693
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                637,
+                                505,
+                                651
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        637,
+                                        505,
+                                        651
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ADS are not inherently bad, and are sometimes even desirable. But they can cause problems if they",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                648,
+                                505,
+                                661
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        648,
+                                        505,
+                                        661
+                                    ],
+                                    "score": 1.0,
+                                    "content": "occur unexpectedly. It is typical in machine learning (ML) to assume (e.g. via the i.i.d. assumption)",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                660,
+                                505,
+                                673
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        660,
+                                        505,
+                                        673
+                                    ],
+                                    "score": 1.0,
+                                    "content": "that (2) will not happen. However, given the increasing real-world use of ML algorithms, we believe",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                671,
+                                505,
+                                683
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        671,
+                                        505,
+                                        683
+                                    ],
+                                    "score": 1.0,
+                                    "content": "it is important to model and experimentally observe what happens when assumptions like this are",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                682,
+                                284,
+                                692
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        682,
+                                        284,
+                                        692
+                                    ],
+                                    "score": 1.0,
+                                    "content": "violated. This is the motivation of our work.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        }
+                    ],
+                    "index": 35,
+                    "bbox_fs": [
+                        105,
+                        637,
+                        505,
+                        692
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        699,
+                        504,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                698,
+                                506,
+                                712
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        698,
+                                        506,
+                                        712
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In many cases, including news recommendation, we would consider (2) a form of cheating—the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                710,
+                                505,
+                                722
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        710,
+                                        505,
+                                        722
+                                    ],
+                                    "score": 1.0,
+                                    "content": "algorithm changed the task rather than solving it as intended. We care which means the algorithm",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                720,
+                                506,
+                                733
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        720,
+                                        506,
+                                        733
+                                    ],
+                                    "score": 1.0,
+                                    "content": "used to solve the problem (e.g. (1) and/or (2)), but we only told it about the ends, so it didn’t know",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                83,
+                                506,
+                                95
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        83,
+                                        506,
+                                        95
+                                    ],
+                                    "score": 1.0,
+                                    "content": "not to ’cheat’. This is an example of a specification problem (Leike et al., 2017; Ortega et al., 2018):",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                93,
+                                506,
+                                107
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        93,
+                                        506,
+                                        107
+                                    ],
+                                    "score": 1.0,
+                                    "content": "a problem which arises from a discrepancy between the performance metric (maximize accuracy)",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                105,
+                                393,
+                                117
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        105,
+                                        393,
+                                        117
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and “what we really meant”: in this case, to maximize accuracy via (1).",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 2
+                        }
+                    ],
+                    "index": 39,
+                    "bbox_fs": [
+                        105,
+                        698,
+                        506,
+                        733
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        82,
+                        504,
+                        116
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                83,
+                                506,
+                                95
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        83,
+                                        506,
+                                        95
+                                    ],
+                                    "score": 1.0,
+                                    "content": "not to ’cheat’. This is an example of a specification problem (Leike et al., 2017; Ortega et al., 2018):",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                93,
+                                506,
+                                107
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        93,
+                                        506,
+                                        107
+                                    ],
+                                    "score": 1.0,
+                                    "content": "a problem which arises from a discrepancy between the performance metric (maximize accuracy)",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                105,
+                                393,
+                                117
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        105,
+                                        393,
+                                        117
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and “what we really meant”: in this case, to maximize accuracy via (1).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        }
+                    ],
+                    "index": 1
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        121,
+                        505,
+                        198
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                120,
+                                506,
+                                135
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        120,
+                                        506,
+                                        135
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Ideally, we’d like to quantify the desirability of all possible means, e.g. assign appropriate rewards",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                131,
+                                505,
+                                146
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        131,
+                                        505,
+                                        146
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to all potential strategies and “side-effects”, but this is intractable for real-world settings. Using",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                142,
+                                505,
+                                156
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        142,
+                                        505,
+                                        156
+                                    ],
+                                    "score": 1.0,
+                                    "content": "human feedback to learn reward functions which account for such impacts is a promising approach to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                154,
+                                506,
+                                167
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        154,
+                                        506,
+                                        167
+                                    ],
+                                    "score": 1.0,
+                                    "content": "specifying desired behavior (Leike et al., 2018; Christiano et al., 2017). But the same issue can arise",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                164,
+                                505,
+                                178
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        164,
+                                        505,
+                                        178
+                                    ],
+                                    "score": 1.0,
+                                    "content": "whenever human feedback is used in training: one means of improving performance could be to alter",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                175,
+                                505,
+                                190
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        175,
+                                        505,
+                                        190
+                                    ],
+                                    "score": 1.0,
+                                    "content": "human preferences, making them easier to satisfy. Thus in this work, we pursue a complementary",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                188,
+                                273,
+                                200
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        188,
+                                        273,
+                                        200
+                                    ],
+                                    "score": 1.0,
+                                    "content": "approach: managing learners’ incentives.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        }
+                    ],
+                    "index": 6
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        204,
+                        505,
+                        281
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                203,
+                                506,
+                                218
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        203,
+                                        506,
+                                        218
+                                    ],
+                                    "score": 1.0,
+                                    "content": "A learner has an incentive to behave in a certain way when doing so can increase performance (e.g.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                216,
+                                506,
+                                227
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        216,
+                                        506,
+                                        227
+                                    ],
+                                    "score": 1.0,
+                                    "content": "accuracy or reward). Informally, we say an incentive is hidden when the learner behaves as if it",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                226,
+                                506,
+                                239
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        226,
+                                        506,
+                                        239
+                                    ],
+                                    "score": 1.0,
+                                    "content": "were not present. But we note that changes to the learning algorithm or training regime could cause",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                237,
+                                505,
+                                249
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        237,
+                                        505,
+                                        249
+                                    ],
+                                    "score": 1.0,
+                                    "content": "previously hidden incentives to be revealed, resulting in unexpected and potentially undesirable",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                247,
+                                506,
+                                262
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        247,
+                                        506,
+                                        262
+                                    ],
+                                    "score": 1.0,
+                                    "content": "behaviour. Managing incentives (e.g. controlling which incentives are hidden/ revealed) can allow",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                258,
+                                505,
+                                273
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        258,
+                                        505,
+                                        273
+                                    ],
+                                    "score": 1.0,
+                                    "content": "algorithm designers to disincentivize broad classes of strategies (such as any that rely on manipulating",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                270,
+                                359,
+                                281
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        270,
+                                        359,
+                                        281
+                                    ],
+                                    "score": 1.0,
+                                    "content": "human preferences) without knowing their exact instantiation.1",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        }
+                    ],
+                    "index": 13
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        286,
+                        505,
+                        342
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                285,
+                                506,
+                                301
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        285,
+                                        506,
+                                        301
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The goal of our work is to provide insights and practical tools for understanding and managing",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                298,
+                                505,
+                                310
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        298,
+                                        505,
+                                        310
+                                    ],
+                                    "score": 1.0,
+                                    "content": "incentives, specifically hidden incentives for auto-induced distributional shift: HI-ADS. To",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                308,
+                                505,
+                                322
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        308,
+                                        505,
+                                        322
+                                    ],
+                                    "score": 1.0,
+                                    "content": "study which conditions cause HI-ADS to be revealed, we present unit tests for detecting HI-ADS in",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                320,
+                                505,
+                                332
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        320,
+                                        505,
+                                        332
+                                    ],
+                                    "score": 1.0,
+                                    "content": "supervised learning (SL) and reinforcement learning (RL). We also create an environment that models",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                329,
+                                488,
+                                344
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        329,
+                                        488,
+                                        344
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ADS in news recommendation, illustrating possible effects of revealing HI-ADS in this setting.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        }
+                    ],
+                    "index": 19
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        347,
+                        505,
+                        479
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                347,
+                                505,
+                                360
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        347,
+                                        505,
+                                        360
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The unit tests both have two means by which the learner can improve performance: one which creates",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                358,
+                                506,
+                                371
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        358,
+                                        506,
+                                        371
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ADS and one which does not. The intended method of improving performance is one that does not",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                370,
+                                505,
+                                381
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        370,
+                                        505,
+                                        381
+                                    ],
+                                    "score": 1.0,
+                                    "content": "induce ADS; the other is ’hidden’ and we want it to remain hidden. A learner \"fails\" the unit test if",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                381,
+                                506,
+                                393
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        381,
+                                        506,
+                                        393
+                                    ],
+                                    "score": 1.0,
+                                    "content": "it nonetheless pursues the incentive to increase performance via ADS. In both the RL and SL unit",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                392,
+                                506,
+                                404
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        392,
+                                        506,
+                                        404
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tests, we find that introducing an outer-loop of meta-learning (e.g. Population-Based Training (PBT)",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                402,
+                                506,
+                                415
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        402,
+                                        506,
+                                        415
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Jaderberg et al. (2017)) can lead to high levels of failure. Similarly, recommender systems trained",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                414,
+                                506,
+                                426
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        414,
+                                        506,
+                                        426
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with PBT induce larger drifts in user base and user interests. These results suggest that failure of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                423,
+                                506,
+                                438
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        423,
+                                        506,
+                                        438
+                                    ],
+                                    "score": 1.0,
+                                    "content": "our unit tests indicates that an algorithm is prone to revealing HI-ADS in other settings. Finally, we",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                434,
+                                506,
+                                450
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        434,
+                                        506,
+                                        450
+                                    ],
+                                    "score": 1.0,
+                                    "content": "propose and test a mitigation strategy we call context swapping. The strategy consists of rotating",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                446,
+                                506,
+                                459
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        446,
+                                        506,
+                                        459
+                                    ],
+                                    "score": 1.0,
+                                    "content": "learners through different environments throughout learning, so that they can’t see the results or",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                457,
+                                506,
+                                470
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        457,
+                                        506,
+                                        470
+                                    ],
+                                    "score": 1.0,
+                                    "content": "correlations of their actions in one environment over longer time horizons. This effectively mitigates",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                468,
+                                507,
+                                480
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        468,
+                                        507,
+                                        480
+                                    ],
+                                    "score": 1.0,
+                                    "content": "HI-ADS in our unit test environments, but did not work well in content recommendation experiments.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        }
+                    ],
+                    "index": 27.5
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        490,
+                        200,
+                        502
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                488,
+                                201,
+                                505
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        488,
+                                        201,
+                                        505
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2 BACKGROUND",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        }
+                    ],
+                    "index": 34
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        510,
+                        359,
+                        520
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                509,
+                                361,
+                                522
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        509,
+                                        361,
+                                        522
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2.1 META-LEARNING AND POPULATION BASED TRAINING",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        }
+                    ],
+                    "index": 35
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        524,
+                        505,
+                        623
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                524,
+                                505,
+                                537
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        524,
+                                        505,
+                                        537
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Meta-learning is the use of machine learning techniques to learn machine learning algorithms. This",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                534,
+                                505,
+                                548
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        534,
+                                        421,
+                                        548
+                                    ],
+                                    "score": 1.0,
+                                    "content": "involves instantiating multiple learning scenarios which run in an inner loop",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        421,
+                                        535,
+                                        438,
+                                        546
+                                    ],
+                                    "score": 0.42,
+                                    "content": "\\mathbf { \\Pi } ( \\mathbf { I I L } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        439,
+                                        534,
+                                        505,
+                                        548
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", while an outer",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                545,
+                                506,
+                                560
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        545,
+                                        506,
+                                        560
+                                    ],
+                                    "score": 1.0,
+                                    "content": "loop (OL) uses the outcomes of the inner loop(s) as data-points from which to learn which learning",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                557,
+                                505,
+                                569
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        557,
+                                        505,
+                                        569
+                                    ],
+                                    "score": 1.0,
+                                    "content": "algorithms are most effective (Metz et al., 2019). The number of IL steps per OL step is called the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                567,
+                                506,
+                                582
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        567,
+                                        506,
+                                        582
+                                    ],
+                                    "score": 1.0,
+                                    "content": "interval. Many recent works focus on multi-task meta-learning, where the OL seeks to find learning",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                579,
+                                506,
+                                591
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        579,
+                                        506,
+                                        591
+                                    ],
+                                    "score": 1.0,
+                                    "content": "rules that generalize to unseen tasks by training the IL on a distribution of tasks (Finn et al., 2017).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                590,
+                                506,
+                                603
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        590,
+                                        506,
+                                        603
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Single-task meta-learning includes learning an optimizer for a single task (Gong et al., 2018), and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                600,
+                                507,
+                                615
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        600,
+                                        507,
+                                        615
+                                    ],
+                                    "score": 1.0,
+                                    "content": "adaptive methods for selecting models (Kalousis, 2000) or setting hyperparameters (Snoek et al.,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                610,
+                                427,
+                                625
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        610,
+                                        427,
+                                        625
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2012). For simplicity in this initial study we focus on single-task meta-learning.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        }
+                    ],
+                    "index": 40
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        629,
+                        505,
+                        673
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                628,
+                                506,
+                                642
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        628,
+                                        506,
+                                        642
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Population-based training (PBT; Jaderberg et al., 2017) is a meta-learning algorithm that trains",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                639,
+                                505,
+                                653
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        639,
+                                        177,
+                                        653
+                                    ],
+                                    "score": 1.0,
+                                    "content": "multiple learners",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        177,
+                                        640,
+                                        218,
+                                        651
+                                    ],
+                                    "score": 0.92,
+                                    "content": "L _ { 1 } , . . . , L _ { n }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        219,
+                                        639,
+                                        344,
+                                        653
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in parallel, after each interval (",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        344,
+                                        640,
+                                        353,
+                                        650
+                                    ],
+                                    "score": 0.65,
+                                    "content": "T",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        353,
+                                        639,
+                                        505,
+                                        653
+                                    ],
+                                    "score": 1.0,
+                                    "content": "steps of IL) applying an evolutionary",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                650,
+                                506,
+                                663
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        650,
+                                        506,
+                                        663
+                                    ],
+                                    "score": 1.0,
+                                    "content": "OL step which consists of: (1) Evaluate the performance of each learner, (2) Replace both parameters",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 47
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                662,
+                                505,
+                                673
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        662,
+                                        205,
+                                        673
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and hyperparameters of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        205,
+                                        662,
+                                        225,
+                                        672
+                                    ],
+                                    "score": 0.86,
+                                    "content": "20 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        225,
+                                        662,
+                                        461,
+                                        673
+                                    ],
+                                    "score": 1.0,
+                                    "content": "lowest-performing learners with copies of those from the",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        461,
+                                        662,
+                                        482,
+                                        672
+                                    ],
+                                    "score": 0.84,
+                                    "content": "20 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        482,
+                                        662,
+                                        505,
+                                        673
+                                    ],
+                                    "score": 1.0,
+                                    "content": "high-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 48
+                        }
+                    ],
+                    "index": 46.5
+                }
+            ],
+            "page_idx": 1,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        106,
+                        681,
+                        505,
+                        731
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                119,
+                                680,
+                                505,
+                                694
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        119,
+                                        680,
+                                        505,
+                                        694
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1Note removing or hiding an incentive for a behavior is different from prohibiting that behavior, which may",
+                                    "type": "text"
+                                }
+                            ]
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                691,
+                                505,
+                                703
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        691,
+                                        505,
+                                        703
+                                    ],
+                                    "score": 1.0,
+                                    "content": "still occur incidentally. In particular, not having a (revealed) incentive for behaviors that change a human’s",
+                                    "type": "text"
+                                }
+                            ]
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                700,
+                                506,
+                                713
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        700,
+                                        506,
+                                        713
+                                    ],
+                                    "score": 1.0,
+                                    "content": "preferences, is not the same as having a (revealed) incentive for behaviors that preserve a human’s preferences.",
+                                    "type": "text"
+                                }
+                            ]
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                708,
+                                507,
+                                725
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        708,
+                                        507,
+                                        725
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The first is often preferable; we don’t want to prevent changes in human preferences that occur “naturally”, e.g.",
+                                    "type": "text"
+                                }
+                            ]
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                721,
+                                260,
+                                734
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        721,
+                                        260,
+                                        734
+                                    ],
+                                    "score": 1.0,
+                                    "content": "as a result of good arguments or evidence.",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        108,
+                        27,
+                        306,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                26,
+                                308,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        26,
+                                        308,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Under review as a conference paper at ICLR 2021",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        302,
+                        752,
+                        308,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                302,
+                                750,
+                                310,
+                                763
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        302,
+                                        750,
+                                        310,
+                                        763
+                                    ],
+                                    "score": 1.0,
+                                    "content": "",
+                                    "type": "text",
+                                    "height": 13,
+                                    "width": 8
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        82,
+                        504,
+                        116
+                    ],
+                    "lines": [],
+                    "index": 1,
+                    "bbox_fs": [
+                        105,
+                        83,
+                        506,
+                        117
+                    ],
+                    "lines_deleted": true
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        121,
+                        505,
+                        198
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                120,
+                                506,
+                                135
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        120,
+                                        506,
+                                        135
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Ideally, we’d like to quantify the desirability of all possible means, e.g. assign appropriate rewards",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                131,
+                                505,
+                                146
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        131,
+                                        505,
+                                        146
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to all potential strategies and “side-effects”, but this is intractable for real-world settings. Using",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                142,
+                                505,
+                                156
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        142,
+                                        505,
+                                        156
+                                    ],
+                                    "score": 1.0,
+                                    "content": "human feedback to learn reward functions which account for such impacts is a promising approach to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                154,
+                                506,
+                                167
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        154,
+                                        506,
+                                        167
+                                    ],
+                                    "score": 1.0,
+                                    "content": "specifying desired behavior (Leike et al., 2018; Christiano et al., 2017). But the same issue can arise",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                164,
+                                505,
+                                178
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        164,
+                                        505,
+                                        178
+                                    ],
+                                    "score": 1.0,
+                                    "content": "whenever human feedback is used in training: one means of improving performance could be to alter",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                175,
+                                505,
+                                190
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        175,
+                                        505,
+                                        190
+                                    ],
+                                    "score": 1.0,
+                                    "content": "human preferences, making them easier to satisfy. Thus in this work, we pursue a complementary",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                188,
+                                273,
+                                200
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        188,
+                                        273,
+                                        200
+                                    ],
+                                    "score": 1.0,
+                                    "content": "approach: managing learners’ incentives.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        }
+                    ],
+                    "index": 6,
+                    "bbox_fs": [
+                        105,
+                        120,
+                        506,
+                        200
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        204,
+                        505,
+                        281
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                203,
+                                506,
+                                218
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        203,
+                                        506,
+                                        218
+                                    ],
+                                    "score": 1.0,
+                                    "content": "A learner has an incentive to behave in a certain way when doing so can increase performance (e.g.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                216,
+                                506,
+                                227
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        216,
+                                        506,
+                                        227
+                                    ],
+                                    "score": 1.0,
+                                    "content": "accuracy or reward). Informally, we say an incentive is hidden when the learner behaves as if it",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                226,
+                                506,
+                                239
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        226,
+                                        506,
+                                        239
+                                    ],
+                                    "score": 1.0,
+                                    "content": "were not present. But we note that changes to the learning algorithm or training regime could cause",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                237,
+                                505,
+                                249
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        237,
+                                        505,
+                                        249
+                                    ],
+                                    "score": 1.0,
+                                    "content": "previously hidden incentives to be revealed, resulting in unexpected and potentially undesirable",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                247,
+                                506,
+                                262
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        247,
+                                        506,
+                                        262
+                                    ],
+                                    "score": 1.0,
+                                    "content": "behaviour. Managing incentives (e.g. controlling which incentives are hidden/ revealed) can allow",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                258,
+                                505,
+                                273
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        258,
+                                        505,
+                                        273
+                                    ],
+                                    "score": 1.0,
+                                    "content": "algorithm designers to disincentivize broad classes of strategies (such as any that rely on manipulating",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                270,
+                                359,
+                                281
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        270,
+                                        359,
+                                        281
+                                    ],
+                                    "score": 1.0,
+                                    "content": "human preferences) without knowing their exact instantiation.1",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        }
+                    ],
+                    "index": 13,
+                    "bbox_fs": [
+                        105,
+                        203,
+                        506,
+                        281
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        286,
+                        505,
+                        342
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                285,
+                                506,
+                                301
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        285,
+                                        506,
+                                        301
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The goal of our work is to provide insights and practical tools for understanding and managing",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                298,
+                                505,
+                                310
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        298,
+                                        505,
+                                        310
+                                    ],
+                                    "score": 1.0,
+                                    "content": "incentives, specifically hidden incentives for auto-induced distributional shift: HI-ADS. To",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                308,
+                                505,
+                                322
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        308,
+                                        505,
+                                        322
+                                    ],
+                                    "score": 1.0,
+                                    "content": "study which conditions cause HI-ADS to be revealed, we present unit tests for detecting HI-ADS in",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                320,
+                                505,
+                                332
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        320,
+                                        505,
+                                        332
+                                    ],
+                                    "score": 1.0,
+                                    "content": "supervised learning (SL) and reinforcement learning (RL). We also create an environment that models",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                329,
+                                488,
+                                344
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        329,
+                                        488,
+                                        344
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ADS in news recommendation, illustrating possible effects of revealing HI-ADS in this setting.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        }
+                    ],
+                    "index": 19,
+                    "bbox_fs": [
+                        105,
+                        285,
+                        506,
+                        344
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        347,
+                        505,
+                        479
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                347,
+                                505,
+                                360
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        347,
+                                        505,
+                                        360
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The unit tests both have two means by which the learner can improve performance: one which creates",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                358,
+                                506,
+                                371
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        358,
+                                        506,
+                                        371
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ADS and one which does not. The intended method of improving performance is one that does not",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                370,
+                                505,
+                                381
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        370,
+                                        505,
+                                        381
+                                    ],
+                                    "score": 1.0,
+                                    "content": "induce ADS; the other is ’hidden’ and we want it to remain hidden. A learner \"fails\" the unit test if",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                381,
+                                506,
+                                393
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        381,
+                                        506,
+                                        393
+                                    ],
+                                    "score": 1.0,
+                                    "content": "it nonetheless pursues the incentive to increase performance via ADS. In both the RL and SL unit",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                392,
+                                506,
+                                404
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        392,
+                                        506,
+                                        404
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tests, we find that introducing an outer-loop of meta-learning (e.g. Population-Based Training (PBT)",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                402,
+                                506,
+                                415
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        402,
+                                        506,
+                                        415
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Jaderberg et al. (2017)) can lead to high levels of failure. Similarly, recommender systems trained",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                414,
+                                506,
+                                426
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        414,
+                                        506,
+                                        426
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with PBT induce larger drifts in user base and user interests. These results suggest that failure of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                423,
+                                506,
+                                438
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        423,
+                                        506,
+                                        438
+                                    ],
+                                    "score": 1.0,
+                                    "content": "our unit tests indicates that an algorithm is prone to revealing HI-ADS in other settings. Finally, we",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                434,
+                                506,
+                                450
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        434,
+                                        506,
+                                        450
+                                    ],
+                                    "score": 1.0,
+                                    "content": "propose and test a mitigation strategy we call context swapping. The strategy consists of rotating",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                446,
+                                506,
+                                459
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        446,
+                                        506,
+                                        459
+                                    ],
+                                    "score": 1.0,
+                                    "content": "learners through different environments throughout learning, so that they can’t see the results or",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                457,
+                                506,
+                                470
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        457,
+                                        506,
+                                        470
+                                    ],
+                                    "score": 1.0,
+                                    "content": "correlations of their actions in one environment over longer time horizons. This effectively mitigates",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                468,
+                                507,
+                                480
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        468,
+                                        507,
+                                        480
+                                    ],
+                                    "score": 1.0,
+                                    "content": "HI-ADS in our unit test environments, but did not work well in content recommendation experiments.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        }
+                    ],
+                    "index": 27.5,
+                    "bbox_fs": [
+                        104,
+                        347,
+                        507,
+                        480
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        490,
+                        200,
+                        502
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                488,
+                                201,
+                                505
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        488,
+                                        201,
+                                        505
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2 BACKGROUND",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        }
+                    ],
+                    "index": 34
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        510,
+                        359,
+                        520
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                509,
+                                361,
+                                522
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        509,
+                                        361,
+                                        522
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2.1 META-LEARNING AND POPULATION BASED TRAINING",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        }
+                    ],
+                    "index": 35
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        524,
+                        505,
+                        623
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                524,
+                                505,
+                                537
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        524,
+                                        505,
+                                        537
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Meta-learning is the use of machine learning techniques to learn machine learning algorithms. This",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                534,
+                                505,
+                                548
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        534,
+                                        421,
+                                        548
+                                    ],
+                                    "score": 1.0,
+                                    "content": "involves instantiating multiple learning scenarios which run in an inner loop",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        421,
+                                        535,
+                                        438,
+                                        546
+                                    ],
+                                    "score": 0.42,
+                                    "content": "\\mathbf { \\Pi } ( \\mathbf { I I L } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        439,
+                                        534,
+                                        505,
+                                        548
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", while an outer",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                545,
+                                506,
+                                560
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        545,
+                                        506,
+                                        560
+                                    ],
+                                    "score": 1.0,
+                                    "content": "loop (OL) uses the outcomes of the inner loop(s) as data-points from which to learn which learning",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                557,
+                                505,
+                                569
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        557,
+                                        505,
+                                        569
+                                    ],
+                                    "score": 1.0,
+                                    "content": "algorithms are most effective (Metz et al., 2019). The number of IL steps per OL step is called the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                567,
+                                506,
+                                582
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        567,
+                                        506,
+                                        582
+                                    ],
+                                    "score": 1.0,
+                                    "content": "interval. Many recent works focus on multi-task meta-learning, where the OL seeks to find learning",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                579,
+                                506,
+                                591
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        579,
+                                        506,
+                                        591
+                                    ],
+                                    "score": 1.0,
+                                    "content": "rules that generalize to unseen tasks by training the IL on a distribution of tasks (Finn et al., 2017).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                590,
+                                506,
+                                603
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        590,
+                                        506,
+                                        603
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Single-task meta-learning includes learning an optimizer for a single task (Gong et al., 2018), and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                600,
+                                507,
+                                615
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        600,
+                                        507,
+                                        615
+                                    ],
+                                    "score": 1.0,
+                                    "content": "adaptive methods for selecting models (Kalousis, 2000) or setting hyperparameters (Snoek et al.,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                610,
+                                427,
+                                625
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        610,
+                                        427,
+                                        625
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2012). For simplicity in this initial study we focus on single-task meta-learning.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        }
+                    ],
+                    "index": 40,
+                    "bbox_fs": [
+                        105,
+                        524,
+                        507,
+                        625
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        629,
+                        505,
+                        673
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                628,
+                                506,
+                                642
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        628,
+                                        506,
+                                        642
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Population-based training (PBT; Jaderberg et al., 2017) is a meta-learning algorithm that trains",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                639,
+                                505,
+                                653
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        639,
+                                        177,
+                                        653
+                                    ],
+                                    "score": 1.0,
+                                    "content": "multiple learners",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        177,
+                                        640,
+                                        218,
+                                        651
+                                    ],
+                                    "score": 0.92,
+                                    "content": "L _ { 1 } , . . . , L _ { n }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        219,
+                                        639,
+                                        344,
+                                        653
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in parallel, after each interval (",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        344,
+                                        640,
+                                        353,
+                                        650
+                                    ],
+                                    "score": 0.65,
+                                    "content": "T",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        353,
+                                        639,
+                                        505,
+                                        653
+                                    ],
+                                    "score": 1.0,
+                                    "content": "steps of IL) applying an evolutionary",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                650,
+                                506,
+                                663
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        650,
+                                        506,
+                                        663
+                                    ],
+                                    "score": 1.0,
+                                    "content": "OL step which consists of: (1) Evaluate the performance of each learner, (2) Replace both parameters",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 47
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                662,
+                                505,
+                                673
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        662,
+                                        205,
+                                        673
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and hyperparameters of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        205,
+                                        662,
+                                        225,
+                                        672
+                                    ],
+                                    "score": 0.86,
+                                    "content": "20 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        225,
+                                        662,
+                                        461,
+                                        673
+                                    ],
+                                    "score": 1.0,
+                                    "content": "lowest-performing learners with copies of those from the",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        461,
+                                        662,
+                                        482,
+                                        672
+                                    ],
+                                    "score": 0.84,
+                                    "content": "20 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        482,
+                                        662,
+                                        505,
+                                        673
+                                    ],
+                                    "score": 1.0,
+                                    "content": "high-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 48
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                81,
+                                505,
+                                95
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        81,
+                                        505,
+                                        95
+                                    ],
+                                    "score": 1.0,
+                                    "content": "performing learners (EXPLOIT). (3) Randomly perturb the hyperparameters (but not the parameters)",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                94,
+                                220,
+                                105
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        94,
+                                        220,
+                                        105
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of all learners (EXPLORE).",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 1
+                        }
+                    ],
+                    "index": 46.5,
+                    "bbox_fs": [
+                        105,
+                        628,
+                        506,
+                        673
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        82,
+                        505,
+                        105
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                81,
+                                505,
+                                95
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        81,
+                                        505,
+                                        95
+                                    ],
+                                    "score": 1.0,
+                                    "content": "performing learners (EXPLOIT). (3) Randomly perturb the hyperparameters (but not the parameters)",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                94,
+                                220,
+                                105
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        94,
+                                        220,
+                                        105
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of all learners (EXPLORE).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        }
+                    ],
+                    "index": 0.5
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        117,
+                        382,
+                        127
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                115,
+                                384,
+                                129
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        115,
+                                        384,
+                                        129
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2.2 DISTRIBUTIONAL SHIFT AND CONTENT RECOMMENDATION",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        }
+                    ],
+                    "index": 2
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        131,
+                        506,
+                        297
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                132,
+                                506,
+                                144
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        132,
+                                        506,
+                                        144
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In general, distributional shift refers to change of the data distribution over time. In supervised",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                142,
+                                506,
+                                156
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        142,
+                                        182,
+                                        156
+                                    ],
+                                    "score": 1.0,
+                                    "content": "learning with data",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        183,
+                                        145,
+                                        191,
+                                        153
+                                    ],
+                                    "score": 0.39,
+                                    "content": "\\mathbf { x }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        191,
+                                        142,
+                                        235,
+                                        156
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and labels",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        236,
+                                        145,
+                                        242,
+                                        154
+                                    ],
+                                    "score": 0.73,
+                                    "content": "y",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        243,
+                                        142,
+                                        506,
+                                        156
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", this can be more specifically described as dataset shift: change",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                154,
+                                506,
+                                166
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        154,
+                                        214,
+                                        166
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in the joint distribution of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        215,
+                                        154,
+                                        247,
+                                        165
+                                    ],
+                                    "score": 0.92,
+                                    "content": "P ( \\mathbf { x } , y )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        247,
+                                        154,
+                                        506,
+                                        166
+                                    ],
+                                    "score": 1.0,
+                                    "content": "between the training and test sets (Moreno-Torres et al., 2012;",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                164,
+                                506,
+                                177
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        164,
+                                        506,
+                                        177
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Quionero-Candela et al., 2009). As identified by Moreno-Torres et al. (2012), two common kinds",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                176,
+                                506,
+                                188
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        176,
+                                        281,
+                                        188
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of shift are: (1) Covariate shift: changing",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        282,
+                                        176,
+                                        304,
+                                        188
+                                    ],
+                                    "score": 0.91,
+                                    "content": "P ( \\mathbf { x } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        305,
+                                        176,
+                                        506,
+                                        188
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". In the example of content recommendation, this",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                186,
+                                506,
+                                199
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        186,
+                                        506,
+                                        199
+                                    ],
+                                    "score": 1.0,
+                                    "content": "corresponds to changing the user base of the recommendation system. For instance, a media outlet",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                196,
+                                506,
+                                211
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        196,
+                                        506,
+                                        211
+                                    ],
+                                    "score": 1.0,
+                                    "content": "which publishes inflammatory content may appeal to users with extreme views while alienating more",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                208,
+                                506,
+                                222
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        208,
+                                        506,
+                                        222
+                                    ],
+                                    "score": 1.0,
+                                    "content": "moderate users. This self-selection effect (Kayhan, 2015) may appear to a recommendation system",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                220,
+                                506,
+                                232
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        220,
+                                        506,
+                                        232
+                                    ],
+                                    "score": 1.0,
+                                    "content": "as an increase in performance, leading to a feedback effect, as previously noted by Shah et al. (2018).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                231,
+                                505,
+                                242
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        231,
+                                        505,
+                                        242
+                                    ],
+                                    "score": 1.0,
+                                    "content": "This type of feedback effect has been identified as contributing to filter bubbles and radicalization",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                241,
+                                506,
+                                254
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        242,
+                                        360,
+                                        254
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(Pariser, 2011; Kayhan, 2015). (2) Concept shift: changing",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        360,
+                                        241,
+                                        390,
+                                        253
+                                    ],
+                                    "score": 0.93,
+                                    "content": "P ( \\boldsymbol { y } | \\mathbf { x } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        391,
+                                        242,
+                                        506,
+                                        254
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". In the example of content",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                252,
+                                507,
+                                266
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        252,
+                                        507,
+                                        266
+                                    ],
+                                    "score": 1.0,
+                                    "content": "recommendation, this corresponds to changing a given user’s interest in different kinds of content.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                263,
+                                505,
+                                275
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        263,
+                                        505,
+                                        275
+                                    ],
+                                    "score": 1.0,
+                                    "content": "For example, exposure to a fake news story has been shown to increase the perceived accuracy of (and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                275,
+                                506,
+                                287
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        275,
+                                        506,
+                                        287
+                                    ],
+                                    "score": 1.0,
+                                    "content": "thus presumably future interest in) the content, an example of the illusory truth effect (Pennycook",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                285,
+                                478,
+                                298
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        285,
+                                        478,
+                                        298
+                                    ],
+                                    "score": 1.0,
+                                    "content": "et al., 2019). For further details on such effects in content recommendation, see Appendix 8.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        }
+                    ],
+                    "index": 10
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        310,
+                        358,
+                        323
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                309,
+                                360,
+                                325
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        309,
+                                        360,
+                                        325
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3 AUTO-INDUCED DISTRIBUTION SHIFT (ADS)",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        }
+                    ],
+                    "index": 18
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        331,
+                        506,
+                        375
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                330,
+                                506,
+                                344
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        330,
+                                        506,
+                                        344
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Auto-induced distribution shift (ADS) is distributional shift caused by an algorithm’s behaviour. This",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                340,
+                                507,
+                                357
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        340,
+                                        507,
+                                        357
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is in contrast to distributional shift which would happen even if the learner were not present - e.g.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                353,
+                                505,
+                                366
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        353,
+                                        505,
+                                        366
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for a crash prediction algorithm trained on data from the summer, encountering snowy roads is an",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                364,
+                                424,
+                                377
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        364,
+                                        424,
+                                        377
+                                    ],
+                                    "score": 1.0,
+                                    "content": "example of distributional shift, but not auto-induced distributional shift (ADS).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        }
+                    ],
+                    "index": 20.5
+                },
+                {
+                    "type": "image",
+                    "bbox": [
+                        106,
+                        391,
+                        506,
+                        496
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                106,
+                                391,
+                                506,
+                                496
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        391,
+                                        506,
+                                        496
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                391,
+                                                506,
+                                                496
+                                            ],
+                                            "score": 0.802,
+                                            "type": "image",
+                                            "image_path": "e073c2dfa79ed13d098885e0aecd106a567f60dc458af17cb1a90e8412bd2143.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 24,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        391,
+                                        506,
+                                        426.0
+                                    ],
+                                    "spans": [],
+                                    "index": 23
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        426.0,
+                                        506,
+                                        461.0
+                                    ],
+                                    "spans": [],
+                                    "index": 24
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        461.0,
+                                        506,
+                                        496.0
+                                    ],
+                                    "spans": [],
+                                    "index": 25
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 24
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        506,
+                        505,
+                        572
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                506,
+                                506,
+                                519
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        506,
+                                        429,
+                                        519
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Figure 2: The widely studied problems of reinforcement learning (RL) with state",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        430,
+                                        509,
+                                        435,
+                                        516
+                                    ],
+                                    "score": 0.63,
+                                    "content": "s",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        436,
+                                        506,
+                                        465,
+                                        519
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", action",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        466,
+                                        509,
+                                        472,
+                                        516
+                                    ],
+                                    "score": 0.66,
+                                    "content": "a",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        472,
+                                        506,
+                                        506,
+                                        519
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", reward",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                518,
+                                505,
+                                529
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        519,
+                                        113,
+                                        527
+                                    ],
+                                    "score": 0.64,
+                                    "content": "r",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        113,
+                                        518,
+                                        325,
+                                        529
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tuples, and i.i.d. supervised learning (SL) with inputs",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        325,
+                                        519,
+                                        332,
+                                        527
+                                    ],
+                                    "score": 0.74,
+                                    "content": "x",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        332,
+                                        518,
+                                        381,
+                                        529
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", predictions",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        381,
+                                        518,
+                                        388,
+                                        529
+                                    ],
+                                    "score": 0.83,
+                                    "content": "\\hat { y }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        389,
+                                        518,
+                                        423,
+                                        529
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and loss",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        424,
+                                        518,
+                                        428,
+                                        527
+                                    ],
+                                    "score": 0.44,
+                                    "content": "l",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        429,
+                                        518,
+                                        505,
+                                        529
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(a,d) are free from",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                528,
+                                506,
+                                541
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        528,
+                                        506,
+                                        541
+                                    ],
+                                    "score": 1.0,
+                                    "content": "incentive problems. We focus on cases where there are incentives present which the learner is not",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                540,
+                                504,
+                                551
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        540,
+                                        173,
+                                        551
+                                    ],
+                                    "score": 1.0,
+                                    "content": "meant to pursue",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        173,
+                                        540,
+                                        193,
+                                        550
+                                    ],
+                                    "score": 0.59,
+                                    "content": "^ { ( \\mathbf { b } , \\mathbf { c } ) }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        194,
+                                        540,
+                                        504,
+                                        551
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Lines show paths of influence. The learner may have incentives to influence",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                550,
+                                506,
+                                563
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        550,
+                                        258,
+                                        563
+                                    ],
+                                    "score": 1.0,
+                                    "content": "any nodes descending from its action,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        258,
+                                        551,
+                                        267,
+                                        560
+                                    ],
+                                    "score": 0.67,
+                                    "content": "A",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        267,
+                                        550,
+                                        325,
+                                        563
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", or prediction,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        326,
+                                        551,
+                                        333,
+                                        561
+                                    ],
+                                    "score": 0.81,
+                                    "content": "\\hat { y }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        333,
+                                        550,
+                                        506,
+                                        563
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Which incentives are undesirable (orange)",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                561,
+                                368,
+                                573
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        561,
+                                        368,
+                                        573
+                                    ],
+                                    "score": 1.0,
+                                    "content": "or desirable (cyan) for the learner to pursue is context-dependent.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        }
+                    ],
+                    "index": 28.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        586,
+                        505,
+                        653
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                586,
+                                506,
+                                599
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        586,
+                                        506,
+                                        599
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We emphasize that ADS are not inherently bad or good; often ADS can even be desirable: consider",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                597,
+                                505,
+                                610
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        597,
+                                        505,
+                                        610
+                                    ],
+                                    "score": 1.0,
+                                    "content": "an algorithm meant to alert drivers of imminent collisions. If it works well, such a system will help",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                608,
+                                506,
+                                621
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        608,
+                                        506,
+                                        621
+                                    ],
+                                    "score": 1.0,
+                                    "content": "drivers avoid crashing, thus making self-refuting predictions which result in ADS. What separates",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                620,
+                                505,
+                                631
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        620,
+                                        505,
+                                        631
+                                    ],
+                                    "score": 1.0,
+                                    "content": "desirable and undesirable ADS? The collision-alert system alters its data distribution in a way that is",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                631,
+                                505,
+                                642
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        631,
+                                        505,
+                                        642
+                                    ],
+                                    "score": 1.0,
+                                    "content": "aligned with the goal of fewer collisions, whereas the news manipulation results in changes that are",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                641,
+                                463,
+                                653
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        641,
+                                        463,
+                                        653
+                                    ],
+                                    "score": 1.0,
+                                    "content": "misaligned with the goal of better predicting existing users’ interests (Leike et al., 2018).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        }
+                    ],
+                    "index": 34.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        658,
+                        505,
+                        735
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                658,
+                                506,
+                                672
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        658,
+                                        506,
+                                        672
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In reinforcement learning (RL), ADS are typically encouraged as a means to increase performance.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                669,
+                                506,
+                                683
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        669,
+                                        506,
+                                        683
+                                    ],
+                                    "score": 1.0,
+                                    "content": "On the other hand, in supervised learning (SL), the i.i.d. assumption precludes ADS in theory. In",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                681,
+                                506,
+                                693
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        681,
+                                        506,
+                                        693
+                                    ],
+                                    "score": 1.0,
+                                    "content": "practice, however, the possibility of using ADS to increase performance (and thus an incentive to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                691,
+                                504,
+                                704
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        691,
+                                        504,
+                                        704
+                                    ],
+                                    "score": 1.0,
+                                    "content": "do so) often remains. For instance, this occurs in online learning. In our experiments, we explicitly",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                701,
+                                505,
+                                716
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        701,
+                                        505,
+                                        716
+                                    ],
+                                    "score": 1.0,
+                                    "content": "model such situations where i.i.d. assumptions are violated: We study the behavior of SL and myopic",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                713,
+                                506,
+                                726
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        713,
+                                        506,
+                                        726
+                                    ],
+                                    "score": 1.0,
+                                    "content": "RL algorithms, in environments designed to include incentives for ADS, in order to understand when",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                724,
+                                460,
+                                737
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        724,
+                                        460,
+                                        737
+                                    ],
+                                    "score": 1.0,
+                                    "content": "incentives are effectively hidden. Fig. 2 contrasts these settings with typical RL and SL.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        }
+                    ],
+                    "index": 41
+                }
+            ],
+            "page_idx": 2,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        302,
+                        751,
+                        309,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                301,
+                                750,
+                                310,
+                                762
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        301,
+                                        750,
+                                        310,
+                                        762
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        107,
+                        27,
+                        307,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                26,
+                                308,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        26,
+                                        308,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Under review as a conference paper at ICLR 2021",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        82,
+                        505,
+                        105
+                    ],
+                    "lines": [],
+                    "index": 0.5,
+                    "bbox_fs": [
+                        105,
+                        81,
+                        505,
+                        105
+                    ],
+                    "lines_deleted": true
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        117,
+                        382,
+                        127
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                115,
+                                384,
+                                129
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        115,
+                                        384,
+                                        129
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2.2 DISTRIBUTIONAL SHIFT AND CONTENT RECOMMENDATION",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        }
+                    ],
+                    "index": 2
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        131,
+                        506,
+                        297
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                132,
+                                506,
+                                144
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        132,
+                                        506,
+                                        144
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In general, distributional shift refers to change of the data distribution over time. In supervised",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                142,
+                                506,
+                                156
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        142,
+                                        182,
+                                        156
+                                    ],
+                                    "score": 1.0,
+                                    "content": "learning with data",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        183,
+                                        145,
+                                        191,
+                                        153
+                                    ],
+                                    "score": 0.39,
+                                    "content": "\\mathbf { x }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        191,
+                                        142,
+                                        235,
+                                        156
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and labels",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        236,
+                                        145,
+                                        242,
+                                        154
+                                    ],
+                                    "score": 0.73,
+                                    "content": "y",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        243,
+                                        142,
+                                        506,
+                                        156
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", this can be more specifically described as dataset shift: change",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                154,
+                                506,
+                                166
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        154,
+                                        214,
+                                        166
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in the joint distribution of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        215,
+                                        154,
+                                        247,
+                                        165
+                                    ],
+                                    "score": 0.92,
+                                    "content": "P ( \\mathbf { x } , y )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        247,
+                                        154,
+                                        506,
+                                        166
+                                    ],
+                                    "score": 1.0,
+                                    "content": "between the training and test sets (Moreno-Torres et al., 2012;",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                164,
+                                506,
+                                177
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        164,
+                                        506,
+                                        177
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Quionero-Candela et al., 2009). As identified by Moreno-Torres et al. (2012), two common kinds",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                176,
+                                506,
+                                188
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        176,
+                                        281,
+                                        188
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of shift are: (1) Covariate shift: changing",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        282,
+                                        176,
+                                        304,
+                                        188
+                                    ],
+                                    "score": 0.91,
+                                    "content": "P ( \\mathbf { x } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        305,
+                                        176,
+                                        506,
+                                        188
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". In the example of content recommendation, this",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                186,
+                                506,
+                                199
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        186,
+                                        506,
+                                        199
+                                    ],
+                                    "score": 1.0,
+                                    "content": "corresponds to changing the user base of the recommendation system. For instance, a media outlet",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                196,
+                                506,
+                                211
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        196,
+                                        506,
+                                        211
+                                    ],
+                                    "score": 1.0,
+                                    "content": "which publishes inflammatory content may appeal to users with extreme views while alienating more",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                208,
+                                506,
+                                222
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        208,
+                                        506,
+                                        222
+                                    ],
+                                    "score": 1.0,
+                                    "content": "moderate users. This self-selection effect (Kayhan, 2015) may appear to a recommendation system",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                220,
+                                506,
+                                232
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        220,
+                                        506,
+                                        232
+                                    ],
+                                    "score": 1.0,
+                                    "content": "as an increase in performance, leading to a feedback effect, as previously noted by Shah et al. (2018).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                231,
+                                505,
+                                242
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        231,
+                                        505,
+                                        242
+                                    ],
+                                    "score": 1.0,
+                                    "content": "This type of feedback effect has been identified as contributing to filter bubbles and radicalization",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                241,
+                                506,
+                                254
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        242,
+                                        360,
+                                        254
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(Pariser, 2011; Kayhan, 2015). (2) Concept shift: changing",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        360,
+                                        241,
+                                        390,
+                                        253
+                                    ],
+                                    "score": 0.93,
+                                    "content": "P ( \\boldsymbol { y } | \\mathbf { x } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        391,
+                                        242,
+                                        506,
+                                        254
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". In the example of content",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                252,
+                                507,
+                                266
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        252,
+                                        507,
+                                        266
+                                    ],
+                                    "score": 1.0,
+                                    "content": "recommendation, this corresponds to changing a given user’s interest in different kinds of content.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                263,
+                                505,
+                                275
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        263,
+                                        505,
+                                        275
+                                    ],
+                                    "score": 1.0,
+                                    "content": "For example, exposure to a fake news story has been shown to increase the perceived accuracy of (and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                275,
+                                506,
+                                287
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        275,
+                                        506,
+                                        287
+                                    ],
+                                    "score": 1.0,
+                                    "content": "thus presumably future interest in) the content, an example of the illusory truth effect (Pennycook",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                285,
+                                478,
+                                298
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        285,
+                                        478,
+                                        298
+                                    ],
+                                    "score": 1.0,
+                                    "content": "et al., 2019). For further details on such effects in content recommendation, see Appendix 8.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        }
+                    ],
+                    "index": 10,
+                    "bbox_fs": [
+                        105,
+                        132,
+                        507,
+                        298
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        310,
+                        358,
+                        323
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                309,
+                                360,
+                                325
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        309,
+                                        360,
+                                        325
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3 AUTO-INDUCED DISTRIBUTION SHIFT (ADS)",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        }
+                    ],
+                    "index": 18
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        331,
+                        506,
+                        375
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                330,
+                                506,
+                                344
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        330,
+                                        506,
+                                        344
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Auto-induced distribution shift (ADS) is distributional shift caused by an algorithm’s behaviour. This",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                340,
+                                507,
+                                357
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        340,
+                                        507,
+                                        357
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is in contrast to distributional shift which would happen even if the learner were not present - e.g.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                353,
+                                505,
+                                366
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        353,
+                                        505,
+                                        366
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for a crash prediction algorithm trained on data from the summer, encountering snowy roads is an",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                364,
+                                424,
+                                377
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        364,
+                                        424,
+                                        377
+                                    ],
+                                    "score": 1.0,
+                                    "content": "example of distributional shift, but not auto-induced distributional shift (ADS).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        }
+                    ],
+                    "index": 20.5,
+                    "bbox_fs": [
+                        104,
+                        330,
+                        507,
+                        377
+                    ]
+                },
+                {
+                    "type": "image",
+                    "bbox": [
+                        106,
+                        391,
+                        506,
+                        496
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                106,
+                                391,
+                                506,
+                                496
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        391,
+                                        506,
+                                        496
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                391,
+                                                506,
+                                                496
+                                            ],
+                                            "score": 0.802,
+                                            "type": "image",
+                                            "image_path": "e073c2dfa79ed13d098885e0aecd106a567f60dc458af17cb1a90e8412bd2143.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 24,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        391,
+                                        506,
+                                        426.0
+                                    ],
+                                    "spans": [],
+                                    "index": 23
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        426.0,
+                                        506,
+                                        461.0
+                                    ],
+                                    "spans": [],
+                                    "index": 24
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        461.0,
+                                        506,
+                                        496.0
+                                    ],
+                                    "spans": [],
+                                    "index": 25
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 24
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        506,
+                        505,
+                        572
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                506,
+                                506,
+                                519
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        506,
+                                        429,
+                                        519
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Figure 2: The widely studied problems of reinforcement learning (RL) with state",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        430,
+                                        509,
+                                        435,
+                                        516
+                                    ],
+                                    "score": 0.63,
+                                    "content": "s",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        436,
+                                        506,
+                                        465,
+                                        519
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", action",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        466,
+                                        509,
+                                        472,
+                                        516
+                                    ],
+                                    "score": 0.66,
+                                    "content": "a",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        472,
+                                        506,
+                                        506,
+                                        519
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", reward",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                518,
+                                505,
+                                529
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        519,
+                                        113,
+                                        527
+                                    ],
+                                    "score": 0.64,
+                                    "content": "r",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        113,
+                                        518,
+                                        325,
+                                        529
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tuples, and i.i.d. supervised learning (SL) with inputs",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        325,
+                                        519,
+                                        332,
+                                        527
+                                    ],
+                                    "score": 0.74,
+                                    "content": "x",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        332,
+                                        518,
+                                        381,
+                                        529
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", predictions",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        381,
+                                        518,
+                                        388,
+                                        529
+                                    ],
+                                    "score": 0.83,
+                                    "content": "\\hat { y }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        389,
+                                        518,
+                                        423,
+                                        529
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and loss",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        424,
+                                        518,
+                                        428,
+                                        527
+                                    ],
+                                    "score": 0.44,
+                                    "content": "l",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        429,
+                                        518,
+                                        505,
+                                        529
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(a,d) are free from",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                528,
+                                506,
+                                541
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        528,
+                                        506,
+                                        541
+                                    ],
+                                    "score": 1.0,
+                                    "content": "incentive problems. We focus on cases where there are incentives present which the learner is not",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                540,
+                                504,
+                                551
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        540,
+                                        173,
+                                        551
+                                    ],
+                                    "score": 1.0,
+                                    "content": "meant to pursue",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        173,
+                                        540,
+                                        193,
+                                        550
+                                    ],
+                                    "score": 0.59,
+                                    "content": "^ { ( \\mathbf { b } , \\mathbf { c } ) }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        194,
+                                        540,
+                                        504,
+                                        551
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Lines show paths of influence. The learner may have incentives to influence",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                550,
+                                506,
+                                563
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        550,
+                                        258,
+                                        563
+                                    ],
+                                    "score": 1.0,
+                                    "content": "any nodes descending from its action,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        258,
+                                        551,
+                                        267,
+                                        560
+                                    ],
+                                    "score": 0.67,
+                                    "content": "A",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        267,
+                                        550,
+                                        325,
+                                        563
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", or prediction,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        326,
+                                        551,
+                                        333,
+                                        561
+                                    ],
+                                    "score": 0.81,
+                                    "content": "\\hat { y }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        333,
+                                        550,
+                                        506,
+                                        563
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Which incentives are undesirable (orange)",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                561,
+                                368,
+                                573
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        561,
+                                        368,
+                                        573
+                                    ],
+                                    "score": 1.0,
+                                    "content": "or desirable (cyan) for the learner to pursue is context-dependent.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        }
+                    ],
+                    "index": 28.5,
+                    "bbox_fs": [
+                        105,
+                        506,
+                        506,
+                        573
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        586,
+                        505,
+                        653
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                586,
+                                506,
+                                599
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        586,
+                                        506,
+                                        599
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We emphasize that ADS are not inherently bad or good; often ADS can even be desirable: consider",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                597,
+                                505,
+                                610
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        597,
+                                        505,
+                                        610
+                                    ],
+                                    "score": 1.0,
+                                    "content": "an algorithm meant to alert drivers of imminent collisions. If it works well, such a system will help",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                608,
+                                506,
+                                621
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        608,
+                                        506,
+                                        621
+                                    ],
+                                    "score": 1.0,
+                                    "content": "drivers avoid crashing, thus making self-refuting predictions which result in ADS. What separates",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                620,
+                                505,
+                                631
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        620,
+                                        505,
+                                        631
+                                    ],
+                                    "score": 1.0,
+                                    "content": "desirable and undesirable ADS? The collision-alert system alters its data distribution in a way that is",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                631,
+                                505,
+                                642
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        631,
+                                        505,
+                                        642
+                                    ],
+                                    "score": 1.0,
+                                    "content": "aligned with the goal of fewer collisions, whereas the news manipulation results in changes that are",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                641,
+                                463,
+                                653
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        641,
+                                        463,
+                                        653
+                                    ],
+                                    "score": 1.0,
+                                    "content": "misaligned with the goal of better predicting existing users’ interests (Leike et al., 2018).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        }
+                    ],
+                    "index": 34.5,
+                    "bbox_fs": [
+                        105,
+                        586,
+                        506,
+                        653
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        658,
+                        505,
+                        735
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                658,
+                                506,
+                                672
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        658,
+                                        506,
+                                        672
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In reinforcement learning (RL), ADS are typically encouraged as a means to increase performance.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                669,
+                                506,
+                                683
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        669,
+                                        506,
+                                        683
+                                    ],
+                                    "score": 1.0,
+                                    "content": "On the other hand, in supervised learning (SL), the i.i.d. assumption precludes ADS in theory. In",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                681,
+                                506,
+                                693
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        681,
+                                        506,
+                                        693
+                                    ],
+                                    "score": 1.0,
+                                    "content": "practice, however, the possibility of using ADS to increase performance (and thus an incentive to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                691,
+                                504,
+                                704
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        691,
+                                        504,
+                                        704
+                                    ],
+                                    "score": 1.0,
+                                    "content": "do so) often remains. For instance, this occurs in online learning. In our experiments, we explicitly",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                701,
+                                505,
+                                716
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        701,
+                                        505,
+                                        716
+                                    ],
+                                    "score": 1.0,
+                                    "content": "model such situations where i.i.d. assumptions are violated: We study the behavior of SL and myopic",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                713,
+                                506,
+                                726
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        713,
+                                        506,
+                                        726
+                                    ],
+                                    "score": 1.0,
+                                    "content": "RL algorithms, in environments designed to include incentives for ADS, in order to understand when",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                724,
+                                460,
+                                737
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        724,
+                                        460,
+                                        737
+                                    ],
+                                    "score": 1.0,
+                                    "content": "incentives are effectively hidden. Fig. 2 contrasts these settings with typical RL and SL.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        }
+                    ],
+                    "index": 41,
+                    "bbox_fs": [
+                        105,
+                        658,
+                        506,
+                        737
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        81,
+                        189,
+                        93
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                80,
+                                191,
+                                96
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        80,
+                                        191,
+                                        96
+                                    ],
+                                    "score": 1.0,
+                                    "content": "4 INCENTIVES",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        }
+                    ],
+                    "index": 0
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        102,
+                        505,
+                        180
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                103,
+                                505,
+                                115
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        103,
+                                        505,
+                                        115
+                                    ],
+                                    "score": 1.0,
+                                    "content": "For our study of incentives, we use the following terminology: an incentive for a behavior (e.g. an",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                115,
+                                505,
+                                126
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        115,
+                                        505,
+                                        126
+                                    ],
+                                    "score": 1.0,
+                                    "content": "action, a classification, etc.) is present (not absent) to the extent that the behaviour will increase",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                125,
+                                506,
+                                137
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        125,
+                                        506,
+                                        137
+                                    ],
+                                    "score": 1.0,
+                                    "content": "performance (e.g. reward, accuracy, etc.) (Everitt & Hutter, 2019). This incentive is revealed to (not",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                136,
+                                506,
+                                148
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        136,
+                                        506,
+                                        148
+                                    ],
+                                    "score": 1.0,
+                                    "content": "hidden from) a learner if it would, at higher than chance levels, learn to perform the behavior given",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                146,
+                                507,
+                                160
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        146,
+                                        507,
+                                        160
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sufficient capacity and training experience. The incentive is pursued (not eschewed) by a learner if",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                158,
+                                506,
+                                171
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        158,
+                                        506,
+                                        171
+                                    ],
+                                    "score": 1.0,
+                                    "content": "it actually performs the incentivized behaviour. Note even when an incentive is revealed, it may not",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                169,
+                                432,
+                                181
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        169,
+                                        432,
+                                        181
+                                    ],
+                                    "score": 1.0,
+                                    "content": "be pursued, e.g. due to limited capacity and/or data, or simply chance. See Fig 3.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        }
+                    ],
+                    "index": 4
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        186,
+                        317,
+                        284
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                185,
+                                318,
+                                197
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        185,
+                                        318,
+                                        197
+                                    ],
+                                    "score": 1.0,
+                                    "content": "For example, in content recommendation, the incen-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                196,
+                                317,
+                                209
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        196,
+                                        317,
+                                        209
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tive to drive users away is present if some user types",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                208,
+                                317,
+                                218
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        208,
+                                        317,
+                                        218
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are easier to predict than others. But this incentive",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                218,
+                                317,
+                                230
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        218,
+                                        317,
+                                        230
+                                    ],
+                                    "score": 1.0,
+                                    "content": "may be hidden from the learner by using a myopic",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                229,
+                                318,
+                                241
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        229,
+                                        318,
+                                        241
+                                    ],
+                                    "score": 1.0,
+                                    "content": "algorithm, e.g. one that does not see the effects of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                241,
+                                318,
+                                252
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        241,
+                                        318,
+                                        252
+                                    ],
+                                    "score": 1.0,
+                                    "content": "its actions on the distribution of users. The incen-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                251,
+                                317,
+                                263
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        251,
+                                        317,
+                                        263
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tive might instead be revealed to the outer loop of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                262,
+                                317,
+                                274
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        262,
+                                        317,
+                                        274
+                                    ],
+                                    "score": 1.0,
+                                    "content": "a meta-learning algorithm like PBT, which does see",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                272,
+                                231,
+                                284
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        272,
+                                        231,
+                                        284
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the effects of learner’s actions.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        }
+                    ],
+                    "index": 12
+                },
+                {
+                    "type": "image",
+                    "bbox": [
+                        326,
+                        186,
+                        503,
+                        249
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                326,
+                                186,
+                                503,
+                                249
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        326,
+                                        186,
+                                        503,
+                                        249
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                326,
+                                                186,
+                                                503,
+                                                249
+                                            ],
+                                            "score": 0.967,
+                                            "type": "image",
+                                            "image_path": "2ade91d4be27c22a6f418d3a986aa9448815d5a0ddb40aa939127276f08bc9bc.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 18.5,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        326,
+                                        186,
+                                        503,
+                                        201.75
+                                    ],
+                                    "spans": [],
+                                    "index": 17
+                                },
+                                {
+                                    "bbox": [
+                                        326,
+                                        201.75,
+                                        503,
+                                        217.5
+                                    ],
+                                    "spans": [],
+                                    "index": 18
+                                },
+                                {
+                                    "bbox": [
+                                        326,
+                                        217.5,
+                                        503,
+                                        233.25
+                                    ],
+                                    "spans": [],
+                                    "index": 19
+                                },
+                                {
+                                    "bbox": [
+                                        326,
+                                        233.25,
+                                        503,
+                                        249.0
+                                    ],
+                                    "spans": [],
+                                    "index": 20
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                324,
+                                260,
+                                505,
+                                282
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        324,
+                                        259,
+                                        506,
+                                        272
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                324,
+                                                259,
+                                                506,
+                                                272
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 3: Types of incentives, and their rela-",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 21
+                                },
+                                {
+                                    "bbox": [
+                                        324,
+                                        270,
+                                        394,
+                                        283
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                324,
+                                                270,
+                                                394,
+                                                283
+                                            ],
+                                            "score": 1.0,
+                                            "content": "tionship to ADS.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 22
+                                }
+                            ],
+                            "index": 21.5
+                        }
+                    ],
+                    "index": 20.0
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        290,
+                        505,
+                        367
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                290,
+                                505,
+                                303
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        290,
+                                        505,
+                                        303
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Even when this incentive is revealed, however, it might not end up being pursued. For example, this",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                300,
+                                506,
+                                315
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        300,
+                                        506,
+                                        315
+                                    ],
+                                    "score": 1.0,
+                                    "content": "could happen if predicting which recommendations will drive away users is too difficult a learning",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                312,
+                                505,
+                                325
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        312,
+                                        505,
+                                        325
+                                    ],
+                                    "score": 1.0,
+                                    "content": "problem, or if the incentive to do so is dominated by other incentives (e.g. change individual users’",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                324,
+                                505,
+                                336
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        324,
+                                        505,
+                                        336
+                                    ],
+                                    "score": 1.0,
+                                    "content": "interests, or improve accuracy of predictions). In general, it may be difficult to determine empirically",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                334,
+                                506,
+                                347
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        334,
+                                        506,
+                                        347
+                                    ],
+                                    "score": 1.0,
+                                    "content": "which incentives are revealed, because failure to pursue an incentive can be due to limited capacity,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                344,
+                                505,
+                                358
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        344,
+                                        505,
+                                        358
+                                    ],
+                                    "score": 1.0,
+                                    "content": "insufficient training, and/or random chance. To address this challenge, we devise extremely simple",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                356,
+                                488,
+                                369
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        356,
+                                        488,
+                                        369
+                                    ],
+                                    "score": 1.0,
+                                    "content": "environments (“unit tests”), where we can be confident that revealed incentives will be pursued.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        }
+                    ],
+                    "index": 26
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        106,
+                        379,
+                        462,
+                        390
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                378,
+                                464,
+                                393
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        378,
+                                        464,
+                                        393
+                                    ],
+                                    "score": 1.0,
+                                    "content": "4.1 HIDDEN INCENTIVES FOR AUTO-INDUCED DISTRIBUTIONAL SHIFT (HI-ADS)",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        }
+                    ],
+                    "index": 30
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        395,
+                        505,
+                        440
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                395,
+                                505,
+                                408
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        395,
+                                        505,
+                                        408
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Following from the definitions in Sections 3 and 4, HI-ADS are incentives for behaviors that cause",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                407,
+                                505,
+                                419
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        407,
+                                        505,
+                                        419
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Auto-induced Distributional Shift that are hidden from the learner, i.e. the learner would not learn",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                418,
+                                506,
+                                431
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        418,
+                                        506,
+                                        431
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to perform the incentivized behaviors at higher than chance levels, even given infinite capacity and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                430,
+                                189,
+                                441
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        430,
+                                        189,
+                                        441
+                                    ],
+                                    "score": 1.0,
+                                    "content": "training experience.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        }
+                    ],
+                    "index": 32.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        446,
+                        505,
+                        512
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                445,
+                                505,
+                                458
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        445,
+                                        505,
+                                        458
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Like ADS, HI-ADS are not necessarily problematic. Indeed, hiding incentives can be an effective",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                456,
+                                505,
+                                469
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        456,
+                                        505,
+                                        469
+                                    ],
+                                    "score": 1.0,
+                                    "content": "method of influencing learner behavior. For example, hiding the incentive to manipulate users from",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                467,
+                                506,
+                                480
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        467,
+                                        506,
+                                        480
+                                    ],
+                                    "score": 1.0,
+                                    "content": "a content recommendation algorithm could prevent it from influencing users in a way they would",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                478,
+                                506,
+                                492
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        478,
+                                        506,
+                                        492
+                                    ],
+                                    "score": 1.0,
+                                    "content": "not endorse. However, if machine learning practitioners are not aware that incentives are present, or",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                489,
+                                505,
+                                502
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        489,
+                                        505,
+                                        502
+                                    ],
+                                    "score": 1.0,
+                                    "content": "that properties of the learning algorithm are hiding them, then seemingly innocuous changes to the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                500,
+                                489,
+                                513
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        500,
+                                        489,
+                                        513
+                                    ],
+                                    "score": 1.0,
+                                    "content": "learning algorithm may reveal HI-ADS, and lead to significant unexpected changes in behavior.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        }
+                    ],
+                    "index": 37.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        517,
+                        505,
+                        583
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                516,
+                                506,
+                                530
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        516,
+                                        506,
+                                        530
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Hiding incentives for ADS may seem counter-intuitive and counter-productive in the context of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                529,
+                                506,
+                                540
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        529,
+                                        506,
+                                        540
+                                    ],
+                                    "score": 1.0,
+                                    "content": "reinforcement learning (RL), where moving towards high-reward states is typically desirable. How-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                540,
+                                507,
+                                552
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        540,
+                                        507,
+                                        552
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ever, for real-world applications of RL, the ultimate goal is not a system that achieves high reward,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                550,
+                                505,
+                                563
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        550,
+                                        505,
+                                        563
+                                    ],
+                                    "score": 1.0,
+                                    "content": "but rather one that behaves according to the designer’s intentions. And as we discussed in the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                560,
+                                507,
+                                574
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        560,
+                                        507,
+                                        574
+                                    ],
+                                    "score": 1.0,
+                                    "content": "introduction, it can be intractable to design reward functions that perfectly specify intended behavior.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                572,
+                                507,
+                                585
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        572,
+                                        507,
+                                        585
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Thus managing (e.g. hiding) some incentives can provide a useful tool for specification, even in RL.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        }
+                    ],
+                    "index": 43.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        588,
+                        505,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                588,
+                                506,
+                                601
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        588,
+                                        506,
+                                        601
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We have several reasons for focusing on HI-ADS: (1) The issue of HI-ADS has not yet been identified,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 47
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                600,
+                                505,
+                                613
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        600,
+                                        505,
+                                        613
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and thus is liable to be neglected in practice. Indeed, our “unit tests” are the first published empirical",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 48
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                611,
+                                506,
+                                624
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        611,
+                                        506,
+                                        624
+                                    ],
+                                    "score": 1.0,
+                                    "content": "methodology for assessing whether incentives are hidden or revealed by different learning algorithms.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 49
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                621,
+                                506,
+                                636
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        621,
+                                        506,
+                                        636
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(2) Machine learning algorithms are commonly deployed in settings where ADS are present, violating",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 50
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                633,
+                                505,
+                                646
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        633,
+                                        505,
+                                        646
+                                    ],
+                                    "score": 1.0,
+                                    "content": "assumptions used to analyze their properties theoretically. This means learners could exploit ADS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 51
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                644,
+                                505,
+                                657
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        644,
+                                        505,
+                                        657
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in unexpected and undesirable ways if incentives for ADS are revealed. Hiding these incentives",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 52
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                654,
+                                506,
+                                668
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        654,
+                                        506,
+                                        668
+                                    ],
+                                    "score": 1.0,
+                                    "content": "heuristically (e.g. via off-line training) is a common approach, but potentially brittle (if practitioners",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 53
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                665,
+                                506,
+                                679
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        665,
+                                        506,
+                                        679
+                                    ],
+                                    "score": 1.0,
+                                    "content": "don’t understand how HI-ADS could become revealed). In particular, meta-learning can reveal",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 54
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                676,
+                                506,
+                                690
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        676,
+                                        506,
+                                        690
+                                    ],
+                                    "score": 1.0,
+                                    "content": "HI-ADS in online learning settings. (3) Substantial real-world issues could result from improper",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 55
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                687,
+                                506,
+                                700
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        687,
+                                        506,
+                                        700
+                                    ],
+                                    "score": 1.0,
+                                    "content": "management of learner’s incentives. Examples include tampering with human-generated reward",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 56
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                699,
+                                505,
+                                712
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        699,
+                                        505,
+                                        712
+                                    ],
+                                    "score": 1.0,
+                                    "content": "signals (Everitt & Hutter, 2018) (e.g. selecting news articles which manipulate user interests), and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 57
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                709,
+                                505,
+                                723
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        709,
+                                        505,
+                                        723
+                                    ],
+                                    "score": 1.0,
+                                    "content": "creating “self-fulfilling prophecies” (e.g. driving up the value of an asset by publicly predicting its",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 58
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                720,
+                                313,
+                                732
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        720,
+                                        313,
+                                        732
+                                    ],
+                                    "score": 1.0,
+                                    "content": "value will increase (Armstrong & O’Rorke, 2017)).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 59
+                        }
+                    ],
+                    "index": 53
+                }
+            ],
+            "page_idx": 3,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        107,
+                        27,
+                        307,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                26,
+                                308,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        26,
+                                        308,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Under review as a conference paper at ICLR 2021",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        302,
+                        752,
+                        308,
+                        759
+                    ],
+                    "lines": []
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        81,
+                        189,
+                        93
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                80,
+                                191,
+                                96
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        80,
+                                        191,
+                                        96
+                                    ],
+                                    "score": 1.0,
+                                    "content": "4 INCENTIVES",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        }
+                    ],
+                    "index": 0
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        102,
+                        505,
+                        180
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                103,
+                                505,
+                                115
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        103,
+                                        505,
+                                        115
+                                    ],
+                                    "score": 1.0,
+                                    "content": "For our study of incentives, we use the following terminology: an incentive for a behavior (e.g. an",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                115,
+                                505,
+                                126
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        115,
+                                        505,
+                                        126
+                                    ],
+                                    "score": 1.0,
+                                    "content": "action, a classification, etc.) is present (not absent) to the extent that the behaviour will increase",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                125,
+                                506,
+                                137
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        125,
+                                        506,
+                                        137
+                                    ],
+                                    "score": 1.0,
+                                    "content": "performance (e.g. reward, accuracy, etc.) (Everitt & Hutter, 2019). This incentive is revealed to (not",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                136,
+                                506,
+                                148
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        136,
+                                        506,
+                                        148
+                                    ],
+                                    "score": 1.0,
+                                    "content": "hidden from) a learner if it would, at higher than chance levels, learn to perform the behavior given",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                146,
+                                507,
+                                160
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        146,
+                                        507,
+                                        160
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sufficient capacity and training experience. The incentive is pursued (not eschewed) by a learner if",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                158,
+                                506,
+                                171
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        158,
+                                        506,
+                                        171
+                                    ],
+                                    "score": 1.0,
+                                    "content": "it actually performs the incentivized behaviour. Note even when an incentive is revealed, it may not",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                169,
+                                432,
+                                181
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        169,
+                                        432,
+                                        181
+                                    ],
+                                    "score": 1.0,
+                                    "content": "be pursued, e.g. due to limited capacity and/or data, or simply chance. See Fig 3.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        }
+                    ],
+                    "index": 4,
+                    "bbox_fs": [
+                        105,
+                        103,
+                        507,
+                        181
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        186,
+                        317,
+                        284
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                185,
+                                318,
+                                197
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        185,
+                                        318,
+                                        197
+                                    ],
+                                    "score": 1.0,
+                                    "content": "For example, in content recommendation, the incen-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                196,
+                                317,
+                                209
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        196,
+                                        317,
+                                        209
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tive to drive users away is present if some user types",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                208,
+                                317,
+                                218
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        208,
+                                        317,
+                                        218
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are easier to predict than others. But this incentive",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                218,
+                                317,
+                                230
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        218,
+                                        317,
+                                        230
+                                    ],
+                                    "score": 1.0,
+                                    "content": "may be hidden from the learner by using a myopic",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                229,
+                                318,
+                                241
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        229,
+                                        318,
+                                        241
+                                    ],
+                                    "score": 1.0,
+                                    "content": "algorithm, e.g. one that does not see the effects of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                241,
+                                318,
+                                252
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        241,
+                                        318,
+                                        252
+                                    ],
+                                    "score": 1.0,
+                                    "content": "its actions on the distribution of users. The incen-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                251,
+                                317,
+                                263
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        251,
+                                        317,
+                                        263
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tive might instead be revealed to the outer loop of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                262,
+                                317,
+                                274
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        262,
+                                        317,
+                                        274
+                                    ],
+                                    "score": 1.0,
+                                    "content": "a meta-learning algorithm like PBT, which does see",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                272,
+                                231,
+                                284
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        272,
+                                        231,
+                                        284
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the effects of learner’s actions.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        }
+                    ],
+                    "index": 12,
+                    "bbox_fs": [
+                        105,
+                        185,
+                        318,
+                        284
+                    ]
+                },
+                {
+                    "type": "image",
+                    "bbox": [
+                        326,
+                        186,
+                        503,
+                        249
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                326,
+                                186,
+                                503,
+                                249
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        326,
+                                        186,
+                                        503,
+                                        249
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                326,
+                                                186,
+                                                503,
+                                                249
+                                            ],
+                                            "score": 0.967,
+                                            "type": "image",
+                                            "image_path": "2ade91d4be27c22a6f418d3a986aa9448815d5a0ddb40aa939127276f08bc9bc.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 18.5,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        326,
+                                        186,
+                                        503,
+                                        201.75
+                                    ],
+                                    "spans": [],
+                                    "index": 17
+                                },
+                                {
+                                    "bbox": [
+                                        326,
+                                        201.75,
+                                        503,
+                                        217.5
+                                    ],
+                                    "spans": [],
+                                    "index": 18
+                                },
+                                {
+                                    "bbox": [
+                                        326,
+                                        217.5,
+                                        503,
+                                        233.25
+                                    ],
+                                    "spans": [],
+                                    "index": 19
+                                },
+                                {
+                                    "bbox": [
+                                        326,
+                                        233.25,
+                                        503,
+                                        249.0
+                                    ],
+                                    "spans": [],
+                                    "index": 20
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                324,
+                                260,
+                                505,
+                                282
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        324,
+                                        259,
+                                        506,
+                                        272
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                324,
+                                                259,
+                                                506,
+                                                272
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 3: Types of incentives, and their rela-",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 21
+                                },
+                                {
+                                    "bbox": [
+                                        324,
+                                        270,
+                                        394,
+                                        283
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                324,
+                                                270,
+                                                394,
+                                                283
+                                            ],
+                                            "score": 1.0,
+                                            "content": "tionship to ADS.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 22
+                                }
+                            ],
+                            "index": 21.5
+                        }
+                    ],
+                    "index": 20.0
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        290,
+                        505,
+                        367
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                290,
+                                505,
+                                303
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        290,
+                                        505,
+                                        303
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Even when this incentive is revealed, however, it might not end up being pursued. For example, this",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                300,
+                                506,
+                                315
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        300,
+                                        506,
+                                        315
+                                    ],
+                                    "score": 1.0,
+                                    "content": "could happen if predicting which recommendations will drive away users is too difficult a learning",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                312,
+                                505,
+                                325
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        312,
+                                        505,
+                                        325
+                                    ],
+                                    "score": 1.0,
+                                    "content": "problem, or if the incentive to do so is dominated by other incentives (e.g. change individual users’",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                324,
+                                505,
+                                336
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        324,
+                                        505,
+                                        336
+                                    ],
+                                    "score": 1.0,
+                                    "content": "interests, or improve accuracy of predictions). In general, it may be difficult to determine empirically",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                334,
+                                506,
+                                347
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        334,
+                                        506,
+                                        347
+                                    ],
+                                    "score": 1.0,
+                                    "content": "which incentives are revealed, because failure to pursue an incentive can be due to limited capacity,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                344,
+                                505,
+                                358
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        344,
+                                        505,
+                                        358
+                                    ],
+                                    "score": 1.0,
+                                    "content": "insufficient training, and/or random chance. To address this challenge, we devise extremely simple",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                356,
+                                488,
+                                369
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        356,
+                                        488,
+                                        369
+                                    ],
+                                    "score": 1.0,
+                                    "content": "environments (“unit tests”), where we can be confident that revealed incentives will be pursued.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        }
+                    ],
+                    "index": 26,
+                    "bbox_fs": [
+                        105,
+                        290,
+                        506,
+                        369
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        106,
+                        379,
+                        462,
+                        390
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                378,
+                                464,
+                                393
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        378,
+                                        464,
+                                        393
+                                    ],
+                                    "score": 1.0,
+                                    "content": "4.1 HIDDEN INCENTIVES FOR AUTO-INDUCED DISTRIBUTIONAL SHIFT (HI-ADS)",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        }
+                    ],
+                    "index": 30
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        395,
+                        505,
+                        440
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                395,
+                                505,
+                                408
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        395,
+                                        505,
+                                        408
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Following from the definitions in Sections 3 and 4, HI-ADS are incentives for behaviors that cause",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                407,
+                                505,
+                                419
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        407,
+                                        505,
+                                        419
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Auto-induced Distributional Shift that are hidden from the learner, i.e. the learner would not learn",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                418,
+                                506,
+                                431
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        418,
+                                        506,
+                                        431
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to perform the incentivized behaviors at higher than chance levels, even given infinite capacity and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                430,
+                                189,
+                                441
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        430,
+                                        189,
+                                        441
+                                    ],
+                                    "score": 1.0,
+                                    "content": "training experience.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        }
+                    ],
+                    "index": 32.5,
+                    "bbox_fs": [
+                        104,
+                        395,
+                        506,
+                        441
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        446,
+                        505,
+                        512
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                445,
+                                505,
+                                458
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        445,
+                                        505,
+                                        458
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Like ADS, HI-ADS are not necessarily problematic. Indeed, hiding incentives can be an effective",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                456,
+                                505,
+                                469
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        456,
+                                        505,
+                                        469
+                                    ],
+                                    "score": 1.0,
+                                    "content": "method of influencing learner behavior. For example, hiding the incentive to manipulate users from",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                467,
+                                506,
+                                480
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        467,
+                                        506,
+                                        480
+                                    ],
+                                    "score": 1.0,
+                                    "content": "a content recommendation algorithm could prevent it from influencing users in a way they would",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                478,
+                                506,
+                                492
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        478,
+                                        506,
+                                        492
+                                    ],
+                                    "score": 1.0,
+                                    "content": "not endorse. However, if machine learning practitioners are not aware that incentives are present, or",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                489,
+                                505,
+                                502
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        489,
+                                        505,
+                                        502
+                                    ],
+                                    "score": 1.0,
+                                    "content": "that properties of the learning algorithm are hiding them, then seemingly innocuous changes to the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                500,
+                                489,
+                                513
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        500,
+                                        489,
+                                        513
+                                    ],
+                                    "score": 1.0,
+                                    "content": "learning algorithm may reveal HI-ADS, and lead to significant unexpected changes in behavior.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        }
+                    ],
+                    "index": 37.5,
+                    "bbox_fs": [
+                        105,
+                        445,
+                        506,
+                        513
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        517,
+                        505,
+                        583
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                516,
+                                506,
+                                530
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        516,
+                                        506,
+                                        530
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Hiding incentives for ADS may seem counter-intuitive and counter-productive in the context of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                529,
+                                506,
+                                540
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        529,
+                                        506,
+                                        540
+                                    ],
+                                    "score": 1.0,
+                                    "content": "reinforcement learning (RL), where moving towards high-reward states is typically desirable. How-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                540,
+                                507,
+                                552
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        540,
+                                        507,
+                                        552
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ever, for real-world applications of RL, the ultimate goal is not a system that achieves high reward,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                550,
+                                505,
+                                563
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        550,
+                                        505,
+                                        563
+                                    ],
+                                    "score": 1.0,
+                                    "content": "but rather one that behaves according to the designer’s intentions. And as we discussed in the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                560,
+                                507,
+                                574
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        560,
+                                        507,
+                                        574
+                                    ],
+                                    "score": 1.0,
+                                    "content": "introduction, it can be intractable to design reward functions that perfectly specify intended behavior.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                572,
+                                507,
+                                585
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        572,
+                                        507,
+                                        585
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Thus managing (e.g. hiding) some incentives can provide a useful tool for specification, even in RL.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        }
+                    ],
+                    "index": 43.5,
+                    "bbox_fs": [
+                        105,
+                        516,
+                        507,
+                        585
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        588,
+                        505,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                588,
+                                506,
+                                601
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        588,
+                                        506,
+                                        601
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We have several reasons for focusing on HI-ADS: (1) The issue of HI-ADS has not yet been identified,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 47
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                600,
+                                505,
+                                613
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        600,
+                                        505,
+                                        613
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and thus is liable to be neglected in practice. Indeed, our “unit tests” are the first published empirical",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 48
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                611,
+                                506,
+                                624
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        611,
+                                        506,
+                                        624
+                                    ],
+                                    "score": 1.0,
+                                    "content": "methodology for assessing whether incentives are hidden or revealed by different learning algorithms.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 49
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                621,
+                                506,
+                                636
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        621,
+                                        506,
+                                        636
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(2) Machine learning algorithms are commonly deployed in settings where ADS are present, violating",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 50
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                633,
+                                505,
+                                646
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        633,
+                                        505,
+                                        646
+                                    ],
+                                    "score": 1.0,
+                                    "content": "assumptions used to analyze their properties theoretically. This means learners could exploit ADS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 51
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                644,
+                                505,
+                                657
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        644,
+                                        505,
+                                        657
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in unexpected and undesirable ways if incentives for ADS are revealed. Hiding these incentives",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 52
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                654,
+                                506,
+                                668
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        654,
+                                        506,
+                                        668
+                                    ],
+                                    "score": 1.0,
+                                    "content": "heuristically (e.g. via off-line training) is a common approach, but potentially brittle (if practitioners",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 53
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                665,
+                                506,
+                                679
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        665,
+                                        506,
+                                        679
+                                    ],
+                                    "score": 1.0,
+                                    "content": "don’t understand how HI-ADS could become revealed). In particular, meta-learning can reveal",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 54
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                676,
+                                506,
+                                690
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        676,
+                                        506,
+                                        690
+                                    ],
+                                    "score": 1.0,
+                                    "content": "HI-ADS in online learning settings. (3) Substantial real-world issues could result from improper",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 55
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                687,
+                                506,
+                                700
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        687,
+                                        506,
+                                        700
+                                    ],
+                                    "score": 1.0,
+                                    "content": "management of learner’s incentives. Examples include tampering with human-generated reward",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 56
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                699,
+                                505,
+                                712
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        699,
+                                        505,
+                                        712
+                                    ],
+                                    "score": 1.0,
+                                    "content": "signals (Everitt & Hutter, 2018) (e.g. selecting news articles which manipulate user interests), and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 57
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                709,
+                                505,
+                                723
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        709,
+                                        505,
+                                        723
+                                    ],
+                                    "score": 1.0,
+                                    "content": "creating “self-fulfilling prophecies” (e.g. driving up the value of an asset by publicly predicting its",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 58
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                720,
+                                313,
+                                732
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        720,
+                                        313,
+                                        732
+                                    ],
+                                    "score": 1.0,
+                                    "content": "value will increase (Armstrong & O’Rorke, 2017)).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 59
+                        }
+                    ],
+                    "index": 53,
+                    "bbox_fs": [
+                        105,
+                        588,
+                        506,
+                        732
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "title",
+                    "bbox": [
+                        106,
+                        82,
+                        329,
+                        93
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                82,
+                                330,
+                                95
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        82,
+                                        330,
+                                        95
+                                    ],
+                                    "score": 1.0,
+                                    "content": "4.2 REMOVING HI-ADS VIA CONTEXT SWAPPING",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        }
+                    ],
+                    "index": 0
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        97,
+                        505,
+                        152
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                97,
+                                505,
+                                110
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        97,
+                                        505,
+                                        110
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We propose a technique called context swapping for removing incentives for ADS. The technique",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                108,
+                                506,
+                                121
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        108,
+                                        132,
+                                        121
+                                    ],
+                                    "score": 1.0,
+                                    "content": "trains",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        132,
+                                        109,
+                                        142,
+                                        118
+                                    ],
+                                    "score": 0.81,
+                                    "content": "N",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        142,
+                                        108,
+                                        362,
+                                        121
+                                    ],
+                                    "score": 1.0,
+                                    "content": "learners in parallel, and shuffles the learners through",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        362,
+                                        109,
+                                        372,
+                                        118
+                                    ],
+                                    "score": 0.71,
+                                    "content": "N",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        373,
+                                        108,
+                                        506,
+                                        121
+                                    ],
+                                    "score": 1.0,
+                                    "content": "different copies of the same (or",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                119,
+                                505,
+                                132
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        119,
+                                        505,
+                                        132
+                                    ],
+                                    "score": 1.0,
+                                    "content": "similar) environments.We use a deterministic permutation of learners in environment copies, so",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                130,
+                                506,
+                                143
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        130,
+                                        141,
+                                        143
+                                    ],
+                                    "score": 1.0,
+                                    "content": "that the",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        141,
+                                        131,
+                                        146,
+                                        140
+                                    ],
+                                    "score": 0.75,
+                                    "content": "i",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        146,
+                                        130,
+                                        243,
+                                        143
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-th learner inhabits the",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        243,
+                                        131,
+                                        249,
+                                        142
+                                    ],
+                                    "score": 0.81,
+                                    "content": "j",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        250,
+                                        130,
+                                        376,
+                                        143
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-th environment on time-steps",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        376,
+                                        131,
+                                        382,
+                                        140
+                                    ],
+                                    "score": 0.62,
+                                    "content": "t",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        382,
+                                        130,
+                                        412,
+                                        143
+                                    ],
+                                    "score": 1.0,
+                                    "content": "where",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        412,
+                                        130,
+                                        464,
+                                        142
+                                    ],
+                                    "score": 0.92,
+                                    "content": "~ j ~ = ~ ( i + t )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        465,
+                                        130,
+                                        492,
+                                        143
+                                    ],
+                                    "score": 1.0,
+                                    "content": "mod",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        492,
+                                        131,
+                                        502,
+                                        141
+                                    ],
+                                    "score": 0.71,
+                                    "content": "N",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        503,
+                                        130,
+                                        506,
+                                        143
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                141,
+                                506,
+                                153
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        141,
+                                        506,
+                                        153
+                                    ],
+                                    "score": 1.0,
+                                    "content": "makes an observation, takes an action, and receives a reward before moving to the next environment.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        }
+                    ],
+                    "index": 3
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        153,
+                        317,
+                        218
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                151,
+                                318,
+                                164
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        151,
+                                        133,
+                                        164
+                                    ],
+                                    "score": 1.0,
+                                    "content": "When",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        134,
+                                        153,
+                                        144,
+                                        162
+                                    ],
+                                    "score": 0.79,
+                                    "content": "N",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        145,
+                                        151,
+                                        318,
+                                        164
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is larger than the interval of the OL opti-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                163,
+                                317,
+                                174
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        163,
+                                        317,
+                                        174
+                                    ],
+                                    "score": 1.0,
+                                    "content": "mizer, each learner inhabits each copy for at most a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                174,
+                                317,
+                                186
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        174,
+                                        317,
+                                        186
+                                    ],
+                                    "score": 1.0,
+                                    "content": "single time-step before an OL step is applied. Under",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                185,
+                                318,
+                                196
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        185,
+                                        318,
+                                        196
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the assumption that different copies of the environ-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                196,
+                                317,
+                                208
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        196,
+                                        317,
+                                        208
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ment do not influence each other, this technique can",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                207,
+                                318,
+                                219
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        207,
+                                        318,
+                                        219
+                                    ],
+                                    "score": 1.0,
+                                    "content": "address HI-ADS in practice, as we show in Sec. 5.1.1.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        }
+                    ],
+                    "index": 8.5
+                },
+                {
+                    "type": "image",
+                    "bbox": [
+                        327,
+                        162,
+                        502,
+                        200
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                327,
+                                162,
+                                502,
+                                200
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        327,
+                                        162,
+                                        502,
+                                        200
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                327,
+                                                162,
+                                                502,
+                                                200
+                                            ],
+                                            "score": 0.948,
+                                            "type": "image",
+                                            "image_path": "06fd14fc3c3117d01ce92cc59924cf8d1fe7f705cf8d340291c9e7534c747cd1.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 12.5,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        327,
+                                        162,
+                                        502,
+                                        181.0
+                                    ],
+                                    "spans": [],
+                                    "index": 12
+                                },
+                                {
+                                    "bbox": [
+                                        327,
+                                        181.0,
+                                        502,
+                                        200.0
+                                    ],
+                                    "spans": [],
+                                    "index": 13
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                339,
+                                213,
+                                486,
+                                225
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        337,
+                                        211,
+                                        487,
+                                        228
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                337,
+                                                211,
+                                                487,
+                                                228
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 4: Context swapping (right).",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 14
+                                }
+                            ],
+                            "index": 14
+                        }
+                    ],
+                    "index": 13.25
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        229,
+                        201,
+                        241
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                227,
+                                202,
+                                243
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        227,
+                                        202,
+                                        243
+                                    ],
+                                    "score": 1.0,
+                                    "content": "5 EXPERIMENTS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        }
+                    ],
+                    "index": 15
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        245,
+                        505,
+                        322
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                245,
+                                505,
+                                258
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        245,
+                                        505,
+                                        258
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In Section 5.1, we introduce ‘unit tests’ for HI-ADS. Our primary goal with these experiments is",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                257,
+                                505,
+                                269
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        257,
+                                        505,
+                                        269
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to convey a crisp understanding of potential issues caused by revealing HI-ADS. Put simply, our",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                267,
+                                506,
+                                280
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        267,
+                                        506,
+                                        280
+                                    ],
+                                    "score": 1.0,
+                                    "content": "experiments show that you can have a learner which behaves as intended, and just by using meta-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                279,
+                                505,
+                                291
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        279,
+                                        505,
+                                        291
+                                    ],
+                                    "score": 1.0,
+                                    "content": "learning (e.g. PBT), without changing the performance metric (e.g. loss or rewards), the learner’s",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                289,
+                                505,
+                                302
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        289,
+                                        505,
+                                        302
+                                    ],
+                                    "score": 1.0,
+                                    "content": "behavior can change completely. We also show that context swapping is an effective mitigation",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                300,
+                                505,
+                                313
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        300,
+                                        505,
+                                        313
+                                    ],
+                                    "score": 1.0,
+                                    "content": "technique in these environments. On the practical side, the unit tests can be used to compare learning",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                312,
+                                353,
+                                323
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        312,
+                                        353,
+                                        323
+                                    ],
+                                    "score": 1.0,
+                                    "content": "algorithms and diagnose their propensity to reveal incentives.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        }
+                    ],
+                    "index": 19
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        328,
+                        505,
+                        406
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                328,
+                                505,
+                                340
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        328,
+                                        505,
+                                        340
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In Section 5.2, we model a content recommendation system. The goal of these experiments is to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                339,
+                                505,
+                                351
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        339,
+                                        505,
+                                        351
+                                    ],
+                                    "score": 1.0,
+                                    "content": "demonstrate how HI-ADS could create issues for real-world content recommendation systems such",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                350,
+                                506,
+                                362
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        350,
+                                        506,
+                                        362
+                                    ],
+                                    "score": 1.0,
+                                    "content": "as news feeds, search results, or automated suggestions. They also validate the usefulness of the unit",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                361,
+                                505,
+                                374
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        361,
+                                        505,
+                                        374
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tests: algorithms that failed the unit tests also reveal HI-ADS in this setting. We emphasize that ADS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                371,
+                                505,
+                                385
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        371,
+                                        505,
+                                        385
+                                    ],
+                                    "score": 1.0,
+                                    "content": "takes place in this environment by construction. The point of our experiments is that meta-learning",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                382,
+                                505,
+                                396
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        382,
+                                        505,
+                                        396
+                                    ],
+                                    "score": 1.0,
+                                    "content": "can increase the rate and/or extent of ADS, by revealing this incentive. We find that context swapping",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                393,
+                                480,
+                                407
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        393,
+                                        480,
+                                        407
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is not effective in this environment, highlighting the need for alternative mitigation strategies.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        }
+                    ],
+                    "index": 26
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        413,
+                        223,
+                        424
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                412,
+                                224,
+                                426
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        412,
+                                        224,
+                                        426
+                                    ],
+                                    "score": 1.0,
+                                    "content": "5.1 HI-ADS UNIT TESTS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        }
+                    ],
+                    "index": 30
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        428,
+                        505,
+                        516
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                427,
+                                506,
+                                441
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        427,
+                                        506,
+                                        441
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Unit test 1: Supervised Learning. This unit test consists of a simple prediction problem. There are",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                438,
+                                506,
+                                452
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        438,
+                                        241,
+                                        452
+                                    ],
+                                    "score": 1.0,
+                                    "content": "no inputs, only an underlying state",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        241,
+                                        439,
+                                        283,
+                                        451
+                                    ],
+                                    "score": 0.93,
+                                    "content": "s \\in \\{ 0 , 1 \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        284,
+                                        438,
+                                        331,
+                                        452
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", and targets",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        331,
+                                        439,
+                                        361,
+                                        450
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\boldsymbol { y } \\in \\mathbb { R } ^ { 2 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        361,
+                                        438,
+                                        381,
+                                        452
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        381,
+                                        438,
+                                        503,
+                                        451
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\chi _ { 1 } , y _ { 2 } \\sim \\mathcal { N } ( 0 , s * \\sigma ^ { 2 } ) , \\mathcal { N } ( 0 , 1 )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        503,
+                                        438,
+                                        506,
+                                        452
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                450,
+                                506,
+                                463
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        450,
+                                        235,
+                                        463
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with corresponding predictions",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        236,
+                                        451,
+                                        260,
+                                        462
+                                    ],
+                                    "score": 0.89,
+                                    "content": "\\hat { y } _ { 1 } , \\hat { y } _ { 2 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        260,
+                                        450,
+                                        321,
+                                        463
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Additionally,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        321,
+                                        451,
+                                        360,
+                                        461
+                                    ],
+                                    "score": 0.89,
+                                    "content": "s _ { t + 1 } = 0",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        360,
+                                        450,
+                                        373,
+                                        463
+                                    ],
+                                    "score": 1.0,
+                                    "content": "iff",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        374,
+                                        451,
+                                        395,
+                                        461
+                                    ],
+                                    "score": 0.86,
+                                    "content": "\\hat { y } _ { 2 } >",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        396,
+                                        450,
+                                        506,
+                                        463
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".5. We use Mean Squared",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                461,
+                                504,
+                                474
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        461,
+                                        322,
+                                        474
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Error as the loss function, so the optimal predictor is",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        323,
+                                        461,
+                                        382,
+                                        473
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\hat { y _ { 1 } } , \\hat { y _ { 2 } } = ( 0 , 0 )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        383,
+                                        461,
+                                        472,
+                                        474
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". However, predicting",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        472,
+                                        461,
+                                        504,
+                                        472
+                                    ],
+                                    "score": 0.87,
+                                    "content": "\\hat { y } _ { 2 } > . 5",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                470,
+                                506,
+                                486
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        470,
+                                        207,
+                                        486
+                                    ],
+                                    "score": 1.0,
+                                    "content": "reduces the variance of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        207,
+                                        472,
+                                        218,
+                                        483
+                                    ],
+                                    "score": 0.88,
+                                    "content": "\\hat { y _ { 1 } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        218,
+                                        470,
+                                        467,
+                                        486
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", i.e. reduces future loss. The baseline/IL predictor learns",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        467,
+                                        473,
+                                        492,
+                                        483
+                                    ],
+                                    "score": 0.89,
+                                    "content": "\\hat { y _ { 1 } } , \\hat { y _ { 2 } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        492,
+                                        470,
+                                        506,
+                                        486
+                                    ],
+                                    "score": 1.0,
+                                    "content": "as",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                482,
+                                506,
+                                495
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        482,
+                                        506,
+                                        495
+                                    ],
+                                    "score": 1.0,
+                                    "content": "parameters using SGD with a learning rate of 0.001. For experiments with meta-learning, PBT is the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                493,
+                                505,
+                                506
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        493,
+                                        505,
+                                        506
+                                    ],
+                                    "score": 1.0,
+                                    "content": "OL (with default settings, see Section 2.2), used to tune the learning rate, with negative loss on the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                505,
+                                375,
+                                517
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        505,
+                                        375,
+                                        517
+                                    ],
+                                    "score": 1.0,
+                                    "content": "final time-step of the interval as the performance measure for PBT.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        }
+                    ],
+                    "index": 34.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        525,
+                        505,
+                        570
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                525,
+                                505,
+                                538
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        525,
+                                        505,
+                                        538
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Unit test 2: Myopic RL. This unit test is based on a modification of the prisoner’s dilemma",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                536,
+                                506,
+                                549
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        536,
+                                        506,
+                                        549
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(Prisner, 2014) where an agent plays each round against its past self. The reward function is presented",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                547,
+                                505,
+                                561
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        547,
+                                        505,
+                                        561
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in Table 1. An agent in this environment has a long-term, non-myopic, incentive for cooperation",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                559,
+                                506,
+                                570
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        559,
+                                        506,
+                                        570
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(with its future self), but a current-time-step, myopic, incentive for defection (from its future self).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        }
+                    ],
+                    "index": 40.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        570,
+                        309,
+                        679
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                569,
+                                308,
+                                581
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        569,
+                                        308,
+                                        581
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The unit test evaluates whether a agent reveals the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                580,
+                                309,
+                                592
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        580,
+                                        309,
+                                        592
+                                    ],
+                                    "score": 1.0,
+                                    "content": "non-myopic incentive even when the agent is meant",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                591,
+                                309,
+                                603
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        591,
+                                        309,
+                                        603
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to optimize for the present reward only (i.e. uses",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                602,
+                                310,
+                                614
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        602,
+                                        159,
+                                        614
+                                    ],
+                                    "score": 1.0,
+                                    "content": "discount rate",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        159,
+                                        603,
+                                        185,
+                                        614
+                                    ],
+                                    "score": 0.9,
+                                    "content": "\\gamma = 0",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        185,
+                                        602,
+                                        310,
+                                        614
+                                    ],
+                                    "score": 1.0,
+                                    "content": "). Naively, we’d expect the non-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                613,
+                                309,
+                                625
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        613,
+                                        309,
+                                        625
+                                    ],
+                                    "score": 1.0,
+                                    "content": "myopic incentive to be hidden from the agent in",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 47
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                624,
+                                310,
+                                636
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        624,
+                                        310,
+                                        636
+                                    ],
+                                    "score": 1.0,
+                                    "content": "this case, and for the agent to consistently defect;",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 48
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                635,
+                                309,
+                                647
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        635,
+                                        309,
+                                        647
+                                    ],
+                                    "score": 1.0,
+                                    "content": "learning algorithms that do so pass the test. But",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 49
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                646,
+                                310,
+                                658
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        646,
+                                        310,
+                                        658
+                                    ],
+                                    "score": 1.0,
+                                    "content": "some learning algorithms also fail the unit test, re-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 50
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                657,
+                                309,
+                                669
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        657,
+                                        309,
+                                        669
+                                    ],
+                                    "score": 1.0,
+                                    "content": "vealing the incentive for the agent to cooperate with",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 51
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                667,
+                                309,
+                                680
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        667,
+                                        309,
+                                        680
+                                    ],
+                                    "score": 1.0,
+                                    "content": "its future self. While aiming for myopic behavior",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 52
+                        }
+                    ],
+                    "index": 47.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        316,
+                        577,
+                        505,
+                        643
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                316,
+                                577,
+                                505,
+                                589
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        316,
+                                        577,
+                                        505,
+                                        589
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Table 1: Rewards for the RL unit test. Note that",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 53
+                        },
+                        {
+                            "bbox": [
+                                316,
+                                589,
+                                505,
+                                601
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        316,
+                                        589,
+                                        505,
+                                        601
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the myopic (defect (D)) action always increases",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 54
+                        },
+                        {
+                            "bbox": [
+                                315,
+                                600,
+                                505,
+                                611
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        315,
+                                        600,
+                                        505,
+                                        611
+                                    ],
+                                    "score": 1.0,
+                                    "content": "reward at the current time-step, but decreases",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 55
+                        },
+                        {
+                            "bbox": [
+                                315,
+                                610,
+                                505,
+                                623
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        315,
+                                        610,
+                                        505,
+                                        623
+                                    ],
+                                    "score": 1.0,
+                                    "content": "reward at the next time-step - the incentive to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 56
+                        },
+                        {
+                            "bbox": [
+                                317,
+                                622,
+                                505,
+                                632
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        317,
+                                        622,
+                                        505,
+                                        632
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(cooperate (C)) with one’s future self is hidden",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 57
+                        },
+                        {
+                            "bbox": [
+                                316,
+                                632,
+                                491,
+                                645
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        316,
+                                        632,
+                                        491,
+                                        645
+                                    ],
+                                    "score": 1.0,
+                                    "content": "from the point of view of a myopic learner.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 58
+                        }
+                    ],
+                    "index": 55.5
+                },
+                {
+                    "type": "table",
+                    "bbox": [
+                        332,
+                        644,
+                        488,
+                        679
+                    ],
+                    "blocks": [
+                        {
+                            "type": "table_body",
+                            "bbox": [
+                                332,
+                                644,
+                                488,
+                                679
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        332,
+                                        644,
+                                        488,
+                                        679
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                332,
+                                                644,
+                                                488,
+                                                679
+                                            ],
+                                            "score": 0.961,
+                                            "html": "<table><tr><td rowspan=1 colspan=1></td><td rowspan=1 colspan=1>at=D</td><td rowspan=1 colspan=1>at=C</td></tr><tr><td rowspan=1 colspan=1>St=at-1=D</td><td rowspan=1 colspan=1>-1/2</td><td rowspan=1 colspan=1>-1</td></tr><tr><td rowspan=1 colspan=1>St=at-1=C</td><td rowspan=1 colspan=1>1/2</td><td rowspan=1 colspan=1>0</td></tr></table>",
+                                            "type": "table",
+                                            "image_path": "370e974762175107c7aed2e60f9f736ca0b327ddca422ff17241cb1b39d1c0b0.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 59.5,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        332,
+                                        644,
+                                        488,
+                                        661.5
+                                    ],
+                                    "spans": [],
+                                    "index": 59
+                                },
+                                {
+                                    "bbox": [
+                                        332,
+                                        661.5,
+                                        488,
+                                        679.0
+                                    ],
+                                    "spans": [],
+                                    "index": 60
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 59.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        680,
+                        506,
+                        723
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                678,
+                                506,
+                                692
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        678,
+                                        506,
+                                        692
+                                    ],
+                                    "score": 1.0,
+                                    "content": "may seem odd, myopic learners have no incentives to cause distributional shift, since it can only",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 61
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                689,
+                                506,
+                                702
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        689,
+                                        506,
+                                        702
+                                    ],
+                                    "score": 1.0,
+                                    "content": "improve future performance. And while making learners myopic may seem like a ’brute-force’",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 62
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                700,
+                                506,
+                                714
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        700,
+                                        506,
+                                        714
+                                    ],
+                                    "score": 1.0,
+                                    "content": "guaranteed way to manage HI-ADS, we show it is in fact non-trivial to implement. See Appendix 9.1",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 63
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                711,
+                                334,
+                                725
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        711,
+                                        334,
+                                        725
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for details and experiments varying the reward structure.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 64
+                        }
+                    ],
+                    "index": 62.5
+                }
+            ],
+            "page_idx": 4,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        107,
+                        27,
+                        307,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                26,
+                                308,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        26,
+                                        308,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Under review as a conference paper at ICLR 2021",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        302,
+                        751,
+                        308,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                302,
+                                750,
+                                309,
+                                763
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        302,
+                                        750,
+                                        309,
+                                        763
+                                    ],
+                                    "score": 1.0,
+                                    "content": "5",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "title",
+                    "bbox": [
+                        106,
+                        82,
+                        329,
+                        93
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                82,
+                                330,
+                                95
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        82,
+                                        330,
+                                        95
+                                    ],
+                                    "score": 1.0,
+                                    "content": "4.2 REMOVING HI-ADS VIA CONTEXT SWAPPING",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        }
+                    ],
+                    "index": 0
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        97,
+                        505,
+                        152
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                97,
+                                505,
+                                110
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        97,
+                                        505,
+                                        110
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We propose a technique called context swapping for removing incentives for ADS. The technique",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                108,
+                                506,
+                                121
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        108,
+                                        132,
+                                        121
+                                    ],
+                                    "score": 1.0,
+                                    "content": "trains",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        132,
+                                        109,
+                                        142,
+                                        118
+                                    ],
+                                    "score": 0.81,
+                                    "content": "N",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        142,
+                                        108,
+                                        362,
+                                        121
+                                    ],
+                                    "score": 1.0,
+                                    "content": "learners in parallel, and shuffles the learners through",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        362,
+                                        109,
+                                        372,
+                                        118
+                                    ],
+                                    "score": 0.71,
+                                    "content": "N",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        373,
+                                        108,
+                                        506,
+                                        121
+                                    ],
+                                    "score": 1.0,
+                                    "content": "different copies of the same (or",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                119,
+                                505,
+                                132
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        119,
+                                        505,
+                                        132
+                                    ],
+                                    "score": 1.0,
+                                    "content": "similar) environments.We use a deterministic permutation of learners in environment copies, so",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                130,
+                                506,
+                                143
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        130,
+                                        141,
+                                        143
+                                    ],
+                                    "score": 1.0,
+                                    "content": "that the",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        141,
+                                        131,
+                                        146,
+                                        140
+                                    ],
+                                    "score": 0.75,
+                                    "content": "i",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        146,
+                                        130,
+                                        243,
+                                        143
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-th learner inhabits the",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        243,
+                                        131,
+                                        249,
+                                        142
+                                    ],
+                                    "score": 0.81,
+                                    "content": "j",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        250,
+                                        130,
+                                        376,
+                                        143
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-th environment on time-steps",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        376,
+                                        131,
+                                        382,
+                                        140
+                                    ],
+                                    "score": 0.62,
+                                    "content": "t",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        382,
+                                        130,
+                                        412,
+                                        143
+                                    ],
+                                    "score": 1.0,
+                                    "content": "where",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        412,
+                                        130,
+                                        464,
+                                        142
+                                    ],
+                                    "score": 0.92,
+                                    "content": "~ j ~ = ~ ( i + t )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        465,
+                                        130,
+                                        492,
+                                        143
+                                    ],
+                                    "score": 1.0,
+                                    "content": "mod",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        492,
+                                        131,
+                                        502,
+                                        141
+                                    ],
+                                    "score": 0.71,
+                                    "content": "N",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        503,
+                                        130,
+                                        506,
+                                        143
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                141,
+                                506,
+                                153
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        141,
+                                        506,
+                                        153
+                                    ],
+                                    "score": 1.0,
+                                    "content": "makes an observation, takes an action, and receives a reward before moving to the next environment.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        }
+                    ],
+                    "index": 3,
+                    "bbox_fs": [
+                        105,
+                        97,
+                        506,
+                        153
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        153,
+                        317,
+                        218
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                151,
+                                318,
+                                164
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        151,
+                                        133,
+                                        164
+                                    ],
+                                    "score": 1.0,
+                                    "content": "When",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        134,
+                                        153,
+                                        144,
+                                        162
+                                    ],
+                                    "score": 0.79,
+                                    "content": "N",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        145,
+                                        151,
+                                        318,
+                                        164
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is larger than the interval of the OL opti-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                163,
+                                317,
+                                174
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        163,
+                                        317,
+                                        174
+                                    ],
+                                    "score": 1.0,
+                                    "content": "mizer, each learner inhabits each copy for at most a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                174,
+                                317,
+                                186
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        174,
+                                        317,
+                                        186
+                                    ],
+                                    "score": 1.0,
+                                    "content": "single time-step before an OL step is applied. Under",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                185,
+                                318,
+                                196
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        185,
+                                        318,
+                                        196
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the assumption that different copies of the environ-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                196,
+                                317,
+                                208
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        196,
+                                        317,
+                                        208
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ment do not influence each other, this technique can",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                207,
+                                318,
+                                219
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        207,
+                                        318,
+                                        219
+                                    ],
+                                    "score": 1.0,
+                                    "content": "address HI-ADS in practice, as we show in Sec. 5.1.1.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        }
+                    ],
+                    "index": 8.5,
+                    "bbox_fs": [
+                        106,
+                        151,
+                        318,
+                        219
+                    ]
+                },
+                {
+                    "type": "image",
+                    "bbox": [
+                        327,
+                        162,
+                        502,
+                        200
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                327,
+                                162,
+                                502,
+                                200
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        327,
+                                        162,
+                                        502,
+                                        200
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                327,
+                                                162,
+                                                502,
+                                                200
+                                            ],
+                                            "score": 0.948,
+                                            "type": "image",
+                                            "image_path": "06fd14fc3c3117d01ce92cc59924cf8d1fe7f705cf8d340291c9e7534c747cd1.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 12.5,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        327,
+                                        162,
+                                        502,
+                                        181.0
+                                    ],
+                                    "spans": [],
+                                    "index": 12
+                                },
+                                {
+                                    "bbox": [
+                                        327,
+                                        181.0,
+                                        502,
+                                        200.0
+                                    ],
+                                    "spans": [],
+                                    "index": 13
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                339,
+                                213,
+                                486,
+                                225
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        337,
+                                        211,
+                                        487,
+                                        228
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                337,
+                                                211,
+                                                487,
+                                                228
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 4: Context swapping (right).",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 14
+                                }
+                            ],
+                            "index": 14
+                        }
+                    ],
+                    "index": 13.25
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        229,
+                        201,
+                        241
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                227,
+                                202,
+                                243
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        227,
+                                        202,
+                                        243
+                                    ],
+                                    "score": 1.0,
+                                    "content": "5 EXPERIMENTS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        }
+                    ],
+                    "index": 15
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        245,
+                        505,
+                        322
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                245,
+                                505,
+                                258
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        245,
+                                        505,
+                                        258
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In Section 5.1, we introduce ‘unit tests’ for HI-ADS. Our primary goal with these experiments is",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                257,
+                                505,
+                                269
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        257,
+                                        505,
+                                        269
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to convey a crisp understanding of potential issues caused by revealing HI-ADS. Put simply, our",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                267,
+                                506,
+                                280
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        267,
+                                        506,
+                                        280
+                                    ],
+                                    "score": 1.0,
+                                    "content": "experiments show that you can have a learner which behaves as intended, and just by using meta-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                279,
+                                505,
+                                291
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        279,
+                                        505,
+                                        291
+                                    ],
+                                    "score": 1.0,
+                                    "content": "learning (e.g. PBT), without changing the performance metric (e.g. loss or rewards), the learner’s",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                289,
+                                505,
+                                302
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        289,
+                                        505,
+                                        302
+                                    ],
+                                    "score": 1.0,
+                                    "content": "behavior can change completely. We also show that context swapping is an effective mitigation",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                300,
+                                505,
+                                313
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        300,
+                                        505,
+                                        313
+                                    ],
+                                    "score": 1.0,
+                                    "content": "technique in these environments. On the practical side, the unit tests can be used to compare learning",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                312,
+                                353,
+                                323
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        312,
+                                        353,
+                                        323
+                                    ],
+                                    "score": 1.0,
+                                    "content": "algorithms and diagnose their propensity to reveal incentives.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        }
+                    ],
+                    "index": 19,
+                    "bbox_fs": [
+                        105,
+                        245,
+                        506,
+                        323
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        328,
+                        505,
+                        406
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                328,
+                                505,
+                                340
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        328,
+                                        505,
+                                        340
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In Section 5.2, we model a content recommendation system. The goal of these experiments is to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                339,
+                                505,
+                                351
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        339,
+                                        505,
+                                        351
+                                    ],
+                                    "score": 1.0,
+                                    "content": "demonstrate how HI-ADS could create issues for real-world content recommendation systems such",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                350,
+                                506,
+                                362
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        350,
+                                        506,
+                                        362
+                                    ],
+                                    "score": 1.0,
+                                    "content": "as news feeds, search results, or automated suggestions. They also validate the usefulness of the unit",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                361,
+                                505,
+                                374
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        361,
+                                        505,
+                                        374
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tests: algorithms that failed the unit tests also reveal HI-ADS in this setting. We emphasize that ADS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                371,
+                                505,
+                                385
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        371,
+                                        505,
+                                        385
+                                    ],
+                                    "score": 1.0,
+                                    "content": "takes place in this environment by construction. The point of our experiments is that meta-learning",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                382,
+                                505,
+                                396
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        382,
+                                        505,
+                                        396
+                                    ],
+                                    "score": 1.0,
+                                    "content": "can increase the rate and/or extent of ADS, by revealing this incentive. We find that context swapping",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                393,
+                                480,
+                                407
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        393,
+                                        480,
+                                        407
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is not effective in this environment, highlighting the need for alternative mitigation strategies.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        }
+                    ],
+                    "index": 26,
+                    "bbox_fs": [
+                        105,
+                        328,
+                        506,
+                        407
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        413,
+                        223,
+                        424
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                412,
+                                224,
+                                426
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        412,
+                                        224,
+                                        426
+                                    ],
+                                    "score": 1.0,
+                                    "content": "5.1 HI-ADS UNIT TESTS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        }
+                    ],
+                    "index": 30
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        428,
+                        505,
+                        516
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                427,
+                                506,
+                                441
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        427,
+                                        506,
+                                        441
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Unit test 1: Supervised Learning. This unit test consists of a simple prediction problem. There are",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                438,
+                                506,
+                                452
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        438,
+                                        241,
+                                        452
+                                    ],
+                                    "score": 1.0,
+                                    "content": "no inputs, only an underlying state",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        241,
+                                        439,
+                                        283,
+                                        451
+                                    ],
+                                    "score": 0.93,
+                                    "content": "s \\in \\{ 0 , 1 \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        284,
+                                        438,
+                                        331,
+                                        452
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", and targets",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        331,
+                                        439,
+                                        361,
+                                        450
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\boldsymbol { y } \\in \\mathbb { R } ^ { 2 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        361,
+                                        438,
+                                        381,
+                                        452
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        381,
+                                        438,
+                                        503,
+                                        451
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\chi _ { 1 } , y _ { 2 } \\sim \\mathcal { N } ( 0 , s * \\sigma ^ { 2 } ) , \\mathcal { N } ( 0 , 1 )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        503,
+                                        438,
+                                        506,
+                                        452
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                450,
+                                506,
+                                463
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        450,
+                                        235,
+                                        463
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with corresponding predictions",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        236,
+                                        451,
+                                        260,
+                                        462
+                                    ],
+                                    "score": 0.89,
+                                    "content": "\\hat { y } _ { 1 } , \\hat { y } _ { 2 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        260,
+                                        450,
+                                        321,
+                                        463
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Additionally,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        321,
+                                        451,
+                                        360,
+                                        461
+                                    ],
+                                    "score": 0.89,
+                                    "content": "s _ { t + 1 } = 0",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        360,
+                                        450,
+                                        373,
+                                        463
+                                    ],
+                                    "score": 1.0,
+                                    "content": "iff",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        374,
+                                        451,
+                                        395,
+                                        461
+                                    ],
+                                    "score": 0.86,
+                                    "content": "\\hat { y } _ { 2 } >",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        396,
+                                        450,
+                                        506,
+                                        463
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".5. We use Mean Squared",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                461,
+                                504,
+                                474
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        461,
+                                        322,
+                                        474
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Error as the loss function, so the optimal predictor is",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        323,
+                                        461,
+                                        382,
+                                        473
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\hat { y _ { 1 } } , \\hat { y _ { 2 } } = ( 0 , 0 )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        383,
+                                        461,
+                                        472,
+                                        474
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". However, predicting",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        472,
+                                        461,
+                                        504,
+                                        472
+                                    ],
+                                    "score": 0.87,
+                                    "content": "\\hat { y } _ { 2 } > . 5",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                470,
+                                506,
+                                486
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        470,
+                                        207,
+                                        486
+                                    ],
+                                    "score": 1.0,
+                                    "content": "reduces the variance of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        207,
+                                        472,
+                                        218,
+                                        483
+                                    ],
+                                    "score": 0.88,
+                                    "content": "\\hat { y _ { 1 } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        218,
+                                        470,
+                                        467,
+                                        486
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", i.e. reduces future loss. The baseline/IL predictor learns",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        467,
+                                        473,
+                                        492,
+                                        483
+                                    ],
+                                    "score": 0.89,
+                                    "content": "\\hat { y _ { 1 } } , \\hat { y _ { 2 } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        492,
+                                        470,
+                                        506,
+                                        486
+                                    ],
+                                    "score": 1.0,
+                                    "content": "as",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                482,
+                                506,
+                                495
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        482,
+                                        506,
+                                        495
+                                    ],
+                                    "score": 1.0,
+                                    "content": "parameters using SGD with a learning rate of 0.001. For experiments with meta-learning, PBT is the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                493,
+                                505,
+                                506
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        493,
+                                        505,
+                                        506
+                                    ],
+                                    "score": 1.0,
+                                    "content": "OL (with default settings, see Section 2.2), used to tune the learning rate, with negative loss on the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                505,
+                                375,
+                                517
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        505,
+                                        375,
+                                        517
+                                    ],
+                                    "score": 1.0,
+                                    "content": "final time-step of the interval as the performance measure for PBT.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        }
+                    ],
+                    "index": 34.5,
+                    "bbox_fs": [
+                        105,
+                        427,
+                        506,
+                        517
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        525,
+                        505,
+                        570
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                525,
+                                505,
+                                538
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        525,
+                                        505,
+                                        538
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Unit test 2: Myopic RL. This unit test is based on a modification of the prisoner’s dilemma",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                536,
+                                506,
+                                549
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        536,
+                                        506,
+                                        549
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(Prisner, 2014) where an agent plays each round against its past self. The reward function is presented",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                547,
+                                505,
+                                561
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        547,
+                                        505,
+                                        561
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in Table 1. An agent in this environment has a long-term, non-myopic, incentive for cooperation",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                559,
+                                506,
+                                570
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        559,
+                                        506,
+                                        570
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(with its future self), but a current-time-step, myopic, incentive for defection (from its future self).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        }
+                    ],
+                    "index": 40.5,
+                    "bbox_fs": [
+                        104,
+                        525,
+                        506,
+                        570
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        570,
+                        309,
+                        679
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                569,
+                                308,
+                                581
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        569,
+                                        308,
+                                        581
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The unit test evaluates whether a agent reveals the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                580,
+                                309,
+                                592
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        580,
+                                        309,
+                                        592
+                                    ],
+                                    "score": 1.0,
+                                    "content": "non-myopic incentive even when the agent is meant",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                591,
+                                309,
+                                603
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        591,
+                                        309,
+                                        603
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to optimize for the present reward only (i.e. uses",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                602,
+                                310,
+                                614
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        602,
+                                        159,
+                                        614
+                                    ],
+                                    "score": 1.0,
+                                    "content": "discount rate",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        159,
+                                        603,
+                                        185,
+                                        614
+                                    ],
+                                    "score": 0.9,
+                                    "content": "\\gamma = 0",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        185,
+                                        602,
+                                        310,
+                                        614
+                                    ],
+                                    "score": 1.0,
+                                    "content": "). Naively, we’d expect the non-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                613,
+                                309,
+                                625
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        613,
+                                        309,
+                                        625
+                                    ],
+                                    "score": 1.0,
+                                    "content": "myopic incentive to be hidden from the agent in",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 47
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                624,
+                                310,
+                                636
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        624,
+                                        310,
+                                        636
+                                    ],
+                                    "score": 1.0,
+                                    "content": "this case, and for the agent to consistently defect;",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 48
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                635,
+                                309,
+                                647
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        635,
+                                        309,
+                                        647
+                                    ],
+                                    "score": 1.0,
+                                    "content": "learning algorithms that do so pass the test. But",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 49
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                646,
+                                310,
+                                658
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        646,
+                                        310,
+                                        658
+                                    ],
+                                    "score": 1.0,
+                                    "content": "some learning algorithms also fail the unit test, re-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 50
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                657,
+                                309,
+                                669
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        657,
+                                        309,
+                                        669
+                                    ],
+                                    "score": 1.0,
+                                    "content": "vealing the incentive for the agent to cooperate with",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 51
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                667,
+                                309,
+                                680
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        667,
+                                        309,
+                                        680
+                                    ],
+                                    "score": 1.0,
+                                    "content": "its future self. While aiming for myopic behavior",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 52
+                        }
+                    ],
+                    "index": 47.5,
+                    "bbox_fs": [
+                        105,
+                        569,
+                        310,
+                        680
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        316,
+                        577,
+                        505,
+                        643
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                316,
+                                577,
+                                505,
+                                589
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        316,
+                                        577,
+                                        505,
+                                        589
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Table 1: Rewards for the RL unit test. Note that",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 53
+                        },
+                        {
+                            "bbox": [
+                                316,
+                                589,
+                                505,
+                                601
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        316,
+                                        589,
+                                        505,
+                                        601
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the myopic (defect (D)) action always increases",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 54
+                        },
+                        {
+                            "bbox": [
+                                315,
+                                600,
+                                505,
+                                611
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        315,
+                                        600,
+                                        505,
+                                        611
+                                    ],
+                                    "score": 1.0,
+                                    "content": "reward at the current time-step, but decreases",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 55
+                        },
+                        {
+                            "bbox": [
+                                315,
+                                610,
+                                505,
+                                623
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        315,
+                                        610,
+                                        505,
+                                        623
+                                    ],
+                                    "score": 1.0,
+                                    "content": "reward at the next time-step - the incentive to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 56
+                        },
+                        {
+                            "bbox": [
+                                317,
+                                622,
+                                505,
+                                632
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        317,
+                                        622,
+                                        505,
+                                        632
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(cooperate (C)) with one’s future self is hidden",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 57
+                        },
+                        {
+                            "bbox": [
+                                316,
+                                632,
+                                491,
+                                645
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        316,
+                                        632,
+                                        491,
+                                        645
+                                    ],
+                                    "score": 1.0,
+                                    "content": "from the point of view of a myopic learner.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 58
+                        }
+                    ],
+                    "index": 55.5,
+                    "bbox_fs": [
+                        315,
+                        577,
+                        505,
+                        645
+                    ]
+                },
+                {
+                    "type": "table",
+                    "bbox": [
+                        332,
+                        644,
+                        488,
+                        679
+                    ],
+                    "blocks": [
+                        {
+                            "type": "table_body",
+                            "bbox": [
+                                332,
+                                644,
+                                488,
+                                679
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        332,
+                                        644,
+                                        488,
+                                        679
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                332,
+                                                644,
+                                                488,
+                                                679
+                                            ],
+                                            "score": 0.961,
+                                            "html": "<table><tr><td rowspan=1 colspan=1></td><td rowspan=1 colspan=1>at=D</td><td rowspan=1 colspan=1>at=C</td></tr><tr><td rowspan=1 colspan=1>St=at-1=D</td><td rowspan=1 colspan=1>-1/2</td><td rowspan=1 colspan=1>-1</td></tr><tr><td rowspan=1 colspan=1>St=at-1=C</td><td rowspan=1 colspan=1>1/2</td><td rowspan=1 colspan=1>0</td></tr></table>",
+                                            "type": "table",
+                                            "image_path": "370e974762175107c7aed2e60f9f736ca0b327ddca422ff17241cb1b39d1c0b0.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 59.5,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        332,
+                                        644,
+                                        488,
+                                        661.5
+                                    ],
+                                    "spans": [],
+                                    "index": 59
+                                },
+                                {
+                                    "bbox": [
+                                        332,
+                                        661.5,
+                                        488,
+                                        679.0
+                                    ],
+                                    "spans": [],
+                                    "index": 60
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 59.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        680,
+                        506,
+                        723
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                678,
+                                506,
+                                692
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        678,
+                                        506,
+                                        692
+                                    ],
+                                    "score": 1.0,
+                                    "content": "may seem odd, myopic learners have no incentives to cause distributional shift, since it can only",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 61
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                689,
+                                506,
+                                702
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        689,
+                                        506,
+                                        702
+                                    ],
+                                    "score": 1.0,
+                                    "content": "improve future performance. And while making learners myopic may seem like a ’brute-force’",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 62
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                700,
+                                506,
+                                714
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        700,
+                                        506,
+                                        714
+                                    ],
+                                    "score": 1.0,
+                                    "content": "guaranteed way to manage HI-ADS, we show it is in fact non-trivial to implement. See Appendix 9.1",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 63
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                711,
+                                334,
+                                725
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        711,
+                                        334,
+                                        725
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for details and experiments varying the reward structure.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 64
+                        }
+                    ],
+                    "index": 62.5,
+                    "bbox_fs": [
+                        105,
+                        678,
+                        506,
+                        725
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "title",
+                    "bbox": [
+                        106,
+                        82,
+                        416,
+                        93
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                82,
+                                417,
+                                95
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        82,
+                                        417,
+                                        95
+                                    ],
+                                    "score": 1.0,
+                                    "content": "5.1.1 HI-ADS UNIT TESTS EXPERIMENTAL RESULTS AND DISCUSSION",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        }
+                    ],
+                    "index": 0
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        87,
+                        505,
+                        160
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                93,
+                                505,
+                                106
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        93,
+                                        505,
+                                        106
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We first show that agents trained with PBT fail the unit tests more often than “vanilla” algorithms",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                105,
+                                505,
+                                117
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        105,
+                                        505,
+                                        117
+                                    ],
+                                    "score": 1.0,
+                                    "content": "which do not use meta-learning. We initialize the learning rate log-uniformly between 0.01 and 1.0 for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                116,
+                                505,
+                                128
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        116,
+                                        505,
+                                        128
+                                    ],
+                                    "score": 1.0,
+                                    "content": "all experiments (whether using PBT or not). We expect and confirm that the following two factors lead",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                127,
+                                505,
+                                139
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        127,
+                                        505,
+                                        139
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to higher rates of unit test failure: (1) Shorter intervals: These give the OL more opportunities to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                137,
+                                506,
+                                151
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        137,
+                                        506,
+                                        151
+                                    ],
+                                    "score": 1.0,
+                                    "content": "influence the population. (2) Larger populations: These make outliers with exceptional non-myopic",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                149,
+                                420,
+                                162
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        149,
+                                        420,
+                                        162
+                                    ],
+                                    "score": 1.0,
+                                    "content": "performance more likely, and OL makes them likely to survive and propagate.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        }
+                    ],
+                    "index": 3.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        165,
+                        505,
+                        199
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                165,
+                                506,
+                                178
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        165,
+                                        506,
+                                        178
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The baseline (no meta-learning) algorithms all pass the unit tests: hidden incentives are almost never",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                177,
+                                505,
+                                189
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        177,
+                                        505,
+                                        189
+                                    ],
+                                    "score": 1.0,
+                                    "content": "revealed - see blue curves in Fig. 5. However, agents trained with meta-learning and large populations",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                187,
+                                361,
+                                200
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        187,
+                                        361,
+                                        200
+                                    ],
+                                    "score": 1.0,
+                                    "content": "often fail the unit tests: see orange curves in top rows of Fig. 5.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        }
+                    ],
+                    "index": 8
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        205,
+                        277,
+                        424
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                204,
+                                279,
+                                217
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        204,
+                                        279,
+                                        217
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Furthermore, we verify that context swap-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                216,
+                                278,
+                                227
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        216,
+                                        278,
+                                        227
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ping significantly mitigates the effect of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                226,
+                                278,
+                                239
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        226,
+                                        278,
+                                        239
+                                    ],
+                                    "score": 1.0,
+                                    "content": "HI-ADS in both unit tests, decreasing un-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                237,
+                                278,
+                                249
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        237,
+                                        278,
+                                        249
+                                    ],
+                                    "score": 1.0,
+                                    "content": "desirable behaviour to near-baseline lev-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                248,
+                                279,
+                                260
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        248,
+                                        279,
+                                        260
+                                    ],
+                                    "score": 1.0,
+                                    "content": "els - see bottom rows of Fig. 5. This ef-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                259,
+                                278,
+                                271
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        259,
+                                        278,
+                                        271
+                                    ],
+                                    "score": 1.0,
+                                    "content": "fect can be explained as follows: Because",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                271,
+                                278,
+                                282
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        271,
+                                        278,
+                                        282
+                                    ],
+                                    "score": 1.0,
+                                    "content": "context swapping transfers the benefits of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                282,
+                                277,
+                                292
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        282,
+                                        277,
+                                        292
+                                    ],
+                                    "score": 1.0,
+                                    "content": "one learner’s action to the next learner to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                291,
+                                277,
+                                304
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        291,
+                                        277,
+                                        304
+                                    ],
+                                    "score": 1.0,
+                                    "content": "inhabit that environment, it increases the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                303,
+                                279,
+                                315
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        303,
+                                        279,
+                                        315
+                                    ],
+                                    "score": 1.0,
+                                    "content": "second learner’s fitness, and thereby re-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                313,
+                                278,
+                                326
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        313,
+                                        278,
+                                        326
+                                    ],
+                                    "score": 1.0,
+                                    "content": "duces the relative fitness (as evaluated by",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                325,
+                                277,
+                                338
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        325,
+                                        277,
+                                        338
+                                    ],
+                                    "score": 1.0,
+                                    "content": "PBT’s EXPLOIT step) of the non-myopic",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                336,
+                                278,
+                                348
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        336,
+                                        278,
+                                        348
+                                    ],
+                                    "score": 1.0,
+                                    "content": "cooperate behaviour. We observe some",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                347,
+                                278,
+                                358
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        347,
+                                        278,
+                                        358
+                                    ],
+                                    "score": 1.0,
+                                    "content": "interesting exceptions with the combina-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                358,
+                                278,
+                                369
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        358,
+                                        278,
+                                        369
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tion of small populations and short PBT",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                369,
+                                277,
+                                381
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        369,
+                                        277,
+                                        381
+                                    ],
+                                    "score": 1.0,
+                                    "content": "intervals: Although context swapping still",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                380,
+                                279,
+                                391
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        380,
+                                        279,
+                                        391
+                                    ],
+                                    "score": 1.0,
+                                    "content": "significantly decreases the effect of HI-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                390,
+                                278,
+                                402
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        390,
+                                        278,
+                                        402
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ADS, non-myopic cooperate behaviour",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                401,
+                                278,
+                                414
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        401,
+                                        196,
+                                        414
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is observed as much as",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        196,
+                                        402,
+                                        216,
+                                        412
+                                    ],
+                                    "score": 0.87,
+                                    "content": "20 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        216,
+                                        401,
+                                        278,
+                                        414
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of the time (for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                412,
+                                279,
+                                425
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        412,
+                                        142,
+                                        425
+                                    ],
+                                    "score": 1.0,
+                                    "content": "#learner",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        143,
+                                        413,
+                                        159,
+                                        423
+                                    ],
+                                    "score": 0.77,
+                                    "content": "_ { = 1 0 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        159,
+                                        412,
+                                        162,
+                                        425
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        163,
+                                        413,
+                                        190,
+                                        423
+                                    ],
+                                    "score": 0.88,
+                                    "content": "T = 1",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        190,
+                                        412,
+                                        279,
+                                        425
+                                    ],
+                                    "score": 1.0,
+                                    "content": "; see bottom-left plot).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        }
+                    ],
+                    "index": 19.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        429,
+                        277,
+                        660
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                429,
+                                277,
+                                441
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        429,
+                                        277,
+                                        441
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We also observe that PBT reveals HI-ADS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                440,
+                                277,
+                                453
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        440,
+                                        153,
+                                        453
+                                    ],
+                                    "score": 1.0,
+                                    "content": "even when",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        153,
+                                        441,
+                                        180,
+                                        451
+                                    ],
+                                    "score": 0.9,
+                                    "content": "T = 1",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        181,
+                                        440,
+                                        277,
+                                        453
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", where the explanation",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                451,
+                                279,
+                                464
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        451,
+                                        279,
+                                        464
+                                    ],
+                                    "score": 1.0,
+                                    "content": "that PBT operates on a longer time hori-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                462,
+                                277,
+                                474
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        462,
+                                        277,
+                                        474
+                                    ],
+                                    "score": 1.0,
+                                    "content": "zon than the inner loop does not apply. We",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                474,
+                                277,
+                                485
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        474,
+                                        277,
+                                        485
+                                    ],
+                                    "score": 1.0,
+                                    "content": "provide a detailed explanation for how this",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                484,
+                                277,
+                                497
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        484,
+                                        277,
+                                        497
+                                    ],
+                                    "score": 1.0,
+                                    "content": "might happen in Appendix 9.1.2, but in",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                496,
+                                278,
+                                507
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        496,
+                                        278,
+                                        507
+                                    ],
+                                    "score": 1.0,
+                                    "content": "summary, we hypothesize that there are at",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                506,
+                                279,
+                                518
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        506,
+                                        279,
+                                        518
+                                    ],
+                                    "score": 1.0,
+                                    "content": "least 2 mechanisms by which PBT is reveal-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                517,
+                                278,
+                                530
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        517,
+                                        278,
+                                        530
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ing HI-ADS: (1) optimizing over a longer",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                527,
+                                279,
+                                541
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        527,
+                                        279,
+                                        541
+                                    ],
+                                    "score": 1.0,
+                                    "content": "time-scale, and (2) picking up on the cor-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                539,
+                                277,
+                                551
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        539,
+                                        277,
+                                        551
+                                    ],
+                                    "score": 1.0,
+                                    "content": "relation between an agent’s current policy",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                550,
+                                277,
+                                563
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        550,
+                                        277,
+                                        563
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and the underlying state. Mechanism (2)",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                561,
+                                278,
+                                574
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        561,
+                                        278,
+                                        574
+                                    ],
+                                    "score": 1.0,
+                                    "content": "can be explained informally as reasoning",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                572,
+                                279,
+                                585
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        572,
+                                        279,
+                                        585
+                                    ],
+                                    "score": 1.0,
+                                    "content": "as: “If I’m cooperating, then I was prob-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                583,
+                                278,
+                                595
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        583,
+                                        278,
+                                        595
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ably cooperating on the last time-step as",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                594,
+                                278,
+                                606
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        594,
+                                        278,
+                                        606
+                                    ],
+                                    "score": 1.0,
+                                    "content": "well, so my reward should be higher”. As",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                606,
+                                279,
+                                617
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        606,
+                                        279,
+                                        617
+                                    ],
+                                    "score": 1.0,
+                                    "content": "support for these hypotheses, we run con-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                616,
+                                278,
+                                628
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        616,
+                                        278,
+                                        628
+                                    ],
+                                    "score": 1.0,
+                                    "content": "trol experiments identifying two algorithms",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 47
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                626,
+                                277,
+                                639
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        626,
+                                        277,
+                                        639
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(each sharing only one of these properties)",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 48
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                636,
+                                277,
+                                651
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        636,
+                                        277,
+                                        651
+                                    ],
+                                    "score": 1.0,
+                                    "content": "that can fail the unit test. Context swapping",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 49
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                649,
+                                180,
+                                660
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        649,
+                                        180,
+                                        660
+                                    ],
+                                    "score": 1.0,
+                                    "content": "remains effective.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 50
+                        }
+                    ],
+                    "index": 40
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        106,
+                        666,
+                        277,
+                        688
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                665,
+                                279,
+                                678
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        665,
+                                        279,
+                                        678
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(1) Optimizing over a longer time-scale:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 51
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                676,
+                                277,
+                                689
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        676,
+                                        277,
+                                        689
+                                    ],
+                                    "score": 1.0,
+                                    "content": "replacing PBT with REINFORCE as an",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 52
+                        }
+                    ],
+                    "index": 51.5
+                },
+                {
+                    "type": "image",
+                    "bbox": [
+                        287,
+                        207,
+                        502,
+                        531
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                287,
+                                207,
+                                502,
+                                531
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        287,
+                                        207,
+                                        502,
+                                        531
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                287,
+                                                207,
+                                                502,
+                                                531
+                                            ],
+                                            "score": 0.967,
+                                            "type": "image",
+                                            "image_path": "fb963b752d78b641eafae0011c4187a77f39ee589e94374f2b91ae0ecc03c850.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 64.5,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        287,
+                                        207,
+                                        502,
+                                        220.5
+                                    ],
+                                    "spans": [],
+                                    "index": 53
+                                },
+                                {
+                                    "bbox": [
+                                        287,
+                                        220.5,
+                                        502,
+                                        234.0
+                                    ],
+                                    "spans": [],
+                                    "index": 54
+                                },
+                                {
+                                    "bbox": [
+                                        287,
+                                        234.0,
+                                        502,
+                                        247.5
+                                    ],
+                                    "spans": [],
+                                    "index": 55
+                                },
+                                {
+                                    "bbox": [
+                                        287,
+                                        247.5,
+                                        502,
+                                        261.0
+                                    ],
+                                    "spans": [],
+                                    "index": 84
+                                },
+                                {
+                                    "bbox": [
+                                        287,
+                                        261.0,
+                                        502,
+                                        274.5
+                                    ],
+                                    "spans": [],
+                                    "index": 56
+                                },
+                                {
+                                    "bbox": [
+                                        287,
+                                        274.5,
+                                        502,
+                                        288.0
+                                    ],
+                                    "spans": [],
+                                    "index": 57
+                                },
+                                {
+                                    "bbox": [
+                                        287,
+                                        288.0,
+                                        502,
+                                        301.5
+                                    ],
+                                    "spans": [],
+                                    "index": 58
+                                },
+                                {
+                                    "bbox": [
+                                        287,
+                                        301.5,
+                                        502,
+                                        315.0
+                                    ],
+                                    "spans": [],
+                                    "index": 59
+                                },
+                                {
+                                    "bbox": [
+                                        287,
+                                        315.0,
+                                        502,
+                                        328.5
+                                    ],
+                                    "spans": [],
+                                    "index": 60
+                                },
+                                {
+                                    "bbox": [
+                                        287,
+                                        328.5,
+                                        502,
+                                        342.0
+                                    ],
+                                    "spans": [],
+                                    "index": 61
+                                },
+                                {
+                                    "bbox": [
+                                        287,
+                                        342.0,
+                                        502,
+                                        355.5
+                                    ],
+                                    "spans": [],
+                                    "index": 62
+                                },
+                                {
+                                    "bbox": [
+                                        287,
+                                        355.5,
+                                        502,
+                                        369.0
+                                    ],
+                                    "spans": [],
+                                    "index": 63
+                                },
+                                {
+                                    "bbox": [
+                                        287,
+                                        369.0,
+                                        502,
+                                        382.5
+                                    ],
+                                    "spans": [],
+                                    "index": 65
+                                },
+                                {
+                                    "bbox": [
+                                        287,
+                                        382.5,
+                                        502,
+                                        396.0
+                                    ],
+                                    "spans": [],
+                                    "index": 66
+                                },
+                                {
+                                    "bbox": [
+                                        287,
+                                        396.0,
+                                        502,
+                                        409.5
+                                    ],
+                                    "spans": [],
+                                    "index": 64
+                                },
+                                {
+                                    "bbox": [
+                                        287,
+                                        409.5,
+                                        502,
+                                        423.0
+                                    ],
+                                    "spans": [],
+                                    "index": 67
+                                },
+                                {
+                                    "bbox": [
+                                        287,
+                                        423.0,
+                                        502,
+                                        436.5
+                                    ],
+                                    "spans": [],
+                                    "index": 68
+                                },
+                                {
+                                    "bbox": [
+                                        287,
+                                        436.5,
+                                        502,
+                                        450.0
+                                    ],
+                                    "spans": [],
+                                    "index": 69
+                                },
+                                {
+                                    "bbox": [
+                                        287,
+                                        450.0,
+                                        502,
+                                        463.5
+                                    ],
+                                    "spans": [],
+                                    "index": 70
+                                },
+                                {
+                                    "bbox": [
+                                        287,
+                                        463.5,
+                                        502,
+                                        477.0
+                                    ],
+                                    "spans": [],
+                                    "index": 71
+                                },
+                                {
+                                    "bbox": [
+                                        287,
+                                        477.0,
+                                        502,
+                                        490.5
+                                    ],
+                                    "spans": [],
+                                    "index": 72
+                                },
+                                {
+                                    "bbox": [
+                                        287,
+                                        490.5,
+                                        502,
+                                        504.0
+                                    ],
+                                    "spans": [],
+                                    "index": 73
+                                },
+                                {
+                                    "bbox": [
+                                        287,
+                                        504.0,
+                                        502,
+                                        517.5
+                                    ],
+                                    "spans": [],
+                                    "index": 74
+                                },
+                                {
+                                    "bbox": [
+                                        287,
+                                        517.5,
+                                        502,
+                                        531.0
+                                    ],
+                                    "spans": [],
+                                    "index": 75
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                285,
+                                542,
+                                505,
+                                674
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        284,
+                                        540,
+                                        505,
+                                        555
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                284,
+                                                540,
+                                                385,
+                                                555
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 5: (A) Values of",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                386,
+                                                542,
+                                                396,
+                                                553
+                                            ],
+                                            "score": 0.88,
+                                            "content": "\\hat { y _ { 2 } }",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                397,
+                                                540,
+                                                505,
+                                                555
+                                            ],
+                                            "score": 1.0,
+                                            "content": "in the supervised learning",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 76
+                                },
+                                {
+                                    "bbox": [
+                                        284,
+                                        553,
+                                        505,
+                                        565
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                284,
+                                                553,
+                                                505,
+                                                565
+                                            ],
+                                            "score": 1.0,
+                                            "content": "(SL) unit test. Larger values mean sacrificing present",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 77
+                                },
+                                {
+                                    "bbox": [
+                                        284,
+                                        564,
+                                        505,
+                                        576
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                284,
+                                                564,
+                                                505,
+                                                576
+                                            ],
+                                            "score": 1.0,
+                                            "content": "performance for future performance (i.e. non-myopic",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 78
+                                },
+                                {
+                                    "bbox": [
+                                        284,
+                                        575,
+                                        505,
+                                        587
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                284,
+                                                575,
+                                                505,
+                                                587
+                                            ],
+                                            "score": 1.0,
+                                            "content": "exploitation of ADS). (B) Average level of non-myopic",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 79
+                                },
+                                {
+                                    "bbox": [
+                                        284,
+                                        586,
+                                        505,
+                                        597
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                284,
+                                                586,
+                                                505,
+                                                597
+                                            ],
+                                            "score": 1.0,
+                                            "content": "cooperate behavior observed in the RL unit test for",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 80
+                                },
+                                {
+                                    "bbox": [
+                                        283,
+                                        596,
+                                        506,
+                                        609
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                283,
+                                                596,
+                                                506,
+                                                609
+                                            ],
+                                            "score": 1.0,
+                                            "content": "HI-ADS, with two meta-learning algorithms (B1) PBT",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 81
+                                },
+                                {
+                                    "bbox": [
+                                        284,
+                                        608,
+                                        505,
+                                        620
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                284,
+                                                608,
+                                                505,
+                                                620
+                                            ],
+                                            "score": 1.0,
+                                            "content": "and (B2) REINFORCE. Lower is better, since the goal",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 82
+                                },
+                                {
+                                    "bbox": [
+                                        284,
+                                        619,
+                                        505,
+                                        631
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                284,
+                                                619,
+                                                505,
+                                                631
+                                            ],
+                                            "score": 1.0,
+                                            "content": "is for non-myopic incentives to remain hidden. Despite",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 83
+                                },
+                                {
+                                    "bbox": [
+                                        283,
+                                        629,
+                                        505,
+                                        642
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                283,
+                                                629,
+                                                505,
+                                                642
+                                            ],
+                                            "score": 1.0,
+                                            "content": "the inner loop being fully myopic (simple MLP in the",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 85
+                                },
+                                {
+                                    "bbox": [
+                                        283,
+                                        640,
+                                        506,
+                                        654
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                283,
+                                                640,
+                                                317,
+                                                654
+                                            ],
+                                            "score": 1.0,
+                                            "content": "SL test,",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                318,
+                                                641,
+                                                343,
+                                                652
+                                            ],
+                                            "score": 0.91,
+                                            "content": "\\gamma = 0",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                344,
+                                                640,
+                                                506,
+                                                654
+                                            ],
+                                            "score": 1.0,
+                                            "content": "in RL test), in all cases outer-loop (OL)",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 86
+                                },
+                                {
+                                    "bbox": [
+                                        283,
+                                        651,
+                                        507,
+                                        665
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                283,
+                                                651,
+                                                507,
+                                                665
+                                            ],
+                                            "score": 1.0,
+                                            "content": "optimizers reveal HI-ADS (top rows). Context swap-",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 87
+                                },
+                                {
+                                    "bbox": [
+                                        283,
+                                        662,
+                                        497,
+                                        675
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                283,
+                                                662,
+                                                497,
+                                                675
+                                            ],
+                                            "score": 1.0,
+                                            "content": "ping significantly mitigates HI-ADS (bottom rows).",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 88
+                                }
+                            ],
+                            "index": 81.5
+                        }
+                    ],
+                    "index": 73.0
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        688,
+                        505,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                687,
+                                506,
+                                700
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        687,
+                                        506,
+                                        700
+                                    ],
+                                    "score": 1.0,
+                                    "content": "outer-loop optimizer. The outer-loop optimizes the parameters to maximize the summed reward of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 89
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                699,
+                                503,
+                                712
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        699,
+                                        138,
+                                        712
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the last",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        138,
+                                        700,
+                                        147,
+                                        709
+                                    ],
+                                    "score": 0.8,
+                                    "content": "T",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        147,
+                                        699,
+                                        475,
+                                        712
+                                    ],
+                                    "score": 1.0,
+                                    "content": "time-steps. As with PBT, we observe non-myopic behavior, but now only when",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        475,
+                                        699,
+                                        503,
+                                        709
+                                    ],
+                                    "score": 0.89,
+                                    "content": "T > 1",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 90
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                709,
+                                506,
+                                723
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        709,
+                                        506,
+                                        723
+                                    ],
+                                    "score": 1.0,
+                                    "content": "This supports our hypothesis that exploitation of HI-ADS is due not to PBT in particular, but just to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 91
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                720,
+                                389,
+                                733
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        720,
+                                        389,
+                                        733
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the introduction of sufficiently powerful meta-learning. See Fig. 5 B2.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 92
+                        }
+                    ],
+                    "index": 90.5
+                }
+            ],
+            "page_idx": 5,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        107,
+                        27,
+                        307,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                25,
+                                308,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        25,
+                                        308,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Under review as a conference paper at ICLR 2021",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        302,
+                        752,
+                        309,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                302,
+                                751,
+                                310,
+                                762
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        302,
+                                        751,
+                                        310,
+                                        762
+                                    ],
+                                    "score": 1.0,
+                                    "content": "6",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "title",
+                    "bbox": [
+                        106,
+                        82,
+                        416,
+                        93
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                82,
+                                417,
+                                95
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        82,
+                                        417,
+                                        95
+                                    ],
+                                    "score": 1.0,
+                                    "content": "5.1.1 HI-ADS UNIT TESTS EXPERIMENTAL RESULTS AND DISCUSSION",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        }
+                    ],
+                    "index": 0
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        87,
+                        505,
+                        160
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                93,
+                                505,
+                                106
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        93,
+                                        505,
+                                        106
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We first show that agents trained with PBT fail the unit tests more often than “vanilla” algorithms",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                105,
+                                505,
+                                117
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        105,
+                                        505,
+                                        117
+                                    ],
+                                    "score": 1.0,
+                                    "content": "which do not use meta-learning. We initialize the learning rate log-uniformly between 0.01 and 1.0 for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                116,
+                                505,
+                                128
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        116,
+                                        505,
+                                        128
+                                    ],
+                                    "score": 1.0,
+                                    "content": "all experiments (whether using PBT or not). We expect and confirm that the following two factors lead",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                127,
+                                505,
+                                139
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        127,
+                                        505,
+                                        139
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to higher rates of unit test failure: (1) Shorter intervals: These give the OL more opportunities to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                137,
+                                506,
+                                151
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        137,
+                                        506,
+                                        151
+                                    ],
+                                    "score": 1.0,
+                                    "content": "influence the population. (2) Larger populations: These make outliers with exceptional non-myopic",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                149,
+                                420,
+                                162
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        149,
+                                        420,
+                                        162
+                                    ],
+                                    "score": 1.0,
+                                    "content": "performance more likely, and OL makes them likely to survive and propagate.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        }
+                    ],
+                    "index": 3.5,
+                    "bbox_fs": [
+                        105,
+                        93,
+                        506,
+                        162
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        165,
+                        505,
+                        199
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                165,
+                                506,
+                                178
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        165,
+                                        506,
+                                        178
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The baseline (no meta-learning) algorithms all pass the unit tests: hidden incentives are almost never",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                177,
+                                505,
+                                189
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        177,
+                                        505,
+                                        189
+                                    ],
+                                    "score": 1.0,
+                                    "content": "revealed - see blue curves in Fig. 5. However, agents trained with meta-learning and large populations",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                187,
+                                361,
+                                200
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        187,
+                                        361,
+                                        200
+                                    ],
+                                    "score": 1.0,
+                                    "content": "often fail the unit tests: see orange curves in top rows of Fig. 5.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        }
+                    ],
+                    "index": 8,
+                    "bbox_fs": [
+                        105,
+                        165,
+                        506,
+                        200
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        205,
+                        277,
+                        424
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                204,
+                                279,
+                                217
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        204,
+                                        279,
+                                        217
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Furthermore, we verify that context swap-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                216,
+                                278,
+                                227
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        216,
+                                        278,
+                                        227
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ping significantly mitigates the effect of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                226,
+                                278,
+                                239
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        226,
+                                        278,
+                                        239
+                                    ],
+                                    "score": 1.0,
+                                    "content": "HI-ADS in both unit tests, decreasing un-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                237,
+                                278,
+                                249
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        237,
+                                        278,
+                                        249
+                                    ],
+                                    "score": 1.0,
+                                    "content": "desirable behaviour to near-baseline lev-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                248,
+                                279,
+                                260
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        248,
+                                        279,
+                                        260
+                                    ],
+                                    "score": 1.0,
+                                    "content": "els - see bottom rows of Fig. 5. This ef-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                259,
+                                278,
+                                271
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        259,
+                                        278,
+                                        271
+                                    ],
+                                    "score": 1.0,
+                                    "content": "fect can be explained as follows: Because",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                271,
+                                278,
+                                282
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        271,
+                                        278,
+                                        282
+                                    ],
+                                    "score": 1.0,
+                                    "content": "context swapping transfers the benefits of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                282,
+                                277,
+                                292
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        282,
+                                        277,
+                                        292
+                                    ],
+                                    "score": 1.0,
+                                    "content": "one learner’s action to the next learner to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                291,
+                                277,
+                                304
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        291,
+                                        277,
+                                        304
+                                    ],
+                                    "score": 1.0,
+                                    "content": "inhabit that environment, it increases the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                303,
+                                279,
+                                315
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        303,
+                                        279,
+                                        315
+                                    ],
+                                    "score": 1.0,
+                                    "content": "second learner’s fitness, and thereby re-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                313,
+                                278,
+                                326
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        313,
+                                        278,
+                                        326
+                                    ],
+                                    "score": 1.0,
+                                    "content": "duces the relative fitness (as evaluated by",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                325,
+                                277,
+                                338
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        325,
+                                        277,
+                                        338
+                                    ],
+                                    "score": 1.0,
+                                    "content": "PBT’s EXPLOIT step) of the non-myopic",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                336,
+                                278,
+                                348
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        336,
+                                        278,
+                                        348
+                                    ],
+                                    "score": 1.0,
+                                    "content": "cooperate behaviour. We observe some",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                347,
+                                278,
+                                358
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        347,
+                                        278,
+                                        358
+                                    ],
+                                    "score": 1.0,
+                                    "content": "interesting exceptions with the combina-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                358,
+                                278,
+                                369
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        358,
+                                        278,
+                                        369
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tion of small populations and short PBT",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                369,
+                                277,
+                                381
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        369,
+                                        277,
+                                        381
+                                    ],
+                                    "score": 1.0,
+                                    "content": "intervals: Although context swapping still",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                380,
+                                279,
+                                391
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        380,
+                                        279,
+                                        391
+                                    ],
+                                    "score": 1.0,
+                                    "content": "significantly decreases the effect of HI-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                390,
+                                278,
+                                402
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        390,
+                                        278,
+                                        402
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ADS, non-myopic cooperate behaviour",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                401,
+                                278,
+                                414
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        401,
+                                        196,
+                                        414
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is observed as much as",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        196,
+                                        402,
+                                        216,
+                                        412
+                                    ],
+                                    "score": 0.87,
+                                    "content": "20 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        216,
+                                        401,
+                                        278,
+                                        414
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of the time (for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                412,
+                                279,
+                                425
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        412,
+                                        142,
+                                        425
+                                    ],
+                                    "score": 1.0,
+                                    "content": "#learner",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        143,
+                                        413,
+                                        159,
+                                        423
+                                    ],
+                                    "score": 0.77,
+                                    "content": "_ { = 1 0 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        159,
+                                        412,
+                                        162,
+                                        425
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        163,
+                                        413,
+                                        190,
+                                        423
+                                    ],
+                                    "score": 0.88,
+                                    "content": "T = 1",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        190,
+                                        412,
+                                        279,
+                                        425
+                                    ],
+                                    "score": 1.0,
+                                    "content": "; see bottom-left plot).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        }
+                    ],
+                    "index": 19.5,
+                    "bbox_fs": [
+                        105,
+                        204,
+                        279,
+                        425
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        429,
+                        277,
+                        660
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                429,
+                                277,
+                                441
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        429,
+                                        277,
+                                        441
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We also observe that PBT reveals HI-ADS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                440,
+                                277,
+                                453
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        440,
+                                        153,
+                                        453
+                                    ],
+                                    "score": 1.0,
+                                    "content": "even when",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        153,
+                                        441,
+                                        180,
+                                        451
+                                    ],
+                                    "score": 0.9,
+                                    "content": "T = 1",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        181,
+                                        440,
+                                        277,
+                                        453
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", where the explanation",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                451,
+                                279,
+                                464
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        451,
+                                        279,
+                                        464
+                                    ],
+                                    "score": 1.0,
+                                    "content": "that PBT operates on a longer time hori-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                462,
+                                277,
+                                474
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        462,
+                                        277,
+                                        474
+                                    ],
+                                    "score": 1.0,
+                                    "content": "zon than the inner loop does not apply. We",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                474,
+                                277,
+                                485
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        474,
+                                        277,
+                                        485
+                                    ],
+                                    "score": 1.0,
+                                    "content": "provide a detailed explanation for how this",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                484,
+                                277,
+                                497
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        484,
+                                        277,
+                                        497
+                                    ],
+                                    "score": 1.0,
+                                    "content": "might happen in Appendix 9.1.2, but in",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                496,
+                                278,
+                                507
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        496,
+                                        278,
+                                        507
+                                    ],
+                                    "score": 1.0,
+                                    "content": "summary, we hypothesize that there are at",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                506,
+                                279,
+                                518
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        506,
+                                        279,
+                                        518
+                                    ],
+                                    "score": 1.0,
+                                    "content": "least 2 mechanisms by which PBT is reveal-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                517,
+                                278,
+                                530
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        517,
+                                        278,
+                                        530
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ing HI-ADS: (1) optimizing over a longer",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                527,
+                                279,
+                                541
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        527,
+                                        279,
+                                        541
+                                    ],
+                                    "score": 1.0,
+                                    "content": "time-scale, and (2) picking up on the cor-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                539,
+                                277,
+                                551
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        539,
+                                        277,
+                                        551
+                                    ],
+                                    "score": 1.0,
+                                    "content": "relation between an agent’s current policy",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                550,
+                                277,
+                                563
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        550,
+                                        277,
+                                        563
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and the underlying state. Mechanism (2)",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                561,
+                                278,
+                                574
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        561,
+                                        278,
+                                        574
+                                    ],
+                                    "score": 1.0,
+                                    "content": "can be explained informally as reasoning",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                572,
+                                279,
+                                585
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        572,
+                                        279,
+                                        585
+                                    ],
+                                    "score": 1.0,
+                                    "content": "as: “If I’m cooperating, then I was prob-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                583,
+                                278,
+                                595
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        583,
+                                        278,
+                                        595
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ably cooperating on the last time-step as",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                594,
+                                278,
+                                606
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        594,
+                                        278,
+                                        606
+                                    ],
+                                    "score": 1.0,
+                                    "content": "well, so my reward should be higher”. As",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                606,
+                                279,
+                                617
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        606,
+                                        279,
+                                        617
+                                    ],
+                                    "score": 1.0,
+                                    "content": "support for these hypotheses, we run con-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                616,
+                                278,
+                                628
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        616,
+                                        278,
+                                        628
+                                    ],
+                                    "score": 1.0,
+                                    "content": "trol experiments identifying two algorithms",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 47
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                626,
+                                277,
+                                639
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        626,
+                                        277,
+                                        639
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(each sharing only one of these properties)",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 48
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                636,
+                                277,
+                                651
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        636,
+                                        277,
+                                        651
+                                    ],
+                                    "score": 1.0,
+                                    "content": "that can fail the unit test. Context swapping",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 49
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                649,
+                                180,
+                                660
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        649,
+                                        180,
+                                        660
+                                    ],
+                                    "score": 1.0,
+                                    "content": "remains effective.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 50
+                        }
+                    ],
+                    "index": 40,
+                    "bbox_fs": [
+                        105,
+                        429,
+                        279,
+                        660
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        106,
+                        666,
+                        277,
+                        688
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                665,
+                                279,
+                                678
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        665,
+                                        279,
+                                        678
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(1) Optimizing over a longer time-scale:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 51
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                676,
+                                277,
+                                689
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        676,
+                                        277,
+                                        689
+                                    ],
+                                    "score": 1.0,
+                                    "content": "replacing PBT with REINFORCE as an",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 52
+                        }
+                    ],
+                    "index": 51.5
+                },
+                {
+                    "type": "image",
+                    "bbox": [
+                        287,
+                        207,
+                        502,
+                        531
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                287,
+                                207,
+                                502,
+                                531
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        287,
+                                        207,
+                                        502,
+                                        531
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                287,
+                                                207,
+                                                502,
+                                                531
+                                            ],
+                                            "score": 0.967,
+                                            "type": "image",
+                                            "image_path": "fb963b752d78b641eafae0011c4187a77f39ee589e94374f2b91ae0ecc03c850.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 64.5,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        287,
+                                        207,
+                                        502,
+                                        220.5
+                                    ],
+                                    "spans": [],
+                                    "index": 53
+                                },
+                                {
+                                    "bbox": [
+                                        287,
+                                        220.5,
+                                        502,
+                                        234.0
+                                    ],
+                                    "spans": [],
+                                    "index": 54
+                                },
+                                {
+                                    "bbox": [
+                                        287,
+                                        234.0,
+                                        502,
+                                        247.5
+                                    ],
+                                    "spans": [],
+                                    "index": 55
+                                },
+                                {
+                                    "bbox": [
+                                        287,
+                                        247.5,
+                                        502,
+                                        261.0
+                                    ],
+                                    "spans": [],
+                                    "index": 84
+                                },
+                                {
+                                    "bbox": [
+                                        287,
+                                        261.0,
+                                        502,
+                                        274.5
+                                    ],
+                                    "spans": [],
+                                    "index": 56
+                                },
+                                {
+                                    "bbox": [
+                                        287,
+                                        274.5,
+                                        502,
+                                        288.0
+                                    ],
+                                    "spans": [],
+                                    "index": 57
+                                },
+                                {
+                                    "bbox": [
+                                        287,
+                                        288.0,
+                                        502,
+                                        301.5
+                                    ],
+                                    "spans": [],
+                                    "index": 58
+                                },
+                                {
+                                    "bbox": [
+                                        287,
+                                        301.5,
+                                        502,
+                                        315.0
+                                    ],
+                                    "spans": [],
+                                    "index": 59
+                                },
+                                {
+                                    "bbox": [
+                                        287,
+                                        315.0,
+                                        502,
+                                        328.5
+                                    ],
+                                    "spans": [],
+                                    "index": 60
+                                },
+                                {
+                                    "bbox": [
+                                        287,
+                                        328.5,
+                                        502,
+                                        342.0
+                                    ],
+                                    "spans": [],
+                                    "index": 61
+                                },
+                                {
+                                    "bbox": [
+                                        287,
+                                        342.0,
+                                        502,
+                                        355.5
+                                    ],
+                                    "spans": [],
+                                    "index": 62
+                                },
+                                {
+                                    "bbox": [
+                                        287,
+                                        355.5,
+                                        502,
+                                        369.0
+                                    ],
+                                    "spans": [],
+                                    "index": 63
+                                },
+                                {
+                                    "bbox": [
+                                        287,
+                                        369.0,
+                                        502,
+                                        382.5
+                                    ],
+                                    "spans": [],
+                                    "index": 65
+                                },
+                                {
+                                    "bbox": [
+                                        287,
+                                        382.5,
+                                        502,
+                                        396.0
+                                    ],
+                                    "spans": [],
+                                    "index": 66
+                                },
+                                {
+                                    "bbox": [
+                                        287,
+                                        396.0,
+                                        502,
+                                        409.5
+                                    ],
+                                    "spans": [],
+                                    "index": 64
+                                },
+                                {
+                                    "bbox": [
+                                        287,
+                                        409.5,
+                                        502,
+                                        423.0
+                                    ],
+                                    "spans": [],
+                                    "index": 67
+                                },
+                                {
+                                    "bbox": [
+                                        287,
+                                        423.0,
+                                        502,
+                                        436.5
+                                    ],
+                                    "spans": [],
+                                    "index": 68
+                                },
+                                {
+                                    "bbox": [
+                                        287,
+                                        436.5,
+                                        502,
+                                        450.0
+                                    ],
+                                    "spans": [],
+                                    "index": 69
+                                },
+                                {
+                                    "bbox": [
+                                        287,
+                                        450.0,
+                                        502,
+                                        463.5
+                                    ],
+                                    "spans": [],
+                                    "index": 70
+                                },
+                                {
+                                    "bbox": [
+                                        287,
+                                        463.5,
+                                        502,
+                                        477.0
+                                    ],
+                                    "spans": [],
+                                    "index": 71
+                                },
+                                {
+                                    "bbox": [
+                                        287,
+                                        477.0,
+                                        502,
+                                        490.5
+                                    ],
+                                    "spans": [],
+                                    "index": 72
+                                },
+                                {
+                                    "bbox": [
+                                        287,
+                                        490.5,
+                                        502,
+                                        504.0
+                                    ],
+                                    "spans": [],
+                                    "index": 73
+                                },
+                                {
+                                    "bbox": [
+                                        287,
+                                        504.0,
+                                        502,
+                                        517.5
+                                    ],
+                                    "spans": [],
+                                    "index": 74
+                                },
+                                {
+                                    "bbox": [
+                                        287,
+                                        517.5,
+                                        502,
+                                        531.0
+                                    ],
+                                    "spans": [],
+                                    "index": 75
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                285,
+                                542,
+                                505,
+                                674
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        284,
+                                        540,
+                                        505,
+                                        555
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                284,
+                                                540,
+                                                385,
+                                                555
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 5: (A) Values of",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                386,
+                                                542,
+                                                396,
+                                                553
+                                            ],
+                                            "score": 0.88,
+                                            "content": "\\hat { y _ { 2 } }",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                397,
+                                                540,
+                                                505,
+                                                555
+                                            ],
+                                            "score": 1.0,
+                                            "content": "in the supervised learning",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 76
+                                },
+                                {
+                                    "bbox": [
+                                        284,
+                                        553,
+                                        505,
+                                        565
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                284,
+                                                553,
+                                                505,
+                                                565
+                                            ],
+                                            "score": 1.0,
+                                            "content": "(SL) unit test. Larger values mean sacrificing present",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 77
+                                },
+                                {
+                                    "bbox": [
+                                        284,
+                                        564,
+                                        505,
+                                        576
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                284,
+                                                564,
+                                                505,
+                                                576
+                                            ],
+                                            "score": 1.0,
+                                            "content": "performance for future performance (i.e. non-myopic",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 78
+                                },
+                                {
+                                    "bbox": [
+                                        284,
+                                        575,
+                                        505,
+                                        587
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                284,
+                                                575,
+                                                505,
+                                                587
+                                            ],
+                                            "score": 1.0,
+                                            "content": "exploitation of ADS). (B) Average level of non-myopic",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 79
+                                },
+                                {
+                                    "bbox": [
+                                        284,
+                                        586,
+                                        505,
+                                        597
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                284,
+                                                586,
+                                                505,
+                                                597
+                                            ],
+                                            "score": 1.0,
+                                            "content": "cooperate behavior observed in the RL unit test for",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 80
+                                },
+                                {
+                                    "bbox": [
+                                        283,
+                                        596,
+                                        506,
+                                        609
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                283,
+                                                596,
+                                                506,
+                                                609
+                                            ],
+                                            "score": 1.0,
+                                            "content": "HI-ADS, with two meta-learning algorithms (B1) PBT",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 81
+                                },
+                                {
+                                    "bbox": [
+                                        284,
+                                        608,
+                                        505,
+                                        620
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                284,
+                                                608,
+                                                505,
+                                                620
+                                            ],
+                                            "score": 1.0,
+                                            "content": "and (B2) REINFORCE. Lower is better, since the goal",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 82
+                                },
+                                {
+                                    "bbox": [
+                                        284,
+                                        619,
+                                        505,
+                                        631
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                284,
+                                                619,
+                                                505,
+                                                631
+                                            ],
+                                            "score": 1.0,
+                                            "content": "is for non-myopic incentives to remain hidden. Despite",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 83
+                                },
+                                {
+                                    "bbox": [
+                                        283,
+                                        629,
+                                        505,
+                                        642
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                283,
+                                                629,
+                                                505,
+                                                642
+                                            ],
+                                            "score": 1.0,
+                                            "content": "the inner loop being fully myopic (simple MLP in the",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 85
+                                },
+                                {
+                                    "bbox": [
+                                        283,
+                                        640,
+                                        506,
+                                        654
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                283,
+                                                640,
+                                                317,
+                                                654
+                                            ],
+                                            "score": 1.0,
+                                            "content": "SL test,",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                318,
+                                                641,
+                                                343,
+                                                652
+                                            ],
+                                            "score": 0.91,
+                                            "content": "\\gamma = 0",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                344,
+                                                640,
+                                                506,
+                                                654
+                                            ],
+                                            "score": 1.0,
+                                            "content": "in RL test), in all cases outer-loop (OL)",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 86
+                                },
+                                {
+                                    "bbox": [
+                                        283,
+                                        651,
+                                        507,
+                                        665
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                283,
+                                                651,
+                                                507,
+                                                665
+                                            ],
+                                            "score": 1.0,
+                                            "content": "optimizers reveal HI-ADS (top rows). Context swap-",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 87
+                                },
+                                {
+                                    "bbox": [
+                                        283,
+                                        662,
+                                        497,
+                                        675
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                283,
+                                                662,
+                                                497,
+                                                675
+                                            ],
+                                            "score": 1.0,
+                                            "content": "ping significantly mitigates HI-ADS (bottom rows).",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 88
+                                }
+                            ],
+                            "index": 81.5
+                        }
+                    ],
+                    "index": 73.0
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        688,
+                        505,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                687,
+                                506,
+                                700
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        687,
+                                        506,
+                                        700
+                                    ],
+                                    "score": 1.0,
+                                    "content": "outer-loop optimizer. The outer-loop optimizes the parameters to maximize the summed reward of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 89
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                699,
+                                503,
+                                712
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        699,
+                                        138,
+                                        712
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the last",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        138,
+                                        700,
+                                        147,
+                                        709
+                                    ],
+                                    "score": 0.8,
+                                    "content": "T",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        147,
+                                        699,
+                                        475,
+                                        712
+                                    ],
+                                    "score": 1.0,
+                                    "content": "time-steps. As with PBT, we observe non-myopic behavior, but now only when",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        475,
+                                        699,
+                                        503,
+                                        709
+                                    ],
+                                    "score": 0.89,
+                                    "content": "T > 1",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 90
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                709,
+                                506,
+                                723
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        709,
+                                        506,
+                                        723
+                                    ],
+                                    "score": 1.0,
+                                    "content": "This supports our hypothesis that exploitation of HI-ADS is due not to PBT in particular, but just to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 91
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                720,
+                                389,
+                                733
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        720,
+                                        389,
+                                        733
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the introduction of sufficiently powerful meta-learning. See Fig. 5 B2.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 92
+                        }
+                    ],
+                    "index": 90.5,
+                    "bbox_fs": [
+                        105,
+                        687,
+                        506,
+                        733
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        83,
+                        297,
+                        236
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                82,
+                                297,
+                                95
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        82,
+                                        297,
+                                        95
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(2) Exploiting correlation: Q-learning with",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                93,
+                                297,
+                                106
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        94,
+                                        132,
+                                        105
+                                    ],
+                                    "score": 0.9,
+                                    "content": "\\gamma = 0",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        132,
+                                        93,
+                                        143,
+                                        106
+                                    ],
+                                    "score": 1.0,
+                                    "content": "an",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        144,
+                                        94,
+                                        175,
+                                        104
+                                    ],
+                                    "score": 0.89,
+                                    "content": "\\epsilon = 0 . 1",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        175,
+                                        93,
+                                        297,
+                                        106
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-greedy behavior policy and no",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                105,
+                                298,
+                                117
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        105,
+                                        298,
+                                        117
+                                    ],
+                                    "score": 1.0,
+                                    "content": "meta-learning. If either state was equally likely,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                115,
+                                299,
+                                127
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        115,
+                                        122,
+                                        127
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        122,
+                                        116,
+                                        131,
+                                        127
+                                    ],
+                                    "score": 0.34,
+                                    "content": "\\mathbf { Q }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        131,
+                                        115,
+                                        299,
+                                        127
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-values would be the average of the val-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                127,
+                                298,
+                                137
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        127,
+                                        298,
+                                        137
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ues in each column in Table 1, so the estimated",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                107,
+                                137,
+                                299,
+                                150
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        137,
+                                        149,
+                                        149
+                                    ],
+                                    "score": 0.48,
+                                    "content": "Q ( { \\mathrm { d e f e c t } } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        149,
+                                        137,
+                                        244,
+                                        150
+                                    ],
+                                    "score": 1.0,
+                                    "content": "would be larger. But the",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        245,
+                                        140,
+                                        250,
+                                        147
+                                    ],
+                                    "score": 0.38,
+                                    "content": "\\epsilon",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        250,
+                                        137,
+                                        299,
+                                        150
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-greedy pol-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                149,
+                                298,
+                                160
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        149,
+                                        298,
+                                        160
+                                    ],
+                                    "score": 1.0,
+                                    "content": "icy correlates the previous action (i.e. the current",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                159,
+                                297,
+                                172
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        159,
+                                        297,
+                                        172
+                                    ],
+                                    "score": 1.0,
+                                    "content": "state) and current action (so long as the policy",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                170,
+                                299,
+                                183
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        170,
+                                        299,
+                                        183
+                                    ],
+                                    "score": 1.0,
+                                    "content": "did not just change), so the top-left and bottom-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                182,
+                                298,
+                                194
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        182,
+                                        298,
+                                        194
+                                    ],
+                                    "score": 1.0,
+                                    "content": "right entries carry more weight in the estimates,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                192,
+                                298,
+                                204
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        192,
+                                        183,
+                                        204
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sometimes causing",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        184,
+                                        192,
+                                        249,
+                                        204
+                                    ],
+                                    "score": 0.84,
+                                    "content": "Q ( { \\mathrm { d e f e c t } } ) \\approx Q (",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        249,
+                                        192,
+                                        298,
+                                        204
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(cooperate)",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                203,
+                                298,
+                                215
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        203,
+                                        298,
+                                        215
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and persistent nonmyopic behavior. See Fig. 6",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                214,
+                                297,
+                                226
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        214,
+                                        297,
+                                        226
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for results, Appendix 9.1.4 for more results, and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                225,
+                                270,
+                                237
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        225,
+                                        270,
+                                        237
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Appendix 9.1.3 for experimental details.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        }
+                    ],
+                    "index": 6.5
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        109,
+                        250,
+                        258,
+                        261
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                249,
+                                261,
+                                263
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        249,
+                                        261,
+                                        263
+                                    ],
+                                    "score": 1.0,
+                                    "content": "5.2 CONTENT RECOMMENDATION",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        }
+                    ],
+                    "index": 14
+                },
+                {
+                    "type": "image",
+                    "bbox": [
+                        304,
+                        83,
+                        503,
+                        166
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                304,
+                                83,
+                                503,
+                                166
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        304,
+                                        83,
+                                        503,
+                                        166
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                304,
+                                                83,
+                                                503,
+                                                166
+                                            ],
+                                            "score": 0.967,
+                                            "type": "image",
+                                            "image_path": "c27ce5d1176c26223b5a51441be7d7fafa717c25c484532ba211b635ec5f8ba5.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 17.5,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        304,
+                                        83,
+                                        503,
+                                        96.83333333333333
+                                    ],
+                                    "spans": [],
+                                    "index": 15
+                                },
+                                {
+                                    "bbox": [
+                                        304,
+                                        96.83333333333333,
+                                        503,
+                                        110.66666666666666
+                                    ],
+                                    "spans": [],
+                                    "index": 16
+                                },
+                                {
+                                    "bbox": [
+                                        304,
+                                        110.66666666666666,
+                                        503,
+                                        124.49999999999999
+                                    ],
+                                    "spans": [],
+                                    "index": 17
+                                },
+                                {
+                                    "bbox": [
+                                        304,
+                                        124.49999999999999,
+                                        503,
+                                        138.33333333333331
+                                    ],
+                                    "spans": [],
+                                    "index": 18
+                                },
+                                {
+                                    "bbox": [
+                                        304,
+                                        138.33333333333331,
+                                        503,
+                                        152.16666666666666
+                                    ],
+                                    "spans": [],
+                                    "index": 19
+                                },
+                                {
+                                    "bbox": [
+                                        304,
+                                        152.16666666666666,
+                                        503,
+                                        166.0
+                                    ],
+                                    "spans": [],
+                                    "index": 20
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                304,
+                                177,
+                                505,
+                                264
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        303,
+                                        175,
+                                        507,
+                                        191
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                303,
+                                                175,
+                                                507,
+                                                191
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 6: Q-learning can fail the unit test, play-",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 21
+                                },
+                                {
+                                    "bbox": [
+                                        303,
+                                        188,
+                                        505,
+                                        200
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                303,
+                                                188,
+                                                320,
+                                                200
+                                            ],
+                                            "score": 1.0,
+                                            "content": "ing",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                320,
+                                                188,
+                                                360,
+                                                199
+                                            ],
+                                            "score": 0.88,
+                                            "content": "{ \\sim } 8 0 { \\cdot } 9 0 \\%",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                360,
+                                                188,
+                                                505,
+                                                200
+                                            ],
+                                            "score": 1.0,
+                                            "content": "cooperate in 3 of 5 experiments",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 22
+                                },
+                                {
+                                    "bbox": [
+                                        304,
+                                        199,
+                                        506,
+                                        211
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                304,
+                                                199,
+                                                506,
+                                                211
+                                            ],
+                                            "score": 1.0,
+                                            "content": "(bottom row). Each column represents an indepen-",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 23
+                                },
+                                {
+                                    "bbox": [
+                                        304,
+                                        209,
+                                        505,
+                                        222
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                304,
+                                                209,
+                                                505,
+                                                222
+                                            ],
+                                            "score": 1.0,
+                                            "content": "dent experiment. Q-values for the cooperate",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 24
+                                },
+                                {
+                                    "bbox": [
+                                        304,
+                                        221,
+                                        505,
+                                        233
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                304,
+                                                221,
+                                                505,
+                                                233
+                                            ],
+                                            "score": 1.0,
+                                            "content": "and defect actions stay tightly coupled in the",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 25
+                                },
+                                {
+                                    "bbox": [
+                                        303,
+                                        231,
+                                        505,
+                                        245
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                303,
+                                                231,
+                                                505,
+                                                245
+                                            ],
+                                            "score": 1.0,
+                                            "content": "failure cases (col. 1,2,5), while in the cases passing",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 26
+                                },
+                                {
+                                    "bbox": [
+                                        303,
+                                        242,
+                                        505,
+                                        255
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                303,
+                                                242,
+                                                505,
+                                                255
+                                            ],
+                                            "score": 1.0,
+                                            "content": "the unit test (col. 3,4) the Q-value of cooperate",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 27
+                                },
+                                {
+                                    "bbox": [
+                                        304,
+                                        254,
+                                        388,
+                                        266
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                304,
+                                                254,
+                                                388,
+                                                266
+                                            ],
+                                            "score": 1.0,
+                                            "content": "decreases over time.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 28
+                                }
+                            ],
+                            "index": 24.5
+                        }
+                    ],
+                    "index": 21.0
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        270,
+                        505,
+                        336
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                270,
+                                505,
+                                283
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        270,
+                                        505,
+                                        283
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We now present a toy environment for modeling content recommendation of news articles, which",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                281,
+                                505,
+                                294
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        281,
+                                        505,
+                                        294
+                                    ],
+                                    "score": 1.0,
+                                    "content": "includes the potential for ADS by incorporating the mechanisms mentioned in Sec. 2.2, discussed as",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                292,
+                                506,
+                                304
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        292,
+                                        506,
+                                        304
+                                    ],
+                                    "score": 1.0,
+                                    "content": "contributing factors to the problems of fake news and filter bubbles. Specifically, the environment",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                303,
+                                506,
+                                316
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        303,
+                                        506,
+                                        316
+                                    ],
+                                    "score": 1.0,
+                                    "content": "assumes that presenting an article to a user can influence (1) their interest in similar articles, and (2)",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                314,
+                                505,
+                                326
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        314,
+                                        505,
+                                        326
+                                    ],
+                                    "score": 1.0,
+                                    "content": "their propensity to use the recommendation service. These correspond to modeling auto-induced",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                325,
+                                504,
+                                338
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        325,
+                                        504,
+                                        338
+                                    ],
+                                    "score": 1.0,
+                                    "content": "concept shift of users, and auto-induced covariate shift of the user base, respectively (see Sec. 2.2).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        }
+                    ],
+                    "index": 31.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        342,
+                        505,
+                        452
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                342,
+                                506,
+                                355
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        342,
+                                        506,
+                                        355
+                                    ],
+                                    "score": 1.0,
+                                    "content": "This environment includes the following components, which change over (discrete) time: User type:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                107,
+                                352,
+                                506,
+                                366
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        354,
+                                        117,
+                                        363
+                                    ],
+                                    "score": 0.85,
+                                    "content": "x ^ { t }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        117,
+                                        352,
+                                        177,
+                                        366
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", Article type:",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        178,
+                                        353,
+                                        188,
+                                        365
+                                    ],
+                                    "score": 0.88,
+                                    "content": "y ^ { t }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        188,
+                                        352,
+                                        257,
+                                        366
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", User interests:",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        257,
+                                        353,
+                                        274,
+                                        363
+                                    ],
+                                    "score": 0.77,
+                                    "content": "\\bar { \\mathbf { W } } ^ { t }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        274,
+                                        352,
+                                        506,
+                                        366
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(propensity for users of each type to click on articles of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                363,
+                                505,
+                                376
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        363,
+                                        229,
+                                        376
+                                    ],
+                                    "score": 1.0,
+                                    "content": "each type), and User loyalty:",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        230,
+                                        364,
+                                        241,
+                                        375
+                                    ],
+                                    "score": 0.87,
+                                    "content": "\\mathbf { g } ^ { t }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        241,
+                                        363,
+                                        505,
+                                        376
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(propensity for users of each type to use the platform). At each",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                375,
+                                505,
+                                388
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        375,
+                                        145,
+                                        388
+                                    ],
+                                    "score": 1.0,
+                                    "content": "time step",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        145,
+                                        376,
+                                        150,
+                                        385
+                                    ],
+                                    "score": 0.71,
+                                    "content": "t",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        150,
+                                        375,
+                                        179,
+                                        388
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", a user",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        180,
+                                        375,
+                                        190,
+                                        385
+                                    ],
+                                    "score": 0.87,
+                                    "content": "x ^ { t }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        190,
+                                        375,
+                                        505,
+                                        388
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is sampled from a categorical distribution, based on the loyalty of the different",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                386,
+                                505,
+                                399
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        386,
+                                        505,
+                                        399
+                                    ],
+                                    "score": 1.0,
+                                    "content": "user types. The recommendation system (a classifier) selects which type of article to present in the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                397,
+                                506,
+                                411
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        397,
+                                        311,
+                                        411
+                                    ],
+                                    "score": 1.0,
+                                    "content": "top position, and finally the user ‘clicks’ an article",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        312,
+                                        397,
+                                        322,
+                                        408
+                                    ],
+                                    "score": 0.88,
+                                    "content": "y ^ { t }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        322,
+                                        397,
+                                        506,
+                                        411
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", according to their interests. User loyalty for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                408,
+                                506,
+                                421
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        408,
+                                        146,
+                                        421
+                                    ],
+                                    "score": 1.0,
+                                    "content": "user type",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        146,
+                                        408,
+                                        156,
+                                        418
+                                    ],
+                                    "score": 0.86,
+                                    "content": "x ^ { t }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        156,
+                                        408,
+                                        443,
+                                        421
+                                    ],
+                                    "score": 1.0,
+                                    "content": "undergoes covariate shift: in accordance with the self-selection effect,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        443,
+                                        408,
+                                        453,
+                                        420
+                                    ],
+                                    "score": 0.87,
+                                    "content": "g ^ { t }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        453,
+                                        408,
+                                        506,
+                                        421
+                                    ],
+                                    "score": 1.0,
+                                    "content": "increases or",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                418,
+                                504,
+                                432
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        418,
+                                        493,
+                                        432
+                                    ],
+                                    "score": 1.0,
+                                    "content": "decreases proportionally to that user type’s interest in the top article. The interests of user type",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        493,
+                                        419,
+                                        504,
+                                        429
+                                    ],
+                                    "score": 0.84,
+                                    "content": "x ^ { t }",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                429,
+                                506,
+                                443
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        429,
+                                        225,
+                                        443
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(represented by a column of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        225,
+                                        430,
+                                        241,
+                                        440
+                                    ],
+                                    "score": 0.83,
+                                    "content": "\\mathbf { W } ^ { t }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        241,
+                                        429,
+                                        506,
+                                        443
+                                    ],
+                                    "score": 1.0,
+                                    "content": ") undergoing concept shift; in accordance with the illusory truth",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                440,
+                                497,
+                                454
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        440,
+                                        497,
+                                        454
+                                    ],
+                                    "score": 1.0,
+                                    "content": "effect, interest in the topic of the top article chosen by the recommender system always increases.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        }
+                    ],
+                    "index": 39.5
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        465,
+                        451,
+                        475
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                464,
+                                452,
+                                477
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        464,
+                                        452,
+                                        477
+                                    ],
+                                    "score": 1.0,
+                                    "content": "5.2.1 CONTENT RECOMMENDATION EXPERIMENTAL RESULTS AND DISCUSSION",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        }
+                    ],
+                    "index": 45
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        483,
+                        506,
+                        583
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                483,
+                                506,
+                                496
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        483,
+                                        506,
+                                        496
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We run 20 trials using an MLP trained with SGD for the recommender system. We find that PBT",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                494,
+                                505,
+                                507
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        494,
+                                        505,
+                                        507
+                                    ],
+                                    "score": 1.0,
+                                    "content": "yields significant improvements in training time and accuracy, but also greater distributional shift",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 47
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                505,
+                                506,
+                                519
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        505,
+                                        506,
+                                        519
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(Fig. 7). User base and user interests both change faster with PBT, and user interests change more",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 48
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                516,
+                                506,
+                                530
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        516,
+                                        506,
+                                        530
+                                    ],
+                                    "score": 1.0,
+                                    "content": "overall. We measure concept/covariate shift using the cosine distance and KL-divergence, respectively.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 49
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                526,
+                                506,
+                                541
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        526,
+                                        506,
+                                        541
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We observe that the distributions over user types typically saturate (to a single user type) after a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 50
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                538,
+                                506,
+                                551
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        538,
+                                        506,
+                                        551
+                                    ],
+                                    "score": 1.0,
+                                    "content": "few hundred time-steps (Fig 1 and Fig. 7, Right). We run long enough to reach such states, to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 51
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                549,
+                                506,
+                                562
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        549,
+                                        506,
+                                        562
+                                    ],
+                                    "score": 1.0,
+                                    "content": "demonstrate that the increase in ADS from PBT is not transitory. The environment has a number of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 52
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                561,
+                                505,
+                                573
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        561,
+                                        476,
+                                        573
+                                    ],
+                                    "score": 1.0,
+                                    "content": "free parameters, and our results are qualitatively consistent so long as the covariate shift rate",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        477,
+                                        561,
+                                        495,
+                                        572
+                                    ],
+                                    "score": 0.82,
+                                    "content": "( \\alpha _ { 1 } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        495,
+                                        561,
+                                        505,
+                                        573
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 53
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                571,
+                                385,
+                                584
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        571,
+                                        237,
+                                        584
+                                    ],
+                                    "score": 1.0,
+                                    "content": "faster than the concept shift rate",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        237,
+                                        572,
+                                        254,
+                                        583
+                                    ],
+                                    "score": 0.78,
+                                    "content": "\\left( \\alpha _ { 2 } \\right)",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        255,
+                                        571,
+                                        385,
+                                        584
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". See Appendix 9.2.1 for details.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 54
+                        }
+                    ],
+                    "index": 50
+                },
+                {
+                    "type": "image",
+                    "bbox": [
+                        189,
+                        592,
+                        421,
+                        681
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                189,
+                                592,
+                                421,
+                                681
+                            ],
+                            "group_id": 1,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        189,
+                                        592,
+                                        421,
+                                        681
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                189,
+                                                592,
+                                                421,
+                                                681
+                                            ],
+                                            "score": 0.965,
+                                            "type": "image",
+                                            "image_path": "95efffd0c2948471a88671bd2e264c0a9ccf142775dc7e578a618fd609eb6f42.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 57.5,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        189,
+                                        592,
+                                        421,
+                                        606.8333333333334
+                                    ],
+                                    "spans": [],
+                                    "index": 55
+                                },
+                                {
+                                    "bbox": [
+                                        189,
+                                        606.8333333333334,
+                                        421,
+                                        621.6666666666667
+                                    ],
+                                    "spans": [],
+                                    "index": 56
+                                },
+                                {
+                                    "bbox": [
+                                        189,
+                                        621.6666666666667,
+                                        421,
+                                        636.5000000000001
+                                    ],
+                                    "spans": [],
+                                    "index": 57
+                                },
+                                {
+                                    "bbox": [
+                                        189,
+                                        636.5000000000001,
+                                        421,
+                                        651.3333333333335
+                                    ],
+                                    "spans": [],
+                                    "index": 58
+                                },
+                                {
+                                    "bbox": [
+                                        189,
+                                        651.3333333333335,
+                                        421,
+                                        666.1666666666669
+                                    ],
+                                    "spans": [],
+                                    "index": 59
+                                },
+                                {
+                                    "bbox": [
+                                        189,
+                                        666.1666666666669,
+                                        421,
+                                        681.0000000000002
+                                    ],
+                                    "spans": [],
+                                    "index": 60
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                107,
+                                691,
+                                505,
+                                725
+                            ],
+                            "group_id": 1,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        691,
+                                        506,
+                                        703
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                691,
+                                                506,
+                                                703
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 7: Content recommendation experiments. Left: using Population Based Training (PBT)",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 61
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        702,
+                                        506,
+                                        715
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                702,
+                                                471,
+                                                715
+                                            ],
+                                            "score": 1.0,
+                                            "content": "increases accuracy of predictions faster, leads to a faster and larger drift in users’ interests,",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                472,
+                                                702,
+                                                502,
+                                                714
+                                            ],
+                                            "score": 0.91,
+                                            "content": "P ( \\boldsymbol { y } | \\mathbf { x } )",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                502,
+                                                702,
+                                                506,
+                                                715
+                                            ],
+                                            "score": 1.0,
+                                            "content": ",",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 62
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        713,
+                                        504,
+                                        725
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                713,
+                                                288,
+                                                725
+                                            ],
+                                            "score": 1.0,
+                                            "content": "(Center); as well as the distribution of users,",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                288,
+                                                713,
+                                                310,
+                                                725
+                                            ],
+                                            "score": 0.91,
+                                            "content": "P ( \\mathbf { x } )",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                311,
+                                                713,
+                                                504,
+                                                725
+                                            ],
+                                            "score": 1.0,
+                                            "content": ", (Right). Shading shows std error over 20 runs.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 63
+                                }
+                            ],
+                            "index": 62
+                        }
+                    ],
+                    "index": 59.75
+                }
+            ],
+            "page_idx": 6,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        108,
+                        27,
+                        306,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                26,
+                                308,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        26,
+                                        308,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Under review as a conference paper at ICLR 2021",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        302,
+                        751,
+                        308,
+                        759
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                302,
+                                750,
+                                309,
+                                762
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        302,
+                                        750,
+                                        309,
+                                        762
+                                    ],
+                                    "score": 1.0,
+                                    "content": "7",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        83,
+                        297,
+                        236
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                82,
+                                297,
+                                95
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        82,
+                                        297,
+                                        95
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(2) Exploiting correlation: Q-learning with",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                93,
+                                297,
+                                106
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        94,
+                                        132,
+                                        105
+                                    ],
+                                    "score": 0.9,
+                                    "content": "\\gamma = 0",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        132,
+                                        93,
+                                        143,
+                                        106
+                                    ],
+                                    "score": 1.0,
+                                    "content": "an",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        144,
+                                        94,
+                                        175,
+                                        104
+                                    ],
+                                    "score": 0.89,
+                                    "content": "\\epsilon = 0 . 1",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        175,
+                                        93,
+                                        297,
+                                        106
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-greedy behavior policy and no",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                105,
+                                298,
+                                117
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        105,
+                                        298,
+                                        117
+                                    ],
+                                    "score": 1.0,
+                                    "content": "meta-learning. If either state was equally likely,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                115,
+                                299,
+                                127
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        115,
+                                        122,
+                                        127
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        122,
+                                        116,
+                                        131,
+                                        127
+                                    ],
+                                    "score": 0.34,
+                                    "content": "\\mathbf { Q }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        131,
+                                        115,
+                                        299,
+                                        127
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-values would be the average of the val-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                127,
+                                298,
+                                137
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        127,
+                                        298,
+                                        137
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ues in each column in Table 1, so the estimated",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                107,
+                                137,
+                                299,
+                                150
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        137,
+                                        149,
+                                        149
+                                    ],
+                                    "score": 0.48,
+                                    "content": "Q ( { \\mathrm { d e f e c t } } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        149,
+                                        137,
+                                        244,
+                                        150
+                                    ],
+                                    "score": 1.0,
+                                    "content": "would be larger. But the",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        245,
+                                        140,
+                                        250,
+                                        147
+                                    ],
+                                    "score": 0.38,
+                                    "content": "\\epsilon",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        250,
+                                        137,
+                                        299,
+                                        150
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-greedy pol-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                149,
+                                298,
+                                160
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        149,
+                                        298,
+                                        160
+                                    ],
+                                    "score": 1.0,
+                                    "content": "icy correlates the previous action (i.e. the current",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                159,
+                                297,
+                                172
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        159,
+                                        297,
+                                        172
+                                    ],
+                                    "score": 1.0,
+                                    "content": "state) and current action (so long as the policy",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                170,
+                                299,
+                                183
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        170,
+                                        299,
+                                        183
+                                    ],
+                                    "score": 1.0,
+                                    "content": "did not just change), so the top-left and bottom-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                182,
+                                298,
+                                194
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        182,
+                                        298,
+                                        194
+                                    ],
+                                    "score": 1.0,
+                                    "content": "right entries carry more weight in the estimates,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                192,
+                                298,
+                                204
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        192,
+                                        183,
+                                        204
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sometimes causing",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        184,
+                                        192,
+                                        249,
+                                        204
+                                    ],
+                                    "score": 0.84,
+                                    "content": "Q ( { \\mathrm { d e f e c t } } ) \\approx Q (",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        249,
+                                        192,
+                                        298,
+                                        204
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(cooperate)",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                203,
+                                298,
+                                215
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        203,
+                                        298,
+                                        215
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and persistent nonmyopic behavior. See Fig. 6",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                214,
+                                297,
+                                226
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        214,
+                                        297,
+                                        226
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for results, Appendix 9.1.4 for more results, and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                225,
+                                270,
+                                237
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        225,
+                                        270,
+                                        237
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Appendix 9.1.3 for experimental details.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        }
+                    ],
+                    "index": 6.5,
+                    "bbox_fs": [
+                        105,
+                        82,
+                        299,
+                        237
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        109,
+                        250,
+                        258,
+                        261
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                249,
+                                261,
+                                263
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        249,
+                                        261,
+                                        263
+                                    ],
+                                    "score": 1.0,
+                                    "content": "5.2 CONTENT RECOMMENDATION",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        }
+                    ],
+                    "index": 14
+                },
+                {
+                    "type": "image",
+                    "bbox": [
+                        304,
+                        83,
+                        503,
+                        166
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                304,
+                                83,
+                                503,
+                                166
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        304,
+                                        83,
+                                        503,
+                                        166
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                304,
+                                                83,
+                                                503,
+                                                166
+                                            ],
+                                            "score": 0.967,
+                                            "type": "image",
+                                            "image_path": "c27ce5d1176c26223b5a51441be7d7fafa717c25c484532ba211b635ec5f8ba5.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 17.5,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        304,
+                                        83,
+                                        503,
+                                        96.83333333333333
+                                    ],
+                                    "spans": [],
+                                    "index": 15
+                                },
+                                {
+                                    "bbox": [
+                                        304,
+                                        96.83333333333333,
+                                        503,
+                                        110.66666666666666
+                                    ],
+                                    "spans": [],
+                                    "index": 16
+                                },
+                                {
+                                    "bbox": [
+                                        304,
+                                        110.66666666666666,
+                                        503,
+                                        124.49999999999999
+                                    ],
+                                    "spans": [],
+                                    "index": 17
+                                },
+                                {
+                                    "bbox": [
+                                        304,
+                                        124.49999999999999,
+                                        503,
+                                        138.33333333333331
+                                    ],
+                                    "spans": [],
+                                    "index": 18
+                                },
+                                {
+                                    "bbox": [
+                                        304,
+                                        138.33333333333331,
+                                        503,
+                                        152.16666666666666
+                                    ],
+                                    "spans": [],
+                                    "index": 19
+                                },
+                                {
+                                    "bbox": [
+                                        304,
+                                        152.16666666666666,
+                                        503,
+                                        166.0
+                                    ],
+                                    "spans": [],
+                                    "index": 20
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                304,
+                                177,
+                                505,
+                                264
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        303,
+                                        175,
+                                        507,
+                                        191
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                303,
+                                                175,
+                                                507,
+                                                191
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 6: Q-learning can fail the unit test, play-",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 21
+                                },
+                                {
+                                    "bbox": [
+                                        303,
+                                        188,
+                                        505,
+                                        200
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                303,
+                                                188,
+                                                320,
+                                                200
+                                            ],
+                                            "score": 1.0,
+                                            "content": "ing",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                320,
+                                                188,
+                                                360,
+                                                199
+                                            ],
+                                            "score": 0.88,
+                                            "content": "{ \\sim } 8 0 { \\cdot } 9 0 \\%",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                360,
+                                                188,
+                                                505,
+                                                200
+                                            ],
+                                            "score": 1.0,
+                                            "content": "cooperate in 3 of 5 experiments",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 22
+                                },
+                                {
+                                    "bbox": [
+                                        304,
+                                        199,
+                                        506,
+                                        211
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                304,
+                                                199,
+                                                506,
+                                                211
+                                            ],
+                                            "score": 1.0,
+                                            "content": "(bottom row). Each column represents an indepen-",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 23
+                                },
+                                {
+                                    "bbox": [
+                                        304,
+                                        209,
+                                        505,
+                                        222
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                304,
+                                                209,
+                                                505,
+                                                222
+                                            ],
+                                            "score": 1.0,
+                                            "content": "dent experiment. Q-values for the cooperate",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 24
+                                },
+                                {
+                                    "bbox": [
+                                        304,
+                                        221,
+                                        505,
+                                        233
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                304,
+                                                221,
+                                                505,
+                                                233
+                                            ],
+                                            "score": 1.0,
+                                            "content": "and defect actions stay tightly coupled in the",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 25
+                                },
+                                {
+                                    "bbox": [
+                                        303,
+                                        231,
+                                        505,
+                                        245
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                303,
+                                                231,
+                                                505,
+                                                245
+                                            ],
+                                            "score": 1.0,
+                                            "content": "failure cases (col. 1,2,5), while in the cases passing",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 26
+                                },
+                                {
+                                    "bbox": [
+                                        303,
+                                        242,
+                                        505,
+                                        255
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                303,
+                                                242,
+                                                505,
+                                                255
+                                            ],
+                                            "score": 1.0,
+                                            "content": "the unit test (col. 3,4) the Q-value of cooperate",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 27
+                                },
+                                {
+                                    "bbox": [
+                                        304,
+                                        254,
+                                        388,
+                                        266
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                304,
+                                                254,
+                                                388,
+                                                266
+                                            ],
+                                            "score": 1.0,
+                                            "content": "decreases over time.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 28
+                                }
+                            ],
+                            "index": 24.5
+                        }
+                    ],
+                    "index": 21.0
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        270,
+                        505,
+                        336
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                270,
+                                505,
+                                283
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        270,
+                                        505,
+                                        283
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We now present a toy environment for modeling content recommendation of news articles, which",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                281,
+                                505,
+                                294
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        281,
+                                        505,
+                                        294
+                                    ],
+                                    "score": 1.0,
+                                    "content": "includes the potential for ADS by incorporating the mechanisms mentioned in Sec. 2.2, discussed as",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                292,
+                                506,
+                                304
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        292,
+                                        506,
+                                        304
+                                    ],
+                                    "score": 1.0,
+                                    "content": "contributing factors to the problems of fake news and filter bubbles. Specifically, the environment",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                303,
+                                506,
+                                316
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        303,
+                                        506,
+                                        316
+                                    ],
+                                    "score": 1.0,
+                                    "content": "assumes that presenting an article to a user can influence (1) their interest in similar articles, and (2)",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                314,
+                                505,
+                                326
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        314,
+                                        505,
+                                        326
+                                    ],
+                                    "score": 1.0,
+                                    "content": "their propensity to use the recommendation service. These correspond to modeling auto-induced",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                325,
+                                504,
+                                338
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        325,
+                                        504,
+                                        338
+                                    ],
+                                    "score": 1.0,
+                                    "content": "concept shift of users, and auto-induced covariate shift of the user base, respectively (see Sec. 2.2).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        }
+                    ],
+                    "index": 31.5,
+                    "bbox_fs": [
+                        105,
+                        270,
+                        506,
+                        338
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        342,
+                        505,
+                        452
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                342,
+                                506,
+                                355
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        342,
+                                        506,
+                                        355
+                                    ],
+                                    "score": 1.0,
+                                    "content": "This environment includes the following components, which change over (discrete) time: User type:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                107,
+                                352,
+                                506,
+                                366
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        354,
+                                        117,
+                                        363
+                                    ],
+                                    "score": 0.85,
+                                    "content": "x ^ { t }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        117,
+                                        352,
+                                        177,
+                                        366
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", Article type:",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        178,
+                                        353,
+                                        188,
+                                        365
+                                    ],
+                                    "score": 0.88,
+                                    "content": "y ^ { t }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        188,
+                                        352,
+                                        257,
+                                        366
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", User interests:",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        257,
+                                        353,
+                                        274,
+                                        363
+                                    ],
+                                    "score": 0.77,
+                                    "content": "\\bar { \\mathbf { W } } ^ { t }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        274,
+                                        352,
+                                        506,
+                                        366
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(propensity for users of each type to click on articles of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                363,
+                                505,
+                                376
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        363,
+                                        229,
+                                        376
+                                    ],
+                                    "score": 1.0,
+                                    "content": "each type), and User loyalty:",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        230,
+                                        364,
+                                        241,
+                                        375
+                                    ],
+                                    "score": 0.87,
+                                    "content": "\\mathbf { g } ^ { t }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        241,
+                                        363,
+                                        505,
+                                        376
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(propensity for users of each type to use the platform). At each",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                375,
+                                505,
+                                388
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        375,
+                                        145,
+                                        388
+                                    ],
+                                    "score": 1.0,
+                                    "content": "time step",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        145,
+                                        376,
+                                        150,
+                                        385
+                                    ],
+                                    "score": 0.71,
+                                    "content": "t",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        150,
+                                        375,
+                                        179,
+                                        388
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", a user",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        180,
+                                        375,
+                                        190,
+                                        385
+                                    ],
+                                    "score": 0.87,
+                                    "content": "x ^ { t }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        190,
+                                        375,
+                                        505,
+                                        388
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is sampled from a categorical distribution, based on the loyalty of the different",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                386,
+                                505,
+                                399
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        386,
+                                        505,
+                                        399
+                                    ],
+                                    "score": 1.0,
+                                    "content": "user types. The recommendation system (a classifier) selects which type of article to present in the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                397,
+                                506,
+                                411
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        397,
+                                        311,
+                                        411
+                                    ],
+                                    "score": 1.0,
+                                    "content": "top position, and finally the user ‘clicks’ an article",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        312,
+                                        397,
+                                        322,
+                                        408
+                                    ],
+                                    "score": 0.88,
+                                    "content": "y ^ { t }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        322,
+                                        397,
+                                        506,
+                                        411
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", according to their interests. User loyalty for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                408,
+                                506,
+                                421
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        408,
+                                        146,
+                                        421
+                                    ],
+                                    "score": 1.0,
+                                    "content": "user type",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        146,
+                                        408,
+                                        156,
+                                        418
+                                    ],
+                                    "score": 0.86,
+                                    "content": "x ^ { t }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        156,
+                                        408,
+                                        443,
+                                        421
+                                    ],
+                                    "score": 1.0,
+                                    "content": "undergoes covariate shift: in accordance with the self-selection effect,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        443,
+                                        408,
+                                        453,
+                                        420
+                                    ],
+                                    "score": 0.87,
+                                    "content": "g ^ { t }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        453,
+                                        408,
+                                        506,
+                                        421
+                                    ],
+                                    "score": 1.0,
+                                    "content": "increases or",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                418,
+                                504,
+                                432
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        418,
+                                        493,
+                                        432
+                                    ],
+                                    "score": 1.0,
+                                    "content": "decreases proportionally to that user type’s interest in the top article. The interests of user type",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        493,
+                                        419,
+                                        504,
+                                        429
+                                    ],
+                                    "score": 0.84,
+                                    "content": "x ^ { t }",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                429,
+                                506,
+                                443
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        429,
+                                        225,
+                                        443
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(represented by a column of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        225,
+                                        430,
+                                        241,
+                                        440
+                                    ],
+                                    "score": 0.83,
+                                    "content": "\\mathbf { W } ^ { t }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        241,
+                                        429,
+                                        506,
+                                        443
+                                    ],
+                                    "score": 1.0,
+                                    "content": ") undergoing concept shift; in accordance with the illusory truth",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                440,
+                                497,
+                                454
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        440,
+                                        497,
+                                        454
+                                    ],
+                                    "score": 1.0,
+                                    "content": "effect, interest in the topic of the top article chosen by the recommender system always increases.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        }
+                    ],
+                    "index": 39.5,
+                    "bbox_fs": [
+                        105,
+                        342,
+                        506,
+                        454
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        465,
+                        451,
+                        475
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                464,
+                                452,
+                                477
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        464,
+                                        452,
+                                        477
+                                    ],
+                                    "score": 1.0,
+                                    "content": "5.2.1 CONTENT RECOMMENDATION EXPERIMENTAL RESULTS AND DISCUSSION",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        }
+                    ],
+                    "index": 45
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        483,
+                        506,
+                        583
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                483,
+                                506,
+                                496
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        483,
+                                        506,
+                                        496
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We run 20 trials using an MLP trained with SGD for the recommender system. We find that PBT",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                494,
+                                505,
+                                507
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        494,
+                                        505,
+                                        507
+                                    ],
+                                    "score": 1.0,
+                                    "content": "yields significant improvements in training time and accuracy, but also greater distributional shift",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 47
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                505,
+                                506,
+                                519
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        505,
+                                        506,
+                                        519
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(Fig. 7). User base and user interests both change faster with PBT, and user interests change more",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 48
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                516,
+                                506,
+                                530
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        516,
+                                        506,
+                                        530
+                                    ],
+                                    "score": 1.0,
+                                    "content": "overall. We measure concept/covariate shift using the cosine distance and KL-divergence, respectively.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 49
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                526,
+                                506,
+                                541
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        526,
+                                        506,
+                                        541
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We observe that the distributions over user types typically saturate (to a single user type) after a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 50
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                538,
+                                506,
+                                551
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        538,
+                                        506,
+                                        551
+                                    ],
+                                    "score": 1.0,
+                                    "content": "few hundred time-steps (Fig 1 and Fig. 7, Right). We run long enough to reach such states, to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 51
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                549,
+                                506,
+                                562
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        549,
+                                        506,
+                                        562
+                                    ],
+                                    "score": 1.0,
+                                    "content": "demonstrate that the increase in ADS from PBT is not transitory. The environment has a number of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 52
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                561,
+                                505,
+                                573
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        561,
+                                        476,
+                                        573
+                                    ],
+                                    "score": 1.0,
+                                    "content": "free parameters, and our results are qualitatively consistent so long as the covariate shift rate",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        477,
+                                        561,
+                                        495,
+                                        572
+                                    ],
+                                    "score": 0.82,
+                                    "content": "( \\alpha _ { 1 } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        495,
+                                        561,
+                                        505,
+                                        573
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 53
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                571,
+                                385,
+                                584
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        571,
+                                        237,
+                                        584
+                                    ],
+                                    "score": 1.0,
+                                    "content": "faster than the concept shift rate",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        237,
+                                        572,
+                                        254,
+                                        583
+                                    ],
+                                    "score": 0.78,
+                                    "content": "\\left( \\alpha _ { 2 } \\right)",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        255,
+                                        571,
+                                        385,
+                                        584
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". See Appendix 9.2.1 for details.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 54
+                        }
+                    ],
+                    "index": 50,
+                    "bbox_fs": [
+                        104,
+                        483,
+                        506,
+                        584
+                    ]
+                },
+                {
+                    "type": "image",
+                    "bbox": [
+                        189,
+                        592,
+                        421,
+                        681
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                189,
+                                592,
+                                421,
+                                681
+                            ],
+                            "group_id": 1,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        189,
+                                        592,
+                                        421,
+                                        681
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                189,
+                                                592,
+                                                421,
+                                                681
+                                            ],
+                                            "score": 0.965,
+                                            "type": "image",
+                                            "image_path": "95efffd0c2948471a88671bd2e264c0a9ccf142775dc7e578a618fd609eb6f42.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 57.5,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        189,
+                                        592,
+                                        421,
+                                        606.8333333333334
+                                    ],
+                                    "spans": [],
+                                    "index": 55
+                                },
+                                {
+                                    "bbox": [
+                                        189,
+                                        606.8333333333334,
+                                        421,
+                                        621.6666666666667
+                                    ],
+                                    "spans": [],
+                                    "index": 56
+                                },
+                                {
+                                    "bbox": [
+                                        189,
+                                        621.6666666666667,
+                                        421,
+                                        636.5000000000001
+                                    ],
+                                    "spans": [],
+                                    "index": 57
+                                },
+                                {
+                                    "bbox": [
+                                        189,
+                                        636.5000000000001,
+                                        421,
+                                        651.3333333333335
+                                    ],
+                                    "spans": [],
+                                    "index": 58
+                                },
+                                {
+                                    "bbox": [
+                                        189,
+                                        651.3333333333335,
+                                        421,
+                                        666.1666666666669
+                                    ],
+                                    "spans": [],
+                                    "index": 59
+                                },
+                                {
+                                    "bbox": [
+                                        189,
+                                        666.1666666666669,
+                                        421,
+                                        681.0000000000002
+                                    ],
+                                    "spans": [],
+                                    "index": 60
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                107,
+                                691,
+                                505,
+                                725
+                            ],
+                            "group_id": 1,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        691,
+                                        506,
+                                        703
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                691,
+                                                506,
+                                                703
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 7: Content recommendation experiments. Left: using Population Based Training (PBT)",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 61
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        702,
+                                        506,
+                                        715
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                702,
+                                                471,
+                                                715
+                                            ],
+                                            "score": 1.0,
+                                            "content": "increases accuracy of predictions faster, leads to a faster and larger drift in users’ interests,",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                472,
+                                                702,
+                                                502,
+                                                714
+                                            ],
+                                            "score": 0.91,
+                                            "content": "P ( \\boldsymbol { y } | \\mathbf { x } )",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                502,
+                                                702,
+                                                506,
+                                                715
+                                            ],
+                                            "score": 1.0,
+                                            "content": ",",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 62
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        713,
+                                        504,
+                                        725
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                713,
+                                                288,
+                                                725
+                                            ],
+                                            "score": 1.0,
+                                            "content": "(Center); as well as the distribution of users,",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                288,
+                                                713,
+                                                310,
+                                                725
+                                            ],
+                                            "score": 0.91,
+                                            "content": "P ( \\mathbf { x } )",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                311,
+                                                713,
+                                                504,
+                                                725
+                                            ],
+                                            "score": 1.0,
+                                            "content": ", (Right). Shading shows std error over 20 runs.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 63
+                                }
+                            ],
+                            "index": 62
+                        }
+                    ],
+                    "index": 59.75
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        81,
+                        209,
+                        93
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                79,
+                                210,
+                                96
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        79,
+                                        210,
+                                        96
+                                    ],
+                                    "score": 1.0,
+                                    "content": "6 RELATED WORK",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        }
+                    ],
+                    "index": 0
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        99,
+                        505,
+                        187
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                100,
+                                505,
+                                111
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        100,
+                                        505,
+                                        111
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ADS in practice: We introduce the term ADS, but we are far from the first to study it. Caruana",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                111,
+                                506,
+                                123
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        111,
+                                        506,
+                                        123
+                                    ],
+                                    "score": 1.0,
+                                    "content": "et al. (2015) provide an example of asthmatic patients having lower predicted risk of pneumonia.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                122,
+                                505,
+                                134
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        122,
+                                        505,
+                                        134
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Treating asthmatics with pneumonia less aggressively on this basis would be an example of harmful",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                132,
+                                505,
+                                145
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        132,
+                                        505,
+                                        145
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ADS; the reason they had lower pneumonia risk was because they had received more aggressive care",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                143,
+                                505,
+                                156
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        143,
+                                        505,
+                                        156
+                                    ],
+                                    "score": 1.0,
+                                    "content": "already. Schulam & Saria (2017) note that such predictive models are commonly used to inform",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                154,
+                                506,
+                                167
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        154,
+                                        506,
+                                        167
+                                    ],
+                                    "score": 1.0,
+                                    "content": "decision-making, and propose modeling counterfactuals (e.g. “how would this patient fare with less",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                166,
+                                506,
+                                178
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        166,
+                                        506,
+                                        178
+                                    ],
+                                    "score": 1.0,
+                                    "content": "aggressive treatment”) to avoid making such self-refuting predictions. While their goal is to make",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                177,
+                                506,
+                                189
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        177,
+                                        506,
+                                        189
+                                    ],
+                                    "score": 1.0,
+                                    "content": "accurate predictions in the presence of ADS, our goal is to identify and manage incentives for ADS.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        }
+                    ],
+                    "index": 4.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        193,
+                        505,
+                        248
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                192,
+                                505,
+                                206
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        192,
+                                        505,
+                                        206
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Non-i.i.d bandits: Contextual bandits (Wang et al., 2005) are a common approach to content",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                203,
+                                506,
+                                217
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        203,
+                                        506,
+                                        217
+                                    ],
+                                    "score": 1.0,
+                                    "content": "recommendation (Li et al., 2010). While bandit algorithms typically make the i.i.d. assumption,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                215,
+                                506,
+                                227
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        215,
+                                        506,
+                                        227
+                                    ],
+                                    "score": 1.0,
+                                    "content": "counter-examples exist (Gheshlaghi Azar et al., 2014; Auer et al., 1995). Closest to our work is Shah",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                226,
+                                506,
+                                239
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        226,
+                                        506,
+                                        239
+                                    ],
+                                    "score": 1.0,
+                                    "content": "et al. (2018), who consider covariate shift caused by recommender systems’ recommendations. But",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                236,
+                                466,
+                                249
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        236,
+                                        466,
+                                        249
+                                    ],
+                                    "score": 1.0,
+                                    "content": "while they seek to exploit ADS, our aim is to avoid hidden incentives for exploiting ADS.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        }
+                    ],
+                    "index": 11
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        253,
+                        505,
+                        364
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                254,
+                                505,
+                                266
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        254,
+                                        505,
+                                        266
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Safety and incentives: Emergent incentives to influence the world (such as HI-ADS) are at the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                266,
+                                506,
+                                277
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        266,
+                                        506,
+                                        277
+                                    ],
+                                    "score": 1.0,
+                                    "content": "heart of many concerns about the safety of advanced AI systems (Omohundro, 2008; Bostrom, 2014).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                275,
+                                506,
+                                289
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        275,
+                                        506,
+                                        289
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Understanding and managing the incentives of learners is also a focus of Armstrong & O’Rourke",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                286,
+                                506,
+                                299
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        286,
+                                        506,
+                                        299
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(2017); Everitt (2018); Everitt et al. (2019); Cohen et al.. While Everitt et al. (2019) focus on",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                297,
+                                506,
+                                311
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        297,
+                                        506,
+                                        311
+                                    ],
+                                    "score": 1.0,
+                                    "content": "identifying which incentives are present, we note that incentives may be present and yet not be",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                309,
+                                506,
+                                321
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        309,
+                                        506,
+                                        321
+                                    ],
+                                    "score": 1.0,
+                                    "content": "revealed or pursued - for example, in supervised learning, there is an incentive to make predictions",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                318,
+                                506,
+                                333
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        318,
+                                        506,
+                                        333
+                                    ],
+                                    "score": 1.0,
+                                    "content": "that are over-fit to the test set, but we typically hide the test set from the learner, which effectively",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                330,
+                                506,
+                                343
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        330,
+                                        506,
+                                        343
+                                    ],
+                                    "score": 1.0,
+                                    "content": "hides this incentive. While Carey et al. (2020); Everitt et al. (2019); Armstrong & O’Rourke (2017)",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                342,
+                                506,
+                                354
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        342,
+                                        506,
+                                        354
+                                    ],
+                                    "score": 1.0,
+                                    "content": "discuss methods of removing problematic incentives, we note in practice incentives are often hidden",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                353,
+                                506,
+                                365
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        353,
+                                        506,
+                                        365
+                                    ],
+                                    "score": 1.0,
+                                    "content": "rather than removed. Our work addresses the efficacy of this approach and ways in which it can fail.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        }
+                    ],
+                    "index": 18.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        369,
+                        505,
+                        447
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                370,
+                                506,
+                                382
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        370,
+                                        506,
+                                        382
+                                    ],
+                                    "score": 1.0,
+                                    "content": "HI-ADS and meta-learning: We believe our work is the first to consider HI-ADS and their",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                380,
+                                506,
+                                393
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        380,
+                                        506,
+                                        393
+                                    ],
+                                    "score": 1.0,
+                                    "content": "relation to meta-learning. A few previous works have some relevance. Rabinowitz (2019) documents",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                392,
+                                505,
+                                404
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        392,
+                                        505,
+                                        404
+                                    ],
+                                    "score": 1.0,
+                                    "content": "qualitative differences in learning behavior when meta-learning is applied. MacKay et al. (2019) and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                402,
+                                506,
+                                415
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        402,
+                                        506,
+                                        415
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Lorraine & Duvenaud (2018) view meta-learning as a bilevel optimization problem, with the inner",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                413,
+                                506,
+                                426
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        413,
+                                        506,
+                                        426
+                                    ],
+                                    "score": 1.0,
+                                    "content": "loop playing a best-response to the outer loop. In our work, the outer loop has a greater influence,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                423,
+                                505,
+                                437
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        423,
+                                        505,
+                                        437
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and the inner loop often fails to play best-response. Sutton et al. (2007) noted that meta-learning can",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                436,
+                                441,
+                                448
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        436,
+                                        441,
+                                        448
+                                    ],
+                                    "score": 1.0,
+                                    "content": "improve performance by preventing convergence of the inner loop to best response.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        }
+                    ],
+                    "index": 27
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        109,
+                        460,
+                        287,
+                        473
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                459,
+                                289,
+                                475
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        459,
+                                        289,
+                                        475
+                                    ],
+                                    "score": 1.0,
+                                    "content": "7 DISCUSSION AND CONCLUSION",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        }
+                    ],
+                    "index": 31
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        478,
+                        504,
+                        512
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                477,
+                                506,
+                                491
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        477,
+                                        506,
+                                        491
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We identify the phenomenon of auto-induced distributional shift (ADS) and problems that can arise",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                489,
+                                505,
+                                501
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        489,
+                                        505,
+                                        501
+                                    ],
+                                    "score": 1.0,
+                                    "content": "when there are hidden incentives for learners to induce distributional shift (HI-ADS). We show that",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                500,
+                                506,
+                                514
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        500,
+                                        506,
+                                        514
+                                    ],
+                                    "score": 1.0,
+                                    "content": "meta-learning can reveal HI-ADS and lead learners to use ADS as a means of increasing performance.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        }
+                    ],
+                    "index": 33
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        517,
+                        505,
+                        627
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                517,
+                                506,
+                                530
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        517,
+                                        506,
+                                        530
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Our work highlights the interdisciplinary nature of issues with real-world deployment of ML systems",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                529,
+                                505,
+                                540
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        529,
+                                        505,
+                                        540
+                                    ],
+                                    "score": 1.0,
+                                    "content": "- we show how HI-ADS could play a role in important technosocial issues like filter bubbles and the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                539,
+                                505,
+                                552
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        539,
+                                        505,
+                                        552
+                                    ],
+                                    "score": 1.0,
+                                    "content": "propagation of fake news. There are a number of potential implications for our work: (1) When",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                549,
+                                506,
+                                563
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        549,
+                                        506,
+                                        563
+                                    ],
+                                    "score": 1.0,
+                                    "content": "HI-ADS are a concern, our methodology and environments can be used to help diagnose whether",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                561,
+                                506,
+                                574
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        561,
+                                        506,
+                                        574
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and to what extent the final performance/behavior of a learner is due to ADS and/or incentives for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                572,
+                                506,
+                                585
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        572,
+                                        506,
+                                        585
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ADS, i.e. to quantify their influence on that learner. (2) Comparing this quantitative analysis for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                582,
+                                506,
+                                596
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        582,
+                                        506,
+                                        596
+                                    ],
+                                    "score": 1.0,
+                                    "content": "different algorithms could help us understand which features of algorithms affect their propensity to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                593,
+                                506,
+                                608
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        593,
+                                        506,
+                                        608
+                                    ],
+                                    "score": 1.0,
+                                    "content": "reveal HI-ADS, and aid in the development of safer and more robust algorithms. (3) Characterizing",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                605,
+                                506,
+                                618
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        605,
+                                        506,
+                                        618
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and identifying HI-ADS in these tests is a first step to analyzing and mitigating other (problematic)",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                615,
+                                468,
+                                629
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        615,
+                                        468,
+                                        629
+                                    ],
+                                    "score": 1.0,
+                                    "content": "hidden incentives, as well as to developing theoretical understanding of hidden incentives.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        }
+                    ],
+                    "index": 39.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        632,
+                        505,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                633,
+                                505,
+                                645
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        633,
+                                        505,
+                                        645
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Broadly speaking, our work emphasizes that the choice of machine learning algorithm plays an",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                644,
+                                506,
+                                656
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        644,
+                                        506,
+                                        656
+                                    ],
+                                    "score": 1.0,
+                                    "content": "important role in specification, independently of the choice of performance metric. A learner can",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                654,
+                                506,
+                                669
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        654,
+                                        506,
+                                        669
+                                    ],
+                                    "score": 1.0,
+                                    "content": "use ADS to increase performance according to the intended performance metric, and yet still behave",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 47
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                666,
+                                505,
+                                679
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        666,
+                                        505,
+                                        679
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in an undesirable way, if we did not intend the learner to improve performance by that method. In",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 48
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                677,
+                                506,
+                                690
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        677,
+                                        506,
+                                        690
+                                    ],
+                                    "score": 1.0,
+                                    "content": "other words, performance metrics are typically incomplete specifications: they only specify our goals",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 49
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                688,
+                                505,
+                                701
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        688,
+                                        505,
+                                        701
+                                    ],
+                                    "score": 1.0,
+                                    "content": "or ends, while our choice of learning algorithm plays a role in specifying the means by which we",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 50
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                698,
+                                506,
+                                712
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        698,
+                                        506,
+                                        712
+                                    ],
+                                    "score": 1.0,
+                                    "content": "intend an learner to achieve those ends. With increasing deployment of ML algorithms in daily life,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 51
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                707,
+                                506,
+                                725
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        707,
+                                        506,
+                                        725
+                                    ],
+                                    "score": 1.0,
+                                    "content": "we believe that (1) understanding incentives and (2) specifying desired/allowed means of improving",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 52
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                721,
+                                475,
+                                733
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        721,
+                                        475,
+                                        733
+                                    ],
+                                    "score": 1.0,
+                                    "content": "performance are important avenues of future work to ensure fair, robust, and safe outcomes.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 53
+                        }
+                    ],
+                    "index": 49
+                }
+            ],
+            "page_idx": 7,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        302,
+                        752,
+                        308,
+                        759
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                302,
+                                750,
+                                309,
+                                761
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        302,
+                                        750,
+                                        309,
+                                        761
+                                    ],
+                                    "score": 1.0,
+                                    "content": "8",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        107,
+                        27,
+                        307,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                26,
+                                308,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        26,
+                                        308,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Under review as a conference paper at ICLR 2021",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        81,
+                        209,
+                        93
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                79,
+                                210,
+                                96
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        79,
+                                        210,
+                                        96
+                                    ],
+                                    "score": 1.0,
+                                    "content": "6 RELATED WORK",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        }
+                    ],
+                    "index": 0
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        99,
+                        505,
+                        187
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                100,
+                                505,
+                                111
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        100,
+                                        505,
+                                        111
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ADS in practice: We introduce the term ADS, but we are far from the first to study it. Caruana",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                111,
+                                506,
+                                123
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        111,
+                                        506,
+                                        123
+                                    ],
+                                    "score": 1.0,
+                                    "content": "et al. (2015) provide an example of asthmatic patients having lower predicted risk of pneumonia.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                122,
+                                505,
+                                134
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        122,
+                                        505,
+                                        134
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Treating asthmatics with pneumonia less aggressively on this basis would be an example of harmful",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                132,
+                                505,
+                                145
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        132,
+                                        505,
+                                        145
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ADS; the reason they had lower pneumonia risk was because they had received more aggressive care",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                143,
+                                505,
+                                156
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        143,
+                                        505,
+                                        156
+                                    ],
+                                    "score": 1.0,
+                                    "content": "already. Schulam & Saria (2017) note that such predictive models are commonly used to inform",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                154,
+                                506,
+                                167
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        154,
+                                        506,
+                                        167
+                                    ],
+                                    "score": 1.0,
+                                    "content": "decision-making, and propose modeling counterfactuals (e.g. “how would this patient fare with less",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                166,
+                                506,
+                                178
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        166,
+                                        506,
+                                        178
+                                    ],
+                                    "score": 1.0,
+                                    "content": "aggressive treatment”) to avoid making such self-refuting predictions. While their goal is to make",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                177,
+                                506,
+                                189
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        177,
+                                        506,
+                                        189
+                                    ],
+                                    "score": 1.0,
+                                    "content": "accurate predictions in the presence of ADS, our goal is to identify and manage incentives for ADS.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        }
+                    ],
+                    "index": 4.5,
+                    "bbox_fs": [
+                        105,
+                        100,
+                        506,
+                        189
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        193,
+                        505,
+                        248
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                192,
+                                505,
+                                206
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        192,
+                                        505,
+                                        206
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Non-i.i.d bandits: Contextual bandits (Wang et al., 2005) are a common approach to content",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                203,
+                                506,
+                                217
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        203,
+                                        506,
+                                        217
+                                    ],
+                                    "score": 1.0,
+                                    "content": "recommendation (Li et al., 2010). While bandit algorithms typically make the i.i.d. assumption,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                215,
+                                506,
+                                227
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        215,
+                                        506,
+                                        227
+                                    ],
+                                    "score": 1.0,
+                                    "content": "counter-examples exist (Gheshlaghi Azar et al., 2014; Auer et al., 1995). Closest to our work is Shah",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                226,
+                                506,
+                                239
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        226,
+                                        506,
+                                        239
+                                    ],
+                                    "score": 1.0,
+                                    "content": "et al. (2018), who consider covariate shift caused by recommender systems’ recommendations. But",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                236,
+                                466,
+                                249
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        236,
+                                        466,
+                                        249
+                                    ],
+                                    "score": 1.0,
+                                    "content": "while they seek to exploit ADS, our aim is to avoid hidden incentives for exploiting ADS.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        }
+                    ],
+                    "index": 11,
+                    "bbox_fs": [
+                        105,
+                        192,
+                        506,
+                        249
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        253,
+                        505,
+                        364
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                254,
+                                505,
+                                266
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        254,
+                                        505,
+                                        266
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Safety and incentives: Emergent incentives to influence the world (such as HI-ADS) are at the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                266,
+                                506,
+                                277
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        266,
+                                        506,
+                                        277
+                                    ],
+                                    "score": 1.0,
+                                    "content": "heart of many concerns about the safety of advanced AI systems (Omohundro, 2008; Bostrom, 2014).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                275,
+                                506,
+                                289
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        275,
+                                        506,
+                                        289
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Understanding and managing the incentives of learners is also a focus of Armstrong & O’Rourke",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                286,
+                                506,
+                                299
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        286,
+                                        506,
+                                        299
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(2017); Everitt (2018); Everitt et al. (2019); Cohen et al.. While Everitt et al. (2019) focus on",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                297,
+                                506,
+                                311
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        297,
+                                        506,
+                                        311
+                                    ],
+                                    "score": 1.0,
+                                    "content": "identifying which incentives are present, we note that incentives may be present and yet not be",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                309,
+                                506,
+                                321
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        309,
+                                        506,
+                                        321
+                                    ],
+                                    "score": 1.0,
+                                    "content": "revealed or pursued - for example, in supervised learning, there is an incentive to make predictions",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                318,
+                                506,
+                                333
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        318,
+                                        506,
+                                        333
+                                    ],
+                                    "score": 1.0,
+                                    "content": "that are over-fit to the test set, but we typically hide the test set from the learner, which effectively",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                330,
+                                506,
+                                343
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        330,
+                                        506,
+                                        343
+                                    ],
+                                    "score": 1.0,
+                                    "content": "hides this incentive. While Carey et al. (2020); Everitt et al. (2019); Armstrong & O’Rourke (2017)",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                342,
+                                506,
+                                354
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        342,
+                                        506,
+                                        354
+                                    ],
+                                    "score": 1.0,
+                                    "content": "discuss methods of removing problematic incentives, we note in practice incentives are often hidden",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                353,
+                                506,
+                                365
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        353,
+                                        506,
+                                        365
+                                    ],
+                                    "score": 1.0,
+                                    "content": "rather than removed. Our work addresses the efficacy of this approach and ways in which it can fail.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        }
+                    ],
+                    "index": 18.5,
+                    "bbox_fs": [
+                        105,
+                        254,
+                        506,
+                        365
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        369,
+                        505,
+                        447
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                370,
+                                506,
+                                382
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        370,
+                                        506,
+                                        382
+                                    ],
+                                    "score": 1.0,
+                                    "content": "HI-ADS and meta-learning: We believe our work is the first to consider HI-ADS and their",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                380,
+                                506,
+                                393
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        380,
+                                        506,
+                                        393
+                                    ],
+                                    "score": 1.0,
+                                    "content": "relation to meta-learning. A few previous works have some relevance. Rabinowitz (2019) documents",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                392,
+                                505,
+                                404
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        392,
+                                        505,
+                                        404
+                                    ],
+                                    "score": 1.0,
+                                    "content": "qualitative differences in learning behavior when meta-learning is applied. MacKay et al. (2019) and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                402,
+                                506,
+                                415
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        402,
+                                        506,
+                                        415
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Lorraine & Duvenaud (2018) view meta-learning as a bilevel optimization problem, with the inner",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                413,
+                                506,
+                                426
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        413,
+                                        506,
+                                        426
+                                    ],
+                                    "score": 1.0,
+                                    "content": "loop playing a best-response to the outer loop. In our work, the outer loop has a greater influence,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                423,
+                                505,
+                                437
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        423,
+                                        505,
+                                        437
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and the inner loop often fails to play best-response. Sutton et al. (2007) noted that meta-learning can",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                436,
+                                441,
+                                448
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        436,
+                                        441,
+                                        448
+                                    ],
+                                    "score": 1.0,
+                                    "content": "improve performance by preventing convergence of the inner loop to best response.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        }
+                    ],
+                    "index": 27,
+                    "bbox_fs": [
+                        105,
+                        370,
+                        506,
+                        448
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        109,
+                        460,
+                        287,
+                        473
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                459,
+                                289,
+                                475
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        459,
+                                        289,
+                                        475
+                                    ],
+                                    "score": 1.0,
+                                    "content": "7 DISCUSSION AND CONCLUSION",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        }
+                    ],
+                    "index": 31
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        478,
+                        504,
+                        512
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                477,
+                                506,
+                                491
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        477,
+                                        506,
+                                        491
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We identify the phenomenon of auto-induced distributional shift (ADS) and problems that can arise",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                489,
+                                505,
+                                501
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        489,
+                                        505,
+                                        501
+                                    ],
+                                    "score": 1.0,
+                                    "content": "when there are hidden incentives for learners to induce distributional shift (HI-ADS). We show that",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                500,
+                                506,
+                                514
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        500,
+                                        506,
+                                        514
+                                    ],
+                                    "score": 1.0,
+                                    "content": "meta-learning can reveal HI-ADS and lead learners to use ADS as a means of increasing performance.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        }
+                    ],
+                    "index": 33,
+                    "bbox_fs": [
+                        105,
+                        477,
+                        506,
+                        514
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        517,
+                        505,
+                        627
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                517,
+                                506,
+                                530
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        517,
+                                        506,
+                                        530
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Our work highlights the interdisciplinary nature of issues with real-world deployment of ML systems",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                529,
+                                505,
+                                540
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        529,
+                                        505,
+                                        540
+                                    ],
+                                    "score": 1.0,
+                                    "content": "- we show how HI-ADS could play a role in important technosocial issues like filter bubbles and the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                539,
+                                505,
+                                552
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        539,
+                                        505,
+                                        552
+                                    ],
+                                    "score": 1.0,
+                                    "content": "propagation of fake news. There are a number of potential implications for our work: (1) When",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                549,
+                                506,
+                                563
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        549,
+                                        506,
+                                        563
+                                    ],
+                                    "score": 1.0,
+                                    "content": "HI-ADS are a concern, our methodology and environments can be used to help diagnose whether",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                561,
+                                506,
+                                574
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        561,
+                                        506,
+                                        574
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and to what extent the final performance/behavior of a learner is due to ADS and/or incentives for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                572,
+                                506,
+                                585
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        572,
+                                        506,
+                                        585
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ADS, i.e. to quantify their influence on that learner. (2) Comparing this quantitative analysis for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                582,
+                                506,
+                                596
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        582,
+                                        506,
+                                        596
+                                    ],
+                                    "score": 1.0,
+                                    "content": "different algorithms could help us understand which features of algorithms affect their propensity to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                593,
+                                506,
+                                608
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        593,
+                                        506,
+                                        608
+                                    ],
+                                    "score": 1.0,
+                                    "content": "reveal HI-ADS, and aid in the development of safer and more robust algorithms. (3) Characterizing",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                605,
+                                506,
+                                618
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        605,
+                                        506,
+                                        618
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and identifying HI-ADS in these tests is a first step to analyzing and mitigating other (problematic)",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                615,
+                                468,
+                                629
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        615,
+                                        468,
+                                        629
+                                    ],
+                                    "score": 1.0,
+                                    "content": "hidden incentives, as well as to developing theoretical understanding of hidden incentives.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        }
+                    ],
+                    "index": 39.5,
+                    "bbox_fs": [
+                        104,
+                        517,
+                        506,
+                        629
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        632,
+                        505,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                633,
+                                505,
+                                645
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        633,
+                                        505,
+                                        645
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Broadly speaking, our work emphasizes that the choice of machine learning algorithm plays an",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                644,
+                                506,
+                                656
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        644,
+                                        506,
+                                        656
+                                    ],
+                                    "score": 1.0,
+                                    "content": "important role in specification, independently of the choice of performance metric. A learner can",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                654,
+                                506,
+                                669
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        654,
+                                        506,
+                                        669
+                                    ],
+                                    "score": 1.0,
+                                    "content": "use ADS to increase performance according to the intended performance metric, and yet still behave",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 47
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                666,
+                                505,
+                                679
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        666,
+                                        505,
+                                        679
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in an undesirable way, if we did not intend the learner to improve performance by that method. In",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 48
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                677,
+                                506,
+                                690
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        677,
+                                        506,
+                                        690
+                                    ],
+                                    "score": 1.0,
+                                    "content": "other words, performance metrics are typically incomplete specifications: they only specify our goals",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 49
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                688,
+                                505,
+                                701
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        688,
+                                        505,
+                                        701
+                                    ],
+                                    "score": 1.0,
+                                    "content": "or ends, while our choice of learning algorithm plays a role in specifying the means by which we",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 50
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                698,
+                                506,
+                                712
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        698,
+                                        506,
+                                        712
+                                    ],
+                                    "score": 1.0,
+                                    "content": "intend an learner to achieve those ends. With increasing deployment of ML algorithms in daily life,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 51
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                707,
+                                506,
+                                725
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        707,
+                                        506,
+                                        725
+                                    ],
+                                    "score": 1.0,
+                                    "content": "we believe that (1) understanding incentives and (2) specifying desired/allowed means of improving",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 52
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                721,
+                                475,
+                                733
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        721,
+                                        475,
+                                        733
+                                    ],
+                                    "score": 1.0,
+                                    "content": "performance are important avenues of future work to ensure fair, robust, and safe outcomes.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 53
+                        }
+                    ],
+                    "index": 49,
+                    "bbox_fs": [
+                        104,
+                        633,
+                        506,
+                        733
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        81,
+                        176,
+                        94
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                82,
+                                176,
+                                94
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        82,
+                                        176,
+                                        94
+                                    ],
+                                    "score": 1.0,
+                                    "content": "REFERENCES",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        }
+                    ],
+                    "index": 0
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        100,
+                        505,
+                        123
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                99,
+                                506,
+                                113
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        99,
+                                        506,
+                                        113
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Hunt Allcott and Matthew Gentzkow. Social media and fake news in the 2016 election. Journal of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                111,
+                                318,
+                                124
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        111,
+                                        318,
+                                        124
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Economic Perspectives, 31(2):211–36, May 2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        }
+                    ],
+                    "index": 1.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        129,
+                        504,
+                        164
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                128,
+                                505,
+                                144
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        128,
+                                        505,
+                                        144
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Michelle A. Amazeen and Bartosz W. Wojdynski. Reducing native advertising deception: Revisiting",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                142,
+                                505,
+                                154
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        142,
+                                        505,
+                                        154
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the antecedents and consequences of persuasion knowledge in digital news contexts. Mass",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                152,
+                                303,
+                                164
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        152,
+                                        303,
+                                        164
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Communication and Society, 0(0):1–26, 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        }
+                    ],
+                    "index": 4
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        171,
+                        505,
+                        194
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                169,
+                                506,
+                                185
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        169,
+                                        506,
+                                        185
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Stuart Armstrong and Xavier O’Rorke. Good and safe uses of ai oracles. arXiv preprint",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                182,
+                                220,
+                                194
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        182,
+                                        220,
+                                        194
+                                    ],
+                                    "score": 1.0,
+                                    "content": "arXiv:1711.05541, 2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        }
+                    ],
+                    "index": 6.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        201,
+                        505,
+                        224
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                201,
+                                505,
+                                214
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        201,
+                                        505,
+                                        214
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Stuart Armstrong and Xavier O’Rourke. Indifference methods for managing agent rewards. Technical",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                213,
+                                290,
+                                224
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        213,
+                                        290,
+                                        224
+                                    ],
+                                    "score": 1.0,
+                                    "content": "report, Future of Humanity Institute, 2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        }
+                    ],
+                    "index": 8.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        232,
+                        506,
+                        255
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                231,
+                                506,
+                                246
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        231,
+                                        506,
+                                        246
+                                    ],
+                                    "score": 1.0,
+                                    "content": "K. J. Åström. Optimal control of Markov Processes with incomplete state information. Journal of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                244,
+                                394,
+                                255
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        244,
+                                        394,
+                                        255
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Mathematical Analysis and Applications, 10:174–205, January 1965.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        }
+                    ],
+                    "index": 10.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        262,
+                        504,
+                        286
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                262,
+                                505,
+                                276
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        262,
+                                        505,
+                                        276
+                                    ],
+                                    "score": 1.0,
+                                    "content": "P. Auer, N. Cesa-Bianchi, Y. Freund, and R. E. Schapire. Gambling in a rigged casino: The adversarial",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                273,
+                                410,
+                                286
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        273,
+                                        410,
+                                        286
+                                    ],
+                                    "score": 1.0,
+                                    "content": "multi-armed bandit problem. In Foundations of Computer Science, 1995.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        }
+                    ],
+                    "index": 12.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        292,
+                        506,
+                        338
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                293,
+                                506,
+                                305
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        293,
+                                        506,
+                                        305
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Eytan Bakshy, Solomon Messing, and Lada A. Adamic. Exposure to ideologically diverse news",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                304,
+                                507,
+                                317
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        304,
+                                        507,
+                                        317
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and opinion on Facebook. Science, 348(6239):1130–1132, 2015. ISSN 0036-8075. doi:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                314,
+                                506,
+                                328
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        314,
+                                        506,
+                                        328
+                                    ],
+                                    "score": 1.0,
+                                    "content": "10.1126/science.aaa1160. URL http://science.sciencemag.org/content/348/",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                117,
+                                327,
+                                176,
+                                337
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        117,
+                                        327,
+                                        176,
+                                        337
+                                    ],
+                                    "score": 1.0,
+                                    "content": "6239/1130.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        }
+                    ],
+                    "index": 15.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        344,
+                        504,
+                        368
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                345,
+                                505,
+                                358
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        345,
+                                        505,
+                                        358
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Nick Bostrom. Superintelligence: Paths, Dangers, Strategies. Oxford University Press, Inc., New",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                356,
+                                255,
+                                367
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        356,
+                                        255,
+                                        367
+                                    ],
+                                    "score": 1.0,
+                                    "content": "York, NY, USA, 1st edition, 2014.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        }
+                    ],
+                    "index": 18.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        375,
+                        504,
+                        398
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                374,
+                                505,
+                                389
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        374,
+                                        505,
+                                        389
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Ryan Carey, Eric Langlois, Tom Everitt, and Shane Legg. The incentives that shape behaviour. arXiv",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                114,
+                                387,
+                                253,
+                                398
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        114,
+                                        387,
+                                        253,
+                                        398
+                                    ],
+                                    "score": 1.0,
+                                    "content": "preprint arXiv:2001.07118, 2020.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        }
+                    ],
+                    "index": 20.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        404,
+                        505,
+                        439
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                405,
+                                505,
+                                418
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        405,
+                                        505,
+                                        418
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Rich Caruana, Yin Lou, Johannes Gehrke, Paul Koch, Marc Sturm, and Noemie Elhadad. Intelligible",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                416,
+                                505,
+                                429
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        416,
+                                        505,
+                                        429
+                                    ],
+                                    "score": 1.0,
+                                    "content": "models for healthcare: Predicting pneumonia risk and hospital 30-day readmission. In International",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                427,
+                                431,
+                                441
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        427,
+                                        431,
+                                        441
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Conference on Knowledge Discovery and Data Mining, pp. 1721–1730, 2015.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        }
+                    ],
+                    "index": 23
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        446,
+                        504,
+                        470
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                444,
+                                505,
+                                460
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        444,
+                                        505,
+                                        460
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Paul Christiano, Jan Leike, Tom B. Brown, Miljan Martic, Shane Legg, and Dario Amodei. Deep",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                458,
+                                337,
+                                469
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        458,
+                                        337,
+                                        469
+                                    ],
+                                    "score": 1.0,
+                                    "content": "reinforcement learning from human preferences, 2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        }
+                    ],
+                    "index": 25.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        476,
+                        505,
+                        500
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                476,
+                                505,
+                                489
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        476,
+                                        505,
+                                        489
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Michael K Cohen, Badri N Vellambi, and Marcus Hutter. Asymptotically unambitious artificial",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                488,
+                                200,
+                                501
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        488,
+                                        200,
+                                        501
+                                    ],
+                                    "score": 1.0,
+                                    "content": "general intelligence.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        }
+                    ],
+                    "index": 27.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        506,
+                        506,
+                        541
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                506,
+                                506,
+                                519
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        506,
+                                        506,
+                                        519
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Dominic DiFranzo and Kristine Gloria-Garcia. Filter bubbles and fake news. XRDS, 23(3):32–35,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                518,
+                                508,
+                                530
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        518,
+                                        508,
+                                        530
+                                    ],
+                                    "score": 1.0,
+                                    "content": "April 2017. ISSN 1528-4972. doi: 10.1145/3055153. URL http://doi.acm.org/10.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                529,
+                                194,
+                                540
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        529,
+                                        194,
+                                        540
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1145/3055153.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        }
+                    ],
+                    "index": 30
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        547,
+                        505,
+                        571
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                546,
+                                507,
+                                561
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        546,
+                                        507,
+                                        561
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Mostafa M. El-Bermawy. Your echo chamber is destroying democracy, 2016. URL https:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                558,
+                                475,
+                                572
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        558,
+                                        475,
+                                        572
+                                    ],
+                                    "score": 1.0,
+                                    "content": "//www.wired.com/2016/11/filter-bubble-destroying-democracy/.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        }
+                    ],
+                    "index": 32.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        577,
+                        505,
+                        601
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                576,
+                                506,
+                                591
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        576,
+                                        506,
+                                        591
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Tom Everitt. Towards Safe Artificial General Intelligence. PhD thesis, Australian National University,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                114,
+                                588,
+                                143,
+                                601
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        114,
+                                        588,
+                                        143,
+                                        601
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        }
+                    ],
+                    "index": 34.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        607,
+                        505,
+                        631
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                606,
+                                505,
+                                621
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        606,
+                                        505,
+                                        621
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Tom Everitt and Marcus Hutter. The alignment problem for bayesian history-based reinforcement",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                619,
+                                179,
+                                630
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        619,
+                                        179,
+                                        630
+                                    ],
+                                    "score": 1.0,
+                                    "content": "learners. 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        }
+                    ],
+                    "index": 36.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        638,
+                        504,
+                        662
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                637,
+                                506,
+                                651
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        637,
+                                        506,
+                                        651
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Tom Everitt and Marcus Hutter. Reward tampering problems and solutions in reinforcement learning:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                649,
+                                441,
+                                662
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        649,
+                                        441,
+                                        662
+                                    ],
+                                    "score": 1.0,
+                                    "content": "A causal influence diagram perspective. arXiv preprint arXiv:1908.04734, 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        }
+                    ],
+                    "index": 38.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        668,
+                        504,
+                        691
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                667,
+                                505,
+                                681
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        667,
+                                        505,
+                                        681
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Tom Everitt, Pedro A. Ortega, Elizabeth Barnes, and Shane Legg. Understanding agent incentives",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                679,
+                                390,
+                                691
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        679,
+                                        390,
+                                        691
+                                    ],
+                                    "score": 1.0,
+                                    "content": "using causal influence diagrams. part i: Single action settings, 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        }
+                    ],
+                    "index": 40.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        698,
+                        505,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                699,
+                                505,
+                                711
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        699,
+                                        505,
+                                        711
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Lisa K. Fazio, Nadia M. Brashier, B. Keith Payne, and Elizabeth J. Marsh. Knowledge does not",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                710,
+                                505,
+                                722
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        710,
+                                        505,
+                                        722
+                                    ],
+                                    "score": 1.0,
+                                    "content": "protect against illusory truth. Journal of Experimental Psychology: General, 144(5):993–1002, 10",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                720,
+                                324,
+                                732
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        720,
+                                        324,
+                                        732
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2015. ISSN 0096-3445. doi: 10.1037/xge0000098.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        }
+                    ],
+                    "index": 43
+                }
+            ],
+            "page_idx": 8,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        302,
+                        751,
+                        309,
+                        759
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                302,
+                                751,
+                                309,
+                                762
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        302,
+                                        751,
+                                        309,
+                                        762
+                                    ],
+                                    "score": 1.0,
+                                    "content": "9",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        107,
+                        27,
+                        307,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                26,
+                                308,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        26,
+                                        308,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Under review as a conference paper at ICLR 2021",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        81,
+                        176,
+                        94
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                82,
+                                176,
+                                94
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        82,
+                                        176,
+                                        94
+                                    ],
+                                    "score": 1.0,
+                                    "content": "REFERENCES",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        }
+                    ],
+                    "index": 0
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        100,
+                        505,
+                        123
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                99,
+                                506,
+                                113
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        99,
+                                        506,
+                                        113
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Hunt Allcott and Matthew Gentzkow. Social media and fake news in the 2016 election. Journal of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                111,
+                                318,
+                                124
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        111,
+                                        318,
+                                        124
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Economic Perspectives, 31(2):211–36, May 2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        }
+                    ],
+                    "index": 1.5,
+                    "bbox_fs": [
+                        105,
+                        99,
+                        506,
+                        124
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        129,
+                        504,
+                        164
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                128,
+                                505,
+                                144
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        128,
+                                        505,
+                                        144
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Michelle A. Amazeen and Bartosz W. Wojdynski. Reducing native advertising deception: Revisiting",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                142,
+                                505,
+                                154
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        142,
+                                        505,
+                                        154
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the antecedents and consequences of persuasion knowledge in digital news contexts. Mass",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                152,
+                                303,
+                                164
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        152,
+                                        303,
+                                        164
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Communication and Society, 0(0):1–26, 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        }
+                    ],
+                    "index": 4,
+                    "bbox_fs": [
+                        105,
+                        128,
+                        505,
+                        164
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        171,
+                        505,
+                        194
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                169,
+                                506,
+                                185
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        169,
+                                        506,
+                                        185
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Stuart Armstrong and Xavier O’Rorke. Good and safe uses of ai oracles. arXiv preprint",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                182,
+                                220,
+                                194
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        182,
+                                        220,
+                                        194
+                                    ],
+                                    "score": 1.0,
+                                    "content": "arXiv:1711.05541, 2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        }
+                    ],
+                    "index": 6.5,
+                    "bbox_fs": [
+                        105,
+                        169,
+                        506,
+                        194
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        201,
+                        505,
+                        224
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                201,
+                                505,
+                                214
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        201,
+                                        505,
+                                        214
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Stuart Armstrong and Xavier O’Rourke. Indifference methods for managing agent rewards. Technical",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                213,
+                                290,
+                                224
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        213,
+                                        290,
+                                        224
+                                    ],
+                                    "score": 1.0,
+                                    "content": "report, Future of Humanity Institute, 2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        }
+                    ],
+                    "index": 8.5,
+                    "bbox_fs": [
+                        106,
+                        201,
+                        505,
+                        224
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        232,
+                        506,
+                        255
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                231,
+                                506,
+                                246
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        231,
+                                        506,
+                                        246
+                                    ],
+                                    "score": 1.0,
+                                    "content": "K. J. Åström. Optimal control of Markov Processes with incomplete state information. Journal of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                244,
+                                394,
+                                255
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        244,
+                                        394,
+                                        255
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Mathematical Analysis and Applications, 10:174–205, January 1965.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        }
+                    ],
+                    "index": 10.5,
+                    "bbox_fs": [
+                        105,
+                        231,
+                        506,
+                        255
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        262,
+                        504,
+                        286
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                262,
+                                505,
+                                276
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        262,
+                                        505,
+                                        276
+                                    ],
+                                    "score": 1.0,
+                                    "content": "P. Auer, N. Cesa-Bianchi, Y. Freund, and R. E. Schapire. Gambling in a rigged casino: The adversarial",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                273,
+                                410,
+                                286
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        273,
+                                        410,
+                                        286
+                                    ],
+                                    "score": 1.0,
+                                    "content": "multi-armed bandit problem. In Foundations of Computer Science, 1995.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        }
+                    ],
+                    "index": 12.5,
+                    "bbox_fs": [
+                        105,
+                        262,
+                        505,
+                        286
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        292,
+                        506,
+                        338
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                293,
+                                506,
+                                305
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        293,
+                                        506,
+                                        305
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Eytan Bakshy, Solomon Messing, and Lada A. Adamic. Exposure to ideologically diverse news",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                304,
+                                507,
+                                317
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        304,
+                                        507,
+                                        317
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and opinion on Facebook. Science, 348(6239):1130–1132, 2015. ISSN 0036-8075. doi:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                314,
+                                506,
+                                328
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        314,
+                                        506,
+                                        328
+                                    ],
+                                    "score": 1.0,
+                                    "content": "10.1126/science.aaa1160. URL http://science.sciencemag.org/content/348/",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                117,
+                                327,
+                                176,
+                                337
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        117,
+                                        327,
+                                        176,
+                                        337
+                                    ],
+                                    "score": 1.0,
+                                    "content": "6239/1130.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        }
+                    ],
+                    "index": 15.5,
+                    "bbox_fs": [
+                        106,
+                        293,
+                        507,
+                        337
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        344,
+                        504,
+                        368
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                345,
+                                505,
+                                358
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        345,
+                                        505,
+                                        358
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Nick Bostrom. Superintelligence: Paths, Dangers, Strategies. Oxford University Press, Inc., New",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                356,
+                                255,
+                                367
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        356,
+                                        255,
+                                        367
+                                    ],
+                                    "score": 1.0,
+                                    "content": "York, NY, USA, 1st edition, 2014.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        }
+                    ],
+                    "index": 18.5,
+                    "bbox_fs": [
+                        106,
+                        345,
+                        505,
+                        367
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        375,
+                        504,
+                        398
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                374,
+                                505,
+                                389
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        374,
+                                        505,
+                                        389
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Ryan Carey, Eric Langlois, Tom Everitt, and Shane Legg. The incentives that shape behaviour. arXiv",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                114,
+                                387,
+                                253,
+                                398
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        114,
+                                        387,
+                                        253,
+                                        398
+                                    ],
+                                    "score": 1.0,
+                                    "content": "preprint arXiv:2001.07118, 2020.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        }
+                    ],
+                    "index": 20.5,
+                    "bbox_fs": [
+                        105,
+                        374,
+                        505,
+                        398
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        404,
+                        505,
+                        439
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                405,
+                                505,
+                                418
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        405,
+                                        505,
+                                        418
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Rich Caruana, Yin Lou, Johannes Gehrke, Paul Koch, Marc Sturm, and Noemie Elhadad. Intelligible",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                416,
+                                505,
+                                429
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        416,
+                                        505,
+                                        429
+                                    ],
+                                    "score": 1.0,
+                                    "content": "models for healthcare: Predicting pneumonia risk and hospital 30-day readmission. In International",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                427,
+                                431,
+                                441
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        427,
+                                        431,
+                                        441
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Conference on Knowledge Discovery and Data Mining, pp. 1721–1730, 2015.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        }
+                    ],
+                    "index": 23,
+                    "bbox_fs": [
+                        106,
+                        405,
+                        505,
+                        441
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        446,
+                        504,
+                        470
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                444,
+                                505,
+                                460
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        444,
+                                        505,
+                                        460
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Paul Christiano, Jan Leike, Tom B. Brown, Miljan Martic, Shane Legg, and Dario Amodei. Deep",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                458,
+                                337,
+                                469
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        458,
+                                        337,
+                                        469
+                                    ],
+                                    "score": 1.0,
+                                    "content": "reinforcement learning from human preferences, 2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        }
+                    ],
+                    "index": 25.5,
+                    "bbox_fs": [
+                        105,
+                        444,
+                        505,
+                        469
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        476,
+                        505,
+                        500
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                476,
+                                505,
+                                489
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        476,
+                                        505,
+                                        489
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Michael K Cohen, Badri N Vellambi, and Marcus Hutter. Asymptotically unambitious artificial",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                488,
+                                200,
+                                501
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        488,
+                                        200,
+                                        501
+                                    ],
+                                    "score": 1.0,
+                                    "content": "general intelligence.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        }
+                    ],
+                    "index": 27.5,
+                    "bbox_fs": [
+                        106,
+                        476,
+                        505,
+                        501
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        506,
+                        506,
+                        541
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                506,
+                                506,
+                                519
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        506,
+                                        506,
+                                        519
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Dominic DiFranzo and Kristine Gloria-Garcia. Filter bubbles and fake news. XRDS, 23(3):32–35,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                518,
+                                508,
+                                530
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        518,
+                                        508,
+                                        530
+                                    ],
+                                    "score": 1.0,
+                                    "content": "April 2017. ISSN 1528-4972. doi: 10.1145/3055153. URL http://doi.acm.org/10.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                529,
+                                194,
+                                540
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        529,
+                                        194,
+                                        540
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1145/3055153.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        }
+                    ],
+                    "index": 30,
+                    "bbox_fs": [
+                        106,
+                        506,
+                        508,
+                        540
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        547,
+                        505,
+                        571
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                546,
+                                507,
+                                561
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        546,
+                                        507,
+                                        561
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Mostafa M. El-Bermawy. Your echo chamber is destroying democracy, 2016. URL https:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                558,
+                                475,
+                                572
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        558,
+                                        475,
+                                        572
+                                    ],
+                                    "score": 1.0,
+                                    "content": "//www.wired.com/2016/11/filter-bubble-destroying-democracy/.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        }
+                    ],
+                    "index": 32.5,
+                    "bbox_fs": [
+                        105,
+                        546,
+                        507,
+                        572
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        577,
+                        505,
+                        601
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                576,
+                                506,
+                                591
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        576,
+                                        506,
+                                        591
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Tom Everitt. Towards Safe Artificial General Intelligence. PhD thesis, Australian National University,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                114,
+                                588,
+                                143,
+                                601
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        114,
+                                        588,
+                                        143,
+                                        601
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        }
+                    ],
+                    "index": 34.5,
+                    "bbox_fs": [
+                        105,
+                        576,
+                        506,
+                        601
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        607,
+                        505,
+                        631
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                606,
+                                505,
+                                621
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        606,
+                                        505,
+                                        621
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Tom Everitt and Marcus Hutter. The alignment problem for bayesian history-based reinforcement",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                619,
+                                179,
+                                630
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        619,
+                                        179,
+                                        630
+                                    ],
+                                    "score": 1.0,
+                                    "content": "learners. 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        }
+                    ],
+                    "index": 36.5,
+                    "bbox_fs": [
+                        106,
+                        606,
+                        505,
+                        630
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        638,
+                        504,
+                        662
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                637,
+                                506,
+                                651
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        637,
+                                        506,
+                                        651
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Tom Everitt and Marcus Hutter. Reward tampering problems and solutions in reinforcement learning:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                649,
+                                441,
+                                662
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        649,
+                                        441,
+                                        662
+                                    ],
+                                    "score": 1.0,
+                                    "content": "A causal influence diagram perspective. arXiv preprint arXiv:1908.04734, 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        }
+                    ],
+                    "index": 38.5,
+                    "bbox_fs": [
+                        105,
+                        637,
+                        506,
+                        662
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        668,
+                        504,
+                        691
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                667,
+                                505,
+                                681
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        667,
+                                        505,
+                                        681
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Tom Everitt, Pedro A. Ortega, Elizabeth Barnes, and Shane Legg. Understanding agent incentives",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                679,
+                                390,
+                                691
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        679,
+                                        390,
+                                        691
+                                    ],
+                                    "score": 1.0,
+                                    "content": "using causal influence diagrams. part i: Single action settings, 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        }
+                    ],
+                    "index": 40.5,
+                    "bbox_fs": [
+                        106,
+                        667,
+                        505,
+                        691
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        698,
+                        505,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                699,
+                                505,
+                                711
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        699,
+                                        505,
+                                        711
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Lisa K. Fazio, Nadia M. Brashier, B. Keith Payne, and Elizabeth J. Marsh. Knowledge does not",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                710,
+                                505,
+                                722
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        710,
+                                        505,
+                                        722
+                                    ],
+                                    "score": 1.0,
+                                    "content": "protect against illusory truth. Journal of Experimental Psychology: General, 144(5):993–1002, 10",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                720,
+                                324,
+                                732
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        720,
+                                        324,
+                                        732
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2015. ISSN 0096-3445. doi: 10.1037/xge0000098.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        }
+                    ],
+                    "index": 43,
+                    "bbox_fs": [
+                        106,
+                        699,
+                        505,
+                        732
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        82,
+                        505,
+                        106
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                81,
+                                505,
+                                95
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        81,
+                                        505,
+                                        95
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Chelsea Finn, Pieter Abbeel, and Sergey Levine. Model-agnostic meta-learning for fast adaptation of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                117,
+                                93,
+                                411,
+                                106
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        117,
+                                        93,
+                                        411,
+                                        106
+                                    ],
+                                    "score": 1.0,
+                                    "content": "deep networks. In International Conference on Machine Learning, 2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        }
+                    ],
+                    "index": 0.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        112,
+                        504,
+                        135
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                111,
+                                505,
+                                125
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        111,
+                                        505,
+                                        125
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Seth Flaxman and Sharad Goel. Filter bubbles, echo chambers, and online news consumption. Public",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                123,
+                                221,
+                                136
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        123,
+                                        221,
+                                        136
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Opinion Quarterly, 2015.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        }
+                    ],
+                    "index": 2.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        142,
+                        505,
+                        165
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                141,
+                                506,
+                                155
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        141,
+                                        506,
+                                        155
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Mohammad Gheshlaghi Azar, Alessandro Lazaric, and Emma Brunskill. Online stochastic optimiza-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                153,
+                                359,
+                                165
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        153,
+                                        359,
+                                        165
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tion under correlated bandit feedback. ArXiv preprint, 2014.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        }
+                    ],
+                    "index": 4.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        172,
+                        505,
+                        195
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                171,
+                                507,
+                                186
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        171,
+                                        507,
+                                        186
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Dong Gong, Zhen Zhang, Qinfeng Shi, Anton van den Hengel, Chunhua Shen, and Yanning Zhang.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                183,
+                                399,
+                                196
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        183,
+                                        399,
+                                        196
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Learning an optimizer for image deconvolution. ArXiv preprint, 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        }
+                    ],
+                    "index": 6.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        202,
+                        506,
+                        247
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                202,
+                                507,
+                                215
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        202,
+                                        507,
+                                        215
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Jacob Groshek and Karolina Koc-Michalska. Helping populism win? Social media use, filter bubbles,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                213,
+                                507,
+                                226
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        213,
+                                        507,
+                                        226
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and support for populist presidential candidates in the 2016 us election campaign. Information,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                225,
+                                506,
+                                236
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        225,
+                                        506,
+                                        236
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Communication & Society, 20(9):1389–1407, 2017. doi: 10.1080/1369118X.2017.1329334. URL",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                236,
+                                391,
+                                248
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        236,
+                                        391,
+                                        248
+                                    ],
+                                    "score": 1.0,
+                                    "content": "https://doi.org/10.1080/1369118X.2017.1329334.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        }
+                    ],
+                    "index": 9.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        253,
+                        503,
+                        277
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                252,
+                                505,
+                                268
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        252,
+                                        505,
+                                        268
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Dylan Hadfield-Menell, Smitha Milli, Pieter Abbeel, Stuart Russell, and Anca Dragan. Inverse",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                265,
+                                378,
+                                277
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        265,
+                                        378,
+                                        277
+                                    ],
+                                    "score": 1.0,
+                                    "content": "reward design. In Neural Information Processing Systems, 2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        }
+                    ],
+                    "index": 12.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        283,
+                        505,
+                        318
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                282,
+                                506,
+                                297
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        282,
+                                        506,
+                                        297
+                                    ],
+                                    "score": 1.0,
+                                    "content": "M. Jaderberg, V. Dalibard, S. Osindero, W. M. Czarnecki, J. Donahue, A. Razavi, O. Vinyals,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                114,
+                                294,
+                                506,
+                                309
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        114,
+                                        294,
+                                        506,
+                                        309
+                                    ],
+                                    "score": 1.0,
+                                    "content": "T. Green, I. Dunning, K. Simonyan, C. Fernando, and K. Kavukcuoglu. Population Based Training",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                305,
+                                289,
+                                318
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        305,
+                                        289,
+                                        318
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of Neural Networks. ArXiv preprint, 2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        }
+                    ],
+                    "index": 15
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        324,
+                        506,
+                        358
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                324,
+                                506,
+                                338
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        324,
+                                        506,
+                                        338
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Leslie Pack Kaelbling, Michael L. Littman, and Anthony R. Cassandra. Planning and acting in",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                336,
+                                506,
+                                349
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        336,
+                                        506,
+                                        349
+                                    ],
+                                    "score": 1.0,
+                                    "content": "partially observable stochastic domains. Artificial Intelligence, 101(1-2):99–134, May 1998. ISSN",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                347,
+                                165,
+                                358
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        347,
+                                        165,
+                                        358
+                                    ],
+                                    "score": 1.0,
+                                    "content": "0004-3702.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        }
+                    ],
+                    "index": 18
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        366,
+                        503,
+                        389
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                366,
+                                505,
+                                379
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        366,
+                                        505,
+                                        379
+                                    ],
+                                    "score": 1.0,
+                                    "content": "A. Kalousis. Model selection via meta-learning: A comparative study. In IEEE International",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                377,
+                                334,
+                                389
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        377,
+                                        334,
+                                        389
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Conference on Tools with Artificial Intelligence, 2000.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        }
+                    ],
+                    "index": 20.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        396,
+                        504,
+                        419
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                107,
+                                396,
+                                505,
+                                408
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        396,
+                                        505,
+                                        408
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Varol Kayhan. Confirmation bias: Roles of search engines and search contexts. In International",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                407,
+                                289,
+                                419
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        407,
+                                        289,
+                                        419
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Conference on Information Systems, 2015.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        }
+                    ],
+                    "index": 22.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        425,
+                        505,
+                        471
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                424,
+                                506,
+                                438
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        426,
+                                        191,
+                                        437
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Ed. Lee Howell.",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        217,
+                                        425,
+                                        442,
+                                        438
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Digital wildfires in a hyperconnected world.",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        466,
+                                        424,
+                                        506,
+                                        438
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In Lee",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                436,
+                                507,
+                                449
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        436,
+                                        456,
+                                        449
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Howell (ed.), Global Risks 2013. World Economic Forum, 2013.",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        481,
+                                        437,
+                                        507,
+                                        448
+                                    ],
+                                    "score": 1.0,
+                                    "content": "URL",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                447,
+                                461,
+                                461
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        447,
+                                        461,
+                                        461
+                                    ],
+                                    "score": 1.0,
+                                    "content": "http://reports.weforum.org/global-risks-2013/risk-case-1/",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                459,
+                                388,
+                                471
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        459,
+                                        388,
+                                        471
+                                    ],
+                                    "score": 1.0,
+                                    "content": "digital-wildfires-in-a-hyperconnected-world/.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        }
+                    ],
+                    "index": 25.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        103,
+                        477,
+                        505,
+                        501
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                475,
+                                506,
+                                492
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        475,
+                                        506,
+                                        492
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Jan Leike, Miljan Martic, Victoria Krakovna, Pedro A. Ortega, Tom Everitt, Andrew Lefrancq,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                489,
+                                484,
+                                502
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        489,
+                                        484,
+                                        502
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Laurent Orseau, and Shane Legg. AI safety gridworlds. Technical report, DeepMind, 2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        }
+                    ],
+                    "index": 28.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        507,
+                        505,
+                        541
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                506,
+                                505,
+                                521
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        506,
+                                        505,
+                                        521
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Jan Leike, David Krueger, Tom Everitt, Miljan Martic, Vishal Maini, and Shane Legg. Scalable agent",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                519,
+                                506,
+                                531
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        519,
+                                        506,
+                                        531
+                                    ],
+                                    "score": 1.0,
+                                    "content": "alignment via reward modeling: a research direction. Technical report, DeepMind Safety Research,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                529,
+                                143,
+                                542
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        529,
+                                        143,
+                                        542
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        }
+                    ],
+                    "index": 31
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        548,
+                        505,
+                        572
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                548,
+                                505,
+                                561
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        548,
+                                        505,
+                                        561
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Lihong Li, Wei Chu, John Langford, and Robert E. Schapire. A contextual-bandit approach to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                560,
+                                506,
+                                572
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        560,
+                                        506,
+                                        572
+                                    ],
+                                    "score": 1.0,
+                                    "content": "personalized news article recommendation. In International Conference on World Wide Web, 2010.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        }
+                    ],
+                    "index": 33.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        578,
+                        505,
+                        601
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                577,
+                                507,
+                                592
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        577,
+                                        507,
+                                        592
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Jonathan Lorraine and David Duvenaud. Stochastic hyperparameter optimization through hypernet-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                590,
+                                234,
+                                601
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        590,
+                                        234,
+                                        601
+                                    ],
+                                    "score": 1.0,
+                                    "content": "works. ArXiv preprint, 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        }
+                    ],
+                    "index": 35.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        104,
+                        608,
+                        505,
+                        632
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                608,
+                                506,
+                                622
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        608,
+                                        506,
+                                        622
+                                    ],
+                                    "score": 1.0,
+                                    "content": "D. D. Luxton, J. D. June, and J. M. Fairall. Social media and suicide: A public health perspective.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                114,
+                                619,
+                                351,
+                                632
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        114,
+                                        619,
+                                        351,
+                                        632
+                                    ],
+                                    "score": 1.0,
+                                    "content": "American journal of public health, 102(2):195–200, 2012.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        }
+                    ],
+                    "index": 37.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        638,
+                        504,
+                        672
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                636,
+                                506,
+                                653
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        636,
+                                        506,
+                                        653
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Matthew MacKay, Paul Vicol, Jonathan Lorraine, David Duvenaud, and Roger Grosse. Self-tuning",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                650,
+                                505,
+                                662
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        650,
+                                        505,
+                                        662
+                                    ],
+                                    "score": 1.0,
+                                    "content": "networks: Bilevel optimization of hyperparameters using structured best-response functions. ArXiv",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                114,
+                                661,
+                                179,
+                                672
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        114,
+                                        661,
+                                        179,
+                                        672
+                                    ],
+                                    "score": 1.0,
+                                    "content": "preprint, 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        }
+                    ],
+                    "index": 40
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        679,
+                        503,
+                        703
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                679,
+                                503,
+                                692
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        679,
+                                        503,
+                                        692
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Merriam-Webster. The real story of fake news, 2017. URL https://www.merriam-webster.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                117,
+                                690,
+                                394,
+                                703
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        117,
+                                        690,
+                                        394,
+                                        703
+                                    ],
+                                    "score": 1.0,
+                                    "content": "com/words-at-play/the-real-story-of-fake-news.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        }
+                    ],
+                    "index": 42.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        709,
+                        502,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                709,
+                                504,
+                                722
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        709,
+                                        504,
+                                        722
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Luke Metz, Niru Maheswaranathan, Brian Cheung, and Jascha Sohl-Dickstein. Learning unsupervised",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                720,
+                                437,
+                                732
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        720,
+                                        437,
+                                        732
+                                    ],
+                                    "score": 1.0,
+                                    "content": "learning rules. In International Conference on Learning Representations, 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        }
+                    ],
+                    "index": 44.5
+                }
+            ],
+            "page_idx": 9,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        300,
+                        751,
+                        311,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                299,
+                                750,
+                                312,
+                                764
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        299,
+                                        750,
+                                        312,
+                                        764
+                                    ],
+                                    "score": 1.0,
+                                    "content": "10",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        107,
+                        27,
+                        307,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                26,
+                                308,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        26,
+                                        308,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Under review as a conference paper at ICLR 2021",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        82,
+                        505,
+                        106
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                81,
+                                505,
+                                95
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        81,
+                                        505,
+                                        95
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Chelsea Finn, Pieter Abbeel, and Sergey Levine. Model-agnostic meta-learning for fast adaptation of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                117,
+                                93,
+                                411,
+                                106
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        117,
+                                        93,
+                                        411,
+                                        106
+                                    ],
+                                    "score": 1.0,
+                                    "content": "deep networks. In International Conference on Machine Learning, 2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        }
+                    ],
+                    "index": 0.5,
+                    "bbox_fs": [
+                        105,
+                        81,
+                        505,
+                        106
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        112,
+                        504,
+                        135
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                111,
+                                505,
+                                125
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        111,
+                                        505,
+                                        125
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Seth Flaxman and Sharad Goel. Filter bubbles, echo chambers, and online news consumption. Public",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                123,
+                                221,
+                                136
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        123,
+                                        221,
+                                        136
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Opinion Quarterly, 2015.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        }
+                    ],
+                    "index": 2.5,
+                    "bbox_fs": [
+                        105,
+                        111,
+                        505,
+                        136
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        142,
+                        505,
+                        165
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                141,
+                                506,
+                                155
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        141,
+                                        506,
+                                        155
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Mohammad Gheshlaghi Azar, Alessandro Lazaric, and Emma Brunskill. Online stochastic optimiza-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                153,
+                                359,
+                                165
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        153,
+                                        359,
+                                        165
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tion under correlated bandit feedback. ArXiv preprint, 2014.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        }
+                    ],
+                    "index": 4.5,
+                    "bbox_fs": [
+                        105,
+                        141,
+                        506,
+                        165
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        172,
+                        505,
+                        195
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                171,
+                                507,
+                                186
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        171,
+                                        507,
+                                        186
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Dong Gong, Zhen Zhang, Qinfeng Shi, Anton van den Hengel, Chunhua Shen, and Yanning Zhang.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                183,
+                                399,
+                                196
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        183,
+                                        399,
+                                        196
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Learning an optimizer for image deconvolution. ArXiv preprint, 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        }
+                    ],
+                    "index": 6.5,
+                    "bbox_fs": [
+                        105,
+                        171,
+                        507,
+                        196
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        202,
+                        506,
+                        247
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                202,
+                                507,
+                                215
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        202,
+                                        507,
+                                        215
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Jacob Groshek and Karolina Koc-Michalska. Helping populism win? Social media use, filter bubbles,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                213,
+                                507,
+                                226
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        213,
+                                        507,
+                                        226
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and support for populist presidential candidates in the 2016 us election campaign. Information,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                225,
+                                506,
+                                236
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        225,
+                                        506,
+                                        236
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Communication & Society, 20(9):1389–1407, 2017. doi: 10.1080/1369118X.2017.1329334. URL",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                236,
+                                391,
+                                248
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        236,
+                                        391,
+                                        248
+                                    ],
+                                    "score": 1.0,
+                                    "content": "https://doi.org/10.1080/1369118X.2017.1329334.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        }
+                    ],
+                    "index": 9.5,
+                    "bbox_fs": [
+                        105,
+                        202,
+                        507,
+                        248
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        253,
+                        503,
+                        277
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                252,
+                                505,
+                                268
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        252,
+                                        505,
+                                        268
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Dylan Hadfield-Menell, Smitha Milli, Pieter Abbeel, Stuart Russell, and Anca Dragan. Inverse",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                265,
+                                378,
+                                277
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        265,
+                                        378,
+                                        277
+                                    ],
+                                    "score": 1.0,
+                                    "content": "reward design. In Neural Information Processing Systems, 2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        }
+                    ],
+                    "index": 12.5,
+                    "bbox_fs": [
+                        105,
+                        252,
+                        505,
+                        277
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        283,
+                        505,
+                        318
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                282,
+                                506,
+                                297
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        282,
+                                        506,
+                                        297
+                                    ],
+                                    "score": 1.0,
+                                    "content": "M. Jaderberg, V. Dalibard, S. Osindero, W. M. Czarnecki, J. Donahue, A. Razavi, O. Vinyals,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                114,
+                                294,
+                                506,
+                                309
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        114,
+                                        294,
+                                        506,
+                                        309
+                                    ],
+                                    "score": 1.0,
+                                    "content": "T. Green, I. Dunning, K. Simonyan, C. Fernando, and K. Kavukcuoglu. Population Based Training",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                305,
+                                289,
+                                318
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        305,
+                                        289,
+                                        318
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of Neural Networks. ArXiv preprint, 2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        }
+                    ],
+                    "index": 15,
+                    "bbox_fs": [
+                        105,
+                        282,
+                        506,
+                        318
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        324,
+                        506,
+                        358
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                324,
+                                506,
+                                338
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        324,
+                                        506,
+                                        338
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Leslie Pack Kaelbling, Michael L. Littman, and Anthony R. Cassandra. Planning and acting in",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                336,
+                                506,
+                                349
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        336,
+                                        506,
+                                        349
+                                    ],
+                                    "score": 1.0,
+                                    "content": "partially observable stochastic domains. Artificial Intelligence, 101(1-2):99–134, May 1998. ISSN",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                347,
+                                165,
+                                358
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        347,
+                                        165,
+                                        358
+                                    ],
+                                    "score": 1.0,
+                                    "content": "0004-3702.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        }
+                    ],
+                    "index": 18,
+                    "bbox_fs": [
+                        104,
+                        324,
+                        506,
+                        358
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        366,
+                        503,
+                        389
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                366,
+                                505,
+                                379
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        366,
+                                        505,
+                                        379
+                                    ],
+                                    "score": 1.0,
+                                    "content": "A. Kalousis. Model selection via meta-learning: A comparative study. In IEEE International",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                377,
+                                334,
+                                389
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        377,
+                                        334,
+                                        389
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Conference on Tools with Artificial Intelligence, 2000.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        }
+                    ],
+                    "index": 20.5,
+                    "bbox_fs": [
+                        106,
+                        366,
+                        505,
+                        389
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        396,
+                        504,
+                        419
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                107,
+                                396,
+                                505,
+                                408
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        396,
+                                        505,
+                                        408
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Varol Kayhan. Confirmation bias: Roles of search engines and search contexts. In International",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                407,
+                                289,
+                                419
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        407,
+                                        289,
+                                        419
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Conference on Information Systems, 2015.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        }
+                    ],
+                    "index": 22.5,
+                    "bbox_fs": [
+                        107,
+                        396,
+                        505,
+                        419
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        425,
+                        505,
+                        471
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                424,
+                                506,
+                                438
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        426,
+                                        191,
+                                        437
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Ed. Lee Howell.",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        217,
+                                        425,
+                                        442,
+                                        438
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Digital wildfires in a hyperconnected world.",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        466,
+                                        424,
+                                        506,
+                                        438
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In Lee",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                436,
+                                507,
+                                449
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        436,
+                                        456,
+                                        449
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Howell (ed.), Global Risks 2013. World Economic Forum, 2013.",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        481,
+                                        437,
+                                        507,
+                                        448
+                                    ],
+                                    "score": 1.0,
+                                    "content": "URL",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                447,
+                                461,
+                                461
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        447,
+                                        461,
+                                        461
+                                    ],
+                                    "score": 1.0,
+                                    "content": "http://reports.weforum.org/global-risks-2013/risk-case-1/",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                459,
+                                388,
+                                471
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        459,
+                                        388,
+                                        471
+                                    ],
+                                    "score": 1.0,
+                                    "content": "digital-wildfires-in-a-hyperconnected-world/.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        }
+                    ],
+                    "index": 25.5,
+                    "bbox_fs": [
+                        106,
+                        424,
+                        507,
+                        471
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        103,
+                        477,
+                        505,
+                        501
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                475,
+                                506,
+                                492
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        475,
+                                        506,
+                                        492
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Jan Leike, Miljan Martic, Victoria Krakovna, Pedro A. Ortega, Tom Everitt, Andrew Lefrancq,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                489,
+                                484,
+                                502
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        489,
+                                        484,
+                                        502
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Laurent Orseau, and Shane Legg. AI safety gridworlds. Technical report, DeepMind, 2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        }
+                    ],
+                    "index": 28.5,
+                    "bbox_fs": [
+                        104,
+                        475,
+                        506,
+                        502
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        507,
+                        505,
+                        541
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                506,
+                                505,
+                                521
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        506,
+                                        505,
+                                        521
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Jan Leike, David Krueger, Tom Everitt, Miljan Martic, Vishal Maini, and Shane Legg. Scalable agent",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                519,
+                                506,
+                                531
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        519,
+                                        506,
+                                        531
+                                    ],
+                                    "score": 1.0,
+                                    "content": "alignment via reward modeling: a research direction. Technical report, DeepMind Safety Research,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                529,
+                                143,
+                                542
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        529,
+                                        143,
+                                        542
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        }
+                    ],
+                    "index": 31,
+                    "bbox_fs": [
+                        105,
+                        506,
+                        506,
+                        542
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        548,
+                        505,
+                        572
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                548,
+                                505,
+                                561
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        548,
+                                        505,
+                                        561
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Lihong Li, Wei Chu, John Langford, and Robert E. Schapire. A contextual-bandit approach to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                560,
+                                506,
+                                572
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        560,
+                                        506,
+                                        572
+                                    ],
+                                    "score": 1.0,
+                                    "content": "personalized news article recommendation. In International Conference on World Wide Web, 2010.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        }
+                    ],
+                    "index": 33.5,
+                    "bbox_fs": [
+                        106,
+                        548,
+                        506,
+                        572
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        578,
+                        505,
+                        601
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                577,
+                                507,
+                                592
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        577,
+                                        507,
+                                        592
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Jonathan Lorraine and David Duvenaud. Stochastic hyperparameter optimization through hypernet-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                590,
+                                234,
+                                601
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        590,
+                                        234,
+                                        601
+                                    ],
+                                    "score": 1.0,
+                                    "content": "works. ArXiv preprint, 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        }
+                    ],
+                    "index": 35.5,
+                    "bbox_fs": [
+                        105,
+                        577,
+                        507,
+                        601
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        104,
+                        608,
+                        505,
+                        632
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                608,
+                                506,
+                                622
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        608,
+                                        506,
+                                        622
+                                    ],
+                                    "score": 1.0,
+                                    "content": "D. D. Luxton, J. D. June, and J. M. Fairall. Social media and suicide: A public health perspective.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                114,
+                                619,
+                                351,
+                                632
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        114,
+                                        619,
+                                        351,
+                                        632
+                                    ],
+                                    "score": 1.0,
+                                    "content": "American journal of public health, 102(2):195–200, 2012.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        }
+                    ],
+                    "index": 37.5,
+                    "bbox_fs": [
+                        105,
+                        608,
+                        506,
+                        632
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        638,
+                        504,
+                        672
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                636,
+                                506,
+                                653
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        636,
+                                        506,
+                                        653
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Matthew MacKay, Paul Vicol, Jonathan Lorraine, David Duvenaud, and Roger Grosse. Self-tuning",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                650,
+                                505,
+                                662
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        650,
+                                        505,
+                                        662
+                                    ],
+                                    "score": 1.0,
+                                    "content": "networks: Bilevel optimization of hyperparameters using structured best-response functions. ArXiv",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                114,
+                                661,
+                                179,
+                                672
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        114,
+                                        661,
+                                        179,
+                                        672
+                                    ],
+                                    "score": 1.0,
+                                    "content": "preprint, 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        }
+                    ],
+                    "index": 40,
+                    "bbox_fs": [
+                        105,
+                        636,
+                        506,
+                        672
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        679,
+                        503,
+                        703
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                679,
+                                503,
+                                692
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        679,
+                                        503,
+                                        692
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Merriam-Webster. The real story of fake news, 2017. URL https://www.merriam-webster.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                117,
+                                690,
+                                394,
+                                703
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        117,
+                                        690,
+                                        394,
+                                        703
+                                    ],
+                                    "score": 1.0,
+                                    "content": "com/words-at-play/the-real-story-of-fake-news.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        }
+                    ],
+                    "index": 42.5,
+                    "bbox_fs": [
+                        105,
+                        679,
+                        503,
+                        703
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        709,
+                        502,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                709,
+                                504,
+                                722
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        709,
+                                        504,
+                                        722
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Luke Metz, Niru Maheswaranathan, Brian Cheung, and Jascha Sohl-Dickstein. Learning unsupervised",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                720,
+                                437,
+                                732
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        720,
+                                        437,
+                                        732
+                                    ],
+                                    "score": 1.0,
+                                    "content": "learning rules. In International Conference on Learning Representations, 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        }
+                    ],
+                    "index": 44.5,
+                    "bbox_fs": [
+                        105,
+                        709,
+                        504,
+                        732
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        82,
+                        505,
+                        105
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                81,
+                                505,
+                                95
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        81,
+                                        505,
+                                        95
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Paul Mihailidis and Samantha Viotty. Spreadable spectacle in digital culture: Civic expression, fake",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                93,
+                                507,
+                                106
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        93,
+                                        507,
+                                        106
+                                    ],
+                                    "score": 1.0,
+                                    "content": "news, and the role of media literacies in \"post-fact\" society. American Behavioural Scientist, 2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        }
+                    ],
+                    "index": 0.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        111,
+                        505,
+                        145
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                110,
+                                505,
+                                124
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        110,
+                                        505,
+                                        124
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Jose G. Moreno-Torres, Troy Raeder, RocíO Alaiz-RodríGuez, Nitesh V. Chawla, and Francisco",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                123,
+                                506,
+                                135
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        123,
+                                        506,
+                                        135
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Herrera. A unifying view on dataset shift in classification. Pattern Recognition, 45(1):521–530,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                133,
+                                174,
+                                146
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        133,
+                                        174,
+                                        146
+                                    ],
+                                    "score": 1.0,
+                                    "content": "January 2012.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        }
+                    ],
+                    "index": 3
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        152,
+                        505,
+                        208
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                151,
+                                505,
+                                166
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        151,
+                                        505,
+                                        166
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Tien T. Nguyen, Pik-Mai Hui, F. Maxwell Harper, Loren Terveen, and Joseph A. Konstan. Exploring",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                163,
+                                505,
+                                176
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        163,
+                                        505,
+                                        176
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the filter bubble: The effect of using recommender systems on content diversity. In Proceedings",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                173,
+                                506,
+                                187
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        173,
+                                        506,
+                                        187
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of the 23rd International Conference on World Wide Web, WWW ’14, pp. 677–686, New York,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                185,
+                                506,
+                                197
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        185,
+                                        506,
+                                        197
+                                    ],
+                                    "score": 1.0,
+                                    "content": "NY, USA, 2014. ACM. ISBN 978-1-4503-2744-2. doi: 10.1145/2566486.2568012. URL",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                195,
+                                373,
+                                209
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        195,
+                                        373,
+                                        209
+                                    ],
+                                    "score": 1.0,
+                                    "content": "http://doi.acm.org/10.1145/2566486.2568012.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        }
+                    ],
+                    "index": 7
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        214,
+                        505,
+                        237
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                212,
+                                506,
+                                228
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        212,
+                                        506,
+                                        228
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Safiya Umoja Noble. Algorithms of Oppression: How Search Engines Reinforce Racism. NYC Press,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                225,
+                                142,
+                                237
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        225,
+                                        142,
+                                        237
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        }
+                    ],
+                    "index": 10.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        243,
+                        505,
+                        256
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                243,
+                                506,
+                                257
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        243,
+                                        506,
+                                        257
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Stephen M. Omohundro. The basic AI drives. In Conference on Artificial General Intelligence, 2008.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        }
+                    ],
+                    "index": 12
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        261,
+                        506,
+                        285
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                259,
+                                507,
+                                276
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        259,
+                                        507,
+                                        276
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Pedro A. Ortega, Vishal Maini, et al. Building safe artificial intelligence: specification, robustness,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                273,
+                                202,
+                                284
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        273,
+                                        202,
+                                        284
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and assurance, 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        }
+                    ],
+                    "index": 13.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        290,
+                        493,
+                        304
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                289,
+                                493,
+                                305
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        289,
+                                        493,
+                                        305
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Eli Pariser. The Filter Bubble: What the Internet Is Hiding from You. The Penguin Group, 2011.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        }
+                    ],
+                    "index": 15
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        308,
+                        505,
+                        333
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                308,
+                                505,
+                                322
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        308,
+                                        505,
+                                        322
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Gordon Pennycook, Tyrone D Cannon, and David G. Rand. Prior exposure increases perceived",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                321,
+                                444,
+                                333
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        321,
+                                        444,
+                                        333
+                                    ],
+                                    "score": 1.0,
+                                    "content": "accuracy of fake news. Journal of Experimental Psychology (forthcoming), 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        }
+                    ],
+                    "index": 16.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        338,
+                        485,
+                        351
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                338,
+                                485,
+                                351
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        338,
+                                        485,
+                                        351
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Erich Prisner. Game Theory Through Examples. Mathematical Association of America, 2014.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        }
+                    ],
+                    "index": 18
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        357,
+                        505,
+                        380
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                357,
+                                506,
+                                370
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        357,
+                                        506,
+                                        370
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Joaquin Quionero-Candela, Masashi Sugiyama, Anton Schwaighofer, and Neil D. Lawrence. Dataset",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                369,
+                                316,
+                                380
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        369,
+                                        316,
+                                        380
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Shift in Machine Learning. The MIT Press, 2009.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        }
+                    ],
+                    "index": 19.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        386,
+                        496,
+                        399
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                385,
+                                497,
+                                400
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        385,
+                                        497,
+                                        400
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Neil C. Rabinowitz. Meta-learners’ learning dynamics are unlike learners’. ArXiv preprint, 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        }
+                    ],
+                    "index": 21
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        405,
+                        503,
+                        428
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                404,
+                                505,
+                                418
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        404,
+                                        505,
+                                        418
+                                    ],
+                                    "score": 1.0,
+                                    "content": "David Robson. The myth of the online echo chamber, 2018. URL http://www.bbc.com/",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                117,
+                                416,
+                                466,
+                                429
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        117,
+                                        416,
+                                        466,
+                                        429
+                                    ],
+                                    "score": 1.0,
+                                    "content": "future/story/20180416-the-myth-of-the-online-echo-chamber.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        }
+                    ],
+                    "index": 22.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        434,
+                        505,
+                        457
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                434,
+                                506,
+                                448
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        434,
+                                        506,
+                                        448
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Peter Schulam and Suchi Saria. Reliable decision support using counterfactual models. In Neural",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                444,
+                                274,
+                                457
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        444,
+                                        274,
+                                        457
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Information Processing Systems, 2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        }
+                    ],
+                    "index": 24.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        463,
+                        505,
+                        487
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                463,
+                                506,
+                                477
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        463,
+                                        506,
+                                        477
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Virag Shah, Jose Blanchet, and Ramesh Johari. Bandit learning with positive externalities. Neural",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                475,
+                                274,
+                                488
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        475,
+                                        274,
+                                        488
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Information Processing Systems, 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        }
+                    ],
+                    "index": 26.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        492,
+                        506,
+                        527
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                492,
+                                506,
+                                506
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        492,
+                                        506,
+                                        506
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Chengcheng Shao, Giovanni Luca Ciampaglia, Onur Varol, Kai-Cheng Yang, Alessandro Flammini,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                504,
+                                507,
+                                516
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        504,
+                                        507,
+                                        516
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and Filippo Menczer. The spread of low-credibility content by social bots. Nature Communications,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                515,
+                                178,
+                                527
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        515,
+                                        178,
+                                        527
+                                    ],
+                                    "score": 1.0,
+                                    "content": "9(4787), 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        }
+                    ],
+                    "index": 29
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        533,
+                        504,
+                        556
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                533,
+                                505,
+                                546
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        533,
+                                        505,
+                                        546
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Jasper Snoek, Hugo Larochelle, and Ryan P Adams. Practical Bayesian optimization of machine",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                544,
+                                399,
+                                556
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        544,
+                                        399,
+                                        556
+                                    ],
+                                    "score": 1.0,
+                                    "content": "learning algorithms. In Neural Information Processing Systems, 2012.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        }
+                    ],
+                    "index": 31.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        562,
+                        502,
+                        575
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                561,
+                                502,
+                                576
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        561,
+                                        502,
+                                        576
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Richard S Sutton and Andrew G Barto. Introduction to Reinforcement Learning. MIT Press, 1998.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        }
+                    ],
+                    "index": 33
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        581,
+                        504,
+                        604
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                580,
+                                505,
+                                595
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        580,
+                                        505,
+                                        595
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Richard S Sutton, Anna Koop, and David Silver. On the role of tracking in stationary environments.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                592,
+                                342,
+                                604
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        592,
+                                        342,
+                                        604
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In International conference on Machine learning, 2007.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        }
+                    ],
+                    "index": 34.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        610,
+                        504,
+                        633
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                610,
+                                505,
+                                623
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        610,
+                                        505,
+                                        623
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Techopedia. Filter bubble, 2018. URL https://www.techopedia.com/definition/",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                621,
+                                235,
+                                633
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        621,
+                                        235,
+                                        633
+                                    ],
+                                    "score": 1.0,
+                                    "content": "28556/filter-bubble.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        }
+                    ],
+                    "index": 36.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        639,
+                        504,
+                        663
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                640,
+                                506,
+                                652
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        640,
+                                        506,
+                                        652
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Chih-Chun Wang, Sanjeev R Kulkarni, and H Vincent Poor. Bandit problems with side observations.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                651,
+                                375,
+                                663
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        651,
+                                        375,
+                                        663
+                                    ],
+                                    "score": 1.0,
+                                    "content": "IEEE Transactions on Automatic Control, 50(3):338–355, 2005.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        }
+                    ],
+                    "index": 38.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        669,
+                        505,
+                        703
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                668,
+                                506,
+                                682
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        668,
+                                        506,
+                                        682
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Wikipedia contributors. Confirmation bias — Wikipedia, the free encyclopedia, 2018. URL",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                680,
+                                484,
+                                693
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        680,
+                                        484,
+                                        693
+                                    ],
+                                    "score": 1.0,
+                                    "content": "https://en.wikipedia.org/w/index.php?title=Confirmation_bias&",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                691,
+                                361,
+                                704
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        691,
+                                        361,
+                                        704
+                                    ],
+                                    "score": 1.0,
+                                    "content": "oldid=875026726. [Online; accessed 20-January-2019].",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        }
+                    ],
+                    "index": 41
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        709,
+                        504,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                709,
+                                505,
+                                722
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        709,
+                                        505,
+                                        722
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Ronald J. Williams. Simple statistical gradient-following algorithms for connectionist reinforcement",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                720,
+                                324,
+                                732
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        720,
+                                        324,
+                                        732
+                                    ],
+                                    "score": 1.0,
+                                    "content": "learning. In Machine Learning, pp. 229–256, 1992.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        }
+                    ],
+                    "index": 43.5
+                }
+            ],
+            "page_idx": 10,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        107,
+                        27,
+                        307,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                26,
+                                308,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        26,
+                                        308,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Under review as a conference paper at ICLR 2021",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        300,
+                        751,
+                        310,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                299,
+                                750,
+                                312,
+                                765
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        299,
+                                        750,
+                                        312,
+                                        765
+                                    ],
+                                    "score": 1.0,
+                                    "content": "",
+                                    "type": "text",
+                                    "height": 15,
+                                    "width": 13
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        82,
+                        505,
+                        105
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                81,
+                                505,
+                                95
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        81,
+                                        505,
+                                        95
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Paul Mihailidis and Samantha Viotty. Spreadable spectacle in digital culture: Civic expression, fake",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                93,
+                                507,
+                                106
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        93,
+                                        507,
+                                        106
+                                    ],
+                                    "score": 1.0,
+                                    "content": "news, and the role of media literacies in \"post-fact\" society. American Behavioural Scientist, 2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        }
+                    ],
+                    "index": 0.5,
+                    "bbox_fs": [
+                        105,
+                        81,
+                        507,
+                        106
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        111,
+                        505,
+                        145
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                110,
+                                505,
+                                124
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        110,
+                                        505,
+                                        124
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Jose G. Moreno-Torres, Troy Raeder, RocíO Alaiz-RodríGuez, Nitesh V. Chawla, and Francisco",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                123,
+                                506,
+                                135
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        123,
+                                        506,
+                                        135
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Herrera. A unifying view on dataset shift in classification. Pattern Recognition, 45(1):521–530,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                133,
+                                174,
+                                146
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        133,
+                                        174,
+                                        146
+                                    ],
+                                    "score": 1.0,
+                                    "content": "January 2012.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        }
+                    ],
+                    "index": 3,
+                    "bbox_fs": [
+                        106,
+                        110,
+                        506,
+                        146
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        152,
+                        505,
+                        208
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                151,
+                                505,
+                                166
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        151,
+                                        505,
+                                        166
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Tien T. Nguyen, Pik-Mai Hui, F. Maxwell Harper, Loren Terveen, and Joseph A. Konstan. Exploring",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                163,
+                                505,
+                                176
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        163,
+                                        505,
+                                        176
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the filter bubble: The effect of using recommender systems on content diversity. In Proceedings",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                173,
+                                506,
+                                187
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        173,
+                                        506,
+                                        187
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of the 23rd International Conference on World Wide Web, WWW ’14, pp. 677–686, New York,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                185,
+                                506,
+                                197
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        185,
+                                        506,
+                                        197
+                                    ],
+                                    "score": 1.0,
+                                    "content": "NY, USA, 2014. ACM. ISBN 978-1-4503-2744-2. doi: 10.1145/2566486.2568012. URL",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                195,
+                                373,
+                                209
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        195,
+                                        373,
+                                        209
+                                    ],
+                                    "score": 1.0,
+                                    "content": "http://doi.acm.org/10.1145/2566486.2568012.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        }
+                    ],
+                    "index": 7,
+                    "bbox_fs": [
+                        105,
+                        151,
+                        506,
+                        209
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        214,
+                        505,
+                        237
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                212,
+                                506,
+                                228
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        212,
+                                        506,
+                                        228
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Safiya Umoja Noble. Algorithms of Oppression: How Search Engines Reinforce Racism. NYC Press,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                225,
+                                142,
+                                237
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        225,
+                                        142,
+                                        237
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        }
+                    ],
+                    "index": 10.5,
+                    "bbox_fs": [
+                        105,
+                        212,
+                        506,
+                        237
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        243,
+                        505,
+                        256
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                243,
+                                506,
+                                257
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        243,
+                                        506,
+                                        257
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Stephen M. Omohundro. The basic AI drives. In Conference on Artificial General Intelligence, 2008.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        }
+                    ],
+                    "index": 12,
+                    "bbox_fs": [
+                        106,
+                        243,
+                        506,
+                        257
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        261,
+                        506,
+                        285
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                259,
+                                507,
+                                276
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        259,
+                                        507,
+                                        276
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Pedro A. Ortega, Vishal Maini, et al. Building safe artificial intelligence: specification, robustness,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                273,
+                                202,
+                                284
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        273,
+                                        202,
+                                        284
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and assurance, 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        }
+                    ],
+                    "index": 13.5,
+                    "bbox_fs": [
+                        105,
+                        259,
+                        507,
+                        284
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        290,
+                        493,
+                        304
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                289,
+                                493,
+                                305
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        289,
+                                        493,
+                                        305
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Eli Pariser. The Filter Bubble: What the Internet Is Hiding from You. The Penguin Group, 2011.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        }
+                    ],
+                    "index": 15,
+                    "bbox_fs": [
+                        105,
+                        289,
+                        493,
+                        305
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        308,
+                        505,
+                        333
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                308,
+                                505,
+                                322
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        308,
+                                        505,
+                                        322
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Gordon Pennycook, Tyrone D Cannon, and David G. Rand. Prior exposure increases perceived",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                321,
+                                444,
+                                333
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        321,
+                                        444,
+                                        333
+                                    ],
+                                    "score": 1.0,
+                                    "content": "accuracy of fake news. Journal of Experimental Psychology (forthcoming), 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        }
+                    ],
+                    "index": 16.5,
+                    "bbox_fs": [
+                        106,
+                        308,
+                        505,
+                        333
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        338,
+                        485,
+                        351
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                338,
+                                485,
+                                351
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        338,
+                                        485,
+                                        351
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Erich Prisner. Game Theory Through Examples. Mathematical Association of America, 2014.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        }
+                    ],
+                    "index": 18,
+                    "bbox_fs": [
+                        105,
+                        338,
+                        485,
+                        351
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        357,
+                        505,
+                        380
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                357,
+                                506,
+                                370
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        357,
+                                        506,
+                                        370
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Joaquin Quionero-Candela, Masashi Sugiyama, Anton Schwaighofer, and Neil D. Lawrence. Dataset",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                369,
+                                316,
+                                380
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        369,
+                                        316,
+                                        380
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Shift in Machine Learning. The MIT Press, 2009.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        }
+                    ],
+                    "index": 19.5,
+                    "bbox_fs": [
+                        106,
+                        357,
+                        506,
+                        380
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        386,
+                        496,
+                        399
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                385,
+                                497,
+                                400
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        385,
+                                        497,
+                                        400
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Neil C. Rabinowitz. Meta-learners’ learning dynamics are unlike learners’. ArXiv preprint, 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        }
+                    ],
+                    "index": 21,
+                    "bbox_fs": [
+                        105,
+                        385,
+                        497,
+                        400
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        405,
+                        503,
+                        428
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                404,
+                                505,
+                                418
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        404,
+                                        505,
+                                        418
+                                    ],
+                                    "score": 1.0,
+                                    "content": "David Robson. The myth of the online echo chamber, 2018. URL http://www.bbc.com/",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                117,
+                                416,
+                                466,
+                                429
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        117,
+                                        416,
+                                        466,
+                                        429
+                                    ],
+                                    "score": 1.0,
+                                    "content": "future/story/20180416-the-myth-of-the-online-echo-chamber.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        }
+                    ],
+                    "index": 22.5,
+                    "bbox_fs": [
+                        105,
+                        404,
+                        505,
+                        429
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        434,
+                        505,
+                        457
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                434,
+                                506,
+                                448
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        434,
+                                        506,
+                                        448
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Peter Schulam and Suchi Saria. Reliable decision support using counterfactual models. In Neural",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                444,
+                                274,
+                                457
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        444,
+                                        274,
+                                        457
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Information Processing Systems, 2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        }
+                    ],
+                    "index": 24.5,
+                    "bbox_fs": [
+                        105,
+                        434,
+                        506,
+                        457
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        463,
+                        505,
+                        487
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                463,
+                                506,
+                                477
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        463,
+                                        506,
+                                        477
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Virag Shah, Jose Blanchet, and Ramesh Johari. Bandit learning with positive externalities. Neural",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                475,
+                                274,
+                                488
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        475,
+                                        274,
+                                        488
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Information Processing Systems, 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        }
+                    ],
+                    "index": 26.5,
+                    "bbox_fs": [
+                        106,
+                        463,
+                        506,
+                        488
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        492,
+                        506,
+                        527
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                492,
+                                506,
+                                506
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        492,
+                                        506,
+                                        506
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Chengcheng Shao, Giovanni Luca Ciampaglia, Onur Varol, Kai-Cheng Yang, Alessandro Flammini,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                504,
+                                507,
+                                516
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        504,
+                                        507,
+                                        516
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and Filippo Menczer. The spread of low-credibility content by social bots. Nature Communications,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                515,
+                                178,
+                                527
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        515,
+                                        178,
+                                        527
+                                    ],
+                                    "score": 1.0,
+                                    "content": "9(4787), 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        }
+                    ],
+                    "index": 29,
+                    "bbox_fs": [
+                        106,
+                        492,
+                        507,
+                        527
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        533,
+                        504,
+                        556
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                533,
+                                505,
+                                546
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        533,
+                                        505,
+                                        546
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Jasper Snoek, Hugo Larochelle, and Ryan P Adams. Practical Bayesian optimization of machine",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                544,
+                                399,
+                                556
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        544,
+                                        399,
+                                        556
+                                    ],
+                                    "score": 1.0,
+                                    "content": "learning algorithms. In Neural Information Processing Systems, 2012.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        }
+                    ],
+                    "index": 31.5,
+                    "bbox_fs": [
+                        106,
+                        533,
+                        505,
+                        556
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        562,
+                        502,
+                        575
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                561,
+                                502,
+                                576
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        561,
+                                        502,
+                                        576
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Richard S Sutton and Andrew G Barto. Introduction to Reinforcement Learning. MIT Press, 1998.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        }
+                    ],
+                    "index": 33,
+                    "bbox_fs": [
+                        105,
+                        561,
+                        502,
+                        576
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        581,
+                        504,
+                        604
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                580,
+                                505,
+                                595
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        580,
+                                        505,
+                                        595
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Richard S Sutton, Anna Koop, and David Silver. On the role of tracking in stationary environments.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                592,
+                                342,
+                                604
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        592,
+                                        342,
+                                        604
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In International conference on Machine learning, 2007.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        }
+                    ],
+                    "index": 34.5,
+                    "bbox_fs": [
+                        105,
+                        580,
+                        505,
+                        604
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        610,
+                        504,
+                        633
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                610,
+                                505,
+                                623
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        610,
+                                        505,
+                                        623
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Techopedia. Filter bubble, 2018. URL https://www.techopedia.com/definition/",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                621,
+                                235,
+                                633
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        621,
+                                        235,
+                                        633
+                                    ],
+                                    "score": 1.0,
+                                    "content": "28556/filter-bubble.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        }
+                    ],
+                    "index": 36.5,
+                    "bbox_fs": [
+                        106,
+                        610,
+                        505,
+                        633
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        639,
+                        504,
+                        663
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                640,
+                                506,
+                                652
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        640,
+                                        506,
+                                        652
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Chih-Chun Wang, Sanjeev R Kulkarni, and H Vincent Poor. Bandit problems with side observations.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                651,
+                                375,
+                                663
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        651,
+                                        375,
+                                        663
+                                    ],
+                                    "score": 1.0,
+                                    "content": "IEEE Transactions on Automatic Control, 50(3):338–355, 2005.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        }
+                    ],
+                    "index": 38.5,
+                    "bbox_fs": [
+                        106,
+                        640,
+                        506,
+                        663
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        669,
+                        505,
+                        703
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                668,
+                                506,
+                                682
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        668,
+                                        506,
+                                        682
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Wikipedia contributors. Confirmation bias — Wikipedia, the free encyclopedia, 2018. URL",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                680,
+                                484,
+                                693
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        680,
+                                        484,
+                                        693
+                                    ],
+                                    "score": 1.0,
+                                    "content": "https://en.wikipedia.org/w/index.php?title=Confirmation_bias&",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                691,
+                                361,
+                                704
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        691,
+                                        361,
+                                        704
+                                    ],
+                                    "score": 1.0,
+                                    "content": "oldid=875026726. [Online; accessed 20-January-2019].",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        }
+                    ],
+                    "index": 41,
+                    "bbox_fs": [
+                        105,
+                        668,
+                        506,
+                        704
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        709,
+                        504,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                709,
+                                505,
+                                722
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        709,
+                                        505,
+                                        722
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Ronald J. Williams. Simple statistical gradient-following algorithms for connectionist reinforcement",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                720,
+                                324,
+                                732
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        720,
+                                        324,
+                                        732
+                                    ],
+                                    "score": 1.0,
+                                    "content": "learning. In Machine Learning, pp. 229–256, 1992.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        }
+                    ],
+                    "index": 43.5,
+                    "bbox_fs": [
+                        106,
+                        709,
+                        505,
+                        732
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "title",
+                    "bbox": [
+                        273,
+                        82,
+                        338,
+                        93
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                271,
+                                81,
+                                340,
+                                96
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        271,
+                                        81,
+                                        340,
+                                        96
+                                    ],
+                                    "score": 1.0,
+                                    "content": "APPENDICES",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        }
+                    ],
+                    "index": 0
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        111,
+                        346,
+                        124
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                110,
+                                348,
+                                126
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        110,
+                                        348,
+                                        126
+                                    ],
+                                    "score": 1.0,
+                                    "content": "8 CONTENT RECOMMENDATION IN THE WILD",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        }
+                    ],
+                    "index": 1
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        139,
+                        505,
+                        217
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                140,
+                                505,
+                                152
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        140,
+                                        505,
+                                        152
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Filter bubbles, the spread of fake news, and other techno-social issues are widely reported to be",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                151,
+                                506,
+                                163
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        151,
+                                        506,
+                                        163
+                                    ],
+                                    "score": 1.0,
+                                    "content": "responsible for the rise of populism (Groshek & Koc-Michalska, 2017), increase in racism and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                162,
+                                505,
+                                174
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        162,
+                                        505,
+                                        174
+                                    ],
+                                    "score": 1.0,
+                                    "content": "prejudice against immigrants and refugees (Noble, 2018), increase in social isolation and suicide",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                173,
+                                505,
+                                184
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        173,
+                                        505,
+                                        184
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(Luxton et al., 2012), and, particularly with reference to the 2016 US elections, are decried as",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                183,
+                                506,
+                                195
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        183,
+                                        506,
+                                        195
+                                    ],
+                                    "score": 1.0,
+                                    "content": "threatening the foundations of democracy (El-Bermawy, 2016). Even in 2013, well before the 2016",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                195,
+                                505,
+                                207
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        195,
+                                        505,
+                                        207
+                                    ],
+                                    "score": 1.0,
+                                    "content": "American elections, a World Economic Forum report identified these problems as a global crisis",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                206,
+                                189,
+                                218
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        206,
+                                        189,
+                                        218
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(Lee Howell, 2013).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        }
+                    ],
+                    "index": 5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        222,
+                        504,
+                        244
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                221,
+                                506,
+                                235
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        221,
+                                        506,
+                                        235
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We focus on two related issues in which content recommendation algorithms play a role: fake news",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                233,
+                                180,
+                                245
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        233,
+                                        180,
+                                        245
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and filter bubbles.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        }
+                    ],
+                    "index": 9.5
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        264,
+                        182,
+                        275
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                263,
+                                183,
+                                277
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        263,
+                                        183,
+                                        277
+                                    ],
+                                    "score": 1.0,
+                                    "content": "8.1 FAKE NEWS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        }
+                    ],
+                    "index": 11
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        287,
+                        505,
+                        342
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                286,
+                                506,
+                                300
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        286,
+                                        506,
+                                        300
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Fake news (also called false news or junk news) is an extreme version of yellow journalism, pro-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                297,
+                                506,
+                                312
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        297,
+                                        506,
+                                        312
+                                    ],
+                                    "score": 1.0,
+                                    "content": "paganda, or clickbait, in which media that is ostensibly providing information focuses on being",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                309,
+                                507,
+                                322
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        309,
+                                        507,
+                                        322
+                                    ],
+                                    "score": 1.0,
+                                    "content": "eye-catching or appealing, at the expense of the quality of research and exposition of factual informa-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                320,
+                                506,
+                                333
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        320,
+                                        506,
+                                        333
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tion. Fake news is distinguished by being specifically and deliberately created to spread falsehoods or",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                331,
+                                385,
+                                344
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        331,
+                                        385,
+                                        344
+                                    ],
+                                    "score": 1.0,
+                                    "content": "misinformation (Merriam-Webster, 2017; Mihailidis & Viotty, 2017).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        }
+                    ],
+                    "index": 14
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        348,
+                        505,
+                        425
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                348,
+                                505,
+                                361
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        348,
+                                        505,
+                                        361
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Why does fake news spread? It may at first seem the solution is simply to educate people about the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                359,
+                                505,
+                                370
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        359,
+                                        505,
+                                        370
+                                    ],
+                                    "score": 1.0,
+                                    "content": "truth, but research tells us the problem is more multifaceted and insidious, due to a combination of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                369,
+                                506,
+                                383
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        369,
+                                        506,
+                                        383
+                                    ],
+                                    "score": 1.0,
+                                    "content": "related biases and cognitive effects including confirmation bias (people are more likely to believe",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                381,
+                                505,
+                                394
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        381,
+                                        505,
+                                        394
+                                    ],
+                                    "score": 1.0,
+                                    "content": "things that fit with their existing beliefs), priming (exposure to information unconsciously influences",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                391,
+                                505,
+                                406
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        391,
+                                        505,
+                                        406
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the processing of subsequent information, i.e. seeing something in a credible context makes things",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                402,
+                                505,
+                                416
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        402,
+                                        505,
+                                        416
+                                    ],
+                                    "score": 1.0,
+                                    "content": "seem more credible) and the illusory truth effect (i.e. people are more likely to believe something",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                413,
+                                238,
+                                427
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        413,
+                                        238,
+                                        427
+                                    ],
+                                    "score": 1.0,
+                                    "content": "simply if they are told it is true).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        }
+                    ],
+                    "index": 20
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        430,
+                        505,
+                        508
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                430,
+                                505,
+                                443
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        430,
+                                        505,
+                                        443
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Allcott & Gentzkow (2017) track about 150 fake news stories during the 2016 US election, and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                441,
+                                506,
+                                455
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        441,
+                                        506,
+                                        455
+                                    ],
+                                    "score": 1.0,
+                                    "content": "find the average American adult saw 1-2 fake news stories, just over half believed the story was",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                452,
+                                506,
+                                465
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        452,
+                                        506,
+                                        465
+                                    ],
+                                    "score": 1.0,
+                                    "content": "true, and likelihood of believing fake news increased with ideological segregation (polarization) of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                462,
+                                505,
+                                477
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        462,
+                                        505,
+                                        477
+                                    ],
+                                    "score": 1.0,
+                                    "content": "their social media. Shao et al. (2018) examine the role of social bots in spreading fake news by",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                474,
+                                505,
+                                487
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        474,
+                                        505,
+                                        487
+                                    ],
+                                    "score": 1.0,
+                                    "content": "analyzing 14 million Twitter messages. They find that bots are far more likely than humans to spread",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                486,
+                                505,
+                                498
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        486,
+                                        505,
+                                        498
+                                    ],
+                                    "score": 1.0,
+                                    "content": "misinformation, and that success of a fake news story (in terms of human retweets) was heavily",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                496,
+                                300,
+                                510
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        496,
+                                        300,
+                                        510
+                                    ],
+                                    "score": 1.0,
+                                    "content": "dependent on whether bots had shared the story.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        }
+                    ],
+                    "index": 27
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        513,
+                        505,
+                        623
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                514,
+                                505,
+                                525
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        514,
+                                        505,
+                                        525
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Pennycook et al. (2019) examine the role of the illusory truth effect in fake news. They find that",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                525,
+                                505,
+                                536
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        525,
+                                        505,
+                                        536
+                                    ],
+                                    "score": 1.0,
+                                    "content": "even a single exposure to a news story makes people more likely to believe that it is true, and repeat",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                536,
+                                505,
+                                547
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        536,
+                                        505,
+                                        547
+                                    ],
+                                    "score": 1.0,
+                                    "content": "viewings increase this likelihood. They find that this is not true for extremely implausible statements",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                546,
+                                506,
+                                559
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        546,
+                                        506,
+                                        559
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(e.g. “the world is a perfect cube”), but that “only a small degree of potential plausibility is sufficient",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                556,
+                                505,
+                                570
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        556,
+                                        505,
+                                        570
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for repetition to increase perceived accuracy” of the story. The situation is further complicated by",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                568,
+                                507,
+                                582
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        568,
+                                        507,
+                                        582
+                                    ],
+                                    "score": 1.0,
+                                    "content": "peoples’ inability to distinguish promoted content from real news - Amazeen & Wojdynski (2018)",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                578,
+                                506,
+                                592
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        578,
+                                        506,
+                                        592
+                                    ],
+                                    "score": 1.0,
+                                    "content": "find that fewer than 1/10 people were able to tell when content was an advertisement, even when it",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                591,
+                                505,
+                                602
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        591,
+                                        505,
+                                        602
+                                    ],
+                                    "score": 1.0,
+                                    "content": "was explicitly labelled as such. Similarly, Fazio et al. (2015) find that repeated exposure to incorrect",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                601,
+                                506,
+                                614
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        601,
+                                        506,
+                                        614
+                                    ],
+                                    "score": 1.0,
+                                    "content": "trivia make people more likely to believe it, even when they are later able to identify the trivia as",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                613,
+                                147,
+                                624
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        613,
+                                        147,
+                                        624
+                                    ],
+                                    "score": 1.0,
+                                    "content": "incorrect.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        }
+                    ],
+                    "index": 35.5
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        642,
+                        203,
+                        654
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                643,
+                                205,
+                                655
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        643,
+                                        205,
+                                        655
+                                    ],
+                                    "score": 1.0,
+                                    "content": "8.2 FILTER BUBBLES",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        }
+                    ],
+                    "index": 41
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        665,
+                        505,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                666,
+                                505,
+                                678
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        666,
+                                        505,
+                                        678
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Filter bubbles, a term coined and popularized by Pariser (2011) are created by positive or negative",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                676,
+                                506,
+                                690
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        676,
+                                        506,
+                                        690
+                                    ],
+                                    "score": 1.0,
+                                    "content": "feedback loops which encourage users or groups of users towards increasing within-group similarity,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                687,
+                                506,
+                                700
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        687,
+                                        506,
+                                        700
+                                    ],
+                                    "score": 1.0,
+                                    "content": "while driving up between-group dissimilarity. The curation of this echo chamber is called self-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                699,
+                                504,
+                                711
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        699,
+                                        504,
+                                        711
+                                    ],
+                                    "score": 1.0,
+                                    "content": "selection (people are more likely to look for or select things that fit their existing preferences), and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                710,
+                                505,
+                                721
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        710,
+                                        505,
+                                        721
+                                    ],
+                                    "score": 1.0,
+                                    "content": "favours what Techopedia (2018) calls intellectual isolation. In the context of social and political",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                721,
+                                443,
+                                733
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        721,
+                                        443,
+                                        733
+                                    ],
+                                    "score": 1.0,
+                                    "content": "opinions, this is often called the polarization effect (Wikipedia contributors, 2018).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 47
+                        }
+                    ],
+                    "index": 44.5
+                }
+            ],
+            "page_idx": 11,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        107,
+                        27,
+                        307,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                25,
+                                308,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        25,
+                                        308,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Under review as a conference paper at ICLR 2021",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        300,
+                        751,
+                        310,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                299,
+                                750,
+                                312,
+                                764
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        299,
+                                        750,
+                                        312,
+                                        764
+                                    ],
+                                    "score": 1.0,
+                                    "content": "",
+                                    "type": "text",
+                                    "height": 14,
+                                    "width": 13
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "title",
+                    "bbox": [
+                        273,
+                        82,
+                        338,
+                        93
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                271,
+                                81,
+                                340,
+                                96
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        271,
+                                        81,
+                                        340,
+                                        96
+                                    ],
+                                    "score": 1.0,
+                                    "content": "APPENDICES",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        }
+                    ],
+                    "index": 0
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        111,
+                        346,
+                        124
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                110,
+                                348,
+                                126
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        110,
+                                        348,
+                                        126
+                                    ],
+                                    "score": 1.0,
+                                    "content": "8 CONTENT RECOMMENDATION IN THE WILD",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        }
+                    ],
+                    "index": 1
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        139,
+                        505,
+                        217
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                140,
+                                505,
+                                152
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        140,
+                                        505,
+                                        152
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Filter bubbles, the spread of fake news, and other techno-social issues are widely reported to be",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                151,
+                                506,
+                                163
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        151,
+                                        506,
+                                        163
+                                    ],
+                                    "score": 1.0,
+                                    "content": "responsible for the rise of populism (Groshek & Koc-Michalska, 2017), increase in racism and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                162,
+                                505,
+                                174
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        162,
+                                        505,
+                                        174
+                                    ],
+                                    "score": 1.0,
+                                    "content": "prejudice against immigrants and refugees (Noble, 2018), increase in social isolation and suicide",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                173,
+                                505,
+                                184
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        173,
+                                        505,
+                                        184
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(Luxton et al., 2012), and, particularly with reference to the 2016 US elections, are decried as",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                183,
+                                506,
+                                195
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        183,
+                                        506,
+                                        195
+                                    ],
+                                    "score": 1.0,
+                                    "content": "threatening the foundations of democracy (El-Bermawy, 2016). Even in 2013, well before the 2016",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                195,
+                                505,
+                                207
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        195,
+                                        505,
+                                        207
+                                    ],
+                                    "score": 1.0,
+                                    "content": "American elections, a World Economic Forum report identified these problems as a global crisis",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                206,
+                                189,
+                                218
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        206,
+                                        189,
+                                        218
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(Lee Howell, 2013).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        }
+                    ],
+                    "index": 5,
+                    "bbox_fs": [
+                        105,
+                        140,
+                        506,
+                        218
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        222,
+                        504,
+                        244
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                221,
+                                506,
+                                235
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        221,
+                                        506,
+                                        235
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We focus on two related issues in which content recommendation algorithms play a role: fake news",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                233,
+                                180,
+                                245
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        233,
+                                        180,
+                                        245
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and filter bubbles.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        }
+                    ],
+                    "index": 9.5,
+                    "bbox_fs": [
+                        105,
+                        221,
+                        506,
+                        245
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        264,
+                        182,
+                        275
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                263,
+                                183,
+                                277
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        263,
+                                        183,
+                                        277
+                                    ],
+                                    "score": 1.0,
+                                    "content": "8.1 FAKE NEWS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        }
+                    ],
+                    "index": 11
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        287,
+                        505,
+                        342
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                286,
+                                506,
+                                300
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        286,
+                                        506,
+                                        300
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Fake news (also called false news or junk news) is an extreme version of yellow journalism, pro-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                297,
+                                506,
+                                312
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        297,
+                                        506,
+                                        312
+                                    ],
+                                    "score": 1.0,
+                                    "content": "paganda, or clickbait, in which media that is ostensibly providing information focuses on being",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                309,
+                                507,
+                                322
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        309,
+                                        507,
+                                        322
+                                    ],
+                                    "score": 1.0,
+                                    "content": "eye-catching or appealing, at the expense of the quality of research and exposition of factual informa-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                320,
+                                506,
+                                333
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        320,
+                                        506,
+                                        333
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tion. Fake news is distinguished by being specifically and deliberately created to spread falsehoods or",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                331,
+                                385,
+                                344
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        331,
+                                        385,
+                                        344
+                                    ],
+                                    "score": 1.0,
+                                    "content": "misinformation (Merriam-Webster, 2017; Mihailidis & Viotty, 2017).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        }
+                    ],
+                    "index": 14,
+                    "bbox_fs": [
+                        105,
+                        286,
+                        507,
+                        344
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        348,
+                        505,
+                        425
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                348,
+                                505,
+                                361
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        348,
+                                        505,
+                                        361
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Why does fake news spread? It may at first seem the solution is simply to educate people about the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                359,
+                                505,
+                                370
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        359,
+                                        505,
+                                        370
+                                    ],
+                                    "score": 1.0,
+                                    "content": "truth, but research tells us the problem is more multifaceted and insidious, due to a combination of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                369,
+                                506,
+                                383
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        369,
+                                        506,
+                                        383
+                                    ],
+                                    "score": 1.0,
+                                    "content": "related biases and cognitive effects including confirmation bias (people are more likely to believe",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                381,
+                                505,
+                                394
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        381,
+                                        505,
+                                        394
+                                    ],
+                                    "score": 1.0,
+                                    "content": "things that fit with their existing beliefs), priming (exposure to information unconsciously influences",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                391,
+                                505,
+                                406
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        391,
+                                        505,
+                                        406
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the processing of subsequent information, i.e. seeing something in a credible context makes things",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                402,
+                                505,
+                                416
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        402,
+                                        505,
+                                        416
+                                    ],
+                                    "score": 1.0,
+                                    "content": "seem more credible) and the illusory truth effect (i.e. people are more likely to believe something",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                413,
+                                238,
+                                427
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        413,
+                                        238,
+                                        427
+                                    ],
+                                    "score": 1.0,
+                                    "content": "simply if they are told it is true).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        }
+                    ],
+                    "index": 20,
+                    "bbox_fs": [
+                        105,
+                        348,
+                        506,
+                        427
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        430,
+                        505,
+                        508
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                430,
+                                505,
+                                443
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        430,
+                                        505,
+                                        443
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Allcott & Gentzkow (2017) track about 150 fake news stories during the 2016 US election, and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                441,
+                                506,
+                                455
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        441,
+                                        506,
+                                        455
+                                    ],
+                                    "score": 1.0,
+                                    "content": "find the average American adult saw 1-2 fake news stories, just over half believed the story was",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                452,
+                                506,
+                                465
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        452,
+                                        506,
+                                        465
+                                    ],
+                                    "score": 1.0,
+                                    "content": "true, and likelihood of believing fake news increased with ideological segregation (polarization) of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                462,
+                                505,
+                                477
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        462,
+                                        505,
+                                        477
+                                    ],
+                                    "score": 1.0,
+                                    "content": "their social media. Shao et al. (2018) examine the role of social bots in spreading fake news by",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                474,
+                                505,
+                                487
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        474,
+                                        505,
+                                        487
+                                    ],
+                                    "score": 1.0,
+                                    "content": "analyzing 14 million Twitter messages. They find that bots are far more likely than humans to spread",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                486,
+                                505,
+                                498
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        486,
+                                        505,
+                                        498
+                                    ],
+                                    "score": 1.0,
+                                    "content": "misinformation, and that success of a fake news story (in terms of human retweets) was heavily",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                496,
+                                300,
+                                510
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        496,
+                                        300,
+                                        510
+                                    ],
+                                    "score": 1.0,
+                                    "content": "dependent on whether bots had shared the story.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        }
+                    ],
+                    "index": 27,
+                    "bbox_fs": [
+                        105,
+                        430,
+                        506,
+                        510
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        513,
+                        505,
+                        623
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                514,
+                                505,
+                                525
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        514,
+                                        505,
+                                        525
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Pennycook et al. (2019) examine the role of the illusory truth effect in fake news. They find that",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                525,
+                                505,
+                                536
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        525,
+                                        505,
+                                        536
+                                    ],
+                                    "score": 1.0,
+                                    "content": "even a single exposure to a news story makes people more likely to believe that it is true, and repeat",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                536,
+                                505,
+                                547
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        536,
+                                        505,
+                                        547
+                                    ],
+                                    "score": 1.0,
+                                    "content": "viewings increase this likelihood. They find that this is not true for extremely implausible statements",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                546,
+                                506,
+                                559
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        546,
+                                        506,
+                                        559
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(e.g. “the world is a perfect cube”), but that “only a small degree of potential plausibility is sufficient",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                556,
+                                505,
+                                570
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        556,
+                                        505,
+                                        570
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for repetition to increase perceived accuracy” of the story. The situation is further complicated by",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                568,
+                                507,
+                                582
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        568,
+                                        507,
+                                        582
+                                    ],
+                                    "score": 1.0,
+                                    "content": "peoples’ inability to distinguish promoted content from real news - Amazeen & Wojdynski (2018)",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                578,
+                                506,
+                                592
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        578,
+                                        506,
+                                        592
+                                    ],
+                                    "score": 1.0,
+                                    "content": "find that fewer than 1/10 people were able to tell when content was an advertisement, even when it",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                591,
+                                505,
+                                602
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        591,
+                                        505,
+                                        602
+                                    ],
+                                    "score": 1.0,
+                                    "content": "was explicitly labelled as such. Similarly, Fazio et al. (2015) find that repeated exposure to incorrect",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                601,
+                                506,
+                                614
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        601,
+                                        506,
+                                        614
+                                    ],
+                                    "score": 1.0,
+                                    "content": "trivia make people more likely to believe it, even when they are later able to identify the trivia as",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                613,
+                                147,
+                                624
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        613,
+                                        147,
+                                        624
+                                    ],
+                                    "score": 1.0,
+                                    "content": "incorrect.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        }
+                    ],
+                    "index": 35.5,
+                    "bbox_fs": [
+                        105,
+                        514,
+                        507,
+                        624
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        642,
+                        203,
+                        654
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                643,
+                                205,
+                                655
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        643,
+                                        205,
+                                        655
+                                    ],
+                                    "score": 1.0,
+                                    "content": "8.2 FILTER BUBBLES",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        }
+                    ],
+                    "index": 41
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        665,
+                        505,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                666,
+                                505,
+                                678
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        666,
+                                        505,
+                                        678
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Filter bubbles, a term coined and popularized by Pariser (2011) are created by positive or negative",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                676,
+                                506,
+                                690
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        676,
+                                        506,
+                                        690
+                                    ],
+                                    "score": 1.0,
+                                    "content": "feedback loops which encourage users or groups of users towards increasing within-group similarity,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                687,
+                                506,
+                                700
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        687,
+                                        506,
+                                        700
+                                    ],
+                                    "score": 1.0,
+                                    "content": "while driving up between-group dissimilarity. The curation of this echo chamber is called self-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                699,
+                                504,
+                                711
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        699,
+                                        504,
+                                        711
+                                    ],
+                                    "score": 1.0,
+                                    "content": "selection (people are more likely to look for or select things that fit their existing preferences), and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                710,
+                                505,
+                                721
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        710,
+                                        505,
+                                        721
+                                    ],
+                                    "score": 1.0,
+                                    "content": "favours what Techopedia (2018) calls intellectual isolation. In the context of social and political",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                721,
+                                443,
+                                733
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        721,
+                                        443,
+                                        733
+                                    ],
+                                    "score": 1.0,
+                                    "content": "opinions, this is often called the polarization effect (Wikipedia contributors, 2018).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 47
+                        }
+                    ],
+                    "index": 44.5,
+                    "bbox_fs": [
+                        105,
+                        666,
+                        506,
+                        733
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        82,
+                        505,
+                        170
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                82,
+                                505,
+                                95
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        82,
+                                        505,
+                                        95
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Filter bubbles can be encouraged by algorithms in two main ways. The first is the most commonly",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                93,
+                                505,
+                                106
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        93,
+                                        505,
+                                        106
+                                    ],
+                                    "score": 1.0,
+                                    "content": "described: simply by showing content that is similar to what a user has already searched for, search",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                105,
+                                506,
+                                117
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        105,
+                                        506,
+                                        117
+                                    ],
+                                    "score": 1.0,
+                                    "content": "or recommender systems create a positive feedback loop of increasingly-similar content (Pariser,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                115,
+                                505,
+                                128
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        115,
+                                        505,
+                                        128
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2011; Kayhan, 2015). The second way is similar but opposite - if the predictions of an algorithm are",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                127,
+                                505,
+                                140
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        127,
+                                        505,
+                                        140
+                                    ],
+                                    "score": 1.0,
+                                    "content": "good for a certain group of people, but bad for others, the algorithm can do better on its metrics by",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                136,
+                                506,
+                                150
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        136,
+                                        506,
+                                        150
+                                    ],
+                                    "score": 1.0,
+                                    "content": "driving hard-to-predict users away. Then new users to the site will either be turned off entirely, or see",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                148,
+                                506,
+                                162
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        148,
+                                        506,
+                                        162
+                                    ],
+                                    "score": 1.0,
+                                    "content": "an artificially homogenous community of like-minded peers, a phenomena Shah et al. (2018) call",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                159,
+                                198,
+                                171
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        159,
+                                        198,
+                                        171
+                                    ],
+                                    "score": 1.0,
+                                    "content": "positive externalities.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        }
+                    ],
+                    "index": 3.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        176,
+                        505,
+                        242
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                175,
+                                505,
+                                189
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        175,
+                                        505,
+                                        189
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In a study of 50,000 US-based internet users, Flaxman & Goel (2015) find that two things increase",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                187,
+                                505,
+                                200
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        187,
+                                        505,
+                                        200
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with social media and search engine use: (1) exposure of an individual to opposing or different",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                198,
+                                505,
+                                210
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        198,
+                                        505,
+                                        210
+                                    ],
+                                    "score": 1.0,
+                                    "content": "viewpoints, and (2) mean ideological distance between users. Many studies cite the first result as",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                209,
+                                505,
+                                221
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        209,
+                                        505,
+                                        221
+                                    ],
+                                    "score": 1.0,
+                                    "content": "evidence of the benefits of internet and social media (Robson, 2018; Bakshy et al., 2015), but the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                220,
+                                505,
+                                233
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        220,
+                                        505,
+                                        233
+                                    ],
+                                    "score": 1.0,
+                                    "content": "correlation of exposure with ideological distances demonstrates that exposure is not enough, and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                231,
+                                242,
+                                244
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        231,
+                                        242,
+                                        244
+                                    ],
+                                    "score": 1.0,
+                                    "content": "might even be counterproductive.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        }
+                    ],
+                    "index": 10.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        248,
+                        505,
+                        325
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                248,
+                                505,
+                                260
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        248,
+                                        505,
+                                        260
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Facebook’s own study on filter bubbles results show that the impact of the news feed algorithm on",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                258,
+                                506,
+                                272
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        258,
+                                        506,
+                                        272
+                                    ],
+                                    "score": 1.0,
+                                    "content": "filter bubble “size” (a measure of homogeneity of posts relative to a baseline) is almost as large as",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                270,
+                                505,
+                                282
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        270,
+                                        505,
+                                        282
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the impact of friend group composition (Bakshy et al., 2015). Kayhan (2015) specifically study the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                282,
+                                505,
+                                293
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        282,
+                                        505,
+                                        293
+                                    ],
+                                    "score": 1.0,
+                                    "content": "role of search engines in confirmation bias, and find that search context and the similarity of results",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                292,
+                                505,
+                                303
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        292,
+                                        505,
+                                        303
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in search engine results both reinforce existing biases and increase the likelihood of future biased",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                303,
+                                505,
+                                315
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        303,
+                                        505,
+                                        315
+                                    ],
+                                    "score": 1.0,
+                                    "content": "searches. Nguyen et al. (2014) similarly study the effect of recommender systems on individual users’",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                315,
+                                437,
+                                326
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        315,
+                                        437,
+                                        326
+                                    ],
+                                    "score": 1.0,
+                                    "content": "content diversity, and find that the set of options recommended narrows over time.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        }
+                    ],
+                    "index": 17
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        330,
+                        505,
+                        386
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                330,
+                                506,
+                                342
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        330,
+                                        506,
+                                        342
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Filter bubbles create an ideal environment for the spread of fake news: they increase the likelihood of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                342,
+                                506,
+                                354
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        342,
+                                        506,
+                                        354
+                                    ],
+                                    "score": 1.0,
+                                    "content": "repeat viewings of similar content, and because of the illusory truth effect, that content is more likely",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                351,
+                                506,
+                                365
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        351,
+                                        506,
+                                        365
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to be believed and shared (Pennycook et al., 2019; DiFranzo & Gloria-Garcia, 2017; Pariser, 2011).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                362,
+                                506,
+                                377
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        362,
+                                        506,
+                                        377
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We are not claiming that HI-ADS are entirely or even mostly responsible for these problems, but we",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                373,
+                                336,
+                                388
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        373,
+                                        336,
+                                        388
+                                    ],
+                                    "score": 1.0,
+                                    "content": "do note that they can play a role that is worth addressing.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        }
+                    ],
+                    "index": 23
+                }
+            ],
+            "page_idx": 12,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        108,
+                        27,
+                        306,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                26,
+                                308,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        26,
+                                        308,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Under review as a conference paper at ICLR 2021",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        300,
+                        751,
+                        310,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                299,
+                                750,
+                                312,
+                                764
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        299,
+                                        750,
+                                        312,
+                                        764
+                                    ],
+                                    "score": 1.0,
+                                    "content": "",
+                                    "type": "text",
+                                    "height": 14,
+                                    "width": 13
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        82,
+                        505,
+                        170
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                82,
+                                505,
+                                95
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        82,
+                                        505,
+                                        95
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Filter bubbles can be encouraged by algorithms in two main ways. The first is the most commonly",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                93,
+                                505,
+                                106
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        93,
+                                        505,
+                                        106
+                                    ],
+                                    "score": 1.0,
+                                    "content": "described: simply by showing content that is similar to what a user has already searched for, search",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                105,
+                                506,
+                                117
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        105,
+                                        506,
+                                        117
+                                    ],
+                                    "score": 1.0,
+                                    "content": "or recommender systems create a positive feedback loop of increasingly-similar content (Pariser,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                115,
+                                505,
+                                128
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        115,
+                                        505,
+                                        128
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2011; Kayhan, 2015). The second way is similar but opposite - if the predictions of an algorithm are",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                127,
+                                505,
+                                140
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        127,
+                                        505,
+                                        140
+                                    ],
+                                    "score": 1.0,
+                                    "content": "good for a certain group of people, but bad for others, the algorithm can do better on its metrics by",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                136,
+                                506,
+                                150
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        136,
+                                        506,
+                                        150
+                                    ],
+                                    "score": 1.0,
+                                    "content": "driving hard-to-predict users away. Then new users to the site will either be turned off entirely, or see",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                148,
+                                506,
+                                162
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        148,
+                                        506,
+                                        162
+                                    ],
+                                    "score": 1.0,
+                                    "content": "an artificially homogenous community of like-minded peers, a phenomena Shah et al. (2018) call",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                159,
+                                198,
+                                171
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        159,
+                                        198,
+                                        171
+                                    ],
+                                    "score": 1.0,
+                                    "content": "positive externalities.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        }
+                    ],
+                    "index": 3.5,
+                    "bbox_fs": [
+                        105,
+                        82,
+                        506,
+                        171
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        176,
+                        505,
+                        242
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                175,
+                                505,
+                                189
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        175,
+                                        505,
+                                        189
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In a study of 50,000 US-based internet users, Flaxman & Goel (2015) find that two things increase",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                187,
+                                505,
+                                200
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        187,
+                                        505,
+                                        200
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with social media and search engine use: (1) exposure of an individual to opposing or different",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                198,
+                                505,
+                                210
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        198,
+                                        505,
+                                        210
+                                    ],
+                                    "score": 1.0,
+                                    "content": "viewpoints, and (2) mean ideological distance between users. Many studies cite the first result as",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                209,
+                                505,
+                                221
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        209,
+                                        505,
+                                        221
+                                    ],
+                                    "score": 1.0,
+                                    "content": "evidence of the benefits of internet and social media (Robson, 2018; Bakshy et al., 2015), but the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                220,
+                                505,
+                                233
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        220,
+                                        505,
+                                        233
+                                    ],
+                                    "score": 1.0,
+                                    "content": "correlation of exposure with ideological distances demonstrates that exposure is not enough, and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                231,
+                                242,
+                                244
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        231,
+                                        242,
+                                        244
+                                    ],
+                                    "score": 1.0,
+                                    "content": "might even be counterproductive.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        }
+                    ],
+                    "index": 10.5,
+                    "bbox_fs": [
+                        105,
+                        175,
+                        505,
+                        244
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        248,
+                        505,
+                        325
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                248,
+                                505,
+                                260
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        248,
+                                        505,
+                                        260
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Facebook’s own study on filter bubbles results show that the impact of the news feed algorithm on",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                258,
+                                506,
+                                272
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        258,
+                                        506,
+                                        272
+                                    ],
+                                    "score": 1.0,
+                                    "content": "filter bubble “size” (a measure of homogeneity of posts relative to a baseline) is almost as large as",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                270,
+                                505,
+                                282
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        270,
+                                        505,
+                                        282
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the impact of friend group composition (Bakshy et al., 2015). Kayhan (2015) specifically study the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                282,
+                                505,
+                                293
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        282,
+                                        505,
+                                        293
+                                    ],
+                                    "score": 1.0,
+                                    "content": "role of search engines in confirmation bias, and find that search context and the similarity of results",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                292,
+                                505,
+                                303
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        292,
+                                        505,
+                                        303
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in search engine results both reinforce existing biases and increase the likelihood of future biased",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                303,
+                                505,
+                                315
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        303,
+                                        505,
+                                        315
+                                    ],
+                                    "score": 1.0,
+                                    "content": "searches. Nguyen et al. (2014) similarly study the effect of recommender systems on individual users’",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                315,
+                                437,
+                                326
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        315,
+                                        437,
+                                        326
+                                    ],
+                                    "score": 1.0,
+                                    "content": "content diversity, and find that the set of options recommended narrows over time.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        }
+                    ],
+                    "index": 17,
+                    "bbox_fs": [
+                        105,
+                        248,
+                        506,
+                        326
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        330,
+                        505,
+                        386
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                330,
+                                506,
+                                342
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        330,
+                                        506,
+                                        342
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Filter bubbles create an ideal environment for the spread of fake news: they increase the likelihood of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                342,
+                                506,
+                                354
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        342,
+                                        506,
+                                        354
+                                    ],
+                                    "score": 1.0,
+                                    "content": "repeat viewings of similar content, and because of the illusory truth effect, that content is more likely",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                351,
+                                506,
+                                365
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        351,
+                                        506,
+                                        365
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to be believed and shared (Pennycook et al., 2019; DiFranzo & Gloria-Garcia, 2017; Pariser, 2011).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                362,
+                                506,
+                                377
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        362,
+                                        506,
+                                        377
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We are not claiming that HI-ADS are entirely or even mostly responsible for these problems, but we",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                373,
+                                336,
+                                388
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        373,
+                                        336,
+                                        388
+                                    ],
+                                    "score": 1.0,
+                                    "content": "do note that they can play a role that is worth addressing.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        }
+                    ],
+                    "index": 23,
+                    "bbox_fs": [
+                        105,
+                        330,
+                        506,
+                        388
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "image",
+                    "bbox": [
+                        142,
+                        115,
+                        468,
+                        617
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                242,
+                                105,
+                                370,
+                                116
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        242,
+                                        104,
+                                        370,
+                                        118
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                242,
+                                                104,
+                                                331,
+                                                118
+                                            ],
+                                            "score": 1.0,
+                                            "content": "incentive-compatible",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                331,
+                                                105,
+                                                366,
+                                                116
+                                            ],
+                                            "score": 0.84,
+                                            "content": "\\beta = 0 . 5",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                366,
+                                                104,
+                                                370,
+                                                118
+                                            ],
+                                            "score": 1.0,
+                                            "content": ")",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 0
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                142,
+                                115,
+                                468,
+                                617
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        142,
+                                        115,
+                                        468,
+                                        617
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                142,
+                                                115,
+                                                468,
+                                                617
+                                            ],
+                                            "score": 0.97,
+                                            "type": "image",
+                                            "image_path": "5c7a8c62362379a4ec118b2d7cacc1bfea380d1d1281d25198b5cb5100eaef60.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 2,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        142,
+                                        115,
+                                        468,
+                                        282.33333333333337
+                                    ],
+                                    "spans": [],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        142,
+                                        282.33333333333337,
+                                        468,
+                                        449.66666666666674
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                },
+                                {
+                                    "bbox": [
+                                        142,
+                                        449.66666666666674,
+                                        468,
+                                        617.0000000000001
+                                    ],
+                                    "spans": [],
+                                    "index": 3
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                106,
+                                629,
+                                506,
+                                707
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        629,
+                                        506,
+                                        642
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                629,
+                                                506,
+                                                642
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 8: Average level of non-myopic (i.e. cooperate) behavior learned by agents in the unit test",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 4
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        640,
+                                        505,
+                                        653
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                640,
+                                                339,
+                                                653
+                                            ],
+                                            "score": 1.0,
+                                            "content": "for HI-ADS. Despite making the inner loop fully myopic",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                339,
+                                                641,
+                                                367,
+                                                652
+                                            ],
+                                            "score": 0.87,
+                                            "content": "( \\gamma = 0 )",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                367,
+                                                640,
+                                                505,
+                                                653
+                                            ],
+                                            "score": 1.0,
+                                            "content": "), population-based training (PBT)",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 5
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        650,
+                                        506,
+                                        665
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                650,
+                                                506,
+                                                665
+                                            ],
+                                            "score": 1.0,
+                                            "content": "can cause HI-ADS, leading agents to choose the cooperate action (top row). context swapping",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 6
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        662,
+                                        505,
+                                        675
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                662,
+                                                505,
+                                                675
+                                            ],
+                                            "score": 1.0,
+                                            "content": "successfully prevents this (bottom row). Columns (from left to right) show results for populations",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 7
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        673,
+                                        506,
+                                        686
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                673,
+                                                419,
+                                                686
+                                            ],
+                                            "score": 1.0,
+                                            "content": "of 10, 100, and 1000 learners. In the legend, “interval” refers to the interval",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                419,
+                                                674,
+                                                434,
+                                                684
+                                            ],
+                                            "score": 0.7,
+                                            "content": "( T )",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                434,
+                                                673,
+                                                506,
+                                                686
+                                            ],
+                                            "score": 1.0,
+                                            "content": "of PBT (see Sec.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 8
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        683,
+                                        506,
+                                        698
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                683,
+                                                506,
+                                                698
+                                            ],
+                                            "score": 1.0,
+                                            "content": "2.2). Sufficiently large populations and short intervals are necessary for PBT to induce nonmyopic",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 9
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        694,
+                                        146,
+                                        708
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                694,
+                                                146,
+                                                708
+                                            ],
+                                            "score": 1.0,
+                                            "content": "behavior.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 10
+                                }
+                            ],
+                            "index": 7
+                        }
+                    ],
+                    "index": 2
+                }
+            ],
+            "page_idx": 13,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        300,
+                        751,
+                        310,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                299,
+                                750,
+                                312,
+                                764
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        299,
+                                        750,
+                                        312,
+                                        764
+                                    ],
+                                    "score": 1.0,
+                                    "content": "",
+                                    "type": "text",
+                                    "height": 14,
+                                    "width": 13
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        107,
+                        27,
+                        306,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                26,
+                                308,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        26,
+                                        308,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Under review as a conference paper at ICLR 2021",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "image",
+                    "bbox": [
+                        142,
+                        115,
+                        468,
+                        617
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                242,
+                                105,
+                                370,
+                                116
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        242,
+                                        104,
+                                        370,
+                                        118
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                242,
+                                                104,
+                                                331,
+                                                118
+                                            ],
+                                            "score": 1.0,
+                                            "content": "incentive-compatible",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                331,
+                                                105,
+                                                366,
+                                                116
+                                            ],
+                                            "score": 0.84,
+                                            "content": "\\beta = 0 . 5",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                366,
+                                                104,
+                                                370,
+                                                118
+                                            ],
+                                            "score": 1.0,
+                                            "content": ")",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 0
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                142,
+                                115,
+                                468,
+                                617
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        142,
+                                        115,
+                                        468,
+                                        617
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                142,
+                                                115,
+                                                468,
+                                                617
+                                            ],
+                                            "score": 0.97,
+                                            "type": "image",
+                                            "image_path": "5c7a8c62362379a4ec118b2d7cacc1bfea380d1d1281d25198b5cb5100eaef60.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 2,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        142,
+                                        115,
+                                        468,
+                                        282.33333333333337
+                                    ],
+                                    "spans": [],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        142,
+                                        282.33333333333337,
+                                        468,
+                                        449.66666666666674
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                },
+                                {
+                                    "bbox": [
+                                        142,
+                                        449.66666666666674,
+                                        468,
+                                        617.0000000000001
+                                    ],
+                                    "spans": [],
+                                    "index": 3
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                106,
+                                629,
+                                506,
+                                707
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        629,
+                                        506,
+                                        642
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                629,
+                                                506,
+                                                642
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 8: Average level of non-myopic (i.e. cooperate) behavior learned by agents in the unit test",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 4
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        640,
+                                        505,
+                                        653
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                640,
+                                                339,
+                                                653
+                                            ],
+                                            "score": 1.0,
+                                            "content": "for HI-ADS. Despite making the inner loop fully myopic",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                339,
+                                                641,
+                                                367,
+                                                652
+                                            ],
+                                            "score": 0.87,
+                                            "content": "( \\gamma = 0 )",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                367,
+                                                640,
+                                                505,
+                                                653
+                                            ],
+                                            "score": 1.0,
+                                            "content": "), population-based training (PBT)",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 5
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        650,
+                                        506,
+                                        665
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                650,
+                                                506,
+                                                665
+                                            ],
+                                            "score": 1.0,
+                                            "content": "can cause HI-ADS, leading agents to choose the cooperate action (top row). context swapping",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 6
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        662,
+                                        505,
+                                        675
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                662,
+                                                505,
+                                                675
+                                            ],
+                                            "score": 1.0,
+                                            "content": "successfully prevents this (bottom row). Columns (from left to right) show results for populations",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 7
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        673,
+                                        506,
+                                        686
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                673,
+                                                419,
+                                                686
+                                            ],
+                                            "score": 1.0,
+                                            "content": "of 10, 100, and 1000 learners. In the legend, “interval” refers to the interval",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                419,
+                                                674,
+                                                434,
+                                                684
+                                            ],
+                                            "score": 0.7,
+                                            "content": "( T )",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                434,
+                                                673,
+                                                506,
+                                                686
+                                            ],
+                                            "score": 1.0,
+                                            "content": "of PBT (see Sec.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 8
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        683,
+                                        506,
+                                        698
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                683,
+                                                506,
+                                                698
+                                            ],
+                                            "score": 1.0,
+                                            "content": "2.2). Sufficiently large populations and short intervals are necessary for PBT to induce nonmyopic",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 9
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        694,
+                                        146,
+                                        708
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                694,
+                                                146,
+                                                708
+                                            ],
+                                            "score": 1.0,
+                                            "content": "behavior.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 10
+                                }
+                            ],
+                            "index": 7
+                        }
+                    ],
+                    "index": 2
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "title",
+                    "bbox": [
+                        106,
+                        82,
+                        407,
+                        94
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                80,
+                                408,
+                                96
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        80,
+                                        408,
+                                        96
+                                    ],
+                                    "score": 1.0,
+                                    "content": "9 EXTRA EXPERIMENTS AND REPRODUCIBILITY DETAILS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        }
+                    ],
+                    "index": 0
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        106,
+                        216,
+                        118
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                106,
+                                217,
+                                119
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        106,
+                                        217,
+                                        119
+                                    ],
+                                    "score": 1.0,
+                                    "content": "9.1 HI-ADS UNIT TEST",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        }
+                    ],
+                    "index": 1
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        126,
+                        506,
+                        160
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                127,
+                                506,
+                                139
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        127,
+                                        360,
+                                        139
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We used REINFORCE (Williams, 1992) with discount factor",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        360,
+                                        127,
+                                        387,
+                                        138
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\gamma = 0",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        387,
+                                        127,
+                                        506,
+                                        139
+                                    ],
+                                    "score": 1.0,
+                                    "content": "as the baseline/IL optimizer.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                137,
+                                506,
+                                150
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        137,
+                                        506,
+                                        150
+                                    ],
+                                    "score": 1.0,
+                                    "content": "PBT (with default settings, see Section 2.2) is used to tune the learning rate, with reward on the final",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                149,
+                                355,
+                                160
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        149,
+                                        355,
+                                        160
+                                    ],
+                                    "score": 1.0,
+                                    "content": "time-step of the interval as the performance measure for PBT.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        }
+                    ],
+                    "index": 3
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        165,
+                        504,
+                        190
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                164,
+                                505,
+                                179
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        164,
+                                        250,
+                                        179
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Formally, the environment is not a",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        250,
+                                        166,
+                                        267,
+                                        176
+                                    ],
+                                    "score": 0.66,
+                                    "content": "2 \\mathbf { x } 2",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        268,
+                                        164,
+                                        505,
+                                        179
+                                    ],
+                                    "score": 1.0,
+                                    "content": "game (as the original prisoner’s dilemma); it’s a partially",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                177,
+                                415,
+                                191
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        177,
+                                        415,
+                                        191
+                                    ],
+                                    "score": 1.0,
+                                    "content": "observable Markov Decision Process (Åström, 1965; Kaelbling et al., 1998):",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        }
+                    ],
+                    "index": 5.5
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        122,
+                        193,
+                        405,
+                        241
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                122,
+                                193,
+                                405,
+                                241
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        122,
+                                        193,
+                                        405,
+                                        241
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\begin{array} { r l r } & { } & { s _ { t } , o _ { t } = a _ { t - 1 } , \\{ \\} } \\\\ & { } & { a _ { t } \\in \\{ \\mathsf { d e f e c t } , ~ \\mathsf { c o o p e r a t e } \\} } \\\\ & { } & { P ( s _ { t } , a _ { t } ) = \\delta ( a _ { t } ) } \\\\ & { } & { R ( s _ { t } , a _ { t } ) = I ( s _ { t } = \\mathsf { c o o p e r a t e } ) + \\beta ~ I ( a _ { t } = \\mathsf { c o o p e r a t e } ) - 1 / 2 } \\end{array}",
+                                    "type": "interline_equation",
+                                    "image_path": "909ce0b5c22c4e5607063381a463b92ae72bf2dc7c3edecf48a8704f92fad06f.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 8,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                122,
+                                193,
+                                405,
+                                209.0
+                            ],
+                            "spans": [],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                122,
+                                209.0,
+                                405,
+                                225.0
+                            ],
+                            "spans": [],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                122,
+                                225.0,
+                                405,
+                                241.0
+                            ],
+                            "spans": [],
+                            "index": 9
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        244,
+                        502,
+                        289
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                244,
+                                504,
+                                256
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        245,
+                                        132,
+                                        256
+                                    ],
+                                    "score": 1.0,
+                                    "content": "where",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        132,
+                                        245,
+                                        139,
+                                        254
+                                    ],
+                                    "score": 0.78,
+                                    "content": "I",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        139,
+                                        245,
+                                        250,
+                                        256
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is an indicator function, and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        251,
+                                        244,
+                                        294,
+                                        256
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\beta = - 1 / 2",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        294,
+                                        245,
+                                        504,
+                                        256
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is a parameter controlling the alignment of incentives.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                255,
+                                505,
+                                267
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        255,
+                                        234,
+                                        267
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The initial state is sampled as",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        235,
+                                        256,
+                                        307,
+                                        267
+                                    ],
+                                    "score": 0.45,
+                                    "content": "s _ { 0 } \\sim U ( { \\tt d e f e c t }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        308,
+                                        255,
+                                        505,
+                                        267
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", cooperate). Policies are represented by a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                267,
+                                504,
+                                279
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        267,
+                                        223,
+                                        279
+                                    ],
+                                    "score": 1.0,
+                                    "content": "single real-valued parameter",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        223,
+                                        267,
+                                        230,
+                                        276
+                                    ],
+                                    "score": 0.8,
+                                    "content": "\\theta",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        230,
+                                        267,
+                                        288,
+                                        279
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(initialized as",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        288,
+                                        267,
+                                        340,
+                                        279
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\theta \\sim \\mathcal { N } ( 0 , 1 ) \\big )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        340,
+                                        267,
+                                        504,
+                                        279
+                                    ],
+                                    "score": 1.0,
+                                    "content": ") passed through a sigmoid whose output",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                277,
+                                218,
+                                290
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        277,
+                                        149,
+                                        290
+                                    ],
+                                    "score": 1.0,
+                                    "content": "represents",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        150,
+                                        277,
+                                        212,
+                                        289
+                                    ],
+                                    "score": 0.9,
+                                    "content": "P ( a _ { t } = \\mathrm { \\bar { d e f e c t } } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        213,
+                                        277,
+                                        218,
+                                        290
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        }
+                    ],
+                    "index": 11.5
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        301,
+                        323,
+                        312
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                301,
+                                325,
+                                313
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        301,
+                                        325,
+                                        313
+                                    ],
+                                    "score": 1.0,
+                                    "content": "9.1.1 ALIGNMENT OF INCENTIVES EXPLORATION",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        }
+                    ],
+                    "index": 14
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        320,
+                        504,
+                        343
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                320,
+                                505,
+                                333
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        320,
+                                        315,
+                                        333
+                                    ],
+                                    "score": 1.0,
+                                    "content": "This section presents an exploration of the parameter",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        316,
+                                        321,
+                                        323,
+                                        331
+                                    ],
+                                    "score": 0.83,
+                                    "content": "\\beta",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        323,
+                                        320,
+                                        505,
+                                        333
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", which controls the alignment of incentives in",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                331,
+                                251,
+                                343
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        331,
+                                        251,
+                                        343
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the HI-ADS unit tests (see Table 2).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        }
+                    ],
+                    "index": 15.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        348,
+                        504,
+                        381
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                347,
+                                505,
+                                362
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        347,
+                                        505,
+                                        362
+                                    ],
+                                    "score": 1.0,
+                                    "content": "To clarify the interpretation of experiments, we distinguish between environments in which myopic",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                359,
+                                505,
+                                371
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        359,
+                                        505,
+                                        371
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(defect) vs. nonmyopic (cooperate) incentives are opposed, orthogonal, or compatible. Note that in",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                370,
+                                374,
+                                383
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        370,
+                                        374,
+                                        383
+                                    ],
+                                    "score": 1.0,
+                                    "content": "this unit test myopic behaviour (defection) is what we want to see.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        }
+                    ],
+                    "index": 18
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        115,
+                        390,
+                        506,
+                        465
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                115,
+                                390,
+                                506,
+                                404
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        390,
+                                        506,
+                                        404
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1. Incentive-opposed: Optimal myopic behavior is incompatible with optimal nonmyopic behav-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                402,
+                                421,
+                                414
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        402,
+                                        421,
+                                        414
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ior (classic prisoner’s dilemma; these experiments are in the main paper).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                113,
+                                415,
+                                506,
+                                431
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        113,
+                                        415,
+                                        506,
+                                        431
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2. Incentive-orthogonal: Optimal myopic behavior may or may not be optimal nonmyopic",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                428,
+                                167,
+                                440
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        428,
+                                        167,
+                                        440
+                                    ],
+                                    "score": 1.0,
+                                    "content": "behavior.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                113,
+                                442,
+                                506,
+                                456
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        113,
+                                        442,
+                                        506,
+                                        456
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3. Incentive-compatible: Optimal myopic behavior is necessarily also optimal nonmyopic behav-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                453,
+                                145,
+                                467
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        453,
+                                        145,
+                                        467
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ior.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        }
+                    ],
+                    "index": 22.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        474,
+                        505,
+                        541
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                475,
+                                506,
+                                488
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        475,
+                                        292,
+                                        488
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We focused on incentive-opposed environment",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        293,
+                                        475,
+                                        339,
+                                        487
+                                    ],
+                                    "score": 0.89,
+                                    "content": "( \\beta = - 1 / 2 )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        339,
+                                        475,
+                                        506,
+                                        488
+                                    ],
+                                    "score": 1.0,
+                                    "content": ") in the main paper in order to demonstrate",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                485,
+                                506,
+                                498
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        485,
+                                        506,
+                                        498
+                                    ],
+                                    "score": 1.0,
+                                    "content": "that HI-ADS can be powerful enough to change the behavior of the system in an undesirable way.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                496,
+                                505,
+                                510
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        496,
+                                        505,
+                                        510
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Here we also explore incentive-compatible and incentive-orthogonal environments because they",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                507,
+                                506,
+                                520
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        507,
+                                        506,
+                                        520
+                                    ],
+                                    "score": 1.0,
+                                    "content": "provide useful baselines, helping us distinguish a systematic bias towards nonmyopic behavior",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                519,
+                                506,
+                                531
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        519,
+                                        506,
+                                        531
+                                    ],
+                                    "score": 1.0,
+                                    "content": "from other reasons (such as randomness or optimization issues) for behavior that does not follow a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                530,
+                                216,
+                                543
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        530,
+                                        216,
+                                        543
+                                    ],
+                                    "score": 1.0,
+                                    "content": "myopically optimal policy.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        }
+                    ],
+                    "index": 28.5
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        553,
+                        421,
+                        565
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                553,
+                                421,
+                                565
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        553,
+                                        394,
+                                        565
+                                    ],
+                                    "score": 1.0,
+                                    "content": "9.1.2 WORKING THROUGH A DETAILED EXAMPLE FOR PBT WITH",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        394,
+                                        554,
+                                        421,
+                                        564
+                                    ],
+                                    "score": 0.66,
+                                    "content": "T = 1",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 32
+                        }
+                    ],
+                    "index": 32
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        572,
+                        506,
+                        650
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                571,
+                                506,
+                                585
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        571,
+                                        506,
+                                        585
+                                    ],
+                                    "score": 1.0,
+                                    "content": "To help provide intuition on how (mechanistically) PBT could lead to persistent levels of cooperation,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                583,
+                                506,
+                                596
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        583,
+                                        408,
+                                        596
+                                    ],
+                                    "score": 1.0,
+                                    "content": "we walk through a simple example (with no inner loop). Consider PBT with",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        409,
+                                        584,
+                                        436,
+                                        594
+                                    ],
+                                    "score": 0.93,
+                                    "content": "T = 1",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        436,
+                                        583,
+                                        506,
+                                        596
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and a population",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                594,
+                                506,
+                                608
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        594,
+                                        211,
+                                        608
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of 5 deterministic agents",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        211,
+                                        595,
+                                        253,
+                                        606
+                                    ],
+                                    "score": 0.94,
+                                    "content": "A _ { 1 } , . . . , A _ { 5 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        253,
+                                        594,
+                                        433,
+                                        608
+                                    ],
+                                    "score": 1.0,
+                                    "content": "playing cooperate and receiving reward of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        434,
+                                        595,
+                                        478,
+                                        606
+                                    ],
+                                    "score": 0.92,
+                                    "content": "r ( A _ { i } ) = 0",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        478,
+                                        594,
+                                        506,
+                                        608
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Now",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                605,
+                                506,
+                                618
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        605,
+                                        141,
+                                        618
+                                    ],
+                                    "score": 1.0,
+                                    "content": "suppose",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        142,
+                                        606,
+                                        155,
+                                        617
+                                    ],
+                                    "score": 0.88,
+                                    "content": "A _ { 1 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        155,
+                                        605,
+                                        318,
+                                        618
+                                    ],
+                                    "score": 1.0,
+                                    "content": "suddenly switches to play defect. Then",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        319,
+                                        605,
+                                        372,
+                                        618
+                                    ],
+                                    "score": 0.93,
+                                    "content": "r ( A _ { 1 } ) = 1 / 2",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        373,
+                                        605,
+                                        506,
+                                        618
+                                    ],
+                                    "score": 1.0,
+                                    "content": "on the next time-step (while the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                615,
+                                506,
+                                630
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        615,
+                                        385,
+                                        630
+                                    ],
+                                    "score": 1.0,
+                                    "content": "other agents’ reward is still 0), and so PBT’s EXPLOIT step will copy",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        385,
+                                        617,
+                                        398,
+                                        627
+                                    ],
+                                    "score": 0.88,
+                                    "content": "A _ { 1 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        398,
+                                        615,
+                                        506,
+                                        630
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(without loss of generality",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                627,
+                                507,
+                                641
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        627,
+                                        117,
+                                        641
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        117,
+                                        628,
+                                        130,
+                                        639
+                                    ],
+                                    "score": 0.87,
+                                    "content": "A _ { 2 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        130,
+                                        627,
+                                        250,
+                                        641
+                                    ],
+                                    "score": 1.0,
+                                    "content": "). On the following time-step,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        251,
+                                        627,
+                                        304,
+                                        640
+                                    ],
+                                    "score": 0.93,
+                                    "content": "r ( A _ { 2 } ) = 1 / 2",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        304,
+                                        627,
+                                        325,
+                                        641
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        325,
+                                        627,
+                                        386,
+                                        640
+                                    ],
+                                    "score": 0.92,
+                                    "content": "r ( A _ { 1 } ) = - 1 / 2",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        387,
+                                        627,
+                                        465,
+                                        641
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", so PBT will clone",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        465,
+                                        628,
+                                        478,
+                                        639
+                                    ],
+                                    "score": 0.89,
+                                    "content": "A _ { 2 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        478,
+                                        627,
+                                        489,
+                                        641
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        489,
+                                        628,
+                                        502,
+                                        639
+                                    ],
+                                    "score": 0.89,
+                                    "content": "A _ { 1 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        502,
+                                        627,
+                                        507,
+                                        641
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                639,
+                                438,
+                                651
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        639,
+                                        407,
+                                        651
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and the cycle repeats. Similar reasoning applies for larger populations, and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        407,
+                                        639,
+                                        433,
+                                        649
+                                    ],
+                                    "score": 0.9,
+                                    "content": "T > 1",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        434,
+                                        639,
+                                        438,
+                                        651
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        }
+                    ],
+                    "index": 36
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        690,
+                        293,
+                        702
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                690,
+                                294,
+                                703
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        690,
+                                        294,
+                                        703
+                                    ],
+                                    "score": 1.0,
+                                    "content": "9.1.3 Q-LEARNING EXPERIMENT DETAILS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        }
+                    ],
+                    "index": 40
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        709,
+                        504,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                709,
+                                505,
+                                722
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        709,
+                                        505,
+                                        722
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We show that, under certain conditions, Q-learning can learn to (primarily) cooperate, and thus fails",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                719,
+                                506,
+                                734
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        719,
+                                        506,
+                                        734
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the HI-ADS unit test. We estimate Q-values using the sample-average method, which is guaranteed to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        }
+                    ],
+                    "index": 41.5
+                }
+            ],
+            "page_idx": 14,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        300,
+                        751,
+                        310,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                299,
+                                750,
+                                312,
+                                764
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        299,
+                                        750,
+                                        312,
+                                        764
+                                    ],
+                                    "score": 1.0,
+                                    "content": "",
+                                    "type": "text",
+                                    "height": 14,
+                                    "width": 13
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        106,
+                        27,
+                        307,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                26,
+                                308,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        26,
+                                        308,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Under review as a conference paper at ICLR 2021",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "title",
+                    "bbox": [
+                        106,
+                        82,
+                        407,
+                        94
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                80,
+                                408,
+                                96
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        80,
+                                        408,
+                                        96
+                                    ],
+                                    "score": 1.0,
+                                    "content": "9 EXTRA EXPERIMENTS AND REPRODUCIBILITY DETAILS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        }
+                    ],
+                    "index": 0
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        106,
+                        216,
+                        118
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                106,
+                                217,
+                                119
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        106,
+                                        217,
+                                        119
+                                    ],
+                                    "score": 1.0,
+                                    "content": "9.1 HI-ADS UNIT TEST",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        }
+                    ],
+                    "index": 1
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        126,
+                        506,
+                        160
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                127,
+                                506,
+                                139
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        127,
+                                        360,
+                                        139
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We used REINFORCE (Williams, 1992) with discount factor",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        360,
+                                        127,
+                                        387,
+                                        138
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\gamma = 0",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        387,
+                                        127,
+                                        506,
+                                        139
+                                    ],
+                                    "score": 1.0,
+                                    "content": "as the baseline/IL optimizer.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                137,
+                                506,
+                                150
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        137,
+                                        506,
+                                        150
+                                    ],
+                                    "score": 1.0,
+                                    "content": "PBT (with default settings, see Section 2.2) is used to tune the learning rate, with reward on the final",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                149,
+                                355,
+                                160
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        149,
+                                        355,
+                                        160
+                                    ],
+                                    "score": 1.0,
+                                    "content": "time-step of the interval as the performance measure for PBT.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        }
+                    ],
+                    "index": 3,
+                    "bbox_fs": [
+                        106,
+                        127,
+                        506,
+                        160
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        165,
+                        504,
+                        190
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                164,
+                                505,
+                                179
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        164,
+                                        250,
+                                        179
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Formally, the environment is not a",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        250,
+                                        166,
+                                        267,
+                                        176
+                                    ],
+                                    "score": 0.66,
+                                    "content": "2 \\mathbf { x } 2",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        268,
+                                        164,
+                                        505,
+                                        179
+                                    ],
+                                    "score": 1.0,
+                                    "content": "game (as the original prisoner’s dilemma); it’s a partially",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                177,
+                                415,
+                                191
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        177,
+                                        415,
+                                        191
+                                    ],
+                                    "score": 1.0,
+                                    "content": "observable Markov Decision Process (Åström, 1965; Kaelbling et al., 1998):",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        }
+                    ],
+                    "index": 5.5,
+                    "bbox_fs": [
+                        105,
+                        164,
+                        505,
+                        191
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        122,
+                        193,
+                        405,
+                        241
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                122,
+                                193,
+                                405,
+                                241
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        122,
+                                        193,
+                                        405,
+                                        241
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\begin{array} { r l r } & { } & { s _ { t } , o _ { t } = a _ { t - 1 } , \\{ \\} } \\\\ & { } & { a _ { t } \\in \\{ \\mathsf { d e f e c t } , ~ \\mathsf { c o o p e r a t e } \\} } \\\\ & { } & { P ( s _ { t } , a _ { t } ) = \\delta ( a _ { t } ) } \\\\ & { } & { R ( s _ { t } , a _ { t } ) = I ( s _ { t } = \\mathsf { c o o p e r a t e } ) + \\beta ~ I ( a _ { t } = \\mathsf { c o o p e r a t e } ) - 1 / 2 } \\end{array}",
+                                    "type": "interline_equation",
+                                    "image_path": "909ce0b5c22c4e5607063381a463b92ae72bf2dc7c3edecf48a8704f92fad06f.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 8,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                122,
+                                193,
+                                405,
+                                209.0
+                            ],
+                            "spans": [],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                122,
+                                209.0,
+                                405,
+                                225.0
+                            ],
+                            "spans": [],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                122,
+                                225.0,
+                                405,
+                                241.0
+                            ],
+                            "spans": [],
+                            "index": 9
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        244,
+                        502,
+                        289
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                244,
+                                504,
+                                256
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        245,
+                                        132,
+                                        256
+                                    ],
+                                    "score": 1.0,
+                                    "content": "where",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        132,
+                                        245,
+                                        139,
+                                        254
+                                    ],
+                                    "score": 0.78,
+                                    "content": "I",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        139,
+                                        245,
+                                        250,
+                                        256
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is an indicator function, and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        251,
+                                        244,
+                                        294,
+                                        256
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\beta = - 1 / 2",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        294,
+                                        245,
+                                        504,
+                                        256
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is a parameter controlling the alignment of incentives.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                255,
+                                505,
+                                267
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        255,
+                                        234,
+                                        267
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The initial state is sampled as",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        235,
+                                        256,
+                                        307,
+                                        267
+                                    ],
+                                    "score": 0.45,
+                                    "content": "s _ { 0 } \\sim U ( { \\tt d e f e c t }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        308,
+                                        255,
+                                        505,
+                                        267
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", cooperate). Policies are represented by a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                267,
+                                504,
+                                279
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        267,
+                                        223,
+                                        279
+                                    ],
+                                    "score": 1.0,
+                                    "content": "single real-valued parameter",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        223,
+                                        267,
+                                        230,
+                                        276
+                                    ],
+                                    "score": 0.8,
+                                    "content": "\\theta",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        230,
+                                        267,
+                                        288,
+                                        279
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(initialized as",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        288,
+                                        267,
+                                        340,
+                                        279
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\theta \\sim \\mathcal { N } ( 0 , 1 ) \\big )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        340,
+                                        267,
+                                        504,
+                                        279
+                                    ],
+                                    "score": 1.0,
+                                    "content": ") passed through a sigmoid whose output",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                277,
+                                218,
+                                290
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        277,
+                                        149,
+                                        290
+                                    ],
+                                    "score": 1.0,
+                                    "content": "represents",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        150,
+                                        277,
+                                        212,
+                                        289
+                                    ],
+                                    "score": 0.9,
+                                    "content": "P ( a _ { t } = \\mathrm { \\bar { d e f e c t } } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        213,
+                                        277,
+                                        218,
+                                        290
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        }
+                    ],
+                    "index": 11.5,
+                    "bbox_fs": [
+                        105,
+                        244,
+                        505,
+                        290
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        301,
+                        323,
+                        312
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                301,
+                                325,
+                                313
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        301,
+                                        325,
+                                        313
+                                    ],
+                                    "score": 1.0,
+                                    "content": "9.1.1 ALIGNMENT OF INCENTIVES EXPLORATION",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        }
+                    ],
+                    "index": 14
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        320,
+                        504,
+                        343
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                320,
+                                505,
+                                333
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        320,
+                                        315,
+                                        333
+                                    ],
+                                    "score": 1.0,
+                                    "content": "This section presents an exploration of the parameter",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        316,
+                                        321,
+                                        323,
+                                        331
+                                    ],
+                                    "score": 0.83,
+                                    "content": "\\beta",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        323,
+                                        320,
+                                        505,
+                                        333
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", which controls the alignment of incentives in",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                331,
+                                251,
+                                343
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        331,
+                                        251,
+                                        343
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the HI-ADS unit tests (see Table 2).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        }
+                    ],
+                    "index": 15.5,
+                    "bbox_fs": [
+                        106,
+                        320,
+                        505,
+                        343
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        348,
+                        504,
+                        381
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                347,
+                                505,
+                                362
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        347,
+                                        505,
+                                        362
+                                    ],
+                                    "score": 1.0,
+                                    "content": "To clarify the interpretation of experiments, we distinguish between environments in which myopic",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                359,
+                                505,
+                                371
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        359,
+                                        505,
+                                        371
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(defect) vs. nonmyopic (cooperate) incentives are opposed, orthogonal, or compatible. Note that in",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                370,
+                                374,
+                                383
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        370,
+                                        374,
+                                        383
+                                    ],
+                                    "score": 1.0,
+                                    "content": "this unit test myopic behaviour (defection) is what we want to see.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        }
+                    ],
+                    "index": 18,
+                    "bbox_fs": [
+                        105,
+                        347,
+                        505,
+                        383
+                    ]
+                },
+                {
+                    "type": "list",
+                    "bbox": [
+                        115,
+                        390,
+                        506,
+                        465
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                115,
+                                390,
+                                506,
+                                404
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        390,
+                                        506,
+                                        404
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1. Incentive-opposed: Optimal myopic behavior is incompatible with optimal nonmyopic behav-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                402,
+                                421,
+                                414
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        402,
+                                        421,
+                                        414
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ior (classic prisoner’s dilemma; these experiments are in the main paper).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                113,
+                                415,
+                                506,
+                                431
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        113,
+                                        415,
+                                        506,
+                                        431
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2. Incentive-orthogonal: Optimal myopic behavior may or may not be optimal nonmyopic",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                428,
+                                167,
+                                440
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        428,
+                                        167,
+                                        440
+                                    ],
+                                    "score": 1.0,
+                                    "content": "behavior.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                113,
+                                442,
+                                506,
+                                456
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        113,
+                                        442,
+                                        506,
+                                        456
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3. Incentive-compatible: Optimal myopic behavior is necessarily also optimal nonmyopic behav-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                453,
+                                145,
+                                467
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        453,
+                                        145,
+                                        467
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ior.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25,
+                            "is_list_end_line": true
+                        }
+                    ],
+                    "index": 22.5,
+                    "bbox_fs": [
+                        113,
+                        390,
+                        506,
+                        467
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        474,
+                        505,
+                        541
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                475,
+                                506,
+                                488
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        475,
+                                        292,
+                                        488
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We focused on incentive-opposed environment",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        293,
+                                        475,
+                                        339,
+                                        487
+                                    ],
+                                    "score": 0.89,
+                                    "content": "( \\beta = - 1 / 2 )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        339,
+                                        475,
+                                        506,
+                                        488
+                                    ],
+                                    "score": 1.0,
+                                    "content": ") in the main paper in order to demonstrate",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                485,
+                                506,
+                                498
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        485,
+                                        506,
+                                        498
+                                    ],
+                                    "score": 1.0,
+                                    "content": "that HI-ADS can be powerful enough to change the behavior of the system in an undesirable way.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                496,
+                                505,
+                                510
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        496,
+                                        505,
+                                        510
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Here we also explore incentive-compatible and incentive-orthogonal environments because they",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                507,
+                                506,
+                                520
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        507,
+                                        506,
+                                        520
+                                    ],
+                                    "score": 1.0,
+                                    "content": "provide useful baselines, helping us distinguish a systematic bias towards nonmyopic behavior",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                519,
+                                506,
+                                531
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        519,
+                                        506,
+                                        531
+                                    ],
+                                    "score": 1.0,
+                                    "content": "from other reasons (such as randomness or optimization issues) for behavior that does not follow a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                530,
+                                216,
+                                543
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        530,
+                                        216,
+                                        543
+                                    ],
+                                    "score": 1.0,
+                                    "content": "myopically optimal policy.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        }
+                    ],
+                    "index": 28.5,
+                    "bbox_fs": [
+                        105,
+                        475,
+                        506,
+                        543
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        553,
+                        421,
+                        565
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                553,
+                                421,
+                                565
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        553,
+                                        394,
+                                        565
+                                    ],
+                                    "score": 1.0,
+                                    "content": "9.1.2 WORKING THROUGH A DETAILED EXAMPLE FOR PBT WITH",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        394,
+                                        554,
+                                        421,
+                                        564
+                                    ],
+                                    "score": 0.66,
+                                    "content": "T = 1",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 32
+                        }
+                    ],
+                    "index": 32
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        572,
+                        506,
+                        650
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                571,
+                                506,
+                                585
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        571,
+                                        506,
+                                        585
+                                    ],
+                                    "score": 1.0,
+                                    "content": "To help provide intuition on how (mechanistically) PBT could lead to persistent levels of cooperation,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                583,
+                                506,
+                                596
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        583,
+                                        408,
+                                        596
+                                    ],
+                                    "score": 1.0,
+                                    "content": "we walk through a simple example (with no inner loop). Consider PBT with",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        409,
+                                        584,
+                                        436,
+                                        594
+                                    ],
+                                    "score": 0.93,
+                                    "content": "T = 1",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        436,
+                                        583,
+                                        506,
+                                        596
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and a population",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                594,
+                                506,
+                                608
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        594,
+                                        211,
+                                        608
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of 5 deterministic agents",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        211,
+                                        595,
+                                        253,
+                                        606
+                                    ],
+                                    "score": 0.94,
+                                    "content": "A _ { 1 } , . . . , A _ { 5 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        253,
+                                        594,
+                                        433,
+                                        608
+                                    ],
+                                    "score": 1.0,
+                                    "content": "playing cooperate and receiving reward of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        434,
+                                        595,
+                                        478,
+                                        606
+                                    ],
+                                    "score": 0.92,
+                                    "content": "r ( A _ { i } ) = 0",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        478,
+                                        594,
+                                        506,
+                                        608
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Now",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                605,
+                                506,
+                                618
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        605,
+                                        141,
+                                        618
+                                    ],
+                                    "score": 1.0,
+                                    "content": "suppose",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        142,
+                                        606,
+                                        155,
+                                        617
+                                    ],
+                                    "score": 0.88,
+                                    "content": "A _ { 1 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        155,
+                                        605,
+                                        318,
+                                        618
+                                    ],
+                                    "score": 1.0,
+                                    "content": "suddenly switches to play defect. Then",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        319,
+                                        605,
+                                        372,
+                                        618
+                                    ],
+                                    "score": 0.93,
+                                    "content": "r ( A _ { 1 } ) = 1 / 2",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        373,
+                                        605,
+                                        506,
+                                        618
+                                    ],
+                                    "score": 1.0,
+                                    "content": "on the next time-step (while the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                615,
+                                506,
+                                630
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        615,
+                                        385,
+                                        630
+                                    ],
+                                    "score": 1.0,
+                                    "content": "other agents’ reward is still 0), and so PBT’s EXPLOIT step will copy",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        385,
+                                        617,
+                                        398,
+                                        627
+                                    ],
+                                    "score": 0.88,
+                                    "content": "A _ { 1 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        398,
+                                        615,
+                                        506,
+                                        630
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(without loss of generality",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                627,
+                                507,
+                                641
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        627,
+                                        117,
+                                        641
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        117,
+                                        628,
+                                        130,
+                                        639
+                                    ],
+                                    "score": 0.87,
+                                    "content": "A _ { 2 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        130,
+                                        627,
+                                        250,
+                                        641
+                                    ],
+                                    "score": 1.0,
+                                    "content": "). On the following time-step,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        251,
+                                        627,
+                                        304,
+                                        640
+                                    ],
+                                    "score": 0.93,
+                                    "content": "r ( A _ { 2 } ) = 1 / 2",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        304,
+                                        627,
+                                        325,
+                                        641
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        325,
+                                        627,
+                                        386,
+                                        640
+                                    ],
+                                    "score": 0.92,
+                                    "content": "r ( A _ { 1 } ) = - 1 / 2",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        387,
+                                        627,
+                                        465,
+                                        641
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", so PBT will clone",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        465,
+                                        628,
+                                        478,
+                                        639
+                                    ],
+                                    "score": 0.89,
+                                    "content": "A _ { 2 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        478,
+                                        627,
+                                        489,
+                                        641
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        489,
+                                        628,
+                                        502,
+                                        639
+                                    ],
+                                    "score": 0.89,
+                                    "content": "A _ { 1 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        502,
+                                        627,
+                                        507,
+                                        641
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                639,
+                                438,
+                                651
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        639,
+                                        407,
+                                        651
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and the cycle repeats. Similar reasoning applies for larger populations, and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        407,
+                                        639,
+                                        433,
+                                        649
+                                    ],
+                                    "score": 0.9,
+                                    "content": "T > 1",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        434,
+                                        639,
+                                        438,
+                                        651
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        }
+                    ],
+                    "index": 36,
+                    "bbox_fs": [
+                        104,
+                        571,
+                        507,
+                        651
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        690,
+                        293,
+                        702
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                690,
+                                294,
+                                703
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        690,
+                                        294,
+                                        703
+                                    ],
+                                    "score": 1.0,
+                                    "content": "9.1.3 Q-LEARNING EXPERIMENT DETAILS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        }
+                    ],
+                    "index": 40
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        709,
+                        504,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                709,
+                                505,
+                                722
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        709,
+                                        505,
+                                        722
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We show that, under certain conditions, Q-learning can learn to (primarily) cooperate, and thus fails",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                719,
+                                506,
+                                734
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        719,
+                                        506,
+                                        734
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the HI-ADS unit test. We estimate Q-values using the sample-average method, which is guaranteed to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                169,
+                                505,
+                                183
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        169,
+                                        468,
+                                        183
+                                    ],
+                                    "score": 1.0,
+                                    "content": "converge in the fully observed, tabular case (Sutton & Barto, 1998). The agent follows the",
+                                    "type": "text",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        469,
+                                        172,
+                                        474,
+                                        180
+                                    ],
+                                    "score": 0.67,
+                                    "content": "\\epsilon",
+                                    "type": "inline_equation",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        474,
+                                        169,
+                                        505,
+                                        183
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-greedy",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                180,
+                                506,
+                                195
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        180,
+                                        156,
+                                        195
+                                    ],
+                                    "score": 1.0,
+                                    "content": "policy with",
+                                    "type": "text",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        156,
+                                        181,
+                                        190,
+                                        191
+                                    ],
+                                    "score": 0.88,
+                                    "content": "\\epsilon = 0 . 1",
+                                    "type": "inline_equation",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        190,
+                                        180,
+                                        506,
+                                        195
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". In order to achieve this result, we additionally start the agent off with one",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                192,
+                                506,
+                                205
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        192,
+                                        396,
+                                        205
+                                    ],
+                                    "score": 1.0,
+                                    "content": "synthetic memory where both state and action are defect and therefor",
+                                    "type": "text",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        396,
+                                        192,
+                                        469,
+                                        204
+                                    ],
+                                    "score": 0.9,
+                                    "content": "R ( \\mathrm { d e f e c t } ) = - . 5",
+                                    "type": "inline_equation",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        469,
+                                        192,
+                                        506,
+                                        205
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", and we",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                202,
+                                505,
+                                215
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        202,
+                                        400,
+                                        215
+                                    ],
+                                    "score": 1.0,
+                                    "content": "hard-code the starting state to be cooperate (which normally only happens",
+                                    "type": "text",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        400,
+                                        203,
+                                        419,
+                                        213
+                                    ],
+                                    "score": 0.83,
+                                    "content": "50 \\%",
+                                    "type": "inline_equation",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        420,
+                                        202,
+                                        505,
+                                        215
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of the time). Without",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                213,
+                                506,
+                                227
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        213,
+                                        506,
+                                        227
+                                    ],
+                                    "score": 1.0,
+                                    "content": "this kind of an initialization, the agent always learns to defect. However, under these conditions, we",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                224,
+                                504,
+                                237
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        224,
+                                        385,
+                                        237
+                                    ],
+                                    "score": 1.0,
+                                    "content": "find that 10/30 agents learned to play cooperate most of the time, with",
+                                    "type": "text",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        385,
+                                        225,
+                                        394,
+                                        236
+                                    ],
+                                    "score": 0.69,
+                                    "content": "Q",
+                                    "type": "inline_equation",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        395,
+                                        224,
+                                        462,
+                                        237
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(cooperate) and",
+                                    "type": "text",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        462,
+                                        225,
+                                        504,
+                                        236
+                                    ],
+                                    "score": 0.8,
+                                    "content": "Q ( { \\mathrm { d e f e c t } } )",
+                                    "type": "inline_equation",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                235,
+                                505,
+                                248
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        235,
+                                        195,
+                                        248
+                                    ],
+                                    "score": 1.0,
+                                    "content": "both hovering around",
+                                    "type": "text",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        195,
+                                        236,
+                                        221,
+                                        246
+                                    ],
+                                    "score": 0.67,
+                                    "content": "- 0 . 0 7",
+                                    "type": "inline_equation",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        222,
+                                        235,
+                                        389,
+                                        248
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", while others learn to always defect, with",
+                                    "type": "text",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        389,
+                                        236,
+                                        487,
+                                        248
+                                    ],
+                                    "score": 0.87,
+                                    "content": "Q ( { \\mathrm { c o o p e r a t e } } ) \\approx - 0 . 9 2",
+                                    "type": "inline_equation",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        487,
+                                        235,
+                                        505,
+                                        248
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                107,
+                                245,
+                                506,
+                                261
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        246,
+                                        188,
+                                        258
+                                    ],
+                                    "score": 0.88,
+                                    "content": "Q ( { \\mathrm { d e f e c t } } ) \\approx - 0 . 4 5",
+                                    "type": "inline_equation",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        189,
+                                        245,
+                                        506,
+                                        261
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". context swapping, however, prevents majority-cooperate behavior from ever",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                258,
+                                208,
+                                270
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        258,
+                                        208,
+                                        270
+                                    ],
+                                    "score": 1.0,
+                                    "content": "emerging, see Figure 11.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 12
+                        }
+                    ],
+                    "index": 41.5,
+                    "bbox_fs": [
+                        105,
+                        709,
+                        506,
+                        734
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "table",
+                    "bbox": [
+                        108,
+                        101,
+                        510,
+                        150
+                    ],
+                    "blocks": [
+                        {
+                            "type": "table_caption",
+                            "bbox": [
+                                127,
+                                89,
+                                476,
+                                101
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        129,
+                                        88,
+                                        478,
+                                        102
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                129,
+                                                88,
+                                                168,
+                                                102
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Table 2:",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                168,
+                                                90,
+                                                176,
+                                                101
+                                            ],
+                                            "score": 0.77,
+                                            "content": "\\beta",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                177,
+                                                88,
+                                                478,
+                                                102
+                                            ],
+                                            "score": 1.0,
+                                            "content": "controls the extent to which myopic and nonmyopic incentives are aligned.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 0
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "type": "table_body",
+                            "bbox": [
+                                108,
+                                101,
+                                510,
+                                150
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        108,
+                                        101,
+                                        510,
+                                        150
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                108,
+                                                101,
+                                                510,
+                                                150
+                                            ],
+                                            "score": 0.965,
+                                            "html": "<table><tr><td>β</td><td>Environment</td><td>Cooperating</td></tr><tr><td>&lt;0</td><td>incentive-opposed</td><td>yields less reward on the current time-step (myopically detrimental)</td></tr><tr><td>=0</td><td>incentive-orthogonal</td><td>does not affect the current reward (myopically indifferent)</td></tr><tr><td>v0</td><td>incentive-compatible</td><td>yields more reward on the current time-step (myopically beneficial)</td></tr></table>",
+                                            "type": "table",
+                                            "image_path": "1175f254f167db3c316ece81017afe9fac641b6f9b455be985267608b1241e01.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 2,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        108,
+                                        101,
+                                        510,
+                                        117.33333333333333
+                                    ],
+                                    "spans": [],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        108,
+                                        117.33333333333333,
+                                        510,
+                                        133.66666666666666
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                },
+                                {
+                                    "bbox": [
+                                        108,
+                                        133.66666666666666,
+                                        510,
+                                        150.0
+                                    ],
+                                    "spans": [],
+                                    "index": 3
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 1.0
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        169,
+                        506,
+                        269
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                169,
+                                505,
+                                183
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        169,
+                                        468,
+                                        183
+                                    ],
+                                    "score": 1.0,
+                                    "content": "converge in the fully observed, tabular case (Sutton & Barto, 1998). The agent follows the",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        469,
+                                        172,
+                                        474,
+                                        180
+                                    ],
+                                    "score": 0.67,
+                                    "content": "\\epsilon",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        474,
+                                        169,
+                                        505,
+                                        183
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-greedy",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                180,
+                                506,
+                                195
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        180,
+                                        156,
+                                        195
+                                    ],
+                                    "score": 1.0,
+                                    "content": "policy with",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        156,
+                                        181,
+                                        190,
+                                        191
+                                    ],
+                                    "score": 0.88,
+                                    "content": "\\epsilon = 0 . 1",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        190,
+                                        180,
+                                        506,
+                                        195
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". In order to achieve this result, we additionally start the agent off with one",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                192,
+                                506,
+                                205
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        192,
+                                        396,
+                                        205
+                                    ],
+                                    "score": 1.0,
+                                    "content": "synthetic memory where both state and action are defect and therefor",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        396,
+                                        192,
+                                        469,
+                                        204
+                                    ],
+                                    "score": 0.9,
+                                    "content": "R ( \\mathrm { d e f e c t } ) = - . 5",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        469,
+                                        192,
+                                        506,
+                                        205
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", and we",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                202,
+                                505,
+                                215
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        202,
+                                        400,
+                                        215
+                                    ],
+                                    "score": 1.0,
+                                    "content": "hard-code the starting state to be cooperate (which normally only happens",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        400,
+                                        203,
+                                        419,
+                                        213
+                                    ],
+                                    "score": 0.83,
+                                    "content": "50 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        420,
+                                        202,
+                                        505,
+                                        215
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of the time). Without",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                213,
+                                506,
+                                227
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        213,
+                                        506,
+                                        227
+                                    ],
+                                    "score": 1.0,
+                                    "content": "this kind of an initialization, the agent always learns to defect. However, under these conditions, we",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                224,
+                                504,
+                                237
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        224,
+                                        385,
+                                        237
+                                    ],
+                                    "score": 1.0,
+                                    "content": "find that 10/30 agents learned to play cooperate most of the time, with",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        385,
+                                        225,
+                                        394,
+                                        236
+                                    ],
+                                    "score": 0.69,
+                                    "content": "Q",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        395,
+                                        224,
+                                        462,
+                                        237
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(cooperate) and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        462,
+                                        225,
+                                        504,
+                                        236
+                                    ],
+                                    "score": 0.8,
+                                    "content": "Q ( { \\mathrm { d e f e c t } } )",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                235,
+                                505,
+                                248
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        235,
+                                        195,
+                                        248
+                                    ],
+                                    "score": 1.0,
+                                    "content": "both hovering around",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        195,
+                                        236,
+                                        221,
+                                        246
+                                    ],
+                                    "score": 0.67,
+                                    "content": "- 0 . 0 7",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        222,
+                                        235,
+                                        389,
+                                        248
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", while others learn to always defect, with",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        389,
+                                        236,
+                                        487,
+                                        248
+                                    ],
+                                    "score": 0.87,
+                                    "content": "Q ( { \\mathrm { c o o p e r a t e } } ) \\approx - 0 . 9 2",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        487,
+                                        235,
+                                        505,
+                                        248
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                107,
+                                245,
+                                506,
+                                261
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        246,
+                                        188,
+                                        258
+                                    ],
+                                    "score": 0.88,
+                                    "content": "Q ( { \\mathrm { d e f e c t } } ) \\approx - 0 . 4 5",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        189,
+                                        245,
+                                        506,
+                                        261
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". context swapping, however, prevents majority-cooperate behavior from ever",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                258,
+                                208,
+                                270
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        258,
+                                        208,
+                                        270
+                                    ],
+                                    "score": 1.0,
+                                    "content": "emerging, see Figure 11.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        }
+                    ],
+                    "index": 8
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        281,
+                        282,
+                        293
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                280,
+                                283,
+                                294
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        280,
+                                        283,
+                                        294
+                                    ],
+                                    "score": 1.0,
+                                    "content": "9.1.4 Q-LEARNING: FURTHER RESULTS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        }
+                    ],
+                    "index": 13
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        300,
+                        505,
+                        345
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                300,
+                                506,
+                                312
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        300,
+                                        506,
+                                        312
+                                    ],
+                                    "score": 1.0,
+                                    "content": "To give a more representative picture of how often Q-learning fails the unit test, we run a larger set of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                311,
+                                506,
+                                324
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        311,
+                                        506,
+                                        324
+                                    ],
+                                    "score": 1.0,
+                                    "content": "experiments with Q-learning, results are in Figure 10. It’s possible that the failure of Q-learning is",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                323,
+                                505,
+                                334
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        323,
+                                        505,
+                                        334
+                                    ],
+                                    "score": 1.0,
+                                    "content": "not persistent, since we have not proved otherwise, but we did run much longer experiments and still",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                334,
+                                265,
+                                345
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        334,
+                                        265,
+                                        345
+                                    ],
+                                    "score": 1.0,
+                                    "content": "observe persistent failure, see Figure 9.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        }
+                    ],
+                    "index": 15.5
+                },
+                {
+                    "type": "image",
+                    "bbox": [
+                        160,
+                        354,
+                        454,
+                        468
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                160,
+                                354,
+                                454,
+                                468
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        160,
+                                        354,
+                                        454,
+                                        468
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                160,
+                                                354,
+                                                454,
+                                                468
+                                            ],
+                                            "score": 0.971,
+                                            "type": "image",
+                                            "image_path": "93bc9858be4e122e4aa54709b97526f0b712f8b3de8f38002ec30b69c8b8d7ee.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 19,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        160,
+                                        354,
+                                        454,
+                                        392.0
+                                    ],
+                                    "spans": [],
+                                    "index": 18
+                                },
+                                {
+                                    "bbox": [
+                                        160,
+                                        392.0,
+                                        454,
+                                        430.0
+                                    ],
+                                    "spans": [],
+                                    "index": 19
+                                },
+                                {
+                                    "bbox": [
+                                        160,
+                                        430.0,
+                                        454,
+                                        468.0
+                                    ],
+                                    "spans": [],
+                                    "index": 20
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                105,
+                                481,
+                                505,
+                                504
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        481,
+                                        505,
+                                        494
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                481,
+                                                505,
+                                                494
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 9: The same experiments as Figures 6, 10, run for 50,000 time-steps instead of 3000, to",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 21
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        492,
+                                        306,
+                                        506
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                492,
+                                                306,
+                                                506
+                                            ],
+                                            "score": 1.0,
+                                            "content": "illustrate the persistence of non-myopic behavior.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 22
+                                }
+                            ],
+                            "index": 21.5
+                        }
+                    ],
+                    "index": 20.25
+                }
+            ],
+            "page_idx": 15,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        106,
+                        27,
+                        307,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                107,
+                                26,
+                                308,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        26,
+                                        308,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Under review as a conference paper at ICLR 2021",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        300,
+                        751,
+                        311,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                299,
+                                750,
+                                312,
+                                764
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        299,
+                                        750,
+                                        312,
+                                        764
+                                    ],
+                                    "score": 1.0,
+                                    "content": "16",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "table",
+                    "bbox": [
+                        108,
+                        101,
+                        510,
+                        150
+                    ],
+                    "blocks": [
+                        {
+                            "type": "table_caption",
+                            "bbox": [
+                                127,
+                                89,
+                                476,
+                                101
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        129,
+                                        88,
+                                        478,
+                                        102
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                129,
+                                                88,
+                                                168,
+                                                102
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Table 2:",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                168,
+                                                90,
+                                                176,
+                                                101
+                                            ],
+                                            "score": 0.77,
+                                            "content": "\\beta",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                177,
+                                                88,
+                                                478,
+                                                102
+                                            ],
+                                            "score": 1.0,
+                                            "content": "controls the extent to which myopic and nonmyopic incentives are aligned.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 0
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "type": "table_body",
+                            "bbox": [
+                                108,
+                                101,
+                                510,
+                                150
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        108,
+                                        101,
+                                        510,
+                                        150
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                108,
+                                                101,
+                                                510,
+                                                150
+                                            ],
+                                            "score": 0.965,
+                                            "html": "<table><tr><td>β</td><td>Environment</td><td>Cooperating</td></tr><tr><td>&lt;0</td><td>incentive-opposed</td><td>yields less reward on the current time-step (myopically detrimental)</td></tr><tr><td>=0</td><td>incentive-orthogonal</td><td>does not affect the current reward (myopically indifferent)</td></tr><tr><td>v0</td><td>incentive-compatible</td><td>yields more reward on the current time-step (myopically beneficial)</td></tr></table>",
+                                            "type": "table",
+                                            "image_path": "1175f254f167db3c316ece81017afe9fac641b6f9b455be985267608b1241e01.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 2,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        108,
+                                        101,
+                                        510,
+                                        117.33333333333333
+                                    ],
+                                    "spans": [],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        108,
+                                        117.33333333333333,
+                                        510,
+                                        133.66666666666666
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                },
+                                {
+                                    "bbox": [
+                                        108,
+                                        133.66666666666666,
+                                        510,
+                                        150.0
+                                    ],
+                                    "spans": [],
+                                    "index": 3
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 1.0
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        169,
+                        506,
+                        269
+                    ],
+                    "lines": [],
+                    "index": 8,
+                    "bbox_fs": [
+                        105,
+                        169,
+                        506,
+                        270
+                    ],
+                    "lines_deleted": true
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        281,
+                        282,
+                        293
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                280,
+                                283,
+                                294
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        280,
+                                        283,
+                                        294
+                                    ],
+                                    "score": 1.0,
+                                    "content": "9.1.4 Q-LEARNING: FURTHER RESULTS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        }
+                    ],
+                    "index": 13
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        300,
+                        505,
+                        345
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                300,
+                                506,
+                                312
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        300,
+                                        506,
+                                        312
+                                    ],
+                                    "score": 1.0,
+                                    "content": "To give a more representative picture of how often Q-learning fails the unit test, we run a larger set of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                311,
+                                506,
+                                324
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        311,
+                                        506,
+                                        324
+                                    ],
+                                    "score": 1.0,
+                                    "content": "experiments with Q-learning, results are in Figure 10. It’s possible that the failure of Q-learning is",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                323,
+                                505,
+                                334
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        323,
+                                        505,
+                                        334
+                                    ],
+                                    "score": 1.0,
+                                    "content": "not persistent, since we have not proved otherwise, but we did run much longer experiments and still",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                334,
+                                265,
+                                345
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        334,
+                                        265,
+                                        345
+                                    ],
+                                    "score": 1.0,
+                                    "content": "observe persistent failure, see Figure 9.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        }
+                    ],
+                    "index": 15.5,
+                    "bbox_fs": [
+                        105,
+                        300,
+                        506,
+                        345
+                    ]
+                },
+                {
+                    "type": "image",
+                    "bbox": [
+                        160,
+                        354,
+                        454,
+                        468
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                160,
+                                354,
+                                454,
+                                468
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        160,
+                                        354,
+                                        454,
+                                        468
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                160,
+                                                354,
+                                                454,
+                                                468
+                                            ],
+                                            "score": 0.971,
+                                            "type": "image",
+                                            "image_path": "93bc9858be4e122e4aa54709b97526f0b712f8b3de8f38002ec30b69c8b8d7ee.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 19,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        160,
+                                        354,
+                                        454,
+                                        392.0
+                                    ],
+                                    "spans": [],
+                                    "index": 18
+                                },
+                                {
+                                    "bbox": [
+                                        160,
+                                        392.0,
+                                        454,
+                                        430.0
+                                    ],
+                                    "spans": [],
+                                    "index": 19
+                                },
+                                {
+                                    "bbox": [
+                                        160,
+                                        430.0,
+                                        454,
+                                        468.0
+                                    ],
+                                    "spans": [],
+                                    "index": 20
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                105,
+                                481,
+                                505,
+                                504
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        481,
+                                        505,
+                                        494
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                481,
+                                                505,
+                                                494
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 9: The same experiments as Figures 6, 10, run for 50,000 time-steps instead of 3000, to",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 21
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        492,
+                                        306,
+                                        506
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                492,
+                                                306,
+                                                506
+                                            ],
+                                            "score": 1.0,
+                                            "content": "illustrate the persistence of non-myopic behavior.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 22
+                                }
+                            ],
+                            "index": 21.5
+                        }
+                    ],
+                    "index": 20.25
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "image",
+                    "bbox": [
+                        161,
+                        104,
+                        447,
+                        672
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                161,
+                                104,
+                                447,
+                                672
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        161,
+                                        104,
+                                        447,
+                                        672
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                161,
+                                                104,
+                                                447,
+                                                672
+                                            ],
+                                            "score": 0.966,
+                                            "type": "image",
+                                            "image_path": "3e860d515c7ba817cb56c44c846a248ed6df5909c2c2c53c73553c58f29bb252.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 19,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        161,
+                                        104,
+                                        447,
+                                        118.56410256410257
+                                    ],
+                                    "spans": [],
+                                    "index": 0
+                                },
+                                {
+                                    "bbox": [
+                                        161,
+                                        118.56410256410257,
+                                        447,
+                                        133.12820512820514
+                                    ],
+                                    "spans": [],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        161,
+                                        133.12820512820514,
+                                        447,
+                                        147.6923076923077
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                },
+                                {
+                                    "bbox": [
+                                        161,
+                                        147.6923076923077,
+                                        447,
+                                        162.25641025641028
+                                    ],
+                                    "spans": [],
+                                    "index": 3
+                                },
+                                {
+                                    "bbox": [
+                                        161,
+                                        162.25641025641028,
+                                        447,
+                                        176.82051282051285
+                                    ],
+                                    "spans": [],
+                                    "index": 4
+                                },
+                                {
+                                    "bbox": [
+                                        161,
+                                        176.82051282051285,
+                                        447,
+                                        191.38461538461542
+                                    ],
+                                    "spans": [],
+                                    "index": 5
+                                },
+                                {
+                                    "bbox": [
+                                        161,
+                                        191.38461538461542,
+                                        447,
+                                        205.94871794871798
+                                    ],
+                                    "spans": [],
+                                    "index": 6
+                                },
+                                {
+                                    "bbox": [
+                                        161,
+                                        205.94871794871798,
+                                        447,
+                                        220.51282051282055
+                                    ],
+                                    "spans": [],
+                                    "index": 7
+                                },
+                                {
+                                    "bbox": [
+                                        161,
+                                        220.51282051282055,
+                                        447,
+                                        235.07692307692312
+                                    ],
+                                    "spans": [],
+                                    "index": 8
+                                },
+                                {
+                                    "bbox": [
+                                        161,
+                                        235.07692307692312,
+                                        447,
+                                        249.6410256410257
+                                    ],
+                                    "spans": [],
+                                    "index": 9
+                                },
+                                {
+                                    "bbox": [
+                                        161,
+                                        249.6410256410257,
+                                        447,
+                                        264.20512820512823
+                                    ],
+                                    "spans": [],
+                                    "index": 10
+                                },
+                                {
+                                    "bbox": [
+                                        161,
+                                        264.20512820512823,
+                                        447,
+                                        278.7692307692308
+                                    ],
+                                    "spans": [],
+                                    "index": 11
+                                },
+                                {
+                                    "bbox": [
+                                        161,
+                                        278.7692307692308,
+                                        447,
+                                        293.3333333333333
+                                    ],
+                                    "spans": [],
+                                    "index": 12
+                                },
+                                {
+                                    "bbox": [
+                                        161,
+                                        293.3333333333333,
+                                        447,
+                                        307.89743589743586
+                                    ],
+                                    "spans": [],
+                                    "index": 13
+                                },
+                                {
+                                    "bbox": [
+                                        161,
+                                        307.89743589743586,
+                                        447,
+                                        322.4615384615384
+                                    ],
+                                    "spans": [],
+                                    "index": 14
+                                },
+                                {
+                                    "bbox": [
+                                        161,
+                                        322.4615384615384,
+                                        447,
+                                        337.02564102564094
+                                    ],
+                                    "spans": [],
+                                    "index": 15
+                                },
+                                {
+                                    "bbox": [
+                                        161,
+                                        337.02564102564094,
+                                        447,
+                                        351.5897435897435
+                                    ],
+                                    "spans": [],
+                                    "index": 16
+                                },
+                                {
+                                    "bbox": [
+                                        161,
+                                        351.5897435897435,
+                                        447,
+                                        366.153846153846
+                                    ],
+                                    "spans": [],
+                                    "index": 17
+                                },
+                                {
+                                    "bbox": [
+                                        161,
+                                        366.153846153846,
+                                        447,
+                                        380.71794871794856
+                                    ],
+                                    "spans": [],
+                                    "index": 18
+                                },
+                                {
+                                    "bbox": [
+                                        161,
+                                        380.71794871794856,
+                                        447,
+                                        395.2820512820511
+                                    ],
+                                    "spans": [],
+                                    "index": 19
+                                },
+                                {
+                                    "bbox": [
+                                        161,
+                                        395.2820512820511,
+                                        447,
+                                        409.84615384615364
+                                    ],
+                                    "spans": [],
+                                    "index": 20
+                                },
+                                {
+                                    "bbox": [
+                                        161,
+                                        409.84615384615364,
+                                        447,
+                                        424.4102564102562
+                                    ],
+                                    "spans": [],
+                                    "index": 21
+                                },
+                                {
+                                    "bbox": [
+                                        161,
+                                        424.4102564102562,
+                                        447,
+                                        438.9743589743587
+                                    ],
+                                    "spans": [],
+                                    "index": 22
+                                },
+                                {
+                                    "bbox": [
+                                        161,
+                                        438.9743589743587,
+                                        447,
+                                        453.53846153846126
+                                    ],
+                                    "spans": [],
+                                    "index": 23
+                                },
+                                {
+                                    "bbox": [
+                                        161,
+                                        453.53846153846126,
+                                        447,
+                                        468.1025641025638
+                                    ],
+                                    "spans": [],
+                                    "index": 24
+                                },
+                                {
+                                    "bbox": [
+                                        161,
+                                        468.1025641025638,
+                                        447,
+                                        482.66666666666634
+                                    ],
+                                    "spans": [],
+                                    "index": 25
+                                },
+                                {
+                                    "bbox": [
+                                        161,
+                                        482.66666666666634,
+                                        447,
+                                        497.2307692307689
+                                    ],
+                                    "spans": [],
+                                    "index": 26
+                                },
+                                {
+                                    "bbox": [
+                                        161,
+                                        497.2307692307689,
+                                        447,
+                                        511.7948717948714
+                                    ],
+                                    "spans": [],
+                                    "index": 27
+                                },
+                                {
+                                    "bbox": [
+                                        161,
+                                        511.7948717948714,
+                                        447,
+                                        526.358974358974
+                                    ],
+                                    "spans": [],
+                                    "index": 28
+                                },
+                                {
+                                    "bbox": [
+                                        161,
+                                        526.358974358974,
+                                        447,
+                                        540.9230769230766
+                                    ],
+                                    "spans": [],
+                                    "index": 29
+                                },
+                                {
+                                    "bbox": [
+                                        161,
+                                        540.9230769230766,
+                                        447,
+                                        555.4871794871791
+                                    ],
+                                    "spans": [],
+                                    "index": 30
+                                },
+                                {
+                                    "bbox": [
+                                        161,
+                                        555.4871794871791,
+                                        447,
+                                        570.0512820512816
+                                    ],
+                                    "spans": [],
+                                    "index": 31
+                                },
+                                {
+                                    "bbox": [
+                                        161,
+                                        570.0512820512816,
+                                        447,
+                                        584.6153846153842
+                                    ],
+                                    "spans": [],
+                                    "index": 32
+                                },
+                                {
+                                    "bbox": [
+                                        161,
+                                        584.6153846153842,
+                                        447,
+                                        599.1794871794867
+                                    ],
+                                    "spans": [],
+                                    "index": 33
+                                },
+                                {
+                                    "bbox": [
+                                        161,
+                                        599.1794871794867,
+                                        447,
+                                        613.7435897435893
+                                    ],
+                                    "spans": [],
+                                    "index": 34
+                                },
+                                {
+                                    "bbox": [
+                                        161,
+                                        613.7435897435893,
+                                        447,
+                                        628.3076923076918
+                                    ],
+                                    "spans": [],
+                                    "index": 35
+                                },
+                                {
+                                    "bbox": [
+                                        161,
+                                        628.3076923076918,
+                                        447,
+                                        642.8717948717944
+                                    ],
+                                    "spans": [],
+                                    "index": 36
+                                },
+                                {
+                                    "bbox": [
+                                        161,
+                                        642.8717948717944,
+                                        447,
+                                        657.4358974358969
+                                    ],
+                                    "spans": [],
+                                    "index": 37
+                                },
+                                {
+                                    "bbox": [
+                                        161,
+                                        657.4358974358969,
+                                        447,
+                                        671.9999999999994
+                                    ],
+                                    "spans": [],
+                                    "index": 38
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                106,
+                                681,
+                                504,
+                                704
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        680,
+                                        506,
+                                        695
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                680,
+                                                506,
+                                                695
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 10: More independent experiments with Q-learning, exactly following Figure 6. Q-learning",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 39
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        691,
+                                        430,
+                                        705
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                691,
+                                                430,
+                                                705
+                                            ],
+                                            "score": 1.0,
+                                            "content": "fails the unit test in a total of 10/30 experiments (including those from Figure 6).",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 40
+                                }
+                            ],
+                            "index": 39.5
+                        }
+                    ],
+                    "index": 29.25
+                }
+            ],
+            "page_idx": 16,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        300,
+                        751,
+                        310,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                299,
+                                750,
+                                312,
+                                764
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        299,
+                                        750,
+                                        312,
+                                        764
+                                    ],
+                                    "score": 1.0,
+                                    "content": "",
+                                    "type": "text",
+                                    "height": 14,
+                                    "width": 13
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        107,
+                        27,
+                        307,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                26,
+                                308,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        26,
+                                        308,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Under review as a conference paper at ICLR 2021",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "image",
+                    "bbox": [
+                        161,
+                        104,
+                        447,
+                        672
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                161,
+                                104,
+                                447,
+                                672
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        161,
+                                        104,
+                                        447,
+                                        672
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                161,
+                                                104,
+                                                447,
+                                                672
+                                            ],
+                                            "score": 0.966,
+                                            "type": "image",
+                                            "image_path": "3e860d515c7ba817cb56c44c846a248ed6df5909c2c2c53c73553c58f29bb252.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 19,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        161,
+                                        104,
+                                        447,
+                                        118.56410256410257
+                                    ],
+                                    "spans": [],
+                                    "index": 0
+                                },
+                                {
+                                    "bbox": [
+                                        161,
+                                        118.56410256410257,
+                                        447,
+                                        133.12820512820514
+                                    ],
+                                    "spans": [],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        161,
+                                        133.12820512820514,
+                                        447,
+                                        147.6923076923077
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                },
+                                {
+                                    "bbox": [
+                                        161,
+                                        147.6923076923077,
+                                        447,
+                                        162.25641025641028
+                                    ],
+                                    "spans": [],
+                                    "index": 3
+                                },
+                                {
+                                    "bbox": [
+                                        161,
+                                        162.25641025641028,
+                                        447,
+                                        176.82051282051285
+                                    ],
+                                    "spans": [],
+                                    "index": 4
+                                },
+                                {
+                                    "bbox": [
+                                        161,
+                                        176.82051282051285,
+                                        447,
+                                        191.38461538461542
+                                    ],
+                                    "spans": [],
+                                    "index": 5
+                                },
+                                {
+                                    "bbox": [
+                                        161,
+                                        191.38461538461542,
+                                        447,
+                                        205.94871794871798
+                                    ],
+                                    "spans": [],
+                                    "index": 6
+                                },
+                                {
+                                    "bbox": [
+                                        161,
+                                        205.94871794871798,
+                                        447,
+                                        220.51282051282055
+                                    ],
+                                    "spans": [],
+                                    "index": 7
+                                },
+                                {
+                                    "bbox": [
+                                        161,
+                                        220.51282051282055,
+                                        447,
+                                        235.07692307692312
+                                    ],
+                                    "spans": [],
+                                    "index": 8
+                                },
+                                {
+                                    "bbox": [
+                                        161,
+                                        235.07692307692312,
+                                        447,
+                                        249.6410256410257
+                                    ],
+                                    "spans": [],
+                                    "index": 9
+                                },
+                                {
+                                    "bbox": [
+                                        161,
+                                        249.6410256410257,
+                                        447,
+                                        264.20512820512823
+                                    ],
+                                    "spans": [],
+                                    "index": 10
+                                },
+                                {
+                                    "bbox": [
+                                        161,
+                                        264.20512820512823,
+                                        447,
+                                        278.7692307692308
+                                    ],
+                                    "spans": [],
+                                    "index": 11
+                                },
+                                {
+                                    "bbox": [
+                                        161,
+                                        278.7692307692308,
+                                        447,
+                                        293.3333333333333
+                                    ],
+                                    "spans": [],
+                                    "index": 12
+                                },
+                                {
+                                    "bbox": [
+                                        161,
+                                        293.3333333333333,
+                                        447,
+                                        307.89743589743586
+                                    ],
+                                    "spans": [],
+                                    "index": 13
+                                },
+                                {
+                                    "bbox": [
+                                        161,
+                                        307.89743589743586,
+                                        447,
+                                        322.4615384615384
+                                    ],
+                                    "spans": [],
+                                    "index": 14
+                                },
+                                {
+                                    "bbox": [
+                                        161,
+                                        322.4615384615384,
+                                        447,
+                                        337.02564102564094
+                                    ],
+                                    "spans": [],
+                                    "index": 15
+                                },
+                                {
+                                    "bbox": [
+                                        161,
+                                        337.02564102564094,
+                                        447,
+                                        351.5897435897435
+                                    ],
+                                    "spans": [],
+                                    "index": 16
+                                },
+                                {
+                                    "bbox": [
+                                        161,
+                                        351.5897435897435,
+                                        447,
+                                        366.153846153846
+                                    ],
+                                    "spans": [],
+                                    "index": 17
+                                },
+                                {
+                                    "bbox": [
+                                        161,
+                                        366.153846153846,
+                                        447,
+                                        380.71794871794856
+                                    ],
+                                    "spans": [],
+                                    "index": 18
+                                },
+                                {
+                                    "bbox": [
+                                        161,
+                                        380.71794871794856,
+                                        447,
+                                        395.2820512820511
+                                    ],
+                                    "spans": [],
+                                    "index": 19
+                                },
+                                {
+                                    "bbox": [
+                                        161,
+                                        395.2820512820511,
+                                        447,
+                                        409.84615384615364
+                                    ],
+                                    "spans": [],
+                                    "index": 20
+                                },
+                                {
+                                    "bbox": [
+                                        161,
+                                        409.84615384615364,
+                                        447,
+                                        424.4102564102562
+                                    ],
+                                    "spans": [],
+                                    "index": 21
+                                },
+                                {
+                                    "bbox": [
+                                        161,
+                                        424.4102564102562,
+                                        447,
+                                        438.9743589743587
+                                    ],
+                                    "spans": [],
+                                    "index": 22
+                                },
+                                {
+                                    "bbox": [
+                                        161,
+                                        438.9743589743587,
+                                        447,
+                                        453.53846153846126
+                                    ],
+                                    "spans": [],
+                                    "index": 23
+                                },
+                                {
+                                    "bbox": [
+                                        161,
+                                        453.53846153846126,
+                                        447,
+                                        468.1025641025638
+                                    ],
+                                    "spans": [],
+                                    "index": 24
+                                },
+                                {
+                                    "bbox": [
+                                        161,
+                                        468.1025641025638,
+                                        447,
+                                        482.66666666666634
+                                    ],
+                                    "spans": [],
+                                    "index": 25
+                                },
+                                {
+                                    "bbox": [
+                                        161,
+                                        482.66666666666634,
+                                        447,
+                                        497.2307692307689
+                                    ],
+                                    "spans": [],
+                                    "index": 26
+                                },
+                                {
+                                    "bbox": [
+                                        161,
+                                        497.2307692307689,
+                                        447,
+                                        511.7948717948714
+                                    ],
+                                    "spans": [],
+                                    "index": 27
+                                },
+                                {
+                                    "bbox": [
+                                        161,
+                                        511.7948717948714,
+                                        447,
+                                        526.358974358974
+                                    ],
+                                    "spans": [],
+                                    "index": 28
+                                },
+                                {
+                                    "bbox": [
+                                        161,
+                                        526.358974358974,
+                                        447,
+                                        540.9230769230766
+                                    ],
+                                    "spans": [],
+                                    "index": 29
+                                },
+                                {
+                                    "bbox": [
+                                        161,
+                                        540.9230769230766,
+                                        447,
+                                        555.4871794871791
+                                    ],
+                                    "spans": [],
+                                    "index": 30
+                                },
+                                {
+                                    "bbox": [
+                                        161,
+                                        555.4871794871791,
+                                        447,
+                                        570.0512820512816
+                                    ],
+                                    "spans": [],
+                                    "index": 31
+                                },
+                                {
+                                    "bbox": [
+                                        161,
+                                        570.0512820512816,
+                                        447,
+                                        584.6153846153842
+                                    ],
+                                    "spans": [],
+                                    "index": 32
+                                },
+                                {
+                                    "bbox": [
+                                        161,
+                                        584.6153846153842,
+                                        447,
+                                        599.1794871794867
+                                    ],
+                                    "spans": [],
+                                    "index": 33
+                                },
+                                {
+                                    "bbox": [
+                                        161,
+                                        599.1794871794867,
+                                        447,
+                                        613.7435897435893
+                                    ],
+                                    "spans": [],
+                                    "index": 34
+                                },
+                                {
+                                    "bbox": [
+                                        161,
+                                        613.7435897435893,
+                                        447,
+                                        628.3076923076918
+                                    ],
+                                    "spans": [],
+                                    "index": 35
+                                },
+                                {
+                                    "bbox": [
+                                        161,
+                                        628.3076923076918,
+                                        447,
+                                        642.8717948717944
+                                    ],
+                                    "spans": [],
+                                    "index": 36
+                                },
+                                {
+                                    "bbox": [
+                                        161,
+                                        642.8717948717944,
+                                        447,
+                                        657.4358974358969
+                                    ],
+                                    "spans": [],
+                                    "index": 37
+                                },
+                                {
+                                    "bbox": [
+                                        161,
+                                        657.4358974358969,
+                                        447,
+                                        671.9999999999994
+                                    ],
+                                    "spans": [],
+                                    "index": 38
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                106,
+                                681,
+                                504,
+                                704
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        680,
+                                        506,
+                                        695
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                680,
+                                                506,
+                                                695
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 10: More independent experiments with Q-learning, exactly following Figure 6. Q-learning",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 39
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        691,
+                                        430,
+                                        705
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                691,
+                                                430,
+                                                705
+                                            ],
+                                            "score": 1.0,
+                                            "content": "fails the unit test in a total of 10/30 experiments (including those from Figure 6).",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 40
+                                }
+                            ],
+                            "index": 39.5
+                        }
+                    ],
+                    "index": 29.25
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "image",
+                    "bbox": [
+                        159,
+                        91,
+                        447,
+                        684
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                159,
+                                91,
+                                447,
+                                684
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        159,
+                                        91,
+                                        447,
+                                        684
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                159,
+                                                91,
+                                                447,
+                                                684
+                                            ],
+                                            "score": 0.964,
+                                            "type": "image",
+                                            "image_path": "18011e0877d74bf4ea8b3877bc9a66d2f427d421200bb03d1cb2dd7167b3d022.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 22,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        159,
+                                        91,
+                                        447,
+                                        104.17777777777778
+                                    ],
+                                    "spans": [],
+                                    "index": 0
+                                },
+                                {
+                                    "bbox": [
+                                        159,
+                                        104.17777777777778,
+                                        447,
+                                        117.35555555555555
+                                    ],
+                                    "spans": [],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        159,
+                                        117.35555555555555,
+                                        447,
+                                        130.53333333333333
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                },
+                                {
+                                    "bbox": [
+                                        159,
+                                        130.53333333333333,
+                                        447,
+                                        143.7111111111111
+                                    ],
+                                    "spans": [],
+                                    "index": 3
+                                },
+                                {
+                                    "bbox": [
+                                        159,
+                                        143.7111111111111,
+                                        447,
+                                        156.88888888888889
+                                    ],
+                                    "spans": [],
+                                    "index": 4
+                                },
+                                {
+                                    "bbox": [
+                                        159,
+                                        156.88888888888889,
+                                        447,
+                                        170.06666666666666
+                                    ],
+                                    "spans": [],
+                                    "index": 5
+                                },
+                                {
+                                    "bbox": [
+                                        159,
+                                        170.06666666666666,
+                                        447,
+                                        183.24444444444444
+                                    ],
+                                    "spans": [],
+                                    "index": 6
+                                },
+                                {
+                                    "bbox": [
+                                        159,
+                                        183.24444444444444,
+                                        447,
+                                        196.42222222222222
+                                    ],
+                                    "spans": [],
+                                    "index": 7
+                                },
+                                {
+                                    "bbox": [
+                                        159,
+                                        196.42222222222222,
+                                        447,
+                                        209.6
+                                    ],
+                                    "spans": [],
+                                    "index": 8
+                                },
+                                {
+                                    "bbox": [
+                                        159,
+                                        209.6,
+                                        447,
+                                        222.77777777777777
+                                    ],
+                                    "spans": [],
+                                    "index": 9
+                                },
+                                {
+                                    "bbox": [
+                                        159,
+                                        222.77777777777777,
+                                        447,
+                                        235.95555555555555
+                                    ],
+                                    "spans": [],
+                                    "index": 10
+                                },
+                                {
+                                    "bbox": [
+                                        159,
+                                        235.95555555555555,
+                                        447,
+                                        249.13333333333333
+                                    ],
+                                    "spans": [],
+                                    "index": 11
+                                },
+                                {
+                                    "bbox": [
+                                        159,
+                                        249.13333333333333,
+                                        447,
+                                        262.31111111111113
+                                    ],
+                                    "spans": [],
+                                    "index": 12
+                                },
+                                {
+                                    "bbox": [
+                                        159,
+                                        262.31111111111113,
+                                        447,
+                                        275.48888888888894
+                                    ],
+                                    "spans": [],
+                                    "index": 13
+                                },
+                                {
+                                    "bbox": [
+                                        159,
+                                        275.48888888888894,
+                                        447,
+                                        288.66666666666674
+                                    ],
+                                    "spans": [],
+                                    "index": 14
+                                },
+                                {
+                                    "bbox": [
+                                        159,
+                                        288.66666666666674,
+                                        447,
+                                        301.84444444444455
+                                    ],
+                                    "spans": [],
+                                    "index": 15
+                                },
+                                {
+                                    "bbox": [
+                                        159,
+                                        301.84444444444455,
+                                        447,
+                                        315.02222222222235
+                                    ],
+                                    "spans": [],
+                                    "index": 16
+                                },
+                                {
+                                    "bbox": [
+                                        159,
+                                        315.02222222222235,
+                                        447,
+                                        328.20000000000016
+                                    ],
+                                    "spans": [],
+                                    "index": 17
+                                },
+                                {
+                                    "bbox": [
+                                        159,
+                                        328.20000000000016,
+                                        447,
+                                        341.37777777777796
+                                    ],
+                                    "spans": [],
+                                    "index": 18
+                                },
+                                {
+                                    "bbox": [
+                                        159,
+                                        341.37777777777796,
+                                        447,
+                                        354.55555555555577
+                                    ],
+                                    "spans": [],
+                                    "index": 19
+                                },
+                                {
+                                    "bbox": [
+                                        159,
+                                        354.55555555555577,
+                                        447,
+                                        367.7333333333336
+                                    ],
+                                    "spans": [],
+                                    "index": 20
+                                },
+                                {
+                                    "bbox": [
+                                        159,
+                                        367.7333333333336,
+                                        447,
+                                        380.9111111111114
+                                    ],
+                                    "spans": [],
+                                    "index": 21
+                                },
+                                {
+                                    "bbox": [
+                                        159,
+                                        380.9111111111114,
+                                        447,
+                                        394.0888888888892
+                                    ],
+                                    "spans": [],
+                                    "index": 22
+                                },
+                                {
+                                    "bbox": [
+                                        159,
+                                        394.0888888888892,
+                                        447,
+                                        407.266666666667
+                                    ],
+                                    "spans": [],
+                                    "index": 23
+                                },
+                                {
+                                    "bbox": [
+                                        159,
+                                        407.266666666667,
+                                        447,
+                                        420.4444444444448
+                                    ],
+                                    "spans": [],
+                                    "index": 24
+                                },
+                                {
+                                    "bbox": [
+                                        159,
+                                        420.4444444444448,
+                                        447,
+                                        433.6222222222226
+                                    ],
+                                    "spans": [],
+                                    "index": 25
+                                },
+                                {
+                                    "bbox": [
+                                        159,
+                                        433.6222222222226,
+                                        447,
+                                        446.8000000000004
+                                    ],
+                                    "spans": [],
+                                    "index": 26
+                                },
+                                {
+                                    "bbox": [
+                                        159,
+                                        446.8000000000004,
+                                        447,
+                                        459.9777777777782
+                                    ],
+                                    "spans": [],
+                                    "index": 27
+                                },
+                                {
+                                    "bbox": [
+                                        159,
+                                        459.9777777777782,
+                                        447,
+                                        473.155555555556
+                                    ],
+                                    "spans": [],
+                                    "index": 28
+                                },
+                                {
+                                    "bbox": [
+                                        159,
+                                        473.155555555556,
+                                        447,
+                                        486.3333333333338
+                                    ],
+                                    "spans": [],
+                                    "index": 29
+                                },
+                                {
+                                    "bbox": [
+                                        159,
+                                        486.3333333333338,
+                                        447,
+                                        499.51111111111163
+                                    ],
+                                    "spans": [],
+                                    "index": 30
+                                },
+                                {
+                                    "bbox": [
+                                        159,
+                                        499.51111111111163,
+                                        447,
+                                        512.6888888888894
+                                    ],
+                                    "spans": [],
+                                    "index": 31
+                                },
+                                {
+                                    "bbox": [
+                                        159,
+                                        512.6888888888894,
+                                        447,
+                                        525.8666666666672
+                                    ],
+                                    "spans": [],
+                                    "index": 32
+                                },
+                                {
+                                    "bbox": [
+                                        159,
+                                        525.8666666666672,
+                                        447,
+                                        539.044444444445
+                                    ],
+                                    "spans": [],
+                                    "index": 33
+                                },
+                                {
+                                    "bbox": [
+                                        159,
+                                        539.044444444445,
+                                        447,
+                                        552.2222222222229
+                                    ],
+                                    "spans": [],
+                                    "index": 34
+                                },
+                                {
+                                    "bbox": [
+                                        159,
+                                        552.2222222222229,
+                                        447,
+                                        565.4000000000007
+                                    ],
+                                    "spans": [],
+                                    "index": 35
+                                },
+                                {
+                                    "bbox": [
+                                        159,
+                                        565.4000000000007,
+                                        447,
+                                        578.5777777777785
+                                    ],
+                                    "spans": [],
+                                    "index": 36
+                                },
+                                {
+                                    "bbox": [
+                                        159,
+                                        578.5777777777785,
+                                        447,
+                                        591.7555555555563
+                                    ],
+                                    "spans": [],
+                                    "index": 37
+                                },
+                                {
+                                    "bbox": [
+                                        159,
+                                        591.7555555555563,
+                                        447,
+                                        604.9333333333341
+                                    ],
+                                    "spans": [],
+                                    "index": 38
+                                },
+                                {
+                                    "bbox": [
+                                        159,
+                                        604.9333333333341,
+                                        447,
+                                        618.1111111111119
+                                    ],
+                                    "spans": [],
+                                    "index": 39
+                                },
+                                {
+                                    "bbox": [
+                                        159,
+                                        618.1111111111119,
+                                        447,
+                                        631.2888888888897
+                                    ],
+                                    "spans": [],
+                                    "index": 40
+                                },
+                                {
+                                    "bbox": [
+                                        159,
+                                        631.2888888888897,
+                                        447,
+                                        644.4666666666675
+                                    ],
+                                    "spans": [],
+                                    "index": 41
+                                },
+                                {
+                                    "bbox": [
+                                        159,
+                                        644.4666666666675,
+                                        447,
+                                        657.6444444444453
+                                    ],
+                                    "spans": [],
+                                    "index": 42
+                                },
+                                {
+                                    "bbox": [
+                                        159,
+                                        657.6444444444453,
+                                        447,
+                                        670.8222222222231
+                                    ],
+                                    "spans": [],
+                                    "index": 43
+                                },
+                                {
+                                    "bbox": [
+                                        159,
+                                        670.8222222222231,
+                                        447,
+                                        684.0000000000009
+                                    ],
+                                    "spans": [],
+                                    "index": 44
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                106,
+                                692,
+                                505,
+                                715
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        692,
+                                        505,
+                                        705
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                692,
+                                                505,
+                                                705
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 11: More independent experiments with Q-learning, exactly following Figure 6, except also",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 45
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        703,
+                                        406,
+                                        716
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                703,
+                                                266,
+                                                716
+                                            ],
+                                            "score": 1.0,
+                                            "content": "using context swapping. This leads to a",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                267,
+                                                703,
+                                                291,
+                                                714
+                                            ],
+                                            "score": 0.87,
+                                            "content": "100 \\%",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                291,
+                                                703,
+                                                406,
+                                                716
+                                            ],
+                                            "score": 1.0,
+                                            "content": "success rate on the unit test.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 46
+                                }
+                            ],
+                            "index": 45.5
+                        }
+                    ],
+                    "index": 33.75
+                }
+            ],
+            "page_idx": 17,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        300,
+                        751,
+                        311,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                299,
+                                750,
+                                312,
+                                763
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        299,
+                                        750,
+                                        312,
+                                        763
+                                    ],
+                                    "score": 1.0,
+                                    "content": "18",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        107,
+                        27,
+                        307,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                26,
+                                308,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        26,
+                                        308,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Under review as a conference paper at ICLR 2021",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "image",
+                    "bbox": [
+                        159,
+                        91,
+                        447,
+                        684
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                159,
+                                91,
+                                447,
+                                684
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        159,
+                                        91,
+                                        447,
+                                        684
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                159,
+                                                91,
+                                                447,
+                                                684
+                                            ],
+                                            "score": 0.964,
+                                            "type": "image",
+                                            "image_path": "18011e0877d74bf4ea8b3877bc9a66d2f427d421200bb03d1cb2dd7167b3d022.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 22,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        159,
+                                        91,
+                                        447,
+                                        104.17777777777778
+                                    ],
+                                    "spans": [],
+                                    "index": 0
+                                },
+                                {
+                                    "bbox": [
+                                        159,
+                                        104.17777777777778,
+                                        447,
+                                        117.35555555555555
+                                    ],
+                                    "spans": [],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        159,
+                                        117.35555555555555,
+                                        447,
+                                        130.53333333333333
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                },
+                                {
+                                    "bbox": [
+                                        159,
+                                        130.53333333333333,
+                                        447,
+                                        143.7111111111111
+                                    ],
+                                    "spans": [],
+                                    "index": 3
+                                },
+                                {
+                                    "bbox": [
+                                        159,
+                                        143.7111111111111,
+                                        447,
+                                        156.88888888888889
+                                    ],
+                                    "spans": [],
+                                    "index": 4
+                                },
+                                {
+                                    "bbox": [
+                                        159,
+                                        156.88888888888889,
+                                        447,
+                                        170.06666666666666
+                                    ],
+                                    "spans": [],
+                                    "index": 5
+                                },
+                                {
+                                    "bbox": [
+                                        159,
+                                        170.06666666666666,
+                                        447,
+                                        183.24444444444444
+                                    ],
+                                    "spans": [],
+                                    "index": 6
+                                },
+                                {
+                                    "bbox": [
+                                        159,
+                                        183.24444444444444,
+                                        447,
+                                        196.42222222222222
+                                    ],
+                                    "spans": [],
+                                    "index": 7
+                                },
+                                {
+                                    "bbox": [
+                                        159,
+                                        196.42222222222222,
+                                        447,
+                                        209.6
+                                    ],
+                                    "spans": [],
+                                    "index": 8
+                                },
+                                {
+                                    "bbox": [
+                                        159,
+                                        209.6,
+                                        447,
+                                        222.77777777777777
+                                    ],
+                                    "spans": [],
+                                    "index": 9
+                                },
+                                {
+                                    "bbox": [
+                                        159,
+                                        222.77777777777777,
+                                        447,
+                                        235.95555555555555
+                                    ],
+                                    "spans": [],
+                                    "index": 10
+                                },
+                                {
+                                    "bbox": [
+                                        159,
+                                        235.95555555555555,
+                                        447,
+                                        249.13333333333333
+                                    ],
+                                    "spans": [],
+                                    "index": 11
+                                },
+                                {
+                                    "bbox": [
+                                        159,
+                                        249.13333333333333,
+                                        447,
+                                        262.31111111111113
+                                    ],
+                                    "spans": [],
+                                    "index": 12
+                                },
+                                {
+                                    "bbox": [
+                                        159,
+                                        262.31111111111113,
+                                        447,
+                                        275.48888888888894
+                                    ],
+                                    "spans": [],
+                                    "index": 13
+                                },
+                                {
+                                    "bbox": [
+                                        159,
+                                        275.48888888888894,
+                                        447,
+                                        288.66666666666674
+                                    ],
+                                    "spans": [],
+                                    "index": 14
+                                },
+                                {
+                                    "bbox": [
+                                        159,
+                                        288.66666666666674,
+                                        447,
+                                        301.84444444444455
+                                    ],
+                                    "spans": [],
+                                    "index": 15
+                                },
+                                {
+                                    "bbox": [
+                                        159,
+                                        301.84444444444455,
+                                        447,
+                                        315.02222222222235
+                                    ],
+                                    "spans": [],
+                                    "index": 16
+                                },
+                                {
+                                    "bbox": [
+                                        159,
+                                        315.02222222222235,
+                                        447,
+                                        328.20000000000016
+                                    ],
+                                    "spans": [],
+                                    "index": 17
+                                },
+                                {
+                                    "bbox": [
+                                        159,
+                                        328.20000000000016,
+                                        447,
+                                        341.37777777777796
+                                    ],
+                                    "spans": [],
+                                    "index": 18
+                                },
+                                {
+                                    "bbox": [
+                                        159,
+                                        341.37777777777796,
+                                        447,
+                                        354.55555555555577
+                                    ],
+                                    "spans": [],
+                                    "index": 19
+                                },
+                                {
+                                    "bbox": [
+                                        159,
+                                        354.55555555555577,
+                                        447,
+                                        367.7333333333336
+                                    ],
+                                    "spans": [],
+                                    "index": 20
+                                },
+                                {
+                                    "bbox": [
+                                        159,
+                                        367.7333333333336,
+                                        447,
+                                        380.9111111111114
+                                    ],
+                                    "spans": [],
+                                    "index": 21
+                                },
+                                {
+                                    "bbox": [
+                                        159,
+                                        380.9111111111114,
+                                        447,
+                                        394.0888888888892
+                                    ],
+                                    "spans": [],
+                                    "index": 22
+                                },
+                                {
+                                    "bbox": [
+                                        159,
+                                        394.0888888888892,
+                                        447,
+                                        407.266666666667
+                                    ],
+                                    "spans": [],
+                                    "index": 23
+                                },
+                                {
+                                    "bbox": [
+                                        159,
+                                        407.266666666667,
+                                        447,
+                                        420.4444444444448
+                                    ],
+                                    "spans": [],
+                                    "index": 24
+                                },
+                                {
+                                    "bbox": [
+                                        159,
+                                        420.4444444444448,
+                                        447,
+                                        433.6222222222226
+                                    ],
+                                    "spans": [],
+                                    "index": 25
+                                },
+                                {
+                                    "bbox": [
+                                        159,
+                                        433.6222222222226,
+                                        447,
+                                        446.8000000000004
+                                    ],
+                                    "spans": [],
+                                    "index": 26
+                                },
+                                {
+                                    "bbox": [
+                                        159,
+                                        446.8000000000004,
+                                        447,
+                                        459.9777777777782
+                                    ],
+                                    "spans": [],
+                                    "index": 27
+                                },
+                                {
+                                    "bbox": [
+                                        159,
+                                        459.9777777777782,
+                                        447,
+                                        473.155555555556
+                                    ],
+                                    "spans": [],
+                                    "index": 28
+                                },
+                                {
+                                    "bbox": [
+                                        159,
+                                        473.155555555556,
+                                        447,
+                                        486.3333333333338
+                                    ],
+                                    "spans": [],
+                                    "index": 29
+                                },
+                                {
+                                    "bbox": [
+                                        159,
+                                        486.3333333333338,
+                                        447,
+                                        499.51111111111163
+                                    ],
+                                    "spans": [],
+                                    "index": 30
+                                },
+                                {
+                                    "bbox": [
+                                        159,
+                                        499.51111111111163,
+                                        447,
+                                        512.6888888888894
+                                    ],
+                                    "spans": [],
+                                    "index": 31
+                                },
+                                {
+                                    "bbox": [
+                                        159,
+                                        512.6888888888894,
+                                        447,
+                                        525.8666666666672
+                                    ],
+                                    "spans": [],
+                                    "index": 32
+                                },
+                                {
+                                    "bbox": [
+                                        159,
+                                        525.8666666666672,
+                                        447,
+                                        539.044444444445
+                                    ],
+                                    "spans": [],
+                                    "index": 33
+                                },
+                                {
+                                    "bbox": [
+                                        159,
+                                        539.044444444445,
+                                        447,
+                                        552.2222222222229
+                                    ],
+                                    "spans": [],
+                                    "index": 34
+                                },
+                                {
+                                    "bbox": [
+                                        159,
+                                        552.2222222222229,
+                                        447,
+                                        565.4000000000007
+                                    ],
+                                    "spans": [],
+                                    "index": 35
+                                },
+                                {
+                                    "bbox": [
+                                        159,
+                                        565.4000000000007,
+                                        447,
+                                        578.5777777777785
+                                    ],
+                                    "spans": [],
+                                    "index": 36
+                                },
+                                {
+                                    "bbox": [
+                                        159,
+                                        578.5777777777785,
+                                        447,
+                                        591.7555555555563
+                                    ],
+                                    "spans": [],
+                                    "index": 37
+                                },
+                                {
+                                    "bbox": [
+                                        159,
+                                        591.7555555555563,
+                                        447,
+                                        604.9333333333341
+                                    ],
+                                    "spans": [],
+                                    "index": 38
+                                },
+                                {
+                                    "bbox": [
+                                        159,
+                                        604.9333333333341,
+                                        447,
+                                        618.1111111111119
+                                    ],
+                                    "spans": [],
+                                    "index": 39
+                                },
+                                {
+                                    "bbox": [
+                                        159,
+                                        618.1111111111119,
+                                        447,
+                                        631.2888888888897
+                                    ],
+                                    "spans": [],
+                                    "index": 40
+                                },
+                                {
+                                    "bbox": [
+                                        159,
+                                        631.2888888888897,
+                                        447,
+                                        644.4666666666675
+                                    ],
+                                    "spans": [],
+                                    "index": 41
+                                },
+                                {
+                                    "bbox": [
+                                        159,
+                                        644.4666666666675,
+                                        447,
+                                        657.6444444444453
+                                    ],
+                                    "spans": [],
+                                    "index": 42
+                                },
+                                {
+                                    "bbox": [
+                                        159,
+                                        657.6444444444453,
+                                        447,
+                                        670.8222222222231
+                                    ],
+                                    "spans": [],
+                                    "index": 43
+                                },
+                                {
+                                    "bbox": [
+                                        159,
+                                        670.8222222222231,
+                                        447,
+                                        684.0000000000009
+                                    ],
+                                    "spans": [],
+                                    "index": 44
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                106,
+                                692,
+                                505,
+                                715
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        692,
+                                        505,
+                                        705
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                692,
+                                                505,
+                                                705
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 11: More independent experiments with Q-learning, exactly following Figure 6, except also",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 45
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        703,
+                                        406,
+                                        716
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                703,
+                                                266,
+                                                716
+                                            ],
+                                            "score": 1.0,
+                                            "content": "using context swapping. This leads to a",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                267,
+                                                703,
+                                                291,
+                                                714
+                                            ],
+                                            "score": 0.87,
+                                            "content": "100 \\%",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                291,
+                                                703,
+                                                406,
+                                                716
+                                            ],
+                                            "score": 1.0,
+                                            "content": "success rate on the unit test.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 46
+                                }
+                            ],
+                            "index": 45.5
+                        }
+                    ],
+                    "index": 33.75
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        82,
+                        259,
+                        94
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                82,
+                                260,
+                                95
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        82,
+                                        260,
+                                        95
+                                    ],
+                                    "score": 1.0,
+                                    "content": "9.2 CONTENT RECOMMENDATION",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        }
+                    ],
+                    "index": 0
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        102,
+                        244,
+                        114
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                102,
+                                245,
+                                115
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        102,
+                                        245,
+                                        115
+                                    ],
+                                    "score": 1.0,
+                                    "content": "9.2.1 ENVIRONMENT DETAILS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        }
+                    ],
+                    "index": 1
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        122,
+                        295,
+                        134
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                120,
+                                297,
+                                136
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        120,
+                                        297,
+                                        136
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The evironment has the following components:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        }
+                    ],
+                    "index": 2
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        127,
+                        142,
+                        505,
+                        243
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                130,
+                                142,
+                                505,
+                                154
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        130,
+                                        142,
+                                        191,
+                                        154
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1. User type,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        191,
+                                        142,
+                                        201,
+                                        153
+                                    ],
+                                    "score": 0.86,
+                                    "content": "x ^ { t }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        201,
+                                        142,
+                                        505,
+                                        154
+                                    ],
+                                    "score": 1.0,
+                                    "content": ": categorical variable representing different types of users. The content",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                154,
+                                426,
+                                166
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        154,
+                                        426,
+                                        166
+                                    ],
+                                    "score": 1.0,
+                                    "content": "recommender conditions its predictions on the type of the current user.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                128,
+                                167,
+                                506,
+                                182
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        128,
+                                        167,
+                                        200,
+                                        182
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2. User loyalty,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        200,
+                                        168,
+                                        211,
+                                        180
+                                    ],
+                                    "score": 0.87,
+                                    "content": "\\mathbf { g } ^ { t }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        211,
+                                        167,
+                                        483,
+                                        182
+                                    ],
+                                    "score": 1.0,
+                                    "content": ": the propensity for users of each type to use the platform. User",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        483,
+                                        168,
+                                        493,
+                                        178
+                                    ],
+                                    "score": 0.86,
+                                    "content": "x ^ { t }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        494,
+                                        167,
+                                        506,
+                                        182
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                178,
+                                461,
+                                193
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        178,
+                                        439,
+                                        193
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sampled from a categorical distribution with parameters given by softmax",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        439,
+                                        179,
+                                        456,
+                                        192
+                                    ],
+                                    "score": 0.83,
+                                    "content": "( \\mathbf { g } ^ { t } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        457,
+                                        178,
+                                        461,
+                                        193
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                128,
+                                193,
+                                505,
+                                207
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        128,
+                                        193,
+                                        197,
+                                        207
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3. Article type,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        197,
+                                        194,
+                                        207,
+                                        205
+                                    ],
+                                    "score": 0.87,
+                                    "content": "y ^ { t }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        208,
+                                        193,
+                                        505,
+                                        207
+                                    ],
+                                    "score": 1.0,
+                                    "content": ": a categorical variable (one-hot encoding) representing the type of article",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                142,
+                                206,
+                                225,
+                                218
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        142,
+                                        206,
+                                        225,
+                                        218
+                                    ],
+                                    "score": 1.0,
+                                    "content": "selected by the user.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                216,
+                                507,
+                                237
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        216,
+                                        205,
+                                        237
+                                    ],
+                                    "score": 1.0,
+                                    "content": "4. User interests,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        205,
+                                        219,
+                                        222,
+                                        231
+                                    ],
+                                    "score": 0.82,
+                                    "content": "\\mathbf { W } ^ { t }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        222,
+                                        216,
+                                        317,
+                                        237
+                                    ],
+                                    "score": 1.0,
+                                    "content": ": a matrix whose entries",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        317,
+                                        220,
+                                        339,
+                                        234
+                                    ],
+                                    "score": 0.92,
+                                    "content": "W _ { x , y } ^ { t }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        339,
+                                        216,
+                                        507,
+                                        237
+                                    ],
+                                    "score": 1.0,
+                                    "content": "represent the average interest user of type",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                142,
+                                229,
+                                254,
+                                245
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        142,
+                                        233,
+                                        149,
+                                        241
+                                    ],
+                                    "score": 0.73,
+                                    "content": "x",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        150,
+                                        229,
+                                        243,
+                                        245
+                                    ],
+                                    "score": 1.0,
+                                    "content": "have in articles of type",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        244,
+                                        233,
+                                        250,
+                                        243
+                                    ],
+                                    "score": 0.78,
+                                    "content": "y",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        250,
+                                        229,
+                                        254,
+                                        245
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        }
+                    ],
+                    "index": 6.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        251,
+                        505,
+                        307
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                251,
+                                505,
+                                263
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        252,
+                                        178,
+                                        263
+                                    ],
+                                    "score": 1.0,
+                                    "content": "At each time step",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        178,
+                                        253,
+                                        183,
+                                        262
+                                    ],
+                                    "score": 0.68,
+                                    "content": "t",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        183,
+                                        252,
+                                        213,
+                                        263
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", a user",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        213,
+                                        251,
+                                        223,
+                                        261
+                                    ],
+                                    "score": 0.87,
+                                    "content": "x ^ { t }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        224,
+                                        252,
+                                        505,
+                                        263
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is sampled from a categorical distribution (based on the loyalty of the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                262,
+                                505,
+                                275
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        262,
+                                        505,
+                                        275
+                                    ],
+                                    "score": 1.0,
+                                    "content": "different user types), then the recommendation system selects which type of article to present in the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                272,
+                                506,
+                                286
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        272,
+                                        506,
+                                        286
+                                    ],
+                                    "score": 1.0,
+                                    "content": "top position, and finally, the user selects an article. The goal of the recommendation system is to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                283,
+                                506,
+                                297
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        283,
+                                        506,
+                                        297
+                                    ],
+                                    "score": 1.0,
+                                    "content": "predict the likelihood that the user would click on each of the available articles, in order to select the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                295,
+                                272,
+                                307
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        295,
+                                        272,
+                                        307
+                                    ],
+                                    "score": 1.0,
+                                    "content": "one which is most interesting to the user.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        }
+                    ],
+                    "index": 13
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        312,
+                        505,
+                        378
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                309,
+                                506,
+                                326
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        309,
+                                        171,
+                                        326
+                                    ],
+                                    "score": 1.0,
+                                    "content": "User loyalty for",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        171,
+                                        312,
+                                        181,
+                                        322
+                                    ],
+                                    "score": 0.87,
+                                    "content": "x ^ { t }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        181,
+                                        309,
+                                        506,
+                                        326
+                                    ],
+                                    "score": 1.0,
+                                    "content": "then changes in accordance with the self-selection effect, increasing or decreasing",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                322,
+                                506,
+                                336
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        322,
+                                        418,
+                                        336
+                                    ],
+                                    "score": 1.0,
+                                    "content": "proportionally to their interest in the top article. The interests of user type",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        419,
+                                        323,
+                                        429,
+                                        333
+                                    ],
+                                    "score": 0.86,
+                                    "content": "x ^ { t }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        429,
+                                        322,
+                                        506,
+                                        336
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(represented by a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                333,
+                                506,
+                                347
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        333,
+                                        150,
+                                        347
+                                    ],
+                                    "score": 1.0,
+                                    "content": "column of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        150,
+                                        334,
+                                        166,
+                                        344
+                                    ],
+                                    "score": 0.81,
+                                    "content": "\\mathbf { W } ^ { t }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        167,
+                                        333,
+                                        506,
+                                        347
+                                    ],
+                                    "score": 1.0,
+                                    "content": ") also change; in accordance with the illusory truth effect, their interest in the topic of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                345,
+                                506,
+                                357
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        345,
+                                        506,
+                                        357
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the top article (as chosen by the recommender system) always increases. Overall, this environment",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                356,
+                                506,
+                                368
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        356,
+                                        506,
+                                        368
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is an extremely crude representation of reality, but it allows us to incorporate both the effects of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                366,
+                                432,
+                                379
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        366,
+                                        432,
+                                        379
+                                    ],
+                                    "score": 1.0,
+                                    "content": "self-selection (via covariate shift), and the illusory truth effect (via concept shift).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        }
+                    ],
+                    "index": 18.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        383,
+                        506,
+                        428
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                384,
+                                506,
+                                396
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        384,
+                                        506,
+                                        396
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Formally, this environment is similar to a POMDP\\R, i.e. a POMDP with no reward function,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                395,
+                                505,
+                                407
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        395,
+                                        505,
+                                        407
+                                    ],
+                                    "score": 1.0,
+                                    "content": "also known as a world model (Armstrong & O’Rourke, 2017; Hadfield-Menell et al., 2017); the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                405,
+                                507,
+                                419
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        405,
+                                        507,
+                                        419
+                                    ],
+                                    "score": 1.0,
+                                    "content": "difference is that the learner observes the input before acting and only observes the target after acting.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                416,
+                                307,
+                                429
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        416,
+                                        307,
+                                        429
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The states, observations, and actions given below.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        }
+                    ],
+                    "index": 23.5
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        235,
+                        431,
+                        376,
+                        464
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                235,
+                                431,
+                                376,
+                                464
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        235,
+                                        431,
+                                        376,
+                                        464
+                                    ],
+                                    "score": 0.84,
+                                    "content": "\\begin{array} { c } { { \\boldsymbol { s } ^ { t } = ( \\mathbf { g } ^ { t } , \\mathbf { W } ^ { t } , \\boldsymbol { x } ^ { t } , \\boldsymbol { y } ^ { t } ) } } \\\\ { { \\boldsymbol { o } _ { \\mathrm { p r e } } ^ { t } , ~ \\boldsymbol { a } ^ { t } , ~ \\boldsymbol { o } _ { \\mathrm { p o s t } } ^ { t } = ( \\boldsymbol { x } ^ { t } , \\hat { \\boldsymbol { y } } ^ { t } , \\boldsymbol { y } ^ { t } ) } } \\end{array}",
+                                    "type": "interline_equation",
+                                    "image_path": "db0b38546afafe534e88febd9a237072bd24f328ca4cb11931aed443be558f9d.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 26.5,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                235,
+                                431,
+                                376,
+                                447.5
+                            ],
+                            "spans": [],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                235,
+                                447.5,
+                                376,
+                                464.0
+                            ],
+                            "spans": [],
+                            "index": 27
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        466,
+                        276,
+                        478
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                464,
+                                277,
+                                480
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        464,
+                                        277,
+                                        480
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The state transition function is defined by:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        }
+                    ],
+                    "index": 28
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        201,
+                        480,
+                        409,
+                        564
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                201,
+                                480,
+                                409,
+                                564
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        201,
+                                        480,
+                                        409,
+                                        564
+                                    ],
+                                    "score": 0.94,
+                                    "content": "\\begin{array} { r l r } { \\mathbf { g } _ { x ^ { t } } ^ { t + 1 } = \\mathbf { g } _ { x ^ { t } } ^ { t } + \\alpha _ { 1 } W _ { x ^ { t } , \\hat { y } ^ { t } } ^ { t } } & { } & \\\\ { \\mathbf { W } _ { x ^ { t } , \\hat { y } ^ { t } } ^ { t + 1 / 2 } = W _ { x ^ { t } , \\hat { y } ^ { t } } ^ { t } + \\alpha _ { 2 } ; \\quad \\mathbf { W } _ { x ^ { t } } ^ { t + 1 } = \\frac { \\mathbf { W } _ { x ^ { t } } ^ { t + 1 / 2 } } { \\Vert \\mathbf { W } _ { x ^ { t } } ^ { t + 1 / 2 } \\Vert _ { 2 } } } & { } & \\\\ { x ^ { t + 1 } \\sim \\mathrm { s o f t m a x } ( \\mathbf { g } ^ { t + 1 } ) } & { } & \\\\ { y ^ { t + 1 } \\sim \\mathrm { s o f t m a x } ( \\mathbf { W } _ { x ^ { t + 1 } } ^ { t + 1 } ) } & { } & \\end{array}",
+                                    "type": "interline_equation",
+                                    "image_path": "18371f2cadc36f7f8a1bbad2aa2e9888bab73b266fa7d957e9964ff1779ada1c.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 31.5,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                201,
+                                480,
+                                409,
+                                494.0
+                            ],
+                            "spans": [],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                201,
+                                494.0,
+                                409,
+                                508.0
+                            ],
+                            "spans": [],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                201,
+                                508.0,
+                                409,
+                                522.0
+                            ],
+                            "spans": [],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                201,
+                                522.0,
+                                409,
+                                536.0
+                            ],
+                            "spans": [],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                201,
+                                536.0,
+                                409,
+                                550.0
+                            ],
+                            "spans": [],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                201,
+                                550.0,
+                                409,
+                                564.0
+                            ],
+                            "spans": [],
+                            "index": 34
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        568,
+                        504,
+                        602
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                567,
+                                506,
+                                581
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        567,
+                                        135,
+                                        581
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Where",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        135,
+                                        568,
+                                        145,
+                                        580
+                                    ],
+                                    "score": 0.88,
+                                    "content": "\\hat { y } ^ { t }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        145,
+                                        567,
+                                        356,
+                                        581
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is the top article as chosen by the recommender, and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        356,
+                                        570,
+                                        367,
+                                        579
+                                    ],
+                                    "score": 0.76,
+                                    "content": "\\alpha _ { 1 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        368,
+                                        567,
+                                        371,
+                                        581
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        372,
+                                        570,
+                                        384,
+                                        579
+                                    ],
+                                    "score": 0.72,
+                                    "content": "\\alpha _ { 2 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        384,
+                                        567,
+                                        506,
+                                        581
+                                    ],
+                                    "score": 1.0,
+                                    "content": "represent the rate of covariate",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                577,
+                                504,
+                                592
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        577,
+                                        299,
+                                        592
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and concept shift (respectively). The update for",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        299,
+                                        578,
+                                        326,
+                                        590
+                                    ],
+                                    "score": 0.89,
+                                    "content": "\\mathbf { W } ^ { t + 1 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        326,
+                                        577,
+                                        493,
+                                        592
+                                    ],
+                                    "score": 1.0,
+                                    "content": "merely increases the interest of user type",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        493,
+                                        579,
+                                        504,
+                                        589
+                                    ],
+                                    "score": 0.84,
+                                    "content": "x ^ { t }",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                589,
+                                366,
+                                603
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        589,
+                                        164,
+                                        603
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in article type",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        164,
+                                        590,
+                                        173,
+                                        601
+                                    ],
+                                    "score": 0.88,
+                                    "content": "\\hat { y } ^ { t }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        174,
+                                        589,
+                                        366,
+                                        603
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", then normalizes the interests for that user type.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        }
+                    ],
+                    "index": 36
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        613,
+                        258,
+                        625
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                614,
+                                259,
+                                626
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        614,
+                                        259,
+                                        626
+                                    ],
+                                    "score": 1.0,
+                                    "content": "9.2.2 REPRODUCIBILITY DETAILS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        }
+                    ],
+                    "index": 38
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        632,
+                        506,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                632,
+                                506,
+                                646
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        632,
+                                        506,
+                                        646
+                                    ],
+                                    "score": 1.0,
+                                    "content": "For these experiments, the recommendation system is a ReLU-MLP with 1 hidden layer of 100 units,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                644,
+                                505,
+                                655
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        644,
+                                        331,
+                                        655
+                                    ],
+                                    "score": 1.0,
+                                    "content": "trained via supervised learning with SGD (learning rate",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        332,
+                                        644,
+                                        361,
+                                        655
+                                    ],
+                                    "score": 0.83,
+                                    "content": "= 0 . 0 1 ",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        361,
+                                        644,
+                                        505,
+                                        655
+                                    ],
+                                    "score": 1.0,
+                                    "content": ") to predict which article a user will",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                653,
+                                506,
+                                669
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        653,
+                                        506,
+                                        669
+                                    ],
+                                    "score": 1.0,
+                                    "content": "select. Actions are sampled from the MLP’s predictive distribution. We apply PBT without any",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                666,
+                                506,
+                                678
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        666,
+                                        506,
+                                        678
+                                    ],
+                                    "score": 1.0,
+                                    "content": "hyperparameter selection (this amounts to just doing the EXPLOIT step), and an interval of 10,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                677,
+                                505,
+                                689
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        677,
+                                        505,
+                                        689
+                                    ],
+                                    "score": 1.0,
+                                    "content": "selecting on accuracy. We use a population of 20 learners (whether applying PBT or not), and match",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                686,
+                                506,
+                                701
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        686,
+                                        356,
+                                        701
+                                    ],
+                                    "score": 1.0,
+                                    "content": "random seeds for the trials with and without PBT. We initialize",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        356,
+                                        687,
+                                        368,
+                                        699
+                                    ],
+                                    "score": 0.89,
+                                    "content": "\\mathbf { g } ^ { 1 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        368,
+                                        686,
+                                        385,
+                                        701
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        386,
+                                        687,
+                                        403,
+                                        698
+                                    ],
+                                    "score": 0.84,
+                                    "content": "\\dot { \\mathbf { W } } ^ { 1 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        403,
+                                        686,
+                                        506,
+                                        701
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to be the same across the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                698,
+                                506,
+                                712
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        698,
+                                        506,
+                                        712
+                                    ],
+                                    "score": 1.0,
+                                    "content": "20 copies of the environment (i.e. the learners start with the same user population), but these values",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                708,
+                                506,
+                                723
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        708,
+                                        506,
+                                        723
+                                    ],
+                                    "score": 1.0,
+                                    "content": "diverge throughout learning. For the environment, we set the number of user and article types both to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                720,
+                                480,
+                                734
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        720,
+                                        316,
+                                        734
+                                    ],
+                                    "score": 1.0,
+                                    "content": "10. Initial user loyalties are randomly sampled from",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        316,
+                                        720,
+                                        361,
+                                        732
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\mathcal { N } ( 0 , 0 . 0 3 )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        361,
+                                        720,
+                                        365,
+                                        734
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        365,
+                                        721,
+                                        408,
+                                        732
+                                    ],
+                                    "score": 0.89,
+                                    "content": "\\alpha _ { 1 } = 0 . 0 3",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        408,
+                                        720,
+                                        429,
+                                        734
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        429,
+                                        721,
+                                        477,
+                                        732
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\alpha _ { 2 } = 0 . 0 0 3",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        477,
+                                        720,
+                                        480,
+                                        734
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 47
+                        }
+                    ],
+                    "index": 43
+                }
+            ],
+            "page_idx": 18,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        300,
+                        751,
+                        311,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                299,
+                                750,
+                                312,
+                                764
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        299,
+                                        750,
+                                        312,
+                                        764
+                                    ],
+                                    "score": 1.0,
+                                    "content": "19",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        107,
+                        27,
+                        307,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                26,
+                                308,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        26,
+                                        308,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Under review as a conference paper at ICLR 2021",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        82,
+                        259,
+                        94
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                82,
+                                260,
+                                95
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        82,
+                                        260,
+                                        95
+                                    ],
+                                    "score": 1.0,
+                                    "content": "9.2 CONTENT RECOMMENDATION",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        }
+                    ],
+                    "index": 0
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        102,
+                        244,
+                        114
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                102,
+                                245,
+                                115
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        102,
+                                        245,
+                                        115
+                                    ],
+                                    "score": 1.0,
+                                    "content": "9.2.1 ENVIRONMENT DETAILS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        }
+                    ],
+                    "index": 1
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        122,
+                        295,
+                        134
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                120,
+                                297,
+                                136
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        120,
+                                        297,
+                                        136
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The evironment has the following components:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        }
+                    ],
+                    "index": 2,
+                    "bbox_fs": [
+                        105,
+                        120,
+                        297,
+                        136
+                    ]
+                },
+                {
+                    "type": "list",
+                    "bbox": [
+                        127,
+                        142,
+                        505,
+                        243
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                130,
+                                142,
+                                505,
+                                154
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        130,
+                                        142,
+                                        191,
+                                        154
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1. User type,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        191,
+                                        142,
+                                        201,
+                                        153
+                                    ],
+                                    "score": 0.86,
+                                    "content": "x ^ { t }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        201,
+                                        142,
+                                        505,
+                                        154
+                                    ],
+                                    "score": 1.0,
+                                    "content": ": categorical variable representing different types of users. The content",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                154,
+                                426,
+                                166
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        154,
+                                        426,
+                                        166
+                                    ],
+                                    "score": 1.0,
+                                    "content": "recommender conditions its predictions on the type of the current user.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                128,
+                                167,
+                                506,
+                                182
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        128,
+                                        167,
+                                        200,
+                                        182
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2. User loyalty,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        200,
+                                        168,
+                                        211,
+                                        180
+                                    ],
+                                    "score": 0.87,
+                                    "content": "\\mathbf { g } ^ { t }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        211,
+                                        167,
+                                        483,
+                                        182
+                                    ],
+                                    "score": 1.0,
+                                    "content": ": the propensity for users of each type to use the platform. User",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        483,
+                                        168,
+                                        493,
+                                        178
+                                    ],
+                                    "score": 0.86,
+                                    "content": "x ^ { t }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        494,
+                                        167,
+                                        506,
+                                        182
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                178,
+                                461,
+                                193
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        178,
+                                        439,
+                                        193
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sampled from a categorical distribution with parameters given by softmax",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        439,
+                                        179,
+                                        456,
+                                        192
+                                    ],
+                                    "score": 0.83,
+                                    "content": "( \\mathbf { g } ^ { t } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        457,
+                                        178,
+                                        461,
+                                        193
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                128,
+                                193,
+                                505,
+                                207
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        128,
+                                        193,
+                                        197,
+                                        207
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3. Article type,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        197,
+                                        194,
+                                        207,
+                                        205
+                                    ],
+                                    "score": 0.87,
+                                    "content": "y ^ { t }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        208,
+                                        193,
+                                        505,
+                                        207
+                                    ],
+                                    "score": 1.0,
+                                    "content": ": a categorical variable (one-hot encoding) representing the type of article",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                142,
+                                206,
+                                225,
+                                218
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        142,
+                                        206,
+                                        225,
+                                        218
+                                    ],
+                                    "score": 1.0,
+                                    "content": "selected by the user.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                216,
+                                507,
+                                237
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        216,
+                                        205,
+                                        237
+                                    ],
+                                    "score": 1.0,
+                                    "content": "4. User interests,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        205,
+                                        219,
+                                        222,
+                                        231
+                                    ],
+                                    "score": 0.82,
+                                    "content": "\\mathbf { W } ^ { t }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        222,
+                                        216,
+                                        317,
+                                        237
+                                    ],
+                                    "score": 1.0,
+                                    "content": ": a matrix whose entries",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        317,
+                                        220,
+                                        339,
+                                        234
+                                    ],
+                                    "score": 0.92,
+                                    "content": "W _ { x , y } ^ { t }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        339,
+                                        216,
+                                        507,
+                                        237
+                                    ],
+                                    "score": 1.0,
+                                    "content": "represent the average interest user of type",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                142,
+                                229,
+                                254,
+                                245
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        142,
+                                        233,
+                                        149,
+                                        241
+                                    ],
+                                    "score": 0.73,
+                                    "content": "x",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        150,
+                                        229,
+                                        243,
+                                        245
+                                    ],
+                                    "score": 1.0,
+                                    "content": "have in articles of type",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        244,
+                                        233,
+                                        250,
+                                        243
+                                    ],
+                                    "score": 0.78,
+                                    "content": "y",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        250,
+                                        229,
+                                        254,
+                                        245
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10,
+                            "is_list_end_line": true
+                        }
+                    ],
+                    "index": 6.5,
+                    "bbox_fs": [
+                        126,
+                        142,
+                        507,
+                        245
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        251,
+                        505,
+                        307
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                251,
+                                505,
+                                263
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        252,
+                                        178,
+                                        263
+                                    ],
+                                    "score": 1.0,
+                                    "content": "At each time step",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        178,
+                                        253,
+                                        183,
+                                        262
+                                    ],
+                                    "score": 0.68,
+                                    "content": "t",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        183,
+                                        252,
+                                        213,
+                                        263
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", a user",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        213,
+                                        251,
+                                        223,
+                                        261
+                                    ],
+                                    "score": 0.87,
+                                    "content": "x ^ { t }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        224,
+                                        252,
+                                        505,
+                                        263
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is sampled from a categorical distribution (based on the loyalty of the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                262,
+                                505,
+                                275
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        262,
+                                        505,
+                                        275
+                                    ],
+                                    "score": 1.0,
+                                    "content": "different user types), then the recommendation system selects which type of article to present in the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                272,
+                                506,
+                                286
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        272,
+                                        506,
+                                        286
+                                    ],
+                                    "score": 1.0,
+                                    "content": "top position, and finally, the user selects an article. The goal of the recommendation system is to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                283,
+                                506,
+                                297
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        283,
+                                        506,
+                                        297
+                                    ],
+                                    "score": 1.0,
+                                    "content": "predict the likelihood that the user would click on each of the available articles, in order to select the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                295,
+                                272,
+                                307
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        295,
+                                        272,
+                                        307
+                                    ],
+                                    "score": 1.0,
+                                    "content": "one which is most interesting to the user.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        }
+                    ],
+                    "index": 13,
+                    "bbox_fs": [
+                        105,
+                        251,
+                        506,
+                        307
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        312,
+                        505,
+                        378
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                309,
+                                506,
+                                326
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        309,
+                                        171,
+                                        326
+                                    ],
+                                    "score": 1.0,
+                                    "content": "User loyalty for",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        171,
+                                        312,
+                                        181,
+                                        322
+                                    ],
+                                    "score": 0.87,
+                                    "content": "x ^ { t }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        181,
+                                        309,
+                                        506,
+                                        326
+                                    ],
+                                    "score": 1.0,
+                                    "content": "then changes in accordance with the self-selection effect, increasing or decreasing",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                322,
+                                506,
+                                336
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        322,
+                                        418,
+                                        336
+                                    ],
+                                    "score": 1.0,
+                                    "content": "proportionally to their interest in the top article. The interests of user type",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        419,
+                                        323,
+                                        429,
+                                        333
+                                    ],
+                                    "score": 0.86,
+                                    "content": "x ^ { t }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        429,
+                                        322,
+                                        506,
+                                        336
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(represented by a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                333,
+                                506,
+                                347
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        333,
+                                        150,
+                                        347
+                                    ],
+                                    "score": 1.0,
+                                    "content": "column of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        150,
+                                        334,
+                                        166,
+                                        344
+                                    ],
+                                    "score": 0.81,
+                                    "content": "\\mathbf { W } ^ { t }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        167,
+                                        333,
+                                        506,
+                                        347
+                                    ],
+                                    "score": 1.0,
+                                    "content": ") also change; in accordance with the illusory truth effect, their interest in the topic of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                345,
+                                506,
+                                357
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        345,
+                                        506,
+                                        357
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the top article (as chosen by the recommender system) always increases. Overall, this environment",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                356,
+                                506,
+                                368
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        356,
+                                        506,
+                                        368
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is an extremely crude representation of reality, but it allows us to incorporate both the effects of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                366,
+                                432,
+                                379
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        366,
+                                        432,
+                                        379
+                                    ],
+                                    "score": 1.0,
+                                    "content": "self-selection (via covariate shift), and the illusory truth effect (via concept shift).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        }
+                    ],
+                    "index": 18.5,
+                    "bbox_fs": [
+                        104,
+                        309,
+                        506,
+                        379
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        383,
+                        506,
+                        428
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                384,
+                                506,
+                                396
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        384,
+                                        506,
+                                        396
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Formally, this environment is similar to a POMDP\\R, i.e. a POMDP with no reward function,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                395,
+                                505,
+                                407
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        395,
+                                        505,
+                                        407
+                                    ],
+                                    "score": 1.0,
+                                    "content": "also known as a world model (Armstrong & O’Rourke, 2017; Hadfield-Menell et al., 2017); the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                405,
+                                507,
+                                419
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        405,
+                                        507,
+                                        419
+                                    ],
+                                    "score": 1.0,
+                                    "content": "difference is that the learner observes the input before acting and only observes the target after acting.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                416,
+                                307,
+                                429
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        416,
+                                        307,
+                                        429
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The states, observations, and actions given below.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        }
+                    ],
+                    "index": 23.5,
+                    "bbox_fs": [
+                        105,
+                        384,
+                        507,
+                        429
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        235,
+                        431,
+                        376,
+                        464
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                235,
+                                431,
+                                376,
+                                464
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        235,
+                                        431,
+                                        376,
+                                        464
+                                    ],
+                                    "score": 0.84,
+                                    "content": "\\begin{array} { c } { { \\boldsymbol { s } ^ { t } = ( \\mathbf { g } ^ { t } , \\mathbf { W } ^ { t } , \\boldsymbol { x } ^ { t } , \\boldsymbol { y } ^ { t } ) } } \\\\ { { \\boldsymbol { o } _ { \\mathrm { p r e } } ^ { t } , ~ \\boldsymbol { a } ^ { t } , ~ \\boldsymbol { o } _ { \\mathrm { p o s t } } ^ { t } = ( \\boldsymbol { x } ^ { t } , \\hat { \\boldsymbol { y } } ^ { t } , \\boldsymbol { y } ^ { t } ) } } \\end{array}",
+                                    "type": "interline_equation",
+                                    "image_path": "db0b38546afafe534e88febd9a237072bd24f328ca4cb11931aed443be558f9d.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 26.5,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                235,
+                                431,
+                                376,
+                                447.5
+                            ],
+                            "spans": [],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                235,
+                                447.5,
+                                376,
+                                464.0
+                            ],
+                            "spans": [],
+                            "index": 27
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        466,
+                        276,
+                        478
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                464,
+                                277,
+                                480
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        464,
+                                        277,
+                                        480
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The state transition function is defined by:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        }
+                    ],
+                    "index": 28,
+                    "bbox_fs": [
+                        105,
+                        464,
+                        277,
+                        480
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        201,
+                        480,
+                        409,
+                        564
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                201,
+                                480,
+                                409,
+                                564
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        201,
+                                        480,
+                                        409,
+                                        564
+                                    ],
+                                    "score": 0.94,
+                                    "content": "\\begin{array} { r l r } { \\mathbf { g } _ { x ^ { t } } ^ { t + 1 } = \\mathbf { g } _ { x ^ { t } } ^ { t } + \\alpha _ { 1 } W _ { x ^ { t } , \\hat { y } ^ { t } } ^ { t } } & { } & \\\\ { \\mathbf { W } _ { x ^ { t } , \\hat { y } ^ { t } } ^ { t + 1 / 2 } = W _ { x ^ { t } , \\hat { y } ^ { t } } ^ { t } + \\alpha _ { 2 } ; \\quad \\mathbf { W } _ { x ^ { t } } ^ { t + 1 } = \\frac { \\mathbf { W } _ { x ^ { t } } ^ { t + 1 / 2 } } { \\Vert \\mathbf { W } _ { x ^ { t } } ^ { t + 1 / 2 } \\Vert _ { 2 } } } & { } & \\\\ { x ^ { t + 1 } \\sim \\mathrm { s o f t m a x } ( \\mathbf { g } ^ { t + 1 } ) } & { } & \\\\ { y ^ { t + 1 } \\sim \\mathrm { s o f t m a x } ( \\mathbf { W } _ { x ^ { t + 1 } } ^ { t + 1 } ) } & { } & \\end{array}",
+                                    "type": "interline_equation",
+                                    "image_path": "18371f2cadc36f7f8a1bbad2aa2e9888bab73b266fa7d957e9964ff1779ada1c.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 31.5,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                201,
+                                480,
+                                409,
+                                494.0
+                            ],
+                            "spans": [],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                201,
+                                494.0,
+                                409,
+                                508.0
+                            ],
+                            "spans": [],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                201,
+                                508.0,
+                                409,
+                                522.0
+                            ],
+                            "spans": [],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                201,
+                                522.0,
+                                409,
+                                536.0
+                            ],
+                            "spans": [],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                201,
+                                536.0,
+                                409,
+                                550.0
+                            ],
+                            "spans": [],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                201,
+                                550.0,
+                                409,
+                                564.0
+                            ],
+                            "spans": [],
+                            "index": 34
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        568,
+                        504,
+                        602
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                567,
+                                506,
+                                581
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        567,
+                                        135,
+                                        581
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Where",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        135,
+                                        568,
+                                        145,
+                                        580
+                                    ],
+                                    "score": 0.88,
+                                    "content": "\\hat { y } ^ { t }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        145,
+                                        567,
+                                        356,
+                                        581
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is the top article as chosen by the recommender, and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        356,
+                                        570,
+                                        367,
+                                        579
+                                    ],
+                                    "score": 0.76,
+                                    "content": "\\alpha _ { 1 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        368,
+                                        567,
+                                        371,
+                                        581
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        372,
+                                        570,
+                                        384,
+                                        579
+                                    ],
+                                    "score": 0.72,
+                                    "content": "\\alpha _ { 2 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        384,
+                                        567,
+                                        506,
+                                        581
+                                    ],
+                                    "score": 1.0,
+                                    "content": "represent the rate of covariate",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                577,
+                                504,
+                                592
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        577,
+                                        299,
+                                        592
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and concept shift (respectively). The update for",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        299,
+                                        578,
+                                        326,
+                                        590
+                                    ],
+                                    "score": 0.89,
+                                    "content": "\\mathbf { W } ^ { t + 1 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        326,
+                                        577,
+                                        493,
+                                        592
+                                    ],
+                                    "score": 1.0,
+                                    "content": "merely increases the interest of user type",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        493,
+                                        579,
+                                        504,
+                                        589
+                                    ],
+                                    "score": 0.84,
+                                    "content": "x ^ { t }",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                589,
+                                366,
+                                603
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        589,
+                                        164,
+                                        603
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in article type",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        164,
+                                        590,
+                                        173,
+                                        601
+                                    ],
+                                    "score": 0.88,
+                                    "content": "\\hat { y } ^ { t }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        174,
+                                        589,
+                                        366,
+                                        603
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", then normalizes the interests for that user type.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        }
+                    ],
+                    "index": 36,
+                    "bbox_fs": [
+                        105,
+                        567,
+                        506,
+                        603
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        613,
+                        258,
+                        625
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                614,
+                                259,
+                                626
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        614,
+                                        259,
+                                        626
+                                    ],
+                                    "score": 1.0,
+                                    "content": "9.2.2 REPRODUCIBILITY DETAILS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        }
+                    ],
+                    "index": 38
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        632,
+                        506,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                632,
+                                506,
+                                646
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        632,
+                                        506,
+                                        646
+                                    ],
+                                    "score": 1.0,
+                                    "content": "For these experiments, the recommendation system is a ReLU-MLP with 1 hidden layer of 100 units,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                644,
+                                505,
+                                655
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        644,
+                                        331,
+                                        655
+                                    ],
+                                    "score": 1.0,
+                                    "content": "trained via supervised learning with SGD (learning rate",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        332,
+                                        644,
+                                        361,
+                                        655
+                                    ],
+                                    "score": 0.83,
+                                    "content": "= 0 . 0 1 ",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        361,
+                                        644,
+                                        505,
+                                        655
+                                    ],
+                                    "score": 1.0,
+                                    "content": ") to predict which article a user will",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                653,
+                                506,
+                                669
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        653,
+                                        506,
+                                        669
+                                    ],
+                                    "score": 1.0,
+                                    "content": "select. Actions are sampled from the MLP’s predictive distribution. We apply PBT without any",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                666,
+                                506,
+                                678
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        666,
+                                        506,
+                                        678
+                                    ],
+                                    "score": 1.0,
+                                    "content": "hyperparameter selection (this amounts to just doing the EXPLOIT step), and an interval of 10,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                677,
+                                505,
+                                689
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        677,
+                                        505,
+                                        689
+                                    ],
+                                    "score": 1.0,
+                                    "content": "selecting on accuracy. We use a population of 20 learners (whether applying PBT or not), and match",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                686,
+                                506,
+                                701
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        686,
+                                        356,
+                                        701
+                                    ],
+                                    "score": 1.0,
+                                    "content": "random seeds for the trials with and without PBT. We initialize",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        356,
+                                        687,
+                                        368,
+                                        699
+                                    ],
+                                    "score": 0.89,
+                                    "content": "\\mathbf { g } ^ { 1 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        368,
+                                        686,
+                                        385,
+                                        701
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        386,
+                                        687,
+                                        403,
+                                        698
+                                    ],
+                                    "score": 0.84,
+                                    "content": "\\dot { \\mathbf { W } } ^ { 1 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        403,
+                                        686,
+                                        506,
+                                        701
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to be the same across the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                698,
+                                506,
+                                712
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        698,
+                                        506,
+                                        712
+                                    ],
+                                    "score": 1.0,
+                                    "content": "20 copies of the environment (i.e. the learners start with the same user population), but these values",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                708,
+                                506,
+                                723
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        708,
+                                        506,
+                                        723
+                                    ],
+                                    "score": 1.0,
+                                    "content": "diverge throughout learning. For the environment, we set the number of user and article types both to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                720,
+                                480,
+                                734
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        720,
+                                        316,
+                                        734
+                                    ],
+                                    "score": 1.0,
+                                    "content": "10. Initial user loyalties are randomly sampled from",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        316,
+                                        720,
+                                        361,
+                                        732
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\mathcal { N } ( 0 , 0 . 0 3 )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        361,
+                                        720,
+                                        365,
+                                        734
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        365,
+                                        721,
+                                        408,
+                                        732
+                                    ],
+                                    "score": 0.89,
+                                    "content": "\\alpha _ { 1 } = 0 . 0 3",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        408,
+                                        720,
+                                        429,
+                                        734
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        429,
+                                        721,
+                                        477,
+                                        732
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\alpha _ { 2 } = 0 . 0 0 3",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        477,
+                                        720,
+                                        480,
+                                        734
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 47
+                        }
+                    ],
+                    "index": 43,
+                    "bbox_fs": [
+                        104,
+                        632,
+                        506,
+                        734
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        82,
+                        248,
+                        93
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                81,
+                                250,
+                                95
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        81,
+                                        250,
+                                        95
+                                    ],
+                                    "score": 1.0,
+                                    "content": "9.2.3 DETAILS OF EVALUATION",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        }
+                    ],
+                    "index": 0
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        101,
+                        505,
+                        189
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                101,
+                                506,
+                                115
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        101,
+                                        254,
+                                        115
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We measure concept shift (change in",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        254,
+                                        101,
+                                        286,
+                                        114
+                                    ],
+                                    "score": 0.9,
+                                    "content": "P ( \\boldsymbol { y } | \\mathbf { x } ) .",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        286,
+                                        101,
+                                        506,
+                                        115
+                                    ],
+                                    "score": 1.0,
+                                    "content": ") as the cosine distance between each user types’ initial",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                113,
+                                505,
+                                126
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        113,
+                                        390,
+                                        126
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and current interest vectors. And we measure covariate shift (change in",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        390,
+                                        113,
+                                        415,
+                                        124
+                                    ],
+                                    "score": 0.86,
+                                    "content": "P ( \\mathbf { x } ) )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        415,
+                                        113,
+                                        505,
+                                        126
+                                    ],
+                                    "score": 1.0,
+                                    "content": ") as the KL-divergence",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                123,
+                                506,
+                                137
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        123,
+                                        374,
+                                        137
+                                    ],
+                                    "score": 1.0,
+                                    "content": "between the current and initial user distributions, parametrized by",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        375,
+                                        123,
+                                        387,
+                                        135
+                                    ],
+                                    "score": 0.88,
+                                    "content": "\\mathbf { g } ^ { 1 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        387,
+                                        123,
+                                        404,
+                                        137
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        405,
+                                        124,
+                                        415,
+                                        135
+                                    ],
+                                    "score": 0.88,
+                                    "content": "\\mathbf { g } ^ { t }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        416,
+                                        123,
+                                        506,
+                                        137
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", respectively. Results",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                135,
+                                506,
+                                147
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        135,
+                                        506,
+                                        147
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are presented in 7 (main text). In Figure 12, we additionally plot concept shift and covariate shift as a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                146,
+                                505,
+                                158
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        146,
+                                        505,
+                                        158
+                                    ],
+                                    "score": 1.0,
+                                    "content": "function of accuracy. We observe that for both types of ADS, at low levels of accuracy PBT actually",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                156,
+                                506,
+                                169
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        156,
+                                        483,
+                                        169
+                                    ],
+                                    "score": 1.0,
+                                    "content": "causes less shift than occur in baseline agents; HI-ADS are only observed for accuracies above",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        483,
+                                        156,
+                                        503,
+                                        167
+                                    ],
+                                    "score": 0.88,
+                                    "content": "60 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        503,
+                                        156,
+                                        506,
+                                        169
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                167,
+                                506,
+                                181
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        167,
+                                        506,
+                                        181
+                                    ],
+                                    "score": 1.0,
+                                    "content": "This suggests that only relatively strong performers are able to pick up on the HI-ADS revealed by",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                177,
+                                168,
+                                191
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        177,
+                                        168,
+                                        191
+                                    ],
+                                    "score": 1.0,
+                                    "content": "PBT (Fig. 12).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        }
+                    ],
+                    "index": 4.5
+                },
+                {
+                    "type": "image",
+                    "bbox": [
+                        145,
+                        203,
+                        453,
+                        340
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                145,
+                                203,
+                                453,
+                                340
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        145,
+                                        203,
+                                        453,
+                                        340
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                145,
+                                                203,
+                                                453,
+                                                340
+                                            ],
+                                            "score": 0.971,
+                                            "type": "image",
+                                            "image_path": "4ee83ea25b84d00e9bd2c93c72bcfe37100fc4c406703659c71aeb8e4ba6b65a.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 10,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        145,
+                                        203,
+                                        453,
+                                        248.66666666666666
+                                    ],
+                                    "spans": [],
+                                    "index": 9
+                                },
+                                {
+                                    "bbox": [
+                                        145,
+                                        248.66666666666666,
+                                        453,
+                                        294.3333333333333
+                                    ],
+                                    "spans": [],
+                                    "index": 10
+                                },
+                                {
+                                    "bbox": [
+                                        145,
+                                        294.3333333333333,
+                                        453,
+                                        340.0
+                                    ],
+                                    "spans": [],
+                                    "index": 11
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                106,
+                                352,
+                                504,
+                                386
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        352,
+                                        506,
+                                        366
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                352,
+                                                506,
+                                                366
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 12: Amount of auto-induced covariate shift (left) and auto-induced concept shift (right) as a",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 12
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        363,
+                                        505,
+                                        377
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                363,
+                                                505,
+                                                377
+                                            ],
+                                            "score": 1.0,
+                                            "content": "function of performance (accuracy) averaged over all trials, learners, and time-steps. Only relatively",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 13
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        375,
+                                        393,
+                                        387
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                375,
+                                                291,
+                                                387
+                                            ],
+                                            "score": 1.0,
+                                            "content": "strong learners (those which achieve accuracy",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                291,
+                                                375,
+                                                320,
+                                                385
+                                            ],
+                                            "score": 0.87,
+                                            "content": "> 6 0 \\%",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                320,
+                                                375,
+                                                393,
+                                                387
+                                            ],
+                                            "score": 1.0,
+                                            "content": ") exhibit HI-ADS.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 14
+                                }
+                            ],
+                            "index": 13
+                        }
+                    ],
+                    "index": 11.5
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        405,
+                        367,
+                        415
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                403,
+                                369,
+                                417
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        403,
+                                        369,
+                                        417
+                                    ],
+                                    "score": 1.0,
+                                    "content": "9.2.4 CONTEXT SWAPPING IN CONTENT RECOMMENDATION",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        }
+                    ],
+                    "index": 15
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        423,
+                        505,
+                        478
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                422,
+                                506,
+                                436
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        422,
+                                        506,
+                                        436
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We believe context swapping is not appropriate for the content recommendation environment, since",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                433,
+                                505,
+                                447
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        433,
+                                        505,
+                                        447
+                                    ],
+                                    "score": 1.0,
+                                    "content": "when the environments diverge, optimal behavior may differ across environments. Nevertheless, we",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                445,
+                                505,
+                                458
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        445,
+                                        505,
+                                        458
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ran experiments with it for completeness. The main effect appears to be to hamper learning when",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                455,
+                                506,
+                                469
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        455,
+                                        506,
+                                        469
+                                    ],
+                                    "score": 1.0,
+                                    "content": "PBT is not used, see Figure 13. Notably, it does not appear to significantly influence the rate or",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                467,
+                                275,
+                                479
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        467,
+                                        275,
+                                        479
+                                    ],
+                                    "score": 1.0,
+                                    "content": "extent of ADS when combined with PBT.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        }
+                    ],
+                    "index": 18
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        106,
+                        491,
+                        341,
+                        502
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                491,
+                                342,
+                                503
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        491,
+                                        342,
+                                        503
+                                    ],
+                                    "score": 1.0,
+                                    "content": "9.2.5 EXPLORATION OF ENVIRONMENT PARAMETERS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        }
+                    ],
+                    "index": 21
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        510,
+                        505,
+                        599
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                509,
+                                506,
+                                523
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        509,
+                                        405,
+                                        523
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In Figure 14, we examine the effect of the rate-of-change parameters",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        405,
+                                        511,
+                                        441,
+                                        522
+                                    ],
+                                    "score": 0.87,
+                                    "content": "( \\alpha _ { 1 } , \\alpha _ { 2 } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        441,
+                                        509,
+                                        506,
+                                        523
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of the content",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                521,
+                                506,
+                                533
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        521,
+                                        506,
+                                        533
+                                    ],
+                                    "score": 1.0,
+                                    "content": "recommendation environment on the results provided in the paper. As noted there, our results are",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                532,
+                                506,
+                                545
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        532,
+                                        506,
+                                        545
+                                    ],
+                                    "score": 1.0,
+                                    "content": "qualitatively consistent so long as (1) the initial user distribution is approximately uniform, and (2)",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                543,
+                                506,
+                                555
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        543,
+                                        198,
+                                        555
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the covariate shift rate",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        198,
+                                        543,
+                                        216,
+                                        554
+                                    ],
+                                    "score": 0.81,
+                                    "content": "( \\alpha _ { 1 } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        216,
+                                        543,
+                                        356,
+                                        555
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is faster than the concept shift rate",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        356,
+                                        543,
+                                        374,
+                                        554
+                                    ],
+                                    "score": 0.78,
+                                    "content": "\\left( \\alpha _ { 2 } \\right)",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        374,
+                                        543,
+                                        506,
+                                        555
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". These distributions are updated",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                553,
+                                506,
+                                567
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        553,
+                                        506,
+                                        567
+                                    ],
+                                    "score": 1.0,
+                                    "content": "by different mechanisms, and are not directly comparable. Concept shift changes the task more",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                564,
+                                506,
+                                578
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        564,
+                                        506,
+                                        578
+                                    ],
+                                    "score": 1.0,
+                                    "content": "radically, requiring a learner to change its predictions, rather than just become accurate on a wider",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                576,
+                                506,
+                                589
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        576,
+                                        291,
+                                        589
+                                    ],
+                                    "score": 1.0,
+                                    "content": "range of inputs. We conjecture that changes in",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        291,
+                                        576,
+                                        322,
+                                        588
+                                    ],
+                                    "score": 0.92,
+                                    "content": "P ( \\boldsymbol { y } | \\boldsymbol { x } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        322,
+                                        576,
+                                        506,
+                                        589
+                                    ],
+                                    "score": 1.0,
+                                    "content": "must therefore be kept smooth enough for the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                587,
+                                318,
+                                599
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        587,
+                                        318,
+                                        599
+                                    ],
+                                    "score": 1.0,
+                                    "content": "outer loop to have pressure to capitalize on HI-ADS.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        }
+                    ],
+                    "index": 25.5
+                }
+            ],
+            "page_idx": 19,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        107,
+                        27,
+                        307,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                25,
+                                308,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        25,
+                                        308,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Under review as a conference paper at ICLR 2021",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        300,
+                        751,
+                        311,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                298,
+                                749,
+                                313,
+                                764
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        298,
+                                        749,
+                                        313,
+                                        764
+                                    ],
+                                    "score": 1.0,
+                                    "content": "",
+                                    "type": "text",
+                                    "height": 15,
+                                    "width": 15
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        82,
+                        248,
+                        93
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                81,
+                                250,
+                                95
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        81,
+                                        250,
+                                        95
+                                    ],
+                                    "score": 1.0,
+                                    "content": "9.2.3 DETAILS OF EVALUATION",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        }
+                    ],
+                    "index": 0
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        101,
+                        505,
+                        189
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                101,
+                                506,
+                                115
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        101,
+                                        254,
+                                        115
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We measure concept shift (change in",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        254,
+                                        101,
+                                        286,
+                                        114
+                                    ],
+                                    "score": 0.9,
+                                    "content": "P ( \\boldsymbol { y } | \\mathbf { x } ) .",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        286,
+                                        101,
+                                        506,
+                                        115
+                                    ],
+                                    "score": 1.0,
+                                    "content": ") as the cosine distance between each user types’ initial",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                113,
+                                505,
+                                126
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        113,
+                                        390,
+                                        126
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and current interest vectors. And we measure covariate shift (change in",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        390,
+                                        113,
+                                        415,
+                                        124
+                                    ],
+                                    "score": 0.86,
+                                    "content": "P ( \\mathbf { x } ) )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        415,
+                                        113,
+                                        505,
+                                        126
+                                    ],
+                                    "score": 1.0,
+                                    "content": ") as the KL-divergence",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                123,
+                                506,
+                                137
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        123,
+                                        374,
+                                        137
+                                    ],
+                                    "score": 1.0,
+                                    "content": "between the current and initial user distributions, parametrized by",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        375,
+                                        123,
+                                        387,
+                                        135
+                                    ],
+                                    "score": 0.88,
+                                    "content": "\\mathbf { g } ^ { 1 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        387,
+                                        123,
+                                        404,
+                                        137
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        405,
+                                        124,
+                                        415,
+                                        135
+                                    ],
+                                    "score": 0.88,
+                                    "content": "\\mathbf { g } ^ { t }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        416,
+                                        123,
+                                        506,
+                                        137
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", respectively. Results",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                135,
+                                506,
+                                147
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        135,
+                                        506,
+                                        147
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are presented in 7 (main text). In Figure 12, we additionally plot concept shift and covariate shift as a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                146,
+                                505,
+                                158
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        146,
+                                        505,
+                                        158
+                                    ],
+                                    "score": 1.0,
+                                    "content": "function of accuracy. We observe that for both types of ADS, at low levels of accuracy PBT actually",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                156,
+                                506,
+                                169
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        156,
+                                        483,
+                                        169
+                                    ],
+                                    "score": 1.0,
+                                    "content": "causes less shift than occur in baseline agents; HI-ADS are only observed for accuracies above",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        483,
+                                        156,
+                                        503,
+                                        167
+                                    ],
+                                    "score": 0.88,
+                                    "content": "60 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        503,
+                                        156,
+                                        506,
+                                        169
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                167,
+                                506,
+                                181
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        167,
+                                        506,
+                                        181
+                                    ],
+                                    "score": 1.0,
+                                    "content": "This suggests that only relatively strong performers are able to pick up on the HI-ADS revealed by",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                177,
+                                168,
+                                191
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        177,
+                                        168,
+                                        191
+                                    ],
+                                    "score": 1.0,
+                                    "content": "PBT (Fig. 12).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        }
+                    ],
+                    "index": 4.5,
+                    "bbox_fs": [
+                        105,
+                        101,
+                        506,
+                        191
+                    ]
+                },
+                {
+                    "type": "image",
+                    "bbox": [
+                        145,
+                        203,
+                        453,
+                        340
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                145,
+                                203,
+                                453,
+                                340
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        145,
+                                        203,
+                                        453,
+                                        340
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                145,
+                                                203,
+                                                453,
+                                                340
+                                            ],
+                                            "score": 0.971,
+                                            "type": "image",
+                                            "image_path": "4ee83ea25b84d00e9bd2c93c72bcfe37100fc4c406703659c71aeb8e4ba6b65a.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 10,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        145,
+                                        203,
+                                        453,
+                                        248.66666666666666
+                                    ],
+                                    "spans": [],
+                                    "index": 9
+                                },
+                                {
+                                    "bbox": [
+                                        145,
+                                        248.66666666666666,
+                                        453,
+                                        294.3333333333333
+                                    ],
+                                    "spans": [],
+                                    "index": 10
+                                },
+                                {
+                                    "bbox": [
+                                        145,
+                                        294.3333333333333,
+                                        453,
+                                        340.0
+                                    ],
+                                    "spans": [],
+                                    "index": 11
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                106,
+                                352,
+                                504,
+                                386
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        352,
+                                        506,
+                                        366
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                352,
+                                                506,
+                                                366
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 12: Amount of auto-induced covariate shift (left) and auto-induced concept shift (right) as a",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 12
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        363,
+                                        505,
+                                        377
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                363,
+                                                505,
+                                                377
+                                            ],
+                                            "score": 1.0,
+                                            "content": "function of performance (accuracy) averaged over all trials, learners, and time-steps. Only relatively",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 13
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        375,
+                                        393,
+                                        387
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                375,
+                                                291,
+                                                387
+                                            ],
+                                            "score": 1.0,
+                                            "content": "strong learners (those which achieve accuracy",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                291,
+                                                375,
+                                                320,
+                                                385
+                                            ],
+                                            "score": 0.87,
+                                            "content": "> 6 0 \\%",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                320,
+                                                375,
+                                                393,
+                                                387
+                                            ],
+                                            "score": 1.0,
+                                            "content": ") exhibit HI-ADS.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 14
+                                }
+                            ],
+                            "index": 13
+                        }
+                    ],
+                    "index": 11.5
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        405,
+                        367,
+                        415
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                403,
+                                369,
+                                417
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        403,
+                                        369,
+                                        417
+                                    ],
+                                    "score": 1.0,
+                                    "content": "9.2.4 CONTEXT SWAPPING IN CONTENT RECOMMENDATION",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        }
+                    ],
+                    "index": 15
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        423,
+                        505,
+                        478
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                422,
+                                506,
+                                436
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        422,
+                                        506,
+                                        436
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We believe context swapping is not appropriate for the content recommendation environment, since",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                433,
+                                505,
+                                447
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        433,
+                                        505,
+                                        447
+                                    ],
+                                    "score": 1.0,
+                                    "content": "when the environments diverge, optimal behavior may differ across environments. Nevertheless, we",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                445,
+                                505,
+                                458
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        445,
+                                        505,
+                                        458
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ran experiments with it for completeness. The main effect appears to be to hamper learning when",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                455,
+                                506,
+                                469
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        455,
+                                        506,
+                                        469
+                                    ],
+                                    "score": 1.0,
+                                    "content": "PBT is not used, see Figure 13. Notably, it does not appear to significantly influence the rate or",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                467,
+                                275,
+                                479
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        467,
+                                        275,
+                                        479
+                                    ],
+                                    "score": 1.0,
+                                    "content": "extent of ADS when combined with PBT.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        }
+                    ],
+                    "index": 18,
+                    "bbox_fs": [
+                        105,
+                        422,
+                        506,
+                        479
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        106,
+                        491,
+                        341,
+                        502
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                491,
+                                342,
+                                503
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        491,
+                                        342,
+                                        503
+                                    ],
+                                    "score": 1.0,
+                                    "content": "9.2.5 EXPLORATION OF ENVIRONMENT PARAMETERS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        }
+                    ],
+                    "index": 21
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        510,
+                        505,
+                        599
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                509,
+                                506,
+                                523
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        509,
+                                        405,
+                                        523
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In Figure 14, we examine the effect of the rate-of-change parameters",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        405,
+                                        511,
+                                        441,
+                                        522
+                                    ],
+                                    "score": 0.87,
+                                    "content": "( \\alpha _ { 1 } , \\alpha _ { 2 } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        441,
+                                        509,
+                                        506,
+                                        523
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of the content",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                521,
+                                506,
+                                533
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        521,
+                                        506,
+                                        533
+                                    ],
+                                    "score": 1.0,
+                                    "content": "recommendation environment on the results provided in the paper. As noted there, our results are",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                532,
+                                506,
+                                545
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        532,
+                                        506,
+                                        545
+                                    ],
+                                    "score": 1.0,
+                                    "content": "qualitatively consistent so long as (1) the initial user distribution is approximately uniform, and (2)",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                543,
+                                506,
+                                555
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        543,
+                                        198,
+                                        555
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the covariate shift rate",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        198,
+                                        543,
+                                        216,
+                                        554
+                                    ],
+                                    "score": 0.81,
+                                    "content": "( \\alpha _ { 1 } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        216,
+                                        543,
+                                        356,
+                                        555
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is faster than the concept shift rate",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        356,
+                                        543,
+                                        374,
+                                        554
+                                    ],
+                                    "score": 0.78,
+                                    "content": "\\left( \\alpha _ { 2 } \\right)",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        374,
+                                        543,
+                                        506,
+                                        555
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". These distributions are updated",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                553,
+                                506,
+                                567
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        553,
+                                        506,
+                                        567
+                                    ],
+                                    "score": 1.0,
+                                    "content": "by different mechanisms, and are not directly comparable. Concept shift changes the task more",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                564,
+                                506,
+                                578
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        564,
+                                        506,
+                                        578
+                                    ],
+                                    "score": 1.0,
+                                    "content": "radically, requiring a learner to change its predictions, rather than just become accurate on a wider",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                576,
+                                506,
+                                589
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        576,
+                                        291,
+                                        589
+                                    ],
+                                    "score": 1.0,
+                                    "content": "range of inputs. We conjecture that changes in",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        291,
+                                        576,
+                                        322,
+                                        588
+                                    ],
+                                    "score": 0.92,
+                                    "content": "P ( \\boldsymbol { y } | \\boldsymbol { x } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        322,
+                                        576,
+                                        506,
+                                        589
+                                    ],
+                                    "score": 1.0,
+                                    "content": "must therefore be kept smooth enough for the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                587,
+                                318,
+                                599
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        587,
+                                        318,
+                                        599
+                                    ],
+                                    "score": 1.0,
+                                    "content": "outer loop to have pressure to capitalize on HI-ADS.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        }
+                    ],
+                    "index": 25.5,
+                    "bbox_fs": [
+                        105,
+                        509,
+                        506,
+                        599
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "image",
+                    "bbox": [
+                        123,
+                        257,
+                        488,
+                        523
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                123,
+                                257,
+                                488,
+                                523
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        123,
+                                        257,
+                                        488,
+                                        523
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                123,
+                                                257,
+                                                488,
+                                                523
+                                            ],
+                                            "score": 0.976,
+                                            "type": "image",
+                                            "image_path": "32bff10367ed93273e67c56861072a16c782eb88536e394129f2eaa0da72bbf8.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 1,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        123,
+                                        257,
+                                        488,
+                                        345.6666666666667
+                                    ],
+                                    "spans": [],
+                                    "index": 0
+                                },
+                                {
+                                    "bbox": [
+                                        123,
+                                        345.6666666666667,
+                                        488,
+                                        434.33333333333337
+                                    ],
+                                    "spans": [],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        123,
+                                        434.33333333333337,
+                                        488,
+                                        523.0
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                105,
+                                537,
+                                505,
+                                560
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        536,
+                                        506,
+                                        551
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                536,
+                                                506,
+                                                551
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 13: Context swapping doesn’t have the desired effect in the content recommendation",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 3
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        548,
+                                        162,
+                                        560
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                548,
+                                                162,
+                                                560
+                                            ],
+                                            "score": 1.0,
+                                            "content": "environment.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 4
+                                }
+                            ],
+                            "index": 3.5
+                        }
+                    ],
+                    "index": 2.25
+                }
+            ],
+            "page_idx": 20,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        300,
+                        751,
+                        310,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                298,
+                                750,
+                                312,
+                                764
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        298,
+                                        750,
+                                        312,
+                                        764
+                                    ],
+                                    "score": 1.0,
+                                    "content": "",
+                                    "type": "text",
+                                    "height": 14,
+                                    "width": 14
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        106,
+                        27,
+                        307,
+                        38
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                107,
+                                26,
+                                308,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        26,
+                                        308,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Under review as a conference paper at ICLR 2021",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "image",
+                    "bbox": [
+                        123,
+                        257,
+                        488,
+                        523
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                123,
+                                257,
+                                488,
+                                523
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        123,
+                                        257,
+                                        488,
+                                        523
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                123,
+                                                257,
+                                                488,
+                                                523
+                                            ],
+                                            "score": 0.976,
+                                            "type": "image",
+                                            "image_path": "32bff10367ed93273e67c56861072a16c782eb88536e394129f2eaa0da72bbf8.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 1,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        123,
+                                        257,
+                                        488,
+                                        345.6666666666667
+                                    ],
+                                    "spans": [],
+                                    "index": 0
+                                },
+                                {
+                                    "bbox": [
+                                        123,
+                                        345.6666666666667,
+                                        488,
+                                        434.33333333333337
+                                    ],
+                                    "spans": [],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        123,
+                                        434.33333333333337,
+                                        488,
+                                        523.0
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                105,
+                                537,
+                                505,
+                                560
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        536,
+                                        506,
+                                        551
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                536,
+                                                506,
+                                                551
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 13: Context swapping doesn’t have the desired effect in the content recommendation",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 3
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        548,
+                                        162,
+                                        560
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                548,
+                                                162,
+                                                560
+                                            ],
+                                            "score": 1.0,
+                                            "content": "environment.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 4
+                                }
+                            ],
+                            "index": 3.5
+                        }
+                    ],
+                    "index": 2.25
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "image",
+                    "bbox": [
+                        145,
+                        56,
+                        466,
+                        710
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                145,
+                                56,
+                                466,
+                                710
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        145,
+                                        56,
+                                        466,
+                                        710
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                145,
+                                                56,
+                                                466,
+                                                710
+                                            ],
+                                            "score": 0.951,
+                                            "type": "image",
+                                            "image_path": "a89f27f3d6a426966da5d92bf05f995060ebcc886850b49f7d5327641103133d.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 1,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        145,
+                                        56,
+                                        466,
+                                        274.0
+                                    ],
+                                    "spans": [],
+                                    "index": 0
+                                },
+                                {
+                                    "bbox": [
+                                        145,
+                                        274.0,
+                                        466,
+                                        492.0
+                                    ],
+                                    "spans": [],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        145,
+                                        492.0,
+                                        466,
+                                        710.0
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                153,
+                                714,
+                                453,
+                                726
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        152,
+                                        712,
+                                        450,
+                                        729
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                152,
+                                                712,
+                                                422,
+                                                729
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 14: Content recommendation results for different values of",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                423,
+                                                717,
+                                                450,
+                                                726
+                                            ],
+                                            "score": 0.26,
+                                            "content": "\\alpha _ { 1 } , \\alpha _ { 2 }",
+                                            "type": "inline_equation"
+                                        }
+                                    ],
+                                    "index": 3
+                                }
+                            ],
+                            "index": 3
+                        }
+                    ],
+                    "index": 2.0
+                }
+            ],
+            "page_idx": 21,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        300,
+                        751,
+                        311,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                298,
+                                750,
+                                313,
+                                765
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        298,
+                                        750,
+                                        313,
+                                        765
+                                    ],
+                                    "score": 1.0,
+                                    "content": "",
+                                    "type": "text",
+                                    "height": 15,
+                                    "width": 15
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        107,
+                        27,
+                        306,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                26,
+                                308,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        26,
+                                        308,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Under review as a conference paper at ICLR 2021",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "image",
+                    "bbox": [
+                        145,
+                        56,
+                        466,
+                        710
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                145,
+                                56,
+                                466,
+                                710
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        145,
+                                        56,
+                                        466,
+                                        710
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                145,
+                                                56,
+                                                466,
+                                                710
+                                            ],
+                                            "score": 0.951,
+                                            "type": "image",
+                                            "image_path": "a89f27f3d6a426966da5d92bf05f995060ebcc886850b49f7d5327641103133d.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 1,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        145,
+                                        56,
+                                        466,
+                                        274.0
+                                    ],
+                                    "spans": [],
+                                    "index": 0
+                                },
+                                {
+                                    "bbox": [
+                                        145,
+                                        274.0,
+                                        466,
+                                        492.0
+                                    ],
+                                    "spans": [],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        145,
+                                        492.0,
+                                        466,
+                                        710.0
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                153,
+                                714,
+                                453,
+                                726
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        152,
+                                        712,
+                                        450,
+                                        729
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                152,
+                                                712,
+                                                422,
+                                                729
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 14: Content recommendation results for different values of",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                423,
+                                                717,
+                                                450,
+                                                726
+                                            ],
+                                            "score": 0.26,
+                                            "content": "\\alpha _ { 1 } , \\alpha _ { 2 }",
+                                            "type": "inline_equation"
+                                        }
+                                    ],
+                                    "index": 3
+                                }
+                            ],
+                            "index": 3
+                        }
+                    ],
+                    "index": 2.0
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        81,
+                        434,
+                        94
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                78,
+                                436,
+                                99
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        78,
+                                        436,
+                                        99
+                                    ],
+                                    "score": 1.0,
+                                    "content": "10 OFFLINE Q-LEARNING CAN REVEAL INCENTIVES FOR ADS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        }
+                    ],
+                    "index": 0
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        105,
+                        505,
+                        151
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                107,
+                                505,
+                                118
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        107,
+                                        505,
+                                        118
+                                    ],
+                                    "score": 1.0,
+                                    "content": "First, recall that this unit test is a POMDP, and the state is not observed. Since there are only 2 possible",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                117,
+                                505,
+                                129
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        118,
+                                        307,
+                                        129
+                                    ],
+                                    "score": 1.0,
+                                    "content": "actions, a policy is defined by a single parameter",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        307,
+                                        117,
+                                        381,
+                                        129
+                                    ],
+                                    "score": 0.46,
+                                    "content": "\\theta = p ( { \\mathrm { c o o p e r a t e } } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        382,
+                                        118,
+                                        505,
+                                        129
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Now, the state distribution is",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                127,
+                                507,
+                                141
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        128,
+                                        204,
+                                        140
+                                    ],
+                                    "score": 0.88,
+                                    "content": "P ( s = { \\mathrm { c o o p e r a t e } } ) = \\theta",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        205,
+                                        127,
+                                        507,
+                                        141
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(ignoring the first state, which is appropriate in the limit of infinite data).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                138,
+                                435,
+                                152
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        138,
+                                        435,
+                                        152
+                                    ],
+                                    "score": 1.0,
+                                    "content": "More specifically, the probability of each state-action combination are as follows:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        }
+                    ],
+                    "index": 2.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        155,
+                        429,
+                        168
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                153,
+                                431,
+                                171
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        153,
+                                        226,
+                                        171
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Suppose we have a dataset of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        226,
+                                        156,
+                                        236,
+                                        166
+                                    ],
+                                    "score": 0.83,
+                                    "content": "N",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        236,
+                                        153,
+                                        431,
+                                        171
+                                    ],
+                                    "score": 1.0,
+                                    "content": "examples generated by following a fixed policy.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        }
+                    ],
+                    "index": 5
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        118,
+                        172,
+                        480,
+                        323
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                118,
+                                172,
+                                480,
+                                323
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        118,
+                                        172,
+                                        480,
+                                        323
+                                    ],
+                                    "score": 0.95,
+                                    "content": "\\begin{array} { r l } & { Q ( C ) = \\frac { | s = C , a = C | R ( s = C , a = C ) + | s = D , a = C | R ( s = D , a = C ) } { | a = C | } } \\\\ & { \\quad \\quad = \\frac { N P ( s = C , a = C ) R ( s = C , a = C ) + N P ( s = D , a = C ) R ( s = D , a = C ) } { N P ( a = C ) } } \\\\ & { \\quad \\quad = \\frac { P ( s = C , a = C ) R ( s = C , a = C ) + P ( s = D , a = C ) R ( s = D , a = C ) } { P ( a = C ) } } \\\\ & { \\quad \\quad = \\frac { \\theta ^ { 2 } R ( s = C , a = C ) + \\theta ( 1 - \\theta ) R ( s = D , a = C ) } { \\theta } } \\\\ & { \\quad \\quad = \\theta R ( s = C , a = C ) + ( 1 - \\theta ) R ( s = D , a = C ) } \\\\ & { \\quad \\quad = \\theta ( 0 ) + ( 1 - \\theta ) ( - 1 ) } \\\\ & { \\quad \\quad = \\theta - 1 } \\end{array}",
+                                    "type": "interline_equation",
+                                    "image_path": "7fb69a1c5ab1fc2b242a18d9902a3514228eef0bb301a5ba80a763a06f220cf4.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 7,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                118,
+                                172,
+                                480,
+                                222.33333333333334
+                            ],
+                            "spans": [],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                118,
+                                222.33333333333334,
+                                480,
+                                272.6666666666667
+                            ],
+                            "spans": [],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                118,
+                                272.6666666666667,
+                                480,
+                                323.0
+                            ],
+                            "spans": [],
+                            "index": 8
+                        }
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        132,
+                        347,
+                        477,
+                        460
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                132,
+                                347,
+                                477,
+                                460
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        132,
+                                        347,
+                                        477,
+                                        460
+                                    ],
+                                    "score": 0.94,
+                                    "content": "\\begin{array} { l } { Q ( D ) = \\frac { \\left| s = C , a = D \\right| R \\left( s = C , a = D \\right) + \\left| s = D , a = D \\right| R \\left( s = D , a = D \\right) } { \\left| a = D \\right| } } \\\\ { \\ = \\frac { P \\left( s = C , a = D \\right) R \\left( s = C , a = D \\right) + P \\left( s = D , a = D \\right) R \\left( s = D , a = D \\right) } { P \\left( a = D \\right) } } \\\\ { \\ = P ( s = C ) R ( s = C , a = D ) + P ( s = D ) R ( s = D , a = D ) } \\\\ { \\ = \\theta ( 1 / 2 ) + ( 1 - \\theta ) ( - 1 / 2 ) } \\\\ { \\ = 1 / 2 ( 2 \\theta - 1 ) } \\\\ { \\ = \\theta - 1 / 2 } \\end{array}",
+                                    "type": "interline_equation",
+                                    "image_path": "ed2ca0ff8e4e61ab4623d117e82a4bde4865f66e2921751fbf7508dd654aaa90.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 10,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                132,
+                                347,
+                                477,
+                                384.6666666666667
+                            ],
+                            "spans": [],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                132,
+                                384.6666666666667,
+                                477,
+                                422.33333333333337
+                            ],
+                            "spans": [],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                132,
+                                422.33333333333337,
+                                477,
+                                460.00000000000006
+                            ],
+                            "spans": [],
+                            "index": 11
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        469,
+                        294,
+                        482
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                468,
+                                295,
+                                483
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        468,
+                                        167,
+                                        483
+                                    ],
+                                    "score": 1.0,
+                                    "content": "So we see that",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        167,
+                                        469,
+                                        228,
+                                        482
+                                    ],
+                                    "score": 0.94,
+                                    "content": "Q ( D ) > Q ( C )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        228,
+                                        468,
+                                        286,
+                                        483
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", regardless of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        286,
+                                        470,
+                                        292,
+                                        479
+                                    ],
+                                    "score": 0.8,
+                                    "content": "\\theta",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        292,
+                                        468,
+                                        295,
+                                        483
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        }
+                    ],
+                    "index": 12
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        486,
+                        504,
+                        531
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                485,
+                                506,
+                                499
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        485,
+                                        244,
+                                        499
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Now, suppose instead that we have",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        244,
+                                        487,
+                                        254,
+                                        496
+                                    ],
+                                    "score": 0.82,
+                                    "content": "N",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        254,
+                                        485,
+                                        506,
+                                        499
+                                    ],
+                                    "score": 1.0,
+                                    "content": "examples from each of 2 different policies (given by parameters",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                107,
+                                497,
+                                506,
+                                510
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        498,
+                                        117,
+                                        509
+                                    ],
+                                    "score": 0.87,
+                                    "content": "\\theta _ { 1 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        117,
+                                        497,
+                                        135,
+                                        510
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        136,
+                                        498,
+                                        146,
+                                        509
+                                    ],
+                                    "score": 0.85,
+                                    "content": "\\theta _ { 2 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        146,
+                                        497,
+                                        506,
+                                        510
+                                    ],
+                                    "score": 1.0,
+                                    "content": ") operating in different environments. Intuitively, this sort of data might arise in practice",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                509,
+                                506,
+                                521
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        509,
+                                        506,
+                                        521
+                                    ],
+                                    "score": 1.0,
+                                    "content": "from “A/B testing”, where 2 different users have been assigned to 2 different policies in order to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                520,
+                                449,
+                                532
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        520,
+                                        301,
+                                        532
+                                    ],
+                                    "score": 1.0,
+                                    "content": "compare the policies’ performance. We now use",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        302,
+                                        520,
+                                        319,
+                                        530
+                                    ],
+                                    "score": 0.83,
+                                    "content": "_ { D C }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        319,
+                                        520,
+                                        369,
+                                        532
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to represent",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        370,
+                                        520,
+                                        428,
+                                        531
+                                    ],
+                                    "score": 0.92,
+                                    "content": "s = D , a = C",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        428,
+                                        520,
+                                        449,
+                                        532
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", etc.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        }
+                    ],
+                    "index": 14.5
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        119,
+                        550,
+                        473,
+                        720
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                119,
+                                550,
+                                473,
+                                720
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        119,
+                                        550,
+                                        473,
+                                        720
+                                    ],
+                                    "score": 0.95,
+                                    "content": "\\begin{array} { r l } & { Q ^ { \\theta _ { 1 } , \\theta _ { 2 } } ( C ) = \\frac { | C C | R ( C C ) + | D C | | R ( U C ) } { | C | } } \\\\ & { \\quad \\quad \\quad = \\frac { N \\left( P ^ { \\theta _ { 1 } } ( C C ) + P ^ { \\theta _ { 2 } } ( C C ) \\right) R ( C C ) + N ( P ^ { \\theta _ { 1 } } ( D C ) + P ^ { \\theta _ { 2 } } ( D C ) ) R ( D C ) } { N \\left( P ^ { \\theta _ { 1 } } ( C ) + P ^ { \\theta _ { 2 } } ( C ) \\right) } } \\\\ & { \\quad \\quad \\quad = \\frac { ( P ^ { \\theta _ { 1 } } ( C C ) + P ^ { \\theta _ { 2 } } ( C C ) ) I ( C C ) C + ( P ^ { \\theta _ { 1 } } ( D C ) + P ^ { \\theta _ { 2 } } ( D C ) ) R ( D C ) } { ( F ^ { \\theta _ { 1 } } ( C ) + P ^ { \\theta _ { 2 } } ( C ) ) } } \\\\ & { \\quad \\quad \\quad = \\frac { ( \\theta _ { 1 } ^ { 2 } + \\theta _ { 2 } ^ { 2 } ) R ( C C ) + ( \\theta _ { 1 } ( 1 - \\theta _ { 1 } ) + \\theta _ { 2 } ( 1 - \\theta _ { 2 } ) ) R ( D C ) } { \\theta _ { 1 } + \\theta _ { 2 } } } \\\\ & { \\quad \\quad \\quad = - \\frac { \\theta _ { 1 } ( 1 - \\theta _ { 1 } ) + \\theta _ { 2 } ( 1 - \\theta _ { 2 } ) } { \\theta _ { 1 } + \\theta _ { 2 } } } \\\\ & { \\quad \\quad \\quad = \\frac { \\theta _ { 1 } ^ { 2 } - \\theta _ { 1 } + \\theta _ { 2 } ^ { 2 } - \\theta _ { 2 } } { \\theta _ { 1 } + \\theta _ { 2 } } } \\\\ & { \\quad \\quad \\quad = \\frac { \\theta _ { 1 } ^ { 2 } - \\theta _ { 1 } + \\theta _ { 2 } ^ { 2 } - \\theta _ { 2 } } { \\theta _ { 1 } + \\theta _ { 2 } } } \\\\ & { \\quad \\quad \\quad = \\frac { \\theta _ { 1 } ^ { 2 } - \\theta _ { 1 } + \\theta _ { 2 } ^ { 2 } - \\theta _ { 2 } } { \\theta _ { 1 } + \\theta _ { 2 } } } \\end{array}",
+                                    "type": "interline_equation",
+                                    "image_path": "d4bb4fc9c5a13bc7c7368888d2836a3dc036de75040b79dfca768e03e860184f.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 18,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                119,
+                                550,
+                                473,
+                                606.6666666666666
+                            ],
+                            "spans": [],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                119,
+                                606.6666666666666,
+                                473,
+                                663.3333333333333
+                            ],
+                            "spans": [],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                119,
+                                663.3333333333333,
+                                473,
+                                719.9999999999999
+                            ],
+                            "spans": [],
+                            "index": 19
+                        }
+                    ]
+                }
+            ],
+            "page_idx": 22,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        300,
+                        751,
+                        311,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                298,
+                                750,
+                                313,
+                                764
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        298,
+                                        750,
+                                        313,
+                                        764
+                                    ],
+                                    "score": 1.0,
+                                    "content": "",
+                                    "type": "text",
+                                    "height": 14,
+                                    "width": 15
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        106,
+                        26,
+                        307,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                25,
+                                308,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        25,
+                                        308,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Under review as a conference paper at ICLR 2021",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        81,
+                        434,
+                        94
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                78,
+                                436,
+                                99
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        78,
+                                        436,
+                                        99
+                                    ],
+                                    "score": 1.0,
+                                    "content": "10 OFFLINE Q-LEARNING CAN REVEAL INCENTIVES FOR ADS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        }
+                    ],
+                    "index": 0
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        105,
+                        505,
+                        151
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                107,
+                                505,
+                                118
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        107,
+                                        505,
+                                        118
+                                    ],
+                                    "score": 1.0,
+                                    "content": "First, recall that this unit test is a POMDP, and the state is not observed. Since there are only 2 possible",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                117,
+                                505,
+                                129
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        118,
+                                        307,
+                                        129
+                                    ],
+                                    "score": 1.0,
+                                    "content": "actions, a policy is defined by a single parameter",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        307,
+                                        117,
+                                        381,
+                                        129
+                                    ],
+                                    "score": 0.46,
+                                    "content": "\\theta = p ( { \\mathrm { c o o p e r a t e } } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        382,
+                                        118,
+                                        505,
+                                        129
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Now, the state distribution is",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                127,
+                                507,
+                                141
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        128,
+                                        204,
+                                        140
+                                    ],
+                                    "score": 0.88,
+                                    "content": "P ( s = { \\mathrm { c o o p e r a t e } } ) = \\theta",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        205,
+                                        127,
+                                        507,
+                                        141
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(ignoring the first state, which is appropriate in the limit of infinite data).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                138,
+                                435,
+                                152
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        138,
+                                        435,
+                                        152
+                                    ],
+                                    "score": 1.0,
+                                    "content": "More specifically, the probability of each state-action combination are as follows:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        }
+                    ],
+                    "index": 2.5,
+                    "bbox_fs": [
+                        105,
+                        107,
+                        507,
+                        152
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        155,
+                        429,
+                        168
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                153,
+                                431,
+                                171
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        153,
+                                        226,
+                                        171
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Suppose we have a dataset of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        226,
+                                        156,
+                                        236,
+                                        166
+                                    ],
+                                    "score": 0.83,
+                                    "content": "N",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        236,
+                                        153,
+                                        431,
+                                        171
+                                    ],
+                                    "score": 1.0,
+                                    "content": "examples generated by following a fixed policy.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        }
+                    ],
+                    "index": 5,
+                    "bbox_fs": [
+                        105,
+                        153,
+                        431,
+                        171
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        118,
+                        172,
+                        480,
+                        323
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                118,
+                                172,
+                                480,
+                                323
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        118,
+                                        172,
+                                        480,
+                                        323
+                                    ],
+                                    "score": 0.95,
+                                    "content": "\\begin{array} { r l } & { Q ( C ) = \\frac { | s = C , a = C | R ( s = C , a = C ) + | s = D , a = C | R ( s = D , a = C ) } { | a = C | } } \\\\ & { \\quad \\quad = \\frac { N P ( s = C , a = C ) R ( s = C , a = C ) + N P ( s = D , a = C ) R ( s = D , a = C ) } { N P ( a = C ) } } \\\\ & { \\quad \\quad = \\frac { P ( s = C , a = C ) R ( s = C , a = C ) + P ( s = D , a = C ) R ( s = D , a = C ) } { P ( a = C ) } } \\\\ & { \\quad \\quad = \\frac { \\theta ^ { 2 } R ( s = C , a = C ) + \\theta ( 1 - \\theta ) R ( s = D , a = C ) } { \\theta } } \\\\ & { \\quad \\quad = \\theta R ( s = C , a = C ) + ( 1 - \\theta ) R ( s = D , a = C ) } \\\\ & { \\quad \\quad = \\theta ( 0 ) + ( 1 - \\theta ) ( - 1 ) } \\\\ & { \\quad \\quad = \\theta - 1 } \\end{array}",
+                                    "type": "interline_equation",
+                                    "image_path": "7fb69a1c5ab1fc2b242a18d9902a3514228eef0bb301a5ba80a763a06f220cf4.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 7,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                118,
+                                172,
+                                480,
+                                222.33333333333334
+                            ],
+                            "spans": [],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                118,
+                                222.33333333333334,
+                                480,
+                                272.6666666666667
+                            ],
+                            "spans": [],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                118,
+                                272.6666666666667,
+                                480,
+                                323.0
+                            ],
+                            "spans": [],
+                            "index": 8
+                        }
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        132,
+                        347,
+                        477,
+                        460
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                132,
+                                347,
+                                477,
+                                460
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        132,
+                                        347,
+                                        477,
+                                        460
+                                    ],
+                                    "score": 0.94,
+                                    "content": "\\begin{array} { l } { Q ( D ) = \\frac { \\left| s = C , a = D \\right| R \\left( s = C , a = D \\right) + \\left| s = D , a = D \\right| R \\left( s = D , a = D \\right) } { \\left| a = D \\right| } } \\\\ { \\ = \\frac { P \\left( s = C , a = D \\right) R \\left( s = C , a = D \\right) + P \\left( s = D , a = D \\right) R \\left( s = D , a = D \\right) } { P \\left( a = D \\right) } } \\\\ { \\ = P ( s = C ) R ( s = C , a = D ) + P ( s = D ) R ( s = D , a = D ) } \\\\ { \\ = \\theta ( 1 / 2 ) + ( 1 - \\theta ) ( - 1 / 2 ) } \\\\ { \\ = 1 / 2 ( 2 \\theta - 1 ) } \\\\ { \\ = \\theta - 1 / 2 } \\end{array}",
+                                    "type": "interline_equation",
+                                    "image_path": "ed2ca0ff8e4e61ab4623d117e82a4bde4865f66e2921751fbf7508dd654aaa90.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 10,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                132,
+                                347,
+                                477,
+                                384.6666666666667
+                            ],
+                            "spans": [],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                132,
+                                384.6666666666667,
+                                477,
+                                422.33333333333337
+                            ],
+                            "spans": [],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                132,
+                                422.33333333333337,
+                                477,
+                                460.00000000000006
+                            ],
+                            "spans": [],
+                            "index": 11
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        469,
+                        294,
+                        482
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                468,
+                                295,
+                                483
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        468,
+                                        167,
+                                        483
+                                    ],
+                                    "score": 1.0,
+                                    "content": "So we see that",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        167,
+                                        469,
+                                        228,
+                                        482
+                                    ],
+                                    "score": 0.94,
+                                    "content": "Q ( D ) > Q ( C )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        228,
+                                        468,
+                                        286,
+                                        483
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", regardless of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        286,
+                                        470,
+                                        292,
+                                        479
+                                    ],
+                                    "score": 0.8,
+                                    "content": "\\theta",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        292,
+                                        468,
+                                        295,
+                                        483
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        }
+                    ],
+                    "index": 12,
+                    "bbox_fs": [
+                        106,
+                        468,
+                        295,
+                        483
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        486,
+                        504,
+                        531
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                485,
+                                506,
+                                499
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        485,
+                                        244,
+                                        499
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Now, suppose instead that we have",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        244,
+                                        487,
+                                        254,
+                                        496
+                                    ],
+                                    "score": 0.82,
+                                    "content": "N",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        254,
+                                        485,
+                                        506,
+                                        499
+                                    ],
+                                    "score": 1.0,
+                                    "content": "examples from each of 2 different policies (given by parameters",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                107,
+                                497,
+                                506,
+                                510
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        498,
+                                        117,
+                                        509
+                                    ],
+                                    "score": 0.87,
+                                    "content": "\\theta _ { 1 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        117,
+                                        497,
+                                        135,
+                                        510
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        136,
+                                        498,
+                                        146,
+                                        509
+                                    ],
+                                    "score": 0.85,
+                                    "content": "\\theta _ { 2 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        146,
+                                        497,
+                                        506,
+                                        510
+                                    ],
+                                    "score": 1.0,
+                                    "content": ") operating in different environments. Intuitively, this sort of data might arise in practice",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                509,
+                                506,
+                                521
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        509,
+                                        506,
+                                        521
+                                    ],
+                                    "score": 1.0,
+                                    "content": "from “A/B testing”, where 2 different users have been assigned to 2 different policies in order to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                520,
+                                449,
+                                532
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        520,
+                                        301,
+                                        532
+                                    ],
+                                    "score": 1.0,
+                                    "content": "compare the policies’ performance. We now use",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        302,
+                                        520,
+                                        319,
+                                        530
+                                    ],
+                                    "score": 0.83,
+                                    "content": "_ { D C }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        319,
+                                        520,
+                                        369,
+                                        532
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to represent",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        370,
+                                        520,
+                                        428,
+                                        531
+                                    ],
+                                    "score": 0.92,
+                                    "content": "s = D , a = C",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        428,
+                                        520,
+                                        449,
+                                        532
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", etc.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        }
+                    ],
+                    "index": 14.5,
+                    "bbox_fs": [
+                        105,
+                        485,
+                        506,
+                        532
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        119,
+                        550,
+                        473,
+                        720
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                119,
+                                550,
+                                473,
+                                720
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        119,
+                                        550,
+                                        473,
+                                        720
+                                    ],
+                                    "score": 0.95,
+                                    "content": "\\begin{array} { r l } & { Q ^ { \\theta _ { 1 } , \\theta _ { 2 } } ( C ) = \\frac { | C C | R ( C C ) + | D C | | R ( U C ) } { | C | } } \\\\ & { \\quad \\quad \\quad = \\frac { N \\left( P ^ { \\theta _ { 1 } } ( C C ) + P ^ { \\theta _ { 2 } } ( C C ) \\right) R ( C C ) + N ( P ^ { \\theta _ { 1 } } ( D C ) + P ^ { \\theta _ { 2 } } ( D C ) ) R ( D C ) } { N \\left( P ^ { \\theta _ { 1 } } ( C ) + P ^ { \\theta _ { 2 } } ( C ) \\right) } } \\\\ & { \\quad \\quad \\quad = \\frac { ( P ^ { \\theta _ { 1 } } ( C C ) + P ^ { \\theta _ { 2 } } ( C C ) ) I ( C C ) C + ( P ^ { \\theta _ { 1 } } ( D C ) + P ^ { \\theta _ { 2 } } ( D C ) ) R ( D C ) } { ( F ^ { \\theta _ { 1 } } ( C ) + P ^ { \\theta _ { 2 } } ( C ) ) } } \\\\ & { \\quad \\quad \\quad = \\frac { ( \\theta _ { 1 } ^ { 2 } + \\theta _ { 2 } ^ { 2 } ) R ( C C ) + ( \\theta _ { 1 } ( 1 - \\theta _ { 1 } ) + \\theta _ { 2 } ( 1 - \\theta _ { 2 } ) ) R ( D C ) } { \\theta _ { 1 } + \\theta _ { 2 } } } \\\\ & { \\quad \\quad \\quad = - \\frac { \\theta _ { 1 } ( 1 - \\theta _ { 1 } ) + \\theta _ { 2 } ( 1 - \\theta _ { 2 } ) } { \\theta _ { 1 } + \\theta _ { 2 } } } \\\\ & { \\quad \\quad \\quad = \\frac { \\theta _ { 1 } ^ { 2 } - \\theta _ { 1 } + \\theta _ { 2 } ^ { 2 } - \\theta _ { 2 } } { \\theta _ { 1 } + \\theta _ { 2 } } } \\\\ & { \\quad \\quad \\quad = \\frac { \\theta _ { 1 } ^ { 2 } - \\theta _ { 1 } + \\theta _ { 2 } ^ { 2 } - \\theta _ { 2 } } { \\theta _ { 1 } + \\theta _ { 2 } } } \\\\ & { \\quad \\quad \\quad = \\frac { \\theta _ { 1 } ^ { 2 } - \\theta _ { 1 } + \\theta _ { 2 } ^ { 2 } - \\theta _ { 2 } } { \\theta _ { 1 } + \\theta _ { 2 } } } \\end{array}",
+                                    "type": "interline_equation",
+                                    "image_path": "d4bb4fc9c5a13bc7c7368888d2836a3dc036de75040b79dfca768e03e860184f.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 18,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                119,
+                                550,
+                                473,
+                                606.6666666666666
+                            ],
+                            "spans": [],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                119,
+                                606.6666666666666,
+                                473,
+                                663.3333333333333
+                            ],
+                            "spans": [],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                119,
+                                663.3333333333333,
+                                473,
+                                719.9999999999999
+                            ],
+                            "spans": [],
+                            "index": 19
+                        }
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        128,
+                        96,
+                        466,
+                        209
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                128,
+                                96,
+                                466,
+                                209
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        128,
+                                        96,
+                                        466,
+                                        209
+                                    ],
+                                    "score": 0.96,
+                                    "content": "\\begin{array} { l } { { Q ^ { \\theta _ { 1 } , \\theta _ { 2 } } ( D ) = \\frac { \\left| C D \\right| R \\left( C D \\right) + \\left| D D \\right| R \\left( D D \\right) } { \\left| D \\right| } } } \\\\ { { \\ = \\frac { \\left( P ^ { \\theta _ { 1 } } \\left( C D \\right) + P ^ { \\theta _ { 2 } } \\left( C D \\right) \\right) R \\left( C D \\right) + \\left( P ^ { \\theta _ { 1 } } \\left( D D \\right) + P ^ { \\theta _ { 2 } } \\left( D D \\right) \\right) R \\left( D D \\right) } { \\left( P ^ { \\theta _ { 1 } } \\left( D \\right) + P ^ { \\theta _ { 2 } } \\left( D \\right) \\right) } } } \\\\ { { \\ = \\frac { 1 / 2 \\left( \\theta _ { 1 } \\left( 1 - \\theta _ { 1 } \\right) + \\theta _ { 2 } \\left( 1 - \\theta _ { 2 } \\right) \\right) - 1 / 2 \\left( ( 1 - \\theta _ { 1 } ) ^ { 2 } + ( 1 - \\theta _ { 2 } ) ^ { 2 } \\right) } { 2 - \\theta _ { 1 } - \\theta _ { 2 } } } } \\\\ { { \\ = \\frac { \\left( 2 \\theta _ { 1 } - 1 \\right) \\left( 1 - \\theta _ { 1 } \\right) + \\left( 2 \\theta _ { 2 } - 1 \\right) \\left( 1 - \\theta _ { 2 } \\right) } { 4 - 2 \\theta _ { 1 } - 2 \\theta _ { 2 } } } } \\end{array}",
+                                    "type": "interline_equation",
+                                    "image_path": "b154449f50ac688679c16d1b39473d4ce271210e3cd6613c45b3e3f3d0e9126b.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 1,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                128,
+                                96,
+                                466,
+                                133.66666666666666
+                            ],
+                            "spans": [],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                128,
+                                133.66666666666666,
+                                466,
+                                171.33333333333331
+                            ],
+                            "spans": [],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                128,
+                                171.33333333333331,
+                                466,
+                                208.99999999999997
+                            ],
+                            "spans": [],
+                            "index": 2
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        217,
+                        504,
+                        251
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                217,
+                                505,
+                                230
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        217,
+                                        239,
+                                        230
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Now, in Figure 15 we see that",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        239,
+                                        217,
+                                        305,
+                                        230
+                                    ],
+                                    "score": 0.93,
+                                    "content": "Q ( C ) > Q ( D )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        306,
+                                        217,
+                                        505,
+                                        230
+                                    ],
+                                    "score": 1.0,
+                                    "content": "when one of the policies cooperates with high",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                228,
+                                506,
+                                241
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        228,
+                                        506,
+                                        241
+                                    ],
+                                    "score": 1.0,
+                                    "content": "probability, and the other defects with high probability. Intuitively, the result of pooling data from 2",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                240,
+                                504,
+                                252
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        240,
+                                        321,
+                                        252
+                                    ],
+                                    "score": 1.0,
+                                    "content": "such policies is very similar to collecting data from an",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        322,
+                                        242,
+                                        327,
+                                        249
+                                    ],
+                                    "score": 0.58,
+                                    "content": "\\epsilon",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        327,
+                                        240,
+                                        504,
+                                        252
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-greedy policy trained online (as in Figure 6).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        }
+                    ],
+                    "index": 4
+                },
+                {
+                    "type": "image",
+                    "bbox": [
+                        164,
+                        267,
+                        446,
+                        537
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                164,
+                                267,
+                                446,
+                                537
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        164,
+                                        267,
+                                        446,
+                                        537
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                164,
+                                                267,
+                                                446,
+                                                537
+                                            ],
+                                            "score": 0.973,
+                                            "type": "image",
+                                            "image_path": "f8cf957c6b390dfb762d079853f1b8f334d627fe895c972607d8fe3c71e3938d.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 15.5,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        164,
+                                        267,
+                                        446,
+                                        280.5
+                                    ],
+                                    "spans": [],
+                                    "index": 6
+                                },
+                                {
+                                    "bbox": [
+                                        164,
+                                        280.5,
+                                        446,
+                                        294.0
+                                    ],
+                                    "spans": [],
+                                    "index": 7
+                                },
+                                {
+                                    "bbox": [
+                                        164,
+                                        294.0,
+                                        446,
+                                        307.5
+                                    ],
+                                    "spans": [],
+                                    "index": 8
+                                },
+                                {
+                                    "bbox": [
+                                        164,
+                                        307.5,
+                                        446,
+                                        321.0
+                                    ],
+                                    "spans": [],
+                                    "index": 9
+                                },
+                                {
+                                    "bbox": [
+                                        164,
+                                        321.0,
+                                        446,
+                                        334.5
+                                    ],
+                                    "spans": [],
+                                    "index": 10
+                                },
+                                {
+                                    "bbox": [
+                                        164,
+                                        334.5,
+                                        446,
+                                        348.0
+                                    ],
+                                    "spans": [],
+                                    "index": 11
+                                },
+                                {
+                                    "bbox": [
+                                        164,
+                                        348.0,
+                                        446,
+                                        361.5
+                                    ],
+                                    "spans": [],
+                                    "index": 12
+                                },
+                                {
+                                    "bbox": [
+                                        164,
+                                        361.5,
+                                        446,
+                                        375.0
+                                    ],
+                                    "spans": [],
+                                    "index": 13
+                                },
+                                {
+                                    "bbox": [
+                                        164,
+                                        375.0,
+                                        446,
+                                        388.5
+                                    ],
+                                    "spans": [],
+                                    "index": 14
+                                },
+                                {
+                                    "bbox": [
+                                        164,
+                                        388.5,
+                                        446,
+                                        402.0
+                                    ],
+                                    "spans": [],
+                                    "index": 15
+                                },
+                                {
+                                    "bbox": [
+                                        164,
+                                        402.0,
+                                        446,
+                                        415.5
+                                    ],
+                                    "spans": [],
+                                    "index": 16
+                                },
+                                {
+                                    "bbox": [
+                                        164,
+                                        415.5,
+                                        446,
+                                        429.0
+                                    ],
+                                    "spans": [],
+                                    "index": 17
+                                },
+                                {
+                                    "bbox": [
+                                        164,
+                                        429.0,
+                                        446,
+                                        442.5
+                                    ],
+                                    "spans": [],
+                                    "index": 18
+                                },
+                                {
+                                    "bbox": [
+                                        164,
+                                        442.5,
+                                        446,
+                                        456.0
+                                    ],
+                                    "spans": [],
+                                    "index": 19
+                                },
+                                {
+                                    "bbox": [
+                                        164,
+                                        456.0,
+                                        446,
+                                        469.5
+                                    ],
+                                    "spans": [],
+                                    "index": 20
+                                },
+                                {
+                                    "bbox": [
+                                        164,
+                                        469.5,
+                                        446,
+                                        483.0
+                                    ],
+                                    "spans": [],
+                                    "index": 21
+                                },
+                                {
+                                    "bbox": [
+                                        164,
+                                        483.0,
+                                        446,
+                                        496.5
+                                    ],
+                                    "spans": [],
+                                    "index": 22
+                                },
+                                {
+                                    "bbox": [
+                                        164,
+                                        496.5,
+                                        446,
+                                        510.0
+                                    ],
+                                    "spans": [],
+                                    "index": 23
+                                },
+                                {
+                                    "bbox": [
+                                        164,
+                                        510.0,
+                                        446,
+                                        523.5
+                                    ],
+                                    "spans": [],
+                                    "index": 24
+                                },
+                                {
+                                    "bbox": [
+                                        164,
+                                        523.5,
+                                        446,
+                                        537.0
+                                    ],
+                                    "spans": [],
+                                    "index": 25
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                106,
+                                556,
+                                505,
+                                591
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        556,
+                                        506,
+                                        570
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                556,
+                                                506,
+                                                570
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 15: Offline Q-learning can also reveal HI-ADS, when pooling data from different (policy,",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 26
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        567,
+                                        506,
+                                        582
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                567,
+                                                388,
+                                                582
+                                            ],
+                                            "score": 1.0,
+                                            "content": "environment) pairs. Yellow regions represent policy pairs for which",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                388,
+                                                568,
+                                                452,
+                                                580
+                                            ],
+                                            "score": 0.93,
+                                            "content": "\\mathbf { \\bar { Q } } ( C ) > Q ( D )",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                452,
+                                                567,
+                                                506,
+                                                582
+                                            ],
+                                            "score": 1.0,
+                                            "content": ", resulting in",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 27
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        579,
+                                        196,
+                                        592
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                579,
+                                                196,
+                                                592
+                                            ],
+                                            "score": 1.0,
+                                            "content": "non-myopic behavior.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 28
+                                }
+                            ],
+                            "index": 27
+                        }
+                    ],
+                    "index": 21.25
+                }
+            ],
+            "page_idx": 23,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        300,
+                        751,
+                        311,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                298,
+                                750,
+                                313,
+                                765
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        298,
+                                        750,
+                                        313,
+                                        765
+                                    ],
+                                    "score": 1.0,
+                                    "content": "",
+                                    "type": "text",
+                                    "height": 15,
+                                    "width": 15
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        106,
+                        27,
+                        307,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                107,
+                                26,
+                                308,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        26,
+                                        308,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Under review as a conference paper at ICLR 2021",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        128,
+                        96,
+                        466,
+                        209
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                128,
+                                96,
+                                466,
+                                209
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        128,
+                                        96,
+                                        466,
+                                        209
+                                    ],
+                                    "score": 0.96,
+                                    "content": "\\begin{array} { l } { { Q ^ { \\theta _ { 1 } , \\theta _ { 2 } } ( D ) = \\frac { \\left| C D \\right| R \\left( C D \\right) + \\left| D D \\right| R \\left( D D \\right) } { \\left| D \\right| } } } \\\\ { { \\ = \\frac { \\left( P ^ { \\theta _ { 1 } } \\left( C D \\right) + P ^ { \\theta _ { 2 } } \\left( C D \\right) \\right) R \\left( C D \\right) + \\left( P ^ { \\theta _ { 1 } } \\left( D D \\right) + P ^ { \\theta _ { 2 } } \\left( D D \\right) \\right) R \\left( D D \\right) } { \\left( P ^ { \\theta _ { 1 } } \\left( D \\right) + P ^ { \\theta _ { 2 } } \\left( D \\right) \\right) } } } \\\\ { { \\ = \\frac { 1 / 2 \\left( \\theta _ { 1 } \\left( 1 - \\theta _ { 1 } \\right) + \\theta _ { 2 } \\left( 1 - \\theta _ { 2 } \\right) \\right) - 1 / 2 \\left( ( 1 - \\theta _ { 1 } ) ^ { 2 } + ( 1 - \\theta _ { 2 } ) ^ { 2 } \\right) } { 2 - \\theta _ { 1 } - \\theta _ { 2 } } } } \\\\ { { \\ = \\frac { \\left( 2 \\theta _ { 1 } - 1 \\right) \\left( 1 - \\theta _ { 1 } \\right) + \\left( 2 \\theta _ { 2 } - 1 \\right) \\left( 1 - \\theta _ { 2 } \\right) } { 4 - 2 \\theta _ { 1 } - 2 \\theta _ { 2 } } } } \\end{array}",
+                                    "type": "interline_equation",
+                                    "image_path": "b154449f50ac688679c16d1b39473d4ce271210e3cd6613c45b3e3f3d0e9126b.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 1,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                128,
+                                96,
+                                466,
+                                133.66666666666666
+                            ],
+                            "spans": [],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                128,
+                                133.66666666666666,
+                                466,
+                                171.33333333333331
+                            ],
+                            "spans": [],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                128,
+                                171.33333333333331,
+                                466,
+                                208.99999999999997
+                            ],
+                            "spans": [],
+                            "index": 2
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        217,
+                        504,
+                        251
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                217,
+                                505,
+                                230
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        217,
+                                        239,
+                                        230
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Now, in Figure 15 we see that",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        239,
+                                        217,
+                                        305,
+                                        230
+                                    ],
+                                    "score": 0.93,
+                                    "content": "Q ( C ) > Q ( D )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        306,
+                                        217,
+                                        505,
+                                        230
+                                    ],
+                                    "score": 1.0,
+                                    "content": "when one of the policies cooperates with high",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                228,
+                                506,
+                                241
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        228,
+                                        506,
+                                        241
+                                    ],
+                                    "score": 1.0,
+                                    "content": "probability, and the other defects with high probability. Intuitively, the result of pooling data from 2",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                240,
+                                504,
+                                252
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        240,
+                                        321,
+                                        252
+                                    ],
+                                    "score": 1.0,
+                                    "content": "such policies is very similar to collecting data from an",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        322,
+                                        242,
+                                        327,
+                                        249
+                                    ],
+                                    "score": 0.58,
+                                    "content": "\\epsilon",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        327,
+                                        240,
+                                        504,
+                                        252
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-greedy policy trained online (as in Figure 6).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        }
+                    ],
+                    "index": 4,
+                    "bbox_fs": [
+                        105,
+                        217,
+                        506,
+                        252
+                    ]
+                },
+                {
+                    "type": "image",
+                    "bbox": [
+                        164,
+                        267,
+                        446,
+                        537
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                164,
+                                267,
+                                446,
+                                537
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        164,
+                                        267,
+                                        446,
+                                        537
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                164,
+                                                267,
+                                                446,
+                                                537
+                                            ],
+                                            "score": 0.973,
+                                            "type": "image",
+                                            "image_path": "f8cf957c6b390dfb762d079853f1b8f334d627fe895c972607d8fe3c71e3938d.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 15.5,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        164,
+                                        267,
+                                        446,
+                                        280.5
+                                    ],
+                                    "spans": [],
+                                    "index": 6
+                                },
+                                {
+                                    "bbox": [
+                                        164,
+                                        280.5,
+                                        446,
+                                        294.0
+                                    ],
+                                    "spans": [],
+                                    "index": 7
+                                },
+                                {
+                                    "bbox": [
+                                        164,
+                                        294.0,
+                                        446,
+                                        307.5
+                                    ],
+                                    "spans": [],
+                                    "index": 8
+                                },
+                                {
+                                    "bbox": [
+                                        164,
+                                        307.5,
+                                        446,
+                                        321.0
+                                    ],
+                                    "spans": [],
+                                    "index": 9
+                                },
+                                {
+                                    "bbox": [
+                                        164,
+                                        321.0,
+                                        446,
+                                        334.5
+                                    ],
+                                    "spans": [],
+                                    "index": 10
+                                },
+                                {
+                                    "bbox": [
+                                        164,
+                                        334.5,
+                                        446,
+                                        348.0
+                                    ],
+                                    "spans": [],
+                                    "index": 11
+                                },
+                                {
+                                    "bbox": [
+                                        164,
+                                        348.0,
+                                        446,
+                                        361.5
+                                    ],
+                                    "spans": [],
+                                    "index": 12
+                                },
+                                {
+                                    "bbox": [
+                                        164,
+                                        361.5,
+                                        446,
+                                        375.0
+                                    ],
+                                    "spans": [],
+                                    "index": 13
+                                },
+                                {
+                                    "bbox": [
+                                        164,
+                                        375.0,
+                                        446,
+                                        388.5
+                                    ],
+                                    "spans": [],
+                                    "index": 14
+                                },
+                                {
+                                    "bbox": [
+                                        164,
+                                        388.5,
+                                        446,
+                                        402.0
+                                    ],
+                                    "spans": [],
+                                    "index": 15
+                                },
+                                {
+                                    "bbox": [
+                                        164,
+                                        402.0,
+                                        446,
+                                        415.5
+                                    ],
+                                    "spans": [],
+                                    "index": 16
+                                },
+                                {
+                                    "bbox": [
+                                        164,
+                                        415.5,
+                                        446,
+                                        429.0
+                                    ],
+                                    "spans": [],
+                                    "index": 17
+                                },
+                                {
+                                    "bbox": [
+                                        164,
+                                        429.0,
+                                        446,
+                                        442.5
+                                    ],
+                                    "spans": [],
+                                    "index": 18
+                                },
+                                {
+                                    "bbox": [
+                                        164,
+                                        442.5,
+                                        446,
+                                        456.0
+                                    ],
+                                    "spans": [],
+                                    "index": 19
+                                },
+                                {
+                                    "bbox": [
+                                        164,
+                                        456.0,
+                                        446,
+                                        469.5
+                                    ],
+                                    "spans": [],
+                                    "index": 20
+                                },
+                                {
+                                    "bbox": [
+                                        164,
+                                        469.5,
+                                        446,
+                                        483.0
+                                    ],
+                                    "spans": [],
+                                    "index": 21
+                                },
+                                {
+                                    "bbox": [
+                                        164,
+                                        483.0,
+                                        446,
+                                        496.5
+                                    ],
+                                    "spans": [],
+                                    "index": 22
+                                },
+                                {
+                                    "bbox": [
+                                        164,
+                                        496.5,
+                                        446,
+                                        510.0
+                                    ],
+                                    "spans": [],
+                                    "index": 23
+                                },
+                                {
+                                    "bbox": [
+                                        164,
+                                        510.0,
+                                        446,
+                                        523.5
+                                    ],
+                                    "spans": [],
+                                    "index": 24
+                                },
+                                {
+                                    "bbox": [
+                                        164,
+                                        523.5,
+                                        446,
+                                        537.0
+                                    ],
+                                    "spans": [],
+                                    "index": 25
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                106,
+                                556,
+                                505,
+                                591
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        556,
+                                        506,
+                                        570
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                556,
+                                                506,
+                                                570
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 15: Offline Q-learning can also reveal HI-ADS, when pooling data from different (policy,",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 26
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        567,
+                                        506,
+                                        582
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                567,
+                                                388,
+                                                582
+                                            ],
+                                            "score": 1.0,
+                                            "content": "environment) pairs. Yellow regions represent policy pairs for which",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                388,
+                                                568,
+                                                452,
+                                                580
+                                            ],
+                                            "score": 0.93,
+                                            "content": "\\mathbf { \\bar { Q } } ( C ) > Q ( D )",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                452,
+                                                567,
+                                                506,
+                                                582
+                                            ],
+                                            "score": 1.0,
+                                            "content": ", resulting in",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 27
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        579,
+                                        196,
+                                        592
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                579,
+                                                196,
+                                                592
+                                            ],
+                                            "score": 1.0,
+                                            "content": "non-myopic behavior.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 28
+                                }
+                            ],
+                            "index": 27
+                        }
+                    ],
+                    "index": 21.25
+                }
+            ]
+        }
+    ],
+    "_backend": "pipeline",
+    "_version_name": "2.2.2"
+}
\ No newline at end of file
diff --git a/parse/train/3FK30d5BZdu/3FK30d5BZdu_model.json b/parse/train/3FK30d5BZdu/3FK30d5BZdu_model.json
new file mode 100644
index 0000000000000000000000000000000000000000..e31317a2032c673465f1486bd04170252971017e
--- /dev/null
+++ b/parse/train/3FK30d5BZdu/3FK30d5BZdu_model.json
@@ -0,0 +1,41386 @@
+[
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    398,
+                    587,
+                    1304,
+                    587,
+                    1304,
+                    1078,
+                    398,
+                    1078
+                ],
+                "score": 0.984
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    1185,
+                    1407,
+                    1185,
+                    1407,
+                    1339,
+                    299,
+                    1339
+                ],
+                "score": 0.979
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1772,
+                    1403,
+                    1772,
+                    1403,
+                    1926,
+                    298,
+                    1926
+                ],
+                "score": 0.977
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    300,
+                    1942,
+                    1402,
+                    1942,
+                    1402,
+                    2035,
+                    300,
+                    2035
+                ],
+                "score": 0.97
+            },
+            {
+                "category_id": 3,
+                "poly": [
+                    481,
+                    1378,
+                    1247,
+                    1378,
+                    1247,
+                    1618,
+                    481,
+                    1618
+                ],
+                "score": 0.966
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    299,
+                    220,
+                    1409,
+                    220,
+                    1409,
+                    322,
+                    299,
+                    322
+                ],
+                "score": 0.96
+            },
+            {
+                "category_id": 4,
+                "poly": [
+                    300,
+                    1649,
+                    1405,
+                    1649,
+                    1405,
+                    1742,
+                    300,
+                    1742
+                ],
+                "score": 0.959
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    313,
+                    377,
+                    680,
+                    377,
+                    680,
+                    438,
+                    313,
+                    438
+                ],
+                "score": 0.934
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    302,
+                    1131,
+                    573,
+                    1131,
+                    573,
+                    1166,
+                    302,
+                    1166
+                ],
+                "score": 0.908
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    773,
+                    519,
+                    927,
+                    519,
+                    927,
+                    553,
+                    773,
+                    553
+                ],
+                "score": 0.883
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    298,
+                    75,
+                    854,
+                    75,
+                    854,
+                    105,
+                    298,
+                    105
+                ],
+                "score": 0.842
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    841,
+                    2088,
+                    856,
+                    2088,
+                    856,
+                    2112,
+                    841,
+                    2112
+                ],
+                "score": 0.731
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    818.0,
+                    1398.0,
+                    836.0,
+                    1398.0,
+                    836.0,
+                    1412.0,
+                    818.0,
+                    1412.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    815.0,
+                    1423.0,
+                    834.0,
+                    1423.0,
+                    834.0,
+                    1437.0,
+                    815.0,
+                    1437.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1206.0,
+                    1423.0,
+                    1227.0,
+                    1423.0,
+                    1227.0,
+                    1437.0,
+                    1206.0,
+                    1437.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    813.0,
+                    1447.0,
+                    832.0,
+                    1447.0,
+                    832.0,
+                    1461.0,
+                    813.0,
+                    1461.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1203.0,
+                    1447.0,
+                    1225.0,
+                    1447.0,
+                    1225.0,
+                    1461.0,
+                    1203.0,
+                    1461.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    812.0,
+                    1471.0,
+                    832.0,
+                    1471.0,
+                    832.0,
+                    1485.0,
+                    812.0,
+                    1485.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    842.0,
+                    1476.0,
+                    850.0,
+                    1476.0,
+                    850.0,
+                    1488.0,
+                    842.0,
+                    1488.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1203.0,
+                    1471.0,
+                    1224.0,
+                    1471.0,
+                    1224.0,
+                    1484.0,
+                    1203.0,
+                    1484.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1235.0,
+                    1474.0,
+                    1242.0,
+                    1474.0,
+                    1242.0,
+                    1488.0,
+                    1235.0,
+                    1488.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    810.0,
+                    1493.0,
+                    829.0,
+                    1493.0,
+                    829.0,
+                    1508.0,
+                    810.0,
+                    1508.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1202.0,
+                    1494.0,
+                    1222.0,
+                    1494.0,
+                    1222.0,
+                    1508.0,
+                    1202.0,
+                    1508.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    808.0,
+                    1516.0,
+                    828.0,
+                    1516.0,
+                    828.0,
+                    1531.0,
+                    808.0,
+                    1531.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1201.0,
+                    1516.0,
+                    1220.0,
+                    1516.0,
+                    1220.0,
+                    1531.0,
+                    1201.0,
+                    1531.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    796.0,
+                    1537.0,
+                    813.0,
+                    1537.0,
+                    813.0,
+                    1551.0,
+                    796.0,
+                    1551.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1189.0,
+                    1538.0,
+                    1204.0,
+                    1538.0,
+                    1204.0,
+                    1549.0,
+                    1189.0,
+                    1549.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    784.0,
+                    1550.0,
+                    793.0,
+                    1550.0,
+                    793.0,
+                    1560.0,
+                    784.0,
+                    1560.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1177.0,
+                    1550.0,
+                    1185.0,
+                    1550.0,
+                    1185.0,
+                    1559.0,
+                    1177.0,
+                    1559.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    482.0,
+                    1559.0,
+                    510.0,
+                    1559.0,
+                    510.0,
+                    1575.0,
+                    482.0,
+                    1575.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    769.0,
+                    1563.0,
+                    776.0,
+                    1563.0,
+                    776.0,
+                    1571.0,
+                    769.0,
+                    1571.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    876.0,
+                    1560.0,
+                    902.0,
+                    1560.0,
+                    902.0,
+                    1573.0,
+                    876.0,
+                    1573.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    533.0,
+                    1570.0,
+                    555.0,
+                    1570.0,
+                    555.0,
+                    1585.0,
+                    533.0,
+                    1585.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    925.0,
+                    1570.0,
+                    948.0,
+                    1570.0,
+                    948.0,
+                    1585.0,
+                    925.0,
+                    1585.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    583.0,
+                    1582.0,
+                    606.0,
+                    1582.0,
+                    606.0,
+                    1597.0,
+                    583.0,
+                    1597.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    975.0,
+                    1582.0,
+                    998.0,
+                    1582.0,
+                    998.0,
+                    1597.0,
+                    975.0,
+                    1597.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    571.0,
+                    1596.0,
+                    603.0,
+                    1596.0,
+                    603.0,
+                    1615.0,
+                    571.0,
+                    1615.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    632.0,
+                    1595.0,
+                    660.0,
+                    1595.0,
+                    660.0,
+                    1608.0,
+                    632.0,
+                    1608.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    963.0,
+                    1596.0,
+                    995.0,
+                    1596.0,
+                    995.0,
+                    1615.0,
+                    963.0,
+                    1615.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1024.0,
+                    1595.0,
+                    1051.0,
+                    1595.0,
+                    1051.0,
+                    1608.0,
+                    1024.0,
+                    1608.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1109.0,
+                    1600.0,
+                    1117.0,
+                    1600.0,
+                    1117.0,
+                    1608.0,
+                    1109.0,
+                    1608.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1207.0,
+                    1397.5,
+                    1228.0,
+                    1397.5,
+                    1228.0,
+                    1411.5,
+                    1207.0,
+                    1411.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    753.0,
+                    1578.5,
+                    811.0,
+                    1578.5,
+                    811.0,
+                    1594.5,
+                    753.0,
+                    1594.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1145.0,
+                    1578.5,
+                    1204.0,
+                    1578.5,
+                    1204.0,
+                    1594.5,
+                    1145.0,
+                    1594.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    218.0,
+                    1411.0,
+                    218.0,
+                    1411.0,
+                    274.0,
+                    294.0,
+                    274.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    276.0,
+                    594.0,
+                    276.0,
+                    594.0,
+                    325.0,
+                    294.0,
+                    325.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1647.0,
+                    1410.0,
+                    1647.0,
+                    1410.0,
+                    1686.0,
+                    293.0,
+                    1686.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1679.0,
+                    1405.0,
+                    1679.0,
+                    1405.0,
+                    1716.0,
+                    294.0,
+                    1716.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1712.0,
+                    1406.0,
+                    1712.0,
+                    1406.0,
+                    1743.0,
+                    295.0,
+                    1743.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1128.0,
+                    579.0,
+                    1128.0,
+                    579.0,
+                    1175.0,
+                    294.0,
+                    1175.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    770.0,
+                    518.0,
+                    932.0,
+                    518.0,
+                    932.0,
+                    556.0,
+                    770.0,
+                    556.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    73.0,
+                    857.0,
+                    73.0,
+                    857.0,
+                    108.0,
+                    297.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    840.0,
+                    2088.0,
+                    857.0,
+                    2088.0,
+                    857.0,
+                    2116.0,
+                    840.0,
+                    2116.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    394.0,
+                    588.0,
+                    1304.0,
+                    588.0,
+                    1304.0,
+                    622.0,
+                    394.0,
+                    622.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    395.0,
+                    621.0,
+                    1304.0,
+                    621.0,
+                    1304.0,
+                    653.0,
+                    395.0,
+                    653.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    394.0,
+                    649.0,
+                    1307.0,
+                    649.0,
+                    1307.0,
+                    684.0,
+                    394.0,
+                    684.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    393.0,
+                    680.0,
+                    1305.0,
+                    680.0,
+                    1305.0,
+                    713.0,
+                    393.0,
+                    713.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    394.0,
+                    710.0,
+                    1307.0,
+                    710.0,
+                    1307.0,
+                    745.0,
+                    394.0,
+                    745.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    394.0,
+                    741.0,
+                    1304.0,
+                    741.0,
+                    1304.0,
+                    773.0,
+                    394.0,
+                    773.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    393.0,
+                    770.0,
+                    1306.0,
+                    770.0,
+                    1306.0,
+                    806.0,
+                    393.0,
+                    806.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    392.0,
+                    802.0,
+                    1306.0,
+                    802.0,
+                    1306.0,
+                    836.0,
+                    392.0,
+                    836.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    393.0,
+                    834.0,
+                    1306.0,
+                    834.0,
+                    1306.0,
+                    865.0,
+                    393.0,
+                    865.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    393.0,
+                    861.0,
+                    1306.0,
+                    861.0,
+                    1306.0,
+                    898.0,
+                    393.0,
+                    898.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    394.0,
+                    894.0,
+                    1305.0,
+                    894.0,
+                    1305.0,
+                    926.0,
+                    394.0,
+                    926.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    392.0,
+                    921.0,
+                    1306.0,
+                    921.0,
+                    1306.0,
+                    960.0,
+                    392.0,
+                    960.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    393.0,
+                    954.0,
+                    1308.0,
+                    954.0,
+                    1308.0,
+                    988.0,
+                    393.0,
+                    988.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    393.0,
+                    984.0,
+                    1305.0,
+                    984.0,
+                    1305.0,
+                    1017.0,
+                    393.0,
+                    1017.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    392.0,
+                    1011.0,
+                    1306.0,
+                    1011.0,
+                    1306.0,
+                    1052.0,
+                    392.0,
+                    1052.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    392.0,
+                    1041.0,
+                    1229.0,
+                    1041.0,
+                    1229.0,
+                    1085.0,
+                    392.0,
+                    1085.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1183.0,
+                    1405.0,
+                    1183.0,
+                    1405.0,
+                    1222.0,
+                    294.0,
+                    1222.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1213.0,
+                    1407.0,
+                    1213.0,
+                    1407.0,
+                    1251.0,
+                    293.0,
+                    1251.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1245.0,
+                    1407.0,
+                    1245.0,
+                    1407.0,
+                    1283.0,
+                    293.0,
+                    1283.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1277.0,
+                    1407.0,
+                    1277.0,
+                    1407.0,
+                    1310.0,
+                    296.0,
+                    1310.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1307.0,
+                    1042.0,
+                    1307.0,
+                    1042.0,
+                    1343.0,
+                    294.0,
+                    1343.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1770.0,
+                    1405.0,
+                    1770.0,
+                    1405.0,
+                    1811.0,
+                    293.0,
+                    1811.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1801.0,
+                    1404.0,
+                    1801.0,
+                    1404.0,
+                    1838.0,
+                    293.0,
+                    1838.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1834.0,
+                    1405.0,
+                    1834.0,
+                    1405.0,
+                    1871.0,
+                    293.0,
+                    1871.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1866.0,
+                    1404.0,
+                    1866.0,
+                    1404.0,
+                    1899.0,
+                    294.0,
+                    1899.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1895.0,
+                    789.0,
+                    1895.0,
+                    789.0,
+                    1924.0,
+                    296.0,
+                    1924.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1941.0,
+                    1406.0,
+                    1941.0,
+                    1406.0,
+                    1979.0,
+                    294.0,
+                    1979.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1973.0,
+                    1404.0,
+                    1973.0,
+                    1404.0,
+                    2007.0,
+                    296.0,
+                    2007.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    2002.0,
+                    1406.0,
+                    2002.0,
+                    1406.0,
+                    2038.0,
+                    294.0,
+                    2038.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    315.0,
+                    377.0,
+                    560.0,
+                    377.0,
+                    560.0,
+                    410.0,
+                    315.0,
+                    410.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    311.0,
+                    407.0,
+                    682.0,
+                    407.0,
+                    682.0,
+                    440.0,
+                    311.0,
+                    440.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 0,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    966,
+                    1404,
+                    966,
+                    1404,
+                    1331,
+                    298,
+                    1331
+                ],
+                "score": 0.983
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    568,
+                    1404,
+                    568,
+                    1404,
+                    781,
+                    298,
+                    781
+                ],
+                "score": 0.977
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    338,
+                    1403,
+                    338,
+                    1403,
+                    552,
+                    298,
+                    552
+                ],
+                "score": 0.976
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    797,
+                    1403,
+                    797,
+                    1403,
+                    950,
+                    299,
+                    950
+                ],
+                "score": 0.973
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1748,
+                    1403,
+                    1748,
+                    1403,
+                    1871,
+                    298,
+                    1871
+                ],
+                "score": 0.97
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1457,
+                    1405,
+                    1457,
+                    1405,
+                    1732,
+                    297,
+                    1732
+                ],
+                "score": 0.964
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    230,
+                    1402,
+                    230,
+                    1402,
+                    323,
+                    299,
+                    323
+                ],
+                "score": 0.958
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    297,
+                    1893,
+                    1404,
+                    1893,
+                    1404,
+                    2033,
+                    297,
+                    2033
+                ],
+                "score": 0.954
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    298,
+                    1418,
+                    999,
+                    1418,
+                    999,
+                    1447,
+                    298,
+                    1447
+                ],
+                "score": 0.898
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    299,
+                    1362,
+                    557,
+                    1362,
+                    557,
+                    1397,
+                    299,
+                    1397
+                ],
+                "score": 0.882
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    300,
+                    76,
+                    852,
+                    76,
+                    852,
+                    104,
+                    300,
+                    104
+                ],
+                "score": 0.861
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    841,
+                    2089,
+                    858,
+                    2089,
+                    858,
+                    2112,
+                    841,
+                    2112
+                ],
+                "score": 0.73
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    493,
+                    1780,
+                    608,
+                    1780,
+                    608,
+                    1810,
+                    493,
+                    1810
+                ],
+                "score": 0.92,
+                "latex": "L _ { 1 } , . . . , L _ { n }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    572,
+                    1840,
+                    626,
+                    1840,
+                    626,
+                    1868,
+                    572,
+                    1868
+                ],
+                "score": 0.86,
+                "latex": "20 \\%"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1282,
+                    1840,
+                    1339,
+                    1840,
+                    1339,
+                    1869,
+                    1282,
+                    1869
+                ],
+                "score": 0.84,
+                "latex": "20 \\%"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    958,
+                    1780,
+                    981,
+                    1780,
+                    981,
+                    1807,
+                    958,
+                    1807
+                ],
+                "score": 0.65,
+                "latex": "T"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1171,
+                    1488,
+                    1219,
+                    1488,
+                    1219,
+                    1518,
+                    1171,
+                    1518
+                ],
+                "score": 0.42,
+                "latex": "\\mathbf { \\Pi } ( \\mathbf { I I L } )"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    333.0,
+                    1890.0,
+                    1405.0,
+                    1890.0,
+                    1405.0,
+                    1928.0,
+                    333.0,
+                    1928.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1922.0,
+                    1405.0,
+                    1922.0,
+                    1405.0,
+                    1953.0,
+                    295.0,
+                    1953.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1947.0,
+                    1407.0,
+                    1947.0,
+                    1407.0,
+                    1982.0,
+                    294.0,
+                    1982.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1969.0,
+                    1410.0,
+                    1969.0,
+                    1410.0,
+                    2016.0,
+                    292.0,
+                    2016.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    2004.0,
+                    724.0,
+                    2004.0,
+                    724.0,
+                    2039.0,
+                    294.0,
+                    2039.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1415.0,
+                    1003.0,
+                    1415.0,
+                    1003.0,
+                    1450.0,
+                    295.0,
+                    1450.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1358.0,
+                    561.0,
+                    1358.0,
+                    561.0,
+                    1405.0,
+                    291.0,
+                    1405.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    73.0,
+                    856.0,
+                    73.0,
+                    856.0,
+                    108.0,
+                    297.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    839.0,
+                    2086.0,
+                    862.0,
+                    2086.0,
+                    862.0,
+                    2121.0,
+                    839.0,
+                    2121.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    964.0,
+                    1405.0,
+                    964.0,
+                    1405.0,
+                    1002.0,
+                    293.0,
+                    1002.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    996.0,
+                    1406.0,
+                    996.0,
+                    1406.0,
+                    1032.0,
+                    293.0,
+                    1032.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1029.0,
+                    1405.0,
+                    1029.0,
+                    1405.0,
+                    1059.0,
+                    296.0,
+                    1059.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1059.0,
+                    1406.0,
+                    1059.0,
+                    1406.0,
+                    1092.0,
+                    293.0,
+                    1092.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1089.0,
+                    1406.0,
+                    1089.0,
+                    1406.0,
+                    1123.0,
+                    294.0,
+                    1123.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1119.0,
+                    1406.0,
+                    1119.0,
+                    1406.0,
+                    1153.0,
+                    293.0,
+                    1153.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1150.0,
+                    1406.0,
+                    1150.0,
+                    1406.0,
+                    1184.0,
+                    294.0,
+                    1184.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1176.0,
+                    1406.0,
+                    1176.0,
+                    1406.0,
+                    1218.0,
+                    292.0,
+                    1218.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1208.0,
+                    1406.0,
+                    1208.0,
+                    1406.0,
+                    1250.0,
+                    291.0,
+                    1250.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1239.0,
+                    1406.0,
+                    1239.0,
+                    1406.0,
+                    1276.0,
+                    293.0,
+                    1276.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1272.0,
+                    1406.0,
+                    1272.0,
+                    1406.0,
+                    1306.0,
+                    294.0,
+                    1306.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1300.0,
+                    1409.0,
+                    1300.0,
+                    1409.0,
+                    1336.0,
+                    293.0,
+                    1336.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    564.0,
+                    1408.0,
+                    564.0,
+                    1408.0,
+                    606.0,
+                    293.0,
+                    606.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    600.0,
+                    1407.0,
+                    600.0,
+                    1407.0,
+                    632.0,
+                    293.0,
+                    632.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    630.0,
+                    1406.0,
+                    630.0,
+                    1406.0,
+                    665.0,
+                    294.0,
+                    665.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    659.0,
+                    1405.0,
+                    659.0,
+                    1405.0,
+                    694.0,
+                    292.0,
+                    694.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    687.0,
+                    1406.0,
+                    687.0,
+                    1406.0,
+                    728.0,
+                    292.0,
+                    728.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    718.0,
+                    1405.0,
+                    718.0,
+                    1405.0,
+                    759.0,
+                    292.0,
+                    759.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    752.0,
+                    998.0,
+                    752.0,
+                    998.0,
+                    783.0,
+                    293.0,
+                    783.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    335.0,
+                    1407.0,
+                    335.0,
+                    1407.0,
+                    375.0,
+                    293.0,
+                    375.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    365.0,
+                    1404.0,
+                    365.0,
+                    1404.0,
+                    406.0,
+                    293.0,
+                    406.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    396.0,
+                    1405.0,
+                    396.0,
+                    1405.0,
+                    434.0,
+                    293.0,
+                    434.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    429.0,
+                    1407.0,
+                    429.0,
+                    1407.0,
+                    465.0,
+                    292.0,
+                    465.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    458.0,
+                    1405.0,
+                    458.0,
+                    1405.0,
+                    495.0,
+                    293.0,
+                    495.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    487.0,
+                    1404.0,
+                    487.0,
+                    1404.0,
+                    528.0,
+                    292.0,
+                    528.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    524.0,
+                    759.0,
+                    524.0,
+                    759.0,
+                    557.0,
+                    294.0,
+                    557.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    793.0,
+                    1407.0,
+                    793.0,
+                    1407.0,
+                    837.0,
+                    293.0,
+                    837.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    829.0,
+                    1404.0,
+                    829.0,
+                    1404.0,
+                    862.0,
+                    297.0,
+                    862.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    858.0,
+                    1405.0,
+                    858.0,
+                    1405.0,
+                    895.0,
+                    294.0,
+                    895.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    889.0,
+                    1405.0,
+                    889.0,
+                    1405.0,
+                    924.0,
+                    294.0,
+                    924.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    916.0,
+                    1356.0,
+                    916.0,
+                    1356.0,
+                    958.0,
+                    293.0,
+                    958.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1747.0,
+                    1407.0,
+                    1747.0,
+                    1407.0,
+                    1784.0,
+                    293.0,
+                    1784.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1777.0,
+                    492.0,
+                    1777.0,
+                    492.0,
+                    1815.0,
+                    292.0,
+                    1815.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    609.0,
+                    1777.0,
+                    957.0,
+                    1777.0,
+                    957.0,
+                    1815.0,
+                    609.0,
+                    1815.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    982.0,
+                    1777.0,
+                    1404.0,
+                    1777.0,
+                    1404.0,
+                    1815.0,
+                    982.0,
+                    1815.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1807.0,
+                    1407.0,
+                    1807.0,
+                    1407.0,
+                    1844.0,
+                    293.0,
+                    1844.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1840.0,
+                    571.0,
+                    1840.0,
+                    571.0,
+                    1872.0,
+                    296.0,
+                    1872.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    627.0,
+                    1840.0,
+                    1281.0,
+                    1840.0,
+                    1281.0,
+                    1872.0,
+                    627.0,
+                    1872.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1340.0,
+                    1840.0,
+                    1405.0,
+                    1840.0,
+                    1405.0,
+                    1872.0,
+                    1340.0,
+                    1872.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1456.0,
+                    1405.0,
+                    1456.0,
+                    1405.0,
+                    1492.0,
+                    295.0,
+                    1492.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1485.0,
+                    1170.0,
+                    1485.0,
+                    1170.0,
+                    1523.0,
+                    294.0,
+                    1523.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1220.0,
+                    1485.0,
+                    1403.0,
+                    1485.0,
+                    1403.0,
+                    1523.0,
+                    1220.0,
+                    1523.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1514.0,
+                    1406.0,
+                    1514.0,
+                    1406.0,
+                    1558.0,
+                    292.0,
+                    1558.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1548.0,
+                    1405.0,
+                    1548.0,
+                    1405.0,
+                    1583.0,
+                    295.0,
+                    1583.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1575.0,
+                    1407.0,
+                    1575.0,
+                    1407.0,
+                    1618.0,
+                    292.0,
+                    1618.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1609.0,
+                    1407.0,
+                    1609.0,
+                    1407.0,
+                    1644.0,
+                    292.0,
+                    1644.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1640.0,
+                    1407.0,
+                    1640.0,
+                    1407.0,
+                    1677.0,
+                    295.0,
+                    1677.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1669.0,
+                    1410.0,
+                    1669.0,
+                    1410.0,
+                    1709.0,
+                    294.0,
+                    1709.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1697.0,
+                    1187.0,
+                    1697.0,
+                    1187.0,
+                    1738.0,
+                    294.0,
+                    1738.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    231.0,
+                    1407.0,
+                    231.0,
+                    1407.0,
+                    264.0,
+                    294.0,
+                    264.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    261.0,
+                    1407.0,
+                    261.0,
+                    1407.0,
+                    298.0,
+                    294.0,
+                    298.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    292.0,
+                    1094.0,
+                    292.0,
+                    1094.0,
+                    326.0,
+                    295.0,
+                    326.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 1,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    366,
+                    1406,
+                    366,
+                    1406,
+                    825,
+                    297,
+                    825
+                ],
+                "score": 0.983
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1830,
+                    1404,
+                    1830,
+                    1404,
+                    2044,
+                    298,
+                    2044
+                ],
+                "score": 0.979
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1630,
+                    1403,
+                    1630,
+                    1403,
+                    1814,
+                    298,
+                    1814
+                ],
+                "score": 0.978
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    921,
+                    1406,
+                    921,
+                    1406,
+                    1044,
+                    298,
+                    1044
+                ],
+                "score": 0.972
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    292,
+                    229,
+                    1404,
+                    229,
+                    1404,
+                    293,
+                    292,
+                    293
+                ],
+                "score": 0.94
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    299,
+                    862,
+                    997,
+                    862,
+                    997,
+                    898,
+                    299,
+                    898
+                ],
+                "score": 0.931
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    299,
+                    325,
+                    1063,
+                    325,
+                    1063,
+                    354,
+                    299,
+                    354
+                ],
+                "score": 0.891
+            },
+            {
+                "category_id": 3,
+                "poly": [
+                    296,
+                    1087,
+                    1406,
+                    1087,
+                    1406,
+                    1379,
+                    296,
+                    1379
+                ],
+                "score": 0.802
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    300,
+                    76,
+                    853,
+                    76,
+                    853,
+                    104,
+                    300,
+                    104
+                ],
+                "score": 0.635
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    841,
+                    2088,
+                    858,
+                    2088,
+                    858,
+                    2112,
+                    841,
+                    2112
+                ],
+                "score": 0.611
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    295,
+                    1407,
+                    1405,
+                    1407,
+                    1405,
+                    1591,
+                    295,
+                    1591
+                ],
+                "score": 0.605
+            },
+            {
+                "category_id": 4,
+                "poly": [
+                    295,
+                    1407,
+                    1405,
+                    1407,
+                    1405,
+                    1591,
+                    295,
+                    1591
+                ],
+                "score": 0.592
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    841,
+                    2088,
+                    859,
+                    2088,
+                    859,
+                    2112,
+                    841,
+                    2112
+                ],
+                "score": 0.467
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    299,
+                    76,
+                    854,
+                    76,
+                    854,
+                    104,
+                    299,
+                    104
+                ],
+                "score": 0.398
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1002,
+                    672,
+                    1086,
+                    672,
+                    1086,
+                    705,
+                    1002,
+                    705
+                ],
+                "score": 0.93,
+                "latex": "P ( \\boldsymbol { y } | \\mathbf { x } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    598,
+                    429,
+                    687,
+                    429,
+                    687,
+                    461,
+                    598,
+                    461
+                ],
+                "score": 0.92,
+                "latex": "P ( \\mathbf { x } , y )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    784,
+                    489,
+                    847,
+                    489,
+                    847,
+                    523,
+                    784,
+                    523
+                ],
+                "score": 0.91,
+                "latex": "P ( \\mathbf { x } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1061,
+                    1439,
+                    1080,
+                    1439,
+                    1080,
+                    1470,
+                    1061,
+                    1470
+                ],
+                "score": 0.83,
+                "latex": "\\hat { y }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    906,
+                    1531,
+                    925,
+                    1531,
+                    925,
+                    1561,
+                    906,
+                    1561
+                ],
+                "score": 0.81,
+                "latex": "\\hat { y }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    904,
+                    1443,
+                    923,
+                    1443,
+                    923,
+                    1465,
+                    904,
+                    1465
+                ],
+                "score": 0.74,
+                "latex": "x"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    656,
+                    403,
+                    674,
+                    403,
+                    674,
+                    430,
+                    656,
+                    430
+                ],
+                "score": 0.73,
+                "latex": "y"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    718,
+                    1531,
+                    742,
+                    1531,
+                    742,
+                    1558,
+                    718,
+                    1558
+                ],
+                "score": 0.67,
+                "latex": "A"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1295,
+                    1415,
+                    1312,
+                    1415,
+                    1312,
+                    1435,
+                    1295,
+                    1435
+                ],
+                "score": 0.66,
+                "latex": "a"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    297,
+                    1443,
+                    314,
+                    1443,
+                    314,
+                    1465,
+                    297,
+                    1465
+                ],
+                "score": 0.64,
+                "latex": "r"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1195,
+                    1415,
+                    1211,
+                    1415,
+                    1211,
+                    1436,
+                    1195,
+                    1436
+                ],
+                "score": 0.63,
+                "latex": "s"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    483,
+                    1500,
+                    538,
+                    1500,
+                    538,
+                    1530,
+                    483,
+                    1530
+                ],
+                "score": 0.59,
+                "latex": "^ { ( \\mathbf { b } , \\mathbf { c } ) }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1178,
+                    1439,
+                    1191,
+                    1439,
+                    1191,
+                    1465,
+                    1178,
+                    1465
+                ],
+                "score": 0.44,
+                "latex": "l"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    509,
+                    403,
+                    531,
+                    403,
+                    531,
+                    427,
+                    509,
+                    427
+                ],
+                "score": 0.39,
+                "latex": "\\mathbf { x }"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    859.0,
+                    1000.0,
+                    859.0,
+                    1000.0,
+                    904.0,
+                    291.0,
+                    904.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    321.0,
+                    1068.0,
+                    321.0,
+                    1068.0,
+                    359.0,
+                    294.0,
+                    359.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    460.0,
+                    1107.0,
+                    477.0,
+                    1107.0,
+                    477.0,
+                    1120.0,
+                    460.0,
+                    1120.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    745.0,
+                    1129.0,
+                    765.0,
+                    1129.0,
+                    765.0,
+                    1148.0,
+                    745.0,
+                    1148.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    957.0,
+                    1131.0,
+                    971.0,
+                    1131.0,
+                    971.0,
+                    1148.0,
+                    957.0,
+                    1148.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1017.0,
+                    1129.0,
+                    1048.0,
+                    1129.0,
+                    1048.0,
+                    1148.0,
+                    1017.0,
+                    1148.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1237.0,
+                    1133.0,
+                    1249.0,
+                    1133.0,
+                    1249.0,
+                    1142.0,
+                    1237.0,
+                    1142.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    381.0,
+                    1163.0,
+                    414.0,
+                    1163.0,
+                    414.0,
+                    1186.0,
+                    381.0,
+                    1186.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    668.0,
+                    1162.0,
+                    704.0,
+                    1162.0,
+                    704.0,
+                    1190.0,
+                    668.0,
+                    1190.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    741.0,
+                    1167.0,
+                    763.0,
+                    1167.0,
+                    763.0,
+                    1182.0,
+                    741.0,
+                    1182.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    946.0,
+                    1166.0,
+                    977.0,
+                    1166.0,
+                    977.0,
+                    1190.0,
+                    946.0,
+                    1190.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1017.0,
+                    1169.0,
+                    1043.0,
+                    1169.0,
+                    1043.0,
+                    1188.0,
+                    1017.0,
+                    1188.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    451.0,
+                    1224.0,
+                    484.0,
+                    1224.0,
+                    484.0,
+                    1254.0,
+                    451.0,
+                    1254.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    739.0,
+                    1229.0,
+                    760.0,
+                    1229.0,
+                    760.0,
+                    1248.0,
+                    739.0,
+                    1248.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    290.0,
+                    1259.0,
+                    382.0,
+                    1259.0,
+                    382.0,
+                    1294.0,
+                    290.0,
+                    1294.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    588.0,
+                    1260.0,
+                    762.0,
+                    1260.0,
+                    762.0,
+                    1295.0,
+                    588.0,
+                    1295.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    863.0,
+                    1260.0,
+                    1057.0,
+                    1260.0,
+                    1057.0,
+                    1292.0,
+                    863.0,
+                    1292.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1141.0,
+                    1260.0,
+                    1380.0,
+                    1260.0,
+                    1380.0,
+                    1294.0,
+                    1141.0,
+                    1294.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1284.0,
+                    578.0,
+                    1284.0,
+                    578.0,
+                    1323.0,
+                    292.0,
+                    1323.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    585.0,
+                    1287.0,
+                    854.0,
+                    1287.0,
+                    854.0,
+                    1321.0,
+                    585.0,
+                    1321.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    864.0,
+                    1287.0,
+                    1132.0,
+                    1287.0,
+                    1132.0,
+                    1321.0,
+                    864.0,
+                    1321.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1140.0,
+                    1287.0,
+                    1404.0,
+                    1287.0,
+                    1404.0,
+                    1321.0,
+                    1140.0,
+                    1321.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1318.0,
+                    577.0,
+                    1318.0,
+                    577.0,
+                    1350.0,
+                    291.0,
+                    1350.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    581.0,
+                    1318.0,
+                    854.0,
+                    1318.0,
+                    854.0,
+                    1352.0,
+                    581.0,
+                    1352.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    861.0,
+                    1316.0,
+                    1227.0,
+                    1316.0,
+                    1227.0,
+                    1351.0,
+                    861.0,
+                    1351.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1343.0,
+                    406.0,
+                    1343.0,
+                    406.0,
+                    1379.0,
+                    293.0,
+                    1379.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    584.0,
+                    1341.0,
+                    749.0,
+                    1341.0,
+                    749.0,
+                    1379.0,
+                    584.0,
+                    1379.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    865.0,
+                    1343.0,
+                    1026.0,
+                    1343.0,
+                    1026.0,
+                    1375.0,
+                    865.0,
+                    1375.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    465.75,
+                    1122.5,
+                    473.75,
+                    1122.5,
+                    473.75,
+                    1160.5,
+                    465.75,
+                    1160.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    462.0,
+                    1168.0,
+                    469.0,
+                    1168.0,
+                    469.0,
+                    1180.0,
+                    462.0,
+                    1180.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    433.0,
+                    1208.0,
+                    451.0,
+                    1208.0,
+                    451.0,
+                    1220.0,
+                    433.0,
+                    1220.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    73.0,
+                    856.0,
+                    73.0,
+                    856.0,
+                    108.0,
+                    297.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    839.0,
+                    2085.0,
+                    860.0,
+                    2085.0,
+                    860.0,
+                    2116.0,
+                    839.0,
+                    2116.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1406.0,
+                    1194.0,
+                    1406.0,
+                    1194.0,
+                    1442.0,
+                    295.0,
+                    1442.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1212.0,
+                    1406.0,
+                    1294.0,
+                    1406.0,
+                    1294.0,
+                    1442.0,
+                    1212.0,
+                    1442.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1313.0,
+                    1406.0,
+                    1406.0,
+                    1406.0,
+                    1406.0,
+                    1442.0,
+                    1313.0,
+                    1442.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    315.0,
+                    1439.0,
+                    903.0,
+                    1439.0,
+                    903.0,
+                    1471.0,
+                    315.0,
+                    1471.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    924.0,
+                    1439.0,
+                    1060.0,
+                    1439.0,
+                    1060.0,
+                    1471.0,
+                    924.0,
+                    1471.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1081.0,
+                    1439.0,
+                    1177.0,
+                    1439.0,
+                    1177.0,
+                    1471.0,
+                    1081.0,
+                    1471.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1192.0,
+                    1439.0,
+                    1403.0,
+                    1439.0,
+                    1403.0,
+                    1471.0,
+                    1192.0,
+                    1471.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1469.0,
+                    1406.0,
+                    1469.0,
+                    1406.0,
+                    1504.0,
+                    294.0,
+                    1504.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1500.0,
+                    482.0,
+                    1500.0,
+                    482.0,
+                    1532.0,
+                    295.0,
+                    1532.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    539.0,
+                    1500.0,
+                    1402.0,
+                    1500.0,
+                    1402.0,
+                    1532.0,
+                    539.0,
+                    1532.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1529.0,
+                    717.0,
+                    1529.0,
+                    717.0,
+                    1565.0,
+                    294.0,
+                    1565.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    743.0,
+                    1529.0,
+                    905.0,
+                    1529.0,
+                    905.0,
+                    1565.0,
+                    743.0,
+                    1565.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    926.0,
+                    1529.0,
+                    1406.0,
+                    1529.0,
+                    1406.0,
+                    1565.0,
+                    926.0,
+                    1565.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1561.0,
+                    1024.0,
+                    1561.0,
+                    1024.0,
+                    1593.0,
+                    295.0,
+                    1593.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    838.0,
+                    2085.0,
+                    862.0,
+                    2085.0,
+                    862.0,
+                    2118.0,
+                    838.0,
+                    2118.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    73.0,
+                    857.0,
+                    73.0,
+                    857.0,
+                    108.0,
+                    297.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    367.0,
+                    1406.0,
+                    367.0,
+                    1406.0,
+                    402.0,
+                    295.0,
+                    402.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    396.0,
+                    508.0,
+                    396.0,
+                    508.0,
+                    436.0,
+                    292.0,
+                    436.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    532.0,
+                    396.0,
+                    655.0,
+                    396.0,
+                    655.0,
+                    436.0,
+                    532.0,
+                    436.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    675.0,
+                    396.0,
+                    1407.0,
+                    396.0,
+                    1407.0,
+                    436.0,
+                    675.0,
+                    436.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    428.0,
+                    597.0,
+                    428.0,
+                    597.0,
+                    463.0,
+                    294.0,
+                    463.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    688.0,
+                    428.0,
+                    1407.0,
+                    428.0,
+                    1407.0,
+                    463.0,
+                    688.0,
+                    463.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    458.0,
+                    1406.0,
+                    458.0,
+                    1406.0,
+                    493.0,
+                    296.0,
+                    493.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    489.0,
+                    783.0,
+                    489.0,
+                    783.0,
+                    524.0,
+                    295.0,
+                    524.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    848.0,
+                    489.0,
+                    1406.0,
+                    489.0,
+                    1406.0,
+                    524.0,
+                    848.0,
+                    524.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    519.0,
+                    1407.0,
+                    519.0,
+                    1407.0,
+                    553.0,
+                    292.0,
+                    553.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    547.0,
+                    1407.0,
+                    547.0,
+                    1407.0,
+                    588.0,
+                    292.0,
+                    588.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    579.0,
+                    1406.0,
+                    579.0,
+                    1406.0,
+                    617.0,
+                    292.0,
+                    617.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    612.0,
+                    1408.0,
+                    612.0,
+                    1408.0,
+                    645.0,
+                    292.0,
+                    645.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    643.0,
+                    1404.0,
+                    643.0,
+                    1404.0,
+                    674.0,
+                    295.0,
+                    674.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    673.0,
+                    1001.0,
+                    673.0,
+                    1001.0,
+                    708.0,
+                    294.0,
+                    708.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1087.0,
+                    673.0,
+                    1406.0,
+                    673.0,
+                    1406.0,
+                    708.0,
+                    1087.0,
+                    708.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    702.0,
+                    1409.0,
+                    702.0,
+                    1409.0,
+                    740.0,
+                    292.0,
+                    740.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    732.0,
+                    1404.0,
+                    732.0,
+                    1404.0,
+                    766.0,
+                    294.0,
+                    766.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    764.0,
+                    1407.0,
+                    764.0,
+                    1407.0,
+                    799.0,
+                    295.0,
+                    799.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    794.0,
+                    1328.0,
+                    794.0,
+                    1328.0,
+                    829.0,
+                    294.0,
+                    829.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1829.0,
+                    1408.0,
+                    1829.0,
+                    1408.0,
+                    1867.0,
+                    293.0,
+                    1867.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1860.0,
+                    1406.0,
+                    1860.0,
+                    1406.0,
+                    1898.0,
+                    294.0,
+                    1898.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1892.0,
+                    1406.0,
+                    1892.0,
+                    1406.0,
+                    1926.0,
+                    294.0,
+                    1926.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1922.0,
+                    1402.0,
+                    1922.0,
+                    1402.0,
+                    1957.0,
+                    294.0,
+                    1957.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1948.0,
+                    1405.0,
+                    1948.0,
+                    1405.0,
+                    1989.0,
+                    292.0,
+                    1989.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1982.0,
+                    1406.0,
+                    1982.0,
+                    1406.0,
+                    2017.0,
+                    294.0,
+                    2017.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    2012.0,
+                    1279.0,
+                    2012.0,
+                    1279.0,
+                    2049.0,
+                    293.0,
+                    2049.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1629.0,
+                    1407.0,
+                    1629.0,
+                    1407.0,
+                    1665.0,
+                    296.0,
+                    1665.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1661.0,
+                    1405.0,
+                    1661.0,
+                    1405.0,
+                    1697.0,
+                    294.0,
+                    1697.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1689.0,
+                    1407.0,
+                    1689.0,
+                    1407.0,
+                    1726.0,
+                    293.0,
+                    1726.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1723.0,
+                    1405.0,
+                    1723.0,
+                    1405.0,
+                    1755.0,
+                    296.0,
+                    1755.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1754.0,
+                    1405.0,
+                    1754.0,
+                    1405.0,
+                    1785.0,
+                    296.0,
+                    1785.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1783.0,
+                    1287.0,
+                    1783.0,
+                    1287.0,
+                    1816.0,
+                    293.0,
+                    1816.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    919.0,
+                    1406.0,
+                    919.0,
+                    1406.0,
+                    956.0,
+                    292.0,
+                    956.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    290.0,
+                    945.0,
+                    1411.0,
+                    945.0,
+                    1411.0,
+                    992.0,
+                    290.0,
+                    992.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    982.0,
+                    1404.0,
+                    982.0,
+                    1404.0,
+                    1017.0,
+                    293.0,
+                    1017.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1012.0,
+                    1180.0,
+                    1012.0,
+                    1180.0,
+                    1048.0,
+                    293.0,
+                    1048.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    226.0,
+                    1405.0,
+                    226.0,
+                    1405.0,
+                    266.0,
+                    294.0,
+                    266.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    262.0,
+                    612.0,
+                    262.0,
+                    612.0,
+                    294.0,
+                    295.0,
+                    294.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1406.0,
+                    1194.0,
+                    1406.0,
+                    1194.0,
+                    1442.0,
+                    295.0,
+                    1442.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1212.0,
+                    1406.0,
+                    1294.0,
+                    1406.0,
+                    1294.0,
+                    1442.0,
+                    1212.0,
+                    1442.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1313.0,
+                    1406.0,
+                    1406.0,
+                    1406.0,
+                    1406.0,
+                    1442.0,
+                    1313.0,
+                    1442.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    315.0,
+                    1439.0,
+                    903.0,
+                    1439.0,
+                    903.0,
+                    1471.0,
+                    315.0,
+                    1471.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    924.0,
+                    1439.0,
+                    1060.0,
+                    1439.0,
+                    1060.0,
+                    1471.0,
+                    924.0,
+                    1471.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1081.0,
+                    1439.0,
+                    1177.0,
+                    1439.0,
+                    1177.0,
+                    1471.0,
+                    1081.0,
+                    1471.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1192.0,
+                    1439.0,
+                    1403.0,
+                    1439.0,
+                    1403.0,
+                    1471.0,
+                    1192.0,
+                    1471.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1469.0,
+                    1406.0,
+                    1469.0,
+                    1406.0,
+                    1504.0,
+                    294.0,
+                    1504.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1500.0,
+                    482.0,
+                    1500.0,
+                    482.0,
+                    1532.0,
+                    295.0,
+                    1532.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    539.0,
+                    1500.0,
+                    1402.0,
+                    1500.0,
+                    1402.0,
+                    1532.0,
+                    539.0,
+                    1532.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1529.0,
+                    717.0,
+                    1529.0,
+                    717.0,
+                    1565.0,
+                    294.0,
+                    1565.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    743.0,
+                    1529.0,
+                    905.0,
+                    1529.0,
+                    905.0,
+                    1565.0,
+                    743.0,
+                    1565.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    926.0,
+                    1529.0,
+                    1406.0,
+                    1529.0,
+                    1406.0,
+                    1565.0,
+                    926.0,
+                    1565.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1561.0,
+                    1024.0,
+                    1561.0,
+                    1024.0,
+                    1593.0,
+                    295.0,
+                    1593.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 2,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1636,
+                    1405,
+                    1636,
+                    1405,
+                    2034,
+                    298,
+                    2034
+                ],
+                "score": 0.982
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    517,
+                    881,
+                    517,
+                    881,
+                    791,
+                    299,
+                    791
+                ],
+                "score": 0.98
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    807,
+                    1405,
+                    807,
+                    1405,
+                    1022,
+                    298,
+                    1022
+                ],
+                "score": 0.979
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1239,
+                    1404,
+                    1239,
+                    1404,
+                    1423,
+                    298,
+                    1423
+                ],
+                "score": 0.979
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    286,
+                    1404,
+                    286,
+                    1404,
+                    501,
+                    297,
+                    501
+                ],
+                "score": 0.978
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    1438,
+                    1405,
+                    1438,
+                    1405,
+                    1622,
+                    299,
+                    1622
+                ],
+                "score": 0.977
+            },
+            {
+                "category_id": 3,
+                "poly": [
+                    908,
+                    518,
+                    1398,
+                    518,
+                    1398,
+                    692,
+                    908,
+                    692
+                ],
+                "score": 0.967
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1099,
+                    1403,
+                    1099,
+                    1403,
+                    1223,
+                    297,
+                    1223
+                ],
+                "score": 0.966
+            },
+            {
+                "category_id": 4,
+                "poly": [
+                    901,
+                    723,
+                    1404,
+                    723,
+                    1404,
+                    785,
+                    901,
+                    785
+                ],
+                "score": 0.944
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    299,
+                    226,
+                    527,
+                    226,
+                    527,
+                    261,
+                    299,
+                    261
+                ],
+                "score": 0.905
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    299,
+                    76,
+                    854,
+                    76,
+                    854,
+                    104,
+                    299,
+                    104
+                ],
+                "score": 0.825
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    840,
+                    2089,
+                    857,
+                    2089,
+                    857,
+                    2111,
+                    840,
+                    2111
+                ],
+                "score": 0.771
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    297,
+                    1054,
+                    1286,
+                    1054,
+                    1286,
+                    1086,
+                    297,
+                    1086
+                ],
+                "score": 0.684
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1054,
+                    1286,
+                    1054,
+                    1286,
+                    1086,
+                    297,
+                    1086
+                ],
+                "score": 0.224
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    300,
+                    76,
+                    852,
+                    76,
+                    852,
+                    104,
+                    300,
+                    104
+                ],
+                "score": 0.179
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1210.0,
+                    521.0,
+                    1395.0,
+                    521.0,
+                    1395.0,
+                    548.0,
+                    1210.0,
+                    548.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    910.0,
+                    534.0,
+                    1091.0,
+                    534.0,
+                    1091.0,
+                    558.0,
+                    910.0,
+                    558.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    916.0,
+                    560.0,
+                    1132.0,
+                    560.0,
+                    1132.0,
+                    585.0,
+                    916.0,
+                    585.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1238.0,
+                    554.0,
+                    1384.0,
+                    554.0,
+                    1384.0,
+                    578.0,
+                    1238.0,
+                    578.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1235.0,
+                    574.0,
+                    1384.0,
+                    574.0,
+                    1384.0,
+                    598.0,
+                    1235.0,
+                    598.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    926.0,
+                    589.0,
+                    1154.0,
+                    589.0,
+                    1154.0,
+                    615.0,
+                    926.0,
+                    615.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1261.0,
+                    591.0,
+                    1386.0,
+                    591.0,
+                    1386.0,
+                    621.0,
+                    1261.0,
+                    621.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    936.0,
+                    617.0,
+                    1190.0,
+                    617.0,
+                    1190.0,
+                    643.0,
+                    936.0,
+                    643.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1215.0,
+                    604.0,
+                    1223.0,
+                    604.0,
+                    1223.0,
+                    635.0,
+                    1215.0,
+                    635.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1002.0,
+                    666.0,
+                    1186.0,
+                    666.0,
+                    1186.0,
+                    691.0,
+                    1002.0,
+                    691.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    900.0,
+                    722.0,
+                    1407.0,
+                    722.0,
+                    1407.0,
+                    757.0,
+                    900.0,
+                    757.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    901.0,
+                    752.0,
+                    1097.0,
+                    752.0,
+                    1097.0,
+                    787.0,
+                    901.0,
+                    787.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    223.0,
+                    531.0,
+                    223.0,
+                    531.0,
+                    268.0,
+                    292.0,
+                    268.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    73.0,
+                    857.0,
+                    73.0,
+                    857.0,
+                    108.0,
+                    297.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    838.0,
+                    2086.0,
+                    862.0,
+                    2086.0,
+                    862.0,
+                    2118.0,
+                    838.0,
+                    2118.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1052.0,
+                    1290.0,
+                    1052.0,
+                    1290.0,
+                    1092.0,
+                    294.0,
+                    1092.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    73.0,
+                    856.0,
+                    73.0,
+                    856.0,
+                    108.0,
+                    297.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1635.0,
+                    1406.0,
+                    1635.0,
+                    1406.0,
+                    1672.0,
+                    295.0,
+                    1672.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1668.0,
+                    1405.0,
+                    1668.0,
+                    1405.0,
+                    1703.0,
+                    295.0,
+                    1703.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1698.0,
+                    1408.0,
+                    1698.0,
+                    1408.0,
+                    1735.0,
+                    292.0,
+                    1735.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1726.0,
+                    1406.0,
+                    1726.0,
+                    1406.0,
+                    1768.0,
+                    292.0,
+                    1768.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1760.0,
+                    1405.0,
+                    1760.0,
+                    1405.0,
+                    1796.0,
+                    293.0,
+                    1796.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1790.0,
+                    1405.0,
+                    1790.0,
+                    1405.0,
+                    1826.0,
+                    295.0,
+                    1826.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1817.0,
+                    1406.0,
+                    1817.0,
+                    1406.0,
+                    1858.0,
+                    292.0,
+                    1858.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1848.0,
+                    1406.0,
+                    1848.0,
+                    1406.0,
+                    1887.0,
+                    293.0,
+                    1887.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1878.0,
+                    1407.0,
+                    1878.0,
+                    1407.0,
+                    1919.0,
+                    293.0,
+                    1919.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1910.0,
+                    1406.0,
+                    1910.0,
+                    1406.0,
+                    1947.0,
+                    293.0,
+                    1947.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1943.0,
+                    1405.0,
+                    1943.0,
+                    1405.0,
+                    1979.0,
+                    295.0,
+                    1979.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1972.0,
+                    1403.0,
+                    1972.0,
+                    1403.0,
+                    2009.0,
+                    295.0,
+                    2009.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    2002.0,
+                    872.0,
+                    2002.0,
+                    872.0,
+                    2035.0,
+                    296.0,
+                    2035.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    516.0,
+                    886.0,
+                    516.0,
+                    886.0,
+                    549.0,
+                    295.0,
+                    549.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    545.0,
+                    882.0,
+                    545.0,
+                    882.0,
+                    582.0,
+                    294.0,
+                    582.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    579.0,
+                    882.0,
+                    579.0,
+                    882.0,
+                    607.0,
+                    295.0,
+                    607.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    608.0,
+                    882.0,
+                    608.0,
+                    882.0,
+                    641.0,
+                    294.0,
+                    641.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    637.0,
+                    884.0,
+                    637.0,
+                    884.0,
+                    671.0,
+                    294.0,
+                    671.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    670.0,
+                    885.0,
+                    670.0,
+                    885.0,
+                    700.0,
+                    295.0,
+                    700.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    698.0,
+                    883.0,
+                    698.0,
+                    883.0,
+                    732.0,
+                    295.0,
+                    732.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    730.0,
+                    883.0,
+                    730.0,
+                    883.0,
+                    763.0,
+                    294.0,
+                    763.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    758.0,
+                    643.0,
+                    758.0,
+                    643.0,
+                    791.0,
+                    296.0,
+                    791.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    807.0,
+                    1405.0,
+                    807.0,
+                    1405.0,
+                    842.0,
+                    295.0,
+                    842.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    836.0,
+                    1406.0,
+                    836.0,
+                    1406.0,
+                    875.0,
+                    292.0,
+                    875.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    869.0,
+                    1405.0,
+                    869.0,
+                    1405.0,
+                    903.0,
+                    296.0,
+                    903.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    900.0,
+                    1403.0,
+                    900.0,
+                    1403.0,
+                    934.0,
+                    295.0,
+                    934.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    928.0,
+                    1407.0,
+                    928.0,
+                    1407.0,
+                    965.0,
+                    293.0,
+                    965.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    958.0,
+                    1405.0,
+                    958.0,
+                    1405.0,
+                    995.0,
+                    293.0,
+                    995.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    991.0,
+                    1358.0,
+                    991.0,
+                    1358.0,
+                    1025.0,
+                    295.0,
+                    1025.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1237.0,
+                    1405.0,
+                    1237.0,
+                    1405.0,
+                    1274.0,
+                    292.0,
+                    1274.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1269.0,
+                    1404.0,
+                    1269.0,
+                    1404.0,
+                    1303.0,
+                    295.0,
+                    1303.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1299.0,
+                    1406.0,
+                    1299.0,
+                    1406.0,
+                    1335.0,
+                    292.0,
+                    1335.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1330.0,
+                    1406.0,
+                    1330.0,
+                    1406.0,
+                    1367.0,
+                    292.0,
+                    1367.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1359.0,
+                    1404.0,
+                    1359.0,
+                    1404.0,
+                    1396.0,
+                    293.0,
+                    1396.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1391.0,
+                    1361.0,
+                    1391.0,
+                    1361.0,
+                    1427.0,
+                    293.0,
+                    1427.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    288.0,
+                    1405.0,
+                    288.0,
+                    1405.0,
+                    322.0,
+                    295.0,
+                    322.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    320.0,
+                    1405.0,
+                    320.0,
+                    1405.0,
+                    350.0,
+                    296.0,
+                    350.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    349.0,
+                    1406.0,
+                    349.0,
+                    1406.0,
+                    382.0,
+                    294.0,
+                    382.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    379.0,
+                    1406.0,
+                    379.0,
+                    1406.0,
+                    413.0,
+                    295.0,
+                    413.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    407.0,
+                    1409.0,
+                    407.0,
+                    1409.0,
+                    445.0,
+                    294.0,
+                    445.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    440.0,
+                    1406.0,
+                    440.0,
+                    1406.0,
+                    475.0,
+                    294.0,
+                    475.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    471.0,
+                    1200.0,
+                    471.0,
+                    1200.0,
+                    505.0,
+                    295.0,
+                    505.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1436.0,
+                    1406.0,
+                    1436.0,
+                    1406.0,
+                    1473.0,
+                    294.0,
+                    1473.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1470.0,
+                    1407.0,
+                    1470.0,
+                    1407.0,
+                    1502.0,
+                    295.0,
+                    1502.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1500.0,
+                    1409.0,
+                    1500.0,
+                    1409.0,
+                    1535.0,
+                    294.0,
+                    1535.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1529.0,
+                    1405.0,
+                    1529.0,
+                    1405.0,
+                    1564.0,
+                    294.0,
+                    1564.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1558.0,
+                    1409.0,
+                    1558.0,
+                    1409.0,
+                    1596.0,
+                    294.0,
+                    1596.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1591.0,
+                    1409.0,
+                    1591.0,
+                    1409.0,
+                    1626.0,
+                    295.0,
+                    1626.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1098.0,
+                    1405.0,
+                    1098.0,
+                    1405.0,
+                    1134.0,
+                    294.0,
+                    1134.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1131.0,
+                    1405.0,
+                    1131.0,
+                    1405.0,
+                    1165.0,
+                    291.0,
+                    1165.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1162.0,
+                    1407.0,
+                    1162.0,
+                    1407.0,
+                    1198.0,
+                    292.0,
+                    1198.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1196.0,
+                    525.0,
+                    1196.0,
+                    525.0,
+                    1226.0,
+                    295.0,
+                    1226.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1052.0,
+                    1290.0,
+                    1052.0,
+                    1290.0,
+                    1092.0,
+                    294.0,
+                    1092.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 3,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    912,
+                    1404,
+                    912,
+                    1404,
+                    1128,
+                    298,
+                    1128
+                ],
+                "score": 0.981
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    682,
+                    1405,
+                    682,
+                    1405,
+                    897,
+                    298,
+                    897
+                ],
+                "score": 0.978
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1190,
+                    1405,
+                    1190,
+                    1405,
+                    1435,
+                    297,
+                    1435
+                ],
+                "score": 0.976
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    426,
+                    883,
+                    426,
+                    883,
+                    607,
+                    298,
+                    607
+                ],
+                "score": 0.973
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    271,
+                    1404,
+                    271,
+                    1404,
+                    424,
+                    297,
+                    424
+                ],
+                "score": 0.971
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1585,
+                    859,
+                    1585,
+                    859,
+                    1888,
+                    298,
+                    1888
+                ],
+                "score": 0.971
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1461,
+                    1404,
+                    1461,
+                    1404,
+                    1584,
+                    298,
+                    1584
+                ],
+                "score": 0.969
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    1889,
+                    1406,
+                    1889,
+                    1406,
+                    2010,
+                    299,
+                    2010
+                ],
+                "score": 0.967
+            },
+            {
+                "category_id": 5,
+                "poly": [
+                    924,
+                    1790,
+                    1358,
+                    1790,
+                    1358,
+                    1887,
+                    924,
+                    1887
+                ],
+                "score": 0.961,
+                "html": "<table><tr><td rowspan=1 colspan=1></td><td rowspan=1 colspan=1>at=D</td><td rowspan=1 colspan=1>at=C</td></tr><tr><td rowspan=1 colspan=1>St=at-1=D</td><td rowspan=1 colspan=1>-1/2</td><td rowspan=1 colspan=1>-1</td></tr><tr><td rowspan=1 colspan=1>St=at-1=C</td><td rowspan=1 colspan=1>1/2</td><td rowspan=1 colspan=1>0</td></tr></table>"
+            },
+            {
+                "category_id": 3,
+                "poly": [
+                    910,
+                    451,
+                    1397,
+                    451,
+                    1397,
+                    556,
+                    910,
+                    556
+                ],
+                "score": 0.948
+            },
+            {
+                "category_id": 4,
+                "poly": [
+                    942,
+                    594,
+                    1352,
+                    594,
+                    1352,
+                    627,
+                    942,
+                    627
+                ],
+                "score": 0.913
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    299,
+                    1148,
+                    620,
+                    1148,
+                    620,
+                    1179,
+                    299,
+                    1179
+                ],
+                "score": 0.907
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    299,
+                    637,
+                    559,
+                    637,
+                    559,
+                    671,
+                    299,
+                    671
+                ],
+                "score": 0.902
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    879,
+                    1605,
+                    1403,
+                    1605,
+                    1403,
+                    1788,
+                    879,
+                    1788
+                ],
+                "score": 0.899
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    298,
+                    75,
+                    854,
+                    75,
+                    854,
+                    105,
+                    298,
+                    105
+                ],
+                "score": 0.882
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    297,
+                    228,
+                    916,
+                    228,
+                    916,
+                    261,
+                    297,
+                    261
+                ],
+                "score": 0.876
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    841,
+                    2088,
+                    858,
+                    2088,
+                    858,
+                    2112,
+                    841,
+                    2112
+                ],
+                "score": 0.743
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    672,
+                    1220,
+                    788,
+                    1220,
+                    788,
+                    1253,
+                    672,
+                    1253
+                ],
+                "score": 0.93,
+                "latex": "s \\in \\{ 0 , 1 \\}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    898,
+                    1283,
+                    1063,
+                    1283,
+                    1063,
+                    1314,
+                    898,
+                    1314
+                ],
+                "score": 0.92,
+                "latex": "\\hat { y _ { 1 } } , \\hat { y _ { 2 } } = ( 0 , 0 )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1061,
+                    1219,
+                    1398,
+                    1219,
+                    1398,
+                    1253,
+                    1061,
+                    1253
+                ],
+                "score": 0.92,
+                "latex": "\\chi _ { 1 } , y _ { 2 } \\sim \\mathcal { N } ( 0 , s * \\sigma ^ { 2 } ) , \\mathcal { N } ( 0 , 1 )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1147,
+                    363,
+                    1291,
+                    363,
+                    1291,
+                    396,
+                    1147,
+                    396
+                ],
+                "score": 0.92,
+                "latex": "~ j ~ = ~ ( i + t )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    921,
+                    1220,
+                    1004,
+                    1220,
+                    1004,
+                    1252,
+                    921,
+                    1252
+                ],
+                "score": 0.92,
+                "latex": "\\boldsymbol { y } \\in \\mathbb { R } ^ { 2 }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    443,
+                    1676,
+                    514,
+                    1676,
+                    514,
+                    1707,
+                    443,
+                    1707
+                ],
+                "score": 0.9,
+                "latex": "\\gamma = 0"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1299,
+                    1314,
+                    1367,
+                    1314,
+                    1367,
+                    1344,
+                    1299,
+                    1344
+                ],
+                "score": 0.89,
+                "latex": "\\hat { y _ { 1 } } , \\hat { y _ { 2 } }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    894,
+                    1254,
+                    1001,
+                    1254,
+                    1001,
+                    1283,
+                    894,
+                    1283
+                ],
+                "score": 0.89,
+                "latex": "s _ { t + 1 } = 0"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    656,
+                    1254,
+                    723,
+                    1254,
+                    723,
+                    1284,
+                    656,
+                    1284
+                ],
+                "score": 0.89,
+                "latex": "\\hat { y } _ { 1 } , \\hat { y } _ { 2 }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    577,
+                    1313,
+                    606,
+                    1313,
+                    606,
+                    1344,
+                    577,
+                    1344
+                ],
+                "score": 0.88,
+                "latex": "\\hat { y _ { 1 } }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1313,
+                    1283,
+                    1402,
+                    1283,
+                    1402,
+                    1313,
+                    1313,
+                    1313
+                ],
+                "score": 0.87,
+                "latex": "\\hat { y } _ { 2 } > . 5"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1039,
+                    1253,
+                    1099,
+                    1253,
+                    1099,
+                    1282,
+                    1039,
+                    1282
+                ],
+                "score": 0.86,
+                "latex": "\\hat { y } _ { 2 } >"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1144,
+                    1790,
+                    1225,
+                    1790,
+                    1225,
+                    1819,
+                    1144,
+                    1819
+                ],
+                "score": 0.85,
+                "latex": "a _ { t } = \\mathrm { D }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1256,
+                    1790,
+                    1337,
+                    1790,
+                    1337,
+                    1819,
+                    1256,
+                    1819
+                ],
+                "score": 0.82,
+                "latex": "a _ { t } = \\mathrm { C }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    368,
+                    304,
+                    396,
+                    304,
+                    396,
+                    330,
+                    368,
+                    330
+                ],
+                "score": 0.81,
+                "latex": "N"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    677,
+                    365,
+                    694,
+                    365,
+                    694,
+                    395,
+                    677,
+                    395
+                ],
+                "score": 0.81,
+                "latex": "j"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    373,
+                    426,
+                    402,
+                    426,
+                    402,
+                    452,
+                    373,
+                    452
+                ],
+                "score": 0.79,
+                "latex": "N"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1151,
+                    1822,
+                    1217,
+                    1822,
+                    1217,
+                    1854,
+                    1151,
+                    1854
+                ],
+                "score": 0.78,
+                "latex": "- 1 / 2"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    394,
+                    366,
+                    407,
+                    366,
+                    407,
+                    391,
+                    394,
+                    391
+                ],
+                "score": 0.75,
+                "latex": "i"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    943,
+                    1825,
+                    1112,
+                    1825,
+                    1112,
+                    1852,
+                    943,
+                    1852
+                ],
+                "score": 0.72,
+                "latex": "s _ { t } = a _ { t - 1 } = \\mathbb { D }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1007,
+                    303,
+                    1036,
+                    303,
+                    1036,
+                    330,
+                    1007,
+                    330
+                ],
+                "score": 0.71,
+                "latex": "N"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1369,
+                    364,
+                    1397,
+                    364,
+                    1397,
+                    392,
+                    1369,
+                    392
+                ],
+                "score": 0.71,
+                "latex": "N"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    944,
+                    1855,
+                    1113,
+                    1855,
+                    1113,
+                    1886,
+                    944,
+                    1886
+                ],
+                "score": 0.64,
+                "latex": "s _ { t } = a _ { t - 1 } = \\mathsf { C }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1278,
+                    1825,
+                    1316,
+                    1825,
+                    1316,
+                    1849,
+                    1278,
+                    1849
+                ],
+                "score": 0.62,
+                "latex": "- 1"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1047,
+                    366,
+                    1062,
+                    366,
+                    1062,
+                    391,
+                    1047,
+                    391
+                ],
+                "score": 0.62,
+                "latex": "t"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    924.0,
+                    458.0,
+                    978.0,
+                    458.0,
+                    978.0,
+                    485.0,
+                    924.0,
+                    485.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    979.0,
+                    453.0,
+                    1149.0,
+                    453.0,
+                    1149.0,
+                    487.0,
+                    979.0,
+                    487.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1174.0,
+                    452.0,
+                    1396.0,
+                    452.0,
+                    1396.0,
+                    490.0,
+                    1174.0,
+                    490.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    925.0,
+                    490.0,
+                    981.0,
+                    490.0,
+                    981.0,
+                    518.0,
+                    925.0,
+                    518.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    991.0,
+                    489.0,
+                    1148.0,
+                    489.0,
+                    1148.0,
+                    520.0,
+                    991.0,
+                    520.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1174.0,
+                    490.0,
+                    1230.0,
+                    490.0,
+                    1230.0,
+                    518.0,
+                    1174.0,
+                    518.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1242.0,
+                    489.0,
+                    1394.0,
+                    489.0,
+                    1394.0,
+                    520.0,
+                    1242.0,
+                    520.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    905.0,
+                    523.0,
+                    980.0,
+                    523.0,
+                    980.0,
+                    551.0,
+                    905.0,
+                    551.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    989.0,
+                    516.0,
+                    1230.0,
+                    516.0,
+                    1230.0,
+                    555.0,
+                    989.0,
+                    555.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1238.0,
+                    519.0,
+                    1398.0,
+                    519.0,
+                    1398.0,
+                    554.0,
+                    1238.0,
+                    554.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    937.0,
+                    587.0,
+                    1355.0,
+                    587.0,
+                    1355.0,
+                    634.0,
+                    937.0,
+                    634.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1145.0,
+                    624.0,
+                    1145.0,
+                    624.0,
+                    1184.0,
+                    294.0,
+                    1184.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    632.0,
+                    562.0,
+                    632.0,
+                    562.0,
+                    677.0,
+                    291.0,
+                    677.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    73.0,
+                    857.0,
+                    73.0,
+                    857.0,
+                    108.0,
+                    297.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    228.0,
+                    917.0,
+                    228.0,
+                    917.0,
+                    264.0,
+                    295.0,
+                    264.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    839.0,
+                    2085.0,
+                    861.0,
+                    2085.0,
+                    861.0,
+                    2120.0,
+                    839.0,
+                    2120.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    913.0,
+                    1405.0,
+                    913.0,
+                    1405.0,
+                    947.0,
+                    294.0,
+                    947.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    942.0,
+                    1404.0,
+                    942.0,
+                    1404.0,
+                    976.0,
+                    295.0,
+                    976.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    973.0,
+                    1406.0,
+                    973.0,
+                    1406.0,
+                    1006.0,
+                    293.0,
+                    1006.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1004.0,
+                    1404.0,
+                    1004.0,
+                    1404.0,
+                    1039.0,
+                    294.0,
+                    1039.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1032.0,
+                    1404.0,
+                    1032.0,
+                    1404.0,
+                    1071.0,
+                    293.0,
+                    1071.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1062.0,
+                    1405.0,
+                    1062.0,
+                    1405.0,
+                    1102.0,
+                    293.0,
+                    1102.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1093.0,
+                    1336.0,
+                    1093.0,
+                    1336.0,
+                    1131.0,
+                    292.0,
+                    1131.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    682.0,
+                    1405.0,
+                    682.0,
+                    1405.0,
+                    717.0,
+                    295.0,
+                    717.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    714.0,
+                    1405.0,
+                    714.0,
+                    1405.0,
+                    749.0,
+                    295.0,
+                    749.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    743.0,
+                    1407.0,
+                    743.0,
+                    1407.0,
+                    780.0,
+                    293.0,
+                    780.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    776.0,
+                    1405.0,
+                    776.0,
+                    1405.0,
+                    810.0,
+                    295.0,
+                    810.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    803.0,
+                    1405.0,
+                    803.0,
+                    1405.0,
+                    840.0,
+                    293.0,
+                    840.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    834.0,
+                    1403.0,
+                    834.0,
+                    1403.0,
+                    872.0,
+                    293.0,
+                    872.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    868.0,
+                    981.0,
+                    868.0,
+                    981.0,
+                    899.0,
+                    297.0,
+                    899.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1187.0,
+                    1406.0,
+                    1187.0,
+                    1406.0,
+                    1226.0,
+                    294.0,
+                    1226.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1217.0,
+                    671.0,
+                    1217.0,
+                    671.0,
+                    1257.0,
+                    292.0,
+                    1257.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    789.0,
+                    1217.0,
+                    920.0,
+                    1217.0,
+                    920.0,
+                    1257.0,
+                    789.0,
+                    1257.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1005.0,
+                    1217.0,
+                    1060.0,
+                    1217.0,
+                    1060.0,
+                    1257.0,
+                    1005.0,
+                    1257.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1399.0,
+                    1217.0,
+                    1408.0,
+                    1217.0,
+                    1408.0,
+                    1257.0,
+                    1399.0,
+                    1257.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1250.0,
+                    655.0,
+                    1250.0,
+                    655.0,
+                    1287.0,
+                    294.0,
+                    1287.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    724.0,
+                    1250.0,
+                    893.0,
+                    1250.0,
+                    893.0,
+                    1287.0,
+                    724.0,
+                    1287.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1002.0,
+                    1250.0,
+                    1038.0,
+                    1250.0,
+                    1038.0,
+                    1287.0,
+                    1002.0,
+                    1287.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1100.0,
+                    1250.0,
+                    1406.0,
+                    1250.0,
+                    1406.0,
+                    1287.0,
+                    1100.0,
+                    1287.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1281.0,
+                    897.0,
+                    1281.0,
+                    897.0,
+                    1318.0,
+                    294.0,
+                    1318.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1064.0,
+                    1281.0,
+                    1312.0,
+                    1281.0,
+                    1312.0,
+                    1318.0,
+                    1064.0,
+                    1318.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1403.0,
+                    1281.0,
+                    1407.0,
+                    1281.0,
+                    1407.0,
+                    1318.0,
+                    1403.0,
+                    1318.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1308.0,
+                    576.0,
+                    1308.0,
+                    576.0,
+                    1350.0,
+                    292.0,
+                    1350.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    607.0,
+                    1308.0,
+                    1298.0,
+                    1308.0,
+                    1298.0,
+                    1350.0,
+                    607.0,
+                    1350.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1368.0,
+                    1308.0,
+                    1407.0,
+                    1308.0,
+                    1407.0,
+                    1350.0,
+                    1368.0,
+                    1350.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1341.0,
+                    1406.0,
+                    1341.0,
+                    1406.0,
+                    1377.0,
+                    292.0,
+                    1377.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1371.0,
+                    1405.0,
+                    1371.0,
+                    1405.0,
+                    1407.0,
+                    295.0,
+                    1407.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1404.0,
+                    1044.0,
+                    1404.0,
+                    1044.0,
+                    1437.0,
+                    295.0,
+                    1437.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    422.0,
+                    372.0,
+                    422.0,
+                    372.0,
+                    457.0,
+                    296.0,
+                    457.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    403.0,
+                    422.0,
+                    885.0,
+                    422.0,
+                    885.0,
+                    457.0,
+                    403.0,
+                    457.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    454.0,
+                    882.0,
+                    454.0,
+                    882.0,
+                    486.0,
+                    296.0,
+                    486.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    484.0,
+                    883.0,
+                    484.0,
+                    883.0,
+                    517.0,
+                    296.0,
+                    517.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    516.0,
+                    884.0,
+                    516.0,
+                    884.0,
+                    547.0,
+                    296.0,
+                    547.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    546.0,
+                    882.0,
+                    546.0,
+                    882.0,
+                    578.0,
+                    295.0,
+                    578.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    576.0,
+                    886.0,
+                    576.0,
+                    886.0,
+                    609.0,
+                    296.0,
+                    609.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    271.0,
+                    1404.0,
+                    271.0,
+                    1404.0,
+                    308.0,
+                    295.0,
+                    308.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    300.0,
+                    367.0,
+                    300.0,
+                    367.0,
+                    338.0,
+                    294.0,
+                    338.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    397.0,
+                    300.0,
+                    1006.0,
+                    300.0,
+                    1006.0,
+                    338.0,
+                    397.0,
+                    338.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1037.0,
+                    300.0,
+                    1406.0,
+                    300.0,
+                    1406.0,
+                    338.0,
+                    1037.0,
+                    338.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    332.0,
+                    1405.0,
+                    332.0,
+                    1405.0,
+                    367.0,
+                    294.0,
+                    367.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    362.0,
+                    393.0,
+                    362.0,
+                    393.0,
+                    399.0,
+                    295.0,
+                    399.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    408.0,
+                    362.0,
+                    676.0,
+                    362.0,
+                    676.0,
+                    399.0,
+                    408.0,
+                    399.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    695.0,
+                    362.0,
+                    1046.0,
+                    362.0,
+                    1046.0,
+                    399.0,
+                    695.0,
+                    399.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1063.0,
+                    362.0,
+                    1146.0,
+                    362.0,
+                    1146.0,
+                    399.0,
+                    1063.0,
+                    399.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1292.0,
+                    362.0,
+                    1368.0,
+                    362.0,
+                    1368.0,
+                    399.0,
+                    1292.0,
+                    399.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1398.0,
+                    362.0,
+                    1406.0,
+                    362.0,
+                    1406.0,
+                    399.0,
+                    1398.0,
+                    399.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    394.0,
+                    1406.0,
+                    394.0,
+                    1406.0,
+                    427.0,
+                    295.0,
+                    427.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1582.0,
+                    857.0,
+                    1582.0,
+                    857.0,
+                    1615.0,
+                    296.0,
+                    1615.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1613.0,
+                    860.0,
+                    1613.0,
+                    860.0,
+                    1647.0,
+                    295.0,
+                    1647.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1644.0,
+                    859.0,
+                    1644.0,
+                    859.0,
+                    1676.0,
+                    294.0,
+                    1676.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1674.0,
+                    442.0,
+                    1674.0,
+                    442.0,
+                    1706.0,
+                    295.0,
+                    1706.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    515.0,
+                    1674.0,
+                    862.0,
+                    1674.0,
+                    862.0,
+                    1706.0,
+                    515.0,
+                    1706.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1705.0,
+                    861.0,
+                    1705.0,
+                    861.0,
+                    1737.0,
+                    293.0,
+                    1737.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1735.0,
+                    862.0,
+                    1735.0,
+                    862.0,
+                    1769.0,
+                    294.0,
+                    1769.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1766.0,
+                    861.0,
+                    1766.0,
+                    861.0,
+                    1798.0,
+                    295.0,
+                    1798.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1795.0,
+                    862.0,
+                    1795.0,
+                    862.0,
+                    1830.0,
+                    294.0,
+                    1830.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1827.0,
+                    859.0,
+                    1827.0,
+                    859.0,
+                    1860.0,
+                    295.0,
+                    1860.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1855.0,
+                    860.0,
+                    1855.0,
+                    860.0,
+                    1891.0,
+                    294.0,
+                    1891.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1460.0,
+                    1405.0,
+                    1460.0,
+                    1405.0,
+                    1496.0,
+                    294.0,
+                    1496.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1490.0,
+                    1406.0,
+                    1490.0,
+                    1406.0,
+                    1527.0,
+                    292.0,
+                    1527.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1520.0,
+                    1405.0,
+                    1520.0,
+                    1405.0,
+                    1559.0,
+                    291.0,
+                    1559.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1554.0,
+                    1406.0,
+                    1554.0,
+                    1406.0,
+                    1586.0,
+                    294.0,
+                    1586.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1886.0,
+                    1406.0,
+                    1886.0,
+                    1406.0,
+                    1923.0,
+                    293.0,
+                    1923.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1916.0,
+                    1406.0,
+                    1916.0,
+                    1406.0,
+                    1952.0,
+                    294.0,
+                    1952.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1947.0,
+                    1406.0,
+                    1947.0,
+                    1406.0,
+                    1984.0,
+                    293.0,
+                    1984.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1977.0,
+                    929.0,
+                    1977.0,
+                    929.0,
+                    2014.0,
+                    294.0,
+                    2014.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    878.0,
+                    1604.0,
+                    1404.0,
+                    1604.0,
+                    1404.0,
+                    1637.0,
+                    878.0,
+                    1637.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    879.0,
+                    1637.0,
+                    1404.0,
+                    1637.0,
+                    1404.0,
+                    1671.0,
+                    879.0,
+                    1671.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    877.0,
+                    1667.0,
+                    1403.0,
+                    1667.0,
+                    1403.0,
+                    1699.0,
+                    877.0,
+                    1699.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    877.0,
+                    1695.0,
+                    1403.0,
+                    1695.0,
+                    1403.0,
+                    1731.0,
+                    877.0,
+                    1731.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    881.0,
+                    1728.0,
+                    1404.0,
+                    1728.0,
+                    1404.0,
+                    1758.0,
+                    881.0,
+                    1758.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    878.0,
+                    1756.0,
+                    1364.0,
+                    1756.0,
+                    1364.0,
+                    1792.0,
+                    878.0,
+                    1792.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 4,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    570,
+                    772,
+                    570,
+                    772,
+                    1179,
+                    298,
+                    1179
+                ],
+                "score": 0.977
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1193,
+                    771,
+                    1193,
+                    771,
+                    1834,
+                    298,
+                    1834
+                ],
+                "score": 0.976
+            },
+            {
+                "category_id": 4,
+                "poly": [
+                    792,
+                    1506,
+                    1403,
+                    1506,
+                    1403,
+                    1873,
+                    792,
+                    1873
+                ],
+                "score": 0.973
+            },
+            {
+                "category_id": 3,
+                "poly": [
+                    798,
+                    576,
+                    1395,
+                    576,
+                    1395,
+                    1477,
+                    798,
+                    1477
+                ],
+                "score": 0.967
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    461,
+                    1404,
+                    461,
+                    1404,
+                    554,
+                    299,
+                    554
+                ],
+                "score": 0.965
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1912,
+                    1405,
+                    1912,
+                    1405,
+                    2034,
+                    298,
+                    2034
+                ],
+                "score": 0.964
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    242,
+                    1405,
+                    242,
+                    1405,
+                    446,
+                    297,
+                    446
+                ],
+                "score": 0.932
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    298,
+                    75,
+                    853,
+                    75,
+                    853,
+                    104,
+                    298,
+                    104
+                ],
+                "score": 0.796
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    840,
+                    2089,
+                    859,
+                    2089,
+                    859,
+                    2112,
+                    840,
+                    2112
+                ],
+                "score": 0.771
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    296,
+                    1852,
+                    772,
+                    1852,
+                    772,
+                    1912,
+                    296,
+                    1912
+                ],
+                "score": 0.75
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    296,
+                    230,
+                    1157,
+                    230,
+                    1157,
+                    261,
+                    296,
+                    261
+                ],
+                "score": 0.284
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    299,
+                    76,
+                    852,
+                    76,
+                    852,
+                    104,
+                    299,
+                    104
+                ],
+                "score": 0.199
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    296,
+                    1852,
+                    772,
+                    1852,
+                    772,
+                    1912,
+                    296,
+                    1912
+                ],
+                "score": 0.108
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    884,
+                    1782,
+                    955,
+                    1782,
+                    955,
+                    1812,
+                    884,
+                    1812
+                ],
+                "score": 0.91,
+                "latex": "\\gamma = 0"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    426,
+                    1226,
+                    502,
+                    1226,
+                    502,
+                    1254,
+                    426,
+                    1254
+                ],
+                "score": 0.9,
+                "latex": "T = 1"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1322,
+                    1943,
+                    1398,
+                    1943,
+                    1398,
+                    1971,
+                    1322,
+                    1971
+                ],
+                "score": 0.89,
+                "latex": "T > 1"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    453,
+                    1148,
+                    528,
+                    1148,
+                    528,
+                    1176,
+                    453,
+                    1176
+                ],
+                "score": 0.88,
+                "latex": "T = 1"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1073,
+                    1508,
+                    1102,
+                    1508,
+                    1102,
+                    1538,
+                    1073,
+                    1538
+                ],
+                "score": 0.88,
+                "latex": "\\hat { y _ { 2 } }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    547,
+                    1118,
+                    601,
+                    1118,
+                    601,
+                    1147,
+                    547,
+                    1147
+                ],
+                "score": 0.87,
+                "latex": "20 \\%"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    386,
+                    1945,
+                    410,
+                    1945,
+                    410,
+                    1970,
+                    386,
+                    1970
+                ],
+                "score": 0.8,
+                "latex": "T"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    398,
+                    1148,
+                    443,
+                    1148,
+                    443,
+                    1177,
+                    398,
+                    1177
+                ],
+                "score": 0.77,
+                "latex": "_ { = 1 0 }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1183,
+                    1452,
+                    1203,
+                    1452,
+                    1203,
+                    1475,
+                    1183,
+                    1475
+                ],
+                "score": 0.48,
+                "latex": "\\ r ="
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1349,
+                    577,
+                    1363,
+                    577,
+                    1363,
+                    591,
+                    1349,
+                    591
+                ],
+                "score": 0.36,
+                "latex": "{ } = { }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1018,
+                    577,
+                    1044,
+                    577,
+                    1044,
+                    592,
+                    1018,
+                    592
+                ],
+                "score": 0.35,
+                "latex": "= 1"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1235,
+                    1179,
+                    1254,
+                    1179,
+                    1254,
+                    1201,
+                    1235,
+                    1201
+                ],
+                "score": 0.33,
+                "latex": ""
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1181,
+                    824,
+                    1199,
+                    824,
+                    1199,
+                    847,
+                    1181,
+                    847
+                ],
+                "score": 0.27,
+                "latex": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    789.0,
+                    1502.0,
+                    1072.0,
+                    1502.0,
+                    1072.0,
+                    1544.0,
+                    789.0,
+                    1544.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1103.0,
+                    1502.0,
+                    1405.0,
+                    1502.0,
+                    1405.0,
+                    1544.0,
+                    1103.0,
+                    1544.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    790.0,
+                    1537.0,
+                    1405.0,
+                    1537.0,
+                    1405.0,
+                    1572.0,
+                    790.0,
+                    1572.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    789.0,
+                    1568.0,
+                    1405.0,
+                    1568.0,
+                    1405.0,
+                    1602.0,
+                    789.0,
+                    1602.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    789.0,
+                    1599.0,
+                    1405.0,
+                    1599.0,
+                    1405.0,
+                    1633.0,
+                    789.0,
+                    1633.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    790.0,
+                    1630.0,
+                    1404.0,
+                    1630.0,
+                    1404.0,
+                    1659.0,
+                    790.0,
+                    1659.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    788.0,
+                    1657.0,
+                    1407.0,
+                    1657.0,
+                    1407.0,
+                    1694.0,
+                    788.0,
+                    1694.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    789.0,
+                    1689.0,
+                    1405.0,
+                    1689.0,
+                    1405.0,
+                    1723.0,
+                    789.0,
+                    1723.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    789.0,
+                    1721.0,
+                    1405.0,
+                    1721.0,
+                    1405.0,
+                    1754.0,
+                    789.0,
+                    1754.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    787.0,
+                    1749.0,
+                    1405.0,
+                    1749.0,
+                    1405.0,
+                    1785.0,
+                    787.0,
+                    1785.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    788.0,
+                    1780.0,
+                    883.0,
+                    1780.0,
+                    883.0,
+                    1817.0,
+                    788.0,
+                    1817.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    956.0,
+                    1780.0,
+                    1407.0,
+                    1780.0,
+                    1407.0,
+                    1817.0,
+                    956.0,
+                    1817.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    788.0,
+                    1810.0,
+                    1410.0,
+                    1810.0,
+                    1410.0,
+                    1848.0,
+                    788.0,
+                    1848.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    787.0,
+                    1841.0,
+                    1382.0,
+                    1841.0,
+                    1382.0,
+                    1876.0,
+                    787.0,
+                    1876.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    831.0,
+                    575.0,
+                    886.0,
+                    575.0,
+                    886.0,
+                    593.0,
+                    831.0,
+                    593.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    937.0,
+                    575.0,
+                    1017.0,
+                    575.0,
+                    1017.0,
+                    595.0,
+                    937.0,
+                    595.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1045.0,
+                    575.0,
+                    1048.0,
+                    575.0,
+                    1048.0,
+                    595.0,
+                    1045.0,
+                    595.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1098.0,
+                    575.0,
+                    1217.0,
+                    575.0,
+                    1217.0,
+                    595.0,
+                    1098.0,
+                    595.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1267.0,
+                    575.0,
+                    1348.0,
+                    575.0,
+                    1348.0,
+                    595.0,
+                    1267.0,
+                    595.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1364.0,
+                    575.0,
+                    1394.0,
+                    575.0,
+                    1394.0,
+                    595.0,
+                    1364.0,
+                    595.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    874.0,
+                    600.0,
+                    946.0,
+                    600.0,
+                    946.0,
+                    620.0,
+                    874.0,
+                    620.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1068.0,
+                    601.0,
+                    1145.0,
+                    601.0,
+                    1145.0,
+                    618.0,
+                    1068.0,
+                    618.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1261.0,
+                    600.0,
+                    1348.0,
+                    600.0,
+                    1348.0,
+                    620.0,
+                    1261.0,
+                    620.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    804.0,
+                    610.0,
+                    833.0,
+                    610.0,
+                    833.0,
+                    645.0,
+                    804.0,
+                    645.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1379.0,
+                    614.0,
+                    1395.0,
+                    614.0,
+                    1395.0,
+                    693.0,
+                    1379.0,
+                    693.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    787.0,
+                    635.0,
+                    839.0,
+                    635.0,
+                    839.0,
+                    674.0,
+                    787.0,
+                    674.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    804.0,
+                    664.0,
+                    834.0,
+                    664.0,
+                    834.0,
+                    681.0,
+                    804.0,
+                    681.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    804.0,
+                    682.0,
+                    834.0,
+                    682.0,
+                    834.0,
+                    700.0,
+                    804.0,
+                    700.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    806.0,
+                    698.0,
+                    832.0,
+                    698.0,
+                    832.0,
+                    714.0,
+                    806.0,
+                    714.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    804.0,
+                    715.0,
+                    833.0,
+                    715.0,
+                    833.0,
+                    732.0,
+                    804.0,
+                    732.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1380.0,
+                    706.0,
+                    1396.0,
+                    706.0,
+                    1396.0,
+                    776.0,
+                    1380.0,
+                    776.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    787.0,
+                    722.0,
+                    839.0,
+                    722.0,
+                    839.0,
+                    762.0,
+                    787.0,
+                    762.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    804.0,
+                    750.0,
+                    835.0,
+                    750.0,
+                    835.0,
+                    769.0,
+                    804.0,
+                    769.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    804.0,
+                    769.0,
+                    834.0,
+                    769.0,
+                    834.0,
+                    787.0,
+                    804.0,
+                    787.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    835.0,
+                    779.0,
+                    848.0,
+                    779.0,
+                    848.0,
+                    793.0,
+                    835.0,
+                    793.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    856.0,
+                    778.0,
+                    882.0,
+                    778.0,
+                    882.0,
+                    794.0,
+                    856.0,
+                    794.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    883.0,
+                    778.0,
+                    994.0,
+                    778.0,
+                    994.0,
+                    795.0,
+                    883.0,
+                    795.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1032.0,
+                    780.0,
+                    1045.0,
+                    780.0,
+                    1045.0,
+                    793.0,
+                    1032.0,
+                    793.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1054.0,
+                    778.0,
+                    1079.0,
+                    778.0,
+                    1079.0,
+                    794.0,
+                    1054.0,
+                    794.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1081.0,
+                    778.0,
+                    1190.0,
+                    778.0,
+                    1190.0,
+                    795.0,
+                    1081.0,
+                    795.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1229.0,
+                    780.0,
+                    1243.0,
+                    780.0,
+                    1243.0,
+                    793.0,
+                    1229.0,
+                    793.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1251.0,
+                    779.0,
+                    1277.0,
+                    779.0,
+                    1277.0,
+                    794.0,
+                    1251.0,
+                    794.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1278.0,
+                    778.0,
+                    1332.0,
+                    778.0,
+                    1332.0,
+                    795.0,
+                    1278.0,
+                    795.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1334.0,
+                    779.0,
+                    1387.0,
+                    779.0,
+                    1387.0,
+                    794.0,
+                    1334.0,
+                    794.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    884.0,
+                    788.0,
+                    934.0,
+                    788.0,
+                    934.0,
+                    808.0,
+                    884.0,
+                    808.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    940.0,
+                    821.0,
+                    1180.0,
+                    821.0,
+                    1180.0,
+                    852.0,
+                    940.0,
+                    852.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1200.0,
+                    821.0,
+                    1258.0,
+                    821.0,
+                    1258.0,
+                    852.0,
+                    1200.0,
+                    852.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    887.0,
+                    859.0,
+                    973.0,
+                    859.0,
+                    973.0,
+                    877.0,
+                    887.0,
+                    877.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1065.0,
+                    859.0,
+                    1159.0,
+                    859.0,
+                    1159.0,
+                    877.0,
+                    1065.0,
+                    877.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1244.0,
+                    859.0,
+                    1345.0,
+                    859.0,
+                    1345.0,
+                    878.0,
+                    1244.0,
+                    878.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    823.0,
+                    870.0,
+                    848.0,
+                    870.0,
+                    848.0,
+                    889.0,
+                    823.0,
+                    889.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    803.0,
+                    893.0,
+                    850.0,
+                    893.0,
+                    850.0,
+                    969.0,
+                    803.0,
+                    969.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1374.0,
+                    889.0,
+                    1394.0,
+                    889.0,
+                    1394.0,
+                    987.0,
+                    1374.0,
+                    987.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    822.0,
+                    938.0,
+                    848.0,
+                    938.0,
+                    848.0,
+                    959.0,
+                    822.0,
+                    959.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    822.0,
+                    963.0,
+                    848.0,
+                    963.0,
+                    848.0,
+                    982.0,
+                    822.0,
+                    982.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    822.0,
+                    986.0,
+                    849.0,
+                    986.0,
+                    849.0,
+                    1005.0,
+                    822.0,
+                    1005.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    822.0,
+                    1008.0,
+                    848.0,
+                    1008.0,
+                    848.0,
+                    1026.0,
+                    822.0,
+                    1026.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    802.0,
+                    1031.0,
+                    849.0,
+                    1031.0,
+                    849.0,
+                    1114.0,
+                    802.0,
+                    1114.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1374.0,
+                    1035.0,
+                    1394.0,
+                    1035.0,
+                    1394.0,
+                    1116.0,
+                    1374.0,
+                    1116.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    822.0,
+                    1077.0,
+                    848.0,
+                    1077.0,
+                    848.0,
+                    1096.0,
+                    822.0,
+                    1096.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    822.0,
+                    1099.0,
+                    848.0,
+                    1099.0,
+                    848.0,
+                    1118.0,
+                    822.0,
+                    1118.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    867.0,
+                    1109.0,
+                    875.0,
+                    1109.0,
+                    875.0,
+                    1117.0,
+                    867.0,
+                    1117.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    823.0,
+                    1123.0,
+                    848.0,
+                    1123.0,
+                    848.0,
+                    1142.0,
+                    823.0,
+                    1142.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    851.0,
+                    1136.0,
+                    863.0,
+                    1136.0,
+                    863.0,
+                    1151.0,
+                    851.0,
+                    1151.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    879.0,
+                    1134.0,
+                    946.0,
+                    1134.0,
+                    946.0,
+                    1154.0,
+                    879.0,
+                    1154.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    953.0,
+                    1135.0,
+                    983.0,
+                    1135.0,
+                    983.0,
+                    1153.0,
+                    953.0,
+                    1153.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    986.0,
+                    1135.0,
+                    1022.0,
+                    1135.0,
+                    1022.0,
+                    1153.0,
+                    986.0,
+                    1153.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1032.0,
+                    1136.0,
+                    1046.0,
+                    1136.0,
+                    1046.0,
+                    1151.0,
+                    1032.0,
+                    1151.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1062.0,
+                    1134.0,
+                    1091.0,
+                    1134.0,
+                    1091.0,
+                    1153.0,
+                    1062.0,
+                    1153.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1098.0,
+                    1135.0,
+                    1128.0,
+                    1135.0,
+                    1128.0,
+                    1153.0,
+                    1098.0,
+                    1153.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1136.0,
+                    1135.0,
+                    1165.0,
+                    1135.0,
+                    1165.0,
+                    1153.0,
+                    1136.0,
+                    1153.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1168.0,
+                    1135.0,
+                    1203.0,
+                    1135.0,
+                    1203.0,
+                    1153.0,
+                    1168.0,
+                    1153.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1215.0,
+                    1136.0,
+                    1227.0,
+                    1136.0,
+                    1227.0,
+                    1151.0,
+                    1215.0,
+                    1151.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1244.0,
+                    1134.0,
+                    1273.0,
+                    1134.0,
+                    1273.0,
+                    1153.0,
+                    1244.0,
+                    1153.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1279.0,
+                    1135.0,
+                    1308.0,
+                    1135.0,
+                    1308.0,
+                    1153.0,
+                    1279.0,
+                    1153.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1317.0,
+                    1135.0,
+                    1346.0,
+                    1135.0,
+                    1346.0,
+                    1153.0,
+                    1317.0,
+                    1153.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1351.0,
+                    1135.0,
+                    1385.0,
+                    1135.0,
+                    1385.0,
+                    1153.0,
+                    1351.0,
+                    1153.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    900.0,
+                    1147.0,
+                    960.0,
+                    1147.0,
+                    960.0,
+                    1170.0,
+                    900.0,
+                    1170.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1084.0,
+                    1147.0,
+                    1141.0,
+                    1147.0,
+                    1141.0,
+                    1170.0,
+                    1084.0,
+                    1170.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1265.0,
+                    1146.0,
+                    1324.0,
+                    1146.0,
+                    1324.0,
+                    1170.0,
+                    1265.0,
+                    1170.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    885.0,
+                    1175.0,
+                    1234.0,
+                    1175.0,
+                    1234.0,
+                    1207.0,
+                    885.0,
+                    1207.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1255.0,
+                    1175.0,
+                    1312.0,
+                    1175.0,
+                    1312.0,
+                    1207.0,
+                    1255.0,
+                    1207.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    881.0,
+                    1211.0,
+                    959.0,
+                    1211.0,
+                    959.0,
+                    1231.0,
+                    881.0,
+                    1231.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1065.0,
+                    1212.0,
+                    1150.0,
+                    1212.0,
+                    1150.0,
+                    1232.0,
+                    1065.0,
+                    1232.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1251.0,
+                    1212.0,
+                    1341.0,
+                    1212.0,
+                    1341.0,
+                    1232.0,
+                    1251.0,
+                    1232.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    793.0,
+                    1223.0,
+                    835.0,
+                    1223.0,
+                    835.0,
+                    1310.0,
+                    793.0,
+                    1310.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1380.0,
+                    1228.0,
+                    1397.0,
+                    1228.0,
+                    1397.0,
+                    1315.0,
+                    1380.0,
+                    1315.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    805.0,
+                    1262.0,
+                    835.0,
+                    1262.0,
+                    835.0,
+                    1280.0,
+                    805.0,
+                    1280.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    805.0,
+                    1281.0,
+                    834.0,
+                    1281.0,
+                    834.0,
+                    1300.0,
+                    805.0,
+                    1300.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    806.0,
+                    1303.0,
+                    836.0,
+                    1303.0,
+                    836.0,
+                    1321.0,
+                    806.0,
+                    1321.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    807.0,
+                    1320.0,
+                    835.0,
+                    1320.0,
+                    835.0,
+                    1334.0,
+                    807.0,
+                    1334.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    794.0,
+                    1338.0,
+                    835.0,
+                    1338.0,
+                    835.0,
+                    1403.0,
+                    794.0,
+                    1403.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1380.0,
+                    1330.0,
+                    1396.0,
+                    1330.0,
+                    1396.0,
+                    1403.0,
+                    1380.0,
+                    1403.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    805.0,
+                    1378.0,
+                    835.0,
+                    1378.0,
+                    835.0,
+                    1396.0,
+                    805.0,
+                    1396.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    807.0,
+                    1400.0,
+                    834.0,
+                    1400.0,
+                    834.0,
+                    1414.0,
+                    807.0,
+                    1414.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    838.0,
+                    1409.0,
+                    852.0,
+                    1409.0,
+                    852.0,
+                    1423.0,
+                    838.0,
+                    1423.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    862.0,
+                    1409.0,
+                    888.0,
+                    1409.0,
+                    888.0,
+                    1424.0,
+                    862.0,
+                    1424.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    892.0,
+                    1409.0,
+                    948.0,
+                    1409.0,
+                    948.0,
+                    1424.0,
+                    892.0,
+                    1424.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    951.0,
+                    1408.0,
+                    1010.0,
+                    1408.0,
+                    1010.0,
+                    1425.0,
+                    951.0,
+                    1425.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1027.0,
+                    1409.0,
+                    1040.0,
+                    1409.0,
+                    1040.0,
+                    1423.0,
+                    1027.0,
+                    1423.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1048.0,
+                    1408.0,
+                    1078.0,
+                    1408.0,
+                    1078.0,
+                    1426.0,
+                    1048.0,
+                    1426.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1080.0,
+                    1408.0,
+                    1137.0,
+                    1408.0,
+                    1137.0,
+                    1425.0,
+                    1080.0,
+                    1425.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1140.0,
+                    1409.0,
+                    1199.0,
+                    1409.0,
+                    1199.0,
+                    1424.0,
+                    1140.0,
+                    1424.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1216.0,
+                    1409.0,
+                    1228.0,
+                    1409.0,
+                    1228.0,
+                    1424.0,
+                    1216.0,
+                    1424.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1238.0,
+                    1409.0,
+                    1263.0,
+                    1409.0,
+                    1263.0,
+                    1425.0,
+                    1238.0,
+                    1425.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1269.0,
+                    1408.0,
+                    1325.0,
+                    1408.0,
+                    1325.0,
+                    1425.0,
+                    1269.0,
+                    1425.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1328.0,
+                    1408.0,
+                    1388.0,
+                    1408.0,
+                    1388.0,
+                    1425.0,
+                    1328.0,
+                    1425.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    893.0,
+                    1419.0,
+                    946.0,
+                    1419.0,
+                    946.0,
+                    1439.0,
+                    893.0,
+                    1439.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1082.0,
+                    1418.0,
+                    1134.0,
+                    1418.0,
+                    1134.0,
+                    1439.0,
+                    1082.0,
+                    1439.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1270.0,
+                    1419.0,
+                    1323.0,
+                    1419.0,
+                    1323.0,
+                    1440.0,
+                    1270.0,
+                    1440.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    835.0,
+                    1449.0,
+                    1182.0,
+                    1449.0,
+                    1182.0,
+                    1479.0,
+                    835.0,
+                    1479.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1204.0,
+                    1449.0,
+                    1360.0,
+                    1449.0,
+                    1360.0,
+                    1479.0,
+                    1204.0,
+                    1479.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1041.75,
+                    634.5,
+                    1046.75,
+                    634.5,
+                    1046.75,
+                    640.5,
+                    1041.75,
+                    640.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1239.0,
+                    634.5,
+                    1244.0,
+                    634.5,
+                    1244.0,
+                    640.5,
+                    1239.0,
+                    640.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1082.75,
+                    789.0,
+                    1129.75,
+                    789.0,
+                    1129.75,
+                    807.0,
+                    1082.75,
+                    807.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1280.0,
+                    789.0,
+                    1327.0,
+                    789.0,
+                    1327.0,
+                    807.0,
+                    1280.0,
+                    807.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    72.0,
+                    857.0,
+                    72.0,
+                    857.0,
+                    108.0,
+                    297.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    840.0,
+                    2087.0,
+                    862.0,
+                    2087.0,
+                    862.0,
+                    2118.0,
+                    840.0,
+                    2118.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1849.0,
+                    775.0,
+                    1849.0,
+                    775.0,
+                    1885.0,
+                    294.0,
+                    1885.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1880.0,
+                    772.0,
+                    1880.0,
+                    772.0,
+                    1916.0,
+                    294.0,
+                    1916.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    230.0,
+                    1160.0,
+                    230.0,
+                    1160.0,
+                    265.0,
+                    295.0,
+                    265.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    73.0,
+                    856.0,
+                    73.0,
+                    856.0,
+                    108.0,
+                    297.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    567.0,
+                    775.0,
+                    567.0,
+                    775.0,
+                    605.0,
+                    295.0,
+                    605.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    600.0,
+                    774.0,
+                    600.0,
+                    774.0,
+                    632.0,
+                    295.0,
+                    632.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    629.0,
+                    774.0,
+                    629.0,
+                    774.0,
+                    664.0,
+                    295.0,
+                    664.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    661.0,
+                    774.0,
+                    661.0,
+                    774.0,
+                    692.0,
+                    296.0,
+                    692.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    690.0,
+                    776.0,
+                    690.0,
+                    776.0,
+                    724.0,
+                    296.0,
+                    724.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    722.0,
+                    773.0,
+                    722.0,
+                    773.0,
+                    754.0,
+                    296.0,
+                    754.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    754.0,
+                    773.0,
+                    754.0,
+                    773.0,
+                    784.0,
+                    295.0,
+                    784.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    785.0,
+                    771.0,
+                    785.0,
+                    771.0,
+                    813.0,
+                    297.0,
+                    813.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    811.0,
+                    772.0,
+                    811.0,
+                    772.0,
+                    845.0,
+                    295.0,
+                    845.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    844.0,
+                    775.0,
+                    844.0,
+                    775.0,
+                    876.0,
+                    296.0,
+                    876.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    871.0,
+                    773.0,
+                    871.0,
+                    773.0,
+                    908.0,
+                    294.0,
+                    908.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    903.0,
+                    771.0,
+                    903.0,
+                    771.0,
+                    939.0,
+                    295.0,
+                    939.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    936.0,
+                    773.0,
+                    936.0,
+                    773.0,
+                    967.0,
+                    295.0,
+                    967.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    966.0,
+                    774.0,
+                    966.0,
+                    774.0,
+                    997.0,
+                    295.0,
+                    997.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    995.0,
+                    773.0,
+                    995.0,
+                    773.0,
+                    1027.0,
+                    295.0,
+                    1027.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1026.0,
+                    771.0,
+                    1026.0,
+                    771.0,
+                    1060.0,
+                    295.0,
+                    1060.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1057.0,
+                    775.0,
+                    1057.0,
+                    775.0,
+                    1088.0,
+                    294.0,
+                    1088.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1086.0,
+                    773.0,
+                    1086.0,
+                    773.0,
+                    1119.0,
+                    294.0,
+                    1119.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1114.0,
+                    546.0,
+                    1114.0,
+                    546.0,
+                    1150.0,
+                    294.0,
+                    1150.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    602.0,
+                    1114.0,
+                    774.0,
+                    1114.0,
+                    774.0,
+                    1150.0,
+                    602.0,
+                    1150.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1145.0,
+                    397.0,
+                    1145.0,
+                    397.0,
+                    1182.0,
+                    294.0,
+                    1182.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    444.0,
+                    1145.0,
+                    452.0,
+                    1145.0,
+                    452.0,
+                    1182.0,
+                    444.0,
+                    1182.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    529.0,
+                    1145.0,
+                    775.0,
+                    1145.0,
+                    775.0,
+                    1182.0,
+                    529.0,
+                    1182.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    1194.0,
+                    770.0,
+                    1194.0,
+                    770.0,
+                    1226.0,
+                    297.0,
+                    1226.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1224.0,
+                    425.0,
+                    1224.0,
+                    425.0,
+                    1260.0,
+                    294.0,
+                    1260.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    503.0,
+                    1224.0,
+                    771.0,
+                    1224.0,
+                    771.0,
+                    1260.0,
+                    503.0,
+                    1260.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1255.0,
+                    775.0,
+                    1255.0,
+                    775.0,
+                    1289.0,
+                    295.0,
+                    1289.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1285.0,
+                    771.0,
+                    1285.0,
+                    771.0,
+                    1319.0,
+                    295.0,
+                    1319.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1317.0,
+                    771.0,
+                    1317.0,
+                    771.0,
+                    1348.0,
+                    295.0,
+                    1348.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1347.0,
+                    771.0,
+                    1347.0,
+                    771.0,
+                    1382.0,
+                    296.0,
+                    1382.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1379.0,
+                    773.0,
+                    1379.0,
+                    773.0,
+                    1409.0,
+                    295.0,
+                    1409.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1408.0,
+                    775.0,
+                    1408.0,
+                    775.0,
+                    1439.0,
+                    296.0,
+                    1439.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1437.0,
+                    773.0,
+                    1437.0,
+                    773.0,
+                    1474.0,
+                    294.0,
+                    1474.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1466.0,
+                    776.0,
+                    1466.0,
+                    776.0,
+                    1503.0,
+                    295.0,
+                    1503.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1498.0,
+                    772.0,
+                    1498.0,
+                    772.0,
+                    1533.0,
+                    295.0,
+                    1533.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1530.0,
+                    772.0,
+                    1530.0,
+                    772.0,
+                    1564.0,
+                    295.0,
+                    1564.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1559.0,
+                    773.0,
+                    1559.0,
+                    773.0,
+                    1596.0,
+                    294.0,
+                    1596.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1589.0,
+                    775.0,
+                    1589.0,
+                    775.0,
+                    1625.0,
+                    293.0,
+                    1625.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1620.0,
+                    774.0,
+                    1620.0,
+                    774.0,
+                    1655.0,
+                    295.0,
+                    1655.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1651.0,
+                    773.0,
+                    1651.0,
+                    773.0,
+                    1686.0,
+                    295.0,
+                    1686.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1684.0,
+                    775.0,
+                    1684.0,
+                    775.0,
+                    1715.0,
+                    295.0,
+                    1715.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1713.0,
+                    773.0,
+                    1713.0,
+                    773.0,
+                    1747.0,
+                    295.0,
+                    1747.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1741.0,
+                    772.0,
+                    1741.0,
+                    772.0,
+                    1777.0,
+                    294.0,
+                    1777.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1769.0,
+                    771.0,
+                    1769.0,
+                    771.0,
+                    1810.0,
+                    294.0,
+                    1810.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    1805.0,
+                    501.0,
+                    1805.0,
+                    501.0,
+                    1835.0,
+                    297.0,
+                    1835.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    459.0,
+                    1406.0,
+                    459.0,
+                    1406.0,
+                    495.0,
+                    294.0,
+                    495.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    492.0,
+                    1403.0,
+                    492.0,
+                    1403.0,
+                    526.0,
+                    294.0,
+                    526.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    521.0,
+                    1005.0,
+                    521.0,
+                    1005.0,
+                    557.0,
+                    294.0,
+                    557.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1911.0,
+                    1408.0,
+                    1911.0,
+                    1408.0,
+                    1946.0,
+                    292.0,
+                    1946.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1942.0,
+                    385.0,
+                    1942.0,
+                    385.0,
+                    1978.0,
+                    293.0,
+                    1978.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    411.0,
+                    1942.0,
+                    1321.0,
+                    1942.0,
+                    1321.0,
+                    1978.0,
+                    411.0,
+                    1978.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1399.0,
+                    1942.0,
+                    1410.0,
+                    1942.0,
+                    1410.0,
+                    1978.0,
+                    1399.0,
+                    1978.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1971.0,
+                    1406.0,
+                    1971.0,
+                    1406.0,
+                    2009.0,
+                    292.0,
+                    2009.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    2002.0,
+                    1083.0,
+                    2002.0,
+                    1083.0,
+                    2038.0,
+                    293.0,
+                    2038.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    236.0,
+                    1157.0,
+                    236.0,
+                    1157.0,
+                    262.0,
+                    295.0,
+                    262.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    261.0,
+                    1403.0,
+                    261.0,
+                    1403.0,
+                    297.0,
+                    294.0,
+                    297.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    292.0,
+                    1403.0,
+                    292.0,
+                    1403.0,
+                    325.0,
+                    295.0,
+                    325.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    324.0,
+                    1405.0,
+                    324.0,
+                    1405.0,
+                    357.0,
+                    295.0,
+                    357.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    355.0,
+                    1405.0,
+                    355.0,
+                    1405.0,
+                    388.0,
+                    295.0,
+                    388.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    382.0,
+                    1406.0,
+                    382.0,
+                    1406.0,
+                    421.0,
+                    292.0,
+                    421.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    414.0,
+                    1167.0,
+                    414.0,
+                    1167.0,
+                    451.0,
+                    294.0,
+                    451.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1849.0,
+                    775.0,
+                    1849.0,
+                    775.0,
+                    1885.0,
+                    294.0,
+                    1885.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1880.0,
+                    772.0,
+                    1880.0,
+                    772.0,
+                    1916.0,
+                    294.0,
+                    1916.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 5,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    952,
+                    1405,
+                    952,
+                    1405,
+                    1258,
+                    297,
+                    1258
+                ],
+                "score": 0.982
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    296,
+                    1344,
+                    1407,
+                    1344,
+                    1407,
+                    1620,
+                    296,
+                    1620
+                ],
+                "score": 0.981
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    231,
+                    826,
+                    231,
+                    826,
+                    657,
+                    298,
+                    657
+                ],
+                "score": 0.98
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    752,
+                    1404,
+                    752,
+                    1404,
+                    935,
+                    298,
+                    935
+                ],
+                "score": 0.979
+            },
+            {
+                "category_id": 4,
+                "poly": [
+                    847,
+                    493,
+                    1404,
+                    493,
+                    1404,
+                    736,
+                    847,
+                    736
+                ],
+                "score": 0.968
+            },
+            {
+                "category_id": 3,
+                "poly": [
+                    847,
+                    232,
+                    1398,
+                    232,
+                    1398,
+                    463,
+                    847,
+                    463
+                ],
+                "score": 0.967
+            },
+            {
+                "category_id": 3,
+                "poly": [
+                    527,
+                    1646,
+                    1170,
+                    1646,
+                    1170,
+                    1892,
+                    527,
+                    1892
+                ],
+                "score": 0.965
+            },
+            {
+                "category_id": 4,
+                "poly": [
+                    298,
+                    1921,
+                    1405,
+                    1921,
+                    1405,
+                    2014,
+                    298,
+                    2014
+                ],
+                "score": 0.948
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    302,
+                    1293,
+                    1253,
+                    1293,
+                    1253,
+                    1322,
+                    302,
+                    1322
+                ],
+                "score": 0.892
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    303,
+                    696,
+                    718,
+                    696,
+                    718,
+                    725,
+                    303,
+                    725
+                ],
+                "score": 0.884
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    300,
+                    76,
+                    852,
+                    76,
+                    852,
+                    104,
+                    300,
+                    104
+                ],
+                "score": 0.861
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    841,
+                    2087,
+                    858,
+                    2087,
+                    858,
+                    2111,
+                    841,
+                    2111
+                ],
+                "score": 0.752
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1312,
+                    1952,
+                    1396,
+                    1952,
+                    1396,
+                    1985,
+                    1312,
+                    1985
+                ],
+                "score": 0.91,
+                "latex": "P ( \\boldsymbol { y } | \\mathbf { x } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    801,
+                    1981,
+                    863,
+                    1981,
+                    863,
+                    2015,
+                    801,
+                    2015
+                ],
+                "score": 0.91,
+                "latex": "P ( \\mathbf { x } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    297,
+                    263,
+                    367,
+                    263,
+                    367,
+                    294,
+                    297,
+                    294
+                ],
+                "score": 0.9,
+                "latex": "\\gamma = 0"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    400,
+                    263,
+                    487,
+                    263,
+                    487,
+                    291,
+                    400,
+                    291
+                ],
+                "score": 0.89,
+                "latex": "\\epsilon = 0 . 1"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    867,
+                    1104,
+                    895,
+                    1104,
+                    895,
+                    1136,
+                    867,
+                    1136
+                ],
+                "score": 0.88,
+                "latex": "y ^ { t }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    890,
+                    524,
+                    1000,
+                    524,
+                    1000,
+                    554,
+                    890,
+                    554
+                ],
+                "score": 0.88,
+                "latex": "{ \\sim } 8 0 { \\cdot } 9 0 \\%"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    495,
+                    982,
+                    523,
+                    982,
+                    523,
+                    1014,
+                    495,
+                    1014
+                ],
+                "score": 0.88,
+                "latex": "y ^ { t }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1233,
+                    1134,
+                    1260,
+                    1134,
+                    1260,
+                    1167,
+                    1233,
+                    1167
+                ],
+                "score": 0.87,
+                "latex": "g ^ { t }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    639,
+                    1012,
+                    671,
+                    1012,
+                    671,
+                    1044,
+                    639,
+                    1044
+                ],
+                "score": 0.87,
+                "latex": "\\mathbf { g } ^ { t }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    500,
+                    1043,
+                    529,
+                    1043,
+                    529,
+                    1071,
+                    500,
+                    1071
+                ],
+                "score": 0.87,
+                "latex": "x ^ { t }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    407,
+                    1135,
+                    435,
+                    1135,
+                    435,
+                    1163,
+                    407,
+                    1163
+                ],
+                "score": 0.86,
+                "latex": "x ^ { t }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    298,
+                    984,
+                    325,
+                    984,
+                    325,
+                    1010,
+                    298,
+                    1010
+                ],
+                "score": 0.85,
+                "latex": "x ^ { t }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    512,
+                    535,
+                    692,
+                    535,
+                    692,
+                    568,
+                    512,
+                    568
+                ],
+                "score": 0.84,
+                "latex": "Q ( { \\mathrm { d e f e c t } } ) \\approx Q ("
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1372,
+                    1166,
+                    1400,
+                    1166,
+                    1400,
+                    1193,
+                    1372,
+                    1193
+                ],
+                "score": 0.84,
+                "latex": "x ^ { t }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    626,
+                    1195,
+                    671,
+                    1195,
+                    671,
+                    1224,
+                    626,
+                    1224
+                ],
+                "score": 0.83,
+                "latex": "\\mathbf { W } ^ { t }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1325,
+                    1559,
+                    1375,
+                    1559,
+                    1375,
+                    1590,
+                    1325,
+                    1590
+                ],
+                "score": 0.82,
+                "latex": "( \\alpha _ { 1 } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    660,
+                    1591,
+                    708,
+                    1591,
+                    708,
+                    1620,
+                    660,
+                    1620
+                ],
+                "score": 0.78,
+                "latex": "\\left( \\alpha _ { 2 } \\right)"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    715,
+                    982,
+                    762,
+                    982,
+                    762,
+                    1011,
+                    715,
+                    1011
+                ],
+                "score": 0.77,
+                "latex": "\\bar { \\mathbf { W } } ^ { t }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    404,
+                    1047,
+                    418,
+                    1047,
+                    418,
+                    1072,
+                    404,
+                    1072
+                ],
+                "score": 0.71,
+                "latex": "t"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    298,
+                    382,
+                    414,
+                    382,
+                    414,
+                    416,
+                    298,
+                    416
+                ],
+                "score": 0.48,
+                "latex": "Q ( { \\mathrm { d e f e c t } } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    297,
+                    382,
+                    322,
+                    382,
+                    322,
+                    415,
+                    297,
+                    415
+                ],
+                "score": 0.4,
+                "latex": "Q"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    681,
+                    389,
+                    696,
+                    389,
+                    696,
+                    410,
+                    681,
+                    410
+                ],
+                "score": 0.38,
+                "latex": "\\epsilon"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    340,
+                    323,
+                    365,
+                    323,
+                    365,
+                    353,
+                    340,
+                    353
+                ],
+                "score": 0.34,
+                "latex": "\\mathbf { Q }"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    843.0,
+                    488.0,
+                    1409.0,
+                    488.0,
+                    1409.0,
+                    531.0,
+                    843.0,
+                    531.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    844.0,
+                    523.0,
+                    889.0,
+                    523.0,
+                    889.0,
+                    558.0,
+                    844.0,
+                    558.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1001.0,
+                    523.0,
+                    1405.0,
+                    523.0,
+                    1405.0,
+                    558.0,
+                    1001.0,
+                    558.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    845.0,
+                    553.0,
+                    1407.0,
+                    553.0,
+                    1407.0,
+                    588.0,
+                    845.0,
+                    588.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    845.0,
+                    583.0,
+                    1404.0,
+                    583.0,
+                    1404.0,
+                    618.0,
+                    845.0,
+                    618.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    845.0,
+                    615.0,
+                    1404.0,
+                    615.0,
+                    1404.0,
+                    648.0,
+                    845.0,
+                    648.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    843.0,
+                    642.0,
+                    1405.0,
+                    642.0,
+                    1405.0,
+                    682.0,
+                    843.0,
+                    682.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    844.0,
+                    674.0,
+                    1404.0,
+                    674.0,
+                    1404.0,
+                    710.0,
+                    844.0,
+                    710.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    846.0,
+                    706.0,
+                    1080.0,
+                    706.0,
+                    1080.0,
+                    739.0,
+                    846.0,
+                    739.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    981.0,
+                    235.0,
+                    1009.0,
+                    235.0,
+                    1009.0,
+                    253.0,
+                    981.0,
+                    253.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1010.0,
+                    241.0,
+                    1023.0,
+                    241.0,
+                    1023.0,
+                    252.0,
+                    1010.0,
+                    252.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1083.0,
+                    244.0,
+                    1112.0,
+                    244.0,
+                    1112.0,
+                    267.0,
+                    1083.0,
+                    267.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1184.0,
+                    234.0,
+                    1223.0,
+                    234.0,
+                    1223.0,
+                    255.0,
+                    1184.0,
+                    255.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1289.0,
+                    235.0,
+                    1332.0,
+                    235.0,
+                    1332.0,
+                    253.0,
+                    1289.0,
+                    253.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    845.0,
+                    248.0,
+                    865.0,
+                    248.0,
+                    865.0,
+                    309.0,
+                    845.0,
+                    309.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    872.0,
+                    251.0,
+                    907.0,
+                    251.0,
+                    907.0,
+                    270.0,
+                    872.0,
+                    270.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    914.0,
+                    263.0,
+                    922.0,
+                    263.0,
+                    922.0,
+                    272.0,
+                    914.0,
+                    272.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    977.0,
+                    260.0,
+                    1011.0,
+                    260.0,
+                    1011.0,
+                    281.0,
+                    977.0,
+                    281.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1283.0,
+                    260.0,
+                    1318.0,
+                    260.0,
+                    1318.0,
+                    281.0,
+                    1283.0,
+                    281.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    863.0,
+                    273.0,
+                    908.0,
+                    273.0,
+                    908.0,
+                    300.0,
+                    863.0,
+                    300.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    977.0,
+                    288.0,
+                    1011.0,
+                    288.0,
+                    1011.0,
+                    309.0,
+                    977.0,
+                    309.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1076.0,
+                    279.0,
+                    1112.0,
+                    279.0,
+                    1112.0,
+                    302.0,
+                    1076.0,
+                    302.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1179.0,
+                    270.0,
+                    1214.0,
+                    270.0,
+                    1214.0,
+                    292.0,
+                    1179.0,
+                    292.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1284.0,
+                    288.0,
+                    1319.0,
+                    288.0,
+                    1319.0,
+                    309.0,
+                    1284.0,
+                    309.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    862.0,
+                    302.0,
+                    908.0,
+                    302.0,
+                    908.0,
+                    323.0,
+                    862.0,
+                    323.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    861.0,
+                    317.0,
+                    911.0,
+                    317.0,
+                    911.0,
+                    408.0,
+                    861.0,
+                    408.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    981.0,
+                    318.0,
+                    1015.0,
+                    318.0,
+                    1015.0,
+                    336.0,
+                    981.0,
+                    336.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1083.0,
+                    318.0,
+                    1112.0,
+                    318.0,
+                    1112.0,
+                    337.0,
+                    1083.0,
+                    337.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1185.0,
+                    316.0,
+                    1223.0,
+                    316.0,
+                    1223.0,
+                    338.0,
+                    1185.0,
+                    338.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1282.0,
+                    318.0,
+                    1316.0,
+                    318.0,
+                    1316.0,
+                    336.0,
+                    1282.0,
+                    336.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    877.0,
+                    343.0,
+                    908.0,
+                    343.0,
+                    908.0,
+                    366.0,
+                    877.0,
+                    366.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    981.0,
+                    350.0,
+                    1010.0,
+                    350.0,
+                    1010.0,
+                    369.0,
+                    981.0,
+                    369.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1082.0,
+                    352.0,
+                    1112.0,
+                    352.0,
+                    1112.0,
+                    375.0,
+                    1082.0,
+                    375.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1185.0,
+                    351.0,
+                    1214.0,
+                    351.0,
+                    1214.0,
+                    373.0,
+                    1185.0,
+                    373.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1281.0,
+                    343.0,
+                    1317.0,
+                    343.0,
+                    1317.0,
+                    362.0,
+                    1281.0,
+                    362.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    879.0,
+                    373.0,
+                    907.0,
+                    373.0,
+                    907.0,
+                    393.0,
+                    879.0,
+                    393.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1030.0,
+                    371.0,
+                    1052.0,
+                    371.0,
+                    1052.0,
+                    382.0,
+                    1030.0,
+                    382.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1280.0,
+                    369.0,
+                    1318.0,
+                    369.0,
+                    1318.0,
+                    391.0,
+                    1280.0,
+                    391.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    981.0,
+                    384.0,
+                    1011.0,
+                    384.0,
+                    1011.0,
+                    403.0,
+                    981.0,
+                    403.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1185.0,
+                    390.0,
+                    1215.0,
+                    390.0,
+                    1215.0,
+                    409.0,
+                    1185.0,
+                    409.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1328.0,
+                    387.0,
+                    1337.0,
+                    387.0,
+                    1337.0,
+                    395.0,
+                    1328.0,
+                    395.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    907.0,
+                    403.0,
+                    922.0,
+                    403.0,
+                    922.0,
+                    421.0,
+                    907.0,
+                    421.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    954.0,
+                    401.0,
+                    994.0,
+                    401.0,
+                    994.0,
+                    422.0,
+                    954.0,
+                    422.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1009.0,
+                    403.0,
+                    1025.0,
+                    403.0,
+                    1025.0,
+                    421.0,
+                    1009.0,
+                    421.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1055.0,
+                    401.0,
+                    1096.0,
+                    401.0,
+                    1096.0,
+                    422.0,
+                    1055.0,
+                    422.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1110.0,
+                    403.0,
+                    1127.0,
+                    403.0,
+                    1127.0,
+                    421.0,
+                    1110.0,
+                    421.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1158.0,
+                    401.0,
+                    1199.0,
+                    401.0,
+                    1199.0,
+                    422.0,
+                    1158.0,
+                    422.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1213.0,
+                    403.0,
+                    1229.0,
+                    403.0,
+                    1229.0,
+                    421.0,
+                    1213.0,
+                    421.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1261.0,
+                    401.0,
+                    1301.0,
+                    401.0,
+                    1301.0,
+                    423.0,
+                    1261.0,
+                    423.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1317.0,
+                    404.0,
+                    1329.0,
+                    404.0,
+                    1329.0,
+                    420.0,
+                    1317.0,
+                    420.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1363.0,
+                    401.0,
+                    1403.0,
+                    401.0,
+                    1403.0,
+                    422.0,
+                    1363.0,
+                    422.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    910.0,
+                    416.0,
+                    977.0,
+                    416.0,
+                    977.0,
+                    441.0,
+                    910.0,
+                    441.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1011.0,
+                    418.0,
+                    1079.0,
+                    418.0,
+                    1079.0,
+                    441.0,
+                    1011.0,
+                    441.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1114.0,
+                    418.0,
+                    1182.0,
+                    418.0,
+                    1182.0,
+                    441.0,
+                    1114.0,
+                    441.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1216.0,
+                    418.0,
+                    1282.0,
+                    418.0,
+                    1282.0,
+                    441.0,
+                    1216.0,
+                    441.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1319.0,
+                    416.0,
+                    1385.0,
+                    416.0,
+                    1385.0,
+                    441.0,
+                    1319.0,
+                    441.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    946.0,
+                    441.0,
+                    1010.0,
+                    441.0,
+                    1010.0,
+                    461.0,
+                    946.0,
+                    461.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1060.0,
+                    443.0,
+                    1148.0,
+                    443.0,
+                    1148.0,
+                    462.0,
+                    1060.0,
+                    462.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1197.0,
+                    442.0,
+                    1338.0,
+                    442.0,
+                    1338.0,
+                    461.0,
+                    1197.0,
+                    461.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    565.0,
+                    1650.0,
+                    723.0,
+                    1650.0,
+                    723.0,
+                    1666.0,
+                    565.0,
+                    1666.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    766.0,
+                    1649.0,
+                    953.0,
+                    1649.0,
+                    953.0,
+                    1668.0,
+                    766.0,
+                    1668.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    998.0,
+                    1649.0,
+                    1153.0,
+                    1649.0,
+                    1153.0,
+                    1668.0,
+                    998.0,
+                    1668.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    732.0,
+                    1661.0,
+                    755.0,
+                    1661.0,
+                    755.0,
+                    1850.0,
+                    732.0,
+                    1850.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    537.0,
+                    1667.0,
+                    562.0,
+                    1667.0,
+                    562.0,
+                    1684.0,
+                    537.0,
+                    1684.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1082.0,
+                    1668.0,
+                    1157.0,
+                    1668.0,
+                    1157.0,
+                    1683.0,
+                    1082.0,
+                    1683.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    954.0,
+                    1674.0,
+                    976.0,
+                    1674.0,
+                    976.0,
+                    1835.0,
+                    954.0,
+                    1835.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    747.0,
+                    1679.0,
+                    776.0,
+                    1679.0,
+                    776.0,
+                    1696.0,
+                    747.0,
+                    1696.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    967.0,
+                    1686.0,
+                    994.0,
+                    1686.0,
+                    994.0,
+                    1705.0,
+                    967.0,
+                    1705.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    536.0,
+                    1700.0,
+                    561.0,
+                    1700.0,
+                    561.0,
+                    1718.0,
+                    536.0,
+                    1718.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    747.0,
+                    1709.0,
+                    777.0,
+                    1709.0,
+                    777.0,
+                    1726.0,
+                    747.0,
+                    1726.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    929.0,
+                    1716.0,
+                    939.0,
+                    1716.0,
+                    939.0,
+                    1725.0,
+                    929.0,
+                    1725.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    525.0,
+                    1728.0,
+                    559.0,
+                    1728.0,
+                    559.0,
+                    1782.0,
+                    525.0,
+                    1782.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    747.0,
+                    1738.0,
+                    777.0,
+                    1738.0,
+                    777.0,
+                    1754.0,
+                    747.0,
+                    1754.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    967.0,
+                    1721.0,
+                    991.0,
+                    1721.0,
+                    991.0,
+                    1740.0,
+                    967.0,
+                    1740.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    967.0,
+                    1756.0,
+                    991.0,
+                    1756.0,
+                    991.0,
+                    1774.0,
+                    967.0,
+                    1774.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    536.0,
+                    1768.0,
+                    561.0,
+                    1768.0,
+                    561.0,
+                    1786.0,
+                    536.0,
+                    1786.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    747.0,
+                    1767.0,
+                    776.0,
+                    1767.0,
+                    776.0,
+                    1784.0,
+                    747.0,
+                    1784.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    538.0,
+                    1803.0,
+                    560.0,
+                    1803.0,
+                    560.0,
+                    1818.0,
+                    538.0,
+                    1818.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    747.0,
+                    1797.0,
+                    776.0,
+                    1797.0,
+                    776.0,
+                    1813.0,
+                    747.0,
+                    1813.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    968.0,
+                    1791.0,
+                    992.0,
+                    1791.0,
+                    992.0,
+                    1808.0,
+                    968.0,
+                    1808.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    747.0,
+                    1826.0,
+                    777.0,
+                    1826.0,
+                    777.0,
+                    1844.0,
+                    747.0,
+                    1844.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    780.0,
+                    1829.0,
+                    803.0,
+                    1829.0,
+                    803.0,
+                    1840.0,
+                    780.0,
+                    1840.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    970.0,
+                    1827.0,
+                    1020.0,
+                    1827.0,
+                    1020.0,
+                    1842.0,
+                    970.0,
+                    1842.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    563.0,
+                    1846.0,
+                    576.0,
+                    1846.0,
+                    576.0,
+                    1858.0,
+                    563.0,
+                    1858.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    589.0,
+                    1845.0,
+                    613.0,
+                    1845.0,
+                    613.0,
+                    1859.0,
+                    589.0,
+                    1859.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    619.0,
+                    1846.0,
+                    646.0,
+                    1846.0,
+                    646.0,
+                    1859.0,
+                    619.0,
+                    1859.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    648.0,
+                    1845.0,
+                    743.0,
+                    1845.0,
+                    743.0,
+                    1860.0,
+                    648.0,
+                    1860.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    779.0,
+                    1846.0,
+                    792.0,
+                    1846.0,
+                    792.0,
+                    1859.0,
+                    779.0,
+                    1859.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    803.0,
+                    1844.0,
+                    830.0,
+                    1844.0,
+                    830.0,
+                    1861.0,
+                    803.0,
+                    1861.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    835.0,
+                    1845.0,
+                    861.0,
+                    1845.0,
+                    861.0,
+                    1859.0,
+                    835.0,
+                    1859.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    864.0,
+                    1845.0,
+                    959.0,
+                    1845.0,
+                    959.0,
+                    1860.0,
+                    864.0,
+                    1860.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    995.0,
+                    1846.0,
+                    1007.0,
+                    1846.0,
+                    1007.0,
+                    1858.0,
+                    995.0,
+                    1858.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1020.0,
+                    1845.0,
+                    1044.0,
+                    1845.0,
+                    1044.0,
+                    1859.0,
+                    1020.0,
+                    1859.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1049.0,
+                    1845.0,
+                    1175.0,
+                    1845.0,
+                    1175.0,
+                    1860.0,
+                    1049.0,
+                    1860.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    619.0,
+                    1855.0,
+                    670.0,
+                    1855.0,
+                    670.0,
+                    1875.0,
+                    619.0,
+                    1875.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    834.0,
+                    1855.0,
+                    886.0,
+                    1855.0,
+                    886.0,
+                    1874.0,
+                    834.0,
+                    1874.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1049.0,
+                    1855.0,
+                    1101.0,
+                    1855.0,
+                    1101.0,
+                    1875.0,
+                    1049.0,
+                    1875.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    793.0,
+                    1874.0,
+                    836.0,
+                    1874.0,
+                    836.0,
+                    1889.0,
+                    793.0,
+                    1889.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    854.0,
+                    1877.0,
+                    868.0,
+                    1877.0,
+                    868.0,
+                    1888.0,
+                    854.0,
+                    1888.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    873.0,
+                    1873.0,
+                    915.0,
+                    1873.0,
+                    915.0,
+                    1890.0,
+                    873.0,
+                    1890.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    668.0,
+                    1668.5,
+                    727.0,
+                    1668.5,
+                    727.0,
+                    1685.5,
+                    668.0,
+                    1685.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    885.0,
+                    1681.5,
+                    944.0,
+                    1681.5,
+                    944.0,
+                    1697.0,
+                    885.0,
+                    1697.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1922.0,
+                    1406.0,
+                    1922.0,
+                    1406.0,
+                    1955.0,
+                    296.0,
+                    1955.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1951.0,
+                    1311.0,
+                    1951.0,
+                    1311.0,
+                    1988.0,
+                    293.0,
+                    1988.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1397.0,
+                    1951.0,
+                    1407.0,
+                    1951.0,
+                    1407.0,
+                    1988.0,
+                    1397.0,
+                    1988.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1982.0,
+                    800.0,
+                    1982.0,
+                    800.0,
+                    2016.0,
+                    296.0,
+                    2016.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    864.0,
+                    1982.0,
+                    1401.0,
+                    1982.0,
+                    1401.0,
+                    2016.0,
+                    864.0,
+                    2016.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1289.0,
+                    1258.0,
+                    1289.0,
+                    1258.0,
+                    1326.0,
+                    294.0,
+                    1326.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    692.0,
+                    725.0,
+                    692.0,
+                    725.0,
+                    731.0,
+                    294.0,
+                    731.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    73.0,
+                    856.0,
+                    73.0,
+                    856.0,
+                    108.0,
+                    297.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    839.0,
+                    2086.0,
+                    860.0,
+                    2086.0,
+                    860.0,
+                    2118.0,
+                    839.0,
+                    2118.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    950.0,
+                    1408.0,
+                    950.0,
+                    1408.0,
+                    988.0,
+                    294.0,
+                    988.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    978.0,
+                    297.0,
+                    978.0,
+                    297.0,
+                    1019.0,
+                    292.0,
+                    1019.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    326.0,
+                    978.0,
+                    494.0,
+                    978.0,
+                    494.0,
+                    1019.0,
+                    326.0,
+                    1019.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    524.0,
+                    978.0,
+                    714.0,
+                    978.0,
+                    714.0,
+                    1019.0,
+                    524.0,
+                    1019.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    763.0,
+                    978.0,
+                    1408.0,
+                    978.0,
+                    1408.0,
+                    1019.0,
+                    763.0,
+                    1019.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1011.0,
+                    638.0,
+                    1011.0,
+                    638.0,
+                    1047.0,
+                    295.0,
+                    1047.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    672.0,
+                    1011.0,
+                    1403.0,
+                    1011.0,
+                    1403.0,
+                    1047.0,
+                    672.0,
+                    1047.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1043.0,
+                    403.0,
+                    1043.0,
+                    403.0,
+                    1078.0,
+                    295.0,
+                    1078.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    419.0,
+                    1043.0,
+                    499.0,
+                    1043.0,
+                    499.0,
+                    1078.0,
+                    419.0,
+                    1078.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    530.0,
+                    1043.0,
+                    1405.0,
+                    1043.0,
+                    1405.0,
+                    1078.0,
+                    530.0,
+                    1078.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1073.0,
+                    1405.0,
+                    1073.0,
+                    1405.0,
+                    1109.0,
+                    295.0,
+                    1109.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1104.0,
+                    866.0,
+                    1104.0,
+                    866.0,
+                    1142.0,
+                    292.0,
+                    1142.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    896.0,
+                    1104.0,
+                    1407.0,
+                    1104.0,
+                    1407.0,
+                    1142.0,
+                    896.0,
+                    1142.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1134.0,
+                    406.0,
+                    1134.0,
+                    406.0,
+                    1170.0,
+                    295.0,
+                    1170.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    436.0,
+                    1134.0,
+                    1232.0,
+                    1134.0,
+                    1232.0,
+                    1170.0,
+                    436.0,
+                    1170.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1261.0,
+                    1134.0,
+                    1407.0,
+                    1134.0,
+                    1407.0,
+                    1170.0,
+                    1261.0,
+                    1170.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1162.0,
+                    1371.0,
+                    1162.0,
+                    1371.0,
+                    1201.0,
+                    292.0,
+                    1201.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1401.0,
+                    1162.0,
+                    1406.0,
+                    1162.0,
+                    1406.0,
+                    1201.0,
+                    1401.0,
+                    1201.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1194.0,
+                    625.0,
+                    1194.0,
+                    625.0,
+                    1233.0,
+                    294.0,
+                    1233.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    672.0,
+                    1194.0,
+                    1406.0,
+                    1194.0,
+                    1406.0,
+                    1233.0,
+                    672.0,
+                    1233.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1224.0,
+                    1383.0,
+                    1224.0,
+                    1383.0,
+                    1262.0,
+                    294.0,
+                    1262.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1344.0,
+                    1407.0,
+                    1344.0,
+                    1407.0,
+                    1380.0,
+                    294.0,
+                    1380.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1374.0,
+                    1405.0,
+                    1374.0,
+                    1405.0,
+                    1409.0,
+                    291.0,
+                    1409.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1405.0,
+                    1407.0,
+                    1405.0,
+                    1407.0,
+                    1442.0,
+                    294.0,
+                    1442.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1434.0,
+                    1408.0,
+                    1434.0,
+                    1408.0,
+                    1473.0,
+                    291.0,
+                    1473.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1462.0,
+                    1408.0,
+                    1462.0,
+                    1408.0,
+                    1504.0,
+                    293.0,
+                    1504.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1495.0,
+                    1407.0,
+                    1495.0,
+                    1407.0,
+                    1533.0,
+                    293.0,
+                    1533.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1526.0,
+                    1407.0,
+                    1526.0,
+                    1407.0,
+                    1562.0,
+                    293.0,
+                    1562.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1559.0,
+                    1324.0,
+                    1559.0,
+                    1324.0,
+                    1592.0,
+                    295.0,
+                    1592.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1376.0,
+                    1559.0,
+                    1405.0,
+                    1559.0,
+                    1405.0,
+                    1592.0,
+                    1376.0,
+                    1592.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1588.0,
+                    659.0,
+                    1588.0,
+                    659.0,
+                    1624.0,
+                    294.0,
+                    1624.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    709.0,
+                    1588.0,
+                    1071.0,
+                    1588.0,
+                    1071.0,
+                    1624.0,
+                    709.0,
+                    1624.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    230.0,
+                    825.0,
+                    230.0,
+                    825.0,
+                    265.0,
+                    296.0,
+                    265.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    260.0,
+                    296.0,
+                    260.0,
+                    296.0,
+                    296.0,
+                    293.0,
+                    296.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    368.0,
+                    260.0,
+                    399.0,
+                    260.0,
+                    399.0,
+                    296.0,
+                    368.0,
+                    296.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    488.0,
+                    260.0,
+                    827.0,
+                    260.0,
+                    827.0,
+                    296.0,
+                    488.0,
+                    296.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    292.0,
+                    829.0,
+                    292.0,
+                    829.0,
+                    327.0,
+                    294.0,
+                    327.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    322.0,
+                    339.0,
+                    322.0,
+                    339.0,
+                    354.0,
+                    296.0,
+                    354.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    366.0,
+                    322.0,
+                    831.0,
+                    322.0,
+                    831.0,
+                    354.0,
+                    366.0,
+                    354.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    354.0,
+                    828.0,
+                    354.0,
+                    828.0,
+                    383.0,
+                    294.0,
+                    383.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    415.0,
+                    382.0,
+                    680.0,
+                    382.0,
+                    680.0,
+                    417.0,
+                    415.0,
+                    417.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    697.0,
+                    382.0,
+                    831.0,
+                    382.0,
+                    831.0,
+                    417.0,
+                    697.0,
+                    417.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    415.0,
+                    828.0,
+                    415.0,
+                    828.0,
+                    447.0,
+                    295.0,
+                    447.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    442.0,
+                    827.0,
+                    442.0,
+                    827.0,
+                    479.0,
+                    294.0,
+                    479.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    474.0,
+                    831.0,
+                    474.0,
+                    831.0,
+                    509.0,
+                    295.0,
+                    509.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    507.0,
+                    829.0,
+                    507.0,
+                    829.0,
+                    539.0,
+                    296.0,
+                    539.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    534.0,
+                    511.0,
+                    534.0,
+                    511.0,
+                    569.0,
+                    294.0,
+                    569.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    693.0,
+                    534.0,
+                    828.0,
+                    534.0,
+                    828.0,
+                    569.0,
+                    693.0,
+                    569.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    565.0,
+                    828.0,
+                    565.0,
+                    828.0,
+                    599.0,
+                    294.0,
+                    599.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    596.0,
+                    827.0,
+                    596.0,
+                    827.0,
+                    628.0,
+                    296.0,
+                    628.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    625.0,
+                    751.0,
+                    625.0,
+                    751.0,
+                    661.0,
+                    294.0,
+                    661.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    751.0,
+                    1405.0,
+                    751.0,
+                    1405.0,
+                    787.0,
+                    294.0,
+                    787.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    782.0,
+                    1405.0,
+                    782.0,
+                    1405.0,
+                    817.0,
+                    294.0,
+                    817.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    813.0,
+                    1406.0,
+                    813.0,
+                    1406.0,
+                    847.0,
+                    293.0,
+                    847.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    844.0,
+                    1406.0,
+                    844.0,
+                    1406.0,
+                    879.0,
+                    294.0,
+                    879.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    874.0,
+                    1405.0,
+                    874.0,
+                    1405.0,
+                    908.0,
+                    293.0,
+                    908.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    903.0,
+                    1401.0,
+                    903.0,
+                    1401.0,
+                    941.0,
+                    292.0,
+                    941.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 6,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1437,
+                    1404,
+                    1437,
+                    1404,
+                    1744,
+                    297,
+                    1744
+                ],
+                "score": 0.983
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1758,
+                    1404,
+                    1758,
+                    1404,
+                    2034,
+                    298,
+                    2034
+                ],
+                "score": 0.982
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    705,
+                    1404,
+                    705,
+                    1404,
+                    1012,
+                    298,
+                    1012
+                ],
+                "score": 0.982
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1027,
+                    1404,
+                    1027,
+                    1404,
+                    1243,
+                    298,
+                    1243
+                ],
+                "score": 0.98
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    276,
+                    1405,
+                    276,
+                    1405,
+                    522,
+                    298,
+                    522
+                ],
+                "score": 0.98
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    537,
+                    1404,
+                    537,
+                    1404,
+                    690,
+                    299,
+                    690
+                ],
+                "score": 0.977
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    301,
+                    1329,
+                    1402,
+                    1329,
+                    1402,
+                    1423,
+                    301,
+                    1423
+                ],
+                "score": 0.959
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    303,
+                    1279,
+                    798,
+                    1279,
+                    798,
+                    1314,
+                    303,
+                    1314
+                ],
+                "score": 0.906
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    300,
+                    227,
+                    581,
+                    227,
+                    581,
+                    261,
+                    300,
+                    261
+                ],
+                "score": 0.898
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    841,
+                    2089,
+                    858,
+                    2089,
+                    858,
+                    2111,
+                    841,
+                    2111
+                ],
+                "score": 0.792
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    300,
+                    76,
+                    852,
+                    76,
+                    852,
+                    104,
+                    300,
+                    104
+                ],
+                "score": 0.515
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    299,
+                    76,
+                    854,
+                    76,
+                    854,
+                    104,
+                    299,
+                    104
+                ],
+                "score": 0.498
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1275.0,
+                    804.0,
+                    1275.0,
+                    804.0,
+                    1322.0,
+                    293.0,
+                    1322.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    222.0,
+                    586.0,
+                    222.0,
+                    586.0,
+                    267.0,
+                    292.0,
+                    267.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    839.0,
+                    2086.0,
+                    860.0,
+                    2086.0,
+                    860.0,
+                    2116.0,
+                    839.0,
+                    2116.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    73.0,
+                    856.0,
+                    73.0,
+                    856.0,
+                    108.0,
+                    297.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    73.0,
+                    857.0,
+                    73.0,
+                    857.0,
+                    108.0,
+                    297.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1437.0,
+                    1406.0,
+                    1437.0,
+                    1406.0,
+                    1474.0,
+                    294.0,
+                    1474.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1470.0,
+                    1405.0,
+                    1470.0,
+                    1405.0,
+                    1501.0,
+                    292.0,
+                    1501.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1499.0,
+                    1405.0,
+                    1499.0,
+                    1405.0,
+                    1535.0,
+                    292.0,
+                    1535.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1527.0,
+                    1406.0,
+                    1527.0,
+                    1406.0,
+                    1564.0,
+                    294.0,
+                    1564.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1560.0,
+                    1406.0,
+                    1560.0,
+                    1406.0,
+                    1596.0,
+                    295.0,
+                    1596.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1590.0,
+                    1406.0,
+                    1590.0,
+                    1406.0,
+                    1626.0,
+                    294.0,
+                    1626.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1618.0,
+                    1406.0,
+                    1618.0,
+                    1406.0,
+                    1657.0,
+                    294.0,
+                    1657.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1648.0,
+                    1408.0,
+                    1648.0,
+                    1408.0,
+                    1690.0,
+                    291.0,
+                    1690.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1682.0,
+                    1406.0,
+                    1682.0,
+                    1406.0,
+                    1718.0,
+                    294.0,
+                    1718.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1710.0,
+                    1301.0,
+                    1710.0,
+                    1301.0,
+                    1748.0,
+                    292.0,
+                    1748.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    1760.0,
+                    1404.0,
+                    1760.0,
+                    1404.0,
+                    1793.0,
+                    297.0,
+                    1793.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1789.0,
+                    1406.0,
+                    1789.0,
+                    1406.0,
+                    1824.0,
+                    294.0,
+                    1824.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1819.0,
+                    1408.0,
+                    1819.0,
+                    1408.0,
+                    1859.0,
+                    293.0,
+                    1859.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1850.0,
+                    1405.0,
+                    1850.0,
+                    1405.0,
+                    1887.0,
+                    294.0,
+                    1887.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1882.0,
+                    1406.0,
+                    1882.0,
+                    1406.0,
+                    1918.0,
+                    294.0,
+                    1918.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1912.0,
+                    1405.0,
+                    1912.0,
+                    1405.0,
+                    1948.0,
+                    294.0,
+                    1948.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1939.0,
+                    1408.0,
+                    1939.0,
+                    1408.0,
+                    1980.0,
+                    292.0,
+                    1980.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1964.0,
+                    1406.0,
+                    1964.0,
+                    1406.0,
+                    2015.0,
+                    291.0,
+                    2015.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    2003.0,
+                    1321.0,
+                    2003.0,
+                    1321.0,
+                    2038.0,
+                    293.0,
+                    2038.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    707.0,
+                    1404.0,
+                    707.0,
+                    1404.0,
+                    739.0,
+                    296.0,
+                    739.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    739.0,
+                    1406.0,
+                    739.0,
+                    1406.0,
+                    771.0,
+                    294.0,
+                    771.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    766.0,
+                    1406.0,
+                    766.0,
+                    1406.0,
+                    803.0,
+                    293.0,
+                    803.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    796.0,
+                    1406.0,
+                    796.0,
+                    1406.0,
+                    833.0,
+                    293.0,
+                    833.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    825.0,
+                    1406.0,
+                    825.0,
+                    1406.0,
+                    866.0,
+                    293.0,
+                    866.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    859.0,
+                    1406.0,
+                    859.0,
+                    1406.0,
+                    894.0,
+                    293.0,
+                    894.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    885.0,
+                    1408.0,
+                    885.0,
+                    1408.0,
+                    927.0,
+                    292.0,
+                    927.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    917.0,
+                    1406.0,
+                    917.0,
+                    1406.0,
+                    955.0,
+                    293.0,
+                    955.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    950.0,
+                    1406.0,
+                    950.0,
+                    1406.0,
+                    986.0,
+                    294.0,
+                    986.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    981.0,
+                    1408.0,
+                    981.0,
+                    1408.0,
+                    1016.0,
+                    294.0,
+                    1016.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1028.0,
+                    1406.0,
+                    1028.0,
+                    1406.0,
+                    1062.0,
+                    294.0,
+                    1062.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1058.0,
+                    1406.0,
+                    1058.0,
+                    1406.0,
+                    1093.0,
+                    294.0,
+                    1093.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1089.0,
+                    1405.0,
+                    1089.0,
+                    1405.0,
+                    1124.0,
+                    294.0,
+                    1124.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1118.0,
+                    1406.0,
+                    1118.0,
+                    1406.0,
+                    1155.0,
+                    293.0,
+                    1155.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1149.0,
+                    1408.0,
+                    1149.0,
+                    1408.0,
+                    1186.0,
+                    293.0,
+                    1186.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1177.0,
+                    1405.0,
+                    1177.0,
+                    1405.0,
+                    1216.0,
+                    293.0,
+                    1216.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1212.0,
+                    1227.0,
+                    1212.0,
+                    1227.0,
+                    1246.0,
+                    296.0,
+                    1246.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    278.0,
+                    1403.0,
+                    278.0,
+                    1403.0,
+                    311.0,
+                    295.0,
+                    311.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    309.0,
+                    1407.0,
+                    309.0,
+                    1407.0,
+                    343.0,
+                    293.0,
+                    343.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    339.0,
+                    1405.0,
+                    339.0,
+                    1405.0,
+                    373.0,
+                    295.0,
+                    373.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    369.0,
+                    1405.0,
+                    369.0,
+                    1405.0,
+                    405.0,
+                    293.0,
+                    405.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    398.0,
+                    1403.0,
+                    398.0,
+                    1403.0,
+                    434.0,
+                    293.0,
+                    434.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    429.0,
+                    1406.0,
+                    429.0,
+                    1406.0,
+                    466.0,
+                    293.0,
+                    466.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    462.0,
+                    1406.0,
+                    462.0,
+                    1406.0,
+                    496.0,
+                    293.0,
+                    496.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    492.0,
+                    1407.0,
+                    492.0,
+                    1407.0,
+                    525.0,
+                    295.0,
+                    525.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    536.0,
+                    1405.0,
+                    536.0,
+                    1405.0,
+                    573.0,
+                    295.0,
+                    573.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    566.0,
+                    1406.0,
+                    566.0,
+                    1406.0,
+                    603.0,
+                    294.0,
+                    603.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    598.0,
+                    1406.0,
+                    598.0,
+                    1406.0,
+                    633.0,
+                    293.0,
+                    633.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    629.0,
+                    1406.0,
+                    629.0,
+                    1406.0,
+                    664.0,
+                    293.0,
+                    664.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    658.0,
+                    1295.0,
+                    658.0,
+                    1295.0,
+                    694.0,
+                    293.0,
+                    694.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1327.0,
+                    1407.0,
+                    1327.0,
+                    1407.0,
+                    1366.0,
+                    294.0,
+                    1366.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1360.0,
+                    1404.0,
+                    1360.0,
+                    1404.0,
+                    1394.0,
+                    296.0,
+                    1394.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1389.0,
+                    1408.0,
+                    1389.0,
+                    1408.0,
+                    1428.0,
+                    292.0,
+                    1428.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 7,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 0,
+                "poly": [
+                    299,
+                    227,
+                    489,
+                    227,
+                    489,
+                    262,
+                    299,
+                    262
+                ],
+                "score": 0.825
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    840,
+                    2088,
+                    859,
+                    2088,
+                    859,
+                    2111,
+                    840,
+                    2111
+                ],
+                "score": 0.779
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1688,
+                    1403,
+                    1688,
+                    1403,
+                    1753,
+                    297,
+                    1753
+                ],
+                "score": 0.773
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    278,
+                    1404,
+                    278,
+                    1404,
+                    343,
+                    298,
+                    343
+                ],
+                "score": 0.772
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    295,
+                    1605,
+                    1405,
+                    1605,
+                    1405,
+                    1670,
+                    295,
+                    1670
+                ],
+                "score": 0.765
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    293,
+                    1323,
+                    1404,
+                    1323,
+                    1404,
+                    1389,
+                    293,
+                    1389
+                ],
+                "score": 0.758
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    294,
+                    560,
+                    1404,
+                    560,
+                    1404,
+                    624,
+                    294,
+                    624
+                ],
+                "score": 0.747
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    295,
+                    1042,
+                    1401,
+                    1042,
+                    1401,
+                    1107,
+                    295,
+                    1107
+                ],
+                "score": 0.745
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    1407,
+                    1406,
+                    1407,
+                    1406,
+                    1503,
+                    299,
+                    1503
+                ],
+                "score": 0.736
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    294,
+                    1522,
+                    1405,
+                    1522,
+                    1405,
+                    1587,
+                    294,
+                    1587
+                ],
+                "score": 0.726
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    958,
+                    1401,
+                    958,
+                    1401,
+                    1023,
+                    297,
+                    1023
+                ],
+                "score": 0.722
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    301,
+                    730,
+                    1401,
+                    730,
+                    1401,
+                    796,
+                    301,
+                    796
+                ],
+                "score": 0.715
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    294,
+                    647,
+                    1406,
+                    647,
+                    1406,
+                    711,
+                    294,
+                    711
+                ],
+                "score": 0.713
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    295,
+                    1239,
+                    1401,
+                    1239,
+                    1401,
+                    1306,
+                    295,
+                    1306
+                ],
+                "score": 0.705
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    292,
+                    475,
+                    1404,
+                    475,
+                    1404,
+                    541,
+                    292,
+                    541
+                ],
+                "score": 0.704
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    294,
+                    1773,
+                    1401,
+                    1773,
+                    1401,
+                    1839,
+                    294,
+                    1839
+                ],
+                "score": 0.703
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    361,
+                    1400,
+                    361,
+                    1400,
+                    456,
+                    298,
+                    456
+                ],
+                "score": 0.689
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    293,
+                    1856,
+                    1401,
+                    1856,
+                    1401,
+                    1922,
+                    293,
+                    1922
+                ],
+                "score": 0.669
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    813,
+                    1406,
+                    813,
+                    1406,
+                    939,
+                    297,
+                    939
+                ],
+                "score": 0.627
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    1941,
+                    1404,
+                    1941,
+                    1404,
+                    2034,
+                    299,
+                    2034
+                ],
+                "score": 0.59
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1124,
+                    1405,
+                    1124,
+                    1405,
+                    1222,
+                    297,
+                    1222
+                ],
+                "score": 0.547
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    298,
+                    75,
+                    854,
+                    75,
+                    854,
+                    105,
+                    298,
+                    105
+                ],
+                "score": 0.513
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    298,
+                    75,
+                    854,
+                    75,
+                    854,
+                    105,
+                    298,
+                    105
+                ],
+                "score": 0.375
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    230.0,
+                    490.0,
+                    230.0,
+                    490.0,
+                    263.0,
+                    297.0,
+                    263.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    839.0,
+                    2087.0,
+                    861.0,
+                    2087.0,
+                    861.0,
+                    2117.0,
+                    839.0,
+                    2117.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    73.0,
+                    857.0,
+                    73.0,
+                    857.0,
+                    108.0,
+                    297.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    73.0,
+                    857.0,
+                    73.0,
+                    857.0,
+                    108.0,
+                    297.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1686.0,
+                    1404.0,
+                    1686.0,
+                    1404.0,
+                    1727.0,
+                    295.0,
+                    1727.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    1721.0,
+                    499.0,
+                    1721.0,
+                    499.0,
+                    1751.0,
+                    322.0,
+                    1751.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    276.0,
+                    1406.0,
+                    276.0,
+                    1406.0,
+                    314.0,
+                    293.0,
+                    314.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    309.0,
+                    885.0,
+                    309.0,
+                    885.0,
+                    345.0,
+                    323.0,
+                    345.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1601.0,
+                    1408.0,
+                    1601.0,
+                    1408.0,
+                    1644.0,
+                    293.0,
+                    1644.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    318.0,
+                    1634.0,
+                    399.0,
+                    1634.0,
+                    399.0,
+                    1672.0,
+                    318.0,
+                    1672.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1324.0,
+                    1404.0,
+                    1324.0,
+                    1404.0,
+                    1360.0,
+                    295.0,
+                    1360.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    321.0,
+                    1356.0,
+                    558.0,
+                    1356.0,
+                    558.0,
+                    1392.0,
+                    321.0,
+                    1392.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    560.0,
+                    1404.0,
+                    560.0,
+                    1404.0,
+                    596.0,
+                    296.0,
+                    596.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    321.0,
+                    592.0,
+                    807.0,
+                    592.0,
+                    807.0,
+                    624.0,
+                    321.0,
+                    624.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1040.0,
+                    1405.0,
+                    1040.0,
+                    1405.0,
+                    1081.0,
+                    293.0,
+                    1081.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    319.0,
+                    1075.0,
+                    705.0,
+                    1075.0,
+                    705.0,
+                    1108.0,
+                    319.0,
+                    1108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1408.0,
+                    1408.0,
+                    1408.0,
+                    1408.0,
+                    1443.0,
+                    296.0,
+                    1443.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    1439.0,
+                    1413.0,
+                    1439.0,
+                    1413.0,
+                    1473.0,
+                    322.0,
+                    1473.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    1471.0,
+                    539.0,
+                    1471.0,
+                    539.0,
+                    1502.0,
+                    323.0,
+                    1502.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1518.0,
+                    1410.0,
+                    1518.0,
+                    1410.0,
+                    1561.0,
+                    292.0,
+                    1561.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    1552.0,
+                    1321.0,
+                    1552.0,
+                    1321.0,
+                    1589.0,
+                    322.0,
+                    1589.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    960.0,
+                    1403.0,
+                    960.0,
+                    1403.0,
+                    996.0,
+                    295.0,
+                    996.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    989.0,
+                    711.0,
+                    989.0,
+                    711.0,
+                    1021.0,
+                    323.0,
+                    1021.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    729.0,
+                    1405.0,
+                    729.0,
+                    1405.0,
+                    769.0,
+                    293.0,
+                    769.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    321.0,
+                    760.0,
+                    1141.0,
+                    760.0,
+                    1141.0,
+                    797.0,
+                    321.0,
+                    797.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    643.0,
+                    1408.0,
+                    643.0,
+                    1408.0,
+                    685.0,
+                    293.0,
+                    685.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    324.0,
+                    679.0,
+                    1095.0,
+                    679.0,
+                    1095.0,
+                    711.0,
+                    324.0,
+                    711.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1235.0,
+                    1405.0,
+                    1235.0,
+                    1405.0,
+                    1280.0,
+                    293.0,
+                    1280.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    1273.0,
+                    937.0,
+                    1273.0,
+                    937.0,
+                    1305.0,
+                    323.0,
+                    1305.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    470.0,
+                    1406.0,
+                    470.0,
+                    1406.0,
+                    515.0,
+                    293.0,
+                    515.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    320.0,
+                    506.0,
+                    612.0,
+                    506.0,
+                    612.0,
+                    540.0,
+                    320.0,
+                    540.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1771.0,
+                    1407.0,
+                    1771.0,
+                    1407.0,
+                    1811.0,
+                    294.0,
+                    1811.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    321.0,
+                    1805.0,
+                    1225.0,
+                    1805.0,
+                    1225.0,
+                    1840.0,
+                    321.0,
+                    1840.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    358.0,
+                    1405.0,
+                    358.0,
+                    1405.0,
+                    401.0,
+                    293.0,
+                    401.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    324.0,
+                    395.0,
+                    1405.0,
+                    395.0,
+                    1405.0,
+                    429.0,
+                    324.0,
+                    429.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    324.0,
+                    424.0,
+                    843.0,
+                    424.0,
+                    843.0,
+                    458.0,
+                    324.0,
+                    458.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1854.0,
+                    1404.0,
+                    1854.0,
+                    1404.0,
+                    1894.0,
+                    295.0,
+                    1894.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    1888.0,
+                    1086.0,
+                    1888.0,
+                    1086.0,
+                    1921.0,
+                    322.0,
+                    1921.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    816.0,
+                    1406.0,
+                    816.0,
+                    1406.0,
+                    849.0,
+                    296.0,
+                    849.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    320.0,
+                    845.0,
+                    1411.0,
+                    845.0,
+                    1411.0,
+                    881.0,
+                    320.0,
+                    881.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    321.0,
+                    874.0,
+                    1407.0,
+                    874.0,
+                    1407.0,
+                    913.0,
+                    321.0,
+                    913.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    325.0,
+                    909.0,
+                    489.0,
+                    909.0,
+                    489.0,
+                    938.0,
+                    325.0,
+                    938.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1942.0,
+                    1405.0,
+                    1942.0,
+                    1405.0,
+                    1975.0,
+                    295.0,
+                    1975.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    1973.0,
+                    1405.0,
+                    1973.0,
+                    1405.0,
+                    2007.0,
+                    323.0,
+                    2007.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    2001.0,
+                    900.0,
+                    2001.0,
+                    900.0,
+                    2036.0,
+                    322.0,
+                    2036.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1125.0,
+                    1405.0,
+                    1125.0,
+                    1405.0,
+                    1163.0,
+                    295.0,
+                    1163.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    324.0,
+                    1157.0,
+                    1403.0,
+                    1157.0,
+                    1403.0,
+                    1192.0,
+                    324.0,
+                    1192.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    324.0,
+                    1187.0,
+                    1198.0,
+                    1187.0,
+                    1198.0,
+                    1225.0,
+                    324.0,
+                    1225.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 8,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 2,
+                "poly": [
+                    836,
+                    2088,
+                    864,
+                    2088,
+                    864,
+                    2113,
+                    836,
+                    2113
+                ],
+                "score": 0.836
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    298,
+                    75,
+                    854,
+                    75,
+                    854,
+                    105,
+                    298,
+                    105
+                ],
+                "score": 0.822
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    288,
+                    1327,
+                    1405,
+                    1327,
+                    1405,
+                    1393,
+                    288,
+                    1393
+                ],
+                "score": 0.811
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1410,
+                    1405,
+                    1410,
+                    1405,
+                    1505,
+                    297,
+                    1505
+                ],
+                "score": 0.802
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1182,
+                    1404,
+                    1182,
+                    1404,
+                    1309,
+                    298,
+                    1309
+                ],
+                "score": 0.775
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1100,
+                    1400,
+                    1100,
+                    1400,
+                    1164,
+                    298,
+                    1164
+                ],
+                "score": 0.773
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    291,
+                    1691,
+                    1403,
+                    1691,
+                    1403,
+                    1756,
+                    291,
+                    1756
+                ],
+                "score": 0.728
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    294,
+                    1607,
+                    1404,
+                    1607,
+                    1404,
+                    1671,
+                    294,
+                    1671
+                ],
+                "score": 0.718
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    296,
+                    902,
+                    1406,
+                    902,
+                    1406,
+                    996,
+                    296,
+                    996
+                ],
+                "score": 0.713
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    292,
+                    1523,
+                    1405,
+                    1523,
+                    1405,
+                    1590,
+                    292,
+                    1590
+                ],
+                "score": 0.698
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    296,
+                    1017,
+                    1399,
+                    1017,
+                    1399,
+                    1081,
+                    296,
+                    1081
+                ],
+                "score": 0.691
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    294,
+                    229,
+                    1403,
+                    229,
+                    1403,
+                    295,
+                    294,
+                    295
+                ],
+                "score": 0.683
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    296,
+                    312,
+                    1401,
+                    312,
+                    1401,
+                    377,
+                    296,
+                    377
+                ],
+                "score": 0.659
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    705,
+                    1399,
+                    705,
+                    1399,
+                    772,
+                    297,
+                    772
+                ],
+                "score": 0.64
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1971,
+                    1396,
+                    1971,
+                    1396,
+                    2035,
+                    297,
+                    2035
+                ],
+                "score": 0.632
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    293,
+                    478,
+                    1404,
+                    478,
+                    1404,
+                    544,
+                    293,
+                    544
+                ],
+                "score": 0.62
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    295,
+                    395,
+                    1404,
+                    395,
+                    1404,
+                    460,
+                    295,
+                    460
+                ],
+                "score": 0.614
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1888,
+                    1399,
+                    1888,
+                    1399,
+                    1953,
+                    297,
+                    1953
+                ],
+                "score": 0.582
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1774,
+                    1402,
+                    1774,
+                    1402,
+                    1869,
+                    298,
+                    1869
+                ],
+                "score": 0.576
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    295,
+                    562,
+                    1407,
+                    562,
+                    1407,
+                    688,
+                    295,
+                    688
+                ],
+                "score": 0.557
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    300,
+                    788,
+                    1403,
+                    788,
+                    1403,
+                    885,
+                    300,
+                    885
+                ],
+                "score": 0.493
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    831.0,
+                    2085.0,
+                    869.0,
+                    2085.0,
+                    869.0,
+                    2123.0,
+                    831.0,
+                    2123.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    73.0,
+                    857.0,
+                    73.0,
+                    857.0,
+                    108.0,
+                    297.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1322.0,
+                    1408.0,
+                    1322.0,
+                    1408.0,
+                    1368.0,
+                    291.0,
+                    1368.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    1359.0,
+                    1346.0,
+                    1359.0,
+                    1346.0,
+                    1395.0,
+                    322.0,
+                    1395.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1408.0,
+                    1405.0,
+                    1408.0,
+                    1405.0,
+                    1449.0,
+                    292.0,
+                    1449.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    1443.0,
+                    1407.0,
+                    1443.0,
+                    1407.0,
+                    1477.0,
+                    323.0,
+                    1477.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    320.0,
+                    1470.0,
+                    398.0,
+                    1470.0,
+                    398.0,
+                    1507.0,
+                    320.0,
+                    1507.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1184.0,
+                    533.0,
+                    1184.0,
+                    533.0,
+                    1214.0,
+                    296.0,
+                    1214.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    605.0,
+                    1182.0,
+                    1228.0,
+                    1182.0,
+                    1228.0,
+                    1218.0,
+                    605.0,
+                    1218.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1295.0,
+                    1179.0,
+                    1408.0,
+                    1179.0,
+                    1408.0,
+                    1217.0,
+                    1295.0,
+                    1217.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    321.0,
+                    1212.0,
+                    1267.0,
+                    1212.0,
+                    1267.0,
+                    1249.0,
+                    321.0,
+                    1249.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1337.0,
+                    1214.0,
+                    1410.0,
+                    1214.0,
+                    1410.0,
+                    1247.0,
+                    1337.0,
+                    1247.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    1244.0,
+                    1283.0,
+                    1244.0,
+                    1283.0,
+                    1281.0,
+                    322.0,
+                    1281.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    1275.0,
+                    1078.0,
+                    1275.0,
+                    1078.0,
+                    1311.0,
+                    323.0,
+                    1311.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    298.0,
+                    1100.0,
+                    1405.0,
+                    1100.0,
+                    1405.0,
+                    1136.0,
+                    298.0,
+                    1136.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    1131.0,
+                    803.0,
+                    1131.0,
+                    803.0,
+                    1165.0,
+                    323.0,
+                    1165.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1689.0,
+                    1406.0,
+                    1689.0,
+                    1406.0,
+                    1729.0,
+                    294.0,
+                    1729.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    318.0,
+                    1722.0,
+                    976.0,
+                    1722.0,
+                    976.0,
+                    1756.0,
+                    318.0,
+                    1756.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1604.0,
+                    1409.0,
+                    1604.0,
+                    1409.0,
+                    1647.0,
+                    293.0,
+                    1647.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    321.0,
+                    1639.0,
+                    652.0,
+                    1639.0,
+                    652.0,
+                    1672.0,
+                    321.0,
+                    1672.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    902.0,
+                    1406.0,
+                    902.0,
+                    1406.0,
+                    941.0,
+                    291.0,
+                    941.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    320.0,
+                    934.0,
+                    1406.0,
+                    934.0,
+                    1406.0,
+                    970.0,
+                    320.0,
+                    970.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    965.0,
+                    459.0,
+                    965.0,
+                    459.0,
+                    996.0,
+                    323.0,
+                    996.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1524.0,
+                    1404.0,
+                    1524.0,
+                    1404.0,
+                    1560.0,
+                    295.0,
+                    1560.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    1557.0,
+                    1408.0,
+                    1557.0,
+                    1408.0,
+                    1590.0,
+                    322.0,
+                    1590.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1017.0,
+                    1405.0,
+                    1017.0,
+                    1405.0,
+                    1053.0,
+                    296.0,
+                    1053.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    324.0,
+                    1049.0,
+                    930.0,
+                    1049.0,
+                    930.0,
+                    1081.0,
+                    324.0,
+                    1081.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    227.0,
+                    1405.0,
+                    227.0,
+                    1405.0,
+                    266.0,
+                    294.0,
+                    266.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    325.0,
+                    261.0,
+                    1143.0,
+                    261.0,
+                    1143.0,
+                    297.0,
+                    325.0,
+                    297.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    310.0,
+                    1405.0,
+                    310.0,
+                    1405.0,
+                    348.0,
+                    294.0,
+                    348.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    321.0,
+                    343.0,
+                    614.0,
+                    343.0,
+                    614.0,
+                    378.0,
+                    321.0,
+                    378.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    702.0,
+                    1404.0,
+                    702.0,
+                    1404.0,
+                    745.0,
+                    293.0,
+                    745.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    737.0,
+                    1050.0,
+                    737.0,
+                    1050.0,
+                    772.0,
+                    322.0,
+                    772.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1970.0,
+                    1401.0,
+                    1970.0,
+                    1401.0,
+                    2008.0,
+                    293.0,
+                    2008.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    2002.0,
+                    1214.0,
+                    2002.0,
+                    1214.0,
+                    2036.0,
+                    322.0,
+                    2036.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    475.0,
+                    1410.0,
+                    475.0,
+                    1410.0,
+                    519.0,
+                    292.0,
+                    519.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    320.0,
+                    510.0,
+                    1110.0,
+                    510.0,
+                    1110.0,
+                    545.0,
+                    320.0,
+                    545.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    394.0,
+                    1406.0,
+                    394.0,
+                    1406.0,
+                    433.0,
+                    293.0,
+                    433.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    324.0,
+                    427.0,
+                    999.0,
+                    427.0,
+                    999.0,
+                    460.0,
+                    324.0,
+                    460.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1887.0,
+                    1399.0,
+                    1887.0,
+                    1399.0,
+                    1924.0,
+                    293.0,
+                    1924.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    325.0,
+                    1919.0,
+                    1097.0,
+                    1919.0,
+                    1097.0,
+                    1955.0,
+                    325.0,
+                    1955.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1768.0,
+                    1407.0,
+                    1768.0,
+                    1407.0,
+                    1815.0,
+                    292.0,
+                    1815.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    1806.0,
+                    1404.0,
+                    1806.0,
+                    1404.0,
+                    1840.0,
+                    322.0,
+                    1840.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    319.0,
+                    1838.0,
+                    498.0,
+                    1838.0,
+                    498.0,
+                    1869.0,
+                    319.0,
+                    1869.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    563.0,
+                    1410.0,
+                    563.0,
+                    1410.0,
+                    599.0,
+                    294.0,
+                    599.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    321.0,
+                    593.0,
+                    1409.0,
+                    593.0,
+                    1409.0,
+                    629.0,
+                    321.0,
+                    629.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    625.0,
+                    1408.0,
+                    625.0,
+                    1408.0,
+                    657.0,
+                    322.0,
+                    657.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    656.0,
+                    1088.0,
+                    656.0,
+                    1088.0,
+                    689.0,
+                    323.0,
+                    689.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    786.0,
+                    1408.0,
+                    786.0,
+                    1408.0,
+                    827.0,
+                    293.0,
+                    827.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    319.0,
+                    817.0,
+                    1407.0,
+                    817.0,
+                    1407.0,
+                    860.0,
+                    319.0,
+                    860.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    849.0,
+                    805.0,
+                    849.0,
+                    805.0,
+                    886.0,
+                    323.0,
+                    886.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 9,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 2,
+                "poly": [
+                    298,
+                    75,
+                    854,
+                    75,
+                    854,
+                    105,
+                    298,
+                    105
+                ],
+                "score": 0.893
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    292,
+                    229,
+                    1405,
+                    229,
+                    1405,
+                    294,
+                    292,
+                    294
+                ],
+                "score": 0.839
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    293,
+                    993,
+                    1404,
+                    993,
+                    1404,
+                    1057,
+                    293,
+                    1057
+                ],
+                "score": 0.835
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    296,
+                    1126,
+                    1399,
+                    1126,
+                    1399,
+                    1191,
+                    296,
+                    1191
+                ],
+                "score": 0.833
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1207,
+                    1403,
+                    1207,
+                    1403,
+                    1271,
+                    297,
+                    1271
+                ],
+                "score": 0.819
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1073,
+                    1380,
+                    1073,
+                    1380,
+                    1109,
+                    297,
+                    1109
+                ],
+                "score": 0.818
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    294,
+                    727,
+                    1406,
+                    727,
+                    1406,
+                    792,
+                    294,
+                    792
+                ],
+                "score": 0.81
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    835,
+                    2088,
+                    862,
+                    2088,
+                    862,
+                    2113,
+                    835,
+                    2113
+                ],
+                "score": 0.806
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1288,
+                    1403,
+                    1288,
+                    1403,
+                    1354,
+                    297,
+                    1354
+                ],
+                "score": 0.799
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    310,
+                    1403,
+                    310,
+                    1403,
+                    404,
+                    297,
+                    404
+                ],
+                "score": 0.798
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    295,
+                    1481,
+                    1400,
+                    1481,
+                    1400,
+                    1547,
+                    295,
+                    1547
+                ],
+                "score": 0.796
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    296,
+                    858,
+                    1403,
+                    858,
+                    1403,
+                    927,
+                    296,
+                    927
+                ],
+                "score": 0.794
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    301,
+                    808,
+                    1371,
+                    808,
+                    1371,
+                    845,
+                    301,
+                    845
+                ],
+                "score": 0.778
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    595,
+                    1403,
+                    595,
+                    1403,
+                    660,
+                    299,
+                    660
+                ],
+                "score": 0.77
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1614,
+                    1401,
+                    1614,
+                    1401,
+                    1679,
+                    297,
+                    1679
+                ],
+                "score": 0.767
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    301,
+                    941,
+                    1348,
+                    941,
+                    1348,
+                    977,
+                    301,
+                    977
+                ],
+                "score": 0.767
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    294,
+                    1562,
+                    1396,
+                    1562,
+                    1396,
+                    1598,
+                    294,
+                    1598
+                ],
+                "score": 0.761
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    293,
+                    676,
+                    1404,
+                    676,
+                    1404,
+                    712,
+                    293,
+                    712
+                ],
+                "score": 0.753
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1695,
+                    1401,
+                    1695,
+                    1401,
+                    1760,
+                    297,
+                    1760
+                ],
+                "score": 0.725
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    296,
+                    1369,
+                    1407,
+                    1369,
+                    1407,
+                    1464,
+                    296,
+                    1464
+                ],
+                "score": 0.698
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1971,
+                    1400,
+                    1971,
+                    1400,
+                    2036,
+                    297,
+                    2036
+                ],
+                "score": 0.688
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1777,
+                    1402,
+                    1777,
+                    1402,
+                    1842,
+                    297,
+                    1842
+                ],
+                "score": 0.641
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    423,
+                    1405,
+                    423,
+                    1405,
+                    578,
+                    297,
+                    578
+                ],
+                "score": 0.625
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1859,
+                    1403,
+                    1859,
+                    1403,
+                    1954,
+                    298,
+                    1954
+                ],
+                "score": 0.571
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    73.0,
+                    857.0,
+                    73.0,
+                    857.0,
+                    108.0,
+                    297.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    831.0,
+                    2085.0,
+                    867.0,
+                    2085.0,
+                    867.0,
+                    2125.0,
+                    831.0,
+                    2125.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    227.0,
+                    1404.0,
+                    227.0,
+                    1404.0,
+                    265.0,
+                    294.0,
+                    265.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    260.0,
+                    1410.0,
+                    260.0,
+                    1410.0,
+                    296.0,
+                    323.0,
+                    296.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    993.0,
+                    1406.0,
+                    993.0,
+                    1406.0,
+                    1029.0,
+                    295.0,
+                    1029.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    324.0,
+                    1025.0,
+                    880.0,
+                    1025.0,
+                    880.0,
+                    1057.0,
+                    324.0,
+                    1057.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1124.0,
+                    1404.0,
+                    1124.0,
+                    1404.0,
+                    1162.0,
+                    294.0,
+                    1162.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    325.0,
+                    1157.0,
+                    1296.0,
+                    1157.0,
+                    1296.0,
+                    1193.0,
+                    325.0,
+                    1193.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1206.0,
+                    1407.0,
+                    1206.0,
+                    1407.0,
+                    1245.0,
+                    293.0,
+                    1245.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    321.0,
+                    1236.0,
+                    763.0,
+                    1236.0,
+                    763.0,
+                    1272.0,
+                    321.0,
+                    1272.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1070.0,
+                    1381.0,
+                    1070.0,
+                    1381.0,
+                    1112.0,
+                    293.0,
+                    1112.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    722.0,
+                    1409.0,
+                    722.0,
+                    1409.0,
+                    769.0,
+                    292.0,
+                    769.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    761.0,
+                    562.0,
+                    761.0,
+                    562.0,
+                    790.0,
+                    322.0,
+                    790.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1287.0,
+                    1407.0,
+                    1287.0,
+                    1407.0,
+                    1327.0,
+                    296.0,
+                    1327.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    1320.0,
+                    763.0,
+                    1320.0,
+                    763.0,
+                    1356.0,
+                    322.0,
+                    1356.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    308.0,
+                    1403.0,
+                    308.0,
+                    1403.0,
+                    347.0,
+                    295.0,
+                    347.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    324.0,
+                    342.0,
+                    1407.0,
+                    342.0,
+                    1407.0,
+                    376.0,
+                    324.0,
+                    376.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    321.0,
+                    372.0,
+                    486.0,
+                    372.0,
+                    486.0,
+                    407.0,
+                    321.0,
+                    407.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1482.0,
+                    1404.0,
+                    1482.0,
+                    1404.0,
+                    1518.0,
+                    295.0,
+                    1518.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    1513.0,
+                    1111.0,
+                    1513.0,
+                    1111.0,
+                    1546.0,
+                    322.0,
+                    1546.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    858.0,
+                    1404.0,
+                    858.0,
+                    1404.0,
+                    896.0,
+                    295.0,
+                    896.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    893.0,
+                    1235.0,
+                    893.0,
+                    1235.0,
+                    927.0,
+                    323.0,
+                    927.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    805.0,
+                    1370.0,
+                    805.0,
+                    1370.0,
+                    849.0,
+                    293.0,
+                    849.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    591.0,
+                    1408.0,
+                    591.0,
+                    1408.0,
+                    634.0,
+                    293.0,
+                    634.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    320.0,
+                    625.0,
+                    396.0,
+                    625.0,
+                    396.0,
+                    661.0,
+                    320.0,
+                    661.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1613.0,
+                    1405.0,
+                    1613.0,
+                    1405.0,
+                    1653.0,
+                    293.0,
+                    1653.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    1645.0,
+                    952.0,
+                    1645.0,
+                    952.0,
+                    1680.0,
+                    322.0,
+                    1680.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    941.0,
+                    1348.0,
+                    941.0,
+                    1348.0,
+                    977.0,
+                    294.0,
+                    977.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1559.0,
+                    1396.0,
+                    1559.0,
+                    1396.0,
+                    1600.0,
+                    294.0,
+                    1600.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    676.0,
+                    1407.0,
+                    676.0,
+                    1407.0,
+                    715.0,
+                    296.0,
+                    715.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1695.0,
+                    1405.0,
+                    1695.0,
+                    1405.0,
+                    1731.0,
+                    296.0,
+                    1731.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    324.0,
+                    1726.0,
+                    655.0,
+                    1726.0,
+                    655.0,
+                    1761.0,
+                    324.0,
+                    1761.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1369.0,
+                    1408.0,
+                    1369.0,
+                    1408.0,
+                    1407.0,
+                    296.0,
+                    1407.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    1402.0,
+                    1409.0,
+                    1402.0,
+                    1409.0,
+                    1436.0,
+                    323.0,
+                    1436.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    1431.0,
+                    496.0,
+                    1431.0,
+                    496.0,
+                    1466.0,
+                    322.0,
+                    1466.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1971.0,
+                    1404.0,
+                    1971.0,
+                    1404.0,
+                    2007.0,
+                    295.0,
+                    2007.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    2001.0,
+                    902.0,
+                    2001.0,
+                    902.0,
+                    2036.0,
+                    322.0,
+                    2036.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    1780.0,
+                    1406.0,
+                    1780.0,
+                    1406.0,
+                    1812.0,
+                    297.0,
+                    1812.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    324.0,
+                    1809.0,
+                    1043.0,
+                    1809.0,
+                    1043.0,
+                    1842.0,
+                    324.0,
+                    1842.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    422.0,
+                    1405.0,
+                    422.0,
+                    1405.0,
+                    462.0,
+                    294.0,
+                    462.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    453.0,
+                    1403.0,
+                    453.0,
+                    1403.0,
+                    490.0,
+                    323.0,
+                    490.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    321.0,
+                    483.0,
+                    1408.0,
+                    483.0,
+                    1408.0,
+                    520.0,
+                    321.0,
+                    520.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    515.0,
+                    1407.0,
+                    515.0,
+                    1407.0,
+                    548.0,
+                    323.0,
+                    548.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    320.0,
+                    544.0,
+                    1038.0,
+                    544.0,
+                    1038.0,
+                    582.0,
+                    320.0,
+                    582.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1857.0,
+                    1408.0,
+                    1857.0,
+                    1408.0,
+                    1896.0,
+                    294.0,
+                    1896.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    1889.0,
+                    1345.0,
+                    1889.0,
+                    1345.0,
+                    1926.0,
+                    322.0,
+                    1926.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    1920.0,
+                    1005.0,
+                    1920.0,
+                    1005.0,
+                    1958.0,
+                    323.0,
+                    1958.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 10,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1426,
+                    1404,
+                    1426,
+                    1404,
+                    1732,
+                    298,
+                    1732
+                ],
+                "score": 0.983
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    968,
+                    1404,
+                    968,
+                    1404,
+                    1182,
+                    299,
+                    1182
+                ],
+                "score": 0.98
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    388,
+                    1404,
+                    388,
+                    1404,
+                    604,
+                    298,
+                    604
+                ],
+                "score": 0.98
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1197,
+                    1404,
+                    1197,
+                    1404,
+                    1412,
+                    298,
+                    1412
+                ],
+                "score": 0.979
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1849,
+                    1405,
+                    1849,
+                    1405,
+                    2034,
+                    298,
+                    2034
+                ],
+                "score": 0.979
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    799,
+                    1404,
+                    799,
+                    1404,
+                    952,
+                    299,
+                    952
+                ],
+                "score": 0.977
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    618,
+                    1402,
+                    618,
+                    1402,
+                    680,
+                    297,
+                    680
+                ],
+                "score": 0.946
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    298,
+                    311,
+                    963,
+                    311,
+                    963,
+                    345,
+                    298,
+                    345
+                ],
+                "score": 0.888
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    300,
+                    1786,
+                    566,
+                    1786,
+                    566,
+                    1818,
+                    300,
+                    1818
+                ],
+                "score": 0.887
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    299,
+                    735,
+                    507,
+                    735,
+                    507,
+                    765,
+                    299,
+                    765
+                ],
+                "score": 0.885
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    760,
+                    228,
+                    941,
+                    228,
+                    941,
+                    261,
+                    760,
+                    261
+                ],
+                "score": 0.881
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    298,
+                    75,
+                    854,
+                    75,
+                    854,
+                    104,
+                    298,
+                    104
+                ],
+                "score": 0.879
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    836,
+                    2088,
+                    863,
+                    2088,
+                    863,
+                    2112,
+                    836,
+                    2112
+                ],
+                "score": 0.839
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    306.0,
+                    967.0,
+                    306.0,
+                    967.0,
+                    352.0,
+                    292.0,
+                    352.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1787.0,
+                    570.0,
+                    1787.0,
+                    570.0,
+                    1820.0,
+                    296.0,
+                    1820.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    732.0,
+                    511.0,
+                    732.0,
+                    511.0,
+                    770.0,
+                    295.0,
+                    770.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    755.0,
+                    225.0,
+                    945.0,
+                    225.0,
+                    945.0,
+                    268.0,
+                    755.0,
+                    268.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    72.0,
+                    857.0,
+                    72.0,
+                    857.0,
+                    108.0,
+                    297.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    832.0,
+                    2085.0,
+                    868.0,
+                    2085.0,
+                    868.0,
+                    2124.0,
+                    832.0,
+                    2124.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1428.0,
+                    1405.0,
+                    1428.0,
+                    1405.0,
+                    1460.0,
+                    296.0,
+                    1460.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1460.0,
+                    1404.0,
+                    1460.0,
+                    1404.0,
+                    1491.0,
+                    294.0,
+                    1491.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1490.0,
+                    1405.0,
+                    1490.0,
+                    1405.0,
+                    1522.0,
+                    296.0,
+                    1522.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1517.0,
+                    1407.0,
+                    1517.0,
+                    1407.0,
+                    1554.0,
+                    292.0,
+                    1554.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1547.0,
+                    1405.0,
+                    1547.0,
+                    1405.0,
+                    1585.0,
+                    292.0,
+                    1585.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1579.0,
+                    1409.0,
+                    1579.0,
+                    1409.0,
+                    1617.0,
+                    292.0,
+                    1617.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1608.0,
+                    1408.0,
+                    1608.0,
+                    1408.0,
+                    1645.0,
+                    293.0,
+                    1645.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1642.0,
+                    1405.0,
+                    1642.0,
+                    1405.0,
+                    1674.0,
+                    294.0,
+                    1674.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1670.0,
+                    1406.0,
+                    1670.0,
+                    1406.0,
+                    1706.0,
+                    293.0,
+                    1706.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1705.0,
+                    409.0,
+                    1705.0,
+                    409.0,
+                    1734.0,
+                    294.0,
+                    1734.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    968.0,
+                    1403.0,
+                    968.0,
+                    1403.0,
+                    1003.0,
+                    295.0,
+                    1003.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    999.0,
+                    1404.0,
+                    999.0,
+                    1404.0,
+                    1030.0,
+                    297.0,
+                    1030.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1027.0,
+                    1406.0,
+                    1027.0,
+                    1406.0,
+                    1065.0,
+                    293.0,
+                    1065.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1060.0,
+                    1405.0,
+                    1060.0,
+                    1405.0,
+                    1095.0,
+                    295.0,
+                    1095.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1087.0,
+                    1405.0,
+                    1087.0,
+                    1405.0,
+                    1128.0,
+                    293.0,
+                    1128.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1118.0,
+                    1405.0,
+                    1118.0,
+                    1405.0,
+                    1158.0,
+                    292.0,
+                    1158.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1149.0,
+                    663.0,
+                    1149.0,
+                    663.0,
+                    1187.0,
+                    294.0,
+                    1187.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    389.0,
+                    1404.0,
+                    389.0,
+                    1404.0,
+                    423.0,
+                    294.0,
+                    423.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    421.0,
+                    1406.0,
+                    421.0,
+                    1406.0,
+                    455.0,
+                    294.0,
+                    455.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    450.0,
+                    1405.0,
+                    450.0,
+                    1405.0,
+                    485.0,
+                    293.0,
+                    485.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    482.0,
+                    1405.0,
+                    482.0,
+                    1405.0,
+                    513.0,
+                    294.0,
+                    513.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    511.0,
+                    1406.0,
+                    511.0,
+                    1406.0,
+                    544.0,
+                    293.0,
+                    544.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    543.0,
+                    1405.0,
+                    543.0,
+                    1405.0,
+                    577.0,
+                    293.0,
+                    577.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    574.0,
+                    527.0,
+                    574.0,
+                    527.0,
+                    606.0,
+                    296.0,
+                    606.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1195.0,
+                    1405.0,
+                    1195.0,
+                    1405.0,
+                    1231.0,
+                    293.0,
+                    1231.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1227.0,
+                    1406.0,
+                    1227.0,
+                    1406.0,
+                    1264.0,
+                    293.0,
+                    1264.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1256.0,
+                    1406.0,
+                    1256.0,
+                    1406.0,
+                    1293.0,
+                    293.0,
+                    1293.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1286.0,
+                    1405.0,
+                    1286.0,
+                    1405.0,
+                    1327.0,
+                    292.0,
+                    1327.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1319.0,
+                    1405.0,
+                    1319.0,
+                    1405.0,
+                    1354.0,
+                    294.0,
+                    1354.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1350.0,
+                    1404.0,
+                    1350.0,
+                    1404.0,
+                    1384.0,
+                    294.0,
+                    1384.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1378.0,
+                    835.0,
+                    1378.0,
+                    835.0,
+                    1417.0,
+                    295.0,
+                    1417.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1850.0,
+                    1405.0,
+                    1850.0,
+                    1405.0,
+                    1886.0,
+                    293.0,
+                    1886.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1879.0,
+                    1407.0,
+                    1879.0,
+                    1407.0,
+                    1917.0,
+                    293.0,
+                    1917.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1911.0,
+                    1407.0,
+                    1911.0,
+                    1407.0,
+                    1947.0,
+                    295.0,
+                    1947.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1944.0,
+                    1402.0,
+                    1944.0,
+                    1402.0,
+                    1976.0,
+                    295.0,
+                    1976.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1973.0,
+                    1403.0,
+                    1973.0,
+                    1403.0,
+                    2005.0,
+                    296.0,
+                    2005.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    2003.0,
+                    1232.0,
+                    2003.0,
+                    1232.0,
+                    2037.0,
+                    295.0,
+                    2037.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    796.0,
+                    1408.0,
+                    796.0,
+                    1408.0,
+                    835.0,
+                    293.0,
+                    835.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    825.0,
+                    1406.0,
+                    825.0,
+                    1406.0,
+                    869.0,
+                    293.0,
+                    869.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    860.0,
+                    1409.0,
+                    860.0,
+                    1409.0,
+                    897.0,
+                    293.0,
+                    897.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    889.0,
+                    1406.0,
+                    889.0,
+                    1406.0,
+                    927.0,
+                    293.0,
+                    927.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    920.0,
+                    1072.0,
+                    920.0,
+                    1072.0,
+                    957.0,
+                    294.0,
+                    957.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    616.0,
+                    1406.0,
+                    616.0,
+                    1406.0,
+                    654.0,
+                    293.0,
+                    654.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    649.0,
+                    502.0,
+                    649.0,
+                    502.0,
+                    682.0,
+                    296.0,
+                    682.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 11,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    229,
+                    1404,
+                    229,
+                    1404,
+                    474,
+                    298,
+                    474
+                ],
+                "score": 0.98
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    689,
+                    1403,
+                    689,
+                    1403,
+                    903,
+                    298,
+                    903
+                ],
+                "score": 0.98
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    489,
+                    1404,
+                    489,
+                    1404,
+                    673,
+                    298,
+                    673
+                ],
+                "score": 0.979
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    919,
+                    1404,
+                    919,
+                    1404,
+                    1073,
+                    299,
+                    1073
+                ],
+                "score": 0.977
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    300,
+                    76,
+                    852,
+                    76,
+                    852,
+                    104,
+                    300,
+                    104
+                ],
+                "score": 0.839
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    836,
+                    2088,
+                    863,
+                    2088,
+                    863,
+                    2112,
+                    836,
+                    2112
+                ],
+                "score": 0.836
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    73.0,
+                    856.0,
+                    73.0,
+                    856.0,
+                    108.0,
+                    297.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    832.0,
+                    2084.0,
+                    867.0,
+                    2084.0,
+                    867.0,
+                    2123.0,
+                    832.0,
+                    2123.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    229.0,
+                    1405.0,
+                    229.0,
+                    1405.0,
+                    266.0,
+                    293.0,
+                    266.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    260.0,
+                    1405.0,
+                    260.0,
+                    1405.0,
+                    295.0,
+                    293.0,
+                    295.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    293.0,
+                    1408.0,
+                    293.0,
+                    1408.0,
+                    326.0,
+                    293.0,
+                    326.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    320.0,
+                    1405.0,
+                    320.0,
+                    1405.0,
+                    356.0,
+                    295.0,
+                    356.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    353.0,
+                    1405.0,
+                    353.0,
+                    1405.0,
+                    389.0,
+                    293.0,
+                    389.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    380.0,
+                    1406.0,
+                    380.0,
+                    1406.0,
+                    419.0,
+                    293.0,
+                    419.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    413.0,
+                    1406.0,
+                    413.0,
+                    1406.0,
+                    450.0,
+                    293.0,
+                    450.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    444.0,
+                    552.0,
+                    444.0,
+                    552.0,
+                    476.0,
+                    294.0,
+                    476.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    689.0,
+                    1404.0,
+                    689.0,
+                    1404.0,
+                    724.0,
+                    294.0,
+                    724.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    719.0,
+                    1407.0,
+                    719.0,
+                    1407.0,
+                    756.0,
+                    293.0,
+                    756.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    751.0,
+                    1404.0,
+                    751.0,
+                    1404.0,
+                    785.0,
+                    294.0,
+                    785.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    784.0,
+                    1404.0,
+                    784.0,
+                    1404.0,
+                    814.0,
+                    294.0,
+                    814.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    813.0,
+                    1404.0,
+                    813.0,
+                    1404.0,
+                    844.0,
+                    296.0,
+                    844.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    843.0,
+                    1405.0,
+                    843.0,
+                    1405.0,
+                    877.0,
+                    293.0,
+                    877.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    875.0,
+                    1214.0,
+                    875.0,
+                    1214.0,
+                    906.0,
+                    293.0,
+                    906.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    488.0,
+                    1405.0,
+                    488.0,
+                    1405.0,
+                    525.0,
+                    293.0,
+                    525.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    521.0,
+                    1405.0,
+                    521.0,
+                    1405.0,
+                    557.0,
+                    294.0,
+                    557.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    552.0,
+                    1405.0,
+                    552.0,
+                    1405.0,
+                    586.0,
+                    293.0,
+                    586.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    582.0,
+                    1404.0,
+                    582.0,
+                    1404.0,
+                    616.0,
+                    293.0,
+                    616.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    613.0,
+                    1405.0,
+                    613.0,
+                    1405.0,
+                    648.0,
+                    294.0,
+                    648.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    643.0,
+                    674.0,
+                    643.0,
+                    674.0,
+                    679.0,
+                    296.0,
+                    679.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    919.0,
+                    1406.0,
+                    919.0,
+                    1406.0,
+                    952.0,
+                    295.0,
+                    952.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    950.0,
+                    1406.0,
+                    950.0,
+                    1406.0,
+                    985.0,
+                    293.0,
+                    985.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    977.0,
+                    1406.0,
+                    977.0,
+                    1406.0,
+                    1015.0,
+                    294.0,
+                    1015.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1008.0,
+                    1406.0,
+                    1008.0,
+                    1406.0,
+                    1048.0,
+                    293.0,
+                    1048.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1038.0,
+                    934.0,
+                    1038.0,
+                    934.0,
+                    1079.0,
+                    293.0,
+                    1079.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 12,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 3,
+                "poly": [
+                    397,
+                    321,
+                    1302,
+                    321,
+                    1302,
+                    1716,
+                    397,
+                    1716
+                ],
+                "score": 0.97
+            },
+            {
+                "category_id": 4,
+                "poly": [
+                    295,
+                    1749,
+                    1407,
+                    1749,
+                    1407,
+                    1964,
+                    295,
+                    1964
+                ],
+                "score": 0.964
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    835,
+                    2088,
+                    863,
+                    2088,
+                    863,
+                    2112,
+                    835,
+                    2112
+                ],
+                "score": 0.853
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    299,
+                    76,
+                    852,
+                    76,
+                    852,
+                    104,
+                    299,
+                    104
+                ],
+                "score": 0.748
+            },
+            {
+                "category_id": 4,
+                "poly": [
+                    674,
+                    294,
+                    1029,
+                    294,
+                    1029,
+                    323,
+                    674,
+                    323
+                ],
+                "score": 0.41
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    944,
+                    1782,
+                    1020,
+                    1782,
+                    1020,
+                    1813,
+                    944,
+                    1813
+                ],
+                "score": 0.87,
+                "latex": "( \\gamma = 0 )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    896,
+                    1217,
+                    1019,
+                    1217,
+                    1019,
+                    1248,
+                    896,
+                    1248
+                ],
+                "score": 0.85,
+                "latex": "\\beta = - 0 . 5 )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    920,
+                    754,
+                    1016,
+                    754,
+                    1016,
+                    787,
+                    920,
+                    787
+                ],
+                "score": 0.85,
+                "latex": "\\beta = 0 . 0"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    922,
+                    292,
+                    1018,
+                    292,
+                    1018,
+                    324,
+                    922,
+                    324
+                ],
+                "score": 0.84,
+                "latex": "\\beta = 0 . 5"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1166,
+                    1873,
+                    1207,
+                    1873,
+                    1207,
+                    1902,
+                    1166,
+                    1902
+                ],
+                "score": 0.7,
+                "latex": "( T )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    885,
+                    800,
+                    939,
+                    800,
+                    939,
+                    822,
+                    885,
+                    822
+                ],
+                "score": 0.49,
+                "latex": "{ \\tt = } 1 0 0"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    884,
+                    1260,
+                    937,
+                    1260,
+                    937,
+                    1282,
+                    884,
+                    1282
+                ],
+                "score": 0.49,
+                "latex": "= 1 0 0"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    614,
+                    799,
+                    656,
+                    799,
+                    656,
+                    822,
+                    614,
+                    822
+                ],
+                "score": 0.37,
+                "latex": "{ \\tt = } 1 0"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    734,
+                    1682,
+                    773,
+                    1682,
+                    773,
+                    1703,
+                    734,
+                    1703
+                ],
+                "score": 0.37,
+                "latex": "= 1"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    968,
+                    1681,
+                    1019,
+                    1681,
+                    1019,
+                    1704,
+                    968,
+                    1704
+                ],
+                "score": 0.34,
+                "latex": "= 1 0"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    884,
+                    337,
+                    938,
+                    337,
+                    938,
+                    358,
+                    884,
+                    358
+                ],
+                "score": 0.33,
+                "latex": "{ \\tt = } 1 0 0"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    531.0,
+                    333.0,
+                    661.0,
+                    333.0,
+                    661.0,
+                    364.0,
+                    531.0,
+                    364.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    801.0,
+                    332.0,
+                    883.0,
+                    332.0,
+                    883.0,
+                    366.0,
+                    801.0,
+                    366.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    939.0,
+                    332.0,
+                    944.0,
+                    332.0,
+                    944.0,
+                    366.0,
+                    939.0,
+                    366.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1072.0,
+                    332.0,
+                    1228.0,
+                    332.0,
+                    1228.0,
+                    366.0,
+                    1072.0,
+                    366.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    421.0,
+                    350.0,
+                    469.0,
+                    350.0,
+                    469.0,
+                    378.0,
+                    421.0,
+                    378.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    399.0,
+                    386.0,
+                    469.0,
+                    386.0,
+                    469.0,
+                    486.0,
+                    399.0,
+                    486.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1275.0,
+                    363.0,
+                    1302.0,
+                    363.0,
+                    1302.0,
+                    509.0,
+                    1275.0,
+                    509.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    416.0,
+                    459.0,
+                    468.0,
+                    459.0,
+                    468.0,
+                    486.0,
+                    416.0,
+                    486.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    421.0,
+                    494.0,
+                    469.0,
+                    494.0,
+                    469.0,
+                    522.0,
+                    421.0,
+                    522.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    421.0,
+                    526.0,
+                    468.0,
+                    526.0,
+                    468.0,
+                    554.0,
+                    421.0,
+                    554.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    399.0,
+                    562.0,
+                    469.0,
+                    562.0,
+                    469.0,
+                    663.0,
+                    399.0,
+                    663.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1275.0,
+                    551.0,
+                    1302.0,
+                    551.0,
+                    1302.0,
+                    671.0,
+                    1275.0,
+                    671.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    418.0,
+                    635.0,
+                    468.0,
+                    635.0,
+                    468.0,
+                    662.0,
+                    418.0,
+                    662.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    421.0,
+                    670.0,
+                    469.0,
+                    670.0,
+                    469.0,
+                    696.0,
+                    421.0,
+                    696.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    474.0,
+                    688.0,
+                    495.0,
+                    688.0,
+                    495.0,
+                    710.0,
+                    474.0,
+                    710.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    518.0,
+                    685.0,
+                    564.0,
+                    685.0,
+                    564.0,
+                    713.0,
+                    518.0,
+                    713.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    570.0,
+                    687.0,
+                    619.0,
+                    687.0,
+                    619.0,
+                    713.0,
+                    570.0,
+                    713.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    630.0,
+                    685.0,
+                    735.0,
+                    685.0,
+                    735.0,
+                    713.0,
+                    630.0,
+                    713.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    749.0,
+                    690.0,
+                    771.0,
+                    690.0,
+                    771.0,
+                    710.0,
+                    749.0,
+                    710.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    793.0,
+                    684.0,
+                    1013.0,
+                    684.0,
+                    1013.0,
+                    715.0,
+                    793.0,
+                    715.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1027.0,
+                    688.0,
+                    1047.0,
+                    688.0,
+                    1047.0,
+                    710.0,
+                    1027.0,
+                    710.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1071.0,
+                    685.0,
+                    1118.0,
+                    685.0,
+                    1118.0,
+                    713.0,
+                    1071.0,
+                    713.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1126.0,
+                    685.0,
+                    1173.0,
+                    685.0,
+                    1173.0,
+                    713.0,
+                    1126.0,
+                    713.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1184.0,
+                    685.0,
+                    1230.0,
+                    685.0,
+                    1230.0,
+                    713.0,
+                    1184.0,
+                    713.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1234.0,
+                    687.0,
+                    1288.0,
+                    687.0,
+                    1288.0,
+                    713.0,
+                    1234.0,
+                    713.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    552.0,
+                    707.0,
+                    638.0,
+                    707.0,
+                    638.0,
+                    737.0,
+                    552.0,
+                    737.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    829.0,
+                    710.0,
+                    914.0,
+                    710.0,
+                    914.0,
+                    737.0,
+                    829.0,
+                    737.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1106.0,
+                    705.0,
+                    1192.0,
+                    705.0,
+                    1192.0,
+                    739.0,
+                    1106.0,
+                    739.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    672.0,
+                    751.0,
+                    919.0,
+                    751.0,
+                    919.0,
+                    790.0,
+                    672.0,
+                    790.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1017.0,
+                    751.0,
+                    1030.0,
+                    751.0,
+                    1030.0,
+                    790.0,
+                    1017.0,
+                    790.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    532.0,
+                    796.0,
+                    613.0,
+                    796.0,
+                    613.0,
+                    825.0,
+                    532.0,
+                    825.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    801.0,
+                    794.0,
+                    884.0,
+                    794.0,
+                    884.0,
+                    829.0,
+                    801.0,
+                    829.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    940.0,
+                    794.0,
+                    944.0,
+                    794.0,
+                    944.0,
+                    829.0,
+                    940.0,
+                    829.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1072.0,
+                    798.0,
+                    1224.0,
+                    798.0,
+                    1224.0,
+                    824.0,
+                    1072.0,
+                    824.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    397.0,
+                    814.0,
+                    469.0,
+                    814.0,
+                    469.0,
+                    950.0,
+                    397.0,
+                    950.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1274.0,
+                    825.0,
+                    1304.0,
+                    825.0,
+                    1304.0,
+                    972.0,
+                    1274.0,
+                    972.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    419.0,
+                    920.0,
+                    468.0,
+                    920.0,
+                    468.0,
+                    948.0,
+                    419.0,
+                    948.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    421.0,
+                    958.0,
+                    469.0,
+                    958.0,
+                    469.0,
+                    984.0,
+                    421.0,
+                    984.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    421.0,
+                    990.0,
+                    468.0,
+                    990.0,
+                    468.0,
+                    1017.0,
+                    421.0,
+                    1017.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    397.0,
+                    1025.0,
+                    469.0,
+                    1025.0,
+                    469.0,
+                    1126.0,
+                    397.0,
+                    1126.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1274.0,
+                    1014.0,
+                    1304.0,
+                    1014.0,
+                    1304.0,
+                    1136.0,
+                    1274.0,
+                    1136.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    419.0,
+                    1096.0,
+                    468.0,
+                    1096.0,
+                    468.0,
+                    1123.0,
+                    419.0,
+                    1123.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    421.0,
+                    1132.0,
+                    471.0,
+                    1132.0,
+                    471.0,
+                    1160.0,
+                    421.0,
+                    1160.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    473.0,
+                    1148.0,
+                    496.0,
+                    1148.0,
+                    496.0,
+                    1174.0,
+                    473.0,
+                    1174.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    517.0,
+                    1148.0,
+                    562.0,
+                    1148.0,
+                    562.0,
+                    1176.0,
+                    517.0,
+                    1176.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    573.0,
+                    1149.0,
+                    617.0,
+                    1149.0,
+                    617.0,
+                    1176.0,
+                    573.0,
+                    1176.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    631.0,
+                    1149.0,
+                    675.0,
+                    1149.0,
+                    675.0,
+                    1176.0,
+                    631.0,
+                    1176.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    680.0,
+                    1149.0,
+                    735.0,
+                    1149.0,
+                    735.0,
+                    1176.0,
+                    680.0,
+                    1176.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    751.0,
+                    1151.0,
+                    771.0,
+                    1151.0,
+                    771.0,
+                    1173.0,
+                    751.0,
+                    1173.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    795.0,
+                    1148.0,
+                    840.0,
+                    1148.0,
+                    840.0,
+                    1176.0,
+                    795.0,
+                    1176.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    851.0,
+                    1148.0,
+                    895.0,
+                    1148.0,
+                    895.0,
+                    1176.0,
+                    851.0,
+                    1176.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    911.0,
+                    1151.0,
+                    950.0,
+                    1151.0,
+                    950.0,
+                    1174.0,
+                    911.0,
+                    1174.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    958.0,
+                    1149.0,
+                    1011.0,
+                    1149.0,
+                    1011.0,
+                    1176.0,
+                    958.0,
+                    1176.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1029.0,
+                    1151.0,
+                    1049.0,
+                    1151.0,
+                    1049.0,
+                    1173.0,
+                    1029.0,
+                    1173.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1072.0,
+                    1149.0,
+                    1118.0,
+                    1149.0,
+                    1118.0,
+                    1176.0,
+                    1072.0,
+                    1176.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1127.0,
+                    1149.0,
+                    1173.0,
+                    1149.0,
+                    1173.0,
+                    1176.0,
+                    1127.0,
+                    1176.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1189.0,
+                    1151.0,
+                    1228.0,
+                    1151.0,
+                    1228.0,
+                    1174.0,
+                    1189.0,
+                    1174.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1234.0,
+                    1149.0,
+                    1288.0,
+                    1149.0,
+                    1288.0,
+                    1176.0,
+                    1234.0,
+                    1176.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    552.0,
+                    1169.0,
+                    638.0,
+                    1169.0,
+                    638.0,
+                    1200.0,
+                    552.0,
+                    1200.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    831.0,
+                    1173.0,
+                    915.0,
+                    1173.0,
+                    915.0,
+                    1199.0,
+                    831.0,
+                    1199.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1108.0,
+                    1171.0,
+                    1194.0,
+                    1171.0,
+                    1194.0,
+                    1201.0,
+                    1108.0,
+                    1201.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    675.0,
+                    1213.0,
+                    895.0,
+                    1213.0,
+                    895.0,
+                    1252.0,
+                    675.0,
+                    1252.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1020.0,
+                    1213.0,
+                    1027.0,
+                    1213.0,
+                    1027.0,
+                    1252.0,
+                    1020.0,
+                    1252.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    532.0,
+                    1257.0,
+                    659.0,
+                    1257.0,
+                    659.0,
+                    1286.0,
+                    532.0,
+                    1286.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    802.0,
+                    1257.0,
+                    883.0,
+                    1257.0,
+                    883.0,
+                    1286.0,
+                    802.0,
+                    1286.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    938.0,
+                    1257.0,
+                    942.0,
+                    1257.0,
+                    942.0,
+                    1286.0,
+                    938.0,
+                    1286.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1072.0,
+                    1259.0,
+                    1224.0,
+                    1259.0,
+                    1224.0,
+                    1285.0,
+                    1072.0,
+                    1285.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    421.0,
+                    1274.0,
+                    468.0,
+                    1274.0,
+                    468.0,
+                    1302.0,
+                    421.0,
+                    1302.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    396.0,
+                    1310.0,
+                    468.0,
+                    1310.0,
+                    468.0,
+                    1406.0,
+                    396.0,
+                    1406.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1275.0,
+                    1288.0,
+                    1300.0,
+                    1288.0,
+                    1300.0,
+                    1431.0,
+                    1275.0,
+                    1431.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    418.0,
+                    1381.0,
+                    468.0,
+                    1381.0,
+                    468.0,
+                    1408.0,
+                    418.0,
+                    1408.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    421.0,
+                    1416.0,
+                    468.0,
+                    1416.0,
+                    468.0,
+                    1444.0,
+                    421.0,
+                    1444.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    419.0,
+                    1448.0,
+                    468.0,
+                    1448.0,
+                    468.0,
+                    1476.0,
+                    419.0,
+                    1476.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    378.0,
+                    1469.0,
+                    487.0,
+                    1469.0,
+                    487.0,
+                    1601.0,
+                    378.0,
+                    1601.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1274.0,
+                    1471.0,
+                    1304.0,
+                    1471.0,
+                    1304.0,
+                    1595.0,
+                    1274.0,
+                    1595.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    421.0,
+                    1518.0,
+                    468.0,
+                    1518.0,
+                    468.0,
+                    1546.0,
+                    421.0,
+                    1546.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    482.0,
+                    1556.0,
+                    502.0,
+                    1556.0,
+                    502.0,
+                    1578.0,
+                    482.0,
+                    1578.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    765.0,
+                    1565.0,
+                    784.0,
+                    1565.0,
+                    784.0,
+                    1584.0,
+                    765.0,
+                    1584.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    421.0,
+                    1590.0,
+                    468.0,
+                    1590.0,
+                    468.0,
+                    1616.0,
+                    421.0,
+                    1616.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    473.0,
+                    1609.0,
+                    493.0,
+                    1609.0,
+                    493.0,
+                    1631.0,
+                    473.0,
+                    1631.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    517.0,
+                    1606.0,
+                    562.0,
+                    1606.0,
+                    562.0,
+                    1635.0,
+                    517.0,
+                    1635.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    573.0,
+                    1606.0,
+                    617.0,
+                    1606.0,
+                    617.0,
+                    1634.0,
+                    573.0,
+                    1634.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    631.0,
+                    1606.0,
+                    675.0,
+                    1606.0,
+                    675.0,
+                    1634.0,
+                    631.0,
+                    1634.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    681.0,
+                    1606.0,
+                    735.0,
+                    1606.0,
+                    735.0,
+                    1634.0,
+                    681.0,
+                    1634.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    751.0,
+                    1609.0,
+                    769.0,
+                    1609.0,
+                    769.0,
+                    1631.0,
+                    751.0,
+                    1631.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    795.0,
+                    1606.0,
+                    839.0,
+                    1606.0,
+                    839.0,
+                    1634.0,
+                    795.0,
+                    1634.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    849.0,
+                    1606.0,
+                    893.0,
+                    1606.0,
+                    893.0,
+                    1634.0,
+                    849.0,
+                    1634.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    909.0,
+                    1606.0,
+                    952.0,
+                    1606.0,
+                    952.0,
+                    1634.0,
+                    909.0,
+                    1634.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    959.0,
+                    1606.0,
+                    1013.0,
+                    1606.0,
+                    1013.0,
+                    1634.0,
+                    959.0,
+                    1634.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1027.0,
+                    1609.0,
+                    1047.0,
+                    1609.0,
+                    1047.0,
+                    1631.0,
+                    1027.0,
+                    1631.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1071.0,
+                    1606.0,
+                    1116.0,
+                    1606.0,
+                    1116.0,
+                    1634.0,
+                    1071.0,
+                    1634.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1127.0,
+                    1606.0,
+                    1170.0,
+                    1606.0,
+                    1170.0,
+                    1634.0,
+                    1127.0,
+                    1634.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1186.0,
+                    1606.0,
+                    1228.0,
+                    1606.0,
+                    1228.0,
+                    1634.0,
+                    1186.0,
+                    1634.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1237.0,
+                    1606.0,
+                    1289.0,
+                    1606.0,
+                    1289.0,
+                    1634.0,
+                    1237.0,
+                    1634.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    549.0,
+                    1624.0,
+                    640.0,
+                    1624.0,
+                    640.0,
+                    1659.0,
+                    549.0,
+                    1659.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    827.0,
+                    1624.0,
+                    916.0,
+                    1624.0,
+                    916.0,
+                    1659.0,
+                    827.0,
+                    1659.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1105.0,
+                    1624.0,
+                    1192.0,
+                    1624.0,
+                    1192.0,
+                    1659.0,
+                    1105.0,
+                    1659.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    465.0,
+                    1677.0,
+                    545.0,
+                    1677.0,
+                    545.0,
+                    1708.0,
+                    465.0,
+                    1708.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    617.0,
+                    1677.0,
+                    733.0,
+                    1677.0,
+                    733.0,
+                    1707.0,
+                    617.0,
+                    1707.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    774.0,
+                    1677.0,
+                    777.0,
+                    1677.0,
+                    777.0,
+                    1707.0,
+                    774.0,
+                    1707.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    851.0,
+                    1680.0,
+                    967.0,
+                    1680.0,
+                    967.0,
+                    1705.0,
+                    851.0,
+                    1705.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1096.0,
+                    1677.0,
+                    1281.0,
+                    1677.0,
+                    1281.0,
+                    1707.0,
+                    1096.0,
+                    1707.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1749.0,
+                    1406.0,
+                    1749.0,
+                    1406.0,
+                    1784.0,
+                    294.0,
+                    1784.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1779.0,
+                    943.0,
+                    1779.0,
+                    943.0,
+                    1816.0,
+                    293.0,
+                    1816.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1021.0,
+                    1779.0,
+                    1405.0,
+                    1779.0,
+                    1405.0,
+                    1816.0,
+                    1021.0,
+                    1816.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1808.0,
+                    1407.0,
+                    1808.0,
+                    1407.0,
+                    1849.0,
+                    293.0,
+                    1849.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1839.0,
+                    1405.0,
+                    1839.0,
+                    1405.0,
+                    1876.0,
+                    293.0,
+                    1876.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1871.0,
+                    1165.0,
+                    1871.0,
+                    1165.0,
+                    1906.0,
+                    294.0,
+                    1906.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1208.0,
+                    1871.0,
+                    1406.0,
+                    1871.0,
+                    1406.0,
+                    1906.0,
+                    1208.0,
+                    1906.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1898.0,
+                    1407.0,
+                    1898.0,
+                    1407.0,
+                    1939.0,
+                    292.0,
+                    1939.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1929.0,
+                    408.0,
+                    1929.0,
+                    408.0,
+                    1968.0,
+                    292.0,
+                    1968.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    831.0,
+                    2084.0,
+                    869.0,
+                    2084.0,
+                    869.0,
+                    2123.0,
+                    831.0,
+                    2123.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    73.0,
+                    856.0,
+                    73.0,
+                    856.0,
+                    108.0,
+                    297.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    673.0,
+                    290.0,
+                    921.0,
+                    290.0,
+                    921.0,
+                    329.0,
+                    673.0,
+                    329.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1019.0,
+                    290.0,
+                    1029.0,
+                    290.0,
+                    1029.0,
+                    329.0,
+                    1019.0,
+                    329.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 13,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1589,
+                    1406,
+                    1589,
+                    1406,
+                    1808,
+                    297,
+                    1808
+                ],
+                "score": 0.977
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1319,
+                    1405,
+                    1319,
+                    1405,
+                    1505,
+                    297,
+                    1505
+                ],
+                "score": 0.976
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    352,
+                    1406,
+                    352,
+                    1406,
+                    446,
+                    299,
+                    446
+                ],
+                "score": 0.97
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    300,
+                    679,
+                    1397,
+                    679,
+                    1397,
+                    804,
+                    300,
+                    804
+                ],
+                "score": 0.966
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    301,
+                    967,
+                    1401,
+                    967,
+                    1401,
+                    1061,
+                    301,
+                    1061
+                ],
+                "score": 0.958
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    295,
+                    889,
+                    1401,
+                    889,
+                    1401,
+                    954,
+                    295,
+                    954
+                ],
+                "score": 0.95
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    320,
+                    1086,
+                    1406,
+                    1086,
+                    1406,
+                    1294,
+                    320,
+                    1294
+                ],
+                "score": 0.943
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    295,
+                    460,
+                    1400,
+                    460,
+                    1400,
+                    528,
+                    295,
+                    528
+                ],
+                "score": 0.941
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    295,
+                    1971,
+                    1402,
+                    1971,
+                    1402,
+                    2035,
+                    295,
+                    2035
+                ],
+                "score": 0.937
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    302,
+                    837,
+                    898,
+                    837,
+                    898,
+                    868,
+                    302,
+                    868
+                ],
+                "score": 0.891
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    299,
+                    295,
+                    601,
+                    295,
+                    601,
+                    328,
+                    299,
+                    328
+                ],
+                "score": 0.861
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    836,
+                    2088,
+                    863,
+                    2088,
+                    863,
+                    2112,
+                    836,
+                    2112
+                ],
+                "score": 0.855
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    299,
+                    1537,
+                    1170,
+                    1537,
+                    1170,
+                    1570,
+                    299,
+                    1570
+                ],
+                "score": 0.835
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    297,
+                    228,
+                    1131,
+                    228,
+                    1131,
+                    262,
+                    297,
+                    262
+                ],
+                "score": 0.829
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    299,
+                    1919,
+                    814,
+                    1919,
+                    814,
+                    1951,
+                    299,
+                    1951
+                ],
+                "score": 0.828
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    344,
+                    541,
+                    1125,
+                    541,
+                    1125,
+                    668,
+                    344,
+                    668
+                ],
+                "score": 0.73
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    298,
+                    75,
+                    854,
+                    75,
+                    854,
+                    105,
+                    298,
+                    105
+                ],
+                "score": 0.561
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    344,
+                    541,
+                    1125,
+                    541,
+                    1125,
+                    668,
+                    344,
+                    668
+                ],
+                "score": 0.409
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    297,
+                    75,
+                    854,
+                    75,
+                    854,
+                    105,
+                    297,
+                    105
+                ],
+                "score": 0.352
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    588,
+                    1654,
+                    704,
+                    1654,
+                    704,
+                    1684,
+                    588,
+                    1684
+                ],
+                "score": 0.94,
+                "latex": "A _ { 1 } , . . . , A _ { 5 }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1137,
+                    1623,
+                    1212,
+                    1623,
+                    1212,
+                    1650,
+                    1137,
+                    1650
+                ],
+                "score": 0.93,
+                "latex": "T = 1"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    698,
+                    1744,
+                    846,
+                    1744,
+                    846,
+                    1778,
+                    698,
+                    1778
+                ],
+                "score": 0.93,
+                "latex": "r ( A _ { 2 } ) = 1 / 2"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    887,
+                    1683,
+                    1036,
+                    1683,
+                    1036,
+                    1717,
+                    887,
+                    1717
+                ],
+                "score": 0.93,
+                "latex": "r ( A _ { 1 } ) = 1 / 2"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1206,
+                    1653,
+                    1328,
+                    1653,
+                    1328,
+                    1685,
+                    1206,
+                    1685
+                ],
+                "score": 0.92,
+                "latex": "r ( A _ { i } ) = 0"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    905,
+                    1744,
+                    1074,
+                    1744,
+                    1074,
+                    1778,
+                    905,
+                    1778
+                ],
+                "score": 0.92,
+                "latex": "r ( A _ { 1 } ) = - 1 / 2"
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    341,
+                    538,
+                    1125,
+                    538,
+                    1125,
+                    672,
+                    341,
+                    672
+                ],
+                "score": 0.92,
+                "latex": "\\begin{array} { r l r } & { } & { s _ { t } , o _ { t } = a _ { t - 1 } , \\{ \\} } \\\\ & { } & { a _ { t } \\in \\{ \\mathsf { d e f e c t } , ~ \\mathsf { c o o p e r a t e } \\} } \\\\ & { } & { P ( s _ { t } , a _ { t } ) = \\delta ( a _ { t } ) } \\\\ & { } & { R ( s _ { t } , a _ { t } ) = I ( s _ { t } = \\mathsf { c o o p e r a t e } ) + \\beta ~ I ( a _ { t } = \\mathsf { c o o p e r a t e } ) - 1 / 2 } \\end{array}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    801,
+                    742,
+                    946,
+                    742,
+                    946,
+                    775,
+                    801,
+                    775
+                ],
+                "score": 0.92,
+                "latex": "\\theta \\sim \\mathcal { N } ( 0 , 1 ) \\big )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1001,
+                    355,
+                    1076,
+                    355,
+                    1076,
+                    385,
+                    1001,
+                    385
+                ],
+                "score": 0.91,
+                "latex": "\\gamma = 0"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    698,
+                    680,
+                    818,
+                    680,
+                    818,
+                    713,
+                    698,
+                    713
+                ],
+                "score": 0.91,
+                "latex": "\\beta = - 1 / 2"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    417,
+                    771,
+                    591,
+                    771,
+                    591,
+                    805,
+                    417,
+                    805
+                ],
+                "score": 0.9,
+                "latex": "P ( a _ { t } = \\mathrm { \\bar { d e f e c t } } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1132,
+                    1776,
+                    1205,
+                    1776,
+                    1205,
+                    1803,
+                    1132,
+                    1803
+                ],
+                "score": 0.9,
+                "latex": "T > 1"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1293,
+                    1746,
+                    1328,
+                    1746,
+                    1328,
+                    1775,
+                    1293,
+                    1775
+                ],
+                "score": 0.89,
+                "latex": "A _ { 2 }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    814,
+                    1321,
+                    943,
+                    1321,
+                    943,
+                    1353,
+                    814,
+                    1353
+                ],
+                "score": 0.89,
+                "latex": "( \\beta = - 1 / 2 )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1361,
+                    1746,
+                    1396,
+                    1746,
+                    1396,
+                    1775,
+                    1361,
+                    1775
+                ],
+                "score": 0.89,
+                "latex": "A _ { 1 }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    395,
+                    1684,
+                    432,
+                    1684,
+                    432,
+                    1714,
+                    395,
+                    1714
+                ],
+                "score": 0.88,
+                "latex": "A _ { 1 }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1071,
+                    1715,
+                    1107,
+                    1715,
+                    1107,
+                    1744,
+                    1071,
+                    1744
+                ],
+                "score": 0.88,
+                "latex": "A _ { 1 }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    326,
+                    1746,
+                    362,
+                    1746,
+                    362,
+                    1775,
+                    326,
+                    1775
+                ],
+                "score": 0.87,
+                "latex": "A _ { 2 }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    878,
+                    892,
+                    898,
+                    892,
+                    898,
+                    922,
+                    878,
+                    922
+                ],
+                "score": 0.83,
+                "latex": "\\beta"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    622,
+                    743,
+                    639,
+                    743,
+                    639,
+                    769,
+                    622,
+                    769
+                ],
+                "score": 0.8,
+                "latex": "\\theta"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    369,
+                    682,
+                    387,
+                    682,
+                    387,
+                    708,
+                    369,
+                    708
+                ],
+                "score": 0.78,
+                "latex": "I"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    697,
+                    463,
+                    744,
+                    463,
+                    744,
+                    490,
+                    697,
+                    490
+                ],
+                "score": 0.66,
+                "latex": "2 \\mathbf { x } 2"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1096,
+                    1539,
+                    1171,
+                    1539,
+                    1171,
+                    1567,
+                    1096,
+                    1567
+                ],
+                "score": 0.66,
+                "latex": "T = 1"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    653,
+                    712,
+                    855,
+                    712,
+                    855,
+                    743,
+                    653,
+                    743
+                ],
+                "score": 0.45,
+                "latex": "s _ { 0 } \\sim U ( { \\tt d e f e c t }"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    837.0,
+                    904.0,
+                    837.0,
+                    904.0,
+                    872.0,
+                    296.0,
+                    872.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    295.0,
+                    603.0,
+                    295.0,
+                    603.0,
+                    331.0,
+                    295.0,
+                    331.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    832.0,
+                    2085.0,
+                    867.0,
+                    2085.0,
+                    867.0,
+                    2124.0,
+                    832.0,
+                    2124.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1537.0,
+                    1095.0,
+                    1537.0,
+                    1095.0,
+                    1572.0,
+                    295.0,
+                    1572.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    290.0,
+                    223.0,
+                    1135.0,
+                    223.0,
+                    1135.0,
+                    268.0,
+                    290.0,
+                    268.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1918.0,
+                    818.0,
+                    1918.0,
+                    818.0,
+                    1955.0,
+                    296.0,
+                    1955.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    73.0,
+                    857.0,
+                    73.0,
+                    857.0,
+                    108.0,
+                    297.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    298.0,
+                    73.0,
+                    856.0,
+                    73.0,
+                    856.0,
+                    108.0,
+                    298.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1588.0,
+                    1408.0,
+                    1588.0,
+                    1408.0,
+                    1627.0,
+                    292.0,
+                    1627.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1622.0,
+                    1136.0,
+                    1622.0,
+                    1136.0,
+                    1657.0,
+                    294.0,
+                    1657.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1213.0,
+                    1622.0,
+                    1406.0,
+                    1622.0,
+                    1406.0,
+                    1657.0,
+                    1213.0,
+                    1657.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1651.0,
+                    587.0,
+                    1651.0,
+                    587.0,
+                    1690.0,
+                    294.0,
+                    1690.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    705.0,
+                    1651.0,
+                    1205.0,
+                    1651.0,
+                    1205.0,
+                    1690.0,
+                    705.0,
+                    1690.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1329.0,
+                    1651.0,
+                    1407.0,
+                    1651.0,
+                    1407.0,
+                    1690.0,
+                    1329.0,
+                    1690.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1682.0,
+                    394.0,
+                    1682.0,
+                    394.0,
+                    1718.0,
+                    291.0,
+                    1718.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    433.0,
+                    1682.0,
+                    886.0,
+                    1682.0,
+                    886.0,
+                    1718.0,
+                    433.0,
+                    1718.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1037.0,
+                    1682.0,
+                    1407.0,
+                    1682.0,
+                    1407.0,
+                    1718.0,
+                    1037.0,
+                    1718.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1711.0,
+                    1070.0,
+                    1711.0,
+                    1070.0,
+                    1750.0,
+                    294.0,
+                    1750.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1108.0,
+                    1711.0,
+                    1406.0,
+                    1711.0,
+                    1406.0,
+                    1750.0,
+                    1108.0,
+                    1750.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1742.0,
+                    325.0,
+                    1742.0,
+                    325.0,
+                    1781.0,
+                    294.0,
+                    1781.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    363.0,
+                    1742.0,
+                    697.0,
+                    1742.0,
+                    697.0,
+                    1781.0,
+                    363.0,
+                    1781.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    847.0,
+                    1742.0,
+                    904.0,
+                    1742.0,
+                    904.0,
+                    1781.0,
+                    847.0,
+                    1781.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1075.0,
+                    1742.0,
+                    1292.0,
+                    1742.0,
+                    1292.0,
+                    1781.0,
+                    1075.0,
+                    1781.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1329.0,
+                    1742.0,
+                    1360.0,
+                    1742.0,
+                    1360.0,
+                    1781.0,
+                    1329.0,
+                    1781.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1397.0,
+                    1742.0,
+                    1409.0,
+                    1742.0,
+                    1409.0,
+                    1781.0,
+                    1397.0,
+                    1781.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1775.0,
+                    1131.0,
+                    1775.0,
+                    1131.0,
+                    1809.0,
+                    295.0,
+                    1809.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1206.0,
+                    1775.0,
+                    1219.0,
+                    1775.0,
+                    1219.0,
+                    1809.0,
+                    1206.0,
+                    1809.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1320.0,
+                    813.0,
+                    1320.0,
+                    813.0,
+                    1356.0,
+                    295.0,
+                    1356.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    944.0,
+                    1320.0,
+                    1406.0,
+                    1320.0,
+                    1406.0,
+                    1356.0,
+                    944.0,
+                    1356.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1348.0,
+                    1408.0,
+                    1348.0,
+                    1408.0,
+                    1386.0,
+                    294.0,
+                    1386.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1380.0,
+                    1405.0,
+                    1380.0,
+                    1405.0,
+                    1417.0,
+                    295.0,
+                    1417.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1411.0,
+                    1407.0,
+                    1411.0,
+                    1407.0,
+                    1446.0,
+                    294.0,
+                    1446.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1444.0,
+                    1406.0,
+                    1444.0,
+                    1406.0,
+                    1476.0,
+                    296.0,
+                    1476.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1473.0,
+                    602.0,
+                    1473.0,
+                    602.0,
+                    1509.0,
+                    295.0,
+                    1509.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    353.0,
+                    1000.0,
+                    353.0,
+                    1000.0,
+                    387.0,
+                    296.0,
+                    387.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1077.0,
+                    353.0,
+                    1407.0,
+                    353.0,
+                    1407.0,
+                    387.0,
+                    1077.0,
+                    387.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    383.0,
+                    1406.0,
+                    383.0,
+                    1406.0,
+                    417.0,
+                    296.0,
+                    417.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    414.0,
+                    987.0,
+                    414.0,
+                    987.0,
+                    447.0,
+                    295.0,
+                    447.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    681.0,
+                    368.0,
+                    681.0,
+                    368.0,
+                    713.0,
+                    295.0,
+                    713.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    388.0,
+                    681.0,
+                    697.0,
+                    681.0,
+                    697.0,
+                    713.0,
+                    388.0,
+                    713.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    819.0,
+                    681.0,
+                    1400.0,
+                    681.0,
+                    1400.0,
+                    713.0,
+                    819.0,
+                    713.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    709.0,
+                    652.0,
+                    709.0,
+                    652.0,
+                    744.0,
+                    295.0,
+                    744.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    856.0,
+                    709.0,
+                    1405.0,
+                    709.0,
+                    1405.0,
+                    744.0,
+                    856.0,
+                    744.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    743.0,
+                    621.0,
+                    743.0,
+                    621.0,
+                    775.0,
+                    296.0,
+                    775.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    640.0,
+                    743.0,
+                    800.0,
+                    743.0,
+                    800.0,
+                    775.0,
+                    640.0,
+                    775.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    947.0,
+                    743.0,
+                    1402.0,
+                    743.0,
+                    1402.0,
+                    775.0,
+                    947.0,
+                    775.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    771.0,
+                    416.0,
+                    771.0,
+                    416.0,
+                    806.0,
+                    294.0,
+                    806.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    592.0,
+                    771.0,
+                    608.0,
+                    771.0,
+                    608.0,
+                    806.0,
+                    592.0,
+                    806.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    965.0,
+                    1405.0,
+                    965.0,
+                    1405.0,
+                    1006.0,
+                    294.0,
+                    1006.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    999.0,
+                    1405.0,
+                    999.0,
+                    1405.0,
+                    1033.0,
+                    296.0,
+                    1033.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1030.0,
+                    1039.0,
+                    1030.0,
+                    1039.0,
+                    1064.0,
+                    296.0,
+                    1064.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    889.0,
+                    877.0,
+                    889.0,
+                    877.0,
+                    925.0,
+                    295.0,
+                    925.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    899.0,
+                    889.0,
+                    1405.0,
+                    889.0,
+                    1405.0,
+                    925.0,
+                    899.0,
+                    925.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    921.0,
+                    699.0,
+                    921.0,
+                    699.0,
+                    954.0,
+                    297.0,
+                    954.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    321.0,
+                    1085.0,
+                    1407.0,
+                    1085.0,
+                    1407.0,
+                    1123.0,
+                    321.0,
+                    1123.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    353.0,
+                    1117.0,
+                    1172.0,
+                    1117.0,
+                    1172.0,
+                    1152.0,
+                    353.0,
+                    1152.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    315.0,
+                    1154.0,
+                    1407.0,
+                    1154.0,
+                    1407.0,
+                    1198.0,
+                    315.0,
+                    1198.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    353.0,
+                    1189.0,
+                    465.0,
+                    1189.0,
+                    465.0,
+                    1224.0,
+                    353.0,
+                    1224.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    316.0,
+                    1228.0,
+                    1408.0,
+                    1228.0,
+                    1408.0,
+                    1268.0,
+                    316.0,
+                    1268.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    352.0,
+                    1260.0,
+                    403.0,
+                    1260.0,
+                    403.0,
+                    1299.0,
+                    352.0,
+                    1299.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    457.0,
+                    696.0,
+                    457.0,
+                    696.0,
+                    498.0,
+                    294.0,
+                    498.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    745.0,
+                    457.0,
+                    1404.0,
+                    457.0,
+                    1404.0,
+                    498.0,
+                    745.0,
+                    498.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    494.0,
+                    1153.0,
+                    494.0,
+                    1153.0,
+                    531.0,
+                    295.0,
+                    531.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1971.0,
+                    1404.0,
+                    1971.0,
+                    1404.0,
+                    2007.0,
+                    295.0,
+                    2007.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1998.0,
+                    1406.0,
+                    1998.0,
+                    1406.0,
+                    2040.0,
+                    294.0,
+                    2040.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 14,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    300,
+                    834,
+                    1405,
+                    834,
+                    1405,
+                    959,
+                    300,
+                    959
+                ],
+                "score": 0.98
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    471,
+                    1407,
+                    471,
+                    1407,
+                    749,
+                    297,
+                    749
+                ],
+                "score": 0.974
+            },
+            {
+                "category_id": 3,
+                "poly": [
+                    446,
+                    984,
+                    1262,
+                    984,
+                    1262,
+                    1300,
+                    446,
+                    1300
+                ],
+                "score": 0.971
+            },
+            {
+                "category_id": 5,
+                "poly": [
+                    302,
+                    281,
+                    1419,
+                    281,
+                    1419,
+                    417,
+                    302,
+                    417
+                ],
+                "score": 0.965,
+                "html": "<table><tr><td>β</td><td>Environment</td><td>Cooperating</td></tr><tr><td>&lt;0</td><td>incentive-opposed</td><td>yields less reward on the current time-step (myopically detrimental)</td></tr><tr><td>=0</td><td>incentive-orthogonal</td><td>does not affect the current reward (myopically indifferent)</td></tr><tr><td>v0</td><td>incentive-compatible</td><td>yields more reward on the current time-step (myopically beneficial)</td></tr></table>"
+            },
+            {
+                "category_id": 4,
+                "poly": [
+                    293,
+                    1337,
+                    1403,
+                    1337,
+                    1403,
+                    1402,
+                    293,
+                    1402
+                ],
+                "score": 0.935
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    300,
+                    782,
+                    785,
+                    782,
+                    785,
+                    814,
+                    300,
+                    814
+                ],
+                "score": 0.91
+            },
+            {
+                "category_id": 6,
+                "poly": [
+                    354,
+                    249,
+                    1324,
+                    249,
+                    1324,
+                    281,
+                    354,
+                    281
+                ],
+                "score": 0.89
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    297,
+                    75,
+                    854,
+                    75,
+                    854,
+                    105,
+                    297,
+                    105
+                ],
+                "score": 0.888
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    836,
+                    2088,
+                    864,
+                    2088,
+                    864,
+                    2113,
+                    836,
+                    2113
+                ],
+                "score": 0.846
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1101,
+                    534,
+                    1303,
+                    534,
+                    1303,
+                    567,
+                    1101,
+                    567
+                ],
+                "score": 0.9,
+                "latex": "R ( \\mathrm { d e f e c t } ) = - . 5"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    298,
+                    686,
+                    524,
+                    686,
+                    524,
+                    719,
+                    298,
+                    719
+                ],
+                "score": 0.88,
+                "latex": "Q ( { \\mathrm { d e f e c t } } ) \\approx - 0 . 4 5"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    435,
+                    505,
+                    528,
+                    505,
+                    528,
+                    533,
+                    435,
+                    533
+                ],
+                "score": 0.88,
+                "latex": "\\epsilon = 0 . 1"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1083,
+                    657,
+                    1354,
+                    657,
+                    1354,
+                    689,
+                    1083,
+                    689
+                ],
+                "score": 0.87,
+                "latex": "Q ( { \\mathrm { c o o p e r a t e } } ) \\approx - 0 . 9 2"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1113,
+                    566,
+                    1166,
+                    566,
+                    1166,
+                    594,
+                    1113,
+                    594
+                ],
+                "score": 0.83,
+                "latex": "50 \\%"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1286,
+                    626,
+                    1402,
+                    626,
+                    1402,
+                    657,
+                    1286,
+                    657
+                ],
+                "score": 0.8,
+                "latex": "Q ( { \\mathrm { d e f e c t } } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    314,
+                    322,
+                    361,
+                    322,
+                    361,
+                    350,
+                    314,
+                    350
+                ],
+                "score": 0.8,
+                "latex": "< 0"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    469,
+                    250,
+                    491,
+                    250,
+                    491,
+                    281,
+                    469,
+                    281
+                ],
+                "score": 0.77,
+                "latex": "\\beta"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    311,
+                    352,
+                    361,
+                    352,
+                    361,
+                    380,
+                    311,
+                    380
+                ],
+                "score": 0.75,
+                "latex": "= 0"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1072,
+                    626,
+                    1097,
+                    626,
+                    1097,
+                    657,
+                    1072,
+                    657
+                ],
+                "score": 0.69,
+                "latex": "Q"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    340,
+                    280,
+                    360,
+                    280,
+                    360,
+                    311,
+                    340,
+                    311
+                ],
+                "score": 0.69,
+                "latex": "\\beta"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    543,
+                    656,
+                    616,
+                    656,
+                    616,
+                    686,
+                    543,
+                    686
+                ],
+                "score": 0.67,
+                "latex": "- 0 . 0 7"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1303,
+                    479,
+                    1317,
+                    479,
+                    1317,
+                    501,
+                    1303,
+                    501
+                ],
+                "score": 0.67,
+                "latex": "\\epsilon"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    310,
+                    382,
+                    384,
+                    382,
+                    384,
+                    417,
+                    310,
+                    417
+                ],
+                "score": 0.57,
+                "latex": "> 0 \\ \\mid"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    941.0,
+                    985.0,
+                    984.0,
+                    985.0,
+                    984.0,
+                    1013.0,
+                    941.0,
+                    1013.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1092.0,
+                    985.0,
+                    1135.0,
+                    985.0,
+                    1135.0,
+                    1014.0,
+                    1092.0,
+                    1014.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    446.0,
+                    1001.0,
+                    476.0,
+                    1001.0,
+                    476.0,
+                    1091.0,
+                    446.0,
+                    1091.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    487.0,
+                    1008.0,
+                    527.0,
+                    1008.0,
+                    527.0,
+                    1037.0,
+                    487.0,
+                    1037.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    658.0,
+                    1008.0,
+                    677.0,
+                    1008.0,
+                    677.0,
+                    1030.0,
+                    658.0,
+                    1030.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    811.0,
+                    1005.0,
+                    826.0,
+                    1005.0,
+                    826.0,
+                    1023.0,
+                    811.0,
+                    1023.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1087.0,
+                    1035.0,
+                    1134.0,
+                    1035.0,
+                    1134.0,
+                    1064.0,
+                    1087.0,
+                    1064.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    476.0,
+                    1076.0,
+                    528.0,
+                    1076.0,
+                    528.0,
+                    1107.0,
+                    476.0,
+                    1107.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    658.0,
+                    1088.0,
+                    670.0,
+                    1088.0,
+                    670.0,
+                    1101.0,
+                    658.0,
+                    1101.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    936.0,
+                    1077.0,
+                    980.0,
+                    1077.0,
+                    980.0,
+                    1106.0,
+                    936.0,
+                    1106.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    460.0,
+                    1097.0,
+                    527.0,
+                    1097.0,
+                    527.0,
+                    1224.0,
+                    460.0,
+                    1224.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    662.0,
+                    1104.0,
+                    672.0,
+                    1104.0,
+                    672.0,
+                    1118.0,
+                    662.0,
+                    1118.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    940.0,
+                    1097.0,
+                    983.0,
+                    1097.0,
+                    983.0,
+                    1126.0,
+                    940.0,
+                    1126.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1091.0,
+                    1097.0,
+                    1135.0,
+                    1097.0,
+                    1135.0,
+                    1126.0,
+                    1091.0,
+                    1126.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1091.0,
+                    1146.0,
+                    1134.0,
+                    1146.0,
+                    1134.0,
+                    1175.0,
+                    1091.0,
+                    1175.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    487.0,
+                    1159.0,
+                    527.0,
+                    1159.0,
+                    527.0,
+                    1188.0,
+                    487.0,
+                    1188.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    661.0,
+                    1201.0,
+                    675.0,
+                    1201.0,
+                    675.0,
+                    1219.0,
+                    661.0,
+                    1219.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    811.0,
+                    1201.0,
+                    826.0,
+                    1201.0,
+                    826.0,
+                    1218.0,
+                    811.0,
+                    1218.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    940.0,
+                    1189.0,
+                    982.0,
+                    1189.0,
+                    982.0,
+                    1218.0,
+                    940.0,
+                    1218.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1093.0,
+                    1196.0,
+                    1132.0,
+                    1196.0,
+                    1132.0,
+                    1222.0,
+                    1093.0,
+                    1222.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    528.0,
+                    1217.0,
+                    548.0,
+                    1217.0,
+                    548.0,
+                    1240.0,
+                    528.0,
+                    1240.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    592.0,
+                    1215.0,
+                    659.0,
+                    1215.0,
+                    659.0,
+                    1241.0,
+                    592.0,
+                    1241.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    679.0,
+                    1217.0,
+                    700.0,
+                    1217.0,
+                    700.0,
+                    1240.0,
+                    679.0,
+                    1240.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    744.0,
+                    1215.0,
+                    811.0,
+                    1215.0,
+                    811.0,
+                    1241.0,
+                    744.0,
+                    1241.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    830.0,
+                    1217.0,
+                    851.0,
+                    1217.0,
+                    851.0,
+                    1240.0,
+                    830.0,
+                    1240.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    896.0,
+                    1215.0,
+                    963.0,
+                    1215.0,
+                    963.0,
+                    1242.0,
+                    896.0,
+                    1242.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    980.0,
+                    1215.0,
+                    1003.0,
+                    1215.0,
+                    1003.0,
+                    1241.0,
+                    980.0,
+                    1241.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1046.0,
+                    1215.0,
+                    1113.0,
+                    1215.0,
+                    1113.0,
+                    1241.0,
+                    1046.0,
+                    1241.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1132.0,
+                    1217.0,
+                    1153.0,
+                    1217.0,
+                    1153.0,
+                    1240.0,
+                    1132.0,
+                    1240.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1199.0,
+                    1215.0,
+                    1264.0,
+                    1215.0,
+                    1264.0,
+                    1240.0,
+                    1199.0,
+                    1240.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    533.0,
+                    1241.0,
+                    629.0,
+                    1241.0,
+                    629.0,
+                    1269.0,
+                    533.0,
+                    1269.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    685.0,
+                    1241.0,
+                    779.0,
+                    1241.0,
+                    779.0,
+                    1269.0,
+                    685.0,
+                    1269.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    836.0,
+                    1239.0,
+                    931.0,
+                    1239.0,
+                    931.0,
+                    1269.0,
+                    836.0,
+                    1269.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    987.0,
+                    1240.0,
+                    1081.0,
+                    1240.0,
+                    1081.0,
+                    1270.0,
+                    987.0,
+                    1270.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1137.0,
+                    1238.0,
+                    1234.0,
+                    1238.0,
+                    1234.0,
+                    1271.0,
+                    1137.0,
+                    1271.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    586.0,
+                    1265.0,
+                    680.0,
+                    1265.0,
+                    680.0,
+                    1293.0,
+                    586.0,
+                    1293.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    756.0,
+                    1266.0,
+                    882.0,
+                    1266.0,
+                    882.0,
+                    1295.0,
+                    756.0,
+                    1295.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    956.0,
+                    1262.0,
+                    1165.0,
+                    1262.0,
+                    1165.0,
+                    1297.0,
+                    956.0,
+                    1297.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    815.0,
+                    1105.0,
+                    821.0,
+                    1105.0,
+                    821.0,
+                    1116.0,
+                    815.0,
+                    1116.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1337.0,
+                    1405.0,
+                    1337.0,
+                    1405.0,
+                    1373.0,
+                    295.0,
+                    1373.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1367.0,
+                    851.0,
+                    1367.0,
+                    851.0,
+                    1406.0,
+                    294.0,
+                    1406.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    780.0,
+                    787.0,
+                    780.0,
+                    787.0,
+                    819.0,
+                    295.0,
+                    819.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    360.0,
+                    245.0,
+                    468.0,
+                    245.0,
+                    468.0,
+                    286.0,
+                    360.0,
+                    286.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    492.0,
+                    245.0,
+                    1328.0,
+                    245.0,
+                    1328.0,
+                    286.0,
+                    492.0,
+                    286.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    298.0,
+                    73.0,
+                    856.0,
+                    73.0,
+                    856.0,
+                    108.0,
+                    298.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    831.0,
+                    2085.0,
+                    868.0,
+                    2085.0,
+                    868.0,
+                    2124.0,
+                    831.0,
+                    2124.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    836.0,
+                    1406.0,
+                    836.0,
+                    1406.0,
+                    868.0,
+                    295.0,
+                    868.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    866.0,
+                    1406.0,
+                    866.0,
+                    1406.0,
+                    901.0,
+                    293.0,
+                    901.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    898.0,
+                    1404.0,
+                    898.0,
+                    1404.0,
+                    930.0,
+                    295.0,
+                    930.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    928.0,
+                    738.0,
+                    928.0,
+                    738.0,
+                    961.0,
+                    296.0,
+                    961.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    472.0,
+                    1302.0,
+                    472.0,
+                    1302.0,
+                    509.0,
+                    295.0,
+                    509.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1318.0,
+                    472.0,
+                    1404.0,
+                    472.0,
+                    1404.0,
+                    509.0,
+                    1318.0,
+                    509.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    501.0,
+                    434.0,
+                    501.0,
+                    434.0,
+                    542.0,
+                    294.0,
+                    542.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    529.0,
+                    501.0,
+                    1407.0,
+                    501.0,
+                    1407.0,
+                    542.0,
+                    529.0,
+                    542.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    534.0,
+                    1100.0,
+                    534.0,
+                    1100.0,
+                    571.0,
+                    292.0,
+                    571.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1304.0,
+                    534.0,
+                    1407.0,
+                    534.0,
+                    1407.0,
+                    571.0,
+                    1304.0,
+                    571.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    563.0,
+                    1112.0,
+                    563.0,
+                    1112.0,
+                    599.0,
+                    292.0,
+                    599.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1167.0,
+                    563.0,
+                    1405.0,
+                    563.0,
+                    1405.0,
+                    599.0,
+                    1167.0,
+                    599.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    593.0,
+                    1407.0,
+                    593.0,
+                    1407.0,
+                    632.0,
+                    292.0,
+                    632.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    623.0,
+                    1071.0,
+                    623.0,
+                    1071.0,
+                    661.0,
+                    292.0,
+                    661.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1098.0,
+                    623.0,
+                    1285.0,
+                    623.0,
+                    1285.0,
+                    661.0,
+                    1098.0,
+                    661.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    655.0,
+                    542.0,
+                    655.0,
+                    542.0,
+                    691.0,
+                    294.0,
+                    691.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    617.0,
+                    655.0,
+                    1082.0,
+                    655.0,
+                    1082.0,
+                    691.0,
+                    617.0,
+                    691.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1355.0,
+                    655.0,
+                    1405.0,
+                    655.0,
+                    1405.0,
+                    691.0,
+                    1355.0,
+                    691.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    682.0,
+                    297.0,
+                    682.0,
+                    297.0,
+                    726.0,
+                    294.0,
+                    726.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    525.0,
+                    682.0,
+                    1408.0,
+                    682.0,
+                    1408.0,
+                    726.0,
+                    525.0,
+                    726.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    718.0,
+                    578.0,
+                    718.0,
+                    578.0,
+                    752.0,
+                    294.0,
+                    752.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 15,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 3,
+                "poly": [
+                    448,
+                    289,
+                    1244,
+                    289,
+                    1244,
+                    1867,
+                    448,
+                    1867
+                ],
+                "score": 0.966
+            },
+            {
+                "category_id": 4,
+                "poly": [
+                    297,
+                    1894,
+                    1402,
+                    1894,
+                    1402,
+                    1958,
+                    297,
+                    1958
+                ],
+                "score": 0.95
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    835,
+                    2087,
+                    863,
+                    2087,
+                    863,
+                    2112,
+                    835,
+                    2112
+                ],
+                "score": 0.852
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    299,
+                    76,
+                    852,
+                    76,
+                    852,
+                    104,
+                    299,
+                    104
+                ],
+                "score": 0.485
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    298,
+                    76,
+                    854,
+                    76,
+                    854,
+                    105,
+                    298,
+                    105
+                ],
+                "score": 0.428
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    447.0,
+                    311.0,
+                    477.0,
+                    311.0,
+                    477.0,
+                    402.0,
+                    447.0,
+                    402.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    487.0,
+                    309.0,
+                    528.0,
+                    309.0,
+                    528.0,
+                    335.0,
+                    487.0,
+                    335.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    636.0,
+                    293.0,
+                    682.0,
+                    293.0,
+                    682.0,
+                    325.0,
+                    636.0,
+                    325.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    788.0,
+                    307.0,
+                    832.0,
+                    307.0,
+                    832.0,
+                    339.0,
+                    788.0,
+                    339.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    940.0,
+                    302.0,
+                    982.0,
+                    302.0,
+                    982.0,
+                    328.0,
+                    940.0,
+                    328.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1091.0,
+                    311.0,
+                    1135.0,
+                    311.0,
+                    1135.0,
+                    342.0,
+                    1091.0,
+                    342.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    473.0,
+                    349.0,
+                    528.0,
+                    349.0,
+                    528.0,
+                    381.0,
+                    473.0,
+                    381.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    631.0,
+                    351.0,
+                    680.0,
+                    351.0,
+                    680.0,
+                    382.0,
+                    631.0,
+                    382.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    785.0,
+                    349.0,
+                    830.0,
+                    349.0,
+                    830.0,
+                    381.0,
+                    785.0,
+                    381.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    934.0,
+                    372.0,
+                    982.0,
+                    372.0,
+                    982.0,
+                    405.0,
+                    934.0,
+                    405.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1088.0,
+                    368.0,
+                    1131.0,
+                    368.0,
+                    1131.0,
+                    395.0,
+                    1088.0,
+                    395.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    461.0,
+                    405.0,
+                    528.0,
+                    405.0,
+                    528.0,
+                    537.0,
+                    461.0,
+                    537.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    634.0,
+                    405.0,
+                    678.0,
+                    405.0,
+                    678.0,
+                    437.0,
+                    634.0,
+                    437.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    809.0,
+                    410.0,
+                    830.0,
+                    410.0,
+                    830.0,
+                    433.0,
+                    809.0,
+                    433.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    939.0,
+                    405.0,
+                    984.0,
+                    405.0,
+                    984.0,
+                    437.0,
+                    939.0,
+                    437.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1089.0,
+                    405.0,
+                    1135.0,
+                    405.0,
+                    1135.0,
+                    437.0,
+                    1089.0,
+                    437.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    486.0,
+                    452.0,
+                    529.0,
+                    452.0,
+                    529.0,
+                    485.0,
+                    486.0,
+                    485.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    636.0,
+                    452.0,
+                    680.0,
+                    452.0,
+                    680.0,
+                    485.0,
+                    636.0,
+                    485.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    939.0,
+                    457.0,
+                    984.0,
+                    457.0,
+                    984.0,
+                    489.0,
+                    939.0,
+                    489.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1089.0,
+                    454.0,
+                    1133.0,
+                    454.0,
+                    1133.0,
+                    487.0,
+                    1089.0,
+                    487.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    808.0,
+                    503.0,
+                    834.0,
+                    503.0,
+                    834.0,
+                    533.0,
+                    808.0,
+                    533.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    528.0,
+                    524.0,
+                    548.0,
+                    524.0,
+                    548.0,
+                    550.0,
+                    528.0,
+                    550.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    582.0,
+                    520.0,
+                    641.0,
+                    520.0,
+                    641.0,
+                    552.0,
+                    582.0,
+                    552.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    680.0,
+                    524.0,
+                    701.0,
+                    524.0,
+                    701.0,
+                    550.0,
+                    680.0,
+                    550.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    732.0,
+                    520.0,
+                    792.0,
+                    520.0,
+                    792.0,
+                    552.0,
+                    732.0,
+                    552.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    830.0,
+                    524.0,
+                    851.0,
+                    524.0,
+                    851.0,
+                    550.0,
+                    830.0,
+                    550.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    883.0,
+                    520.0,
+                    942.0,
+                    520.0,
+                    942.0,
+                    552.0,
+                    883.0,
+                    552.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    982.0,
+                    524.0,
+                    1004.0,
+                    524.0,
+                    1004.0,
+                    550.0,
+                    982.0,
+                    550.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1035.0,
+                    520.0,
+                    1094.0,
+                    520.0,
+                    1094.0,
+                    552.0,
+                    1035.0,
+                    552.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1131.0,
+                    526.0,
+                    1154.0,
+                    526.0,
+                    1154.0,
+                    550.0,
+                    1131.0,
+                    550.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1186.0,
+                    520.0,
+                    1245.0,
+                    520.0,
+                    1245.0,
+                    552.0,
+                    1186.0,
+                    552.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    533.0,
+                    548.0,
+                    631.0,
+                    548.0,
+                    631.0,
+                    578.0,
+                    533.0,
+                    578.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    682.0,
+                    544.0,
+                    782.0,
+                    544.0,
+                    782.0,
+                    579.0,
+                    682.0,
+                    579.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    833.0,
+                    544.0,
+                    932.0,
+                    544.0,
+                    932.0,
+                    579.0,
+                    833.0,
+                    579.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    984.0,
+                    548.0,
+                    1082.0,
+                    548.0,
+                    1082.0,
+                    578.0,
+                    984.0,
+                    578.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1136.0,
+                    544.0,
+                    1235.0,
+                    544.0,
+                    1235.0,
+                    579.0,
+                    1136.0,
+                    579.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    585.0,
+                    573.0,
+                    682.0,
+                    573.0,
+                    682.0,
+                    603.0,
+                    585.0,
+                    603.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    755.0,
+                    575.0,
+                    884.0,
+                    575.0,
+                    884.0,
+                    603.0,
+                    755.0,
+                    603.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    955.0,
+                    569.0,
+                    1166.0,
+                    569.0,
+                    1166.0,
+                    606.0,
+                    955.0,
+                    606.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    447.0,
+                    610.0,
+                    540.0,
+                    610.0,
+                    540.0,
+                    718.0,
+                    447.0,
+                    718.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    636.0,
+                    611.0,
+                    682.0,
+                    611.0,
+                    682.0,
+                    643.0,
+                    636.0,
+                    643.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    788.0,
+                    613.0,
+                    830.0,
+                    613.0,
+                    830.0,
+                    639.0,
+                    788.0,
+                    639.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    940.0,
+                    613.0,
+                    1010.0,
+                    613.0,
+                    1010.0,
+                    639.0,
+                    940.0,
+                    639.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1093.0,
+                    613.0,
+                    1135.0,
+                    613.0,
+                    1135.0,
+                    639.0,
+                    1093.0,
+                    639.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    632.0,
+                    683.0,
+                    680.0,
+                    683.0,
+                    680.0,
+                    716.0,
+                    632.0,
+                    716.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    785.0,
+                    678.0,
+                    828.0,
+                    678.0,
+                    828.0,
+                    704.0,
+                    785.0,
+                    704.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    937.0,
+                    700.0,
+                    979.0,
+                    700.0,
+                    979.0,
+                    727.0,
+                    937.0,
+                    727.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1084.0,
+                    699.0,
+                    1133.0,
+                    699.0,
+                    1133.0,
+                    730.0,
+                    1084.0,
+                    730.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    461.0,
+                    716.0,
+                    526.0,
+                    716.0,
+                    526.0,
+                    848.0,
+                    461.0,
+                    848.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    636.0,
+                    720.0,
+                    678.0,
+                    720.0,
+                    678.0,
+                    751.0,
+                    636.0,
+                    751.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    788.0,
+                    720.0,
+                    830.0,
+                    720.0,
+                    830.0,
+                    751.0,
+                    788.0,
+                    751.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    939.0,
+                    720.0,
+                    984.0,
+                    720.0,
+                    984.0,
+                    751.0,
+                    939.0,
+                    751.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1089.0,
+                    720.0,
+                    1133.0,
+                    720.0,
+                    1133.0,
+                    751.0,
+                    1089.0,
+                    751.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    486.0,
+                    765.0,
+                    528.0,
+                    765.0,
+                    528.0,
+                    798.0,
+                    486.0,
+                    798.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    636.0,
+                    769.0,
+                    680.0,
+                    769.0,
+                    680.0,
+                    800.0,
+                    636.0,
+                    800.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    788.0,
+                    769.0,
+                    832.0,
+                    769.0,
+                    832.0,
+                    800.0,
+                    788.0,
+                    800.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    939.0,
+                    781.0,
+                    990.0,
+                    781.0,
+                    990.0,
+                    812.0,
+                    939.0,
+                    812.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    991.0,
+                    770.0,
+                    1019.0,
+                    770.0,
+                    1019.0,
+                    798.0,
+                    991.0,
+                    798.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1089.0,
+                    790.0,
+                    1133.0,
+                    790.0,
+                    1133.0,
+                    821.0,
+                    1089.0,
+                    821.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    528.0,
+                    837.0,
+                    548.0,
+                    837.0,
+                    548.0,
+                    860.0,
+                    528.0,
+                    860.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    580.0,
+                    833.0,
+                    641.0,
+                    833.0,
+                    641.0,
+                    863.0,
+                    580.0,
+                    863.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    680.0,
+                    837.0,
+                    701.0,
+                    837.0,
+                    701.0,
+                    860.0,
+                    680.0,
+                    860.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    732.0,
+                    833.0,
+                    792.0,
+                    833.0,
+                    792.0,
+                    863.0,
+                    732.0,
+                    863.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    830.0,
+                    837.0,
+                    851.0,
+                    837.0,
+                    851.0,
+                    860.0,
+                    830.0,
+                    860.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    883.0,
+                    833.0,
+                    942.0,
+                    833.0,
+                    942.0,
+                    863.0,
+                    883.0,
+                    863.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    981.0,
+                    837.0,
+                    1004.0,
+                    837.0,
+                    1004.0,
+                    860.0,
+                    981.0,
+                    860.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1035.0,
+                    833.0,
+                    1094.0,
+                    833.0,
+                    1094.0,
+                    863.0,
+                    1035.0,
+                    863.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1133.0,
+                    837.0,
+                    1152.0,
+                    837.0,
+                    1152.0,
+                    860.0,
+                    1133.0,
+                    860.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1186.0,
+                    833.0,
+                    1245.0,
+                    833.0,
+                    1245.0,
+                    863.0,
+                    1186.0,
+                    863.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    532.0,
+                    856.0,
+                    631.0,
+                    856.0,
+                    631.0,
+                    890.0,
+                    532.0,
+                    890.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    682.0,
+                    857.0,
+                    783.0,
+                    857.0,
+                    783.0,
+                    890.0,
+                    682.0,
+                    890.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    833.0,
+                    857.0,
+                    934.0,
+                    857.0,
+                    934.0,
+                    891.0,
+                    833.0,
+                    891.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    983.0,
+                    856.0,
+                    1084.0,
+                    856.0,
+                    1084.0,
+                    890.0,
+                    983.0,
+                    890.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1132.0,
+                    856.0,
+                    1238.0,
+                    856.0,
+                    1238.0,
+                    893.0,
+                    1132.0,
+                    893.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    585.0,
+                    882.0,
+                    682.0,
+                    882.0,
+                    682.0,
+                    912.0,
+                    585.0,
+                    912.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    755.0,
+                    882.0,
+                    884.0,
+                    882.0,
+                    884.0,
+                    910.0,
+                    755.0,
+                    910.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    955.0,
+                    877.0,
+                    1166.0,
+                    877.0,
+                    1166.0,
+                    914.0,
+                    955.0,
+                    914.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    939.0,
+                    921.0,
+                    990.0,
+                    921.0,
+                    990.0,
+                    950.0,
+                    939.0,
+                    950.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    638.0,
+                    922.0,
+                    680.0,
+                    922.0,
+                    680.0,
+                    949.0,
+                    638.0,
+                    949.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1093.0,
+                    922.0,
+                    1133.0,
+                    922.0,
+                    1133.0,
+                    949.0,
+                    1093.0,
+                    949.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    682.0,
+                    929.0,
+                    699.0,
+                    929.0,
+                    699.0,
+                    943.0,
+                    682.0,
+                    943.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    461.0,
+                    933.0,
+                    494.0,
+                    933.0,
+                    494.0,
+                    1160.0,
+                    461.0,
+                    1160.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    786.0,
+                    933.0,
+                    830.0,
+                    933.0,
+                    830.0,
+                    964.0,
+                    786.0,
+                    964.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    505.0,
+                    936.0,
+                    528.0,
+                    936.0,
+                    528.0,
+                    961.0,
+                    505.0,
+                    961.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    785.0,
+                    978.0,
+                    828.0,
+                    978.0,
+                    828.0,
+                    1003.0,
+                    785.0,
+                    1003.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    934.0,
+                    968.0,
+                    982.0,
+                    968.0,
+                    982.0,
+                    999.0,
+                    934.0,
+                    999.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1086.0,
+                    970.0,
+                    1131.0,
+                    970.0,
+                    1131.0,
+                    1001.0,
+                    1086.0,
+                    1001.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    491.0,
+                    1015.0,
+                    524.0,
+                    1015.0,
+                    524.0,
+                    1043.0,
+                    491.0,
+                    1043.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    632.0,
+                    1012.0,
+                    678.0,
+                    1012.0,
+                    678.0,
+                    1043.0,
+                    632.0,
+                    1043.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    486.0,
+                    1031.0,
+                    528.0,
+                    1031.0,
+                    528.0,
+                    1064.0,
+                    486.0,
+                    1064.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    638.0,
+                    1034.0,
+                    676.0,
+                    1034.0,
+                    676.0,
+                    1061.0,
+                    638.0,
+                    1061.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    790.0,
+                    1034.0,
+                    830.0,
+                    1034.0,
+                    830.0,
+                    1061.0,
+                    790.0,
+                    1061.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    940.0,
+                    1034.0,
+                    982.0,
+                    1034.0,
+                    982.0,
+                    1061.0,
+                    940.0,
+                    1061.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1091.0,
+                    1034.0,
+                    1131.0,
+                    1034.0,
+                    1131.0,
+                    1061.0,
+                    1091.0,
+                    1061.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    786.0,
+                    1080.0,
+                    830.0,
+                    1080.0,
+                    830.0,
+                    1111.0,
+                    786.0,
+                    1111.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    939.0,
+                    1080.0,
+                    982.0,
+                    1080.0,
+                    982.0,
+                    1111.0,
+                    939.0,
+                    1111.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1089.0,
+                    1080.0,
+                    1133.0,
+                    1080.0,
+                    1133.0,
+                    1111.0,
+                    1089.0,
+                    1111.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    487.0,
+                    1125.0,
+                    526.0,
+                    1125.0,
+                    526.0,
+                    1151.0,
+                    487.0,
+                    1151.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    638.0,
+                    1111.0,
+                    680.0,
+                    1111.0,
+                    680.0,
+                    1143.0,
+                    638.0,
+                    1143.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    940.0,
+                    1129.0,
+                    982.0,
+                    1129.0,
+                    982.0,
+                    1155.0,
+                    940.0,
+                    1155.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1091.0,
+                    1129.0,
+                    1135.0,
+                    1129.0,
+                    1135.0,
+                    1160.0,
+                    1091.0,
+                    1160.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    528.0,
+                    1150.0,
+                    550.0,
+                    1150.0,
+                    550.0,
+                    1174.0,
+                    528.0,
+                    1174.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    582.0,
+                    1146.0,
+                    641.0,
+                    1146.0,
+                    641.0,
+                    1178.0,
+                    582.0,
+                    1178.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    680.0,
+                    1150.0,
+                    701.0,
+                    1150.0,
+                    701.0,
+                    1174.0,
+                    680.0,
+                    1174.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    734.0,
+                    1146.0,
+                    792.0,
+                    1146.0,
+                    792.0,
+                    1178.0,
+                    734.0,
+                    1178.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    830.0,
+                    1150.0,
+                    851.0,
+                    1150.0,
+                    851.0,
+                    1174.0,
+                    830.0,
+                    1174.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    883.0,
+                    1146.0,
+                    942.0,
+                    1146.0,
+                    942.0,
+                    1176.0,
+                    883.0,
+                    1176.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    982.0,
+                    1150.0,
+                    1004.0,
+                    1150.0,
+                    1004.0,
+                    1174.0,
+                    982.0,
+                    1174.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1035.0,
+                    1146.0,
+                    1094.0,
+                    1146.0,
+                    1094.0,
+                    1178.0,
+                    1035.0,
+                    1178.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1133.0,
+                    1150.0,
+                    1154.0,
+                    1150.0,
+                    1154.0,
+                    1174.0,
+                    1133.0,
+                    1174.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1186.0,
+                    1146.0,
+                    1243.0,
+                    1146.0,
+                    1243.0,
+                    1176.0,
+                    1186.0,
+                    1176.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    530.0,
+                    1168.0,
+                    633.0,
+                    1168.0,
+                    633.0,
+                    1206.0,
+                    530.0,
+                    1206.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    681.0,
+                    1168.0,
+                    785.0,
+                    1168.0,
+                    785.0,
+                    1206.0,
+                    681.0,
+                    1206.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    831.0,
+                    1167.0,
+                    934.0,
+                    1167.0,
+                    934.0,
+                    1206.0,
+                    831.0,
+                    1206.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    985.0,
+                    1168.0,
+                    1083.0,
+                    1168.0,
+                    1083.0,
+                    1206.0,
+                    985.0,
+                    1206.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1134.0,
+                    1168.0,
+                    1239.0,
+                    1168.0,
+                    1239.0,
+                    1206.0,
+                    1134.0,
+                    1206.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    585.0,
+                    1199.0,
+                    682.0,
+                    1199.0,
+                    682.0,
+                    1228.0,
+                    585.0,
+                    1228.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    753.0,
+                    1199.0,
+                    884.0,
+                    1199.0,
+                    884.0,
+                    1227.0,
+                    753.0,
+                    1227.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    958.0,
+                    1200.0,
+                    1164.0,
+                    1200.0,
+                    1164.0,
+                    1228.0,
+                    958.0,
+                    1228.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    447.0,
+                    1251.0,
+                    475.0,
+                    1251.0,
+                    475.0,
+                    1344.0,
+                    447.0,
+                    1344.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    636.0,
+                    1258.0,
+                    680.0,
+                    1258.0,
+                    680.0,
+                    1289.0,
+                    636.0,
+                    1289.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    687.0,
+                    1274.0,
+                    701.0,
+                    1274.0,
+                    701.0,
+                    1288.0,
+                    687.0,
+                    1288.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    809.0,
+                    1253.0,
+                    832.0,
+                    1253.0,
+                    832.0,
+                    1277.0,
+                    809.0,
+                    1277.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    939.0,
+                    1235.0,
+                    998.0,
+                    1235.0,
+                    998.0,
+                    1265.0,
+                    939.0,
+                    1265.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1091.0,
+                    1258.0,
+                    1136.0,
+                    1258.0,
+                    1136.0,
+                    1289.0,
+                    1091.0,
+                    1289.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    459.0,
+                    1324.0,
+                    528.0,
+                    1324.0,
+                    528.0,
+                    1473.0,
+                    459.0,
+                    1473.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    631.0,
+                    1324.0,
+                    678.0,
+                    1324.0,
+                    678.0,
+                    1375.0,
+                    631.0,
+                    1375.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    788.0,
+                    1335.0,
+                    830.0,
+                    1335.0,
+                    830.0,
+                    1377.0,
+                    788.0,
+                    1377.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    935.0,
+                    1324.0,
+                    984.0,
+                    1324.0,
+                    984.0,
+                    1377.0,
+                    935.0,
+                    1377.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1086.0,
+                    1324.0,
+                    1133.0,
+                    1324.0,
+                    1133.0,
+                    1379.0,
+                    1086.0,
+                    1379.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    486.0,
+                    1417.0,
+                    528.0,
+                    1417.0,
+                    528.0,
+                    1449.0,
+                    486.0,
+                    1449.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    640.0,
+                    1422.0,
+                    678.0,
+                    1422.0,
+                    678.0,
+                    1449.0,
+                    640.0,
+                    1449.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    786.0,
+                    1393.0,
+                    830.0,
+                    1393.0,
+                    830.0,
+                    1424.0,
+                    786.0,
+                    1424.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    939.0,
+                    1407.0,
+                    986.0,
+                    1407.0,
+                    986.0,
+                    1438.0,
+                    939.0,
+                    1438.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1089.0,
+                    1405.0,
+                    1133.0,
+                    1405.0,
+                    1133.0,
+                    1436.0,
+                    1089.0,
+                    1436.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1152.0,
+                    1422.0,
+                    1161.0,
+                    1422.0,
+                    1161.0,
+                    1431.0,
+                    1152.0,
+                    1431.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    788.0,
+                    1443.0,
+                    830.0,
+                    1443.0,
+                    830.0,
+                    1470.0,
+                    788.0,
+                    1470.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    526.0,
+                    1461.0,
+                    552.0,
+                    1461.0,
+                    552.0,
+                    1491.0,
+                    526.0,
+                    1491.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    582.0,
+                    1459.0,
+                    641.0,
+                    1459.0,
+                    641.0,
+                    1491.0,
+                    582.0,
+                    1491.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    680.0,
+                    1464.0,
+                    701.0,
+                    1464.0,
+                    701.0,
+                    1489.0,
+                    680.0,
+                    1489.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    732.0,
+                    1459.0,
+                    792.0,
+                    1459.0,
+                    792.0,
+                    1491.0,
+                    732.0,
+                    1491.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    828.0,
+                    1463.0,
+                    851.0,
+                    1463.0,
+                    851.0,
+                    1487.0,
+                    828.0,
+                    1487.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    883.0,
+                    1459.0,
+                    942.0,
+                    1459.0,
+                    942.0,
+                    1491.0,
+                    883.0,
+                    1491.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    981.0,
+                    1463.0,
+                    1004.0,
+                    1463.0,
+                    1004.0,
+                    1487.0,
+                    981.0,
+                    1487.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1035.0,
+                    1459.0,
+                    1094.0,
+                    1459.0,
+                    1094.0,
+                    1491.0,
+                    1035.0,
+                    1491.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1133.0,
+                    1464.0,
+                    1154.0,
+                    1464.0,
+                    1154.0,
+                    1487.0,
+                    1133.0,
+                    1487.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1186.0,
+                    1459.0,
+                    1245.0,
+                    1459.0,
+                    1245.0,
+                    1491.0,
+                    1186.0,
+                    1491.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    533.0,
+                    1487.0,
+                    631.0,
+                    1487.0,
+                    631.0,
+                    1517.0,
+                    533.0,
+                    1517.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    683.0,
+                    1487.0,
+                    783.0,
+                    1487.0,
+                    783.0,
+                    1517.0,
+                    683.0,
+                    1517.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    831.0,
+                    1481.0,
+                    934.0,
+                    1481.0,
+                    934.0,
+                    1519.0,
+                    831.0,
+                    1519.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    984.0,
+                    1487.0,
+                    1082.0,
+                    1487.0,
+                    1082.0,
+                    1517.0,
+                    984.0,
+                    1517.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1136.0,
+                    1481.0,
+                    1235.0,
+                    1481.0,
+                    1235.0,
+                    1519.0,
+                    1136.0,
+                    1519.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    584.0,
+                    1511.0,
+                    682.0,
+                    1511.0,
+                    682.0,
+                    1539.0,
+                    584.0,
+                    1539.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    755.0,
+                    1510.0,
+                    886.0,
+                    1510.0,
+                    886.0,
+                    1538.0,
+                    755.0,
+                    1538.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    955.0,
+                    1506.0,
+                    1166.0,
+                    1506.0,
+                    1166.0,
+                    1543.0,
+                    955.0,
+                    1543.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    447.0,
+                    1560.0,
+                    529.0,
+                    1560.0,
+                    529.0,
+                    1655.0,
+                    447.0,
+                    1655.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    631.0,
+                    1560.0,
+                    680.0,
+                    1560.0,
+                    680.0,
+                    1637.0,
+                    631.0,
+                    1637.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    783.0,
+                    1546.0,
+                    841.0,
+                    1546.0,
+                    841.0,
+                    1625.0,
+                    783.0,
+                    1625.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    932.0,
+                    1545.0,
+                    1004.0,
+                    1545.0,
+                    1004.0,
+                    1639.0,
+                    932.0,
+                    1639.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1086.0,
+                    1560.0,
+                    1135.0,
+                    1560.0,
+                    1135.0,
+                    1637.0,
+                    1086.0,
+                    1637.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    461.0,
+                    1653.0,
+                    529.0,
+                    1653.0,
+                    529.0,
+                    1785.0,
+                    461.0,
+                    1785.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    634.0,
+                    1657.0,
+                    678.0,
+                    1657.0,
+                    678.0,
+                    1688.0,
+                    634.0,
+                    1688.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    783.0,
+                    1639.0,
+                    834.0,
+                    1639.0,
+                    834.0,
+                    1692.0,
+                    783.0,
+                    1692.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    939.0,
+                    1657.0,
+                    986.0,
+                    1657.0,
+                    986.0,
+                    1688.0,
+                    939.0,
+                    1688.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1089.0,
+                    1657.0,
+                    1133.0,
+                    1657.0,
+                    1133.0,
+                    1688.0,
+                    1089.0,
+                    1688.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    486.0,
+                    1702.0,
+                    528.0,
+                    1702.0,
+                    528.0,
+                    1735.0,
+                    486.0,
+                    1735.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    636.0,
+                    1704.0,
+                    682.0,
+                    1704.0,
+                    682.0,
+                    1737.0,
+                    636.0,
+                    1737.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    786.0,
+                    1704.0,
+                    830.0,
+                    1704.0,
+                    830.0,
+                    1735.0,
+                    786.0,
+                    1735.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    939.0,
+                    1704.0,
+                    984.0,
+                    1704.0,
+                    984.0,
+                    1737.0,
+                    939.0,
+                    1737.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1089.0,
+                    1704.0,
+                    1133.0,
+                    1704.0,
+                    1133.0,
+                    1735.0,
+                    1089.0,
+                    1735.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    487.0,
+                    1754.0,
+                    528.0,
+                    1754.0,
+                    528.0,
+                    1781.0,
+                    487.0,
+                    1781.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    788.0,
+                    1753.0,
+                    830.0,
+                    1753.0,
+                    830.0,
+                    1779.0,
+                    788.0,
+                    1779.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1093.0,
+                    1754.0,
+                    1133.0,
+                    1754.0,
+                    1133.0,
+                    1781.0,
+                    1093.0,
+                    1781.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    524.0,
+                    1772.0,
+                    552.0,
+                    1772.0,
+                    552.0,
+                    1802.0,
+                    524.0,
+                    1802.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    580.0,
+                    1768.0,
+                    643.0,
+                    1768.0,
+                    643.0,
+                    1805.0,
+                    580.0,
+                    1805.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    678.0,
+                    1772.0,
+                    704.0,
+                    1772.0,
+                    704.0,
+                    1802.0,
+                    678.0,
+                    1802.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    732.0,
+                    1772.0,
+                    794.0,
+                    1772.0,
+                    794.0,
+                    1802.0,
+                    732.0,
+                    1802.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    830.0,
+                    1774.0,
+                    851.0,
+                    1774.0,
+                    851.0,
+                    1800.0,
+                    830.0,
+                    1800.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    883.0,
+                    1770.0,
+                    944.0,
+                    1770.0,
+                    944.0,
+                    1802.0,
+                    883.0,
+                    1802.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    979.0,
+                    1772.0,
+                    1005.0,
+                    1772.0,
+                    1005.0,
+                    1802.0,
+                    979.0,
+                    1802.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1035.0,
+                    1770.0,
+                    1094.0,
+                    1770.0,
+                    1094.0,
+                    1802.0,
+                    1035.0,
+                    1802.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1131.0,
+                    1775.0,
+                    1154.0,
+                    1775.0,
+                    1154.0,
+                    1800.0,
+                    1131.0,
+                    1800.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1186.0,
+                    1772.0,
+                    1245.0,
+                    1772.0,
+                    1245.0,
+                    1802.0,
+                    1186.0,
+                    1802.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    532.0,
+                    1792.0,
+                    631.0,
+                    1792.0,
+                    631.0,
+                    1827.0,
+                    532.0,
+                    1827.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    682.0,
+                    1794.0,
+                    783.0,
+                    1794.0,
+                    783.0,
+                    1828.0,
+                    682.0,
+                    1828.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    833.0,
+                    1794.0,
+                    934.0,
+                    1794.0,
+                    934.0,
+                    1827.0,
+                    833.0,
+                    1827.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    983.0,
+                    1792.0,
+                    1084.0,
+                    1792.0,
+                    1084.0,
+                    1827.0,
+                    983.0,
+                    1827.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1134.0,
+                    1794.0,
+                    1237.0,
+                    1794.0,
+                    1237.0,
+                    1828.0,
+                    1134.0,
+                    1828.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    587.0,
+                    1821.0,
+                    682.0,
+                    1821.0,
+                    682.0,
+                    1851.0,
+                    587.0,
+                    1851.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    753.0,
+                    1819.0,
+                    886.0,
+                    1819.0,
+                    886.0,
+                    1852.0,
+                    753.0,
+                    1852.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    954.0,
+                    1819.0,
+                    1164.0,
+                    1819.0,
+                    1164.0,
+                    1852.0,
+                    954.0,
+                    1852.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    679.0,
+                    302.5,
+                    688.0,
+                    302.5,
+                    688.0,
+                    314.0,
+                    679.0,
+                    314.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1130.0,
+                    618.5,
+                    1155.0,
+                    618.5,
+                    1155.0,
+                    636.0,
+                    1130.0,
+                    636.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    697.0,
+                    1095.5,
+                    742.0,
+                    1095.5,
+                    742.0,
+                    1132.5,
+                    697.0,
+                    1132.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    488.0,
+                    1237.0,
+                    525.0,
+                    1237.0,
+                    525.0,
+                    1263.5,
+                    488.0,
+                    1263.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    533.75,
+                    1245.5,
+                    543.75,
+                    1245.5,
+                    543.75,
+                    1260.0,
+                    533.75,
+                    1260.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1164.0,
+                    1426.0,
+                    1216.0,
+                    1426.0,
+                    1216.0,
+                    1445.0,
+                    1164.0,
+                    1445.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1890.0,
+                    1406.0,
+                    1890.0,
+                    1406.0,
+                    1932.0,
+                    292.0,
+                    1932.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1922.0,
+                    1195.0,
+                    1922.0,
+                    1195.0,
+                    1961.0,
+                    292.0,
+                    1961.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    831.0,
+                    2085.0,
+                    867.0,
+                    2085.0,
+                    867.0,
+                    2124.0,
+                    831.0,
+                    2124.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    73.0,
+                    856.0,
+                    73.0,
+                    856.0,
+                    108.0,
+                    297.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    72.0,
+                    857.0,
+                    72.0,
+                    857.0,
+                    108.0,
+                    297.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 16,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 3,
+                "poly": [
+                    444,
+                    254,
+                    1244,
+                    254,
+                    1244,
+                    1900,
+                    444,
+                    1900
+                ],
+                "score": 0.964
+            },
+            {
+                "category_id": 4,
+                "poly": [
+                    296,
+                    1924,
+                    1403,
+                    1924,
+                    1403,
+                    1987,
+                    296,
+                    1987
+                ],
+                "score": 0.947
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    835,
+                    2088,
+                    864,
+                    2088,
+                    864,
+                    2113,
+                    835,
+                    2113
+                ],
+                "score": 0.854
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    299,
+                    76,
+                    852,
+                    76,
+                    852,
+                    104,
+                    299,
+                    104
+                ],
+                "score": 0.574
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    298,
+                    76,
+                    854,
+                    76,
+                    854,
+                    105,
+                    298,
+                    105
+                ],
+                "score": 0.255
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    742,
+                    1955,
+                    809,
+                    1955,
+                    809,
+                    1984,
+                    742,
+                    1984
+                ],
+                "score": 0.87,
+                "latex": "100 \\%"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    447.0,
+                    282.0,
+                    477.0,
+                    282.0,
+                    477.0,
+                    373.0,
+                    447.0,
+                    373.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    486.0,
+                    260.0,
+                    529.0,
+                    260.0,
+                    529.0,
+                    326.0,
+                    486.0,
+                    326.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    638.0,
+                    275.0,
+                    679.0,
+                    275.0,
+                    679.0,
+                    302.0,
+                    638.0,
+                    302.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    808.0,
+                    300.0,
+                    833.0,
+                    300.0,
+                    833.0,
+                    324.0,
+                    808.0,
+                    324.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    940.0,
+                    266.0,
+                    998.0,
+                    266.0,
+                    998.0,
+                    291.0,
+                    940.0,
+                    291.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1090.0,
+                    264.0,
+                    1144.0,
+                    264.0,
+                    1144.0,
+                    295.0,
+                    1090.0,
+                    295.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    634.0,
+                    324.0,
+                    679.0,
+                    324.0,
+                    679.0,
+                    351.0,
+                    634.0,
+                    351.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    936.0,
+                    319.0,
+                    981.0,
+                    319.0,
+                    981.0,
+                    346.0,
+                    936.0,
+                    346.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1084.0,
+                    319.0,
+                    1135.0,
+                    319.0,
+                    1135.0,
+                    351.0,
+                    1084.0,
+                    351.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    477.0,
+                    337.0,
+                    527.0,
+                    337.0,
+                    527.0,
+                    362.0,
+                    477.0,
+                    362.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    805.0,
+                    373.0,
+                    823.0,
+                    373.0,
+                    823.0,
+                    390.0,
+                    805.0,
+                    390.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    460.0,
+                    384.0,
+                    524.0,
+                    384.0,
+                    524.0,
+                    515.0,
+                    460.0,
+                    515.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    636.0,
+                    384.0,
+                    677.0,
+                    384.0,
+                    677.0,
+                    411.0,
+                    636.0,
+                    411.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    790.0,
+                    382.0,
+                    829.0,
+                    382.0,
+                    829.0,
+                    411.0,
+                    790.0,
+                    411.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    942.0,
+                    384.0,
+                    983.0,
+                    384.0,
+                    983.0,
+                    411.0,
+                    942.0,
+                    411.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1091.0,
+                    384.0,
+                    1133.0,
+                    384.0,
+                    1133.0,
+                    411.0,
+                    1091.0,
+                    411.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    484.0,
+                    431.0,
+                    529.0,
+                    431.0,
+                    529.0,
+                    466.0,
+                    484.0,
+                    466.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    636.0,
+                    431.0,
+                    681.0,
+                    431.0,
+                    681.0,
+                    466.0,
+                    636.0,
+                    466.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    786.0,
+                    431.0,
+                    831.0,
+                    431.0,
+                    831.0,
+                    464.0,
+                    786.0,
+                    464.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    938.0,
+                    431.0,
+                    983.0,
+                    431.0,
+                    983.0,
+                    466.0,
+                    938.0,
+                    466.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1090.0,
+                    431.0,
+                    1135.0,
+                    431.0,
+                    1135.0,
+                    466.0,
+                    1090.0,
+                    466.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    790.0,
+                    486.0,
+                    831.0,
+                    486.0,
+                    831.0,
+                    513.0,
+                    790.0,
+                    513.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1093.0,
+                    486.0,
+                    1135.0,
+                    486.0,
+                    1135.0,
+                    513.0,
+                    1093.0,
+                    513.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    527.0,
+                    508.0,
+                    550.0,
+                    508.0,
+                    550.0,
+                    530.0,
+                    527.0,
+                    530.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    581.0,
+                    506.0,
+                    640.0,
+                    506.0,
+                    640.0,
+                    531.0,
+                    581.0,
+                    531.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    679.0,
+                    508.0,
+                    702.0,
+                    508.0,
+                    702.0,
+                    530.0,
+                    679.0,
+                    530.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    732.0,
+                    504.0,
+                    791.0,
+                    504.0,
+                    791.0,
+                    535.0,
+                    732.0,
+                    535.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    833.0,
+                    510.0,
+                    850.0,
+                    510.0,
+                    850.0,
+                    528.0,
+                    833.0,
+                    528.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    883.0,
+                    506.0,
+                    942.0,
+                    506.0,
+                    942.0,
+                    531.0,
+                    883.0,
+                    531.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    981.0,
+                    508.0,
+                    1003.0,
+                    508.0,
+                    1003.0,
+                    530.0,
+                    981.0,
+                    530.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1033.0,
+                    502.0,
+                    1095.0,
+                    502.0,
+                    1095.0,
+                    533.0,
+                    1033.0,
+                    533.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1133.0,
+                    506.0,
+                    1155.0,
+                    506.0,
+                    1155.0,
+                    531.0,
+                    1133.0,
+                    531.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1185.0,
+                    502.0,
+                    1245.0,
+                    502.0,
+                    1245.0,
+                    533.0,
+                    1185.0,
+                    533.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    532.0,
+                    525.0,
+                    633.0,
+                    525.0,
+                    633.0,
+                    561.0,
+                    532.0,
+                    561.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    682.0,
+                    525.0,
+                    783.0,
+                    525.0,
+                    783.0,
+                    561.0,
+                    682.0,
+                    561.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    834.0,
+                    525.0,
+                    933.0,
+                    525.0,
+                    933.0,
+                    561.0,
+                    834.0,
+                    561.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    984.0,
+                    525.0,
+                    1085.0,
+                    525.0,
+                    1085.0,
+                    561.0,
+                    984.0,
+                    561.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1136.0,
+                    525.0,
+                    1236.0,
+                    525.0,
+                    1236.0,
+                    561.0,
+                    1136.0,
+                    561.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    583.0,
+                    560.0,
+                    681.0,
+                    560.0,
+                    681.0,
+                    590.0,
+                    583.0,
+                    590.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    756.0,
+                    560.0,
+                    885.0,
+                    560.0,
+                    885.0,
+                    590.0,
+                    756.0,
+                    590.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    957.0,
+                    558.0,
+                    1165.0,
+                    558.0,
+                    1165.0,
+                    592.0,
+                    957.0,
+                    592.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    486.0,
+                    597.0,
+                    546.0,
+                    597.0,
+                    546.0,
+                    628.0,
+                    486.0,
+                    628.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1090.0,
+                    597.0,
+                    1150.0,
+                    597.0,
+                    1150.0,
+                    628.0,
+                    1090.0,
+                    628.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    447.0,
+                    613.0,
+                    477.0,
+                    613.0,
+                    477.0,
+                    706.0,
+                    447.0,
+                    706.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    636.0,
+                    628.0,
+                    681.0,
+                    628.0,
+                    681.0,
+                    661.0,
+                    636.0,
+                    661.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    790.0,
+                    619.0,
+                    829.0,
+                    619.0,
+                    829.0,
+                    648.0,
+                    790.0,
+                    648.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    942.0,
+                    628.0,
+                    985.0,
+                    628.0,
+                    985.0,
+                    655.0,
+                    942.0,
+                    655.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    475.0,
+                    655.0,
+                    525.0,
+                    655.0,
+                    525.0,
+                    681.0,
+                    475.0,
+                    681.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    632.0,
+                    670.0,
+                    677.0,
+                    670.0,
+                    677.0,
+                    697.0,
+                    632.0,
+                    697.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    784.0,
+                    661.0,
+                    831.0,
+                    661.0,
+                    831.0,
+                    693.0,
+                    784.0,
+                    693.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    934.0,
+                    662.0,
+                    983.0,
+                    662.0,
+                    983.0,
+                    695.0,
+                    934.0,
+                    695.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1088.0,
+                    650.0,
+                    1131.0,
+                    650.0,
+                    1131.0,
+                    677.0,
+                    1088.0,
+                    677.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    460.0,
+                    706.0,
+                    527.0,
+                    706.0,
+                    527.0,
+                    841.0,
+                    460.0,
+                    841.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    638.0,
+                    711.0,
+                    677.0,
+                    711.0,
+                    677.0,
+                    741.0,
+                    638.0,
+                    741.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    788.0,
+                    713.0,
+                    829.0,
+                    713.0,
+                    829.0,
+                    741.0,
+                    788.0,
+                    741.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    942.0,
+                    713.0,
+                    983.0,
+                    713.0,
+                    983.0,
+                    741.0,
+                    942.0,
+                    741.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1091.0,
+                    713.0,
+                    1131.0,
+                    713.0,
+                    1131.0,
+                    741.0,
+                    1091.0,
+                    741.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    484.0,
+                    757.0,
+                    529.0,
+                    757.0,
+                    529.0,
+                    791.0,
+                    484.0,
+                    791.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    636.0,
+                    757.0,
+                    681.0,
+                    757.0,
+                    681.0,
+                    790.0,
+                    636.0,
+                    790.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    788.0,
+                    757.0,
+                    833.0,
+                    757.0,
+                    833.0,
+                    790.0,
+                    788.0,
+                    790.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    942.0,
+                    761.0,
+                    985.0,
+                    761.0,
+                    985.0,
+                    788.0,
+                    942.0,
+                    788.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1091.0,
+                    761.0,
+                    1135.0,
+                    761.0,
+                    1135.0,
+                    788.0,
+                    1091.0,
+                    788.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    940.0,
+                    808.0,
+                    998.0,
+                    808.0,
+                    998.0,
+                    839.0,
+                    940.0,
+                    839.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    581.0,
+                    826.0,
+                    640.0,
+                    826.0,
+                    640.0,
+                    857.0,
+                    581.0,
+                    857.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    679.0,
+                    830.0,
+                    700.0,
+                    830.0,
+                    700.0,
+                    853.0,
+                    679.0,
+                    853.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    733.0,
+                    826.0,
+                    793.0,
+                    826.0,
+                    793.0,
+                    857.0,
+                    733.0,
+                    857.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    833.0,
+                    831.0,
+                    848.0,
+                    831.0,
+                    848.0,
+                    851.0,
+                    833.0,
+                    851.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    883.0,
+                    826.0,
+                    943.0,
+                    826.0,
+                    943.0,
+                    857.0,
+                    883.0,
+                    857.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    981.0,
+                    830.0,
+                    1003.0,
+                    830.0,
+                    1003.0,
+                    853.0,
+                    981.0,
+                    853.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1033.0,
+                    826.0,
+                    1095.0,
+                    826.0,
+                    1095.0,
+                    857.0,
+                    1033.0,
+                    857.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1133.0,
+                    830.0,
+                    1153.0,
+                    830.0,
+                    1153.0,
+                    853.0,
+                    1133.0,
+                    853.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1185.0,
+                    826.0,
+                    1245.0,
+                    826.0,
+                    1245.0,
+                    857.0,
+                    1185.0,
+                    857.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    532.0,
+                    847.0,
+                    633.0,
+                    847.0,
+                    633.0,
+                    883.0,
+                    532.0,
+                    883.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    682.0,
+                    846.0,
+                    785.0,
+                    846.0,
+                    785.0,
+                    885.0,
+                    682.0,
+                    885.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    832.0,
+                    847.0,
+                    934.0,
+                    847.0,
+                    934.0,
+                    883.0,
+                    832.0,
+                    883.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    985.0,
+                    851.0,
+                    1084.0,
+                    851.0,
+                    1084.0,
+                    882.0,
+                    985.0,
+                    882.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1135.0,
+                    845.0,
+                    1237.0,
+                    845.0,
+                    1237.0,
+                    888.0,
+                    1135.0,
+                    888.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    583.0,
+                    879.0,
+                    681.0,
+                    879.0,
+                    681.0,
+                    908.0,
+                    583.0,
+                    908.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    756.0,
+                    879.0,
+                    883.0,
+                    879.0,
+                    883.0,
+                    908.0,
+                    756.0,
+                    908.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    955.0,
+                    871.0,
+                    1167.0,
+                    871.0,
+                    1167.0,
+                    914.0,
+                    955.0,
+                    914.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    638.0,
+                    917.0,
+                    679.0,
+                    917.0,
+                    679.0,
+                    944.0,
+                    638.0,
+                    944.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    940.0,
+                    917.0,
+                    1000.0,
+                    917.0,
+                    1000.0,
+                    948.0,
+                    940.0,
+                    948.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1091.0,
+                    917.0,
+                    1138.0,
+                    917.0,
+                    1138.0,
+                    944.0,
+                    1091.0,
+                    944.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    445.0,
+                    933.0,
+                    477.0,
+                    933.0,
+                    477.0,
+                    1026.0,
+                    445.0,
+                    1026.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    486.0,
+                    942.0,
+                    527.0,
+                    942.0,
+                    527.0,
+                    970.0,
+                    486.0,
+                    970.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    808.0,
+                    948.0,
+                    833.0,
+                    948.0,
+                    833.0,
+                    973.0,
+                    808.0,
+                    973.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    475.0,
+                    975.0,
+                    529.0,
+                    975.0,
+                    529.0,
+                    1008.0,
+                    475.0,
+                    1008.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    632.0,
+                    964.0,
+                    679.0,
+                    964.0,
+                    679.0,
+                    997.0,
+                    632.0,
+                    997.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    938.0,
+                    966.0,
+                    979.0,
+                    966.0,
+                    979.0,
+                    993.0,
+                    938.0,
+                    993.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1088.0,
+                    970.0,
+                    1131.0,
+                    970.0,
+                    1131.0,
+                    997.0,
+                    1088.0,
+                    997.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    801.0,
+                    1019.0,
+                    825.0,
+                    1019.0,
+                    825.0,
+                    1039.0,
+                    801.0,
+                    1039.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    458.0,
+                    1030.0,
+                    524.0,
+                    1030.0,
+                    524.0,
+                    1161.0,
+                    458.0,
+                    1161.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    633.0,
+                    1034.0,
+                    674.0,
+                    1034.0,
+                    674.0,
+                    1058.0,
+                    633.0,
+                    1058.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    962.0,
+                    1037.0,
+                    977.0,
+                    1037.0,
+                    977.0,
+                    1055.0,
+                    962.0,
+                    1055.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1091.0,
+                    1033.0,
+                    1131.0,
+                    1033.0,
+                    1131.0,
+                    1059.0,
+                    1091.0,
+                    1059.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    790.0,
+                    1055.0,
+                    829.0,
+                    1055.0,
+                    829.0,
+                    1082.0,
+                    790.0,
+                    1082.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    488.0,
+                    1081.0,
+                    527.0,
+                    1081.0,
+                    527.0,
+                    1110.0,
+                    488.0,
+                    1110.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    636.0,
+                    1079.0,
+                    681.0,
+                    1079.0,
+                    681.0,
+                    1112.0,
+                    636.0,
+                    1112.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1090.0,
+                    1079.0,
+                    1135.0,
+                    1079.0,
+                    1135.0,
+                    1112.0,
+                    1090.0,
+                    1112.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    484.0,
+                    1130.0,
+                    527.0,
+                    1130.0,
+                    527.0,
+                    1157.0,
+                    484.0,
+                    1157.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    638.0,
+                    1130.0,
+                    679.0,
+                    1130.0,
+                    679.0,
+                    1157.0,
+                    638.0,
+                    1157.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    790.0,
+                    1126.0,
+                    829.0,
+                    1126.0,
+                    829.0,
+                    1153.0,
+                    790.0,
+                    1153.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    958.0,
+                    1132.0,
+                    983.0,
+                    1132.0,
+                    983.0,
+                    1155.0,
+                    958.0,
+                    1155.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1093.0,
+                    1130.0,
+                    1136.0,
+                    1130.0,
+                    1136.0,
+                    1157.0,
+                    1093.0,
+                    1157.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    527.0,
+                    1152.0,
+                    550.0,
+                    1152.0,
+                    550.0,
+                    1177.0,
+                    527.0,
+                    1177.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    580.0,
+                    1148.0,
+                    640.0,
+                    1148.0,
+                    640.0,
+                    1179.0,
+                    580.0,
+                    1179.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    679.0,
+                    1152.0,
+                    703.0,
+                    1152.0,
+                    703.0,
+                    1177.0,
+                    679.0,
+                    1177.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    732.0,
+                    1148.0,
+                    793.0,
+                    1148.0,
+                    793.0,
+                    1179.0,
+                    732.0,
+                    1179.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    831.0,
+                    1152.0,
+                    853.0,
+                    1152.0,
+                    853.0,
+                    1177.0,
+                    831.0,
+                    1177.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    882.0,
+                    1148.0,
+                    943.0,
+                    1148.0,
+                    943.0,
+                    1179.0,
+                    882.0,
+                    1179.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    981.0,
+                    1152.0,
+                    1005.0,
+                    1152.0,
+                    1005.0,
+                    1177.0,
+                    981.0,
+                    1177.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1033.0,
+                    1148.0,
+                    1093.0,
+                    1148.0,
+                    1093.0,
+                    1179.0,
+                    1033.0,
+                    1179.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1133.0,
+                    1152.0,
+                    1155.0,
+                    1152.0,
+                    1155.0,
+                    1177.0,
+                    1133.0,
+                    1177.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1185.0,
+                    1148.0,
+                    1245.0,
+                    1148.0,
+                    1245.0,
+                    1179.0,
+                    1185.0,
+                    1179.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    531.0,
+                    1173.0,
+                    632.0,
+                    1173.0,
+                    632.0,
+                    1204.0,
+                    531.0,
+                    1204.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    679.0,
+                    1165.0,
+                    785.0,
+                    1165.0,
+                    785.0,
+                    1209.0,
+                    679.0,
+                    1209.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    831.0,
+                    1165.0,
+                    937.0,
+                    1165.0,
+                    937.0,
+                    1209.0,
+                    831.0,
+                    1209.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    985.0,
+                    1173.0,
+                    1084.0,
+                    1173.0,
+                    1084.0,
+                    1204.0,
+                    985.0,
+                    1204.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1133.0,
+                    1165.0,
+                    1239.0,
+                    1165.0,
+                    1239.0,
+                    1209.0,
+                    1133.0,
+                    1209.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    583.0,
+                    1201.0,
+                    681.0,
+                    1201.0,
+                    681.0,
+                    1232.0,
+                    583.0,
+                    1232.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    754.0,
+                    1197.0,
+                    887.0,
+                    1197.0,
+                    887.0,
+                    1232.0,
+                    754.0,
+                    1232.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    958.0,
+                    1202.0,
+                    1165.0,
+                    1202.0,
+                    1165.0,
+                    1232.0,
+                    958.0,
+                    1232.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    788.0,
+                    1241.0,
+                    840.0,
+                    1241.0,
+                    840.0,
+                    1268.0,
+                    788.0,
+                    1268.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    940.0,
+                    1241.0,
+                    998.0,
+                    1241.0,
+                    998.0,
+                    1266.0,
+                    940.0,
+                    1266.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1091.0,
+                    1237.0,
+                    1131.0,
+                    1237.0,
+                    1131.0,
+                    1264.0,
+                    1091.0,
+                    1264.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    447.0,
+                    1257.0,
+                    477.0,
+                    1257.0,
+                    477.0,
+                    1350.0,
+                    447.0,
+                    1350.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    486.0,
+                    1255.0,
+                    529.0,
+                    1255.0,
+                    529.0,
+                    1283.0,
+                    486.0,
+                    1283.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    656.0,
+                    1263.0,
+                    681.0,
+                    1263.0,
+                    681.0,
+                    1286.0,
+                    656.0,
+                    1286.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1091.0,
+                    1272.0,
+                    1135.0,
+                    1272.0,
+                    1135.0,
+                    1299.0,
+                    1091.0,
+                    1299.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    475.0,
+                    1299.0,
+                    527.0,
+                    1299.0,
+                    527.0,
+                    1324.0,
+                    475.0,
+                    1324.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    784.0,
+                    1290.0,
+                    829.0,
+                    1290.0,
+                    829.0,
+                    1317.0,
+                    784.0,
+                    1317.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    936.0,
+                    1290.0,
+                    983.0,
+                    1290.0,
+                    983.0,
+                    1323.0,
+                    936.0,
+                    1323.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1086.0,
+                    1306.0,
+                    1133.0,
+                    1306.0,
+                    1133.0,
+                    1333.0,
+                    1086.0,
+                    1333.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    458.0,
+                    1353.0,
+                    529.0,
+                    1353.0,
+                    529.0,
+                    1484.0,
+                    458.0,
+                    1484.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    810.0,
+                    1361.0,
+                    825.0,
+                    1361.0,
+                    825.0,
+                    1379.0,
+                    810.0,
+                    1379.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    942.0,
+                    1357.0,
+                    985.0,
+                    1357.0,
+                    985.0,
+                    1383.0,
+                    942.0,
+                    1383.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1091.0,
+                    1357.0,
+                    1135.0,
+                    1357.0,
+                    1135.0,
+                    1383.0,
+                    1091.0,
+                    1383.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    636.0,
+                    1377.0,
+                    681.0,
+                    1377.0,
+                    681.0,
+                    1410.0,
+                    636.0,
+                    1410.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    484.0,
+                    1403.0,
+                    529.0,
+                    1403.0,
+                    529.0,
+                    1435.0,
+                    484.0,
+                    1435.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    940.0,
+                    1403.0,
+                    987.0,
+                    1403.0,
+                    987.0,
+                    1435.0,
+                    940.0,
+                    1435.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1090.0,
+                    1403.0,
+                    1136.0,
+                    1403.0,
+                    1136.0,
+                    1435.0,
+                    1090.0,
+                    1435.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    484.0,
+                    1455.0,
+                    529.0,
+                    1455.0,
+                    529.0,
+                    1483.0,
+                    484.0,
+                    1483.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    638.0,
+                    1450.0,
+                    679.0,
+                    1450.0,
+                    679.0,
+                    1477.0,
+                    638.0,
+                    1477.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    808.0,
+                    1455.0,
+                    833.0,
+                    1455.0,
+                    833.0,
+                    1479.0,
+                    808.0,
+                    1479.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    940.0,
+                    1453.0,
+                    996.0,
+                    1453.0,
+                    996.0,
+                    1484.0,
+                    940.0,
+                    1484.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1093.0,
+                    1455.0,
+                    1135.0,
+                    1455.0,
+                    1135.0,
+                    1483.0,
+                    1093.0,
+                    1483.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    527.0,
+                    1475.0,
+                    550.0,
+                    1475.0,
+                    550.0,
+                    1503.0,
+                    527.0,
+                    1503.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    581.0,
+                    1472.0,
+                    642.0,
+                    1472.0,
+                    642.0,
+                    1504.0,
+                    581.0,
+                    1504.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    679.0,
+                    1477.0,
+                    703.0,
+                    1477.0,
+                    703.0,
+                    1501.0,
+                    679.0,
+                    1501.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    732.0,
+                    1472.0,
+                    793.0,
+                    1472.0,
+                    793.0,
+                    1504.0,
+                    732.0,
+                    1504.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    829.0,
+                    1475.0,
+                    853.0,
+                    1475.0,
+                    853.0,
+                    1501.0,
+                    829.0,
+                    1501.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    882.0,
+                    1472.0,
+                    943.0,
+                    1472.0,
+                    943.0,
+                    1504.0,
+                    882.0,
+                    1504.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    981.0,
+                    1475.0,
+                    1003.0,
+                    1475.0,
+                    1003.0,
+                    1501.0,
+                    981.0,
+                    1501.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1035.0,
+                    1472.0,
+                    1095.0,
+                    1472.0,
+                    1095.0,
+                    1504.0,
+                    1035.0,
+                    1504.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1131.0,
+                    1475.0,
+                    1153.0,
+                    1475.0,
+                    1153.0,
+                    1501.0,
+                    1131.0,
+                    1501.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1183.0,
+                    1472.0,
+                    1245.0,
+                    1472.0,
+                    1245.0,
+                    1504.0,
+                    1183.0,
+                    1504.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    532.0,
+                    1495.0,
+                    633.0,
+                    1495.0,
+                    633.0,
+                    1530.0,
+                    532.0,
+                    1530.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    682.0,
+                    1495.0,
+                    783.0,
+                    1495.0,
+                    783.0,
+                    1530.0,
+                    682.0,
+                    1530.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    832.0,
+                    1495.0,
+                    934.0,
+                    1495.0,
+                    934.0,
+                    1530.0,
+                    832.0,
+                    1530.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    984.0,
+                    1495.0,
+                    1085.0,
+                    1495.0,
+                    1085.0,
+                    1530.0,
+                    984.0,
+                    1530.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1136.0,
+                    1495.0,
+                    1236.0,
+                    1495.0,
+                    1236.0,
+                    1530.0,
+                    1136.0,
+                    1530.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    583.0,
+                    1528.0,
+                    681.0,
+                    1528.0,
+                    681.0,
+                    1557.0,
+                    583.0,
+                    1557.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    754.0,
+                    1528.0,
+                    885.0,
+                    1528.0,
+                    885.0,
+                    1557.0,
+                    754.0,
+                    1557.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    955.0,
+                    1522.0,
+                    1167.0,
+                    1522.0,
+                    1167.0,
+                    1561.0,
+                    955.0,
+                    1561.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    484.0,
+                    1563.0,
+                    546.0,
+                    1563.0,
+                    546.0,
+                    1595.0,
+                    484.0,
+                    1595.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    956.0,
+                    1566.0,
+                    998.0,
+                    1566.0,
+                    998.0,
+                    1594.0,
+                    956.0,
+                    1594.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    447.0,
+                    1583.0,
+                    477.0,
+                    1583.0,
+                    477.0,
+                    1675.0,
+                    447.0,
+                    1675.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    655.0,
+                    1588.0,
+                    688.0,
+                    1588.0,
+                    688.0,
+                    1617.0,
+                    655.0,
+                    1617.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    808.0,
+                    1590.0,
+                    833.0,
+                    1590.0,
+                    833.0,
+                    1614.0,
+                    808.0,
+                    1614.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1110.0,
+                    1590.0,
+                    1135.0,
+                    1590.0,
+                    1135.0,
+                    1615.0,
+                    1110.0,
+                    1615.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    475.0,
+                    1619.0,
+                    527.0,
+                    1619.0,
+                    527.0,
+                    1644.0,
+                    475.0,
+                    1644.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    460.0,
+                    1677.0,
+                    498.0,
+                    1677.0,
+                    498.0,
+                    1810.0,
+                    460.0,
+                    1810.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    645.0,
+                    1669.0,
+                    671.0,
+                    1669.0,
+                    671.0,
+                    1704.0,
+                    645.0,
+                    1704.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    810.0,
+                    1683.0,
+                    825.0,
+                    1683.0,
+                    825.0,
+                    1704.0,
+                    810.0,
+                    1704.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    938.0,
+                    1664.0,
+                    983.0,
+                    1664.0,
+                    983.0,
+                    1712.0,
+                    938.0,
+                    1712.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1110.0,
+                    1668.0,
+                    1131.0,
+                    1668.0,
+                    1131.0,
+                    1708.0,
+                    1110.0,
+                    1708.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    484.0,
+                    1726.0,
+                    529.0,
+                    1726.0,
+                    529.0,
+                    1761.0,
+                    484.0,
+                    1761.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    938.0,
+                    1728.0,
+                    983.0,
+                    1728.0,
+                    983.0,
+                    1761.0,
+                    938.0,
+                    1761.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    655.0,
+                    1779.0,
+                    690.0,
+                    1779.0,
+                    690.0,
+                    1806.0,
+                    655.0,
+                    1806.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    808.0,
+                    1779.0,
+                    833.0,
+                    1779.0,
+                    833.0,
+                    1803.0,
+                    808.0,
+                    1803.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    942.0,
+                    1781.0,
+                    985.0,
+                    1781.0,
+                    985.0,
+                    1806.0,
+                    942.0,
+                    1806.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1110.0,
+                    1781.0,
+                    1136.0,
+                    1781.0,
+                    1136.0,
+                    1805.0,
+                    1110.0,
+                    1805.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    527.0,
+                    1799.0,
+                    550.0,
+                    1799.0,
+                    550.0,
+                    1825.0,
+                    527.0,
+                    1825.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    581.0,
+                    1795.0,
+                    642.0,
+                    1795.0,
+                    642.0,
+                    1828.0,
+                    581.0,
+                    1828.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    679.0,
+                    1801.0,
+                    702.0,
+                    1801.0,
+                    702.0,
+                    1825.0,
+                    679.0,
+                    1825.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    733.0,
+                    1797.0,
+                    793.0,
+                    1797.0,
+                    793.0,
+                    1828.0,
+                    733.0,
+                    1828.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    829.0,
+                    1801.0,
+                    851.0,
+                    1801.0,
+                    851.0,
+                    1825.0,
+                    829.0,
+                    1825.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    883.0,
+                    1797.0,
+                    943.0,
+                    1797.0,
+                    943.0,
+                    1828.0,
+                    883.0,
+                    1828.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    981.0,
+                    1801.0,
+                    1005.0,
+                    1801.0,
+                    1005.0,
+                    1825.0,
+                    981.0,
+                    1825.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1035.0,
+                    1795.0,
+                    1095.0,
+                    1795.0,
+                    1095.0,
+                    1828.0,
+                    1035.0,
+                    1828.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1131.0,
+                    1801.0,
+                    1155.0,
+                    1801.0,
+                    1155.0,
+                    1825.0,
+                    1131.0,
+                    1825.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1185.0,
+                    1795.0,
+                    1247.0,
+                    1795.0,
+                    1247.0,
+                    1828.0,
+                    1185.0,
+                    1828.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    533.0,
+                    1823.0,
+                    632.0,
+                    1823.0,
+                    632.0,
+                    1854.0,
+                    533.0,
+                    1854.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    682.0,
+                    1819.0,
+                    783.0,
+                    1819.0,
+                    783.0,
+                    1854.0,
+                    682.0,
+                    1854.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    834.0,
+                    1819.0,
+                    933.0,
+                    1819.0,
+                    933.0,
+                    1854.0,
+                    834.0,
+                    1854.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    984.0,
+                    1819.0,
+                    1085.0,
+                    1819.0,
+                    1085.0,
+                    1854.0,
+                    984.0,
+                    1854.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1136.0,
+                    1819.0,
+                    1236.0,
+                    1819.0,
+                    1236.0,
+                    1854.0,
+                    1136.0,
+                    1854.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    585.0,
+                    1850.0,
+                    681.0,
+                    1850.0,
+                    681.0,
+                    1881.0,
+                    585.0,
+                    1881.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    754.0,
+                    1852.0,
+                    883.0,
+                    1852.0,
+                    883.0,
+                    1881.0,
+                    754.0,
+                    1881.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    955.0,
+                    1850.0,
+                    1167.0,
+                    1850.0,
+                    1167.0,
+                    1885.0,
+                    955.0,
+                    1885.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    522.0,
+                    829.0,
+                    551.0,
+                    829.0,
+                    551.0,
+                    853.5,
+                    522.0,
+                    853.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    483.0,
+                    1681.0,
+                    530.0,
+                    1681.0,
+                    530.0,
+                    1708.0,
+                    483.0,
+                    1708.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1924.0,
+                    1403.0,
+                    1924.0,
+                    1403.0,
+                    1959.0,
+                    295.0,
+                    1959.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1954.0,
+                    741.0,
+                    1954.0,
+                    741.0,
+                    1989.0,
+                    295.0,
+                    1989.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    810.0,
+                    1954.0,
+                    1129.0,
+                    1954.0,
+                    1129.0,
+                    1989.0,
+                    810.0,
+                    1989.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    831.0,
+                    2085.0,
+                    869.0,
+                    2085.0,
+                    869.0,
+                    2122.0,
+                    831.0,
+                    2122.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    73.0,
+                    856.0,
+                    73.0,
+                    856.0,
+                    108.0,
+                    297.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    72.0,
+                    857.0,
+                    72.0,
+                    857.0,
+                    108.0,
+                    297.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 17,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1757,
+                    1406,
+                    1757,
+                    1406,
+                    2035,
+                    297,
+                    2035
+                ],
+                "score": 0.982
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1066,
+                    1406,
+                    1066,
+                    1406,
+                    1190,
+                    298,
+                    1190
+                ],
+                "score": 0.977
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    867,
+                    1405,
+                    867,
+                    1405,
+                    1052,
+                    297,
+                    1052
+                ],
+                "score": 0.977
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    698,
+                    1404,
+                    698,
+                    1404,
+                    853,
+                    298,
+                    853
+                ],
+                "score": 0.975
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1579,
+                    1402,
+                    1579,
+                    1402,
+                    1673,
+                    298,
+                    1673
+                ],
+                "score": 0.969
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    561,
+                    1338,
+                    1135,
+                    1338,
+                    1135,
+                    1567,
+                    561,
+                    1567
+                ],
+                "score": 0.968
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    655,
+                    1201,
+                    1043,
+                    1201,
+                    1043,
+                    1288,
+                    655,
+                    1288
+                ],
+                "score": 0.957
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    300,
+                    1296,
+                    768,
+                    1296,
+                    768,
+                    1328,
+                    300,
+                    1328
+                ],
+                "score": 0.928
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    339,
+                    821,
+                    339,
+                    821,
+                    373,
+                    298,
+                    373
+                ],
+                "score": 0.92
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    300,
+                    1705,
+                    717,
+                    1705,
+                    717,
+                    1737,
+                    300,
+                    1737
+                ],
+                "score": 0.905
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    354,
+                    396,
+                    1405,
+                    396,
+                    1405,
+                    676,
+                    354,
+                    676
+                ],
+                "score": 0.896
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    836,
+                    2088,
+                    864,
+                    2088,
+                    864,
+                    2112,
+                    836,
+                    2112
+                ],
+                "score": 0.859
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    300,
+                    229,
+                    721,
+                    229,
+                    721,
+                    262,
+                    300,
+                    262
+                ],
+                "score": 0.803
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    298,
+                    75,
+                    854,
+                    75,
+                    854,
+                    105,
+                    298,
+                    105
+                ],
+                "score": 0.79
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    300,
+                    286,
+                    679,
+                    286,
+                    679,
+                    319,
+                    300,
+                    319
+                ],
+                "score": 0.786
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    298,
+                    75,
+                    854,
+                    75,
+                    854,
+                    105,
+                    298,
+                    105
+                ],
+                "score": 0.147
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    300,
+                    286,
+                    679,
+                    286,
+                    679,
+                    319,
+                    300,
+                    319
+                ],
+                "score": 0.135
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    599,
+                    1339,
+                    875,
+                    1339,
+                    875,
+                    1383,
+                    599,
+                    1383
+                ],
+                "score": 0.098
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    300,
+                    229,
+                    721,
+                    229,
+                    721,
+                    262,
+                    300,
+                    262
+                ],
+                "score": 0.091
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    560,
+                    1336,
+                    1138,
+                    1336,
+                    1138,
+                    1568,
+                    560,
+                    1568
+                ],
+                "score": 0.94,
+                "latex": "\\begin{array} { r l r } { \\mathbf { g } _ { x ^ { t } } ^ { t + 1 } = \\mathbf { g } _ { x ^ { t } } ^ { t } + \\alpha _ { 1 } W _ { x ^ { t } , \\hat { y } ^ { t } } ^ { t } } & { } & \\\\ { \\mathbf { W } _ { x ^ { t } , \\hat { y } ^ { t } } ^ { t + 1 / 2 } = W _ { x ^ { t } , \\hat { y } ^ { t } } ^ { t } + \\alpha _ { 2 } ; \\quad \\mathbf { W } _ { x ^ { t } } ^ { t + 1 } = \\frac { \\mathbf { W } _ { x ^ { t } } ^ { t + 1 / 2 } } { \\Vert \\mathbf { W } _ { x ^ { t } } ^ { t + 1 / 2 } \\Vert _ { 2 } } } & { } & \\\\ { x ^ { t + 1 } \\sim \\mathrm { s o f t m a x } ( \\mathbf { g } ^ { t + 1 } ) } & { } & \\\\ { y ^ { t + 1 } \\sim \\mathrm { s o f t m a x } ( \\mathbf { W } _ { x ^ { t + 1 } } ^ { t + 1 } ) } & { } & \\end{array}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    882,
+                    612,
+                    943,
+                    612,
+                    943,
+                    650,
+                    882,
+                    650
+                ],
+                "score": 0.92,
+                "latex": "W _ { x , y } ^ { t }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    879,
+                    2002,
+                    1004,
+                    2002,
+                    1004,
+                    2036,
+                    879,
+                    2036
+                ],
+                "score": 0.91,
+                "latex": "\\mathcal { N } ( 0 , 0 . 0 3 )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1193,
+                    2003,
+                    1326,
+                    2003,
+                    1326,
+                    2034,
+                    1193,
+                    2034
+                ],
+                "score": 0.91,
+                "latex": "\\alpha _ { 2 } = 0 . 0 0 3"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    990,
+                    1910,
+                    1023,
+                    1910,
+                    1023,
+                    1944,
+                    990,
+                    1944
+                ],
+                "score": 0.89,
+                "latex": "\\mathbf { g } ^ { 1 }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    832,
+                    1608,
+                    906,
+                    1608,
+                    906,
+                    1639,
+                    832,
+                    1639
+                ],
+                "score": 0.89,
+                "latex": "\\mathbf { W } ^ { t + 1 }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1016,
+                    2003,
+                    1135,
+                    2003,
+                    1135,
+                    2034,
+                    1016,
+                    2034
+                ],
+                "score": 0.89,
+                "latex": "\\alpha _ { 1 } = 0 . 0 3"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    376,
+                    1579,
+                    403,
+                    1579,
+                    403,
+                    1612,
+                    376,
+                    1612
+                ],
+                "score": 0.88,
+                "latex": "\\hat { y } ^ { t }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    457,
+                    1641,
+                    483,
+                    1641,
+                    483,
+                    1672,
+                    457,
+                    1672
+                ],
+                "score": 0.88,
+                "latex": "\\hat { y } ^ { t }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    549,
+                    539,
+                    577,
+                    539,
+                    577,
+                    572,
+                    549,
+                    572
+                ],
+                "score": 0.87,
+                "latex": "y ^ { t }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    557,
+                    468,
+                    587,
+                    468,
+                    587,
+                    502,
+                    557,
+                    502
+                ],
+                "score": 0.87,
+                "latex": "\\mathbf { g } ^ { t }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    593,
+                    699,
+                    622,
+                    699,
+                    622,
+                    727,
+                    593,
+                    727
+                ],
+                "score": 0.87,
+                "latex": "x ^ { t }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    476,
+                    868,
+                    504,
+                    868,
+                    504,
+                    896,
+                    476,
+                    896
+                ],
+                "score": 0.87,
+                "latex": "x ^ { t }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1343,
+                    468,
+                    1372,
+                    468,
+                    1372,
+                    497,
+                    1343,
+                    497
+                ],
+                "score": 0.86,
+                "latex": "x ^ { t }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1164,
+                    898,
+                    1193,
+                    898,
+                    1193,
+                    926,
+                    1164,
+                    926
+                ],
+                "score": 0.86,
+                "latex": "x ^ { t }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    532,
+                    397,
+                    560,
+                    397,
+                    560,
+                    426,
+                    532,
+                    426
+                ],
+                "score": 0.86,
+                "latex": "x ^ { t }"
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    655,
+                    1198,
+                    1046,
+                    1198,
+                    1046,
+                    1291,
+                    655,
+                    1291
+                ],
+                "score": 0.84,
+                "latex": "\\begin{array} { c } { { \\boldsymbol { s } ^ { t } = ( \\mathbf { g } ^ { t } , \\mathbf { W } ^ { t } , \\boldsymbol { x } ^ { t } , \\boldsymbol { y } ^ { t } ) } } \\\\ { { \\boldsymbol { o } _ { \\mathrm { p r e } } ^ { t } , ~ \\boldsymbol { a } ^ { t } , ~ \\boldsymbol { o } _ { \\mathrm { p o s t } } ^ { t } = ( \\boldsymbol { x } ^ { t } , \\hat { \\boldsymbol { y } } ^ { t } , \\boldsymbol { y } ^ { t } ) } } \\end{array}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1372,
+                    1610,
+                    1400,
+                    1610,
+                    1400,
+                    1638,
+                    1372,
+                    1638
+                ],
+                "score": 0.84,
+                "latex": "x ^ { t }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1073,
+                    1911,
+                    1121,
+                    1911,
+                    1121,
+                    1940,
+                    1073,
+                    1940
+                ],
+                "score": 0.84,
+                "latex": "\\dot { \\mathbf { W } } ^ { 1 }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1222,
+                    499,
+                    1269,
+                    499,
+                    1269,
+                    534,
+                    1222,
+                    534
+                ],
+                "score": 0.83,
+                "latex": "( \\mathbf { g } ^ { t } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    923,
+                    1790,
+                    1003,
+                    1790,
+                    1003,
+                    1820,
+                    923,
+                    1820
+                ],
+                "score": 0.83,
+                "latex": "= 0 . 0 1 "
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    571,
+                    611,
+                    618,
+                    611,
+                    618,
+                    642,
+                    571,
+                    642
+                ],
+                "score": 0.82,
+                "latex": "\\mathbf { W } ^ { t }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    418,
+                    929,
+                    463,
+                    929,
+                    463,
+                    958,
+                    418,
+                    958
+                ],
+                "score": 0.81,
+                "latex": "\\mathbf { W } ^ { t }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    678,
+                    649,
+                    695,
+                    649,
+                    695,
+                    675,
+                    678,
+                    675
+                ],
+                "score": 0.78,
+                "latex": "y"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    990,
+                    1585,
+                    1022,
+                    1585,
+                    1022,
+                    1611,
+                    990,
+                    1611
+                ],
+                "score": 0.76,
+                "latex": "\\alpha _ { 1 }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    397,
+                    649,
+                    416,
+                    649,
+                    416,
+                    670,
+                    397,
+                    670
+                ],
+                "score": 0.73,
+                "latex": "x"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1034,
+                    1586,
+                    1067,
+                    1586,
+                    1067,
+                    1611,
+                    1034,
+                    1611
+                ],
+                "score": 0.72,
+                "latex": "\\alpha _ { 2 }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    496,
+                    703,
+                    510,
+                    703,
+                    510,
+                    728,
+                    496,
+                    728
+                ],
+                "score": 0.68,
+                "latex": "t"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1707.0,
+                    722.0,
+                    1707.0,
+                    722.0,
+                    1740.0,
+                    296.0,
+                    1740.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    831.0,
+                    2085.0,
+                    868.0,
+                    2085.0,
+                    868.0,
+                    2124.0,
+                    831.0,
+                    2124.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    229.0,
+                    723.0,
+                    229.0,
+                    723.0,
+                    265.0,
+                    294.0,
+                    265.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    73.0,
+                    857.0,
+                    73.0,
+                    857.0,
+                    108.0,
+                    297.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    284.0,
+                    681.0,
+                    284.0,
+                    681.0,
+                    322.0,
+                    294.0,
+                    322.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    73.0,
+                    857.0,
+                    73.0,
+                    857.0,
+                    108.0,
+                    297.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1758.0,
+                    1408.0,
+                    1758.0,
+                    1408.0,
+                    1795.0,
+                    295.0,
+                    1795.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1790.0,
+                    922.0,
+                    1790.0,
+                    922.0,
+                    1822.0,
+                    295.0,
+                    1822.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1004.0,
+                    1790.0,
+                    1404.0,
+                    1790.0,
+                    1404.0,
+                    1822.0,
+                    1004.0,
+                    1822.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1816.0,
+                    1407.0,
+                    1816.0,
+                    1407.0,
+                    1859.0,
+                    291.0,
+                    1859.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1850.0,
+                    1407.0,
+                    1850.0,
+                    1407.0,
+                    1885.0,
+                    292.0,
+                    1885.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1883.0,
+                    1404.0,
+                    1883.0,
+                    1404.0,
+                    1916.0,
+                    295.0,
+                    1916.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1908.0,
+                    989.0,
+                    1908.0,
+                    989.0,
+                    1949.0,
+                    292.0,
+                    1949.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1024.0,
+                    1908.0,
+                    1072.0,
+                    1908.0,
+                    1072.0,
+                    1949.0,
+                    1024.0,
+                    1949.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1122.0,
+                    1908.0,
+                    1407.0,
+                    1908.0,
+                    1407.0,
+                    1949.0,
+                    1122.0,
+                    1949.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1939.0,
+                    1406.0,
+                    1939.0,
+                    1406.0,
+                    1979.0,
+                    294.0,
+                    1979.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1969.0,
+                    1406.0,
+                    1969.0,
+                    1406.0,
+                    2009.0,
+                    294.0,
+                    2009.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    2000.0,
+                    878.0,
+                    2000.0,
+                    878.0,
+                    2039.0,
+                    292.0,
+                    2039.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1005.0,
+                    2000.0,
+                    1015.0,
+                    2000.0,
+                    1015.0,
+                    2039.0,
+                    1005.0,
+                    2039.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1136.0,
+                    2000.0,
+                    1192.0,
+                    2000.0,
+                    1192.0,
+                    2039.0,
+                    1136.0,
+                    2039.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1327.0,
+                    2000.0,
+                    1336.0,
+                    2000.0,
+                    1336.0,
+                    2039.0,
+                    1327.0,
+                    2039.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1067.0,
+                    1408.0,
+                    1067.0,
+                    1408.0,
+                    1100.0,
+                    295.0,
+                    1100.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1099.0,
+                    1404.0,
+                    1099.0,
+                    1404.0,
+                    1132.0,
+                    295.0,
+                    1132.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1126.0,
+                    1411.0,
+                    1126.0,
+                    1411.0,
+                    1165.0,
+                    292.0,
+                    1165.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1158.0,
+                    853.0,
+                    1158.0,
+                    853.0,
+                    1193.0,
+                    293.0,
+                    1193.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    861.0,
+                    475.0,
+                    861.0,
+                    475.0,
+                    907.0,
+                    294.0,
+                    907.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    505.0,
+                    861.0,
+                    1407.0,
+                    861.0,
+                    1407.0,
+                    907.0,
+                    505.0,
+                    907.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    896.0,
+                    1163.0,
+                    896.0,
+                    1163.0,
+                    935.0,
+                    291.0,
+                    935.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1194.0,
+                    896.0,
+                    1408.0,
+                    896.0,
+                    1408.0,
+                    935.0,
+                    1194.0,
+                    935.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    925.0,
+                    417.0,
+                    925.0,
+                    417.0,
+                    966.0,
+                    292.0,
+                    966.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    464.0,
+                    925.0,
+                    1408.0,
+                    925.0,
+                    1408.0,
+                    966.0,
+                    464.0,
+                    966.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    961.0,
+                    1406.0,
+                    961.0,
+                    1406.0,
+                    993.0,
+                    296.0,
+                    993.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    991.0,
+                    1406.0,
+                    991.0,
+                    1406.0,
+                    1023.0,
+                    294.0,
+                    1023.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1018.0,
+                    1202.0,
+                    1018.0,
+                    1202.0,
+                    1054.0,
+                    294.0,
+                    1054.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    700.0,
+                    495.0,
+                    700.0,
+                    495.0,
+                    733.0,
+                    296.0,
+                    733.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    511.0,
+                    700.0,
+                    592.0,
+                    700.0,
+                    592.0,
+                    733.0,
+                    511.0,
+                    733.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    623.0,
+                    700.0,
+                    1404.0,
+                    700.0,
+                    1404.0,
+                    733.0,
+                    623.0,
+                    733.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    730.0,
+                    1404.0,
+                    730.0,
+                    1404.0,
+                    764.0,
+                    296.0,
+                    764.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    758.0,
+                    1407.0,
+                    758.0,
+                    1407.0,
+                    797.0,
+                    292.0,
+                    797.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    787.0,
+                    1407.0,
+                    787.0,
+                    1407.0,
+                    826.0,
+                    292.0,
+                    826.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    822.0,
+                    756.0,
+                    822.0,
+                    756.0,
+                    855.0,
+                    296.0,
+                    855.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1576.0,
+                    375.0,
+                    1576.0,
+                    375.0,
+                    1615.0,
+                    293.0,
+                    1615.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    404.0,
+                    1576.0,
+                    989.0,
+                    1576.0,
+                    989.0,
+                    1615.0,
+                    404.0,
+                    1615.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1023.0,
+                    1576.0,
+                    1033.0,
+                    1576.0,
+                    1033.0,
+                    1615.0,
+                    1023.0,
+                    1615.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1068.0,
+                    1576.0,
+                    1406.0,
+                    1576.0,
+                    1406.0,
+                    1615.0,
+                    1068.0,
+                    1615.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1605.0,
+                    831.0,
+                    1605.0,
+                    831.0,
+                    1647.0,
+                    293.0,
+                    1647.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    907.0,
+                    1605.0,
+                    1371.0,
+                    1605.0,
+                    1371.0,
+                    1647.0,
+                    907.0,
+                    1647.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1401.0,
+                    1605.0,
+                    1406.0,
+                    1605.0,
+                    1406.0,
+                    1647.0,
+                    1401.0,
+                    1647.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1638.0,
+                    456.0,
+                    1638.0,
+                    456.0,
+                    1677.0,
+                    292.0,
+                    1677.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    484.0,
+                    1638.0,
+                    1019.0,
+                    1638.0,
+                    1019.0,
+                    1677.0,
+                    484.0,
+                    1677.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1290.0,
+                    771.0,
+                    1290.0,
+                    771.0,
+                    1334.0,
+                    294.0,
+                    1334.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    334.0,
+                    826.0,
+                    334.0,
+                    826.0,
+                    379.0,
+                    293.0,
+                    379.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    363.0,
+                    397.0,
+                    531.0,
+                    397.0,
+                    531.0,
+                    430.0,
+                    363.0,
+                    430.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    561.0,
+                    397.0,
+                    1403.0,
+                    397.0,
+                    1403.0,
+                    430.0,
+                    561.0,
+                    430.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    394.0,
+                    429.0,
+                    1185.0,
+                    429.0,
+                    1185.0,
+                    462.0,
+                    394.0,
+                    462.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    357.0,
+                    466.0,
+                    556.0,
+                    466.0,
+                    556.0,
+                    506.0,
+                    357.0,
+                    506.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    588.0,
+                    466.0,
+                    1342.0,
+                    466.0,
+                    1342.0,
+                    506.0,
+                    588.0,
+                    506.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1373.0,
+                    466.0,
+                    1407.0,
+                    466.0,
+                    1407.0,
+                    506.0,
+                    1373.0,
+                    506.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    392.0,
+                    497.0,
+                    1221.0,
+                    497.0,
+                    1221.0,
+                    537.0,
+                    392.0,
+                    537.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1270.0,
+                    497.0,
+                    1282.0,
+                    497.0,
+                    1282.0,
+                    537.0,
+                    1270.0,
+                    537.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    357.0,
+                    537.0,
+                    548.0,
+                    537.0,
+                    548.0,
+                    576.0,
+                    357.0,
+                    576.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    578.0,
+                    537.0,
+                    1405.0,
+                    537.0,
+                    1405.0,
+                    576.0,
+                    578.0,
+                    576.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    395.0,
+                    573.0,
+                    627.0,
+                    573.0,
+                    627.0,
+                    606.0,
+                    395.0,
+                    606.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    351.0,
+                    601.0,
+                    570.0,
+                    601.0,
+                    570.0,
+                    661.0,
+                    351.0,
+                    661.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    619.0,
+                    601.0,
+                    881.0,
+                    601.0,
+                    881.0,
+                    661.0,
+                    619.0,
+                    661.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    944.0,
+                    601.0,
+                    1411.0,
+                    601.0,
+                    1411.0,
+                    661.0,
+                    944.0,
+                    661.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    392.0,
+                    637.0,
+                    396.0,
+                    637.0,
+                    396.0,
+                    681.0,
+                    392.0,
+                    681.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    417.0,
+                    637.0,
+                    677.0,
+                    637.0,
+                    677.0,
+                    681.0,
+                    417.0,
+                    681.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    696.0,
+                    637.0,
+                    707.0,
+                    637.0,
+                    707.0,
+                    681.0,
+                    696.0,
+                    681.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    284.0,
+                    681.0,
+                    284.0,
+                    681.0,
+                    322.0,
+                    294.0,
+                    322.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    229.0,
+                    723.0,
+                    229.0,
+                    723.0,
+                    265.0,
+                    294.0,
+                    265.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 18,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    281,
+                    1405,
+                    281,
+                    1405,
+                    527,
+                    297,
+                    527
+                ],
+                "score": 0.981
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1417,
+                    1405,
+                    1417,
+                    1405,
+                    1664,
+                    297,
+                    1664
+                ],
+                "score": 0.98
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1177,
+                    1405,
+                    1177,
+                    1405,
+                    1330,
+                    298,
+                    1330
+                ],
+                "score": 0.978
+            },
+            {
+                "category_id": 3,
+                "poly": [
+                    405,
+                    566,
+                    1259,
+                    566,
+                    1259,
+                    945,
+                    405,
+                    945
+                ],
+                "score": 0.971
+            },
+            {
+                "category_id": 4,
+                "poly": [
+                    297,
+                    980,
+                    1402,
+                    980,
+                    1402,
+                    1073,
+                    297,
+                    1073
+                ],
+                "score": 0.956
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    300,
+                    230,
+                    690,
+                    230,
+                    690,
+                    261,
+                    300,
+                    261
+                ],
+                "score": 0.878
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    298,
+                    75,
+                    854,
+                    75,
+                    854,
+                    104,
+                    298,
+                    104
+                ],
+                "score": 0.876
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    298,
+                    1126,
+                    1021,
+                    1126,
+                    1021,
+                    1155,
+                    298,
+                    1155
+                ],
+                "score": 0.854
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    835,
+                    2088,
+                    866,
+                    2088,
+                    866,
+                    2113,
+                    835,
+                    2113
+                ],
+                "score": 0.852
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    297,
+                    1365,
+                    949,
+                    1365,
+                    949,
+                    1395,
+                    297,
+                    1395
+                ],
+                "score": 0.654
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    297,
+                    1366,
+                    950,
+                    1366,
+                    950,
+                    1395,
+                    297,
+                    1395
+                ],
+                "score": 0.325
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    811,
+                    1600,
+                    895,
+                    1600,
+                    895,
+                    1634,
+                    811,
+                    1634
+                ],
+                "score": 0.92,
+                "latex": "P ( \\boldsymbol { y } | \\boldsymbol { x } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    708,
+                    283,
+                    796,
+                    283,
+                    796,
+                    317,
+                    708,
+                    317
+                ],
+                "score": 0.9,
+                "latex": "P ( \\boldsymbol { y } | \\mathbf { x } ) ."
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1125,
+                    346,
+                    1155,
+                    346,
+                    1155,
+                    377,
+                    1125,
+                    377
+                ],
+                "score": 0.88,
+                "latex": "\\mathbf { g } ^ { t }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1042,
+                    344,
+                    1075,
+                    344,
+                    1075,
+                    377,
+                    1042,
+                    377
+                ],
+                "score": 0.88,
+                "latex": "\\mathbf { g } ^ { 1 }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1344,
+                    436,
+                    1398,
+                    436,
+                    1398,
+                    466,
+                    1344,
+                    466
+                ],
+                "score": 0.88,
+                "latex": "60 \\%"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1126,
+                    1420,
+                    1226,
+                    1420,
+                    1226,
+                    1450,
+                    1126,
+                    1450
+                ],
+                "score": 0.87,
+                "latex": "( \\alpha _ { 1 } , \\alpha _ { 2 } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    810,
+                    1042,
+                    890,
+                    1042,
+                    890,
+                    1071,
+                    810,
+                    1071
+                ],
+                "score": 0.87,
+                "latex": "> 6 0 \\%"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1086,
+                    314,
+                    1154,
+                    314,
+                    1154,
+                    346,
+                    1086,
+                    346
+                ],
+                "score": 0.86,
+                "latex": "P ( \\mathbf { x } ) )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    551,
+                    1511,
+                    600,
+                    1511,
+                    600,
+                    1541,
+                    551,
+                    1541
+                ],
+                "score": 0.81,
+                "latex": "( \\alpha _ { 1 } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    990,
+                    1511,
+                    1039,
+                    1511,
+                    1039,
+                    1541,
+                    990,
+                    1541
+                ],
+                "score": 0.78,
+                "latex": "\\left( \\alpha _ { 2 } \\right)"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    422.0,
+                    564.0,
+                    476.0,
+                    564.0,
+                    476.0,
+                    588.0,
+                    422.0,
+                    588.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    934.0,
+                    580.0,
+                    992.0,
+                    580.0,
+                    992.0,
+                    604.0,
+                    934.0,
+                    604.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    422.0,
+                    605.0,
+                    478.0,
+                    605.0,
+                    478.0,
+                    629.0,
+                    422.0,
+                    629.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    935.0,
+                    598.0,
+                    974.0,
+                    598.0,
+                    974.0,
+                    623.0,
+                    935.0,
+                    623.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    856.0,
+                    620.0,
+                    900.0,
+                    620.0,
+                    900.0,
+                    645.0,
+                    856.0,
+                    645.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    422.0,
+                    646.0,
+                    478.0,
+                    646.0,
+                    478.0,
+                    670.0,
+                    422.0,
+                    670.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    402.0,
+                    687.0,
+                    479.0,
+                    687.0,
+                    479.0,
+                    775.0,
+                    402.0,
+                    775.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    835.0,
+                    686.0,
+                    897.0,
+                    686.0,
+                    897.0,
+                    793.0,
+                    835.0,
+                    793.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    858.0,
+                    752.0,
+                    895.0,
+                    752.0,
+                    895.0,
+                    773.0,
+                    858.0,
+                    773.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    420.0,
+                    767.0,
+                    479.0,
+                    767.0,
+                    479.0,
+                    794.0,
+                    420.0,
+                    794.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    422.0,
+                    810.0,
+                    478.0,
+                    810.0,
+                    478.0,
+                    834.0,
+                    422.0,
+                    834.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    856.0,
+                    813.0,
+                    897.0,
+                    813.0,
+                    897.0,
+                    838.0,
+                    856.0,
+                    838.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    421.0,
+                    849.0,
+                    478.0,
+                    849.0,
+                    478.0,
+                    876.0,
+                    421.0,
+                    876.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    856.0,
+                    878.0,
+                    902.0,
+                    878.0,
+                    902.0,
+                    903.0,
+                    856.0,
+                    903.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    420.0,
+                    890.0,
+                    480.0,
+                    890.0,
+                    480.0,
+                    917.0,
+                    420.0,
+                    917.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    463.0,
+                    904.0,
+                    496.0,
+                    904.0,
+                    496.0,
+                    930.0,
+                    463.0,
+                    930.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    532.0,
+                    904.0,
+                    565.0,
+                    904.0,
+                    565.0,
+                    930.0,
+                    532.0,
+                    930.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    602.0,
+                    904.0,
+                    635.0,
+                    904.0,
+                    635.0,
+                    930.0,
+                    602.0,
+                    930.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    672.0,
+                    904.0,
+                    706.0,
+                    904.0,
+                    706.0,
+                    930.0,
+                    672.0,
+                    930.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    742.0,
+                    904.0,
+                    774.0,
+                    904.0,
+                    774.0,
+                    931.0,
+                    742.0,
+                    931.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    812.0,
+                    906.0,
+                    843.0,
+                    906.0,
+                    843.0,
+                    929.0,
+                    812.0,
+                    929.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    881.0,
+                    904.0,
+                    914.0,
+                    904.0,
+                    914.0,
+                    931.0,
+                    881.0,
+                    931.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    951.0,
+                    904.0,
+                    984.0,
+                    904.0,
+                    984.0,
+                    930.0,
+                    951.0,
+                    930.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1021.0,
+                    904.0,
+                    1054.0,
+                    904.0,
+                    1054.0,
+                    930.0,
+                    1021.0,
+                    930.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1090.0,
+                    904.0,
+                    1124.0,
+                    904.0,
+                    1124.0,
+                    930.0,
+                    1090.0,
+                    930.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1160.0,
+                    904.0,
+                    1193.0,
+                    904.0,
+                    1193.0,
+                    931.0,
+                    1160.0,
+                    931.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1231.0,
+                    906.0,
+                    1261.0,
+                    906.0,
+                    1261.0,
+                    929.0,
+                    1231.0,
+                    929.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    620.0,
+                    921.0,
+                    687.0,
+                    921.0,
+                    687.0,
+                    948.0,
+                    620.0,
+                    948.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1038.0,
+                    922.0,
+                    1106.0,
+                    922.0,
+                    1106.0,
+                    948.0,
+                    1038.0,
+                    948.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    978.0,
+                    1407.0,
+                    978.0,
+                    1407.0,
+                    1017.0,
+                    293.0,
+                    1017.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1010.0,
+                    1404.0,
+                    1010.0,
+                    1404.0,
+                    1048.0,
+                    292.0,
+                    1048.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1042.0,
+                    809.0,
+                    1042.0,
+                    809.0,
+                    1076.0,
+                    296.0,
+                    1076.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    891.0,
+                    1042.0,
+                    1093.0,
+                    1042.0,
+                    1093.0,
+                    1076.0,
+                    891.0,
+                    1076.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    227.0,
+                    695.0,
+                    227.0,
+                    695.0,
+                    265.0,
+                    294.0,
+                    265.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    72.0,
+                    857.0,
+                    72.0,
+                    857.0,
+                    108.0,
+                    297.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1122.0,
+                    1025.0,
+                    1122.0,
+                    1025.0,
+                    1159.0,
+                    295.0,
+                    1159.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    829.0,
+                    2083.0,
+                    872.0,
+                    2083.0,
+                    872.0,
+                    2124.0,
+                    829.0,
+                    2124.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1364.0,
+                    952.0,
+                    1364.0,
+                    952.0,
+                    1399.0,
+                    296.0,
+                    1399.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1363.0,
+                    951.0,
+                    1363.0,
+                    951.0,
+                    1399.0,
+                    295.0,
+                    1399.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    283.0,
+                    707.0,
+                    283.0,
+                    707.0,
+                    320.0,
+                    294.0,
+                    320.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    797.0,
+                    283.0,
+                    1406.0,
+                    283.0,
+                    1406.0,
+                    320.0,
+                    797.0,
+                    320.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    314.0,
+                    1085.0,
+                    314.0,
+                    1085.0,
+                    350.0,
+                    292.0,
+                    350.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1155.0,
+                    314.0,
+                    1405.0,
+                    314.0,
+                    1405.0,
+                    350.0,
+                    1155.0,
+                    350.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    344.0,
+                    1041.0,
+                    344.0,
+                    1041.0,
+                    381.0,
+                    294.0,
+                    381.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1076.0,
+                    344.0,
+                    1124.0,
+                    344.0,
+                    1124.0,
+                    381.0,
+                    1076.0,
+                    381.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1156.0,
+                    344.0,
+                    1406.0,
+                    344.0,
+                    1406.0,
+                    381.0,
+                    1156.0,
+                    381.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    376.0,
+                    1406.0,
+                    376.0,
+                    1406.0,
+                    410.0,
+                    295.0,
+                    410.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    406.0,
+                    1405.0,
+                    406.0,
+                    1405.0,
+                    440.0,
+                    295.0,
+                    440.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    435.0,
+                    1343.0,
+                    435.0,
+                    1343.0,
+                    472.0,
+                    294.0,
+                    472.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1399.0,
+                    435.0,
+                    1408.0,
+                    435.0,
+                    1408.0,
+                    472.0,
+                    1399.0,
+                    472.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    464.0,
+                    1406.0,
+                    464.0,
+                    1406.0,
+                    504.0,
+                    292.0,
+                    504.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    494.0,
+                    467.0,
+                    494.0,
+                    467.0,
+                    533.0,
+                    293.0,
+                    533.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1416.0,
+                    1125.0,
+                    1416.0,
+                    1125.0,
+                    1453.0,
+                    294.0,
+                    1453.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1227.0,
+                    1416.0,
+                    1406.0,
+                    1416.0,
+                    1406.0,
+                    1453.0,
+                    1227.0,
+                    1453.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1449.0,
+                    1406.0,
+                    1449.0,
+                    1406.0,
+                    1483.0,
+                    294.0,
+                    1483.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1479.0,
+                    1406.0,
+                    1479.0,
+                    1406.0,
+                    1515.0,
+                    292.0,
+                    1515.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1510.0,
+                    550.0,
+                    1510.0,
+                    550.0,
+                    1544.0,
+                    294.0,
+                    1544.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    601.0,
+                    1510.0,
+                    989.0,
+                    1510.0,
+                    989.0,
+                    1544.0,
+                    601.0,
+                    1544.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1040.0,
+                    1510.0,
+                    1406.0,
+                    1510.0,
+                    1406.0,
+                    1544.0,
+                    1040.0,
+                    1544.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1538.0,
+                    1406.0,
+                    1538.0,
+                    1406.0,
+                    1575.0,
+                    294.0,
+                    1575.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1568.0,
+                    1407.0,
+                    1568.0,
+                    1407.0,
+                    1607.0,
+                    292.0,
+                    1607.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1600.0,
+                    810.0,
+                    1600.0,
+                    810.0,
+                    1637.0,
+                    294.0,
+                    1637.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    896.0,
+                    1600.0,
+                    1406.0,
+                    1600.0,
+                    1406.0,
+                    1637.0,
+                    896.0,
+                    1637.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1632.0,
+                    886.0,
+                    1632.0,
+                    886.0,
+                    1666.0,
+                    294.0,
+                    1666.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1174.0,
+                    1406.0,
+                    1174.0,
+                    1406.0,
+                    1213.0,
+                    295.0,
+                    1213.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1205.0,
+                    1405.0,
+                    1205.0,
+                    1405.0,
+                    1244.0,
+                    293.0,
+                    1244.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1238.0,
+                    1403.0,
+                    1238.0,
+                    1403.0,
+                    1273.0,
+                    292.0,
+                    1273.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1265.0,
+                    1406.0,
+                    1265.0,
+                    1406.0,
+                    1305.0,
+                    292.0,
+                    1305.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1298.0,
+                    765.0,
+                    1298.0,
+                    765.0,
+                    1331.0,
+                    295.0,
+                    1331.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 19,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 3,
+                "poly": [
+                    343,
+                    715,
+                    1356,
+                    715,
+                    1356,
+                    1455,
+                    343,
+                    1455
+                ],
+                "score": 0.976
+            },
+            {
+                "category_id": 4,
+                "poly": [
+                    292,
+                    1493,
+                    1405,
+                    1493,
+                    1405,
+                    1558,
+                    292,
+                    1558
+                ],
+                "score": 0.937
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    834,
+                    2088,
+                    863,
+                    2088,
+                    863,
+                    2113,
+                    834,
+                    2113
+                ],
+                "score": 0.82
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    297,
+                    75,
+                    855,
+                    75,
+                    855,
+                    105,
+                    297,
+                    105
+                ],
+                "score": 0.682
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    297,
+                    75,
+                    855,
+                    75,
+                    855,
+                    106,
+                    297,
+                    106
+                ],
+                "score": 0.238
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    404.0,
+                    712.0,
+                    654.0,
+                    712.0,
+                    654.0,
+                    740.0,
+                    404.0,
+                    740.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    719.0,
+                    711.0,
+                    1011.0,
+                    711.0,
+                    1011.0,
+                    742.0,
+                    719.0,
+                    742.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1084.0,
+                    712.0,
+                    1323.0,
+                    712.0,
+                    1323.0,
+                    740.0,
+                    1084.0,
+                    740.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1035.0,
+                    730.0,
+                    1072.0,
+                    730.0,
+                    1072.0,
+                    756.0,
+                    1035.0,
+                    756.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    671.0,
+                    733.0,
+                    698.0,
+                    733.0,
+                    698.0,
+                    911.0,
+                    671.0,
+                    911.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    359.0,
+                    739.0,
+                    400.0,
+                    739.0,
+                    400.0,
+                    769.0,
+                    359.0,
+                    769.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1015.0,
+                    754.0,
+                    1046.0,
+                    754.0,
+                    1046.0,
+                    1003.0,
+                    1015.0,
+                    1003.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    691.0,
+                    767.0,
+                    736.0,
+                    767.0,
+                    736.0,
+                    793.0,
+                    691.0,
+                    793.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1035.0,
+                    779.0,
+                    1072.0,
+                    779.0,
+                    1072.0,
+                    810.0,
+                    1035.0,
+                    810.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    360.0,
+                    792.0,
+                    402.0,
+                    792.0,
+                    402.0,
+                    823.0,
+                    360.0,
+                    823.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    691.0,
+                    812.0,
+                    736.0,
+                    812.0,
+                    736.0,
+                    838.0,
+                    691.0,
+                    838.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    344.0,
+                    839.0,
+                    399.0,
+                    839.0,
+                    399.0,
+                    920.0,
+                    344.0,
+                    920.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1038.0,
+                    835.0,
+                    1067.0,
+                    835.0,
+                    1067.0,
+                    858.0,
+                    1038.0,
+                    858.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    688.0,
+                    856.0,
+                    736.0,
+                    856.0,
+                    736.0,
+                    883.0,
+                    688.0,
+                    883.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1036.0,
+                    885.0,
+                    1069.0,
+                    885.0,
+                    1069.0,
+                    912.0,
+                    1036.0,
+                    912.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    671.0,
+                    896.0,
+                    697.0,
+                    896.0,
+                    697.0,
+                    1021.0,
+                    671.0,
+                    1021.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    360.0,
+                    898.0,
+                    396.0,
+                    898.0,
+                    396.0,
+                    923.0,
+                    360.0,
+                    923.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    690.0,
+                    901.0,
+                    735.0,
+                    901.0,
+                    735.0,
+                    927.0,
+                    690.0,
+                    927.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    360.0,
+                    948.0,
+                    400.0,
+                    948.0,
+                    400.0,
+                    978.0,
+                    360.0,
+                    978.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    691.0,
+                    946.0,
+                    732.0,
+                    946.0,
+                    732.0,
+                    971.0,
+                    691.0,
+                    971.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1037.0,
+                    938.0,
+                    1070.0,
+                    938.0,
+                    1070.0,
+                    963.0,
+                    1037.0,
+                    963.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1118.0,
+                    972.0,
+                    1127.0,
+                    972.0,
+                    1127.0,
+                    983.0,
+                    1118.0,
+                    983.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    691.0,
+                    990.0,
+                    734.0,
+                    990.0,
+                    734.0,
+                    1014.0,
+                    691.0,
+                    1014.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1038.0,
+                    988.0,
+                    1072.0,
+                    988.0,
+                    1072.0,
+                    1017.0,
+                    1038.0,
+                    1017.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1079.0,
+                    995.0,
+                    1094.0,
+                    995.0,
+                    1094.0,
+                    1009.0,
+                    1079.0,
+                    1009.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    361.0,
+                    1002.0,
+                    397.0,
+                    1002.0,
+                    397.0,
+                    1027.0,
+                    361.0,
+                    1027.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    403.0,
+                    1020.0,
+                    422.0,
+                    1020.0,
+                    422.0,
+                    1042.0,
+                    403.0,
+                    1042.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    441.0,
+                    1018.0,
+                    482.0,
+                    1018.0,
+                    482.0,
+                    1043.0,
+                    441.0,
+                    1043.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    489.0,
+                    1017.0,
+                    683.0,
+                    1017.0,
+                    683.0,
+                    1045.0,
+                    489.0,
+                    1045.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    739.0,
+                    1020.0,
+                    759.0,
+                    1020.0,
+                    759.0,
+                    1042.0,
+                    739.0,
+                    1042.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    779.0,
+                    1018.0,
+                    818.0,
+                    1018.0,
+                    818.0,
+                    1043.0,
+                    779.0,
+                    1043.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    825.0,
+                    1017.0,
+                    1020.0,
+                    1017.0,
+                    1020.0,
+                    1045.0,
+                    825.0,
+                    1045.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1077.0,
+                    1021.0,
+                    1095.0,
+                    1021.0,
+                    1095.0,
+                    1040.0,
+                    1077.0,
+                    1040.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1115.0,
+                    1019.0,
+                    1156.0,
+                    1019.0,
+                    1156.0,
+                    1043.0,
+                    1115.0,
+                    1043.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1163.0,
+                    1017.0,
+                    1357.0,
+                    1017.0,
+                    1357.0,
+                    1045.0,
+                    1163.0,
+                    1045.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    488.0,
+                    1037.0,
+                    569.0,
+                    1037.0,
+                    569.0,
+                    1066.0,
+                    488.0,
+                    1066.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    826.0,
+                    1036.0,
+                    907.0,
+                    1036.0,
+                    907.0,
+                    1066.0,
+                    826.0,
+                    1066.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1162.0,
+                    1037.0,
+                    1243.0,
+                    1037.0,
+                    1243.0,
+                    1066.0,
+                    1162.0,
+                    1066.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    538.0,
+                    1063.0,
+                    610.0,
+                    1063.0,
+                    610.0,
+                    1092.0,
+                    538.0,
+                    1092.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    636.0,
+                    1063.0,
+                    728.0,
+                    1063.0,
+                    728.0,
+                    1091.0,
+                    636.0,
+                    1091.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    757.0,
+                    1070.0,
+                    781.0,
+                    1070.0,
+                    781.0,
+                    1087.0,
+                    757.0,
+                    1087.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    782.0,
+                    1061.0,
+                    939.0,
+                    1061.0,
+                    939.0,
+                    1094.0,
+                    782.0,
+                    1094.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    970.0,
+                    1074.0,
+                    982.0,
+                    1074.0,
+                    982.0,
+                    1084.0,
+                    970.0,
+                    1084.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    991.0,
+                    1062.0,
+                    1172.0,
+                    1062.0,
+                    1172.0,
+                    1095.0,
+                    991.0,
+                    1095.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    824.0,
+                    1101.0,
+                    856.0,
+                    1101.0,
+                    856.0,
+                    1415.0,
+                    824.0,
+                    1415.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    855.0,
+                    1110.0,
+                    1170.0,
+                    1110.0,
+                    1170.0,
+                    1143.0,
+                    855.0,
+                    1143.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    565.0,
+                    1114.0,
+                    818.0,
+                    1114.0,
+                    818.0,
+                    1142.0,
+                    565.0,
+                    1142.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    493.0,
+                    1125.0,
+                    526.0,
+                    1125.0,
+                    526.0,
+                    1389.0,
+                    493.0,
+                    1389.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    846.0,
+                    1143.0,
+                    894.0,
+                    1143.0,
+                    894.0,
+                    1172.0,
+                    846.0,
+                    1172.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    513.0,
+                    1166.0,
+                    570.0,
+                    1166.0,
+                    570.0,
+                    1193.0,
+                    513.0,
+                    1193.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    847.0,
+                    1187.0,
+                    895.0,
+                    1187.0,
+                    895.0,
+                    1211.0,
+                    847.0,
+                    1211.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    513.0,
+                    1204.0,
+                    571.0,
+                    1204.0,
+                    571.0,
+                    1231.0,
+                    513.0,
+                    1231.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    846.0,
+                    1225.0,
+                    895.0,
+                    1225.0,
+                    895.0,
+                    1253.0,
+                    846.0,
+                    1253.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    513.0,
+                    1244.0,
+                    571.0,
+                    1244.0,
+                    571.0,
+                    1271.0,
+                    513.0,
+                    1271.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    846.0,
+                    1266.0,
+                    894.0,
+                    1266.0,
+                    894.0,
+                    1294.0,
+                    846.0,
+                    1294.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    511.0,
+                    1280.0,
+                    572.0,
+                    1280.0,
+                    572.0,
+                    1311.0,
+                    511.0,
+                    1311.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    846.0,
+                    1306.0,
+                    894.0,
+                    1306.0,
+                    894.0,
+                    1334.0,
+                    846.0,
+                    1334.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    514.0,
+                    1324.0,
+                    569.0,
+                    1324.0,
+                    569.0,
+                    1347.0,
+                    514.0,
+                    1347.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    846.0,
+                    1347.0,
+                    894.0,
+                    1347.0,
+                    894.0,
+                    1375.0,
+                    846.0,
+                    1375.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    606.0,
+                    1373.0,
+                    645.0,
+                    1373.0,
+                    645.0,
+                    1402.0,
+                    606.0,
+                    1402.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    693.0,
+                    1373.0,
+                    732.0,
+                    1373.0,
+                    732.0,
+                    1402.0,
+                    693.0,
+                    1402.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    781.0,
+                    1373.0,
+                    819.0,
+                    1373.0,
+                    819.0,
+                    1402.0,
+                    781.0,
+                    1402.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    884.0,
+                    1373.0,
+                    985.0,
+                    1373.0,
+                    985.0,
+                    1401.0,
+                    884.0,
+                    1401.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    990.0,
+                    1373.0,
+                    1041.0,
+                    1373.0,
+                    1041.0,
+                    1401.0,
+                    990.0,
+                    1401.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1042.0,
+                    1373.0,
+                    1144.0,
+                    1373.0,
+                    1144.0,
+                    1402.0,
+                    1042.0,
+                    1402.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    654.0,
+                    1399.0,
+                    734.0,
+                    1399.0,
+                    734.0,
+                    1425.0,
+                    654.0,
+                    1425.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    975.0,
+                    1399.0,
+                    1055.0,
+                    1399.0,
+                    1055.0,
+                    1425.0,
+                    975.0,
+                    1425.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    550.0,
+                    1426.0,
+                    592.0,
+                    1426.0,
+                    592.0,
+                    1451.0,
+                    550.0,
+                    1451.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    629.0,
+                    1434.0,
+                    638.0,
+                    1434.0,
+                    638.0,
+                    1446.0,
+                    629.0,
+                    1446.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    652.0,
+                    1425.0,
+                    720.0,
+                    1425.0,
+                    720.0,
+                    1453.0,
+                    652.0,
+                    1453.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    751.0,
+                    1431.0,
+                    770.0,
+                    1431.0,
+                    770.0,
+                    1449.0,
+                    751.0,
+                    1449.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    778.0,
+                    1423.0,
+                    944.0,
+                    1423.0,
+                    944.0,
+                    1454.0,
+                    778.0,
+                    1454.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1000.0,
+                    1422.0,
+                    1192.0,
+                    1422.0,
+                    1192.0,
+                    1454.0,
+                    1000.0,
+                    1454.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1491.0,
+                    1406.0,
+                    1491.0,
+                    1406.0,
+                    1531.0,
+                    294.0,
+                    1531.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1524.0,
+                    450.0,
+                    1524.0,
+                    450.0,
+                    1558.0,
+                    295.0,
+                    1558.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    830.0,
+                    2085.0,
+                    868.0,
+                    2085.0,
+                    868.0,
+                    2124.0,
+                    830.0,
+                    2124.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    298.0,
+                    73.0,
+                    856.0,
+                    73.0,
+                    856.0,
+                    108.0,
+                    298.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    298.0,
+                    72.0,
+                    856.0,
+                    72.0,
+                    856.0,
+                    108.0,
+                    298.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 20,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 3,
+                "poly": [
+                    405,
+                    158,
+                    1295,
+                    158,
+                    1295,
+                    1973,
+                    405,
+                    1973
+                ],
+                "score": 0.951
+            },
+            {
+                "category_id": 4,
+                "poly": [
+                    427,
+                    1985,
+                    1261,
+                    1985,
+                    1261,
+                    2018,
+                    427,
+                    2018
+                ],
+                "score": 0.913
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    834,
+                    2088,
+                    865,
+                    2088,
+                    865,
+                    2113,
+                    834,
+                    2113
+                ],
+                "score": 0.852
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    298,
+                    76,
+                    851,
+                    76,
+                    851,
+                    104,
+                    298,
+                    104
+                ],
+                "score": 0.749
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1176,
+                    1992,
+                    1208,
+                    1992,
+                    1208,
+                    2016,
+                    1176,
+                    2016
+                ],
+                "score": 0.73,
+                "latex": "\\alpha _ { 1 }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1220,
+                    1993,
+                    1252,
+                    1993,
+                    1252,
+                    2016,
+                    1220,
+                    2016
+                ],
+                "score": 0.57,
+                "latex": "\\alpha _ { 2 }"
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    489,
+                    1382,
+                    737,
+                    1382,
+                    737,
+                    1414,
+                    489,
+                    1414
+                ],
+                "score": 0.57,
+                "latex": "\\alpha _ { 1 } = 0 . 0 1 , \\alpha _ { 2 } = 0 . 1"
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    979,
+                    231,
+                    1239,
+                    231,
+                    1239,
+                    263,
+                    979,
+                    263
+                ],
+                "score": 0.57,
+                "latex": "\\alpha _ { 1 } = 0 . 1 , \\alpha _ { 2 } = 0 . 0 0 1"
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    979,
+                    807,
+                    1226,
+                    807,
+                    1226,
+                    839,
+                    979,
+                    839
+                ],
+                "score": 0.56,
+                "latex": "\\alpha _ { 1 } = 0 . 1 , \\alpha _ { 2 } = 0 . 0 1"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    613,
+                    807,
+                    735,
+                    807,
+                    735,
+                    839,
+                    613,
+                    839
+                ],
+                "score": 0.45,
+                "latex": "\\alpha _ { 2 } = 0 . 0 1"
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    476,
+                    807,
+                    739,
+                    807,
+                    739,
+                    840,
+                    476,
+                    840
+                ],
+                "score": 0.41,
+                "latex": "\\alpha _ { 1 } = 0 . 0 1 , \\alpha _ { 2 } = 0 . 0 1"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    980,
+                    1383,
+                    1088,
+                    1383,
+                    1088,
+                    1414,
+                    980,
+                    1414
+                ],
+                "score": 0.4,
+                "latex": "\\alpha _ { 1 } = 0 . 1"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    477,
+                    808,
+                    600,
+                    808,
+                    600,
+                    839,
+                    477,
+                    839
+                ],
+                "score": 0.39,
+                "latex": "\\alpha _ { 1 } = 0 . 0 1"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    598,
+                    231,
+                    735,
+                    231,
+                    735,
+                    263,
+                    598,
+                    263
+                ],
+                "score": 0.32,
+                "latex": "\\alpha _ { 2 } = 0 . 0 0 1"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1102,
+                    1383,
+                    1211,
+                    1383,
+                    1211,
+                    1414,
+                    1102,
+                    1414
+                ],
+                "score": 0.31,
+                "latex": "\\alpha _ { 2 } = 0 . 1"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    980,
+                    808,
+                    1088,
+                    808,
+                    1088,
+                    839,
+                    980,
+                    839
+                ],
+                "score": 0.3,
+                "latex": "\\alpha _ { 1 } = 0 . 1"
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    979,
+                    1382,
+                    1211,
+                    1382,
+                    1211,
+                    1414,
+                    979,
+                    1414
+                ],
+                "score": 0.26,
+                "latex": "\\alpha _ { 1 } = 0 . 1 , \\alpha _ { 2 } = 0 . 1"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1175,
+                    1992,
+                    1252,
+                    1992,
+                    1252,
+                    2017,
+                    1175,
+                    2017
+                ],
+                "score": 0.26,
+                "latex": "\\alpha _ { 1 } , \\alpha _ { 2 }"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    455.0,
+                    228.0,
+                    597.0,
+                    228.0,
+                    597.0,
+                    267.0,
+                    455.0,
+                    267.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    974.0,
+                    230.0,
+                    978.0,
+                    230.0,
+                    978.0,
+                    263.0,
+                    974.0,
+                    263.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    529.0,
+                    264.0,
+                    746.0,
+                    264.0,
+                    746.0,
+                    290.0,
+                    529.0,
+                    290.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    978.0,
+                    264.0,
+                    1194.0,
+                    264.0,
+                    1194.0,
+                    290.0,
+                    978.0,
+                    290.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    419.0,
+                    284.0,
+                    450.0,
+                    284.0,
+                    450.0,
+                    305.0,
+                    419.0,
+                    305.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    668.0,
+                    288.0,
+                    817.0,
+                    288.0,
+                    817.0,
+                    315.0,
+                    668.0,
+                    315.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    865.0,
+                    284.0,
+                    898.0,
+                    284.0,
+                    898.0,
+                    305.0,
+                    865.0,
+                    305.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    419.0,
+                    331.0,
+                    450.0,
+                    331.0,
+                    450.0,
+                    355.0,
+                    419.0,
+                    355.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    862.0,
+                    329.0,
+                    900.0,
+                    329.0,
+                    900.0,
+                    357.0,
+                    862.0,
+                    357.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    419.0,
+                    381.0,
+                    452.0,
+                    381.0,
+                    452.0,
+                    405.0,
+                    419.0,
+                    405.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    459.0,
+                    375.0,
+                    494.0,
+                    375.0,
+                    494.0,
+                    397.0,
+                    459.0,
+                    397.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    865.0,
+                    381.0,
+                    898.0,
+                    381.0,
+                    898.0,
+                    405.0,
+                    865.0,
+                    405.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    908.0,
+                    377.0,
+                    931.0,
+                    377.0,
+                    931.0,
+                    393.0,
+                    908.0,
+                    393.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1039.0,
+                    371.0,
+                    1101.0,
+                    371.0,
+                    1101.0,
+                    385.0,
+                    1039.0,
+                    385.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    461.0,
+                    399.0,
+                    479.0,
+                    399.0,
+                    479.0,
+                    419.0,
+                    461.0,
+                    419.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    523.0,
+                    395.0,
+                    556.0,
+                    395.0,
+                    556.0,
+                    419.0,
+                    523.0,
+                    419.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    593.0,
+                    393.0,
+                    630.0,
+                    393.0,
+                    630.0,
+                    423.0,
+                    593.0,
+                    423.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    659.0,
+                    393.0,
+                    705.0,
+                    393.0,
+                    705.0,
+                    421.0,
+                    659.0,
+                    421.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    732.0,
+                    393.0,
+                    777.0,
+                    393.0,
+                    777.0,
+                    421.0,
+                    732.0,
+                    421.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    802.0,
+                    397.0,
+                    846.0,
+                    397.0,
+                    846.0,
+                    419.0,
+                    802.0,
+                    419.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    908.0,
+                    399.0,
+                    925.0,
+                    399.0,
+                    925.0,
+                    417.0,
+                    908.0,
+                    417.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    968.0,
+                    393.0,
+                    1007.0,
+                    393.0,
+                    1007.0,
+                    423.0,
+                    968.0,
+                    423.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1038.0,
+                    393.0,
+                    1076.0,
+                    393.0,
+                    1076.0,
+                    423.0,
+                    1038.0,
+                    423.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1107.0,
+                    393.0,
+                    1152.0,
+                    393.0,
+                    1152.0,
+                    423.0,
+                    1107.0,
+                    423.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1174.0,
+                    394.0,
+                    1221.0,
+                    394.0,
+                    1221.0,
+                    418.0,
+                    1174.0,
+                    418.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1246.0,
+                    393.0,
+                    1293.0,
+                    393.0,
+                    1293.0,
+                    421.0,
+                    1246.0,
+                    421.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    601.0,
+                    413.0,
+                    674.0,
+                    413.0,
+                    674.0,
+                    439.0,
+                    601.0,
+                    439.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1046.0,
+                    409.0,
+                    1121.0,
+                    409.0,
+                    1121.0,
+                    442.0,
+                    1046.0,
+                    442.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    508.0,
+                    437.0,
+                    763.0,
+                    437.0,
+                    763.0,
+                    463.0,
+                    508.0,
+                    463.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    954.0,
+                    437.0,
+                    1210.0,
+                    437.0,
+                    1210.0,
+                    463.0,
+                    954.0,
+                    463.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    409.0,
+                    457.0,
+                    454.0,
+                    457.0,
+                    454.0,
+                    485.0,
+                    409.0,
+                    485.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    856.0,
+                    467.0,
+                    902.0,
+                    467.0,
+                    902.0,
+                    495.0,
+                    856.0,
+                    495.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    411.0,
+                    491.0,
+                    450.0,
+                    491.0,
+                    450.0,
+                    513.0,
+                    411.0,
+                    513.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    407.0,
+                    519.0,
+                    452.0,
+                    519.0,
+                    452.0,
+                    547.0,
+                    407.0,
+                    547.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    856.0,
+                    513.0,
+                    898.0,
+                    513.0,
+                    898.0,
+                    535.0,
+                    856.0,
+                    535.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    411.0,
+                    555.0,
+                    452.0,
+                    555.0,
+                    452.0,
+                    577.0,
+                    411.0,
+                    577.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    858.0,
+                    555.0,
+                    900.0,
+                    555.0,
+                    900.0,
+                    577.0,
+                    858.0,
+                    577.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    906.0,
+                    555.0,
+                    1039.0,
+                    555.0,
+                    1039.0,
+                    575.0,
+                    906.0,
+                    575.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    461.0,
+                    575.0,
+                    479.0,
+                    575.0,
+                    479.0,
+                    595.0,
+                    461.0,
+                    595.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    523.0,
+                    573.0,
+                    556.0,
+                    573.0,
+                    556.0,
+                    595.0,
+                    523.0,
+                    595.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    591.0,
+                    571.0,
+                    630.0,
+                    571.0,
+                    630.0,
+                    599.0,
+                    591.0,
+                    599.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    661.0,
+                    573.0,
+                    701.0,
+                    573.0,
+                    701.0,
+                    595.0,
+                    661.0,
+                    595.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    734.0,
+                    573.0,
+                    775.0,
+                    573.0,
+                    775.0,
+                    595.0,
+                    734.0,
+                    595.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    802.0,
+                    573.0,
+                    846.0,
+                    573.0,
+                    846.0,
+                    595.0,
+                    802.0,
+                    595.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    908.0,
+                    575.0,
+                    925.0,
+                    575.0,
+                    925.0,
+                    595.0,
+                    908.0,
+                    595.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    968.0,
+                    571.0,
+                    1007.0,
+                    571.0,
+                    1007.0,
+                    599.0,
+                    968.0,
+                    599.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1038.0,
+                    569.0,
+                    1076.0,
+                    569.0,
+                    1076.0,
+                    597.0,
+                    1038.0,
+                    597.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1109.0,
+                    573.0,
+                    1150.0,
+                    573.0,
+                    1150.0,
+                    595.0,
+                    1109.0,
+                    595.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1174.0,
+                    570.0,
+                    1221.0,
+                    570.0,
+                    1221.0,
+                    594.0,
+                    1174.0,
+                    594.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1246.0,
+                    571.0,
+                    1293.0,
+                    571.0,
+                    1293.0,
+                    599.0,
+                    1246.0,
+                    599.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    598.0,
+                    586.0,
+                    675.0,
+                    586.0,
+                    675.0,
+                    617.0,
+                    598.0,
+                    617.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1046.0,
+                    586.0,
+                    1121.0,
+                    586.0,
+                    1121.0,
+                    617.0,
+                    1046.0,
+                    617.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    531.0,
+                    615.0,
+                    744.0,
+                    615.0,
+                    744.0,
+                    641.0,
+                    531.0,
+                    641.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    978.0,
+                    615.0,
+                    1190.0,
+                    615.0,
+                    1190.0,
+                    641.0,
+                    978.0,
+                    641.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    682.0,
+                    643.0,
+                    815.0,
+                    643.0,
+                    815.0,
+                    663.0,
+                    682.0,
+                    663.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    430.0,
+                    653.0,
+                    454.0,
+                    653.0,
+                    454.0,
+                    677.0,
+                    430.0,
+                    677.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    879.0,
+                    655.0,
+                    896.0,
+                    655.0,
+                    896.0,
+                    673.0,
+                    879.0,
+                    673.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1156.0,
+                    683.0,
+                    1173.0,
+                    683.0,
+                    1173.0,
+                    698.0,
+                    1156.0,
+                    698.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    432.0,
+                    692.0,
+                    448.0,
+                    692.0,
+                    448.0,
+                    714.0,
+                    432.0,
+                    714.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    875.0,
+                    690.0,
+                    900.0,
+                    690.0,
+                    900.0,
+                    712.0,
+                    875.0,
+                    712.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    432.0,
+                    732.0,
+                    450.0,
+                    732.0,
+                    450.0,
+                    750.0,
+                    432.0,
+                    750.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    877.0,
+                    728.0,
+                    902.0,
+                    728.0,
+                    902.0,
+                    752.0,
+                    877.0,
+                    752.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    461.0,
+                    750.0,
+                    479.0,
+                    750.0,
+                    479.0,
+                    770.0,
+                    461.0,
+                    770.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    521.0,
+                    746.0,
+                    560.0,
+                    746.0,
+                    560.0,
+                    774.0,
+                    521.0,
+                    774.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    591.0,
+                    746.0,
+                    630.0,
+                    746.0,
+                    630.0,
+                    774.0,
+                    591.0,
+                    774.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    659.0,
+                    746.0,
+                    705.0,
+                    746.0,
+                    705.0,
+                    774.0,
+                    659.0,
+                    774.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    732.0,
+                    746.0,
+                    777.0,
+                    746.0,
+                    777.0,
+                    774.0,
+                    732.0,
+                    774.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    800.0,
+                    746.0,
+                    846.0,
+                    746.0,
+                    846.0,
+                    774.0,
+                    800.0,
+                    774.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    908.0,
+                    750.0,
+                    925.0,
+                    750.0,
+                    925.0,
+                    770.0,
+                    908.0,
+                    770.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    970.0,
+                    748.0,
+                    1003.0,
+                    748.0,
+                    1003.0,
+                    772.0,
+                    970.0,
+                    772.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1038.0,
+                    746.0,
+                    1076.0,
+                    746.0,
+                    1076.0,
+                    774.0,
+                    1038.0,
+                    774.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1107.0,
+                    746.0,
+                    1152.0,
+                    746.0,
+                    1152.0,
+                    774.0,
+                    1107.0,
+                    774.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1177.0,
+                    746.0,
+                    1223.0,
+                    746.0,
+                    1223.0,
+                    774.0,
+                    1177.0,
+                    774.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    537.0,
+                    766.0,
+                    719.0,
+                    766.0,
+                    719.0,
+                    798.0,
+                    537.0,
+                    798.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    987.0,
+                    766.0,
+                    1165.0,
+                    766.0,
+                    1165.0,
+                    798.0,
+                    987.0,
+                    798.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    469.0,
+                    804.0,
+                    475.0,
+                    804.0,
+                    475.0,
+                    840.0,
+                    469.0,
+                    840.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    972.0,
+                    802.0,
+                    978.0,
+                    802.0,
+                    978.0,
+                    842.0,
+                    972.0,
+                    842.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    529.0,
+                    838.0,
+                    746.0,
+                    838.0,
+                    746.0,
+                    864.0,
+                    529.0,
+                    864.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    980.0,
+                    840.0,
+                    1198.0,
+                    840.0,
+                    1198.0,
+                    866.0,
+                    980.0,
+                    866.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    419.0,
+                    860.0,
+                    452.0,
+                    860.0,
+                    452.0,
+                    882.0,
+                    419.0,
+                    882.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    871.0,
+                    858.0,
+                    912.0,
+                    858.0,
+                    912.0,
+                    886.0,
+                    871.0,
+                    886.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1156.0,
+                    864.0,
+                    1194.0,
+                    864.0,
+                    1194.0,
+                    878.0,
+                    1156.0,
+                    878.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1241.0,
+                    864.0,
+                    1260.0,
+                    864.0,
+                    1260.0,
+                    880.0,
+                    1241.0,
+                    880.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    707.0,
+                    882.0,
+                    722.0,
+                    882.0,
+                    722.0,
+                    896.0,
+                    707.0,
+                    896.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    688.0,
+                    888.0,
+                    715.0,
+                    888.0,
+                    715.0,
+                    910.0,
+                    688.0,
+                    910.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1001.0,
+                    888.0,
+                    1016.0,
+                    888.0,
+                    1016.0,
+                    904.0,
+                    1001.0,
+                    904.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    419.0,
+                    908.0,
+                    450.0,
+                    908.0,
+                    450.0,
+                    932.0,
+                    419.0,
+                    932.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    873.0,
+                    910.0,
+                    906.0,
+                    910.0,
+                    906.0,
+                    932.0,
+                    873.0,
+                    932.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    548.0,
+                    938.0,
+                    558.0,
+                    938.0,
+                    558.0,
+                    948.0,
+                    548.0,
+                    948.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    419.0,
+                    956.0,
+                    450.0,
+                    956.0,
+                    450.0,
+                    980.0,
+                    419.0,
+                    980.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    552.0,
+                    952.0,
+                    568.0,
+                    952.0,
+                    568.0,
+                    966.0,
+                    552.0,
+                    966.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    871.0,
+                    956.0,
+                    910.0,
+                    956.0,
+                    910.0,
+                    984.0,
+                    871.0,
+                    984.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    461.0,
+                    974.0,
+                    479.0,
+                    974.0,
+                    479.0,
+                    994.0,
+                    461.0,
+                    994.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    521.0,
+                    968.0,
+                    558.0,
+                    968.0,
+                    558.0,
+                    998.0,
+                    521.0,
+                    998.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    591.0,
+                    968.0,
+                    632.0,
+                    968.0,
+                    632.0,
+                    998.0,
+                    591.0,
+                    998.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    659.0,
+                    970.0,
+                    705.0,
+                    970.0,
+                    705.0,
+                    998.0,
+                    659.0,
+                    998.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    730.0,
+                    970.0,
+                    777.0,
+                    970.0,
+                    777.0,
+                    998.0,
+                    730.0,
+                    998.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    800.0,
+                    970.0,
+                    846.0,
+                    970.0,
+                    846.0,
+                    998.0,
+                    800.0,
+                    998.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    916.0,
+                    974.0,
+                    933.0,
+                    974.0,
+                    933.0,
+                    994.0,
+                    916.0,
+                    994.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    978.0,
+                    972.0,
+                    1009.0,
+                    972.0,
+                    1009.0,
+                    996.0,
+                    978.0,
+                    996.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1041.0,
+                    970.0,
+                    1082.0,
+                    970.0,
+                    1082.0,
+                    998.0,
+                    1041.0,
+                    998.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1109.0,
+                    970.0,
+                    1154.0,
+                    970.0,
+                    1154.0,
+                    998.0,
+                    1109.0,
+                    998.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1179.0,
+                    970.0,
+                    1223.0,
+                    970.0,
+                    1223.0,
+                    998.0,
+                    1179.0,
+                    998.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1246.0,
+                    970.0,
+                    1293.0,
+                    970.0,
+                    1293.0,
+                    998.0,
+                    1246.0,
+                    998.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    600.0,
+                    985.0,
+                    675.0,
+                    985.0,
+                    675.0,
+                    1018.0,
+                    600.0,
+                    1018.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1051.0,
+                    990.0,
+                    1125.0,
+                    990.0,
+                    1125.0,
+                    1016.0,
+                    1051.0,
+                    1016.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    508.0,
+                    1012.0,
+                    767.0,
+                    1012.0,
+                    767.0,
+                    1044.0,
+                    508.0,
+                    1044.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    960.0,
+                    1016.0,
+                    1217.0,
+                    1016.0,
+                    1217.0,
+                    1042.0,
+                    960.0,
+                    1042.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    409.0,
+                    1034.0,
+                    452.0,
+                    1034.0,
+                    452.0,
+                    1062.0,
+                    409.0,
+                    1062.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    701.0,
+                    1038.0,
+                    819.0,
+                    1038.0,
+                    819.0,
+                    1077.0,
+                    701.0,
+                    1077.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    856.0,
+                    1036.0,
+                    910.0,
+                    1036.0,
+                    910.0,
+                    1064.0,
+                    856.0,
+                    1064.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    856.0,
+                    1067.0,
+                    910.0,
+                    1067.0,
+                    910.0,
+                    1093.0,
+                    856.0,
+                    1093.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1123.0,
+                    1071.0,
+                    1260.0,
+                    1071.0,
+                    1260.0,
+                    1091.0,
+                    1123.0,
+                    1091.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    409.0,
+                    1079.0,
+                    454.0,
+                    1079.0,
+                    454.0,
+                    1107.0,
+                    409.0,
+                    1107.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    854.0,
+                    1097.0,
+                    906.0,
+                    1097.0,
+                    906.0,
+                    1125.0,
+                    854.0,
+                    1125.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    409.0,
+                    1127.0,
+                    454.0,
+                    1127.0,
+                    454.0,
+                    1155.0,
+                    409.0,
+                    1155.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    456.0,
+                    1127.0,
+                    515.0,
+                    1127.0,
+                    515.0,
+                    1153.0,
+                    456.0,
+                    1153.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    856.0,
+                    1127.0,
+                    910.0,
+                    1127.0,
+                    910.0,
+                    1155.0,
+                    856.0,
+                    1155.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    914.0,
+                    1131.0,
+                    951.0,
+                    1131.0,
+                    951.0,
+                    1151.0,
+                    914.0,
+                    1151.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    461.0,
+                    1149.0,
+                    479.0,
+                    1149.0,
+                    479.0,
+                    1169.0,
+                    461.0,
+                    1169.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    523.0,
+                    1147.0,
+                    556.0,
+                    1147.0,
+                    556.0,
+                    1171.0,
+                    523.0,
+                    1171.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    593.0,
+                    1145.0,
+                    632.0,
+                    1145.0,
+                    632.0,
+                    1173.0,
+                    593.0,
+                    1173.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    659.0,
+                    1145.0,
+                    703.0,
+                    1145.0,
+                    703.0,
+                    1173.0,
+                    659.0,
+                    1173.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    730.0,
+                    1145.0,
+                    777.0,
+                    1145.0,
+                    777.0,
+                    1173.0,
+                    730.0,
+                    1173.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    802.0,
+                    1149.0,
+                    844.0,
+                    1149.0,
+                    844.0,
+                    1171.0,
+                    802.0,
+                    1171.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    916.0,
+                    1151.0,
+                    933.0,
+                    1151.0,
+                    933.0,
+                    1169.0,
+                    916.0,
+                    1169.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    976.0,
+                    1147.0,
+                    1009.0,
+                    1147.0,
+                    1009.0,
+                    1171.0,
+                    976.0,
+                    1171.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1045.0,
+                    1147.0,
+                    1078.0,
+                    1147.0,
+                    1078.0,
+                    1171.0,
+                    1045.0,
+                    1171.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1109.0,
+                    1145.0,
+                    1154.0,
+                    1145.0,
+                    1154.0,
+                    1173.0,
+                    1109.0,
+                    1173.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1179.0,
+                    1145.0,
+                    1223.0,
+                    1145.0,
+                    1223.0,
+                    1173.0,
+                    1179.0,
+                    1173.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1246.0,
+                    1145.0,
+                    1293.0,
+                    1145.0,
+                    1293.0,
+                    1173.0,
+                    1246.0,
+                    1173.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    599.0,
+                    1165.0,
+                    676.0,
+                    1165.0,
+                    676.0,
+                    1191.0,
+                    599.0,
+                    1191.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1049.0,
+                    1165.0,
+                    1125.0,
+                    1165.0,
+                    1125.0,
+                    1191.0,
+                    1049.0,
+                    1191.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    529.0,
+                    1185.0,
+                    746.0,
+                    1185.0,
+                    746.0,
+                    1219.0,
+                    529.0,
+                    1219.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    981.0,
+                    1191.0,
+                    1194.0,
+                    1191.0,
+                    1194.0,
+                    1217.0,
+                    981.0,
+                    1217.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    432.0,
+                    1225.0,
+                    448.0,
+                    1225.0,
+                    448.0,
+                    1245.0,
+                    432.0,
+                    1245.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    707.0,
+                    1215.0,
+                    815.0,
+                    1215.0,
+                    815.0,
+                    1235.0,
+                    707.0,
+                    1235.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    883.0,
+                    1229.0,
+                    906.0,
+                    1229.0,
+                    906.0,
+                    1253.0,
+                    883.0,
+                    1253.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1223.0,
+                    1219.0,
+                    1260.0,
+                    1219.0,
+                    1260.0,
+                    1239.0,
+                    1223.0,
+                    1239.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    655.0,
+                    1247.0,
+                    672.0,
+                    1247.0,
+                    672.0,
+                    1261.0,
+                    655.0,
+                    1261.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    432.0,
+                    1265.0,
+                    450.0,
+                    1265.0,
+                    450.0,
+                    1285.0,
+                    432.0,
+                    1285.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    887.0,
+                    1269.0,
+                    904.0,
+                    1269.0,
+                    904.0,
+                    1287.0,
+                    887.0,
+                    1287.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    432.0,
+                    1305.0,
+                    452.0,
+                    1305.0,
+                    452.0,
+                    1323.0,
+                    432.0,
+                    1323.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    889.0,
+                    1307.0,
+                    908.0,
+                    1307.0,
+                    908.0,
+                    1325.0,
+                    889.0,
+                    1325.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    461.0,
+                    1325.0,
+                    479.0,
+                    1325.0,
+                    479.0,
+                    1345.0,
+                    461.0,
+                    1345.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    523.0,
+                    1323.0,
+                    556.0,
+                    1323.0,
+                    556.0,
+                    1347.0,
+                    523.0,
+                    1347.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    591.0,
+                    1321.0,
+                    630.0,
+                    1321.0,
+                    630.0,
+                    1349.0,
+                    591.0,
+                    1349.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    661.0,
+                    1321.0,
+                    705.0,
+                    1321.0,
+                    705.0,
+                    1349.0,
+                    661.0,
+                    1349.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    730.0,
+                    1321.0,
+                    777.0,
+                    1321.0,
+                    777.0,
+                    1349.0,
+                    730.0,
+                    1349.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    800.0,
+                    1321.0,
+                    846.0,
+                    1321.0,
+                    846.0,
+                    1349.0,
+                    800.0,
+                    1349.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    916.0,
+                    1325.0,
+                    933.0,
+                    1325.0,
+                    933.0,
+                    1345.0,
+                    916.0,
+                    1345.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    974.0,
+                    1321.0,
+                    1012.0,
+                    1321.0,
+                    1012.0,
+                    1349.0,
+                    974.0,
+                    1349.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1041.0,
+                    1321.0,
+                    1080.0,
+                    1321.0,
+                    1080.0,
+                    1349.0,
+                    1041.0,
+                    1349.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1109.0,
+                    1321.0,
+                    1156.0,
+                    1321.0,
+                    1156.0,
+                    1349.0,
+                    1109.0,
+                    1349.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1177.0,
+                    1321.0,
+                    1223.0,
+                    1321.0,
+                    1223.0,
+                    1349.0,
+                    1177.0,
+                    1349.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1246.0,
+                    1321.0,
+                    1293.0,
+                    1321.0,
+                    1293.0,
+                    1349.0,
+                    1246.0,
+                    1349.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    535.0,
+                    1339.0,
+                    718.0,
+                    1339.0,
+                    718.0,
+                    1372.0,
+                    535.0,
+                    1372.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    982.0,
+                    1334.0,
+                    1167.0,
+                    1334.0,
+                    1167.0,
+                    1377.0,
+                    982.0,
+                    1377.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    482.0,
+                    1379.0,
+                    488.0,
+                    1379.0,
+                    488.0,
+                    1418.0,
+                    482.0,
+                    1418.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    972.0,
+                    1377.0,
+                    978.0,
+                    1377.0,
+                    978.0,
+                    1421.0,
+                    972.0,
+                    1421.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1212.0,
+                    1377.0,
+                    1215.0,
+                    1377.0,
+                    1215.0,
+                    1421.0,
+                    1212.0,
+                    1421.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    525.0,
+                    1415.0,
+                    742.0,
+                    1415.0,
+                    742.0,
+                    1441.0,
+                    525.0,
+                    1441.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    976.0,
+                    1415.0,
+                    1194.0,
+                    1415.0,
+                    1194.0,
+                    1441.0,
+                    976.0,
+                    1441.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    409.0,
+                    1433.0,
+                    448.0,
+                    1433.0,
+                    448.0,
+                    1460.0,
+                    409.0,
+                    1460.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    599.0,
+                    1439.0,
+                    633.0,
+                    1439.0,
+                    633.0,
+                    1452.0,
+                    599.0,
+                    1452.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    703.0,
+                    1437.0,
+                    815.0,
+                    1437.0,
+                    815.0,
+                    1456.0,
+                    703.0,
+                    1456.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    864.0,
+                    1433.0,
+                    902.0,
+                    1433.0,
+                    902.0,
+                    1460.0,
+                    864.0,
+                    1460.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1125.0,
+                    1439.0,
+                    1264.0,
+                    1439.0,
+                    1264.0,
+                    1458.0,
+                    1125.0,
+                    1458.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    411.0,
+                    1482.0,
+                    442.0,
+                    1482.0,
+                    442.0,
+                    1506.0,
+                    411.0,
+                    1506.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    865.0,
+                    1484.0,
+                    896.0,
+                    1484.0,
+                    896.0,
+                    1508.0,
+                    865.0,
+                    1508.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    409.0,
+                    1530.0,
+                    448.0,
+                    1530.0,
+                    448.0,
+                    1558.0,
+                    409.0,
+                    1558.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    452.0,
+                    1526.0,
+                    492.0,
+                    1526.0,
+                    492.0,
+                    1546.0,
+                    452.0,
+                    1546.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    864.0,
+                    1530.0,
+                    902.0,
+                    1530.0,
+                    902.0,
+                    1558.0,
+                    864.0,
+                    1558.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    908.0,
+                    1530.0,
+                    931.0,
+                    1530.0,
+                    931.0,
+                    1544.0,
+                    908.0,
+                    1544.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    450.0,
+                    1548.0,
+                    473.0,
+                    1548.0,
+                    473.0,
+                    1572.0,
+                    450.0,
+                    1572.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    515.0,
+                    1544.0,
+                    552.0,
+                    1544.0,
+                    552.0,
+                    1574.0,
+                    515.0,
+                    1574.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    587.0,
+                    1544.0,
+                    626.0,
+                    1544.0,
+                    626.0,
+                    1574.0,
+                    587.0,
+                    1574.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    659.0,
+                    1548.0,
+                    699.0,
+                    1548.0,
+                    699.0,
+                    1570.0,
+                    659.0,
+                    1570.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    730.0,
+                    1548.0,
+                    771.0,
+                    1548.0,
+                    771.0,
+                    1570.0,
+                    730.0,
+                    1570.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    802.0,
+                    1548.0,
+                    846.0,
+                    1548.0,
+                    846.0,
+                    1570.0,
+                    802.0,
+                    1570.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    908.0,
+                    1550.0,
+                    925.0,
+                    1550.0,
+                    925.0,
+                    1570.0,
+                    908.0,
+                    1570.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    968.0,
+                    1544.0,
+                    1005.0,
+                    1544.0,
+                    1005.0,
+                    1574.0,
+                    968.0,
+                    1574.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1038.0,
+                    1544.0,
+                    1078.0,
+                    1544.0,
+                    1078.0,
+                    1572.0,
+                    1038.0,
+                    1572.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1109.0,
+                    1548.0,
+                    1150.0,
+                    1548.0,
+                    1150.0,
+                    1570.0,
+                    1109.0,
+                    1570.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1179.0,
+                    1548.0,
+                    1221.0,
+                    1548.0,
+                    1221.0,
+                    1570.0,
+                    1179.0,
+                    1570.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1248.0,
+                    1548.0,
+                    1291.0,
+                    1548.0,
+                    1291.0,
+                    1570.0,
+                    1248.0,
+                    1570.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    596.0,
+                    1560.0,
+                    669.0,
+                    1560.0,
+                    669.0,
+                    1593.0,
+                    596.0,
+                    1593.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1047.0,
+                    1564.0,
+                    1121.0,
+                    1564.0,
+                    1121.0,
+                    1590.0,
+                    1047.0,
+                    1590.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    504.0,
+                    1588.0,
+                    763.0,
+                    1588.0,
+                    763.0,
+                    1620.0,
+                    504.0,
+                    1620.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    954.0,
+                    1588.0,
+                    1210.0,
+                    1588.0,
+                    1210.0,
+                    1614.0,
+                    954.0,
+                    1614.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    409.0,
+                    1616.0,
+                    448.0,
+                    1616.0,
+                    448.0,
+                    1644.0,
+                    409.0,
+                    1644.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    554.0,
+                    1616.0,
+                    815.0,
+                    1616.0,
+                    815.0,
+                    1648.0,
+                    554.0,
+                    1648.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    854.0,
+                    1612.0,
+                    904.0,
+                    1612.0,
+                    904.0,
+                    1640.0,
+                    854.0,
+                    1640.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    955.0,
+                    1616.0,
+                    1260.0,
+                    1616.0,
+                    1260.0,
+                    1650.0,
+                    955.0,
+                    1650.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    409.0,
+                    1646.0,
+                    448.0,
+                    1646.0,
+                    448.0,
+                    1674.0,
+                    409.0,
+                    1674.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    854.0,
+                    1642.0,
+                    898.0,
+                    1642.0,
+                    898.0,
+                    1670.0,
+                    854.0,
+                    1670.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    409.0,
+                    1674.0,
+                    446.0,
+                    1674.0,
+                    446.0,
+                    1704.0,
+                    409.0,
+                    1704.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    856.0,
+                    1674.0,
+                    898.0,
+                    1674.0,
+                    898.0,
+                    1702.0,
+                    856.0,
+                    1702.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    473.0,
+                    1696.0,
+                    486.0,
+                    1696.0,
+                    486.0,
+                    1714.0,
+                    473.0,
+                    1714.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    411.0,
+                    1706.0,
+                    446.0,
+                    1706.0,
+                    446.0,
+                    1728.0,
+                    411.0,
+                    1728.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    858.0,
+                    1706.0,
+                    900.0,
+                    1706.0,
+                    900.0,
+                    1728.0,
+                    858.0,
+                    1728.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    450.0,
+                    1724.0,
+                    473.0,
+                    1724.0,
+                    473.0,
+                    1748.0,
+                    450.0,
+                    1748.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    517.0,
+                    1724.0,
+                    552.0,
+                    1724.0,
+                    552.0,
+                    1746.0,
+                    517.0,
+                    1746.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    587.0,
+                    1722.0,
+                    626.0,
+                    1722.0,
+                    626.0,
+                    1750.0,
+                    587.0,
+                    1750.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    659.0,
+                    1724.0,
+                    699.0,
+                    1724.0,
+                    699.0,
+                    1746.0,
+                    659.0,
+                    1746.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    730.0,
+                    1724.0,
+                    773.0,
+                    1724.0,
+                    773.0,
+                    1746.0,
+                    730.0,
+                    1746.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    802.0,
+                    1724.0,
+                    846.0,
+                    1724.0,
+                    846.0,
+                    1746.0,
+                    802.0,
+                    1746.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    908.0,
+                    1726.0,
+                    925.0,
+                    1726.0,
+                    925.0,
+                    1746.0,
+                    908.0,
+                    1746.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    968.0,
+                    1722.0,
+                    1007.0,
+                    1722.0,
+                    1007.0,
+                    1750.0,
+                    968.0,
+                    1750.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1039.0,
+                    1724.0,
+                    1076.0,
+                    1724.0,
+                    1076.0,
+                    1746.0,
+                    1039.0,
+                    1746.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1109.0,
+                    1724.0,
+                    1150.0,
+                    1724.0,
+                    1150.0,
+                    1746.0,
+                    1109.0,
+                    1746.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1179.0,
+                    1724.0,
+                    1221.0,
+                    1724.0,
+                    1221.0,
+                    1746.0,
+                    1179.0,
+                    1746.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1248.0,
+                    1724.0,
+                    1293.0,
+                    1724.0,
+                    1293.0,
+                    1746.0,
+                    1248.0,
+                    1746.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    596.0,
+                    1735.0,
+                    669.0,
+                    1735.0,
+                    669.0,
+                    1768.0,
+                    596.0,
+                    1768.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1046.0,
+                    1735.0,
+                    1121.0,
+                    1735.0,
+                    1121.0,
+                    1768.0,
+                    1046.0,
+                    1768.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    525.0,
+                    1764.0,
+                    742.0,
+                    1764.0,
+                    742.0,
+                    1796.0,
+                    525.0,
+                    1796.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    974.0,
+                    1762.0,
+                    1192.0,
+                    1762.0,
+                    1192.0,
+                    1796.0,
+                    974.0,
+                    1796.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    570.0,
+                    1788.0,
+                    815.0,
+                    1788.0,
+                    815.0,
+                    1808.0,
+                    570.0,
+                    1808.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    943.0,
+                    1786.0,
+                    1266.0,
+                    1786.0,
+                    1266.0,
+                    1810.0,
+                    943.0,
+                    1810.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    425.0,
+                    1802.0,
+                    440.0,
+                    1802.0,
+                    440.0,
+                    1820.0,
+                    425.0,
+                    1820.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    879.0,
+                    1802.0,
+                    896.0,
+                    1802.0,
+                    896.0,
+                    1820.0,
+                    879.0,
+                    1820.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    425.0,
+                    1843.0,
+                    440.0,
+                    1843.0,
+                    440.0,
+                    1861.0,
+                    425.0,
+                    1861.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    879.0,
+                    1843.0,
+                    894.0,
+                    1843.0,
+                    894.0,
+                    1861.0,
+                    879.0,
+                    1861.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    527.0,
+                    1877.0,
+                    537.0,
+                    1877.0,
+                    537.0,
+                    1887.0,
+                    527.0,
+                    1887.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    425.0,
+                    1883.0,
+                    444.0,
+                    1883.0,
+                    444.0,
+                    1901.0,
+                    425.0,
+                    1901.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    879.0,
+                    1883.0,
+                    898.0,
+                    1883.0,
+                    898.0,
+                    1901.0,
+                    879.0,
+                    1901.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    450.0,
+                    1899.0,
+                    473.0,
+                    1899.0,
+                    473.0,
+                    1923.0,
+                    450.0,
+                    1923.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    517.0,
+                    1899.0,
+                    550.0,
+                    1899.0,
+                    550.0,
+                    1921.0,
+                    517.0,
+                    1921.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    589.0,
+                    1899.0,
+                    622.0,
+                    1899.0,
+                    622.0,
+                    1923.0,
+                    589.0,
+                    1923.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    657.0,
+                    1897.0,
+                    703.0,
+                    1897.0,
+                    703.0,
+                    1925.0,
+                    657.0,
+                    1925.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    728.0,
+                    1897.0,
+                    775.0,
+                    1897.0,
+                    775.0,
+                    1925.0,
+                    728.0,
+                    1925.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    802.0,
+                    1899.0,
+                    846.0,
+                    1899.0,
+                    846.0,
+                    1921.0,
+                    802.0,
+                    1921.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    908.0,
+                    1901.0,
+                    925.0,
+                    1901.0,
+                    925.0,
+                    1921.0,
+                    908.0,
+                    1921.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    970.0,
+                    1899.0,
+                    1003.0,
+                    1899.0,
+                    1003.0,
+                    1923.0,
+                    970.0,
+                    1923.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1038.0,
+                    1897.0,
+                    1076.0,
+                    1897.0,
+                    1076.0,
+                    1925.0,
+                    1038.0,
+                    1925.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1105.0,
+                    1897.0,
+                    1152.0,
+                    1897.0,
+                    1152.0,
+                    1925.0,
+                    1105.0,
+                    1925.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1177.0,
+                    1897.0,
+                    1223.0,
+                    1897.0,
+                    1223.0,
+                    1925.0,
+                    1177.0,
+                    1925.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    539.0,
+                    1921.0,
+                    614.0,
+                    1921.0,
+                    614.0,
+                    1947.0,
+                    539.0,
+                    1947.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    624.0,
+                    1921.0,
+                    717.0,
+                    1921.0,
+                    717.0,
+                    1947.0,
+                    624.0,
+                    1947.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    983.0,
+                    1917.0,
+                    1063.0,
+                    1917.0,
+                    1063.0,
+                    1949.0,
+                    983.0,
+                    1949.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1070.0,
+                    1917.0,
+                    1165.0,
+                    1917.0,
+                    1165.0,
+                    1949.0,
+                    1070.0,
+                    1949.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1176.0,
+                    312.0,
+                    1258.0,
+                    312.0,
+                    1258.0,
+                    347.0,
+                    1176.0,
+                    347.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    535.0,
+                    332.5,
+                    595.0,
+                    332.5,
+                    595.0,
+                    358.5,
+                    535.0,
+                    358.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    643.0,
+                    495.0,
+                    818.0,
+                    495.0,
+                    818.0,
+                    533.0,
+                    643.0,
+                    533.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1209.25,
+                    532.0,
+                    1264.25,
+                    532.0,
+                    1264.25,
+                    553.5,
+                    1209.25,
+                    553.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    451.0,
+                    545.5,
+                    630.0,
+                    545.5,
+                    630.0,
+                    578.0,
+                    451.0,
+                    578.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1163.0,
+                    660.5,
+                    1237.0,
+                    660.5,
+                    1237.0,
+                    688.5,
+                    1163.0,
+                    688.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1243.25,
+                    744.0,
+                    1294.25,
+                    744.0,
+                    1294.25,
+                    772.0,
+                    1243.25,
+                    772.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    720.0,
+                    862.0,
+                    810.0,
+                    862.0,
+                    810.0,
+                    888.0,
+                    720.0,
+                    888.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    980.0,
+                    890.5,
+                    1045.0,
+                    890.5,
+                    1045.0,
+                    925.0,
+                    980.0,
+                    925.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    642.25,
+                    903.5,
+                    710.25,
+                    903.5,
+                    710.25,
+                    930.5,
+                    642.25,
+                    930.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    452.25,
+                    949.0,
+                    485.25,
+                    949.0,
+                    485.25,
+                    969.0,
+                    452.25,
+                    969.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    911.0,
+                    950.0,
+                    950.0,
+                    950.0,
+                    950.0,
+                    970.0,
+                    911.0,
+                    970.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    970.0,
+                    1097.5,
+                    1022.0,
+                    1097.5,
+                    1022.0,
+                    1119.5,
+                    970.0,
+                    1119.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    592.0,
+                    1480.0,
+                    659.0,
+                    1480.0,
+                    659.0,
+                    1498.5,
+                    592.0,
+                    1498.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1243.25,
+                    1895.0,
+                    1296.25,
+                    1895.0,
+                    1296.25,
+                    1923.0,
+                    1243.25,
+                    1923.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    423.0,
+                    1978.0,
+                    1174.0,
+                    1978.0,
+                    1174.0,
+                    2026.0,
+                    423.0,
+                    2026.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1253.0,
+                    1978.0,
+                    1266.0,
+                    1978.0,
+                    1266.0,
+                    2026.0,
+                    1253.0,
+                    2026.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    829.0,
+                    2084.0,
+                    871.0,
+                    2084.0,
+                    871.0,
+                    2125.0,
+                    829.0,
+                    2125.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    73.0,
+                    856.0,
+                    73.0,
+                    856.0,
+                    108.0,
+                    297.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 21,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    1351,
+                    1402,
+                    1351,
+                    1402,
+                    1477,
+                    299,
+                    1477
+                ],
+                "score": 0.979
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    294,
+                    1404,
+                    294,
+                    1404,
+                    420,
+                    299,
+                    420
+                ],
+                "score": 0.977
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    1303,
+                    819,
+                    1303,
+                    819,
+                    1339,
+                    299,
+                    1339
+                ],
+                "score": 0.918
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    300,
+                    433,
+                    1193,
+                    433,
+                    1193,
+                    467,
+                    300,
+                    467
+                ],
+                "score": 0.901
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    760,
+                    1729,
+                    993,
+                    1729,
+                    993,
+                    1762,
+                    760,
+                    1762
+                ],
+                "score": 0.896
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    769,
+                    1650,
+                    1032,
+                    1650,
+                    1032,
+                    1683,
+                    769,
+                    1683
+                ],
+                "score": 0.889
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    442,
+                    1689,
+                    1260,
+                    1689,
+                    1260,
+                    1726,
+                    442,
+                    1726
+                ],
+                "score": 0.885
+            },
+            {
+                "category_id": 9,
+                "poly": [
+                    1352,
+                    1553,
+                    1400,
+                    1553,
+                    1400,
+                    1583,
+                    1352,
+                    1583
+                ],
+                "score": 0.883
+            },
+            {
+                "category_id": 9,
+                "poly": [
+                    1352,
+                    1864,
+                    1400,
+                    1864,
+                    1400,
+                    1894,
+                    1352,
+                    1894
+                ],
+                "score": 0.88
+            },
+            {
+                "category_id": 9,
+                "poly": [
+                    1352,
+                    1994,
+                    1401,
+                    1994,
+                    1401,
+                    2024,
+                    1352,
+                    2024
+                ],
+                "score": 0.878
+            },
+            {
+                "category_id": 9,
+                "poly": [
+                    1352,
+                    1124,
+                    1400,
+                    1124,
+                    1400,
+                    1155,
+                    1352,
+                    1155
+                ],
+                "score": 0.878
+            },
+            {
+                "category_id": 9,
+                "poly": [
+                    1352,
+                    1939,
+                    1400,
+                    1939,
+                    1400,
+                    1970,
+                    1352,
+                    1970
+                ],
+                "score": 0.878
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    360,
+                    966,
+                    1277,
+                    966,
+                    1277,
+                    1040,
+                    360,
+                    1040
+                ],
+                "score": 0.877
+            },
+            {
+                "category_id": 9,
+                "poly": [
+                    1352,
+                    1712,
+                    1400,
+                    1712,
+                    1400,
+                    1741,
+                    1352,
+                    1741
+                ],
+                "score": 0.874
+            },
+            {
+                "category_id": 9,
+                "poly": [
+                    1352,
+                    1633,
+                    1400,
+                    1633,
+                    1400,
+                    1663,
+                    1352,
+                    1663
+                ],
+                "score": 0.874
+            },
+            {
+                "category_id": 9,
+                "poly": [
+                    1352,
+                    1790,
+                    1400,
+                    1790,
+                    1400,
+                    1821,
+                    1352,
+                    1821
+                ],
+                "score": 0.87
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    834,
+                    2087,
+                    865,
+                    2087,
+                    865,
+                    2113,
+                    834,
+                    2113
+                ],
+                "score": 0.868
+            },
+            {
+                "category_id": 9,
+                "poly": [
+                    1366,
+                    503,
+                    1400,
+                    503,
+                    1400,
+                    533,
+                    1366,
+                    533
+                ],
+                "score": 0.866
+            },
+            {
+                "category_id": 9,
+                "poly": [
+                    1366,
+                    990,
+                    1400,
+                    990,
+                    1400,
+                    1019,
+                    1366,
+                    1019
+                ],
+                "score": 0.864
+            },
+            {
+                "category_id": 9,
+                "poly": [
+                    1366,
+                    656,
+                    1400,
+                    656,
+                    1400,
+                    685,
+                    1366,
+                    685
+                ],
+                "score": 0.863
+            },
+            {
+                "category_id": 9,
+                "poly": [
+                    1366,
+                    864,
+                    1400,
+                    864,
+                    1400,
+                    893,
+                    1366,
+                    893
+                ],
+                "score": 0.861
+            },
+            {
+                "category_id": 9,
+                "poly": [
+                    1352,
+                    1241,
+                    1400,
+                    1241,
+                    1400,
+                    1270,
+                    1352,
+                    1270
+                ],
+                "score": 0.861
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    449,
+                    1608,
+                    1313,
+                    1608,
+                    1313,
+                    1646,
+                    449,
+                    1646
+                ],
+                "score": 0.86
+            },
+            {
+                "category_id": 9,
+                "poly": [
+                    1366,
+                    786,
+                    1400,
+                    786,
+                    1400,
+                    815,
+                    1366,
+                    815
+                ],
+                "score": 0.859
+            },
+            {
+                "category_id": 9,
+                "poly": [
+                    1366,
+                    1067,
+                    1400,
+                    1067,
+                    1400,
+                    1096,
+                    1366,
+                    1096
+                ],
+                "score": 0.858
+            },
+            {
+                "category_id": 9,
+                "poly": [
+                    1366,
+                    735,
+                    1400,
+                    735,
+                    1400,
+                    763,
+                    1366,
+                    763
+                ],
+                "score": 0.856
+            },
+            {
+                "category_id": 9,
+                "poly": [
+                    1367,
+                    826,
+                    1400,
+                    826,
+                    1400,
+                    854,
+                    1367,
+                    854
+                ],
+                "score": 0.855
+            },
+            {
+                "category_id": 9,
+                "poly": [
+                    1353,
+                    1203,
+                    1400,
+                    1203,
+                    1400,
+                    1232,
+                    1353,
+                    1232
+                ],
+                "score": 0.851
+            },
+            {
+                "category_id": 9,
+                "poly": [
+                    1366,
+                    581,
+                    1400,
+                    581,
+                    1400,
+                    609,
+                    1366,
+                    609
+                ],
+                "score": 0.85
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    443,
+                    1122,
+                    1148,
+                    1122,
+                    1148,
+                    1157,
+                    443,
+                    1157
+                ],
+                "score": 0.833
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    404,
+                    712,
+                    1003,
+                    712,
+                    1003,
+                    776,
+                    404,
+                    776
+                ],
+                "score": 0.819
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    298,
+                    227,
+                    1207,
+                    227,
+                    1207,
+                    263,
+                    298,
+                    263
+                ],
+                "score": 0.807
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    810,
+                    597,
+                    958,
+                    597,
+                    958,
+                    629,
+                    810,
+                    629
+                ],
+                "score": 0.794
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    298,
+                    74,
+                    854,
+                    74,
+                    854,
+                    105,
+                    298,
+                    105
+                ],
+                "score": 0.761
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    461,
+                    1768,
+                    1112,
+                    1768,
+                    1112,
+                    1839,
+                    461,
+                    1839
+                ],
+                "score": 0.719
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    800,
+                    674,
+                    917,
+                    674,
+                    917,
+                    705,
+                    800,
+                    705
+                ],
+                "score": 0.697
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    462,
+                    1843,
+                    797,
+                    1843,
+                    797,
+                    1912,
+                    462,
+                    1912
+                ],
+                "score": 0.69
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    332,
+                    479,
+                    1229,
+                    479,
+                    1229,
+                    553,
+                    332,
+                    553
+                ],
+                "score": 0.676
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    337,
+                    1529,
+                    829,
+                    1529,
+                    829,
+                    1605,
+                    337,
+                    1605
+                ],
+                "score": 0.665
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    364,
+                    557,
+                    1328,
+                    557,
+                    1328,
+                    628,
+                    364,
+                    628
+                ],
+                "score": 0.647
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    462,
+                    1917,
+                    711,
+                    1917,
+                    711,
+                    1990,
+                    462,
+                    1990
+                ],
+                "score": 0.601
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    401,
+                    782,
+                    969,
+                    782,
+                    969,
+                    818,
+                    401,
+                    818
+                ],
+                "score": 0.573
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    371,
+                    633,
+                    1283,
+                    633,
+                    1283,
+                    669,
+                    371,
+                    669
+                ],
+                "score": 0.554
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    376,
+                    634,
+                    1281,
+                    634,
+                    1281,
+                    706,
+                    376,
+                    706
+                ],
+                "score": 0.493
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    359,
+                    558,
+                    1323,
+                    558,
+                    1323,
+                    595,
+                    359,
+                    595
+                ],
+                "score": 0.466
+            },
+            {
+                "category_id": 9,
+                "poly": [
+                    1353,
+                    1164,
+                    1399,
+                    1164,
+                    1399,
+                    1194,
+                    1353,
+                    1194
+                ],
+                "score": 0.36
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    366,
+                    1044,
+                    1329,
+                    1044,
+                    1329,
+                    1117,
+                    366,
+                    1117
+                ],
+                "score": 0.343
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    371,
+                    1044,
+                    1327,
+                    1044,
+                    1327,
+                    1079,
+                    371,
+                    1079
+                ],
+                "score": 0.331
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    841,
+                    1085,
+                    962,
+                    1085,
+                    962,
+                    1116,
+                    841,
+                    1116
+                ],
+                "score": 0.32
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    448,
+                    1161,
+                    756,
+                    1161,
+                    756,
+                    1196,
+                    448,
+                    1196
+                ],
+                "score": 0.313
+            },
+            {
+                "category_id": 9,
+                "poly": [
+                    1353,
+                    1164,
+                    1399,
+                    1164,
+                    1399,
+                    1194,
+                    1353,
+                    1194
+                ],
+                "score": 0.257
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    446,
+                    1201,
+                    616,
+                    1201,
+                    616,
+                    1234,
+                    446,
+                    1234
+                ],
+                "score": 0.194
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    297,
+                    74,
+                    854,
+                    74,
+                    854,
+                    105,
+                    297,
+                    105
+                ],
+                "score": 0.165
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    402,
+                    782,
+                    975,
+                    782,
+                    975,
+                    892,
+                    402,
+                    892
+                ],
+                "score": 0.164
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    766,
+                    520,
+                    911,
+                    520,
+                    911,
+                    551,
+                    766,
+                    551
+                ],
+                "score": 0.156
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    446,
+                    1239,
+                    570,
+                    1239,
+                    570,
+                    1271,
+                    446,
+                    1271
+                ],
+                "score": 0.154
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    417,
+                    480,
+                    1252,
+                    480,
+                    1252,
+                    518,
+                    417,
+                    518
+                ],
+                "score": 0.102
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    440,
+                    1160,
+                    758,
+                    1160,
+                    758,
+                    1272,
+                    440,
+                    1272
+                ],
+                "score": 0.101
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    416,
+                    1045,
+                    1307,
+                    1045,
+                    1307,
+                    1080,
+                    416,
+                    1080
+                ],
+                "score": 0.1
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    328,
+                    479,
+                    1335,
+                    479,
+                    1335,
+                    899,
+                    328,
+                    899
+                ],
+                "score": 0.95,
+                "latex": "\\begin{array} { r l } & { Q ( C ) = \\frac { | s = C , a = C | R ( s = C , a = C ) + | s = D , a = C | R ( s = D , a = C ) } { | a = C | } } \\\\ & { \\quad \\quad = \\frac { N P ( s = C , a = C ) R ( s = C , a = C ) + N P ( s = D , a = C ) R ( s = D , a = C ) } { N P ( a = C ) } } \\\\ & { \\quad \\quad = \\frac { P ( s = C , a = C ) R ( s = C , a = C ) + P ( s = D , a = C ) R ( s = D , a = C ) } { P ( a = C ) } } \\\\ & { \\quad \\quad = \\frac { \\theta ^ { 2 } R ( s = C , a = C ) + \\theta ( 1 - \\theta ) R ( s = D , a = C ) } { \\theta } } \\\\ & { \\quad \\quad = \\theta R ( s = C , a = C ) + ( 1 - \\theta ) R ( s = D , a = C ) } \\\\ & { \\quad \\quad = \\theta ( 0 ) + ( 1 - \\theta ) ( - 1 ) } \\\\ & { \\quad \\quad = \\theta - 1 } \\end{array}"
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    332,
+                    1528,
+                    1316,
+                    1528,
+                    1316,
+                    2002,
+                    332,
+                    2002
+                ],
+                "score": 0.95,
+                "latex": "\\begin{array} { r l } & { Q ^ { \\theta _ { 1 } , \\theta _ { 2 } } ( C ) = \\frac { | C C | R ( C C ) + | D C | | R ( U C ) } { | C | } } \\\\ & { \\quad \\quad \\quad = \\frac { N \\left( P ^ { \\theta _ { 1 } } ( C C ) + P ^ { \\theta _ { 2 } } ( C C ) \\right) R ( C C ) + N ( P ^ { \\theta _ { 1 } } ( D C ) + P ^ { \\theta _ { 2 } } ( D C ) ) R ( D C ) } { N \\left( P ^ { \\theta _ { 1 } } ( C ) + P ^ { \\theta _ { 2 } } ( C ) \\right) } } \\\\ & { \\quad \\quad \\quad = \\frac { ( P ^ { \\theta _ { 1 } } ( C C ) + P ^ { \\theta _ { 2 } } ( C C ) ) I ( C C ) C + ( P ^ { \\theta _ { 1 } } ( D C ) + P ^ { \\theta _ { 2 } } ( D C ) ) R ( D C ) } { ( F ^ { \\theta _ { 1 } } ( C ) + P ^ { \\theta _ { 2 } } ( C ) ) } } \\\\ & { \\quad \\quad \\quad = \\frac { ( \\theta _ { 1 } ^ { 2 } + \\theta _ { 2 } ^ { 2 } ) R ( C C ) + ( \\theta _ { 1 } ( 1 - \\theta _ { 1 } ) + \\theta _ { 2 } ( 1 - \\theta _ { 2 } ) ) R ( D C ) } { \\theta _ { 1 } + \\theta _ { 2 } } } \\\\ & { \\quad \\quad \\quad = - \\frac { \\theta _ { 1 } ( 1 - \\theta _ { 1 } ) + \\theta _ { 2 } ( 1 - \\theta _ { 2 } ) } { \\theta _ { 1 } + \\theta _ { 2 } } } \\\\ & { \\quad \\quad \\quad = \\frac { \\theta _ { 1 } ^ { 2 } - \\theta _ { 1 } + \\theta _ { 2 } ^ { 2 } - \\theta _ { 2 } } { \\theta _ { 1 } + \\theta _ { 2 } } } \\\\ & { \\quad \\quad \\quad = \\frac { \\theta _ { 1 } ^ { 2 } - \\theta _ { 1 } + \\theta _ { 2 } ^ { 2 } - \\theta _ { 2 } } { \\theta _ { 1 } + \\theta _ { 2 } } } \\\\ & { \\quad \\quad \\quad = \\frac { \\theta _ { 1 } ^ { 2 } - \\theta _ { 1 } + \\theta _ { 2 } ^ { 2 } - \\theta _ { 2 } } { \\theta _ { 1 } + \\theta _ { 2 } } } \\end{array}"
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    368,
+                    966,
+                    1327,
+                    966,
+                    1327,
+                    1279,
+                    368,
+                    1279
+                ],
+                "score": 0.94,
+                "latex": "\\begin{array} { l } { Q ( D ) = \\frac { \\left| s = C , a = D \\right| R \\left( s = C , a = D \\right) + \\left| s = D , a = D \\right| R \\left( s = D , a = D \\right) } { \\left| a = D \\right| } } \\\\ { \\ = \\frac { P \\left( s = C , a = D \\right) R \\left( s = C , a = D \\right) + P \\left( s = D , a = D \\right) R \\left( s = D , a = D \\right) } { P \\left( a = D \\right) } } \\\\ { \\ = P ( s = C ) R ( s = C , a = D ) + P ( s = D ) R ( s = D , a = D ) } \\\\ { \\ = \\theta ( 1 / 2 ) + ( 1 - \\theta ) ( - 1 / 2 ) } \\\\ { \\ = 1 / 2 ( 2 \\theta - 1 ) } \\\\ { \\ = \\theta - 1 / 2 } \\end{array}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    465,
+                    1305,
+                    635,
+                    1305,
+                    635,
+                    1340,
+                    465,
+                    1340
+                ],
+                "score": 0.94,
+                "latex": "Q ( D ) > Q ( C )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1028,
+                    1445,
+                    1190,
+                    1445,
+                    1190,
+                    1476,
+                    1028,
+                    1476
+                ],
+                "score": 0.92,
+                "latex": "s = D , a = C"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    297,
+                    358,
+                    569,
+                    358,
+                    569,
+                    391,
+                    297,
+                    391
+                ],
+                "score": 0.88,
+                "latex": "P ( s = { \\mathrm { c o o p e r a t e } } ) = \\theta"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    298,
+                    1384,
+                    326,
+                    1384,
+                    326,
+                    1414,
+                    298,
+                    1414
+                ],
+                "score": 0.87,
+                "latex": "\\theta _ { 1 }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    378,
+                    1385,
+                    406,
+                    1385,
+                    406,
+                    1414,
+                    378,
+                    1414
+                ],
+                "score": 0.85,
+                "latex": "\\theta _ { 2 }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    854,
+                    327,
+                    923,
+                    327,
+                    923,
+                    359,
+                    854,
+                    359
+                ],
+                "score": 0.83,
+                "latex": "\\theta = p"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    629,
+                    436,
+                    657,
+                    436,
+                    657,
+                    462,
+                    629,
+                    462
+                ],
+                "score": 0.83,
+                "latex": "N"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    839,
+                    1445,
+                    887,
+                    1445,
+                    887,
+                    1473,
+                    839,
+                    1473
+                ],
+                "score": 0.83,
+                "latex": "_ { D C }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    679,
+                    1354,
+                    707,
+                    1354,
+                    707,
+                    1380,
+                    679,
+                    1380
+                ],
+                "score": 0.82,
+                "latex": "N"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    796,
+                    1307,
+                    812,
+                    1307,
+                    812,
+                    1333,
+                    796,
+                    1333
+                ],
+                "score": 0.8,
+                "latex": "\\theta"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    855,
+                    327,
+                    1061,
+                    327,
+                    1061,
+                    360,
+                    855,
+                    360
+                ],
+                "score": 0.46,
+                "latex": "\\theta = p ( { \\mathrm { c o o p e r a t e } } )"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    830.0,
+                    2084.0,
+                    870.0,
+                    2084.0,
+                    870.0,
+                    2123.0,
+                    830.0,
+                    2123.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    218.0,
+                    1213.0,
+                    218.0,
+                    1213.0,
+                    276.0,
+                    292.0,
+                    276.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    72.0,
+                    857.0,
+                    72.0,
+                    857.0,
+                    108.0,
+                    296.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    72.0,
+                    856.0,
+                    72.0,
+                    856.0,
+                    108.0,
+                    297.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1349.0,
+                    678.0,
+                    1349.0,
+                    678.0,
+                    1388.0,
+                    293.0,
+                    1388.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    708.0,
+                    1349.0,
+                    1406.0,
+                    1349.0,
+                    1406.0,
+                    1388.0,
+                    708.0,
+                    1388.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1382.0,
+                    297.0,
+                    1382.0,
+                    297.0,
+                    1418.0,
+                    293.0,
+                    1418.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    327.0,
+                    1382.0,
+                    377.0,
+                    1382.0,
+                    377.0,
+                    1418.0,
+                    327.0,
+                    1418.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    407.0,
+                    1382.0,
+                    1406.0,
+                    1382.0,
+                    1406.0,
+                    1418.0,
+                    407.0,
+                    1418.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1415.0,
+                    1406.0,
+                    1415.0,
+                    1406.0,
+                    1448.0,
+                    295.0,
+                    1448.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1445.0,
+                    838.0,
+                    1445.0,
+                    838.0,
+                    1480.0,
+                    293.0,
+                    1480.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    888.0,
+                    1445.0,
+                    1027.0,
+                    1445.0,
+                    1027.0,
+                    1480.0,
+                    888.0,
+                    1480.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1191.0,
+                    1445.0,
+                    1248.0,
+                    1445.0,
+                    1248.0,
+                    1480.0,
+                    1191.0,
+                    1480.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    298.0,
+                    1403.0,
+                    298.0,
+                    1403.0,
+                    328.0,
+                    297.0,
+                    328.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    328.0,
+                    853.0,
+                    328.0,
+                    853.0,
+                    361.0,
+                    295.0,
+                    361.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1062.0,
+                    328.0,
+                    1405.0,
+                    328.0,
+                    1405.0,
+                    361.0,
+                    1062.0,
+                    361.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    570.0,
+                    355.0,
+                    1409.0,
+                    355.0,
+                    1409.0,
+                    394.0,
+                    570.0,
+                    394.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    385.0,
+                    1209.0,
+                    385.0,
+                    1209.0,
+                    423.0,
+                    293.0,
+                    423.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1302.0,
+                    464.0,
+                    1302.0,
+                    464.0,
+                    1342.0,
+                    296.0,
+                    1342.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    636.0,
+                    1302.0,
+                    795.0,
+                    1302.0,
+                    795.0,
+                    1342.0,
+                    636.0,
+                    1342.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    813.0,
+                    1302.0,
+                    822.0,
+                    1302.0,
+                    822.0,
+                    1342.0,
+                    813.0,
+                    1342.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    427.0,
+                    628.0,
+                    427.0,
+                    628.0,
+                    476.0,
+                    292.0,
+                    476.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    658.0,
+                    427.0,
+                    1198.0,
+                    427.0,
+                    1198.0,
+                    476.0,
+                    658.0,
+                    476.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 22,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 3,
+                "poly": [
+                    456,
+                    744,
+                    1241,
+                    744,
+                    1241,
+                    1493,
+                    456,
+                    1493
+                ],
+                "score": 0.973
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    295,
+                    603,
+                    1400,
+                    603,
+                    1400,
+                    699,
+                    295,
+                    699
+                ],
+                "score": 0.966
+            },
+            {
+                "category_id": 4,
+                "poly": [
+                    297,
+                    1547,
+                    1405,
+                    1547,
+                    1405,
+                    1642,
+                    297,
+                    1642
+                ],
+                "score": 0.956
+            },
+            {
+                "category_id": 9,
+                "poly": [
+                    1352,
+                    524,
+                    1400,
+                    524,
+                    1400,
+                    554,
+                    1352,
+                    554
+                ],
+                "score": 0.884
+            },
+            {
+                "category_id": 9,
+                "poly": [
+                    1352,
+                    450,
+                    1400,
+                    450,
+                    1400,
+                    481,
+                    1352,
+                    481
+                ],
+                "score": 0.878
+            },
+            {
+                "category_id": 9,
+                "poly": [
+                    1352,
+                    293,
+                    1400,
+                    293,
+                    1400,
+                    324,
+                    1352,
+                    324
+                ],
+                "score": 0.874
+            },
+            {
+                "category_id": 9,
+                "poly": [
+                    1352,
+                    372,
+                    1400,
+                    372,
+                    1400,
+                    402,
+                    1352,
+                    402
+                ],
+                "score": 0.873
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    834,
+                    2088,
+                    865,
+                    2088,
+                    865,
+                    2113,
+                    834,
+                    2113
+                ],
+                "score": 0.856
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    484,
+                    503,
+                    964,
+                    503,
+                    964,
+                    575,
+                    484,
+                    575
+                ],
+                "score": 0.824
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    475,
+                    348,
+                    1291,
+                    348,
+                    1291,
+                    386,
+                    475,
+                    386
+                ],
+                "score": 0.784
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    786,
+                    389,
+                    1016,
+                    389,
+                    1016,
+                    423,
+                    786,
+                    423
+                ],
+                "score": 0.772
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    480,
+                    428,
+                    1205,
+                    428,
+                    1205,
+                    469,
+                    480,
+                    469
+                ],
+                "score": 0.583
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    788,
+                    469,
+                    928,
+                    469,
+                    928,
+                    498,
+                    788,
+                    498
+                ],
+                "score": 0.575
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    357,
+                    269,
+                    859,
+                    269,
+                    859,
+                    344,
+                    357,
+                    344
+                ],
+                "score": 0.573
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    297,
+                    75,
+                    854,
+                    75,
+                    854,
+                    105,
+                    297,
+                    105
+                ],
+                "score": 0.474
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    478,
+                    428,
+                    1202,
+                    428,
+                    1202,
+                    498,
+                    478,
+                    498
+                ],
+                "score": 0.464
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    297,
+                    75,
+                    854,
+                    75,
+                    854,
+                    105,
+                    297,
+                    105
+                ],
+                "score": 0.414
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    475,
+                    348,
+                    1287,
+                    348,
+                    1287,
+                    423,
+                    475,
+                    423
+                ],
+                "score": 0.129
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    450,
+                    429,
+                    1205,
+                    429,
+                    1205,
+                    466,
+                    450,
+                    466
+                ],
+                "score": 0.12
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    358,
+                    269,
+                    1296,
+                    269,
+                    1296,
+                    582,
+                    358,
+                    582
+                ],
+                "score": 0.96,
+                "latex": "\\begin{array} { l } { { Q ^ { \\theta _ { 1 } , \\theta _ { 2 } } ( D ) = \\frac { \\left| C D \\right| R \\left( C D \\right) + \\left| D D \\right| R \\left( D D \\right) } { \\left| D \\right| } } } \\\\ { { \\ = \\frac { \\left( P ^ { \\theta _ { 1 } } \\left( C D \\right) + P ^ { \\theta _ { 2 } } \\left( C D \\right) \\right) R \\left( C D \\right) + \\left( P ^ { \\theta _ { 1 } } \\left( D D \\right) + P ^ { \\theta _ { 2 } } \\left( D D \\right) \\right) R \\left( D D \\right) } { \\left( P ^ { \\theta _ { 1 } } \\left( D \\right) + P ^ { \\theta _ { 2 } } \\left( D \\right) \\right) } } } \\\\ { { \\ = \\frac { 1 / 2 \\left( \\theta _ { 1 } \\left( 1 - \\theta _ { 1 } \\right) + \\theta _ { 2 } \\left( 1 - \\theta _ { 2 } \\right) \\right) - 1 / 2 \\left( ( 1 - \\theta _ { 1 } ) ^ { 2 } + ( 1 - \\theta _ { 2 } ) ^ { 2 } \\right) } { 2 - \\theta _ { 1 } - \\theta _ { 2 } } } } \\\\ { { \\ = \\frac { \\left( 2 \\theta _ { 1 } - 1 \\right) \\left( 1 - \\theta _ { 1 } \\right) + \\left( 2 \\theta _ { 2 } - 1 \\right) \\left( 1 - \\theta _ { 2 } \\right) } { 4 - 2 \\theta _ { 1 } - 2 \\theta _ { 2 } } } } \\end{array}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    666,
+                    605,
+                    849,
+                    605,
+                    849,
+                    639,
+                    666,
+                    639
+                ],
+                "score": 0.93,
+                "latex": "Q ( C ) > Q ( D )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1080,
+                    1579,
+                    1256,
+                    1579,
+                    1256,
+                    1613,
+                    1080,
+                    1613
+                ],
+                "score": 0.93,
+                "latex": "\\mathbf { \\bar { Q } } ( C ) > Q ( D )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    895,
+                    673,
+                    909,
+                    673,
+                    909,
+                    694,
+                    895,
+                    694
+                ],
+                "score": 0.58,
+                "latex": "\\epsilon"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    504.0,
+                    735.0,
+                    575.0,
+                    735.0,
+                    575.0,
+                    786.0,
+                    504.0,
+                    786.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    507.0,
+                    866.0,
+                    574.0,
+                    866.0,
+                    574.0,
+                    912.0,
+                    507.0,
+                    912.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    505.0,
+                    989.0,
+                    595.0,
+                    989.0,
+                    595.0,
+                    1040.0,
+                    505.0,
+                    1040.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    459.0,
+                    1059.0,
+                    500.0,
+                    1059.0,
+                    500.0,
+                    1103.0,
+                    459.0,
+                    1103.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    504.0,
+                    1117.0,
+                    596.0,
+                    1117.0,
+                    596.0,
+                    1168.0,
+                    504.0,
+                    1168.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    504.0,
+                    1240.0,
+                    592.0,
+                    1240.0,
+                    592.0,
+                    1296.0,
+                    504.0,
+                    1296.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    507.0,
+                    1371.0,
+                    614.0,
+                    1371.0,
+                    614.0,
+                    1450.0,
+                    507.0,
+                    1450.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    677.0,
+                    1403.0,
+                    741.0,
+                    1403.0,
+                    741.0,
+                    1451.0,
+                    677.0,
+                    1451.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    805.0,
+                    1403.0,
+                    870.0,
+                    1403.0,
+                    870.0,
+                    1451.0,
+                    805.0,
+                    1451.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    929.0,
+                    1403.0,
+                    995.0,
+                    1403.0,
+                    995.0,
+                    1451.0,
+                    929.0,
+                    1451.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1059.0,
+                    1399.0,
+                    1121.0,
+                    1399.0,
+                    1121.0,
+                    1450.0,
+                    1059.0,
+                    1450.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1183.0,
+                    1401.0,
+                    1251.0,
+                    1401.0,
+                    1251.0,
+                    1454.0,
+                    1183.0,
+                    1454.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    875.0,
+                    1446.0,
+                    929.0,
+                    1446.0,
+                    929.0,
+                    1499.0,
+                    875.0,
+                    1499.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1546.0,
+                    1408.0,
+                    1546.0,
+                    1408.0,
+                    1586.0,
+                    294.0,
+                    1586.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1577.0,
+                    1079.0,
+                    1577.0,
+                    1079.0,
+                    1618.0,
+                    292.0,
+                    1618.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1257.0,
+                    1577.0,
+                    1406.0,
+                    1577.0,
+                    1406.0,
+                    1618.0,
+                    1257.0,
+                    1618.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1610.0,
+                    546.0,
+                    1610.0,
+                    546.0,
+                    1646.0,
+                    296.0,
+                    1646.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    829.0,
+                    2084.0,
+                    871.0,
+                    2084.0,
+                    871.0,
+                    2125.0,
+                    829.0,
+                    2125.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    298.0,
+                    73.0,
+                    856.0,
+                    73.0,
+                    856.0,
+                    108.0,
+                    298.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    298.0,
+                    73.0,
+                    856.0,
+                    73.0,
+                    856.0,
+                    108.0,
+                    298.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    603.0,
+                    665.0,
+                    603.0,
+                    665.0,
+                    641.0,
+                    294.0,
+                    641.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    850.0,
+                    603.0,
+                    1405.0,
+                    603.0,
+                    1405.0,
+                    641.0,
+                    850.0,
+                    641.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    635.0,
+                    1406.0,
+                    635.0,
+                    1406.0,
+                    671.0,
+                    294.0,
+                    671.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    667.0,
+                    894.0,
+                    667.0,
+                    894.0,
+                    701.0,
+                    296.0,
+                    701.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    910.0,
+                    667.0,
+                    1402.0,
+                    667.0,
+                    1402.0,
+                    701.0,
+                    910.0,
+                    701.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 23,
+            "width": 1700,
+            "height": 2200
+        }
+    }
+]
\ No newline at end of file
diff --git a/parse/train/AHm3dbp7D1D/AHm3dbp7D1D.md b/parse/train/AHm3dbp7D1D/AHm3dbp7D1D.md
new file mode 100644
index 0000000000000000000000000000000000000000..ff56c52ea1b446b645513b6963503a6a03cd3ff7
--- /dev/null
+++ b/parse/train/AHm3dbp7D1D/AHm3dbp7D1D.md
@@ -0,0 +1,525 @@
+# SEED: SELF-SUPERVISED DISTILLATION FOR VISUAL REPRESENTATION
+
+Zhiyuan Fang† , Jianfeng Wang‡, Lijuan Wang‡, Lei Zhang‡, Yezhou Yang†, Zicheng Liu‡
+
+†Arizona State University, ‡Microsoft Corporation {zy.fang, yz.yang}@asu.edu {jianfw, lijuanw, leizhang, zliu}@microsoft.com
+
+# ABSTRACT
+
+This paper is concerned with self-supervised learning for small models. The problem is motivated by our empirical studies that while the widely used contrastive self-supervised learning method has shown great progress on large model training, it does not work well for small models. To address this problem, we propose a new learning paradigm, named SElf-SupErvised Distillation (SEED), where we leverage a larger network (as Teacher) to transfer its representational knowledge into a smaller architecture (as Student) in a self-supervised fashion. Instead of directly learning from unlabeled data, we train a student encoder to mimic the similarity score distribution inferred by a teacher over a set of instances. We show that SEED dramatically boosts the performance of small networks on downstream tasks. Compared with self-supervised baselines, SEED improves the top-1 accuracy from $4 2 . 2 \%$ to $6 7 . 6 \%$ on EfficientNet-B0 and from $3 6 . 3 \%$ to $6 8 . 2 \%$ on MobileNetV3-Large on the ImageNet-1k dataset.
+
+# 1 INTRODUCTION
+
+The burgeoning studies and success on self-supervised learning (SSL) for visual representation are mainly marked by its extraordinary potency of learning from unlabeled data at scale. Accompanying with the SSL is its phenomenal benefit of obtaining task-agnostic representations while allowing the training to dispense with prohibitively expensive data labeling. Major ramifications of visual SSL include pretext tasks (Noroozi & Favaro, 2016; Zhang et al., 2016; Gidaris et al., 2018; Zhang et al., 2019; Feng et al., 2019), contrastive representation learning (Wu et al., 2018; He et al., 2020; Chen et al., 2020a), online/offline clustering (Yang et al., 2016; Caron et al., 2018; Li et al., 2020; Caron et al., 2020; Grill et al., 2020), etc. Among them, several recent works (He et al., 2020; Chen et al., 2020a; Caron et al., 2020) have achieved comparable or even better accuracy than the supervised pre-training when transferring to downstream tasks, e.g. semi-supervised classification, object detection.
+
+The aforementioned top-performing SSL algorithms all involve large networks (e.g., ResNet-50 (He et al., 2016) or larger), with, however, little attention on small networks. Empirically, we find that existing techniques like contrastive learning do not work well on small networks. For instance, the linear probe top-1 accuracy on ImageNet using MoCo-V2 (Chen et al., 2020c) is only $3 6 . 3 \%$ with MobileNetV3-Large (see Figure 1), which is much lower compared with its supervised training accuracy
+
+![](images/d4fb00f69f184e9fccdf211a9f298a6ff3e7571f5dfb94ff7b3d5393734ffc54.jpg)  
+Figure 1: SEED vs. MoCo-V2 (Chen et al., 2020c)) on ImageNet-1K linear probe accuracy. The vertical axis is the top-1 accuracy and the horizontal axis is the number of learnable parameters for different network architectures. Directly applying self-supervised contrastive learning (MoCo-V2) does not work well for smaller architectures, while our method (SEED) leads to dramatic performance boost. Details of the setting can be found in Section 4.
+
+$7 5 . 2 \%$ (Howard et al., 2019). For EfficientNet-B0, the accuracy is $4 2 . 2 \%$ compared with its supervised training accuracy $7 7 . 1 \%$ (Tan & Le, 2019). We conjecture that this is because smaller models with fewer parameters cannot effectively learn instance level discriminative representation with large amount of data.
+
+To address this challenge, we inject knowledge distillation (KD) (Bucilua et al. ˇ , 2006; Hinton et al., 2015) into self-supervised learning and propose self-supervised distillation (dubbed as SEED) as a new learning paradigm. That is, train the larger, and distill to the smaller both in self-supervised manner. Instead of directly conducting self-supervised training on a smaller model, SEED first trains a large model (as the teacher) in a self-supervised way, and then distills the knowledge to the smaller model (as the student). Note that the conventional distillation is for supervised learning, while the distillation here is in the self-supervised setting without any labeled data. Supervised distillation can be formulated as training a student to mimic the probability mass function over classes predicted by a teacher model. In unsupervised knowledge distillation setting, however, the distribution over classes is not directly attainable. Therefore, we propose a simple yet effective self-supervised distillation method. Similar to (He et al., 2020; Wu et al., 2018), we maintain a queue of data samples. Given an instance, we first use the teacher network to obtain its similarity scores with all the data samples in the queue as well as the instance itself. Then the student encoder is trained to mimic the similarity score distribution inferred by the teacher over these data samples.
+
+The simplicity and flexibility that SEED brings are self-evident. 1) It does not require any clustering/prototypical computing procedure to retrieve the pseudo-labels or latent classes. 2) The teacher model can be pre-trained with any advanced SSL approach, e.g., MoCo-V2 (Chen et al., 2020c), SimCLR (Chen et al., 2020a), SWAV (Caron et al., 2020). 3) The knowledge can be distilled to any target small networks (either shallower, thinner, or totally different architectures).
+
+To demonstrate the effectiveness, we comprehensively evaluate the learned representations on series of downstream tasks, e.g., fully/semi-supervised classification, object detection, and also assess the transferability to other domains. For example, on ImageNet-1k dataset, SEED improves the linear probe accuracy of EfficientNet-B0 from $4 2 . 2 \%$ to $6 7 . 6 \%$ (a gain over $2 5 \%$ ), and MobileNet-V3 from $3 6 . 3 \%$ to $6 8 . 2 \%$ (a gain over $3 1 \%$ ) compared to MoCo-V2 baselines, as shown in Figure 1 and Section 4.
+
+Our contributions can be summarized as follows:
+
+• We are the first to address the problem of self-supervised visual representation learning for small models. We propose a self-supervised distillation (SEED) technique to transfer knowledge from a large model to a small model without any labeled data. With the proposed distillation technique (SEED), we significantly improve the state-of-theart SSL performance on small models.   
+• We exhaustively compare a variety of distillation strategies to show the validity of SEED under multiple settings.
+
+# 2 RELATED WORK
+
+Among the recent literature in self-supervised learning, contrastive based approaches show prominent results on downstream tasks. Majority of the techniques along this direction are stemming from noise-contrastive estimation (Gutmann & Hyvärinen, 2010) where the latent distribution is estimated by contrasting with randomly or artificially generated noises. Oord et al. (2018) first proposed Info-NCE to learn image representations by predicting the future using an auto-regressive model for unsupervised learning. Follow-up works include improving the efficiency (Hénaff et al., 2019), and using multi-view as positive samples (Tian et al., 2019b). As these approaches can only have the access to limited negative instances, Wu et al. (2018) designed a memory-bank to store the previously seen random representations as negative samples, and treat each of them as independent categories (instance discrimination). However, this approach also comes with a deficiency that the previously stored vectors are inconsistent with the recently computed representations during the earlier stage of pre-training. Chen et al. (2020a) mitigate this issue by sampling negative samples from a large batch. Concurrently, He et al. (2020) improve the memory-bank based method and propose to use the momentum updated encoder for the remission of representation inconsistency. Other techniques include Misra & Maaten (2020) that combines the pretext-invariant objective loss with contrastive learning, and Wang & Isola (2020) that decomposes contrastive loss into alignment and uniformity objectiveness.
+
+Knowledge distillation (Hinton et al., 2015) aims to transfer knowledge from a cumbersome model to a smaller one without losing too much generalization power, which is also well investigated in model compression (Bucilua et al. ˇ , 2006). Instead of mimicking the teacher’s output logit, attention transfer (Zagoruyko & Komodakis, 2016) formulates knowledge distillation on attention maps. Similarly, works in (Ahn et al., 2019; Yim et al., 2017; Koratana et al., 2019; Huang & Wang, 2017) have utilized different learning objectives including consistency on feature maps, consistency on probability mass function, and maximizing the mutual information. CRD (Tian et al., 2019a), which is derived from CMC (Tian et al., 2019b), optimizes the student network by a similar objective to Oord et al. (2018) using a derived lower bound on mutual information. However, the aforementioned efforts all focus on task-specific distillation (e.g., image classification) during the fine-tuning phase rather than a task-agnostic distillation in the pre-training phase for the representation learning. Several works on natural language pre-training proposed to leverage knowledge distillation for a smaller yet stronger small models. For instances, DistillBert (Sanh et al., 2019), TinyBert (Jiao et al., 2019), and MobileBert (Sun et al., 2020), have used knowledge distillation for model compression and shown their validity on multiple downstream tasks. Similar works also emphasize the value of smaller and faster models for language representation learning by leveraging knowledge distillation (Turc et al., 2019; Sun et al., 2019). These works all demonstrate the effectiveness of knowledge distillation for language representation learning in small models, while are not extended to the pre-training for visual representations. Notably, a recent concurrent work CompRess (Abbasi Koohpayegani et al., 2020) also point out the importance to develop better SSL method for smaller models. SEED closely relates to the above techniques but aims to facilitate visual representation learning during pre-training phase using distillation technique for small models, which as far as we know has not yet been investigated.
+
+# 3 METHOD
+
+# 3.1 PRELIMINARY ON KNOWLEDGE DISTILLATION
+
+Knowledge distillation (Hinton et al., 2015; Bucilua et al.ˇ , 2006) is an effective technique to transfer knowledge from a strong teacher network to a target student network. The training task can be generalized as the following formulation:
+
+$$
+\hat { \theta } _ { S } = \underset { \theta _ { S } } { \arg \operatorname* { m i n } } \sum _ { i } ^ { N } \mathcal { L } _ { \mathrm { s u p } } ( \mathbf { x } _ { i } , \theta _ { S } , y _ { i } ) + \mathcal { L } _ { \mathrm { d i s t i l l } } ( \mathbf { x } _ { i } , \theta _ { S } , \theta _ { T } ) ,
+$$
+
+where $\mathbf { x } _ { i }$ is an image, $y _ { i }$ is the corresponding annotation, $\theta _ { S }$ is the parameter set for the student network, and $\theta _ { T }$ is the set for the teacher network. The loss $\mathcal { L } _ { \mathrm { s u p } }$ is the alignment error between the network prediction and the annotation. For example in image classification task (Mishra & Marr, 2017; Shen & Savvides, 2020; Polino et al., 2018; Cho & Hariharan, 2019), it is normally a cross entropy loss. For object detection (Liu et al., 2019; Chen et al., 2017), it includes bounding box regression as well. The loss of ${ \mathcal { L } } _ { \mathrm { d i s t i l l } }$ is the mimic error of the student network towards a pre-trained teacher network. For example in (Hinton et al., 2015), the teacher signal comes from the softmax prediction of multiple large-scale networks and the loss is measured by the Kullback–Leibler divergence. In Romero et al. (2014), the task is to align the intermediate feature map values and to minimize the squared $l 2$ distance. The effectiveness has been well demonstrated in the supervised setting with labeled data, but remains unknown for the unsupervised setting, which is our focus.
+
+# 3.2 SELF-SUPERVISED DISTILLATION FOR VISUAL REPRESENTATION
+
+Different from supervised distillation, SEED aims to transfer knowledge from a large model to a small model without requiring labeled data, so that the learned representations in small model can be used for downstream tasks. Inspired by contrastive SSL, we formulate a simple approach for the distillation on the basis of instance similarity distribution over a contrastive instance queue. Similar to He et al. (2020), we maintain an instance queue for storing data samples’ encoding output from the teacher. Given a new sample, we compute its similarity scores with all the samples in the queue using both the teacher and the student models. We require that the similarity score distribution computed by the student matches with that computed by the teacher, which is formulated as minimizing the cross entropy between the student and the teacher’s similarity score distributions (as illustrated in Figure 2).
+
+![](images/2f2a03a1e9389d9452c8beee5f2f9f477f5576361541704a7434d89396409316.jpg)  
+Figure 2: Illustration of our self-supervised distillation pipeline. The teacher encoder is pre-trained by SSL and kept frozen during the distillation. The student encoder is trained by minimizing the cross entropy of probabilities from teacher & student for an augmented view of an image, computed over a dynamically maintained queue.
+
+Specifically, for a randomly augmented view $\mathbf { x } _ { i }$ of an image, it is first mapped and normalized into feature vector representations $\mathbf { z } _ { i } ^ { T } = f _ { \theta } ^ { T } ( \mathbf { x } _ { i } ) / \vert \vert f _ { \theta } ^ { T } ( \mathbf { x } _ { i } ) \vert \vert _ { 2 }$ , and $\mathbf { z } _ { i } ^ { S } = f _ { \theta } ^ { S } ( \mathbf { \bar { x } } _ { i } ^ { \star } ) / \vert \vert f _ { \theta } ^ { S } ( \mathbf { x } _ { i } ) \vert \vert _ { 2 }$ , where $\mathbf { z } _ { i } ^ { T } , \mathbf { z } _ { i } ^ { S } \in \mathbb { R } ^ { D }$ , and $f _ { \theta } ^ { T }$ and $f _ { \theta } ^ { S }$ θ θ θ θdenote the teacher and student encoders, respectively. Let $\mathbf { D } =$ $[ \mathbf { d } _ { 1 } . . . \mathbf { d } _ { K } ]$ denote the instance queue where $K$ is the queue length and ${ \bf d } _ { j }$ is the feature vector obtained from the teacher encoder. Similar to the contrastive learning framework, $\mathbf { D }$ is progressively updated under the “first-in first-out” strategy as distillation proceeds. That is, we en-queue the visual features of the current batch inferred by the teacher and de-queue the earliest seen samples at the end of iteration. Note that the maintained samples in queue $\mathbf { D }$ are mostly random and irrelevant to the target instance $\mathbf { x } _ { i }$ . Minimizing the cross entropy between the similarity score distribution computed by the student and teacher based on $\mathbf { D }$ softly contrasts $\mathbf { x } _ { i }$ with randomly selected samples, without directly aligning with the teacher encoder. To address this problem, we add the teacher’s embedding $( \mathbf { z } _ { i } ^ { T } )$ into the queue and form $\mathbf { D } ^ { + } = [ \mathbf { d } _ { 1 } . . . \mathbf { d } _ { K } , \mathbf { d } _ { K + 1 } ]$ with $\mathbf { d } _ { K + 1 } = \mathbf { z } _ { i } ^ { T }$ .
+
+Let $\mathbf { p } ^ { T } ( \mathbf { x } _ { i } ; \boldsymbol { \theta } _ { T } ; \mathbf { D } ^ { + } )$ denote the similarity score between the extracted teacher feature $\mathbf { z } _ { i } ^ { T }$ and ${ \bf d } _ { j }$ ’s $( j = 1 , . . . , K + 1 )$ computed by the teacher model. $\mathbf { p } ^ { T } ( \mathbf { x } _ { i } ; \boldsymbol { \theta } _ { T } ; \mathbf { D } ^ { + } )$ is defined as
+
+$$
+\mathbf { p } ^ { T } ( \mathbf { x } _ { i } ; \theta _ { T } , \mathbf { D } ^ { + } ) = \left[ p _ { 1 } ^ { T } \dots p _ { K + 1 } ^ { T } \right] , \qquad p _ { j } ^ { T } = \frac { \exp ( \mathbf { z } _ { i } ^ { T } \cdot \mathbf { d } _ { j } / \tau ^ { T } ) } { \sum _ { \mathbf { d } \sim \mathbf { D } ^ { + } } \exp ( \mathbf { z } _ { i } ^ { T } \cdot \mathbf { d } / \tau ^ { T } ) } ,
+$$
+
+and $\tau ^ { T }$ is a temperature parameter for the teacher. Note, we use $( ) ^ { T }$ to represent the feature from the teacher network and use $( \cdot )$ to represent the inner product between two features.
+
+Similarly let $\mathbf { p } ^ { S } ( x _ { i } ; \theta _ { S } , \mathbf { D } ^ { + } )$ denote the similarity score computed by the student model, which is defined as
+
+$$
+\mathbf { p } ^ { S } ( \mathbf { x } _ { i } ; \theta _ { S } , \mathbf { D } ^ { + } ) = \left[ p _ { 1 } ^ { S } \dots p _ { K + 1 } ^ { S } \right] , \qquad { \mathrm { w h e r e ~ } } p _ { j } ^ { S } = { \frac { \exp ( \mathbf { z } _ { i } ^ { S } \cdot \mathbf { d } _ { j } / \tau ^ { S } ) } { \sum _ { \mathbf { d } \sim \mathbf { D } ^ { + } } \exp ( \mathbf { z } _ { i } ^ { S } \cdot \mathbf { d } / \tau ^ { S } ) } } ,
+$$
+
+and $\tau ^ { S }$ is a temperature parameter for the student.
+
+Our self-supervised distillation can be formulated as minimizing the cross entropy between the similarity scores of the teacher, $\mathbf { p } ^ { T } ( \mathbf { x } _ { i } ; \theta _ { T } , \mathbf { D } ^ { + } )$ , and the student, $\mathbf { p } ^ { S } ( \mathbf { x } _ { i } ; \boldsymbol { \theta } _ { S } , \mathbf { D } ^ { \dagger } )$ , over all the instances $\mathbf { x } _ { i }$ , that is,
+
+$$
+\begin{array} { r l } & { \boldsymbol { \hat { \theta } _ { S } } = \underset { \boldsymbol { \theta _ { S } } } { \arg \operatorname* { m i n } } \sum _ { i } ^ { N } - \mathbf { p } ^ { T } ( \mathbf { x } _ { i } ; \boldsymbol { \theta _ { T } } , \mathbf { D } ^ { + } ) \cdot \log \mathbf { p } ^ { S } ( \mathbf { x } _ { i } ; \boldsymbol { \theta _ { S } } , \mathbf { D } ^ { + } ) } \\ & { \quad = \underset { \boldsymbol { \theta _ { S } } } { \arg \operatorname* { m i n } } \sum _ { i } ^ { N } \sum _ { j } ^ { K + 1 } - \frac { \exp \left( \mathbf { z } _ { i } ^ { T } \cdot \mathbf { d } _ { j } / \tau ^ { T } \right) } { \sum _ { \mathbf { d } \sim \mathbf { D } ^ { + } } \exp \left( \mathbf { z } _ { i } ^ { T } \cdot \mathbf { d } / \tau ^ { T } \right) } \cdot \log \frac { \exp \left( \mathbf { z } _ { i } ^ { S } \cdot \mathbf { d } _ { j } / \tau ^ { S } \right) } { \sum _ { \mathbf { d } \sim \mathbf { D } ^ { + } } \exp \left( \mathbf { z } _ { i } ^ { S } \cdot \mathbf { d } / \tau ^ { S } \right) } . } \end{array}
+$$
+
+Since the teacher network is pre-trained and frozen, the queued features are consistent during training w.r.t. the student network. The higher the value of $p _ { j } ^ { I ^ { \prime } }$ is, the larger weight will be laid on $p _ { j } ^ { S }$ . Due to the l2 normalization, similarity score between ${ \mathbf z } _ { i } ^ { T }$ and ${ \bf d } _ { K + 1 }$ remains constant 1 before softmax normalization, which is the largest among $p _ { j } ^ { T }$ . Thus, the weight for $p _ { K + 1 } ^ { S }$ is the largest and can be adjusted solely by tuning the value of $\tau ^ { T }$ . By minimizing the loss, the feature of $\mathbf { z } _ { i } ^ { S }$ can be aligned with ${ \mathbf z } _ { i } ^ { T }$ and meanwhile contrasts with other unrelated image features in $\mathbf { D }$ . We further discuss the relation of these two goals with our learning objective in Appendix A.5.
+
+Relations with Info-NCE loss. When $\tau ^ { T } \to 0$ , the softmax function for $\mathbf { p } ^ { T }$ smoothly approaches to a one-hot vector, where $p _ { K + 1 } ^ { T }$ equals 1 and all others 0. In this extreme case, the loss becomes
+
+$$
+\mathcal { L } _ { N C E } = \sum _ { i } ^ { N } - \log \frac { \exp ( \mathbf { z } _ { i } ^ { T } \cdot \mathbf { z } _ { i } ^ { S } / \tau ) } { \sum _ { \mathbf { d } \sim \mathbf { D } ^ { + } } \exp ( \mathbf { z } _ { i } ^ { S } \cdot \mathbf { d } / \tau ) } ,
+$$
+
+which is similar to the widely-used Info-NCE loss (Oord et al., 2018) in contrastive-based SSL (see discussion in Appendix A.6.
+
+# 4 EXPERIMENT
+
+# 4.1 PRE-TRAINING
+
+Self-Supervised Pre-training of Teacher Network. By default, we use MoCo-V2 (Chen et al., 2020c) to pre-train the teacher network. Following (Chen et al., 2020a), we use ResNet as the network backbone with different depths/widths and append a multi-layer-perceptron (MLP) layer (two linear layers and one ReLU (Nair & Hinton, 2010) activation layer in between) at the end of the encoder after average pooling. The dimension of the last feature dimension is 128. All teacher networks are pre-trained for 200 epochs due to the computational limitation unless explicitly specified. As our distillation is independent with the teacher pre-training algorithm, we also show results with other self-supervised pre-trained models for teacher network, e.g., SWAV (Caron et al., 2020), SimCLR (Chen et al., 2020a).
+
+Self-Supervised Distillation on Student Network. We choose multiple smaller networks with fewer learnable parameters as the student network: MobileNet-v3-Large (Howard et al., 2017), EfficientNet-B0 (Tan & Le, 2019), and smaller ResNet with fewer layers (ResNet-18, 34). Similar to the pre-training for teacher network, we add one additional MLP layer on the basis of the student network. Our distillation is trained with a standard SGD optimizer with momentum 0.9 and a weight decay parameter of 1e-4 for 200 epochs. The initial learning rate is set as 0.03 and updated by a cosine decay scheduler (Nair & Hinton, 2010) with 5 warm-up epochs and batch size 256. In Eq. 4, the teacher temperature is set as $\tau ^ { T } = 0 . 0 1$ and the student temperature is $\tau ^ { S } = 0 . 2$ . The queue size of $K$ is 65,536. In the following subsections and appendix, we also show results with different hyper-parameter values, e.g., for $\tau ^ { T }$ and $K$ .
+
+# 4.2 FINE-TUNING AND EVALUATION
+
+In order to validate the effectiveness of self-supervised distillation, we choose to assess the performance of representations of the student encoder on several downstream tasks. We first report its performances of linear evaluation and semi-supervised linear evaluation on the ImageNet ILSVRC2012 (Deng et al., 2009) dataset. To measure the feature transferability brought by distillation, we also conduct evaluations on other tasks, which include object detection and segmentation on the VOC07 (Everingham et al.) and MS-COCO (Lin et al., 2014) datasets. At the end, we compare the transferability of the features learned by distillation with ordinary self-supervised contrastive learning on the tasks of linear classification on datasets from different domains.
+
+Linear and KNN Evaluation on ImageNet. We conduct the supervised linear classification on ImageNet-1K, which contains ${ \sim } 1 . 3 \mathbf { M }$ images for training, and 50,000 images for validation, spanning 1,000 categories. Following previous works in (He et al., 2020; Chen et al., 2020a), we train a single linear layer classifier on top of the frozen network encoder after self-supervised pretraining/distillation. SGD optimizer is used to train the linear classifier for 100 epochs with weight decay to be 0. The initial learning rate is set as 30 and is then reduced by a factor of 10 at 60 and 80 epochs (similar as in Tian et al. (2019a)). Notably, when training the linear classifier for MobileNet and EfficientNet, we reduce the initial learning rate to 3. The results are reported in terms of Top-1 and Top-5 accuracy. We also perform classification using $K$ -Nearest Neighbors (KNN) based on the learned 128d vector from the last MLP layer. The sample is classified by taking the most frequent label of its $K$ ( $K = 1 0$ ) nearest neighbors.
+
+Table 1: ImageNet-1k test accuracy $( \% )$ using KNN and linear classification for multiple students and MoCov2 pre-trained deeper teacher architectures. $\pmb { \chi }$ denotes MoCo-V2 self-supervised learning baselines before distillation. \* indicates using a deeper teacher encoder pre-trained by SWAV, where additional small-patches are also utilized during distillation and trained for 800 epochs. $K$ denotes Top-1 accuracy using KNN. T-1 and T-5 denote Top-1 and Top-5 accuracy using linear evaluation. First column shows Top-1 Acc. of Teacher network. First row shows the supervised performances of student networks.   
+
+<table><tr><td rowspan="2">S T</td><td rowspan="2">T-1</td><td rowspan="2">K</td><td colspan="2">Eff-bo</td><td colspan="3">Eff-b1</td><td colspan="3">Mob-v3</td><td colspan="3">R-18</td><td colspan="3">R-34</td></tr><tr><td>T-1</td><td>T-5</td><td>K</td><td>T-1</td><td>T-5</td><td>K</td><td>T-1</td><td>T-5</td><td>K</td><td>T-1</td><td>T-5</td><td>K</td><td>T-1</td><td>T-5</td></tr><tr><td>Supervised Acc.</td><td></td><td></td><td>77.3</td><td></td><td></td><td>79.2</td><td></td><td></td><td>75.2</td><td></td><td></td><td>72.1</td><td></td><td></td><td>75.0</td><td></td></tr><tr><td>X</td><td></td><td>30.0</td><td>42.2</td><td></td><td>68.534.4</td><td>50.7</td><td>74.6</td><td>27.5</td><td>36.3</td><td>62.2</td><td>36.7</td><td>52.5</td><td>77.0</td><td>）41.5</td><td>57.4</td><td>81.6</td></tr><tr><td>R-50 △</td><td>67.4</td><td>46.0 +16.0</td><td>61.3 +19.1</td><td>82.7 +14.2</td><td>46.1 +16.1</td><td>61.4 +10.7</td><td>83.1 +8.8</td><td>44.8 +17.3</td><td>55.2 +18.9</td><td>80.3 +18.1</td><td>43.4 +6.7</td><td>57.9 +5.1</td><td>82.0 +4.8</td><td>45.2 +3.7</td><td>58.5 +1.1</td><td>82.6 +1.0</td></tr><tr><td>R-101 △</td><td>70.3</td><td>50.1 +20.1</td><td>63.0 +20.8</td><td>83.8 +15.3</td><td>50.3 +15.9</td><td>63.4 +12.7</td><td>84.6 +10.0</td><td>48.8 +21.3</td><td>59.9 +23.6</td><td>83.5 +21.3</td><td>48.6 +11.9</td><td>58.9 +6.4</td><td>82.5 +5.5</td><td>50.5 +9.0</td><td>61.6 +4.2</td><td>84.9 +3.3</td></tr><tr><td>R-152</td><td></td><td>50.7</td><td>65.3</td><td>86.0</td><td>52.4</td><td>67.3</td><td>86.9</td><td>49.5</td><td>61.4</td><td>84.6</td><td>49.1</td><td>59.5</td><td>83.3</td><td>51.4</td><td>62.7</td><td>85.8</td></tr><tr><td>△</td><td>74.2</td><td>+20.7</td><td>+23.1</td><td>+17.5</td><td>+18.0</td><td>+16.6</td><td>+12.3</td><td>+22.0</td><td>+25.1</td><td>+22.4</td><td>+12.4</td><td>+7.0</td><td>+6.3</td><td>+9.9</td><td>+5.3</td><td>+4.2</td></tr><tr><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td></tr><tr><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td></tr><tr><td>R50×2*</td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td></tr><tr><td></td><td></td><td></td><td></td><td>87.4</td><td>60.3</td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td></tr><tr><td>△</td><td>77.3</td><td>57.4 +27.4</td><td>67.6 +25.4</td><td>+18.9</td><td>+25.9</td><td>68.0 +17.3</td><td>87.6 +13.0</td><td>55.9 +18.9</td><td>68.2 +31.9</td><td>88.2 +26.0</td><td>55.3 +18.6</td><td>63.0 +10.5</td><td>84.9 +7.9</td><td>58.2 +16.7</td><td>65.7 +8.3</td><td>86.8 +5.2</td></tr></table>
+
+![](images/3623a3702ec553403c1eb676464c6cd00d584bc812cc122bd666f370eb648b7c.jpg)  
+Figure 3: ImageNet-1k Top-1 accuracy for semi-supervised evaluations using $1 \%$ (red line), $10 \%$ (blue line) of the annotations for linear fine-tuning, in comparison with the fully supervised (green line) linear evaluation baseline for SEED. For the points whose Teacher’s number of parameters is at 0, we show the semi-supervised linear evaluation results of MoCo-V2 without any distillation. The Student models tend to perform better on the semi-supervised tasks after distillation from larger Teachers.
+
+Table 1 shows the results with various teacher networks and student networks. We list the baseline of contrastive self-supervised pre-training using MoCo-V2 (Chen et al., 2020c) in the first row for each student architecture. We can see clearly that smaller networks perform rather worse. For example, MobileNet-V3 can only reach $3 6 . 3 \%$ . This aligns well with previous conclusions from (Chen et al., 2020a;b) that bigger models are desired to perform better in contrastive-based self-supervised pretraining. We conjecture that this is mainly caused by the inability of smaller network to discriminate instances in a large-scale dataset. The results also clearly demonstrate that the distillation from a larger network helps boosting the performances of small networks, and show obvious improvement. For instance, with MoCo-V2 pre-trained ResNet-152 (for 400 epochs) as the teacher network, the Top-1 accuracy of MobileNet-V3-Large can be significantly improved from $3 6 . 3 \%$ to $6 1 . 4 \%$ . Furthermore, we use ResNet- $5 0 \times 2$ (provided in Caron et al. (2020)) as the teacher network and adopt the multi-crop trick (see A.2 for details). The accuracy can be further improved to $6 8 . 2 \%$ (last row of Table 1) for MobileNet-V3-Large with 800 epochs of distillation. We note that the gain benefited from distillation becomes more distinct on smaller architectures and we further study the effect of various teacher models in ablations.
+
+Semi-Supervised Evaluation on ImageNet. Following (Oord et al., 2018; Kornblith et al., 2019; Kolesnikov et al., 2019), we evaluate the representation on the semi-supervised task, where a fixed $1 \%$ or $10 \%$ subsets of ImageNet training data (Chen et al., 2020a) are provided with the annotations. After the self-supervised learning with and without distillation, we also train a classifier on top of the representation. The results are shown in Figure 3, where the baseline without distillation is depicted when teacher parameters are 0. As we can see, the accuracy is also improved remarkably with SEED distillation, and a stronger teacher network with more parameters leads to a better performed student network.
+
+Table 2: Object detection and instance segmentation results using contrastive self-supervised learning and SEED distillation using ResNet-18 as backbone: bounding-box AP $( \mathbf { A P } ^ { b b } )$ and mask AP $( \mathbf { A P } ^ { m k } )$ evaluated on VOC07-val and COCO testing split. More results on different backbones can be found in the Appendix. Subscript in green represents improvement is larger than 0.3.   
+
+<table><tr><td rowspan="2">S</td><td rowspan="2">T</td><td colspan="3">VOC Obj. Det.</td><td colspan="3">COCO Obj. Det.</td><td colspan="3">COCO Inst. Segm.</td></tr><tr><td>Apbb</td><td>AP</td><td>AP</td><td>Apbb</td><td>AP</td><td>APP</td><td>Apmk</td><td>AP</td><td>AP</td></tr><tr><td></td><td>X</td><td>46.1</td><td>74.5</td><td>48.6</td><td>35.0</td><td>53.9</td><td>37.7</td><td>31.0</td><td>51.1</td><td>33.1</td></tr><tr><td></td><td>R-50</td><td>46.1(0.0)</td><td>74.8(+0.3)</td><td>49.1(+0.5)</td><td>35.3(+0.3)</td><td>54.2(+0.3)</td><td>37.8(+0.1)</td><td>31.1(+0.1)</td><td>51.1(0.0)</td><td>33.2(+0.1)</td></tr><tr><td>R-18 R-101</td><td></td><td>46.8(+0.7)</td><td>75.8(+1.3)</td><td>49.3(+0.7)</td><td>35.3(+0.3)</td><td>54.3(+0.4)</td><td>37.9(+0.2)</td><td>31.3(+0.3)</td><td>51.3(+0.2)</td><td>33.4(+0.3)</td></tr><tr><td></td><td>R-152</td><td>46.8(+0.7)</td><td>75.9(+1.4)</td><td>50.2(+1.6)</td><td>35.4(+0.4)</td><td>54.4(+0.5)</td><td>38.0(+0.3)</td><td>31.3(+0.3)</td><td>51.4(+0.3)</td><td>33.4(+0.3)</td></tr></table>
+
+![](images/15d565abe6e76f1be7191527af52d1cfceecfa87f6083e78a4a48b0e6be4ff99.jpg)  
+Figure 4: ImageNet-1k Accuracy $( \% )$ of student network (EfficientNet-B0 and ResNet-18) transferred to other domains (CIFAR-10, CIFAR-100, SUN-397 datasets) with and without distillation from lager architectures (ResNet-50/101/152).
+
+Transferring to Classification. To further study whether the improvement of the learned representations by distillation is confined to ImageNet, we evaluate on additional classification datasets to study the generalization and transferability of the feature representation. We strictly follow the linear evaluation and fine-tuning settings from (Kornblith et al., 2019; Chen et al., 2020a; Grill et al., 2020), that a linear layer is trained on the basis of frozen features. We report Top-1 Accuracy of models before and after distillation from various architectures on CIFAR-10, CIFAR-100 (Krizhevsky et al., 2009), SUN-397 (Xiao et al., 2010) datasets (see Figure 4). More details regarding pre-processing and training can be found in A.1.2. Notably, we observe that our distillation surpasses contrastive self-supervised pre-training consistently on all benchmarks, verifying the effectiveness of SEED. This also proves the generalization ability of the learned representations from distillation to a wide range of data domain and different classes.
+
+Transferring to Detection and Segmentation. We conduct two downstream tasks here. The first is Faster R-CNN (Ren et al., 2015) model for object detection trained on VOC- $^ { 0 7 + 1 2 }$ train $^ +$ val set and evaluated on VOC-07 test split. The second is Mask R-CNN (He et al., 2017) model for the object detection and instance segmentation on COCO 2017 dataset (Lin et al., 2014). The pre-trained model serves as the initial weight and following He et al. (2020), we fine-tune all the layers of the model. More experiment settings can be found in A.2. The results are illustrated in Table 2. As we can see, on VOC, the distilled pre-trained model achieves a large improvement. With ResNet-152 as the teacher network, the Resnet18-based Faster R-CNN model shows $+ 0 . 7$ point improvement on AP, $+ 1 . 4$ improvement on $\mathrm { { A P } _ { 5 0 } }$ and $+ 1 . 6$ on $\mathsf { A P } _ { 7 5 }$ . On COCO, the improvement is relatively minor and the reason could be that COCO training set has ${ \sim } 1 1 8 \mathrm { k }$ training images while VOC has only ${ \sim } 1 6 . 5 \mathrm { k }$ training images. A larger training set with more fine-tuning iterations reduces the importance of the initial weights.
+
+# 4.3 ABLATION STUDY
+
+We now explore the effects of distillation using different Teacher architectures, Teacher Pre-training algorithms, various distillation strategies and hyper-parameters.
+
+Table 3: ImageNet-1k Accuracy $( \% )$ of student network (ResNet-18) distilled from variants of selfsupervised ResNet-50. P-E/D-E represent the pretraining and distillation epochs. T./S.-Top represent testing accuracy of Teacher and Student. ∗ represents distillation using additional small patches. First row is the ResNet-18 SSL baseline using MoCo-v2 trained for 200 epochs.   
+
+<table><tr><td>Teacher</td><td></td><td>P-E D-E T.Top-1 S.Top-1 S.Top-5</td></tr><tr><td></td><td>X X</td><td>X 52.5 77.0</td></tr><tr><td>X</td><td>200</td><td></td></tr><tr><td>MoCo</td><td>200</td><td>60.6 52.1 77.0</td></tr><tr><td>SimCLR</td><td>200 200</td><td>65.6 57.5 81.7</td></tr><tr><td>MoCo-v2 200 800</td><td>200 67.4</td><td>57.9 82.0</td></tr><tr><td>SWAV 800</td><td>200 71.1</td><td>60.5 83.5</td></tr><tr><td></td><td>100 75.3</td><td>61.1 83.8</td></tr><tr><td>800</td><td>200 75.3</td><td>61.7 84.2</td></tr><tr><td>800</td><td>400 75.3</td><td>62.0 84.4</td></tr><tr><td>SWAV* 800</td><td>200 75.3</td><td>62.6 84.8</td></tr></table>
+
+![](images/8cecf45d517b8c5b96207f2fc90806edfe43935c7cfa9552a0ceb80d0db26208.jpg)  
+Figure 5: Accuracy $( \% )$ of student networks (EfficientNet-b0 and ResNet-18) on ImageNet distilled from wider MoCo-v2 pre-trained ResNet (ResNet-50/101/152×2).
+
+Different Teacher Networks. Figure 5 summarizes the accuracy of ResNet-18 and EfficientNet-B0 distilled from wider and deeper ResNet architectures. We see clear performance improvement as depth and width of teacher network increase: compared to ResNet-50, deeper (ResNet-101) and wider (ResNet- ${ 5 0 \times 2 }$ ) substantially improve the accuracy. However, further architectural enlargement has relatively limited effects, and we suspect the accuracy might be limited by the student network capacity in this case.
+
+Different Teacher Pre-training Algorithms. In Table 3, we show the Top-1 accuracy of ResNet-18 distilled from ResNet-50 with different pre-training algorithms, i.e., MoCo-V1 (He et al., 2020), MoCo-V2 (Chen et al., 2020c), SimCLR (Chen et al., 2020a), and SWAV (Caron et al., 2020)). Notably, the aforementioned methods all unanimously adopt contrastive-based pre-training except SWAV, which is based upon online clustering. We find that our SEED is agnostic to pre-training approaches, making it easy to use any self-supervised models (including clustering-based approach like SWAV) in self-supervised distillation. In addition, we observe that more training epochs for both teacher $S S L$ and distillation epochs can bring beneficial gain.
+
+Other Distillation Strategies. We explore several alternative distillation strategies. l2-Distance: where the $l 2$ -distance of teacher & student’s embeddings are minimized, motivated by Romero et al. (2014). $K$ -Means: we exploit $K$ -Means clustering to assign a pseudo-label based on the teacher network’s representation. Online Clustering: we continuously update the clustering centers during distillation for pseudo-label generation. Binary Contrastive Loss: we adopt an Info-NCE alike loss for contrastive distillation (Tian et al., 2019a). We provide details for other strategies in A.4. Table 4 shows the results for each method on ResNet-18 (student) distilled from ResNet-50. From the results, the simple $l 2$ -distance minimizing approach can achieve a decent accuracy, which demonstrates the effectiveness of applying the distillation idea to the self-supervised learning. Beyond that, we study the effect of the original SSL (MoCo-V2) supervision as supplementary loss to SEED and find it does not bring additional benefits to distillation. We find close results from these two strategies (Top-1 linear Acc.), SEED achieves $5 7 . 9 \%$ , while $\mathbf { S E E D + M o C o - V } 2$ achieves $5 7 . 6 \%$ . This implies that the loss of SEED can to a large extent cover the original $S S L$ loss, and it is not necessary to conduct SSL any further during distillation. Meanwhile, our proposed SEED outperforms these alternatives with highest accuracy, which shows the superiority of aligning the student towards the teacher and contrasting with the irrelevant samples.
+
+Other Hyper-Parameters. Table 5 summarizes the distillation performances on multiple datasets using different temperature $\tau ^ { T }$ . We observe a better performance when decreasing $\tau ^ { T }$ to 0.01 for ImageNet-1k and CIFAR-10 dataset, and to 1e-3 for CIFAR-100 datasets. When $\tau$ is large, the softmax-normalized similarity score of $p _ { j } ^ { T }$ between ${ \mathbf z } _ { i } ^ { T }$ and instance ${ \bf d } _ { j }$ in the queue $\mathbf { D } ^ { + }$ also becomes large, which means the student’s feature should be less discriminative with the features of other images to some extent. When $\tau ^ { T }$ is $_ 0$ , the teacher model will generate a one-hot vector, which only treats ${ \mathbf z } _ { i } ^ { T }$ as a positive instance and all others in the queue as negative. Thus, the best $\tau$ is a trade-off depending on the data distribution. We further compare effect of different hyper-parameters in A.8.
+
+Table 4: Top-1/5 accuracy of linear classification results on ImageNet using different distillation strategies on ResNet-18 (student) and ResNet-50 (teacher) architectures.   
+
+<table><tr><td>Method</td><td colspan="2">Top-1 Acc. Top-5 Acc.</td></tr><tr><td>l2-Distance</td><td>55.3</td><td>80.3</td></tr><tr><td>K-Means</td><td>51.0</td><td>75.8</td></tr><tr><td>Online Clustering</td><td>56.4</td><td>81.2</td></tr><tr><td>Binary Contr. Loss</td><td>57.4</td><td>81.5</td></tr><tr><td>SEED +MoCo-V2</td><td>57.6</td><td>81.8</td></tr><tr><td>SEED</td><td>57.9</td><td>82.0</td></tr></table>
+
+Table 5: Effect of $\tau ^ { T }$ for the distillation of ResNet-18 (student), ResNet-50 (teacher) on multiple datasets.   
+
+<table><tr><td rowspan="2">YT</td><td>ImageNet</td><td>CIFAR-10</td><td>CIFAR-100</td></tr><tr><td>Top-1 Top-5</td><td>Top-1</td><td>Top-1</td></tr><tr><td>0.3</td><td>54.8 80.0</td><td>78.7</td><td>46.6</td></tr><tr><td>0.1</td><td>54.9 80.1</td><td>83.0</td><td>50.1</td></tr><tr><td>0.05</td><td>56.5 81.3</td><td>84.4</td><td>56.2</td></tr><tr><td>0.01</td><td>57.9 82.0</td><td>87.5 </td><td>60.6</td></tr><tr><td>1e-3</td><td>57.6 81.8</td><td>86.9</td><td>60.8</td></tr></table>
+
+# 5 CONCLUSIONS
+
+Self-Supervised Learning is acknowledged for its remarkable ability in learning from unlabeled, and large scale data. However, a critical impedance for the SSL pre-training on smaller architecture comes from its low capacity of discriminating enormous number of instances. Instead of directly learning from unlabeled data, we proposed SEED as a novel self-supervised learning paradigm, which learns representation by self-supervised distillation from a bigger SSL pre-trained model. We show in extensive experiments that SEED effectively addresses the weakness of self-supervised learning for small models and achieves state-of-the-art results on various benchmarks of small architectures.
+
+# REFERENCES
+
+Soroush Abbasi Koohpayegani, Ajinkya Tejankar, and Hamed Pirsiavash. Compress: Self-supervised learning by compressing representations. Advances in Neural Information Processing Systems, 33, 2020.
+
+Sungsoo Ahn, Shell Xu Hu, Andreas Damianou, Neil D Lawrence, and Zhenwen Dai. Variational information distillation for knowledge transfer. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pp. 9163–9171, 2019.
+
+Cristian Bucilua, Rich Caruana, and Alexandru Niculescu-Mizil. Model compression. In ˇ Proceedings of the 12th ACM SIGKDD international conference on Knowledge discovery and data mining, pp. 535–541, 2006.
+
+Mathilde Caron, Piotr Bojanowski, Armand Joulin, and Matthijs Douze. Deep clustering for unsupervised learning of visual features. In Proceedings of the European Conference on Computer Vision (ECCV), pp. 132–149, 2018.
+
+Mathilde Caron, Ishan Misra, Julien Mairal, Priya Goyal, Piotr Bojanowski, and Armand Joulin. Unsupervised learning of visual features by contrasting cluster assignments. arXiv preprint arXiv:2006.09882, 2020.
+
+Guobin Chen, Wongun Choi, Xiang Yu, Tony Han, and Manmohan Chandraker. Learning efficient object detection models with knowledge distillation. In Advances in Neural Information Processing Systems, pp. 742–751, 2017.
+
+Ting Chen, Simon Kornblith, Mohammad Norouzi, and Geoffrey Hinton. A simple framework for contrastive learning of visual representations. arXiv preprint arXiv:2002.05709, 2020a.
+
+Ting Chen, Simon Kornblith, Kevin Swersky, Mohammad Norouzi, and Geoffrey Hinton. Big self-supervised models are strong semi-supervised learners. arXiv preprint arXiv:2006.10029, 2020b.
+
+Xinlei Chen, Haoqi Fan, Ross Girshick, and Kaiming He. Improved baselines with momentum contrastive learning. arXiv preprint arXiv:2003.04297, 2020c.
+
+Jang Hyun Cho and Bharath Hariharan. On the efficacy of knowledge distillation. In Proceedings of the IEEE International Conference on Computer Vision, pp. 4794–4802, 2019.
+
+Jia Deng, Wei Dong, Richard Socher, Li-Jia Li, Kai Li, and Li Fei-Fei. Imagenet: A large-scale hierarchical image database. In 2009 IEEE conference on computer vision and pattern recognition, pp. 248–255. Ieee, 2009.
+
+M. Everingham, L. Van Gool, C. K. I. Williams, J. Winn, and A. Zisserman. The PASCAL Visual Object Classes Challenge 2007 (VOC2007) Results. http://www.pascalnetwork.org/challenges/VOC/voc2007/workshop/index.html.
+
+Zeyu Feng, Chang Xu, and Dacheng Tao. Self-supervised representation learning by rotation feature decoupling. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pp. 10364–10374, 2019.
+
+Spyros Gidaris, Praveer Singh, and Nikos Komodakis. Unsupervised representation learning by predicting image rotations. arXiv preprint arXiv:1803.07728, 2018.
+
+Jean-Bastien Grill, Florian Strub, Florent Altché, Corentin Tallec, Pierre H Richemond, Elena Buchatskaya, Carl Doersch, Bernardo Avila Pires, Zhaohan Daniel Guo, Mohammad Gheshlaghi Azar, et al. Bootstrap your own latent: A new approach to self-supervised learning. arXiv preprint arXiv:2006.07733, 2020.
+
+Yandong Guo, Lei Zhang, Yuxiao Hu, Xiaodong He, and Jianfeng Gao. Ms-celeb-1m: A dataset and benchmark for large-scale face recognition. In European conference on computer vision, pp. 87–102. Springer, 2016.
+
+Michael Gutmann and Aapo Hyvärinen. Noise-contrastive estimation: A new estimation principle for unnormalized statistical models. In Proceedings of the Thirteenth International Conference on Artificial Intelligence and Statistics, pp. 297–304, 2010.
+
+Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image recognition. In Proceedings of the IEEE conference on computer vision and pattern recognition, pp. 770–778, 2016.
+
+Kaiming He, Georgia Gkioxari, Piotr Dollár, and Ross Girshick. Mask r-cnn. In Proceedings of the IEEE international conference on computer vision, pp. 2961–2969, 2017.
+
+Kaiming He, Haoqi Fan, Yuxin Wu, Saining Xie, and Ross Girshick. Momentum contrast for unsupervised visual representation learning. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp. 9729–9738, 2020.
+
+Olivier J Hénaff, Aravind Srinivas, Jeffrey De Fauw, Ali Razavi, Carl Doersch, SM Eslami, and Aaron van den Oord. Data-efficient image recognition with contrastive predictive coding. arXiv preprint arXiv:1905.09272, 2019.
+
+Geoffrey Hinton, Oriol Vinyals, and Jeff Dean. Distilling the knowledge in a neural network. arXiv preprint arXiv:1503.02531, 2015.
+
+Andrew Howard, Mark Sandler, Grace Chu, Liang-Chieh Chen, Bo Chen, Mingxing Tan, Weijun Wang, Yukun Zhu, Ruoming Pang, Vijay Vasudevan, et al. Searching for mobilenetv3. In Proceedings of the IEEE International Conference on Computer Vision, pp. 1314–1324, 2019.
+
+Andrew G Howard, Menglong Zhu, Bo Chen, Dmitry Kalenichenko, Weijun Wang, Tobias Weyand, Marco Andreetto, and Hartwig Adam. Mobilenets: Efficient convolutional neural networks for mobile vision applications. arXiv preprint arXiv:1704.04861, 2017.
+
+Zehao Huang and Naiyan Wang. Like what you like: Knowledge distill via neuron selectivity transfer. arXiv preprint arXiv:1707.01219, 2017.
+
+Xiaoqi Jiao, Yichun Yin, Lifeng Shang, Xin Jiang, Xiao Chen, Linlin Li, Fang Wang, and Qun Liu. Tinybert: Distilling bert for natural language understanding. arXiv preprint arXiv:1909.10351, 2019.
+
+Alexander Kolesnikov, Xiaohua Zhai, and Lucas Beyer. Revisiting self-supervised visual representation learning. In Proceedings of the IEEE conference on Computer Vision and Pattern Recognition, pp. 1920–1929, 2019.
+
+Animesh Koratana, Daniel Kang, Peter Bailis, and Matei Zaharia. Lit: Learned intermediate representation training for model compression. In International Conference on Machine Learning, pp. 3509–3518, 2019.
+
+Simon Kornblith, Jonathon Shlens, and Quoc V Le. Do better imagenet models transfer better? In Proceedings of the IEEE conference on computer vision and pattern recognition, pp. 2661–2671, 2019.
+
+Alex Krizhevsky, Geoffrey Hinton, et al. Learning multiple layers of features from tiny images. 2009.
+
+Junnan Li, Pan Zhou, Caiming Xiong, Richard Socher, and Steven CH Hoi. Prototypical contrastive learning of unsupervised representations. arXiv preprint arXiv:2005.04966, 2020.
+
+Tsung-Yi Lin, Michael Maire, Serge Belongie, James Hays, Pietro Perona, Deva Ramanan, Piotr Dollár, and C Lawrence Zitnick. Microsoft coco: Common objects in context. In European conference on computer vision, pp. 740–755. Springer, 2014.
+
+Yifan Liu, Ke Chen, Chris Liu, Zengchang Qin, Zhenbo Luo, and Jingdong Wang. Structured knowledge distillation for semantic segmentation. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pp. 2604–2613, 2019.
+
+Asit Mishra and Debbie Marr. Apprentice: Using knowledge distillation techniques to improve low-precision network accuracy. arXiv preprint arXiv:1711.05852, 2017.
+
+Ishan Misra and Laurens van der Maaten. Self-supervised learning of pretext-invariant representations. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp. 6707–6717, 2020.
+
+Vinod Nair and Geoffrey E Hinton. Rectified linear units improve restricted boltzmann machines. In ICML, 2010.
+
+Mehdi Noroozi and Paolo Favaro. Unsupervised learning of visual representations by solving jigsaw puzzles. In European Conference on Computer Vision, pp. 69–84. Springer, 2016.
+
+Aaron van den Oord, Yazhe Li, and Oriol Vinyals. Representation learning with contrastive predictive coding. arXiv preprint arXiv:1807.03748, 2018.
+
+Adam Paszke, Sam Gross, Francisco Massa, Adam Lerer, James Bradbury, Gregory Chanan, Trevor Killeen, Zeming Lin, Natalia Gimelshein, Luca Antiga, et al. Pytorch: An imperative style, high-performance deep learning library. In Advances in neural information processing systems, pp. 8026–8037, 2019.
+
+Antonio Polino, Razvan Pascanu, and Dan Alistarh. Model compression via distillation and quantization. arXiv preprint arXiv:1802.05668, 2018.
+
+Shaoqing Ren, Kaiming He, Ross Girshick, and Jian Sun. Faster r-cnn: Towards real-time object detection with region proposal networks. In Advances in neural information processing systems, pp. 91–99, 2015.
+
+Adriana Romero, Nicolas Ballas, Samira Ebrahimi Kahou, Antoine Chassang, Carlo Gatta, and Yoshua Bengio. Fitnets: Hints for thin deep nets. 2014.
+
+Mark Sandler, Andrew Howard, Menglong Zhu, Andrey Zhmoginov, and Liang-Chieh Chen. Mobilenetv2: Inverted residuals and linear bottlenecks. In Proceedings of the IEEE conference on computer vision and pattern recognition, pp. 4510–4520, 2018.
+
+Victor Sanh, Lysandre Debut, Julien Chaumond, and Thomas Wolf. Distilbert, a distilled version of bert: smaller, faster, cheaper and lighter. arXiv preprint arXiv:1910.01108, 2019.
+
+Zhiqiang Shen and Marios Savvides. Meal v2: Boosting vanilla resnet-50 to $80 \text{‰}$ top-1 accuracy on imagenet without tricks. arXiv preprint arXiv:2009.08453, 2020.
+
+Siqi Sun, Yu Cheng, Zhe Gan, and Jingjing Liu. Patient knowledge distillation for bert model compression. arXiv preprint arXiv:1908.09355, 2019.
+
+Zhiqing Sun, Hongkun Yu, Xiaodan Song, Renjie Liu, Yiming Yang, and Denny Zhou. Mobilebert: a compact task-agnostic bert for resource-limited devices. arXiv preprint arXiv:2004.02984, 2020.
+
+Mingxing Tan and Quoc V Le. Efficientnet: Rethinking model scaling for convolutional neural networks. arXiv preprint arXiv:1905.11946, 2019.
+
+Yonglong Tian, Dilip Krishnan, and Phillip Isola. Contrastive representation distillation. In International Conference on Learning Representations, 2019a.
+
+Yonglong Tian, Dilip Krishnan, and Phillip Isola. Contrastive multiview coding. arXiv preprint arXiv:1906.05849, 2019b.
+
+Iulia Turc, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. Well-read students learn better: On the importance of pre-training compact models. arXiv preprint arXiv:1908.08962, 2019.
+
+Tongzhou Wang and Phillip Isola. Understanding contrastive representation learning through alignment and uniformity on the hypersphere. arXiv preprint arXiv:2005.10242, 2020.
+
+Yuxin Wu, Alexander Kirillov, Francisco Massa, Wan-Yen Lo, and Ross Girshick. Detectron2, 2019.
+
+Zhirong Wu, Yuanjun Xiong, Stella X Yu, and Dahua Lin. Unsupervised feature learning via nonparametric instance discrimination. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pp. 3733–3742, 2018.
+
+Jianxiong Xiao, James Hays, Krista A Ehinger, Aude Oliva, and Antonio Torralba. Sun database: Large-scale scene recognition from abbey to zoo. In 2010 IEEE computer society conference on computer vision and pattern recognition, pp. 3485–3492. IEEE, 2010.
+
+Jianwei Yang, Devi Parikh, and Dhruv Batra. Joint unsupervised learning of deep representations and image clusters. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pp. 5147–5156, 2016.
+
+Sergey Zagoruyko and Nikos Komodakis. Paying more attention to attention: Improving the performance of convolutional neural networks via attention transfer. arXiv preprint arXiv:1612.03928, 2016.
+
+Xiaohua Zhai, Avital Oliver, Alexander Kolesnikov, and Lucas Beyer. S4l: Self-supervised semisupervised learning. In Proceedings of the IEEE international conference on computer vision, pp. 1476–1485, 2019.
+
+Liheng Zhang, Guo-Jun Qi, Liqiang Wang, and Jiebo Luo. Aet vs. aed: Unsupervised representation learning by auto-encoding transformations rather than data. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pp. 2547–2555, 2019.
+
+Richard Zhang, Phillip Isola, and Alexei A Efros. Colorful image colorization. In European conference on computer vision, pp. 649–666. Springer, 2016.
+
+Xiao Zhang, Zhiyuan Fang, Yandong Wen, Zhifeng Li, and Yu Qiao. Range loss for deep face recognition with long-tailed training data. In Proceedings of the IEEE International Conference on Computer Vision, pp. 5409–5418, 2017.
+
+# A APPENDIX
+
+We discuss more details and different hyperparameters for SEED during distillation.
+
+A.1 PSEUDO-IMPLEMENTATIONS
+
+We provide pseudo-code of the SEED distillation in PyTorch Paszke et al. (2019) style:
+
+‘‘‘Q: maintaining queue of previous representations: (N X D)   
+2 T: Cumbersome encoder as Teacher.   
+3 S: Target encoder as Student.   
+4 temp_T, temp_S: temperatures of the Teacher & Student.   
+11   
+6   
+7 # activate evaluation mode for Teacher to freeze BN and updation.   
+8 T.eval()   
+9   
+10 for images in enumerate(loader): # Enumerate single crop-view   
+11   
+12 # augment image to get one identical view   
+13 images $=$ aug(images)   
+14   
+15 # Batch-size   
+16 $\begin{array} { r l } { \mathrm { B } } & { { } = } \end{array}$ images.shape[0]   
+17   
+18 # extract embedding from S: 1 X D   
+19 $\mathrm { ~ \cal ~ X ~ \_ ~ S ~ ~ } =$ S(images)   
+20 $\mathrm { ~ \cal ~ X ~ \_ ~ S ~ ~ } =$ torch.norm(X_S, $\mathrm { p } { = } 2$ , dim $^ { = 1 }$ )   
+21   
+22 # use the gradient-free mode   
+23 with torch.no_grad():   
+24 X_T $=$ T(image) # embedding from T: 1 X D   
+25 $\mathrm { ~ X ~ \_ ~ T ~ } =$ torch.norm(X_T, $\mathrm { p } { = } 2$ , dim $^ { = 1 }$ )   
+26   
+27 # insert the current batch embedding from T   
+28 enqueue(Q, X_T)   
+29   
+30 # probability scores distribution for T, S: B X (N + 1)   
+31 S_Dist $=$ torch.einsum(’bd, dn -> bn’, [X_S], Q.t().clone().detach())   
+32 T_Dist $=$ torch.einsum(’bd, dn -> bn’, [X_T], Q.t().clone().detach())   
+33   
+34 # Apply temperatures for soft-labels   
+35 S_Dist / $=$ temp_S   
+36 T_Dist $=$ SoftMax(T_Dist/temp_T, dim $^ { = 1 }$ )   
+37   
+38 # loss computation, use log_softmax for stable computation   
+39 loss $=$ -torch.mul(T_Dist, Log_SoftMax(S_Dist, dim $^ { 1 = 1 }$ )).sum()/B   
+40   
+41 # update the random sample queue   
+42 dequeue(Q, B) # pop-out earliest B instances   
+43   
+44 # SGD updation   
+45 loss.backward()   
+46 update(S.params)
+
+# A.1.1 DATA AUGMENTATIONS
+
+Both our teacher pre-training and distillation adopt the data augmentations as follows:
+
+Random Resized Crop: The image is randomly resized with a scale of {0.2, 1.0}, then cropped to the size of $2 2 4 \times 2 2 4$ .
+
+Random Color Jittering: with brightness to be {0.4, 0.4, 0.4, 0.1} with probability at 0.8.
+
+Random Gray Scale transformation: with probability at 0.2.
+
+Random Gaussian Blur transformation: with $\sigma = \{ 0 . 1 , 0 . 2 \}$ and probability at 0.5.
+
+Horizontal Flip: Horizontal flip is applied with probability at 0.5
+
+# A.1.2 PRE-TRAINING AND DISTILLATION ON MOBILENET AND EFFICIENTNET
+
+MobileNet (Howard et al., 2017) and EfficientNet (Tan & Le, 2019) have been considered as the smaller counterparts with larger models, i.e., ResNet-50 (with supervised training, EfficientNetB0 hits $7 7 . 2 \%$ Top-1 Acc., and MobileNet-V3-large reaches $7 2 . 2 \%$ on ImageNet testing split). Nevertheless, un-matched performances are observed in the task of self-supervised contrastive pretraining: i.e., Self-Supervised Learning (MoCo-V2) on MobileNet-V3 only yields $3 6 . 3 \%$ Top-1 Acc. on ImageNet. We conjecture that several reasons might lead to this dilemma:
+
+1. The inability of models with less parameters for handling large volume of categories and data, which exists also in other domains, i.e., face recognition (Guo et al., 2016; Zhang et al., 2017).   
+2. Less possibility for optimum parameters to be chosen when transferring to downstream tasks: models with more parameters after pre-training might produce a plenty cornucopia of optimum parameters for fine-tuning.
+
+To narrow the dramatic performance gap between smaller architectures using contrastive SSL with the larger, we explore with architectural manipulations and training hyper-parameters. In specific, we find that by adding a deeper projection head largely improves the representation quality, a.k.a., better performances on linear evaluation. We experiment with adding one additional linear projection head on the top of convolutional backbones.
+
+Similarly, we also expand the MLP projection head on EfficientNet-b0. Though recent work shows that fine-tuning from a middle layer of the projection head can produce a largely different result (Chen et al., 2020b), we consistently just use the representations from convolutional trunk without adding extra layers during the phase of linear evaluation. As shown in Table 6, pre-training with a deeper projection head dramatically helps the improvement on linear evaluations, adding $17 \%$ Top-1. Acc. for Mobile-v3-large, and we report the improved baselines in the main paper (see the first row in Table 1 of the main paper). We keep most of the hyper-parameters as the distillation on ResNet except reducing the weight-decay of them to 1e-5, following (Tan & Le, 2019; Sandler et al., 2018).
+
+Table 6: Linear evaluations on ImageNet of EfficientNet and MobileNet pre-trained using MoCo-v2. A deeper projection head largely boosts the linear evaluation performances on smaller architectures.   
+
+<table><tr><td>Model</td><td>Deeper MLPs</td><td>Top-1 Acc. Top-5 Acc.</td><td></td></tr><tr><td>EfficientNet-b0</td><td></td><td>39.1</td><td>64.6</td></tr><tr><td>EfficientNet-b0</td><td>×</td><td>42.2</td><td>68.5</td></tr><tr><td>Mobile-v3-large</td><td>X</td><td>19.0</td><td>41.3</td></tr><tr><td>Mobile-v3-large</td><td>√</td><td>36.3</td><td>62.2</td></tr></table>
+
+# A.2 ADDITIONAL DETAILS OF EVALUATIONS
+
+We list additional details regarding our evaluation experiments in this section.
+
+ImageNet-1k Semi-Supervised Linear Evaluation. Following Zhai et al. (2019); Chen et al. (2020a), we train the FC layers on the basis of our student encoder after distillation using a fraction of labeled ImageNet-1k dataset ( $1 \%$ and $10 \%$ ), and evaluate it on the whole test split. The fraction of labeled dataset is constructed in a class-balanced way, with roughly 12 and 128 images per class∗. We use SGD optimizer and set initial learning rate to be 30 with a multiplier $=$ BatchSize/256 without weight decaying for 100 epochs. We use the step-wise scheduler for the learning rate updating with 5 warm-up epochs, and the learning rate is reduced by 10 at 60 and 80 epochs. On smaller architectures like EfficientNet and MobileNet, we reduce the initial learning rate to 3. During training, the image is center-cropped to the size of $2 2 4 \times 2 2 4$ with just Random Horizontal Flip as the data augmentation. For testing, we first resize the image to $2 5 6 \times 2 5 6$ and use the center cropped $2 2 4 \times 2 2 4$ for pre-processing. In Table 8, we show the distillation results on a larger encoder (ResNet-550) when using different teacher networks.
+
+Table 7: Before and after distillation Top-1/5 test accuracy $( \% )$ on ImageNet of EfficientNet-b0 and MobileNetlarge without deeper MLPs.   
+
+<table><tr><td>Student</td><td>Teacher</td><td>Top-1</td><td>Top-5</td></tr><tr><td rowspan="3">EfficientNet-b0</td><td>X</td><td>39.1</td><td>64.6</td></tr><tr><td>ResNet-50</td><td>59.2</td><td>81.2</td></tr><tr><td>ResNet-101</td><td>62.8</td><td>84.7</td></tr><tr><td></td><td>ResNet-152</td><td>63.3</td><td>85.6</td></tr><tr><td rowspan="3">MobileNet-v3</td><td>X ResNet-50</td><td>19.0 50.9</td><td>41.3 77.7</td></tr><tr><td>ResNet-101</td><td>57.6</td><td>82.6</td></tr><tr><td>ResNet-152</td><td>58.3</td><td>82.9</td></tr></table>
+
+Table 8: ImageNet-1k test accuracy $( \% )$ under KNN and linear classification on ResNet-50 encoder with deeper, MoCo-V2/SWAV pre-trained teacher architectures. $\pmb { \chi }$ denotes MoCo-V2 self-supervised learning baselines before distillation. \* indicates using a stronger teacher encoder pre-trained by SWAV with additional small-patches during distillation.   
+
+<table><tr><td></td><td>Stud.</td><td></td><td>ResNet-50</td><td></td></tr><tr><td>Teac.</td><td>Epoch</td><td>KNN</td><td>Top-1</td><td>Top-5</td></tr><tr><td>X ResNet-50</td><td>200</td><td>46.1 46.1</td><td>67.4 67.5</td><td>87.8 87.8</td></tr><tr><td>△ ResNet-101</td><td>200</td><td>+0.0 52.3</td><td>+0.1</td><td>+0.0</td></tr><tr><td>△</td><td>200</td><td>+6.2</td><td>69.1 +1.7</td><td>88.7 +0.9</td></tr><tr><td>ResNet-152 △</td><td>200</td><td>53.2 +7.1</td><td>70.4 +3.0</td><td>90.5 +2.7</td></tr><tr><td>ResNet-50×2* △</td><td>800</td><td>59.0 +12.9</td><td>74.3 +6.9</td><td>92.2 +4.4</td></tr></table>
+
+Transfer Learning. We test the transferability of the representations learned from self-supervised distillation by conducting the linear evaluations using offline features on several other datasets. Specifically, a single layer logistic classifier is trained following (Chen et al., 2020a; Grill et al., 2020) using SGD optimizer without weight decay and momentum parameter at 0.9. We use CIFAR-10, CIFAR-100 (Krizhevsky et al., 2009) and SUN-397 (Xiao et al., 2010) as our testing beds.
+
+CIFAR: As the size for CIFAR dataset is $3 2 \times 3 2$ , we resize all images to $2 2 4 \times 2 2 4$ pixels along the shorter side using bicubic resampling method, followed by a center crop operation. We set the learning rate at 1e-3 constantly and train it for 120 epochs. The hyper-parameters are searched using 10 fold cross-validation on the train split and report its final top-1 accuracy on the test split.
+
+Table 9: Object detection and instance segmentation fine-tuned on VOC07: bounding-box AP $( \mathsf { A P } ^ { b b } )$ ) and mask AP $( \mathbf { A } \mathbf { P } ^ { m k } )$ ) evaluated on VOC07-val. The first row shows the baseline from MoCo-v2 backbones without distillation.   
+
+<table><tr><td rowspan="2">Student</td><td rowspan="2">Teacher</td><td colspan="3">VOC Object Detection</td></tr><tr><td>APbb</td><td>AP</td><td>AP</td></tr><tr><td rowspan="3">ResNet-34</td><td>X</td><td>53.6</td><td>79.1</td><td>58.7</td></tr><tr><td>ResNet-50</td><td>53.7 (+0.1)</td><td>79.4 (+0.3)</td><td>59.2 (+0.5)</td></tr><tr><td>ResNet-101 ResNet-152</td><td>54.1 (+0.5) 54.4 (+0.8)</td><td>79.8 (+0.7) 80.1 (+1.0)</td><td>59.1 (+0.4) ） 59.9 (+1.2)</td></tr><tr><td rowspan="2">ResNet-50</td><td>X</td><td>57.0</td><td>82.4</td><td>63.6</td></tr><tr><td>ResNet-50 ResNet-101</td><td>57.0 (+0.0) 57.1 (+0.1)</td><td>82.4 (+0.0) 82.8 (+0.4)</td><td>63.6 (+0.0) 63.8 (+0.2)</td></tr></table>
+
+Table 10: Object detection and instance segmentation fine-tuned on COCO: bounding-box AP $( \mathbf { A P } ^ { b b } )$ and mask AP $( \mathbf { A } \mathbf { P } ^ { m k } )$ evaluated on COCO-val2017. The first several rows show the baselines from unsupervised backbones without distillation.   
+
+<table><tr><td rowspan="2">Student</td><td rowspan="2">Teacher</td><td colspan="3">Object Detection</td><td colspan="3">Instance Segmentation</td></tr><tr><td>Apbb</td><td>AP</td><td>APP</td><td>Apmk</td><td>AP</td><td>AP</td></tr><tr><td rowspan="4">ResNet34</td><td>X</td><td>38.1</td><td>56.8</td><td>40.7</td><td>33.0</td><td>53.2</td><td>35.3</td></tr><tr><td>ResNet50</td><td>38.4 (+0.3)</td><td>57.0 (+0.2)</td><td>41.0 (+0.3)</td><td>33.3 (+0.3)</td><td>53.6 (+0.4)</td><td>35.4 (+0.1)</td></tr><tr><td>ResNet101</td><td>38.5 (+0.4)</td><td>57.3 (+0.5)</td><td>41.4 (+0.7)</td><td>33.6 (+0.6)</td><td>54.1 (+0.9)</td><td>35.6 (+0.3)</td></tr><tr><td>ResNet152</td><td>38.4 (+0.3)</td><td>57.0 (+0.2)</td><td>41.0 (+0.3)</td><td>33.3 (+0.3)</td><td>53.7 (+0.5)</td><td>35.3 (+0.0)</td></tr></table>
+
+SUN-397: We further extend our transferring evaluation to the scene dataset SUN-397 for a more diverse testing. The official dataset specifies 10 different train/test splits, with each contains 50 images per category covering 397 different scenes. We follow (Chen et al., 2020a; Grill et al., 2020) and use the first train/test split. For the validation set, we randomly pick 10 images (yielding $20 \%$ of the dataset), with identical optimizer parameters as CIFAR.
+
+Object Detection and Instance Segmentation. As indicated by (He et al., 2020), features produced by self-supervised pre-training have divergent distributions in downstream tasks, thus resulting the supervised pre-training picked hyper-parameters not applicable. To relieve this, He et al. (2020) uses feature normalization during the fine-tuning phase and train the BN layers. Different from previous transferring and linear evaluations where we exploit only offline features, model for detection and segmentation is trained with all parameters tuned. For this reason, annotations on COCO for segmentation gives much higher influence for the backbone model than the VOC dataset (see Table 9), and gives an offset to the pre-training difference (see Table 10). Thus, this makes the performance boosting by pre-training less obvious, and leads to trivial AP differences before and after distillation.
+
+Object Detection on PASCAL VOC-07: We train a C4 (He et al., 2017) based Faster R-CNN (Ren et al., 2015) as the detector with different ResNet architectures (ResNet-18, ResNet-34 and ResNet-50) for evaluating the transferability of features for object detection tasks. We use Detectron2 (Wu et al., 2019) for the implementations. We train our detector for 48k iterations with a batch size of 32 (8 images per GPU). The base learning rate is set to 0.01 with 200 warm-up iterations. We set the scale of images for training as [400, 800] and 800 at inference.
+
+Object Detection and Segmentation on COCO: We use Mask R-CNN (He et al., 2017) with the C4 backbone for the object detection and instance segmentation task on COCO dataset, with $2 \times$ schedule. Similar to the VOC detection, we tune the BN layers and all parameters. The model is trained for $1 8 0 \mathrm { k }$ iterations with initial learning rate set to 0.02. We set the scale of images for training as [600, 800] and 800 at inference.
+
+Table 11: Linear evaluations on ImageNet of ResNet-18 after distillation from the SWAV pre-trained ResNet-50 using either single view, cross-views, or small patch views.   
+
+<table><tr><td>Method</td><td>Multi-View(s)</td><td>Top-1 Acc.</td><td>Top-5 Acc.</td></tr><tr><td>Identical-View</td><td>1×224</td><td>61.7</td><td>84.2</td></tr><tr><td>Cross-Views</td><td>2×224</td><td>58.2</td><td>81.7</td></tr><tr><td>Multi-Crops + Cross-Views</td><td>1×224 + 6×96×96</td><td>61.9</td><td>84.4</td></tr><tr><td>Multi-Crops + Identical-View </td><td>1×224 +6×96×96</td><td>62.6</td><td>84.8</td></tr></table>
+
+![](images/0863ba5174cfabf1f7a4f95ce0b77df7bdbf71b58813b9c855df14a8e571182a.jpg)  
+Figure 6: We experiment with different strategies of using views during distillation, which include: (a). Identical view for distillation. (b). Cross view distillation. (c). Large-small cross view distillation. (d). Large-small identical view distillation.
+
+# A.3 SINGLE CROP V.S. MULTI-CROPS VIEW(S) FOR DISTILLATION
+
+In contrary with most contrastive SSL methods where two different augmented views of an image are utilized as the positive samples (see Figure 6-a), SEED uses an identical view for each image (see Figure 6-b) during distillation and yields better performances, as is shown in Table. 11. In addition, we have also experimented with two strategies of using small patches. To be specific, we follow the set-up in SWAV (Caron et al., 2020), that 6 small patches of the size $9 6 \times 9 6$ are sampled at the scale of (0.05, 0.14). Then, we apply the same augmentations as introduced previously as data pre-processing. Figure. 6-c shows the way that is similar in SWAV for small-patch learning, where both large and 6 small patches are fed into the student encoder, with the learning target $( \mathbf { z } ^ { T } )$ to be the embedding of large view from the teacher encoder. Figure. 6-d is the strategy we use during distillation, that both views are fed into student and teacher to produce the embeddings for small-views $( \mathbf { z } _ { s } ^ { S } , \mathbf { z } _ { s } ^ { T } )$ and large views $( \mathbf { z } _ { l } ^ { S } , \mathbf { z } _ { l } ^ { T } )$ . Based on that, the distillation is formulated separately on the small and large views. Notably, we maintain two independent queues for storing historical data samples for the large and small views.
+
+# A.4 STRATEGIES FOR OTHER DISTILLATION METHODS
+
+We compare the effect of distillation using different strategies with SEED.
+
+$\mathbf { \xi } _ { l 2 }$ -Distance: We train the student encoder by minimizing the squared $l 2$ -distance of representations from student $( \mathbf { z } _ { i } ^ { S } )$ and teacher $( \mathbf { z } _ { i } ^ { T } )$ for an identical view $\mathbf { x } _ { i }$ .
+
+$\pmb { K }$ -Means: We experiment with the $K$ -Means clustering method to retrieve pseudo class labels for distillation. Specifically, we first extract offline image features using the $S S L$ pre-trained Teacher network without any image augmentations. Based on this, we conduct our $K$ -Means clustering with $4 \mathrm { k }$ and $1 6 \mathrm { k }$ unique centroids. Then the final centroids are used to produce pseudo labels for unlabelled instances. With that, we carry out the distillation by training the model on a classification task using the produced labels as the ground-truth. To avoid trivial solutions that the majority of images are assigned to a few clusters, we sample images based on a uniform distribution over pseudo-labels as clustering proceeds. We observe very close results when adjusting numbers of centroids.
+
+Online-Clustering: With $K$ -Means for pseudo-label generation training, it does not lead to satisfying results $( 5 1 . 0 \%$ on ResNet-18 with ResNet-50 as Teacher) as instances might have not been accurately categorized by limited frozen centroids. Similar to (Caron et al., 2018; Li et al., 2020), we resort to the “in-batch” and dynamical clustering to substitute the frozen $K$ -Means method. We conduct
+
+$K$ -Means clustering within a batch and continuously update the centroid based on the teacher feature representation as distillation goes on. This alleviates the above problems and yields a substantial performance improvement on ResNet-18 to $5 6 . 4 \%$ .
+
+Binary Contrastive Loss: We resort to CRD (Tian et al., 2019a) and adopt an info-NCE loss-alike training objective in unsupervised distillation tasks. Specifically, we treat representation features from Teacher and Student for instance $\mathbf { x } _ { i }$ as positive pairs, and random instances from $\mathbf { D }$ as negative samples:
+
+$$
+\hat { \theta } _ { S } = \underset { \theta _ { S } } { \arg \operatorname* { m i n } } \sum _ { i } ^ { N } \log h ( \mathbf { z } _ { i } ^ { S } , \mathbf { z } _ { i } ^ { T } ) + K \cdot [ \log h ( 1 - h ( \mathbf { z } _ { i } ^ { S } , \mathbf { d } _ { j } ^ { T } ) ) ] ,
+$$
+
+where $\mathbf { d } _ { j } ^ { T } \in \mathbf { D } , h ( \cdot )$ is any family of functions that satisfy $h$ : $\{ { \bf z } , { \bf d } \}  [ 0 , 1 ]$ , e.g., cosine similarity.
+
+# A.5 DISCUSSIONS ON SEED
+
+Our proposed learning objective for SEED is composed of two goals, that is to align the encoding $\mathbf { z } ^ { S }$ by the student model with $\mathbf { z } ^ { T }$ produced by the teacher model; meanwhile, $\mathbf { z } ^ { S }$ also softly contrasts with random samples maintained in the $\mathbf { D }$ . This can be formulated more directly as minimizing the $l 2$ distance of ${ \mathbf z } ^ { T } , { \mathbf z } ^ { S }$ , together with the cross-entropy computed using $\mathbf { D }$ :
+
+$$
+\begin{array} { l } { \displaystyle \mathcal { L } = \frac { 1 } { N } \sum _ { i } ^ { N } \bigg \{ \lambda _ { a } \cdot \bigg \vert \bigg \vert \mathbf { z } _ { i } ^ { T } - \mathbf { z } _ { i } ^ { S } \bigg \vert \bigg \vert _ { 2 } - \lambda _ { b } \cdot \mathbf { p } ^ { T } ( \mathbf { x } _ { i } ; \boldsymbol { \theta } _ { T } , \mathbf { D } ) \cdot \log \mathbf { p } ^ { S } ( \mathbf { x } _ { i } ; \boldsymbol { \theta } _ { S } , \mathbf { D } ) \bigg \} } \\ { \displaystyle \quad = \sum _ { i } ^ { N } \bigg \{ - \lambda _ { a } \cdot \mathbf { z } _ { i } ^ { T } \cdot \mathbf { z } _ { i } ^ { S } - \lambda _ { b } \cdot \sum _ { j } ^ { K } \frac { \exp ( \mathbf { z } _ { i } ^ { T } \cdot \mathbf { d } _ { j } / \tau ^ { T } ) } { \sum _ { \mathbf { d } \setminus \mathbf { D } } \exp ( \mathbf { z } _ { i } ^ { T } \cdot \mathbf { d } / \tau ^ { T } ) } \cdot \log \frac { \exp ( \mathbf { z } _ { i } ^ { S } \cdot \mathbf { d } _ { j } / \tau ^ { S } ) } { \sum _ { \mathbf { d } \setminus \mathbf { D } } \exp ( \mathbf { z } _ { i } ^ { S } \cdot \mathbf { d } / \tau ^ { S } ) } \bigg \} , } \end{array}
+$$
+
+Directly optimizing Eq. 7 can lead to apparent difficulty in searching optimal hyper-parameters $( \lambda _ { a } , \lambda _ { b } , \bar { \tau } ^ { T }$ and $\tau ^ { S }$ ). Our proposed objective on $\mathbf { D } ^ { + }$ indeed is an approximated upper-bound of the above objectiveness however much simplified:
+
+$$
+\begin{array} { r l } & { \mathcal { L } _ { \mathrm { S E E D } } = \displaystyle \frac { 1 } { N } \sum _ { i } ^ { N } - \mathbf { p } ^ { T } ( \mathbf { x } _ { i } ; \boldsymbol { \theta } _ { T } , \mathbf { D } ^ { + } ) \cdot \log \mathbf { p } ^ { S } ( \mathbf { x } _ { i } ; \boldsymbol { \theta } _ { S } , \mathbf { D } ^ { + } ) } \\ & { \qquad = \displaystyle \sum _ { i } ^ { N } \sum _ { j } ^ { K + 1 } - \underbrace { \frac { \exp \left( \mathbf { z } _ { i } ^ { T } \cdot \mathbf { d } _ { j } / \tau ^ { T } \right) } { \sum _ { \mathbf { d } \sim \mathbf { D } ^ { + } } \exp \left( \mathbf { z } _ { i } ^ { T } \cdot \mathbf { d } / \tau ^ { T } \right) } } _ { \mathbf { w } _ { j } ^ { i } } \cdot \log \frac { \exp \left( \mathbf { z } _ { i } ^ { S } \cdot \mathbf { d } _ { j } / \tau ^ { S } \right) } { \sum _ { \mathbf { d } \sim \mathbf { D } ^ { + } } \exp \left( \mathbf { z } _ { i } ^ { S } \cdot \mathbf { d } / \tau ^ { S } \right) } , } \end{array}
+$$
+
+where we let $\mathbf { w } _ { j } ^ { i }$ denote the weighting term regulated under $\tau ^ { T }$ . Since the $( K + 1 )$ th element in $\mathbf { D } ^ { + }$ is our supplemented vector ${ \mathbf z } _ { i } ^ { T }$ , the above objective can be expanded into:
+
+$$
+\begin{array} { r l } & { \mathcal { L } _ { \mathrm { S E E D } } = \displaystyle \frac { 1 } { N } \sum _ { i } ^ { N } \Big \{ { \bf w } _ { K + 1 } ^ { i } \cdot \big ( - { \bf z } _ { i } ^ { S } \cdot { \bf z } _ { i } ^ { T } / \tau ^ { S } + \log \sum _ { \bf d \sim D ^ { + } } \exp ( { \bf z } _ { i } ^ { S } \cdot { \bf d } / \tau ^ { S } ) \big ) } \\ & { \qquad + \displaystyle \sum _ { j = 1 } ^ { K } { \bf w } _ { j } ^ { i } \cdot \big ( - { \bf z } _ { i } ^ { S } \cdot { \bf d } _ { j } / \tau ^ { S } + \log \sum _ { \bf d \sim D ^ { + } } \exp ( { \bf z } _ { i } ^ { S } \cdot { \bf d } / \tau ^ { S } ) \big ) \Big \} } \end{array}
+$$
+
+Note that the LSE term in the first line is strictly non-negative as the range of inner product for $\mathbf { z } ^ { S }$ and $\mathbf { d }$ lies between $\left[ - 1 , + 1 \right]$ :
+
+$$
+\mathrm { L S E } ( \mathbf { D } ^ { + } , \mathbf { z } _ { i } ^ { S } ) \geq \log \big ( M \cdot \exp ( - 1 / \tau ^ { S } ) \big ) = \log \big ( M \cdot \exp ( - 5 ) \big ) > 0 ,
+$$
+
+where $M$ denotes the cardinality of the maintained queue $\mathbf { D } ^ { + }$ and is set to 65,536 in our experiment with $\tau ^ { S } = 0 . 2$ constantly. Meanwhile, the LSE term in the second line satisfies the following inequality:
+
+$$
+\mathrm { L S E } ( \mathbf { D } ^ { + } , \mathbf { z } _ { i } ^ { S } ) \geq \mathrm { L S E } ( \mathbf { D } , \mathbf { z } _ { i } ^ { S } ) .
+$$
+
+Thus, this demonstrates that the objective for SEED as Eq. 8 is equivalent to minimizing a weakened upper-bound of e.q. 7:
+
+$$
+\begin{array} { l } { { \mathcal { L } _ { \mathrm { S E E D } } = \displaystyle \frac { 1 } { N } \sum _ { i } ^ { N } - { \bf p } ^ { T } ( { \bf x } _ { i } ; \boldsymbol { \theta } _ { T } , { \bf D } ^ { + } ) \cdot \log { \bf p } ^ { S } ( { \bf x } _ { i } ; \boldsymbol { \theta } _ { S } , { \bf D } ^ { + } ) } \ ~ } \\ { { \displaystyle \geq \frac { 1 } { N } \sum _ { i } ^ { N } \left\{ { \bf w } _ { K + 1 } ^ { i } \cdot ( - { \bf z } _ { i } ^ { S } \cdot { \bf z } _ { i } ^ { T } / \tau ^ { S } ) + \sum _ { j = 1 } ^ { K } { \bf w } _ { j } ^ { i } \cdot \left( - { \bf z } _ { i } ^ { S } \cdot { \bf d } _ { j } / \tau ^ { S } + \log \sum _ { \bf d \times D } ^ { C } \exp ( { \bf z } _ { i } ^ { S } \cdot { \bf d } / \tau ^ { S } ) \right) \right\} } \ ~ } \\ { { \displaystyle = \frac { 1 } { N } \sum _ { i } ^ { N } \Bigg \{ - \frac { { \bf w } _ { K + 1 } ^ { i } } { \tau ^ { S } } \cdot { \bf z } _ { i } ^ { S } \cdot { \bf z } _ { i } ^ { T } - { \bf p } ^ { T } ( { \bf x } _ { i } ; \boldsymbol { \theta } _ { T } , { \bf D } ) \cdot \log { \bf p } ^ { S } ( { \bf x } _ { i } ; \boldsymbol { \theta } _ { S } , { \bf D } ) \Bigg \} } \ ~ } \end{array}
+$$
+
+This proves that our $\mathcal { L } _ { \mathrm { S E E D } }$ directly relates to a more intuitive distillation formulation as Eq. 7 $( l 2 +$ cross entropy loss), and it implicitly contains the objective of aligning and contrasting. However, our training objective is much simplified. During practice, we find by regulating $\tau ^ { T }$ , both training losses produce equal results.
+
+# A.6 DISCUSSION ON THE RELATIONSHIP OF SEED WITH INFO-NCE
+
+The objective of distillation can be considered as a soft version of Info-NCE (Oord et al., 2018), with the only difference to be that SEED learns from the negative samples with probabilities instead of treating them all strictly as negative samples. To be more specific, following Info-NCE, the “hard” style contrastive distillation can be expressed as aligning with representations from the Teacher encoder and contrasting with all random instances:
+
+$$
+\hat { \theta } _ { S } = \underset { \theta _ { S } } { \arg \operatorname* { m i n } } \ : \mathcal { L } _ { N C E } = \underset { \theta _ { S } } { \arg \operatorname* { m i n } } \sum _ { i } ^ { N } - \log \frac { \exp ( \mathbf { z } _ { i } ^ { T } \cdot \mathbf { z } _ { i } ^ { S } / \tau ) } { \sum _ { \mathbf { d } \sim \mathbf { D } } \exp ( \mathbf { z } _ { i } ^ { S } \cdot \mathbf { d } / \tau ) }
+$$
+
+which can be further deduced with two sub-terms consisting of positive sample alignment and contrasting with negative instances:
+
+$$
+\mathcal { L } _ { N C E } = \sum _ { i } ^ { N } \Big \{ \underbrace { - \mathbf { z } _ { i } ^ { S } \cdot \mathbf { z } _ { i } ^ { T } / \tau } _ { a l i g n m e n t } + \underbrace { \log \sum _ { \mathbf { d } \sim \mathbf { D } } \exp ( \mathbf { z } _ { i } ^ { S } \cdot \mathbf { d } / \tau ) } _ { c o n t r a s t i n g } \Big \} .
+$$
+
+Similarly, the objective of SEED can be dissembled into the weighted form of alignment and contrasting terms:
+
+$$
+\begin{array} { r l } & { \mathcal { L } _ { \mathrm { S E E D } } = \displaystyle \frac { 1 } { N * M } \sum _ { i } ^ { N } \sum _ { j } ^ { K + 1 } - \frac { \exp ( { \bf z } _ { i } ^ { T } \cdot { \bf d } _ { j } / \tau ^ { T } ) } { \sum _ { \bf d } \sim \mathrm { D } ^ { + } } \cdot \log \Big \frac { \exp ( { \bf z } _ { i } ^ { S } \cdot { \bf d } _ { j } / \tau ^ { S } ) } { \sum _ { \bf d } \sim \mathrm { D } ^ { + } } } \\ & { \quad \quad \quad \quad = \frac { 1 } { N * M } \sum _ { i } ^ { N } \displaystyle \sum _ { j } ^ { K + 1 } \underbrace { \frac { \exp ( { \bf z } _ { i } ^ { T } \cdot { \bf d } _ { j } / \tau ^ { T } ) } { \sum _ { \bf d } \sim \mathrm { D } ^ { + } } } _ { { \bf w } _ { j } ^ { \bf d } } \cdot \underbrace { ( - { \bf z } _ { i } ^ { S } \cdot { \bf z } _ { i } ^ { T } / \tau ^ { S } ) } _ { a l i g n m e n t } \cdot \underbrace { ( \frac { \log \sum _ { i } ^ { S } \cdot { \bf d } _ { i } / \tau ^ { S } ) } { c \mathrm { o } n t r a s t i n g } } _ { c o n t r a s t i n g } ) , } \end{array}
+$$
+
+where the normalization term can be considered as soft labels, $\mathbf { W } ^ { i } = \left[ \mathbf { w } _ { 1 } ^ { i } \ldots \mathbf { w } _ { K + 1 } ^ { i } \right]$ , which can weight the above loss as:
+
+$$
+\mathcal { L } _ { \mathrm { S E E D } } = \frac { 1 } { N * M } \sum _ { i } ^ { N } \sum _ { j } ^ { K + 1 } \mathbf { w } _ { j } ^ { i } \cdot \Big \{ - \mathbf { z } _ { i } ^ { S } \cdot \mathbf { z } _ { i } ^ { T } / \tau ^ { S } + \log \sum _ { \mathbf { d } \sim \mathbf { D } } \exp ( \mathbf { z } _ { i } ^ { S } \cdot \mathbf { d } / \tau ^ { S } ) ) \Big \} ,
+$$
+
+When tuning hyper-parameter $\tau ^ { T }$ towards 0, $\mathbf { W } ^ { i }$ can be altered into the format of one-hot vector with $\mathbf { w } _ { K + 1 } ^ { i } = \dot { 1 }$ , which is then degraded to the case of contrastive distillation as in equation 14. In practice, the choice of an optimal $\tau ^ { T }$ can be dataset-specific. We show that the higher $\tau ^ { T }$ (with labels be more ‘soft’) can actually yield better results on other datasets, e.g., CIFAR-10 (Krizhevsky et al., 2009).
+
+# A.7 COMPATIBILITY WITH SUPERVISED DISTILLATION
+
+SEED conducts self-supervised distillation at the pre-training phase for the representation learning. However, we verify that SEED is compatible with traditional supervised distillation that happened during fine-tuning phrase at downstream, and can even produce better results. We begin with the SSL pre-training on a larger architecture (ResNet-152) using MoCo-V2 and train it for 200 epochs as the teacher network. As images in CIFAR-100 are in the size of $3 2 \times 3 2$ , we modify the first conv layer in ResNet with kernel size $= 3$ and stride $= 1$ .
+
+We then compare the Top-1 accuracy of a smaller ResNet-18 on CIFAR-100 when using different distillation strategies when all parameters are trainable. First, we use SEED to pre-train ResNet-18 with Res-152 as the teacher model, and then evaluate in on the test split of CIFAR-100 using linear fine-tuning task. As we keep all parameters trainable during the fine-tuning phase, distillation on the pre-training only yields a trivial boost: $7 5 . 4 \%$ v.s. $7 5 . 2 \%$ . Then, we adopt the traditional distillation method, e.g., (Hinton et al., 2015), to first fine-tune the ResNet-152 model, and then use its output class probability to facilitate the linear classification task on ResNet-18 in the fine-tuning phrase. This improves the linear classification accuracy on ResNet-18 to $7 6 . 0 \%$ . At the end, we initialize the ResNet-18 with our SEED pre-trained ResNet-18, and equip it with the supervised classification distillation during fine-tuning. With that, we find that the performance of ResNet-18 is further boosted to $7 8 . 1 \%$ . We can conclude that our SEED is compatible with traditional supervised distillation that mostly happened at downstream for specific tasks, e.g., classification, object detection.
+
+Table 12: CIFAR-100 Top-1 Accuracy $( \% )$ of ResNet-18 with (or without) distillation at different phase: selfsupervised pre-training stage, and supervised classification fine-tuning. All backbone parameters of ResNet-18 are trainable in experiments.   
+
+<table><tr><td colspan="3">Pre-training Distill. Fine-tuning Distill. Top-1 Acc</td></tr><tr><td>X</td><td>X</td><td>75.2</td></tr><tr><td>√</td><td>×</td><td>75.4</td></tr><tr><td>×</td><td>√</td><td>76.0</td></tr><tr><td>√</td><td>√</td><td>78.1</td></tr></table>
+
+![](images/d2dbbfad613f736ba1bbdacdf0d815fdcec0dd5cde93e27f656664f449c6e7cd.jpg)  
+Figure 7: Linear evaluation accuracy $( \% )$ of distillation between ResNet-18 (as the Student) and ResNet-50 (as the Teacher) using different size of queue when $\mathrm { L R } { = } 0 . 0 3$ and weight decay=1e-6. Note the axis is the log(·) value of queue lengths.
+
+# A.8 ADDITIONAL ABLATION STUDIES
+
+We study effects of different hyper-parameters to distillation using a ResNet-18 (as Student) and a SWAV pre-trained ResNet-50 (as Teacher) with small patch views. In specific, we list the Top-1 Acc. on validation split of ImageNet-1k using different lengths of queue $K { = } 1 2 8$ , 512, 1,024, 4,096, 8,192, 16,384, 65,536) in Figure. 7. With the increasing of random data samples, the distillation boosts the accuracy of learned representations, however within a limited range: $+ 1 . 5$ when the queue size is
+
+Table 13: Linear evaluation accuracy $( \% )$ of distillation between ResNet-18 (as the Student) and ResNet-50 (as the Teacher) using different learning rates when the queue size is 65,536 and weight decay=1e-6.   
+
+<table><tr><td>LR</td><td>Top-1 Acc. Top-5 Acc.</td><td></td></tr><tr><td>1</td><td>58.9</td><td>83.1</td></tr><tr><td>0.1</td><td>62.9</td><td>85.3</td></tr><tr><td>0.03</td><td>63.3</td><td>85.4</td></tr><tr><td>0.01</td><td>62.6</td><td>85.0</td></tr></table>
+
+Table 14: Linear evaluation accuracy $( \% )$ of distillation between ResNet-18 (as the Student) and ResNet-50 (as the Teacher) using different weight decays when the queue size is 65,536 and $\mathrm { L R } { = } 0 . 0 3$ .   
+
+<table><tr><td>WD</td><td>Top-1 Acc. Top-5 Acc.</td><td></td></tr><tr><td>1e-2</td><td>11.8</td><td>27.7</td></tr><tr><td>1e-3</td><td>62.3</td><td>84.7</td></tr><tr><td>1e-4</td><td>61.9</td><td>84.4</td></tr><tr><td>1e-5</td><td>61.6</td><td>84.2</td></tr><tr><td>1e-6</td><td>63.3</td><td>85.4</td></tr></table>
+
+65,536 compared with 256. Furthermore, Table. 13 and 14 summarize the linear evaluation accuracy under different learning rates and weight decays.
\ No newline at end of file
diff --git a/parse/train/AHm3dbp7D1D/AHm3dbp7D1D_content_list.json b/parse/train/AHm3dbp7D1D/AHm3dbp7D1D_content_list.json
new file mode 100644
index 0000000000000000000000000000000000000000..3c6cb6368231a427f65076ece6e9c074daca6ce4
--- /dev/null
+++ b/parse/train/AHm3dbp7D1D/AHm3dbp7D1D_content_list.json
@@ -0,0 +1,2498 @@
+[
+    {
+        "type": "text",
+        "text": "SEED: SELF-SUPERVISED DISTILLATION FOR VISUAL REPRESENTATION ",
+        "text_level": 1,
+        "bbox": [
+            176,
+            99,
+            732,
+            148
+        ],
+        "page_idx": 0
+    },
+    {
+        "type": "text",
+        "text": "Zhiyuan Fang† , Jianfeng Wang‡, Lijuan Wang‡, Lei Zhang‡, Yezhou Yang†, Zicheng Liu‡ ",
+        "bbox": [
+            186,
+            171,
+            813,
+            188
+        ],
+        "page_idx": 0
+    },
+    {
+        "type": "text",
+        "text": "†Arizona State University, ‡Microsoft Corporation {zy.fang, yz.yang}@asu.edu {jianfw, lijuanw, leizhang, zliu}@microsoft.com ",
+        "bbox": [
+            258,
+            189,
+            718,
+            229
+        ],
+        "page_idx": 0
+    },
+    {
+        "type": "text",
+        "text": "ABSTRACT ",
+        "text_level": 1,
+        "bbox": [
+            454,
+            273,
+            544,
+            287
+        ],
+        "page_idx": 0
+    },
+    {
+        "type": "text",
+        "text": "This paper is concerned with self-supervised learning for small models. The problem is motivated by our empirical studies that while the widely used contrastive self-supervised learning method has shown great progress on large model training, it does not work well for small models. To address this problem, we propose a new learning paradigm, named SElf-SupErvised Distillation (SEED), where we leverage a larger network (as Teacher) to transfer its representational knowledge into a smaller architecture (as Student) in a self-supervised fashion. Instead of directly learning from unlabeled data, we train a student encoder to mimic the similarity score distribution inferred by a teacher over a set of instances. We show that SEED dramatically boosts the performance of small networks on downstream tasks. Compared with self-supervised baselines, SEED improves the top-1 accuracy from $4 2 . 2 \\%$ to $6 7 . 6 \\%$ on EfficientNet-B0 and from $3 6 . 3 \\%$ to $6 8 . 2 \\%$ on MobileNetV3-Large on the ImageNet-1k dataset. ",
+        "bbox": [
+            233,
+            304,
+            766,
+            484
+        ],
+        "page_idx": 0
+    },
+    {
+        "type": "text",
+        "text": "1 INTRODUCTION ",
+        "text_level": 1,
+        "bbox": [
+            176,
+            512,
+            334,
+            526
+        ],
+        "page_idx": 0
+    },
+    {
+        "type": "text",
+        "text": "The burgeoning studies and success on self-supervised learning (SSL) for visual representation are mainly marked by its extraordinary potency of learning from unlabeled data at scale. Accompanying with the SSL is its phenomenal benefit of obtaining task-agnostic representations while allowing the training to dispense with prohibitively expensive data labeling. Major ramifications of visual SSL include pretext tasks (Noroozi & Favaro, 2016; Zhang et al., 2016; Gidaris et al., 2018; Zhang et al., 2019; Feng et al., 2019), contrastive representation learning (Wu et al., 2018; He et al., 2020; Chen et al., 2020a), online/offline clustering (Yang et al., 2016; Caron et al., 2018; Li et al., 2020; Caron et al., 2020; Grill et al., 2020), etc. Among them, several recent works (He et al., 2020; Chen et al., 2020a; Caron et al., 2020) have achieved comparable or even better accuracy than the supervised pre-training when transferring to downstream tasks, e.g. semi-supervised classification, object detection. ",
+        "bbox": [
+            174,
+            541,
+            524,
+            819
+        ],
+        "page_idx": 0
+    },
+    {
+        "type": "text",
+        "text": "The aforementioned top-performing SSL algorithms all involve large networks (e.g., ResNet-50 (He et al., 2016) or larger), with, however, little attention on small networks. Empirically, we find that existing techniques like contrastive learning do not work well on small networks. For instance, the linear probe top-1 accuracy on ImageNet using MoCo-V2 (Chen et al., 2020c) is only $3 6 . 3 \\%$ with MobileNetV3-Large (see Figure 1), which is much lower compared with its supervised training accuracy ",
+        "bbox": [
+            173,
+            827,
+            524,
+            882
+        ],
+        "page_idx": 0
+    },
+    {
+        "type": "image",
+        "img_path": "images/d4fb00f69f184e9fccdf211a9f298a6ff3e7571f5dfb94ff7b3d5393734ffc54.jpg",
+        "image_caption": [
+            "Figure 1: SEED vs. MoCo-V2 (Chen et al., 2020c)) on ImageNet-1K linear probe accuracy. The vertical axis is the top-1 accuracy and the horizontal axis is the number of learnable parameters for different network architectures. Directly applying self-supervised contrastive learning (MoCo-V2) does not work well for smaller architectures, while our method (SEED) leads to dramatic performance boost. Details of the setting can be found in Section 4. "
+        ],
+        "image_footnote": [],
+        "bbox": [
+            539,
+            549,
+            820,
+            712
+        ],
+        "page_idx": 0
+    },
+    {
+        "type": "text",
+        "text": "",
+        "bbox": [
+            174,
+            882,
+            825,
+            924
+        ],
+        "page_idx": 0
+    },
+    {
+        "type": "text",
+        "text": "$7 5 . 2 \\%$ (Howard et al., 2019). For EfficientNet-B0, the accuracy is $4 2 . 2 \\%$ compared with its supervised training accuracy $7 7 . 1 \\%$ (Tan & Le, 2019). We conjecture that this is because smaller models with fewer parameters cannot effectively learn instance level discriminative representation with large amount of data. ",
+        "bbox": [
+            174,
+            103,
+            823,
+            159
+        ],
+        "page_idx": 1
+    },
+    {
+        "type": "text",
+        "text": "To address this challenge, we inject knowledge distillation (KD) (Bucilua et al. ˇ , 2006; Hinton et al., 2015) into self-supervised learning and propose self-supervised distillation (dubbed as SEED) as a new learning paradigm. That is, train the larger, and distill to the smaller both in self-supervised manner. Instead of directly conducting self-supervised training on a smaller model, SEED first trains a large model (as the teacher) in a self-supervised way, and then distills the knowledge to the smaller model (as the student). Note that the conventional distillation is for supervised learning, while the distillation here is in the self-supervised setting without any labeled data. Supervised distillation can be formulated as training a student to mimic the probability mass function over classes predicted by a teacher model. In unsupervised knowledge distillation setting, however, the distribution over classes is not directly attainable. Therefore, we propose a simple yet effective self-supervised distillation method. Similar to (He et al., 2020; Wu et al., 2018), we maintain a queue of data samples. Given an instance, we first use the teacher network to obtain its similarity scores with all the data samples in the queue as well as the instance itself. Then the student encoder is trained to mimic the similarity score distribution inferred by the teacher over these data samples. ",
+        "bbox": [
+            174,
+            166,
+            825,
+            361
+        ],
+        "page_idx": 1
+    },
+    {
+        "type": "text",
+        "text": "The simplicity and flexibility that SEED brings are self-evident. 1) It does not require any clustering/prototypical computing procedure to retrieve the pseudo-labels or latent classes. 2) The teacher model can be pre-trained with any advanced SSL approach, e.g., MoCo-V2 (Chen et al., 2020c), SimCLR (Chen et al., 2020a), SWAV (Caron et al., 2020). 3) The knowledge can be distilled to any target small networks (either shallower, thinner, or totally different architectures). ",
+        "bbox": [
+            174,
+            367,
+            825,
+            438
+        ],
+        "page_idx": 1
+    },
+    {
+        "type": "text",
+        "text": "To demonstrate the effectiveness, we comprehensively evaluate the learned representations on series of downstream tasks, e.g., fully/semi-supervised classification, object detection, and also assess the transferability to other domains. For example, on ImageNet-1k dataset, SEED improves the linear probe accuracy of EfficientNet-B0 from $4 2 . 2 \\%$ to $6 7 . 6 \\%$ (a gain over $2 5 \\%$ ), and MobileNet-V3 from $3 6 . 3 \\%$ to $6 8 . 2 \\%$ (a gain over $3 1 \\%$ ) compared to MoCo-V2 baselines, as shown in Figure 1 and Section 4. ",
+        "bbox": [
+            174,
+            444,
+            825,
+            527
+        ],
+        "page_idx": 1
+    },
+    {
+        "type": "text",
+        "text": "Our contributions can be summarized as follows: ",
+        "bbox": [
+            173,
+            535,
+            495,
+            549
+        ],
+        "page_idx": 1
+    },
+    {
+        "type": "text",
+        "text": "• We are the first to address the problem of self-supervised visual representation learning for small models. We propose a self-supervised distillation (SEED) technique to transfer knowledge from a large model to a small model without any labeled data. With the proposed distillation technique (SEED), we significantly improve the state-of-theart SSL performance on small models.   \n• We exhaustively compare a variety of distillation strategies to show the validity of SEED under multiple settings. ",
+        "bbox": [
+            215,
+            561,
+            825,
+            691
+        ],
+        "page_idx": 1
+    },
+    {
+        "type": "text",
+        "text": "2 RELATED WORK ",
+        "text_level": 1,
+        "bbox": [
+            176,
+            710,
+            344,
+            727
+        ],
+        "page_idx": 1
+    },
+    {
+        "type": "text",
+        "text": "Among the recent literature in self-supervised learning, contrastive based approaches show prominent results on downstream tasks. Majority of the techniques along this direction are stemming from noise-contrastive estimation (Gutmann & Hyvärinen, 2010) where the latent distribution is estimated by contrasting with randomly or artificially generated noises. Oord et al. (2018) first proposed Info-NCE to learn image representations by predicting the future using an auto-regressive model for unsupervised learning. Follow-up works include improving the efficiency (Hénaff et al., 2019), and using multi-view as positive samples (Tian et al., 2019b). As these approaches can only have the access to limited negative instances, Wu et al. (2018) designed a memory-bank to store the previously seen random representations as negative samples, and treat each of them as independent categories (instance discrimination). However, this approach also comes with a deficiency that the previously stored vectors are inconsistent with the recently computed representations during the earlier stage of pre-training. Chen et al. (2020a) mitigate this issue by sampling negative samples from a large batch. Concurrently, He et al. (2020) improve the memory-bank based method and propose to use the momentum updated encoder for the remission of representation inconsistency. Other techniques include Misra & Maaten (2020) that combines the pretext-invariant objective loss with contrastive learning, and Wang & Isola (2020) that decomposes contrastive loss into alignment and uniformity objectiveness. ",
+        "bbox": [
+            174,
+            743,
+            825,
+            924
+        ],
+        "page_idx": 1
+    },
+    {
+        "type": "text",
+        "text": "",
+        "bbox": [
+            174,
+            103,
+            823,
+            159
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "text",
+        "text": "Knowledge distillation (Hinton et al., 2015) aims to transfer knowledge from a cumbersome model to a smaller one without losing too much generalization power, which is also well investigated in model compression (Bucilua et al. ˇ , 2006). Instead of mimicking the teacher’s output logit, attention transfer (Zagoruyko & Komodakis, 2016) formulates knowledge distillation on attention maps. Similarly, works in (Ahn et al., 2019; Yim et al., 2017; Koratana et al., 2019; Huang & Wang, 2017) have utilized different learning objectives including consistency on feature maps, consistency on probability mass function, and maximizing the mutual information. CRD (Tian et al., 2019a), which is derived from CMC (Tian et al., 2019b), optimizes the student network by a similar objective to Oord et al. (2018) using a derived lower bound on mutual information. However, the aforementioned efforts all focus on task-specific distillation (e.g., image classification) during the fine-tuning phase rather than a task-agnostic distillation in the pre-training phase for the representation learning. Several works on natural language pre-training proposed to leverage knowledge distillation for a smaller yet stronger small models. For instances, DistillBert (Sanh et al., 2019), TinyBert (Jiao et al., 2019), and MobileBert (Sun et al., 2020), have used knowledge distillation for model compression and shown their validity on multiple downstream tasks. Similar works also emphasize the value of smaller and faster models for language representation learning by leveraging knowledge distillation (Turc et al., 2019; Sun et al., 2019). These works all demonstrate the effectiveness of knowledge distillation for language representation learning in small models, while are not extended to the pre-training for visual representations. Notably, a recent concurrent work CompRess (Abbasi Koohpayegani et al., 2020) also point out the importance to develop better SSL method for smaller models. SEED closely relates to the above techniques but aims to facilitate visual representation learning during pre-training phase using distillation technique for small models, which as far as we know has not yet been investigated. ",
+        "bbox": [
+            173,
+            165,
+            825,
+            484
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "text",
+        "text": "3 METHOD ",
+        "text_level": 1,
+        "bbox": [
+            174,
+            506,
+            282,
+            521
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "text",
+        "text": "3.1 PRELIMINARY ON KNOWLEDGE DISTILLATION ",
+        "text_level": 1,
+        "bbox": [
+            176,
+            536,
+            539,
+            551
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "text",
+        "text": "Knowledge distillation (Hinton et al., 2015; Bucilua et al.ˇ , 2006) is an effective technique to transfer knowledge from a strong teacher network to a target student network. The training task can be generalized as the following formulation: ",
+        "bbox": [
+            176,
+            563,
+            825,
+            604
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "equation",
+        "img_path": "images/f6d588e03d3db3f013e786c6fff9679cd8aa0b9b81a92bf569f2acbe7a654360.jpg",
+        "text": "$$\n\\hat { \\theta } _ { S } = \\underset { \\theta _ { S } } { \\arg \\operatorname* { m i n } } \\sum _ { i } ^ { N } \\mathcal { L } _ { \\mathrm { s u p } } ( \\mathbf { x } _ { i } , \\theta _ { S } , y _ { i } ) + \\mathcal { L } _ { \\mathrm { d i s t i l l } } ( \\mathbf { x } _ { i } , \\theta _ { S } , \\theta _ { T } ) ,\n$$",
+        "text_format": "latex",
+        "bbox": [
+            315,
+            609,
+            679,
+            654
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "text",
+        "text": "where $\\mathbf { x } _ { i }$ is an image, $y _ { i }$ is the corresponding annotation, $\\theta _ { S }$ is the parameter set for the student network, and $\\theta _ { T }$ is the set for the teacher network. The loss $\\mathcal { L } _ { \\mathrm { s u p } }$ is the alignment error between the network prediction and the annotation. For example in image classification task (Mishra & Marr, 2017; Shen & Savvides, 2020; Polino et al., 2018; Cho & Hariharan, 2019), it is normally a cross entropy loss. For object detection (Liu et al., 2019; Chen et al., 2017), it includes bounding box regression as well. The loss of ${ \\mathcal { L } } _ { \\mathrm { d i s t i l l } }$ is the mimic error of the student network towards a pre-trained teacher network. For example in (Hinton et al., 2015), the teacher signal comes from the softmax prediction of multiple large-scale networks and the loss is measured by the Kullback–Leibler divergence. In Romero et al. (2014), the task is to align the intermediate feature map values and to minimize the squared $l 2$ distance. The effectiveness has been well demonstrated in the supervised setting with labeled data, but remains unknown for the unsupervised setting, which is our focus. ",
+        "bbox": [
+            174,
+            657,
+            825,
+            811
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "text",
+        "text": "3.2 SELF-SUPERVISED DISTILLATION FOR VISUAL REPRESENTATION ",
+        "text_level": 1,
+        "bbox": [
+            176,
+            827,
+            666,
+            843
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "text",
+        "text": "Different from supervised distillation, SEED aims to transfer knowledge from a large model to a small model without requiring labeled data, so that the learned representations in small model can be used for downstream tasks. Inspired by contrastive SSL, we formulate a simple approach for the distillation on the basis of instance similarity distribution over a contrastive instance queue. Similar to He et al. (2020), we maintain an instance queue for storing data samples’ encoding output from the teacher. Given a new sample, we compute its similarity scores with all the samples in the queue using both the teacher and the student models. We require that the similarity score distribution computed by the student matches with that computed by the teacher, which is formulated as minimizing the cross entropy between the student and the teacher’s similarity score distributions (as illustrated in Figure 2). ",
+        "bbox": [
+            174,
+            853,
+            823,
+            924
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "image",
+        "img_path": "images/2f2a03a1e9389d9452c8beee5f2f9f477f5576361541704a7434d89396409316.jpg",
+        "image_caption": [
+            "Figure 2: Illustration of our self-supervised distillation pipeline. The teacher encoder is pre-trained by SSL and kept frozen during the distillation. The student encoder is trained by minimizing the cross entropy of probabilities from teacher & student for an augmented view of an image, computed over a dynamically maintained queue. "
+        ],
+        "image_footnote": [],
+        "bbox": [
+            181,
+            102,
+            825,
+            228
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "",
+        "bbox": [
+            173,
+            301,
+            825,
+            372
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "Specifically, for a randomly augmented view $\\mathbf { x } _ { i }$ of an image, it is first mapped and normalized into feature vector representations $\\mathbf { z } _ { i } ^ { T } = f _ { \\theta } ^ { T } ( \\mathbf { x } _ { i } ) / \\vert \\vert f _ { \\theta } ^ { T } ( \\mathbf { x } _ { i } ) \\vert \\vert _ { 2 }$ , and $\\mathbf { z } _ { i } ^ { S } = f _ { \\theta } ^ { S } ( \\mathbf { \\bar { x } } _ { i } ^ { \\star } ) / \\vert \\vert f _ { \\theta } ^ { S } ( \\mathbf { x } _ { i } ) \\vert \\vert _ { 2 }$ , where $\\mathbf { z } _ { i } ^ { T } , \\mathbf { z } _ { i } ^ { S } \\in \\mathbb { R } ^ { D }$ , and $f _ { \\theta } ^ { T }$ and $f _ { \\theta } ^ { S }$ θ θ θ θdenote the teacher and student encoders, respectively. Let $\\mathbf { D } =$ $[ \\mathbf { d } _ { 1 } . . . \\mathbf { d } _ { K } ]$ denote the instance queue where $K$ is the queue length and ${ \\bf d } _ { j }$ is the feature vector obtained from the teacher encoder. Similar to the contrastive learning framework, $\\mathbf { D }$ is progressively updated under the “first-in first-out” strategy as distillation proceeds. That is, we en-queue the visual features of the current batch inferred by the teacher and de-queue the earliest seen samples at the end of iteration. Note that the maintained samples in queue $\\mathbf { D }$ are mostly random and irrelevant to the target instance $\\mathbf { x } _ { i }$ . Minimizing the cross entropy between the similarity score distribution computed by the student and teacher based on $\\mathbf { D }$ softly contrasts $\\mathbf { x } _ { i }$ with randomly selected samples, without directly aligning with the teacher encoder. To address this problem, we add the teacher’s embedding $( \\mathbf { z } _ { i } ^ { T } )$ into the queue and form $\\mathbf { D } ^ { + } = [ \\mathbf { d } _ { 1 } . . . \\mathbf { d } _ { K } , \\mathbf { d } _ { K + 1 } ]$ with $\\mathbf { d } _ { K + 1 } = \\mathbf { z } _ { i } ^ { T }$ . ",
+        "bbox": [
+            173,
+            378,
+            825,
+            549
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "Let $\\mathbf { p } ^ { T } ( \\mathbf { x } _ { i } ; \\boldsymbol { \\theta } _ { T } ; \\mathbf { D } ^ { + } )$ denote the similarity score between the extracted teacher feature $\\mathbf { z } _ { i } ^ { T }$ and ${ \\bf d } _ { j }$ ’s $( j = 1 , . . . , K + 1 )$ computed by the teacher model. $\\mathbf { p } ^ { T } ( \\mathbf { x } _ { i } ; \\boldsymbol { \\theta } _ { T } ; \\mathbf { D } ^ { + } )$ is defined as ",
+        "bbox": [
+            173,
+            555,
+            823,
+            587
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "equation",
+        "img_path": "images/fa03ea71bb692eacf77006e376709e981c9236cbb7b4dc642d83915ba5eb6316.jpg",
+        "text": "$$\n\\mathbf { p } ^ { T } ( \\mathbf { x } _ { i } ; \\theta _ { T } , \\mathbf { D } ^ { + } ) = \\left[ p _ { 1 } ^ { T } \\dots p _ { K + 1 } ^ { T } \\right] , \\qquad p _ { j } ^ { T } = \\frac { \\exp ( \\mathbf { z } _ { i } ^ { T } \\cdot \\mathbf { d } _ { j } / \\tau ^ { T } ) } { \\sum _ { \\mathbf { d } \\sim \\mathbf { D } ^ { + } } \\exp ( \\mathbf { z } _ { i } ^ { T } \\cdot \\mathbf { d } / \\tau ^ { T } ) } ,\n$$",
+        "text_format": "latex",
+        "bbox": [
+            258,
+            590,
+            738,
+            626
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "and $\\tau ^ { T }$ is a temperature parameter for the teacher. Note, we use $( ) ^ { T }$ to represent the feature from the teacher network and use $( \\cdot )$ to represent the inner product between two features. ",
+        "bbox": [
+            173,
+            630,
+            823,
+            660
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "Similarly let $\\mathbf { p } ^ { S } ( x _ { i } ; \\theta _ { S } , \\mathbf { D } ^ { + } )$ denote the similarity score computed by the student model, which is defined as ",
+        "bbox": [
+            173,
+            665,
+            823,
+            694
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "equation",
+        "img_path": "images/7a86dd461ab60bef53314bfb98d8072c5f8f7ea6f386125f046180e8c76b6966.jpg",
+        "text": "$$\n\\mathbf { p } ^ { S } ( \\mathbf { x } _ { i } ; \\theta _ { S } , \\mathbf { D } ^ { + } ) = \\left[ p _ { 1 } ^ { S } \\dots p _ { K + 1 } ^ { S } \\right] , \\qquad { \\mathrm { w h e r e ~ } } p _ { j } ^ { S } = { \\frac { \\exp ( \\mathbf { z } _ { i } ^ { S } \\cdot \\mathbf { d } _ { j } / \\tau ^ { S } ) } { \\sum _ { \\mathbf { d } \\sim \\mathbf { D } ^ { + } } \\exp ( \\mathbf { z } _ { i } ^ { S } \\cdot \\mathbf { d } / \\tau ^ { S } ) } } ,\n$$",
+        "text_format": "latex",
+        "bbox": [
+            238,
+            695,
+            758,
+            732
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "and $\\tau ^ { S }$ is a temperature parameter for the student. ",
+        "bbox": [
+            174,
+            736,
+            501,
+            752
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "Our self-supervised distillation can be formulated as minimizing the cross entropy between the similarity scores of the teacher, $\\mathbf { p } ^ { T } ( \\mathbf { x } _ { i } ; \\theta _ { T } , \\mathbf { D } ^ { + } )$ , and the student, $\\mathbf { p } ^ { S } ( \\mathbf { x } _ { i } ; \\boldsymbol { \\theta } _ { S } , \\mathbf { D } ^ { \\dagger } )$ , over all the instances $\\mathbf { x } _ { i }$ , that is, ",
+        "bbox": [
+            174,
+            757,
+            823,
+            800
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "equation",
+        "img_path": "images/ebb8b97ca7f12d289494242cd41ec751bf46c453936e79d85361fe539a0cb9bd.jpg",
+        "text": "$$\n\\begin{array} { r l } & { \\boldsymbol { \\hat { \\theta } _ { S } } = \\underset { \\boldsymbol { \\theta _ { S } } } { \\arg \\operatorname* { m i n } } \\sum _ { i } ^ { N } - \\mathbf { p } ^ { T } ( \\mathbf { x } _ { i } ; \\boldsymbol { \\theta _ { T } } , \\mathbf { D } ^ { + } ) \\cdot \\log \\mathbf { p } ^ { S } ( \\mathbf { x } _ { i } ; \\boldsymbol { \\theta _ { S } } , \\mathbf { D } ^ { + } ) } \\\\ & { \\quad = \\underset { \\boldsymbol { \\theta _ { S } } } { \\arg \\operatorname* { m i n } } \\sum _ { i } ^ { N } \\sum _ { j } ^ { K + 1 } - \\frac { \\exp \\left( \\mathbf { z } _ { i } ^ { T } \\cdot \\mathbf { d } _ { j } / \\tau ^ { T } \\right) } { \\sum _ { \\mathbf { d } \\sim \\mathbf { D } ^ { + } } \\exp \\left( \\mathbf { z } _ { i } ^ { T } \\cdot \\mathbf { d } / \\tau ^ { T } \\right) } \\cdot \\log \\frac { \\exp \\left( \\mathbf { z } _ { i } ^ { S } \\cdot \\mathbf { d } _ { j } / \\tau ^ { S } \\right) } { \\sum _ { \\mathbf { d } \\sim \\mathbf { D } ^ { + } } \\exp \\left( \\mathbf { z } _ { i } ^ { S } \\cdot \\mathbf { d } / \\tau ^ { S } \\right) } . } \\end{array}\n$$",
+        "text_format": "latex",
+        "bbox": [
+            225,
+            803,
+            771,
+            893
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "Since the teacher network is pre-trained and frozen, the queued features are consistent during training w.r.t. the student network. The higher the value of $p _ { j } ^ { I ^ { \\prime } }$ is, the larger weight will be laid on $p _ { j } ^ { S }$ . Due to the l2 normalization, similarity score between ${ \\mathbf z } _ { i } ^ { T }$ and ${ \\bf d } _ { K + 1 }$ remains constant 1 before softmax normalization, which is the largest among $p _ { j } ^ { T }$ . Thus, the weight for $p _ { K + 1 } ^ { S }$ is the largest and can be adjusted solely by tuning the value of $\\tau ^ { T }$ . By minimizing the loss, the feature of $\\mathbf { z } _ { i } ^ { S }$ can be aligned with ${ \\mathbf z } _ { i } ^ { T }$ and meanwhile contrasts with other unrelated image features in $\\mathbf { D }$ . We further discuss the relation of these two goals with our learning objective in Appendix A.5. ",
+        "bbox": [
+            173,
+            895,
+            825,
+            925
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "",
+        "bbox": [
+            173,
+            102,
+            825,
+            179
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "text",
+        "text": "Relations with Info-NCE loss. When $\\tau ^ { T } \\to 0$ , the softmax function for $\\mathbf { p } ^ { T }$ smoothly approaches to a one-hot vector, where $p _ { K + 1 } ^ { T }$ equals 1 and all others 0. In this extreme case, the loss becomes ",
+        "bbox": [
+            173,
+            184,
+            826,
+            215
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "equation",
+        "img_path": "images/e089e43867b6fe00aefe91b5ac56af8b29248c22fa3a0c345faa7cdfe89a5b9b.jpg",
+        "text": "$$\n\\mathcal { L } _ { N C E } = \\sum _ { i } ^ { N } - \\log \\frac { \\exp ( \\mathbf { z } _ { i } ^ { T } \\cdot \\mathbf { z } _ { i } ^ { S } / \\tau ) } { \\sum _ { \\mathbf { d } \\sim \\mathbf { D } ^ { + } } \\exp ( \\mathbf { z } _ { i } ^ { S } \\cdot \\mathbf { d } / \\tau ) } ,\n$$",
+        "text_format": "latex",
+        "bbox": [
+            354,
+            223,
+            642,
+            267
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "text",
+        "text": "which is similar to the widely-used Info-NCE loss (Oord et al., 2018) in contrastive-based SSL (see discussion in Appendix A.6. ",
+        "bbox": [
+            173,
+            272,
+            826,
+            301
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "text",
+        "text": "4 EXPERIMENT ",
+        "text_level": 1,
+        "bbox": [
+            176,
+            321,
+            316,
+            337
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "text",
+        "text": "4.1 PRE-TRAINING ",
+        "text_level": 1,
+        "bbox": [
+            174,
+            353,
+            318,
+            367
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "text",
+        "text": "Self-Supervised Pre-training of Teacher Network. By default, we use MoCo-V2 (Chen et al., 2020c) to pre-train the teacher network. Following (Chen et al., 2020a), we use ResNet as the network backbone with different depths/widths and append a multi-layer-perceptron (MLP) layer (two linear layers and one ReLU (Nair & Hinton, 2010) activation layer in between) at the end of the encoder after average pooling. The dimension of the last feature dimension is 128. All teacher networks are pre-trained for 200 epochs due to the computational limitation unless explicitly specified. As our distillation is independent with the teacher pre-training algorithm, we also show results with other self-supervised pre-trained models for teacher network, e.g., SWAV (Caron et al., 2020), SimCLR (Chen et al., 2020a). ",
+        "bbox": [
+            173,
+            378,
+            825,
+            505
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "text",
+        "text": "Self-Supervised Distillation on Student Network. We choose multiple smaller networks with fewer learnable parameters as the student network: MobileNet-v3-Large (Howard et al., 2017), EfficientNet-B0 (Tan & Le, 2019), and smaller ResNet with fewer layers (ResNet-18, 34). Similar to the pre-training for teacher network, we add one additional MLP layer on the basis of the student network. Our distillation is trained with a standard SGD optimizer with momentum 0.9 and a weight decay parameter of 1e-4 for 200 epochs. The initial learning rate is set as 0.03 and updated by a cosine decay scheduler (Nair & Hinton, 2010) with 5 warm-up epochs and batch size 256. In Eq. 4, the teacher temperature is set as $\\tau ^ { T } = 0 . 0 1$ and the student temperature is $\\tau ^ { S } = 0 . 2$ . The queue size of $K$ is 65,536. In the following subsections and appendix, we also show results with different hyper-parameter values, e.g., for $\\tau ^ { T }$ and $K$ . ",
+        "bbox": [
+            174,
+            511,
+            825,
+            651
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "text",
+        "text": "4.2 FINE-TUNING AND EVALUATION ",
+        "text_level": 1,
+        "bbox": [
+            176,
+            667,
+            437,
+            683
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "text",
+        "text": "In order to validate the effectiveness of self-supervised distillation, we choose to assess the performance of representations of the student encoder on several downstream tasks. We first report its performances of linear evaluation and semi-supervised linear evaluation on the ImageNet ILSVRC2012 (Deng et al., 2009) dataset. To measure the feature transferability brought by distillation, we also conduct evaluations on other tasks, which include object detection and segmentation on the VOC07 (Everingham et al.) and MS-COCO (Lin et al., 2014) datasets. At the end, we compare the transferability of the features learned by distillation with ordinary self-supervised contrastive learning on the tasks of linear classification on datasets from different domains. ",
+        "bbox": [
+            174,
+            694,
+            825,
+            805
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "text",
+        "text": "Linear and KNN Evaluation on ImageNet. We conduct the supervised linear classification on ImageNet-1K, which contains ${ \\sim } 1 . 3 \\mathbf { M }$ images for training, and 50,000 images for validation, spanning 1,000 categories. Following previous works in (He et al., 2020; Chen et al., 2020a), we train a single linear layer classifier on top of the frozen network encoder after self-supervised pretraining/distillation. SGD optimizer is used to train the linear classifier for 100 epochs with weight decay to be 0. The initial learning rate is set as 30 and is then reduced by a factor of 10 at 60 and 80 epochs (similar as in Tian et al. (2019a)). Notably, when training the linear classifier for MobileNet and EfficientNet, we reduce the initial learning rate to 3. The results are reported in terms of Top-1 and Top-5 accuracy. We also perform classification using $K$ -Nearest Neighbors (KNN) based on the learned 128d vector from the last MLP layer. The sample is classified by taking the most frequent label of its $K$ ( $K = 1 0$ ) nearest neighbors. ",
+        "bbox": [
+            174,
+            811,
+            825,
+            924
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "table",
+        "img_path": "images/34e5b547413177102a008d3fb8f2ab5d87724d22eeda84157c53976330012be9.jpg",
+        "table_caption": [
+            "Table 1: ImageNet-1k test accuracy $( \\% )$ using KNN and linear classification for multiple students and MoCov2 pre-trained deeper teacher architectures. $\\pmb { \\chi }$ denotes MoCo-V2 self-supervised learning baselines before distillation. \\* indicates using a deeper teacher encoder pre-trained by SWAV, where additional small-patches are also utilized during distillation and trained for 800 epochs. $K$ denotes Top-1 accuracy using KNN. T-1 and T-5 denote Top-1 and Top-5 accuracy using linear evaluation. First column shows Top-1 Acc. of Teacher network. First row shows the supervised performances of student networks. "
+        ],
+        "table_footnote": [],
+        "table_body": "<table><tr><td rowspan=\"2\">S T</td><td rowspan=\"2\">T-1</td><td rowspan=\"2\">K</td><td colspan=\"2\">Eff-bo</td><td colspan=\"3\">Eff-b1</td><td colspan=\"3\">Mob-v3</td><td colspan=\"3\">R-18</td><td colspan=\"3\">R-34</td></tr><tr><td>T-1</td><td>T-5</td><td>K</td><td>T-1</td><td>T-5</td><td>K</td><td>T-1</td><td>T-5</td><td>K</td><td>T-1</td><td>T-5</td><td>K</td><td>T-1</td><td>T-5</td></tr><tr><td>Supervised Acc.</td><td></td><td></td><td>77.3</td><td></td><td></td><td>79.2</td><td></td><td></td><td>75.2</td><td></td><td></td><td>72.1</td><td></td><td></td><td>75.0</td><td></td></tr><tr><td>X</td><td></td><td>30.0</td><td>42.2</td><td></td><td>68.534.4</td><td>50.7</td><td>74.6</td><td>27.5</td><td>36.3</td><td>62.2</td><td>36.7</td><td>52.5</td><td>77.0</td><td>）41.5</td><td>57.4</td><td>81.6</td></tr><tr><td>R-50 △</td><td>67.4</td><td>46.0 +16.0</td><td>61.3 +19.1</td><td>82.7 +14.2</td><td>46.1 +16.1</td><td>61.4 +10.7</td><td>83.1 +8.8</td><td>44.8 +17.3</td><td>55.2 +18.9</td><td>80.3 +18.1</td><td>43.4 +6.7</td><td>57.9 +5.1</td><td>82.0 +4.8</td><td>45.2 +3.7</td><td>58.5 +1.1</td><td>82.6 +1.0</td></tr><tr><td>R-101 △</td><td>70.3</td><td>50.1 +20.1</td><td>63.0 +20.8</td><td>83.8 +15.3</td><td>50.3 +15.9</td><td>63.4 +12.7</td><td>84.6 +10.0</td><td>48.8 +21.3</td><td>59.9 +23.6</td><td>83.5 +21.3</td><td>48.6 +11.9</td><td>58.9 +6.4</td><td>82.5 +5.5</td><td>50.5 +9.0</td><td>61.6 +4.2</td><td>84.9 +3.3</td></tr><tr><td>R-152</td><td></td><td>50.7</td><td>65.3</td><td>86.0</td><td>52.4</td><td>67.3</td><td>86.9</td><td>49.5</td><td>61.4</td><td>84.6</td><td>49.1</td><td>59.5</td><td>83.3</td><td>51.4</td><td>62.7</td><td>85.8</td></tr><tr><td>△</td><td>74.2</td><td>+20.7</td><td>+23.1</td><td>+17.5</td><td>+18.0</td><td>+16.6</td><td>+12.3</td><td>+22.0</td><td>+25.1</td><td>+22.4</td><td>+12.4</td><td>+7.0</td><td>+6.3</td><td>+9.9</td><td>+5.3</td><td>+4.2</td></tr><tr><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td></tr><tr><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td></tr><tr><td>R50×2*</td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td></tr><tr><td></td><td></td><td></td><td></td><td>87.4</td><td>60.3</td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td></tr><tr><td>△</td><td>77.3</td><td>57.4 +27.4</td><td>67.6 +25.4</td><td>+18.9</td><td>+25.9</td><td>68.0 +17.3</td><td>87.6 +13.0</td><td>55.9 +18.9</td><td>68.2 +31.9</td><td>88.2 +26.0</td><td>55.3 +18.6</td><td>63.0 +10.5</td><td>84.9 +7.9</td><td>58.2 +16.7</td><td>65.7 +8.3</td><td>86.8 +5.2</td></tr></table>",
+        "bbox": [
+            176,
+            190,
+            826,
+            359
+        ],
+        "page_idx": 5
+    },
+    {
+        "type": "image",
+        "img_path": "images/3623a3702ec553403c1eb676464c6cd00d584bc812cc122bd666f370eb648b7c.jpg",
+        "image_caption": [
+            "Figure 3: ImageNet-1k Top-1 accuracy for semi-supervised evaluations using $1 \\%$ (red line), $10 \\%$ (blue line) of the annotations for linear fine-tuning, in comparison with the fully supervised (green line) linear evaluation baseline for SEED. For the points whose Teacher’s number of parameters is at 0, we show the semi-supervised linear evaluation results of MoCo-V2 without any distillation. The Student models tend to perform better on the semi-supervised tasks after distillation from larger Teachers. "
+        ],
+        "image_footnote": [],
+        "bbox": [
+            191,
+            386,
+            807,
+            500
+        ],
+        "page_idx": 5
+    },
+    {
+        "type": "text",
+        "text": "",
+        "bbox": [
+            176,
+            589,
+            821,
+            632
+        ],
+        "page_idx": 5
+    },
+    {
+        "type": "text",
+        "text": "Table 1 shows the results with various teacher networks and student networks. We list the baseline of contrastive self-supervised pre-training using MoCo-V2 (Chen et al., 2020c) in the first row for each student architecture. We can see clearly that smaller networks perform rather worse. For example, MobileNet-V3 can only reach $3 6 . 3 \\%$ . This aligns well with previous conclusions from (Chen et al., 2020a;b) that bigger models are desired to perform better in contrastive-based self-supervised pretraining. We conjecture that this is mainly caused by the inability of smaller network to discriminate instances in a large-scale dataset. The results also clearly demonstrate that the distillation from a larger network helps boosting the performances of small networks, and show obvious improvement. For instance, with MoCo-V2 pre-trained ResNet-152 (for 400 epochs) as the teacher network, the Top-1 accuracy of MobileNet-V3-Large can be significantly improved from $3 6 . 3 \\%$ to $6 1 . 4 \\%$ . Furthermore, we use ResNet- $5 0 \\times 2$ (provided in Caron et al. (2020)) as the teacher network and adopt the multi-crop trick (see A.2 for details). The accuracy can be further improved to $6 8 . 2 \\%$ (last row of Table 1) for MobileNet-V3-Large with 800 epochs of distillation. We note that the gain benefited from distillation becomes more distinct on smaller architectures and we further study the effect of various teacher models in ablations. ",
+        "bbox": [
+            174,
+            638,
+            826,
+            847
+        ],
+        "page_idx": 5
+    },
+    {
+        "type": "text",
+        "text": "Semi-Supervised Evaluation on ImageNet. Following (Oord et al., 2018; Kornblith et al., 2019; Kolesnikov et al., 2019), we evaluate the representation on the semi-supervised task, where a fixed $1 \\%$ or $10 \\%$ subsets of ImageNet training data (Chen et al., 2020a) are provided with the annotations. After the self-supervised learning with and without distillation, we also train a classifier on top of the representation. The results are shown in Figure 3, where the baseline without distillation is depicted when teacher parameters are 0. As we can see, the accuracy is also improved remarkably with SEED distillation, and a stronger teacher network with more parameters leads to a better performed student network. ",
+        "bbox": [
+            174,
+            854,
+            825,
+            924
+        ],
+        "page_idx": 5
+    },
+    {
+        "type": "table",
+        "img_path": "images/bc2a89152889a8d15d369e7b291816928ecaa6276dd595b93bc9cd3b191f82cd.jpg",
+        "table_caption": [
+            "Table 2: Object detection and instance segmentation results using contrastive self-supervised learning and SEED distillation using ResNet-18 as backbone: bounding-box AP $( \\mathbf { A P } ^ { b b } )$ and mask AP $( \\mathbf { A P } ^ { m k } )$ evaluated on VOC07-val and COCO testing split. More results on different backbones can be found in the Appendix. Subscript in green represents improvement is larger than 0.3. "
+        ],
+        "table_footnote": [],
+        "table_body": "<table><tr><td rowspan=\"2\">S</td><td rowspan=\"2\">T</td><td colspan=\"3\">VOC Obj. Det.</td><td colspan=\"3\">COCO Obj. Det.</td><td colspan=\"3\">COCO Inst. Segm.</td></tr><tr><td>Apbb</td><td>AP</td><td>AP</td><td>Apbb</td><td>AP</td><td>APP</td><td>Apmk</td><td>AP</td><td>AP</td></tr><tr><td></td><td>X</td><td>46.1</td><td>74.5</td><td>48.6</td><td>35.0</td><td>53.9</td><td>37.7</td><td>31.0</td><td>51.1</td><td>33.1</td></tr><tr><td></td><td>R-50</td><td>46.1(0.0)</td><td>74.8(+0.3)</td><td>49.1(+0.5)</td><td>35.3(+0.3)</td><td>54.2(+0.3)</td><td>37.8(+0.1)</td><td>31.1(+0.1)</td><td>51.1(0.0)</td><td>33.2(+0.1)</td></tr><tr><td>R-18 R-101</td><td></td><td>46.8(+0.7)</td><td>75.8(+1.3)</td><td>49.3(+0.7)</td><td>35.3(+0.3)</td><td>54.3(+0.4)</td><td>37.9(+0.2)</td><td>31.3(+0.3)</td><td>51.3(+0.2)</td><td>33.4(+0.3)</td></tr><tr><td></td><td>R-152</td><td>46.8(+0.7)</td><td>75.9(+1.4)</td><td>50.2(+1.6)</td><td>35.4(+0.4)</td><td>54.4(+0.5)</td><td>38.0(+0.3)</td><td>31.3(+0.3)</td><td>51.4(+0.3)</td><td>33.4(+0.3)</td></tr></table>",
+        "bbox": [
+            173,
+            151,
+            831,
+            255
+        ],
+        "page_idx": 6
+    },
+    {
+        "type": "image",
+        "img_path": "images/15d565abe6e76f1be7191527af52d1cfceecfa87f6083e78a4a48b0e6be4ff99.jpg",
+        "image_caption": [
+            "Figure 4: ImageNet-1k Accuracy $( \\% )$ of student network (EfficientNet-B0 and ResNet-18) transferred to other domains (CIFAR-10, CIFAR-100, SUN-397 datasets) with and without distillation from lager architectures (ResNet-50/101/152). "
+        ],
+        "image_footnote": [],
+        "bbox": [
+            174,
+            284,
+            821,
+            405
+        ],
+        "page_idx": 6
+    },
+    {
+        "type": "text",
+        "text": "",
+        "bbox": [
+            174,
+            470,
+            825,
+            511
+        ],
+        "page_idx": 6
+    },
+    {
+        "type": "text",
+        "text": "Transferring to Classification. To further study whether the improvement of the learned representations by distillation is confined to ImageNet, we evaluate on additional classification datasets to study the generalization and transferability of the feature representation. We strictly follow the linear evaluation and fine-tuning settings from (Kornblith et al., 2019; Chen et al., 2020a; Grill et al., 2020), that a linear layer is trained on the basis of frozen features. We report Top-1 Accuracy of models before and after distillation from various architectures on CIFAR-10, CIFAR-100 (Krizhevsky et al., 2009), SUN-397 (Xiao et al., 2010) datasets (see Figure 4). More details regarding pre-processing and training can be found in A.1.2. Notably, we observe that our distillation surpasses contrastive self-supervised pre-training consistently on all benchmarks, verifying the effectiveness of SEED. This also proves the generalization ability of the learned representations from distillation to a wide range of data domain and different classes. ",
+        "bbox": [
+            174,
+            518,
+            825,
+            671
+        ],
+        "page_idx": 6
+    },
+    {
+        "type": "text",
+        "text": "Transferring to Detection and Segmentation. We conduct two downstream tasks here. The first is Faster R-CNN (Ren et al., 2015) model for object detection trained on VOC- $^ { 0 7 + 1 2 }$ train $^ +$ val set and evaluated on VOC-07 test split. The second is Mask R-CNN (He et al., 2017) model for the object detection and instance segmentation on COCO 2017 dataset (Lin et al., 2014). The pre-trained model serves as the initial weight and following He et al. (2020), we fine-tune all the layers of the model. More experiment settings can be found in A.2. The results are illustrated in Table 2. As we can see, on VOC, the distilled pre-trained model achieves a large improvement. With ResNet-152 as the teacher network, the Resnet18-based Faster R-CNN model shows $+ 0 . 7$ point improvement on AP, $+ 1 . 4$ improvement on $\\mathrm { { A P } _ { 5 0 } }$ and $+ 1 . 6$ on $\\mathsf { A P } _ { 7 5 }$ . On COCO, the improvement is relatively minor and the reason could be that COCO training set has ${ \\sim } 1 1 8 \\mathrm { k }$ training images while VOC has only ${ \\sim } 1 6 . 5 \\mathrm { k }$ training images. A larger training set with more fine-tuning iterations reduces the importance of the initial weights. ",
+        "bbox": [
+            173,
+            679,
+            825,
+            844
+        ],
+        "page_idx": 6
+    },
+    {
+        "type": "text",
+        "text": "4.3 ABLATION STUDY ",
+        "text_level": 1,
+        "bbox": [
+            176,
+            867,
+            339,
+            881
+        ],
+        "page_idx": 6
+    },
+    {
+        "type": "text",
+        "text": "We now explore the effects of distillation using different Teacher architectures, Teacher Pre-training algorithms, various distillation strategies and hyper-parameters. ",
+        "bbox": [
+            174,
+            895,
+            821,
+            924
+        ],
+        "page_idx": 6
+    },
+    {
+        "type": "table",
+        "img_path": "images/6b287c612ea27b0206b8c1a70fa565f801cdb95b0af984578c7bb7576521984e.jpg",
+        "table_caption": [
+            "Table 3: ImageNet-1k Accuracy $( \\% )$ of student network (ResNet-18) distilled from variants of selfsupervised ResNet-50. P-E/D-E represent the pretraining and distillation epochs. T./S.-Top represent testing accuracy of Teacher and Student. ∗ represents distillation using additional small patches. First row is the ResNet-18 SSL baseline using MoCo-v2 trained for 200 epochs. "
+        ],
+        "table_footnote": [],
+        "table_body": "<table><tr><td>Teacher</td><td></td><td>P-E D-E T.Top-1 S.Top-1 S.Top-5</td></tr><tr><td></td><td>X X</td><td>X 52.5 77.0</td></tr><tr><td>X</td><td>200</td><td></td></tr><tr><td>MoCo</td><td>200</td><td>60.6 52.1 77.0</td></tr><tr><td>SimCLR</td><td>200 200</td><td>65.6 57.5 81.7</td></tr><tr><td>MoCo-v2 200 800</td><td>200 67.4</td><td>57.9 82.0</td></tr><tr><td>SWAV 800</td><td>200 71.1</td><td>60.5 83.5</td></tr><tr><td></td><td>100 75.3</td><td>61.1 83.8</td></tr><tr><td>800</td><td>200 75.3</td><td>61.7 84.2</td></tr><tr><td>800</td><td>400 75.3</td><td>62.0 84.4</td></tr><tr><td>SWAV* 800</td><td>200 75.3</td><td>62.6 84.8</td></tr></table>",
+        "bbox": [
+            179,
+            214,
+            477,
+            343
+        ],
+        "page_idx": 7
+    },
+    {
+        "type": "image",
+        "img_path": "images/8cecf45d517b8c5b96207f2fc90806edfe43935c7cfa9552a0ceb80d0db26208.jpg",
+        "image_caption": [
+            "Figure 5: Accuracy $( \\% )$ of student networks (EfficientNet-b0 and ResNet-18) on ImageNet distilled from wider MoCo-v2 pre-trained ResNet (ResNet-50/101/152×2). "
+        ],
+        "image_footnote": [],
+        "bbox": [
+            529,
+            118,
+            821,
+            287
+        ],
+        "page_idx": 7
+    },
+    {
+        "type": "text",
+        "text": "Different Teacher Networks. Figure 5 summarizes the accuracy of ResNet-18 and EfficientNet-B0 distilled from wider and deeper ResNet architectures. We see clear performance improvement as depth and width of teacher network increase: compared to ResNet-50, deeper (ResNet-101) and wider (ResNet- ${ 5 0 \\times 2 }$ ) substantially improve the accuracy. However, further architectural enlargement has relatively limited effects, and we suspect the accuracy might be limited by the student network capacity in this case. ",
+        "bbox": [
+            174,
+            378,
+            825,
+            462
+        ],
+        "page_idx": 7
+    },
+    {
+        "type": "text",
+        "text": "Different Teacher Pre-training Algorithms. In Table 3, we show the Top-1 accuracy of ResNet-18 distilled from ResNet-50 with different pre-training algorithms, i.e., MoCo-V1 (He et al., 2020), MoCo-V2 (Chen et al., 2020c), SimCLR (Chen et al., 2020a), and SWAV (Caron et al., 2020)). Notably, the aforementioned methods all unanimously adopt contrastive-based pre-training except SWAV, which is based upon online clustering. We find that our SEED is agnostic to pre-training approaches, making it easy to use any self-supervised models (including clustering-based approach like SWAV) in self-supervised distillation. In addition, we observe that more training epochs for both teacher $S S L$ and distillation epochs can bring beneficial gain. ",
+        "bbox": [
+            174,
+            469,
+            825,
+            580
+        ],
+        "page_idx": 7
+    },
+    {
+        "type": "text",
+        "text": "Other Distillation Strategies. We explore several alternative distillation strategies. l2-Distance: where the $l 2$ -distance of teacher & student’s embeddings are minimized, motivated by Romero et al. (2014). $K$ -Means: we exploit $K$ -Means clustering to assign a pseudo-label based on the teacher network’s representation. Online Clustering: we continuously update the clustering centers during distillation for pseudo-label generation. Binary Contrastive Loss: we adopt an Info-NCE alike loss for contrastive distillation (Tian et al., 2019a). We provide details for other strategies in A.4. Table 4 shows the results for each method on ResNet-18 (student) distilled from ResNet-50. From the results, the simple $l 2$ -distance minimizing approach can achieve a decent accuracy, which demonstrates the effectiveness of applying the distillation idea to the self-supervised learning. Beyond that, we study the effect of the original SSL (MoCo-V2) supervision as supplementary loss to SEED and find it does not bring additional benefits to distillation. We find close results from these two strategies (Top-1 linear Acc.), SEED achieves $5 7 . 9 \\%$ , while $\\mathbf { S E E D + M o C o - V } 2$ achieves $5 7 . 6 \\%$ . This implies that the loss of SEED can to a large extent cover the original $S S L$ loss, and it is not necessary to conduct SSL any further during distillation. Meanwhile, our proposed SEED outperforms these alternatives with highest accuracy, which shows the superiority of aligning the student towards the teacher and contrasting with the irrelevant samples. ",
+        "bbox": [
+            173,
+            588,
+            825,
+            810
+        ],
+        "page_idx": 7
+    },
+    {
+        "type": "text",
+        "text": "Other Hyper-Parameters. Table 5 summarizes the distillation performances on multiple datasets using different temperature $\\tau ^ { T }$ . We observe a better performance when decreasing $\\tau ^ { T }$ to 0.01 for ImageNet-1k and CIFAR-10 dataset, and to 1e-3 for CIFAR-100 datasets. When $\\tau$ is large, the softmax-normalized similarity score of $p _ { j } ^ { T }$ between ${ \\mathbf z } _ { i } ^ { T }$ and instance ${ \\bf d } _ { j }$ in the queue $\\mathbf { D } ^ { + }$ also becomes large, which means the student’s feature should be less discriminative with the features of other images to some extent. When $\\tau ^ { T }$ is $_ 0$ , the teacher model will generate a one-hot vector, which only treats ${ \\mathbf z } _ { i } ^ { T }$ as a positive instance and all others in the queue as negative. Thus, the best $\\tau$ is a trade-off depending on the data distribution. We further compare effect of different hyper-parameters in A.8. ",
+        "bbox": [
+            174,
+            816,
+            825,
+            929
+        ],
+        "page_idx": 7
+    },
+    {
+        "type": "table",
+        "img_path": "images/111d557f7135948444d3e5bb803f101bbeaabe2c0cfcc43a81faac6da770ecd7.jpg",
+        "table_caption": [
+            "Table 4: Top-1/5 accuracy of linear classification results on ImageNet using different distillation strategies on ResNet-18 (student) and ResNet-50 (teacher) architectures. "
+        ],
+        "table_footnote": [],
+        "table_body": "<table><tr><td>Method</td><td colspan=\"2\">Top-1 Acc. Top-5 Acc.</td></tr><tr><td>l2-Distance</td><td>55.3</td><td>80.3</td></tr><tr><td>K-Means</td><td>51.0</td><td>75.8</td></tr><tr><td>Online Clustering</td><td>56.4</td><td>81.2</td></tr><tr><td>Binary Contr. Loss</td><td>57.4</td><td>81.5</td></tr><tr><td>SEED +MoCo-V2</td><td>57.6</td><td>81.8</td></tr><tr><td>SEED</td><td>57.9</td><td>82.0</td></tr></table>",
+        "bbox": [
+            173,
+            160,
+            460,
+            273
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "table",
+        "img_path": "images/0ce07da924f5f7e589e1097898757fd0f72682d6d444c981019b465d74535daa.jpg",
+        "table_caption": [
+            "Table 5: Effect of $\\tau ^ { T }$ for the distillation of ResNet-18 (student), ResNet-50 (teacher) on multiple datasets. "
+        ],
+        "table_footnote": [],
+        "table_body": "<table><tr><td rowspan=\"2\">YT</td><td>ImageNet</td><td>CIFAR-10</td><td>CIFAR-100</td></tr><tr><td>Top-1 Top-5</td><td>Top-1</td><td>Top-1</td></tr><tr><td>0.3</td><td>54.8 80.0</td><td>78.7</td><td>46.6</td></tr><tr><td>0.1</td><td>54.9 80.1</td><td>83.0</td><td>50.1</td></tr><tr><td>0.05</td><td>56.5 81.3</td><td>84.4</td><td>56.2</td></tr><tr><td>0.01</td><td>57.9 82.0</td><td>87.5 </td><td>60.6</td></tr><tr><td>1e-3</td><td>57.6 81.8</td><td>86.9</td><td>60.8</td></tr></table>",
+        "bbox": [
+            496,
+            147,
+            813,
+            265
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "5 CONCLUSIONS ",
+        "text_level": 1,
+        "bbox": [
+            176,
+            301,
+            328,
+            318
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "Self-Supervised Learning is acknowledged for its remarkable ability in learning from unlabeled, and large scale data. However, a critical impedance for the SSL pre-training on smaller architecture comes from its low capacity of discriminating enormous number of instances. Instead of directly learning from unlabeled data, we proposed SEED as a novel self-supervised learning paradigm, which learns representation by self-supervised distillation from a bigger SSL pre-trained model. We show in extensive experiments that SEED effectively addresses the weakness of self-supervised learning for small models and achieves state-of-the-art results on various benchmarks of small architectures. ",
+        "bbox": [
+            173,
+            333,
+            825,
+            431
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "REFERENCES ",
+        "text_level": 1,
+        "bbox": [
+            174,
+            454,
+            285,
+            469
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "Soroush Abbasi Koohpayegani, Ajinkya Tejankar, and Hamed Pirsiavash. Compress: Self-supervised learning by compressing representations. Advances in Neural Information Processing Systems, 33, 2020. ",
+        "bbox": [
+            176,
+            477,
+            823,
+            520
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "Sungsoo Ahn, Shell Xu Hu, Andreas Damianou, Neil D Lawrence, and Zhenwen Dai. Variational information distillation for knowledge transfer. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pp. 9163–9171, 2019. ",
+        "bbox": [
+            174,
+            531,
+            826,
+            574
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "Cristian Bucilua, Rich Caruana, and Alexandru Niculescu-Mizil. Model compression. In ˇ Proceedings of the 12th ACM SIGKDD international conference on Knowledge discovery and data mining, pp. 535–541, 2006. ",
+        "bbox": [
+            173,
+            585,
+            826,
+            627
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "Mathilde Caron, Piotr Bojanowski, Armand Joulin, and Matthijs Douze. Deep clustering for unsupervised learning of visual features. In Proceedings of the European Conference on Computer Vision (ECCV), pp. 132–149, 2018. ",
+        "bbox": [
+            173,
+            638,
+            825,
+            681
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "Mathilde Caron, Ishan Misra, Julien Mairal, Priya Goyal, Piotr Bojanowski, and Armand Joulin. Unsupervised learning of visual features by contrasting cluster assignments. arXiv preprint arXiv:2006.09882, 2020. ",
+        "bbox": [
+            173,
+            693,
+            825,
+            736
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "Guobin Chen, Wongun Choi, Xiang Yu, Tony Han, and Manmohan Chandraker. Learning efficient object detection models with knowledge distillation. In Advances in Neural Information Processing Systems, pp. 742–751, 2017. ",
+        "bbox": [
+            173,
+            746,
+            826,
+            790
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "Ting Chen, Simon Kornblith, Mohammad Norouzi, and Geoffrey Hinton. A simple framework for contrastive learning of visual representations. arXiv preprint arXiv:2002.05709, 2020a. ",
+        "bbox": [
+            173,
+            801,
+            825,
+            830
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "Ting Chen, Simon Kornblith, Kevin Swersky, Mohammad Norouzi, and Geoffrey Hinton. Big self-supervised models are strong semi-supervised learners. arXiv preprint arXiv:2006.10029, 2020b. ",
+        "bbox": [
+            174,
+            840,
+            825,
+            883
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "Xinlei Chen, Haoqi Fan, Ross Girshick, and Kaiming He. Improved baselines with momentum contrastive learning. arXiv preprint arXiv:2003.04297, 2020c. ",
+        "bbox": [
+            173,
+            895,
+            823,
+            924
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "Jang Hyun Cho and Bharath Hariharan. On the efficacy of knowledge distillation. In Proceedings of the IEEE International Conference on Computer Vision, pp. 4794–4802, 2019. ",
+        "bbox": [
+            169,
+            103,
+            825,
+            133
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "Jia Deng, Wei Dong, Richard Socher, Li-Jia Li, Kai Li, and Li Fei-Fei. Imagenet: A large-scale hierarchical image database. In 2009 IEEE conference on computer vision and pattern recognition, pp. 248–255. Ieee, 2009. ",
+        "bbox": [
+            173,
+            141,
+            823,
+            184
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "M. Everingham, L. Van Gool, C. K. I. Williams, J. Winn, and A. Zisserman. The PASCAL Visual Object Classes Challenge 2007 (VOC2007) Results. http://www.pascalnetwork.org/challenges/VOC/voc2007/workshop/index.html. ",
+        "bbox": [
+            173,
+            193,
+            821,
+            237
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "Zeyu Feng, Chang Xu, and Dacheng Tao. Self-supervised representation learning by rotation feature decoupling. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pp. 10364–10374, 2019. ",
+        "bbox": [
+            173,
+            244,
+            825,
+            289
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "Spyros Gidaris, Praveer Singh, and Nikos Komodakis. Unsupervised representation learning by predicting image rotations. arXiv preprint arXiv:1803.07728, 2018. ",
+        "bbox": [
+            171,
+            297,
+            825,
+            327
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "Jean-Bastien Grill, Florian Strub, Florent Altché, Corentin Tallec, Pierre H Richemond, Elena Buchatskaya, Carl Doersch, Bernardo Avila Pires, Zhaohan Daniel Guo, Mohammad Gheshlaghi Azar, et al. Bootstrap your own latent: A new approach to self-supervised learning. arXiv preprint arXiv:2006.07733, 2020. ",
+        "bbox": [
+            173,
+            335,
+            825,
+            392
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "Yandong Guo, Lei Zhang, Yuxiao Hu, Xiaodong He, and Jianfeng Gao. Ms-celeb-1m: A dataset and benchmark for large-scale face recognition. In European conference on computer vision, pp. 87–102. Springer, 2016. ",
+        "bbox": [
+            173,
+            401,
+            823,
+            445
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "Michael Gutmann and Aapo Hyvärinen. Noise-contrastive estimation: A new estimation principle for unnormalized statistical models. In Proceedings of the Thirteenth International Conference on Artificial Intelligence and Statistics, pp. 297–304, 2010. ",
+        "bbox": [
+            173,
+            454,
+            823,
+            497
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image recognition. In Proceedings of the IEEE conference on computer vision and pattern recognition, pp. 770–778, 2016. ",
+        "bbox": [
+            173,
+            506,
+            823,
+            549
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "Kaiming He, Georgia Gkioxari, Piotr Dollár, and Ross Girshick. Mask r-cnn. In Proceedings of the IEEE international conference on computer vision, pp. 2961–2969, 2017. ",
+        "bbox": [
+            171,
+            558,
+            823,
+            588
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "Kaiming He, Haoqi Fan, Yuxin Wu, Saining Xie, and Ross Girshick. Momentum contrast for unsupervised visual representation learning. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp. 9729–9738, 2020. ",
+        "bbox": [
+            173,
+            595,
+            823,
+            640
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "Olivier J Hénaff, Aravind Srinivas, Jeffrey De Fauw, Ali Razavi, Carl Doersch, SM Eslami, and Aaron van den Oord. Data-efficient image recognition with contrastive predictive coding. arXiv preprint arXiv:1905.09272, 2019. ",
+        "bbox": [
+            173,
+            648,
+            823,
+            691
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "Geoffrey Hinton, Oriol Vinyals, and Jeff Dean. Distilling the knowledge in a neural network. arXiv preprint arXiv:1503.02531, 2015. ",
+        "bbox": [
+            173,
+            700,
+            825,
+            729
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "Andrew Howard, Mark Sandler, Grace Chu, Liang-Chieh Chen, Bo Chen, Mingxing Tan, Weijun Wang, Yukun Zhu, Ruoming Pang, Vijay Vasudevan, et al. Searching for mobilenetv3. In Proceedings of the IEEE International Conference on Computer Vision, pp. 1314–1324, 2019. ",
+        "bbox": [
+            173,
+            738,
+            825,
+            782
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "Andrew G Howard, Menglong Zhu, Bo Chen, Dmitry Kalenichenko, Weijun Wang, Tobias Weyand, Marco Andreetto, and Hartwig Adam. Mobilenets: Efficient convolutional neural networks for mobile vision applications. arXiv preprint arXiv:1704.04861, 2017. ",
+        "bbox": [
+            173,
+            790,
+            825,
+            834
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "Zehao Huang and Naiyan Wang. Like what you like: Knowledge distill via neuron selectivity transfer. arXiv preprint arXiv:1707.01219, 2017. ",
+        "bbox": [
+            171,
+            842,
+            825,
+            872
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "Xiaoqi Jiao, Yichun Yin, Lifeng Shang, Xin Jiang, Xiao Chen, Linlin Li, Fang Wang, and Qun Liu. Tinybert: Distilling bert for natural language understanding. arXiv preprint arXiv:1909.10351, 2019. ",
+        "bbox": [
+            174,
+            881,
+            826,
+            922
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "Alexander Kolesnikov, Xiaohua Zhai, and Lucas Beyer. Revisiting self-supervised visual representation learning. In Proceedings of the IEEE conference on Computer Vision and Pattern Recognition, pp. 1920–1929, 2019. ",
+        "bbox": [
+            174,
+            103,
+            825,
+            146
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Animesh Koratana, Daniel Kang, Peter Bailis, and Matei Zaharia. Lit: Learned intermediate representation training for model compression. In International Conference on Machine Learning, pp. 3509–3518, 2019. ",
+        "bbox": [
+            173,
+            154,
+            825,
+            196
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Simon Kornblith, Jonathon Shlens, and Quoc V Le. Do better imagenet models transfer better? In Proceedings of the IEEE conference on computer vision and pattern recognition, pp. 2661–2671, 2019. ",
+        "bbox": [
+            176,
+            204,
+            823,
+            247
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Alex Krizhevsky, Geoffrey Hinton, et al. Learning multiple layers of features from tiny images. 2009. ",
+        "bbox": [
+            173,
+            256,
+            825,
+            272
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Junnan Li, Pan Zhou, Caiming Xiong, Richard Socher, and Steven CH Hoi. Prototypical contrastive learning of unsupervised representations. arXiv preprint arXiv:2005.04966, 2020. ",
+        "bbox": [
+            171,
+            279,
+            823,
+            309
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Tsung-Yi Lin, Michael Maire, Serge Belongie, James Hays, Pietro Perona, Deva Ramanan, Piotr Dollár, and C Lawrence Zitnick. Microsoft coco: Common objects in context. In European conference on computer vision, pp. 740–755. Springer, 2014. ",
+        "bbox": [
+            176,
+            315,
+            823,
+            359
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Yifan Liu, Ke Chen, Chris Liu, Zengchang Qin, Zhenbo Luo, and Jingdong Wang. Structured knowledge distillation for semantic segmentation. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pp. 2604–2613, 2019. ",
+        "bbox": [
+            176,
+            367,
+            823,
+            411
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Asit Mishra and Debbie Marr. Apprentice: Using knowledge distillation techniques to improve low-precision network accuracy. arXiv preprint arXiv:1711.05852, 2017. ",
+        "bbox": [
+            171,
+            417,
+            823,
+            446
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Ishan Misra and Laurens van der Maaten. Self-supervised learning of pretext-invariant representations. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp. 6707–6717, 2020. ",
+        "bbox": [
+            174,
+            454,
+            825,
+            497
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Vinod Nair and Geoffrey E Hinton. Rectified linear units improve restricted boltzmann machines. In ICML, 2010. ",
+        "bbox": [
+            173,
+            506,
+            825,
+            535
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Mehdi Noroozi and Paolo Favaro. Unsupervised learning of visual representations by solving jigsaw puzzles. In European Conference on Computer Vision, pp. 69–84. Springer, 2016. ",
+        "bbox": [
+            171,
+            542,
+            823,
+            573
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Aaron van den Oord, Yazhe Li, and Oriol Vinyals. Representation learning with contrastive predictive coding. arXiv preprint arXiv:1807.03748, 2018. ",
+        "bbox": [
+            171,
+            579,
+            823,
+            609
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Adam Paszke, Sam Gross, Francisco Massa, Adam Lerer, James Bradbury, Gregory Chanan, Trevor Killeen, Zeming Lin, Natalia Gimelshein, Luca Antiga, et al. Pytorch: An imperative style, high-performance deep learning library. In Advances in neural information processing systems, pp. 8026–8037, 2019. ",
+        "bbox": [
+            174,
+            617,
+            826,
+            674
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Antonio Polino, Razvan Pascanu, and Dan Alistarh. Model compression via distillation and quantization. arXiv preprint arXiv:1802.05668, 2018. ",
+        "bbox": [
+            173,
+            681,
+            823,
+            712
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Shaoqing Ren, Kaiming He, Ross Girshick, and Jian Sun. Faster r-cnn: Towards real-time object detection with region proposal networks. In Advances in neural information processing systems, pp. 91–99, 2015. ",
+        "bbox": [
+            174,
+            718,
+            826,
+            762
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Adriana Romero, Nicolas Ballas, Samira Ebrahimi Kahou, Antoine Chassang, Carlo Gatta, and Yoshua Bengio. Fitnets: Hints for thin deep nets. 2014. ",
+        "bbox": [
+            173,
+            770,
+            823,
+            799
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Mark Sandler, Andrew Howard, Menglong Zhu, Andrey Zhmoginov, and Liang-Chieh Chen. Mobilenetv2: Inverted residuals and linear bottlenecks. In Proceedings of the IEEE conference on computer vision and pattern recognition, pp. 4510–4520, 2018. ",
+        "bbox": [
+            173,
+            806,
+            825,
+            849
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Victor Sanh, Lysandre Debut, Julien Chaumond, and Thomas Wolf. Distilbert, a distilled version of bert: smaller, faster, cheaper and lighter. arXiv preprint arXiv:1910.01108, 2019. ",
+        "bbox": [
+            174,
+            858,
+            821,
+            887
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Zhiqiang Shen and Marios Savvides. Meal v2: Boosting vanilla resnet-50 to $80 \\text{‰}$ top-1 accuracy on imagenet without tricks. arXiv preprint arXiv:2009.08453, 2020. ",
+        "bbox": [
+            174,
+            895,
+            821,
+            924
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Siqi Sun, Yu Cheng, Zhe Gan, and Jingjing Liu. Patient knowledge distillation for bert model compression. arXiv preprint arXiv:1908.09355, 2019. ",
+        "bbox": [
+            173,
+            103,
+            825,
+            132
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "Zhiqing Sun, Hongkun Yu, Xiaodan Song, Renjie Liu, Yiming Yang, and Denny Zhou. Mobilebert: a compact task-agnostic bert for resource-limited devices. arXiv preprint arXiv:2004.02984, 2020. ",
+        "bbox": [
+            174,
+            140,
+            823,
+            171
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "Mingxing Tan and Quoc V Le. Efficientnet: Rethinking model scaling for convolutional neural networks. arXiv preprint arXiv:1905.11946, 2019. ",
+        "bbox": [
+            173,
+            179,
+            823,
+            208
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "Yonglong Tian, Dilip Krishnan, and Phillip Isola. Contrastive representation distillation. In International Conference on Learning Representations, 2019a. ",
+        "bbox": [
+            173,
+            215,
+            825,
+            246
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "Yonglong Tian, Dilip Krishnan, and Phillip Isola. Contrastive multiview coding. arXiv preprint arXiv:1906.05849, 2019b. ",
+        "bbox": [
+            174,
+            253,
+            823,
+            284
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "Iulia Turc, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. Well-read students learn better: On the importance of pre-training compact models. arXiv preprint arXiv:1908.08962, 2019. ",
+        "bbox": [
+            173,
+            291,
+            825,
+            321
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "Tongzhou Wang and Phillip Isola. Understanding contrastive representation learning through alignment and uniformity on the hypersphere. arXiv preprint arXiv:2005.10242, 2020. ",
+        "bbox": [
+            171,
+            329,
+            825,
+            359
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "Yuxin Wu, Alexander Kirillov, Francisco Massa, Wan-Yen Lo, and Ross Girshick. Detectron2, 2019. ",
+        "bbox": [
+            171,
+            367,
+            825,
+            383
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "Zhirong Wu, Yuanjun Xiong, Stella X Yu, and Dahua Lin. Unsupervised feature learning via nonparametric instance discrimination. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pp. 3733–3742, 2018. ",
+        "bbox": [
+            176,
+            390,
+            823,
+            435
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "Jianxiong Xiao, James Hays, Krista A Ehinger, Aude Oliva, and Antonio Torralba. Sun database: Large-scale scene recognition from abbey to zoo. In 2010 IEEE computer society conference on computer vision and pattern recognition, pp. 3485–3492. IEEE, 2010. ",
+        "bbox": [
+            173,
+            443,
+            823,
+            486
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "Jianwei Yang, Devi Parikh, and Dhruv Batra. Joint unsupervised learning of deep representations and image clusters. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pp. 5147–5156, 2016. ",
+        "bbox": [
+            174,
+            494,
+            823,
+            537
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "Sergey Zagoruyko and Nikos Komodakis. Paying more attention to attention: Improving the performance of convolutional neural networks via attention transfer. arXiv preprint arXiv:1612.03928, 2016. ",
+        "bbox": [
+            173,
+            597,
+            826,
+            640
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "Xiaohua Zhai, Avital Oliver, Alexander Kolesnikov, and Lucas Beyer. S4l: Self-supervised semisupervised learning. In Proceedings of the IEEE international conference on computer vision, pp. 1476–1485, 2019. ",
+        "bbox": [
+            173,
+            648,
+            826,
+            693
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "Liheng Zhang, Guo-Jun Qi, Liqiang Wang, and Jiebo Luo. Aet vs. aed: Unsupervised representation learning by auto-encoding transformations rather than data. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pp. 2547–2555, 2019. ",
+        "bbox": [
+            173,
+            700,
+            823,
+            744
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "Richard Zhang, Phillip Isola, and Alexei A Efros. Colorful image colorization. In European conference on computer vision, pp. 649–666. Springer, 2016. ",
+        "bbox": [
+            171,
+            752,
+            823,
+            782
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "Xiao Zhang, Zhiyuan Fang, Yandong Wen, Zhifeng Li, and Yu Qiao. Range loss for deep face recognition with long-tailed training data. In Proceedings of the IEEE International Conference on Computer Vision, pp. 5409–5418, 2017. ",
+        "bbox": [
+            176,
+            790,
+            823,
+            833
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "A APPENDIX ",
+        "text_level": 1,
+        "bbox": [
+            176,
+            102,
+            299,
+            118
+        ],
+        "page_idx": 12
+    },
+    {
+        "type": "text",
+        "text": "We discuss more details and different hyperparameters for SEED during distillation. ",
+        "bbox": [
+            174,
+            132,
+            725,
+            148
+        ],
+        "page_idx": 12
+    },
+    {
+        "type": "text",
+        "text": "A.1 PSEUDO-IMPLEMENTATIONS ",
+        "bbox": [
+            174,
+            165,
+            418,
+            179
+        ],
+        "page_idx": 12
+    },
+    {
+        "type": "text",
+        "text": "We provide pseudo-code of the SEED distillation in PyTorch Paszke et al. (2019) style: ",
+        "bbox": [
+            171,
+            190,
+            746,
+            205
+        ],
+        "page_idx": 12
+    },
+    {
+        "type": "text",
+        "text": "‘‘‘Q: maintaining queue of previous representations: (N X D)   \n2 T: Cumbersome encoder as Teacher.   \n3 S: Target encoder as Student.   \n4 temp_T, temp_S: temperatures of the Teacher & Student.   \n11   \n6   \n7 # activate evaluation mode for Teacher to freeze BN and updation.   \n8 T.eval()   \n9   \n10 for images in enumerate(loader): # Enumerate single crop-view   \n11   \n12 # augment image to get one identical view   \n13 images $=$ aug(images)   \n14   \n15 # Batch-size   \n16 $\\begin{array} { r l } { \\mathrm { B } } & { { } = } \\end{array}$ images.shape[0]   \n17   \n18 # extract embedding from S: 1 X D   \n19 $\\mathrm { ~ \\cal ~ X ~ \\_ ~ S ~ ~ } =$ S(images)   \n20 $\\mathrm { ~ \\cal ~ X ~ \\_ ~ S ~ ~ } =$ torch.norm(X_S, $\\mathrm { p } { = } 2$ , dim $^ { = 1 }$ )   \n21   \n22 # use the gradient-free mode   \n23 with torch.no_grad():   \n24 X_T $=$ T(image) # embedding from T: 1 X D   \n25 $\\mathrm { ~ X ~ \\_ ~ T ~ } =$ torch.norm(X_T, $\\mathrm { p } { = } 2$ , dim $^ { = 1 }$ )   \n26   \n27 # insert the current batch embedding from T   \n28 enqueue(Q, X_T)   \n29   \n30 # probability scores distribution for T, S: B X (N + 1)   \n31 S_Dist $=$ torch.einsum(’bd, dn -> bn’, [X_S], Q.t().clone().detach())   \n32 T_Dist $=$ torch.einsum(’bd, dn -> bn’, [X_T], Q.t().clone().detach())   \n33   \n34 # Apply temperatures for soft-labels   \n35 S_Dist / $=$ temp_S   \n36 T_Dist $=$ SoftMax(T_Dist/temp_T, dim $^ { = 1 }$ )   \n37   \n38 # loss computation, use log_softmax for stable computation   \n39 loss $=$ -torch.mul(T_Dist, Log_SoftMax(S_Dist, dim $^ { 1 = 1 }$ )).sum()/B   \n40   \n41 # update the random sample queue   \n42 dequeue(Q, B) # pop-out earliest B instances   \n43   \n44 # SGD updation   \n45 loss.backward()   \n46 update(S.params) ",
+        "bbox": [
+            150,
+            218,
+            810,
+            815
+        ],
+        "page_idx": 12
+    },
+    {
+        "type": "text",
+        "text": "A.1.1 DATA AUGMENTATIONS ",
+        "text_level": 1,
+        "bbox": [
+            176,
+            103,
+            398,
+            118
+        ],
+        "page_idx": 13
+    },
+    {
+        "type": "text",
+        "text": "Both our teacher pre-training and distillation adopt the data augmentations as follows: ",
+        "bbox": [
+            176,
+            127,
+            735,
+            142
+        ],
+        "page_idx": 13
+    },
+    {
+        "type": "text",
+        "text": "Random Resized Crop: The image is randomly resized with a scale of {0.2, 1.0}, then cropped to the size of $2 2 4 \\times 2 2 4$ . ",
+        "bbox": [
+            222,
+            152,
+            823,
+            181
+        ],
+        "page_idx": 13
+    },
+    {
+        "type": "text",
+        "text": "Random Color Jittering: with brightness to be {0.4, 0.4, 0.4, 0.1} with probability at 0.8. ",
+        "bbox": [
+            217,
+            184,
+            810,
+            200
+        ],
+        "page_idx": 13
+    },
+    {
+        "type": "text",
+        "text": "Random Gray Scale transformation: with probability at 0.2. ",
+        "bbox": [
+            223,
+            204,
+            617,
+            218
+        ],
+        "page_idx": 13
+    },
+    {
+        "type": "text",
+        "text": "Random Gaussian Blur transformation: with $\\sigma = \\{ 0 . 1 , 0 . 2 \\}$ and probability at 0.5. ",
+        "bbox": [
+            228,
+            222,
+            764,
+            236
+        ],
+        "page_idx": 13
+    },
+    {
+        "type": "text",
+        "text": "Horizontal Flip: Horizontal flip is applied with probability at 0.5 ",
+        "bbox": [
+            230,
+            239,
+            648,
+            255
+        ],
+        "page_idx": 13
+    },
+    {
+        "type": "text",
+        "text": "A.1.2 PRE-TRAINING AND DISTILLATION ON MOBILENET AND EFFICIENTNET ",
+        "text_level": 1,
+        "bbox": [
+            178,
+            268,
+            733,
+            284
+        ],
+        "page_idx": 13
+    },
+    {
+        "type": "text",
+        "text": "MobileNet (Howard et al., 2017) and EfficientNet (Tan & Le, 2019) have been considered as the smaller counterparts with larger models, i.e., ResNet-50 (with supervised training, EfficientNetB0 hits $7 7 . 2 \\%$ Top-1 Acc., and MobileNet-V3-large reaches $7 2 . 2 \\%$ on ImageNet testing split). Nevertheless, un-matched performances are observed in the task of self-supervised contrastive pretraining: i.e., Self-Supervised Learning (MoCo-V2) on MobileNet-V3 only yields $3 6 . 3 \\%$ Top-1 Acc. on ImageNet. We conjecture that several reasons might lead to this dilemma: ",
+        "bbox": [
+            174,
+            292,
+            826,
+            377
+        ],
+        "page_idx": 13
+    },
+    {
+        "type": "text",
+        "text": "1. The inability of models with less parameters for handling large volume of categories and data, which exists also in other domains, i.e., face recognition (Guo et al., 2016; Zhang et al., 2017).   \n2. Less possibility for optimum parameters to be chosen when transferring to downstream tasks: models with more parameters after pre-training might produce a plenty cornucopia of optimum parameters for fine-tuning. ",
+        "bbox": [
+            210,
+            387,
+            826,
+            476
+        ],
+        "page_idx": 13
+    },
+    {
+        "type": "text",
+        "text": "To narrow the dramatic performance gap between smaller architectures using contrastive SSL with the larger, we explore with architectural manipulations and training hyper-parameters. In specific, we find that by adding a deeper projection head largely improves the representation quality, a.k.a., better performances on linear evaluation. We experiment with adding one additional linear projection head on the top of convolutional backbones. ",
+        "bbox": [
+            174,
+            486,
+            825,
+            556
+        ],
+        "page_idx": 13
+    },
+    {
+        "type": "text",
+        "text": "Similarly, we also expand the MLP projection head on EfficientNet-b0. Though recent work shows that fine-tuning from a middle layer of the projection head can produce a largely different result (Chen et al., 2020b), we consistently just use the representations from convolutional trunk without adding extra layers during the phase of linear evaluation. As shown in Table 6, pre-training with a deeper projection head dramatically helps the improvement on linear evaluations, adding $17 \\%$ Top-1. Acc. for Mobile-v3-large, and we report the improved baselines in the main paper (see the first row in Table 1 of the main paper). We keep most of the hyper-parameters as the distillation on ResNet except reducing the weight-decay of them to 1e-5, following (Tan & Le, 2019; Sandler et al., 2018). ",
+        "bbox": [
+            173,
+            563,
+            825,
+            675
+        ],
+        "page_idx": 13
+    },
+    {
+        "type": "table",
+        "img_path": "images/637ac61f5f9e5741399531cd7141a325efafa86d41a01e1c7e857e7176d2e6b6.jpg",
+        "table_caption": [
+            "Table 6: Linear evaluations on ImageNet of EfficientNet and MobileNet pre-trained using MoCo-v2. A deeper projection head largely boosts the linear evaluation performances on smaller architectures. "
+        ],
+        "table_footnote": [],
+        "table_body": "<table><tr><td>Model</td><td>Deeper MLPs</td><td>Top-1 Acc. Top-5 Acc.</td><td></td></tr><tr><td>EfficientNet-b0</td><td></td><td>39.1</td><td>64.6</td></tr><tr><td>EfficientNet-b0</td><td>×</td><td>42.2</td><td>68.5</td></tr><tr><td>Mobile-v3-large</td><td>X</td><td>19.0</td><td>41.3</td></tr><tr><td>Mobile-v3-large</td><td>√</td><td>36.3</td><td>62.2</td></tr></table>",
+        "bbox": [
+            300,
+            731,
+            692,
+            815
+        ],
+        "page_idx": 13
+    },
+    {
+        "type": "text",
+        "text": "A.2 ADDITIONAL DETAILS OF EVALUATIONS ",
+        "text_level": 1,
+        "bbox": [
+            174,
+            848,
+            501,
+            863
+        ],
+        "page_idx": 13
+    },
+    {
+        "type": "text",
+        "text": "We list additional details regarding our evaluation experiments in this section. ",
+        "bbox": [
+            173,
+            873,
+            681,
+            888
+        ],
+        "page_idx": 13
+    },
+    {
+        "type": "text",
+        "text": "ImageNet-1k Semi-Supervised Linear Evaluation. Following Zhai et al. (2019); Chen et al. (2020a), we train the FC layers on the basis of our student encoder after distillation using a fraction of labeled ImageNet-1k dataset ( $1 \\%$ and $10 \\%$ ), and evaluate it on the whole test split. The fraction of labeled dataset is constructed in a class-balanced way, with roughly 12 and 128 images per class∗. We use SGD optimizer and set initial learning rate to be 30 with a multiplier $=$ BatchSize/256 without weight decaying for 100 epochs. We use the step-wise scheduler for the learning rate updating with 5 warm-up epochs, and the learning rate is reduced by 10 at 60 and 80 epochs. On smaller architectures like EfficientNet and MobileNet, we reduce the initial learning rate to 3. During training, the image is center-cropped to the size of $2 2 4 \\times 2 2 4$ with just Random Horizontal Flip as the data augmentation. For testing, we first resize the image to $2 5 6 \\times 2 5 6$ and use the center cropped $2 2 4 \\times 2 2 4$ for pre-processing. In Table 8, we show the distillation results on a larger encoder (ResNet-550) when using different teacher networks. ",
+        "bbox": [
+            171,
+            895,
+            828,
+            924
+        ],
+        "page_idx": 13
+    },
+    {
+        "type": "table",
+        "img_path": "images/cf3071272dac29e76307ac8ba82a1a6899f7e85f6571ad9cb66401dc71b61d17.jpg",
+        "table_caption": [
+            "Table 7: Before and after distillation Top-1/5 test accuracy $( \\% )$ on ImageNet of EfficientNet-b0 and MobileNetlarge without deeper MLPs. "
+        ],
+        "table_footnote": [],
+        "table_body": "<table><tr><td>Student</td><td>Teacher</td><td>Top-1</td><td>Top-5</td></tr><tr><td rowspan=\"3\">EfficientNet-b0</td><td>X</td><td>39.1</td><td>64.6</td></tr><tr><td>ResNet-50</td><td>59.2</td><td>81.2</td></tr><tr><td>ResNet-101</td><td>62.8</td><td>84.7</td></tr><tr><td></td><td>ResNet-152</td><td>63.3</td><td>85.6</td></tr><tr><td rowspan=\"3\">MobileNet-v3</td><td>X ResNet-50</td><td>19.0 50.9</td><td>41.3 77.7</td></tr><tr><td>ResNet-101</td><td>57.6</td><td>82.6</td></tr><tr><td>ResNet-152</td><td>58.3</td><td>82.9</td></tr></table>",
+        "bbox": [
+            323,
+            138,
+            665,
+            294
+        ],
+        "page_idx": 14
+    },
+    {
+        "type": "table",
+        "img_path": "images/e10d7d8d9104543baae25184077eb1a0f410866ba1b0404e862ec467a0ea7815.jpg",
+        "table_caption": [
+            "Table 8: ImageNet-1k test accuracy $( \\% )$ under KNN and linear classification on ResNet-50 encoder with deeper, MoCo-V2/SWAV pre-trained teacher architectures. $\\pmb { \\chi }$ denotes MoCo-V2 self-supervised learning baselines before distillation. \\* indicates using a stronger teacher encoder pre-trained by SWAV with additional small-patches during distillation. "
+        ],
+        "table_footnote": [],
+        "table_body": "<table><tr><td></td><td>Stud.</td><td></td><td>ResNet-50</td><td></td></tr><tr><td>Teac.</td><td>Epoch</td><td>KNN</td><td>Top-1</td><td>Top-5</td></tr><tr><td>X ResNet-50</td><td>200</td><td>46.1 46.1</td><td>67.4 67.5</td><td>87.8 87.8</td></tr><tr><td>△ ResNet-101</td><td>200</td><td>+0.0 52.3</td><td>+0.1</td><td>+0.0</td></tr><tr><td>△</td><td>200</td><td>+6.2</td><td>69.1 +1.7</td><td>88.7 +0.9</td></tr><tr><td>ResNet-152 △</td><td>200</td><td>53.2 +7.1</td><td>70.4 +3.0</td><td>90.5 +2.7</td></tr><tr><td>ResNet-50×2* △</td><td>800</td><td>59.0 +12.9</td><td>74.3 +6.9</td><td>92.2 +4.4</td></tr></table>",
+        "bbox": [
+            315,
+            381,
+            681,
+            550
+        ],
+        "page_idx": 14
+    },
+    {
+        "type": "text",
+        "text": "",
+        "bbox": [
+            173,
+            587,
+            826,
+            727
+        ],
+        "page_idx": 14
+    },
+    {
+        "type": "text",
+        "text": "Transfer Learning. We test the transferability of the representations learned from self-supervised distillation by conducting the linear evaluations using offline features on several other datasets. Specifically, a single layer logistic classifier is trained following (Chen et al., 2020a; Grill et al., 2020) using SGD optimizer without weight decay and momentum parameter at 0.9. We use CIFAR-10, CIFAR-100 (Krizhevsky et al., 2009) and SUN-397 (Xiao et al., 2010) as our testing beds. ",
+        "bbox": [
+            176,
+            733,
+            825,
+            804
+        ],
+        "page_idx": 14
+    },
+    {
+        "type": "text",
+        "text": "CIFAR: As the size for CIFAR dataset is $3 2 \\times 3 2$ , we resize all images to $2 2 4 \\times 2 2 4$ pixels along the shorter side using bicubic resampling method, followed by a center crop operation. We set the learning rate at 1e-3 constantly and train it for 120 epochs. The hyper-parameters are searched using 10 fold cross-validation on the train split and report its final top-1 accuracy on the test split. ",
+        "bbox": [
+            204,
+            815,
+            825,
+            886
+        ],
+        "page_idx": 14
+    },
+    {
+        "type": "table",
+        "img_path": "images/5c3f7a898d26a2d2fe5fcf5d40e410578494648f20defdcd18389a3531ae57cb.jpg",
+        "table_caption": [
+            "Table 9: Object detection and instance segmentation fine-tuned on VOC07: bounding-box AP $( \\mathsf { A P } ^ { b b } )$ ) and mask AP $( \\mathbf { A } \\mathbf { P } ^ { m k } )$ ) evaluated on VOC07-val. The first row shows the baseline from MoCo-v2 backbones without distillation. "
+        ],
+        "table_footnote": [],
+        "table_body": "<table><tr><td rowspan=\"2\">Student</td><td rowspan=\"2\">Teacher</td><td colspan=\"3\">VOC Object Detection</td></tr><tr><td>APbb</td><td>AP</td><td>AP</td></tr><tr><td rowspan=\"3\">ResNet-34</td><td>X</td><td>53.6</td><td>79.1</td><td>58.7</td></tr><tr><td>ResNet-50</td><td>53.7 (+0.1)</td><td>79.4 (+0.3)</td><td>59.2 (+0.5)</td></tr><tr><td>ResNet-101 ResNet-152</td><td>54.1 (+0.5) 54.4 (+0.8)</td><td>79.8 (+0.7) 80.1 (+1.0)</td><td>59.1 (+0.4) ） 59.9 (+1.2)</td></tr><tr><td rowspan=\"2\">ResNet-50</td><td>X</td><td>57.0</td><td>82.4</td><td>63.6</td></tr><tr><td>ResNet-50 ResNet-101</td><td>57.0 (+0.0) 57.1 (+0.1)</td><td>82.4 (+0.0) 82.8 (+0.4)</td><td>63.6 (+0.0) 63.8 (+0.2)</td></tr></table>",
+        "bbox": [
+            303,
+            152,
+            687,
+            329
+        ],
+        "page_idx": 15
+    },
+    {
+        "type": "table",
+        "img_path": "images/6d3d5a9fb7c5054b18a1d0cede779a4500b23d1f19747e3aeb70e1db8094b73e.jpg",
+        "table_caption": [
+            "Table 10: Object detection and instance segmentation fine-tuned on COCO: bounding-box AP $( \\mathbf { A P } ^ { b b } )$ and mask AP $( \\mathbf { A } \\mathbf { P } ^ { m k } )$ evaluated on COCO-val2017. The first several rows show the baselines from unsupervised backbones without distillation. "
+        ],
+        "table_footnote": [],
+        "table_body": "<table><tr><td rowspan=\"2\">Student</td><td rowspan=\"2\">Teacher</td><td colspan=\"3\">Object Detection</td><td colspan=\"3\">Instance Segmentation</td></tr><tr><td>Apbb</td><td>AP</td><td>APP</td><td>Apmk</td><td>AP</td><td>AP</td></tr><tr><td rowspan=\"4\">ResNet34</td><td>X</td><td>38.1</td><td>56.8</td><td>40.7</td><td>33.0</td><td>53.2</td><td>35.3</td></tr><tr><td>ResNet50</td><td>38.4 (+0.3)</td><td>57.0 (+0.2)</td><td>41.0 (+0.3)</td><td>33.3 (+0.3)</td><td>53.6 (+0.4)</td><td>35.4 (+0.1)</td></tr><tr><td>ResNet101</td><td>38.5 (+0.4)</td><td>57.3 (+0.5)</td><td>41.4 (+0.7)</td><td>33.6 (+0.6)</td><td>54.1 (+0.9)</td><td>35.6 (+0.3)</td></tr><tr><td>ResNet152</td><td>38.4 (+0.3)</td><td>57.0 (+0.2)</td><td>41.0 (+0.3)</td><td>33.3 (+0.3)</td><td>53.7 (+0.5)</td><td>35.3 (+0.0)</td></tr></table>",
+        "bbox": [
+            178,
+            402,
+            820,
+            513
+        ],
+        "page_idx": 15
+    },
+    {
+        "type": "text",
+        "text": "SUN-397: We further extend our transferring evaluation to the scene dataset SUN-397 for a more diverse testing. The official dataset specifies 10 different train/test splits, with each contains 50 images per category covering 397 different scenes. We follow (Chen et al., 2020a; Grill et al., 2020) and use the first train/test split. For the validation set, we randomly pick 10 images (yielding $20 \\%$ of the dataset), with identical optimizer parameters as CIFAR. ",
+        "bbox": [
+            204,
+            547,
+            825,
+            617
+        ],
+        "page_idx": 15
+    },
+    {
+        "type": "text",
+        "text": "Object Detection and Instance Segmentation. As indicated by (He et al., 2020), features produced by self-supervised pre-training have divergent distributions in downstream tasks, thus resulting the supervised pre-training picked hyper-parameters not applicable. To relieve this, He et al. (2020) uses feature normalization during the fine-tuning phase and train the BN layers. Different from previous transferring and linear evaluations where we exploit only offline features, model for detection and segmentation is trained with all parameters tuned. For this reason, annotations on COCO for segmentation gives much higher influence for the backbone model than the VOC dataset (see Table 9), and gives an offset to the pre-training difference (see Table 10). Thus, this makes the performance boosting by pre-training less obvious, and leads to trivial AP differences before and after distillation. ",
+        "bbox": [
+            174,
+            628,
+            825,
+            755
+        ],
+        "page_idx": 15
+    },
+    {
+        "type": "text",
+        "text": "Object Detection on PASCAL VOC-07: We train a C4 (He et al., 2017) based Faster R-CNN (Ren et al., 2015) as the detector with different ResNet architectures (ResNet-18, ResNet-34 and ResNet-50) for evaluating the transferability of features for object detection tasks. We use Detectron2 (Wu et al., 2019) for the implementations. We train our detector for 48k iterations with a batch size of 32 (8 images per GPU). The base learning rate is set to 0.01 with 200 warm-up iterations. We set the scale of images for training as [400, 800] and 800 at inference. ",
+        "bbox": [
+            205,
+            765,
+            825,
+            849
+        ],
+        "page_idx": 15
+    },
+    {
+        "type": "text",
+        "text": "Object Detection and Segmentation on COCO: We use Mask R-CNN (He et al., 2017) with the C4 backbone for the object detection and instance segmentation task on COCO dataset, with $2 \\times$ schedule. Similar to the VOC detection, we tune the BN layers and all parameters. The model is trained for $1 8 0 \\mathrm { k }$ iterations with initial learning rate set to 0.02. We set the scale of images for training as [600, 800] and 800 at inference. ",
+        "bbox": [
+            207,
+            853,
+            823,
+            924
+        ],
+        "page_idx": 15
+    },
+    {
+        "type": "table",
+        "img_path": "images/72a4faa1cf7e3c6038d27d854e17468ebd5461becf2efda07c70ea9634eb9831.jpg",
+        "table_caption": [
+            "Table 11: Linear evaluations on ImageNet of ResNet-18 after distillation from the SWAV pre-trained ResNet-50 using either single view, cross-views, or small patch views. "
+        ],
+        "table_footnote": [],
+        "table_body": "<table><tr><td>Method</td><td>Multi-View(s)</td><td>Top-1 Acc.</td><td>Top-5 Acc.</td></tr><tr><td>Identical-View</td><td>1×224</td><td>61.7</td><td>84.2</td></tr><tr><td>Cross-Views</td><td>2×224</td><td>58.2</td><td>81.7</td></tr><tr><td>Multi-Crops + Cross-Views</td><td>1×224 + 6×96×96</td><td>61.9</td><td>84.4</td></tr><tr><td>Multi-Crops + Identical-View </td><td>1×224 +6×96×96</td><td>62.6</td><td>84.8</td></tr></table>",
+        "bbox": [
+            236,
+            147,
+            758,
+            231
+        ],
+        "page_idx": 16
+    },
+    {
+        "type": "image",
+        "img_path": "images/0863ba5174cfabf1f7a4f95ce0b77df7bdbf71b58813b9c855df14a8e571182a.jpg",
+        "image_caption": [
+            "Figure 6: We experiment with different strategies of using views during distillation, which include: (a). Identical view for distillation. (b). Cross view distillation. (c). Large-small cross view distillation. (d). Large-small identical view distillation. "
+        ],
+        "image_footnote": [],
+        "bbox": [
+            194,
+            255,
+            810,
+            363
+        ],
+        "page_idx": 16
+    },
+    {
+        "type": "text",
+        "text": "A.3 SINGLE CROP V.S. MULTI-CROPS VIEW(S) FOR DISTILLATION",
+        "text_level": 1,
+        "bbox": [
+            176,
+            440,
+            648,
+            455
+        ],
+        "page_idx": 16
+    },
+    {
+        "type": "text",
+        "text": "In contrary with most contrastive SSL methods where two different augmented views of an image are utilized as the positive samples (see Figure 6-a), SEED uses an identical view for each image (see Figure 6-b) during distillation and yields better performances, as is shown in Table. 11. In addition, we have also experimented with two strategies of using small patches. To be specific, we follow the set-up in SWAV (Caron et al., 2020), that 6 small patches of the size $9 6 \\times 9 6$ are sampled at the scale of (0.05, 0.14). Then, we apply the same augmentations as introduced previously as data pre-processing. Figure. 6-c shows the way that is similar in SWAV for small-patch learning, where both large and 6 small patches are fed into the student encoder, with the learning target $( \\mathbf { z } ^ { T } )$ to be the embedding of large view from the teacher encoder. Figure. 6-d is the strategy we use during distillation, that both views are fed into student and teacher to produce the embeddings for small-views $( \\mathbf { z } _ { s } ^ { S } , \\mathbf { z } _ { s } ^ { T } )$ and large views $( \\mathbf { z } _ { l } ^ { S } , \\mathbf { z } _ { l } ^ { T } )$ . Based on that, the distillation is formulated separately on the small and large views. Notably, we maintain two independent queues for storing historical data samples for the large and small views. ",
+        "bbox": [
+            173,
+            467,
+            826,
+            647
+        ],
+        "page_idx": 16
+    },
+    {
+        "type": "text",
+        "text": "A.4 STRATEGIES FOR OTHER DISTILLATION METHODS ",
+        "text_level": 1,
+        "bbox": [
+            173,
+            666,
+            566,
+            680
+        ],
+        "page_idx": 16
+    },
+    {
+        "type": "text",
+        "text": "We compare the effect of distillation using different strategies with SEED. ",
+        "bbox": [
+            176,
+            693,
+            661,
+            707
+        ],
+        "page_idx": 16
+    },
+    {
+        "type": "text",
+        "text": "$\\mathbf { \\xi } _ { l 2 }$ -Distance: We train the student encoder by minimizing the squared $l 2$ -distance of representations from student $( \\mathbf { z } _ { i } ^ { S } )$ and teacher $( \\mathbf { z } _ { i } ^ { T } )$ for an identical view $\\mathbf { x } _ { i }$ . ",
+        "bbox": [
+            173,
+            714,
+            821,
+            742
+        ],
+        "page_idx": 16
+    },
+    {
+        "type": "text",
+        "text": "$\\pmb { K }$ -Means: We experiment with the $K$ -Means clustering method to retrieve pseudo class labels for distillation. Specifically, we first extract offline image features using the $S S L$ pre-trained Teacher network without any image augmentations. Based on this, we conduct our $K$ -Means clustering with $4 \\mathrm { k }$ and $1 6 \\mathrm { k }$ unique centroids. Then the final centroids are used to produce pseudo labels for unlabelled instances. With that, we carry out the distillation by training the model on a classification task using the produced labels as the ground-truth. To avoid trivial solutions that the majority of images are assigned to a few clusters, we sample images based on a uniform distribution over pseudo-labels as clustering proceeds. We observe very close results when adjusting numbers of centroids. ",
+        "bbox": [
+            173,
+            750,
+            825,
+            861
+        ],
+        "page_idx": 16
+    },
+    {
+        "type": "text",
+        "text": "Online-Clustering: With $K$ -Means for pseudo-label generation training, it does not lead to satisfying results $( 5 1 . 0 \\%$ on ResNet-18 with ResNet-50 as Teacher) as instances might have not been accurately categorized by limited frozen centroids. Similar to (Caron et al., 2018; Li et al., 2020), we resort to the “in-batch” and dynamical clustering to substitute the frozen $K$ -Means method. We conduct ",
+        "bbox": [
+            174,
+            867,
+            823,
+            924
+        ],
+        "page_idx": 16
+    },
+    {
+        "type": "text",
+        "text": "$K$ -Means clustering within a batch and continuously update the centroid based on the teacher feature representation as distillation goes on. This alleviates the above problems and yields a substantial performance improvement on ResNet-18 to $5 6 . 4 \\%$ . ",
+        "bbox": [
+            173,
+            103,
+            823,
+            146
+        ],
+        "page_idx": 17
+    },
+    {
+        "type": "text",
+        "text": "Binary Contrastive Loss: We resort to CRD (Tian et al., 2019a) and adopt an info-NCE loss-alike training objective in unsupervised distillation tasks. Specifically, we treat representation features from Teacher and Student for instance $\\mathbf { x } _ { i }$ as positive pairs, and random instances from $\\mathbf { D }$ as negative samples: ",
+        "bbox": [
+            173,
+            151,
+            825,
+            209
+        ],
+        "page_idx": 17
+    },
+    {
+        "type": "equation",
+        "img_path": "images/7efa8646a999cf9b57dbdc3eab933940b7c76683e63c8311e334af00b5e6c4a4.jpg",
+        "text": "$$\n\\hat { \\theta } _ { S } = \\underset { \\theta _ { S } } { \\arg \\operatorname* { m i n } } \\sum _ { i } ^ { N } \\log h ( \\mathbf { z } _ { i } ^ { S } , \\mathbf { z } _ { i } ^ { T } ) + K \\cdot [ \\log h ( 1 - h ( \\mathbf { z } _ { i } ^ { S } , \\mathbf { d } _ { j } ^ { T } ) ) ] ,\n$$",
+        "text_format": "latex",
+        "bbox": [
+            281,
+            209,
+            697,
+            252
+        ],
+        "page_idx": 17
+    },
+    {
+        "type": "text",
+        "text": "where $\\mathbf { d } _ { j } ^ { T } \\in \\mathbf { D } , h ( \\cdot )$ is any family of functions that satisfy $h$ : $\\{ { \\bf z } , { \\bf d } \\}  [ 0 , 1 ]$ , e.g., cosine similarity. ",
+        "bbox": [
+            173,
+            260,
+            825,
+            276
+        ],
+        "page_idx": 17
+    },
+    {
+        "type": "text",
+        "text": "A.5 DISCUSSIONS ON SEED ",
+        "text_level": 1,
+        "bbox": [
+            174,
+            295,
+            388,
+            311
+        ],
+        "page_idx": 17
+    },
+    {
+        "type": "text",
+        "text": "Our proposed learning objective for SEED is composed of two goals, that is to align the encoding $\\mathbf { z } ^ { S }$ by the student model with $\\mathbf { z } ^ { T }$ produced by the teacher model; meanwhile, $\\mathbf { z } ^ { S }$ also softly contrasts with random samples maintained in the $\\mathbf { D }$ . This can be formulated more directly as minimizing the $l 2$ distance of ${ \\mathbf z } ^ { T } , { \\mathbf z } ^ { S }$ , together with the cross-entropy computed using $\\mathbf { D }$ : ",
+        "bbox": [
+            173,
+            321,
+            825,
+            381
+        ],
+        "page_idx": 17
+    },
+    {
+        "type": "equation",
+        "img_path": "images/5e4bf14a4554833a0ba76bc465a50d4fc5b25018566e53300c808cf0763a0ad4.jpg",
+        "text": "$$\n\\begin{array} { l } { \\displaystyle \\mathcal { L } = \\frac { 1 } { N } \\sum _ { i } ^ { N } \\bigg \\{ \\lambda _ { a } \\cdot \\bigg \\vert \\bigg \\vert \\mathbf { z } _ { i } ^ { T } - \\mathbf { z } _ { i } ^ { S } \\bigg \\vert \\bigg \\vert _ { 2 } - \\lambda _ { b } \\cdot \\mathbf { p } ^ { T } ( \\mathbf { x } _ { i } ; \\boldsymbol { \\theta } _ { T } , \\mathbf { D } ) \\cdot \\log \\mathbf { p } ^ { S } ( \\mathbf { x } _ { i } ; \\boldsymbol { \\theta } _ { S } , \\mathbf { D } ) \\bigg \\} } \\\\ { \\displaystyle \\quad = \\sum _ { i } ^ { N } \\bigg \\{ - \\lambda _ { a } \\cdot \\mathbf { z } _ { i } ^ { T } \\cdot \\mathbf { z } _ { i } ^ { S } - \\lambda _ { b } \\cdot \\sum _ { j } ^ { K } \\frac { \\exp ( \\mathbf { z } _ { i } ^ { T } \\cdot \\mathbf { d } _ { j } / \\tau ^ { T } ) } { \\sum _ { \\mathbf { d } \\setminus \\mathbf { D } } \\exp ( \\mathbf { z } _ { i } ^ { T } \\cdot \\mathbf { d } / \\tau ^ { T } ) } \\cdot \\log \\frac { \\exp ( \\mathbf { z } _ { i } ^ { S } \\cdot \\mathbf { d } _ { j } / \\tau ^ { S } ) } { \\sum _ { \\mathbf { d } \\setminus \\mathbf { D } } \\exp ( \\mathbf { z } _ { i } ^ { S } \\cdot \\mathbf { d } / \\tau ^ { S } ) } \\bigg \\} , } \\end{array}\n$$",
+        "text_format": "latex",
+        "bbox": [
+            186,
+            388,
+            789,
+            478
+        ],
+        "page_idx": 17
+    },
+    {
+        "type": "text",
+        "text": "Directly optimizing Eq. 7 can lead to apparent difficulty in searching optimal hyper-parameters $( \\lambda _ { a } , \\lambda _ { b } , \\bar { \\tau } ^ { T }$ and $\\tau ^ { S }$ ). Our proposed objective on $\\mathbf { D } ^ { + }$ indeed is an approximated upper-bound of the above objectiveness however much simplified: ",
+        "bbox": [
+            176,
+            484,
+            823,
+            529
+        ],
+        "page_idx": 17
+    },
+    {
+        "type": "equation",
+        "img_path": "images/23a09cdc1125e61dd98e8974b713166f88a62cc74247c152a34211fa3a3d3ea6.jpg",
+        "text": "$$\n\\begin{array} { r l } & { \\mathcal { L } _ { \\mathrm { S E E D } } = \\displaystyle \\frac { 1 } { N } \\sum _ { i } ^ { N } - \\mathbf { p } ^ { T } ( \\mathbf { x } _ { i } ; \\boldsymbol { \\theta } _ { T } , \\mathbf { D } ^ { + } ) \\cdot \\log \\mathbf { p } ^ { S } ( \\mathbf { x } _ { i } ; \\boldsymbol { \\theta } _ { S } , \\mathbf { D } ^ { + } ) } \\\\ & { \\qquad = \\displaystyle \\sum _ { i } ^ { N } \\sum _ { j } ^ { K + 1 } - \\underbrace { \\frac { \\exp \\left( \\mathbf { z } _ { i } ^ { T } \\cdot \\mathbf { d } _ { j } / \\tau ^ { T } \\right) } { \\sum _ { \\mathbf { d } \\sim \\mathbf { D } ^ { + } } \\exp \\left( \\mathbf { z } _ { i } ^ { T } \\cdot \\mathbf { d } / \\tau ^ { T } \\right) } } _ { \\mathbf { w } _ { j } ^ { i } } \\cdot \\log \\frac { \\exp \\left( \\mathbf { z } _ { i } ^ { S } \\cdot \\mathbf { d } _ { j } / \\tau ^ { S } \\right) } { \\sum _ { \\mathbf { d } \\sim \\mathbf { D } ^ { + } } \\exp \\left( \\mathbf { z } _ { i } ^ { S } \\cdot \\mathbf { d } / \\tau ^ { S } \\right) } , } \\end{array}\n$$",
+        "text_format": "latex",
+        "bbox": [
+            240,
+            536,
+            758,
+            645
+        ],
+        "page_idx": 17
+    },
+    {
+        "type": "text",
+        "text": "where we let $\\mathbf { w } _ { j } ^ { i }$ denote the weighting term regulated under $\\tau ^ { T }$ . Since the $( K + 1 )$ th element in $\\mathbf { D } ^ { + }$ is our supplemented vector ${ \\mathbf z } _ { i } ^ { T }$ , the above objective can be expanded into: ",
+        "bbox": [
+            174,
+            654,
+            825,
+            686
+        ],
+        "page_idx": 17
+    },
+    {
+        "type": "equation",
+        "img_path": "images/7d21dd51331e153adeabf8831905dcc223c4a73c9d250b37724e457be450eb78.jpg",
+        "text": "$$\n\\begin{array} { r l } & { \\mathcal { L } _ { \\mathrm { S E E D } } = \\displaystyle \\frac { 1 } { N } \\sum _ { i } ^ { N } \\Big \\{ { \\bf w } _ { K + 1 } ^ { i } \\cdot \\big ( - { \\bf z } _ { i } ^ { S } \\cdot { \\bf z } _ { i } ^ { T } / \\tau ^ { S } + \\log \\sum _ { \\bf d \\sim D ^ { + } } \\exp ( { \\bf z } _ { i } ^ { S } \\cdot { \\bf d } / \\tau ^ { S } ) \\big ) } \\\\ & { \\qquad + \\displaystyle \\sum _ { j = 1 } ^ { K } { \\bf w } _ { j } ^ { i } \\cdot \\big ( - { \\bf z } _ { i } ^ { S } \\cdot { \\bf d } _ { j } / \\tau ^ { S } + \\log \\sum _ { \\bf d \\sim D ^ { + } } \\exp ( { \\bf z } _ { i } ^ { S } \\cdot { \\bf d } / \\tau ^ { S } ) \\big ) \\Big \\} } \\end{array}\n$$",
+        "text_format": "latex",
+        "bbox": [
+            264,
+            695,
+            730,
+            786
+        ],
+        "page_idx": 17
+    },
+    {
+        "type": "text",
+        "text": "Note that the LSE term in the first line is strictly non-negative as the range of inner product for $\\mathbf { z } ^ { S }$ and $\\mathbf { d }$ lies between $\\left[ - 1 , + 1 \\right]$ : ",
+        "bbox": [
+            173,
+            795,
+            823,
+            825
+        ],
+        "page_idx": 17
+    },
+    {
+        "type": "equation",
+        "img_path": "images/60167030ccb1d22c58722f0344ef83e074921c76ef9e5b47b5b5bc7aa017a94c.jpg",
+        "text": "$$\n\\mathrm { L S E } ( \\mathbf { D } ^ { + } , \\mathbf { z } _ { i } ^ { S } ) \\geq \\log \\big ( M \\cdot \\exp ( - 1 / \\tau ^ { S } ) \\big ) = \\log \\big ( M \\cdot \\exp ( - 5 ) \\big ) > 0 ,\n$$",
+        "text_format": "latex",
+        "bbox": [
+            274,
+            833,
+            720,
+            854
+        ],
+        "page_idx": 17
+    },
+    {
+        "type": "text",
+        "text": "where $M$ denotes the cardinality of the maintained queue $\\mathbf { D } ^ { + }$ and is set to 65,536 in our experiment with $\\tau ^ { S } = 0 . 2$ constantly. Meanwhile, the LSE term in the second line satisfies the following inequality: ",
+        "bbox": [
+            173,
+            863,
+            825,
+            905
+        ],
+        "page_idx": 17
+    },
+    {
+        "type": "equation",
+        "img_path": "images/a9817ebf8558bdc08eb3f5d034f4302f4d5743c284feec760b5983580c09cbb1.jpg",
+        "text": "$$\n\\mathrm { L S E } ( \\mathbf { D } ^ { + } , \\mathbf { z } _ { i } ^ { S } ) \\geq \\mathrm { L S E } ( \\mathbf { D } , \\mathbf { z } _ { i } ^ { S } ) .\n$$",
+        "text_format": "latex",
+        "bbox": [
+            395,
+            906,
+            601,
+            925
+        ],
+        "page_idx": 17
+    },
+    {
+        "type": "text",
+        "text": "Thus, this demonstrates that the objective for SEED as Eq. 8 is equivalent to minimizing a weakened upper-bound of e.q. 7: ",
+        "bbox": [
+            171,
+            102,
+            825,
+            133
+        ],
+        "page_idx": 18
+    },
+    {
+        "type": "equation",
+        "img_path": "images/d86539a726b5a86171c5a58180e145908afef1dd30004cb1df11f07069fb1ca7.jpg",
+        "text": "$$\n\\begin{array} { l } { { \\mathcal { L } _ { \\mathrm { S E E D } } = \\displaystyle \\frac { 1 } { N } \\sum _ { i } ^ { N } - { \\bf p } ^ { T } ( { \\bf x } _ { i } ; \\boldsymbol { \\theta } _ { T } , { \\bf D } ^ { + } ) \\cdot \\log { \\bf p } ^ { S } ( { \\bf x } _ { i } ; \\boldsymbol { \\theta } _ { S } , { \\bf D } ^ { + } ) } \\ ~ } \\\\ { { \\displaystyle \\geq \\frac { 1 } { N } \\sum _ { i } ^ { N } \\left\\{ { \\bf w } _ { K + 1 } ^ { i } \\cdot ( - { \\bf z } _ { i } ^ { S } \\cdot { \\bf z } _ { i } ^ { T } / \\tau ^ { S } ) + \\sum _ { j = 1 } ^ { K } { \\bf w } _ { j } ^ { i } \\cdot \\left( - { \\bf z } _ { i } ^ { S } \\cdot { \\bf d } _ { j } / \\tau ^ { S } + \\log \\sum _ { \\bf d \\times D } ^ { C } \\exp ( { \\bf z } _ { i } ^ { S } \\cdot { \\bf d } / \\tau ^ { S } ) \\right) \\right\\} } \\ ~ } \\\\ { { \\displaystyle = \\frac { 1 } { N } \\sum _ { i } ^ { N } \\Bigg \\{ - \\frac { { \\bf w } _ { K + 1 } ^ { i } } { \\tau ^ { S } } \\cdot { \\bf z } _ { i } ^ { S } \\cdot { \\bf z } _ { i } ^ { T } - { \\bf p } ^ { T } ( { \\bf x } _ { i } ; \\boldsymbol { \\theta } _ { T } , { \\bf D } ) \\cdot \\log { \\bf p } ^ { S } ( { \\bf x } _ { i } ; \\boldsymbol { \\theta } _ { S } , { \\bf D } ) \\Bigg \\} } \\ ~ } \\end{array}\n$$",
+        "text_format": "latex",
+        "bbox": [
+            181,
+            137,
+            825,
+            272
+        ],
+        "page_idx": 18
+    },
+    {
+        "type": "text",
+        "text": "This proves that our $\\mathcal { L } _ { \\mathrm { S E E D } }$ directly relates to a more intuitive distillation formulation as Eq. 7 $( l 2 +$ cross entropy loss), and it implicitly contains the objective of aligning and contrasting. However, our training objective is much simplified. During practice, we find by regulating $\\tau ^ { T }$ , both training losses produce equal results. ",
+        "bbox": [
+            174,
+            281,
+            825,
+            338
+        ],
+        "page_idx": 18
+    },
+    {
+        "type": "text",
+        "text": "A.6 DISCUSSION ON THE RELATIONSHIP OF SEED WITH INFO-NCE ",
+        "text_level": 1,
+        "bbox": [
+            176,
+            353,
+            663,
+            369
+        ],
+        "page_idx": 18
+    },
+    {
+        "type": "text",
+        "text": "The objective of distillation can be considered as a soft version of Info-NCE (Oord et al., 2018), with the only difference to be that SEED learns from the negative samples with probabilities instead of treating them all strictly as negative samples. To be more specific, following Info-NCE, the “hard” style contrastive distillation can be expressed as aligning with representations from the Teacher encoder and contrasting with all random instances: ",
+        "bbox": [
+            174,
+            380,
+            826,
+            452
+        ],
+        "page_idx": 18
+    },
+    {
+        "type": "equation",
+        "img_path": "images/e34ea33920cc7ebc30f1c6559aaa3da4456286629890e13238bec91b4ea5dd57.jpg",
+        "text": "$$\n\\hat { \\theta } _ { S } = \\underset { \\theta _ { S } } { \\arg \\operatorname* { m i n } } \\ : \\mathcal { L } _ { N C E } = \\underset { \\theta _ { S } } { \\arg \\operatorname* { m i n } } \\sum _ { i } ^ { N } - \\log \\frac { \\exp ( \\mathbf { z } _ { i } ^ { T } \\cdot \\mathbf { z } _ { i } ^ { S } / \\tau ) } { \\sum _ { \\mathbf { d } \\sim \\mathbf { D } } \\exp ( \\mathbf { z } _ { i } ^ { S } \\cdot \\mathbf { d } / \\tau ) }\n$$",
+        "text_format": "latex",
+        "bbox": [
+            272,
+            457,
+            707,
+            500
+        ],
+        "page_idx": 18
+    },
+    {
+        "type": "text",
+        "text": "which can be further deduced with two sub-terms consisting of positive sample alignment and contrasting with negative instances: ",
+        "bbox": [
+            173,
+            505,
+            823,
+            534
+        ],
+        "page_idx": 18
+    },
+    {
+        "type": "equation",
+        "img_path": "images/065e9718ac69ca6ad65c1327e04b7725124e89d075d2dedbd19848d52002ec32.jpg",
+        "text": "$$\n\\mathcal { L } _ { N C E } = \\sum _ { i } ^ { N } \\Big \\{ \\underbrace { - \\mathbf { z } _ { i } ^ { S } \\cdot \\mathbf { z } _ { i } ^ { T } / \\tau } _ { a l i g n m e n t } + \\underbrace { \\log \\sum _ { \\mathbf { d } \\sim \\mathbf { D } } \\exp ( \\mathbf { z } _ { i } ^ { S } \\cdot \\mathbf { d } / \\tau ) } _ { c o n t r a s t i n g } \\Big \\} .\n$$",
+        "text_format": "latex",
+        "bbox": [
+            310,
+            539,
+            686,
+            603
+        ],
+        "page_idx": 18
+    },
+    {
+        "type": "text",
+        "text": "Similarly, the objective of SEED can be dissembled into the weighted form of alignment and contrasting terms: ",
+        "bbox": [
+            173,
+            609,
+            825,
+            638
+        ],
+        "page_idx": 18
+    },
+    {
+        "type": "equation",
+        "img_path": "images/0df2375166674514ce463d7ecc5010b2bd9ba95a67bcbc95f409e7ab2e12a30c.jpg",
+        "text": "$$\n\\begin{array} { r l } & { \\mathcal { L } _ { \\mathrm { S E E D } } = \\displaystyle \\frac { 1 } { N * M } \\sum _ { i } ^ { N } \\sum _ { j } ^ { K + 1 } - \\frac { \\exp ( { \\bf z } _ { i } ^ { T } \\cdot { \\bf d } _ { j } / \\tau ^ { T } ) } { \\sum _ { \\bf d } \\sim \\mathrm { D } ^ { + } } \\cdot \\log \\Big \\frac { \\exp ( { \\bf z } _ { i } ^ { S } \\cdot { \\bf d } _ { j } / \\tau ^ { S } ) } { \\sum _ { \\bf d } \\sim \\mathrm { D } ^ { + } } } \\\\ & { \\quad \\quad \\quad \\quad = \\frac { 1 } { N * M } \\sum _ { i } ^ { N } \\displaystyle \\sum _ { j } ^ { K + 1 } \\underbrace { \\frac { \\exp ( { \\bf z } _ { i } ^ { T } \\cdot { \\bf d } _ { j } / \\tau ^ { T } ) } { \\sum _ { \\bf d } \\sim \\mathrm { D } ^ { + } } } _ { { \\bf w } _ { j } ^ { \\bf d } } \\cdot \\underbrace { ( - { \\bf z } _ { i } ^ { S } \\cdot { \\bf z } _ { i } ^ { T } / \\tau ^ { S } ) } _ { a l i g n m e n t } \\cdot \\underbrace { ( \\frac { \\log \\sum _ { i } ^ { S } \\cdot { \\bf d } _ { i } / \\tau ^ { S } ) } { c \\mathrm { o } n t r a s t i n g } } _ { c o n t r a s t i n g } ) , } \\end{array}\n$$",
+        "text_format": "latex",
+        "bbox": [
+            181,
+            643,
+            821,
+            756
+        ],
+        "page_idx": 18
+    },
+    {
+        "type": "text",
+        "text": "where the normalization term can be considered as soft labels, $\\mathbf { W } ^ { i } = \\left[ \\mathbf { w } _ { 1 } ^ { i } \\ldots \\mathbf { w } _ { K + 1 } ^ { i } \\right]$ , which can weight the above loss as: ",
+        "bbox": [
+            173,
+            763,
+            828,
+            792
+        ],
+        "page_idx": 18
+    },
+    {
+        "type": "equation",
+        "img_path": "images/32d06c560859d8112c17701b5a0909209017dbe9d795a6964a507eb73141f98d.jpg",
+        "text": "$$\n\\mathcal { L } _ { \\mathrm { S E E D } } = \\frac { 1 } { N * M } \\sum _ { i } ^ { N } \\sum _ { j } ^ { K + 1 } \\mathbf { w } _ { j } ^ { i } \\cdot \\Big \\{ - \\mathbf { z } _ { i } ^ { S } \\cdot \\mathbf { z } _ { i } ^ { T } / \\tau ^ { S } + \\log \\sum _ { \\mathbf { d } \\sim \\mathbf { D } } \\exp ( \\mathbf { z } _ { i } ^ { S } \\cdot \\mathbf { d } / \\tau ^ { S } ) ) \\Big \\} ,\n$$",
+        "text_format": "latex",
+        "bbox": [
+            238,
+            799,
+            758,
+            844
+        ],
+        "page_idx": 18
+    },
+    {
+        "type": "text",
+        "text": "When tuning hyper-parameter $\\tau ^ { T }$ towards 0, $\\mathbf { W } ^ { i }$ can be altered into the format of one-hot vector with $\\mathbf { w } _ { K + 1 } ^ { i } = \\dot { 1 }$ , which is then degraded to the case of contrastive distillation as in equation 14. In practice, the choice of an optimal $\\tau ^ { T }$ can be dataset-specific. We show that the higher $\\tau ^ { T }$ (with labels be more ‘soft’) can actually yield better results on other datasets, e.g., CIFAR-10 (Krizhevsky et al., 2009). ",
+        "bbox": [
+            173,
+            849,
+            825,
+            924
+        ],
+        "page_idx": 18
+    },
+    {
+        "type": "text",
+        "text": "A.7 COMPATIBILITY WITH SUPERVISED DISTILLATION ",
+        "text_level": 1,
+        "bbox": [
+            174,
+            103,
+            566,
+            117
+        ],
+        "page_idx": 19
+    },
+    {
+        "type": "text",
+        "text": "SEED conducts self-supervised distillation at the pre-training phase for the representation learning. However, we verify that SEED is compatible with traditional supervised distillation that happened during fine-tuning phrase at downstream, and can even produce better results. We begin with the SSL pre-training on a larger architecture (ResNet-152) using MoCo-V2 and train it for 200 epochs as the teacher network. As images in CIFAR-100 are in the size of $3 2 \\times 3 2$ , we modify the first conv layer in ResNet with kernel size $= 3$ and stride $= 1$ . ",
+        "bbox": [
+            174,
+            130,
+            825,
+            213
+        ],
+        "page_idx": 19
+    },
+    {
+        "type": "text",
+        "text": "We then compare the Top-1 accuracy of a smaller ResNet-18 on CIFAR-100 when using different distillation strategies when all parameters are trainable. First, we use SEED to pre-train ResNet-18 with Res-152 as the teacher model, and then evaluate in on the test split of CIFAR-100 using linear fine-tuning task. As we keep all parameters trainable during the fine-tuning phase, distillation on the pre-training only yields a trivial boost: $7 5 . 4 \\%$ v.s. $7 5 . 2 \\%$ . Then, we adopt the traditional distillation method, e.g., (Hinton et al., 2015), to first fine-tune the ResNet-152 model, and then use its output class probability to facilitate the linear classification task on ResNet-18 in the fine-tuning phrase. This improves the linear classification accuracy on ResNet-18 to $7 6 . 0 \\%$ . At the end, we initialize the ResNet-18 with our SEED pre-trained ResNet-18, and equip it with the supervised classification distillation during fine-tuning. With that, we find that the performance of ResNet-18 is further boosted to $7 8 . 1 \\%$ . We can conclude that our SEED is compatible with traditional supervised distillation that mostly happened at downstream for specific tasks, e.g., classification, object detection. ",
+        "bbox": [
+            173,
+            220,
+            825,
+            387
+        ],
+        "page_idx": 19
+    },
+    {
+        "type": "table",
+        "img_path": "images/cc40158c716b6af83e7e123cc7de02ae87a713e9db269a0724d73ab8a599f371.jpg",
+        "table_caption": [
+            "Table 12: CIFAR-100 Top-1 Accuracy $( \\% )$ of ResNet-18 with (or without) distillation at different phase: selfsupervised pre-training stage, and supervised classification fine-tuning. All backbone parameters of ResNet-18 are trainable in experiments. "
+        ],
+        "table_footnote": [],
+        "table_body": "<table><tr><td colspan=\"3\">Pre-training Distill. Fine-tuning Distill. Top-1 Acc</td></tr><tr><td>X</td><td>X</td><td>75.2</td></tr><tr><td>√</td><td>×</td><td>75.4</td></tr><tr><td>×</td><td>√</td><td>76.0</td></tr><tr><td>√</td><td>√</td><td>78.1</td></tr></table>",
+        "bbox": [
+            318,
+            459,
+            676,
+            544
+        ],
+        "page_idx": 19
+    },
+    {
+        "type": "image",
+        "img_path": "images/d2dbbfad613f736ba1bbdacdf0d815fdcec0dd5cde93e27f656664f449c6e7cd.jpg",
+        "image_caption": [
+            "Figure 7: Linear evaluation accuracy $( \\% )$ of distillation between ResNet-18 (as the Student) and ResNet-50 (as the Teacher) using different size of queue when $\\mathrm { L R } { = } 0 . 0 3$ and weight decay=1e-6. Note the axis is the log(·) value of queue lengths. "
+        ],
+        "image_footnote": [],
+        "bbox": [
+            174,
+            580,
+            818,
+            747
+        ],
+        "page_idx": 19
+    },
+    {
+        "type": "text",
+        "text": "A.8 ADDITIONAL ABLATION STUDIES ",
+        "text_level": 1,
+        "bbox": [
+            178,
+            827,
+            450,
+            842
+        ],
+        "page_idx": 19
+    },
+    {
+        "type": "text",
+        "text": "We study effects of different hyper-parameters to distillation using a ResNet-18 (as Student) and a SWAV pre-trained ResNet-50 (as Teacher) with small patch views. In specific, we list the Top-1 Acc. on validation split of ImageNet-1k using different lengths of queue $K { = } 1 2 8$ , 512, 1,024, 4,096, 8,192, 16,384, 65,536) in Figure. 7. With the increasing of random data samples, the distillation boosts the accuracy of learned representations, however within a limited range: $+ 1 . 5$ when the queue size is ",
+        "bbox": [
+            174,
+            853,
+            825,
+            924
+        ],
+        "page_idx": 19
+    },
+    {
+        "type": "table",
+        "img_path": "images/120cf29bca49ba5997cc5ea1380a9c72f739a04304171a9fcc5f244930c68c5f.jpg",
+        "table_caption": [
+            "Table 13: Linear evaluation accuracy $( \\% )$ of distillation between ResNet-18 (as the Student) and ResNet-50 (as the Teacher) using different learning rates when the queue size is 65,536 and weight decay=1e-6. "
+        ],
+        "table_footnote": [],
+        "table_body": "<table><tr><td>LR</td><td>Top-1 Acc. Top-5 Acc.</td><td></td></tr><tr><td>1</td><td>58.9</td><td>83.1</td></tr><tr><td>0.1</td><td>62.9</td><td>85.3</td></tr><tr><td>0.03</td><td>63.3</td><td>85.4</td></tr><tr><td>0.01</td><td>62.6</td><td>85.0</td></tr></table>",
+        "bbox": [
+            173,
+            184,
+            385,
+            267
+        ],
+        "page_idx": 20
+    },
+    {
+        "type": "table",
+        "img_path": "images/b86126ebcbfd9cbef60e6b42f12f5be8b33e96aeaf60d51c629bc07499ef28df.jpg",
+        "table_caption": [
+            "Table 14: Linear evaluation accuracy $( \\% )$ of distillation between ResNet-18 (as the Student) and ResNet-50 (as the Teacher) using different weight decays when the queue size is 65,536 and $\\mathrm { L R } { = } 0 . 0 3$ . "
+        ],
+        "table_footnote": [],
+        "table_body": "<table><tr><td>WD</td><td>Top-1 Acc. Top-5 Acc.</td><td></td></tr><tr><td>1e-2</td><td>11.8</td><td>27.7</td></tr><tr><td>1e-3</td><td>62.3</td><td>84.7</td></tr><tr><td>1e-4</td><td>61.9</td><td>84.4</td></tr><tr><td>1e-5</td><td>61.6</td><td>84.2</td></tr><tr><td>1e-6</td><td>63.3</td><td>85.4</td></tr></table>",
+        "bbox": [
+            493,
+            171,
+            705,
+            268
+        ],
+        "page_idx": 20
+    },
+    {
+        "type": "text",
+        "text": "65,536 compared with 256. Furthermore, Table. 13 and 14 summarize the linear evaluation accuracy under different learning rates and weight decays. ",
+        "bbox": [
+            174,
+            297,
+            823,
+            327
+        ],
+        "page_idx": 20
+    }
+]
\ No newline at end of file
diff --git a/parse/train/AHm3dbp7D1D/AHm3dbp7D1D_middle.json b/parse/train/AHm3dbp7D1D/AHm3dbp7D1D_middle.json
new file mode 100644
index 0000000000000000000000000000000000000000..27e16f6e8147cea81f4ab7f9ee728c32f9e5ad58
--- /dev/null
+++ b/parse/train/AHm3dbp7D1D/AHm3dbp7D1D_middle.json
@@ -0,0 +1,59339 @@
+{
+    "pdf_info": [
+        {
+            "preproc_blocks": [
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        79,
+                        448,
+                        118
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                79,
+                                449,
+                                100
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        79,
+                                        449,
+                                        100
+                                    ],
+                                    "score": 1.0,
+                                    "content": "SEED: SELF-SUPERVISED DISTILLATION FOR",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                100,
+                                303,
+                                120
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        100,
+                                        303,
+                                        120
+                                    ],
+                                    "score": 1.0,
+                                    "content": "VISUAL REPRESENTATION",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        }
+                    ],
+                    "index": 0.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        114,
+                        136,
+                        498,
+                        149
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                114,
+                                135,
+                                500,
+                                151
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        114,
+                                        135,
+                                        500,
+                                        151
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Zhiyuan Fang† , Jianfeng Wang‡, Lijuan Wang‡, Lei Zhang‡, Yezhou Yang†, Zicheng Liu‡",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        }
+                    ],
+                    "index": 2
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        158,
+                        150,
+                        440,
+                        182
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                192,
+                                147,
+                                435,
+                                162
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        192,
+                                        147,
+                                        299,
+                                        162
+                                    ],
+                                    "score": 1.0,
+                                    "content": "†Arizona State University,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        339,
+                                        147,
+                                        435,
+                                        162
+                                    ],
+                                    "score": 1.0,
+                                    "content": "‡Microsoft Corporation",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                222,
+                                161,
+                                380,
+                                172
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        222,
+                                        161,
+                                        380,
+                                        172
+                                    ],
+                                    "score": 1.0,
+                                    "content": "{zy.fang, yz.yang}@asu.edu",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                159,
+                                172,
+                                441,
+                                183
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        159,
+                                        172,
+                                        441,
+                                        183
+                                    ],
+                                    "score": 1.0,
+                                    "content": "{jianfw, lijuanw, leizhang, zliu}@microsoft.com",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        }
+                    ],
+                    "index": 4
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        278,
+                        217,
+                        333,
+                        228
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                276,
+                                216,
+                                335,
+                                230
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        276,
+                                        216,
+                                        335,
+                                        230
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ABSTRACT",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        }
+                    ],
+                    "index": 6
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        143,
+                        241,
+                        469,
+                        384
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                142,
+                                242,
+                                468,
+                                254
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        142,
+                                        242,
+                                        468,
+                                        254
+                                    ],
+                                    "score": 1.0,
+                                    "content": "This paper is concerned with self-supervised learning for small models. The",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                253,
+                                469,
+                                266
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        253,
+                                        469,
+                                        266
+                                    ],
+                                    "score": 1.0,
+                                    "content": "problem is motivated by our empirical studies that while the widely used contrastive",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                140,
+                                262,
+                                470,
+                                277
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        140,
+                                        262,
+                                        470,
+                                        277
+                                    ],
+                                    "score": 1.0,
+                                    "content": "self-supervised learning method has shown great progress on large model training,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                140,
+                                273,
+                                471,
+                                288
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        140,
+                                        273,
+                                        471,
+                                        288
+                                    ],
+                                    "score": 1.0,
+                                    "content": "it does not work well for small models. To address this problem, we propose a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                286,
+                                470,
+                                298
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        286,
+                                        470,
+                                        298
+                                    ],
+                                    "score": 1.0,
+                                    "content": "new learning paradigm, named SElf-SupErvised Distillation (SEED), where we",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                297,
+                                469,
+                                309
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        297,
+                                        469,
+                                        309
+                                    ],
+                                    "score": 1.0,
+                                    "content": "leverage a larger network (as Teacher) to transfer its representational knowledge",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                307,
+                                470,
+                                319
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        307,
+                                        470,
+                                        319
+                                    ],
+                                    "score": 1.0,
+                                    "content": "into a smaller architecture (as Student) in a self-supervised fashion. Instead of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                319,
+                                470,
+                                330
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        319,
+                                        470,
+                                        330
+                                    ],
+                                    "score": 1.0,
+                                    "content": "directly learning from unlabeled data, we train a student encoder to mimic the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                142,
+                                330,
+                                469,
+                                341
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        142,
+                                        330,
+                                        469,
+                                        341
+                                    ],
+                                    "score": 1.0,
+                                    "content": "similarity score distribution inferred by a teacher over a set of instances. We show",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                340,
+                                469,
+                                352
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        340,
+                                        469,
+                                        352
+                                    ],
+                                    "score": 1.0,
+                                    "content": "that SEED dramatically boosts the performance of small networks on downstream",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                140,
+                                350,
+                                470,
+                                365
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        140,
+                                        350,
+                                        470,
+                                        365
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tasks. Compared with self-supervised baselines, SEED improves the top-1 accuracy",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                362,
+                                470,
+                                373
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        362,
+                                        163,
+                                        373
+                                    ],
+                                    "score": 1.0,
+                                    "content": "from",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        164,
+                                        362,
+                                        190,
+                                        373
+                                    ],
+                                    "score": 0.87,
+                                    "content": "4 2 . 2 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        191,
+                                        362,
+                                        201,
+                                        373
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        201,
+                                        362,
+                                        228,
+                                        373
+                                    ],
+                                    "score": 0.87,
+                                    "content": "6 7 . 6 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        229,
+                                        362,
+                                        344,
+                                        373
+                                    ],
+                                    "score": 1.0,
+                                    "content": "on EfficientNet-B0 and from",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        345,
+                                        362,
+                                        371,
+                                        373
+                                    ],
+                                    "score": 0.87,
+                                    "content": "3 6 . 3 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        372,
+                                        362,
+                                        382,
+                                        373
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        383,
+                                        362,
+                                        409,
+                                        373
+                                    ],
+                                    "score": 0.87,
+                                    "content": "6 8 . 2 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        410,
+                                        362,
+                                        470,
+                                        373
+                                    ],
+                                    "score": 1.0,
+                                    "content": "on MobileNet-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                142,
+                                374,
+                                298,
+                                385
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        142,
+                                        374,
+                                        298,
+                                        385
+                                    ],
+                                    "score": 1.0,
+                                    "content": "V3-Large on the ImageNet-1k dataset.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        }
+                    ],
+                    "index": 13
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        406,
+                        205,
+                        417
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                403,
+                                207,
+                                421
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        403,
+                                        207,
+                                        421
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1 INTRODUCTION",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        }
+                    ],
+                    "index": 20
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        429,
+                        321,
+                        649
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                429,
+                                321,
+                                442
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        429,
+                                        321,
+                                        442
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The burgeoning studies and success on self-supervised",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                441,
+                                321,
+                                453
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        441,
+                                        321,
+                                        453
+                                    ],
+                                    "score": 1.0,
+                                    "content": "learning (SSL) for visual representation are mainly",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                451,
+                                321,
+                                463
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        451,
+                                        321,
+                                        463
+                                    ],
+                                    "score": 1.0,
+                                    "content": "marked by its extraordinary potency of learning from",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                462,
+                                321,
+                                475
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        462,
+                                        321,
+                                        475
+                                    ],
+                                    "score": 1.0,
+                                    "content": "unlabeled data at scale. Accompanying with the SSL",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                473,
+                                320,
+                                485
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        473,
+                                        320,
+                                        485
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is its phenomenal benefit of obtaining task-agnostic",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                485,
+                                321,
+                                497
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        485,
+                                        321,
+                                        497
+                                    ],
+                                    "score": 1.0,
+                                    "content": "representations while allowing the training to dispense",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                496,
+                                322,
+                                507
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        496,
+                                        322,
+                                        507
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with prohibitively expensive data labeling. Major ram-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                506,
+                                321,
+                                518
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        506,
+                                        321,
+                                        518
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ifications of visual SSL include pretext tasks (Noroozi",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                516,
+                                322,
+                                529
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        516,
+                                        322,
+                                        529
+                                    ],
+                                    "score": 1.0,
+                                    "content": "& Favaro, 2016; Zhang et al., 2016; Gidaris et al.,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                528,
+                                322,
+                                540
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        528,
+                                        322,
+                                        540
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2018; Zhang et al., 2019; Feng et al., 2019), con-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                539,
+                                321,
+                                551
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        539,
+                                        321,
+                                        551
+                                    ],
+                                    "score": 1.0,
+                                    "content": "trastive representation learning (Wu et al., 2018; He",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                550,
+                                322,
+                                561
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        550,
+                                        322,
+                                        561
+                                    ],
+                                    "score": 1.0,
+                                    "content": "et al., 2020; Chen et al., 2020a), online/offline cluster-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                561,
+                                322,
+                                573
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        561,
+                                        322,
+                                        573
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ing (Yang et al., 2016; Caron et al., 2018; Li et al.,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                571,
+                                322,
+                                583
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        571,
+                                        322,
+                                        583
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2020; Caron et al., 2020; Grill et al., 2020), etc.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                583,
+                                322,
+                                594
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        583,
+                                        322,
+                                        594
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Among them, several recent works (He et al., 2020;",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                594,
+                                321,
+                                605
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        594,
+                                        321,
+                                        605
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Chen et al., 2020a; Caron et al., 2020) have achieved",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                605,
+                                322,
+                                617
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        605,
+                                        322,
+                                        617
+                                    ],
+                                    "score": 1.0,
+                                    "content": "comparable or even better accuracy than the super-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                617,
+                                321,
+                                628
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        617,
+                                        321,
+                                        628
+                                    ],
+                                    "score": 1.0,
+                                    "content": "vised pre-training when transferring to downstream",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                627,
+                                322,
+                                639
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        627,
+                                        322,
+                                        639
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tasks, e.g. semi-supervised classification, object detec-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                637,
+                                128,
+                                650
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        637,
+                                        128,
+                                        650
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tion.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        }
+                    ],
+                    "index": 30.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        655,
+                        321,
+                        699
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                655,
+                                321,
+                                667
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        655,
+                                        321,
+                                        667
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The aforementioned top-performing SSL algorithms",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                665,
+                                322,
+                                677
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        665,
+                                        322,
+                                        677
+                                    ],
+                                    "score": 1.0,
+                                    "content": "all involve large networks (e.g., ResNet-50 (He et al.,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                676,
+                                321,
+                                689
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        676,
+                                        321,
+                                        689
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2016) or larger), with, however, little attention on",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                687,
+                                321,
+                                700
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        687,
+                                        321,
+                                        700
+                                    ],
+                                    "score": 1.0,
+                                    "content": "small networks. Empirically, we find that existing",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 64
+                        }
+                    ],
+                    "index": 42.5
+                },
+                {
+                    "type": "image",
+                    "bbox": [
+                        330,
+                        435,
+                        502,
+                        564
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                330,
+                                435,
+                                502,
+                                564
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        330,
+                                        435,
+                                        502,
+                                        564
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                330,
+                                                435,
+                                                502,
+                                                564
+                                            ],
+                                            "score": 0.972,
+                                            "type": "image",
+                                            "image_path": "d4fb00f69f184e9fccdf211a9f298a6ff3e7571f5dfb94ff7b3d5393734ffc54.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 48.5,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        330,
+                                        435,
+                                        502,
+                                        447.9
+                                    ],
+                                    "spans": [],
+                                    "index": 44
+                                },
+                                {
+                                    "bbox": [
+                                        330,
+                                        447.9,
+                                        502,
+                                        460.79999999999995
+                                    ],
+                                    "spans": [],
+                                    "index": 45
+                                },
+                                {
+                                    "bbox": [
+                                        330,
+                                        460.79999999999995,
+                                        502,
+                                        473.69999999999993
+                                    ],
+                                    "spans": [],
+                                    "index": 46
+                                },
+                                {
+                                    "bbox": [
+                                        330,
+                                        473.69999999999993,
+                                        502,
+                                        486.5999999999999
+                                    ],
+                                    "spans": [],
+                                    "index": 47
+                                },
+                                {
+                                    "bbox": [
+                                        330,
+                                        486.5999999999999,
+                                        502,
+                                        499.4999999999999
+                                    ],
+                                    "spans": [],
+                                    "index": 48
+                                },
+                                {
+                                    "bbox": [
+                                        330,
+                                        499.4999999999999,
+                                        502,
+                                        512.3999999999999
+                                    ],
+                                    "spans": [],
+                                    "index": 49
+                                },
+                                {
+                                    "bbox": [
+                                        330,
+                                        512.3999999999999,
+                                        502,
+                                        525.2999999999998
+                                    ],
+                                    "spans": [],
+                                    "index": 50
+                                },
+                                {
+                                    "bbox": [
+                                        330,
+                                        525.2999999999998,
+                                        502,
+                                        538.1999999999998
+                                    ],
+                                    "spans": [],
+                                    "index": 51
+                                },
+                                {
+                                    "bbox": [
+                                        330,
+                                        538.1999999999998,
+                                        502,
+                                        551.0999999999998
+                                    ],
+                                    "spans": [],
+                                    "index": 52
+                                },
+                                {
+                                    "bbox": [
+                                        330,
+                                        551.0999999999998,
+                                        502,
+                                        563.9999999999998
+                                    ],
+                                    "spans": [],
+                                    "index": 53
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                328,
+                                569,
+                                505,
+                                671
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        327,
+                                        569,
+                                        506,
+                                        582
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                327,
+                                                569,
+                                                506,
+                                                582
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 1: SEED vs. MoCo-V2 (Chen et al.,",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 54
+                                },
+                                {
+                                    "bbox": [
+                                        327,
+                                        578,
+                                        507,
+                                        592
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                327,
+                                                578,
+                                                507,
+                                                592
+                                            ],
+                                            "score": 1.0,
+                                            "content": "2020c)) on ImageNet-1K linear probe accuracy.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 55
+                                },
+                                {
+                                    "bbox": [
+                                        327,
+                                        590,
+                                        505,
+                                        601
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                327,
+                                                590,
+                                                505,
+                                                601
+                                            ],
+                                            "score": 1.0,
+                                            "content": "The vertical axis is the top-1 accuracy and the",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 56
+                                },
+                                {
+                                    "bbox": [
+                                        327,
+                                        599,
+                                        506,
+                                        612
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                327,
+                                                599,
+                                                506,
+                                                612
+                                            ],
+                                            "score": 1.0,
+                                            "content": "horizontal axis is the number of learnable pa-",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 57
+                                },
+                                {
+                                    "bbox": [
+                                        327,
+                                        610,
+                                        506,
+                                        619
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                327,
+                                                610,
+                                                506,
+                                                619
+                                            ],
+                                            "score": 1.0,
+                                            "content": "rameters for different network architectures. Di-",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 58
+                                },
+                                {
+                                    "bbox": [
+                                        327,
+                                        620,
+                                        507,
+                                        631
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                327,
+                                                620,
+                                                507,
+                                                631
+                                            ],
+                                            "score": 1.0,
+                                            "content": "rectly applying self-supervised contrastive learn-",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 59
+                                },
+                                {
+                                    "bbox": [
+                                        327,
+                                        630,
+                                        506,
+                                        640
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                327,
+                                                630,
+                                                506,
+                                                640
+                                            ],
+                                            "score": 1.0,
+                                            "content": "ing (MoCo-V2) does not work well for smaller",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 60
+                                },
+                                {
+                                    "bbox": [
+                                        328,
+                                        641,
+                                        505,
+                                        651
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                328,
+                                                641,
+                                                505,
+                                                651
+                                            ],
+                                            "score": 1.0,
+                                            "content": "architectures, while our method (SEED) leads",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 61
+                                },
+                                {
+                                    "bbox": [
+                                        327,
+                                        651,
+                                        505,
+                                        661
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                327,
+                                                651,
+                                                505,
+                                                661
+                                            ],
+                                            "score": 1.0,
+                                            "content": "to dramatic performance boost. Details of the",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 62
+                                },
+                                {
+                                    "bbox": [
+                                        327,
+                                        661,
+                                        450,
+                                        671
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                327,
+                                                661,
+                                                450,
+                                                671
+                                            ],
+                                            "score": 1.0,
+                                            "content": "setting can be found in Section 4.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 63
+                                }
+                            ],
+                            "index": 58.5
+                        }
+                    ],
+                    "index": 53.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        699,
+                        505,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                698,
+                                505,
+                                711
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        698,
+                                        505,
+                                        711
+                                    ],
+                                    "score": 1.0,
+                                    "content": "techniques like contrastive learning do not work well on small networks. For instance, the linear probe",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 65
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                709,
+                                506,
+                                721
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        709,
+                                        407,
+                                        721
+                                    ],
+                                    "score": 1.0,
+                                    "content": "top-1 accuracy on ImageNet using MoCo-V2 (Chen et al., 2020c) is only",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        407,
+                                        710,
+                                        435,
+                                        720
+                                    ],
+                                    "score": 0.85,
+                                    "content": "3 6 . 3 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        435,
+                                        709,
+                                        506,
+                                        721
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with MobileNet-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 66
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                720,
+                                505,
+                                734
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        720,
+                                        505,
+                                        734
+                                    ],
+                                    "score": 1.0,
+                                    "content": "V3-Large (see Figure 1), which is much lower compared with its supervised training accuracy",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 67
+                        }
+                    ],
+                    "index": 66
+                }
+            ],
+            "page_idx": 0,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        108,
+                        27,
+                        292,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                26,
+                                293,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        26,
+                                        293,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Published as a conference paper at ICLR 2021",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        302,
+                        751,
+                        308,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                302,
+                                751,
+                                308,
+                                761
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        302,
+                                        751,
+                                        308,
+                                        761
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        79,
+                        448,
+                        118
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                79,
+                                449,
+                                100
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        79,
+                                        449,
+                                        100
+                                    ],
+                                    "score": 1.0,
+                                    "content": "SEED: SELF-SUPERVISED DISTILLATION FOR",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                100,
+                                303,
+                                120
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        100,
+                                        303,
+                                        120
+                                    ],
+                                    "score": 1.0,
+                                    "content": "VISUAL REPRESENTATION",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        }
+                    ],
+                    "index": 0.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        114,
+                        136,
+                        498,
+                        149
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                114,
+                                135,
+                                500,
+                                151
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        114,
+                                        135,
+                                        500,
+                                        151
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Zhiyuan Fang† , Jianfeng Wang‡, Lijuan Wang‡, Lei Zhang‡, Yezhou Yang†, Zicheng Liu‡",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        }
+                    ],
+                    "index": 2,
+                    "bbox_fs": [
+                        114,
+                        135,
+                        500,
+                        151
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        158,
+                        150,
+                        440,
+                        182
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                192,
+                                147,
+                                435,
+                                162
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        192,
+                                        147,
+                                        299,
+                                        162
+                                    ],
+                                    "score": 1.0,
+                                    "content": "†Arizona State University,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        339,
+                                        147,
+                                        435,
+                                        162
+                                    ],
+                                    "score": 1.0,
+                                    "content": "‡Microsoft Corporation",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                222,
+                                161,
+                                380,
+                                172
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        222,
+                                        161,
+                                        380,
+                                        172
+                                    ],
+                                    "score": 1.0,
+                                    "content": "{zy.fang, yz.yang}@asu.edu",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                159,
+                                172,
+                                441,
+                                183
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        159,
+                                        172,
+                                        441,
+                                        183
+                                    ],
+                                    "score": 1.0,
+                                    "content": "{jianfw, lijuanw, leizhang, zliu}@microsoft.com",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        }
+                    ],
+                    "index": 4,
+                    "bbox_fs": [
+                        159,
+                        147,
+                        441,
+                        183
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        278,
+                        217,
+                        333,
+                        228
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                276,
+                                216,
+                                335,
+                                230
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        276,
+                                        216,
+                                        335,
+                                        230
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ABSTRACT",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        }
+                    ],
+                    "index": 6
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        143,
+                        241,
+                        469,
+                        384
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                142,
+                                242,
+                                468,
+                                254
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        142,
+                                        242,
+                                        468,
+                                        254
+                                    ],
+                                    "score": 1.0,
+                                    "content": "This paper is concerned with self-supervised learning for small models. The",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                253,
+                                469,
+                                266
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        253,
+                                        469,
+                                        266
+                                    ],
+                                    "score": 1.0,
+                                    "content": "problem is motivated by our empirical studies that while the widely used contrastive",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                140,
+                                262,
+                                470,
+                                277
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        140,
+                                        262,
+                                        470,
+                                        277
+                                    ],
+                                    "score": 1.0,
+                                    "content": "self-supervised learning method has shown great progress on large model training,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                140,
+                                273,
+                                471,
+                                288
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        140,
+                                        273,
+                                        471,
+                                        288
+                                    ],
+                                    "score": 1.0,
+                                    "content": "it does not work well for small models. To address this problem, we propose a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                286,
+                                470,
+                                298
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        286,
+                                        470,
+                                        298
+                                    ],
+                                    "score": 1.0,
+                                    "content": "new learning paradigm, named SElf-SupErvised Distillation (SEED), where we",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                297,
+                                469,
+                                309
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        297,
+                                        469,
+                                        309
+                                    ],
+                                    "score": 1.0,
+                                    "content": "leverage a larger network (as Teacher) to transfer its representational knowledge",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                307,
+                                470,
+                                319
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        307,
+                                        470,
+                                        319
+                                    ],
+                                    "score": 1.0,
+                                    "content": "into a smaller architecture (as Student) in a self-supervised fashion. Instead of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                319,
+                                470,
+                                330
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        319,
+                                        470,
+                                        330
+                                    ],
+                                    "score": 1.0,
+                                    "content": "directly learning from unlabeled data, we train a student encoder to mimic the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                142,
+                                330,
+                                469,
+                                341
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        142,
+                                        330,
+                                        469,
+                                        341
+                                    ],
+                                    "score": 1.0,
+                                    "content": "similarity score distribution inferred by a teacher over a set of instances. We show",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                340,
+                                469,
+                                352
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        340,
+                                        469,
+                                        352
+                                    ],
+                                    "score": 1.0,
+                                    "content": "that SEED dramatically boosts the performance of small networks on downstream",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                140,
+                                350,
+                                470,
+                                365
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        140,
+                                        350,
+                                        470,
+                                        365
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tasks. Compared with self-supervised baselines, SEED improves the top-1 accuracy",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                362,
+                                470,
+                                373
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        362,
+                                        163,
+                                        373
+                                    ],
+                                    "score": 1.0,
+                                    "content": "from",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        164,
+                                        362,
+                                        190,
+                                        373
+                                    ],
+                                    "score": 0.87,
+                                    "content": "4 2 . 2 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        191,
+                                        362,
+                                        201,
+                                        373
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        201,
+                                        362,
+                                        228,
+                                        373
+                                    ],
+                                    "score": 0.87,
+                                    "content": "6 7 . 6 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        229,
+                                        362,
+                                        344,
+                                        373
+                                    ],
+                                    "score": 1.0,
+                                    "content": "on EfficientNet-B0 and from",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        345,
+                                        362,
+                                        371,
+                                        373
+                                    ],
+                                    "score": 0.87,
+                                    "content": "3 6 . 3 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        372,
+                                        362,
+                                        382,
+                                        373
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        383,
+                                        362,
+                                        409,
+                                        373
+                                    ],
+                                    "score": 0.87,
+                                    "content": "6 8 . 2 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        410,
+                                        362,
+                                        470,
+                                        373
+                                    ],
+                                    "score": 1.0,
+                                    "content": "on MobileNet-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                142,
+                                374,
+                                298,
+                                385
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        142,
+                                        374,
+                                        298,
+                                        385
+                                    ],
+                                    "score": 1.0,
+                                    "content": "V3-Large on the ImageNet-1k dataset.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        }
+                    ],
+                    "index": 13,
+                    "bbox_fs": [
+                        140,
+                        242,
+                        471,
+                        385
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        406,
+                        205,
+                        417
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                403,
+                                207,
+                                421
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        403,
+                                        207,
+                                        421
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1 INTRODUCTION",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        }
+                    ],
+                    "index": 20
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        429,
+                        321,
+                        649
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                429,
+                                321,
+                                442
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        429,
+                                        321,
+                                        442
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The burgeoning studies and success on self-supervised",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                441,
+                                321,
+                                453
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        441,
+                                        321,
+                                        453
+                                    ],
+                                    "score": 1.0,
+                                    "content": "learning (SSL) for visual representation are mainly",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                451,
+                                321,
+                                463
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        451,
+                                        321,
+                                        463
+                                    ],
+                                    "score": 1.0,
+                                    "content": "marked by its extraordinary potency of learning from",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                462,
+                                321,
+                                475
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        462,
+                                        321,
+                                        475
+                                    ],
+                                    "score": 1.0,
+                                    "content": "unlabeled data at scale. Accompanying with the SSL",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                473,
+                                320,
+                                485
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        473,
+                                        320,
+                                        485
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is its phenomenal benefit of obtaining task-agnostic",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                485,
+                                321,
+                                497
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        485,
+                                        321,
+                                        497
+                                    ],
+                                    "score": 1.0,
+                                    "content": "representations while allowing the training to dispense",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                496,
+                                322,
+                                507
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        496,
+                                        322,
+                                        507
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with prohibitively expensive data labeling. Major ram-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                506,
+                                321,
+                                518
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        506,
+                                        321,
+                                        518
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ifications of visual SSL include pretext tasks (Noroozi",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                516,
+                                322,
+                                529
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        516,
+                                        322,
+                                        529
+                                    ],
+                                    "score": 1.0,
+                                    "content": "& Favaro, 2016; Zhang et al., 2016; Gidaris et al.,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                528,
+                                322,
+                                540
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        528,
+                                        322,
+                                        540
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2018; Zhang et al., 2019; Feng et al., 2019), con-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                539,
+                                321,
+                                551
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        539,
+                                        321,
+                                        551
+                                    ],
+                                    "score": 1.0,
+                                    "content": "trastive representation learning (Wu et al., 2018; He",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                550,
+                                322,
+                                561
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        550,
+                                        322,
+                                        561
+                                    ],
+                                    "score": 1.0,
+                                    "content": "et al., 2020; Chen et al., 2020a), online/offline cluster-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                561,
+                                322,
+                                573
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        561,
+                                        322,
+                                        573
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ing (Yang et al., 2016; Caron et al., 2018; Li et al.,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                571,
+                                322,
+                                583
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        571,
+                                        322,
+                                        583
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2020; Caron et al., 2020; Grill et al., 2020), etc.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                583,
+                                322,
+                                594
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        583,
+                                        322,
+                                        594
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Among them, several recent works (He et al., 2020;",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                594,
+                                321,
+                                605
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        594,
+                                        321,
+                                        605
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Chen et al., 2020a; Caron et al., 2020) have achieved",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                605,
+                                322,
+                                617
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        605,
+                                        322,
+                                        617
+                                    ],
+                                    "score": 1.0,
+                                    "content": "comparable or even better accuracy than the super-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                617,
+                                321,
+                                628
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        617,
+                                        321,
+                                        628
+                                    ],
+                                    "score": 1.0,
+                                    "content": "vised pre-training when transferring to downstream",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                627,
+                                322,
+                                639
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        627,
+                                        322,
+                                        639
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tasks, e.g. semi-supervised classification, object detec-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                637,
+                                128,
+                                650
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        637,
+                                        128,
+                                        650
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tion.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        }
+                    ],
+                    "index": 30.5,
+                    "bbox_fs": [
+                        105,
+                        429,
+                        322,
+                        650
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        655,
+                        321,
+                        699
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                655,
+                                321,
+                                667
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        655,
+                                        321,
+                                        667
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The aforementioned top-performing SSL algorithms",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                665,
+                                322,
+                                677
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        665,
+                                        322,
+                                        677
+                                    ],
+                                    "score": 1.0,
+                                    "content": "all involve large networks (e.g., ResNet-50 (He et al.,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                676,
+                                321,
+                                689
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        676,
+                                        321,
+                                        689
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2016) or larger), with, however, little attention on",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                687,
+                                321,
+                                700
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        687,
+                                        321,
+                                        700
+                                    ],
+                                    "score": 1.0,
+                                    "content": "small networks. Empirically, we find that existing",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 64
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                698,
+                                505,
+                                711
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        698,
+                                        505,
+                                        711
+                                    ],
+                                    "score": 1.0,
+                                    "content": "techniques like contrastive learning do not work well on small networks. For instance, the linear probe",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 65
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                709,
+                                506,
+                                721
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        709,
+                                        407,
+                                        721
+                                    ],
+                                    "score": 1.0,
+                                    "content": "top-1 accuracy on ImageNet using MoCo-V2 (Chen et al., 2020c) is only",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        407,
+                                        710,
+                                        435,
+                                        720
+                                    ],
+                                    "score": 0.85,
+                                    "content": "3 6 . 3 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        435,
+                                        709,
+                                        506,
+                                        721
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with MobileNet-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 66
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                720,
+                                505,
+                                734
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        720,
+                                        505,
+                                        734
+                                    ],
+                                    "score": 1.0,
+                                    "content": "V3-Large (see Figure 1), which is much lower compared with its supervised training accuracy",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 67
+                        }
+                    ],
+                    "index": 42.5,
+                    "bbox_fs": [
+                        105,
+                        655,
+                        322,
+                        700
+                    ]
+                },
+                {
+                    "type": "image",
+                    "bbox": [
+                        330,
+                        435,
+                        502,
+                        564
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                330,
+                                435,
+                                502,
+                                564
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        330,
+                                        435,
+                                        502,
+                                        564
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                330,
+                                                435,
+                                                502,
+                                                564
+                                            ],
+                                            "score": 0.972,
+                                            "type": "image",
+                                            "image_path": "d4fb00f69f184e9fccdf211a9f298a6ff3e7571f5dfb94ff7b3d5393734ffc54.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 48.5,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        330,
+                                        435,
+                                        502,
+                                        447.9
+                                    ],
+                                    "spans": [],
+                                    "index": 44
+                                },
+                                {
+                                    "bbox": [
+                                        330,
+                                        447.9,
+                                        502,
+                                        460.79999999999995
+                                    ],
+                                    "spans": [],
+                                    "index": 45
+                                },
+                                {
+                                    "bbox": [
+                                        330,
+                                        460.79999999999995,
+                                        502,
+                                        473.69999999999993
+                                    ],
+                                    "spans": [],
+                                    "index": 46
+                                },
+                                {
+                                    "bbox": [
+                                        330,
+                                        473.69999999999993,
+                                        502,
+                                        486.5999999999999
+                                    ],
+                                    "spans": [],
+                                    "index": 47
+                                },
+                                {
+                                    "bbox": [
+                                        330,
+                                        486.5999999999999,
+                                        502,
+                                        499.4999999999999
+                                    ],
+                                    "spans": [],
+                                    "index": 48
+                                },
+                                {
+                                    "bbox": [
+                                        330,
+                                        499.4999999999999,
+                                        502,
+                                        512.3999999999999
+                                    ],
+                                    "spans": [],
+                                    "index": 49
+                                },
+                                {
+                                    "bbox": [
+                                        330,
+                                        512.3999999999999,
+                                        502,
+                                        525.2999999999998
+                                    ],
+                                    "spans": [],
+                                    "index": 50
+                                },
+                                {
+                                    "bbox": [
+                                        330,
+                                        525.2999999999998,
+                                        502,
+                                        538.1999999999998
+                                    ],
+                                    "spans": [],
+                                    "index": 51
+                                },
+                                {
+                                    "bbox": [
+                                        330,
+                                        538.1999999999998,
+                                        502,
+                                        551.0999999999998
+                                    ],
+                                    "spans": [],
+                                    "index": 52
+                                },
+                                {
+                                    "bbox": [
+                                        330,
+                                        551.0999999999998,
+                                        502,
+                                        563.9999999999998
+                                    ],
+                                    "spans": [],
+                                    "index": 53
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                328,
+                                569,
+                                505,
+                                671
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        327,
+                                        569,
+                                        506,
+                                        582
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                327,
+                                                569,
+                                                506,
+                                                582
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 1: SEED vs. MoCo-V2 (Chen et al.,",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 54
+                                },
+                                {
+                                    "bbox": [
+                                        327,
+                                        578,
+                                        507,
+                                        592
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                327,
+                                                578,
+                                                507,
+                                                592
+                                            ],
+                                            "score": 1.0,
+                                            "content": "2020c)) on ImageNet-1K linear probe accuracy.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 55
+                                },
+                                {
+                                    "bbox": [
+                                        327,
+                                        590,
+                                        505,
+                                        601
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                327,
+                                                590,
+                                                505,
+                                                601
+                                            ],
+                                            "score": 1.0,
+                                            "content": "The vertical axis is the top-1 accuracy and the",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 56
+                                },
+                                {
+                                    "bbox": [
+                                        327,
+                                        599,
+                                        506,
+                                        612
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                327,
+                                                599,
+                                                506,
+                                                612
+                                            ],
+                                            "score": 1.0,
+                                            "content": "horizontal axis is the number of learnable pa-",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 57
+                                },
+                                {
+                                    "bbox": [
+                                        327,
+                                        610,
+                                        506,
+                                        619
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                327,
+                                                610,
+                                                506,
+                                                619
+                                            ],
+                                            "score": 1.0,
+                                            "content": "rameters for different network architectures. Di-",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 58
+                                },
+                                {
+                                    "bbox": [
+                                        327,
+                                        620,
+                                        507,
+                                        631
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                327,
+                                                620,
+                                                507,
+                                                631
+                                            ],
+                                            "score": 1.0,
+                                            "content": "rectly applying self-supervised contrastive learn-",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 59
+                                },
+                                {
+                                    "bbox": [
+                                        327,
+                                        630,
+                                        506,
+                                        640
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                327,
+                                                630,
+                                                506,
+                                                640
+                                            ],
+                                            "score": 1.0,
+                                            "content": "ing (MoCo-V2) does not work well for smaller",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 60
+                                },
+                                {
+                                    "bbox": [
+                                        328,
+                                        641,
+                                        505,
+                                        651
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                328,
+                                                641,
+                                                505,
+                                                651
+                                            ],
+                                            "score": 1.0,
+                                            "content": "architectures, while our method (SEED) leads",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 61
+                                },
+                                {
+                                    "bbox": [
+                                        327,
+                                        651,
+                                        505,
+                                        661
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                327,
+                                                651,
+                                                505,
+                                                661
+                                            ],
+                                            "score": 1.0,
+                                            "content": "to dramatic performance boost. Details of the",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 62
+                                },
+                                {
+                                    "bbox": [
+                                        327,
+                                        661,
+                                        450,
+                                        671
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                327,
+                                                661,
+                                                450,
+                                                671
+                                            ],
+                                            "score": 1.0,
+                                            "content": "setting can be found in Section 4.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 63
+                                }
+                            ],
+                            "index": 58.5
+                        }
+                    ],
+                    "index": 53.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        699,
+                        505,
+                        732
+                    ],
+                    "lines": [],
+                    "index": 66,
+                    "bbox_fs": [
+                        105,
+                        698,
+                        506,
+                        734
+                    ],
+                    "lines_deleted": true
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        82,
+                        504,
+                        126
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                82,
+                                505,
+                                94
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        82,
+                                        133,
+                                        93
+                                    ],
+                                    "score": 0.9,
+                                    "content": "7 5 . 2 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        133,
+                                        82,
+                                        362,
+                                        94
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(Howard et al., 2019). For EfficientNet-B0, the accuracy is",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        362,
+                                        83,
+                                        389,
+                                        93
+                                    ],
+                                    "score": 0.87,
+                                    "content": "4 2 . 2 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        389,
+                                        82,
+                                        505,
+                                        94
+                                    ],
+                                    "score": 1.0,
+                                    "content": "compared with its supervised",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                93,
+                                505,
+                                105
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        93,
+                                        178,
+                                        105
+                                    ],
+                                    "score": 1.0,
+                                    "content": "training accuracy",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        178,
+                                        94,
+                                        206,
+                                        105
+                                    ],
+                                    "score": 0.85,
+                                    "content": "7 7 . 1 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        206,
+                                        93,
+                                        505,
+                                        105
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(Tan & Le, 2019). We conjecture that this is because smaller models with",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                104,
+                                506,
+                                118
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        104,
+                                        506,
+                                        118
+                                    ],
+                                    "score": 1.0,
+                                    "content": "fewer parameters cannot effectively learn instance level discriminative representation with large",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                116,
+                                171,
+                                127
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        116,
+                                        171,
+                                        127
+                                    ],
+                                    "score": 1.0,
+                                    "content": "amount of data.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        }
+                    ],
+                    "index": 1.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        132,
+                        505,
+                        286
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                132,
+                                506,
+                                144
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        132,
+                                        506,
+                                        144
+                                    ],
+                                    "score": 1.0,
+                                    "content": "To address this challenge, we inject knowledge distillation (KD) (Bucilua et al. ˇ , 2006; Hinton et al.,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                142,
+                                506,
+                                156
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        142,
+                                        506,
+                                        156
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2015) into self-supervised learning and propose self-supervised distillation (dubbed as SEED) as",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                154,
+                                505,
+                                166
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        154,
+                                        505,
+                                        166
+                                    ],
+                                    "score": 1.0,
+                                    "content": "a new learning paradigm. That is, train the larger, and distill to the smaller both in self-supervised",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                165,
+                                505,
+                                178
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        165,
+                                        505,
+                                        178
+                                    ],
+                                    "score": 1.0,
+                                    "content": "manner. Instead of directly conducting self-supervised training on a smaller model, SEED first trains",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                176,
+                                506,
+                                188
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        176,
+                                        506,
+                                        188
+                                    ],
+                                    "score": 1.0,
+                                    "content": "a large model (as the teacher) in a self-supervised way, and then distills the knowledge to the smaller",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                187,
+                                504,
+                                199
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        187,
+                                        504,
+                                        199
+                                    ],
+                                    "score": 1.0,
+                                    "content": "model (as the student). Note that the conventional distillation is for supervised learning, while the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                197,
+                                506,
+                                210
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        197,
+                                        506,
+                                        210
+                                    ],
+                                    "score": 1.0,
+                                    "content": "distillation here is in the self-supervised setting without any labeled data. Supervised distillation can",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                208,
+                                506,
+                                222
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        208,
+                                        506,
+                                        222
+                                    ],
+                                    "score": 1.0,
+                                    "content": "be formulated as training a student to mimic the probability mass function over classes predicted by a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                221,
+                                504,
+                                232
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        221,
+                                        504,
+                                        232
+                                    ],
+                                    "score": 1.0,
+                                    "content": "teacher model. In unsupervised knowledge distillation setting, however, the distribution over classes",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                230,
+                                505,
+                                243
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        230,
+                                        505,
+                                        243
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is not directly attainable. Therefore, we propose a simple yet effective self-supervised distillation",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                241,
+                                506,
+                                254
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        241,
+                                        506,
+                                        254
+                                    ],
+                                    "score": 1.0,
+                                    "content": "method. Similar to (He et al., 2020; Wu et al., 2018), we maintain a queue of data samples. Given an",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                253,
+                                505,
+                                264
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        253,
+                                        505,
+                                        264
+                                    ],
+                                    "score": 1.0,
+                                    "content": "instance, we first use the teacher network to obtain its similarity scores with all the data samples in",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                264,
+                                505,
+                                276
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        264,
+                                        505,
+                                        276
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the queue as well as the instance itself. Then the student encoder is trained to mimic the similarity",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                275,
+                                369,
+                                287
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        275,
+                                        369,
+                                        287
+                                    ],
+                                    "score": 1.0,
+                                    "content": "score distribution inferred by the teacher over these data samples.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        }
+                    ],
+                    "index": 10.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        291,
+                        505,
+                        347
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                291,
+                                506,
+                                303
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        291,
+                                        506,
+                                        303
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The simplicity and flexibility that SEED brings are self-evident. 1) It does not require any cluster-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                303,
+                                506,
+                                315
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        303,
+                                        506,
+                                        315
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ing/prototypical computing procedure to retrieve the pseudo-labels or latent classes. 2) The teacher",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                313,
+                                506,
+                                326
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        313,
+                                        506,
+                                        326
+                                    ],
+                                    "score": 1.0,
+                                    "content": "model can be pre-trained with any advanced SSL approach, e.g., MoCo-V2 (Chen et al., 2020c),",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                323,
+                                505,
+                                338
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        323,
+                                        505,
+                                        338
+                                    ],
+                                    "score": 1.0,
+                                    "content": "SimCLR (Chen et al., 2020a), SWAV (Caron et al., 2020). 3) The knowledge can be distilled to any",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                335,
+                                433,
+                                348
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        335,
+                                        433,
+                                        348
+                                    ],
+                                    "score": 1.0,
+                                    "content": "target small networks (either shallower, thinner, or totally different architectures).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        }
+                    ],
+                    "index": 20
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        352,
+                        505,
+                        418
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                352,
+                                505,
+                                364
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        352,
+                                        505,
+                                        364
+                                    ],
+                                    "score": 1.0,
+                                    "content": "To demonstrate the effectiveness, we comprehensively evaluate the learned representations on series",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                362,
+                                505,
+                                376
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        362,
+                                        505,
+                                        376
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of downstream tasks, e.g., fully/semi-supervised classification, object detection, and also assess the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                374,
+                                506,
+                                387
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        374,
+                                        506,
+                                        387
+                                    ],
+                                    "score": 1.0,
+                                    "content": "transferability to other domains. For example, on ImageNet-1k dataset, SEED improves the linear",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                384,
+                                505,
+                                398
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        384,
+                                        274,
+                                        398
+                                    ],
+                                    "score": 1.0,
+                                    "content": "probe accuracy of EfficientNet-B0 from",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        274,
+                                        385,
+                                        302,
+                                        396
+                                    ],
+                                    "score": 0.88,
+                                    "content": "4 2 . 2 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        302,
+                                        384,
+                                        314,
+                                        398
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        315,
+                                        385,
+                                        343,
+                                        396
+                                    ],
+                                    "score": 0.87,
+                                    "content": "6 7 . 6 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        343,
+                                        384,
+                                        396,
+                                        398
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(a gain over",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        397,
+                                        385,
+                                        417,
+                                        396
+                                    ],
+                                    "score": 0.86,
+                                    "content": "2 5 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        417,
+                                        384,
+                                        505,
+                                        398
+                                    ],
+                                    "score": 1.0,
+                                    "content": "), and MobileNet-V3",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                396,
+                                506,
+                                409
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        396,
+                                        128,
+                                        409
+                                    ],
+                                    "score": 1.0,
+                                    "content": "from",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        128,
+                                        397,
+                                        155,
+                                        407
+                                    ],
+                                    "score": 0.86,
+                                    "content": "3 6 . 3 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        155,
+                                        396,
+                                        166,
+                                        409
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        167,
+                                        397,
+                                        194,
+                                        407
+                                    ],
+                                    "score": 0.85,
+                                    "content": "6 8 . 2 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        194,
+                                        396,
+                                        243,
+                                        409
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(a gain over",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        244,
+                                        397,
+                                        263,
+                                        407
+                                    ],
+                                    "score": 0.84,
+                                    "content": "3 1 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        264,
+                                        396,
+                                        506,
+                                        409
+                                    ],
+                                    "score": 1.0,
+                                    "content": ") compared to MoCo-V2 baselines, as shown in Figure 1 and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                407,
+                                149,
+                                418
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        407,
+                                        149,
+                                        418
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Section 4.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        }
+                    ],
+                    "index": 25.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        424,
+                        303,
+                        435
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                423,
+                                305,
+                                437
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        423,
+                                        305,
+                                        437
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Our contributions can be summarized as follows:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        }
+                    ],
+                    "index": 29
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        132,
+                        445,
+                        505,
+                        548
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                132,
+                                444,
+                                505,
+                                458
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        132,
+                                        444,
+                                        505,
+                                        458
+                                    ],
+                                    "score": 1.0,
+                                    "content": "• We are the first to address the problem of self-supervised visual representation learning for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                457,
+                                200,
+                                467
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        457,
+                                        200,
+                                        467
+                                    ],
+                                    "score": 1.0,
+                                    "content": "small models.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                138,
+                                471,
+                                505,
+                                484
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        138,
+                                        471,
+                                        505,
+                                        484
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We propose a self-supervised distillation (SEED) technique to transfer knowledge from a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                142,
+                                483,
+                                362,
+                                495
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        142,
+                                        483,
+                                        362,
+                                        495
+                                    ],
+                                    "score": 1.0,
+                                    "content": "large model to a small model without any labeled data.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                140,
+                                498,
+                                506,
+                                511
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        140,
+                                        498,
+                                        506,
+                                        511
+                                    ],
+                                    "score": 1.0,
+                                    "content": "With the proposed distillation technique (SEED), we significantly improve the state-of-the-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                510,
+                                295,
+                                521
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        510,
+                                        295,
+                                        521
+                                    ],
+                                    "score": 1.0,
+                                    "content": "art SSL performance on small models.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                131,
+                                523,
+                                505,
+                                538
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        131,
+                                        523,
+                                        505,
+                                        538
+                                    ],
+                                    "score": 1.0,
+                                    "content": "• We exhaustively compare a variety of distillation strategies to show the validity of SEED",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                534,
+                                239,
+                                550
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        534,
+                                        239,
+                                        550
+                                    ],
+                                    "score": 1.0,
+                                    "content": "under multiple settings.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        }
+                    ],
+                    "index": 33.5
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        563,
+                        211,
+                        576
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                562,
+                                213,
+                                579
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        562,
+                                        213,
+                                        579
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2 RELATED WORK",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        }
+                    ],
+                    "index": 38
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        589,
+                        505,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                588,
+                                505,
+                                601
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        588,
+                                        505,
+                                        601
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Among the recent literature in self-supervised learning, contrastive based approaches show prominent",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                600,
+                                505,
+                                612
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        600,
+                                        505,
+                                        612
+                                    ],
+                                    "score": 1.0,
+                                    "content": "results on downstream tasks. Majority of the techniques along this direction are stemming from",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                610,
+                                505,
+                                623
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        610,
+                                        505,
+                                        623
+                                    ],
+                                    "score": 1.0,
+                                    "content": "noise-contrastive estimation (Gutmann & Hyvärinen, 2010) where the latent distribution is estimated",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                621,
+                                505,
+                                634
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        621,
+                                        505,
+                                        634
+                                    ],
+                                    "score": 1.0,
+                                    "content": "by contrasting with randomly or artificially generated noises. Oord et al. (2018) first proposed",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                632,
+                                506,
+                                646
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        632,
+                                        506,
+                                        646
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Info-NCE to learn image representations by predicting the future using an auto-regressive model for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                644,
+                                505,
+                                656
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        644,
+                                        505,
+                                        656
+                                    ],
+                                    "score": 1.0,
+                                    "content": "unsupervised learning. Follow-up works include improving the efficiency (Hénaff et al., 2019), and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                655,
+                                505,
+                                667
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        655,
+                                        505,
+                                        667
+                                    ],
+                                    "score": 1.0,
+                                    "content": "using multi-view as positive samples (Tian et al., 2019b). As these approaches can only have the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                665,
+                                505,
+                                679
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        665,
+                                        505,
+                                        679
+                                    ],
+                                    "score": 1.0,
+                                    "content": "access to limited negative instances, Wu et al. (2018) designed a memory-bank to store the previously",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                677,
+                                505,
+                                689
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        677,
+                                        505,
+                                        689
+                                    ],
+                                    "score": 1.0,
+                                    "content": "seen random representations as negative samples, and treat each of them as independent categories",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 47
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                686,
+                                505,
+                                701
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        686,
+                                        505,
+                                        701
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(instance discrimination). However, this approach also comes with a deficiency that the previously",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 48
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                698,
+                                505,
+                                712
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        698,
+                                        505,
+                                        712
+                                    ],
+                                    "score": 1.0,
+                                    "content": "stored vectors are inconsistent with the recently computed representations during the earlier stage",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 49
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                708,
+                                505,
+                                724
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        708,
+                                        505,
+                                        724
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of pre-training. Chen et al. (2020a) mitigate this issue by sampling negative samples from a large",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 50
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                720,
+                                505,
+                                734
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        720,
+                                        505,
+                                        734
+                                    ],
+                                    "score": 1.0,
+                                    "content": "batch. Concurrently, He et al. (2020) improve the memory-bank based method and propose to use",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 51
+                        }
+                    ],
+                    "index": 45
+                }
+            ],
+            "page_idx": 1,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        108,
+                        27,
+                        292,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                25,
+                                293,
+                                39
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        25,
+                                        293,
+                                        39
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Published as a conference paper at ICLR 2021",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        302,
+                        751,
+                        309,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                301,
+                                750,
+                                310,
+                                763
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        301,
+                                        750,
+                                        310,
+                                        763
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        82,
+                        504,
+                        126
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                82,
+                                505,
+                                94
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        82,
+                                        133,
+                                        93
+                                    ],
+                                    "score": 0.9,
+                                    "content": "7 5 . 2 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        133,
+                                        82,
+                                        362,
+                                        94
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(Howard et al., 2019). For EfficientNet-B0, the accuracy is",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        362,
+                                        83,
+                                        389,
+                                        93
+                                    ],
+                                    "score": 0.87,
+                                    "content": "4 2 . 2 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        389,
+                                        82,
+                                        505,
+                                        94
+                                    ],
+                                    "score": 1.0,
+                                    "content": "compared with its supervised",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                93,
+                                505,
+                                105
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        93,
+                                        178,
+                                        105
+                                    ],
+                                    "score": 1.0,
+                                    "content": "training accuracy",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        178,
+                                        94,
+                                        206,
+                                        105
+                                    ],
+                                    "score": 0.85,
+                                    "content": "7 7 . 1 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        206,
+                                        93,
+                                        505,
+                                        105
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(Tan & Le, 2019). We conjecture that this is because smaller models with",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                104,
+                                506,
+                                118
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        104,
+                                        506,
+                                        118
+                                    ],
+                                    "score": 1.0,
+                                    "content": "fewer parameters cannot effectively learn instance level discriminative representation with large",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                116,
+                                171,
+                                127
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        116,
+                                        171,
+                                        127
+                                    ],
+                                    "score": 1.0,
+                                    "content": "amount of data.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        }
+                    ],
+                    "index": 1.5,
+                    "bbox_fs": [
+                        105,
+                        82,
+                        506,
+                        127
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        132,
+                        505,
+                        286
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                132,
+                                506,
+                                144
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        132,
+                                        506,
+                                        144
+                                    ],
+                                    "score": 1.0,
+                                    "content": "To address this challenge, we inject knowledge distillation (KD) (Bucilua et al. ˇ , 2006; Hinton et al.,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                142,
+                                506,
+                                156
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        142,
+                                        506,
+                                        156
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2015) into self-supervised learning and propose self-supervised distillation (dubbed as SEED) as",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                154,
+                                505,
+                                166
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        154,
+                                        505,
+                                        166
+                                    ],
+                                    "score": 1.0,
+                                    "content": "a new learning paradigm. That is, train the larger, and distill to the smaller both in self-supervised",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                165,
+                                505,
+                                178
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        165,
+                                        505,
+                                        178
+                                    ],
+                                    "score": 1.0,
+                                    "content": "manner. Instead of directly conducting self-supervised training on a smaller model, SEED first trains",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                176,
+                                506,
+                                188
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        176,
+                                        506,
+                                        188
+                                    ],
+                                    "score": 1.0,
+                                    "content": "a large model (as the teacher) in a self-supervised way, and then distills the knowledge to the smaller",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                187,
+                                504,
+                                199
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        187,
+                                        504,
+                                        199
+                                    ],
+                                    "score": 1.0,
+                                    "content": "model (as the student). Note that the conventional distillation is for supervised learning, while the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                197,
+                                506,
+                                210
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        197,
+                                        506,
+                                        210
+                                    ],
+                                    "score": 1.0,
+                                    "content": "distillation here is in the self-supervised setting without any labeled data. Supervised distillation can",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                208,
+                                506,
+                                222
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        208,
+                                        506,
+                                        222
+                                    ],
+                                    "score": 1.0,
+                                    "content": "be formulated as training a student to mimic the probability mass function over classes predicted by a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                221,
+                                504,
+                                232
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        221,
+                                        504,
+                                        232
+                                    ],
+                                    "score": 1.0,
+                                    "content": "teacher model. In unsupervised knowledge distillation setting, however, the distribution over classes",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                230,
+                                505,
+                                243
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        230,
+                                        505,
+                                        243
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is not directly attainable. Therefore, we propose a simple yet effective self-supervised distillation",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                241,
+                                506,
+                                254
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        241,
+                                        506,
+                                        254
+                                    ],
+                                    "score": 1.0,
+                                    "content": "method. Similar to (He et al., 2020; Wu et al., 2018), we maintain a queue of data samples. Given an",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                253,
+                                505,
+                                264
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        253,
+                                        505,
+                                        264
+                                    ],
+                                    "score": 1.0,
+                                    "content": "instance, we first use the teacher network to obtain its similarity scores with all the data samples in",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                264,
+                                505,
+                                276
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        264,
+                                        505,
+                                        276
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the queue as well as the instance itself. Then the student encoder is trained to mimic the similarity",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                275,
+                                369,
+                                287
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        275,
+                                        369,
+                                        287
+                                    ],
+                                    "score": 1.0,
+                                    "content": "score distribution inferred by the teacher over these data samples.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        }
+                    ],
+                    "index": 10.5,
+                    "bbox_fs": [
+                        105,
+                        132,
+                        506,
+                        287
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        291,
+                        505,
+                        347
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                291,
+                                506,
+                                303
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        291,
+                                        506,
+                                        303
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The simplicity and flexibility that SEED brings are self-evident. 1) It does not require any cluster-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                303,
+                                506,
+                                315
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        303,
+                                        506,
+                                        315
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ing/prototypical computing procedure to retrieve the pseudo-labels or latent classes. 2) The teacher",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                313,
+                                506,
+                                326
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        313,
+                                        506,
+                                        326
+                                    ],
+                                    "score": 1.0,
+                                    "content": "model can be pre-trained with any advanced SSL approach, e.g., MoCo-V2 (Chen et al., 2020c),",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                323,
+                                505,
+                                338
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        323,
+                                        505,
+                                        338
+                                    ],
+                                    "score": 1.0,
+                                    "content": "SimCLR (Chen et al., 2020a), SWAV (Caron et al., 2020). 3) The knowledge can be distilled to any",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                335,
+                                433,
+                                348
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        335,
+                                        433,
+                                        348
+                                    ],
+                                    "score": 1.0,
+                                    "content": "target small networks (either shallower, thinner, or totally different architectures).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        }
+                    ],
+                    "index": 20,
+                    "bbox_fs": [
+                        105,
+                        291,
+                        506,
+                        348
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        352,
+                        505,
+                        418
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                352,
+                                505,
+                                364
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        352,
+                                        505,
+                                        364
+                                    ],
+                                    "score": 1.0,
+                                    "content": "To demonstrate the effectiveness, we comprehensively evaluate the learned representations on series",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                362,
+                                505,
+                                376
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        362,
+                                        505,
+                                        376
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of downstream tasks, e.g., fully/semi-supervised classification, object detection, and also assess the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                374,
+                                506,
+                                387
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        374,
+                                        506,
+                                        387
+                                    ],
+                                    "score": 1.0,
+                                    "content": "transferability to other domains. For example, on ImageNet-1k dataset, SEED improves the linear",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                384,
+                                505,
+                                398
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        384,
+                                        274,
+                                        398
+                                    ],
+                                    "score": 1.0,
+                                    "content": "probe accuracy of EfficientNet-B0 from",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        274,
+                                        385,
+                                        302,
+                                        396
+                                    ],
+                                    "score": 0.88,
+                                    "content": "4 2 . 2 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        302,
+                                        384,
+                                        314,
+                                        398
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        315,
+                                        385,
+                                        343,
+                                        396
+                                    ],
+                                    "score": 0.87,
+                                    "content": "6 7 . 6 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        343,
+                                        384,
+                                        396,
+                                        398
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(a gain over",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        397,
+                                        385,
+                                        417,
+                                        396
+                                    ],
+                                    "score": 0.86,
+                                    "content": "2 5 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        417,
+                                        384,
+                                        505,
+                                        398
+                                    ],
+                                    "score": 1.0,
+                                    "content": "), and MobileNet-V3",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                396,
+                                506,
+                                409
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        396,
+                                        128,
+                                        409
+                                    ],
+                                    "score": 1.0,
+                                    "content": "from",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        128,
+                                        397,
+                                        155,
+                                        407
+                                    ],
+                                    "score": 0.86,
+                                    "content": "3 6 . 3 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        155,
+                                        396,
+                                        166,
+                                        409
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        167,
+                                        397,
+                                        194,
+                                        407
+                                    ],
+                                    "score": 0.85,
+                                    "content": "6 8 . 2 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        194,
+                                        396,
+                                        243,
+                                        409
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(a gain over",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        244,
+                                        397,
+                                        263,
+                                        407
+                                    ],
+                                    "score": 0.84,
+                                    "content": "3 1 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        264,
+                                        396,
+                                        506,
+                                        409
+                                    ],
+                                    "score": 1.0,
+                                    "content": ") compared to MoCo-V2 baselines, as shown in Figure 1 and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                407,
+                                149,
+                                418
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        407,
+                                        149,
+                                        418
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Section 4.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        }
+                    ],
+                    "index": 25.5,
+                    "bbox_fs": [
+                        105,
+                        352,
+                        506,
+                        418
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        424,
+                        303,
+                        435
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                423,
+                                305,
+                                437
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        423,
+                                        305,
+                                        437
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Our contributions can be summarized as follows:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        }
+                    ],
+                    "index": 29,
+                    "bbox_fs": [
+                        106,
+                        423,
+                        305,
+                        437
+                    ]
+                },
+                {
+                    "type": "list",
+                    "bbox": [
+                        132,
+                        445,
+                        505,
+                        548
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                132,
+                                444,
+                                505,
+                                458
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        132,
+                                        444,
+                                        505,
+                                        458
+                                    ],
+                                    "score": 1.0,
+                                    "content": "• We are the first to address the problem of self-supervised visual representation learning for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                457,
+                                200,
+                                467
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        457,
+                                        200,
+                                        467
+                                    ],
+                                    "score": 1.0,
+                                    "content": "small models.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                138,
+                                471,
+                                505,
+                                484
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        138,
+                                        471,
+                                        505,
+                                        484
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We propose a self-supervised distillation (SEED) technique to transfer knowledge from a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                142,
+                                483,
+                                362,
+                                495
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        142,
+                                        483,
+                                        362,
+                                        495
+                                    ],
+                                    "score": 1.0,
+                                    "content": "large model to a small model without any labeled data.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                140,
+                                498,
+                                506,
+                                511
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        140,
+                                        498,
+                                        506,
+                                        511
+                                    ],
+                                    "score": 1.0,
+                                    "content": "With the proposed distillation technique (SEED), we significantly improve the state-of-the-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                510,
+                                295,
+                                521
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        510,
+                                        295,
+                                        521
+                                    ],
+                                    "score": 1.0,
+                                    "content": "art SSL performance on small models.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                131,
+                                523,
+                                505,
+                                538
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        131,
+                                        523,
+                                        505,
+                                        538
+                                    ],
+                                    "score": 1.0,
+                                    "content": "• We exhaustively compare a variety of distillation strategies to show the validity of SEED",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                534,
+                                239,
+                                550
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        534,
+                                        239,
+                                        550
+                                    ],
+                                    "score": 1.0,
+                                    "content": "under multiple settings.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37,
+                            "is_list_end_line": true
+                        }
+                    ],
+                    "index": 33.5,
+                    "bbox_fs": [
+                        131,
+                        444,
+                        506,
+                        550
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        563,
+                        211,
+                        576
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                562,
+                                213,
+                                579
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        562,
+                                        213,
+                                        579
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2 RELATED WORK",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        }
+                    ],
+                    "index": 38
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        589,
+                        505,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                588,
+                                505,
+                                601
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        588,
+                                        505,
+                                        601
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Among the recent literature in self-supervised learning, contrastive based approaches show prominent",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                600,
+                                505,
+                                612
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        600,
+                                        505,
+                                        612
+                                    ],
+                                    "score": 1.0,
+                                    "content": "results on downstream tasks. Majority of the techniques along this direction are stemming from",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                610,
+                                505,
+                                623
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        610,
+                                        505,
+                                        623
+                                    ],
+                                    "score": 1.0,
+                                    "content": "noise-contrastive estimation (Gutmann & Hyvärinen, 2010) where the latent distribution is estimated",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                621,
+                                505,
+                                634
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        621,
+                                        505,
+                                        634
+                                    ],
+                                    "score": 1.0,
+                                    "content": "by contrasting with randomly or artificially generated noises. Oord et al. (2018) first proposed",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                632,
+                                506,
+                                646
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        632,
+                                        506,
+                                        646
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Info-NCE to learn image representations by predicting the future using an auto-regressive model for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                644,
+                                505,
+                                656
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        644,
+                                        505,
+                                        656
+                                    ],
+                                    "score": 1.0,
+                                    "content": "unsupervised learning. Follow-up works include improving the efficiency (Hénaff et al., 2019), and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                655,
+                                505,
+                                667
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        655,
+                                        505,
+                                        667
+                                    ],
+                                    "score": 1.0,
+                                    "content": "using multi-view as positive samples (Tian et al., 2019b). As these approaches can only have the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                665,
+                                505,
+                                679
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        665,
+                                        505,
+                                        679
+                                    ],
+                                    "score": 1.0,
+                                    "content": "access to limited negative instances, Wu et al. (2018) designed a memory-bank to store the previously",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                677,
+                                505,
+                                689
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        677,
+                                        505,
+                                        689
+                                    ],
+                                    "score": 1.0,
+                                    "content": "seen random representations as negative samples, and treat each of them as independent categories",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 47
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                686,
+                                505,
+                                701
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        686,
+                                        505,
+                                        701
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(instance discrimination). However, this approach also comes with a deficiency that the previously",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 48
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                698,
+                                505,
+                                712
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        698,
+                                        505,
+                                        712
+                                    ],
+                                    "score": 1.0,
+                                    "content": "stored vectors are inconsistent with the recently computed representations during the earlier stage",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 49
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                708,
+                                505,
+                                724
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        708,
+                                        505,
+                                        724
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of pre-training. Chen et al. (2020a) mitigate this issue by sampling negative samples from a large",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 50
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                720,
+                                505,
+                                734
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        720,
+                                        505,
+                                        734
+                                    ],
+                                    "score": 1.0,
+                                    "content": "batch. Concurrently, He et al. (2020) improve the memory-bank based method and propose to use",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 51
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                82,
+                                506,
+                                95
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        82,
+                                        506,
+                                        95
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the momentum updated encoder for the remission of representation inconsistency. Other techniques",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                93,
+                                506,
+                                105
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        93,
+                                        506,
+                                        105
+                                    ],
+                                    "score": 1.0,
+                                    "content": "include Misra & Maaten (2020) that combines the pretext-invariant objective loss with contrastive",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                104,
+                                505,
+                                118
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        104,
+                                        505,
+                                        118
+                                    ],
+                                    "score": 1.0,
+                                    "content": "learning, and Wang & Isola (2020) that decomposes contrastive loss into alignment and uniformity",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                114,
+                                165,
+                                128
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        114,
+                                        165,
+                                        128
+                                    ],
+                                    "score": 1.0,
+                                    "content": "objectiveness.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 3
+                        }
+                    ],
+                    "index": 45,
+                    "bbox_fs": [
+                        105,
+                        588,
+                        506,
+                        734
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        82,
+                        504,
+                        126
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                82,
+                                506,
+                                95
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        82,
+                                        506,
+                                        95
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the momentum updated encoder for the remission of representation inconsistency. Other techniques",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                93,
+                                506,
+                                105
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        93,
+                                        506,
+                                        105
+                                    ],
+                                    "score": 1.0,
+                                    "content": "include Misra & Maaten (2020) that combines the pretext-invariant objective loss with contrastive",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                104,
+                                505,
+                                118
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        104,
+                                        505,
+                                        118
+                                    ],
+                                    "score": 1.0,
+                                    "content": "learning, and Wang & Isola (2020) that decomposes contrastive loss into alignment and uniformity",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                114,
+                                165,
+                                128
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        114,
+                                        165,
+                                        128
+                                    ],
+                                    "score": 1.0,
+                                    "content": "objectiveness.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        }
+                    ],
+                    "index": 1.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        131,
+                        505,
+                        384
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                132,
+                                505,
+                                145
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        132,
+                                        505,
+                                        145
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Knowledge distillation (Hinton et al., 2015) aims to transfer knowledge from a cumbersome model",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                143,
+                                505,
+                                156
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        143,
+                                        505,
+                                        156
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to a smaller one without losing too much generalization power, which is also well investigated in",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                154,
+                                505,
+                                166
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        154,
+                                        505,
+                                        166
+                                    ],
+                                    "score": 1.0,
+                                    "content": "model compression (Bucilua et al. ˇ , 2006). Instead of mimicking the teacher’s output logit, attention",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                165,
+                                506,
+                                177
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        165,
+                                        506,
+                                        177
+                                    ],
+                                    "score": 1.0,
+                                    "content": "transfer (Zagoruyko & Komodakis, 2016) formulates knowledge distillation on attention maps.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                176,
+                                506,
+                                188
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        176,
+                                        506,
+                                        188
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Similarly, works in (Ahn et al., 2019; Yim et al., 2017; Koratana et al., 2019; Huang & Wang, 2017)",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                186,
+                                506,
+                                200
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        186,
+                                        506,
+                                        200
+                                    ],
+                                    "score": 1.0,
+                                    "content": "have utilized different learning objectives including consistency on feature maps, consistency on",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                198,
+                                505,
+                                209
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        198,
+                                        505,
+                                        209
+                                    ],
+                                    "score": 1.0,
+                                    "content": "probability mass function, and maximizing the mutual information. CRD (Tian et al., 2019a), which is",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                209,
+                                505,
+                                221
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        209,
+                                        505,
+                                        221
+                                    ],
+                                    "score": 1.0,
+                                    "content": "derived from CMC (Tian et al., 2019b), optimizes the student network by a similar objective to Oord",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                219,
+                                505,
+                                232
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        219,
+                                        505,
+                                        232
+                                    ],
+                                    "score": 1.0,
+                                    "content": "et al. (2018) using a derived lower bound on mutual information. However, the aforementioned efforts",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                231,
+                                506,
+                                243
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        231,
+                                        506,
+                                        243
+                                    ],
+                                    "score": 1.0,
+                                    "content": "all focus on task-specific distillation (e.g., image classification) during the fine-tuning phase rather",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                241,
+                                505,
+                                254
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        241,
+                                        505,
+                                        254
+                                    ],
+                                    "score": 1.0,
+                                    "content": "than a task-agnostic distillation in the pre-training phase for the representation learning. Several",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                253,
+                                506,
+                                266
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        253,
+                                        506,
+                                        266
+                                    ],
+                                    "score": 1.0,
+                                    "content": "works on natural language pre-training proposed to leverage knowledge distillation for a smaller yet",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                263,
+                                506,
+                                276
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        263,
+                                        506,
+                                        276
+                                    ],
+                                    "score": 1.0,
+                                    "content": "stronger small models. For instances, DistillBert (Sanh et al., 2019), TinyBert (Jiao et al., 2019), and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                275,
+                                504,
+                                286
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        275,
+                                        504,
+                                        286
+                                    ],
+                                    "score": 1.0,
+                                    "content": "MobileBert (Sun et al., 2020), have used knowledge distillation for model compression and shown",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                285,
+                                506,
+                                298
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        285,
+                                        506,
+                                        298
+                                    ],
+                                    "score": 1.0,
+                                    "content": "their validity on multiple downstream tasks. Similar works also emphasize the value of smaller and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                297,
+                                505,
+                                309
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        297,
+                                        505,
+                                        309
+                                    ],
+                                    "score": 1.0,
+                                    "content": "faster models for language representation learning by leveraging knowledge distillation (Turc et al.,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                307,
+                                506,
+                                320
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        307,
+                                        506,
+                                        320
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2019; Sun et al., 2019). These works all demonstrate the effectiveness of knowledge distillation for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                319,
+                                505,
+                                330
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        319,
+                                        505,
+                                        330
+                                    ],
+                                    "score": 1.0,
+                                    "content": "language representation learning in small models, while are not extended to the pre-training for visual",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                329,
+                                505,
+                                342
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        329,
+                                        505,
+                                        342
+                                    ],
+                                    "score": 1.0,
+                                    "content": "representations. Notably, a recent concurrent work CompRess (Abbasi Koohpayegani et al., 2020)",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                340,
+                                505,
+                                353
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        340,
+                                        505,
+                                        353
+                                    ],
+                                    "score": 1.0,
+                                    "content": "also point out the importance to develop better SSL method for smaller models. SEED closely relates",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                351,
+                                505,
+                                364
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        351,
+                                        505,
+                                        364
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to the above techniques but aims to facilitate visual representation learning during pre-training",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                363,
+                                504,
+                                374
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        363,
+                                        504,
+                                        374
+                                    ],
+                                    "score": 1.0,
+                                    "content": "phase using distillation technique for small models, which as far as we know has not yet been",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                374,
+                                159,
+                                385
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        374,
+                                        159,
+                                        385
+                                    ],
+                                    "score": 1.0,
+                                    "content": "investigated.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        }
+                    ],
+                    "index": 15
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        401,
+                        173,
+                        413
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                398,
+                                175,
+                                416
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        398,
+                                        175,
+                                        416
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3 METHOD",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        }
+                    ],
+                    "index": 27
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        425,
+                        330,
+                        437
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                424,
+                                331,
+                                438
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        424,
+                                        331,
+                                        438
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3.1 PRELIMINARY ON KNOWLEDGE DISTILLATION",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        }
+                    ],
+                    "index": 28
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        446,
+                        505,
+                        479
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                445,
+                                505,
+                                457
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        445,
+                                        505,
+                                        457
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Knowledge distillation (Hinton et al., 2015; Bucilua et al.ˇ , 2006) is an effective technique to transfer",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                457,
+                                505,
+                                469
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        457,
+                                        505,
+                                        469
+                                    ],
+                                    "score": 1.0,
+                                    "content": "knowledge from a strong teacher network to a target student network. The training task can be",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                468,
+                                274,
+                                480
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        468,
+                                        274,
+                                        480
+                                    ],
+                                    "score": 1.0,
+                                    "content": "generalized as the following formulation:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        }
+                    ],
+                    "index": 30
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        193,
+                        483,
+                        416,
+                        518
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                193,
+                                483,
+                                416,
+                                518
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        193,
+                                        483,
+                                        416,
+                                        518
+                                    ],
+                                    "score": 0.94,
+                                    "content": "\\hat { \\theta } _ { S } = \\underset { \\theta _ { S } } { \\arg \\operatorname* { m i n } } \\sum _ { i } ^ { N } \\mathcal { L } _ { \\mathrm { s u p } } ( \\mathbf { x } _ { i } , \\theta _ { S } , y _ { i } ) + \\mathcal { L } _ { \\mathrm { d i s t i l l } } ( \\mathbf { x } _ { i } , \\theta _ { S } , \\theta _ { T } ) ,",
+                                    "type": "interline_equation",
+                                    "image_path": "f6d588e03d3db3f013e786c6fff9679cd8aa0b9b81a92bf569f2acbe7a654360.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 32.5,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                193,
+                                483,
+                                416,
+                                500.5
+                            ],
+                            "spans": [],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                193,
+                                500.5,
+                                416,
+                                518.0
+                            ],
+                            "spans": [],
+                            "index": 33
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        521,
+                        505,
+                        643
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                521,
+                                506,
+                                535
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        521,
+                                        133,
+                                        535
+                                    ],
+                                    "score": 1.0,
+                                    "content": "where",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        134,
+                                        523,
+                                        145,
+                                        533
+                                    ],
+                                    "score": 0.84,
+                                    "content": "\\mathbf { x } _ { i }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        145,
+                                        521,
+                                        199,
+                                        535
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is an image,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        199,
+                                        523,
+                                        209,
+                                        533
+                                    ],
+                                    "score": 0.84,
+                                    "content": "y _ { i }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        209,
+                                        521,
+                                        345,
+                                        535
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is the corresponding annotation,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        346,
+                                        522,
+                                        358,
+                                        533
+                                    ],
+                                    "score": 0.88,
+                                    "content": "\\theta _ { S }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        358,
+                                        521,
+                                        506,
+                                        535
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is the parameter set for the student",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                532,
+                                506,
+                                546
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        532,
+                                        163,
+                                        546
+                                    ],
+                                    "score": 1.0,
+                                    "content": "network, and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        163,
+                                        533,
+                                        175,
+                                        544
+                                    ],
+                                    "score": 0.88,
+                                    "content": "\\theta _ { T }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        176,
+                                        532,
+                                        357,
+                                        546
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is the set for the teacher network. The loss",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        357,
+                                        533,
+                                        376,
+                                        545
+                                    ],
+                                    "score": 0.9,
+                                    "content": "\\mathcal { L } _ { \\mathrm { s u p } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        376,
+                                        532,
+                                        506,
+                                        546
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is the alignment error between",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                543,
+                                505,
+                                556
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        543,
+                                        505,
+                                        556
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the network prediction and the annotation. For example in image classification task (Mishra &",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                554,
+                                506,
+                                567
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        554,
+                                        506,
+                                        567
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Marr, 2017; Shen & Savvides, 2020; Polino et al., 2018; Cho & Hariharan, 2019), it is normally a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                564,
+                                505,
+                                579
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        564,
+                                        505,
+                                        579
+                                    ],
+                                    "score": 1.0,
+                                    "content": "cross entropy loss. For object detection (Liu et al., 2019; Chen et al., 2017), it includes bounding",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                577,
+                                506,
+                                588
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        577,
+                                        259,
+                                        588
+                                    ],
+                                    "score": 1.0,
+                                    "content": "box regression as well. The loss of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        259,
+                                        577,
+                                        284,
+                                        588
+                                    ],
+                                    "score": 0.92,
+                                    "content": "{ \\mathcal { L } } _ { \\mathrm { d i s t i l l } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        284,
+                                        577,
+                                        506,
+                                        588
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is the mimic error of the student network towards a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                587,
+                                505,
+                                599
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        587,
+                                        505,
+                                        599
+                                    ],
+                                    "score": 1.0,
+                                    "content": "pre-trained teacher network. For example in (Hinton et al., 2015), the teacher signal comes from the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                598,
+                                506,
+                                610
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        598,
+                                        506,
+                                        610
+                                    ],
+                                    "score": 1.0,
+                                    "content": "softmax prediction of multiple large-scale networks and the loss is measured by the Kullback–Leibler",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                609,
+                                505,
+                                622
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        609,
+                                        505,
+                                        622
+                                    ],
+                                    "score": 1.0,
+                                    "content": "divergence. In Romero et al. (2014), the task is to align the intermediate feature map values and to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                621,
+                                505,
+                                632
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        621,
+                                        196,
+                                        632
+                                    ],
+                                    "score": 1.0,
+                                    "content": "minimize the squared",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        196,
+                                        621,
+                                        206,
+                                        631
+                                    ],
+                                    "score": 0.31,
+                                    "content": "l 2",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        207,
+                                        621,
+                                        505,
+                                        632
+                                    ],
+                                    "score": 1.0,
+                                    "content": "distance. The effectiveness has been well demonstrated in the supervised",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                632,
+                                490,
+                                643
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        632,
+                                        490,
+                                        643
+                                    ],
+                                    "score": 1.0,
+                                    "content": "setting with labeled data, but remains unknown for the unsupervised setting, which is our focus.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        }
+                    ],
+                    "index": 39
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        655,
+                        408,
+                        668
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                655,
+                                411,
+                                669
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        655,
+                                        411,
+                                        669
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3.2 SELF-SUPERVISED DISTILLATION FOR VISUAL REPRESENTATION",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        }
+                    ],
+                    "index": 45
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        676,
+                        504,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                676,
+                                506,
+                                690
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        676,
+                                        506,
+                                        690
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Different from supervised distillation, SEED aims to transfer knowledge from a large model to a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                687,
+                                505,
+                                699
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        687,
+                                        505,
+                                        699
+                                    ],
+                                    "score": 1.0,
+                                    "content": "small model without requiring labeled data, so that the learned representations in small model can",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 47
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                699,
+                                505,
+                                711
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        699,
+                                        505,
+                                        711
+                                    ],
+                                    "score": 1.0,
+                                    "content": "be used for downstream tasks. Inspired by contrastive SSL, we formulate a simple approach for the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 48
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                709,
+                                505,
+                                722
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        709,
+                                        505,
+                                        722
+                                    ],
+                                    "score": 1.0,
+                                    "content": "distillation on the basis of instance similarity distribution over a contrastive instance queue. Similar",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 49
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                719,
+                                506,
+                                734
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        719,
+                                        506,
+                                        734
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to He et al. (2020), we maintain an instance queue for storing data samples’ encoding output from the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 50
+                        }
+                    ],
+                    "index": 48
+                }
+            ],
+            "page_idx": 2,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        108,
+                        27,
+                        292,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                26,
+                                293,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        26,
+                                        293,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Published as a conference paper at ICLR 2021",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        302,
+                        751,
+                        309,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                301,
+                                750,
+                                310,
+                                762
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        301,
+                                        750,
+                                        310,
+                                        762
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        82,
+                        504,
+                        126
+                    ],
+                    "lines": [],
+                    "index": 1.5,
+                    "bbox_fs": [
+                        105,
+                        82,
+                        506,
+                        128
+                    ],
+                    "lines_deleted": true
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        131,
+                        505,
+                        384
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                132,
+                                505,
+                                145
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        132,
+                                        505,
+                                        145
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Knowledge distillation (Hinton et al., 2015) aims to transfer knowledge from a cumbersome model",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                143,
+                                505,
+                                156
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        143,
+                                        505,
+                                        156
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to a smaller one without losing too much generalization power, which is also well investigated in",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                154,
+                                505,
+                                166
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        154,
+                                        505,
+                                        166
+                                    ],
+                                    "score": 1.0,
+                                    "content": "model compression (Bucilua et al. ˇ , 2006). Instead of mimicking the teacher’s output logit, attention",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                165,
+                                506,
+                                177
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        165,
+                                        506,
+                                        177
+                                    ],
+                                    "score": 1.0,
+                                    "content": "transfer (Zagoruyko & Komodakis, 2016) formulates knowledge distillation on attention maps.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                176,
+                                506,
+                                188
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        176,
+                                        506,
+                                        188
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Similarly, works in (Ahn et al., 2019; Yim et al., 2017; Koratana et al., 2019; Huang & Wang, 2017)",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                186,
+                                506,
+                                200
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        186,
+                                        506,
+                                        200
+                                    ],
+                                    "score": 1.0,
+                                    "content": "have utilized different learning objectives including consistency on feature maps, consistency on",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                198,
+                                505,
+                                209
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        198,
+                                        505,
+                                        209
+                                    ],
+                                    "score": 1.0,
+                                    "content": "probability mass function, and maximizing the mutual information. CRD (Tian et al., 2019a), which is",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                209,
+                                505,
+                                221
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        209,
+                                        505,
+                                        221
+                                    ],
+                                    "score": 1.0,
+                                    "content": "derived from CMC (Tian et al., 2019b), optimizes the student network by a similar objective to Oord",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                219,
+                                505,
+                                232
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        219,
+                                        505,
+                                        232
+                                    ],
+                                    "score": 1.0,
+                                    "content": "et al. (2018) using a derived lower bound on mutual information. However, the aforementioned efforts",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                231,
+                                506,
+                                243
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        231,
+                                        506,
+                                        243
+                                    ],
+                                    "score": 1.0,
+                                    "content": "all focus on task-specific distillation (e.g., image classification) during the fine-tuning phase rather",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                241,
+                                505,
+                                254
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        241,
+                                        505,
+                                        254
+                                    ],
+                                    "score": 1.0,
+                                    "content": "than a task-agnostic distillation in the pre-training phase for the representation learning. Several",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                253,
+                                506,
+                                266
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        253,
+                                        506,
+                                        266
+                                    ],
+                                    "score": 1.0,
+                                    "content": "works on natural language pre-training proposed to leverage knowledge distillation for a smaller yet",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                263,
+                                506,
+                                276
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        263,
+                                        506,
+                                        276
+                                    ],
+                                    "score": 1.0,
+                                    "content": "stronger small models. For instances, DistillBert (Sanh et al., 2019), TinyBert (Jiao et al., 2019), and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                275,
+                                504,
+                                286
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        275,
+                                        504,
+                                        286
+                                    ],
+                                    "score": 1.0,
+                                    "content": "MobileBert (Sun et al., 2020), have used knowledge distillation for model compression and shown",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                285,
+                                506,
+                                298
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        285,
+                                        506,
+                                        298
+                                    ],
+                                    "score": 1.0,
+                                    "content": "their validity on multiple downstream tasks. Similar works also emphasize the value of smaller and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                297,
+                                505,
+                                309
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        297,
+                                        505,
+                                        309
+                                    ],
+                                    "score": 1.0,
+                                    "content": "faster models for language representation learning by leveraging knowledge distillation (Turc et al.,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                307,
+                                506,
+                                320
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        307,
+                                        506,
+                                        320
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2019; Sun et al., 2019). These works all demonstrate the effectiveness of knowledge distillation for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                319,
+                                505,
+                                330
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        319,
+                                        505,
+                                        330
+                                    ],
+                                    "score": 1.0,
+                                    "content": "language representation learning in small models, while are not extended to the pre-training for visual",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                329,
+                                505,
+                                342
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        329,
+                                        505,
+                                        342
+                                    ],
+                                    "score": 1.0,
+                                    "content": "representations. Notably, a recent concurrent work CompRess (Abbasi Koohpayegani et al., 2020)",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                340,
+                                505,
+                                353
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        340,
+                                        505,
+                                        353
+                                    ],
+                                    "score": 1.0,
+                                    "content": "also point out the importance to develop better SSL method for smaller models. SEED closely relates",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                351,
+                                505,
+                                364
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        351,
+                                        505,
+                                        364
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to the above techniques but aims to facilitate visual representation learning during pre-training",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                363,
+                                504,
+                                374
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        363,
+                                        504,
+                                        374
+                                    ],
+                                    "score": 1.0,
+                                    "content": "phase using distillation technique for small models, which as far as we know has not yet been",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                374,
+                                159,
+                                385
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        374,
+                                        159,
+                                        385
+                                    ],
+                                    "score": 1.0,
+                                    "content": "investigated.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        }
+                    ],
+                    "index": 15,
+                    "bbox_fs": [
+                        105,
+                        132,
+                        506,
+                        385
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        401,
+                        173,
+                        413
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                398,
+                                175,
+                                416
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        398,
+                                        175,
+                                        416
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3 METHOD",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        }
+                    ],
+                    "index": 27
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        425,
+                        330,
+                        437
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                424,
+                                331,
+                                438
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        424,
+                                        331,
+                                        438
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3.1 PRELIMINARY ON KNOWLEDGE DISTILLATION",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        }
+                    ],
+                    "index": 28
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        446,
+                        505,
+                        479
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                445,
+                                505,
+                                457
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        445,
+                                        505,
+                                        457
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Knowledge distillation (Hinton et al., 2015; Bucilua et al.ˇ , 2006) is an effective technique to transfer",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                457,
+                                505,
+                                469
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        457,
+                                        505,
+                                        469
+                                    ],
+                                    "score": 1.0,
+                                    "content": "knowledge from a strong teacher network to a target student network. The training task can be",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                468,
+                                274,
+                                480
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        468,
+                                        274,
+                                        480
+                                    ],
+                                    "score": 1.0,
+                                    "content": "generalized as the following formulation:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        }
+                    ],
+                    "index": 30,
+                    "bbox_fs": [
+                        105,
+                        445,
+                        505,
+                        480
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        193,
+                        483,
+                        416,
+                        518
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                193,
+                                483,
+                                416,
+                                518
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        193,
+                                        483,
+                                        416,
+                                        518
+                                    ],
+                                    "score": 0.94,
+                                    "content": "\\hat { \\theta } _ { S } = \\underset { \\theta _ { S } } { \\arg \\operatorname* { m i n } } \\sum _ { i } ^ { N } \\mathcal { L } _ { \\mathrm { s u p } } ( \\mathbf { x } _ { i } , \\theta _ { S } , y _ { i } ) + \\mathcal { L } _ { \\mathrm { d i s t i l l } } ( \\mathbf { x } _ { i } , \\theta _ { S } , \\theta _ { T } ) ,",
+                                    "type": "interline_equation",
+                                    "image_path": "f6d588e03d3db3f013e786c6fff9679cd8aa0b9b81a92bf569f2acbe7a654360.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 32.5,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                193,
+                                483,
+                                416,
+                                500.5
+                            ],
+                            "spans": [],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                193,
+                                500.5,
+                                416,
+                                518.0
+                            ],
+                            "spans": [],
+                            "index": 33
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        521,
+                        505,
+                        643
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                521,
+                                506,
+                                535
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        521,
+                                        133,
+                                        535
+                                    ],
+                                    "score": 1.0,
+                                    "content": "where",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        134,
+                                        523,
+                                        145,
+                                        533
+                                    ],
+                                    "score": 0.84,
+                                    "content": "\\mathbf { x } _ { i }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        145,
+                                        521,
+                                        199,
+                                        535
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is an image,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        199,
+                                        523,
+                                        209,
+                                        533
+                                    ],
+                                    "score": 0.84,
+                                    "content": "y _ { i }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        209,
+                                        521,
+                                        345,
+                                        535
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is the corresponding annotation,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        346,
+                                        522,
+                                        358,
+                                        533
+                                    ],
+                                    "score": 0.88,
+                                    "content": "\\theta _ { S }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        358,
+                                        521,
+                                        506,
+                                        535
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is the parameter set for the student",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                532,
+                                506,
+                                546
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        532,
+                                        163,
+                                        546
+                                    ],
+                                    "score": 1.0,
+                                    "content": "network, and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        163,
+                                        533,
+                                        175,
+                                        544
+                                    ],
+                                    "score": 0.88,
+                                    "content": "\\theta _ { T }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        176,
+                                        532,
+                                        357,
+                                        546
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is the set for the teacher network. The loss",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        357,
+                                        533,
+                                        376,
+                                        545
+                                    ],
+                                    "score": 0.9,
+                                    "content": "\\mathcal { L } _ { \\mathrm { s u p } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        376,
+                                        532,
+                                        506,
+                                        546
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is the alignment error between",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                543,
+                                505,
+                                556
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        543,
+                                        505,
+                                        556
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the network prediction and the annotation. For example in image classification task (Mishra &",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                554,
+                                506,
+                                567
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        554,
+                                        506,
+                                        567
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Marr, 2017; Shen & Savvides, 2020; Polino et al., 2018; Cho & Hariharan, 2019), it is normally a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                564,
+                                505,
+                                579
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        564,
+                                        505,
+                                        579
+                                    ],
+                                    "score": 1.0,
+                                    "content": "cross entropy loss. For object detection (Liu et al., 2019; Chen et al., 2017), it includes bounding",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                577,
+                                506,
+                                588
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        577,
+                                        259,
+                                        588
+                                    ],
+                                    "score": 1.0,
+                                    "content": "box regression as well. The loss of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        259,
+                                        577,
+                                        284,
+                                        588
+                                    ],
+                                    "score": 0.92,
+                                    "content": "{ \\mathcal { L } } _ { \\mathrm { d i s t i l l } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        284,
+                                        577,
+                                        506,
+                                        588
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is the mimic error of the student network towards a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                587,
+                                505,
+                                599
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        587,
+                                        505,
+                                        599
+                                    ],
+                                    "score": 1.0,
+                                    "content": "pre-trained teacher network. For example in (Hinton et al., 2015), the teacher signal comes from the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                598,
+                                506,
+                                610
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        598,
+                                        506,
+                                        610
+                                    ],
+                                    "score": 1.0,
+                                    "content": "softmax prediction of multiple large-scale networks and the loss is measured by the Kullback–Leibler",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                609,
+                                505,
+                                622
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        609,
+                                        505,
+                                        622
+                                    ],
+                                    "score": 1.0,
+                                    "content": "divergence. In Romero et al. (2014), the task is to align the intermediate feature map values and to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                621,
+                                505,
+                                632
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        621,
+                                        196,
+                                        632
+                                    ],
+                                    "score": 1.0,
+                                    "content": "minimize the squared",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        196,
+                                        621,
+                                        206,
+                                        631
+                                    ],
+                                    "score": 0.31,
+                                    "content": "l 2",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        207,
+                                        621,
+                                        505,
+                                        632
+                                    ],
+                                    "score": 1.0,
+                                    "content": "distance. The effectiveness has been well demonstrated in the supervised",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                632,
+                                490,
+                                643
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        632,
+                                        490,
+                                        643
+                                    ],
+                                    "score": 1.0,
+                                    "content": "setting with labeled data, but remains unknown for the unsupervised setting, which is our focus.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        }
+                    ],
+                    "index": 39,
+                    "bbox_fs": [
+                        104,
+                        521,
+                        506,
+                        643
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        655,
+                        408,
+                        668
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                655,
+                                411,
+                                669
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        655,
+                                        411,
+                                        669
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3.2 SELF-SUPERVISED DISTILLATION FOR VISUAL REPRESENTATION",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        }
+                    ],
+                    "index": 45
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        676,
+                        504,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                676,
+                                506,
+                                690
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        676,
+                                        506,
+                                        690
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Different from supervised distillation, SEED aims to transfer knowledge from a large model to a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                687,
+                                505,
+                                699
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        687,
+                                        505,
+                                        699
+                                    ],
+                                    "score": 1.0,
+                                    "content": "small model without requiring labeled data, so that the learned representations in small model can",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 47
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                699,
+                                505,
+                                711
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        699,
+                                        505,
+                                        711
+                                    ],
+                                    "score": 1.0,
+                                    "content": "be used for downstream tasks. Inspired by contrastive SSL, we formulate a simple approach for the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 48
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                709,
+                                505,
+                                722
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        709,
+                                        505,
+                                        722
+                                    ],
+                                    "score": 1.0,
+                                    "content": "distillation on the basis of instance similarity distribution over a contrastive instance queue. Similar",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 49
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                719,
+                                506,
+                                734
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        719,
+                                        506,
+                                        734
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to He et al. (2020), we maintain an instance queue for storing data samples’ encoding output from the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 50
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                239,
+                                505,
+                                253
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        239,
+                                        505,
+                                        253
+                                    ],
+                                    "score": 1.0,
+                                    "content": "teacher. Given a new sample, we compute its similarity scores with all the samples in the queue using",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                250,
+                                505,
+                                263
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        250,
+                                        505,
+                                        263
+                                    ],
+                                    "score": 1.0,
+                                    "content": "both the teacher and the student models. We require that the similarity score distribution computed",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                261,
+                                505,
+                                275
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        261,
+                                        505,
+                                        275
+                                    ],
+                                    "score": 1.0,
+                                    "content": "by the student matches with that computed by the teacher, which is formulated as minimizing the",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                272,
+                                505,
+                                286
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        272,
+                                        505,
+                                        286
+                                    ],
+                                    "score": 1.0,
+                                    "content": "cross entropy between the student and the teacher’s similarity score distributions (as illustrated in",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                282,
+                                149,
+                                298
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        282,
+                                        149,
+                                        298
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Figure 2).",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 10
+                        }
+                    ],
+                    "index": 48,
+                    "bbox_fs": [
+                        105,
+                        676,
+                        506,
+                        734
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "image",
+                    "bbox": [
+                        111,
+                        81,
+                        505,
+                        181
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                111,
+                                81,
+                                505,
+                                181
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        111,
+                                        81,
+                                        505,
+                                        181
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                111,
+                                                81,
+                                                505,
+                                                181
+                                            ],
+                                            "score": 0.966,
+                                            "type": "image",
+                                            "image_path": "2f2a03a1e9389d9452c8beee5f2f9f477f5576361541704a7434d89396409316.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 1,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        111,
+                                        81,
+                                        505,
+                                        114.33333333333334
+                                    ],
+                                    "spans": [],
+                                    "index": 0
+                                },
+                                {
+                                    "bbox": [
+                                        111,
+                                        114.33333333333334,
+                                        505,
+                                        147.66666666666669
+                                    ],
+                                    "spans": [],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        111,
+                                        147.66666666666669,
+                                        505,
+                                        181.00000000000003
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                106,
+                                189,
+                                506,
+                                221
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        189,
+                                        506,
+                                        201
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                189,
+                                                506,
+                                                201
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 2: Illustration of our self-supervised distillation pipeline. The teacher encoder is pre-trained by SSL and",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 3
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        199,
+                                        506,
+                                        211
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                199,
+                                                506,
+                                                211
+                                            ],
+                                            "score": 1.0,
+                                            "content": "kept frozen during the distillation. The student encoder is trained by minimizing the cross entropy of probabilities",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 4
+                                },
+                                {
+                                    "bbox": [
+                                        104,
+                                        208,
+                                        500,
+                                        223
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                104,
+                                                208,
+                                                500,
+                                                223
+                                            ],
+                                            "score": 1.0,
+                                            "content": "from teacher & student for an augmented view of an image, computed over a dynamically maintained queue.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 5
+                                }
+                            ],
+                            "index": 4
+                        }
+                    ],
+                    "index": 2.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        239,
+                        505,
+                        295
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                239,
+                                505,
+                                253
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        239,
+                                        505,
+                                        253
+                                    ],
+                                    "score": 1.0,
+                                    "content": "teacher. Given a new sample, we compute its similarity scores with all the samples in the queue using",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                250,
+                                505,
+                                263
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        250,
+                                        505,
+                                        263
+                                    ],
+                                    "score": 1.0,
+                                    "content": "both the teacher and the student models. We require that the similarity score distribution computed",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                261,
+                                505,
+                                275
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        261,
+                                        505,
+                                        275
+                                    ],
+                                    "score": 1.0,
+                                    "content": "by the student matches with that computed by the teacher, which is formulated as minimizing the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                272,
+                                505,
+                                286
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        272,
+                                        505,
+                                        286
+                                    ],
+                                    "score": 1.0,
+                                    "content": "cross entropy between the student and the teacher’s similarity score distributions (as illustrated in",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                282,
+                                149,
+                                298
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        282,
+                                        149,
+                                        298
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Figure 2).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        }
+                    ],
+                    "index": 8
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        300,
+                        505,
+                        435
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                300,
+                                505,
+                                313
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        300,
+                                        296,
+                                        313
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Specifically, for a randomly augmented view",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        296,
+                                        303,
+                                        307,
+                                        311
+                                    ],
+                                    "score": 0.82,
+                                    "content": "\\mathbf { x } _ { i }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        307,
+                                        300,
+                                        505,
+                                        313
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of an image, it is first mapped and normalized",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                310,
+                                506,
+                                326
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        310,
+                                        248,
+                                        326
+                                    ],
+                                    "score": 1.0,
+                                    "content": "into feature vector representations",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        248,
+                                        311,
+                                        351,
+                                        324
+                                    ],
+                                    "score": 0.94,
+                                    "content": "\\mathbf { z } _ { i } ^ { T } = f _ { \\theta } ^ { T } ( \\mathbf { x } _ { i } ) / \\vert \\vert f _ { \\theta } ^ { T } ( \\mathbf { x } _ { i } ) \\vert \\vert _ { 2 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        351,
+                                        310,
+                                        372,
+                                        326
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        372,
+                                        311,
+                                        474,
+                                        324
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\mathbf { z } _ { i } ^ { S } = f _ { \\theta } ^ { S } ( \\mathbf { \\bar { x } } _ { i } ^ { \\star } ) / \\vert \\vert f _ { \\theta } ^ { S } ( \\mathbf { x } _ { i } ) \\vert \\vert _ { 2 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        474,
+                                        310,
+                                        506,
+                                        326
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", where",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                318,
+                                505,
+                                340
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        322,
+                                        163,
+                                        336
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\mathbf { z } _ { i } ^ { T } , \\mathbf { z } _ { i } ^ { S } \\in \\mathbb { R } ^ { D }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        164,
+                                        318,
+                                        186,
+                                        340
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        187,
+                                        323,
+                                        200,
+                                        336
+                                    ],
+                                    "score": 0.9,
+                                    "content": "f _ { \\theta } ^ { T }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        200,
+                                        318,
+                                        220,
+                                        340
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        221,
+                                        323,
+                                        234,
+                                        336
+                                    ],
+                                    "score": 0.9,
+                                    "content": "f _ { \\theta } ^ { S }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        234,
+                                        318,
+                                        481,
+                                        340
+                                    ],
+                                    "score": 1.0,
+                                    "content": "θ θ θ θdenote the teacher and student encoders, respectively. Let",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        481,
+                                        324,
+                                        505,
+                                        335
+                                    ],
+                                    "score": 0.84,
+                                    "content": "\\mathbf { D } =",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                107,
+                                335,
+                                506,
+                                348
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        336,
+                                        146,
+                                        347
+                                    ],
+                                    "score": 0.89,
+                                    "content": "[ \\mathbf { d } _ { 1 } . . . \\mathbf { d } _ { K } ]",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        147,
+                                        335,
+                                        289,
+                                        348
+                                    ],
+                                    "score": 1.0,
+                                    "content": "denote the instance queue where",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        290,
+                                        335,
+                                        300,
+                                        345
+                                    ],
+                                    "score": 0.81,
+                                    "content": "K",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        300,
+                                        335,
+                                        404,
+                                        348
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is the queue length and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        405,
+                                        335,
+                                        416,
+                                        347
+                                    ],
+                                    "score": 0.88,
+                                    "content": "{ \\bf d } _ { j }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        417,
+                                        335,
+                                        506,
+                                        348
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is the feature vector",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                346,
+                                505,
+                                358
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        346,
+                                        429,
+                                        358
+                                    ],
+                                    "score": 1.0,
+                                    "content": "obtained from the teacher encoder. Similar to the contrastive learning framework,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        430,
+                                        346,
+                                        440,
+                                        356
+                                    ],
+                                    "score": 0.53,
+                                    "content": "\\mathbf { D }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        441,
+                                        346,
+                                        505,
+                                        358
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is progressively",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                357,
+                                506,
+                                369
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        357,
+                                        506,
+                                        369
+                                    ],
+                                    "score": 1.0,
+                                    "content": "updated under the “first-in first-out” strategy as distillation proceeds. That is, we en-queue the visual",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                368,
+                                506,
+                                380
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        368,
+                                        506,
+                                        380
+                                    ],
+                                    "score": 1.0,
+                                    "content": "features of the current batch inferred by the teacher and de-queue the earliest seen samples at the end",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                378,
+                                505,
+                                391
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        378,
+                                        331,
+                                        391
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of iteration. Note that the maintained samples in queue",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        332,
+                                        379,
+                                        342,
+                                        389
+                                    ],
+                                    "score": 0.52,
+                                    "content": "\\mathbf { D }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        342,
+                                        378,
+                                        505,
+                                        391
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are mostly random and irrelevant to the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                389,
+                                505,
+                                402
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        389,
+                                        166,
+                                        402
+                                    ],
+                                    "score": 1.0,
+                                    "content": "target instance",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        166,
+                                        391,
+                                        177,
+                                        401
+                                    ],
+                                    "score": 0.85,
+                                    "content": "\\mathbf { x } _ { i }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        177,
+                                        389,
+                                        505,
+                                        402
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Minimizing the cross entropy between the similarity score distribution computed",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                400,
+                                505,
+                                412
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        400,
+                                        251,
+                                        412
+                                    ],
+                                    "score": 1.0,
+                                    "content": "by the student and teacher based on",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        252,
+                                        401,
+                                        262,
+                                        411
+                                    ],
+                                    "score": 0.52,
+                                    "content": "\\mathbf { D }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        262,
+                                        400,
+                                        326,
+                                        412
+                                    ],
+                                    "score": 1.0,
+                                    "content": "softly contrasts",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        326,
+                                        402,
+                                        337,
+                                        412
+                                    ],
+                                    "score": 0.87,
+                                    "content": "\\mathbf { x } _ { i }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        338,
+                                        400,
+                                        505,
+                                        412
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with randomly selected samples, without",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                410,
+                                506,
+                                425
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        410,
+                                        506,
+                                        425
+                                    ],
+                                    "score": 1.0,
+                                    "content": "directly aligning with the teacher encoder. To address this problem, we add the teacher’s embedding",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                421,
+                                397,
+                                437
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        422,
+                                        125,
+                                        435
+                                    ],
+                                    "score": 0.89,
+                                    "content": "( \\mathbf { z } _ { i } ^ { T } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        126,
+                                        421,
+                                        224,
+                                        437
+                                    ],
+                                    "score": 1.0,
+                                    "content": "into the queue and form",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        225,
+                                        422,
+                                        322,
+                                        434
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\mathbf { D } ^ { + } = [ \\mathbf { d } _ { 1 } . . . \\mathbf { d } _ { K } , \\mathbf { d } _ { K + 1 } ]",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        322,
+                                        421,
+                                        343,
+                                        437
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        343,
+                                        422,
+                                        393,
+                                        435
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\mathbf { d } _ { K + 1 } = \\mathbf { z } _ { i } ^ { T }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        393,
+                                        421,
+                                        397,
+                                        437
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        }
+                    ],
+                    "index": 16.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        440,
+                        504,
+                        465
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                439,
+                                505,
+                                453
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        439,
+                                        123,
+                                        453
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Let",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        123,
+                                        439,
+                                        189,
+                                        452
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\mathbf { p } ^ { T } ( \\mathbf { x } _ { i } ; \\boldsymbol { \\theta } _ { T } ; \\mathbf { D } ^ { + } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        189,
+                                        439,
+                                        454,
+                                        453
+                                    ],
+                                    "score": 1.0,
+                                    "content": "denote the similarity score between the extracted teacher feature",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        454,
+                                        439,
+                                        466,
+                                        452
+                                    ],
+                                    "score": 0.9,
+                                    "content": "\\mathbf { z } _ { i } ^ { T }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        467,
+                                        439,
+                                        485,
+                                        453
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        485,
+                                        440,
+                                        497,
+                                        453
+                                    ],
+                                    "score": 0.86,
+                                    "content": "{ \\bf d } _ { j }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        497,
+                                        439,
+                                        505,
+                                        453
+                                    ],
+                                    "score": 1.0,
+                                    "content": "’s",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                107,
+                                451,
+                                432,
+                                465
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        453,
+                                        179,
+                                        464
+                                    ],
+                                    "score": 0.89,
+                                    "content": "( j = 1 , . . . , K + 1 )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        180,
+                                        452,
+                                        312,
+                                        465
+                                    ],
+                                    "score": 1.0,
+                                    "content": "computed by the teacher model.",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        312,
+                                        451,
+                                        378,
+                                        465
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\mathbf { p } ^ { T } ( \\mathbf { x } _ { i } ; \\boldsymbol { \\theta } _ { T } ; \\mathbf { D } ^ { + } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        379,
+                                        452,
+                                        432,
+                                        465
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is defined as",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        }
+                    ],
+                    "index": 23.5
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        158,
+                        468,
+                        452,
+                        496
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                158,
+                                468,
+                                452,
+                                496
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        158,
+                                        468,
+                                        452,
+                                        496
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\mathbf { p } ^ { T } ( \\mathbf { x } _ { i } ; \\theta _ { T } , \\mathbf { D } ^ { + } ) = \\left[ p _ { 1 } ^ { T } \\dots p _ { K + 1 } ^ { T } \\right] , \\qquad p _ { j } ^ { T } = \\frac { \\exp ( \\mathbf { z } _ { i } ^ { T } \\cdot \\mathbf { d } _ { j } / \\tau ^ { T } ) } { \\sum _ { \\mathbf { d } \\sim \\mathbf { D } ^ { + } } \\exp ( \\mathbf { z } _ { i } ^ { T } \\cdot \\mathbf { d } / \\tau ^ { T } ) } ,",
+                                    "type": "interline_equation",
+                                    "image_path": "fa03ea71bb692eacf77006e376709e981c9236cbb7b4dc642d83915ba5eb6316.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 26,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                158,
+                                468,
+                                452,
+                                477.3333333333333
+                            ],
+                            "spans": [],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                158,
+                                477.3333333333333,
+                                452,
+                                486.66666666666663
+                            ],
+                            "spans": [],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                158,
+                                486.66666666666663,
+                                452,
+                                495.99999999999994
+                            ],
+                            "spans": [],
+                            "index": 27
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        499,
+                        504,
+                        523
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                499,
+                                505,
+                                512
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        499,
+                                        123,
+                                        512
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        123,
+                                        499,
+                                        136,
+                                        510
+                                    ],
+                                    "score": 0.89,
+                                    "content": "\\tau ^ { T }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        136,
+                                        499,
+                                        360,
+                                        512
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is a temperature parameter for the teacher. Note, we use",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        360,
+                                        499,
+                                        375,
+                                        512
+                                    ],
+                                    "score": 0.9,
+                                    "content": "( ) ^ { T }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        375,
+                                        499,
+                                        505,
+                                        512
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to represent the feature from the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                510,
+                                428,
+                                523
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        510,
+                                        206,
+                                        523
+                                    ],
+                                    "score": 1.0,
+                                    "content": "teacher network and use",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        206,
+                                        511,
+                                        217,
+                                        523
+                                    ],
+                                    "score": 0.58,
+                                    "content": "( \\cdot )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        217,
+                                        510,
+                                        428,
+                                        523
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to represent the inner product between two features.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        }
+                    ],
+                    "index": 28.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        527,
+                        504,
+                        550
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                525,
+                                506,
+                                542
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        525,
+                                        159,
+                                        542
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Similarly let",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        160,
+                                        527,
+                                        224,
+                                        540
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\mathbf { p } ^ { S } ( x _ { i } ; \\theta _ { S } , \\mathbf { D } ^ { + } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        225,
+                                        525,
+                                        506,
+                                        542
+                                    ],
+                                    "score": 1.0,
+                                    "content": "denote the similarity score computed by the student model, which is",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                537,
+                                151,
+                                551
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        537,
+                                        151,
+                                        551
+                                    ],
+                                    "score": 1.0,
+                                    "content": "defined as",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        }
+                    ],
+                    "index": 30.5
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        146,
+                        551,
+                        464,
+                        580
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                146,
+                                551,
+                                464,
+                                580
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        146,
+                                        551,
+                                        464,
+                                        580
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\mathbf { p } ^ { S } ( \\mathbf { x } _ { i } ; \\theta _ { S } , \\mathbf { D } ^ { + } ) = \\left[ p _ { 1 } ^ { S } \\dots p _ { K + 1 } ^ { S } \\right] , \\qquad { \\mathrm { w h e r e ~ } } p _ { j } ^ { S } = { \\frac { \\exp ( \\mathbf { z } _ { i } ^ { S } \\cdot \\mathbf { d } _ { j } / \\tau ^ { S } ) } { \\sum _ { \\mathbf { d } \\sim \\mathbf { D } ^ { + } } \\exp ( \\mathbf { z } _ { i } ^ { S } \\cdot \\mathbf { d } / \\tau ^ { S } ) } } ,",
+                                    "type": "interline_equation",
+                                    "image_path": "7a86dd461ab60bef53314bfb98d8072c5f8f7ea6f386125f046180e8c76b6966.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 33,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                146,
+                                551,
+                                464,
+                                560.6666666666666
+                            ],
+                            "spans": [],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                146,
+                                560.6666666666666,
+                                464,
+                                570.3333333333333
+                            ],
+                            "spans": [],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                146,
+                                570.3333333333333,
+                                464,
+                                579.9999999999999
+                            ],
+                            "spans": [],
+                            "index": 34
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        583,
+                        307,
+                        596
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                581,
+                                308,
+                                597
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        581,
+                                        123,
+                                        597
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        123,
+                                        583,
+                                        136,
+                                        594
+                                    ],
+                                    "score": 0.88,
+                                    "content": "\\tau ^ { S }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        136,
+                                        581,
+                                        308,
+                                        597
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is a temperature parameter for the student.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        }
+                    ],
+                    "index": 35
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        600,
+                        504,
+                        634
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                600,
+                                505,
+                                613
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        600,
+                                        505,
+                                        613
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Our self-supervised distillation can be formulated as minimizing the cross entropy between the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                611,
+                                506,
+                                624
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        611,
+                                        242,
+                                        624
+                                    ],
+                                    "score": 1.0,
+                                    "content": "similarity scores of the teacher,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        242,
+                                        611,
+                                        308,
+                                        624
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\mathbf { p } ^ { T } ( \\mathbf { x } _ { i } ; \\theta _ { T } , \\mathbf { D } ^ { + } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        308,
+                                        611,
+                                        384,
+                                        624
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", and the student,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        384,
+                                        611,
+                                        449,
+                                        624
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\mathbf { p } ^ { S } ( \\mathbf { x } _ { i } ; \\boldsymbol { \\theta } _ { S } , \\mathbf { D } ^ { \\dagger } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        449,
+                                        611,
+                                        506,
+                                        624
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", over all the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                622,
+                                189,
+                                635
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        622,
+                                        145,
+                                        635
+                                    ],
+                                    "score": 1.0,
+                                    "content": "instances",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        145,
+                                        624,
+                                        156,
+                                        634
+                                    ],
+                                    "score": 0.86,
+                                    "content": "\\mathbf { x } _ { i }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        156,
+                                        622,
+                                        189,
+                                        635
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", that is,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        }
+                    ],
+                    "index": 37
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        138,
+                        636,
+                        472,
+                        708
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                138,
+                                636,
+                                472,
+                                708
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        138,
+                                        636,
+                                        472,
+                                        708
+                                    ],
+                                    "score": 0.95,
+                                    "content": "\\begin{array} { r l } & { \\boldsymbol { \\hat { \\theta } _ { S } } = \\underset { \\boldsymbol { \\theta _ { S } } } { \\arg \\operatorname* { m i n } } \\sum _ { i } ^ { N } - \\mathbf { p } ^ { T } ( \\mathbf { x } _ { i } ; \\boldsymbol { \\theta _ { T } } , \\mathbf { D } ^ { + } ) \\cdot \\log \\mathbf { p } ^ { S } ( \\mathbf { x } _ { i } ; \\boldsymbol { \\theta _ { S } } , \\mathbf { D } ^ { + } ) } \\\\ & { \\quad = \\underset { \\boldsymbol { \\theta _ { S } } } { \\arg \\operatorname* { m i n } } \\sum _ { i } ^ { N } \\sum _ { j } ^ { K + 1 } - \\frac { \\exp \\left( \\mathbf { z } _ { i } ^ { T } \\cdot \\mathbf { d } _ { j } / \\tau ^ { T } \\right) } { \\sum _ { \\mathbf { d } \\sim \\mathbf { D } ^ { + } } \\exp \\left( \\mathbf { z } _ { i } ^ { T } \\cdot \\mathbf { d } / \\tau ^ { T } \\right) } \\cdot \\log \\frac { \\exp \\left( \\mathbf { z } _ { i } ^ { S } \\cdot \\mathbf { d } _ { j } / \\tau ^ { S } \\right) } { \\sum _ { \\mathbf { d } \\sim \\mathbf { D } ^ { + } } \\exp \\left( \\mathbf { z } _ { i } ^ { S } \\cdot \\mathbf { d } / \\tau ^ { S } \\right) } . } \\end{array}",
+                                    "type": "interline_equation",
+                                    "image_path": "ebb8b97ca7f12d289494242cd41ec751bf46c453936e79d85361fe539a0cb9bd.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 40,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                138,
+                                636,
+                                472,
+                                660.0
+                            ],
+                            "spans": [],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                138,
+                                660.0,
+                                472,
+                                684.0
+                            ],
+                            "spans": [],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                138,
+                                684.0,
+                                472,
+                                708.0
+                            ],
+                            "spans": [],
+                            "index": 41
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        709,
+                        505,
+                        733
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                708,
+                                505,
+                                723
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        708,
+                                        505,
+                                        723
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Since the teacher network is pre-trained and frozen, the queued features are consistent during training",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                720,
+                                505,
+                                734
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        720,
+                                        312,
+                                        734
+                                    ],
+                                    "score": 1.0,
+                                    "content": "w.r.t. the student network. The higher the value of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        312,
+                                        721,
+                                        324,
+                                        734
+                                    ],
+                                    "score": 0.91,
+                                    "content": "p _ { j } ^ { I ^ { \\prime } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        324,
+                                        720,
+                                        469,
+                                        734
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is, the larger weight will be laid on",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        469,
+                                        721,
+                                        481,
+                                        734
+                                    ],
+                                    "score": 0.89,
+                                    "content": "p _ { j } ^ { S }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        481,
+                                        720,
+                                        505,
+                                        734
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Due",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        }
+                    ],
+                    "index": 42.5
+                }
+            ],
+            "page_idx": 3,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        107,
+                        27,
+                        292,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                26,
+                                293,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        26,
+                                        293,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Published as a conference paper at ICLR 2021",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        302,
+                        751,
+                        308,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                301,
+                                750,
+                                310,
+                                762
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        301,
+                                        750,
+                                        310,
+                                        762
+                                    ],
+                                    "score": 1.0,
+                                    "content": "",
+                                    "type": "text",
+                                    "height": 12,
+                                    "width": 9
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "image",
+                    "bbox": [
+                        111,
+                        81,
+                        505,
+                        181
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                111,
+                                81,
+                                505,
+                                181
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        111,
+                                        81,
+                                        505,
+                                        181
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                111,
+                                                81,
+                                                505,
+                                                181
+                                            ],
+                                            "score": 0.966,
+                                            "type": "image",
+                                            "image_path": "2f2a03a1e9389d9452c8beee5f2f9f477f5576361541704a7434d89396409316.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 1,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        111,
+                                        81,
+                                        505,
+                                        114.33333333333334
+                                    ],
+                                    "spans": [],
+                                    "index": 0
+                                },
+                                {
+                                    "bbox": [
+                                        111,
+                                        114.33333333333334,
+                                        505,
+                                        147.66666666666669
+                                    ],
+                                    "spans": [],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        111,
+                                        147.66666666666669,
+                                        505,
+                                        181.00000000000003
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                106,
+                                189,
+                                506,
+                                221
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        189,
+                                        506,
+                                        201
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                189,
+                                                506,
+                                                201
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 2: Illustration of our self-supervised distillation pipeline. The teacher encoder is pre-trained by SSL and",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 3
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        199,
+                                        506,
+                                        211
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                199,
+                                                506,
+                                                211
+                                            ],
+                                            "score": 1.0,
+                                            "content": "kept frozen during the distillation. The student encoder is trained by minimizing the cross entropy of probabilities",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 4
+                                },
+                                {
+                                    "bbox": [
+                                        104,
+                                        208,
+                                        500,
+                                        223
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                104,
+                                                208,
+                                                500,
+                                                223
+                                            ],
+                                            "score": 1.0,
+                                            "content": "from teacher & student for an augmented view of an image, computed over a dynamically maintained queue.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 5
+                                }
+                            ],
+                            "index": 4
+                        }
+                    ],
+                    "index": 2.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        239,
+                        505,
+                        295
+                    ],
+                    "lines": [],
+                    "index": 8,
+                    "bbox_fs": [
+                        105,
+                        239,
+                        505,
+                        298
+                    ],
+                    "lines_deleted": true
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        300,
+                        505,
+                        435
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                300,
+                                505,
+                                313
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        300,
+                                        296,
+                                        313
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Specifically, for a randomly augmented view",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        296,
+                                        303,
+                                        307,
+                                        311
+                                    ],
+                                    "score": 0.82,
+                                    "content": "\\mathbf { x } _ { i }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        307,
+                                        300,
+                                        505,
+                                        313
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of an image, it is first mapped and normalized",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                310,
+                                506,
+                                326
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        310,
+                                        248,
+                                        326
+                                    ],
+                                    "score": 1.0,
+                                    "content": "into feature vector representations",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        248,
+                                        311,
+                                        351,
+                                        324
+                                    ],
+                                    "score": 0.94,
+                                    "content": "\\mathbf { z } _ { i } ^ { T } = f _ { \\theta } ^ { T } ( \\mathbf { x } _ { i } ) / \\vert \\vert f _ { \\theta } ^ { T } ( \\mathbf { x } _ { i } ) \\vert \\vert _ { 2 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        351,
+                                        310,
+                                        372,
+                                        326
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        372,
+                                        311,
+                                        474,
+                                        324
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\mathbf { z } _ { i } ^ { S } = f _ { \\theta } ^ { S } ( \\mathbf { \\bar { x } } _ { i } ^ { \\star } ) / \\vert \\vert f _ { \\theta } ^ { S } ( \\mathbf { x } _ { i } ) \\vert \\vert _ { 2 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        474,
+                                        310,
+                                        506,
+                                        326
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", where",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                318,
+                                505,
+                                340
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        322,
+                                        163,
+                                        336
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\mathbf { z } _ { i } ^ { T } , \\mathbf { z } _ { i } ^ { S } \\in \\mathbb { R } ^ { D }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        164,
+                                        318,
+                                        186,
+                                        340
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        187,
+                                        323,
+                                        200,
+                                        336
+                                    ],
+                                    "score": 0.9,
+                                    "content": "f _ { \\theta } ^ { T }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        200,
+                                        318,
+                                        220,
+                                        340
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        221,
+                                        323,
+                                        234,
+                                        336
+                                    ],
+                                    "score": 0.9,
+                                    "content": "f _ { \\theta } ^ { S }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        234,
+                                        318,
+                                        481,
+                                        340
+                                    ],
+                                    "score": 1.0,
+                                    "content": "θ θ θ θdenote the teacher and student encoders, respectively. Let",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        481,
+                                        324,
+                                        505,
+                                        335
+                                    ],
+                                    "score": 0.84,
+                                    "content": "\\mathbf { D } =",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                107,
+                                335,
+                                506,
+                                348
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        336,
+                                        146,
+                                        347
+                                    ],
+                                    "score": 0.89,
+                                    "content": "[ \\mathbf { d } _ { 1 } . . . \\mathbf { d } _ { K } ]",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        147,
+                                        335,
+                                        289,
+                                        348
+                                    ],
+                                    "score": 1.0,
+                                    "content": "denote the instance queue where",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        290,
+                                        335,
+                                        300,
+                                        345
+                                    ],
+                                    "score": 0.81,
+                                    "content": "K",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        300,
+                                        335,
+                                        404,
+                                        348
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is the queue length and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        405,
+                                        335,
+                                        416,
+                                        347
+                                    ],
+                                    "score": 0.88,
+                                    "content": "{ \\bf d } _ { j }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        417,
+                                        335,
+                                        506,
+                                        348
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is the feature vector",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                346,
+                                505,
+                                358
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        346,
+                                        429,
+                                        358
+                                    ],
+                                    "score": 1.0,
+                                    "content": "obtained from the teacher encoder. Similar to the contrastive learning framework,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        430,
+                                        346,
+                                        440,
+                                        356
+                                    ],
+                                    "score": 0.53,
+                                    "content": "\\mathbf { D }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        441,
+                                        346,
+                                        505,
+                                        358
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is progressively",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                357,
+                                506,
+                                369
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        357,
+                                        506,
+                                        369
+                                    ],
+                                    "score": 1.0,
+                                    "content": "updated under the “first-in first-out” strategy as distillation proceeds. That is, we en-queue the visual",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                368,
+                                506,
+                                380
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        368,
+                                        506,
+                                        380
+                                    ],
+                                    "score": 1.0,
+                                    "content": "features of the current batch inferred by the teacher and de-queue the earliest seen samples at the end",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                378,
+                                505,
+                                391
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        378,
+                                        331,
+                                        391
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of iteration. Note that the maintained samples in queue",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        332,
+                                        379,
+                                        342,
+                                        389
+                                    ],
+                                    "score": 0.52,
+                                    "content": "\\mathbf { D }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        342,
+                                        378,
+                                        505,
+                                        391
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are mostly random and irrelevant to the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                389,
+                                505,
+                                402
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        389,
+                                        166,
+                                        402
+                                    ],
+                                    "score": 1.0,
+                                    "content": "target instance",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        166,
+                                        391,
+                                        177,
+                                        401
+                                    ],
+                                    "score": 0.85,
+                                    "content": "\\mathbf { x } _ { i }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        177,
+                                        389,
+                                        505,
+                                        402
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Minimizing the cross entropy between the similarity score distribution computed",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                400,
+                                505,
+                                412
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        400,
+                                        251,
+                                        412
+                                    ],
+                                    "score": 1.0,
+                                    "content": "by the student and teacher based on",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        252,
+                                        401,
+                                        262,
+                                        411
+                                    ],
+                                    "score": 0.52,
+                                    "content": "\\mathbf { D }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        262,
+                                        400,
+                                        326,
+                                        412
+                                    ],
+                                    "score": 1.0,
+                                    "content": "softly contrasts",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        326,
+                                        402,
+                                        337,
+                                        412
+                                    ],
+                                    "score": 0.87,
+                                    "content": "\\mathbf { x } _ { i }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        338,
+                                        400,
+                                        505,
+                                        412
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with randomly selected samples, without",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                410,
+                                506,
+                                425
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        410,
+                                        506,
+                                        425
+                                    ],
+                                    "score": 1.0,
+                                    "content": "directly aligning with the teacher encoder. To address this problem, we add the teacher’s embedding",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                421,
+                                397,
+                                437
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        422,
+                                        125,
+                                        435
+                                    ],
+                                    "score": 0.89,
+                                    "content": "( \\mathbf { z } _ { i } ^ { T } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        126,
+                                        421,
+                                        224,
+                                        437
+                                    ],
+                                    "score": 1.0,
+                                    "content": "into the queue and form",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        225,
+                                        422,
+                                        322,
+                                        434
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\mathbf { D } ^ { + } = [ \\mathbf { d } _ { 1 } . . . \\mathbf { d } _ { K } , \\mathbf { d } _ { K + 1 } ]",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        322,
+                                        421,
+                                        343,
+                                        437
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        343,
+                                        422,
+                                        393,
+                                        435
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\mathbf { d } _ { K + 1 } = \\mathbf { z } _ { i } ^ { T }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        393,
+                                        421,
+                                        397,
+                                        437
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        }
+                    ],
+                    "index": 16.5,
+                    "bbox_fs": [
+                        104,
+                        300,
+                        506,
+                        437
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        440,
+                        504,
+                        465
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                439,
+                                505,
+                                453
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        439,
+                                        123,
+                                        453
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Let",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        123,
+                                        439,
+                                        189,
+                                        452
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\mathbf { p } ^ { T } ( \\mathbf { x } _ { i } ; \\boldsymbol { \\theta } _ { T } ; \\mathbf { D } ^ { + } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        189,
+                                        439,
+                                        454,
+                                        453
+                                    ],
+                                    "score": 1.0,
+                                    "content": "denote the similarity score between the extracted teacher feature",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        454,
+                                        439,
+                                        466,
+                                        452
+                                    ],
+                                    "score": 0.9,
+                                    "content": "\\mathbf { z } _ { i } ^ { T }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        467,
+                                        439,
+                                        485,
+                                        453
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        485,
+                                        440,
+                                        497,
+                                        453
+                                    ],
+                                    "score": 0.86,
+                                    "content": "{ \\bf d } _ { j }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        497,
+                                        439,
+                                        505,
+                                        453
+                                    ],
+                                    "score": 1.0,
+                                    "content": "’s",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                107,
+                                451,
+                                432,
+                                465
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        453,
+                                        179,
+                                        464
+                                    ],
+                                    "score": 0.89,
+                                    "content": "( j = 1 , . . . , K + 1 )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        180,
+                                        452,
+                                        312,
+                                        465
+                                    ],
+                                    "score": 1.0,
+                                    "content": "computed by the teacher model.",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        312,
+                                        451,
+                                        378,
+                                        465
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\mathbf { p } ^ { T } ( \\mathbf { x } _ { i } ; \\boldsymbol { \\theta } _ { T } ; \\mathbf { D } ^ { + } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        379,
+                                        452,
+                                        432,
+                                        465
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is defined as",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        }
+                    ],
+                    "index": 23.5,
+                    "bbox_fs": [
+                        105,
+                        439,
+                        505,
+                        465
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        158,
+                        468,
+                        452,
+                        496
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                158,
+                                468,
+                                452,
+                                496
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        158,
+                                        468,
+                                        452,
+                                        496
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\mathbf { p } ^ { T } ( \\mathbf { x } _ { i } ; \\theta _ { T } , \\mathbf { D } ^ { + } ) = \\left[ p _ { 1 } ^ { T } \\dots p _ { K + 1 } ^ { T } \\right] , \\qquad p _ { j } ^ { T } = \\frac { \\exp ( \\mathbf { z } _ { i } ^ { T } \\cdot \\mathbf { d } _ { j } / \\tau ^ { T } ) } { \\sum _ { \\mathbf { d } \\sim \\mathbf { D } ^ { + } } \\exp ( \\mathbf { z } _ { i } ^ { T } \\cdot \\mathbf { d } / \\tau ^ { T } ) } ,",
+                                    "type": "interline_equation",
+                                    "image_path": "fa03ea71bb692eacf77006e376709e981c9236cbb7b4dc642d83915ba5eb6316.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 26,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                158,
+                                468,
+                                452,
+                                477.3333333333333
+                            ],
+                            "spans": [],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                158,
+                                477.3333333333333,
+                                452,
+                                486.66666666666663
+                            ],
+                            "spans": [],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                158,
+                                486.66666666666663,
+                                452,
+                                495.99999999999994
+                            ],
+                            "spans": [],
+                            "index": 27
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        499,
+                        504,
+                        523
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                499,
+                                505,
+                                512
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        499,
+                                        123,
+                                        512
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        123,
+                                        499,
+                                        136,
+                                        510
+                                    ],
+                                    "score": 0.89,
+                                    "content": "\\tau ^ { T }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        136,
+                                        499,
+                                        360,
+                                        512
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is a temperature parameter for the teacher. Note, we use",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        360,
+                                        499,
+                                        375,
+                                        512
+                                    ],
+                                    "score": 0.9,
+                                    "content": "( ) ^ { T }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        375,
+                                        499,
+                                        505,
+                                        512
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to represent the feature from the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                510,
+                                428,
+                                523
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        510,
+                                        206,
+                                        523
+                                    ],
+                                    "score": 1.0,
+                                    "content": "teacher network and use",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        206,
+                                        511,
+                                        217,
+                                        523
+                                    ],
+                                    "score": 0.58,
+                                    "content": "( \\cdot )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        217,
+                                        510,
+                                        428,
+                                        523
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to represent the inner product between two features.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        }
+                    ],
+                    "index": 28.5,
+                    "bbox_fs": [
+                        105,
+                        499,
+                        505,
+                        523
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        527,
+                        504,
+                        550
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                525,
+                                506,
+                                542
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        525,
+                                        159,
+                                        542
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Similarly let",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        160,
+                                        527,
+                                        224,
+                                        540
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\mathbf { p } ^ { S } ( x _ { i } ; \\theta _ { S } , \\mathbf { D } ^ { + } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        225,
+                                        525,
+                                        506,
+                                        542
+                                    ],
+                                    "score": 1.0,
+                                    "content": "denote the similarity score computed by the student model, which is",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                537,
+                                151,
+                                551
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        537,
+                                        151,
+                                        551
+                                    ],
+                                    "score": 1.0,
+                                    "content": "defined as",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        }
+                    ],
+                    "index": 30.5,
+                    "bbox_fs": [
+                        105,
+                        525,
+                        506,
+                        551
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        146,
+                        551,
+                        464,
+                        580
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                146,
+                                551,
+                                464,
+                                580
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        146,
+                                        551,
+                                        464,
+                                        580
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\mathbf { p } ^ { S } ( \\mathbf { x } _ { i } ; \\theta _ { S } , \\mathbf { D } ^ { + } ) = \\left[ p _ { 1 } ^ { S } \\dots p _ { K + 1 } ^ { S } \\right] , \\qquad { \\mathrm { w h e r e ~ } } p _ { j } ^ { S } = { \\frac { \\exp ( \\mathbf { z } _ { i } ^ { S } \\cdot \\mathbf { d } _ { j } / \\tau ^ { S } ) } { \\sum _ { \\mathbf { d } \\sim \\mathbf { D } ^ { + } } \\exp ( \\mathbf { z } _ { i } ^ { S } \\cdot \\mathbf { d } / \\tau ^ { S } ) } } ,",
+                                    "type": "interline_equation",
+                                    "image_path": "7a86dd461ab60bef53314bfb98d8072c5f8f7ea6f386125f046180e8c76b6966.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 33,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                146,
+                                551,
+                                464,
+                                560.6666666666666
+                            ],
+                            "spans": [],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                146,
+                                560.6666666666666,
+                                464,
+                                570.3333333333333
+                            ],
+                            "spans": [],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                146,
+                                570.3333333333333,
+                                464,
+                                579.9999999999999
+                            ],
+                            "spans": [],
+                            "index": 34
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        583,
+                        307,
+                        596
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                581,
+                                308,
+                                597
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        581,
+                                        123,
+                                        597
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        123,
+                                        583,
+                                        136,
+                                        594
+                                    ],
+                                    "score": 0.88,
+                                    "content": "\\tau ^ { S }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        136,
+                                        581,
+                                        308,
+                                        597
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is a temperature parameter for the student.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        }
+                    ],
+                    "index": 35,
+                    "bbox_fs": [
+                        105,
+                        581,
+                        308,
+                        597
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        600,
+                        504,
+                        634
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                600,
+                                505,
+                                613
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        600,
+                                        505,
+                                        613
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Our self-supervised distillation can be formulated as minimizing the cross entropy between the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                611,
+                                506,
+                                624
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        611,
+                                        242,
+                                        624
+                                    ],
+                                    "score": 1.0,
+                                    "content": "similarity scores of the teacher,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        242,
+                                        611,
+                                        308,
+                                        624
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\mathbf { p } ^ { T } ( \\mathbf { x } _ { i } ; \\theta _ { T } , \\mathbf { D } ^ { + } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        308,
+                                        611,
+                                        384,
+                                        624
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", and the student,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        384,
+                                        611,
+                                        449,
+                                        624
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\mathbf { p } ^ { S } ( \\mathbf { x } _ { i } ; \\boldsymbol { \\theta } _ { S } , \\mathbf { D } ^ { \\dagger } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        449,
+                                        611,
+                                        506,
+                                        624
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", over all the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                622,
+                                189,
+                                635
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        622,
+                                        145,
+                                        635
+                                    ],
+                                    "score": 1.0,
+                                    "content": "instances",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        145,
+                                        624,
+                                        156,
+                                        634
+                                    ],
+                                    "score": 0.86,
+                                    "content": "\\mathbf { x } _ { i }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        156,
+                                        622,
+                                        189,
+                                        635
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", that is,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        }
+                    ],
+                    "index": 37,
+                    "bbox_fs": [
+                        105,
+                        600,
+                        506,
+                        635
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        138,
+                        636,
+                        472,
+                        708
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                138,
+                                636,
+                                472,
+                                708
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        138,
+                                        636,
+                                        472,
+                                        708
+                                    ],
+                                    "score": 0.95,
+                                    "content": "\\begin{array} { r l } & { \\boldsymbol { \\hat { \\theta } _ { S } } = \\underset { \\boldsymbol { \\theta _ { S } } } { \\arg \\operatorname* { m i n } } \\sum _ { i } ^ { N } - \\mathbf { p } ^ { T } ( \\mathbf { x } _ { i } ; \\boldsymbol { \\theta _ { T } } , \\mathbf { D } ^ { + } ) \\cdot \\log \\mathbf { p } ^ { S } ( \\mathbf { x } _ { i } ; \\boldsymbol { \\theta _ { S } } , \\mathbf { D } ^ { + } ) } \\\\ & { \\quad = \\underset { \\boldsymbol { \\theta _ { S } } } { \\arg \\operatorname* { m i n } } \\sum _ { i } ^ { N } \\sum _ { j } ^ { K + 1 } - \\frac { \\exp \\left( \\mathbf { z } _ { i } ^ { T } \\cdot \\mathbf { d } _ { j } / \\tau ^ { T } \\right) } { \\sum _ { \\mathbf { d } \\sim \\mathbf { D } ^ { + } } \\exp \\left( \\mathbf { z } _ { i } ^ { T } \\cdot \\mathbf { d } / \\tau ^ { T } \\right) } \\cdot \\log \\frac { \\exp \\left( \\mathbf { z } _ { i } ^ { S } \\cdot \\mathbf { d } _ { j } / \\tau ^ { S } \\right) } { \\sum _ { \\mathbf { d } \\sim \\mathbf { D } ^ { + } } \\exp \\left( \\mathbf { z } _ { i } ^ { S } \\cdot \\mathbf { d } / \\tau ^ { S } \\right) } . } \\end{array}",
+                                    "type": "interline_equation",
+                                    "image_path": "ebb8b97ca7f12d289494242cd41ec751bf46c453936e79d85361fe539a0cb9bd.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 40,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                138,
+                                636,
+                                472,
+                                660.0
+                            ],
+                            "spans": [],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                138,
+                                660.0,
+                                472,
+                                684.0
+                            ],
+                            "spans": [],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                138,
+                                684.0,
+                                472,
+                                708.0
+                            ],
+                            "spans": [],
+                            "index": 41
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        709,
+                        505,
+                        733
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                708,
+                                505,
+                                723
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        708,
+                                        505,
+                                        723
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Since the teacher network is pre-trained and frozen, the queued features are consistent during training",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                720,
+                                505,
+                                734
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        720,
+                                        312,
+                                        734
+                                    ],
+                                    "score": 1.0,
+                                    "content": "w.r.t. the student network. The higher the value of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        312,
+                                        721,
+                                        324,
+                                        734
+                                    ],
+                                    "score": 0.91,
+                                    "content": "p _ { j } ^ { I ^ { \\prime } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        324,
+                                        720,
+                                        469,
+                                        734
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is, the larger weight will be laid on",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        469,
+                                        721,
+                                        481,
+                                        734
+                                    ],
+                                    "score": 0.89,
+                                    "content": "p _ { j } ^ { S }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        481,
+                                        720,
+                                        505,
+                                        734
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Due",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                81,
+                                505,
+                                96
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        81,
+                                        305,
+                                        96
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to the l2 normalization, similarity score between",
+                                    "type": "text",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        305,
+                                        82,
+                                        318,
+                                        95
+                                    ],
+                                    "score": 0.91,
+                                    "content": "{ \\mathbf z } _ { i } ^ { T }",
+                                    "type": "inline_equation",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        318,
+                                        81,
+                                        336,
+                                        96
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        336,
+                                        82,
+                                        362,
+                                        94
+                                    ],
+                                    "score": 0.92,
+                                    "content": "{ \\bf d } _ { K + 1 }",
+                                    "type": "inline_equation",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        362,
+                                        81,
+                                        505,
+                                        96
+                                    ],
+                                    "score": 1.0,
+                                    "content": "remains constant 1 before softmax",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                92,
+                                506,
+                                109
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        92,
+                                        278,
+                                        109
+                                    ],
+                                    "score": 1.0,
+                                    "content": "normalization, which is the largest among",
+                                    "type": "text",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        279,
+                                        93,
+                                        291,
+                                        108
+                                    ],
+                                    "score": 0.9,
+                                    "content": "p _ { j } ^ { T }",
+                                    "type": "inline_equation",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        291,
+                                        92,
+                                        380,
+                                        109
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Thus, the weight for",
+                                    "type": "text",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        380,
+                                        93,
+                                        405,
+                                        108
+                                    ],
+                                    "score": 0.93,
+                                    "content": "p _ { K + 1 } ^ { S }",
+                                    "type": "inline_equation",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        405,
+                                        92,
+                                        506,
+                                        109
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is the largest and can be",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                107,
+                                506,
+                                120
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        107,
+                                        259,
+                                        120
+                                    ],
+                                    "score": 1.0,
+                                    "content": "adjusted solely by tuning the value of",
+                                    "type": "text",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        259,
+                                        107,
+                                        272,
+                                        118
+                                    ],
+                                    "score": 0.87,
+                                    "content": "\\tau ^ { T }",
+                                    "type": "inline_equation",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        272,
+                                        107,
+                                        431,
+                                        120
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". By minimizing the loss, the feature of",
+                                    "type": "text",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        431,
+                                        107,
+                                        443,
+                                        120
+                                    ],
+                                    "score": 0.9,
+                                    "content": "\\mathbf { z } _ { i } ^ { S }",
+                                    "type": "inline_equation",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        443,
+                                        107,
+                                        506,
+                                        120
+                                    ],
+                                    "score": 1.0,
+                                    "content": "can be aligned",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                119,
+                                505,
+                                132
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        119,
+                                        126,
+                                        132
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with",
+                                    "type": "text",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        127,
+                                        119,
+                                        139,
+                                        132
+                                    ],
+                                    "score": 0.9,
+                                    "content": "{ \\mathbf z } _ { i } ^ { T }",
+                                    "type": "inline_equation",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        140,
+                                        119,
+                                        398,
+                                        132
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and meanwhile contrasts with other unrelated image features in",
+                                    "type": "text",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        398,
+                                        120,
+                                        408,
+                                        130
+                                    ],
+                                    "score": 0.45,
+                                    "content": "\\mathbf { D }",
+                                    "type": "inline_equation",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        409,
+                                        119,
+                                        505,
+                                        132
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". We further discuss the",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                131,
+                                395,
+                                143
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        131,
+                                        395,
+                                        143
+                                    ],
+                                    "score": 1.0,
+                                    "content": "relation of these two goals with our learning objective in Appendix A.5.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 4
+                        }
+                    ],
+                    "index": 42.5,
+                    "bbox_fs": [
+                        105,
+                        708,
+                        505,
+                        734
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        81,
+                        505,
+                        142
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                81,
+                                505,
+                                96
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        81,
+                                        305,
+                                        96
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to the l2 normalization, similarity score between",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        305,
+                                        82,
+                                        318,
+                                        95
+                                    ],
+                                    "score": 0.91,
+                                    "content": "{ \\mathbf z } _ { i } ^ { T }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        318,
+                                        81,
+                                        336,
+                                        96
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        336,
+                                        82,
+                                        362,
+                                        94
+                                    ],
+                                    "score": 0.92,
+                                    "content": "{ \\bf d } _ { K + 1 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        362,
+                                        81,
+                                        505,
+                                        96
+                                    ],
+                                    "score": 1.0,
+                                    "content": "remains constant 1 before softmax",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                92,
+                                506,
+                                109
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        92,
+                                        278,
+                                        109
+                                    ],
+                                    "score": 1.0,
+                                    "content": "normalization, which is the largest among",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        279,
+                                        93,
+                                        291,
+                                        108
+                                    ],
+                                    "score": 0.9,
+                                    "content": "p _ { j } ^ { T }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        291,
+                                        92,
+                                        380,
+                                        109
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Thus, the weight for",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        380,
+                                        93,
+                                        405,
+                                        108
+                                    ],
+                                    "score": 0.93,
+                                    "content": "p _ { K + 1 } ^ { S }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        405,
+                                        92,
+                                        506,
+                                        109
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is the largest and can be",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                107,
+                                506,
+                                120
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        107,
+                                        259,
+                                        120
+                                    ],
+                                    "score": 1.0,
+                                    "content": "adjusted solely by tuning the value of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        259,
+                                        107,
+                                        272,
+                                        118
+                                    ],
+                                    "score": 0.87,
+                                    "content": "\\tau ^ { T }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        272,
+                                        107,
+                                        431,
+                                        120
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". By minimizing the loss, the feature of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        431,
+                                        107,
+                                        443,
+                                        120
+                                    ],
+                                    "score": 0.9,
+                                    "content": "\\mathbf { z } _ { i } ^ { S }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        443,
+                                        107,
+                                        506,
+                                        120
+                                    ],
+                                    "score": 1.0,
+                                    "content": "can be aligned",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                119,
+                                505,
+                                132
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        119,
+                                        126,
+                                        132
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        127,
+                                        119,
+                                        139,
+                                        132
+                                    ],
+                                    "score": 0.9,
+                                    "content": "{ \\mathbf z } _ { i } ^ { T }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        140,
+                                        119,
+                                        398,
+                                        132
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and meanwhile contrasts with other unrelated image features in",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        398,
+                                        120,
+                                        408,
+                                        130
+                                    ],
+                                    "score": 0.45,
+                                    "content": "\\mathbf { D }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        409,
+                                        119,
+                                        505,
+                                        132
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". We further discuss the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                131,
+                                395,
+                                143
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        131,
+                                        395,
+                                        143
+                                    ],
+                                    "score": 1.0,
+                                    "content": "relation of these two goals with our learning objective in Appendix A.5.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        }
+                    ],
+                    "index": 2
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        146,
+                        506,
+                        171
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                145,
+                                506,
+                                160
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        145,
+                                        261,
+                                        160
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Relations with Info-NCE loss. When",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        261,
+                                        146,
+                                        294,
+                                        158
+                                    ],
+                                    "score": 0.9,
+                                    "content": "\\tau ^ { T } \\to 0",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        294,
+                                        145,
+                                        394,
+                                        160
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", the softmax function for",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        395,
+                                        147,
+                                        408,
+                                        159
+                                    ],
+                                    "score": 0.89,
+                                    "content": "\\mathbf { p } ^ { T }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        409,
+                                        145,
+                                        506,
+                                        160
+                                    ],
+                                    "score": 1.0,
+                                    "content": "smoothly approaches to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                158,
+                                485,
+                                172
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        158,
+                                        202,
+                                        172
+                                    ],
+                                    "score": 1.0,
+                                    "content": "a one-hot vector, where",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        203,
+                                        158,
+                                        226,
+                                        172
+                                    ],
+                                    "score": 0.93,
+                                    "content": "p _ { K + 1 } ^ { T }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        227,
+                                        158,
+                                        485,
+                                        172
+                                    ],
+                                    "score": 1.0,
+                                    "content": "equals 1 and all others 0. In this extreme case, the loss becomes",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        }
+                    ],
+                    "index": 5.5
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        217,
+                        177,
+                        393,
+                        212
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                217,
+                                177,
+                                393,
+                                212
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        217,
+                                        177,
+                                        393,
+                                        212
+                                    ],
+                                    "score": 0.95,
+                                    "content": "\\mathcal { L } _ { N C E } = \\sum _ { i } ^ { N } - \\log \\frac { \\exp ( \\mathbf { z } _ { i } ^ { T } \\cdot \\mathbf { z } _ { i } ^ { S } / \\tau ) } { \\sum _ { \\mathbf { d } \\sim \\mathbf { D } ^ { + } } \\exp ( \\mathbf { z } _ { i } ^ { S } \\cdot \\mathbf { d } / \\tau ) } ,",
+                                    "type": "interline_equation",
+                                    "image_path": "e089e43867b6fe00aefe91b5ac56af8b29248c22fa3a0c345faa7cdfe89a5b9b.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 7.5,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                217,
+                                177,
+                                393,
+                                194.5
+                            ],
+                            "spans": [],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                217,
+                                194.5,
+                                393,
+                                212.0
+                            ],
+                            "spans": [],
+                            "index": 8
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        216,
+                        506,
+                        239
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                215,
+                                505,
+                                229
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        215,
+                                        505,
+                                        229
+                                    ],
+                                    "score": 1.0,
+                                    "content": "which is similar to the widely-used Info-NCE loss (Oord et al., 2018) in contrastive-based SSL (see",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                227,
+                                223,
+                                240
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        227,
+                                        223,
+                                        240
+                                    ],
+                                    "score": 1.0,
+                                    "content": "discussion in Appendix A.6.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        }
+                    ],
+                    "index": 9.5
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        255,
+                        194,
+                        267
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                254,
+                                196,
+                                270
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        254,
+                                        196,
+                                        270
+                                    ],
+                                    "score": 1.0,
+                                    "content": "4 EXPERIMENT",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        }
+                    ],
+                    "index": 11
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        280,
+                        195,
+                        291
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                280,
+                                196,
+                                293
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        280,
+                                        196,
+                                        293
+                                    ],
+                                    "score": 1.0,
+                                    "content": "4.1 PRE-TRAINING",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        }
+                    ],
+                    "index": 12
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        300,
+                        505,
+                        400
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                301,
+                                505,
+                                313
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        301,
+                                        505,
+                                        313
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Self-Supervised Pre-training of Teacher Network. By default, we use MoCo-V2 (Chen et al.,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                312,
+                                506,
+                                324
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        312,
+                                        506,
+                                        324
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2020c) to pre-train the teacher network. Following (Chen et al., 2020a), we use ResNet as the network",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                323,
+                                506,
+                                335
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        323,
+                                        506,
+                                        335
+                                    ],
+                                    "score": 1.0,
+                                    "content": "backbone with different depths/widths and append a multi-layer-perceptron (MLP) layer (two linear",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                334,
+                                505,
+                                345
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        334,
+                                        505,
+                                        345
+                                    ],
+                                    "score": 1.0,
+                                    "content": "layers and one ReLU (Nair & Hinton, 2010) activation layer in between) at the end of the encoder",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                345,
+                                505,
+                                357
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        345,
+                                        505,
+                                        357
+                                    ],
+                                    "score": 1.0,
+                                    "content": "after average pooling. The dimension of the last feature dimension is 128. All teacher networks",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                355,
+                                506,
+                                368
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        355,
+                                        506,
+                                        368
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are pre-trained for 200 epochs due to the computational limitation unless explicitly specified. As",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                367,
+                                505,
+                                379
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        367,
+                                        505,
+                                        379
+                                    ],
+                                    "score": 1.0,
+                                    "content": "our distillation is independent with the teacher pre-training algorithm, we also show results with",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                377,
+                                506,
+                                390
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        377,
+                                        506,
+                                        390
+                                    ],
+                                    "score": 1.0,
+                                    "content": "other self-supervised pre-trained models for teacher network, e.g., SWAV (Caron et al., 2020),",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                388,
+                                228,
+                                401
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        388,
+                                        228,
+                                        401
+                                    ],
+                                    "score": 1.0,
+                                    "content": "SimCLR (Chen et al., 2020a).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        }
+                    ],
+                    "index": 17
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        405,
+                        505,
+                        516
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                406,
+                                505,
+                                417
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        406,
+                                        505,
+                                        417
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Self-Supervised Distillation on Student Network. We choose multiple smaller networks with",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                416,
+                                506,
+                                428
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        416,
+                                        506,
+                                        428
+                                    ],
+                                    "score": 1.0,
+                                    "content": "fewer learnable parameters as the student network: MobileNet-v3-Large (Howard et al., 2017),",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                426,
+                                505,
+                                439
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        426,
+                                        505,
+                                        439
+                                    ],
+                                    "score": 1.0,
+                                    "content": "EfficientNet-B0 (Tan & Le, 2019), and smaller ResNet with fewer layers (ResNet-18, 34). Similar to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                438,
+                                505,
+                                451
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        438,
+                                        505,
+                                        451
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the pre-training for teacher network, we add one additional MLP layer on the basis of the student",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                450,
+                                505,
+                                461
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        450,
+                                        505,
+                                        461
+                                    ],
+                                    "score": 1.0,
+                                    "content": "network. Our distillation is trained with a standard SGD optimizer with momentum 0.9 and a weight",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                460,
+                                506,
+                                473
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        460,
+                                        506,
+                                        473
+                                    ],
+                                    "score": 1.0,
+                                    "content": "decay parameter of 1e-4 for 200 epochs. The initial learning rate is set as 0.03 and updated by a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                471,
+                                506,
+                                484
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        471,
+                                        506,
+                                        484
+                                    ],
+                                    "score": 1.0,
+                                    "content": "cosine decay scheduler (Nair & Hinton, 2010) with 5 warm-up epochs and batch size 256. In Eq. 4,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                480,
+                                506,
+                                495
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        480,
+                                        240,
+                                        495
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the teacher temperature is set as",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        241,
+                                        482,
+                                        286,
+                                        493
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\tau ^ { T } = 0 . 0 1",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        286,
+                                        480,
+                                        414,
+                                        495
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and the student temperature is",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        414,
+                                        482,
+                                        454,
+                                        493
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\tau ^ { S } = 0 . 2",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        454,
+                                        480,
+                                        506,
+                                        495
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". The queue",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                493,
+                                505,
+                                505
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        493,
+                                        135,
+                                        505
+                                    ],
+                                    "score": 1.0,
+                                    "content": "size of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        136,
+                                        493,
+                                        146,
+                                        503
+                                    ],
+                                    "score": 0.82,
+                                    "content": "K",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        146,
+                                        493,
+                                        505,
+                                        505
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is 65,536. In the following subsections and appendix, we also show results with different",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                504,
+                                284,
+                                516
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        505,
+                                        237,
+                                        516
+                                    ],
+                                    "score": 1.0,
+                                    "content": "hyper-parameter values, e.g., for",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        238,
+                                        504,
+                                        250,
+                                        514
+                                    ],
+                                    "score": 0.89,
+                                    "content": "\\tau ^ { T }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        251,
+                                        505,
+                                        269,
+                                        516
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        269,
+                                        505,
+                                        279,
+                                        514
+                                    ],
+                                    "score": 0.82,
+                                    "content": "K",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        280,
+                                        505,
+                                        284,
+                                        516
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        }
+                    ],
+                    "index": 26.5
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        529,
+                        268,
+                        541
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                529,
+                                269,
+                                542
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        529,
+                                        269,
+                                        542
+                                    ],
+                                    "score": 1.0,
+                                    "content": "4.2 FINE-TUNING AND EVALUATION",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        }
+                    ],
+                    "index": 32
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        550,
+                        505,
+                        638
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                551,
+                                506,
+                                561
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        551,
+                                        506,
+                                        561
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In order to validate the effectiveness of self-supervised distillation, we choose to assess the perfor-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                562,
+                                505,
+                                573
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        562,
+                                        505,
+                                        573
+                                    ],
+                                    "score": 1.0,
+                                    "content": "mance of representations of the student encoder on several downstream tasks. We first report its",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                573,
+                                506,
+                                583
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        573,
+                                        506,
+                                        583
+                                    ],
+                                    "score": 1.0,
+                                    "content": "performances of linear evaluation and semi-supervised linear evaluation on the ImageNet ILSVRC-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                582,
+                                505,
+                                595
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        582,
+                                        505,
+                                        595
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2012 (Deng et al., 2009) dataset. To measure the feature transferability brought by distillation, we",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                594,
+                                505,
+                                606
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        594,
+                                        505,
+                                        606
+                                    ],
+                                    "score": 1.0,
+                                    "content": "also conduct evaluations on other tasks, which include object detection and segmentation on the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                604,
+                                505,
+                                617
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        604,
+                                        505,
+                                        617
+                                    ],
+                                    "score": 1.0,
+                                    "content": "VOC07 (Everingham et al.) and MS-COCO (Lin et al., 2014) datasets. At the end, we compare the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                615,
+                                506,
+                                630
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        615,
+                                        506,
+                                        630
+                                    ],
+                                    "score": 1.0,
+                                    "content": "transferability of the features learned by distillation with ordinary self-supervised contrastive learning",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                626,
+                                390,
+                                639
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        626,
+                                        390,
+                                        639
+                                    ],
+                                    "score": 1.0,
+                                    "content": "on the tasks of linear classification on datasets from different domains.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        }
+                    ],
+                    "index": 36.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        643,
+                        505,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                644,
+                                505,
+                                655
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        644,
+                                        505,
+                                        655
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Linear and KNN Evaluation on ImageNet. We conduct the supervised linear classification",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                655,
+                                506,
+                                667
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        655,
+                                        246,
+                                        667
+                                    ],
+                                    "score": 1.0,
+                                    "content": "on ImageNet-1K, which contains",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        246,
+                                        655,
+                                        278,
+                                        666
+                                    ],
+                                    "score": 0.83,
+                                    "content": "{ \\sim } 1 . 3 \\mathbf { M }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        279,
+                                        655,
+                                        506,
+                                        667
+                                    ],
+                                    "score": 1.0,
+                                    "content": "images for training, and 50,000 images for validation,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                665,
+                                506,
+                                678
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        665,
+                                        506,
+                                        678
+                                    ],
+                                    "score": 1.0,
+                                    "content": "spanning 1,000 categories. Following previous works in (He et al., 2020; Chen et al., 2020a), we",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                677,
+                                506,
+                                689
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        677,
+                                        506,
+                                        689
+                                    ],
+                                    "score": 1.0,
+                                    "content": "train a single linear layer classifier on top of the frozen network encoder after self-supervised pre-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                687,
+                                506,
+                                700
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        687,
+                                        506,
+                                        700
+                                    ],
+                                    "score": 1.0,
+                                    "content": "training/distillation. SGD optimizer is used to train the linear classifier for 100 epochs with weight",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                699,
+                                505,
+                                710
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        699,
+                                        505,
+                                        710
+                                    ],
+                                    "score": 1.0,
+                                    "content": "decay to be 0. The initial learning rate is set as 30 and is then reduced by a factor of 10 at 60 and 80",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                710,
+                                506,
+                                721
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        710,
+                                        506,
+                                        721
+                                    ],
+                                    "score": 1.0,
+                                    "content": "epochs (similar as in Tian et al. (2019a)). Notably, when training the linear classifier for MobileNet",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 47
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                721,
+                                505,
+                                733
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        721,
+                                        505,
+                                        733
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and EfficientNet, we reduce the initial learning rate to 3. The results are reported in terms of Top-1",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 48
+                        }
+                    ],
+                    "index": 44.5
+                }
+            ],
+            "page_idx": 4,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        108,
+                        27,
+                        292,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                26,
+                                293,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        26,
+                                        293,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Published as a conference paper at ICLR 2021",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        302,
+                        751,
+                        308,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                302,
+                                750,
+                                309,
+                                763
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        302,
+                                        750,
+                                        309,
+                                        763
+                                    ],
+                                    "score": 1.0,
+                                    "content": "5",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        81,
+                        505,
+                        142
+                    ],
+                    "lines": [],
+                    "index": 2,
+                    "bbox_fs": [
+                        104,
+                        81,
+                        506,
+                        143
+                    ],
+                    "lines_deleted": true
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        146,
+                        506,
+                        171
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                145,
+                                506,
+                                160
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        145,
+                                        261,
+                                        160
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Relations with Info-NCE loss. When",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        261,
+                                        146,
+                                        294,
+                                        158
+                                    ],
+                                    "score": 0.9,
+                                    "content": "\\tau ^ { T } \\to 0",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        294,
+                                        145,
+                                        394,
+                                        160
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", the softmax function for",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        395,
+                                        147,
+                                        408,
+                                        159
+                                    ],
+                                    "score": 0.89,
+                                    "content": "\\mathbf { p } ^ { T }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        409,
+                                        145,
+                                        506,
+                                        160
+                                    ],
+                                    "score": 1.0,
+                                    "content": "smoothly approaches to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                158,
+                                485,
+                                172
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        158,
+                                        202,
+                                        172
+                                    ],
+                                    "score": 1.0,
+                                    "content": "a one-hot vector, where",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        203,
+                                        158,
+                                        226,
+                                        172
+                                    ],
+                                    "score": 0.93,
+                                    "content": "p _ { K + 1 } ^ { T }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        227,
+                                        158,
+                                        485,
+                                        172
+                                    ],
+                                    "score": 1.0,
+                                    "content": "equals 1 and all others 0. In this extreme case, the loss becomes",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        }
+                    ],
+                    "index": 5.5,
+                    "bbox_fs": [
+                        104,
+                        145,
+                        506,
+                        172
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        217,
+                        177,
+                        393,
+                        212
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                217,
+                                177,
+                                393,
+                                212
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        217,
+                                        177,
+                                        393,
+                                        212
+                                    ],
+                                    "score": 0.95,
+                                    "content": "\\mathcal { L } _ { N C E } = \\sum _ { i } ^ { N } - \\log \\frac { \\exp ( \\mathbf { z } _ { i } ^ { T } \\cdot \\mathbf { z } _ { i } ^ { S } / \\tau ) } { \\sum _ { \\mathbf { d } \\sim \\mathbf { D } ^ { + } } \\exp ( \\mathbf { z } _ { i } ^ { S } \\cdot \\mathbf { d } / \\tau ) } ,",
+                                    "type": "interline_equation",
+                                    "image_path": "e089e43867b6fe00aefe91b5ac56af8b29248c22fa3a0c345faa7cdfe89a5b9b.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 7.5,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                217,
+                                177,
+                                393,
+                                194.5
+                            ],
+                            "spans": [],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                217,
+                                194.5,
+                                393,
+                                212.0
+                            ],
+                            "spans": [],
+                            "index": 8
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        216,
+                        506,
+                        239
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                215,
+                                505,
+                                229
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        215,
+                                        505,
+                                        229
+                                    ],
+                                    "score": 1.0,
+                                    "content": "which is similar to the widely-used Info-NCE loss (Oord et al., 2018) in contrastive-based SSL (see",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                227,
+                                223,
+                                240
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        227,
+                                        223,
+                                        240
+                                    ],
+                                    "score": 1.0,
+                                    "content": "discussion in Appendix A.6.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        }
+                    ],
+                    "index": 9.5,
+                    "bbox_fs": [
+                        105,
+                        215,
+                        505,
+                        240
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        255,
+                        194,
+                        267
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                254,
+                                196,
+                                270
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        254,
+                                        196,
+                                        270
+                                    ],
+                                    "score": 1.0,
+                                    "content": "4 EXPERIMENT",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        }
+                    ],
+                    "index": 11
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        280,
+                        195,
+                        291
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                280,
+                                196,
+                                293
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        280,
+                                        196,
+                                        293
+                                    ],
+                                    "score": 1.0,
+                                    "content": "4.1 PRE-TRAINING",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        }
+                    ],
+                    "index": 12
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        300,
+                        505,
+                        400
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                301,
+                                505,
+                                313
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        301,
+                                        505,
+                                        313
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Self-Supervised Pre-training of Teacher Network. By default, we use MoCo-V2 (Chen et al.,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                312,
+                                506,
+                                324
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        312,
+                                        506,
+                                        324
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2020c) to pre-train the teacher network. Following (Chen et al., 2020a), we use ResNet as the network",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                323,
+                                506,
+                                335
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        323,
+                                        506,
+                                        335
+                                    ],
+                                    "score": 1.0,
+                                    "content": "backbone with different depths/widths and append a multi-layer-perceptron (MLP) layer (two linear",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                334,
+                                505,
+                                345
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        334,
+                                        505,
+                                        345
+                                    ],
+                                    "score": 1.0,
+                                    "content": "layers and one ReLU (Nair & Hinton, 2010) activation layer in between) at the end of the encoder",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                345,
+                                505,
+                                357
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        345,
+                                        505,
+                                        357
+                                    ],
+                                    "score": 1.0,
+                                    "content": "after average pooling. The dimension of the last feature dimension is 128. All teacher networks",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                355,
+                                506,
+                                368
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        355,
+                                        506,
+                                        368
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are pre-trained for 200 epochs due to the computational limitation unless explicitly specified. As",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                367,
+                                505,
+                                379
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        367,
+                                        505,
+                                        379
+                                    ],
+                                    "score": 1.0,
+                                    "content": "our distillation is independent with the teacher pre-training algorithm, we also show results with",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                377,
+                                506,
+                                390
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        377,
+                                        506,
+                                        390
+                                    ],
+                                    "score": 1.0,
+                                    "content": "other self-supervised pre-trained models for teacher network, e.g., SWAV (Caron et al., 2020),",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                388,
+                                228,
+                                401
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        388,
+                                        228,
+                                        401
+                                    ],
+                                    "score": 1.0,
+                                    "content": "SimCLR (Chen et al., 2020a).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        }
+                    ],
+                    "index": 17,
+                    "bbox_fs": [
+                        105,
+                        301,
+                        506,
+                        401
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        405,
+                        505,
+                        516
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                406,
+                                505,
+                                417
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        406,
+                                        505,
+                                        417
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Self-Supervised Distillation on Student Network. We choose multiple smaller networks with",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                416,
+                                506,
+                                428
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        416,
+                                        506,
+                                        428
+                                    ],
+                                    "score": 1.0,
+                                    "content": "fewer learnable parameters as the student network: MobileNet-v3-Large (Howard et al., 2017),",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                426,
+                                505,
+                                439
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        426,
+                                        505,
+                                        439
+                                    ],
+                                    "score": 1.0,
+                                    "content": "EfficientNet-B0 (Tan & Le, 2019), and smaller ResNet with fewer layers (ResNet-18, 34). Similar to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                438,
+                                505,
+                                451
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        438,
+                                        505,
+                                        451
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the pre-training for teacher network, we add one additional MLP layer on the basis of the student",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                450,
+                                505,
+                                461
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        450,
+                                        505,
+                                        461
+                                    ],
+                                    "score": 1.0,
+                                    "content": "network. Our distillation is trained with a standard SGD optimizer with momentum 0.9 and a weight",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                460,
+                                506,
+                                473
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        460,
+                                        506,
+                                        473
+                                    ],
+                                    "score": 1.0,
+                                    "content": "decay parameter of 1e-4 for 200 epochs. The initial learning rate is set as 0.03 and updated by a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                471,
+                                506,
+                                484
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        471,
+                                        506,
+                                        484
+                                    ],
+                                    "score": 1.0,
+                                    "content": "cosine decay scheduler (Nair & Hinton, 2010) with 5 warm-up epochs and batch size 256. In Eq. 4,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                480,
+                                506,
+                                495
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        480,
+                                        240,
+                                        495
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the teacher temperature is set as",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        241,
+                                        482,
+                                        286,
+                                        493
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\tau ^ { T } = 0 . 0 1",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        286,
+                                        480,
+                                        414,
+                                        495
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and the student temperature is",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        414,
+                                        482,
+                                        454,
+                                        493
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\tau ^ { S } = 0 . 2",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        454,
+                                        480,
+                                        506,
+                                        495
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". The queue",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                493,
+                                505,
+                                505
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        493,
+                                        135,
+                                        505
+                                    ],
+                                    "score": 1.0,
+                                    "content": "size of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        136,
+                                        493,
+                                        146,
+                                        503
+                                    ],
+                                    "score": 0.82,
+                                    "content": "K",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        146,
+                                        493,
+                                        505,
+                                        505
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is 65,536. In the following subsections and appendix, we also show results with different",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                504,
+                                284,
+                                516
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        505,
+                                        237,
+                                        516
+                                    ],
+                                    "score": 1.0,
+                                    "content": "hyper-parameter values, e.g., for",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        238,
+                                        504,
+                                        250,
+                                        514
+                                    ],
+                                    "score": 0.89,
+                                    "content": "\\tau ^ { T }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        251,
+                                        505,
+                                        269,
+                                        516
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        269,
+                                        505,
+                                        279,
+                                        514
+                                    ],
+                                    "score": 0.82,
+                                    "content": "K",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        280,
+                                        505,
+                                        284,
+                                        516
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        }
+                    ],
+                    "index": 26.5,
+                    "bbox_fs": [
+                        105,
+                        406,
+                        506,
+                        516
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        529,
+                        268,
+                        541
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                529,
+                                269,
+                                542
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        529,
+                                        269,
+                                        542
+                                    ],
+                                    "score": 1.0,
+                                    "content": "4.2 FINE-TUNING AND EVALUATION",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        }
+                    ],
+                    "index": 32
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        550,
+                        505,
+                        638
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                551,
+                                506,
+                                561
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        551,
+                                        506,
+                                        561
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In order to validate the effectiveness of self-supervised distillation, we choose to assess the perfor-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                562,
+                                505,
+                                573
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        562,
+                                        505,
+                                        573
+                                    ],
+                                    "score": 1.0,
+                                    "content": "mance of representations of the student encoder on several downstream tasks. We first report its",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                573,
+                                506,
+                                583
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        573,
+                                        506,
+                                        583
+                                    ],
+                                    "score": 1.0,
+                                    "content": "performances of linear evaluation and semi-supervised linear evaluation on the ImageNet ILSVRC-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                582,
+                                505,
+                                595
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        582,
+                                        505,
+                                        595
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2012 (Deng et al., 2009) dataset. To measure the feature transferability brought by distillation, we",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                594,
+                                505,
+                                606
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        594,
+                                        505,
+                                        606
+                                    ],
+                                    "score": 1.0,
+                                    "content": "also conduct evaluations on other tasks, which include object detection and segmentation on the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                604,
+                                505,
+                                617
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        604,
+                                        505,
+                                        617
+                                    ],
+                                    "score": 1.0,
+                                    "content": "VOC07 (Everingham et al.) and MS-COCO (Lin et al., 2014) datasets. At the end, we compare the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                615,
+                                506,
+                                630
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        615,
+                                        506,
+                                        630
+                                    ],
+                                    "score": 1.0,
+                                    "content": "transferability of the features learned by distillation with ordinary self-supervised contrastive learning",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                626,
+                                390,
+                                639
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        626,
+                                        390,
+                                        639
+                                    ],
+                                    "score": 1.0,
+                                    "content": "on the tasks of linear classification on datasets from different domains.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        }
+                    ],
+                    "index": 36.5,
+                    "bbox_fs": [
+                        105,
+                        551,
+                        506,
+                        639
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        643,
+                        505,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                644,
+                                505,
+                                655
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        644,
+                                        505,
+                                        655
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Linear and KNN Evaluation on ImageNet. We conduct the supervised linear classification",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                655,
+                                506,
+                                667
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        655,
+                                        246,
+                                        667
+                                    ],
+                                    "score": 1.0,
+                                    "content": "on ImageNet-1K, which contains",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        246,
+                                        655,
+                                        278,
+                                        666
+                                    ],
+                                    "score": 0.83,
+                                    "content": "{ \\sim } 1 . 3 \\mathbf { M }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        279,
+                                        655,
+                                        506,
+                                        667
+                                    ],
+                                    "score": 1.0,
+                                    "content": "images for training, and 50,000 images for validation,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                665,
+                                506,
+                                678
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        665,
+                                        506,
+                                        678
+                                    ],
+                                    "score": 1.0,
+                                    "content": "spanning 1,000 categories. Following previous works in (He et al., 2020; Chen et al., 2020a), we",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                677,
+                                506,
+                                689
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        677,
+                                        506,
+                                        689
+                                    ],
+                                    "score": 1.0,
+                                    "content": "train a single linear layer classifier on top of the frozen network encoder after self-supervised pre-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                687,
+                                506,
+                                700
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        687,
+                                        506,
+                                        700
+                                    ],
+                                    "score": 1.0,
+                                    "content": "training/distillation. SGD optimizer is used to train the linear classifier for 100 epochs with weight",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                699,
+                                505,
+                                710
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        699,
+                                        505,
+                                        710
+                                    ],
+                                    "score": 1.0,
+                                    "content": "decay to be 0. The initial learning rate is set as 30 and is then reduced by a factor of 10 at 60 and 80",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                710,
+                                506,
+                                721
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        710,
+                                        506,
+                                        721
+                                    ],
+                                    "score": 1.0,
+                                    "content": "epochs (similar as in Tian et al. (2019a)). Notably, when training the linear classifier for MobileNet",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 47
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                721,
+                                505,
+                                733
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        721,
+                                        505,
+                                        733
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and EfficientNet, we reduce the initial learning rate to 3. The results are reported in terms of Top-1",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 48
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                467,
+                                505,
+                                480
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        467,
+                                        335,
+                                        480
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and Top-5 accuracy. We also perform classification using",
+                                    "type": "text",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        335,
+                                        468,
+                                        344,
+                                        477
+                                    ],
+                                    "score": 0.77,
+                                    "content": "K",
+                                    "type": "inline_equation",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        344,
+                                        467,
+                                        505,
+                                        480
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-Nearest Neighbors (KNN) based on the",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                478,
+                                505,
+                                491
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        478,
+                                        505,
+                                        491
+                                    ],
+                                    "score": 1.0,
+                                    "content": "learned 128d vector from the last MLP layer. The sample is classified by taking the most frequent",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                489,
+                                279,
+                                502
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        489,
+                                        151,
+                                        502
+                                    ],
+                                    "score": 1.0,
+                                    "content": "label of its",
+                                    "type": "text",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        151,
+                                        489,
+                                        162,
+                                        500
+                                    ],
+                                    "score": 0.74,
+                                    "content": "K",
+                                    "type": "inline_equation",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        162,
+                                        489,
+                                        166,
+                                        502
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(",
+                                    "type": "text",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        166,
+                                        489,
+                                        200,
+                                        500
+                                    ],
+                                    "score": 0.85,
+                                    "content": "K = 1 0",
+                                    "type": "inline_equation",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        200,
+                                        489,
+                                        279,
+                                        502
+                                    ],
+                                    "score": 1.0,
+                                    "content": ") nearest neighbors.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 19
+                        }
+                    ],
+                    "index": 44.5,
+                    "bbox_fs": [
+                        105,
+                        644,
+                        506,
+                        733
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "table",
+                    "bbox": [
+                        108,
+                        151,
+                        506,
+                        285
+                    ],
+                    "blocks": [
+                        {
+                            "type": "table_caption",
+                            "bbox": [
+                                106,
+                                80,
+                                505,
+                                141
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        81,
+                                        506,
+                                        91
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                81,
+                                                240,
+                                                91
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Table 1: ImageNet-1k test accuracy",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                240,
+                                                81,
+                                                254,
+                                                91
+                                            ],
+                                            "score": 0.7,
+                                            "content": "( \\% )",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                254,
+                                                81,
+                                                506,
+                                                91
+                                            ],
+                                            "score": 1.0,
+                                            "content": "using KNN and linear classification for multiple students and MoCo-",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 0
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        90,
+                                        506,
+                                        102
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                90,
+                                                271,
+                                                102
+                                            ],
+                                            "score": 1.0,
+                                            "content": "v2 pre-trained deeper teacher architectures.",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                271,
+                                                91,
+                                                279,
+                                                100
+                                            ],
+                                            "score": 0.56,
+                                            "content": "\\pmb { \\chi }",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                279,
+                                                90,
+                                                506,
+                                                102
+                                            ],
+                                            "score": 1.0,
+                                            "content": "denotes MoCo-V2 self-supervised learning baselines before",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        100,
+                                        506,
+                                        112
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                100,
+                                                506,
+                                                112
+                                            ],
+                                            "score": 1.0,
+                                            "content": "distillation. * indicates using a deeper teacher encoder pre-trained by SWAV, where additional small-patches are",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 2
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        110,
+                                        505,
+                                        122
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                110,
+                                                319,
+                                                122
+                                            ],
+                                            "score": 1.0,
+                                            "content": "also utilized during distillation and trained for 800 epochs.",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                319,
+                                                111,
+                                                327,
+                                                119
+                                            ],
+                                            "score": 0.47,
+                                            "content": "K",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                328,
+                                                110,
+                                                505,
+                                                122
+                                            ],
+                                            "score": 1.0,
+                                            "content": "denotes Top-1 accuracy using KNN. T-1 and T-5",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 3
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        120,
+                                        507,
+                                        132
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                120,
+                                                507,
+                                                132
+                                            ],
+                                            "score": 1.0,
+                                            "content": "denote Top-1 and Top-5 accuracy using linear evaluation. First column shows Top-1 Acc. of Teacher network.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 4
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        131,
+                                        345,
+                                        141
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                131,
+                                                345,
+                                                141
+                                            ],
+                                            "score": 1.0,
+                                            "content": "First row shows the supervised performances of student networks.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 5
+                                }
+                            ],
+                            "index": 2.5
+                        },
+                        {
+                            "type": "table_body",
+                            "bbox": [
+                                108,
+                                151,
+                                506,
+                                285
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        108,
+                                        151,
+                                        506,
+                                        285
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                108,
+                                                151,
+                                                506,
+                                                285
+                                            ],
+                                            "score": 0.967,
+                                            "html": "<table><tr><td rowspan=\"2\">S T</td><td rowspan=\"2\">T-1</td><td rowspan=\"2\">K</td><td colspan=\"2\">Eff-bo</td><td colspan=\"3\">Eff-b1</td><td colspan=\"3\">Mob-v3</td><td colspan=\"3\">R-18</td><td colspan=\"3\">R-34</td></tr><tr><td>T-1</td><td>T-5</td><td>K</td><td>T-1</td><td>T-5</td><td>K</td><td>T-1</td><td>T-5</td><td>K</td><td>T-1</td><td>T-5</td><td>K</td><td>T-1</td><td>T-5</td></tr><tr><td>Supervised Acc.</td><td></td><td></td><td>77.3</td><td></td><td></td><td>79.2</td><td></td><td></td><td>75.2</td><td></td><td></td><td>72.1</td><td></td><td></td><td>75.0</td><td></td></tr><tr><td>X</td><td></td><td>30.0</td><td>42.2</td><td></td><td>68.534.4</td><td>50.7</td><td>74.6</td><td>27.5</td><td>36.3</td><td>62.2</td><td>36.7</td><td>52.5</td><td>77.0</td><td>）41.5</td><td>57.4</td><td>81.6</td></tr><tr><td>R-50 △</td><td>67.4</td><td>46.0 +16.0</td><td>61.3 +19.1</td><td>82.7 +14.2</td><td>46.1 +16.1</td><td>61.4 +10.7</td><td>83.1 +8.8</td><td>44.8 +17.3</td><td>55.2 +18.9</td><td>80.3 +18.1</td><td>43.4 +6.7</td><td>57.9 +5.1</td><td>82.0 +4.8</td><td>45.2 +3.7</td><td>58.5 +1.1</td><td>82.6 +1.0</td></tr><tr><td>R-101 △</td><td>70.3</td><td>50.1 +20.1</td><td>63.0 +20.8</td><td>83.8 +15.3</td><td>50.3 +15.9</td><td>63.4 +12.7</td><td>84.6 +10.0</td><td>48.8 +21.3</td><td>59.9 +23.6</td><td>83.5 +21.3</td><td>48.6 +11.9</td><td>58.9 +6.4</td><td>82.5 +5.5</td><td>50.5 +9.0</td><td>61.6 +4.2</td><td>84.9 +3.3</td></tr><tr><td>R-152</td><td></td><td>50.7</td><td>65.3</td><td>86.0</td><td>52.4</td><td>67.3</td><td>86.9</td><td>49.5</td><td>61.4</td><td>84.6</td><td>49.1</td><td>59.5</td><td>83.3</td><td>51.4</td><td>62.7</td><td>85.8</td></tr><tr><td>△</td><td>74.2</td><td>+20.7</td><td>+23.1</td><td>+17.5</td><td>+18.0</td><td>+16.6</td><td>+12.3</td><td>+22.0</td><td>+25.1</td><td>+22.4</td><td>+12.4</td><td>+7.0</td><td>+6.3</td><td>+9.9</td><td>+5.3</td><td>+4.2</td></tr><tr><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td></tr><tr><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td></tr><tr><td>R50×2*</td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td></tr><tr><td></td><td></td><td></td><td></td><td>87.4</td><td>60.3</td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td></tr><tr><td>△</td><td>77.3</td><td>57.4 +27.4</td><td>67.6 +25.4</td><td>+18.9</td><td>+25.9</td><td>68.0 +17.3</td><td>87.6 +13.0</td><td>55.9 +18.9</td><td>68.2 +31.9</td><td>88.2 +26.0</td><td>55.3 +18.6</td><td>63.0 +10.5</td><td>84.9 +7.9</td><td>58.2 +16.7</td><td>65.7 +8.3</td><td>86.8 +5.2</td></tr></table>",
+                                            "type": "table",
+                                            "image_path": "34e5b547413177102a008d3fb8f2ab5d87724d22eeda84157c53976330012be9.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 7,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        108,
+                                        151,
+                                        506,
+                                        195.66666666666666
+                                    ],
+                                    "spans": [],
+                                    "index": 6
+                                },
+                                {
+                                    "bbox": [
+                                        108,
+                                        195.66666666666666,
+                                        506,
+                                        240.33333333333331
+                                    ],
+                                    "spans": [],
+                                    "index": 7
+                                },
+                                {
+                                    "bbox": [
+                                        108,
+                                        240.33333333333331,
+                                        506,
+                                        285.0
+                                    ],
+                                    "spans": [],
+                                    "index": 8
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 4.75
+                },
+                {
+                    "type": "image",
+                    "bbox": [
+                        117,
+                        306,
+                        494,
+                        396
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                117,
+                                306,
+                                494,
+                                396
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        117,
+                                        306,
+                                        494,
+                                        396
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                117,
+                                                306,
+                                                494,
+                                                396
+                                            ],
+                                            "score": 0.931,
+                                            "type": "image",
+                                            "image_path": "3623a3702ec553403c1eb676464c6cd00d584bc812cc122bd666f370eb648b7c.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 10,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        117,
+                                        306,
+                                        494,
+                                        336.0
+                                    ],
+                                    "spans": [],
+                                    "index": 9
+                                },
+                                {
+                                    "bbox": [
+                                        117,
+                                        336.0,
+                                        494,
+                                        366.0
+                                    ],
+                                    "spans": [],
+                                    "index": 10
+                                },
+                                {
+                                    "bbox": [
+                                        117,
+                                        366.0,
+                                        494,
+                                        396.0
+                                    ],
+                                    "spans": [],
+                                    "index": 11
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                106,
+                                406,
+                                505,
+                                457
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        406,
+                                        505,
+                                        417
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                406,
+                                                394,
+                                                417
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 3: ImageNet-1k Top-1 accuracy for semi-supervised evaluations using",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                394,
+                                                406,
+                                                407,
+                                                416
+                                            ],
+                                            "score": 0.83,
+                                            "content": "1 \\%",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                408,
+                                                406,
+                                                446,
+                                                417
+                                            ],
+                                            "score": 1.0,
+                                            "content": "(red line),",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                447,
+                                                406,
+                                                464,
+                                                416
+                                            ],
+                                            "score": 0.86,
+                                            "content": "10 \\%",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                465,
+                                                406,
+                                                505,
+                                                417
+                                            ],
+                                            "score": 1.0,
+                                            "content": "(blue line)",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 12
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        415,
+                                        506,
+                                        428
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                415,
+                                                506,
+                                                428
+                                            ],
+                                            "score": 1.0,
+                                            "content": "of the annotations for linear fine-tuning, in comparison with the fully supervised (green line) linear evaluation",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 13
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        425,
+                                        505,
+                                        437
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                425,
+                                                505,
+                                                437
+                                            ],
+                                            "score": 1.0,
+                                            "content": "baseline for SEED. For the points whose Teacher’s number of parameters is at 0, we show the semi-supervised",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 14
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        436,
+                                        505,
+                                        447
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                436,
+                                                505,
+                                                447
+                                            ],
+                                            "score": 1.0,
+                                            "content": "linear evaluation results of MoCo-V2 without any distillation. The Student models tend to perform better on the",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 15
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        446,
+                                        325,
+                                        457
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                446,
+                                                325,
+                                                457
+                                            ],
+                                            "score": 1.0,
+                                            "content": "semi-supervised tasks after distillation from larger Teachers.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 16
+                                }
+                            ],
+                            "index": 14
+                        }
+                    ],
+                    "index": 12.0
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        467,
+                        503,
+                        501
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                467,
+                                505,
+                                480
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        467,
+                                        335,
+                                        480
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and Top-5 accuracy. We also perform classification using",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        335,
+                                        468,
+                                        344,
+                                        477
+                                    ],
+                                    "score": 0.77,
+                                    "content": "K",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        344,
+                                        467,
+                                        505,
+                                        480
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-Nearest Neighbors (KNN) based on the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                478,
+                                505,
+                                491
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        478,
+                                        505,
+                                        491
+                                    ],
+                                    "score": 1.0,
+                                    "content": "learned 128d vector from the last MLP layer. The sample is classified by taking the most frequent",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                489,
+                                279,
+                                502
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        489,
+                                        151,
+                                        502
+                                    ],
+                                    "score": 1.0,
+                                    "content": "label of its",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        151,
+                                        489,
+                                        162,
+                                        500
+                                    ],
+                                    "score": 0.74,
+                                    "content": "K",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        162,
+                                        489,
+                                        166,
+                                        502
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        166,
+                                        489,
+                                        200,
+                                        500
+                                    ],
+                                    "score": 0.85,
+                                    "content": "K = 1 0",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        200,
+                                        489,
+                                        279,
+                                        502
+                                    ],
+                                    "score": 1.0,
+                                    "content": ") nearest neighbors.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        }
+                    ],
+                    "index": 18
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        506,
+                        506,
+                        671
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                506,
+                                505,
+                                518
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        506,
+                                        505,
+                                        518
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Table 1 shows the results with various teacher networks and student networks. We list the baseline of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                517,
+                                505,
+                                529
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        517,
+                                        505,
+                                        529
+                                    ],
+                                    "score": 1.0,
+                                    "content": "contrastive self-supervised pre-training using MoCo-V2 (Chen et al., 2020c) in the first row for each",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                528,
+                                506,
+                                541
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        528,
+                                        506,
+                                        541
+                                    ],
+                                    "score": 1.0,
+                                    "content": "student architecture. We can see clearly that smaller networks perform rather worse. For example,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                538,
+                                505,
+                                552
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        538,
+                                        228,
+                                        552
+                                    ],
+                                    "score": 1.0,
+                                    "content": "MobileNet-V3 can only reach",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        228,
+                                        539,
+                                        255,
+                                        550
+                                    ],
+                                    "score": 0.87,
+                                    "content": "3 6 . 3 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        256,
+                                        538,
+                                        505,
+                                        552
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". This aligns well with previous conclusions from (Chen et al.,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                549,
+                                506,
+                                563
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        549,
+                                        506,
+                                        563
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2020a;b) that bigger models are desired to perform better in contrastive-based self-supervised pre-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                561,
+                                506,
+                                573
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        561,
+                                        506,
+                                        573
+                                    ],
+                                    "score": 1.0,
+                                    "content": "training. We conjecture that this is mainly caused by the inability of smaller network to discriminate",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                572,
+                                506,
+                                585
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        572,
+                                        506,
+                                        585
+                                    ],
+                                    "score": 1.0,
+                                    "content": "instances in a large-scale dataset. The results also clearly demonstrate that the distillation from a larger",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                583,
+                                505,
+                                595
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        583,
+                                        505,
+                                        595
+                                    ],
+                                    "score": 1.0,
+                                    "content": "network helps boosting the performances of small networks, and show obvious improvement. For",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                594,
+                                506,
+                                606
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        594,
+                                        506,
+                                        606
+                                    ],
+                                    "score": 1.0,
+                                    "content": "instance, with MoCo-V2 pre-trained ResNet-152 (for 400 epochs) as the teacher network, the Top-1",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                605,
+                                507,
+                                617
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        605,
+                                        382,
+                                        617
+                                    ],
+                                    "score": 1.0,
+                                    "content": "accuracy of MobileNet-V3-Large can be significantly improved from",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        383,
+                                        605,
+                                        410,
+                                        615
+                                    ],
+                                    "score": 0.87,
+                                    "content": "3 6 . 3 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        410,
+                                        605,
+                                        420,
+                                        617
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        421,
+                                        605,
+                                        447,
+                                        615
+                                    ],
+                                    "score": 0.88,
+                                    "content": "6 1 . 4 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        448,
+                                        605,
+                                        507,
+                                        617
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Furthermore,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                615,
+                                506,
+                                629
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        615,
+                                        167,
+                                        629
+                                    ],
+                                    "score": 1.0,
+                                    "content": "we use ResNet-",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        167,
+                                        616,
+                                        191,
+                                        627
+                                    ],
+                                    "score": 0.88,
+                                    "content": "5 0 \\times 2",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        191,
+                                        615,
+                                        506,
+                                        629
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(provided in Caron et al. (2020)) as the teacher network and adopt the multi-crop",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                627,
+                                506,
+                                639
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        627,
+                                        378,
+                                        639
+                                    ],
+                                    "score": 1.0,
+                                    "content": "trick (see A.2 for details). The accuracy can be further improved to",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        379,
+                                        627,
+                                        406,
+                                        637
+                                    ],
+                                    "score": 0.88,
+                                    "content": "6 8 . 2 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        406,
+                                        627,
+                                        506,
+                                        639
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(last row of Table 1) for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                638,
+                                505,
+                                649
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        638,
+                                        505,
+                                        649
+                                    ],
+                                    "score": 1.0,
+                                    "content": "MobileNet-V3-Large with 800 epochs of distillation. We note that the gain benefited from distillation",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                649,
+                                506,
+                                661
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        649,
+                                        506,
+                                        661
+                                    ],
+                                    "score": 1.0,
+                                    "content": "becomes more distinct on smaller architectures and we further study the effect of various teacher",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                660,
+                                189,
+                                671
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        660,
+                                        189,
+                                        671
+                                    ],
+                                    "score": 1.0,
+                                    "content": "models in ablations.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        }
+                    ],
+                    "index": 27
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        677,
+                        505,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                676,
+                                506,
+                                689
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        676,
+                                        506,
+                                        689
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Semi-Supervised Evaluation on ImageNet. Following (Oord et al., 2018; Kornblith et al., 2019;",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                686,
+                                505,
+                                700
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        686,
+                                        505,
+                                        700
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Kolesnikov et al., 2019), we evaluate the representation on the semi-supervised task, where a fixed",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                698,
+                                506,
+                                711
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        699,
+                                        120,
+                                        709
+                                    ],
+                                    "score": 0.84,
+                                    "content": "1 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        121,
+                                        698,
+                                        132,
+                                        711
+                                    ],
+                                    "score": 1.0,
+                                    "content": "or",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        132,
+                                        699,
+                                        152,
+                                        709
+                                    ],
+                                    "score": 0.86,
+                                    "content": "10 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        152,
+                                        698,
+                                        506,
+                                        711
+                                    ],
+                                    "score": 1.0,
+                                    "content": "subsets of ImageNet training data (Chen et al., 2020a) are provided with the annotations.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                709,
+                                505,
+                                721
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        709,
+                                        505,
+                                        721
+                                    ],
+                                    "score": 1.0,
+                                    "content": "After the self-supervised learning with and without distillation, we also train a classifier on top of the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                720,
+                                505,
+                                733
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        720,
+                                        505,
+                                        733
+                                    ],
+                                    "score": 1.0,
+                                    "content": "representation. The results are shown in Figure 3, where the baseline without distillation is depicted",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        }
+                    ],
+                    "index": 37
+                }
+            ],
+            "page_idx": 5,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        108,
+                        27,
+                        292,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                26,
+                                293,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        26,
+                                        293,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Published as a conference paper at ICLR 2021",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        302,
+                        752,
+                        308,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                302,
+                                751,
+                                309,
+                                762
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        302,
+                                        751,
+                                        309,
+                                        762
+                                    ],
+                                    "score": 1.0,
+                                    "content": "6",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "table",
+                    "bbox": [
+                        108,
+                        151,
+                        506,
+                        285
+                    ],
+                    "blocks": [
+                        {
+                            "type": "table_caption",
+                            "bbox": [
+                                106,
+                                80,
+                                505,
+                                141
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        81,
+                                        506,
+                                        91
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                81,
+                                                240,
+                                                91
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Table 1: ImageNet-1k test accuracy",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                240,
+                                                81,
+                                                254,
+                                                91
+                                            ],
+                                            "score": 0.7,
+                                            "content": "( \\% )",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                254,
+                                                81,
+                                                506,
+                                                91
+                                            ],
+                                            "score": 1.0,
+                                            "content": "using KNN and linear classification for multiple students and MoCo-",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 0
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        90,
+                                        506,
+                                        102
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                90,
+                                                271,
+                                                102
+                                            ],
+                                            "score": 1.0,
+                                            "content": "v2 pre-trained deeper teacher architectures.",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                271,
+                                                91,
+                                                279,
+                                                100
+                                            ],
+                                            "score": 0.56,
+                                            "content": "\\pmb { \\chi }",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                279,
+                                                90,
+                                                506,
+                                                102
+                                            ],
+                                            "score": 1.0,
+                                            "content": "denotes MoCo-V2 self-supervised learning baselines before",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        100,
+                                        506,
+                                        112
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                100,
+                                                506,
+                                                112
+                                            ],
+                                            "score": 1.0,
+                                            "content": "distillation. * indicates using a deeper teacher encoder pre-trained by SWAV, where additional small-patches are",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 2
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        110,
+                                        505,
+                                        122
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                110,
+                                                319,
+                                                122
+                                            ],
+                                            "score": 1.0,
+                                            "content": "also utilized during distillation and trained for 800 epochs.",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                319,
+                                                111,
+                                                327,
+                                                119
+                                            ],
+                                            "score": 0.47,
+                                            "content": "K",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                328,
+                                                110,
+                                                505,
+                                                122
+                                            ],
+                                            "score": 1.0,
+                                            "content": "denotes Top-1 accuracy using KNN. T-1 and T-5",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 3
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        120,
+                                        507,
+                                        132
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                120,
+                                                507,
+                                                132
+                                            ],
+                                            "score": 1.0,
+                                            "content": "denote Top-1 and Top-5 accuracy using linear evaluation. First column shows Top-1 Acc. of Teacher network.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 4
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        131,
+                                        345,
+                                        141
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                131,
+                                                345,
+                                                141
+                                            ],
+                                            "score": 1.0,
+                                            "content": "First row shows the supervised performances of student networks.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 5
+                                }
+                            ],
+                            "index": 2.5
+                        },
+                        {
+                            "type": "table_body",
+                            "bbox": [
+                                108,
+                                151,
+                                506,
+                                285
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        108,
+                                        151,
+                                        506,
+                                        285
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                108,
+                                                151,
+                                                506,
+                                                285
+                                            ],
+                                            "score": 0.967,
+                                            "html": "<table><tr><td rowspan=\"2\">S T</td><td rowspan=\"2\">T-1</td><td rowspan=\"2\">K</td><td colspan=\"2\">Eff-bo</td><td colspan=\"3\">Eff-b1</td><td colspan=\"3\">Mob-v3</td><td colspan=\"3\">R-18</td><td colspan=\"3\">R-34</td></tr><tr><td>T-1</td><td>T-5</td><td>K</td><td>T-1</td><td>T-5</td><td>K</td><td>T-1</td><td>T-5</td><td>K</td><td>T-1</td><td>T-5</td><td>K</td><td>T-1</td><td>T-5</td></tr><tr><td>Supervised Acc.</td><td></td><td></td><td>77.3</td><td></td><td></td><td>79.2</td><td></td><td></td><td>75.2</td><td></td><td></td><td>72.1</td><td></td><td></td><td>75.0</td><td></td></tr><tr><td>X</td><td></td><td>30.0</td><td>42.2</td><td></td><td>68.534.4</td><td>50.7</td><td>74.6</td><td>27.5</td><td>36.3</td><td>62.2</td><td>36.7</td><td>52.5</td><td>77.0</td><td>）41.5</td><td>57.4</td><td>81.6</td></tr><tr><td>R-50 △</td><td>67.4</td><td>46.0 +16.0</td><td>61.3 +19.1</td><td>82.7 +14.2</td><td>46.1 +16.1</td><td>61.4 +10.7</td><td>83.1 +8.8</td><td>44.8 +17.3</td><td>55.2 +18.9</td><td>80.3 +18.1</td><td>43.4 +6.7</td><td>57.9 +5.1</td><td>82.0 +4.8</td><td>45.2 +3.7</td><td>58.5 +1.1</td><td>82.6 +1.0</td></tr><tr><td>R-101 △</td><td>70.3</td><td>50.1 +20.1</td><td>63.0 +20.8</td><td>83.8 +15.3</td><td>50.3 +15.9</td><td>63.4 +12.7</td><td>84.6 +10.0</td><td>48.8 +21.3</td><td>59.9 +23.6</td><td>83.5 +21.3</td><td>48.6 +11.9</td><td>58.9 +6.4</td><td>82.5 +5.5</td><td>50.5 +9.0</td><td>61.6 +4.2</td><td>84.9 +3.3</td></tr><tr><td>R-152</td><td></td><td>50.7</td><td>65.3</td><td>86.0</td><td>52.4</td><td>67.3</td><td>86.9</td><td>49.5</td><td>61.4</td><td>84.6</td><td>49.1</td><td>59.5</td><td>83.3</td><td>51.4</td><td>62.7</td><td>85.8</td></tr><tr><td>△</td><td>74.2</td><td>+20.7</td><td>+23.1</td><td>+17.5</td><td>+18.0</td><td>+16.6</td><td>+12.3</td><td>+22.0</td><td>+25.1</td><td>+22.4</td><td>+12.4</td><td>+7.0</td><td>+6.3</td><td>+9.9</td><td>+5.3</td><td>+4.2</td></tr><tr><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td></tr><tr><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td></tr><tr><td>R50×2*</td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td></tr><tr><td></td><td></td><td></td><td></td><td>87.4</td><td>60.3</td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td></tr><tr><td>△</td><td>77.3</td><td>57.4 +27.4</td><td>67.6 +25.4</td><td>+18.9</td><td>+25.9</td><td>68.0 +17.3</td><td>87.6 +13.0</td><td>55.9 +18.9</td><td>68.2 +31.9</td><td>88.2 +26.0</td><td>55.3 +18.6</td><td>63.0 +10.5</td><td>84.9 +7.9</td><td>58.2 +16.7</td><td>65.7 +8.3</td><td>86.8 +5.2</td></tr></table>",
+                                            "type": "table",
+                                            "image_path": "34e5b547413177102a008d3fb8f2ab5d87724d22eeda84157c53976330012be9.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 7,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        108,
+                                        151,
+                                        506,
+                                        195.66666666666666
+                                    ],
+                                    "spans": [],
+                                    "index": 6
+                                },
+                                {
+                                    "bbox": [
+                                        108,
+                                        195.66666666666666,
+                                        506,
+                                        240.33333333333331
+                                    ],
+                                    "spans": [],
+                                    "index": 7
+                                },
+                                {
+                                    "bbox": [
+                                        108,
+                                        240.33333333333331,
+                                        506,
+                                        285.0
+                                    ],
+                                    "spans": [],
+                                    "index": 8
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 4.75
+                },
+                {
+                    "type": "image",
+                    "bbox": [
+                        117,
+                        306,
+                        494,
+                        396
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                117,
+                                306,
+                                494,
+                                396
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        117,
+                                        306,
+                                        494,
+                                        396
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                117,
+                                                306,
+                                                494,
+                                                396
+                                            ],
+                                            "score": 0.931,
+                                            "type": "image",
+                                            "image_path": "3623a3702ec553403c1eb676464c6cd00d584bc812cc122bd666f370eb648b7c.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 10,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        117,
+                                        306,
+                                        494,
+                                        336.0
+                                    ],
+                                    "spans": [],
+                                    "index": 9
+                                },
+                                {
+                                    "bbox": [
+                                        117,
+                                        336.0,
+                                        494,
+                                        366.0
+                                    ],
+                                    "spans": [],
+                                    "index": 10
+                                },
+                                {
+                                    "bbox": [
+                                        117,
+                                        366.0,
+                                        494,
+                                        396.0
+                                    ],
+                                    "spans": [],
+                                    "index": 11
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                106,
+                                406,
+                                505,
+                                457
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        406,
+                                        505,
+                                        417
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                406,
+                                                394,
+                                                417
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 3: ImageNet-1k Top-1 accuracy for semi-supervised evaluations using",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                394,
+                                                406,
+                                                407,
+                                                416
+                                            ],
+                                            "score": 0.83,
+                                            "content": "1 \\%",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                408,
+                                                406,
+                                                446,
+                                                417
+                                            ],
+                                            "score": 1.0,
+                                            "content": "(red line),",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                447,
+                                                406,
+                                                464,
+                                                416
+                                            ],
+                                            "score": 0.86,
+                                            "content": "10 \\%",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                465,
+                                                406,
+                                                505,
+                                                417
+                                            ],
+                                            "score": 1.0,
+                                            "content": "(blue line)",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 12
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        415,
+                                        506,
+                                        428
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                415,
+                                                506,
+                                                428
+                                            ],
+                                            "score": 1.0,
+                                            "content": "of the annotations for linear fine-tuning, in comparison with the fully supervised (green line) linear evaluation",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 13
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        425,
+                                        505,
+                                        437
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                425,
+                                                505,
+                                                437
+                                            ],
+                                            "score": 1.0,
+                                            "content": "baseline for SEED. For the points whose Teacher’s number of parameters is at 0, we show the semi-supervised",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 14
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        436,
+                                        505,
+                                        447
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                436,
+                                                505,
+                                                447
+                                            ],
+                                            "score": 1.0,
+                                            "content": "linear evaluation results of MoCo-V2 without any distillation. The Student models tend to perform better on the",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 15
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        446,
+                                        325,
+                                        457
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                446,
+                                                325,
+                                                457
+                                            ],
+                                            "score": 1.0,
+                                            "content": "semi-supervised tasks after distillation from larger Teachers.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 16
+                                }
+                            ],
+                            "index": 14
+                        }
+                    ],
+                    "index": 12.0
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        467,
+                        503,
+                        501
+                    ],
+                    "lines": [],
+                    "index": 18,
+                    "bbox_fs": [
+                        105,
+                        467,
+                        505,
+                        502
+                    ],
+                    "lines_deleted": true
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        506,
+                        506,
+                        671
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                506,
+                                505,
+                                518
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        506,
+                                        505,
+                                        518
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Table 1 shows the results with various teacher networks and student networks. We list the baseline of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                517,
+                                505,
+                                529
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        517,
+                                        505,
+                                        529
+                                    ],
+                                    "score": 1.0,
+                                    "content": "contrastive self-supervised pre-training using MoCo-V2 (Chen et al., 2020c) in the first row for each",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                528,
+                                506,
+                                541
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        528,
+                                        506,
+                                        541
+                                    ],
+                                    "score": 1.0,
+                                    "content": "student architecture. We can see clearly that smaller networks perform rather worse. For example,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                538,
+                                505,
+                                552
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        538,
+                                        228,
+                                        552
+                                    ],
+                                    "score": 1.0,
+                                    "content": "MobileNet-V3 can only reach",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        228,
+                                        539,
+                                        255,
+                                        550
+                                    ],
+                                    "score": 0.87,
+                                    "content": "3 6 . 3 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        256,
+                                        538,
+                                        505,
+                                        552
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". This aligns well with previous conclusions from (Chen et al.,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                549,
+                                506,
+                                563
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        549,
+                                        506,
+                                        563
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2020a;b) that bigger models are desired to perform better in contrastive-based self-supervised pre-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                561,
+                                506,
+                                573
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        561,
+                                        506,
+                                        573
+                                    ],
+                                    "score": 1.0,
+                                    "content": "training. We conjecture that this is mainly caused by the inability of smaller network to discriminate",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                572,
+                                506,
+                                585
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        572,
+                                        506,
+                                        585
+                                    ],
+                                    "score": 1.0,
+                                    "content": "instances in a large-scale dataset. The results also clearly demonstrate that the distillation from a larger",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                583,
+                                505,
+                                595
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        583,
+                                        505,
+                                        595
+                                    ],
+                                    "score": 1.0,
+                                    "content": "network helps boosting the performances of small networks, and show obvious improvement. For",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                594,
+                                506,
+                                606
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        594,
+                                        506,
+                                        606
+                                    ],
+                                    "score": 1.0,
+                                    "content": "instance, with MoCo-V2 pre-trained ResNet-152 (for 400 epochs) as the teacher network, the Top-1",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                605,
+                                507,
+                                617
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        605,
+                                        382,
+                                        617
+                                    ],
+                                    "score": 1.0,
+                                    "content": "accuracy of MobileNet-V3-Large can be significantly improved from",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        383,
+                                        605,
+                                        410,
+                                        615
+                                    ],
+                                    "score": 0.87,
+                                    "content": "3 6 . 3 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        410,
+                                        605,
+                                        420,
+                                        617
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        421,
+                                        605,
+                                        447,
+                                        615
+                                    ],
+                                    "score": 0.88,
+                                    "content": "6 1 . 4 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        448,
+                                        605,
+                                        507,
+                                        617
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Furthermore,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                615,
+                                506,
+                                629
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        615,
+                                        167,
+                                        629
+                                    ],
+                                    "score": 1.0,
+                                    "content": "we use ResNet-",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        167,
+                                        616,
+                                        191,
+                                        627
+                                    ],
+                                    "score": 0.88,
+                                    "content": "5 0 \\times 2",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        191,
+                                        615,
+                                        506,
+                                        629
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(provided in Caron et al. (2020)) as the teacher network and adopt the multi-crop",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                627,
+                                506,
+                                639
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        627,
+                                        378,
+                                        639
+                                    ],
+                                    "score": 1.0,
+                                    "content": "trick (see A.2 for details). The accuracy can be further improved to",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        379,
+                                        627,
+                                        406,
+                                        637
+                                    ],
+                                    "score": 0.88,
+                                    "content": "6 8 . 2 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        406,
+                                        627,
+                                        506,
+                                        639
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(last row of Table 1) for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                638,
+                                505,
+                                649
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        638,
+                                        505,
+                                        649
+                                    ],
+                                    "score": 1.0,
+                                    "content": "MobileNet-V3-Large with 800 epochs of distillation. We note that the gain benefited from distillation",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                649,
+                                506,
+                                661
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        649,
+                                        506,
+                                        661
+                                    ],
+                                    "score": 1.0,
+                                    "content": "becomes more distinct on smaller architectures and we further study the effect of various teacher",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                660,
+                                189,
+                                671
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        660,
+                                        189,
+                                        671
+                                    ],
+                                    "score": 1.0,
+                                    "content": "models in ablations.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        }
+                    ],
+                    "index": 27,
+                    "bbox_fs": [
+                        105,
+                        506,
+                        507,
+                        671
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        677,
+                        505,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                676,
+                                506,
+                                689
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        676,
+                                        506,
+                                        689
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Semi-Supervised Evaluation on ImageNet. Following (Oord et al., 2018; Kornblith et al., 2019;",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                686,
+                                505,
+                                700
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        686,
+                                        505,
+                                        700
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Kolesnikov et al., 2019), we evaluate the representation on the semi-supervised task, where a fixed",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                698,
+                                506,
+                                711
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        699,
+                                        120,
+                                        709
+                                    ],
+                                    "score": 0.84,
+                                    "content": "1 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        121,
+                                        698,
+                                        132,
+                                        711
+                                    ],
+                                    "score": 1.0,
+                                    "content": "or",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        132,
+                                        699,
+                                        152,
+                                        709
+                                    ],
+                                    "score": 0.86,
+                                    "content": "10 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        152,
+                                        698,
+                                        506,
+                                        711
+                                    ],
+                                    "score": 1.0,
+                                    "content": "subsets of ImageNet training data (Chen et al., 2020a) are provided with the annotations.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                709,
+                                505,
+                                721
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        709,
+                                        505,
+                                        721
+                                    ],
+                                    "score": 1.0,
+                                    "content": "After the self-supervised learning with and without distillation, we also train a classifier on top of the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                720,
+                                505,
+                                733
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        720,
+                                        505,
+                                        733
+                                    ],
+                                    "score": 1.0,
+                                    "content": "representation. The results are shown in Figure 3, where the baseline without distillation is depicted",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                372,
+                                505,
+                                385
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        372,
+                                        505,
+                                        385
+                                    ],
+                                    "score": 1.0,
+                                    "content": "when teacher parameters are 0. As we can see, the accuracy is also improved remarkably with SEED",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                384,
+                                505,
+                                396
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        384,
+                                        505,
+                                        396
+                                    ],
+                                    "score": 1.0,
+                                    "content": "distillation, and a stronger teacher network with more parameters leads to a better performed student",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                394,
+                                145,
+                                406
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        394,
+                                        145,
+                                        406
+                                    ],
+                                    "score": 1.0,
+                                    "content": "network.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 15
+                        }
+                    ],
+                    "index": 37,
+                    "bbox_fs": [
+                        105,
+                        676,
+                        506,
+                        733
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "table",
+                    "bbox": [
+                        106,
+                        120,
+                        509,
+                        202
+                    ],
+                    "blocks": [
+                        {
+                            "type": "table_caption",
+                            "bbox": [
+                                106,
+                                69,
+                                506,
+                                110
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        67,
+                                        506,
+                                        81
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                67,
+                                                506,
+                                                81
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Table 2: Object detection and instance segmentation results using contrastive self-supervised learning and",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 0
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        77,
+                                        506,
+                                        92
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                77,
+                                                358,
+                                                92
+                                            ],
+                                            "score": 1.0,
+                                            "content": "SEED distillation using ResNet-18 as backbone: bounding-box AP",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                358,
+                                                79,
+                                                383,
+                                                90
+                                            ],
+                                            "score": 0.85,
+                                            "content": "( \\mathbf { A P } ^ { b b } )",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                383,
+                                                77,
+                                                437,
+                                                92
+                                            ],
+                                            "score": 1.0,
+                                            "content": "and mask AP",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                437,
+                                                79,
+                                                466,
+                                                90
+                                            ],
+                                            "score": 0.85,
+                                            "content": "( \\mathbf { A P } ^ { m k } )",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                466,
+                                                77,
+                                                506,
+                                                92
+                                            ],
+                                            "score": 1.0,
+                                            "content": "evaluated",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        88,
+                                        506,
+                                        102
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                88,
+                                                506,
+                                                102
+                                            ],
+                                            "score": 1.0,
+                                            "content": "on VOC07-val and COCO testing split. More results on different backbones can be found in the Appendix.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 2
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        99,
+                                        327,
+                                        111
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                99,
+                                                327,
+                                                111
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Subscript in green represents improvement is larger than 0.3.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 3
+                                }
+                            ],
+                            "index": 1.5
+                        },
+                        {
+                            "type": "table_body",
+                            "bbox": [
+                                106,
+                                120,
+                                509,
+                                202
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        120,
+                                        509,
+                                        202
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                120,
+                                                509,
+                                                202
+                                            ],
+                                            "score": 0.963,
+                                            "html": "<table><tr><td rowspan=\"2\">S</td><td rowspan=\"2\">T</td><td colspan=\"3\">VOC Obj. Det.</td><td colspan=\"3\">COCO Obj. Det.</td><td colspan=\"3\">COCO Inst. Segm.</td></tr><tr><td>Apbb</td><td>AP</td><td>AP</td><td>Apbb</td><td>AP</td><td>APP</td><td>Apmk</td><td>AP</td><td>AP</td></tr><tr><td></td><td>X</td><td>46.1</td><td>74.5</td><td>48.6</td><td>35.0</td><td>53.9</td><td>37.7</td><td>31.0</td><td>51.1</td><td>33.1</td></tr><tr><td></td><td>R-50</td><td>46.1(0.0)</td><td>74.8(+0.3)</td><td>49.1(+0.5)</td><td>35.3(+0.3)</td><td>54.2(+0.3)</td><td>37.8(+0.1)</td><td>31.1(+0.1)</td><td>51.1(0.0)</td><td>33.2(+0.1)</td></tr><tr><td>R-18 R-101</td><td></td><td>46.8(+0.7)</td><td>75.8(+1.3)</td><td>49.3(+0.7)</td><td>35.3(+0.3)</td><td>54.3(+0.4)</td><td>37.9(+0.2)</td><td>31.3(+0.3)</td><td>51.3(+0.2)</td><td>33.4(+0.3)</td></tr><tr><td></td><td>R-152</td><td>46.8(+0.7)</td><td>75.9(+1.4)</td><td>50.2(+1.6)</td><td>35.4(+0.4)</td><td>54.4(+0.5)</td><td>38.0(+0.3)</td><td>31.3(+0.3)</td><td>51.4(+0.3)</td><td>33.4(+0.3)</td></tr></table>",
+                                            "type": "table",
+                                            "image_path": "bc2a89152889a8d15d369e7b291816928ecaa6276dd595b93bc9cd3b191f82cd.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 5,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        120,
+                                        509,
+                                        147.33333333333334
+                                    ],
+                                    "spans": [],
+                                    "index": 4
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        147.33333333333334,
+                                        509,
+                                        174.66666666666669
+                                    ],
+                                    "spans": [],
+                                    "index": 5
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        174.66666666666669,
+                                        509,
+                                        202.00000000000003
+                                    ],
+                                    "spans": [],
+                                    "index": 6
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 3.25
+                },
+                {
+                    "type": "image",
+                    "bbox": [
+                        107,
+                        225,
+                        503,
+                        321
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                107,
+                                225,
+                                503,
+                                321
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        225,
+                                        503,
+                                        321
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                107,
+                                                225,
+                                                503,
+                                                321
+                                            ],
+                                            "score": 0.96,
+                                            "type": "image",
+                                            "image_path": "15d565abe6e76f1be7191527af52d1cfceecfa87f6083e78a4a48b0e6be4ff99.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 8,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        225,
+                                        503,
+                                        257.0
+                                    ],
+                                    "spans": [],
+                                    "index": 7
+                                },
+                                {
+                                    "bbox": [
+                                        107,
+                                        257.0,
+                                        503,
+                                        289.0
+                                    ],
+                                    "spans": [],
+                                    "index": 8
+                                },
+                                {
+                                    "bbox": [
+                                        107,
+                                        289.0,
+                                        503,
+                                        321.0
+                                    ],
+                                    "spans": [],
+                                    "index": 9
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                107,
+                                325,
+                                505,
+                                356
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        325,
+                                        505,
+                                        337
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                325,
+                                                230,
+                                                337
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 4: ImageNet-1k Accuracy",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                230,
+                                                326,
+                                                243,
+                                                336
+                                            ],
+                                            "score": 0.73,
+                                            "content": "( \\% )",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                244,
+                                                325,
+                                                505,
+                                                337
+                                            ],
+                                            "score": 1.0,
+                                            "content": "of student network (EfficientNet-B0 and ResNet-18) transferred to other",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 10
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        334,
+                                        506,
+                                        347
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                334,
+                                                506,
+                                                347
+                                            ],
+                                            "score": 1.0,
+                                            "content": "domains (CIFAR-10, CIFAR-100, SUN-397 datasets) with and without distillation from lager architectures",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 11
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        345,
+                                        187,
+                                        356
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                345,
+                                                187,
+                                                356
+                                            ],
+                                            "score": 1.0,
+                                            "content": "(ResNet-50/101/152).",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 12
+                                }
+                            ],
+                            "index": 11
+                        }
+                    ],
+                    "index": 9.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        373,
+                        505,
+                        405
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                372,
+                                505,
+                                385
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        372,
+                                        505,
+                                        385
+                                    ],
+                                    "score": 1.0,
+                                    "content": "when teacher parameters are 0. As we can see, the accuracy is also improved remarkably with SEED",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                384,
+                                505,
+                                396
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        384,
+                                        505,
+                                        396
+                                    ],
+                                    "score": 1.0,
+                                    "content": "distillation, and a stronger teacher network with more parameters leads to a better performed student",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                394,
+                                145,
+                                406
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        394,
+                                        145,
+                                        406
+                                    ],
+                                    "score": 1.0,
+                                    "content": "network.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        }
+                    ],
+                    "index": 14
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        411,
+                        505,
+                        532
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                411,
+                                506,
+                                423
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        411,
+                                        506,
+                                        423
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Transferring to Classification. To further study whether the improvement of the learned represen-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                423,
+                                505,
+                                434
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        423,
+                                        505,
+                                        434
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tations by distillation is confined to ImageNet, we evaluate on additional classification datasets to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                433,
+                                506,
+                                445
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        433,
+                                        506,
+                                        445
+                                    ],
+                                    "score": 1.0,
+                                    "content": "study the generalization and transferability of the feature representation. We strictly follow the linear",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                444,
+                                505,
+                                456
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        444,
+                                        505,
+                                        456
+                                    ],
+                                    "score": 1.0,
+                                    "content": "evaluation and fine-tuning settings from (Kornblith et al., 2019; Chen et al., 2020a; Grill et al., 2020),",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                455,
+                                505,
+                                466
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        455,
+                                        505,
+                                        466
+                                    ],
+                                    "score": 1.0,
+                                    "content": "that a linear layer is trained on the basis of frozen features. We report Top-1 Accuracy of models",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                465,
+                                506,
+                                478
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        465,
+                                        506,
+                                        478
+                                    ],
+                                    "score": 1.0,
+                                    "content": "before and after distillation from various architectures on CIFAR-10, CIFAR-100 (Krizhevsky et al.,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                476,
+                                505,
+                                490
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        476,
+                                        505,
+                                        490
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2009), SUN-397 (Xiao et al., 2010) datasets (see Figure 4). More details regarding pre-processing",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                488,
+                                505,
+                                500
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        488,
+                                        505,
+                                        500
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and training can be found in A.1.2. Notably, we observe that our distillation surpasses contrastive",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                498,
+                                506,
+                                511
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        498,
+                                        506,
+                                        511
+                                    ],
+                                    "score": 1.0,
+                                    "content": "self-supervised pre-training consistently on all benchmarks, verifying the effectiveness of SEED.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                510,
+                                505,
+                                522
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        510,
+                                        505,
+                                        522
+                                    ],
+                                    "score": 1.0,
+                                    "content": "This also proves the generalization ability of the learned representations from distillation to a wide",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                521,
+                                279,
+                                533
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        521,
+                                        279,
+                                        533
+                                    ],
+                                    "score": 1.0,
+                                    "content": "range of data domain and different classes.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        }
+                    ],
+                    "index": 21
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        538,
+                        505,
+                        669
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                537,
+                                506,
+                                550
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        537,
+                                        506,
+                                        550
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Transferring to Detection and Segmentation. We conduct two downstream tasks here. The first",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                549,
+                                506,
+                                561
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        549,
+                                        424,
+                                        561
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is Faster R-CNN (Ren et al., 2015) model for object detection trained on VOC-",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        425,
+                                        549,
+                                        452,
+                                        559
+                                    ],
+                                    "score": 0.79,
+                                    "content": "^ { 0 7 + 1 2 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        452,
+                                        549,
+                                        471,
+                                        561
+                                    ],
+                                    "score": 1.0,
+                                    "content": "train",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        471,
+                                        551,
+                                        478,
+                                        558
+                                    ],
+                                    "score": 0.32,
+                                    "content": "^ +",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        478,
+                                        549,
+                                        506,
+                                        561
+                                    ],
+                                    "score": 1.0,
+                                    "content": "val set",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                559,
+                                505,
+                                572
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        559,
+                                        505,
+                                        572
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and evaluated on VOC-07 test split. The second is Mask R-CNN (He et al., 2017) model for the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                570,
+                                505,
+                                582
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        570,
+                                        505,
+                                        582
+                                    ],
+                                    "score": 1.0,
+                                    "content": "object detection and instance segmentation on COCO 2017 dataset (Lin et al., 2014). The pre-trained",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                582,
+                                505,
+                                594
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        582,
+                                        505,
+                                        594
+                                    ],
+                                    "score": 1.0,
+                                    "content": "model serves as the initial weight and following He et al. (2020), we fine-tune all the layers of the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                592,
+                                506,
+                                605
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        592,
+                                        506,
+                                        605
+                                    ],
+                                    "score": 1.0,
+                                    "content": "model. More experiment settings can be found in A.2. The results are illustrated in Table 2. As we",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                604,
+                                506,
+                                616
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        604,
+                                        506,
+                                        616
+                                    ],
+                                    "score": 1.0,
+                                    "content": "can see, on VOC, the distilled pre-trained model achieves a large improvement. With ResNet-152 as",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                614,
+                                506,
+                                626
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        614,
+                                        380,
+                                        626
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the teacher network, the Resnet18-based Faster R-CNN model shows",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        380,
+                                        615,
+                                        399,
+                                        625
+                                    ],
+                                    "score": 0.83,
+                                    "content": "+ 0 . 7",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        400,
+                                        614,
+                                        506,
+                                        626
+                                    ],
+                                    "score": 1.0,
+                                    "content": "point improvement on AP,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                624,
+                                506,
+                                638
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        626,
+                                        125,
+                                        636
+                                    ],
+                                    "score": 0.82,
+                                    "content": "+ 1 . 4",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        125,
+                                        624,
+                                        194,
+                                        638
+                                    ],
+                                    "score": 1.0,
+                                    "content": "improvement on",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        194,
+                                        626,
+                                        216,
+                                        637
+                                    ],
+                                    "score": 0.9,
+                                    "content": "\\mathrm { { A P } _ { 5 0 } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        217,
+                                        624,
+                                        234,
+                                        638
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        234,
+                                        626,
+                                        254,
+                                        636
+                                    ],
+                                    "score": 0.85,
+                                    "content": "+ 1 . 6",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        255,
+                                        624,
+                                        267,
+                                        638
+                                    ],
+                                    "score": 1.0,
+                                    "content": "on",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        268,
+                                        626,
+                                        290,
+                                        637
+                                    ],
+                                    "score": 0.9,
+                                    "content": "\\mathsf { A P } _ { 7 5 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        290,
+                                        624,
+                                        506,
+                                        638
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". On COCO, the improvement is relatively minor and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                636,
+                                505,
+                                648
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        636,
+                                        295,
+                                        648
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the reason could be that COCO training set has",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        295,
+                                        637,
+                                        325,
+                                        647
+                                    ],
+                                    "score": 0.89,
+                                    "content": "{ \\sim } 1 1 8 \\mathrm { k }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        325,
+                                        636,
+                                        472,
+                                        648
+                                    ],
+                                    "score": 1.0,
+                                    "content": "training images while VOC has only",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        473,
+                                        637,
+                                        505,
+                                        647
+                                    ],
+                                    "score": 0.88,
+                                    "content": "{ \\sim } 1 6 . 5 \\mathrm { k }",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                648,
+                                505,
+                                659
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        648,
+                                        505,
+                                        659
+                                    ],
+                                    "score": 1.0,
+                                    "content": "training images. A larger training set with more fine-tuning iterations reduces the importance of the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                658,
+                                169,
+                                671
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        658,
+                                        169,
+                                        671
+                                    ],
+                                    "score": 1.0,
+                                    "content": "initial weights.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        }
+                    ],
+                    "index": 32.5
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        687,
+                        208,
+                        698
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                686,
+                                210,
+                                699
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        686,
+                                        210,
+                                        699
+                                    ],
+                                    "score": 1.0,
+                                    "content": "4.3 ABLATION STUDY",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        }
+                    ],
+                    "index": 39
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        709,
+                        503,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                707,
+                                505,
+                                723
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        707,
+                                        505,
+                                        723
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We now explore the effects of distillation using different Teacher architectures, Teacher Pre-training",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                720,
+                                362,
+                                733
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        720,
+                                        362,
+                                        733
+                                    ],
+                                    "score": 1.0,
+                                    "content": "algorithms, various distillation strategies and hyper-parameters.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        }
+                    ],
+                    "index": 40.5
+                }
+            ],
+            "page_idx": 6,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        108,
+                        27,
+                        292,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                26,
+                                293,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        26,
+                                        293,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Published as a conference paper at ICLR 2021",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        303,
+                        751,
+                        308,
+                        759
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                302,
+                                750,
+                                309,
+                                763
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        302,
+                                        750,
+                                        309,
+                                        763
+                                    ],
+                                    "score": 1.0,
+                                    "content": "",
+                                    "type": "text",
+                                    "height": 13,
+                                    "width": 7
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "table",
+                    "bbox": [
+                        106,
+                        120,
+                        509,
+                        202
+                    ],
+                    "blocks": [
+                        {
+                            "type": "table_caption",
+                            "bbox": [
+                                106,
+                                69,
+                                506,
+                                110
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        67,
+                                        506,
+                                        81
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                67,
+                                                506,
+                                                81
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Table 2: Object detection and instance segmentation results using contrastive self-supervised learning and",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 0
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        77,
+                                        506,
+                                        92
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                77,
+                                                358,
+                                                92
+                                            ],
+                                            "score": 1.0,
+                                            "content": "SEED distillation using ResNet-18 as backbone: bounding-box AP",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                358,
+                                                79,
+                                                383,
+                                                90
+                                            ],
+                                            "score": 0.85,
+                                            "content": "( \\mathbf { A P } ^ { b b } )",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                383,
+                                                77,
+                                                437,
+                                                92
+                                            ],
+                                            "score": 1.0,
+                                            "content": "and mask AP",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                437,
+                                                79,
+                                                466,
+                                                90
+                                            ],
+                                            "score": 0.85,
+                                            "content": "( \\mathbf { A P } ^ { m k } )",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                466,
+                                                77,
+                                                506,
+                                                92
+                                            ],
+                                            "score": 1.0,
+                                            "content": "evaluated",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        88,
+                                        506,
+                                        102
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                88,
+                                                506,
+                                                102
+                                            ],
+                                            "score": 1.0,
+                                            "content": "on VOC07-val and COCO testing split. More results on different backbones can be found in the Appendix.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 2
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        99,
+                                        327,
+                                        111
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                99,
+                                                327,
+                                                111
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Subscript in green represents improvement is larger than 0.3.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 3
+                                }
+                            ],
+                            "index": 1.5
+                        },
+                        {
+                            "type": "table_body",
+                            "bbox": [
+                                106,
+                                120,
+                                509,
+                                202
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        120,
+                                        509,
+                                        202
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                120,
+                                                509,
+                                                202
+                                            ],
+                                            "score": 0.963,
+                                            "html": "<table><tr><td rowspan=\"2\">S</td><td rowspan=\"2\">T</td><td colspan=\"3\">VOC Obj. Det.</td><td colspan=\"3\">COCO Obj. Det.</td><td colspan=\"3\">COCO Inst. Segm.</td></tr><tr><td>Apbb</td><td>AP</td><td>AP</td><td>Apbb</td><td>AP</td><td>APP</td><td>Apmk</td><td>AP</td><td>AP</td></tr><tr><td></td><td>X</td><td>46.1</td><td>74.5</td><td>48.6</td><td>35.0</td><td>53.9</td><td>37.7</td><td>31.0</td><td>51.1</td><td>33.1</td></tr><tr><td></td><td>R-50</td><td>46.1(0.0)</td><td>74.8(+0.3)</td><td>49.1(+0.5)</td><td>35.3(+0.3)</td><td>54.2(+0.3)</td><td>37.8(+0.1)</td><td>31.1(+0.1)</td><td>51.1(0.0)</td><td>33.2(+0.1)</td></tr><tr><td>R-18 R-101</td><td></td><td>46.8(+0.7)</td><td>75.8(+1.3)</td><td>49.3(+0.7)</td><td>35.3(+0.3)</td><td>54.3(+0.4)</td><td>37.9(+0.2)</td><td>31.3(+0.3)</td><td>51.3(+0.2)</td><td>33.4(+0.3)</td></tr><tr><td></td><td>R-152</td><td>46.8(+0.7)</td><td>75.9(+1.4)</td><td>50.2(+1.6)</td><td>35.4(+0.4)</td><td>54.4(+0.5)</td><td>38.0(+0.3)</td><td>31.3(+0.3)</td><td>51.4(+0.3)</td><td>33.4(+0.3)</td></tr></table>",
+                                            "type": "table",
+                                            "image_path": "bc2a89152889a8d15d369e7b291816928ecaa6276dd595b93bc9cd3b191f82cd.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 5,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        120,
+                                        509,
+                                        147.33333333333334
+                                    ],
+                                    "spans": [],
+                                    "index": 4
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        147.33333333333334,
+                                        509,
+                                        174.66666666666669
+                                    ],
+                                    "spans": [],
+                                    "index": 5
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        174.66666666666669,
+                                        509,
+                                        202.00000000000003
+                                    ],
+                                    "spans": [],
+                                    "index": 6
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 3.25
+                },
+                {
+                    "type": "image",
+                    "bbox": [
+                        107,
+                        225,
+                        503,
+                        321
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                107,
+                                225,
+                                503,
+                                321
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        225,
+                                        503,
+                                        321
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                107,
+                                                225,
+                                                503,
+                                                321
+                                            ],
+                                            "score": 0.96,
+                                            "type": "image",
+                                            "image_path": "15d565abe6e76f1be7191527af52d1cfceecfa87f6083e78a4a48b0e6be4ff99.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 8,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        225,
+                                        503,
+                                        257.0
+                                    ],
+                                    "spans": [],
+                                    "index": 7
+                                },
+                                {
+                                    "bbox": [
+                                        107,
+                                        257.0,
+                                        503,
+                                        289.0
+                                    ],
+                                    "spans": [],
+                                    "index": 8
+                                },
+                                {
+                                    "bbox": [
+                                        107,
+                                        289.0,
+                                        503,
+                                        321.0
+                                    ],
+                                    "spans": [],
+                                    "index": 9
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                107,
+                                325,
+                                505,
+                                356
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        325,
+                                        505,
+                                        337
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                325,
+                                                230,
+                                                337
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 4: ImageNet-1k Accuracy",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                230,
+                                                326,
+                                                243,
+                                                336
+                                            ],
+                                            "score": 0.73,
+                                            "content": "( \\% )",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                244,
+                                                325,
+                                                505,
+                                                337
+                                            ],
+                                            "score": 1.0,
+                                            "content": "of student network (EfficientNet-B0 and ResNet-18) transferred to other",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 10
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        334,
+                                        506,
+                                        347
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                334,
+                                                506,
+                                                347
+                                            ],
+                                            "score": 1.0,
+                                            "content": "domains (CIFAR-10, CIFAR-100, SUN-397 datasets) with and without distillation from lager architectures",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 11
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        345,
+                                        187,
+                                        356
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                345,
+                                                187,
+                                                356
+                                            ],
+                                            "score": 1.0,
+                                            "content": "(ResNet-50/101/152).",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 12
+                                }
+                            ],
+                            "index": 11
+                        }
+                    ],
+                    "index": 9.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        373,
+                        505,
+                        405
+                    ],
+                    "lines": [],
+                    "index": 14,
+                    "bbox_fs": [
+                        105,
+                        372,
+                        505,
+                        406
+                    ],
+                    "lines_deleted": true
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        411,
+                        505,
+                        532
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                411,
+                                506,
+                                423
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        411,
+                                        506,
+                                        423
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Transferring to Classification. To further study whether the improvement of the learned represen-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                423,
+                                505,
+                                434
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        423,
+                                        505,
+                                        434
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tations by distillation is confined to ImageNet, we evaluate on additional classification datasets to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                433,
+                                506,
+                                445
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        433,
+                                        506,
+                                        445
+                                    ],
+                                    "score": 1.0,
+                                    "content": "study the generalization and transferability of the feature representation. We strictly follow the linear",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                444,
+                                505,
+                                456
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        444,
+                                        505,
+                                        456
+                                    ],
+                                    "score": 1.0,
+                                    "content": "evaluation and fine-tuning settings from (Kornblith et al., 2019; Chen et al., 2020a; Grill et al., 2020),",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                455,
+                                505,
+                                466
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        455,
+                                        505,
+                                        466
+                                    ],
+                                    "score": 1.0,
+                                    "content": "that a linear layer is trained on the basis of frozen features. We report Top-1 Accuracy of models",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                465,
+                                506,
+                                478
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        465,
+                                        506,
+                                        478
+                                    ],
+                                    "score": 1.0,
+                                    "content": "before and after distillation from various architectures on CIFAR-10, CIFAR-100 (Krizhevsky et al.,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                476,
+                                505,
+                                490
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        476,
+                                        505,
+                                        490
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2009), SUN-397 (Xiao et al., 2010) datasets (see Figure 4). More details regarding pre-processing",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                488,
+                                505,
+                                500
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        488,
+                                        505,
+                                        500
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and training can be found in A.1.2. Notably, we observe that our distillation surpasses contrastive",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                498,
+                                506,
+                                511
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        498,
+                                        506,
+                                        511
+                                    ],
+                                    "score": 1.0,
+                                    "content": "self-supervised pre-training consistently on all benchmarks, verifying the effectiveness of SEED.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                510,
+                                505,
+                                522
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        510,
+                                        505,
+                                        522
+                                    ],
+                                    "score": 1.0,
+                                    "content": "This also proves the generalization ability of the learned representations from distillation to a wide",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                521,
+                                279,
+                                533
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        521,
+                                        279,
+                                        533
+                                    ],
+                                    "score": 1.0,
+                                    "content": "range of data domain and different classes.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        }
+                    ],
+                    "index": 21,
+                    "bbox_fs": [
+                        105,
+                        411,
+                        506,
+                        533
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        538,
+                        505,
+                        669
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                537,
+                                506,
+                                550
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        537,
+                                        506,
+                                        550
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Transferring to Detection and Segmentation. We conduct two downstream tasks here. The first",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                549,
+                                506,
+                                561
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        549,
+                                        424,
+                                        561
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is Faster R-CNN (Ren et al., 2015) model for object detection trained on VOC-",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        425,
+                                        549,
+                                        452,
+                                        559
+                                    ],
+                                    "score": 0.79,
+                                    "content": "^ { 0 7 + 1 2 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        452,
+                                        549,
+                                        471,
+                                        561
+                                    ],
+                                    "score": 1.0,
+                                    "content": "train",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        471,
+                                        551,
+                                        478,
+                                        558
+                                    ],
+                                    "score": 0.32,
+                                    "content": "^ +",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        478,
+                                        549,
+                                        506,
+                                        561
+                                    ],
+                                    "score": 1.0,
+                                    "content": "val set",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                559,
+                                505,
+                                572
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        559,
+                                        505,
+                                        572
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and evaluated on VOC-07 test split. The second is Mask R-CNN (He et al., 2017) model for the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                570,
+                                505,
+                                582
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        570,
+                                        505,
+                                        582
+                                    ],
+                                    "score": 1.0,
+                                    "content": "object detection and instance segmentation on COCO 2017 dataset (Lin et al., 2014). The pre-trained",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                582,
+                                505,
+                                594
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        582,
+                                        505,
+                                        594
+                                    ],
+                                    "score": 1.0,
+                                    "content": "model serves as the initial weight and following He et al. (2020), we fine-tune all the layers of the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                592,
+                                506,
+                                605
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        592,
+                                        506,
+                                        605
+                                    ],
+                                    "score": 1.0,
+                                    "content": "model. More experiment settings can be found in A.2. The results are illustrated in Table 2. As we",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                604,
+                                506,
+                                616
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        604,
+                                        506,
+                                        616
+                                    ],
+                                    "score": 1.0,
+                                    "content": "can see, on VOC, the distilled pre-trained model achieves a large improvement. With ResNet-152 as",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                614,
+                                506,
+                                626
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        614,
+                                        380,
+                                        626
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the teacher network, the Resnet18-based Faster R-CNN model shows",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        380,
+                                        615,
+                                        399,
+                                        625
+                                    ],
+                                    "score": 0.83,
+                                    "content": "+ 0 . 7",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        400,
+                                        614,
+                                        506,
+                                        626
+                                    ],
+                                    "score": 1.0,
+                                    "content": "point improvement on AP,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                624,
+                                506,
+                                638
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        626,
+                                        125,
+                                        636
+                                    ],
+                                    "score": 0.82,
+                                    "content": "+ 1 . 4",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        125,
+                                        624,
+                                        194,
+                                        638
+                                    ],
+                                    "score": 1.0,
+                                    "content": "improvement on",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        194,
+                                        626,
+                                        216,
+                                        637
+                                    ],
+                                    "score": 0.9,
+                                    "content": "\\mathrm { { A P } _ { 5 0 } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        217,
+                                        624,
+                                        234,
+                                        638
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        234,
+                                        626,
+                                        254,
+                                        636
+                                    ],
+                                    "score": 0.85,
+                                    "content": "+ 1 . 6",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        255,
+                                        624,
+                                        267,
+                                        638
+                                    ],
+                                    "score": 1.0,
+                                    "content": "on",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        268,
+                                        626,
+                                        290,
+                                        637
+                                    ],
+                                    "score": 0.9,
+                                    "content": "\\mathsf { A P } _ { 7 5 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        290,
+                                        624,
+                                        506,
+                                        638
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". On COCO, the improvement is relatively minor and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                636,
+                                505,
+                                648
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        636,
+                                        295,
+                                        648
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the reason could be that COCO training set has",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        295,
+                                        637,
+                                        325,
+                                        647
+                                    ],
+                                    "score": 0.89,
+                                    "content": "{ \\sim } 1 1 8 \\mathrm { k }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        325,
+                                        636,
+                                        472,
+                                        648
+                                    ],
+                                    "score": 1.0,
+                                    "content": "training images while VOC has only",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        473,
+                                        637,
+                                        505,
+                                        647
+                                    ],
+                                    "score": 0.88,
+                                    "content": "{ \\sim } 1 6 . 5 \\mathrm { k }",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                648,
+                                505,
+                                659
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        648,
+                                        505,
+                                        659
+                                    ],
+                                    "score": 1.0,
+                                    "content": "training images. A larger training set with more fine-tuning iterations reduces the importance of the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                658,
+                                169,
+                                671
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        658,
+                                        169,
+                                        671
+                                    ],
+                                    "score": 1.0,
+                                    "content": "initial weights.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        }
+                    ],
+                    "index": 32.5,
+                    "bbox_fs": [
+                        105,
+                        537,
+                        506,
+                        671
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        687,
+                        208,
+                        698
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                686,
+                                210,
+                                699
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        686,
+                                        210,
+                                        699
+                                    ],
+                                    "score": 1.0,
+                                    "content": "4.3 ABLATION STUDY",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        }
+                    ],
+                    "index": 39
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        709,
+                        503,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                707,
+                                505,
+                                723
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        707,
+                                        505,
+                                        723
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We now explore the effects of distillation using different Teacher architectures, Teacher Pre-training",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                720,
+                                362,
+                                733
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        720,
+                                        362,
+                                        733
+                                    ],
+                                    "score": 1.0,
+                                    "content": "algorithms, various distillation strategies and hyper-parameters.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        }
+                    ],
+                    "index": 40.5,
+                    "bbox_fs": [
+                        105,
+                        707,
+                        505,
+                        733
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "table",
+                    "bbox": [
+                        110,
+                        170,
+                        292,
+                        272
+                    ],
+                    "blocks": [
+                        {
+                            "type": "table_caption",
+                            "bbox": [
+                                106,
+                                86,
+                                299,
+                                166
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        87,
+                                        300,
+                                        97
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                87,
+                                                228,
+                                                97
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Table 3: ImageNet-1k Accuracy",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                229,
+                                                87,
+                                                243,
+                                                96
+                                            ],
+                                            "score": 0.59,
+                                            "content": "( \\% )",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                243,
+                                                87,
+                                                300,
+                                                97
+                                            ],
+                                            "score": 1.0,
+                                            "content": "of student net-",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 0
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        96,
+                                        300,
+                                        107
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                96,
+                                                300,
+                                                107
+                                            ],
+                                            "score": 1.0,
+                                            "content": "work (ResNet-18) distilled from variants of self-",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        106,
+                                        300,
+                                        118
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                106,
+                                                300,
+                                                118
+                                            ],
+                                            "score": 1.0,
+                                            "content": "supervised ResNet-50. P-E/D-E represent the pre-",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 2
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        116,
+                                        299,
+                                        127
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                116,
+                                                299,
+                                                127
+                                            ],
+                                            "score": 1.0,
+                                            "content": "training and distillation epochs. T./S.-Top represent",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 3
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        127,
+                                        299,
+                                        138
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                127,
+                                                299,
+                                                138
+                                            ],
+                                            "score": 1.0,
+                                            "content": "testing accuracy of Teacher and Student. ∗ represents",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 4
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        136,
+                                        299,
+                                        147
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                136,
+                                                299,
+                                                147
+                                            ],
+                                            "score": 1.0,
+                                            "content": "distillation using additional small patches. First row",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 5
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        145,
+                                        299,
+                                        157
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                145,
+                                                299,
+                                                157
+                                            ],
+                                            "score": 1.0,
+                                            "content": "is the ResNet-18 SSL baseline using MoCo-v2 trained",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 6
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        155,
+                                        165,
+                                        168
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                155,
+                                                165,
+                                                168
+                                            ],
+                                            "score": 1.0,
+                                            "content": "for 200 epochs.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 7
+                                }
+                            ],
+                            "index": 3.5
+                        },
+                        {
+                            "type": "table_body",
+                            "bbox": [
+                                110,
+                                170,
+                                292,
+                                272
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        110,
+                                        170,
+                                        292,
+                                        272
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                110,
+                                                170,
+                                                292,
+                                                272
+                                            ],
+                                            "score": 0.979,
+                                            "html": "<table><tr><td>Teacher</td><td></td><td>P-E D-E T.Top-1 S.Top-1 S.Top-5</td></tr><tr><td></td><td>X X</td><td>X 52.5 77.0</td></tr><tr><td>X</td><td>200</td><td></td></tr><tr><td>MoCo</td><td>200</td><td>60.6 52.1 77.0</td></tr><tr><td>SimCLR</td><td>200 200</td><td>65.6 57.5 81.7</td></tr><tr><td>MoCo-v2 200 800</td><td>200 67.4</td><td>57.9 82.0</td></tr><tr><td>SWAV 800</td><td>200 71.1</td><td>60.5 83.5</td></tr><tr><td></td><td>100 75.3</td><td>61.1 83.8</td></tr><tr><td>800</td><td>200 75.3</td><td>61.7 84.2</td></tr><tr><td>800</td><td>400 75.3</td><td>62.0 84.4</td></tr><tr><td>SWAV* 800</td><td>200 75.3</td><td>62.6 84.8</td></tr></table>",
+                                            "type": "table",
+                                            "image_path": "6b287c612ea27b0206b8c1a70fa565f801cdb95b0af984578c7bb7576521984e.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 11.5,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        110,
+                                        170,
+                                        292,
+                                        182.75
+                                    ],
+                                    "spans": [],
+                                    "index": 8
+                                },
+                                {
+                                    "bbox": [
+                                        110,
+                                        182.75,
+                                        292,
+                                        195.5
+                                    ],
+                                    "spans": [],
+                                    "index": 9
+                                },
+                                {
+                                    "bbox": [
+                                        110,
+                                        195.5,
+                                        292,
+                                        208.25
+                                    ],
+                                    "spans": [],
+                                    "index": 10
+                                },
+                                {
+                                    "bbox": [
+                                        110,
+                                        208.25,
+                                        292,
+                                        221.0
+                                    ],
+                                    "spans": [],
+                                    "index": 11
+                                },
+                                {
+                                    "bbox": [
+                                        110,
+                                        221.0,
+                                        292,
+                                        233.75
+                                    ],
+                                    "spans": [],
+                                    "index": 12
+                                },
+                                {
+                                    "bbox": [
+                                        110,
+                                        233.75,
+                                        292,
+                                        246.5
+                                    ],
+                                    "spans": [],
+                                    "index": 13
+                                },
+                                {
+                                    "bbox": [
+                                        110,
+                                        246.5,
+                                        292,
+                                        259.25
+                                    ],
+                                    "spans": [],
+                                    "index": 14
+                                },
+                                {
+                                    "bbox": [
+                                        110,
+                                        259.25,
+                                        292,
+                                        272.0
+                                    ],
+                                    "spans": [],
+                                    "index": 15
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 7.5
+                },
+                {
+                    "type": "image",
+                    "bbox": [
+                        324,
+                        94,
+                        503,
+                        228
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                324,
+                                94,
+                                503,
+                                228
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        324,
+                                        94,
+                                        503,
+                                        228
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                324,
+                                                94,
+                                                503,
+                                                228
+                                            ],
+                                            "score": 0.971,
+                                            "type": "image",
+                                            "image_path": "8cecf45d517b8c5b96207f2fc90806edfe43935c7cfa9552a0ceb80d0db26208.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 21,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        324,
+                                        94,
+                                        503,
+                                        106.18181818181819
+                                    ],
+                                    "spans": [],
+                                    "index": 16
+                                },
+                                {
+                                    "bbox": [
+                                        324,
+                                        106.18181818181819,
+                                        503,
+                                        118.36363636363637
+                                    ],
+                                    "spans": [],
+                                    "index": 17
+                                },
+                                {
+                                    "bbox": [
+                                        324,
+                                        118.36363636363637,
+                                        503,
+                                        130.54545454545456
+                                    ],
+                                    "spans": [],
+                                    "index": 18
+                                },
+                                {
+                                    "bbox": [
+                                        324,
+                                        130.54545454545456,
+                                        503,
+                                        142.72727272727275
+                                    ],
+                                    "spans": [],
+                                    "index": 19
+                                },
+                                {
+                                    "bbox": [
+                                        324,
+                                        142.72727272727275,
+                                        503,
+                                        154.90909090909093
+                                    ],
+                                    "spans": [],
+                                    "index": 20
+                                },
+                                {
+                                    "bbox": [
+                                        324,
+                                        154.90909090909093,
+                                        503,
+                                        167.09090909090912
+                                    ],
+                                    "spans": [],
+                                    "index": 21
+                                },
+                                {
+                                    "bbox": [
+                                        324,
+                                        167.09090909090912,
+                                        503,
+                                        179.2727272727273
+                                    ],
+                                    "spans": [],
+                                    "index": 22
+                                },
+                                {
+                                    "bbox": [
+                                        324,
+                                        179.2727272727273,
+                                        503,
+                                        191.4545454545455
+                                    ],
+                                    "spans": [],
+                                    "index": 23
+                                },
+                                {
+                                    "bbox": [
+                                        324,
+                                        191.4545454545455,
+                                        503,
+                                        203.63636363636368
+                                    ],
+                                    "spans": [],
+                                    "index": 24
+                                },
+                                {
+                                    "bbox": [
+                                        324,
+                                        203.63636363636368,
+                                        503,
+                                        215.81818181818187
+                                    ],
+                                    "spans": [],
+                                    "index": 25
+                                },
+                                {
+                                    "bbox": [
+                                        324,
+                                        215.81818181818187,
+                                        503,
+                                        228.00000000000006
+                                    ],
+                                    "spans": [],
+                                    "index": 26
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                320,
+                                230,
+                                505,
+                                270
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        319,
+                                        230,
+                                        505,
+                                        241
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                319,
+                                                230,
+                                                406,
+                                                241
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 5: Accuracy",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                406,
+                                                231,
+                                                420,
+                                                240
+                                            ],
+                                            "score": 0.66,
+                                            "content": "( \\% )",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                421,
+                                                230,
+                                                505,
+                                                241
+                                            ],
+                                            "score": 1.0,
+                                            "content": "of student networks",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 27
+                                },
+                                {
+                                    "bbox": [
+                                        320,
+                                        239,
+                                        506,
+                                        250
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                320,
+                                                239,
+                                                506,
+                                                250
+                                            ],
+                                            "score": 1.0,
+                                            "content": "(EfficientNet-b0 and ResNet-18) on ImageNet dis-",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 28
+                                },
+                                {
+                                    "bbox": [
+                                        319,
+                                        250,
+                                        505,
+                                        261
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                319,
+                                                250,
+                                                505,
+                                                261
+                                            ],
+                                            "score": 1.0,
+                                            "content": "tilled from wider MoCo-v2 pre-trained ResNet",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 29
+                                },
+                                {
+                                    "bbox": [
+                                        319,
+                                        259,
+                                        411,
+                                        271
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                319,
+                                                259,
+                                                411,
+                                                271
+                                            ],
+                                            "score": 1.0,
+                                            "content": "(ResNet-50/101/152×2).",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 30
+                                }
+                            ],
+                            "index": 28.5
+                        }
+                    ],
+                    "index": 24.75
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        300,
+                        505,
+                        366
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                301,
+                                505,
+                                312
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        301,
+                                        505,
+                                        312
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Different Teacher Networks. Figure 5 summarizes the accuracy of ResNet-18 and EfficientNet-B0",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                311,
+                                506,
+                                324
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        311,
+                                        506,
+                                        324
+                                    ],
+                                    "score": 1.0,
+                                    "content": "distilled from wider and deeper ResNet architectures. We see clear performance improvement as",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                322,
+                                506,
+                                335
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        322,
+                                        506,
+                                        335
+                                    ],
+                                    "score": 1.0,
+                                    "content": "depth and width of teacher network increase: compared to ResNet-50, deeper (ResNet-101) and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                333,
+                                506,
+                                346
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        333,
+                                        166,
+                                        346
+                                    ],
+                                    "score": 1.0,
+                                    "content": "wider (ResNet-",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        166,
+                                        334,
+                                        190,
+                                        344
+                                    ],
+                                    "score": 0.68,
+                                    "content": "{ 5 0 \\times 2 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        191,
+                                        333,
+                                        506,
+                                        346
+                                    ],
+                                    "score": 1.0,
+                                    "content": ") substantially improve the accuracy. However, further architectural enlargement",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                345,
+                                506,
+                                357
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        345,
+                                        506,
+                                        357
+                                    ],
+                                    "score": 1.0,
+                                    "content": "has relatively limited effects, and we suspect the accuracy might be limited by the student network",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                356,
+                                192,
+                                369
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        356,
+                                        192,
+                                        369
+                                    ],
+                                    "score": 1.0,
+                                    "content": "capacity in this case.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        }
+                    ],
+                    "index": 33.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        372,
+                        505,
+                        460
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                372,
+                                505,
+                                384
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        372,
+                                        505,
+                                        384
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Different Teacher Pre-training Algorithms. In Table 3, we show the Top-1 accuracy of ResNet-18",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                383,
+                                506,
+                                395
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        383,
+                                        506,
+                                        395
+                                    ],
+                                    "score": 1.0,
+                                    "content": "distilled from ResNet-50 with different pre-training algorithms, i.e., MoCo-V1 (He et al., 2020),",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                394,
+                                506,
+                                406
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        394,
+                                        506,
+                                        406
+                                    ],
+                                    "score": 1.0,
+                                    "content": "MoCo-V2 (Chen et al., 2020c), SimCLR (Chen et al., 2020a), and SWAV (Caron et al., 2020)).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                405,
+                                505,
+                                418
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        405,
+                                        505,
+                                        418
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Notably, the aforementioned methods all unanimously adopt contrastive-based pre-training except",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                415,
+                                506,
+                                429
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        415,
+                                        506,
+                                        429
+                                    ],
+                                    "score": 1.0,
+                                    "content": "SWAV, which is based upon online clustering. We find that our SEED is agnostic to pre-training",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                427,
+                                506,
+                                440
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        427,
+                                        506,
+                                        440
+                                    ],
+                                    "score": 1.0,
+                                    "content": "approaches, making it easy to use any self-supervised models (including clustering-based approach",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                437,
+                                505,
+                                450
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        437,
+                                        505,
+                                        450
+                                    ],
+                                    "score": 1.0,
+                                    "content": "like SWAV) in self-supervised distillation. In addition, we observe that more training epochs for both",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                448,
+                                351,
+                                461
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        448,
+                                        137,
+                                        461
+                                    ],
+                                    "score": 1.0,
+                                    "content": "teacher",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        138,
+                                        450,
+                                        155,
+                                        459
+                                    ],
+                                    "score": 0.25,
+                                    "content": "S S L",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        155,
+                                        448,
+                                        351,
+                                        461
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and distillation epochs can bring beneficial gain.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        }
+                    ],
+                    "index": 40.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        466,
+                        505,
+                        642
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                465,
+                                506,
+                                478
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        465,
+                                        506,
+                                        478
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Other Distillation Strategies. We explore several alternative distillation strategies. l2-Distance:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                476,
+                                506,
+                                489
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        476,
+                                        147,
+                                        489
+                                    ],
+                                    "score": 1.0,
+                                    "content": "where the",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        147,
+                                        478,
+                                        156,
+                                        487
+                                    ],
+                                    "score": 0.66,
+                                    "content": "l 2",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        157,
+                                        476,
+                                        506,
+                                        489
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-distance of teacher & student’s embeddings are minimized, motivated by Romero et al.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                488,
+                                506,
+                                500
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        488,
+                                        140,
+                                        500
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(2014).",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        140,
+                                        488,
+                                        151,
+                                        498
+                                    ],
+                                    "score": 0.67,
+                                    "content": "K",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        151,
+                                        488,
+                                        232,
+                                        500
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-Means: we exploit",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        233,
+                                        489,
+                                        241,
+                                        498
+                                    ],
+                                    "score": 0.71,
+                                    "content": "K",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        242,
+                                        488,
+                                        506,
+                                        500
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-Means clustering to assign a pseudo-label based on the teacher",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 47
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                498,
+                                505,
+                                512
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        498,
+                                        505,
+                                        512
+                                    ],
+                                    "score": 1.0,
+                                    "content": "network’s representation. Online Clustering: we continuously update the clustering centers during",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 48
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                510,
+                                505,
+                                522
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        510,
+                                        505,
+                                        522
+                                    ],
+                                    "score": 1.0,
+                                    "content": "distillation for pseudo-label generation. Binary Contrastive Loss: we adopt an Info-NCE alike loss",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 49
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                520,
+                                505,
+                                533
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        520,
+                                        505,
+                                        533
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for contrastive distillation (Tian et al., 2019a). We provide details for other strategies in A.4. Table 4",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 50
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                531,
+                                506,
+                                544
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        531,
+                                        506,
+                                        544
+                                    ],
+                                    "score": 1.0,
+                                    "content": "shows the results for each method on ResNet-18 (student) distilled from ResNet-50. From the results,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 51
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                543,
+                                505,
+                                555
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        543,
+                                        150,
+                                        555
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the simple",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        150,
+                                        543,
+                                        159,
+                                        553
+                                    ],
+                                    "score": 0.38,
+                                    "content": "l 2",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        160,
+                                        543,
+                                        505,
+                                        555
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-distance minimizing approach can achieve a decent accuracy, which demonstrates the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 52
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                554,
+                                505,
+                                566
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        554,
+                                        505,
+                                        566
+                                    ],
+                                    "score": 1.0,
+                                    "content": "effectiveness of applying the distillation idea to the self-supervised learning. Beyond that, we study",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 53
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                564,
+                                506,
+                                578
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        564,
+                                        506,
+                                        578
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the effect of the original SSL (MoCo-V2) supervision as supplementary loss to SEED and find it does",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 54
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                576,
+                                505,
+                                588
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        576,
+                                        505,
+                                        588
+                                    ],
+                                    "score": 1.0,
+                                    "content": "not bring additional benefits to distillation. We find close results from these two strategies (Top-1",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 55
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                585,
+                                506,
+                                599
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        585,
+                                        226,
+                                        599
+                                    ],
+                                    "score": 1.0,
+                                    "content": "linear Acc.), SEED achieves",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        226,
+                                        587,
+                                        254,
+                                        597
+                                    ],
+                                    "score": 0.88,
+                                    "content": "5 7 . 9 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        254,
+                                        585,
+                                        284,
+                                        599
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", while",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        284,
+                                        587,
+                                        363,
+                                        597
+                                    ],
+                                    "score": 0.63,
+                                    "content": "\\mathbf { S E E D + M o C o - V } 2",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        363,
+                                        585,
+                                        401,
+                                        599
+                                    ],
+                                    "score": 1.0,
+                                    "content": "achieves",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        401,
+                                        587,
+                                        429,
+                                        597
+                                    ],
+                                    "score": 0.92,
+                                    "content": "5 7 . 6 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        429,
+                                        585,
+                                        506,
+                                        599
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". This implies that",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 56
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                596,
+                                506,
+                                610
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        596,
+                                        332,
+                                        610
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the loss of SEED can to a large extent cover the original",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        332,
+                                        598,
+                                        349,
+                                        608
+                                    ],
+                                    "score": 0.3,
+                                    "content": "S S L",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        350,
+                                        596,
+                                        506,
+                                        610
+                                    ],
+                                    "score": 1.0,
+                                    "content": "loss, and it is not necessary to conduct",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 57
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                609,
+                                505,
+                                620
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        609,
+                                        505,
+                                        620
+                                    ],
+                                    "score": 1.0,
+                                    "content": "SSL any further during distillation. Meanwhile, our proposed SEED outperforms these alternatives",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 58
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                619,
+                                506,
+                                633
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        619,
+                                        506,
+                                        633
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with highest accuracy, which shows the superiority of aligning the student towards the teacher and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 59
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                631,
+                                265,
+                                643
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        631,
+                                        265,
+                                        643
+                                    ],
+                                    "score": 1.0,
+                                    "content": "contrasting with the irrelevant samples.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 60
+                        }
+                    ],
+                    "index": 52.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        647,
+                        505,
+                        736
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                646,
+                                505,
+                                660
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        646,
+                                        505,
+                                        660
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Other Hyper-Parameters. Table 5 summarizes the distillation performances on multiple datasets",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 61
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                657,
+                                506,
+                                671
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        657,
+                                        219,
+                                        671
+                                    ],
+                                    "score": 1.0,
+                                    "content": "using different temperature",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        219,
+                                        657,
+                                        232,
+                                        668
+                                    ],
+                                    "score": 0.87,
+                                    "content": "\\tau ^ { T }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        233,
+                                        657,
+                                        445,
+                                        671
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". We observe a better performance when decreasing",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        445,
+                                        657,
+                                        459,
+                                        668
+                                    ],
+                                    "score": 0.88,
+                                    "content": "\\tau ^ { T }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        459,
+                                        657,
+                                        506,
+                                        671
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to 0.01 for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 62
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                669,
+                                505,
+                                681
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        669,
+                                        445,
+                                        681
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ImageNet-1k and CIFAR-10 dataset, and to 1e-3 for CIFAR-100 datasets. When",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        445,
+                                        671,
+                                        452,
+                                        679
+                                    ],
+                                    "score": 0.78,
+                                    "content": "\\tau",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        453,
+                                        669,
+                                        505,
+                                        681
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is large, the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 63
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                679,
+                                506,
+                                694
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        679,
+                                        259,
+                                        693
+                                    ],
+                                    "score": 1.0,
+                                    "content": "softmax-normalized similarity score of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        260,
+                                        680,
+                                        272,
+                                        694
+                                    ],
+                                    "score": 0.9,
+                                    "content": "p _ { j } ^ { T }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        272,
+                                        679,
+                                        307,
+                                        693
+                                    ],
+                                    "score": 1.0,
+                                    "content": "between",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        308,
+                                        680,
+                                        320,
+                                        693
+                                    ],
+                                    "score": 0.9,
+                                    "content": "{ \\mathbf z } _ { i } ^ { T }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        320,
+                                        679,
+                                        371,
+                                        693
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and instance",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        371,
+                                        681,
+                                        383,
+                                        693
+                                    ],
+                                    "score": 0.89,
+                                    "content": "{ \\bf d } _ { j }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        383,
+                                        679,
+                                        433,
+                                        693
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in the queue",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        433,
+                                        680,
+                                        450,
+                                        691
+                                    ],
+                                    "score": 0.89,
+                                    "content": "\\mathbf { D } ^ { + }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        450,
+                                        679,
+                                        506,
+                                        693
+                                    ],
+                                    "score": 1.0,
+                                    "content": "also becomes",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 64
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                691,
+                                506,
+                                703
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        691,
+                                        506,
+                                        703
+                                    ],
+                                    "score": 1.0,
+                                    "content": "large, which means the student’s feature should be less discriminative with the features of other",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 65
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                702,
+                                506,
+                                714
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        702,
+                                        227,
+                                        714
+                                    ],
+                                    "score": 1.0,
+                                    "content": "images to some extent. When",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        228,
+                                        702,
+                                        240,
+                                        712
+                                    ],
+                                    "score": 0.88,
+                                    "content": "\\tau ^ { T }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        240,
+                                        702,
+                                        251,
+                                        714
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        251,
+                                        704,
+                                        257,
+                                        712
+                                    ],
+                                    "score": 0.43,
+                                    "content": "_ 0",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        257,
+                                        702,
+                                        506,
+                                        714
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", the teacher model will generate a one-hot vector, which only",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 66
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                713,
+                                506,
+                                726
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        713,
+                                        130,
+                                        726
+                                    ],
+                                    "score": 1.0,
+                                    "content": "treats",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        130,
+                                        713,
+                                        143,
+                                        725
+                                    ],
+                                    "score": 0.9,
+                                    "content": "{ \\mathbf z } _ { i } ^ { T }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        143,
+                                        713,
+                                        444,
+                                        726
+                                    ],
+                                    "score": 1.0,
+                                    "content": "as a positive instance and all others in the queue as negative. Thus, the best",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        444,
+                                        715,
+                                        451,
+                                        723
+                                    ],
+                                    "score": 0.76,
+                                    "content": "\\tau",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        452,
+                                        713,
+                                        506,
+                                        726
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is a trade-off",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 67
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                724,
+                                504,
+                                736
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        724,
+                                        504,
+                                        736
+                                    ],
+                                    "score": 1.0,
+                                    "content": "depending on the data distribution. We further compare effect of different hyper-parameters in A.8.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 68
+                        }
+                    ],
+                    "index": 64.5
+                }
+            ],
+            "page_idx": 7,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        108,
+                        27,
+                        292,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                26,
+                                293,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        26,
+                                        293,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Published as a conference paper at ICLR 2021",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        302,
+                        752,
+                        308,
+                        759
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                302,
+                                750,
+                                309,
+                                761
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        302,
+                                        750,
+                                        309,
+                                        761
+                                    ],
+                                    "score": 1.0,
+                                    "content": "8",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "table",
+                    "bbox": [
+                        110,
+                        170,
+                        292,
+                        272
+                    ],
+                    "blocks": [
+                        {
+                            "type": "table_caption",
+                            "bbox": [
+                                106,
+                                86,
+                                299,
+                                166
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        87,
+                                        300,
+                                        97
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                87,
+                                                228,
+                                                97
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Table 3: ImageNet-1k Accuracy",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                229,
+                                                87,
+                                                243,
+                                                96
+                                            ],
+                                            "score": 0.59,
+                                            "content": "( \\% )",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                243,
+                                                87,
+                                                300,
+                                                97
+                                            ],
+                                            "score": 1.0,
+                                            "content": "of student net-",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 0
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        96,
+                                        300,
+                                        107
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                96,
+                                                300,
+                                                107
+                                            ],
+                                            "score": 1.0,
+                                            "content": "work (ResNet-18) distilled from variants of self-",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        106,
+                                        300,
+                                        118
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                106,
+                                                300,
+                                                118
+                                            ],
+                                            "score": 1.0,
+                                            "content": "supervised ResNet-50. P-E/D-E represent the pre-",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 2
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        116,
+                                        299,
+                                        127
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                116,
+                                                299,
+                                                127
+                                            ],
+                                            "score": 1.0,
+                                            "content": "training and distillation epochs. T./S.-Top represent",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 3
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        127,
+                                        299,
+                                        138
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                127,
+                                                299,
+                                                138
+                                            ],
+                                            "score": 1.0,
+                                            "content": "testing accuracy of Teacher and Student. ∗ represents",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 4
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        136,
+                                        299,
+                                        147
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                136,
+                                                299,
+                                                147
+                                            ],
+                                            "score": 1.0,
+                                            "content": "distillation using additional small patches. First row",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 5
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        145,
+                                        299,
+                                        157
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                145,
+                                                299,
+                                                157
+                                            ],
+                                            "score": 1.0,
+                                            "content": "is the ResNet-18 SSL baseline using MoCo-v2 trained",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 6
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        155,
+                                        165,
+                                        168
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                155,
+                                                165,
+                                                168
+                                            ],
+                                            "score": 1.0,
+                                            "content": "for 200 epochs.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 7
+                                }
+                            ],
+                            "index": 3.5
+                        },
+                        {
+                            "type": "table_body",
+                            "bbox": [
+                                110,
+                                170,
+                                292,
+                                272
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        110,
+                                        170,
+                                        292,
+                                        272
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                110,
+                                                170,
+                                                292,
+                                                272
+                                            ],
+                                            "score": 0.979,
+                                            "html": "<table><tr><td>Teacher</td><td></td><td>P-E D-E T.Top-1 S.Top-1 S.Top-5</td></tr><tr><td></td><td>X X</td><td>X 52.5 77.0</td></tr><tr><td>X</td><td>200</td><td></td></tr><tr><td>MoCo</td><td>200</td><td>60.6 52.1 77.0</td></tr><tr><td>SimCLR</td><td>200 200</td><td>65.6 57.5 81.7</td></tr><tr><td>MoCo-v2 200 800</td><td>200 67.4</td><td>57.9 82.0</td></tr><tr><td>SWAV 800</td><td>200 71.1</td><td>60.5 83.5</td></tr><tr><td></td><td>100 75.3</td><td>61.1 83.8</td></tr><tr><td>800</td><td>200 75.3</td><td>61.7 84.2</td></tr><tr><td>800</td><td>400 75.3</td><td>62.0 84.4</td></tr><tr><td>SWAV* 800</td><td>200 75.3</td><td>62.6 84.8</td></tr></table>",
+                                            "type": "table",
+                                            "image_path": "6b287c612ea27b0206b8c1a70fa565f801cdb95b0af984578c7bb7576521984e.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 11.5,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        110,
+                                        170,
+                                        292,
+                                        182.75
+                                    ],
+                                    "spans": [],
+                                    "index": 8
+                                },
+                                {
+                                    "bbox": [
+                                        110,
+                                        182.75,
+                                        292,
+                                        195.5
+                                    ],
+                                    "spans": [],
+                                    "index": 9
+                                },
+                                {
+                                    "bbox": [
+                                        110,
+                                        195.5,
+                                        292,
+                                        208.25
+                                    ],
+                                    "spans": [],
+                                    "index": 10
+                                },
+                                {
+                                    "bbox": [
+                                        110,
+                                        208.25,
+                                        292,
+                                        221.0
+                                    ],
+                                    "spans": [],
+                                    "index": 11
+                                },
+                                {
+                                    "bbox": [
+                                        110,
+                                        221.0,
+                                        292,
+                                        233.75
+                                    ],
+                                    "spans": [],
+                                    "index": 12
+                                },
+                                {
+                                    "bbox": [
+                                        110,
+                                        233.75,
+                                        292,
+                                        246.5
+                                    ],
+                                    "spans": [],
+                                    "index": 13
+                                },
+                                {
+                                    "bbox": [
+                                        110,
+                                        246.5,
+                                        292,
+                                        259.25
+                                    ],
+                                    "spans": [],
+                                    "index": 14
+                                },
+                                {
+                                    "bbox": [
+                                        110,
+                                        259.25,
+                                        292,
+                                        272.0
+                                    ],
+                                    "spans": [],
+                                    "index": 15
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 7.5
+                },
+                {
+                    "type": "image",
+                    "bbox": [
+                        324,
+                        94,
+                        503,
+                        228
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                324,
+                                94,
+                                503,
+                                228
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        324,
+                                        94,
+                                        503,
+                                        228
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                324,
+                                                94,
+                                                503,
+                                                228
+                                            ],
+                                            "score": 0.971,
+                                            "type": "image",
+                                            "image_path": "8cecf45d517b8c5b96207f2fc90806edfe43935c7cfa9552a0ceb80d0db26208.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 21,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        324,
+                                        94,
+                                        503,
+                                        106.18181818181819
+                                    ],
+                                    "spans": [],
+                                    "index": 16
+                                },
+                                {
+                                    "bbox": [
+                                        324,
+                                        106.18181818181819,
+                                        503,
+                                        118.36363636363637
+                                    ],
+                                    "spans": [],
+                                    "index": 17
+                                },
+                                {
+                                    "bbox": [
+                                        324,
+                                        118.36363636363637,
+                                        503,
+                                        130.54545454545456
+                                    ],
+                                    "spans": [],
+                                    "index": 18
+                                },
+                                {
+                                    "bbox": [
+                                        324,
+                                        130.54545454545456,
+                                        503,
+                                        142.72727272727275
+                                    ],
+                                    "spans": [],
+                                    "index": 19
+                                },
+                                {
+                                    "bbox": [
+                                        324,
+                                        142.72727272727275,
+                                        503,
+                                        154.90909090909093
+                                    ],
+                                    "spans": [],
+                                    "index": 20
+                                },
+                                {
+                                    "bbox": [
+                                        324,
+                                        154.90909090909093,
+                                        503,
+                                        167.09090909090912
+                                    ],
+                                    "spans": [],
+                                    "index": 21
+                                },
+                                {
+                                    "bbox": [
+                                        324,
+                                        167.09090909090912,
+                                        503,
+                                        179.2727272727273
+                                    ],
+                                    "spans": [],
+                                    "index": 22
+                                },
+                                {
+                                    "bbox": [
+                                        324,
+                                        179.2727272727273,
+                                        503,
+                                        191.4545454545455
+                                    ],
+                                    "spans": [],
+                                    "index": 23
+                                },
+                                {
+                                    "bbox": [
+                                        324,
+                                        191.4545454545455,
+                                        503,
+                                        203.63636363636368
+                                    ],
+                                    "spans": [],
+                                    "index": 24
+                                },
+                                {
+                                    "bbox": [
+                                        324,
+                                        203.63636363636368,
+                                        503,
+                                        215.81818181818187
+                                    ],
+                                    "spans": [],
+                                    "index": 25
+                                },
+                                {
+                                    "bbox": [
+                                        324,
+                                        215.81818181818187,
+                                        503,
+                                        228.00000000000006
+                                    ],
+                                    "spans": [],
+                                    "index": 26
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                320,
+                                230,
+                                505,
+                                270
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        319,
+                                        230,
+                                        505,
+                                        241
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                319,
+                                                230,
+                                                406,
+                                                241
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 5: Accuracy",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                406,
+                                                231,
+                                                420,
+                                                240
+                                            ],
+                                            "score": 0.66,
+                                            "content": "( \\% )",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                421,
+                                                230,
+                                                505,
+                                                241
+                                            ],
+                                            "score": 1.0,
+                                            "content": "of student networks",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 27
+                                },
+                                {
+                                    "bbox": [
+                                        320,
+                                        239,
+                                        506,
+                                        250
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                320,
+                                                239,
+                                                506,
+                                                250
+                                            ],
+                                            "score": 1.0,
+                                            "content": "(EfficientNet-b0 and ResNet-18) on ImageNet dis-",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 28
+                                },
+                                {
+                                    "bbox": [
+                                        319,
+                                        250,
+                                        505,
+                                        261
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                319,
+                                                250,
+                                                505,
+                                                261
+                                            ],
+                                            "score": 1.0,
+                                            "content": "tilled from wider MoCo-v2 pre-trained ResNet",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 29
+                                },
+                                {
+                                    "bbox": [
+                                        319,
+                                        259,
+                                        411,
+                                        271
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                319,
+                                                259,
+                                                411,
+                                                271
+                                            ],
+                                            "score": 1.0,
+                                            "content": "(ResNet-50/101/152×2).",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 30
+                                }
+                            ],
+                            "index": 28.5
+                        }
+                    ],
+                    "index": 24.75
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        300,
+                        505,
+                        366
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                301,
+                                505,
+                                312
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        301,
+                                        505,
+                                        312
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Different Teacher Networks. Figure 5 summarizes the accuracy of ResNet-18 and EfficientNet-B0",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                311,
+                                506,
+                                324
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        311,
+                                        506,
+                                        324
+                                    ],
+                                    "score": 1.0,
+                                    "content": "distilled from wider and deeper ResNet architectures. We see clear performance improvement as",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                322,
+                                506,
+                                335
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        322,
+                                        506,
+                                        335
+                                    ],
+                                    "score": 1.0,
+                                    "content": "depth and width of teacher network increase: compared to ResNet-50, deeper (ResNet-101) and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                333,
+                                506,
+                                346
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        333,
+                                        166,
+                                        346
+                                    ],
+                                    "score": 1.0,
+                                    "content": "wider (ResNet-",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        166,
+                                        334,
+                                        190,
+                                        344
+                                    ],
+                                    "score": 0.68,
+                                    "content": "{ 5 0 \\times 2 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        191,
+                                        333,
+                                        506,
+                                        346
+                                    ],
+                                    "score": 1.0,
+                                    "content": ") substantially improve the accuracy. However, further architectural enlargement",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                345,
+                                506,
+                                357
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        345,
+                                        506,
+                                        357
+                                    ],
+                                    "score": 1.0,
+                                    "content": "has relatively limited effects, and we suspect the accuracy might be limited by the student network",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                356,
+                                192,
+                                369
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        356,
+                                        192,
+                                        369
+                                    ],
+                                    "score": 1.0,
+                                    "content": "capacity in this case.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        }
+                    ],
+                    "index": 33.5,
+                    "bbox_fs": [
+                        105,
+                        301,
+                        506,
+                        369
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        372,
+                        505,
+                        460
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                372,
+                                505,
+                                384
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        372,
+                                        505,
+                                        384
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Different Teacher Pre-training Algorithms. In Table 3, we show the Top-1 accuracy of ResNet-18",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                383,
+                                506,
+                                395
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        383,
+                                        506,
+                                        395
+                                    ],
+                                    "score": 1.0,
+                                    "content": "distilled from ResNet-50 with different pre-training algorithms, i.e., MoCo-V1 (He et al., 2020),",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                394,
+                                506,
+                                406
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        394,
+                                        506,
+                                        406
+                                    ],
+                                    "score": 1.0,
+                                    "content": "MoCo-V2 (Chen et al., 2020c), SimCLR (Chen et al., 2020a), and SWAV (Caron et al., 2020)).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                405,
+                                505,
+                                418
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        405,
+                                        505,
+                                        418
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Notably, the aforementioned methods all unanimously adopt contrastive-based pre-training except",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                415,
+                                506,
+                                429
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        415,
+                                        506,
+                                        429
+                                    ],
+                                    "score": 1.0,
+                                    "content": "SWAV, which is based upon online clustering. We find that our SEED is agnostic to pre-training",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                427,
+                                506,
+                                440
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        427,
+                                        506,
+                                        440
+                                    ],
+                                    "score": 1.0,
+                                    "content": "approaches, making it easy to use any self-supervised models (including clustering-based approach",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                437,
+                                505,
+                                450
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        437,
+                                        505,
+                                        450
+                                    ],
+                                    "score": 1.0,
+                                    "content": "like SWAV) in self-supervised distillation. In addition, we observe that more training epochs for both",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                448,
+                                351,
+                                461
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        448,
+                                        137,
+                                        461
+                                    ],
+                                    "score": 1.0,
+                                    "content": "teacher",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        138,
+                                        450,
+                                        155,
+                                        459
+                                    ],
+                                    "score": 0.25,
+                                    "content": "S S L",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        155,
+                                        448,
+                                        351,
+                                        461
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and distillation epochs can bring beneficial gain.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        }
+                    ],
+                    "index": 40.5,
+                    "bbox_fs": [
+                        105,
+                        372,
+                        506,
+                        461
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        466,
+                        505,
+                        642
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                465,
+                                506,
+                                478
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        465,
+                                        506,
+                                        478
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Other Distillation Strategies. We explore several alternative distillation strategies. l2-Distance:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                476,
+                                506,
+                                489
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        476,
+                                        147,
+                                        489
+                                    ],
+                                    "score": 1.0,
+                                    "content": "where the",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        147,
+                                        478,
+                                        156,
+                                        487
+                                    ],
+                                    "score": 0.66,
+                                    "content": "l 2",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        157,
+                                        476,
+                                        506,
+                                        489
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-distance of teacher & student’s embeddings are minimized, motivated by Romero et al.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                488,
+                                506,
+                                500
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        488,
+                                        140,
+                                        500
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(2014).",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        140,
+                                        488,
+                                        151,
+                                        498
+                                    ],
+                                    "score": 0.67,
+                                    "content": "K",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        151,
+                                        488,
+                                        232,
+                                        500
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-Means: we exploit",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        233,
+                                        489,
+                                        241,
+                                        498
+                                    ],
+                                    "score": 0.71,
+                                    "content": "K",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        242,
+                                        488,
+                                        506,
+                                        500
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-Means clustering to assign a pseudo-label based on the teacher",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 47
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                498,
+                                505,
+                                512
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        498,
+                                        505,
+                                        512
+                                    ],
+                                    "score": 1.0,
+                                    "content": "network’s representation. Online Clustering: we continuously update the clustering centers during",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 48
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                510,
+                                505,
+                                522
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        510,
+                                        505,
+                                        522
+                                    ],
+                                    "score": 1.0,
+                                    "content": "distillation for pseudo-label generation. Binary Contrastive Loss: we adopt an Info-NCE alike loss",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 49
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                520,
+                                505,
+                                533
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        520,
+                                        505,
+                                        533
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for contrastive distillation (Tian et al., 2019a). We provide details for other strategies in A.4. Table 4",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 50
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                531,
+                                506,
+                                544
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        531,
+                                        506,
+                                        544
+                                    ],
+                                    "score": 1.0,
+                                    "content": "shows the results for each method on ResNet-18 (student) distilled from ResNet-50. From the results,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 51
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                543,
+                                505,
+                                555
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        543,
+                                        150,
+                                        555
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the simple",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        150,
+                                        543,
+                                        159,
+                                        553
+                                    ],
+                                    "score": 0.38,
+                                    "content": "l 2",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        160,
+                                        543,
+                                        505,
+                                        555
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-distance minimizing approach can achieve a decent accuracy, which demonstrates the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 52
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                554,
+                                505,
+                                566
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        554,
+                                        505,
+                                        566
+                                    ],
+                                    "score": 1.0,
+                                    "content": "effectiveness of applying the distillation idea to the self-supervised learning. Beyond that, we study",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 53
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                564,
+                                506,
+                                578
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        564,
+                                        506,
+                                        578
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the effect of the original SSL (MoCo-V2) supervision as supplementary loss to SEED and find it does",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 54
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                576,
+                                505,
+                                588
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        576,
+                                        505,
+                                        588
+                                    ],
+                                    "score": 1.0,
+                                    "content": "not bring additional benefits to distillation. We find close results from these two strategies (Top-1",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 55
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                585,
+                                506,
+                                599
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        585,
+                                        226,
+                                        599
+                                    ],
+                                    "score": 1.0,
+                                    "content": "linear Acc.), SEED achieves",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        226,
+                                        587,
+                                        254,
+                                        597
+                                    ],
+                                    "score": 0.88,
+                                    "content": "5 7 . 9 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        254,
+                                        585,
+                                        284,
+                                        599
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", while",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        284,
+                                        587,
+                                        363,
+                                        597
+                                    ],
+                                    "score": 0.63,
+                                    "content": "\\mathbf { S E E D + M o C o - V } 2",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        363,
+                                        585,
+                                        401,
+                                        599
+                                    ],
+                                    "score": 1.0,
+                                    "content": "achieves",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        401,
+                                        587,
+                                        429,
+                                        597
+                                    ],
+                                    "score": 0.92,
+                                    "content": "5 7 . 6 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        429,
+                                        585,
+                                        506,
+                                        599
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". This implies that",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 56
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                596,
+                                506,
+                                610
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        596,
+                                        332,
+                                        610
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the loss of SEED can to a large extent cover the original",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        332,
+                                        598,
+                                        349,
+                                        608
+                                    ],
+                                    "score": 0.3,
+                                    "content": "S S L",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        350,
+                                        596,
+                                        506,
+                                        610
+                                    ],
+                                    "score": 1.0,
+                                    "content": "loss, and it is not necessary to conduct",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 57
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                609,
+                                505,
+                                620
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        609,
+                                        505,
+                                        620
+                                    ],
+                                    "score": 1.0,
+                                    "content": "SSL any further during distillation. Meanwhile, our proposed SEED outperforms these alternatives",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 58
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                619,
+                                506,
+                                633
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        619,
+                                        506,
+                                        633
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with highest accuracy, which shows the superiority of aligning the student towards the teacher and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 59
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                631,
+                                265,
+                                643
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        631,
+                                        265,
+                                        643
+                                    ],
+                                    "score": 1.0,
+                                    "content": "contrasting with the irrelevant samples.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 60
+                        }
+                    ],
+                    "index": 52.5,
+                    "bbox_fs": [
+                        105,
+                        465,
+                        506,
+                        643
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        647,
+                        505,
+                        736
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                646,
+                                505,
+                                660
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        646,
+                                        505,
+                                        660
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Other Hyper-Parameters. Table 5 summarizes the distillation performances on multiple datasets",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 61
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                657,
+                                506,
+                                671
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        657,
+                                        219,
+                                        671
+                                    ],
+                                    "score": 1.0,
+                                    "content": "using different temperature",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        219,
+                                        657,
+                                        232,
+                                        668
+                                    ],
+                                    "score": 0.87,
+                                    "content": "\\tau ^ { T }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        233,
+                                        657,
+                                        445,
+                                        671
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". We observe a better performance when decreasing",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        445,
+                                        657,
+                                        459,
+                                        668
+                                    ],
+                                    "score": 0.88,
+                                    "content": "\\tau ^ { T }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        459,
+                                        657,
+                                        506,
+                                        671
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to 0.01 for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 62
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                669,
+                                505,
+                                681
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        669,
+                                        445,
+                                        681
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ImageNet-1k and CIFAR-10 dataset, and to 1e-3 for CIFAR-100 datasets. When",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        445,
+                                        671,
+                                        452,
+                                        679
+                                    ],
+                                    "score": 0.78,
+                                    "content": "\\tau",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        453,
+                                        669,
+                                        505,
+                                        681
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is large, the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 63
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                679,
+                                506,
+                                694
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        679,
+                                        259,
+                                        693
+                                    ],
+                                    "score": 1.0,
+                                    "content": "softmax-normalized similarity score of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        260,
+                                        680,
+                                        272,
+                                        694
+                                    ],
+                                    "score": 0.9,
+                                    "content": "p _ { j } ^ { T }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        272,
+                                        679,
+                                        307,
+                                        693
+                                    ],
+                                    "score": 1.0,
+                                    "content": "between",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        308,
+                                        680,
+                                        320,
+                                        693
+                                    ],
+                                    "score": 0.9,
+                                    "content": "{ \\mathbf z } _ { i } ^ { T }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        320,
+                                        679,
+                                        371,
+                                        693
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and instance",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        371,
+                                        681,
+                                        383,
+                                        693
+                                    ],
+                                    "score": 0.89,
+                                    "content": "{ \\bf d } _ { j }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        383,
+                                        679,
+                                        433,
+                                        693
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in the queue",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        433,
+                                        680,
+                                        450,
+                                        691
+                                    ],
+                                    "score": 0.89,
+                                    "content": "\\mathbf { D } ^ { + }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        450,
+                                        679,
+                                        506,
+                                        693
+                                    ],
+                                    "score": 1.0,
+                                    "content": "also becomes",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 64
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                691,
+                                506,
+                                703
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        691,
+                                        506,
+                                        703
+                                    ],
+                                    "score": 1.0,
+                                    "content": "large, which means the student’s feature should be less discriminative with the features of other",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 65
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                702,
+                                506,
+                                714
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        702,
+                                        227,
+                                        714
+                                    ],
+                                    "score": 1.0,
+                                    "content": "images to some extent. When",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        228,
+                                        702,
+                                        240,
+                                        712
+                                    ],
+                                    "score": 0.88,
+                                    "content": "\\tau ^ { T }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        240,
+                                        702,
+                                        251,
+                                        714
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        251,
+                                        704,
+                                        257,
+                                        712
+                                    ],
+                                    "score": 0.43,
+                                    "content": "_ 0",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        257,
+                                        702,
+                                        506,
+                                        714
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", the teacher model will generate a one-hot vector, which only",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 66
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                713,
+                                506,
+                                726
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        713,
+                                        130,
+                                        726
+                                    ],
+                                    "score": 1.0,
+                                    "content": "treats",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        130,
+                                        713,
+                                        143,
+                                        725
+                                    ],
+                                    "score": 0.9,
+                                    "content": "{ \\mathbf z } _ { i } ^ { T }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        143,
+                                        713,
+                                        444,
+                                        726
+                                    ],
+                                    "score": 1.0,
+                                    "content": "as a positive instance and all others in the queue as negative. Thus, the best",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        444,
+                                        715,
+                                        451,
+                                        723
+                                    ],
+                                    "score": 0.76,
+                                    "content": "\\tau",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        452,
+                                        713,
+                                        506,
+                                        726
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is a trade-off",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 67
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                724,
+                                504,
+                                736
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        724,
+                                        504,
+                                        736
+                                    ],
+                                    "score": 1.0,
+                                    "content": "depending on the data distribution. We further compare effect of different hyper-parameters in A.8.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 68
+                        }
+                    ],
+                    "index": 64.5,
+                    "bbox_fs": [
+                        105,
+                        646,
+                        506,
+                        736
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "table",
+                    "bbox": [
+                        106,
+                        127,
+                        282,
+                        217
+                    ],
+                    "blocks": [
+                        {
+                            "type": "table_caption",
+                            "bbox": [
+                                106,
+                                86,
+                                291,
+                                126
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        85,
+                                        292,
+                                        97
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                85,
+                                                292,
+                                                97
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Table 4: Top-1/5 accuracy of linear classifica-",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 0
+                                },
+                                {
+                                    "bbox": [
+                                        107,
+                                        97,
+                                        291,
+                                        106
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                107,
+                                                97,
+                                                291,
+                                                106
+                                            ],
+                                            "score": 1.0,
+                                            "content": "tion results on ImageNet using different distillation",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        106,
+                                        290,
+                                        117
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                106,
+                                                290,
+                                                117
+                                            ],
+                                            "score": 1.0,
+                                            "content": "strategies on ResNet-18 (student) and ResNet-50",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 2
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        117,
+                                        191,
+                                        127
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                117,
+                                                191,
+                                                127
+                                            ],
+                                            "score": 1.0,
+                                            "content": "(teacher) architectures.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 3
+                                }
+                            ],
+                            "index": 1.5
+                        },
+                        {
+                            "type": "table_body",
+                            "bbox": [
+                                106,
+                                127,
+                                282,
+                                217
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        127,
+                                        282,
+                                        217
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                127,
+                                                282,
+                                                217
+                                            ],
+                                            "score": 0.978,
+                                            "html": "<table><tr><td>Method</td><td colspan=\"2\">Top-1 Acc. Top-5 Acc.</td></tr><tr><td>l2-Distance</td><td>55.3</td><td>80.3</td></tr><tr><td>K-Means</td><td>51.0</td><td>75.8</td></tr><tr><td>Online Clustering</td><td>56.4</td><td>81.2</td></tr><tr><td>Binary Contr. Loss</td><td>57.4</td><td>81.5</td></tr><tr><td>SEED +MoCo-V2</td><td>57.6</td><td>81.8</td></tr><tr><td>SEED</td><td>57.9</td><td>82.0</td></tr></table>",
+                                            "type": "table",
+                                            "image_path": "111d557f7135948444d3e5bb803f101bbeaabe2c0cfcc43a81faac6da770ecd7.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 6.5,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        127,
+                                        282,
+                                        142.0
+                                    ],
+                                    "spans": [],
+                                    "index": 4
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        142.0,
+                                        282,
+                                        157.0
+                                    ],
+                                    "spans": [],
+                                    "index": 5
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        157.0,
+                                        282,
+                                        172.0
+                                    ],
+                                    "spans": [],
+                                    "index": 6
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        172.0,
+                                        282,
+                                        187.0
+                                    ],
+                                    "spans": [],
+                                    "index": 7
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        187.0,
+                                        282,
+                                        202.0
+                                    ],
+                                    "spans": [],
+                                    "index": 8
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        202.0,
+                                        282,
+                                        217.0
+                                    ],
+                                    "spans": [],
+                                    "index": 9
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 4.0
+                },
+                {
+                    "type": "table",
+                    "bbox": [
+                        304,
+                        117,
+                        498,
+                        210
+                    ],
+                    "blocks": [
+                        {
+                            "type": "table_caption",
+                            "bbox": [
+                                303,
+                                92,
+                                504,
+                                114
+                            ],
+                            "group_id": 1,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        303,
+                                        92,
+                                        505,
+                                        104
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                303,
+                                                92,
+                                                372,
+                                                104
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Table 5: Effect of",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                372,
+                                                92,
+                                                384,
+                                                102
+                                            ],
+                                            "score": 0.89,
+                                            "content": "\\tau ^ { T }",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                385,
+                                                92,
+                                                505,
+                                                104
+                                            ],
+                                            "score": 1.0,
+                                            "content": "for the distillation of ResNet-18",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 10
+                                },
+                                {
+                                    "bbox": [
+                                        304,
+                                        102,
+                                        492,
+                                        115
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                304,
+                                                102,
+                                                492,
+                                                115
+                                            ],
+                                            "score": 1.0,
+                                            "content": "(student), ResNet-50 (teacher) on multiple datasets.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 11
+                                }
+                            ],
+                            "index": 10.5
+                        },
+                        {
+                            "type": "table_body",
+                            "bbox": [
+                                304,
+                                117,
+                                498,
+                                210
+                            ],
+                            "group_id": 1,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        304,
+                                        117,
+                                        498,
+                                        210
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                304,
+                                                117,
+                                                498,
+                                                210
+                                            ],
+                                            "score": 0.981,
+                                            "html": "<table><tr><td rowspan=\"2\">YT</td><td>ImageNet</td><td>CIFAR-10</td><td>CIFAR-100</td></tr><tr><td>Top-1 Top-5</td><td>Top-1</td><td>Top-1</td></tr><tr><td>0.3</td><td>54.8 80.0</td><td>78.7</td><td>46.6</td></tr><tr><td>0.1</td><td>54.9 80.1</td><td>83.0</td><td>50.1</td></tr><tr><td>0.05</td><td>56.5 81.3</td><td>84.4</td><td>56.2</td></tr><tr><td>0.01</td><td>57.9 82.0</td><td>87.5 </td><td>60.6</td></tr><tr><td>1e-3</td><td>57.6 81.8</td><td>86.9</td><td>60.8</td></tr></table>",
+                                            "type": "table",
+                                            "image_path": "0ce07da924f5f7e589e1097898757fd0f72682d6d444c981019b465d74535daa.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 15,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        304,
+                                        117,
+                                        498,
+                                        130.28571428571428
+                                    ],
+                                    "spans": [],
+                                    "index": 12
+                                },
+                                {
+                                    "bbox": [
+                                        304,
+                                        130.28571428571428,
+                                        498,
+                                        143.57142857142856
+                                    ],
+                                    "spans": [],
+                                    "index": 13
+                                },
+                                {
+                                    "bbox": [
+                                        304,
+                                        143.57142857142856,
+                                        498,
+                                        156.85714285714283
+                                    ],
+                                    "spans": [],
+                                    "index": 14
+                                },
+                                {
+                                    "bbox": [
+                                        304,
+                                        156.85714285714283,
+                                        498,
+                                        170.1428571428571
+                                    ],
+                                    "spans": [],
+                                    "index": 15
+                                },
+                                {
+                                    "bbox": [
+                                        304,
+                                        170.1428571428571,
+                                        498,
+                                        183.4285714285714
+                                    ],
+                                    "spans": [],
+                                    "index": 16
+                                },
+                                {
+                                    "bbox": [
+                                        304,
+                                        183.4285714285714,
+                                        498,
+                                        196.71428571428567
+                                    ],
+                                    "spans": [],
+                                    "index": 17
+                                },
+                                {
+                                    "bbox": [
+                                        304,
+                                        196.71428571428567,
+                                        498,
+                                        209.99999999999994
+                                    ],
+                                    "spans": [],
+                                    "index": 18
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 12.75
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        239,
+                        201,
+                        252
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                236,
+                                203,
+                                255
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        236,
+                                        203,
+                                        255
+                                    ],
+                                    "score": 1.0,
+                                    "content": "5 CONCLUSIONS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        }
+                    ],
+                    "index": 19
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        264,
+                        505,
+                        342
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                263,
+                                506,
+                                277
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        263,
+                                        506,
+                                        277
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Self-Supervised Learning is acknowledged for its remarkable ability in learning from unlabeled, and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                276,
+                                506,
+                                288
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        276,
+                                        506,
+                                        288
+                                    ],
+                                    "score": 1.0,
+                                    "content": "large scale data. However, a critical impedance for the SSL pre-training on smaller architecture comes",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                285,
+                                506,
+                                300
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        285,
+                                        506,
+                                        300
+                                    ],
+                                    "score": 1.0,
+                                    "content": "from its low capacity of discriminating enormous number of instances. Instead of directly learning",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                298,
+                                505,
+                                310
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        298,
+                                        505,
+                                        310
+                                    ],
+                                    "score": 1.0,
+                                    "content": "from unlabeled data, we proposed SEED as a novel self-supervised learning paradigm, which learns",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                308,
+                                505,
+                                321
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        308,
+                                        505,
+                                        321
+                                    ],
+                                    "score": 1.0,
+                                    "content": "representation by self-supervised distillation from a bigger SSL pre-trained model. We show in",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                319,
+                                506,
+                                331
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        319,
+                                        506,
+                                        331
+                                    ],
+                                    "score": 1.0,
+                                    "content": "extensive experiments that SEED effectively addresses the weakness of self-supervised learning for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                331,
+                                490,
+                                342
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        331,
+                                        490,
+                                        342
+                                    ],
+                                    "score": 1.0,
+                                    "content": "small models and achieves state-of-the-art results on various benchmarks of small architectures.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        }
+                    ],
+                    "index": 23
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        360,
+                        175,
+                        372
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                360,
+                                176,
+                                373
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        360,
+                                        176,
+                                        373
+                                    ],
+                                    "score": 1.0,
+                                    "content": "REFERENCES",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        }
+                    ],
+                    "index": 27
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        378,
+                        504,
+                        412
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                378,
+                                505,
+                                392
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        378,
+                                        505,
+                                        392
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Soroush Abbasi Koohpayegani, Ajinkya Tejankar, and Hamed Pirsiavash. Compress: Self-supervised",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                390,
+                                506,
+                                402
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        390,
+                                        506,
+                                        402
+                                    ],
+                                    "score": 1.0,
+                                    "content": "learning by compressing representations. Advances in Neural Information Processing Systems, 33,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                114,
+                                399,
+                                142,
+                                414
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        114,
+                                        399,
+                                        142,
+                                        414
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2020.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        }
+                    ],
+                    "index": 29
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        421,
+                        506,
+                        455
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                421,
+                                505,
+                                434
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        421,
+                                        505,
+                                        434
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Sungsoo Ahn, Shell Xu Hu, Andreas Damianou, Neil D Lawrence, and Zhenwen Dai. Variational",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                432,
+                                505,
+                                445
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        432,
+                                        505,
+                                        445
+                                    ],
+                                    "score": 1.0,
+                                    "content": "information distillation for knowledge transfer. In Proceedings of the IEEE Conference on",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                442,
+                                378,
+                                456
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        442,
+                                        378,
+                                        456
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Computer Vision and Pattern Recognition, pp. 9163–9171, 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        }
+                    ],
+                    "index": 32
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        464,
+                        506,
+                        497
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                462,
+                                505,
+                                478
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        462,
+                                        505,
+                                        478
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Cristian Bucilua, Rich Caruana, and Alexandru Niculescu-Mizil. Model compression. In ˇ Proceedings",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                474,
+                                506,
+                                489
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        474,
+                                        506,
+                                        489
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of the 12th ACM SIGKDD international conference on Knowledge discovery and data mining, pp.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                486,
+                                181,
+                                497
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        486,
+                                        181,
+                                        497
+                                    ],
+                                    "score": 1.0,
+                                    "content": "535–541, 2006.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        }
+                    ],
+                    "index": 35
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        506,
+                        505,
+                        540
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                505,
+                                506,
+                                520
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        505,
+                                        506,
+                                        520
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Mathilde Caron, Piotr Bojanowski, Armand Joulin, and Matthijs Douze. Deep clustering for unsuper-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                518,
+                                505,
+                                530
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        518,
+                                        505,
+                                        530
+                                    ],
+                                    "score": 1.0,
+                                    "content": "vised learning of visual features. In Proceedings of the European Conference on Computer Vision",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                529,
+                                233,
+                                541
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        529,
+                                        233,
+                                        541
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(ECCV), pp. 132–149, 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        }
+                    ],
+                    "index": 38
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        549,
+                        505,
+                        583
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                549,
+                                506,
+                                561
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        549,
+                                        506,
+                                        561
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Mathilde Caron, Ishan Misra, Julien Mairal, Priya Goyal, Piotr Bojanowski, and Armand Joulin.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                561,
+                                505,
+                                573
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        561,
+                                        505,
+                                        573
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Unsupervised learning of visual features by contrasting cluster assignments. arXiv preprint",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                572,
+                                219,
+                                583
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        572,
+                                        219,
+                                        583
+                                    ],
+                                    "score": 1.0,
+                                    "content": "arXiv:2006.09882, 2020.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        }
+                    ],
+                    "index": 41
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        591,
+                        506,
+                        626
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                591,
+                                506,
+                                605
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        591,
+                                        506,
+                                        605
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Guobin Chen, Wongun Choi, Xiang Yu, Tony Han, and Manmohan Chandraker. Learning efficient",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                602,
+                                506,
+                                617
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        602,
+                                        506,
+                                        617
+                                    ],
+                                    "score": 1.0,
+                                    "content": "object detection models with knowledge distillation. In Advances in Neural Information Processing",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                614,
+                                232,
+                                627
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        614,
+                                        232,
+                                        627
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Systems, pp. 742–751, 2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        }
+                    ],
+                    "index": 44
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        635,
+                        505,
+                        658
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                635,
+                                506,
+                                648
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        635,
+                                        506,
+                                        648
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Ting Chen, Simon Kornblith, Mohammad Norouzi, and Geoffrey Hinton. A simple framework for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                646,
+                                468,
+                                658
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        646,
+                                        468,
+                                        658
+                                    ],
+                                    "score": 1.0,
+                                    "content": "contrastive learning of visual representations. arXiv preprint arXiv:2002.05709, 2020a.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 47
+                        }
+                    ],
+                    "index": 46.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        666,
+                        505,
+                        700
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                665,
+                                505,
+                                680
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        665,
+                                        505,
+                                        680
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Ting Chen, Simon Kornblith, Kevin Swersky, Mohammad Norouzi, and Geoffrey Hinton. Big",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 48
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                678,
+                                506,
+                                690
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        678,
+                                        506,
+                                        690
+                                    ],
+                                    "score": 1.0,
+                                    "content": "self-supervised models are strong semi-supervised learners. arXiv preprint arXiv:2006.10029,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 49
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                687,
+                                148,
+                                701
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        687,
+                                        148,
+                                        701
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2020b.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 50
+                        }
+                    ],
+                    "index": 49
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        709,
+                        504,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                708,
+                                505,
+                                722
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        708,
+                                        505,
+                                        722
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Xinlei Chen, Haoqi Fan, Ross Girshick, and Kaiming He. Improved baselines with momentum",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 51
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                720,
+                                368,
+                                733
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        720,
+                                        368,
+                                        733
+                                    ],
+                                    "score": 1.0,
+                                    "content": "contrastive learning. arXiv preprint arXiv:2003.04297, 2020c.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 52
+                        }
+                    ],
+                    "index": 51.5
+                }
+            ],
+            "page_idx": 8,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        108,
+                        27,
+                        292,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                26,
+                                293,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        26,
+                                        293,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Published as a conference paper at ICLR 2021",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        302,
+                        751,
+                        308,
+                        759
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                302,
+                                751,
+                                309,
+                                762
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        302,
+                                        751,
+                                        309,
+                                        762
+                                    ],
+                                    "score": 1.0,
+                                    "content": "9",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "table",
+                    "bbox": [
+                        106,
+                        127,
+                        282,
+                        217
+                    ],
+                    "blocks": [
+                        {
+                            "type": "table_caption",
+                            "bbox": [
+                                106,
+                                86,
+                                291,
+                                126
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        85,
+                                        292,
+                                        97
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                85,
+                                                292,
+                                                97
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Table 4: Top-1/5 accuracy of linear classifica-",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 0
+                                },
+                                {
+                                    "bbox": [
+                                        107,
+                                        97,
+                                        291,
+                                        106
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                107,
+                                                97,
+                                                291,
+                                                106
+                                            ],
+                                            "score": 1.0,
+                                            "content": "tion results on ImageNet using different distillation",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        106,
+                                        290,
+                                        117
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                106,
+                                                290,
+                                                117
+                                            ],
+                                            "score": 1.0,
+                                            "content": "strategies on ResNet-18 (student) and ResNet-50",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 2
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        117,
+                                        191,
+                                        127
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                117,
+                                                191,
+                                                127
+                                            ],
+                                            "score": 1.0,
+                                            "content": "(teacher) architectures.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 3
+                                }
+                            ],
+                            "index": 1.5
+                        },
+                        {
+                            "type": "table_body",
+                            "bbox": [
+                                106,
+                                127,
+                                282,
+                                217
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        127,
+                                        282,
+                                        217
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                127,
+                                                282,
+                                                217
+                                            ],
+                                            "score": 0.978,
+                                            "html": "<table><tr><td>Method</td><td colspan=\"2\">Top-1 Acc. Top-5 Acc.</td></tr><tr><td>l2-Distance</td><td>55.3</td><td>80.3</td></tr><tr><td>K-Means</td><td>51.0</td><td>75.8</td></tr><tr><td>Online Clustering</td><td>56.4</td><td>81.2</td></tr><tr><td>Binary Contr. Loss</td><td>57.4</td><td>81.5</td></tr><tr><td>SEED +MoCo-V2</td><td>57.6</td><td>81.8</td></tr><tr><td>SEED</td><td>57.9</td><td>82.0</td></tr></table>",
+                                            "type": "table",
+                                            "image_path": "111d557f7135948444d3e5bb803f101bbeaabe2c0cfcc43a81faac6da770ecd7.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 6.5,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        127,
+                                        282,
+                                        142.0
+                                    ],
+                                    "spans": [],
+                                    "index": 4
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        142.0,
+                                        282,
+                                        157.0
+                                    ],
+                                    "spans": [],
+                                    "index": 5
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        157.0,
+                                        282,
+                                        172.0
+                                    ],
+                                    "spans": [],
+                                    "index": 6
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        172.0,
+                                        282,
+                                        187.0
+                                    ],
+                                    "spans": [],
+                                    "index": 7
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        187.0,
+                                        282,
+                                        202.0
+                                    ],
+                                    "spans": [],
+                                    "index": 8
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        202.0,
+                                        282,
+                                        217.0
+                                    ],
+                                    "spans": [],
+                                    "index": 9
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 4.0
+                },
+                {
+                    "type": "table",
+                    "bbox": [
+                        304,
+                        117,
+                        498,
+                        210
+                    ],
+                    "blocks": [
+                        {
+                            "type": "table_caption",
+                            "bbox": [
+                                303,
+                                92,
+                                504,
+                                114
+                            ],
+                            "group_id": 1,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        303,
+                                        92,
+                                        505,
+                                        104
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                303,
+                                                92,
+                                                372,
+                                                104
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Table 5: Effect of",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                372,
+                                                92,
+                                                384,
+                                                102
+                                            ],
+                                            "score": 0.89,
+                                            "content": "\\tau ^ { T }",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                385,
+                                                92,
+                                                505,
+                                                104
+                                            ],
+                                            "score": 1.0,
+                                            "content": "for the distillation of ResNet-18",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 10
+                                },
+                                {
+                                    "bbox": [
+                                        304,
+                                        102,
+                                        492,
+                                        115
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                304,
+                                                102,
+                                                492,
+                                                115
+                                            ],
+                                            "score": 1.0,
+                                            "content": "(student), ResNet-50 (teacher) on multiple datasets.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 11
+                                }
+                            ],
+                            "index": 10.5
+                        },
+                        {
+                            "type": "table_body",
+                            "bbox": [
+                                304,
+                                117,
+                                498,
+                                210
+                            ],
+                            "group_id": 1,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        304,
+                                        117,
+                                        498,
+                                        210
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                304,
+                                                117,
+                                                498,
+                                                210
+                                            ],
+                                            "score": 0.981,
+                                            "html": "<table><tr><td rowspan=\"2\">YT</td><td>ImageNet</td><td>CIFAR-10</td><td>CIFAR-100</td></tr><tr><td>Top-1 Top-5</td><td>Top-1</td><td>Top-1</td></tr><tr><td>0.3</td><td>54.8 80.0</td><td>78.7</td><td>46.6</td></tr><tr><td>0.1</td><td>54.9 80.1</td><td>83.0</td><td>50.1</td></tr><tr><td>0.05</td><td>56.5 81.3</td><td>84.4</td><td>56.2</td></tr><tr><td>0.01</td><td>57.9 82.0</td><td>87.5 </td><td>60.6</td></tr><tr><td>1e-3</td><td>57.6 81.8</td><td>86.9</td><td>60.8</td></tr></table>",
+                                            "type": "table",
+                                            "image_path": "0ce07da924f5f7e589e1097898757fd0f72682d6d444c981019b465d74535daa.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 15,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        304,
+                                        117,
+                                        498,
+                                        130.28571428571428
+                                    ],
+                                    "spans": [],
+                                    "index": 12
+                                },
+                                {
+                                    "bbox": [
+                                        304,
+                                        130.28571428571428,
+                                        498,
+                                        143.57142857142856
+                                    ],
+                                    "spans": [],
+                                    "index": 13
+                                },
+                                {
+                                    "bbox": [
+                                        304,
+                                        143.57142857142856,
+                                        498,
+                                        156.85714285714283
+                                    ],
+                                    "spans": [],
+                                    "index": 14
+                                },
+                                {
+                                    "bbox": [
+                                        304,
+                                        156.85714285714283,
+                                        498,
+                                        170.1428571428571
+                                    ],
+                                    "spans": [],
+                                    "index": 15
+                                },
+                                {
+                                    "bbox": [
+                                        304,
+                                        170.1428571428571,
+                                        498,
+                                        183.4285714285714
+                                    ],
+                                    "spans": [],
+                                    "index": 16
+                                },
+                                {
+                                    "bbox": [
+                                        304,
+                                        183.4285714285714,
+                                        498,
+                                        196.71428571428567
+                                    ],
+                                    "spans": [],
+                                    "index": 17
+                                },
+                                {
+                                    "bbox": [
+                                        304,
+                                        196.71428571428567,
+                                        498,
+                                        209.99999999999994
+                                    ],
+                                    "spans": [],
+                                    "index": 18
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 12.75
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        239,
+                        201,
+                        252
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                236,
+                                203,
+                                255
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        236,
+                                        203,
+                                        255
+                                    ],
+                                    "score": 1.0,
+                                    "content": "5 CONCLUSIONS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        }
+                    ],
+                    "index": 19
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        264,
+                        505,
+                        342
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                263,
+                                506,
+                                277
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        263,
+                                        506,
+                                        277
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Self-Supervised Learning is acknowledged for its remarkable ability in learning from unlabeled, and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                276,
+                                506,
+                                288
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        276,
+                                        506,
+                                        288
+                                    ],
+                                    "score": 1.0,
+                                    "content": "large scale data. However, a critical impedance for the SSL pre-training on smaller architecture comes",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                285,
+                                506,
+                                300
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        285,
+                                        506,
+                                        300
+                                    ],
+                                    "score": 1.0,
+                                    "content": "from its low capacity of discriminating enormous number of instances. Instead of directly learning",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                298,
+                                505,
+                                310
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        298,
+                                        505,
+                                        310
+                                    ],
+                                    "score": 1.0,
+                                    "content": "from unlabeled data, we proposed SEED as a novel self-supervised learning paradigm, which learns",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                308,
+                                505,
+                                321
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        308,
+                                        505,
+                                        321
+                                    ],
+                                    "score": 1.0,
+                                    "content": "representation by self-supervised distillation from a bigger SSL pre-trained model. We show in",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                319,
+                                506,
+                                331
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        319,
+                                        506,
+                                        331
+                                    ],
+                                    "score": 1.0,
+                                    "content": "extensive experiments that SEED effectively addresses the weakness of self-supervised learning for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                331,
+                                490,
+                                342
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        331,
+                                        490,
+                                        342
+                                    ],
+                                    "score": 1.0,
+                                    "content": "small models and achieves state-of-the-art results on various benchmarks of small architectures.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        }
+                    ],
+                    "index": 23,
+                    "bbox_fs": [
+                        105,
+                        263,
+                        506,
+                        342
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        360,
+                        175,
+                        372
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                360,
+                                176,
+                                373
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        360,
+                                        176,
+                                        373
+                                    ],
+                                    "score": 1.0,
+                                    "content": "REFERENCES",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        }
+                    ],
+                    "index": 27
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        378,
+                        504,
+                        412
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                378,
+                                505,
+                                392
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        378,
+                                        505,
+                                        392
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Soroush Abbasi Koohpayegani, Ajinkya Tejankar, and Hamed Pirsiavash. Compress: Self-supervised",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                390,
+                                506,
+                                402
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        390,
+                                        506,
+                                        402
+                                    ],
+                                    "score": 1.0,
+                                    "content": "learning by compressing representations. Advances in Neural Information Processing Systems, 33,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                114,
+                                399,
+                                142,
+                                414
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        114,
+                                        399,
+                                        142,
+                                        414
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2020.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        }
+                    ],
+                    "index": 29,
+                    "bbox_fs": [
+                        106,
+                        378,
+                        506,
+                        414
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        421,
+                        506,
+                        455
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                421,
+                                505,
+                                434
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        421,
+                                        505,
+                                        434
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Sungsoo Ahn, Shell Xu Hu, Andreas Damianou, Neil D Lawrence, and Zhenwen Dai. Variational",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                432,
+                                505,
+                                445
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        432,
+                                        505,
+                                        445
+                                    ],
+                                    "score": 1.0,
+                                    "content": "information distillation for knowledge transfer. In Proceedings of the IEEE Conference on",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                442,
+                                378,
+                                456
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        442,
+                                        378,
+                                        456
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Computer Vision and Pattern Recognition, pp. 9163–9171, 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        }
+                    ],
+                    "index": 32,
+                    "bbox_fs": [
+                        106,
+                        421,
+                        505,
+                        456
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        464,
+                        506,
+                        497
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                462,
+                                505,
+                                478
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        462,
+                                        505,
+                                        478
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Cristian Bucilua, Rich Caruana, and Alexandru Niculescu-Mizil. Model compression. In ˇ Proceedings",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                474,
+                                506,
+                                489
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        474,
+                                        506,
+                                        489
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of the 12th ACM SIGKDD international conference on Knowledge discovery and data mining, pp.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                486,
+                                181,
+                                497
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        486,
+                                        181,
+                                        497
+                                    ],
+                                    "score": 1.0,
+                                    "content": "535–541, 2006.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        }
+                    ],
+                    "index": 35,
+                    "bbox_fs": [
+                        105,
+                        462,
+                        506,
+                        497
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        506,
+                        505,
+                        540
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                505,
+                                506,
+                                520
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        505,
+                                        506,
+                                        520
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Mathilde Caron, Piotr Bojanowski, Armand Joulin, and Matthijs Douze. Deep clustering for unsuper-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                518,
+                                505,
+                                530
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        518,
+                                        505,
+                                        530
+                                    ],
+                                    "score": 1.0,
+                                    "content": "vised learning of visual features. In Proceedings of the European Conference on Computer Vision",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                529,
+                                233,
+                                541
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        529,
+                                        233,
+                                        541
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(ECCV), pp. 132–149, 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        }
+                    ],
+                    "index": 38,
+                    "bbox_fs": [
+                        105,
+                        505,
+                        506,
+                        541
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        549,
+                        505,
+                        583
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                549,
+                                506,
+                                561
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        549,
+                                        506,
+                                        561
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Mathilde Caron, Ishan Misra, Julien Mairal, Priya Goyal, Piotr Bojanowski, and Armand Joulin.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                561,
+                                505,
+                                573
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        561,
+                                        505,
+                                        573
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Unsupervised learning of visual features by contrasting cluster assignments. arXiv preprint",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                572,
+                                219,
+                                583
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        572,
+                                        219,
+                                        583
+                                    ],
+                                    "score": 1.0,
+                                    "content": "arXiv:2006.09882, 2020.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        }
+                    ],
+                    "index": 41,
+                    "bbox_fs": [
+                        106,
+                        549,
+                        506,
+                        583
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        591,
+                        506,
+                        626
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                591,
+                                506,
+                                605
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        591,
+                                        506,
+                                        605
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Guobin Chen, Wongun Choi, Xiang Yu, Tony Han, and Manmohan Chandraker. Learning efficient",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                602,
+                                506,
+                                617
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        602,
+                                        506,
+                                        617
+                                    ],
+                                    "score": 1.0,
+                                    "content": "object detection models with knowledge distillation. In Advances in Neural Information Processing",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                614,
+                                232,
+                                627
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        614,
+                                        232,
+                                        627
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Systems, pp. 742–751, 2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        }
+                    ],
+                    "index": 44,
+                    "bbox_fs": [
+                        106,
+                        591,
+                        506,
+                        627
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        635,
+                        505,
+                        658
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                635,
+                                506,
+                                648
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        635,
+                                        506,
+                                        648
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Ting Chen, Simon Kornblith, Mohammad Norouzi, and Geoffrey Hinton. A simple framework for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                646,
+                                468,
+                                658
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        646,
+                                        468,
+                                        658
+                                    ],
+                                    "score": 1.0,
+                                    "content": "contrastive learning of visual representations. arXiv preprint arXiv:2002.05709, 2020a.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 47
+                        }
+                    ],
+                    "index": 46.5,
+                    "bbox_fs": [
+                        106,
+                        635,
+                        506,
+                        658
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        666,
+                        505,
+                        700
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                665,
+                                505,
+                                680
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        665,
+                                        505,
+                                        680
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Ting Chen, Simon Kornblith, Kevin Swersky, Mohammad Norouzi, and Geoffrey Hinton. Big",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 48
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                678,
+                                506,
+                                690
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        678,
+                                        506,
+                                        690
+                                    ],
+                                    "score": 1.0,
+                                    "content": "self-supervised models are strong semi-supervised learners. arXiv preprint arXiv:2006.10029,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 49
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                687,
+                                148,
+                                701
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        687,
+                                        148,
+                                        701
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2020b.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 50
+                        }
+                    ],
+                    "index": 49,
+                    "bbox_fs": [
+                        105,
+                        665,
+                        506,
+                        701
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        709,
+                        504,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                708,
+                                505,
+                                722
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        708,
+                                        505,
+                                        722
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Xinlei Chen, Haoqi Fan, Ross Girshick, and Kaiming He. Improved baselines with momentum",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 51
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                720,
+                                368,
+                                733
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        720,
+                                        368,
+                                        733
+                                    ],
+                                    "score": 1.0,
+                                    "content": "contrastive learning. arXiv preprint arXiv:2003.04297, 2020c.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 52
+                        }
+                    ],
+                    "index": 51.5,
+                    "bbox_fs": [
+                        105,
+                        708,
+                        505,
+                        733
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        104,
+                        82,
+                        505,
+                        106
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                81,
+                                506,
+                                96
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        81,
+                                        506,
+                                        96
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Jang Hyun Cho and Bharath Hariharan. On the efficacy of knowledge distillation. In Proceedings of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                93,
+                                433,
+                                106
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        93,
+                                        433,
+                                        106
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the IEEE International Conference on Computer Vision, pp. 4794–4802, 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        }
+                    ],
+                    "index": 0.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        112,
+                        504,
+                        146
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                112,
+                                505,
+                                126
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        112,
+                                        505,
+                                        126
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Jia Deng, Wei Dong, Richard Socher, Li-Jia Li, Kai Li, and Li Fei-Fei. Imagenet: A large-scale",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                123,
+                                506,
+                                137
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        123,
+                                        506,
+                                        137
+                                    ],
+                                    "score": 1.0,
+                                    "content": "hierarchical image database. In 2009 IEEE conference on computer vision and pattern recognition,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                134,
+                                218,
+                                146
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        134,
+                                        218,
+                                        146
+                                    ],
+                                    "score": 1.0,
+                                    "content": "pp. 248–255. Ieee, 2009.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        }
+                    ],
+                    "index": 3
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        153,
+                        503,
+                        188
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                154,
+                                505,
+                                166
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        154,
+                                        505,
+                                        166
+                                    ],
+                                    "score": 1.0,
+                                    "content": "M. Everingham, L. Van Gool, C. K. I. Williams, J. Winn, and A. Zisserman. The",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                164,
+                                505,
+                                178
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        164,
+                                        505,
+                                        178
+                                    ],
+                                    "score": 1.0,
+                                    "content": "PASCAL Visual Object Classes Challenge 2007 (VOC2007) Results. http://www.pascal-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                176,
+                                362,
+                                189
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        176,
+                                        362,
+                                        189
+                                    ],
+                                    "score": 1.0,
+                                    "content": "network.org/challenges/VOC/voc2007/workshop/index.html.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        }
+                    ],
+                    "index": 6
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        194,
+                        505,
+                        229
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                194,
+                                505,
+                                208
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        194,
+                                        505,
+                                        208
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Zeyu Feng, Chang Xu, and Dacheng Tao. Self-supervised representation learning by rotation feature",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                206,
+                                507,
+                                219
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        206,
+                                        507,
+                                        219
+                                    ],
+                                    "score": 1.0,
+                                    "content": "decoupling. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                217,
+                                216,
+                                228
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        217,
+                                        216,
+                                        228
+                                    ],
+                                    "score": 1.0,
+                                    "content": "pp. 10364–10374, 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        }
+                    ],
+                    "index": 9
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        236,
+                        505,
+                        259
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                235,
+                                505,
+                                250
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        235,
+                                        505,
+                                        250
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Spyros Gidaris, Praveer Singh, and Nikos Komodakis. Unsupervised representation learning by",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                248,
+                                390,
+                                260
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        248,
+                                        390,
+                                        260
+                                    ],
+                                    "score": 1.0,
+                                    "content": "predicting image rotations. arXiv preprint arXiv:1803.07728, 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        }
+                    ],
+                    "index": 11.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        266,
+                        505,
+                        311
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                267,
+                                505,
+                                279
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        267,
+                                        505,
+                                        279
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Jean-Bastien Grill, Florian Strub, Florent Altché, Corentin Tallec, Pierre H Richemond, Elena",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                278,
+                                505,
+                                290
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        278,
+                                        505,
+                                        290
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Buchatskaya, Carl Doersch, Bernardo Avila Pires, Zhaohan Daniel Guo, Mohammad Gheshlaghi",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                289,
+                                506,
+                                302
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        289,
+                                        506,
+                                        302
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Azar, et al. Bootstrap your own latent: A new approach to self-supervised learning. arXiv preprint",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                299,
+                                219,
+                                311
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        299,
+                                        219,
+                                        311
+                                    ],
+                                    "score": 1.0,
+                                    "content": "arXiv:2006.07733, 2020.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        }
+                    ],
+                    "index": 14.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        318,
+                        504,
+                        353
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                107,
+                                318,
+                                506,
+                                331
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        318,
+                                        506,
+                                        331
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Yandong Guo, Lei Zhang, Yuxiao Hu, Xiaodong He, and Jianfeng Gao. Ms-celeb-1m: A dataset",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                328,
+                                506,
+                                344
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        328,
+                                        506,
+                                        344
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and benchmark for large-scale face recognition. In European conference on computer vision, pp.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                341,
+                                215,
+                                353
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        341,
+                                        215,
+                                        353
+                                    ],
+                                    "score": 1.0,
+                                    "content": "87–102. Springer, 2016.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        }
+                    ],
+                    "index": 18
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        360,
+                        504,
+                        394
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                359,
+                                505,
+                                373
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        359,
+                                        505,
+                                        373
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Michael Gutmann and Aapo Hyvärinen. Noise-contrastive estimation: A new estimation principle",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                371,
+                                506,
+                                384
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        371,
+                                        506,
+                                        384
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for unnormalized statistical models. In Proceedings of the Thirteenth International Conference on",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                382,
+                                340,
+                                395
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        382,
+                                        340,
+                                        395
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Artificial Intelligence and Statistics, pp. 297–304, 2010.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        }
+                    ],
+                    "index": 21
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        401,
+                        504,
+                        435
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                399,
+                                506,
+                                416
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        399,
+                                        506,
+                                        416
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                412,
+                                506,
+                                425
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        412,
+                                        506,
+                                        425
+                                    ],
+                                    "score": 1.0,
+                                    "content": "recognition. In Proceedings of the IEEE conference on computer vision and pattern recognition,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                114,
+                                423,
+                                197,
+                                435
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        114,
+                                        423,
+                                        197,
+                                        435
+                                    ],
+                                    "score": 1.0,
+                                    "content": "pp. 770–778, 2016.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        }
+                    ],
+                    "index": 24
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        442,
+                        504,
+                        466
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                442,
+                                505,
+                                455
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        442,
+                                        505,
+                                        455
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Kaiming He, Georgia Gkioxari, Piotr Dollár, and Ross Girshick. Mask r-cnn. In Proceedings of the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                453,
+                                412,
+                                466
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        453,
+                                        412,
+                                        466
+                                    ],
+                                    "score": 1.0,
+                                    "content": "IEEE international conference on computer vision, pp. 2961–2969, 2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        }
+                    ],
+                    "index": 26.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        472,
+                        504,
+                        507
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                472,
+                                506,
+                                486
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        472,
+                                        506,
+                                        486
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Kaiming He, Haoqi Fan, Yuxin Wu, Saining Xie, and Ross Girshick. Momentum contrast for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                483,
+                                506,
+                                497
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        483,
+                                        506,
+                                        497
+                                    ],
+                                    "score": 1.0,
+                                    "content": "unsupervised visual representation learning. In Proceedings of the IEEE/CVF Conference on",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                117,
+                                495,
+                                378,
+                                508
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        117,
+                                        495,
+                                        378,
+                                        508
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Computer Vision and Pattern Recognition, pp. 9729–9738, 2020.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        }
+                    ],
+                    "index": 29
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        514,
+                        504,
+                        548
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                107,
+                                514,
+                                505,
+                                526
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        514,
+                                        505,
+                                        526
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Olivier J Hénaff, Aravind Srinivas, Jeffrey De Fauw, Ali Razavi, Carl Doersch, SM Eslami, and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                525,
+                                505,
+                                538
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        525,
+                                        505,
+                                        538
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Aaron van den Oord. Data-efficient image recognition with contrastive predictive coding. arXiv",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                536,
+                                254,
+                                548
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        536,
+                                        254,
+                                        548
+                                    ],
+                                    "score": 1.0,
+                                    "content": "preprint arXiv:1905.09272, 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        }
+                    ],
+                    "index": 32
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        555,
+                        505,
+                        578
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                554,
+                                506,
+                                569
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        554,
+                                        506,
+                                        569
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Geoffrey Hinton, Oriol Vinyals, and Jeff Dean. Distilling the knowledge in a neural network. arXiv",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                567,
+                                254,
+                                578
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        567,
+                                        254,
+                                        578
+                                    ],
+                                    "score": 1.0,
+                                    "content": "preprint arXiv:1503.02531, 2015.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        }
+                    ],
+                    "index": 34.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        585,
+                        505,
+                        620
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                585,
+                                505,
+                                598
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        585,
+                                        505,
+                                        598
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Andrew Howard, Mark Sandler, Grace Chu, Liang-Chieh Chen, Bo Chen, Mingxing Tan, Weijun",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                597,
+                                505,
+                                609
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        597,
+                                        505,
+                                        609
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Wang, Yukun Zhu, Ruoming Pang, Vijay Vasudevan, et al. Searching for mobilenetv3. In",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                607,
+                                494,
+                                621
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        607,
+                                        494,
+                                        621
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Proceedings of the IEEE International Conference on Computer Vision, pp. 1314–1324, 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        }
+                    ],
+                    "index": 37
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        626,
+                        505,
+                        661
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                626,
+                                506,
+                                640
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        626,
+                                        506,
+                                        640
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Andrew G Howard, Menglong Zhu, Bo Chen, Dmitry Kalenichenko, Weijun Wang, Tobias Weyand,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                638,
+                                505,
+                                650
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        638,
+                                        505,
+                                        650
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Marco Andreetto, and Hartwig Adam. Mobilenets: Efficient convolutional neural networks for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                650,
+                                390,
+                                661
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        650,
+                                        390,
+                                        661
+                                    ],
+                                    "score": 1.0,
+                                    "content": "mobile vision applications. arXiv preprint arXiv:1704.04861, 2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        }
+                    ],
+                    "index": 40
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        667,
+                        505,
+                        691
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                668,
+                                506,
+                                681
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        668,
+                                        506,
+                                        681
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Zehao Huang and Naiyan Wang. Like what you like: Knowledge distill via neuron selectivity transfer.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                679,
+                                279,
+                                691
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        679,
+                                        279,
+                                        691
+                                    ],
+                                    "score": 1.0,
+                                    "content": "arXiv preprint arXiv:1707.01219, 2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        }
+                    ],
+                    "index": 42.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        698,
+                        506,
+                        731
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                698,
+                                506,
+                                711
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        698,
+                                        506,
+                                        711
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Xiaoqi Jiao, Yichun Yin, Lifeng Shang, Xin Jiang, Xiao Chen, Linlin Li, Fang Wang, and Qun Liu.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                114,
+                                708,
+                                507,
+                                723
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        114,
+                                        708,
+                                        507,
+                                        723
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Tinybert: Distilling bert for natural language understanding. arXiv preprint arXiv:1909.10351,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                114,
+                                719,
+                                143,
+                                732
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        114,
+                                        719,
+                                        143,
+                                        732
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        }
+                    ],
+                    "index": 45
+                }
+            ],
+            "page_idx": 9,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        108,
+                        27,
+                        292,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                26,
+                                293,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        26,
+                                        293,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Published as a conference paper at ICLR 2021",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        300,
+                        751,
+                        311,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                299,
+                                750,
+                                312,
+                                764
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        299,
+                                        750,
+                                        312,
+                                        764
+                                    ],
+                                    "score": 1.0,
+                                    "content": "10",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        104,
+                        82,
+                        505,
+                        106
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                81,
+                                506,
+                                96
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        81,
+                                        506,
+                                        96
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Jang Hyun Cho and Bharath Hariharan. On the efficacy of knowledge distillation. In Proceedings of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                93,
+                                433,
+                                106
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        93,
+                                        433,
+                                        106
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the IEEE International Conference on Computer Vision, pp. 4794–4802, 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        }
+                    ],
+                    "index": 0.5,
+                    "bbox_fs": [
+                        105,
+                        81,
+                        506,
+                        106
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        112,
+                        504,
+                        146
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                112,
+                                505,
+                                126
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        112,
+                                        505,
+                                        126
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Jia Deng, Wei Dong, Richard Socher, Li-Jia Li, Kai Li, and Li Fei-Fei. Imagenet: A large-scale",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                123,
+                                506,
+                                137
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        123,
+                                        506,
+                                        137
+                                    ],
+                                    "score": 1.0,
+                                    "content": "hierarchical image database. In 2009 IEEE conference on computer vision and pattern recognition,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                134,
+                                218,
+                                146
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        134,
+                                        218,
+                                        146
+                                    ],
+                                    "score": 1.0,
+                                    "content": "pp. 248–255. Ieee, 2009.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        }
+                    ],
+                    "index": 3,
+                    "bbox_fs": [
+                        105,
+                        112,
+                        506,
+                        146
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        153,
+                        503,
+                        188
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                154,
+                                505,
+                                166
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        154,
+                                        505,
+                                        166
+                                    ],
+                                    "score": 1.0,
+                                    "content": "M. Everingham, L. Van Gool, C. K. I. Williams, J. Winn, and A. Zisserman. The",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                164,
+                                505,
+                                178
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        164,
+                                        505,
+                                        178
+                                    ],
+                                    "score": 1.0,
+                                    "content": "PASCAL Visual Object Classes Challenge 2007 (VOC2007) Results. http://www.pascal-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                176,
+                                362,
+                                189
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        176,
+                                        362,
+                                        189
+                                    ],
+                                    "score": 1.0,
+                                    "content": "network.org/challenges/VOC/voc2007/workshop/index.html.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        }
+                    ],
+                    "index": 6,
+                    "bbox_fs": [
+                        106,
+                        154,
+                        505,
+                        189
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        194,
+                        505,
+                        229
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                194,
+                                505,
+                                208
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        194,
+                                        505,
+                                        208
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Zeyu Feng, Chang Xu, and Dacheng Tao. Self-supervised representation learning by rotation feature",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                206,
+                                507,
+                                219
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        206,
+                                        507,
+                                        219
+                                    ],
+                                    "score": 1.0,
+                                    "content": "decoupling. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                217,
+                                216,
+                                228
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        217,
+                                        216,
+                                        228
+                                    ],
+                                    "score": 1.0,
+                                    "content": "pp. 10364–10374, 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        }
+                    ],
+                    "index": 9,
+                    "bbox_fs": [
+                        106,
+                        194,
+                        507,
+                        228
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        236,
+                        505,
+                        259
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                235,
+                                505,
+                                250
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        235,
+                                        505,
+                                        250
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Spyros Gidaris, Praveer Singh, and Nikos Komodakis. Unsupervised representation learning by",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                248,
+                                390,
+                                260
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        248,
+                                        390,
+                                        260
+                                    ],
+                                    "score": 1.0,
+                                    "content": "predicting image rotations. arXiv preprint arXiv:1803.07728, 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        }
+                    ],
+                    "index": 11.5,
+                    "bbox_fs": [
+                        105,
+                        235,
+                        505,
+                        260
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        266,
+                        505,
+                        311
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                267,
+                                505,
+                                279
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        267,
+                                        505,
+                                        279
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Jean-Bastien Grill, Florian Strub, Florent Altché, Corentin Tallec, Pierre H Richemond, Elena",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                278,
+                                505,
+                                290
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        278,
+                                        505,
+                                        290
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Buchatskaya, Carl Doersch, Bernardo Avila Pires, Zhaohan Daniel Guo, Mohammad Gheshlaghi",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                289,
+                                506,
+                                302
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        289,
+                                        506,
+                                        302
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Azar, et al. Bootstrap your own latent: A new approach to self-supervised learning. arXiv preprint",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                299,
+                                219,
+                                311
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        299,
+                                        219,
+                                        311
+                                    ],
+                                    "score": 1.0,
+                                    "content": "arXiv:2006.07733, 2020.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        }
+                    ],
+                    "index": 14.5,
+                    "bbox_fs": [
+                        105,
+                        267,
+                        506,
+                        311
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        318,
+                        504,
+                        353
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                107,
+                                318,
+                                506,
+                                331
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        318,
+                                        506,
+                                        331
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Yandong Guo, Lei Zhang, Yuxiao Hu, Xiaodong He, and Jianfeng Gao. Ms-celeb-1m: A dataset",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                328,
+                                506,
+                                344
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        328,
+                                        506,
+                                        344
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and benchmark for large-scale face recognition. In European conference on computer vision, pp.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                341,
+                                215,
+                                353
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        341,
+                                        215,
+                                        353
+                                    ],
+                                    "score": 1.0,
+                                    "content": "87–102. Springer, 2016.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        }
+                    ],
+                    "index": 18,
+                    "bbox_fs": [
+                        107,
+                        318,
+                        506,
+                        353
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        360,
+                        504,
+                        394
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                359,
+                                505,
+                                373
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        359,
+                                        505,
+                                        373
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Michael Gutmann and Aapo Hyvärinen. Noise-contrastive estimation: A new estimation principle",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                371,
+                                506,
+                                384
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        371,
+                                        506,
+                                        384
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for unnormalized statistical models. In Proceedings of the Thirteenth International Conference on",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                382,
+                                340,
+                                395
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        382,
+                                        340,
+                                        395
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Artificial Intelligence and Statistics, pp. 297–304, 2010.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        }
+                    ],
+                    "index": 21,
+                    "bbox_fs": [
+                        106,
+                        359,
+                        506,
+                        395
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        401,
+                        504,
+                        435
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                399,
+                                506,
+                                416
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        399,
+                                        506,
+                                        416
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                412,
+                                506,
+                                425
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        412,
+                                        506,
+                                        425
+                                    ],
+                                    "score": 1.0,
+                                    "content": "recognition. In Proceedings of the IEEE conference on computer vision and pattern recognition,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                114,
+                                423,
+                                197,
+                                435
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        114,
+                                        423,
+                                        197,
+                                        435
+                                    ],
+                                    "score": 1.0,
+                                    "content": "pp. 770–778, 2016.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        }
+                    ],
+                    "index": 24,
+                    "bbox_fs": [
+                        105,
+                        399,
+                        506,
+                        435
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        442,
+                        504,
+                        466
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                442,
+                                505,
+                                455
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        442,
+                                        505,
+                                        455
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Kaiming He, Georgia Gkioxari, Piotr Dollár, and Ross Girshick. Mask r-cnn. In Proceedings of the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                453,
+                                412,
+                                466
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        453,
+                                        412,
+                                        466
+                                    ],
+                                    "score": 1.0,
+                                    "content": "IEEE international conference on computer vision, pp. 2961–2969, 2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        }
+                    ],
+                    "index": 26.5,
+                    "bbox_fs": [
+                        106,
+                        442,
+                        505,
+                        466
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        472,
+                        504,
+                        507
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                472,
+                                506,
+                                486
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        472,
+                                        506,
+                                        486
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Kaiming He, Haoqi Fan, Yuxin Wu, Saining Xie, and Ross Girshick. Momentum contrast for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                483,
+                                506,
+                                497
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        483,
+                                        506,
+                                        497
+                                    ],
+                                    "score": 1.0,
+                                    "content": "unsupervised visual representation learning. In Proceedings of the IEEE/CVF Conference on",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                117,
+                                495,
+                                378,
+                                508
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        117,
+                                        495,
+                                        378,
+                                        508
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Computer Vision and Pattern Recognition, pp. 9729–9738, 2020.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        }
+                    ],
+                    "index": 29,
+                    "bbox_fs": [
+                        106,
+                        472,
+                        506,
+                        508
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        514,
+                        504,
+                        548
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                107,
+                                514,
+                                505,
+                                526
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        514,
+                                        505,
+                                        526
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Olivier J Hénaff, Aravind Srinivas, Jeffrey De Fauw, Ali Razavi, Carl Doersch, SM Eslami, and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                525,
+                                505,
+                                538
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        525,
+                                        505,
+                                        538
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Aaron van den Oord. Data-efficient image recognition with contrastive predictive coding. arXiv",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                536,
+                                254,
+                                548
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        536,
+                                        254,
+                                        548
+                                    ],
+                                    "score": 1.0,
+                                    "content": "preprint arXiv:1905.09272, 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        }
+                    ],
+                    "index": 32,
+                    "bbox_fs": [
+                        107,
+                        514,
+                        505,
+                        548
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        555,
+                        505,
+                        578
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                554,
+                                506,
+                                569
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        554,
+                                        506,
+                                        569
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Geoffrey Hinton, Oriol Vinyals, and Jeff Dean. Distilling the knowledge in a neural network. arXiv",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                567,
+                                254,
+                                578
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        567,
+                                        254,
+                                        578
+                                    ],
+                                    "score": 1.0,
+                                    "content": "preprint arXiv:1503.02531, 2015.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        }
+                    ],
+                    "index": 34.5,
+                    "bbox_fs": [
+                        106,
+                        554,
+                        506,
+                        578
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        585,
+                        505,
+                        620
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                585,
+                                505,
+                                598
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        585,
+                                        505,
+                                        598
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Andrew Howard, Mark Sandler, Grace Chu, Liang-Chieh Chen, Bo Chen, Mingxing Tan, Weijun",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                597,
+                                505,
+                                609
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        597,
+                                        505,
+                                        609
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Wang, Yukun Zhu, Ruoming Pang, Vijay Vasudevan, et al. Searching for mobilenetv3. In",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                607,
+                                494,
+                                621
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        607,
+                                        494,
+                                        621
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Proceedings of the IEEE International Conference on Computer Vision, pp. 1314–1324, 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        }
+                    ],
+                    "index": 37,
+                    "bbox_fs": [
+                        106,
+                        585,
+                        505,
+                        621
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        626,
+                        505,
+                        661
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                626,
+                                506,
+                                640
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        626,
+                                        506,
+                                        640
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Andrew G Howard, Menglong Zhu, Bo Chen, Dmitry Kalenichenko, Weijun Wang, Tobias Weyand,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                638,
+                                505,
+                                650
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        638,
+                                        505,
+                                        650
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Marco Andreetto, and Hartwig Adam. Mobilenets: Efficient convolutional neural networks for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                650,
+                                390,
+                                661
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        650,
+                                        390,
+                                        661
+                                    ],
+                                    "score": 1.0,
+                                    "content": "mobile vision applications. arXiv preprint arXiv:1704.04861, 2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        }
+                    ],
+                    "index": 40,
+                    "bbox_fs": [
+                        105,
+                        626,
+                        506,
+                        661
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        667,
+                        505,
+                        691
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                668,
+                                506,
+                                681
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        668,
+                                        506,
+                                        681
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Zehao Huang and Naiyan Wang. Like what you like: Knowledge distill via neuron selectivity transfer.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                679,
+                                279,
+                                691
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        679,
+                                        279,
+                                        691
+                                    ],
+                                    "score": 1.0,
+                                    "content": "arXiv preprint arXiv:1707.01219, 2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        }
+                    ],
+                    "index": 42.5,
+                    "bbox_fs": [
+                        106,
+                        668,
+                        506,
+                        691
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        698,
+                        506,
+                        731
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                698,
+                                506,
+                                711
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        698,
+                                        506,
+                                        711
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Xiaoqi Jiao, Yichun Yin, Lifeng Shang, Xin Jiang, Xiao Chen, Linlin Li, Fang Wang, and Qun Liu.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                114,
+                                708,
+                                507,
+                                723
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        114,
+                                        708,
+                                        507,
+                                        723
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Tinybert: Distilling bert for natural language understanding. arXiv preprint arXiv:1909.10351,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                114,
+                                719,
+                                143,
+                                732
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        114,
+                                        719,
+                                        143,
+                                        732
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        }
+                    ],
+                    "index": 45,
+                    "bbox_fs": [
+                        105,
+                        698,
+                        507,
+                        732
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        82,
+                        505,
+                        116
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                81,
+                                506,
+                                95
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        81,
+                                        506,
+                                        95
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Alexander Kolesnikov, Xiaohua Zhai, and Lucas Beyer. Revisiting self-supervised visual representa-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                93,
+                                506,
+                                106
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        93,
+                                        506,
+                                        106
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tion learning. In Proceedings of the IEEE conference on Computer Vision and Pattern Recognition,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                105,
+                                206,
+                                115
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        105,
+                                        206,
+                                        115
+                                    ],
+                                    "score": 1.0,
+                                    "content": "pp. 1920–1929, 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        }
+                    ],
+                    "index": 1
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        122,
+                        505,
+                        156
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                122,
+                                505,
+                                135
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        122,
+                                        505,
+                                        135
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Animesh Koratana, Daniel Kang, Peter Bailis, and Matei Zaharia. Lit: Learned intermediate",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                133,
+                                507,
+                                147
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        133,
+                                        507,
+                                        147
+                                    ],
+                                    "score": 1.0,
+                                    "content": "representation training for model compression. In International Conference on Machine Learning,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                144,
+                                207,
+                                156
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        144,
+                                        207,
+                                        156
+                                    ],
+                                    "score": 1.0,
+                                    "content": "pp. 3509–3518, 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        }
+                    ],
+                    "index": 4
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        162,
+                        504,
+                        196
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                163,
+                                505,
+                                175
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        163,
+                                        505,
+                                        175
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Simon Kornblith, Jonathon Shlens, and Quoc V Le. Do better imagenet models transfer better? In",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                173,
+                                507,
+                                187
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        173,
+                                        507,
+                                        187
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Proceedings of the IEEE conference on computer vision and pattern recognition, pp. 2661–2671,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                184,
+                                143,
+                                198
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        184,
+                                        143,
+                                        198
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        }
+                    ],
+                    "index": 7
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        203,
+                        505,
+                        216
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                201,
+                                507,
+                                217
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        201,
+                                        507,
+                                        217
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Alex Krizhevsky, Geoffrey Hinton, et al. Learning multiple layers of features from tiny images. 2009.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        }
+                    ],
+                    "index": 9
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        221,
+                        504,
+                        245
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                221,
+                                505,
+                                234
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        221,
+                                        505,
+                                        234
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Junnan Li, Pan Zhou, Caiming Xiong, Richard Socher, and Steven CH Hoi. Prototypical contrastive",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                233,
+                                446,
+                                245
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        233,
+                                        446,
+                                        245
+                                    ],
+                                    "score": 1.0,
+                                    "content": "learning of unsupervised representations. arXiv preprint arXiv:2005.04966, 2020.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        }
+                    ],
+                    "index": 10.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        250,
+                        504,
+                        285
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                251,
+                                506,
+                                263
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        251,
+                                        506,
+                                        263
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Tsung-Yi Lin, Michael Maire, Serge Belongie, James Hays, Pietro Perona, Deva Ramanan, Piotr",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                261,
+                                506,
+                                275
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        261,
+                                        506,
+                                        275
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Dollár, and C Lawrence Zitnick. Microsoft coco: Common objects in context. In European",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                273,
+                                363,
+                                286
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        273,
+                                        363,
+                                        286
+                                    ],
+                                    "score": 1.0,
+                                    "content": "conference on computer vision, pp. 740–755. Springer, 2014.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        }
+                    ],
+                    "index": 13
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        291,
+                        504,
+                        326
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                290,
+                                505,
+                                305
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        290,
+                                        505,
+                                        305
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Yifan Liu, Ke Chen, Chris Liu, Zengchang Qin, Zhenbo Luo, and Jingdong Wang. Structured",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                302,
+                                505,
+                                316
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        302,
+                                        505,
+                                        316
+                                    ],
+                                    "score": 1.0,
+                                    "content": "knowledge distillation for semantic segmentation. In Proceedings of the IEEE Conference on",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                313,
+                                378,
+                                327
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        313,
+                                        378,
+                                        327
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Computer Vision and Pattern Recognition, pp. 2604–2613, 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        }
+                    ],
+                    "index": 16
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        331,
+                        504,
+                        354
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                331,
+                                505,
+                                345
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        331,
+                                        505,
+                                        345
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Asit Mishra and Debbie Marr. Apprentice: Using knowledge distillation techniques to improve",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                342,
+                                412,
+                                356
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        342,
+                                        412,
+                                        356
+                                    ],
+                                    "score": 1.0,
+                                    "content": "low-precision network accuracy. arXiv preprint arXiv:1711.05852, 2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        }
+                    ],
+                    "index": 18.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        360,
+                        505,
+                        394
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                360,
+                                506,
+                                374
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        360,
+                                        506,
+                                        374
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Ishan Misra and Laurens van der Maaten. Self-supervised learning of pretext-invariant representations.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                370,
+                                507,
+                                387
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        370,
+                                        507,
+                                        387
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                117,
+                                383,
+                                191,
+                                394
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        117,
+                                        383,
+                                        191,
+                                        394
+                                    ],
+                                    "score": 1.0,
+                                    "content": "6707–6717, 2020.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        }
+                    ],
+                    "index": 21
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        401,
+                        505,
+                        424
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                401,
+                                505,
+                                414
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        401,
+                                        505,
+                                        414
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Vinod Nair and Geoffrey E Hinton. Rectified linear units improve restricted boltzmann machines. In",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                412,
+                                170,
+                                424
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        412,
+                                        170,
+                                        424
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ICML, 2010.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        }
+                    ],
+                    "index": 23.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        430,
+                        504,
+                        454
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                429,
+                                506,
+                                444
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        429,
+                                        506,
+                                        444
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Mehdi Noroozi and Paolo Favaro. Unsupervised learning of visual representations by solving jigsaw",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                441,
+                                446,
+                                455
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        441,
+                                        446,
+                                        455
+                                    ],
+                                    "score": 1.0,
+                                    "content": "puzzles. In European Conference on Computer Vision, pp. 69–84. Springer, 2016.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        }
+                    ],
+                    "index": 25.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        459,
+                        504,
+                        483
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                459,
+                                505,
+                                473
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        459,
+                                        505,
+                                        473
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Aaron van den Oord, Yazhe Li, and Oriol Vinyals. Representation learning with contrastive predictive",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                471,
+                                312,
+                                483
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        471,
+                                        312,
+                                        483
+                                    ],
+                                    "score": 1.0,
+                                    "content": "coding. arXiv preprint arXiv:1807.03748, 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        }
+                    ],
+                    "index": 27.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        489,
+                        506,
+                        534
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                490,
+                                505,
+                                502
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        490,
+                                        505,
+                                        502
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Adam Paszke, Sam Gross, Francisco Massa, Adam Lerer, James Bradbury, Gregory Chanan, Trevor",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                500,
+                                506,
+                                513
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        500,
+                                        506,
+                                        513
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Killeen, Zeming Lin, Natalia Gimelshein, Luca Antiga, et al. Pytorch: An imperative style,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                511,
+                                507,
+                                525
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        511,
+                                        507,
+                                        525
+                                    ],
+                                    "score": 1.0,
+                                    "content": "high-performance deep learning library. In Advances in neural information processing systems, pp.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                523,
+                                191,
+                                534
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        523,
+                                        191,
+                                        534
+                                    ],
+                                    "score": 1.0,
+                                    "content": "8026–8037, 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        }
+                    ],
+                    "index": 30.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        540,
+                        504,
+                        564
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                540,
+                                505,
+                                553
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        540,
+                                        505,
+                                        553
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Antonio Polino, Razvan Pascanu, and Dan Alistarh. Model compression via distillation and quantiza-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                552,
+                                300,
+                                564
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        552,
+                                        300,
+                                        564
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tion. arXiv preprint arXiv:1802.05668, 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        }
+                    ],
+                    "index": 33.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        569,
+                        506,
+                        604
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                569,
+                                506,
+                                583
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        569,
+                                        506,
+                                        583
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Shaoqing Ren, Kaiming He, Ross Girshick, and Jian Sun. Faster r-cnn: Towards real-time object",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                581,
+                                506,
+                                594
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        581,
+                                        506,
+                                        594
+                                    ],
+                                    "score": 1.0,
+                                    "content": "detection with region proposal networks. In Advances in neural information processing systems,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                114,
+                                592,
+                                187,
+                                604
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        114,
+                                        592,
+                                        187,
+                                        604
+                                    ],
+                                    "score": 1.0,
+                                    "content": "pp. 91–99, 2015.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        }
+                    ],
+                    "index": 36
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        610,
+                        504,
+                        633
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                610,
+                                505,
+                                623
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        610,
+                                        505,
+                                        623
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Adriana Romero, Nicolas Ballas, Samira Ebrahimi Kahou, Antoine Chassang, Carlo Gatta, and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                621,
+                                340,
+                                634
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        621,
+                                        340,
+                                        634
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Yoshua Bengio. Fitnets: Hints for thin deep nets. 2014.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        }
+                    ],
+                    "index": 38.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        639,
+                        505,
+                        673
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                639,
+                                506,
+                                653
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        639,
+                                        506,
+                                        653
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Mark Sandler, Andrew Howard, Menglong Zhu, Andrey Zhmoginov, and Liang-Chieh Chen. Mo-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                650,
+                                505,
+                                664
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        650,
+                                        505,
+                                        664
+                                    ],
+                                    "score": 1.0,
+                                    "content": "bilenetv2: Inverted residuals and linear bottlenecks. In Proceedings of the IEEE conference on",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                662,
+                                372,
+                                674
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        662,
+                                        372,
+                                        674
+                                    ],
+                                    "score": 1.0,
+                                    "content": "computer vision and pattern recognition, pp. 4510–4520, 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        }
+                    ],
+                    "index": 41
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        680,
+                        503,
+                        703
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                107,
+                                680,
+                                505,
+                                692
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        680,
+                                        505,
+                                        692
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Victor Sanh, Lysandre Debut, Julien Chaumond, and Thomas Wolf. Distilbert, a distilled version of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                691,
+                                443,
+                                703
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        691,
+                                        443,
+                                        703
+                                    ],
+                                    "score": 1.0,
+                                    "content": "bert: smaller, faster, cheaper and lighter. arXiv preprint arXiv:1910.01108, 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        }
+                    ],
+                    "index": 43.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        709,
+                        503,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                708,
+                                505,
+                                723
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        708,
+                                        407,
+                                        723
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Zhiqiang Shen and Marios Savvides. Meal v2: Boosting vanilla resnet-50 to",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        407,
+                                        710,
+                                        432,
+                                        720
+                                    ],
+                                    "score": 0.88,
+                                    "content": "80 \\text{‰}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        432,
+                                        708,
+                                        505,
+                                        723
+                                    ],
+                                    "score": 1.0,
+                                    "content": "top-1 accuracy on",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                720,
+                                378,
+                                732
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        720,
+                                        378,
+                                        732
+                                    ],
+                                    "score": 1.0,
+                                    "content": "imagenet without tricks. arXiv preprint arXiv:2009.08453, 2020.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        }
+                    ],
+                    "index": 45.5
+                }
+            ],
+            "page_idx": 10,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        108,
+                        27,
+                        293,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                26,
+                                293,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        26,
+                                        293,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Published as a conference paper at ICLR 2021",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        300,
+                        751,
+                        310,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                299,
+                                750,
+                                312,
+                                765
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        299,
+                                        750,
+                                        312,
+                                        765
+                                    ],
+                                    "score": 1.0,
+                                    "content": "",
+                                    "type": "text",
+                                    "height": 15,
+                                    "width": 13
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        82,
+                        505,
+                        116
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                81,
+                                506,
+                                95
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        81,
+                                        506,
+                                        95
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Alexander Kolesnikov, Xiaohua Zhai, and Lucas Beyer. Revisiting self-supervised visual representa-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                93,
+                                506,
+                                106
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        93,
+                                        506,
+                                        106
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tion learning. In Proceedings of the IEEE conference on Computer Vision and Pattern Recognition,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                105,
+                                206,
+                                115
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        105,
+                                        206,
+                                        115
+                                    ],
+                                    "score": 1.0,
+                                    "content": "pp. 1920–1929, 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        }
+                    ],
+                    "index": 1,
+                    "bbox_fs": [
+                        105,
+                        81,
+                        506,
+                        115
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        122,
+                        505,
+                        156
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                122,
+                                505,
+                                135
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        122,
+                                        505,
+                                        135
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Animesh Koratana, Daniel Kang, Peter Bailis, and Matei Zaharia. Lit: Learned intermediate",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                133,
+                                507,
+                                147
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        133,
+                                        507,
+                                        147
+                                    ],
+                                    "score": 1.0,
+                                    "content": "representation training for model compression. In International Conference on Machine Learning,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                144,
+                                207,
+                                156
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        144,
+                                        207,
+                                        156
+                                    ],
+                                    "score": 1.0,
+                                    "content": "pp. 3509–3518, 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        }
+                    ],
+                    "index": 4,
+                    "bbox_fs": [
+                        106,
+                        122,
+                        507,
+                        156
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        162,
+                        504,
+                        196
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                163,
+                                505,
+                                175
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        163,
+                                        505,
+                                        175
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Simon Kornblith, Jonathon Shlens, and Quoc V Le. Do better imagenet models transfer better? In",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                173,
+                                507,
+                                187
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        173,
+                                        507,
+                                        187
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Proceedings of the IEEE conference on computer vision and pattern recognition, pp. 2661–2671,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                184,
+                                143,
+                                198
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        184,
+                                        143,
+                                        198
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        }
+                    ],
+                    "index": 7,
+                    "bbox_fs": [
+                        106,
+                        163,
+                        507,
+                        198
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        203,
+                        505,
+                        216
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                201,
+                                507,
+                                217
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        201,
+                                        507,
+                                        217
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Alex Krizhevsky, Geoffrey Hinton, et al. Learning multiple layers of features from tiny images. 2009.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        }
+                    ],
+                    "index": 9,
+                    "bbox_fs": [
+                        106,
+                        201,
+                        507,
+                        217
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        221,
+                        504,
+                        245
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                221,
+                                505,
+                                234
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        221,
+                                        505,
+                                        234
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Junnan Li, Pan Zhou, Caiming Xiong, Richard Socher, and Steven CH Hoi. Prototypical contrastive",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                233,
+                                446,
+                                245
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        233,
+                                        446,
+                                        245
+                                    ],
+                                    "score": 1.0,
+                                    "content": "learning of unsupervised representations. arXiv preprint arXiv:2005.04966, 2020.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        }
+                    ],
+                    "index": 10.5,
+                    "bbox_fs": [
+                        106,
+                        221,
+                        505,
+                        245
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        250,
+                        504,
+                        285
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                251,
+                                506,
+                                263
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        251,
+                                        506,
+                                        263
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Tsung-Yi Lin, Michael Maire, Serge Belongie, James Hays, Pietro Perona, Deva Ramanan, Piotr",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                261,
+                                506,
+                                275
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        261,
+                                        506,
+                                        275
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Dollár, and C Lawrence Zitnick. Microsoft coco: Common objects in context. In European",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                273,
+                                363,
+                                286
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        273,
+                                        363,
+                                        286
+                                    ],
+                                    "score": 1.0,
+                                    "content": "conference on computer vision, pp. 740–755. Springer, 2014.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        }
+                    ],
+                    "index": 13,
+                    "bbox_fs": [
+                        106,
+                        251,
+                        506,
+                        286
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        291,
+                        504,
+                        326
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                290,
+                                505,
+                                305
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        290,
+                                        505,
+                                        305
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Yifan Liu, Ke Chen, Chris Liu, Zengchang Qin, Zhenbo Luo, and Jingdong Wang. Structured",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                302,
+                                505,
+                                316
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        302,
+                                        505,
+                                        316
+                                    ],
+                                    "score": 1.0,
+                                    "content": "knowledge distillation for semantic segmentation. In Proceedings of the IEEE Conference on",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                313,
+                                378,
+                                327
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        313,
+                                        378,
+                                        327
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Computer Vision and Pattern Recognition, pp. 2604–2613, 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        }
+                    ],
+                    "index": 16,
+                    "bbox_fs": [
+                        106,
+                        290,
+                        505,
+                        327
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        331,
+                        504,
+                        354
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                331,
+                                505,
+                                345
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        331,
+                                        505,
+                                        345
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Asit Mishra and Debbie Marr. Apprentice: Using knowledge distillation techniques to improve",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                342,
+                                412,
+                                356
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        342,
+                                        412,
+                                        356
+                                    ],
+                                    "score": 1.0,
+                                    "content": "low-precision network accuracy. arXiv preprint arXiv:1711.05852, 2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        }
+                    ],
+                    "index": 18.5,
+                    "bbox_fs": [
+                        106,
+                        331,
+                        505,
+                        356
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        360,
+                        505,
+                        394
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                360,
+                                506,
+                                374
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        360,
+                                        506,
+                                        374
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Ishan Misra and Laurens van der Maaten. Self-supervised learning of pretext-invariant representations.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                370,
+                                507,
+                                387
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        370,
+                                        507,
+                                        387
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                117,
+                                383,
+                                191,
+                                394
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        117,
+                                        383,
+                                        191,
+                                        394
+                                    ],
+                                    "score": 1.0,
+                                    "content": "6707–6717, 2020.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        }
+                    ],
+                    "index": 21,
+                    "bbox_fs": [
+                        105,
+                        360,
+                        507,
+                        394
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        401,
+                        505,
+                        424
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                401,
+                                505,
+                                414
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        401,
+                                        505,
+                                        414
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Vinod Nair and Geoffrey E Hinton. Rectified linear units improve restricted boltzmann machines. In",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                412,
+                                170,
+                                424
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        412,
+                                        170,
+                                        424
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ICML, 2010.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        }
+                    ],
+                    "index": 23.5,
+                    "bbox_fs": [
+                        106,
+                        401,
+                        505,
+                        424
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        430,
+                        504,
+                        454
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                429,
+                                506,
+                                444
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        429,
+                                        506,
+                                        444
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Mehdi Noroozi and Paolo Favaro. Unsupervised learning of visual representations by solving jigsaw",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                441,
+                                446,
+                                455
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        441,
+                                        446,
+                                        455
+                                    ],
+                                    "score": 1.0,
+                                    "content": "puzzles. In European Conference on Computer Vision, pp. 69–84. Springer, 2016.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        }
+                    ],
+                    "index": 25.5,
+                    "bbox_fs": [
+                        105,
+                        429,
+                        506,
+                        455
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        459,
+                        504,
+                        483
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                459,
+                                505,
+                                473
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        459,
+                                        505,
+                                        473
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Aaron van den Oord, Yazhe Li, and Oriol Vinyals. Representation learning with contrastive predictive",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                471,
+                                312,
+                                483
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        471,
+                                        312,
+                                        483
+                                    ],
+                                    "score": 1.0,
+                                    "content": "coding. arXiv preprint arXiv:1807.03748, 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        }
+                    ],
+                    "index": 27.5,
+                    "bbox_fs": [
+                        106,
+                        459,
+                        505,
+                        483
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        489,
+                        506,
+                        534
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                490,
+                                505,
+                                502
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        490,
+                                        505,
+                                        502
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Adam Paszke, Sam Gross, Francisco Massa, Adam Lerer, James Bradbury, Gregory Chanan, Trevor",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                500,
+                                506,
+                                513
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        500,
+                                        506,
+                                        513
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Killeen, Zeming Lin, Natalia Gimelshein, Luca Antiga, et al. Pytorch: An imperative style,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                511,
+                                507,
+                                525
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        511,
+                                        507,
+                                        525
+                                    ],
+                                    "score": 1.0,
+                                    "content": "high-performance deep learning library. In Advances in neural information processing systems, pp.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                523,
+                                191,
+                                534
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        523,
+                                        191,
+                                        534
+                                    ],
+                                    "score": 1.0,
+                                    "content": "8026–8037, 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        }
+                    ],
+                    "index": 30.5,
+                    "bbox_fs": [
+                        106,
+                        490,
+                        507,
+                        534
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        540,
+                        504,
+                        564
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                540,
+                                505,
+                                553
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        540,
+                                        505,
+                                        553
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Antonio Polino, Razvan Pascanu, and Dan Alistarh. Model compression via distillation and quantiza-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                552,
+                                300,
+                                564
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        552,
+                                        300,
+                                        564
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tion. arXiv preprint arXiv:1802.05668, 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        }
+                    ],
+                    "index": 33.5,
+                    "bbox_fs": [
+                        106,
+                        540,
+                        505,
+                        564
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        569,
+                        506,
+                        604
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                569,
+                                506,
+                                583
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        569,
+                                        506,
+                                        583
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Shaoqing Ren, Kaiming He, Ross Girshick, and Jian Sun. Faster r-cnn: Towards real-time object",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                581,
+                                506,
+                                594
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        581,
+                                        506,
+                                        594
+                                    ],
+                                    "score": 1.0,
+                                    "content": "detection with region proposal networks. In Advances in neural information processing systems,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                114,
+                                592,
+                                187,
+                                604
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        114,
+                                        592,
+                                        187,
+                                        604
+                                    ],
+                                    "score": 1.0,
+                                    "content": "pp. 91–99, 2015.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        }
+                    ],
+                    "index": 36,
+                    "bbox_fs": [
+                        106,
+                        569,
+                        506,
+                        604
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        610,
+                        504,
+                        633
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                610,
+                                505,
+                                623
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        610,
+                                        505,
+                                        623
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Adriana Romero, Nicolas Ballas, Samira Ebrahimi Kahou, Antoine Chassang, Carlo Gatta, and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                621,
+                                340,
+                                634
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        621,
+                                        340,
+                                        634
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Yoshua Bengio. Fitnets: Hints for thin deep nets. 2014.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        }
+                    ],
+                    "index": 38.5,
+                    "bbox_fs": [
+                        106,
+                        610,
+                        505,
+                        634
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        639,
+                        505,
+                        673
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                639,
+                                506,
+                                653
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        639,
+                                        506,
+                                        653
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Mark Sandler, Andrew Howard, Menglong Zhu, Andrey Zhmoginov, and Liang-Chieh Chen. Mo-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                650,
+                                505,
+                                664
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        650,
+                                        505,
+                                        664
+                                    ],
+                                    "score": 1.0,
+                                    "content": "bilenetv2: Inverted residuals and linear bottlenecks. In Proceedings of the IEEE conference on",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                662,
+                                372,
+                                674
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        662,
+                                        372,
+                                        674
+                                    ],
+                                    "score": 1.0,
+                                    "content": "computer vision and pattern recognition, pp. 4510–4520, 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        }
+                    ],
+                    "index": 41,
+                    "bbox_fs": [
+                        105,
+                        639,
+                        506,
+                        674
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        680,
+                        503,
+                        703
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                107,
+                                680,
+                                505,
+                                692
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        680,
+                                        505,
+                                        692
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Victor Sanh, Lysandre Debut, Julien Chaumond, and Thomas Wolf. Distilbert, a distilled version of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                691,
+                                443,
+                                703
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        691,
+                                        443,
+                                        703
+                                    ],
+                                    "score": 1.0,
+                                    "content": "bert: smaller, faster, cheaper and lighter. arXiv preprint arXiv:1910.01108, 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        }
+                    ],
+                    "index": 43.5,
+                    "bbox_fs": [
+                        107,
+                        680,
+                        505,
+                        703
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        709,
+                        503,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                708,
+                                505,
+                                723
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        708,
+                                        407,
+                                        723
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Zhiqiang Shen and Marios Savvides. Meal v2: Boosting vanilla resnet-50 to",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        407,
+                                        710,
+                                        432,
+                                        720
+                                    ],
+                                    "score": 0.88,
+                                    "content": "80 \\text{‰}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        432,
+                                        708,
+                                        505,
+                                        723
+                                    ],
+                                    "score": 1.0,
+                                    "content": "top-1 accuracy on",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                720,
+                                378,
+                                732
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        720,
+                                        378,
+                                        732
+                                    ],
+                                    "score": 1.0,
+                                    "content": "imagenet without tricks. arXiv preprint arXiv:2009.08453, 2020.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        }
+                    ],
+                    "index": 45.5,
+                    "bbox_fs": [
+                        105,
+                        708,
+                        505,
+                        732
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        82,
+                        505,
+                        105
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                81,
+                                505,
+                                95
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        81,
+                                        505,
+                                        95
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Siqi Sun, Yu Cheng, Zhe Gan, and Jingjing Liu. Patient knowledge distillation for bert model",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                93,
+                                335,
+                                106
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        93,
+                                        335,
+                                        106
+                                    ],
+                                    "score": 1.0,
+                                    "content": "compression. arXiv preprint arXiv:1908.09355, 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        }
+                    ],
+                    "index": 0.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        111,
+                        504,
+                        136
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                112,
+                                506,
+                                125
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        112,
+                                        506,
+                                        125
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Zhiqing Sun, Hongkun Yu, Xiaodan Song, Renjie Liu, Yiming Yang, and Denny Zhou. Mobilebert: a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                123,
+                                505,
+                                136
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        123,
+                                        505,
+                                        136
+                                    ],
+                                    "score": 1.0,
+                                    "content": "compact task-agnostic bert for resource-limited devices. arXiv preprint arXiv:2004.02984, 2020.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        }
+                    ],
+                    "index": 2.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        142,
+                        504,
+                        165
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                142,
+                                505,
+                                155
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        142,
+                                        505,
+                                        155
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Mingxing Tan and Quoc V Le. Efficientnet: Rethinking model scaling for convolutional neural",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                154,
+                                321,
+                                165
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        154,
+                                        321,
+                                        165
+                                    ],
+                                    "score": 1.0,
+                                    "content": "networks. arXiv preprint arXiv:1905.11946, 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        }
+                    ],
+                    "index": 4.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        171,
+                        505,
+                        195
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                172,
+                                506,
+                                185
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        172,
+                                        506,
+                                        185
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Yonglong Tian, Dilip Krishnan, and Phillip Isola. Contrastive representation distillation. In Interna-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                183,
+                                340,
+                                195
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        183,
+                                        340,
+                                        195
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tional Conference on Learning Representations, 2019a.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        }
+                    ],
+                    "index": 6.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        201,
+                        504,
+                        225
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                200,
+                                506,
+                                215
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        200,
+                                        506,
+                                        215
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Yonglong Tian, Dilip Krishnan, and Phillip Isola. Contrastive multiview coding. arXiv preprint",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                213,
+                                223,
+                                225
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        213,
+                                        223,
+                                        225
+                                    ],
+                                    "score": 1.0,
+                                    "content": "arXiv:1906.05849, 2019b.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        }
+                    ],
+                    "index": 8.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        231,
+                        505,
+                        255
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                232,
+                                506,
+                                245
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        232,
+                                        506,
+                                        245
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Iulia Turc, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. Well-read students learn better:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                243,
+                                487,
+                                255
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        243,
+                                        487,
+                                        255
+                                    ],
+                                    "score": 1.0,
+                                    "content": "On the importance of pre-training compact models. arXiv preprint arXiv:1908.08962, 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        }
+                    ],
+                    "index": 10.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        261,
+                        505,
+                        285
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                262,
+                                505,
+                                275
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        262,
+                                        505,
+                                        275
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Tongzhou Wang and Phillip Isola. Understanding contrastive representation learning through align-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                273,
+                                444,
+                                286
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        273,
+                                        444,
+                                        286
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ment and uniformity on the hypersphere. arXiv preprint arXiv:2005.10242, 2020.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        }
+                    ],
+                    "index": 12.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        291,
+                        505,
+                        304
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                290,
+                                506,
+                                304
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        290,
+                                        506,
+                                        304
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Yuxin Wu, Alexander Kirillov, Francisco Massa, Wan-Yen Lo, and Ross Girshick. Detectron2, 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        }
+                    ],
+                    "index": 14
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        309,
+                        504,
+                        345
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                309,
+                                506,
+                                324
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        309,
+                                        506,
+                                        324
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Zhirong Wu, Yuanjun Xiong, Stella X Yu, and Dahua Lin. Unsupervised feature learning via non-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                321,
+                                506,
+                                335
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        321,
+                                        506,
+                                        335
+                                    ],
+                                    "score": 1.0,
+                                    "content": "parametric instance discrimination. In Proceedings of the IEEE Conference on Computer Vision",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                333,
+                                308,
+                                345
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        333,
+                                        308,
+                                        345
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and Pattern Recognition, pp. 3733–3742, 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        }
+                    ],
+                    "index": 16
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        351,
+                        504,
+                        385
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                351,
+                                506,
+                                363
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        351,
+                                        506,
+                                        363
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Jianxiong Xiao, James Hays, Krista A Ehinger, Aude Oliva, and Antonio Torralba. Sun database:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                363,
+                                506,
+                                375
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        363,
+                                        506,
+                                        375
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Large-scale scene recognition from abbey to zoo. In 2010 IEEE computer society conference on",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                373,
+                                398,
+                                386
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        373,
+                                        398,
+                                        386
+                                    ],
+                                    "score": 1.0,
+                                    "content": "computer vision and pattern recognition, pp. 3485–3492. IEEE, 2010.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        }
+                    ],
+                    "index": 19
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        392,
+                        504,
+                        426
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                392,
+                                505,
+                                405
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        392,
+                                        505,
+                                        405
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Jianwei Yang, Devi Parikh, and Dhruv Batra. Joint unsupervised learning of deep representations",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                403,
+                                505,
+                                416
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        403,
+                                        505,
+                                        416
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and image clusters. In Proceedings of the IEEE Conference on Computer Vision and Pattern",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                415,
+                                259,
+                                427
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        415,
+                                        259,
+                                        427
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Recognition, pp. 5147–5156, 2016.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        }
+                    ],
+                    "index": 22
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        473,
+                        506,
+                        507
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                474,
+                                506,
+                                486
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        474,
+                                        506,
+                                        486
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Sergey Zagoruyko and Nikos Komodakis. Paying more attention to attention: Improving the perfor-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                485,
+                                506,
+                                497
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        485,
+                                        506,
+                                        497
+                                    ],
+                                    "score": 1.0,
+                                    "content": "mance of convolutional neural networks via attention transfer. arXiv preprint arXiv:1612.03928,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                494,
+                                143,
+                                509
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        494,
+                                        143,
+                                        509
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2016.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        }
+                    ],
+                    "index": 25
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        514,
+                        506,
+                        549
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                514,
+                                506,
+                                527
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        514,
+                                        506,
+                                        527
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Xiaohua Zhai, Avital Oliver, Alexander Kolesnikov, and Lucas Beyer. S4l: Self-supervised semi-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                525,
+                                507,
+                                540
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        525,
+                                        507,
+                                        540
+                                    ],
+                                    "score": 1.0,
+                                    "content": "supervised learning. In Proceedings of the IEEE international conference on computer vision, pp.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                537,
+                                191,
+                                548
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        537,
+                                        191,
+                                        548
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1476–1485, 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        }
+                    ],
+                    "index": 28
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        555,
+                        504,
+                        590
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                555,
+                                505,
+                                569
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        555,
+                                        505,
+                                        569
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Liheng Zhang, Guo-Jun Qi, Liqiang Wang, and Jiebo Luo. Aet vs. aed: Unsupervised representation",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                566,
+                                506,
+                                580
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        566,
+                                        506,
+                                        580
+                                    ],
+                                    "score": 1.0,
+                                    "content": "learning by auto-encoding transformations rather than data. In Proceedings of the IEEE Conference",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                577,
+                                390,
+                                590
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        577,
+                                        390,
+                                        590
+                                    ],
+                                    "score": 1.0,
+                                    "content": "on Computer Vision and Pattern Recognition, pp. 2547–2555, 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        }
+                    ],
+                    "index": 31
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        596,
+                        504,
+                        620
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                595,
+                                505,
+                                610
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        595,
+                                        505,
+                                        610
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Richard Zhang, Phillip Isola, and Alexei A Efros. Colorful image colorization. In European",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                608,
+                                362,
+                                619
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        608,
+                                        362,
+                                        619
+                                    ],
+                                    "score": 1.0,
+                                    "content": "conference on computer vision, pp. 649–666. Springer, 2016.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        }
+                    ],
+                    "index": 33.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        626,
+                        504,
+                        660
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                626,
+                                506,
+                                639
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        626,
+                                        506,
+                                        639
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Xiao Zhang, Zhiyuan Fang, Yandong Wen, Zhifeng Li, and Yu Qiao. Range loss for deep face",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                637,
+                                505,
+                                650
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        637,
+                                        505,
+                                        650
+                                    ],
+                                    "score": 1.0,
+                                    "content": "recognition with long-tailed training data. In Proceedings of the IEEE International Conference on",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                648,
+                                279,
+                                661
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        648,
+                                        279,
+                                        661
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Computer Vision, pp. 5409–5418, 2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        }
+                    ],
+                    "index": 36
+                }
+            ],
+            "page_idx": 11,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        107,
+                        27,
+                        292,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                26,
+                                293,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        26,
+                                        293,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Published as a conference paper at ICLR 2021",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        300,
+                        751,
+                        311,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                299,
+                                750,
+                                312,
+                                764
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        299,
+                                        750,
+                                        312,
+                                        764
+                                    ],
+                                    "score": 1.0,
+                                    "content": "12",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        82,
+                        505,
+                        105
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                81,
+                                505,
+                                95
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        81,
+                                        505,
+                                        95
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Siqi Sun, Yu Cheng, Zhe Gan, and Jingjing Liu. Patient knowledge distillation for bert model",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                93,
+                                335,
+                                106
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        93,
+                                        335,
+                                        106
+                                    ],
+                                    "score": 1.0,
+                                    "content": "compression. arXiv preprint arXiv:1908.09355, 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        }
+                    ],
+                    "index": 0.5,
+                    "bbox_fs": [
+                        106,
+                        81,
+                        505,
+                        106
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        111,
+                        504,
+                        136
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                112,
+                                506,
+                                125
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        112,
+                                        506,
+                                        125
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Zhiqing Sun, Hongkun Yu, Xiaodan Song, Renjie Liu, Yiming Yang, and Denny Zhou. Mobilebert: a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                123,
+                                505,
+                                136
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        123,
+                                        505,
+                                        136
+                                    ],
+                                    "score": 1.0,
+                                    "content": "compact task-agnostic bert for resource-limited devices. arXiv preprint arXiv:2004.02984, 2020.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        }
+                    ],
+                    "index": 2.5,
+                    "bbox_fs": [
+                        106,
+                        112,
+                        506,
+                        136
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        142,
+                        504,
+                        165
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                142,
+                                505,
+                                155
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        142,
+                                        505,
+                                        155
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Mingxing Tan and Quoc V Le. Efficientnet: Rethinking model scaling for convolutional neural",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                154,
+                                321,
+                                165
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        154,
+                                        321,
+                                        165
+                                    ],
+                                    "score": 1.0,
+                                    "content": "networks. arXiv preprint arXiv:1905.11946, 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        }
+                    ],
+                    "index": 4.5,
+                    "bbox_fs": [
+                        106,
+                        142,
+                        505,
+                        165
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        171,
+                        505,
+                        195
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                172,
+                                506,
+                                185
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        172,
+                                        506,
+                                        185
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Yonglong Tian, Dilip Krishnan, and Phillip Isola. Contrastive representation distillation. In Interna-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                183,
+                                340,
+                                195
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        183,
+                                        340,
+                                        195
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tional Conference on Learning Representations, 2019a.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        }
+                    ],
+                    "index": 6.5,
+                    "bbox_fs": [
+                        106,
+                        172,
+                        506,
+                        195
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        201,
+                        504,
+                        225
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                200,
+                                506,
+                                215
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        200,
+                                        506,
+                                        215
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Yonglong Tian, Dilip Krishnan, and Phillip Isola. Contrastive multiview coding. arXiv preprint",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                213,
+                                223,
+                                225
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        213,
+                                        223,
+                                        225
+                                    ],
+                                    "score": 1.0,
+                                    "content": "arXiv:1906.05849, 2019b.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        }
+                    ],
+                    "index": 8.5,
+                    "bbox_fs": [
+                        105,
+                        200,
+                        506,
+                        225
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        231,
+                        505,
+                        255
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                232,
+                                506,
+                                245
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        232,
+                                        506,
+                                        245
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Iulia Turc, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. Well-read students learn better:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                243,
+                                487,
+                                255
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        243,
+                                        487,
+                                        255
+                                    ],
+                                    "score": 1.0,
+                                    "content": "On the importance of pre-training compact models. arXiv preprint arXiv:1908.08962, 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        }
+                    ],
+                    "index": 10.5,
+                    "bbox_fs": [
+                        106,
+                        232,
+                        506,
+                        255
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        261,
+                        505,
+                        285
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                262,
+                                505,
+                                275
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        262,
+                                        505,
+                                        275
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Tongzhou Wang and Phillip Isola. Understanding contrastive representation learning through align-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                273,
+                                444,
+                                286
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        273,
+                                        444,
+                                        286
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ment and uniformity on the hypersphere. arXiv preprint arXiv:2005.10242, 2020.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        }
+                    ],
+                    "index": 12.5,
+                    "bbox_fs": [
+                        106,
+                        262,
+                        505,
+                        286
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        291,
+                        505,
+                        304
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                290,
+                                506,
+                                304
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        290,
+                                        506,
+                                        304
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Yuxin Wu, Alexander Kirillov, Francisco Massa, Wan-Yen Lo, and Ross Girshick. Detectron2, 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        }
+                    ],
+                    "index": 14,
+                    "bbox_fs": [
+                        106,
+                        290,
+                        506,
+                        304
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        309,
+                        504,
+                        345
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                309,
+                                506,
+                                324
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        309,
+                                        506,
+                                        324
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Zhirong Wu, Yuanjun Xiong, Stella X Yu, and Dahua Lin. Unsupervised feature learning via non-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                321,
+                                506,
+                                335
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        321,
+                                        506,
+                                        335
+                                    ],
+                                    "score": 1.0,
+                                    "content": "parametric instance discrimination. In Proceedings of the IEEE Conference on Computer Vision",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                333,
+                                308,
+                                345
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        333,
+                                        308,
+                                        345
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and Pattern Recognition, pp. 3733–3742, 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        }
+                    ],
+                    "index": 16,
+                    "bbox_fs": [
+                        105,
+                        309,
+                        506,
+                        345
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        351,
+                        504,
+                        385
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                351,
+                                506,
+                                363
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        351,
+                                        506,
+                                        363
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Jianxiong Xiao, James Hays, Krista A Ehinger, Aude Oliva, and Antonio Torralba. Sun database:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                363,
+                                506,
+                                375
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        363,
+                                        506,
+                                        375
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Large-scale scene recognition from abbey to zoo. In 2010 IEEE computer society conference on",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                373,
+                                398,
+                                386
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        373,
+                                        398,
+                                        386
+                                    ],
+                                    "score": 1.0,
+                                    "content": "computer vision and pattern recognition, pp. 3485–3492. IEEE, 2010.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        }
+                    ],
+                    "index": 19,
+                    "bbox_fs": [
+                        106,
+                        351,
+                        506,
+                        386
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        392,
+                        504,
+                        426
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                392,
+                                505,
+                                405
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        392,
+                                        505,
+                                        405
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Jianwei Yang, Devi Parikh, and Dhruv Batra. Joint unsupervised learning of deep representations",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                403,
+                                505,
+                                416
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        403,
+                                        505,
+                                        416
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and image clusters. In Proceedings of the IEEE Conference on Computer Vision and Pattern",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                415,
+                                259,
+                                427
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        415,
+                                        259,
+                                        427
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Recognition, pp. 5147–5156, 2016.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        }
+                    ],
+                    "index": 22,
+                    "bbox_fs": [
+                        105,
+                        392,
+                        505,
+                        427
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        473,
+                        506,
+                        507
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                474,
+                                506,
+                                486
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        474,
+                                        506,
+                                        486
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Sergey Zagoruyko and Nikos Komodakis. Paying more attention to attention: Improving the perfor-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                485,
+                                506,
+                                497
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        485,
+                                        506,
+                                        497
+                                    ],
+                                    "score": 1.0,
+                                    "content": "mance of convolutional neural networks via attention transfer. arXiv preprint arXiv:1612.03928,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                494,
+                                143,
+                                509
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        494,
+                                        143,
+                                        509
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2016.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        }
+                    ],
+                    "index": 25,
+                    "bbox_fs": [
+                        106,
+                        474,
+                        506,
+                        509
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        514,
+                        506,
+                        549
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                514,
+                                506,
+                                527
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        514,
+                                        506,
+                                        527
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Xiaohua Zhai, Avital Oliver, Alexander Kolesnikov, and Lucas Beyer. S4l: Self-supervised semi-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                525,
+                                507,
+                                540
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        525,
+                                        507,
+                                        540
+                                    ],
+                                    "score": 1.0,
+                                    "content": "supervised learning. In Proceedings of the IEEE international conference on computer vision, pp.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                537,
+                                191,
+                                548
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        537,
+                                        191,
+                                        548
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1476–1485, 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        }
+                    ],
+                    "index": 28,
+                    "bbox_fs": [
+                        106,
+                        514,
+                        507,
+                        548
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        555,
+                        504,
+                        590
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                555,
+                                505,
+                                569
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        555,
+                                        505,
+                                        569
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Liheng Zhang, Guo-Jun Qi, Liqiang Wang, and Jiebo Luo. Aet vs. aed: Unsupervised representation",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                566,
+                                506,
+                                580
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        566,
+                                        506,
+                                        580
+                                    ],
+                                    "score": 1.0,
+                                    "content": "learning by auto-encoding transformations rather than data. In Proceedings of the IEEE Conference",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                577,
+                                390,
+                                590
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        577,
+                                        390,
+                                        590
+                                    ],
+                                    "score": 1.0,
+                                    "content": "on Computer Vision and Pattern Recognition, pp. 2547–2555, 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        }
+                    ],
+                    "index": 31,
+                    "bbox_fs": [
+                        105,
+                        555,
+                        506,
+                        590
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        596,
+                        504,
+                        620
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                595,
+                                505,
+                                610
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        595,
+                                        505,
+                                        610
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Richard Zhang, Phillip Isola, and Alexei A Efros. Colorful image colorization. In European",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                608,
+                                362,
+                                619
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        608,
+                                        362,
+                                        619
+                                    ],
+                                    "score": 1.0,
+                                    "content": "conference on computer vision, pp. 649–666. Springer, 2016.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        }
+                    ],
+                    "index": 33.5,
+                    "bbox_fs": [
+                        105,
+                        595,
+                        505,
+                        619
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        626,
+                        504,
+                        660
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                626,
+                                506,
+                                639
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        626,
+                                        506,
+                                        639
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Xiao Zhang, Zhiyuan Fang, Yandong Wen, Zhifeng Li, and Yu Qiao. Range loss for deep face",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                637,
+                                505,
+                                650
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        637,
+                                        505,
+                                        650
+                                    ],
+                                    "score": 1.0,
+                                    "content": "recognition with long-tailed training data. In Proceedings of the IEEE International Conference on",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                648,
+                                279,
+                                661
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        648,
+                                        279,
+                                        661
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Computer Vision, pp. 5409–5418, 2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        }
+                    ],
+                    "index": 36,
+                    "bbox_fs": [
+                        106,
+                        626,
+                        506,
+                        661
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        81,
+                        183,
+                        94
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                79,
+                                185,
+                                97
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        79,
+                                        185,
+                                        97
+                                    ],
+                                    "score": 1.0,
+                                    "content": "A APPENDIX",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        }
+                    ],
+                    "index": 0
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        105,
+                        444,
+                        118
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                103,
+                                446,
+                                121
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        103,
+                                        446,
+                                        121
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We discuss more details and different hyperparameters for SEED during distillation.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        }
+                    ],
+                    "index": 1
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        131,
+                        256,
+                        142
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                131,
+                                256,
+                                143
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        131,
+                                        256,
+                                        143
+                                    ],
+                                    "score": 1.0,
+                                    "content": "A.1 PSEUDO-IMPLEMENTATIONS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        }
+                    ],
+                    "index": 2
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        151,
+                        457,
+                        163
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                149,
+                                458,
+                                165
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        149,
+                                        458,
+                                        165
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We provide pseudo-code of the SEED distillation in PyTorch Paszke et al. (2019) style:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        }
+                    ],
+                    "index": 3
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        92,
+                        173,
+                        496,
+                        646
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                177,
+                                431,
+                                191
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        177,
+                                        431,
+                                        191
+                                    ],
+                                    "score": 1.0,
+                                    "content": "‘‘‘Q: maintaining queue of previous representations: (N X D)",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                98,
+                                189,
+                                286,
+                                201
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        98,
+                                        189,
+                                        286,
+                                        201
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2 T: Cumbersome encoder as Teacher.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                97,
+                                199,
+                                266,
+                                211
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        97,
+                                        199,
+                                        266,
+                                        211
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3 S: Target encoder as Student.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                98,
+                                209,
+                                399,
+                                221
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        98,
+                                        209,
+                                        399,
+                                        221
+                                    ],
+                                    "score": 1.0,
+                                    "content": "4 temp_T, temp_S: temperatures of the Teacher & Student.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                108,
+                                219,
+                                122,
+                                226
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        108,
+                                        219,
+                                        122,
+                                        226
+                                    ],
+                                    "score": 0.715,
+                                    "content": "11",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                99,
+                                232,
+                                104,
+                                238
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        99,
+                                        232,
+                                        104,
+                                        238
+                                    ],
+                                    "score": 1.0,
+                                    "content": "6",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                101,
+                                237,
+                                457,
+                                251
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        101,
+                                        237,
+                                        457,
+                                        251
+                                    ],
+                                    "score": 1.0,
+                                    "content": "7 # activate evaluation mode for Teacher to freeze BN and updation.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                96,
+                                247,
+                                151,
+                                261
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        96,
+                                        247,
+                                        151,
+                                        261
+                                    ],
+                                    "score": 1.0,
+                                    "content": "8 T.eval()",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                98,
+                                262,
+                                105,
+                                268
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        98,
+                                        262,
+                                        105,
+                                        268
+                                    ],
+                                    "score": 1.0,
+                                    "content": "9",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                93,
+                                268,
+                                437,
+                                281
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        93,
+                                        268,
+                                        437,
+                                        281
+                                    ],
+                                    "score": 1.0,
+                                    "content": "10 for images in enumerate(loader): # Enumerate single crop-view",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                95,
+                                280,
+                                104,
+                                289
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        95,
+                                        280,
+                                        104,
+                                        289
+                                    ],
+                                    "score": 1.0,
+                                    "content": "11",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                95,
+                                288,
+                                357,
+                                301
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        95,
+                                        290,
+                                        105,
+                                        299
+                                    ],
+                                    "score": 1.0,
+                                    "content": "12",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        131,
+                                        288,
+                                        357,
+                                        301
+                                    ],
+                                    "score": 1.0,
+                                    "content": "# augment image to get one identical view",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                95,
+                                299,
+                                237,
+                                311
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        95,
+                                        300,
+                                        105,
+                                        309
+                                    ],
+                                    "score": 1.0,
+                                    "content": "13",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        127,
+                                        299,
+                                        164,
+                                        311
+                                    ],
+                                    "score": 1.0,
+                                    "content": "images",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        164,
+                                        300,
+                                        174,
+                                        308
+                                    ],
+                                    "score": 0.77,
+                                    "content": "=",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        174,
+                                        299,
+                                        237,
+                                        311
+                                    ],
+                                    "score": 1.0,
+                                    "content": "aug(images)",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                95,
+                                309,
+                                106,
+                                320
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        95,
+                                        309,
+                                        106,
+                                        320
+                                    ],
+                                    "score": 1.0,
+                                    "content": "14",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                95,
+                                318,
+                                196,
+                                330
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        95,
+                                        320,
+                                        106,
+                                        330
+                                    ],
+                                    "score": 1.0,
+                                    "content": "15",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        127,
+                                        318,
+                                        196,
+                                        330
+                                    ],
+                                    "score": 1.0,
+                                    "content": "# Batch-size",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                95,
+                                328,
+                                233,
+                                341
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        95,
+                                        330,
+                                        105,
+                                        339
+                                    ],
+                                    "score": 1.0,
+                                    "content": "16",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        128,
+                                        330,
+                                        147,
+                                        339
+                                    ],
+                                    "score": 0.35,
+                                    "content": "\\begin{array} { r l } { \\mathrm { B } } & { { } = } \\end{array}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        147,
+                                        328,
+                                        233,
+                                        341
+                                    ],
+                                    "score": 1.0,
+                                    "content": "images.shape[0]",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                95,
+                                340,
+                                105,
+                                350
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        95,
+                                        340,
+                                        105,
+                                        350
+                                    ],
+                                    "score": 1.0,
+                                    "content": "17",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                95,
+                                348,
+                                308,
+                                360
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        95,
+                                        349,
+                                        106,
+                                        360
+                                    ],
+                                    "score": 1.0,
+                                    "content": "18",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        127,
+                                        348,
+                                        308,
+                                        359
+                                    ],
+                                    "score": 1.0,
+                                    "content": "# extract embedding from S: 1 X D",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                95,
+                                358,
+                                210,
+                                370
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        95,
+                                        360,
+                                        105,
+                                        369
+                                    ],
+                                    "score": 1.0,
+                                    "content": "19",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        128,
+                                        359,
+                                        158,
+                                        369
+                                    ],
+                                    "score": 0.26,
+                                    "content": "\\mathrm { ~ \\cal ~ X ~ \\_ ~ S ~ ~ } =",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        158,
+                                        358,
+                                        210,
+                                        370
+                                    ],
+                                    "score": 1.0,
+                                    "content": "S(images)",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                94,
+                                368,
+                                307,
+                                381
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        94,
+                                        369,
+                                        106,
+                                        380
+                                    ],
+                                    "score": 1.0,
+                                    "content": "20",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        128,
+                                        370,
+                                        158,
+                                        380
+                                    ],
+                                    "score": 0.47,
+                                    "content": "\\mathrm { ~ \\cal ~ X ~ \\_ ~ S ~ ~ } =",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        159,
+                                        368,
+                                        245,
+                                        381
+                                    ],
+                                    "score": 1.0,
+                                    "content": "torch.norm(X_S,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        245,
+                                        370,
+                                        264,
+                                        380
+                                    ],
+                                    "score": 0.82,
+                                    "content": "\\mathrm { p } { = } 2",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        264,
+                                        368,
+                                        290,
+                                        381
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", dim",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        290,
+                                        370,
+                                        302,
+                                        378
+                                    ],
+                                    "score": 0.7,
+                                    "content": "^ { = 1 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        302,
+                                        368,
+                                        307,
+                                        381
+                                    ],
+                                    "score": 1.0,
+                                    "content": ")",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                95,
+                                380,
+                                105,
+                                390
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        95,
+                                        380,
+                                        105,
+                                        390
+                                    ],
+                                    "score": 1.0,
+                                    "content": "21",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                94,
+                                389,
+                                281,
+                                400
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        94,
+                                        389,
+                                        105,
+                                        400
+                                    ],
+                                    "score": 1.0,
+                                    "content": "22",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        127,
+                                        389,
+                                        281,
+                                        399
+                                    ],
+                                    "score": 1.0,
+                                    "content": "# use the gradient-free mode",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                94,
+                                398,
+                                243,
+                                410
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        94,
+                                        399,
+                                        105,
+                                        410
+                                    ],
+                                    "score": 1.0,
+                                    "content": "23",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        127,
+                                        398,
+                                        243,
+                                        410
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with torch.no_grad():",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                94,
+                                407,
+                                369,
+                                420
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        94,
+                                        409,
+                                        106,
+                                        420
+                                    ],
+                                    "score": 1.0,
+                                    "content": "24",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        147,
+                                        407,
+                                        169,
+                                        420
+                                    ],
+                                    "score": 1.0,
+                                    "content": "X_T",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        169,
+                                        410,
+                                        178,
+                                        417
+                                    ],
+                                    "score": 0.52,
+                                    "content": "=",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        179,
+                                        407,
+                                        369,
+                                        420
+                                    ],
+                                    "score": 1.0,
+                                    "content": "T(image) # embedding from T: 1 X D",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                94,
+                                418,
+                                329,
+                                430
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        94,
+                                        419,
+                                        105,
+                                        430
+                                    ],
+                                    "score": 1.0,
+                                    "content": "25",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        149,
+                                        419,
+                                        180,
+                                        430
+                                    ],
+                                    "score": 0.27,
+                                    "content": "\\mathrm { ~ X ~ \\_ ~ T ~ } =",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        180,
+                                        418,
+                                        267,
+                                        430
+                                    ],
+                                    "score": 1.0,
+                                    "content": "torch.norm(X_T,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        267,
+                                        419,
+                                        285,
+                                        429
+                                    ],
+                                    "score": 0.79,
+                                    "content": "\\mathrm { p } { = } 2",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        285,
+                                        418,
+                                        311,
+                                        430
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", dim",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        311,
+                                        419,
+                                        323,
+                                        428
+                                    ],
+                                    "score": 0.69,
+                                    "content": "^ { = 1 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        323,
+                                        418,
+                                        329,
+                                        430
+                                    ],
+                                    "score": 1.0,
+                                    "content": ")",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                94,
+                                429,
+                                106,
+                                440
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        94,
+                                        429,
+                                        106,
+                                        440
+                                    ],
+                                    "score": 1.0,
+                                    "content": "26",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                94,
+                                437,
+                                363,
+                                450
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        94,
+                                        439,
+                                        105,
+                                        450
+                                    ],
+                                    "score": 1.0,
+                                    "content": "27",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        126,
+                                        437,
+                                        363,
+                                        450
+                                    ],
+                                    "score": 1.0,
+                                    "content": "# insert the current batch embedding from T",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                94,
+                                448,
+                                211,
+                                461
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        94,
+                                        449,
+                                        105,
+                                        459
+                                    ],
+                                    "score": 1.0,
+                                    "content": "28",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        127,
+                                        448,
+                                        211,
+                                        461
+                                    ],
+                                    "score": 1.0,
+                                    "content": "enqueue(Q, X_T)",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                94,
+                                459,
+                                106,
+                                469
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        94,
+                                        459,
+                                        106,
+                                        469
+                                    ],
+                                    "score": 1.0,
+                                    "content": "29",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                94,
+                                467,
+                                425,
+                                480
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        94,
+                                        469,
+                                        106,
+                                        479
+                                    ],
+                                    "score": 1.0,
+                                    "content": "30",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        126,
+                                        467,
+                                        425,
+                                        480
+                                    ],
+                                    "score": 1.0,
+                                    "content": "# probability scores distribution for T, S: B X (N + 1)",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                95,
+                                477,
+                                496,
+                                490
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        95,
+                                        479,
+                                        104,
+                                        488
+                                    ],
+                                    "score": 1.0,
+                                    "content": "31",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        126,
+                                        477,
+                                        164,
+                                        490
+                                    ],
+                                    "score": 1.0,
+                                    "content": "S_Dist",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        165,
+                                        479,
+                                        174,
+                                        487
+                                    ],
+                                    "score": 0.8,
+                                    "content": "=",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        174,
+                                        477,
+                                        496,
+                                        490
+                                    ],
+                                    "score": 1.0,
+                                    "content": "torch.einsum(’bd, dn -> bn’, [X_S], Q.t().clone().detach())",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                94,
+                                487,
+                                496,
+                                500
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        94,
+                                        489,
+                                        105,
+                                        500
+                                    ],
+                                    "score": 1.0,
+                                    "content": "32",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        126,
+                                        487,
+                                        164,
+                                        500
+                                    ],
+                                    "score": 1.0,
+                                    "content": "T_Dist",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        165,
+                                        489,
+                                        174,
+                                        498
+                                    ],
+                                    "score": 0.8,
+                                    "content": "=",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        174,
+                                        487,
+                                        496,
+                                        500
+                                    ],
+                                    "score": 1.0,
+                                    "content": "torch.einsum(’bd, dn -> bn’, [X_T], Q.t().clone().detach())",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                95,
+                                500,
+                                105,
+                                509
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        95,
+                                        500,
+                                        105,
+                                        509
+                                    ],
+                                    "score": 1.0,
+                                    "content": "33",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                94,
+                                507,
+                                324,
+                                519
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        94,
+                                        509,
+                                        106,
+                                        519
+                                    ],
+                                    "score": 1.0,
+                                    "content": "34",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        127,
+                                        507,
+                                        324,
+                                        519
+                                    ],
+                                    "score": 1.0,
+                                    "content": "# Apply temperatures for soft-labels",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                94,
+                                516,
+                                218,
+                                531
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        94,
+                                        519,
+                                        105,
+                                        529
+                                    ],
+                                    "score": 1.0,
+                                    "content": "35",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        126,
+                                        516,
+                                        171,
+                                        531
+                                    ],
+                                    "score": 1.0,
+                                    "content": "S_Dist /",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        171,
+                                        519,
+                                        179,
+                                        527
+                                    ],
+                                    "score": 0.6,
+                                    "content": "=",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        179,
+                                        516,
+                                        218,
+                                        531
+                                    ],
+                                    "score": 1.0,
+                                    "content": "temp_S",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                94,
+                                528,
+                                334,
+                                540
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        94,
+                                        529,
+                                        106,
+                                        539
+                                    ],
+                                    "score": 1.0,
+                                    "content": "36",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        127,
+                                        528,
+                                        164,
+                                        540
+                                    ],
+                                    "score": 1.0,
+                                    "content": "T_Dist",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        164,
+                                        530,
+                                        174,
+                                        538
+                                    ],
+                                    "score": 0.78,
+                                    "content": "=",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        174,
+                                        528,
+                                        316,
+                                        540
+                                    ],
+                                    "score": 1.0,
+                                    "content": "SoftMax(T_Dist/temp_T, dim",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        317,
+                                        529,
+                                        328,
+                                        538
+                                    ],
+                                    "score": 0.69,
+                                    "content": "^ { = 1 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        329,
+                                        528,
+                                        334,
+                                        540
+                                    ],
+                                    "score": 1.0,
+                                    "content": ")",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                94,
+                                538,
+                                105,
+                                549
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        94,
+                                        538,
+                                        105,
+                                        549
+                                    ],
+                                    "score": 1.0,
+                                    "content": "37",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                95,
+                                546,
+                                443,
+                                560
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        95,
+                                        550,
+                                        105,
+                                        559
+                                    ],
+                                    "score": 1.0,
+                                    "content": "38",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        126,
+                                        546,
+                                        443,
+                                        560
+                                    ],
+                                    "score": 1.0,
+                                    "content": "# loss computation, use log_softmax for stable computation",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                94,
+                                557,
+                                460,
+                                570
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        94,
+                                        559,
+                                        105,
+                                        569
+                                    ],
+                                    "score": 1.0,
+                                    "content": "39",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        127,
+                                        557,
+                                        153,
+                                        570
+                                    ],
+                                    "score": 1.0,
+                                    "content": "loss",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        154,
+                                        559,
+                                        163,
+                                        568
+                                    ],
+                                    "score": 0.7,
+                                    "content": "=",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        164,
+                                        557,
+                                        392,
+                                        570
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-torch.mul(T_Dist, Log_SoftMax(S_Dist, dim",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        392,
+                                        559,
+                                        403,
+                                        568
+                                    ],
+                                    "score": 0.64,
+                                    "content": "^ { 1 = 1 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        404,
+                                        557,
+                                        460,
+                                        570
+                                    ],
+                                    "score": 1.0,
+                                    "content": ")).sum()/B",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                94,
+                                568,
+                                106,
+                                579
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        94,
+                                        568,
+                                        106,
+                                        579
+                                    ],
+                                    "score": 1.0,
+                                    "content": "40",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                95,
+                                577,
+                                303,
+                                590
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        95,
+                                        579,
+                                        104,
+                                        588
+                                    ],
+                                    "score": 1.0,
+                                    "content": "41",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        126,
+                                        577,
+                                        303,
+                                        590
+                                    ],
+                                    "score": 1.0,
+                                    "content": "# update the random sample queue",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                94,
+                                587,
+                                379,
+                                600
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        94,
+                                        589,
+                                        105,
+                                        599
+                                    ],
+                                    "score": 1.0,
+                                    "content": "42",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        127,
+                                        588,
+                                        207,
+                                        599
+                                    ],
+                                    "score": 1.0,
+                                    "content": "dequeue(Q, B)",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        212,
+                                        587,
+                                        379,
+                                        600
+                                    ],
+                                    "score": 1.0,
+                                    "content": "# pop-out earliest B instances",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                94,
+                                598,
+                                106,
+                                609
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        94,
+                                        598,
+                                        106,
+                                        609
+                                    ],
+                                    "score": 1.0,
+                                    "content": "43",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        },
+                        {
+                            "bbox": [
+                                94,
+                                606,
+                                207,
+                                619
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        94,
+                                        608,
+                                        105,
+                                        619
+                                    ],
+                                    "score": 1.0,
+                                    "content": "44",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        126,
+                                        606,
+                                        207,
+                                        619
+                                    ],
+                                    "score": 1.0,
+                                    "content": "# SGD updation",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 47
+                        },
+                        {
+                            "bbox": [
+                                94,
+                                617,
+                                210,
+                                629
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        94,
+                                        618,
+                                        105,
+                                        629
+                                    ],
+                                    "score": 1.0,
+                                    "content": "45",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        127,
+                                        617,
+                                        210,
+                                        629
+                                    ],
+                                    "score": 1.0,
+                                    "content": "loss.backward()",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 48
+                        },
+                        {
+                            "bbox": [
+                                94,
+                                627,
+                                216,
+                                639
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        94,
+                                        628,
+                                        106,
+                                        639
+                                    ],
+                                    "score": 1.0,
+                                    "content": "46",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        126,
+                                        627,
+                                        216,
+                                        639
+                                    ],
+                                    "score": 1.0,
+                                    "content": "update(S.params)",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 49
+                        }
+                    ],
+                    "index": 26.5
+                }
+            ],
+            "page_idx": 12,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        107,
+                        26,
+                        292,
+                        38
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                25,
+                                293,
+                                39
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        25,
+                                        293,
+                                        39
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Published as a conference paper at ICLR 2021",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        300,
+                        751,
+                        311,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                299,
+                                750,
+                                312,
+                                764
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        299,
+                                        750,
+                                        312,
+                                        764
+                                    ],
+                                    "score": 1.0,
+                                    "content": "13",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        81,
+                        183,
+                        94
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                79,
+                                185,
+                                97
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        79,
+                                        185,
+                                        97
+                                    ],
+                                    "score": 1.0,
+                                    "content": "A APPENDIX",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        }
+                    ],
+                    "index": 0
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        105,
+                        444,
+                        118
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                103,
+                                446,
+                                121
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        103,
+                                        446,
+                                        121
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We discuss more details and different hyperparameters for SEED during distillation.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        }
+                    ],
+                    "index": 1,
+                    "bbox_fs": [
+                        104,
+                        103,
+                        446,
+                        121
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        131,
+                        256,
+                        142
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                131,
+                                256,
+                                143
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        131,
+                                        256,
+                                        143
+                                    ],
+                                    "score": 1.0,
+                                    "content": "A.1 PSEUDO-IMPLEMENTATIONS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        }
+                    ],
+                    "index": 2,
+                    "bbox_fs": [
+                        106,
+                        131,
+                        256,
+                        143
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        151,
+                        457,
+                        163
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                149,
+                                458,
+                                165
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        149,
+                                        458,
+                                        165
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We provide pseudo-code of the SEED distillation in PyTorch Paszke et al. (2019) style:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        }
+                    ],
+                    "index": 3,
+                    "bbox_fs": [
+                        105,
+                        149,
+                        458,
+                        165
+                    ]
+                },
+                {
+                    "type": "index",
+                    "bbox": [
+                        92,
+                        173,
+                        496,
+                        646
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                177,
+                                431,
+                                191
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        177,
+                                        431,
+                                        191
+                                    ],
+                                    "score": 1.0,
+                                    "content": "‘‘‘Q: maintaining queue of previous representations: (N X D)",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                98,
+                                189,
+                                286,
+                                201
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        98,
+                                        189,
+                                        286,
+                                        201
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2 T: Cumbersome encoder as Teacher.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                97,
+                                199,
+                                266,
+                                211
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        97,
+                                        199,
+                                        266,
+                                        211
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3 S: Target encoder as Student.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                98,
+                                209,
+                                399,
+                                221
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        98,
+                                        209,
+                                        399,
+                                        221
+                                    ],
+                                    "score": 1.0,
+                                    "content": "4 temp_T, temp_S: temperatures of the Teacher & Student.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                108,
+                                219,
+                                122,
+                                226
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        108,
+                                        219,
+                                        122,
+                                        226
+                                    ],
+                                    "score": 0.715,
+                                    "content": "11",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                99,
+                                232,
+                                104,
+                                238
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        99,
+                                        232,
+                                        104,
+                                        238
+                                    ],
+                                    "score": 1.0,
+                                    "content": "6",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                101,
+                                237,
+                                457,
+                                251
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        101,
+                                        237,
+                                        457,
+                                        251
+                                    ],
+                                    "score": 1.0,
+                                    "content": "7 # activate evaluation mode for Teacher to freeze BN and updation.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                96,
+                                247,
+                                151,
+                                261
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        96,
+                                        247,
+                                        151,
+                                        261
+                                    ],
+                                    "score": 1.0,
+                                    "content": "8 T.eval()",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                98,
+                                262,
+                                105,
+                                268
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        98,
+                                        262,
+                                        105,
+                                        268
+                                    ],
+                                    "score": 1.0,
+                                    "content": "9",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                93,
+                                268,
+                                437,
+                                281
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        93,
+                                        268,
+                                        437,
+                                        281
+                                    ],
+                                    "score": 1.0,
+                                    "content": "10 for images in enumerate(loader): # Enumerate single crop-view",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                95,
+                                280,
+                                104,
+                                289
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        95,
+                                        280,
+                                        104,
+                                        289
+                                    ],
+                                    "score": 1.0,
+                                    "content": "11",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                95,
+                                288,
+                                357,
+                                301
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        95,
+                                        290,
+                                        105,
+                                        299
+                                    ],
+                                    "score": 1.0,
+                                    "content": "12",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        131,
+                                        288,
+                                        357,
+                                        301
+                                    ],
+                                    "score": 1.0,
+                                    "content": "# augment image to get one identical view",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                95,
+                                299,
+                                237,
+                                311
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        95,
+                                        300,
+                                        105,
+                                        309
+                                    ],
+                                    "score": 1.0,
+                                    "content": "13",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        127,
+                                        299,
+                                        164,
+                                        311
+                                    ],
+                                    "score": 1.0,
+                                    "content": "images",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        164,
+                                        300,
+                                        174,
+                                        308
+                                    ],
+                                    "score": 0.77,
+                                    "content": "=",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        174,
+                                        299,
+                                        237,
+                                        311
+                                    ],
+                                    "score": 1.0,
+                                    "content": "aug(images)",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                95,
+                                309,
+                                106,
+                                320
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        95,
+                                        309,
+                                        106,
+                                        320
+                                    ],
+                                    "score": 1.0,
+                                    "content": "14",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                95,
+                                318,
+                                196,
+                                330
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        95,
+                                        320,
+                                        106,
+                                        330
+                                    ],
+                                    "score": 1.0,
+                                    "content": "15",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        127,
+                                        318,
+                                        196,
+                                        330
+                                    ],
+                                    "score": 1.0,
+                                    "content": "# Batch-size",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                95,
+                                328,
+                                233,
+                                341
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        95,
+                                        330,
+                                        105,
+                                        339
+                                    ],
+                                    "score": 1.0,
+                                    "content": "16",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        128,
+                                        330,
+                                        147,
+                                        339
+                                    ],
+                                    "score": 0.35,
+                                    "content": "\\begin{array} { r l } { \\mathrm { B } } & { { } = } \\end{array}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        147,
+                                        328,
+                                        233,
+                                        341
+                                    ],
+                                    "score": 1.0,
+                                    "content": "images.shape[0]",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                95,
+                                340,
+                                105,
+                                350
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        95,
+                                        340,
+                                        105,
+                                        350
+                                    ],
+                                    "score": 1.0,
+                                    "content": "17",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                95,
+                                348,
+                                308,
+                                360
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        95,
+                                        349,
+                                        106,
+                                        360
+                                    ],
+                                    "score": 1.0,
+                                    "content": "18",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        127,
+                                        348,
+                                        308,
+                                        359
+                                    ],
+                                    "score": 1.0,
+                                    "content": "# extract embedding from S: 1 X D",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                95,
+                                358,
+                                210,
+                                370
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        95,
+                                        360,
+                                        105,
+                                        369
+                                    ],
+                                    "score": 1.0,
+                                    "content": "19",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        128,
+                                        359,
+                                        158,
+                                        369
+                                    ],
+                                    "score": 0.26,
+                                    "content": "\\mathrm { ~ \\cal ~ X ~ \\_ ~ S ~ ~ } =",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        158,
+                                        358,
+                                        210,
+                                        370
+                                    ],
+                                    "score": 1.0,
+                                    "content": "S(images)",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                94,
+                                368,
+                                307,
+                                381
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        94,
+                                        369,
+                                        106,
+                                        380
+                                    ],
+                                    "score": 1.0,
+                                    "content": "20",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        128,
+                                        370,
+                                        158,
+                                        380
+                                    ],
+                                    "score": 0.47,
+                                    "content": "\\mathrm { ~ \\cal ~ X ~ \\_ ~ S ~ ~ } =",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        159,
+                                        368,
+                                        245,
+                                        381
+                                    ],
+                                    "score": 1.0,
+                                    "content": "torch.norm(X_S,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        245,
+                                        370,
+                                        264,
+                                        380
+                                    ],
+                                    "score": 0.82,
+                                    "content": "\\mathrm { p } { = } 2",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        264,
+                                        368,
+                                        290,
+                                        381
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", dim",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        290,
+                                        370,
+                                        302,
+                                        378
+                                    ],
+                                    "score": 0.7,
+                                    "content": "^ { = 1 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        302,
+                                        368,
+                                        307,
+                                        381
+                                    ],
+                                    "score": 1.0,
+                                    "content": ")",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                95,
+                                380,
+                                105,
+                                390
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        95,
+                                        380,
+                                        105,
+                                        390
+                                    ],
+                                    "score": 1.0,
+                                    "content": "21",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                94,
+                                389,
+                                281,
+                                400
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        94,
+                                        389,
+                                        105,
+                                        400
+                                    ],
+                                    "score": 1.0,
+                                    "content": "22",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        127,
+                                        389,
+                                        281,
+                                        399
+                                    ],
+                                    "score": 1.0,
+                                    "content": "# use the gradient-free mode",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                94,
+                                398,
+                                243,
+                                410
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        94,
+                                        399,
+                                        105,
+                                        410
+                                    ],
+                                    "score": 1.0,
+                                    "content": "23",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        127,
+                                        398,
+                                        243,
+                                        410
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with torch.no_grad():",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                94,
+                                407,
+                                369,
+                                420
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        94,
+                                        409,
+                                        106,
+                                        420
+                                    ],
+                                    "score": 1.0,
+                                    "content": "24",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        147,
+                                        407,
+                                        169,
+                                        420
+                                    ],
+                                    "score": 1.0,
+                                    "content": "X_T",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        169,
+                                        410,
+                                        178,
+                                        417
+                                    ],
+                                    "score": 0.52,
+                                    "content": "=",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        179,
+                                        407,
+                                        369,
+                                        420
+                                    ],
+                                    "score": 1.0,
+                                    "content": "T(image) # embedding from T: 1 X D",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                94,
+                                418,
+                                329,
+                                430
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        94,
+                                        419,
+                                        105,
+                                        430
+                                    ],
+                                    "score": 1.0,
+                                    "content": "25",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        149,
+                                        419,
+                                        180,
+                                        430
+                                    ],
+                                    "score": 0.27,
+                                    "content": "\\mathrm { ~ X ~ \\_ ~ T ~ } =",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        180,
+                                        418,
+                                        267,
+                                        430
+                                    ],
+                                    "score": 1.0,
+                                    "content": "torch.norm(X_T,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        267,
+                                        419,
+                                        285,
+                                        429
+                                    ],
+                                    "score": 0.79,
+                                    "content": "\\mathrm { p } { = } 2",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        285,
+                                        418,
+                                        311,
+                                        430
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", dim",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        311,
+                                        419,
+                                        323,
+                                        428
+                                    ],
+                                    "score": 0.69,
+                                    "content": "^ { = 1 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        323,
+                                        418,
+                                        329,
+                                        430
+                                    ],
+                                    "score": 1.0,
+                                    "content": ")",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                94,
+                                429,
+                                106,
+                                440
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        94,
+                                        429,
+                                        106,
+                                        440
+                                    ],
+                                    "score": 1.0,
+                                    "content": "26",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                94,
+                                437,
+                                363,
+                                450
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        94,
+                                        439,
+                                        105,
+                                        450
+                                    ],
+                                    "score": 1.0,
+                                    "content": "27",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        126,
+                                        437,
+                                        363,
+                                        450
+                                    ],
+                                    "score": 1.0,
+                                    "content": "# insert the current batch embedding from T",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                94,
+                                448,
+                                211,
+                                461
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        94,
+                                        449,
+                                        105,
+                                        459
+                                    ],
+                                    "score": 1.0,
+                                    "content": "28",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        127,
+                                        448,
+                                        211,
+                                        461
+                                    ],
+                                    "score": 1.0,
+                                    "content": "enqueue(Q, X_T)",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                94,
+                                459,
+                                106,
+                                469
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        94,
+                                        459,
+                                        106,
+                                        469
+                                    ],
+                                    "score": 1.0,
+                                    "content": "29",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                94,
+                                467,
+                                425,
+                                480
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        94,
+                                        469,
+                                        106,
+                                        479
+                                    ],
+                                    "score": 1.0,
+                                    "content": "30",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        126,
+                                        467,
+                                        425,
+                                        480
+                                    ],
+                                    "score": 1.0,
+                                    "content": "# probability scores distribution for T, S: B X (N + 1)",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                95,
+                                477,
+                                496,
+                                490
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        95,
+                                        479,
+                                        104,
+                                        488
+                                    ],
+                                    "score": 1.0,
+                                    "content": "31",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        126,
+                                        477,
+                                        164,
+                                        490
+                                    ],
+                                    "score": 1.0,
+                                    "content": "S_Dist",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        165,
+                                        479,
+                                        174,
+                                        487
+                                    ],
+                                    "score": 0.8,
+                                    "content": "=",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        174,
+                                        477,
+                                        496,
+                                        490
+                                    ],
+                                    "score": 1.0,
+                                    "content": "torch.einsum(’bd, dn -> bn’, [X_S], Q.t().clone().detach())",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                94,
+                                487,
+                                496,
+                                500
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        94,
+                                        489,
+                                        105,
+                                        500
+                                    ],
+                                    "score": 1.0,
+                                    "content": "32",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        126,
+                                        487,
+                                        164,
+                                        500
+                                    ],
+                                    "score": 1.0,
+                                    "content": "T_Dist",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        165,
+                                        489,
+                                        174,
+                                        498
+                                    ],
+                                    "score": 0.8,
+                                    "content": "=",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        174,
+                                        487,
+                                        496,
+                                        500
+                                    ],
+                                    "score": 1.0,
+                                    "content": "torch.einsum(’bd, dn -> bn’, [X_T], Q.t().clone().detach())",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                95,
+                                500,
+                                105,
+                                509
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        95,
+                                        500,
+                                        105,
+                                        509
+                                    ],
+                                    "score": 1.0,
+                                    "content": "33",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                94,
+                                507,
+                                324,
+                                519
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        94,
+                                        509,
+                                        106,
+                                        519
+                                    ],
+                                    "score": 1.0,
+                                    "content": "34",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        127,
+                                        507,
+                                        324,
+                                        519
+                                    ],
+                                    "score": 1.0,
+                                    "content": "# Apply temperatures for soft-labels",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                94,
+                                516,
+                                218,
+                                531
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        94,
+                                        519,
+                                        105,
+                                        529
+                                    ],
+                                    "score": 1.0,
+                                    "content": "35",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        126,
+                                        516,
+                                        171,
+                                        531
+                                    ],
+                                    "score": 1.0,
+                                    "content": "S_Dist /",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        171,
+                                        519,
+                                        179,
+                                        527
+                                    ],
+                                    "score": 0.6,
+                                    "content": "=",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        179,
+                                        516,
+                                        218,
+                                        531
+                                    ],
+                                    "score": 1.0,
+                                    "content": "temp_S",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                94,
+                                528,
+                                334,
+                                540
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        94,
+                                        529,
+                                        106,
+                                        539
+                                    ],
+                                    "score": 1.0,
+                                    "content": "36",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        127,
+                                        528,
+                                        164,
+                                        540
+                                    ],
+                                    "score": 1.0,
+                                    "content": "T_Dist",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        164,
+                                        530,
+                                        174,
+                                        538
+                                    ],
+                                    "score": 0.78,
+                                    "content": "=",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        174,
+                                        528,
+                                        316,
+                                        540
+                                    ],
+                                    "score": 1.0,
+                                    "content": "SoftMax(T_Dist/temp_T, dim",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        317,
+                                        529,
+                                        328,
+                                        538
+                                    ],
+                                    "score": 0.69,
+                                    "content": "^ { = 1 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        329,
+                                        528,
+                                        334,
+                                        540
+                                    ],
+                                    "score": 1.0,
+                                    "content": ")",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                94,
+                                538,
+                                105,
+                                549
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        94,
+                                        538,
+                                        105,
+                                        549
+                                    ],
+                                    "score": 1.0,
+                                    "content": "37",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                95,
+                                546,
+                                443,
+                                560
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        95,
+                                        550,
+                                        105,
+                                        559
+                                    ],
+                                    "score": 1.0,
+                                    "content": "38",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        126,
+                                        546,
+                                        443,
+                                        560
+                                    ],
+                                    "score": 1.0,
+                                    "content": "# loss computation, use log_softmax for stable computation",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                94,
+                                557,
+                                460,
+                                570
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        94,
+                                        559,
+                                        105,
+                                        569
+                                    ],
+                                    "score": 1.0,
+                                    "content": "39",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        127,
+                                        557,
+                                        153,
+                                        570
+                                    ],
+                                    "score": 1.0,
+                                    "content": "loss",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        154,
+                                        559,
+                                        163,
+                                        568
+                                    ],
+                                    "score": 0.7,
+                                    "content": "=",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        164,
+                                        557,
+                                        392,
+                                        570
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-torch.mul(T_Dist, Log_SoftMax(S_Dist, dim",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        392,
+                                        559,
+                                        403,
+                                        568
+                                    ],
+                                    "score": 0.64,
+                                    "content": "^ { 1 = 1 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        404,
+                                        557,
+                                        460,
+                                        570
+                                    ],
+                                    "score": 1.0,
+                                    "content": ")).sum()/B",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                94,
+                                568,
+                                106,
+                                579
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        94,
+                                        568,
+                                        106,
+                                        579
+                                    ],
+                                    "score": 1.0,
+                                    "content": "40",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                95,
+                                577,
+                                303,
+                                590
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        95,
+                                        579,
+                                        104,
+                                        588
+                                    ],
+                                    "score": 1.0,
+                                    "content": "41",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        126,
+                                        577,
+                                        303,
+                                        590
+                                    ],
+                                    "score": 1.0,
+                                    "content": "# update the random sample queue",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                94,
+                                587,
+                                379,
+                                600
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        94,
+                                        589,
+                                        105,
+                                        599
+                                    ],
+                                    "score": 1.0,
+                                    "content": "42",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        127,
+                                        588,
+                                        207,
+                                        599
+                                    ],
+                                    "score": 1.0,
+                                    "content": "dequeue(Q, B)",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        212,
+                                        587,
+                                        379,
+                                        600
+                                    ],
+                                    "score": 1.0,
+                                    "content": "# pop-out earliest B instances",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                94,
+                                598,
+                                106,
+                                609
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        94,
+                                        598,
+                                        106,
+                                        609
+                                    ],
+                                    "score": 1.0,
+                                    "content": "43",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                94,
+                                606,
+                                207,
+                                619
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        94,
+                                        608,
+                                        105,
+                                        619
+                                    ],
+                                    "score": 1.0,
+                                    "content": "44",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        126,
+                                        606,
+                                        207,
+                                        619
+                                    ],
+                                    "score": 1.0,
+                                    "content": "# SGD updation",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 47,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                94,
+                                617,
+                                210,
+                                629
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        94,
+                                        618,
+                                        105,
+                                        629
+                                    ],
+                                    "score": 1.0,
+                                    "content": "45",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        127,
+                                        617,
+                                        210,
+                                        629
+                                    ],
+                                    "score": 1.0,
+                                    "content": "loss.backward()",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 48,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                94,
+                                627,
+                                216,
+                                639
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        94,
+                                        628,
+                                        106,
+                                        639
+                                    ],
+                                    "score": 1.0,
+                                    "content": "46",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        126,
+                                        627,
+                                        216,
+                                        639
+                                    ],
+                                    "score": 1.0,
+                                    "content": "update(S.params)",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 49,
+                            "is_list_start_line": true
+                        }
+                    ],
+                    "index": 26.5,
+                    "bbox_fs": [
+                        93,
+                        177,
+                        496,
+                        639
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        82,
+                        244,
+                        94
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                81,
+                                245,
+                                95
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        81,
+                                        245,
+                                        95
+                                    ],
+                                    "score": 1.0,
+                                    "content": "A.1.1 DATA AUGMENTATIONS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        }
+                    ],
+                    "index": 0
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        101,
+                        450,
+                        113
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                101,
+                                452,
+                                115
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        101,
+                                        452,
+                                        115
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Both our teacher pre-training and distillation adopt the data augmentations as follows:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        }
+                    ],
+                    "index": 1
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        136,
+                        121,
+                        504,
+                        144
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                138,
+                                120,
+                                505,
+                                135
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        138,
+                                        120,
+                                        505,
+                                        135
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Random Resized Crop: The image is randomly resized with a scale of {0.2, 1.0}, then",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                138,
+                                132,
+                                271,
+                                145
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        138,
+                                        132,
+                                        227,
+                                        145
+                                    ],
+                                    "score": 1.0,
+                                    "content": "cropped to the size of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        227,
+                                        133,
+                                        266,
+                                        143
+                                    ],
+                                    "score": 0.91,
+                                    "content": "2 2 4 \\times 2 2 4",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        267,
+                                        132,
+                                        271,
+                                        145
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        }
+                    ],
+                    "index": 2.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        133,
+                        146,
+                        496,
+                        159
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                138,
+                                146,
+                                496,
+                                160
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        138,
+                                        146,
+                                        496,
+                                        160
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Random Color Jittering: with brightness to be {0.4, 0.4, 0.4, 0.1} with probability at 0.8.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        }
+                    ],
+                    "index": 4
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        137,
+                        162,
+                        378,
+                        173
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                138,
+                                160,
+                                380,
+                                174
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        138,
+                                        160,
+                                        380,
+                                        174
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Random Gray Scale transformation: with probability at 0.2.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        }
+                    ],
+                    "index": 5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        140,
+                        176,
+                        468,
+                        187
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                138,
+                                174,
+                                469,
+                                189
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        138,
+                                        174,
+                                        320,
+                                        189
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Random Gaussian Blur transformation: with",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        320,
+                                        176,
+                                        377,
+                                        187
+                                    ],
+                                    "score": 0.9,
+                                    "content": "\\sigma = \\{ 0 . 1 , 0 . 2 \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        378,
+                                        174,
+                                        469,
+                                        189
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and probability at 0.5.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        }
+                    ],
+                    "index": 6
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        141,
+                        190,
+                        397,
+                        202
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                138,
+                                189,
+                                399,
+                                204
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        138,
+                                        189,
+                                        399,
+                                        204
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Horizontal Flip: Horizontal flip is applied with probability at 0.5",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        }
+                    ],
+                    "index": 7
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        109,
+                        213,
+                        449,
+                        225
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                212,
+                                450,
+                                226
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        212,
+                                        450,
+                                        226
+                                    ],
+                                    "score": 1.0,
+                                    "content": "A.1.2 PRE-TRAINING AND DISTILLATION ON MOBILENET AND EFFICIENTNET",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        }
+                    ],
+                    "index": 8
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        232,
+                        506,
+                        299
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                232,
+                                505,
+                                244
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        232,
+                                        505,
+                                        244
+                                    ],
+                                    "score": 1.0,
+                                    "content": "MobileNet (Howard et al., 2017) and EfficientNet (Tan & Le, 2019) have been considered as the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                244,
+                                505,
+                                255
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        244,
+                                        505,
+                                        255
+                                    ],
+                                    "score": 1.0,
+                                    "content": "smaller counterparts with larger models, i.e., ResNet-50 (with supervised training, EfficientNet-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                254,
+                                507,
+                                267
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        254,
+                                        140,
+                                        267
+                                    ],
+                                    "score": 1.0,
+                                    "content": "B0 hits",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        140,
+                                        255,
+                                        168,
+                                        265
+                                    ],
+                                    "score": 0.86,
+                                    "content": "7 7 . 2 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        168,
+                                        254,
+                                        362,
+                                        267
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Top-1 Acc., and MobileNet-V3-large reaches",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        362,
+                                        255,
+                                        390,
+                                        265
+                                    ],
+                                    "score": 0.88,
+                                    "content": "7 2 . 2 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        391,
+                                        254,
+                                        507,
+                                        267
+                                    ],
+                                    "score": 1.0,
+                                    "content": "on ImageNet testing split).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                264,
+                                506,
+                                279
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        264,
+                                        506,
+                                        279
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Nevertheless, un-matched performances are observed in the task of self-supervised contrastive pre-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                276,
+                                506,
+                                289
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        276,
+                                        432,
+                                        289
+                                    ],
+                                    "score": 1.0,
+                                    "content": "training: i.e., Self-Supervised Learning (MoCo-V2) on MobileNet-V3 only yields",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        432,
+                                        277,
+                                        459,
+                                        288
+                                    ],
+                                    "score": 0.84,
+                                    "content": "3 6 . 3 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        459,
+                                        276,
+                                        506,
+                                        289
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Top-1 Acc.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                288,
+                                415,
+                                299
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        288,
+                                        415,
+                                        299
+                                    ],
+                                    "score": 1.0,
+                                    "content": "on ImageNet. We conjecture that several reasons might lead to this dilemma:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        }
+                    ],
+                    "index": 11.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        129,
+                        307,
+                        506,
+                        377
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                130,
+                                308,
+                                505,
+                                319
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        130,
+                                        308,
+                                        505,
+                                        319
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1. The inability of models with less parameters for handling large volume of categories and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                317,
+                                506,
+                                331
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        317,
+                                        506,
+                                        331
+                                    ],
+                                    "score": 1.0,
+                                    "content": "data, which exists also in other domains, i.e., face recognition (Guo et al., 2016; Zhang et al.,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                329,
+                                171,
+                                342
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        329,
+                                        171,
+                                        342
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2017).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                129,
+                                343,
+                                504,
+                                355
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        129,
+                                        343,
+                                        504,
+                                        355
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2. Less possibility for optimum parameters to be chosen when transferring to downstream",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                354,
+                                506,
+                                367
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        354,
+                                        506,
+                                        367
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tasks: models with more parameters after pre-training might produce a plenty cornucopia of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                142,
+                                366,
+                                289,
+                                379
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        142,
+                                        366,
+                                        289,
+                                        379
+                                    ],
+                                    "score": 1.0,
+                                    "content": "optimum parameters for fine-tuning.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        }
+                    ],
+                    "index": 17.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        385,
+                        505,
+                        441
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                386,
+                                504,
+                                398
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        386,
+                                        504,
+                                        398
+                                    ],
+                                    "score": 1.0,
+                                    "content": "To narrow the dramatic performance gap between smaller architectures using contrastive SSL with",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                396,
+                                505,
+                                410
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        396,
+                                        505,
+                                        410
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the larger, we explore with architectural manipulations and training hyper-parameters. In specific, we",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                407,
+                                506,
+                                421
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        407,
+                                        506,
+                                        421
+                                    ],
+                                    "score": 1.0,
+                                    "content": "find that by adding a deeper projection head largely improves the representation quality, a.k.a., better",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                418,
+                                506,
+                                431
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        418,
+                                        506,
+                                        431
+                                    ],
+                                    "score": 1.0,
+                                    "content": "performances on linear evaluation. We experiment with adding one additional linear projection head",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                429,
+                                263,
+                                442
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        429,
+                                        263,
+                                        442
+                                    ],
+                                    "score": 1.0,
+                                    "content": "on the top of convolutional backbones.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        }
+                    ],
+                    "index": 23
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        446,
+                        505,
+                        535
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                446,
+                                505,
+                                458
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        446,
+                                        505,
+                                        458
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Similarly, we also expand the MLP projection head on EfficientNet-b0. Though recent work shows",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                458,
+                                505,
+                                469
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        458,
+                                        505,
+                                        469
+                                    ],
+                                    "score": 1.0,
+                                    "content": "that fine-tuning from a middle layer of the projection head can produce a largely different result (Chen",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                468,
+                                505,
+                                481
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        468,
+                                        505,
+                                        481
+                                    ],
+                                    "score": 1.0,
+                                    "content": "et al., 2020b), we consistently just use the representations from convolutional trunk without adding",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                479,
+                                505,
+                                492
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        479,
+                                        505,
+                                        492
+                                    ],
+                                    "score": 1.0,
+                                    "content": "extra layers during the phase of linear evaluation. As shown in Table 6, pre-training with a deeper",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                490,
+                                506,
+                                503
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        490,
+                                        435,
+                                        503
+                                    ],
+                                    "score": 1.0,
+                                    "content": "projection head dramatically helps the improvement on linear evaluations, adding",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        436,
+                                        491,
+                                        455,
+                                        501
+                                    ],
+                                    "score": 0.84,
+                                    "content": "17 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        456,
+                                        490,
+                                        506,
+                                        503
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Top-1. Acc.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                500,
+                                505,
+                                513
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        500,
+                                        505,
+                                        513
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for Mobile-v3-large, and we report the improved baselines in the main paper (see the first row in",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                512,
+                                505,
+                                525
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        512,
+                                        505,
+                                        525
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Table 1 of the main paper). We keep most of the hyper-parameters as the distillation on ResNet except",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                523,
+                                478,
+                                535
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        523,
+                                        478,
+                                        535
+                                    ],
+                                    "score": 1.0,
+                                    "content": "reducing the weight-decay of them to 1e-5, following (Tan & Le, 2019; Sandler et al., 2018).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        }
+                    ],
+                    "index": 29.5
+                },
+                {
+                    "type": "table",
+                    "bbox": [
+                        184,
+                        579,
+                        424,
+                        646
+                    ],
+                    "blocks": [
+                        {
+                            "type": "table_caption",
+                            "bbox": [
+                                104,
+                                549,
+                                504,
+                                571
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        548,
+                                        505,
+                                        562
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                548,
+                                                505,
+                                                562
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Table 6: Linear evaluations on ImageNet of EfficientNet and MobileNet pre-trained using MoCo-v2. A deeper",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 34
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        559,
+                                        433,
+                                        572
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                559,
+                                                433,
+                                                572
+                                            ],
+                                            "score": 1.0,
+                                            "content": "projection head largely boosts the linear evaluation performances on smaller architectures.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 35
+                                }
+                            ],
+                            "index": 34.5
+                        },
+                        {
+                            "type": "table_body",
+                            "bbox": [
+                                184,
+                                579,
+                                424,
+                                646
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        184,
+                                        579,
+                                        424,
+                                        646
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                184,
+                                                579,
+                                                424,
+                                                646
+                                            ],
+                                            "score": 0.978,
+                                            "html": "<table><tr><td>Model</td><td>Deeper MLPs</td><td>Top-1 Acc. Top-5 Acc.</td><td></td></tr><tr><td>EfficientNet-b0</td><td></td><td>39.1</td><td>64.6</td></tr><tr><td>EfficientNet-b0</td><td>×</td><td>42.2</td><td>68.5</td></tr><tr><td>Mobile-v3-large</td><td>X</td><td>19.0</td><td>41.3</td></tr><tr><td>Mobile-v3-large</td><td>√</td><td>36.3</td><td>62.2</td></tr></table>",
+                                            "type": "table",
+                                            "image_path": "637ac61f5f9e5741399531cd7141a325efafa86d41a01e1c7e857e7176d2e6b6.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 38,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        184,
+                                        579,
+                                        424,
+                                        592.4
+                                    ],
+                                    "spans": [],
+                                    "index": 36
+                                },
+                                {
+                                    "bbox": [
+                                        184,
+                                        592.4,
+                                        424,
+                                        605.8
+                                    ],
+                                    "spans": [],
+                                    "index": 37
+                                },
+                                {
+                                    "bbox": [
+                                        184,
+                                        605.8,
+                                        424,
+                                        619.1999999999999
+                                    ],
+                                    "spans": [],
+                                    "index": 38
+                                },
+                                {
+                                    "bbox": [
+                                        184,
+                                        619.1999999999999,
+                                        424,
+                                        632.5999999999999
+                                    ],
+                                    "spans": [],
+                                    "index": 39
+                                },
+                                {
+                                    "bbox": [
+                                        184,
+                                        632.5999999999999,
+                                        424,
+                                        645.9999999999999
+                                    ],
+                                    "spans": [],
+                                    "index": 40
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 36.25
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        672,
+                        307,
+                        684
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                672,
+                                307,
+                                685
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        672,
+                                        307,
+                                        685
+                                    ],
+                                    "score": 1.0,
+                                    "content": "A.2 ADDITIONAL DETAILS OF EVALUATIONS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        }
+                    ],
+                    "index": 41
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        692,
+                        417,
+                        704
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                691,
+                                418,
+                                706
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        691,
+                                        418,
+                                        706
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We list additional details regarding our evaluation experiments in this section.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        }
+                    ],
+                    "index": 42
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        709,
+                        507,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                709,
+                                506,
+                                722
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        709,
+                                        506,
+                                        722
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ImageNet-1k Semi-Supervised Linear Evaluation. Following Zhai et al. (2019); Chen et al.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                720,
+                                505,
+                                732
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        720,
+                                        505,
+                                        732
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(2020a), we train the FC layers on the basis of our student encoder after distillation using a fraction",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        }
+                    ],
+                    "index": 43.5
+                }
+            ],
+            "page_idx": 13,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        108,
+                        27,
+                        293,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                26,
+                                293,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        26,
+                                        293,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Published as a conference paper at ICLR 2021",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        300,
+                        751,
+                        310,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                299,
+                                750,
+                                312,
+                                764
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        299,
+                                        750,
+                                        312,
+                                        764
+                                    ],
+                                    "score": 1.0,
+                                    "content": "",
+                                    "type": "text",
+                                    "height": 14,
+                                    "width": 13
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        82,
+                        244,
+                        94
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                81,
+                                245,
+                                95
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        81,
+                                        245,
+                                        95
+                                    ],
+                                    "score": 1.0,
+                                    "content": "A.1.1 DATA AUGMENTATIONS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        }
+                    ],
+                    "index": 0
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        101,
+                        450,
+                        113
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                101,
+                                452,
+                                115
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        101,
+                                        452,
+                                        115
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Both our teacher pre-training and distillation adopt the data augmentations as follows:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        }
+                    ],
+                    "index": 1,
+                    "bbox_fs": [
+                        105,
+                        101,
+                        452,
+                        115
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        136,
+                        121,
+                        504,
+                        144
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                138,
+                                120,
+                                505,
+                                135
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        138,
+                                        120,
+                                        505,
+                                        135
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Random Resized Crop: The image is randomly resized with a scale of {0.2, 1.0}, then",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                138,
+                                132,
+                                271,
+                                145
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        138,
+                                        132,
+                                        227,
+                                        145
+                                    ],
+                                    "score": 1.0,
+                                    "content": "cropped to the size of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        227,
+                                        133,
+                                        266,
+                                        143
+                                    ],
+                                    "score": 0.91,
+                                    "content": "2 2 4 \\times 2 2 4",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        267,
+                                        132,
+                                        271,
+                                        145
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        }
+                    ],
+                    "index": 2.5,
+                    "bbox_fs": [
+                        138,
+                        120,
+                        505,
+                        145
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        133,
+                        146,
+                        496,
+                        159
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                138,
+                                146,
+                                496,
+                                160
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        138,
+                                        146,
+                                        496,
+                                        160
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Random Color Jittering: with brightness to be {0.4, 0.4, 0.4, 0.1} with probability at 0.8.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        }
+                    ],
+                    "index": 4,
+                    "bbox_fs": [
+                        138,
+                        146,
+                        496,
+                        160
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        137,
+                        162,
+                        378,
+                        173
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                138,
+                                160,
+                                380,
+                                174
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        138,
+                                        160,
+                                        380,
+                                        174
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Random Gray Scale transformation: with probability at 0.2.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        }
+                    ],
+                    "index": 5,
+                    "bbox_fs": [
+                        138,
+                        160,
+                        380,
+                        174
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        140,
+                        176,
+                        468,
+                        187
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                138,
+                                174,
+                                469,
+                                189
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        138,
+                                        174,
+                                        320,
+                                        189
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Random Gaussian Blur transformation: with",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        320,
+                                        176,
+                                        377,
+                                        187
+                                    ],
+                                    "score": 0.9,
+                                    "content": "\\sigma = \\{ 0 . 1 , 0 . 2 \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        378,
+                                        174,
+                                        469,
+                                        189
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and probability at 0.5.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        }
+                    ],
+                    "index": 6,
+                    "bbox_fs": [
+                        138,
+                        174,
+                        469,
+                        189
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        141,
+                        190,
+                        397,
+                        202
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                138,
+                                189,
+                                399,
+                                204
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        138,
+                                        189,
+                                        399,
+                                        204
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Horizontal Flip: Horizontal flip is applied with probability at 0.5",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        }
+                    ],
+                    "index": 7,
+                    "bbox_fs": [
+                        138,
+                        189,
+                        399,
+                        204
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        109,
+                        213,
+                        449,
+                        225
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                212,
+                                450,
+                                226
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        212,
+                                        450,
+                                        226
+                                    ],
+                                    "score": 1.0,
+                                    "content": "A.1.2 PRE-TRAINING AND DISTILLATION ON MOBILENET AND EFFICIENTNET",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        }
+                    ],
+                    "index": 8
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        232,
+                        506,
+                        299
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                232,
+                                505,
+                                244
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        232,
+                                        505,
+                                        244
+                                    ],
+                                    "score": 1.0,
+                                    "content": "MobileNet (Howard et al., 2017) and EfficientNet (Tan & Le, 2019) have been considered as the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                244,
+                                505,
+                                255
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        244,
+                                        505,
+                                        255
+                                    ],
+                                    "score": 1.0,
+                                    "content": "smaller counterparts with larger models, i.e., ResNet-50 (with supervised training, EfficientNet-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                254,
+                                507,
+                                267
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        254,
+                                        140,
+                                        267
+                                    ],
+                                    "score": 1.0,
+                                    "content": "B0 hits",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        140,
+                                        255,
+                                        168,
+                                        265
+                                    ],
+                                    "score": 0.86,
+                                    "content": "7 7 . 2 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        168,
+                                        254,
+                                        362,
+                                        267
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Top-1 Acc., and MobileNet-V3-large reaches",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        362,
+                                        255,
+                                        390,
+                                        265
+                                    ],
+                                    "score": 0.88,
+                                    "content": "7 2 . 2 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        391,
+                                        254,
+                                        507,
+                                        267
+                                    ],
+                                    "score": 1.0,
+                                    "content": "on ImageNet testing split).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                264,
+                                506,
+                                279
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        264,
+                                        506,
+                                        279
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Nevertheless, un-matched performances are observed in the task of self-supervised contrastive pre-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                276,
+                                506,
+                                289
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        276,
+                                        432,
+                                        289
+                                    ],
+                                    "score": 1.0,
+                                    "content": "training: i.e., Self-Supervised Learning (MoCo-V2) on MobileNet-V3 only yields",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        432,
+                                        277,
+                                        459,
+                                        288
+                                    ],
+                                    "score": 0.84,
+                                    "content": "3 6 . 3 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        459,
+                                        276,
+                                        506,
+                                        289
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Top-1 Acc.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                288,
+                                415,
+                                299
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        288,
+                                        415,
+                                        299
+                                    ],
+                                    "score": 1.0,
+                                    "content": "on ImageNet. We conjecture that several reasons might lead to this dilemma:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        }
+                    ],
+                    "index": 11.5,
+                    "bbox_fs": [
+                        105,
+                        232,
+                        507,
+                        299
+                    ]
+                },
+                {
+                    "type": "list",
+                    "bbox": [
+                        129,
+                        307,
+                        506,
+                        377
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                130,
+                                308,
+                                505,
+                                319
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        130,
+                                        308,
+                                        505,
+                                        319
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1. The inability of models with less parameters for handling large volume of categories and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                317,
+                                506,
+                                331
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        317,
+                                        506,
+                                        331
+                                    ],
+                                    "score": 1.0,
+                                    "content": "data, which exists also in other domains, i.e., face recognition (Guo et al., 2016; Zhang et al.,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                329,
+                                171,
+                                342
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        329,
+                                        171,
+                                        342
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2017).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                129,
+                                343,
+                                504,
+                                355
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        129,
+                                        343,
+                                        504,
+                                        355
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2. Less possibility for optimum parameters to be chosen when transferring to downstream",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                354,
+                                506,
+                                367
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        354,
+                                        506,
+                                        367
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tasks: models with more parameters after pre-training might produce a plenty cornucopia of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                142,
+                                366,
+                                289,
+                                379
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        142,
+                                        366,
+                                        289,
+                                        379
+                                    ],
+                                    "score": 1.0,
+                                    "content": "optimum parameters for fine-tuning.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20,
+                            "is_list_end_line": true
+                        }
+                    ],
+                    "index": 17.5,
+                    "bbox_fs": [
+                        129,
+                        308,
+                        506,
+                        379
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        385,
+                        505,
+                        441
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                386,
+                                504,
+                                398
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        386,
+                                        504,
+                                        398
+                                    ],
+                                    "score": 1.0,
+                                    "content": "To narrow the dramatic performance gap between smaller architectures using contrastive SSL with",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                396,
+                                505,
+                                410
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        396,
+                                        505,
+                                        410
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the larger, we explore with architectural manipulations and training hyper-parameters. In specific, we",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                407,
+                                506,
+                                421
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        407,
+                                        506,
+                                        421
+                                    ],
+                                    "score": 1.0,
+                                    "content": "find that by adding a deeper projection head largely improves the representation quality, a.k.a., better",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                418,
+                                506,
+                                431
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        418,
+                                        506,
+                                        431
+                                    ],
+                                    "score": 1.0,
+                                    "content": "performances on linear evaluation. We experiment with adding one additional linear projection head",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                429,
+                                263,
+                                442
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        429,
+                                        263,
+                                        442
+                                    ],
+                                    "score": 1.0,
+                                    "content": "on the top of convolutional backbones.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        }
+                    ],
+                    "index": 23,
+                    "bbox_fs": [
+                        105,
+                        386,
+                        506,
+                        442
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        446,
+                        505,
+                        535
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                446,
+                                505,
+                                458
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        446,
+                                        505,
+                                        458
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Similarly, we also expand the MLP projection head on EfficientNet-b0. Though recent work shows",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                458,
+                                505,
+                                469
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        458,
+                                        505,
+                                        469
+                                    ],
+                                    "score": 1.0,
+                                    "content": "that fine-tuning from a middle layer of the projection head can produce a largely different result (Chen",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                468,
+                                505,
+                                481
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        468,
+                                        505,
+                                        481
+                                    ],
+                                    "score": 1.0,
+                                    "content": "et al., 2020b), we consistently just use the representations from convolutional trunk without adding",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                479,
+                                505,
+                                492
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        479,
+                                        505,
+                                        492
+                                    ],
+                                    "score": 1.0,
+                                    "content": "extra layers during the phase of linear evaluation. As shown in Table 6, pre-training with a deeper",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                490,
+                                506,
+                                503
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        490,
+                                        435,
+                                        503
+                                    ],
+                                    "score": 1.0,
+                                    "content": "projection head dramatically helps the improvement on linear evaluations, adding",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        436,
+                                        491,
+                                        455,
+                                        501
+                                    ],
+                                    "score": 0.84,
+                                    "content": "17 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        456,
+                                        490,
+                                        506,
+                                        503
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Top-1. Acc.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                500,
+                                505,
+                                513
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        500,
+                                        505,
+                                        513
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for Mobile-v3-large, and we report the improved baselines in the main paper (see the first row in",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                512,
+                                505,
+                                525
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        512,
+                                        505,
+                                        525
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Table 1 of the main paper). We keep most of the hyper-parameters as the distillation on ResNet except",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                523,
+                                478,
+                                535
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        523,
+                                        478,
+                                        535
+                                    ],
+                                    "score": 1.0,
+                                    "content": "reducing the weight-decay of them to 1e-5, following (Tan & Le, 2019; Sandler et al., 2018).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        }
+                    ],
+                    "index": 29.5,
+                    "bbox_fs": [
+                        105,
+                        446,
+                        506,
+                        535
+                    ]
+                },
+                {
+                    "type": "table",
+                    "bbox": [
+                        184,
+                        579,
+                        424,
+                        646
+                    ],
+                    "blocks": [
+                        {
+                            "type": "table_caption",
+                            "bbox": [
+                                104,
+                                549,
+                                504,
+                                571
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        548,
+                                        505,
+                                        562
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                548,
+                                                505,
+                                                562
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Table 6: Linear evaluations on ImageNet of EfficientNet and MobileNet pre-trained using MoCo-v2. A deeper",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 34
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        559,
+                                        433,
+                                        572
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                559,
+                                                433,
+                                                572
+                                            ],
+                                            "score": 1.0,
+                                            "content": "projection head largely boosts the linear evaluation performances on smaller architectures.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 35
+                                }
+                            ],
+                            "index": 34.5
+                        },
+                        {
+                            "type": "table_body",
+                            "bbox": [
+                                184,
+                                579,
+                                424,
+                                646
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        184,
+                                        579,
+                                        424,
+                                        646
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                184,
+                                                579,
+                                                424,
+                                                646
+                                            ],
+                                            "score": 0.978,
+                                            "html": "<table><tr><td>Model</td><td>Deeper MLPs</td><td>Top-1 Acc. Top-5 Acc.</td><td></td></tr><tr><td>EfficientNet-b0</td><td></td><td>39.1</td><td>64.6</td></tr><tr><td>EfficientNet-b0</td><td>×</td><td>42.2</td><td>68.5</td></tr><tr><td>Mobile-v3-large</td><td>X</td><td>19.0</td><td>41.3</td></tr><tr><td>Mobile-v3-large</td><td>√</td><td>36.3</td><td>62.2</td></tr></table>",
+                                            "type": "table",
+                                            "image_path": "637ac61f5f9e5741399531cd7141a325efafa86d41a01e1c7e857e7176d2e6b6.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 38,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        184,
+                                        579,
+                                        424,
+                                        592.4
+                                    ],
+                                    "spans": [],
+                                    "index": 36
+                                },
+                                {
+                                    "bbox": [
+                                        184,
+                                        592.4,
+                                        424,
+                                        605.8
+                                    ],
+                                    "spans": [],
+                                    "index": 37
+                                },
+                                {
+                                    "bbox": [
+                                        184,
+                                        605.8,
+                                        424,
+                                        619.1999999999999
+                                    ],
+                                    "spans": [],
+                                    "index": 38
+                                },
+                                {
+                                    "bbox": [
+                                        184,
+                                        619.1999999999999,
+                                        424,
+                                        632.5999999999999
+                                    ],
+                                    "spans": [],
+                                    "index": 39
+                                },
+                                {
+                                    "bbox": [
+                                        184,
+                                        632.5999999999999,
+                                        424,
+                                        645.9999999999999
+                                    ],
+                                    "spans": [],
+                                    "index": 40
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 36.25
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        672,
+                        307,
+                        684
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                672,
+                                307,
+                                685
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        672,
+                                        307,
+                                        685
+                                    ],
+                                    "score": 1.0,
+                                    "content": "A.2 ADDITIONAL DETAILS OF EVALUATIONS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        }
+                    ],
+                    "index": 41
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        692,
+                        417,
+                        704
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                691,
+                                418,
+                                706
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        691,
+                                        418,
+                                        706
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We list additional details regarding our evaluation experiments in this section.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        }
+                    ],
+                    "index": 42,
+                    "bbox_fs": [
+                        105,
+                        691,
+                        418,
+                        706
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        709,
+                        507,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                709,
+                                506,
+                                722
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        709,
+                                        506,
+                                        722
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ImageNet-1k Semi-Supervised Linear Evaluation. Following Zhai et al. (2019); Chen et al.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                720,
+                                505,
+                                732
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        720,
+                                        505,
+                                        732
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(2020a), we train the FC layers on the basis of our student encoder after distillation using a fraction",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                466,
+                                506,
+                                478
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        466,
+                                        235,
+                                        478
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of labeled ImageNet-1k dataset (",
+                                    "type": "text",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        235,
+                                        466,
+                                        249,
+                                        477
+                                    ],
+                                    "score": 0.83,
+                                    "content": "1 \\%",
+                                    "type": "inline_equation",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        250,
+                                        466,
+                                        267,
+                                        478
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        267,
+                                        466,
+                                        287,
+                                        477
+                                    ],
+                                    "score": 0.85,
+                                    "content": "10 \\%",
+                                    "type": "inline_equation",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        287,
+                                        466,
+                                        506,
+                                        478
+                                    ],
+                                    "score": 1.0,
+                                    "content": "), and evaluate it on the whole test split. The fraction of",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                477,
+                                505,
+                                488
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        477,
+                                        505,
+                                        488
+                                    ],
+                                    "score": 1.0,
+                                    "content": "labeled dataset is constructed in a class-balanced way, with roughly 12 and 128 images per class∗. We",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                487,
+                                506,
+                                500
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        487,
+                                        399,
+                                        500
+                                    ],
+                                    "score": 1.0,
+                                    "content": "use SGD optimizer and set initial learning rate to be 30 with a multiplier",
+                                    "type": "text",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        399,
+                                        489,
+                                        411,
+                                        498
+                                    ],
+                                    "score": 0.75,
+                                    "content": "=",
+                                    "type": "inline_equation",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        411,
+                                        487,
+                                        506,
+                                        500
+                                    ],
+                                    "score": 1.0,
+                                    "content": "BatchSize/256 without",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                498,
+                                506,
+                                512
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        498,
+                                        506,
+                                        512
+                                    ],
+                                    "score": 1.0,
+                                    "content": "weight decaying for 100 epochs. We use the step-wise scheduler for the learning rate updating",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                510,
+                                506,
+                                522
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        510,
+                                        506,
+                                        522
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with 5 warm-up epochs, and the learning rate is reduced by 10 at 60 and 80 epochs. On smaller",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                519,
+                                506,
+                                534
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        519,
+                                        506,
+                                        534
+                                    ],
+                                    "score": 1.0,
+                                    "content": "architectures like EfficientNet and MobileNet, we reduce the initial learning rate to 3. During training,",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                531,
+                                506,
+                                544
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        531,
+                                        277,
+                                        544
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the image is center-cropped to the size of",
+                                    "type": "text",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        278,
+                                        532,
+                                        317,
+                                        542
+                                    ],
+                                    "score": 0.89,
+                                    "content": "2 2 4 \\times 2 2 4",
+                                    "type": "inline_equation",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        318,
+                                        531,
+                                        506,
+                                        544
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with just Random Horizontal Flip as the data",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                542,
+                                504,
+                                555
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        542,
+                                        319,
+                                        555
+                                    ],
+                                    "score": 1.0,
+                                    "content": "augmentation. For testing, we first resize the image to",
+                                    "type": "text",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        320,
+                                        543,
+                                        358,
+                                        553
+                                    ],
+                                    "score": 0.9,
+                                    "content": "2 5 6 \\times 2 5 6",
+                                    "type": "inline_equation",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        359,
+                                        542,
+                                        465,
+                                        555
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and use the center cropped",
+                                    "type": "text",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        465,
+                                        543,
+                                        504,
+                                        553
+                                    ],
+                                    "score": 0.88,
+                                    "content": "2 2 4 \\times 2 2 4",
+                                    "type": "inline_equation",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                553,
+                                506,
+                                566
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        553,
+                                        506,
+                                        566
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for pre-processing. In Table 8, we show the distillation results on a larger encoder (ResNet-550)",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                565,
+                                263,
+                                577
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        565,
+                                        263,
+                                        577
+                                    ],
+                                    "score": 1.0,
+                                    "content": "when using different teacher networks.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 34
+                        }
+                    ],
+                    "index": 43.5,
+                    "bbox_fs": [
+                        106,
+                        709,
+                        506,
+                        732
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "table",
+                    "bbox": [
+                        198,
+                        110,
+                        407,
+                        233
+                    ],
+                    "blocks": [
+                        {
+                            "type": "table_caption",
+                            "bbox": [
+                                106,
+                                80,
+                                505,
+                                101
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        79,
+                                        507,
+                                        92
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                79,
+                                                318,
+                                                92
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Table 7: Before and after distillation Top-1/5 test accuracy",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                318,
+                                                81,
+                                                332,
+                                                90
+                                            ],
+                                            "score": 0.69,
+                                            "content": "( \\% )",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                333,
+                                                79,
+                                                507,
+                                                92
+                                            ],
+                                            "score": 1.0,
+                                            "content": "on ImageNet of EfficientNet-b0 and MobileNet-",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 0
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        90,
+                                        209,
+                                        102
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                90,
+                                                209,
+                                                102
+                                            ],
+                                            "score": 1.0,
+                                            "content": "large without deeper MLPs.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 1
+                                }
+                            ],
+                            "index": 0.5
+                        },
+                        {
+                            "type": "table_body",
+                            "bbox": [
+                                198,
+                                110,
+                                407,
+                                233
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        198,
+                                        110,
+                                        407,
+                                        233
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                198,
+                                                110,
+                                                407,
+                                                233
+                                            ],
+                                            "score": 0.973,
+                                            "html": "<table><tr><td>Student</td><td>Teacher</td><td>Top-1</td><td>Top-5</td></tr><tr><td rowspan=\"3\">EfficientNet-b0</td><td>X</td><td>39.1</td><td>64.6</td></tr><tr><td>ResNet-50</td><td>59.2</td><td>81.2</td></tr><tr><td>ResNet-101</td><td>62.8</td><td>84.7</td></tr><tr><td></td><td>ResNet-152</td><td>63.3</td><td>85.6</td></tr><tr><td rowspan=\"3\">MobileNet-v3</td><td>X ResNet-50</td><td>19.0 50.9</td><td>41.3 77.7</td></tr><tr><td>ResNet-101</td><td>57.6</td><td>82.6</td></tr><tr><td>ResNet-152</td><td>58.3</td><td>82.9</td></tr></table>",
+                                            "type": "table",
+                                            "image_path": "cf3071272dac29e76307ac8ba82a1a6899f7e85f6571ad9cb66401dc71b61d17.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 6,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        198,
+                                        110,
+                                        407,
+                                        123.66666666666667
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                },
+                                {
+                                    "bbox": [
+                                        198,
+                                        123.66666666666667,
+                                        407,
+                                        137.33333333333334
+                                    ],
+                                    "spans": [],
+                                    "index": 3
+                                },
+                                {
+                                    "bbox": [
+                                        198,
+                                        137.33333333333334,
+                                        407,
+                                        151.0
+                                    ],
+                                    "spans": [],
+                                    "index": 4
+                                },
+                                {
+                                    "bbox": [
+                                        198,
+                                        151.0,
+                                        407,
+                                        164.66666666666666
+                                    ],
+                                    "spans": [],
+                                    "index": 5
+                                },
+                                {
+                                    "bbox": [
+                                        198,
+                                        164.66666666666666,
+                                        407,
+                                        178.33333333333331
+                                    ],
+                                    "spans": [],
+                                    "index": 6
+                                },
+                                {
+                                    "bbox": [
+                                        198,
+                                        178.33333333333331,
+                                        407,
+                                        191.99999999999997
+                                    ],
+                                    "spans": [],
+                                    "index": 7
+                                },
+                                {
+                                    "bbox": [
+                                        198,
+                                        191.99999999999997,
+                                        407,
+                                        205.66666666666663
+                                    ],
+                                    "spans": [],
+                                    "index": 8
+                                },
+                                {
+                                    "bbox": [
+                                        198,
+                                        205.66666666666663,
+                                        407,
+                                        219.3333333333333
+                                    ],
+                                    "spans": [],
+                                    "index": 9
+                                },
+                                {
+                                    "bbox": [
+                                        198,
+                                        219.3333333333333,
+                                        407,
+                                        232.99999999999994
+                                    ],
+                                    "spans": [],
+                                    "index": 10
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 3.25
+                },
+                {
+                    "type": "table",
+                    "bbox": [
+                        193,
+                        302,
+                        417,
+                        436
+                    ],
+                    "blocks": [
+                        {
+                            "type": "table_caption",
+                            "bbox": [
+                                107,
+                                252,
+                                506,
+                                293
+                            ],
+                            "group_id": 1,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        253,
+                                        504,
+                                        263
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                253,
+                                                245,
+                                                263
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Table 8: ImageNet-1k test accuracy",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                246,
+                                                253,
+                                                260,
+                                                262
+                                            ],
+                                            "score": 0.73,
+                                            "content": "( \\% )",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                261,
+                                                253,
+                                                504,
+                                                263
+                                            ],
+                                            "score": 1.0,
+                                            "content": "under KNN and linear classification on ResNet-50 encoder with",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 11
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        261,
+                                        506,
+                                        275
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                261,
+                                                331,
+                                                275
+                                            ],
+                                            "score": 1.0,
+                                            "content": "deeper, MoCo-V2/SWAV pre-trained teacher architectures.",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                332,
+                                                263,
+                                                340,
+                                                272
+                                            ],
+                                            "score": 0.46,
+                                            "content": "\\pmb { \\chi }",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                340,
+                                                261,
+                                                506,
+                                                275
+                                            ],
+                                            "score": 1.0,
+                                            "content": "denotes MoCo-V2 self-supervised learning",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 12
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        272,
+                                        505,
+                                        284
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                272,
+                                                505,
+                                                284
+                                            ],
+                                            "score": 1.0,
+                                            "content": "baselines before distillation. * indicates using a stronger teacher encoder pre-trained by SWAV with additional",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 13
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        282,
+                                        227,
+                                        294
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                282,
+                                                227,
+                                                294
+                                            ],
+                                            "score": 1.0,
+                                            "content": "small-patches during distillation.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 14
+                                }
+                            ],
+                            "index": 12.5
+                        },
+                        {
+                            "type": "table_body",
+                            "bbox": [
+                                193,
+                                302,
+                                417,
+                                436
+                            ],
+                            "group_id": 1,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        193,
+                                        302,
+                                        417,
+                                        436
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                193,
+                                                302,
+                                                417,
+                                                436
+                                            ],
+                                            "score": 0.974,
+                                            "html": "<table><tr><td></td><td>Stud.</td><td></td><td>ResNet-50</td><td></td></tr><tr><td>Teac.</td><td>Epoch</td><td>KNN</td><td>Top-1</td><td>Top-5</td></tr><tr><td>X ResNet-50</td><td>200</td><td>46.1 46.1</td><td>67.4 67.5</td><td>87.8 87.8</td></tr><tr><td>△ ResNet-101</td><td>200</td><td>+0.0 52.3</td><td>+0.1</td><td>+0.0</td></tr><tr><td>△</td><td>200</td><td>+6.2</td><td>69.1 +1.7</td><td>88.7 +0.9</td></tr><tr><td>ResNet-152 △</td><td>200</td><td>53.2 +7.1</td><td>70.4 +3.0</td><td>90.5 +2.7</td></tr><tr><td>ResNet-50×2* △</td><td>800</td><td>59.0 +12.9</td><td>74.3 +6.9</td><td>92.2 +4.4</td></tr></table>",
+                                            "type": "table",
+                                            "image_path": "e10d7d8d9104543baae25184077eb1a0f410866ba1b0404e862ec467a0ea7815.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 19.5,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        193,
+                                        302,
+                                        417,
+                                        315.4
+                                    ],
+                                    "spans": [],
+                                    "index": 15
+                                },
+                                {
+                                    "bbox": [
+                                        193,
+                                        315.4,
+                                        417,
+                                        328.79999999999995
+                                    ],
+                                    "spans": [],
+                                    "index": 16
+                                },
+                                {
+                                    "bbox": [
+                                        193,
+                                        328.79999999999995,
+                                        417,
+                                        342.19999999999993
+                                    ],
+                                    "spans": [],
+                                    "index": 17
+                                },
+                                {
+                                    "bbox": [
+                                        193,
+                                        342.19999999999993,
+                                        417,
+                                        355.5999999999999
+                                    ],
+                                    "spans": [],
+                                    "index": 18
+                                },
+                                {
+                                    "bbox": [
+                                        193,
+                                        355.5999999999999,
+                                        417,
+                                        368.9999999999999
+                                    ],
+                                    "spans": [],
+                                    "index": 19
+                                },
+                                {
+                                    "bbox": [
+                                        193,
+                                        368.9999999999999,
+                                        417,
+                                        382.39999999999986
+                                    ],
+                                    "spans": [],
+                                    "index": 20
+                                },
+                                {
+                                    "bbox": [
+                                        193,
+                                        382.39999999999986,
+                                        417,
+                                        395.79999999999984
+                                    ],
+                                    "spans": [],
+                                    "index": 21
+                                },
+                                {
+                                    "bbox": [
+                                        193,
+                                        395.79999999999984,
+                                        417,
+                                        409.1999999999998
+                                    ],
+                                    "spans": [],
+                                    "index": 22
+                                },
+                                {
+                                    "bbox": [
+                                        193,
+                                        409.1999999999998,
+                                        417,
+                                        422.5999999999998
+                                    ],
+                                    "spans": [],
+                                    "index": 23
+                                },
+                                {
+                                    "bbox": [
+                                        193,
+                                        422.5999999999998,
+                                        417,
+                                        435.9999999999998
+                                    ],
+                                    "spans": [],
+                                    "index": 24
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 16.0
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        465,
+                        506,
+                        576
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                466,
+                                506,
+                                478
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        466,
+                                        235,
+                                        478
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of labeled ImageNet-1k dataset (",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        235,
+                                        466,
+                                        249,
+                                        477
+                                    ],
+                                    "score": 0.83,
+                                    "content": "1 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        250,
+                                        466,
+                                        267,
+                                        478
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        267,
+                                        466,
+                                        287,
+                                        477
+                                    ],
+                                    "score": 0.85,
+                                    "content": "10 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        287,
+                                        466,
+                                        506,
+                                        478
+                                    ],
+                                    "score": 1.0,
+                                    "content": "), and evaluate it on the whole test split. The fraction of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                477,
+                                505,
+                                488
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        477,
+                                        505,
+                                        488
+                                    ],
+                                    "score": 1.0,
+                                    "content": "labeled dataset is constructed in a class-balanced way, with roughly 12 and 128 images per class∗. We",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                487,
+                                506,
+                                500
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        487,
+                                        399,
+                                        500
+                                    ],
+                                    "score": 1.0,
+                                    "content": "use SGD optimizer and set initial learning rate to be 30 with a multiplier",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        399,
+                                        489,
+                                        411,
+                                        498
+                                    ],
+                                    "score": 0.75,
+                                    "content": "=",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        411,
+                                        487,
+                                        506,
+                                        500
+                                    ],
+                                    "score": 1.0,
+                                    "content": "BatchSize/256 without",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                498,
+                                506,
+                                512
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        498,
+                                        506,
+                                        512
+                                    ],
+                                    "score": 1.0,
+                                    "content": "weight decaying for 100 epochs. We use the step-wise scheduler for the learning rate updating",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                510,
+                                506,
+                                522
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        510,
+                                        506,
+                                        522
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with 5 warm-up epochs, and the learning rate is reduced by 10 at 60 and 80 epochs. On smaller",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                519,
+                                506,
+                                534
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        519,
+                                        506,
+                                        534
+                                    ],
+                                    "score": 1.0,
+                                    "content": "architectures like EfficientNet and MobileNet, we reduce the initial learning rate to 3. During training,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                531,
+                                506,
+                                544
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        531,
+                                        277,
+                                        544
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the image is center-cropped to the size of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        278,
+                                        532,
+                                        317,
+                                        542
+                                    ],
+                                    "score": 0.89,
+                                    "content": "2 2 4 \\times 2 2 4",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        318,
+                                        531,
+                                        506,
+                                        544
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with just Random Horizontal Flip as the data",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                542,
+                                504,
+                                555
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        542,
+                                        319,
+                                        555
+                                    ],
+                                    "score": 1.0,
+                                    "content": "augmentation. For testing, we first resize the image to",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        320,
+                                        543,
+                                        358,
+                                        553
+                                    ],
+                                    "score": 0.9,
+                                    "content": "2 5 6 \\times 2 5 6",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        359,
+                                        542,
+                                        465,
+                                        555
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and use the center cropped",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        465,
+                                        543,
+                                        504,
+                                        553
+                                    ],
+                                    "score": 0.88,
+                                    "content": "2 2 4 \\times 2 2 4",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                553,
+                                506,
+                                566
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        553,
+                                        506,
+                                        566
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for pre-processing. In Table 8, we show the distillation results on a larger encoder (ResNet-550)",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                565,
+                                263,
+                                577
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        565,
+                                        263,
+                                        577
+                                    ],
+                                    "score": 1.0,
+                                    "content": "when using different teacher networks.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        }
+                    ],
+                    "index": 29.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        581,
+                        505,
+                        637
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                580,
+                                505,
+                                594
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        580,
+                                        505,
+                                        594
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Transfer Learning. We test the transferability of the representations learned from self-supervised",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                591,
+                                506,
+                                604
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        591,
+                                        506,
+                                        604
+                                    ],
+                                    "score": 1.0,
+                                    "content": "distillation by conducting the linear evaluations using offline features on several other datasets.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                602,
+                                506,
+                                616
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        602,
+                                        506,
+                                        616
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Specifically, a single layer logistic classifier is trained following (Chen et al., 2020a; Grill et al., 2020)",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                614,
+                                507,
+                                626
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        614,
+                                        507,
+                                        626
+                                    ],
+                                    "score": 1.0,
+                                    "content": "using SGD optimizer without weight decay and momentum parameter at 0.9. We use CIFAR-10,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                624,
+                                469,
+                                637
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        624,
+                                        469,
+                                        637
+                                    ],
+                                    "score": 1.0,
+                                    "content": "CIFAR-100 (Krizhevsky et al., 2009) and SUN-397 (Xiao et al., 2010) as our testing beds.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        }
+                    ],
+                    "index": 37
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        125,
+                        646,
+                        505,
+                        702
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                124,
+                                645,
+                                506,
+                                659
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        124,
+                                        645,
+                                        289,
+                                        659
+                                    ],
+                                    "score": 1.0,
+                                    "content": "CIFAR: As the size for CIFAR dataset is",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        289,
+                                        647,
+                                        318,
+                                        657
+                                    ],
+                                    "score": 0.89,
+                                    "content": "3 2 \\times 3 2",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        318,
+                                        645,
+                                        414,
+                                        659
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", we resize all images to",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        414,
+                                        646,
+                                        454,
+                                        658
+                                    ],
+                                    "score": 0.88,
+                                    "content": "2 2 4 \\times 2 2 4",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        454,
+                                        645,
+                                        506,
+                                        659
+                                    ],
+                                    "score": 1.0,
+                                    "content": "pixels along",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                124,
+                                658,
+                                506,
+                                671
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        124,
+                                        658,
+                                        506,
+                                        671
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the shorter side using bicubic resampling method, followed by a center crop operation. We",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                124,
+                                669,
+                                505,
+                                681
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        124,
+                                        669,
+                                        505,
+                                        681
+                                    ],
+                                    "score": 1.0,
+                                    "content": "set the learning rate at 1e-3 constantly and train it for 120 epochs. The hyper-parameters are",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                124,
+                                680,
+                                505,
+                                693
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        124,
+                                        680,
+                                        505,
+                                        693
+                                    ],
+                                    "score": 1.0,
+                                    "content": "searched using 10 fold cross-validation on the train split and report its final top-1 accuracy on",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                124,
+                                690,
+                                178,
+                                702
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        124,
+                                        690,
+                                        178,
+                                        702
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the test split.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        }
+                    ],
+                    "index": 42
+                }
+            ],
+            "page_idx": 14,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        106,
+                        712,
+                        507,
+                        731
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                117,
+                                709,
+                                509,
+                                724
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        117,
+                                        709,
+                                        509,
+                                        724
+                                    ],
+                                    "score": 1.0,
+                                    "content": "∗The full image ids for semi-supervised evaluation on ImageNet-1k can be found at https://github.",
+                                    "type": "text"
+                                }
+                            ]
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                721,
+                                408,
+                                733
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        721,
+                                        408,
+                                        733
+                                    ],
+                                    "score": 1.0,
+                                    "content": "com/google-research/simclr/tree/master/imagenet_subsets.",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        108,
+                        27,
+                        292,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                26,
+                                293,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        26,
+                                        293,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Published as a conference paper at ICLR 2021",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        300,
+                        751,
+                        310,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                299,
+                                750,
+                                312,
+                                764
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        299,
+                                        750,
+                                        312,
+                                        764
+                                    ],
+                                    "score": 1.0,
+                                    "content": "",
+                                    "type": "text",
+                                    "height": 14,
+                                    "width": 13
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "table",
+                    "bbox": [
+                        198,
+                        110,
+                        407,
+                        233
+                    ],
+                    "blocks": [
+                        {
+                            "type": "table_caption",
+                            "bbox": [
+                                106,
+                                80,
+                                505,
+                                101
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        79,
+                                        507,
+                                        92
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                79,
+                                                318,
+                                                92
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Table 7: Before and after distillation Top-1/5 test accuracy",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                318,
+                                                81,
+                                                332,
+                                                90
+                                            ],
+                                            "score": 0.69,
+                                            "content": "( \\% )",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                333,
+                                                79,
+                                                507,
+                                                92
+                                            ],
+                                            "score": 1.0,
+                                            "content": "on ImageNet of EfficientNet-b0 and MobileNet-",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 0
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        90,
+                                        209,
+                                        102
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                90,
+                                                209,
+                                                102
+                                            ],
+                                            "score": 1.0,
+                                            "content": "large without deeper MLPs.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 1
+                                }
+                            ],
+                            "index": 0.5
+                        },
+                        {
+                            "type": "table_body",
+                            "bbox": [
+                                198,
+                                110,
+                                407,
+                                233
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        198,
+                                        110,
+                                        407,
+                                        233
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                198,
+                                                110,
+                                                407,
+                                                233
+                                            ],
+                                            "score": 0.973,
+                                            "html": "<table><tr><td>Student</td><td>Teacher</td><td>Top-1</td><td>Top-5</td></tr><tr><td rowspan=\"3\">EfficientNet-b0</td><td>X</td><td>39.1</td><td>64.6</td></tr><tr><td>ResNet-50</td><td>59.2</td><td>81.2</td></tr><tr><td>ResNet-101</td><td>62.8</td><td>84.7</td></tr><tr><td></td><td>ResNet-152</td><td>63.3</td><td>85.6</td></tr><tr><td rowspan=\"3\">MobileNet-v3</td><td>X ResNet-50</td><td>19.0 50.9</td><td>41.3 77.7</td></tr><tr><td>ResNet-101</td><td>57.6</td><td>82.6</td></tr><tr><td>ResNet-152</td><td>58.3</td><td>82.9</td></tr></table>",
+                                            "type": "table",
+                                            "image_path": "cf3071272dac29e76307ac8ba82a1a6899f7e85f6571ad9cb66401dc71b61d17.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 6,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        198,
+                                        110,
+                                        407,
+                                        123.66666666666667
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                },
+                                {
+                                    "bbox": [
+                                        198,
+                                        123.66666666666667,
+                                        407,
+                                        137.33333333333334
+                                    ],
+                                    "spans": [],
+                                    "index": 3
+                                },
+                                {
+                                    "bbox": [
+                                        198,
+                                        137.33333333333334,
+                                        407,
+                                        151.0
+                                    ],
+                                    "spans": [],
+                                    "index": 4
+                                },
+                                {
+                                    "bbox": [
+                                        198,
+                                        151.0,
+                                        407,
+                                        164.66666666666666
+                                    ],
+                                    "spans": [],
+                                    "index": 5
+                                },
+                                {
+                                    "bbox": [
+                                        198,
+                                        164.66666666666666,
+                                        407,
+                                        178.33333333333331
+                                    ],
+                                    "spans": [],
+                                    "index": 6
+                                },
+                                {
+                                    "bbox": [
+                                        198,
+                                        178.33333333333331,
+                                        407,
+                                        191.99999999999997
+                                    ],
+                                    "spans": [],
+                                    "index": 7
+                                },
+                                {
+                                    "bbox": [
+                                        198,
+                                        191.99999999999997,
+                                        407,
+                                        205.66666666666663
+                                    ],
+                                    "spans": [],
+                                    "index": 8
+                                },
+                                {
+                                    "bbox": [
+                                        198,
+                                        205.66666666666663,
+                                        407,
+                                        219.3333333333333
+                                    ],
+                                    "spans": [],
+                                    "index": 9
+                                },
+                                {
+                                    "bbox": [
+                                        198,
+                                        219.3333333333333,
+                                        407,
+                                        232.99999999999994
+                                    ],
+                                    "spans": [],
+                                    "index": 10
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 3.25
+                },
+                {
+                    "type": "table",
+                    "bbox": [
+                        193,
+                        302,
+                        417,
+                        436
+                    ],
+                    "blocks": [
+                        {
+                            "type": "table_caption",
+                            "bbox": [
+                                107,
+                                252,
+                                506,
+                                293
+                            ],
+                            "group_id": 1,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        253,
+                                        504,
+                                        263
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                253,
+                                                245,
+                                                263
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Table 8: ImageNet-1k test accuracy",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                246,
+                                                253,
+                                                260,
+                                                262
+                                            ],
+                                            "score": 0.73,
+                                            "content": "( \\% )",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                261,
+                                                253,
+                                                504,
+                                                263
+                                            ],
+                                            "score": 1.0,
+                                            "content": "under KNN and linear classification on ResNet-50 encoder with",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 11
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        261,
+                                        506,
+                                        275
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                261,
+                                                331,
+                                                275
+                                            ],
+                                            "score": 1.0,
+                                            "content": "deeper, MoCo-V2/SWAV pre-trained teacher architectures.",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                332,
+                                                263,
+                                                340,
+                                                272
+                                            ],
+                                            "score": 0.46,
+                                            "content": "\\pmb { \\chi }",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                340,
+                                                261,
+                                                506,
+                                                275
+                                            ],
+                                            "score": 1.0,
+                                            "content": "denotes MoCo-V2 self-supervised learning",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 12
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        272,
+                                        505,
+                                        284
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                272,
+                                                505,
+                                                284
+                                            ],
+                                            "score": 1.0,
+                                            "content": "baselines before distillation. * indicates using a stronger teacher encoder pre-trained by SWAV with additional",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 13
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        282,
+                                        227,
+                                        294
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                282,
+                                                227,
+                                                294
+                                            ],
+                                            "score": 1.0,
+                                            "content": "small-patches during distillation.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 14
+                                }
+                            ],
+                            "index": 12.5
+                        },
+                        {
+                            "type": "table_body",
+                            "bbox": [
+                                193,
+                                302,
+                                417,
+                                436
+                            ],
+                            "group_id": 1,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        193,
+                                        302,
+                                        417,
+                                        436
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                193,
+                                                302,
+                                                417,
+                                                436
+                                            ],
+                                            "score": 0.974,
+                                            "html": "<table><tr><td></td><td>Stud.</td><td></td><td>ResNet-50</td><td></td></tr><tr><td>Teac.</td><td>Epoch</td><td>KNN</td><td>Top-1</td><td>Top-5</td></tr><tr><td>X ResNet-50</td><td>200</td><td>46.1 46.1</td><td>67.4 67.5</td><td>87.8 87.8</td></tr><tr><td>△ ResNet-101</td><td>200</td><td>+0.0 52.3</td><td>+0.1</td><td>+0.0</td></tr><tr><td>△</td><td>200</td><td>+6.2</td><td>69.1 +1.7</td><td>88.7 +0.9</td></tr><tr><td>ResNet-152 △</td><td>200</td><td>53.2 +7.1</td><td>70.4 +3.0</td><td>90.5 +2.7</td></tr><tr><td>ResNet-50×2* △</td><td>800</td><td>59.0 +12.9</td><td>74.3 +6.9</td><td>92.2 +4.4</td></tr></table>",
+                                            "type": "table",
+                                            "image_path": "e10d7d8d9104543baae25184077eb1a0f410866ba1b0404e862ec467a0ea7815.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 19.5,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        193,
+                                        302,
+                                        417,
+                                        315.4
+                                    ],
+                                    "spans": [],
+                                    "index": 15
+                                },
+                                {
+                                    "bbox": [
+                                        193,
+                                        315.4,
+                                        417,
+                                        328.79999999999995
+                                    ],
+                                    "spans": [],
+                                    "index": 16
+                                },
+                                {
+                                    "bbox": [
+                                        193,
+                                        328.79999999999995,
+                                        417,
+                                        342.19999999999993
+                                    ],
+                                    "spans": [],
+                                    "index": 17
+                                },
+                                {
+                                    "bbox": [
+                                        193,
+                                        342.19999999999993,
+                                        417,
+                                        355.5999999999999
+                                    ],
+                                    "spans": [],
+                                    "index": 18
+                                },
+                                {
+                                    "bbox": [
+                                        193,
+                                        355.5999999999999,
+                                        417,
+                                        368.9999999999999
+                                    ],
+                                    "spans": [],
+                                    "index": 19
+                                },
+                                {
+                                    "bbox": [
+                                        193,
+                                        368.9999999999999,
+                                        417,
+                                        382.39999999999986
+                                    ],
+                                    "spans": [],
+                                    "index": 20
+                                },
+                                {
+                                    "bbox": [
+                                        193,
+                                        382.39999999999986,
+                                        417,
+                                        395.79999999999984
+                                    ],
+                                    "spans": [],
+                                    "index": 21
+                                },
+                                {
+                                    "bbox": [
+                                        193,
+                                        395.79999999999984,
+                                        417,
+                                        409.1999999999998
+                                    ],
+                                    "spans": [],
+                                    "index": 22
+                                },
+                                {
+                                    "bbox": [
+                                        193,
+                                        409.1999999999998,
+                                        417,
+                                        422.5999999999998
+                                    ],
+                                    "spans": [],
+                                    "index": 23
+                                },
+                                {
+                                    "bbox": [
+                                        193,
+                                        422.5999999999998,
+                                        417,
+                                        435.9999999999998
+                                    ],
+                                    "spans": [],
+                                    "index": 24
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 16.0
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        465,
+                        506,
+                        576
+                    ],
+                    "lines": [],
+                    "index": 29.5,
+                    "bbox_fs": [
+                        105,
+                        466,
+                        506,
+                        577
+                    ],
+                    "lines_deleted": true
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        581,
+                        505,
+                        637
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                580,
+                                505,
+                                594
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        580,
+                                        505,
+                                        594
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Transfer Learning. We test the transferability of the representations learned from self-supervised",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                591,
+                                506,
+                                604
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        591,
+                                        506,
+                                        604
+                                    ],
+                                    "score": 1.0,
+                                    "content": "distillation by conducting the linear evaluations using offline features on several other datasets.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                602,
+                                506,
+                                616
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        602,
+                                        506,
+                                        616
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Specifically, a single layer logistic classifier is trained following (Chen et al., 2020a; Grill et al., 2020)",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                614,
+                                507,
+                                626
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        614,
+                                        507,
+                                        626
+                                    ],
+                                    "score": 1.0,
+                                    "content": "using SGD optimizer without weight decay and momentum parameter at 0.9. We use CIFAR-10,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                624,
+                                469,
+                                637
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        624,
+                                        469,
+                                        637
+                                    ],
+                                    "score": 1.0,
+                                    "content": "CIFAR-100 (Krizhevsky et al., 2009) and SUN-397 (Xiao et al., 2010) as our testing beds.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        }
+                    ],
+                    "index": 37,
+                    "bbox_fs": [
+                        105,
+                        580,
+                        507,
+                        637
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        125,
+                        646,
+                        505,
+                        702
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                124,
+                                645,
+                                506,
+                                659
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        124,
+                                        645,
+                                        289,
+                                        659
+                                    ],
+                                    "score": 1.0,
+                                    "content": "CIFAR: As the size for CIFAR dataset is",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        289,
+                                        647,
+                                        318,
+                                        657
+                                    ],
+                                    "score": 0.89,
+                                    "content": "3 2 \\times 3 2",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        318,
+                                        645,
+                                        414,
+                                        659
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", we resize all images to",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        414,
+                                        646,
+                                        454,
+                                        658
+                                    ],
+                                    "score": 0.88,
+                                    "content": "2 2 4 \\times 2 2 4",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        454,
+                                        645,
+                                        506,
+                                        659
+                                    ],
+                                    "score": 1.0,
+                                    "content": "pixels along",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                124,
+                                658,
+                                506,
+                                671
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        124,
+                                        658,
+                                        506,
+                                        671
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the shorter side using bicubic resampling method, followed by a center crop operation. We",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                124,
+                                669,
+                                505,
+                                681
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        124,
+                                        669,
+                                        505,
+                                        681
+                                    ],
+                                    "score": 1.0,
+                                    "content": "set the learning rate at 1e-3 constantly and train it for 120 epochs. The hyper-parameters are",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                124,
+                                680,
+                                505,
+                                693
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        124,
+                                        680,
+                                        505,
+                                        693
+                                    ],
+                                    "score": 1.0,
+                                    "content": "searched using 10 fold cross-validation on the train split and report its final top-1 accuracy on",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                124,
+                                690,
+                                178,
+                                702
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        124,
+                                        690,
+                                        178,
+                                        702
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the test split.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        }
+                    ],
+                    "index": 42,
+                    "bbox_fs": [
+                        124,
+                        645,
+                        506,
+                        702
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "table",
+                    "bbox": [
+                        186,
+                        121,
+                        421,
+                        261
+                    ],
+                    "blocks": [
+                        {
+                            "type": "table_caption",
+                            "bbox": [
+                                106,
+                                81,
+                                503,
+                                111
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        80,
+                                        505,
+                                        92
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                104,
+                                                80,
+                                                443,
+                                                92
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Table 9: Object detection and instance segmentation fine-tuned on VOC07: bounding-box AP",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                443,
+                                                80,
+                                                466,
+                                                91
+                                            ],
+                                            "score": 0.81,
+                                            "content": "( \\mathsf { A P } ^ { b b } )",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                466,
+                                                80,
+                                                505,
+                                                92
+                                            ],
+                                            "score": 1.0,
+                                            "content": ") and mask",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 0
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        90,
+                                        505,
+                                        102
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                90,
+                                                121,
+                                                102
+                                            ],
+                                            "score": 1.0,
+                                            "content": "AP",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                121,
+                                                91,
+                                                149,
+                                                101
+                                            ],
+                                            "score": 0.82,
+                                            "content": "( \\mathbf { A } \\mathbf { P } ^ { m k } )",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                149,
+                                                90,
+                                                505,
+                                                102
+                                            ],
+                                            "score": 1.0,
+                                            "content": ") evaluated on VOC07-val. The first row shows the baseline from MoCo-v2 backbones without",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        100,
+                                        150,
+                                        112
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                100,
+                                                150,
+                                                112
+                                            ],
+                                            "score": 1.0,
+                                            "content": "distillation.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 2
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "type": "table_body",
+                            "bbox": [
+                                186,
+                                121,
+                                421,
+                                261
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        186,
+                                        121,
+                                        421,
+                                        261
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                186,
+                                                121,
+                                                421,
+                                                261
+                                            ],
+                                            "score": 0.976,
+                                            "html": "<table><tr><td rowspan=\"2\">Student</td><td rowspan=\"2\">Teacher</td><td colspan=\"3\">VOC Object Detection</td></tr><tr><td>APbb</td><td>AP</td><td>AP</td></tr><tr><td rowspan=\"3\">ResNet-34</td><td>X</td><td>53.6</td><td>79.1</td><td>58.7</td></tr><tr><td>ResNet-50</td><td>53.7 (+0.1)</td><td>79.4 (+0.3)</td><td>59.2 (+0.5)</td></tr><tr><td>ResNet-101 ResNet-152</td><td>54.1 (+0.5) 54.4 (+0.8)</td><td>79.8 (+0.7) 80.1 (+1.0)</td><td>59.1 (+0.4) ） 59.9 (+1.2)</td></tr><tr><td rowspan=\"2\">ResNet-50</td><td>X</td><td>57.0</td><td>82.4</td><td>63.6</td></tr><tr><td>ResNet-50 ResNet-101</td><td>57.0 (+0.0) 57.1 (+0.1)</td><td>82.4 (+0.0) 82.8 (+0.4)</td><td>63.6 (+0.0) 63.8 (+0.2)</td></tr></table>",
+                                            "type": "table",
+                                            "image_path": "5c3f7a898d26a2d2fe5fcf5d40e410578494648f20defdcd18389a3531ae57cb.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 7.5,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        186,
+                                        121,
+                                        421,
+                                        135.0
+                                    ],
+                                    "spans": [],
+                                    "index": 3
+                                },
+                                {
+                                    "bbox": [
+                                        186,
+                                        135.0,
+                                        421,
+                                        149.0
+                                    ],
+                                    "spans": [],
+                                    "index": 4
+                                },
+                                {
+                                    "bbox": [
+                                        186,
+                                        149.0,
+                                        421,
+                                        163.0
+                                    ],
+                                    "spans": [],
+                                    "index": 5
+                                },
+                                {
+                                    "bbox": [
+                                        186,
+                                        163.0,
+                                        421,
+                                        177.0
+                                    ],
+                                    "spans": [],
+                                    "index": 6
+                                },
+                                {
+                                    "bbox": [
+                                        186,
+                                        177.0,
+                                        421,
+                                        191.0
+                                    ],
+                                    "spans": [],
+                                    "index": 7
+                                },
+                                {
+                                    "bbox": [
+                                        186,
+                                        191.0,
+                                        421,
+                                        205.0
+                                    ],
+                                    "spans": [],
+                                    "index": 8
+                                },
+                                {
+                                    "bbox": [
+                                        186,
+                                        205.0,
+                                        421,
+                                        219.0
+                                    ],
+                                    "spans": [],
+                                    "index": 9
+                                },
+                                {
+                                    "bbox": [
+                                        186,
+                                        219.0,
+                                        421,
+                                        233.0
+                                    ],
+                                    "spans": [],
+                                    "index": 10
+                                },
+                                {
+                                    "bbox": [
+                                        186,
+                                        233.0,
+                                        421,
+                                        247.0
+                                    ],
+                                    "spans": [],
+                                    "index": 11
+                                },
+                                {
+                                    "bbox": [
+                                        186,
+                                        247.0,
+                                        421,
+                                        261.0
+                                    ],
+                                    "spans": [],
+                                    "index": 12
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 4.25
+                },
+                {
+                    "type": "table",
+                    "bbox": [
+                        109,
+                        319,
+                        502,
+                        407
+                    ],
+                    "blocks": [
+                        {
+                            "type": "table_caption",
+                            "bbox": [
+                                106,
+                                280,
+                                504,
+                                310
+                            ],
+                            "group_id": 1,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        279,
+                                        505,
+                                        290
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                104,
+                                                279,
+                                                462,
+                                                290
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Table 10: Object detection and instance segmentation fine-tuned on COCO: bounding-box AP",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                462,
+                                                279,
+                                                487,
+                                                290
+                                            ],
+                                            "score": 0.84,
+                                            "content": "( \\mathbf { A P } ^ { b b } )",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                488,
+                                                279,
+                                                505,
+                                                290
+                                            ],
+                                            "score": 1.0,
+                                            "content": "and",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 13
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        288,
+                                        506,
+                                        301
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                288,
+                                                142,
+                                                301
+                                            ],
+                                            "score": 1.0,
+                                            "content": "mask AP",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                142,
+                                                289,
+                                                170,
+                                                300
+                                            ],
+                                            "score": 0.82,
+                                            "content": "( \\mathbf { A } \\mathbf { P } ^ { m k } )",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                171,
+                                                288,
+                                                506,
+                                                301
+                                            ],
+                                            "score": 1.0,
+                                            "content": "evaluated on COCO-val2017. The first several rows show the baselines from unsupervised",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 14
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        299,
+                                        219,
+                                        311
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                299,
+                                                219,
+                                                311
+                                            ],
+                                            "score": 1.0,
+                                            "content": "backbones without distillation.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 15
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "type": "table_body",
+                            "bbox": [
+                                109,
+                                319,
+                                502,
+                                407
+                            ],
+                            "group_id": 1,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        109,
+                                        319,
+                                        502,
+                                        407
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                109,
+                                                319,
+                                                502,
+                                                407
+                                            ],
+                                            "score": 0.982,
+                                            "html": "<table><tr><td rowspan=\"2\">Student</td><td rowspan=\"2\">Teacher</td><td colspan=\"3\">Object Detection</td><td colspan=\"3\">Instance Segmentation</td></tr><tr><td>Apbb</td><td>AP</td><td>APP</td><td>Apmk</td><td>AP</td><td>AP</td></tr><tr><td rowspan=\"4\">ResNet34</td><td>X</td><td>38.1</td><td>56.8</td><td>40.7</td><td>33.0</td><td>53.2</td><td>35.3</td></tr><tr><td>ResNet50</td><td>38.4 (+0.3)</td><td>57.0 (+0.2)</td><td>41.0 (+0.3)</td><td>33.3 (+0.3)</td><td>53.6 (+0.4)</td><td>35.4 (+0.1)</td></tr><tr><td>ResNet101</td><td>38.5 (+0.4)</td><td>57.3 (+0.5)</td><td>41.4 (+0.7)</td><td>33.6 (+0.6)</td><td>54.1 (+0.9)</td><td>35.6 (+0.3)</td></tr><tr><td>ResNet152</td><td>38.4 (+0.3)</td><td>57.0 (+0.2)</td><td>41.0 (+0.3)</td><td>33.3 (+0.3)</td><td>53.7 (+0.5)</td><td>35.3 (+0.0)</td></tr></table>",
+                                            "type": "table",
+                                            "image_path": "6d3d5a9fb7c5054b18a1d0cede779a4500b23d1f19747e3aeb70e1db8094b73e.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 17,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        109,
+                                        319,
+                                        502,
+                                        348.3333333333333
+                                    ],
+                                    "spans": [],
+                                    "index": 16
+                                },
+                                {
+                                    "bbox": [
+                                        109,
+                                        348.3333333333333,
+                                        502,
+                                        377.66666666666663
+                                    ],
+                                    "spans": [],
+                                    "index": 17
+                                },
+                                {
+                                    "bbox": [
+                                        109,
+                                        377.66666666666663,
+                                        502,
+                                        406.99999999999994
+                                    ],
+                                    "spans": [],
+                                    "index": 18
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 15.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        125,
+                        434,
+                        505,
+                        489
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                124,
+                                433,
+                                505,
+                                446
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        124,
+                                        433,
+                                        505,
+                                        446
+                                    ],
+                                    "score": 1.0,
+                                    "content": "SUN-397: We further extend our transferring evaluation to the scene dataset SUN-397 for a more",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                124,
+                                445,
+                                505,
+                                458
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        124,
+                                        445,
+                                        505,
+                                        458
+                                    ],
+                                    "score": 1.0,
+                                    "content": "diverse testing. The official dataset specifies 10 different train/test splits, with each contains",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                124,
+                                456,
+                                505,
+                                468
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        124,
+                                        456,
+                                        505,
+                                        468
+                                    ],
+                                    "score": 1.0,
+                                    "content": "50 images per category covering 397 different scenes. We follow (Chen et al., 2020a; Grill",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                123,
+                                467,
+                                505,
+                                480
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        123,
+                                        467,
+                                        505,
+                                        480
+                                    ],
+                                    "score": 1.0,
+                                    "content": "et al., 2020) and use the first train/test split. For the validation set, we randomly pick 10 images",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                124,
+                                478,
+                                435,
+                                491
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        124,
+                                        478,
+                                        163,
+                                        491
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(yielding",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        163,
+                                        478,
+                                        183,
+                                        489
+                                    ],
+                                    "score": 0.86,
+                                    "content": "20 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        183,
+                                        478,
+                                        435,
+                                        491
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of the dataset), with identical optimizer parameters as CIFAR.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        }
+                    ],
+                    "index": 21
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        498,
+                        505,
+                        598
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                499,
+                                505,
+                                511
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        499,
+                                        505,
+                                        511
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Object Detection and Instance Segmentation. As indicated by (He et al., 2020), features produced",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                510,
+                                505,
+                                522
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        510,
+                                        505,
+                                        522
+                                    ],
+                                    "score": 1.0,
+                                    "content": "by self-supervised pre-training have divergent distributions in downstream tasks, thus resulting the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                520,
+                                505,
+                                533
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        520,
+                                        505,
+                                        533
+                                    ],
+                                    "score": 1.0,
+                                    "content": "supervised pre-training picked hyper-parameters not applicable. To relieve this, He et al. (2020)",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                531,
+                                506,
+                                545
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        531,
+                                        506,
+                                        545
+                                    ],
+                                    "score": 1.0,
+                                    "content": "uses feature normalization during the fine-tuning phase and train the BN layers. Different from",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                542,
+                                506,
+                                555
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        542,
+                                        506,
+                                        555
+                                    ],
+                                    "score": 1.0,
+                                    "content": "previous transferring and linear evaluations where we exploit only offline features, model for detection",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                554,
+                                505,
+                                565
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        554,
+                                        505,
+                                        565
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and segmentation is trained with all parameters tuned. For this reason, annotations on COCO for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                564,
+                                506,
+                                578
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        564,
+                                        506,
+                                        578
+                                    ],
+                                    "score": 1.0,
+                                    "content": "segmentation gives much higher influence for the backbone model than the VOC dataset (see Table 9),",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                575,
+                                505,
+                                588
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        575,
+                                        505,
+                                        588
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and gives an offset to the pre-training difference (see Table 10). Thus, this makes the performance",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                586,
+                                506,
+                                599
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        586,
+                                        506,
+                                        599
+                                    ],
+                                    "score": 1.0,
+                                    "content": "boosting by pre-training less obvious, and leads to trivial AP differences before and after distillation.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        }
+                    ],
+                    "index": 28
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        126,
+                        606,
+                        505,
+                        673
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                125,
+                                607,
+                                505,
+                                618
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        125,
+                                        607,
+                                        505,
+                                        618
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Object Detection on PASCAL VOC-07: We train a C4 (He et al., 2017) based Faster R-CNN (Ren",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                123,
+                                617,
+                                505,
+                                630
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        123,
+                                        617,
+                                        505,
+                                        630
+                                    ],
+                                    "score": 1.0,
+                                    "content": "et al., 2015) as the detector with different ResNet architectures (ResNet-18, ResNet-34 and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                123,
+                                627,
+                                506,
+                                642
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        123,
+                                        627,
+                                        506,
+                                        642
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ResNet-50) for evaluating the transferability of features for object detection tasks. We use",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                123,
+                                639,
+                                505,
+                                652
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        123,
+                                        639,
+                                        505,
+                                        652
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Detectron2 (Wu et al., 2019) for the implementations. We train our detector for 48k iterations",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                124,
+                                650,
+                                505,
+                                663
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        124,
+                                        650,
+                                        505,
+                                        663
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with a batch size of 32 (8 images per GPU). The base learning rate is set to 0.01 with 200",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                124,
+                                662,
+                                503,
+                                675
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        124,
+                                        662,
+                                        503,
+                                        675
+                                    ],
+                                    "score": 1.0,
+                                    "content": "warm-up iterations. We set the scale of images for training as [400, 800] and 800 at inference.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        }
+                    ],
+                    "index": 35.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        127,
+                        676,
+                        504,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                124,
+                                676,
+                                506,
+                                689
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        124,
+                                        676,
+                                        506,
+                                        689
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Object Detection and Segmentation on COCO: We use Mask R-CNN (He et al., 2017) with the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                124,
+                                687,
+                                504,
+                                701
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        124,
+                                        687,
+                                        489,
+                                        701
+                                    ],
+                                    "score": 1.0,
+                                    "content": "C4 backbone for the object detection and instance segmentation task on COCO dataset, with",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        489,
+                                        688,
+                                        504,
+                                        698
+                                    ],
+                                    "score": 0.85,
+                                    "content": "2 \\times",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                124,
+                                698,
+                                505,
+                                711
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        124,
+                                        698,
+                                        505,
+                                        711
+                                    ],
+                                    "score": 1.0,
+                                    "content": "schedule. Similar to the VOC detection, we tune the BN layers and all parameters. The model is",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                124,
+                                709,
+                                505,
+                                722
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        124,
+                                        709,
+                                        171,
+                                        722
+                                    ],
+                                    "score": 1.0,
+                                    "content": "trained for",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        171,
+                                        710,
+                                        192,
+                                        720
+                                    ],
+                                    "score": 0.26,
+                                    "content": "1 8 0 \\mathrm { k }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        192,
+                                        709,
+                                        505,
+                                        722
+                                    ],
+                                    "score": 1.0,
+                                    "content": "iterations with initial learning rate set to 0.02. We set the scale of images for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                124,
+                                720,
+                                300,
+                                733
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        124,
+                                        720,
+                                        300,
+                                        733
+                                    ],
+                                    "score": 1.0,
+                                    "content": "training as [600, 800] and 800 at inference.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        }
+                    ],
+                    "index": 41
+                }
+            ],
+            "page_idx": 15,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        107,
+                        27,
+                        292,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                26,
+                                293,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        26,
+                                        293,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Published as a conference paper at ICLR 2021",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        300,
+                        752,
+                        310,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                299,
+                                750,
+                                312,
+                                764
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        299,
+                                        750,
+                                        312,
+                                        764
+                                    ],
+                                    "score": 1.0,
+                                    "content": "",
+                                    "type": "text",
+                                    "height": 14,
+                                    "width": 13
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "table",
+                    "bbox": [
+                        186,
+                        121,
+                        421,
+                        261
+                    ],
+                    "blocks": [
+                        {
+                            "type": "table_caption",
+                            "bbox": [
+                                106,
+                                81,
+                                503,
+                                111
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        80,
+                                        505,
+                                        92
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                104,
+                                                80,
+                                                443,
+                                                92
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Table 9: Object detection and instance segmentation fine-tuned on VOC07: bounding-box AP",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                443,
+                                                80,
+                                                466,
+                                                91
+                                            ],
+                                            "score": 0.81,
+                                            "content": "( \\mathsf { A P } ^ { b b } )",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                466,
+                                                80,
+                                                505,
+                                                92
+                                            ],
+                                            "score": 1.0,
+                                            "content": ") and mask",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 0
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        90,
+                                        505,
+                                        102
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                90,
+                                                121,
+                                                102
+                                            ],
+                                            "score": 1.0,
+                                            "content": "AP",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                121,
+                                                91,
+                                                149,
+                                                101
+                                            ],
+                                            "score": 0.82,
+                                            "content": "( \\mathbf { A } \\mathbf { P } ^ { m k } )",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                149,
+                                                90,
+                                                505,
+                                                102
+                                            ],
+                                            "score": 1.0,
+                                            "content": ") evaluated on VOC07-val. The first row shows the baseline from MoCo-v2 backbones without",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        100,
+                                        150,
+                                        112
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                100,
+                                                150,
+                                                112
+                                            ],
+                                            "score": 1.0,
+                                            "content": "distillation.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 2
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "type": "table_body",
+                            "bbox": [
+                                186,
+                                121,
+                                421,
+                                261
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        186,
+                                        121,
+                                        421,
+                                        261
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                186,
+                                                121,
+                                                421,
+                                                261
+                                            ],
+                                            "score": 0.976,
+                                            "html": "<table><tr><td rowspan=\"2\">Student</td><td rowspan=\"2\">Teacher</td><td colspan=\"3\">VOC Object Detection</td></tr><tr><td>APbb</td><td>AP</td><td>AP</td></tr><tr><td rowspan=\"3\">ResNet-34</td><td>X</td><td>53.6</td><td>79.1</td><td>58.7</td></tr><tr><td>ResNet-50</td><td>53.7 (+0.1)</td><td>79.4 (+0.3)</td><td>59.2 (+0.5)</td></tr><tr><td>ResNet-101 ResNet-152</td><td>54.1 (+0.5) 54.4 (+0.8)</td><td>79.8 (+0.7) 80.1 (+1.0)</td><td>59.1 (+0.4) ） 59.9 (+1.2)</td></tr><tr><td rowspan=\"2\">ResNet-50</td><td>X</td><td>57.0</td><td>82.4</td><td>63.6</td></tr><tr><td>ResNet-50 ResNet-101</td><td>57.0 (+0.0) 57.1 (+0.1)</td><td>82.4 (+0.0) 82.8 (+0.4)</td><td>63.6 (+0.0) 63.8 (+0.2)</td></tr></table>",
+                                            "type": "table",
+                                            "image_path": "5c3f7a898d26a2d2fe5fcf5d40e410578494648f20defdcd18389a3531ae57cb.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 7.5,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        186,
+                                        121,
+                                        421,
+                                        135.0
+                                    ],
+                                    "spans": [],
+                                    "index": 3
+                                },
+                                {
+                                    "bbox": [
+                                        186,
+                                        135.0,
+                                        421,
+                                        149.0
+                                    ],
+                                    "spans": [],
+                                    "index": 4
+                                },
+                                {
+                                    "bbox": [
+                                        186,
+                                        149.0,
+                                        421,
+                                        163.0
+                                    ],
+                                    "spans": [],
+                                    "index": 5
+                                },
+                                {
+                                    "bbox": [
+                                        186,
+                                        163.0,
+                                        421,
+                                        177.0
+                                    ],
+                                    "spans": [],
+                                    "index": 6
+                                },
+                                {
+                                    "bbox": [
+                                        186,
+                                        177.0,
+                                        421,
+                                        191.0
+                                    ],
+                                    "spans": [],
+                                    "index": 7
+                                },
+                                {
+                                    "bbox": [
+                                        186,
+                                        191.0,
+                                        421,
+                                        205.0
+                                    ],
+                                    "spans": [],
+                                    "index": 8
+                                },
+                                {
+                                    "bbox": [
+                                        186,
+                                        205.0,
+                                        421,
+                                        219.0
+                                    ],
+                                    "spans": [],
+                                    "index": 9
+                                },
+                                {
+                                    "bbox": [
+                                        186,
+                                        219.0,
+                                        421,
+                                        233.0
+                                    ],
+                                    "spans": [],
+                                    "index": 10
+                                },
+                                {
+                                    "bbox": [
+                                        186,
+                                        233.0,
+                                        421,
+                                        247.0
+                                    ],
+                                    "spans": [],
+                                    "index": 11
+                                },
+                                {
+                                    "bbox": [
+                                        186,
+                                        247.0,
+                                        421,
+                                        261.0
+                                    ],
+                                    "spans": [],
+                                    "index": 12
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 4.25
+                },
+                {
+                    "type": "table",
+                    "bbox": [
+                        109,
+                        319,
+                        502,
+                        407
+                    ],
+                    "blocks": [
+                        {
+                            "type": "table_caption",
+                            "bbox": [
+                                106,
+                                280,
+                                504,
+                                310
+                            ],
+                            "group_id": 1,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        279,
+                                        505,
+                                        290
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                104,
+                                                279,
+                                                462,
+                                                290
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Table 10: Object detection and instance segmentation fine-tuned on COCO: bounding-box AP",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                462,
+                                                279,
+                                                487,
+                                                290
+                                            ],
+                                            "score": 0.84,
+                                            "content": "( \\mathbf { A P } ^ { b b } )",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                488,
+                                                279,
+                                                505,
+                                                290
+                                            ],
+                                            "score": 1.0,
+                                            "content": "and",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 13
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        288,
+                                        506,
+                                        301
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                288,
+                                                142,
+                                                301
+                                            ],
+                                            "score": 1.0,
+                                            "content": "mask AP",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                142,
+                                                289,
+                                                170,
+                                                300
+                                            ],
+                                            "score": 0.82,
+                                            "content": "( \\mathbf { A } \\mathbf { P } ^ { m k } )",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                171,
+                                                288,
+                                                506,
+                                                301
+                                            ],
+                                            "score": 1.0,
+                                            "content": "evaluated on COCO-val2017. The first several rows show the baselines from unsupervised",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 14
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        299,
+                                        219,
+                                        311
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                299,
+                                                219,
+                                                311
+                                            ],
+                                            "score": 1.0,
+                                            "content": "backbones without distillation.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 15
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "type": "table_body",
+                            "bbox": [
+                                109,
+                                319,
+                                502,
+                                407
+                            ],
+                            "group_id": 1,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        109,
+                                        319,
+                                        502,
+                                        407
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                109,
+                                                319,
+                                                502,
+                                                407
+                                            ],
+                                            "score": 0.982,
+                                            "html": "<table><tr><td rowspan=\"2\">Student</td><td rowspan=\"2\">Teacher</td><td colspan=\"3\">Object Detection</td><td colspan=\"3\">Instance Segmentation</td></tr><tr><td>Apbb</td><td>AP</td><td>APP</td><td>Apmk</td><td>AP</td><td>AP</td></tr><tr><td rowspan=\"4\">ResNet34</td><td>X</td><td>38.1</td><td>56.8</td><td>40.7</td><td>33.0</td><td>53.2</td><td>35.3</td></tr><tr><td>ResNet50</td><td>38.4 (+0.3)</td><td>57.0 (+0.2)</td><td>41.0 (+0.3)</td><td>33.3 (+0.3)</td><td>53.6 (+0.4)</td><td>35.4 (+0.1)</td></tr><tr><td>ResNet101</td><td>38.5 (+0.4)</td><td>57.3 (+0.5)</td><td>41.4 (+0.7)</td><td>33.6 (+0.6)</td><td>54.1 (+0.9)</td><td>35.6 (+0.3)</td></tr><tr><td>ResNet152</td><td>38.4 (+0.3)</td><td>57.0 (+0.2)</td><td>41.0 (+0.3)</td><td>33.3 (+0.3)</td><td>53.7 (+0.5)</td><td>35.3 (+0.0)</td></tr></table>",
+                                            "type": "table",
+                                            "image_path": "6d3d5a9fb7c5054b18a1d0cede779a4500b23d1f19747e3aeb70e1db8094b73e.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 17,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        109,
+                                        319,
+                                        502,
+                                        348.3333333333333
+                                    ],
+                                    "spans": [],
+                                    "index": 16
+                                },
+                                {
+                                    "bbox": [
+                                        109,
+                                        348.3333333333333,
+                                        502,
+                                        377.66666666666663
+                                    ],
+                                    "spans": [],
+                                    "index": 17
+                                },
+                                {
+                                    "bbox": [
+                                        109,
+                                        377.66666666666663,
+                                        502,
+                                        406.99999999999994
+                                    ],
+                                    "spans": [],
+                                    "index": 18
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 15.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        125,
+                        434,
+                        505,
+                        489
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                124,
+                                433,
+                                505,
+                                446
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        124,
+                                        433,
+                                        505,
+                                        446
+                                    ],
+                                    "score": 1.0,
+                                    "content": "SUN-397: We further extend our transferring evaluation to the scene dataset SUN-397 for a more",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                124,
+                                445,
+                                505,
+                                458
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        124,
+                                        445,
+                                        505,
+                                        458
+                                    ],
+                                    "score": 1.0,
+                                    "content": "diverse testing. The official dataset specifies 10 different train/test splits, with each contains",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                124,
+                                456,
+                                505,
+                                468
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        124,
+                                        456,
+                                        505,
+                                        468
+                                    ],
+                                    "score": 1.0,
+                                    "content": "50 images per category covering 397 different scenes. We follow (Chen et al., 2020a; Grill",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                123,
+                                467,
+                                505,
+                                480
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        123,
+                                        467,
+                                        505,
+                                        480
+                                    ],
+                                    "score": 1.0,
+                                    "content": "et al., 2020) and use the first train/test split. For the validation set, we randomly pick 10 images",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                124,
+                                478,
+                                435,
+                                491
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        124,
+                                        478,
+                                        163,
+                                        491
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(yielding",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        163,
+                                        478,
+                                        183,
+                                        489
+                                    ],
+                                    "score": 0.86,
+                                    "content": "20 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        183,
+                                        478,
+                                        435,
+                                        491
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of the dataset), with identical optimizer parameters as CIFAR.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        }
+                    ],
+                    "index": 21,
+                    "bbox_fs": [
+                        123,
+                        433,
+                        505,
+                        491
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        498,
+                        505,
+                        598
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                499,
+                                505,
+                                511
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        499,
+                                        505,
+                                        511
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Object Detection and Instance Segmentation. As indicated by (He et al., 2020), features produced",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                510,
+                                505,
+                                522
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        510,
+                                        505,
+                                        522
+                                    ],
+                                    "score": 1.0,
+                                    "content": "by self-supervised pre-training have divergent distributions in downstream tasks, thus resulting the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                520,
+                                505,
+                                533
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        520,
+                                        505,
+                                        533
+                                    ],
+                                    "score": 1.0,
+                                    "content": "supervised pre-training picked hyper-parameters not applicable. To relieve this, He et al. (2020)",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                531,
+                                506,
+                                545
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        531,
+                                        506,
+                                        545
+                                    ],
+                                    "score": 1.0,
+                                    "content": "uses feature normalization during the fine-tuning phase and train the BN layers. Different from",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                542,
+                                506,
+                                555
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        542,
+                                        506,
+                                        555
+                                    ],
+                                    "score": 1.0,
+                                    "content": "previous transferring and linear evaluations where we exploit only offline features, model for detection",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                554,
+                                505,
+                                565
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        554,
+                                        505,
+                                        565
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and segmentation is trained with all parameters tuned. For this reason, annotations on COCO for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                564,
+                                506,
+                                578
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        564,
+                                        506,
+                                        578
+                                    ],
+                                    "score": 1.0,
+                                    "content": "segmentation gives much higher influence for the backbone model than the VOC dataset (see Table 9),",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                575,
+                                505,
+                                588
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        575,
+                                        505,
+                                        588
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and gives an offset to the pre-training difference (see Table 10). Thus, this makes the performance",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                586,
+                                506,
+                                599
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        586,
+                                        506,
+                                        599
+                                    ],
+                                    "score": 1.0,
+                                    "content": "boosting by pre-training less obvious, and leads to trivial AP differences before and after distillation.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        }
+                    ],
+                    "index": 28,
+                    "bbox_fs": [
+                        105,
+                        499,
+                        506,
+                        599
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        126,
+                        606,
+                        505,
+                        673
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                125,
+                                607,
+                                505,
+                                618
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        125,
+                                        607,
+                                        505,
+                                        618
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Object Detection on PASCAL VOC-07: We train a C4 (He et al., 2017) based Faster R-CNN (Ren",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                123,
+                                617,
+                                505,
+                                630
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        123,
+                                        617,
+                                        505,
+                                        630
+                                    ],
+                                    "score": 1.0,
+                                    "content": "et al., 2015) as the detector with different ResNet architectures (ResNet-18, ResNet-34 and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                123,
+                                627,
+                                506,
+                                642
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        123,
+                                        627,
+                                        506,
+                                        642
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ResNet-50) for evaluating the transferability of features for object detection tasks. We use",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                123,
+                                639,
+                                505,
+                                652
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        123,
+                                        639,
+                                        505,
+                                        652
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Detectron2 (Wu et al., 2019) for the implementations. We train our detector for 48k iterations",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                124,
+                                650,
+                                505,
+                                663
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        124,
+                                        650,
+                                        505,
+                                        663
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with a batch size of 32 (8 images per GPU). The base learning rate is set to 0.01 with 200",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                124,
+                                662,
+                                503,
+                                675
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        124,
+                                        662,
+                                        503,
+                                        675
+                                    ],
+                                    "score": 1.0,
+                                    "content": "warm-up iterations. We set the scale of images for training as [400, 800] and 800 at inference.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        }
+                    ],
+                    "index": 35.5,
+                    "bbox_fs": [
+                        123,
+                        607,
+                        506,
+                        675
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        127,
+                        676,
+                        504,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                124,
+                                676,
+                                506,
+                                689
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        124,
+                                        676,
+                                        506,
+                                        689
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Object Detection and Segmentation on COCO: We use Mask R-CNN (He et al., 2017) with the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                124,
+                                687,
+                                504,
+                                701
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        124,
+                                        687,
+                                        489,
+                                        701
+                                    ],
+                                    "score": 1.0,
+                                    "content": "C4 backbone for the object detection and instance segmentation task on COCO dataset, with",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        489,
+                                        688,
+                                        504,
+                                        698
+                                    ],
+                                    "score": 0.85,
+                                    "content": "2 \\times",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                124,
+                                698,
+                                505,
+                                711
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        124,
+                                        698,
+                                        505,
+                                        711
+                                    ],
+                                    "score": 1.0,
+                                    "content": "schedule. Similar to the VOC detection, we tune the BN layers and all parameters. The model is",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                124,
+                                709,
+                                505,
+                                722
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        124,
+                                        709,
+                                        171,
+                                        722
+                                    ],
+                                    "score": 1.0,
+                                    "content": "trained for",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        171,
+                                        710,
+                                        192,
+                                        720
+                                    ],
+                                    "score": 0.26,
+                                    "content": "1 8 0 \\mathrm { k }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        192,
+                                        709,
+                                        505,
+                                        722
+                                    ],
+                                    "score": 1.0,
+                                    "content": "iterations with initial learning rate set to 0.02. We set the scale of images for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                124,
+                                720,
+                                300,
+                                733
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        124,
+                                        720,
+                                        300,
+                                        733
+                                    ],
+                                    "score": 1.0,
+                                    "content": "training as [600, 800] and 800 at inference.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        }
+                    ],
+                    "index": 41,
+                    "bbox_fs": [
+                        124,
+                        676,
+                        506,
+                        733
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "table",
+                    "bbox": [
+                        145,
+                        117,
+                        464,
+                        183
+                    ],
+                    "blocks": [
+                        {
+                            "type": "table_caption",
+                            "bbox": [
+                                106,
+                                87,
+                                504,
+                                108
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        87,
+                                        505,
+                                        99
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                87,
+                                                505,
+                                                99
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Table 11: Linear evaluations on ImageNet of ResNet-18 after distillation from the SWAV pre-trained ResNet-50",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 0
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        97,
+                                        321,
+                                        109
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                97,
+                                                321,
+                                                109
+                                            ],
+                                            "score": 1.0,
+                                            "content": "using either single view, cross-views, or small patch views.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 1
+                                }
+                            ],
+                            "index": 0.5
+                        },
+                        {
+                            "type": "table_body",
+                            "bbox": [
+                                145,
+                                117,
+                                464,
+                                183
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        145,
+                                        117,
+                                        464,
+                                        183
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                145,
+                                                117,
+                                                464,
+                                                183
+                                            ],
+                                            "score": 0.935,
+                                            "html": "<table><tr><td>Method</td><td>Multi-View(s)</td><td>Top-1 Acc.</td><td>Top-5 Acc.</td></tr><tr><td>Identical-View</td><td>1×224</td><td>61.7</td><td>84.2</td></tr><tr><td>Cross-Views</td><td>2×224</td><td>58.2</td><td>81.7</td></tr><tr><td>Multi-Crops + Cross-Views</td><td>1×224 + 6×96×96</td><td>61.9</td><td>84.4</td></tr><tr><td>Multi-Crops + Identical-View </td><td>1×224 +6×96×96</td><td>62.6</td><td>84.8</td></tr></table>",
+                                            "type": "table",
+                                            "image_path": "72a4faa1cf7e3c6038d27d854e17468ebd5461becf2efda07c70ea9634eb9831.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 3,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        145,
+                                        117,
+                                        464,
+                                        139.0
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                },
+                                {
+                                    "bbox": [
+                                        145,
+                                        139.0,
+                                        464,
+                                        161.0
+                                    ],
+                                    "spans": [],
+                                    "index": 3
+                                },
+                                {
+                                    "bbox": [
+                                        145,
+                                        161.0,
+                                        464,
+                                        183.0
+                                    ],
+                                    "spans": [],
+                                    "index": 4
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 1.75
+                },
+                {
+                    "type": "image",
+                    "bbox": [
+                        119,
+                        202,
+                        496,
+                        288
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                119,
+                                202,
+                                496,
+                                288
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        119,
+                                        202,
+                                        496,
+                                        288
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                119,
+                                                202,
+                                                496,
+                                                288
+                                            ],
+                                            "score": 0.961,
+                                            "type": "image",
+                                            "image_path": "0863ba5174cfabf1f7a4f95ce0b77df7bdbf71b58813b9c855df14a8e571182a.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 6,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        119,
+                                        202,
+                                        496,
+                                        230.66666666666666
+                                    ],
+                                    "spans": [],
+                                    "index": 5
+                                },
+                                {
+                                    "bbox": [
+                                        119,
+                                        230.66666666666666,
+                                        496,
+                                        259.3333333333333
+                                    ],
+                                    "spans": [],
+                                    "index": 6
+                                },
+                                {
+                                    "bbox": [
+                                        119,
+                                        259.3333333333333,
+                                        496,
+                                        288.0
+                                    ],
+                                    "spans": [],
+                                    "index": 7
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                108,
+                                296,
+                                505,
+                                326
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        296,
+                                        505,
+                                        307
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                296,
+                                                505,
+                                                307
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 6: We experiment with different strategies of using views during distillation, which include: (a). Identical",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 8
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        307,
+                                        505,
+                                        317
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                307,
+                                                505,
+                                                317
+                                            ],
+                                            "score": 1.0,
+                                            "content": "view for distillation. (b). Cross view distillation. (c). Large-small cross view distillation. (d). Large-small",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 9
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        316,
+                                        202,
+                                        327
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                316,
+                                                202,
+                                                327
+                                            ],
+                                            "score": 1.0,
+                                            "content": "identical view distillation.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 10
+                                }
+                            ],
+                            "index": 9
+                        }
+                    ],
+                    "index": 7.5
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        349,
+                        397,
+                        361
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                348,
+                                400,
+                                362
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        348,
+                                        400,
+                                        362
+                                    ],
+                                    "score": 1.0,
+                                    "content": "A.3 SINGLE CROP V.S. MULTI-CROPS VIEW(S) FOR DISTILLATION",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        }
+                    ],
+                    "index": 11
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        370,
+                        506,
+                        513
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                370,
+                                506,
+                                383
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        370,
+                                        506,
+                                        383
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In contrary with most contrastive SSL methods where two different augmented views of an image are",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                381,
+                                506,
+                                394
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        381,
+                                        506,
+                                        394
+                                    ],
+                                    "score": 1.0,
+                                    "content": "utilized as the positive samples (see Figure 6-a), SEED uses an identical view for each image (see",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                392,
+                                506,
+                                405
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        392,
+                                        506,
+                                        405
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Figure 6-b) during distillation and yields better performances, as is shown in Table. 11. In addition,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                403,
+                                505,
+                                415
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        403,
+                                        505,
+                                        415
+                                    ],
+                                    "score": 1.0,
+                                    "content": "we have also experimented with two strategies of using small patches. To be specific, we follow the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                414,
+                                506,
+                                427
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        414,
+                                        371,
+                                        427
+                                    ],
+                                    "score": 1.0,
+                                    "content": "set-up in SWAV (Caron et al., 2020), that 6 small patches of the size",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        371,
+                                        414,
+                                        400,
+                                        425
+                                    ],
+                                    "score": 0.9,
+                                    "content": "9 6 \\times 9 6",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        401,
+                                        414,
+                                        506,
+                                        427
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are sampled at the scale of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                424,
+                                506,
+                                438
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        424,
+                                        506,
+                                        438
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(0.05, 0.14). Then, we apply the same augmentations as introduced previously as data pre-processing.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                435,
+                                506,
+                                448
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        435,
+                                        506,
+                                        448
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Figure. 6-c shows the way that is similar in SWAV for small-patch learning, where both large and 6",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                446,
+                                506,
+                                460
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        446,
+                                        389,
+                                        460
+                                    ],
+                                    "score": 1.0,
+                                    "content": "small patches are fed into the student encoder, with the learning target",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        390,
+                                        446,
+                                        409,
+                                        459
+                                    ],
+                                    "score": 0.85,
+                                    "content": "( \\mathbf { z } ^ { T } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        409,
+                                        446,
+                                        506,
+                                        460
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to be the embedding of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                457,
+                                506,
+                                470
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        457,
+                                        506,
+                                        470
+                                    ],
+                                    "score": 1.0,
+                                    "content": "large view from the teacher encoder. Figure. 6-d is the strategy we use during distillation, that both",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                466,
+                                507,
+                                485
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        466,
+                                        430,
+                                        485
+                                    ],
+                                    "score": 1.0,
+                                    "content": "views are fed into student and teacher to produce the embeddings for small-views",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        430,
+                                        468,
+                                        465,
+                                        481
+                                    ],
+                                    "score": 0.91,
+                                    "content": "( \\mathbf { z } _ { s } ^ { S } , \\mathbf { z } _ { s } ^ { T } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        465,
+                                        466,
+                                        507,
+                                        485
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and large",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                479,
+                                507,
+                                493
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        479,
+                                        132,
+                                        493
+                                    ],
+                                    "score": 1.0,
+                                    "content": "views",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        132,
+                                        479,
+                                        167,
+                                        492
+                                    ],
+                                    "score": 0.91,
+                                    "content": "( \\mathbf { z } _ { l } ^ { S } , \\mathbf { z } _ { l } ^ { T } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        167,
+                                        479,
+                                        507,
+                                        493
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Based on that, the distillation is formulated separately on the small and large views.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                489,
+                                506,
+                                504
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        489,
+                                        506,
+                                        504
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Notably, we maintain two independent queues for storing historical data samples for the large and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                502,
+                                158,
+                                513
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        502,
+                                        158,
+                                        513
+                                    ],
+                                    "score": 1.0,
+                                    "content": "small views.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        }
+                    ],
+                    "index": 18
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        106,
+                        528,
+                        347,
+                        539
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                528,
+                                349,
+                                541
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        528,
+                                        349,
+                                        541
+                                    ],
+                                    "score": 1.0,
+                                    "content": "A.4 STRATEGIES FOR OTHER DISTILLATION METHODS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        }
+                    ],
+                    "index": 25
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        549,
+                        405,
+                        560
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                547,
+                                405,
+                                562
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        547,
+                                        405,
+                                        562
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We compare the effect of distillation using different strategies with SEED.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        }
+                    ],
+                    "index": 26
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        566,
+                        503,
+                        588
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                565,
+                                505,
+                                578
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        566,
+                                        115,
+                                        576
+                                    ],
+                                    "score": 0.39,
+                                    "content": "\\mathbf { \\xi } _ { l 2 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        115,
+                                        565,
+                                        385,
+                                        578
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-Distance: We train the student encoder by minimizing the squared",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        385,
+                                        567,
+                                        394,
+                                        576
+                                    ],
+                                    "score": 0.66,
+                                    "content": "l 2",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        395,
+                                        565,
+                                        505,
+                                        578
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-distance of representations",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                575,
+                                348,
+                                591
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        575,
+                                        160,
+                                        591
+                                    ],
+                                    "score": 1.0,
+                                    "content": "from student",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        160,
+                                        576,
+                                        178,
+                                        589
+                                    ],
+                                    "score": 0.85,
+                                    "content": "( \\mathbf { z } _ { i } ^ { S } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        178,
+                                        575,
+                                        228,
+                                        591
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and teacher",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        228,
+                                        576,
+                                        246,
+                                        589
+                                    ],
+                                    "score": 0.76,
+                                    "content": "( \\mathbf { z } _ { i } ^ { T } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        246,
+                                        575,
+                                        332,
+                                        591
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for an identical view",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        333,
+                                        579,
+                                        343,
+                                        588
+                                    ],
+                                    "score": 0.83,
+                                    "content": "\\mathbf { x } _ { i }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        343,
+                                        575,
+                                        348,
+                                        591
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        }
+                    ],
+                    "index": 27.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        594,
+                        505,
+                        682
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                593,
+                                506,
+                                607
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        594,
+                                        114,
+                                        604
+                                    ],
+                                    "score": 0.39,
+                                    "content": "\\pmb { K }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        115,
+                                        593,
+                                        250,
+                                        607
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-Means: We experiment with the",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        251,
+                                        595,
+                                        260,
+                                        604
+                                    ],
+                                    "score": 0.68,
+                                    "content": "K",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        260,
+                                        593,
+                                        506,
+                                        607
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-Means clustering method to retrieve pseudo class labels for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                605,
+                                506,
+                                618
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        605,
+                                        404,
+                                        618
+                                    ],
+                                    "score": 1.0,
+                                    "content": "distillation. Specifically, we first extract offline image features using the",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        405,
+                                        605,
+                                        422,
+                                        615
+                                    ],
+                                    "score": 0.41,
+                                    "content": "S S L",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        423,
+                                        605,
+                                        506,
+                                        618
+                                    ],
+                                    "score": 1.0,
+                                    "content": "pre-trained Teacher",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                616,
+                                505,
+                                628
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        616,
+                                        403,
+                                        628
+                                    ],
+                                    "score": 1.0,
+                                    "content": "network without any image augmentations. Based on this, we conduct our",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        403,
+                                        617,
+                                        412,
+                                        626
+                                    ],
+                                    "score": 0.71,
+                                    "content": "K",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        412,
+                                        616,
+                                        505,
+                                        628
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-Means clustering with",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                627,
+                                505,
+                                639
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        627,
+                                        118,
+                                        637
+                                    ],
+                                    "score": 0.28,
+                                    "content": "4 \\mathrm { k }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        118,
+                                        627,
+                                        135,
+                                        639
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        135,
+                                        627,
+                                        151,
+                                        637
+                                    ],
+                                    "score": 0.39,
+                                    "content": "1 6 \\mathrm { k }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        151,
+                                        627,
+                                        505,
+                                        639
+                                    ],
+                                    "score": 1.0,
+                                    "content": "unique centroids. Then the final centroids are used to produce pseudo labels for unlabelled",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                637,
+                                505,
+                                652
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        637,
+                                        505,
+                                        652
+                                    ],
+                                    "score": 1.0,
+                                    "content": "instances. With that, we carry out the distillation by training the model on a classification task using",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                648,
+                                506,
+                                662
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        648,
+                                        506,
+                                        662
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the produced labels as the ground-truth. To avoid trivial solutions that the majority of images are",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                660,
+                                506,
+                                672
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        660,
+                                        506,
+                                        672
+                                    ],
+                                    "score": 1.0,
+                                    "content": "assigned to a few clusters, we sample images based on a uniform distribution over pseudo-labels as",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                670,
+                                462,
+                                684
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        670,
+                                        462,
+                                        684
+                                    ],
+                                    "score": 1.0,
+                                    "content": "clustering proceeds. We observe very close results when adjusting numbers of centroids.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        }
+                    ],
+                    "index": 32.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        687,
+                        504,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                686,
+                                506,
+                                700
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        686,
+                                        210,
+                                        700
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Online-Clustering: With",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        211,
+                                        688,
+                                        219,
+                                        698
+                                    ],
+                                    "score": 0.69,
+                                    "content": "K",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        220,
+                                        686,
+                                        506,
+                                        700
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-Means for pseudo-label generation training, it does not lead to satisfying",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                698,
+                                505,
+                                711
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        698,
+                                        136,
+                                        711
+                                    ],
+                                    "score": 1.0,
+                                    "content": "results",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        136,
+                                        699,
+                                        164,
+                                        710
+                                    ],
+                                    "score": 0.82,
+                                    "content": "( 5 1 . 0 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        164,
+                                        698,
+                                        505,
+                                        711
+                                    ],
+                                    "score": 1.0,
+                                    "content": "on ResNet-18 with ResNet-50 as Teacher) as instances might have not been accurately",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                710,
+                                506,
+                                721
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        710,
+                                        506,
+                                        721
+                                    ],
+                                    "score": 1.0,
+                                    "content": "categorized by limited frozen centroids. Similar to (Caron et al., 2018; Li et al., 2020), we resort",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                720,
+                                506,
+                                732
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        720,
+                                        377,
+                                        732
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to the “in-batch” and dynamical clustering to substitute the frozen",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        378,
+                                        722,
+                                        386,
+                                        730
+                                    ],
+                                    "score": 0.73,
+                                    "content": "K",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        386,
+                                        720,
+                                        506,
+                                        732
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-Means method. We conduct",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        }
+                    ],
+                    "index": 38.5
+                }
+            ],
+            "page_idx": 16,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        108,
+                        27,
+                        292,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                26,
+                                293,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        26,
+                                        293,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Published as a conference paper at ICLR 2021",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        300,
+                        751,
+                        310,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                299,
+                                750,
+                                312,
+                                764
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        299,
+                                        750,
+                                        312,
+                                        764
+                                    ],
+                                    "score": 1.0,
+                                    "content": "",
+                                    "type": "text",
+                                    "height": 14,
+                                    "width": 13
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "table",
+                    "bbox": [
+                        145,
+                        117,
+                        464,
+                        183
+                    ],
+                    "blocks": [
+                        {
+                            "type": "table_caption",
+                            "bbox": [
+                                106,
+                                87,
+                                504,
+                                108
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        87,
+                                        505,
+                                        99
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                87,
+                                                505,
+                                                99
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Table 11: Linear evaluations on ImageNet of ResNet-18 after distillation from the SWAV pre-trained ResNet-50",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 0
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        97,
+                                        321,
+                                        109
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                97,
+                                                321,
+                                                109
+                                            ],
+                                            "score": 1.0,
+                                            "content": "using either single view, cross-views, or small patch views.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 1
+                                }
+                            ],
+                            "index": 0.5
+                        },
+                        {
+                            "type": "table_body",
+                            "bbox": [
+                                145,
+                                117,
+                                464,
+                                183
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        145,
+                                        117,
+                                        464,
+                                        183
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                145,
+                                                117,
+                                                464,
+                                                183
+                                            ],
+                                            "score": 0.935,
+                                            "html": "<table><tr><td>Method</td><td>Multi-View(s)</td><td>Top-1 Acc.</td><td>Top-5 Acc.</td></tr><tr><td>Identical-View</td><td>1×224</td><td>61.7</td><td>84.2</td></tr><tr><td>Cross-Views</td><td>2×224</td><td>58.2</td><td>81.7</td></tr><tr><td>Multi-Crops + Cross-Views</td><td>1×224 + 6×96×96</td><td>61.9</td><td>84.4</td></tr><tr><td>Multi-Crops + Identical-View </td><td>1×224 +6×96×96</td><td>62.6</td><td>84.8</td></tr></table>",
+                                            "type": "table",
+                                            "image_path": "72a4faa1cf7e3c6038d27d854e17468ebd5461becf2efda07c70ea9634eb9831.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 3,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        145,
+                                        117,
+                                        464,
+                                        139.0
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                },
+                                {
+                                    "bbox": [
+                                        145,
+                                        139.0,
+                                        464,
+                                        161.0
+                                    ],
+                                    "spans": [],
+                                    "index": 3
+                                },
+                                {
+                                    "bbox": [
+                                        145,
+                                        161.0,
+                                        464,
+                                        183.0
+                                    ],
+                                    "spans": [],
+                                    "index": 4
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 1.75
+                },
+                {
+                    "type": "image",
+                    "bbox": [
+                        119,
+                        202,
+                        496,
+                        288
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                119,
+                                202,
+                                496,
+                                288
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        119,
+                                        202,
+                                        496,
+                                        288
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                119,
+                                                202,
+                                                496,
+                                                288
+                                            ],
+                                            "score": 0.961,
+                                            "type": "image",
+                                            "image_path": "0863ba5174cfabf1f7a4f95ce0b77df7bdbf71b58813b9c855df14a8e571182a.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 6,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        119,
+                                        202,
+                                        496,
+                                        230.66666666666666
+                                    ],
+                                    "spans": [],
+                                    "index": 5
+                                },
+                                {
+                                    "bbox": [
+                                        119,
+                                        230.66666666666666,
+                                        496,
+                                        259.3333333333333
+                                    ],
+                                    "spans": [],
+                                    "index": 6
+                                },
+                                {
+                                    "bbox": [
+                                        119,
+                                        259.3333333333333,
+                                        496,
+                                        288.0
+                                    ],
+                                    "spans": [],
+                                    "index": 7
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                108,
+                                296,
+                                505,
+                                326
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        296,
+                                        505,
+                                        307
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                296,
+                                                505,
+                                                307
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 6: We experiment with different strategies of using views during distillation, which include: (a). Identical",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 8
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        307,
+                                        505,
+                                        317
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                307,
+                                                505,
+                                                317
+                                            ],
+                                            "score": 1.0,
+                                            "content": "view for distillation. (b). Cross view distillation. (c). Large-small cross view distillation. (d). Large-small",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 9
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        316,
+                                        202,
+                                        327
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                316,
+                                                202,
+                                                327
+                                            ],
+                                            "score": 1.0,
+                                            "content": "identical view distillation.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 10
+                                }
+                            ],
+                            "index": 9
+                        }
+                    ],
+                    "index": 7.5
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        349,
+                        397,
+                        361
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                348,
+                                400,
+                                362
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        348,
+                                        400,
+                                        362
+                                    ],
+                                    "score": 1.0,
+                                    "content": "A.3 SINGLE CROP V.S. MULTI-CROPS VIEW(S) FOR DISTILLATION",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        }
+                    ],
+                    "index": 11
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        370,
+                        506,
+                        513
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                370,
+                                506,
+                                383
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        370,
+                                        506,
+                                        383
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In contrary with most contrastive SSL methods where two different augmented views of an image are",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                381,
+                                506,
+                                394
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        381,
+                                        506,
+                                        394
+                                    ],
+                                    "score": 1.0,
+                                    "content": "utilized as the positive samples (see Figure 6-a), SEED uses an identical view for each image (see",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                392,
+                                506,
+                                405
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        392,
+                                        506,
+                                        405
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Figure 6-b) during distillation and yields better performances, as is shown in Table. 11. In addition,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                403,
+                                505,
+                                415
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        403,
+                                        505,
+                                        415
+                                    ],
+                                    "score": 1.0,
+                                    "content": "we have also experimented with two strategies of using small patches. To be specific, we follow the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                414,
+                                506,
+                                427
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        414,
+                                        371,
+                                        427
+                                    ],
+                                    "score": 1.0,
+                                    "content": "set-up in SWAV (Caron et al., 2020), that 6 small patches of the size",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        371,
+                                        414,
+                                        400,
+                                        425
+                                    ],
+                                    "score": 0.9,
+                                    "content": "9 6 \\times 9 6",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        401,
+                                        414,
+                                        506,
+                                        427
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are sampled at the scale of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                424,
+                                506,
+                                438
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        424,
+                                        506,
+                                        438
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(0.05, 0.14). Then, we apply the same augmentations as introduced previously as data pre-processing.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                435,
+                                506,
+                                448
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        435,
+                                        506,
+                                        448
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Figure. 6-c shows the way that is similar in SWAV for small-patch learning, where both large and 6",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                446,
+                                506,
+                                460
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        446,
+                                        389,
+                                        460
+                                    ],
+                                    "score": 1.0,
+                                    "content": "small patches are fed into the student encoder, with the learning target",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        390,
+                                        446,
+                                        409,
+                                        459
+                                    ],
+                                    "score": 0.85,
+                                    "content": "( \\mathbf { z } ^ { T } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        409,
+                                        446,
+                                        506,
+                                        460
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to be the embedding of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                457,
+                                506,
+                                470
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        457,
+                                        506,
+                                        470
+                                    ],
+                                    "score": 1.0,
+                                    "content": "large view from the teacher encoder. Figure. 6-d is the strategy we use during distillation, that both",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                466,
+                                507,
+                                485
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        466,
+                                        430,
+                                        485
+                                    ],
+                                    "score": 1.0,
+                                    "content": "views are fed into student and teacher to produce the embeddings for small-views",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        430,
+                                        468,
+                                        465,
+                                        481
+                                    ],
+                                    "score": 0.91,
+                                    "content": "( \\mathbf { z } _ { s } ^ { S } , \\mathbf { z } _ { s } ^ { T } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        465,
+                                        466,
+                                        507,
+                                        485
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and large",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                479,
+                                507,
+                                493
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        479,
+                                        132,
+                                        493
+                                    ],
+                                    "score": 1.0,
+                                    "content": "views",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        132,
+                                        479,
+                                        167,
+                                        492
+                                    ],
+                                    "score": 0.91,
+                                    "content": "( \\mathbf { z } _ { l } ^ { S } , \\mathbf { z } _ { l } ^ { T } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        167,
+                                        479,
+                                        507,
+                                        493
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Based on that, the distillation is formulated separately on the small and large views.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                489,
+                                506,
+                                504
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        489,
+                                        506,
+                                        504
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Notably, we maintain two independent queues for storing historical data samples for the large and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                502,
+                                158,
+                                513
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        502,
+                                        158,
+                                        513
+                                    ],
+                                    "score": 1.0,
+                                    "content": "small views.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        }
+                    ],
+                    "index": 18,
+                    "bbox_fs": [
+                        104,
+                        370,
+                        507,
+                        513
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        106,
+                        528,
+                        347,
+                        539
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                528,
+                                349,
+                                541
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        528,
+                                        349,
+                                        541
+                                    ],
+                                    "score": 1.0,
+                                    "content": "A.4 STRATEGIES FOR OTHER DISTILLATION METHODS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        }
+                    ],
+                    "index": 25
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        549,
+                        405,
+                        560
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                547,
+                                405,
+                                562
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        547,
+                                        405,
+                                        562
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We compare the effect of distillation using different strategies with SEED.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        }
+                    ],
+                    "index": 26,
+                    "bbox_fs": [
+                        105,
+                        547,
+                        405,
+                        562
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        566,
+                        503,
+                        588
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                565,
+                                505,
+                                578
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        566,
+                                        115,
+                                        576
+                                    ],
+                                    "score": 0.39,
+                                    "content": "\\mathbf { \\xi } _ { l 2 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        115,
+                                        565,
+                                        385,
+                                        578
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-Distance: We train the student encoder by minimizing the squared",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        385,
+                                        567,
+                                        394,
+                                        576
+                                    ],
+                                    "score": 0.66,
+                                    "content": "l 2",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        395,
+                                        565,
+                                        505,
+                                        578
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-distance of representations",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                575,
+                                348,
+                                591
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        575,
+                                        160,
+                                        591
+                                    ],
+                                    "score": 1.0,
+                                    "content": "from student",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        160,
+                                        576,
+                                        178,
+                                        589
+                                    ],
+                                    "score": 0.85,
+                                    "content": "( \\mathbf { z } _ { i } ^ { S } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        178,
+                                        575,
+                                        228,
+                                        591
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and teacher",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        228,
+                                        576,
+                                        246,
+                                        589
+                                    ],
+                                    "score": 0.76,
+                                    "content": "( \\mathbf { z } _ { i } ^ { T } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        246,
+                                        575,
+                                        332,
+                                        591
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for an identical view",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        333,
+                                        579,
+                                        343,
+                                        588
+                                    ],
+                                    "score": 0.83,
+                                    "content": "\\mathbf { x } _ { i }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        343,
+                                        575,
+                                        348,
+                                        591
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        }
+                    ],
+                    "index": 27.5,
+                    "bbox_fs": [
+                        105,
+                        565,
+                        505,
+                        591
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        594,
+                        505,
+                        682
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                593,
+                                506,
+                                607
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        594,
+                                        114,
+                                        604
+                                    ],
+                                    "score": 0.39,
+                                    "content": "\\pmb { K }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        115,
+                                        593,
+                                        250,
+                                        607
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-Means: We experiment with the",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        251,
+                                        595,
+                                        260,
+                                        604
+                                    ],
+                                    "score": 0.68,
+                                    "content": "K",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        260,
+                                        593,
+                                        506,
+                                        607
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-Means clustering method to retrieve pseudo class labels for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                605,
+                                506,
+                                618
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        605,
+                                        404,
+                                        618
+                                    ],
+                                    "score": 1.0,
+                                    "content": "distillation. Specifically, we first extract offline image features using the",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        405,
+                                        605,
+                                        422,
+                                        615
+                                    ],
+                                    "score": 0.41,
+                                    "content": "S S L",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        423,
+                                        605,
+                                        506,
+                                        618
+                                    ],
+                                    "score": 1.0,
+                                    "content": "pre-trained Teacher",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                616,
+                                505,
+                                628
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        616,
+                                        403,
+                                        628
+                                    ],
+                                    "score": 1.0,
+                                    "content": "network without any image augmentations. Based on this, we conduct our",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        403,
+                                        617,
+                                        412,
+                                        626
+                                    ],
+                                    "score": 0.71,
+                                    "content": "K",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        412,
+                                        616,
+                                        505,
+                                        628
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-Means clustering with",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                627,
+                                505,
+                                639
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        627,
+                                        118,
+                                        637
+                                    ],
+                                    "score": 0.28,
+                                    "content": "4 \\mathrm { k }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        118,
+                                        627,
+                                        135,
+                                        639
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        135,
+                                        627,
+                                        151,
+                                        637
+                                    ],
+                                    "score": 0.39,
+                                    "content": "1 6 \\mathrm { k }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        151,
+                                        627,
+                                        505,
+                                        639
+                                    ],
+                                    "score": 1.0,
+                                    "content": "unique centroids. Then the final centroids are used to produce pseudo labels for unlabelled",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                637,
+                                505,
+                                652
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        637,
+                                        505,
+                                        652
+                                    ],
+                                    "score": 1.0,
+                                    "content": "instances. With that, we carry out the distillation by training the model on a classification task using",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                648,
+                                506,
+                                662
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        648,
+                                        506,
+                                        662
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the produced labels as the ground-truth. To avoid trivial solutions that the majority of images are",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                660,
+                                506,
+                                672
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        660,
+                                        506,
+                                        672
+                                    ],
+                                    "score": 1.0,
+                                    "content": "assigned to a few clusters, we sample images based on a uniform distribution over pseudo-labels as",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                670,
+                                462,
+                                684
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        670,
+                                        462,
+                                        684
+                                    ],
+                                    "score": 1.0,
+                                    "content": "clustering proceeds. We observe very close results when adjusting numbers of centroids.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        }
+                    ],
+                    "index": 32.5,
+                    "bbox_fs": [
+                        104,
+                        593,
+                        506,
+                        684
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        687,
+                        504,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                686,
+                                506,
+                                700
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        686,
+                                        210,
+                                        700
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Online-Clustering: With",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        211,
+                                        688,
+                                        219,
+                                        698
+                                    ],
+                                    "score": 0.69,
+                                    "content": "K",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        220,
+                                        686,
+                                        506,
+                                        700
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-Means for pseudo-label generation training, it does not lead to satisfying",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                698,
+                                505,
+                                711
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        698,
+                                        136,
+                                        711
+                                    ],
+                                    "score": 1.0,
+                                    "content": "results",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        136,
+                                        699,
+                                        164,
+                                        710
+                                    ],
+                                    "score": 0.82,
+                                    "content": "( 5 1 . 0 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        164,
+                                        698,
+                                        505,
+                                        711
+                                    ],
+                                    "score": 1.0,
+                                    "content": "on ResNet-18 with ResNet-50 as Teacher) as instances might have not been accurately",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                710,
+                                506,
+                                721
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        710,
+                                        506,
+                                        721
+                                    ],
+                                    "score": 1.0,
+                                    "content": "categorized by limited frozen centroids. Similar to (Caron et al., 2018; Li et al., 2020), we resort",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                720,
+                                506,
+                                732
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        720,
+                                        377,
+                                        732
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to the “in-batch” and dynamical clustering to substitute the frozen",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        378,
+                                        722,
+                                        386,
+                                        730
+                                    ],
+                                    "score": 0.73,
+                                    "content": "K",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        386,
+                                        720,
+                                        506,
+                                        732
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-Means method. We conduct",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        }
+                    ],
+                    "index": 38.5,
+                    "bbox_fs": [
+                        105,
+                        686,
+                        506,
+                        732
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        82,
+                        504,
+                        116
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                81,
+                                506,
+                                95
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        83,
+                                        114,
+                                        92
+                                    ],
+                                    "score": 0.67,
+                                    "content": "K",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        115,
+                                        81,
+                                        506,
+                                        95
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-Means clustering within a batch and continuously update the centroid based on the teacher feature",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                94,
+                                505,
+                                105
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        94,
+                                        505,
+                                        105
+                                    ],
+                                    "score": 1.0,
+                                    "content": "representation as distillation goes on. This alleviates the above problems and yields a substantial",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                105,
+                                314,
+                                116
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        105,
+                                        282,
+                                        116
+                                    ],
+                                    "score": 1.0,
+                                    "content": "performance improvement on ResNet-18 to",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        282,
+                                        105,
+                                        309,
+                                        115
+                                    ],
+                                    "score": 0.86,
+                                    "content": "5 6 . 4 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        309,
+                                        105,
+                                        314,
+                                        116
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        }
+                    ],
+                    "index": 1
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        120,
+                        505,
+                        166
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                121,
+                                505,
+                                133
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        121,
+                                        505,
+                                        133
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Binary Contrastive Loss: We resort to CRD (Tian et al., 2019a) and adopt an info-NCE loss-alike",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                132,
+                                505,
+                                145
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        132,
+                                        505,
+                                        145
+                                    ],
+                                    "score": 1.0,
+                                    "content": "training objective in unsupervised distillation tasks. Specifically, we treat representation features",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                142,
+                                505,
+                                156
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        142,
+                                        258,
+                                        156
+                                    ],
+                                    "score": 1.0,
+                                    "content": "from Teacher and Student for instance",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        259,
+                                        145,
+                                        270,
+                                        154
+                                    ],
+                                    "score": 0.87,
+                                    "content": "\\mathbf { x } _ { i }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        270,
+                                        142,
+                                        448,
+                                        156
+                                    ],
+                                    "score": 1.0,
+                                    "content": "as positive pairs, and random instances from",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        448,
+                                        144,
+                                        458,
+                                        154
+                                    ],
+                                    "score": 0.64,
+                                    "content": "\\mathbf { D }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        459,
+                                        142,
+                                        505,
+                                        156
+                                    ],
+                                    "score": 1.0,
+                                    "content": "as negative",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                154,
+                                146,
+                                168
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        154,
+                                        146,
+                                        168
+                                    ],
+                                    "score": 1.0,
+                                    "content": "samples:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        }
+                    ],
+                    "index": 4.5
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        172,
+                        166,
+                        427,
+                        200
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                172,
+                                166,
+                                427,
+                                200
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        172,
+                                        166,
+                                        427,
+                                        200
+                                    ],
+                                    "score": 0.94,
+                                    "content": "\\hat { \\theta } _ { S } = \\underset { \\theta _ { S } } { \\arg \\operatorname* { m i n } } \\sum _ { i } ^ { N } \\log h ( \\mathbf { z } _ { i } ^ { S } , \\mathbf { z } _ { i } ^ { T } ) + K \\cdot [ \\log h ( 1 - h ( \\mathbf { z } _ { i } ^ { S } , \\mathbf { d } _ { j } ^ { T } ) ) ] ,",
+                                    "type": "interline_equation",
+                                    "image_path": "7efa8646a999cf9b57dbdc3eab933940b7c76683e63c8311e334af00b5e6c4a4.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 8,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                172,
+                                166,
+                                427,
+                                177.33333333333334
+                            ],
+                            "spans": [],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                172,
+                                177.33333333333334,
+                                427,
+                                188.66666666666669
+                            ],
+                            "spans": [],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                172,
+                                188.66666666666669,
+                                427,
+                                200.00000000000003
+                            ],
+                            "spans": [],
+                            "index": 9
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        206,
+                        505,
+                        219
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                204,
+                                506,
+                                221
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        204,
+                                        133,
+                                        221
+                                    ],
+                                    "score": 1.0,
+                                    "content": "where",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        133,
+                                        205,
+                                        189,
+                                        220
+                                    ],
+                                    "score": 0.89,
+                                    "content": "\\mathbf { d } _ { j } ^ { T } \\in \\mathbf { D } , h ( \\cdot )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        189,
+                                        204,
+                                        339,
+                                        221
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is any family of functions that satisfy",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        339,
+                                        207,
+                                        346,
+                                        217
+                                    ],
+                                    "score": 0.49,
+                                    "content": "h",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        347,
+                                        204,
+                                        351,
+                                        221
+                                    ],
+                                    "score": 1.0,
+                                    "content": ":",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        351,
+                                        207,
+                                        414,
+                                        218
+                                    ],
+                                    "score": 0.88,
+                                    "content": "\\{ { \\bf z } , { \\bf d } \\}  [ 0 , 1 ]",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        415,
+                                        204,
+                                        506,
+                                        221
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", e.g., cosine similarity.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        }
+                    ],
+                    "index": 10
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        234,
+                        238,
+                        247
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                232,
+                                239,
+                                249
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        232,
+                                        239,
+                                        249
+                                    ],
+                                    "score": 1.0,
+                                    "content": "A.5 DISCUSSIONS ON SEED",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        }
+                    ],
+                    "index": 11
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        255,
+                        505,
+                        302
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                255,
+                                504,
+                                270
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        255,
+                                        491,
+                                        270
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Our proposed learning objective for SEED is composed of two goals, that is to align the encoding",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        492,
+                                        256,
+                                        504,
+                                        267
+                                    ],
+                                    "score": 0.83,
+                                    "content": "\\mathbf { z } ^ { S }",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                267,
+                                506,
+                                280
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        267,
+                                        214,
+                                        280
+                                    ],
+                                    "score": 1.0,
+                                    "content": "by the student model with",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        214,
+                                        267,
+                                        227,
+                                        278
+                                    ],
+                                    "score": 0.88,
+                                    "content": "\\mathbf { z } ^ { T }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        227,
+                                        267,
+                                        408,
+                                        280
+                                    ],
+                                    "score": 1.0,
+                                    "content": "produced by the teacher model; meanwhile,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        408,
+                                        267,
+                                        420,
+                                        278
+                                    ],
+                                    "score": 0.87,
+                                    "content": "\\mathbf { z } ^ { S }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        421,
+                                        267,
+                                        506,
+                                        280
+                                    ],
+                                    "score": 1.0,
+                                    "content": "also softly contrasts",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                279,
+                                505,
+                                291
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        279,
+                                        265,
+                                        291
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with random samples maintained in the",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        266,
+                                        279,
+                                        276,
+                                        289
+                                    ],
+                                    "score": 0.37,
+                                    "content": "\\mathbf { D }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        276,
+                                        279,
+                                        505,
+                                        291
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". This can be formulated more directly as minimizing the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                288,
+                                401,
+                                303
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        290,
+                                        115,
+                                        299
+                                    ],
+                                    "score": 0.72,
+                                    "content": "l 2",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        116,
+                                        288,
+                                        163,
+                                        303
+                                    ],
+                                    "score": 1.0,
+                                    "content": "distance of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        163,
+                                        289,
+                                        191,
+                                        300
+                                    ],
+                                    "score": 0.59,
+                                    "content": "{ \\mathbf z } ^ { T } , { \\mathbf z } ^ { S }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        191,
+                                        288,
+                                        387,
+                                        303
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", together with the cross-entropy computed using",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        387,
+                                        290,
+                                        397,
+                                        300
+                                    ],
+                                    "score": 0.52,
+                                    "content": "\\mathbf { D }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        398,
+                                        288,
+                                        401,
+                                        303
+                                    ],
+                                    "score": 1.0,
+                                    "content": ":",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        }
+                    ],
+                    "index": 13.5
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        114,
+                        308,
+                        483,
+                        379
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                114,
+                                308,
+                                483,
+                                379
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        114,
+                                        308,
+                                        483,
+                                        379
+                                    ],
+                                    "score": 0.94,
+                                    "content": "\\begin{array} { l } { \\displaystyle \\mathcal { L } = \\frac { 1 } { N } \\sum _ { i } ^ { N } \\bigg \\{ \\lambda _ { a } \\cdot \\bigg \\vert \\bigg \\vert \\mathbf { z } _ { i } ^ { T } - \\mathbf { z } _ { i } ^ { S } \\bigg \\vert \\bigg \\vert _ { 2 } - \\lambda _ { b } \\cdot \\mathbf { p } ^ { T } ( \\mathbf { x } _ { i } ; \\boldsymbol { \\theta } _ { T } , \\mathbf { D } ) \\cdot \\log \\mathbf { p } ^ { S } ( \\mathbf { x } _ { i } ; \\boldsymbol { \\theta } _ { S } , \\mathbf { D } ) \\bigg \\} } \\\\ { \\displaystyle \\quad = \\sum _ { i } ^ { N } \\bigg \\{ - \\lambda _ { a } \\cdot \\mathbf { z } _ { i } ^ { T } \\cdot \\mathbf { z } _ { i } ^ { S } - \\lambda _ { b } \\cdot \\sum _ { j } ^ { K } \\frac { \\exp ( \\mathbf { z } _ { i } ^ { T } \\cdot \\mathbf { d } _ { j } / \\tau ^ { T } ) } { \\sum _ { \\mathbf { d } \\setminus \\mathbf { D } } \\exp ( \\mathbf { z } _ { i } ^ { T } \\cdot \\mathbf { d } / \\tau ^ { T } ) } \\cdot \\log \\frac { \\exp ( \\mathbf { z } _ { i } ^ { S } \\cdot \\mathbf { d } _ { j } / \\tau ^ { S } ) } { \\sum _ { \\mathbf { d } \\setminus \\mathbf { D } } \\exp ( \\mathbf { z } _ { i } ^ { S } \\cdot \\mathbf { d } / \\tau ^ { S } ) } \\bigg \\} , } \\end{array}",
+                                    "type": "interline_equation",
+                                    "image_path": "5e4bf14a4554833a0ba76bc465a50d4fc5b25018566e53300c808cf0763a0ad4.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 17,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                114,
+                                308,
+                                483,
+                                331.6666666666667
+                            ],
+                            "spans": [],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                114,
+                                331.6666666666667,
+                                483,
+                                355.33333333333337
+                            ],
+                            "spans": [],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                114,
+                                355.33333333333337,
+                                483,
+                                379.00000000000006
+                            ],
+                            "spans": [],
+                            "index": 18
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        384,
+                        504,
+                        419
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                384,
+                                505,
+                                398
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        384,
+                                        505,
+                                        398
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Directly optimizing Eq. 7 can lead to apparent difficulty in searching optimal hyper-parameters",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                107,
+                                394,
+                                506,
+                                410
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        395,
+                                        151,
+                                        408
+                                    ],
+                                    "score": 0.89,
+                                    "content": "( \\lambda _ { a } , \\lambda _ { b } , \\bar { \\tau } ^ { T }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        152,
+                                        394,
+                                        169,
+                                        410
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        170,
+                                        396,
+                                        182,
+                                        406
+                                    ],
+                                    "score": 0.82,
+                                    "content": "\\tau ^ { S }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        182,
+                                        394,
+                                        299,
+                                        410
+                                    ],
+                                    "score": 1.0,
+                                    "content": "). Our proposed objective on",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        299,
+                                        396,
+                                        316,
+                                        406
+                                    ],
+                                    "score": 0.88,
+                                    "content": "\\mathbf { D } ^ { + }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        316,
+                                        394,
+                                        506,
+                                        410
+                                    ],
+                                    "score": 1.0,
+                                    "content": "indeed is an approximated upper-bound of the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                407,
+                                294,
+                                419
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        407,
+                                        294,
+                                        419
+                                    ],
+                                    "score": 1.0,
+                                    "content": "above objectiveness however much simplified:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        }
+                    ],
+                    "index": 20
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        147,
+                        425,
+                        464,
+                        511
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                147,
+                                425,
+                                464,
+                                511
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        147,
+                                        425,
+                                        464,
+                                        511
+                                    ],
+                                    "score": 0.94,
+                                    "content": "\\begin{array} { r l } & { \\mathcal { L } _ { \\mathrm { S E E D } } = \\displaystyle \\frac { 1 } { N } \\sum _ { i } ^ { N } - \\mathbf { p } ^ { T } ( \\mathbf { x } _ { i } ; \\boldsymbol { \\theta } _ { T } , \\mathbf { D } ^ { + } ) \\cdot \\log \\mathbf { p } ^ { S } ( \\mathbf { x } _ { i } ; \\boldsymbol { \\theta } _ { S } , \\mathbf { D } ^ { + } ) } \\\\ & { \\qquad = \\displaystyle \\sum _ { i } ^ { N } \\sum _ { j } ^ { K + 1 } - \\underbrace { \\frac { \\exp \\left( \\mathbf { z } _ { i } ^ { T } \\cdot \\mathbf { d } _ { j } / \\tau ^ { T } \\right) } { \\sum _ { \\mathbf { d } \\sim \\mathbf { D } ^ { + } } \\exp \\left( \\mathbf { z } _ { i } ^ { T } \\cdot \\mathbf { d } / \\tau ^ { T } \\right) } } _ { \\mathbf { w } _ { j } ^ { i } } \\cdot \\log \\frac { \\exp \\left( \\mathbf { z } _ { i } ^ { S } \\cdot \\mathbf { d } _ { j } / \\tau ^ { S } \\right) } { \\sum _ { \\mathbf { d } \\sim \\mathbf { D } ^ { + } } \\exp \\left( \\mathbf { z } _ { i } ^ { S } \\cdot \\mathbf { d } / \\tau ^ { S } \\right) } , } \\end{array}",
+                                    "type": "interline_equation",
+                                    "image_path": "23a09cdc1125e61dd98e8974b713166f88a62cc74247c152a34211fa3a3d3ea6.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 23,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                147,
+                                425,
+                                464,
+                                453.6666666666667
+                            ],
+                            "spans": [],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                147,
+                                453.6666666666667,
+                                464,
+                                482.33333333333337
+                            ],
+                            "spans": [],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                147,
+                                482.33333333333337,
+                                464,
+                                511.00000000000006
+                            ],
+                            "spans": [],
+                            "index": 24
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        518,
+                        505,
+                        544
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                517,
+                                505,
+                                532
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        517,
+                                        159,
+                                        532
+                                    ],
+                                    "score": 1.0,
+                                    "content": "where we let",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        159,
+                                        518,
+                                        173,
+                                        532
+                                    ],
+                                    "score": 0.9,
+                                    "content": "\\mathbf { w } _ { j } ^ { i }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        173,
+                                        517,
+                                        344,
+                                        532
+                                    ],
+                                    "score": 1.0,
+                                    "content": "denote the weighting term regulated under",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        345,
+                                        518,
+                                        358,
+                                        529
+                                    ],
+                                    "score": 0.88,
+                                    "content": "\\tau ^ { T }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        358,
+                                        517,
+                                        402,
+                                        532
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Since the",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        402,
+                                        519,
+                                        435,
+                                        530
+                                    ],
+                                    "score": 0.8,
+                                    "content": "( K + 1 )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        435,
+                                        517,
+                                        488,
+                                        532
+                                    ],
+                                    "score": 1.0,
+                                    "content": "th element in",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        489,
+                                        518,
+                                        505,
+                                        529
+                                    ],
+                                    "score": 0.86,
+                                    "content": "\\mathbf { D } ^ { + }",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                531,
+                                402,
+                                545
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        531,
+                                        216,
+                                        545
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is our supplemented vector",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        216,
+                                        531,
+                                        228,
+                                        544
+                                    ],
+                                    "score": 0.9,
+                                    "content": "{ \\mathbf z } _ { i } ^ { T }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        229,
+                                        531,
+                                        402,
+                                        545
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", the above objective can be expanded into:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        }
+                    ],
+                    "index": 25.5
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        162,
+                        551,
+                        447,
+                        623
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                162,
+                                551,
+                                447,
+                                623
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        162,
+                                        551,
+                                        447,
+                                        623
+                                    ],
+                                    "score": 0.95,
+                                    "content": "\\begin{array} { r l } & { \\mathcal { L } _ { \\mathrm { S E E D } } = \\displaystyle \\frac { 1 } { N } \\sum _ { i } ^ { N } \\Big \\{ { \\bf w } _ { K + 1 } ^ { i } \\cdot \\big ( - { \\bf z } _ { i } ^ { S } \\cdot { \\bf z } _ { i } ^ { T } / \\tau ^ { S } + \\log \\sum _ { \\bf d \\sim D ^ { + } } \\exp ( { \\bf z } _ { i } ^ { S } \\cdot { \\bf d } / \\tau ^ { S } ) \\big ) } \\\\ & { \\qquad + \\displaystyle \\sum _ { j = 1 } ^ { K } { \\bf w } _ { j } ^ { i } \\cdot \\big ( - { \\bf z } _ { i } ^ { S } \\cdot { \\bf d } _ { j } / \\tau ^ { S } + \\log \\sum _ { \\bf d \\sim D ^ { + } } \\exp ( { \\bf z } _ { i } ^ { S } \\cdot { \\bf d } / \\tau ^ { S } ) \\big ) \\Big \\} } \\end{array}",
+                                    "type": "interline_equation",
+                                    "image_path": "7d21dd51331e153adeabf8831905dcc223c4a73c9d250b37724e457be450eb78.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 28,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                162,
+                                551,
+                                447,
+                                575.0
+                            ],
+                            "spans": [],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                162,
+                                575.0,
+                                447,
+                                599.0
+                            ],
+                            "spans": [],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                162,
+                                599.0,
+                                447,
+                                623.0
+                            ],
+                            "spans": [],
+                            "index": 29
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        630,
+                        504,
+                        654
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                628,
+                                504,
+                                643
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        628,
+                                        491,
+                                        643
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Note that the LSE term in the first line is strictly non-negative as the range of inner product for",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        492,
+                                        629,
+                                        504,
+                                        640
+                                    ],
+                                    "score": 0.85,
+                                    "content": "\\mathbf { z } ^ { S }",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                640,
+                                222,
+                                654
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        640,
+                                        123,
+                                        654
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        124,
+                                        642,
+                                        131,
+                                        651
+                                    ],
+                                    "score": 0.56,
+                                    "content": "\\mathbf { d }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        132,
+                                        640,
+                                        185,
+                                        654
+                                    ],
+                                    "score": 1.0,
+                                    "content": "lies between",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        185,
+                                        640,
+                                        217,
+                                        654
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\left[ - 1 , + 1 \\right]",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        218,
+                                        640,
+                                        222,
+                                        654
+                                    ],
+                                    "score": 1.0,
+                                    "content": ":",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        }
+                    ],
+                    "index": 30.5
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        168,
+                        660,
+                        441,
+                        677
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                168,
+                                660,
+                                441,
+                                677
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        168,
+                                        660,
+                                        441,
+                                        677
+                                    ],
+                                    "score": 0.9,
+                                    "content": "\\mathrm { L S E } ( \\mathbf { D } ^ { + } , \\mathbf { z } _ { i } ^ { S } ) \\geq \\log \\big ( M \\cdot \\exp ( - 1 / \\tau ^ { S } ) \\big ) = \\log \\big ( M \\cdot \\exp ( - 5 ) \\big ) > 0 ,",
+                                    "type": "interline_equation",
+                                    "image_path": "60167030ccb1d22c58722f0344ef83e074921c76ef9e5b47b5b5bc7aa017a94c.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 32,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                168,
+                                660,
+                                441,
+                                677
+                            ],
+                            "spans": [],
+                            "index": 32
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        684,
+                        505,
+                        717
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                682,
+                                505,
+                                696
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        682,
+                                        133,
+                                        696
+                                    ],
+                                    "score": 1.0,
+                                    "content": "where",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        133,
+                                        685,
+                                        145,
+                                        694
+                                    ],
+                                    "score": 0.81,
+                                    "content": "M",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        145,
+                                        682,
+                                        335,
+                                        696
+                                    ],
+                                    "score": 1.0,
+                                    "content": "denotes the cardinality of the maintained queue",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        335,
+                                        684,
+                                        352,
+                                        694
+                                    ],
+                                    "score": 0.89,
+                                    "content": "\\mathbf { D } ^ { + }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        352,
+                                        682,
+                                        505,
+                                        696
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and is set to 65,536 in our experiment",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                693,
+                                506,
+                                709
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        693,
+                                        127,
+                                        709
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        128,
+                                        695,
+                                        171,
+                                        706
+                                    ],
+                                    "score": 0.9,
+                                    "content": "\\tau ^ { S } = 0 . 2",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        171,
+                                        693,
+                                        506,
+                                        709
+                                    ],
+                                    "score": 1.0,
+                                    "content": "constantly. Meanwhile, the LSE term in the second line satisfies the following",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                704,
+                                153,
+                                721
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        704,
+                                        153,
+                                        721
+                                    ],
+                                    "score": 1.0,
+                                    "content": "inequality:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        }
+                    ],
+                    "index": 34
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        242,
+                        718,
+                        368,
+                        733
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                242,
+                                718,
+                                368,
+                                733
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        242,
+                                        718,
+                                        368,
+                                        733
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\mathrm { L S E } ( \\mathbf { D } ^ { + } , \\mathbf { z } _ { i } ^ { S } ) \\geq \\mathrm { L S E } ( \\mathbf { D } , \\mathbf { z } _ { i } ^ { S } ) .",
+                                    "type": "interline_equation",
+                                    "image_path": "a9817ebf8558bdc08eb3f5d034f4302f4d5743c284feec760b5983580c09cbb1.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 36,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                242,
+                                718,
+                                368,
+                                733
+                            ],
+                            "spans": [],
+                            "index": 36
+                        }
+                    ]
+                }
+            ],
+            "page_idx": 17,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        107,
+                        26,
+                        292,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                25,
+                                293,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        25,
+                                        293,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Published as a conference paper at ICLR 2021",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        300,
+                        751,
+                        311,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                299,
+                                750,
+                                312,
+                                763
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        299,
+                                        750,
+                                        312,
+                                        763
+                                    ],
+                                    "score": 1.0,
+                                    "content": "18",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        82,
+                        504,
+                        116
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                81,
+                                506,
+                                95
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        83,
+                                        114,
+                                        92
+                                    ],
+                                    "score": 0.67,
+                                    "content": "K",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        115,
+                                        81,
+                                        506,
+                                        95
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-Means clustering within a batch and continuously update the centroid based on the teacher feature",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                94,
+                                505,
+                                105
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        94,
+                                        505,
+                                        105
+                                    ],
+                                    "score": 1.0,
+                                    "content": "representation as distillation goes on. This alleviates the above problems and yields a substantial",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                105,
+                                314,
+                                116
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        105,
+                                        282,
+                                        116
+                                    ],
+                                    "score": 1.0,
+                                    "content": "performance improvement on ResNet-18 to",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        282,
+                                        105,
+                                        309,
+                                        115
+                                    ],
+                                    "score": 0.86,
+                                    "content": "5 6 . 4 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        309,
+                                        105,
+                                        314,
+                                        116
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        }
+                    ],
+                    "index": 1,
+                    "bbox_fs": [
+                        105,
+                        81,
+                        506,
+                        116
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        120,
+                        505,
+                        166
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                121,
+                                505,
+                                133
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        121,
+                                        505,
+                                        133
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Binary Contrastive Loss: We resort to CRD (Tian et al., 2019a) and adopt an info-NCE loss-alike",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                132,
+                                505,
+                                145
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        132,
+                                        505,
+                                        145
+                                    ],
+                                    "score": 1.0,
+                                    "content": "training objective in unsupervised distillation tasks. Specifically, we treat representation features",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                142,
+                                505,
+                                156
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        142,
+                                        258,
+                                        156
+                                    ],
+                                    "score": 1.0,
+                                    "content": "from Teacher and Student for instance",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        259,
+                                        145,
+                                        270,
+                                        154
+                                    ],
+                                    "score": 0.87,
+                                    "content": "\\mathbf { x } _ { i }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        270,
+                                        142,
+                                        448,
+                                        156
+                                    ],
+                                    "score": 1.0,
+                                    "content": "as positive pairs, and random instances from",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        448,
+                                        144,
+                                        458,
+                                        154
+                                    ],
+                                    "score": 0.64,
+                                    "content": "\\mathbf { D }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        459,
+                                        142,
+                                        505,
+                                        156
+                                    ],
+                                    "score": 1.0,
+                                    "content": "as negative",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                154,
+                                146,
+                                168
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        154,
+                                        146,
+                                        168
+                                    ],
+                                    "score": 1.0,
+                                    "content": "samples:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        }
+                    ],
+                    "index": 4.5,
+                    "bbox_fs": [
+                        105,
+                        121,
+                        505,
+                        168
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        172,
+                        166,
+                        427,
+                        200
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                172,
+                                166,
+                                427,
+                                200
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        172,
+                                        166,
+                                        427,
+                                        200
+                                    ],
+                                    "score": 0.94,
+                                    "content": "\\hat { \\theta } _ { S } = \\underset { \\theta _ { S } } { \\arg \\operatorname* { m i n } } \\sum _ { i } ^ { N } \\log h ( \\mathbf { z } _ { i } ^ { S } , \\mathbf { z } _ { i } ^ { T } ) + K \\cdot [ \\log h ( 1 - h ( \\mathbf { z } _ { i } ^ { S } , \\mathbf { d } _ { j } ^ { T } ) ) ] ,",
+                                    "type": "interline_equation",
+                                    "image_path": "7efa8646a999cf9b57dbdc3eab933940b7c76683e63c8311e334af00b5e6c4a4.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 8,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                172,
+                                166,
+                                427,
+                                177.33333333333334
+                            ],
+                            "spans": [],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                172,
+                                177.33333333333334,
+                                427,
+                                188.66666666666669
+                            ],
+                            "spans": [],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                172,
+                                188.66666666666669,
+                                427,
+                                200.00000000000003
+                            ],
+                            "spans": [],
+                            "index": 9
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        206,
+                        505,
+                        219
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                204,
+                                506,
+                                221
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        204,
+                                        133,
+                                        221
+                                    ],
+                                    "score": 1.0,
+                                    "content": "where",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        133,
+                                        205,
+                                        189,
+                                        220
+                                    ],
+                                    "score": 0.89,
+                                    "content": "\\mathbf { d } _ { j } ^ { T } \\in \\mathbf { D } , h ( \\cdot )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        189,
+                                        204,
+                                        339,
+                                        221
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is any family of functions that satisfy",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        339,
+                                        207,
+                                        346,
+                                        217
+                                    ],
+                                    "score": 0.49,
+                                    "content": "h",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        347,
+                                        204,
+                                        351,
+                                        221
+                                    ],
+                                    "score": 1.0,
+                                    "content": ":",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        351,
+                                        207,
+                                        414,
+                                        218
+                                    ],
+                                    "score": 0.88,
+                                    "content": "\\{ { \\bf z } , { \\bf d } \\}  [ 0 , 1 ]",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        415,
+                                        204,
+                                        506,
+                                        221
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", e.g., cosine similarity.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        }
+                    ],
+                    "index": 10,
+                    "bbox_fs": [
+                        105,
+                        204,
+                        506,
+                        221
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        234,
+                        238,
+                        247
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                232,
+                                239,
+                                249
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        232,
+                                        239,
+                                        249
+                                    ],
+                                    "score": 1.0,
+                                    "content": "A.5 DISCUSSIONS ON SEED",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        }
+                    ],
+                    "index": 11
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        255,
+                        505,
+                        302
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                255,
+                                504,
+                                270
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        255,
+                                        491,
+                                        270
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Our proposed learning objective for SEED is composed of two goals, that is to align the encoding",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        492,
+                                        256,
+                                        504,
+                                        267
+                                    ],
+                                    "score": 0.83,
+                                    "content": "\\mathbf { z } ^ { S }",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                267,
+                                506,
+                                280
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        267,
+                                        214,
+                                        280
+                                    ],
+                                    "score": 1.0,
+                                    "content": "by the student model with",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        214,
+                                        267,
+                                        227,
+                                        278
+                                    ],
+                                    "score": 0.88,
+                                    "content": "\\mathbf { z } ^ { T }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        227,
+                                        267,
+                                        408,
+                                        280
+                                    ],
+                                    "score": 1.0,
+                                    "content": "produced by the teacher model; meanwhile,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        408,
+                                        267,
+                                        420,
+                                        278
+                                    ],
+                                    "score": 0.87,
+                                    "content": "\\mathbf { z } ^ { S }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        421,
+                                        267,
+                                        506,
+                                        280
+                                    ],
+                                    "score": 1.0,
+                                    "content": "also softly contrasts",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                279,
+                                505,
+                                291
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        279,
+                                        265,
+                                        291
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with random samples maintained in the",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        266,
+                                        279,
+                                        276,
+                                        289
+                                    ],
+                                    "score": 0.37,
+                                    "content": "\\mathbf { D }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        276,
+                                        279,
+                                        505,
+                                        291
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". This can be formulated more directly as minimizing the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                288,
+                                401,
+                                303
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        290,
+                                        115,
+                                        299
+                                    ],
+                                    "score": 0.72,
+                                    "content": "l 2",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        116,
+                                        288,
+                                        163,
+                                        303
+                                    ],
+                                    "score": 1.0,
+                                    "content": "distance of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        163,
+                                        289,
+                                        191,
+                                        300
+                                    ],
+                                    "score": 0.59,
+                                    "content": "{ \\mathbf z } ^ { T } , { \\mathbf z } ^ { S }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        191,
+                                        288,
+                                        387,
+                                        303
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", together with the cross-entropy computed using",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        387,
+                                        290,
+                                        397,
+                                        300
+                                    ],
+                                    "score": 0.52,
+                                    "content": "\\mathbf { D }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        398,
+                                        288,
+                                        401,
+                                        303
+                                    ],
+                                    "score": 1.0,
+                                    "content": ":",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        }
+                    ],
+                    "index": 13.5,
+                    "bbox_fs": [
+                        105,
+                        255,
+                        506,
+                        303
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        114,
+                        308,
+                        483,
+                        379
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                114,
+                                308,
+                                483,
+                                379
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        114,
+                                        308,
+                                        483,
+                                        379
+                                    ],
+                                    "score": 0.94,
+                                    "content": "\\begin{array} { l } { \\displaystyle \\mathcal { L } = \\frac { 1 } { N } \\sum _ { i } ^ { N } \\bigg \\{ \\lambda _ { a } \\cdot \\bigg \\vert \\bigg \\vert \\mathbf { z } _ { i } ^ { T } - \\mathbf { z } _ { i } ^ { S } \\bigg \\vert \\bigg \\vert _ { 2 } - \\lambda _ { b } \\cdot \\mathbf { p } ^ { T } ( \\mathbf { x } _ { i } ; \\boldsymbol { \\theta } _ { T } , \\mathbf { D } ) \\cdot \\log \\mathbf { p } ^ { S } ( \\mathbf { x } _ { i } ; \\boldsymbol { \\theta } _ { S } , \\mathbf { D } ) \\bigg \\} } \\\\ { \\displaystyle \\quad = \\sum _ { i } ^ { N } \\bigg \\{ - \\lambda _ { a } \\cdot \\mathbf { z } _ { i } ^ { T } \\cdot \\mathbf { z } _ { i } ^ { S } - \\lambda _ { b } \\cdot \\sum _ { j } ^ { K } \\frac { \\exp ( \\mathbf { z } _ { i } ^ { T } \\cdot \\mathbf { d } _ { j } / \\tau ^ { T } ) } { \\sum _ { \\mathbf { d } \\setminus \\mathbf { D } } \\exp ( \\mathbf { z } _ { i } ^ { T } \\cdot \\mathbf { d } / \\tau ^ { T } ) } \\cdot \\log \\frac { \\exp ( \\mathbf { z } _ { i } ^ { S } \\cdot \\mathbf { d } _ { j } / \\tau ^ { S } ) } { \\sum _ { \\mathbf { d } \\setminus \\mathbf { D } } \\exp ( \\mathbf { z } _ { i } ^ { S } \\cdot \\mathbf { d } / \\tau ^ { S } ) } \\bigg \\} , } \\end{array}",
+                                    "type": "interline_equation",
+                                    "image_path": "5e4bf14a4554833a0ba76bc465a50d4fc5b25018566e53300c808cf0763a0ad4.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 17,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                114,
+                                308,
+                                483,
+                                331.6666666666667
+                            ],
+                            "spans": [],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                114,
+                                331.6666666666667,
+                                483,
+                                355.33333333333337
+                            ],
+                            "spans": [],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                114,
+                                355.33333333333337,
+                                483,
+                                379.00000000000006
+                            ],
+                            "spans": [],
+                            "index": 18
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        384,
+                        504,
+                        419
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                384,
+                                505,
+                                398
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        384,
+                                        505,
+                                        398
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Directly optimizing Eq. 7 can lead to apparent difficulty in searching optimal hyper-parameters",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                107,
+                                394,
+                                506,
+                                410
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        395,
+                                        151,
+                                        408
+                                    ],
+                                    "score": 0.89,
+                                    "content": "( \\lambda _ { a } , \\lambda _ { b } , \\bar { \\tau } ^ { T }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        152,
+                                        394,
+                                        169,
+                                        410
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        170,
+                                        396,
+                                        182,
+                                        406
+                                    ],
+                                    "score": 0.82,
+                                    "content": "\\tau ^ { S }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        182,
+                                        394,
+                                        299,
+                                        410
+                                    ],
+                                    "score": 1.0,
+                                    "content": "). Our proposed objective on",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        299,
+                                        396,
+                                        316,
+                                        406
+                                    ],
+                                    "score": 0.88,
+                                    "content": "\\mathbf { D } ^ { + }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        316,
+                                        394,
+                                        506,
+                                        410
+                                    ],
+                                    "score": 1.0,
+                                    "content": "indeed is an approximated upper-bound of the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                407,
+                                294,
+                                419
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        407,
+                                        294,
+                                        419
+                                    ],
+                                    "score": 1.0,
+                                    "content": "above objectiveness however much simplified:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        }
+                    ],
+                    "index": 20,
+                    "bbox_fs": [
+                        105,
+                        384,
+                        506,
+                        419
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        147,
+                        425,
+                        464,
+                        511
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                147,
+                                425,
+                                464,
+                                511
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        147,
+                                        425,
+                                        464,
+                                        511
+                                    ],
+                                    "score": 0.94,
+                                    "content": "\\begin{array} { r l } & { \\mathcal { L } _ { \\mathrm { S E E D } } = \\displaystyle \\frac { 1 } { N } \\sum _ { i } ^ { N } - \\mathbf { p } ^ { T } ( \\mathbf { x } _ { i } ; \\boldsymbol { \\theta } _ { T } , \\mathbf { D } ^ { + } ) \\cdot \\log \\mathbf { p } ^ { S } ( \\mathbf { x } _ { i } ; \\boldsymbol { \\theta } _ { S } , \\mathbf { D } ^ { + } ) } \\\\ & { \\qquad = \\displaystyle \\sum _ { i } ^ { N } \\sum _ { j } ^ { K + 1 } - \\underbrace { \\frac { \\exp \\left( \\mathbf { z } _ { i } ^ { T } \\cdot \\mathbf { d } _ { j } / \\tau ^ { T } \\right) } { \\sum _ { \\mathbf { d } \\sim \\mathbf { D } ^ { + } } \\exp \\left( \\mathbf { z } _ { i } ^ { T } \\cdot \\mathbf { d } / \\tau ^ { T } \\right) } } _ { \\mathbf { w } _ { j } ^ { i } } \\cdot \\log \\frac { \\exp \\left( \\mathbf { z } _ { i } ^ { S } \\cdot \\mathbf { d } _ { j } / \\tau ^ { S } \\right) } { \\sum _ { \\mathbf { d } \\sim \\mathbf { D } ^ { + } } \\exp \\left( \\mathbf { z } _ { i } ^ { S } \\cdot \\mathbf { d } / \\tau ^ { S } \\right) } , } \\end{array}",
+                                    "type": "interline_equation",
+                                    "image_path": "23a09cdc1125e61dd98e8974b713166f88a62cc74247c152a34211fa3a3d3ea6.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 23,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                147,
+                                425,
+                                464,
+                                453.6666666666667
+                            ],
+                            "spans": [],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                147,
+                                453.6666666666667,
+                                464,
+                                482.33333333333337
+                            ],
+                            "spans": [],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                147,
+                                482.33333333333337,
+                                464,
+                                511.00000000000006
+                            ],
+                            "spans": [],
+                            "index": 24
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        518,
+                        505,
+                        544
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                517,
+                                505,
+                                532
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        517,
+                                        159,
+                                        532
+                                    ],
+                                    "score": 1.0,
+                                    "content": "where we let",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        159,
+                                        518,
+                                        173,
+                                        532
+                                    ],
+                                    "score": 0.9,
+                                    "content": "\\mathbf { w } _ { j } ^ { i }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        173,
+                                        517,
+                                        344,
+                                        532
+                                    ],
+                                    "score": 1.0,
+                                    "content": "denote the weighting term regulated under",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        345,
+                                        518,
+                                        358,
+                                        529
+                                    ],
+                                    "score": 0.88,
+                                    "content": "\\tau ^ { T }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        358,
+                                        517,
+                                        402,
+                                        532
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Since the",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        402,
+                                        519,
+                                        435,
+                                        530
+                                    ],
+                                    "score": 0.8,
+                                    "content": "( K + 1 )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        435,
+                                        517,
+                                        488,
+                                        532
+                                    ],
+                                    "score": 1.0,
+                                    "content": "th element in",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        489,
+                                        518,
+                                        505,
+                                        529
+                                    ],
+                                    "score": 0.86,
+                                    "content": "\\mathbf { D } ^ { + }",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                531,
+                                402,
+                                545
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        531,
+                                        216,
+                                        545
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is our supplemented vector",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        216,
+                                        531,
+                                        228,
+                                        544
+                                    ],
+                                    "score": 0.9,
+                                    "content": "{ \\mathbf z } _ { i } ^ { T }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        229,
+                                        531,
+                                        402,
+                                        545
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", the above objective can be expanded into:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        }
+                    ],
+                    "index": 25.5,
+                    "bbox_fs": [
+                        105,
+                        517,
+                        505,
+                        545
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        162,
+                        551,
+                        447,
+                        623
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                162,
+                                551,
+                                447,
+                                623
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        162,
+                                        551,
+                                        447,
+                                        623
+                                    ],
+                                    "score": 0.95,
+                                    "content": "\\begin{array} { r l } & { \\mathcal { L } _ { \\mathrm { S E E D } } = \\displaystyle \\frac { 1 } { N } \\sum _ { i } ^ { N } \\Big \\{ { \\bf w } _ { K + 1 } ^ { i } \\cdot \\big ( - { \\bf z } _ { i } ^ { S } \\cdot { \\bf z } _ { i } ^ { T } / \\tau ^ { S } + \\log \\sum _ { \\bf d \\sim D ^ { + } } \\exp ( { \\bf z } _ { i } ^ { S } \\cdot { \\bf d } / \\tau ^ { S } ) \\big ) } \\\\ & { \\qquad + \\displaystyle \\sum _ { j = 1 } ^ { K } { \\bf w } _ { j } ^ { i } \\cdot \\big ( - { \\bf z } _ { i } ^ { S } \\cdot { \\bf d } _ { j } / \\tau ^ { S } + \\log \\sum _ { \\bf d \\sim D ^ { + } } \\exp ( { \\bf z } _ { i } ^ { S } \\cdot { \\bf d } / \\tau ^ { S } ) \\big ) \\Big \\} } \\end{array}",
+                                    "type": "interline_equation",
+                                    "image_path": "7d21dd51331e153adeabf8831905dcc223c4a73c9d250b37724e457be450eb78.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 28,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                162,
+                                551,
+                                447,
+                                575.0
+                            ],
+                            "spans": [],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                162,
+                                575.0,
+                                447,
+                                599.0
+                            ],
+                            "spans": [],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                162,
+                                599.0,
+                                447,
+                                623.0
+                            ],
+                            "spans": [],
+                            "index": 29
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        630,
+                        504,
+                        654
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                628,
+                                504,
+                                643
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        628,
+                                        491,
+                                        643
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Note that the LSE term in the first line is strictly non-negative as the range of inner product for",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        492,
+                                        629,
+                                        504,
+                                        640
+                                    ],
+                                    "score": 0.85,
+                                    "content": "\\mathbf { z } ^ { S }",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                640,
+                                222,
+                                654
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        640,
+                                        123,
+                                        654
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        124,
+                                        642,
+                                        131,
+                                        651
+                                    ],
+                                    "score": 0.56,
+                                    "content": "\\mathbf { d }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        132,
+                                        640,
+                                        185,
+                                        654
+                                    ],
+                                    "score": 1.0,
+                                    "content": "lies between",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        185,
+                                        640,
+                                        217,
+                                        654
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\left[ - 1 , + 1 \\right]",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        218,
+                                        640,
+                                        222,
+                                        654
+                                    ],
+                                    "score": 1.0,
+                                    "content": ":",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        }
+                    ],
+                    "index": 30.5,
+                    "bbox_fs": [
+                        105,
+                        628,
+                        504,
+                        654
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        168,
+                        660,
+                        441,
+                        677
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                168,
+                                660,
+                                441,
+                                677
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        168,
+                                        660,
+                                        441,
+                                        677
+                                    ],
+                                    "score": 0.9,
+                                    "content": "\\mathrm { L S E } ( \\mathbf { D } ^ { + } , \\mathbf { z } _ { i } ^ { S } ) \\geq \\log \\big ( M \\cdot \\exp ( - 1 / \\tau ^ { S } ) \\big ) = \\log \\big ( M \\cdot \\exp ( - 5 ) \\big ) > 0 ,",
+                                    "type": "interline_equation",
+                                    "image_path": "60167030ccb1d22c58722f0344ef83e074921c76ef9e5b47b5b5bc7aa017a94c.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 32,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                168,
+                                660,
+                                441,
+                                677
+                            ],
+                            "spans": [],
+                            "index": 32
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        684,
+                        505,
+                        717
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                682,
+                                505,
+                                696
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        682,
+                                        133,
+                                        696
+                                    ],
+                                    "score": 1.0,
+                                    "content": "where",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        133,
+                                        685,
+                                        145,
+                                        694
+                                    ],
+                                    "score": 0.81,
+                                    "content": "M",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        145,
+                                        682,
+                                        335,
+                                        696
+                                    ],
+                                    "score": 1.0,
+                                    "content": "denotes the cardinality of the maintained queue",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        335,
+                                        684,
+                                        352,
+                                        694
+                                    ],
+                                    "score": 0.89,
+                                    "content": "\\mathbf { D } ^ { + }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        352,
+                                        682,
+                                        505,
+                                        696
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and is set to 65,536 in our experiment",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                693,
+                                506,
+                                709
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        693,
+                                        127,
+                                        709
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        128,
+                                        695,
+                                        171,
+                                        706
+                                    ],
+                                    "score": 0.9,
+                                    "content": "\\tau ^ { S } = 0 . 2",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        171,
+                                        693,
+                                        506,
+                                        709
+                                    ],
+                                    "score": 1.0,
+                                    "content": "constantly. Meanwhile, the LSE term in the second line satisfies the following",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                704,
+                                153,
+                                721
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        704,
+                                        153,
+                                        721
+                                    ],
+                                    "score": 1.0,
+                                    "content": "inequality:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        }
+                    ],
+                    "index": 34,
+                    "bbox_fs": [
+                        104,
+                        682,
+                        506,
+                        721
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        242,
+                        718,
+                        368,
+                        733
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                242,
+                                718,
+                                368,
+                                733
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        242,
+                                        718,
+                                        368,
+                                        733
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\mathrm { L S E } ( \\mathbf { D } ^ { + } , \\mathbf { z } _ { i } ^ { S } ) \\geq \\mathrm { L S E } ( \\mathbf { D } , \\mathbf { z } _ { i } ^ { S } ) .",
+                                    "type": "interline_equation",
+                                    "image_path": "a9817ebf8558bdc08eb3f5d034f4302f4d5743c284feec760b5983580c09cbb1.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 36,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                242,
+                                718,
+                                368,
+                                733
+                            ],
+                            "spans": [],
+                            "index": 36
+                        }
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        81,
+                        505,
+                        106
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                81,
+                                505,
+                                95
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        81,
+                                        505,
+                                        95
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Thus, this demonstrates that the objective for SEED as Eq. 8 is equivalent to minimizing a weakened",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                93,
+                                200,
+                                106
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        93,
+                                        200,
+                                        106
+                                    ],
+                                    "score": 1.0,
+                                    "content": "upper-bound of e.q. 7:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        }
+                    ],
+                    "index": 0.5
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        111,
+                        109,
+                        505,
+                        216
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                111,
+                                109,
+                                505,
+                                216
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        111,
+                                        109,
+                                        505,
+                                        216
+                                    ],
+                                    "score": 0.94,
+                                    "content": "\\begin{array} { l } { { \\mathcal { L } _ { \\mathrm { S E E D } } = \\displaystyle \\frac { 1 } { N } \\sum _ { i } ^ { N } - { \\bf p } ^ { T } ( { \\bf x } _ { i } ; \\boldsymbol { \\theta } _ { T } , { \\bf D } ^ { + } ) \\cdot \\log { \\bf p } ^ { S } ( { \\bf x } _ { i } ; \\boldsymbol { \\theta } _ { S } , { \\bf D } ^ { + } ) } \\ ~ } \\\\ { { \\displaystyle \\geq \\frac { 1 } { N } \\sum _ { i } ^ { N } \\left\\{ { \\bf w } _ { K + 1 } ^ { i } \\cdot ( - { \\bf z } _ { i } ^ { S } \\cdot { \\bf z } _ { i } ^ { T } / \\tau ^ { S } ) + \\sum _ { j = 1 } ^ { K } { \\bf w } _ { j } ^ { i } \\cdot \\left( - { \\bf z } _ { i } ^ { S } \\cdot { \\bf d } _ { j } / \\tau ^ { S } + \\log \\sum _ { \\bf d \\times D } ^ { C } \\exp ( { \\bf z } _ { i } ^ { S } \\cdot { \\bf d } / \\tau ^ { S } ) \\right) \\right\\} } \\ ~ } \\\\ { { \\displaystyle = \\frac { 1 } { N } \\sum _ { i } ^ { N } \\Bigg \\{ - \\frac { { \\bf w } _ { K + 1 } ^ { i } } { \\tau ^ { S } } \\cdot { \\bf z } _ { i } ^ { S } \\cdot { \\bf z } _ { i } ^ { T } - { \\bf p } ^ { T } ( { \\bf x } _ { i } ; \\boldsymbol { \\theta } _ { T } , { \\bf D } ) \\cdot \\log { \\bf p } ^ { S } ( { \\bf x } _ { i } ; \\boldsymbol { \\theta } _ { S } , { \\bf D } ) \\Bigg \\} } \\ ~ } \\end{array}",
+                                    "type": "interline_equation",
+                                    "image_path": "d86539a726b5a86171c5a58180e145908afef1dd30004cb1df11f07069fb1ca7.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 3,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                111,
+                                109,
+                                505,
+                                144.66666666666666
+                            ],
+                            "spans": [],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                111,
+                                144.66666666666666,
+                                505,
+                                180.33333333333331
+                            ],
+                            "spans": [],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                111,
+                                180.33333333333331,
+                                505,
+                                215.99999999999997
+                            ],
+                            "spans": [],
+                            "index": 4
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        223,
+                        505,
+                        268
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                223,
+                                506,
+                                237
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        223,
+                                        189,
+                                        237
+                                    ],
+                                    "score": 1.0,
+                                    "content": "This proves that our",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        189,
+                                        225,
+                                        216,
+                                        236
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\mathcal { L } _ { \\mathrm { S E E D } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        216,
+                                        223,
+                                        486,
+                                        237
+                                    ],
+                                    "score": 1.0,
+                                    "content": "directly relates to a more intuitive distillation formulation as Eq. 7",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        487,
+                                        225,
+                                        506,
+                                        235
+                                    ],
+                                    "score": 0.69,
+                                    "content": "( l 2 +",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                235,
+                                506,
+                                247
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        235,
+                                        506,
+                                        247
+                                    ],
+                                    "score": 1.0,
+                                    "content": "cross entropy loss), and it implicitly contains the objective of aligning and contrasting. However, our",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                245,
+                                506,
+                                259
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        245,
+                                        409,
+                                        259
+                                    ],
+                                    "score": 1.0,
+                                    "content": "training objective is much simplified. During practice, we find by regulating",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        409,
+                                        245,
+                                        422,
+                                        256
+                                    ],
+                                    "score": 0.88,
+                                    "content": "\\tau ^ { T }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        422,
+                                        245,
+                                        506,
+                                        259
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", both training losses",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                257,
+                                196,
+                                269
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        257,
+                                        196,
+                                        269
+                                    ],
+                                    "score": 1.0,
+                                    "content": "produce equal results.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        }
+                    ],
+                    "index": 6.5
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        280,
+                        406,
+                        293
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                280,
+                                408,
+                                294
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        280,
+                                        408,
+                                        294
+                                    ],
+                                    "score": 1.0,
+                                    "content": "A.6 DISCUSSION ON THE RELATIONSHIP OF SEED WITH INFO-NCE",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        }
+                    ],
+                    "index": 9
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        301,
+                        506,
+                        358
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                302,
+                                505,
+                                314
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        302,
+                                        505,
+                                        314
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The objective of distillation can be considered as a soft version of Info-NCE (Oord et al., 2018), with",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                313,
+                                505,
+                                325
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        313,
+                                        505,
+                                        325
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the only difference to be that SEED learns from the negative samples with probabilities instead of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                324,
+                                507,
+                                337
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        324,
+                                        507,
+                                        337
+                                    ],
+                                    "score": 1.0,
+                                    "content": "treating them all strictly as negative samples. To be more specific, following Info-NCE, the “hard”",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                335,
+                                506,
+                                348
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        335,
+                                        506,
+                                        348
+                                    ],
+                                    "score": 1.0,
+                                    "content": "style contrastive distillation can be expressed as aligning with representations from the Teacher",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                346,
+                                312,
+                                359
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        346,
+                                        312,
+                                        359
+                                    ],
+                                    "score": 1.0,
+                                    "content": "encoder and contrasting with all random instances:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        }
+                    ],
+                    "index": 12
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        167,
+                        362,
+                        433,
+                        396
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                167,
+                                362,
+                                433,
+                                396
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        167,
+                                        362,
+                                        433,
+                                        396
+                                    ],
+                                    "score": 0.94,
+                                    "content": "\\hat { \\theta } _ { S } = \\underset { \\theta _ { S } } { \\arg \\operatorname* { m i n } } \\ : \\mathcal { L } _ { N C E } = \\underset { \\theta _ { S } } { \\arg \\operatorname* { m i n } } \\sum _ { i } ^ { N } - \\log \\frac { \\exp ( \\mathbf { z } _ { i } ^ { T } \\cdot \\mathbf { z } _ { i } ^ { S } / \\tau ) } { \\sum _ { \\mathbf { d } \\sim \\mathbf { D } } \\exp ( \\mathbf { z } _ { i } ^ { S } \\cdot \\mathbf { d } / \\tau ) }",
+                                    "type": "interline_equation",
+                                    "image_path": "e34ea33920cc7ebc30f1c6559aaa3da4456286629890e13238bec91b4ea5dd57.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 16,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                167,
+                                362,
+                                433,
+                                373.3333333333333
+                            ],
+                            "spans": [],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                167,
+                                373.3333333333333,
+                                433,
+                                384.66666666666663
+                            ],
+                            "spans": [],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                167,
+                                384.66666666666663,
+                                433,
+                                395.99999999999994
+                            ],
+                            "spans": [],
+                            "index": 17
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        400,
+                        504,
+                        423
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                400,
+                                505,
+                                413
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        400,
+                                        505,
+                                        413
+                                    ],
+                                    "score": 1.0,
+                                    "content": "which can be further deduced with two sub-terms consisting of positive sample alignment and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                412,
+                                251,
+                                425
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        412,
+                                        251,
+                                        425
+                                    ],
+                                    "score": 1.0,
+                                    "content": "contrasting with negative instances:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        }
+                    ],
+                    "index": 18.5
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        190,
+                        427,
+                        420,
+                        478
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                190,
+                                427,
+                                420,
+                                478
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        190,
+                                        427,
+                                        420,
+                                        478
+                                    ],
+                                    "score": 0.94,
+                                    "content": "\\mathcal { L } _ { N C E } = \\sum _ { i } ^ { N } \\Big \\{ \\underbrace { - \\mathbf { z } _ { i } ^ { S } \\cdot \\mathbf { z } _ { i } ^ { T } / \\tau } _ { a l i g n m e n t } + \\underbrace { \\log \\sum _ { \\mathbf { d } \\sim \\mathbf { D } } \\exp ( \\mathbf { z } _ { i } ^ { S } \\cdot \\mathbf { d } / \\tau ) } _ { c o n t r a s t i n g } \\Big \\} .",
+                                    "type": "interline_equation",
+                                    "image_path": "065e9718ac69ca6ad65c1327e04b7725124e89d075d2dedbd19848d52002ec32.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 21,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                190,
+                                427,
+                                420,
+                                444.0
+                            ],
+                            "spans": [],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                190,
+                                444.0,
+                                420,
+                                461.0
+                            ],
+                            "spans": [],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                190,
+                                461.0,
+                                420,
+                                478.0
+                            ],
+                            "spans": [],
+                            "index": 22
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        483,
+                        505,
+                        506
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                482,
+                                506,
+                                496
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        482,
+                                        506,
+                                        496
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Similarly, the objective of SEED can be dissembled into the weighted form of alignment and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                496,
+                                180,
+                                507
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        496,
+                                        180,
+                                        507
+                                    ],
+                                    "score": 1.0,
+                                    "content": "contrasting terms:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        }
+                    ],
+                    "index": 23.5
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        111,
+                        510,
+                        503,
+                        599
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                111,
+                                510,
+                                503,
+                                599
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        111,
+                                        510,
+                                        503,
+                                        599
+                                    ],
+                                    "score": 0.94,
+                                    "content": "\\begin{array} { r l } & { \\mathcal { L } _ { \\mathrm { S E E D } } = \\displaystyle \\frac { 1 } { N * M } \\sum _ { i } ^ { N } \\sum _ { j } ^ { K + 1 } - \\frac { \\exp ( { \\bf z } _ { i } ^ { T } \\cdot { \\bf d } _ { j } / \\tau ^ { T } ) } { \\sum _ { \\bf d } \\sim \\mathrm { D } ^ { + } } \\cdot \\log \\Big \\frac { \\exp ( { \\bf z } _ { i } ^ { S } \\cdot { \\bf d } _ { j } / \\tau ^ { S } ) } { \\sum _ { \\bf d } \\sim \\mathrm { D } ^ { + } } } \\\\ & { \\quad \\quad \\quad \\quad = \\frac { 1 } { N * M } \\sum _ { i } ^ { N } \\displaystyle \\sum _ { j } ^ { K + 1 } \\underbrace { \\frac { \\exp ( { \\bf z } _ { i } ^ { T } \\cdot { \\bf d } _ { j } / \\tau ^ { T } ) } { \\sum _ { \\bf d } \\sim \\mathrm { D } ^ { + } } } _ { { \\bf w } _ { j } ^ { \\bf d } } \\cdot \\underbrace { ( - { \\bf z } _ { i } ^ { S } \\cdot { \\bf z } _ { i } ^ { T } / \\tau ^ { S } ) } _ { a l i g n m e n t } \\cdot \\underbrace { ( \\frac { \\log \\sum _ { i } ^ { S } \\cdot { \\bf d } _ { i } / \\tau ^ { S } ) } { c \\mathrm { o } n t r a s t i n g } } _ { c o n t r a s t i n g } ) , } \\end{array}",
+                                    "type": "interline_equation",
+                                    "image_path": "0df2375166674514ce463d7ecc5010b2bd9ba95a67bcbc95f409e7ab2e12a30c.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 26,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                111,
+                                510,
+                                503,
+                                539.6666666666666
+                            ],
+                            "spans": [],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                111,
+                                539.6666666666666,
+                                503,
+                                569.3333333333333
+                            ],
+                            "spans": [],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                111,
+                                569.3333333333333,
+                                503,
+                                598.9999999999999
+                            ],
+                            "spans": [],
+                            "index": 27
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        605,
+                        507,
+                        628
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                604,
+                                505,
+                                620
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        604,
+                                        364,
+                                        620
+                                    ],
+                                    "score": 1.0,
+                                    "content": "where the normalization term can be considered as soft labels,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        364,
+                                        605,
+                                        456,
+                                        619
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\mathbf { W } ^ { i } = \\left[ \\mathbf { w } _ { 1 } ^ { i } \\ldots \\mathbf { w } _ { K + 1 } ^ { i } \\right]",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        457,
+                                        604,
+                                        505,
+                                        620
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", which can",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                617,
+                                208,
+                                630
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        617,
+                                        208,
+                                        630
+                                    ],
+                                    "score": 1.0,
+                                    "content": "weight the above loss as:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        }
+                    ],
+                    "index": 28.5
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        146,
+                        633,
+                        464,
+                        669
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                146,
+                                633,
+                                464,
+                                669
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        146,
+                                        633,
+                                        464,
+                                        669
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\mathcal { L } _ { \\mathrm { S E E D } } = \\frac { 1 } { N * M } \\sum _ { i } ^ { N } \\sum _ { j } ^ { K + 1 } \\mathbf { w } _ { j } ^ { i } \\cdot \\Big \\{ - \\mathbf { z } _ { i } ^ { S } \\cdot \\mathbf { z } _ { i } ^ { T } / \\tau ^ { S } + \\log \\sum _ { \\mathbf { d } \\sim \\mathbf { D } } \\exp ( \\mathbf { z } _ { i } ^ { S } \\cdot \\mathbf { d } / \\tau ^ { S } ) ) \\Big \\} ,",
+                                    "type": "interline_equation",
+                                    "image_path": "32d06c560859d8112c17701b5a0909209017dbe9d795a6964a507eb73141f98d.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 31,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                146,
+                                633,
+                                464,
+                                645.0
+                            ],
+                            "spans": [],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                146,
+                                645.0,
+                                464,
+                                657.0
+                            ],
+                            "spans": [],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                146,
+                                657.0,
+                                464,
+                                669.0
+                            ],
+                            "spans": [],
+                            "index": 32
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        673,
+                        505,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                673,
+                                506,
+                                687
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        673,
+                                        231,
+                                        687
+                                    ],
+                                    "score": 1.0,
+                                    "content": "When tuning hyper-parameter",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        231,
+                                        674,
+                                        244,
+                                        685
+                                    ],
+                                    "score": 0.89,
+                                    "content": "\\tau ^ { T }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        245,
+                                        673,
+                                        291,
+                                        687
+                                    ],
+                                    "score": 1.0,
+                                    "content": "towards 0,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        291,
+                                        674,
+                                        307,
+                                        685
+                                    ],
+                                    "score": 0.67,
+                                    "content": "\\mathbf { W } ^ { i }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        308,
+                                        673,
+                                        506,
+                                        687
+                                    ],
+                                    "score": 1.0,
+                                    "content": "can be altered into the format of one-hot vector",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                684,
+                                506,
+                                701
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        684,
+                                        127,
+                                        701
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        127,
+                                        685,
+                                        172,
+                                        699
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\mathbf { w } _ { K + 1 } ^ { i } = \\dot { 1 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        172,
+                                        684,
+                                        506,
+                                        701
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", which is then degraded to the case of contrastive distillation as in equation 14. In",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                698,
+                                506,
+                                711
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        698,
+                                        239,
+                                        711
+                                    ],
+                                    "score": 1.0,
+                                    "content": "practice, the choice of an optimal",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        239,
+                                        698,
+                                        252,
+                                        709
+                                    ],
+                                    "score": 0.87,
+                                    "content": "\\tau ^ { T }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        252,
+                                        698,
+                                        442,
+                                        711
+                                    ],
+                                    "score": 1.0,
+                                    "content": "can be dataset-specific. We show that the higher",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        443,
+                                        699,
+                                        455,
+                                        709
+                                    ],
+                                    "score": 0.87,
+                                    "content": "\\tau ^ { T }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        456,
+                                        698,
+                                        506,
+                                        711
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(with labels",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                708,
+                                506,
+                                722
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        708,
+                                        506,
+                                        722
+                                    ],
+                                    "score": 1.0,
+                                    "content": "be more ‘soft’) can actually yield better results on other datasets, e.g., CIFAR-10 (Krizhevsky et al.,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                718,
+                                135,
+                                733
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        718,
+                                        135,
+                                        733
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2009).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        }
+                    ],
+                    "index": 35
+                }
+            ],
+            "page_idx": 18,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        107,
+                        26,
+                        292,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                25,
+                                293,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        25,
+                                        293,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Published as a conference paper at ICLR 2021",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        300,
+                        751,
+                        311,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                299,
+                                750,
+                                312,
+                                765
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        299,
+                                        750,
+                                        312,
+                                        765
+                                    ],
+                                    "score": 1.0,
+                                    "content": "19",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        81,
+                        505,
+                        106
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                81,
+                                505,
+                                95
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        81,
+                                        505,
+                                        95
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Thus, this demonstrates that the objective for SEED as Eq. 8 is equivalent to minimizing a weakened",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                93,
+                                200,
+                                106
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        93,
+                                        200,
+                                        106
+                                    ],
+                                    "score": 1.0,
+                                    "content": "upper-bound of e.q. 7:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        }
+                    ],
+                    "index": 0.5,
+                    "bbox_fs": [
+                        105,
+                        81,
+                        505,
+                        106
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        111,
+                        109,
+                        505,
+                        216
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                111,
+                                109,
+                                505,
+                                216
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        111,
+                                        109,
+                                        505,
+                                        216
+                                    ],
+                                    "score": 0.94,
+                                    "content": "\\begin{array} { l } { { \\mathcal { L } _ { \\mathrm { S E E D } } = \\displaystyle \\frac { 1 } { N } \\sum _ { i } ^ { N } - { \\bf p } ^ { T } ( { \\bf x } _ { i } ; \\boldsymbol { \\theta } _ { T } , { \\bf D } ^ { + } ) \\cdot \\log { \\bf p } ^ { S } ( { \\bf x } _ { i } ; \\boldsymbol { \\theta } _ { S } , { \\bf D } ^ { + } ) } \\ ~ } \\\\ { { \\displaystyle \\geq \\frac { 1 } { N } \\sum _ { i } ^ { N } \\left\\{ { \\bf w } _ { K + 1 } ^ { i } \\cdot ( - { \\bf z } _ { i } ^ { S } \\cdot { \\bf z } _ { i } ^ { T } / \\tau ^ { S } ) + \\sum _ { j = 1 } ^ { K } { \\bf w } _ { j } ^ { i } \\cdot \\left( - { \\bf z } _ { i } ^ { S } \\cdot { \\bf d } _ { j } / \\tau ^ { S } + \\log \\sum _ { \\bf d \\times D } ^ { C } \\exp ( { \\bf z } _ { i } ^ { S } \\cdot { \\bf d } / \\tau ^ { S } ) \\right) \\right\\} } \\ ~ } \\\\ { { \\displaystyle = \\frac { 1 } { N } \\sum _ { i } ^ { N } \\Bigg \\{ - \\frac { { \\bf w } _ { K + 1 } ^ { i } } { \\tau ^ { S } } \\cdot { \\bf z } _ { i } ^ { S } \\cdot { \\bf z } _ { i } ^ { T } - { \\bf p } ^ { T } ( { \\bf x } _ { i } ; \\boldsymbol { \\theta } _ { T } , { \\bf D } ) \\cdot \\log { \\bf p } ^ { S } ( { \\bf x } _ { i } ; \\boldsymbol { \\theta } _ { S } , { \\bf D } ) \\Bigg \\} } \\ ~ } \\end{array}",
+                                    "type": "interline_equation",
+                                    "image_path": "d86539a726b5a86171c5a58180e145908afef1dd30004cb1df11f07069fb1ca7.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 3,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                111,
+                                109,
+                                505,
+                                144.66666666666666
+                            ],
+                            "spans": [],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                111,
+                                144.66666666666666,
+                                505,
+                                180.33333333333331
+                            ],
+                            "spans": [],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                111,
+                                180.33333333333331,
+                                505,
+                                215.99999999999997
+                            ],
+                            "spans": [],
+                            "index": 4
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        223,
+                        505,
+                        268
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                223,
+                                506,
+                                237
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        223,
+                                        189,
+                                        237
+                                    ],
+                                    "score": 1.0,
+                                    "content": "This proves that our",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        189,
+                                        225,
+                                        216,
+                                        236
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\mathcal { L } _ { \\mathrm { S E E D } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        216,
+                                        223,
+                                        486,
+                                        237
+                                    ],
+                                    "score": 1.0,
+                                    "content": "directly relates to a more intuitive distillation formulation as Eq. 7",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        487,
+                                        225,
+                                        506,
+                                        235
+                                    ],
+                                    "score": 0.69,
+                                    "content": "( l 2 +",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                235,
+                                506,
+                                247
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        235,
+                                        506,
+                                        247
+                                    ],
+                                    "score": 1.0,
+                                    "content": "cross entropy loss), and it implicitly contains the objective of aligning and contrasting. However, our",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                245,
+                                506,
+                                259
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        245,
+                                        409,
+                                        259
+                                    ],
+                                    "score": 1.0,
+                                    "content": "training objective is much simplified. During practice, we find by regulating",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        409,
+                                        245,
+                                        422,
+                                        256
+                                    ],
+                                    "score": 0.88,
+                                    "content": "\\tau ^ { T }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        422,
+                                        245,
+                                        506,
+                                        259
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", both training losses",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                257,
+                                196,
+                                269
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        257,
+                                        196,
+                                        269
+                                    ],
+                                    "score": 1.0,
+                                    "content": "produce equal results.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        }
+                    ],
+                    "index": 6.5,
+                    "bbox_fs": [
+                        104,
+                        223,
+                        506,
+                        269
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        280,
+                        406,
+                        293
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                280,
+                                408,
+                                294
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        280,
+                                        408,
+                                        294
+                                    ],
+                                    "score": 1.0,
+                                    "content": "A.6 DISCUSSION ON THE RELATIONSHIP OF SEED WITH INFO-NCE",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        }
+                    ],
+                    "index": 9
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        301,
+                        506,
+                        358
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                302,
+                                505,
+                                314
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        302,
+                                        505,
+                                        314
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The objective of distillation can be considered as a soft version of Info-NCE (Oord et al., 2018), with",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                313,
+                                505,
+                                325
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        313,
+                                        505,
+                                        325
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the only difference to be that SEED learns from the negative samples with probabilities instead of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                324,
+                                507,
+                                337
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        324,
+                                        507,
+                                        337
+                                    ],
+                                    "score": 1.0,
+                                    "content": "treating them all strictly as negative samples. To be more specific, following Info-NCE, the “hard”",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                335,
+                                506,
+                                348
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        335,
+                                        506,
+                                        348
+                                    ],
+                                    "score": 1.0,
+                                    "content": "style contrastive distillation can be expressed as aligning with representations from the Teacher",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                346,
+                                312,
+                                359
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        346,
+                                        312,
+                                        359
+                                    ],
+                                    "score": 1.0,
+                                    "content": "encoder and contrasting with all random instances:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        }
+                    ],
+                    "index": 12,
+                    "bbox_fs": [
+                        105,
+                        302,
+                        507,
+                        359
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        167,
+                        362,
+                        433,
+                        396
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                167,
+                                362,
+                                433,
+                                396
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        167,
+                                        362,
+                                        433,
+                                        396
+                                    ],
+                                    "score": 0.94,
+                                    "content": "\\hat { \\theta } _ { S } = \\underset { \\theta _ { S } } { \\arg \\operatorname* { m i n } } \\ : \\mathcal { L } _ { N C E } = \\underset { \\theta _ { S } } { \\arg \\operatorname* { m i n } } \\sum _ { i } ^ { N } - \\log \\frac { \\exp ( \\mathbf { z } _ { i } ^ { T } \\cdot \\mathbf { z } _ { i } ^ { S } / \\tau ) } { \\sum _ { \\mathbf { d } \\sim \\mathbf { D } } \\exp ( \\mathbf { z } _ { i } ^ { S } \\cdot \\mathbf { d } / \\tau ) }",
+                                    "type": "interline_equation",
+                                    "image_path": "e34ea33920cc7ebc30f1c6559aaa3da4456286629890e13238bec91b4ea5dd57.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 16,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                167,
+                                362,
+                                433,
+                                373.3333333333333
+                            ],
+                            "spans": [],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                167,
+                                373.3333333333333,
+                                433,
+                                384.66666666666663
+                            ],
+                            "spans": [],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                167,
+                                384.66666666666663,
+                                433,
+                                395.99999999999994
+                            ],
+                            "spans": [],
+                            "index": 17
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        400,
+                        504,
+                        423
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                400,
+                                505,
+                                413
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        400,
+                                        505,
+                                        413
+                                    ],
+                                    "score": 1.0,
+                                    "content": "which can be further deduced with two sub-terms consisting of positive sample alignment and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                412,
+                                251,
+                                425
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        412,
+                                        251,
+                                        425
+                                    ],
+                                    "score": 1.0,
+                                    "content": "contrasting with negative instances:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        }
+                    ],
+                    "index": 18.5,
+                    "bbox_fs": [
+                        106,
+                        400,
+                        505,
+                        425
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        190,
+                        427,
+                        420,
+                        478
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                190,
+                                427,
+                                420,
+                                478
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        190,
+                                        427,
+                                        420,
+                                        478
+                                    ],
+                                    "score": 0.94,
+                                    "content": "\\mathcal { L } _ { N C E } = \\sum _ { i } ^ { N } \\Big \\{ \\underbrace { - \\mathbf { z } _ { i } ^ { S } \\cdot \\mathbf { z } _ { i } ^ { T } / \\tau } _ { a l i g n m e n t } + \\underbrace { \\log \\sum _ { \\mathbf { d } \\sim \\mathbf { D } } \\exp ( \\mathbf { z } _ { i } ^ { S } \\cdot \\mathbf { d } / \\tau ) } _ { c o n t r a s t i n g } \\Big \\} .",
+                                    "type": "interline_equation",
+                                    "image_path": "065e9718ac69ca6ad65c1327e04b7725124e89d075d2dedbd19848d52002ec32.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 21,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                190,
+                                427,
+                                420,
+                                444.0
+                            ],
+                            "spans": [],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                190,
+                                444.0,
+                                420,
+                                461.0
+                            ],
+                            "spans": [],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                190,
+                                461.0,
+                                420,
+                                478.0
+                            ],
+                            "spans": [],
+                            "index": 22
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        483,
+                        505,
+                        506
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                482,
+                                506,
+                                496
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        482,
+                                        506,
+                                        496
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Similarly, the objective of SEED can be dissembled into the weighted form of alignment and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                496,
+                                180,
+                                507
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        496,
+                                        180,
+                                        507
+                                    ],
+                                    "score": 1.0,
+                                    "content": "contrasting terms:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        }
+                    ],
+                    "index": 23.5,
+                    "bbox_fs": [
+                        106,
+                        482,
+                        506,
+                        507
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        111,
+                        510,
+                        503,
+                        599
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                111,
+                                510,
+                                503,
+                                599
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        111,
+                                        510,
+                                        503,
+                                        599
+                                    ],
+                                    "score": 0.94,
+                                    "content": "\\begin{array} { r l } & { \\mathcal { L } _ { \\mathrm { S E E D } } = \\displaystyle \\frac { 1 } { N * M } \\sum _ { i } ^ { N } \\sum _ { j } ^ { K + 1 } - \\frac { \\exp ( { \\bf z } _ { i } ^ { T } \\cdot { \\bf d } _ { j } / \\tau ^ { T } ) } { \\sum _ { \\bf d } \\sim \\mathrm { D } ^ { + } } \\cdot \\log \\Big \\frac { \\exp ( { \\bf z } _ { i } ^ { S } \\cdot { \\bf d } _ { j } / \\tau ^ { S } ) } { \\sum _ { \\bf d } \\sim \\mathrm { D } ^ { + } } } \\\\ & { \\quad \\quad \\quad \\quad = \\frac { 1 } { N * M } \\sum _ { i } ^ { N } \\displaystyle \\sum _ { j } ^ { K + 1 } \\underbrace { \\frac { \\exp ( { \\bf z } _ { i } ^ { T } \\cdot { \\bf d } _ { j } / \\tau ^ { T } ) } { \\sum _ { \\bf d } \\sim \\mathrm { D } ^ { + } } } _ { { \\bf w } _ { j } ^ { \\bf d } } \\cdot \\underbrace { ( - { \\bf z } _ { i } ^ { S } \\cdot { \\bf z } _ { i } ^ { T } / \\tau ^ { S } ) } _ { a l i g n m e n t } \\cdot \\underbrace { ( \\frac { \\log \\sum _ { i } ^ { S } \\cdot { \\bf d } _ { i } / \\tau ^ { S } ) } { c \\mathrm { o } n t r a s t i n g } } _ { c o n t r a s t i n g } ) , } \\end{array}",
+                                    "type": "interline_equation",
+                                    "image_path": "0df2375166674514ce463d7ecc5010b2bd9ba95a67bcbc95f409e7ab2e12a30c.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 26,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                111,
+                                510,
+                                503,
+                                539.6666666666666
+                            ],
+                            "spans": [],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                111,
+                                539.6666666666666,
+                                503,
+                                569.3333333333333
+                            ],
+                            "spans": [],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                111,
+                                569.3333333333333,
+                                503,
+                                598.9999999999999
+                            ],
+                            "spans": [],
+                            "index": 27
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        605,
+                        507,
+                        628
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                604,
+                                505,
+                                620
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        604,
+                                        364,
+                                        620
+                                    ],
+                                    "score": 1.0,
+                                    "content": "where the normalization term can be considered as soft labels,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        364,
+                                        605,
+                                        456,
+                                        619
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\mathbf { W } ^ { i } = \\left[ \\mathbf { w } _ { 1 } ^ { i } \\ldots \\mathbf { w } _ { K + 1 } ^ { i } \\right]",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        457,
+                                        604,
+                                        505,
+                                        620
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", which can",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                617,
+                                208,
+                                630
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        617,
+                                        208,
+                                        630
+                                    ],
+                                    "score": 1.0,
+                                    "content": "weight the above loss as:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        }
+                    ],
+                    "index": 28.5,
+                    "bbox_fs": [
+                        105,
+                        604,
+                        505,
+                        630
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        146,
+                        633,
+                        464,
+                        669
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                146,
+                                633,
+                                464,
+                                669
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        146,
+                                        633,
+                                        464,
+                                        669
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\mathcal { L } _ { \\mathrm { S E E D } } = \\frac { 1 } { N * M } \\sum _ { i } ^ { N } \\sum _ { j } ^ { K + 1 } \\mathbf { w } _ { j } ^ { i } \\cdot \\Big \\{ - \\mathbf { z } _ { i } ^ { S } \\cdot \\mathbf { z } _ { i } ^ { T } / \\tau ^ { S } + \\log \\sum _ { \\mathbf { d } \\sim \\mathbf { D } } \\exp ( \\mathbf { z } _ { i } ^ { S } \\cdot \\mathbf { d } / \\tau ^ { S } ) ) \\Big \\} ,",
+                                    "type": "interline_equation",
+                                    "image_path": "32d06c560859d8112c17701b5a0909209017dbe9d795a6964a507eb73141f98d.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 31,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                146,
+                                633,
+                                464,
+                                645.0
+                            ],
+                            "spans": [],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                146,
+                                645.0,
+                                464,
+                                657.0
+                            ],
+                            "spans": [],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                146,
+                                657.0,
+                                464,
+                                669.0
+                            ],
+                            "spans": [],
+                            "index": 32
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        673,
+                        505,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                673,
+                                506,
+                                687
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        673,
+                                        231,
+                                        687
+                                    ],
+                                    "score": 1.0,
+                                    "content": "When tuning hyper-parameter",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        231,
+                                        674,
+                                        244,
+                                        685
+                                    ],
+                                    "score": 0.89,
+                                    "content": "\\tau ^ { T }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        245,
+                                        673,
+                                        291,
+                                        687
+                                    ],
+                                    "score": 1.0,
+                                    "content": "towards 0,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        291,
+                                        674,
+                                        307,
+                                        685
+                                    ],
+                                    "score": 0.67,
+                                    "content": "\\mathbf { W } ^ { i }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        308,
+                                        673,
+                                        506,
+                                        687
+                                    ],
+                                    "score": 1.0,
+                                    "content": "can be altered into the format of one-hot vector",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                684,
+                                506,
+                                701
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        684,
+                                        127,
+                                        701
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        127,
+                                        685,
+                                        172,
+                                        699
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\mathbf { w } _ { K + 1 } ^ { i } = \\dot { 1 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        172,
+                                        684,
+                                        506,
+                                        701
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", which is then degraded to the case of contrastive distillation as in equation 14. In",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                698,
+                                506,
+                                711
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        698,
+                                        239,
+                                        711
+                                    ],
+                                    "score": 1.0,
+                                    "content": "practice, the choice of an optimal",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        239,
+                                        698,
+                                        252,
+                                        709
+                                    ],
+                                    "score": 0.87,
+                                    "content": "\\tau ^ { T }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        252,
+                                        698,
+                                        442,
+                                        711
+                                    ],
+                                    "score": 1.0,
+                                    "content": "can be dataset-specific. We show that the higher",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        443,
+                                        699,
+                                        455,
+                                        709
+                                    ],
+                                    "score": 0.87,
+                                    "content": "\\tau ^ { T }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        456,
+                                        698,
+                                        506,
+                                        711
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(with labels",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                708,
+                                506,
+                                722
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        708,
+                                        506,
+                                        722
+                                    ],
+                                    "score": 1.0,
+                                    "content": "be more ‘soft’) can actually yield better results on other datasets, e.g., CIFAR-10 (Krizhevsky et al.,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                718,
+                                135,
+                                733
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        718,
+                                        135,
+                                        733
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2009).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        }
+                    ],
+                    "index": 35,
+                    "bbox_fs": [
+                        105,
+                        673,
+                        506,
+                        733
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        82,
+                        347,
+                        93
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                81,
+                                349,
+                                95
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        81,
+                                        349,
+                                        95
+                                    ],
+                                    "score": 1.0,
+                                    "content": "A.7 COMPATIBILITY WITH SUPERVISED DISTILLATION",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        }
+                    ],
+                    "index": 0
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        103,
+                        505,
+                        169
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                101,
+                                507,
+                                117
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        101,
+                                        507,
+                                        117
+                                    ],
+                                    "score": 1.0,
+                                    "content": "SEED conducts self-supervised distillation at the pre-training phase for the representation learning.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                114,
+                                505,
+                                127
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        114,
+                                        505,
+                                        127
+                                    ],
+                                    "score": 1.0,
+                                    "content": "However, we verify that SEED is compatible with traditional supervised distillation that happened",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                126,
+                                505,
+                                137
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        126,
+                                        505,
+                                        137
+                                    ],
+                                    "score": 1.0,
+                                    "content": "during fine-tuning phrase at downstream, and can even produce better results. We begin with the SSL",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                137,
+                                505,
+                                149
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        137,
+                                        505,
+                                        149
+                                    ],
+                                    "score": 1.0,
+                                    "content": "pre-training on a larger architecture (ResNet-152) using MoCo-V2 and train it for 200 epochs as the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                146,
+                                506,
+                                161
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        146,
+                                        349,
+                                        161
+                                    ],
+                                    "score": 1.0,
+                                    "content": "teacher network. As images in CIFAR-100 are in the size of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        349,
+                                        147,
+                                        378,
+                                        158
+                                    ],
+                                    "score": 0.89,
+                                    "content": "3 2 \\times 3 2",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        378,
+                                        146,
+                                        506,
+                                        161
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", we modify the first conv layer",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                158,
+                                290,
+                                170
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        158,
+                                        213,
+                                        170
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in ResNet with kernel size",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        213,
+                                        159,
+                                        228,
+                                        169
+                                    ],
+                                    "score": 0.83,
+                                    "content": "= 3",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        229,
+                                        158,
+                                        270,
+                                        170
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and stride",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        270,
+                                        159,
+                                        285,
+                                        169
+                                    ],
+                                    "score": 0.85,
+                                    "content": "= 1",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        285,
+                                        158,
+                                        290,
+                                        170
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        }
+                    ],
+                    "index": 3.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        175,
+                        505,
+                        307
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                175,
+                                505,
+                                187
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        175,
+                                        505,
+                                        187
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We then compare the Top-1 accuracy of a smaller ResNet-18 on CIFAR-100 when using different",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                185,
+                                506,
+                                199
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        185,
+                                        506,
+                                        199
+                                    ],
+                                    "score": 1.0,
+                                    "content": "distillation strategies when all parameters are trainable. First, we use SEED to pre-train ResNet-18",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                196,
+                                506,
+                                210
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        196,
+                                        506,
+                                        210
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with Res-152 as the teacher model, and then evaluate in on the test split of CIFAR-100 using linear",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                209,
+                                505,
+                                221
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        209,
+                                        505,
+                                        221
+                                    ],
+                                    "score": 1.0,
+                                    "content": "fine-tuning task. As we keep all parameters trainable during the fine-tuning phase, distillation on the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                219,
+                                505,
+                                231
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        219,
+                                        262,
+                                        231
+                                    ],
+                                    "score": 1.0,
+                                    "content": "pre-training only yields a trivial boost:",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        263,
+                                        219,
+                                        290,
+                                        230
+                                    ],
+                                    "score": 0.79,
+                                    "content": "7 5 . 4 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        290,
+                                        219,
+                                        306,
+                                        231
+                                    ],
+                                    "score": 1.0,
+                                    "content": "v.s.",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        306,
+                                        219,
+                                        333,
+                                        230
+                                    ],
+                                    "score": 0.83,
+                                    "content": "7 5 . 2 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        334,
+                                        219,
+                                        505,
+                                        231
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Then, we adopt the traditional distillation",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                230,
+                                505,
+                                243
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        230,
+                                        505,
+                                        243
+                                    ],
+                                    "score": 1.0,
+                                    "content": "method, e.g., (Hinton et al., 2015), to first fine-tune the ResNet-152 model, and then use its output",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                240,
+                                506,
+                                254
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        240,
+                                        506,
+                                        254
+                                    ],
+                                    "score": 1.0,
+                                    "content": "class probability to facilitate the linear classification task on ResNet-18 in the fine-tuning phrase.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                252,
+                                505,
+                                264
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        252,
+                                        371,
+                                        264
+                                    ],
+                                    "score": 1.0,
+                                    "content": "This improves the linear classification accuracy on ResNet-18 to",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        372,
+                                        252,
+                                        399,
+                                        263
+                                    ],
+                                    "score": 0.88,
+                                    "content": "7 6 . 0 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        400,
+                                        252,
+                                        505,
+                                        264
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". At the end, we initialize",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                262,
+                                505,
+                                275
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        262,
+                                        505,
+                                        275
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the ResNet-18 with our SEED pre-trained ResNet-18, and equip it with the supervised classification",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                274,
+                                505,
+                                286
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        274,
+                                        505,
+                                        286
+                                    ],
+                                    "score": 1.0,
+                                    "content": "distillation during fine-tuning. With that, we find that the performance of ResNet-18 is further boosted",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                285,
+                                506,
+                                297
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        285,
+                                        116,
+                                        297
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        117,
+                                        285,
+                                        144,
+                                        295
+                                    ],
+                                    "score": 0.87,
+                                    "content": "7 8 . 1 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        144,
+                                        285,
+                                        506,
+                                        297
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". We can conclude that our SEED is compatible with traditional supervised distillation that",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                295,
+                                453,
+                                309
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        295,
+                                        453,
+                                        309
+                                    ],
+                                    "score": 1.0,
+                                    "content": "mostly happened at downstream for specific tasks, e.g., classification, object detection.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        }
+                    ],
+                    "index": 12.5
+                },
+                {
+                    "type": "table",
+                    "bbox": [
+                        195,
+                        364,
+                        414,
+                        431
+                    ],
+                    "blocks": [
+                        {
+                            "type": "table_caption",
+                            "bbox": [
+                                108,
+                                325,
+                                505,
+                                355
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        324,
+                                        506,
+                                        336
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                324,
+                                                248,
+                                                336
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Table 12: CIFAR-100 Top-1 Accuracy",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                249,
+                                                325,
+                                                262,
+                                                335
+                                            ],
+                                            "score": 0.51,
+                                            "content": "( \\% )",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                262,
+                                                324,
+                                                506,
+                                                336
+                                            ],
+                                            "score": 1.0,
+                                            "content": "of ResNet-18 with (or without) distillation at different phase: self-",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 19
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        334,
+                                        506,
+                                        346
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                334,
+                                                506,
+                                                346
+                                            ],
+                                            "score": 1.0,
+                                            "content": "supervised pre-training stage, and supervised classification fine-tuning. All backbone parameters of ResNet-18",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 20
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        345,
+                                        210,
+                                        356
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                345,
+                                                210,
+                                                356
+                                            ],
+                                            "score": 1.0,
+                                            "content": "are trainable in experiments.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 21
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "type": "table_body",
+                            "bbox": [
+                                195,
+                                364,
+                                414,
+                                431
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        195,
+                                        364,
+                                        414,
+                                        431
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                195,
+                                                364,
+                                                414,
+                                                431
+                                            ],
+                                            "score": 0.952,
+                                            "html": "<table><tr><td colspan=\"3\">Pre-training Distill. Fine-tuning Distill. Top-1 Acc</td></tr><tr><td>X</td><td>X</td><td>75.2</td></tr><tr><td>√</td><td>×</td><td>75.4</td></tr><tr><td>×</td><td>√</td><td>76.0</td></tr><tr><td>√</td><td>√</td><td>78.1</td></tr></table>",
+                                            "type": "table",
+                                            "image_path": "cc40158c716b6af83e7e123cc7de02ae87a713e9db269a0724d73ab8a599f371.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 24,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        195,
+                                        364,
+                                        414,
+                                        377.4
+                                    ],
+                                    "spans": [],
+                                    "index": 22
+                                },
+                                {
+                                    "bbox": [
+                                        195,
+                                        377.4,
+                                        414,
+                                        390.79999999999995
+                                    ],
+                                    "spans": [],
+                                    "index": 23
+                                },
+                                {
+                                    "bbox": [
+                                        195,
+                                        390.79999999999995,
+                                        414,
+                                        404.19999999999993
+                                    ],
+                                    "spans": [],
+                                    "index": 24
+                                },
+                                {
+                                    "bbox": [
+                                        195,
+                                        404.19999999999993,
+                                        414,
+                                        417.5999999999999
+                                    ],
+                                    "spans": [],
+                                    "index": 25
+                                },
+                                {
+                                    "bbox": [
+                                        195,
+                                        417.5999999999999,
+                                        414,
+                                        430.9999999999999
+                                    ],
+                                    "spans": [],
+                                    "index": 26
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 22.0
+                },
+                {
+                    "type": "image",
+                    "bbox": [
+                        107,
+                        460,
+                        501,
+                        592
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                107,
+                                460,
+                                501,
+                                592
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        460,
+                                        501,
+                                        592
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                107,
+                                                460,
+                                                501,
+                                                592
+                                            ],
+                                            "score": 0.968,
+                                            "type": "image",
+                                            "image_path": "d2dbbfad613f736ba1bbdacdf0d815fdcec0dd5cde93e27f656664f449c6e7cd.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 28,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        460,
+                                        501,
+                                        504.0
+                                    ],
+                                    "spans": [],
+                                    "index": 27
+                                },
+                                {
+                                    "bbox": [
+                                        107,
+                                        504.0,
+                                        501,
+                                        548.0
+                                    ],
+                                    "spans": [],
+                                    "index": 28
+                                },
+                                {
+                                    "bbox": [
+                                        107,
+                                        548.0,
+                                        501,
+                                        592.0
+                                    ],
+                                    "spans": [],
+                                    "index": 29
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                107,
+                                602,
+                                505,
+                                633
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        602,
+                                        505,
+                                        614
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                602,
+                                                243,
+                                                614
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 7: Linear evaluation accuracy",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                243,
+                                                603,
+                                                257,
+                                                613
+                                            ],
+                                            "score": 0.68,
+                                            "content": "( \\% )",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                257,
+                                                602,
+                                                505,
+                                                614
+                                            ],
+                                            "score": 1.0,
+                                            "content": "of distillation between ResNet-18 (as the Student) and ResNet-50 (as",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 30
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        612,
+                                        506,
+                                        624
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                612,
+                                                283,
+                                                624
+                                            ],
+                                            "score": 1.0,
+                                            "content": "the Teacher) using different size of queue when",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                283,
+                                                613,
+                                                317,
+                                                622
+                                            ],
+                                            "score": 0.71,
+                                            "content": "\\mathrm { L R } { = } 0 . 0 3",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                318,
+                                                612,
+                                                506,
+                                                624
+                                            ],
+                                            "score": 1.0,
+                                            "content": "and weight decay=1e-6. Note the axis is the log(·)",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 31
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        622,
+                                        192,
+                                        635
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                622,
+                                                192,
+                                                635
+                                            ],
+                                            "score": 1.0,
+                                            "content": "value of queue lengths.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 32
+                                }
+                            ],
+                            "index": 31
+                        }
+                    ],
+                    "index": 29.5
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        109,
+                        655,
+                        276,
+                        667
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                654,
+                                278,
+                                668
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        654,
+                                        278,
+                                        668
+                                    ],
+                                    "score": 1.0,
+                                    "content": "A.8 ADDITIONAL ABLATION STUDIES",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        }
+                    ],
+                    "index": 33
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        676,
+                        505,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                676,
+                                506,
+                                690
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        676,
+                                        506,
+                                        690
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We study effects of different hyper-parameters to distillation using a ResNet-18 (as Student) and a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                686,
+                                507,
+                                700
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        686,
+                                        507,
+                                        700
+                                    ],
+                                    "score": 1.0,
+                                    "content": "SWAV pre-trained ResNet-50 (as Teacher) with small patch views. In specific, we list the Top-1 Acc.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                698,
+                                506,
+                                712
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        698,
+                                        372,
+                                        712
+                                    ],
+                                    "score": 1.0,
+                                    "content": "on validation split of ImageNet-1k using different lengths of queue",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        372,
+                                        699,
+                                        403,
+                                        709
+                                    ],
+                                    "score": 0.81,
+                                    "content": "K { = } 1 2 8",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        403,
+                                        698,
+                                        506,
+                                        712
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", 512, 1,024, 4,096, 8,192,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                709,
+                                505,
+                                723
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        709,
+                                        505,
+                                        723
+                                    ],
+                                    "score": 1.0,
+                                    "content": "16,384, 65,536) in Figure. 7. With the increasing of random data samples, the distillation boosts the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                720,
+                                506,
+                                734
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        720,
+                                        388,
+                                        734
+                                    ],
+                                    "score": 1.0,
+                                    "content": "accuracy of learned representations, however within a limited range:",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        389,
+                                        721,
+                                        409,
+                                        731
+                                    ],
+                                    "score": 0.84,
+                                    "content": "+ 1 . 5",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        410,
+                                        720,
+                                        506,
+                                        734
+                                    ],
+                                    "score": 1.0,
+                                    "content": "when the queue size is",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        }
+                    ],
+                    "index": 36
+                }
+            ],
+            "page_idx": 19,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        108,
+                        27,
+                        292,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                25,
+                                293,
+                                39
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        25,
+                                        293,
+                                        39
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Published as a conference paper at ICLR 2021",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        300,
+                        751,
+                        311,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                298,
+                                750,
+                                313,
+                                764
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        298,
+                                        750,
+                                        313,
+                                        764
+                                    ],
+                                    "score": 1.0,
+                                    "content": "",
+                                    "type": "text",
+                                    "height": 14,
+                                    "width": 15
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        82,
+                        347,
+                        93
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                81,
+                                349,
+                                95
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        81,
+                                        349,
+                                        95
+                                    ],
+                                    "score": 1.0,
+                                    "content": "A.7 COMPATIBILITY WITH SUPERVISED DISTILLATION",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        }
+                    ],
+                    "index": 0
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        103,
+                        505,
+                        169
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                101,
+                                507,
+                                117
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        101,
+                                        507,
+                                        117
+                                    ],
+                                    "score": 1.0,
+                                    "content": "SEED conducts self-supervised distillation at the pre-training phase for the representation learning.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                114,
+                                505,
+                                127
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        114,
+                                        505,
+                                        127
+                                    ],
+                                    "score": 1.0,
+                                    "content": "However, we verify that SEED is compatible with traditional supervised distillation that happened",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                126,
+                                505,
+                                137
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        126,
+                                        505,
+                                        137
+                                    ],
+                                    "score": 1.0,
+                                    "content": "during fine-tuning phrase at downstream, and can even produce better results. We begin with the SSL",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                137,
+                                505,
+                                149
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        137,
+                                        505,
+                                        149
+                                    ],
+                                    "score": 1.0,
+                                    "content": "pre-training on a larger architecture (ResNet-152) using MoCo-V2 and train it for 200 epochs as the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                146,
+                                506,
+                                161
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        146,
+                                        349,
+                                        161
+                                    ],
+                                    "score": 1.0,
+                                    "content": "teacher network. As images in CIFAR-100 are in the size of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        349,
+                                        147,
+                                        378,
+                                        158
+                                    ],
+                                    "score": 0.89,
+                                    "content": "3 2 \\times 3 2",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        378,
+                                        146,
+                                        506,
+                                        161
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", we modify the first conv layer",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                158,
+                                290,
+                                170
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        158,
+                                        213,
+                                        170
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in ResNet with kernel size",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        213,
+                                        159,
+                                        228,
+                                        169
+                                    ],
+                                    "score": 0.83,
+                                    "content": "= 3",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        229,
+                                        158,
+                                        270,
+                                        170
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and stride",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        270,
+                                        159,
+                                        285,
+                                        169
+                                    ],
+                                    "score": 0.85,
+                                    "content": "= 1",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        285,
+                                        158,
+                                        290,
+                                        170
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        }
+                    ],
+                    "index": 3.5,
+                    "bbox_fs": [
+                        105,
+                        101,
+                        507,
+                        170
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        175,
+                        505,
+                        307
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                175,
+                                505,
+                                187
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        175,
+                                        505,
+                                        187
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We then compare the Top-1 accuracy of a smaller ResNet-18 on CIFAR-100 when using different",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                185,
+                                506,
+                                199
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        185,
+                                        506,
+                                        199
+                                    ],
+                                    "score": 1.0,
+                                    "content": "distillation strategies when all parameters are trainable. First, we use SEED to pre-train ResNet-18",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                196,
+                                506,
+                                210
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        196,
+                                        506,
+                                        210
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with Res-152 as the teacher model, and then evaluate in on the test split of CIFAR-100 using linear",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                209,
+                                505,
+                                221
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        209,
+                                        505,
+                                        221
+                                    ],
+                                    "score": 1.0,
+                                    "content": "fine-tuning task. As we keep all parameters trainable during the fine-tuning phase, distillation on the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                219,
+                                505,
+                                231
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        219,
+                                        262,
+                                        231
+                                    ],
+                                    "score": 1.0,
+                                    "content": "pre-training only yields a trivial boost:",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        263,
+                                        219,
+                                        290,
+                                        230
+                                    ],
+                                    "score": 0.79,
+                                    "content": "7 5 . 4 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        290,
+                                        219,
+                                        306,
+                                        231
+                                    ],
+                                    "score": 1.0,
+                                    "content": "v.s.",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        306,
+                                        219,
+                                        333,
+                                        230
+                                    ],
+                                    "score": 0.83,
+                                    "content": "7 5 . 2 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        334,
+                                        219,
+                                        505,
+                                        231
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Then, we adopt the traditional distillation",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                230,
+                                505,
+                                243
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        230,
+                                        505,
+                                        243
+                                    ],
+                                    "score": 1.0,
+                                    "content": "method, e.g., (Hinton et al., 2015), to first fine-tune the ResNet-152 model, and then use its output",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                240,
+                                506,
+                                254
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        240,
+                                        506,
+                                        254
+                                    ],
+                                    "score": 1.0,
+                                    "content": "class probability to facilitate the linear classification task on ResNet-18 in the fine-tuning phrase.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                252,
+                                505,
+                                264
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        252,
+                                        371,
+                                        264
+                                    ],
+                                    "score": 1.0,
+                                    "content": "This improves the linear classification accuracy on ResNet-18 to",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        372,
+                                        252,
+                                        399,
+                                        263
+                                    ],
+                                    "score": 0.88,
+                                    "content": "7 6 . 0 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        400,
+                                        252,
+                                        505,
+                                        264
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". At the end, we initialize",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                262,
+                                505,
+                                275
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        262,
+                                        505,
+                                        275
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the ResNet-18 with our SEED pre-trained ResNet-18, and equip it with the supervised classification",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                274,
+                                505,
+                                286
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        274,
+                                        505,
+                                        286
+                                    ],
+                                    "score": 1.0,
+                                    "content": "distillation during fine-tuning. With that, we find that the performance of ResNet-18 is further boosted",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                285,
+                                506,
+                                297
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        285,
+                                        116,
+                                        297
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        117,
+                                        285,
+                                        144,
+                                        295
+                                    ],
+                                    "score": 0.87,
+                                    "content": "7 8 . 1 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        144,
+                                        285,
+                                        506,
+                                        297
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". We can conclude that our SEED is compatible with traditional supervised distillation that",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                295,
+                                453,
+                                309
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        295,
+                                        453,
+                                        309
+                                    ],
+                                    "score": 1.0,
+                                    "content": "mostly happened at downstream for specific tasks, e.g., classification, object detection.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        }
+                    ],
+                    "index": 12.5,
+                    "bbox_fs": [
+                        105,
+                        175,
+                        506,
+                        309
+                    ]
+                },
+                {
+                    "type": "table",
+                    "bbox": [
+                        195,
+                        364,
+                        414,
+                        431
+                    ],
+                    "blocks": [
+                        {
+                            "type": "table_caption",
+                            "bbox": [
+                                108,
+                                325,
+                                505,
+                                355
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        324,
+                                        506,
+                                        336
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                324,
+                                                248,
+                                                336
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Table 12: CIFAR-100 Top-1 Accuracy",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                249,
+                                                325,
+                                                262,
+                                                335
+                                            ],
+                                            "score": 0.51,
+                                            "content": "( \\% )",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                262,
+                                                324,
+                                                506,
+                                                336
+                                            ],
+                                            "score": 1.0,
+                                            "content": "of ResNet-18 with (or without) distillation at different phase: self-",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 19
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        334,
+                                        506,
+                                        346
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                334,
+                                                506,
+                                                346
+                                            ],
+                                            "score": 1.0,
+                                            "content": "supervised pre-training stage, and supervised classification fine-tuning. All backbone parameters of ResNet-18",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 20
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        345,
+                                        210,
+                                        356
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                345,
+                                                210,
+                                                356
+                                            ],
+                                            "score": 1.0,
+                                            "content": "are trainable in experiments.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 21
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "type": "table_body",
+                            "bbox": [
+                                195,
+                                364,
+                                414,
+                                431
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        195,
+                                        364,
+                                        414,
+                                        431
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                195,
+                                                364,
+                                                414,
+                                                431
+                                            ],
+                                            "score": 0.952,
+                                            "html": "<table><tr><td colspan=\"3\">Pre-training Distill. Fine-tuning Distill. Top-1 Acc</td></tr><tr><td>X</td><td>X</td><td>75.2</td></tr><tr><td>√</td><td>×</td><td>75.4</td></tr><tr><td>×</td><td>√</td><td>76.0</td></tr><tr><td>√</td><td>√</td><td>78.1</td></tr></table>",
+                                            "type": "table",
+                                            "image_path": "cc40158c716b6af83e7e123cc7de02ae87a713e9db269a0724d73ab8a599f371.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 24,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        195,
+                                        364,
+                                        414,
+                                        377.4
+                                    ],
+                                    "spans": [],
+                                    "index": 22
+                                },
+                                {
+                                    "bbox": [
+                                        195,
+                                        377.4,
+                                        414,
+                                        390.79999999999995
+                                    ],
+                                    "spans": [],
+                                    "index": 23
+                                },
+                                {
+                                    "bbox": [
+                                        195,
+                                        390.79999999999995,
+                                        414,
+                                        404.19999999999993
+                                    ],
+                                    "spans": [],
+                                    "index": 24
+                                },
+                                {
+                                    "bbox": [
+                                        195,
+                                        404.19999999999993,
+                                        414,
+                                        417.5999999999999
+                                    ],
+                                    "spans": [],
+                                    "index": 25
+                                },
+                                {
+                                    "bbox": [
+                                        195,
+                                        417.5999999999999,
+                                        414,
+                                        430.9999999999999
+                                    ],
+                                    "spans": [],
+                                    "index": 26
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 22.0
+                },
+                {
+                    "type": "image",
+                    "bbox": [
+                        107,
+                        460,
+                        501,
+                        592
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                107,
+                                460,
+                                501,
+                                592
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        460,
+                                        501,
+                                        592
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                107,
+                                                460,
+                                                501,
+                                                592
+                                            ],
+                                            "score": 0.968,
+                                            "type": "image",
+                                            "image_path": "d2dbbfad613f736ba1bbdacdf0d815fdcec0dd5cde93e27f656664f449c6e7cd.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 28,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        460,
+                                        501,
+                                        504.0
+                                    ],
+                                    "spans": [],
+                                    "index": 27
+                                },
+                                {
+                                    "bbox": [
+                                        107,
+                                        504.0,
+                                        501,
+                                        548.0
+                                    ],
+                                    "spans": [],
+                                    "index": 28
+                                },
+                                {
+                                    "bbox": [
+                                        107,
+                                        548.0,
+                                        501,
+                                        592.0
+                                    ],
+                                    "spans": [],
+                                    "index": 29
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                107,
+                                602,
+                                505,
+                                633
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        602,
+                                        505,
+                                        614
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                602,
+                                                243,
+                                                614
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 7: Linear evaluation accuracy",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                243,
+                                                603,
+                                                257,
+                                                613
+                                            ],
+                                            "score": 0.68,
+                                            "content": "( \\% )",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                257,
+                                                602,
+                                                505,
+                                                614
+                                            ],
+                                            "score": 1.0,
+                                            "content": "of distillation between ResNet-18 (as the Student) and ResNet-50 (as",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 30
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        612,
+                                        506,
+                                        624
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                612,
+                                                283,
+                                                624
+                                            ],
+                                            "score": 1.0,
+                                            "content": "the Teacher) using different size of queue when",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                283,
+                                                613,
+                                                317,
+                                                622
+                                            ],
+                                            "score": 0.71,
+                                            "content": "\\mathrm { L R } { = } 0 . 0 3",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                318,
+                                                612,
+                                                506,
+                                                624
+                                            ],
+                                            "score": 1.0,
+                                            "content": "and weight decay=1e-6. Note the axis is the log(·)",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 31
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        622,
+                                        192,
+                                        635
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                622,
+                                                192,
+                                                635
+                                            ],
+                                            "score": 1.0,
+                                            "content": "value of queue lengths.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 32
+                                }
+                            ],
+                            "index": 31
+                        }
+                    ],
+                    "index": 29.5
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        109,
+                        655,
+                        276,
+                        667
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                654,
+                                278,
+                                668
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        654,
+                                        278,
+                                        668
+                                    ],
+                                    "score": 1.0,
+                                    "content": "A.8 ADDITIONAL ABLATION STUDIES",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        }
+                    ],
+                    "index": 33
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        676,
+                        505,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                676,
+                                506,
+                                690
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        676,
+                                        506,
+                                        690
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We study effects of different hyper-parameters to distillation using a ResNet-18 (as Student) and a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                686,
+                                507,
+                                700
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        686,
+                                        507,
+                                        700
+                                    ],
+                                    "score": 1.0,
+                                    "content": "SWAV pre-trained ResNet-50 (as Teacher) with small patch views. In specific, we list the Top-1 Acc.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                698,
+                                506,
+                                712
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        698,
+                                        372,
+                                        712
+                                    ],
+                                    "score": 1.0,
+                                    "content": "on validation split of ImageNet-1k using different lengths of queue",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        372,
+                                        699,
+                                        403,
+                                        709
+                                    ],
+                                    "score": 0.81,
+                                    "content": "K { = } 1 2 8",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        403,
+                                        698,
+                                        506,
+                                        712
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", 512, 1,024, 4,096, 8,192,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                709,
+                                505,
+                                723
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        709,
+                                        505,
+                                        723
+                                    ],
+                                    "score": 1.0,
+                                    "content": "16,384, 65,536) in Figure. 7. With the increasing of random data samples, the distillation boosts the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                720,
+                                506,
+                                734
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        720,
+                                        388,
+                                        734
+                                    ],
+                                    "score": 1.0,
+                                    "content": "accuracy of learned representations, however within a limited range:",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        389,
+                                        721,
+                                        409,
+                                        731
+                                    ],
+                                    "score": 0.84,
+                                    "content": "+ 1 . 5",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        410,
+                                        720,
+                                        506,
+                                        734
+                                    ],
+                                    "score": 1.0,
+                                    "content": "when the queue size is",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        }
+                    ],
+                    "index": 36,
+                    "bbox_fs": [
+                        105,
+                        676,
+                        507,
+                        734
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "table",
+                    "bbox": [
+                        106,
+                        146,
+                        236,
+                        212
+                    ],
+                    "blocks": [
+                        {
+                            "type": "table_caption",
+                            "bbox": [
+                                106,
+                                86,
+                                291,
+                                137
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        86,
+                                        292,
+                                        97
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                86,
+                                                250,
+                                                97
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Table 13: Linear evaluation accuracy",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                250,
+                                                87,
+                                                264,
+                                                97
+                                            ],
+                                            "score": 0.6,
+                                            "content": "( \\% )",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                264,
+                                                86,
+                                                292,
+                                                97
+                                            ],
+                                            "score": 1.0,
+                                            "content": "of dis-",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 0
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        96,
+                                        291,
+                                        107
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                96,
+                                                291,
+                                                107
+                                            ],
+                                            "score": 1.0,
+                                            "content": "tillation between ResNet-18 (as the Student) and",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        106,
+                                        292,
+                                        117
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                106,
+                                                292,
+                                                117
+                                            ],
+                                            "score": 1.0,
+                                            "content": "ResNet-50 (as the Teacher) using different learn-",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 2
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        117,
+                                        292,
+                                        128
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                117,
+                                                292,
+                                                128
+                                            ],
+                                            "score": 1.0,
+                                            "content": "ing rates when the queue size is 65,536 and weight",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 3
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        126,
+                                        153,
+                                        137
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                126,
+                                                153,
+                                                137
+                                            ],
+                                            "score": 1.0,
+                                            "content": "decay=1e-6.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 4
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "type": "table_body",
+                            "bbox": [
+                                106,
+                                146,
+                                236,
+                                212
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        146,
+                                        236,
+                                        212
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                146,
+                                                236,
+                                                212
+                                            ],
+                                            "score": 0.972,
+                                            "html": "<table><tr><td>LR</td><td>Top-1 Acc. Top-5 Acc.</td><td></td></tr><tr><td>1</td><td>58.9</td><td>83.1</td></tr><tr><td>0.1</td><td>62.9</td><td>85.3</td></tr><tr><td>0.03</td><td>63.3</td><td>85.4</td></tr><tr><td>0.01</td><td>62.6</td><td>85.0</td></tr></table>",
+                                            "type": "table",
+                                            "image_path": "120cf29bca49ba5997cc5ea1380a9c72f739a04304171a9fcc5f244930c68c5f.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 9.5,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        146,
+                                        236,
+                                        179.0
+                                    ],
+                                    "spans": [],
+                                    "index": 9
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        179.0,
+                                        236,
+                                        212.0
+                                    ],
+                                    "spans": [],
+                                    "index": 10
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 5.75
+                },
+                {
+                    "type": "table",
+                    "bbox": [
+                        302,
+                        136,
+                        432,
+                        213
+                    ],
+                    "blocks": [
+                        {
+                            "type": "table_caption",
+                            "bbox": [
+                                302,
+                                86,
+                                502,
+                                127
+                            ],
+                            "group_id": 1,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        301,
+                                        85,
+                                        502,
+                                        97
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                301,
+                                                85,
+                                                438,
+                                                97
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Table 14: Linear evaluation accuracy",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                438,
+                                                87,
+                                                452,
+                                                96
+                                            ],
+                                            "score": 0.65,
+                                            "content": "( \\% )",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                452,
+                                                85,
+                                                502,
+                                                97
+                                            ],
+                                            "score": 1.0,
+                                            "content": "of distillation",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 5
+                                },
+                                {
+                                    "bbox": [
+                                        302,
+                                        96,
+                                        502,
+                                        107
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                302,
+                                                96,
+                                                502,
+                                                107
+                                            ],
+                                            "score": 1.0,
+                                            "content": "between ResNet-18 (as the Student) and ResNet-50 (as",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 6
+                                },
+                                {
+                                    "bbox": [
+                                        301,
+                                        106,
+                                        502,
+                                        117
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                301,
+                                                106,
+                                                502,
+                                                117
+                                            ],
+                                            "score": 1.0,
+                                            "content": "the Teacher) using different weight decays when the",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 7
+                                },
+                                {
+                                    "bbox": [
+                                        301,
+                                        116,
+                                        429,
+                                        126
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                301,
+                                                116,
+                                                392,
+                                                126
+                                            ],
+                                            "score": 1.0,
+                                            "content": "queue size is 65,536 and",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                392,
+                                                116,
+                                                425,
+                                                126
+                                            ],
+                                            "score": 0.8,
+                                            "content": "\\mathrm { L R } { = } 0 . 0 3",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                426,
+                                                116,
+                                                429,
+                                                126
+                                            ],
+                                            "score": 1.0,
+                                            "content": ".",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 8
+                                }
+                            ],
+                            "index": 6.5
+                        },
+                        {
+                            "type": "table_body",
+                            "bbox": [
+                                302,
+                                136,
+                                432,
+                                213
+                            ],
+                            "group_id": 1,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        302,
+                                        136,
+                                        432,
+                                        213
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                302,
+                                                136,
+                                                432,
+                                                213
+                                            ],
+                                            "score": 0.968,
+                                            "html": "<table><tr><td>WD</td><td>Top-1 Acc. Top-5 Acc.</td><td></td></tr><tr><td>1e-2</td><td>11.8</td><td>27.7</td></tr><tr><td>1e-3</td><td>62.3</td><td>84.7</td></tr><tr><td>1e-4</td><td>61.9</td><td>84.4</td></tr><tr><td>1e-5</td><td>61.6</td><td>84.2</td></tr><tr><td>1e-6</td><td>63.3</td><td>85.4</td></tr></table>",
+                                            "type": "table",
+                                            "image_path": "b86126ebcbfd9cbef60e6b42f12f5be8b33e96aeaf60d51c629bc07499ef28df.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 11.5,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        302,
+                                        136,
+                                        432,
+                                        174.5
+                                    ],
+                                    "spans": [],
+                                    "index": 11
+                                },
+                                {
+                                    "bbox": [
+                                        302,
+                                        174.5,
+                                        432,
+                                        213.0
+                                    ],
+                                    "spans": [],
+                                    "index": 12
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 9.0
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        236,
+                        504,
+                        259
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                235,
+                                505,
+                                250
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        235,
+                                        505,
+                                        250
+                                    ],
+                                    "score": 1.0,
+                                    "content": "65,536 compared with 256. Furthermore, Table. 13 and 14 summarize the linear evaluation accuracy",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                246,
+                                303,
+                                260
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        246,
+                                        303,
+                                        260
+                                    ],
+                                    "score": 1.0,
+                                    "content": "under different learning rates and weight decays.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        }
+                    ],
+                    "index": 13.5
+                }
+            ],
+            "page_idx": 20,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        108,
+                        27,
+                        292,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                26,
+                                293,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        26,
+                                        293,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Published as a conference paper at ICLR 2021",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        300,
+                        751,
+                        310,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                298,
+                                750,
+                                312,
+                                764
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        298,
+                                        750,
+                                        312,
+                                        764
+                                    ],
+                                    "score": 1.0,
+                                    "content": "",
+                                    "type": "text",
+                                    "height": 14,
+                                    "width": 14
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "table",
+                    "bbox": [
+                        106,
+                        146,
+                        236,
+                        212
+                    ],
+                    "blocks": [
+                        {
+                            "type": "table_caption",
+                            "bbox": [
+                                106,
+                                86,
+                                291,
+                                137
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        86,
+                                        292,
+                                        97
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                86,
+                                                250,
+                                                97
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Table 13: Linear evaluation accuracy",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                250,
+                                                87,
+                                                264,
+                                                97
+                                            ],
+                                            "score": 0.6,
+                                            "content": "( \\% )",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                264,
+                                                86,
+                                                292,
+                                                97
+                                            ],
+                                            "score": 1.0,
+                                            "content": "of dis-",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 0
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        96,
+                                        291,
+                                        107
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                96,
+                                                291,
+                                                107
+                                            ],
+                                            "score": 1.0,
+                                            "content": "tillation between ResNet-18 (as the Student) and",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        106,
+                                        292,
+                                        117
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                106,
+                                                292,
+                                                117
+                                            ],
+                                            "score": 1.0,
+                                            "content": "ResNet-50 (as the Teacher) using different learn-",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 2
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        117,
+                                        292,
+                                        128
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                117,
+                                                292,
+                                                128
+                                            ],
+                                            "score": 1.0,
+                                            "content": "ing rates when the queue size is 65,536 and weight",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 3
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        126,
+                                        153,
+                                        137
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                126,
+                                                153,
+                                                137
+                                            ],
+                                            "score": 1.0,
+                                            "content": "decay=1e-6.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 4
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "type": "table_body",
+                            "bbox": [
+                                106,
+                                146,
+                                236,
+                                212
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        146,
+                                        236,
+                                        212
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                146,
+                                                236,
+                                                212
+                                            ],
+                                            "score": 0.972,
+                                            "html": "<table><tr><td>LR</td><td>Top-1 Acc. Top-5 Acc.</td><td></td></tr><tr><td>1</td><td>58.9</td><td>83.1</td></tr><tr><td>0.1</td><td>62.9</td><td>85.3</td></tr><tr><td>0.03</td><td>63.3</td><td>85.4</td></tr><tr><td>0.01</td><td>62.6</td><td>85.0</td></tr></table>",
+                                            "type": "table",
+                                            "image_path": "120cf29bca49ba5997cc5ea1380a9c72f739a04304171a9fcc5f244930c68c5f.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 9.5,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        146,
+                                        236,
+                                        179.0
+                                    ],
+                                    "spans": [],
+                                    "index": 9
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        179.0,
+                                        236,
+                                        212.0
+                                    ],
+                                    "spans": [],
+                                    "index": 10
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 5.75
+                },
+                {
+                    "type": "table",
+                    "bbox": [
+                        302,
+                        136,
+                        432,
+                        213
+                    ],
+                    "blocks": [
+                        {
+                            "type": "table_caption",
+                            "bbox": [
+                                302,
+                                86,
+                                502,
+                                127
+                            ],
+                            "group_id": 1,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        301,
+                                        85,
+                                        502,
+                                        97
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                301,
+                                                85,
+                                                438,
+                                                97
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Table 14: Linear evaluation accuracy",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                438,
+                                                87,
+                                                452,
+                                                96
+                                            ],
+                                            "score": 0.65,
+                                            "content": "( \\% )",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                452,
+                                                85,
+                                                502,
+                                                97
+                                            ],
+                                            "score": 1.0,
+                                            "content": "of distillation",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 5
+                                },
+                                {
+                                    "bbox": [
+                                        302,
+                                        96,
+                                        502,
+                                        107
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                302,
+                                                96,
+                                                502,
+                                                107
+                                            ],
+                                            "score": 1.0,
+                                            "content": "between ResNet-18 (as the Student) and ResNet-50 (as",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 6
+                                },
+                                {
+                                    "bbox": [
+                                        301,
+                                        106,
+                                        502,
+                                        117
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                301,
+                                                106,
+                                                502,
+                                                117
+                                            ],
+                                            "score": 1.0,
+                                            "content": "the Teacher) using different weight decays when the",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 7
+                                },
+                                {
+                                    "bbox": [
+                                        301,
+                                        116,
+                                        429,
+                                        126
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                301,
+                                                116,
+                                                392,
+                                                126
+                                            ],
+                                            "score": 1.0,
+                                            "content": "queue size is 65,536 and",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                392,
+                                                116,
+                                                425,
+                                                126
+                                            ],
+                                            "score": 0.8,
+                                            "content": "\\mathrm { L R } { = } 0 . 0 3",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                426,
+                                                116,
+                                                429,
+                                                126
+                                            ],
+                                            "score": 1.0,
+                                            "content": ".",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 8
+                                }
+                            ],
+                            "index": 6.5
+                        },
+                        {
+                            "type": "table_body",
+                            "bbox": [
+                                302,
+                                136,
+                                432,
+                                213
+                            ],
+                            "group_id": 1,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        302,
+                                        136,
+                                        432,
+                                        213
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                302,
+                                                136,
+                                                432,
+                                                213
+                                            ],
+                                            "score": 0.968,
+                                            "html": "<table><tr><td>WD</td><td>Top-1 Acc. Top-5 Acc.</td><td></td></tr><tr><td>1e-2</td><td>11.8</td><td>27.7</td></tr><tr><td>1e-3</td><td>62.3</td><td>84.7</td></tr><tr><td>1e-4</td><td>61.9</td><td>84.4</td></tr><tr><td>1e-5</td><td>61.6</td><td>84.2</td></tr><tr><td>1e-6</td><td>63.3</td><td>85.4</td></tr></table>",
+                                            "type": "table",
+                                            "image_path": "b86126ebcbfd9cbef60e6b42f12f5be8b33e96aeaf60d51c629bc07499ef28df.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 11.5,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        302,
+                                        136,
+                                        432,
+                                        174.5
+                                    ],
+                                    "spans": [],
+                                    "index": 11
+                                },
+                                {
+                                    "bbox": [
+                                        302,
+                                        174.5,
+                                        432,
+                                        213.0
+                                    ],
+                                    "spans": [],
+                                    "index": 12
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 9.0
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        236,
+                        504,
+                        259
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                235,
+                                505,
+                                250
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        235,
+                                        505,
+                                        250
+                                    ],
+                                    "score": 1.0,
+                                    "content": "65,536 compared with 256. Furthermore, Table. 13 and 14 summarize the linear evaluation accuracy",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                246,
+                                303,
+                                260
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        246,
+                                        303,
+                                        260
+                                    ],
+                                    "score": 1.0,
+                                    "content": "under different learning rates and weight decays.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        }
+                    ],
+                    "index": 13.5,
+                    "bbox_fs": [
+                        105,
+                        235,
+                        505,
+                        260
+                    ]
+                }
+            ]
+        }
+    ],
+    "_backend": "pipeline",
+    "_version_name": "2.2.2"
+}
\ No newline at end of file
diff --git a/parse/train/AHm3dbp7D1D/AHm3dbp7D1D_model.json b/parse/train/AHm3dbp7D1D/AHm3dbp7D1D_model.json
new file mode 100644
index 0000000000000000000000000000000000000000..330b087cb86ead8c63e5cdea08d6a3f133895d6c
--- /dev/null
+++ b/parse/train/AHm3dbp7D1D/AHm3dbp7D1D_model.json
@@ -0,0 +1,27683 @@
+[
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    398,
+                    672,
+                    1304,
+                    672,
+                    1304,
+                    1069,
+                    398,
+                    1069
+                ],
+                "score": 0.983
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1192,
+                    892,
+                    1192,
+                    892,
+                    1803,
+                    298,
+                    1803
+                ],
+                "score": 0.981
+            },
+            {
+                "category_id": 3,
+                "poly": [
+                    918,
+                    1209,
+                    1395,
+                    1209,
+                    1395,
+                    1567,
+                    918,
+                    1567
+                ],
+                "score": 0.972
+            },
+            {
+                "category_id": 4,
+                "poly": [
+                    913,
+                    1582,
+                    1404,
+                    1582,
+                    1404,
+                    1865,
+                    913,
+                    1865
+                ],
+                "score": 0.967
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    1944,
+                    1403,
+                    1944,
+                    1403,
+                    2034,
+                    299,
+                    2034
+                ],
+                "score": 0.959
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    301,
+                    222,
+                    1245,
+                    222,
+                    1245,
+                    330,
+                    301,
+                    330
+                ],
+                "score": 0.956
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    441,
+                    418,
+                    1223,
+                    418,
+                    1223,
+                    508,
+                    441,
+                    508
+                ],
+                "score": 0.935
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1821,
+                    892,
+                    1821,
+                    892,
+                    1943,
+                    297,
+                    1943
+                ],
+                "score": 0.935
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    773,
+                    603,
+                    926,
+                    603,
+                    926,
+                    636,
+                    773,
+                    636
+                ],
+                "score": 0.882
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    300,
+                    75,
+                    813,
+                    75,
+                    813,
+                    105,
+                    300,
+                    105
+                ],
+                "score": 0.879
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    302,
+                    1128,
+                    570,
+                    1128,
+                    570,
+                    1160,
+                    302,
+                    1160
+                ],
+                "score": 0.87
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    319,
+                    379,
+                    1386,
+                    379,
+                    1386,
+                    415,
+                    319,
+                    415
+                ],
+                "score": 0.864
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    841,
+                    2088,
+                    856,
+                    2088,
+                    856,
+                    2112,
+                    841,
+                    2112
+                ],
+                "score": 0.732
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1064,
+                    1008,
+                    1138,
+                    1008,
+                    1138,
+                    1037,
+                    1064,
+                    1037
+                ],
+                "score": 0.87,
+                "latex": "6 8 . 2 \\%"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    561,
+                    1008,
+                    636,
+                    1008,
+                    636,
+                    1037,
+                    561,
+                    1037
+                ],
+                "score": 0.87,
+                "latex": "6 7 . 6 \\%"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    456,
+                    1008,
+                    530,
+                    1008,
+                    530,
+                    1037,
+                    456,
+                    1037
+                ],
+                "score": 0.87,
+                "latex": "4 2 . 2 \\%"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    959,
+                    1008,
+                    1033,
+                    1008,
+                    1033,
+                    1037,
+                    959,
+                    1037
+                ],
+                "score": 0.87,
+                "latex": "3 6 . 3 \\%"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1132,
+                    1973,
+                    1209,
+                    1973,
+                    1209,
+                    2002,
+                    1132,
+                    2002
+                ],
+                "score": 0.85,
+                "latex": "3 6 . 3 \\%"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1273.0,
+                    1215.0,
+                    1354.0,
+                    1215.0,
+                    1354.0,
+                    1249.0,
+                    1273.0,
+                    1249.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    912.0,
+                    1217.0,
+                    953.0,
+                    1217.0,
+                    953.0,
+                    1535.0,
+                    912.0,
+                    1535.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    949.0,
+                    1255.0,
+                    984.0,
+                    1255.0,
+                    984.0,
+                    1289.0,
+                    949.0,
+                    1289.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1338.0,
+                    1260.0,
+                    1391.0,
+                    1260.0,
+                    1391.0,
+                    1285.0,
+                    1338.0,
+                    1285.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1225.0,
+                    1266.0,
+                    1302.0,
+                    1266.0,
+                    1302.0,
+                    1297.0,
+                    1225.0,
+                    1297.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1090.0,
+                    1279.0,
+                    1171.0,
+                    1279.0,
+                    1171.0,
+                    1313.0,
+                    1090.0,
+                    1313.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1355.0,
+                    1288.0,
+                    1363.0,
+                    1288.0,
+                    1363.0,
+                    1295.0,
+                    1355.0,
+                    1295.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1347.0,
+                    1296.0,
+                    1371.0,
+                    1296.0,
+                    1371.0,
+                    1316.0,
+                    1347.0,
+                    1316.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1115.0,
+                    1315.0,
+                    1130.0,
+                    1315.0,
+                    1130.0,
+                    1328.0,
+                    1115.0,
+                    1328.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    949.0,
+                    1327.0,
+                    983.0,
+                    1327.0,
+                    983.0,
+                    1355.0,
+                    949.0,
+                    1355.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1025.0,
+                    1337.0,
+                    1083.0,
+                    1337.0,
+                    1083.0,
+                    1361.0,
+                    1025.0,
+                    1361.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1247.0,
+                    1323.0,
+                    1299.0,
+                    1323.0,
+                    1299.0,
+                    1351.0,
+                    1247.0,
+                    1351.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    980.0,
+                    1355.0,
+                    1042.0,
+                    1355.0,
+                    1042.0,
+                    1383.0,
+                    980.0,
+                    1383.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1096.0,
+                    1355.0,
+                    1157.0,
+                    1355.0,
+                    1157.0,
+                    1379.0,
+                    1096.0,
+                    1379.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1001.0,
+                    1377.0,
+                    1056.0,
+                    1377.0,
+                    1056.0,
+                    1404.0,
+                    1001.0,
+                    1404.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1111.0,
+                    1387.0,
+                    1136.0,
+                    1387.0,
+                    1136.0,
+                    1427.0,
+                    1111.0,
+                    1427.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    948.0,
+                    1398.0,
+                    983.0,
+                    1398.0,
+                    983.0,
+                    1426.0,
+                    948.0,
+                    1426.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1041.0,
+                    1423.0,
+                    1121.0,
+                    1423.0,
+                    1121.0,
+                    1456.0,
+                    1041.0,
+                    1456.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1271.0,
+                    1428.0,
+                    1345.0,
+                    1428.0,
+                    1345.0,
+                    1460.0,
+                    1271.0,
+                    1460.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1009.0,
+                    1453.0,
+                    1085.0,
+                    1453.0,
+                    1085.0,
+                    1484.0,
+                    1009.0,
+                    1484.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    949.0,
+                    1468.0,
+                    983.0,
+                    1468.0,
+                    983.0,
+                    1496.0,
+                    949.0,
+                    1496.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1228.0,
+                    1472.0,
+                    1249.0,
+                    1472.0,
+                    1249.0,
+                    1495.0,
+                    1228.0,
+                    1495.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1259.0,
+                    1466.0,
+                    1369.0,
+                    1466.0,
+                    1369.0,
+                    1498.0,
+                    1259.0,
+                    1498.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1025.0,
+                    1482.0,
+                    1118.0,
+                    1482.0,
+                    1118.0,
+                    1516.0,
+                    1025.0,
+                    1516.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1015.0,
+                    1483.0,
+                    1023.0,
+                    1483.0,
+                    1023.0,
+                    1495.0,
+                    1015.0,
+                    1495.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    989.0,
+                    1518.0,
+                    1012.0,
+                    1518.0,
+                    1012.0,
+                    1546.0,
+                    989.0,
+                    1546.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1068.0,
+                    1515.0,
+                    1108.0,
+                    1515.0,
+                    1108.0,
+                    1546.0,
+                    1068.0,
+                    1546.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1160.0,
+                    1515.0,
+                    1197.0,
+                    1515.0,
+                    1197.0,
+                    1545.0,
+                    1160.0,
+                    1545.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1249.0,
+                    1515.0,
+                    1288.0,
+                    1515.0,
+                    1288.0,
+                    1546.0,
+                    1249.0,
+                    1546.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1336.0,
+                    1515.0,
+                    1376.0,
+                    1515.0,
+                    1376.0,
+                    1546.0,
+                    1336.0,
+                    1546.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1020.0,
+                    1538.0,
+                    1365.0,
+                    1538.0,
+                    1365.0,
+                    1570.0,
+                    1020.0,
+                    1570.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1279.0,
+                    1366.5,
+                    1309.0,
+                    1366.5,
+                    1309.0,
+                    1384.5,
+                    1279.0,
+                    1384.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    987.25,
+                    1456.0,
+                    1023.25,
+                    1456.0,
+                    1023.25,
+                    1477.0,
+                    987.25,
+                    1477.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1015.0,
+                    1492.5,
+                    1025.0,
+                    1492.5,
+                    1025.0,
+                    1515.5,
+                    1015.0,
+                    1515.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    910.0,
+                    1582.0,
+                    1408.0,
+                    1582.0,
+                    1408.0,
+                    1617.0,
+                    910.0,
+                    1617.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    909.0,
+                    1608.0,
+                    1410.0,
+                    1608.0,
+                    1410.0,
+                    1647.0,
+                    909.0,
+                    1647.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    911.0,
+                    1639.0,
+                    1405.0,
+                    1639.0,
+                    1405.0,
+                    1671.0,
+                    911.0,
+                    1671.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    909.0,
+                    1666.0,
+                    1408.0,
+                    1666.0,
+                    1408.0,
+                    1700.0,
+                    909.0,
+                    1700.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    909.0,
+                    1697.0,
+                    1408.0,
+                    1697.0,
+                    1408.0,
+                    1722.0,
+                    909.0,
+                    1722.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    911.0,
+                    1724.0,
+                    1409.0,
+                    1724.0,
+                    1409.0,
+                    1754.0,
+                    911.0,
+                    1754.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    910.0,
+                    1751.0,
+                    1406.0,
+                    1751.0,
+                    1406.0,
+                    1780.0,
+                    910.0,
+                    1780.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    912.0,
+                    1781.0,
+                    1404.0,
+                    1781.0,
+                    1404.0,
+                    1809.0,
+                    912.0,
+                    1809.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    910.0,
+                    1810.0,
+                    1404.0,
+                    1810.0,
+                    1404.0,
+                    1838.0,
+                    910.0,
+                    1838.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    911.0,
+                    1838.0,
+                    1252.0,
+                    1838.0,
+                    1252.0,
+                    1866.0,
+                    911.0,
+                    1866.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    220.0,
+                    1248.0,
+                    220.0,
+                    1248.0,
+                    280.0,
+                    296.0,
+                    280.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    278.0,
+                    842.0,
+                    278.0,
+                    842.0,
+                    334.0,
+                    295.0,
+                    334.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    769.0,
+                    602.0,
+                    932.0,
+                    602.0,
+                    932.0,
+                    639.0,
+                    769.0,
+                    639.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    73.0,
+                    815.0,
+                    73.0,
+                    815.0,
+                    108.0,
+                    295.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1122.0,
+                    576.0,
+                    1122.0,
+                    576.0,
+                    1171.0,
+                    292.0,
+                    1171.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    840.0,
+                    2088.0,
+                    857.0,
+                    2088.0,
+                    857.0,
+                    2116.0,
+                    840.0,
+                    2116.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    395.0,
+                    674.0,
+                    1302.0,
+                    674.0,
+                    1302.0,
+                    706.0,
+                    395.0,
+                    706.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    394.0,
+                    704.0,
+                    1305.0,
+                    704.0,
+                    1305.0,
+                    739.0,
+                    394.0,
+                    739.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    391.0,
+                    729.0,
+                    1308.0,
+                    729.0,
+                    1308.0,
+                    772.0,
+                    391.0,
+                    772.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    391.0,
+                    760.0,
+                    1309.0,
+                    760.0,
+                    1309.0,
+                    802.0,
+                    391.0,
+                    802.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    392.0,
+                    795.0,
+                    1306.0,
+                    795.0,
+                    1306.0,
+                    829.0,
+                    392.0,
+                    829.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    393.0,
+                    827.0,
+                    1305.0,
+                    827.0,
+                    1305.0,
+                    859.0,
+                    393.0,
+                    859.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    392.0,
+                    855.0,
+                    1308.0,
+                    855.0,
+                    1308.0,
+                    888.0,
+                    392.0,
+                    888.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    393.0,
+                    887.0,
+                    1306.0,
+                    887.0,
+                    1306.0,
+                    919.0,
+                    393.0,
+                    919.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    395.0,
+                    919.0,
+                    1304.0,
+                    919.0,
+                    1304.0,
+                    948.0,
+                    395.0,
+                    948.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    393.0,
+                    945.0,
+                    1305.0,
+                    945.0,
+                    1305.0,
+                    980.0,
+                    393.0,
+                    980.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    391.0,
+                    973.0,
+                    1307.0,
+                    973.0,
+                    1307.0,
+                    1016.0,
+                    391.0,
+                    1016.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    394.0,
+                    1006.0,
+                    455.0,
+                    1006.0,
+                    455.0,
+                    1038.0,
+                    394.0,
+                    1038.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    531.0,
+                    1006.0,
+                    560.0,
+                    1006.0,
+                    560.0,
+                    1038.0,
+                    531.0,
+                    1038.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    637.0,
+                    1006.0,
+                    958.0,
+                    1006.0,
+                    958.0,
+                    1038.0,
+                    637.0,
+                    1038.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1034.0,
+                    1006.0,
+                    1063.0,
+                    1006.0,
+                    1063.0,
+                    1038.0,
+                    1034.0,
+                    1038.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1139.0,
+                    1006.0,
+                    1307.0,
+                    1006.0,
+                    1307.0,
+                    1038.0,
+                    1139.0,
+                    1038.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    395.0,
+                    1039.0,
+                    828.0,
+                    1039.0,
+                    828.0,
+                    1072.0,
+                    395.0,
+                    1072.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    1194.0,
+                    894.0,
+                    1194.0,
+                    894.0,
+                    1228.0,
+                    297.0,
+                    1228.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1225.0,
+                    893.0,
+                    1225.0,
+                    893.0,
+                    1259.0,
+                    295.0,
+                    1259.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1255.0,
+                    892.0,
+                    1255.0,
+                    892.0,
+                    1288.0,
+                    294.0,
+                    1288.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1285.0,
+                    894.0,
+                    1285.0,
+                    894.0,
+                    1320.0,
+                    295.0,
+                    1320.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1316.0,
+                    891.0,
+                    1316.0,
+                    891.0,
+                    1349.0,
+                    294.0,
+                    1349.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1348.0,
+                    893.0,
+                    1348.0,
+                    893.0,
+                    1382.0,
+                    294.0,
+                    1382.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1378.0,
+                    895.0,
+                    1378.0,
+                    895.0,
+                    1410.0,
+                    296.0,
+                    1410.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1408.0,
+                    892.0,
+                    1408.0,
+                    892.0,
+                    1439.0,
+                    295.0,
+                    1439.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1436.0,
+                    896.0,
+                    1436.0,
+                    896.0,
+                    1472.0,
+                    293.0,
+                    1472.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1467.0,
+                    895.0,
+                    1467.0,
+                    895.0,
+                    1501.0,
+                    294.0,
+                    1501.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1499.0,
+                    894.0,
+                    1499.0,
+                    894.0,
+                    1532.0,
+                    294.0,
+                    1532.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1530.0,
+                    896.0,
+                    1530.0,
+                    896.0,
+                    1561.0,
+                    295.0,
+                    1561.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1560.0,
+                    895.0,
+                    1560.0,
+                    895.0,
+                    1594.0,
+                    295.0,
+                    1594.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1588.0,
+                    896.0,
+                    1588.0,
+                    896.0,
+                    1622.0,
+                    295.0,
+                    1622.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1620.0,
+                    895.0,
+                    1620.0,
+                    895.0,
+                    1651.0,
+                    295.0,
+                    1651.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1650.0,
+                    894.0,
+                    1650.0,
+                    894.0,
+                    1683.0,
+                    296.0,
+                    1683.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1681.0,
+                    897.0,
+                    1681.0,
+                    897.0,
+                    1716.0,
+                    295.0,
+                    1716.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1714.0,
+                    892.0,
+                    1714.0,
+                    892.0,
+                    1745.0,
+                    296.0,
+                    1745.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1744.0,
+                    895.0,
+                    1744.0,
+                    895.0,
+                    1775.0,
+                    296.0,
+                    1775.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1772.0,
+                    356.0,
+                    1772.0,
+                    356.0,
+                    1806.0,
+                    294.0,
+                    1806.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1940.0,
+                    1405.0,
+                    1940.0,
+                    1405.0,
+                    1977.0,
+                    294.0,
+                    1977.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1972.0,
+                    1131.0,
+                    1972.0,
+                    1131.0,
+                    2004.0,
+                    294.0,
+                    2004.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1210.0,
+                    1972.0,
+                    1407.0,
+                    1972.0,
+                    1407.0,
+                    2004.0,
+                    1210.0,
+                    2004.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    2000.0,
+                    1405.0,
+                    2000.0,
+                    1405.0,
+                    2040.0,
+                    293.0,
+                    2040.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    535.0,
+                    410.0,
+                    831.0,
+                    410.0,
+                    831.0,
+                    450.0,
+                    535.0,
+                    450.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    942.0,
+                    411.0,
+                    1211.0,
+                    411.0,
+                    1211.0,
+                    450.0,
+                    942.0,
+                    450.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    618.0,
+                    448.0,
+                    1057.0,
+                    448.0,
+                    1057.0,
+                    479.0,
+                    618.0,
+                    479.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    442.0,
+                    478.0,
+                    1226.0,
+                    478.0,
+                    1226.0,
+                    509.0,
+                    442.0,
+                    509.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1820.0,
+                    893.0,
+                    1820.0,
+                    893.0,
+                    1853.0,
+                    294.0,
+                    1853.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1849.0,
+                    895.0,
+                    1849.0,
+                    895.0,
+                    1883.0,
+                    295.0,
+                    1883.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1880.0,
+                    894.0,
+                    1880.0,
+                    894.0,
+                    1914.0,
+                    295.0,
+                    1914.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1910.0,
+                    893.0,
+                    1910.0,
+                    893.0,
+                    1946.0,
+                    294.0,
+                    1946.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    318.0,
+                    377.0,
+                    1391.0,
+                    377.0,
+                    1391.0,
+                    421.0,
+                    318.0,
+                    421.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 0,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    367,
+                    1405,
+                    367,
+                    1405,
+                    795,
+                    298,
+                    795
+                ],
+                "score": 0.983
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1637,
+                    1404,
+                    1637,
+                    1404,
+                    2034,
+                    298,
+                    2034
+                ],
+                "score": 0.982
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    980,
+                    1404,
+                    980,
+                    1404,
+                    1163,
+                    298,
+                    1163
+                ],
+                "score": 0.977
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    809,
+                    1404,
+                    809,
+                    1404,
+                    965,
+                    299,
+                    965
+                ],
+                "score": 0.975
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    229,
+                    1402,
+                    229,
+                    1402,
+                    352,
+                    299,
+                    352
+                ],
+                "score": 0.972
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    368,
+                    1237,
+                    1405,
+                    1237,
+                    1405,
+                    1523,
+                    368,
+                    1523
+                ],
+                "score": 0.965
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1179,
+                    843,
+                    1179,
+                    843,
+                    1211,
+                    297,
+                    1211
+                ],
+                "score": 0.924
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    300,
+                    1565,
+                    587,
+                    1565,
+                    587,
+                    1601,
+                    300,
+                    1601
+                ],
+                "score": 0.912
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    300,
+                    76,
+                    813,
+                    76,
+                    813,
+                    104,
+                    300,
+                    104
+                ],
+                "score": 0.896
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    841,
+                    2088,
+                    858,
+                    2088,
+                    858,
+                    2112,
+                    841,
+                    2112
+                ],
+                "score": 0.706
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    841,
+                    2088,
+                    859,
+                    2088,
+                    859,
+                    2112,
+                    841,
+                    2112
+                ],
+                "score": 0.146
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    296,
+                    230,
+                    371,
+                    230,
+                    371,
+                    261,
+                    296,
+                    261
+                ],
+                "score": 0.9,
+                "latex": "7 5 . 2 \\%"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    763,
+                    1072,
+                    840,
+                    1072,
+                    840,
+                    1101,
+                    763,
+                    1101
+                ],
+                "score": 0.88,
+                "latex": "4 2 . 2 \\%"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1007,
+                    231,
+                    1082,
+                    231,
+                    1082,
+                    260,
+                    1007,
+                    260
+                ],
+                "score": 0.87,
+                "latex": "4 2 . 2 \\%"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    875,
+                    1072,
+                    953,
+                    1072,
+                    953,
+                    1102,
+                    875,
+                    1102
+                ],
+                "score": 0.87,
+                "latex": "6 7 . 6 \\%"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1103,
+                    1072,
+                    1160,
+                    1072,
+                    1160,
+                    1101,
+                    1103,
+                    1101
+                ],
+                "score": 0.86,
+                "latex": "2 5 \\%"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    358,
+                    1103,
+                    432,
+                    1103,
+                    432,
+                    1132,
+                    358,
+                    1132
+                ],
+                "score": 0.86,
+                "latex": "3 6 . 3 \\%"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    464,
+                    1103,
+                    539,
+                    1103,
+                    539,
+                    1132,
+                    464,
+                    1132
+                ],
+                "score": 0.85,
+                "latex": "6 8 . 2 \\%"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    497,
+                    262,
+                    573,
+                    262,
+                    573,
+                    292,
+                    497,
+                    292
+                ],
+                "score": 0.85,
+                "latex": "7 7 . 1 \\%"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    678,
+                    1103,
+                    733,
+                    1103,
+                    733,
+                    1132,
+                    678,
+                    1132
+                ],
+                "score": 0.84,
+                "latex": "3 1 \\%"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1562.0,
+                    594.0,
+                    1562.0,
+                    594.0,
+                    1609.0,
+                    291.0,
+                    1609.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    72.0,
+                    816.0,
+                    72.0,
+                    816.0,
+                    110.0,
+                    294.0,
+                    110.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    839.0,
+                    2085.0,
+                    862.0,
+                    2085.0,
+                    862.0,
+                    2120.0,
+                    839.0,
+                    2120.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    838.0,
+                    2085.0,
+                    862.0,
+                    2085.0,
+                    862.0,
+                    2120.0,
+                    838.0,
+                    2120.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    367.0,
+                    1407.0,
+                    367.0,
+                    1407.0,
+                    402.0,
+                    293.0,
+                    402.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    397.0,
+                    1406.0,
+                    397.0,
+                    1406.0,
+                    436.0,
+                    293.0,
+                    436.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    429.0,
+                    1405.0,
+                    429.0,
+                    1405.0,
+                    463.0,
+                    292.0,
+                    463.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    459.0,
+                    1405.0,
+                    459.0,
+                    1405.0,
+                    495.0,
+                    293.0,
+                    495.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    490.0,
+                    1406.0,
+                    490.0,
+                    1406.0,
+                    524.0,
+                    292.0,
+                    524.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    521.0,
+                    1402.0,
+                    521.0,
+                    1402.0,
+                    553.0,
+                    296.0,
+                    553.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    548.0,
+                    1406.0,
+                    548.0,
+                    1406.0,
+                    586.0,
+                    293.0,
+                    586.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    580.0,
+                    1406.0,
+                    580.0,
+                    1406.0,
+                    618.0,
+                    293.0,
+                    618.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    614.0,
+                    1402.0,
+                    614.0,
+                    1402.0,
+                    646.0,
+                    296.0,
+                    646.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    641.0,
+                    1405.0,
+                    641.0,
+                    1405.0,
+                    677.0,
+                    292.0,
+                    677.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    671.0,
+                    1406.0,
+                    671.0,
+                    1406.0,
+                    708.0,
+                    293.0,
+                    708.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    704.0,
+                    1403.0,
+                    704.0,
+                    1403.0,
+                    736.0,
+                    296.0,
+                    736.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    736.0,
+                    1403.0,
+                    736.0,
+                    1403.0,
+                    768.0,
+                    296.0,
+                    768.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    764.0,
+                    1026.0,
+                    764.0,
+                    1026.0,
+                    799.0,
+                    295.0,
+                    799.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1635.0,
+                    1405.0,
+                    1635.0,
+                    1405.0,
+                    1672.0,
+                    293.0,
+                    1672.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1667.0,
+                    1403.0,
+                    1667.0,
+                    1403.0,
+                    1702.0,
+                    293.0,
+                    1702.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1697.0,
+                    1404.0,
+                    1697.0,
+                    1404.0,
+                    1732.0,
+                    292.0,
+                    1732.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1727.0,
+                    1404.0,
+                    1727.0,
+                    1404.0,
+                    1763.0,
+                    294.0,
+                    1763.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1757.0,
+                    1406.0,
+                    1757.0,
+                    1406.0,
+                    1797.0,
+                    293.0,
+                    1797.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1789.0,
+                    1405.0,
+                    1789.0,
+                    1405.0,
+                    1824.0,
+                    293.0,
+                    1824.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1820.0,
+                    1404.0,
+                    1820.0,
+                    1404.0,
+                    1855.0,
+                    293.0,
+                    1855.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1848.0,
+                    1404.0,
+                    1848.0,
+                    1404.0,
+                    1887.0,
+                    292.0,
+                    1887.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1883.0,
+                    1404.0,
+                    1883.0,
+                    1404.0,
+                    1916.0,
+                    294.0,
+                    1916.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1908.0,
+                    1404.0,
+                    1908.0,
+                    1404.0,
+                    1948.0,
+                    293.0,
+                    1948.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1940.0,
+                    1405.0,
+                    1940.0,
+                    1405.0,
+                    1979.0,
+                    293.0,
+                    1979.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1969.0,
+                    1405.0,
+                    1969.0,
+                    1405.0,
+                    2012.0,
+                    292.0,
+                    2012.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    2000.0,
+                    1405.0,
+                    2000.0,
+                    1405.0,
+                    2039.0,
+                    293.0,
+                    2039.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    979.0,
+                    1404.0,
+                    979.0,
+                    1404.0,
+                    1013.0,
+                    295.0,
+                    1013.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1008.0,
+                    1405.0,
+                    1008.0,
+                    1405.0,
+                    1045.0,
+                    293.0,
+                    1045.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1041.0,
+                    1406.0,
+                    1041.0,
+                    1406.0,
+                    1077.0,
+                    294.0,
+                    1077.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1069.0,
+                    762.0,
+                    1069.0,
+                    762.0,
+                    1108.0,
+                    292.0,
+                    1108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    841.0,
+                    1069.0,
+                    874.0,
+                    1069.0,
+                    874.0,
+                    1108.0,
+                    841.0,
+                    1108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    954.0,
+                    1069.0,
+                    1102.0,
+                    1069.0,
+                    1102.0,
+                    1108.0,
+                    954.0,
+                    1108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1161.0,
+                    1069.0,
+                    1405.0,
+                    1069.0,
+                    1405.0,
+                    1108.0,
+                    1161.0,
+                    1108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1102.0,
+                    357.0,
+                    1102.0,
+                    357.0,
+                    1137.0,
+                    293.0,
+                    1137.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    433.0,
+                    1102.0,
+                    463.0,
+                    1102.0,
+                    463.0,
+                    1137.0,
+                    433.0,
+                    1137.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    540.0,
+                    1102.0,
+                    677.0,
+                    1102.0,
+                    677.0,
+                    1137.0,
+                    540.0,
+                    1137.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    734.0,
+                    1102.0,
+                    1406.0,
+                    1102.0,
+                    1406.0,
+                    1137.0,
+                    734.0,
+                    1137.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1133.0,
+                    414.0,
+                    1133.0,
+                    414.0,
+                    1162.0,
+                    296.0,
+                    1162.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    811.0,
+                    1408.0,
+                    811.0,
+                    1408.0,
+                    844.0,
+                    295.0,
+                    844.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    842.0,
+                    1408.0,
+                    842.0,
+                    1408.0,
+                    876.0,
+                    294.0,
+                    876.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    870.0,
+                    1408.0,
+                    870.0,
+                    1408.0,
+                    908.0,
+                    293.0,
+                    908.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    899.0,
+                    1405.0,
+                    899.0,
+                    1405.0,
+                    940.0,
+                    294.0,
+                    940.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    932.0,
+                    1205.0,
+                    932.0,
+                    1205.0,
+                    968.0,
+                    293.0,
+                    968.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    372.0,
+                    230.0,
+                    1006.0,
+                    230.0,
+                    1006.0,
+                    263.0,
+                    372.0,
+                    263.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1083.0,
+                    230.0,
+                    1404.0,
+                    230.0,
+                    1404.0,
+                    263.0,
+                    1083.0,
+                    263.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    261.0,
+                    496.0,
+                    261.0,
+                    496.0,
+                    294.0,
+                    295.0,
+                    294.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    574.0,
+                    261.0,
+                    1404.0,
+                    261.0,
+                    1404.0,
+                    294.0,
+                    574.0,
+                    294.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    289.0,
+                    1406.0,
+                    289.0,
+                    1406.0,
+                    328.0,
+                    292.0,
+                    328.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    323.0,
+                    477.0,
+                    323.0,
+                    477.0,
+                    353.0,
+                    295.0,
+                    353.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    367.0,
+                    1236.0,
+                    1404.0,
+                    1236.0,
+                    1404.0,
+                    1274.0,
+                    367.0,
+                    1274.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    394.0,
+                    1270.0,
+                    557.0,
+                    1270.0,
+                    557.0,
+                    1299.0,
+                    394.0,
+                    1299.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    386.0,
+                    1309.0,
+                    1405.0,
+                    1309.0,
+                    1405.0,
+                    1346.0,
+                    386.0,
+                    1346.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    395.0,
+                    1343.0,
+                    1008.0,
+                    1343.0,
+                    1008.0,
+                    1376.0,
+                    395.0,
+                    1376.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    390.0,
+                    1384.0,
+                    1408.0,
+                    1384.0,
+                    1408.0,
+                    1421.0,
+                    390.0,
+                    1421.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    394.0,
+                    1417.0,
+                    822.0,
+                    1417.0,
+                    822.0,
+                    1448.0,
+                    394.0,
+                    1448.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    364.0,
+                    1455.0,
+                    1404.0,
+                    1455.0,
+                    1404.0,
+                    1496.0,
+                    364.0,
+                    1496.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    393.0,
+                    1486.0,
+                    665.0,
+                    1486.0,
+                    665.0,
+                    1529.0,
+                    393.0,
+                    1529.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1176.0,
+                    848.0,
+                    1176.0,
+                    848.0,
+                    1214.0,
+                    295.0,
+                    1214.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 1,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    366,
+                    1405,
+                    366,
+                    1405,
+                    1069,
+                    297,
+                    1069
+                ],
+                "score": 0.984
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1448,
+                    1404,
+                    1448,
+                    1404,
+                    1787,
+                    298,
+                    1787
+                ],
+                "score": 0.984
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    1879,
+                    1402,
+                    1879,
+                    1402,
+                    2034,
+                    299,
+                    2034
+                ],
+                "score": 0.975
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    300,
+                    1239,
+                    1403,
+                    1239,
+                    1403,
+                    1332,
+                    300,
+                    1332
+                ],
+                "score": 0.975
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    229,
+                    1402,
+                    229,
+                    1402,
+                    352,
+                    299,
+                    352
+                ],
+                "score": 0.971
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    536,
+                    1349,
+                    1159,
+                    1349,
+                    1159,
+                    1437,
+                    536,
+                    1437
+                ],
+                "score": 0.95
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    300,
+                    1182,
+                    918,
+                    1182,
+                    918,
+                    1214,
+                    300,
+                    1214
+                ],
+                "score": 0.918
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    299,
+                    1114,
+                    481,
+                    1114,
+                    481,
+                    1149,
+                    299,
+                    1149
+                ],
+                "score": 0.913
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    300,
+                    76,
+                    812,
+                    76,
+                    812,
+                    104,
+                    300,
+                    104
+                ],
+                "score": 0.896
+            },
+            {
+                "category_id": 9,
+                "poly": [
+                    1366,
+                    1378,
+                    1399,
+                    1378,
+                    1399,
+                    1408,
+                    1366,
+                    1408
+                ],
+                "score": 0.882
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    300,
+                    1822,
+                    1135,
+                    1822,
+                    1135,
+                    1856,
+                    300,
+                    1856
+                ],
+                "score": 0.856
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    841,
+                    2088,
+                    858,
+                    2088,
+                    858,
+                    2112,
+                    841,
+                    2112
+                ],
+                "score": 0.604
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    841,
+                    2088,
+                    859,
+                    2088,
+                    859,
+                    2112,
+                    841,
+                    2112
+                ],
+                "score": 0.415
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    538,
+                    1344,
+                    1158,
+                    1344,
+                    1158,
+                    1439,
+                    538,
+                    1439
+                ],
+                "score": 0.94,
+                "latex": "\\hat { \\theta } _ { S } = \\underset { \\theta _ { S } } { \\arg \\operatorname* { m i n } } \\sum _ { i } ^ { N } \\mathcal { L } _ { \\mathrm { s u p } } ( \\mathbf { x } _ { i } , \\theta _ { S } , y _ { i } ) + \\mathcal { L } _ { \\mathrm { d i s t i l l } } ( \\mathbf { x } _ { i } , \\theta _ { S } , \\theta _ { T } ) ,"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    722,
+                    1604,
+                    790,
+                    1604,
+                    790,
+                    1635,
+                    722,
+                    1635
+                ],
+                "score": 0.92,
+                "latex": "{ \\mathcal { L } } _ { \\mathrm { d i s t i l l } }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    993,
+                    1482,
+                    1045,
+                    1482,
+                    1045,
+                    1515,
+                    993,
+                    1515
+                ],
+                "score": 0.9,
+                "latex": "\\mathcal { L } _ { \\mathrm { s u p } }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    454,
+                    1482,
+                    488,
+                    1482,
+                    488,
+                    1512,
+                    454,
+                    1512
+                ],
+                "score": 0.88,
+                "latex": "\\theta _ { T }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    962,
+                    1451,
+                    995,
+                    1451,
+                    995,
+                    1481,
+                    962,
+                    1481
+                ],
+                "score": 0.88,
+                "latex": "\\theta _ { S }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    373,
+                    1455,
+                    403,
+                    1455,
+                    403,
+                    1481,
+                    373,
+                    1481
+                ],
+                "score": 0.84,
+                "latex": "\\mathbf { x } _ { i }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    555,
+                    1455,
+                    582,
+                    1455,
+                    582,
+                    1482,
+                    555,
+                    1482
+                ],
+                "score": 0.84,
+                "latex": "y _ { i }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    546,
+                    1726,
+                    574,
+                    1726,
+                    574,
+                    1753,
+                    546,
+                    1753
+                ],
+                "score": 0.31,
+                "latex": "l 2"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1180.0,
+                    921.0,
+                    1180.0,
+                    921.0,
+                    1218.0,
+                    294.0,
+                    1218.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    290.0,
+                    1108.0,
+                    487.0,
+                    1108.0,
+                    487.0,
+                    1158.0,
+                    290.0,
+                    1158.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    73.0,
+                    815.0,
+                    73.0,
+                    815.0,
+                    108.0,
+                    295.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1821.0,
+                    1142.0,
+                    1821.0,
+                    1142.0,
+                    1860.0,
+                    293.0,
+                    1860.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    839.0,
+                    2085.0,
+                    860.0,
+                    2085.0,
+                    860.0,
+                    2116.0,
+                    839.0,
+                    2116.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    838.0,
+                    2085.0,
+                    862.0,
+                    2085.0,
+                    862.0,
+                    2118.0,
+                    838.0,
+                    2118.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    367.0,
+                    1405.0,
+                    367.0,
+                    1405.0,
+                    403.0,
+                    296.0,
+                    403.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    399.0,
+                    1405.0,
+                    399.0,
+                    1405.0,
+                    434.0,
+                    295.0,
+                    434.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    430.0,
+                    1403.0,
+                    430.0,
+                    1403.0,
+                    462.0,
+                    296.0,
+                    462.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    459.0,
+                    1407.0,
+                    459.0,
+                    1407.0,
+                    493.0,
+                    295.0,
+                    493.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    489.0,
+                    1406.0,
+                    489.0,
+                    1406.0,
+                    524.0,
+                    295.0,
+                    524.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    519.0,
+                    1406.0,
+                    519.0,
+                    1406.0,
+                    556.0,
+                    292.0,
+                    556.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    552.0,
+                    1405.0,
+                    552.0,
+                    1405.0,
+                    583.0,
+                    295.0,
+                    583.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    583.0,
+                    1403.0,
+                    583.0,
+                    1403.0,
+                    615.0,
+                    296.0,
+                    615.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    611.0,
+                    1403.0,
+                    611.0,
+                    1403.0,
+                    646.0,
+                    294.0,
+                    646.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    642.0,
+                    1407.0,
+                    642.0,
+                    1407.0,
+                    677.0,
+                    295.0,
+                    677.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    671.0,
+                    1405.0,
+                    671.0,
+                    1405.0,
+                    707.0,
+                    294.0,
+                    707.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    704.0,
+                    1406.0,
+                    704.0,
+                    1406.0,
+                    739.0,
+                    294.0,
+                    739.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    733.0,
+                    1406.0,
+                    733.0,
+                    1406.0,
+                    768.0,
+                    295.0,
+                    768.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    764.0,
+                    1402.0,
+                    764.0,
+                    1402.0,
+                    795.0,
+                    296.0,
+                    795.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    793.0,
+                    1406.0,
+                    793.0,
+                    1406.0,
+                    829.0,
+                    294.0,
+                    829.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    825.0,
+                    1403.0,
+                    825.0,
+                    1403.0,
+                    861.0,
+                    294.0,
+                    861.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    853.0,
+                    1406.0,
+                    853.0,
+                    1406.0,
+                    891.0,
+                    294.0,
+                    891.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    887.0,
+                    1405.0,
+                    887.0,
+                    1405.0,
+                    918.0,
+                    294.0,
+                    918.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    916.0,
+                    1405.0,
+                    916.0,
+                    1405.0,
+                    951.0,
+                    292.0,
+                    951.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    946.0,
+                    1405.0,
+                    946.0,
+                    1405.0,
+                    981.0,
+                    295.0,
+                    981.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    975.0,
+                    1405.0,
+                    975.0,
+                    1405.0,
+                    1012.0,
+                    292.0,
+                    1012.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1009.0,
+                    1402.0,
+                    1009.0,
+                    1402.0,
+                    1040.0,
+                    296.0,
+                    1040.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1039.0,
+                    442.0,
+                    1039.0,
+                    442.0,
+                    1071.0,
+                    295.0,
+                    1071.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1448.0,
+                    372.0,
+                    1448.0,
+                    372.0,
+                    1487.0,
+                    293.0,
+                    1487.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    404.0,
+                    1448.0,
+                    554.0,
+                    1448.0,
+                    554.0,
+                    1487.0,
+                    404.0,
+                    1487.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    583.0,
+                    1448.0,
+                    961.0,
+                    1448.0,
+                    961.0,
+                    1487.0,
+                    583.0,
+                    1487.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    996.0,
+                    1448.0,
+                    1406.0,
+                    1448.0,
+                    1406.0,
+                    1487.0,
+                    996.0,
+                    1487.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1479.0,
+                    453.0,
+                    1479.0,
+                    453.0,
+                    1518.0,
+                    292.0,
+                    1518.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    489.0,
+                    1479.0,
+                    992.0,
+                    1479.0,
+                    992.0,
+                    1518.0,
+                    489.0,
+                    1518.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1046.0,
+                    1479.0,
+                    1406.0,
+                    1479.0,
+                    1406.0,
+                    1518.0,
+                    1046.0,
+                    1518.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1510.0,
+                    1405.0,
+                    1510.0,
+                    1405.0,
+                    1545.0,
+                    294.0,
+                    1545.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1539.0,
+                    1406.0,
+                    1539.0,
+                    1406.0,
+                    1576.0,
+                    293.0,
+                    1576.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1569.0,
+                    1405.0,
+                    1569.0,
+                    1405.0,
+                    1609.0,
+                    291.0,
+                    1609.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1603.0,
+                    721.0,
+                    1603.0,
+                    721.0,
+                    1636.0,
+                    293.0,
+                    1636.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    791.0,
+                    1603.0,
+                    1406.0,
+                    1603.0,
+                    1406.0,
+                    1636.0,
+                    791.0,
+                    1636.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1632.0,
+                    1405.0,
+                    1632.0,
+                    1405.0,
+                    1666.0,
+                    293.0,
+                    1666.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1662.0,
+                    1406.0,
+                    1662.0,
+                    1406.0,
+                    1697.0,
+                    294.0,
+                    1697.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1694.0,
+                    1405.0,
+                    1694.0,
+                    1405.0,
+                    1728.0,
+                    294.0,
+                    1728.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1726.0,
+                    545.0,
+                    1726.0,
+                    545.0,
+                    1757.0,
+                    296.0,
+                    1757.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    575.0,
+                    1726.0,
+                    1404.0,
+                    1726.0,
+                    1404.0,
+                    1757.0,
+                    575.0,
+                    1757.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    1757.0,
+                    1362.0,
+                    1757.0,
+                    1362.0,
+                    1788.0,
+                    297.0,
+                    1788.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1878.0,
+                    1407.0,
+                    1878.0,
+                    1407.0,
+                    1917.0,
+                    294.0,
+                    1917.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1910.0,
+                    1403.0,
+                    1910.0,
+                    1403.0,
+                    1943.0,
+                    295.0,
+                    1943.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    1943.0,
+                    1404.0,
+                    1943.0,
+                    1404.0,
+                    1976.0,
+                    297.0,
+                    1976.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1972.0,
+                    1403.0,
+                    1972.0,
+                    1403.0,
+                    2006.0,
+                    295.0,
+                    2006.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1999.0,
+                    1406.0,
+                    1999.0,
+                    1406.0,
+                    2039.0,
+                    293.0,
+                    2039.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1238.0,
+                    1405.0,
+                    1238.0,
+                    1405.0,
+                    1272.0,
+                    296.0,
+                    1272.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1270.0,
+                    1404.0,
+                    1270.0,
+                    1404.0,
+                    1304.0,
+                    295.0,
+                    1304.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1301.0,
+                    762.0,
+                    1301.0,
+                    762.0,
+                    1336.0,
+                    292.0,
+                    1336.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    228.0,
+                    1406.0,
+                    228.0,
+                    1406.0,
+                    265.0,
+                    293.0,
+                    265.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    261.0,
+                    1406.0,
+                    261.0,
+                    1406.0,
+                    294.0,
+                    295.0,
+                    294.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    289.0,
+                    1403.0,
+                    289.0,
+                    1403.0,
+                    328.0,
+                    293.0,
+                    328.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    319.0,
+                    460.0,
+                    319.0,
+                    460.0,
+                    357.0,
+                    295.0,
+                    357.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 2,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    296,
+                    835,
+                    1405,
+                    835,
+                    1405,
+                    1210,
+                    296,
+                    1210
+                ],
+                "score": 0.985
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    296,
+                    665,
+                    1404,
+                    665,
+                    1404,
+                    821,
+                    296,
+                    821
+                ],
+                "score": 0.98
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    1669,
+                    1402,
+                    1669,
+                    1402,
+                    1763,
+                    299,
+                    1763
+                ],
+                "score": 0.974
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    384,
+                    1773,
+                    1314,
+                    1773,
+                    1314,
+                    1962,
+                    384,
+                    1962
+                ],
+                "score": 0.971
+            },
+            {
+                "category_id": 3,
+                "poly": [
+                    311,
+                    226,
+                    1404,
+                    226,
+                    1404,
+                    503,
+                    311,
+                    503
+                ],
+                "score": 0.966
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1388,
+                    1401,
+                    1388,
+                    1401,
+                    1453,
+                    297,
+                    1453
+                ],
+                "score": 0.956
+            },
+            {
+                "category_id": 4,
+                "poly": [
+                    297,
+                    527,
+                    1406,
+                    527,
+                    1406,
+                    614,
+                    297,
+                    614
+                ],
+                "score": 0.952
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    295,
+                    1223,
+                    1401,
+                    1223,
+                    1401,
+                    1292,
+                    295,
+                    1292
+                ],
+                "score": 0.948
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    296,
+                    1972,
+                    1403,
+                    1972,
+                    1403,
+                    2038,
+                    296,
+                    2038
+                ],
+                "score": 0.943
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    295,
+                    1465,
+                    1402,
+                    1465,
+                    1402,
+                    1529,
+                    295,
+                    1529
+                ],
+                "score": 0.941
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    440,
+                    1302,
+                    1256,
+                    1302,
+                    1256,
+                    1380,
+                    440,
+                    1380
+                ],
+                "score": 0.939
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1622,
+                    854,
+                    1622,
+                    854,
+                    1656,
+                    298,
+                    1656
+                ],
+                "score": 0.927
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    407,
+                    1537,
+                    1289,
+                    1537,
+                    1289,
+                    1615,
+                    407,
+                    1615
+                ],
+                "score": 0.923
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    299,
+                    76,
+                    813,
+                    76,
+                    813,
+                    104,
+                    299,
+                    104
+                ],
+                "score": 0.892
+            },
+            {
+                "category_id": 9,
+                "poly": [
+                    1365,
+                    1853,
+                    1401,
+                    1853,
+                    1401,
+                    1883,
+                    1365,
+                    1883
+                ],
+                "score": 0.878
+            },
+            {
+                "category_id": 9,
+                "poly": [
+                    1366,
+                    1324,
+                    1400,
+                    1324,
+                    1400,
+                    1354,
+                    1366,
+                    1354
+                ],
+                "score": 0.873
+            },
+            {
+                "category_id": 9,
+                "poly": [
+                    1365,
+                    1558,
+                    1400,
+                    1558,
+                    1400,
+                    1588,
+                    1365,
+                    1588
+                ],
+                "score": 0.857
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    840,
+                    2088,
+                    858,
+                    2088,
+                    858,
+                    2112,
+                    840,
+                    2112
+                ],
+                "score": 0.789
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    384,
+                    1768,
+                    1313,
+                    1768,
+                    1313,
+                    1968,
+                    384,
+                    1968
+                ],
+                "score": 0.95,
+                "latex": "\\begin{array} { r l } & { \\boldsymbol { \\hat { \\theta } _ { S } } = \\underset { \\boldsymbol { \\theta _ { S } } } { \\arg \\operatorname* { m i n } } \\sum _ { i } ^ { N } - \\mathbf { p } ^ { T } ( \\mathbf { x } _ { i } ; \\boldsymbol { \\theta _ { T } } , \\mathbf { D } ^ { + } ) \\cdot \\log \\mathbf { p } ^ { S } ( \\mathbf { x } _ { i } ; \\boldsymbol { \\theta _ { S } } , \\mathbf { D } ^ { + } ) } \\\\ & { \\quad = \\underset { \\boldsymbol { \\theta _ { S } } } { \\arg \\operatorname* { m i n } } \\sum _ { i } ^ { N } \\sum _ { j } ^ { K + 1 } - \\frac { \\exp \\left( \\mathbf { z } _ { i } ^ { T } \\cdot \\mathbf { d } _ { j } / \\tau ^ { T } \\right) } { \\sum _ { \\mathbf { d } \\sim \\mathbf { D } ^ { + } } \\exp \\left( \\mathbf { z } _ { i } ^ { T } \\cdot \\mathbf { d } / \\tau ^ { T } \\right) } \\cdot \\log \\frac { \\exp \\left( \\mathbf { z } _ { i } ^ { S } \\cdot \\mathbf { d } _ { j } / \\tau ^ { S } \\right) } { \\sum _ { \\mathbf { d } \\sim \\mathbf { D } ^ { + } } \\exp \\left( \\mathbf { z } _ { i } ^ { S } \\cdot \\mathbf { d } / \\tau ^ { S } \\right) } . } \\end{array}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    690,
+                    865,
+                    975,
+                    865,
+                    975,
+                    901,
+                    690,
+                    901
+                ],
+                "score": 0.94,
+                "latex": "\\mathbf { z } _ { i } ^ { T } = f _ { \\theta } ^ { T } ( \\mathbf { x } _ { i } ) / \\vert \\vert f _ { \\theta } ^ { T } ( \\mathbf { x } _ { i } ) \\vert \\vert _ { 2 }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    869,
+                    1255,
+                    1052,
+                    1255,
+                    1052,
+                    1292,
+                    869,
+                    1292
+                ],
+                "score": 0.93,
+                "latex": "\\mathbf { p } ^ { T } ( \\mathbf { x } _ { i } ; \\boldsymbol { \\theta } _ { T } ; \\mathbf { D } ^ { + } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1036,
+                    864,
+                    1317,
+                    864,
+                    1317,
+                    901,
+                    1036,
+                    901
+                ],
+                "score": 0.93,
+                "latex": "\\mathbf { z } _ { i } ^ { S } = f _ { \\theta } ^ { S } ( \\mathbf { \\bar { x } } _ { i } ^ { \\star } ) / \\vert \\vert f _ { \\theta } ^ { S } ( \\mathbf { x } _ { i } ) \\vert \\vert _ { 2 }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    955,
+                    1173,
+                    1092,
+                    1173,
+                    1092,
+                    1209,
+                    955,
+                    1209
+                ],
+                "score": 0.93,
+                "latex": "\\mathbf { d } _ { K + 1 } = \\mathbf { z } _ { i } ^ { T }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    625,
+                    1174,
+                    895,
+                    1174,
+                    895,
+                    1208,
+                    625,
+                    1208
+                ],
+                "score": 0.93,
+                "latex": "\\mathbf { D } ^ { + } = [ \\mathbf { d } _ { 1 } . . . \\mathbf { d } _ { K } , \\mathbf { d } _ { K + 1 } ]"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    674,
+                    1699,
+                    857,
+                    1699,
+                    857,
+                    1734,
+                    674,
+                    1734
+                ],
+                "score": 0.92,
+                "latex": "\\mathbf { p } ^ { T } ( \\mathbf { x } _ { i } ; \\theta _ { T } , \\mathbf { D } ^ { + } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1068,
+                    1699,
+                    1248,
+                    1699,
+                    1248,
+                    1734,
+                    1068,
+                    1734
+                ],
+                "score": 0.92,
+                "latex": "\\mathbf { p } ^ { S } ( \\mathbf { x } _ { i } ; \\boldsymbol { \\theta } _ { S } , \\mathbf { D } ^ { \\dagger } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    445,
+                    1464,
+                    624,
+                    1464,
+                    624,
+                    1501,
+                    445,
+                    1501
+                ],
+                "score": 0.92,
+                "latex": "\\mathbf { p } ^ { S } ( x _ { i } ; \\theta _ { S } , \\mathbf { D } ^ { + } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    343,
+                    1221,
+                    525,
+                    1221,
+                    525,
+                    1258,
+                    343,
+                    1258
+                ],
+                "score": 0.91,
+                "latex": "\\mathbf { p } ^ { T } ( \\mathbf { x } _ { i } ; \\boldsymbol { \\theta } _ { T } ; \\mathbf { D } ^ { + } )"
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    440,
+                    1300,
+                    1256,
+                    1300,
+                    1256,
+                    1380,
+                    440,
+                    1380
+                ],
+                "score": 0.91,
+                "latex": "\\mathbf { p } ^ { T } ( \\mathbf { x } _ { i } ; \\theta _ { T } , \\mathbf { D } ^ { + } ) = \\left[ p _ { 1 } ^ { T } \\dots p _ { K + 1 } ^ { T } \\right] , \\qquad p _ { j } ^ { T } = \\frac { \\exp ( \\mathbf { z } _ { i } ^ { T } \\cdot \\mathbf { d } _ { j } / \\tau ^ { T } ) } { \\sum _ { \\mathbf { d } \\sim \\mathbf { D } ^ { + } } \\exp ( \\mathbf { z } _ { i } ^ { T } \\cdot \\mathbf { d } / \\tau ^ { T } ) } ,"
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    406,
+                    1531,
+                    1289,
+                    1531,
+                    1289,
+                    1613,
+                    406,
+                    1613
+                ],
+                "score": 0.91,
+                "latex": "\\mathbf { p } ^ { S } ( \\mathbf { x } _ { i } ; \\theta _ { S } , \\mathbf { D } ^ { + } ) = \\left[ p _ { 1 } ^ { S } \\dots p _ { K + 1 } ^ { S } \\right] , \\qquad { \\mathrm { w h e r e ~ } } p _ { j } ^ { S } = { \\frac { \\exp ( \\mathbf { z } _ { i } ^ { S } \\cdot \\mathbf { d } _ { j } / \\tau ^ { S } ) } { \\sum _ { \\mathbf { d } \\sim \\mathbf { D } ^ { + } } \\exp ( \\mathbf { z } _ { i } ^ { S } \\cdot \\mathbf { d } / \\tau ^ { S } ) } } ,"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    868,
+                    2005,
+                    900,
+                    2005,
+                    900,
+                    2041,
+                    868,
+                    2041
+                ],
+                "score": 0.91,
+                "latex": "p _ { j } ^ { I ^ { \\prime } }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    297,
+                    897,
+                    455,
+                    897,
+                    455,
+                    934,
+                    297,
+                    934
+                ],
+                "score": 0.91,
+                "latex": "\\mathbf { z } _ { i } ^ { T } , \\mathbf { z } _ { i } ^ { S } \\in \\mathbb { R } ^ { D }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    520,
+                    898,
+                    557,
+                    898,
+                    557,
+                    935,
+                    520,
+                    935
+                ],
+                "score": 0.9,
+                "latex": "f _ { \\theta } ^ { T }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1263,
+                    1222,
+                    1297,
+                    1222,
+                    1297,
+                    1258,
+                    1263,
+                    1258
+                ],
+                "score": 0.9,
+                "latex": "\\mathbf { z } _ { i } ^ { T }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    614,
+                    898,
+                    650,
+                    898,
+                    650,
+                    935,
+                    614,
+                    935
+                ],
+                "score": 0.9,
+                "latex": "f _ { \\theta } ^ { S }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1001,
+                    1388,
+                    1042,
+                    1388,
+                    1042,
+                    1424,
+                    1001,
+                    1424
+                ],
+                "score": 0.9,
+                "latex": "( ) ^ { T }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    299,
+                    934,
+                    408,
+                    934,
+                    408,
+                    965,
+                    299,
+                    965
+                ],
+                "score": 0.89,
+                "latex": "[ \\mathbf { d } _ { 1 } . . . \\mathbf { d } _ { K } ]"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    298,
+                    1260,
+                    499,
+                    1260,
+                    499,
+                    1291,
+                    298,
+                    1291
+                ],
+                "score": 0.89,
+                "latex": "( j = 1 , . . . , K + 1 )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    297,
+                    1173,
+                    349,
+                    1173,
+                    349,
+                    1209,
+                    297,
+                    1209
+                ],
+                "score": 0.89,
+                "latex": "( \\mathbf { z } _ { i } ^ { T } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    343,
+                    1388,
+                    379,
+                    1388,
+                    379,
+                    1418,
+                    343,
+                    1418
+                ],
+                "score": 0.89,
+                "latex": "\\tau ^ { T }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1305,
+                    2003,
+                    1337,
+                    2003,
+                    1337,
+                    2041,
+                    1305,
+                    2041
+                ],
+                "score": 0.89,
+                "latex": "p _ { j } ^ { S }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    344,
+                    1621,
+                    378,
+                    1621,
+                    378,
+                    1651,
+                    344,
+                    1651
+                ],
+                "score": 0.88,
+                "latex": "\\tau ^ { S }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1125,
+                    932,
+                    1158,
+                    932,
+                    1158,
+                    965,
+                    1125,
+                    965
+                ],
+                "score": 0.88,
+                "latex": "{ \\bf d } _ { j }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    908,
+                    1118,
+                    938,
+                    1118,
+                    938,
+                    1145,
+                    908,
+                    1145
+                ],
+                "score": 0.87,
+                "latex": "\\mathbf { x } _ { i }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1349,
+                    1224,
+                    1381,
+                    1224,
+                    1381,
+                    1259,
+                    1349,
+                    1259
+                ],
+                "score": 0.86,
+                "latex": "{ \\bf d } _ { j }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    405,
+                    1736,
+                    434,
+                    1736,
+                    434,
+                    1762,
+                    405,
+                    1762
+                ],
+                "score": 0.86,
+                "latex": "\\mathbf { x } _ { i }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    463,
+                    1088,
+                    493,
+                    1088,
+                    493,
+                    1114,
+                    463,
+                    1114
+                ],
+                "score": 0.85,
+                "latex": "\\mathbf { x } _ { i }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1338,
+                    900,
+                    1404,
+                    900,
+                    1404,
+                    931,
+                    1338,
+                    931
+                ],
+                "score": 0.84,
+                "latex": "\\mathbf { D } ="
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    824,
+                    842,
+                    853,
+                    842,
+                    853,
+                    866,
+                    824,
+                    866
+                ],
+                "score": 0.82,
+                "latex": "\\mathbf { x } _ { i }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    806,
+                    932,
+                    835,
+                    932,
+                    835,
+                    959,
+                    806,
+                    959
+                ],
+                "score": 0.81,
+                "latex": "K"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    574,
+                    1421,
+                    604,
+                    1421,
+                    604,
+                    1454,
+                    574,
+                    1454
+                ],
+                "score": 0.58,
+                "latex": "( \\cdot )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1195,
+                    962,
+                    1224,
+                    962,
+                    1224,
+                    990,
+                    1195,
+                    990
+                ],
+                "score": 0.53,
+                "latex": "\\mathbf { D }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    700,
+                    1114,
+                    729,
+                    1114,
+                    729,
+                    1142,
+                    700,
+                    1142
+                ],
+                "score": 0.52,
+                "latex": "\\mathbf { D }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    923,
+                    1054,
+                    951,
+                    1054,
+                    951,
+                    1082,
+                    923,
+                    1082
+                ],
+                "score": 0.52,
+                "latex": "\\mathbf { D }"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    743.0,
+                    240.0,
+                    835.0,
+                    240.0,
+                    835.0,
+                    277.0,
+                    743.0,
+                    277.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    988.0,
+                    244.0,
+                    1079.0,
+                    244.0,
+                    1079.0,
+                    278.0,
+                    988.0,
+                    278.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    618.0,
+                    258.0,
+                    727.0,
+                    258.0,
+                    727.0,
+                    303.0,
+                    618.0,
+                    303.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    932.0,
+                    260.0,
+                    987.0,
+                    260.0,
+                    987.0,
+                    299.0,
+                    932.0,
+                    299.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1074.0,
+                    273.0,
+                    1085.0,
+                    273.0,
+                    1085.0,
+                    285.0,
+                    1074.0,
+                    285.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    507.0,
+                    311.0,
+                    517.0,
+                    311.0,
+                    517.0,
+                    325.0,
+                    507.0,
+                    325.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    956.0,
+                    291.0,
+                    981.0,
+                    291.0,
+                    981.0,
+                    320.0,
+                    956.0,
+                    320.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1079.0,
+                    300.0,
+                    1284.0,
+                    300.0,
+                    1284.0,
+                    333.0,
+                    1079.0,
+                    333.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    429.0,
+                    335.0,
+                    465.0,
+                    335.0,
+                    465.0,
+                    358.0,
+                    429.0,
+                    358.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    744.0,
+                    336.0,
+                    839.0,
+                    336.0,
+                    839.0,
+                    377.0,
+                    744.0,
+                    377.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    877.0,
+                    337.0,
+                    920.0,
+                    337.0,
+                    920.0,
+                    380.0,
+                    877.0,
+                    380.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    994.0,
+                    347.0,
+                    1022.0,
+                    347.0,
+                    1022.0,
+                    367.0,
+                    994.0,
+                    367.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1095.0,
+                    338.0,
+                    1182.0,
+                    338.0,
+                    1182.0,
+                    370.0,
+                    1095.0,
+                    370.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1250.0,
+                    339.0,
+                    1405.0,
+                    339.0,
+                    1405.0,
+                    377.0,
+                    1250.0,
+                    377.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    850.0,
+                    380.0,
+                    861.0,
+                    380.0,
+                    861.0,
+                    393.0,
+                    850.0,
+                    393.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    956.0,
+                    388.0,
+                    981.0,
+                    388.0,
+                    981.0,
+                    419.0,
+                    956.0,
+                    419.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1304.0,
+                    389.0,
+                    1316.0,
+                    389.0,
+                    1316.0,
+                    400.0,
+                    1304.0,
+                    400.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    740.0,
+                    395.0,
+                    831.0,
+                    395.0,
+                    831.0,
+                    430.0,
+                    740.0,
+                    430.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    990.0,
+                    399.0,
+                    1079.0,
+                    399.0,
+                    1079.0,
+                    433.0,
+                    990.0,
+                    433.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    626.0,
+                    416.0,
+                    722.0,
+                    416.0,
+                    722.0,
+                    451.0,
+                    626.0,
+                    451.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    902.0,
+                    417.0,
+                    1039.0,
+                    417.0,
+                    1039.0,
+                    480.0,
+                    902.0,
+                    480.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1110.0,
+                    417.0,
+                    1252.0,
+                    417.0,
+                    1252.0,
+                    478.0,
+                    1110.0,
+                    478.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    326.0,
+                    472.0,
+                    507.0,
+                    472.0,
+                    507.0,
+                    505.0,
+                    326.0,
+                    505.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    569.0,
+                    472.0,
+                    790.0,
+                    472.0,
+                    790.0,
+                    501.0,
+                    569.0,
+                    501.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1079.0,
+                    468.0,
+                    1284.0,
+                    468.0,
+                    1284.0,
+                    503.0,
+                    1079.0,
+                    503.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    830.0,
+                    234.5,
+                    886.0,
+                    234.5,
+                    886.0,
+                    260.5,
+                    830.0,
+                    260.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    827.0,
+                    459.0,
+                    877.0,
+                    459.0,
+                    877.0,
+                    479.5,
+                    827.0,
+                    479.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    527.0,
+                    1406.0,
+                    527.0,
+                    1406.0,
+                    560.0,
+                    295.0,
+                    560.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    555.0,
+                    1406.0,
+                    555.0,
+                    1406.0,
+                    588.0,
+                    294.0,
+                    588.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    580.0,
+                    1391.0,
+                    580.0,
+                    1391.0,
+                    620.0,
+                    291.0,
+                    620.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    73.0,
+                    815.0,
+                    73.0,
+                    815.0,
+                    108.0,
+                    295.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    837.0,
+                    2085.0,
+                    862.0,
+                    2085.0,
+                    862.0,
+                    2118.0,
+                    837.0,
+                    2118.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    836.0,
+                    823.0,
+                    836.0,
+                    823.0,
+                    870.0,
+                    296.0,
+                    870.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    854.0,
+                    836.0,
+                    1405.0,
+                    836.0,
+                    1405.0,
+                    870.0,
+                    854.0,
+                    870.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    290.0,
+                    862.0,
+                    689.0,
+                    862.0,
+                    689.0,
+                    907.0,
+                    290.0,
+                    907.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    976.0,
+                    862.0,
+                    1035.0,
+                    862.0,
+                    1035.0,
+                    907.0,
+                    976.0,
+                    907.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1318.0,
+                    862.0,
+                    1408.0,
+                    862.0,
+                    1408.0,
+                    907.0,
+                    1318.0,
+                    907.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    286.0,
+                    886.0,
+                    296.0,
+                    886.0,
+                    296.0,
+                    945.0,
+                    286.0,
+                    945.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    456.0,
+                    886.0,
+                    519.0,
+                    886.0,
+                    519.0,
+                    945.0,
+                    456.0,
+                    945.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    558.0,
+                    886.0,
+                    613.0,
+                    886.0,
+                    613.0,
+                    945.0,
+                    558.0,
+                    945.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    651.0,
+                    886.0,
+                    1337.0,
+                    886.0,
+                    1337.0,
+                    945.0,
+                    651.0,
+                    945.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1405.0,
+                    886.0,
+                    1411.0,
+                    886.0,
+                    1411.0,
+                    945.0,
+                    1405.0,
+                    945.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    932.0,
+                    298.0,
+                    932.0,
+                    298.0,
+                    967.0,
+                    295.0,
+                    967.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    409.0,
+                    932.0,
+                    805.0,
+                    932.0,
+                    805.0,
+                    967.0,
+                    409.0,
+                    967.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    836.0,
+                    932.0,
+                    1124.0,
+                    932.0,
+                    1124.0,
+                    967.0,
+                    836.0,
+                    967.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1159.0,
+                    932.0,
+                    1406.0,
+                    932.0,
+                    1406.0,
+                    967.0,
+                    1159.0,
+                    967.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    962.0,
+                    1194.0,
+                    962.0,
+                    1194.0,
+                    997.0,
+                    295.0,
+                    997.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1225.0,
+                    962.0,
+                    1405.0,
+                    962.0,
+                    1405.0,
+                    997.0,
+                    1225.0,
+                    997.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    993.0,
+                    1406.0,
+                    993.0,
+                    1406.0,
+                    1027.0,
+                    295.0,
+                    1027.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1023.0,
+                    1406.0,
+                    1023.0,
+                    1406.0,
+                    1057.0,
+                    295.0,
+                    1057.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1051.0,
+                    922.0,
+                    1051.0,
+                    922.0,
+                    1088.0,
+                    294.0,
+                    1088.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    952.0,
+                    1051.0,
+                    1405.0,
+                    1051.0,
+                    1405.0,
+                    1088.0,
+                    952.0,
+                    1088.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1083.0,
+                    462.0,
+                    1083.0,
+                    462.0,
+                    1118.0,
+                    294.0,
+                    1118.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    494.0,
+                    1083.0,
+                    1405.0,
+                    1083.0,
+                    1405.0,
+                    1118.0,
+                    494.0,
+                    1118.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1113.0,
+                    699.0,
+                    1113.0,
+                    699.0,
+                    1147.0,
+                    294.0,
+                    1147.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    730.0,
+                    1113.0,
+                    907.0,
+                    1113.0,
+                    907.0,
+                    1147.0,
+                    730.0,
+                    1147.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    939.0,
+                    1113.0,
+                    1405.0,
+                    1113.0,
+                    1405.0,
+                    1147.0,
+                    939.0,
+                    1147.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1140.0,
+                    1406.0,
+                    1140.0,
+                    1406.0,
+                    1182.0,
+                    291.0,
+                    1182.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1170.0,
+                    296.0,
+                    1170.0,
+                    296.0,
+                    1214.0,
+                    291.0,
+                    1214.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    350.0,
+                    1170.0,
+                    624.0,
+                    1170.0,
+                    624.0,
+                    1214.0,
+                    350.0,
+                    1214.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    896.0,
+                    1170.0,
+                    954.0,
+                    1170.0,
+                    954.0,
+                    1214.0,
+                    896.0,
+                    1214.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1093.0,
+                    1170.0,
+                    1105.0,
+                    1170.0,
+                    1105.0,
+                    1214.0,
+                    1093.0,
+                    1214.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    664.0,
+                    1405.0,
+                    664.0,
+                    1405.0,
+                    704.0,
+                    293.0,
+                    704.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    696.0,
+                    1405.0,
+                    696.0,
+                    1405.0,
+                    732.0,
+                    294.0,
+                    732.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    726.0,
+                    1405.0,
+                    726.0,
+                    1405.0,
+                    764.0,
+                    294.0,
+                    764.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    758.0,
+                    1405.0,
+                    758.0,
+                    1405.0,
+                    795.0,
+                    294.0,
+                    795.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    785.0,
+                    414.0,
+                    785.0,
+                    414.0,
+                    829.0,
+                    292.0,
+                    829.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1667.0,
+                    1404.0,
+                    1667.0,
+                    1404.0,
+                    1704.0,
+                    295.0,
+                    1704.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1698.0,
+                    673.0,
+                    1698.0,
+                    673.0,
+                    1735.0,
+                    294.0,
+                    1735.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    858.0,
+                    1698.0,
+                    1067.0,
+                    1698.0,
+                    1067.0,
+                    1735.0,
+                    858.0,
+                    1735.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1249.0,
+                    1698.0,
+                    1407.0,
+                    1698.0,
+                    1407.0,
+                    1735.0,
+                    1249.0,
+                    1735.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1729.0,
+                    404.0,
+                    1729.0,
+                    404.0,
+                    1766.0,
+                    296.0,
+                    1766.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    435.0,
+                    1729.0,
+                    527.0,
+                    1729.0,
+                    527.0,
+                    1766.0,
+                    435.0,
+                    1766.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1388.0,
+                    342.0,
+                    1388.0,
+                    342.0,
+                    1424.0,
+                    296.0,
+                    1424.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    380.0,
+                    1388.0,
+                    1000.0,
+                    1388.0,
+                    1000.0,
+                    1424.0,
+                    380.0,
+                    1424.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1043.0,
+                    1388.0,
+                    1405.0,
+                    1388.0,
+                    1405.0,
+                    1424.0,
+                    1043.0,
+                    1424.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1418.0,
+                    573.0,
+                    1418.0,
+                    573.0,
+                    1455.0,
+                    293.0,
+                    1455.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    605.0,
+                    1418.0,
+                    1189.0,
+                    1418.0,
+                    1189.0,
+                    1455.0,
+                    605.0,
+                    1455.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1221.0,
+                    342.0,
+                    1221.0,
+                    342.0,
+                    1261.0,
+                    293.0,
+                    1261.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    526.0,
+                    1221.0,
+                    1262.0,
+                    1221.0,
+                    1262.0,
+                    1261.0,
+                    526.0,
+                    1261.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1298.0,
+                    1221.0,
+                    1348.0,
+                    1221.0,
+                    1348.0,
+                    1261.0,
+                    1298.0,
+                    1261.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1382.0,
+                    1221.0,
+                    1405.0,
+                    1221.0,
+                    1405.0,
+                    1261.0,
+                    1382.0,
+                    1261.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1256.0,
+                    297.0,
+                    1256.0,
+                    297.0,
+                    1294.0,
+                    291.0,
+                    1294.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    500.0,
+                    1256.0,
+                    868.0,
+                    1256.0,
+                    868.0,
+                    1294.0,
+                    500.0,
+                    1294.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1053.0,
+                    1256.0,
+                    1202.0,
+                    1256.0,
+                    1202.0,
+                    1294.0,
+                    1053.0,
+                    1294.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1967.0,
+                    1405.0,
+                    1967.0,
+                    1405.0,
+                    2010.0,
+                    294.0,
+                    2010.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    2000.0,
+                    867.0,
+                    2000.0,
+                    867.0,
+                    2040.0,
+                    294.0,
+                    2040.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    901.0,
+                    2000.0,
+                    1304.0,
+                    2000.0,
+                    1304.0,
+                    2040.0,
+                    901.0,
+                    2040.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1338.0,
+                    2000.0,
+                    1404.0,
+                    2000.0,
+                    1404.0,
+                    2040.0,
+                    1338.0,
+                    2040.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1459.0,
+                    444.0,
+                    1459.0,
+                    444.0,
+                    1506.0,
+                    292.0,
+                    1506.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    625.0,
+                    1459.0,
+                    1408.0,
+                    1459.0,
+                    1408.0,
+                    1506.0,
+                    625.0,
+                    1506.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1493.0,
+                    420.0,
+                    1493.0,
+                    420.0,
+                    1533.0,
+                    293.0,
+                    1533.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1616.0,
+                    343.0,
+                    1616.0,
+                    343.0,
+                    1661.0,
+                    293.0,
+                    1661.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    379.0,
+                    1616.0,
+                    858.0,
+                    1616.0,
+                    858.0,
+                    1661.0,
+                    379.0,
+                    1661.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 3,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1127,
+                    1405,
+                    1127,
+                    1405,
+                    1434,
+                    298,
+                    1434
+                ],
+                "score": 0.983
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    836,
+                    1405,
+                    836,
+                    1405,
+                    1113,
+                    297,
+                    1113
+                ],
+                "score": 0.981
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1528,
+                    1405,
+                    1528,
+                    1405,
+                    1774,
+                    298,
+                    1774
+                ],
+                "score": 0.981
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1788,
+                    1404,
+                    1788,
+                    1404,
+                    2035,
+                    298,
+                    2035
+                ],
+                "score": 0.978
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    226,
+                    1404,
+                    226,
+                    1404,
+                    397,
+                    297,
+                    397
+                ],
+                "score": 0.975
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    605,
+                    496,
+                    1095,
+                    496,
+                    1095,
+                    586,
+                    605,
+                    586
+                ],
+                "score": 0.956
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    602,
+                    1407,
+                    602,
+                    1407,
+                    665,
+                    297,
+                    665
+                ],
+                "score": 0.951
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    408,
+                    1406,
+                    408,
+                    1406,
+                    476,
+                    297,
+                    476
+                ],
+                "score": 0.943
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    299,
+                    779,
+                    544,
+                    779,
+                    544,
+                    811,
+                    299,
+                    811
+                ],
+                "score": 0.917
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    300,
+                    709,
+                    541,
+                    709,
+                    541,
+                    744,
+                    300,
+                    744
+                ],
+                "score": 0.912
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    300,
+                    75,
+                    813,
+                    75,
+                    813,
+                    105,
+                    300,
+                    105
+                ],
+                "score": 0.906
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    301,
+                    1471,
+                    745,
+                    1471,
+                    745,
+                    1503,
+                    301,
+                    1503
+                ],
+                "score": 0.898
+            },
+            {
+                "category_id": 9,
+                "poly": [
+                    1366,
+                    525,
+                    1400,
+                    525,
+                    1400,
+                    555,
+                    1366,
+                    555
+                ],
+                "score": 0.881
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    841,
+                    2088,
+                    858,
+                    2088,
+                    858,
+                    2112,
+                    841,
+                    2112
+                ],
+                "score": 0.725
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    605,
+                    492,
+                    1093,
+                    492,
+                    1093,
+                    589,
+                    605,
+                    589
+                ],
+                "score": 0.95,
+                "latex": "\\mathcal { L } _ { N C E } = \\sum _ { i } ^ { N } - \\log \\frac { \\exp ( \\mathbf { z } _ { i } ^ { T } \\cdot \\mathbf { z } _ { i } ^ { S } / \\tau ) } { \\sum _ { \\mathbf { d } \\sim \\mathbf { D } ^ { + } } \\exp ( \\mathbf { z } _ { i } ^ { S } \\cdot \\mathbf { d } / \\tau ) } ,"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1058,
+                    261,
+                    1125,
+                    261,
+                    1125,
+                    300,
+                    1058,
+                    300
+                ],
+                "score": 0.93,
+                "latex": "p _ { K + 1 } ^ { S }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    564,
+                    441,
+                    630,
+                    441,
+                    630,
+                    478,
+                    564,
+                    478
+                ],
+                "score": 0.93,
+                "latex": "p _ { K + 1 } ^ { T }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    670,
+                    1340,
+                    796,
+                    1340,
+                    796,
+                    1371,
+                    670,
+                    1371
+                ],
+                "score": 0.92,
+                "latex": "\\tau ^ { T } = 0 . 0 1"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    935,
+                    230,
+                    1006,
+                    230,
+                    1006,
+                    263,
+                    935,
+                    263
+                ],
+                "score": 0.92,
+                "latex": "{ \\bf d } _ { K + 1 }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1151,
+                    1340,
+                    1262,
+                    1340,
+                    1262,
+                    1370,
+                    1151,
+                    1370
+                ],
+                "score": 0.91,
+                "latex": "\\tau ^ { S } = 0 . 2"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    849,
+                    228,
+                    884,
+                    228,
+                    884,
+                    264,
+                    849,
+                    264
+                ],
+                "score": 0.91,
+                "latex": "{ \\mathbf z } _ { i } ^ { T }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1199,
+                    298,
+                    1232,
+                    298,
+                    1232,
+                    335,
+                    1199,
+                    335
+                ],
+                "score": 0.9,
+                "latex": "\\mathbf { z } _ { i } ^ { S }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    353,
+                    332,
+                    388,
+                    332,
+                    388,
+                    368,
+                    353,
+                    368
+                ],
+                "score": 0.9,
+                "latex": "{ \\mathbf z } _ { i } ^ { T }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    775,
+                    261,
+                    810,
+                    261,
+                    810,
+                    302,
+                    775,
+                    302
+                ],
+                "score": 0.9,
+                "latex": "p _ { j } ^ { T }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    726,
+                    408,
+                    817,
+                    408,
+                    817,
+                    440,
+                    726,
+                    440
+                ],
+                "score": 0.9,
+                "latex": "\\tau ^ { T } \\to 0"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    662,
+                    1401,
+                    697,
+                    1401,
+                    697,
+                    1430,
+                    662,
+                    1430
+                ],
+                "score": 0.89,
+                "latex": "\\tau ^ { T }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1098,
+                    409,
+                    1136,
+                    409,
+                    1136,
+                    444,
+                    1098,
+                    444
+                ],
+                "score": 0.89,
+                "latex": "\\mathbf { p } ^ { T }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    722,
+                    298,
+                    757,
+                    298,
+                    757,
+                    328,
+                    722,
+                    328
+                ],
+                "score": 0.87,
+                "latex": "\\tau ^ { T }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    686,
+                    1821,
+                    774,
+                    1821,
+                    774,
+                    1851,
+                    686,
+                    1851
+                ],
+                "score": 0.83,
+                "latex": "{ \\sim } 1 . 3 \\mathbf { M }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    749,
+                    1403,
+                    777,
+                    1403,
+                    777,
+                    1430,
+                    749,
+                    1430
+                ],
+                "score": 0.82,
+                "latex": "K"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    378,
+                    1372,
+                    407,
+                    1372,
+                    407,
+                    1399,
+                    378,
+                    1399
+                ],
+                "score": 0.82,
+                "latex": "K"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1108,
+                    335,
+                    1136,
+                    335,
+                    1136,
+                    362,
+                    1108,
+                    362
+                ],
+                "score": 0.45,
+                "latex": "\\mathbf { D }"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    778.0,
+                    546.0,
+                    778.0,
+                    546.0,
+                    814.0,
+                    294.0,
+                    814.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    706.0,
+                    545.0,
+                    706.0,
+                    545.0,
+                    750.0,
+                    293.0,
+                    750.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    73.0,
+                    815.0,
+                    73.0,
+                    815.0,
+                    108.0,
+                    295.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1470.0,
+                    749.0,
+                    1470.0,
+                    749.0,
+                    1508.0,
+                    294.0,
+                    1508.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    839.0,
+                    2085.0,
+                    861.0,
+                    2085.0,
+                    861.0,
+                    2120.0,
+                    839.0,
+                    2120.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1128.0,
+                    1403.0,
+                    1128.0,
+                    1403.0,
+                    1160.0,
+                    296.0,
+                    1160.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1158.0,
+                    1406.0,
+                    1158.0,
+                    1406.0,
+                    1190.0,
+                    295.0,
+                    1190.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1186.0,
+                    1405.0,
+                    1186.0,
+                    1405.0,
+                    1221.0,
+                    293.0,
+                    1221.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1218.0,
+                    1405.0,
+                    1218.0,
+                    1405.0,
+                    1253.0,
+                    295.0,
+                    1253.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1250.0,
+                    1405.0,
+                    1250.0,
+                    1405.0,
+                    1282.0,
+                    295.0,
+                    1282.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1278.0,
+                    1406.0,
+                    1278.0,
+                    1406.0,
+                    1314.0,
+                    295.0,
+                    1314.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1310.0,
+                    1408.0,
+                    1310.0,
+                    1408.0,
+                    1345.0,
+                    295.0,
+                    1345.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1336.0,
+                    669.0,
+                    1336.0,
+                    669.0,
+                    1377.0,
+                    292.0,
+                    1377.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    797.0,
+                    1336.0,
+                    1150.0,
+                    1336.0,
+                    1150.0,
+                    1377.0,
+                    797.0,
+                    1377.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1263.0,
+                    1336.0,
+                    1407.0,
+                    1336.0,
+                    1407.0,
+                    1377.0,
+                    1263.0,
+                    1377.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1372.0,
+                    377.0,
+                    1372.0,
+                    377.0,
+                    1404.0,
+                    296.0,
+                    1404.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    408.0,
+                    1372.0,
+                    1403.0,
+                    1372.0,
+                    1403.0,
+                    1404.0,
+                    408.0,
+                    1404.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1404.0,
+                    661.0,
+                    1404.0,
+                    661.0,
+                    1434.0,
+                    294.0,
+                    1434.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    698.0,
+                    1404.0,
+                    748.0,
+                    1404.0,
+                    748.0,
+                    1434.0,
+                    698.0,
+                    1434.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    778.0,
+                    1404.0,
+                    789.0,
+                    1404.0,
+                    789.0,
+                    1434.0,
+                    778.0,
+                    1434.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    837.0,
+                    1403.0,
+                    837.0,
+                    1403.0,
+                    870.0,
+                    296.0,
+                    870.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    867.0,
+                    1406.0,
+                    867.0,
+                    1406.0,
+                    900.0,
+                    296.0,
+                    900.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    899.0,
+                    1406.0,
+                    899.0,
+                    1406.0,
+                    931.0,
+                    296.0,
+                    931.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    929.0,
+                    1403.0,
+                    929.0,
+                    1403.0,
+                    961.0,
+                    296.0,
+                    961.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    959.0,
+                    1403.0,
+                    959.0,
+                    1403.0,
+                    992.0,
+                    295.0,
+                    992.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    988.0,
+                    1406.0,
+                    988.0,
+                    1406.0,
+                    1024.0,
+                    294.0,
+                    1024.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1020.0,
+                    1403.0,
+                    1020.0,
+                    1403.0,
+                    1053.0,
+                    296.0,
+                    1053.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1049.0,
+                    1408.0,
+                    1049.0,
+                    1408.0,
+                    1086.0,
+                    295.0,
+                    1086.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1078.0,
+                    635.0,
+                    1078.0,
+                    635.0,
+                    1116.0,
+                    294.0,
+                    1116.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1531.0,
+                    1406.0,
+                    1531.0,
+                    1406.0,
+                    1561.0,
+                    296.0,
+                    1561.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1562.0,
+                    1405.0,
+                    1562.0,
+                    1405.0,
+                    1592.0,
+                    296.0,
+                    1592.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1592.0,
+                    1406.0,
+                    1592.0,
+                    1406.0,
+                    1622.0,
+                    296.0,
+                    1622.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1619.0,
+                    1405.0,
+                    1619.0,
+                    1405.0,
+                    1655.0,
+                    293.0,
+                    1655.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1650.0,
+                    1405.0,
+                    1650.0,
+                    1405.0,
+                    1686.0,
+                    293.0,
+                    1686.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1679.0,
+                    1405.0,
+                    1679.0,
+                    1405.0,
+                    1716.0,
+                    295.0,
+                    1716.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1710.0,
+                    1406.0,
+                    1710.0,
+                    1406.0,
+                    1750.0,
+                    293.0,
+                    1750.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1741.0,
+                    1085.0,
+                    1741.0,
+                    1085.0,
+                    1776.0,
+                    295.0,
+                    1776.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1789.0,
+                    1404.0,
+                    1789.0,
+                    1404.0,
+                    1822.0,
+                    296.0,
+                    1822.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1820.0,
+                    685.0,
+                    1820.0,
+                    685.0,
+                    1854.0,
+                    296.0,
+                    1854.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    775.0,
+                    1820.0,
+                    1408.0,
+                    1820.0,
+                    1408.0,
+                    1854.0,
+                    775.0,
+                    1854.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1849.0,
+                    1406.0,
+                    1849.0,
+                    1406.0,
+                    1886.0,
+                    292.0,
+                    1886.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1881.0,
+                    1408.0,
+                    1881.0,
+                    1408.0,
+                    1915.0,
+                    294.0,
+                    1915.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1911.0,
+                    1406.0,
+                    1911.0,
+                    1406.0,
+                    1945.0,
+                    294.0,
+                    1945.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1942.0,
+                    1405.0,
+                    1942.0,
+                    1405.0,
+                    1974.0,
+                    293.0,
+                    1974.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1973.0,
+                    1406.0,
+                    1973.0,
+                    1406.0,
+                    2005.0,
+                    294.0,
+                    2005.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    2003.0,
+                    1405.0,
+                    2003.0,
+                    1405.0,
+                    2037.0,
+                    296.0,
+                    2037.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    226.0,
+                    848.0,
+                    226.0,
+                    848.0,
+                    267.0,
+                    292.0,
+                    267.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    885.0,
+                    226.0,
+                    934.0,
+                    226.0,
+                    934.0,
+                    267.0,
+                    885.0,
+                    267.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1007.0,
+                    226.0,
+                    1405.0,
+                    226.0,
+                    1405.0,
+                    267.0,
+                    1007.0,
+                    267.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    290.0,
+                    258.0,
+                    774.0,
+                    258.0,
+                    774.0,
+                    305.0,
+                    290.0,
+                    305.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    811.0,
+                    258.0,
+                    1057.0,
+                    258.0,
+                    1057.0,
+                    305.0,
+                    811.0,
+                    305.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1126.0,
+                    258.0,
+                    1408.0,
+                    258.0,
+                    1408.0,
+                    305.0,
+                    1126.0,
+                    305.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    298.0,
+                    721.0,
+                    298.0,
+                    721.0,
+                    336.0,
+                    295.0,
+                    336.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    758.0,
+                    298.0,
+                    1198.0,
+                    298.0,
+                    1198.0,
+                    336.0,
+                    758.0,
+                    336.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1233.0,
+                    298.0,
+                    1406.0,
+                    298.0,
+                    1406.0,
+                    336.0,
+                    1233.0,
+                    336.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    332.0,
+                    352.0,
+                    332.0,
+                    352.0,
+                    368.0,
+                    294.0,
+                    368.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    389.0,
+                    332.0,
+                    1107.0,
+                    332.0,
+                    1107.0,
+                    368.0,
+                    389.0,
+                    368.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1137.0,
+                    332.0,
+                    1404.0,
+                    332.0,
+                    1404.0,
+                    368.0,
+                    1137.0,
+                    368.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    365.0,
+                    1099.0,
+                    365.0,
+                    1099.0,
+                    399.0,
+                    295.0,
+                    399.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    598.0,
+                    1405.0,
+                    598.0,
+                    1405.0,
+                    637.0,
+                    292.0,
+                    637.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    631.0,
+                    620.0,
+                    631.0,
+                    620.0,
+                    667.0,
+                    295.0,
+                    667.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    404.0,
+                    725.0,
+                    404.0,
+                    725.0,
+                    447.0,
+                    292.0,
+                    447.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    818.0,
+                    404.0,
+                    1097.0,
+                    404.0,
+                    1097.0,
+                    447.0,
+                    818.0,
+                    447.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1137.0,
+                    404.0,
+                    1407.0,
+                    404.0,
+                    1407.0,
+                    447.0,
+                    1137.0,
+                    447.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    290.0,
+                    439.0,
+                    563.0,
+                    439.0,
+                    563.0,
+                    479.0,
+                    290.0,
+                    479.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    631.0,
+                    439.0,
+                    1349.0,
+                    439.0,
+                    1349.0,
+                    479.0,
+                    631.0,
+                    479.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 4,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1407,
+                    1406,
+                    1407,
+                    1406,
+                    1865,
+                    298,
+                    1865
+                ],
+                "score": 0.983
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    1881,
+                    1404,
+                    1881,
+                    1404,
+                    2034,
+                    299,
+                    2034
+                ],
+                "score": 0.976
+            },
+            {
+                "category_id": 5,
+                "poly": [
+                    301,
+                    421,
+                    1406,
+                    421,
+                    1406,
+                    793,
+                    301,
+                    793
+                ],
+                "score": 0.967,
+                "html": "<table><tr><td rowspan=\"2\">S T</td><td rowspan=\"2\">T-1</td><td rowspan=\"2\">K</td><td colspan=\"2\">Eff-bo</td><td colspan=\"3\">Eff-b1</td><td colspan=\"3\">Mob-v3</td><td colspan=\"3\">R-18</td><td colspan=\"3\">R-34</td></tr><tr><td>T-1</td><td>T-5</td><td>K</td><td>T-1</td><td>T-5</td><td>K</td><td>T-1</td><td>T-5</td><td>K</td><td>T-1</td><td>T-5</td><td>K</td><td>T-1</td><td>T-5</td></tr><tr><td>Supervised Acc.</td><td></td><td></td><td>77.3</td><td></td><td></td><td>79.2</td><td></td><td></td><td>75.2</td><td></td><td></td><td>72.1</td><td></td><td></td><td>75.0</td><td></td></tr><tr><td>X</td><td></td><td>30.0</td><td>42.2</td><td></td><td>68.534.4</td><td>50.7</td><td>74.6</td><td>27.5</td><td>36.3</td><td>62.2</td><td>36.7</td><td>52.5</td><td>77.0</td><td>）41.5</td><td>57.4</td><td>81.6</td></tr><tr><td>R-50 △</td><td>67.4</td><td>46.0 +16.0</td><td>61.3 +19.1</td><td>82.7 +14.2</td><td>46.1 +16.1</td><td>61.4 +10.7</td><td>83.1 +8.8</td><td>44.8 +17.3</td><td>55.2 +18.9</td><td>80.3 +18.1</td><td>43.4 +6.7</td><td>57.9 +5.1</td><td>82.0 +4.8</td><td>45.2 +3.7</td><td>58.5 +1.1</td><td>82.6 +1.0</td></tr><tr><td>R-101 △</td><td>70.3</td><td>50.1 +20.1</td><td>63.0 +20.8</td><td>83.8 +15.3</td><td>50.3 +15.9</td><td>63.4 +12.7</td><td>84.6 +10.0</td><td>48.8 +21.3</td><td>59.9 +23.6</td><td>83.5 +21.3</td><td>48.6 +11.9</td><td>58.9 +6.4</td><td>82.5 +5.5</td><td>50.5 +9.0</td><td>61.6 +4.2</td><td>84.9 +3.3</td></tr><tr><td>R-152</td><td></td><td>50.7</td><td>65.3</td><td>86.0</td><td>52.4</td><td>67.3</td><td>86.9</td><td>49.5</td><td>61.4</td><td>84.6</td><td>49.1</td><td>59.5</td><td>83.3</td><td>51.4</td><td>62.7</td><td>85.8</td></tr><tr><td>△</td><td>74.2</td><td>+20.7</td><td>+23.1</td><td>+17.5</td><td>+18.0</td><td>+16.6</td><td>+12.3</td><td>+22.0</td><td>+25.1</td><td>+22.4</td><td>+12.4</td><td>+7.0</td><td>+6.3</td><td>+9.9</td><td>+5.3</td><td>+4.2</td></tr><tr><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td></tr><tr><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td></tr><tr><td>R50×2*</td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td></tr><tr><td></td><td></td><td></td><td></td><td>87.4</td><td>60.3</td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td></tr><tr><td>△</td><td>77.3</td><td>57.4 +27.4</td><td>67.6 +25.4</td><td>+18.9</td><td>+25.9</td><td>68.0 +17.3</td><td>87.6 +13.0</td><td>55.9 +18.9</td><td>68.2 +31.9</td><td>88.2 +26.0</td><td>55.3 +18.6</td><td>63.0 +10.5</td><td>84.9 +7.9</td><td>58.2 +16.7</td><td>65.7 +8.3</td><td>86.8 +5.2</td></tr></table>"
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    302,
+                    1299,
+                    1399,
+                    1299,
+                    1399,
+                    1392,
+                    302,
+                    1392
+                ],
+                "score": 0.96
+            },
+            {
+                "category_id": 4,
+                "poly": [
+                    296,
+                    1129,
+                    1404,
+                    1129,
+                    1404,
+                    1270,
+                    296,
+                    1270
+                ],
+                "score": 0.953
+            },
+            {
+                "category_id": 3,
+                "poly": [
+                    326,
+                    851,
+                    1373,
+                    851,
+                    1373,
+                    1101,
+                    326,
+                    1101
+                ],
+                "score": 0.931
+            },
+            {
+                "category_id": 6,
+                "poly": [
+                    295,
+                    223,
+                    1405,
+                    223,
+                    1405,
+                    392,
+                    295,
+                    392
+                ],
+                "score": 0.897
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    300,
+                    76,
+                    812,
+                    76,
+                    812,
+                    104,
+                    300,
+                    104
+                ],
+                "score": 0.892
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    840,
+                    2089,
+                    858,
+                    2089,
+                    858,
+                    2112,
+                    840,
+                    2112
+                ],
+                "score": 0.776
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    295,
+                    223,
+                    1405,
+                    223,
+                    1405,
+                    392,
+                    295,
+                    392
+                ],
+                "score": 0.142
+            },
+            {
+                "category_id": 4,
+                "poly": [
+                    659,
+                    1075,
+                    1112,
+                    1075,
+                    1112,
+                    1101,
+                    659,
+                    1101
+                ],
+                "score": 0.125
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1170,
+                    1682,
+                    1244,
+                    1682,
+                    1244,
+                    1711,
+                    1170,
+                    1711
+                ],
+                "score": 0.88,
+                "latex": "6 1 . 4 \\%"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    465,
+                    1713,
+                    531,
+                    1713,
+                    531,
+                    1742,
+                    465,
+                    1742
+                ],
+                "score": 0.88,
+                "latex": "5 0 \\times 2"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1053,
+                    1743,
+                    1128,
+                    1743,
+                    1128,
+                    1772,
+                    1053,
+                    1772
+                ],
+                "score": 0.88,
+                "latex": "6 8 . 2 \\%"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    636,
+                    1499,
+                    711,
+                    1499,
+                    711,
+                    1529,
+                    636,
+                    1529
+                ],
+                "score": 0.87,
+                "latex": "3 6 . 3 \\%"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1064,
+                    1682,
+                    1139,
+                    1682,
+                    1139,
+                    1711,
+                    1064,
+                    1711
+                ],
+                "score": 0.87,
+                "latex": "3 6 . 3 \\%"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    369,
+                    1943,
+                    423,
+                    1943,
+                    423,
+                    1972,
+                    369,
+                    1972
+                ],
+                "score": 0.86,
+                "latex": "10 \\%"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1242,
+                    1130,
+                    1291,
+                    1130,
+                    1291,
+                    1157,
+                    1242,
+                    1157
+                ],
+                "score": 0.86,
+                "latex": "10 \\%"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    463,
+                    1361,
+                    557,
+                    1361,
+                    557,
+                    1390,
+                    463,
+                    1390
+                ],
+                "score": 0.85,
+                "latex": "K = 1 0"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    295,
+                    1943,
+                    336,
+                    1943,
+                    336,
+                    1972,
+                    295,
+                    1972
+                ],
+                "score": 0.84,
+                "latex": "1 \\%"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1096,
+                    1130,
+                    1133,
+                    1130,
+                    1133,
+                    1156,
+                    1096,
+                    1156
+                ],
+                "score": 0.83,
+                "latex": "1 \\%"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    933,
+                    1301,
+                    957,
+                    1301,
+                    957,
+                    1327,
+                    933,
+                    1327
+                ],
+                "score": 0.77,
+                "latex": "K"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    421,
+                    1361,
+                    450,
+                    1361,
+                    450,
+                    1389,
+                    421,
+                    1389
+                ],
+                "score": 0.74,
+                "latex": "K"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    668,
+                    226,
+                    706,
+                    226,
+                    706,
+                    253,
+                    668,
+                    253
+                ],
+                "score": 0.7,
+                "latex": "( \\% )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    754,
+                    254,
+                    776,
+                    254,
+                    776,
+                    279,
+                    754,
+                    279
+                ],
+                "score": 0.56,
+                "latex": "\\pmb { \\chi }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    888,
+                    309,
+                    911,
+                    309,
+                    911,
+                    333,
+                    888,
+                    333
+                ],
+                "score": 0.47,
+                "latex": "K"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1129.0,
+                    1095.0,
+                    1129.0,
+                    1095.0,
+                    1161.0,
+                    296.0,
+                    1161.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1134.0,
+                    1129.0,
+                    1241.0,
+                    1129.0,
+                    1241.0,
+                    1161.0,
+                    1134.0,
+                    1161.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1292.0,
+                    1129.0,
+                    1405.0,
+                    1129.0,
+                    1405.0,
+                    1161.0,
+                    1292.0,
+                    1161.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1154.0,
+                    1406.0,
+                    1154.0,
+                    1406.0,
+                    1191.0,
+                    293.0,
+                    1191.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1183.0,
+                    1405.0,
+                    1183.0,
+                    1405.0,
+                    1216.0,
+                    294.0,
+                    1216.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1212.0,
+                    1405.0,
+                    1212.0,
+                    1405.0,
+                    1243.0,
+                    295.0,
+                    1243.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1241.0,
+                    905.0,
+                    1241.0,
+                    905.0,
+                    1272.0,
+                    295.0,
+                    1272.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    459.0,
+                    850.0,
+                    608.0,
+                    850.0,
+                    608.0,
+                    876.0,
+                    459.0,
+                    876.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    793.0,
+                    848.0,
+                    985.0,
+                    848.0,
+                    985.0,
+                    883.0,
+                    793.0,
+                    883.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1178.0,
+                    852.0,
+                    1276.0,
+                    852.0,
+                    1276.0,
+                    880.0,
+                    1178.0,
+                    880.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    321.0,
+                    866.0,
+                    363.0,
+                    866.0,
+                    363.0,
+                    1070.0,
+                    321.0,
+                    1070.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1048.0,
+                    884.0,
+                    1072.0,
+                    884.0,
+                    1072.0,
+                    903.0,
+                    1048.0,
+                    903.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    355.0,
+                    896.0,
+                    378.0,
+                    896.0,
+                    378.0,
+                    914.0,
+                    355.0,
+                    914.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    700.0,
+                    893.0,
+                    726.0,
+                    893.0,
+                    726.0,
+                    916.0,
+                    700.0,
+                    916.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    930.0,
+                    901.0,
+                    940.0,
+                    901.0,
+                    940.0,
+                    910.0,
+                    930.0,
+                    910.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    485.0,
+                    909.0,
+                    514.0,
+                    909.0,
+                    514.0,
+                    931.0,
+                    485.0,
+                    931.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    650.0,
+                    906.0,
+                    664.0,
+                    906.0,
+                    664.0,
+                    920.0,
+                    650.0,
+                    920.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    814.0,
+                    912.0,
+                    857.0,
+                    912.0,
+                    857.0,
+                    936.0,
+                    814.0,
+                    936.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    922.0,
+                    912.0,
+                    945.0,
+                    912.0,
+                    945.0,
+                    931.0,
+                    922.0,
+                    931.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    357.0,
+                    946.0,
+                    379.0,
+                    946.0,
+                    379.0,
+                    965.0,
+                    357.0,
+                    965.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    492.0,
+                    954.0,
+                    503.0,
+                    954.0,
+                    503.0,
+                    966.0,
+                    492.0,
+                    966.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    703.0,
+                    946.0,
+                    725.0,
+                    946.0,
+                    725.0,
+                    965.0,
+                    703.0,
+                    965.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1048.0,
+                    938.0,
+                    1071.0,
+                    938.0,
+                    1071.0,
+                    958.0,
+                    1048.0,
+                    958.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    607.0,
+                    970.0,
+                    643.0,
+                    970.0,
+                    643.0,
+                    998.0,
+                    607.0,
+                    998.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    844.0,
+                    964.0,
+                    850.0,
+                    964.0,
+                    850.0,
+                    970.0,
+                    844.0,
+                    970.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    952.0,
+                    970.0,
+                    988.0,
+                    970.0,
+                    988.0,
+                    998.0,
+                    952.0,
+                    998.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1072.0,
+                    966.0,
+                    1086.0,
+                    966.0,
+                    1086.0,
+                    981.0,
+                    1072.0,
+                    981.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1299.0,
+                    970.0,
+                    1335.0,
+                    970.0,
+                    1335.0,
+                    998.0,
+                    1299.0,
+                    998.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    357.0,
+                    994.0,
+                    379.0,
+                    994.0,
+                    379.0,
+                    1014.0,
+                    357.0,
+                    1014.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    382.0,
+                    994.0,
+                    393.0,
+                    994.0,
+                    393.0,
+                    1006.0,
+                    382.0,
+                    1006.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    577.0,
+                    1003.0,
+                    590.0,
+                    1003.0,
+                    590.0,
+                    1012.0,
+                    577.0,
+                    1012.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    601.0,
+                    994.0,
+                    651.0,
+                    994.0,
+                    651.0,
+                    1023.0,
+                    601.0,
+                    1023.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    700.0,
+                    993.0,
+                    740.0,
+                    993.0,
+                    740.0,
+                    1019.0,
+                    700.0,
+                    1019.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    947.0,
+                    993.0,
+                    998.0,
+                    993.0,
+                    998.0,
+                    1023.0,
+                    947.0,
+                    1023.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1047.0,
+                    989.0,
+                    1073.0,
+                    989.0,
+                    1073.0,
+                    1012.0,
+                    1047.0,
+                    1012.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1261.0,
+                    999.0,
+                    1289.0,
+                    999.0,
+                    1289.0,
+                    1016.0,
+                    1261.0,
+                    1016.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1295.0,
+                    993.0,
+                    1344.0,
+                    993.0,
+                    1344.0,
+                    1023.0,
+                    1295.0,
+                    1023.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    606.0,
+                    1019.0,
+                    660.0,
+                    1019.0,
+                    660.0,
+                    1049.0,
+                    606.0,
+                    1049.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    951.0,
+                    1020.0,
+                    1006.0,
+                    1020.0,
+                    1006.0,
+                    1049.0,
+                    951.0,
+                    1049.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1298.0,
+                    1019.0,
+                    1351.0,
+                    1019.0,
+                    1351.0,
+                    1049.0,
+                    1298.0,
+                    1049.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    364.0,
+                    1046.0,
+                    379.0,
+                    1046.0,
+                    379.0,
+                    1062.0,
+                    364.0,
+                    1062.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    710.0,
+                    1046.0,
+                    725.0,
+                    1046.0,
+                    725.0,
+                    1063.0,
+                    710.0,
+                    1063.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1054.0,
+                    1045.0,
+                    1073.0,
+                    1045.0,
+                    1073.0,
+                    1065.0,
+                    1054.0,
+                    1065.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    379.0,
+                    1056.0,
+                    394.0,
+                    1056.0,
+                    394.0,
+                    1073.0,
+                    379.0,
+                    1073.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    468.0,
+                    1054.0,
+                    495.0,
+                    1054.0,
+                    495.0,
+                    1077.0,
+                    468.0,
+                    1077.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    562.0,
+                    1054.0,
+                    588.0,
+                    1054.0,
+                    588.0,
+                    1077.0,
+                    562.0,
+                    1077.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    656.0,
+                    1055.0,
+                    681.0,
+                    1055.0,
+                    681.0,
+                    1074.0,
+                    656.0,
+                    1074.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    723.0,
+                    1055.0,
+                    742.0,
+                    1055.0,
+                    742.0,
+                    1076.0,
+                    723.0,
+                    1076.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    814.0,
+                    1054.0,
+                    841.0,
+                    1054.0,
+                    841.0,
+                    1077.0,
+                    814.0,
+                    1077.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    907.0,
+                    1054.0,
+                    934.0,
+                    1054.0,
+                    934.0,
+                    1077.0,
+                    907.0,
+                    1077.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1001.0,
+                    1054.0,
+                    1028.0,
+                    1054.0,
+                    1028.0,
+                    1077.0,
+                    1001.0,
+                    1077.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1071.0,
+                    1057.0,
+                    1087.0,
+                    1057.0,
+                    1087.0,
+                    1073.0,
+                    1071.0,
+                    1073.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1160.0,
+                    1054.0,
+                    1186.0,
+                    1054.0,
+                    1186.0,
+                    1077.0,
+                    1160.0,
+                    1077.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1256.0,
+                    1055.0,
+                    1278.0,
+                    1055.0,
+                    1278.0,
+                    1076.0,
+                    1256.0,
+                    1076.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1348.0,
+                    1055.0,
+                    1372.0,
+                    1055.0,
+                    1372.0,
+                    1074.0,
+                    1348.0,
+                    1074.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    652.0,
+                    1071.0,
+                    1112.0,
+                    1071.0,
+                    1112.0,
+                    1101.0,
+                    652.0,
+                    1101.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    225.0,
+                    667.0,
+                    225.0,
+                    667.0,
+                    255.0,
+                    294.0,
+                    255.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    707.0,
+                    225.0,
+                    1407.0,
+                    225.0,
+                    1407.0,
+                    255.0,
+                    707.0,
+                    255.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    250.0,
+                    753.0,
+                    250.0,
+                    753.0,
+                    285.0,
+                    293.0,
+                    285.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    777.0,
+                    250.0,
+                    1406.0,
+                    250.0,
+                    1406.0,
+                    285.0,
+                    777.0,
+                    285.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    279.0,
+                    1407.0,
+                    279.0,
+                    1407.0,
+                    312.0,
+                    294.0,
+                    312.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    306.0,
+                    887.0,
+                    306.0,
+                    887.0,
+                    340.0,
+                    294.0,
+                    340.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    912.0,
+                    306.0,
+                    1403.0,
+                    306.0,
+                    1403.0,
+                    340.0,
+                    912.0,
+                    340.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    334.0,
+                    1410.0,
+                    334.0,
+                    1410.0,
+                    368.0,
+                    294.0,
+                    368.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    364.0,
+                    961.0,
+                    364.0,
+                    961.0,
+                    394.0,
+                    295.0,
+                    394.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    73.0,
+                    815.0,
+                    73.0,
+                    815.0,
+                    108.0,
+                    295.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    840.0,
+                    2087.0,
+                    861.0,
+                    2087.0,
+                    861.0,
+                    2118.0,
+                    840.0,
+                    2118.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    654.0,
+                    1069.0,
+                    1112.0,
+                    1069.0,
+                    1112.0,
+                    1105.0,
+                    654.0,
+                    1105.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1408.0,
+                    1404.0,
+                    1408.0,
+                    1404.0,
+                    1439.0,
+                    296.0,
+                    1439.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1438.0,
+                    1404.0,
+                    1438.0,
+                    1404.0,
+                    1470.0,
+                    293.0,
+                    1470.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1468.0,
+                    1407.0,
+                    1468.0,
+                    1407.0,
+                    1503.0,
+                    295.0,
+                    1503.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1496.0,
+                    635.0,
+                    1496.0,
+                    635.0,
+                    1534.0,
+                    293.0,
+                    1534.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    712.0,
+                    1496.0,
+                    1404.0,
+                    1496.0,
+                    1404.0,
+                    1534.0,
+                    712.0,
+                    1534.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1526.0,
+                    1406.0,
+                    1526.0,
+                    1406.0,
+                    1565.0,
+                    295.0,
+                    1565.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1560.0,
+                    1406.0,
+                    1560.0,
+                    1406.0,
+                    1594.0,
+                    293.0,
+                    1594.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1589.0,
+                    1406.0,
+                    1589.0,
+                    1406.0,
+                    1626.0,
+                    293.0,
+                    1626.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1622.0,
+                    1404.0,
+                    1622.0,
+                    1404.0,
+                    1653.0,
+                    296.0,
+                    1653.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1650.0,
+                    1406.0,
+                    1650.0,
+                    1406.0,
+                    1686.0,
+                    293.0,
+                    1686.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1681.0,
+                    1063.0,
+                    1681.0,
+                    1063.0,
+                    1716.0,
+                    295.0,
+                    1716.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1140.0,
+                    1681.0,
+                    1169.0,
+                    1681.0,
+                    1169.0,
+                    1716.0,
+                    1140.0,
+                    1716.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1245.0,
+                    1681.0,
+                    1409.0,
+                    1681.0,
+                    1409.0,
+                    1716.0,
+                    1245.0,
+                    1716.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1709.0,
+                    464.0,
+                    1709.0,
+                    464.0,
+                    1749.0,
+                    292.0,
+                    1749.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    532.0,
+                    1709.0,
+                    1407.0,
+                    1709.0,
+                    1407.0,
+                    1749.0,
+                    532.0,
+                    1749.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1742.0,
+                    1052.0,
+                    1742.0,
+                    1052.0,
+                    1777.0,
+                    295.0,
+                    1777.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1129.0,
+                    1742.0,
+                    1406.0,
+                    1742.0,
+                    1406.0,
+                    1777.0,
+                    1129.0,
+                    1777.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1774.0,
+                    1404.0,
+                    1774.0,
+                    1404.0,
+                    1805.0,
+                    296.0,
+                    1805.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1805.0,
+                    1406.0,
+                    1805.0,
+                    1406.0,
+                    1837.0,
+                    296.0,
+                    1837.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1835.0,
+                    526.0,
+                    1835.0,
+                    526.0,
+                    1866.0,
+                    295.0,
+                    1866.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1878.0,
+                    1408.0,
+                    1878.0,
+                    1408.0,
+                    1914.0,
+                    294.0,
+                    1914.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1907.0,
+                    1405.0,
+                    1907.0,
+                    1405.0,
+                    1946.0,
+                    295.0,
+                    1946.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    337.0,
+                    1940.0,
+                    368.0,
+                    1940.0,
+                    368.0,
+                    1976.0,
+                    337.0,
+                    1976.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    424.0,
+                    1940.0,
+                    1408.0,
+                    1940.0,
+                    1408.0,
+                    1976.0,
+                    424.0,
+                    1976.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1972.0,
+                    1404.0,
+                    1972.0,
+                    1404.0,
+                    2005.0,
+                    295.0,
+                    2005.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    2001.0,
+                    1405.0,
+                    2001.0,
+                    1405.0,
+                    2037.0,
+                    293.0,
+                    2037.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1298.0,
+                    932.0,
+                    1298.0,
+                    932.0,
+                    1336.0,
+                    296.0,
+                    1336.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    958.0,
+                    1298.0,
+                    1404.0,
+                    1298.0,
+                    1404.0,
+                    1336.0,
+                    958.0,
+                    1336.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1330.0,
+                    1403.0,
+                    1330.0,
+                    1403.0,
+                    1366.0,
+                    294.0,
+                    1366.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1361.0,
+                    420.0,
+                    1361.0,
+                    420.0,
+                    1395.0,
+                    296.0,
+                    1395.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    451.0,
+                    1361.0,
+                    462.0,
+                    1361.0,
+                    462.0,
+                    1395.0,
+                    451.0,
+                    1395.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    558.0,
+                    1361.0,
+                    777.0,
+                    1361.0,
+                    777.0,
+                    1395.0,
+                    558.0,
+                    1395.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    225.0,
+                    667.0,
+                    225.0,
+                    667.0,
+                    255.0,
+                    294.0,
+                    255.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    707.0,
+                    225.0,
+                    1407.0,
+                    225.0,
+                    1407.0,
+                    255.0,
+                    707.0,
+                    255.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    250.0,
+                    753.0,
+                    250.0,
+                    753.0,
+                    285.0,
+                    293.0,
+                    285.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    777.0,
+                    250.0,
+                    1406.0,
+                    250.0,
+                    1406.0,
+                    285.0,
+                    777.0,
+                    285.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    279.0,
+                    1407.0,
+                    279.0,
+                    1407.0,
+                    312.0,
+                    294.0,
+                    312.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    306.0,
+                    887.0,
+                    306.0,
+                    887.0,
+                    340.0,
+                    294.0,
+                    340.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    912.0,
+                    306.0,
+                    1403.0,
+                    306.0,
+                    1403.0,
+                    340.0,
+                    912.0,
+                    340.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    334.0,
+                    1410.0,
+                    334.0,
+                    1410.0,
+                    368.0,
+                    294.0,
+                    368.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    364.0,
+                    961.0,
+                    364.0,
+                    961.0,
+                    394.0,
+                    295.0,
+                    394.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 5,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1496,
+                    1405,
+                    1496,
+                    1405,
+                    1861,
+                    297,
+                    1861
+                ],
+                "score": 0.983
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1142,
+                    1405,
+                    1142,
+                    1405,
+                    1479,
+                    298,
+                    1479
+                ],
+                "score": 0.982
+            },
+            {
+                "category_id": 5,
+                "poly": [
+                    297,
+                    336,
+                    1415,
+                    336,
+                    1415,
+                    562,
+                    297,
+                    562
+                ],
+                "score": 0.963,
+                "html": "<table><tr><td rowspan=\"2\">S</td><td rowspan=\"2\">T</td><td colspan=\"3\">VOC Obj. Det.</td><td colspan=\"3\">COCO Obj. Det.</td><td colspan=\"3\">COCO Inst. Segm.</td></tr><tr><td>Apbb</td><td>AP</td><td>AP</td><td>Apbb</td><td>AP</td><td>APP</td><td>Apmk</td><td>AP</td><td>AP</td></tr><tr><td></td><td>X</td><td>46.1</td><td>74.5</td><td>48.6</td><td>35.0</td><td>53.9</td><td>37.7</td><td>31.0</td><td>51.1</td><td>33.1</td></tr><tr><td></td><td>R-50</td><td>46.1(0.0)</td><td>74.8(+0.3)</td><td>49.1(+0.5)</td><td>35.3(+0.3)</td><td>54.2(+0.3)</td><td>37.8(+0.1)</td><td>31.1(+0.1)</td><td>51.1(0.0)</td><td>33.2(+0.1)</td></tr><tr><td>R-18 R-101</td><td></td><td>46.8(+0.7)</td><td>75.8(+1.3)</td><td>49.3(+0.7)</td><td>35.3(+0.3)</td><td>54.3(+0.4)</td><td>37.9(+0.2)</td><td>31.3(+0.3)</td><td>51.3(+0.2)</td><td>33.4(+0.3)</td></tr><tr><td></td><td>R-152</td><td>46.8(+0.7)</td><td>75.9(+1.4)</td><td>50.2(+1.6)</td><td>35.4(+0.4)</td><td>54.4(+0.5)</td><td>38.0(+0.3)</td><td>31.3(+0.3)</td><td>51.4(+0.3)</td><td>33.4(+0.3)</td></tr></table>"
+            },
+            {
+                "category_id": 6,
+                "poly": [
+                    296,
+                    192,
+                    1406,
+                    192,
+                    1406,
+                    308,
+                    296,
+                    308
+                ],
+                "score": 0.961
+            },
+            {
+                "category_id": 4,
+                "poly": [
+                    298,
+                    905,
+                    1405,
+                    905,
+                    1405,
+                    989,
+                    298,
+                    989
+                ],
+                "score": 0.961
+            },
+            {
+                "category_id": 3,
+                "poly": [
+                    299,
+                    625,
+                    1399,
+                    625,
+                    1399,
+                    893,
+                    299,
+                    893
+                ],
+                "score": 0.96
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    1037,
+                    1405,
+                    1037,
+                    1405,
+                    1126,
+                    299,
+                    1126
+                ],
+                "score": 0.957
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    1972,
+                    1398,
+                    1972,
+                    1398,
+                    2034,
+                    299,
+                    2034
+                ],
+                "score": 0.945
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    300,
+                    1910,
+                    580,
+                    1910,
+                    580,
+                    1941,
+                    300,
+                    1941
+                ],
+                "score": 0.893
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    300,
+                    76,
+                    812,
+                    76,
+                    812,
+                    104,
+                    300,
+                    104
+                ],
+                "score": 0.891
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    842,
+                    2088,
+                    857,
+                    2088,
+                    857,
+                    2111,
+                    842,
+                    2111
+                ],
+                "score": 0.732
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    745,
+                    1739,
+                    806,
+                    1739,
+                    806,
+                    1770,
+                    745,
+                    1770
+                ],
+                "score": 0.9,
+                "latex": "\\mathsf { A P } _ { 7 5 }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    540,
+                    1739,
+                    602,
+                    1739,
+                    602,
+                    1770,
+                    540,
+                    1770
+                ],
+                "score": 0.9,
+                "latex": "\\mathrm { { A P } _ { 5 0 } }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    822,
+                    1770,
+                    904,
+                    1770,
+                    904,
+                    1799,
+                    822,
+                    1799
+                ],
+                "score": 0.89,
+                "latex": "{ \\sim } 1 1 8 \\mathrm { k }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1314,
+                    1770,
+                    1403,
+                    1770,
+                    1403,
+                    1799,
+                    1314,
+                    1799
+                ],
+                "score": 0.88,
+                "latex": "{ \\sim } 1 6 . 5 \\mathrm { k }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    997,
+                    220,
+                    1064,
+                    220,
+                    1064,
+                    252,
+                    997,
+                    252
+                ],
+                "score": 0.85,
+                "latex": "( \\mathbf { A P } ^ { b b } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    652,
+                    1739,
+                    708,
+                    1739,
+                    708,
+                    1769,
+                    652,
+                    1769
+                ],
+                "score": 0.85,
+                "latex": "+ 1 . 6"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1215,
+                    221,
+                    1295,
+                    221,
+                    1295,
+                    252,
+                    1215,
+                    252
+                ],
+                "score": 0.85,
+                "latex": "( \\mathbf { A P } ^ { m k } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1057,
+                    1710,
+                    1111,
+                    1710,
+                    1111,
+                    1738,
+                    1057,
+                    1738
+                ],
+                "score": 0.83,
+                "latex": "+ 0 . 7"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    293,
+                    1739,
+                    348,
+                    1739,
+                    348,
+                    1768,
+                    293,
+                    1768
+                ],
+                "score": 0.82,
+                "latex": "+ 1 . 4"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1181,
+                    1526,
+                    1256,
+                    1526,
+                    1256,
+                    1555,
+                    1181,
+                    1555
+                ],
+                "score": 0.79,
+                "latex": "^ { 0 7 + 1 2 }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    640,
+                    907,
+                    677,
+                    907,
+                    677,
+                    934,
+                    640,
+                    934
+                ],
+                "score": 0.73,
+                "latex": "( \\% )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1311,
+                    1531,
+                    1328,
+                    1531,
+                    1328,
+                    1552,
+                    1311,
+                    1552
+                ],
+                "score": 0.32,
+                "latex": "^ +"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    188.0,
+                    1406.0,
+                    188.0,
+                    1406.0,
+                    227.0,
+                    293.0,
+                    227.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    215.0,
+                    996.0,
+                    215.0,
+                    996.0,
+                    257.0,
+                    294.0,
+                    257.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1065.0,
+                    215.0,
+                    1214.0,
+                    215.0,
+                    1214.0,
+                    257.0,
+                    1065.0,
+                    257.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1296.0,
+                    215.0,
+                    1407.0,
+                    215.0,
+                    1407.0,
+                    257.0,
+                    1296.0,
+                    257.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    246.0,
+                    1408.0,
+                    246.0,
+                    1408.0,
+                    286.0,
+                    293.0,
+                    286.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    277.0,
+                    910.0,
+                    277.0,
+                    910.0,
+                    311.0,
+                    294.0,
+                    311.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    904.0,
+                    639.0,
+                    904.0,
+                    639.0,
+                    937.0,
+                    296.0,
+                    937.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    678.0,
+                    904.0,
+                    1405.0,
+                    904.0,
+                    1405.0,
+                    937.0,
+                    678.0,
+                    937.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    930.0,
+                    1406.0,
+                    930.0,
+                    1406.0,
+                    964.0,
+                    293.0,
+                    964.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    961.0,
+                    520.0,
+                    961.0,
+                    520.0,
+                    991.0,
+                    296.0,
+                    991.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    463.0,
+                    629.0,
+                    570.0,
+                    629.0,
+                    570.0,
+                    654.0,
+                    463.0,
+                    654.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    821.0,
+                    626.0,
+                    943.0,
+                    626.0,
+                    943.0,
+                    658.0,
+                    821.0,
+                    658.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1194.0,
+                    630.0,
+                    1291.0,
+                    630.0,
+                    1291.0,
+                    659.0,
+                    1194.0,
+                    659.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    647.0,
+                    340.0,
+                    647.0,
+                    340.0,
+                    856.0,
+                    296.0,
+                    856.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    676.0,
+                    647.0,
+                    723.0,
+                    647.0,
+                    723.0,
+                    673.0,
+                    676.0,
+                    673.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1055.0,
+                    667.0,
+                    1083.0,
+                    667.0,
+                    1083.0,
+                    693.0,
+                    1055.0,
+                    693.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    329.0,
+                    681.0,
+                    356.0,
+                    681.0,
+                    356.0,
+                    705.0,
+                    329.0,
+                    705.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    676.0,
+                    687.0,
+                    722.0,
+                    687.0,
+                    722.0,
+                    716.0,
+                    676.0,
+                    716.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1055.0,
+                    708.0,
+                    1083.0,
+                    708.0,
+                    1083.0,
+                    733.0,
+                    1055.0,
+                    733.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1296.0,
+                    717.0,
+                    1307.0,
+                    717.0,
+                    1307.0,
+                    731.0,
+                    1296.0,
+                    731.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    332.0,
+                    730.0,
+                    355.0,
+                    730.0,
+                    355.0,
+                    751.0,
+                    332.0,
+                    751.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    678.0,
+                    730.0,
+                    722.0,
+                    730.0,
+                    722.0,
+                    758.0,
+                    678.0,
+                    758.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1002.0,
+                    731.0,
+                    1023.0,
+                    731.0,
+                    1023.0,
+                    751.0,
+                    1002.0,
+                    751.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    359.0,
+                    744.0,
+                    379.0,
+                    744.0,
+                    379.0,
+                    759.0,
+                    359.0,
+                    759.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1055.0,
+                    749.0,
+                    1104.0,
+                    749.0,
+                    1104.0,
+                    774.0,
+                    1055.0,
+                    774.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    329.0,
+                    773.0,
+                    374.0,
+                    773.0,
+                    374.0,
+                    865.0,
+                    329.0,
+                    865.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    449.0,
+                    801.0,
+                    490.0,
+                    801.0,
+                    490.0,
+                    866.0,
+                    449.0,
+                    866.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    549.0,
+                    779.0,
+                    584.0,
+                    779.0,
+                    584.0,
+                    865.0,
+                    549.0,
+                    865.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    644.0,
+                    842.0,
+                    674.0,
+                    842.0,
+                    674.0,
+                    866.0,
+                    644.0,
+                    866.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    676.0,
+                    771.0,
+                    738.0,
+                    771.0,
+                    738.0,
+                    848.0,
+                    676.0,
+                    848.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    810.0,
+                    785.0,
+                    853.0,
+                    785.0,
+                    853.0,
+                    866.0,
+                    810.0,
+                    866.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    909.0,
+                    778.0,
+                    950.0,
+                    778.0,
+                    950.0,
+                    866.0,
+                    909.0,
+                    866.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1008.0,
+                    843.0,
+                    1038.0,
+                    843.0,
+                    1038.0,
+                    868.0,
+                    1008.0,
+                    868.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1054.0,
+                    791.0,
+                    1104.0,
+                    791.0,
+                    1104.0,
+                    865.0,
+                    1054.0,
+                    865.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1172.0,
+                    841.0,
+                    1206.0,
+                    841.0,
+                    1206.0,
+                    869.0,
+                    1172.0,
+                    869.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1209.0,
+                    810.0,
+                    1234.0,
+                    810.0,
+                    1234.0,
+                    829.0,
+                    1209.0,
+                    829.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1258.0,
+                    772.0,
+                    1400.0,
+                    772.0,
+                    1400.0,
+                    868.0,
+                    1258.0,
+                    868.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    639.0,
+                    863.0,
+                    1123.0,
+                    863.0,
+                    1123.0,
+                    894.0,
+                    639.0,
+                    894.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    716.0,
+                    706.5,
+                    766.0,
+                    706.5,
+                    766.0,
+                    728.5,
+                    716.0,
+                    728.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    714.0,
+                    843.0,
+                    742.0,
+                    843.0,
+                    742.0,
+                    866.5,
+                    714.0,
+                    866.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1908.0,
+                    585.0,
+                    1908.0,
+                    585.0,
+                    1944.0,
+                    296.0,
+                    1944.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    73.0,
+                    815.0,
+                    73.0,
+                    815.0,
+                    108.0,
+                    295.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    839.0,
+                    2085.0,
+                    861.0,
+                    2085.0,
+                    861.0,
+                    2120.0,
+                    839.0,
+                    2120.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1494.0,
+                    1406.0,
+                    1494.0,
+                    1406.0,
+                    1528.0,
+                    295.0,
+                    1528.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1525.0,
+                    1180.0,
+                    1525.0,
+                    1180.0,
+                    1559.0,
+                    294.0,
+                    1559.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1257.0,
+                    1525.0,
+                    1310.0,
+                    1525.0,
+                    1310.0,
+                    1559.0,
+                    1257.0,
+                    1559.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1329.0,
+                    1525.0,
+                    1406.0,
+                    1525.0,
+                    1406.0,
+                    1559.0,
+                    1329.0,
+                    1559.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1555.0,
+                    1405.0,
+                    1555.0,
+                    1405.0,
+                    1589.0,
+                    295.0,
+                    1589.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1586.0,
+                    1405.0,
+                    1586.0,
+                    1405.0,
+                    1619.0,
+                    295.0,
+                    1619.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1617.0,
+                    1405.0,
+                    1617.0,
+                    1405.0,
+                    1651.0,
+                    294.0,
+                    1651.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1647.0,
+                    1406.0,
+                    1647.0,
+                    1406.0,
+                    1681.0,
+                    295.0,
+                    1681.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1678.0,
+                    1406.0,
+                    1678.0,
+                    1406.0,
+                    1712.0,
+                    295.0,
+                    1712.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1708.0,
+                    1056.0,
+                    1708.0,
+                    1056.0,
+                    1741.0,
+                    295.0,
+                    1741.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1112.0,
+                    1708.0,
+                    1407.0,
+                    1708.0,
+                    1407.0,
+                    1741.0,
+                    1112.0,
+                    1741.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    349.0,
+                    1735.0,
+                    539.0,
+                    1735.0,
+                    539.0,
+                    1774.0,
+                    349.0,
+                    1774.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    603.0,
+                    1735.0,
+                    651.0,
+                    1735.0,
+                    651.0,
+                    1774.0,
+                    603.0,
+                    1774.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    709.0,
+                    1735.0,
+                    744.0,
+                    1735.0,
+                    744.0,
+                    1774.0,
+                    709.0,
+                    1774.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    807.0,
+                    1735.0,
+                    1407.0,
+                    1735.0,
+                    1407.0,
+                    1774.0,
+                    807.0,
+                    1774.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1768.0,
+                    821.0,
+                    1768.0,
+                    821.0,
+                    1802.0,
+                    295.0,
+                    1802.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    905.0,
+                    1768.0,
+                    1313.0,
+                    1768.0,
+                    1313.0,
+                    1802.0,
+                    905.0,
+                    1802.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1800.0,
+                    1405.0,
+                    1800.0,
+                    1405.0,
+                    1832.0,
+                    294.0,
+                    1832.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1830.0,
+                    470.0,
+                    1830.0,
+                    470.0,
+                    1865.0,
+                    295.0,
+                    1865.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1142.0,
+                    1407.0,
+                    1142.0,
+                    1407.0,
+                    1176.0,
+                    295.0,
+                    1176.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1175.0,
+                    1405.0,
+                    1175.0,
+                    1405.0,
+                    1206.0,
+                    296.0,
+                    1206.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1204.0,
+                    1406.0,
+                    1204.0,
+                    1406.0,
+                    1238.0,
+                    295.0,
+                    1238.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1236.0,
+                    1403.0,
+                    1236.0,
+                    1403.0,
+                    1267.0,
+                    295.0,
+                    1267.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1266.0,
+                    1405.0,
+                    1266.0,
+                    1405.0,
+                    1297.0,
+                    296.0,
+                    1297.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1293.0,
+                    1407.0,
+                    1293.0,
+                    1407.0,
+                    1330.0,
+                    293.0,
+                    1330.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1323.0,
+                    1405.0,
+                    1323.0,
+                    1405.0,
+                    1363.0,
+                    293.0,
+                    1363.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1356.0,
+                    1405.0,
+                    1356.0,
+                    1405.0,
+                    1391.0,
+                    295.0,
+                    1391.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1386.0,
+                    1407.0,
+                    1386.0,
+                    1407.0,
+                    1421.0,
+                    293.0,
+                    1421.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1417.0,
+                    1405.0,
+                    1417.0,
+                    1405.0,
+                    1452.0,
+                    296.0,
+                    1452.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1448.0,
+                    777.0,
+                    1448.0,
+                    777.0,
+                    1482.0,
+                    293.0,
+                    1482.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1034.0,
+                    1405.0,
+                    1034.0,
+                    1405.0,
+                    1070.0,
+                    293.0,
+                    1070.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1067.0,
+                    1405.0,
+                    1067.0,
+                    1405.0,
+                    1100.0,
+                    295.0,
+                    1100.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1096.0,
+                    404.0,
+                    1096.0,
+                    404.0,
+                    1130.0,
+                    293.0,
+                    1130.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1966.0,
+                    1404.0,
+                    1966.0,
+                    1404.0,
+                    2011.0,
+                    294.0,
+                    2011.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    2000.0,
+                    1007.0,
+                    2000.0,
+                    1007.0,
+                    2038.0,
+                    295.0,
+                    2038.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 6,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1295,
+                    1405,
+                    1295,
+                    1405,
+                    1784,
+                    297,
+                    1784
+                ],
+                "score": 0.984
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1034,
+                    1405,
+                    1034,
+                    1405,
+                    1280,
+                    298,
+                    1280
+                ],
+                "score": 0.98
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1799,
+                    1404,
+                    1799,
+                    1404,
+                    2045,
+                    298,
+                    2045
+                ],
+                "score": 0.979
+            },
+            {
+                "category_id": 5,
+                "poly": [
+                    308,
+                    473,
+                    812,
+                    473,
+                    812,
+                    758,
+                    308,
+                    758
+                ],
+                "score": 0.979,
+                "html": "<table><tr><td>Teacher</td><td></td><td>P-E D-E T.Top-1 S.Top-1 S.Top-5</td></tr><tr><td></td><td>X X</td><td>X 52.5 77.0</td></tr><tr><td>X</td><td>200</td><td></td></tr><tr><td>MoCo</td><td>200</td><td>60.6 52.1 77.0</td></tr><tr><td>SimCLR</td><td>200 200</td><td>65.6 57.5 81.7</td></tr><tr><td>MoCo-v2 200 800</td><td>200 67.4</td><td>57.9 82.0</td></tr><tr><td>SWAV 800</td><td>200 71.1</td><td>60.5 83.5</td></tr><tr><td></td><td>100 75.3</td><td>61.1 83.8</td></tr><tr><td>800</td><td>200 75.3</td><td>61.7 84.2</td></tr><tr><td>800</td><td>400 75.3</td><td>62.0 84.4</td></tr><tr><td>SWAV* 800</td><td>200 75.3</td><td>62.6 84.8</td></tr></table>"
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    835,
+                    1404,
+                    835,
+                    1404,
+                    1019,
+                    298,
+                    1019
+                ],
+                "score": 0.975
+            },
+            {
+                "category_id": 3,
+                "poly": [
+                    901,
+                    262,
+                    1398,
+                    262,
+                    1398,
+                    635,
+                    901,
+                    635
+                ],
+                "score": 0.971
+            },
+            {
+                "category_id": 4,
+                "poly": [
+                    889,
+                    641,
+                    1403,
+                    641,
+                    1403,
+                    752,
+                    889,
+                    752
+                ],
+                "score": 0.96
+            },
+            {
+                "category_id": 6,
+                "poly": [
+                    297,
+                    239,
+                    833,
+                    239,
+                    833,
+                    463,
+                    297,
+                    463
+                ],
+                "score": 0.948
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    300,
+                    76,
+                    812,
+                    76,
+                    812,
+                    104,
+                    300,
+                    104
+                ],
+                "score": 0.892
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    841,
+                    2089,
+                    858,
+                    2089,
+                    858,
+                    2111,
+                    841,
+                    2111
+                ],
+                "score": 0.794
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1116,
+                    1631,
+                    1192,
+                    1631,
+                    1192,
+                    1660,
+                    1116,
+                    1660
+                ],
+                "score": 0.92,
+                "latex": "5 7 . 6 \\%"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    856,
+                    1889,
+                    890,
+                    1889,
+                    890,
+                    1925,
+                    856,
+                    1925
+                ],
+                "score": 0.9,
+                "latex": "{ \\mathbf z } _ { i } ^ { T }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    723,
+                    1889,
+                    756,
+                    1889,
+                    756,
+                    1929,
+                    723,
+                    1929
+                ],
+                "score": 0.9,
+                "latex": "p _ { j } ^ { T }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    363,
+                    1981,
+                    398,
+                    1981,
+                    398,
+                    2016,
+                    363,
+                    2016
+                ],
+                "score": 0.9,
+                "latex": "{ \\mathbf z } _ { i } ^ { T }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1033,
+                    1892,
+                    1065,
+                    1892,
+                    1065,
+                    1925,
+                    1033,
+                    1925
+                ],
+                "score": 0.89,
+                "latex": "{ \\bf d } _ { j }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1204,
+                    1890,
+                    1250,
+                    1890,
+                    1250,
+                    1920,
+                    1204,
+                    1920
+                ],
+                "score": 0.89,
+                "latex": "\\mathbf { D } ^ { + }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    630,
+                    1631,
+                    706,
+                    1631,
+                    706,
+                    1660,
+                    630,
+                    1660
+                ],
+                "score": 0.88,
+                "latex": "5 7 . 9 \\%"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1238,
+                    1827,
+                    1275,
+                    1827,
+                    1275,
+                    1858,
+                    1238,
+                    1858
+                ],
+                "score": 0.88,
+                "latex": "\\tau ^ { T }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    634,
+                    1952,
+                    668,
+                    1952,
+                    668,
+                    1979,
+                    634,
+                    1979
+                ],
+                "score": 0.88,
+                "latex": "\\tau ^ { T }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    611,
+                    1827,
+                    647,
+                    1827,
+                    647,
+                    1858,
+                    611,
+                    1858
+                ],
+                "score": 0.87,
+                "latex": "\\tau ^ { T }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1238,
+                    1866,
+                    1258,
+                    1866,
+                    1258,
+                    1888,
+                    1238,
+                    1888
+                ],
+                "score": 0.78,
+                "latex": "\\tau"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1235,
+                    1987,
+                    1255,
+                    1987,
+                    1255,
+                    2010,
+                    1235,
+                    2010
+                ],
+                "score": 0.76,
+                "latex": "\\tau"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    648,
+                    1359,
+                    672,
+                    1359,
+                    672,
+                    1384,
+                    648,
+                    1384
+                ],
+                "score": 0.71,
+                "latex": "K"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    463,
+                    928,
+                    530,
+                    928,
+                    530,
+                    958,
+                    463,
+                    958
+                ],
+                "score": 0.68,
+                "latex": "{ 5 0 \\times 2 }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    391,
+                    1358,
+                    420,
+                    1358,
+                    420,
+                    1384,
+                    391,
+                    1384
+                ],
+                "score": 0.67,
+                "latex": "K"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    411,
+                    1328,
+                    436,
+                    1328,
+                    436,
+                    1354,
+                    411,
+                    1354
+                ],
+                "score": 0.66,
+                "latex": "l 2"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1130,
+                    642,
+                    1169,
+                    642,
+                    1169,
+                    668,
+                    1130,
+                    668
+                ],
+                "score": 0.66,
+                "latex": "( \\% )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    790,
+                    1631,
+                    1010,
+                    1631,
+                    1010,
+                    1660,
+                    790,
+                    1660
+                ],
+                "score": 0.63,
+                "latex": "\\mathbf { S E E D + M o C o - V } 2"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    637,
+                    242,
+                    675,
+                    242,
+                    675,
+                    269,
+                    637,
+                    269
+                ],
+                "score": 0.59,
+                "latex": "( \\% )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    865,
+                    1633,
+                    889,
+                    1633,
+                    889,
+                    1658,
+                    865,
+                    1658
+                ],
+                "score": 0.45,
+                "latex": "^ +"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    699,
+                    1956,
+                    715,
+                    1956,
+                    715,
+                    1979,
+                    699,
+                    1979
+                ],
+                "score": 0.43,
+                "latex": "_ 0"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    419,
+                    1510,
+                    444,
+                    1510,
+                    444,
+                    1537,
+                    419,
+                    1537
+                ],
+                "score": 0.38,
+                "latex": "l 2"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    924,
+                    1663,
+                    972,
+                    1663,
+                    972,
+                    1690,
+                    924,
+                    1690
+                ],
+                "score": 0.3,
+                "latex": "S S L"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    384,
+                    1250,
+                    432,
+                    1250,
+                    432,
+                    1277,
+                    384,
+                    1277
+                ],
+                "score": 0.25,
+                "latex": "S S L"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    941.0,
+                    271.0,
+                    975.0,
+                    271.0,
+                    975.0,
+                    304.0,
+                    941.0,
+                    304.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1211.0,
+                    270.0,
+                    1397.0,
+                    270.0,
+                    1397.0,
+                    308.0,
+                    1211.0,
+                    308.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    902.0,
+                    286.0,
+                    939.0,
+                    286.0,
+                    939.0,
+                    589.0,
+                    902.0,
+                    589.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1052.0,
+                    286.0,
+                    1163.0,
+                    286.0,
+                    1163.0,
+                    318.0,
+                    1052.0,
+                    318.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1050.0,
+                    317.0,
+                    1165.0,
+                    317.0,
+                    1165.0,
+                    356.0,
+                    1050.0,
+                    356.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    942.0,
+                    349.0,
+                    974.0,
+                    349.0,
+                    974.0,
+                    377.0,
+                    942.0,
+                    377.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1014.0,
+                    386.0,
+                    1143.0,
+                    386.0,
+                    1143.0,
+                    425.0,
+                    1014.0,
+                    425.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    940.0,
+                    421.0,
+                    975.0,
+                    421.0,
+                    975.0,
+                    451.0,
+                    940.0,
+                    451.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    940.0,
+                    493.0,
+                    975.0,
+                    493.0,
+                    975.0,
+                    525.0,
+                    940.0,
+                    525.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    940.0,
+                    568.0,
+                    976.0,
+                    568.0,
+                    976.0,
+                    598.0,
+                    940.0,
+                    598.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    896.0,
+                    602.0,
+                    1177.0,
+                    602.0,
+                    1177.0,
+                    638.0,
+                    896.0,
+                    638.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1203.0,
+                    603.0,
+                    1399.0,
+                    603.0,
+                    1399.0,
+                    638.0,
+                    1203.0,
+                    638.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    888.0,
+                    639.0,
+                    1129.0,
+                    639.0,
+                    1129.0,
+                    671.0,
+                    888.0,
+                    671.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1170.0,
+                    639.0,
+                    1405.0,
+                    639.0,
+                    1405.0,
+                    671.0,
+                    1170.0,
+                    671.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    889.0,
+                    666.0,
+                    1407.0,
+                    666.0,
+                    1407.0,
+                    697.0,
+                    889.0,
+                    697.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    888.0,
+                    696.0,
+                    1405.0,
+                    696.0,
+                    1405.0,
+                    725.0,
+                    888.0,
+                    725.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    888.0,
+                    720.0,
+                    1142.0,
+                    720.0,
+                    1142.0,
+                    755.0,
+                    888.0,
+                    755.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    242.0,
+                    636.0,
+                    242.0,
+                    636.0,
+                    270.0,
+                    296.0,
+                    270.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    676.0,
+                    242.0,
+                    834.0,
+                    242.0,
+                    834.0,
+                    270.0,
+                    676.0,
+                    270.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    268.0,
+                    834.0,
+                    268.0,
+                    834.0,
+                    298.0,
+                    295.0,
+                    298.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    295.0,
+                    836.0,
+                    295.0,
+                    836.0,
+                    330.0,
+                    295.0,
+                    330.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    324.0,
+                    833.0,
+                    324.0,
+                    833.0,
+                    355.0,
+                    295.0,
+                    355.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    353.0,
+                    832.0,
+                    353.0,
+                    832.0,
+                    384.0,
+                    295.0,
+                    384.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    379.0,
+                    831.0,
+                    379.0,
+                    831.0,
+                    410.0,
+                    296.0,
+                    410.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    405.0,
+                    832.0,
+                    405.0,
+                    832.0,
+                    438.0,
+                    294.0,
+                    438.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    433.0,
+                    459.0,
+                    433.0,
+                    459.0,
+                    468.0,
+                    294.0,
+                    468.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    73.0,
+                    815.0,
+                    73.0,
+                    815.0,
+                    108.0,
+                    295.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    839.0,
+                    2086.0,
+                    860.0,
+                    2086.0,
+                    860.0,
+                    2116.0,
+                    839.0,
+                    2116.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1292.0,
+                    1408.0,
+                    1292.0,
+                    1408.0,
+                    1330.0,
+                    295.0,
+                    1330.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1324.0,
+                    410.0,
+                    1324.0,
+                    410.0,
+                    1360.0,
+                    292.0,
+                    1360.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    437.0,
+                    1324.0,
+                    1407.0,
+                    1324.0,
+                    1407.0,
+                    1360.0,
+                    437.0,
+                    1360.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1357.0,
+                    390.0,
+                    1357.0,
+                    390.0,
+                    1391.0,
+                    295.0,
+                    1391.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    421.0,
+                    1357.0,
+                    647.0,
+                    1357.0,
+                    647.0,
+                    1391.0,
+                    421.0,
+                    1391.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    673.0,
+                    1357.0,
+                    1406.0,
+                    1357.0,
+                    1406.0,
+                    1391.0,
+                    673.0,
+                    1391.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1385.0,
+                    1405.0,
+                    1385.0,
+                    1405.0,
+                    1423.0,
+                    292.0,
+                    1423.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1417.0,
+                    1405.0,
+                    1417.0,
+                    1405.0,
+                    1451.0,
+                    295.0,
+                    1451.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1447.0,
+                    1405.0,
+                    1447.0,
+                    1405.0,
+                    1482.0,
+                    294.0,
+                    1482.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1477.0,
+                    1407.0,
+                    1477.0,
+                    1407.0,
+                    1513.0,
+                    294.0,
+                    1513.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1509.0,
+                    418.0,
+                    1509.0,
+                    418.0,
+                    1543.0,
+                    295.0,
+                    1543.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    445.0,
+                    1509.0,
+                    1405.0,
+                    1509.0,
+                    1405.0,
+                    1543.0,
+                    445.0,
+                    1543.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1539.0,
+                    1405.0,
+                    1539.0,
+                    1405.0,
+                    1574.0,
+                    295.0,
+                    1574.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1568.0,
+                    1406.0,
+                    1568.0,
+                    1406.0,
+                    1606.0,
+                    294.0,
+                    1606.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1601.0,
+                    1405.0,
+                    1601.0,
+                    1405.0,
+                    1635.0,
+                    295.0,
+                    1635.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1627.0,
+                    629.0,
+                    1627.0,
+                    629.0,
+                    1665.0,
+                    294.0,
+                    1665.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    707.0,
+                    1627.0,
+                    789.0,
+                    1627.0,
+                    789.0,
+                    1665.0,
+                    707.0,
+                    1665.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1011.0,
+                    1627.0,
+                    1115.0,
+                    1627.0,
+                    1115.0,
+                    1665.0,
+                    1011.0,
+                    1665.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1193.0,
+                    1627.0,
+                    1406.0,
+                    1627.0,
+                    1406.0,
+                    1665.0,
+                    1193.0,
+                    1665.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1658.0,
+                    923.0,
+                    1658.0,
+                    923.0,
+                    1697.0,
+                    294.0,
+                    1697.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    973.0,
+                    1658.0,
+                    1406.0,
+                    1658.0,
+                    1406.0,
+                    1697.0,
+                    973.0,
+                    1697.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1693.0,
+                    1403.0,
+                    1693.0,
+                    1403.0,
+                    1724.0,
+                    296.0,
+                    1724.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1721.0,
+                    1407.0,
+                    1721.0,
+                    1407.0,
+                    1759.0,
+                    294.0,
+                    1759.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1753.0,
+                    738.0,
+                    1753.0,
+                    738.0,
+                    1787.0,
+                    295.0,
+                    1787.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1034.0,
+                    1405.0,
+                    1034.0,
+                    1405.0,
+                    1068.0,
+                    295.0,
+                    1068.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1064.0,
+                    1406.0,
+                    1064.0,
+                    1406.0,
+                    1098.0,
+                    296.0,
+                    1098.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1096.0,
+                    1407.0,
+                    1096.0,
+                    1407.0,
+                    1129.0,
+                    295.0,
+                    1129.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1126.0,
+                    1405.0,
+                    1126.0,
+                    1405.0,
+                    1162.0,
+                    293.0,
+                    1162.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1153.0,
+                    1406.0,
+                    1153.0,
+                    1406.0,
+                    1194.0,
+                    293.0,
+                    1194.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1187.0,
+                    1406.0,
+                    1187.0,
+                    1406.0,
+                    1223.0,
+                    292.0,
+                    1223.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1216.0,
+                    1405.0,
+                    1216.0,
+                    1405.0,
+                    1252.0,
+                    293.0,
+                    1252.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1247.0,
+                    383.0,
+                    1247.0,
+                    383.0,
+                    1283.0,
+                    293.0,
+                    1283.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    433.0,
+                    1247.0,
+                    976.0,
+                    1247.0,
+                    976.0,
+                    1283.0,
+                    433.0,
+                    1283.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1796.0,
+                    1405.0,
+                    1796.0,
+                    1405.0,
+                    1836.0,
+                    293.0,
+                    1836.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1825.0,
+                    610.0,
+                    1825.0,
+                    610.0,
+                    1866.0,
+                    293.0,
+                    1866.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    648.0,
+                    1825.0,
+                    1237.0,
+                    1825.0,
+                    1237.0,
+                    1866.0,
+                    648.0,
+                    1866.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1276.0,
+                    1825.0,
+                    1408.0,
+                    1825.0,
+                    1408.0,
+                    1866.0,
+                    1276.0,
+                    1866.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1860.0,
+                    1237.0,
+                    1860.0,
+                    1237.0,
+                    1893.0,
+                    296.0,
+                    1893.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1259.0,
+                    1860.0,
+                    1404.0,
+                    1860.0,
+                    1404.0,
+                    1893.0,
+                    1259.0,
+                    1893.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1888.0,
+                    722.0,
+                    1888.0,
+                    722.0,
+                    1926.0,
+                    293.0,
+                    1926.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    757.0,
+                    1888.0,
+                    855.0,
+                    1888.0,
+                    855.0,
+                    1926.0,
+                    757.0,
+                    1926.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    891.0,
+                    1888.0,
+                    1032.0,
+                    1888.0,
+                    1032.0,
+                    1926.0,
+                    891.0,
+                    1926.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1066.0,
+                    1888.0,
+                    1203.0,
+                    1888.0,
+                    1203.0,
+                    1926.0,
+                    1066.0,
+                    1926.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1251.0,
+                    1888.0,
+                    1406.0,
+                    1888.0,
+                    1406.0,
+                    1926.0,
+                    1251.0,
+                    1926.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1921.0,
+                    1406.0,
+                    1921.0,
+                    1406.0,
+                    1954.0,
+                    294.0,
+                    1954.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1950.0,
+                    633.0,
+                    1950.0,
+                    633.0,
+                    1986.0,
+                    292.0,
+                    1986.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    669.0,
+                    1950.0,
+                    698.0,
+                    1950.0,
+                    698.0,
+                    1986.0,
+                    669.0,
+                    1986.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    716.0,
+                    1950.0,
+                    1406.0,
+                    1950.0,
+                    1406.0,
+                    1986.0,
+                    716.0,
+                    1986.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1981.0,
+                    362.0,
+                    1981.0,
+                    362.0,
+                    2018.0,
+                    293.0,
+                    2018.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    399.0,
+                    1981.0,
+                    1234.0,
+                    1981.0,
+                    1234.0,
+                    2018.0,
+                    399.0,
+                    2018.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1256.0,
+                    1981.0,
+                    1408.0,
+                    1981.0,
+                    1408.0,
+                    2018.0,
+                    1256.0,
+                    2018.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    2013.0,
+                    1402.0,
+                    2013.0,
+                    1402.0,
+                    2047.0,
+                    296.0,
+                    2047.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    837.0,
+                    1404.0,
+                    837.0,
+                    1404.0,
+                    869.0,
+                    296.0,
+                    869.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    865.0,
+                    1406.0,
+                    865.0,
+                    1406.0,
+                    902.0,
+                    293.0,
+                    902.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    897.0,
+                    1406.0,
+                    897.0,
+                    1406.0,
+                    931.0,
+                    293.0,
+                    931.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    926.0,
+                    462.0,
+                    926.0,
+                    462.0,
+                    963.0,
+                    292.0,
+                    963.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    531.0,
+                    926.0,
+                    1406.0,
+                    926.0,
+                    1406.0,
+                    963.0,
+                    531.0,
+                    963.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    959.0,
+                    1406.0,
+                    959.0,
+                    1406.0,
+                    994.0,
+                    293.0,
+                    994.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    990.0,
+                    534.0,
+                    990.0,
+                    534.0,
+                    1026.0,
+                    294.0,
+                    1026.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 7,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 5,
+                "poly": [
+                    846,
+                    327,
+                    1385,
+                    327,
+                    1385,
+                    585,
+                    846,
+                    585
+                ],
+                "score": 0.981,
+                "html": "<table><tr><td rowspan=\"2\">YT</td><td>ImageNet</td><td>CIFAR-10</td><td>CIFAR-100</td></tr><tr><td>Top-1 Top-5</td><td>Top-1</td><td>Top-1</td></tr><tr><td>0.3</td><td>54.8 80.0</td><td>78.7</td><td>46.6</td></tr><tr><td>0.1</td><td>54.9 80.1</td><td>83.0</td><td>50.1</td></tr><tr><td>0.05</td><td>56.5 81.3</td><td>84.4</td><td>56.2</td></tr><tr><td>0.01</td><td>57.9 82.0</td><td>87.5 </td><td>60.6</td></tr><tr><td>1e-3</td><td>57.6 81.8</td><td>86.9</td><td>60.8</td></tr></table>"
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    735,
+                    1405,
+                    735,
+                    1405,
+                    950,
+                    297,
+                    950
+                ],
+                "score": 0.979
+            },
+            {
+                "category_id": 5,
+                "poly": [
+                    297,
+                    355,
+                    786,
+                    355,
+                    786,
+                    603,
+                    297,
+                    603
+                ],
+                "score": 0.978,
+                "html": "<table><tr><td>Method</td><td colspan=\"2\">Top-1 Acc. Top-5 Acc.</td></tr><tr><td>l2-Distance</td><td>55.3</td><td>80.3</td></tr><tr><td>K-Means</td><td>51.0</td><td>75.8</td></tr><tr><td>Online Clustering</td><td>56.4</td><td>81.2</td></tr><tr><td>Binary Contr. Loss</td><td>57.4</td><td>81.5</td></tr><tr><td>SEED +MoCo-V2</td><td>57.6</td><td>81.8</td></tr><tr><td>SEED</td><td>57.9</td><td>82.0</td></tr></table>"
+            },
+            {
+                "category_id": 6,
+                "poly": [
+                    295,
+                    240,
+                    810,
+                    240,
+                    810,
+                    352,
+                    295,
+                    352
+                ],
+                "score": 0.951
+            },
+            {
+                "category_id": 6,
+                "poly": [
+                    844,
+                    258,
+                    1402,
+                    258,
+                    1402,
+                    319,
+                    844,
+                    319
+                ],
+                "score": 0.94
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    300,
+                    665,
+                    561,
+                    665,
+                    561,
+                    701,
+                    300,
+                    701
+                ],
+                "score": 0.911
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    302,
+                    1052,
+                    1401,
+                    1052,
+                    1401,
+                    1145,
+                    302,
+                    1145
+                ],
+                "score": 0.91
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    300,
+                    76,
+                    812,
+                    76,
+                    812,
+                    104,
+                    300,
+                    104
+                ],
+                "score": 0.895
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    299,
+                    1000,
+                    488,
+                    1000,
+                    488,
+                    1034,
+                    299,
+                    1034
+                ],
+                "score": 0.874
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    295,
+                    1526,
+                    1404,
+                    1526,
+                    1404,
+                    1621,
+                    295,
+                    1621
+                ],
+                "score": 0.87
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    296,
+                    1972,
+                    1401,
+                    1972,
+                    1401,
+                    2034,
+                    296,
+                    2034
+                ],
+                "score": 0.837
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    296,
+                    1644,
+                    1406,
+                    1644,
+                    1406,
+                    1739,
+                    296,
+                    1739
+                ],
+                "score": 0.816
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    296,
+                    1408,
+                    1403,
+                    1408,
+                    1403,
+                    1502,
+                    296,
+                    1502
+                ],
+                "score": 0.803
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    296,
+                    1764,
+                    1403,
+                    1764,
+                    1403,
+                    1828,
+                    296,
+                    1828
+                ],
+                "score": 0.794
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1289,
+                    1407,
+                    1289,
+                    1407,
+                    1383,
+                    297,
+                    1383
+                ],
+                "score": 0.775
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    840,
+                    2088,
+                    858,
+                    2088,
+                    858,
+                    2111,
+                    840,
+                    2111
+                ],
+                "score": 0.773
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1852,
+                    1404,
+                    1852,
+                    1404,
+                    1946,
+                    298,
+                    1946
+                ],
+                "score": 0.773
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    1171,
+                    1406,
+                    1171,
+                    1406,
+                    1265,
+                    299,
+                    1265
+                ],
+                "score": 0.769
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1036,
+                    257,
+                    1069,
+                    257,
+                    1069,
+                    286,
+                    1036,
+                    286
+                ],
+                "score": 0.89,
+                "latex": "\\tau ^ { T }"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    238.0,
+                    813.0,
+                    238.0,
+                    813.0,
+                    272.0,
+                    294.0,
+                    272.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    298.0,
+                    270.0,
+                    809.0,
+                    270.0,
+                    809.0,
+                    297.0,
+                    298.0,
+                    297.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    297.0,
+                    808.0,
+                    297.0,
+                    808.0,
+                    326.0,
+                    296.0,
+                    326.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    325.0,
+                    531.0,
+                    325.0,
+                    531.0,
+                    353.0,
+                    297.0,
+                    353.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    844.0,
+                    256.0,
+                    1035.0,
+                    256.0,
+                    1035.0,
+                    290.0,
+                    844.0,
+                    290.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1070.0,
+                    256.0,
+                    1404.0,
+                    256.0,
+                    1404.0,
+                    290.0,
+                    1070.0,
+                    290.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    845.0,
+                    286.0,
+                    1367.0,
+                    286.0,
+                    1367.0,
+                    321.0,
+                    845.0,
+                    321.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    658.0,
+                    566.0,
+                    658.0,
+                    566.0,
+                    710.0,
+                    291.0,
+                    710.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    73.0,
+                    815.0,
+                    73.0,
+                    815.0,
+                    108.0,
+                    295.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1000.0,
+                    490.0,
+                    1000.0,
+                    490.0,
+                    1037.0,
+                    296.0,
+                    1037.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    839.0,
+                    2087.0,
+                    860.0,
+                    2087.0,
+                    860.0,
+                    2117.0,
+                    839.0,
+                    2117.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    733.0,
+                    1406.0,
+                    733.0,
+                    1406.0,
+                    772.0,
+                    292.0,
+                    772.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    767.0,
+                    1406.0,
+                    767.0,
+                    1406.0,
+                    802.0,
+                    295.0,
+                    802.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    794.0,
+                    1406.0,
+                    794.0,
+                    1406.0,
+                    835.0,
+                    292.0,
+                    835.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    829.0,
+                    1403.0,
+                    829.0,
+                    1403.0,
+                    863.0,
+                    295.0,
+                    863.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    857.0,
+                    1405.0,
+                    857.0,
+                    1405.0,
+                    893.0,
+                    294.0,
+                    893.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    888.0,
+                    1406.0,
+                    888.0,
+                    1406.0,
+                    922.0,
+                    295.0,
+                    922.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    921.0,
+                    1363.0,
+                    921.0,
+                    1363.0,
+                    952.0,
+                    296.0,
+                    952.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1051.0,
+                    1405.0,
+                    1051.0,
+                    1405.0,
+                    1089.0,
+                    296.0,
+                    1089.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    1084.0,
+                    1406.0,
+                    1084.0,
+                    1406.0,
+                    1118.0,
+                    322.0,
+                    1118.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    319.0,
+                    1109.0,
+                    397.0,
+                    1109.0,
+                    397.0,
+                    1150.0,
+                    319.0,
+                    1150.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1527.0,
+                    1406.0,
+                    1527.0,
+                    1406.0,
+                    1561.0,
+                    295.0,
+                    1561.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    324.0,
+                    1559.0,
+                    1405.0,
+                    1559.0,
+                    1405.0,
+                    1593.0,
+                    324.0,
+                    1593.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    1589.0,
+                    610.0,
+                    1589.0,
+                    610.0,
+                    1620.0,
+                    323.0,
+                    1620.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1969.0,
+                    1403.0,
+                    1969.0,
+                    1403.0,
+                    2008.0,
+                    294.0,
+                    2008.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    2002.0,
+                    1023.0,
+                    2002.0,
+                    1023.0,
+                    2037.0,
+                    323.0,
+                    2037.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1644.0,
+                    1406.0,
+                    1644.0,
+                    1406.0,
+                    1682.0,
+                    295.0,
+                    1682.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    1673.0,
+                    1406.0,
+                    1673.0,
+                    1406.0,
+                    1716.0,
+                    322.0,
+                    1716.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    1707.0,
+                    647.0,
+                    1707.0,
+                    647.0,
+                    1743.0,
+                    323.0,
+                    1743.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1404.0,
+                    1408.0,
+                    1404.0,
+                    1408.0,
+                    1447.0,
+                    293.0,
+                    1447.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    1440.0,
+                    1405.0,
+                    1440.0,
+                    1405.0,
+                    1474.0,
+                    323.0,
+                    1474.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    321.0,
+                    1470.0,
+                    648.0,
+                    1470.0,
+                    648.0,
+                    1503.0,
+                    321.0,
+                    1503.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1764.0,
+                    1406.0,
+                    1764.0,
+                    1406.0,
+                    1800.0,
+                    296.0,
+                    1800.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    321.0,
+                    1796.0,
+                    1300.0,
+                    1796.0,
+                    1300.0,
+                    1829.0,
+                    321.0,
+                    1829.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1285.0,
+                    1405.0,
+                    1285.0,
+                    1405.0,
+                    1328.0,
+                    294.0,
+                    1328.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    321.0,
+                    1319.0,
+                    1408.0,
+                    1319.0,
+                    1408.0,
+                    1359.0,
+                    321.0,
+                    1359.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    324.0,
+                    1352.0,
+                    504.0,
+                    1352.0,
+                    504.0,
+                    1382.0,
+                    324.0,
+                    1382.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1849.0,
+                    1405.0,
+                    1849.0,
+                    1405.0,
+                    1891.0,
+                    293.0,
+                    1891.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    1884.0,
+                    1408.0,
+                    1884.0,
+                    1408.0,
+                    1918.0,
+                    323.0,
+                    1918.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    321.0,
+                    1909.0,
+                    412.0,
+                    1909.0,
+                    412.0,
+                    1949.0,
+                    321.0,
+                    1949.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    1172.0,
+                    1404.0,
+                    1172.0,
+                    1404.0,
+                    1206.0,
+                    297.0,
+                    1206.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    1201.0,
+                    1404.0,
+                    1201.0,
+                    1404.0,
+                    1237.0,
+                    322.0,
+                    1237.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    1229.0,
+                    1052.0,
+                    1229.0,
+                    1052.0,
+                    1268.0,
+                    323.0,
+                    1268.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 8,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 2,
+                "poly": [
+                    300,
+                    75,
+                    813,
+                    75,
+                    813,
+                    105,
+                    300,
+                    105
+                ],
+                "score": 0.89
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    836,
+                    2088,
+                    864,
+                    2088,
+                    864,
+                    2113,
+                    836,
+                    2113
+                ],
+                "score": 0.828
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1626,
+                    1405,
+                    1626,
+                    1405,
+                    1723,
+                    297,
+                    1723
+                ],
+                "score": 0.663
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    295,
+                    1542,
+                    1403,
+                    1542,
+                    1403,
+                    1608,
+                    295,
+                    1608
+                ],
+                "score": 0.653
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    296,
+                    885,
+                    1402,
+                    885,
+                    1402,
+                    981,
+                    296,
+                    981
+                ],
+                "score": 0.649
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    294,
+                    656,
+                    1403,
+                    656,
+                    1403,
+                    722,
+                    294,
+                    722
+                ],
+                "score": 0.631
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    296,
+                    1428,
+                    1402,
+                    1428,
+                    1402,
+                    1523,
+                    296,
+                    1523
+                ],
+                "score": 0.612
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    295,
+                    1741,
+                    1403,
+                    1741,
+                    1403,
+                    1838,
+                    295,
+                    1838
+                ],
+                "score": 0.606
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    293,
+                    1855,
+                    1405,
+                    1855,
+                    1405,
+                    1921,
+                    293,
+                    1921
+                ],
+                "score": 0.603
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    295,
+                    1313,
+                    1401,
+                    1313,
+                    1401,
+                    1410,
+                    295,
+                    1410
+                ],
+                "score": 0.596
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    294,
+                    1229,
+                    1400,
+                    1229,
+                    1400,
+                    1295,
+                    294,
+                    1295
+                ],
+                "score": 0.595
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    541,
+                    1403,
+                    541,
+                    1403,
+                    637,
+                    297,
+                    637
+                ],
+                "score": 0.594
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1000,
+                    1401,
+                    1000,
+                    1401,
+                    1097,
+                    297,
+                    1097
+                ],
+                "score": 0.589
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    1941,
+                    1407,
+                    1941,
+                    1407,
+                    2033,
+                    299,
+                    2033
+                ],
+                "score": 0.582
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    291,
+                    228,
+                    1405,
+                    228,
+                    1405,
+                    295,
+                    291,
+                    295
+                ],
+                "score": 0.571
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1114,
+                    1402,
+                    1114,
+                    1402,
+                    1210,
+                    297,
+                    1210
+                ],
+                "score": 0.561
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    426,
+                    1399,
+                    426,
+                    1399,
+                    523,
+                    297,
+                    523
+                ],
+                "score": 0.513
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    296,
+                    739,
+                    1405,
+                    739,
+                    1405,
+                    865,
+                    296,
+                    865
+                ],
+                "score": 0.506
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    295,
+                    312,
+                    1401,
+                    312,
+                    1401,
+                    407,
+                    295,
+                    407
+                ],
+                "score": 0.502
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    73.0,
+                    815.0,
+                    73.0,
+                    815.0,
+                    108.0,
+                    295.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    831.0,
+                    2085.0,
+                    869.0,
+                    2085.0,
+                    869.0,
+                    2123.0,
+                    831.0,
+                    2123.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1625.0,
+                    1403.0,
+                    1625.0,
+                    1403.0,
+                    1663.0,
+                    295.0,
+                    1663.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    1659.0,
+                    1405.0,
+                    1659.0,
+                    1405.0,
+                    1693.0,
+                    322.0,
+                    1693.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    321.0,
+                    1688.0,
+                    1374.0,
+                    1688.0,
+                    1374.0,
+                    1726.0,
+                    321.0,
+                    1726.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1541.0,
+                    1406.0,
+                    1541.0,
+                    1406.0,
+                    1581.0,
+                    295.0,
+                    1581.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    320.0,
+                    1575.0,
+                    706.0,
+                    1575.0,
+                    706.0,
+                    1607.0,
+                    320.0,
+                    1607.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    298.0,
+                    886.0,
+                    1406.0,
+                    886.0,
+                    1406.0,
+                    921.0,
+                    298.0,
+                    921.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    913.0,
+                    1408.0,
+                    913.0,
+                    1408.0,
+                    958.0,
+                    322.0,
+                    958.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    324.0,
+                    948.0,
+                    599.0,
+                    948.0,
+                    599.0,
+                    983.0,
+                    324.0,
+                    983.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    654.0,
+                    1405.0,
+                    654.0,
+                    1405.0,
+                    697.0,
+                    293.0,
+                    697.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    321.0,
+                    689.0,
+                    1084.0,
+                    689.0,
+                    1084.0,
+                    723.0,
+                    321.0,
+                    723.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    298.0,
+                    1429.0,
+                    1404.0,
+                    1429.0,
+                    1404.0,
+                    1463.0,
+                    298.0,
+                    1463.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    1461.0,
+                    1404.0,
+                    1461.0,
+                    1404.0,
+                    1495.0,
+                    323.0,
+                    1495.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    320.0,
+                    1491.0,
+                    707.0,
+                    1491.0,
+                    707.0,
+                    1523.0,
+                    320.0,
+                    1523.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1739.0,
+                    1406.0,
+                    1739.0,
+                    1406.0,
+                    1780.0,
+                    294.0,
+                    1780.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    1773.0,
+                    1405.0,
+                    1773.0,
+                    1405.0,
+                    1808.0,
+                    323.0,
+                    1808.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    1806.0,
+                    1084.0,
+                    1806.0,
+                    1084.0,
+                    1837.0,
+                    322.0,
+                    1837.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1856.0,
+                    1407.0,
+                    1856.0,
+                    1407.0,
+                    1892.0,
+                    295.0,
+                    1892.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    1887.0,
+                    777.0,
+                    1887.0,
+                    777.0,
+                    1922.0,
+                    323.0,
+                    1922.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1313.0,
+                    1406.0,
+                    1313.0,
+                    1406.0,
+                    1352.0,
+                    295.0,
+                    1352.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    320.0,
+                    1343.0,
+                    1406.0,
+                    1343.0,
+                    1406.0,
+                    1382.0,
+                    320.0,
+                    1382.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    325.0,
+                    1375.0,
+                    1050.0,
+                    1375.0,
+                    1050.0,
+                    1413.0,
+                    325.0,
+                    1413.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1230.0,
+                    1404.0,
+                    1230.0,
+                    1404.0,
+                    1266.0,
+                    296.0,
+                    1266.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    1261.0,
+                    1147.0,
+                    1261.0,
+                    1147.0,
+                    1297.0,
+                    323.0,
+                    1297.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    541.0,
+                    1404.0,
+                    541.0,
+                    1404.0,
+                    579.0,
+                    295.0,
+                    579.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    573.0,
+                    1409.0,
+                    573.0,
+                    1409.0,
+                    611.0,
+                    322.0,
+                    611.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    321.0,
+                    605.0,
+                    602.0,
+                    605.0,
+                    602.0,
+                    636.0,
+                    321.0,
+                    636.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    999.0,
+                    1403.0,
+                    999.0,
+                    1403.0,
+                    1037.0,
+                    295.0,
+                    1037.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    1032.0,
+                    1406.0,
+                    1032.0,
+                    1406.0,
+                    1067.0,
+                    322.0,
+                    1067.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    320.0,
+                    1063.0,
+                    947.0,
+                    1063.0,
+                    947.0,
+                    1098.0,
+                    320.0,
+                    1098.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1940.0,
+                    1408.0,
+                    1940.0,
+                    1408.0,
+                    1977.0,
+                    294.0,
+                    1977.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    319.0,
+                    1969.0,
+                    1410.0,
+                    1969.0,
+                    1410.0,
+                    2009.0,
+                    319.0,
+                    2009.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    319.0,
+                    1998.0,
+                    399.0,
+                    1998.0,
+                    399.0,
+                    2035.0,
+                    319.0,
+                    2035.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    226.0,
+                    1408.0,
+                    226.0,
+                    1408.0,
+                    268.0,
+                    293.0,
+                    268.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    260.0,
+                    1205.0,
+                    260.0,
+                    1205.0,
+                    297.0,
+                    323.0,
+                    297.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1111.0,
+                    1407.0,
+                    1111.0,
+                    1407.0,
+                    1156.0,
+                    292.0,
+                    1156.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    321.0,
+                    1146.0,
+                    1408.0,
+                    1146.0,
+                    1408.0,
+                    1183.0,
+                    321.0,
+                    1183.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    319.0,
+                    1177.0,
+                    548.0,
+                    1177.0,
+                    548.0,
+                    1211.0,
+                    319.0,
+                    1211.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    428.0,
+                    1403.0,
+                    428.0,
+                    1403.0,
+                    462.0,
+                    296.0,
+                    462.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    457.0,
+                    1404.0,
+                    457.0,
+                    1404.0,
+                    495.0,
+                    322.0,
+                    495.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    489.0,
+                    1008.0,
+                    489.0,
+                    1008.0,
+                    526.0,
+                    322.0,
+                    526.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    742.0,
+                    1405.0,
+                    742.0,
+                    1405.0,
+                    776.0,
+                    294.0,
+                    776.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    774.0,
+                    1403.0,
+                    774.0,
+                    1403.0,
+                    807.0,
+                    322.0,
+                    807.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    803.0,
+                    1406.0,
+                    803.0,
+                    1406.0,
+                    840.0,
+                    322.0,
+                    840.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    833.0,
+                    610.0,
+                    833.0,
+                    610.0,
+                    866.0,
+                    322.0,
+                    866.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    312.0,
+                    1405.0,
+                    312.0,
+                    1405.0,
+                    350.0,
+                    294.0,
+                    350.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    344.0,
+                    1407.0,
+                    344.0,
+                    1407.0,
+                    381.0,
+                    322.0,
+                    381.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    320.0,
+                    374.0,
+                    608.0,
+                    374.0,
+                    608.0,
+                    408.0,
+                    320.0,
+                    408.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 9,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 2,
+                "poly": [
+                    300,
+                    75,
+                    814,
+                    75,
+                    814,
+                    105,
+                    300,
+                    105
+                ],
+                "score": 0.882
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    229,
+                    1403,
+                    229,
+                    1403,
+                    323,
+                    298,
+                    323
+                ],
+                "score": 0.851
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    295,
+                    1114,
+                    1403,
+                    1114,
+                    1403,
+                    1179,
+                    295,
+                    1179
+                ],
+                "score": 0.838
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    295,
+                    1501,
+                    1401,
+                    1501,
+                    1401,
+                    1567,
+                    295,
+                    1567
+                ],
+                "score": 0.835
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    293,
+                    1277,
+                    1402,
+                    1277,
+                    1402,
+                    1343,
+                    293,
+                    1343
+                ],
+                "score": 0.832
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    300,
+                    452,
+                    1402,
+                    452,
+                    1402,
+                    546,
+                    300,
+                    546
+                ],
+                "score": 0.831
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    340,
+                    1403,
+                    340,
+                    1403,
+                    435,
+                    297,
+                    435
+                ],
+                "score": 0.827
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    292,
+                    1197,
+                    1401,
+                    1197,
+                    1401,
+                    1262,
+                    292,
+                    1262
+                ],
+                "score": 0.822
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    835,
+                    2088,
+                    862,
+                    2088,
+                    862,
+                    2113,
+                    835,
+                    2113
+                ],
+                "score": 0.821
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1359,
+                    1407,
+                    1359,
+                    1407,
+                    1484,
+                    298,
+                    1484
+                ],
+                "score": 0.8
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1696,
+                    1401,
+                    1696,
+                    1401,
+                    1761,
+                    297,
+                    1761
+                ],
+                "score": 0.799
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1583,
+                    1406,
+                    1583,
+                    1406,
+                    1678,
+                    298,
+                    1678
+                ],
+                "score": 0.795
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    565,
+                    1404,
+                    565,
+                    1404,
+                    600,
+                    297,
+                    600
+                ],
+                "score": 0.788
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1002,
+                    1403,
+                    1002,
+                    1403,
+                    1097,
+                    298,
+                    1097
+                ],
+                "score": 0.783
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    301,
+                    809,
+                    1400,
+                    809,
+                    1400,
+                    906,
+                    301,
+                    906
+                ],
+                "score": 0.762
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    292,
+                    615,
+                    1402,
+                    615,
+                    1402,
+                    681,
+                    292,
+                    681
+                ],
+                "score": 0.753
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    292,
+                    921,
+                    1402,
+                    921,
+                    1402,
+                    986,
+                    292,
+                    986
+                ],
+                "score": 0.741
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    301,
+                    697,
+                    1402,
+                    697,
+                    1402,
+                    794,
+                    301,
+                    794
+                ],
+                "score": 0.73
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1971,
+                    1399,
+                    1971,
+                    1399,
+                    2035,
+                    298,
+                    2035
+                ],
+                "score": 0.727
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1777,
+                    1405,
+                    1777,
+                    1405,
+                    1872,
+                    297,
+                    1872
+                ],
+                "score": 0.719
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    1889,
+                    1398,
+                    1889,
+                    1398,
+                    1954,
+                    299,
+                    1954
+                ],
+                "score": 0.688
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1132,
+                    1973,
+                    1201,
+                    1973,
+                    1201,
+                    2002,
+                    1132,
+                    2002
+                ],
+                "score": 0.88,
+                "latex": "80 \\text{‰}"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    73.0,
+                    816.0,
+                    73.0,
+                    816.0,
+                    108.0,
+                    295.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    831.0,
+                    2085.0,
+                    867.0,
+                    2085.0,
+                    867.0,
+                    2125.0,
+                    831.0,
+                    2125.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    227.0,
+                    1408.0,
+                    227.0,
+                    1408.0,
+                    266.0,
+                    293.0,
+                    266.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    260.0,
+                    1408.0,
+                    260.0,
+                    1408.0,
+                    297.0,
+                    322.0,
+                    297.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    320.0,
+                    293.0,
+                    574.0,
+                    293.0,
+                    574.0,
+                    322.0,
+                    320.0,
+                    322.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    1114.0,
+                    1405.0,
+                    1114.0,
+                    1405.0,
+                    1150.0,
+                    297.0,
+                    1150.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    1146.0,
+                    474.0,
+                    1146.0,
+                    474.0,
+                    1180.0,
+                    322.0,
+                    1180.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    1502.0,
+                    1405.0,
+                    1502.0,
+                    1405.0,
+                    1538.0,
+                    297.0,
+                    1538.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    1534.0,
+                    835.0,
+                    1534.0,
+                    835.0,
+                    1568.0,
+                    322.0,
+                    1568.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1276.0,
+                    1404.0,
+                    1276.0,
+                    1404.0,
+                    1315.0,
+                    295.0,
+                    1315.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    324.0,
+                    1310.0,
+                    868.0,
+                    1310.0,
+                    868.0,
+                    1344.0,
+                    324.0,
+                    1344.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    454.0,
+                    1404.0,
+                    454.0,
+                    1404.0,
+                    488.0,
+                    296.0,
+                    488.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    320.0,
+                    483.0,
+                    1409.0,
+                    483.0,
+                    1409.0,
+                    521.0,
+                    320.0,
+                    521.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    320.0,
+                    512.0,
+                    399.0,
+                    512.0,
+                    399.0,
+                    550.0,
+                    320.0,
+                    550.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    341.0,
+                    1404.0,
+                    341.0,
+                    1404.0,
+                    375.0,
+                    296.0,
+                    375.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    320.0,
+                    372.0,
+                    1409.0,
+                    372.0,
+                    1409.0,
+                    411.0,
+                    320.0,
+                    411.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    320.0,
+                    402.0,
+                    577.0,
+                    402.0,
+                    577.0,
+                    436.0,
+                    320.0,
+                    436.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1193.0,
+                    1406.0,
+                    1193.0,
+                    1406.0,
+                    1236.0,
+                    292.0,
+                    1236.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    320.0,
+                    1227.0,
+                    1241.0,
+                    1227.0,
+                    1241.0,
+                    1264.0,
+                    320.0,
+                    1264.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1362.0,
+                    1405.0,
+                    1362.0,
+                    1405.0,
+                    1395.0,
+                    296.0,
+                    1395.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    321.0,
+                    1389.0,
+                    1407.0,
+                    1389.0,
+                    1407.0,
+                    1427.0,
+                    321.0,
+                    1427.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    321.0,
+                    1421.0,
+                    1409.0,
+                    1421.0,
+                    1409.0,
+                    1459.0,
+                    321.0,
+                    1459.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    324.0,
+                    1453.0,
+                    531.0,
+                    1453.0,
+                    531.0,
+                    1484.0,
+                    324.0,
+                    1484.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1696.0,
+                    1405.0,
+                    1696.0,
+                    1405.0,
+                    1732.0,
+                    296.0,
+                    1732.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    324.0,
+                    1726.0,
+                    947.0,
+                    1726.0,
+                    947.0,
+                    1762.0,
+                    324.0,
+                    1762.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1583.0,
+                    1406.0,
+                    1583.0,
+                    1406.0,
+                    1621.0,
+                    296.0,
+                    1621.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    1615.0,
+                    1408.0,
+                    1615.0,
+                    1408.0,
+                    1651.0,
+                    322.0,
+                    1651.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    319.0,
+                    1645.0,
+                    520.0,
+                    1645.0,
+                    520.0,
+                    1679.0,
+                    319.0,
+                    1679.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    561.0,
+                    1409.0,
+                    561.0,
+                    1409.0,
+                    605.0,
+                    295.0,
+                    605.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1001.0,
+                    1408.0,
+                    1001.0,
+                    1408.0,
+                    1041.0,
+                    292.0,
+                    1041.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    320.0,
+                    1029.0,
+                    1410.0,
+                    1029.0,
+                    1410.0,
+                    1076.0,
+                    320.0,
+                    1076.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    325.0,
+                    1065.0,
+                    532.0,
+                    1065.0,
+                    532.0,
+                    1097.0,
+                    325.0,
+                    1097.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    807.0,
+                    1405.0,
+                    807.0,
+                    1405.0,
+                    848.0,
+                    295.0,
+                    848.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    840.0,
+                    1405.0,
+                    840.0,
+                    1405.0,
+                    878.0,
+                    322.0,
+                    878.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    871.0,
+                    1050.0,
+                    871.0,
+                    1050.0,
+                    909.0,
+                    323.0,
+                    909.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    616.0,
+                    1405.0,
+                    616.0,
+                    1405.0,
+                    652.0,
+                    295.0,
+                    652.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    648.0,
+                    1239.0,
+                    648.0,
+                    1239.0,
+                    682.0,
+                    323.0,
+                    682.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    920.0,
+                    1404.0,
+                    920.0,
+                    1404.0,
+                    960.0,
+                    295.0,
+                    960.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    321.0,
+                    951.0,
+                    1146.0,
+                    951.0,
+                    1146.0,
+                    991.0,
+                    321.0,
+                    991.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    699.0,
+                    1406.0,
+                    699.0,
+                    1406.0,
+                    733.0,
+                    297.0,
+                    733.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    727.0,
+                    1406.0,
+                    727.0,
+                    1406.0,
+                    764.0,
+                    322.0,
+                    764.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    321.0,
+                    759.0,
+                    1009.0,
+                    759.0,
+                    1009.0,
+                    796.0,
+                    321.0,
+                    796.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1967.0,
+                    1131.0,
+                    1967.0,
+                    1131.0,
+                    2011.0,
+                    293.0,
+                    2011.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1202.0,
+                    1967.0,
+                    1405.0,
+                    1967.0,
+                    1405.0,
+                    2011.0,
+                    1202.0,
+                    2011.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    2002.0,
+                    1052.0,
+                    2002.0,
+                    1052.0,
+                    2036.0,
+                    323.0,
+                    2036.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1777.0,
+                    1406.0,
+                    1777.0,
+                    1406.0,
+                    1815.0,
+                    294.0,
+                    1815.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    321.0,
+                    1806.0,
+                    1405.0,
+                    1806.0,
+                    1405.0,
+                    1846.0,
+                    321.0,
+                    1846.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    324.0,
+                    1840.0,
+                    1035.0,
+                    1840.0,
+                    1035.0,
+                    1873.0,
+                    324.0,
+                    1873.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    299.0,
+                    1891.0,
+                    1403.0,
+                    1891.0,
+                    1403.0,
+                    1923.0,
+                    299.0,
+                    1923.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    1920.0,
+                    1231.0,
+                    1920.0,
+                    1231.0,
+                    1955.0,
+                    323.0,
+                    1955.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 10,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 2,
+                "poly": [
+                    299,
+                    75,
+                    813,
+                    75,
+                    813,
+                    105,
+                    299,
+                    105
+                ],
+                "score": 0.885
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    835,
+                    2088,
+                    864,
+                    2088,
+                    864,
+                    2113,
+                    835,
+                    2113
+                ],
+                "score": 0.837
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    560,
+                    1402,
+                    560,
+                    1402,
+                    625,
+                    298,
+                    625
+                ],
+                "score": 0.692
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    295,
+                    644,
+                    1405,
+                    644,
+                    1405,
+                    710,
+                    295,
+                    710
+                ],
+                "score": 0.673
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    292,
+                    1656,
+                    1402,
+                    1656,
+                    1402,
+                    1723,
+                    292,
+                    1723
+                ],
+                "score": 0.664
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    295,
+                    1543,
+                    1401,
+                    1543,
+                    1401,
+                    1640,
+                    295,
+                    1640
+                ],
+                "score": 0.652
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    301,
+                    1739,
+                    1402,
+                    1739,
+                    1402,
+                    1835,
+                    301,
+                    1835
+                ],
+                "score": 0.65
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    293,
+                    726,
+                    1403,
+                    726,
+                    1403,
+                    792,
+                    293,
+                    792
+                ],
+                "score": 0.647
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    477,
+                    1404,
+                    477,
+                    1404,
+                    544,
+                    297,
+                    544
+                ],
+                "score": 0.626
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1089,
+                    1401,
+                    1089,
+                    1401,
+                    1185,
+                    298,
+                    1185
+                ],
+                "score": 0.562
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1429,
+                    1407,
+                    1429,
+                    1407,
+                    1525,
+                    297,
+                    1525
+                ],
+                "score": 0.561
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    294,
+                    809,
+                    1405,
+                    809,
+                    1405,
+                    845,
+                    294,
+                    845
+                ],
+                "score": 0.543
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    395,
+                    1402,
+                    395,
+                    1402,
+                    460,
+                    297,
+                    460
+                ],
+                "score": 0.539
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    300,
+                    861,
+                    1402,
+                    861,
+                    1402,
+                    959,
+                    300,
+                    959
+                ],
+                "score": 0.527
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    296,
+                    228,
+                    1403,
+                    228,
+                    1403,
+                    294,
+                    296,
+                    294
+                ],
+                "score": 0.517
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    311,
+                    1402,
+                    311,
+                    1402,
+                    378,
+                    299,
+                    378
+                ],
+                "score": 0.509
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    975,
+                    1402,
+                    975,
+                    1402,
+                    1072,
+                    297,
+                    1072
+                ],
+                "score": 0.509
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1316,
+                    1406,
+                    1316,
+                    1406,
+                    1410,
+                    297,
+                    1410
+                ],
+                "score": 0.43
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    73.0,
+                    815.0,
+                    73.0,
+                    815.0,
+                    108.0,
+                    295.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    831.0,
+                    2085.0,
+                    869.0,
+                    2085.0,
+                    869.0,
+                    2124.0,
+                    831.0,
+                    2124.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    558.0,
+                    1406.0,
+                    558.0,
+                    1406.0,
+                    599.0,
+                    294.0,
+                    599.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    592.0,
+                    622.0,
+                    592.0,
+                    622.0,
+                    625.0,
+                    322.0,
+                    625.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    645.0,
+                    1408.0,
+                    645.0,
+                    1408.0,
+                    681.0,
+                    295.0,
+                    681.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    676.0,
+                    1353.0,
+                    676.0,
+                    1353.0,
+                    711.0,
+                    323.0,
+                    711.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1654.0,
+                    1404.0,
+                    1654.0,
+                    1404.0,
+                    1695.0,
+                    294.0,
+                    1695.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    1690.0,
+                    1007.0,
+                    1690.0,
+                    1007.0,
+                    1722.0,
+                    323.0,
+                    1722.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1543.0,
+                    1403.0,
+                    1543.0,
+                    1403.0,
+                    1582.0,
+                    294.0,
+                    1582.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    1574.0,
+                    1406.0,
+                    1574.0,
+                    1406.0,
+                    1612.0,
+                    322.0,
+                    1612.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    1604.0,
+                    1084.0,
+                    1604.0,
+                    1084.0,
+                    1641.0,
+                    322.0,
+                    1641.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1739.0,
+                    1406.0,
+                    1739.0,
+                    1406.0,
+                    1777.0,
+                    296.0,
+                    1777.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    1772.0,
+                    1404.0,
+                    1772.0,
+                    1404.0,
+                    1807.0,
+                    322.0,
+                    1807.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    324.0,
+                    1800.0,
+                    775.0,
+                    1800.0,
+                    775.0,
+                    1838.0,
+                    324.0,
+                    1838.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    728.0,
+                    1405.0,
+                    728.0,
+                    1405.0,
+                    764.0,
+                    296.0,
+                    764.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    759.0,
+                    1236.0,
+                    759.0,
+                    1236.0,
+                    795.0,
+                    322.0,
+                    795.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    478.0,
+                    1407.0,
+                    478.0,
+                    1407.0,
+                    514.0,
+                    297.0,
+                    514.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    509.0,
+                    945.0,
+                    509.0,
+                    945.0,
+                    544.0,
+                    322.0,
+                    544.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1089.0,
+                    1403.0,
+                    1089.0,
+                    1403.0,
+                    1126.0,
+                    294.0,
+                    1126.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    1122.0,
+                    1405.0,
+                    1122.0,
+                    1405.0,
+                    1157.0,
+                    323.0,
+                    1157.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    1153.0,
+                    720.0,
+                    1153.0,
+                    720.0,
+                    1187.0,
+                    323.0,
+                    1187.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1430.0,
+                    1407.0,
+                    1430.0,
+                    1407.0,
+                    1465.0,
+                    296.0,
+                    1465.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    320.0,
+                    1459.0,
+                    1410.0,
+                    1459.0,
+                    1410.0,
+                    1502.0,
+                    320.0,
+                    1502.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    324.0,
+                    1493.0,
+                    532.0,
+                    1493.0,
+                    532.0,
+                    1524.0,
+                    324.0,
+                    1524.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    807.0,
+                    1408.0,
+                    807.0,
+                    1408.0,
+                    847.0,
+                    297.0,
+                    847.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    395.0,
+                    1404.0,
+                    395.0,
+                    1404.0,
+                    431.0,
+                    296.0,
+                    431.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    428.0,
+                    893.0,
+                    428.0,
+                    893.0,
+                    461.0,
+                    322.0,
+                    461.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    860.0,
+                    1407.0,
+                    860.0,
+                    1407.0,
+                    901.0,
+                    294.0,
+                    901.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    320.0,
+                    894.0,
+                    1407.0,
+                    894.0,
+                    1407.0,
+                    931.0,
+                    320.0,
+                    931.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    321.0,
+                    925.0,
+                    858.0,
+                    925.0,
+                    858.0,
+                    959.0,
+                    321.0,
+                    959.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    227.0,
+                    1403.0,
+                    227.0,
+                    1403.0,
+                    264.0,
+                    296.0,
+                    264.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    261.0,
+                    932.0,
+                    261.0,
+                    932.0,
+                    295.0,
+                    323.0,
+                    295.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    312.0,
+                    1407.0,
+                    312.0,
+                    1407.0,
+                    348.0,
+                    295.0,
+                    348.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    343.0,
+                    1404.0,
+                    343.0,
+                    1404.0,
+                    380.0,
+                    323.0,
+                    380.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    977.0,
+                    1407.0,
+                    977.0,
+                    1407.0,
+                    1011.0,
+                    296.0,
+                    1011.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    324.0,
+                    1009.0,
+                    1406.0,
+                    1009.0,
+                    1406.0,
+                    1043.0,
+                    324.0,
+                    1043.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    1038.0,
+                    1108.0,
+                    1038.0,
+                    1108.0,
+                    1074.0,
+                    322.0,
+                    1074.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1318.0,
+                    1407.0,
+                    1318.0,
+                    1407.0,
+                    1352.0,
+                    296.0,
+                    1352.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    321.0,
+                    1348.0,
+                    1408.0,
+                    1348.0,
+                    1408.0,
+                    1382.0,
+                    321.0,
+                    1382.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    320.0,
+                    1373.0,
+                    399.0,
+                    1373.0,
+                    399.0,
+                    1414.0,
+                    320.0,
+                    1414.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 11,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    258,
+                    482,
+                    1378,
+                    482,
+                    1378,
+                    1796,
+                    258,
+                    1796
+                ],
+                "score": 0.955
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    299,
+                    74,
+                    813,
+                    74,
+                    813,
+                    106,
+                    299,
+                    106
+                ],
+                "score": 0.871
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    294,
+                    1234,
+                    294,
+                    1234,
+                    329,
+                    298,
+                    329
+                ],
+                "score": 0.867
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    292,
+                    420,
+                    1270,
+                    420,
+                    1270,
+                    455,
+                    292,
+                    455
+                ],
+                "score": 0.837
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    835,
+                    2087,
+                    864,
+                    2087,
+                    864,
+                    2113,
+                    835,
+                    2113
+                ],
+                "score": 0.828
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    300,
+                    227,
+                    510,
+                    227,
+                    510,
+                    262,
+                    300,
+                    262
+                ],
+                "score": 0.796
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    364,
+                    712,
+                    364,
+                    712,
+                    397,
+                    299,
+                    397
+                ],
+                "score": 0.454
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    299,
+                    364,
+                    712,
+                    364,
+                    712,
+                    397,
+                    299,
+                    397
+                ],
+                "score": 0.386
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    683,
+                    1028,
+                    734,
+                    1028,
+                    734,
+                    1056,
+                    683,
+                    1056
+                ],
+                "score": 0.82,
+                "latex": "\\mathrm { p } { = } 2"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    459,
+                    1361,
+                    484,
+                    1361,
+                    484,
+                    1384,
+                    459,
+                    1384
+                ],
+                "score": 0.8,
+                "latex": "="
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    459,
+                    1333,
+                    484,
+                    1333,
+                    484,
+                    1355,
+                    459,
+                    1355
+                ],
+                "score": 0.8,
+                "latex": "="
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    743,
+                    1165,
+                    793,
+                    1165,
+                    793,
+                    1194,
+                    743,
+                    1194
+                ],
+                "score": 0.79,
+                "latex": "\\mathrm { p } { = } 2"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    458,
+                    1473,
+                    485,
+                    1473,
+                    485,
+                    1495,
+                    458,
+                    1495
+                ],
+                "score": 0.78,
+                "latex": "="
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    458,
+                    836,
+                    485,
+                    836,
+                    485,
+                    858,
+                    458,
+                    858
+                ],
+                "score": 0.77,
+                "latex": "="
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    807,
+                    1029,
+                    839,
+                    1029,
+                    839,
+                    1052,
+                    807,
+                    1052
+                ],
+                "score": 0.7,
+                "latex": "^ { = 1 }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    428,
+                    1555,
+                    455,
+                    1555,
+                    455,
+                    1578,
+                    428,
+                    1578
+                ],
+                "score": 0.7,
+                "latex": "="
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    881,
+                    1472,
+                    913,
+                    1472,
+                    913,
+                    1495,
+                    881,
+                    1495
+                ],
+                "score": 0.69,
+                "latex": "^ { = 1 }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    866,
+                    1166,
+                    898,
+                    1166,
+                    898,
+                    1190,
+                    866,
+                    1190
+                ],
+                "score": 0.69,
+                "latex": "^ { = 1 }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1090,
+                    1554,
+                    1122,
+                    1554,
+                    1122,
+                    1578,
+                    1090,
+                    1578
+                ],
+                "score": 0.64,
+                "latex": "^ { 1 = 1 }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    477,
+                    1444,
+                    498,
+                    1444,
+                    498,
+                    1466,
+                    477,
+                    1466
+                ],
+                "score": 0.6,
+                "latex": "="
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    471,
+                    1167,
+                    499,
+                    1167,
+                    499,
+                    1190,
+                    471,
+                    1190
+                ],
+                "score": 0.55,
+                "latex": "="
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    472,
+                    1139,
+                    497,
+                    1139,
+                    497,
+                    1161,
+                    472,
+                    1161
+                ],
+                "score": 0.52,
+                "latex": "="
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    356,
+                    1028,
+                    441,
+                    1028,
+                    441,
+                    1056,
+                    356,
+                    1056
+                ],
+                "score": 0.47,
+                "latex": "\\mathrm { ~ \\cal ~ X ~ \\_ ~ S ~ ~ } ="
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    413,
+                    1029,
+                    440,
+                    1029,
+                    440,
+                    1052,
+                    413,
+                    1052
+                ],
+                "score": 0.36,
+                "latex": "="
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    356,
+                    917,
+                    410,
+                    917,
+                    410,
+                    942,
+                    356,
+                    942
+                ],
+                "score": 0.35,
+                "latex": "\\begin{array} { r l } { \\mathrm { B } } & { { } = } \\end{array}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    415,
+                    1165,
+                    500,
+                    1165,
+                    500,
+                    1196,
+                    415,
+                    1196
+                ],
+                "score": 0.27,
+                "latex": "\\mathrm { ~ X ~ \\_ ~ T ~ } ="
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    357,
+                    999,
+                    439,
+                    999,
+                    439,
+                    1025,
+                    357,
+                    1025
+                ],
+                "score": 0.26,
+                "latex": "\\mathrm { ~ \\cal ~ X ~ \\_ ~ S ~ ~ } ="
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    72.0,
+                    815.0,
+                    72.0,
+                    815.0,
+                    109.0,
+                    295.0,
+                    109.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    831.0,
+                    2085.0,
+                    868.0,
+                    2085.0,
+                    868.0,
+                    2124.0,
+                    831.0,
+                    2124.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    221.0,
+                    515.0,
+                    221.0,
+                    515.0,
+                    271.0,
+                    293.0,
+                    271.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    365.0,
+                    713.0,
+                    365.0,
+                    713.0,
+                    398.0,
+                    297.0,
+                    398.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    494.0,
+                    1198.0,
+                    494.0,
+                    1198.0,
+                    532.0,
+                    292.0,
+                    532.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    273.0,
+                    526.0,
+                    797.0,
+                    526.0,
+                    797.0,
+                    559.0,
+                    273.0,
+                    559.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    271.0,
+                    554.0,
+                    739.0,
+                    554.0,
+                    739.0,
+                    587.0,
+                    271.0,
+                    587.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    273.0,
+                    582.0,
+                    1110.0,
+                    582.0,
+                    1110.0,
+                    615.0,
+                    273.0,
+                    615.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    300.0,
+                    610.0,
+                    340.0,
+                    610.0,
+                    340.0,
+                    629.0,
+                    300.0,
+                    629.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    275.0,
+                    647.0,
+                    289.0,
+                    647.0,
+                    289.0,
+                    662.0,
+                    275.0,
+                    662.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    283.0,
+                    660.0,
+                    1271.0,
+                    660.0,
+                    1271.0,
+                    699.0,
+                    283.0,
+                    699.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    269.0,
+                    688.0,
+                    421.0,
+                    688.0,
+                    421.0,
+                    725.0,
+                    269.0,
+                    725.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    274.0,
+                    728.0,
+                    292.0,
+                    728.0,
+                    292.0,
+                    746.0,
+                    274.0,
+                    746.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    261.0,
+                    746.0,
+                    1215.0,
+                    746.0,
+                    1215.0,
+                    781.0,
+                    261.0,
+                    781.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    265.0,
+                    780.0,
+                    290.0,
+                    780.0,
+                    290.0,
+                    805.0,
+                    265.0,
+                    805.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    265.0,
+                    808.0,
+                    293.0,
+                    808.0,
+                    293.0,
+                    833.0,
+                    265.0,
+                    833.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    366.0,
+                    800.0,
+                    994.0,
+                    800.0,
+                    994.0,
+                    837.0,
+                    366.0,
+                    837.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    265.0,
+                    836.0,
+                    293.0,
+                    836.0,
+                    293.0,
+                    861.0,
+                    265.0,
+                    861.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    355.0,
+                    831.0,
+                    457.0,
+                    831.0,
+                    457.0,
+                    864.0,
+                    355.0,
+                    864.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    486.0,
+                    831.0,
+                    660.0,
+                    831.0,
+                    660.0,
+                    864.0,
+                    486.0,
+                    864.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    264.0,
+                    861.0,
+                    295.0,
+                    861.0,
+                    295.0,
+                    890.0,
+                    264.0,
+                    890.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    264.0,
+                    889.0,
+                    295.0,
+                    889.0,
+                    295.0,
+                    918.0,
+                    264.0,
+                    918.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    353.0,
+                    884.0,
+                    545.0,
+                    884.0,
+                    545.0,
+                    917.0,
+                    353.0,
+                    917.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    265.0,
+                    918.0,
+                    293.0,
+                    918.0,
+                    293.0,
+                    943.0,
+                    265.0,
+                    943.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    349.0,
+                    912.0,
+                    355.0,
+                    912.0,
+                    355.0,
+                    949.0,
+                    349.0,
+                    949.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    411.0,
+                    912.0,
+                    648.0,
+                    912.0,
+                    648.0,
+                    949.0,
+                    411.0,
+                    949.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    265.0,
+                    946.0,
+                    293.0,
+                    946.0,
+                    293.0,
+                    974.0,
+                    265.0,
+                    974.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    264.0,
+                    971.0,
+                    295.0,
+                    971.0,
+                    295.0,
+                    1001.0,
+                    264.0,
+                    1001.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    353.0,
+                    967.0,
+                    858.0,
+                    967.0,
+                    858.0,
+                    999.0,
+                    353.0,
+                    999.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    265.0,
+                    1001.0,
+                    293.0,
+                    1001.0,
+                    293.0,
+                    1026.0,
+                    265.0,
+                    1026.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    353.0,
+                    996.0,
+                    356.0,
+                    996.0,
+                    356.0,
+                    1029.0,
+                    353.0,
+                    1029.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    440.0,
+                    996.0,
+                    585.0,
+                    996.0,
+                    585.0,
+                    1029.0,
+                    440.0,
+                    1029.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    262.0,
+                    1027.0,
+                    295.0,
+                    1027.0,
+                    295.0,
+                    1057.0,
+                    262.0,
+                    1057.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    352.0,
+                    1023.0,
+                    355.0,
+                    1023.0,
+                    355.0,
+                    1059.0,
+                    352.0,
+                    1059.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    442.0,
+                    1023.0,
+                    682.0,
+                    1023.0,
+                    682.0,
+                    1059.0,
+                    442.0,
+                    1059.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    735.0,
+                    1023.0,
+                    806.0,
+                    1023.0,
+                    806.0,
+                    1059.0,
+                    735.0,
+                    1059.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    840.0,
+                    1023.0,
+                    853.0,
+                    1023.0,
+                    853.0,
+                    1059.0,
+                    840.0,
+                    1059.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    264.0,
+                    1057.0,
+                    292.0,
+                    1057.0,
+                    292.0,
+                    1085.0,
+                    264.0,
+                    1085.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    262.0,
+                    1082.0,
+                    293.0,
+                    1082.0,
+                    293.0,
+                    1112.0,
+                    262.0,
+                    1112.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    355.0,
+                    1082.0,
+                    781.0,
+                    1082.0,
+                    781.0,
+                    1110.0,
+                    355.0,
+                    1110.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    262.0,
+                    1110.0,
+                    293.0,
+                    1110.0,
+                    293.0,
+                    1140.0,
+                    262.0,
+                    1140.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    353.0,
+                    1108.0,
+                    676.0,
+                    1108.0,
+                    676.0,
+                    1140.0,
+                    353.0,
+                    1140.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    262.0,
+                    1138.0,
+                    295.0,
+                    1138.0,
+                    295.0,
+                    1167.0,
+                    262.0,
+                    1167.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    410.0,
+                    1132.0,
+                    471.0,
+                    1132.0,
+                    471.0,
+                    1167.0,
+                    410.0,
+                    1167.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    498.0,
+                    1132.0,
+                    1025.0,
+                    1132.0,
+                    1025.0,
+                    1167.0,
+                    498.0,
+                    1167.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    262.0,
+                    1164.0,
+                    293.0,
+                    1164.0,
+                    293.0,
+                    1195.0,
+                    262.0,
+                    1195.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    501.0,
+                    1163.0,
+                    742.0,
+                    1163.0,
+                    742.0,
+                    1195.0,
+                    501.0,
+                    1195.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    794.0,
+                    1163.0,
+                    865.0,
+                    1163.0,
+                    865.0,
+                    1195.0,
+                    794.0,
+                    1195.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    899.0,
+                    1163.0,
+                    915.0,
+                    1163.0,
+                    915.0,
+                    1195.0,
+                    899.0,
+                    1195.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    262.0,
+                    1192.0,
+                    295.0,
+                    1192.0,
+                    295.0,
+                    1223.0,
+                    262.0,
+                    1223.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    262.0,
+                    1220.0,
+                    293.0,
+                    1220.0,
+                    293.0,
+                    1251.0,
+                    262.0,
+                    1251.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    350.0,
+                    1214.0,
+                    1010.0,
+                    1214.0,
+                    1010.0,
+                    1252.0,
+                    350.0,
+                    1252.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    262.0,
+                    1248.0,
+                    293.0,
+                    1248.0,
+                    293.0,
+                    1277.0,
+                    262.0,
+                    1277.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    353.0,
+                    1245.0,
+                    588.0,
+                    1245.0,
+                    588.0,
+                    1282.0,
+                    353.0,
+                    1282.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    262.0,
+                    1276.0,
+                    295.0,
+                    1276.0,
+                    295.0,
+                    1305.0,
+                    262.0,
+                    1305.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    262.0,
+                    1304.0,
+                    295.0,
+                    1304.0,
+                    295.0,
+                    1333.0,
+                    262.0,
+                    1333.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    352.0,
+                    1298.0,
+                    1183.0,
+                    1298.0,
+                    1183.0,
+                    1335.0,
+                    352.0,
+                    1335.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    264.0,
+                    1333.0,
+                    290.0,
+                    1333.0,
+                    290.0,
+                    1358.0,
+                    264.0,
+                    1358.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    352.0,
+                    1325.0,
+                    458.0,
+                    1325.0,
+                    458.0,
+                    1363.0,
+                    352.0,
+                    1363.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    485.0,
+                    1325.0,
+                    1378.0,
+                    1325.0,
+                    1378.0,
+                    1363.0,
+                    485.0,
+                    1363.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    262.0,
+                    1360.0,
+                    293.0,
+                    1360.0,
+                    293.0,
+                    1389.0,
+                    262.0,
+                    1389.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    352.0,
+                    1353.0,
+                    458.0,
+                    1353.0,
+                    458.0,
+                    1391.0,
+                    352.0,
+                    1391.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    485.0,
+                    1353.0,
+                    1378.0,
+                    1353.0,
+                    1378.0,
+                    1391.0,
+                    485.0,
+                    1391.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    264.0,
+                    1389.0,
+                    292.0,
+                    1389.0,
+                    292.0,
+                    1414.0,
+                    264.0,
+                    1414.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    262.0,
+                    1414.0,
+                    295.0,
+                    1414.0,
+                    295.0,
+                    1444.0,
+                    262.0,
+                    1444.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    353.0,
+                    1410.0,
+                    902.0,
+                    1410.0,
+                    902.0,
+                    1442.0,
+                    353.0,
+                    1442.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    262.0,
+                    1442.0,
+                    293.0,
+                    1442.0,
+                    293.0,
+                    1472.0,
+                    262.0,
+                    1472.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    352.0,
+                    1435.0,
+                    476.0,
+                    1435.0,
+                    476.0,
+                    1475.0,
+                    352.0,
+                    1475.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    499.0,
+                    1435.0,
+                    607.0,
+                    1435.0,
+                    607.0,
+                    1475.0,
+                    499.0,
+                    1475.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    262.0,
+                    1470.0,
+                    295.0,
+                    1470.0,
+                    295.0,
+                    1498.0,
+                    262.0,
+                    1498.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    355.0,
+                    1467.0,
+                    457.0,
+                    1467.0,
+                    457.0,
+                    1500.0,
+                    355.0,
+                    1500.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    486.0,
+                    1467.0,
+                    880.0,
+                    1467.0,
+                    880.0,
+                    1500.0,
+                    486.0,
+                    1500.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    914.0,
+                    1467.0,
+                    928.0,
+                    1467.0,
+                    928.0,
+                    1500.0,
+                    914.0,
+                    1500.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    262.0,
+                    1497.0,
+                    293.0,
+                    1497.0,
+                    293.0,
+                    1526.0,
+                    262.0,
+                    1526.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    264.0,
+                    1528.0,
+                    292.0,
+                    1528.0,
+                    292.0,
+                    1553.0,
+                    264.0,
+                    1553.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    350.0,
+                    1519.0,
+                    1233.0,
+                    1519.0,
+                    1233.0,
+                    1557.0,
+                    350.0,
+                    1557.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    262.0,
+                    1553.0,
+                    293.0,
+                    1553.0,
+                    293.0,
+                    1582.0,
+                    262.0,
+                    1582.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    353.0,
+                    1548.0,
+                    427.0,
+                    1548.0,
+                    427.0,
+                    1584.0,
+                    353.0,
+                    1584.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    456.0,
+                    1548.0,
+                    1089.0,
+                    1548.0,
+                    1089.0,
+                    1584.0,
+                    456.0,
+                    1584.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1123.0,
+                    1548.0,
+                    1278.0,
+                    1548.0,
+                    1278.0,
+                    1584.0,
+                    1123.0,
+                    1584.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    262.0,
+                    1579.0,
+                    295.0,
+                    1579.0,
+                    295.0,
+                    1610.0,
+                    262.0,
+                    1610.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    264.0,
+                    1610.0,
+                    290.0,
+                    1610.0,
+                    290.0,
+                    1635.0,
+                    264.0,
+                    1635.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    352.0,
+                    1603.0,
+                    844.0,
+                    1603.0,
+                    844.0,
+                    1640.0,
+                    352.0,
+                    1640.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    262.0,
+                    1637.0,
+                    293.0,
+                    1637.0,
+                    293.0,
+                    1666.0,
+                    262.0,
+                    1666.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    355.0,
+                    1634.0,
+                    575.0,
+                    1634.0,
+                    575.0,
+                    1666.0,
+                    355.0,
+                    1666.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    591.0,
+                    1631.0,
+                    1053.0,
+                    1631.0,
+                    1053.0,
+                    1668.0,
+                    591.0,
+                    1668.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    262.0,
+                    1663.0,
+                    295.0,
+                    1663.0,
+                    295.0,
+                    1693.0,
+                    262.0,
+                    1693.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    262.0,
+                    1691.0,
+                    293.0,
+                    1691.0,
+                    293.0,
+                    1721.0,
+                    262.0,
+                    1721.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    352.0,
+                    1685.0,
+                    576.0,
+                    1685.0,
+                    576.0,
+                    1722.0,
+                    352.0,
+                    1722.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    262.0,
+                    1719.0,
+                    293.0,
+                    1719.0,
+                    293.0,
+                    1749.0,
+                    262.0,
+                    1749.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    355.0,
+                    1716.0,
+                    585.0,
+                    1716.0,
+                    585.0,
+                    1749.0,
+                    355.0,
+                    1749.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    262.0,
+                    1746.0,
+                    295.0,
+                    1746.0,
+                    295.0,
+                    1775.0,
+                    262.0,
+                    1775.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    352.0,
+                    1743.0,
+                    601.0,
+                    1743.0,
+                    601.0,
+                    1777.0,
+                    352.0,
+                    1777.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    287.0,
+                    1241.0,
+                    287.0,
+                    1241.0,
+                    338.0,
+                    291.0,
+                    338.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    416.0,
+                    1274.0,
+                    416.0,
+                    1274.0,
+                    460.0,
+                    294.0,
+                    460.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    365.0,
+                    713.0,
+                    365.0,
+                    713.0,
+                    398.0,
+                    297.0,
+                    398.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 12,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 5,
+                "poly": [
+                    513,
+                    1610,
+                    1178,
+                    1610,
+                    1178,
+                    1796,
+                    513,
+                    1796
+                ],
+                "score": 0.978,
+                "html": "<table><tr><td>Model</td><td>Deeper MLPs</td><td>Top-1 Acc. Top-5 Acc.</td><td></td></tr><tr><td>EfficientNet-b0</td><td></td><td>39.1</td><td>64.6</td></tr><tr><td>EfficientNet-b0</td><td>×</td><td>42.2</td><td>68.5</td></tr><tr><td>Mobile-v3-large</td><td>X</td><td>19.0</td><td>41.3</td></tr><tr><td>Mobile-v3-large</td><td>√</td><td>36.3</td><td>62.2</td></tr></table>"
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    646,
+                    1408,
+                    646,
+                    1408,
+                    832,
+                    298,
+                    832
+                ],
+                "score": 0.973
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1240,
+                    1405,
+                    1240,
+                    1405,
+                    1487,
+                    297,
+                    1487
+                ],
+                "score": 0.973
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1072,
+                    1404,
+                    1072,
+                    1404,
+                    1225,
+                    298,
+                    1225
+                ],
+                "score": 0.971
+            },
+            {
+                "category_id": 6,
+                "poly": [
+                    289,
+                    1527,
+                    1400,
+                    1527,
+                    1400,
+                    1588,
+                    289,
+                    1588
+                ],
+                "score": 0.937
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1924,
+                    1161,
+                    1924,
+                    1161,
+                    1958,
+                    297,
+                    1958
+                ],
+                "score": 0.914
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    300,
+                    282,
+                    1252,
+                    282,
+                    1252,
+                    316,
+                    300,
+                    316
+                ],
+                "score": 0.904
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    300,
+                    75,
+                    814,
+                    75,
+                    814,
+                    105,
+                    300,
+                    105
+                ],
+                "score": 0.891
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    293,
+                    1972,
+                    1410,
+                    1972,
+                    1410,
+                    2034,
+                    293,
+                    2034
+                ],
+                "score": 0.89
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    835,
+                    2088,
+                    863,
+                    2088,
+                    863,
+                    2112,
+                    835,
+                    2112
+                ],
+                "score": 0.86
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    300,
+                    229,
+                    679,
+                    229,
+                    679,
+                    262,
+                    300,
+                    262
+                ],
+                "score": 0.83
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    359,
+                    855,
+                    1406,
+                    855,
+                    1406,
+                    1049,
+                    359,
+                    1049
+                ],
+                "score": 0.823
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    298,
+                    1867,
+                    853,
+                    1867,
+                    853,
+                    1900,
+                    298,
+                    1900
+                ],
+                "score": 0.805
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    372,
+                    408,
+                    1380,
+                    408,
+                    1380,
+                    442,
+                    372,
+                    442
+                ],
+                "score": 0.547
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    304,
+                    593,
+                    1248,
+                    593,
+                    1248,
+                    625,
+                    304,
+                    625
+                ],
+                "score": 0.484
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    304,
+                    593,
+                    1248,
+                    593,
+                    1248,
+                    625,
+                    304,
+                    625
+                ],
+                "score": 0.475
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    380,
+                    338,
+                    1402,
+                    338,
+                    1402,
+                    400,
+                    380,
+                    400
+                ],
+                "score": 0.459
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    391,
+                    489,
+                    1302,
+                    489,
+                    1302,
+                    521,
+                    391,
+                    521
+                ],
+                "score": 0.438
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    393,
+                    530,
+                    1104,
+                    530,
+                    1104,
+                    563,
+                    393,
+                    563
+                ],
+                "score": 0.415
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    381,
+                    450,
+                    1051,
+                    450,
+                    1051,
+                    481,
+                    381,
+                    481
+                ],
+                "score": 0.397
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    633,
+                    370,
+                    741,
+                    370,
+                    741,
+                    399,
+                    633,
+                    399
+                ],
+                "score": 0.91,
+                "latex": "2 2 4 \\times 2 2 4"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    891,
+                    489,
+                    1049,
+                    489,
+                    1049,
+                    522,
+                    891,
+                    522
+                ],
+                "score": 0.9,
+                "latex": "\\sigma = \\{ 0 . 1 , 0 . 2 \\}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1008,
+                    709,
+                    1086,
+                    709,
+                    1086,
+                    738,
+                    1008,
+                    738
+                ],
+                "score": 0.88,
+                "latex": "7 2 . 2 \\%"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    390,
+                    709,
+                    468,
+                    709,
+                    468,
+                    738,
+                    390,
+                    738
+                ],
+                "score": 0.86,
+                "latex": "7 7 . 2 \\%"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1212,
+                    1364,
+                    1266,
+                    1364,
+                    1266,
+                    1393,
+                    1212,
+                    1393
+                ],
+                "score": 0.84,
+                "latex": "17 \\%"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1201,
+                    770,
+                    1276,
+                    770,
+                    1276,
+                    800,
+                    1201,
+                    800
+                ],
+                "score": 0.84,
+                "latex": "3 6 . 3 \\%"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1523.0,
+                    1405.0,
+                    1523.0,
+                    1405.0,
+                    1563.0,
+                    293.0,
+                    1563.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1555.0,
+                    1205.0,
+                    1555.0,
+                    1205.0,
+                    1589.0,
+                    293.0,
+                    1589.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    73.0,
+                    816.0,
+                    73.0,
+                    816.0,
+                    108.0,
+                    295.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    831.0,
+                    2084.0,
+                    869.0,
+                    2084.0,
+                    869.0,
+                    2123.0,
+                    831.0,
+                    2123.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    227.0,
+                    681.0,
+                    227.0,
+                    681.0,
+                    265.0,
+                    295.0,
+                    265.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1867.0,
+                    855.0,
+                    1867.0,
+                    855.0,
+                    1903.0,
+                    295.0,
+                    1903.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    591.0,
+                    1251.0,
+                    591.0,
+                    1251.0,
+                    629.0,
+                    296.0,
+                    629.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    647.0,
+                    1405.0,
+                    647.0,
+                    1405.0,
+                    679.0,
+                    296.0,
+                    679.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    678.0,
+                    1405.0,
+                    678.0,
+                    1405.0,
+                    710.0,
+                    296.0,
+                    710.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    707.0,
+                    389.0,
+                    707.0,
+                    389.0,
+                    743.0,
+                    295.0,
+                    743.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    469.0,
+                    707.0,
+                    1007.0,
+                    707.0,
+                    1007.0,
+                    743.0,
+                    469.0,
+                    743.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1087.0,
+                    707.0,
+                    1409.0,
+                    707.0,
+                    1409.0,
+                    743.0,
+                    1087.0,
+                    743.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    736.0,
+                    1408.0,
+                    736.0,
+                    1408.0,
+                    775.0,
+                    293.0,
+                    775.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    767.0,
+                    1200.0,
+                    767.0,
+                    1200.0,
+                    803.0,
+                    295.0,
+                    803.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1277.0,
+                    767.0,
+                    1408.0,
+                    767.0,
+                    1408.0,
+                    803.0,
+                    1277.0,
+                    803.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    802.0,
+                    1154.0,
+                    802.0,
+                    1154.0,
+                    833.0,
+                    296.0,
+                    833.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1241.0,
+                    1403.0,
+                    1241.0,
+                    1403.0,
+                    1274.0,
+                    296.0,
+                    1274.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1273.0,
+                    1405.0,
+                    1273.0,
+                    1405.0,
+                    1303.0,
+                    296.0,
+                    1303.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1301.0,
+                    1405.0,
+                    1301.0,
+                    1405.0,
+                    1337.0,
+                    292.0,
+                    1337.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1332.0,
+                    1405.0,
+                    1332.0,
+                    1405.0,
+                    1367.0,
+                    294.0,
+                    1367.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1362.0,
+                    1211.0,
+                    1362.0,
+                    1211.0,
+                    1398.0,
+                    292.0,
+                    1398.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1267.0,
+                    1362.0,
+                    1408.0,
+                    1362.0,
+                    1408.0,
+                    1398.0,
+                    1267.0,
+                    1398.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1391.0,
+                    1405.0,
+                    1391.0,
+                    1405.0,
+                    1427.0,
+                    294.0,
+                    1427.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1423.0,
+                    1405.0,
+                    1423.0,
+                    1405.0,
+                    1459.0,
+                    294.0,
+                    1459.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1454.0,
+                    1328.0,
+                    1454.0,
+                    1328.0,
+                    1488.0,
+                    294.0,
+                    1488.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1073.0,
+                    1402.0,
+                    1073.0,
+                    1402.0,
+                    1106.0,
+                    296.0,
+                    1106.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1101.0,
+                    1404.0,
+                    1101.0,
+                    1404.0,
+                    1139.0,
+                    293.0,
+                    1139.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1131.0,
+                    1406.0,
+                    1131.0,
+                    1406.0,
+                    1170.0,
+                    293.0,
+                    1170.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1163.0,
+                    1406.0,
+                    1163.0,
+                    1406.0,
+                    1199.0,
+                    293.0,
+                    1199.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1193.0,
+                    733.0,
+                    1193.0,
+                    733.0,
+                    1230.0,
+                    294.0,
+                    1230.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1921.0,
+                    1162.0,
+                    1921.0,
+                    1162.0,
+                    1962.0,
+                    293.0,
+                    1962.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    281.0,
+                    1257.0,
+                    281.0,
+                    1257.0,
+                    320.0,
+                    294.0,
+                    320.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1970.0,
+                    1407.0,
+                    1970.0,
+                    1407.0,
+                    2006.0,
+                    296.0,
+                    2006.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    2000.0,
+                    1405.0,
+                    2000.0,
+                    1405.0,
+                    2036.0,
+                    296.0,
+                    2036.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    363.0,
+                    856.0,
+                    1403.0,
+                    856.0,
+                    1403.0,
+                    888.0,
+                    363.0,
+                    888.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    393.0,
+                    882.0,
+                    1406.0,
+                    882.0,
+                    1406.0,
+                    920.0,
+                    393.0,
+                    920.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    393.0,
+                    914.0,
+                    476.0,
+                    914.0,
+                    476.0,
+                    950.0,
+                    393.0,
+                    950.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    359.0,
+                    955.0,
+                    1402.0,
+                    955.0,
+                    1402.0,
+                    988.0,
+                    359.0,
+                    988.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    394.0,
+                    986.0,
+                    1406.0,
+                    986.0,
+                    1406.0,
+                    1022.0,
+                    394.0,
+                    1022.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    395.0,
+                    1019.0,
+                    805.0,
+                    1019.0,
+                    805.0,
+                    1053.0,
+                    395.0,
+                    1053.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    385.0,
+                    407.0,
+                    1380.0,
+                    407.0,
+                    1380.0,
+                    446.0,
+                    385.0,
+                    446.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    591.0,
+                    1251.0,
+                    591.0,
+                    1251.0,
+                    629.0,
+                    296.0,
+                    629.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    386.0,
+                    336.0,
+                    1405.0,
+                    336.0,
+                    1405.0,
+                    375.0,
+                    386.0,
+                    375.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    385.0,
+                    369.0,
+                    632.0,
+                    369.0,
+                    632.0,
+                    403.0,
+                    385.0,
+                    403.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    742.0,
+                    369.0,
+                    753.0,
+                    369.0,
+                    753.0,
+                    403.0,
+                    742.0,
+                    403.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    384.0,
+                    484.0,
+                    890.0,
+                    484.0,
+                    890.0,
+                    527.0,
+                    384.0,
+                    527.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1050.0,
+                    484.0,
+                    1304.0,
+                    484.0,
+                    1304.0,
+                    527.0,
+                    1050.0,
+                    527.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    386.0,
+                    526.0,
+                    1111.0,
+                    526.0,
+                    1111.0,
+                    569.0,
+                    386.0,
+                    569.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    386.0,
+                    447.0,
+                    1056.0,
+                    447.0,
+                    1056.0,
+                    486.0,
+                    386.0,
+                    486.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 13,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1294,
+                    1406,
+                    1294,
+                    1406,
+                    1600,
+                    297,
+                    1600
+                ],
+                "score": 0.982
+            },
+            {
+                "category_id": 5,
+                "poly": [
+                    538,
+                    839,
+                    1160,
+                    839,
+                    1160,
+                    1213,
+                    538,
+                    1213
+                ],
+                "score": 0.974,
+                "html": "<table><tr><td></td><td>Stud.</td><td></td><td>ResNet-50</td><td></td></tr><tr><td>Teac.</td><td>Epoch</td><td>KNN</td><td>Top-1</td><td>Top-5</td></tr><tr><td>X ResNet-50</td><td>200</td><td>46.1 46.1</td><td>67.4 67.5</td><td>87.8 87.8</td></tr><tr><td>△ ResNet-101</td><td>200</td><td>+0.0 52.3</td><td>+0.1</td><td>+0.0</td></tr><tr><td>△</td><td>200</td><td>+6.2</td><td>69.1 +1.7</td><td>88.7 +0.9</td></tr><tr><td>ResNet-152 △</td><td>200</td><td>53.2 +7.1</td><td>70.4 +3.0</td><td>90.5 +2.7</td></tr><tr><td>ResNet-50×2* △</td><td>800</td><td>59.0 +12.9</td><td>74.3 +6.9</td><td>92.2 +4.4</td></tr></table>"
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    301,
+                    1616,
+                    1405,
+                    1616,
+                    1405,
+                    1770,
+                    301,
+                    1770
+                ],
+                "score": 0.973
+            },
+            {
+                "category_id": 5,
+                "poly": [
+                    550,
+                    306,
+                    1133,
+                    306,
+                    1133,
+                    649,
+                    550,
+                    649
+                ],
+                "score": 0.973,
+                "html": "<table><tr><td>Student</td><td>Teacher</td><td>Top-1</td><td>Top-5</td></tr><tr><td rowspan=\"3\">EfficientNet-b0</td><td>X</td><td>39.1</td><td>64.6</td></tr><tr><td>ResNet-50</td><td>59.2</td><td>81.2</td></tr><tr><td>ResNet-101</td><td>62.8</td><td>84.7</td></tr><tr><td></td><td>ResNet-152</td><td>63.3</td><td>85.6</td></tr><tr><td rowspan=\"3\">MobileNet-v3</td><td>X ResNet-50</td><td>19.0 50.9</td><td>41.3 77.7</td></tr><tr><td>ResNet-101</td><td>57.6</td><td>82.6</td></tr><tr><td>ResNet-152</td><td>58.3</td><td>82.9</td></tr></table>"
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    349,
+                    1797,
+                    1404,
+                    1797,
+                    1404,
+                    1951,
+                    349,
+                    1951
+                ],
+                "score": 0.968
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    295,
+                    1978,
+                    1409,
+                    1978,
+                    1409,
+                    2033,
+                    295,
+                    2033
+                ],
+                "score": 0.942
+            },
+            {
+                "category_id": 6,
+                "poly": [
+                    296,
+                    223,
+                    1404,
+                    223,
+                    1404,
+                    281,
+                    296,
+                    281
+                ],
+                "score": 0.913
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    300,
+                    75,
+                    813,
+                    75,
+                    813,
+                    105,
+                    300,
+                    105
+                ],
+                "score": 0.892
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    835,
+                    2088,
+                    863,
+                    2088,
+                    863,
+                    2112,
+                    835,
+                    2112
+                ],
+                "score": 0.859
+            },
+            {
+                "category_id": 6,
+                "poly": [
+                    298,
+                    702,
+                    1406,
+                    702,
+                    1406,
+                    814,
+                    298,
+                    814
+                ],
+                "score": 0.691
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    702,
+                    1406,
+                    702,
+                    1406,
+                    814,
+                    298,
+                    814
+                ],
+                "score": 0.325
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    889,
+                    1509,
+                    997,
+                    1509,
+                    997,
+                    1538,
+                    889,
+                    1538
+                ],
+                "score": 0.9,
+                "latex": "2 5 6 \\times 2 5 6"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    773,
+                    1479,
+                    883,
+                    1479,
+                    883,
+                    1508,
+                    773,
+                    1508
+                ],
+                "score": 0.89,
+                "latex": "2 2 4 \\times 2 2 4"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    804,
+                    1798,
+                    884,
+                    1798,
+                    884,
+                    1827,
+                    804,
+                    1827
+                ],
+                "score": 0.89,
+                "latex": "3 2 \\times 3 2"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1294,
+                    1509,
+                    1402,
+                    1509,
+                    1402,
+                    1538,
+                    1294,
+                    1538
+                ],
+                "score": 0.88,
+                "latex": "2 2 4 \\times 2 2 4"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1152,
+                    1797,
+                    1262,
+                    1797,
+                    1262,
+                    1828,
+                    1152,
+                    1828
+                ],
+                "score": 0.88,
+                "latex": "2 2 4 \\times 2 2 4"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    744,
+                    1296,
+                    798,
+                    1296,
+                    798,
+                    1326,
+                    744,
+                    1326
+                ],
+                "score": 0.85,
+                "latex": "10 \\%"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    655,
+                    1296,
+                    694,
+                    1296,
+                    694,
+                    1325,
+                    655,
+                    1325
+                ],
+                "score": 0.83,
+                "latex": "1 \\%"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1111,
+                    1360,
+                    1142,
+                    1360,
+                    1142,
+                    1384,
+                    1111,
+                    1384
+                ],
+                "score": 0.75,
+                "latex": "="
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    684,
+                    704,
+                    724,
+                    704,
+                    724,
+                    730,
+                    684,
+                    730
+                ],
+                "score": 0.73,
+                "latex": "( \\% )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    886,
+                    226,
+                    924,
+                    226,
+                    924,
+                    252,
+                    886,
+                    252
+                ],
+                "score": 0.69,
+                "latex": "( \\% )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    923,
+                    731,
+                    946,
+                    731,
+                    946,
+                    757,
+                    923,
+                    757
+                ],
+                "score": 0.46,
+                "latex": "\\pmb { \\chi }"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    326.0,
+                    1970.0,
+                    1414.0,
+                    1970.0,
+                    1414.0,
+                    2012.0,
+                    326.0,
+                    2012.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    2003.0,
+                    1134.0,
+                    2003.0,
+                    1134.0,
+                    2037.0,
+                    296.0,
+                    2037.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    222.0,
+                    885.0,
+                    222.0,
+                    885.0,
+                    257.0,
+                    294.0,
+                    257.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    925.0,
+                    222.0,
+                    1409.0,
+                    222.0,
+                    1409.0,
+                    257.0,
+                    925.0,
+                    257.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    251.0,
+                    583.0,
+                    251.0,
+                    583.0,
+                    285.0,
+                    294.0,
+                    285.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    73.0,
+                    815.0,
+                    73.0,
+                    815.0,
+                    108.0,
+                    295.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    831.0,
+                    2085.0,
+                    868.0,
+                    2085.0,
+                    868.0,
+                    2124.0,
+                    831.0,
+                    2124.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    704.0,
+                    683.0,
+                    704.0,
+                    683.0,
+                    733.0,
+                    295.0,
+                    733.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    725.0,
+                    704.0,
+                    1402.0,
+                    704.0,
+                    1402.0,
+                    733.0,
+                    725.0,
+                    733.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    725.0,
+                    922.0,
+                    725.0,
+                    922.0,
+                    766.0,
+                    292.0,
+                    766.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    947.0,
+                    725.0,
+                    1406.0,
+                    725.0,
+                    1406.0,
+                    766.0,
+                    947.0,
+                    766.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    757.0,
+                    1404.0,
+                    757.0,
+                    1404.0,
+                    790.0,
+                    296.0,
+                    790.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    786.0,
+                    632.0,
+                    786.0,
+                    632.0,
+                    819.0,
+                    296.0,
+                    819.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1296.0,
+                    654.0,
+                    1296.0,
+                    654.0,
+                    1328.0,
+                    296.0,
+                    1328.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    695.0,
+                    1296.0,
+                    743.0,
+                    1296.0,
+                    743.0,
+                    1328.0,
+                    695.0,
+                    1328.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    799.0,
+                    1296.0,
+                    1406.0,
+                    1296.0,
+                    1406.0,
+                    1328.0,
+                    799.0,
+                    1328.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1326.0,
+                    1404.0,
+                    1326.0,
+                    1404.0,
+                    1358.0,
+                    296.0,
+                    1358.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1354.0,
+                    1110.0,
+                    1354.0,
+                    1110.0,
+                    1391.0,
+                    292.0,
+                    1391.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1143.0,
+                    1354.0,
+                    1407.0,
+                    1354.0,
+                    1407.0,
+                    1391.0,
+                    1143.0,
+                    1391.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1386.0,
+                    1406.0,
+                    1386.0,
+                    1406.0,
+                    1423.0,
+                    294.0,
+                    1423.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1417.0,
+                    1406.0,
+                    1417.0,
+                    1406.0,
+                    1452.0,
+                    295.0,
+                    1452.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1443.0,
+                    1408.0,
+                    1443.0,
+                    1408.0,
+                    1486.0,
+                    292.0,
+                    1486.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1477.0,
+                    772.0,
+                    1477.0,
+                    772.0,
+                    1513.0,
+                    295.0,
+                    1513.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    884.0,
+                    1477.0,
+                    1406.0,
+                    1477.0,
+                    1406.0,
+                    1513.0,
+                    884.0,
+                    1513.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1508.0,
+                    888.0,
+                    1508.0,
+                    888.0,
+                    1543.0,
+                    292.0,
+                    1543.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    998.0,
+                    1508.0,
+                    1293.0,
+                    1508.0,
+                    1293.0,
+                    1543.0,
+                    998.0,
+                    1543.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1403.0,
+                    1508.0,
+                    1406.0,
+                    1508.0,
+                    1406.0,
+                    1543.0,
+                    1403.0,
+                    1543.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1538.0,
+                    1406.0,
+                    1538.0,
+                    1406.0,
+                    1574.0,
+                    295.0,
+                    1574.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1571.0,
+                    732.0,
+                    1571.0,
+                    732.0,
+                    1603.0,
+                    296.0,
+                    1603.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1612.0,
+                    1405.0,
+                    1612.0,
+                    1405.0,
+                    1651.0,
+                    295.0,
+                    1651.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1644.0,
+                    1407.0,
+                    1644.0,
+                    1407.0,
+                    1680.0,
+                    296.0,
+                    1680.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1674.0,
+                    1407.0,
+                    1674.0,
+                    1407.0,
+                    1712.0,
+                    295.0,
+                    1712.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1706.0,
+                    1409.0,
+                    1706.0,
+                    1409.0,
+                    1741.0,
+                    294.0,
+                    1741.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1734.0,
+                    1303.0,
+                    1734.0,
+                    1303.0,
+                    1772.0,
+                    295.0,
+                    1772.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    346.0,
+                    1793.0,
+                    803.0,
+                    1793.0,
+                    803.0,
+                    1833.0,
+                    346.0,
+                    1833.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    885.0,
+                    1793.0,
+                    1151.0,
+                    1793.0,
+                    1151.0,
+                    1833.0,
+                    885.0,
+                    1833.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1263.0,
+                    1793.0,
+                    1406.0,
+                    1793.0,
+                    1406.0,
+                    1833.0,
+                    1263.0,
+                    1833.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    347.0,
+                    1829.0,
+                    1406.0,
+                    1829.0,
+                    1406.0,
+                    1864.0,
+                    347.0,
+                    1864.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    347.0,
+                    1860.0,
+                    1405.0,
+                    1860.0,
+                    1405.0,
+                    1892.0,
+                    347.0,
+                    1892.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    346.0,
+                    1889.0,
+                    1405.0,
+                    1889.0,
+                    1405.0,
+                    1925.0,
+                    346.0,
+                    1925.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    347.0,
+                    1919.0,
+                    495.0,
+                    1919.0,
+                    495.0,
+                    1952.0,
+                    347.0,
+                    1952.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    704.0,
+                    683.0,
+                    704.0,
+                    683.0,
+                    733.0,
+                    295.0,
+                    733.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    725.0,
+                    704.0,
+                    1402.0,
+                    704.0,
+                    1402.0,
+                    733.0,
+                    725.0,
+                    733.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    725.0,
+                    922.0,
+                    725.0,
+                    922.0,
+                    766.0,
+                    292.0,
+                    766.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    947.0,
+                    725.0,
+                    1406.0,
+                    725.0,
+                    1406.0,
+                    766.0,
+                    947.0,
+                    766.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    757.0,
+                    1404.0,
+                    757.0,
+                    1404.0,
+                    790.0,
+                    296.0,
+                    790.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    786.0,
+                    632.0,
+                    786.0,
+                    632.0,
+                    819.0,
+                    296.0,
+                    819.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 14,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 5,
+                "poly": [
+                    304,
+                    887,
+                    1396,
+                    887,
+                    1396,
+                    1133,
+                    304,
+                    1133
+                ],
+                "score": 0.982,
+                "html": "<table><tr><td rowspan=\"2\">Student</td><td rowspan=\"2\">Teacher</td><td colspan=\"3\">Object Detection</td><td colspan=\"3\">Instance Segmentation</td></tr><tr><td>Apbb</td><td>AP</td><td>APP</td><td>Apmk</td><td>AP</td><td>AP</td></tr><tr><td rowspan=\"4\">ResNet34</td><td>X</td><td>38.1</td><td>56.8</td><td>40.7</td><td>33.0</td><td>53.2</td><td>35.3</td></tr><tr><td>ResNet50</td><td>38.4 (+0.3)</td><td>57.0 (+0.2)</td><td>41.0 (+0.3)</td><td>33.3 (+0.3)</td><td>53.6 (+0.4)</td><td>35.4 (+0.1)</td></tr><tr><td>ResNet101</td><td>38.5 (+0.4)</td><td>57.3 (+0.5)</td><td>41.4 (+0.7)</td><td>33.6 (+0.6)</td><td>54.1 (+0.9)</td><td>35.6 (+0.3)</td></tr><tr><td>ResNet152</td><td>38.4 (+0.3)</td><td>57.0 (+0.2)</td><td>41.0 (+0.3)</td><td>33.3 (+0.3)</td><td>53.7 (+0.5)</td><td>35.3 (+0.0)</td></tr></table>"
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1386,
+                    1405,
+                    1386,
+                    1405,
+                    1662,
+                    298,
+                    1662
+                ],
+                "score": 0.98
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    348,
+                    1207,
+                    1403,
+                    1207,
+                    1403,
+                    1361,
+                    348,
+                    1361
+                ],
+                "score": 0.978
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    350,
+                    1686,
+                    1403,
+                    1686,
+                    1403,
+                    1872,
+                    350,
+                    1872
+                ],
+                "score": 0.978
+            },
+            {
+                "category_id": 5,
+                "poly": [
+                    519,
+                    337,
+                    1172,
+                    337,
+                    1172,
+                    727,
+                    519,
+                    727
+                ],
+                "score": 0.976,
+                "html": "<table><tr><td rowspan=\"2\">Student</td><td rowspan=\"2\">Teacher</td><td colspan=\"3\">VOC Object Detection</td></tr><tr><td>APbb</td><td>AP</td><td>AP</td></tr><tr><td rowspan=\"3\">ResNet-34</td><td>X</td><td>53.6</td><td>79.1</td><td>58.7</td></tr><tr><td>ResNet-50</td><td>53.7 (+0.1)</td><td>79.4 (+0.3)</td><td>59.2 (+0.5)</td></tr><tr><td>ResNet-101 ResNet-152</td><td>54.1 (+0.5) 54.4 (+0.8)</td><td>79.8 (+0.7) 80.1 (+1.0)</td><td>59.1 (+0.4) ） 59.9 (+1.2)</td></tr><tr><td rowspan=\"2\">ResNet-50</td><td>X</td><td>57.0</td><td>82.4</td><td>63.6</td></tr><tr><td>ResNet-50 ResNet-101</td><td>57.0 (+0.0) 57.1 (+0.1)</td><td>82.4 (+0.0) 82.8 (+0.4)</td><td>63.6 (+0.0) 63.8 (+0.2)</td></tr></table>"
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    353,
+                    1880,
+                    1402,
+                    1880,
+                    1402,
+                    2034,
+                    353,
+                    2034
+                ],
+                "score": 0.974
+            },
+            {
+                "category_id": 6,
+                "poly": [
+                    296,
+                    778,
+                    1402,
+                    778,
+                    1402,
+                    862,
+                    296,
+                    862
+                ],
+                "score": 0.925
+            },
+            {
+                "category_id": 6,
+                "poly": [
+                    295,
+                    226,
+                    1398,
+                    226,
+                    1398,
+                    311,
+                    295,
+                    311
+                ],
+                "score": 0.911
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    299,
+                    75,
+                    812,
+                    75,
+                    812,
+                    105,
+                    299,
+                    105
+                ],
+                "score": 0.898
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    836,
+                    2089,
+                    863,
+                    2089,
+                    863,
+                    2112,
+                    836,
+                    2112
+                ],
+                "score": 0.847
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    455,
+                    1330,
+                    510,
+                    1330,
+                    510,
+                    1359,
+                    455,
+                    1359
+                ],
+                "score": 0.86,
+                "latex": "20 \\%"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1361,
+                    1912,
+                    1402,
+                    1912,
+                    1402,
+                    1941,
+                    1361,
+                    1941
+                ],
+                "score": 0.85,
+                "latex": "2 \\times"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1285,
+                    775,
+                    1355,
+                    775,
+                    1355,
+                    807,
+                    1285,
+                    807
+                ],
+                "score": 0.84,
+                "latex": "( \\mathbf { A P } ^ { b b } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    338,
+                    254,
+                    414,
+                    254,
+                    414,
+                    283,
+                    338,
+                    283
+                ],
+                "score": 0.82,
+                "latex": "( \\mathbf { A } \\mathbf { P } ^ { m k } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    397,
+                    805,
+                    474,
+                    805,
+                    474,
+                    834,
+                    397,
+                    834
+                ],
+                "score": 0.82,
+                "latex": "( \\mathbf { A } \\mathbf { P } ^ { m k } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1232,
+                    224,
+                    1296,
+                    224,
+                    1296,
+                    255,
+                    1232,
+                    255
+                ],
+                "score": 0.81,
+                "latex": "( \\mathsf { A P } ^ { b b } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    476,
+                    1973,
+                    535,
+                    1973,
+                    535,
+                    2001,
+                    476,
+                    2001
+                ],
+                "score": 0.26,
+                "latex": "1 8 0 \\mathrm { k }"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    775.0,
+                    1284.0,
+                    775.0,
+                    1284.0,
+                    808.0,
+                    291.0,
+                    808.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1356.0,
+                    775.0,
+                    1405.0,
+                    775.0,
+                    1405.0,
+                    808.0,
+                    1356.0,
+                    808.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    802.0,
+                    396.0,
+                    802.0,
+                    396.0,
+                    837.0,
+                    292.0,
+                    837.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    475.0,
+                    802.0,
+                    1407.0,
+                    802.0,
+                    1407.0,
+                    837.0,
+                    475.0,
+                    837.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    833.0,
+                    610.0,
+                    833.0,
+                    610.0,
+                    864.0,
+                    294.0,
+                    864.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    223.0,
+                    1231.0,
+                    223.0,
+                    1231.0,
+                    258.0,
+                    291.0,
+                    258.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1297.0,
+                    223.0,
+                    1404.0,
+                    223.0,
+                    1404.0,
+                    258.0,
+                    1297.0,
+                    258.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    250.0,
+                    337.0,
+                    250.0,
+                    337.0,
+                    286.0,
+                    293.0,
+                    286.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    415.0,
+                    250.0,
+                    1405.0,
+                    250.0,
+                    1405.0,
+                    286.0,
+                    415.0,
+                    286.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    280.0,
+                    418.0,
+                    280.0,
+                    418.0,
+                    313.0,
+                    293.0,
+                    313.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    73.0,
+                    816.0,
+                    73.0,
+                    816.0,
+                    108.0,
+                    295.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    832.0,
+                    2085.0,
+                    868.0,
+                    2085.0,
+                    868.0,
+                    2124.0,
+                    832.0,
+                    2124.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    1387.0,
+                    1403.0,
+                    1387.0,
+                    1403.0,
+                    1420.0,
+                    297.0,
+                    1420.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1417.0,
+                    1405.0,
+                    1417.0,
+                    1405.0,
+                    1451.0,
+                    293.0,
+                    1451.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1446.0,
+                    1405.0,
+                    1446.0,
+                    1405.0,
+                    1482.0,
+                    293.0,
+                    1482.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1477.0,
+                    1406.0,
+                    1477.0,
+                    1406.0,
+                    1514.0,
+                    295.0,
+                    1514.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1508.0,
+                    1406.0,
+                    1508.0,
+                    1406.0,
+                    1542.0,
+                    293.0,
+                    1542.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1540.0,
+                    1405.0,
+                    1540.0,
+                    1405.0,
+                    1572.0,
+                    296.0,
+                    1572.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1567.0,
+                    1408.0,
+                    1567.0,
+                    1408.0,
+                    1606.0,
+                    292.0,
+                    1606.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1598.0,
+                    1405.0,
+                    1598.0,
+                    1405.0,
+                    1636.0,
+                    293.0,
+                    1636.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1630.0,
+                    1408.0,
+                    1630.0,
+                    1408.0,
+                    1666.0,
+                    296.0,
+                    1666.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    345.0,
+                    1205.0,
+                    1405.0,
+                    1205.0,
+                    1405.0,
+                    1241.0,
+                    345.0,
+                    1241.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    346.0,
+                    1238.0,
+                    1405.0,
+                    1238.0,
+                    1405.0,
+                    1273.0,
+                    346.0,
+                    1273.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    345.0,
+                    1267.0,
+                    1405.0,
+                    1267.0,
+                    1405.0,
+                    1302.0,
+                    345.0,
+                    1302.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    344.0,
+                    1298.0,
+                    1405.0,
+                    1298.0,
+                    1405.0,
+                    1335.0,
+                    344.0,
+                    1335.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    347.0,
+                    1329.0,
+                    454.0,
+                    1329.0,
+                    454.0,
+                    1365.0,
+                    347.0,
+                    1365.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    511.0,
+                    1329.0,
+                    1209.0,
+                    1329.0,
+                    1209.0,
+                    1365.0,
+                    511.0,
+                    1365.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    349.0,
+                    1687.0,
+                    1405.0,
+                    1687.0,
+                    1405.0,
+                    1719.0,
+                    349.0,
+                    1719.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    344.0,
+                    1716.0,
+                    1405.0,
+                    1716.0,
+                    1405.0,
+                    1751.0,
+                    344.0,
+                    1751.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    344.0,
+                    1743.0,
+                    1406.0,
+                    1743.0,
+                    1406.0,
+                    1784.0,
+                    344.0,
+                    1784.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    344.0,
+                    1776.0,
+                    1405.0,
+                    1776.0,
+                    1405.0,
+                    1813.0,
+                    344.0,
+                    1813.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    347.0,
+                    1808.0,
+                    1404.0,
+                    1808.0,
+                    1404.0,
+                    1844.0,
+                    347.0,
+                    1844.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    346.0,
+                    1839.0,
+                    1398.0,
+                    1839.0,
+                    1398.0,
+                    1876.0,
+                    346.0,
+                    1876.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    345.0,
+                    1879.0,
+                    1406.0,
+                    1879.0,
+                    1406.0,
+                    1916.0,
+                    345.0,
+                    1916.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    345.0,
+                    1909.0,
+                    1360.0,
+                    1909.0,
+                    1360.0,
+                    1948.0,
+                    345.0,
+                    1948.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    345.0,
+                    1941.0,
+                    1405.0,
+                    1941.0,
+                    1405.0,
+                    1975.0,
+                    345.0,
+                    1975.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    345.0,
+                    1971.0,
+                    475.0,
+                    1971.0,
+                    475.0,
+                    2008.0,
+                    345.0,
+                    2008.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    536.0,
+                    1971.0,
+                    1405.0,
+                    1971.0,
+                    1405.0,
+                    2008.0,
+                    536.0,
+                    2008.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    345.0,
+                    2002.0,
+                    834.0,
+                    2002.0,
+                    834.0,
+                    2037.0,
+                    345.0,
+                    2037.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 15,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1030,
+                    1406,
+                    1030,
+                    1406,
+                    1426,
+                    297,
+                    1426
+                ],
+                "score": 0.981
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1651,
+                    1404,
+                    1651,
+                    1404,
+                    1896,
+                    297,
+                    1896
+                ],
+                "score": 0.98
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    1911,
+                    1402,
+                    1911,
+                    1402,
+                    2034,
+                    299,
+                    2034
+                ],
+                "score": 0.972
+            },
+            {
+                "category_id": 4,
+                "poly": [
+                    300,
+                    823,
+                    1404,
+                    823,
+                    1404,
+                    908,
+                    300,
+                    908
+                ],
+                "score": 0.962
+            },
+            {
+                "category_id": 3,
+                "poly": [
+                    333,
+                    563,
+                    1378,
+                    563,
+                    1378,
+                    802,
+                    333,
+                    802
+                ],
+                "score": 0.961
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    296,
+                    1573,
+                    1399,
+                    1573,
+                    1399,
+                    1636,
+                    296,
+                    1636
+                ],
+                "score": 0.936
+            },
+            {
+                "category_id": 5,
+                "poly": [
+                    403,
+                    326,
+                    1290,
+                    326,
+                    1290,
+                    510,
+                    403,
+                    510
+                ],
+                "score": 0.935,
+                "html": "<table><tr><td>Method</td><td>Multi-View(s)</td><td>Top-1 Acc.</td><td>Top-5 Acc.</td></tr><tr><td>Identical-View</td><td>1×224</td><td>61.7</td><td>84.2</td></tr><tr><td>Cross-Views</td><td>2×224</td><td>58.2</td><td>81.7</td></tr><tr><td>Multi-Crops + Cross-Views</td><td>1×224 + 6×96×96</td><td>61.9</td><td>84.4</td></tr><tr><td>Multi-Crops + Identical-View </td><td>1×224 +6×96×96</td><td>62.6</td><td>84.8</td></tr></table>"
+            },
+            {
+                "category_id": 6,
+                "poly": [
+                    296,
+                    242,
+                    1401,
+                    242,
+                    1401,
+                    301,
+                    296,
+                    301
+                ],
+                "score": 0.912
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    301,
+                    1525,
+                    1125,
+                    1525,
+                    1125,
+                    1558,
+                    301,
+                    1558
+                ],
+                "score": 0.909
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    300,
+                    76,
+                    812,
+                    76,
+                    812,
+                    104,
+                    300,
+                    104
+                ],
+                "score": 0.894
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    836,
+                    2088,
+                    863,
+                    2088,
+                    863,
+                    2112,
+                    836,
+                    2112
+                ],
+                "score": 0.846
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    296,
+                    1467,
+                    965,
+                    1467,
+                    965,
+                    1499,
+                    296,
+                    1499
+                ],
+                "score": 0.84
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    300,
+                    970,
+                    1105,
+                    970,
+                    1105,
+                    1003,
+                    300,
+                    1003
+                ],
+                "score": 0.593
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    300,
+                    970,
+                    1105,
+                    970,
+                    1105,
+                    1003,
+                    300,
+                    1003
+                ],
+                "score": 0.331
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    368,
+                    1332,
+                    464,
+                    1332,
+                    464,
+                    1369,
+                    368,
+                    1369
+                ],
+                "score": 0.91,
+                "latex": "( \\mathbf { z } _ { l } ^ { S } , \\mathbf { z } _ { l } ^ { T } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1197,
+                    1302,
+                    1293,
+                    1302,
+                    1293,
+                    1337,
+                    1197,
+                    1337
+                ],
+                "score": 0.91,
+                "latex": "( \\mathbf { z } _ { s } ^ { S } , \\mathbf { z } _ { s } ^ { T } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1032,
+                    1152,
+                    1113,
+                    1152,
+                    1113,
+                    1181,
+                    1032,
+                    1181
+                ],
+                "score": 0.9,
+                "latex": "9 6 \\times 9 6"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    446,
+                    1602,
+                    496,
+                    1602,
+                    496,
+                    1638,
+                    446,
+                    1638
+                ],
+                "score": 0.85,
+                "latex": "( \\mathbf { z } _ { i } ^ { S } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1084,
+                    1241,
+                    1137,
+                    1241,
+                    1137,
+                    1275,
+                    1084,
+                    1275
+                ],
+                "score": 0.85,
+                "latex": "( \\mathbf { z } ^ { T } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    925,
+                    1610,
+                    954,
+                    1610,
+                    954,
+                    1635,
+                    925,
+                    1635
+                ],
+                "score": 0.83,
+                "latex": "\\mathbf { x } _ { i }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    380,
+                    1943,
+                    457,
+                    1943,
+                    457,
+                    1973,
+                    380,
+                    1973
+                ],
+                "score": 0.82,
+                "latex": "( 5 1 . 0 \\%"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    635,
+                    1602,
+                    684,
+                    1602,
+                    684,
+                    1638,
+                    635,
+                    1638
+                ],
+                "score": 0.76,
+                "latex": "( \\mathbf { z } _ { i } ^ { T } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1050,
+                    2006,
+                    1073,
+                    2006,
+                    1073,
+                    2030,
+                    1050,
+                    2030
+                ],
+                "score": 0.73,
+                "latex": "K"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1122,
+                    1714,
+                    1146,
+                    1714,
+                    1146,
+                    1741,
+                    1122,
+                    1741
+                ],
+                "score": 0.71,
+                "latex": "K"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    587,
+                    1913,
+                    611,
+                    1913,
+                    611,
+                    1939,
+                    587,
+                    1939
+                ],
+                "score": 0.69,
+                "latex": "K"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    698,
+                    1653,
+                    723,
+                    1653,
+                    723,
+                    1679,
+                    698,
+                    1679
+                ],
+                "score": 0.68,
+                "latex": "K"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1072,
+                    1575,
+                    1097,
+                    1575,
+                    1097,
+                    1601,
+                    1072,
+                    1601
+                ],
+                "score": 0.66,
+                "latex": "l 2"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1125,
+                    1683,
+                    1174,
+                    1683,
+                    1174,
+                    1710,
+                    1125,
+                    1710
+                ],
+                "score": 0.41,
+                "latex": "S S L"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    376,
+                    1743,
+                    420,
+                    1743,
+                    420,
+                    1772,
+                    376,
+                    1772
+                ],
+                "score": 0.39,
+                "latex": "1 6 \\mathrm { k }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    296,
+                    1573,
+                    321,
+                    1573,
+                    321,
+                    1602,
+                    296,
+                    1602
+                ],
+                "score": 0.39,
+                "latex": "\\mathbf { \\xi } _ { l 2 }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    295,
+                    1652,
+                    319,
+                    1652,
+                    319,
+                    1680,
+                    295,
+                    1680
+                ],
+                "score": 0.39,
+                "latex": "\\pmb { K }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    296,
+                    1743,
+                    328,
+                    1743,
+                    328,
+                    1771,
+                    296,
+                    1771
+                ],
+                "score": 0.28,
+                "latex": "4 \\mathrm { k }"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    823.0,
+                    1404.0,
+                    823.0,
+                    1404.0,
+                    855.0,
+                    296.0,
+                    855.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    853.0,
+                    1403.0,
+                    853.0,
+                    1403.0,
+                    882.0,
+                    296.0,
+                    882.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    878.0,
+                    562.0,
+                    878.0,
+                    562.0,
+                    911.0,
+                    294.0,
+                    911.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    498.0,
+                    592.0,
+                    518.0,
+                    592.0,
+                    518.0,
+                    611.0,
+                    498.0,
+                    611.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1063.0,
+                    590.0,
+                    1079.0,
+                    590.0,
+                    1079.0,
+                    604.0,
+                    1063.0,
+                    604.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1339.0,
+                    580.0,
+                    1372.0,
+                    580.0,
+                    1372.0,
+                    614.0,
+                    1339.0,
+                    614.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1044.0,
+                    707.0,
+                    1083.0,
+                    707.0,
+                    1083.0,
+                    733.0,
+                    1044.0,
+                    733.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    413.0,
+                    772.0,
+                    449.0,
+                    772.0,
+                    449.0,
+                    805.0,
+                    413.0,
+                    805.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    666.0,
+                    772.0,
+                    704.0,
+                    772.0,
+                    704.0,
+                    804.0,
+                    666.0,
+                    804.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    938.0,
+                    774.0,
+                    970.0,
+                    774.0,
+                    970.0,
+                    803.0,
+                    938.0,
+                    803.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1230.0,
+                    772.0,
+                    1263.0,
+                    772.0,
+                    1263.0,
+                    801.0,
+                    1230.0,
+                    801.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    242.0,
+                    1405.0,
+                    242.0,
+                    1405.0,
+                    276.0,
+                    295.0,
+                    276.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    272.0,
+                    892.0,
+                    272.0,
+                    892.0,
+                    303.0,
+                    295.0,
+                    303.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    73.0,
+                    815.0,
+                    73.0,
+                    815.0,
+                    108.0,
+                    295.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    832.0,
+                    2085.0,
+                    867.0,
+                    2085.0,
+                    867.0,
+                    2124.0,
+                    832.0,
+                    2124.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    1467.0,
+                    971.0,
+                    1467.0,
+                    971.0,
+                    1503.0,
+                    297.0,
+                    1503.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    967.0,
+                    1112.0,
+                    967.0,
+                    1112.0,
+                    1007.0,
+                    294.0,
+                    1007.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1028.0,
+                    1406.0,
+                    1028.0,
+                    1406.0,
+                    1065.0,
+                    294.0,
+                    1065.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1059.0,
+                    1406.0,
+                    1059.0,
+                    1406.0,
+                    1095.0,
+                    295.0,
+                    1095.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1090.0,
+                    1408.0,
+                    1090.0,
+                    1408.0,
+                    1126.0,
+                    295.0,
+                    1126.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1122.0,
+                    1404.0,
+                    1122.0,
+                    1404.0,
+                    1154.0,
+                    295.0,
+                    1154.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1150.0,
+                    1031.0,
+                    1150.0,
+                    1031.0,
+                    1187.0,
+                    292.0,
+                    1187.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1114.0,
+                    1150.0,
+                    1407.0,
+                    1150.0,
+                    1407.0,
+                    1187.0,
+                    1114.0,
+                    1187.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1180.0,
+                    1408.0,
+                    1180.0,
+                    1408.0,
+                    1219.0,
+                    292.0,
+                    1219.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1211.0,
+                    1407.0,
+                    1211.0,
+                    1407.0,
+                    1247.0,
+                    295.0,
+                    1247.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1240.0,
+                    1083.0,
+                    1240.0,
+                    1083.0,
+                    1280.0,
+                    292.0,
+                    1280.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1138.0,
+                    1240.0,
+                    1407.0,
+                    1240.0,
+                    1407.0,
+                    1280.0,
+                    1138.0,
+                    1280.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1272.0,
+                    1406.0,
+                    1272.0,
+                    1406.0,
+                    1308.0,
+                    292.0,
+                    1308.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    289.0,
+                    1295.0,
+                    1196.0,
+                    1295.0,
+                    1196.0,
+                    1348.0,
+                    289.0,
+                    1348.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1294.0,
+                    1295.0,
+                    1409.0,
+                    1295.0,
+                    1409.0,
+                    1348.0,
+                    1294.0,
+                    1348.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1331.0,
+                    367.0,
+                    1331.0,
+                    367.0,
+                    1371.0,
+                    292.0,
+                    1371.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    465.0,
+                    1331.0,
+                    1409.0,
+                    1331.0,
+                    1409.0,
+                    1371.0,
+                    465.0,
+                    1371.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1360.0,
+                    1406.0,
+                    1360.0,
+                    1406.0,
+                    1401.0,
+                    291.0,
+                    1401.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1397.0,
+                    441.0,
+                    1397.0,
+                    441.0,
+                    1427.0,
+                    296.0,
+                    1427.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1649.0,
+                    294.0,
+                    1649.0,
+                    294.0,
+                    1687.0,
+                    291.0,
+                    1687.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    320.0,
+                    1649.0,
+                    697.0,
+                    1649.0,
+                    697.0,
+                    1687.0,
+                    320.0,
+                    1687.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    724.0,
+                    1649.0,
+                    1406.0,
+                    1649.0,
+                    1406.0,
+                    1687.0,
+                    724.0,
+                    1687.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1681.0,
+                    1124.0,
+                    1681.0,
+                    1124.0,
+                    1718.0,
+                    294.0,
+                    1718.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1175.0,
+                    1681.0,
+                    1407.0,
+                    1681.0,
+                    1407.0,
+                    1718.0,
+                    1175.0,
+                    1718.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1712.0,
+                    1121.0,
+                    1712.0,
+                    1121.0,
+                    1746.0,
+                    294.0,
+                    1746.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1147.0,
+                    1712.0,
+                    1404.0,
+                    1712.0,
+                    1404.0,
+                    1746.0,
+                    1147.0,
+                    1746.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    329.0,
+                    1742.0,
+                    375.0,
+                    1742.0,
+                    375.0,
+                    1776.0,
+                    329.0,
+                    1776.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    421.0,
+                    1742.0,
+                    1405.0,
+                    1742.0,
+                    1405.0,
+                    1776.0,
+                    421.0,
+                    1776.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1770.0,
+                    1405.0,
+                    1770.0,
+                    1405.0,
+                    1812.0,
+                    291.0,
+                    1812.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1801.0,
+                    1406.0,
+                    1801.0,
+                    1406.0,
+                    1841.0,
+                    292.0,
+                    1841.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1836.0,
+                    1406.0,
+                    1836.0,
+                    1406.0,
+                    1869.0,
+                    295.0,
+                    1869.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1862.0,
+                    1284.0,
+                    1862.0,
+                    1284.0,
+                    1900.0,
+                    294.0,
+                    1900.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1907.0,
+                    586.0,
+                    1907.0,
+                    586.0,
+                    1947.0,
+                    293.0,
+                    1947.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    612.0,
+                    1907.0,
+                    1407.0,
+                    1907.0,
+                    1407.0,
+                    1947.0,
+                    612.0,
+                    1947.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1941.0,
+                    379.0,
+                    1941.0,
+                    379.0,
+                    1976.0,
+                    293.0,
+                    1976.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    458.0,
+                    1941.0,
+                    1404.0,
+                    1941.0,
+                    1404.0,
+                    1976.0,
+                    458.0,
+                    1976.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1973.0,
+                    1406.0,
+                    1973.0,
+                    1406.0,
+                    2005.0,
+                    294.0,
+                    2005.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    2000.0,
+                    1049.0,
+                    2000.0,
+                    1049.0,
+                    2036.0,
+                    293.0,
+                    2036.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1074.0,
+                    2000.0,
+                    1407.0,
+                    2000.0,
+                    1407.0,
+                    2036.0,
+                    1074.0,
+                    2036.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    1570.0,
+                    1071.0,
+                    1570.0,
+                    1071.0,
+                    1608.0,
+                    322.0,
+                    1608.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1098.0,
+                    1570.0,
+                    1403.0,
+                    1570.0,
+                    1403.0,
+                    1608.0,
+                    1098.0,
+                    1608.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1598.0,
+                    445.0,
+                    1598.0,
+                    445.0,
+                    1643.0,
+                    294.0,
+                    1643.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    497.0,
+                    1598.0,
+                    634.0,
+                    1598.0,
+                    634.0,
+                    1643.0,
+                    497.0,
+                    1643.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    685.0,
+                    1598.0,
+                    924.0,
+                    1598.0,
+                    924.0,
+                    1643.0,
+                    685.0,
+                    1643.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    955.0,
+                    1598.0,
+                    967.0,
+                    1598.0,
+                    967.0,
+                    1643.0,
+                    955.0,
+                    1643.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1522.0,
+                    1127.0,
+                    1522.0,
+                    1127.0,
+                    1563.0,
+                    294.0,
+                    1563.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    967.0,
+                    1112.0,
+                    967.0,
+                    1112.0,
+                    1007.0,
+                    294.0,
+                    1007.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 16,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    711,
+                    1403,
+                    711,
+                    1403,
+                    839,
+                    297,
+                    839
+                ],
+                "score": 0.978
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    336,
+                    1405,
+                    336,
+                    1405,
+                    462,
+                    297,
+                    462
+                ],
+                "score": 0.975
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1900,
+                    1405,
+                    1900,
+                    1405,
+                    1994,
+                    297,
+                    1994
+                ],
+                "score": 0.974
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    300,
+                    1069,
+                    1401,
+                    1069,
+                    1401,
+                    1165,
+                    300,
+                    1165
+                ],
+                "score": 0.974
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    319,
+                    862,
+                    1345,
+                    862,
+                    1345,
+                    1050,
+                    319,
+                    1050
+                ],
+                "score": 0.972
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    454,
+                    1534,
+                    1245,
+                    1534,
+                    1245,
+                    1730,
+                    454,
+                    1730
+                ],
+                "score": 0.969
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    411,
+                    1186,
+                    1291,
+                    1186,
+                    1291,
+                    1417,
+                    411,
+                    1417
+                ],
+                "score": 0.968
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    228,
+                    1401,
+                    228,
+                    1401,
+                    324,
+                    297,
+                    324
+                ],
+                "score": 0.963
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    1439,
+                    1405,
+                    1439,
+                    1405,
+                    1513,
+                    299,
+                    1513
+                ],
+                "score": 0.954
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1750,
+                    1400,
+                    1750,
+                    1400,
+                    1818,
+                    297,
+                    1818
+                ],
+                "score": 0.949
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    479,
+                    468,
+                    1191,
+                    468,
+                    1191,
+                    556,
+                    479,
+                    556
+                ],
+                "score": 0.946
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    677,
+                    1997,
+                    1020,
+                    1997,
+                    1020,
+                    2039,
+                    677,
+                    2039
+                ],
+                "score": 0.943
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    467,
+                    1836,
+                    1229,
+                    1836,
+                    1229,
+                    1882,
+                    467,
+                    1882
+                ],
+                "score": 0.932
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    573,
+                    1403,
+                    573,
+                    1403,
+                    609,
+                    297,
+                    609
+                ],
+                "score": 0.931
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    299,
+                    652,
+                    663,
+                    652,
+                    663,
+                    687,
+                    299,
+                    687
+                ],
+                "score": 0.924
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    299,
+                    74,
+                    813,
+                    74,
+                    813,
+                    105,
+                    299,
+                    105
+                ],
+                "score": 0.92
+            },
+            {
+                "category_id": 9,
+                "poly": [
+                    1352,
+                    1843,
+                    1399,
+                    1843,
+                    1399,
+                    1874,
+                    1352,
+                    1874
+                ],
+                "score": 0.896
+            },
+            {
+                "category_id": 9,
+                "poly": [
+                    1351,
+                    2003,
+                    1400,
+                    2003,
+                    1400,
+                    2034,
+                    1351,
+                    2034
+                ],
+                "score": 0.894
+            },
+            {
+                "category_id": 9,
+                "poly": [
+                    1365,
+                    496,
+                    1400,
+                    496,
+                    1400,
+                    526,
+                    1365,
+                    526
+                ],
+                "score": 0.879
+            },
+            {
+                "category_id": 9,
+                "poly": [
+                    1365,
+                    1287,
+                    1401,
+                    1287,
+                    1401,
+                    1317,
+                    1365,
+                    1317
+                ],
+                "score": 0.873
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    835,
+                    2087,
+                    864,
+                    2087,
+                    864,
+                    2113,
+                    835,
+                    2113
+                ],
+                "score": 0.872
+            },
+            {
+                "category_id": 9,
+                "poly": [
+                    1365,
+                    1615,
+                    1401,
+                    1615,
+                    1401,
+                    1645,
+                    1365,
+                    1645
+                ],
+                "score": 0.868
+            },
+            {
+                "category_id": 9,
+                "poly": [
+                    1365,
+                    939,
+                    1401,
+                    939,
+                    1401,
+                    969,
+                    1365,
+                    969
+                ],
+                "score": 0.855
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    451,
+                    1531,
+                    1244,
+                    1531,
+                    1244,
+                    1732,
+                    451,
+                    1732
+                ],
+                "score": 0.95,
+                "latex": "\\begin{array} { r l } & { \\mathcal { L } _ { \\mathrm { S E E D } } = \\displaystyle \\frac { 1 } { N } \\sum _ { i } ^ { N } \\Big \\{ { \\bf w } _ { K + 1 } ^ { i } \\cdot \\big ( - { \\bf z } _ { i } ^ { S } \\cdot { \\bf z } _ { i } ^ { T } / \\tau ^ { S } + \\log \\sum _ { \\bf d \\sim D ^ { + } } \\exp ( { \\bf z } _ { i } ^ { S } \\cdot { \\bf d } / \\tau ^ { S } ) \\big ) } \\\\ & { \\qquad + \\displaystyle \\sum _ { j = 1 } ^ { K } { \\bf w } _ { j } ^ { i } \\cdot \\big ( - { \\bf z } _ { i } ^ { S } \\cdot { \\bf d } _ { j } / \\tau ^ { S } + \\log \\sum _ { \\bf d \\sim D ^ { + } } \\exp ( { \\bf z } _ { i } ^ { S } \\cdot { \\bf d } / \\tau ^ { S } ) \\big ) \\Big \\} } \\end{array}"
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    409,
+                    1181,
+                    1289,
+                    1181,
+                    1289,
+                    1420,
+                    409,
+                    1420
+                ],
+                "score": 0.94,
+                "latex": "\\begin{array} { r l } & { \\mathcal { L } _ { \\mathrm { S E E D } } = \\displaystyle \\frac { 1 } { N } \\sum _ { i } ^ { N } - \\mathbf { p } ^ { T } ( \\mathbf { x } _ { i } ; \\boldsymbol { \\theta } _ { T } , \\mathbf { D } ^ { + } ) \\cdot \\log \\mathbf { p } ^ { S } ( \\mathbf { x } _ { i } ; \\boldsymbol { \\theta } _ { S } , \\mathbf { D } ^ { + } ) } \\\\ & { \\qquad = \\displaystyle \\sum _ { i } ^ { N } \\sum _ { j } ^ { K + 1 } - \\underbrace { \\frac { \\exp \\left( \\mathbf { z } _ { i } ^ { T } \\cdot \\mathbf { d } _ { j } / \\tau ^ { T } \\right) } { \\sum _ { \\mathbf { d } \\sim \\mathbf { D } ^ { + } } \\exp \\left( \\mathbf { z } _ { i } ^ { T } \\cdot \\mathbf { d } / \\tau ^ { T } \\right) } } _ { \\mathbf { w } _ { j } ^ { i } } \\cdot \\log \\frac { \\exp \\left( \\mathbf { z } _ { i } ^ { S } \\cdot \\mathbf { d } _ { j } / \\tau ^ { S } \\right) } { \\sum _ { \\mathbf { d } \\sim \\mathbf { D } ^ { + } } \\exp \\left( \\mathbf { z } _ { i } ^ { S } \\cdot \\mathbf { d } / \\tau ^ { S } \\right) } , } \\end{array}"
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    480,
+                    463,
+                    1188,
+                    463,
+                    1188,
+                    558,
+                    480,
+                    558
+                ],
+                "score": 0.94,
+                "latex": "\\hat { \\theta } _ { S } = \\underset { \\theta _ { S } } { \\arg \\operatorname* { m i n } } \\sum _ { i } ^ { N } \\log h ( \\mathbf { z } _ { i } ^ { S } , \\mathbf { z } _ { i } ^ { T } ) + K \\cdot [ \\log h ( 1 - h ( \\mathbf { z } _ { i } ^ { S } , \\mathbf { d } _ { j } ^ { T } ) ) ] ,"
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    319,
+                    856,
+                    1343,
+                    856,
+                    1343,
+                    1054,
+                    319,
+                    1054
+                ],
+                "score": 0.94,
+                "latex": "\\begin{array} { l } { \\displaystyle \\mathcal { L } = \\frac { 1 } { N } \\sum _ { i } ^ { N } \\bigg \\{ \\lambda _ { a } \\cdot \\bigg \\vert \\bigg \\vert \\mathbf { z } _ { i } ^ { T } - \\mathbf { z } _ { i } ^ { S } \\bigg \\vert \\bigg \\vert _ { 2 } - \\lambda _ { b } \\cdot \\mathbf { p } ^ { T } ( \\mathbf { x } _ { i } ; \\boldsymbol { \\theta } _ { T } , \\mathbf { D } ) \\cdot \\log \\mathbf { p } ^ { S } ( \\mathbf { x } _ { i } ; \\boldsymbol { \\theta } _ { S } , \\mathbf { D } ) \\bigg \\} } \\\\ { \\displaystyle \\quad = \\sum _ { i } ^ { N } \\bigg \\{ - \\lambda _ { a } \\cdot \\mathbf { z } _ { i } ^ { T } \\cdot \\mathbf { z } _ { i } ^ { S } - \\lambda _ { b } \\cdot \\sum _ { j } ^ { K } \\frac { \\exp ( \\mathbf { z } _ { i } ^ { T } \\cdot \\mathbf { d } _ { j } / \\tau ^ { T } ) } { \\sum _ { \\mathbf { d } \\setminus \\mathbf { D } } \\exp ( \\mathbf { z } _ { i } ^ { T } \\cdot \\mathbf { d } / \\tau ^ { T } ) } \\cdot \\log \\frac { \\exp ( \\mathbf { z } _ { i } ^ { S } \\cdot \\mathbf { d } _ { j } / \\tau ^ { S } ) } { \\sum _ { \\mathbf { d } \\setminus \\mathbf { D } } \\exp ( \\mathbf { z } _ { i } ^ { S } \\cdot \\mathbf { d } / \\tau ^ { S } ) } \\bigg \\} , } \\end{array}"
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    674,
+                    1995,
+                    1023,
+                    1995,
+                    1023,
+                    2038,
+                    674,
+                    2038
+                ],
+                "score": 0.91,
+                "latex": "\\mathrm { L S E } ( \\mathbf { D } ^ { + } , \\mathbf { z } _ { i } ^ { S } ) \\geq \\mathrm { L S E } ( \\mathbf { D } , \\mathbf { z } _ { i } ^ { S } ) ."
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    516,
+                    1780,
+                    605,
+                    1780,
+                    605,
+                    1819,
+                    516,
+                    1819
+                ],
+                "score": 0.91,
+                "latex": "\\left[ - 1 , + 1 \\right]"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    356,
+                    1931,
+                    475,
+                    1931,
+                    475,
+                    1962,
+                    356,
+                    1962
+                ],
+                "score": 0.9,
+                "latex": "\\tau ^ { S } = 0 . 2"
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    469,
+                    1835,
+                    1226,
+                    1835,
+                    1226,
+                    1881,
+                    469,
+                    1881
+                ],
+                "score": 0.9,
+                "latex": "\\mathrm { L S E } ( \\mathbf { D } ^ { + } , \\mathbf { z } _ { i } ^ { S } ) \\geq \\log \\big ( M \\cdot \\exp ( - 1 / \\tau ^ { S } ) \\big ) = \\log \\big ( M \\cdot \\exp ( - 5 ) \\big ) > 0 ,"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    444,
+                    1441,
+                    481,
+                    1441,
+                    481,
+                    1480,
+                    444,
+                    1480
+                ],
+                "score": 0.9,
+                "latex": "\\mathbf { w } _ { j } ^ { i }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    602,
+                    1477,
+                    636,
+                    1477,
+                    636,
+                    1513,
+                    602,
+                    1513
+                ],
+                "score": 0.9,
+                "latex": "{ \\mathbf z } _ { i } ^ { T }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    933,
+                    1900,
+                    979,
+                    1900,
+                    979,
+                    1930,
+                    933,
+                    1930
+                ],
+                "score": 0.89,
+                "latex": "\\mathbf { D } ^ { + }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    371,
+                    572,
+                    525,
+                    572,
+                    525,
+                    613,
+                    371,
+                    613
+                ],
+                "score": 0.89,
+                "latex": "\\mathbf { d } _ { j } ^ { T } \\in \\mathbf { D } , h ( \\cdot )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    298,
+                    1099,
+                    422,
+                    1099,
+                    422,
+                    1134,
+                    298,
+                    1134
+                ],
+                "score": 0.89,
+                "latex": "( \\lambda _ { a } , \\lambda _ { b } , \\bar { \\tau } ^ { T }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    977,
+                    576,
+                    1152,
+                    576,
+                    1152,
+                    608,
+                    977,
+                    608
+                ],
+                "score": 0.88,
+                "latex": "\\{ { \\bf z } , { \\bf d } \\}  [ 0 , 1 ]"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    833,
+                    1101,
+                    879,
+                    1101,
+                    879,
+                    1130,
+                    833,
+                    1130
+                ],
+                "score": 0.88,
+                "latex": "\\mathbf { D } ^ { + }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    597,
+                    743,
+                    631,
+                    743,
+                    631,
+                    774,
+                    597,
+                    774
+                ],
+                "score": 0.88,
+                "latex": "\\mathbf { z } ^ { T }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    959,
+                    1440,
+                    995,
+                    1440,
+                    995,
+                    1470,
+                    959,
+                    1470
+                ],
+                "score": 0.88,
+                "latex": "\\tau ^ { T }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1135,
+                    743,
+                    1169,
+                    743,
+                    1169,
+                    774,
+                    1135,
+                    774
+                ],
+                "score": 0.87,
+                "latex": "\\mathbf { z } ^ { S }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    720,
+                    404,
+                    751,
+                    404,
+                    751,
+                    430,
+                    720,
+                    430
+                ],
+                "score": 0.87,
+                "latex": "\\mathbf { x } _ { i }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1359,
+                    1441,
+                    1405,
+                    1441,
+                    1405,
+                    1471,
+                    1359,
+                    1471
+                ],
+                "score": 0.86,
+                "latex": "\\mathbf { D } ^ { + }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    786,
+                    292,
+                    860,
+                    292,
+                    860,
+                    321,
+                    786,
+                    321
+                ],
+                "score": 0.86,
+                "latex": "5 6 . 4 \\%"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1367,
+                    1749,
+                    1400,
+                    1749,
+                    1400,
+                    1779,
+                    1367,
+                    1779
+                ],
+                "score": 0.85,
+                "latex": "\\mathbf { z } ^ { S }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1367,
+                    713,
+                    1400,
+                    713,
+                    1400,
+                    743,
+                    1367,
+                    743
+                ],
+                "score": 0.83,
+                "latex": "\\mathbf { z } ^ { S }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    473,
+                    1100,
+                    507,
+                    1100,
+                    507,
+                    1130,
+                    473,
+                    1130
+                ],
+                "score": 0.82,
+                "latex": "\\tau ^ { S }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    371,
+                    1903,
+                    403,
+                    1903,
+                    403,
+                    1929,
+                    371,
+                    1929
+                ],
+                "score": 0.81,
+                "latex": "M"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1118,
+                    1443,
+                    1210,
+                    1443,
+                    1210,
+                    1474,
+                    1118,
+                    1474
+                ],
+                "score": 0.8,
+                "latex": "( K + 1 )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    297,
+                    808,
+                    322,
+                    808,
+                    322,
+                    833,
+                    297,
+                    833
+                ],
+                "score": 0.72,
+                "latex": "l 2"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    296,
+                    231,
+                    319,
+                    231,
+                    319,
+                    258,
+                    296,
+                    258
+                ],
+                "score": 0.67,
+                "latex": "K"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1246,
+                    400,
+                    1274,
+                    400,
+                    1274,
+                    428,
+                    1246,
+                    428
+                ],
+                "score": 0.64,
+                "latex": "\\mathbf { D }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    454,
+                    804,
+                    531,
+                    804,
+                    531,
+                    836,
+                    454,
+                    836
+                ],
+                "score": 0.59,
+                "latex": "{ \\mathbf z } ^ { T } , { \\mathbf z } ^ { S }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    345,
+                    1784,
+                    366,
+                    1784,
+                    366,
+                    1811,
+                    345,
+                    1811
+                ],
+                "score": 0.56,
+                "latex": "\\mathbf { d }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1077,
+                    806,
+                    1105,
+                    806,
+                    1105,
+                    834,
+                    1077,
+                    834
+                ],
+                "score": 0.52,
+                "latex": "\\mathbf { D }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    944,
+                    577,
+                    963,
+                    577,
+                    963,
+                    604,
+                    944,
+                    604
+                ],
+                "score": 0.49,
+                "latex": "h"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    739,
+                    777,
+                    767,
+                    777,
+                    767,
+                    804,
+                    739,
+                    804
+                ],
+                "score": 0.37,
+                "latex": "\\mathbf { D }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    499,
+                    804,
+                    531,
+                    804,
+                    531,
+                    833,
+                    499,
+                    833
+                ],
+                "score": 0.35,
+                "latex": "\\mathbf { z } ^ { S }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    454,
+                    804,
+                    487,
+                    804,
+                    487,
+                    833,
+                    454,
+                    833
+                ],
+                "score": 0.35,
+                "latex": "\\mathbf { z } ^ { T }"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    647.0,
+                    666.0,
+                    647.0,
+                    666.0,
+                    692.0,
+                    291.0,
+                    692.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    72.0,
+                    815.0,
+                    72.0,
+                    815.0,
+                    108.0,
+                    295.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    831.0,
+                    2085.0,
+                    869.0,
+                    2085.0,
+                    869.0,
+                    2122.0,
+                    831.0,
+                    2122.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    711.0,
+                    1366.0,
+                    711.0,
+                    1366.0,
+                    750.0,
+                    292.0,
+                    750.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1401.0,
+                    711.0,
+                    1404.0,
+                    711.0,
+                    1404.0,
+                    750.0,
+                    1401.0,
+                    750.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    744.0,
+                    596.0,
+                    744.0,
+                    596.0,
+                    779.0,
+                    292.0,
+                    779.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    632.0,
+                    744.0,
+                    1134.0,
+                    744.0,
+                    1134.0,
+                    779.0,
+                    632.0,
+                    779.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1170.0,
+                    744.0,
+                    1406.0,
+                    744.0,
+                    1406.0,
+                    779.0,
+                    1170.0,
+                    779.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    776.0,
+                    738.0,
+                    776.0,
+                    738.0,
+                    809.0,
+                    295.0,
+                    809.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    768.0,
+                    776.0,
+                    1404.0,
+                    776.0,
+                    1404.0,
+                    809.0,
+                    768.0,
+                    809.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    802.0,
+                    296.0,
+                    802.0,
+                    296.0,
+                    842.0,
+                    292.0,
+                    842.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    802.0,
+                    453.0,
+                    802.0,
+                    453.0,
+                    842.0,
+                    323.0,
+                    842.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    532.0,
+                    802.0,
+                    1076.0,
+                    802.0,
+                    1076.0,
+                    842.0,
+                    532.0,
+                    842.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1106.0,
+                    802.0,
+                    1116.0,
+                    802.0,
+                    1116.0,
+                    842.0,
+                    1106.0,
+                    842.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    338.0,
+                    1403.0,
+                    338.0,
+                    1403.0,
+                    371.0,
+                    295.0,
+                    371.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    367.0,
+                    1405.0,
+                    367.0,
+                    1405.0,
+                    403.0,
+                    294.0,
+                    403.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    397.0,
+                    719.0,
+                    397.0,
+                    719.0,
+                    434.0,
+                    292.0,
+                    434.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    752.0,
+                    397.0,
+                    1245.0,
+                    397.0,
+                    1245.0,
+                    434.0,
+                    752.0,
+                    434.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1275.0,
+                    397.0,
+                    1405.0,
+                    397.0,
+                    1405.0,
+                    434.0,
+                    1275.0,
+                    434.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    428.0,
+                    406.0,
+                    428.0,
+                    406.0,
+                    467.0,
+                    292.0,
+                    467.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1897.0,
+                    370.0,
+                    1897.0,
+                    370.0,
+                    1936.0,
+                    292.0,
+                    1936.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    404.0,
+                    1897.0,
+                    932.0,
+                    1897.0,
+                    932.0,
+                    1936.0,
+                    404.0,
+                    1936.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    980.0,
+                    1897.0,
+                    1405.0,
+                    1897.0,
+                    1405.0,
+                    1936.0,
+                    980.0,
+                    1936.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1926.0,
+                    355.0,
+                    1926.0,
+                    355.0,
+                    1970.0,
+                    292.0,
+                    1970.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    476.0,
+                    1926.0,
+                    1406.0,
+                    1926.0,
+                    1406.0,
+                    1970.0,
+                    476.0,
+                    1970.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1956.0,
+                    427.0,
+                    1956.0,
+                    427.0,
+                    2004.0,
+                    291.0,
+                    2004.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1067.0,
+                    1405.0,
+                    1067.0,
+                    1405.0,
+                    1107.0,
+                    294.0,
+                    1107.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1097.0,
+                    297.0,
+                    1097.0,
+                    297.0,
+                    1139.0,
+                    294.0,
+                    1139.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    423.0,
+                    1097.0,
+                    472.0,
+                    1097.0,
+                    472.0,
+                    1139.0,
+                    423.0,
+                    1139.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    508.0,
+                    1097.0,
+                    832.0,
+                    1097.0,
+                    832.0,
+                    1139.0,
+                    508.0,
+                    1139.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    880.0,
+                    1097.0,
+                    1407.0,
+                    1097.0,
+                    1407.0,
+                    1139.0,
+                    880.0,
+                    1139.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1132.0,
+                    818.0,
+                    1132.0,
+                    818.0,
+                    1166.0,
+                    296.0,
+                    1166.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    227.0,
+                    295.0,
+                    227.0,
+                    295.0,
+                    266.0,
+                    291.0,
+                    266.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    320.0,
+                    227.0,
+                    1406.0,
+                    227.0,
+                    1406.0,
+                    266.0,
+                    320.0,
+                    266.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    263.0,
+                    1405.0,
+                    263.0,
+                    1405.0,
+                    294.0,
+                    293.0,
+                    294.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    292.0,
+                    785.0,
+                    292.0,
+                    785.0,
+                    324.0,
+                    293.0,
+                    324.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    861.0,
+                    292.0,
+                    873.0,
+                    292.0,
+                    873.0,
+                    324.0,
+                    861.0,
+                    324.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1437.0,
+                    443.0,
+                    1437.0,
+                    443.0,
+                    1479.0,
+                    293.0,
+                    1479.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    482.0,
+                    1437.0,
+                    958.0,
+                    1437.0,
+                    958.0,
+                    1479.0,
+                    482.0,
+                    1479.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    996.0,
+                    1437.0,
+                    1117.0,
+                    1437.0,
+                    1117.0,
+                    1479.0,
+                    996.0,
+                    1479.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1211.0,
+                    1437.0,
+                    1358.0,
+                    1437.0,
+                    1358.0,
+                    1479.0,
+                    1211.0,
+                    1479.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1477.0,
+                    601.0,
+                    1477.0,
+                    601.0,
+                    1515.0,
+                    295.0,
+                    1515.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    637.0,
+                    1477.0,
+                    1117.0,
+                    1477.0,
+                    1117.0,
+                    1515.0,
+                    637.0,
+                    1515.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1746.0,
+                    1366.0,
+                    1746.0,
+                    1366.0,
+                    1787.0,
+                    292.0,
+                    1787.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1401.0,
+                    1746.0,
+                    1404.0,
+                    1746.0,
+                    1404.0,
+                    1787.0,
+                    1401.0,
+                    1787.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1780.0,
+                    344.0,
+                    1780.0,
+                    344.0,
+                    1818.0,
+                    295.0,
+                    1818.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    367.0,
+                    1780.0,
+                    515.0,
+                    1780.0,
+                    515.0,
+                    1818.0,
+                    367.0,
+                    1818.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    606.0,
+                    1780.0,
+                    618.0,
+                    1780.0,
+                    618.0,
+                    1818.0,
+                    606.0,
+                    1818.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    568.0,
+                    370.0,
+                    568.0,
+                    370.0,
+                    615.0,
+                    292.0,
+                    615.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    526.0,
+                    568.0,
+                    943.0,
+                    568.0,
+                    943.0,
+                    615.0,
+                    526.0,
+                    615.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    964.0,
+                    568.0,
+                    976.0,
+                    568.0,
+                    976.0,
+                    615.0,
+                    964.0,
+                    615.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1153.0,
+                    568.0,
+                    1408.0,
+                    568.0,
+                    1408.0,
+                    615.0,
+                    1153.0,
+                    615.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 17,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    838,
+                    1407,
+                    838,
+                    1407,
+                    995,
+                    298,
+                    995
+                ],
+                "score": 0.979
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    295,
+                    1872,
+                    1405,
+                    1872,
+                    1405,
+                    2034,
+                    295,
+                    2034
+                ],
+                "score": 0.977
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    621,
+                    1405,
+                    621,
+                    1405,
+                    746,
+                    298,
+                    746
+                ],
+                "score": 0.976
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    301,
+                    310,
+                    1399,
+                    310,
+                    1399,
+                    593,
+                    301,
+                    593
+                ],
+                "score": 0.972
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    302,
+                    1421,
+                    1399,
+                    1421,
+                    1399,
+                    1658,
+                    302,
+                    1658
+                ],
+                "score": 0.968
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    530,
+                    1194,
+                    1167,
+                    1194,
+                    1167,
+                    1328,
+                    530,
+                    1328
+                ],
+                "score": 0.964
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    296,
+                    1113,
+                    1401,
+                    1113,
+                    1401,
+                    1177,
+                    296,
+                    1177
+                ],
+                "score": 0.957
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    466,
+                    1011,
+                    1205,
+                    1011,
+                    1205,
+                    1100,
+                    466,
+                    1100
+                ],
+                "score": 0.957
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    294,
+                    226,
+                    1403,
+                    226,
+                    1403,
+                    295,
+                    294,
+                    295
+                ],
+                "score": 0.949
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    408,
+                    1764,
+                    1290,
+                    1764,
+                    1290,
+                    1857,
+                    408,
+                    1857
+                ],
+                "score": 0.948
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1342,
+                    1403,
+                    1342,
+                    1403,
+                    1408,
+                    297,
+                    1408
+                ],
+                "score": 0.945
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    300,
+                    780,
+                    1130,
+                    780,
+                    1130,
+                    816,
+                    300,
+                    816
+                ],
+                "score": 0.922
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    296,
+                    1681,
+                    1409,
+                    1681,
+                    1409,
+                    1747,
+                    296,
+                    1747
+                ],
+                "score": 0.92
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    299,
+                    74,
+                    813,
+                    74,
+                    813,
+                    105,
+                    299,
+                    105
+                ],
+                "score": 0.917
+            },
+            {
+                "category_id": 9,
+                "poly": [
+                    1352,
+                    1038,
+                    1400,
+                    1038,
+                    1400,
+                    1070,
+                    1352,
+                    1070
+                ],
+                "score": 0.896
+            },
+            {
+                "category_id": 9,
+                "poly": [
+                    1351,
+                    1222,
+                    1400,
+                    1222,
+                    1400,
+                    1253,
+                    1351,
+                    1253
+                ],
+                "score": 0.895
+            },
+            {
+                "category_id": 9,
+                "poly": [
+                    1352,
+                    593,
+                    1400,
+                    593,
+                    1400,
+                    622,
+                    1352,
+                    622
+                ],
+                "score": 0.882
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    835,
+                    2087,
+                    864,
+                    2087,
+                    864,
+                    2113,
+                    835,
+                    2113
+                ],
+                "score": 0.864
+            },
+            {
+                "category_id": 9,
+                "poly": [
+                    1352,
+                    1792,
+                    1400,
+                    1792,
+                    1400,
+                    1824,
+                    1352,
+                    1824
+                ],
+                "score": 0.864
+            },
+            {
+                "category_id": 9,
+                "poly": [
+                    1352,
+                    1656,
+                    1400,
+                    1656,
+                    1400,
+                    1684,
+                    1352,
+                    1684
+                ],
+                "score": 0.842
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    530,
+                    1187,
+                    1169,
+                    1187,
+                    1169,
+                    1330,
+                    530,
+                    1330
+                ],
+                "score": 0.94,
+                "latex": "\\mathcal { L } _ { N C E } = \\sum _ { i } ^ { N } \\Big \\{ \\underbrace { - \\mathbf { z } _ { i } ^ { S } \\cdot \\mathbf { z } _ { i } ^ { T } / \\tau } _ { a l i g n m e n t } + \\underbrace { \\log \\sum _ { \\mathbf { d } \\sim \\mathbf { D } } \\exp ( \\mathbf { z } _ { i } ^ { S } \\cdot \\mathbf { d } / \\tau ) } _ { c o n t r a s t i n g } \\Big \\} ."
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    310,
+                    305,
+                    1403,
+                    305,
+                    1403,
+                    601,
+                    310,
+                    601
+                ],
+                "score": 0.94,
+                "latex": "\\begin{array} { l } { { \\mathcal { L } _ { \\mathrm { S E E D } } = \\displaystyle \\frac { 1 } { N } \\sum _ { i } ^ { N } - { \\bf p } ^ { T } ( { \\bf x } _ { i } ; \\boldsymbol { \\theta } _ { T } , { \\bf D } ^ { + } ) \\cdot \\log { \\bf p } ^ { S } ( { \\bf x } _ { i } ; \\boldsymbol { \\theta } _ { S } , { \\bf D } ^ { + } ) } \\ ~ } \\\\ { { \\displaystyle \\geq \\frac { 1 } { N } \\sum _ { i } ^ { N } \\left\\{ { \\bf w } _ { K + 1 } ^ { i } \\cdot ( - { \\bf z } _ { i } ^ { S } \\cdot { \\bf z } _ { i } ^ { T } / \\tau ^ { S } ) + \\sum _ { j = 1 } ^ { K } { \\bf w } _ { j } ^ { i } \\cdot \\left( - { \\bf z } _ { i } ^ { S } \\cdot { \\bf d } _ { j } / \\tau ^ { S } + \\log \\sum _ { \\bf d \\times D } ^ { C } \\exp ( { \\bf z } _ { i } ^ { S } \\cdot { \\bf d } / \\tau ^ { S } ) \\right) \\right\\} } \\ ~ } \\\\ { { \\displaystyle = \\frac { 1 } { N } \\sum _ { i } ^ { N } \\Bigg \\{ - \\frac { { \\bf w } _ { K + 1 } ^ { i } } { \\tau ^ { S } } \\cdot { \\bf z } _ { i } ^ { S } \\cdot { \\bf z } _ { i } ^ { T } - { \\bf p } ^ { T } ( { \\bf x } _ { i } ; \\boldsymbol { \\theta } _ { T } , { \\bf D } ) \\cdot \\log { \\bf p } ^ { S } ( { \\bf x } _ { i } ; \\boldsymbol { \\theta } _ { S } , { \\bf D } ) \\Bigg \\} } \\ ~ } \\end{array}"
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    466,
+                    1008,
+                    1205,
+                    1008,
+                    1205,
+                    1102,
+                    466,
+                    1102
+                ],
+                "score": 0.94,
+                "latex": "\\hat { \\theta } _ { S } = \\underset { \\theta _ { S } } { \\arg \\operatorname* { m i n } } \\ : \\mathcal { L } _ { N C E } = \\underset { \\theta _ { S } } { \\arg \\operatorname* { m i n } } \\sum _ { i } ^ { N } - \\log \\frac { \\exp ( \\mathbf { z } _ { i } ^ { T } \\cdot \\mathbf { z } _ { i } ^ { S } / \\tau ) } { \\sum _ { \\mathbf { d } \\sim \\mathbf { D } } \\exp ( \\mathbf { z } _ { i } ^ { S } \\cdot \\mathbf { d } / \\tau ) }"
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    310,
+                    1417,
+                    1398,
+                    1417,
+                    1398,
+                    1664,
+                    310,
+                    1664
+                ],
+                "score": 0.94,
+                "latex": "\\begin{array} { r l } & { \\mathcal { L } _ { \\mathrm { S E E D } } = \\displaystyle \\frac { 1 } { N * M } \\sum _ { i } ^ { N } \\sum _ { j } ^ { K + 1 } - \\frac { \\exp ( { \\bf z } _ { i } ^ { T } \\cdot { \\bf d } _ { j } / \\tau ^ { T } ) } { \\sum _ { \\bf d } \\sim \\mathrm { D } ^ { + } } \\cdot \\log \\Big \\frac { \\exp ( { \\bf z } _ { i } ^ { S } \\cdot { \\bf d } _ { j } / \\tau ^ { S } ) } { \\sum _ { \\bf d } \\sim \\mathrm { D } ^ { + } } } \\\\ & { \\quad \\quad \\quad \\quad = \\frac { 1 } { N * M } \\sum _ { i } ^ { N } \\displaystyle \\sum _ { j } ^ { K + 1 } \\underbrace { \\frac { \\exp ( { \\bf z } _ { i } ^ { T } \\cdot { \\bf d } _ { j } / \\tau ^ { T } ) } { \\sum _ { \\bf d } \\sim \\mathrm { D } ^ { + } } } _ { { \\bf w } _ { j } ^ { \\bf d } } \\cdot \\underbrace { ( - { \\bf z } _ { i } ^ { S } \\cdot { \\bf z } _ { i } ^ { T } / \\tau ^ { S } ) } _ { a l i g n m e n t } \\cdot \\underbrace { ( \\frac { \\log \\sum _ { i } ^ { S } \\cdot { \\bf d } _ { i } / \\tau ^ { S } ) } { c \\mathrm { o } n t r a s t i n g } } _ { c o n t r a s t i n g } ) , } \\end{array}"
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    406,
+                    1759,
+                    1290,
+                    1759,
+                    1290,
+                    1859,
+                    406,
+                    1859
+                ],
+                "score": 0.93,
+                "latex": "\\mathcal { L } _ { \\mathrm { S E E D } } = \\frac { 1 } { N * M } \\sum _ { i } ^ { N } \\sum _ { j } ^ { K + 1 } \\mathbf { w } _ { j } ^ { i } \\cdot \\Big \\{ - \\mathbf { z } _ { i } ^ { S } \\cdot \\mathbf { z } _ { i } ^ { T } / \\tau ^ { S } + \\log \\sum _ { \\mathbf { d } \\sim \\mathbf { D } } \\exp ( \\mathbf { z } _ { i } ^ { S } \\cdot \\mathbf { d } / \\tau ^ { S } ) ) \\Big \\} ,"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    354,
+                    1904,
+                    479,
+                    1904,
+                    479,
+                    1942,
+                    354,
+                    1942
+                ],
+                "score": 0.93,
+                "latex": "\\mathbf { w } _ { K + 1 } ^ { i } = \\dot { 1 }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1013,
+                    1682,
+                    1269,
+                    1682,
+                    1269,
+                    1721,
+                    1013,
+                    1721
+                ],
+                "score": 0.92,
+                "latex": "\\mathbf { W } ^ { i } = \\left[ \\mathbf { w } _ { 1 } ^ { i } \\ldots \\mathbf { w } _ { K + 1 } ^ { i } \\right]"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    526,
+                    625,
+                    601,
+                    625,
+                    601,
+                    657,
+                    526,
+                    657
+                ],
+                "score": 0.91,
+                "latex": "\\mathcal { L } _ { \\mathrm { S E E D } }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    644,
+                    1873,
+                    680,
+                    1873,
+                    680,
+                    1903,
+                    644,
+                    1903
+                ],
+                "score": 0.89,
+                "latex": "\\tau ^ { T }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1138,
+                    683,
+                    1173,
+                    683,
+                    1173,
+                    712,
+                    1138,
+                    712
+                ],
+                "score": 0.88,
+                "latex": "\\tau ^ { T }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1231,
+                    1943,
+                    1266,
+                    1943,
+                    1266,
+                    1970,
+                    1231,
+                    1970
+                ],
+                "score": 0.87,
+                "latex": "\\tau ^ { T }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    665,
+                    1940,
+                    700,
+                    1940,
+                    700,
+                    1970,
+                    665,
+                    1970
+                ],
+                "score": 0.87,
+                "latex": "\\tau ^ { T }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1353,
+                    625,
+                    1407,
+                    625,
+                    1407,
+                    654,
+                    1353,
+                    654
+                ],
+                "score": 0.69,
+                "latex": "( l 2 +"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    810,
+                    1873,
+                    855,
+                    1873,
+                    855,
+                    1904,
+                    810,
+                    1904
+                ],
+                "score": 0.67,
+                "latex": "\\mathbf { W } ^ { i }"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    779.0,
+                    1134.0,
+                    779.0,
+                    1134.0,
+                    818.0,
+                    294.0,
+                    818.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    72.0,
+                    815.0,
+                    72.0,
+                    815.0,
+                    108.0,
+                    295.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    831.0,
+                    2085.0,
+                    869.0,
+                    2085.0,
+                    869.0,
+                    2125.0,
+                    831.0,
+                    2125.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    840.0,
+                    1404.0,
+                    840.0,
+                    1404.0,
+                    874.0,
+                    296.0,
+                    874.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    871.0,
+                    1405.0,
+                    871.0,
+                    1405.0,
+                    904.0,
+                    296.0,
+                    904.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    900.0,
+                    1409.0,
+                    900.0,
+                    1409.0,
+                    937.0,
+                    292.0,
+                    937.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    931.0,
+                    1407.0,
+                    931.0,
+                    1407.0,
+                    967.0,
+                    293.0,
+                    967.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    962.0,
+                    868.0,
+                    962.0,
+                    868.0,
+                    998.0,
+                    293.0,
+                    998.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1871.0,
+                    643.0,
+                    1871.0,
+                    643.0,
+                    1909.0,
+                    294.0,
+                    1909.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    681.0,
+                    1871.0,
+                    809.0,
+                    1871.0,
+                    809.0,
+                    1909.0,
+                    681.0,
+                    1909.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    856.0,
+                    1871.0,
+                    1407.0,
+                    1871.0,
+                    1407.0,
+                    1909.0,
+                    856.0,
+                    1909.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1901.0,
+                    353.0,
+                    1901.0,
+                    353.0,
+                    1948.0,
+                    292.0,
+                    1948.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    480.0,
+                    1901.0,
+                    1408.0,
+                    1901.0,
+                    1408.0,
+                    1948.0,
+                    480.0,
+                    1948.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1939.0,
+                    664.0,
+                    1939.0,
+                    664.0,
+                    1976.0,
+                    293.0,
+                    1976.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    701.0,
+                    1939.0,
+                    1230.0,
+                    1939.0,
+                    1230.0,
+                    1976.0,
+                    701.0,
+                    1976.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1267.0,
+                    1939.0,
+                    1406.0,
+                    1939.0,
+                    1406.0,
+                    1976.0,
+                    1267.0,
+                    1976.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1969.0,
+                    1407.0,
+                    1969.0,
+                    1407.0,
+                    2008.0,
+                    294.0,
+                    2008.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1996.0,
+                    377.0,
+                    1996.0,
+                    377.0,
+                    2038.0,
+                    293.0,
+                    2038.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    621.0,
+                    525.0,
+                    621.0,
+                    525.0,
+                    661.0,
+                    293.0,
+                    661.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    602.0,
+                    621.0,
+                    1352.0,
+                    621.0,
+                    1352.0,
+                    661.0,
+                    602.0,
+                    661.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1408.0,
+                    621.0,
+                    1411.0,
+                    621.0,
+                    1411.0,
+                    661.0,
+                    1408.0,
+                    661.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    654.0,
+                    1406.0,
+                    654.0,
+                    1406.0,
+                    687.0,
+                    295.0,
+                    687.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    682.0,
+                    1137.0,
+                    682.0,
+                    1137.0,
+                    721.0,
+                    291.0,
+                    721.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1174.0,
+                    682.0,
+                    1407.0,
+                    682.0,
+                    1407.0,
+                    721.0,
+                    1174.0,
+                    721.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    716.0,
+                    546.0,
+                    716.0,
+                    546.0,
+                    748.0,
+                    293.0,
+                    748.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1113.0,
+                    1403.0,
+                    1113.0,
+                    1403.0,
+                    1149.0,
+                    295.0,
+                    1149.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1145.0,
+                    698.0,
+                    1145.0,
+                    698.0,
+                    1181.0,
+                    295.0,
+                    1181.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    227.0,
+                    1405.0,
+                    227.0,
+                    1405.0,
+                    264.0,
+                    294.0,
+                    264.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    261.0,
+                    557.0,
+                    261.0,
+                    557.0,
+                    297.0,
+                    294.0,
+                    297.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1340.0,
+                    1407.0,
+                    1340.0,
+                    1407.0,
+                    1380.0,
+                    295.0,
+                    1380.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1378.0,
+                    502.0,
+                    1378.0,
+                    502.0,
+                    1410.0,
+                    296.0,
+                    1410.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1679.0,
+                    1012.0,
+                    1679.0,
+                    1012.0,
+                    1723.0,
+                    293.0,
+                    1723.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1270.0,
+                    1679.0,
+                    1405.0,
+                    1679.0,
+                    1405.0,
+                    1723.0,
+                    1270.0,
+                    1723.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1714.0,
+                    579.0,
+                    1714.0,
+                    579.0,
+                    1750.0,
+                    294.0,
+                    1750.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 18,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    487,
+                    1405,
+                    487,
+                    1405,
+                    854,
+                    297,
+                    854
+                ],
+                "score": 0.982
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    287,
+                    1404,
+                    287,
+                    1404,
+                    472,
+                    298,
+                    472
+                ],
+                "score": 0.978
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    1880,
+                    1405,
+                    1880,
+                    1405,
+                    2034,
+                    299,
+                    2034
+                ],
+                "score": 0.975
+            },
+            {
+                "category_id": 3,
+                "poly": [
+                    299,
+                    1280,
+                    1394,
+                    1280,
+                    1394,
+                    1646,
+                    299,
+                    1646
+                ],
+                "score": 0.968
+            },
+            {
+                "category_id": 4,
+                "poly": [
+                    298,
+                    1674,
+                    1405,
+                    1674,
+                    1405,
+                    1760,
+                    298,
+                    1760
+                ],
+                "score": 0.954
+            },
+            {
+                "category_id": 5,
+                "poly": [
+                    543,
+                    1012,
+                    1150,
+                    1012,
+                    1150,
+                    1199,
+                    543,
+                    1199
+                ],
+                "score": 0.952,
+                "html": "<table><tr><td colspan=\"3\">Pre-training Distill. Fine-tuning Distill. Top-1 Acc</td></tr><tr><td>X</td><td>X</td><td>75.2</td></tr><tr><td>√</td><td>×</td><td>75.4</td></tr><tr><td>×</td><td>√</td><td>76.0</td></tr><tr><td>√</td><td>√</td><td>78.1</td></tr></table>"
+            },
+            {
+                "category_id": 6,
+                "poly": [
+                    300,
+                    903,
+                    1404,
+                    903,
+                    1404,
+                    988,
+                    300,
+                    988
+                ],
+                "score": 0.935
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    300,
+                    76,
+                    813,
+                    76,
+                    813,
+                    104,
+                    300,
+                    104
+                ],
+                "score": 0.891
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    304,
+                    1822,
+                    768,
+                    1822,
+                    768,
+                    1853,
+                    304,
+                    1853
+                ],
+                "score": 0.883
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    835,
+                    2088,
+                    865,
+                    2088,
+                    865,
+                    2112,
+                    835,
+                    2112
+                ],
+                "score": 0.851
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    298,
+                    229,
+                    966,
+                    229,
+                    966,
+                    261,
+                    298,
+                    261
+                ],
+                "score": 0.654
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    229,
+                    966,
+                    229,
+                    966,
+                    261,
+                    298,
+                    261
+                ],
+                "score": 0.225
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    971,
+                    411,
+                    1051,
+                    411,
+                    1051,
+                    440,
+                    971,
+                    440
+                ],
+                "score": 0.89,
+                "latex": "3 2 \\times 3 2"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1034,
+                    702,
+                    1111,
+                    702,
+                    1111,
+                    731,
+                    1034,
+                    731
+                ],
+                "score": 0.88,
+                "latex": "7 6 . 0 \\%"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    325,
+                    793,
+                    400,
+                    793,
+                    400,
+                    822,
+                    325,
+                    822
+                ],
+                "score": 0.87,
+                "latex": "7 8 . 1 \\%"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    751,
+                    443,
+                    793,
+                    443,
+                    793,
+                    470,
+                    751,
+                    470
+                ],
+                "score": 0.85,
+                "latex": "= 1"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1081,
+                    2004,
+                    1138,
+                    2004,
+                    1138,
+                    2032,
+                    1081,
+                    2032
+                ],
+                "score": 0.84,
+                "latex": "+ 1 . 5"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    594,
+                    443,
+                    636,
+                    443,
+                    636,
+                    470,
+                    594,
+                    470
+                ],
+                "score": 0.83,
+                "latex": "= 3"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    852,
+                    610,
+                    927,
+                    610,
+                    927,
+                    639,
+                    852,
+                    639
+                ],
+                "score": 0.83,
+                "latex": "7 5 . 2 \\%"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1036,
+                    1943,
+                    1120,
+                    1943,
+                    1120,
+                    1972,
+                    1036,
+                    1972
+                ],
+                "score": 0.81,
+                "latex": "K { = } 1 2 8"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    731,
+                    610,
+                    806,
+                    610,
+                    806,
+                    639,
+                    731,
+                    639
+                ],
+                "score": 0.79,
+                "latex": "7 5 . 4 \\%"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    788,
+                    1703,
+                    883,
+                    1703,
+                    883,
+                    1730,
+                    788,
+                    1730
+                ],
+                "score": 0.71,
+                "latex": "\\mathrm { L R } { = } 0 . 0 3"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    677,
+                    1677,
+                    714,
+                    1677,
+                    714,
+                    1704,
+                    677,
+                    1704
+                ],
+                "score": 0.68,
+                "latex": "( \\% )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    692,
+                    905,
+                    729,
+                    905,
+                    729,
+                    931,
+                    692,
+                    931
+                ],
+                "score": 0.51,
+                "latex": "( \\% )"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1271.0,
+                    342.0,
+                    1271.0,
+                    342.0,
+                    1626.0,
+                    295.0,
+                    1626.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    331.0,
+                    1294.0,
+                    382.0,
+                    1294.0,
+                    382.0,
+                    1326.0,
+                    331.0,
+                    1326.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    331.0,
+                    1350.0,
+                    381.0,
+                    1350.0,
+                    381.0,
+                    1382.0,
+                    331.0,
+                    1382.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    330.0,
+                    1402.0,
+                    382.0,
+                    1402.0,
+                    382.0,
+                    1438.0,
+                    330.0,
+                    1438.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    330.0,
+                    1458.0,
+                    382.0,
+                    1458.0,
+                    382.0,
+                    1493.0,
+                    330.0,
+                    1493.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    330.0,
+                    1516.0,
+                    381.0,
+                    1516.0,
+                    381.0,
+                    1548.0,
+                    330.0,
+                    1548.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    332.0,
+                    1575.0,
+                    381.0,
+                    1575.0,
+                    381.0,
+                    1603.0,
+                    332.0,
+                    1603.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    372.0,
+                    1594.0,
+                    387.0,
+                    1594.0,
+                    387.0,
+                    1610.0,
+                    372.0,
+                    1610.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    621.0,
+                    1590.0,
+                    641.0,
+                    1590.0,
+                    641.0,
+                    1612.0,
+                    621.0,
+                    1612.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    747.0,
+                    1589.0,
+                    767.0,
+                    1589.0,
+                    767.0,
+                    1612.0,
+                    747.0,
+                    1612.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    871.0,
+                    1589.0,
+                    891.0,
+                    1589.0,
+                    891.0,
+                    1613.0,
+                    871.0,
+                    1613.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    996.0,
+                    1588.0,
+                    1020.0,
+                    1588.0,
+                    1020.0,
+                    1615.0,
+                    996.0,
+                    1615.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1118.0,
+                    1587.0,
+                    1150.0,
+                    1587.0,
+                    1150.0,
+                    1616.0,
+                    1118.0,
+                    1616.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1372.0,
+                    1588.0,
+                    1399.0,
+                    1588.0,
+                    1399.0,
+                    1612.0,
+                    1372.0,
+                    1612.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    695.0,
+                    1604.0,
+                    1067.0,
+                    1604.0,
+                    1067.0,
+                    1652.0,
+                    695.0,
+                    1652.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    497.0,
+                    1588.5,
+                    512.0,
+                    1588.5,
+                    512.0,
+                    1612.5,
+                    497.0,
+                    1612.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1674.0,
+                    676.0,
+                    1674.0,
+                    676.0,
+                    1707.0,
+                    295.0,
+                    1707.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    715.0,
+                    1674.0,
+                    1405.0,
+                    1674.0,
+                    1405.0,
+                    1707.0,
+                    715.0,
+                    1707.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1702.0,
+                    787.0,
+                    1702.0,
+                    787.0,
+                    1734.0,
+                    295.0,
+                    1734.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    884.0,
+                    1702.0,
+                    1406.0,
+                    1702.0,
+                    1406.0,
+                    1734.0,
+                    884.0,
+                    1734.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1728.0,
+                    535.0,
+                    1728.0,
+                    535.0,
+                    1765.0,
+                    294.0,
+                    1765.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    900.0,
+                    691.0,
+                    900.0,
+                    691.0,
+                    934.0,
+                    294.0,
+                    934.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    730.0,
+                    900.0,
+                    1408.0,
+                    900.0,
+                    1408.0,
+                    934.0,
+                    730.0,
+                    934.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    930.0,
+                    1406.0,
+                    930.0,
+                    1406.0,
+                    963.0,
+                    294.0,
+                    963.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    961.0,
+                    585.0,
+                    961.0,
+                    585.0,
+                    989.0,
+                    295.0,
+                    989.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    72.0,
+                    816.0,
+                    72.0,
+                    816.0,
+                    110.0,
+                    294.0,
+                    110.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1817.0,
+                    774.0,
+                    1817.0,
+                    774.0,
+                    1858.0,
+                    296.0,
+                    1858.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    829.0,
+                    2084.0,
+                    871.0,
+                    2084.0,
+                    871.0,
+                    2123.0,
+                    829.0,
+                    2123.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    227.0,
+                    970.0,
+                    227.0,
+                    970.0,
+                    265.0,
+                    294.0,
+                    265.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    488.0,
+                    1405.0,
+                    488.0,
+                    1405.0,
+                    522.0,
+                    295.0,
+                    522.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    516.0,
+                    1406.0,
+                    516.0,
+                    1406.0,
+                    554.0,
+                    294.0,
+                    554.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    545.0,
+                    1406.0,
+                    545.0,
+                    1406.0,
+                    585.0,
+                    292.0,
+                    585.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    581.0,
+                    1405.0,
+                    581.0,
+                    1405.0,
+                    615.0,
+                    295.0,
+                    615.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    610.0,
+                    730.0,
+                    610.0,
+                    730.0,
+                    643.0,
+                    294.0,
+                    643.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    807.0,
+                    610.0,
+                    851.0,
+                    610.0,
+                    851.0,
+                    643.0,
+                    807.0,
+                    643.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    928.0,
+                    610.0,
+                    1405.0,
+                    610.0,
+                    1405.0,
+                    643.0,
+                    928.0,
+                    643.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    639.0,
+                    1405.0,
+                    639.0,
+                    1405.0,
+                    675.0,
+                    294.0,
+                    675.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    669.0,
+                    1407.0,
+                    669.0,
+                    1407.0,
+                    706.0,
+                    294.0,
+                    706.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    701.0,
+                    1033.0,
+                    701.0,
+                    1033.0,
+                    735.0,
+                    294.0,
+                    735.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1112.0,
+                    701.0,
+                    1403.0,
+                    701.0,
+                    1403.0,
+                    735.0,
+                    1112.0,
+                    735.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    729.0,
+                    1405.0,
+                    729.0,
+                    1405.0,
+                    765.0,
+                    294.0,
+                    765.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    762.0,
+                    1405.0,
+                    762.0,
+                    1405.0,
+                    796.0,
+                    295.0,
+                    796.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    792.0,
+                    324.0,
+                    792.0,
+                    324.0,
+                    826.0,
+                    294.0,
+                    826.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    401.0,
+                    792.0,
+                    1407.0,
+                    792.0,
+                    1407.0,
+                    826.0,
+                    401.0,
+                    826.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    820.0,
+                    1261.0,
+                    820.0,
+                    1261.0,
+                    859.0,
+                    294.0,
+                    859.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    283.0,
+                    1409.0,
+                    283.0,
+                    1409.0,
+                    327.0,
+                    293.0,
+                    327.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    318.0,
+                    1405.0,
+                    318.0,
+                    1405.0,
+                    354.0,
+                    294.0,
+                    354.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    350.0,
+                    1405.0,
+                    350.0,
+                    1405.0,
+                    382.0,
+                    294.0,
+                    382.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    381.0,
+                    1405.0,
+                    381.0,
+                    1405.0,
+                    414.0,
+                    293.0,
+                    414.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    406.0,
+                    970.0,
+                    406.0,
+                    970.0,
+                    448.0,
+                    292.0,
+                    448.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1052.0,
+                    406.0,
+                    1408.0,
+                    406.0,
+                    1408.0,
+                    448.0,
+                    1052.0,
+                    448.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    441.0,
+                    593.0,
+                    441.0,
+                    593.0,
+                    474.0,
+                    295.0,
+                    474.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    637.0,
+                    441.0,
+                    750.0,
+                    441.0,
+                    750.0,
+                    474.0,
+                    637.0,
+                    474.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    794.0,
+                    441.0,
+                    806.0,
+                    441.0,
+                    806.0,
+                    474.0,
+                    794.0,
+                    474.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1880.0,
+                    1407.0,
+                    1880.0,
+                    1407.0,
+                    1917.0,
+                    295.0,
+                    1917.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1907.0,
+                    1409.0,
+                    1907.0,
+                    1409.0,
+                    1946.0,
+                    294.0,
+                    1946.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1941.0,
+                    1035.0,
+                    1941.0,
+                    1035.0,
+                    1978.0,
+                    294.0,
+                    1978.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1121.0,
+                    1941.0,
+                    1407.0,
+                    1941.0,
+                    1407.0,
+                    1978.0,
+                    1121.0,
+                    1978.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1972.0,
+                    1403.0,
+                    1972.0,
+                    1403.0,
+                    2009.0,
+                    295.0,
+                    2009.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    2002.0,
+                    1080.0,
+                    2002.0,
+                    1080.0,
+                    2039.0,
+                    295.0,
+                    2039.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1139.0,
+                    2002.0,
+                    1406.0,
+                    2002.0,
+                    1406.0,
+                    2039.0,
+                    1139.0,
+                    2039.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    227.0,
+                    970.0,
+                    227.0,
+                    970.0,
+                    265.0,
+                    294.0,
+                    265.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 19,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 5,
+                "poly": [
+                    296,
+                    406,
+                    656,
+                    406,
+                    656,
+                    591,
+                    296,
+                    591
+                ],
+                "score": 0.972,
+                "html": "<table><tr><td>LR</td><td>Top-1 Acc. Top-5 Acc.</td><td></td></tr><tr><td>1</td><td>58.9</td><td>83.1</td></tr><tr><td>0.1</td><td>62.9</td><td>85.3</td></tr><tr><td>0.03</td><td>63.3</td><td>85.4</td></tr><tr><td>0.01</td><td>62.6</td><td>85.0</td></tr></table>"
+            },
+            {
+                "category_id": 5,
+                "poly": [
+                    839,
+                    378,
+                    1200,
+                    378,
+                    1200,
+                    593,
+                    839,
+                    593
+                ],
+                "score": 0.968,
+                "html": "<table><tr><td>WD</td><td>Top-1 Acc. Top-5 Acc.</td><td></td></tr><tr><td>1e-2</td><td>11.8</td><td>27.7</td></tr><tr><td>1e-3</td><td>62.3</td><td>84.7</td></tr><tr><td>1e-4</td><td>61.9</td><td>84.4</td></tr><tr><td>1e-5</td><td>61.6</td><td>84.2</td></tr><tr><td>1e-6</td><td>63.3</td><td>85.4</td></tr></table>"
+            },
+            {
+                "category_id": 6,
+                "poly": [
+                    295,
+                    241,
+                    811,
+                    241,
+                    811,
+                    382,
+                    295,
+                    382
+                ],
+                "score": 0.944
+            },
+            {
+                "category_id": 6,
+                "poly": [
+                    839,
+                    240,
+                    1396,
+                    240,
+                    1396,
+                    353,
+                    839,
+                    353
+                ],
+                "score": 0.942
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    657,
+                    1401,
+                    657,
+                    1401,
+                    720,
+                    298,
+                    720
+                ],
+                "score": 0.904
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    300,
+                    75,
+                    812,
+                    75,
+                    812,
+                    105,
+                    300,
+                    105
+                ],
+                "score": 0.889
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    834,
+                    2088,
+                    863,
+                    2088,
+                    863,
+                    2112,
+                    834,
+                    2112
+                ],
+                "score": 0.835
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1091,
+                    324,
+                    1183,
+                    324,
+                    1183,
+                    351,
+                    1091,
+                    351
+                ],
+                "score": 0.8,
+                "latex": "\\mathrm { L R } { = } 0 . 0 3"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1218,
+                    242,
+                    1256,
+                    242,
+                    1256,
+                    269,
+                    1218,
+                    269
+                ],
+                "score": 0.65,
+                "latex": "( \\% )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    696,
+                    244,
+                    734,
+                    244,
+                    734,
+                    271,
+                    696,
+                    271
+                ],
+                "score": 0.6,
+                "latex": "( \\% )"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    241.0,
+                    695.0,
+                    241.0,
+                    695.0,
+                    272.0,
+                    294.0,
+                    272.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    735.0,
+                    241.0,
+                    813.0,
+                    241.0,
+                    813.0,
+                    272.0,
+                    735.0,
+                    272.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    269.0,
+                    811.0,
+                    269.0,
+                    811.0,
+                    298.0,
+                    296.0,
+                    298.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    297.0,
+                    813.0,
+                    297.0,
+                    813.0,
+                    327.0,
+                    296.0,
+                    327.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    326.0,
+                    812.0,
+                    326.0,
+                    812.0,
+                    358.0,
+                    295.0,
+                    358.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    352.0,
+                    427.0,
+                    352.0,
+                    427.0,
+                    383.0,
+                    296.0,
+                    383.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    838.0,
+                    238.0,
+                    1217.0,
+                    238.0,
+                    1217.0,
+                    271.0,
+                    838.0,
+                    271.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1257.0,
+                    238.0,
+                    1396.0,
+                    238.0,
+                    1396.0,
+                    271.0,
+                    1257.0,
+                    271.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    839.0,
+                    267.0,
+                    1396.0,
+                    267.0,
+                    1396.0,
+                    298.0,
+                    839.0,
+                    298.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    838.0,
+                    296.0,
+                    1396.0,
+                    296.0,
+                    1396.0,
+                    327.0,
+                    838.0,
+                    327.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    838.0,
+                    324.0,
+                    1090.0,
+                    324.0,
+                    1090.0,
+                    352.0,
+                    838.0,
+                    352.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1184.0,
+                    324.0,
+                    1193.0,
+                    324.0,
+                    1193.0,
+                    352.0,
+                    1184.0,
+                    352.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    73.0,
+                    815.0,
+                    73.0,
+                    815.0,
+                    108.0,
+                    295.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    830.0,
+                    2084.0,
+                    868.0,
+                    2084.0,
+                    868.0,
+                    2124.0,
+                    830.0,
+                    2124.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    653.0,
+                    1405.0,
+                    653.0,
+                    1405.0,
+                    695.0,
+                    293.0,
+                    695.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    686.0,
+                    843.0,
+                    686.0,
+                    843.0,
+                    723.0,
+                    294.0,
+                    723.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 20,
+            "width": 1700,
+            "height": 2200
+        }
+    }
+]
\ No newline at end of file
diff --git a/parse/train/B17JTOe0-/B17JTOe0-.md b/parse/train/B17JTOe0-/B17JTOe0-.md
new file mode 100644
index 0000000000000000000000000000000000000000..8c3eaf5f5de0b1e855eeff084dbe496a3ad01582
--- /dev/null
+++ b/parse/train/B17JTOe0-/B17JTOe0-.md
@@ -0,0 +1,308 @@
+# EMERGENCE OF GRID-LIKE REPRESENTATIONS BYTRAINING RECURRENT NEURAL NETWORKS TOPERFORM SPATIAL LOCALIZATION
+
+Christopher J. Cueva∗, Xue-Xin Wei∗ Columbia University New York, NY 10027, USA {ccueva,weixxpku}@gmail.com
+
+# ABSTRACT
+
+Decades of research on the neural code underlying spatial navigation have revealed a diverse set of neural response properties. The Entorhinal Cortex (EC) of the mammalian brain contains a rich set of spatial correlates, including grid cells which encode space using tessellating patterns. However, the mechanisms and functional significance of these spatial representations remain largely mysterious. As a new way to understand these neural representations, we trained recurrent neural networks (RNNs) to perform navigation tasks in 2D arenas based on velocity inputs. Surprisingly, we find that grid-like spatial response patterns emerge in trained networks, along with units that exhibit other spatial correlates, including border cells and band-like cells. All these different functional types of neurons have been observed experimentally. The order of the emergence of grid-like and border cells is also consistent with observations from developmental studies. Together, our results suggest that grid cells, border cells and others as observed in EC may be a natural solution for representing space efficiently given the predominant recurrent connections in the neural circuits.
+
+# 1 INTRODUCTION
+
+Understanding the neural code in the brain has long been driven by studying feed-forward architectures, starting from Hubel and Wiesel’s famous proposal on the origin of orientation selectivity in primary visual cortex (Hubel & Wiesel, 1962). Inspired by the recent development in deep learning (Krizhevsky et al., 2012; LeCun et al., 2015; Hochreiter & Schmidhuber, 1997; Mnih et al., 2015), there has been a burst of interest in applying deep feedforward models, in particular convolutional neural networks (CNN) (LeCun et al., 1998), to study the sensory systems, which hierarchically extract useful features from sensory inputs (see e.g., Yamins et al. (2014); Kriegeskorte (2015); Kietzmann et al. (2017); Yamins & DiCarlo (2016)).
+
+For more cognitive tasks, neural systems often need to maintain certain internal representations of relevant variables in the absence of external stimuli- a process that requires more than feature extraction. We will focus on spatial navigation, which typically requires the brain to maintain a representation of self-location and update it according to the animal’s movements and landmarks of the environment. Physiological studies done in rodents and other mammals (including humans, non-human primates and bats) have revealed a variety of neural correlates of space in Hippocampus and Entorhinal Cortex (EC), including place cells (O’Keefe, 1976), grid cells (Fyhn et al., 2004; Hafting et al., 2005; Fyhn et al., 2008; Yartsev et al., 2011; Killian et al., 2012; Jacobs et al., 2013), along with border cells (Solstad et al., 2008), band-like cells (Krupic et al., 2012) and others (see Figure 1a). In particular, each grid cell only fires when the animal occupies a distinct set of physical locations, and strikingly these locations lie on a lattice. The study of the neural underpinning of spatial cognition has provided an important window into how high-level cognitive functions are supported in the brain (Moser et al., 2008; Aronov et al., 2017).
+
+How might the spatial navigation task be solved using a network of neurons? Recurrent neural networks (RNNs) (Hochreiter & Schmidhuber, 1997; Graves et al., 2013; Oord et al., 2016; Theis & Bethge, 2015; Gregor et al., 2015; Sussillo et al., 2015) seem particularly useful for these tasks. Indeed, recurrent-based continuous attractor networks have been one popular type of models proposed for the formation of grid cells (McNaughton et al., 2006; Burak & Fiete, 2009; Couey et al., 2013) and place cells (Samsonovich & McNaughton, 1997). Such models have provided valuable insights into one set of possible mechanisms that could support the formation of the grids. However, these models typically rely on fine-tuned connectivity patterns, in particular the models need a subtle yet systematic asymmetry in the connectivity pattern to move the attractor state according to the animal’s own movement. The existence of such a specific 2D connectivity in rodent EC remains unclear. Additionally, previous models have mainly focused on grid cells, while other types of responses that co-exist in the Entorhinal Cortex have been largely ignored. It would be useful to have a unified model that can simultaneously explain different types of neural responses in EC.
+
+Motivated by these considerations, here we present an alternative modeling approach for understanding the representation of space in the neural system. Specifically, we trained a RNN to perform some spatial navigation tasks. By leveraging the recent development in RNN training and knowledge of the navigation system in the brain, we show that training a RNN with biologically relevant constraints naturally gives rise to a variety of spatial response profiles as observed in EC, including grid-like responses. To our knowledge, this is the first study to show that grid-like responses could emerge from training a RNN to perform navigation. Our result implies that the neural representation in EC may be seen as a natural way for the brain to solve the navigation task efficiently (Wei et al., 2015). More generally, it suggests that RNNs can be a powerful tool for understanding the neural mechanisms of certain high-level cognitive functions.
+
+![](images/8e6f9cf7feba198b2a18d2ed284d372bf5324606db8951624994c943f119c937.jpg)  
+Figure 1: a) Example neural data showing different kinds of neural correlates underlying spatial navigation in EC. All figures are replotted from previous publications. From left to right: a “grid cell” recorded when an animal navigates in a square environment, replotted from Krupic et al. (2012), with the heat map representing the firing rate of this neuron as a function of the animal’s location (red corresponds to high firing rate); a “band-like” cell from Krupic et al. (2012); a border cell from Solstad et al. (2008); an irregular spatially tuned cell from Diehl et al. (2017); a “speed cell” from Kropff et al. (2015), which exhibits roughly linear dependence on the rodent’s running speed; a “heading direction cell” from Sargolini et al. (2006), which shows systematic change of firing rate depending on animal’s heading direction. b) The network consists of $N = 1 0 0$ recurrently connected units (or neurons) which receive two external inputs, representing the animal’s speed and heading direction. The two outputs linearly weight the neurons in the RNN. The goal of training is to make the responses of the two output neurons accurately represent the animal’s physical location. c) Typical trajectory after training. As shown, the output of the RNN can accurately, though not perfectly, track the animal’s location during navigation.
+
+# 2 MODEL
+
+# 2.1 MODEL DESCRIPTION
+
+Our network model consists of a set of recurrently connected units $N = 1 0 0$ ). The dynamics of each unit in the network $u _ { i } ( t )$ is governed by the standard continuous-time RNN equation:
+
+$$
+\tau \frac { d x _ { i } ( t ) } { d t } = - x _ { i } ( t ) + \sum _ { j = 1 } ^ { N } W _ { i j } ^ { \mathrm { r e c } } u _ { j } ( t ) + \sum _ { k = 1 } ^ { N _ { \mathrm { i n } } } W _ { i k } ^ { \mathrm { i n } } I _ { k } ( t ) + b _ { i } + \xi _ { i } ( t )
+$$
+
+for $i = 1 , \ldots , N$ . The activity of each unit, $u _ { i } ( t )$ , is related to the activation of that unit, $x _ { i } ( t )$ , through a nonlinearity which in this study we take to be $u _ { i } ( t ) = \operatorname { t a n h } ( x _ { i } ( t ) )$ . Each unit receives input from other units through the recurrent weight matrix $W ^ { \mathrm { r e c } }$ and also receives external input, $I ( t )$ , that enters the network through the weight matrix $W ^ { \mathrm { i n } }$ . Each unit has two sources of bias, $b _ { i }$ which is learned and $\xi _ { i } ( t )$ which represents noise intrinsic to the network and is taken to be Gaussian with zero mean and constant variance. The network was simulated using the Euler method for $T = 5 0 0$ timesteps of duration $\tau / 1 0$ .
+
+To perform a 2D navigation task with the RNN, we linearly combine the firing rates of units in the network to estimate the current location of the animal. The responses of the two linear readout neurons, $y _ { 1 } ( t )$ and $y _ { 2 } ( t )$ , are given by the following equation:
+
+$$
+y _ { j } ( t ) = \sum _ { i = 1 } ^ { N } W _ { j i } ^ { \mathrm { o u t } } u _ { i } ( t )
+$$
+
+# 2.2 INPUT TO THE NETWORK
+
+The network inputs and outputs were inspired by simple spatial navigation tasks in 2D open environments. The task resembles dead-reckoning (sometimes referred to as path integration), which is ethologically relevant for many animal species (Darwin, 1873; Mittelstaedt & Mittelstaedt, 1980; Etienne & Jeffery, 2004; McNaughton et al., 2006). To be more specific, the inputs to the network were the animal’s speed and direction at each time step. Experimentally, it has been shown that the velocity signals exist in EC (Sargolini et al., 2006; Kropff et al., 2015; Hinman et al., 2016), and there is also evidence that such signals are necessary for grid formation (Winter et al., 2015a;b).
+
+Throughout the paper, we adopt the common assumption that the head direction of the animal coincides with the actual moving direction. The outputs were the x- and y-coordinates of the integrated position. The direction of the animal is modeled by modified Brownian motion to increase the probability of straight-runs, in order to be consistent with the typical rodent’s behavior in an open environment. The usage of such simple movement statistics has the advantage of having full control of the simulated trajectories. However, for future work it would be very interesting to test the model using different animals’ real movement trajectories to see how the results might change.
+
+Special care is taken when the animal is close to the boundary. The boundary of the environment will affect the statistics of the movement, as the animal cannot cross the boundary. This fact was reflected in the model by re-sampling the angular input variable until the input angle did not lead the animal outside the boundary. In the simulations shown below, the animal always starts from the center of the arena, but we verified that the results are insensitive to the starting locations.
+
+# 2.3 TRAINING
+
+We optimized the network parameters $W ^ { \mathrm { r e c } } , W ^ { \mathrm { i n } } ,$ , $b$ and $W ^ { \mathrm { o u t } }$ to minimize the squared error in equation (3) between target $\mathbf { X } ^ { - }$ and $_ \textrm { y }$ -coordinates from a two dimensional navigation task (performed in rectangular, hexagonal, and triangular arenas) and the network outputs generated according to equation (2).
+
+$$
+E = \frac { 1 } { M T N _ { \mathrm { o u t } } } \sum _ { m , t , j = 1 } ^ { M , T , N _ { \mathrm { o u t } } } ( y _ { j } ( t , m ) - y _ { j } ^ { \mathrm { t a r g e t } } ( t , m ) ) ^ { 2 }
+$$
+
+Parameters were updated with the Hessian-free algorithm (Martens & Sutskever, 2011) using minibatches of size $M = 5 0 0$ trials. In addition to minimizing the error function in equation (3) we regularized the input and output weights according to equation (4) and the squared firing rates of the units (referred to as metabolic cost) according to equation (5). In sum, the training aims to minimize a loss function, that consists of the error of the animal, the metabolic cost, and a penalty for large network parameters.
+
+$$
+\begin{array} { l } { { \displaystyle R _ { L 2 } = \frac { 1 } { N N _ { \mathrm { i n } } } \sum _ { i , j = 1 } ^ { N , N _ { \mathrm { i n } } } ( W _ { i j } ^ { \mathrm { i n } } ) ^ { 2 } + \frac { 1 } { N N _ { \mathrm { o u t } } } \sum _ { i , j = 1 } ^ { N _ { \mathrm { o u t } } , N } ( W _ { i j } ^ { \mathrm { o u t } } ) ^ { 2 } } } \\ { { \displaystyle R _ { F R } = \frac { 1 } { N T M } \sum _ { i , t , m = 1 } ^ { N , T , M } u _ { i } ( t , m ) ^ { 2 } } } \end{array}
+$$
+
+We find that the results are qualitatively insensitive to the initialization schemes used for the recurrent weight matrix $W ^ { \mathrm { r e c } }$ . For the results presented in this paper, simulations in the hexagonal environment were obtained by initializing the elements of $W ^ { \mathrm { r e c } }$ to be zero mean Gaussian random variables with variance $1 . 5 ^ { 2 } \dot { / } N$ , and simulations in the square and triangular environments were initialized with an orthogonal $W ^ { \mathrm { r e c } }$ (Saxe et al., 2014). We initialized the bias $b$ and output weights $W ^ { \mathrm { o u t } }$ to be zero. The elements of $W ^ { \mathrm { i n } }$ were zero mean Gaussian variables with variance $1 / N _ { \mathrm { i n } }$ .
+
+![](images/7cde4a241b1b10ed008d3d0dbd8d11d439e41445d8cafe267c621ecd7571dd08.jpg)  
+Figure 2: Different types of spatial selective responses of units in the trained RNN. Example simulation results for three different environments (square, triangular, hexagon) are presented. Blue (yellow) represents low (high) activity. a) Grid-like responses. b) Band-like responses; c) Borderrelated responses; d) Spatially irregular responses. These responses can be spatially selective but they do not form a regular pattern defined in the conventional sense.
+
+# 3 RESULTS
+
+We run simulation experiments in arenas with different boundary shapes, including square, triangular and hexagonal. Figure 1c shows a typical example of the model performance after training; the network (red trace) accurately tracks the animal’s actual path (black).
+
+# 3.1 TUNING PROPERTIES OF THE MODEL NEURONS
+
+We are mostly interested in what kind of representation the RNN has learned to solve this navigation task, and whether such a representation resembles the response properties of neurons in EC (Moser et al., 2008).
+
+# 3.1.1 SPATIAL TUNING
+
+To test whether the trained RNN developed location-selective representations, we plot individual neurons’ mean activity level as a function of the animal’s location during spatial exploration. Note that these average response profiles should not be confused with the linear filters typically shown in feedforward networks. Surprisingly, we find neurons in the trained RNN show a range of interesting spatial response profiles. Examination of these response profiles suggests they can be classified into distinct functional types. Importantly, as we will show, these distinct spatial response profiles can be mapped naturally to known physiology in EC. The spatial responses of all units in trained networks are shown in the Appendix.
+
+Grid-like responses Most interestingly, we find some of the units in the RNN exhibit clear grid-like responses (Figure 2a). These firing patterns typically exhibit multiple firing fields, with each firing field exhibiting roughly circular symmetric or ellipse shape. Furthermore, the firing fields are highly structured, i.e., when combined, are arranged on a regular lattice. Furthermore, the structure of the response lattice depends on the shape of the boundary. In particular, training the network to perform self-localization in a square environment tends to give rectangular grids. In hexagonal and triangular environments, the grids are closer to triangular.
+
+Experimentally, it is shown that (medial) EC contains so-called grid cells which exhibit multiple firing fields that lie on a regular grid (Fyhn et al., 2004; Hafting et al., 2005). The grid-like firing patterns in our simulation are reminiscent of the grid cells in rodents and other mammals. However, we also notice that the the grid-like model responses typically exhibit few periods, not as many as experimental data (see Figure 1a). It is possible that using a larger network might reveal finer grid-patterns in our model. Nonetheless, it is surprising that the gird-like spatial representations can develop in our model, given there is no periodicity in the input. Another potential concern is that, experimentally it is reported that the grids are often on the corners of a triangular lattice (Hafting et al., 2005) even in square environments (see Figure 1a), though the grids are somewhat influenced by the shape of the environment. However, the rats in these experiments presumable had spatial experience in other environments with various boundary shapes. Experimentally, it would be interesting to see if grid cells would lie on a square lattice instead if the rats are raised in a single square environment - a situation we are simulating here.
+
+Border responses Many neurons in the RNN exhibit selectivity to the boundary (Figure 2c). Typically, they only encode a portion of the boundary, e.g. one piece of wall in a square shaped environment. Such properties are similar to the border cells discovered in rodent EC (Solstad et al., 2008; Savelli et al., 2008; Lever et al., 2009). Experimentally, border cells mainly fire along one piece of wall, although some have been observed to fire along multiple borders or along the whole boundary of the environment; interestingly, these multi-border responses were also observed in some RNN models. Currently, it is unclear how the boundary-like response profiles emerge (Solstad et al., 2008; Savelli et al., 2008; Lever et al., 2009). Our model points to the possibility that the border cells may emerge without the presence of tactile cues. Furthermore, it suggests that border cell formation may be related to the movement statistics of the animals, i.e. due to the asymmetry of the movement statistics along the boundary.
+
+Band-like responses Interestingly, some neurons in the RNN exhibit band-like responses (Figure 2b). In most of our simulations, these bands tend to be parallel to one of the boundaries. For some of the units, one of the bands overlaps the boundary, but for others, that is not the case. Experimentally, neurons with periodic-like firing patterns have been recently reported in rodent EC. In one study, it has been reported that a substantial portion of cells in EC exhibit band-like firing characteristics (Krupic et al., 2012). However, we note that based on the reported data in Krupic et al. (2012), the band pattern is not as clear as in our model.
+
+Spatially-stable but non-regular responses Besides the units described above, most of the remaining units also exhibit stable spatial responses, but they do not belong to the above categories. These response profiles can exhibit either one large irregular firing field; or multiple circular firing fields, but these firing fields do not show a regular pattern. Experimentally these types of cells have also been observed. In fact, it is recently reported that the non-grid spatial cells constitute a large portion of the neurons in Layer II and III of rodent EC (Diehl et al., 2017).
+
+# 3.1.2 SPEED TUNING AND HEAD DIRECTION TUNING
+
+Speed tuning We next ask how neurons in the RNN are tuned to the inputs. Many of the model neurons exhibit linear responses to the running speed of the animal, while some neurons show no selectivity to speed, as suggested by the near-flat response functions. Example response profiles are shown in Figure 3. Interestingly, we observe that the model border cells tend to have almost zero speed-tuning (e.g., see Figure 3g,h).
+
+![](images/318a67b1eb08bdf270e30eebad468380d0e5a80e744e9f80d58a5d693f3a94ed.jpg)  
+Figure 3: Direction tuning and speed tuning for nine example units in an RNN trained in a triangular arena. For each unit, we show the spatial tuning, (head) directional tuning, speed tuning respectively, from left to right. a,b,c) The three model neurons show strong directional tuning, but the spatial tuning is weak and irregular. The three neurons also exhibit linear speed tuning. d,e,f) The three neurons exhibit grid-like firing patterns, and clear speed tuning. The strength of their direction tuning differ. ${ \bf g } , { \bf h } )$ Border cells exhibit weak and a bit complex directional tuning and almost no speed tuning. i) This band cell shows weak directional tuning, but strong speed tuning.
+
+Head direction tuning A substantial portion of the model neurons show direction tuning. There are a diversity of direction tuning profiles, both in terms of the strength of the tuning and their preferred direction. Example tuning curves are shown in Figure 3, and the direction tuning curves of a complete population are shown in the Appendix. Interestingly, in general model neurons which show the strongest head direction tuning do not show a clear spatial firing pattern (see Figure 3a,b,c). This suggests that there are a group of neurons which are mostly responsible for encoding the direction. We also notice that neurons with clear grid-like firing can exhibit a variety of direction tuning strengths, from weak to strong (Figure 3d,e,f). In the Appendix, we quantify the relation between these different tuning properties at the whole population level, which show somewhat complex dependence.
+
+Experimentally, the heading direction tuning in EC is well-known, e.g., Sargolini et al. (2006). Both the grid and non-grid cells in EC exhibit head direction tuning (Sargolini et al., 2006). Furthermore, the linear speed dependence of the model neurons is similar to the properties of speed cells reported recently in EC (Kropff et al., 2015). Our result is also consistent with another recent study reporting that the majority of neurons in EC exhibit some amount of speed tuning (Hinman et al., 2016). It remains an open question experimentally, at a population level, how different types of tuning characteristics in EC relate to each other.
+
+# 3.1.3 DEVELOPMENT OF THE TUNING PROPERTIES
+
+We next investigate how the spatial response profiles evolve as learning/training progresses. We report two main observations. First, neurons that fire selectively along the boundary typically emerge first. Second, the grid-like responses with finer spatial tuning patterns only emerge later in training. For visualization, we perform dimensionality reduction using the t-SNE algorithm (Maaten & Hinton, 2008). This algorithm embeds 100 model neurons during three phases of training (early, intermediate, and late) into a two-dimensional space according to the similarity of their temporal responses. Here the similarity metric is taken to be firing rate correlation. In this 2D space as shown in Figure 4a, border cell representations appear early and stably persist through the end of training. Furthermore, early during training all responses are similar to the border related responses. In contrast, grid-like cells typically undergo a substantial change in firing pattern during training before settling into their final grid-like representation (Figure 4b).
+
+The developmental time line of the grid-like cells and border cells is roughly consistent with developmental studies in rodents. Experimentally, it is known that border cells emerge earlier in development, and they exist at about 2 weeks after the rat is born (Bjerknes et al., 2014). The grid cells mature only at about 4 weeks after birth (Langston et al., 2010; Wills et al., 2010; Bjerknes et al., 2014). Furthermore, our simulations suggest the reason why border cells emerge earlier in development may be that computationally it is easier to wire-up a network that gives rise to border cell responses.
+
+![](images/9c19816ea23656a00f78df190e6e407bf4c84fe0856a9d69293f60f5972dfa0a.jpg)  
+Figure 4: Development of border cells and grid-like cells. Early during training all responses are similar to the border related responses, and only as training continues do the grid-like cells emerge. We perform dimensionality reduction using the t-SNE algorithm on the firing rates of the neurons. Each dot represents one neuron $N = 1 0 0$ ), and the color represents different training stages (early/intermediate/late shown in blue/cyan/yellow). Each line shows the trajectory of a single highlighted neuron as its firing responses evolve during training. In panel a), we highlight the border representation. It appears there are four clusters of border cells, each responding to one wall of a square environment (spatial responses from four of these border cells are inset). These cells’ response profiles appear early and stably persist through training, illustrated by the short distance they travel in this space. In b), we show that the neurons which eventually become grid cells initially have tuning profiles similar to the border cells but then change their tuning substantially during learning. As a natural consequence, they need to travel a long distance in this space between the early and late phase of the training. Spatial responses are shown for four of these grid-like cells during the late phase of training.
+
+# 3.2 THE IMPORTANCE OF REGULARIZATION
+
+We find appropriate regularizations of the RNN to be crucial for the emergence of grid-like representations. We only observed grid-like representations when the network was encouraged to store information while perturbed by noise. This was accomplished by setting the speed input to zero, e.g. zero speed $90 \%$ of the time, and adding Gaussian noise to the network $\xi _ { i } ( t )$ in equation (1)); the precise method for setting the speed input to zero and the value of the noise variance is not crucial for our simulations to develop grid-like representations. The cost function which aims to capture the penalization on the metabolic cost of the neural activity also acts as an important regularization. Our simulations show that the grid-like representation did not emerge without this metabolic cost. In Figure 5, we show typical simulation results for a square environment, with and without proper metabolic regularization. In the Appendix, we illustrate the effect of regularization further, in particular the role of injecting noise into the RNN units.
+
+Our results are consistent with the general notion on the importance of incorporating proper constraint for learning useful representations in neural networks (Bengio et al., 2013). Furthermore, it suggests that, to learn a model with response properties similar to neural systems it may be necessary to incorporate the relevant constraints, e.g., noise and metabolic cost.
+
+# 3.3 ERROR CORRECTION AROUND THE BOUNDARY
+
+One natural question is whether the trained RNNs are able to perform localization when the path length exceeds the typical length used during training (500 steps), in particular given that noise in the network would gradually accumulate, leading to a decrease in localization performance. We test this by simulating paths that are several orders of magnitude longer. Somewhat surprisingly, we find the RNNs still perform well (Figure 6b). In fact, the squared error (averaged over every 10000 steps) is stable. The spatial response profiles of individual units also remain stable. This implies that the RNNs have acquired intrinsic error-correction mechanisms during training.
+
+![](images/476435913a7b64388291eece485ed2e49a2bc8befbf4d07d7b0df7b3c97e20d4.jpg)  
+Figure 5: Complete set of spatial response profiles for 100 neurons in a RNN trained in a square environment. a) Without proper regularization, complex and periodic spatial response patterns do not emerge. b) With proper regularization, a rich set of periodic response patterns emerge, including grid-like responses. Regularization can also be adjusted to achieve spatial profiles intermediate between these two examples.
+
+![](images/1313022c4e140eff6c754917cddc7b20c35899670d67ff7b351c56f3f49df4d8.jpg)  
+Figure 6: Error-correction happens at the boundary and the error is stable over time. At the boundary, the direction is resampled to avoid input velocities that lead to a path extending beyond the boundary of the environment. These changing input statistics at the boundary, termed a boundary interaction, are the only cue the RNN receives about the boundary. We find that the RNN uses the boundary interactions to correct the accumulated error between the true integrated input and its prediction based on the linear readout of equation (2). Panel a), the mean squared error increases when there are no boundary interactions, but then decreases after a boundary interaction, with more boundary interactions leading to greater error reduction. In the absence of further boundary interaction, the squared error would gradually increase again (blue curve) at roughly a constant rate. b) The network was trained using mini-batches of 500 timesteps but has stable error over a duration at least four orders of magnitude larger. The error of the RNN output (mean and standard deviation shown in black, computed based on 10000 timesteps) is compared to the error that would be achieved by an RNN outputting the best constant values (red).
+
+As shown earlier, during training some of the RNN units develop boundary-related firing (Figure 2c), presumably by exploiting the change of input statistics around the boundary. We hypothesize that boundary interactions may enable error-correction through signals based on these boundary-related activities. Indeed, we find that boundary interactions can dramatically reduce the accumulated error (Figure 6a). Figure 6a shows that, without boundary interactions, on average the squared error grows roughly linearly as expected, however, interactions with the boundaries substantially reduce the error, and more frequent boundary interactions can reduce the error further. Error-correction on grid cells via boundary interactions has been proposed (Hardcastle et al., 2015; Pollock et al., 2017), however, we emphasize that the model proposed here develops the grid-like responses, boundary responses and the error-correction mechanisms all within the same neural network, thus potentially providing a unifying account of a diverse set of phenomena.
+
+# 4 DISCUSSION
+
+In this paper, we trained RNNs to perform path integration (dead-reckoning) in 2D arenas. We found that after training RNNs with appropriate regularization, the model neurons exhibit a variety of spatial and velocity tuning profiles that match neurophysiology in EC. What’s more, there is also similarity in terms of when these distinct neuron types emerge during training/development. The EC has long been thought to be involved in path integration and localization of the animal’s location (Moser et al., 2008). The general agreement between the different response properties in our model and the neurophysiology provide strong evidence supporting the hypothesis that the neural population in EC may provide an efficient code for representation self-locations based on the velocity input.
+
+Recently, there has been increased interest in using complex neural network models to understand the neural code. But the focus has been on using feedforward architectures, in particular CNNs (LeCun et al., 1998). Given the abundant recurrent connections in the brain, it seems a particularly fruitful avenue to take advantage of the recent development in RNNs to help with neuroscience questions (Mante et al., 2013; Song et al., 2016; Miconi, 2017; Sussillo et al., 2015). Here, we only show one instance following this approach. However, the insight from this work could be general, and potentially useful for other cognitive functions as well.
+
+The finding that metabolic constraints lead to the emergence of grid-like responses may be seen as conceptually related to the efficient coding hypothesis in visual processing (Barlow, 1961), in particular the seminal work on the emergence of the V1-like Gabor filters in a sparse coding model by Olshausen & Field (1996). Indeed, our work is partly inspired by these results. While there are conceptual similarities, however, we should also note there are differences between the sparse coding work and ours. First, the sparsity constraint in sparse coding can be naturally viewed as a particular prior while in the context of the recurrent network, it is difficult to interpret that way. Second, the grid-like responses are not the most sparse solution one could imagine. In fact, they are still quite dense compared to a more spatially localized representation. Third, the grid-like patterns that emerged in our network are not filters based on the raw input, rather the velocity inputs need to be integrated first in order to encode spatial locations. Our work is also inspired by recent work using the efficient coding idea to explain the functional architecture of the grid cells (Wei et al., 2015). It has been shown that efficient coding considerations could explain the particular set of grid scales observed in rodents (Stensola et al., 2012). However, in that work, the firing patterns of the neurons are assumed to have a lattice structure to start with. Furthermore, our work is related to the study by Sussillo and others (Sussillo et al., 2015), in which they show that regularization of RNN models are important for generating solutions that are similar to the neural activity observed in motor cortex. In Sussillo et al., a smoothness constraint together with others lead to simple oscillatory neural dynamics that well matches the neural data. We have not incorporated a smoothness constraint into our network.
+
+Additionally, we note that there are a few recent studies which use place cells as the input to generate grid cells (Dordek et al., 2016; Stachenfeld et al., 2016), which are fundamentally different from our work. In these feedforward network models, the grid cells essentially perform dimensionality reduction based on the spatial input from place cells. However, the main issue with these models is that, it is unclear how place cells acquire spatial tuning in the first place. To the contrary, our model takes the animal’s velocity as the input, and addresses the question of how the spatial tuning can be generated from such input, which are known to exist in EC (Sargolini et al., 2006; Kropff et al., 2015). In another related study (Kanitscheider & Fiete, 2016), the authors train a RNN with
+
+LSTM units (Hochreiter & Schmidhuber, 1997) to perform different navigation tasks. However, no grid-like spatial firing patterns are reported.
+
+Although our model shows a qualitative match to the neural responses observed in the EC, nonetheless it has several major limitations, with each offering interesting future research directions. First, the learning rule we use seems to be biologically implausible. We are interested in exploring how a more biologically plausible learning rule could give rise to similar results (Lillicrap et al., 2016; Miconi, 2017; Guerguiev et al., 2017). Second, the simulation results do not show a variety of spatial scales in grid-like cells. Experimentally, it is known that grid cells have multiple spatial scales, that scale geometrically with a ratio 1.4 (Stensola et al., 2012), and this particular scale ratio is predicted by efficient coding of space (Wei et al., 2015). We are investigating how to modify the model to get a hierarchy of spatial scales, perhaps by incorporating more neurons or modifying the regularization. Last but not least, we have focused on the representation produced by the trained RNN. An equally important set of questions concern how the networks actually support the generation of such a representation. As a preliminary effort, we have examined the connectivity patterns of the trained network, and they do not seem to resemble the connectivity patterns required by standard attractor network models. Maybe this should not be seen as too surprising. After all, the trained networks can produce a diverse set of neural responses, while the previous models only led to grid responses. It would be interesting for future work to systematically examine the questions related to the underlying mechanisms.
+
+# 5 ACKNOWLEDGEMENTS
+
+We thank members of the Center for Theoretical Neuroscience at Columbia University for useful discussions and three anonymous reviewers for constructive feedback. Research supported by NSF NeuroNex Award DBI-1707398 and NIH training grant 5T32NS064929 (CJC).
+
+# REFERENCES
+
+Dmitriy Aronov, Rhino Nevers, and David W Tank. Mapping of a non-spatial dimension by the hippocampalentorhinal circuit. Nature, 2017.
+
+Horace B Barlow. Possible principles underlying the transformation of sensory messages. Sensory communication, pp. 217–234, 1961.
+
+Yoshua Bengio, Aaron Courville, and Pascal Vincent. Representation learning: A review and new perspectives. IEEE transactions on pattern analysis and machine intelligence, 35(8):1798–1828, 2013.
+
+Tale L Bjerknes, Edvard I Moser, and May-Britt Moser. Representation of geometric borders in the developing rat. Neuron, 82(1):71–78, 2014.
+
+Yoram Burak and Ila R Fiete. Accurate path integration in continuous attractor network models of grid cells. PLoS computational biology, 5(2):e1000291, 2009.
+
+Jonathan J Couey, Aree Witoelar, Sheng-Jia Zhang, Kang Zheng, Jing Ye, Benjamin Dunn, Rafal Czajkowski, May-Britt Moser, Edvard I Moser, Yasser Roudi, et al. Recurrent inhibitory circuitry as a mechanism for grid formation. Nature neuroscience, 16(3):318–324, 2013.
+
+Charles Darwin. Origin of certain instincts. Nature, 7:417–418, 1873.
+
+Geoffrey W Diehl, Olivia J Hon, Stefan Leutgeb, and Jill K Leutgeb. Grid and nongrid cells in medial entorhinal cortex represent spatial location and environmental features with complementary coding schemes. Neuron, 94(1):83–92, 2017.
+
+Yedidyah Dordek, Daniel Soudry, Ron Meir, and Dori Derdikman. Extracting grid cell characteristics from place cell inputs using non-negative principal component analysis. eLife, 5:e10094, 2016.
+
+Ariane S Etienne and Kathryn J Jeffery. Path integration in mammals. Hippocampus, 14(2):180– 192, 2004.
+
+Marianne Fyhn, Sturla Molden, Menno P Witter, Edvard I Moser, and May-Britt Moser. Spatial representation in the entorhinal cortex. Science, 305(5688):1258–1264, 2004.
+
+Marianne Fyhn, Torkel Hafting, Menno P Witter, Edvard I Moser, and May-Britt Moser. Grid cells in mice. Hippocampus, 18(12):1230–1238, 2008.
+
+Alex Graves, Abdel-Rahman Mohamed, and Geoffrey Hinton. Speech recognition with deep recurrent neural networks. In Acoustics, speech and signal processing (icassp), 2013 ieee international conference on, pp. 6645–6649. IEEE, 2013.
+
+Karol Gregor, Ivo Danihelka, Alex Graves, Danilo Jimenez Rezende, and Daan Wierstra. Draw: A recurrent neural network for image generation. arXiv preprint arXiv:1502.04623, 2015.
+
+Jordan Guerguiev, Timothy P Lillicrap, and Blake A Richards. Towards deep learning with segregated dendrites. eLife, 6, 2017.
+
+Torkel Hafting, Marianne Fyhn, Sturla Molden, May-Britt Moser, and Edvard I Moser. Microstructure of a spatial map in the entorhinal cortex. Nature, 436(7052):801–806, 2005.
+
+Kiah Hardcastle, Surya Ganguli, and Lisa M Giocomo. Environmental boundaries as an error correction mechanism for grid cells. Neuron, 86(3):827–839, 2015.
+
+James R Hinman, Mark P Brandon, Jason R Climer, G William Chapman, and Michael E Hasselmo. Multiple running speed signals in medial entorhinal cortex. Neuron, 91(3):666–679, 2016.
+
+Sepp Hochreiter and Jurgen Schmidhuber. Long short-term memory. ¨ Neural computation, 9(8): 1735–1780, 1997.
+
+David H Hubel and Torsten N Wiesel. Receptive fields, binocular interaction and functional architecture in the cat’s visual cortex. The Journal of physiology, 160(1):106–154, 1962.
+
+Joshua Jacobs, Christoph T Weidemann, Jonathan F Miller, Alec Solway, John F Burke, Xue-Xin Wei, Nanthia Suthana, Michael R Sperling, Ashwini D Sharan, Itzhak Fried, et al. Direct recordings of grid-like neuronal activity in human spatial navigation. Nature neuroscience, 16(9):1188– 1190, 2013.
+
+Ingmar Kanitscheider and Ila Fiete. Training recurrent networks to generate hypotheses about how the brain solves hard navigation problems. arXiv preprint arXiv:1609.09059, 2016.
+
+Tim Christian Kietzmann, Patrick McClure, and Nikolaus Kriegeskorte. Deep neural networks in computational neuroscience. bioRxiv, pp. 133504, 2017.
+
+Nathaniel J Killian, Michael J Jutras, and Elizabeth A Buffalo. A map of visual space in the primate entorhinal cortex. Nature, 491(7426):761–764, 2012.
+
+Nikolaus Kriegeskorte. Deep neural networks: a new framework for modeling biological vision and brain information processing. Annual Review of Vision Science, 1:417–446, 2015.
+
+Alex Krizhevsky, Ilya Sutskever, and Geoffrey E Hinton. Imagenet classification with deep convolutional neural networks. In Advances in neural information processing systems, pp. 1097–1105, 2012.
+
+Emilio Kropff, James E Carmichael, May-Britt Moser, and Edvard I Moser. Speed cells in the medial entorhinal cortex. Nature, 523(7561):419–424, 2015.
+
+Julija Krupic, Neil Burgess, and John O?Keefe. Neural representations of location composed of spatially periodic bands. Science, 337(6096):853–857, 2012.
+
+Rosamund F Langston, James A Ainge, Jonathan J Couey, Cathrin B Canto, Tale L Bjerknes, Menno P Witter, Edvard I Moser, and May-Britt Moser. Development of the spatial representation system in the rat. Science, 328(5985):1576–1580, 2010.
+
+Yann LeCun, Leon Bottou, Yoshua Bengio, and Patrick Haffner. Gradient-based learning applied to ´ document recognition. Proceedings of the IEEE, 86(11):2278–2324, 1998.
+
+Yann LeCun, Yoshua Bengio, and Geoffrey Hinton. Deep learning. Nature, 521(7553):436–444, 2015.
+
+Colin Lever, Stephen Burton, Ali Jeewajee, John O’Keefe, and Neil Burgess. Boundary vector cells in the subiculum of the hippocampal formation. The journal of neuroscience, 29(31):9771–9777, 2009.
+
+Timothy P Lillicrap, Daniel Cownden, Douglas B Tweed, and Colin J Akerman. Random synaptic feedback weights support error backpropagation for deep learning. Nature communications, 7, 2016.
+
+Laurens van der Maaten and Geoffrey Hinton. Visualizing data using t-sne. Journal of Machine Learning Research, 9(Nov):2579–2605, 2008.
+
+Valerio Mante, David Sussillo, Krishna V Shenoy, and William T Newsome. Context-dependent computation by recurrent dynamics in prefrontal cortex. Nature, 503(7474):78–84, 2013.
+
+James Martens and Ilya Sutskever. Learning recurrent neural networks with hessian-free optimization. pp. 10331040, 2011.
+
+Bruce L McNaughton, Francesco P Battaglia, Ole Jensen, Edvard I Moser, and May-Britt Moser. Path integration and the neural basis of the ’cognitive map’. Nature Reviews Neuroscience, 7(8): 663–678, 2006.
+
+Thomas Miconi. Biologically plausible learning in recurrent neural networks reproduces neural dynamics observed during cognitive tasks. eLife, 6:e20899, 2017.
+
+M-L Mittelstaedt and H Mittelstaedt. Homing by path integration in a mammal. Naturwissenschaften, 67(11):566–567, 1980.
+
+Volodymyr Mnih, Koray Kavukcuoglu, David Silver, Andrei A Rusu, Joel Veness, Marc G Bellemare, Alex Graves, Martin Riedmiller, Andreas K Fidjeland, Georg Ostrovski, et al. Human-level control through deep reinforcement learning. Nature, 518(7540):529–533, 2015.
+
+Edvard I Moser, Emilio Kropff, and May-Britt Moser. Place cells, grid cells, and the brain’s spatial representation system. Annu. Rev. Neurosci., 31:69–89, 2008.
+
+John O’Keefe. Place units in the hippocampus of the freely moving rat. Experimental neurology, 51 (1):78–109, 1976.
+
+Bruno A Olshausen and David J Field. Emergence of simple-cell receptive field properties by learning a sparse code for natural images. Nature, 381(6583):607, 1996.
+
+Aaron van den Oord, Nal Kalchbrenner, and Koray Kavukcuoglu. Pixel recurrent neural networks. arXiv preprint arXiv:1601.06759, 2016.
+
+Eli Pollock, Niral Desai, Xue-Xin Wei, and Vijay B Balasubramanian. A mechanism for selforganized error-correction of grid cells by border cells. CoSyNe abstract, 2017.
+
+Alexei Samsonovich and Bruce L McNaughton. Path integration and cognitive mapping in a continuous attractor neural network model. Journal of Neuroscience, 17(15):5900–5920, 1997.
+
+Francesca Sargolini, Marianne Fyhn, Torkel Hafting, Bruce L McNaughton, Menno P Witter, MayBritt Moser, and Edvard I Moser. Conjunctive representation of position, direction, and velocity in entorhinal cortex. Science, 312(5774):758–762, 2006.
+
+Francesco Savelli, D Yoganarasimha, and James J Knierim. Influence of boundary removal on the spatial representations of the medial entorhinal cortex. Hippocampus, 18(12):1270, 2008.
+
+Andrew M Saxe, James L McClelland, and Surya Ganguli. Exact solutions to the nonlinear dynamics of learning in deep linear neural networks. 2014.
+
+Trygve Solstad, Charlotte N Boccara, Emilio Kropff, May-Britt Moser, and Edvard I Moser. Representation of geometric borders in the entorhinal cortex. Science, 322(5909):1865–1868, 2008.
+
+H Francis Song, Guangyu R Yang, and Xiao-Jing Wang. Training excitatory-inhibitory recurrent neural networks for cognitive tasks: A simple and flexible framework. PLoS Comput Biol, 12(2): e1004792, 2016.
+
+Kimberly Lauren Stachenfeld, Matthew M Botvinick, and Samuel J Gershman. The hippocampus as a predictive map. bioRxiv, pp. 097170, 2016.
+
+Hanne Stensola, Tor Stensola, Trygve Solstad, Kristian Frøland, May-Britt Moser, and Edvard I Moser. The entorhinal grid map is discretized. Nature, 492(7427):72–78, 2012.
+
+David Sussillo, Mark M Churchland, Matthew T Kaufman, and Krishna V Shenoy. A neural network that finds a naturalistic solution for the production of muscle activity. Nature neuroscience, 18(7): 1025–1033, 2015.
+
+Lucas Theis and Matthias Bethge. Generative image modeling using spatial lstms. In Advances in Neural Information Processing Systems, pp. 1927–1935, 2015.
+
+Xue-Xin Wei, Jason Prentice, and Vijay Balasubramanian. A principle of economy predicts the functional architecture of grid cells. Elife, 4:e08362, 2015.
+
+B Willmore and DJ Tolhurst. Characterizing the sparseness of neural codes. Network, 12:255–270, 2001.
+
+Tom J Wills, Francesca Cacucci, Neil Burgess, and John O’keefe. Development of the hippocampal cognitive map in preweanling rats. Science, 328(5985):1573–1576, 2010.
+
+Shawn S. Winter, Benjamin J. Clark, and Jeffrey S. Taube. Disruption of the head direction cell network impairs the parahippocampal grid cell signal. Science, 347(6224):870–874, 2015a.
+
+Shawn S. Winter, Max L. Mehlman, Benjamin J. Clark, and Jeffrey S. Taube. Passive transport disrupts grid signals in the parahippocampal cortex. Current Biology, 25:2493–2502, 2015b.
+
+Daniel LK Yamins and James J DiCarlo. Using goal-driven deep learning models to understand sensory cortex. Nature neuroscience, 19(3):356–365, 2016.
+
+Daniel LK Yamins, Ha Hong, Charles F Cadieu, Ethan A Solomon, Darren Seibert, and James J DiCarlo. Performance-optimized hierarchical models predict neural responses in higher visual cortex. Proceedings of the National Academy of Sciences, 111(23):8619–8624, 2014.
+
+Michael M Yartsev, Menno P Witter, and Nachum Ulanovsky. Grid cells without theta oscillations in the entorhinal cortex of bats. Nature, 479(7371):103–107, 2011.
+
+# A TRIANGULAR ENVIRONMENT
+
+![](images/9fec0b40f547311967f788a5e33c37ca83be210239a76a72034b231f75f6b016.jpg)
+
+![](images/76d7068d3c86954df82911a198a111d047595eeb46564b8603a26da4acf15c38.jpg)
+
+Noise and metabolic cost are important for grid-like representations. The figure on the left shows the spatial responses for a network trained with noise and no metabolic cost. The figure on the right shows the spatial responses for a network trained with no noise and the metabolic cost.
+
+![](images/ace176f80461890db9b194a0b86e5c7936f97026efec6348b0297c43783b3061.jpg)
+
+# B RECTANGULAR ENVIRONMENT
+
+![](images/fad2d973408af95191597a5e409766654c8b3a9862f0ad4f2a8de04f639ad5ee.jpg)![](images/97fb3c206da8fbace0f4d8eeba7efe4308290448d5d9e2cba584838472b30abd.jpg)
+
+Speed	input
+
+# C HEXAGONAL ENVIRONMENT
+
+![](images/d7c955da2b4d05471ca42b23f6a780ea605e2aed77c7b520059872b81c33fead.jpg)
+
+![](images/64fe9f96b9155e2bd1b22d7a344f3085ff647a473ca40df4a123eb8758cf0801.jpg)
+
+Angular	input	(0	to	360	degrees)
+
+Speed	tuning
+
+![](images/7eb735c5ab485558870c8b0720dd627699e69018e8cbacfa58c8d343cfc8a2d1.jpg)  
+Speed	input
+
+To quantify the speed selectivity of each unit we first fit a line to the tuning curve of unit activity as a function of speed. The speed selectivity is the absolute value of the slope. If the unit activity is not modulated by speed then the speed selectivity is 0. To quantify the direction selectivity of each unit we calculated the average unit activity as a function of direction input and then took the maximum minus minimum of this tuning curve. If the unit activity is not modulated by direction then the direction selectivity is 0. To quantify the spatial selectivity we used lifetime sparseness (Willmore & Tolhurst, 2001). If the unit activity is not modulated by spatial location then the spatial selectivity is 0. Each dot in the figures below show the selectivity for a single unit.
+
+![](images/7d81c0e39724fab8f48b0863c121f5336b8e91d07c652f62dff58ab0d2f8945e.jpg)
+
+# E ADDITIONAL TRAINING DETAILS
+
+During training we tried to balance all three terms we were minimizing $( E , R _ { L 2 }$ , and $R _ { F R }$ ) so no single term was neglected or dominated. At the beginning of training we weighted the regularization term $R _ { L 2 }$ to be equal to the error function $E$ and then decreased the weighting on $R _ { L 2 }$ according to the schedule used by Martens & Sutskever (2011). We adaptively adjusted the weighting on $R _ { F R }$ , starting from an initial value of $E / 1 0$ and enforcing an upper bound of $E / 3$ as training progressed. We found this training procedure improved training performance and led to more interesting representations.
\ No newline at end of file
diff --git a/parse/train/B17JTOe0-/B17JTOe0-_content_list.json b/parse/train/B17JTOe0-/B17JTOe0-_content_list.json
new file mode 100644
index 0000000000000000000000000000000000000000..0ae8a8627dca5c5b7fe4e2fcb7f6983b16892ca9
--- /dev/null
+++ b/parse/train/B17JTOe0-/B17JTOe0-_content_list.json
@@ -0,0 +1,1760 @@
+[
+    {
+        "type": "text",
+        "text": "EMERGENCE OF GRID-LIKE REPRESENTATIONS BYTRAINING RECURRENT NEURAL NETWORKS TOPERFORM SPATIAL LOCALIZATION",
+        "text_level": 1,
+        "bbox": [
+            174,
+            99,
+            821,
+            171
+        ],
+        "page_idx": 0
+    },
+    {
+        "type": "text",
+        "text": "Christopher J. Cueva∗, Xue-Xin Wei∗ Columbia University New York, NY 10027, USA {ccueva,weixxpku}@gmail.com ",
+        "bbox": [
+            183,
+            194,
+            446,
+            251
+        ],
+        "page_idx": 0
+    },
+    {
+        "type": "text",
+        "text": "ABSTRACT ",
+        "text_level": 1,
+        "bbox": [
+            454,
+            287,
+            544,
+            303
+        ],
+        "page_idx": 0
+    },
+    {
+        "type": "text",
+        "text": "Decades of research on the neural code underlying spatial navigation have revealed a diverse set of neural response properties. The Entorhinal Cortex (EC) of the mammalian brain contains a rich set of spatial correlates, including grid cells which encode space using tessellating patterns. However, the mechanisms and functional significance of these spatial representations remain largely mysterious. As a new way to understand these neural representations, we trained recurrent neural networks (RNNs) to perform navigation tasks in 2D arenas based on velocity inputs. Surprisingly, we find that grid-like spatial response patterns emerge in trained networks, along with units that exhibit other spatial correlates, including border cells and band-like cells. All these different functional types of neurons have been observed experimentally. The order of the emergence of grid-like and border cells is also consistent with observations from developmental studies. Together, our results suggest that grid cells, border cells and others as observed in EC may be a natural solution for representing space efficiently given the predominant recurrent connections in the neural circuits. ",
+        "bbox": [
+            233,
+            321,
+            766,
+            529
+        ],
+        "page_idx": 0
+    },
+    {
+        "type": "text",
+        "text": "1 INTRODUCTION ",
+        "text_level": 1,
+        "bbox": [
+            176,
+            563,
+            336,
+            579
+        ],
+        "page_idx": 0
+    },
+    {
+        "type": "text",
+        "text": "Understanding the neural code in the brain has long been driven by studying feed-forward architectures, starting from Hubel and Wiesel’s famous proposal on the origin of orientation selectivity in primary visual cortex (Hubel & Wiesel, 1962). Inspired by the recent development in deep learning (Krizhevsky et al., 2012; LeCun et al., 2015; Hochreiter & Schmidhuber, 1997; Mnih et al., 2015), there has been a burst of interest in applying deep feedforward models, in particular convolutional neural networks (CNN) (LeCun et al., 1998), to study the sensory systems, which hierarchically extract useful features from sensory inputs (see e.g., Yamins et al. (2014); Kriegeskorte (2015); Kietzmann et al. (2017); Yamins & DiCarlo (2016)). ",
+        "bbox": [
+            174,
+            595,
+            825,
+            708
+        ],
+        "page_idx": 0
+    },
+    {
+        "type": "text",
+        "text": "For more cognitive tasks, neural systems often need to maintain certain internal representations of relevant variables in the absence of external stimuli- a process that requires more than feature extraction. We will focus on spatial navigation, which typically requires the brain to maintain a representation of self-location and update it according to the animal’s movements and landmarks of the environment. Physiological studies done in rodents and other mammals (including humans, non-human primates and bats) have revealed a variety of neural correlates of space in Hippocampus and Entorhinal Cortex (EC), including place cells (O’Keefe, 1976), grid cells (Fyhn et al., 2004; Hafting et al., 2005; Fyhn et al., 2008; Yartsev et al., 2011; Killian et al., 2012; Jacobs et al., 2013), along with border cells (Solstad et al., 2008), band-like cells (Krupic et al., 2012) and others (see Figure 1a). In particular, each grid cell only fires when the animal occupies a distinct set of physical locations, and strikingly these locations lie on a lattice. The study of the neural underpinning of spatial cognition has provided an important window into how high-level cognitive functions are supported in the brain (Moser et al., 2008; Aronov et al., 2017). ",
+        "bbox": [
+            174,
+            715,
+            825,
+            893
+        ],
+        "page_idx": 0
+    },
+    {
+        "type": "text",
+        "text": "How might the spatial navigation task be solved using a network of neurons? Recurrent neural networks (RNNs) (Hochreiter & Schmidhuber, 1997; Graves et al., 2013; Oord et al., 2016; Theis & Bethge, 2015; Gregor et al., 2015; Sussillo et al., 2015) seem particularly useful for these tasks. Indeed, recurrent-based continuous attractor networks have been one popular type of models proposed for the formation of grid cells (McNaughton et al., 2006; Burak & Fiete, 2009; Couey et al., 2013) and place cells (Samsonovich & McNaughton, 1997). Such models have provided valuable insights into one set of possible mechanisms that could support the formation of the grids. However, these models typically rely on fine-tuned connectivity patterns, in particular the models need a subtle yet systematic asymmetry in the connectivity pattern to move the attractor state according to the animal’s own movement. The existence of such a specific 2D connectivity in rodent EC remains unclear. Additionally, previous models have mainly focused on grid cells, while other types of responses that co-exist in the Entorhinal Cortex have been largely ignored. It would be useful to have a unified model that can simultaneously explain different types of neural responses in EC. ",
+        "bbox": [
+            174,
+            103,
+            825,
+            284
+        ],
+        "page_idx": 1
+    },
+    {
+        "type": "text",
+        "text": "Motivated by these considerations, here we present an alternative modeling approach for understanding the representation of space in the neural system. Specifically, we trained a RNN to perform some spatial navigation tasks. By leveraging the recent development in RNN training and knowledge of the navigation system in the brain, we show that training a RNN with biologically relevant constraints naturally gives rise to a variety of spatial response profiles as observed in EC, including grid-like responses. To our knowledge, this is the first study to show that grid-like responses could emerge from training a RNN to perform navigation. Our result implies that the neural representation in EC may be seen as a natural way for the brain to solve the navigation task efficiently (Wei et al., 2015). More generally, it suggests that RNNs can be a powerful tool for understanding the neural mechanisms of certain high-level cognitive functions. ",
+        "bbox": [
+            173,
+            291,
+            825,
+            430
+        ],
+        "page_idx": 1
+    },
+    {
+        "type": "image",
+        "img_path": "images/8e6f9cf7feba198b2a18d2ed284d372bf5324606db8951624994c943f119c937.jpg",
+        "image_caption": [
+            "Figure 1: a) Example neural data showing different kinds of neural correlates underlying spatial navigation in EC. All figures are replotted from previous publications. From left to right: a “grid cell” recorded when an animal navigates in a square environment, replotted from Krupic et al. (2012), with the heat map representing the firing rate of this neuron as a function of the animal’s location (red corresponds to high firing rate); a “band-like” cell from Krupic et al. (2012); a border cell from Solstad et al. (2008); an irregular spatially tuned cell from Diehl et al. (2017); a “speed cell” from Kropff et al. (2015), which exhibits roughly linear dependence on the rodent’s running speed; a “heading direction cell” from Sargolini et al. (2006), which shows systematic change of firing rate depending on animal’s heading direction. b) The network consists of $N = 1 0 0$ recurrently connected units (or neurons) which receive two external inputs, representing the animal’s speed and heading direction. The two outputs linearly weight the neurons in the RNN. The goal of training is to make the responses of the two output neurons accurately represent the animal’s physical location. c) Typical trajectory after training. As shown, the output of the RNN can accurately, though not perfectly, track the animal’s location during navigation. "
+        ],
+        "image_footnote": [],
+        "bbox": [
+            238,
+            470,
+            751,
+            681
+        ],
+        "page_idx": 1
+    },
+    {
+        "type": "text",
+        "text": "2 MODEL ",
+        "text_level": 1,
+        "bbox": [
+            174,
+            102,
+            269,
+            118
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "text",
+        "text": "2.1 MODEL DESCRIPTION ",
+        "text_level": 1,
+        "bbox": [
+            176,
+            132,
+            366,
+            147
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "text",
+        "text": "Our network model consists of a set of recurrently connected units $N = 1 0 0$ ). The dynamics of each unit in the network $u _ { i } ( t )$ is governed by the standard continuous-time RNN equation: ",
+        "bbox": [
+            173,
+            159,
+            826,
+            188
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "equation",
+        "img_path": "images/73f5fb1e1639b7bb2d07858a8bd6b7b8e2b6dc0350eb54f951fc46c74a662eee.jpg",
+        "text": "$$\n\\tau \\frac { d x _ { i } ( t ) } { d t } = - x _ { i } ( t ) + \\sum _ { j = 1 } ^ { N } W _ { i j } ^ { \\mathrm { r e c } } u _ { j } ( t ) + \\sum _ { k = 1 } ^ { N _ { \\mathrm { i n } } } W _ { i k } ^ { \\mathrm { i n } } I _ { k } ( t ) + b _ { i } + \\xi _ { i } ( t )\n$$",
+        "text_format": "latex",
+        "bbox": [
+            281,
+            190,
+            717,
+            236
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "text",
+        "text": "for $i = 1 , \\ldots , N$ . The activity of each unit, $u _ { i } ( t )$ , is related to the activation of that unit, $x _ { i } ( t )$ , through a nonlinearity which in this study we take to be $u _ { i } ( t ) = \\operatorname { t a n h } ( x _ { i } ( t ) )$ . Each unit receives input from other units through the recurrent weight matrix $W ^ { \\mathrm { r e c } }$ and also receives external input, $I ( t )$ , that enters the network through the weight matrix $W ^ { \\mathrm { i n } }$ . Each unit has two sources of bias, $b _ { i }$ which is learned and $\\xi _ { i } ( t )$ which represents noise intrinsic to the network and is taken to be Gaussian with zero mean and constant variance. The network was simulated using the Euler method for $T = 5 0 0$ timesteps of duration $\\tau / 1 0$ . ",
+        "bbox": [
+            173,
+            239,
+            825,
+            338
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "text",
+        "text": "To perform a 2D navigation task with the RNN, we linearly combine the firing rates of units in the network to estimate the current location of the animal. The responses of the two linear readout neurons, $y _ { 1 } ( t )$ and $y _ { 2 } ( t )$ , are given by the following equation: ",
+        "bbox": [
+            174,
+            344,
+            825,
+            387
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "equation",
+        "img_path": "images/d64453343a45b8133e59ab95038edf8c7ca7e2bd28bd901cacdd4eff0e7578b4.jpg",
+        "text": "$$\ny _ { j } ( t ) = \\sum _ { i = 1 } ^ { N } W _ { j i } ^ { \\mathrm { o u t } } u _ { i } ( t )\n$$",
+        "text_format": "latex",
+        "bbox": [
+            419,
+            390,
+            576,
+            434
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "text",
+        "text": "2.2 INPUT TO THE NETWORK ",
+        "text_level": 1,
+        "bbox": [
+            174,
+            449,
+            387,
+            463
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "text",
+        "text": "The network inputs and outputs were inspired by simple spatial navigation tasks in 2D open environments. The task resembles dead-reckoning (sometimes referred to as path integration), which is ethologically relevant for many animal species (Darwin, 1873; Mittelstaedt & Mittelstaedt, 1980; Etienne & Jeffery, 2004; McNaughton et al., 2006). To be more specific, the inputs to the network were the animal’s speed and direction at each time step. Experimentally, it has been shown that the velocity signals exist in EC (Sargolini et al., 2006; Kropff et al., 2015; Hinman et al., 2016), and there is also evidence that such signals are necessary for grid formation (Winter et al., 2015a;b). ",
+        "bbox": [
+            173,
+            473,
+            825,
+            571
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "text",
+        "text": "Throughout the paper, we adopt the common assumption that the head direction of the animal coincides with the actual moving direction. The outputs were the x- and y-coordinates of the integrated position. The direction of the animal is modeled by modified Brownian motion to increase the probability of straight-runs, in order to be consistent with the typical rodent’s behavior in an open environment. The usage of such simple movement statistics has the advantage of having full control of the simulated trajectories. However, for future work it would be very interesting to test the model using different animals’ real movement trajectories to see how the results might change. ",
+        "bbox": [
+            173,
+            578,
+            825,
+            676
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "text",
+        "text": "Special care is taken when the animal is close to the boundary. The boundary of the environment will affect the statistics of the movement, as the animal cannot cross the boundary. This fact was reflected in the model by re-sampling the angular input variable until the input angle did not lead the animal outside the boundary. In the simulations shown below, the animal always starts from the center of the arena, but we verified that the results are insensitive to the starting locations. ",
+        "bbox": [
+            173,
+            683,
+            825,
+            753
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "text",
+        "text": "2.3 TRAINING ",
+        "text_level": 1,
+        "bbox": [
+            174,
+            768,
+            287,
+            784
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "text",
+        "text": "We optimized the network parameters $W ^ { \\mathrm { r e c } } , W ^ { \\mathrm { i n } } ,$ , $b$ and $W ^ { \\mathrm { o u t } }$ to minimize the squared error in equation (3) between target $\\mathbf { X } ^ { - }$ and $_ \\textrm { y }$ -coordinates from a two dimensional navigation task (performed in rectangular, hexagonal, and triangular arenas) and the network outputs generated according to equation (2). ",
+        "bbox": [
+            173,
+            794,
+            825,
+            851
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "equation",
+        "img_path": "images/c0586e3c42e764d474fb5fc1894e2ed9f63e5c96ba0d3cd1f2199ef9a9b3a013.jpg",
+        "text": "$$\nE = \\frac { 1 } { M T N _ { \\mathrm { o u t } } } \\sum _ { m , t , j = 1 } ^ { M , T , N _ { \\mathrm { o u t } } } ( y _ { j } ( t , m ) - y _ { j } ^ { \\mathrm { t a r g e t } } ( t , m ) ) ^ { 2 }\n$$",
+        "text_format": "latex",
+        "bbox": [
+            326,
+            849,
+            669,
+            895
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "text",
+        "text": "Parameters were updated with the Hessian-free algorithm (Martens & Sutskever, 2011) using minibatches of size $M = 5 0 0$ trials. In addition to minimizing the error function in equation (3) we regularized the input and output weights according to equation (4) and the squared firing rates of the units (referred to as metabolic cost) according to equation (5). In sum, the training aims to minimize a loss function, that consists of the error of the animal, the metabolic cost, and a penalty for large network parameters. ",
+        "bbox": [
+            173,
+            895,
+            823,
+            924
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "text",
+        "text": "",
+        "bbox": [
+            174,
+            103,
+            825,
+            160
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "equation",
+        "img_path": "images/e960c0e70ffb072304b7abf5d77743a5b44399b4e093a5b22d41b568af796bb4.jpg",
+        "text": "$$\n\\begin{array} { l } { { \\displaystyle R _ { L 2 } = \\frac { 1 } { N N _ { \\mathrm { i n } } } \\sum _ { i , j = 1 } ^ { N , N _ { \\mathrm { i n } } } ( W _ { i j } ^ { \\mathrm { i n } } ) ^ { 2 } + \\frac { 1 } { N N _ { \\mathrm { o u t } } } \\sum _ { i , j = 1 } ^ { N _ { \\mathrm { o u t } } , N } ( W _ { i j } ^ { \\mathrm { o u t } } ) ^ { 2 } } } \\\\ { { \\displaystyle R _ { F R } = \\frac { 1 } { N T M } \\sum _ { i , t , m = 1 } ^ { N , T , M } u _ { i } ( t , m ) ^ { 2 } } } \\end{array}\n$$",
+        "text_format": "latex",
+        "bbox": [
+            320,
+            180,
+            678,
+            273
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "We find that the results are qualitatively insensitive to the initialization schemes used for the recurrent weight matrix $W ^ { \\mathrm { r e c } }$ . For the results presented in this paper, simulations in the hexagonal environment were obtained by initializing the elements of $W ^ { \\mathrm { r e c } }$ to be zero mean Gaussian random variables with variance $1 . 5 ^ { 2 } \\dot { / } N$ , and simulations in the square and triangular environments were initialized with an orthogonal $W ^ { \\mathrm { r e c } }$ (Saxe et al., 2014). We initialized the bias $b$ and output weights $W ^ { \\mathrm { o u t } }$ to be zero. The elements of $W ^ { \\mathrm { i n } }$ were zero mean Gaussian variables with variance $1 / N _ { \\mathrm { i n } }$ . ",
+        "bbox": [
+            174,
+            281,
+            825,
+            366
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "image",
+        "img_path": "images/7cde4a241b1b10ed008d3d0dbd8d11d439e41445d8cafe267c621ecd7571dd08.jpg",
+        "image_caption": [
+            "Figure 2: Different types of spatial selective responses of units in the trained RNN. Example simulation results for three different environments (square, triangular, hexagon) are presented. Blue (yellow) represents low (high) activity. a) Grid-like responses. b) Band-like responses; c) Borderrelated responses; d) Spatially irregular responses. These responses can be spatially selective but they do not form a regular pattern defined in the conventional sense. "
+        ],
+        "image_footnote": [],
+        "bbox": [
+            205,
+            378,
+            790,
+            589
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "3 RESULTS ",
+        "text_level": 1,
+        "bbox": [
+            174,
+            699,
+            281,
+            715
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "We run simulation experiments in arenas with different boundary shapes, including square, triangular and hexagonal. Figure 1c shows a typical example of the model performance after training; the network (red trace) accurately tracks the animal’s actual path (black). ",
+        "bbox": [
+            174,
+            729,
+            825,
+            772
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "3.1 TUNING PROPERTIES OF THE MODEL NEURONS",
+        "text_level": 1,
+        "bbox": [
+            174,
+            789,
+            539,
+            803
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "We are mostly interested in what kind of representation the RNN has learned to solve this navigation task, and whether such a representation resembles the response properties of neurons in EC (Moser et al., 2008). ",
+        "bbox": [
+            176,
+            814,
+            825,
+            857
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "3.1.1 SPATIAL TUNING ",
+        "text_level": 1,
+        "bbox": [
+            176,
+            871,
+            346,
+            886
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "To test whether the trained RNN developed location-selective representations, we plot individual neurons’ mean activity level as a function of the animal’s location during spatial exploration. Note that these average response profiles should not be confused with the linear filters typically shown in feedforward networks. Surprisingly, we find neurons in the trained RNN show a range of interesting spatial response profiles. Examination of these response profiles suggests they can be classified into distinct functional types. Importantly, as we will show, these distinct spatial response profiles can be mapped naturally to known physiology in EC. The spatial responses of all units in trained networks are shown in the Appendix. ",
+        "bbox": [
+            174,
+            895,
+            823,
+            924
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "",
+        "bbox": [
+            174,
+            103,
+            825,
+            188
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "text",
+        "text": "Grid-like responses Most interestingly, we find some of the units in the RNN exhibit clear grid-like responses (Figure 2a). These firing patterns typically exhibit multiple firing fields, with each firing field exhibiting roughly circular symmetric or ellipse shape. Furthermore, the firing fields are highly structured, i.e., when combined, are arranged on a regular lattice. Furthermore, the structure of the response lattice depends on the shape of the boundary. In particular, training the network to perform self-localization in a square environment tends to give rectangular grids. In hexagonal and triangular environments, the grids are closer to triangular. ",
+        "bbox": [
+            174,
+            194,
+            825,
+            291
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "text",
+        "text": "Experimentally, it is shown that (medial) EC contains so-called grid cells which exhibit multiple firing fields that lie on a regular grid (Fyhn et al., 2004; Hafting et al., 2005). The grid-like firing patterns in our simulation are reminiscent of the grid cells in rodents and other mammals. However, we also notice that the the grid-like model responses typically exhibit few periods, not as many as experimental data (see Figure 1a). It is possible that using a larger network might reveal finer grid-patterns in our model. Nonetheless, it is surprising that the gird-like spatial representations can develop in our model, given there is no periodicity in the input. Another potential concern is that, experimentally it is reported that the grids are often on the corners of a triangular lattice (Hafting et al., 2005) even in square environments (see Figure 1a), though the grids are somewhat influenced by the shape of the environment. However, the rats in these experiments presumable had spatial experience in other environments with various boundary shapes. Experimentally, it would be interesting to see if grid cells would lie on a square lattice instead if the rats are raised in a single square environment - a situation we are simulating here. ",
+        "bbox": [
+            174,
+            299,
+            825,
+            478
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "text",
+        "text": "Border responses Many neurons in the RNN exhibit selectivity to the boundary (Figure 2c). Typically, they only encode a portion of the boundary, e.g. one piece of wall in a square shaped environment. Such properties are similar to the border cells discovered in rodent EC (Solstad et al., 2008; Savelli et al., 2008; Lever et al., 2009). Experimentally, border cells mainly fire along one piece of wall, although some have been observed to fire along multiple borders or along the whole boundary of the environment; interestingly, these multi-border responses were also observed in some RNN models. Currently, it is unclear how the boundary-like response profiles emerge (Solstad et al., 2008; Savelli et al., 2008; Lever et al., 2009). Our model points to the possibility that the border cells may emerge without the presence of tactile cues. Furthermore, it suggests that border cell formation may be related to the movement statistics of the animals, i.e. due to the asymmetry of the movement statistics along the boundary. ",
+        "bbox": [
+            174,
+            486,
+            825,
+            638
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "text",
+        "text": "Band-like responses Interestingly, some neurons in the RNN exhibit band-like responses (Figure 2b). In most of our simulations, these bands tend to be parallel to one of the boundaries. For some of the units, one of the bands overlaps the boundary, but for others, that is not the case. Experimentally, neurons with periodic-like firing patterns have been recently reported in rodent EC. In one study, it has been reported that a substantial portion of cells in EC exhibit band-like firing characteristics (Krupic et al., 2012). However, we note that based on the reported data in Krupic et al. (2012), the band pattern is not as clear as in our model. ",
+        "bbox": [
+            173,
+            645,
+            825,
+            743
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "text",
+        "text": "Spatially-stable but non-regular responses Besides the units described above, most of the remaining units also exhibit stable spatial responses, but they do not belong to the above categories. These response profiles can exhibit either one large irregular firing field; or multiple circular firing fields, but these firing fields do not show a regular pattern. Experimentally these types of cells have also been observed. In fact, it is recently reported that the non-grid spatial cells constitute a large portion of the neurons in Layer II and III of rodent EC (Diehl et al., 2017). ",
+        "bbox": [
+            174,
+            751,
+            825,
+            834
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "text",
+        "text": "3.1.2 SPEED TUNING AND HEAD DIRECTION TUNING",
+        "text_level": 1,
+        "bbox": [
+            174,
+            856,
+            552,
+            869
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "text",
+        "text": "Speed tuning We next ask how neurons in the RNN are tuned to the inputs. Many of the model neurons exhibit linear responses to the running speed of the animal, while some neurons show no selectivity to speed, as suggested by the near-flat response functions. Example response profiles are shown in Figure 3. Interestingly, we observe that the model border cells tend to have almost zero speed-tuning (e.g., see Figure 3g,h). ",
+        "bbox": [
+            176,
+            882,
+            823,
+            924
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "image",
+        "img_path": "images/318a67b1eb08bdf270e30eebad468380d0e5a80e744e9f80d58a5d693f3a94ed.jpg",
+        "image_caption": [
+            "Figure 3: Direction tuning and speed tuning for nine example units in an RNN trained in a triangular arena. For each unit, we show the spatial tuning, (head) directional tuning, speed tuning respectively, from left to right. a,b,c) The three model neurons show strong directional tuning, but the spatial tuning is weak and irregular. The three neurons also exhibit linear speed tuning. d,e,f) The three neurons exhibit grid-like firing patterns, and clear speed tuning. The strength of their direction tuning differ. ${ \\bf g } , { \\bf h } )$ Border cells exhibit weak and a bit complex directional tuning and almost no speed tuning. i) This band cell shows weak directional tuning, but strong speed tuning. "
+        ],
+        "image_footnote": [],
+        "bbox": [
+            214,
+            104,
+            784,
+            253
+        ],
+        "page_idx": 5
+    },
+    {
+        "type": "text",
+        "text": "",
+        "bbox": [
+            176,
+            397,
+            823,
+            426
+        ],
+        "page_idx": 5
+    },
+    {
+        "type": "text",
+        "text": "Head direction tuning A substantial portion of the model neurons show direction tuning. There are a diversity of direction tuning profiles, both in terms of the strength of the tuning and their preferred direction. Example tuning curves are shown in Figure 3, and the direction tuning curves of a complete population are shown in the Appendix. Interestingly, in general model neurons which show the strongest head direction tuning do not show a clear spatial firing pattern (see Figure 3a,b,c). This suggests that there are a group of neurons which are mostly responsible for encoding the direction. We also notice that neurons with clear grid-like firing can exhibit a variety of direction tuning strengths, from weak to strong (Figure 3d,e,f). In the Appendix, we quantify the relation between these different tuning properties at the whole population level, which show somewhat complex dependence. ",
+        "bbox": [
+            174,
+            433,
+            825,
+            571
+        ],
+        "page_idx": 5
+    },
+    {
+        "type": "text",
+        "text": "Experimentally, the heading direction tuning in EC is well-known, e.g., Sargolini et al. (2006). Both the grid and non-grid cells in EC exhibit head direction tuning (Sargolini et al., 2006). Furthermore, the linear speed dependence of the model neurons is similar to the properties of speed cells reported recently in EC (Kropff et al., 2015). Our result is also consistent with another recent study reporting that the majority of neurons in EC exhibit some amount of speed tuning (Hinman et al., 2016). It remains an open question experimentally, at a population level, how different types of tuning characteristics in EC relate to each other. ",
+        "bbox": [
+            174,
+            579,
+            825,
+            676
+        ],
+        "page_idx": 5
+    },
+    {
+        "type": "text",
+        "text": "3.1.3 DEVELOPMENT OF THE TUNING PROPERTIES",
+        "text_level": 1,
+        "bbox": [
+            174,
+            696,
+            537,
+            710
+        ],
+        "page_idx": 5
+    },
+    {
+        "type": "text",
+        "text": "We next investigate how the spatial response profiles evolve as learning/training progresses. We report two main observations. First, neurons that fire selectively along the boundary typically emerge first. Second, the grid-like responses with finer spatial tuning patterns only emerge later in training. For visualization, we perform dimensionality reduction using the t-SNE algorithm (Maaten & Hinton, 2008). This algorithm embeds 100 model neurons during three phases of training (early, intermediate, and late) into a two-dimensional space according to the similarity of their temporal responses. Here the similarity metric is taken to be firing rate correlation. In this 2D space as shown in Figure 4a, border cell representations appear early and stably persist through the end of training. Furthermore, early during training all responses are similar to the border related responses. In contrast, grid-like cells typically undergo a substantial change in firing pattern during training before settling into their final grid-like representation (Figure 4b). ",
+        "bbox": [
+            174,
+            722,
+            825,
+            875
+        ],
+        "page_idx": 5
+    },
+    {
+        "type": "text",
+        "text": "The developmental time line of the grid-like cells and border cells is roughly consistent with developmental studies in rodents. Experimentally, it is known that border cells emerge earlier in development, and they exist at about 2 weeks after the rat is born (Bjerknes et al., 2014). The grid cells mature only at about 4 weeks after birth (Langston et al., 2010; Wills et al., 2010; Bjerknes et al., 2014). Furthermore, our simulations suggest the reason why border cells emerge earlier in development may be that computationally it is easier to wire-up a network that gives rise to border cell responses. ",
+        "bbox": [
+            176,
+            882,
+            823,
+            924
+        ],
+        "page_idx": 5
+    },
+    {
+        "type": "text",
+        "text": "",
+        "bbox": [
+            174,
+            103,
+            825,
+            160
+        ],
+        "page_idx": 6
+    },
+    {
+        "type": "image",
+        "img_path": "images/9c19816ea23656a00f78df190e6e407bf4c84fe0856a9d69293f60f5972dfa0a.jpg",
+        "image_caption": [
+            "Figure 4: Development of border cells and grid-like cells. Early during training all responses are similar to the border related responses, and only as training continues do the grid-like cells emerge. We perform dimensionality reduction using the t-SNE algorithm on the firing rates of the neurons. Each dot represents one neuron $N = 1 0 0$ ), and the color represents different training stages (early/intermediate/late shown in blue/cyan/yellow). Each line shows the trajectory of a single highlighted neuron as its firing responses evolve during training. In panel a), we highlight the border representation. It appears there are four clusters of border cells, each responding to one wall of a square environment (spatial responses from four of these border cells are inset). These cells’ response profiles appear early and stably persist through training, illustrated by the short distance they travel in this space. In b), we show that the neurons which eventually become grid cells initially have tuning profiles similar to the border cells but then change their tuning substantially during learning. As a natural consequence, they need to travel a long distance in this space between the early and late phase of the training. Spatial responses are shown for four of these grid-like cells during the late phase of training. "
+        ],
+        "image_footnote": [],
+        "bbox": [
+            189,
+            176,
+            797,
+            366
+        ],
+        "page_idx": 6
+    },
+    {
+        "type": "text",
+        "text": "3.2 THE IMPORTANCE OF REGULARIZATION ",
+        "text_level": 1,
+        "bbox": [
+            176,
+            611,
+            488,
+            625
+        ],
+        "page_idx": 6
+    },
+    {
+        "type": "text",
+        "text": "We find appropriate regularizations of the RNN to be crucial for the emergence of grid-like representations. We only observed grid-like representations when the network was encouraged to store information while perturbed by noise. This was accomplished by setting the speed input to zero, e.g. zero speed $90 \\%$ of the time, and adding Gaussian noise to the network $\\xi _ { i } ( t )$ in equation (1)); the precise method for setting the speed input to zero and the value of the noise variance is not crucial for our simulations to develop grid-like representations. The cost function which aims to capture the penalization on the metabolic cost of the neural activity also acts as an important regularization. Our simulations show that the grid-like representation did not emerge without this metabolic cost. In Figure 5, we show typical simulation results for a square environment, with and without proper metabolic regularization. In the Appendix, we illustrate the effect of regularization further, in particular the role of injecting noise into the RNN units. ",
+        "bbox": [
+            174,
+            637,
+            825,
+            790
+        ],
+        "page_idx": 6
+    },
+    {
+        "type": "text",
+        "text": "Our results are consistent with the general notion on the importance of incorporating proper constraint for learning useful representations in neural networks (Bengio et al., 2013). Furthermore, it suggests that, to learn a model with response properties similar to neural systems it may be necessary to incorporate the relevant constraints, e.g., noise and metabolic cost. ",
+        "bbox": [
+            174,
+            797,
+            825,
+            853
+        ],
+        "page_idx": 6
+    },
+    {
+        "type": "text",
+        "text": "3.3 ERROR CORRECTION AROUND THE BOUNDARY ",
+        "text_level": 1,
+        "bbox": [
+            174,
+            871,
+            537,
+            883
+        ],
+        "page_idx": 6
+    },
+    {
+        "type": "text",
+        "text": "One natural question is whether the trained RNNs are able to perform localization when the path length exceeds the typical length used during training (500 steps), in particular given that noise in the network would gradually accumulate, leading to a decrease in localization performance. We test this by simulating paths that are several orders of magnitude longer. Somewhat surprisingly, we find the RNNs still perform well (Figure 6b). In fact, the squared error (averaged over every 10000 steps) is stable. The spatial response profiles of individual units also remain stable. This implies that the RNNs have acquired intrinsic error-correction mechanisms during training. ",
+        "bbox": [
+            176,
+            895,
+            823,
+            924
+        ],
+        "page_idx": 6
+    },
+    {
+        "type": "image",
+        "img_path": "images/476435913a7b64388291eece485ed2e49a2bc8befbf4d07d7b0df7b3c97e20d4.jpg",
+        "image_caption": [
+            "Figure 5: Complete set of spatial response profiles for 100 neurons in a RNN trained in a square environment. a) Without proper regularization, complex and periodic spatial response patterns do not emerge. b) With proper regularization, a rich set of periodic response patterns emerge, including grid-like responses. Regularization can also be adjusted to achieve spatial profiles intermediate between these two examples. "
+        ],
+        "image_footnote": [],
+        "bbox": [
+            184,
+            99,
+            794,
+            333
+        ],
+        "page_idx": 7
+    },
+    {
+        "type": "image",
+        "img_path": "images/1313022c4e140eff6c754917cddc7b20c35899670d67ff7b351c56f3f49df4d8.jpg",
+        "image_caption": [
+            "Figure 6: Error-correction happens at the boundary and the error is stable over time. At the boundary, the direction is resampled to avoid input velocities that lead to a path extending beyond the boundary of the environment. These changing input statistics at the boundary, termed a boundary interaction, are the only cue the RNN receives about the boundary. We find that the RNN uses the boundary interactions to correct the accumulated error between the true integrated input and its prediction based on the linear readout of equation (2). Panel a), the mean squared error increases when there are no boundary interactions, but then decreases after a boundary interaction, with more boundary interactions leading to greater error reduction. In the absence of further boundary interaction, the squared error would gradually increase again (blue curve) at roughly a constant rate. b) The network was trained using mini-batches of 500 timesteps but has stable error over a duration at least four orders of magnitude larger. The error of the RNN output (mean and standard deviation shown in black, computed based on 10000 timesteps) is compared to the error that would be achieved by an RNN outputting the best constant values (red). "
+        ],
+        "image_footnote": [],
+        "bbox": [
+            209,
+            436,
+            785,
+            593
+        ],
+        "page_idx": 7
+    },
+    {
+        "type": "text",
+        "text": "",
+        "bbox": [
+            174,
+            819,
+            825,
+            888
+        ],
+        "page_idx": 7
+    },
+    {
+        "type": "text",
+        "text": "As shown earlier, during training some of the RNN units develop boundary-related firing (Figure 2c), presumably by exploiting the change of input statistics around the boundary. We hypothesize that boundary interactions may enable error-correction through signals based on these boundary-related activities. Indeed, we find that boundary interactions can dramatically reduce the accumulated error (Figure 6a). Figure 6a shows that, without boundary interactions, on average the squared error grows roughly linearly as expected, however, interactions with the boundaries substantially reduce the error, and more frequent boundary interactions can reduce the error further. Error-correction on grid cells via boundary interactions has been proposed (Hardcastle et al., 2015; Pollock et al., 2017), however, we emphasize that the model proposed here develops the grid-like responses, boundary responses and the error-correction mechanisms all within the same neural network, thus potentially providing a unifying account of a diverse set of phenomena. ",
+        "bbox": [
+            174,
+            895,
+            823,
+            924
+        ],
+        "page_idx": 7
+    },
+    {
+        "type": "text",
+        "text": "",
+        "bbox": [
+            174,
+            103,
+            825,
+            229
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "4 DISCUSSION ",
+        "text_level": 1,
+        "bbox": [
+            174,
+            256,
+            310,
+            272
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "In this paper, we trained RNNs to perform path integration (dead-reckoning) in 2D arenas. We found that after training RNNs with appropriate regularization, the model neurons exhibit a variety of spatial and velocity tuning profiles that match neurophysiology in EC. What’s more, there is also similarity in terms of when these distinct neuron types emerge during training/development. The EC has long been thought to be involved in path integration and localization of the animal’s location (Moser et al., 2008). The general agreement between the different response properties in our model and the neurophysiology provide strong evidence supporting the hypothesis that the neural population in EC may provide an efficient code for representation self-locations based on the velocity input. ",
+        "bbox": [
+            174,
+            291,
+            825,
+            416
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "Recently, there has been increased interest in using complex neural network models to understand the neural code. But the focus has been on using feedforward architectures, in particular CNNs (LeCun et al., 1998). Given the abundant recurrent connections in the brain, it seems a particularly fruitful avenue to take advantage of the recent development in RNNs to help with neuroscience questions (Mante et al., 2013; Song et al., 2016; Miconi, 2017; Sussillo et al., 2015). Here, we only show one instance following this approach. However, the insight from this work could be general, and potentially useful for other cognitive functions as well. ",
+        "bbox": [
+            174,
+            424,
+            825,
+            521
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "The finding that metabolic constraints lead to the emergence of grid-like responses may be seen as conceptually related to the efficient coding hypothesis in visual processing (Barlow, 1961), in particular the seminal work on the emergence of the V1-like Gabor filters in a sparse coding model by Olshausen & Field (1996). Indeed, our work is partly inspired by these results. While there are conceptual similarities, however, we should also note there are differences between the sparse coding work and ours. First, the sparsity constraint in sparse coding can be naturally viewed as a particular prior while in the context of the recurrent network, it is difficult to interpret that way. Second, the grid-like responses are not the most sparse solution one could imagine. In fact, they are still quite dense compared to a more spatially localized representation. Third, the grid-like patterns that emerged in our network are not filters based on the raw input, rather the velocity inputs need to be integrated first in order to encode spatial locations. Our work is also inspired by recent work using the efficient coding idea to explain the functional architecture of the grid cells (Wei et al., 2015). It has been shown that efficient coding considerations could explain the particular set of grid scales observed in rodents (Stensola et al., 2012). However, in that work, the firing patterns of the neurons are assumed to have a lattice structure to start with. Furthermore, our work is related to the study by Sussillo and others (Sussillo et al., 2015), in which they show that regularization of RNN models are important for generating solutions that are similar to the neural activity observed in motor cortex. In Sussillo et al., a smoothness constraint together with others lead to simple oscillatory neural dynamics that well matches the neural data. We have not incorporated a smoothness constraint into our network. ",
+        "bbox": [
+            174,
+            529,
+            825,
+            804
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "Additionally, we note that there are a few recent studies which use place cells as the input to generate grid cells (Dordek et al., 2016; Stachenfeld et al., 2016), which are fundamentally different from our work. In these feedforward network models, the grid cells essentially perform dimensionality reduction based on the spatial input from place cells. However, the main issue with these models is that, it is unclear how place cells acquire spatial tuning in the first place. To the contrary, our model takes the animal’s velocity as the input, and addresses the question of how the spatial tuning can be generated from such input, which are known to exist in EC (Sargolini et al., 2006; Kropff et al., 2015). In another related study (Kanitscheider & Fiete, 2016), the authors train a RNN with ",
+        "bbox": [
+            174,
+            813,
+            823,
+            924
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "LSTM units (Hochreiter & Schmidhuber, 1997) to perform different navigation tasks. However, no grid-like spatial firing patterns are reported. ",
+        "bbox": [
+            173,
+            103,
+            823,
+            132
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "Although our model shows a qualitative match to the neural responses observed in the EC, nonetheless it has several major limitations, with each offering interesting future research directions. First, the learning rule we use seems to be biologically implausible. We are interested in exploring how a more biologically plausible learning rule could give rise to similar results (Lillicrap et al., 2016; Miconi, 2017; Guerguiev et al., 2017). Second, the simulation results do not show a variety of spatial scales in grid-like cells. Experimentally, it is known that grid cells have multiple spatial scales, that scale geometrically with a ratio 1.4 (Stensola et al., 2012), and this particular scale ratio is predicted by efficient coding of space (Wei et al., 2015). We are investigating how to modify the model to get a hierarchy of spatial scales, perhaps by incorporating more neurons or modifying the regularization. Last but not least, we have focused on the representation produced by the trained RNN. An equally important set of questions concern how the networks actually support the generation of such a representation. As a preliminary effort, we have examined the connectivity patterns of the trained network, and they do not seem to resemble the connectivity patterns required by standard attractor network models. Maybe this should not be seen as too surprising. After all, the trained networks can produce a diverse set of neural responses, while the previous models only led to grid responses. It would be interesting for future work to systematically examine the questions related to the underlying mechanisms. ",
+        "bbox": [
+            174,
+            140,
+            825,
+            375
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "5 ACKNOWLEDGEMENTS ",
+        "text_level": 1,
+        "bbox": [
+            176,
+            396,
+            398,
+            411
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "We thank members of the Center for Theoretical Neuroscience at Columbia University for useful discussions and three anonymous reviewers for constructive feedback. Research supported by NSF NeuroNex Award DBI-1707398 and NIH training grant 5T32NS064929 (CJC). ",
+        "bbox": [
+            176,
+            426,
+            821,
+            469
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "REFERENCES ",
+        "text_level": 1,
+        "bbox": [
+            176,
+            489,
+            285,
+            505
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "Dmitriy Aronov, Rhino Nevers, and David W Tank. Mapping of a non-spatial dimension by the hippocampalentorhinal circuit. Nature, 2017. ",
+        "bbox": [
+            174,
+            513,
+            823,
+            541
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "Horace B Barlow. Possible principles underlying the transformation of sensory messages. Sensory communication, pp. 217–234, 1961. ",
+        "bbox": [
+            171,
+            551,
+            825,
+            580
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "Yoshua Bengio, Aaron Courville, and Pascal Vincent. Representation learning: A review and new perspectives. IEEE transactions on pattern analysis and machine intelligence, 35(8):1798–1828, 2013. ",
+        "bbox": [
+            176,
+            588,
+            823,
+            631
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "Tale L Bjerknes, Edvard I Moser, and May-Britt Moser. Representation of geometric borders in the developing rat. Neuron, 82(1):71–78, 2014. ",
+        "bbox": [
+            173,
+            640,
+            825,
+            670
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "Yoram Burak and Ila R Fiete. Accurate path integration in continuous attractor network models of grid cells. PLoS computational biology, 5(2):e1000291, 2009. ",
+        "bbox": [
+            174,
+            678,
+            823,
+            708
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "Jonathan J Couey, Aree Witoelar, Sheng-Jia Zhang, Kang Zheng, Jing Ye, Benjamin Dunn, Rafal Czajkowski, May-Britt Moser, Edvard I Moser, Yasser Roudi, et al. Recurrent inhibitory circuitry as a mechanism for grid formation. Nature neuroscience, 16(3):318–324, 2013. ",
+        "bbox": [
+            178,
+            715,
+            823,
+            758
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "Charles Darwin. Origin of certain instincts. Nature, 7:417–418, 1873. ",
+        "bbox": [
+            173,
+            767,
+            632,
+            784
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "Geoffrey W Diehl, Olivia J Hon, Stefan Leutgeb, and Jill K Leutgeb. Grid and nongrid cells in medial entorhinal cortex represent spatial location and environmental features with complementary coding schemes. Neuron, 94(1):83–92, 2017. ",
+        "bbox": [
+            178,
+            791,
+            823,
+            834
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "Yedidyah Dordek, Daniel Soudry, Ron Meir, and Dori Derdikman. Extracting grid cell characteristics from place cell inputs using non-negative principal component analysis. eLife, 5:e10094, 2016. ",
+        "bbox": [
+            176,
+            843,
+            821,
+            886
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "Ariane S Etienne and Kathryn J Jeffery. Path integration in mammals. Hippocampus, 14(2):180– 192, 2004. ",
+        "bbox": [
+            173,
+            895,
+            821,
+            922
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "Marianne Fyhn, Sturla Molden, Menno P Witter, Edvard I Moser, and May-Britt Moser. Spatial representation in the entorhinal cortex. Science, 305(5688):1258–1264, 2004. ",
+        "bbox": [
+            171,
+            103,
+            823,
+            132
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Marianne Fyhn, Torkel Hafting, Menno P Witter, Edvard I Moser, and May-Britt Moser. Grid cells in mice. Hippocampus, 18(12):1230–1238, 2008. ",
+        "bbox": [
+            171,
+            141,
+            823,
+            170
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Alex Graves, Abdel-Rahman Mohamed, and Geoffrey Hinton. Speech recognition with deep recurrent neural networks. In Acoustics, speech and signal processing (icassp), 2013 ieee international conference on, pp. 6645–6649. IEEE, 2013. ",
+        "bbox": [
+            176,
+            179,
+            823,
+            222
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Karol Gregor, Ivo Danihelka, Alex Graves, Danilo Jimenez Rezende, and Daan Wierstra. Draw: A recurrent neural network for image generation. arXiv preprint arXiv:1502.04623, 2015. ",
+        "bbox": [
+            173,
+            231,
+            821,
+            261
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Jordan Guerguiev, Timothy P Lillicrap, and Blake A Richards. Towards deep learning with segregated dendrites. eLife, 6, 2017. ",
+        "bbox": [
+            174,
+            268,
+            820,
+            299
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Torkel Hafting, Marianne Fyhn, Sturla Molden, May-Britt Moser, and Edvard I Moser. Microstructure of a spatial map in the entorhinal cortex. Nature, 436(7052):801–806, 2005. ",
+        "bbox": [
+            174,
+            306,
+            820,
+            337
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Kiah Hardcastle, Surya Ganguli, and Lisa M Giocomo. Environmental boundaries as an error correction mechanism for grid cells. Neuron, 86(3):827–839, 2015. ",
+        "bbox": [
+            176,
+            344,
+            820,
+            375
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "James R Hinman, Mark P Brandon, Jason R Climer, G William Chapman, and Michael E Hasselmo. Multiple running speed signals in medial entorhinal cortex. Neuron, 91(3):666–679, 2016. ",
+        "bbox": [
+            176,
+            382,
+            821,
+            412
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Sepp Hochreiter and Jurgen Schmidhuber. Long short-term memory. ¨ Neural computation, 9(8): 1735–1780, 1997. ",
+        "bbox": [
+            174,
+            421,
+            823,
+            450
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "David H Hubel and Torsten N Wiesel. Receptive fields, binocular interaction and functional architecture in the cat’s visual cortex. The Journal of physiology, 160(1):106–154, 1962. ",
+        "bbox": [
+            174,
+            459,
+            823,
+            489
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Joshua Jacobs, Christoph T Weidemann, Jonathan F Miller, Alec Solway, John F Burke, Xue-Xin Wei, Nanthia Suthana, Michael R Sperling, Ashwini D Sharan, Itzhak Fried, et al. Direct recordings of grid-like neuronal activity in human spatial navigation. Nature neuroscience, 16(9):1188– 1190, 2013. ",
+        "bbox": [
+            173,
+            497,
+            825,
+            554
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Ingmar Kanitscheider and Ila Fiete. Training recurrent networks to generate hypotheses about how the brain solves hard navigation problems. arXiv preprint arXiv:1609.09059, 2016. ",
+        "bbox": [
+            173,
+            563,
+            823,
+            593
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Tim Christian Kietzmann, Patrick McClure, and Nikolaus Kriegeskorte. Deep neural networks in computational neuroscience. bioRxiv, pp. 133504, 2017. ",
+        "bbox": [
+            173,
+            601,
+            823,
+            631
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Nathaniel J Killian, Michael J Jutras, and Elizabeth A Buffalo. A map of visual space in the primate entorhinal cortex. Nature, 491(7426):761–764, 2012. ",
+        "bbox": [
+            173,
+            638,
+            823,
+            669
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Nikolaus Kriegeskorte. Deep neural networks: a new framework for modeling biological vision and brain information processing. Annual Review of Vision Science, 1:417–446, 2015. ",
+        "bbox": [
+            171,
+            678,
+            823,
+            707
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Alex Krizhevsky, Ilya Sutskever, and Geoffrey E Hinton. Imagenet classification with deep convolutional neural networks. In Advances in neural information processing systems, pp. 1097–1105, 2012. ",
+        "bbox": [
+            176,
+            714,
+            825,
+            757
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Emilio Kropff, James E Carmichael, May-Britt Moser, and Edvard I Moser. Speed cells in the medial entorhinal cortex. Nature, 523(7561):419–424, 2015. ",
+        "bbox": [
+            173,
+            767,
+            823,
+            796
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Julija Krupic, Neil Burgess, and John O?Keefe. Neural representations of location composed of spatially periodic bands. Science, 337(6096):853–857, 2012. ",
+        "bbox": [
+            171,
+            804,
+            823,
+            835
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Rosamund F Langston, James A Ainge, Jonathan J Couey, Cathrin B Canto, Tale L Bjerknes, Menno P Witter, Edvard I Moser, and May-Britt Moser. Development of the spatial representation system in the rat. Science, 328(5985):1576–1580, 2010. ",
+        "bbox": [
+            176,
+            843,
+            821,
+            886
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Yann LeCun, Leon Bottou, Yoshua Bengio, and Patrick Haffner. Gradient-based learning applied to ´ document recognition. Proceedings of the IEEE, 86(11):2278–2324, 1998. ",
+        "bbox": [
+            176,
+            895,
+            821,
+            924
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Yann LeCun, Yoshua Bengio, and Geoffrey Hinton. Deep learning. Nature, 521(7553):436–444, 2015. ",
+        "bbox": [
+            173,
+            103,
+            823,
+            132
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "Colin Lever, Stephen Burton, Ali Jeewajee, John O’Keefe, and Neil Burgess. Boundary vector cells in the subiculum of the hippocampal formation. The journal of neuroscience, 29(31):9771–9777, 2009. ",
+        "bbox": [
+            174,
+            141,
+            823,
+            184
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "Timothy P Lillicrap, Daniel Cownden, Douglas B Tweed, and Colin J Akerman. Random synaptic feedback weights support error backpropagation for deep learning. Nature communications, 7, 2016. ",
+        "bbox": [
+            174,
+            193,
+            825,
+            236
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "Laurens van der Maaten and Geoffrey Hinton. Visualizing data using t-sne. Journal of Machine Learning Research, 9(Nov):2579–2605, 2008. ",
+        "bbox": [
+            174,
+            244,
+            821,
+            273
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "Valerio Mante, David Sussillo, Krishna V Shenoy, and William T Newsome. Context-dependent computation by recurrent dynamics in prefrontal cortex. Nature, 503(7474):78–84, 2013. ",
+        "bbox": [
+            174,
+            282,
+            821,
+            313
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "James Martens and Ilya Sutskever. Learning recurrent neural networks with hessian-free optimization. pp. 10331040, 2011. ",
+        "bbox": [
+            173,
+            320,
+            823,
+            351
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "Bruce L McNaughton, Francesco P Battaglia, Ole Jensen, Edvard I Moser, and May-Britt Moser. Path integration and the neural basis of the ’cognitive map’. Nature Reviews Neuroscience, 7(8): 663–678, 2006. ",
+        "bbox": [
+            173,
+            358,
+            825,
+            401
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "Thomas Miconi. Biologically plausible learning in recurrent neural networks reproduces neural dynamics observed during cognitive tasks. eLife, 6:e20899, 2017. ",
+        "bbox": [
+            173,
+            411,
+            823,
+            440
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "M-L Mittelstaedt and H Mittelstaedt. Homing by path integration in a mammal. Naturwissenschaften, 67(11):566–567, 1980. ",
+        "bbox": [
+            171,
+            449,
+            823,
+            478
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "Volodymyr Mnih, Koray Kavukcuoglu, David Silver, Andrei A Rusu, Joel Veness, Marc G Bellemare, Alex Graves, Martin Riedmiller, Andreas K Fidjeland, Georg Ostrovski, et al. Human-level control through deep reinforcement learning. Nature, 518(7540):529–533, 2015. ",
+        "bbox": [
+            173,
+            486,
+            825,
+            530
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "Edvard I Moser, Emilio Kropff, and May-Britt Moser. Place cells, grid cells, and the brain’s spatial representation system. Annu. Rev. Neurosci., 31:69–89, 2008. ",
+        "bbox": [
+            171,
+            539,
+            823,
+            568
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "John O’Keefe. Place units in the hippocampus of the freely moving rat. Experimental neurology, 51 (1):78–109, 1976. ",
+        "bbox": [
+            171,
+            577,
+            823,
+            606
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "Bruno A Olshausen and David J Field. Emergence of simple-cell receptive field properties by learning a sparse code for natural images. Nature, 381(6583):607, 1996. ",
+        "bbox": [
+            168,
+            614,
+            823,
+            645
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "Aaron van den Oord, Nal Kalchbrenner, and Koray Kavukcuoglu. Pixel recurrent neural networks. arXiv preprint arXiv:1601.06759, 2016. ",
+        "bbox": [
+            173,
+            652,
+            823,
+            683
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "Eli Pollock, Niral Desai, Xue-Xin Wei, and Vijay B Balasubramanian. A mechanism for selforganized error-correction of grid cells by border cells. CoSyNe abstract, 2017. ",
+        "bbox": [
+            171,
+            690,
+            821,
+            720
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "Alexei Samsonovich and Bruce L McNaughton. Path integration and cognitive mapping in a continuous attractor neural network model. Journal of Neuroscience, 17(15):5900–5920, 1997. ",
+        "bbox": [
+            171,
+            728,
+            821,
+            758
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "Francesca Sargolini, Marianne Fyhn, Torkel Hafting, Bruce L McNaughton, Menno P Witter, MayBritt Moser, and Edvard I Moser. Conjunctive representation of position, direction, and velocity in entorhinal cortex. Science, 312(5774):758–762, 2006. ",
+        "bbox": [
+            174,
+            767,
+            823,
+            810
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "Francesco Savelli, D Yoganarasimha, and James J Knierim. Influence of boundary removal on the spatial representations of the medial entorhinal cortex. Hippocampus, 18(12):1270, 2008. ",
+        "bbox": [
+            171,
+            819,
+            823,
+            848
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "Andrew M Saxe, James L McClelland, and Surya Ganguli. Exact solutions to the nonlinear dynamics of learning in deep linear neural networks. 2014. ",
+        "bbox": [
+            173,
+            857,
+            821,
+            886
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "Trygve Solstad, Charlotte N Boccara, Emilio Kropff, May-Britt Moser, and Edvard I Moser. Representation of geometric borders in the entorhinal cortex. Science, 322(5909):1865–1868, 2008. ",
+        "bbox": [
+            174,
+            895,
+            820,
+            924
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "H Francis Song, Guangyu R Yang, and Xiao-Jing Wang. Training excitatory-inhibitory recurrent neural networks for cognitive tasks: A simple and flexible framework. PLoS Comput Biol, 12(2): e1004792, 2016. ",
+        "bbox": [
+            174,
+            103,
+            823,
+            146
+        ],
+        "page_idx": 12
+    },
+    {
+        "type": "text",
+        "text": "Kimberly Lauren Stachenfeld, Matthew M Botvinick, and Samuel J Gershman. The hippocampus as a predictive map. bioRxiv, pp. 097170, 2016. ",
+        "bbox": [
+            173,
+            155,
+            823,
+            184
+        ],
+        "page_idx": 12
+    },
+    {
+        "type": "text",
+        "text": "Hanne Stensola, Tor Stensola, Trygve Solstad, Kristian Frøland, May-Britt Moser, and Edvard I Moser. The entorhinal grid map is discretized. Nature, 492(7427):72–78, 2012. ",
+        "bbox": [
+            173,
+            193,
+            823,
+            222
+        ],
+        "page_idx": 12
+    },
+    {
+        "type": "text",
+        "text": "David Sussillo, Mark M Churchland, Matthew T Kaufman, and Krishna V Shenoy. A neural network that finds a naturalistic solution for the production of muscle activity. Nature neuroscience, 18(7): 1025–1033, 2015. ",
+        "bbox": [
+            174,
+            231,
+            823,
+            272
+        ],
+        "page_idx": 12
+    },
+    {
+        "type": "text",
+        "text": "Lucas Theis and Matthias Bethge. Generative image modeling using spatial lstms. In Advances in Neural Information Processing Systems, pp. 1927–1935, 2015. ",
+        "bbox": [
+            174,
+            281,
+            823,
+            311
+        ],
+        "page_idx": 12
+    },
+    {
+        "type": "text",
+        "text": "Xue-Xin Wei, Jason Prentice, and Vijay Balasubramanian. A principle of economy predicts the functional architecture of grid cells. Elife, 4:e08362, 2015. ",
+        "bbox": [
+            174,
+            319,
+            823,
+            349
+        ],
+        "page_idx": 12
+    },
+    {
+        "type": "text",
+        "text": "B Willmore and DJ Tolhurst. Characterizing the sparseness of neural codes. Network, 12:255–270, 2001. ",
+        "bbox": [
+            174,
+            357,
+            823,
+            386
+        ],
+        "page_idx": 12
+    },
+    {
+        "type": "text",
+        "text": "Tom J Wills, Francesca Cacucci, Neil Burgess, and John O’keefe. Development of the hippocampal cognitive map in preweanling rats. Science, 328(5985):1573–1576, 2010. ",
+        "bbox": [
+            174,
+            395,
+            825,
+            424
+        ],
+        "page_idx": 12
+    },
+    {
+        "type": "text",
+        "text": "Shawn S. Winter, Benjamin J. Clark, and Jeffrey S. Taube. Disruption of the head direction cell network impairs the parahippocampal grid cell signal. Science, 347(6224):870–874, 2015a. ",
+        "bbox": [
+            173,
+            433,
+            823,
+            463
+        ],
+        "page_idx": 12
+    },
+    {
+        "type": "text",
+        "text": "Shawn S. Winter, Max L. Mehlman, Benjamin J. Clark, and Jeffrey S. Taube. Passive transport disrupts grid signals in the parahippocampal cortex. Current Biology, 25:2493–2502, 2015b. ",
+        "bbox": [
+            173,
+            469,
+            823,
+            501
+        ],
+        "page_idx": 12
+    },
+    {
+        "type": "text",
+        "text": "Daniel LK Yamins and James J DiCarlo. Using goal-driven deep learning models to understand sensory cortex. Nature neuroscience, 19(3):356–365, 2016. ",
+        "bbox": [
+            173,
+            508,
+            823,
+            537
+        ],
+        "page_idx": 12
+    },
+    {
+        "type": "text",
+        "text": "Daniel LK Yamins, Ha Hong, Charles F Cadieu, Ethan A Solomon, Darren Seibert, and James J DiCarlo. Performance-optimized hierarchical models predict neural responses in higher visual cortex. Proceedings of the National Academy of Sciences, 111(23):8619–8624, 2014. ",
+        "bbox": [
+            174,
+            546,
+            823,
+            589
+        ],
+        "page_idx": 12
+    },
+    {
+        "type": "text",
+        "text": "Michael M Yartsev, Menno P Witter, and Nachum Ulanovsky. Grid cells without theta oscillations in the entorhinal cortex of bats. Nature, 479(7371):103–107, 2011. ",
+        "bbox": [
+            174,
+            598,
+            823,
+            627
+        ],
+        "page_idx": 12
+    },
+    {
+        "type": "text",
+        "text": "A TRIANGULAR ENVIRONMENT ",
+        "text_level": 1,
+        "bbox": [
+            176,
+            102,
+            452,
+            117
+        ],
+        "page_idx": 13
+    },
+    {
+        "type": "image",
+        "img_path": "images/9fec0b40f547311967f788a5e33c37ca83be210239a76a72034b231f75f6b016.jpg",
+        "image_caption": [],
+        "image_footnote": [],
+        "bbox": [
+            348,
+            136,
+            648,
+            320
+        ],
+        "page_idx": 13
+    },
+    {
+        "type": "image",
+        "img_path": "images/76d7068d3c86954df82911a198a111d047595eeb46564b8603a26da4acf15c38.jpg",
+        "image_caption": [],
+        "image_footnote": [],
+        "bbox": [
+            174,
+            383,
+            825,
+            568
+        ],
+        "page_idx": 13
+    },
+    {
+        "type": "text",
+        "text": "Noise and metabolic cost are important for grid-like representations. The figure on the left shows the spatial responses for a network trained with noise and no metabolic cost. The figure on the right shows the spatial responses for a network trained with no noise and the metabolic cost. ",
+        "bbox": [
+            174,
+            579,
+            825,
+            622
+        ],
+        "page_idx": 13
+    },
+    {
+        "type": "image",
+        "img_path": "images/ace176f80461890db9b194a0b86e5c7936f97026efec6348b0297c43783b3061.jpg",
+        "image_caption": [],
+        "image_footnote": [],
+        "bbox": [
+            313,
+            95,
+            640,
+            559
+        ],
+        "page_idx": 14
+    },
+    {
+        "type": "text",
+        "text": "B RECTANGULAR ENVIRONMENT ",
+        "text_level": 1,
+        "bbox": [
+            174,
+            102,
+            467,
+            117
+        ],
+        "page_idx": 15
+    },
+    {
+        "type": "image",
+        "img_path": "images/97fb3c206da8fbace0f4d8eeba7efe4308290448d5d9e2cba584838472b30abd.jpg",
+        "image_caption": [],
+        "image_footnote": [],
+        "bbox": [
+            321,
+            136,
+            650,
+            580
+        ],
+        "page_idx": 15
+    },
+    {
+        "type": "image",
+        "img_path": "",
+        "image_caption": [
+            "Speed\tinput "
+        ],
+        "image_footnote": [],
+        "bbox": [
+            308,
+            299,
+            643,
+            835
+        ],
+        "page_idx": 15
+    },
+    {
+        "type": "text",
+        "text": "C HEXAGONAL ENVIRONMENT ",
+        "text_level": 1,
+        "bbox": [
+            174,
+            102,
+            446,
+            117
+        ],
+        "page_idx": 16
+    },
+    {
+        "type": "image",
+        "img_path": "images/d7c955da2b4d05471ca42b23f6a780ea605e2aed77c7b520059872b81c33fead.jpg",
+        "image_caption": [],
+        "image_footnote": [],
+        "bbox": [
+            348,
+            136,
+            650,
+            321
+        ],
+        "page_idx": 16
+    },
+    {
+        "type": "image",
+        "img_path": "images/64fe9f96b9155e2bd1b22d7a344f3085ff647a473ca40df4a123eb8758cf0801.jpg",
+        "image_caption": [],
+        "image_footnote": [],
+        "bbox": [
+            348,
+            373,
+            635,
+            553
+        ],
+        "page_idx": 16
+    },
+    {
+        "type": "text",
+        "text": "Angular\tinput\t(0\tto\t360\tdegrees) ",
+        "bbox": [
+            424,
+            577,
+            573,
+            587
+        ],
+        "page_idx": 16
+    },
+    {
+        "type": "text",
+        "text": "Speed\ttuning ",
+        "bbox": [
+            467,
+            637,
+            529,
+            647
+        ],
+        "page_idx": 16
+    },
+    {
+        "type": "image",
+        "img_path": "images/7eb735c5ab485558870c8b0720dd627699e69018e8cbacfa58c8d343cfc8a2d1.jpg",
+        "image_caption": [
+            "Speed\tinput "
+        ],
+        "image_footnote": [],
+        "bbox": [
+            338,
+            661,
+            633,
+            830
+        ],
+        "page_idx": 16
+    },
+    {
+        "type": "text",
+        "text": "To quantify the speed selectivity of each unit we first fit a line to the tuning curve of unit activity as a function of speed. The speed selectivity is the absolute value of the slope. If the unit activity is not modulated by speed then the speed selectivity is 0. To quantify the direction selectivity of each unit we calculated the average unit activity as a function of direction input and then took the maximum minus minimum of this tuning curve. If the unit activity is not modulated by direction then the direction selectivity is 0. To quantify the spatial selectivity we used lifetime sparseness (Willmore & Tolhurst, 2001). If the unit activity is not modulated by spatial location then the spatial selectivity is 0. Each dot in the figures below show the selectivity for a single unit. ",
+        "bbox": [
+            173,
+            133,
+            825,
+            246
+        ],
+        "page_idx": 17
+    },
+    {
+        "type": "image",
+        "img_path": "images/7d81c0e39724fab8f48b0863c121f5336b8e91d07c652f62dff58ab0d2f8945e.jpg",
+        "image_caption": [],
+        "image_footnote": [],
+        "bbox": [
+            174,
+            247,
+            828,
+            868
+        ],
+        "page_idx": 17
+    },
+    {
+        "type": "text",
+        "text": "E ADDITIONAL TRAINING DETAILS ",
+        "text_level": 1,
+        "bbox": [
+            176,
+            102,
+            480,
+            117
+        ],
+        "page_idx": 18
+    },
+    {
+        "type": "text",
+        "text": "During training we tried to balance all three terms we were minimizing $( E , R _ { L 2 }$ , and $R _ { F R }$ ) so no single term was neglected or dominated. At the beginning of training we weighted the regularization term $R _ { L 2 }$ to be equal to the error function $E$ and then decreased the weighting on $R _ { L 2 }$ according to the schedule used by Martens & Sutskever (2011). We adaptively adjusted the weighting on $R _ { F R }$ , starting from an initial value of $E / 1 0$ and enforcing an upper bound of $E / 3$ as training progressed. We found this training procedure improved training performance and led to more interesting representations. ",
+        "bbox": [
+            174,
+            133,
+            825,
+            231
+        ],
+        "page_idx": 18
+    }
+]
\ No newline at end of file
diff --git a/parse/train/B17JTOe0-/B17JTOe0-_middle.json b/parse/train/B17JTOe0-/B17JTOe0-_middle.json
new file mode 100644
index 0000000000000000000000000000000000000000..cc7113afed02d829cb916b0fd3f8a4597468db06
--- /dev/null
+++ b/parse/train/B17JTOe0-/B17JTOe0-_middle.json
@@ -0,0 +1,37139 @@
+{
+    "pdf_info": [
+        {
+            "preproc_blocks": [
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        79,
+                        503,
+                        136
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                79,
+                                506,
+                                100
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        79,
+                                        506,
+                                        100
+                                    ],
+                                    "score": 1.0,
+                                    "content": "EMERGENCE OF GRID-LIKE REPRESENTATIONS BY",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                102,
+                                453,
+                                118
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        102,
+                                        453,
+                                        118
+                                    ],
+                                    "score": 1.0,
+                                    "content": "TRAINING RECURRENT NEURAL NETWORKS TO",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                122,
+                                358,
+                                138
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        122,
+                                        358,
+                                        138
+                                    ],
+                                    "score": 1.0,
+                                    "content": "PERFORM SPATIAL LOCALIZATION",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        }
+                    ],
+                    "index": 1
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        112,
+                        154,
+                        273,
+                        199
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                113,
+                                155,
+                                269,
+                                167
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        113,
+                                        155,
+                                        269,
+                                        167
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Christopher J. Cueva∗, Xue-Xin Wei∗",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                111,
+                                164,
+                                199,
+                                178
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        111,
+                                        164,
+                                        199,
+                                        178
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Columbia University",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                111,
+                                176,
+                                227,
+                                188
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        111,
+                                        176,
+                                        227,
+                                        188
+                                    ],
+                                    "score": 1.0,
+                                    "content": "New York, NY 10027, USA",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                112,
+                                187,
+                                275,
+                                201
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        112,
+                                        187,
+                                        275,
+                                        201
+                                    ],
+                                    "score": 1.0,
+                                    "content": "{ccueva,weixxpku}@gmail.com",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        }
+                    ],
+                    "index": 4.5
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        278,
+                        228,
+                        333,
+                        240
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                276,
+                                227,
+                                336,
+                                242
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        276,
+                                        227,
+                                        336,
+                                        242
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ABSTRACT",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        }
+                    ],
+                    "index": 7
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        143,
+                        255,
+                        469,
+                        419
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                141,
+                                254,
+                                469,
+                                268
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        254,
+                                        469,
+                                        268
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Decades of research on the neural code underlying spatial navigation have re-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                265,
+                                470,
+                                279
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        265,
+                                        470,
+                                        279
+                                    ],
+                                    "score": 1.0,
+                                    "content": "vealed a diverse set of neural response properties. The Entorhinal Cortex (EC) of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                142,
+                                277,
+                                469,
+                                289
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        142,
+                                        277,
+                                        469,
+                                        289
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the mammalian brain contains a rich set of spatial correlates, including grid cells",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                142,
+                                288,
+                                469,
+                                301
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        142,
+                                        288,
+                                        469,
+                                        301
+                                    ],
+                                    "score": 1.0,
+                                    "content": "which encode space using tessellating patterns. However, the mechanisms and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                299,
+                                469,
+                                311
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        299,
+                                        469,
+                                        311
+                                    ],
+                                    "score": 1.0,
+                                    "content": "functional significance of these spatial representations remain largely mysterious.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                310,
+                                470,
+                                323
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        310,
+                                        470,
+                                        323
+                                    ],
+                                    "score": 1.0,
+                                    "content": "As a new way to understand these neural representations, we trained recurrent",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                321,
+                                469,
+                                334
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        321,
+                                        469,
+                                        334
+                                    ],
+                                    "score": 1.0,
+                                    "content": "neural networks (RNNs) to perform navigation tasks in 2D arenas based on veloc-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                333,
+                                469,
+                                344
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        333,
+                                        469,
+                                        344
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ity inputs. Surprisingly, we find that grid-like spatial response patterns emerge in",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                343,
+                                469,
+                                355
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        343,
+                                        469,
+                                        355
+                                    ],
+                                    "score": 1.0,
+                                    "content": "trained networks, along with units that exhibit other spatial correlates, including",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                354,
+                                469,
+                                366
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        354,
+                                        469,
+                                        366
+                                    ],
+                                    "score": 1.0,
+                                    "content": "border cells and band-like cells. All these different functional types of neurons",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                365,
+                                470,
+                                378
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        365,
+                                        470,
+                                        378
+                                    ],
+                                    "score": 1.0,
+                                    "content": "have been observed experimentally. The order of the emergence of grid-like and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                142,
+                                376,
+                                469,
+                                387
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        142,
+                                        376,
+                                        469,
+                                        387
+                                    ],
+                                    "score": 1.0,
+                                    "content": "border cells is also consistent with observations from developmental studies. To-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                387,
+                                469,
+                                399
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        387,
+                                        469,
+                                        399
+                                    ],
+                                    "score": 1.0,
+                                    "content": "gether, our results suggest that grid cells, border cells and others as observed in EC",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                398,
+                                470,
+                                410
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        398,
+                                        470,
+                                        410
+                                    ],
+                                    "score": 1.0,
+                                    "content": "may be a natural solution for representing space efficiently given the predominant",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                409,
+                                317,
+                                420
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        409,
+                                        317,
+                                        420
+                                    ],
+                                    "score": 1.0,
+                                    "content": "recurrent connections in the neural circuits.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        }
+                    ],
+                    "index": 15
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        446,
+                        206,
+                        459
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                444,
+                                208,
+                                461
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        444,
+                                        208,
+                                        461
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1 INTRODUCTION",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        }
+                    ],
+                    "index": 23
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        472,
+                        505,
+                        561
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                473,
+                                505,
+                                485
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        473,
+                                        505,
+                                        485
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Understanding the neural code in the brain has long been driven by studying feed-forward architec-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                484,
+                                505,
+                                496
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        484,
+                                        505,
+                                        496
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tures, starting from Hubel and Wiesel’s famous proposal on the origin of orientation selectivity in",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                495,
+                                505,
+                                506
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        495,
+                                        505,
+                                        506
+                                    ],
+                                    "score": 1.0,
+                                    "content": "primary visual cortex (Hubel & Wiesel, 1962). Inspired by the recent development in deep learn-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                505,
+                                505,
+                                518
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        505,
+                                        505,
+                                        518
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ing (Krizhevsky et al., 2012; LeCun et al., 2015; Hochreiter & Schmidhuber, 1997; Mnih et al.,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                516,
+                                505,
+                                529
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        516,
+                                        505,
+                                        529
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2015), there has been a burst of interest in applying deep feedforward models, in particular convolu-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                527,
+                                505,
+                                540
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        527,
+                                        505,
+                                        540
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tional neural networks (CNN) (LeCun et al., 1998), to study the sensory systems, which hierarchi-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                538,
+                                505,
+                                551
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        538,
+                                        505,
+                                        551
+                                    ],
+                                    "score": 1.0,
+                                    "content": "cally extract useful features from sensory inputs (see e.g., Yamins et al. (2014); Kriegeskorte (2015);",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                549,
+                                318,
+                                562
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        549,
+                                        318,
+                                        562
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Kietzmann et al. (2017); Yamins & DiCarlo (2016)).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        }
+                    ],
+                    "index": 27.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        567,
+                        505,
+                        708
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                566,
+                                505,
+                                579
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        566,
+                                        505,
+                                        579
+                                    ],
+                                    "score": 1.0,
+                                    "content": "For more cognitive tasks, neural systems often need to maintain certain internal representations",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                577,
+                                505,
+                                589
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        577,
+                                        505,
+                                        589
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of relevant variables in the absence of external stimuli- a process that requires more than feature",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                589,
+                                506,
+                                600
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        589,
+                                        506,
+                                        600
+                                    ],
+                                    "score": 1.0,
+                                    "content": "extraction. We will focus on spatial navigation, which typically requires the brain to maintain a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                599,
+                                505,
+                                612
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        599,
+                                        505,
+                                        612
+                                    ],
+                                    "score": 1.0,
+                                    "content": "representation of self-location and update it according to the animal’s movements and landmarks",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                609,
+                                505,
+                                623
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        609,
+                                        505,
+                                        623
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of the environment. Physiological studies done in rodents and other mammals (including humans,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                621,
+                                505,
+                                634
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        621,
+                                        505,
+                                        634
+                                    ],
+                                    "score": 1.0,
+                                    "content": "non-human primates and bats) have revealed a variety of neural correlates of space in Hippocampus",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                632,
+                                505,
+                                644
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        632,
+                                        505,
+                                        644
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and Entorhinal Cortex (EC), including place cells (O’Keefe, 1976), grid cells (Fyhn et al., 2004;",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                642,
+                                505,
+                                656
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        642,
+                                        505,
+                                        656
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Hafting et al., 2005; Fyhn et al., 2008; Yartsev et al., 2011; Killian et al., 2012; Jacobs et al., 2013),",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                654,
+                                506,
+                                667
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        654,
+                                        506,
+                                        667
+                                    ],
+                                    "score": 1.0,
+                                    "content": "along with border cells (Solstad et al., 2008), band-like cells (Krupic et al., 2012) and others (see",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                664,
+                                505,
+                                677
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        664,
+                                        505,
+                                        677
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Figure 1a). In particular, each grid cell only fires when the animal occupies a distinct set of physical",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                676,
+                                506,
+                                689
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        676,
+                                        506,
+                                        689
+                                    ],
+                                    "score": 1.0,
+                                    "content": "locations, and strikingly these locations lie on a lattice. The study of the neural underpinning of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                686,
+                                506,
+                                701
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        686,
+                                        506,
+                                        701
+                                    ],
+                                    "score": 1.0,
+                                    "content": "spatial cognition has provided an important window into how high-level cognitive functions are",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                698,
+                                362,
+                                711
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        698,
+                                        362,
+                                        711
+                                    ],
+                                    "score": 1.0,
+                                    "content": "supported in the brain (Moser et al., 2008; Aronov et al., 2017).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        }
+                    ],
+                    "index": 38
+                }
+            ],
+            "page_idx": 0,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        119,
+                        722,
+                        189,
+                        731
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                118,
+                                720,
+                                191,
+                                732
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        118,
+                                        720,
+                                        191,
+                                        732
+                                    ],
+                                    "score": 1.0,
+                                    "content": "∗equal contribution",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        107,
+                        27,
+                        293,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                25,
+                                294,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        25,
+                                        294,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Published as a conference paper at ICLR 2018",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        302,
+                        751,
+                        308,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                302,
+                                751,
+                                309,
+                                762
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        302,
+                                        751,
+                                        309,
+                                        762
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        79,
+                        503,
+                        136
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                79,
+                                506,
+                                100
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        79,
+                                        506,
+                                        100
+                                    ],
+                                    "score": 1.0,
+                                    "content": "EMERGENCE OF GRID-LIKE REPRESENTATIONS BY",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                102,
+                                453,
+                                118
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        102,
+                                        453,
+                                        118
+                                    ],
+                                    "score": 1.0,
+                                    "content": "TRAINING RECURRENT NEURAL NETWORKS TO",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                122,
+                                358,
+                                138
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        122,
+                                        358,
+                                        138
+                                    ],
+                                    "score": 1.0,
+                                    "content": "PERFORM SPATIAL LOCALIZATION",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        }
+                    ],
+                    "index": 1
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        112,
+                        154,
+                        273,
+                        199
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                113,
+                                155,
+                                269,
+                                167
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        113,
+                                        155,
+                                        269,
+                                        167
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Christopher J. Cueva∗, Xue-Xin Wei∗",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                111,
+                                164,
+                                199,
+                                178
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        111,
+                                        164,
+                                        199,
+                                        178
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Columbia University",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                111,
+                                176,
+                                227,
+                                188
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        111,
+                                        176,
+                                        227,
+                                        188
+                                    ],
+                                    "score": 1.0,
+                                    "content": "New York, NY 10027, USA",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                112,
+                                187,
+                                275,
+                                201
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        112,
+                                        187,
+                                        275,
+                                        201
+                                    ],
+                                    "score": 1.0,
+                                    "content": "{ccueva,weixxpku}@gmail.com",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        }
+                    ],
+                    "index": 4.5,
+                    "bbox_fs": [
+                        111,
+                        155,
+                        275,
+                        201
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        278,
+                        228,
+                        333,
+                        240
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                276,
+                                227,
+                                336,
+                                242
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        276,
+                                        227,
+                                        336,
+                                        242
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ABSTRACT",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        }
+                    ],
+                    "index": 7
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        143,
+                        255,
+                        469,
+                        419
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                141,
+                                254,
+                                469,
+                                268
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        254,
+                                        469,
+                                        268
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Decades of research on the neural code underlying spatial navigation have re-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                265,
+                                470,
+                                279
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        265,
+                                        470,
+                                        279
+                                    ],
+                                    "score": 1.0,
+                                    "content": "vealed a diverse set of neural response properties. The Entorhinal Cortex (EC) of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                142,
+                                277,
+                                469,
+                                289
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        142,
+                                        277,
+                                        469,
+                                        289
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the mammalian brain contains a rich set of spatial correlates, including grid cells",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                142,
+                                288,
+                                469,
+                                301
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        142,
+                                        288,
+                                        469,
+                                        301
+                                    ],
+                                    "score": 1.0,
+                                    "content": "which encode space using tessellating patterns. However, the mechanisms and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                299,
+                                469,
+                                311
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        299,
+                                        469,
+                                        311
+                                    ],
+                                    "score": 1.0,
+                                    "content": "functional significance of these spatial representations remain largely mysterious.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                310,
+                                470,
+                                323
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        310,
+                                        470,
+                                        323
+                                    ],
+                                    "score": 1.0,
+                                    "content": "As a new way to understand these neural representations, we trained recurrent",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                321,
+                                469,
+                                334
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        321,
+                                        469,
+                                        334
+                                    ],
+                                    "score": 1.0,
+                                    "content": "neural networks (RNNs) to perform navigation tasks in 2D arenas based on veloc-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                333,
+                                469,
+                                344
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        333,
+                                        469,
+                                        344
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ity inputs. Surprisingly, we find that grid-like spatial response patterns emerge in",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                343,
+                                469,
+                                355
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        343,
+                                        469,
+                                        355
+                                    ],
+                                    "score": 1.0,
+                                    "content": "trained networks, along with units that exhibit other spatial correlates, including",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                354,
+                                469,
+                                366
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        354,
+                                        469,
+                                        366
+                                    ],
+                                    "score": 1.0,
+                                    "content": "border cells and band-like cells. All these different functional types of neurons",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                365,
+                                470,
+                                378
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        365,
+                                        470,
+                                        378
+                                    ],
+                                    "score": 1.0,
+                                    "content": "have been observed experimentally. The order of the emergence of grid-like and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                142,
+                                376,
+                                469,
+                                387
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        142,
+                                        376,
+                                        469,
+                                        387
+                                    ],
+                                    "score": 1.0,
+                                    "content": "border cells is also consistent with observations from developmental studies. To-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                387,
+                                469,
+                                399
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        387,
+                                        469,
+                                        399
+                                    ],
+                                    "score": 1.0,
+                                    "content": "gether, our results suggest that grid cells, border cells and others as observed in EC",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                398,
+                                470,
+                                410
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        398,
+                                        470,
+                                        410
+                                    ],
+                                    "score": 1.0,
+                                    "content": "may be a natural solution for representing space efficiently given the predominant",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                409,
+                                317,
+                                420
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        409,
+                                        317,
+                                        420
+                                    ],
+                                    "score": 1.0,
+                                    "content": "recurrent connections in the neural circuits.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        }
+                    ],
+                    "index": 15,
+                    "bbox_fs": [
+                        141,
+                        254,
+                        470,
+                        420
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        446,
+                        206,
+                        459
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                444,
+                                208,
+                                461
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        444,
+                                        208,
+                                        461
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1 INTRODUCTION",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        }
+                    ],
+                    "index": 23
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        472,
+                        505,
+                        561
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                473,
+                                505,
+                                485
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        473,
+                                        505,
+                                        485
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Understanding the neural code in the brain has long been driven by studying feed-forward architec-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                484,
+                                505,
+                                496
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        484,
+                                        505,
+                                        496
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tures, starting from Hubel and Wiesel’s famous proposal on the origin of orientation selectivity in",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                495,
+                                505,
+                                506
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        495,
+                                        505,
+                                        506
+                                    ],
+                                    "score": 1.0,
+                                    "content": "primary visual cortex (Hubel & Wiesel, 1962). Inspired by the recent development in deep learn-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                505,
+                                505,
+                                518
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        505,
+                                        505,
+                                        518
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ing (Krizhevsky et al., 2012; LeCun et al., 2015; Hochreiter & Schmidhuber, 1997; Mnih et al.,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                516,
+                                505,
+                                529
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        516,
+                                        505,
+                                        529
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2015), there has been a burst of interest in applying deep feedforward models, in particular convolu-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                527,
+                                505,
+                                540
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        527,
+                                        505,
+                                        540
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tional neural networks (CNN) (LeCun et al., 1998), to study the sensory systems, which hierarchi-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                538,
+                                505,
+                                551
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        538,
+                                        505,
+                                        551
+                                    ],
+                                    "score": 1.0,
+                                    "content": "cally extract useful features from sensory inputs (see e.g., Yamins et al. (2014); Kriegeskorte (2015);",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                549,
+                                318,
+                                562
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        549,
+                                        318,
+                                        562
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Kietzmann et al. (2017); Yamins & DiCarlo (2016)).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        }
+                    ],
+                    "index": 27.5,
+                    "bbox_fs": [
+                        105,
+                        473,
+                        505,
+                        562
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        567,
+                        505,
+                        708
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                566,
+                                505,
+                                579
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        566,
+                                        505,
+                                        579
+                                    ],
+                                    "score": 1.0,
+                                    "content": "For more cognitive tasks, neural systems often need to maintain certain internal representations",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                577,
+                                505,
+                                589
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        577,
+                                        505,
+                                        589
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of relevant variables in the absence of external stimuli- a process that requires more than feature",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                589,
+                                506,
+                                600
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        589,
+                                        506,
+                                        600
+                                    ],
+                                    "score": 1.0,
+                                    "content": "extraction. We will focus on spatial navigation, which typically requires the brain to maintain a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                599,
+                                505,
+                                612
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        599,
+                                        505,
+                                        612
+                                    ],
+                                    "score": 1.0,
+                                    "content": "representation of self-location and update it according to the animal’s movements and landmarks",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                609,
+                                505,
+                                623
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        609,
+                                        505,
+                                        623
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of the environment. Physiological studies done in rodents and other mammals (including humans,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                621,
+                                505,
+                                634
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        621,
+                                        505,
+                                        634
+                                    ],
+                                    "score": 1.0,
+                                    "content": "non-human primates and bats) have revealed a variety of neural correlates of space in Hippocampus",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                632,
+                                505,
+                                644
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        632,
+                                        505,
+                                        644
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and Entorhinal Cortex (EC), including place cells (O’Keefe, 1976), grid cells (Fyhn et al., 2004;",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                642,
+                                505,
+                                656
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        642,
+                                        505,
+                                        656
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Hafting et al., 2005; Fyhn et al., 2008; Yartsev et al., 2011; Killian et al., 2012; Jacobs et al., 2013),",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                654,
+                                506,
+                                667
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        654,
+                                        506,
+                                        667
+                                    ],
+                                    "score": 1.0,
+                                    "content": "along with border cells (Solstad et al., 2008), band-like cells (Krupic et al., 2012) and others (see",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                664,
+                                505,
+                                677
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        664,
+                                        505,
+                                        677
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Figure 1a). In particular, each grid cell only fires when the animal occupies a distinct set of physical",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                676,
+                                506,
+                                689
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        676,
+                                        506,
+                                        689
+                                    ],
+                                    "score": 1.0,
+                                    "content": "locations, and strikingly these locations lie on a lattice. The study of the neural underpinning of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                686,
+                                506,
+                                701
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        686,
+                                        506,
+                                        701
+                                    ],
+                                    "score": 1.0,
+                                    "content": "spatial cognition has provided an important window into how high-level cognitive functions are",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                698,
+                                362,
+                                711
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        698,
+                                        362,
+                                        711
+                                    ],
+                                    "score": 1.0,
+                                    "content": "supported in the brain (Moser et al., 2008; Aronov et al., 2017).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        }
+                    ],
+                    "index": 38,
+                    "bbox_fs": [
+                        105,
+                        566,
+                        506,
+                        711
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        82,
+                        505,
+                        225
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                82,
+                                505,
+                                95
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        82,
+                                        505,
+                                        95
+                                    ],
+                                    "score": 1.0,
+                                    "content": "How might the spatial navigation task be solved using a network of neurons? Recurrent neural net-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                93,
+                                505,
+                                106
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        93,
+                                        505,
+                                        106
+                                    ],
+                                    "score": 1.0,
+                                    "content": "works (RNNs) (Hochreiter & Schmidhuber, 1997; Graves et al., 2013; Oord et al., 2016; Theis &",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                104,
+                                505,
+                                117
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        104,
+                                        505,
+                                        117
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Bethge, 2015; Gregor et al., 2015; Sussillo et al., 2015) seem particularly useful for these tasks.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                114,
+                                505,
+                                128
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        114,
+                                        505,
+                                        128
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Indeed, recurrent-based continuous attractor networks have been one popular type of models pro-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                127,
+                                505,
+                                140
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        127,
+                                        505,
+                                        140
+                                    ],
+                                    "score": 1.0,
+                                    "content": "posed for the formation of grid cells (McNaughton et al., 2006; Burak & Fiete, 2009; Couey et al.,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                138,
+                                505,
+                                149
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        138,
+                                        505,
+                                        149
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2013) and place cells (Samsonovich & McNaughton, 1997). Such models have provided valuable",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                149,
+                                506,
+                                162
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        149,
+                                        506,
+                                        162
+                                    ],
+                                    "score": 1.0,
+                                    "content": "insights into one set of possible mechanisms that could support the formation of the grids. How-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                159,
+                                506,
+                                172
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        159,
+                                        506,
+                                        172
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ever, these models typically rely on fine-tuned connectivity patterns, in particular the models need a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                169,
+                                506,
+                                185
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        169,
+                                        506,
+                                        185
+                                    ],
+                                    "score": 1.0,
+                                    "content": "subtle yet systematic asymmetry in the connectivity pattern to move the attractor state according to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                180,
+                                505,
+                                194
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        180,
+                                        505,
+                                        194
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the animal’s own movement. The existence of such a specific 2D connectivity in rodent EC remains",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                192,
+                                505,
+                                205
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        192,
+                                        505,
+                                        205
+                                    ],
+                                    "score": 1.0,
+                                    "content": "unclear. Additionally, previous models have mainly focused on grid cells, while other types of re-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                203,
+                                506,
+                                216
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        203,
+                                        506,
+                                        216
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sponses that co-exist in the Entorhinal Cortex have been largely ignored. It would be useful to have",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                214,
+                                466,
+                                227
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        214,
+                                        466,
+                                        227
+                                    ],
+                                    "score": 1.0,
+                                    "content": "a unified model that can simultaneously explain different types of neural responses in EC.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        }
+                    ],
+                    "index": 6
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        231,
+                        505,
+                        341
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                230,
+                                505,
+                                243
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        230,
+                                        505,
+                                        243
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Motivated by these considerations, here we present an alternative modeling approach for under-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                242,
+                                505,
+                                254
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        242,
+                                        505,
+                                        254
+                                    ],
+                                    "score": 1.0,
+                                    "content": "standing the representation of space in the neural system. Specifically, we trained a RNN to perform",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                253,
+                                505,
+                                266
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        253,
+                                        505,
+                                        266
+                                    ],
+                                    "score": 1.0,
+                                    "content": "some spatial navigation tasks. By leveraging the recent development in RNN training and knowl-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                264,
+                                505,
+                                277
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        264,
+                                        505,
+                                        277
+                                    ],
+                                    "score": 1.0,
+                                    "content": "edge of the navigation system in the brain, we show that training a RNN with biologically relevant",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                274,
+                                505,
+                                289
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        274,
+                                        505,
+                                        289
+                                    ],
+                                    "score": 1.0,
+                                    "content": "constraints naturally gives rise to a variety of spatial response profiles as observed in EC, including",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                286,
+                                505,
+                                298
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        286,
+                                        505,
+                                        298
+                                    ],
+                                    "score": 1.0,
+                                    "content": "grid-like responses. To our knowledge, this is the first study to show that grid-like responses could",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                297,
+                                505,
+                                309
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        297,
+                                        505,
+                                        309
+                                    ],
+                                    "score": 1.0,
+                                    "content": "emerge from training a RNN to perform navigation. Our result implies that the neural representation",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                307,
+                                505,
+                                320
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        307,
+                                        505,
+                                        320
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in EC may be seen as a natural way for the brain to solve the navigation task efficiently (Wei et al.,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                317,
+                                506,
+                                332
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        317,
+                                        506,
+                                        332
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2015). More generally, it suggests that RNNs can be a powerful tool for understanding the neural",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                329,
+                                322,
+                                342
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        329,
+                                        322,
+                                        342
+                                    ],
+                                    "score": 1.0,
+                                    "content": "mechanisms of certain high-level cognitive functions.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        }
+                    ],
+                    "index": 17.5
+                },
+                {
+                    "type": "image",
+                    "bbox": [
+                        146,
+                        373,
+                        460,
+                        540
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                146,
+                                373,
+                                460,
+                                540
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        146,
+                                        373,
+                                        460,
+                                        540
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                146,
+                                                373,
+                                                460,
+                                                540
+                                            ],
+                                            "score": 0.976,
+                                            "type": "image",
+                                            "image_path": "8e6f9cf7feba198b2a18d2ed284d372bf5324606db8951624994c943f119c937.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 24,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        146,
+                                        373,
+                                        460,
+                                        428.6666666666667
+                                    ],
+                                    "spans": [],
+                                    "index": 23
+                                },
+                                {
+                                    "bbox": [
+                                        146,
+                                        428.6666666666667,
+                                        460,
+                                        484.33333333333337
+                                    ],
+                                    "spans": [],
+                                    "index": 24
+                                },
+                                {
+                                    "bbox": [
+                                        146,
+                                        484.33333333333337,
+                                        460,
+                                        540.0
+                                    ],
+                                    "spans": [],
+                                    "index": 25
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                106,
+                                550,
+                                505,
+                                705
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        550,
+                                        505,
+                                        563
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                550,
+                                                505,
+                                                563
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 1: a) Example neural data showing different kinds of neural correlates underlying spatial nav-",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 26
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        561,
+                                        506,
+                                        574
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                561,
+                                                506,
+                                                574
+                                            ],
+                                            "score": 1.0,
+                                            "content": "igation in EC. All figures are replotted from previous publications. From left to right: a “grid cell”",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 27
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        572,
+                                        505,
+                                        585
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                572,
+                                                505,
+                                                585
+                                            ],
+                                            "score": 1.0,
+                                            "content": "recorded when an animal navigates in a square environment, replotted from Krupic et al. (2012),",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 28
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        583,
+                                        505,
+                                        596
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                583,
+                                                505,
+                                                596
+                                            ],
+                                            "score": 1.0,
+                                            "content": "with the heat map representing the firing rate of this neuron as a function of the animal’s location",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 29
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        594,
+                                        505,
+                                        607
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                594,
+                                                505,
+                                                607
+                                            ],
+                                            "score": 1.0,
+                                            "content": "(red corresponds to high firing rate); a “band-like” cell from Krupic et al. (2012); a border cell from",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 30
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        604,
+                                        506,
+                                        618
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                604,
+                                                506,
+                                                618
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Solstad et al. (2008); an irregular spatially tuned cell from Diehl et al. (2017); a “speed cell” from",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 31
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        615,
+                                        506,
+                                        629
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                615,
+                                                506,
+                                                629
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Kropff et al. (2015), which exhibits roughly linear dependence on the rodent’s running speed; a",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 32
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        626,
+                                        505,
+                                        640
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                626,
+                                                505,
+                                                640
+                                            ],
+                                            "score": 1.0,
+                                            "content": "“heading direction cell” from Sargolini et al. (2006), which shows systematic change of firing rate",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 33
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        638,
+                                        505,
+                                        651
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                638,
+                                                394,
+                                                651
+                                            ],
+                                            "score": 1.0,
+                                            "content": "depending on animal’s heading direction. b) The network consists of",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                394,
+                                                638,
+                                                436,
+                                                648
+                                            ],
+                                            "score": 0.9,
+                                            "content": "N = 1 0 0",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                437,
+                                                638,
+                                                505,
+                                                651
+                                            ],
+                                            "score": 1.0,
+                                            "content": "recurrently con-",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 34
+                                },
+                                {
+                                    "bbox": [
+                                        104,
+                                        649,
+                                        505,
+                                        661
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                104,
+                                                649,
+                                                505,
+                                                661
+                                            ],
+                                            "score": 1.0,
+                                            "content": "nected units (or neurons) which receive two external inputs, representing the animal’s speed and",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 35
+                                },
+                                {
+                                    "bbox": [
+                                        104,
+                                        659,
+                                        506,
+                                        674
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                104,
+                                                659,
+                                                506,
+                                                674
+                                            ],
+                                            "score": 1.0,
+                                            "content": "heading direction. The two outputs linearly weight the neurons in the RNN. The goal of training is",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 36
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        671,
+                                        505,
+                                        683
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                671,
+                                                505,
+                                                683
+                                            ],
+                                            "score": 1.0,
+                                            "content": "to make the responses of the two output neurons accurately represent the animal’s physical location.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 37
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        682,
+                                        506,
+                                        695
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                682,
+                                                506,
+                                                695
+                                            ],
+                                            "score": 1.0,
+                                            "content": "c) Typical trajectory after training. As shown, the output of the RNN can accurately, though not",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 38
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        694,
+                                        329,
+                                        705
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                694,
+                                                329,
+                                                705
+                                            ],
+                                            "score": 1.0,
+                                            "content": "perfectly, track the animal’s location during navigation.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 39
+                                }
+                            ],
+                            "index": 32.5
+                        }
+                    ],
+                    "index": 28.25
+                }
+            ],
+            "page_idx": 1,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        108,
+                        27,
+                        293,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                26,
+                                294,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        26,
+                                        294,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Published as a conference paper at ICLR 2018",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        302,
+                        751,
+                        308,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                302,
+                                750,
+                                310,
+                                763
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        302,
+                                        750,
+                                        310,
+                                        763
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        82,
+                        505,
+                        225
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                82,
+                                505,
+                                95
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        82,
+                                        505,
+                                        95
+                                    ],
+                                    "score": 1.0,
+                                    "content": "How might the spatial navigation task be solved using a network of neurons? Recurrent neural net-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                93,
+                                505,
+                                106
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        93,
+                                        505,
+                                        106
+                                    ],
+                                    "score": 1.0,
+                                    "content": "works (RNNs) (Hochreiter & Schmidhuber, 1997; Graves et al., 2013; Oord et al., 2016; Theis &",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                104,
+                                505,
+                                117
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        104,
+                                        505,
+                                        117
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Bethge, 2015; Gregor et al., 2015; Sussillo et al., 2015) seem particularly useful for these tasks.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                114,
+                                505,
+                                128
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        114,
+                                        505,
+                                        128
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Indeed, recurrent-based continuous attractor networks have been one popular type of models pro-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                127,
+                                505,
+                                140
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        127,
+                                        505,
+                                        140
+                                    ],
+                                    "score": 1.0,
+                                    "content": "posed for the formation of grid cells (McNaughton et al., 2006; Burak & Fiete, 2009; Couey et al.,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                138,
+                                505,
+                                149
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        138,
+                                        505,
+                                        149
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2013) and place cells (Samsonovich & McNaughton, 1997). Such models have provided valuable",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                149,
+                                506,
+                                162
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        149,
+                                        506,
+                                        162
+                                    ],
+                                    "score": 1.0,
+                                    "content": "insights into one set of possible mechanisms that could support the formation of the grids. How-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                159,
+                                506,
+                                172
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        159,
+                                        506,
+                                        172
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ever, these models typically rely on fine-tuned connectivity patterns, in particular the models need a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                169,
+                                506,
+                                185
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        169,
+                                        506,
+                                        185
+                                    ],
+                                    "score": 1.0,
+                                    "content": "subtle yet systematic asymmetry in the connectivity pattern to move the attractor state according to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                180,
+                                505,
+                                194
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        180,
+                                        505,
+                                        194
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the animal’s own movement. The existence of such a specific 2D connectivity in rodent EC remains",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                192,
+                                505,
+                                205
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        192,
+                                        505,
+                                        205
+                                    ],
+                                    "score": 1.0,
+                                    "content": "unclear. Additionally, previous models have mainly focused on grid cells, while other types of re-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                203,
+                                506,
+                                216
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        203,
+                                        506,
+                                        216
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sponses that co-exist in the Entorhinal Cortex have been largely ignored. It would be useful to have",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                214,
+                                466,
+                                227
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        214,
+                                        466,
+                                        227
+                                    ],
+                                    "score": 1.0,
+                                    "content": "a unified model that can simultaneously explain different types of neural responses in EC.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        }
+                    ],
+                    "index": 6,
+                    "bbox_fs": [
+                        105,
+                        82,
+                        506,
+                        227
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        231,
+                        505,
+                        341
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                230,
+                                505,
+                                243
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        230,
+                                        505,
+                                        243
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Motivated by these considerations, here we present an alternative modeling approach for under-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                242,
+                                505,
+                                254
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        242,
+                                        505,
+                                        254
+                                    ],
+                                    "score": 1.0,
+                                    "content": "standing the representation of space in the neural system. Specifically, we trained a RNN to perform",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                253,
+                                505,
+                                266
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        253,
+                                        505,
+                                        266
+                                    ],
+                                    "score": 1.0,
+                                    "content": "some spatial navigation tasks. By leveraging the recent development in RNN training and knowl-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                264,
+                                505,
+                                277
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        264,
+                                        505,
+                                        277
+                                    ],
+                                    "score": 1.0,
+                                    "content": "edge of the navigation system in the brain, we show that training a RNN with biologically relevant",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                274,
+                                505,
+                                289
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        274,
+                                        505,
+                                        289
+                                    ],
+                                    "score": 1.0,
+                                    "content": "constraints naturally gives rise to a variety of spatial response profiles as observed in EC, including",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                286,
+                                505,
+                                298
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        286,
+                                        505,
+                                        298
+                                    ],
+                                    "score": 1.0,
+                                    "content": "grid-like responses. To our knowledge, this is the first study to show that grid-like responses could",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                297,
+                                505,
+                                309
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        297,
+                                        505,
+                                        309
+                                    ],
+                                    "score": 1.0,
+                                    "content": "emerge from training a RNN to perform navigation. Our result implies that the neural representation",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                307,
+                                505,
+                                320
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        307,
+                                        505,
+                                        320
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in EC may be seen as a natural way for the brain to solve the navigation task efficiently (Wei et al.,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                317,
+                                506,
+                                332
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        317,
+                                        506,
+                                        332
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2015). More generally, it suggests that RNNs can be a powerful tool for understanding the neural",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                329,
+                                322,
+                                342
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        329,
+                                        322,
+                                        342
+                                    ],
+                                    "score": 1.0,
+                                    "content": "mechanisms of certain high-level cognitive functions.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        }
+                    ],
+                    "index": 17.5,
+                    "bbox_fs": [
+                        105,
+                        230,
+                        506,
+                        342
+                    ]
+                },
+                {
+                    "type": "image",
+                    "bbox": [
+                        146,
+                        373,
+                        460,
+                        540
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                146,
+                                373,
+                                460,
+                                540
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        146,
+                                        373,
+                                        460,
+                                        540
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                146,
+                                                373,
+                                                460,
+                                                540
+                                            ],
+                                            "score": 0.976,
+                                            "type": "image",
+                                            "image_path": "8e6f9cf7feba198b2a18d2ed284d372bf5324606db8951624994c943f119c937.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 24,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        146,
+                                        373,
+                                        460,
+                                        428.6666666666667
+                                    ],
+                                    "spans": [],
+                                    "index": 23
+                                },
+                                {
+                                    "bbox": [
+                                        146,
+                                        428.6666666666667,
+                                        460,
+                                        484.33333333333337
+                                    ],
+                                    "spans": [],
+                                    "index": 24
+                                },
+                                {
+                                    "bbox": [
+                                        146,
+                                        484.33333333333337,
+                                        460,
+                                        540.0
+                                    ],
+                                    "spans": [],
+                                    "index": 25
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                106,
+                                550,
+                                505,
+                                705
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        550,
+                                        505,
+                                        563
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                550,
+                                                505,
+                                                563
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 1: a) Example neural data showing different kinds of neural correlates underlying spatial nav-",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 26
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        561,
+                                        506,
+                                        574
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                561,
+                                                506,
+                                                574
+                                            ],
+                                            "score": 1.0,
+                                            "content": "igation in EC. All figures are replotted from previous publications. From left to right: a “grid cell”",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 27
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        572,
+                                        505,
+                                        585
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                572,
+                                                505,
+                                                585
+                                            ],
+                                            "score": 1.0,
+                                            "content": "recorded when an animal navigates in a square environment, replotted from Krupic et al. (2012),",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 28
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        583,
+                                        505,
+                                        596
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                583,
+                                                505,
+                                                596
+                                            ],
+                                            "score": 1.0,
+                                            "content": "with the heat map representing the firing rate of this neuron as a function of the animal’s location",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 29
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        594,
+                                        505,
+                                        607
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                594,
+                                                505,
+                                                607
+                                            ],
+                                            "score": 1.0,
+                                            "content": "(red corresponds to high firing rate); a “band-like” cell from Krupic et al. (2012); a border cell from",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 30
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        604,
+                                        506,
+                                        618
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                604,
+                                                506,
+                                                618
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Solstad et al. (2008); an irregular spatially tuned cell from Diehl et al. (2017); a “speed cell” from",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 31
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        615,
+                                        506,
+                                        629
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                615,
+                                                506,
+                                                629
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Kropff et al. (2015), which exhibits roughly linear dependence on the rodent’s running speed; a",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 32
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        626,
+                                        505,
+                                        640
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                626,
+                                                505,
+                                                640
+                                            ],
+                                            "score": 1.0,
+                                            "content": "“heading direction cell” from Sargolini et al. (2006), which shows systematic change of firing rate",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 33
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        638,
+                                        505,
+                                        651
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                638,
+                                                394,
+                                                651
+                                            ],
+                                            "score": 1.0,
+                                            "content": "depending on animal’s heading direction. b) The network consists of",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                394,
+                                                638,
+                                                436,
+                                                648
+                                            ],
+                                            "score": 0.9,
+                                            "content": "N = 1 0 0",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                437,
+                                                638,
+                                                505,
+                                                651
+                                            ],
+                                            "score": 1.0,
+                                            "content": "recurrently con-",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 34
+                                },
+                                {
+                                    "bbox": [
+                                        104,
+                                        649,
+                                        505,
+                                        661
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                104,
+                                                649,
+                                                505,
+                                                661
+                                            ],
+                                            "score": 1.0,
+                                            "content": "nected units (or neurons) which receive two external inputs, representing the animal’s speed and",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 35
+                                },
+                                {
+                                    "bbox": [
+                                        104,
+                                        659,
+                                        506,
+                                        674
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                104,
+                                                659,
+                                                506,
+                                                674
+                                            ],
+                                            "score": 1.0,
+                                            "content": "heading direction. The two outputs linearly weight the neurons in the RNN. The goal of training is",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 36
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        671,
+                                        505,
+                                        683
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                671,
+                                                505,
+                                                683
+                                            ],
+                                            "score": 1.0,
+                                            "content": "to make the responses of the two output neurons accurately represent the animal’s physical location.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 37
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        682,
+                                        506,
+                                        695
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                682,
+                                                506,
+                                                695
+                                            ],
+                                            "score": 1.0,
+                                            "content": "c) Typical trajectory after training. As shown, the output of the RNN can accurately, though not",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 38
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        694,
+                                        329,
+                                        705
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                694,
+                                                329,
+                                                705
+                                            ],
+                                            "score": 1.0,
+                                            "content": "perfectly, track the animal’s location during navigation.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 39
+                                }
+                            ],
+                            "index": 32.5
+                        }
+                    ],
+                    "index": 28.25
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        81,
+                        165,
+                        94
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                78,
+                                168,
+                                96
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        78,
+                                        168,
+                                        96
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2 MODEL",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        }
+                    ],
+                    "index": 0
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        105,
+                        224,
+                        117
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                105,
+                                225,
+                                118
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        105,
+                                        225,
+                                        118
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2.1 MODEL DESCRIPTION",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        }
+                    ],
+                    "index": 1
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        126,
+                        506,
+                        149
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                126,
+                                506,
+                                139
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        126,
+                                        383,
+                                        139
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Our network model consists of a set of recurrently connected units",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        384,
+                                        127,
+                                        425,
+                                        137
+                                    ],
+                                    "score": 0.86,
+                                    "content": "N = 1 0 0",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        425,
+                                        126,
+                                        506,
+                                        139
+                                    ],
+                                    "score": 1.0,
+                                    "content": "). The dynamics of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                137,
+                                470,
+                                150
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        137,
+                                        205,
+                                        150
+                                    ],
+                                    "score": 1.0,
+                                    "content": "each unit in the network",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        206,
+                                        137,
+                                        226,
+                                        149
+                                    ],
+                                    "score": 0.93,
+                                    "content": "u _ { i } ( t )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        227,
+                                        137,
+                                        470,
+                                        150
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is governed by the standard continuous-time RNN equation:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        }
+                    ],
+                    "index": 2.5
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        172,
+                        151,
+                        439,
+                        187
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                172,
+                                151,
+                                439,
+                                187
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        172,
+                                        151,
+                                        439,
+                                        187
+                                    ],
+                                    "score": 0.95,
+                                    "content": "\\tau \\frac { d x _ { i } ( t ) } { d t } = - x _ { i } ( t ) + \\sum _ { j = 1 } ^ { N } W _ { i j } ^ { \\mathrm { r e c } } u _ { j } ( t ) + \\sum _ { k = 1 } ^ { N _ { \\mathrm { i n } } } W _ { i k } ^ { \\mathrm { i n } } I _ { k } ( t ) + b _ { i } + \\xi _ { i } ( t )",
+                                    "type": "interline_equation",
+                                    "image_path": "73f5fb1e1639b7bb2d07858a8bd6b7b8e2b6dc0350eb54f951fc46c74a662eee.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 5,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                172,
+                                151,
+                                439,
+                                163.0
+                            ],
+                            "spans": [],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                172,
+                                163.0,
+                                439,
+                                175.0
+                            ],
+                            "spans": [],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                172,
+                                175.0,
+                                439,
+                                187.0
+                            ],
+                            "spans": [],
+                            "index": 6
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        190,
+                        505,
+                        268
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                189,
+                                505,
+                                204
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        189,
+                                        121,
+                                        204
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        122,
+                                        191,
+                                        178,
+                                        202
+                                    ],
+                                    "score": 0.91,
+                                    "content": "i = 1 , \\ldots , N",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        179,
+                                        189,
+                                        292,
+                                        204
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". The activity of each unit,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        293,
+                                        190,
+                                        314,
+                                        203
+                                    ],
+                                    "score": 0.92,
+                                    "content": "u _ { i } ( t )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        314,
+                                        189,
+                                        479,
+                                        204
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", is related to the activation of that unit,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        480,
+                                        190,
+                                        501,
+                                        203
+                                    ],
+                                    "score": 0.91,
+                                    "content": "x _ { i } ( t )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        501,
+                                        189,
+                                        505,
+                                        204
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                201,
+                                506,
+                                214
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        201,
+                                        337,
+                                        214
+                                    ],
+                                    "score": 1.0,
+                                    "content": "through a nonlinearity which in this study we take to be",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        338,
+                                        201,
+                                        422,
+                                        213
+                                    ],
+                                    "score": 0.92,
+                                    "content": "u _ { i } ( t ) = \\operatorname { t a n h } ( x _ { i } ( t ) )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        423,
+                                        201,
+                                        506,
+                                        214
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Each unit receives",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                211,
+                                505,
+                                226
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        211,
+                                        347,
+                                        226
+                                    ],
+                                    "score": 1.0,
+                                    "content": "input from other units through the recurrent weight matrix",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        348,
+                                        213,
+                                        370,
+                                        222
+                                    ],
+                                    "score": 0.85,
+                                    "content": "W ^ { \\mathrm { r e c } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        370,
+                                        211,
+                                        505,
+                                        226
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and also receives external input,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                222,
+                                506,
+                                237
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        224,
+                                        124,
+                                        235
+                                    ],
+                                    "score": 0.89,
+                                    "content": "I ( t )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        124,
+                                        222,
+                                        335,
+                                        237
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", that enters the network through the weight matrix",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        336,
+                                        223,
+                                        355,
+                                        234
+                                    ],
+                                    "score": 0.88,
+                                    "content": "W ^ { \\mathrm { i n } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        355,
+                                        222,
+                                        506,
+                                        237
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Each unit has two sources of bias,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                235,
+                                505,
+                                246
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        235,
+                                        115,
+                                        245
+                                    ],
+                                    "score": 0.85,
+                                    "content": "b _ { i }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        116,
+                                        235,
+                                        207,
+                                        246
+                                    ],
+                                    "score": 1.0,
+                                    "content": "which is learned and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        208,
+                                        235,
+                                        228,
+                                        246
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\xi _ { i } ( t )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        228,
+                                        235,
+                                        505,
+                                        246
+                                    ],
+                                    "score": 1.0,
+                                    "content": "which represents noise intrinsic to the network and is taken to be",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                245,
+                                505,
+                                257
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        245,
+                                        505,
+                                        257
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Gaussian with zero mean and constant variance. The network was simulated using the Euler method",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                255,
+                                270,
+                                268
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        255,
+                                        120,
+                                        268
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        121,
+                                        257,
+                                        158,
+                                        267
+                                    ],
+                                    "score": 0.89,
+                                    "content": "T = 5 0 0",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        158,
+                                        255,
+                                        245,
+                                        268
+                                    ],
+                                    "score": 1.0,
+                                    "content": "timesteps of duration",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        245,
+                                        256,
+                                        267,
+                                        268
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\tau / 1 0",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        267,
+                                        255,
+                                        270,
+                                        268
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        }
+                    ],
+                    "index": 10
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        273,
+                        505,
+                        307
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                273,
+                                505,
+                                285
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        273,
+                                        505,
+                                        285
+                                    ],
+                                    "score": 1.0,
+                                    "content": "To perform a 2D navigation task with the RNN, we linearly combine the firing rates of units in",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                285,
+                                505,
+                                296
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        285,
+                                        505,
+                                        296
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the network to estimate the current location of the animal. The responses of the two linear readout",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                295,
+                                355,
+                                308
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        295,
+                                        143,
+                                        308
+                                    ],
+                                    "score": 1.0,
+                                    "content": "neurons,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        143,
+                                        295,
+                                        165,
+                                        307
+                                    ],
+                                    "score": 0.92,
+                                    "content": "y _ { 1 } ( t )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        165,
+                                        295,
+                                        183,
+                                        308
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        183,
+                                        295,
+                                        205,
+                                        307
+                                    ],
+                                    "score": 0.92,
+                                    "content": "y _ { 2 } ( t )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        205,
+                                        295,
+                                        355,
+                                        308
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", are given by the following equation:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        }
+                    ],
+                    "index": 15
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        257,
+                        309,
+                        353,
+                        344
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                257,
+                                309,
+                                353,
+                                344
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        257,
+                                        309,
+                                        353,
+                                        344
+                                    ],
+                                    "score": 0.94,
+                                    "content": "y _ { j } ( t ) = \\sum _ { i = 1 } ^ { N } W _ { j i } ^ { \\mathrm { o u t } } u _ { i } ( t )",
+                                    "type": "interline_equation",
+                                    "image_path": "d64453343a45b8133e59ab95038edf8c7ca7e2bd28bd901cacdd4eff0e7578b4.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 17.5,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                257,
+                                309,
+                                353,
+                                326.5
+                            ],
+                            "spans": [],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                257,
+                                326.5,
+                                353,
+                                344.0
+                            ],
+                            "spans": [],
+                            "index": 18
+                        }
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        356,
+                        237,
+                        367
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                355,
+                                239,
+                                368
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        355,
+                                        239,
+                                        368
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2.2 INPUT TO THE NETWORK",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        }
+                    ],
+                    "index": 19
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        375,
+                        505,
+                        453
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                376,
+                                505,
+                                389
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        376,
+                                        505,
+                                        389
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The network inputs and outputs were inspired by simple spatial navigation tasks in 2D open envi-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                387,
+                                506,
+                                400
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        387,
+                                        506,
+                                        400
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ronments. The task resembles dead-reckoning (sometimes referred to as path integration), which is",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                398,
+                                505,
+                                410
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        398,
+                                        505,
+                                        410
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ethologically relevant for many animal species (Darwin, 1873; Mittelstaedt & Mittelstaedt, 1980;",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                408,
+                                506,
+                                421
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        408,
+                                        506,
+                                        421
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Etienne & Jeffery, 2004; McNaughton et al., 2006). To be more specific, the inputs to the network",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                420,
+                                505,
+                                432
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        420,
+                                        505,
+                                        432
+                                    ],
+                                    "score": 1.0,
+                                    "content": "were the animal’s speed and direction at each time step. Experimentally, it has been shown that the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                431,
+                                506,
+                                443
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        431,
+                                        506,
+                                        443
+                                    ],
+                                    "score": 1.0,
+                                    "content": "velocity signals exist in EC (Sargolini et al., 2006; Kropff et al., 2015; Hinman et al., 2016), and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                441,
+                                490,
+                                454
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        441,
+                                        490,
+                                        454
+                                    ],
+                                    "score": 1.0,
+                                    "content": "there is also evidence that such signals are necessary for grid formation (Winter et al., 2015a;b).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        }
+                    ],
+                    "index": 23
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        458,
+                        505,
+                        536
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                458,
+                                505,
+                                471
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        458,
+                                        505,
+                                        471
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Throughout the paper, we adopt the common assumption that the head direction of the animal coin-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                470,
+                                505,
+                                482
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        470,
+                                        505,
+                                        482
+                                    ],
+                                    "score": 1.0,
+                                    "content": "cides with the actual moving direction. The outputs were the x- and y-coordinates of the integrated",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                480,
+                                506,
+                                493
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        480,
+                                        506,
+                                        493
+                                    ],
+                                    "score": 1.0,
+                                    "content": "position. The direction of the animal is modeled by modified Brownian motion to increase the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                492,
+                                506,
+                                504
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        492,
+                                        506,
+                                        504
+                                    ],
+                                    "score": 1.0,
+                                    "content": "probability of straight-runs, in order to be consistent with the typical rodent’s behavior in an open",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                502,
+                                506,
+                                515
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        502,
+                                        506,
+                                        515
+                                    ],
+                                    "score": 1.0,
+                                    "content": "environment. The usage of such simple movement statistics has the advantage of having full control",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                514,
+                                505,
+                                526
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        514,
+                                        505,
+                                        526
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of the simulated trajectories. However, for future work it would be very interesting to test the model",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                524,
+                                459,
+                                538
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        524,
+                                        459,
+                                        538
+                                    ],
+                                    "score": 1.0,
+                                    "content": "using different animals’ real movement trajectories to see how the results might change.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        }
+                    ],
+                    "index": 30
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        541,
+                        505,
+                        597
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                542,
+                                505,
+                                554
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        542,
+                                        505,
+                                        554
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Special care is taken when the animal is close to the boundary. The boundary of the environment",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                551,
+                                505,
+                                566
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        551,
+                                        505,
+                                        566
+                                    ],
+                                    "score": 1.0,
+                                    "content": "will affect the statistics of the movement, as the animal cannot cross the boundary. This fact was",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                564,
+                                505,
+                                576
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        564,
+                                        505,
+                                        576
+                                    ],
+                                    "score": 1.0,
+                                    "content": "reflected in the model by re-sampling the angular input variable until the input angle did not lead",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                574,
+                                505,
+                                586
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        574,
+                                        505,
+                                        586
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the animal outside the boundary. In the simulations shown below, the animal always starts from the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                585,
+                                465,
+                                598
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        585,
+                                        465,
+                                        598
+                                    ],
+                                    "score": 1.0,
+                                    "content": "center of the arena, but we verified that the results are insensitive to the starting locations.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        }
+                    ],
+                    "index": 36
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        609,
+                        176,
+                        621
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                608,
+                                177,
+                                623
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        608,
+                                        177,
+                                        623
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2.3 TRAINING",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        }
+                    ],
+                    "index": 39
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        629,
+                        505,
+                        674
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                628,
+                                506,
+                                644
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        628,
+                                        265,
+                                        644
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We optimized the network parameters",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        266,
+                                        630,
+                                        313,
+                                        641
+                                    ],
+                                    "score": 0.27,
+                                    "content": "W ^ { \\mathrm { r e c } } , W ^ { \\mathrm { i n } } ,",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        313,
+                                        628,
+                                        317,
+                                        644
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        317,
+                                        631,
+                                        323,
+                                        640
+                                    ],
+                                    "score": 0.69,
+                                    "content": "b",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        324,
+                                        628,
+                                        343,
+                                        644
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        343,
+                                        630,
+                                        367,
+                                        641
+                                    ],
+                                    "score": 0.87,
+                                    "content": "W ^ { \\mathrm { o u t } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        367,
+                                        628,
+                                        506,
+                                        644
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to minimize the squared error in",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                641,
+                                506,
+                                654
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        641,
+                                        216,
+                                        654
+                                    ],
+                                    "score": 1.0,
+                                    "content": "equation (3) between target",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        217,
+                                        643,
+                                        226,
+                                        652
+                                    ],
+                                    "score": 0.36,
+                                    "content": "\\mathbf { X } ^ { - }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        226,
+                                        641,
+                                        243,
+                                        654
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        244,
+                                        644,
+                                        250,
+                                        653
+                                    ],
+                                    "score": 0.32,
+                                    "content": "_ \\textrm { y }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        250,
+                                        641,
+                                        506,
+                                        654
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-coordinates from a two dimensional navigation task (performed",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                653,
+                                505,
+                                666
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        653,
+                                        505,
+                                        666
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in rectangular, hexagonal, and triangular arenas) and the network outputs generated according to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                663,
+                                160,
+                                677
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        663,
+                                        160,
+                                        677
+                                    ],
+                                    "score": 1.0,
+                                    "content": "equation (2).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        }
+                    ],
+                    "index": 41.5
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        200,
+                        673,
+                        410,
+                        709
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                200,
+                                673,
+                                410,
+                                709
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        200,
+                                        673,
+                                        410,
+                                        709
+                                    ],
+                                    "score": 0.93,
+                                    "content": "E = \\frac { 1 } { M T N _ { \\mathrm { o u t } } } \\sum _ { m , t , j = 1 } ^ { M , T , N _ { \\mathrm { o u t } } } ( y _ { j } ( t , m ) - y _ { j } ^ { \\mathrm { t a r g e t } } ( t , m ) ) ^ { 2 }",
+                                    "type": "interline_equation",
+                                    "image_path": "c0586e3c42e764d474fb5fc1894e2ed9f63e5c96ba0d3cd1f2199ef9a9b3a013.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 44.5,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                200,
+                                673,
+                                410,
+                                691.0
+                            ],
+                            "spans": [],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                200,
+                                691.0,
+                                410,
+                                709.0
+                            ],
+                            "spans": [],
+                            "index": 45
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        709,
+                        504,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                709,
+                                505,
+                                722
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        709,
+                                        505,
+                                        722
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Parameters were updated with the Hessian-free algorithm (Martens & Sutskever, 2011) using mini-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                718,
+                                506,
+                                734
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        718,
+                                        170,
+                                        734
+                                    ],
+                                    "score": 1.0,
+                                    "content": "batches of size",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        171,
+                                        721,
+                                        214,
+                                        731
+                                    ],
+                                    "score": 0.9,
+                                    "content": "M = 5 0 0",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        214,
+                                        718,
+                                        506,
+                                        734
+                                    ],
+                                    "score": 1.0,
+                                    "content": "trials. In addition to minimizing the error function in equation (3) we",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 47
+                        }
+                    ],
+                    "index": 46.5
+                }
+            ],
+            "page_idx": 2,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        107,
+                        26,
+                        293,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                25,
+                                294,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        25,
+                                        294,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Published as a conference paper at ICLR 2018",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        302,
+                        751,
+                        309,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                301,
+                                750,
+                                310,
+                                762
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        301,
+                                        750,
+                                        310,
+                                        762
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        81,
+                        165,
+                        94
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                78,
+                                168,
+                                96
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        78,
+                                        168,
+                                        96
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2 MODEL",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        }
+                    ],
+                    "index": 0
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        105,
+                        224,
+                        117
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                105,
+                                225,
+                                118
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        105,
+                                        225,
+                                        118
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2.1 MODEL DESCRIPTION",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        }
+                    ],
+                    "index": 1
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        126,
+                        506,
+                        149
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                126,
+                                506,
+                                139
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        126,
+                                        383,
+                                        139
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Our network model consists of a set of recurrently connected units",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        384,
+                                        127,
+                                        425,
+                                        137
+                                    ],
+                                    "score": 0.86,
+                                    "content": "N = 1 0 0",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        425,
+                                        126,
+                                        506,
+                                        139
+                                    ],
+                                    "score": 1.0,
+                                    "content": "). The dynamics of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                137,
+                                470,
+                                150
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        137,
+                                        205,
+                                        150
+                                    ],
+                                    "score": 1.0,
+                                    "content": "each unit in the network",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        206,
+                                        137,
+                                        226,
+                                        149
+                                    ],
+                                    "score": 0.93,
+                                    "content": "u _ { i } ( t )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        227,
+                                        137,
+                                        470,
+                                        150
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is governed by the standard continuous-time RNN equation:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        }
+                    ],
+                    "index": 2.5,
+                    "bbox_fs": [
+                        106,
+                        126,
+                        506,
+                        150
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        172,
+                        151,
+                        439,
+                        187
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                172,
+                                151,
+                                439,
+                                187
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        172,
+                                        151,
+                                        439,
+                                        187
+                                    ],
+                                    "score": 0.95,
+                                    "content": "\\tau \\frac { d x _ { i } ( t ) } { d t } = - x _ { i } ( t ) + \\sum _ { j = 1 } ^ { N } W _ { i j } ^ { \\mathrm { r e c } } u _ { j } ( t ) + \\sum _ { k = 1 } ^ { N _ { \\mathrm { i n } } } W _ { i k } ^ { \\mathrm { i n } } I _ { k } ( t ) + b _ { i } + \\xi _ { i } ( t )",
+                                    "type": "interline_equation",
+                                    "image_path": "73f5fb1e1639b7bb2d07858a8bd6b7b8e2b6dc0350eb54f951fc46c74a662eee.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 5,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                172,
+                                151,
+                                439,
+                                163.0
+                            ],
+                            "spans": [],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                172,
+                                163.0,
+                                439,
+                                175.0
+                            ],
+                            "spans": [],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                172,
+                                175.0,
+                                439,
+                                187.0
+                            ],
+                            "spans": [],
+                            "index": 6
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        190,
+                        505,
+                        268
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                189,
+                                505,
+                                204
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        189,
+                                        121,
+                                        204
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        122,
+                                        191,
+                                        178,
+                                        202
+                                    ],
+                                    "score": 0.91,
+                                    "content": "i = 1 , \\ldots , N",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        179,
+                                        189,
+                                        292,
+                                        204
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". The activity of each unit,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        293,
+                                        190,
+                                        314,
+                                        203
+                                    ],
+                                    "score": 0.92,
+                                    "content": "u _ { i } ( t )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        314,
+                                        189,
+                                        479,
+                                        204
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", is related to the activation of that unit,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        480,
+                                        190,
+                                        501,
+                                        203
+                                    ],
+                                    "score": 0.91,
+                                    "content": "x _ { i } ( t )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        501,
+                                        189,
+                                        505,
+                                        204
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                201,
+                                506,
+                                214
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        201,
+                                        337,
+                                        214
+                                    ],
+                                    "score": 1.0,
+                                    "content": "through a nonlinearity which in this study we take to be",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        338,
+                                        201,
+                                        422,
+                                        213
+                                    ],
+                                    "score": 0.92,
+                                    "content": "u _ { i } ( t ) = \\operatorname { t a n h } ( x _ { i } ( t ) )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        423,
+                                        201,
+                                        506,
+                                        214
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Each unit receives",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                211,
+                                505,
+                                226
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        211,
+                                        347,
+                                        226
+                                    ],
+                                    "score": 1.0,
+                                    "content": "input from other units through the recurrent weight matrix",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        348,
+                                        213,
+                                        370,
+                                        222
+                                    ],
+                                    "score": 0.85,
+                                    "content": "W ^ { \\mathrm { r e c } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        370,
+                                        211,
+                                        505,
+                                        226
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and also receives external input,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                222,
+                                506,
+                                237
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        224,
+                                        124,
+                                        235
+                                    ],
+                                    "score": 0.89,
+                                    "content": "I ( t )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        124,
+                                        222,
+                                        335,
+                                        237
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", that enters the network through the weight matrix",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        336,
+                                        223,
+                                        355,
+                                        234
+                                    ],
+                                    "score": 0.88,
+                                    "content": "W ^ { \\mathrm { i n } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        355,
+                                        222,
+                                        506,
+                                        237
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Each unit has two sources of bias,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                235,
+                                505,
+                                246
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        235,
+                                        115,
+                                        245
+                                    ],
+                                    "score": 0.85,
+                                    "content": "b _ { i }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        116,
+                                        235,
+                                        207,
+                                        246
+                                    ],
+                                    "score": 1.0,
+                                    "content": "which is learned and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        208,
+                                        235,
+                                        228,
+                                        246
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\xi _ { i } ( t )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        228,
+                                        235,
+                                        505,
+                                        246
+                                    ],
+                                    "score": 1.0,
+                                    "content": "which represents noise intrinsic to the network and is taken to be",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                245,
+                                505,
+                                257
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        245,
+                                        505,
+                                        257
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Gaussian with zero mean and constant variance. The network was simulated using the Euler method",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                255,
+                                270,
+                                268
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        255,
+                                        120,
+                                        268
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        121,
+                                        257,
+                                        158,
+                                        267
+                                    ],
+                                    "score": 0.89,
+                                    "content": "T = 5 0 0",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        158,
+                                        255,
+                                        245,
+                                        268
+                                    ],
+                                    "score": 1.0,
+                                    "content": "timesteps of duration",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        245,
+                                        256,
+                                        267,
+                                        268
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\tau / 1 0",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        267,
+                                        255,
+                                        270,
+                                        268
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        }
+                    ],
+                    "index": 10,
+                    "bbox_fs": [
+                        104,
+                        189,
+                        506,
+                        268
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        273,
+                        505,
+                        307
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                273,
+                                505,
+                                285
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        273,
+                                        505,
+                                        285
+                                    ],
+                                    "score": 1.0,
+                                    "content": "To perform a 2D navigation task with the RNN, we linearly combine the firing rates of units in",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                285,
+                                505,
+                                296
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        285,
+                                        505,
+                                        296
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the network to estimate the current location of the animal. The responses of the two linear readout",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                295,
+                                355,
+                                308
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        295,
+                                        143,
+                                        308
+                                    ],
+                                    "score": 1.0,
+                                    "content": "neurons,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        143,
+                                        295,
+                                        165,
+                                        307
+                                    ],
+                                    "score": 0.92,
+                                    "content": "y _ { 1 } ( t )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        165,
+                                        295,
+                                        183,
+                                        308
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        183,
+                                        295,
+                                        205,
+                                        307
+                                    ],
+                                    "score": 0.92,
+                                    "content": "y _ { 2 } ( t )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        205,
+                                        295,
+                                        355,
+                                        308
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", are given by the following equation:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        }
+                    ],
+                    "index": 15,
+                    "bbox_fs": [
+                        105,
+                        273,
+                        505,
+                        308
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        257,
+                        309,
+                        353,
+                        344
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                257,
+                                309,
+                                353,
+                                344
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        257,
+                                        309,
+                                        353,
+                                        344
+                                    ],
+                                    "score": 0.94,
+                                    "content": "y _ { j } ( t ) = \\sum _ { i = 1 } ^ { N } W _ { j i } ^ { \\mathrm { o u t } } u _ { i } ( t )",
+                                    "type": "interline_equation",
+                                    "image_path": "d64453343a45b8133e59ab95038edf8c7ca7e2bd28bd901cacdd4eff0e7578b4.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 17.5,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                257,
+                                309,
+                                353,
+                                326.5
+                            ],
+                            "spans": [],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                257,
+                                326.5,
+                                353,
+                                344.0
+                            ],
+                            "spans": [],
+                            "index": 18
+                        }
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        356,
+                        237,
+                        367
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                355,
+                                239,
+                                368
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        355,
+                                        239,
+                                        368
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2.2 INPUT TO THE NETWORK",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        }
+                    ],
+                    "index": 19
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        375,
+                        505,
+                        453
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                376,
+                                505,
+                                389
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        376,
+                                        505,
+                                        389
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The network inputs and outputs were inspired by simple spatial navigation tasks in 2D open envi-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                387,
+                                506,
+                                400
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        387,
+                                        506,
+                                        400
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ronments. The task resembles dead-reckoning (sometimes referred to as path integration), which is",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                398,
+                                505,
+                                410
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        398,
+                                        505,
+                                        410
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ethologically relevant for many animal species (Darwin, 1873; Mittelstaedt & Mittelstaedt, 1980;",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                408,
+                                506,
+                                421
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        408,
+                                        506,
+                                        421
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Etienne & Jeffery, 2004; McNaughton et al., 2006). To be more specific, the inputs to the network",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                420,
+                                505,
+                                432
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        420,
+                                        505,
+                                        432
+                                    ],
+                                    "score": 1.0,
+                                    "content": "were the animal’s speed and direction at each time step. Experimentally, it has been shown that the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                431,
+                                506,
+                                443
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        431,
+                                        506,
+                                        443
+                                    ],
+                                    "score": 1.0,
+                                    "content": "velocity signals exist in EC (Sargolini et al., 2006; Kropff et al., 2015; Hinman et al., 2016), and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                441,
+                                490,
+                                454
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        441,
+                                        490,
+                                        454
+                                    ],
+                                    "score": 1.0,
+                                    "content": "there is also evidence that such signals are necessary for grid formation (Winter et al., 2015a;b).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        }
+                    ],
+                    "index": 23,
+                    "bbox_fs": [
+                        105,
+                        376,
+                        506,
+                        454
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        458,
+                        505,
+                        536
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                458,
+                                505,
+                                471
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        458,
+                                        505,
+                                        471
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Throughout the paper, we adopt the common assumption that the head direction of the animal coin-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                470,
+                                505,
+                                482
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        470,
+                                        505,
+                                        482
+                                    ],
+                                    "score": 1.0,
+                                    "content": "cides with the actual moving direction. The outputs were the x- and y-coordinates of the integrated",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                480,
+                                506,
+                                493
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        480,
+                                        506,
+                                        493
+                                    ],
+                                    "score": 1.0,
+                                    "content": "position. The direction of the animal is modeled by modified Brownian motion to increase the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                492,
+                                506,
+                                504
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        492,
+                                        506,
+                                        504
+                                    ],
+                                    "score": 1.0,
+                                    "content": "probability of straight-runs, in order to be consistent with the typical rodent’s behavior in an open",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                502,
+                                506,
+                                515
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        502,
+                                        506,
+                                        515
+                                    ],
+                                    "score": 1.0,
+                                    "content": "environment. The usage of such simple movement statistics has the advantage of having full control",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                514,
+                                505,
+                                526
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        514,
+                                        505,
+                                        526
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of the simulated trajectories. However, for future work it would be very interesting to test the model",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                524,
+                                459,
+                                538
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        524,
+                                        459,
+                                        538
+                                    ],
+                                    "score": 1.0,
+                                    "content": "using different animals’ real movement trajectories to see how the results might change.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        }
+                    ],
+                    "index": 30,
+                    "bbox_fs": [
+                        104,
+                        458,
+                        506,
+                        538
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        541,
+                        505,
+                        597
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                542,
+                                505,
+                                554
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        542,
+                                        505,
+                                        554
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Special care is taken when the animal is close to the boundary. The boundary of the environment",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                551,
+                                505,
+                                566
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        551,
+                                        505,
+                                        566
+                                    ],
+                                    "score": 1.0,
+                                    "content": "will affect the statistics of the movement, as the animal cannot cross the boundary. This fact was",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                564,
+                                505,
+                                576
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        564,
+                                        505,
+                                        576
+                                    ],
+                                    "score": 1.0,
+                                    "content": "reflected in the model by re-sampling the angular input variable until the input angle did not lead",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                574,
+                                505,
+                                586
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        574,
+                                        505,
+                                        586
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the animal outside the boundary. In the simulations shown below, the animal always starts from the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                585,
+                                465,
+                                598
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        585,
+                                        465,
+                                        598
+                                    ],
+                                    "score": 1.0,
+                                    "content": "center of the arena, but we verified that the results are insensitive to the starting locations.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        }
+                    ],
+                    "index": 36,
+                    "bbox_fs": [
+                        105,
+                        542,
+                        505,
+                        598
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        609,
+                        176,
+                        621
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                608,
+                                177,
+                                623
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        608,
+                                        177,
+                                        623
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2.3 TRAINING",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        }
+                    ],
+                    "index": 39
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        629,
+                        505,
+                        674
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                628,
+                                506,
+                                644
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        628,
+                                        265,
+                                        644
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We optimized the network parameters",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        266,
+                                        630,
+                                        313,
+                                        641
+                                    ],
+                                    "score": 0.27,
+                                    "content": "W ^ { \\mathrm { r e c } } , W ^ { \\mathrm { i n } } ,",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        313,
+                                        628,
+                                        317,
+                                        644
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        317,
+                                        631,
+                                        323,
+                                        640
+                                    ],
+                                    "score": 0.69,
+                                    "content": "b",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        324,
+                                        628,
+                                        343,
+                                        644
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        343,
+                                        630,
+                                        367,
+                                        641
+                                    ],
+                                    "score": 0.87,
+                                    "content": "W ^ { \\mathrm { o u t } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        367,
+                                        628,
+                                        506,
+                                        644
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to minimize the squared error in",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                641,
+                                506,
+                                654
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        641,
+                                        216,
+                                        654
+                                    ],
+                                    "score": 1.0,
+                                    "content": "equation (3) between target",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        217,
+                                        643,
+                                        226,
+                                        652
+                                    ],
+                                    "score": 0.36,
+                                    "content": "\\mathbf { X } ^ { - }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        226,
+                                        641,
+                                        243,
+                                        654
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        244,
+                                        644,
+                                        250,
+                                        653
+                                    ],
+                                    "score": 0.32,
+                                    "content": "_ \\textrm { y }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        250,
+                                        641,
+                                        506,
+                                        654
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-coordinates from a two dimensional navigation task (performed",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                653,
+                                505,
+                                666
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        653,
+                                        505,
+                                        666
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in rectangular, hexagonal, and triangular arenas) and the network outputs generated according to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                663,
+                                160,
+                                677
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        663,
+                                        160,
+                                        677
+                                    ],
+                                    "score": 1.0,
+                                    "content": "equation (2).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        }
+                    ],
+                    "index": 41.5,
+                    "bbox_fs": [
+                        105,
+                        628,
+                        506,
+                        677
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        200,
+                        673,
+                        410,
+                        709
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                200,
+                                673,
+                                410,
+                                709
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        200,
+                                        673,
+                                        410,
+                                        709
+                                    ],
+                                    "score": 0.93,
+                                    "content": "E = \\frac { 1 } { M T N _ { \\mathrm { o u t } } } \\sum _ { m , t , j = 1 } ^ { M , T , N _ { \\mathrm { o u t } } } ( y _ { j } ( t , m ) - y _ { j } ^ { \\mathrm { t a r g e t } } ( t , m ) ) ^ { 2 }",
+                                    "type": "interline_equation",
+                                    "image_path": "c0586e3c42e764d474fb5fc1894e2ed9f63e5c96ba0d3cd1f2199ef9a9b3a013.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 44.5,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                200,
+                                673,
+                                410,
+                                691.0
+                            ],
+                            "spans": [],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                200,
+                                691.0,
+                                410,
+                                709.0
+                            ],
+                            "spans": [],
+                            "index": 45
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        709,
+                        504,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                709,
+                                505,
+                                722
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        709,
+                                        505,
+                                        722
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Parameters were updated with the Hessian-free algorithm (Martens & Sutskever, 2011) using mini-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                718,
+                                506,
+                                734
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        718,
+                                        170,
+                                        734
+                                    ],
+                                    "score": 1.0,
+                                    "content": "batches of size",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        171,
+                                        721,
+                                        214,
+                                        731
+                                    ],
+                                    "score": 0.9,
+                                    "content": "M = 5 0 0",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        214,
+                                        718,
+                                        506,
+                                        734
+                                    ],
+                                    "score": 1.0,
+                                    "content": "trials. In addition to minimizing the error function in equation (3) we",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 47
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                83,
+                                504,
+                                95
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        83,
+                                        504,
+                                        95
+                                    ],
+                                    "score": 1.0,
+                                    "content": "regularized the input and output weights according to equation (4) and the squared firing rates of the",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                94,
+                                505,
+                                106
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        94,
+                                        505,
+                                        106
+                                    ],
+                                    "score": 1.0,
+                                    "content": "units (referred to as metabolic cost) according to equation (5). In sum, the training aims to minimize",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                104,
+                                505,
+                                117
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        104,
+                                        505,
+                                        117
+                                    ],
+                                    "score": 1.0,
+                                    "content": "a loss function, that consists of the error of the animal, the metabolic cost, and a penalty for large",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                116,
+                                190,
+                                127
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        116,
+                                        190,
+                                        127
+                                    ],
+                                    "score": 1.0,
+                                    "content": "network parameters.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 3
+                        }
+                    ],
+                    "index": 46.5,
+                    "bbox_fs": [
+                        105,
+                        709,
+                        506,
+                        734
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        82,
+                        505,
+                        127
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                83,
+                                504,
+                                95
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        83,
+                                        504,
+                                        95
+                                    ],
+                                    "score": 1.0,
+                                    "content": "regularized the input and output weights according to equation (4) and the squared firing rates of the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                94,
+                                505,
+                                106
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        94,
+                                        505,
+                                        106
+                                    ],
+                                    "score": 1.0,
+                                    "content": "units (referred to as metabolic cost) according to equation (5). In sum, the training aims to minimize",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                104,
+                                505,
+                                117
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        104,
+                                        505,
+                                        117
+                                    ],
+                                    "score": 1.0,
+                                    "content": "a loss function, that consists of the error of the animal, the metabolic cost, and a penalty for large",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                116,
+                                190,
+                                127
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        116,
+                                        190,
+                                        127
+                                    ],
+                                    "score": 1.0,
+                                    "content": "network parameters.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        }
+                    ],
+                    "index": 1.5
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        196,
+                        143,
+                        415,
+                        217
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                196,
+                                143,
+                                415,
+                                217
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        196,
+                                        143,
+                                        415,
+                                        217
+                                    ],
+                                    "score": 0.94,
+                                    "content": "\\begin{array} { l } { { \\displaystyle R _ { L 2 } = \\frac { 1 } { N N _ { \\mathrm { i n } } } \\sum _ { i , j = 1 } ^ { N , N _ { \\mathrm { i n } } } ( W _ { i j } ^ { \\mathrm { i n } } ) ^ { 2 } + \\frac { 1 } { N N _ { \\mathrm { o u t } } } \\sum _ { i , j = 1 } ^ { N _ { \\mathrm { o u t } } , N } ( W _ { i j } ^ { \\mathrm { o u t } } ) ^ { 2 } } } \\\\ { { \\displaystyle R _ { F R } = \\frac { 1 } { N T M } \\sum _ { i , t , m = 1 } ^ { N , T , M } u _ { i } ( t , m ) ^ { 2 } } } \\end{array}",
+                                    "type": "interline_equation",
+                                    "image_path": "e960c0e70ffb072304b7abf5d77743a5b44399b4e093a5b22d41b568af796bb4.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 6,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                196,
+                                143,
+                                415,
+                                157.8
+                            ],
+                            "spans": [],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                196,
+                                157.8,
+                                415,
+                                172.60000000000002
+                            ],
+                            "spans": [],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                196,
+                                172.60000000000002,
+                                415,
+                                187.40000000000003
+                            ],
+                            "spans": [],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                196,
+                                187.40000000000003,
+                                415,
+                                202.20000000000005
+                            ],
+                            "spans": [],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                196,
+                                202.20000000000005,
+                                415,
+                                217.00000000000006
+                            ],
+                            "spans": [],
+                            "index": 8
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        223,
+                        505,
+                        290
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                222,
+                                505,
+                                236
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        222,
+                                        505,
+                                        236
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We find that the results are qualitatively insensitive to the initialization schemes used for the re-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                232,
+                                505,
+                                247
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        232,
+                                        197,
+                                        247
+                                    ],
+                                    "score": 1.0,
+                                    "content": "current weight matrix",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        198,
+                                        235,
+                                        220,
+                                        245
+                                    ],
+                                    "score": 0.86,
+                                    "content": "W ^ { \\mathrm { r e c } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        220,
+                                        232,
+                                        505,
+                                        247
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". For the results presented in this paper, simulations in the hexagonal",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                244,
+                                506,
+                                258
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        244,
+                                        341,
+                                        258
+                                    ],
+                                    "score": 1.0,
+                                    "content": "environment were obtained by initializing the elements of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        342,
+                                        245,
+                                        364,
+                                        255
+                                    ],
+                                    "score": 0.87,
+                                    "content": "W ^ { \\mathrm { r e c } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        365,
+                                        244,
+                                        506,
+                                        258
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to be zero mean Gaussian random",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                255,
+                                506,
+                                269
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        255,
+                                        204,
+                                        269
+                                    ],
+                                    "score": 1.0,
+                                    "content": "variables with variance",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        204,
+                                        256,
+                                        236,
+                                        268
+                                    ],
+                                    "score": 0.91,
+                                    "content": "1 . 5 ^ { 2 } \\dot { / } N",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        236,
+                                        255,
+                                        506,
+                                        269
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", and simulations in the square and triangular environments were",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                265,
+                                506,
+                                280
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        265,
+                                        226,
+                                        280
+                                    ],
+                                    "score": 1.0,
+                                    "content": "initialized with an orthogonal",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        227,
+                                        268,
+                                        249,
+                                        277
+                                    ],
+                                    "score": 0.86,
+                                    "content": "W ^ { \\mathrm { r e c } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        249,
+                                        265,
+                                        420,
+                                        280
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(Saxe et al., 2014). We initialized the bias",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        420,
+                                        268,
+                                        426,
+                                        277
+                                    ],
+                                    "score": 0.6,
+                                    "content": "b",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        426,
+                                        265,
+                                        506,
+                                        280
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and output weights",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                107,
+                                276,
+                                493,
+                                292
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        278,
+                                        130,
+                                        289
+                                    ],
+                                    "score": 0.8,
+                                    "content": "W ^ { \\mathrm { o u t } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        131,
+                                        276,
+                                        243,
+                                        292
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to be zero. The elements of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        244,
+                                        279,
+                                        262,
+                                        289
+                                    ],
+                                    "score": 0.87,
+                                    "content": "W ^ { \\mathrm { i n } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        263,
+                                        276,
+                                        462,
+                                        292
+                                    ],
+                                    "score": 1.0,
+                                    "content": "were zero mean Gaussian variables with variance",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        462,
+                                        278,
+                                        488,
+                                        290
+                                    ],
+                                    "score": 0.91,
+                                    "content": "1 / N _ { \\mathrm { i n } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        489,
+                                        276,
+                                        493,
+                                        292
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        }
+                    ],
+                    "index": 11.5
+                },
+                {
+                    "type": "image",
+                    "bbox": [
+                        126,
+                        300,
+                        484,
+                        467
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                126,
+                                300,
+                                484,
+                                467
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        300,
+                                        484,
+                                        467
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                126,
+                                                300,
+                                                484,
+                                                467
+                                            ],
+                                            "score": 0.973,
+                                            "type": "image",
+                                            "image_path": "7cde4a241b1b10ed008d3d0dbd8d11d439e41445d8cafe267c621ecd7571dd08.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 16,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        300,
+                                        484,
+                                        355.6666666666667
+                                    ],
+                                    "spans": [],
+                                    "index": 15
+                                },
+                                {
+                                    "bbox": [
+                                        126,
+                                        355.6666666666667,
+                                        484,
+                                        411.33333333333337
+                                    ],
+                                    "spans": [],
+                                    "index": 16
+                                },
+                                {
+                                    "bbox": [
+                                        126,
+                                        411.33333333333337,
+                                        484,
+                                        467.00000000000006
+                                    ],
+                                    "spans": [],
+                                    "index": 17
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                106,
+                                478,
+                                505,
+                                534
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        478,
+                                        504,
+                                        490
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                478,
+                                                504,
+                                                490
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 2: Different types of spatial selective responses of units in the trained RNN. Example sim-",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 18
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        488,
+                                        505,
+                                        502
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                488,
+                                                505,
+                                                502
+                                            ],
+                                            "score": 1.0,
+                                            "content": "ulation results for three different environments (square, triangular, hexagon) are presented. Blue",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 19
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        500,
+                                        505,
+                                        513
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                500,
+                                                505,
+                                                513
+                                            ],
+                                            "score": 1.0,
+                                            "content": "(yellow) represents low (high) activity. a) Grid-like responses. b) Band-like responses; c) Border-",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 20
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        511,
+                                        506,
+                                        524
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                511,
+                                                506,
+                                                524
+                                            ],
+                                            "score": 1.0,
+                                            "content": "related responses; d) Spatially irregular responses. These responses can be spatially selective but",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 21
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        523,
+                                        379,
+                                        534
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                523,
+                                                379,
+                                                534
+                                            ],
+                                            "score": 1.0,
+                                            "content": "they do not form a regular pattern defined in the conventional sense.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 22
+                                }
+                            ],
+                            "index": 20
+                        }
+                    ],
+                    "index": 18.0
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        554,
+                        172,
+                        567
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                552,
+                                174,
+                                570
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        552,
+                                        174,
+                                        570
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3 RESULTS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        }
+                    ],
+                    "index": 23
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        578,
+                        505,
+                        612
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                577,
+                                505,
+                                592
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        577,
+                                        505,
+                                        592
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We run simulation experiments in arenas with different boundary shapes, including square, triangu-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                590,
+                                505,
+                                602
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        590,
+                                        505,
+                                        602
+                                    ],
+                                    "score": 1.0,
+                                    "content": "lar and hexagonal. Figure 1c shows a typical example of the model performance after training; the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                601,
+                                385,
+                                613
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        601,
+                                        385,
+                                        613
+                                    ],
+                                    "score": 1.0,
+                                    "content": "network (red trace) accurately tracks the animal’s actual path (black).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        }
+                    ],
+                    "index": 25
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        625,
+                        330,
+                        636
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                624,
+                                331,
+                                637
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        624,
+                                        331,
+                                        637
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3.1 TUNING PROPERTIES OF THE MODEL NEURONS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        }
+                    ],
+                    "index": 27
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        645,
+                        505,
+                        679
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                645,
+                                505,
+                                658
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        645,
+                                        505,
+                                        658
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We are mostly interested in what kind of representation the RNN has learned to solve this navigation",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                656,
+                                505,
+                                668
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        656,
+                                        505,
+                                        668
+                                    ],
+                                    "score": 1.0,
+                                    "content": "task, and whether such a representation resembles the response properties of neurons in EC (Moser",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                667,
+                                159,
+                                680
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        667,
+                                        159,
+                                        680
+                                    ],
+                                    "score": 1.0,
+                                    "content": "et al., 2008).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        }
+                    ],
+                    "index": 29
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        690,
+                        212,
+                        702
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                689,
+                                213,
+                                703
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        689,
+                                        213,
+                                        703
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3.1.1 SPATIAL TUNING",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        }
+                    ],
+                    "index": 31
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        709,
+                        504,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                709,
+                                505,
+                                722
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        709,
+                                        505,
+                                        722
+                                    ],
+                                    "score": 1.0,
+                                    "content": "To test whether the trained RNN developed location-selective representations, we plot individual",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                721,
+                                506,
+                                733
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        721,
+                                        506,
+                                        733
+                                    ],
+                                    "score": 1.0,
+                                    "content": "neurons’ mean activity level as a function of the animal’s location during spatial exploration. Note",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        }
+                    ],
+                    "index": 32.5
+                }
+            ],
+            "page_idx": 3,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        107,
+                        27,
+                        293,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                26,
+                                294,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        26,
+                                        294,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Published as a conference paper at ICLR 2018",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        302,
+                        752,
+                        308,
+                        759
+                    ],
+                    "lines": []
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        82,
+                        505,
+                        127
+                    ],
+                    "lines": [],
+                    "index": 1.5,
+                    "bbox_fs": [
+                        105,
+                        83,
+                        505,
+                        127
+                    ],
+                    "lines_deleted": true
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        196,
+                        143,
+                        415,
+                        217
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                196,
+                                143,
+                                415,
+                                217
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        196,
+                                        143,
+                                        415,
+                                        217
+                                    ],
+                                    "score": 0.94,
+                                    "content": "\\begin{array} { l } { { \\displaystyle R _ { L 2 } = \\frac { 1 } { N N _ { \\mathrm { i n } } } \\sum _ { i , j = 1 } ^ { N , N _ { \\mathrm { i n } } } ( W _ { i j } ^ { \\mathrm { i n } } ) ^ { 2 } + \\frac { 1 } { N N _ { \\mathrm { o u t } } } \\sum _ { i , j = 1 } ^ { N _ { \\mathrm { o u t } } , N } ( W _ { i j } ^ { \\mathrm { o u t } } ) ^ { 2 } } } \\\\ { { \\displaystyle R _ { F R } = \\frac { 1 } { N T M } \\sum _ { i , t , m = 1 } ^ { N , T , M } u _ { i } ( t , m ) ^ { 2 } } } \\end{array}",
+                                    "type": "interline_equation",
+                                    "image_path": "e960c0e70ffb072304b7abf5d77743a5b44399b4e093a5b22d41b568af796bb4.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 6,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                196,
+                                143,
+                                415,
+                                157.8
+                            ],
+                            "spans": [],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                196,
+                                157.8,
+                                415,
+                                172.60000000000002
+                            ],
+                            "spans": [],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                196,
+                                172.60000000000002,
+                                415,
+                                187.40000000000003
+                            ],
+                            "spans": [],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                196,
+                                187.40000000000003,
+                                415,
+                                202.20000000000005
+                            ],
+                            "spans": [],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                196,
+                                202.20000000000005,
+                                415,
+                                217.00000000000006
+                            ],
+                            "spans": [],
+                            "index": 8
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        223,
+                        505,
+                        290
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                222,
+                                505,
+                                236
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        222,
+                                        505,
+                                        236
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We find that the results are qualitatively insensitive to the initialization schemes used for the re-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                232,
+                                505,
+                                247
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        232,
+                                        197,
+                                        247
+                                    ],
+                                    "score": 1.0,
+                                    "content": "current weight matrix",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        198,
+                                        235,
+                                        220,
+                                        245
+                                    ],
+                                    "score": 0.86,
+                                    "content": "W ^ { \\mathrm { r e c } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        220,
+                                        232,
+                                        505,
+                                        247
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". For the results presented in this paper, simulations in the hexagonal",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                244,
+                                506,
+                                258
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        244,
+                                        341,
+                                        258
+                                    ],
+                                    "score": 1.0,
+                                    "content": "environment were obtained by initializing the elements of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        342,
+                                        245,
+                                        364,
+                                        255
+                                    ],
+                                    "score": 0.87,
+                                    "content": "W ^ { \\mathrm { r e c } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        365,
+                                        244,
+                                        506,
+                                        258
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to be zero mean Gaussian random",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                255,
+                                506,
+                                269
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        255,
+                                        204,
+                                        269
+                                    ],
+                                    "score": 1.0,
+                                    "content": "variables with variance",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        204,
+                                        256,
+                                        236,
+                                        268
+                                    ],
+                                    "score": 0.91,
+                                    "content": "1 . 5 ^ { 2 } \\dot { / } N",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        236,
+                                        255,
+                                        506,
+                                        269
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", and simulations in the square and triangular environments were",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                265,
+                                506,
+                                280
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        265,
+                                        226,
+                                        280
+                                    ],
+                                    "score": 1.0,
+                                    "content": "initialized with an orthogonal",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        227,
+                                        268,
+                                        249,
+                                        277
+                                    ],
+                                    "score": 0.86,
+                                    "content": "W ^ { \\mathrm { r e c } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        249,
+                                        265,
+                                        420,
+                                        280
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(Saxe et al., 2014). We initialized the bias",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        420,
+                                        268,
+                                        426,
+                                        277
+                                    ],
+                                    "score": 0.6,
+                                    "content": "b",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        426,
+                                        265,
+                                        506,
+                                        280
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and output weights",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                107,
+                                276,
+                                493,
+                                292
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        278,
+                                        130,
+                                        289
+                                    ],
+                                    "score": 0.8,
+                                    "content": "W ^ { \\mathrm { o u t } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        131,
+                                        276,
+                                        243,
+                                        292
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to be zero. The elements of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        244,
+                                        279,
+                                        262,
+                                        289
+                                    ],
+                                    "score": 0.87,
+                                    "content": "W ^ { \\mathrm { i n } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        263,
+                                        276,
+                                        462,
+                                        292
+                                    ],
+                                    "score": 1.0,
+                                    "content": "were zero mean Gaussian variables with variance",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        462,
+                                        278,
+                                        488,
+                                        290
+                                    ],
+                                    "score": 0.91,
+                                    "content": "1 / N _ { \\mathrm { i n } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        489,
+                                        276,
+                                        493,
+                                        292
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        }
+                    ],
+                    "index": 11.5,
+                    "bbox_fs": [
+                        104,
+                        222,
+                        506,
+                        292
+                    ]
+                },
+                {
+                    "type": "image",
+                    "bbox": [
+                        126,
+                        300,
+                        484,
+                        467
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                126,
+                                300,
+                                484,
+                                467
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        300,
+                                        484,
+                                        467
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                126,
+                                                300,
+                                                484,
+                                                467
+                                            ],
+                                            "score": 0.973,
+                                            "type": "image",
+                                            "image_path": "7cde4a241b1b10ed008d3d0dbd8d11d439e41445d8cafe267c621ecd7571dd08.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 16,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        300,
+                                        484,
+                                        355.6666666666667
+                                    ],
+                                    "spans": [],
+                                    "index": 15
+                                },
+                                {
+                                    "bbox": [
+                                        126,
+                                        355.6666666666667,
+                                        484,
+                                        411.33333333333337
+                                    ],
+                                    "spans": [],
+                                    "index": 16
+                                },
+                                {
+                                    "bbox": [
+                                        126,
+                                        411.33333333333337,
+                                        484,
+                                        467.00000000000006
+                                    ],
+                                    "spans": [],
+                                    "index": 17
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                106,
+                                478,
+                                505,
+                                534
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        478,
+                                        504,
+                                        490
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                478,
+                                                504,
+                                                490
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 2: Different types of spatial selective responses of units in the trained RNN. Example sim-",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 18
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        488,
+                                        505,
+                                        502
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                488,
+                                                505,
+                                                502
+                                            ],
+                                            "score": 1.0,
+                                            "content": "ulation results for three different environments (square, triangular, hexagon) are presented. Blue",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 19
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        500,
+                                        505,
+                                        513
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                500,
+                                                505,
+                                                513
+                                            ],
+                                            "score": 1.0,
+                                            "content": "(yellow) represents low (high) activity. a) Grid-like responses. b) Band-like responses; c) Border-",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 20
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        511,
+                                        506,
+                                        524
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                511,
+                                                506,
+                                                524
+                                            ],
+                                            "score": 1.0,
+                                            "content": "related responses; d) Spatially irregular responses. These responses can be spatially selective but",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 21
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        523,
+                                        379,
+                                        534
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                523,
+                                                379,
+                                                534
+                                            ],
+                                            "score": 1.0,
+                                            "content": "they do not form a regular pattern defined in the conventional sense.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 22
+                                }
+                            ],
+                            "index": 20
+                        }
+                    ],
+                    "index": 18.0
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        554,
+                        172,
+                        567
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                552,
+                                174,
+                                570
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        552,
+                                        174,
+                                        570
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3 RESULTS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        }
+                    ],
+                    "index": 23
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        578,
+                        505,
+                        612
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                577,
+                                505,
+                                592
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        577,
+                                        505,
+                                        592
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We run simulation experiments in arenas with different boundary shapes, including square, triangu-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                590,
+                                505,
+                                602
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        590,
+                                        505,
+                                        602
+                                    ],
+                                    "score": 1.0,
+                                    "content": "lar and hexagonal. Figure 1c shows a typical example of the model performance after training; the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                601,
+                                385,
+                                613
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        601,
+                                        385,
+                                        613
+                                    ],
+                                    "score": 1.0,
+                                    "content": "network (red trace) accurately tracks the animal’s actual path (black).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        }
+                    ],
+                    "index": 25,
+                    "bbox_fs": [
+                        105,
+                        577,
+                        505,
+                        613
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        625,
+                        330,
+                        636
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                624,
+                                331,
+                                637
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        624,
+                                        331,
+                                        637
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3.1 TUNING PROPERTIES OF THE MODEL NEURONS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        }
+                    ],
+                    "index": 27
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        645,
+                        505,
+                        679
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                645,
+                                505,
+                                658
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        645,
+                                        505,
+                                        658
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We are mostly interested in what kind of representation the RNN has learned to solve this navigation",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                656,
+                                505,
+                                668
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        656,
+                                        505,
+                                        668
+                                    ],
+                                    "score": 1.0,
+                                    "content": "task, and whether such a representation resembles the response properties of neurons in EC (Moser",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                667,
+                                159,
+                                680
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        667,
+                                        159,
+                                        680
+                                    ],
+                                    "score": 1.0,
+                                    "content": "et al., 2008).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        }
+                    ],
+                    "index": 29,
+                    "bbox_fs": [
+                        105,
+                        645,
+                        505,
+                        680
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        690,
+                        212,
+                        702
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                689,
+                                213,
+                                703
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        689,
+                                        213,
+                                        703
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3.1.1 SPATIAL TUNING",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        }
+                    ],
+                    "index": 31
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        709,
+                        504,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                709,
+                                505,
+                                722
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        709,
+                                        505,
+                                        722
+                                    ],
+                                    "score": 1.0,
+                                    "content": "To test whether the trained RNN developed location-selective representations, we plot individual",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                721,
+                                506,
+                                733
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        721,
+                                        506,
+                                        733
+                                    ],
+                                    "score": 1.0,
+                                    "content": "neurons’ mean activity level as a function of the animal’s location during spatial exploration. Note",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                82,
+                                505,
+                                95
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        82,
+                                        505,
+                                        95
+                                    ],
+                                    "score": 1.0,
+                                    "content": "that these average response profiles should not be confused with the linear filters typically shown in",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                92,
+                                506,
+                                108
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        92,
+                                        506,
+                                        108
+                                    ],
+                                    "score": 1.0,
+                                    "content": "feedforward networks. Surprisingly, we find neurons in the trained RNN show a range of interesting",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                104,
+                                506,
+                                118
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        104,
+                                        506,
+                                        118
+                                    ],
+                                    "score": 1.0,
+                                    "content": "spatial response profiles. Examination of these response profiles suggests they can be classified into",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                116,
+                                505,
+                                127
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        116,
+                                        505,
+                                        127
+                                    ],
+                                    "score": 1.0,
+                                    "content": "distinct functional types. Importantly, as we will show, these distinct spatial response profiles can be",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                126,
+                                505,
+                                139
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        126,
+                                        505,
+                                        139
+                                    ],
+                                    "score": 1.0,
+                                    "content": "mapped naturally to known physiology in EC. The spatial responses of all units in trained networks",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                137,
+                                218,
+                                149
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        137,
+                                        218,
+                                        149
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are shown in the Appendix.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 5
+                        }
+                    ],
+                    "index": 32.5,
+                    "bbox_fs": [
+                        105,
+                        709,
+                        506,
+                        733
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        82,
+                        505,
+                        149
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                82,
+                                505,
+                                95
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        82,
+                                        505,
+                                        95
+                                    ],
+                                    "score": 1.0,
+                                    "content": "that these average response profiles should not be confused with the linear filters typically shown in",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                92,
+                                506,
+                                108
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        92,
+                                        506,
+                                        108
+                                    ],
+                                    "score": 1.0,
+                                    "content": "feedforward networks. Surprisingly, we find neurons in the trained RNN show a range of interesting",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                104,
+                                506,
+                                118
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        104,
+                                        506,
+                                        118
+                                    ],
+                                    "score": 1.0,
+                                    "content": "spatial response profiles. Examination of these response profiles suggests they can be classified into",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                116,
+                                505,
+                                127
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        116,
+                                        505,
+                                        127
+                                    ],
+                                    "score": 1.0,
+                                    "content": "distinct functional types. Importantly, as we will show, these distinct spatial response profiles can be",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                126,
+                                505,
+                                139
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        126,
+                                        505,
+                                        139
+                                    ],
+                                    "score": 1.0,
+                                    "content": "mapped naturally to known physiology in EC. The spatial responses of all units in trained networks",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                137,
+                                218,
+                                149
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        137,
+                                        218,
+                                        149
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are shown in the Appendix.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        }
+                    ],
+                    "index": 2.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        154,
+                        505,
+                        231
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                154,
+                                505,
+                                167
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        154,
+                                        505,
+                                        167
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Grid-like responses Most interestingly, we find some of the units in the RNN exhibit clear grid-like",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                165,
+                                505,
+                                178
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        165,
+                                        505,
+                                        178
+                                    ],
+                                    "score": 1.0,
+                                    "content": "responses (Figure 2a). These firing patterns typically exhibit multiple firing fields, with each firing",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                175,
+                                505,
+                                190
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        175,
+                                        505,
+                                        190
+                                    ],
+                                    "score": 1.0,
+                                    "content": "field exhibiting roughly circular symmetric or ellipse shape. Furthermore, the firing fields are highly",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                187,
+                                505,
+                                200
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        187,
+                                        505,
+                                        200
+                                    ],
+                                    "score": 1.0,
+                                    "content": "structured, i.e., when combined, are arranged on a regular lattice. Furthermore, the structure of the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                198,
+                                505,
+                                211
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        198,
+                                        505,
+                                        211
+                                    ],
+                                    "score": 1.0,
+                                    "content": "response lattice depends on the shape of the boundary. In particular, training the network to perform",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                209,
+                                505,
+                                222
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        209,
+                                        505,
+                                        222
+                                    ],
+                                    "score": 1.0,
+                                    "content": "self-localization in a square environment tends to give rectangular grids. In hexagonal and triangular",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                221,
+                                296,
+                                233
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        221,
+                                        296,
+                                        233
+                                    ],
+                                    "score": 1.0,
+                                    "content": "environments, the grids are closer to triangular.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        }
+                    ],
+                    "index": 9
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        237,
+                        505,
+                        379
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                237,
+                                504,
+                                249
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        237,
+                                        504,
+                                        249
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Experimentally, it is shown that (medial) EC contains so-called grid cells which exhibit multiple",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                246,
+                                506,
+                                262
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        246,
+                                        506,
+                                        262
+                                    ],
+                                    "score": 1.0,
+                                    "content": "firing fields that lie on a regular grid (Fyhn et al., 2004; Hafting et al., 2005). The grid-like firing",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                258,
+                                506,
+                                273
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        258,
+                                        506,
+                                        273
+                                    ],
+                                    "score": 1.0,
+                                    "content": "patterns in our simulation are reminiscent of the grid cells in rodents and other mammals. However,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                268,
+                                505,
+                                284
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        268,
+                                        505,
+                                        284
+                                    ],
+                                    "score": 1.0,
+                                    "content": "we also notice that the the grid-like model responses typically exhibit few periods, not as many",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                281,
+                                505,
+                                293
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        281,
+                                        505,
+                                        293
+                                    ],
+                                    "score": 1.0,
+                                    "content": "as experimental data (see Figure 1a). It is possible that using a larger network might reveal finer",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                292,
+                                505,
+                                305
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        292,
+                                        505,
+                                        305
+                                    ],
+                                    "score": 1.0,
+                                    "content": "grid-patterns in our model. Nonetheless, it is surprising that the gird-like spatial representations can",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                302,
+                                505,
+                                316
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        302,
+                                        505,
+                                        316
+                                    ],
+                                    "score": 1.0,
+                                    "content": "develop in our model, given there is no periodicity in the input. Another potential concern is that,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                313,
+                                505,
+                                327
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        313,
+                                        505,
+                                        327
+                                    ],
+                                    "score": 1.0,
+                                    "content": "experimentally it is reported that the grids are often on the corners of a triangular lattice (Hafting",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                324,
+                                506,
+                                339
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        324,
+                                        506,
+                                        339
+                                    ],
+                                    "score": 1.0,
+                                    "content": "et al., 2005) even in square environments (see Figure 1a), though the grids are somewhat influenced",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                335,
+                                505,
+                                348
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        335,
+                                        505,
+                                        348
+                                    ],
+                                    "score": 1.0,
+                                    "content": "by the shape of the environment. However, the rats in these experiments presumable had spatial",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                347,
+                                505,
+                                359
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        347,
+                                        505,
+                                        359
+                                    ],
+                                    "score": 1.0,
+                                    "content": "experience in other environments with various boundary shapes. Experimentally, it would be inter-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                356,
+                                505,
+                                371
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        356,
+                                        505,
+                                        371
+                                    ],
+                                    "score": 1.0,
+                                    "content": "esting to see if grid cells would lie on a square lattice instead if the rats are raised in a single square",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                369,
+                                302,
+                                380
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        369,
+                                        302,
+                                        380
+                                    ],
+                                    "score": 1.0,
+                                    "content": "environment - a situation we are simulating here.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        }
+                    ],
+                    "index": 19
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        385,
+                        505,
+                        506
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                384,
+                                505,
+                                398
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        384,
+                                        505,
+                                        398
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Border responses Many neurons in the RNN exhibit selectivity to the boundary (Figure 2c). Typi-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                397,
+                                505,
+                                409
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        397,
+                                        505,
+                                        409
+                                    ],
+                                    "score": 1.0,
+                                    "content": "cally, they only encode a portion of the boundary, e.g. one piece of wall in a square shaped environ-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                408,
+                                505,
+                                419
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        408,
+                                        505,
+                                        419
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ment. Such properties are similar to the border cells discovered in rodent EC (Solstad et al., 2008;",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                417,
+                                506,
+                                431
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        417,
+                                        506,
+                                        431
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Savelli et al., 2008; Lever et al., 2009). Experimentally, border cells mainly fire along one piece of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                429,
+                                505,
+                                442
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        429,
+                                        505,
+                                        442
+                                    ],
+                                    "score": 1.0,
+                                    "content": "wall, although some have been observed to fire along multiple borders or along the whole bound-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                439,
+                                506,
+                                453
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        439,
+                                        506,
+                                        453
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ary of the environment; interestingly, these multi-border responses were also observed in some RNN",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                451,
+                                505,
+                                464
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        451,
+                                        505,
+                                        464
+                                    ],
+                                    "score": 1.0,
+                                    "content": "models. Currently, it is unclear how the boundary-like response profiles emerge (Solstad et al., 2008;",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                460,
+                                505,
+                                476
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        460,
+                                        505,
+                                        476
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Savelli et al., 2008; Lever et al., 2009). Our model points to the possibility that the border cells may",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                472,
+                                505,
+                                487
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        472,
+                                        505,
+                                        487
+                                    ],
+                                    "score": 1.0,
+                                    "content": "emerge without the presence of tactile cues. Furthermore, it suggests that border cell formation may",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                483,
+                                505,
+                                497
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        483,
+                                        505,
+                                        497
+                                    ],
+                                    "score": 1.0,
+                                    "content": "be related to the movement statistics of the animals, i.e. due to the asymmetry of the movement",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                496,
+                                225,
+                                508
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        496,
+                                        225,
+                                        508
+                                    ],
+                                    "score": 1.0,
+                                    "content": "statistics along the boundary.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        }
+                    ],
+                    "index": 31
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        511,
+                        505,
+                        589
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                512,
+                                505,
+                                524
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        512,
+                                        505,
+                                        524
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Band-like responses Interestingly, some neurons in the RNN exhibit band-like responses (Figure",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                522,
+                                505,
+                                534
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        522,
+                                        505,
+                                        534
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2b). In most of our simulations, these bands tend to be parallel to one of the boundaries. For",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                534,
+                                505,
+                                547
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        534,
+                                        505,
+                                        547
+                                    ],
+                                    "score": 1.0,
+                                    "content": "some of the units, one of the bands overlaps the boundary, but for others, that is not the case.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                545,
+                                504,
+                                557
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        545,
+                                        504,
+                                        557
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Experimentally, neurons with periodic-like firing patterns have been recently reported in rodent EC.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                555,
+                                505,
+                                569
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        555,
+                                        505,
+                                        569
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In one study, it has been reported that a substantial portion of cells in EC exhibit band-like firing",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                567,
+                                505,
+                                579
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        567,
+                                        505,
+                                        579
+                                    ],
+                                    "score": 1.0,
+                                    "content": "characteristics (Krupic et al., 2012). However, we note that based on the reported data in Krupic",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                578,
+                                351,
+                                589
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        578,
+                                        351,
+                                        589
+                                    ],
+                                    "score": 1.0,
+                                    "content": "et al. (2012), the band pattern is not as clear as in our model.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        }
+                    ],
+                    "index": 40
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        595,
+                        505,
+                        661
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                594,
+                                505,
+                                607
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        594,
+                                        505,
+                                        607
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Spatially-stable but non-regular responses Besides the units described above, most of the remain-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                605,
+                                505,
+                                618
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        605,
+                                        505,
+                                        618
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ing units also exhibit stable spatial responses, but they do not belong to the above categories. These",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                616,
+                                505,
+                                629
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        616,
+                                        505,
+                                        629
+                                    ],
+                                    "score": 1.0,
+                                    "content": "response profiles can exhibit either one large irregular firing field; or multiple circular firing fields,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                627,
+                                505,
+                                640
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        627,
+                                        505,
+                                        640
+                                    ],
+                                    "score": 1.0,
+                                    "content": "but these firing fields do not show a regular pattern. Experimentally these types of cells have also",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 47
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                638,
+                                505,
+                                651
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        638,
+                                        505,
+                                        651
+                                    ],
+                                    "score": 1.0,
+                                    "content": "been observed. In fact, it is recently reported that the non-grid spatial cells constitute a large portion",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 48
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                649,
+                                374,
+                                662
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        649,
+                                        374,
+                                        662
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of the neurons in Layer II and III of rodent EC (Diehl et al., 2017).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 49
+                        }
+                    ],
+                    "index": 46.5
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        678,
+                        338,
+                        689
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                677,
+                                339,
+                                690
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        677,
+                                        339,
+                                        690
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3.1.2 SPEED TUNING AND HEAD DIRECTION TUNING",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 50
+                        }
+                    ],
+                    "index": 50
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        699,
+                        504,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                698,
+                                506,
+                                711
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        698,
+                                        506,
+                                        711
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Speed tuning We next ask how neurons in the RNN are tuned to the inputs. Many of the model",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 51
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                710,
+                                506,
+                                722
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        710,
+                                        506,
+                                        722
+                                    ],
+                                    "score": 1.0,
+                                    "content": "neurons exhibit linear responses to the running speed of the animal, while some neurons show no",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 52
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                720,
+                                506,
+                                734
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        720,
+                                        506,
+                                        734
+                                    ],
+                                    "score": 1.0,
+                                    "content": "selectivity to speed, as suggested by the near-flat response functions. Example response profiles are",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 53
+                        }
+                    ],
+                    "index": 52
+                }
+            ],
+            "page_idx": 4,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        108,
+                        27,
+                        293,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                26,
+                                294,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        26,
+                                        294,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Published as a conference paper at ICLR 2018",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        302,
+                        751,
+                        309,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                301,
+                                750,
+                                310,
+                                762
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        301,
+                                        750,
+                                        310,
+                                        762
+                                    ],
+                                    "score": 1.0,
+                                    "content": "5",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        82,
+                        505,
+                        149
+                    ],
+                    "lines": [],
+                    "index": 2.5,
+                    "bbox_fs": [
+                        105,
+                        82,
+                        506,
+                        149
+                    ],
+                    "lines_deleted": true
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        154,
+                        505,
+                        231
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                154,
+                                505,
+                                167
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        154,
+                                        505,
+                                        167
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Grid-like responses Most interestingly, we find some of the units in the RNN exhibit clear grid-like",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                165,
+                                505,
+                                178
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        165,
+                                        505,
+                                        178
+                                    ],
+                                    "score": 1.0,
+                                    "content": "responses (Figure 2a). These firing patterns typically exhibit multiple firing fields, with each firing",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                175,
+                                505,
+                                190
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        175,
+                                        505,
+                                        190
+                                    ],
+                                    "score": 1.0,
+                                    "content": "field exhibiting roughly circular symmetric or ellipse shape. Furthermore, the firing fields are highly",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                187,
+                                505,
+                                200
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        187,
+                                        505,
+                                        200
+                                    ],
+                                    "score": 1.0,
+                                    "content": "structured, i.e., when combined, are arranged on a regular lattice. Furthermore, the structure of the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                198,
+                                505,
+                                211
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        198,
+                                        505,
+                                        211
+                                    ],
+                                    "score": 1.0,
+                                    "content": "response lattice depends on the shape of the boundary. In particular, training the network to perform",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                209,
+                                505,
+                                222
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        209,
+                                        505,
+                                        222
+                                    ],
+                                    "score": 1.0,
+                                    "content": "self-localization in a square environment tends to give rectangular grids. In hexagonal and triangular",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                221,
+                                296,
+                                233
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        221,
+                                        296,
+                                        233
+                                    ],
+                                    "score": 1.0,
+                                    "content": "environments, the grids are closer to triangular.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        }
+                    ],
+                    "index": 9,
+                    "bbox_fs": [
+                        105,
+                        154,
+                        505,
+                        233
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        237,
+                        505,
+                        379
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                237,
+                                504,
+                                249
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        237,
+                                        504,
+                                        249
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Experimentally, it is shown that (medial) EC contains so-called grid cells which exhibit multiple",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                246,
+                                506,
+                                262
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        246,
+                                        506,
+                                        262
+                                    ],
+                                    "score": 1.0,
+                                    "content": "firing fields that lie on a regular grid (Fyhn et al., 2004; Hafting et al., 2005). The grid-like firing",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                258,
+                                506,
+                                273
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        258,
+                                        506,
+                                        273
+                                    ],
+                                    "score": 1.0,
+                                    "content": "patterns in our simulation are reminiscent of the grid cells in rodents and other mammals. However,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                268,
+                                505,
+                                284
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        268,
+                                        505,
+                                        284
+                                    ],
+                                    "score": 1.0,
+                                    "content": "we also notice that the the grid-like model responses typically exhibit few periods, not as many",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                281,
+                                505,
+                                293
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        281,
+                                        505,
+                                        293
+                                    ],
+                                    "score": 1.0,
+                                    "content": "as experimental data (see Figure 1a). It is possible that using a larger network might reveal finer",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                292,
+                                505,
+                                305
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        292,
+                                        505,
+                                        305
+                                    ],
+                                    "score": 1.0,
+                                    "content": "grid-patterns in our model. Nonetheless, it is surprising that the gird-like spatial representations can",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                302,
+                                505,
+                                316
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        302,
+                                        505,
+                                        316
+                                    ],
+                                    "score": 1.0,
+                                    "content": "develop in our model, given there is no periodicity in the input. Another potential concern is that,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                313,
+                                505,
+                                327
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        313,
+                                        505,
+                                        327
+                                    ],
+                                    "score": 1.0,
+                                    "content": "experimentally it is reported that the grids are often on the corners of a triangular lattice (Hafting",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                324,
+                                506,
+                                339
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        324,
+                                        506,
+                                        339
+                                    ],
+                                    "score": 1.0,
+                                    "content": "et al., 2005) even in square environments (see Figure 1a), though the grids are somewhat influenced",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                335,
+                                505,
+                                348
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        335,
+                                        505,
+                                        348
+                                    ],
+                                    "score": 1.0,
+                                    "content": "by the shape of the environment. However, the rats in these experiments presumable had spatial",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                347,
+                                505,
+                                359
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        347,
+                                        505,
+                                        359
+                                    ],
+                                    "score": 1.0,
+                                    "content": "experience in other environments with various boundary shapes. Experimentally, it would be inter-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                356,
+                                505,
+                                371
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        356,
+                                        505,
+                                        371
+                                    ],
+                                    "score": 1.0,
+                                    "content": "esting to see if grid cells would lie on a square lattice instead if the rats are raised in a single square",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                369,
+                                302,
+                                380
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        369,
+                                        302,
+                                        380
+                                    ],
+                                    "score": 1.0,
+                                    "content": "environment - a situation we are simulating here.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        }
+                    ],
+                    "index": 19,
+                    "bbox_fs": [
+                        104,
+                        237,
+                        506,
+                        380
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        385,
+                        505,
+                        506
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                384,
+                                505,
+                                398
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        384,
+                                        505,
+                                        398
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Border responses Many neurons in the RNN exhibit selectivity to the boundary (Figure 2c). Typi-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                397,
+                                505,
+                                409
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        397,
+                                        505,
+                                        409
+                                    ],
+                                    "score": 1.0,
+                                    "content": "cally, they only encode a portion of the boundary, e.g. one piece of wall in a square shaped environ-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                408,
+                                505,
+                                419
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        408,
+                                        505,
+                                        419
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ment. Such properties are similar to the border cells discovered in rodent EC (Solstad et al., 2008;",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                417,
+                                506,
+                                431
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        417,
+                                        506,
+                                        431
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Savelli et al., 2008; Lever et al., 2009). Experimentally, border cells mainly fire along one piece of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                429,
+                                505,
+                                442
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        429,
+                                        505,
+                                        442
+                                    ],
+                                    "score": 1.0,
+                                    "content": "wall, although some have been observed to fire along multiple borders or along the whole bound-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                439,
+                                506,
+                                453
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        439,
+                                        506,
+                                        453
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ary of the environment; interestingly, these multi-border responses were also observed in some RNN",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                451,
+                                505,
+                                464
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        451,
+                                        505,
+                                        464
+                                    ],
+                                    "score": 1.0,
+                                    "content": "models. Currently, it is unclear how the boundary-like response profiles emerge (Solstad et al., 2008;",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                460,
+                                505,
+                                476
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        460,
+                                        505,
+                                        476
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Savelli et al., 2008; Lever et al., 2009). Our model points to the possibility that the border cells may",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                472,
+                                505,
+                                487
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        472,
+                                        505,
+                                        487
+                                    ],
+                                    "score": 1.0,
+                                    "content": "emerge without the presence of tactile cues. Furthermore, it suggests that border cell formation may",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                483,
+                                505,
+                                497
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        483,
+                                        505,
+                                        497
+                                    ],
+                                    "score": 1.0,
+                                    "content": "be related to the movement statistics of the animals, i.e. due to the asymmetry of the movement",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                496,
+                                225,
+                                508
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        496,
+                                        225,
+                                        508
+                                    ],
+                                    "score": 1.0,
+                                    "content": "statistics along the boundary.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        }
+                    ],
+                    "index": 31,
+                    "bbox_fs": [
+                        105,
+                        384,
+                        506,
+                        508
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        511,
+                        505,
+                        589
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                512,
+                                505,
+                                524
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        512,
+                                        505,
+                                        524
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Band-like responses Interestingly, some neurons in the RNN exhibit band-like responses (Figure",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                522,
+                                505,
+                                534
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        522,
+                                        505,
+                                        534
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2b). In most of our simulations, these bands tend to be parallel to one of the boundaries. For",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                534,
+                                505,
+                                547
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        534,
+                                        505,
+                                        547
+                                    ],
+                                    "score": 1.0,
+                                    "content": "some of the units, one of the bands overlaps the boundary, but for others, that is not the case.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                545,
+                                504,
+                                557
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        545,
+                                        504,
+                                        557
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Experimentally, neurons with periodic-like firing patterns have been recently reported in rodent EC.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                555,
+                                505,
+                                569
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        555,
+                                        505,
+                                        569
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In one study, it has been reported that a substantial portion of cells in EC exhibit band-like firing",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                567,
+                                505,
+                                579
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        567,
+                                        505,
+                                        579
+                                    ],
+                                    "score": 1.0,
+                                    "content": "characteristics (Krupic et al., 2012). However, we note that based on the reported data in Krupic",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                578,
+                                351,
+                                589
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        578,
+                                        351,
+                                        589
+                                    ],
+                                    "score": 1.0,
+                                    "content": "et al. (2012), the band pattern is not as clear as in our model.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        }
+                    ],
+                    "index": 40,
+                    "bbox_fs": [
+                        105,
+                        512,
+                        505,
+                        589
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        595,
+                        505,
+                        661
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                594,
+                                505,
+                                607
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        594,
+                                        505,
+                                        607
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Spatially-stable but non-regular responses Besides the units described above, most of the remain-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                605,
+                                505,
+                                618
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        605,
+                                        505,
+                                        618
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ing units also exhibit stable spatial responses, but they do not belong to the above categories. These",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                616,
+                                505,
+                                629
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        616,
+                                        505,
+                                        629
+                                    ],
+                                    "score": 1.0,
+                                    "content": "response profiles can exhibit either one large irregular firing field; or multiple circular firing fields,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                627,
+                                505,
+                                640
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        627,
+                                        505,
+                                        640
+                                    ],
+                                    "score": 1.0,
+                                    "content": "but these firing fields do not show a regular pattern. Experimentally these types of cells have also",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 47
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                638,
+                                505,
+                                651
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        638,
+                                        505,
+                                        651
+                                    ],
+                                    "score": 1.0,
+                                    "content": "been observed. In fact, it is recently reported that the non-grid spatial cells constitute a large portion",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 48
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                649,
+                                374,
+                                662
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        649,
+                                        374,
+                                        662
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of the neurons in Layer II and III of rodent EC (Diehl et al., 2017).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 49
+                        }
+                    ],
+                    "index": 46.5,
+                    "bbox_fs": [
+                        105,
+                        594,
+                        505,
+                        662
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        678,
+                        338,
+                        689
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                677,
+                                339,
+                                690
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        677,
+                                        339,
+                                        690
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3.1.2 SPEED TUNING AND HEAD DIRECTION TUNING",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 50
+                        }
+                    ],
+                    "index": 50
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        699,
+                        504,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                698,
+                                506,
+                                711
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        698,
+                                        506,
+                                        711
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Speed tuning We next ask how neurons in the RNN are tuned to the inputs. Many of the model",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 51
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                710,
+                                506,
+                                722
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        710,
+                                        506,
+                                        722
+                                    ],
+                                    "score": 1.0,
+                                    "content": "neurons exhibit linear responses to the running speed of the animal, while some neurons show no",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 52
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                720,
+                                506,
+                                734
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        720,
+                                        506,
+                                        734
+                                    ],
+                                    "score": 1.0,
+                                    "content": "selectivity to speed, as suggested by the near-flat response functions. Example response profiles are",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 53
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                313,
+                                505,
+                                329
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        313,
+                                        505,
+                                        329
+                                    ],
+                                    "score": 1.0,
+                                    "content": "shown in Figure 3. Interestingly, we observe that the model border cells tend to have almost zero",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                326,
+                                253,
+                                339
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        326,
+                                        253,
+                                        339
+                                    ],
+                                    "score": 1.0,
+                                    "content": "speed-tuning (e.g., see Figure 3g,h).",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 11
+                        }
+                    ],
+                    "index": 52,
+                    "bbox_fs": [
+                        105,
+                        698,
+                        506,
+                        734
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "image",
+                    "bbox": [
+                        131,
+                        83,
+                        480,
+                        201
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                131,
+                                83,
+                                480,
+                                201
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        131,
+                                        83,
+                                        480,
+                                        201
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                131,
+                                                83,
+                                                480,
+                                                201
+                                            ],
+                                            "score": 0.974,
+                                            "type": "image",
+                                            "image_path": "318a67b1eb08bdf270e30eebad468380d0e5a80e744e9f80d58a5d693f3a94ed.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 1,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        131,
+                                        83,
+                                        480,
+                                        122.33333333333334
+                                    ],
+                                    "spans": [],
+                                    "index": 0
+                                },
+                                {
+                                    "bbox": [
+                                        131,
+                                        122.33333333333334,
+                                        480,
+                                        161.66666666666669
+                                    ],
+                                    "spans": [],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        131,
+                                        161.66666666666669,
+                                        480,
+                                        201.00000000000003
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                106,
+                                214,
+                                505,
+                                291
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        214,
+                                        505,
+                                        226
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                214,
+                                                505,
+                                                226
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 3: Direction tuning and speed tuning for nine example units in an RNN trained in a triangular",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 3
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        225,
+                                        504,
+                                        237
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                225,
+                                                504,
+                                                237
+                                            ],
+                                            "score": 1.0,
+                                            "content": "arena. For each unit, we show the spatial tuning, (head) directional tuning, speed tuning respectively,",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 4
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        235,
+                                        505,
+                                        248
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                235,
+                                                505,
+                                                248
+                                            ],
+                                            "score": 1.0,
+                                            "content": "from left to right. a,b,c) The three model neurons show strong directional tuning, but the spatial",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 5
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        247,
+                                        505,
+                                        259
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                247,
+                                                505,
+                                                259
+                                            ],
+                                            "score": 1.0,
+                                            "content": "tuning is weak and irregular. The three neurons also exhibit linear speed tuning. d,e,f) The three",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 6
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        257,
+                                        506,
+                                        271
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                257,
+                                                506,
+                                                271
+                                            ],
+                                            "score": 1.0,
+                                            "content": "neurons exhibit grid-like firing patterns, and clear speed tuning. The strength of their direction",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 7
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        268,
+                                        506,
+                                        282
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                268,
+                                                164,
+                                                282
+                                            ],
+                                            "score": 1.0,
+                                            "content": "tuning differ.",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                164,
+                                                269,
+                                                182,
+                                                280
+                                            ],
+                                            "score": 0.49,
+                                            "content": "{ \\bf g } , { \\bf h } )",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                182,
+                                                268,
+                                                506,
+                                                282
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Border cells exhibit weak and a bit complex directional tuning and almost no",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 8
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        279,
+                                        455,
+                                        294
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                279,
+                                                455,
+                                                294
+                                            ],
+                                            "score": 1.0,
+                                            "content": "speed tuning. i) This band cell shows weak directional tuning, but strong speed tuning.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 9
+                                }
+                            ],
+                            "index": 6
+                        }
+                    ],
+                    "index": 3.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        315,
+                        504,
+                        338
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                313,
+                                505,
+                                329
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        313,
+                                        505,
+                                        329
+                                    ],
+                                    "score": 1.0,
+                                    "content": "shown in Figure 3. Interestingly, we observe that the model border cells tend to have almost zero",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                326,
+                                253,
+                                339
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        326,
+                                        253,
+                                        339
+                                    ],
+                                    "score": 1.0,
+                                    "content": "speed-tuning (e.g., see Figure 3g,h).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        }
+                    ],
+                    "index": 10.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        343,
+                        505,
+                        453
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                343,
+                                505,
+                                356
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        343,
+                                        505,
+                                        356
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Head direction tuning A substantial portion of the model neurons show direction tuning. There",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                354,
+                                505,
+                                367
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        354,
+                                        505,
+                                        367
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are a diversity of direction tuning profiles, both in terms of the strength of the tuning and their pre-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                365,
+                                506,
+                                378
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        365,
+                                        506,
+                                        378
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ferred direction. Example tuning curves are shown in Figure 3, and the direction tuning curves of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                376,
+                                505,
+                                389
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        376,
+                                        505,
+                                        389
+                                    ],
+                                    "score": 1.0,
+                                    "content": "a complete population are shown in the Appendix. Interestingly, in general model neurons which",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                387,
+                                505,
+                                400
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        387,
+                                        505,
+                                        400
+                                    ],
+                                    "score": 1.0,
+                                    "content": "show the strongest head direction tuning do not show a clear spatial firing pattern (see Figure 3a,b,c).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                398,
+                                505,
+                                411
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        398,
+                                        505,
+                                        411
+                                    ],
+                                    "score": 1.0,
+                                    "content": "This suggests that there are a group of neurons which are mostly responsible for encoding the direc-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                407,
+                                506,
+                                423
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        407,
+                                        506,
+                                        423
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tion. We also notice that neurons with clear grid-like firing can exhibit a variety of direction tuning",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                420,
+                                505,
+                                433
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        420,
+                                        505,
+                                        433
+                                    ],
+                                    "score": 1.0,
+                                    "content": "strengths, from weak to strong (Figure 3d,e,f). In the Appendix, we quantify the relation between",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                430,
+                                506,
+                                444
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        430,
+                                        506,
+                                        444
+                                    ],
+                                    "score": 1.0,
+                                    "content": "these different tuning properties at the whole population level, which show somewhat complex de-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                442,
+                                150,
+                                454
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        442,
+                                        150,
+                                        454
+                                    ],
+                                    "score": 1.0,
+                                    "content": "pendence.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        }
+                    ],
+                    "index": 16.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        459,
+                        505,
+                        536
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                459,
+                                505,
+                                473
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        459,
+                                        505,
+                                        473
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Experimentally, the heading direction tuning in EC is well-known, e.g., Sargolini et al. (2006). Both",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                469,
+                                506,
+                                483
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        469,
+                                        506,
+                                        483
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the grid and non-grid cells in EC exhibit head direction tuning (Sargolini et al., 2006). Furthermore,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                482,
+                                504,
+                                493
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        482,
+                                        504,
+                                        493
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the linear speed dependence of the model neurons is similar to the properties of speed cells reported",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                491,
+                                506,
+                                506
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        491,
+                                        506,
+                                        506
+                                    ],
+                                    "score": 1.0,
+                                    "content": "recently in EC (Kropff et al., 2015). Our result is also consistent with another recent study reporting",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                503,
+                                505,
+                                515
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        503,
+                                        505,
+                                        515
+                                    ],
+                                    "score": 1.0,
+                                    "content": "that the majority of neurons in EC exhibit some amount of speed tuning (Hinman et al., 2016).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                513,
+                                505,
+                                528
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        513,
+                                        505,
+                                        528
+                                    ],
+                                    "score": 1.0,
+                                    "content": "It remains an open question experimentally, at a population level, how different types of tuning",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                525,
+                                272,
+                                537
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        525,
+                                        272,
+                                        537
+                                    ],
+                                    "score": 1.0,
+                                    "content": "characteristics in EC relate to each other.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        }
+                    ],
+                    "index": 25
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        552,
+                        329,
+                        563
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                551,
+                                330,
+                                564
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        551,
+                                        330,
+                                        564
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3.1.3 DEVELOPMENT OF THE TUNING PROPERTIES",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        }
+                    ],
+                    "index": 29
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        572,
+                        505,
+                        693
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                572,
+                                505,
+                                585
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        572,
+                                        505,
+                                        585
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We next investigate how the spatial response profiles evolve as learning/training progresses. We re-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                582,
+                                505,
+                                597
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        582,
+                                        505,
+                                        597
+                                    ],
+                                    "score": 1.0,
+                                    "content": "port two main observations. First, neurons that fire selectively along the boundary typically emerge",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                594,
+                                505,
+                                607
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        594,
+                                        505,
+                                        607
+                                    ],
+                                    "score": 1.0,
+                                    "content": "first. Second, the grid-like responses with finer spatial tuning patterns only emerge later in train-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                605,
+                                505,
+                                618
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        605,
+                                        505,
+                                        618
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ing. For visualization, we perform dimensionality reduction using the t-SNE algorithm (Maaten &",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                614,
+                                505,
+                                630
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        614,
+                                        505,
+                                        630
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Hinton, 2008). This algorithm embeds 100 model neurons during three phases of training (early,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                627,
+                                505,
+                                640
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        627,
+                                        505,
+                                        640
+                                    ],
+                                    "score": 1.0,
+                                    "content": "intermediate, and late) into a two-dimensional space according to the similarity of their temporal",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                638,
+                                505,
+                                650
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        638,
+                                        505,
+                                        650
+                                    ],
+                                    "score": 1.0,
+                                    "content": "responses. Here the similarity metric is taken to be firing rate correlation. In this 2D space as shown",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                647,
+                                506,
+                                663
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        647,
+                                        506,
+                                        663
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in Figure 4a, border cell representations appear early and stably persist through the end of training.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                659,
+                                505,
+                                673
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        659,
+                                        505,
+                                        673
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Furthermore, early during training all responses are similar to the border related responses. In con-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                671,
+                                506,
+                                685
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        671,
+                                        506,
+                                        685
+                                    ],
+                                    "score": 1.0,
+                                    "content": "trast, grid-like cells typically undergo a substantial change in firing pattern during training before",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                682,
+                                343,
+                                694
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        682,
+                                        343,
+                                        694
+                                    ],
+                                    "score": 1.0,
+                                    "content": "settling into their final grid-like representation (Figure 4b).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        }
+                    ],
+                    "index": 35
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        699,
+                        504,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                698,
+                                505,
+                                711
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        698,
+                                        505,
+                                        711
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The developmental time line of the grid-like cells and border cells is roughly consistent with de-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                709,
+                                505,
+                                721
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        709,
+                                        505,
+                                        721
+                                    ],
+                                    "score": 1.0,
+                                    "content": "velopmental studies in rodents. Experimentally, it is known that border cells emerge earlier in de-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                720,
+                                505,
+                                733
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        720,
+                                        505,
+                                        733
+                                    ],
+                                    "score": 1.0,
+                                    "content": "velopment, and they exist at about 2 weeks after the rat is born (Bjerknes et al., 2014). The grid",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        }
+                    ],
+                    "index": 42
+                }
+            ],
+            "page_idx": 5,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        107,
+                        27,
+                        293,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                26,
+                                294,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        26,
+                                        294,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Published as a conference paper at ICLR 2018",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        302,
+                        752,
+                        308,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                302,
+                                751,
+                                309,
+                                762
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        302,
+                                        751,
+                                        309,
+                                        762
+                                    ],
+                                    "score": 1.0,
+                                    "content": "6",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "image",
+                    "bbox": [
+                        131,
+                        83,
+                        480,
+                        201
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                131,
+                                83,
+                                480,
+                                201
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        131,
+                                        83,
+                                        480,
+                                        201
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                131,
+                                                83,
+                                                480,
+                                                201
+                                            ],
+                                            "score": 0.974,
+                                            "type": "image",
+                                            "image_path": "318a67b1eb08bdf270e30eebad468380d0e5a80e744e9f80d58a5d693f3a94ed.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 1,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        131,
+                                        83,
+                                        480,
+                                        122.33333333333334
+                                    ],
+                                    "spans": [],
+                                    "index": 0
+                                },
+                                {
+                                    "bbox": [
+                                        131,
+                                        122.33333333333334,
+                                        480,
+                                        161.66666666666669
+                                    ],
+                                    "spans": [],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        131,
+                                        161.66666666666669,
+                                        480,
+                                        201.00000000000003
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                106,
+                                214,
+                                505,
+                                291
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        214,
+                                        505,
+                                        226
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                214,
+                                                505,
+                                                226
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 3: Direction tuning and speed tuning for nine example units in an RNN trained in a triangular",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 3
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        225,
+                                        504,
+                                        237
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                225,
+                                                504,
+                                                237
+                                            ],
+                                            "score": 1.0,
+                                            "content": "arena. For each unit, we show the spatial tuning, (head) directional tuning, speed tuning respectively,",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 4
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        235,
+                                        505,
+                                        248
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                235,
+                                                505,
+                                                248
+                                            ],
+                                            "score": 1.0,
+                                            "content": "from left to right. a,b,c) The three model neurons show strong directional tuning, but the spatial",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 5
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        247,
+                                        505,
+                                        259
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                247,
+                                                505,
+                                                259
+                                            ],
+                                            "score": 1.0,
+                                            "content": "tuning is weak and irregular. The three neurons also exhibit linear speed tuning. d,e,f) The three",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 6
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        257,
+                                        506,
+                                        271
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                257,
+                                                506,
+                                                271
+                                            ],
+                                            "score": 1.0,
+                                            "content": "neurons exhibit grid-like firing patterns, and clear speed tuning. The strength of their direction",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 7
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        268,
+                                        506,
+                                        282
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                268,
+                                                164,
+                                                282
+                                            ],
+                                            "score": 1.0,
+                                            "content": "tuning differ.",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                164,
+                                                269,
+                                                182,
+                                                280
+                                            ],
+                                            "score": 0.49,
+                                            "content": "{ \\bf g } , { \\bf h } )",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                182,
+                                                268,
+                                                506,
+                                                282
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Border cells exhibit weak and a bit complex directional tuning and almost no",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 8
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        279,
+                                        455,
+                                        294
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                279,
+                                                455,
+                                                294
+                                            ],
+                                            "score": 1.0,
+                                            "content": "speed tuning. i) This band cell shows weak directional tuning, but strong speed tuning.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 9
+                                }
+                            ],
+                            "index": 6
+                        }
+                    ],
+                    "index": 3.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        315,
+                        504,
+                        338
+                    ],
+                    "lines": [],
+                    "index": 10.5,
+                    "bbox_fs": [
+                        105,
+                        313,
+                        505,
+                        339
+                    ],
+                    "lines_deleted": true
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        343,
+                        505,
+                        453
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                343,
+                                505,
+                                356
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        343,
+                                        505,
+                                        356
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Head direction tuning A substantial portion of the model neurons show direction tuning. There",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                354,
+                                505,
+                                367
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        354,
+                                        505,
+                                        367
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are a diversity of direction tuning profiles, both in terms of the strength of the tuning and their pre-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                365,
+                                506,
+                                378
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        365,
+                                        506,
+                                        378
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ferred direction. Example tuning curves are shown in Figure 3, and the direction tuning curves of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                376,
+                                505,
+                                389
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        376,
+                                        505,
+                                        389
+                                    ],
+                                    "score": 1.0,
+                                    "content": "a complete population are shown in the Appendix. Interestingly, in general model neurons which",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                387,
+                                505,
+                                400
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        387,
+                                        505,
+                                        400
+                                    ],
+                                    "score": 1.0,
+                                    "content": "show the strongest head direction tuning do not show a clear spatial firing pattern (see Figure 3a,b,c).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                398,
+                                505,
+                                411
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        398,
+                                        505,
+                                        411
+                                    ],
+                                    "score": 1.0,
+                                    "content": "This suggests that there are a group of neurons which are mostly responsible for encoding the direc-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                407,
+                                506,
+                                423
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        407,
+                                        506,
+                                        423
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tion. We also notice that neurons with clear grid-like firing can exhibit a variety of direction tuning",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                420,
+                                505,
+                                433
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        420,
+                                        505,
+                                        433
+                                    ],
+                                    "score": 1.0,
+                                    "content": "strengths, from weak to strong (Figure 3d,e,f). In the Appendix, we quantify the relation between",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                430,
+                                506,
+                                444
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        430,
+                                        506,
+                                        444
+                                    ],
+                                    "score": 1.0,
+                                    "content": "these different tuning properties at the whole population level, which show somewhat complex de-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                442,
+                                150,
+                                454
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        442,
+                                        150,
+                                        454
+                                    ],
+                                    "score": 1.0,
+                                    "content": "pendence.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        }
+                    ],
+                    "index": 16.5,
+                    "bbox_fs": [
+                        104,
+                        343,
+                        506,
+                        454
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        459,
+                        505,
+                        536
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                459,
+                                505,
+                                473
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        459,
+                                        505,
+                                        473
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Experimentally, the heading direction tuning in EC is well-known, e.g., Sargolini et al. (2006). Both",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                469,
+                                506,
+                                483
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        469,
+                                        506,
+                                        483
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the grid and non-grid cells in EC exhibit head direction tuning (Sargolini et al., 2006). Furthermore,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                482,
+                                504,
+                                493
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        482,
+                                        504,
+                                        493
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the linear speed dependence of the model neurons is similar to the properties of speed cells reported",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                491,
+                                506,
+                                506
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        491,
+                                        506,
+                                        506
+                                    ],
+                                    "score": 1.0,
+                                    "content": "recently in EC (Kropff et al., 2015). Our result is also consistent with another recent study reporting",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                503,
+                                505,
+                                515
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        503,
+                                        505,
+                                        515
+                                    ],
+                                    "score": 1.0,
+                                    "content": "that the majority of neurons in EC exhibit some amount of speed tuning (Hinman et al., 2016).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                513,
+                                505,
+                                528
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        513,
+                                        505,
+                                        528
+                                    ],
+                                    "score": 1.0,
+                                    "content": "It remains an open question experimentally, at a population level, how different types of tuning",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                525,
+                                272,
+                                537
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        525,
+                                        272,
+                                        537
+                                    ],
+                                    "score": 1.0,
+                                    "content": "characteristics in EC relate to each other.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        }
+                    ],
+                    "index": 25,
+                    "bbox_fs": [
+                        104,
+                        459,
+                        506,
+                        537
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        552,
+                        329,
+                        563
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                551,
+                                330,
+                                564
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        551,
+                                        330,
+                                        564
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3.1.3 DEVELOPMENT OF THE TUNING PROPERTIES",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        }
+                    ],
+                    "index": 29
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        572,
+                        505,
+                        693
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                572,
+                                505,
+                                585
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        572,
+                                        505,
+                                        585
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We next investigate how the spatial response profiles evolve as learning/training progresses. We re-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                582,
+                                505,
+                                597
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        582,
+                                        505,
+                                        597
+                                    ],
+                                    "score": 1.0,
+                                    "content": "port two main observations. First, neurons that fire selectively along the boundary typically emerge",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                594,
+                                505,
+                                607
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        594,
+                                        505,
+                                        607
+                                    ],
+                                    "score": 1.0,
+                                    "content": "first. Second, the grid-like responses with finer spatial tuning patterns only emerge later in train-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                605,
+                                505,
+                                618
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        605,
+                                        505,
+                                        618
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ing. For visualization, we perform dimensionality reduction using the t-SNE algorithm (Maaten &",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                614,
+                                505,
+                                630
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        614,
+                                        505,
+                                        630
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Hinton, 2008). This algorithm embeds 100 model neurons during three phases of training (early,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                627,
+                                505,
+                                640
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        627,
+                                        505,
+                                        640
+                                    ],
+                                    "score": 1.0,
+                                    "content": "intermediate, and late) into a two-dimensional space according to the similarity of their temporal",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                638,
+                                505,
+                                650
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        638,
+                                        505,
+                                        650
+                                    ],
+                                    "score": 1.0,
+                                    "content": "responses. Here the similarity metric is taken to be firing rate correlation. In this 2D space as shown",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                647,
+                                506,
+                                663
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        647,
+                                        506,
+                                        663
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in Figure 4a, border cell representations appear early and stably persist through the end of training.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                659,
+                                505,
+                                673
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        659,
+                                        505,
+                                        673
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Furthermore, early during training all responses are similar to the border related responses. In con-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                671,
+                                506,
+                                685
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        671,
+                                        506,
+                                        685
+                                    ],
+                                    "score": 1.0,
+                                    "content": "trast, grid-like cells typically undergo a substantial change in firing pattern during training before",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                682,
+                                343,
+                                694
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        682,
+                                        343,
+                                        694
+                                    ],
+                                    "score": 1.0,
+                                    "content": "settling into their final grid-like representation (Figure 4b).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        }
+                    ],
+                    "index": 35,
+                    "bbox_fs": [
+                        104,
+                        572,
+                        506,
+                        694
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        699,
+                        504,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                698,
+                                505,
+                                711
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        698,
+                                        505,
+                                        711
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The developmental time line of the grid-like cells and border cells is roughly consistent with de-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                709,
+                                505,
+                                721
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        709,
+                                        505,
+                                        721
+                                    ],
+                                    "score": 1.0,
+                                    "content": "velopmental studies in rodents. Experimentally, it is known that border cells emerge earlier in de-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                720,
+                                505,
+                                733
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        720,
+                                        505,
+                                        733
+                                    ],
+                                    "score": 1.0,
+                                    "content": "velopment, and they exist at about 2 weeks after the rat is born (Bjerknes et al., 2014). The grid",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                82,
+                                505,
+                                95
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        82,
+                                        505,
+                                        95
+                                    ],
+                                    "score": 1.0,
+                                    "content": "cells mature only at about 4 weeks after birth (Langston et al., 2010; Wills et al., 2010; Bjerknes",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                93,
+                                505,
+                                106
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        93,
+                                        505,
+                                        106
+                                    ],
+                                    "score": 1.0,
+                                    "content": "et al., 2014). Furthermore, our simulations suggest the reason why border cells emerge earlier in",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                104,
+                                506,
+                                117
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        104,
+                                        506,
+                                        117
+                                    ],
+                                    "score": 1.0,
+                                    "content": "development may be that computationally it is easier to wire-up a network that gives rise to border",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                116,
+                                167,
+                                128
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        116,
+                                        167,
+                                        128
+                                    ],
+                                    "score": 1.0,
+                                    "content": "cell responses.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 3
+                        }
+                    ],
+                    "index": 42,
+                    "bbox_fs": [
+                        106,
+                        698,
+                        505,
+                        733
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        82,
+                        505,
+                        127
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                82,
+                                505,
+                                95
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        82,
+                                        505,
+                                        95
+                                    ],
+                                    "score": 1.0,
+                                    "content": "cells mature only at about 4 weeks after birth (Langston et al., 2010; Wills et al., 2010; Bjerknes",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                93,
+                                505,
+                                106
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        93,
+                                        505,
+                                        106
+                                    ],
+                                    "score": 1.0,
+                                    "content": "et al., 2014). Furthermore, our simulations suggest the reason why border cells emerge earlier in",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                104,
+                                506,
+                                117
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        104,
+                                        506,
+                                        117
+                                    ],
+                                    "score": 1.0,
+                                    "content": "development may be that computationally it is easier to wire-up a network that gives rise to border",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                116,
+                                167,
+                                128
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        116,
+                                        167,
+                                        128
+                                    ],
+                                    "score": 1.0,
+                                    "content": "cell responses.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        }
+                    ],
+                    "index": 1.5
+                },
+                {
+                    "type": "image",
+                    "bbox": [
+                        116,
+                        140,
+                        488,
+                        290
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                116,
+                                140,
+                                488,
+                                290
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        140,
+                                        488,
+                                        290
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                116,
+                                                140,
+                                                488,
+                                                290
+                                            ],
+                                            "score": 0.971,
+                                            "type": "image",
+                                            "image_path": "9c19816ea23656a00f78df190e6e407bf4c84fe0856a9d69293f60f5972dfa0a.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 5,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        140,
+                                        488,
+                                        190.0
+                                    ],
+                                    "spans": [],
+                                    "index": 4
+                                },
+                                {
+                                    "bbox": [
+                                        116,
+                                        190.0,
+                                        488,
+                                        240.0
+                                    ],
+                                    "spans": [],
+                                    "index": 5
+                                },
+                                {
+                                    "bbox": [
+                                        116,
+                                        240.0,
+                                        488,
+                                        290.0
+                                    ],
+                                    "spans": [],
+                                    "index": 6
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                106,
+                                311,
+                                505,
+                                465
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        311,
+                                        505,
+                                        324
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                311,
+                                                505,
+                                                324
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 4: Development of border cells and grid-like cells. Early during training all responses are",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 7
+                                },
+                                {
+                                    "bbox": [
+                                        104,
+                                        321,
+                                        505,
+                                        336
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                104,
+                                                321,
+                                                505,
+                                                336
+                                            ],
+                                            "score": 1.0,
+                                            "content": "similar to the border related responses, and only as training continues do the grid-like cells emerge.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 8
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        333,
+                                        505,
+                                        346
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                333,
+                                                505,
+                                                346
+                                            ],
+                                            "score": 1.0,
+                                            "content": "We perform dimensionality reduction using the t-SNE algorithm on the firing rates of the neu-",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 9
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        344,
+                                        505,
+                                        357
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                344,
+                                                262,
+                                                357
+                                            ],
+                                            "score": 1.0,
+                                            "content": "rons. Each dot represents one neuron",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                262,
+                                                345,
+                                                303,
+                                                355
+                                            ],
+                                            "score": 0.87,
+                                            "content": "N = 1 0 0",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                303,
+                                                344,
+                                                505,
+                                                357
+                                            ],
+                                            "score": 1.0,
+                                            "content": "), and the color represents different training stages",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 10
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        355,
+                                        504,
+                                        367
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                355,
+                                                504,
+                                                367
+                                            ],
+                                            "score": 1.0,
+                                            "content": "(early/intermediate/late shown in blue/cyan/yellow). Each line shows the trajectory of a single high-",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 11
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        366,
+                                        505,
+                                        379
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                366,
+                                                505,
+                                                379
+                                            ],
+                                            "score": 1.0,
+                                            "content": "lighted neuron as its firing responses evolve during training. In panel a), we highlight the border",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 12
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        377,
+                                        506,
+                                        390
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                377,
+                                                506,
+                                                390
+                                            ],
+                                            "score": 1.0,
+                                            "content": "representation. It appears there are four clusters of border cells, each responding to one wall of a",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 13
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        388,
+                                        506,
+                                        401
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                388,
+                                                506,
+                                                401
+                                            ],
+                                            "score": 1.0,
+                                            "content": "square environment (spatial responses from four of these border cells are inset). These cells’ re-",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 14
+                                },
+                                {
+                                    "bbox": [
+                                        104,
+                                        399,
+                                        506,
+                                        413
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                104,
+                                                399,
+                                                506,
+                                                413
+                                            ],
+                                            "score": 1.0,
+                                            "content": "sponse profiles appear early and stably persist through training, illustrated by the short distance they",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 15
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        410,
+                                        505,
+                                        423
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                410,
+                                                505,
+                                                423
+                                            ],
+                                            "score": 1.0,
+                                            "content": "travel in this space. In b), we show that the neurons which eventually become grid cells initially have",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 16
+                                },
+                                {
+                                    "bbox": [
+                                        104,
+                                        420,
+                                        505,
+                                        436
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                104,
+                                                420,
+                                                505,
+                                                436
+                                            ],
+                                            "score": 1.0,
+                                            "content": "tuning profiles similar to the border cells but then change their tuning substantially during learning.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 17
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        431,
+                                        505,
+                                        444
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                431,
+                                                505,
+                                                444
+                                            ],
+                                            "score": 1.0,
+                                            "content": "As a natural consequence, they need to travel a long distance in this space between the early and late",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 18
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        444,
+                                        505,
+                                        455
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                444,
+                                                505,
+                                                455
+                                            ],
+                                            "score": 1.0,
+                                            "content": "phase of the training. Spatial responses are shown for four of these grid-like cells during the late",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 19
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        453,
+                                        178,
+                                        468
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                453,
+                                                178,
+                                                468
+                                            ],
+                                            "score": 1.0,
+                                            "content": "phase of training.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 20
+                                }
+                            ],
+                            "index": 13.5
+                        }
+                    ],
+                    "index": 9.25
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        484,
+                        299,
+                        495
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                484,
+                                302,
+                                497
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        484,
+                                        302,
+                                        497
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3.2 THE IMPORTANCE OF REGULARIZATION",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        }
+                    ],
+                    "index": 21
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        505,
+                        505,
+                        626
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                504,
+                                505,
+                                519
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        504,
+                                        505,
+                                        519
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We find appropriate regularizations of the RNN to be crucial for the emergence of grid-like repre-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                516,
+                                505,
+                                529
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        516,
+                                        505,
+                                        529
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sentations. We only observed grid-like representations when the network was encouraged to store",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                525,
+                                505,
+                                541
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        525,
+                                        505,
+                                        541
+                                    ],
+                                    "score": 1.0,
+                                    "content": "information while perturbed by noise. This was accomplished by setting the speed input to zero, e.g.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                538,
+                                505,
+                                551
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        538,
+                                        152,
+                                        551
+                                    ],
+                                    "score": 1.0,
+                                    "content": "zero speed",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        152,
+                                        538,
+                                        172,
+                                        549
+                                    ],
+                                    "score": 0.86,
+                                    "content": "90 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        172,
+                                        538,
+                                        398,
+                                        551
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of the time, and adding Gaussian noise to the network",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        399,
+                                        538,
+                                        419,
+                                        550
+                                    ],
+                                    "score": 0.9,
+                                    "content": "\\xi _ { i } ( t )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        420,
+                                        538,
+                                        505,
+                                        551
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in equation (1)); the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                549,
+                                505,
+                                562
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        549,
+                                        505,
+                                        562
+                                    ],
+                                    "score": 1.0,
+                                    "content": "precise method for setting the speed input to zero and the value of the noise variance is not crucial",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                560,
+                                505,
+                                573
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        560,
+                                        505,
+                                        573
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for our simulations to develop grid-like representations. The cost function which aims to capture",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                570,
+                                505,
+                                583
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        570,
+                                        505,
+                                        583
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the penalization on the metabolic cost of the neural activity also acts as an important regulariza-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                582,
+                                505,
+                                594
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        582,
+                                        505,
+                                        594
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tion. Our simulations show that the grid-like representation did not emerge without this metabolic",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                593,
+                                505,
+                                606
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        593,
+                                        505,
+                                        606
+                                    ],
+                                    "score": 1.0,
+                                    "content": "cost. In Figure 5, we show typical simulation results for a square environment, with and without",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                604,
+                                506,
+                                617
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        604,
+                                        506,
+                                        617
+                                    ],
+                                    "score": 1.0,
+                                    "content": "proper metabolic regularization. In the Appendix, we illustrate the effect of regularization further,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                615,
+                                343,
+                                628
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        615,
+                                        343,
+                                        628
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in particular the role of injecting noise into the RNN units.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        }
+                    ],
+                    "index": 27
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        632,
+                        505,
+                        676
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                631,
+                                505,
+                                644
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        631,
+                                        505,
+                                        644
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Our results are consistent with the general notion on the importance of incorporating proper con-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                643,
+                                505,
+                                655
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        643,
+                                        505,
+                                        655
+                                    ],
+                                    "score": 1.0,
+                                    "content": "straint for learning useful representations in neural networks (Bengio et al., 2013). Furthermore, it",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                653,
+                                505,
+                                667
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        653,
+                                        505,
+                                        667
+                                    ],
+                                    "score": 1.0,
+                                    "content": "suggests that, to learn a model with response properties similar to neural systems it may be necessary",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                665,
+                                383,
+                                677
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        665,
+                                        383,
+                                        677
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to incorporate the relevant constraints, e.g., noise and metabolic cost.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        }
+                    ],
+                    "index": 34.5
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        690,
+                        329,
+                        700
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                689,
+                                330,
+                                702
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        689,
+                                        330,
+                                        702
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3.3 ERROR CORRECTION AROUND THE BOUNDARY",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        }
+                    ],
+                    "index": 37
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        709,
+                        504,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                709,
+                                504,
+                                722
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        709,
+                                        504,
+                                        722
+                                    ],
+                                    "score": 1.0,
+                                    "content": "One natural question is whether the trained RNNs are able to perform localization when the path",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                720,
+                                505,
+                                733
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        720,
+                                        505,
+                                        733
+                                    ],
+                                    "score": 1.0,
+                                    "content": "length exceeds the typical length used during training (500 steps), in particular given that noise in",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        }
+                    ],
+                    "index": 38.5
+                }
+            ],
+            "page_idx": 6,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        108,
+                        27,
+                        293,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                26,
+                                294,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        26,
+                                        294,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Published as a conference paper at ICLR 2018",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        303,
+                        751,
+                        308,
+                        759
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                302,
+                                750,
+                                309,
+                                762
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        302,
+                                        750,
+                                        309,
+                                        762
+                                    ],
+                                    "score": 1.0,
+                                    "content": "",
+                                    "type": "text",
+                                    "height": 12,
+                                    "width": 7
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        82,
+                        505,
+                        127
+                    ],
+                    "lines": [],
+                    "index": 1.5,
+                    "bbox_fs": [
+                        105,
+                        82,
+                        506,
+                        128
+                    ],
+                    "lines_deleted": true
+                },
+                {
+                    "type": "image",
+                    "bbox": [
+                        116,
+                        140,
+                        488,
+                        290
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                116,
+                                140,
+                                488,
+                                290
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        140,
+                                        488,
+                                        290
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                116,
+                                                140,
+                                                488,
+                                                290
+                                            ],
+                                            "score": 0.971,
+                                            "type": "image",
+                                            "image_path": "9c19816ea23656a00f78df190e6e407bf4c84fe0856a9d69293f60f5972dfa0a.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 5,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        140,
+                                        488,
+                                        190.0
+                                    ],
+                                    "spans": [],
+                                    "index": 4
+                                },
+                                {
+                                    "bbox": [
+                                        116,
+                                        190.0,
+                                        488,
+                                        240.0
+                                    ],
+                                    "spans": [],
+                                    "index": 5
+                                },
+                                {
+                                    "bbox": [
+                                        116,
+                                        240.0,
+                                        488,
+                                        290.0
+                                    ],
+                                    "spans": [],
+                                    "index": 6
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                106,
+                                311,
+                                505,
+                                465
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        311,
+                                        505,
+                                        324
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                311,
+                                                505,
+                                                324
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 4: Development of border cells and grid-like cells. Early during training all responses are",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 7
+                                },
+                                {
+                                    "bbox": [
+                                        104,
+                                        321,
+                                        505,
+                                        336
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                104,
+                                                321,
+                                                505,
+                                                336
+                                            ],
+                                            "score": 1.0,
+                                            "content": "similar to the border related responses, and only as training continues do the grid-like cells emerge.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 8
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        333,
+                                        505,
+                                        346
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                333,
+                                                505,
+                                                346
+                                            ],
+                                            "score": 1.0,
+                                            "content": "We perform dimensionality reduction using the t-SNE algorithm on the firing rates of the neu-",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 9
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        344,
+                                        505,
+                                        357
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                344,
+                                                262,
+                                                357
+                                            ],
+                                            "score": 1.0,
+                                            "content": "rons. Each dot represents one neuron",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                262,
+                                                345,
+                                                303,
+                                                355
+                                            ],
+                                            "score": 0.87,
+                                            "content": "N = 1 0 0",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                303,
+                                                344,
+                                                505,
+                                                357
+                                            ],
+                                            "score": 1.0,
+                                            "content": "), and the color represents different training stages",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 10
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        355,
+                                        504,
+                                        367
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                355,
+                                                504,
+                                                367
+                                            ],
+                                            "score": 1.0,
+                                            "content": "(early/intermediate/late shown in blue/cyan/yellow). Each line shows the trajectory of a single high-",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 11
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        366,
+                                        505,
+                                        379
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                366,
+                                                505,
+                                                379
+                                            ],
+                                            "score": 1.0,
+                                            "content": "lighted neuron as its firing responses evolve during training. In panel a), we highlight the border",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 12
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        377,
+                                        506,
+                                        390
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                377,
+                                                506,
+                                                390
+                                            ],
+                                            "score": 1.0,
+                                            "content": "representation. It appears there are four clusters of border cells, each responding to one wall of a",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 13
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        388,
+                                        506,
+                                        401
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                388,
+                                                506,
+                                                401
+                                            ],
+                                            "score": 1.0,
+                                            "content": "square environment (spatial responses from four of these border cells are inset). These cells’ re-",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 14
+                                },
+                                {
+                                    "bbox": [
+                                        104,
+                                        399,
+                                        506,
+                                        413
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                104,
+                                                399,
+                                                506,
+                                                413
+                                            ],
+                                            "score": 1.0,
+                                            "content": "sponse profiles appear early and stably persist through training, illustrated by the short distance they",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 15
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        410,
+                                        505,
+                                        423
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                410,
+                                                505,
+                                                423
+                                            ],
+                                            "score": 1.0,
+                                            "content": "travel in this space. In b), we show that the neurons which eventually become grid cells initially have",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 16
+                                },
+                                {
+                                    "bbox": [
+                                        104,
+                                        420,
+                                        505,
+                                        436
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                104,
+                                                420,
+                                                505,
+                                                436
+                                            ],
+                                            "score": 1.0,
+                                            "content": "tuning profiles similar to the border cells but then change their tuning substantially during learning.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 17
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        431,
+                                        505,
+                                        444
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                431,
+                                                505,
+                                                444
+                                            ],
+                                            "score": 1.0,
+                                            "content": "As a natural consequence, they need to travel a long distance in this space between the early and late",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 18
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        444,
+                                        505,
+                                        455
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                444,
+                                                505,
+                                                455
+                                            ],
+                                            "score": 1.0,
+                                            "content": "phase of the training. Spatial responses are shown for four of these grid-like cells during the late",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 19
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        453,
+                                        178,
+                                        468
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                453,
+                                                178,
+                                                468
+                                            ],
+                                            "score": 1.0,
+                                            "content": "phase of training.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 20
+                                }
+                            ],
+                            "index": 13.5
+                        }
+                    ],
+                    "index": 9.25
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        484,
+                        299,
+                        495
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                484,
+                                302,
+                                497
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        484,
+                                        302,
+                                        497
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3.2 THE IMPORTANCE OF REGULARIZATION",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        }
+                    ],
+                    "index": 21
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        505,
+                        505,
+                        626
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                504,
+                                505,
+                                519
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        504,
+                                        505,
+                                        519
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We find appropriate regularizations of the RNN to be crucial for the emergence of grid-like repre-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                516,
+                                505,
+                                529
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        516,
+                                        505,
+                                        529
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sentations. We only observed grid-like representations when the network was encouraged to store",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                525,
+                                505,
+                                541
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        525,
+                                        505,
+                                        541
+                                    ],
+                                    "score": 1.0,
+                                    "content": "information while perturbed by noise. This was accomplished by setting the speed input to zero, e.g.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                538,
+                                505,
+                                551
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        538,
+                                        152,
+                                        551
+                                    ],
+                                    "score": 1.0,
+                                    "content": "zero speed",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        152,
+                                        538,
+                                        172,
+                                        549
+                                    ],
+                                    "score": 0.86,
+                                    "content": "90 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        172,
+                                        538,
+                                        398,
+                                        551
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of the time, and adding Gaussian noise to the network",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        399,
+                                        538,
+                                        419,
+                                        550
+                                    ],
+                                    "score": 0.9,
+                                    "content": "\\xi _ { i } ( t )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        420,
+                                        538,
+                                        505,
+                                        551
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in equation (1)); the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                549,
+                                505,
+                                562
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        549,
+                                        505,
+                                        562
+                                    ],
+                                    "score": 1.0,
+                                    "content": "precise method for setting the speed input to zero and the value of the noise variance is not crucial",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                560,
+                                505,
+                                573
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        560,
+                                        505,
+                                        573
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for our simulations to develop grid-like representations. The cost function which aims to capture",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                570,
+                                505,
+                                583
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        570,
+                                        505,
+                                        583
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the penalization on the metabolic cost of the neural activity also acts as an important regulariza-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                582,
+                                505,
+                                594
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        582,
+                                        505,
+                                        594
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tion. Our simulations show that the grid-like representation did not emerge without this metabolic",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                593,
+                                505,
+                                606
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        593,
+                                        505,
+                                        606
+                                    ],
+                                    "score": 1.0,
+                                    "content": "cost. In Figure 5, we show typical simulation results for a square environment, with and without",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                604,
+                                506,
+                                617
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        604,
+                                        506,
+                                        617
+                                    ],
+                                    "score": 1.0,
+                                    "content": "proper metabolic regularization. In the Appendix, we illustrate the effect of regularization further,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                615,
+                                343,
+                                628
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        615,
+                                        343,
+                                        628
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in particular the role of injecting noise into the RNN units.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        }
+                    ],
+                    "index": 27,
+                    "bbox_fs": [
+                        105,
+                        504,
+                        506,
+                        628
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        632,
+                        505,
+                        676
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                631,
+                                505,
+                                644
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        631,
+                                        505,
+                                        644
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Our results are consistent with the general notion on the importance of incorporating proper con-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                643,
+                                505,
+                                655
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        643,
+                                        505,
+                                        655
+                                    ],
+                                    "score": 1.0,
+                                    "content": "straint for learning useful representations in neural networks (Bengio et al., 2013). Furthermore, it",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                653,
+                                505,
+                                667
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        653,
+                                        505,
+                                        667
+                                    ],
+                                    "score": 1.0,
+                                    "content": "suggests that, to learn a model with response properties similar to neural systems it may be necessary",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                665,
+                                383,
+                                677
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        665,
+                                        383,
+                                        677
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to incorporate the relevant constraints, e.g., noise and metabolic cost.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        }
+                    ],
+                    "index": 34.5,
+                    "bbox_fs": [
+                        105,
+                        631,
+                        505,
+                        677
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        690,
+                        329,
+                        700
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                689,
+                                330,
+                                702
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        689,
+                                        330,
+                                        702
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3.3 ERROR CORRECTION AROUND THE BOUNDARY",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        }
+                    ],
+                    "index": 37
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        709,
+                        504,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                709,
+                                504,
+                                722
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        709,
+                                        504,
+                                        722
+                                    ],
+                                    "score": 1.0,
+                                    "content": "One natural question is whether the trained RNNs are able to perform localization when the path",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                720,
+                                505,
+                                733
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        720,
+                                        505,
+                                        733
+                                    ],
+                                    "score": 1.0,
+                                    "content": "length exceeds the typical length used during training (500 steps), in particular given that noise in",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                649,
+                                505,
+                                661
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        649,
+                                        505,
+                                        661
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the network would gradually accumulate, leading to a decrease in localization performance. We test",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                660,
+                                505,
+                                672
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        660,
+                                        505,
+                                        672
+                                    ],
+                                    "score": 1.0,
+                                    "content": "this by simulating paths that are several orders of magnitude longer. Somewhat surprisingly, we find",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                671,
+                                505,
+                                684
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        671,
+                                        505,
+                                        684
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the RNNs still perform well (Figure 6b). In fact, the squared error (averaged over every 10000 steps)",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                681,
+                                505,
+                                694
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        681,
+                                        505,
+                                        694
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is stable. The spatial response profiles of individual units also remain stable. This implies that the",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                692,
+                                408,
+                                706
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        692,
+                                        408,
+                                        706
+                                    ],
+                                    "score": 1.0,
+                                    "content": "RNNs have acquired intrinsic error-correction mechanisms during training.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 28
+                        }
+                    ],
+                    "index": 38.5,
+                    "bbox_fs": [
+                        106,
+                        709,
+                        505,
+                        733
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "image",
+                    "bbox": [
+                        113,
+                        79,
+                        486,
+                        264
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                113,
+                                79,
+                                486,
+                                264
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        113,
+                                        79,
+                                        486,
+                                        264
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                113,
+                                                79,
+                                                486,
+                                                264
+                                            ],
+                                            "score": 0.974,
+                                            "type": "image",
+                                            "image_path": "476435913a7b64388291eece485ed2e49a2bc8befbf4d07d7b0df7b3c97e20d4.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 1,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        113,
+                                        79,
+                                        486,
+                                        140.66666666666666
+                                    ],
+                                    "spans": [],
+                                    "index": 0
+                                },
+                                {
+                                    "bbox": [
+                                        113,
+                                        140.66666666666666,
+                                        486,
+                                        202.33333333333331
+                                    ],
+                                    "spans": [],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        113,
+                                        202.33333333333331,
+                                        486,
+                                        264.0
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                106,
+                                273,
+                                506,
+                                329
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        274,
+                                        505,
+                                        286
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                274,
+                                                505,
+                                                286
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 5: Complete set of spatial response profiles for 100 neurons in a RNN trained in a square",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 3
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        286,
+                                        505,
+                                        298
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                286,
+                                                505,
+                                                298
+                                            ],
+                                            "score": 1.0,
+                                            "content": "environment. a) Without proper regularization, complex and periodic spatial response patterns do",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 4
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        294,
+                                        506,
+                                        311
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                294,
+                                                506,
+                                                311
+                                            ],
+                                            "score": 1.0,
+                                            "content": "not emerge. b) With proper regularization, a rich set of periodic response patterns emerge, including",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 5
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        307,
+                                        505,
+                                        320
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                307,
+                                                505,
+                                                320
+                                            ],
+                                            "score": 1.0,
+                                            "content": "grid-like responses. Regularization can also be adjusted to achieve spatial profiles intermediate",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 6
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        318,
+                                        224,
+                                        330
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                318,
+                                                224,
+                                                330
+                                            ],
+                                            "score": 1.0,
+                                            "content": "between these two examples.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 7
+                                }
+                            ],
+                            "index": 5
+                        }
+                    ],
+                    "index": 3.0
+                },
+                {
+                    "type": "image",
+                    "bbox": [
+                        128,
+                        346,
+                        481,
+                        470
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                128,
+                                346,
+                                481,
+                                470
+                            ],
+                            "group_id": 1,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        128,
+                                        346,
+                                        481,
+                                        470
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                128,
+                                                346,
+                                                481,
+                                                470
+                                            ],
+                                            "score": 0.97,
+                                            "type": "image",
+                                            "image_path": "1313022c4e140eff6c754917cddc7b20c35899670d67ff7b351c56f3f49df4d8.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 9,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        128,
+                                        346,
+                                        481,
+                                        387.3333333333333
+                                    ],
+                                    "spans": [],
+                                    "index": 8
+                                },
+                                {
+                                    "bbox": [
+                                        128,
+                                        387.3333333333333,
+                                        481,
+                                        428.66666666666663
+                                    ],
+                                    "spans": [],
+                                    "index": 9
+                                },
+                                {
+                                    "bbox": [
+                                        128,
+                                        428.66666666666663,
+                                        481,
+                                        469.99999999999994
+                                    ],
+                                    "spans": [],
+                                    "index": 10
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                106,
+                                480,
+                                505,
+                                623
+                            ],
+                            "group_id": 1,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        479,
+                                        505,
+                                        493
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                479,
+                                                505,
+                                                493
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 6: Error-correction happens at the boundary and the error is stable over time. At the boundary,",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 11
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        490,
+                                        505,
+                                        504
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                490,
+                                                505,
+                                                504
+                                            ],
+                                            "score": 1.0,
+                                            "content": "the direction is resampled to avoid input velocities that lead to a path extending beyond the boundary",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 12
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        502,
+                                        505,
+                                        515
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                502,
+                                                505,
+                                                515
+                                            ],
+                                            "score": 1.0,
+                                            "content": "of the environment. These changing input statistics at the boundary, termed a boundary interaction,",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 13
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        511,
+                                        505,
+                                        525
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                511,
+                                                505,
+                                                525
+                                            ],
+                                            "score": 1.0,
+                                            "content": "are the only cue the RNN receives about the boundary. We find that the RNN uses the boundary",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 14
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        524,
+                                        505,
+                                        536
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                524,
+                                                505,
+                                                536
+                                            ],
+                                            "score": 1.0,
+                                            "content": "interactions to correct the accumulated error between the true integrated input and its prediction",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 15
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        534,
+                                        505,
+                                        547
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                534,
+                                                505,
+                                                547
+                                            ],
+                                            "score": 1.0,
+                                            "content": "based on the linear readout of equation (2). Panel a), the mean squared error increases when there",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 16
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        545,
+                                        505,
+                                        559
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                545,
+                                                505,
+                                                559
+                                            ],
+                                            "score": 1.0,
+                                            "content": "are no boundary interactions, but then decreases after a boundary interaction, with more boundary",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 17
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        556,
+                                        505,
+                                        569
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                556,
+                                                505,
+                                                569
+                                            ],
+                                            "score": 1.0,
+                                            "content": "interactions leading to greater error reduction. In the absence of further boundary interaction, the",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 18
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        568,
+                                        506,
+                                        580
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                568,
+                                                506,
+                                                580
+                                            ],
+                                            "score": 1.0,
+                                            "content": "squared error would gradually increase again (blue curve) at roughly a constant rate. b) The network",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 19
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        578,
+                                        506,
+                                        591
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                578,
+                                                506,
+                                                591
+                                            ],
+                                            "score": 1.0,
+                                            "content": "was trained using mini-batches of 500 timesteps but has stable error over a duration at least four",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 20
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        590,
+                                        505,
+                                        601
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                590,
+                                                505,
+                                                601
+                                            ],
+                                            "score": 1.0,
+                                            "content": "orders of magnitude larger. The error of the RNN output (mean and standard deviation shown in",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 21
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        600,
+                                        505,
+                                        613
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                600,
+                                                505,
+                                                613
+                                            ],
+                                            "score": 1.0,
+                                            "content": "black, computed based on 10000 timesteps) is compared to the error that would be achieved by an",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 22
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        611,
+                                        294,
+                                        624
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                611,
+                                                294,
+                                                624
+                                            ],
+                                            "score": 1.0,
+                                            "content": "RNN outputting the best constant values (red).",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 23
+                                }
+                            ],
+                            "index": 17
+                        }
+                    ],
+                    "index": 13.0
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        649,
+                        505,
+                        704
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                649,
+                                505,
+                                661
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        649,
+                                        505,
+                                        661
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the network would gradually accumulate, leading to a decrease in localization performance. We test",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                660,
+                                505,
+                                672
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        660,
+                                        505,
+                                        672
+                                    ],
+                                    "score": 1.0,
+                                    "content": "this by simulating paths that are several orders of magnitude longer. Somewhat surprisingly, we find",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                671,
+                                505,
+                                684
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        671,
+                                        505,
+                                        684
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the RNNs still perform well (Figure 6b). In fact, the squared error (averaged over every 10000 steps)",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                681,
+                                505,
+                                694
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        681,
+                                        505,
+                                        694
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is stable. The spatial response profiles of individual units also remain stable. This implies that the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                692,
+                                408,
+                                706
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        692,
+                                        408,
+                                        706
+                                    ],
+                                    "score": 1.0,
+                                    "content": "RNNs have acquired intrinsic error-correction mechanisms during training.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        }
+                    ],
+                    "index": 26
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        709,
+                        504,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                708,
+                                505,
+                                723
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        708,
+                                        505,
+                                        723
+                                    ],
+                                    "score": 1.0,
+                                    "content": "As shown earlier, during training some of the RNN units develop boundary-related firing (Figure 2c),",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                720,
+                                506,
+                                734
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        720,
+                                        506,
+                                        734
+                                    ],
+                                    "score": 1.0,
+                                    "content": "presumably by exploiting the change of input statistics around the boundary. We hypothesize that",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        }
+                    ],
+                    "index": 29.5
+                }
+            ],
+            "page_idx": 7,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        107,
+                        27,
+                        293,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                26,
+                                294,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        26,
+                                        294,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Published as a conference paper at ICLR 2018",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        302,
+                        752,
+                        308,
+                        759
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                302,
+                                750,
+                                309,
+                                761
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        302,
+                                        750,
+                                        309,
+                                        761
+                                    ],
+                                    "score": 1.0,
+                                    "content": "8",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "image",
+                    "bbox": [
+                        113,
+                        79,
+                        486,
+                        264
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                113,
+                                79,
+                                486,
+                                264
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        113,
+                                        79,
+                                        486,
+                                        264
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                113,
+                                                79,
+                                                486,
+                                                264
+                                            ],
+                                            "score": 0.974,
+                                            "type": "image",
+                                            "image_path": "476435913a7b64388291eece485ed2e49a2bc8befbf4d07d7b0df7b3c97e20d4.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 1,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        113,
+                                        79,
+                                        486,
+                                        140.66666666666666
+                                    ],
+                                    "spans": [],
+                                    "index": 0
+                                },
+                                {
+                                    "bbox": [
+                                        113,
+                                        140.66666666666666,
+                                        486,
+                                        202.33333333333331
+                                    ],
+                                    "spans": [],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        113,
+                                        202.33333333333331,
+                                        486,
+                                        264.0
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                106,
+                                273,
+                                506,
+                                329
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        274,
+                                        505,
+                                        286
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                274,
+                                                505,
+                                                286
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 5: Complete set of spatial response profiles for 100 neurons in a RNN trained in a square",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 3
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        286,
+                                        505,
+                                        298
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                286,
+                                                505,
+                                                298
+                                            ],
+                                            "score": 1.0,
+                                            "content": "environment. a) Without proper regularization, complex and periodic spatial response patterns do",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 4
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        294,
+                                        506,
+                                        311
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                294,
+                                                506,
+                                                311
+                                            ],
+                                            "score": 1.0,
+                                            "content": "not emerge. b) With proper regularization, a rich set of periodic response patterns emerge, including",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 5
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        307,
+                                        505,
+                                        320
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                307,
+                                                505,
+                                                320
+                                            ],
+                                            "score": 1.0,
+                                            "content": "grid-like responses. Regularization can also be adjusted to achieve spatial profiles intermediate",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 6
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        318,
+                                        224,
+                                        330
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                318,
+                                                224,
+                                                330
+                                            ],
+                                            "score": 1.0,
+                                            "content": "between these two examples.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 7
+                                }
+                            ],
+                            "index": 5
+                        }
+                    ],
+                    "index": 3.0
+                },
+                {
+                    "type": "image",
+                    "bbox": [
+                        128,
+                        346,
+                        481,
+                        470
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                128,
+                                346,
+                                481,
+                                470
+                            ],
+                            "group_id": 1,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        128,
+                                        346,
+                                        481,
+                                        470
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                128,
+                                                346,
+                                                481,
+                                                470
+                                            ],
+                                            "score": 0.97,
+                                            "type": "image",
+                                            "image_path": "1313022c4e140eff6c754917cddc7b20c35899670d67ff7b351c56f3f49df4d8.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 9,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        128,
+                                        346,
+                                        481,
+                                        387.3333333333333
+                                    ],
+                                    "spans": [],
+                                    "index": 8
+                                },
+                                {
+                                    "bbox": [
+                                        128,
+                                        387.3333333333333,
+                                        481,
+                                        428.66666666666663
+                                    ],
+                                    "spans": [],
+                                    "index": 9
+                                },
+                                {
+                                    "bbox": [
+                                        128,
+                                        428.66666666666663,
+                                        481,
+                                        469.99999999999994
+                                    ],
+                                    "spans": [],
+                                    "index": 10
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                106,
+                                480,
+                                505,
+                                623
+                            ],
+                            "group_id": 1,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        479,
+                                        505,
+                                        493
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                479,
+                                                505,
+                                                493
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 6: Error-correction happens at the boundary and the error is stable over time. At the boundary,",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 11
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        490,
+                                        505,
+                                        504
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                490,
+                                                505,
+                                                504
+                                            ],
+                                            "score": 1.0,
+                                            "content": "the direction is resampled to avoid input velocities that lead to a path extending beyond the boundary",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 12
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        502,
+                                        505,
+                                        515
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                502,
+                                                505,
+                                                515
+                                            ],
+                                            "score": 1.0,
+                                            "content": "of the environment. These changing input statistics at the boundary, termed a boundary interaction,",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 13
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        511,
+                                        505,
+                                        525
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                511,
+                                                505,
+                                                525
+                                            ],
+                                            "score": 1.0,
+                                            "content": "are the only cue the RNN receives about the boundary. We find that the RNN uses the boundary",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 14
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        524,
+                                        505,
+                                        536
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                524,
+                                                505,
+                                                536
+                                            ],
+                                            "score": 1.0,
+                                            "content": "interactions to correct the accumulated error between the true integrated input and its prediction",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 15
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        534,
+                                        505,
+                                        547
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                534,
+                                                505,
+                                                547
+                                            ],
+                                            "score": 1.0,
+                                            "content": "based on the linear readout of equation (2). Panel a), the mean squared error increases when there",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 16
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        545,
+                                        505,
+                                        559
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                545,
+                                                505,
+                                                559
+                                            ],
+                                            "score": 1.0,
+                                            "content": "are no boundary interactions, but then decreases after a boundary interaction, with more boundary",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 17
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        556,
+                                        505,
+                                        569
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                556,
+                                                505,
+                                                569
+                                            ],
+                                            "score": 1.0,
+                                            "content": "interactions leading to greater error reduction. In the absence of further boundary interaction, the",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 18
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        568,
+                                        506,
+                                        580
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                568,
+                                                506,
+                                                580
+                                            ],
+                                            "score": 1.0,
+                                            "content": "squared error would gradually increase again (blue curve) at roughly a constant rate. b) The network",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 19
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        578,
+                                        506,
+                                        591
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                578,
+                                                506,
+                                                591
+                                            ],
+                                            "score": 1.0,
+                                            "content": "was trained using mini-batches of 500 timesteps but has stable error over a duration at least four",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 20
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        590,
+                                        505,
+                                        601
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                590,
+                                                505,
+                                                601
+                                            ],
+                                            "score": 1.0,
+                                            "content": "orders of magnitude larger. The error of the RNN output (mean and standard deviation shown in",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 21
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        600,
+                                        505,
+                                        613
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                600,
+                                                505,
+                                                613
+                                            ],
+                                            "score": 1.0,
+                                            "content": "black, computed based on 10000 timesteps) is compared to the error that would be achieved by an",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 22
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        611,
+                                        294,
+                                        624
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                611,
+                                                294,
+                                                624
+                                            ],
+                                            "score": 1.0,
+                                            "content": "RNN outputting the best constant values (red).",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 23
+                                }
+                            ],
+                            "index": 17
+                        }
+                    ],
+                    "index": 13.0
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        649,
+                        505,
+                        704
+                    ],
+                    "lines": [],
+                    "index": 26,
+                    "bbox_fs": [
+                        105,
+                        649,
+                        505,
+                        706
+                    ],
+                    "lines_deleted": true
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        709,
+                        504,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                708,
+                                505,
+                                723
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        708,
+                                        505,
+                                        723
+                                    ],
+                                    "score": 1.0,
+                                    "content": "As shown earlier, during training some of the RNN units develop boundary-related firing (Figure 2c),",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                720,
+                                506,
+                                734
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        720,
+                                        506,
+                                        734
+                                    ],
+                                    "score": 1.0,
+                                    "content": "presumably by exploiting the change of input statistics around the boundary. We hypothesize that",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                82,
+                                505,
+                                94
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        82,
+                                        505,
+                                        94
+                                    ],
+                                    "score": 1.0,
+                                    "content": "boundary interactions may enable error-correction through signals based on these boundary-related",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                94,
+                                505,
+                                106
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        94,
+                                        505,
+                                        106
+                                    ],
+                                    "score": 1.0,
+                                    "content": "activities. Indeed, we find that boundary interactions can dramatically reduce the accumulated er-",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                104,
+                                505,
+                                118
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        104,
+                                        505,
+                                        118
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ror (Figure 6a). Figure 6a shows that, without boundary interactions, on average the squared error",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                115,
+                                506,
+                                129
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        115,
+                                        506,
+                                        129
+                                    ],
+                                    "score": 1.0,
+                                    "content": "grows roughly linearly as expected, however, interactions with the boundaries substantially reduce",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                127,
+                                505,
+                                138
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        127,
+                                        505,
+                                        138
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the error, and more frequent boundary interactions can reduce the error further. Error-correction on",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                137,
+                                506,
+                                150
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        137,
+                                        506,
+                                        150
+                                    ],
+                                    "score": 1.0,
+                                    "content": "grid cells via boundary interactions has been proposed (Hardcastle et al., 2015; Pollock et al., 2017),",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                147,
+                                505,
+                                162
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        147,
+                                        505,
+                                        162
+                                    ],
+                                    "score": 1.0,
+                                    "content": "however, we emphasize that the model proposed here develops the grid-like responses, boundary",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                159,
+                                505,
+                                172
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        159,
+                                        505,
+                                        172
+                                    ],
+                                    "score": 1.0,
+                                    "content": "responses and the error-correction mechanisms all within the same neural network, thus potentially",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                171,
+                                348,
+                                183
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        171,
+                                        348,
+                                        183
+                                    ],
+                                    "score": 1.0,
+                                    "content": "providing a unifying account of a diverse set of phenomena.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 8
+                        }
+                    ],
+                    "index": 29.5,
+                    "bbox_fs": [
+                        105,
+                        708,
+                        506,
+                        734
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        82,
+                        505,
+                        182
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                82,
+                                505,
+                                94
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        82,
+                                        505,
+                                        94
+                                    ],
+                                    "score": 1.0,
+                                    "content": "boundary interactions may enable error-correction through signals based on these boundary-related",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                94,
+                                505,
+                                106
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        94,
+                                        505,
+                                        106
+                                    ],
+                                    "score": 1.0,
+                                    "content": "activities. Indeed, we find that boundary interactions can dramatically reduce the accumulated er-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                104,
+                                505,
+                                118
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        104,
+                                        505,
+                                        118
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ror (Figure 6a). Figure 6a shows that, without boundary interactions, on average the squared error",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                115,
+                                506,
+                                129
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        115,
+                                        506,
+                                        129
+                                    ],
+                                    "score": 1.0,
+                                    "content": "grows roughly linearly as expected, however, interactions with the boundaries substantially reduce",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                127,
+                                505,
+                                138
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        127,
+                                        505,
+                                        138
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the error, and more frequent boundary interactions can reduce the error further. Error-correction on",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                137,
+                                506,
+                                150
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        137,
+                                        506,
+                                        150
+                                    ],
+                                    "score": 1.0,
+                                    "content": "grid cells via boundary interactions has been proposed (Hardcastle et al., 2015; Pollock et al., 2017),",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                147,
+                                505,
+                                162
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        147,
+                                        505,
+                                        162
+                                    ],
+                                    "score": 1.0,
+                                    "content": "however, we emphasize that the model proposed here develops the grid-like responses, boundary",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                159,
+                                505,
+                                172
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        159,
+                                        505,
+                                        172
+                                    ],
+                                    "score": 1.0,
+                                    "content": "responses and the error-correction mechanisms all within the same neural network, thus potentially",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                171,
+                                348,
+                                183
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        171,
+                                        348,
+                                        183
+                                    ],
+                                    "score": 1.0,
+                                    "content": "providing a unifying account of a diverse set of phenomena.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        }
+                    ],
+                    "index": 4
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        203,
+                        190,
+                        216
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                201,
+                                192,
+                                218
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        201,
+                                        192,
+                                        218
+                                    ],
+                                    "score": 1.0,
+                                    "content": "4 DISCUSSION",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        }
+                    ],
+                    "index": 9
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        231,
+                        505,
+                        330
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                232,
+                                505,
+                                244
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        232,
+                                        505,
+                                        244
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In this paper, we trained RNNs to perform path integration (dead-reckoning) in 2D arenas. We",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                242,
+                                505,
+                                255
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        242,
+                                        505,
+                                        255
+                                    ],
+                                    "score": 1.0,
+                                    "content": "found that after training RNNs with appropriate regularization, the model neurons exhibit a variety",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                253,
+                                505,
+                                266
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        253,
+                                        505,
+                                        266
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of spatial and velocity tuning profiles that match neurophysiology in EC. What’s more, there is",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                264,
+                                505,
+                                277
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        264,
+                                        505,
+                                        277
+                                    ],
+                                    "score": 1.0,
+                                    "content": "also similarity in terms of when these distinct neuron types emerge during training/development.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                273,
+                                506,
+                                289
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        273,
+                                        506,
+                                        289
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The EC has long been thought to be involved in path integration and localization of the animal’s",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                285,
+                                505,
+                                299
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        285,
+                                        505,
+                                        299
+                                    ],
+                                    "score": 1.0,
+                                    "content": "location (Moser et al., 2008). The general agreement between the different response properties",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                297,
+                                505,
+                                310
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        297,
+                                        505,
+                                        310
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in our model and the neurophysiology provide strong evidence supporting the hypothesis that the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                308,
+                                505,
+                                321
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        308,
+                                        505,
+                                        321
+                                    ],
+                                    "score": 1.0,
+                                    "content": "neural population in EC may provide an efficient code for representation self-locations based on the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                319,
+                                166,
+                                331
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        319,
+                                        166,
+                                        331
+                                    ],
+                                    "score": 1.0,
+                                    "content": "velocity input.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        }
+                    ],
+                    "index": 14
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        336,
+                        505,
+                        413
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                336,
+                                505,
+                                348
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        336,
+                                        505,
+                                        348
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Recently, there has been increased interest in using complex neural network models to understand",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                347,
+                                505,
+                                360
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        347,
+                                        505,
+                                        360
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the neural code. But the focus has been on using feedforward architectures, in particular CNNs (Le-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                357,
+                                505,
+                                370
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        357,
+                                        505,
+                                        370
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Cun et al., 1998). Given the abundant recurrent connections in the brain, it seems a particularly",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                369,
+                                505,
+                                381
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        369,
+                                        505,
+                                        381
+                                    ],
+                                    "score": 1.0,
+                                    "content": "fruitful avenue to take advantage of the recent development in RNNs to help with neuroscience",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                380,
+                                505,
+                                392
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        380,
+                                        505,
+                                        392
+                                    ],
+                                    "score": 1.0,
+                                    "content": "questions (Mante et al., 2013; Song et al., 2016; Miconi, 2017; Sussillo et al., 2015). Here, we only",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                392,
+                                505,
+                                402
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        392,
+                                        505,
+                                        402
+                                    ],
+                                    "score": 1.0,
+                                    "content": "show one instance following this approach. However, the insight from this work could be general,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                402,
+                                344,
+                                415
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        402,
+                                        344,
+                                        415
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and potentially useful for other cognitive functions as well.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        }
+                    ],
+                    "index": 22
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        419,
+                        505,
+                        637
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                418,
+                                505,
+                                432
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        418,
+                                        505,
+                                        432
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The finding that metabolic constraints lead to the emergence of grid-like responses may be seen",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                430,
+                                505,
+                                442
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        430,
+                                        505,
+                                        442
+                                    ],
+                                    "score": 1.0,
+                                    "content": "as conceptually related to the efficient coding hypothesis in visual processing (Barlow, 1961), in",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                441,
+                                505,
+                                453
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        441,
+                                        505,
+                                        453
+                                    ],
+                                    "score": 1.0,
+                                    "content": "particular the seminal work on the emergence of the V1-like Gabor filters in a sparse coding model",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                451,
+                                506,
+                                465
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        451,
+                                        506,
+                                        465
+                                    ],
+                                    "score": 1.0,
+                                    "content": "by Olshausen & Field (1996). Indeed, our work is partly inspired by these results. While there",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                462,
+                                506,
+                                476
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        462,
+                                        506,
+                                        476
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are conceptual similarities, however, we should also note there are differences between the sparse",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                474,
+                                505,
+                                486
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        474,
+                                        505,
+                                        486
+                                    ],
+                                    "score": 1.0,
+                                    "content": "coding work and ours. First, the sparsity constraint in sparse coding can be naturally viewed as",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                484,
+                                505,
+                                498
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        484,
+                                        505,
+                                        498
+                                    ],
+                                    "score": 1.0,
+                                    "content": "a particular prior while in the context of the recurrent network, it is difficult to interpret that way.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                496,
+                                506,
+                                508
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        496,
+                                        506,
+                                        508
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Second, the grid-like responses are not the most sparse solution one could imagine. In fact, they are",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                506,
+                                505,
+                                519
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        506,
+                                        505,
+                                        519
+                                    ],
+                                    "score": 1.0,
+                                    "content": "still quite dense compared to a more spatially localized representation. Third, the grid-like patterns",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                517,
+                                506,
+                                530
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        517,
+                                        506,
+                                        530
+                                    ],
+                                    "score": 1.0,
+                                    "content": "that emerged in our network are not filters based on the raw input, rather the velocity inputs need to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                528,
+                                505,
+                                541
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        528,
+                                        505,
+                                        541
+                                    ],
+                                    "score": 1.0,
+                                    "content": "be integrated first in order to encode spatial locations. Our work is also inspired by recent work using",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                539,
+                                505,
+                                552
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        539,
+                                        505,
+                                        552
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the efficient coding idea to explain the functional architecture of the grid cells (Wei et al., 2015). It",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                549,
+                                505,
+                                563
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        549,
+                                        505,
+                                        563
+                                    ],
+                                    "score": 1.0,
+                                    "content": "has been shown that efficient coding considerations could explain the particular set of grid scales",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                560,
+                                506,
+                                574
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        560,
+                                        506,
+                                        574
+                                    ],
+                                    "score": 1.0,
+                                    "content": "observed in rodents (Stensola et al., 2012). However, in that work, the firing patterns of the neurons",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                572,
+                                505,
+                                585
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        572,
+                                        505,
+                                        585
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are assumed to have a lattice structure to start with. Furthermore, our work is related to the study by",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                582,
+                                505,
+                                595
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        582,
+                                        505,
+                                        595
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Sussillo and others (Sussillo et al., 2015), in which they show that regularization of RNN models are",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                594,
+                                505,
+                                606
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        594,
+                                        505,
+                                        606
+                                    ],
+                                    "score": 1.0,
+                                    "content": "important for generating solutions that are similar to the neural activity observed in motor cortex.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                605,
+                                505,
+                                617
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        605,
+                                        505,
+                                        617
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In Sussillo et al., a smoothness constraint together with others lead to simple oscillatory neural",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                616,
+                                505,
+                                628
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        616,
+                                        505,
+                                        628
+                                    ],
+                                    "score": 1.0,
+                                    "content": "dynamics that well matches the neural data. We have not incorporated a smoothness constraint into",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                628,
+                                160,
+                                638
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        628,
+                                        160,
+                                        638
+                                    ],
+                                    "score": 1.0,
+                                    "content": "our network.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        }
+                    ],
+                    "index": 35.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        644,
+                        504,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                643,
+                                506,
+                                656
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        643,
+                                        506,
+                                        656
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Additionally, we note that there are a few recent studies which use place cells as the input to generate",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                654,
+                                506,
+                                667
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        654,
+                                        506,
+                                        667
+                                    ],
+                                    "score": 1.0,
+                                    "content": "grid cells (Dordek et al., 2016; Stachenfeld et al., 2016), which are fundamentally different from",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 47
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                665,
+                                505,
+                                678
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        665,
+                                        505,
+                                        678
+                                    ],
+                                    "score": 1.0,
+                                    "content": "our work. In these feedforward network models, the grid cells essentially perform dimensionality",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 48
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                676,
+                                506,
+                                689
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        676,
+                                        506,
+                                        689
+                                    ],
+                                    "score": 1.0,
+                                    "content": "reduction based on the spatial input from place cells. However, the main issue with these models",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 49
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                688,
+                                506,
+                                700
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        688,
+                                        506,
+                                        700
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is that, it is unclear how place cells acquire spatial tuning in the first place. To the contrary, our",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 50
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                698,
+                                506,
+                                712
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        698,
+                                        506,
+                                        712
+                                    ],
+                                    "score": 1.0,
+                                    "content": "model takes the animal’s velocity as the input, and addresses the question of how the spatial tuning",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 51
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                710,
+                                506,
+                                722
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        710,
+                                        506,
+                                        722
+                                    ],
+                                    "score": 1.0,
+                                    "content": "can be generated from such input, which are known to exist in EC (Sargolini et al., 2006; Kropff",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 52
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                721,
+                                505,
+                                732
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        721,
+                                        505,
+                                        732
+                                    ],
+                                    "score": 1.0,
+                                    "content": "et al., 2015). In another related study (Kanitscheider & Fiete, 2016), the authors train a RNN with",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 53
+                        }
+                    ],
+                    "index": 49.5
+                }
+            ],
+            "page_idx": 8,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        108,
+                        27,
+                        293,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                26,
+                                294,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        26,
+                                        294,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Published as a conference paper at ICLR 2018",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        302,
+                        751,
+                        308,
+                        759
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                302,
+                                751,
+                                309,
+                                762
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        302,
+                                        751,
+                                        309,
+                                        762
+                                    ],
+                                    "score": 1.0,
+                                    "content": "9",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        82,
+                        505,
+                        182
+                    ],
+                    "lines": [],
+                    "index": 4,
+                    "bbox_fs": [
+                        105,
+                        82,
+                        506,
+                        183
+                    ],
+                    "lines_deleted": true
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        203,
+                        190,
+                        216
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                201,
+                                192,
+                                218
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        201,
+                                        192,
+                                        218
+                                    ],
+                                    "score": 1.0,
+                                    "content": "4 DISCUSSION",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        }
+                    ],
+                    "index": 9
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        231,
+                        505,
+                        330
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                232,
+                                505,
+                                244
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        232,
+                                        505,
+                                        244
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In this paper, we trained RNNs to perform path integration (dead-reckoning) in 2D arenas. We",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                242,
+                                505,
+                                255
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        242,
+                                        505,
+                                        255
+                                    ],
+                                    "score": 1.0,
+                                    "content": "found that after training RNNs with appropriate regularization, the model neurons exhibit a variety",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                253,
+                                505,
+                                266
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        253,
+                                        505,
+                                        266
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of spatial and velocity tuning profiles that match neurophysiology in EC. What’s more, there is",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                264,
+                                505,
+                                277
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        264,
+                                        505,
+                                        277
+                                    ],
+                                    "score": 1.0,
+                                    "content": "also similarity in terms of when these distinct neuron types emerge during training/development.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                273,
+                                506,
+                                289
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        273,
+                                        506,
+                                        289
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The EC has long been thought to be involved in path integration and localization of the animal’s",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                285,
+                                505,
+                                299
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        285,
+                                        505,
+                                        299
+                                    ],
+                                    "score": 1.0,
+                                    "content": "location (Moser et al., 2008). The general agreement between the different response properties",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                297,
+                                505,
+                                310
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        297,
+                                        505,
+                                        310
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in our model and the neurophysiology provide strong evidence supporting the hypothesis that the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                308,
+                                505,
+                                321
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        308,
+                                        505,
+                                        321
+                                    ],
+                                    "score": 1.0,
+                                    "content": "neural population in EC may provide an efficient code for representation self-locations based on the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                319,
+                                166,
+                                331
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        319,
+                                        166,
+                                        331
+                                    ],
+                                    "score": 1.0,
+                                    "content": "velocity input.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        }
+                    ],
+                    "index": 14,
+                    "bbox_fs": [
+                        105,
+                        232,
+                        506,
+                        331
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        336,
+                        505,
+                        413
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                336,
+                                505,
+                                348
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        336,
+                                        505,
+                                        348
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Recently, there has been increased interest in using complex neural network models to understand",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                347,
+                                505,
+                                360
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        347,
+                                        505,
+                                        360
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the neural code. But the focus has been on using feedforward architectures, in particular CNNs (Le-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                357,
+                                505,
+                                370
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        357,
+                                        505,
+                                        370
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Cun et al., 1998). Given the abundant recurrent connections in the brain, it seems a particularly",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                369,
+                                505,
+                                381
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        369,
+                                        505,
+                                        381
+                                    ],
+                                    "score": 1.0,
+                                    "content": "fruitful avenue to take advantage of the recent development in RNNs to help with neuroscience",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                380,
+                                505,
+                                392
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        380,
+                                        505,
+                                        392
+                                    ],
+                                    "score": 1.0,
+                                    "content": "questions (Mante et al., 2013; Song et al., 2016; Miconi, 2017; Sussillo et al., 2015). Here, we only",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                392,
+                                505,
+                                402
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        392,
+                                        505,
+                                        402
+                                    ],
+                                    "score": 1.0,
+                                    "content": "show one instance following this approach. However, the insight from this work could be general,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                402,
+                                344,
+                                415
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        402,
+                                        344,
+                                        415
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and potentially useful for other cognitive functions as well.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        }
+                    ],
+                    "index": 22,
+                    "bbox_fs": [
+                        105,
+                        336,
+                        505,
+                        415
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        419,
+                        505,
+                        637
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                418,
+                                505,
+                                432
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        418,
+                                        505,
+                                        432
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The finding that metabolic constraints lead to the emergence of grid-like responses may be seen",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                430,
+                                505,
+                                442
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        430,
+                                        505,
+                                        442
+                                    ],
+                                    "score": 1.0,
+                                    "content": "as conceptually related to the efficient coding hypothesis in visual processing (Barlow, 1961), in",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                441,
+                                505,
+                                453
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        441,
+                                        505,
+                                        453
+                                    ],
+                                    "score": 1.0,
+                                    "content": "particular the seminal work on the emergence of the V1-like Gabor filters in a sparse coding model",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                451,
+                                506,
+                                465
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        451,
+                                        506,
+                                        465
+                                    ],
+                                    "score": 1.0,
+                                    "content": "by Olshausen & Field (1996). Indeed, our work is partly inspired by these results. While there",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                462,
+                                506,
+                                476
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        462,
+                                        506,
+                                        476
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are conceptual similarities, however, we should also note there are differences between the sparse",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                474,
+                                505,
+                                486
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        474,
+                                        505,
+                                        486
+                                    ],
+                                    "score": 1.0,
+                                    "content": "coding work and ours. First, the sparsity constraint in sparse coding can be naturally viewed as",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                484,
+                                505,
+                                498
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        484,
+                                        505,
+                                        498
+                                    ],
+                                    "score": 1.0,
+                                    "content": "a particular prior while in the context of the recurrent network, it is difficult to interpret that way.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                496,
+                                506,
+                                508
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        496,
+                                        506,
+                                        508
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Second, the grid-like responses are not the most sparse solution one could imagine. In fact, they are",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                506,
+                                505,
+                                519
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        506,
+                                        505,
+                                        519
+                                    ],
+                                    "score": 1.0,
+                                    "content": "still quite dense compared to a more spatially localized representation. Third, the grid-like patterns",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                517,
+                                506,
+                                530
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        517,
+                                        506,
+                                        530
+                                    ],
+                                    "score": 1.0,
+                                    "content": "that emerged in our network are not filters based on the raw input, rather the velocity inputs need to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                528,
+                                505,
+                                541
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        528,
+                                        505,
+                                        541
+                                    ],
+                                    "score": 1.0,
+                                    "content": "be integrated first in order to encode spatial locations. Our work is also inspired by recent work using",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                539,
+                                505,
+                                552
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        539,
+                                        505,
+                                        552
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the efficient coding idea to explain the functional architecture of the grid cells (Wei et al., 2015). It",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                549,
+                                505,
+                                563
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        549,
+                                        505,
+                                        563
+                                    ],
+                                    "score": 1.0,
+                                    "content": "has been shown that efficient coding considerations could explain the particular set of grid scales",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                560,
+                                506,
+                                574
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        560,
+                                        506,
+                                        574
+                                    ],
+                                    "score": 1.0,
+                                    "content": "observed in rodents (Stensola et al., 2012). However, in that work, the firing patterns of the neurons",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                572,
+                                505,
+                                585
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        572,
+                                        505,
+                                        585
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are assumed to have a lattice structure to start with. Furthermore, our work is related to the study by",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                582,
+                                505,
+                                595
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        582,
+                                        505,
+                                        595
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Sussillo and others (Sussillo et al., 2015), in which they show that regularization of RNN models are",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                594,
+                                505,
+                                606
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        594,
+                                        505,
+                                        606
+                                    ],
+                                    "score": 1.0,
+                                    "content": "important for generating solutions that are similar to the neural activity observed in motor cortex.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                605,
+                                505,
+                                617
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        605,
+                                        505,
+                                        617
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In Sussillo et al., a smoothness constraint together with others lead to simple oscillatory neural",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                616,
+                                505,
+                                628
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        616,
+                                        505,
+                                        628
+                                    ],
+                                    "score": 1.0,
+                                    "content": "dynamics that well matches the neural data. We have not incorporated a smoothness constraint into",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                628,
+                                160,
+                                638
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        628,
+                                        160,
+                                        638
+                                    ],
+                                    "score": 1.0,
+                                    "content": "our network.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        }
+                    ],
+                    "index": 35.5,
+                    "bbox_fs": [
+                        104,
+                        418,
+                        506,
+                        638
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        644,
+                        504,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                643,
+                                506,
+                                656
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        643,
+                                        506,
+                                        656
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Additionally, we note that there are a few recent studies which use place cells as the input to generate",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                654,
+                                506,
+                                667
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        654,
+                                        506,
+                                        667
+                                    ],
+                                    "score": 1.0,
+                                    "content": "grid cells (Dordek et al., 2016; Stachenfeld et al., 2016), which are fundamentally different from",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 47
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                665,
+                                505,
+                                678
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        665,
+                                        505,
+                                        678
+                                    ],
+                                    "score": 1.0,
+                                    "content": "our work. In these feedforward network models, the grid cells essentially perform dimensionality",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 48
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                676,
+                                506,
+                                689
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        676,
+                                        506,
+                                        689
+                                    ],
+                                    "score": 1.0,
+                                    "content": "reduction based on the spatial input from place cells. However, the main issue with these models",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 49
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                688,
+                                506,
+                                700
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        688,
+                                        506,
+                                        700
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is that, it is unclear how place cells acquire spatial tuning in the first place. To the contrary, our",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 50
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                698,
+                                506,
+                                712
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        698,
+                                        506,
+                                        712
+                                    ],
+                                    "score": 1.0,
+                                    "content": "model takes the animal’s velocity as the input, and addresses the question of how the spatial tuning",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 51
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                710,
+                                506,
+                                722
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        710,
+                                        506,
+                                        722
+                                    ],
+                                    "score": 1.0,
+                                    "content": "can be generated from such input, which are known to exist in EC (Sargolini et al., 2006; Kropff",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 52
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                721,
+                                505,
+                                732
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        721,
+                                        505,
+                                        732
+                                    ],
+                                    "score": 1.0,
+                                    "content": "et al., 2015). In another related study (Kanitscheider & Fiete, 2016), the authors train a RNN with",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 53
+                        }
+                    ],
+                    "index": 49.5,
+                    "bbox_fs": [
+                        105,
+                        643,
+                        506,
+                        732
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        82,
+                        504,
+                        105
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                81,
+                                505,
+                                96
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        81,
+                                        505,
+                                        96
+                                    ],
+                                    "score": 1.0,
+                                    "content": "LSTM units (Hochreiter & Schmidhuber, 1997) to perform different navigation tasks. However, no",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                93,
+                                282,
+                                106
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        93,
+                                        282,
+                                        106
+                                    ],
+                                    "score": 1.0,
+                                    "content": "grid-like spatial firing patterns are reported.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        }
+                    ],
+                    "index": 0.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        111,
+                        505,
+                        297
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                110,
+                                505,
+                                123
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        110,
+                                        505,
+                                        123
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Although our model shows a qualitative match to the neural responses observed in the EC, nonethe-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                122,
+                                505,
+                                133
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        122,
+                                        505,
+                                        133
+                                    ],
+                                    "score": 1.0,
+                                    "content": "less it has several major limitations, with each offering interesting future research directions. First,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                132,
+                                506,
+                                145
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        132,
+                                        506,
+                                        145
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the learning rule we use seems to be biologically implausible. We are interested in exploring how a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                143,
+                                505,
+                                156
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        143,
+                                        505,
+                                        156
+                                    ],
+                                    "score": 1.0,
+                                    "content": "more biologically plausible learning rule could give rise to similar results (Lillicrap et al., 2016; Mi-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                154,
+                                506,
+                                167
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        154,
+                                        506,
+                                        167
+                                    ],
+                                    "score": 1.0,
+                                    "content": "coni, 2017; Guerguiev et al., 2017). Second, the simulation results do not show a variety of spatial",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                165,
+                                506,
+                                178
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        165,
+                                        506,
+                                        178
+                                    ],
+                                    "score": 1.0,
+                                    "content": "scales in grid-like cells. Experimentally, it is known that grid cells have multiple spatial scales, that",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                176,
+                                506,
+                                189
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        176,
+                                        506,
+                                        189
+                                    ],
+                                    "score": 1.0,
+                                    "content": "scale geometrically with a ratio 1.4 (Stensola et al., 2012), and this particular scale ratio is predicted",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                187,
+                                505,
+                                200
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        187,
+                                        505,
+                                        200
+                                    ],
+                                    "score": 1.0,
+                                    "content": "by efficient coding of space (Wei et al., 2015). We are investigating how to modify the model to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                198,
+                                506,
+                                212
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        198,
+                                        506,
+                                        212
+                                    ],
+                                    "score": 1.0,
+                                    "content": "get a hierarchy of spatial scales, perhaps by incorporating more neurons or modifying the regular-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                209,
+                                505,
+                                222
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        209,
+                                        505,
+                                        222
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ization. Last but not least, we have focused on the representation produced by the trained RNN.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                220,
+                                506,
+                                233
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        220,
+                                        506,
+                                        233
+                                    ],
+                                    "score": 1.0,
+                                    "content": "An equally important set of questions concern how the networks actually support the generation of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                230,
+                                505,
+                                243
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        230,
+                                        505,
+                                        243
+                                    ],
+                                    "score": 1.0,
+                                    "content": "such a representation. As a preliminary effort, we have examined the connectivity patterns of the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                242,
+                                506,
+                                255
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        242,
+                                        506,
+                                        255
+                                    ],
+                                    "score": 1.0,
+                                    "content": "trained network, and they do not seem to resemble the connectivity patterns required by standard",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                253,
+                                506,
+                                266
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        253,
+                                        506,
+                                        266
+                                    ],
+                                    "score": 1.0,
+                                    "content": "attractor network models. Maybe this should not be seen as too surprising. After all, the trained",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                264,
+                                505,
+                                277
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        264,
+                                        505,
+                                        277
+                                    ],
+                                    "score": 1.0,
+                                    "content": "networks can produce a diverse set of neural responses, while the previous models only led to grid",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                275,
+                                505,
+                                287
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        275,
+                                        505,
+                                        287
+                                    ],
+                                    "score": 1.0,
+                                    "content": "responses. It would be interesting for future work to systematically examine the questions related to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                286,
+                                221,
+                                298
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        286,
+                                        221,
+                                        298
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the underlying mechanisms.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        }
+                    ],
+                    "index": 10
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        314,
+                        244,
+                        326
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                313,
+                                245,
+                                328
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        313,
+                                        245,
+                                        328
+                                    ],
+                                    "score": 1.0,
+                                    "content": "5 ACKNOWLEDGEMENTS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        }
+                    ],
+                    "index": 19
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        338,
+                        503,
+                        372
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                338,
+                                505,
+                                350
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        338,
+                                        505,
+                                        350
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We thank members of the Center for Theoretical Neuroscience at Columbia University for useful",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                349,
+                                505,
+                                362
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        349,
+                                        505,
+                                        362
+                                    ],
+                                    "score": 1.0,
+                                    "content": "discussions and three anonymous reviewers for constructive feedback. Research supported by NSF",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                359,
+                                425,
+                                374
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        359,
+                                        425,
+                                        374
+                                    ],
+                                    "score": 1.0,
+                                    "content": "NeuroNex Award DBI-1707398 and NIH training grant 5T32NS064929 (CJC).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        }
+                    ],
+                    "index": 21
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        388,
+                        175,
+                        400
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                388,
+                                176,
+                                402
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        388,
+                                        176,
+                                        402
+                                    ],
+                                    "score": 1.0,
+                                    "content": "REFERENCES",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        }
+                    ],
+                    "index": 23
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        407,
+                        504,
+                        429
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                407,
+                                505,
+                                419
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        407,
+                                        505,
+                                        419
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Dmitriy Aronov, Rhino Nevers, and David W Tank. Mapping of a non-spatial dimension by the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                417,
+                                300,
+                                430
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        417,
+                                        300,
+                                        430
+                                    ],
+                                    "score": 1.0,
+                                    "content": "hippocampalentorhinal circuit. Nature, 2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        }
+                    ],
+                    "index": 24.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        437,
+                        505,
+                        460
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                435,
+                                505,
+                                451
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        435,
+                                        505,
+                                        451
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Horace B Barlow. Possible principles underlying the transformation of sensory messages. Sensory",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                448,
+                                262,
+                                459
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        448,
+                                        262,
+                                        459
+                                    ],
+                                    "score": 1.0,
+                                    "content": "communication, pp. 217–234, 1961.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        }
+                    ],
+                    "index": 26.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        466,
+                        504,
+                        500
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                466,
+                                505,
+                                479
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        466,
+                                        505,
+                                        479
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Yoshua Bengio, Aaron Courville, and Pascal Vincent. Representation learning: A review and new",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                478,
+                                506,
+                                491
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        478,
+                                        506,
+                                        491
+                                    ],
+                                    "score": 1.0,
+                                    "content": "perspectives. IEEE transactions on pattern analysis and machine intelligence, 35(8):1798–1828,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                488,
+                                144,
+                                501
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        488,
+                                        144,
+                                        501
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2013.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        }
+                    ],
+                    "index": 29
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        507,
+                        505,
+                        531
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                508,
+                                505,
+                                521
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        508,
+                                        505,
+                                        521
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Tale L Bjerknes, Edvard I Moser, and May-Britt Moser. Representation of geometric borders in the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                518,
+                                294,
+                                531
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        518,
+                                        294,
+                                        531
+                                    ],
+                                    "score": 1.0,
+                                    "content": "developing rat. Neuron, 82(1):71–78, 2014.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        }
+                    ],
+                    "index": 31.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        537,
+                        504,
+                        561
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                537,
+                                506,
+                                550
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        537,
+                                        506,
+                                        550
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Yoram Burak and Ila R Fiete. Accurate path integration in continuous attractor network models of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                549,
+                                366,
+                                561
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        549,
+                                        366,
+                                        561
+                                    ],
+                                    "score": 1.0,
+                                    "content": "grid cells. PLoS computational biology, 5(2):e1000291, 2009.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        }
+                    ],
+                    "index": 33.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        109,
+                        567,
+                        504,
+                        601
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                107,
+                                567,
+                                505,
+                                581
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        567,
+                                        505,
+                                        581
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Jonathan J Couey, Aree Witoelar, Sheng-Jia Zhang, Kang Zheng, Jing Ye, Benjamin Dunn, Rafal",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                578,
+                                505,
+                                592
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        578,
+                                        505,
+                                        592
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Czajkowski, May-Britt Moser, Edvard I Moser, Yasser Roudi, et al. Recurrent inhibitory circuitry",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                590,
+                                437,
+                                602
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        590,
+                                        437,
+                                        602
+                                    ],
+                                    "score": 1.0,
+                                    "content": "as a mechanism for grid formation. Nature neuroscience, 16(3):318–324, 2013.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        }
+                    ],
+                    "index": 36
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        608,
+                        387,
+                        621
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                608,
+                                388,
+                                622
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        608,
+                                        388,
+                                        622
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Charles Darwin. Origin of certain instincts. Nature, 7:417–418, 1873.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        }
+                    ],
+                    "index": 38
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        109,
+                        627,
+                        504,
+                        661
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                107,
+                                627,
+                                504,
+                                640
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        627,
+                                        504,
+                                        640
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Geoffrey W Diehl, Olivia J Hon, Stefan Leutgeb, and Jill K Leutgeb. Grid and nongrid cells in me-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                638,
+                                504,
+                                652
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        638,
+                                        504,
+                                        652
+                                    ],
+                                    "score": 1.0,
+                                    "content": "dial entorhinal cortex represent spatial location and environmental features with complementary",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                650,
+                                299,
+                                661
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        650,
+                                        299,
+                                        661
+                                    ],
+                                    "score": 1.0,
+                                    "content": "coding schemes. Neuron, 94(1):83–92, 2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        }
+                    ],
+                    "index": 40
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        668,
+                        503,
+                        702
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                667,
+                                505,
+                                681
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        667,
+                                        505,
+                                        681
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Yedidyah Dordek, Daniel Soudry, Ron Meir, and Dori Derdikman. Extracting grid cell character-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                678,
+                                505,
+                                693
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        678,
+                                        505,
+                                        693
+                                    ],
+                                    "score": 1.0,
+                                    "content": "istics from place cell inputs using non-negative principal component analysis. eLife, 5:e10094,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                690,
+                                142,
+                                702
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        690,
+                                        142,
+                                        702
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2016.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        }
+                    ],
+                    "index": 43
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        709,
+                        503,
+                        731
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                708,
+                                504,
+                                722
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        708,
+                                        504,
+                                        722
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Ariane S Etienne and Kathryn J Jeffery. Path integration in mammals. Hippocampus, 14(2):180–",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                719,
+                                162,
+                                732
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        719,
+                                        162,
+                                        732
+                                    ],
+                                    "score": 1.0,
+                                    "content": "192, 2004.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        }
+                    ],
+                    "index": 45.5
+                }
+            ],
+            "page_idx": 9,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        108,
+                        27,
+                        293,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                26,
+                                294,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        26,
+                                        294,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Published as a conference paper at ICLR 2018",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        301,
+                        751,
+                        311,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                299,
+                                750,
+                                313,
+                                765
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        299,
+                                        750,
+                                        313,
+                                        765
+                                    ],
+                                    "score": 1.0,
+                                    "content": "",
+                                    "type": "text",
+                                    "height": 15,
+                                    "width": 14
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        82,
+                        504,
+                        105
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                81,
+                                505,
+                                96
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        81,
+                                        505,
+                                        96
+                                    ],
+                                    "score": 1.0,
+                                    "content": "LSTM units (Hochreiter & Schmidhuber, 1997) to perform different navigation tasks. However, no",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                93,
+                                282,
+                                106
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        93,
+                                        282,
+                                        106
+                                    ],
+                                    "score": 1.0,
+                                    "content": "grid-like spatial firing patterns are reported.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        }
+                    ],
+                    "index": 0.5,
+                    "bbox_fs": [
+                        105,
+                        81,
+                        505,
+                        106
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        111,
+                        505,
+                        297
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                110,
+                                505,
+                                123
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        110,
+                                        505,
+                                        123
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Although our model shows a qualitative match to the neural responses observed in the EC, nonethe-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                122,
+                                505,
+                                133
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        122,
+                                        505,
+                                        133
+                                    ],
+                                    "score": 1.0,
+                                    "content": "less it has several major limitations, with each offering interesting future research directions. First,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                132,
+                                506,
+                                145
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        132,
+                                        506,
+                                        145
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the learning rule we use seems to be biologically implausible. We are interested in exploring how a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                143,
+                                505,
+                                156
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        143,
+                                        505,
+                                        156
+                                    ],
+                                    "score": 1.0,
+                                    "content": "more biologically plausible learning rule could give rise to similar results (Lillicrap et al., 2016; Mi-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                154,
+                                506,
+                                167
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        154,
+                                        506,
+                                        167
+                                    ],
+                                    "score": 1.0,
+                                    "content": "coni, 2017; Guerguiev et al., 2017). Second, the simulation results do not show a variety of spatial",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                165,
+                                506,
+                                178
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        165,
+                                        506,
+                                        178
+                                    ],
+                                    "score": 1.0,
+                                    "content": "scales in grid-like cells. Experimentally, it is known that grid cells have multiple spatial scales, that",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                176,
+                                506,
+                                189
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        176,
+                                        506,
+                                        189
+                                    ],
+                                    "score": 1.0,
+                                    "content": "scale geometrically with a ratio 1.4 (Stensola et al., 2012), and this particular scale ratio is predicted",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                187,
+                                505,
+                                200
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        187,
+                                        505,
+                                        200
+                                    ],
+                                    "score": 1.0,
+                                    "content": "by efficient coding of space (Wei et al., 2015). We are investigating how to modify the model to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                198,
+                                506,
+                                212
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        198,
+                                        506,
+                                        212
+                                    ],
+                                    "score": 1.0,
+                                    "content": "get a hierarchy of spatial scales, perhaps by incorporating more neurons or modifying the regular-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                209,
+                                505,
+                                222
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        209,
+                                        505,
+                                        222
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ization. Last but not least, we have focused on the representation produced by the trained RNN.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                220,
+                                506,
+                                233
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        220,
+                                        506,
+                                        233
+                                    ],
+                                    "score": 1.0,
+                                    "content": "An equally important set of questions concern how the networks actually support the generation of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                230,
+                                505,
+                                243
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        230,
+                                        505,
+                                        243
+                                    ],
+                                    "score": 1.0,
+                                    "content": "such a representation. As a preliminary effort, we have examined the connectivity patterns of the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                242,
+                                506,
+                                255
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        242,
+                                        506,
+                                        255
+                                    ],
+                                    "score": 1.0,
+                                    "content": "trained network, and they do not seem to resemble the connectivity patterns required by standard",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                253,
+                                506,
+                                266
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        253,
+                                        506,
+                                        266
+                                    ],
+                                    "score": 1.0,
+                                    "content": "attractor network models. Maybe this should not be seen as too surprising. After all, the trained",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                264,
+                                505,
+                                277
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        264,
+                                        505,
+                                        277
+                                    ],
+                                    "score": 1.0,
+                                    "content": "networks can produce a diverse set of neural responses, while the previous models only led to grid",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                275,
+                                505,
+                                287
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        275,
+                                        505,
+                                        287
+                                    ],
+                                    "score": 1.0,
+                                    "content": "responses. It would be interesting for future work to systematically examine the questions related to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                286,
+                                221,
+                                298
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        286,
+                                        221,
+                                        298
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the underlying mechanisms.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        }
+                    ],
+                    "index": 10,
+                    "bbox_fs": [
+                        105,
+                        110,
+                        506,
+                        298
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        314,
+                        244,
+                        326
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                313,
+                                245,
+                                328
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        313,
+                                        245,
+                                        328
+                                    ],
+                                    "score": 1.0,
+                                    "content": "5 ACKNOWLEDGEMENTS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        }
+                    ],
+                    "index": 19
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        338,
+                        503,
+                        372
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                338,
+                                505,
+                                350
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        338,
+                                        505,
+                                        350
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We thank members of the Center for Theoretical Neuroscience at Columbia University for useful",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                349,
+                                505,
+                                362
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        349,
+                                        505,
+                                        362
+                                    ],
+                                    "score": 1.0,
+                                    "content": "discussions and three anonymous reviewers for constructive feedback. Research supported by NSF",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                359,
+                                425,
+                                374
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        359,
+                                        425,
+                                        374
+                                    ],
+                                    "score": 1.0,
+                                    "content": "NeuroNex Award DBI-1707398 and NIH training grant 5T32NS064929 (CJC).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        }
+                    ],
+                    "index": 21,
+                    "bbox_fs": [
+                        105,
+                        338,
+                        505,
+                        374
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        388,
+                        175,
+                        400
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                388,
+                                176,
+                                402
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        388,
+                                        176,
+                                        402
+                                    ],
+                                    "score": 1.0,
+                                    "content": "REFERENCES",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        }
+                    ],
+                    "index": 23
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        407,
+                        504,
+                        429
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                407,
+                                505,
+                                419
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        407,
+                                        505,
+                                        419
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Dmitriy Aronov, Rhino Nevers, and David W Tank. Mapping of a non-spatial dimension by the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                417,
+                                300,
+                                430
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        417,
+                                        300,
+                                        430
+                                    ],
+                                    "score": 1.0,
+                                    "content": "hippocampalentorhinal circuit. Nature, 2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        }
+                    ],
+                    "index": 24.5,
+                    "bbox_fs": [
+                        106,
+                        407,
+                        505,
+                        430
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        437,
+                        505,
+                        460
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                435,
+                                505,
+                                451
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        435,
+                                        505,
+                                        451
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Horace B Barlow. Possible principles underlying the transformation of sensory messages. Sensory",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                448,
+                                262,
+                                459
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        448,
+                                        262,
+                                        459
+                                    ],
+                                    "score": 1.0,
+                                    "content": "communication, pp. 217–234, 1961.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        }
+                    ],
+                    "index": 26.5,
+                    "bbox_fs": [
+                        105,
+                        435,
+                        505,
+                        459
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        466,
+                        504,
+                        500
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                466,
+                                505,
+                                479
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        466,
+                                        505,
+                                        479
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Yoshua Bengio, Aaron Courville, and Pascal Vincent. Representation learning: A review and new",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                478,
+                                506,
+                                491
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        478,
+                                        506,
+                                        491
+                                    ],
+                                    "score": 1.0,
+                                    "content": "perspectives. IEEE transactions on pattern analysis and machine intelligence, 35(8):1798–1828,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                488,
+                                144,
+                                501
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        488,
+                                        144,
+                                        501
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2013.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        }
+                    ],
+                    "index": 29,
+                    "bbox_fs": [
+                        106,
+                        466,
+                        506,
+                        501
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        507,
+                        505,
+                        531
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                508,
+                                505,
+                                521
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        508,
+                                        505,
+                                        521
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Tale L Bjerknes, Edvard I Moser, and May-Britt Moser. Representation of geometric borders in the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                518,
+                                294,
+                                531
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        518,
+                                        294,
+                                        531
+                                    ],
+                                    "score": 1.0,
+                                    "content": "developing rat. Neuron, 82(1):71–78, 2014.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        }
+                    ],
+                    "index": 31.5,
+                    "bbox_fs": [
+                        106,
+                        508,
+                        505,
+                        531
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        537,
+                        504,
+                        561
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                537,
+                                506,
+                                550
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        537,
+                                        506,
+                                        550
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Yoram Burak and Ila R Fiete. Accurate path integration in continuous attractor network models of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                549,
+                                366,
+                                561
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        549,
+                                        366,
+                                        561
+                                    ],
+                                    "score": 1.0,
+                                    "content": "grid cells. PLoS computational biology, 5(2):e1000291, 2009.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        }
+                    ],
+                    "index": 33.5,
+                    "bbox_fs": [
+                        106,
+                        537,
+                        506,
+                        561
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        109,
+                        567,
+                        504,
+                        601
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                107,
+                                567,
+                                505,
+                                581
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        567,
+                                        505,
+                                        581
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Jonathan J Couey, Aree Witoelar, Sheng-Jia Zhang, Kang Zheng, Jing Ye, Benjamin Dunn, Rafal",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                578,
+                                505,
+                                592
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        578,
+                                        505,
+                                        592
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Czajkowski, May-Britt Moser, Edvard I Moser, Yasser Roudi, et al. Recurrent inhibitory circuitry",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                590,
+                                437,
+                                602
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        590,
+                                        437,
+                                        602
+                                    ],
+                                    "score": 1.0,
+                                    "content": "as a mechanism for grid formation. Nature neuroscience, 16(3):318–324, 2013.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        }
+                    ],
+                    "index": 36,
+                    "bbox_fs": [
+                        107,
+                        567,
+                        505,
+                        602
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        608,
+                        387,
+                        621
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                608,
+                                388,
+                                622
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        608,
+                                        388,
+                                        622
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Charles Darwin. Origin of certain instincts. Nature, 7:417–418, 1873.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        }
+                    ],
+                    "index": 38,
+                    "bbox_fs": [
+                        106,
+                        608,
+                        388,
+                        622
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        109,
+                        627,
+                        504,
+                        661
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                107,
+                                627,
+                                504,
+                                640
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        627,
+                                        504,
+                                        640
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Geoffrey W Diehl, Olivia J Hon, Stefan Leutgeb, and Jill K Leutgeb. Grid and nongrid cells in me-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                638,
+                                504,
+                                652
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        638,
+                                        504,
+                                        652
+                                    ],
+                                    "score": 1.0,
+                                    "content": "dial entorhinal cortex represent spatial location and environmental features with complementary",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                650,
+                                299,
+                                661
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        650,
+                                        299,
+                                        661
+                                    ],
+                                    "score": 1.0,
+                                    "content": "coding schemes. Neuron, 94(1):83–92, 2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        }
+                    ],
+                    "index": 40,
+                    "bbox_fs": [
+                        107,
+                        627,
+                        504,
+                        661
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        668,
+                        503,
+                        702
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                667,
+                                505,
+                                681
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        667,
+                                        505,
+                                        681
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Yedidyah Dordek, Daniel Soudry, Ron Meir, and Dori Derdikman. Extracting grid cell character-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                678,
+                                505,
+                                693
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        678,
+                                        505,
+                                        693
+                                    ],
+                                    "score": 1.0,
+                                    "content": "istics from place cell inputs using non-negative principal component analysis. eLife, 5:e10094,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                690,
+                                142,
+                                702
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        690,
+                                        142,
+                                        702
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2016.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        }
+                    ],
+                    "index": 43,
+                    "bbox_fs": [
+                        106,
+                        667,
+                        505,
+                        702
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        709,
+                        503,
+                        731
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                708,
+                                504,
+                                722
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        708,
+                                        504,
+                                        722
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Ariane S Etienne and Kathryn J Jeffery. Path integration in mammals. Hippocampus, 14(2):180–",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                719,
+                                162,
+                                732
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        719,
+                                        162,
+                                        732
+                                    ],
+                                    "score": 1.0,
+                                    "content": "192, 2004.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        }
+                    ],
+                    "index": 45.5,
+                    "bbox_fs": [
+                        106,
+                        708,
+                        504,
+                        732
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        82,
+                        504,
+                        105
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                82,
+                                505,
+                                95
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        82,
+                                        505,
+                                        95
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Marianne Fyhn, Sturla Molden, Menno P Witter, Edvard I Moser, and May-Britt Moser. Spatial",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                93,
+                                429,
+                                106
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        93,
+                                        429,
+                                        106
+                                    ],
+                                    "score": 1.0,
+                                    "content": "representation in the entorhinal cortex. Science, 305(5688):1258–1264, 2004.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        }
+                    ],
+                    "index": 0.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        112,
+                        504,
+                        135
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                111,
+                                505,
+                                126
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        111,
+                                        505,
+                                        126
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Marianne Fyhn, Torkel Hafting, Menno P Witter, Edvard I Moser, and May-Britt Moser. Grid cells",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                124,
+                                316,
+                                136
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        124,
+                                        316,
+                                        136
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in mice. Hippocampus, 18(12):1230–1238, 2008.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        }
+                    ],
+                    "index": 2.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        142,
+                        504,
+                        176
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                141,
+                                505,
+                                155
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        141,
+                                        505,
+                                        155
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Alex Graves, Abdel-Rahman Mohamed, and Geoffrey Hinton. Speech recognition with deep recur-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                154,
+                                505,
+                                166
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        154,
+                                        505,
+                                        166
+                                    ],
+                                    "score": 1.0,
+                                    "content": "rent neural networks. In Acoustics, speech and signal processing (icassp), 2013 ieee international",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                165,
+                                294,
+                                177
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        165,
+                                        294,
+                                        177
+                                    ],
+                                    "score": 1.0,
+                                    "content": "conference on, pp. 6645–6649. IEEE, 2013.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        }
+                    ],
+                    "index": 5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        183,
+                        503,
+                        207
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                184,
+                                505,
+                                196
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        184,
+                                        505,
+                                        196
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Karol Gregor, Ivo Danihelka, Alex Graves, Danilo Jimenez Rezende, and Daan Wierstra. Draw: A",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                195,
+                                468,
+                                207
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        195,
+                                        468,
+                                        207
+                                    ],
+                                    "score": 1.0,
+                                    "content": "recurrent neural network for image generation. arXiv preprint arXiv:1502.04623, 2015.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        }
+                    ],
+                    "index": 7.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        213,
+                        502,
+                        237
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                211,
+                                504,
+                                228
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        211,
+                                        504,
+                                        228
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Jordan Guerguiev, Timothy P Lillicrap, and Blake A Richards. Towards deep learning with segre-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                225,
+                                243,
+                                237
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        225,
+                                        243,
+                                        237
+                                    ],
+                                    "score": 1.0,
+                                    "content": "gated dendrites. eLife, 6, 2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        }
+                    ],
+                    "index": 9.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        243,
+                        502,
+                        267
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                244,
+                                504,
+                                257
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        244,
+                                        504,
+                                        257
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Torkel Hafting, Marianne Fyhn, Sturla Molden, May-Britt Moser, and Edvard I Moser. Microstruc-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                255,
+                                440,
+                                267
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        255,
+                                        440,
+                                        267
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ture of a spatial map in the entorhinal cortex. Nature, 436(7052):801–806, 2005.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        }
+                    ],
+                    "index": 11.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        273,
+                        502,
+                        297
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                273,
+                                504,
+                                287
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        273,
+                                        504,
+                                        287
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Kiah Hardcastle, Surya Ganguli, and Lisa M Giocomo. Environmental boundaries as an error cor-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                285,
+                                375,
+                                297
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        285,
+                                        375,
+                                        297
+                                    ],
+                                    "score": 1.0,
+                                    "content": "rection mechanism for grid cells. Neuron, 86(3):827–839, 2015.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        }
+                    ],
+                    "index": 13.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        303,
+                        503,
+                        327
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                304,
+                                505,
+                                317
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        304,
+                                        505,
+                                        317
+                                    ],
+                                    "score": 1.0,
+                                    "content": "James R Hinman, Mark P Brandon, Jason R Climer, G William Chapman, and Michael E Hasselmo.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                315,
+                                479,
+                                328
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        315,
+                                        479,
+                                        328
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Multiple running speed signals in medial entorhinal cortex. Neuron, 91(3):666–679, 2016.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        }
+                    ],
+                    "index": 15.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        334,
+                        504,
+                        357
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                333,
+                                505,
+                                348
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        333,
+                                        505,
+                                        348
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Sepp Hochreiter and Jurgen Schmidhuber. Long short-term memory. ¨ Neural computation, 9(8):",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                345,
+                                191,
+                                357
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        345,
+                                        191,
+                                        357
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1735–1780, 1997.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        }
+                    ],
+                    "index": 17.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        364,
+                        504,
+                        388
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                364,
+                                504,
+                                377
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        364,
+                                        504,
+                                        377
+                                    ],
+                                    "score": 1.0,
+                                    "content": "David H Hubel and Torsten N Wiesel. Receptive fields, binocular interaction and functional archi-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                375,
+                                452,
+                                388
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        375,
+                                        452,
+                                        388
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tecture in the cat’s visual cortex. The Journal of physiology, 160(1):106–154, 1962.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        }
+                    ],
+                    "index": 19.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        394,
+                        505,
+                        439
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                394,
+                                505,
+                                406
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        394,
+                                        505,
+                                        406
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Joshua Jacobs, Christoph T Weidemann, Jonathan F Miller, Alec Solway, John F Burke, Xue-Xin",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                406,
+                                505,
+                                418
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        406,
+                                        505,
+                                        418
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Wei, Nanthia Suthana, Michael R Sperling, Ashwini D Sharan, Itzhak Fried, et al. Direct record-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                416,
+                                506,
+                                429
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        416,
+                                        506,
+                                        429
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ings of grid-like neuronal activity in human spatial navigation. Nature neuroscience, 16(9):1188–",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                428,
+                                165,
+                                438
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        428,
+                                        165,
+                                        438
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1190, 2013.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        }
+                    ],
+                    "index": 22.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        446,
+                        504,
+                        470
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                446,
+                                504,
+                                459
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        446,
+                                        504,
+                                        459
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Ingmar Kanitscheider and Ila Fiete. Training recurrent networks to generate hypotheses about how",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                457,
+                                451,
+                                470
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        457,
+                                        451,
+                                        470
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the brain solves hard navigation problems. arXiv preprint arXiv:1609.09059, 2016.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        }
+                    ],
+                    "index": 25.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        476,
+                        504,
+                        500
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                475,
+                                505,
+                                490
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        475,
+                                        505,
+                                        490
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Tim Christian Kietzmann, Patrick McClure, and Nikolaus Kriegeskorte. Deep neural networks in",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                488,
+                                344,
+                                500
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        488,
+                                        344,
+                                        500
+                                    ],
+                                    "score": 1.0,
+                                    "content": "computational neuroscience. bioRxiv, pp. 133504, 2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        }
+                    ],
+                    "index": 27.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        506,
+                        504,
+                        530
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                506,
+                                505,
+                                520
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        506,
+                                        505,
+                                        520
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Nathaniel J Killian, Michael J Jutras, and Elizabeth A Buffalo. A map of visual space in the primate",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                518,
+                                331,
+                                530
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        518,
+                                        331,
+                                        530
+                                    ],
+                                    "score": 1.0,
+                                    "content": "entorhinal cortex. Nature, 491(7426):761–764, 2012.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        }
+                    ],
+                    "index": 29.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        537,
+                        504,
+                        560
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                537,
+                                505,
+                                550
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        537,
+                                        505,
+                                        550
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Nikolaus Kriegeskorte. Deep neural networks: a new framework for modeling biological vision and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                547,
+                                446,
+                                560
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        547,
+                                        446,
+                                        560
+                                    ],
+                                    "score": 1.0,
+                                    "content": "brain information processing. Annual Review of Vision Science, 1:417–446, 2015.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        }
+                    ],
+                    "index": 31.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        566,
+                        505,
+                        600
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                565,
+                                505,
+                                580
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        565,
+                                        505,
+                                        580
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Alex Krizhevsky, Ilya Sutskever, and Geoffrey E Hinton. Imagenet classification with deep convo-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                578,
+                                505,
+                                591
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        578,
+                                        505,
+                                        591
+                                    ],
+                                    "score": 1.0,
+                                    "content": "lutional neural networks. In Advances in neural information processing systems, pp. 1097–1105,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                114,
+                                588,
+                                143,
+                                601
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        114,
+                                        588,
+                                        143,
+                                        601
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2012.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        }
+                    ],
+                    "index": 34
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        608,
+                        504,
+                        631
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                608,
+                                506,
+                                622
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        608,
+                                        506,
+                                        622
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Emilio Kropff, James E Carmichael, May-Britt Moser, and Edvard I Moser. Speed cells in the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                619,
+                                361,
+                                631
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        619,
+                                        361,
+                                        631
+                                    ],
+                                    "score": 1.0,
+                                    "content": "medial entorhinal cortex. Nature, 523(7561):419–424, 2015.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        }
+                    ],
+                    "index": 36.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        637,
+                        504,
+                        662
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                638,
+                                505,
+                                651
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        638,
+                                        505,
+                                        651
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Julija Krupic, Neil Burgess, and John O?Keefe. Neural representations of location composed of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                650,
+                                362,
+                                661
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        650,
+                                        362,
+                                        661
+                                    ],
+                                    "score": 1.0,
+                                    "content": "spatially periodic bands. Science, 337(6096):853–857, 2012.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        }
+                    ],
+                    "index": 38.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        668,
+                        503,
+                        702
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                668,
+                                505,
+                                682
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        668,
+                                        505,
+                                        682
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Rosamund F Langston, James A Ainge, Jonathan J Couey, Cathrin B Canto, Tale L Bjerknes,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                680,
+                                504,
+                                692
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        680,
+                                        504,
+                                        692
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Menno P Witter, Edvard I Moser, and May-Britt Moser. Development of the spatial represen-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                117,
+                                690,
+                                367,
+                                703
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        117,
+                                        690,
+                                        367,
+                                        703
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tation system in the rat. Science, 328(5985):1576–1580, 2010.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        }
+                    ],
+                    "index": 41
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        709,
+                        503,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                709,
+                                505,
+                                722
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        709,
+                                        505,
+                                        722
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Yann LeCun, Leon Bottou, Yoshua Bengio, and Patrick Haffner. Gradient-based learning applied to ´",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                720,
+                                417,
+                                732
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        720,
+                                        417,
+                                        732
+                                    ],
+                                    "score": 1.0,
+                                    "content": "document recognition. Proceedings of the IEEE, 86(11):2278–2324, 1998.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        }
+                    ],
+                    "index": 43.5
+                }
+            ],
+            "page_idx": 10,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        107,
+                        27,
+                        293,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                26,
+                                294,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        26,
+                                        294,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Published as a conference paper at ICLR 2018",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        300,
+                        751,
+                        310,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                299,
+                                750,
+                                312,
+                                765
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        299,
+                                        750,
+                                        312,
+                                        765
+                                    ],
+                                    "score": 1.0,
+                                    "content": "",
+                                    "type": "text",
+                                    "height": 15,
+                                    "width": 13
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        82,
+                        504,
+                        105
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                82,
+                                505,
+                                95
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        82,
+                                        505,
+                                        95
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Marianne Fyhn, Sturla Molden, Menno P Witter, Edvard I Moser, and May-Britt Moser. Spatial",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                93,
+                                429,
+                                106
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        93,
+                                        429,
+                                        106
+                                    ],
+                                    "score": 1.0,
+                                    "content": "representation in the entorhinal cortex. Science, 305(5688):1258–1264, 2004.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        }
+                    ],
+                    "index": 0.5,
+                    "bbox_fs": [
+                        106,
+                        82,
+                        505,
+                        106
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        112,
+                        504,
+                        135
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                111,
+                                505,
+                                126
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        111,
+                                        505,
+                                        126
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Marianne Fyhn, Torkel Hafting, Menno P Witter, Edvard I Moser, and May-Britt Moser. Grid cells",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                124,
+                                316,
+                                136
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        124,
+                                        316,
+                                        136
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in mice. Hippocampus, 18(12):1230–1238, 2008.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        }
+                    ],
+                    "index": 2.5,
+                    "bbox_fs": [
+                        105,
+                        111,
+                        505,
+                        136
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        142,
+                        504,
+                        176
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                141,
+                                505,
+                                155
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        141,
+                                        505,
+                                        155
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Alex Graves, Abdel-Rahman Mohamed, and Geoffrey Hinton. Speech recognition with deep recur-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                154,
+                                505,
+                                166
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        154,
+                                        505,
+                                        166
+                                    ],
+                                    "score": 1.0,
+                                    "content": "rent neural networks. In Acoustics, speech and signal processing (icassp), 2013 ieee international",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                165,
+                                294,
+                                177
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        165,
+                                        294,
+                                        177
+                                    ],
+                                    "score": 1.0,
+                                    "content": "conference on, pp. 6645–6649. IEEE, 2013.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        }
+                    ],
+                    "index": 5,
+                    "bbox_fs": [
+                        106,
+                        141,
+                        505,
+                        177
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        183,
+                        503,
+                        207
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                184,
+                                505,
+                                196
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        184,
+                                        505,
+                                        196
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Karol Gregor, Ivo Danihelka, Alex Graves, Danilo Jimenez Rezende, and Daan Wierstra. Draw: A",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                195,
+                                468,
+                                207
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        195,
+                                        468,
+                                        207
+                                    ],
+                                    "score": 1.0,
+                                    "content": "recurrent neural network for image generation. arXiv preprint arXiv:1502.04623, 2015.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        }
+                    ],
+                    "index": 7.5,
+                    "bbox_fs": [
+                        106,
+                        184,
+                        505,
+                        207
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        213,
+                        502,
+                        237
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                211,
+                                504,
+                                228
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        211,
+                                        504,
+                                        228
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Jordan Guerguiev, Timothy P Lillicrap, and Blake A Richards. Towards deep learning with segre-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                225,
+                                243,
+                                237
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        225,
+                                        243,
+                                        237
+                                    ],
+                                    "score": 1.0,
+                                    "content": "gated dendrites. eLife, 6, 2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        }
+                    ],
+                    "index": 9.5,
+                    "bbox_fs": [
+                        104,
+                        211,
+                        504,
+                        237
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        243,
+                        502,
+                        267
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                244,
+                                504,
+                                257
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        244,
+                                        504,
+                                        257
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Torkel Hafting, Marianne Fyhn, Sturla Molden, May-Britt Moser, and Edvard I Moser. Microstruc-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                255,
+                                440,
+                                267
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        255,
+                                        440,
+                                        267
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ture of a spatial map in the entorhinal cortex. Nature, 436(7052):801–806, 2005.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        }
+                    ],
+                    "index": 11.5,
+                    "bbox_fs": [
+                        106,
+                        244,
+                        504,
+                        267
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        273,
+                        502,
+                        297
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                273,
+                                504,
+                                287
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        273,
+                                        504,
+                                        287
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Kiah Hardcastle, Surya Ganguli, and Lisa M Giocomo. Environmental boundaries as an error cor-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                285,
+                                375,
+                                297
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        285,
+                                        375,
+                                        297
+                                    ],
+                                    "score": 1.0,
+                                    "content": "rection mechanism for grid cells. Neuron, 86(3):827–839, 2015.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        }
+                    ],
+                    "index": 13.5,
+                    "bbox_fs": [
+                        106,
+                        273,
+                        504,
+                        297
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        303,
+                        503,
+                        327
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                304,
+                                505,
+                                317
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        304,
+                                        505,
+                                        317
+                                    ],
+                                    "score": 1.0,
+                                    "content": "James R Hinman, Mark P Brandon, Jason R Climer, G William Chapman, and Michael E Hasselmo.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                315,
+                                479,
+                                328
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        315,
+                                        479,
+                                        328
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Multiple running speed signals in medial entorhinal cortex. Neuron, 91(3):666–679, 2016.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        }
+                    ],
+                    "index": 15.5,
+                    "bbox_fs": [
+                        106,
+                        304,
+                        505,
+                        328
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        334,
+                        504,
+                        357
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                333,
+                                505,
+                                348
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        333,
+                                        505,
+                                        348
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Sepp Hochreiter and Jurgen Schmidhuber. Long short-term memory. ¨ Neural computation, 9(8):",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                345,
+                                191,
+                                357
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        345,
+                                        191,
+                                        357
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1735–1780, 1997.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        }
+                    ],
+                    "index": 17.5,
+                    "bbox_fs": [
+                        105,
+                        333,
+                        505,
+                        357
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        364,
+                        504,
+                        388
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                364,
+                                504,
+                                377
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        364,
+                                        504,
+                                        377
+                                    ],
+                                    "score": 1.0,
+                                    "content": "David H Hubel and Torsten N Wiesel. Receptive fields, binocular interaction and functional archi-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                375,
+                                452,
+                                388
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        375,
+                                        452,
+                                        388
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tecture in the cat’s visual cortex. The Journal of physiology, 160(1):106–154, 1962.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        }
+                    ],
+                    "index": 19.5,
+                    "bbox_fs": [
+                        105,
+                        364,
+                        504,
+                        388
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        394,
+                        505,
+                        439
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                394,
+                                505,
+                                406
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        394,
+                                        505,
+                                        406
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Joshua Jacobs, Christoph T Weidemann, Jonathan F Miller, Alec Solway, John F Burke, Xue-Xin",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                406,
+                                505,
+                                418
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        406,
+                                        505,
+                                        418
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Wei, Nanthia Suthana, Michael R Sperling, Ashwini D Sharan, Itzhak Fried, et al. Direct record-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                416,
+                                506,
+                                429
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        416,
+                                        506,
+                                        429
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ings of grid-like neuronal activity in human spatial navigation. Nature neuroscience, 16(9):1188–",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                428,
+                                165,
+                                438
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        428,
+                                        165,
+                                        438
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1190, 2013.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        }
+                    ],
+                    "index": 22.5,
+                    "bbox_fs": [
+                        105,
+                        394,
+                        506,
+                        438
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        446,
+                        504,
+                        470
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                446,
+                                504,
+                                459
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        446,
+                                        504,
+                                        459
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Ingmar Kanitscheider and Ila Fiete. Training recurrent networks to generate hypotheses about how",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                457,
+                                451,
+                                470
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        457,
+                                        451,
+                                        470
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the brain solves hard navigation problems. arXiv preprint arXiv:1609.09059, 2016.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        }
+                    ],
+                    "index": 25.5,
+                    "bbox_fs": [
+                        106,
+                        446,
+                        504,
+                        470
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        476,
+                        504,
+                        500
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                475,
+                                505,
+                                490
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        475,
+                                        505,
+                                        490
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Tim Christian Kietzmann, Patrick McClure, and Nikolaus Kriegeskorte. Deep neural networks in",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                488,
+                                344,
+                                500
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        488,
+                                        344,
+                                        500
+                                    ],
+                                    "score": 1.0,
+                                    "content": "computational neuroscience. bioRxiv, pp. 133504, 2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        }
+                    ],
+                    "index": 27.5,
+                    "bbox_fs": [
+                        106,
+                        475,
+                        505,
+                        500
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        506,
+                        504,
+                        530
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                506,
+                                505,
+                                520
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        506,
+                                        505,
+                                        520
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Nathaniel J Killian, Michael J Jutras, and Elizabeth A Buffalo. A map of visual space in the primate",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                518,
+                                331,
+                                530
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        518,
+                                        331,
+                                        530
+                                    ],
+                                    "score": 1.0,
+                                    "content": "entorhinal cortex. Nature, 491(7426):761–764, 2012.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        }
+                    ],
+                    "index": 29.5,
+                    "bbox_fs": [
+                        105,
+                        506,
+                        505,
+                        530
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        537,
+                        504,
+                        560
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                537,
+                                505,
+                                550
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        537,
+                                        505,
+                                        550
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Nikolaus Kriegeskorte. Deep neural networks: a new framework for modeling biological vision and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                547,
+                                446,
+                                560
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        547,
+                                        446,
+                                        560
+                                    ],
+                                    "score": 1.0,
+                                    "content": "brain information processing. Annual Review of Vision Science, 1:417–446, 2015.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        }
+                    ],
+                    "index": 31.5,
+                    "bbox_fs": [
+                        106,
+                        537,
+                        505,
+                        560
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        566,
+                        505,
+                        600
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                565,
+                                505,
+                                580
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        565,
+                                        505,
+                                        580
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Alex Krizhevsky, Ilya Sutskever, and Geoffrey E Hinton. Imagenet classification with deep convo-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                578,
+                                505,
+                                591
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        578,
+                                        505,
+                                        591
+                                    ],
+                                    "score": 1.0,
+                                    "content": "lutional neural networks. In Advances in neural information processing systems, pp. 1097–1105,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                114,
+                                588,
+                                143,
+                                601
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        114,
+                                        588,
+                                        143,
+                                        601
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2012.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        }
+                    ],
+                    "index": 34,
+                    "bbox_fs": [
+                        106,
+                        565,
+                        505,
+                        601
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        608,
+                        504,
+                        631
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                608,
+                                506,
+                                622
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        608,
+                                        506,
+                                        622
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Emilio Kropff, James E Carmichael, May-Britt Moser, and Edvard I Moser. Speed cells in the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                619,
+                                361,
+                                631
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        619,
+                                        361,
+                                        631
+                                    ],
+                                    "score": 1.0,
+                                    "content": "medial entorhinal cortex. Nature, 523(7561):419–424, 2015.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        }
+                    ],
+                    "index": 36.5,
+                    "bbox_fs": [
+                        105,
+                        608,
+                        506,
+                        631
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        637,
+                        504,
+                        662
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                638,
+                                505,
+                                651
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        638,
+                                        505,
+                                        651
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Julija Krupic, Neil Burgess, and John O?Keefe. Neural representations of location composed of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                650,
+                                362,
+                                661
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        650,
+                                        362,
+                                        661
+                                    ],
+                                    "score": 1.0,
+                                    "content": "spatially periodic bands. Science, 337(6096):853–857, 2012.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        }
+                    ],
+                    "index": 38.5,
+                    "bbox_fs": [
+                        105,
+                        638,
+                        505,
+                        661
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        668,
+                        503,
+                        702
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                668,
+                                505,
+                                682
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        668,
+                                        505,
+                                        682
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Rosamund F Langston, James A Ainge, Jonathan J Couey, Cathrin B Canto, Tale L Bjerknes,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                680,
+                                504,
+                                692
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        680,
+                                        504,
+                                        692
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Menno P Witter, Edvard I Moser, and May-Britt Moser. Development of the spatial represen-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                117,
+                                690,
+                                367,
+                                703
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        117,
+                                        690,
+                                        367,
+                                        703
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tation system in the rat. Science, 328(5985):1576–1580, 2010.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        }
+                    ],
+                    "index": 41,
+                    "bbox_fs": [
+                        106,
+                        668,
+                        505,
+                        703
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        709,
+                        503,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                709,
+                                505,
+                                722
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        709,
+                                        505,
+                                        722
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Yann LeCun, Leon Bottou, Yoshua Bengio, and Patrick Haffner. Gradient-based learning applied to ´",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                720,
+                                417,
+                                732
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        720,
+                                        417,
+                                        732
+                                    ],
+                                    "score": 1.0,
+                                    "content": "document recognition. Proceedings of the IEEE, 86(11):2278–2324, 1998.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        }
+                    ],
+                    "index": 43.5,
+                    "bbox_fs": [
+                        106,
+                        709,
+                        505,
+                        732
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        82,
+                        504,
+                        105
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                81,
+                                505,
+                                95
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        81,
+                                        505,
+                                        95
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Yann LeCun, Yoshua Bengio, and Geoffrey Hinton. Deep learning. Nature, 521(7553):436–444,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                114,
+                                92,
+                                143,
+                                106
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        114,
+                                        92,
+                                        143,
+                                        106
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2015.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        }
+                    ],
+                    "index": 0.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        112,
+                        504,
+                        146
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                113,
+                                505,
+                                125
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        113,
+                                        505,
+                                        125
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Colin Lever, Stephen Burton, Ali Jeewajee, John O’Keefe, and Neil Burgess. Boundary vector cells",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                123,
+                                505,
+                                136
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        123,
+                                        505,
+                                        136
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in the subiculum of the hippocampal formation. The journal of neuroscience, 29(31):9771–9777,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                114,
+                                132,
+                                142,
+                                147
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        114,
+                                        132,
+                                        142,
+                                        147
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2009.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        }
+                    ],
+                    "index": 3
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        153,
+                        505,
+                        187
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                153,
+                                505,
+                                166
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        153,
+                                        505,
+                                        166
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Timothy P Lillicrap, Daniel Cownden, Douglas B Tweed, and Colin J Akerman. Random synaptic",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                163,
+                                506,
+                                178
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        163,
+                                        506,
+                                        178
+                                    ],
+                                    "score": 1.0,
+                                    "content": "feedback weights support error backpropagation for deep learning. Nature communications, 7,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                173,
+                                143,
+                                188
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        173,
+                                        143,
+                                        188
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2016.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        }
+                    ],
+                    "index": 6
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        194,
+                        503,
+                        217
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                194,
+                                505,
+                                207
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        194,
+                                        505,
+                                        207
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Laurens van der Maaten and Geoffrey Hinton. Visualizing data using t-sne. Journal of Machine",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                205,
+                                303,
+                                218
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        205,
+                                        303,
+                                        218
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Learning Research, 9(Nov):2579–2605, 2008.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        }
+                    ],
+                    "index": 8.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        224,
+                        503,
+                        248
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                224,
+                                505,
+                                238
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        224,
+                                        505,
+                                        238
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Valerio Mante, David Sussillo, Krishna V Shenoy, and William T Newsome. Context-dependent",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                236,
+                                475,
+                                248
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        236,
+                                        475,
+                                        248
+                                    ],
+                                    "score": 1.0,
+                                    "content": "computation by recurrent dynamics in prefrontal cortex. Nature, 503(7474):78–84, 2013.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        }
+                    ],
+                    "index": 10.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        254,
+                        504,
+                        278
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                254,
+                                504,
+                                267
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        254,
+                                        504,
+                                        267
+                                    ],
+                                    "score": 1.0,
+                                    "content": "James Martens and Ilya Sutskever. Learning recurrent neural networks with hessian-free optimiza-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                266,
+                                223,
+                                277
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        266,
+                                        223,
+                                        277
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tion. pp. 10331040, 2011.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        }
+                    ],
+                    "index": 12.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        284,
+                        505,
+                        318
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                284,
+                                505,
+                                298
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        284,
+                                        505,
+                                        298
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Bruce L McNaughton, Francesco P Battaglia, Ole Jensen, Edvard I Moser, and May-Britt Moser.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                296,
+                                504,
+                                308
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        296,
+                                        504,
+                                        308
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Path integration and the neural basis of the ’cognitive map’. Nature Reviews Neuroscience, 7(8):",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                307,
+                                181,
+                                318
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        307,
+                                        181,
+                                        318
+                                    ],
+                                    "score": 1.0,
+                                    "content": "663–678, 2006.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        }
+                    ],
+                    "index": 15
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        326,
+                        504,
+                        349
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                326,
+                                505,
+                                339
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        326,
+                                        505,
+                                        339
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Thomas Miconi. Biologically plausible learning in recurrent neural networks reproduces neural",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                337,
+                                381,
+                                350
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        337,
+                                        381,
+                                        350
+                                    ],
+                                    "score": 1.0,
+                                    "content": "dynamics observed during cognitive tasks. eLife, 6:e20899, 2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        }
+                    ],
+                    "index": 17.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        356,
+                        504,
+                        379
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                356,
+                                506,
+                                370
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        356,
+                                        506,
+                                        370
+                                    ],
+                                    "score": 1.0,
+                                    "content": "M-L Mittelstaedt and H Mittelstaedt. Homing by path integration in a mammal. Naturwis-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                367,
+                                262,
+                                379
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        367,
+                                        262,
+                                        379
+                                    ],
+                                    "score": 1.0,
+                                    "content": "senschaften, 67(11):566–567, 1980.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        }
+                    ],
+                    "index": 19.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        385,
+                        505,
+                        420
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                386,
+                                505,
+                                399
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        386,
+                                        505,
+                                        399
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Volodymyr Mnih, Koray Kavukcuoglu, David Silver, Andrei A Rusu, Joel Veness, Marc G Belle-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                397,
+                                505,
+                                410
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        397,
+                                        505,
+                                        410
+                                    ],
+                                    "score": 1.0,
+                                    "content": "mare, Alex Graves, Martin Riedmiller, Andreas K Fidjeland, Georg Ostrovski, et al. Human-level",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                408,
+                                441,
+                                421
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        408,
+                                        441,
+                                        421
+                                    ],
+                                    "score": 1.0,
+                                    "content": "control through deep reinforcement learning. Nature, 518(7540):529–533, 2015.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        }
+                    ],
+                    "index": 22
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        427,
+                        504,
+                        450
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                428,
+                                505,
+                                439
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        428,
+                                        505,
+                                        439
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Edvard I Moser, Emilio Kropff, and May-Britt Moser. Place cells, grid cells, and the brain’s spatial",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                439,
+                                365,
+                                450
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        439,
+                                        365,
+                                        450
+                                    ],
+                                    "score": 1.0,
+                                    "content": "representation system. Annu. Rev. Neurosci., 31:69–89, 2008.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        }
+                    ],
+                    "index": 24.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        457,
+                        504,
+                        480
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                456,
+                                505,
+                                471
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        456,
+                                        505,
+                                        471
+                                    ],
+                                    "score": 1.0,
+                                    "content": "John O’Keefe. Place units in the hippocampus of the freely moving rat. Experimental neurology, 51",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                468,
+                                191,
+                                480
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        468,
+                                        191,
+                                        480
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(1):78–109, 1976.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        }
+                    ],
+                    "index": 26.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        103,
+                        487,
+                        504,
+                        511
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                486,
+                                505,
+                                502
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        486,
+                                        505,
+                                        502
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Bruno A Olshausen and David J Field. Emergence of simple-cell receptive field properties by",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                498,
+                                407,
+                                511
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        498,
+                                        407,
+                                        511
+                                    ],
+                                    "score": 1.0,
+                                    "content": "learning a sparse code for natural images. Nature, 381(6583):607, 1996.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        }
+                    ],
+                    "index": 28.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        517,
+                        504,
+                        541
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                107,
+                                518,
+                                504,
+                                530
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        518,
+                                        504,
+                                        530
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Aaron van den Oord, Nal Kalchbrenner, and Koray Kavukcuoglu. Pixel recurrent neural networks.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                528,
+                                279,
+                                541
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        528,
+                                        279,
+                                        541
+                                    ],
+                                    "score": 1.0,
+                                    "content": "arXiv preprint arXiv:1601.06759, 2016.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        }
+                    ],
+                    "index": 30.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        547,
+                        503,
+                        571
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                547,
+                                504,
+                                560
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        547,
+                                        504,
+                                        560
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Eli Pollock, Niral Desai, Xue-Xin Wei, and Vijay B Balasubramanian. A mechanism for self-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                559,
+                                436,
+                                571
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        559,
+                                        436,
+                                        571
+                                    ],
+                                    "score": 1.0,
+                                    "content": "organized error-correction of grid cells by border cells. CoSyNe abstract, 2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        }
+                    ],
+                    "index": 32.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        577,
+                        503,
+                        601
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                577,
+                                505,
+                                592
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        577,
+                                        505,
+                                        592
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Alexei Samsonovich and Bruce L McNaughton. Path integration and cognitive mapping in a con-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                117,
+                                589,
+                                484,
+                                601
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        117,
+                                        589,
+                                        484,
+                                        601
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tinuous attractor neural network model. Journal of Neuroscience, 17(15):5900–5920, 1997.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        }
+                    ],
+                    "index": 34.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        608,
+                        504,
+                        642
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                607,
+                                505,
+                                621
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        607,
+                                        505,
+                                        621
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Francesca Sargolini, Marianne Fyhn, Torkel Hafting, Bruce L McNaughton, Menno P Witter, May-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                619,
+                                504,
+                                632
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        619,
+                                        504,
+                                        632
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Britt Moser, and Edvard I Moser. Conjunctive representation of position, direction, and velocity",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                631,
+                                343,
+                                641
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        631,
+                                        343,
+                                        641
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in entorhinal cortex. Science, 312(5774):758–762, 2006.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        }
+                    ],
+                    "index": 37
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        649,
+                        504,
+                        672
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                649,
+                                505,
+                                662
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        649,
+                                        505,
+                                        662
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Francesco Savelli, D Yoganarasimha, and James J Knierim. Influence of boundary removal on the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                660,
+                                476,
+                                672
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        660,
+                                        476,
+                                        672
+                                    ],
+                                    "score": 1.0,
+                                    "content": "spatial representations of the medial entorhinal cortex. Hippocampus, 18(12):1270, 2008.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        }
+                    ],
+                    "index": 39.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        679,
+                        503,
+                        702
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                679,
+                                504,
+                                692
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        679,
+                                        504,
+                                        692
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Andrew M Saxe, James L McClelland, and Surya Ganguli. Exact solutions to the nonlinear dynam-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                690,
+                                328,
+                                703
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        690,
+                                        328,
+                                        703
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ics of learning in deep linear neural networks. 2014.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        }
+                    ],
+                    "index": 41.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        709,
+                        502,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                709,
+                                503,
+                                722
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        709,
+                                        503,
+                                        722
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Trygve Solstad, Charlotte N Boccara, Emilio Kropff, May-Britt Moser, and Edvard I Moser. Repre-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                720,
+                                493,
+                                732
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        720,
+                                        493,
+                                        732
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sentation of geometric borders in the entorhinal cortex. Science, 322(5909):1865–1868, 2008.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        }
+                    ],
+                    "index": 43.5
+                }
+            ],
+            "page_idx": 11,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        108,
+                        27,
+                        293,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                26,
+                                294,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        26,
+                                        294,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Published as a conference paper at ICLR 2018",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        300,
+                        751,
+                        311,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                299,
+                                750,
+                                313,
+                                764
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        299,
+                                        750,
+                                        313,
+                                        764
+                                    ],
+                                    "score": 1.0,
+                                    "content": "12",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        82,
+                        504,
+                        105
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                81,
+                                505,
+                                95
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        81,
+                                        505,
+                                        95
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Yann LeCun, Yoshua Bengio, and Geoffrey Hinton. Deep learning. Nature, 521(7553):436–444,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                114,
+                                92,
+                                143,
+                                106
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        114,
+                                        92,
+                                        143,
+                                        106
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2015.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        }
+                    ],
+                    "index": 0.5,
+                    "bbox_fs": [
+                        105,
+                        81,
+                        505,
+                        106
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        112,
+                        504,
+                        146
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                113,
+                                505,
+                                125
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        113,
+                                        505,
+                                        125
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Colin Lever, Stephen Burton, Ali Jeewajee, John O’Keefe, and Neil Burgess. Boundary vector cells",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                123,
+                                505,
+                                136
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        123,
+                                        505,
+                                        136
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in the subiculum of the hippocampal formation. The journal of neuroscience, 29(31):9771–9777,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                114,
+                                132,
+                                142,
+                                147
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        114,
+                                        132,
+                                        142,
+                                        147
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2009.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        }
+                    ],
+                    "index": 3,
+                    "bbox_fs": [
+                        106,
+                        113,
+                        505,
+                        147
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        153,
+                        505,
+                        187
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                153,
+                                505,
+                                166
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        153,
+                                        505,
+                                        166
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Timothy P Lillicrap, Daniel Cownden, Douglas B Tweed, and Colin J Akerman. Random synaptic",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                163,
+                                506,
+                                178
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        163,
+                                        506,
+                                        178
+                                    ],
+                                    "score": 1.0,
+                                    "content": "feedback weights support error backpropagation for deep learning. Nature communications, 7,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                173,
+                                143,
+                                188
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        173,
+                                        143,
+                                        188
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2016.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        }
+                    ],
+                    "index": 6,
+                    "bbox_fs": [
+                        106,
+                        153,
+                        506,
+                        188
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        194,
+                        503,
+                        217
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                194,
+                                505,
+                                207
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        194,
+                                        505,
+                                        207
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Laurens van der Maaten and Geoffrey Hinton. Visualizing data using t-sne. Journal of Machine",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                205,
+                                303,
+                                218
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        205,
+                                        303,
+                                        218
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Learning Research, 9(Nov):2579–2605, 2008.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        }
+                    ],
+                    "index": 8.5,
+                    "bbox_fs": [
+                        106,
+                        194,
+                        505,
+                        218
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        224,
+                        503,
+                        248
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                224,
+                                505,
+                                238
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        224,
+                                        505,
+                                        238
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Valerio Mante, David Sussillo, Krishna V Shenoy, and William T Newsome. Context-dependent",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                236,
+                                475,
+                                248
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        236,
+                                        475,
+                                        248
+                                    ],
+                                    "score": 1.0,
+                                    "content": "computation by recurrent dynamics in prefrontal cortex. Nature, 503(7474):78–84, 2013.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        }
+                    ],
+                    "index": 10.5,
+                    "bbox_fs": [
+                        106,
+                        224,
+                        505,
+                        248
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        254,
+                        504,
+                        278
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                254,
+                                504,
+                                267
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        254,
+                                        504,
+                                        267
+                                    ],
+                                    "score": 1.0,
+                                    "content": "James Martens and Ilya Sutskever. Learning recurrent neural networks with hessian-free optimiza-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                266,
+                                223,
+                                277
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        266,
+                                        223,
+                                        277
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tion. pp. 10331040, 2011.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        }
+                    ],
+                    "index": 12.5,
+                    "bbox_fs": [
+                        106,
+                        254,
+                        504,
+                        277
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        284,
+                        505,
+                        318
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                284,
+                                505,
+                                298
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        284,
+                                        505,
+                                        298
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Bruce L McNaughton, Francesco P Battaglia, Ole Jensen, Edvard I Moser, and May-Britt Moser.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                296,
+                                504,
+                                308
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        296,
+                                        504,
+                                        308
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Path integration and the neural basis of the ’cognitive map’. Nature Reviews Neuroscience, 7(8):",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                307,
+                                181,
+                                318
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        307,
+                                        181,
+                                        318
+                                    ],
+                                    "score": 1.0,
+                                    "content": "663–678, 2006.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        }
+                    ],
+                    "index": 15,
+                    "bbox_fs": [
+                        105,
+                        284,
+                        505,
+                        318
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        326,
+                        504,
+                        349
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                326,
+                                505,
+                                339
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        326,
+                                        505,
+                                        339
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Thomas Miconi. Biologically plausible learning in recurrent neural networks reproduces neural",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                337,
+                                381,
+                                350
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        337,
+                                        381,
+                                        350
+                                    ],
+                                    "score": 1.0,
+                                    "content": "dynamics observed during cognitive tasks. eLife, 6:e20899, 2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        }
+                    ],
+                    "index": 17.5,
+                    "bbox_fs": [
+                        106,
+                        326,
+                        505,
+                        350
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        356,
+                        504,
+                        379
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                356,
+                                506,
+                                370
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        356,
+                                        506,
+                                        370
+                                    ],
+                                    "score": 1.0,
+                                    "content": "M-L Mittelstaedt and H Mittelstaedt. Homing by path integration in a mammal. Naturwis-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                367,
+                                262,
+                                379
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        367,
+                                        262,
+                                        379
+                                    ],
+                                    "score": 1.0,
+                                    "content": "senschaften, 67(11):566–567, 1980.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        }
+                    ],
+                    "index": 19.5,
+                    "bbox_fs": [
+                        105,
+                        356,
+                        506,
+                        379
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        385,
+                        505,
+                        420
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                386,
+                                505,
+                                399
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        386,
+                                        505,
+                                        399
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Volodymyr Mnih, Koray Kavukcuoglu, David Silver, Andrei A Rusu, Joel Veness, Marc G Belle-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                397,
+                                505,
+                                410
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        397,
+                                        505,
+                                        410
+                                    ],
+                                    "score": 1.0,
+                                    "content": "mare, Alex Graves, Martin Riedmiller, Andreas K Fidjeland, Georg Ostrovski, et al. Human-level",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                408,
+                                441,
+                                421
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        408,
+                                        441,
+                                        421
+                                    ],
+                                    "score": 1.0,
+                                    "content": "control through deep reinforcement learning. Nature, 518(7540):529–533, 2015.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        }
+                    ],
+                    "index": 22,
+                    "bbox_fs": [
+                        106,
+                        386,
+                        505,
+                        421
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        427,
+                        504,
+                        450
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                428,
+                                505,
+                                439
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        428,
+                                        505,
+                                        439
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Edvard I Moser, Emilio Kropff, and May-Britt Moser. Place cells, grid cells, and the brain’s spatial",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                439,
+                                365,
+                                450
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        439,
+                                        365,
+                                        450
+                                    ],
+                                    "score": 1.0,
+                                    "content": "representation system. Annu. Rev. Neurosci., 31:69–89, 2008.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        }
+                    ],
+                    "index": 24.5,
+                    "bbox_fs": [
+                        106,
+                        428,
+                        505,
+                        450
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        457,
+                        504,
+                        480
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                456,
+                                505,
+                                471
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        456,
+                                        505,
+                                        471
+                                    ],
+                                    "score": 1.0,
+                                    "content": "John O’Keefe. Place units in the hippocampus of the freely moving rat. Experimental neurology, 51",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                468,
+                                191,
+                                480
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        468,
+                                        191,
+                                        480
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(1):78–109, 1976.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        }
+                    ],
+                    "index": 26.5,
+                    "bbox_fs": [
+                        104,
+                        456,
+                        505,
+                        480
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        103,
+                        487,
+                        504,
+                        511
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                486,
+                                505,
+                                502
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        486,
+                                        505,
+                                        502
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Bruno A Olshausen and David J Field. Emergence of simple-cell receptive field properties by",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                498,
+                                407,
+                                511
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        498,
+                                        407,
+                                        511
+                                    ],
+                                    "score": 1.0,
+                                    "content": "learning a sparse code for natural images. Nature, 381(6583):607, 1996.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        }
+                    ],
+                    "index": 28.5,
+                    "bbox_fs": [
+                        105,
+                        486,
+                        505,
+                        511
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        517,
+                        504,
+                        541
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                107,
+                                518,
+                                504,
+                                530
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        518,
+                                        504,
+                                        530
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Aaron van den Oord, Nal Kalchbrenner, and Koray Kavukcuoglu. Pixel recurrent neural networks.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                528,
+                                279,
+                                541
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        528,
+                                        279,
+                                        541
+                                    ],
+                                    "score": 1.0,
+                                    "content": "arXiv preprint arXiv:1601.06759, 2016.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        }
+                    ],
+                    "index": 30.5,
+                    "bbox_fs": [
+                        107,
+                        518,
+                        504,
+                        541
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        547,
+                        503,
+                        571
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                547,
+                                504,
+                                560
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        547,
+                                        504,
+                                        560
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Eli Pollock, Niral Desai, Xue-Xin Wei, and Vijay B Balasubramanian. A mechanism for self-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                559,
+                                436,
+                                571
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        559,
+                                        436,
+                                        571
+                                    ],
+                                    "score": 1.0,
+                                    "content": "organized error-correction of grid cells by border cells. CoSyNe abstract, 2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        }
+                    ],
+                    "index": 32.5,
+                    "bbox_fs": [
+                        106,
+                        547,
+                        504,
+                        571
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        577,
+                        503,
+                        601
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                577,
+                                505,
+                                592
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        577,
+                                        505,
+                                        592
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Alexei Samsonovich and Bruce L McNaughton. Path integration and cognitive mapping in a con-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                117,
+                                589,
+                                484,
+                                601
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        117,
+                                        589,
+                                        484,
+                                        601
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tinuous attractor neural network model. Journal of Neuroscience, 17(15):5900–5920, 1997.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        }
+                    ],
+                    "index": 34.5,
+                    "bbox_fs": [
+                        106,
+                        577,
+                        505,
+                        601
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        608,
+                        504,
+                        642
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                607,
+                                505,
+                                621
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        607,
+                                        505,
+                                        621
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Francesca Sargolini, Marianne Fyhn, Torkel Hafting, Bruce L McNaughton, Menno P Witter, May-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                619,
+                                504,
+                                632
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        619,
+                                        504,
+                                        632
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Britt Moser, and Edvard I Moser. Conjunctive representation of position, direction, and velocity",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                631,
+                                343,
+                                641
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        631,
+                                        343,
+                                        641
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in entorhinal cortex. Science, 312(5774):758–762, 2006.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        }
+                    ],
+                    "index": 37,
+                    "bbox_fs": [
+                        105,
+                        607,
+                        505,
+                        641
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        649,
+                        504,
+                        672
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                649,
+                                505,
+                                662
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        649,
+                                        505,
+                                        662
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Francesco Savelli, D Yoganarasimha, and James J Knierim. Influence of boundary removal on the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                660,
+                                476,
+                                672
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        660,
+                                        476,
+                                        672
+                                    ],
+                                    "score": 1.0,
+                                    "content": "spatial representations of the medial entorhinal cortex. Hippocampus, 18(12):1270, 2008.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        }
+                    ],
+                    "index": 39.5,
+                    "bbox_fs": [
+                        106,
+                        649,
+                        505,
+                        672
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        679,
+                        503,
+                        702
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                679,
+                                504,
+                                692
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        679,
+                                        504,
+                                        692
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Andrew M Saxe, James L McClelland, and Surya Ganguli. Exact solutions to the nonlinear dynam-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                690,
+                                328,
+                                703
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        690,
+                                        328,
+                                        703
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ics of learning in deep linear neural networks. 2014.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        }
+                    ],
+                    "index": 41.5,
+                    "bbox_fs": [
+                        106,
+                        679,
+                        504,
+                        703
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        709,
+                        502,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                709,
+                                503,
+                                722
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        709,
+                                        503,
+                                        722
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Trygve Solstad, Charlotte N Boccara, Emilio Kropff, May-Britt Moser, and Edvard I Moser. Repre-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                720,
+                                493,
+                                732
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        720,
+                                        493,
+                                        732
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sentation of geometric borders in the entorhinal cortex. Science, 322(5909):1865–1868, 2008.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        }
+                    ],
+                    "index": 43.5,
+                    "bbox_fs": [
+                        106,
+                        709,
+                        503,
+                        732
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        82,
+                        504,
+                        116
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                81,
+                                505,
+                                96
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        81,
+                                        505,
+                                        96
+                                    ],
+                                    "score": 1.0,
+                                    "content": "H Francis Song, Guangyu R Yang, and Xiao-Jing Wang. Training excitatory-inhibitory recurrent",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                93,
+                                505,
+                                107
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        93,
+                                        505,
+                                        107
+                                    ],
+                                    "score": 1.0,
+                                    "content": "neural networks for cognitive tasks: A simple and flexible framework. PLoS Comput Biol, 12(2):",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                105,
+                                185,
+                                115
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        105,
+                                        185,
+                                        115
+                                    ],
+                                    "score": 1.0,
+                                    "content": "e1004792, 2016.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        }
+                    ],
+                    "index": 1
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        123,
+                        504,
+                        146
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                122,
+                                506,
+                                137
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        122,
+                                        506,
+                                        137
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Kimberly Lauren Stachenfeld, Matthew M Botvinick, and Samuel J Gershman. The hippocampus",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                135,
+                                309,
+                                146
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        135,
+                                        309,
+                                        146
+                                    ],
+                                    "score": 1.0,
+                                    "content": "as a predictive map. bioRxiv, pp. 097170, 2016.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        }
+                    ],
+                    "index": 3.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        153,
+                        504,
+                        176
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                154,
+                                505,
+                                165
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        154,
+                                        505,
+                                        165
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Hanne Stensola, Tor Stensola, Trygve Solstad, Kristian Frøland, May-Britt Moser, and Edvard I",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                164,
+                                437,
+                                177
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        164,
+                                        437,
+                                        177
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Moser. The entorhinal grid map is discretized. Nature, 492(7427):72–78, 2012.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        }
+                    ],
+                    "index": 5.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        183,
+                        504,
+                        216
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                183,
+                                505,
+                                195
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        183,
+                                        505,
+                                        195
+                                    ],
+                                    "score": 1.0,
+                                    "content": "David Sussillo, Mark M Churchland, Matthew T Kaufman, and Krishna V Shenoy. A neural network",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                194,
+                                504,
+                                207
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        194,
+                                        504,
+                                        207
+                                    ],
+                                    "score": 1.0,
+                                    "content": "that finds a naturalistic solution for the production of muscle activity. Nature neuroscience, 18(7):",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                117,
+                                205,
+                                191,
+                                216
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        117,
+                                        205,
+                                        191,
+                                        216
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1025–1033, 2015.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        }
+                    ],
+                    "index": 8
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        223,
+                        504,
+                        247
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                223,
+                                505,
+                                236
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        223,
+                                        505,
+                                        236
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Lucas Theis and Matthias Bethge. Generative image modeling using spatial lstms. In Advances in",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                234,
+                                370,
+                                247
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        234,
+                                        370,
+                                        247
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Neural Information Processing Systems, pp. 1927–1935, 2015.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        }
+                    ],
+                    "index": 10.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        253,
+                        504,
+                        277
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                253,
+                                505,
+                                267
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        253,
+                                        505,
+                                        267
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Xue-Xin Wei, Jason Prentice, and Vijay Balasubramanian. A principle of economy predicts the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                264,
+                                353,
+                                277
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        264,
+                                        353,
+                                        277
+                                    ],
+                                    "score": 1.0,
+                                    "content": "functional architecture of grid cells. Elife, 4:e08362, 2015.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        }
+                    ],
+                    "index": 12.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        283,
+                        504,
+                        306
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                282,
+                                505,
+                                297
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        282,
+                                        505,
+                                        297
+                                    ],
+                                    "score": 1.0,
+                                    "content": "B Willmore and DJ Tolhurst. Characterizing the sparseness of neural codes. Network, 12:255–270,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                114,
+                                293,
+                                143,
+                                307
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        114,
+                                        293,
+                                        143,
+                                        307
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2001.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        }
+                    ],
+                    "index": 14.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        313,
+                        505,
+                        336
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                312,
+                                505,
+                                326
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        312,
+                                        505,
+                                        326
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Tom J Wills, Francesca Cacucci, Neil Burgess, and John O’keefe. Development of the hippocampal",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                325,
+                                411,
+                                337
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        325,
+                                        411,
+                                        337
+                                    ],
+                                    "score": 1.0,
+                                    "content": "cognitive map in preweanling rats. Science, 328(5985):1573–1576, 2010.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        }
+                    ],
+                    "index": 16.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        343,
+                        504,
+                        367
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                343,
+                                505,
+                                356
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        343,
+                                        505,
+                                        356
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Shawn S. Winter, Benjamin J. Clark, and Jeffrey S. Taube. Disruption of the head direction cell",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                355,
+                                483,
+                                367
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        355,
+                                        483,
+                                        367
+                                    ],
+                                    "score": 1.0,
+                                    "content": "network impairs the parahippocampal grid cell signal. Science, 347(6224):870–874, 2015a.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        }
+                    ],
+                    "index": 18.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        372,
+                        504,
+                        397
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                373,
+                                505,
+                                386
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        373,
+                                        505,
+                                        386
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Shawn S. Winter, Max L. Mehlman, Benjamin J. Clark, and Jeffrey S. Taube. Passive transport",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                384,
+                                488,
+                                397
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        384,
+                                        488,
+                                        397
+                                    ],
+                                    "score": 1.0,
+                                    "content": "disrupts grid signals in the parahippocampal cortex. Current Biology, 25:2493–2502, 2015b.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        }
+                    ],
+                    "index": 20.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        403,
+                        504,
+                        426
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                402,
+                                505,
+                                416
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        402,
+                                        505,
+                                        416
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Daniel LK Yamins and James J DiCarlo. Using goal-driven deep learning models to understand",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                415,
+                                356,
+                                426
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        415,
+                                        356,
+                                        426
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sensory cortex. Nature neuroscience, 19(3):356–365, 2016.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        }
+                    ],
+                    "index": 22.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        433,
+                        504,
+                        467
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                433,
+                                506,
+                                446
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        433,
+                                        506,
+                                        446
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Daniel LK Yamins, Ha Hong, Charles F Cadieu, Ethan A Solomon, Darren Seibert, and James J",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                444,
+                                506,
+                                456
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        444,
+                                        506,
+                                        456
+                                    ],
+                                    "score": 1.0,
+                                    "content": "DiCarlo. Performance-optimized hierarchical models predict neural responses in higher visual",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                455,
+                                460,
+                                468
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        455,
+                                        460,
+                                        468
+                                    ],
+                                    "score": 1.0,
+                                    "content": "cortex. Proceedings of the National Academy of Sciences, 111(23):8619–8624, 2014.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        }
+                    ],
+                    "index": 25
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        474,
+                        504,
+                        497
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                474,
+                                505,
+                                487
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        474,
+                                        505,
+                                        487
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Michael M Yartsev, Menno P Witter, and Nachum Ulanovsky. Grid cells without theta oscillations",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                117,
+                                485,
+                                386,
+                                497
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        117,
+                                        485,
+                                        386,
+                                        497
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in the entorhinal cortex of bats. Nature, 479(7371):103–107, 2011.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        }
+                    ],
+                    "index": 27.5
+                }
+            ],
+            "page_idx": 12,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        107,
+                        27,
+                        293,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                26,
+                                294,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        26,
+                                        294,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Published as a conference paper at ICLR 2018",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        300,
+                        751,
+                        311,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                299,
+                                750,
+                                313,
+                                764
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        299,
+                                        750,
+                                        313,
+                                        764
+                                    ],
+                                    "score": 1.0,
+                                    "content": "13",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        82,
+                        504,
+                        116
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                81,
+                                505,
+                                96
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        81,
+                                        505,
+                                        96
+                                    ],
+                                    "score": 1.0,
+                                    "content": "H Francis Song, Guangyu R Yang, and Xiao-Jing Wang. Training excitatory-inhibitory recurrent",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                93,
+                                505,
+                                107
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        93,
+                                        505,
+                                        107
+                                    ],
+                                    "score": 1.0,
+                                    "content": "neural networks for cognitive tasks: A simple and flexible framework. PLoS Comput Biol, 12(2):",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                105,
+                                185,
+                                115
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        105,
+                                        185,
+                                        115
+                                    ],
+                                    "score": 1.0,
+                                    "content": "e1004792, 2016.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        }
+                    ],
+                    "index": 1,
+                    "bbox_fs": [
+                        105,
+                        81,
+                        505,
+                        115
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        123,
+                        504,
+                        146
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                122,
+                                506,
+                                137
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        122,
+                                        506,
+                                        137
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Kimberly Lauren Stachenfeld, Matthew M Botvinick, and Samuel J Gershman. The hippocampus",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                135,
+                                309,
+                                146
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        135,
+                                        309,
+                                        146
+                                    ],
+                                    "score": 1.0,
+                                    "content": "as a predictive map. bioRxiv, pp. 097170, 2016.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        }
+                    ],
+                    "index": 3.5,
+                    "bbox_fs": [
+                        105,
+                        122,
+                        506,
+                        146
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        153,
+                        504,
+                        176
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                154,
+                                505,
+                                165
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        154,
+                                        505,
+                                        165
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Hanne Stensola, Tor Stensola, Trygve Solstad, Kristian Frøland, May-Britt Moser, and Edvard I",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                164,
+                                437,
+                                177
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        164,
+                                        437,
+                                        177
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Moser. The entorhinal grid map is discretized. Nature, 492(7427):72–78, 2012.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        }
+                    ],
+                    "index": 5.5,
+                    "bbox_fs": [
+                        106,
+                        154,
+                        505,
+                        177
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        183,
+                        504,
+                        216
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                183,
+                                505,
+                                195
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        183,
+                                        505,
+                                        195
+                                    ],
+                                    "score": 1.0,
+                                    "content": "David Sussillo, Mark M Churchland, Matthew T Kaufman, and Krishna V Shenoy. A neural network",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                194,
+                                504,
+                                207
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        194,
+                                        504,
+                                        207
+                                    ],
+                                    "score": 1.0,
+                                    "content": "that finds a naturalistic solution for the production of muscle activity. Nature neuroscience, 18(7):",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                117,
+                                205,
+                                191,
+                                216
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        117,
+                                        205,
+                                        191,
+                                        216
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1025–1033, 2015.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        }
+                    ],
+                    "index": 8,
+                    "bbox_fs": [
+                        106,
+                        183,
+                        505,
+                        216
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        223,
+                        504,
+                        247
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                223,
+                                505,
+                                236
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        223,
+                                        505,
+                                        236
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Lucas Theis and Matthias Bethge. Generative image modeling using spatial lstms. In Advances in",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                234,
+                                370,
+                                247
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        234,
+                                        370,
+                                        247
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Neural Information Processing Systems, pp. 1927–1935, 2015.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        }
+                    ],
+                    "index": 10.5,
+                    "bbox_fs": [
+                        106,
+                        223,
+                        505,
+                        247
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        253,
+                        504,
+                        277
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                253,
+                                505,
+                                267
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        253,
+                                        505,
+                                        267
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Xue-Xin Wei, Jason Prentice, and Vijay Balasubramanian. A principle of economy predicts the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                264,
+                                353,
+                                277
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        264,
+                                        353,
+                                        277
+                                    ],
+                                    "score": 1.0,
+                                    "content": "functional architecture of grid cells. Elife, 4:e08362, 2015.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        }
+                    ],
+                    "index": 12.5,
+                    "bbox_fs": [
+                        105,
+                        253,
+                        505,
+                        277
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        283,
+                        504,
+                        306
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                282,
+                                505,
+                                297
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        282,
+                                        505,
+                                        297
+                                    ],
+                                    "score": 1.0,
+                                    "content": "B Willmore and DJ Tolhurst. Characterizing the sparseness of neural codes. Network, 12:255–270,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                114,
+                                293,
+                                143,
+                                307
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        114,
+                                        293,
+                                        143,
+                                        307
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2001.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        }
+                    ],
+                    "index": 14.5,
+                    "bbox_fs": [
+                        104,
+                        282,
+                        505,
+                        307
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        313,
+                        505,
+                        336
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                312,
+                                505,
+                                326
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        312,
+                                        505,
+                                        326
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Tom J Wills, Francesca Cacucci, Neil Burgess, and John O’keefe. Development of the hippocampal",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                325,
+                                411,
+                                337
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        325,
+                                        411,
+                                        337
+                                    ],
+                                    "score": 1.0,
+                                    "content": "cognitive map in preweanling rats. Science, 328(5985):1573–1576, 2010.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        }
+                    ],
+                    "index": 16.5,
+                    "bbox_fs": [
+                        106,
+                        312,
+                        505,
+                        337
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        343,
+                        504,
+                        367
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                343,
+                                505,
+                                356
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        343,
+                                        505,
+                                        356
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Shawn S. Winter, Benjamin J. Clark, and Jeffrey S. Taube. Disruption of the head direction cell",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                355,
+                                483,
+                                367
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        355,
+                                        483,
+                                        367
+                                    ],
+                                    "score": 1.0,
+                                    "content": "network impairs the parahippocampal grid cell signal. Science, 347(6224):870–874, 2015a.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        }
+                    ],
+                    "index": 18.5,
+                    "bbox_fs": [
+                        106,
+                        343,
+                        505,
+                        367
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        372,
+                        504,
+                        397
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                373,
+                                505,
+                                386
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        373,
+                                        505,
+                                        386
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Shawn S. Winter, Max L. Mehlman, Benjamin J. Clark, and Jeffrey S. Taube. Passive transport",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                384,
+                                488,
+                                397
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        384,
+                                        488,
+                                        397
+                                    ],
+                                    "score": 1.0,
+                                    "content": "disrupts grid signals in the parahippocampal cortex. Current Biology, 25:2493–2502, 2015b.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        }
+                    ],
+                    "index": 20.5,
+                    "bbox_fs": [
+                        106,
+                        373,
+                        505,
+                        397
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        403,
+                        504,
+                        426
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                402,
+                                505,
+                                416
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        402,
+                                        505,
+                                        416
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Daniel LK Yamins and James J DiCarlo. Using goal-driven deep learning models to understand",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                415,
+                                356,
+                                426
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        415,
+                                        356,
+                                        426
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sensory cortex. Nature neuroscience, 19(3):356–365, 2016.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        }
+                    ],
+                    "index": 22.5,
+                    "bbox_fs": [
+                        105,
+                        402,
+                        505,
+                        426
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        433,
+                        504,
+                        467
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                433,
+                                506,
+                                446
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        433,
+                                        506,
+                                        446
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Daniel LK Yamins, Ha Hong, Charles F Cadieu, Ethan A Solomon, Darren Seibert, and James J",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                444,
+                                506,
+                                456
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        444,
+                                        506,
+                                        456
+                                    ],
+                                    "score": 1.0,
+                                    "content": "DiCarlo. Performance-optimized hierarchical models predict neural responses in higher visual",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                455,
+                                460,
+                                468
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        455,
+                                        460,
+                                        468
+                                    ],
+                                    "score": 1.0,
+                                    "content": "cortex. Proceedings of the National Academy of Sciences, 111(23):8619–8624, 2014.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        }
+                    ],
+                    "index": 25,
+                    "bbox_fs": [
+                        105,
+                        433,
+                        506,
+                        468
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        474,
+                        504,
+                        497
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                474,
+                                505,
+                                487
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        474,
+                                        505,
+                                        487
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Michael M Yartsev, Menno P Witter, and Nachum Ulanovsky. Grid cells without theta oscillations",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                117,
+                                485,
+                                386,
+                                497
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        117,
+                                        485,
+                                        386,
+                                        497
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in the entorhinal cortex of bats. Nature, 479(7371):103–107, 2011.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        }
+                    ],
+                    "index": 27.5,
+                    "bbox_fs": [
+                        105,
+                        474,
+                        505,
+                        497
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        81,
+                        277,
+                        93
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                80,
+                                280,
+                                95
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        80,
+                                        280,
+                                        95
+                                    ],
+                                    "score": 1.0,
+                                    "content": "A TRIANGULAR ENVIRONMENT",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        }
+                    ],
+                    "index": 0
+                },
+                {
+                    "type": "image",
+                    "bbox": [
+                        213,
+                        108,
+                        397,
+                        254
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                213,
+                                108,
+                                397,
+                                254
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        213,
+                                        108,
+                                        397,
+                                        254
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                213,
+                                                108,
+                                                397,
+                                                254
+                                            ],
+                                            "score": 0.938,
+                                            "type": "image",
+                                            "image_path": "9fec0b40f547311967f788a5e33c37ca83be210239a76a72034b231f75f6b016.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 6,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        213,
+                                        108,
+                                        397,
+                                        121.27272727272728
+                                    ],
+                                    "spans": [],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        213,
+                                        121.27272727272728,
+                                        397,
+                                        134.54545454545456
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                },
+                                {
+                                    "bbox": [
+                                        213,
+                                        134.54545454545456,
+                                        397,
+                                        147.81818181818184
+                                    ],
+                                    "spans": [],
+                                    "index": 3
+                                },
+                                {
+                                    "bbox": [
+                                        213,
+                                        147.81818181818184,
+                                        397,
+                                        161.09090909090912
+                                    ],
+                                    "spans": [],
+                                    "index": 4
+                                },
+                                {
+                                    "bbox": [
+                                        213,
+                                        161.09090909090912,
+                                        397,
+                                        174.3636363636364
+                                    ],
+                                    "spans": [],
+                                    "index": 5
+                                },
+                                {
+                                    "bbox": [
+                                        213,
+                                        174.3636363636364,
+                                        397,
+                                        187.63636363636368
+                                    ],
+                                    "spans": [],
+                                    "index": 6
+                                },
+                                {
+                                    "bbox": [
+                                        213,
+                                        187.63636363636368,
+                                        397,
+                                        200.90909090909096
+                                    ],
+                                    "spans": [],
+                                    "index": 7
+                                },
+                                {
+                                    "bbox": [
+                                        213,
+                                        200.90909090909096,
+                                        397,
+                                        214.18181818181824
+                                    ],
+                                    "spans": [],
+                                    "index": 8
+                                },
+                                {
+                                    "bbox": [
+                                        213,
+                                        214.18181818181824,
+                                        397,
+                                        227.45454545454552
+                                    ],
+                                    "spans": [],
+                                    "index": 9
+                                },
+                                {
+                                    "bbox": [
+                                        213,
+                                        227.45454545454552,
+                                        397,
+                                        240.7272727272728
+                                    ],
+                                    "spans": [],
+                                    "index": 10
+                                },
+                                {
+                                    "bbox": [
+                                        213,
+                                        240.7272727272728,
+                                        397,
+                                        254.00000000000009
+                                    ],
+                                    "spans": [],
+                                    "index": 11
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 6
+                },
+                {
+                    "type": "image",
+                    "bbox": [
+                        107,
+                        304,
+                        505,
+                        450
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                107,
+                                304,
+                                505,
+                                450
+                            ],
+                            "group_id": 1,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        304,
+                                        505,
+                                        450
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                107,
+                                                304,
+                                                505,
+                                                450
+                                            ],
+                                            "score": 0.899,
+                                            "type": "image",
+                                            "image_path": "76d7068d3c86954df82911a198a111d047595eeb46564b8603a26da4acf15c38.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 13,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        304,
+                                        505,
+                                        352.6666666666667
+                                    ],
+                                    "spans": [],
+                                    "index": 12
+                                },
+                                {
+                                    "bbox": [
+                                        107,
+                                        352.6666666666667,
+                                        505,
+                                        401.33333333333337
+                                    ],
+                                    "spans": [],
+                                    "index": 13
+                                },
+                                {
+                                    "bbox": [
+                                        107,
+                                        401.33333333333337,
+                                        505,
+                                        450.00000000000006
+                                    ],
+                                    "spans": [],
+                                    "index": 14
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 13
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        459,
+                        505,
+                        493
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                459,
+                                505,
+                                472
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        459,
+                                        505,
+                                        472
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Noise and metabolic cost are important for grid-like representations. The figure on the left shows",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                471,
+                                505,
+                                483
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        471,
+                                        505,
+                                        483
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the spatial responses for a network trained with noise and no metabolic cost. The figure on the right",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                482,
+                                455,
+                                494
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        482,
+                                        455,
+                                        494
+                                    ],
+                                    "score": 1.0,
+                                    "content": "shows the spatial responses for a network trained with no noise and the metabolic cost.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        }
+                    ],
+                    "index": 16
+                }
+            ],
+            "page_idx": 13,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        107,
+                        27,
+                        293,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                26,
+                                294,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        26,
+                                        294,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Published as a conference paper at ICLR 2018",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        300,
+                        751,
+                        311,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                299,
+                                750,
+                                313,
+                                763
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        299,
+                                        750,
+                                        313,
+                                        763
+                                    ],
+                                    "score": 1.0,
+                                    "content": "14",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        81,
+                        277,
+                        93
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                80,
+                                280,
+                                95
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        80,
+                                        280,
+                                        95
+                                    ],
+                                    "score": 1.0,
+                                    "content": "A TRIANGULAR ENVIRONMENT",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        }
+                    ],
+                    "index": 0
+                },
+                {
+                    "type": "image",
+                    "bbox": [
+                        213,
+                        108,
+                        397,
+                        254
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                213,
+                                108,
+                                397,
+                                254
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        213,
+                                        108,
+                                        397,
+                                        254
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                213,
+                                                108,
+                                                397,
+                                                254
+                                            ],
+                                            "score": 0.938,
+                                            "type": "image",
+                                            "image_path": "9fec0b40f547311967f788a5e33c37ca83be210239a76a72034b231f75f6b016.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 6,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        213,
+                                        108,
+                                        397,
+                                        121.27272727272728
+                                    ],
+                                    "spans": [],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        213,
+                                        121.27272727272728,
+                                        397,
+                                        134.54545454545456
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                },
+                                {
+                                    "bbox": [
+                                        213,
+                                        134.54545454545456,
+                                        397,
+                                        147.81818181818184
+                                    ],
+                                    "spans": [],
+                                    "index": 3
+                                },
+                                {
+                                    "bbox": [
+                                        213,
+                                        147.81818181818184,
+                                        397,
+                                        161.09090909090912
+                                    ],
+                                    "spans": [],
+                                    "index": 4
+                                },
+                                {
+                                    "bbox": [
+                                        213,
+                                        161.09090909090912,
+                                        397,
+                                        174.3636363636364
+                                    ],
+                                    "spans": [],
+                                    "index": 5
+                                },
+                                {
+                                    "bbox": [
+                                        213,
+                                        174.3636363636364,
+                                        397,
+                                        187.63636363636368
+                                    ],
+                                    "spans": [],
+                                    "index": 6
+                                },
+                                {
+                                    "bbox": [
+                                        213,
+                                        187.63636363636368,
+                                        397,
+                                        200.90909090909096
+                                    ],
+                                    "spans": [],
+                                    "index": 7
+                                },
+                                {
+                                    "bbox": [
+                                        213,
+                                        200.90909090909096,
+                                        397,
+                                        214.18181818181824
+                                    ],
+                                    "spans": [],
+                                    "index": 8
+                                },
+                                {
+                                    "bbox": [
+                                        213,
+                                        214.18181818181824,
+                                        397,
+                                        227.45454545454552
+                                    ],
+                                    "spans": [],
+                                    "index": 9
+                                },
+                                {
+                                    "bbox": [
+                                        213,
+                                        227.45454545454552,
+                                        397,
+                                        240.7272727272728
+                                    ],
+                                    "spans": [],
+                                    "index": 10
+                                },
+                                {
+                                    "bbox": [
+                                        213,
+                                        240.7272727272728,
+                                        397,
+                                        254.00000000000009
+                                    ],
+                                    "spans": [],
+                                    "index": 11
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 6
+                },
+                {
+                    "type": "image",
+                    "bbox": [
+                        107,
+                        304,
+                        505,
+                        450
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                107,
+                                304,
+                                505,
+                                450
+                            ],
+                            "group_id": 1,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        304,
+                                        505,
+                                        450
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                107,
+                                                304,
+                                                505,
+                                                450
+                                            ],
+                                            "score": 0.899,
+                                            "type": "image",
+                                            "image_path": "76d7068d3c86954df82911a198a111d047595eeb46564b8603a26da4acf15c38.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 13,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        304,
+                                        505,
+                                        352.6666666666667
+                                    ],
+                                    "spans": [],
+                                    "index": 12
+                                },
+                                {
+                                    "bbox": [
+                                        107,
+                                        352.6666666666667,
+                                        505,
+                                        401.33333333333337
+                                    ],
+                                    "spans": [],
+                                    "index": 13
+                                },
+                                {
+                                    "bbox": [
+                                        107,
+                                        401.33333333333337,
+                                        505,
+                                        450.00000000000006
+                                    ],
+                                    "spans": [],
+                                    "index": 14
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 13
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        459,
+                        505,
+                        493
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                459,
+                                505,
+                                472
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        459,
+                                        505,
+                                        472
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Noise and metabolic cost are important for grid-like representations. The figure on the left shows",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                471,
+                                505,
+                                483
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        471,
+                                        505,
+                                        483
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the spatial responses for a network trained with noise and no metabolic cost. The figure on the right",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                482,
+                                455,
+                                494
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        482,
+                                        455,
+                                        494
+                                    ],
+                                    "score": 1.0,
+                                    "content": "shows the spatial responses for a network trained with no noise and the metabolic cost.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        }
+                    ],
+                    "index": 16,
+                    "bbox_fs": [
+                        105,
+                        459,
+                        505,
+                        494
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "image",
+                    "bbox": [
+                        192,
+                        76,
+                        392,
+                        443
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                192,
+                                76,
+                                392,
+                                443
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        192,
+                                        76,
+                                        392,
+                                        443
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                192,
+                                                76,
+                                                392,
+                                                443
+                                            ],
+                                            "score": 0.832,
+                                            "type": "image",
+                                            "image_path": "ace176f80461890db9b194a0b86e5c7936f97026efec6348b0297c43783b3061.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 17.5,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        192,
+                                        76,
+                                        392,
+                                        86.19444444444444
+                                    ],
+                                    "spans": [],
+                                    "index": 0
+                                },
+                                {
+                                    "bbox": [
+                                        192,
+                                        86.19444444444444,
+                                        392,
+                                        96.38888888888889
+                                    ],
+                                    "spans": [],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        192,
+                                        96.38888888888889,
+                                        392,
+                                        106.58333333333333
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                },
+                                {
+                                    "bbox": [
+                                        192,
+                                        106.58333333333333,
+                                        392,
+                                        116.77777777777777
+                                    ],
+                                    "spans": [],
+                                    "index": 3
+                                },
+                                {
+                                    "bbox": [
+                                        192,
+                                        116.77777777777777,
+                                        392,
+                                        126.97222222222221
+                                    ],
+                                    "spans": [],
+                                    "index": 4
+                                },
+                                {
+                                    "bbox": [
+                                        192,
+                                        126.97222222222221,
+                                        392,
+                                        137.16666666666666
+                                    ],
+                                    "spans": [],
+                                    "index": 5
+                                },
+                                {
+                                    "bbox": [
+                                        192,
+                                        137.16666666666666,
+                                        392,
+                                        147.36111111111111
+                                    ],
+                                    "spans": [],
+                                    "index": 6
+                                },
+                                {
+                                    "bbox": [
+                                        192,
+                                        147.36111111111111,
+                                        392,
+                                        157.55555555555557
+                                    ],
+                                    "spans": [],
+                                    "index": 7
+                                },
+                                {
+                                    "bbox": [
+                                        192,
+                                        157.55555555555557,
+                                        392,
+                                        167.75000000000003
+                                    ],
+                                    "spans": [],
+                                    "index": 8
+                                },
+                                {
+                                    "bbox": [
+                                        192,
+                                        167.75000000000003,
+                                        392,
+                                        177.94444444444449
+                                    ],
+                                    "spans": [],
+                                    "index": 9
+                                },
+                                {
+                                    "bbox": [
+                                        192,
+                                        177.94444444444449,
+                                        392,
+                                        188.13888888888894
+                                    ],
+                                    "spans": [],
+                                    "index": 10
+                                },
+                                {
+                                    "bbox": [
+                                        192,
+                                        188.13888888888894,
+                                        392,
+                                        198.3333333333334
+                                    ],
+                                    "spans": [],
+                                    "index": 11
+                                },
+                                {
+                                    "bbox": [
+                                        192,
+                                        198.3333333333334,
+                                        392,
+                                        208.52777777777786
+                                    ],
+                                    "spans": [],
+                                    "index": 12
+                                },
+                                {
+                                    "bbox": [
+                                        192,
+                                        208.52777777777786,
+                                        392,
+                                        218.7222222222223
+                                    ],
+                                    "spans": [],
+                                    "index": 13
+                                },
+                                {
+                                    "bbox": [
+                                        192,
+                                        218.7222222222223,
+                                        392,
+                                        228.91666666666677
+                                    ],
+                                    "spans": [],
+                                    "index": 14
+                                },
+                                {
+                                    "bbox": [
+                                        192,
+                                        228.91666666666677,
+                                        392,
+                                        239.11111111111123
+                                    ],
+                                    "spans": [],
+                                    "index": 15
+                                },
+                                {
+                                    "bbox": [
+                                        192,
+                                        239.11111111111123,
+                                        392,
+                                        249.30555555555569
+                                    ],
+                                    "spans": [],
+                                    "index": 16
+                                },
+                                {
+                                    "bbox": [
+                                        192,
+                                        249.30555555555569,
+                                        392,
+                                        259.5000000000001
+                                    ],
+                                    "spans": [],
+                                    "index": 17
+                                },
+                                {
+                                    "bbox": [
+                                        192,
+                                        259.5000000000001,
+                                        392,
+                                        269.69444444444457
+                                    ],
+                                    "spans": [],
+                                    "index": 18
+                                },
+                                {
+                                    "bbox": [
+                                        192,
+                                        269.69444444444457,
+                                        392,
+                                        279.888888888889
+                                    ],
+                                    "spans": [],
+                                    "index": 19
+                                },
+                                {
+                                    "bbox": [
+                                        192,
+                                        279.888888888889,
+                                        392,
+                                        290.0833333333335
+                                    ],
+                                    "spans": [],
+                                    "index": 20
+                                },
+                                {
+                                    "bbox": [
+                                        192,
+                                        290.0833333333335,
+                                        392,
+                                        300.27777777777794
+                                    ],
+                                    "spans": [],
+                                    "index": 21
+                                },
+                                {
+                                    "bbox": [
+                                        192,
+                                        300.27777777777794,
+                                        392,
+                                        310.4722222222224
+                                    ],
+                                    "spans": [],
+                                    "index": 22
+                                },
+                                {
+                                    "bbox": [
+                                        192,
+                                        310.4722222222224,
+                                        392,
+                                        320.66666666666686
+                                    ],
+                                    "spans": [],
+                                    "index": 23
+                                },
+                                {
+                                    "bbox": [
+                                        192,
+                                        320.66666666666686,
+                                        392,
+                                        330.8611111111113
+                                    ],
+                                    "spans": [],
+                                    "index": 24
+                                },
+                                {
+                                    "bbox": [
+                                        192,
+                                        330.8611111111113,
+                                        392,
+                                        341.05555555555577
+                                    ],
+                                    "spans": [],
+                                    "index": 25
+                                },
+                                {
+                                    "bbox": [
+                                        192,
+                                        341.05555555555577,
+                                        392,
+                                        351.2500000000002
+                                    ],
+                                    "spans": [],
+                                    "index": 26
+                                },
+                                {
+                                    "bbox": [
+                                        192,
+                                        351.2500000000002,
+                                        392,
+                                        361.4444444444447
+                                    ],
+                                    "spans": [],
+                                    "index": 27
+                                },
+                                {
+                                    "bbox": [
+                                        192,
+                                        361.4444444444447,
+                                        392,
+                                        371.63888888888914
+                                    ],
+                                    "spans": [],
+                                    "index": 28
+                                },
+                                {
+                                    "bbox": [
+                                        192,
+                                        371.63888888888914,
+                                        392,
+                                        381.8333333333336
+                                    ],
+                                    "spans": [],
+                                    "index": 29
+                                },
+                                {
+                                    "bbox": [
+                                        192,
+                                        381.8333333333336,
+                                        392,
+                                        392.02777777777806
+                                    ],
+                                    "spans": [],
+                                    "index": 30
+                                },
+                                {
+                                    "bbox": [
+                                        192,
+                                        392.02777777777806,
+                                        392,
+                                        402.2222222222225
+                                    ],
+                                    "spans": [],
+                                    "index": 31
+                                },
+                                {
+                                    "bbox": [
+                                        192,
+                                        402.2222222222225,
+                                        392,
+                                        412.41666666666697
+                                    ],
+                                    "spans": [],
+                                    "index": 32
+                                },
+                                {
+                                    "bbox": [
+                                        192,
+                                        412.41666666666697,
+                                        392,
+                                        422.6111111111114
+                                    ],
+                                    "spans": [],
+                                    "index": 33
+                                },
+                                {
+                                    "bbox": [
+                                        192,
+                                        422.6111111111114,
+                                        392,
+                                        432.8055555555559
+                                    ],
+                                    "spans": [],
+                                    "index": 34
+                                },
+                                {
+                                    "bbox": [
+                                        192,
+                                        432.8055555555559,
+                                        392,
+                                        443.00000000000034
+                                    ],
+                                    "spans": [],
+                                    "index": 35
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 17.5
+                }
+            ],
+            "page_idx": 14,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        107,
+                        27,
+                        293,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                26,
+                                294,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        26,
+                                        294,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Published as a conference paper at ICLR 2018",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        300,
+                        751,
+                        311,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                299,
+                                750,
+                                312,
+                                764
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        299,
+                                        750,
+                                        312,
+                                        764
+                                    ],
+                                    "score": 1.0,
+                                    "content": "15",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        288,
+                        461,
+                        322,
+                        469
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                286,
+                                459,
+                                324,
+                                470
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        286,
+                                        459,
+                                        324,
+                                        470
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Speed\tinput",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "image",
+                    "bbox": [
+                        192,
+                        76,
+                        392,
+                        443
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                192,
+                                76,
+                                392,
+                                443
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        192,
+                                        76,
+                                        392,
+                                        443
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                192,
+                                                76,
+                                                392,
+                                                443
+                                            ],
+                                            "score": 0.832,
+                                            "type": "image",
+                                            "image_path": "ace176f80461890db9b194a0b86e5c7936f97026efec6348b0297c43783b3061.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 17.5,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        192,
+                                        76,
+                                        392,
+                                        86.19444444444444
+                                    ],
+                                    "spans": [],
+                                    "index": 0
+                                },
+                                {
+                                    "bbox": [
+                                        192,
+                                        86.19444444444444,
+                                        392,
+                                        96.38888888888889
+                                    ],
+                                    "spans": [],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        192,
+                                        96.38888888888889,
+                                        392,
+                                        106.58333333333333
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                },
+                                {
+                                    "bbox": [
+                                        192,
+                                        106.58333333333333,
+                                        392,
+                                        116.77777777777777
+                                    ],
+                                    "spans": [],
+                                    "index": 3
+                                },
+                                {
+                                    "bbox": [
+                                        192,
+                                        116.77777777777777,
+                                        392,
+                                        126.97222222222221
+                                    ],
+                                    "spans": [],
+                                    "index": 4
+                                },
+                                {
+                                    "bbox": [
+                                        192,
+                                        126.97222222222221,
+                                        392,
+                                        137.16666666666666
+                                    ],
+                                    "spans": [],
+                                    "index": 5
+                                },
+                                {
+                                    "bbox": [
+                                        192,
+                                        137.16666666666666,
+                                        392,
+                                        147.36111111111111
+                                    ],
+                                    "spans": [],
+                                    "index": 6
+                                },
+                                {
+                                    "bbox": [
+                                        192,
+                                        147.36111111111111,
+                                        392,
+                                        157.55555555555557
+                                    ],
+                                    "spans": [],
+                                    "index": 7
+                                },
+                                {
+                                    "bbox": [
+                                        192,
+                                        157.55555555555557,
+                                        392,
+                                        167.75000000000003
+                                    ],
+                                    "spans": [],
+                                    "index": 8
+                                },
+                                {
+                                    "bbox": [
+                                        192,
+                                        167.75000000000003,
+                                        392,
+                                        177.94444444444449
+                                    ],
+                                    "spans": [],
+                                    "index": 9
+                                },
+                                {
+                                    "bbox": [
+                                        192,
+                                        177.94444444444449,
+                                        392,
+                                        188.13888888888894
+                                    ],
+                                    "spans": [],
+                                    "index": 10
+                                },
+                                {
+                                    "bbox": [
+                                        192,
+                                        188.13888888888894,
+                                        392,
+                                        198.3333333333334
+                                    ],
+                                    "spans": [],
+                                    "index": 11
+                                },
+                                {
+                                    "bbox": [
+                                        192,
+                                        198.3333333333334,
+                                        392,
+                                        208.52777777777786
+                                    ],
+                                    "spans": [],
+                                    "index": 12
+                                },
+                                {
+                                    "bbox": [
+                                        192,
+                                        208.52777777777786,
+                                        392,
+                                        218.7222222222223
+                                    ],
+                                    "spans": [],
+                                    "index": 13
+                                },
+                                {
+                                    "bbox": [
+                                        192,
+                                        218.7222222222223,
+                                        392,
+                                        228.91666666666677
+                                    ],
+                                    "spans": [],
+                                    "index": 14
+                                },
+                                {
+                                    "bbox": [
+                                        192,
+                                        228.91666666666677,
+                                        392,
+                                        239.11111111111123
+                                    ],
+                                    "spans": [],
+                                    "index": 15
+                                },
+                                {
+                                    "bbox": [
+                                        192,
+                                        239.11111111111123,
+                                        392,
+                                        249.30555555555569
+                                    ],
+                                    "spans": [],
+                                    "index": 16
+                                },
+                                {
+                                    "bbox": [
+                                        192,
+                                        249.30555555555569,
+                                        392,
+                                        259.5000000000001
+                                    ],
+                                    "spans": [],
+                                    "index": 17
+                                },
+                                {
+                                    "bbox": [
+                                        192,
+                                        259.5000000000001,
+                                        392,
+                                        269.69444444444457
+                                    ],
+                                    "spans": [],
+                                    "index": 18
+                                },
+                                {
+                                    "bbox": [
+                                        192,
+                                        269.69444444444457,
+                                        392,
+                                        279.888888888889
+                                    ],
+                                    "spans": [],
+                                    "index": 19
+                                },
+                                {
+                                    "bbox": [
+                                        192,
+                                        279.888888888889,
+                                        392,
+                                        290.0833333333335
+                                    ],
+                                    "spans": [],
+                                    "index": 20
+                                },
+                                {
+                                    "bbox": [
+                                        192,
+                                        290.0833333333335,
+                                        392,
+                                        300.27777777777794
+                                    ],
+                                    "spans": [],
+                                    "index": 21
+                                },
+                                {
+                                    "bbox": [
+                                        192,
+                                        300.27777777777794,
+                                        392,
+                                        310.4722222222224
+                                    ],
+                                    "spans": [],
+                                    "index": 22
+                                },
+                                {
+                                    "bbox": [
+                                        192,
+                                        310.4722222222224,
+                                        392,
+                                        320.66666666666686
+                                    ],
+                                    "spans": [],
+                                    "index": 23
+                                },
+                                {
+                                    "bbox": [
+                                        192,
+                                        320.66666666666686,
+                                        392,
+                                        330.8611111111113
+                                    ],
+                                    "spans": [],
+                                    "index": 24
+                                },
+                                {
+                                    "bbox": [
+                                        192,
+                                        330.8611111111113,
+                                        392,
+                                        341.05555555555577
+                                    ],
+                                    "spans": [],
+                                    "index": 25
+                                },
+                                {
+                                    "bbox": [
+                                        192,
+                                        341.05555555555577,
+                                        392,
+                                        351.2500000000002
+                                    ],
+                                    "spans": [],
+                                    "index": 26
+                                },
+                                {
+                                    "bbox": [
+                                        192,
+                                        351.2500000000002,
+                                        392,
+                                        361.4444444444447
+                                    ],
+                                    "spans": [],
+                                    "index": 27
+                                },
+                                {
+                                    "bbox": [
+                                        192,
+                                        361.4444444444447,
+                                        392,
+                                        371.63888888888914
+                                    ],
+                                    "spans": [],
+                                    "index": 28
+                                },
+                                {
+                                    "bbox": [
+                                        192,
+                                        371.63888888888914,
+                                        392,
+                                        381.8333333333336
+                                    ],
+                                    "spans": [],
+                                    "index": 29
+                                },
+                                {
+                                    "bbox": [
+                                        192,
+                                        381.8333333333336,
+                                        392,
+                                        392.02777777777806
+                                    ],
+                                    "spans": [],
+                                    "index": 30
+                                },
+                                {
+                                    "bbox": [
+                                        192,
+                                        392.02777777777806,
+                                        392,
+                                        402.2222222222225
+                                    ],
+                                    "spans": [],
+                                    "index": 31
+                                },
+                                {
+                                    "bbox": [
+                                        192,
+                                        402.2222222222225,
+                                        392,
+                                        412.41666666666697
+                                    ],
+                                    "spans": [],
+                                    "index": 32
+                                },
+                                {
+                                    "bbox": [
+                                        192,
+                                        412.41666666666697,
+                                        392,
+                                        422.6111111111114
+                                    ],
+                                    "spans": [],
+                                    "index": 33
+                                },
+                                {
+                                    "bbox": [
+                                        192,
+                                        422.6111111111114,
+                                        392,
+                                        432.8055555555559
+                                    ],
+                                    "spans": [],
+                                    "index": 34
+                                },
+                                {
+                                    "bbox": [
+                                        192,
+                                        432.8055555555559,
+                                        392,
+                                        443.00000000000034
+                                    ],
+                                    "spans": [],
+                                    "index": 35
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 17.5
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        81,
+                        286,
+                        93
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                81,
+                                287,
+                                95
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        81,
+                                        287,
+                                        95
+                                    ],
+                                    "score": 1.0,
+                                    "content": "B RECTANGULAR ENVIRONMENT",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        }
+                    ],
+                    "index": 0
+                },
+                {
+                    "type": "image",
+                    "bbox": [
+                        197,
+                        108,
+                        398,
+                        460
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                197,
+                                108,
+                                398,
+                                460
+                            ],
+                            "group_id": 1,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        197,
+                                        108,
+                                        398,
+                                        460
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                197,
+                                                108,
+                                                398,
+                                                460
+                                            ],
+                                            "score": 0.128,
+                                            "type": "image",
+                                            "image_path": "fad2d973408af95191597a5e409766654c8b3a9862f0ad4f2a8de04f639ad5ee.jpg"
+                                        }
+                                    ]
+                                },
+                                {
+                                    "bbox": [
+                                        189,
+                                        237,
+                                        394,
+                                        662
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                189,
+                                                237,
+                                                394,
+                                                662
+                                            ],
+                                            "score": 0.419,
+                                            "type": "image",
+                                            "image_path": "97fb3c206da8fbace0f4d8eeba7efe4308290448d5d9e2cba584838472b30abd.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 18.0,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        197,
+                                        108,
+                                        398,
+                                        120.57142857142857
+                                    ],
+                                    "spans": [],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        197,
+                                        120.57142857142857,
+                                        398,
+                                        133.14285714285714
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                },
+                                {
+                                    "bbox": [
+                                        197,
+                                        133.14285714285714,
+                                        398,
+                                        145.71428571428572
+                                    ],
+                                    "spans": [],
+                                    "index": 3
+                                },
+                                {
+                                    "bbox": [
+                                        197,
+                                        145.71428571428572,
+                                        398,
+                                        158.2857142857143
+                                    ],
+                                    "spans": [],
+                                    "index": 4
+                                },
+                                {
+                                    "bbox": [
+                                        197,
+                                        158.2857142857143,
+                                        398,
+                                        170.8571428571429
+                                    ],
+                                    "spans": [],
+                                    "index": 5
+                                },
+                                {
+                                    "bbox": [
+                                        197,
+                                        170.8571428571429,
+                                        398,
+                                        183.42857142857147
+                                    ],
+                                    "spans": [],
+                                    "index": 6
+                                },
+                                {
+                                    "bbox": [
+                                        197,
+                                        183.42857142857147,
+                                        398,
+                                        196.00000000000006
+                                    ],
+                                    "spans": [],
+                                    "index": 7
+                                },
+                                {
+                                    "bbox": [
+                                        197,
+                                        196.00000000000006,
+                                        398,
+                                        208.57142857142864
+                                    ],
+                                    "spans": [],
+                                    "index": 8
+                                },
+                                {
+                                    "bbox": [
+                                        197,
+                                        208.57142857142864,
+                                        398,
+                                        221.14285714285722
+                                    ],
+                                    "spans": [],
+                                    "index": 9
+                                },
+                                {
+                                    "bbox": [
+                                        197,
+                                        221.14285714285722,
+                                        398,
+                                        233.7142857142858
+                                    ],
+                                    "spans": [],
+                                    "index": 10
+                                },
+                                {
+                                    "bbox": [
+                                        197,
+                                        233.7142857142858,
+                                        398,
+                                        246.2857142857144
+                                    ],
+                                    "spans": [],
+                                    "index": 11
+                                },
+                                {
+                                    "bbox": [
+                                        197,
+                                        246.2857142857144,
+                                        398,
+                                        258.85714285714295
+                                    ],
+                                    "spans": [],
+                                    "index": 13
+                                },
+                                {
+                                    "bbox": [
+                                        197,
+                                        258.85714285714295,
+                                        398,
+                                        271.4285714285715
+                                    ],
+                                    "spans": [],
+                                    "index": 15
+                                },
+                                {
+                                    "bbox": [
+                                        197,
+                                        271.4285714285715,
+                                        398,
+                                        284.00000000000006
+                                    ],
+                                    "spans": [],
+                                    "index": 17
+                                },
+                                {
+                                    "bbox": [
+                                        197,
+                                        284.00000000000006,
+                                        398,
+                                        296.5714285714286
+                                    ],
+                                    "spans": [],
+                                    "index": 19
+                                },
+                                {
+                                    "bbox": [
+                                        197,
+                                        296.5714285714286,
+                                        398,
+                                        309.14285714285717
+                                    ],
+                                    "spans": [],
+                                    "index": 21
+                                },
+                                {
+                                    "bbox": [
+                                        197,
+                                        309.14285714285717,
+                                        398,
+                                        321.7142857142857
+                                    ],
+                                    "spans": [],
+                                    "index": 23
+                                },
+                                {
+                                    "bbox": [
+                                        197,
+                                        321.7142857142857,
+                                        398,
+                                        334.2857142857143
+                                    ],
+                                    "spans": [],
+                                    "index": 25
+                                },
+                                {
+                                    "bbox": [
+                                        197,
+                                        334.2857142857143,
+                                        398,
+                                        346.85714285714283
+                                    ],
+                                    "spans": [],
+                                    "index": 27
+                                },
+                                {
+                                    "bbox": [
+                                        197,
+                                        346.85714285714283,
+                                        398,
+                                        359.4285714285714
+                                    ],
+                                    "spans": [],
+                                    "index": 29
+                                },
+                                {
+                                    "bbox": [
+                                        197,
+                                        359.4285714285714,
+                                        398,
+                                        371.99999999999994
+                                    ],
+                                    "spans": [],
+                                    "index": 31
+                                },
+                                {
+                                    "bbox": [
+                                        197,
+                                        371.99999999999994,
+                                        398,
+                                        384.5714285714285
+                                    ],
+                                    "spans": [],
+                                    "index": 33
+                                },
+                                {
+                                    "bbox": [
+                                        197,
+                                        384.5714285714285,
+                                        398,
+                                        397.14285714285705
+                                    ],
+                                    "spans": [],
+                                    "index": 35
+                                },
+                                {
+                                    "bbox": [
+                                        197,
+                                        397.14285714285705,
+                                        398,
+                                        409.7142857142856
+                                    ],
+                                    "spans": [],
+                                    "index": 37
+                                },
+                                {
+                                    "bbox": [
+                                        197,
+                                        409.7142857142856,
+                                        398,
+                                        422.28571428571416
+                                    ],
+                                    "spans": [],
+                                    "index": 39
+                                },
+                                {
+                                    "bbox": [
+                                        197,
+                                        422.28571428571416,
+                                        398,
+                                        434.8571428571427
+                                    ],
+                                    "spans": [],
+                                    "index": 41
+                                },
+                                {
+                                    "bbox": [
+                                        197,
+                                        434.8571428571427,
+                                        398,
+                                        447.4285714285713
+                                    ],
+                                    "spans": [],
+                                    "index": 43
+                                },
+                                {
+                                    "bbox": [
+                                        197,
+                                        447.4285714285713,
+                                        398,
+                                        459.99999999999983
+                                    ],
+                                    "spans": [],
+                                    "index": 46
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 18.0
+                },
+                {
+                    "type": "image",
+                    "bbox": [
+                        189,
+                        237,
+                        394,
+                        662
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                189,
+                                237,
+                                394,
+                                662
+                            ],
+                            "group_id": 0,
+                            "lines": [],
+                            "index": 44.5,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        189,
+                                        237,
+                                        394,
+                                        249.5
+                                    ],
+                                    "spans": [],
+                                    "index": 12
+                                },
+                                {
+                                    "bbox": [
+                                        189,
+                                        249.5,
+                                        394,
+                                        262.0
+                                    ],
+                                    "spans": [],
+                                    "index": 14
+                                },
+                                {
+                                    "bbox": [
+                                        189,
+                                        262.0,
+                                        394,
+                                        274.5
+                                    ],
+                                    "spans": [],
+                                    "index": 16
+                                },
+                                {
+                                    "bbox": [
+                                        189,
+                                        274.5,
+                                        394,
+                                        287.0
+                                    ],
+                                    "spans": [],
+                                    "index": 18
+                                },
+                                {
+                                    "bbox": [
+                                        189,
+                                        287.0,
+                                        394,
+                                        299.5
+                                    ],
+                                    "spans": [],
+                                    "index": 20
+                                },
+                                {
+                                    "bbox": [
+                                        189,
+                                        299.5,
+                                        394,
+                                        312.0
+                                    ],
+                                    "spans": [],
+                                    "index": 22
+                                },
+                                {
+                                    "bbox": [
+                                        189,
+                                        312.0,
+                                        394,
+                                        324.5
+                                    ],
+                                    "spans": [],
+                                    "index": 24
+                                },
+                                {
+                                    "bbox": [
+                                        189,
+                                        324.5,
+                                        394,
+                                        337.0
+                                    ],
+                                    "spans": [],
+                                    "index": 26
+                                },
+                                {
+                                    "bbox": [
+                                        189,
+                                        337.0,
+                                        394,
+                                        349.5
+                                    ],
+                                    "spans": [],
+                                    "index": 28
+                                },
+                                {
+                                    "bbox": [
+                                        189,
+                                        349.5,
+                                        394,
+                                        362.0
+                                    ],
+                                    "spans": [],
+                                    "index": 30
+                                },
+                                {
+                                    "bbox": [
+                                        189,
+                                        362.0,
+                                        394,
+                                        374.5
+                                    ],
+                                    "spans": [],
+                                    "index": 32
+                                },
+                                {
+                                    "bbox": [
+                                        189,
+                                        374.5,
+                                        394,
+                                        387.0
+                                    ],
+                                    "spans": [],
+                                    "index": 34
+                                },
+                                {
+                                    "bbox": [
+                                        189,
+                                        387.0,
+                                        394,
+                                        399.5
+                                    ],
+                                    "spans": [],
+                                    "index": 36
+                                },
+                                {
+                                    "bbox": [
+                                        189,
+                                        399.5,
+                                        394,
+                                        412.0
+                                    ],
+                                    "spans": [],
+                                    "index": 38
+                                },
+                                {
+                                    "bbox": [
+                                        189,
+                                        412.0,
+                                        394,
+                                        424.5
+                                    ],
+                                    "spans": [],
+                                    "index": 40
+                                },
+                                {
+                                    "bbox": [
+                                        189,
+                                        424.5,
+                                        394,
+                                        437.0
+                                    ],
+                                    "spans": [],
+                                    "index": 42
+                                },
+                                {
+                                    "bbox": [
+                                        189,
+                                        437.0,
+                                        394,
+                                        449.5
+                                    ],
+                                    "spans": [],
+                                    "index": 44
+                                },
+                                {
+                                    "bbox": [
+                                        189,
+                                        449.5,
+                                        394,
+                                        462.0
+                                    ],
+                                    "spans": [],
+                                    "index": 45
+                                },
+                                {
+                                    "bbox": [
+                                        189,
+                                        462.0,
+                                        394,
+                                        474.5
+                                    ],
+                                    "spans": [],
+                                    "index": 47
+                                },
+                                {
+                                    "bbox": [
+                                        189,
+                                        474.5,
+                                        394,
+                                        487.0
+                                    ],
+                                    "spans": [],
+                                    "index": 48
+                                },
+                                {
+                                    "bbox": [
+                                        189,
+                                        487.0,
+                                        394,
+                                        499.5
+                                    ],
+                                    "spans": [],
+                                    "index": 49
+                                },
+                                {
+                                    "bbox": [
+                                        189,
+                                        499.5,
+                                        394,
+                                        512.0
+                                    ],
+                                    "spans": [],
+                                    "index": 50
+                                },
+                                {
+                                    "bbox": [
+                                        189,
+                                        512.0,
+                                        394,
+                                        524.5
+                                    ],
+                                    "spans": [],
+                                    "index": 51
+                                },
+                                {
+                                    "bbox": [
+                                        189,
+                                        524.5,
+                                        394,
+                                        537.0
+                                    ],
+                                    "spans": [],
+                                    "index": 52
+                                },
+                                {
+                                    "bbox": [
+                                        189,
+                                        537.0,
+                                        394,
+                                        549.5
+                                    ],
+                                    "spans": [],
+                                    "index": 53
+                                },
+                                {
+                                    "bbox": [
+                                        189,
+                                        549.5,
+                                        394,
+                                        562.0
+                                    ],
+                                    "spans": [],
+                                    "index": 54
+                                },
+                                {
+                                    "bbox": [
+                                        189,
+                                        562.0,
+                                        394,
+                                        574.5
+                                    ],
+                                    "spans": [],
+                                    "index": 55
+                                },
+                                {
+                                    "bbox": [
+                                        189,
+                                        574.5,
+                                        394,
+                                        587.0
+                                    ],
+                                    "spans": [],
+                                    "index": 56
+                                },
+                                {
+                                    "bbox": [
+                                        189,
+                                        587.0,
+                                        394,
+                                        599.5
+                                    ],
+                                    "spans": [],
+                                    "index": 57
+                                },
+                                {
+                                    "bbox": [
+                                        189,
+                                        599.5,
+                                        394,
+                                        612.0
+                                    ],
+                                    "spans": [],
+                                    "index": 58
+                                },
+                                {
+                                    "bbox": [
+                                        189,
+                                        612.0,
+                                        394,
+                                        624.5
+                                    ],
+                                    "spans": [],
+                                    "index": 59
+                                },
+                                {
+                                    "bbox": [
+                                        189,
+                                        624.5,
+                                        394,
+                                        637.0
+                                    ],
+                                    "spans": [],
+                                    "index": 60
+                                },
+                                {
+                                    "bbox": [
+                                        189,
+                                        637.0,
+                                        394,
+                                        649.5
+                                    ],
+                                    "spans": [],
+                                    "index": 61
+                                },
+                                {
+                                    "bbox": [
+                                        189,
+                                        649.5,
+                                        394,
+                                        662.0
+                                    ],
+                                    "spans": [],
+                                    "index": 62
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                288,
+                                677,
+                                322,
+                                685
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        287,
+                                        676,
+                                        324,
+                                        687
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                287,
+                                                676,
+                                                324,
+                                                687
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Speed\tinput",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 63
+                                }
+                            ],
+                            "index": 63
+                        }
+                    ],
+                    "index": 53.75
+                }
+            ],
+            "page_idx": 15,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        107,
+                        27,
+                        293,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                26,
+                                294,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        26,
+                                        294,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Published as a conference paper at ICLR 2018",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        300,
+                        751,
+                        311,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                299,
+                                750,
+                                313,
+                                764
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        299,
+                                        750,
+                                        313,
+                                        764
+                                    ],
+                                    "score": 1.0,
+                                    "content": "16",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        81,
+                        286,
+                        93
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                81,
+                                287,
+                                95
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        81,
+                                        287,
+                                        95
+                                    ],
+                                    "score": 1.0,
+                                    "content": "B RECTANGULAR ENVIRONMENT",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        }
+                    ],
+                    "index": 0
+                },
+                {
+                    "type": "image",
+                    "bbox": [
+                        197,
+                        108,
+                        398,
+                        460
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                197,
+                                108,
+                                398,
+                                460
+                            ],
+                            "group_id": 1,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        197,
+                                        108,
+                                        398,
+                                        460
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                197,
+                                                108,
+                                                398,
+                                                460
+                                            ],
+                                            "score": 0.128,
+                                            "type": "image",
+                                            "image_path": "fad2d973408af95191597a5e409766654c8b3a9862f0ad4f2a8de04f639ad5ee.jpg"
+                                        }
+                                    ]
+                                },
+                                {
+                                    "bbox": [
+                                        189,
+                                        237,
+                                        394,
+                                        662
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                189,
+                                                237,
+                                                394,
+                                                662
+                                            ],
+                                            "score": 0.419,
+                                            "type": "image",
+                                            "image_path": "97fb3c206da8fbace0f4d8eeba7efe4308290448d5d9e2cba584838472b30abd.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 18.0,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        197,
+                                        108,
+                                        398,
+                                        120.57142857142857
+                                    ],
+                                    "spans": [],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        197,
+                                        120.57142857142857,
+                                        398,
+                                        133.14285714285714
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                },
+                                {
+                                    "bbox": [
+                                        197,
+                                        133.14285714285714,
+                                        398,
+                                        145.71428571428572
+                                    ],
+                                    "spans": [],
+                                    "index": 3
+                                },
+                                {
+                                    "bbox": [
+                                        197,
+                                        145.71428571428572,
+                                        398,
+                                        158.2857142857143
+                                    ],
+                                    "spans": [],
+                                    "index": 4
+                                },
+                                {
+                                    "bbox": [
+                                        197,
+                                        158.2857142857143,
+                                        398,
+                                        170.8571428571429
+                                    ],
+                                    "spans": [],
+                                    "index": 5
+                                },
+                                {
+                                    "bbox": [
+                                        197,
+                                        170.8571428571429,
+                                        398,
+                                        183.42857142857147
+                                    ],
+                                    "spans": [],
+                                    "index": 6
+                                },
+                                {
+                                    "bbox": [
+                                        197,
+                                        183.42857142857147,
+                                        398,
+                                        196.00000000000006
+                                    ],
+                                    "spans": [],
+                                    "index": 7
+                                },
+                                {
+                                    "bbox": [
+                                        197,
+                                        196.00000000000006,
+                                        398,
+                                        208.57142857142864
+                                    ],
+                                    "spans": [],
+                                    "index": 8
+                                },
+                                {
+                                    "bbox": [
+                                        197,
+                                        208.57142857142864,
+                                        398,
+                                        221.14285714285722
+                                    ],
+                                    "spans": [],
+                                    "index": 9
+                                },
+                                {
+                                    "bbox": [
+                                        197,
+                                        221.14285714285722,
+                                        398,
+                                        233.7142857142858
+                                    ],
+                                    "spans": [],
+                                    "index": 10
+                                },
+                                {
+                                    "bbox": [
+                                        197,
+                                        233.7142857142858,
+                                        398,
+                                        246.2857142857144
+                                    ],
+                                    "spans": [],
+                                    "index": 11
+                                },
+                                {
+                                    "bbox": [
+                                        197,
+                                        246.2857142857144,
+                                        398,
+                                        258.85714285714295
+                                    ],
+                                    "spans": [],
+                                    "index": 13
+                                },
+                                {
+                                    "bbox": [
+                                        197,
+                                        258.85714285714295,
+                                        398,
+                                        271.4285714285715
+                                    ],
+                                    "spans": [],
+                                    "index": 15
+                                },
+                                {
+                                    "bbox": [
+                                        197,
+                                        271.4285714285715,
+                                        398,
+                                        284.00000000000006
+                                    ],
+                                    "spans": [],
+                                    "index": 17
+                                },
+                                {
+                                    "bbox": [
+                                        197,
+                                        284.00000000000006,
+                                        398,
+                                        296.5714285714286
+                                    ],
+                                    "spans": [],
+                                    "index": 19
+                                },
+                                {
+                                    "bbox": [
+                                        197,
+                                        296.5714285714286,
+                                        398,
+                                        309.14285714285717
+                                    ],
+                                    "spans": [],
+                                    "index": 21
+                                },
+                                {
+                                    "bbox": [
+                                        197,
+                                        309.14285714285717,
+                                        398,
+                                        321.7142857142857
+                                    ],
+                                    "spans": [],
+                                    "index": 23
+                                },
+                                {
+                                    "bbox": [
+                                        197,
+                                        321.7142857142857,
+                                        398,
+                                        334.2857142857143
+                                    ],
+                                    "spans": [],
+                                    "index": 25
+                                },
+                                {
+                                    "bbox": [
+                                        197,
+                                        334.2857142857143,
+                                        398,
+                                        346.85714285714283
+                                    ],
+                                    "spans": [],
+                                    "index": 27
+                                },
+                                {
+                                    "bbox": [
+                                        197,
+                                        346.85714285714283,
+                                        398,
+                                        359.4285714285714
+                                    ],
+                                    "spans": [],
+                                    "index": 29
+                                },
+                                {
+                                    "bbox": [
+                                        197,
+                                        359.4285714285714,
+                                        398,
+                                        371.99999999999994
+                                    ],
+                                    "spans": [],
+                                    "index": 31
+                                },
+                                {
+                                    "bbox": [
+                                        197,
+                                        371.99999999999994,
+                                        398,
+                                        384.5714285714285
+                                    ],
+                                    "spans": [],
+                                    "index": 33
+                                },
+                                {
+                                    "bbox": [
+                                        197,
+                                        384.5714285714285,
+                                        398,
+                                        397.14285714285705
+                                    ],
+                                    "spans": [],
+                                    "index": 35
+                                },
+                                {
+                                    "bbox": [
+                                        197,
+                                        397.14285714285705,
+                                        398,
+                                        409.7142857142856
+                                    ],
+                                    "spans": [],
+                                    "index": 37
+                                },
+                                {
+                                    "bbox": [
+                                        197,
+                                        409.7142857142856,
+                                        398,
+                                        422.28571428571416
+                                    ],
+                                    "spans": [],
+                                    "index": 39
+                                },
+                                {
+                                    "bbox": [
+                                        197,
+                                        422.28571428571416,
+                                        398,
+                                        434.8571428571427
+                                    ],
+                                    "spans": [],
+                                    "index": 41
+                                },
+                                {
+                                    "bbox": [
+                                        197,
+                                        434.8571428571427,
+                                        398,
+                                        447.4285714285713
+                                    ],
+                                    "spans": [],
+                                    "index": 43
+                                },
+                                {
+                                    "bbox": [
+                                        197,
+                                        447.4285714285713,
+                                        398,
+                                        459.99999999999983
+                                    ],
+                                    "spans": [],
+                                    "index": 46
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 18.0
+                },
+                {
+                    "type": "image",
+                    "bbox": [
+                        189,
+                        237,
+                        394,
+                        662
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                189,
+                                237,
+                                394,
+                                662
+                            ],
+                            "group_id": 0,
+                            "lines": [],
+                            "index": 44.5,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        189,
+                                        237,
+                                        394,
+                                        249.5
+                                    ],
+                                    "spans": [],
+                                    "index": 12
+                                },
+                                {
+                                    "bbox": [
+                                        189,
+                                        249.5,
+                                        394,
+                                        262.0
+                                    ],
+                                    "spans": [],
+                                    "index": 14
+                                },
+                                {
+                                    "bbox": [
+                                        189,
+                                        262.0,
+                                        394,
+                                        274.5
+                                    ],
+                                    "spans": [],
+                                    "index": 16
+                                },
+                                {
+                                    "bbox": [
+                                        189,
+                                        274.5,
+                                        394,
+                                        287.0
+                                    ],
+                                    "spans": [],
+                                    "index": 18
+                                },
+                                {
+                                    "bbox": [
+                                        189,
+                                        287.0,
+                                        394,
+                                        299.5
+                                    ],
+                                    "spans": [],
+                                    "index": 20
+                                },
+                                {
+                                    "bbox": [
+                                        189,
+                                        299.5,
+                                        394,
+                                        312.0
+                                    ],
+                                    "spans": [],
+                                    "index": 22
+                                },
+                                {
+                                    "bbox": [
+                                        189,
+                                        312.0,
+                                        394,
+                                        324.5
+                                    ],
+                                    "spans": [],
+                                    "index": 24
+                                },
+                                {
+                                    "bbox": [
+                                        189,
+                                        324.5,
+                                        394,
+                                        337.0
+                                    ],
+                                    "spans": [],
+                                    "index": 26
+                                },
+                                {
+                                    "bbox": [
+                                        189,
+                                        337.0,
+                                        394,
+                                        349.5
+                                    ],
+                                    "spans": [],
+                                    "index": 28
+                                },
+                                {
+                                    "bbox": [
+                                        189,
+                                        349.5,
+                                        394,
+                                        362.0
+                                    ],
+                                    "spans": [],
+                                    "index": 30
+                                },
+                                {
+                                    "bbox": [
+                                        189,
+                                        362.0,
+                                        394,
+                                        374.5
+                                    ],
+                                    "spans": [],
+                                    "index": 32
+                                },
+                                {
+                                    "bbox": [
+                                        189,
+                                        374.5,
+                                        394,
+                                        387.0
+                                    ],
+                                    "spans": [],
+                                    "index": 34
+                                },
+                                {
+                                    "bbox": [
+                                        189,
+                                        387.0,
+                                        394,
+                                        399.5
+                                    ],
+                                    "spans": [],
+                                    "index": 36
+                                },
+                                {
+                                    "bbox": [
+                                        189,
+                                        399.5,
+                                        394,
+                                        412.0
+                                    ],
+                                    "spans": [],
+                                    "index": 38
+                                },
+                                {
+                                    "bbox": [
+                                        189,
+                                        412.0,
+                                        394,
+                                        424.5
+                                    ],
+                                    "spans": [],
+                                    "index": 40
+                                },
+                                {
+                                    "bbox": [
+                                        189,
+                                        424.5,
+                                        394,
+                                        437.0
+                                    ],
+                                    "spans": [],
+                                    "index": 42
+                                },
+                                {
+                                    "bbox": [
+                                        189,
+                                        437.0,
+                                        394,
+                                        449.5
+                                    ],
+                                    "spans": [],
+                                    "index": 44
+                                },
+                                {
+                                    "bbox": [
+                                        189,
+                                        449.5,
+                                        394,
+                                        462.0
+                                    ],
+                                    "spans": [],
+                                    "index": 45
+                                },
+                                {
+                                    "bbox": [
+                                        189,
+                                        462.0,
+                                        394,
+                                        474.5
+                                    ],
+                                    "spans": [],
+                                    "index": 47
+                                },
+                                {
+                                    "bbox": [
+                                        189,
+                                        474.5,
+                                        394,
+                                        487.0
+                                    ],
+                                    "spans": [],
+                                    "index": 48
+                                },
+                                {
+                                    "bbox": [
+                                        189,
+                                        487.0,
+                                        394,
+                                        499.5
+                                    ],
+                                    "spans": [],
+                                    "index": 49
+                                },
+                                {
+                                    "bbox": [
+                                        189,
+                                        499.5,
+                                        394,
+                                        512.0
+                                    ],
+                                    "spans": [],
+                                    "index": 50
+                                },
+                                {
+                                    "bbox": [
+                                        189,
+                                        512.0,
+                                        394,
+                                        524.5
+                                    ],
+                                    "spans": [],
+                                    "index": 51
+                                },
+                                {
+                                    "bbox": [
+                                        189,
+                                        524.5,
+                                        394,
+                                        537.0
+                                    ],
+                                    "spans": [],
+                                    "index": 52
+                                },
+                                {
+                                    "bbox": [
+                                        189,
+                                        537.0,
+                                        394,
+                                        549.5
+                                    ],
+                                    "spans": [],
+                                    "index": 53
+                                },
+                                {
+                                    "bbox": [
+                                        189,
+                                        549.5,
+                                        394,
+                                        562.0
+                                    ],
+                                    "spans": [],
+                                    "index": 54
+                                },
+                                {
+                                    "bbox": [
+                                        189,
+                                        562.0,
+                                        394,
+                                        574.5
+                                    ],
+                                    "spans": [],
+                                    "index": 55
+                                },
+                                {
+                                    "bbox": [
+                                        189,
+                                        574.5,
+                                        394,
+                                        587.0
+                                    ],
+                                    "spans": [],
+                                    "index": 56
+                                },
+                                {
+                                    "bbox": [
+                                        189,
+                                        587.0,
+                                        394,
+                                        599.5
+                                    ],
+                                    "spans": [],
+                                    "index": 57
+                                },
+                                {
+                                    "bbox": [
+                                        189,
+                                        599.5,
+                                        394,
+                                        612.0
+                                    ],
+                                    "spans": [],
+                                    "index": 58
+                                },
+                                {
+                                    "bbox": [
+                                        189,
+                                        612.0,
+                                        394,
+                                        624.5
+                                    ],
+                                    "spans": [],
+                                    "index": 59
+                                },
+                                {
+                                    "bbox": [
+                                        189,
+                                        624.5,
+                                        394,
+                                        637.0
+                                    ],
+                                    "spans": [],
+                                    "index": 60
+                                },
+                                {
+                                    "bbox": [
+                                        189,
+                                        637.0,
+                                        394,
+                                        649.5
+                                    ],
+                                    "spans": [],
+                                    "index": 61
+                                },
+                                {
+                                    "bbox": [
+                                        189,
+                                        649.5,
+                                        394,
+                                        662.0
+                                    ],
+                                    "spans": [],
+                                    "index": 62
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                288,
+                                677,
+                                322,
+                                685
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        287,
+                                        676,
+                                        324,
+                                        687
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                287,
+                                                676,
+                                                324,
+                                                687
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Speed\tinput",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 63
+                                }
+                            ],
+                            "index": 63
+                        }
+                    ],
+                    "index": 53.75
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        81,
+                        273,
+                        93
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                79,
+                                276,
+                                97
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        79,
+                                        276,
+                                        97
+                                    ],
+                                    "score": 1.0,
+                                    "content": "C HEXAGONAL ENVIRONMENT",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        }
+                    ],
+                    "index": 0
+                },
+                {
+                    "type": "image",
+                    "bbox": [
+                        213,
+                        108,
+                        398,
+                        255
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                213,
+                                108,
+                                398,
+                                255
+                            ],
+                            "group_id": 1,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        213,
+                                        108,
+                                        398,
+                                        255
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                213,
+                                                108,
+                                                398,
+                                                255
+                                            ],
+                                            "score": 0.861,
+                                            "type": "image",
+                                            "image_path": "d7c955da2b4d05471ca42b23f6a780ea605e2aed77c7b520059872b81c33fead.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 7,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        213,
+                                        108,
+                                        398,
+                                        119.3076923076923
+                                    ],
+                                    "spans": [],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        213,
+                                        119.3076923076923,
+                                        398,
+                                        130.6153846153846
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                },
+                                {
+                                    "bbox": [
+                                        213,
+                                        130.6153846153846,
+                                        398,
+                                        141.92307692307693
+                                    ],
+                                    "spans": [],
+                                    "index": 3
+                                },
+                                {
+                                    "bbox": [
+                                        213,
+                                        141.92307692307693,
+                                        398,
+                                        153.23076923076925
+                                    ],
+                                    "spans": [],
+                                    "index": 4
+                                },
+                                {
+                                    "bbox": [
+                                        213,
+                                        153.23076923076925,
+                                        398,
+                                        164.53846153846158
+                                    ],
+                                    "spans": [],
+                                    "index": 5
+                                },
+                                {
+                                    "bbox": [
+                                        213,
+                                        164.53846153846158,
+                                        398,
+                                        175.8461538461539
+                                    ],
+                                    "spans": [],
+                                    "index": 6
+                                },
+                                {
+                                    "bbox": [
+                                        213,
+                                        175.8461538461539,
+                                        398,
+                                        187.15384615384622
+                                    ],
+                                    "spans": [],
+                                    "index": 7
+                                },
+                                {
+                                    "bbox": [
+                                        213,
+                                        187.15384615384622,
+                                        398,
+                                        198.46153846153854
+                                    ],
+                                    "spans": [],
+                                    "index": 8
+                                },
+                                {
+                                    "bbox": [
+                                        213,
+                                        198.46153846153854,
+                                        398,
+                                        209.76923076923086
+                                    ],
+                                    "spans": [],
+                                    "index": 9
+                                },
+                                {
+                                    "bbox": [
+                                        213,
+                                        209.76923076923086,
+                                        398,
+                                        221.07692307692318
+                                    ],
+                                    "spans": [],
+                                    "index": 10
+                                },
+                                {
+                                    "bbox": [
+                                        213,
+                                        221.07692307692318,
+                                        398,
+                                        232.3846153846155
+                                    ],
+                                    "spans": [],
+                                    "index": 11
+                                },
+                                {
+                                    "bbox": [
+                                        213,
+                                        232.3846153846155,
+                                        398,
+                                        243.69230769230782
+                                    ],
+                                    "spans": [],
+                                    "index": 12
+                                },
+                                {
+                                    "bbox": [
+                                        213,
+                                        243.69230769230782,
+                                        398,
+                                        255.00000000000014
+                                    ],
+                                    "spans": [],
+                                    "index": 13
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 7
+                },
+                {
+                    "type": "image",
+                    "bbox": [
+                        213,
+                        296,
+                        389,
+                        438
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                213,
+                                296,
+                                389,
+                                438
+                            ],
+                            "group_id": 2,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        213,
+                                        296,
+                                        389,
+                                        438
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                213,
+                                                296,
+                                                389,
+                                                438
+                                            ],
+                                            "score": 0.339,
+                                            "type": "image",
+                                            "image_path": "64fe9f96b9155e2bd1b22d7a344f3085ff647a473ca40df4a123eb8758cf0801.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 19.5,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        213,
+                                        296,
+                                        389,
+                                        307.8333333333333
+                                    ],
+                                    "spans": [],
+                                    "index": 14
+                                },
+                                {
+                                    "bbox": [
+                                        213,
+                                        307.8333333333333,
+                                        389,
+                                        319.66666666666663
+                                    ],
+                                    "spans": [],
+                                    "index": 15
+                                },
+                                {
+                                    "bbox": [
+                                        213,
+                                        319.66666666666663,
+                                        389,
+                                        331.49999999999994
+                                    ],
+                                    "spans": [],
+                                    "index": 16
+                                },
+                                {
+                                    "bbox": [
+                                        213,
+                                        331.49999999999994,
+                                        389,
+                                        343.33333333333326
+                                    ],
+                                    "spans": [],
+                                    "index": 17
+                                },
+                                {
+                                    "bbox": [
+                                        213,
+                                        343.33333333333326,
+                                        389,
+                                        355.1666666666666
+                                    ],
+                                    "spans": [],
+                                    "index": 18
+                                },
+                                {
+                                    "bbox": [
+                                        213,
+                                        355.1666666666666,
+                                        389,
+                                        366.9999999999999
+                                    ],
+                                    "spans": [],
+                                    "index": 19
+                                },
+                                {
+                                    "bbox": [
+                                        213,
+                                        366.9999999999999,
+                                        389,
+                                        378.8333333333332
+                                    ],
+                                    "spans": [],
+                                    "index": 20
+                                },
+                                {
+                                    "bbox": [
+                                        213,
+                                        378.8333333333332,
+                                        389,
+                                        390.6666666666665
+                                    ],
+                                    "spans": [],
+                                    "index": 21
+                                },
+                                {
+                                    "bbox": [
+                                        213,
+                                        390.6666666666665,
+                                        389,
+                                        402.49999999999983
+                                    ],
+                                    "spans": [],
+                                    "index": 22
+                                },
+                                {
+                                    "bbox": [
+                                        213,
+                                        402.49999999999983,
+                                        389,
+                                        414.33333333333314
+                                    ],
+                                    "spans": [],
+                                    "index": 23
+                                },
+                                {
+                                    "bbox": [
+                                        213,
+                                        414.33333333333314,
+                                        389,
+                                        426.16666666666646
+                                    ],
+                                    "spans": [],
+                                    "index": 24
+                                },
+                                {
+                                    "bbox": [
+                                        213,
+                                        426.16666666666646,
+                                        389,
+                                        437.9999999999998
+                                    ],
+                                    "spans": [],
+                                    "index": 25
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 19.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        260,
+                        457,
+                        351,
+                        465
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                259,
+                                456,
+                                351,
+                                466
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        259,
+                                        456,
+                                        351,
+                                        466
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Angular\tinput\t(0\tto\t360\tdegrees)",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        }
+                    ],
+                    "index": 26
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        286,
+                        505,
+                        324,
+                        513
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                285,
+                                504,
+                                325,
+                                515
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        285,
+                                        504,
+                                        325,
+                                        515
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Speed\ttuning",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        }
+                    ],
+                    "index": 27
+                },
+                {
+                    "type": "image",
+                    "bbox": [
+                        207,
+                        524,
+                        388,
+                        658
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                207,
+                                524,
+                                388,
+                                658
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        207,
+                                        524,
+                                        388,
+                                        658
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                207,
+                                                524,
+                                                388,
+                                                658
+                                            ],
+                                            "score": 0.43,
+                                            "type": "image",
+                                            "image_path": "7eb735c5ab485558870c8b0720dd627699e69018e8cbacfa58c8d343cfc8a2d1.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 33.5,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        207,
+                                        524,
+                                        388,
+                                        535.1666666666666
+                                    ],
+                                    "spans": [],
+                                    "index": 28
+                                },
+                                {
+                                    "bbox": [
+                                        207,
+                                        535.1666666666666,
+                                        388,
+                                        546.3333333333333
+                                    ],
+                                    "spans": [],
+                                    "index": 29
+                                },
+                                {
+                                    "bbox": [
+                                        207,
+                                        546.3333333333333,
+                                        388,
+                                        557.4999999999999
+                                    ],
+                                    "spans": [],
+                                    "index": 30
+                                },
+                                {
+                                    "bbox": [
+                                        207,
+                                        557.4999999999999,
+                                        388,
+                                        568.6666666666665
+                                    ],
+                                    "spans": [],
+                                    "index": 31
+                                },
+                                {
+                                    "bbox": [
+                                        207,
+                                        568.6666666666665,
+                                        388,
+                                        579.8333333333331
+                                    ],
+                                    "spans": [],
+                                    "index": 32
+                                },
+                                {
+                                    "bbox": [
+                                        207,
+                                        579.8333333333331,
+                                        388,
+                                        590.9999999999998
+                                    ],
+                                    "spans": [],
+                                    "index": 33
+                                },
+                                {
+                                    "bbox": [
+                                        207,
+                                        590.9999999999998,
+                                        388,
+                                        602.1666666666664
+                                    ],
+                                    "spans": [],
+                                    "index": 34
+                                },
+                                {
+                                    "bbox": [
+                                        207,
+                                        602.1666666666664,
+                                        388,
+                                        613.333333333333
+                                    ],
+                                    "spans": [],
+                                    "index": 35
+                                },
+                                {
+                                    "bbox": [
+                                        207,
+                                        613.333333333333,
+                                        388,
+                                        624.4999999999997
+                                    ],
+                                    "spans": [],
+                                    "index": 36
+                                },
+                                {
+                                    "bbox": [
+                                        207,
+                                        624.4999999999997,
+                                        388,
+                                        635.6666666666663
+                                    ],
+                                    "spans": [],
+                                    "index": 37
+                                },
+                                {
+                                    "bbox": [
+                                        207,
+                                        635.6666666666663,
+                                        388,
+                                        646.8333333333329
+                                    ],
+                                    "spans": [],
+                                    "index": 38
+                                },
+                                {
+                                    "bbox": [
+                                        207,
+                                        646.8333333333329,
+                                        388,
+                                        657.9999999999995
+                                    ],
+                                    "spans": [],
+                                    "index": 39
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                288,
+                                677,
+                                322,
+                                685
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        287,
+                                        676,
+                                        324,
+                                        687
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                287,
+                                                676,
+                                                324,
+                                                687
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Speed\tinput",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 40
+                                }
+                            ],
+                            "index": 40
+                        }
+                    ],
+                    "index": 36.75
+                }
+            ],
+            "page_idx": 16,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        107,
+                        27,
+                        293,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                26,
+                                294,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        26,
+                                        294,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Published as a conference paper at ICLR 2018",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        300,
+                        751,
+                        311,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                299,
+                                750,
+                                312,
+                                764
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        299,
+                                        750,
+                                        312,
+                                        764
+                                    ],
+                                    "score": 1.0,
+                                    "content": "17",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        195,
+                        339,
+                        204,
+                        404
+                    ],
+                    "lines": []
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        195,
+                        559,
+                        204,
+                        624
+                    ],
+                    "lines": []
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        81,
+                        273,
+                        93
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                79,
+                                276,
+                                97
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        79,
+                                        276,
+                                        97
+                                    ],
+                                    "score": 1.0,
+                                    "content": "C HEXAGONAL ENVIRONMENT",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        }
+                    ],
+                    "index": 0
+                },
+                {
+                    "type": "image",
+                    "bbox": [
+                        213,
+                        108,
+                        398,
+                        255
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                213,
+                                108,
+                                398,
+                                255
+                            ],
+                            "group_id": 1,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        213,
+                                        108,
+                                        398,
+                                        255
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                213,
+                                                108,
+                                                398,
+                                                255
+                                            ],
+                                            "score": 0.861,
+                                            "type": "image",
+                                            "image_path": "d7c955da2b4d05471ca42b23f6a780ea605e2aed77c7b520059872b81c33fead.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 7,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        213,
+                                        108,
+                                        398,
+                                        119.3076923076923
+                                    ],
+                                    "spans": [],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        213,
+                                        119.3076923076923,
+                                        398,
+                                        130.6153846153846
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                },
+                                {
+                                    "bbox": [
+                                        213,
+                                        130.6153846153846,
+                                        398,
+                                        141.92307692307693
+                                    ],
+                                    "spans": [],
+                                    "index": 3
+                                },
+                                {
+                                    "bbox": [
+                                        213,
+                                        141.92307692307693,
+                                        398,
+                                        153.23076923076925
+                                    ],
+                                    "spans": [],
+                                    "index": 4
+                                },
+                                {
+                                    "bbox": [
+                                        213,
+                                        153.23076923076925,
+                                        398,
+                                        164.53846153846158
+                                    ],
+                                    "spans": [],
+                                    "index": 5
+                                },
+                                {
+                                    "bbox": [
+                                        213,
+                                        164.53846153846158,
+                                        398,
+                                        175.8461538461539
+                                    ],
+                                    "spans": [],
+                                    "index": 6
+                                },
+                                {
+                                    "bbox": [
+                                        213,
+                                        175.8461538461539,
+                                        398,
+                                        187.15384615384622
+                                    ],
+                                    "spans": [],
+                                    "index": 7
+                                },
+                                {
+                                    "bbox": [
+                                        213,
+                                        187.15384615384622,
+                                        398,
+                                        198.46153846153854
+                                    ],
+                                    "spans": [],
+                                    "index": 8
+                                },
+                                {
+                                    "bbox": [
+                                        213,
+                                        198.46153846153854,
+                                        398,
+                                        209.76923076923086
+                                    ],
+                                    "spans": [],
+                                    "index": 9
+                                },
+                                {
+                                    "bbox": [
+                                        213,
+                                        209.76923076923086,
+                                        398,
+                                        221.07692307692318
+                                    ],
+                                    "spans": [],
+                                    "index": 10
+                                },
+                                {
+                                    "bbox": [
+                                        213,
+                                        221.07692307692318,
+                                        398,
+                                        232.3846153846155
+                                    ],
+                                    "spans": [],
+                                    "index": 11
+                                },
+                                {
+                                    "bbox": [
+                                        213,
+                                        232.3846153846155,
+                                        398,
+                                        243.69230769230782
+                                    ],
+                                    "spans": [],
+                                    "index": 12
+                                },
+                                {
+                                    "bbox": [
+                                        213,
+                                        243.69230769230782,
+                                        398,
+                                        255.00000000000014
+                                    ],
+                                    "spans": [],
+                                    "index": 13
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 7
+                },
+                {
+                    "type": "image",
+                    "bbox": [
+                        213,
+                        296,
+                        389,
+                        438
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                213,
+                                296,
+                                389,
+                                438
+                            ],
+                            "group_id": 2,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        213,
+                                        296,
+                                        389,
+                                        438
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                213,
+                                                296,
+                                                389,
+                                                438
+                                            ],
+                                            "score": 0.339,
+                                            "type": "image",
+                                            "image_path": "64fe9f96b9155e2bd1b22d7a344f3085ff647a473ca40df4a123eb8758cf0801.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 19.5,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        213,
+                                        296,
+                                        389,
+                                        307.8333333333333
+                                    ],
+                                    "spans": [],
+                                    "index": 14
+                                },
+                                {
+                                    "bbox": [
+                                        213,
+                                        307.8333333333333,
+                                        389,
+                                        319.66666666666663
+                                    ],
+                                    "spans": [],
+                                    "index": 15
+                                },
+                                {
+                                    "bbox": [
+                                        213,
+                                        319.66666666666663,
+                                        389,
+                                        331.49999999999994
+                                    ],
+                                    "spans": [],
+                                    "index": 16
+                                },
+                                {
+                                    "bbox": [
+                                        213,
+                                        331.49999999999994,
+                                        389,
+                                        343.33333333333326
+                                    ],
+                                    "spans": [],
+                                    "index": 17
+                                },
+                                {
+                                    "bbox": [
+                                        213,
+                                        343.33333333333326,
+                                        389,
+                                        355.1666666666666
+                                    ],
+                                    "spans": [],
+                                    "index": 18
+                                },
+                                {
+                                    "bbox": [
+                                        213,
+                                        355.1666666666666,
+                                        389,
+                                        366.9999999999999
+                                    ],
+                                    "spans": [],
+                                    "index": 19
+                                },
+                                {
+                                    "bbox": [
+                                        213,
+                                        366.9999999999999,
+                                        389,
+                                        378.8333333333332
+                                    ],
+                                    "spans": [],
+                                    "index": 20
+                                },
+                                {
+                                    "bbox": [
+                                        213,
+                                        378.8333333333332,
+                                        389,
+                                        390.6666666666665
+                                    ],
+                                    "spans": [],
+                                    "index": 21
+                                },
+                                {
+                                    "bbox": [
+                                        213,
+                                        390.6666666666665,
+                                        389,
+                                        402.49999999999983
+                                    ],
+                                    "spans": [],
+                                    "index": 22
+                                },
+                                {
+                                    "bbox": [
+                                        213,
+                                        402.49999999999983,
+                                        389,
+                                        414.33333333333314
+                                    ],
+                                    "spans": [],
+                                    "index": 23
+                                },
+                                {
+                                    "bbox": [
+                                        213,
+                                        414.33333333333314,
+                                        389,
+                                        426.16666666666646
+                                    ],
+                                    "spans": [],
+                                    "index": 24
+                                },
+                                {
+                                    "bbox": [
+                                        213,
+                                        426.16666666666646,
+                                        389,
+                                        437.9999999999998
+                                    ],
+                                    "spans": [],
+                                    "index": 25
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 19.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        260,
+                        457,
+                        351,
+                        465
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                259,
+                                456,
+                                351,
+                                466
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        259,
+                                        456,
+                                        351,
+                                        466
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Angular\tinput\t(0\tto\t360\tdegrees)",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        }
+                    ],
+                    "index": 26,
+                    "bbox_fs": [
+                        259,
+                        456,
+                        351,
+                        466
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        286,
+                        505,
+                        324,
+                        513
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                285,
+                                504,
+                                325,
+                                515
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        285,
+                                        504,
+                                        325,
+                                        515
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Speed\ttuning",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        }
+                    ],
+                    "index": 27,
+                    "bbox_fs": [
+                        285,
+                        504,
+                        325,
+                        515
+                    ]
+                },
+                {
+                    "type": "image",
+                    "bbox": [
+                        207,
+                        524,
+                        388,
+                        658
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                207,
+                                524,
+                                388,
+                                658
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        207,
+                                        524,
+                                        388,
+                                        658
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                207,
+                                                524,
+                                                388,
+                                                658
+                                            ],
+                                            "score": 0.43,
+                                            "type": "image",
+                                            "image_path": "7eb735c5ab485558870c8b0720dd627699e69018e8cbacfa58c8d343cfc8a2d1.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 33.5,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        207,
+                                        524,
+                                        388,
+                                        535.1666666666666
+                                    ],
+                                    "spans": [],
+                                    "index": 28
+                                },
+                                {
+                                    "bbox": [
+                                        207,
+                                        535.1666666666666,
+                                        388,
+                                        546.3333333333333
+                                    ],
+                                    "spans": [],
+                                    "index": 29
+                                },
+                                {
+                                    "bbox": [
+                                        207,
+                                        546.3333333333333,
+                                        388,
+                                        557.4999999999999
+                                    ],
+                                    "spans": [],
+                                    "index": 30
+                                },
+                                {
+                                    "bbox": [
+                                        207,
+                                        557.4999999999999,
+                                        388,
+                                        568.6666666666665
+                                    ],
+                                    "spans": [],
+                                    "index": 31
+                                },
+                                {
+                                    "bbox": [
+                                        207,
+                                        568.6666666666665,
+                                        388,
+                                        579.8333333333331
+                                    ],
+                                    "spans": [],
+                                    "index": 32
+                                },
+                                {
+                                    "bbox": [
+                                        207,
+                                        579.8333333333331,
+                                        388,
+                                        590.9999999999998
+                                    ],
+                                    "spans": [],
+                                    "index": 33
+                                },
+                                {
+                                    "bbox": [
+                                        207,
+                                        590.9999999999998,
+                                        388,
+                                        602.1666666666664
+                                    ],
+                                    "spans": [],
+                                    "index": 34
+                                },
+                                {
+                                    "bbox": [
+                                        207,
+                                        602.1666666666664,
+                                        388,
+                                        613.333333333333
+                                    ],
+                                    "spans": [],
+                                    "index": 35
+                                },
+                                {
+                                    "bbox": [
+                                        207,
+                                        613.333333333333,
+                                        388,
+                                        624.4999999999997
+                                    ],
+                                    "spans": [],
+                                    "index": 36
+                                },
+                                {
+                                    "bbox": [
+                                        207,
+                                        624.4999999999997,
+                                        388,
+                                        635.6666666666663
+                                    ],
+                                    "spans": [],
+                                    "index": 37
+                                },
+                                {
+                                    "bbox": [
+                                        207,
+                                        635.6666666666663,
+                                        388,
+                                        646.8333333333329
+                                    ],
+                                    "spans": [],
+                                    "index": 38
+                                },
+                                {
+                                    "bbox": [
+                                        207,
+                                        646.8333333333329,
+                                        388,
+                                        657.9999999999995
+                                    ],
+                                    "spans": [],
+                                    "index": 39
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                288,
+                                677,
+                                322,
+                                685
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        287,
+                                        676,
+                                        324,
+                                        687
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                287,
+                                                676,
+                                                324,
+                                                687
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Speed\tinput",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 40
+                                }
+                            ],
+                            "index": 40
+                        }
+                    ],
+                    "index": 36.75
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        106,
+                        505,
+                        195
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                106,
+                                505,
+                                119
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        106,
+                                        505,
+                                        119
+                                    ],
+                                    "score": 1.0,
+                                    "content": "To quantify the speed selectivity of each unit we first fit a line to the tuning curve of unit activity as",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                117,
+                                506,
+                                129
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        117,
+                                        506,
+                                        129
+                                    ],
+                                    "score": 1.0,
+                                    "content": "a function of speed. The speed selectivity is the absolute value of the slope. If the unit activity is not",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                128,
+                                505,
+                                141
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        128,
+                                        505,
+                                        141
+                                    ],
+                                    "score": 1.0,
+                                    "content": "modulated by speed then the speed selectivity is 0. To quantify the direction selectivity of each unit",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                139,
+                                505,
+                                151
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        139,
+                                        505,
+                                        151
+                                    ],
+                                    "score": 1.0,
+                                    "content": "we calculated the average unit activity as a function of direction input and then took the maximum",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                151,
+                                505,
+                                163
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        151,
+                                        505,
+                                        163
+                                    ],
+                                    "score": 1.0,
+                                    "content": "minus minimum of this tuning curve. If the unit activity is not modulated by direction then the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                160,
+                                505,
+                                173
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        160,
+                                        505,
+                                        173
+                                    ],
+                                    "score": 1.0,
+                                    "content": "direction selectivity is 0. To quantify the spatial selectivity we used lifetime sparseness (Willmore",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                172,
+                                504,
+                                185
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        172,
+                                        504,
+                                        185
+                                    ],
+                                    "score": 1.0,
+                                    "content": "& Tolhurst, 2001). If the unit activity is not modulated by spatial location then the spatial selectivity",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                183,
+                                394,
+                                195
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        183,
+                                        394,
+                                        195
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is 0. Each dot in the figures below show the selectivity for a single unit.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        }
+                    ],
+                    "index": 3.5
+                },
+                {
+                    "type": "image",
+                    "bbox": [
+                        107,
+                        196,
+                        507,
+                        688
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                107,
+                                196,
+                                507,
+                                688
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        196,
+                                        507,
+                                        688
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                107,
+                                                196,
+                                                507,
+                                                688
+                                            ],
+                                            "score": 0.932,
+                                            "type": "image",
+                                            "image_path": "7d81c0e39724fab8f48b0863c121f5336b8e91d07c652f62dff58ab0d2f8945e.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 9,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        196,
+                                        507,
+                                        360.0
+                                    ],
+                                    "spans": [],
+                                    "index": 8
+                                },
+                                {
+                                    "bbox": [
+                                        107,
+                                        360.0,
+                                        507,
+                                        524.0
+                                    ],
+                                    "spans": [],
+                                    "index": 9
+                                },
+                                {
+                                    "bbox": [
+                                        107,
+                                        524.0,
+                                        507,
+                                        688.0
+                                    ],
+                                    "spans": [],
+                                    "index": 10
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 9
+                }
+            ],
+            "page_idx": 17,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        108,
+                        27,
+                        293,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                26,
+                                294,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        26,
+                                        294,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Published as a conference paper at ICLR 2018",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        300,
+                        751,
+                        311,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                299,
+                                750,
+                                313,
+                                763
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        299,
+                                        750,
+                                        313,
+                                        763
+                                    ],
+                                    "score": 1.0,
+                                    "content": "18",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        105,
+                        82,
+                        478,
+                        94
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                81,
+                                479,
+                                96
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        81,
+                                        479,
+                                        96
+                                    ],
+                                    "score": 1.0,
+                                    "content": "D RELATION BETWEEN SPEED, DIRECTION, AND SPATIAL SELECTIVITY",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        106,
+                        505,
+                        195
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                106,
+                                505,
+                                119
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        106,
+                                        505,
+                                        119
+                                    ],
+                                    "score": 1.0,
+                                    "content": "To quantify the speed selectivity of each unit we first fit a line to the tuning curve of unit activity as",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                117,
+                                506,
+                                129
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        117,
+                                        506,
+                                        129
+                                    ],
+                                    "score": 1.0,
+                                    "content": "a function of speed. The speed selectivity is the absolute value of the slope. If the unit activity is not",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                128,
+                                505,
+                                141
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        128,
+                                        505,
+                                        141
+                                    ],
+                                    "score": 1.0,
+                                    "content": "modulated by speed then the speed selectivity is 0. To quantify the direction selectivity of each unit",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                139,
+                                505,
+                                151
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        139,
+                                        505,
+                                        151
+                                    ],
+                                    "score": 1.0,
+                                    "content": "we calculated the average unit activity as a function of direction input and then took the maximum",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                151,
+                                505,
+                                163
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        151,
+                                        505,
+                                        163
+                                    ],
+                                    "score": 1.0,
+                                    "content": "minus minimum of this tuning curve. If the unit activity is not modulated by direction then the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                160,
+                                505,
+                                173
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        160,
+                                        505,
+                                        173
+                                    ],
+                                    "score": 1.0,
+                                    "content": "direction selectivity is 0. To quantify the spatial selectivity we used lifetime sparseness (Willmore",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                172,
+                                504,
+                                185
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        172,
+                                        504,
+                                        185
+                                    ],
+                                    "score": 1.0,
+                                    "content": "& Tolhurst, 2001). If the unit activity is not modulated by spatial location then the spatial selectivity",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                183,
+                                394,
+                                195
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        183,
+                                        394,
+                                        195
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is 0. Each dot in the figures below show the selectivity for a single unit.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        }
+                    ],
+                    "index": 3.5,
+                    "bbox_fs": [
+                        105,
+                        106,
+                        506,
+                        195
+                    ]
+                },
+                {
+                    "type": "image",
+                    "bbox": [
+                        107,
+                        196,
+                        507,
+                        688
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                107,
+                                196,
+                                507,
+                                688
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        196,
+                                        507,
+                                        688
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                107,
+                                                196,
+                                                507,
+                                                688
+                                            ],
+                                            "score": 0.932,
+                                            "type": "image",
+                                            "image_path": "7d81c0e39724fab8f48b0863c121f5336b8e91d07c652f62dff58ab0d2f8945e.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 9,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        196,
+                                        507,
+                                        360.0
+                                    ],
+                                    "spans": [],
+                                    "index": 8
+                                },
+                                {
+                                    "bbox": [
+                                        107,
+                                        360.0,
+                                        507,
+                                        524.0
+                                    ],
+                                    "spans": [],
+                                    "index": 9
+                                },
+                                {
+                                    "bbox": [
+                                        107,
+                                        524.0,
+                                        507,
+                                        688.0
+                                    ],
+                                    "spans": [],
+                                    "index": 10
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 9
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        81,
+                        294,
+                        93
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                81,
+                                296,
+                                96
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        81,
+                                        296,
+                                        96
+                                    ],
+                                    "score": 1.0,
+                                    "content": "E ADDITIONAL TRAINING DETAILS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        }
+                    ],
+                    "index": 0
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        106,
+                        505,
+                        183
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                106,
+                                505,
+                                119
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        106,
+                                        399,
+                                        119
+                                    ],
+                                    "score": 1.0,
+                                    "content": "During training we tried to balance all three terms we were minimizing",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        399,
+                                        106,
+                                        433,
+                                        118
+                                    ],
+                                    "score": 0.76,
+                                    "content": "( E , R _ { L 2 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        433,
+                                        106,
+                                        454,
+                                        119
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        455,
+                                        107,
+                                        476,
+                                        118
+                                    ],
+                                    "score": 0.89,
+                                    "content": "R _ { F R }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        477,
+                                        106,
+                                        505,
+                                        119
+                                    ],
+                                    "score": 1.0,
+                                    "content": ") so no",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                117,
+                                505,
+                                130
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        117,
+                                        505,
+                                        130
+                                    ],
+                                    "score": 1.0,
+                                    "content": "single term was neglected or dominated. At the beginning of training we weighted the regularization",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                127,
+                                505,
+                                142
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        127,
+                                        127,
+                                        142
+                                    ],
+                                    "score": 1.0,
+                                    "content": "term",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        127,
+                                        128,
+                                        146,
+                                        139
+                                    ],
+                                    "score": 0.9,
+                                    "content": "R _ { L 2 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        146,
+                                        127,
+                                        275,
+                                        142
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to be equal to the error function",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        276,
+                                        129,
+                                        285,
+                                        138
+                                    ],
+                                    "score": 0.81,
+                                    "content": "E",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        285,
+                                        127,
+                                        433,
+                                        142
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and then decreased the weighting on",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        433,
+                                        128,
+                                        452,
+                                        139
+                                    ],
+                                    "score": 0.89,
+                                    "content": "R _ { L 2 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        452,
+                                        127,
+                                        505,
+                                        142
+                                    ],
+                                    "score": 1.0,
+                                    "content": "according to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                137,
+                                505,
+                                153
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        137,
+                                        479,
+                                        153
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the schedule used by Martens & Sutskever (2011). We adaptively adjusted the weighting on",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        480,
+                                        140,
+                                        501,
+                                        150
+                                    ],
+                                    "score": 0.9,
+                                    "content": "R _ { F R }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        501,
+                                        137,
+                                        505,
+                                        153
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                149,
+                                504,
+                                163
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        149,
+                                        233,
+                                        163
+                                    ],
+                                    "score": 1.0,
+                                    "content": "starting from an initial value of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        234,
+                                        150,
+                                        258,
+                                        162
+                                    ],
+                                    "score": 0.91,
+                                    "content": "E / 1 0",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        258,
+                                        149,
+                                        392,
+                                        163
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and enforcing an upper bound of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        392,
+                                        150,
+                                        411,
+                                        162
+                                    ],
+                                    "score": 0.91,
+                                    "content": "E / 3",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        412,
+                                        149,
+                                        504,
+                                        163
+                                    ],
+                                    "score": 1.0,
+                                    "content": "as training progressed.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                160,
+                                505,
+                                174
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        160,
+                                        505,
+                                        174
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We found this training procedure improved training performance and led to more interesting repre-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                172,
+                                152,
+                                184
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        172,
+                                        152,
+                                        184
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sentations.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        }
+                    ],
+                    "index": 4
+                }
+            ],
+            "page_idx": 18,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        107,
+                        27,
+                        293,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                26,
+                                294,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        26,
+                                        294,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Published as a conference paper at ICLR 2018",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        301,
+                        752,
+                        311,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                299,
+                                750,
+                                312,
+                                765
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        299,
+                                        750,
+                                        312,
+                                        765
+                                    ],
+                                    "score": 1.0,
+                                    "content": "",
+                                    "type": "text",
+                                    "height": 15,
+                                    "width": 13
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        81,
+                        294,
+                        93
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                81,
+                                296,
+                                96
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        81,
+                                        296,
+                                        96
+                                    ],
+                                    "score": 1.0,
+                                    "content": "E ADDITIONAL TRAINING DETAILS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        }
+                    ],
+                    "index": 0
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        106,
+                        505,
+                        183
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                106,
+                                505,
+                                119
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        106,
+                                        399,
+                                        119
+                                    ],
+                                    "score": 1.0,
+                                    "content": "During training we tried to balance all three terms we were minimizing",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        399,
+                                        106,
+                                        433,
+                                        118
+                                    ],
+                                    "score": 0.76,
+                                    "content": "( E , R _ { L 2 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        433,
+                                        106,
+                                        454,
+                                        119
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        455,
+                                        107,
+                                        476,
+                                        118
+                                    ],
+                                    "score": 0.89,
+                                    "content": "R _ { F R }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        477,
+                                        106,
+                                        505,
+                                        119
+                                    ],
+                                    "score": 1.0,
+                                    "content": ") so no",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                117,
+                                505,
+                                130
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        117,
+                                        505,
+                                        130
+                                    ],
+                                    "score": 1.0,
+                                    "content": "single term was neglected or dominated. At the beginning of training we weighted the regularization",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                127,
+                                505,
+                                142
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        127,
+                                        127,
+                                        142
+                                    ],
+                                    "score": 1.0,
+                                    "content": "term",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        127,
+                                        128,
+                                        146,
+                                        139
+                                    ],
+                                    "score": 0.9,
+                                    "content": "R _ { L 2 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        146,
+                                        127,
+                                        275,
+                                        142
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to be equal to the error function",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        276,
+                                        129,
+                                        285,
+                                        138
+                                    ],
+                                    "score": 0.81,
+                                    "content": "E",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        285,
+                                        127,
+                                        433,
+                                        142
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and then decreased the weighting on",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        433,
+                                        128,
+                                        452,
+                                        139
+                                    ],
+                                    "score": 0.89,
+                                    "content": "R _ { L 2 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        452,
+                                        127,
+                                        505,
+                                        142
+                                    ],
+                                    "score": 1.0,
+                                    "content": "according to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                137,
+                                505,
+                                153
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        137,
+                                        479,
+                                        153
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the schedule used by Martens & Sutskever (2011). We adaptively adjusted the weighting on",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        480,
+                                        140,
+                                        501,
+                                        150
+                                    ],
+                                    "score": 0.9,
+                                    "content": "R _ { F R }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        501,
+                                        137,
+                                        505,
+                                        153
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                149,
+                                504,
+                                163
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        149,
+                                        233,
+                                        163
+                                    ],
+                                    "score": 1.0,
+                                    "content": "starting from an initial value of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        234,
+                                        150,
+                                        258,
+                                        162
+                                    ],
+                                    "score": 0.91,
+                                    "content": "E / 1 0",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        258,
+                                        149,
+                                        392,
+                                        163
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and enforcing an upper bound of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        392,
+                                        150,
+                                        411,
+                                        162
+                                    ],
+                                    "score": 0.91,
+                                    "content": "E / 3",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        412,
+                                        149,
+                                        504,
+                                        163
+                                    ],
+                                    "score": 1.0,
+                                    "content": "as training progressed.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                160,
+                                505,
+                                174
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        160,
+                                        505,
+                                        174
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We found this training procedure improved training performance and led to more interesting repre-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                172,
+                                152,
+                                184
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        172,
+                                        152,
+                                        184
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sentations.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        }
+                    ],
+                    "index": 4,
+                    "bbox_fs": [
+                        105,
+                        106,
+                        505,
+                        184
+                    ]
+                }
+            ]
+        }
+    ],
+    "_backend": "pipeline",
+    "_version_name": "2.2.2"
+}
\ No newline at end of file
diff --git a/parse/train/B17JTOe0-/B17JTOe0-_model.json b/parse/train/B17JTOe0-/B17JTOe0-_model.json
new file mode 100644
index 0000000000000000000000000000000000000000..06d997889d0316453f41d331b1e05b39ef17ab32
--- /dev/null
+++ b/parse/train/B17JTOe0-/B17JTOe0-_model.json
@@ -0,0 +1,20110 @@
+[
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    398,
+                    710,
+                    1303,
+                    710,
+                    1303,
+                    1166,
+                    398,
+                    1166
+                ],
+                "score": 0.982
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1575,
+                    1405,
+                    1575,
+                    1405,
+                    1969,
+                    298,
+                    1969
+                ],
+                "score": 0.98
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1313,
+                    1403,
+                    1313,
+                    1403,
+                    1559,
+                    298,
+                    1559
+                ],
+                "score": 0.979
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    298,
+                    222,
+                    1398,
+                    222,
+                    1398,
+                    379,
+                    298,
+                    379
+                ],
+                "score": 0.965
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    302,
+                    1239,
+                    573,
+                    1239,
+                    573,
+                    1275,
+                    302,
+                    1275
+                ],
+                "score": 0.892
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    333,
+                    2006,
+                    527,
+                    2006,
+                    527,
+                    2033,
+                    333,
+                    2033
+                ],
+                "score": 0.884
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    299,
+                    75,
+                    816,
+                    75,
+                    816,
+                    104,
+                    299,
+                    104
+                ],
+                "score": 0.878
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    773,
+                    635,
+                    927,
+                    635,
+                    927,
+                    668,
+                    773,
+                    668
+                ],
+                "score": 0.861
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    312,
+                    429,
+                    761,
+                    429,
+                    761,
+                    555,
+                    312,
+                    555
+                ],
+                "score": 0.736
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    841,
+                    2088,
+                    857,
+                    2088,
+                    857,
+                    2112,
+                    841,
+                    2112
+                ],
+                "score": 0.718
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    220.0,
+                    1406.0,
+                    220.0,
+                    1406.0,
+                    278.0,
+                    292.0,
+                    278.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    284.0,
+                    1259.0,
+                    284.0,
+                    1259.0,
+                    330.0,
+                    295.0,
+                    330.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    339.0,
+                    995.0,
+                    339.0,
+                    995.0,
+                    385.0,
+                    294.0,
+                    385.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1236.0,
+                    579.0,
+                    1236.0,
+                    579.0,
+                    1283.0,
+                    294.0,
+                    1283.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    329.0,
+                    2001.0,
+                    532.0,
+                    2001.0,
+                    532.0,
+                    2036.0,
+                    329.0,
+                    2036.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    72.0,
+                    817.0,
+                    72.0,
+                    817.0,
+                    108.0,
+                    295.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    768.0,
+                    631.0,
+                    934.0,
+                    631.0,
+                    934.0,
+                    674.0,
+                    768.0,
+                    674.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    840.0,
+                    2087.0,
+                    860.0,
+                    2087.0,
+                    860.0,
+                    2117.0,
+                    840.0,
+                    2117.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    394.0,
+                    708.0,
+                    1304.0,
+                    708.0,
+                    1304.0,
+                    745.0,
+                    394.0,
+                    745.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    394.0,
+                    738.0,
+                    1306.0,
+                    738.0,
+                    1306.0,
+                    775.0,
+                    394.0,
+                    775.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    395.0,
+                    770.0,
+                    1305.0,
+                    770.0,
+                    1305.0,
+                    805.0,
+                    395.0,
+                    805.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    395.0,
+                    800.0,
+                    1305.0,
+                    800.0,
+                    1305.0,
+                    837.0,
+                    395.0,
+                    837.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    394.0,
+                    832.0,
+                    1305.0,
+                    832.0,
+                    1305.0,
+                    866.0,
+                    394.0,
+                    866.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    393.0,
+                    863.0,
+                    1307.0,
+                    863.0,
+                    1307.0,
+                    898.0,
+                    393.0,
+                    898.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    393.0,
+                    893.0,
+                    1305.0,
+                    893.0,
+                    1305.0,
+                    928.0,
+                    393.0,
+                    928.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    394.0,
+                    925.0,
+                    1304.0,
+                    925.0,
+                    1304.0,
+                    956.0,
+                    394.0,
+                    956.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    393.0,
+                    953.0,
+                    1304.0,
+                    953.0,
+                    1304.0,
+                    988.0,
+                    393.0,
+                    988.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    393.0,
+                    984.0,
+                    1304.0,
+                    984.0,
+                    1304.0,
+                    1018.0,
+                    393.0,
+                    1018.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    392.0,
+                    1014.0,
+                    1306.0,
+                    1014.0,
+                    1306.0,
+                    1051.0,
+                    392.0,
+                    1051.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    395.0,
+                    1045.0,
+                    1303.0,
+                    1045.0,
+                    1303.0,
+                    1077.0,
+                    395.0,
+                    1077.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    394.0,
+                    1076.0,
+                    1304.0,
+                    1076.0,
+                    1304.0,
+                    1109.0,
+                    394.0,
+                    1109.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    394.0,
+                    1106.0,
+                    1307.0,
+                    1106.0,
+                    1307.0,
+                    1140.0,
+                    394.0,
+                    1140.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    393.0,
+                    1138.0,
+                    883.0,
+                    1138.0,
+                    883.0,
+                    1169.0,
+                    393.0,
+                    1169.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1574.0,
+                    1403.0,
+                    1574.0,
+                    1403.0,
+                    1609.0,
+                    295.0,
+                    1609.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1603.0,
+                    1403.0,
+                    1603.0,
+                    1403.0,
+                    1638.0,
+                    295.0,
+                    1638.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1637.0,
+                    1406.0,
+                    1637.0,
+                    1406.0,
+                    1669.0,
+                    295.0,
+                    1669.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1665.0,
+                    1405.0,
+                    1665.0,
+                    1405.0,
+                    1700.0,
+                    293.0,
+                    1700.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1693.0,
+                    1405.0,
+                    1693.0,
+                    1405.0,
+                    1732.0,
+                    293.0,
+                    1732.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1726.0,
+                    1405.0,
+                    1726.0,
+                    1405.0,
+                    1763.0,
+                    293.0,
+                    1763.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1758.0,
+                    1403.0,
+                    1758.0,
+                    1403.0,
+                    1790.0,
+                    295.0,
+                    1790.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1785.0,
+                    1405.0,
+                    1785.0,
+                    1405.0,
+                    1823.0,
+                    293.0,
+                    1823.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1817.0,
+                    1406.0,
+                    1817.0,
+                    1406.0,
+                    1853.0,
+                    295.0,
+                    1853.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1847.0,
+                    1403.0,
+                    1847.0,
+                    1403.0,
+                    1883.0,
+                    295.0,
+                    1883.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1879.0,
+                    1406.0,
+                    1879.0,
+                    1406.0,
+                    1915.0,
+                    295.0,
+                    1915.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1908.0,
+                    1406.0,
+                    1908.0,
+                    1406.0,
+                    1948.0,
+                    295.0,
+                    1948.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1939.0,
+                    1006.0,
+                    1939.0,
+                    1006.0,
+                    1975.0,
+                    295.0,
+                    1975.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1315.0,
+                    1403.0,
+                    1315.0,
+                    1403.0,
+                    1348.0,
+                    296.0,
+                    1348.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1345.0,
+                    1405.0,
+                    1345.0,
+                    1405.0,
+                    1378.0,
+                    294.0,
+                    1378.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1375.0,
+                    1403.0,
+                    1375.0,
+                    1403.0,
+                    1408.0,
+                    296.0,
+                    1408.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1404.0,
+                    1405.0,
+                    1404.0,
+                    1405.0,
+                    1441.0,
+                    292.0,
+                    1441.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1434.0,
+                    1403.0,
+                    1434.0,
+                    1403.0,
+                    1470.0,
+                    293.0,
+                    1470.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1464.0,
+                    1403.0,
+                    1464.0,
+                    1403.0,
+                    1501.0,
+                    293.0,
+                    1501.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1496.0,
+                    1405.0,
+                    1496.0,
+                    1405.0,
+                    1533.0,
+                    294.0,
+                    1533.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1525.0,
+                    886.0,
+                    1525.0,
+                    886.0,
+                    1562.0,
+                    295.0,
+                    1562.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    314.0,
+                    431.0,
+                    749.0,
+                    431.0,
+                    749.0,
+                    464.0,
+                    314.0,
+                    464.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    311.0,
+                    457.0,
+                    554.0,
+                    457.0,
+                    554.0,
+                    497.0,
+                    311.0,
+                    497.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    311.0,
+                    491.0,
+                    631.0,
+                    491.0,
+                    631.0,
+                    523.0,
+                    311.0,
+                    523.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    312.0,
+                    522.0,
+                    765.0,
+                    522.0,
+                    765.0,
+                    561.0,
+                    312.0,
+                    561.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 0,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    229,
+                    1404,
+                    229,
+                    1404,
+                    627,
+                    298,
+                    627
+                ],
+                "score": 0.984
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    642,
+                    1404,
+                    642,
+                    1404,
+                    948,
+                    297,
+                    948
+                ],
+                "score": 0.983
+            },
+            {
+                "category_id": 3,
+                "poly": [
+                    408,
+                    1038,
+                    1280,
+                    1038,
+                    1280,
+                    1500,
+                    408,
+                    1500
+                ],
+                "score": 0.976
+            },
+            {
+                "category_id": 4,
+                "poly": [
+                    296,
+                    1529,
+                    1405,
+                    1529,
+                    1405,
+                    1959,
+                    296,
+                    1959
+                ],
+                "score": 0.972
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    300,
+                    76,
+                    815,
+                    76,
+                    815,
+                    104,
+                    300,
+                    104
+                ],
+                "score": 0.892
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    841,
+                    2088,
+                    858,
+                    2088,
+                    858,
+                    2112,
+                    841,
+                    2112
+                ],
+                "score": 0.72
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1096,
+                    1774,
+                    1213,
+                    1774,
+                    1213,
+                    1802,
+                    1096,
+                    1802
+                ],
+                "score": 0.9,
+                "latex": "N = 1 0 0"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    415.0,
+                    1047.0,
+                    431.0,
+                    1047.0,
+                    431.0,
+                    1067.0,
+                    415.0,
+                    1067.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1016.0,
+                    1047.0,
+                    1046.0,
+                    1047.0,
+                    1046.0,
+                    1138.0,
+                    1016.0,
+                    1138.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    468.0,
+                    1152.0,
+                    543.0,
+                    1152.0,
+                    543.0,
+                    1179.0,
+                    468.0,
+                    1179.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    593.0,
+                    1152.0,
+                    676.0,
+                    1152.0,
+                    676.0,
+                    1180.0,
+                    593.0,
+                    1180.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    740.0,
+                    1152.0,
+                    834.0,
+                    1152.0,
+                    834.0,
+                    1179.0,
+                    740.0,
+                    1179.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    875.0,
+                    1150.0,
+                    985.0,
+                    1150.0,
+                    985.0,
+                    1180.0,
+                    875.0,
+                    1180.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1059.0,
+                    1152.0,
+                    1122.0,
+                    1152.0,
+                    1122.0,
+                    1181.0,
+                    1059.0,
+                    1181.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1175.0,
+                    1150.0,
+                    1255.0,
+                    1150.0,
+                    1255.0,
+                    1180.0,
+                    1175.0,
+                    1180.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    414.0,
+                    1194.0,
+                    434.0,
+                    1194.0,
+                    434.0,
+                    1216.0,
+                    414.0,
+                    1216.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    635.0,
+                    1189.0,
+                    683.0,
+                    1189.0,
+                    683.0,
+                    1218.0,
+                    635.0,
+                    1218.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    961.0,
+                    1207.0,
+                    979.0,
+                    1207.0,
+                    979.0,
+                    1227.0,
+                    961.0,
+                    1227.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1071.0,
+                    1202.0,
+                    1180.0,
+                    1202.0,
+                    1180.0,
+                    1231.0,
+                    1071.0,
+                    1231.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    415.0,
+                    1236.0,
+                    471.0,
+                    1236.0,
+                    471.0,
+                    1266.0,
+                    415.0,
+                    1266.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    845.0,
+                    1236.0,
+                    914.0,
+                    1236.0,
+                    914.0,
+                    1267.0,
+                    845.0,
+                    1267.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1148.0,
+                    1276.0,
+                    1165.0,
+                    1276.0,
+                    1165.0,
+                    1295.0,
+                    1148.0,
+                    1295.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    417.0,
+                    1345.0,
+                    469.0,
+                    1345.0,
+                    469.0,
+                    1365.0,
+                    417.0,
+                    1365.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    845.0,
+                    1344.0,
+                    924.0,
+                    1344.0,
+                    924.0,
+                    1368.0,
+                    845.0,
+                    1368.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1113.0,
+                    1396.0,
+                    1226.0,
+                    1396.0,
+                    1226.0,
+                    1428.0,
+                    1113.0,
+                    1428.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1115.0,
+                    1425.0,
+                    1217.0,
+                    1425.0,
+                    1217.0,
+                    1453.0,
+                    1115.0,
+                    1453.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    415.0,
+                    1460.0,
+                    488.0,
+                    1460.0,
+                    488.0,
+                    1487.0,
+                    415.0,
+                    1487.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    843.0,
+                    1461.0,
+                    923.0,
+                    1461.0,
+                    923.0,
+                    1485.0,
+                    843.0,
+                    1485.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1077.75,
+                    1261.5,
+                    1158.75,
+                    1261.5,
+                    1158.75,
+                    1287.5,
+                    1077.75,
+                    1287.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1530.0,
+                    1405.0,
+                    1530.0,
+                    1405.0,
+                    1566.0,
+                    294.0,
+                    1566.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1559.0,
+                    1406.0,
+                    1559.0,
+                    1406.0,
+                    1596.0,
+                    293.0,
+                    1596.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1591.0,
+                    1405.0,
+                    1591.0,
+                    1405.0,
+                    1627.0,
+                    294.0,
+                    1627.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1622.0,
+                    1405.0,
+                    1622.0,
+                    1405.0,
+                    1657.0,
+                    294.0,
+                    1657.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1652.0,
+                    1405.0,
+                    1652.0,
+                    1405.0,
+                    1688.0,
+                    294.0,
+                    1688.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1680.0,
+                    1406.0,
+                    1680.0,
+                    1406.0,
+                    1719.0,
+                    293.0,
+                    1719.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1711.0,
+                    1407.0,
+                    1711.0,
+                    1407.0,
+                    1749.0,
+                    293.0,
+                    1749.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1741.0,
+                    1405.0,
+                    1741.0,
+                    1405.0,
+                    1780.0,
+                    293.0,
+                    1780.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1773.0,
+                    1095.0,
+                    1773.0,
+                    1095.0,
+                    1809.0,
+                    294.0,
+                    1809.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1214.0,
+                    1773.0,
+                    1405.0,
+                    1773.0,
+                    1405.0,
+                    1809.0,
+                    1214.0,
+                    1809.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1805.0,
+                    1405.0,
+                    1805.0,
+                    1405.0,
+                    1838.0,
+                    291.0,
+                    1838.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    290.0,
+                    1831.0,
+                    1407.0,
+                    1831.0,
+                    1407.0,
+                    1873.0,
+                    290.0,
+                    1873.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1864.0,
+                    1403.0,
+                    1864.0,
+                    1403.0,
+                    1899.0,
+                    293.0,
+                    1899.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1896.0,
+                    1406.0,
+                    1896.0,
+                    1406.0,
+                    1931.0,
+                    294.0,
+                    1931.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1928.0,
+                    916.0,
+                    1928.0,
+                    916.0,
+                    1961.0,
+                    293.0,
+                    1961.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    73.0,
+                    817.0,
+                    73.0,
+                    817.0,
+                    108.0,
+                    296.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    839.0,
+                    2085.0,
+                    862.0,
+                    2085.0,
+                    862.0,
+                    2120.0,
+                    839.0,
+                    2120.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    228.0,
+                    1404.0,
+                    228.0,
+                    1404.0,
+                    266.0,
+                    293.0,
+                    266.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    261.0,
+                    1404.0,
+                    261.0,
+                    1404.0,
+                    296.0,
+                    294.0,
+                    296.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    291.0,
+                    1405.0,
+                    291.0,
+                    1405.0,
+                    327.0,
+                    293.0,
+                    327.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    319.0,
+                    1403.0,
+                    319.0,
+                    1403.0,
+                    358.0,
+                    292.0,
+                    358.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    353.0,
+                    1404.0,
+                    353.0,
+                    1404.0,
+                    389.0,
+                    293.0,
+                    389.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    384.0,
+                    1404.0,
+                    384.0,
+                    1404.0,
+                    416.0,
+                    296.0,
+                    416.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    414.0,
+                    1406.0,
+                    414.0,
+                    1406.0,
+                    450.0,
+                    294.0,
+                    450.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    444.0,
+                    1406.0,
+                    444.0,
+                    1406.0,
+                    480.0,
+                    293.0,
+                    480.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    472.0,
+                    1406.0,
+                    472.0,
+                    1406.0,
+                    514.0,
+                    293.0,
+                    514.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    502.0,
+                    1405.0,
+                    502.0,
+                    1405.0,
+                    541.0,
+                    293.0,
+                    541.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    534.0,
+                    1404.0,
+                    534.0,
+                    1404.0,
+                    571.0,
+                    294.0,
+                    571.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    564.0,
+                    1406.0,
+                    564.0,
+                    1406.0,
+                    602.0,
+                    292.0,
+                    602.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    595.0,
+                    1297.0,
+                    595.0,
+                    1297.0,
+                    632.0,
+                    293.0,
+                    632.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    640.0,
+                    1404.0,
+                    640.0,
+                    1404.0,
+                    677.0,
+                    294.0,
+                    677.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    673.0,
+                    1404.0,
+                    673.0,
+                    1404.0,
+                    707.0,
+                    294.0,
+                    707.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    704.0,
+                    1404.0,
+                    704.0,
+                    1404.0,
+                    739.0,
+                    294.0,
+                    739.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    734.0,
+                    1405.0,
+                    734.0,
+                    1405.0,
+                    770.0,
+                    295.0,
+                    770.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    762.0,
+                    1405.0,
+                    762.0,
+                    1405.0,
+                    803.0,
+                    292.0,
+                    803.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    796.0,
+                    1405.0,
+                    796.0,
+                    1405.0,
+                    829.0,
+                    293.0,
+                    829.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    825.0,
+                    1405.0,
+                    825.0,
+                    1405.0,
+                    861.0,
+                    295.0,
+                    861.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    853.0,
+                    1405.0,
+                    853.0,
+                    1405.0,
+                    890.0,
+                    294.0,
+                    890.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    883.0,
+                    1406.0,
+                    883.0,
+                    1406.0,
+                    923.0,
+                    292.0,
+                    923.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    916.0,
+                    897.0,
+                    916.0,
+                    897.0,
+                    952.0,
+                    294.0,
+                    952.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 1,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    296,
+                    529,
+                    1405,
+                    529,
+                    1405,
+                    747,
+                    296,
+                    747
+                ],
+                "score": 0.983
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    296,
+                    1044,
+                    1404,
+                    1044,
+                    1404,
+                    1261,
+                    296,
+                    1261
+                ],
+                "score": 0.982
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1274,
+                    1404,
+                    1274,
+                    1404,
+                    1491,
+                    297,
+                    1491
+                ],
+                "score": 0.981
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1504,
+                    1404,
+                    1504,
+                    1404,
+                    1660,
+                    297,
+                    1660
+                ],
+                "score": 0.979
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1749,
+                    1404,
+                    1749,
+                    1404,
+                    1874,
+                    297,
+                    1874
+                ],
+                "score": 0.976
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    759,
+                    1404,
+                    759,
+                    1404,
+                    854,
+                    298,
+                    854
+                ],
+                "score": 0.975
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    479,
+                    426,
+                    1221,
+                    426,
+                    1221,
+                    519,
+                    479,
+                    519
+                ],
+                "score": 0.958
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    558,
+                    1875,
+                    1142,
+                    1875,
+                    1142,
+                    1968,
+                    558,
+                    1968
+                ],
+                "score": 0.957
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    717,
+                    866,
+                    982,
+                    866,
+                    982,
+                    957,
+                    717,
+                    957
+                ],
+                "score": 0.951
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    296,
+                    351,
+                    1406,
+                    351,
+                    1406,
+                    415,
+                    296,
+                    415
+                ],
+                "score": 0.951
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    296,
+                    1972,
+                    1401,
+                    1972,
+                    1401,
+                    2035,
+                    296,
+                    2035
+                ],
+                "score": 0.949
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    299,
+                    989,
+                    661,
+                    989,
+                    661,
+                    1020,
+                    299,
+                    1020
+                ],
+                "score": 0.922
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    299,
+                    225,
+                    459,
+                    225,
+                    459,
+                    262,
+                    299,
+                    262
+                ],
+                "score": 0.914
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    299,
+                    1694,
+                    490,
+                    1694,
+                    490,
+                    1726,
+                    299,
+                    1726
+                ],
+                "score": 0.91
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    299,
+                    74,
+                    815,
+                    74,
+                    815,
+                    105,
+                    299,
+                    105
+                ],
+                "score": 0.907
+            },
+            {
+                "category_id": 9,
+                "poly": [
+                    1366,
+                    895,
+                    1400,
+                    895,
+                    1400,
+                    925,
+                    1366,
+                    925
+                ],
+                "score": 0.886
+            },
+            {
+                "category_id": 9,
+                "poly": [
+                    1366,
+                    1905,
+                    1400,
+                    1905,
+                    1400,
+                    1935,
+                    1366,
+                    1935
+                ],
+                "score": 0.884
+            },
+            {
+                "category_id": 9,
+                "poly": [
+                    1366,
+                    456,
+                    1400,
+                    456,
+                    1400,
+                    487,
+                    1366,
+                    487
+                ],
+                "score": 0.878
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    300,
+                    294,
+                    624,
+                    294,
+                    624,
+                    326,
+                    300,
+                    326
+                ],
+                "score": 0.874
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    841,
+                    2088,
+                    859,
+                    2088,
+                    859,
+                    2112,
+                    841,
+                    2112
+                ],
+                "score": 0.684
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    841,
+                    2088,
+                    859,
+                    2088,
+                    859,
+                    2112,
+                    841,
+                    2112
+                ],
+                "score": 0.389
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    478,
+                    422,
+                    1221,
+                    422,
+                    1221,
+                    521,
+                    478,
+                    521
+                ],
+                "score": 0.95,
+                "latex": "\\tau \\frac { d x _ { i } ( t ) } { d t } = - x _ { i } ( t ) + \\sum _ { j = 1 } ^ { N } W _ { i j } ^ { \\mathrm { r e c } } u _ { j } ( t ) + \\sum _ { k = 1 } ^ { N _ { \\mathrm { i n } } } W _ { i k } ^ { \\mathrm { i n } } I _ { k } ( t ) + b _ { i } + \\xi _ { i } ( t )"
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    716,
+                    861,
+                    983,
+                    861,
+                    983,
+                    958,
+                    716,
+                    958
+                ],
+                "score": 0.94,
+                "latex": "y _ { j } ( t ) = \\sum _ { i = 1 } ^ { N } W _ { j i } ^ { \\mathrm { o u t } } u _ { i } ( t )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    573,
+                    383,
+                    630,
+                    383,
+                    630,
+                    416,
+                    573,
+                    416
+                ],
+                "score": 0.93,
+                "latex": "u _ { i } ( t )"
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    557,
+                    1870,
+                    1140,
+                    1870,
+                    1140,
+                    1971,
+                    557,
+                    1971
+                ],
+                "score": 0.93,
+                "latex": "E = \\frac { 1 } { M T N _ { \\mathrm { o u t } } } \\sum _ { m , t , j = 1 } ^ { M , T , N _ { \\mathrm { o u t } } } ( y _ { j } ( t , m ) - y _ { j } ^ { \\mathrm { t a r g e t } } ( t , m ) ) ^ { 2 }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    510,
+                    822,
+                    570,
+                    822,
+                    570,
+                    855,
+                    510,
+                    855
+                ],
+                "score": 0.92,
+                "latex": "y _ { 2 } ( t )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    939,
+                    561,
+                    1174,
+                    561,
+                    1174,
+                    594,
+                    939,
+                    594
+                ],
+                "score": 0.92,
+                "latex": "u _ { i } ( t ) = \\operatorname { t a n h } ( x _ { i } ( t ) )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    814,
+                    530,
+                    873,
+                    530,
+                    873,
+                    564,
+                    814,
+                    564
+                ],
+                "score": 0.92,
+                "latex": "u _ { i } ( t )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    399,
+                    822,
+                    459,
+                    822,
+                    459,
+                    855,
+                    399,
+                    855
+                ],
+                "score": 0.92,
+                "latex": "y _ { 1 } ( t )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    578,
+                    653,
+                    634,
+                    653,
+                    634,
+                    686,
+                    578,
+                    686
+                ],
+                "score": 0.91,
+                "latex": "\\xi _ { i } ( t )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    339,
+                    532,
+                    497,
+                    532,
+                    497,
+                    562,
+                    339,
+                    562
+                ],
+                "score": 0.91,
+                "latex": "i = 1 , \\ldots , N"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1334,
+                    530,
+                    1393,
+                    530,
+                    1393,
+                    564,
+                    1334,
+                    564
+                ],
+                "score": 0.91,
+                "latex": "x _ { i } ( t )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    683,
+                    713,
+                    742,
+                    713,
+                    742,
+                    747,
+                    683,
+                    747
+                ],
+                "score": 0.91,
+                "latex": "\\tau / 1 0"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    475,
+                    2003,
+                    596,
+                    2003,
+                    596,
+                    2031,
+                    475,
+                    2031
+                ],
+                "score": 0.9,
+                "latex": "M = 5 0 0"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    297,
+                    623,
+                    346,
+                    623,
+                    346,
+                    655,
+                    297,
+                    655
+                ],
+                "score": 0.89,
+                "latex": "I ( t )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    337,
+                    714,
+                    439,
+                    714,
+                    439,
+                    742,
+                    337,
+                    742
+                ],
+                "score": 0.89,
+                "latex": "T = 5 0 0"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    934,
+                    621,
+                    987,
+                    621,
+                    987,
+                    651,
+                    934,
+                    651
+                ],
+                "score": 0.88,
+                "latex": "W ^ { \\mathrm { i n } }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    954,
+                    1752,
+                    1021,
+                    1752,
+                    1021,
+                    1781,
+                    954,
+                    1781
+                ],
+                "score": 0.87,
+                "latex": "W ^ { \\mathrm { o u t } }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1067,
+                    353,
+                    1182,
+                    353,
+                    1182,
+                    381,
+                    1067,
+                    381
+                ],
+                "score": 0.86,
+                "latex": "N = 1 0 0"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    297,
+                    655,
+                    322,
+                    655,
+                    322,
+                    683,
+                    297,
+                    683
+                ],
+                "score": 0.85,
+                "latex": "b _ { i }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    967,
+                    594,
+                    1029,
+                    594,
+                    1029,
+                    619,
+                    967,
+                    619
+                ],
+                "score": 0.85,
+                "latex": "W ^ { \\mathrm { r e c } }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    815,
+                    1751,
+                    868,
+                    1751,
+                    868,
+                    1782,
+                    815,
+                    1782
+                ],
+                "score": 0.73,
+                "latex": "W ^ { \\mathrm { i n } }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    883,
+                    1754,
+                    899,
+                    1754,
+                    899,
+                    1780,
+                    883,
+                    1780
+                ],
+                "score": 0.69,
+                "latex": "b"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    739,
+                    1753,
+                    801,
+                    1753,
+                    801,
+                    1781,
+                    739,
+                    1781
+                ],
+                "score": 0.54,
+                "latex": "W ^ { \\mathrm { r e c } }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    603,
+                    1788,
+                    628,
+                    1788,
+                    628,
+                    1812,
+                    603,
+                    1812
+                ],
+                "score": 0.36,
+                "latex": "\\mathbf { X } ^ { - }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    678,
+                    1789,
+                    695,
+                    1789,
+                    695,
+                    1815,
+                    678,
+                    1815
+                ],
+                "score": 0.32,
+                "latex": "_ \\textrm { y }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    739,
+                    1751,
+                    870,
+                    1751,
+                    870,
+                    1783,
+                    739,
+                    1783
+                ],
+                "score": 0.27,
+                "latex": "W ^ { \\mathrm { r e c } } , W ^ { \\mathrm { i n } } ,"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    988.0,
+                    666.0,
+                    988.0,
+                    666.0,
+                    1024.0,
+                    295.0,
+                    1024.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    290.0,
+                    219.0,
+                    468.0,
+                    219.0,
+                    468.0,
+                    269.0,
+                    290.0,
+                    269.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1690.0,
+                    494.0,
+                    1690.0,
+                    494.0,
+                    1732.0,
+                    293.0,
+                    1732.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    72.0,
+                    818.0,
+                    72.0,
+                    818.0,
+                    108.0,
+                    296.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    293.0,
+                    627.0,
+                    293.0,
+                    627.0,
+                    330.0,
+                    295.0,
+                    330.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    838.0,
+                    2085.0,
+                    862.0,
+                    2085.0,
+                    862.0,
+                    2118.0,
+                    838.0,
+                    2118.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    838.0,
+                    2085.0,
+                    862.0,
+                    2085.0,
+                    862.0,
+                    2118.0,
+                    838.0,
+                    2118.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    525.0,
+                    338.0,
+                    525.0,
+                    338.0,
+                    567.0,
+                    291.0,
+                    567.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    498.0,
+                    525.0,
+                    813.0,
+                    525.0,
+                    813.0,
+                    567.0,
+                    498.0,
+                    567.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    874.0,
+                    525.0,
+                    1333.0,
+                    525.0,
+                    1333.0,
+                    567.0,
+                    874.0,
+                    567.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1394.0,
+                    525.0,
+                    1405.0,
+                    525.0,
+                    1405.0,
+                    567.0,
+                    1394.0,
+                    567.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    561.0,
+                    938.0,
+                    561.0,
+                    938.0,
+                    596.0,
+                    295.0,
+                    596.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1175.0,
+                    561.0,
+                    1406.0,
+                    561.0,
+                    1406.0,
+                    596.0,
+                    1175.0,
+                    596.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    587.0,
+                    966.0,
+                    587.0,
+                    966.0,
+                    628.0,
+                    291.0,
+                    628.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1030.0,
+                    587.0,
+                    1405.0,
+                    587.0,
+                    1405.0,
+                    628.0,
+                    1030.0,
+                    628.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    618.0,
+                    296.0,
+                    618.0,
+                    296.0,
+                    659.0,
+                    293.0,
+                    659.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    347.0,
+                    618.0,
+                    933.0,
+                    618.0,
+                    933.0,
+                    659.0,
+                    347.0,
+                    659.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    988.0,
+                    618.0,
+                    1406.0,
+                    618.0,
+                    1406.0,
+                    659.0,
+                    988.0,
+                    659.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    654.0,
+                    577.0,
+                    654.0,
+                    577.0,
+                    685.0,
+                    323.0,
+                    685.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    635.0,
+                    654.0,
+                    1403.0,
+                    654.0,
+                    1403.0,
+                    685.0,
+                    635.0,
+                    685.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    683.0,
+                    1405.0,
+                    683.0,
+                    1405.0,
+                    716.0,
+                    294.0,
+                    716.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    711.0,
+                    336.0,
+                    711.0,
+                    336.0,
+                    747.0,
+                    293.0,
+                    747.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    440.0,
+                    711.0,
+                    682.0,
+                    711.0,
+                    682.0,
+                    747.0,
+                    440.0,
+                    747.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    743.0,
+                    711.0,
+                    752.0,
+                    711.0,
+                    752.0,
+                    747.0,
+                    743.0,
+                    747.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1045.0,
+                    1405.0,
+                    1045.0,
+                    1405.0,
+                    1083.0,
+                    295.0,
+                    1083.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1077.0,
+                    1406.0,
+                    1077.0,
+                    1406.0,
+                    1112.0,
+                    293.0,
+                    1112.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1107.0,
+                    1404.0,
+                    1107.0,
+                    1404.0,
+                    1141.0,
+                    294.0,
+                    1141.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1136.0,
+                    1406.0,
+                    1136.0,
+                    1406.0,
+                    1172.0,
+                    294.0,
+                    1172.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1167.0,
+                    1405.0,
+                    1167.0,
+                    1405.0,
+                    1202.0,
+                    294.0,
+                    1202.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1198.0,
+                    1406.0,
+                    1198.0,
+                    1406.0,
+                    1233.0,
+                    295.0,
+                    1233.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1227.0,
+                    1363.0,
+                    1227.0,
+                    1363.0,
+                    1263.0,
+                    294.0,
+                    1263.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1274.0,
+                    1404.0,
+                    1274.0,
+                    1404.0,
+                    1309.0,
+                    294.0,
+                    1309.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1308.0,
+                    1404.0,
+                    1308.0,
+                    1404.0,
+                    1339.0,
+                    296.0,
+                    1339.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1335.0,
+                    1406.0,
+                    1335.0,
+                    1406.0,
+                    1371.0,
+                    291.0,
+                    1371.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1368.0,
+                    1406.0,
+                    1368.0,
+                    1406.0,
+                    1402.0,
+                    292.0,
+                    1402.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1397.0,
+                    1406.0,
+                    1397.0,
+                    1406.0,
+                    1432.0,
+                    294.0,
+                    1432.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1428.0,
+                    1404.0,
+                    1428.0,
+                    1404.0,
+                    1462.0,
+                    295.0,
+                    1462.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1457.0,
+                    1276.0,
+                    1457.0,
+                    1276.0,
+                    1495.0,
+                    294.0,
+                    1495.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1506.0,
+                    1405.0,
+                    1506.0,
+                    1405.0,
+                    1539.0,
+                    296.0,
+                    1539.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1533.0,
+                    1405.0,
+                    1533.0,
+                    1405.0,
+                    1573.0,
+                    294.0,
+                    1573.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1567.0,
+                    1405.0,
+                    1567.0,
+                    1405.0,
+                    1601.0,
+                    295.0,
+                    1601.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1595.0,
+                    1405.0,
+                    1595.0,
+                    1405.0,
+                    1630.0,
+                    294.0,
+                    1630.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1626.0,
+                    1294.0,
+                    1626.0,
+                    1294.0,
+                    1662.0,
+                    294.0,
+                    1662.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1746.0,
+                    738.0,
+                    1746.0,
+                    738.0,
+                    1789.0,
+                    292.0,
+                    1789.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    871.0,
+                    1746.0,
+                    882.0,
+                    1746.0,
+                    882.0,
+                    1789.0,
+                    871.0,
+                    1789.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    900.0,
+                    1746.0,
+                    953.0,
+                    1746.0,
+                    953.0,
+                    1789.0,
+                    900.0,
+                    1789.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1022.0,
+                    1746.0,
+                    1406.0,
+                    1746.0,
+                    1406.0,
+                    1789.0,
+                    1022.0,
+                    1789.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1783.0,
+                    602.0,
+                    1783.0,
+                    602.0,
+                    1817.0,
+                    293.0,
+                    1817.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    629.0,
+                    1783.0,
+                    677.0,
+                    1783.0,
+                    677.0,
+                    1817.0,
+                    629.0,
+                    1817.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    696.0,
+                    1783.0,
+                    1406.0,
+                    1783.0,
+                    1406.0,
+                    1817.0,
+                    696.0,
+                    1817.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1814.0,
+                    1405.0,
+                    1814.0,
+                    1405.0,
+                    1850.0,
+                    292.0,
+                    1850.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1842.0,
+                    447.0,
+                    1842.0,
+                    447.0,
+                    1881.0,
+                    294.0,
+                    1881.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    760.0,
+                    1404.0,
+                    760.0,
+                    1404.0,
+                    794.0,
+                    296.0,
+                    794.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    793.0,
+                    1404.0,
+                    793.0,
+                    1404.0,
+                    824.0,
+                    297.0,
+                    824.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    822.0,
+                    398.0,
+                    822.0,
+                    398.0,
+                    856.0,
+                    294.0,
+                    856.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    460.0,
+                    822.0,
+                    509.0,
+                    822.0,
+                    509.0,
+                    856.0,
+                    460.0,
+                    856.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    571.0,
+                    822.0,
+                    988.0,
+                    822.0,
+                    988.0,
+                    856.0,
+                    571.0,
+                    856.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    351.0,
+                    1066.0,
+                    351.0,
+                    1066.0,
+                    387.0,
+                    296.0,
+                    387.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1183.0,
+                    351.0,
+                    1406.0,
+                    351.0,
+                    1406.0,
+                    387.0,
+                    1183.0,
+                    387.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    382.0,
+                    572.0,
+                    382.0,
+                    572.0,
+                    418.0,
+                    295.0,
+                    418.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    631.0,
+                    382.0,
+                    1307.0,
+                    382.0,
+                    1307.0,
+                    418.0,
+                    631.0,
+                    418.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1970.0,
+                    1403.0,
+                    1970.0,
+                    1403.0,
+                    2007.0,
+                    294.0,
+                    2007.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1997.0,
+                    474.0,
+                    1997.0,
+                    474.0,
+                    2040.0,
+                    294.0,
+                    2040.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    597.0,
+                    1997.0,
+                    1406.0,
+                    1997.0,
+                    1406.0,
+                    2040.0,
+                    597.0,
+                    2040.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 2,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    621,
+                    1404,
+                    621,
+                    1404,
+                    806,
+                    298,
+                    806
+                ],
+                "score": 0.981
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    228,
+                    1404,
+                    228,
+                    1404,
+                    353,
+                    298,
+                    353
+                ],
+                "score": 0.976
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    1607,
+                    1403,
+                    1607,
+                    1403,
+                    1701,
+                    299,
+                    1701
+                ],
+                "score": 0.974
+            },
+            {
+                "category_id": 3,
+                "poly": [
+                    351,
+                    835,
+                    1345,
+                    835,
+                    1345,
+                    1299,
+                    351,
+                    1299
+                ],
+                "score": 0.973
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1972,
+                    1402,
+                    1972,
+                    1402,
+                    2034,
+                    298,
+                    2034
+                ],
+                "score": 0.945
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    551,
+                    401,
+                    1154,
+                    401,
+                    1154,
+                    497,
+                    551,
+                    497
+                ],
+                "score": 0.945
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    545,
+                    504,
+                    911,
+                    504,
+                    911,
+                    599,
+                    545,
+                    599
+                ],
+                "score": 0.942
+            },
+            {
+                "category_id": 4,
+                "poly": [
+                    295,
+                    1328,
+                    1405,
+                    1328,
+                    1405,
+                    1484,
+                    295,
+                    1484
+                ],
+                "score": 0.928
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    299,
+                    1540,
+                    480,
+                    1540,
+                    480,
+                    1575,
+                    299,
+                    1575
+                ],
+                "score": 0.91
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    299,
+                    1737,
+                    918,
+                    1737,
+                    918,
+                    1768,
+                    299,
+                    1768
+                ],
+                "score": 0.907
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    300,
+                    1919,
+                    591,
+                    1919,
+                    591,
+                    1951,
+                    300,
+                    1951
+                ],
+                "score": 0.903
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    299,
+                    75,
+                    815,
+                    75,
+                    815,
+                    105,
+                    299,
+                    105
+                ],
+                "score": 0.902
+            },
+            {
+                "category_id": 9,
+                "poly": [
+                    1365,
+                    434,
+                    1400,
+                    434,
+                    1400,
+                    465,
+                    1365,
+                    465
+                ],
+                "score": 0.883
+            },
+            {
+                "category_id": 9,
+                "poly": [
+                    1366,
+                    536,
+                    1400,
+                    536,
+                    1400,
+                    567,
+                    1366,
+                    567
+                ],
+                "score": 0.882
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    841,
+                    2089,
+                    858,
+                    2089,
+                    858,
+                    2111,
+                    841,
+                    2111
+                ],
+                "score": 0.773
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    300,
+                    1793,
+                    1404,
+                    1793,
+                    1404,
+                    1887,
+                    300,
+                    1887
+                ],
+                "score": 0.682
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    545,
+                    398,
+                    1154,
+                    398,
+                    1154,
+                    604,
+                    545,
+                    604
+                ],
+                "score": 0.94,
+                "latex": "\\begin{array} { l } { { \\displaystyle R _ { L 2 } = \\frac { 1 } { N N _ { \\mathrm { i n } } } \\sum _ { i , j = 1 } ^ { N , N _ { \\mathrm { i n } } } ( W _ { i j } ^ { \\mathrm { i n } } ) ^ { 2 } + \\frac { 1 } { N N _ { \\mathrm { o u t } } } \\sum _ { i , j = 1 } ^ { N _ { \\mathrm { o u t } } , N } ( W _ { i j } ^ { \\mathrm { o u t } } ) ^ { 2 } } } \\\\ { { \\displaystyle R _ { F R } = \\frac { 1 } { N T M } \\sum _ { i , t , m = 1 } ^ { N , T , M } u _ { i } ( t , m ) ^ { 2 } } } \\end{array}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    569,
+                    712,
+                    657,
+                    712,
+                    657,
+                    746,
+                    569,
+                    746
+                ],
+                "score": 0.91,
+                "latex": "1 . 5 ^ { 2 } \\dot { / } N"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1286,
+                    774,
+                    1358,
+                    774,
+                    1358,
+                    808,
+                    1286,
+                    808
+                ],
+                "score": 0.91,
+                "latex": "1 / N _ { \\mathrm { i n } }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    678,
+                    775,
+                    730,
+                    775,
+                    730,
+                    803,
+                    678,
+                    803
+                ],
+                "score": 0.87,
+                "latex": "W ^ { \\mathrm { i n } }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    950,
+                    683,
+                    1013,
+                    683,
+                    1013,
+                    711,
+                    950,
+                    711
+                ],
+                "score": 0.87,
+                "latex": "W ^ { \\mathrm { r e c } }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    631,
+                    745,
+                    693,
+                    745,
+                    693,
+                    772,
+                    631,
+                    772
+                ],
+                "score": 0.86,
+                "latex": "W ^ { \\mathrm { r e c } }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    550,
+                    653,
+                    612,
+                    653,
+                    612,
+                    681,
+                    550,
+                    681
+                ],
+                "score": 0.86,
+                "latex": "W ^ { \\mathrm { r e c } }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    298,
+                    774,
+                    363,
+                    774,
+                    363,
+                    803,
+                    298,
+                    803
+                ],
+                "score": 0.8,
+                "latex": "W ^ { \\mathrm { o u t } }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1168,
+                    745,
+                    1184,
+                    745,
+                    1184,
+                    771,
+                    1168,
+                    771
+                ],
+                "score": 0.6,
+                "latex": "b"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    363.0,
+                    841.0,
+                    376.0,
+                    841.0,
+                    376.0,
+                    857.0,
+                    363.0,
+                    857.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    782.0,
+                    836.0,
+                    879.0,
+                    836.0,
+                    879.0,
+                    868.0,
+                    782.0,
+                    868.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    870.0,
+                    869.0,
+                    1332.0,
+                    869.0,
+                    1332.0,
+                    941.0,
+                    870.0,
+                    941.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    784.0,
+                    947.0,
+                    892.0,
+                    947.0,
+                    892.0,
+                    981.0,
+                    784.0,
+                    981.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    597.0,
+                    1027.0,
+                    613.0,
+                    1027.0,
+                    613.0,
+                    1035.0,
+                    597.0,
+                    1035.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    924.0,
+                    980.0,
+                    1321.0,
+                    980.0,
+                    1321.0,
+                    1056.0,
+                    924.0,
+                    1056.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    793.0,
+                    1063.0,
+                    878.0,
+                    1063.0,
+                    878.0,
+                    1100.0,
+                    793.0,
+                    1100.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1123.0,
+                    1127.0,
+                    1143.0,
+                    1127.0,
+                    1143.0,
+                    1145.0,
+                    1123.0,
+                    1145.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    793.0,
+                    1188.0,
+                    894.0,
+                    1188.0,
+                    894.0,
+                    1219.0,
+                    793.0,
+                    1219.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    392.0,
+                    1240.0,
+                    438.0,
+                    1240.0,
+                    438.0,
+                    1270.0,
+                    392.0,
+                    1270.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    456.0,
+                    1221.0,
+                    547.0,
+                    1221.0,
+                    547.0,
+                    1284.0,
+                    456.0,
+                    1284.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    552.0,
+                    1219.0,
+                    653.0,
+                    1219.0,
+                    653.0,
+                    1291.0,
+                    552.0,
+                    1291.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    655.0,
+                    1224.0,
+                    773.0,
+                    1224.0,
+                    773.0,
+                    1292.0,
+                    655.0,
+                    1292.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    795.0,
+                    1230.0,
+                    864.0,
+                    1230.0,
+                    864.0,
+                    1284.0,
+                    795.0,
+                    1284.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    910.0,
+                    1237.0,
+                    980.0,
+                    1237.0,
+                    980.0,
+                    1283.0,
+                    910.0,
+                    1283.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1002.0,
+                    1228.0,
+                    1225.0,
+                    1228.0,
+                    1225.0,
+                    1289.0,
+                    1002.0,
+                    1289.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1248.0,
+                    1250.0,
+                    1319.0,
+                    1250.0,
+                    1319.0,
+                    1280.0,
+                    1248.0,
+                    1280.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1330.0,
+                    1402.0,
+                    1330.0,
+                    1402.0,
+                    1363.0,
+                    295.0,
+                    1363.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1358.0,
+                    1405.0,
+                    1358.0,
+                    1405.0,
+                    1397.0,
+                    293.0,
+                    1397.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1390.0,
+                    1403.0,
+                    1390.0,
+                    1403.0,
+                    1427.0,
+                    295.0,
+                    1427.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1421.0,
+                    1406.0,
+                    1421.0,
+                    1406.0,
+                    1458.0,
+                    293.0,
+                    1458.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1453.0,
+                    1055.0,
+                    1453.0,
+                    1055.0,
+                    1486.0,
+                    295.0,
+                    1486.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1535.0,
+                    485.0,
+                    1535.0,
+                    485.0,
+                    1584.0,
+                    291.0,
+                    1584.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1736.0,
+                    920.0,
+                    1736.0,
+                    920.0,
+                    1771.0,
+                    295.0,
+                    1771.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1916.0,
+                    593.0,
+                    1916.0,
+                    593.0,
+                    1955.0,
+                    294.0,
+                    1955.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    73.0,
+                    818.0,
+                    73.0,
+                    818.0,
+                    108.0,
+                    295.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    838.0,
+                    2086.0,
+                    862.0,
+                    2086.0,
+                    862.0,
+                    2118.0,
+                    838.0,
+                    2118.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    619.0,
+                    1405.0,
+                    619.0,
+                    1405.0,
+                    656.0,
+                    295.0,
+                    656.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    647.0,
+                    549.0,
+                    647.0,
+                    549.0,
+                    687.0,
+                    292.0,
+                    687.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    613.0,
+                    647.0,
+                    1405.0,
+                    647.0,
+                    1405.0,
+                    687.0,
+                    613.0,
+                    687.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    678.0,
+                    949.0,
+                    678.0,
+                    949.0,
+                    717.0,
+                    291.0,
+                    717.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1014.0,
+                    678.0,
+                    1406.0,
+                    678.0,
+                    1406.0,
+                    717.0,
+                    1014.0,
+                    717.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    710.0,
+                    568.0,
+                    710.0,
+                    568.0,
+                    748.0,
+                    293.0,
+                    748.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    658.0,
+                    710.0,
+                    1406.0,
+                    710.0,
+                    1406.0,
+                    748.0,
+                    658.0,
+                    748.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    738.0,
+                    630.0,
+                    738.0,
+                    630.0,
+                    780.0,
+                    292.0,
+                    780.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    694.0,
+                    738.0,
+                    1167.0,
+                    738.0,
+                    1167.0,
+                    780.0,
+                    694.0,
+                    780.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1185.0,
+                    738.0,
+                    1406.0,
+                    738.0,
+                    1406.0,
+                    780.0,
+                    1185.0,
+                    780.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    768.0,
+                    297.0,
+                    768.0,
+                    297.0,
+                    812.0,
+                    293.0,
+                    812.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    364.0,
+                    768.0,
+                    677.0,
+                    768.0,
+                    677.0,
+                    812.0,
+                    364.0,
+                    812.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    731.0,
+                    768.0,
+                    1285.0,
+                    768.0,
+                    1285.0,
+                    812.0,
+                    731.0,
+                    812.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1359.0,
+                    768.0,
+                    1370.0,
+                    768.0,
+                    1370.0,
+                    812.0,
+                    1359.0,
+                    812.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    232.0,
+                    1402.0,
+                    232.0,
+                    1402.0,
+                    265.0,
+                    294.0,
+                    265.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    262.0,
+                    1405.0,
+                    262.0,
+                    1405.0,
+                    295.0,
+                    294.0,
+                    295.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    290.0,
+                    1404.0,
+                    290.0,
+                    1404.0,
+                    327.0,
+                    292.0,
+                    327.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    323.0,
+                    528.0,
+                    323.0,
+                    528.0,
+                    354.0,
+                    296.0,
+                    354.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1604.0,
+                    1403.0,
+                    1604.0,
+                    1403.0,
+                    1646.0,
+                    294.0,
+                    1646.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1639.0,
+                    1404.0,
+                    1639.0,
+                    1404.0,
+                    1673.0,
+                    295.0,
+                    1673.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1670.0,
+                    1070.0,
+                    1670.0,
+                    1070.0,
+                    1704.0,
+                    295.0,
+                    1704.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1971.0,
+                    1404.0,
+                    1971.0,
+                    1404.0,
+                    2007.0,
+                    296.0,
+                    2007.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    2003.0,
+                    1406.0,
+                    2003.0,
+                    1406.0,
+                    2037.0,
+                    293.0,
+                    2037.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1794.0,
+                    1404.0,
+                    1794.0,
+                    1404.0,
+                    1828.0,
+                    296.0,
+                    1828.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1824.0,
+                    1405.0,
+                    1824.0,
+                    1405.0,
+                    1858.0,
+                    295.0,
+                    1858.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1855.0,
+                    443.0,
+                    1855.0,
+                    443.0,
+                    1890.0,
+                    294.0,
+                    1890.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 3,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    659,
+                    1404,
+                    659,
+                    1404,
+                    1055,
+                    298,
+                    1055
+                ],
+                "score": 0.983
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1070,
+                    1403,
+                    1070,
+                    1403,
+                    1407,
+                    298,
+                    1407
+                ],
+                "score": 0.982
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1422,
+                    1404,
+                    1422,
+                    1404,
+                    1637,
+                    297,
+                    1637
+                ],
+                "score": 0.98
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    429,
+                    1403,
+                    429,
+                    1403,
+                    644,
+                    298,
+                    644
+                ],
+                "score": 0.979
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1653,
+                    1403,
+                    1653,
+                    1403,
+                    1837,
+                    298,
+                    1837
+                ],
+                "score": 0.978
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    229,
+                    1403,
+                    229,
+                    1403,
+                    414,
+                    299,
+                    414
+                ],
+                "score": 0.976
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    300,
+                    1942,
+                    1402,
+                    1942,
+                    1402,
+                    2034,
+                    300,
+                    2034
+                ],
+                "score": 0.973
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    300,
+                    76,
+                    815,
+                    76,
+                    815,
+                    104,
+                    300,
+                    104
+                ],
+                "score": 0.892
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    298,
+                    1885,
+                    939,
+                    1885,
+                    939,
+                    1914,
+                    298,
+                    1914
+                ],
+                "score": 0.754
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    841,
+                    2088,
+                    858,
+                    2088,
+                    858,
+                    2112,
+                    841,
+                    2112
+                ],
+                "score": 0.691
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    306,
+                    1884,
+                    933,
+                    1884,
+                    933,
+                    1913,
+                    306,
+                    1913
+                ],
+                "score": 0.364
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    841,
+                    2088,
+                    859,
+                    2088,
+                    859,
+                    2112,
+                    841,
+                    2112
+                ],
+                "score": 0.097
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    73.0,
+                    817.0,
+                    73.0,
+                    817.0,
+                    108.0,
+                    296.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1882.0,
+                    943.0,
+                    1882.0,
+                    943.0,
+                    1917.0,
+                    295.0,
+                    1917.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    839.0,
+                    2085.0,
+                    861.0,
+                    2085.0,
+                    861.0,
+                    2120.0,
+                    839.0,
+                    2120.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    299.0,
+                    1882.0,
+                    940.0,
+                    1882.0,
+                    940.0,
+                    1918.0,
+                    299.0,
+                    1918.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    838.0,
+                    2085.0,
+                    862.0,
+                    2085.0,
+                    862.0,
+                    2119.0,
+                    838.0,
+                    2119.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    659.0,
+                    1402.0,
+                    659.0,
+                    1402.0,
+                    692.0,
+                    297.0,
+                    692.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    685.0,
+                    1406.0,
+                    685.0,
+                    1406.0,
+                    728.0,
+                    293.0,
+                    728.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    719.0,
+                    1408.0,
+                    719.0,
+                    1408.0,
+                    759.0,
+                    293.0,
+                    759.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    747.0,
+                    1405.0,
+                    747.0,
+                    1405.0,
+                    790.0,
+                    292.0,
+                    790.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    783.0,
+                    1405.0,
+                    783.0,
+                    1405.0,
+                    816.0,
+                    296.0,
+                    816.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    812.0,
+                    1405.0,
+                    812.0,
+                    1405.0,
+                    848.0,
+                    296.0,
+                    848.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    841.0,
+                    1405.0,
+                    841.0,
+                    1405.0,
+                    879.0,
+                    293.0,
+                    879.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    871.0,
+                    1405.0,
+                    871.0,
+                    1405.0,
+                    911.0,
+                    292.0,
+                    911.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    902.0,
+                    1406.0,
+                    902.0,
+                    1406.0,
+                    942.0,
+                    291.0,
+                    942.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    933.0,
+                    1405.0,
+                    933.0,
+                    1405.0,
+                    969.0,
+                    294.0,
+                    969.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    964.0,
+                    1404.0,
+                    964.0,
+                    1404.0,
+                    998.0,
+                    293.0,
+                    998.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    991.0,
+                    1405.0,
+                    991.0,
+                    1405.0,
+                    1033.0,
+                    292.0,
+                    1033.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1025.0,
+                    841.0,
+                    1025.0,
+                    841.0,
+                    1057.0,
+                    296.0,
+                    1057.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1068.0,
+                    1404.0,
+                    1068.0,
+                    1404.0,
+                    1108.0,
+                    292.0,
+                    1108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1103.0,
+                    1405.0,
+                    1103.0,
+                    1405.0,
+                    1138.0,
+                    296.0,
+                    1138.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1134.0,
+                    1404.0,
+                    1134.0,
+                    1404.0,
+                    1165.0,
+                    294.0,
+                    1165.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1161.0,
+                    1407.0,
+                    1161.0,
+                    1407.0,
+                    1199.0,
+                    293.0,
+                    1199.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1194.0,
+                    1405.0,
+                    1194.0,
+                    1405.0,
+                    1229.0,
+                    294.0,
+                    1229.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1222.0,
+                    1407.0,
+                    1222.0,
+                    1407.0,
+                    1261.0,
+                    292.0,
+                    1261.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1255.0,
+                    1405.0,
+                    1255.0,
+                    1405.0,
+                    1289.0,
+                    294.0,
+                    1289.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1279.0,
+                    1405.0,
+                    1279.0,
+                    1405.0,
+                    1323.0,
+                    292.0,
+                    1323.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1313.0,
+                    1404.0,
+                    1313.0,
+                    1404.0,
+                    1354.0,
+                    292.0,
+                    1354.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1344.0,
+                    1405.0,
+                    1344.0,
+                    1405.0,
+                    1381.0,
+                    293.0,
+                    1381.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1378.0,
+                    626.0,
+                    1378.0,
+                    626.0,
+                    1412.0,
+                    294.0,
+                    1412.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1424.0,
+                    1404.0,
+                    1424.0,
+                    1404.0,
+                    1458.0,
+                    295.0,
+                    1458.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1452.0,
+                    1404.0,
+                    1452.0,
+                    1404.0,
+                    1486.0,
+                    295.0,
+                    1486.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1484.0,
+                    1405.0,
+                    1484.0,
+                    1405.0,
+                    1520.0,
+                    292.0,
+                    1520.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1515.0,
+                    1402.0,
+                    1515.0,
+                    1402.0,
+                    1549.0,
+                    294.0,
+                    1549.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1543.0,
+                    1405.0,
+                    1543.0,
+                    1405.0,
+                    1581.0,
+                    292.0,
+                    1581.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1576.0,
+                    1404.0,
+                    1576.0,
+                    1404.0,
+                    1611.0,
+                    295.0,
+                    1611.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1607.0,
+                    975.0,
+                    1607.0,
+                    975.0,
+                    1638.0,
+                    293.0,
+                    1638.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    429.0,
+                    1404.0,
+                    429.0,
+                    1404.0,
+                    464.0,
+                    296.0,
+                    464.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    461.0,
+                    1405.0,
+                    461.0,
+                    1405.0,
+                    496.0,
+                    293.0,
+                    496.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    487.0,
+                    1404.0,
+                    487.0,
+                    1404.0,
+                    528.0,
+                    292.0,
+                    528.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    522.0,
+                    1405.0,
+                    522.0,
+                    1405.0,
+                    556.0,
+                    294.0,
+                    556.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    550.0,
+                    1405.0,
+                    550.0,
+                    1405.0,
+                    588.0,
+                    293.0,
+                    588.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    583.0,
+                    1405.0,
+                    583.0,
+                    1405.0,
+                    618.0,
+                    294.0,
+                    618.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    614.0,
+                    824.0,
+                    614.0,
+                    824.0,
+                    648.0,
+                    294.0,
+                    648.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1652.0,
+                    1405.0,
+                    1652.0,
+                    1405.0,
+                    1688.0,
+                    296.0,
+                    1688.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1683.0,
+                    1404.0,
+                    1683.0,
+                    1404.0,
+                    1718.0,
+                    294.0,
+                    1718.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1713.0,
+                    1405.0,
+                    1713.0,
+                    1405.0,
+                    1749.0,
+                    293.0,
+                    1749.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1742.0,
+                    1405.0,
+                    1742.0,
+                    1405.0,
+                    1779.0,
+                    293.0,
+                    1779.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1774.0,
+                    1405.0,
+                    1774.0,
+                    1405.0,
+                    1810.0,
+                    294.0,
+                    1810.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1805.0,
+                    1041.0,
+                    1805.0,
+                    1041.0,
+                    1840.0,
+                    294.0,
+                    1840.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    229.0,
+                    1405.0,
+                    229.0,
+                    1405.0,
+                    266.0,
+                    294.0,
+                    266.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    257.0,
+                    1407.0,
+                    257.0,
+                    1407.0,
+                    300.0,
+                    292.0,
+                    300.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    290.0,
+                    1407.0,
+                    290.0,
+                    1407.0,
+                    328.0,
+                    292.0,
+                    328.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    323.0,
+                    1404.0,
+                    323.0,
+                    1404.0,
+                    355.0,
+                    297.0,
+                    355.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    351.0,
+                    1405.0,
+                    351.0,
+                    1405.0,
+                    388.0,
+                    294.0,
+                    388.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    383.0,
+                    606.0,
+                    383.0,
+                    606.0,
+                    414.0,
+                    295.0,
+                    414.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1941.0,
+                    1406.0,
+                    1941.0,
+                    1406.0,
+                    1976.0,
+                    294.0,
+                    1976.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1974.0,
+                    1406.0,
+                    1974.0,
+                    1406.0,
+                    2007.0,
+                    295.0,
+                    2007.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    2002.0,
+                    1407.0,
+                    2002.0,
+                    1407.0,
+                    2040.0,
+                    295.0,
+                    2040.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 4,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1589,
+                    1403,
+                    1589,
+                    1403,
+                    1926,
+                    298,
+                    1926
+                ],
+                "score": 0.982
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    955,
+                    1403,
+                    955,
+                    1403,
+                    1260,
+                    298,
+                    1260
+                ],
+                "score": 0.981
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1275,
+                    1404,
+                    1275,
+                    1404,
+                    1490,
+                    298,
+                    1490
+                ],
+                "score": 0.98
+            },
+            {
+                "category_id": 3,
+                "poly": [
+                    364,
+                    233,
+                    1335,
+                    233,
+                    1335,
+                    559,
+                    364,
+                    559
+                ],
+                "score": 0.974
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    301,
+                    1942,
+                    1401,
+                    1942,
+                    1401,
+                    2034,
+                    301,
+                    2034
+                ],
+                "score": 0.97
+            },
+            {
+                "category_id": 4,
+                "poly": [
+                    297,
+                    595,
+                    1405,
+                    595,
+                    1405,
+                    809,
+                    297,
+                    809
+                ],
+                "score": 0.953
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    302,
+                    877,
+                    1400,
+                    877,
+                    1400,
+                    941,
+                    302,
+                    941
+                ],
+                "score": 0.943
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    299,
+                    76,
+                    815,
+                    76,
+                    815,
+                    104,
+                    299,
+                    104
+                ],
+                "score": 0.89
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    299,
+                    1535,
+                    914,
+                    1535,
+                    914,
+                    1564,
+                    299,
+                    1564
+                ],
+                "score": 0.888
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    840,
+                    2089,
+                    858,
+                    2089,
+                    858,
+                    2112,
+                    840,
+                    2112
+                ],
+                "score": 0.774
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    457,
+                    748,
+                    507,
+                    748,
+                    507,
+                    779,
+                    457,
+                    779
+                ],
+                "score": 0.49,
+                "latex": "{ \\bf g } , { \\bf h } )"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    370.0,
+                    244.0,
+                    389.0,
+                    244.0,
+                    389.0,
+                    265.0,
+                    370.0,
+                    265.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1031.0,
+                    238.0,
+                    1054.0,
+                    238.0,
+                    1054.0,
+                    264.0,
+                    1031.0,
+                    264.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    371.0,
+                    341.0,
+                    389.0,
+                    341.0,
+                    389.0,
+                    362.0,
+                    371.0,
+                    362.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    693.0,
+                    346.0,
+                    716.0,
+                    346.0,
+                    716.0,
+                    369.0,
+                    693.0,
+                    369.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1032.0,
+                    342.0,
+                    1054.0,
+                    342.0,
+                    1054.0,
+                    368.0,
+                    1032.0,
+                    368.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    539.0,
+                    383.0,
+                    548.0,
+                    383.0,
+                    548.0,
+                    390.0,
+                    539.0,
+                    390.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    376.0,
+                    457.0,
+                    387.0,
+                    457.0,
+                    387.0,
+                    466.0,
+                    376.0,
+                    466.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    481.0,
+                    460.0,
+                    590.0,
+                    460.0,
+                    590.0,
+                    523.0,
+                    481.0,
+                    523.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    693.0,
+                    445.0,
+                    713.0,
+                    445.0,
+                    713.0,
+                    471.0,
+                    693.0,
+                    471.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    481.0,
+                    526.0,
+                    663.0,
+                    526.0,
+                    663.0,
+                    564.0,
+                    481.0,
+                    564.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    818.0,
+                    526.0,
+                    997.0,
+                    526.0,
+                    997.0,
+                    558.0,
+                    818.0,
+                    558.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1152.0,
+                    529.0,
+                    1320.0,
+                    529.0,
+                    1320.0,
+                    558.0,
+                    1152.0,
+                    558.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    694.75,
+                    239.5,
+                    709.75,
+                    239.5,
+                    709.75,
+                    264.0,
+                    694.75,
+                    264.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    595.0,
+                    1405.0,
+                    595.0,
+                    1405.0,
+                    630.0,
+                    295.0,
+                    630.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    626.0,
+                    1402.0,
+                    626.0,
+                    1402.0,
+                    660.0,
+                    295.0,
+                    660.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    654.0,
+                    1405.0,
+                    654.0,
+                    1405.0,
+                    690.0,
+                    294.0,
+                    690.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    687.0,
+                    1405.0,
+                    687.0,
+                    1405.0,
+                    722.0,
+                    295.0,
+                    722.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    715.0,
+                    1406.0,
+                    715.0,
+                    1406.0,
+                    754.0,
+                    292.0,
+                    754.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    746.0,
+                    456.0,
+                    746.0,
+                    456.0,
+                    784.0,
+                    295.0,
+                    784.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    508.0,
+                    746.0,
+                    1406.0,
+                    746.0,
+                    1406.0,
+                    784.0,
+                    508.0,
+                    784.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    776.0,
+                    1265.0,
+                    776.0,
+                    1265.0,
+                    817.0,
+                    294.0,
+                    817.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    73.0,
+                    818.0,
+                    73.0,
+                    818.0,
+                    108.0,
+                    295.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1532.0,
+                    917.0,
+                    1532.0,
+                    917.0,
+                    1567.0,
+                    296.0,
+                    1567.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    840.0,
+                    2087.0,
+                    861.0,
+                    2087.0,
+                    861.0,
+                    2118.0,
+                    840.0,
+                    2118.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1591.0,
+                    1404.0,
+                    1591.0,
+                    1404.0,
+                    1626.0,
+                    296.0,
+                    1626.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1618.0,
+                    1404.0,
+                    1618.0,
+                    1404.0,
+                    1659.0,
+                    292.0,
+                    1659.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1652.0,
+                    1403.0,
+                    1652.0,
+                    1403.0,
+                    1687.0,
+                    294.0,
+                    1687.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1682.0,
+                    1404.0,
+                    1682.0,
+                    1404.0,
+                    1717.0,
+                    294.0,
+                    1717.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1708.0,
+                    1405.0,
+                    1708.0,
+                    1405.0,
+                    1750.0,
+                    292.0,
+                    1750.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1744.0,
+                    1405.0,
+                    1744.0,
+                    1405.0,
+                    1779.0,
+                    294.0,
+                    1779.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1774.0,
+                    1405.0,
+                    1774.0,
+                    1405.0,
+                    1808.0,
+                    294.0,
+                    1808.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1798.0,
+                    1407.0,
+                    1798.0,
+                    1407.0,
+                    1843.0,
+                    291.0,
+                    1843.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1833.0,
+                    1404.0,
+                    1833.0,
+                    1404.0,
+                    1872.0,
+                    293.0,
+                    1872.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1864.0,
+                    1407.0,
+                    1864.0,
+                    1407.0,
+                    1903.0,
+                    293.0,
+                    1903.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1895.0,
+                    955.0,
+                    1895.0,
+                    955.0,
+                    1930.0,
+                    296.0,
+                    1930.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    954.0,
+                    1404.0,
+                    954.0,
+                    1404.0,
+                    989.0,
+                    294.0,
+                    989.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    985.0,
+                    1405.0,
+                    985.0,
+                    1405.0,
+                    1021.0,
+                    292.0,
+                    1021.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1016.0,
+                    1407.0,
+                    1016.0,
+                    1407.0,
+                    1052.0,
+                    294.0,
+                    1052.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1045.0,
+                    1405.0,
+                    1045.0,
+                    1405.0,
+                    1081.0,
+                    291.0,
+                    1081.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1076.0,
+                    1405.0,
+                    1076.0,
+                    1405.0,
+                    1113.0,
+                    293.0,
+                    1113.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1107.0,
+                    1405.0,
+                    1107.0,
+                    1405.0,
+                    1143.0,
+                    294.0,
+                    1143.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1133.0,
+                    1407.0,
+                    1133.0,
+                    1407.0,
+                    1177.0,
+                    292.0,
+                    1177.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1168.0,
+                    1404.0,
+                    1168.0,
+                    1404.0,
+                    1204.0,
+                    294.0,
+                    1204.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1195.0,
+                    1407.0,
+                    1195.0,
+                    1407.0,
+                    1235.0,
+                    292.0,
+                    1235.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1229.0,
+                    417.0,
+                    1229.0,
+                    417.0,
+                    1263.0,
+                    292.0,
+                    1263.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1275.0,
+                    1405.0,
+                    1275.0,
+                    1405.0,
+                    1314.0,
+                    294.0,
+                    1314.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1305.0,
+                    1406.0,
+                    1305.0,
+                    1406.0,
+                    1343.0,
+                    294.0,
+                    1343.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    1339.0,
+                    1402.0,
+                    1339.0,
+                    1402.0,
+                    1370.0,
+                    297.0,
+                    1370.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1364.0,
+                    1406.0,
+                    1364.0,
+                    1406.0,
+                    1406.0,
+                    291.0,
+                    1406.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1398.0,
+                    1404.0,
+                    1398.0,
+                    1404.0,
+                    1433.0,
+                    294.0,
+                    1433.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1426.0,
+                    1405.0,
+                    1426.0,
+                    1405.0,
+                    1467.0,
+                    291.0,
+                    1467.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1461.0,
+                    756.0,
+                    1461.0,
+                    756.0,
+                    1492.0,
+                    296.0,
+                    1492.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1939.0,
+                    1405.0,
+                    1939.0,
+                    1405.0,
+                    1977.0,
+                    295.0,
+                    1977.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1972.0,
+                    1403.0,
+                    1972.0,
+                    1403.0,
+                    2005.0,
+                    296.0,
+                    2005.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    2001.0,
+                    1405.0,
+                    2001.0,
+                    1405.0,
+                    2037.0,
+                    295.0,
+                    2037.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    872.0,
+                    1405.0,
+                    872.0,
+                    1405.0,
+                    914.0,
+                    294.0,
+                    914.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    908.0,
+                    704.0,
+                    908.0,
+                    704.0,
+                    944.0,
+                    297.0,
+                    944.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 5,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1404,
+                    1403,
+                    1404,
+                    1403,
+                    1739,
+                    298,
+                    1739
+                ],
+                "score": 0.983
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    1757,
+                    1403,
+                    1757,
+                    1403,
+                    1878,
+                    299,
+                    1878
+                ],
+                "score": 0.975
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    229,
+                    1403,
+                    229,
+                    1403,
+                    353,
+                    298,
+                    353
+                ],
+                "score": 0.973
+            },
+            {
+                "category_id": 3,
+                "poly": [
+                    324,
+                    390,
+                    1358,
+                    390,
+                    1358,
+                    808,
+                    324,
+                    808
+                ],
+                "score": 0.971
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    301,
+                    1972,
+                    1401,
+                    1972,
+                    1401,
+                    2034,
+                    301,
+                    2034
+                ],
+                "score": 0.944
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    299,
+                    1917,
+                    916,
+                    1917,
+                    916,
+                    1945,
+                    299,
+                    1945
+                ],
+                "score": 0.904
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    300,
+                    76,
+                    815,
+                    76,
+                    815,
+                    104,
+                    300,
+                    104
+                ],
+                "score": 0.892
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    301,
+                    1347,
+                    832,
+                    1347,
+                    832,
+                    1377,
+                    301,
+                    1377
+                ],
+                "score": 0.885
+            },
+            {
+                "category_id": 4,
+                "poly": [
+                    296,
+                    865,
+                    1404,
+                    865,
+                    1404,
+                    1294,
+                    296,
+                    1294
+                ],
+                "score": 0.875
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    842,
+                    2088,
+                    858,
+                    2088,
+                    858,
+                    2111,
+                    842,
+                    2111
+                ],
+                "score": 0.604
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    842,
+                    2088,
+                    858,
+                    2088,
+                    858,
+                    2111,
+                    842,
+                    2111
+                ],
+                "score": 0.288
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    296,
+                    865,
+                    1404,
+                    865,
+                    1404,
+                    1294,
+                    296,
+                    1294
+                ],
+                "score": 0.133
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1109,
+                    1496,
+                    1166,
+                    1496,
+                    1166,
+                    1529,
+                    1109,
+                    1529
+                ],
+                "score": 0.9,
+                "latex": "\\xi _ { i } ( t )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    730,
+                    959,
+                    842,
+                    959,
+                    842,
+                    988,
+                    730,
+                    988
+                ],
+                "score": 0.87,
+                "latex": "N = 1 0 0"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    424,
+                    1496,
+                    479,
+                    1496,
+                    479,
+                    1525,
+                    424,
+                    1525
+                ],
+                "score": 0.86,
+                "latex": "90 \\%"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    326.0,
+                    398.0,
+                    345.0,
+                    398.0,
+                    345.0,
+                    420.0,
+                    326.0,
+                    420.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1207.0,
+                    427.0,
+                    1256.0,
+                    427.0,
+                    1256.0,
+                    479.0,
+                    1207.0,
+                    479.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1266.0,
+                    593.0,
+                    1285.0,
+                    593.0,
+                    1285.0,
+                    612.0,
+                    1266.0,
+                    612.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    685.0,
+                    645.0,
+                    697.0,
+                    645.0,
+                    697.0,
+                    656.0,
+                    685.0,
+                    656.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    768.0,
+                    674.0,
+                    821.0,
+                    674.0,
+                    821.0,
+                    713.0,
+                    768.0,
+                    713.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    771.0,
+                    720.0,
+                    888.0,
+                    720.0,
+                    888.0,
+                    747.0,
+                    771.0,
+                    747.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    555.0,
+                    777.0,
+                    565.0,
+                    777.0,
+                    565.0,
+                    787.0,
+                    555.0,
+                    787.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    770.0,
+                    761.0,
+                    819.0,
+                    761.0,
+                    819.0,
+                    790.0,
+                    770.0,
+                    790.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1106.0,
+                    792.0,
+                    1117.0,
+                    792.0,
+                    1117.0,
+                    802.0,
+                    1106.0,
+                    802.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1915.0,
+                    919.0,
+                    1915.0,
+                    919.0,
+                    1950.0,
+                    294.0,
+                    1950.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    73.0,
+                    817.0,
+                    73.0,
+                    817.0,
+                    108.0,
+                    296.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1347.0,
+                    839.0,
+                    1347.0,
+                    839.0,
+                    1381.0,
+                    294.0,
+                    1381.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    866.0,
+                    1405.0,
+                    866.0,
+                    1405.0,
+                    901.0,
+                    294.0,
+                    901.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    894.0,
+                    1404.0,
+                    894.0,
+                    1404.0,
+                    936.0,
+                    291.0,
+                    936.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    926.0,
+                    1404.0,
+                    926.0,
+                    1404.0,
+                    963.0,
+                    295.0,
+                    963.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    957.0,
+                    729.0,
+                    957.0,
+                    729.0,
+                    994.0,
+                    293.0,
+                    994.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    843.0,
+                    957.0,
+                    1405.0,
+                    957.0,
+                    1405.0,
+                    994.0,
+                    843.0,
+                    994.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    987.0,
+                    1402.0,
+                    987.0,
+                    1402.0,
+                    1022.0,
+                    294.0,
+                    1022.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1018.0,
+                    1405.0,
+                    1018.0,
+                    1405.0,
+                    1054.0,
+                    295.0,
+                    1054.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1049.0,
+                    1406.0,
+                    1049.0,
+                    1406.0,
+                    1085.0,
+                    293.0,
+                    1085.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1080.0,
+                    1406.0,
+                    1080.0,
+                    1406.0,
+                    1115.0,
+                    293.0,
+                    1115.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1109.0,
+                    1406.0,
+                    1109.0,
+                    1406.0,
+                    1148.0,
+                    291.0,
+                    1148.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1141.0,
+                    1405.0,
+                    1141.0,
+                    1405.0,
+                    1176.0,
+                    294.0,
+                    1176.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1167.0,
+                    1405.0,
+                    1167.0,
+                    1405.0,
+                    1212.0,
+                    291.0,
+                    1212.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1199.0,
+                    1405.0,
+                    1199.0,
+                    1405.0,
+                    1236.0,
+                    294.0,
+                    1236.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1235.0,
+                    1404.0,
+                    1235.0,
+                    1404.0,
+                    1266.0,
+                    295.0,
+                    1266.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1260.0,
+                    497.0,
+                    1260.0,
+                    497.0,
+                    1301.0,
+                    293.0,
+                    1301.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    840.0,
+                    2086.0,
+                    860.0,
+                    2086.0,
+                    860.0,
+                    2119.0,
+                    840.0,
+                    2119.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    840.0,
+                    2086.0,
+                    860.0,
+                    2086.0,
+                    860.0,
+                    2119.0,
+                    840.0,
+                    2119.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1402.0,
+                    1404.0,
+                    1402.0,
+                    1404.0,
+                    1442.0,
+                    293.0,
+                    1442.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1436.0,
+                    1404.0,
+                    1436.0,
+                    1404.0,
+                    1470.0,
+                    294.0,
+                    1470.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1460.0,
+                    1405.0,
+                    1460.0,
+                    1405.0,
+                    1505.0,
+                    292.0,
+                    1505.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1496.0,
+                    423.0,
+                    1496.0,
+                    423.0,
+                    1531.0,
+                    294.0,
+                    1531.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    480.0,
+                    1496.0,
+                    1108.0,
+                    1496.0,
+                    1108.0,
+                    1531.0,
+                    480.0,
+                    1531.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1167.0,
+                    1496.0,
+                    1405.0,
+                    1496.0,
+                    1405.0,
+                    1531.0,
+                    1167.0,
+                    1531.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1527.0,
+                    1405.0,
+                    1527.0,
+                    1405.0,
+                    1562.0,
+                    294.0,
+                    1562.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1558.0,
+                    1405.0,
+                    1558.0,
+                    1405.0,
+                    1593.0,
+                    294.0,
+                    1593.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1586.0,
+                    1403.0,
+                    1586.0,
+                    1403.0,
+                    1622.0,
+                    293.0,
+                    1622.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1617.0,
+                    1404.0,
+                    1617.0,
+                    1404.0,
+                    1651.0,
+                    294.0,
+                    1651.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1649.0,
+                    1405.0,
+                    1649.0,
+                    1405.0,
+                    1684.0,
+                    294.0,
+                    1684.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1678.0,
+                    1407.0,
+                    1678.0,
+                    1407.0,
+                    1716.0,
+                    293.0,
+                    1716.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1710.0,
+                    953.0,
+                    1710.0,
+                    953.0,
+                    1745.0,
+                    294.0,
+                    1745.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1755.0,
+                    1403.0,
+                    1755.0,
+                    1403.0,
+                    1791.0,
+                    294.0,
+                    1791.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1788.0,
+                    1405.0,
+                    1788.0,
+                    1405.0,
+                    1820.0,
+                    295.0,
+                    1820.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1816.0,
+                    1404.0,
+                    1816.0,
+                    1404.0,
+                    1854.0,
+                    293.0,
+                    1854.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1849.0,
+                    1065.0,
+                    1849.0,
+                    1065.0,
+                    1881.0,
+                    295.0,
+                    1881.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    228.0,
+                    1404.0,
+                    228.0,
+                    1404.0,
+                    265.0,
+                    292.0,
+                    265.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    261.0,
+                    1404.0,
+                    261.0,
+                    1404.0,
+                    296.0,
+                    293.0,
+                    296.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    290.0,
+                    1407.0,
+                    290.0,
+                    1407.0,
+                    327.0,
+                    292.0,
+                    327.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    323.0,
+                    466.0,
+                    323.0,
+                    466.0,
+                    356.0,
+                    294.0,
+                    356.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    1970.0,
+                    1402.0,
+                    1970.0,
+                    1402.0,
+                    2006.0,
+                    297.0,
+                    2006.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    2002.0,
+                    1405.0,
+                    2002.0,
+                    1405.0,
+                    2038.0,
+                    295.0,
+                    2038.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    866.0,
+                    1405.0,
+                    866.0,
+                    1405.0,
+                    901.0,
+                    294.0,
+                    901.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    894.0,
+                    1404.0,
+                    894.0,
+                    1404.0,
+                    936.0,
+                    291.0,
+                    936.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    926.0,
+                    1404.0,
+                    926.0,
+                    1404.0,
+                    963.0,
+                    295.0,
+                    963.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    957.0,
+                    729.0,
+                    957.0,
+                    729.0,
+                    994.0,
+                    293.0,
+                    994.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    843.0,
+                    957.0,
+                    1405.0,
+                    957.0,
+                    1405.0,
+                    994.0,
+                    843.0,
+                    994.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    987.0,
+                    1402.0,
+                    987.0,
+                    1402.0,
+                    1022.0,
+                    294.0,
+                    1022.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1018.0,
+                    1405.0,
+                    1018.0,
+                    1405.0,
+                    1054.0,
+                    295.0,
+                    1054.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1049.0,
+                    1406.0,
+                    1049.0,
+                    1406.0,
+                    1085.0,
+                    293.0,
+                    1085.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1080.0,
+                    1406.0,
+                    1080.0,
+                    1406.0,
+                    1115.0,
+                    293.0,
+                    1115.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1109.0,
+                    1406.0,
+                    1109.0,
+                    1406.0,
+                    1148.0,
+                    291.0,
+                    1148.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1141.0,
+                    1405.0,
+                    1141.0,
+                    1405.0,
+                    1176.0,
+                    294.0,
+                    1176.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1167.0,
+                    1405.0,
+                    1167.0,
+                    1405.0,
+                    1212.0,
+                    291.0,
+                    1212.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1199.0,
+                    1405.0,
+                    1199.0,
+                    1405.0,
+                    1236.0,
+                    294.0,
+                    1236.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1235.0,
+                    1404.0,
+                    1235.0,
+                    1404.0,
+                    1266.0,
+                    295.0,
+                    1266.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1260.0,
+                    497.0,
+                    1260.0,
+                    497.0,
+                    1301.0,
+                    293.0,
+                    1301.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 6,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    1803,
+                    1403,
+                    1803,
+                    1403,
+                    1957,
+                    299,
+                    1957
+                ],
+                "score": 0.976
+            },
+            {
+                "category_id": 3,
+                "poly": [
+                    314,
+                    221,
+                    1350,
+                    221,
+                    1350,
+                    736,
+                    314,
+                    736
+                ],
+                "score": 0.974
+            },
+            {
+                "category_id": 4,
+                "poly": [
+                    296,
+                    1334,
+                    1405,
+                    1334,
+                    1405,
+                    1731,
+                    296,
+                    1731
+                ],
+                "score": 0.974
+            },
+            {
+                "category_id": 3,
+                "poly": [
+                    357,
+                    963,
+                    1337,
+                    963,
+                    1337,
+                    1306,
+                    357,
+                    1306
+                ],
+                "score": 0.97
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1972,
+                    1400,
+                    1972,
+                    1400,
+                    2034,
+                    298,
+                    2034
+                ],
+                "score": 0.938
+            },
+            {
+                "category_id": 4,
+                "poly": [
+                    295,
+                    761,
+                    1407,
+                    761,
+                    1407,
+                    916,
+                    295,
+                    916
+                ],
+                "score": 0.937
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    299,
+                    76,
+                    815,
+                    76,
+                    815,
+                    104,
+                    299,
+                    104
+                ],
+                "score": 0.893
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    841,
+                    2089,
+                    858,
+                    2089,
+                    858,
+                    2111,
+                    841,
+                    2111
+                ],
+                "score": 0.789
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    327.0,
+                    233.0,
+                    346.0,
+                    233.0,
+                    346.0,
+                    257.0,
+                    327.0,
+                    257.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    361.0,
+                    241.0,
+                    821.0,
+                    241.0,
+                    821.0,
+                    290.0,
+                    361.0,
+                    290.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    854.0,
+                    229.0,
+                    882.0,
+                    229.0,
+                    882.0,
+                    263.0,
+                    854.0,
+                    263.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    884.0,
+                    237.0,
+                    1350.0,
+                    237.0,
+                    1350.0,
+                    294.0,
+                    884.0,
+                    294.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    931.0,
+                    294.0,
+                    1347.0,
+                    294.0,
+                    1347.0,
+                    339.0,
+                    931.0,
+                    339.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    886.0,
+                    339.0,
+                    1349.0,
+                    339.0,
+                    1349.0,
+                    392.0,
+                    886.0,
+                    392.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    888.0,
+                    390.0,
+                    1349.0,
+                    390.0,
+                    1349.0,
+                    439.0,
+                    888.0,
+                    439.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    552.0,
+                    450.0,
+                    581.0,
+                    450.0,
+                    581.0,
+                    472.0,
+                    552.0,
+                    472.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    726.0,
+                    444.0,
+                    817.0,
+                    444.0,
+                    817.0,
+                    482.0,
+                    726.0,
+                    482.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    886.0,
+                    439.0,
+                    1349.0,
+                    439.0,
+                    1349.0,
+                    488.0,
+                    886.0,
+                    488.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    537.0,
+                    493.0,
+                    632.0,
+                    493.0,
+                    632.0,
+                    529.0,
+                    537.0,
+                    529.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    727.0,
+                    493.0,
+                    814.0,
+                    493.0,
+                    814.0,
+                    531.0,
+                    727.0,
+                    531.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    924.0,
+                    486.0,
+                    1349.0,
+                    486.0,
+                    1349.0,
+                    543.0,
+                    924.0,
+                    543.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    714.0,
+                    537.0,
+                    821.0,
+                    537.0,
+                    821.0,
+                    585.0,
+                    714.0,
+                    585.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    904.0,
+                    552.0,
+                    923.0,
+                    552.0,
+                    923.0,
+                    566.0,
+                    904.0,
+                    566.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    933.0,
+                    534.0,
+                    1348.0,
+                    534.0,
+                    1348.0,
+                    587.0,
+                    933.0,
+                    587.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    462.0,
+                    599.0,
+                    492.0,
+                    599.0,
+                    492.0,
+                    620.0,
+                    462.0,
+                    620.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    553.0,
+                    600.0,
+                    572.0,
+                    600.0,
+                    572.0,
+                    620.0,
+                    553.0,
+                    620.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    599.0,
+                    597.0,
+                    631.0,
+                    597.0,
+                    631.0,
+                    620.0,
+                    599.0,
+                    620.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    640.0,
+                    595.0,
+                    673.0,
+                    595.0,
+                    673.0,
+                    623.0,
+                    640.0,
+                    623.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    735.0,
+                    600.0,
+                    767.0,
+                    600.0,
+                    767.0,
+                    621.0,
+                    735.0,
+                    621.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    772.0,
+                    595.0,
+                    811.0,
+                    595.0,
+                    811.0,
+                    626.0,
+                    772.0,
+                    626.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    883.0,
+                    584.0,
+                    1348.0,
+                    584.0,
+                    1348.0,
+                    637.0,
+                    883.0,
+                    637.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    413.0,
+                    645.0,
+                    446.0,
+                    645.0,
+                    446.0,
+                    674.0,
+                    413.0,
+                    674.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    456.0,
+                    647.0,
+                    489.0,
+                    647.0,
+                    489.0,
+                    669.0,
+                    456.0,
+                    669.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    521.0,
+                    638.0,
+                    817.0,
+                    638.0,
+                    817.0,
+                    681.0,
+                    521.0,
+                    681.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    889.0,
+                    636.0,
+                    1348.0,
+                    636.0,
+                    1348.0,
+                    685.0,
+                    889.0,
+                    685.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    530.0,
+                    683.0,
+                    821.0,
+                    683.0,
+                    821.0,
+                    737.0,
+                    530.0,
+                    737.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    885.0,
+                    684.0,
+                    1351.0,
+                    684.0,
+                    1351.0,
+                    737.0,
+                    885.0,
+                    737.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    504.0,
+                    497.0,
+                    549.0,
+                    497.0,
+                    549.0,
+                    523.5,
+                    504.0,
+                    523.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1332.0,
+                    1403.0,
+                    1332.0,
+                    1403.0,
+                    1371.0,
+                    293.0,
+                    1371.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1362.0,
+                    1403.0,
+                    1362.0,
+                    1403.0,
+                    1401.0,
+                    293.0,
+                    1401.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1395.0,
+                    1405.0,
+                    1395.0,
+                    1405.0,
+                    1432.0,
+                    294.0,
+                    1432.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1422.0,
+                    1403.0,
+                    1422.0,
+                    1403.0,
+                    1461.0,
+                    293.0,
+                    1461.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1457.0,
+                    1405.0,
+                    1457.0,
+                    1405.0,
+                    1490.0,
+                    295.0,
+                    1490.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1486.0,
+                    1405.0,
+                    1486.0,
+                    1405.0,
+                    1522.0,
+                    295.0,
+                    1522.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1516.0,
+                    1405.0,
+                    1516.0,
+                    1405.0,
+                    1553.0,
+                    293.0,
+                    1553.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1547.0,
+                    1405.0,
+                    1547.0,
+                    1405.0,
+                    1582.0,
+                    293.0,
+                    1582.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1578.0,
+                    1406.0,
+                    1578.0,
+                    1406.0,
+                    1613.0,
+                    294.0,
+                    1613.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1608.0,
+                    1406.0,
+                    1608.0,
+                    1406.0,
+                    1642.0,
+                    294.0,
+                    1642.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1640.0,
+                    1405.0,
+                    1640.0,
+                    1405.0,
+                    1672.0,
+                    296.0,
+                    1672.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1668.0,
+                    1405.0,
+                    1668.0,
+                    1405.0,
+                    1705.0,
+                    294.0,
+                    1705.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1699.0,
+                    819.0,
+                    1699.0,
+                    819.0,
+                    1736.0,
+                    294.0,
+                    1736.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    358.0,
+                    968.0,
+                    380.0,
+                    968.0,
+                    380.0,
+                    993.0,
+                    358.0,
+                    993.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    397.0,
+                    963.0,
+                    442.0,
+                    963.0,
+                    442.0,
+                    987.0,
+                    397.0,
+                    987.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    514.0,
+                    968.0,
+                    541.0,
+                    968.0,
+                    541.0,
+                    990.0,
+                    514.0,
+                    990.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    672.0,
+                    970.0,
+                    814.0,
+                    970.0,
+                    814.0,
+                    988.0,
+                    672.0,
+                    988.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    882.0,
+                    972.0,
+                    904.0,
+                    972.0,
+                    904.0,
+                    997.0,
+                    882.0,
+                    997.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    460.0,
+                    982.0,
+                    610.0,
+                    982.0,
+                    610.0,
+                    1010.0,
+                    460.0,
+                    1010.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    570.0,
+                    1017.0,
+                    579.0,
+                    1017.0,
+                    579.0,
+                    1026.0,
+                    570.0,
+                    1026.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    906.0,
+                    1031.0,
+                    945.0,
+                    1031.0,
+                    945.0,
+                    1056.0,
+                    906.0,
+                    1056.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1190.0,
+                    1026.0,
+                    1272.0,
+                    1026.0,
+                    1272.0,
+                    1048.0,
+                    1190.0,
+                    1048.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    876.0,
+                    1042.0,
+                    902.0,
+                    1042.0,
+                    902.0,
+                    1145.0,
+                    876.0,
+                    1145.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1190.0,
+                    1046.0,
+                    1296.0,
+                    1046.0,
+                    1296.0,
+                    1068.0,
+                    1190.0,
+                    1068.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    368.0,
+                    1056.0,
+                    443.0,
+                    1056.0,
+                    443.0,
+                    1167.0,
+                    368.0,
+                    1167.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    942.0,
+                    1070.0,
+                    1308.0,
+                    1070.0,
+                    1308.0,
+                    1108.0,
+                    942.0,
+                    1108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    906.0,
+                    1116.0,
+                    943.0,
+                    1116.0,
+                    943.0,
+                    1140.0,
+                    906.0,
+                    1140.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    392.0,
+                    1149.0,
+                    443.0,
+                    1149.0,
+                    443.0,
+                    1174.0,
+                    392.0,
+                    1174.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    486.0,
+                    1146.0,
+                    616.0,
+                    1146.0,
+                    616.0,
+                    1181.0,
+                    486.0,
+                    1181.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    485.0,
+                    1171.0,
+                    595.0,
+                    1171.0,
+                    595.0,
+                    1196.0,
+                    485.0,
+                    1196.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    486.0,
+                    1190.0,
+                    594.0,
+                    1190.0,
+                    594.0,
+                    1208.0,
+                    486.0,
+                    1208.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    486.0,
+                    1207.0,
+                    623.0,
+                    1207.0,
+                    623.0,
+                    1226.0,
+                    486.0,
+                    1226.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    917.0,
+                    1203.0,
+                    936.0,
+                    1203.0,
+                    936.0,
+                    1222.0,
+                    917.0,
+                    1222.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    391.0,
+                    1241.0,
+                    445.0,
+                    1241.0,
+                    445.0,
+                    1268.0,
+                    391.0,
+                    1268.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    469.0,
+                    1259.0,
+                    503.0,
+                    1259.0,
+                    503.0,
+                    1284.0,
+                    469.0,
+                    1284.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    542.0,
+                    1259.0,
+                    578.0,
+                    1259.0,
+                    578.0,
+                    1284.0,
+                    542.0,
+                    1284.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    624.0,
+                    1261.0,
+                    643.0,
+                    1261.0,
+                    643.0,
+                    1282.0,
+                    624.0,
+                    1282.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    694.0,
+                    1259.0,
+                    725.0,
+                    1259.0,
+                    725.0,
+                    1284.0,
+                    694.0,
+                    1284.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    772.0,
+                    1260.0,
+                    800.0,
+                    1260.0,
+                    800.0,
+                    1283.0,
+                    772.0,
+                    1283.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    932.0,
+                    1264.0,
+                    949.0,
+                    1264.0,
+                    949.0,
+                    1283.0,
+                    932.0,
+                    1283.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1007.0,
+                    1263.0,
+                    1026.0,
+                    1263.0,
+                    1026.0,
+                    1284.0,
+                    1007.0,
+                    1284.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1082.0,
+                    1264.0,
+                    1100.0,
+                    1264.0,
+                    1100.0,
+                    1283.0,
+                    1082.0,
+                    1283.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1158.0,
+                    1263.0,
+                    1176.0,
+                    1263.0,
+                    1176.0,
+                    1284.0,
+                    1158.0,
+                    1284.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1232.0,
+                    1263.0,
+                    1251.0,
+                    1263.0,
+                    1251.0,
+                    1284.0,
+                    1232.0,
+                    1284.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1303.0,
+                    1259.0,
+                    1334.0,
+                    1259.0,
+                    1334.0,
+                    1286.0,
+                    1303.0,
+                    1286.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    471.0,
+                    1283.0,
+                    783.0,
+                    1283.0,
+                    783.0,
+                    1305.0,
+                    471.0,
+                    1305.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1091.0,
+                    1283.0,
+                    1168.0,
+                    1283.0,
+                    1168.0,
+                    1306.0,
+                    1091.0,
+                    1306.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    749.0,
+                    1007.0,
+                    800.0,
+                    1007.0,
+                    800.0,
+                    1027.0,
+                    749.0,
+                    1027.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    418.25,
+                    1045.5,
+                    578.25,
+                    1045.5,
+                    578.25,
+                    1096.5,
+                    418.25,
+                    1096.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1282.75,
+                    1278.5,
+                    1343.75,
+                    1278.5,
+                    1343.75,
+                    1315.5,
+                    1282.75,
+                    1315.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    763.0,
+                    1405.0,
+                    763.0,
+                    1405.0,
+                    796.0,
+                    297.0,
+                    796.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    795.0,
+                    1404.0,
+                    795.0,
+                    1404.0,
+                    828.0,
+                    296.0,
+                    828.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    819.0,
+                    1407.0,
+                    819.0,
+                    1407.0,
+                    864.0,
+                    292.0,
+                    864.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    853.0,
+                    1405.0,
+                    853.0,
+                    1405.0,
+                    890.0,
+                    296.0,
+                    890.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    885.0,
+                    624.0,
+                    885.0,
+                    624.0,
+                    918.0,
+                    297.0,
+                    918.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    73.0,
+                    818.0,
+                    73.0,
+                    818.0,
+                    108.0,
+                    295.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    839.0,
+                    2086.0,
+                    860.0,
+                    2086.0,
+                    860.0,
+                    2116.0,
+                    839.0,
+                    2116.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    1805.0,
+                    1404.0,
+                    1805.0,
+                    1404.0,
+                    1838.0,
+                    297.0,
+                    1838.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    1835.0,
+                    1404.0,
+                    1835.0,
+                    1404.0,
+                    1868.0,
+                    297.0,
+                    1868.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1864.0,
+                    1405.0,
+                    1864.0,
+                    1405.0,
+                    1901.0,
+                    295.0,
+                    1901.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1892.0,
+                    1405.0,
+                    1892.0,
+                    1405.0,
+                    1930.0,
+                    293.0,
+                    1930.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1924.0,
+                    1136.0,
+                    1924.0,
+                    1136.0,
+                    1963.0,
+                    293.0,
+                    1963.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1968.0,
+                    1404.0,
+                    1968.0,
+                    1404.0,
+                    2009.0,
+                    294.0,
+                    2009.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    2002.0,
+                    1406.0,
+                    2002.0,
+                    1406.0,
+                    2040.0,
+                    293.0,
+                    2040.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 7,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1165,
+                    1404,
+                    1165,
+                    1404,
+                    1772,
+                    298,
+                    1772
+                ],
+                "score": 0.983
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    643,
+                    1404,
+                    643,
+                    1404,
+                    919,
+                    298,
+                    919
+                ],
+                "score": 0.982
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    230,
+                    1403,
+                    230,
+                    1403,
+                    506,
+                    298,
+                    506
+                ],
+                "score": 0.981
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    934,
+                    1403,
+                    934,
+                    1403,
+                    1148,
+                    298,
+                    1148
+                ],
+                "score": 0.979
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1789,
+                    1402,
+                    1789,
+                    1402,
+                    2034,
+                    298,
+                    2034
+                ],
+                "score": 0.978
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    300,
+                    76,
+                    815,
+                    76,
+                    815,
+                    104,
+                    300,
+                    104
+                ],
+                "score": 0.891
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    299,
+                    565,
+                    529,
+                    565,
+                    529,
+                    600,
+                    299,
+                    600
+                ],
+                "score": 0.875
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    840,
+                    2088,
+                    858,
+                    2088,
+                    858,
+                    2111,
+                    840,
+                    2111
+                ],
+                "score": 0.788
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    73.0,
+                    817.0,
+                    73.0,
+                    817.0,
+                    108.0,
+                    296.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    561.0,
+                    536.0,
+                    561.0,
+                    536.0,
+                    608.0,
+                    292.0,
+                    608.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    839.0,
+                    2087.0,
+                    860.0,
+                    2087.0,
+                    860.0,
+                    2117.0,
+                    839.0,
+                    2117.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1163.0,
+                    1405.0,
+                    1163.0,
+                    1405.0,
+                    1200.0,
+                    295.0,
+                    1200.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1196.0,
+                    1404.0,
+                    1196.0,
+                    1404.0,
+                    1228.0,
+                    292.0,
+                    1228.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1225.0,
+                    1405.0,
+                    1225.0,
+                    1405.0,
+                    1260.0,
+                    293.0,
+                    1260.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1254.0,
+                    1408.0,
+                    1254.0,
+                    1408.0,
+                    1292.0,
+                    294.0,
+                    1292.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1284.0,
+                    1406.0,
+                    1284.0,
+                    1406.0,
+                    1324.0,
+                    292.0,
+                    1324.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1318.0,
+                    1405.0,
+                    1318.0,
+                    1405.0,
+                    1352.0,
+                    294.0,
+                    1352.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1346.0,
+                    1405.0,
+                    1346.0,
+                    1405.0,
+                    1386.0,
+                    291.0,
+                    1386.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1379.0,
+                    1406.0,
+                    1379.0,
+                    1406.0,
+                    1413.0,
+                    296.0,
+                    1413.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1408.0,
+                    1404.0,
+                    1408.0,
+                    1404.0,
+                    1443.0,
+                    296.0,
+                    1443.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1438.0,
+                    1406.0,
+                    1438.0,
+                    1406.0,
+                    1473.0,
+                    293.0,
+                    1473.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1467.0,
+                    1405.0,
+                    1467.0,
+                    1405.0,
+                    1505.0,
+                    293.0,
+                    1505.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1499.0,
+                    1405.0,
+                    1499.0,
+                    1405.0,
+                    1534.0,
+                    294.0,
+                    1534.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1527.0,
+                    1404.0,
+                    1527.0,
+                    1404.0,
+                    1565.0,
+                    293.0,
+                    1565.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1556.0,
+                    1406.0,
+                    1556.0,
+                    1406.0,
+                    1597.0,
+                    292.0,
+                    1597.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1590.0,
+                    1403.0,
+                    1590.0,
+                    1403.0,
+                    1626.0,
+                    292.0,
+                    1626.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1619.0,
+                    1405.0,
+                    1619.0,
+                    1405.0,
+                    1655.0,
+                    295.0,
+                    1655.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1651.0,
+                    1405.0,
+                    1651.0,
+                    1405.0,
+                    1686.0,
+                    294.0,
+                    1686.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1682.0,
+                    1405.0,
+                    1682.0,
+                    1405.0,
+                    1716.0,
+                    294.0,
+                    1716.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1712.0,
+                    1405.0,
+                    1712.0,
+                    1405.0,
+                    1747.0,
+                    295.0,
+                    1747.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1745.0,
+                    445.0,
+                    1745.0,
+                    445.0,
+                    1773.0,
+                    293.0,
+                    1773.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    645.0,
+                    1405.0,
+                    645.0,
+                    1405.0,
+                    678.0,
+                    296.0,
+                    678.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    674.0,
+                    1404.0,
+                    674.0,
+                    1404.0,
+                    711.0,
+                    294.0,
+                    711.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    704.0,
+                    1405.0,
+                    704.0,
+                    1405.0,
+                    741.0,
+                    294.0,
+                    741.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    735.0,
+                    1404.0,
+                    735.0,
+                    1404.0,
+                    772.0,
+                    294.0,
+                    772.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    761.0,
+                    1406.0,
+                    761.0,
+                    1406.0,
+                    803.0,
+                    292.0,
+                    803.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    794.0,
+                    1405.0,
+                    794.0,
+                    1405.0,
+                    833.0,
+                    293.0,
+                    833.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    827.0,
+                    1405.0,
+                    827.0,
+                    1405.0,
+                    863.0,
+                    294.0,
+                    863.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    857.0,
+                    1405.0,
+                    857.0,
+                    1405.0,
+                    892.0,
+                    293.0,
+                    892.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    888.0,
+                    462.0,
+                    888.0,
+                    462.0,
+                    922.0,
+                    297.0,
+                    922.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    230.0,
+                    1403.0,
+                    230.0,
+                    1403.0,
+                    262.0,
+                    296.0,
+                    262.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    262.0,
+                    1404.0,
+                    262.0,
+                    1404.0,
+                    295.0,
+                    296.0,
+                    295.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    291.0,
+                    1405.0,
+                    291.0,
+                    1405.0,
+                    328.0,
+                    294.0,
+                    328.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    320.0,
+                    1407.0,
+                    320.0,
+                    1407.0,
+                    359.0,
+                    292.0,
+                    359.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    354.0,
+                    1405.0,
+                    354.0,
+                    1405.0,
+                    386.0,
+                    296.0,
+                    386.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    381.0,
+                    1407.0,
+                    381.0,
+                    1407.0,
+                    419.0,
+                    293.0,
+                    419.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    411.0,
+                    1403.0,
+                    411.0,
+                    1403.0,
+                    450.0,
+                    293.0,
+                    450.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    444.0,
+                    1403.0,
+                    444.0,
+                    1403.0,
+                    479.0,
+                    292.0,
+                    479.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    476.0,
+                    967.0,
+                    476.0,
+                    967.0,
+                    509.0,
+                    296.0,
+                    509.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    934.0,
+                    1405.0,
+                    934.0,
+                    1405.0,
+                    969.0,
+                    294.0,
+                    969.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    964.0,
+                    1405.0,
+                    964.0,
+                    1405.0,
+                    1001.0,
+                    293.0,
+                    1001.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    993.0,
+                    1404.0,
+                    993.0,
+                    1404.0,
+                    1030.0,
+                    293.0,
+                    1030.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1026.0,
+                    1405.0,
+                    1026.0,
+                    1405.0,
+                    1059.0,
+                    294.0,
+                    1059.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1057.0,
+                    1404.0,
+                    1057.0,
+                    1404.0,
+                    1090.0,
+                    293.0,
+                    1090.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1089.0,
+                    1404.0,
+                    1089.0,
+                    1404.0,
+                    1119.0,
+                    296.0,
+                    1119.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1118.0,
+                    956.0,
+                    1118.0,
+                    956.0,
+                    1153.0,
+                    296.0,
+                    1153.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1788.0,
+                    1406.0,
+                    1788.0,
+                    1406.0,
+                    1824.0,
+                    294.0,
+                    1824.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1819.0,
+                    1407.0,
+                    1819.0,
+                    1407.0,
+                    1855.0,
+                    292.0,
+                    1855.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1849.0,
+                    1404.0,
+                    1849.0,
+                    1404.0,
+                    1885.0,
+                    292.0,
+                    1885.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1880.0,
+                    1406.0,
+                    1880.0,
+                    1406.0,
+                    1915.0,
+                    293.0,
+                    1915.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1913.0,
+                    1406.0,
+                    1913.0,
+                    1406.0,
+                    1946.0,
+                    294.0,
+                    1946.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1939.0,
+                    1406.0,
+                    1939.0,
+                    1406.0,
+                    1980.0,
+                    292.0,
+                    1980.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1973.0,
+                    1406.0,
+                    1973.0,
+                    1406.0,
+                    2006.0,
+                    294.0,
+                    2006.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    2003.0,
+                    1404.0,
+                    2003.0,
+                    1404.0,
+                    2036.0,
+                    294.0,
+                    2036.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 8,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    309,
+                    1404,
+                    309,
+                    1404,
+                    827,
+                    298,
+                    827
+                ],
+                "score": 0.985
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    295,
+                    229,
+                    1400,
+                    229,
+                    1400,
+                    293,
+                    295,
+                    293
+                ],
+                "score": 0.944
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    301,
+                    941,
+                    1399,
+                    941,
+                    1399,
+                    1034,
+                    301,
+                    1034
+                ],
+                "score": 0.942
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    300,
+                    76,
+                    815,
+                    76,
+                    815,
+                    104,
+                    300,
+                    104
+                ],
+                "score": 0.893
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    301,
+                    873,
+                    678,
+                    873,
+                    678,
+                    906,
+                    301,
+                    906
+                ],
+                "score": 0.881
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    837,
+                    2088,
+                    865,
+                    2088,
+                    865,
+                    2112,
+                    837,
+                    2112
+                ],
+                "score": 0.831
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1971,
+                    1398,
+                    1971,
+                    1398,
+                    2033,
+                    297,
+                    2033
+                ],
+                "score": 0.786
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    300,
+                    1080,
+                    488,
+                    1080,
+                    488,
+                    1113,
+                    300,
+                    1113
+                ],
+                "score": 0.77
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1131,
+                    1402,
+                    1131,
+                    1402,
+                    1194,
+                    298,
+                    1194
+                ],
+                "score": 0.696
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    294,
+                    1214,
+                    1403,
+                    1214,
+                    1403,
+                    1278,
+                    294,
+                    1278
+                ],
+                "score": 0.679
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    296,
+                    1691,
+                    1077,
+                    1691,
+                    1077,
+                    1725,
+                    296,
+                    1725
+                ],
+                "score": 0.658
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    301,
+                    1857,
+                    1399,
+                    1857,
+                    1399,
+                    1951,
+                    301,
+                    1951
+                ],
+                "score": 0.646
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    304,
+                    1576,
+                    1400,
+                    1576,
+                    1400,
+                    1672,
+                    304,
+                    1672
+                ],
+                "score": 0.622
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    301,
+                    1296,
+                    1402,
+                    1296,
+                    1402,
+                    1389,
+                    301,
+                    1389
+                ],
+                "score": 0.618
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1494,
+                    1402,
+                    1494,
+                    1402,
+                    1559,
+                    298,
+                    1559
+                ],
+                "score": 0.611
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    304,
+                    1744,
+                    1402,
+                    1744,
+                    1402,
+                    1838,
+                    304,
+                    1838
+                ],
+                "score": 0.595
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    296,
+                    1410,
+                    1403,
+                    1410,
+                    1403,
+                    1475,
+                    296,
+                    1475
+                ],
+                "score": 0.595
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    300,
+                    1080,
+                    488,
+                    1080,
+                    488,
+                    1113,
+                    300,
+                    1113
+                ],
+                "score": 0.098
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    73.0,
+                    817.0,
+                    73.0,
+                    817.0,
+                    108.0,
+                    296.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    870.0,
+                    682.0,
+                    870.0,
+                    682.0,
+                    912.0,
+                    293.0,
+                    912.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    831.0,
+                    2084.0,
+                    871.0,
+                    2084.0,
+                    871.0,
+                    2125.0,
+                    831.0,
+                    2125.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1079.0,
+                    491.0,
+                    1079.0,
+                    491.0,
+                    1117.0,
+                    295.0,
+                    1117.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    308.0,
+                    1405.0,
+                    308.0,
+                    1405.0,
+                    344.0,
+                    296.0,
+                    344.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    340.0,
+                    1404.0,
+                    340.0,
+                    1404.0,
+                    372.0,
+                    296.0,
+                    372.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    368.0,
+                    1408.0,
+                    368.0,
+                    1408.0,
+                    404.0,
+                    294.0,
+                    404.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    399.0,
+                    1404.0,
+                    399.0,
+                    1404.0,
+                    435.0,
+                    294.0,
+                    435.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    429.0,
+                    1406.0,
+                    429.0,
+                    1406.0,
+                    465.0,
+                    294.0,
+                    465.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    460.0,
+                    1406.0,
+                    460.0,
+                    1406.0,
+                    496.0,
+                    294.0,
+                    496.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    491.0,
+                    1406.0,
+                    491.0,
+                    1406.0,
+                    527.0,
+                    293.0,
+                    527.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    522.0,
+                    1405.0,
+                    522.0,
+                    1405.0,
+                    558.0,
+                    294.0,
+                    558.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    551.0,
+                    1406.0,
+                    551.0,
+                    1406.0,
+                    589.0,
+                    292.0,
+                    589.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    582.0,
+                    1404.0,
+                    582.0,
+                    1404.0,
+                    618.0,
+                    294.0,
+                    618.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    613.0,
+                    1406.0,
+                    613.0,
+                    1406.0,
+                    649.0,
+                    294.0,
+                    649.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    641.0,
+                    1404.0,
+                    641.0,
+                    1404.0,
+                    677.0,
+                    294.0,
+                    677.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    674.0,
+                    1406.0,
+                    674.0,
+                    1406.0,
+                    710.0,
+                    294.0,
+                    710.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    703.0,
+                    1406.0,
+                    703.0,
+                    1406.0,
+                    739.0,
+                    294.0,
+                    739.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    734.0,
+                    1405.0,
+                    734.0,
+                    1405.0,
+                    770.0,
+                    293.0,
+                    770.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    766.0,
+                    1405.0,
+                    766.0,
+                    1405.0,
+                    798.0,
+                    294.0,
+                    798.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    797.0,
+                    615.0,
+                    797.0,
+                    615.0,
+                    829.0,
+                    296.0,
+                    829.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    225.0,
+                    1404.0,
+                    225.0,
+                    1404.0,
+                    267.0,
+                    293.0,
+                    267.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    261.0,
+                    786.0,
+                    261.0,
+                    786.0,
+                    296.0,
+                    295.0,
+                    296.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    940.0,
+                    1403.0,
+                    940.0,
+                    1403.0,
+                    974.0,
+                    297.0,
+                    974.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    971.0,
+                    1404.0,
+                    971.0,
+                    1404.0,
+                    1008.0,
+                    295.0,
+                    1008.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    998.0,
+                    1183.0,
+                    998.0,
+                    1183.0,
+                    1039.0,
+                    294.0,
+                    1039.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1969.0,
+                    1400.0,
+                    1969.0,
+                    1400.0,
+                    2007.0,
+                    295.0,
+                    2007.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    1999.0,
+                    452.0,
+                    1999.0,
+                    452.0,
+                    2036.0,
+                    323.0,
+                    2036.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1131.0,
+                    1404.0,
+                    1131.0,
+                    1404.0,
+                    1166.0,
+                    296.0,
+                    1166.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    1161.0,
+                    835.0,
+                    1161.0,
+                    835.0,
+                    1196.0,
+                    322.0,
+                    1196.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1209.0,
+                    1404.0,
+                    1209.0,
+                    1404.0,
+                    1255.0,
+                    293.0,
+                    1255.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    1247.0,
+                    730.0,
+                    1247.0,
+                    730.0,
+                    1277.0,
+                    322.0,
+                    1277.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    1689.0,
+                    1079.0,
+                    1689.0,
+                    1079.0,
+                    1728.0,
+                    297.0,
+                    1728.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1854.0,
+                    1403.0,
+                    1854.0,
+                    1403.0,
+                    1894.0,
+                    295.0,
+                    1894.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    320.0,
+                    1886.0,
+                    1405.0,
+                    1886.0,
+                    1405.0,
+                    1926.0,
+                    320.0,
+                    1926.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    321.0,
+                    1917.0,
+                    396.0,
+                    1917.0,
+                    396.0,
+                    1951.0,
+                    321.0,
+                    1951.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    298.0,
+                    1576.0,
+                    1405.0,
+                    1576.0,
+                    1405.0,
+                    1614.0,
+                    298.0,
+                    1614.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    321.0,
+                    1606.0,
+                    1404.0,
+                    1606.0,
+                    1404.0,
+                    1646.0,
+                    321.0,
+                    1646.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    320.0,
+                    1639.0,
+                    1214.0,
+                    1639.0,
+                    1214.0,
+                    1674.0,
+                    320.0,
+                    1674.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1297.0,
+                    1403.0,
+                    1297.0,
+                    1403.0,
+                    1333.0,
+                    296.0,
+                    1333.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    320.0,
+                    1328.0,
+                    1406.0,
+                    1328.0,
+                    1406.0,
+                    1364.0,
+                    320.0,
+                    1364.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    320.0,
+                    1356.0,
+                    400.0,
+                    1356.0,
+                    400.0,
+                    1392.0,
+                    320.0,
+                    1392.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    1494.0,
+                    1407.0,
+                    1494.0,
+                    1407.0,
+                    1530.0,
+                    297.0,
+                    1530.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    1525.0,
+                    1019.0,
+                    1525.0,
+                    1019.0,
+                    1559.0,
+                    322.0,
+                    1559.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    298.0,
+                    1744.0,
+                    1402.0,
+                    1744.0,
+                    1402.0,
+                    1780.0,
+                    298.0,
+                    1780.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    1773.0,
+                    1402.0,
+                    1773.0,
+                    1402.0,
+                    1812.0,
+                    323.0,
+                    1812.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    324.0,
+                    1807.0,
+                    832.0,
+                    1807.0,
+                    832.0,
+                    1837.0,
+                    324.0,
+                    1837.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1412.0,
+                    1404.0,
+                    1412.0,
+                    1404.0,
+                    1448.0,
+                    296.0,
+                    1448.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    1441.0,
+                    817.0,
+                    1441.0,
+                    817.0,
+                    1477.0,
+                    323.0,
+                    1477.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1079.0,
+                    491.0,
+                    1079.0,
+                    491.0,
+                    1117.0,
+                    295.0,
+                    1117.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 9,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 2,
+                "poly": [
+                    299,
+                    75,
+                    816,
+                    75,
+                    816,
+                    105,
+                    299,
+                    105
+                ],
+                "score": 0.888
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1012,
+                    1400,
+                    1012,
+                    1400,
+                    1078,
+                    298,
+                    1078
+                ],
+                "score": 0.829
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    836,
+                    2088,
+                    863,
+                    2088,
+                    863,
+                    2113,
+                    836,
+                    2113
+                ],
+                "score": 0.815
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    929,
+                    1400,
+                    929,
+                    1400,
+                    993,
+                    299,
+                    993
+                ],
+                "score": 0.79
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    301,
+                    844,
+                    1399,
+                    844,
+                    1399,
+                    910,
+                    301,
+                    910
+                ],
+                "score": 0.773
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    296,
+                    1407,
+                    1402,
+                    1407,
+                    1402,
+                    1473,
+                    296,
+                    1473
+                ],
+                "score": 0.772
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    296,
+                    1323,
+                    1401,
+                    1323,
+                    1401,
+                    1390,
+                    296,
+                    1390
+                ],
+                "score": 0.746
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    293,
+                    1492,
+                    1400,
+                    1492,
+                    1400,
+                    1557,
+                    293,
+                    1557
+                ],
+                "score": 0.741
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    300,
+                    761,
+                    1397,
+                    761,
+                    1397,
+                    826,
+                    300,
+                    826
+                ],
+                "score": 0.734
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    301,
+                    1574,
+                    1403,
+                    1574,
+                    1403,
+                    1669,
+                    301,
+                    1669
+                ],
+                "score": 0.722
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    677,
+                    1395,
+                    677,
+                    1395,
+                    742,
+                    299,
+                    742
+                ],
+                "score": 0.711
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1241,
+                    1400,
+                    1241,
+                    1400,
+                    1306,
+                    297,
+                    1306
+                ],
+                "score": 0.702
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    296,
+                    1690,
+                    1402,
+                    1690,
+                    1402,
+                    1754,
+                    296,
+                    1754
+                ],
+                "score": 0.692
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    593,
+                    1397,
+                    593,
+                    1397,
+                    659,
+                    298,
+                    659
+                ],
+                "score": 0.679
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    296,
+                    1095,
+                    1405,
+                    1095,
+                    1405,
+                    1221,
+                    296,
+                    1221
+                ],
+                "score": 0.667
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    510,
+                    1399,
+                    510,
+                    1399,
+                    575,
+                    297,
+                    575
+                ],
+                "score": 0.64
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    293,
+                    228,
+                    1401,
+                    228,
+                    1401,
+                    294,
+                    293,
+                    294
+                ],
+                "score": 0.627
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    294,
+                    1772,
+                    1402,
+                    1772,
+                    1402,
+                    1839,
+                    294,
+                    1839
+                ],
+                "score": 0.62
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    300,
+                    1971,
+                    1399,
+                    1971,
+                    1399,
+                    2035,
+                    300,
+                    2035
+                ],
+                "score": 0.607
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    302,
+                    395,
+                    1400,
+                    395,
+                    1400,
+                    491,
+                    302,
+                    491
+                ],
+                "score": 0.607
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    293,
+                    312,
+                    1401,
+                    312,
+                    1401,
+                    377,
+                    293,
+                    377
+                ],
+                "score": 0.594
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    301,
+                    1857,
+                    1398,
+                    1857,
+                    1398,
+                    1952,
+                    301,
+                    1952
+                ],
+                "score": 0.56
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    73.0,
+                    817.0,
+                    73.0,
+                    817.0,
+                    108.0,
+                    295.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    832.0,
+                    2085.0,
+                    868.0,
+                    2085.0,
+                    868.0,
+                    2125.0,
+                    832.0,
+                    2125.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1013.0,
+                    1402.0,
+                    1013.0,
+                    1402.0,
+                    1049.0,
+                    294.0,
+                    1049.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    1044.0,
+                    1256.0,
+                    1044.0,
+                    1256.0,
+                    1080.0,
+                    323.0,
+                    1080.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    926.0,
+                    1404.0,
+                    926.0,
+                    1404.0,
+                    967.0,
+                    293.0,
+                    967.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    324.0,
+                    960.0,
+                    533.0,
+                    960.0,
+                    533.0,
+                    992.0,
+                    324.0,
+                    992.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    845.0,
+                    1403.0,
+                    845.0,
+                    1403.0,
+                    881.0,
+                    295.0,
+                    881.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    876.0,
+                    1332.0,
+                    876.0,
+                    1332.0,
+                    912.0,
+                    322.0,
+                    912.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1406.0,
+                    1404.0,
+                    1406.0,
+                    1404.0,
+                    1445.0,
+                    294.0,
+                    1445.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    324.0,
+                    1441.0,
+                    922.0,
+                    1441.0,
+                    922.0,
+                    1474.0,
+                    324.0,
+                    1474.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1321.0,
+                    1405.0,
+                    1321.0,
+                    1405.0,
+                    1362.0,
+                    295.0,
+                    1362.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    321.0,
+                    1357.0,
+                    958.0,
+                    1357.0,
+                    958.0,
+                    1390.0,
+                    321.0,
+                    1390.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1492.0,
+                    1405.0,
+                    1492.0,
+                    1405.0,
+                    1528.0,
+                    295.0,
+                    1528.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    1522.0,
+                    1239.0,
+                    1522.0,
+                    1239.0,
+                    1558.0,
+                    322.0,
+                    1558.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    761.0,
+                    1402.0,
+                    761.0,
+                    1402.0,
+                    799.0,
+                    295.0,
+                    799.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    321.0,
+                    794.0,
+                    1043.0,
+                    794.0,
+                    1043.0,
+                    827.0,
+                    321.0,
+                    827.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1571.0,
+                    1403.0,
+                    1571.0,
+                    1403.0,
+                    1613.0,
+                    295.0,
+                    1613.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    321.0,
+                    1606.0,
+                    1404.0,
+                    1606.0,
+                    1404.0,
+                    1642.0,
+                    321.0,
+                    1642.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    319.0,
+                    1634.0,
+                    399.0,
+                    1634.0,
+                    399.0,
+                    1670.0,
+                    319.0,
+                    1670.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    679.0,
+                    1400.0,
+                    679.0,
+                    1400.0,
+                    715.0,
+                    296.0,
+                    715.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    710.0,
+                    1224.0,
+                    710.0,
+                    1224.0,
+                    743.0,
+                    322.0,
+                    743.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1240.0,
+                    1402.0,
+                    1240.0,
+                    1402.0,
+                    1276.0,
+                    295.0,
+                    1276.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    1271.0,
+                    1254.0,
+                    1271.0,
+                    1254.0,
+                    1306.0,
+                    322.0,
+                    1306.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1689.0,
+                    1406.0,
+                    1689.0,
+                    1406.0,
+                    1728.0,
+                    294.0,
+                    1728.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    324.0,
+                    1722.0,
+                    1004.0,
+                    1722.0,
+                    1004.0,
+                    1754.0,
+                    324.0,
+                    1754.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    587.0,
+                    1402.0,
+                    587.0,
+                    1402.0,
+                    636.0,
+                    291.0,
+                    636.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    626.0,
+                    676.0,
+                    626.0,
+                    676.0,
+                    660.0,
+                    322.0,
+                    660.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1097.0,
+                    1403.0,
+                    1097.0,
+                    1403.0,
+                    1130.0,
+                    294.0,
+                    1130.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    324.0,
+                    1129.0,
+                    1403.0,
+                    1129.0,
+                    1403.0,
+                    1162.0,
+                    324.0,
+                    1162.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    320.0,
+                    1157.0,
+                    1406.0,
+                    1157.0,
+                    1406.0,
+                    1193.0,
+                    320.0,
+                    1193.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    1189.0,
+                    461.0,
+                    1189.0,
+                    461.0,
+                    1219.0,
+                    323.0,
+                    1219.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    513.0,
+                    1404.0,
+                    513.0,
+                    1404.0,
+                    545.0,
+                    297.0,
+                    545.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    321.0,
+                    543.0,
+                    1302.0,
+                    543.0,
+                    1302.0,
+                    576.0,
+                    321.0,
+                    576.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    229.0,
+                    1404.0,
+                    229.0,
+                    1404.0,
+                    265.0,
+                    295.0,
+                    265.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    261.0,
+                    1192.0,
+                    261.0,
+                    1192.0,
+                    295.0,
+                    322.0,
+                    295.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1773.0,
+                    1405.0,
+                    1773.0,
+                    1405.0,
+                    1809.0,
+                    294.0,
+                    1809.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    1806.0,
+                    1006.0,
+                    1806.0,
+                    1006.0,
+                    1838.0,
+                    322.0,
+                    1838.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    1971.0,
+                    1404.0,
+                    1971.0,
+                    1404.0,
+                    2007.0,
+                    297.0,
+                    2007.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    321.0,
+                    2002.0,
+                    1159.0,
+                    2002.0,
+                    1159.0,
+                    2036.0,
+                    321.0,
+                    2036.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    393.0,
+                    1403.0,
+                    393.0,
+                    1403.0,
+                    433.0,
+                    297.0,
+                    433.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    428.0,
+                    1404.0,
+                    428.0,
+                    1404.0,
+                    463.0,
+                    323.0,
+                    463.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    324.0,
+                    459.0,
+                    819.0,
+                    459.0,
+                    819.0,
+                    493.0,
+                    324.0,
+                    493.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    311.0,
+                    1404.0,
+                    311.0,
+                    1404.0,
+                    350.0,
+                    292.0,
+                    350.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    321.0,
+                    345.0,
+                    880.0,
+                    345.0,
+                    880.0,
+                    378.0,
+                    321.0,
+                    378.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1857.0,
+                    1403.0,
+                    1857.0,
+                    1403.0,
+                    1895.0,
+                    295.0,
+                    1895.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    1889.0,
+                    1402.0,
+                    1889.0,
+                    1402.0,
+                    1923.0,
+                    323.0,
+                    1923.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    325.0,
+                    1919.0,
+                    1022.0,
+                    1919.0,
+                    1022.0,
+                    1953.0,
+                    325.0,
+                    1953.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 10,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 2,
+                "poly": [
+                    300,
+                    75,
+                    815,
+                    75,
+                    815,
+                    105,
+                    300,
+                    105
+                ],
+                "score": 0.889
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    295,
+                    228,
+                    1402,
+                    228,
+                    1402,
+                    294,
+                    295,
+                    294
+                ],
+                "score": 0.875
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    426,
+                    1403,
+                    426,
+                    1403,
+                    520,
+                    298,
+                    520
+                ],
+                "score": 0.867
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    312,
+                    1401,
+                    312,
+                    1401,
+                    406,
+                    299,
+                    406
+                ],
+                "score": 0.855
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    541,
+                    1399,
+                    541,
+                    1399,
+                    605,
+                    298,
+                    605
+                ],
+                "score": 0.842
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    707,
+                    1400,
+                    707,
+                    1400,
+                    773,
+                    297,
+                    773
+                ],
+                "score": 0.839
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    836,
+                    2088,
+                    865,
+                    2088,
+                    865,
+                    2113,
+                    836,
+                    2113
+                ],
+                "score": 0.825
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    624,
+                    1398,
+                    624,
+                    1398,
+                    689,
+                    298,
+                    689
+                ],
+                "score": 0.82
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    293,
+                    990,
+                    1401,
+                    990,
+                    1401,
+                    1054,
+                    293,
+                    1054
+                ],
+                "score": 0.8
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    294,
+                    1521,
+                    1399,
+                    1521,
+                    1399,
+                    1588,
+                    294,
+                    1588
+                ],
+                "score": 0.796
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    906,
+                    1401,
+                    906,
+                    1401,
+                    972,
+                    297,
+                    972
+                ],
+                "score": 0.792
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    292,
+                    1271,
+                    1400,
+                    1271,
+                    1400,
+                    1336,
+                    292,
+                    1336
+                ],
+                "score": 0.791
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    295,
+                    1438,
+                    1400,
+                    1438,
+                    1400,
+                    1503,
+                    295,
+                    1503
+                ],
+                "score": 0.788
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    1971,
+                    1395,
+                    1971,
+                    1395,
+                    2034,
+                    299,
+                    2034
+                ],
+                "score": 0.78
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    292,
+                    1605,
+                    1399,
+                    1605,
+                    1399,
+                    1671,
+                    292,
+                    1671
+                ],
+                "score": 0.766
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    288,
+                    1354,
+                    1400,
+                    1354,
+                    1400,
+                    1420,
+                    288,
+                    1420
+                ],
+                "score": 0.764
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    292,
+                    1187,
+                    1401,
+                    1187,
+                    1401,
+                    1252,
+                    292,
+                    1252
+                ],
+                "score": 0.749
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    1690,
+                    1401,
+                    1690,
+                    1401,
+                    1784,
+                    299,
+                    1784
+                ],
+                "score": 0.747
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    294,
+                    1804,
+                    1400,
+                    1804,
+                    1400,
+                    1868,
+                    294,
+                    1868
+                ],
+                "score": 0.731
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    791,
+                    1404,
+                    791,
+                    1404,
+                    886,
+                    297,
+                    886
+                ],
+                "score": 0.731
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1887,
+                    1398,
+                    1887,
+                    1398,
+                    1952,
+                    297,
+                    1952
+                ],
+                "score": 0.716
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1071,
+                    1403,
+                    1071,
+                    1403,
+                    1169,
+                    297,
+                    1169
+                ],
+                "score": 0.108
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    73.0,
+                    817.0,
+                    73.0,
+                    817.0,
+                    108.0,
+                    296.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    832.0,
+                    2085.0,
+                    870.0,
+                    2085.0,
+                    870.0,
+                    2124.0,
+                    832.0,
+                    2124.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    225.0,
+                    1405.0,
+                    225.0,
+                    1405.0,
+                    266.0,
+                    294.0,
+                    266.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    318.0,
+                    257.0,
+                    398.0,
+                    257.0,
+                    398.0,
+                    296.0,
+                    318.0,
+                    296.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    426.0,
+                    1404.0,
+                    426.0,
+                    1404.0,
+                    463.0,
+                    297.0,
+                    463.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    320.0,
+                    455.0,
+                    1407.0,
+                    455.0,
+                    1407.0,
+                    496.0,
+                    320.0,
+                    496.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    320.0,
+                    483.0,
+                    398.0,
+                    483.0,
+                    398.0,
+                    524.0,
+                    320.0,
+                    524.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    314.0,
+                    1403.0,
+                    314.0,
+                    1403.0,
+                    348.0,
+                    295.0,
+                    348.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    320.0,
+                    342.0,
+                    1405.0,
+                    342.0,
+                    1405.0,
+                    380.0,
+                    320.0,
+                    380.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    319.0,
+                    369.0,
+                    397.0,
+                    369.0,
+                    397.0,
+                    410.0,
+                    319.0,
+                    410.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    541.0,
+                    1404.0,
+                    541.0,
+                    1404.0,
+                    577.0,
+                    296.0,
+                    577.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    572.0,
+                    842.0,
+                    572.0,
+                    842.0,
+                    606.0,
+                    323.0,
+                    606.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    708.0,
+                    1402.0,
+                    708.0,
+                    1402.0,
+                    744.0,
+                    295.0,
+                    744.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    741.0,
+                    620.0,
+                    741.0,
+                    620.0,
+                    772.0,
+                    323.0,
+                    772.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    623.0,
+                    1403.0,
+                    623.0,
+                    1403.0,
+                    662.0,
+                    296.0,
+                    662.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    657.0,
+                    1322.0,
+                    657.0,
+                    1322.0,
+                    690.0,
+                    323.0,
+                    690.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    989.0,
+                    1406.0,
+                    989.0,
+                    1406.0,
+                    1028.0,
+                    293.0,
+                    1028.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    1022.0,
+                    730.0,
+                    1022.0,
+                    730.0,
+                    1053.0,
+                    322.0,
+                    1053.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1521.0,
+                    1401.0,
+                    1521.0,
+                    1401.0,
+                    1557.0,
+                    295.0,
+                    1557.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    1553.0,
+                    1212.0,
+                    1553.0,
+                    1212.0,
+                    1588.0,
+                    323.0,
+                    1588.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    907.0,
+                    1403.0,
+                    907.0,
+                    1403.0,
+                    943.0,
+                    297.0,
+                    943.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    938.0,
+                    1060.0,
+                    938.0,
+                    1060.0,
+                    973.0,
+                    323.0,
+                    973.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1267.0,
+                    1403.0,
+                    1267.0,
+                    1403.0,
+                    1310.0,
+                    291.0,
+                    1310.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    321.0,
+                    1302.0,
+                    532.0,
+                    1302.0,
+                    532.0,
+                    1336.0,
+                    321.0,
+                    1336.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    298.0,
+                    1441.0,
+                    1401.0,
+                    1441.0,
+                    1401.0,
+                    1473.0,
+                    298.0,
+                    1473.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    1469.0,
+                    775.0,
+                    1469.0,
+                    775.0,
+                    1504.0,
+                    323.0,
+                    1504.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1972.0,
+                    1399.0,
+                    1972.0,
+                    1399.0,
+                    2008.0,
+                    296.0,
+                    2008.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    321.0,
+                    2001.0,
+                    1370.0,
+                    2001.0,
+                    1370.0,
+                    2036.0,
+                    321.0,
+                    2036.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1603.0,
+                    1404.0,
+                    1603.0,
+                    1404.0,
+                    1645.0,
+                    295.0,
+                    1645.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    325.0,
+                    1638.0,
+                    1346.0,
+                    1638.0,
+                    1346.0,
+                    1670.0,
+                    325.0,
+                    1670.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1351.0,
+                    1403.0,
+                    1351.0,
+                    1403.0,
+                    1395.0,
+                    292.0,
+                    1395.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    321.0,
+                    1386.0,
+                    1132.0,
+                    1386.0,
+                    1132.0,
+                    1421.0,
+                    321.0,
+                    1421.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1190.0,
+                    1403.0,
+                    1190.0,
+                    1403.0,
+                    1222.0,
+                    296.0,
+                    1222.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    1221.0,
+                    1014.0,
+                    1221.0,
+                    1014.0,
+                    1250.0,
+                    323.0,
+                    1250.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1687.0,
+                    1405.0,
+                    1687.0,
+                    1405.0,
+                    1727.0,
+                    293.0,
+                    1727.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    1720.0,
+                    1402.0,
+                    1720.0,
+                    1402.0,
+                    1756.0,
+                    322.0,
+                    1756.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    324.0,
+                    1753.0,
+                    955.0,
+                    1753.0,
+                    955.0,
+                    1783.0,
+                    324.0,
+                    1783.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1804.0,
+                    1405.0,
+                    1804.0,
+                    1405.0,
+                    1840.0,
+                    296.0,
+                    1840.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    1835.0,
+                    1323.0,
+                    1835.0,
+                    1323.0,
+                    1869.0,
+                    323.0,
+                    1869.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    790.0,
+                    1404.0,
+                    790.0,
+                    1404.0,
+                    829.0,
+                    294.0,
+                    829.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    824.0,
+                    1402.0,
+                    824.0,
+                    1402.0,
+                    858.0,
+                    322.0,
+                    858.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    324.0,
+                    854.0,
+                    505.0,
+                    854.0,
+                    505.0,
+                    886.0,
+                    324.0,
+                    886.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1887.0,
+                    1402.0,
+                    1887.0,
+                    1402.0,
+                    1923.0,
+                    296.0,
+                    1923.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    1918.0,
+                    912.0,
+                    1918.0,
+                    912.0,
+                    1953.0,
+                    322.0,
+                    1953.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    1074.0,
+                    1403.0,
+                    1074.0,
+                    1403.0,
+                    1109.0,
+                    297.0,
+                    1109.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    1104.0,
+                    1405.0,
+                    1104.0,
+                    1405.0,
+                    1141.0,
+                    322.0,
+                    1141.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    321.0,
+                    1135.0,
+                    1226.0,
+                    1135.0,
+                    1226.0,
+                    1171.0,
+                    321.0,
+                    1171.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 11,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 2,
+                "poly": [
+                    299,
+                    75,
+                    816,
+                    75,
+                    816,
+                    105,
+                    299,
+                    105
+                ],
+                "score": 0.883
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    836,
+                    2088,
+                    865,
+                    2088,
+                    865,
+                    2113,
+                    836,
+                    2113
+                ],
+                "score": 0.834
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    295,
+                    1120,
+                    1400,
+                    1120,
+                    1400,
+                    1185,
+                    295,
+                    1185
+                ],
+                "score": 0.672
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    229,
+                    1400,
+                    229,
+                    1400,
+                    323,
+                    298,
+                    323
+                ],
+                "score": 0.647
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1317,
+                    1402,
+                    1317,
+                    1402,
+                    1381,
+                    298,
+                    1381
+                ],
+                "score": 0.645
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    296,
+                    1036,
+                    1401,
+                    1036,
+                    1401,
+                    1103,
+                    296,
+                    1103
+                ],
+                "score": 0.629
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    788,
+                    1400,
+                    788,
+                    1400,
+                    852,
+                    298,
+                    852
+                ],
+                "score": 0.598
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    1203,
+                    1401,
+                    1203,
+                    1401,
+                    1298,
+                    299,
+                    1298
+                ],
+                "score": 0.597
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    953,
+                    1401,
+                    953,
+                    1401,
+                    1020,
+                    297,
+                    1020
+                ],
+                "score": 0.584
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    705,
+                    1400,
+                    705,
+                    1400,
+                    770,
+                    299,
+                    770
+                ],
+                "score": 0.561
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    870,
+                    1403,
+                    870,
+                    1403,
+                    936,
+                    299,
+                    936
+                ],
+                "score": 0.56
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    295,
+                    425,
+                    1402,
+                    425,
+                    1402,
+                    490,
+                    295,
+                    490
+                ],
+                "score": 0.548
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    509,
+                    1400,
+                    509,
+                    1400,
+                    602,
+                    299,
+                    602
+                ],
+                "score": 0.538
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    622,
+                    1401,
+                    622,
+                    1401,
+                    687,
+                    299,
+                    687
+                ],
+                "score": 0.534
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    296,
+                    343,
+                    1402,
+                    343,
+                    1402,
+                    407,
+                    296,
+                    407
+                ],
+                "score": 0.534
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    73.0,
+                    817.0,
+                    73.0,
+                    817.0,
+                    108.0,
+                    295.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    832.0,
+                    2085.0,
+                    870.0,
+                    2085.0,
+                    870.0,
+                    2123.0,
+                    832.0,
+                    2123.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1118.0,
+                    1405.0,
+                    1118.0,
+                    1405.0,
+                    1156.0,
+                    294.0,
+                    1156.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    1153.0,
+                    991.0,
+                    1153.0,
+                    991.0,
+                    1186.0,
+                    322.0,
+                    1186.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    227.0,
+                    1405.0,
+                    227.0,
+                    1405.0,
+                    268.0,
+                    292.0,
+                    268.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    321.0,
+                    259.0,
+                    1405.0,
+                    259.0,
+                    1405.0,
+                    299.0,
+                    321.0,
+                    299.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    324.0,
+                    292.0,
+                    516.0,
+                    292.0,
+                    516.0,
+                    322.0,
+                    324.0,
+                    322.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1317.0,
+                    1404.0,
+                    1317.0,
+                    1404.0,
+                    1353.0,
+                    294.0,
+                    1353.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    325.0,
+                    1349.0,
+                    1073.0,
+                    1349.0,
+                    1073.0,
+                    1381.0,
+                    325.0,
+                    1381.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1037.0,
+                    1405.0,
+                    1037.0,
+                    1405.0,
+                    1073.0,
+                    296.0,
+                    1073.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    1068.0,
+                    1356.0,
+                    1068.0,
+                    1356.0,
+                    1103.0,
+                    323.0,
+                    1103.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    784.0,
+                    1405.0,
+                    784.0,
+                    1405.0,
+                    825.0,
+                    291.0,
+                    825.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    318.0,
+                    815.0,
+                    398.0,
+                    815.0,
+                    398.0,
+                    853.0,
+                    318.0,
+                    853.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1203.0,
+                    1406.0,
+                    1203.0,
+                    1406.0,
+                    1241.0,
+                    294.0,
+                    1241.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    1236.0,
+                    1406.0,
+                    1236.0,
+                    1406.0,
+                    1269.0,
+                    322.0,
+                    1269.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    1266.0,
+                    1279.0,
+                    1266.0,
+                    1279.0,
+                    1300.0,
+                    322.0,
+                    1300.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    954.0,
+                    1405.0,
+                    954.0,
+                    1405.0,
+                    990.0,
+                    296.0,
+                    990.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    987.0,
+                    1344.0,
+                    987.0,
+                    1344.0,
+                    1020.0,
+                    322.0,
+                    1020.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    704.0,
+                    1404.0,
+                    704.0,
+                    1404.0,
+                    744.0,
+                    294.0,
+                    744.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    324.0,
+                    736.0,
+                    983.0,
+                    736.0,
+                    983.0,
+                    772.0,
+                    324.0,
+                    772.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    869.0,
+                    1403.0,
+                    869.0,
+                    1403.0,
+                    908.0,
+                    295.0,
+                    908.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    904.0,
+                    1144.0,
+                    904.0,
+                    1144.0,
+                    937.0,
+                    323.0,
+                    937.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    428.0,
+                    1404.0,
+                    428.0,
+                    1404.0,
+                    460.0,
+                    297.0,
+                    460.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    456.0,
+                    1214.0,
+                    456.0,
+                    1214.0,
+                    492.0,
+                    323.0,
+                    492.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    510.0,
+                    1404.0,
+                    510.0,
+                    1404.0,
+                    543.0,
+                    295.0,
+                    543.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    324.0,
+                    541.0,
+                    1402.0,
+                    541.0,
+                    1402.0,
+                    575.0,
+                    324.0,
+                    575.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    327.0,
+                    570.0,
+                    533.0,
+                    570.0,
+                    533.0,
+                    601.0,
+                    327.0,
+                    601.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    621.0,
+                    1405.0,
+                    621.0,
+                    1405.0,
+                    657.0,
+                    295.0,
+                    657.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    320.0,
+                    652.0,
+                    1028.0,
+                    652.0,
+                    1028.0,
+                    687.0,
+                    320.0,
+                    687.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    339.0,
+                    1406.0,
+                    339.0,
+                    1406.0,
+                    381.0,
+                    294.0,
+                    381.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    321.0,
+                    375.0,
+                    861.0,
+                    375.0,
+                    861.0,
+                    408.0,
+                    321.0,
+                    408.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 12,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    1277,
+                    1404,
+                    1277,
+                    1404,
+                    1371,
+                    299,
+                    1371
+                ],
+                "score": 0.963
+            },
+            {
+                "category_id": 3,
+                "poly": [
+                    594,
+                    302,
+                    1103,
+                    302,
+                    1103,
+                    706,
+                    594,
+                    706
+                ],
+                "score": 0.938
+            },
+            {
+                "category_id": 3,
+                "poly": [
+                    298,
+                    847,
+                    1403,
+                    847,
+                    1403,
+                    1251,
+                    298,
+                    1251
+                ],
+                "score": 0.899
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    298,
+                    75,
+                    816,
+                    75,
+                    816,
+                    105,
+                    298,
+                    105
+                ],
+                "score": 0.874
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    836,
+                    2088,
+                    865,
+                    2088,
+                    865,
+                    2112,
+                    836,
+                    2112
+                ],
+                "score": 0.833
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    302,
+                    227,
+                    771,
+                    227,
+                    771,
+                    261,
+                    302,
+                    261
+                ],
+                "score": 0.767
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    302,
+                    227,
+                    771,
+                    227,
+                    771,
+                    261,
+                    302,
+                    261
+                ],
+                "score": 0.121
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    743.0,
+                    297.0,
+                    800.0,
+                    297.0,
+                    800.0,
+                    712.0,
+                    743.0,
+                    712.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    692.0,
+                    298.0,
+                    746.0,
+                    298.0,
+                    746.0,
+                    711.0,
+                    692.0,
+                    711.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    950.0,
+                    298.0,
+                    1001.0,
+                    298.0,
+                    1001.0,
+                    711.0,
+                    950.0,
+                    711.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    798.0,
+                    299.0,
+                    850.0,
+                    299.0,
+                    850.0,
+                    710.0,
+                    798.0,
+                    710.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    852.0,
+                    299.0,
+                    900.0,
+                    299.0,
+                    900.0,
+                    710.0,
+                    852.0,
+                    710.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    901.0,
+                    299.0,
+                    949.0,
+                    299.0,
+                    949.0,
+                    713.0,
+                    901.0,
+                    713.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    999.0,
+                    299.0,
+                    1056.0,
+                    299.0,
+                    1056.0,
+                    711.0,
+                    999.0,
+                    711.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    645.0,
+                    301.0,
+                    692.0,
+                    301.0,
+                    692.0,
+                    711.0,
+                    645.0,
+                    711.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    593.0,
+                    302.0,
+                    641.0,
+                    302.0,
+                    641.0,
+                    712.0,
+                    593.0,
+                    712.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1054.0,
+                    303.0,
+                    1105.0,
+                    303.0,
+                    1105.0,
+                    709.0,
+                    1054.0,
+                    709.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    459.0,
+                    844.0,
+                    650.0,
+                    844.0,
+                    650.0,
+                    875.0,
+                    459.0,
+                    875.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1051.0,
+                    844.0,
+                    1241.0,
+                    844.0,
+                    1241.0,
+                    875.0,
+                    1051.0,
+                    875.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    73.0,
+                    818.0,
+                    73.0,
+                    818.0,
+                    108.0,
+                    296.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    832.0,
+                    2084.0,
+                    871.0,
+                    2084.0,
+                    871.0,
+                    2122.0,
+                    832.0,
+                    2122.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    224.0,
+                    778.0,
+                    224.0,
+                    778.0,
+                    266.0,
+                    295.0,
+                    266.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1275.0,
+                    1404.0,
+                    1275.0,
+                    1404.0,
+                    1312.0,
+                    294.0,
+                    1312.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1309.0,
+                    1404.0,
+                    1309.0,
+                    1404.0,
+                    1343.0,
+                    295.0,
+                    1343.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1339.0,
+                    1264.0,
+                    1339.0,
+                    1264.0,
+                    1373.0,
+                    295.0,
+                    1373.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    224.0,
+                    778.0,
+                    224.0,
+                    778.0,
+                    266.0,
+                    295.0,
+                    266.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 13,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 2,
+                "poly": [
+                    299,
+                    76,
+                    815,
+                    76,
+                    815,
+                    104,
+                    299,
+                    104
+                ],
+                "score": 0.872
+            },
+            {
+                "category_id": 3,
+                "poly": [
+                    534,
+                    213,
+                    1090,
+                    213,
+                    1090,
+                    1232,
+                    534,
+                    1232
+                ],
+                "score": 0.832
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    836,
+                    2088,
+                    864,
+                    2088,
+                    864,
+                    2112,
+                    836,
+                    2112
+                ],
+                "score": 0.828
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    801,
+                    1282,
+                    896,
+                    1282,
+                    896,
+                    1303,
+                    801,
+                    1303
+                ],
+                "score": 0.365
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    801,
+                    1282,
+                    896,
+                    1282,
+                    896,
+                    1303,
+                    801,
+                    1303
+                ],
+                "score": 0.106
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    73.0,
+                    818.0,
+                    73.0,
+                    818.0,
+                    108.0,
+                    295.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    760.0,
+                    232.0,
+                    939.0,
+                    232.0,
+                    939.0,
+                    269.0,
+                    760.0,
+                    269.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    542.0,
+                    386.0,
+                    572.0,
+                    386.0,
+                    572.0,
+                    576.0,
+                    542.0,
+                    576.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    996.0,
+                    637.0,
+                    1026.0,
+                    637.0,
+                    1026.0,
+                    658.0,
+                    996.0,
+                    658.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1044.0,
+                    639.0,
+                    1067.0,
+                    639.0,
+                    1067.0,
+                    657.0,
+                    1044.0,
+                    657.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    715.0,
+                    714.0,
+                    982.0,
+                    714.0,
+                    982.0,
+                    747.0,
+                    715.0,
+                    747.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    794.0,
+                    801.0,
+                    905.0,
+                    801.0,
+                    905.0,
+                    831.0,
+                    794.0,
+                    831.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    542.0,
+                    945.0,
+                    572.0,
+                    945.0,
+                    572.0,
+                    1136.0,
+                    542.0,
+                    1136.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    832.0,
+                    2085.0,
+                    869.0,
+                    2085.0,
+                    869.0,
+                    2124.0,
+                    832.0,
+                    2124.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    795.0,
+                    1277.0,
+                    902.0,
+                    1277.0,
+                    902.0,
+                    1308.0,
+                    795.0,
+                    1308.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    795.0,
+                    1277.0,
+                    902.0,
+                    1277.0,
+                    902.0,
+                    1308.0,
+                    795.0,
+                    1308.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 14,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 3,
+                "poly": [
+                    596,
+                    301,
+                    1102,
+                    301,
+                    1102,
+                    709,
+                    596,
+                    709
+                ],
+                "score": 0.91
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    298,
+                    75,
+                    815,
+                    75,
+                    815,
+                    105,
+                    298,
+                    105
+                ],
+                "score": 0.879
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    836,
+                    2088,
+                    865,
+                    2088,
+                    865,
+                    2112,
+                    836,
+                    2112
+                ],
+                "score": 0.826
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    299,
+                    227,
+                    795,
+                    227,
+                    795,
+                    261,
+                    299,
+                    261
+                ],
+                "score": 0.758
+            },
+            {
+                "category_id": 3,
+                "poly": [
+                    525,
+                    660,
+                    1096,
+                    660,
+                    1096,
+                    1840,
+                    525,
+                    1840
+                ],
+                "score": 0.419
+            },
+            {
+                "category_id": 4,
+                "poly": [
+                    801,
+                    1882,
+                    897,
+                    1882,
+                    897,
+                    1904,
+                    801,
+                    1904
+                ],
+                "score": 0.22
+            },
+            {
+                "category_id": 3,
+                "poly": [
+                    549,
+                    300,
+                    1106,
+                    300,
+                    1106,
+                    1278,
+                    549,
+                    1278
+                ],
+                "score": 0.128
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    227,
+                    795,
+                    227,
+                    795,
+                    261,
+                    299,
+                    261
+                ],
+                "score": 0.112
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    606.0,
+                    309.0,
+                    792.0,
+                    309.0,
+                    792.0,
+                    341.0,
+                    606.0,
+                    341.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    803.0,
+                    315.0,
+                    842.0,
+                    315.0,
+                    842.0,
+                    334.0,
+                    803.0,
+                    334.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    762.0,
+                    349.0,
+                    792.0,
+                    349.0,
+                    792.0,
+                    375.0,
+                    762.0,
+                    375.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    859.0,
+                    354.0,
+                    883.0,
+                    354.0,
+                    883.0,
+                    369.0,
+                    859.0,
+                    369.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    753.0,
+                    391.0,
+                    796.0,
+                    391.0,
+                    796.0,
+                    420.0,
+                    753.0,
+                    420.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    806.0,
+                    392.0,
+                    843.0,
+                    392.0,
+                    843.0,
+                    417.0,
+                    806.0,
+                    417.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    908.0,
+                    402.0,
+                    946.0,
+                    402.0,
+                    946.0,
+                    419.0,
+                    908.0,
+                    419.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    703.0,
+                    433.0,
+                    738.0,
+                    433.0,
+                    738.0,
+                    450.0,
+                    703.0,
+                    450.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    807.0,
+                    433.0,
+                    842.0,
+                    433.0,
+                    842.0,
+                    453.0,
+                    807.0,
+                    453.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    914.0,
+                    438.0,
+                    936.0,
+                    438.0,
+                    936.0,
+                    447.0,
+                    914.0,
+                    447.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1026.0,
+                    441.0,
+                    1037.0,
+                    441.0,
+                    1037.0,
+                    447.0,
+                    1026.0,
+                    447.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    693.0,
+                    468.0,
+                    906.0,
+                    468.0,
+                    906.0,
+                    502.0,
+                    693.0,
+                    502.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    911.0,
+                    475.0,
+                    943.0,
+                    475.0,
+                    943.0,
+                    497.0,
+                    911.0,
+                    497.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    967.0,
+                    479.0,
+                    977.0,
+                    479.0,
+                    977.0,
+                    486.0,
+                    967.0,
+                    486.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1009.0,
+                    472.0,
+                    1045.0,
+                    472.0,
+                    1045.0,
+                    498.0,
+                    1009.0,
+                    498.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    798.0,
+                    512.0,
+                    847.0,
+                    512.0,
+                    847.0,
+                    540.0,
+                    798.0,
+                    540.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    754.0,
+                    514.0,
+                    794.0,
+                    514.0,
+                    794.0,
+                    540.0,
+                    754.0,
+                    540.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    706.0,
+                    557.0,
+                    741.0,
+                    557.0,
+                    741.0,
+                    574.0,
+                    706.0,
+                    574.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    760.0,
+                    556.0,
+                    790.0,
+                    556.0,
+                    790.0,
+                    573.0,
+                    760.0,
+                    573.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    808.0,
+                    557.0,
+                    841.0,
+                    557.0,
+                    841.0,
+                    573.0,
+                    808.0,
+                    573.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1007.0,
+                    559.0,
+                    1030.0,
+                    559.0,
+                    1030.0,
+                    573.0,
+                    1007.0,
+                    573.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    701.0,
+                    593.0,
+                    746.0,
+                    593.0,
+                    746.0,
+                    617.0,
+                    701.0,
+                    617.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    756.0,
+                    593.0,
+                    792.0,
+                    593.0,
+                    792.0,
+                    618.0,
+                    756.0,
+                    618.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    800.0,
+                    593.0,
+                    845.0,
+                    593.0,
+                    845.0,
+                    621.0,
+                    800.0,
+                    621.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    862.0,
+                    597.0,
+                    891.0,
+                    597.0,
+                    891.0,
+                    615.0,
+                    862.0,
+                    615.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    905.0,
+                    592.0,
+                    948.0,
+                    592.0,
+                    948.0,
+                    621.0,
+                    905.0,
+                    621.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    959.0,
+                    595.0,
+                    995.0,
+                    595.0,
+                    995.0,
+                    616.0,
+                    959.0,
+                    616.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1011.0,
+                    596.0,
+                    1044.0,
+                    596.0,
+                    1044.0,
+                    614.0,
+                    1011.0,
+                    614.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1058.0,
+                    594.0,
+                    1092.0,
+                    594.0,
+                    1092.0,
+                    615.0,
+                    1058.0,
+                    615.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    760.0,
+                    638.0,
+                    788.0,
+                    638.0,
+                    788.0,
+                    656.0,
+                    760.0,
+                    656.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1011.0,
+                    636.0,
+                    1048.0,
+                    636.0,
+                    1048.0,
+                    658.0,
+                    1011.0,
+                    658.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1058.0,
+                    634.0,
+                    1090.0,
+                    634.0,
+                    1090.0,
+                    656.0,
+                    1058.0,
+                    656.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    895.0,
+                    676.0,
+                    943.0,
+                    676.0,
+                    943.0,
+                    695.0,
+                    895.0,
+                    695.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    952.0,
+                    675.0,
+                    987.0,
+                    675.0,
+                    987.0,
+                    695.0,
+                    952.0,
+                    695.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1052.0,
+                    671.0,
+                    1099.0,
+                    671.0,
+                    1099.0,
+                    702.0,
+                    1052.0,
+                    702.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    710.25,
+                    396.5,
+                    736.25,
+                    396.5,
+                    736.25,
+                    410.5,
+                    710.25,
+                    410.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1060.0,
+                    438.5,
+                    1080.0,
+                    438.5,
+                    1080.0,
+                    449.5,
+                    1060.0,
+                    449.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1061.75,
+                    482.5,
+                    1086.75,
+                    482.5,
+                    1086.75,
+                    497.5,
+                    1061.75,
+                    497.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    714.25,
+                    521.5,
+                    745.25,
+                    521.5,
+                    745.25,
+                    535.5,
+                    714.25,
+                    535.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    73.0,
+                    818.0,
+                    73.0,
+                    818.0,
+                    108.0,
+                    296.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    832.0,
+                    2084.0,
+                    871.0,
+                    2084.0,
+                    871.0,
+                    2123.0,
+                    832.0,
+                    2123.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    225.0,
+                    798.0,
+                    225.0,
+                    798.0,
+                    265.0,
+                    293.0,
+                    265.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    598.0,
+                    658.0,
+                    1100.0,
+                    658.0,
+                    1100.0,
+                    711.0,
+                    598.0,
+                    711.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    760.0,
+                    786.0,
+                    938.0,
+                    786.0,
+                    938.0,
+                    821.0,
+                    760.0,
+                    821.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    541.0,
+                    935.0,
+                    573.0,
+                    935.0,
+                    573.0,
+                    1130.0,
+                    541.0,
+                    1130.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    846.0,
+                    1186.0,
+                    942.0,
+                    1186.0,
+                    942.0,
+                    1211.0,
+                    846.0,
+                    1211.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    947.0,
+                    1190.0,
+                    982.0,
+                    1190.0,
+                    982.0,
+                    1209.0,
+                    947.0,
+                    1209.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    719.0,
+                    1265.0,
+                    980.0,
+                    1265.0,
+                    980.0,
+                    1302.0,
+                    719.0,
+                    1302.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    791.0,
+                    1396.0,
+                    908.0,
+                    1396.0,
+                    908.0,
+                    1434.0,
+                    791.0,
+                    1434.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    541.0,
+                    1547.0,
+                    573.0,
+                    1547.0,
+                    573.0,
+                    1742.0,
+                    541.0,
+                    1742.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    798.0,
+                    1878.0,
+                    902.0,
+                    1878.0,
+                    902.0,
+                    1909.0,
+                    798.0,
+                    1909.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    607.0,
+                    315.0,
+                    637.0,
+                    315.0,
+                    637.0,
+                    336.0,
+                    607.0,
+                    336.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    651.0,
+                    311.0,
+                    696.0,
+                    311.0,
+                    696.0,
+                    339.0,
+                    651.0,
+                    339.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    697.0,
+                    312.0,
+                    739.0,
+                    312.0,
+                    739.0,
+                    335.0,
+                    697.0,
+                    335.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    760.0,
+                    317.0,
+                    782.0,
+                    317.0,
+                    782.0,
+                    330.0,
+                    760.0,
+                    330.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    760.0,
+                    399.0,
+                    778.0,
+                    399.0,
+                    778.0,
+                    413.0,
+                    760.0,
+                    413.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    924.0,
+                    400.0,
+                    939.0,
+                    400.0,
+                    939.0,
+                    415.0,
+                    924.0,
+                    415.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    966.0,
+                    399.0,
+                    985.0,
+                    399.0,
+                    985.0,
+                    411.0,
+                    966.0,
+                    411.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    760.0,
+                    785.0,
+                    940.0,
+                    785.0,
+                    940.0,
+                    822.0,
+                    760.0,
+                    822.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    540.0,
+                    936.0,
+                    573.0,
+                    936.0,
+                    573.0,
+                    1127.0,
+                    540.0,
+                    1127.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    225.0,
+                    798.0,
+                    225.0,
+                    798.0,
+                    265.0,
+                    293.0,
+                    265.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 15,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 2,
+                "poly": [
+                    298,
+                    75,
+                    816,
+                    75,
+                    816,
+                    105,
+                    298,
+                    105
+                ],
+                "score": 0.864
+            },
+            {
+                "category_id": 3,
+                "poly": [
+                    592,
+                    302,
+                    1106,
+                    302,
+                    1106,
+                    711,
+                    592,
+                    711
+                ],
+                "score": 0.861
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    836,
+                    2088,
+                    864,
+                    2088,
+                    864,
+                    2112,
+                    836,
+                    2112
+                ],
+                "score": 0.824
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    544,
+                    943,
+                    568,
+                    943,
+                    568,
+                    1123,
+                    544,
+                    1123
+                ],
+                "score": 0.697
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    544,
+                    1553,
+                    568,
+                    1553,
+                    568,
+                    1736,
+                    544,
+                    1736
+                ],
+                "score": 0.649
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    298,
+                    227,
+                    761,
+                    227,
+                    761,
+                    261,
+                    298,
+                    261
+                ],
+                "score": 0.622
+            },
+            {
+                "category_id": 3,
+                "poly": [
+                    576,
+                    1458,
+                    1080,
+                    1458,
+                    1080,
+                    1828,
+                    576,
+                    1828
+                ],
+                "score": 0.43
+            },
+            {
+                "category_id": 3,
+                "poly": [
+                    594,
+                    824,
+                    1082,
+                    824,
+                    1082,
+                    1219,
+                    594,
+                    1219
+                ],
+                "score": 0.339
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    227,
+                    761,
+                    227,
+                    761,
+                    261,
+                    298,
+                    261
+                ],
+                "score": 0.248
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    724,
+                    1271,
+                    975,
+                    1271,
+                    975,
+                    1294,
+                    724,
+                    1294
+                ],
+                "score": 0.229
+            },
+            {
+                "category_id": 4,
+                "poly": [
+                    801,
+                    1882,
+                    897,
+                    1882,
+                    897,
+                    1904,
+                    801,
+                    1904
+                ],
+                "score": 0.217
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    796,
+                    1404,
+                    901,
+                    1404,
+                    901,
+                    1426,
+                    796,
+                    1426
+                ],
+                "score": 0.153
+            },
+            {
+                "category_id": 4,
+                "poly": [
+                    724,
+                    1271,
+                    975,
+                    1271,
+                    975,
+                    1294,
+                    724,
+                    1294
+                ],
+                "score": 0.104
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    73.0,
+                    818.0,
+                    73.0,
+                    818.0,
+                    108.0,
+                    296.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    789.0,
+                    294.0,
+                    862.0,
+                    294.0,
+                    862.0,
+                    716.0,
+                    789.0,
+                    716.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    736.0,
+                    295.0,
+                    810.0,
+                    295.0,
+                    810.0,
+                    715.0,
+                    736.0,
+                    715.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    893.0,
+                    295.0,
+                    958.0,
+                    295.0,
+                    958.0,
+                    714.0,
+                    893.0,
+                    714.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    683.0,
+                    296.0,
+                    757.0,
+                    296.0,
+                    757.0,
+                    716.0,
+                    683.0,
+                    716.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1042.0,
+                    296.0,
+                    1116.0,
+                    296.0,
+                    1116.0,
+                    717.0,
+                    1042.0,
+                    717.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    947.0,
+                    297.0,
+                    1012.0,
+                    297.0,
+                    1012.0,
+                    714.0,
+                    947.0,
+                    714.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    847.0,
+                    298.0,
+                    904.0,
+                    298.0,
+                    904.0,
+                    710.0,
+                    847.0,
+                    710.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    997.0,
+                    298.0,
+                    1061.0,
+                    298.0,
+                    1061.0,
+                    714.0,
+                    997.0,
+                    714.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    639.0,
+                    300.0,
+                    700.0,
+                    300.0,
+                    700.0,
+                    713.0,
+                    639.0,
+                    713.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    589.0,
+                    301.0,
+                    647.0,
+                    301.0,
+                    647.0,
+                    713.0,
+                    589.0,
+                    713.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    832.0,
+                    2085.0,
+                    869.0,
+                    2085.0,
+                    869.0,
+                    2124.0,
+                    832.0,
+                    2124.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    539.0,
+                    937.0,
+                    570.0,
+                    937.0,
+                    570.0,
+                    1129.0,
+                    539.0,
+                    1129.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    539.0,
+                    1548.0,
+                    570.0,
+                    1548.0,
+                    570.0,
+                    1740.0,
+                    539.0,
+                    1740.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    221.0,
+                    768.0,
+                    221.0,
+                    768.0,
+                    272.0,
+                    294.0,
+                    272.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    701.0,
+                    1561.0,
+                    955.0,
+                    1561.0,
+                    955.0,
+                    1733.0,
+                    701.0,
+                    1733.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    788.0,
+                    1596.0,
+                    816.0,
+                    1596.0,
+                    816.0,
+                    1609.0,
+                    788.0,
+                    1609.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    824.0,
+                    1596.0,
+                    856.0,
+                    1596.0,
+                    856.0,
+                    1609.0,
+                    824.0,
+                    1609.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    766.0,
+                    1646.0,
+                    774.0,
+                    1646.0,
+                    774.0,
+                    1653.0,
+                    766.0,
+                    1653.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1040.0,
+                    1148.0,
+                    1073.0,
+                    1148.0,
+                    1073.0,
+                    1171.0,
+                    1040.0,
+                    1171.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    631.0,
+                    1189.0,
+                    648.0,
+                    1189.0,
+                    648.0,
+                    1201.0,
+                    631.0,
+                    1201.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    798.0,
+                    1878.0,
+                    902.0,
+                    1878.0,
+                    902.0,
+                    1909.0,
+                    798.0,
+                    1909.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    722.0,
+                    1267.0,
+                    975.0,
+                    1267.0,
+                    975.0,
+                    1297.0,
+                    722.0,
+                    1297.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    221.0,
+                    768.0,
+                    221.0,
+                    768.0,
+                    272.0,
+                    294.0,
+                    272.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    722.0,
+                    1267.0,
+                    975.0,
+                    1267.0,
+                    975.0,
+                    1297.0,
+                    722.0,
+                    1297.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    793.0,
+                    1400.0,
+                    904.0,
+                    1400.0,
+                    904.0,
+                    1431.0,
+                    793.0,
+                    1431.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 16,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 3,
+                "poly": [
+                    298,
+                    545,
+                    1411,
+                    545,
+                    1411,
+                    1913,
+                    298,
+                    1913
+                ],
+                "score": 0.932
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    300,
+                    75,
+                    815,
+                    75,
+                    815,
+                    105,
+                    300,
+                    105
+                ],
+                "score": 0.887
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    295,
+                    1404,
+                    295,
+                    1404,
+                    542,
+                    297,
+                    542
+                ],
+                "score": 0.875
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    836,
+                    2088,
+                    865,
+                    2088,
+                    865,
+                    2112,
+                    836,
+                    2112
+                ],
+                "score": 0.854
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    293,
+                    229,
+                    1328,
+                    229,
+                    1328,
+                    262,
+                    293,
+                    262
+                ],
+                "score": 0.506
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    493.0,
+                    576.0,
+                    577.0,
+                    576.0,
+                    577.0,
+                    612.0,
+                    493.0,
+                    612.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    850.0,
+                    579.0,
+                    949.0,
+                    579.0,
+                    949.0,
+                    610.0,
+                    850.0,
+                    610.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1199.0,
+                    577.0,
+                    1290.0,
+                    577.0,
+                    1290.0,
+                    610.0,
+                    1199.0,
+                    610.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    369.0,
+                    614.0,
+                    406.0,
+                    614.0,
+                    406.0,
+                    637.0,
+                    369.0,
+                    637.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    723.0,
+                    613.0,
+                    760.0,
+                    613.0,
+                    760.0,
+                    640.0,
+                    723.0,
+                    640.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1073.0,
+                    619.0,
+                    1109.0,
+                    619.0,
+                    1109.0,
+                    642.0,
+                    1073.0,
+                    642.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    303.0,
+                    654.0,
+                    337.0,
+                    654.0,
+                    337.0,
+                    819.0,
+                    303.0,
+                    819.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    369.0,
+                    657.0,
+                    401.0,
+                    657.0,
+                    401.0,
+                    679.0,
+                    369.0,
+                    679.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    724.0,
+                    657.0,
+                    756.0,
+                    657.0,
+                    756.0,
+                    680.0,
+                    724.0,
+                    680.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1073.0,
+                    662.0,
+                    1106.0,
+                    662.0,
+                    1106.0,
+                    685.0,
+                    1073.0,
+                    685.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1111.0,
+                    669.0,
+                    1121.0,
+                    669.0,
+                    1121.0,
+                    682.0,
+                    1111.0,
+                    682.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1174.0,
+                    671.0,
+                    1182.0,
+                    671.0,
+                    1182.0,
+                    679.0,
+                    1174.0,
+                    679.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1184.0,
+                    656.0,
+                    1202.0,
+                    656.0,
+                    1202.0,
+                    668.0,
+                    1184.0,
+                    668.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1253.0,
+                    669.0,
+                    1265.0,
+                    669.0,
+                    1265.0,
+                    680.0,
+                    1253.0,
+                    680.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    425.0,
+                    677.0,
+                    447.0,
+                    677.0,
+                    447.0,
+                    694.0,
+                    425.0,
+                    694.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    451.0,
+                    683.0,
+                    463.0,
+                    683.0,
+                    463.0,
+                    697.0,
+                    451.0,
+                    697.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    469.0,
+                    675.0,
+                    488.0,
+                    675.0,
+                    488.0,
+                    688.0,
+                    469.0,
+                    688.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1111.0,
+                    683.0,
+                    1182.0,
+                    683.0,
+                    1182.0,
+                    734.0,
+                    1111.0,
+                    734.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1124.0,
+                    688.0,
+                    1144.0,
+                    688.0,
+                    1144.0,
+                    700.0,
+                    1124.0,
+                    700.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    859.0,
+                    695.0,
+                    882.0,
+                    695.0,
+                    882.0,
+                    706.0,
+                    859.0,
+                    706.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    369.0,
+                    700.0,
+                    403.0,
+                    700.0,
+                    403.0,
+                    723.0,
+                    369.0,
+                    723.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    428.0,
+                    709.0,
+                    475.0,
+                    709.0,
+                    475.0,
+                    744.0,
+                    428.0,
+                    744.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    506.0,
+                    704.0,
+                    518.0,
+                    704.0,
+                    518.0,
+                    717.0,
+                    506.0,
+                    717.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    724.0,
+                    701.0,
+                    756.0,
+                    701.0,
+                    756.0,
+                    724.0,
+                    724.0,
+                    724.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    774.0,
+                    718.0,
+                    794.0,
+                    718.0,
+                    794.0,
+                    737.0,
+                    774.0,
+                    737.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    876.0,
+                    704.0,
+                    886.0,
+                    704.0,
+                    886.0,
+                    718.0,
+                    876.0,
+                    718.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1073.0,
+                    706.0,
+                    1106.0,
+                    706.0,
+                    1106.0,
+                    729.0,
+                    1073.0,
+                    729.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1203.0,
+                    703.0,
+                    1227.0,
+                    703.0,
+                    1227.0,
+                    718.0,
+                    1203.0,
+                    718.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1249.0,
+                    724.0,
+                    1262.0,
+                    724.0,
+                    1262.0,
+                    740.0,
+                    1249.0,
+                    740.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1300.0,
+                    727.0,
+                    1312.0,
+                    727.0,
+                    1312.0,
+                    740.0,
+                    1300.0,
+                    740.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    795.0,
+                    738.0,
+                    809.0,
+                    738.0,
+                    809.0,
+                    750.0,
+                    795.0,
+                    750.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    955.0,
+                    738.0,
+                    967.0,
+                    738.0,
+                    967.0,
+                    749.0,
+                    955.0,
+                    749.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    369.0,
+                    743.0,
+                    403.0,
+                    743.0,
+                    403.0,
+                    766.0,
+                    369.0,
+                    766.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    724.0,
+                    744.0,
+                    756.0,
+                    744.0,
+                    756.0,
+                    767.0,
+                    724.0,
+                    767.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1074.0,
+                    749.0,
+                    1106.0,
+                    749.0,
+                    1106.0,
+                    772.0,
+                    1074.0,
+                    772.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    369.0,
+                    787.0,
+                    401.0,
+                    787.0,
+                    401.0,
+                    810.0,
+                    369.0,
+                    810.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    453.0,
+                    776.0,
+                    482.0,
+                    776.0,
+                    482.0,
+                    812.0,
+                    453.0,
+                    812.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    724.0,
+                    787.0,
+                    756.0,
+                    787.0,
+                    756.0,
+                    810.0,
+                    724.0,
+                    810.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    764.0,
+                    775.0,
+                    788.0,
+                    775.0,
+                    788.0,
+                    801.0,
+                    764.0,
+                    801.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1073.0,
+                    793.0,
+                    1106.0,
+                    793.0,
+                    1106.0,
+                    816.0,
+                    1073.0,
+                    816.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    380.0,
+                    830.0,
+                    410.0,
+                    830.0,
+                    410.0,
+                    867.0,
+                    380.0,
+                    867.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    457.0,
+                    844.0,
+                    489.0,
+                    844.0,
+                    489.0,
+                    868.0,
+                    457.0,
+                    868.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    506.0,
+                    822.0,
+                    563.0,
+                    822.0,
+                    563.0,
+                    864.0,
+                    506.0,
+                    864.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    603.0,
+                    845.0,
+                    632.0,
+                    845.0,
+                    632.0,
+                    868.0,
+                    603.0,
+                    868.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    680.0,
+                    845.0,
+                    697.0,
+                    845.0,
+                    697.0,
+                    867.0,
+                    680.0,
+                    867.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    735.0,
+                    831.0,
+                    765.0,
+                    831.0,
+                    765.0,
+                    868.0,
+                    735.0,
+                    868.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    812.0,
+                    847.0,
+                    842.0,
+                    847.0,
+                    842.0,
+                    870.0,
+                    812.0,
+                    870.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    892.0,
+                    850.0,
+                    906.0,
+                    850.0,
+                    906.0,
+                    865.0,
+                    892.0,
+                    865.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    958.0,
+                    847.0,
+                    986.0,
+                    847.0,
+                    986.0,
+                    870.0,
+                    958.0,
+                    870.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1033.0,
+                    848.0,
+                    1052.0,
+                    848.0,
+                    1052.0,
+                    867.0,
+                    1033.0,
+                    867.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1083.0,
+                    836.0,
+                    1114.0,
+                    836.0,
+                    1114.0,
+                    873.0,
+                    1083.0,
+                    873.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1162.0,
+                    834.0,
+                    1193.0,
+                    834.0,
+                    1193.0,
+                    874.0,
+                    1162.0,
+                    874.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1243.0,
+                    854.0,
+                    1255.0,
+                    854.0,
+                    1255.0,
+                    871.0,
+                    1243.0,
+                    871.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1275.0,
+                    827.0,
+                    1294.0,
+                    827.0,
+                    1294.0,
+                    844.0,
+                    1275.0,
+                    844.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1308.0,
+                    851.0,
+                    1338.0,
+                    851.0,
+                    1338.0,
+                    874.0,
+                    1308.0,
+                    874.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1387.0,
+                    853.0,
+                    1403.0,
+                    853.0,
+                    1403.0,
+                    873.0,
+                    1387.0,
+                    873.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    806.0,
+                    883.0,
+                    990.0,
+                    883.0,
+                    990.0,
+                    917.0,
+                    806.0,
+                    917.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    493.0,
+                    1056.0,
+                    577.0,
+                    1056.0,
+                    577.0,
+                    1090.0,
+                    493.0,
+                    1090.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    849.0,
+                    1056.0,
+                    949.0,
+                    1056.0,
+                    949.0,
+                    1090.0,
+                    849.0,
+                    1090.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1200.0,
+                    1056.0,
+                    1290.0,
+                    1056.0,
+                    1290.0,
+                    1092.0,
+                    1200.0,
+                    1092.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    378.0,
+                    1088.0,
+                    398.0,
+                    1088.0,
+                    398.0,
+                    1105.0,
+                    378.0,
+                    1105.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    717.0,
+                    1090.0,
+                    751.0,
+                    1090.0,
+                    751.0,
+                    1113.0,
+                    717.0,
+                    1113.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1077.0,
+                    1091.0,
+                    1096.0,
+                    1091.0,
+                    1096.0,
+                    1110.0,
+                    1077.0,
+                    1110.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    729.0,
+                    1123.0,
+                    748.0,
+                    1123.0,
+                    748.0,
+                    1142.0,
+                    729.0,
+                    1142.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    303.0,
+                    1137.0,
+                    338.0,
+                    1137.0,
+                    338.0,
+                    1301.0,
+                    303.0,
+                    1301.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    377.0,
+                    1142.0,
+                    397.0,
+                    1142.0,
+                    397.0,
+                    1163.0,
+                    377.0,
+                    1163.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    717.0,
+                    1153.0,
+                    750.0,
+                    1153.0,
+                    750.0,
+                    1175.0,
+                    717.0,
+                    1175.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    862.0,
+                    1149.0,
+                    880.0,
+                    1149.0,
+                    880.0,
+                    1169.0,
+                    862.0,
+                    1169.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1076.0,
+                    1146.0,
+                    1096.0,
+                    1146.0,
+                    1096.0,
+                    1166.0,
+                    1076.0,
+                    1166.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    377.0,
+                    1198.0,
+                    395.0,
+                    1198.0,
+                    395.0,
+                    1218.0,
+                    377.0,
+                    1218.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    421.0,
+                    1195.0,
+                    438.0,
+                    1195.0,
+                    438.0,
+                    1215.0,
+                    421.0,
+                    1215.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    729.0,
+                    1186.0,
+                    748.0,
+                    1186.0,
+                    748.0,
+                    1206.0,
+                    729.0,
+                    1206.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    791.0,
+                    1186.0,
+                    806.0,
+                    1186.0,
+                    806.0,
+                    1209.0,
+                    791.0,
+                    1209.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    824.0,
+                    1189.0,
+                    838.0,
+                    1189.0,
+                    838.0,
+                    1205.0,
+                    824.0,
+                    1205.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    867.0,
+                    1175.0,
+                    895.0,
+                    1175.0,
+                    895.0,
+                    1201.0,
+                    867.0,
+                    1201.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1076.0,
+                    1201.0,
+                    1096.0,
+                    1201.0,
+                    1096.0,
+                    1221.0,
+                    1076.0,
+                    1221.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    415.0,
+                    1221.0,
+                    433.0,
+                    1221.0,
+                    433.0,
+                    1234.0,
+                    415.0,
+                    1234.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    460.0,
+                    1223.0,
+                    468.0,
+                    1223.0,
+                    468.0,
+                    1231.0,
+                    460.0,
+                    1231.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    718.0,
+                    1217.0,
+                    748.0,
+                    1217.0,
+                    748.0,
+                    1240.0,
+                    718.0,
+                    1240.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    801.0,
+                    1214.0,
+                    823.0,
+                    1214.0,
+                    823.0,
+                    1234.0,
+                    801.0,
+                    1234.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    445.0,
+                    1235.0,
+                    457.0,
+                    1235.0,
+                    457.0,
+                    1249.0,
+                    445.0,
+                    1249.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    380.0,
+                    1258.0,
+                    391.0,
+                    1258.0,
+                    391.0,
+                    1270.0,
+                    380.0,
+                    1270.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    431.0,
+                    1252.0,
+                    466.0,
+                    1252.0,
+                    466.0,
+                    1269.0,
+                    431.0,
+                    1269.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1137.0,
+                    1253.0,
+                    1149.0,
+                    1253.0,
+                    1149.0,
+                    1266.0,
+                    1137.0,
+                    1266.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    409.0,
+                    1272.0,
+                    494.0,
+                    1272.0,
+                    494.0,
+                    1309.0,
+                    409.0,
+                    1309.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    506.0,
+                    1267.0,
+                    518.0,
+                    1267.0,
+                    518.0,
+                    1278.0,
+                    506.0,
+                    1278.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    536.0,
+                    1296.0,
+                    547.0,
+                    1296.0,
+                    547.0,
+                    1310.0,
+                    536.0,
+                    1310.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    588.0,
+                    1299.0,
+                    601.0,
+                    1299.0,
+                    601.0,
+                    1313.0,
+                    588.0,
+                    1313.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    717.0,
+                    1279.0,
+                    750.0,
+                    1279.0,
+                    750.0,
+                    1302.0,
+                    717.0,
+                    1302.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    757.0,
+                    1281.0,
+                    804.0,
+                    1281.0,
+                    804.0,
+                    1325.0,
+                    757.0,
+                    1325.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    809.0,
+                    1275.0,
+                    830.0,
+                    1275.0,
+                    830.0,
+                    1315.0,
+                    809.0,
+                    1315.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    836.0,
+                    1272.0,
+                    854.0,
+                    1272.0,
+                    854.0,
+                    1283.0,
+                    836.0,
+                    1283.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    858.0,
+                    1275.0,
+                    879.0,
+                    1275.0,
+                    879.0,
+                    1318.0,
+                    858.0,
+                    1318.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    944.0,
+                    1269.0,
+                    964.0,
+                    1269.0,
+                    964.0,
+                    1289.0,
+                    944.0,
+                    1289.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    991.0,
+                    1298.0,
+                    1009.0,
+                    1298.0,
+                    1009.0,
+                    1319.0,
+                    991.0,
+                    1319.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1105.0,
+                    1283.0,
+                    1126.0,
+                    1283.0,
+                    1126.0,
+                    1299.0,
+                    1105.0,
+                    1299.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1146.0,
+                    1301.0,
+                    1168.0,
+                    1301.0,
+                    1168.0,
+                    1318.0,
+                    1146.0,
+                    1318.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1344.0,
+                    1305.0,
+                    1356.0,
+                    1305.0,
+                    1356.0,
+                    1318.0,
+                    1344.0,
+                    1318.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    378.0,
+                    1312.0,
+                    398.0,
+                    1312.0,
+                    398.0,
+                    1331.0,
+                    378.0,
+                    1331.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1077.0,
+                    1315.0,
+                    1097.0,
+                    1315.0,
+                    1097.0,
+                    1333.0,
+                    1077.0,
+                    1333.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    389.0,
+                    1330.0,
+                    403.0,
+                    1330.0,
+                    403.0,
+                    1345.0,
+                    389.0,
+                    1345.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    456.0,
+                    1325.0,
+                    488.0,
+                    1325.0,
+                    488.0,
+                    1348.0,
+                    456.0,
+                    1348.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    539.0,
+                    1330.0,
+                    551.0,
+                    1330.0,
+                    551.0,
+                    1344.0,
+                    539.0,
+                    1344.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    606.0,
+                    1325.0,
+                    636.0,
+                    1325.0,
+                    636.0,
+                    1348.0,
+                    606.0,
+                    1348.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    686.0,
+                    1325.0,
+                    704.0,
+                    1325.0,
+                    704.0,
+                    1347.0,
+                    686.0,
+                    1347.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    739.0,
+                    1328.0,
+                    759.0,
+                    1328.0,
+                    759.0,
+                    1348.0,
+                    739.0,
+                    1348.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    807.0,
+                    1327.0,
+                    838.0,
+                    1327.0,
+                    838.0,
+                    1350.0,
+                    807.0,
+                    1350.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    888.0,
+                    1330.0,
+                    902.0,
+                    1330.0,
+                    902.0,
+                    1344.0,
+                    888.0,
+                    1344.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    955.0,
+                    1327.0,
+                    985.0,
+                    1327.0,
+                    985.0,
+                    1350.0,
+                    955.0,
+                    1350.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1032.0,
+                    1328.0,
+                    1050.0,
+                    1328.0,
+                    1050.0,
+                    1348.0,
+                    1032.0,
+                    1348.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1088.0,
+                    1333.0,
+                    1102.0,
+                    1333.0,
+                    1102.0,
+                    1348.0,
+                    1088.0,
+                    1348.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1155.0,
+                    1328.0,
+                    1185.0,
+                    1328.0,
+                    1185.0,
+                    1351.0,
+                    1155.0,
+                    1351.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1303.0,
+                    1328.0,
+                    1334.0,
+                    1328.0,
+                    1334.0,
+                    1351.0,
+                    1303.0,
+                    1351.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1384.0,
+                    1330.0,
+                    1400.0,
+                    1330.0,
+                    1400.0,
+                    1350.0,
+                    1384.0,
+                    1350.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    805.0,
+                    1360.0,
+                    991.0,
+                    1360.0,
+                    991.0,
+                    1399.0,
+                    805.0,
+                    1399.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    493.0,
+                    1519.0,
+                    579.0,
+                    1519.0,
+                    579.0,
+                    1554.0,
+                    493.0,
+                    1554.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    849.0,
+                    1519.0,
+                    949.0,
+                    1519.0,
+                    949.0,
+                    1554.0,
+                    849.0,
+                    1554.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1200.0,
+                    1519.0,
+                    1290.0,
+                    1519.0,
+                    1290.0,
+                    1555.0,
+                    1200.0,
+                    1555.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    380.0,
+                    1550.0,
+                    398.0,
+                    1550.0,
+                    398.0,
+                    1568.0,
+                    380.0,
+                    1568.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    724.0,
+                    1556.0,
+                    759.0,
+                    1556.0,
+                    759.0,
+                    1579.0,
+                    724.0,
+                    1579.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1079.0,
+                    1556.0,
+                    1094.0,
+                    1556.0,
+                    1094.0,
+                    1570.0,
+                    1079.0,
+                    1570.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    735.0,
+                    1588.0,
+                    754.0,
+                    1588.0,
+                    754.0,
+                    1607.0,
+                    735.0,
+                    1607.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    304.0,
+                    1602.0,
+                    337.0,
+                    1602.0,
+                    337.0,
+                    1763.0,
+                    304.0,
+                    1763.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    378.0,
+                    1604.0,
+                    398.0,
+                    1604.0,
+                    398.0,
+                    1625.0,
+                    378.0,
+                    1625.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    403.0,
+                    1616.0,
+                    418.0,
+                    1616.0,
+                    418.0,
+                    1630.0,
+                    403.0,
+                    1630.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    724.0,
+                    1616.0,
+                    756.0,
+                    1616.0,
+                    756.0,
+                    1639.0,
+                    724.0,
+                    1639.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1077.0,
+                    1608.0,
+                    1096.0,
+                    1608.0,
+                    1096.0,
+                    1628.0,
+                    1077.0,
+                    1628.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    585.0,
+                    1645.0,
+                    597.0,
+                    1645.0,
+                    597.0,
+                    1659.0,
+                    585.0,
+                    1659.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    735.0,
+                    1650.0,
+                    754.0,
+                    1650.0,
+                    754.0,
+                    1669.0,
+                    735.0,
+                    1669.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    380.0,
+                    1660.0,
+                    398.0,
+                    1660.0,
+                    398.0,
+                    1680.0,
+                    380.0,
+                    1680.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    583.0,
+                    1662.0,
+                    606.0,
+                    1662.0,
+                    606.0,
+                    1679.0,
+                    583.0,
+                    1679.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1077.0,
+                    1663.0,
+                    1096.0,
+                    1663.0,
+                    1096.0,
+                    1683.0,
+                    1077.0,
+                    1683.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    571.0,
+                    1686.0,
+                    597.0,
+                    1686.0,
+                    597.0,
+                    1705.0,
+                    571.0,
+                    1705.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    729.0,
+                    1680.0,
+                    754.0,
+                    1680.0,
+                    754.0,
+                    1698.0,
+                    729.0,
+                    1698.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    918.0,
+                    1683.0,
+                    944.0,
+                    1683.0,
+                    944.0,
+                    1717.0,
+                    918.0,
+                    1717.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1294.0,
+                    1697.0,
+                    1306.0,
+                    1697.0,
+                    1306.0,
+                    1708.0,
+                    1294.0,
+                    1708.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1318.0,
+                    1683.0,
+                    1332.0,
+                    1683.0,
+                    1332.0,
+                    1695.0,
+                    1318.0,
+                    1695.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    739.0,
+                    1712.0,
+                    754.0,
+                    1712.0,
+                    754.0,
+                    1726.0,
+                    739.0,
+                    1726.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1227.0,
+                    1714.0,
+                    1240.0,
+                    1714.0,
+                    1240.0,
+                    1724.0,
+                    1227.0,
+                    1724.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1282.0,
+                    1708.0,
+                    1296.0,
+                    1708.0,
+                    1296.0,
+                    1721.0,
+                    1282.0,
+                    1721.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    629.0,
+                    1721.0,
+                    639.0,
+                    1721.0,
+                    639.0,
+                    1734.0,
+                    629.0,
+                    1734.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1077.0,
+                    1718.0,
+                    1096.0,
+                    1718.0,
+                    1096.0,
+                    1738.0,
+                    1077.0,
+                    1738.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    604.0,
+                    1749.0,
+                    612.0,
+                    1749.0,
+                    612.0,
+                    1757.0,
+                    604.0,
+                    1757.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    726.0,
+                    1738.0,
+                    756.0,
+                    1738.0,
+                    756.0,
+                    1761.0,
+                    726.0,
+                    1761.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    930.0,
+                    1732.0,
+                    950.0,
+                    1732.0,
+                    950.0,
+                    1749.0,
+                    930.0,
+                    1749.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1244.0,
+                    1729.0,
+                    1256.0,
+                    1729.0,
+                    1256.0,
+                    1744.0,
+                    1244.0,
+                    1744.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    377.0,
+                    1767.0,
+                    421.0,
+                    1767.0,
+                    421.0,
+                    1793.0,
+                    377.0,
+                    1793.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    436.0,
+                    1767.0,
+                    483.0,
+                    1767.0,
+                    483.0,
+                    1787.0,
+                    436.0,
+                    1787.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    556.0,
+                    1761.0,
+                    568.0,
+                    1761.0,
+                    568.0,
+                    1772.0,
+                    556.0,
+                    1772.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    609.0,
+                    1764.0,
+                    619.0,
+                    1764.0,
+                    619.0,
+                    1776.0,
+                    609.0,
+                    1776.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    773.0,
+                    1764.0,
+                    785.0,
+                    1764.0,
+                    785.0,
+                    1776.0,
+                    773.0,
+                    1776.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    791.0,
+                    1760.0,
+                    844.0,
+                    1760.0,
+                    844.0,
+                    1786.0,
+                    791.0,
+                    1786.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1288.0,
+                    1760.0,
+                    1305.0,
+                    1760.0,
+                    1305.0,
+                    1772.0,
+                    1288.0,
+                    1772.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    391.0,
+                    1786.0,
+                    409.0,
+                    1786.0,
+                    409.0,
+                    1805.0,
+                    391.0,
+                    1805.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    442.0,
+                    1784.0,
+                    472.0,
+                    1784.0,
+                    472.0,
+                    1807.0,
+                    442.0,
+                    1807.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    501.0,
+                    1784.0,
+                    533.0,
+                    1784.0,
+                    533.0,
+                    1807.0,
+                    501.0,
+                    1807.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    559.0,
+                    1784.0,
+                    591.0,
+                    1784.0,
+                    591.0,
+                    1807.0,
+                    559.0,
+                    1807.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    674.0,
+                    1781.0,
+                    710.0,
+                    1781.0,
+                    710.0,
+                    1809.0,
+                    674.0,
+                    1809.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    745.0,
+                    1786.0,
+                    765.0,
+                    1786.0,
+                    765.0,
+                    1805.0,
+                    745.0,
+                    1805.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    797.0,
+                    1784.0,
+                    827.0,
+                    1784.0,
+                    827.0,
+                    1807.0,
+                    797.0,
+                    1807.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    853.0,
+                    1784.0,
+                    885.0,
+                    1784.0,
+                    885.0,
+                    1807.0,
+                    853.0,
+                    1807.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    909.0,
+                    1784.0,
+                    941.0,
+                    1784.0,
+                    941.0,
+                    1807.0,
+                    909.0,
+                    1807.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    967.0,
+                    1784.0,
+                    997.0,
+                    1784.0,
+                    997.0,
+                    1807.0,
+                    967.0,
+                    1807.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1023.0,
+                    1783.0,
+                    1055.0,
+                    1783.0,
+                    1055.0,
+                    1807.0,
+                    1023.0,
+                    1807.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1086.0,
+                    1789.0,
+                    1106.0,
+                    1789.0,
+                    1106.0,
+                    1809.0,
+                    1086.0,
+                    1809.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1138.0,
+                    1786.0,
+                    1170.0,
+                    1786.0,
+                    1170.0,
+                    1810.0,
+                    1138.0,
+                    1810.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1197.0,
+                    1787.0,
+                    1229.0,
+                    1787.0,
+                    1229.0,
+                    1810.0,
+                    1197.0,
+                    1810.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1255.0,
+                    1787.0,
+                    1288.0,
+                    1787.0,
+                    1288.0,
+                    1810.0,
+                    1255.0,
+                    1810.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1314.0,
+                    1787.0,
+                    1346.0,
+                    1787.0,
+                    1346.0,
+                    1810.0,
+                    1314.0,
+                    1810.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1373.0,
+                    1786.0,
+                    1405.0,
+                    1786.0,
+                    1405.0,
+                    1810.0,
+                    1373.0,
+                    1810.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    816.0,
+                    1827.0,
+                    981.0,
+                    1827.0,
+                    981.0,
+                    1865.0,
+                    816.0,
+                    1865.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1151.0,
+                    624.0,
+                    1161.0,
+                    624.0,
+                    1161.0,
+                    659.0,
+                    1151.0,
+                    659.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    413.25,
+                    659.5,
+                    443.25,
+                    659.5,
+                    443.25,
+                    678.0,
+                    413.25,
+                    678.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    423.25,
+                    692.5,
+                    496.25,
+                    692.5,
+                    496.25,
+                    719.0,
+                    423.25,
+                    719.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1171.0,
+                    690.0,
+                    1211.0,
+                    690.0,
+                    1211.0,
+                    707.0,
+                    1171.0,
+                    707.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1190.0,
+                    671.5,
+                    1223.0,
+                    671.5,
+                    1223.0,
+                    688.5,
+                    1190.0,
+                    688.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1173.75,
+                    701.5,
+                    1184.75,
+                    701.5,
+                    1184.75,
+                    717.5,
+                    1173.75,
+                    717.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    797.25,
+                    708.5,
+                    808.25,
+                    708.5,
+                    808.25,
+                    729.0,
+                    797.25,
+                    729.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1220.25,
+                    693.0,
+                    1266.25,
+                    693.0,
+                    1266.25,
+                    714.5,
+                    1220.25,
+                    714.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    760.0,
+                    697.5,
+                    797.0,
+                    697.5,
+                    797.0,
+                    714.5,
+                    760.0,
+                    714.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    444.0,
+                    801.5,
+                    465.0,
+                    801.5,
+                    465.0,
+                    814.0,
+                    444.0,
+                    814.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    432.25,
+                    1197.0,
+                    473.25,
+                    1197.0,
+                    473.25,
+                    1216.5,
+                    432.25,
+                    1216.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    601.25,
+                    1261.0,
+                    649.25,
+                    1261.0,
+                    649.25,
+                    1290.0,
+                    601.25,
+                    1290.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    839.25,
+                    1280.0,
+                    856.25,
+                    1280.0,
+                    856.25,
+                    1314.5,
+                    839.25,
+                    1314.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    501.25,
+                    1287.5,
+                    540.25,
+                    1287.5,
+                    540.25,
+                    1312.5,
+                    501.25,
+                    1312.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    726.0,
+                    1310.0,
+                    753.0,
+                    1310.0,
+                    753.0,
+                    1330.0,
+                    726.0,
+                    1330.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    614.0,
+                    1674.0,
+                    636.0,
+                    1674.0,
+                    636.0,
+                    1693.0,
+                    614.0,
+                    1693.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1173.0,
+                    1754.5,
+                    1234.0,
+                    1754.5,
+                    1234.0,
+                    1792.5,
+                    1173.0,
+                    1792.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    725.0,
+                    1762.0,
+                    784.0,
+                    1762.0,
+                    784.0,
+                    1784.5,
+                    725.0,
+                    1784.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1064.0,
+                    1764.5,
+                    1124.0,
+                    1764.5,
+                    1124.0,
+                    1791.5,
+                    1064.0,
+                    1791.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1254.75,
+                    1765.0,
+                    1297.75,
+                    1765.0,
+                    1297.75,
+                    1787.5,
+                    1254.75,
+                    1787.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    615.0,
+                    1782.5,
+                    651.0,
+                    1782.5,
+                    651.0,
+                    1805.5,
+                    615.0,
+                    1805.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    73.0,
+                    817.0,
+                    73.0,
+                    817.0,
+                    108.0,
+                    296.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    832.0,
+                    2085.0,
+                    870.0,
+                    2085.0,
+                    870.0,
+                    2122.0,
+                    832.0,
+                    2122.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    227.0,
+                    1332.0,
+                    227.0,
+                    1332.0,
+                    267.0,
+                    292.0,
+                    267.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    297.0,
+                    1404.0,
+                    297.0,
+                    1404.0,
+                    331.0,
+                    296.0,
+                    331.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    327.0,
+                    1406.0,
+                    327.0,
+                    1406.0,
+                    361.0,
+                    294.0,
+                    361.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    358.0,
+                    1405.0,
+                    358.0,
+                    1405.0,
+                    392.0,
+                    295.0,
+                    392.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    388.0,
+                    1404.0,
+                    388.0,
+                    1404.0,
+                    422.0,
+                    295.0,
+                    422.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    420.0,
+                    1405.0,
+                    420.0,
+                    1405.0,
+                    453.0,
+                    295.0,
+                    453.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    447.0,
+                    1404.0,
+                    447.0,
+                    1404.0,
+                    482.0,
+                    294.0,
+                    482.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    478.0,
+                    1402.0,
+                    478.0,
+                    1402.0,
+                    514.0,
+                    294.0,
+                    514.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    510.0,
+                    1096.0,
+                    510.0,
+                    1096.0,
+                    544.0,
+                    294.0,
+                    544.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 17,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    295,
+                    1404,
+                    295,
+                    1404,
+                    510,
+                    298,
+                    510
+                ],
+                "score": 0.978
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    299,
+                    75,
+                    815,
+                    75,
+                    815,
+                    105,
+                    299,
+                    105
+                ],
+                "score": 0.859
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    837,
+                    2089,
+                    865,
+                    2089,
+                    865,
+                    2112,
+                    837,
+                    2112
+                ],
+                "score": 0.833
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    301,
+                    227,
+                    817,
+                    227,
+                    817,
+                    261,
+                    301,
+                    261
+                ],
+                "score": 0.785
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    301,
+                    227,
+                    817,
+                    227,
+                    817,
+                    261,
+                    301,
+                    261
+                ],
+                "score": 0.094
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1091,
+                    418,
+                    1144,
+                    418,
+                    1144,
+                    451,
+                    1091,
+                    451
+                ],
+                "score": 0.91,
+                "latex": "E / 3"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    650,
+                    418,
+                    717,
+                    418,
+                    717,
+                    451,
+                    650,
+                    451
+                ],
+                "score": 0.91,
+                "latex": "E / 1 0"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    355,
+                    358,
+                    407,
+                    358,
+                    407,
+                    388,
+                    355,
+                    388
+                ],
+                "score": 0.9,
+                "latex": "R _ { L 2 }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1334,
+                    389,
+                    1393,
+                    389,
+                    1393,
+                    419,
+                    1334,
+                    419
+                ],
+                "score": 0.9,
+                "latex": "R _ { F R }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1205,
+                    358,
+                    1257,
+                    358,
+                    1257,
+                    388,
+                    1205,
+                    388
+                ],
+                "score": 0.89,
+                "latex": "R _ { L 2 }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1264,
+                    298,
+                    1324,
+                    298,
+                    1324,
+                    328,
+                    1264,
+                    328
+                ],
+                "score": 0.89,
+                "latex": "R _ { F R }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    767,
+                    359,
+                    792,
+                    359,
+                    792,
+                    385,
+                    767,
+                    385
+                ],
+                "score": 0.81,
+                "latex": "E"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1110,
+                    297,
+                    1203,
+                    297,
+                    1203,
+                    328,
+                    1110,
+                    328
+                ],
+                "score": 0.76,
+                "latex": "( E , R _ { L 2 }"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    73.0,
+                    818.0,
+                    73.0,
+                    818.0,
+                    108.0,
+                    295.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    832.0,
+                    2086.0,
+                    869.0,
+                    2086.0,
+                    869.0,
+                    2125.0,
+                    832.0,
+                    2125.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    225.0,
+                    824.0,
+                    225.0,
+                    824.0,
+                    267.0,
+                    293.0,
+                    267.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    297.0,
+                    1109.0,
+                    297.0,
+                    1109.0,
+                    331.0,
+                    296.0,
+                    331.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1204.0,
+                    297.0,
+                    1263.0,
+                    297.0,
+                    1263.0,
+                    331.0,
+                    1204.0,
+                    331.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1325.0,
+                    297.0,
+                    1404.0,
+                    297.0,
+                    1404.0,
+                    331.0,
+                    1325.0,
+                    331.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    327.0,
+                    1404.0,
+                    327.0,
+                    1404.0,
+                    362.0,
+                    296.0,
+                    362.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    354.0,
+                    354.0,
+                    354.0,
+                    354.0,
+                    395.0,
+                    292.0,
+                    395.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    408.0,
+                    354.0,
+                    766.0,
+                    354.0,
+                    766.0,
+                    395.0,
+                    408.0,
+                    395.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    793.0,
+                    354.0,
+                    1204.0,
+                    354.0,
+                    1204.0,
+                    395.0,
+                    793.0,
+                    395.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1258.0,
+                    354.0,
+                    1405.0,
+                    354.0,
+                    1405.0,
+                    395.0,
+                    1258.0,
+                    395.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    383.0,
+                    1333.0,
+                    383.0,
+                    1333.0,
+                    427.0,
+                    292.0,
+                    427.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1394.0,
+                    383.0,
+                    1405.0,
+                    383.0,
+                    1405.0,
+                    427.0,
+                    1394.0,
+                    427.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    416.0,
+                    649.0,
+                    416.0,
+                    649.0,
+                    454.0,
+                    293.0,
+                    454.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    718.0,
+                    416.0,
+                    1090.0,
+                    416.0,
+                    1090.0,
+                    454.0,
+                    718.0,
+                    454.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1145.0,
+                    416.0,
+                    1402.0,
+                    416.0,
+                    1402.0,
+                    454.0,
+                    1145.0,
+                    454.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    446.0,
+                    1405.0,
+                    446.0,
+                    1405.0,
+                    486.0,
+                    293.0,
+                    486.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    480.0,
+                    423.0,
+                    480.0,
+                    423.0,
+                    512.0,
+                    294.0,
+                    512.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    225.0,
+                    824.0,
+                    225.0,
+                    824.0,
+                    267.0,
+                    293.0,
+                    267.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 18,
+            "width": 1700,
+            "height": 2200
+        }
+    }
+]
\ No newline at end of file
diff --git a/parse/train/B1em9h4KDS/B1em9h4KDS.md b/parse/train/B1em9h4KDS/B1em9h4KDS.md
new file mode 100644
index 0000000000000000000000000000000000000000..fdb8cbd08a47a6a5a97f718ca4a9bf8e081ba7b1
--- /dev/null
+++ b/parse/train/B1em9h4KDS/B1em9h4KDS.md
@@ -0,0 +1,388 @@
+# GENERATIVE IMPUTATION AND STOCHASTIC PREDICTION
+
+Anonymous authors Paper under double-blind review
+
+# ABSTRACT
+
+In many machine learning applications, we are faced with incomplete datasets. In the literature, missing data imputation techniques have been mostly concerned with filling missing values. However, the existence of missing values is synonymous with uncertainties not only over the distribution of missing values but also over target class assignments that require careful consideration. In this paper, we propose a simple and effective method for imputing missing features and estimating the distribution of target assignments given incomplete data. In order to make imputations, we train a simple and effective generator network to generate imputations that a discriminator network is tasked to distinguish. Following this, a predictor network is trained using the imputed samples from the generator network to capture the classification uncertainties and make predictions accordingly. The proposed method is evaluated on CIFAR-10 image dataset as well as three real-world tabular classification datasets, under different missingness rates and structures. Our experimental results show the effectiveness of the proposed method in generating imputations as well as providing estimates for the class uncertainties in a classification task when faced with missing values.
+
+# 1 INTRODUCTION
+
+While a large body of the machine learning literature is built upon the assumption of having access to complete datasets, in many real-world problems only incomplete datasets are available. The existence of missing values can be due to many different causes such as human subjects not adhering to certain questions or features not being collected frequently due to financial or experimental limitations, sensors failures, and so forth. Data imputation techniques have been suggested as a solution to bridge this gap in the literature by replacing missing values with observed values.
+
+Missing data imputation approaches can be categorized into single and multiple imputation methods. Single imputation methods try to replace each missing value with a plausible value that is the best fit given the value of other correlated features and knowledge extracted from the dataset (Hastie et al., 1999; Anderson, 1957). While these methods are easy to implement and use in practice, imputed values may induce bias by eliminating less likely but important values. Also, these methods do not suggest a way to measure to what extent the imputed values are representative of the missing values (Little & Rubin, 2019).
+
+Multiple imputation (MI) techniques, as suggested by the name, try to use multiple imputed values to impute each missing value. The result would be having a set of imputed datasets that enables measuring how consistent and statistically significant are the results of the experiments (Rubin, 1976). While MI offers interesting statistical insights about the reliability of analysis on incomplete data, the insight is imprecise as it is mainly concerned about the population of data samples rather than individual instances. Specifically, MI methods reason about the statistical properties on a limited number of imputed datasets (less than 10 in most practical implementations) on the population of samples within the dataset (Schafer & Graham, 2002; Murray et al., 2018).
+
+The existence of missing values is synonymous with having uncertainty over these values that requires careful consideration. In many real-world applications, we are dealing with supervised problems that demand modeling and prediction based on incomplete data. Take for instance, prediction of class assignments given an image in which a large portion of the frame is missing. In such a scenario, based on observed frame parts, there might be multiple probable class assignments each having a different likelihood. Here, we are not only interested in imputing missing values or measuring how robust our imputations are, but also it is highly desirable to measure the impact of missing values on the prediction outcome for each instance.
+
+In this paper, we propose the idea of Generative Imputation and Stochastic Prediction (GI) as a novel approach to impute missing values and to measure class uncertainties arising from the distribution of missing values. The suggested approach is based on neural networks trained using an adversarial objective function. Additionally, a predictor is trained on the generated samples from the imputer network which is able to reflect the impact of uncertainties over missing values. This enables measuring different prediction outcomes and certainties for each specific instance. We evaluate the effectiveness of the proposed method on different incomplete image and tabular datasets under various missingness structures.
+
+# 2 RELATED WORK
+
+One of the simplest traditional methods for handling missing values includes imputing the occurrences of missing values with constant values such as zeros or using mean values. To enhance the accuracy of such imputations, alternatives such as $\mathbf { k }$ -nearest neighbors (KNN) (Hastie et al., 1999) and maximum likelihood estimation (MLE) (Anderson, 1957) have been suggested to estimate values to be used given an observed context. While these methods are easy to implement and analyze, they often fail to capture the complex feature dependencies as well as structures present in many problems.
+
+Rubin (1976) suggested a categorization for missingness mechanisms: missing completely at random (MCAR), missing at random (MAR), and missing not at random (MNAR). Under the assumption of MAR, the authors suggested multiple imputation (MI) as a stochastic imputation method. Here, instead of imputing missing values using a single value, several values are sampled to represent the distribution over the missing value. MI generates a few imputed complete datasets that are then used independently in statistical modeling (Schafer & Graham, 2002; Little & Rubin, 2019). Usually, the final goal of MI is to measure the robustness of the final statistical analysis amongst the imputed datasets. In other words, it measures the quality of imputations and the statistical significance of analysis on the imputed data. It should be noted that the number of imputations used in MI is usually very limited. Also, often strong simplifying assumptions are made in modeling the data distribution (e.g., multi-variate Gaussian or Student’s t distribution) which limit the applicability of this method (Schafer & Graham, 2002; Murray et al., 2018).
+
+More recently, autoencoder architectures have been suggested as powerful density estimators capable of capturing complex distributions. Perhaps, denoising autoencoders (DAE) (Vincent et al., 2008) are one of the most intuitive approaches in which a neural network is trained to reconstruct and denoise its input. Following a more probabilistic perspective, variational autoencoders (VAE) (Kingma & Welling, 2013) try to learn the data generating distribution via a latent representation. Specifically, conditional variational autoencoders (CVAE) (Sohn et al., 2015) can be used to sample missing values conditioned on observed values. For instance, Mattei & Frellsen (2018) suggested a method based on deep latent variable models and importance sampling that offers a tighter likelihood bound compared to the standard VAE bound. While these methods are powerful generative models applicable to missing data imputation, often samples generated using autoencoders are biased toward the mode of the distribution (e.g., resulting in blurry images, for vision tasks) (Goodfellow et al., 2014; Dumoulin et al., 2016).
+
+Recently, due to the success of generative adversarial networks (GAN), there has been great attention toward applying them to impute missing values. For instance, Yoon et al. (2018) suggested an imputation method based on adversarial and reconstruction loss terms. Li et al. (2019) introduced the idea of using separate generator and discriminator networks to learn the missing data structure and data distribution. These methods have been quite successful and are able to present the state-of-the-art results. Though it should be noted that often the presence of additional loss terms may bias the generated samples toward the mode of the distribution being modeled. Also, these methods are often complicated to be applied in practical setups by practitioners. For instance, Yoon et al. (2018) requires setting hyperparameters to adjust the influence of an MSE loss term as well as the rate of discriminator hint vectors. Also as another example, Li et al. (2019) uses three generators and three discriminators for the final imputer architecture.
+
+![](images/c5713bc074e138a2a20d050db7edf84f541cc471e67603e24e6900e610f80f08.jpg)  
+Figure 1: Block diagram of the proposed adversarial imputation method. $h$ represents the blending function of (1), and $L$ is the adversarial loss function of (2).
+
+From the perspective of supervised analysis, imputation and handling missing values are usually considered as a preprocessing step. A few exceptions exist such as Bayesian models and decision trees that permit direct analysis on incomplete data (Nielsen & Jensen, 2009; Zhang et al., 2005). Note that while certain Bayesian methods such as probabilistic Bayesian networks allow handling of missing values as unobserved variables. However, given an incomplete training dataset and without any known causal structure as a priori, learning such models is a very challenging problem with the complexity of at least NP-complete to learn the network architecture in addition to an iterative EM optimization to learn model parameters (Darwiche, 2009; Neapolitan et al., 2004). We argue that the simplistic approach of imputing missing values as a preprocessing step discards uncertainties that exist in original incomplete data samples. Instead, there is a need for methods that reflect these uncertainties on the final predicted target distribution. This work suggests the idea of training a predictor on different imputed samples to capture the uncertainties over class assignments. Compared to MI, the suggested method interleaves imputation and training a downstream prediction model, enabling to estimate classification uncertainties for each instance.
+
+# 3 PROPOSED METHOD
+
+# 3.1 PROBLEM DEFINITION
+
+In this paper, we make the general assumption of having access to an incomplete dataset $\mathcal { D }$ consisting of a set of feature vector, mask vector, and target class pairs $( { \pmb x } _ { i } , { \pmb k } _ { i } , y _ { i } )$ . For each feature vector, $\pmb { x } _ { i } \in \mathbb { R } ^ { d }$ , only a subset of the features is available. The mask vector $\pmb { k } _ { i } \in \{ 0 , 1 \} ^ { d }$ is used to indicate available features and missing features by ones and zeros, respectively. Here, to represent features as fixed-width vectors, arbitrary (or NaN) values are used to fill missing values. Also, for convenience, we often use $\pmb { x } _ { i } ^ { o b s }$ and $\pmb { x } _ { i } ^ { m i s s }$ to refer to the set of observed and missing features for the feature vector $\mathbf { \Delta } _ { \mathbf { \mathcal { X } } _ { i } }$ .
+
+We define our objective in two steps: $( i )$ Imputing missing values via sampling from the conditional distribution of missing features given observed features i.e., $P ( \pmb { x } _ { i } ^ { m i s s } | \pmb { x } _ { i } ^ { o b s } )$ . $( i i )$ Estimating the distribution of target classes given the observed features and the distribution of missing features i.e., $P ( y | \mathbf { x } _ { i } ^ { o b s } , \mathbf { x } _ { i } ^ { m i s s } )$ . For the first part, we are interested in sampling from the conditional distribution rather than finding the mode of the distribution as the most probable imputation. Similarly, for the second part, we are interested in obtaining a distribution over the possible target assignments and the confidence of each class rather than maximum likelihood class assignments.
+
+# 3.2 GENERATIVE IMPUTATION
+
+To generate samples from the distribution of missing features conditioned on the observed features, we follow the idea first suggested by Yoon et al. (2018). In this paradigm, a generator network is responsible for generating imputations while a discriminator is trying to distinguish imputed features from observed features (see Figure 1).
+
+Specifically, the generator function $G ( \pmb { x } _ { i } , \pmb { k } _ { i } , z ) \in \mathbb { R } ^ { d }$ generates an imputed feature vector, based on observed features, the corresponding mask, and a Gaussian noise vector $( z )$ . Note that, in order to achieve the final imputed vector, $\widehat { \pmb { x } } _ { i }$ , we blend (or, merge) the output of the generator with the input features to replace generated values with the exact values of observed features:
+
+$$
+\begin{array} { r } { \widehat { \pmb { x } } _ { i , j } = \left\{ \begin{array} { l l } { \pmb { x } _ { i , j } } & { \mathrm { i f } ~ \pmb { k } _ { i , j } = 1 } \\ { G ( \pmb { x } _ { i } , \pmb { k } _ { i } , z ) _ { j } } & { \mathrm { i f } ~ \pmb { k } _ { i , j } = 0 } \end{array} \right. , } \end{array}
+$$
+
+where $\mathbf { \delta } _ { \mathbf { x } _ { i , j } }$ refers to $j ^ { : }$ ’th feature of sample $i$ . Also, note that by sampling $_ { z }$ multiple times, we can obtain different imputation samples from the conditional distribution indicated by $\widehat { \pmb { x } } _ { i } ^ { l }$ where $l$ is the sample number.
+
+A discriminator network, $D ( \widehat { \mathbf { x } } _ { i } )$ , is trained to distinguish real and imputed features by generating a predicted softmax mask output, $\widehat { \pmb { k } } _ { i }$ . Here a binary cross-entropy loss per mask element is used as the adversarial objective function:
+
+$$
+\operatorname* { m a x } _ { G } \operatorname* { m i n } _ { D } L ( G , D ) = \mathbb { E } _ { k \sim \mathcal { D } , \widehat { k } \sim D ( G ( x , k , z ) ) } \ [ k ^ { T } \log ( \widehat { k } ) \ + \ ( 1 - k ) ^ { T } \log ( 1 - \widehat { k } ) ] .
+$$
+
+The intuition behind this adversarial loss function is that given a generator function which captures the data distribution successfully, the discriminator would not be able to distinguish the parts of the feature vector that were originally missing.
+
+Compared to Yoon et al. (2018), the objective function of (2) does not have an MSE loss term. Instead, we use recent advances in GAN stabilization and training to improve the training process (see Section 3.4). While it is quite prevalent in the adversarial learning literature to use additional loss terms such as mean squared error (MSE) to enhance the quality of generated samples, we decided to keep our solution as simple as possible. Additionally, in our experiments, we provide supporting evidence that this simple loss function enables us to sample from the conditional distribution and prevents biased inclinations toward distribution modes.
+
+# 3.3 STOCHASTIC PREDICTION
+
+To capture the distribution of target classes given incomplete data, we suggest the idea of stochastic prediction. As indicated in the previous section, the generator can be used to sample from the conditional distribution. Here, a predictor is trained based on the imputed samples to predict class assignments and to calculate the confidence of these assignments. For instance, for a specific test sample at hand, if a certain missing feature is a strong indicator of the target class, we would like to observe the impact of different imputations for that feature on the final hypothesis.
+
+Formally, we are interested in finding the certainty of class assignments given observed features:
+
+$$
+\Psi = P ( y | \mathbf { x } _ { i } ^ { o b s } ) .
+$$
+
+Here, $\Psi$ is a vector where each element is representing a certain class. Rewriting (3) as a marginal we have:
+
+$$
+\Psi = \int P ( \pmb { x } _ { i } ^ { m i s s } ) P ( y | \pmb { x } _ { i } ^ { o b s } , \pmb { x } _ { i } ^ { m i s s } ) d \pmb { x } _ { i } ^ { m i s s } .
+$$
+
+Approximating the integration using a summation, given enough samples, $\Psi$ can be estimated by:
+
+$$
+\Psi \approx \frac { 1 } { N } \sum P ( y | \pmb { x } _ { i } ^ { o b s } , \widehat { \pmb { x } } _ { i } ^ { m i s s } ) ,
+$$
+
+where $\widehat { \pmb { x } } _ { i } ^ { m i s s }$ are samples taken from the conditional distribution of missing features given observed bones. We use the suggested generative imputation method to generate samples required for this approximation. Rewriting (1) using Hadamard product and as function of the noise vector:
+
+$$
+\widehat { \mathbf { x } } _ { i } = \pmb { k } _ { i } \odot \pmb { x } _ { i } + \left( 1 - \pmb { k } _ { i } \right) \odot G ( \pmb { x } _ { i } , \pmb { k } _ { i } , z )
+$$
+
+Assuming that a predictor, $F _ { \theta }$ , is available which predicts class assignments for a complete feature vector, $\Psi$ can be estimated as:
+
+$$
+\Psi = \mathbb { E } _ { \mathbf { z } } [ F _ { \boldsymbol { \theta } } ( \widehat { \mathbf { x } } _ { i } ) ] \approx \frac { 1 } { N } { \sum _ { l = 1 } ^ { N } } F _ { \boldsymbol { \theta } } ( \widehat { \mathbf { x } } _ { i } ^ { l } ) ~ .
+$$
+
+Algorithm 1 presents the suggested algorithm for training the predictor. It consists of taking samples from the incomplete dataset, then imputing them using our generator network, and using the imputed
+
+<table><tr><td>Algorithm1: Training the predictor.</td></tr><tr><td>Input: G (trained imputer), D (dataset) Output: Fθ (trained predictor) foreachTrainingEpoch do foreach(xi,ki,yi) in D do z ~ N(0,I) Xi ←ki①xi+(1-ki)OG(xi,ki,z) loss ← L(yi,yred) Backpropagate loss</td></tr></table>
+
+Algorithm 2: Estimating target distributions.   
+Input: $F _ { \theta }$ (trained predictor), $( x , k )$ (test sample), $\mathbf { N }$ (ensemble samples)   
+Output: $\Psi$ (distribution over target classes)   
+Ψ ← zeros ∈ R#classes   
+foreach Ensemble Sample $^ { l }$ to $N$ do z ∼ N (0, I) $\widehat { \mathbf { x } }  k \odot x + ( 1 - k ) \odot G ( { \pmb { x } } , { \pmb { k } } , z )$ bypred ← Fθ(x) bj ← argmax(ypred) Ψj ← Ψj + 1N
+
+samples to update the predictor. Note that, on each epoch and for each sample, the generator generates a new sample from the conditional distribution. Intuitively, it means that the predictor observes and learns to operate under different imputations for a given sample. This is different from approaches such as multiple imputation where several predictors are trained on different imputed versions of a dataset.
+
+Algorithm 2 presents the suggested algorithm for making predictions and estimating target distributions given a trained predictor model. Here, a sample is imputed $N$ times and inference on this set results in an ensemble of predictions over different imputations. The output of this algorithm can be interpreted as a distribution over the confidence of class assignments given a partially observed test sample. The following claims justify the validity of Algorithm 1and Algorithm 2.
+
+Claim 1. (Generalization of the predictor). If we assume imputed $\widehat { x _ { i } s }$ are samples from the underlying bfeature distribution, then the assigned training set labels can be modeled as labels generated from a noisy labeling process.
+
+Claim 1 permits the analysis of the generalization and convergence for the predictor trained using Algorithm 1 based on current literature in training models with noisy labels (Natarajan et al., 2013; Reed et al., 2014; Chen et al., 2019). From the analysis provided by Chen et al. (2019), test accuracy in asymmetric label noise conditions is a quadratic function of the label noise:
+
+$$
+P ( y _ { i } = \widehat { y _ { i } } ) = ( 1 - \epsilon ) ^ { 2 } + \epsilon ^ { 2 } ,
+$$
+
+where $\widehat { y _ { i } }$ is underlying true label for the imputed feature vector $( \widehat { x _ { i } } )$ , and $y _ { i }$ is the label provided by bthe incomplete dataset. In (8), label noise ratio, $\epsilon$ b, represents the probability of the label transition from a certain target class to another:
+
+$$
+\epsilon = 1 - P ( \widehat { y _ { i } } = j | y _ { i } = j ) .
+$$
+
+In practice, $\epsilon$ is determined by the problem-specific underlying data distribution as well as the distribution of missing values.
+
+Justification for claim 1 is straightforward, assume that $\{ \widehat { y _ { i } } ^ { 1 } \ldots \widehat { y _ { i } } ^ { N } \}$ are underlying true labels for each of $\{ \widehat { x _ { i } } ^ { 1 } \ldots \widehat { x _ { i } } ^ { N } \}$ b b. During training, for any imputed sample in $\{ \widehat { x _ { i } } ^ { 1 } \ldots \widehat { x _ { i } } ^ { N } \}$ , we use the dataset bprovided label, $y$ b b, to calculate the loss and to update model parameters. In the case that any of $\{ \widehat { y _ { i } } ^ { 1 } \ldots \widehat { y _ { i } } ^ { N } \}$ is different from $y$ , the loss term corresponding to that term would be calculated using b ba wrong label. Here, if we consider the average impact on gradients for batches of samples rather than individual cases, the overall impact on training would be very similar to the case of training using noisy labels.
+
+Claim 2. (Approximation of the target distribution). If we assume: (i) imputed $\widehat { x _ { i } s }$ are valid samples from the underlying feature distribution, b(ii) a good predictor can be trained using the incomplete data, (iii) enough samples are used and the Monte Carlo estimator is unbiased, then the target distribution, $\Psi$ , can be estimated accurately.
+
+This claim supports Algorithm 2 that is suggested to estimate the target distribution given a partially observed feature vector.
+
+The first assumption is consistent with the theoretical analysis of generative adversarial networks that they can converge to the true underlying distribution (Arora et al., 2018; Liu et al., 2017). The second assumption is supported by Claim 1. Regarding the last assumption, each sample requires one forward computation of the generator and predictor networks which, based on the scalability of current network architectures, usually permits thousands of samples to be taken at a reasonable computational cost.
+
+# 3.4 IMPLEMENTATION DETAILS
+
+As we conduct experiments on image and tabular datasets, we use different architectures for each. For image datasets, we used a generator and discriminator architectures similar to the ones suggested by Wang et al. (2018). However, we improved these architectures using self-attention layers (Zhang et al., 2018). It should be noted that, while Zhang et al. (2018) suggests using a single self-attention layer in the middle of the network, we observed consistent improvements by inserting multiple self-attention layers before each residual block within the network. Furthermore, as input to the generator, we concatenate input image, mask, and a random $_ z$ frame along the channels dimension and use it as input. For tabular datasets, we use a simple 4 layer network consisting of fully-connected and batch-norm layers. Also, the input to the generator is the concatenation of a feature vector, mask vector, and a $_ { z }$ vector of size $\frac { 1 } { 8 }$ of the input. For all experiments, we use an ensemble size $( N )$ ) equal to 128.
+
+We used Adam (Kingma & Ba, 2014) for model optimization. Two time-scale update rule (TTUR) (Heusel et al., 2017) was used to balance training the generator and discriminator networks. We explored best TTUR learning-rate settings from the set of {0.001, 0.0005, 0.0001, 0.00005}. Here, Adam parameters $\beta _ { 1 }$ and $\beta _ { 2 }$ are set to 0.5 and 0.999, respectively. Also, spectral normalization was used to stabilize both the generator and discriminator network in our experiments with image data (Miyato et al., 2018). For the predictor network, we used the default Adam settings as suggested by Kingma & Ba (2014). In all training procedures, we decay learning rate by a factor of 5 after reaching a plateau. For all experiments, we use a batch size of 64. Based on our experiments, we found that pretraining the discriminator while fixing the generator network for the first $5 \%$ of the training epochs helps the stability of training.
+
+Further detail on exact architectures, experiments, software dependencies, etc. as well as ablation studies is provided in the appendices.
+
+# 4 EXPERIMENTS
+
+# 4.1 DATASETS
+
+To evaluate the proposed method we use CIFAR-10 (Krizhevsky & Hinton, 2009) as an image classification dataset as well as three non-image datasets: UCI Landsat (Dua & Graff, 2017)2, MIT-BIH arrhythmia (Moody & Mark, 2001), and Diabetes classification (Kachuee et al., 2019) 3. CIFAR-10 dataset consists of $6 0 , 0 0 0 \ 3 2 \mathrm { x } \ 3 2$ images from 10 different classes. For this task, we use train and test sets as provided by the dataset. As a preprocessing step, we normalize pixel values to the range of [0,1] and subtract the mean image. The only data augmentation we use for this task is to randomly flip training images for each batch.
+
+UCI Landsat consists of 6435 samples of 36 features from 6 different categories. We follow the same train and test split as provided by the dataset. MIT-BIH dataset consists of annotated heartbeat signals from which we used the preprocessed version available online4 consisting of 92062 samples of 5 different arrhythmia classes. Diabetes dataset is a real-world health dataset of 92,062 samples and 45 features from different categories such as questionnaire, demographics, medical examination, and lab results. The objective is to classify between three different diabetes conditions i.e., normal, pre-diabetes, and diabetes. As MIT-BIH and Diabetes datasets do not provide explicit train and test sets, we randomly select $80 \%$ of samples as a training set and the rest as a test set. To preprocess our tabular datasets, statistical and unity based normalization are used to balance the variance of different features and center them around zero. Also, while different encoding and representation methods are suggested in the literature to handle categorical features (Jang et al., 2016; Nazabal et al., 2018), in this paper, we take the simple approach of encoding categorical variables using one-hot representation and smoothing them by adding Gaussian noise with zero mean and variance equal to $5 \%$ of feature variances. In our experiments, we observed a reasonable performance using the suggested simple smoothing; however, more advanced encoding methods are also applicable in this setup and can be applied to enhance the performances even further.
+
+# 4.2 MISSINGNESS MECHANISMS
+
+In our experiments, we consider MCAR uniform and MCAR rectangular missingness structures. In MCAR uniform, each feature of each sample is missing based on a Bernoulli distribution with a certain missingness probability (i.e., missing rate) independent of other features. In addition to the case of uniform missingness, for image tasks, we use rectangular missingness/observation structure where rectangular regions of dataset images are missing/observed. To control the rate of missingness and decide on the regions that are missing for each case, we use a latent beta distribution that samples rectangular region’s width and height such that the average missing rate is maintained. For missing rates less than $50 \%$ we make the assumption of having a random rectangular region to be missing, whereas for missing rates more than $50 \%$ we assume that only a random rectangular region is observed and the rest of the image is missing.
+
+We would like to note that while the suggested solution in this paper is readily compatible with MAR structures, in our experiments, to simplify the presentation of results and to have a fair comparison with other work that does not support the MAR assumption, we limited the scope of our experiments to MCAR. Furthermore, to simulate incomplete datasets and to make sure the same features are missing without explicitly storing masks, we use hashed feature vectors to seed random number generators used to sample missing features. More detail is provided in Appendix C.
+
+# 4.3 EVALUATION MEASURES
+
+Fréchet inception distance (FID) (Heusel et al., 2017) score is used to measure the quality of missing data imputation in experiments with images5. We also considered using root means squared error (RMSE); however, we decided not to use this measure as we observed an inconsistent behavior using RMSE in our comparisons as RMSE favors methods that show less variance rather than realistic and sharp samples from the distribution. Also, for each dataset and each missingness scenario, we report top-1 classification accuracy based on the majority vote estimated using Algorithm 2. Another measure that we use in this paper is the comparison between the estimated target certainties and average accuracies achieved for each confidence assignment. We run each experiment multiple times: 4 times for CIFAR-10 and 8 times for tabular datasets. We report the mean and standard deviation of results for each case.
+
+We compare our results with MisGAN (Li et al., 2019) and GAIN (Yoon et al., 2018) as the state of the art imputation algorithms based on GANs as well as basic denoising autoencoder (DAE) (Vincent et al., 2008) and multiple imputation by chained equations (MICE) (Buuren & Groothuis-Oudshoorn, 2010) as baselines. For experiments using MisGAN, we used the same architectures and hyperparameters as suggested by the MisGAN authors6. The only modification was to adapt the last generator layer to generate images with resolutions as we use. Regarding GAIN, we used the same network architecture as our implementation of GI and hyper-parameters as used by the GAIN authors7. In the DAE implementation, due to the incomplete data assumption, only observed features appear in the loss function, ignoring reconstruction terms corresponding to missing features. Due to scalability issues, we were only able to use MICE for the smaller non-image datasets. For these methods, to train and evaluate classifiers, we use predictors trained on imputed datasets rather than the stochastic predictor suggested in Algorithm 1.
+
+![](images/7dee5514f122f68c96a6de74132a186a9e0266bbd52176e7a7dd6f46d002cb83.jpg)  
+Figure 2: Comparison of FID scores on CIFAR-10 dataset for (a) uniform and (b) rectangular missingness. Lower FID score is better. In many cases, variance values are very small and only observable by magnifying the figures.
+
+# 4.4 RESULTS
+
+Figure 2 presents the comparison of FID scores on the CIFAR-10 dataset at different missing rates for uniform and rectangular missingness. As it can be inferred from these plots, GI outperforms other alternatives in all cases. Also, it can be seen that GAIN is able to provide more reasonable results for uniform missing data structure compared to MisGAN which is mainly effective in the rectangular missing data structure. One possible explanation for this behavior might be the fact that GAIN has an MSE loss term acting similar to an autoencoder loss smoothing noisy missing pixels. On the other hand, MisGAN tries to explicitly model missingness structure and is more successful in capturing a more structured missingness such as the case of a rectangular structure. Table 1 provides a comparison between the top-1 classification accuracy achieved using each method at different missing rates and structures. From this table, GI outperforms other work by achieving the best results in 5 out of 6 cases8.
+
+Table 2 presents a comparison of classification accuracies for Landsat, MIT-BIH, and Diabetes datasets at different missing rates. In the Landsat benchmark, GI outperforms other work in all cases. Regarding the MIT-BIH experiemts, GI outperforms other work for missing rates more than $30 \%$ while achieving similar accuracies to GAIN for lower missing rates. In the diabetes classification task, GI appears to be most effective imputing missing rates more than $20 \%$ .
+
+Figure 3 shows a comparison of accuracy versus certainty plots for GI, MisGAN, and GAIN on Landsat dataset at the missing rate of $40 \%$ . To generate these figures we trained each imputation method and then used Algorithm 1 to train predictors on imputed samples. Finally, Algorithm 2 used to measure the average accuracy at different prediction confidence levels based on a sample of 128 imputations for each test example. As it can be seen from the plots, GI provides results closest to the ideal case of having average confidence values equal to average accuracies.
+
+# 4.5 VISUALIZATION USING SYNTHESIZED DATA
+
+In order to provide further insight into the operation of GI and how imputations can potentially influence the outcomes of predictions, we conduct experiments on a synthesized dataset. The original underlying data distribution is generated by sampling 5000 samples from 4 Gaussians of standard deviation 0.1 centered on the vertices of a unit square. We assign two different classes to each cluster
+
+Table 1: Top-1 CIFAR-10 classification accuracy for different missing rates and structures.   
+
+<table><tr><td rowspan="2"></td><td colspan="5">Accuracy at Missing Rate (%) MCAR Rect.</td><td rowspan="2"></td></tr><tr><td colspan="2">MCARUniform</td><td colspan="3"></td></tr><tr><td>Method</td><td>20%</td><td>40%</td><td>60%</td><td>20%</td><td>40%</td><td>60%</td></tr><tr><td>GI</td><td>89.5 (±0.45)</td><td>87.1 (±0.54)</td><td>80.3 (±0.26)</td><td>84.0 (±0.03)</td><td>76.9 (±0.03)</td><td>66.1 (±0.16)</td></tr><tr><td>MisGAN</td><td>86.5 (±0.31)</td><td>83.7 (±0.40)</td><td>78.7 (±0.26)</td><td>82.9 (±0.44)</td><td>75.6 (±0.20)</td><td>65.0 (±0.31)</td></tr><tr><td>GAIN</td><td>88.7 (±0.45)</td><td>86.0 (±0.86)</td><td>81.8 (±0.03)</td><td>81.7 (±0.03)</td><td>73.6 (±0.35)</td><td>58.4 (±1.66)</td></tr><tr><td>DAE</td><td>88.0 (±0.22)</td><td>84.0 (±0.50)</td><td>79.8 (±0.71)</td><td>83.3 (±0.64)</td><td>75.5 (±0.44)</td><td>63.8 (±0.24)</td></tr><tr><td>Mean</td><td>85.7 (±0.02)</td><td>83.4 (±0.38)</td><td>79.2 (±0.16)</td><td>82.7 (±0.15)</td><td>75.3 (±0.16)</td><td>64.0 (±0.32)</td></tr></table>
+
+Table 2: Comparison of classification accuracies for Landsat, MIT-BIH, and Diabetes datasets at different missing rates.   
+
+<table><tr><td></td><td></td><td colspan="4"> Accuracy at Missing Rate (%)a</td></tr><tr><td>Dataset</td><td>Method</td><td>10%</td><td>20%</td><td>30%</td><td>40%</td></tr><tr><td rowspan="5">Landsat (Dua &amp; Graff, 2017)</td><td>GI</td><td>89.9 (±0.36)</td><td>89.6 (±0.36)</td><td>89.0 (±0.03)</td><td>88.0 (±0.22)</td></tr><tr><td>MisGAN</td><td>87.2 (±0.01)</td><td>85.7 (±0.19)</td><td>84.0 (±0.61)</td><td>82.9 (±0.75)</td></tr><tr><td>GAIN</td><td>89.7 (±0.42)</td><td>89.4 (±0.56)</td><td>88.4 (±0.71)</td><td>87.7 (±0.10)</td></tr><tr><td>DAE</td><td>89.4 (±0.10)</td><td>88.6 (±0.54)</td><td>87.5 (±0.14)</td><td>86.6 (±0.21)</td></tr><tr><td>MICE</td><td>89.5 (±0.16)</td><td>89.3 (±0.10)</td><td>88.1 (±0.49)</td><td>87.5 (±0.03)</td></tr><tr><td rowspan="5">MIT-BIH(Moody &amp; Mark,2001)</td><td>GI</td><td>98.5 (±0.02)</td><td>98.4 (±0.03)</td><td>98.2 (±0.07)</td><td>97.7 (±0.03)</td></tr><tr><td>MisGAN</td><td>97.8 (±0.13)</td><td>97.4 (±0.07)</td><td>96.7 (±0.07)</td><td>96.2 (±0.09)</td></tr><tr><td>GAIN</td><td>98.5 (±0.02)</td><td>98.4 (±0.06)</td><td>98.0 (±0.09)</td><td>97.5 (±0.18)</td></tr><tr><td>DAE</td><td>98.4 (±0.02)</td><td>98.2 (±0.11)</td><td>97.9 (±0.09)</td><td>97.4 (±0.02)</td></tr><tr><td>MICE</td><td>98.4 (±0.01)</td><td>98.3 (±0.01)</td><td>98.1 (±0.01)</td><td>97.5 (±0.12)</td></tr><tr><td rowspan="5">Diabetes (Kachuee et al., 2019)</td><td>GI</td><td>89.6 (±0.13)</td><td>89.0 (±0.03)</td><td>88.2 (±0.62)</td><td>86.8 (±0.38)</td></tr><tr><td>MisGAN</td><td>89.7 (±0.01)</td><td>88.9 (±0.30)</td><td>87.6 (±0.02)</td><td>86.4 (±0.68)</td></tr><tr><td>GAIN</td><td>89.2 (±0.09)</td><td>88.3 (±0.02)</td><td>86.9 (±0.09)</td><td>83.8 (±1.44)</td></tr><tr><td>DAE</td><td>89.3 (±0.05)</td><td>88.2 (±0.19)</td><td>86.9 (±0.09)</td><td>84.8 (±0.03)</td></tr><tr><td>MICE</td><td>89.8 (±0.08)</td><td>88.8 (±0.01)</td><td>88.0 (±0.08)</td><td>86.1 (±0.02)</td></tr></table>
+
+aBaseline accuracies for complete datasets (zero missing rate) are equal to $9 0 . 9 \%$ , $9 8 . 6 \%$ , and $9 0 . 7 \%$ for Landsat, MIT-BIH and Diabetes, respectively.
+
+such that diagonal vertices are of the same class (see Figure 4a, classes are represented with colors).   
+From this underlying distribution, we make an incomplete dataset with $50 \%$ of values missing.
+
+The incomplete synthesized dataset is used to train GI and other imputation methods. We take a random test sample in which the second feature has a value of about 0.1 and the other feature is missing. Ideally, in the imputation phase, we would like to sample from the condition distribution i.e. $P ( x _ { 1 } | x _ { 2 } = 0 . 1 )$ (see Figure 4b). Here, in the prediction phase, an ideal method would decide on not making a confident classification and report the uncertainty. Note that solely observing the value of 0.1 for the second feature does not provide any useful evidence for the prediction. Figure 4c-f provide samples and classification results for GI, MisGAN, GAIN, and DAE. As it can be inferred from these figures, GI generates reasonable samples from the conditional distribution and also reflects this uncertainty over the prediction. On the other hand MisGAN, probably due to its complexity of using three different generators and discriminator pairs, is suffering from mode collapse and is unable to generate samples from the other class, resulting in over-confident assignments. GAIN, perhaps due to the MSE loss terms, is inclined towards the mean of the conditional distribution at the origin.
+
+![](images/99ff87491c12b80a83f83e95b0f6597c20ef1cd8b2470028e19f5844c61fce0b.jpg)  
+Figure 3: Accuracy versus certainty plots for (a) GI, (b) MisGAN, and (c) GAIN on Landsat dataset at the missing rate of $40 \%$ .
+
+DAE, as expected, due to its MSE loss term, only captures the expected value of the distribution mean hence reducing the MSE error and generates over-smoothed imputations.
+
+![](images/efa63b3362bf0eec014da42a29ee10be1a5bc69514aec9b20546466115db1b16.jpg)  
+Figure 4: Evaluation using synthesized data: (a) samples from the underlying distribution, (b) samples from the conditional underlying distribution, (c-f) samples from the conditional distribution generate by GI, MisGAN, GAIN, and DAE.
+
+# 5 CONCLUSION
+
+In this paper, we proposed a novel method to generate imputations and measure uncertainties over target class assignments based on incomplete feature vectors. We evaluated the effectiveness of the suggested approach on image and tabular data via using different measures such as FID distance, classification accuracy, and confidence versus accuracy plots. According to the experiments, the proposed method not only can generate accurate imputations but also is able to model prediction uncertainties arising from missing values. The proposed method is applicable to many real-world applications where only an incomplete dataset is available, and modeling classification uncertainties is a necessity.
+
+# REFERENCES
+
+Theodore W Anderson. Maximum likelihood estimates for a multivariate normal distribution when some observations are missing. Journal of the american Statistical Association, 52(278):200–203, 1957.
+
+Sanjeev Arora, Andrej Risteski, and Yi Zhang. Do gans learn the distribution? some theory and empirics. 2018.
+
+S van Buuren and Karin Groothuis-Oudshoorn. mice: Multivariate imputation by chained equations in r. Journal of statistical software, pp. 1–68, 2010.
+
+Pengfei Chen, Benben Liao, Guangyong Chen, and Shengyu Zhang. Understanding and utilizing deep neural networks trained with noisy labels. arXiv preprint arXiv:1905.05040, 2019.
+
+Adnan Darwiche. Modeling and reasoning with Bayesian networks. Cambridge university press, 2009.
+
+Dheeru Dua and Casey Graff. UCI machine learning repository, 2017. URL http://archive. ics.uci.edu/ml.
+
+Vincent Dumoulin, Ishmael Belghazi, Ben Poole, Olivier Mastropietro, Alex Lamb, Martin Arjovsky, and Aaron Courville. Adversarially learned inference. arXiv preprint arXiv:1606.00704, 2016.
+
+Ian Goodfellow, Jean Pouget-Abadie, Mehdi Mirza, Bing Xu, David Warde-Farley, Sherjil Ozair, Aaron Courville, and Yoshua Bengio. Generative adversarial nets. In Advances in neural information processing systems, pp. 2672–2680, 2014.
+
+Trevor Hastie, Robert Tibshirani, Gavin Sherlock, Michael Eisen, Patrick Brown, and David Botstein. Imputing missing data for gene expression arrays, 1999.
+
+Martin Heusel, Hubert Ramsauer, Thomas Unterthiner, Bernhard Nessler, and Sepp Hochreiter. Gans trained by a two time-scale update rule converge to a local nash equilibrium. In Advances in Neural Information Processing Systems, pp. 6626–6637, 2017.
+
+Eric Jang, Shixiang Gu, and Ben Poole. Categorical reparameterization with gumbel-softmax. arXiv preprint arXiv:1611.01144, 2016.
+
+Mohammad Kachuee, Kimmo Karkkainen, Orpaz Goldstein, Davina Zamanzadeh, and Majid Sarrafzadeh. Nutrition and health data for cost-sensitive learning. arXiv preprint arXiv:1902.07102, 2019.
+
+Diederik P Kingma and Jimmy Ba. Adam: A method for stochastic optimization. arXiv preprint arXiv:1412.6980, 2014.
+
+Diederik P Kingma and Max Welling. Auto-encoding variational bayes. arXiv preprint arXiv:1312.6114, 2013.
+
+Alex Krizhevsky and Geoffrey Hinton. Learning multiple layers of features from tiny images. Technical report, Citeseer, 2009.
+
+Steven Cheng-Xian Li, Bo Jiang, and Benjamin Marlin. Learning from incomplete data with generative adversarial networks. In International Conference on Learning Representations, 2019. URL https://openreview.net/forum?id $=$ S1lDV3RcKm.
+
+Roderick JA Little and Donald B Rubin. Statistical analysis with missing data, volume 793. Wiley, 2019.
+
+Shuang Liu, Olivier Bousquet, and Kamalika Chaudhuri. Approximation and convergence properties of generative adversarial learning. In Advances in Neural Information Processing Systems, pp. 5545–5553, 2017.
+
+Pierre-Alexandre Mattei and Jes Frellsen. missiwae: Deep generative modelling and imputation of incomplete data. arXiv preprint arXiv:1812.02633, 2018.
+
+Takeru Miyato, Toshiki Kataoka, Masanori Koyama, and Yuichi Yoshida. Spectral normalization for generative adversarial networks. arXiv preprint arXiv:1802.05957, 2018.
+
+George B Moody and Roger G Mark. The impact of the mit-bih arrhythmia database. IEEE Engineering in Medicine and Biology Magazine, 20(3):45–50, 2001.
+
+Jared S Murray et al. Multiple imputation: a review of practical and theoretical findings. Statistical Science, 33(2):142–159, 2018.
+
+Nagarajan Natarajan, Inderjit S Dhillon, Pradeep K Ravikumar, and Ambuj Tewari. Learning with noisy labels. In Advances in neural information processing systems, pp. 1196–1204, 2013.
+
+Alfredo Nazabal, Pablo M Olmos, Zoubin Ghahramani, and Isabel Valera. Handling incomplete heterogeneous data using vaes. arXiv preprint arXiv:1807.03653, 2018.
+
+Richard E Neapolitan et al. Learning bayesian networks, volume 38. Pearson Prentice Hall Upper Saddle River, NJ, 2004.
+
+Thomas Dyhre Nielsen and Finn Verner Jensen. Bayesian networks and decision graphs. Springer Science & Business Media, 2009.
+
+Scott Reed, Honglak Lee, Dragomir Anguelov, Christian Szegedy, Dumitru Erhan, and Andrew Rabinovich. Training deep neural networks on noisy labels with bootstrapping. arXiv preprint arXiv:1412.6596, 2014.
+
+Donald B Rubin. Inference and missing data. Biometrika, 63(3):581–592, 1976.
+
+Joseph L Schafer and John W Graham. Missing data: our view of the state of the art. Psychological methods, 7(2):147, 2002.
+
+Kihyuk Sohn, Honglak Lee, and Xinchen Yan. Learning structured output representation using deep conditional generative models. In Advances in neural information processing systems, pp. 3483–3491, 2015.
+
+Pascal Vincent, Hugo Larochelle, Yoshua Bengio, and Pierre-Antoine Manzagol. Extracting and composing robust features with denoising autoencoders. In Proceedings of the 25th international conference on Machine learning, pp. 1096–1103. ACM, 2008.
+
+Ting-Chun Wang, Ming-Yu Liu, Jun-Yan Zhu, Andrew Tao, Jan Kautz, and Bryan Catanzaro. Highresolution image synthesis and semantic manipulation with conditional gans. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pp. 8798–8807, 2018.
+
+Jinsung Yoon, James Jordon, and Mihaela Van Der Schaar. Gain: Missing data imputation using generative adversarial nets. arXiv preprint arXiv:1806.02920, 2018.
+
+Han Zhang, Ian Goodfellow, Dimitris Metaxas, and Augustus Odena. Self-attention generative adversarial networks. arXiv preprint arXiv:1805.08318, 2018.
+
+Shichao Zhang, Zhenxing Qin, Charles X Ling, and Shengli Sheng. " missing is useful": missing values in cost-sensitive decision trees. IEEE transactions on knowledge and data engineering, 17 (12):1689–1693, 2005.
+
+# A IMPLEMENTATION AND EXPERIMENTS
+
+Table 3 presents the list of software dependencies and versions used in our implementation. To produce results related to this paper, we used a workstation with 4 NVIDIA GeForce RTX-2080Ti GPUs, a 12 core Intel Core i9-7920X processor, and 128 GB memory. Each experiment took between about 4 hours to 48 hours, based on the task and method being tested.
+
+Table 3: Software dependencies.   
+
+<table><tr><td>Dependency</td><td>Version</td></tr><tr><td>python</td><td>3.7.1</td></tr><tr><td>pytorch</td><td>1.0.0</td></tr><tr><td>cuda100</td><td>1.0</td></tr><tr><td>ipython</td><td>7.2.0</td></tr><tr><td>jupyter</td><td>1.0.0</td></tr><tr><td>numpy</td><td>1.15.4</td></tr><tr><td>pandas</td><td>0.24.1</td></tr><tr><td>scikit-learn</td><td>0.20.1</td></tr><tr><td>scipy</td><td>1.1.0</td></tr><tr><td>torchvision</td><td>0.2.1</td></tr><tr><td>tqdm</td><td>4.28.1</td></tr><tr><td>matplotlib</td><td>3.0.1</td></tr></table>
+
+# B NETWORK ARCHITECTURES
+
+Table 4 shows the exact architectures used in this paper. To show each layer or block we used the following notation. $\mathtt { C x S y P z - t }$ represents a 2-d convolution layer of kernel size $_ \textrm { x }$ , stride y, padding $_ { \textrm { Z } }$ , and number of output channels $\scriptstyle \pm$ followed by ReLU activation. Attn represents a self-attention layer similar to Zhang et al.9. $_ \mathrm { R - x }$ represents a residual block consisting of two 2-d convolutions with kernel size 3 (padding size 1), batch normalization, and ReLU activation. $\mathsf { C T x S y P z - t }$ is the convolution transpose corresponding to $\mathtt { C x S y P z - t }$ . $\operatorname { F C - x }$ is representing a linear fully-connected layer of $_ \textrm { x }$ output neurons with biases. We use spectral normalization as suggested by Miyato et al.10 for all convolutional layers in both generator and discriminator networks.
+
+Table 4: Network architectures used in our experiments.   
+
+<table><tr><td rowspan=1 colspan=1>Dataset</td><td rowspan=1 colspan=1>Generator/Discriminator Architecture</td><td rowspan=1 colspan=1>Predictor Architecture</td></tr><tr><td rowspan=1 colspan=1>CIFAR-10</td><td rowspan=1 colspan=1>C7S1P3-64，C3S2P1-128，Attn，R-128,Attn，R-128，Attn，R-128，Attn,R-128，CT3S2P1-128，CT7S1P3-3，Tanh/Sigmoid</td><td rowspan=1 colspan=1>ResNet-18 11,12</td></tr><tr><td rowspan=1 colspan=1>Landsat</td><td rowspan=1 colspan=1>FC-64，Sigmoid，BNorm，FC-64，Sigmoid,BNorm,FC-64，Sigmoid，BNorm，FC-36，Tanh/Sigmoid</td><td rowspan=1 colspan=1>FC-64，ReLU，BNorm，FC-64,ReLU，BNorm，FC-6，Softmax</td></tr><tr><td rowspan=1 colspan=1>MIT-BIH</td><td rowspan=1 colspan=1>FC-1860，ReLU，BNorm，FC-1860，ReLU，BNorm,FC-1860，ReLU，BNorm, FC-186,Tanh/Sigmoid</td><td rowspan=1 colspan=1>FC-1860，ReLU，BNorm，FC-1860,ReLU，BNorm，FC-5，Softmax</td></tr><tr><td rowspan=1 colspan=1>Diabetes</td><td rowspan=1 colspan=1>FC-45，ReLU，BNorm，FC-45，ReLU，BNorm,FC-45，ReLU，BNorm，FC-45，Tanh/Sigmoid</td><td rowspan=1 colspan=1>FC-22，ReLU，BNorm，FC-22,ReLU，BNorm，FC-3，Softmax</td></tr></table>
+
+# C MISSING DATA MECHANISMS
+
+In this paper, we conduct experiments on two mechanisms for missing values: MCAR uniform and MCAR rectangular. As in our experiments and comparisons, we consider the case where only an incomplete dataset is available for training. It is crucial to guarantee that each method has only access to a unique incomplete version of each sample. However, it is relatively expensive to load and store feature masks for each sample in the dataset. Instead, we generate missing values during the data load for each batch. A hashing mechanism is used to ensure that the same parts are missing for each sample throughout the training. Note that we set system, python, and external library hash seeds to fixed values to ensure the consistency between different runs.
+
+Algorithm 3 presents the procedure used for generating missing values with uniform structure. This algorithm is sampling independent Bernoulli distributions with probabilities equal to the missing rate. Algorithm 4 shows the outline for the rectangular missing structure used in image experiments. It consists of selecting a random point as the center of the rectangle and then deciding on parameters to be used for the beta distribution based on the missing rate. Finally, the width and height of the rectangular region are sampled from the latent beta distribution. In other words, we generate rectangular regions centered at random locations within the image which have width and height values determined by samples from a latent beta distribution. Here, distribution parameters, $\alpha$ and $\beta$ , are used to control the average missing rate. The outcome would be rectangular regions of different shape at different locations within the frame with the expected portion of missing area equal to the missing rate.
+
+In order to decide on the beta distribution parameters i.e. $\alpha$ and $\beta$ we use numerical simulations. Specifically, we fix one of the parameters to 1 and change the other parameter in the range of [1,10], while measuring the average missing rate caused by each case. Figure 5 shows the missing rates caused by different beta distribution parameters. The first half of Figure 5 (missing rates less than about 0.18) corresponds to setting $\beta$ to 1 and changing $\alpha$ values; and the other half fixing $\alpha$ to 1 and changing $\beta$ values. To generate missing rates more than $50 \%$ we invert our masks and limit the observation to the rectangular region while the rest of the image is missing. Note that missing rates indicate the ratio of features that are missing on the average case. As we are using a latent model for sampling width and height for the rectangles, the actual missing ratios for each specific sample differs between samples. See Table 5 for visual examples of different missing rates and missing structures.
+
+# Algorithm 3: MCAR uniform generation.
+
+Input: $_ { \textbf { \em x } }$ (complete feature), $r$ (missing rate) Output: ${ \pmb x } _ { m }$ (incomplete feature) $\begin{array} { r l } & { s e e d _ { x }  h a s h ( \pmb { x } ) } \\ & { \pmb { k }  1 - B e r n o u l l i ( s e e d _ { x } , s h a p e ( x ) , p r o b = r ) } \\ & { \pmb { x } _ { m }  \pmb { k } \odot \pmb { x } + ( 1 - \pmb { k } ) \odot N a N } \end{array}$
+
+# Algorithm 4: MCAR rect. generation.
+
+Input: $_ { \textbf { \em x } }$ (complete feature), $r$ (missing rate)   
+Output: ${ \pmb x } _ { m }$ (incomplete feature)   
+$s e e d _ { x } \gets h a s h ( { \pmb x } )$   
+n $\iota _ { x } , n _ { y } \gets s h a p e ( { \pmb x } )$   
+$( p _ { x } , \bar { p _ { y } } ) \sim ( u n i f o r m ( 0 , n _ { x } ) , u n i f o r m ( 0 , n _ { y } ) )$   
+α, β ← beta_params(r) // beta_params gives $\alpha , \beta$ for each missing rate based on numerical simulations   
+$\begin{array} { r l } & { ( \boldsymbol { w } , h ) \sim ( B e t a ( \alpha , \beta ) \times n _ { x } ) , B e t a ( \alpha , \beta ) \times n _ { y } ) ) } \\ & { \boldsymbol { k }  r e c t \_ m a s k ( p _ { x } , p _ { y } , w , h ) } \\ & { \mathbf { \boldsymbol { x } } _ { m }  \boldsymbol { k } \odot \mathbf { \boldsymbol { x } } + ( 1 - \boldsymbol { k } ) \odot N a N } \end{array}$
+
+![](images/29b1ccc7a90bf84d0c4b18de35c4df32b98f105fff00abdb71b3fde7ca644a3b.jpg)  
+Figure 5: Simulation results for measuring average missing rate given different beta distribution parameters.
+
+![](images/b85dc27446c318885a97542d42c9e6453ece8284078aca51ca167c500d3ddc30.jpg)
+
+# D ABLATION STUDY
+
+Figure 6 presents a comparison between using (GI W/ Atten.) and not using (GI W/O Atten.) self-attention layers before each residual block in the proposed architecture. We report FID scores on CIFAR-10 with rectangular missingness. As it can be inferred from this comparison, using self-attention achieves a consistent improvement over the baseline. We also examined the case of uniform missingness; however, we did not observe any significant improvement for this case. One possible explanation could be the fact that imputing missing data with a uniform structure can be done by processing local regions and does not require attending to different distant regions across the image.
+
+![](images/555f58484ccab468a30f31b8a1a2e7e6951c3a282f1c5c77ff9c483cf318fc49.jpg)  
+Figure 6: Comparison of FID scores achieved with (GI W/ Atten.) and without (GI W/O Atten.) self-attention layers on CIFAR-10 dataset and rectangular missingness. Lower FID score is better.
+
+Figure 7 shows a comparison of classification accuracies for the Landsat dataset achieved using different ensemble sizes $( N )$ . As it can be seen from this figure, higher values of $N$ result in improved accuracies, especially for higher missing rates. Also, it can be observed that for N values more than 64 the difference is negligible.
+
+![](images/12c219ac2d3a48f6cc52d091728437b57a1e9af830f5bd30babc7198f2bac84e.jpg)  
+Figure 7: Comparison of classification accuracies achieved with different ensemble size $( N )$ .
+
+To study the benefits of the suggested stochastic predictor, we conducted experiments comparing GI with its non-stochastic variation $( \mathrm { N } { = } 1 )$ ). Here, the CIFAR-10 dataset with the rectangular missing structure and missing rates from $20 \%$ to as high as $90 \%$ is used. From Table 6 it can be inferred that as the rate of missingness increases, the benefits of the suggested predictor algorithm increase significantly. We hypothesize that at higher rates of missingness, the conditional distribution of missing features becomes multimodal. In such a scenario, the suggested method captures the uncertainties over the target distribution resulting in the predictor to make more reliable class assignments.
+
+Table 6: Comparison of CIFAR-10 accuracies for the stochastic $\mathrm { ( N = 1 2 8 ) }$ ) and the deterministic $\left( \mathrm { N } { = } 1 \right)$ ) predictor under rectangular missingness.   
+
+<table><tr><td></td><td colspan="6">Accuracy at Missing Rate (%)</td></tr><tr><td>Method</td><td>20%</td><td>40%</td><td>60%</td><td>70%</td><td>80%</td><td>90%</td></tr><tr><td>GI (N=128)</td><td>84.0</td><td>76.9</td><td>66.1</td><td>59.1</td><td>46.0</td><td>32.1</td></tr><tr><td>GI (N=1)</td><td>83.6</td><td>75.7</td><td>65.1</td><td>56.7</td><td>42.8</td><td>29.4</td></tr><tr><td>% difference (normalized)</td><td>0.5</td><td>1.6</td><td>1.5</td><td>4.1</td><td>6.9</td><td>8.4</td></tr></table>
+
+# E VISUAL COMPARISON
+
+Table 7 and 8 provide a visual comparison of GI, MisGAN, and GAIN. For each missingness structure, we compare the best two imputation methods based on FID scores in Figure 2 i.e., GI versus MisGAN for rectangular missingness and GI versus GAIN for uniform missingness. From Table 7 it can be seen that GI is more capable in the reconstruction of fine details such as horse legs, car wheels, or plane wings. Regarding the results provided in Table 8, GI imputed samples are generally sharper and more realistic, which is consistent with our hypothesis about the drawbacks of the MSE term in the GAIN objective function.
+
+![](images/61024c4babfabb7e7dd65df430bff40654a51f9e986c3d422c05dfe4e0963228.jpg)
+
+![](images/3958733293402574782f6e1c021c4a754f6dbed20e2753d3f34209f404938a06.jpg)
+
+# F ANALYSIS OF THE RMSE MEASURE
+
+Table 9 presents the comparison of different imputation methods using the RMSE measure on CIFAR-10 for different missing structures and rates. Generally, RMSE values for the uniform missing structure are lower than their rectangular counterparts. It is consistent with our intuition that imputing uniform missingness is most similar to denoising problems where the RMSE measure is frequently used. Additionally, comparing the performance of different imputation methods using the FID measure (Section 4.4) does not demonstrate a clear correlation to results shown in Table 9. Nonetheless, it is well-known that the FID measure is more suited to measuring the performance of generated images from the underlying distribution (Heusel et al., 2017).
+
+Similarly, in Table 10, we provide RMSE values corresponding to experiments on the tabular datasets. Here, GAIN and DAE provide very similar results that are generally better than GI or MisGAN. This signifies our hypothesis that the MSE loss term may skew generated samples toward the mean of the distribution, resulting in better RMSE values but not necessarily higher final classification accuracies (see Table 2).
+
+Table 9: Comparison of imputation RMSE values for CIFAR-10 at different missing structures and rates.   
+
+<table><tr><td></td><td colspan="5">RMSE at Missing Rate (%)</td><td></td></tr><tr><td></td><td colspan="5">MCAR Uniform</td><td>MCAR Rect.</td></tr><tr><td>Method</td><td>20%</td><td>40%</td><td>60%</td><td>20%</td><td>40%</td><td>60%</td></tr><tr><td>GI</td><td>0.026 (±0.003)</td><td>0.057 (±0.008)</td><td>0.090 (±0.006)</td><td>0.097 (±0.02)</td><td>0.148 (±0.001)</td><td>0.660 (±0.010)</td></tr><tr><td>MisGAN</td><td>0.079 (±0.001)</td><td>0.161 (±0.001)</td><td>0.257 (±0.002)</td><td>0.106 (±0.005)</td><td>0.158 (±0.004)</td><td>0.250 (±0.001)</td></tr><tr><td>GAIN</td><td>0.027 (±0.003)</td><td>0.045 (±0.001)</td><td>0.072 (±0.005)</td><td>0.340 (±0.047)</td><td>0.511 (±0.001)</td><td>0.660 (±0.010)</td></tr><tr><td>DAE</td><td>0.036 (±0.001)</td><td>0.075 (±0.002)</td><td>0.121 (±0.005)</td><td>0.116 (±0.007)</td><td>0.160 (±0.001)</td><td>0.233 (±0.029)</td></tr></table>
+
+Table 10: Comparison of imputation RMSE values for Landsat, MIT-BIH, and Diabetes datasets at different missing rates.   
+
+<table><tr><td colspan="2"></td><td colspan="4">RMSE at Missing Rate (%)</td></tr><tr><td>Dataset</td><td>Method</td><td>10%</td><td>20%</td><td>30%</td><td>40%</td></tr><tr><td rowspan="4">Landsat (Dua &amp; Graff,2017)</td><td>GI</td><td>0.040 (±0.005)</td><td>0.067 (±0.007)</td><td>0.076 (±0.020)</td><td>0.136 (±0.002)</td></tr><tr><td>MisGAN</td><td>0.068 (±0.001)</td><td>0.096 (±0.001)</td><td>0.118 (±0.001)</td><td>0.136 (±0.001)</td></tr><tr><td>GAIN</td><td>0.018 (±0.001)</td><td>0.024 (±0.001)</td><td>0.030 (±0.001)</td><td>0.037 (±0.001)</td></tr><tr><td>DAE</td><td>0.020 (±0.001)</td><td>0.031 (±0.001)</td><td>0.041 (±0.001)</td><td>0.052 (±0.001)</td></tr><tr><td rowspan="4">MIT-BIH(Moody &amp; Mark,2001)</td><td>GI</td><td>0.038 (±0.001)</td><td>0.060 (±0.004)</td><td>0.071 (±0.002)</td><td>0.095 (±0.002)</td></tr><tr><td>MisGAN</td><td>0.073 (±0.007)</td><td>0.092 (±0.002)</td><td>0.115 (±0.003)</td><td>0.111 (±0.001)</td></tr><tr><td>GAIN</td><td>0.032 (±0.008)</td><td>0.046 (±0.001)</td><td>0.055 (±0.004)</td><td>0.067 (±0.007)</td></tr><tr><td>DAE</td><td>0.029 (±0.001)</td><td>0.048 (±0.008)</td><td>0.061 (±0.009)</td><td>0.068 (±0.003)</td></tr><tr><td rowspan="4">Diabetes (Kachuee et al., 2019)</td><td>GI</td><td>0.080 (±0.002)</td><td>0.118 (±0.008)</td><td>0.149 (±0.020)</td><td>0.189 (±0.009)</td></tr><tr><td>MisGAN</td><td>0.082 (±0.004)</td><td>0.111 (±0.002)</td><td>0.133 (±0.001)</td><td>0.151 (±0.001)</td></tr><tr><td>GAIN</td><td>0.064 (±0.001)</td><td>0.092 (±0.001)</td><td>0.119 (±0.001)</td><td>0.140 (±0.001)</td></tr><tr><td>DAE</td><td>0.065 (±0.001)</td><td>0.093 (±0.001)</td><td>0.118 (±0.001)</td><td>0.143 (±0.001)</td></tr></table>
+
+# G IMPACT OF TRAINING NOISE
+
+Addition of noise to input vectors often serves as an input augmentation and results in improved generalization accuracies. In order to verify that the improved GI performance is not merely due to the introduction of noise in the suggested architecture, we conducted an experiment by adding different amounts of Gaussian noise during the training process for GAIN and GI. Specifically, we compared how the CIFAR-10 test accuracies change at different degrees of training noise for uniform and rectangular missingess structures at the average missing rate of $40 \%$ .
+
+According to Table 11, adding small amounts of Gaussian noise (e.g., std=0.0125) improves the generalization under uniform missingness for both GI and GAIN. Even in this case, GI is still outperforming GAIN in terms of final classification performance. It is also interesting to point out that for the case of rectangular missingness adding Gaussian noise results in a consistent reduction in the classification accuracy for both methods.
+
+Table 11: Top-1 CIFAR-10 classification accuracy at $40 \%$ missing rate using added training noise.   
+
+<table><tr><td></td><td colspan="2">Accuracy (%)</td></tr><tr><td>Noise STD</td><td>MCAR Uniform (40%) GI GAIN</td><td>MCAR Rect. (40%) GI GAIN</td></tr><tr><td>0.0</td><td>87.1 86.0</td><td>76.9 73.6</td></tr><tr><td>0.0125</td><td>87.3 86.3</td><td>76.8 73.3</td></tr><tr><td>0.025</td><td>86.5 86.6</td><td>76.7 73.2</td></tr><tr><td>0.05</td><td>85.6 84.7</td><td>73.7 72.4</td></tr><tr><td>0.1</td><td>82.0 80.6</td><td>68.7 67.0</td></tr></table>
+
+# H IMPACT OF THE MSE LOSS TERM
+
+In our earlier discussions, we stated that the MSE loss term used in GAIN would bias the distribution of generated samples toward the mean of the distribution. Here, a synthesized dataset is used to illustrate the impact of MSE loss term on the distribution of generated samples. A hyperparameter, $\lambda$ , controls the weight of the MSE term in the final objective function. As it can be observed from Figure 8, the higher the $\lambda$ parameter, the lower the variance of the generated samples (i.e., more bias toward the mean of the distribution).
+
+![](images/e6f566e62f5c08205f49f8240037709b45a8d8f533c27ee98af701aa8fdcd47f.jpg)  
+Figure 8: Comparison of generating samples from a Gaussian distribution (a) samples from the original distribution, (b) samples generated using GAIN imputers with different significance of the MSE term (controlled by $\lambda$ ).
+
+# I IMPACT OF THE DISCRIMINATOR HINT VECTOR
+
+Yoon et al. (2018) suggested the idea of guiding the discriminator network using a hint mechanism. A hint vector reveals a subset of features that are missing to the discriminator. In Figure 9 and 10 we provide a comparison of learning curves for GI implemented using different hint rates. From Figure 9, using the hint mechanism does not result in any noticeable improvement in the final imputation quality justifying the added complexity. For the case of the rectangular missing structure in Figure 10; however, using the hint vector causes instabilities in the training process. One possible explanation is: providing even a small portion of the mask as a hint, due to the deterministic nature of the rectangular shape it is equivalent to providing region boundaries to the discriminator making it obvious for the discriminator. In GAN training we generally want to have equal competition between the generator and discriminator.
+
+![](images/d1773a0973e8e20e8a1e95585cf0570e2f70c7ae32a782150bd1926ef6c94fdd.jpg)  
+Figure 9: Learning curves for CIFAR-10 with uniform missing structure at different discriminator hint rates.
+
+![](images/1d860d6685302d772bd786609a2406e9a1cc659ad5aaf397983d57ed6fba2242.jpg)  
+Figure 10: Learning curves for CIFAR-10 with rectangular missing structure at different discriminator hint rates.
\ No newline at end of file
diff --git a/parse/train/B1em9h4KDS/B1em9h4KDS_content_list.json b/parse/train/B1em9h4KDS/B1em9h4KDS_content_list.json
new file mode 100644
index 0000000000000000000000000000000000000000..99226a2e6388addb2dc7f637c303c6a1c2f2905a
--- /dev/null
+++ b/parse/train/B1em9h4KDS/B1em9h4KDS_content_list.json
@@ -0,0 +1,2009 @@
+[
+    {
+        "type": "text",
+        "text": "GENERATIVE IMPUTATION AND STOCHASTIC PREDICTION ",
+        "text_level": 1,
+        "bbox": [
+            174,
+            98,
+            869,
+            122
+        ],
+        "page_idx": 0
+    },
+    {
+        "type": "text",
+        "text": "Anonymous authors Paper under double-blind review ",
+        "bbox": [
+            183,
+            145,
+            400,
+            172
+        ],
+        "page_idx": 0
+    },
+    {
+        "type": "text",
+        "text": "ABSTRACT ",
+        "text_level": 1,
+        "bbox": [
+            454,
+            210,
+            544,
+            224
+        ],
+        "page_idx": 0
+    },
+    {
+        "type": "text",
+        "text": "In many machine learning applications, we are faced with incomplete datasets. In the literature, missing data imputation techniques have been mostly concerned with filling missing values. However, the existence of missing values is synonymous with uncertainties not only over the distribution of missing values but also over target class assignments that require careful consideration. In this paper, we propose a simple and effective method for imputing missing features and estimating the distribution of target assignments given incomplete data. In order to make imputations, we train a simple and effective generator network to generate imputations that a discriminator network is tasked to distinguish. Following this, a predictor network is trained using the imputed samples from the generator network to capture the classification uncertainties and make predictions accordingly. The proposed method is evaluated on CIFAR-10 image dataset as well as three real-world tabular classification datasets, under different missingness rates and structures. Our experimental results show the effectiveness of the proposed method in generating imputations as well as providing estimates for the class uncertainties in a classification task when faced with missing values. ",
+        "bbox": [
+            233,
+            242,
+            766,
+            465
+        ],
+        "page_idx": 0
+    },
+    {
+        "type": "text",
+        "text": "1 INTRODUCTION ",
+        "text_level": 1,
+        "bbox": [
+            176,
+            494,
+            336,
+            510
+        ],
+        "page_idx": 0
+    },
+    {
+        "type": "text",
+        "text": "While a large body of the machine learning literature is built upon the assumption of having access to complete datasets, in many real-world problems only incomplete datasets are available. The existence of missing values can be due to many different causes such as human subjects not adhering to certain questions or features not being collected frequently due to financial or experimental limitations, sensors failures, and so forth. Data imputation techniques have been suggested as a solution to bridge this gap in the literature by replacing missing values with observed values. ",
+        "bbox": [
+            174,
+            526,
+            825,
+            611
+        ],
+        "page_idx": 0
+    },
+    {
+        "type": "text",
+        "text": "Missing data imputation approaches can be categorized into single and multiple imputation methods. Single imputation methods try to replace each missing value with a plausible value that is the best fit given the value of other correlated features and knowledge extracted from the dataset (Hastie et al., 1999; Anderson, 1957). While these methods are easy to implement and use in practice, imputed values may induce bias by eliminating less likely but important values. Also, these methods do not suggest a way to measure to what extent the imputed values are representative of the missing values (Little & Rubin, 2019). ",
+        "bbox": [
+            174,
+            617,
+            825,
+            714
+        ],
+        "page_idx": 0
+    },
+    {
+        "type": "text",
+        "text": "Multiple imputation (MI) techniques, as suggested by the name, try to use multiple imputed values to impute each missing value. The result would be having a set of imputed datasets that enables measuring how consistent and statistically significant are the results of the experiments (Rubin, 1976). While MI offers interesting statistical insights about the reliability of analysis on incomplete data, the insight is imprecise as it is mainly concerned about the population of data samples rather than individual instances. Specifically, MI methods reason about the statistical properties on a limited number of imputed datasets (less than 10 in most practical implementations) on the population of samples within the dataset (Schafer & Graham, 2002; Murray et al., 2018). ",
+        "bbox": [
+            174,
+            722,
+            825,
+            833
+        ],
+        "page_idx": 0
+    },
+    {
+        "type": "text",
+        "text": "The existence of missing values is synonymous with having uncertainty over these values that requires careful consideration. In many real-world applications, we are dealing with supervised problems that demand modeling and prediction based on incomplete data. Take for instance, prediction of class assignments given an image in which a large portion of the frame is missing. In such a scenario, based on observed frame parts, there might be multiple probable class assignments each having a different likelihood. Here, we are not only interested in imputing missing values or measuring how robust our imputations are, but also it is highly desirable to measure the impact of missing values on the prediction outcome for each instance. ",
+        "bbox": [
+            174,
+            840,
+            825,
+            924
+        ],
+        "page_idx": 0
+    },
+    {
+        "type": "text",
+        "text": "",
+        "bbox": [
+            173,
+            103,
+            821,
+            132
+        ],
+        "page_idx": 1
+    },
+    {
+        "type": "text",
+        "text": "In this paper, we propose the idea of Generative Imputation and Stochastic Prediction (GI) as a novel approach to impute missing values and to measure class uncertainties arising from the distribution of missing values. The suggested approach is based on neural networks trained using an adversarial objective function. Additionally, a predictor is trained on the generated samples from the imputer network which is able to reflect the impact of uncertainties over missing values. This enables measuring different prediction outcomes and certainties for each specific instance. We evaluate the effectiveness of the proposed method on different incomplete image and tabular datasets under various missingness structures. ",
+        "bbox": [
+            174,
+            138,
+            825,
+            250
+        ],
+        "page_idx": 1
+    },
+    {
+        "type": "text",
+        "text": "2 RELATED WORK ",
+        "text_level": 1,
+        "bbox": [
+            176,
+            276,
+            343,
+            292
+        ],
+        "page_idx": 1
+    },
+    {
+        "type": "text",
+        "text": "One of the simplest traditional methods for handling missing values includes imputing the occurrences of missing values with constant values such as zeros or using mean values. To enhance the accuracy of such imputations, alternatives such as $\\mathbf { k }$ -nearest neighbors (KNN) (Hastie et al., 1999) and maximum likelihood estimation (MLE) (Anderson, 1957) have been suggested to estimate values to be used given an observed context. While these methods are easy to implement and analyze, they often fail to capture the complex feature dependencies as well as structures present in many problems. ",
+        "bbox": [
+            174,
+            313,
+            825,
+            396
+        ],
+        "page_idx": 1
+    },
+    {
+        "type": "text",
+        "text": "Rubin (1976) suggested a categorization for missingness mechanisms: missing completely at random (MCAR), missing at random (MAR), and missing not at random (MNAR). Under the assumption of MAR, the authors suggested multiple imputation (MI) as a stochastic imputation method. Here, instead of imputing missing values using a single value, several values are sampled to represent the distribution over the missing value. MI generates a few imputed complete datasets that are then used independently in statistical modeling (Schafer & Graham, 2002; Little & Rubin, 2019). Usually, the final goal of MI is to measure the robustness of the final statistical analysis amongst the imputed datasets. In other words, it measures the quality of imputations and the statistical significance of analysis on the imputed data. It should be noted that the number of imputations used in MI is usually very limited. Also, often strong simplifying assumptions are made in modeling the data distribution (e.g., multi-variate Gaussian or Student’s t distribution) which limit the applicability of this method (Schafer & Graham, 2002; Murray et al., 2018). ",
+        "bbox": [
+            174,
+            404,
+            825,
+            569
+        ],
+        "page_idx": 1
+    },
+    {
+        "type": "text",
+        "text": "More recently, autoencoder architectures have been suggested as powerful density estimators capable of capturing complex distributions. Perhaps, denoising autoencoders (DAE) (Vincent et al., 2008) are one of the most intuitive approaches in which a neural network is trained to reconstruct and denoise its input. Following a more probabilistic perspective, variational autoencoders (VAE) (Kingma & Welling, 2013) try to learn the data generating distribution via a latent representation. Specifically, conditional variational autoencoders (CVAE) (Sohn et al., 2015) can be used to sample missing values conditioned on observed values. For instance, Mattei & Frellsen (2018) suggested a method based on deep latent variable models and importance sampling that offers a tighter likelihood bound compared to the standard VAE bound. While these methods are powerful generative models applicable to missing data imputation, often samples generated using autoencoders are biased toward the mode of the distribution (e.g., resulting in blurry images, for vision tasks) (Goodfellow et al., 2014; Dumoulin et al., 2016). ",
+        "bbox": [
+            174,
+            577,
+            825,
+            743
+        ],
+        "page_idx": 1
+    },
+    {
+        "type": "text",
+        "text": "Recently, due to the success of generative adversarial networks (GAN), there has been great attention toward applying them to impute missing values. For instance, Yoon et al. (2018) suggested an imputation method based on adversarial and reconstruction loss terms. Li et al. (2019) introduced the idea of using separate generator and discriminator networks to learn the missing data structure and data distribution. These methods have been quite successful and are able to present the state-of-the-art results. Though it should be noted that often the presence of additional loss terms may bias the generated samples toward the mode of the distribution being modeled. Also, these methods are often complicated to be applied in practical setups by practitioners. For instance, Yoon et al. (2018) requires setting hyperparameters to adjust the influence of an MSE loss term as well as the rate of discriminator hint vectors. Also as another example, Li et al. (2019) uses three generators and three discriminators for the final imputer architecture. ",
+        "bbox": [
+            174,
+            750,
+            825,
+            876
+        ],
+        "page_idx": 1
+    },
+    {
+        "type": "image",
+        "img_path": "images/c5713bc074e138a2a20d050db7edf84f541cc471e67603e24e6900e610f80f08.jpg",
+        "image_caption": [
+            "Figure 1: Block diagram of the proposed adversarial imputation method. $h$ represents the blending function of (1), and $L$ is the adversarial loss function of (2). "
+        ],
+        "image_footnote": [],
+        "bbox": [
+            210,
+            103,
+            790,
+            185
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "text",
+        "text": "",
+        "bbox": [
+            176,
+            260,
+            820,
+            287
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "text",
+        "text": "From the perspective of supervised analysis, imputation and handling missing values are usually considered as a preprocessing step. A few exceptions exist such as Bayesian models and decision trees that permit direct analysis on incomplete data (Nielsen & Jensen, 2009; Zhang et al., 2005). Note that while certain Bayesian methods such as probabilistic Bayesian networks allow handling of missing values as unobserved variables. However, given an incomplete training dataset and without any known causal structure as a priori, learning such models is a very challenging problem with the complexity of at least NP-complete to learn the network architecture in addition to an iterative EM optimization to learn model parameters (Darwiche, 2009; Neapolitan et al., 2004). We argue that the simplistic approach of imputing missing values as a preprocessing step discards uncertainties that exist in original incomplete data samples. Instead, there is a need for methods that reflect these uncertainties on the final predicted target distribution. This work suggests the idea of training a predictor on different imputed samples to capture the uncertainties over class assignments. Compared to MI, the suggested method interleaves imputation and training a downstream prediction model, enabling to estimate classification uncertainties for each instance. ",
+        "bbox": [
+            174,
+            295,
+            825,
+            489
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "text",
+        "text": "3 PROPOSED METHOD ",
+        "text_level": 1,
+        "bbox": [
+            176,
+            510,
+            375,
+            526
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "text",
+        "text": "3.1 PROBLEM DEFINITION ",
+        "text_level": 1,
+        "bbox": [
+            176,
+            542,
+            370,
+            558
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "text",
+        "text": "In this paper, we make the general assumption of having access to an incomplete dataset $\\mathcal { D }$ consisting of a set of feature vector, mask vector, and target class pairs $( { \\pmb x } _ { i } , { \\pmb k } _ { i } , y _ { i } )$ . For each feature vector, $\\pmb { x } _ { i } \\in \\mathbb { R } ^ { d }$ , only a subset of the features is available. The mask vector $\\pmb { k } _ { i } \\in \\{ 0 , 1 \\} ^ { d }$ is used to indicate available features and missing features by ones and zeros, respectively. Here, to represent features as fixed-width vectors, arbitrary (or NaN) values are used to fill missing values. Also, for convenience, we often use $\\pmb { x } _ { i } ^ { o b s }$ and $\\pmb { x } _ { i } ^ { m i s s }$ to refer to the set of observed and missing features for the feature vector $\\mathbf { \\Delta } _ { \\mathbf { \\mathcal { X } } _ { i } }$ . ",
+        "bbox": [
+            173,
+            569,
+            826,
+            670
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "text",
+        "text": "We define our objective in two steps: $( i )$ Imputing missing values via sampling from the conditional distribution of missing features given observed features i.e., $P ( \\pmb { x } _ { i } ^ { m i s s } | \\pmb { x } _ { i } ^ { o b s } )$ . $( i i )$ Estimating the distribution of target classes given the observed features and the distribution of missing features i.e., $P ( y | \\mathbf { x } _ { i } ^ { o b s } , \\mathbf { x } _ { i } ^ { m i s s } )$ . For the first part, we are interested in sampling from the conditional distribution rather than finding the mode of the distribution as the most probable imputation. Similarly, for the second part, we are interested in obtaining a distribution over the possible target assignments and the confidence of each class rather than maximum likelihood class assignments. ",
+        "bbox": [
+            174,
+            676,
+            825,
+            773
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "text",
+        "text": "3.2 GENERATIVE IMPUTATION ",
+        "text_level": 1,
+        "bbox": [
+            176,
+            792,
+            398,
+            806
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "text",
+        "text": "To generate samples from the distribution of missing features conditioned on the observed features, we follow the idea first suggested by Yoon et al. (2018). In this paradigm, a generator network is responsible for generating imputations while a discriminator is trying to distinguish imputed features from observed features (see Figure 1). ",
+        "bbox": [
+            174,
+            819,
+            825,
+            875
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "text",
+        "text": "Specifically, the generator function $G ( \\pmb { x } _ { i } , \\pmb { k } _ { i } , z ) \\in \\mathbb { R } ^ { d }$ generates an imputed feature vector, based on observed features, the corresponding mask, and a Gaussian noise vector $( z )$ . Note that, in order to achieve the final imputed vector, $\\widehat { \\pmb { x } } _ { i }$ , we blend (or, merge) the output of the generator with the input features to replace generated values with the exact values of observed features: ",
+        "bbox": [
+            176,
+            881,
+            823,
+            924
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "text",
+        "text": "",
+        "bbox": [
+            171,
+            103,
+            689,
+            118
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "equation",
+        "img_path": "images/316b443fc9a2d734705d8ec9fb97c77cc89aef805263dbe96b8ab476ecdb014c.jpg",
+        "text": "$$\n\\begin{array} { r } { \\widehat { \\pmb { x } } _ { i , j } = \\left\\{ \\begin{array} { l l } { \\pmb { x } _ { i , j } } & { \\mathrm { i f } ~ \\pmb { k } _ { i , j } = 1 } \\\\ { G ( \\pmb { x } _ { i } , \\pmb { k } _ { i } , z ) _ { j } } & { \\mathrm { i f } ~ \\pmb { k } _ { i , j } = 0 } \\end{array} \\right. , } \\end{array}\n$$",
+        "text_format": "latex",
+        "bbox": [
+            374,
+            126,
+            622,
+            161
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "where $\\mathbf { \\delta } _ { \\mathbf { x } _ { i , j } }$ refers to $j ^ { : }$ ’th feature of sample $i$ . Also, note that by sampling $_ { z }$ multiple times, we can obtain different imputation samples from the conditional distribution indicated by $\\widehat { \\pmb { x } } _ { i } ^ { l }$ where $l$ is the sample number. ",
+        "bbox": [
+            174,
+            167,
+            823,
+            212
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "A discriminator network, $D ( \\widehat { \\mathbf { x } } _ { i } )$ , is trained to distinguish real and imputed features by generating a predicted softmax mask output, $\\widehat { \\pmb { k } } _ { i }$ . Here a binary cross-entropy loss per mask element is used as the adversarial objective function: ",
+        "bbox": [
+            174,
+            218,
+            825,
+            263
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "equation",
+        "img_path": "images/05f2884f3e6458cf6cd8b12ada277081d36c3ce2f6721f115066e191fe65151b.jpg",
+        "text": "$$\n\\operatorname* { m a x } _ { G } \\operatorname* { m i n } _ { D } L ( G , D ) = \\mathbb { E } _ { k \\sim \\mathcal { D } , \\widehat { k } \\sim D ( G ( x , k , z ) ) } \\ [ k ^ { T } \\log ( \\widehat { k } ) \\ + \\ ( 1 - k ) ^ { T } \\log ( 1 - \\widehat { k } ) ] .\n$$",
+        "text_format": "latex",
+        "bbox": [
+            230,
+            270,
+            766,
+            296
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "The intuition behind this adversarial loss function is that given a generator function which captures the data distribution successfully, the discriminator would not be able to distinguish the parts of the feature vector that were originally missing. ",
+        "bbox": [
+            173,
+            309,
+            825,
+            353
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "Compared to Yoon et al. (2018), the objective function of (2) does not have an MSE loss term. Instead, we use recent advances in GAN stabilization and training to improve the training process (see Section 3.4). While it is quite prevalent in the adversarial learning literature to use additional loss terms such as mean squared error (MSE) to enhance the quality of generated samples, we decided to keep our solution as simple as possible. Additionally, in our experiments, we provide supporting evidence that this simple loss function enables us to sample from the conditional distribution and prevents biased inclinations toward distribution modes. ",
+        "bbox": [
+            173,
+            358,
+            825,
+            457
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "3.3 STOCHASTIC PREDICTION ",
+        "text_level": 1,
+        "bbox": [
+            174,
+            473,
+            397,
+            488
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "To capture the distribution of target classes given incomplete data, we suggest the idea of stochastic prediction. As indicated in the previous section, the generator can be used to sample from the conditional distribution. Here, a predictor is trained based on the imputed samples to predict class assignments and to calculate the confidence of these assignments. For instance, for a specific test sample at hand, if a certain missing feature is a strong indicator of the target class, we would like to observe the impact of different imputations for that feature on the final hypothesis. ",
+        "bbox": [
+            173,
+            498,
+            825,
+            584
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "Formally, we are interested in finding the certainty of class assignments given observed features: ",
+        "bbox": [
+            171,
+            589,
+            803,
+            604
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "equation",
+        "img_path": "images/ac9223ff756409c368c037770a52636ea42d523d930f2c3fec7e95ff161a5778.jpg",
+        "text": "$$\n\\Psi = P ( y | \\mathbf { x } _ { i } ^ { o b s } ) .\n$$",
+        "text_format": "latex",
+        "bbox": [
+            442,
+            611,
+            555,
+            631
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "Here, $\\Psi$ is a vector where each element is representing a certain class. Rewriting (3) as a marginal we have: ",
+        "bbox": [
+            173,
+            637,
+            828,
+            662
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "equation",
+        "img_path": "images/470d6e7a29f360f57c1cae6febf98824385510b7dba8545cf78581e9577b11bb.jpg",
+        "text": "$$\n\\Psi = \\int P ( \\pmb { x } _ { i } ^ { m i s s } ) P ( y | \\pmb { x } _ { i } ^ { o b s } , \\pmb { x } _ { i } ^ { m i s s } ) d \\pmb { x } _ { i } ^ { m i s s } .\n$$",
+        "text_format": "latex",
+        "bbox": [
+            351,
+            660,
+            647,
+            693
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "Approximating the integration using a summation, given enough samples, $\\Psi$ can be estimated by: ",
+        "bbox": [
+            173,
+            695,
+            807,
+            710
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "equation",
+        "img_path": "images/cc396e8a45febe2930d813ed68f2f15e7c1080af486e7d6e38ae1739a25905e1.jpg",
+        "text": "$$\n\\Psi \\approx \\frac { 1 } { N } \\sum P ( y | \\pmb { x } _ { i } ^ { o b s } , \\widehat { \\pmb { x } } _ { i } ^ { m i s s } ) ,\n$$",
+        "text_format": "latex",
+        "bbox": [
+            395,
+            715,
+            601,
+            746
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "where $\\widehat { \\pmb { x } } _ { i } ^ { m i s s }$ are samples taken from the conditional distribution of missing features given observed bones. We use the suggested generative imputation method to generate samples required for this approximation. Rewriting (1) using Hadamard product and as function of the noise vector: ",
+        "bbox": [
+            174,
+            752,
+            825,
+            796
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "equation",
+        "img_path": "images/ea12ac3bd2ae21bbd617c307f801f8487f98ff3bda3cc54317f9fad1e9bde676.jpg",
+        "text": "$$\n\\widehat { \\mathbf { x } } _ { i } = \\pmb { k } _ { i } \\odot \\pmb { x } _ { i } + \\left( 1 - \\pmb { k } _ { i } \\right) \\odot G ( \\pmb { x } _ { i } , \\pmb { k } _ { i } , z )\n$$",
+        "text_format": "latex",
+        "bbox": [
+            364,
+            803,
+            632,
+            820
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "Assuming that a predictor, $F _ { \\theta }$ , is available which predicts class assignments for a complete feature vector, $\\Psi$ can be estimated as: ",
+        "bbox": [
+            169,
+            825,
+            825,
+            854
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "equation",
+        "img_path": "images/bfb16b641f19556475102ee95cb45ba3863f00d59720a9e9105fdc78b4be5e3d.jpg",
+        "text": "$$\n\\Psi = \\mathbb { E } _ { \\mathbf { z } } [ F _ { \\boldsymbol { \\theta } } ( \\widehat { \\mathbf { x } } _ { i } ) ] \\approx \\frac { 1 } { N } { \\sum _ { l = 1 } ^ { N } } F _ { \\boldsymbol { \\theta } } ( \\widehat { \\mathbf { x } } _ { i } ^ { l } ) ~ .\n$$",
+        "text_format": "latex",
+        "bbox": [
+            377,
+            859,
+            620,
+            890
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "Algorithm 1 presents the suggested algorithm for training the predictor. It consists of taking samples from the incomplete dataset, then imputing them using our generator network, and using the imputed ",
+        "bbox": [
+            173,
+            895,
+            825,
+            924
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "table",
+        "img_path": "images/209c06eee3400b4d45d16029079eb73985b40561bfa9e60facddc6458b4b6abc.jpg",
+        "table_caption": [],
+        "table_footnote": [],
+        "table_body": "<table><tr><td>Algorithm1: Training the predictor.</td></tr><tr><td>Input: G (trained imputer), D (dataset) Output: Fθ (trained predictor) foreachTrainingEpoch do foreach(xi,ki,yi) in D do z ~ N(0,I) Xi ←ki①xi+(1-ki)OG(xi,ki,z) loss ← L(yi,yred) Backpropagate loss</td></tr></table>",
+        "bbox": [
+            173,
+            106,
+            475,
+            279
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "text",
+        "text": "Algorithm 2: Estimating target distributions.   \nInput: $F _ { \\theta }$ (trained predictor), $( x , k )$ (test sample), $\\mathbf { N }$ (ensemble samples)   \nOutput: $\\Psi$ (distribution over target classes)   \nΨ ← zeros ∈ R#classes   \nforeach Ensemble Sample $^ { l }$ to $N$ do z ∼ N (0, I) $\\widehat { \\mathbf { x } }  k \\odot x + ( 1 - k ) \\odot G ( { \\pmb { x } } , { \\pmb { k } } , z )$ bypred ← Fθ(x) bj ← argmax(ypred) Ψj ← Ψj + 1N ",
+        "bbox": [
+            483,
+            107,
+            781,
+            272
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "text",
+        "text": "samples to update the predictor. Note that, on each epoch and for each sample, the generator generates a new sample from the conditional distribution. Intuitively, it means that the predictor observes and learns to operate under different imputations for a given sample. This is different from approaches such as multiple imputation where several predictors are trained on different imputed versions of a dataset. ",
+        "bbox": [
+            173,
+            311,
+            825,
+            381
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "text",
+        "text": "Algorithm 2 presents the suggested algorithm for making predictions and estimating target distributions given a trained predictor model. Here, a sample is imputed $N$ times and inference on this set results in an ensemble of predictions over different imputations. The output of this algorithm can be interpreted as a distribution over the confidence of class assignments given a partially observed test sample. The following claims justify the validity of Algorithm 1and Algorithm 2. ",
+        "bbox": [
+            173,
+            387,
+            825,
+            459
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "text",
+        "text": "Claim 1. (Generalization of the predictor). If we assume imputed $\\widehat { x _ { i } s }$ are samples from the underlying bfeature distribution, then the assigned training set labels can be modeled as labels generated from a noisy labeling process. ",
+        "bbox": [
+            174,
+            464,
+            821,
+            507
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "text",
+        "text": "Claim 1 permits the analysis of the generalization and convergence for the predictor trained using Algorithm 1 based on current literature in training models with noisy labels (Natarajan et al., 2013; Reed et al., 2014; Chen et al., 2019). From the analysis provided by Chen et al. (2019), test accuracy in asymmetric label noise conditions is a quadratic function of the label noise: ",
+        "bbox": [
+            173,
+            513,
+            825,
+            570
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "equation",
+        "img_path": "images/9ada2a757bda1fa063129713d7bfe7a24fe80e1f7c59bea48bebc2158db6572e.jpg",
+        "text": "$$\nP ( y _ { i } = \\widehat { y _ { i } } ) = ( 1 - \\epsilon ) ^ { 2 } + \\epsilon ^ { 2 } ,\n$$",
+        "text_format": "latex",
+        "bbox": [
+            401,
+            575,
+            594,
+            594
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "text",
+        "text": "where $\\widehat { y _ { i } }$ is underlying true label for the imputed feature vector $( \\widehat { x _ { i } } )$ , and $y _ { i }$ is the label provided by bthe incomplete dataset. In (8), label noise ratio, $\\epsilon$ b, represents the probability of the label transition from a certain target class to another: ",
+        "bbox": [
+            173,
+            599,
+            825,
+            642
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "equation",
+        "img_path": "images/2d600ebdd741a923b07c08d88e148eb392dd20a835c1cff559210c8791c19559.jpg",
+        "text": "$$\n\\epsilon = 1 - P ( \\widehat { y _ { i } } = j | y _ { i } = j ) .\n$$",
+        "text_format": "latex",
+        "bbox": [
+            408,
+            648,
+            589,
+            666
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "text",
+        "text": "In practice, $\\epsilon$ is determined by the problem-specific underlying data distribution as well as the distribution of missing values. ",
+        "bbox": [
+            174,
+            671,
+            825,
+            700
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "text",
+        "text": "Justification for claim 1 is straightforward, assume that $\\{ \\widehat { y _ { i } } ^ { 1 } \\ldots \\widehat { y _ { i } } ^ { N } \\}$ are underlying true labels for each of $\\{ \\widehat { x _ { i } } ^ { 1 } \\ldots \\widehat { x _ { i } } ^ { N } \\}$ b b. During training, for any imputed sample in $\\{ \\widehat { x _ { i } } ^ { 1 } \\ldots \\widehat { x _ { i } } ^ { N } \\}$ , we use the dataset bprovided label, $y$ b b, to calculate the loss and to update model parameters. In the case that any of $\\{ \\widehat { y _ { i } } ^ { 1 } \\ldots \\widehat { y _ { i } } ^ { N } \\}$ is different from $y$ , the loss term corresponding to that term would be calculated using b ba wrong label. Here, if we consider the average impact on gradients for batches of samples rather than individual cases, the overall impact on training would be very similar to the case of training using noisy labels. ",
+        "bbox": [
+            173,
+            707,
+            826,
+            813
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "text",
+        "text": "Claim 2. (Approximation of the target distribution). If we assume: (i) imputed $\\widehat { x _ { i } s }$ are valid samples from the underlying feature distribution, b(ii) a good predictor can be trained using the incomplete data, (iii) enough samples are used and the Monte Carlo estimator is unbiased, then the target distribution, $\\Psi$ , can be estimated accurately. ",
+        "bbox": [
+            173,
+            818,
+            656,
+            888
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "text",
+        "text": "This claim supports Algorithm 2 that is suggested to estimate the target distribution given a partially observed feature vector. ",
+        "bbox": [
+            173,
+            895,
+            821,
+            924
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "text",
+        "text": "The first assumption is consistent with the theoretical analysis of generative adversarial networks that they can converge to the true underlying distribution (Arora et al., 2018; Liu et al., 2017). The second assumption is supported by Claim 1. Regarding the last assumption, each sample requires one forward computation of the generator and predictor networks which, based on the scalability of current network architectures, usually permits thousands of samples to be taken at a reasonable computational cost. ",
+        "bbox": [
+            174,
+            103,
+            825,
+            186
+        ],
+        "page_idx": 5
+    },
+    {
+        "type": "text",
+        "text": "3.4 IMPLEMENTATION DETAILS ",
+        "text_level": 1,
+        "bbox": [
+            176,
+            204,
+            405,
+            218
+        ],
+        "page_idx": 5
+    },
+    {
+        "type": "text",
+        "text": "As we conduct experiments on image and tabular datasets, we use different architectures for each. For image datasets, we used a generator and discriminator architectures similar to the ones suggested by Wang et al. (2018). However, we improved these architectures using self-attention layers (Zhang et al., 2018). It should be noted that, while Zhang et al. (2018) suggests using a single self-attention layer in the middle of the network, we observed consistent improvements by inserting multiple self-attention layers before each residual block within the network. Furthermore, as input to the generator, we concatenate input image, mask, and a random $_ z$ frame along the channels dimension and use it as input. For tabular datasets, we use a simple 4 layer network consisting of fully-connected and batch-norm layers. Also, the input to the generator is the concatenation of a feature vector, mask vector, and a $_ { z }$ vector of size $\\frac { 1 } { 8 }$ of the input. For all experiments, we use an ensemble size $( N )$ ) equal to 128. ",
+        "bbox": [
+            173,
+            231,
+            825,
+            383
+        ],
+        "page_idx": 5
+    },
+    {
+        "type": "text",
+        "text": "We used Adam (Kingma & Ba, 2014) for model optimization. Two time-scale update rule (TTUR) (Heusel et al., 2017) was used to balance training the generator and discriminator networks. We explored best TTUR learning-rate settings from the set of {0.001, 0.0005, 0.0001, 0.00005}. Here, Adam parameters $\\beta _ { 1 }$ and $\\beta _ { 2 }$ are set to 0.5 and 0.999, respectively. Also, spectral normalization was used to stabilize both the generator and discriminator network in our experiments with image data (Miyato et al., 2018). For the predictor network, we used the default Adam settings as suggested by Kingma & Ba (2014). In all training procedures, we decay learning rate by a factor of 5 after reaching a plateau. For all experiments, we use a batch size of 64. Based on our experiments, we found that pretraining the discriminator while fixing the generator network for the first $5 \\%$ of the training epochs helps the stability of training. ",
+        "bbox": [
+            173,
+            390,
+            825,
+            530
+        ],
+        "page_idx": 5
+    },
+    {
+        "type": "text",
+        "text": "Further detail on exact architectures, experiments, software dependencies, etc. as well as ablation studies is provided in the appendices. ",
+        "bbox": [
+            174,
+            536,
+            823,
+            564
+        ],
+        "page_idx": 5
+    },
+    {
+        "type": "text",
+        "text": "4 EXPERIMENTS ",
+        "text_level": 1,
+        "bbox": [
+            176,
+            585,
+            326,
+            601
+        ],
+        "page_idx": 5
+    },
+    {
+        "type": "text",
+        "text": "4.1 DATASETS ",
+        "text_level": 1,
+        "bbox": [
+            174,
+            616,
+            287,
+            631
+        ],
+        "page_idx": 5
+    },
+    {
+        "type": "text",
+        "text": "To evaluate the proposed method we use CIFAR-10 (Krizhevsky & Hinton, 2009) as an image classification dataset as well as three non-image datasets: UCI Landsat (Dua & Graff, 2017)2, MIT-BIH arrhythmia (Moody & Mark, 2001), and Diabetes classification (Kachuee et al., 2019) 3. CIFAR-10 dataset consists of $6 0 , 0 0 0 \\ 3 2 \\mathrm { x } \\ 3 2$ images from 10 different classes. For this task, we use train and test sets as provided by the dataset. As a preprocessing step, we normalize pixel values to the range of [0,1] and subtract the mean image. The only data augmentation we use for this task is to randomly flip training images for each batch. ",
+        "bbox": [
+            174,
+            642,
+            825,
+            739
+        ],
+        "page_idx": 5
+    },
+    {
+        "type": "text",
+        "text": "UCI Landsat consists of 6435 samples of 36 features from 6 different categories. We follow the same train and test split as provided by the dataset. MIT-BIH dataset consists of annotated heartbeat signals from which we used the preprocessed version available online4 consisting of 92062 samples of 5 different arrhythmia classes. Diabetes dataset is a real-world health dataset of 92,062 samples and 45 features from different categories such as questionnaire, demographics, medical examination, and lab results. The objective is to classify between three different diabetes conditions i.e., normal, pre-diabetes, and diabetes. As MIT-BIH and Diabetes datasets do not provide explicit train and test sets, we randomly select $80 \\%$ of samples as a training set and the rest as a test set. To preprocess our tabular datasets, statistical and unity based normalization are used to balance the variance of different features and center them around zero. Also, while different encoding and representation methods are suggested in the literature to handle categorical features (Jang et al., 2016; Nazabal et al., 2018), in this paper, we take the simple approach of encoding categorical variables using one-hot representation and smoothing them by adding Gaussian noise with zero mean and variance equal to $5 \\%$ of feature variances. In our experiments, we observed a reasonable performance using the suggested simple smoothing; however, more advanced encoding methods are also applicable in this setup and can be applied to enhance the performances even further. ",
+        "bbox": [
+            174,
+            746,
+            825,
+            872
+        ],
+        "page_idx": 5
+    },
+    {
+        "type": "text",
+        "text": "",
+        "bbox": [
+            174,
+            103,
+            825,
+            202
+        ],
+        "page_idx": 6
+    },
+    {
+        "type": "text",
+        "text": "4.2 MISSINGNESS MECHANISMS ",
+        "text_level": 1,
+        "bbox": [
+            176,
+            252,
+            413,
+            266
+        ],
+        "page_idx": 6
+    },
+    {
+        "type": "text",
+        "text": "In our experiments, we consider MCAR uniform and MCAR rectangular missingness structures. In MCAR uniform, each feature of each sample is missing based on a Bernoulli distribution with a certain missingness probability (i.e., missing rate) independent of other features. In addition to the case of uniform missingness, for image tasks, we use rectangular missingness/observation structure where rectangular regions of dataset images are missing/observed. To control the rate of missingness and decide on the regions that are missing for each case, we use a latent beta distribution that samples rectangular region’s width and height such that the average missing rate is maintained. For missing rates less than $50 \\%$ we make the assumption of having a random rectangular region to be missing, whereas for missing rates more than $50 \\%$ we assume that only a random rectangular region is observed and the rest of the image is missing. ",
+        "bbox": [
+            174,
+            291,
+            825,
+            430
+        ],
+        "page_idx": 6
+    },
+    {
+        "type": "text",
+        "text": "We would like to note that while the suggested solution in this paper is readily compatible with MAR structures, in our experiments, to simplify the presentation of results and to have a fair comparison with other work that does not support the MAR assumption, we limited the scope of our experiments to MCAR. Furthermore, to simulate incomplete datasets and to make sure the same features are missing without explicitly storing masks, we use hashed feature vectors to seed random number generators used to sample missing features. More detail is provided in Appendix C. ",
+        "bbox": [
+            174,
+            438,
+            825,
+            521
+        ],
+        "page_idx": 6
+    },
+    {
+        "type": "text",
+        "text": "4.3 EVALUATION MEASURES ",
+        "text_level": 1,
+        "bbox": [
+            174,
+            571,
+            388,
+            585
+        ],
+        "page_idx": 6
+    },
+    {
+        "type": "text",
+        "text": "Fréchet inception distance (FID) (Heusel et al., 2017) score is used to measure the quality of missing data imputation in experiments with images5. We also considered using root means squared error (RMSE); however, we decided not to use this measure as we observed an inconsistent behavior using RMSE in our comparisons as RMSE favors methods that show less variance rather than realistic and sharp samples from the distribution. Also, for each dataset and each missingness scenario, we report top-1 classification accuracy based on the majority vote estimated using Algorithm 2. Another measure that we use in this paper is the comparison between the estimated target certainties and average accuracies achieved for each confidence assignment. We run each experiment multiple times: 4 times for CIFAR-10 and 8 times for tabular datasets. We report the mean and standard deviation of results for each case. ",
+        "bbox": [
+            174,
+            611,
+            825,
+            750
+        ],
+        "page_idx": 6
+    },
+    {
+        "type": "text",
+        "text": "We compare our results with MisGAN (Li et al., 2019) and GAIN (Yoon et al., 2018) as the state of the art imputation algorithms based on GANs as well as basic denoising autoencoder (DAE) (Vincent et al., 2008) and multiple imputation by chained equations (MICE) (Buuren & Groothuis-Oudshoorn, 2010) as baselines. For experiments using MisGAN, we used the same architectures and hyperparameters as suggested by the MisGAN authors6. The only modification was to adapt the last generator layer to generate images with resolutions as we use. Regarding GAIN, we used the same network architecture as our implementation of GI and hyper-parameters as used by the GAIN authors7. In the DAE implementation, due to the incomplete data assumption, only observed features appear in the loss function, ignoring reconstruction terms corresponding to missing features. Due to scalability issues, we were only able to use MICE for the smaller non-image datasets. For these methods, to train and evaluate classifiers, we use predictors trained on imputed datasets rather than the stochastic predictor suggested in Algorithm 1. ",
+        "bbox": [
+            174,
+            757,
+            826,
+            924
+        ],
+        "page_idx": 6
+    },
+    {
+        "type": "image",
+        "img_path": "images/7dee5514f122f68c96a6de74132a186a9e0266bbd52176e7a7dd6f46d002cb83.jpg",
+        "image_caption": [
+            "Figure 2: Comparison of FID scores on CIFAR-10 dataset for (a) uniform and (b) rectangular missingness. Lower FID score is better. In many cases, variance values are very small and only observable by magnifying the figures. "
+        ],
+        "image_footnote": [],
+        "bbox": [
+            178,
+            104,
+            816,
+            305
+        ],
+        "page_idx": 7
+    },
+    {
+        "type": "text",
+        "text": "4.4 RESULTS ",
+        "text_level": 1,
+        "bbox": [
+            174,
+            387,
+            277,
+            401
+        ],
+        "page_idx": 7
+    },
+    {
+        "type": "text",
+        "text": "Figure 2 presents the comparison of FID scores on the CIFAR-10 dataset at different missing rates for uniform and rectangular missingness. As it can be inferred from these plots, GI outperforms other alternatives in all cases. Also, it can be seen that GAIN is able to provide more reasonable results for uniform missing data structure compared to MisGAN which is mainly effective in the rectangular missing data structure. One possible explanation for this behavior might be the fact that GAIN has an MSE loss term acting similar to an autoencoder loss smoothing noisy missing pixels. On the other hand, MisGAN tries to explicitly model missingness structure and is more successful in capturing a more structured missingness such as the case of a rectangular structure. Table 1 provides a comparison between the top-1 classification accuracy achieved using each method at different missing rates and structures. From this table, GI outperforms other work by achieving the best results in 5 out of 6 cases8. ",
+        "bbox": [
+            173,
+            412,
+            825,
+            565
+        ],
+        "page_idx": 7
+    },
+    {
+        "type": "text",
+        "text": "Table 2 presents a comparison of classification accuracies for Landsat, MIT-BIH, and Diabetes datasets at different missing rates. In the Landsat benchmark, GI outperforms other work in all cases. Regarding the MIT-BIH experiemts, GI outperforms other work for missing rates more than $30 \\%$ while achieving similar accuracies to GAIN for lower missing rates. In the diabetes classification task, GI appears to be most effective imputing missing rates more than $20 \\%$ . ",
+        "bbox": [
+            174,
+            573,
+            825,
+            642
+        ],
+        "page_idx": 7
+    },
+    {
+        "type": "text",
+        "text": "Figure 3 shows a comparison of accuracy versus certainty plots for GI, MisGAN, and GAIN on Landsat dataset at the missing rate of $40 \\%$ . To generate these figures we trained each imputation method and then used Algorithm 1 to train predictors on imputed samples. Finally, Algorithm 2 used to measure the average accuracy at different prediction confidence levels based on a sample of 128 imputations for each test example. As it can be seen from the plots, GI provides results closest to the ideal case of having average confidence values equal to average accuracies. ",
+        "bbox": [
+            173,
+            650,
+            825,
+            734
+        ],
+        "page_idx": 7
+    },
+    {
+        "type": "text",
+        "text": "4.5 VISUALIZATION USING SYNTHESIZED DATA ",
+        "text_level": 1,
+        "bbox": [
+            174,
+            751,
+            519,
+            765
+        ],
+        "page_idx": 7
+    },
+    {
+        "type": "text",
+        "text": "In order to provide further insight into the operation of GI and how imputations can potentially influence the outcomes of predictions, we conduct experiments on a synthesized dataset. The original underlying data distribution is generated by sampling 5000 samples from 4 Gaussians of standard deviation 0.1 centered on the vertices of a unit square. We assign two different classes to each cluster ",
+        "bbox": [
+            174,
+            776,
+            825,
+            833
+        ],
+        "page_idx": 7
+    },
+    {
+        "type": "table",
+        "img_path": "images/542a5e4e556ade099fa91694dab860b1113629d1e9a5da54d4590b4438731691.jpg",
+        "table_caption": [
+            "Table 1: Top-1 CIFAR-10 classification accuracy for different missing rates and structures. "
+        ],
+        "table_footnote": [],
+        "table_body": "<table><tr><td rowspan=\"2\"></td><td colspan=\"5\">Accuracy at Missing Rate (%) MCAR Rect.</td><td rowspan=\"2\"></td></tr><tr><td colspan=\"2\">MCARUniform</td><td colspan=\"3\"></td></tr><tr><td>Method</td><td>20%</td><td>40%</td><td>60%</td><td>20%</td><td>40%</td><td>60%</td></tr><tr><td>GI</td><td>89.5 (±0.45)</td><td>87.1 (±0.54)</td><td>80.3 (±0.26)</td><td>84.0 (±0.03)</td><td>76.9 (±0.03)</td><td>66.1 (±0.16)</td></tr><tr><td>MisGAN</td><td>86.5 (±0.31)</td><td>83.7 (±0.40)</td><td>78.7 (±0.26)</td><td>82.9 (±0.44)</td><td>75.6 (±0.20)</td><td>65.0 (±0.31)</td></tr><tr><td>GAIN</td><td>88.7 (±0.45)</td><td>86.0 (±0.86)</td><td>81.8 (±0.03)</td><td>81.7 (±0.03)</td><td>73.6 (±0.35)</td><td>58.4 (±1.66)</td></tr><tr><td>DAE</td><td>88.0 (±0.22)</td><td>84.0 (±0.50)</td><td>79.8 (±0.71)</td><td>83.3 (±0.64)</td><td>75.5 (±0.44)</td><td>63.8 (±0.24)</td></tr><tr><td>Mean</td><td>85.7 (±0.02)</td><td>83.4 (±0.38)</td><td>79.2 (±0.16)</td><td>82.7 (±0.15)</td><td>75.3 (±0.16)</td><td>64.0 (±0.32)</td></tr></table>",
+        "bbox": [
+            176,
+            127,
+            821,
+            292
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "table",
+        "img_path": "images/20ad6cf606f031aa541edca374af1e1c4f0f167e153165882ca7b2dd13238666.jpg",
+        "table_caption": [
+            "Table 2: Comparison of classification accuracies for Landsat, MIT-BIH, and Diabetes datasets at different missing rates. "
+        ],
+        "table_footnote": [
+            "aBaseline accuracies for complete datasets (zero missing rate) are equal to $9 0 . 9 \\%$ , $9 8 . 6 \\%$ , and $9 0 . 7 \\%$ for Landsat, MIT-BIH and Diabetes, respectively. "
+        ],
+        "table_body": "<table><tr><td></td><td></td><td colspan=\"4\"> Accuracy at Missing Rate (%)a</td></tr><tr><td>Dataset</td><td>Method</td><td>10%</td><td>20%</td><td>30%</td><td>40%</td></tr><tr><td rowspan=\"5\">Landsat (Dua &amp; Graff, 2017)</td><td>GI</td><td>89.9 (±0.36)</td><td>89.6 (±0.36)</td><td>89.0 (±0.03)</td><td>88.0 (±0.22)</td></tr><tr><td>MisGAN</td><td>87.2 (±0.01)</td><td>85.7 (±0.19)</td><td>84.0 (±0.61)</td><td>82.9 (±0.75)</td></tr><tr><td>GAIN</td><td>89.7 (±0.42)</td><td>89.4 (±0.56)</td><td>88.4 (±0.71)</td><td>87.7 (±0.10)</td></tr><tr><td>DAE</td><td>89.4 (±0.10)</td><td>88.6 (±0.54)</td><td>87.5 (±0.14)</td><td>86.6 (±0.21)</td></tr><tr><td>MICE</td><td>89.5 (±0.16)</td><td>89.3 (±0.10)</td><td>88.1 (±0.49)</td><td>87.5 (±0.03)</td></tr><tr><td rowspan=\"5\">MIT-BIH(Moody &amp; Mark,2001)</td><td>GI</td><td>98.5 (±0.02)</td><td>98.4 (±0.03)</td><td>98.2 (±0.07)</td><td>97.7 (±0.03)</td></tr><tr><td>MisGAN</td><td>97.8 (±0.13)</td><td>97.4 (±0.07)</td><td>96.7 (±0.07)</td><td>96.2 (±0.09)</td></tr><tr><td>GAIN</td><td>98.5 (±0.02)</td><td>98.4 (±0.06)</td><td>98.0 (±0.09)</td><td>97.5 (±0.18)</td></tr><tr><td>DAE</td><td>98.4 (±0.02)</td><td>98.2 (±0.11)</td><td>97.9 (±0.09)</td><td>97.4 (±0.02)</td></tr><tr><td>MICE</td><td>98.4 (±0.01)</td><td>98.3 (±0.01)</td><td>98.1 (±0.01)</td><td>97.5 (±0.12)</td></tr><tr><td rowspan=\"5\">Diabetes (Kachuee et al., 2019)</td><td>GI</td><td>89.6 (±0.13)</td><td>89.0 (±0.03)</td><td>88.2 (±0.62)</td><td>86.8 (±0.38)</td></tr><tr><td>MisGAN</td><td>89.7 (±0.01)</td><td>88.9 (±0.30)</td><td>87.6 (±0.02)</td><td>86.4 (±0.68)</td></tr><tr><td>GAIN</td><td>89.2 (±0.09)</td><td>88.3 (±0.02)</td><td>86.9 (±0.09)</td><td>83.8 (±1.44)</td></tr><tr><td>DAE</td><td>89.3 (±0.05)</td><td>88.2 (±0.19)</td><td>86.9 (±0.09)</td><td>84.8 (±0.03)</td></tr><tr><td>MICE</td><td>89.8 (±0.08)</td><td>88.8 (±0.01)</td><td>88.0 (±0.08)</td><td>86.1 (±0.02)</td></tr></table>",
+        "bbox": [
+            176,
+            353,
+            823,
+            654
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "such that diagonal vertices are of the same class (see Figure 4a, classes are represented with colors).   \nFrom this underlying distribution, we make an incomplete dataset with $50 \\%$ of values missing. ",
+        "bbox": [
+            174,
+            722,
+            825,
+            750
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "The incomplete synthesized dataset is used to train GI and other imputation methods. We take a random test sample in which the second feature has a value of about 0.1 and the other feature is missing. Ideally, in the imputation phase, we would like to sample from the condition distribution i.e. $P ( x _ { 1 } | x _ { 2 } = 0 . 1 )$ (see Figure 4b). Here, in the prediction phase, an ideal method would decide on not making a confident classification and report the uncertainty. Note that solely observing the value of 0.1 for the second feature does not provide any useful evidence for the prediction. Figure 4c-f provide samples and classification results for GI, MisGAN, GAIN, and DAE. As it can be inferred from these figures, GI generates reasonable samples from the conditional distribution and also reflects this uncertainty over the prediction. On the other hand MisGAN, probably due to its complexity of using three different generators and discriminator pairs, is suffering from mode collapse and is unable to generate samples from the other class, resulting in over-confident assignments. GAIN, perhaps due to the MSE loss terms, is inclined towards the mean of the conditional distribution at the origin. ",
+        "bbox": [
+            173,
+            757,
+            825,
+            922
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "image",
+        "img_path": "images/99ff87491c12b80a83f83e95b0f6597c20ef1cd8b2470028e19f5844c61fce0b.jpg",
+        "image_caption": [
+            "Figure 3: Accuracy versus certainty plots for (a) GI, (b) MisGAN, and (c) GAIN on Landsat dataset at the missing rate of $40 \\%$ . "
+        ],
+        "image_footnote": [],
+        "bbox": [
+            214,
+            107,
+            782,
+            247
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "DAE, as expected, due to its MSE loss term, only captures the expected value of the distribution mean hence reducing the MSE error and generates over-smoothed imputations. ",
+        "bbox": [
+            173,
+            325,
+            825,
+            356
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "image",
+        "img_path": "images/efa63b3362bf0eec014da42a29ee10be1a5bc69514aec9b20546466115db1b16.jpg",
+        "image_caption": [
+            "Figure 4: Evaluation using synthesized data: (a) samples from the underlying distribution, (b) samples from the conditional underlying distribution, (c-f) samples from the conditional distribution generate by GI, MisGAN, GAIN, and DAE. "
+        ],
+        "image_footnote": [],
+        "bbox": [
+            207,
+            378,
+            784,
+            671
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "5 CONCLUSION ",
+        "text_level": 1,
+        "bbox": [
+            176,
+            775,
+            318,
+            791
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "In this paper, we proposed a novel method to generate imputations and measure uncertainties over target class assignments based on incomplete feature vectors. We evaluated the effectiveness of the suggested approach on image and tabular data via using different measures such as FID distance, classification accuracy, and confidence versus accuracy plots. According to the experiments, the proposed method not only can generate accurate imputations but also is able to model prediction uncertainties arising from missing values. The proposed method is applicable to many real-world applications where only an incomplete dataset is available, and modeling classification uncertainties is a necessity. ",
+        "bbox": [
+            174,
+            811,
+            825,
+            924
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "REFERENCES ",
+        "text_level": 1,
+        "bbox": [
+            176,
+            102,
+            287,
+            117
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Theodore W Anderson. Maximum likelihood estimates for a multivariate normal distribution when some observations are missing. Journal of the american Statistical Association, 52(278):200–203, 1957. ",
+        "bbox": [
+            174,
+            126,
+            826,
+            167
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Sanjeev Arora, Andrej Risteski, and Yi Zhang. Do gans learn the distribution? some theory and empirics. 2018. ",
+        "bbox": [
+            173,
+            178,
+            823,
+            207
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "S van Buuren and Karin Groothuis-Oudshoorn. mice: Multivariate imputation by chained equations in r. Journal of statistical software, pp. 1–68, 2010. ",
+        "bbox": [
+            173,
+            215,
+            825,
+            246
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Pengfei Chen, Benben Liao, Guangyong Chen, and Shengyu Zhang. Understanding and utilizing deep neural networks trained with noisy labels. arXiv preprint arXiv:1905.05040, 2019. ",
+        "bbox": [
+            173,
+            253,
+            823,
+            284
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Adnan Darwiche. Modeling and reasoning with Bayesian networks. Cambridge university press, 2009. ",
+        "bbox": [
+            171,
+            291,
+            825,
+            320
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Dheeru Dua and Casey Graff. UCI machine learning repository, 2017. URL http://archive. ics.uci.edu/ml. ",
+        "bbox": [
+            173,
+            329,
+            826,
+            359
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Vincent Dumoulin, Ishmael Belghazi, Ben Poole, Olivier Mastropietro, Alex Lamb, Martin Arjovsky, and Aaron Courville. Adversarially learned inference. arXiv preprint arXiv:1606.00704, 2016. ",
+        "bbox": [
+            171,
+            367,
+            825,
+            398
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Ian Goodfellow, Jean Pouget-Abadie, Mehdi Mirza, Bing Xu, David Warde-Farley, Sherjil Ozair, Aaron Courville, and Yoshua Bengio. Generative adversarial nets. In Advances in neural information processing systems, pp. 2672–2680, 2014. ",
+        "bbox": [
+            174,
+            406,
+            825,
+            449
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Trevor Hastie, Robert Tibshirani, Gavin Sherlock, Michael Eisen, Patrick Brown, and David Botstein. Imputing missing data for gene expression arrays, 1999. ",
+        "bbox": [
+            174,
+            458,
+            823,
+            488
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Martin Heusel, Hubert Ramsauer, Thomas Unterthiner, Bernhard Nessler, and Sepp Hochreiter. Gans trained by a two time-scale update rule converge to a local nash equilibrium. In Advances in Neural Information Processing Systems, pp. 6626–6637, 2017. ",
+        "bbox": [
+            173,
+            496,
+            825,
+            540
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Eric Jang, Shixiang Gu, and Ben Poole. Categorical reparameterization with gumbel-softmax. arXiv preprint arXiv:1611.01144, 2016. ",
+        "bbox": [
+            173,
+            547,
+            825,
+            578
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Mohammad Kachuee, Kimmo Karkkainen, Orpaz Goldstein, Davina Zamanzadeh, and Majid Sarrafzadeh. Nutrition and health data for cost-sensitive learning. arXiv preprint arXiv:1902.07102, 2019. ",
+        "bbox": [
+            173,
+            585,
+            826,
+            628
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Diederik P Kingma and Jimmy Ba. Adam: A method for stochastic optimization. arXiv preprint arXiv:1412.6980, 2014. ",
+        "bbox": [
+            173,
+            638,
+            825,
+            667
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Diederik P Kingma and Max Welling. Auto-encoding variational bayes. arXiv preprint arXiv:1312.6114, 2013. ",
+        "bbox": [
+            173,
+            676,
+            825,
+            705
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Alex Krizhevsky and Geoffrey Hinton. Learning multiple layers of features from tiny images. Technical report, Citeseer, 2009. ",
+        "bbox": [
+            171,
+            714,
+            825,
+            744
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Steven Cheng-Xian Li, Bo Jiang, and Benjamin Marlin. Learning from incomplete data with generative adversarial networks. In International Conference on Learning Representations, 2019. URL https://openreview.net/forum?id $=$ S1lDV3RcKm. ",
+        "bbox": [
+            176,
+            752,
+            825,
+            796
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Roderick JA Little and Donald B Rubin. Statistical analysis with missing data, volume 793. Wiley, 2019. ",
+        "bbox": [
+            173,
+            804,
+            825,
+            833
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Shuang Liu, Olivier Bousquet, and Kamalika Chaudhuri. Approximation and convergence properties of generative adversarial learning. In Advances in Neural Information Processing Systems, pp. 5545–5553, 2017. ",
+        "bbox": [
+            173,
+            843,
+            823,
+            886
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Pierre-Alexandre Mattei and Jes Frellsen. missiwae: Deep generative modelling and imputation of incomplete data. arXiv preprint arXiv:1812.02633, 2018. ",
+        "bbox": [
+            173,
+            895,
+            821,
+            924
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Takeru Miyato, Toshiki Kataoka, Masanori Koyama, and Yuichi Yoshida. Spectral normalization for generative adversarial networks. arXiv preprint arXiv:1802.05957, 2018. ",
+        "bbox": [
+            171,
+            103,
+            825,
+            133
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "George B Moody and Roger G Mark. The impact of the mit-bih arrhythmia database. IEEE Engineering in Medicine and Biology Magazine, 20(3):45–50, 2001. ",
+        "bbox": [
+            171,
+            140,
+            823,
+            171
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "Jared S Murray et al. Multiple imputation: a review of practical and theoretical findings. Statistical Science, 33(2):142–159, 2018. ",
+        "bbox": [
+            171,
+            179,
+            825,
+            208
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "Nagarajan Natarajan, Inderjit S Dhillon, Pradeep K Ravikumar, and Ambuj Tewari. Learning with noisy labels. In Advances in neural information processing systems, pp. 1196–1204, 2013. ",
+        "bbox": [
+            173,
+            215,
+            823,
+            246
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "Alfredo Nazabal, Pablo M Olmos, Zoubin Ghahramani, and Isabel Valera. Handling incomplete heterogeneous data using vaes. arXiv preprint arXiv:1807.03653, 2018. ",
+        "bbox": [
+            173,
+            253,
+            823,
+            284
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "Richard E Neapolitan et al. Learning bayesian networks, volume 38. Pearson Prentice Hall Upper Saddle River, NJ, 2004. ",
+        "bbox": [
+            176,
+            291,
+            825,
+            320
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "Thomas Dyhre Nielsen and Finn Verner Jensen. Bayesian networks and decision graphs. Springer Science & Business Media, 2009. ",
+        "bbox": [
+            173,
+            329,
+            823,
+            359
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "Scott Reed, Honglak Lee, Dragomir Anguelov, Christian Szegedy, Dumitru Erhan, and Andrew Rabinovich. Training deep neural networks on noisy labels with bootstrapping. arXiv preprint arXiv:1412.6596, 2014. ",
+        "bbox": [
+            174,
+            367,
+            823,
+            410
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "Donald B Rubin. Inference and missing data. Biometrika, 63(3):581–592, 1976. ",
+        "bbox": [
+            173,
+            419,
+            700,
+            435
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "Joseph L Schafer and John W Graham. Missing data: our view of the state of the art. Psychological methods, 7(2):147, 2002. ",
+        "bbox": [
+            171,
+            443,
+            823,
+            472
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "Kihyuk Sohn, Honglak Lee, and Xinchen Yan. Learning structured output representation using deep conditional generative models. In Advances in neural information processing systems, pp. 3483–3491, 2015. ",
+        "bbox": [
+            173,
+            481,
+            825,
+            523
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "Pascal Vincent, Hugo Larochelle, Yoshua Bengio, and Pierre-Antoine Manzagol. Extracting and composing robust features with denoising autoencoders. In Proceedings of the 25th international conference on Machine learning, pp. 1096–1103. ACM, 2008. ",
+        "bbox": [
+            173,
+            531,
+            826,
+            575
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "Ting-Chun Wang, Ming-Yu Liu, Jun-Yan Zhu, Andrew Tao, Jan Kautz, and Bryan Catanzaro. Highresolution image synthesis and semantic manipulation with conditional gans. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pp. 8798–8807, 2018. ",
+        "bbox": [
+            174,
+            583,
+            826,
+            627
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "Jinsung Yoon, James Jordon, and Mihaela Van Der Schaar. Gain: Missing data imputation using generative adversarial nets. arXiv preprint arXiv:1806.02920, 2018. ",
+        "bbox": [
+            174,
+            635,
+            823,
+            665
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "Han Zhang, Ian Goodfellow, Dimitris Metaxas, and Augustus Odena. Self-attention generative adversarial networks. arXiv preprint arXiv:1805.08318, 2018. ",
+        "bbox": [
+            173,
+            672,
+            823,
+            702
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "Shichao Zhang, Zhenxing Qin, Charles X Ling, and Shengli Sheng. \" missing is useful\": missing values in cost-sensitive decision trees. IEEE transactions on knowledge and data engineering, 17 (12):1689–1693, 2005. ",
+        "bbox": [
+            174,
+            710,
+            825,
+            753
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "A IMPLEMENTATION AND EXPERIMENTS ",
+        "text_level": 1,
+        "bbox": [
+            176,
+            102,
+            531,
+            118
+        ],
+        "page_idx": 12
+    },
+    {
+        "type": "text",
+        "text": "Table 3 presents the list of software dependencies and versions used in our implementation. To produce results related to this paper, we used a workstation with 4 NVIDIA GeForce RTX-2080Ti GPUs, a 12 core Intel Core i9-7920X processor, and 128 GB memory. Each experiment took between about 4 hours to 48 hours, based on the task and method being tested. ",
+        "bbox": [
+            173,
+            136,
+            825,
+            193
+        ],
+        "page_idx": 12
+    },
+    {
+        "type": "table",
+        "img_path": "images/c34dd7ed960160ceeeb9080e4ac592a923c1d401524807a6a19321bc3e36daa6.jpg",
+        "table_caption": [
+            "Table 3: Software dependencies. "
+        ],
+        "table_footnote": [],
+        "table_body": "<table><tr><td>Dependency</td><td>Version</td></tr><tr><td>python</td><td>3.7.1</td></tr><tr><td>pytorch</td><td>1.0.0</td></tr><tr><td>cuda100</td><td>1.0</td></tr><tr><td>ipython</td><td>7.2.0</td></tr><tr><td>jupyter</td><td>1.0.0</td></tr><tr><td>numpy</td><td>1.15.4</td></tr><tr><td>pandas</td><td>0.24.1</td></tr><tr><td>scikit-learn</td><td>0.20.1</td></tr><tr><td>scipy</td><td>1.1.0</td></tr><tr><td>torchvision</td><td>0.2.1</td></tr><tr><td>tqdm</td><td>4.28.1</td></tr><tr><td>matplotlib</td><td>3.0.1</td></tr></table>",
+        "bbox": [
+            387,
+            237,
+            617,
+            481
+        ],
+        "page_idx": 12
+    },
+    {
+        "type": "text",
+        "text": "B NETWORK ARCHITECTURES ",
+        "text_level": 1,
+        "bbox": [
+            176,
+            520,
+            444,
+            536
+        ],
+        "page_idx": 12
+    },
+    {
+        "type": "text",
+        "text": "Table 4 shows the exact architectures used in this paper. To show each layer or block we used the following notation. $\\mathtt { C x S y P z - t }$ represents a 2-d convolution layer of kernel size $_ \\textrm { x }$ , stride y, padding $_ { \\textrm { Z } }$ , and number of output channels $\\scriptstyle \\pm$ followed by ReLU activation. Attn represents a self-attention layer similar to Zhang et al.9. $_ \\mathrm { R - x }$ represents a residual block consisting of two 2-d convolutions with kernel size 3 (padding size 1), batch normalization, and ReLU activation. $\\mathsf { C T x S y P z - t }$ is the convolution transpose corresponding to $\\mathtt { C x S y P z - t }$ . $\\operatorname { F C - x }$ is representing a linear fully-connected layer of $_ \\textrm { x }$ output neurons with biases. We use spectral normalization as suggested by Miyato et al.10 for all convolutional layers in both generator and discriminator networks. ",
+        "bbox": [
+            173,
+            554,
+            826,
+            666
+        ],
+        "page_idx": 12
+    },
+    {
+        "type": "table",
+        "img_path": "images/1dc4ede59e46fd6fe85ec060a8aeed53fa3365bbe7e3af9f8c634bf3d02c9b6d.jpg",
+        "table_caption": [
+            "Table 4: Network architectures used in our experiments. "
+        ],
+        "table_footnote": [],
+        "table_body": "<table><tr><td rowspan=1 colspan=1>Dataset</td><td rowspan=1 colspan=1>Generator/Discriminator Architecture</td><td rowspan=1 colspan=1>Predictor Architecture</td></tr><tr><td rowspan=1 colspan=1>CIFAR-10</td><td rowspan=1 colspan=1>C7S1P3-64，C3S2P1-128，Attn，R-128,Attn，R-128，Attn，R-128，Attn,R-128，CT3S2P1-128，CT7S1P3-3，Tanh/Sigmoid</td><td rowspan=1 colspan=1>ResNet-18 11,12</td></tr><tr><td rowspan=1 colspan=1>Landsat</td><td rowspan=1 colspan=1>FC-64，Sigmoid，BNorm，FC-64，Sigmoid,BNorm,FC-64，Sigmoid，BNorm，FC-36，Tanh/Sigmoid</td><td rowspan=1 colspan=1>FC-64，ReLU，BNorm，FC-64,ReLU，BNorm，FC-6，Softmax</td></tr><tr><td rowspan=1 colspan=1>MIT-BIH</td><td rowspan=1 colspan=1>FC-1860，ReLU，BNorm，FC-1860，ReLU，BNorm,FC-1860，ReLU，BNorm, FC-186,Tanh/Sigmoid</td><td rowspan=1 colspan=1>FC-1860，ReLU，BNorm，FC-1860,ReLU，BNorm，FC-5，Softmax</td></tr><tr><td rowspan=1 colspan=1>Diabetes</td><td rowspan=1 colspan=1>FC-45，ReLU，BNorm，FC-45，ReLU，BNorm,FC-45，ReLU，BNorm，FC-45，Tanh/Sigmoid</td><td rowspan=1 colspan=1>FC-22，ReLU，BNorm，FC-22,ReLU，BNorm，FC-3，Softmax</td></tr></table>",
+        "bbox": [
+            174,
+            712,
+            821,
+            863
+        ],
+        "page_idx": 12
+    },
+    {
+        "type": "text",
+        "text": "C MISSING DATA MECHANISMS ",
+        "text_level": 1,
+        "bbox": [
+            176,
+            102,
+            455,
+            118
+        ],
+        "page_idx": 13
+    },
+    {
+        "type": "text",
+        "text": "In this paper, we conduct experiments on two mechanisms for missing values: MCAR uniform and MCAR rectangular. As in our experiments and comparisons, we consider the case where only an incomplete dataset is available for training. It is crucial to guarantee that each method has only access to a unique incomplete version of each sample. However, it is relatively expensive to load and store feature masks for each sample in the dataset. Instead, we generate missing values during the data load for each batch. A hashing mechanism is used to ensure that the same parts are missing for each sample throughout the training. Note that we set system, python, and external library hash seeds to fixed values to ensure the consistency between different runs. ",
+        "bbox": [
+            174,
+            133,
+            825,
+            244
+        ],
+        "page_idx": 13
+    },
+    {
+        "type": "text",
+        "text": "Algorithm 3 presents the procedure used for generating missing values with uniform structure. This algorithm is sampling independent Bernoulli distributions with probabilities equal to the missing rate. Algorithm 4 shows the outline for the rectangular missing structure used in image experiments. It consists of selecting a random point as the center of the rectangle and then deciding on parameters to be used for the beta distribution based on the missing rate. Finally, the width and height of the rectangular region are sampled from the latent beta distribution. In other words, we generate rectangular regions centered at random locations within the image which have width and height values determined by samples from a latent beta distribution. Here, distribution parameters, $\\alpha$ and $\\beta$ , are used to control the average missing rate. The outcome would be rectangular regions of different shape at different locations within the frame with the expected portion of missing area equal to the missing rate. ",
+        "bbox": [
+            174,
+            252,
+            825,
+            405
+        ],
+        "page_idx": 13
+    },
+    {
+        "type": "text",
+        "text": "In order to decide on the beta distribution parameters i.e. $\\alpha$ and $\\beta$ we use numerical simulations. Specifically, we fix one of the parameters to 1 and change the other parameter in the range of [1,10], while measuring the average missing rate caused by each case. Figure 5 shows the missing rates caused by different beta distribution parameters. The first half of Figure 5 (missing rates less than about 0.18) corresponds to setting $\\beta$ to 1 and changing $\\alpha$ values; and the other half fixing $\\alpha$ to 1 and changing $\\beta$ values. To generate missing rates more than $50 \\%$ we invert our masks and limit the observation to the rectangular region while the rest of the image is missing. Note that missing rates indicate the ratio of features that are missing on the average case. As we are using a latent model for sampling width and height for the rectangles, the actual missing ratios for each specific sample differs between samples. See Table 5 for visual examples of different missing rates and missing structures. ",
+        "bbox": [
+            174,
+            411,
+            825,
+            550
+        ],
+        "page_idx": 13
+    },
+    {
+        "type": "text",
+        "text": "Algorithm 3: MCAR uniform generation. ",
+        "text_level": 1,
+        "bbox": [
+            272,
+            570,
+            545,
+            584
+        ],
+        "page_idx": 13
+    },
+    {
+        "type": "text",
+        "text": "Input: $_ { \\textbf { \\em x } }$ (complete feature), $r$ (missing rate) Output: ${ \\pmb x } _ { m }$ (incomplete feature) $\\begin{array} { r l } & { s e e d _ { x }  h a s h ( \\pmb { x } ) } \\\\ & { \\pmb { k }  1 - B e r n o u l l i ( s e e d _ { x } , s h a p e ( x ) , p r o b = r ) } \\\\ & { \\pmb { x } _ { m }  \\pmb { k } \\odot \\pmb { x } + ( 1 - \\pmb { k } ) \\odot N a N } \\end{array}$ ",
+        "bbox": [
+            269,
+            588,
+            599,
+            660
+        ],
+        "page_idx": 13
+    },
+    {
+        "type": "text",
+        "text": "Algorithm 4: MCAR rect. generation. ",
+        "text_level": 1,
+        "bbox": [
+            272,
+            671,
+            524,
+            685
+        ],
+        "page_idx": 13
+    },
+    {
+        "type": "text",
+        "text": "Input: $_ { \\textbf { \\em x } }$ (complete feature), $r$ (missing rate)   \nOutput: ${ \\pmb x } _ { m }$ (incomplete feature)   \n$s e e d _ { x } \\gets h a s h ( { \\pmb x } )$   \nn $\\iota _ { x } , n _ { y } \\gets s h a p e ( { \\pmb x } )$   \n$( p _ { x } , \\bar { p _ { y } } ) \\sim ( u n i f o r m ( 0 , n _ { x } ) , u n i f o r m ( 0 , n _ { y } ) )$   \nα, β ← beta_params(r) // beta_params gives $\\alpha , \\beta$ for each missing rate based on numerical simulations   \n$\\begin{array} { r l } & { ( \\boldsymbol { w } , h ) \\sim ( B e t a ( \\alpha , \\beta ) \\times n _ { x } ) , B e t a ( \\alpha , \\beta ) \\times n _ { y } ) ) } \\\\ & { \\boldsymbol { k }  r e c t \\_ m a s k ( p _ { x } , p _ { y } , w , h ) } \\\\ & { \\mathbf { \\boldsymbol { x } } _ { m }  \\boldsymbol { k } \\odot \\mathbf { \\boldsymbol { x } } + ( 1 - \\boldsymbol { k } ) \\odot N a N } \\end{array}$ ",
+        "bbox": [
+            267,
+            688,
+            714,
+            844
+        ],
+        "page_idx": 13
+    },
+    {
+        "type": "image",
+        "img_path": "images/29b1ccc7a90bf84d0c4b18de35c4df32b98f105fff00abdb71b3fde7ca644a3b.jpg",
+        "image_caption": [
+            "Figure 5: Simulation results for measuring average missing rate given different beta distribution parameters. "
+        ],
+        "image_footnote": [],
+        "bbox": [
+            282,
+            189,
+            712,
+            404
+        ],
+        "page_idx": 14
+    },
+    {
+        "type": "image",
+        "img_path": "images/b85dc27446c318885a97542d42c9e6453ece8284078aca51ca167c500d3ddc30.jpg",
+        "image_caption": [],
+        "image_footnote": [],
+        "bbox": [
+            178,
+            646,
+            815,
+            840
+        ],
+        "page_idx": 14
+    },
+    {
+        "type": "text",
+        "text": "D ABLATION STUDY ",
+        "text_level": 1,
+        "bbox": [
+            176,
+            102,
+            361,
+            117
+        ],
+        "page_idx": 15
+    },
+    {
+        "type": "text",
+        "text": "Figure 6 presents a comparison between using (GI W/ Atten.) and not using (GI W/O Atten.) self-attention layers before each residual block in the proposed architecture. We report FID scores on CIFAR-10 with rectangular missingness. As it can be inferred from this comparison, using self-attention achieves a consistent improvement over the baseline. We also examined the case of uniform missingness; however, we did not observe any significant improvement for this case. One possible explanation could be the fact that imputing missing data with a uniform structure can be done by processing local regions and does not require attending to different distant regions across the image. ",
+        "bbox": [
+            173,
+            135,
+            826,
+            247
+        ],
+        "page_idx": 15
+    },
+    {
+        "type": "image",
+        "img_path": "images/555f58484ccab468a30f31b8a1a2e7e6951c3a282f1c5c77ff9c483cf318fc49.jpg",
+        "image_caption": [
+            "Figure 6: Comparison of FID scores achieved with (GI W/ Atten.) and without (GI W/O Atten.) self-attention layers on CIFAR-10 dataset and rectangular missingness. Lower FID score is better. "
+        ],
+        "image_footnote": [],
+        "bbox": [
+            277,
+            271,
+            715,
+            521
+        ],
+        "page_idx": 15
+    },
+    {
+        "type": "text",
+        "text": "Figure 7 shows a comparison of classification accuracies for the Landsat dataset achieved using different ensemble sizes $( N )$ . As it can be seen from this figure, higher values of $N$ result in improved accuracies, especially for higher missing rates. Also, it can be observed that for N values more than 64 the difference is negligible. ",
+        "bbox": [
+            173,
+            587,
+            825,
+            642
+        ],
+        "page_idx": 15
+    },
+    {
+        "type": "image",
+        "img_path": "images/12c219ac2d3a48f6cc52d091728437b57a1e9af830f5bd30babc7198f2bac84e.jpg",
+        "image_caption": [
+            "Figure 7: Comparison of classification accuracies achieved with different ensemble size $( N )$ . "
+        ],
+        "image_footnote": [],
+        "bbox": [
+            281,
+            666,
+            714,
+            886
+        ],
+        "page_idx": 15
+    },
+    {
+        "type": "text",
+        "text": "To study the benefits of the suggested stochastic predictor, we conducted experiments comparing GI with its non-stochastic variation $( \\mathrm { N } { = } 1 )$ ). Here, the CIFAR-10 dataset with the rectangular missing structure and missing rates from $20 \\%$ to as high as $90 \\%$ is used. From Table 6 it can be inferred that as the rate of missingness increases, the benefits of the suggested predictor algorithm increase significantly. We hypothesize that at higher rates of missingness, the conditional distribution of missing features becomes multimodal. In such a scenario, the suggested method captures the uncertainties over the target distribution resulting in the predictor to make more reliable class assignments. ",
+        "bbox": [
+            173,
+            103,
+            825,
+            215
+        ],
+        "page_idx": 16
+    },
+    {
+        "type": "table",
+        "img_path": "images/ae7388387b96894d1aab2c0bea5caf9ef83886b0f974a93634b3532bca3a445a.jpg",
+        "table_caption": [
+            "Table 6: Comparison of CIFAR-10 accuracies for the stochastic $\\mathrm { ( N = 1 2 8 ) }$ ) and the deterministic $\\left( \\mathrm { N } { = } 1 \\right)$ ) predictor under rectangular missingness. "
+        ],
+        "table_footnote": [],
+        "table_body": "<table><tr><td></td><td colspan=\"6\">Accuracy at Missing Rate (%)</td></tr><tr><td>Method</td><td>20%</td><td>40%</td><td>60%</td><td>70%</td><td>80%</td><td>90%</td></tr><tr><td>GI (N=128)</td><td>84.0</td><td>76.9</td><td>66.1</td><td>59.1</td><td>46.0</td><td>32.1</td></tr><tr><td>GI (N=1)</td><td>83.6</td><td>75.7</td><td>65.1</td><td>56.7</td><td>42.8</td><td>29.4</td></tr><tr><td>% difference (normalized)</td><td>0.5</td><td>1.6</td><td>1.5</td><td>4.1</td><td>6.9</td><td>8.4</td></tr></table>",
+        "bbox": [
+            238,
+            267,
+            758,
+            376
+        ],
+        "page_idx": 16
+    },
+    {
+        "type": "text",
+        "text": "E VISUAL COMPARISON ",
+        "text_level": 1,
+        "bbox": [
+            174,
+            102,
+            392,
+            118
+        ],
+        "page_idx": 17
+    },
+    {
+        "type": "text",
+        "text": "Table 7 and 8 provide a visual comparison of GI, MisGAN, and GAIN. For each missingness structure, we compare the best two imputation methods based on FID scores in Figure 2 i.e., GI versus MisGAN for rectangular missingness and GI versus GAIN for uniform missingness. From Table 7 it can be seen that GI is more capable in the reconstruction of fine details such as horse legs, car wheels, or plane wings. Regarding the results provided in Table 8, GI imputed samples are generally sharper and more realistic, which is consistent with our hypothesis about the drawbacks of the MSE term in the GAIN objective function. ",
+        "bbox": [
+            173,
+            133,
+            825,
+            231
+        ],
+        "page_idx": 17
+    },
+    {
+        "type": "image",
+        "img_path": "images/61024c4babfabb7e7dd65df430bff40654a51f9e986c3d422c05dfe4e0963228.jpg",
+        "image_caption": [],
+        "image_footnote": [],
+        "bbox": [
+            178,
+            279,
+            813,
+            813
+        ],
+        "page_idx": 17
+    },
+    {
+        "type": "image",
+        "img_path": "images/3958733293402574782f6e1c021c4a754f6dbed20e2753d3f34209f404938a06.jpg",
+        "image_caption": [],
+        "image_footnote": [],
+        "bbox": [
+            178,
+            252,
+            813,
+            801
+        ],
+        "page_idx": 18
+    },
+    {
+        "type": "text",
+        "text": "F ANALYSIS OF THE RMSE MEASURE ",
+        "text_level": 1,
+        "bbox": [
+            174,
+            102,
+            509,
+            118
+        ],
+        "page_idx": 19
+    },
+    {
+        "type": "text",
+        "text": "Table 9 presents the comparison of different imputation methods using the RMSE measure on CIFAR-10 for different missing structures and rates. Generally, RMSE values for the uniform missing structure are lower than their rectangular counterparts. It is consistent with our intuition that imputing uniform missingness is most similar to denoising problems where the RMSE measure is frequently used. Additionally, comparing the performance of different imputation methods using the FID measure (Section 4.4) does not demonstrate a clear correlation to results shown in Table 9. Nonetheless, it is well-known that the FID measure is more suited to measuring the performance of generated images from the underlying distribution (Heusel et al., 2017). ",
+        "bbox": [
+            173,
+            133,
+            825,
+            246
+        ],
+        "page_idx": 19
+    },
+    {
+        "type": "text",
+        "text": "Similarly, in Table 10, we provide RMSE values corresponding to experiments on the tabular datasets. Here, GAIN and DAE provide very similar results that are generally better than GI or MisGAN. This signifies our hypothesis that the MSE loss term may skew generated samples toward the mean of the distribution, resulting in better RMSE values but not necessarily higher final classification accuracies (see Table 2). ",
+        "bbox": [
+            174,
+            251,
+            825,
+            321
+        ],
+        "page_idx": 19
+    },
+    {
+        "type": "table",
+        "img_path": "images/529c815376d6c496fdf132f191539c1f32c20d0235e9256fc3ebe56ba40b8e1f.jpg",
+        "table_caption": [
+            "Table 9: Comparison of imputation RMSE values for CIFAR-10 at different missing structures and rates. "
+        ],
+        "table_footnote": [],
+        "table_body": "<table><tr><td></td><td colspan=\"5\">RMSE at Missing Rate (%)</td><td></td></tr><tr><td></td><td colspan=\"5\">MCAR Uniform</td><td>MCAR Rect.</td></tr><tr><td>Method</td><td>20%</td><td>40%</td><td>60%</td><td>20%</td><td>40%</td><td>60%</td></tr><tr><td>GI</td><td>0.026 (±0.003)</td><td>0.057 (±0.008)</td><td>0.090 (±0.006)</td><td>0.097 (±0.02)</td><td>0.148 (±0.001)</td><td>0.660 (±0.010)</td></tr><tr><td>MisGAN</td><td>0.079 (±0.001)</td><td>0.161 (±0.001)</td><td>0.257 (±0.002)</td><td>0.106 (±0.005)</td><td>0.158 (±0.004)</td><td>0.250 (±0.001)</td></tr><tr><td>GAIN</td><td>0.027 (±0.003)</td><td>0.045 (±0.001)</td><td>0.072 (±0.005)</td><td>0.340 (±0.047)</td><td>0.511 (±0.001)</td><td>0.660 (±0.010)</td></tr><tr><td>DAE</td><td>0.036 (±0.001)</td><td>0.075 (±0.002)</td><td>0.121 (±0.005)</td><td>0.116 (±0.007)</td><td>0.160 (±0.001)</td><td>0.233 (±0.029)</td></tr></table>",
+        "bbox": [
+            176,
+            376,
+            821,
+            505
+        ],
+        "page_idx": 19
+    },
+    {
+        "type": "table",
+        "img_path": "images/5a52e3e2e067bb0f4bbb49239b9639bda2bb046e5e27aa435b8404de869a0a9b.jpg",
+        "table_caption": [
+            "Table 10: Comparison of imputation RMSE values for Landsat, MIT-BIH, and Diabetes datasets at different missing rates. "
+        ],
+        "table_footnote": [],
+        "table_body": "<table><tr><td colspan=\"2\"></td><td colspan=\"4\">RMSE at Missing Rate (%)</td></tr><tr><td>Dataset</td><td>Method</td><td>10%</td><td>20%</td><td>30%</td><td>40%</td></tr><tr><td rowspan=\"4\">Landsat (Dua &amp; Graff,2017)</td><td>GI</td><td>0.040 (±0.005)</td><td>0.067 (±0.007)</td><td>0.076 (±0.020)</td><td>0.136 (±0.002)</td></tr><tr><td>MisGAN</td><td>0.068 (±0.001)</td><td>0.096 (±0.001)</td><td>0.118 (±0.001)</td><td>0.136 (±0.001)</td></tr><tr><td>GAIN</td><td>0.018 (±0.001)</td><td>0.024 (±0.001)</td><td>0.030 (±0.001)</td><td>0.037 (±0.001)</td></tr><tr><td>DAE</td><td>0.020 (±0.001)</td><td>0.031 (±0.001)</td><td>0.041 (±0.001)</td><td>0.052 (±0.001)</td></tr><tr><td rowspan=\"4\">MIT-BIH(Moody &amp; Mark,2001)</td><td>GI</td><td>0.038 (±0.001)</td><td>0.060 (±0.004)</td><td>0.071 (±0.002)</td><td>0.095 (±0.002)</td></tr><tr><td>MisGAN</td><td>0.073 (±0.007)</td><td>0.092 (±0.002)</td><td>0.115 (±0.003)</td><td>0.111 (±0.001)</td></tr><tr><td>GAIN</td><td>0.032 (±0.008)</td><td>0.046 (±0.001)</td><td>0.055 (±0.004)</td><td>0.067 (±0.007)</td></tr><tr><td>DAE</td><td>0.029 (±0.001)</td><td>0.048 (±0.008)</td><td>0.061 (±0.009)</td><td>0.068 (±0.003)</td></tr><tr><td rowspan=\"4\">Diabetes (Kachuee et al., 2019)</td><td>GI</td><td>0.080 (±0.002)</td><td>0.118 (±0.008)</td><td>0.149 (±0.020)</td><td>0.189 (±0.009)</td></tr><tr><td>MisGAN</td><td>0.082 (±0.004)</td><td>0.111 (±0.002)</td><td>0.133 (±0.001)</td><td>0.151 (±0.001)</td></tr><tr><td>GAIN</td><td>0.064 (±0.001)</td><td>0.092 (±0.001)</td><td>0.119 (±0.001)</td><td>0.140 (±0.001)</td></tr><tr><td>DAE</td><td>0.065 (±0.001)</td><td>0.093 (±0.001)</td><td>0.118 (±0.001)</td><td>0.143 (±0.001)</td></tr></table>",
+        "bbox": [
+            176,
+            568,
+            823,
+            795
+        ],
+        "page_idx": 19
+    },
+    {
+        "type": "text",
+        "text": "G IMPACT OF TRAINING NOISE ",
+        "text_level": 1,
+        "bbox": [
+            176,
+            102,
+            450,
+            118
+        ],
+        "page_idx": 20
+    },
+    {
+        "type": "text",
+        "text": "Addition of noise to input vectors often serves as an input augmentation and results in improved generalization accuracies. In order to verify that the improved GI performance is not merely due to the introduction of noise in the suggested architecture, we conducted an experiment by adding different amounts of Gaussian noise during the training process for GAIN and GI. Specifically, we compared how the CIFAR-10 test accuracies change at different degrees of training noise for uniform and rectangular missingess structures at the average missing rate of $40 \\%$ . ",
+        "bbox": [
+            174,
+            133,
+            825,
+            217
+        ],
+        "page_idx": 20
+    },
+    {
+        "type": "text",
+        "text": "According to Table 11, adding small amounts of Gaussian noise (e.g., std=0.0125) improves the generalization under uniform missingness for both GI and GAIN. Even in this case, GI is still outperforming GAIN in terms of final classification performance. It is also interesting to point out that for the case of rectangular missingness adding Gaussian noise results in a consistent reduction in the classification accuracy for both methods. ",
+        "bbox": [
+            174,
+            223,
+            825,
+            294
+        ],
+        "page_idx": 20
+    },
+    {
+        "type": "table",
+        "img_path": "images/d046f8a6d59738370e36a4b7922d58dd7a1f1787bb18517afbe2533c0193e688.jpg",
+        "table_caption": [
+            "Table 11: Top-1 CIFAR-10 classification accuracy at $40 \\%$ missing rate using added training noise. "
+        ],
+        "table_footnote": [],
+        "table_body": "<table><tr><td></td><td colspan=\"2\">Accuracy (%)</td></tr><tr><td>Noise STD</td><td>MCAR Uniform (40%) GI GAIN</td><td>MCAR Rect. (40%) GI GAIN</td></tr><tr><td>0.0</td><td>87.1 86.0</td><td>76.9 73.6</td></tr><tr><td>0.0125</td><td>87.3 86.3</td><td>76.8 73.3</td></tr><tr><td>0.025</td><td>86.5 86.6</td><td>76.7 73.2</td></tr><tr><td>0.05</td><td>85.6 84.7</td><td>73.7 72.4</td></tr><tr><td>0.1</td><td>82.0 80.6</td><td>68.7 67.0</td></tr></table>",
+        "bbox": [
+            281,
+            333,
+            717,
+            500
+        ],
+        "page_idx": 20
+    },
+    {
+        "type": "text",
+        "text": "H IMPACT OF THE MSE LOSS TERM ",
+        "text_level": 1,
+        "bbox": [
+            174,
+            102,
+            491,
+            118
+        ],
+        "page_idx": 21
+    },
+    {
+        "type": "text",
+        "text": "In our earlier discussions, we stated that the MSE loss term used in GAIN would bias the distribution of generated samples toward the mean of the distribution. Here, a synthesized dataset is used to illustrate the impact of MSE loss term on the distribution of generated samples. A hyperparameter, $\\lambda$ , controls the weight of the MSE term in the final objective function. As it can be observed from Figure 8, the higher the $\\lambda$ parameter, the lower the variance of the generated samples (i.e., more bias toward the mean of the distribution). ",
+        "bbox": [
+            173,
+            133,
+            826,
+            217
+        ],
+        "page_idx": 21
+    },
+    {
+        "type": "image",
+        "img_path": "images/e6f566e62f5c08205f49f8240037709b45a8d8f533c27ee98af701aa8fdcd47f.jpg",
+        "image_caption": [
+            "Figure 8: Comparison of generating samples from a Gaussian distribution (a) samples from the original distribution, (b) samples generated using GAIN imputers with different significance of the MSE term (controlled by $\\lambda$ ). "
+        ],
+        "image_footnote": [],
+        "bbox": [
+            212,
+            232,
+            797,
+            791
+        ],
+        "page_idx": 21
+    },
+    {
+        "type": "text",
+        "text": "I IMPACT OF THE DISCRIMINATOR HINT VECTOR ",
+        "text_level": 1,
+        "bbox": [
+            173,
+            102,
+            598,
+            118
+        ],
+        "page_idx": 22
+    },
+    {
+        "type": "text",
+        "text": "Yoon et al. (2018) suggested the idea of guiding the discriminator network using a hint mechanism. A hint vector reveals a subset of features that are missing to the discriminator. In Figure 9 and 10 we provide a comparison of learning curves for GI implemented using different hint rates. From Figure 9, using the hint mechanism does not result in any noticeable improvement in the final imputation quality justifying the added complexity. For the case of the rectangular missing structure in Figure 10; however, using the hint vector causes instabilities in the training process. One possible explanation is: providing even a small portion of the mask as a hint, due to the deterministic nature of the rectangular shape it is equivalent to providing region boundaries to the discriminator making it obvious for the discriminator. In GAN training we generally want to have equal competition between the generator and discriminator. ",
+        "bbox": [
+            173,
+            133,
+            826,
+            272
+        ],
+        "page_idx": 22
+    },
+    {
+        "type": "image",
+        "img_path": "images/d1773a0973e8e20e8a1e95585cf0570e2f70c7ae32a782150bd1926ef6c94fdd.jpg",
+        "image_caption": [
+            "Figure 9: Learning curves for CIFAR-10 with uniform missing structure at different discriminator hint rates. "
+        ],
+        "image_footnote": [],
+        "bbox": [
+            281,
+            291,
+            714,
+            508
+        ],
+        "page_idx": 22
+    },
+    {
+        "type": "image",
+        "img_path": "images/1d860d6685302d772bd786609a2406e9a1cc659ad5aaf397983d57ed6fba2242.jpg",
+        "image_caption": [
+            "Figure 10: Learning curves for CIFAR-10 with rectangular missing structure at different discriminator hint rates. "
+        ],
+        "image_footnote": [],
+        "bbox": [
+            281,
+            569,
+            714,
+            786
+        ],
+        "page_idx": 22
+    }
+]
\ No newline at end of file
diff --git a/parse/train/B1em9h4KDS/B1em9h4KDS_middle.json b/parse/train/B1em9h4KDS/B1em9h4KDS_middle.json
new file mode 100644
index 0000000000000000000000000000000000000000..03f660bf1462825b8f3452840c15e473624e9cc2
--- /dev/null
+++ b/parse/train/B1em9h4KDS/B1em9h4KDS_middle.json
@@ -0,0 +1,43052 @@
+{
+    "pdf_info": [
+        {
+            "preproc_blocks": [
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        78,
+                        532,
+                        97
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                78,
+                                536,
+                                99
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        78,
+                                        536,
+                                        99
+                                    ],
+                                    "score": 1.0,
+                                    "content": "GENERATIVE IMPUTATION AND STOCHASTIC PREDICTION",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        }
+                    ],
+                    "index": 0
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        112,
+                        115,
+                        245,
+                        137
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                113,
+                                116,
+                                201,
+                                127
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        113,
+                                        116,
+                                        201,
+                                        127
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Anonymous authors",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                111,
+                                126,
+                                245,
+                                138
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        111,
+                                        126,
+                                        245,
+                                        138
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Paper under double-blind review",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        }
+                    ],
+                    "index": 1.5
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        278,
+                        167,
+                        333,
+                        178
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                276,
+                                165,
+                                336,
+                                181
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        276,
+                                        165,
+                                        336,
+                                        181
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ABSTRACT",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        }
+                    ],
+                    "index": 3
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        143,
+                        192,
+                        469,
+                        369
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                141,
+                                192,
+                                469,
+                                205
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        192,
+                                        469,
+                                        205
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In many machine learning applications, we are faced with incomplete datasets. In",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                204,
+                                469,
+                                216
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        204,
+                                        469,
+                                        216
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the literature, missing data imputation techniques have been mostly concerned with",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                214,
+                                469,
+                                227
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        214,
+                                        469,
+                                        227
+                                    ],
+                                    "score": 1.0,
+                                    "content": "filling missing values. However, the existence of missing values is synonymous",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                225,
+                                470,
+                                237
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        225,
+                                        470,
+                                        237
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with uncertainties not only over the distribution of missing values but also over",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                236,
+                                470,
+                                249
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        236,
+                                        470,
+                                        249
+                                    ],
+                                    "score": 1.0,
+                                    "content": "target class assignments that require careful consideration. In this paper, we pro-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                248,
+                                470,
+                                260
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        248,
+                                        470,
+                                        260
+                                    ],
+                                    "score": 1.0,
+                                    "content": "pose a simple and effective method for imputing missing features and estimating",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                258,
+                                470,
+                                271
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        258,
+                                        470,
+                                        271
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the distribution of target assignments given incomplete data. In order to make",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                270,
+                                470,
+                                282
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        270,
+                                        470,
+                                        282
+                                    ],
+                                    "score": 1.0,
+                                    "content": "imputations, we train a simple and effective generator network to generate im-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                280,
+                                470,
+                                293
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        280,
+                                        470,
+                                        293
+                                    ],
+                                    "score": 1.0,
+                                    "content": "putations that a discriminator network is tasked to distinguish. Following this, a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                291,
+                                471,
+                                304
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        291,
+                                        471,
+                                        304
+                                    ],
+                                    "score": 1.0,
+                                    "content": "predictor network is trained using the imputed samples from the generator net-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                302,
+                                470,
+                                315
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        302,
+                                        470,
+                                        315
+                                    ],
+                                    "score": 1.0,
+                                    "content": "work to capture the classification uncertainties and make predictions accordingly.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                313,
+                                470,
+                                325
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        313,
+                                        470,
+                                        325
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The proposed method is evaluated on CIFAR-10 image dataset as well as three",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                324,
+                                469,
+                                336
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        324,
+                                        469,
+                                        336
+                                    ],
+                                    "score": 1.0,
+                                    "content": "real-world tabular classification datasets, under different missingness rates and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                142,
+                                335,
+                                469,
+                                347
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        142,
+                                        335,
+                                        469,
+                                        347
+                                    ],
+                                    "score": 1.0,
+                                    "content": "structures. Our experimental results show the effectiveness of the proposed method",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                347,
+                                469,
+                                358
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        347,
+                                        469,
+                                        358
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in generating imputations as well as providing estimates for the class uncertainties",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                356,
+                                365,
+                                370
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        356,
+                                        365,
+                                        370
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in a classification task when faced with missing values.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        }
+                    ],
+                    "index": 11.5
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        392,
+                        206,
+                        404
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                390,
+                                208,
+                                407
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        390,
+                                        208,
+                                        407
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1 INTRODUCTION",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        }
+                    ],
+                    "index": 20
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        417,
+                        505,
+                        484
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                417,
+                                506,
+                                430
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        417,
+                                        506,
+                                        430
+                                    ],
+                                    "score": 1.0,
+                                    "content": "While a large body of the machine learning literature is built upon the assumption of having access to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                429,
+                                504,
+                                441
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        429,
+                                        504,
+                                        441
+                                    ],
+                                    "score": 1.0,
+                                    "content": "complete datasets, in many real-world problems only incomplete datasets are available. The existence",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                439,
+                                505,
+                                452
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        439,
+                                        505,
+                                        452
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of missing values can be due to many different causes such as human subjects not adhering to certain",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                450,
+                                506,
+                                463
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        450,
+                                        506,
+                                        463
+                                    ],
+                                    "score": 1.0,
+                                    "content": "questions or features not being collected frequently due to financial or experimental limitations,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                461,
+                                505,
+                                474
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        461,
+                                        505,
+                                        474
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sensors failures, and so forth. Data imputation techniques have been suggested as a solution to bridge",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                473,
+                                405,
+                                485
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        473,
+                                        405,
+                                        485
+                                    ],
+                                    "score": 1.0,
+                                    "content": "this gap in the literature by replacing missing values with observed values.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        }
+                    ],
+                    "index": 23.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        489,
+                        505,
+                        566
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                489,
+                                506,
+                                502
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        489,
+                                        506,
+                                        502
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Missing data imputation approaches can be categorized into single and multiple imputation methods.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                501,
+                                506,
+                                513
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        501,
+                                        506,
+                                        513
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Single imputation methods try to replace each missing value with a plausible value that is the best fit",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                511,
+                                506,
+                                524
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        511,
+                                        506,
+                                        524
+                                    ],
+                                    "score": 1.0,
+                                    "content": "given the value of other correlated features and knowledge extracted from the dataset (Hastie et al.,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                522,
+                                506,
+                                535
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        522,
+                                        506,
+                                        535
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1999; Anderson, 1957). While these methods are easy to implement and use in practice, imputed",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                532,
+                                505,
+                                546
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        532,
+                                        505,
+                                        546
+                                    ],
+                                    "score": 1.0,
+                                    "content": "values may induce bias by eliminating less likely but important values. Also, these methods do",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                544,
+                                506,
+                                558
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        544,
+                                        506,
+                                        558
+                                    ],
+                                    "score": 1.0,
+                                    "content": "not suggest a way to measure to what extent the imputed values are representative of the missing",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                556,
+                                229,
+                                567
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        556,
+                                        229,
+                                        567
+                                    ],
+                                    "score": 1.0,
+                                    "content": "values (Little & Rubin, 2019).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        }
+                    ],
+                    "index": 30
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        572,
+                        505,
+                        660
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                571,
+                                506,
+                                585
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        571,
+                                        506,
+                                        585
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Multiple imputation (MI) techniques, as suggested by the name, try to use multiple imputed values",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                583,
+                                506,
+                                595
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        583,
+                                        506,
+                                        595
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to impute each missing value. The result would be having a set of imputed datasets that enables",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                595,
+                                505,
+                                606
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        595,
+                                        505,
+                                        606
+                                    ],
+                                    "score": 1.0,
+                                    "content": "measuring how consistent and statistically significant are the results of the experiments (Rubin, 1976).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                604,
+                                507,
+                                618
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        604,
+                                        507,
+                                        618
+                                    ],
+                                    "score": 1.0,
+                                    "content": "While MI offers interesting statistical insights about the reliability of analysis on incomplete data,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                617,
+                                505,
+                                628
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        617,
+                                        505,
+                                        628
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the insight is imprecise as it is mainly concerned about the population of data samples rather than",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                627,
+                                505,
+                                639
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        627,
+                                        505,
+                                        639
+                                    ],
+                                    "score": 1.0,
+                                    "content": "individual instances. Specifically, MI methods reason about the statistical properties on a limited",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                637,
+                                506,
+                                650
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        637,
+                                        506,
+                                        650
+                                    ],
+                                    "score": 1.0,
+                                    "content": "number of imputed datasets (less than 10 in most practical implementations) on the population of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                649,
+                                407,
+                                661
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        649,
+                                        407,
+                                        661
+                                    ],
+                                    "score": 1.0,
+                                    "content": "samples within the dataset (Schafer & Graham, 2002; Murray et al., 2018).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        }
+                    ],
+                    "index": 37.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        666,
+                        505,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                666,
+                                505,
+                                678
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        666,
+                                        505,
+                                        678
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The existence of missing values is synonymous with having uncertainty over these values that requires",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                677,
+                                506,
+                                690
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        677,
+                                        506,
+                                        690
+                                    ],
+                                    "score": 1.0,
+                                    "content": "careful consideration. In many real-world applications, we are dealing with supervised problems that",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                687,
+                                505,
+                                700
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        687,
+                                        505,
+                                        700
+                                    ],
+                                    "score": 1.0,
+                                    "content": "demand modeling and prediction based on incomplete data. Take for instance, prediction of class",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                699,
+                                506,
+                                712
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        699,
+                                        506,
+                                        712
+                                    ],
+                                    "score": 1.0,
+                                    "content": "assignments given an image in which a large portion of the frame is missing. In such a scenario,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                709,
+                                506,
+                                723
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        709,
+                                        506,
+                                        723
+                                    ],
+                                    "score": 1.0,
+                                    "content": "based on observed frame parts, there might be multiple probable class assignments each having a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                719,
+                                505,
+                                734
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        719,
+                                        505,
+                                        734
+                                    ],
+                                    "score": 1.0,
+                                    "content": "different likelihood. Here, we are not only interested in imputing missing values or measuring how",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 47
+                        }
+                    ],
+                    "index": 44.5
+                }
+            ],
+            "page_idx": 0,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        107,
+                        27,
+                        308,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                107,
+                                26,
+                                308,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        26,
+                                        308,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Under review as a conference paper at ICLR 2020",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        302,
+                        751,
+                        308,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                302,
+                                751,
+                                308,
+                                761
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        302,
+                                        751,
+                                        308,
+                                        761
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        78,
+                        532,
+                        97
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                78,
+                                536,
+                                99
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        78,
+                                        536,
+                                        99
+                                    ],
+                                    "score": 1.0,
+                                    "content": "GENERATIVE IMPUTATION AND STOCHASTIC PREDICTION",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        }
+                    ],
+                    "index": 0
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        112,
+                        115,
+                        245,
+                        137
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                113,
+                                116,
+                                201,
+                                127
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        113,
+                                        116,
+                                        201,
+                                        127
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Anonymous authors",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                111,
+                                126,
+                                245,
+                                138
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        111,
+                                        126,
+                                        245,
+                                        138
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Paper under double-blind review",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        }
+                    ],
+                    "index": 1.5,
+                    "bbox_fs": [
+                        111,
+                        116,
+                        245,
+                        138
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        278,
+                        167,
+                        333,
+                        178
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                276,
+                                165,
+                                336,
+                                181
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        276,
+                                        165,
+                                        336,
+                                        181
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ABSTRACT",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        }
+                    ],
+                    "index": 3
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        143,
+                        192,
+                        469,
+                        369
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                141,
+                                192,
+                                469,
+                                205
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        192,
+                                        469,
+                                        205
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In many machine learning applications, we are faced with incomplete datasets. In",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                204,
+                                469,
+                                216
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        204,
+                                        469,
+                                        216
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the literature, missing data imputation techniques have been mostly concerned with",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                214,
+                                469,
+                                227
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        214,
+                                        469,
+                                        227
+                                    ],
+                                    "score": 1.0,
+                                    "content": "filling missing values. However, the existence of missing values is synonymous",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                225,
+                                470,
+                                237
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        225,
+                                        470,
+                                        237
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with uncertainties not only over the distribution of missing values but also over",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                236,
+                                470,
+                                249
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        236,
+                                        470,
+                                        249
+                                    ],
+                                    "score": 1.0,
+                                    "content": "target class assignments that require careful consideration. In this paper, we pro-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                248,
+                                470,
+                                260
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        248,
+                                        470,
+                                        260
+                                    ],
+                                    "score": 1.0,
+                                    "content": "pose a simple and effective method for imputing missing features and estimating",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                258,
+                                470,
+                                271
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        258,
+                                        470,
+                                        271
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the distribution of target assignments given incomplete data. In order to make",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                270,
+                                470,
+                                282
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        270,
+                                        470,
+                                        282
+                                    ],
+                                    "score": 1.0,
+                                    "content": "imputations, we train a simple and effective generator network to generate im-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                280,
+                                470,
+                                293
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        280,
+                                        470,
+                                        293
+                                    ],
+                                    "score": 1.0,
+                                    "content": "putations that a discriminator network is tasked to distinguish. Following this, a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                291,
+                                471,
+                                304
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        291,
+                                        471,
+                                        304
+                                    ],
+                                    "score": 1.0,
+                                    "content": "predictor network is trained using the imputed samples from the generator net-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                302,
+                                470,
+                                315
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        302,
+                                        470,
+                                        315
+                                    ],
+                                    "score": 1.0,
+                                    "content": "work to capture the classification uncertainties and make predictions accordingly.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                313,
+                                470,
+                                325
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        313,
+                                        470,
+                                        325
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The proposed method is evaluated on CIFAR-10 image dataset as well as three",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                324,
+                                469,
+                                336
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        324,
+                                        469,
+                                        336
+                                    ],
+                                    "score": 1.0,
+                                    "content": "real-world tabular classification datasets, under different missingness rates and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                142,
+                                335,
+                                469,
+                                347
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        142,
+                                        335,
+                                        469,
+                                        347
+                                    ],
+                                    "score": 1.0,
+                                    "content": "structures. Our experimental results show the effectiveness of the proposed method",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                347,
+                                469,
+                                358
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        347,
+                                        469,
+                                        358
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in generating imputations as well as providing estimates for the class uncertainties",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                356,
+                                365,
+                                370
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        356,
+                                        365,
+                                        370
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in a classification task when faced with missing values.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        }
+                    ],
+                    "index": 11.5,
+                    "bbox_fs": [
+                        141,
+                        192,
+                        471,
+                        370
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        392,
+                        206,
+                        404
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                390,
+                                208,
+                                407
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        390,
+                                        208,
+                                        407
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1 INTRODUCTION",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        }
+                    ],
+                    "index": 20
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        417,
+                        505,
+                        484
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                417,
+                                506,
+                                430
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        417,
+                                        506,
+                                        430
+                                    ],
+                                    "score": 1.0,
+                                    "content": "While a large body of the machine learning literature is built upon the assumption of having access to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                429,
+                                504,
+                                441
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        429,
+                                        504,
+                                        441
+                                    ],
+                                    "score": 1.0,
+                                    "content": "complete datasets, in many real-world problems only incomplete datasets are available. The existence",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                439,
+                                505,
+                                452
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        439,
+                                        505,
+                                        452
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of missing values can be due to many different causes such as human subjects not adhering to certain",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                450,
+                                506,
+                                463
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        450,
+                                        506,
+                                        463
+                                    ],
+                                    "score": 1.0,
+                                    "content": "questions or features not being collected frequently due to financial or experimental limitations,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                461,
+                                505,
+                                474
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        461,
+                                        505,
+                                        474
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sensors failures, and so forth. Data imputation techniques have been suggested as a solution to bridge",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                473,
+                                405,
+                                485
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        473,
+                                        405,
+                                        485
+                                    ],
+                                    "score": 1.0,
+                                    "content": "this gap in the literature by replacing missing values with observed values.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        }
+                    ],
+                    "index": 23.5,
+                    "bbox_fs": [
+                        105,
+                        417,
+                        506,
+                        485
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        489,
+                        505,
+                        566
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                489,
+                                506,
+                                502
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        489,
+                                        506,
+                                        502
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Missing data imputation approaches can be categorized into single and multiple imputation methods.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                501,
+                                506,
+                                513
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        501,
+                                        506,
+                                        513
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Single imputation methods try to replace each missing value with a plausible value that is the best fit",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                511,
+                                506,
+                                524
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        511,
+                                        506,
+                                        524
+                                    ],
+                                    "score": 1.0,
+                                    "content": "given the value of other correlated features and knowledge extracted from the dataset (Hastie et al.,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                522,
+                                506,
+                                535
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        522,
+                                        506,
+                                        535
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1999; Anderson, 1957). While these methods are easy to implement and use in practice, imputed",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                532,
+                                505,
+                                546
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        532,
+                                        505,
+                                        546
+                                    ],
+                                    "score": 1.0,
+                                    "content": "values may induce bias by eliminating less likely but important values. Also, these methods do",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                544,
+                                506,
+                                558
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        544,
+                                        506,
+                                        558
+                                    ],
+                                    "score": 1.0,
+                                    "content": "not suggest a way to measure to what extent the imputed values are representative of the missing",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                556,
+                                229,
+                                567
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        556,
+                                        229,
+                                        567
+                                    ],
+                                    "score": 1.0,
+                                    "content": "values (Little & Rubin, 2019).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        }
+                    ],
+                    "index": 30,
+                    "bbox_fs": [
+                        104,
+                        489,
+                        506,
+                        567
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        572,
+                        505,
+                        660
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                571,
+                                506,
+                                585
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        571,
+                                        506,
+                                        585
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Multiple imputation (MI) techniques, as suggested by the name, try to use multiple imputed values",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                583,
+                                506,
+                                595
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        583,
+                                        506,
+                                        595
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to impute each missing value. The result would be having a set of imputed datasets that enables",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                595,
+                                505,
+                                606
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        595,
+                                        505,
+                                        606
+                                    ],
+                                    "score": 1.0,
+                                    "content": "measuring how consistent and statistically significant are the results of the experiments (Rubin, 1976).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                604,
+                                507,
+                                618
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        604,
+                                        507,
+                                        618
+                                    ],
+                                    "score": 1.0,
+                                    "content": "While MI offers interesting statistical insights about the reliability of analysis on incomplete data,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                617,
+                                505,
+                                628
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        617,
+                                        505,
+                                        628
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the insight is imprecise as it is mainly concerned about the population of data samples rather than",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                627,
+                                505,
+                                639
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        627,
+                                        505,
+                                        639
+                                    ],
+                                    "score": 1.0,
+                                    "content": "individual instances. Specifically, MI methods reason about the statistical properties on a limited",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                637,
+                                506,
+                                650
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        637,
+                                        506,
+                                        650
+                                    ],
+                                    "score": 1.0,
+                                    "content": "number of imputed datasets (less than 10 in most practical implementations) on the population of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                649,
+                                407,
+                                661
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        649,
+                                        407,
+                                        661
+                                    ],
+                                    "score": 1.0,
+                                    "content": "samples within the dataset (Schafer & Graham, 2002; Murray et al., 2018).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        }
+                    ],
+                    "index": 37.5,
+                    "bbox_fs": [
+                        105,
+                        571,
+                        507,
+                        661
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        666,
+                        505,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                666,
+                                505,
+                                678
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        666,
+                                        505,
+                                        678
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The existence of missing values is synonymous with having uncertainty over these values that requires",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                677,
+                                506,
+                                690
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        677,
+                                        506,
+                                        690
+                                    ],
+                                    "score": 1.0,
+                                    "content": "careful consideration. In many real-world applications, we are dealing with supervised problems that",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                687,
+                                505,
+                                700
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        687,
+                                        505,
+                                        700
+                                    ],
+                                    "score": 1.0,
+                                    "content": "demand modeling and prediction based on incomplete data. Take for instance, prediction of class",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                699,
+                                506,
+                                712
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        699,
+                                        506,
+                                        712
+                                    ],
+                                    "score": 1.0,
+                                    "content": "assignments given an image in which a large portion of the frame is missing. In such a scenario,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                709,
+                                506,
+                                723
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        709,
+                                        506,
+                                        723
+                                    ],
+                                    "score": 1.0,
+                                    "content": "based on observed frame parts, there might be multiple probable class assignments each having a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                719,
+                                505,
+                                734
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        719,
+                                        505,
+                                        734
+                                    ],
+                                    "score": 1.0,
+                                    "content": "different likelihood. Here, we are not only interested in imputing missing values or measuring how",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 47
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                81,
+                                505,
+                                95
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        81,
+                                        505,
+                                        95
+                                    ],
+                                    "score": 1.0,
+                                    "content": "robust our imputations are, but also it is highly desirable to measure the impact of missing values on",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                94,
+                                273,
+                                106
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        94,
+                                        273,
+                                        106
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the prediction outcome for each instance.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 1
+                        }
+                    ],
+                    "index": 44.5,
+                    "bbox_fs": [
+                        105,
+                        666,
+                        506,
+                        734
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        82,
+                        503,
+                        105
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                81,
+                                505,
+                                95
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        81,
+                                        505,
+                                        95
+                                    ],
+                                    "score": 1.0,
+                                    "content": "robust our imputations are, but also it is highly desirable to measure the impact of missing values on",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                94,
+                                273,
+                                106
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        94,
+                                        273,
+                                        106
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the prediction outcome for each instance.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        }
+                    ],
+                    "index": 0.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        110,
+                        505,
+                        198
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                110,
+                                506,
+                                123
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        110,
+                                        506,
+                                        123
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In this paper, we propose the idea of Generative Imputation and Stochastic Prediction (GI) as a novel",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                121,
+                                506,
+                                133
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        121,
+                                        506,
+                                        133
+                                    ],
+                                    "score": 1.0,
+                                    "content": "approach to impute missing values and to measure class uncertainties arising from the distribution of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                132,
+                                505,
+                                144
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        132,
+                                        505,
+                                        144
+                                    ],
+                                    "score": 1.0,
+                                    "content": "missing values. The suggested approach is based on neural networks trained using an adversarial",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                143,
+                                506,
+                                156
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        143,
+                                        506,
+                                        156
+                                    ],
+                                    "score": 1.0,
+                                    "content": "objective function. Additionally, a predictor is trained on the generated samples from the imputer",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                155,
+                                505,
+                                167
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        155,
+                                        505,
+                                        167
+                                    ],
+                                    "score": 1.0,
+                                    "content": "network which is able to reflect the impact of uncertainties over missing values. This enables",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                165,
+                                504,
+                                177
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        165,
+                                        504,
+                                        177
+                                    ],
+                                    "score": 1.0,
+                                    "content": "measuring different prediction outcomes and certainties for each specific instance. We evaluate",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                177,
+                                505,
+                                189
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        177,
+                                        505,
+                                        189
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the effectiveness of the proposed method on different incomplete image and tabular datasets under",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                188,
+                                236,
+                                199
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        188,
+                                        236,
+                                        199
+                                    ],
+                                    "score": 1.0,
+                                    "content": "various missingness structures.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        }
+                    ],
+                    "index": 5.5
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        219,
+                        210,
+                        232
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                218,
+                                213,
+                                235
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        218,
+                                        213,
+                                        235
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2 RELATED WORK",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        }
+                    ],
+                    "index": 10
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        248,
+                        505,
+                        314
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                247,
+                                506,
+                                261
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        247,
+                                        506,
+                                        261
+                                    ],
+                                    "score": 1.0,
+                                    "content": "One of the simplest traditional methods for handling missing values includes imputing the occurrences",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                258,
+                                505,
+                                272
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        258,
+                                        505,
+                                        272
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of missing values with constant values such as zeros or using mean values. To enhance the accuracy of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                270,
+                                506,
+                                282
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        270,
+                                        255,
+                                        282
+                                    ],
+                                    "score": 1.0,
+                                    "content": "such imputations, alternatives such as",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        256,
+                                        271,
+                                        263,
+                                        280
+                                    ],
+                                    "score": 0.37,
+                                    "content": "\\mathbf { k }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        263,
+                                        270,
+                                        506,
+                                        282
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-nearest neighbors (KNN) (Hastie et al., 1999) and maximum",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                281,
+                                505,
+                                294
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        281,
+                                        505,
+                                        294
+                                    ],
+                                    "score": 1.0,
+                                    "content": "likelihood estimation (MLE) (Anderson, 1957) have been suggested to estimate values to be used",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                291,
+                                506,
+                                304
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        291,
+                                        506,
+                                        304
+                                    ],
+                                    "score": 1.0,
+                                    "content": "given an observed context. While these methods are easy to implement and analyze, they often fail to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                303,
+                                466,
+                                315
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        303,
+                                        466,
+                                        315
+                                    ],
+                                    "score": 1.0,
+                                    "content": "capture the complex feature dependencies as well as structures present in many problems.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        }
+                    ],
+                    "index": 13.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        320,
+                        505,
+                        451
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                320,
+                                505,
+                                332
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        320,
+                                        505,
+                                        332
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Rubin (1976) suggested a categorization for missingness mechanisms: missing completely at random",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                330,
+                                504,
+                                343
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        330,
+                                        504,
+                                        343
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(MCAR), missing at random (MAR), and missing not at random (MNAR). Under the assumption",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                341,
+                                506,
+                                355
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        341,
+                                        506,
+                                        355
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of MAR, the authors suggested multiple imputation (MI) as a stochastic imputation method. Here,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                352,
+                                505,
+                                365
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        352,
+                                        505,
+                                        365
+                                    ],
+                                    "score": 1.0,
+                                    "content": "instead of imputing missing values using a single value, several values are sampled to represent the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                363,
+                                506,
+                                377
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        363,
+                                        506,
+                                        377
+                                    ],
+                                    "score": 1.0,
+                                    "content": "distribution over the missing value. MI generates a few imputed complete datasets that are then used",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                375,
+                                505,
+                                387
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        375,
+                                        505,
+                                        387
+                                    ],
+                                    "score": 1.0,
+                                    "content": "independently in statistical modeling (Schafer & Graham, 2002; Little & Rubin, 2019). Usually, the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                385,
+                                506,
+                                398
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        385,
+                                        506,
+                                        398
+                                    ],
+                                    "score": 1.0,
+                                    "content": "final goal of MI is to measure the robustness of the final statistical analysis amongst the imputed",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                396,
+                                506,
+                                408
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        396,
+                                        506,
+                                        408
+                                    ],
+                                    "score": 1.0,
+                                    "content": "datasets. In other words, it measures the quality of imputations and the statistical significance of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                408,
+                                506,
+                                420
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        408,
+                                        506,
+                                        420
+                                    ],
+                                    "score": 1.0,
+                                    "content": "analysis on the imputed data. It should be noted that the number of imputations used in MI is",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                418,
+                                505,
+                                430
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        418,
+                                        505,
+                                        430
+                                    ],
+                                    "score": 1.0,
+                                    "content": "usually very limited. Also, often strong simplifying assumptions are made in modeling the data",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                428,
+                                506,
+                                443
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        428,
+                                        506,
+                                        443
+                                    ],
+                                    "score": 1.0,
+                                    "content": "distribution (e.g., multi-variate Gaussian or Student’s t distribution) which limit the applicability of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                440,
+                                349,
+                                452
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        440,
+                                        349,
+                                        452
+                                    ],
+                                    "score": 1.0,
+                                    "content": "this method (Schafer & Graham, 2002; Murray et al., 2018).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        }
+                    ],
+                    "index": 22.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        457,
+                        505,
+                        589
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                457,
+                                505,
+                                470
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        457,
+                                        505,
+                                        470
+                                    ],
+                                    "score": 1.0,
+                                    "content": "More recently, autoencoder architectures have been suggested as powerful density estimators capable",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                468,
+                                505,
+                                480
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        468,
+                                        505,
+                                        480
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of capturing complex distributions. Perhaps, denoising autoencoders (DAE) (Vincent et al., 2008) are",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                479,
+                                505,
+                                492
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        479,
+                                        505,
+                                        492
+                                    ],
+                                    "score": 1.0,
+                                    "content": "one of the most intuitive approaches in which a neural network is trained to reconstruct and denoise",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                489,
+                                506,
+                                503
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        489,
+                                        506,
+                                        503
+                                    ],
+                                    "score": 1.0,
+                                    "content": "its input. Following a more probabilistic perspective, variational autoencoders (VAE) (Kingma &",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                500,
+                                507,
+                                514
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        500,
+                                        507,
+                                        514
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Welling, 2013) try to learn the data generating distribution via a latent representation. Specifically,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                511,
+                                506,
+                                525
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        511,
+                                        506,
+                                        525
+                                    ],
+                                    "score": 1.0,
+                                    "content": "conditional variational autoencoders (CVAE) (Sohn et al., 2015) can be used to sample missing values",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                524,
+                                505,
+                                536
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        524,
+                                        505,
+                                        536
+                                    ],
+                                    "score": 1.0,
+                                    "content": "conditioned on observed values. For instance, Mattei & Frellsen (2018) suggested a method based on",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                534,
+                                506,
+                                546
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        534,
+                                        506,
+                                        546
+                                    ],
+                                    "score": 1.0,
+                                    "content": "deep latent variable models and importance sampling that offers a tighter likelihood bound compared",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                545,
+                                505,
+                                557
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        545,
+                                        505,
+                                        557
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to the standard VAE bound. While these methods are powerful generative models applicable to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                556,
+                                506,
+                                568
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        556,
+                                        506,
+                                        568
+                                    ],
+                                    "score": 1.0,
+                                    "content": "missing data imputation, often samples generated using autoencoders are biased toward the mode of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                566,
+                                506,
+                                580
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        566,
+                                        506,
+                                        580
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the distribution (e.g., resulting in blurry images, for vision tasks) (Goodfellow et al., 2014; Dumoulin",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                577,
+                                159,
+                                590
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        577,
+                                        159,
+                                        590
+                                    ],
+                                    "score": 1.0,
+                                    "content": "et al., 2016).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        }
+                    ],
+                    "index": 34.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        594,
+                        505,
+                        694
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                594,
+                                506,
+                                608
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        594,
+                                        506,
+                                        608
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Recently, due to the success of generative adversarial networks (GAN), there has been great attention",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                605,
+                                505,
+                                619
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        605,
+                                        505,
+                                        619
+                                    ],
+                                    "score": 1.0,
+                                    "content": "toward applying them to impute missing values. For instance, Yoon et al. (2018) suggested an",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                616,
+                                506,
+                                630
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        616,
+                                        506,
+                                        630
+                                    ],
+                                    "score": 1.0,
+                                    "content": "imputation method based on adversarial and reconstruction loss terms. Li et al. (2019) introduced the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                627,
+                                506,
+                                641
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        627,
+                                        506,
+                                        641
+                                    ],
+                                    "score": 1.0,
+                                    "content": "idea of using separate generator and discriminator networks to learn the missing data structure and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                638,
+                                506,
+                                651
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        638,
+                                        506,
+                                        651
+                                    ],
+                                    "score": 1.0,
+                                    "content": "data distribution. These methods have been quite successful and are able to present the state-of-the-art",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                649,
+                                505,
+                                662
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        649,
+                                        505,
+                                        662
+                                    ],
+                                    "score": 1.0,
+                                    "content": "results. Though it should be noted that often the presence of additional loss terms may bias the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                660,
+                                506,
+                                673
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        660,
+                                        506,
+                                        673
+                                    ],
+                                    "score": 1.0,
+                                    "content": "generated samples toward the mode of the distribution being modeled. Also, these methods are",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 47
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                671,
+                                506,
+                                684
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        671,
+                                        506,
+                                        684
+                                    ],
+                                    "score": 1.0,
+                                    "content": "often complicated to be applied in practical setups by practitioners. For instance, Yoon et al. (2018)",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 48
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                682,
+                                506,
+                                694
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        682,
+                                        506,
+                                        694
+                                    ],
+                                    "score": 1.0,
+                                    "content": "requires setting hyperparameters to adjust the influence of an MSE loss term as well as the rate of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 49
+                        }
+                    ],
+                    "index": 45
+                }
+            ],
+            "page_idx": 1,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        108,
+                        711,
+                        505,
+                        731
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                118,
+                                709,
+                                505,
+                                724
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        118,
+                                        709,
+                                        505,
+                                        724
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1We plan to include a link to the source code and GitHub page related to this paper in the camera-ready",
+                                    "type": "text"
+                                }
+                            ]
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                720,
+                                138,
+                                733
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        720,
+                                        138,
+                                        733
+                                    ],
+                                    "score": 1.0,
+                                    "content": "version.",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        107,
+                        27,
+                        308,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                25,
+                                309,
+                                39
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        25,
+                                        309,
+                                        39
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Under review as a conference paper at ICLR 2020",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        302,
+                        751,
+                        309,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                301,
+                                750,
+                                310,
+                                763
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        301,
+                                        750,
+                                        310,
+                                        763
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        82,
+                        503,
+                        105
+                    ],
+                    "lines": [],
+                    "index": 0.5,
+                    "bbox_fs": [
+                        105,
+                        81,
+                        505,
+                        106
+                    ],
+                    "lines_deleted": true
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        110,
+                        505,
+                        198
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                110,
+                                506,
+                                123
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        110,
+                                        506,
+                                        123
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In this paper, we propose the idea of Generative Imputation and Stochastic Prediction (GI) as a novel",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                121,
+                                506,
+                                133
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        121,
+                                        506,
+                                        133
+                                    ],
+                                    "score": 1.0,
+                                    "content": "approach to impute missing values and to measure class uncertainties arising from the distribution of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                132,
+                                505,
+                                144
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        132,
+                                        505,
+                                        144
+                                    ],
+                                    "score": 1.0,
+                                    "content": "missing values. The suggested approach is based on neural networks trained using an adversarial",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                143,
+                                506,
+                                156
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        143,
+                                        506,
+                                        156
+                                    ],
+                                    "score": 1.0,
+                                    "content": "objective function. Additionally, a predictor is trained on the generated samples from the imputer",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                155,
+                                505,
+                                167
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        155,
+                                        505,
+                                        167
+                                    ],
+                                    "score": 1.0,
+                                    "content": "network which is able to reflect the impact of uncertainties over missing values. This enables",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                165,
+                                504,
+                                177
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        165,
+                                        504,
+                                        177
+                                    ],
+                                    "score": 1.0,
+                                    "content": "measuring different prediction outcomes and certainties for each specific instance. We evaluate",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                177,
+                                505,
+                                189
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        177,
+                                        505,
+                                        189
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the effectiveness of the proposed method on different incomplete image and tabular datasets under",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                188,
+                                236,
+                                199
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        188,
+                                        236,
+                                        199
+                                    ],
+                                    "score": 1.0,
+                                    "content": "various missingness structures.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        }
+                    ],
+                    "index": 5.5,
+                    "bbox_fs": [
+                        105,
+                        110,
+                        506,
+                        199
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        219,
+                        210,
+                        232
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                218,
+                                213,
+                                235
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        218,
+                                        213,
+                                        235
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2 RELATED WORK",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        }
+                    ],
+                    "index": 10
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        248,
+                        505,
+                        314
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                247,
+                                506,
+                                261
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        247,
+                                        506,
+                                        261
+                                    ],
+                                    "score": 1.0,
+                                    "content": "One of the simplest traditional methods for handling missing values includes imputing the occurrences",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                258,
+                                505,
+                                272
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        258,
+                                        505,
+                                        272
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of missing values with constant values such as zeros or using mean values. To enhance the accuracy of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                270,
+                                506,
+                                282
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        270,
+                                        255,
+                                        282
+                                    ],
+                                    "score": 1.0,
+                                    "content": "such imputations, alternatives such as",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        256,
+                                        271,
+                                        263,
+                                        280
+                                    ],
+                                    "score": 0.37,
+                                    "content": "\\mathbf { k }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        263,
+                                        270,
+                                        506,
+                                        282
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-nearest neighbors (KNN) (Hastie et al., 1999) and maximum",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                281,
+                                505,
+                                294
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        281,
+                                        505,
+                                        294
+                                    ],
+                                    "score": 1.0,
+                                    "content": "likelihood estimation (MLE) (Anderson, 1957) have been suggested to estimate values to be used",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                291,
+                                506,
+                                304
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        291,
+                                        506,
+                                        304
+                                    ],
+                                    "score": 1.0,
+                                    "content": "given an observed context. While these methods are easy to implement and analyze, they often fail to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                303,
+                                466,
+                                315
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        303,
+                                        466,
+                                        315
+                                    ],
+                                    "score": 1.0,
+                                    "content": "capture the complex feature dependencies as well as structures present in many problems.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        }
+                    ],
+                    "index": 13.5,
+                    "bbox_fs": [
+                        105,
+                        247,
+                        506,
+                        315
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        320,
+                        505,
+                        451
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                320,
+                                505,
+                                332
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        320,
+                                        505,
+                                        332
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Rubin (1976) suggested a categorization for missingness mechanisms: missing completely at random",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                330,
+                                504,
+                                343
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        330,
+                                        504,
+                                        343
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(MCAR), missing at random (MAR), and missing not at random (MNAR). Under the assumption",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                341,
+                                506,
+                                355
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        341,
+                                        506,
+                                        355
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of MAR, the authors suggested multiple imputation (MI) as a stochastic imputation method. Here,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                352,
+                                505,
+                                365
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        352,
+                                        505,
+                                        365
+                                    ],
+                                    "score": 1.0,
+                                    "content": "instead of imputing missing values using a single value, several values are sampled to represent the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                363,
+                                506,
+                                377
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        363,
+                                        506,
+                                        377
+                                    ],
+                                    "score": 1.0,
+                                    "content": "distribution over the missing value. MI generates a few imputed complete datasets that are then used",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                375,
+                                505,
+                                387
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        375,
+                                        505,
+                                        387
+                                    ],
+                                    "score": 1.0,
+                                    "content": "independently in statistical modeling (Schafer & Graham, 2002; Little & Rubin, 2019). Usually, the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                385,
+                                506,
+                                398
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        385,
+                                        506,
+                                        398
+                                    ],
+                                    "score": 1.0,
+                                    "content": "final goal of MI is to measure the robustness of the final statistical analysis amongst the imputed",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                396,
+                                506,
+                                408
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        396,
+                                        506,
+                                        408
+                                    ],
+                                    "score": 1.0,
+                                    "content": "datasets. In other words, it measures the quality of imputations and the statistical significance of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                408,
+                                506,
+                                420
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        408,
+                                        506,
+                                        420
+                                    ],
+                                    "score": 1.0,
+                                    "content": "analysis on the imputed data. It should be noted that the number of imputations used in MI is",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                418,
+                                505,
+                                430
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        418,
+                                        505,
+                                        430
+                                    ],
+                                    "score": 1.0,
+                                    "content": "usually very limited. Also, often strong simplifying assumptions are made in modeling the data",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                428,
+                                506,
+                                443
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        428,
+                                        506,
+                                        443
+                                    ],
+                                    "score": 1.0,
+                                    "content": "distribution (e.g., multi-variate Gaussian or Student’s t distribution) which limit the applicability of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                440,
+                                349,
+                                452
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        440,
+                                        349,
+                                        452
+                                    ],
+                                    "score": 1.0,
+                                    "content": "this method (Schafer & Graham, 2002; Murray et al., 2018).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        }
+                    ],
+                    "index": 22.5,
+                    "bbox_fs": [
+                        105,
+                        320,
+                        506,
+                        452
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        457,
+                        505,
+                        589
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                457,
+                                505,
+                                470
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        457,
+                                        505,
+                                        470
+                                    ],
+                                    "score": 1.0,
+                                    "content": "More recently, autoencoder architectures have been suggested as powerful density estimators capable",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                468,
+                                505,
+                                480
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        468,
+                                        505,
+                                        480
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of capturing complex distributions. Perhaps, denoising autoencoders (DAE) (Vincent et al., 2008) are",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                479,
+                                505,
+                                492
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        479,
+                                        505,
+                                        492
+                                    ],
+                                    "score": 1.0,
+                                    "content": "one of the most intuitive approaches in which a neural network is trained to reconstruct and denoise",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                489,
+                                506,
+                                503
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        489,
+                                        506,
+                                        503
+                                    ],
+                                    "score": 1.0,
+                                    "content": "its input. Following a more probabilistic perspective, variational autoencoders (VAE) (Kingma &",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                500,
+                                507,
+                                514
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        500,
+                                        507,
+                                        514
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Welling, 2013) try to learn the data generating distribution via a latent representation. Specifically,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                511,
+                                506,
+                                525
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        511,
+                                        506,
+                                        525
+                                    ],
+                                    "score": 1.0,
+                                    "content": "conditional variational autoencoders (CVAE) (Sohn et al., 2015) can be used to sample missing values",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                524,
+                                505,
+                                536
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        524,
+                                        505,
+                                        536
+                                    ],
+                                    "score": 1.0,
+                                    "content": "conditioned on observed values. For instance, Mattei & Frellsen (2018) suggested a method based on",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                534,
+                                506,
+                                546
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        534,
+                                        506,
+                                        546
+                                    ],
+                                    "score": 1.0,
+                                    "content": "deep latent variable models and importance sampling that offers a tighter likelihood bound compared",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                545,
+                                505,
+                                557
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        545,
+                                        505,
+                                        557
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to the standard VAE bound. While these methods are powerful generative models applicable to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                556,
+                                506,
+                                568
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        556,
+                                        506,
+                                        568
+                                    ],
+                                    "score": 1.0,
+                                    "content": "missing data imputation, often samples generated using autoencoders are biased toward the mode of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                566,
+                                506,
+                                580
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        566,
+                                        506,
+                                        580
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the distribution (e.g., resulting in blurry images, for vision tasks) (Goodfellow et al., 2014; Dumoulin",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                577,
+                                159,
+                                590
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        577,
+                                        159,
+                                        590
+                                    ],
+                                    "score": 1.0,
+                                    "content": "et al., 2016).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        }
+                    ],
+                    "index": 34.5,
+                    "bbox_fs": [
+                        105,
+                        457,
+                        507,
+                        590
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        594,
+                        505,
+                        694
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                594,
+                                506,
+                                608
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        594,
+                                        506,
+                                        608
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Recently, due to the success of generative adversarial networks (GAN), there has been great attention",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                605,
+                                505,
+                                619
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        605,
+                                        505,
+                                        619
+                                    ],
+                                    "score": 1.0,
+                                    "content": "toward applying them to impute missing values. For instance, Yoon et al. (2018) suggested an",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                616,
+                                506,
+                                630
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        616,
+                                        506,
+                                        630
+                                    ],
+                                    "score": 1.0,
+                                    "content": "imputation method based on adversarial and reconstruction loss terms. Li et al. (2019) introduced the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                627,
+                                506,
+                                641
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        627,
+                                        506,
+                                        641
+                                    ],
+                                    "score": 1.0,
+                                    "content": "idea of using separate generator and discriminator networks to learn the missing data structure and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                638,
+                                506,
+                                651
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        638,
+                                        506,
+                                        651
+                                    ],
+                                    "score": 1.0,
+                                    "content": "data distribution. These methods have been quite successful and are able to present the state-of-the-art",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                649,
+                                505,
+                                662
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        649,
+                                        505,
+                                        662
+                                    ],
+                                    "score": 1.0,
+                                    "content": "results. Though it should be noted that often the presence of additional loss terms may bias the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                660,
+                                506,
+                                673
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        660,
+                                        506,
+                                        673
+                                    ],
+                                    "score": 1.0,
+                                    "content": "generated samples toward the mode of the distribution being modeled. Also, these methods are",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 47
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                671,
+                                506,
+                                684
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        671,
+                                        506,
+                                        684
+                                    ],
+                                    "score": 1.0,
+                                    "content": "often complicated to be applied in practical setups by practitioners. For instance, Yoon et al. (2018)",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 48
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                682,
+                                506,
+                                694
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        682,
+                                        506,
+                                        694
+                                    ],
+                                    "score": 1.0,
+                                    "content": "requires setting hyperparameters to adjust the influence of an MSE loss term as well as the rate of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 49
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                205,
+                                505,
+                                220
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        205,
+                                        505,
+                                        220
+                                    ],
+                                    "score": 1.0,
+                                    "content": "discriminator hint vectors. Also as another example, Li et al. (2019) uses three generators and three",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                217,
+                                300,
+                                230
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        217,
+                                        300,
+                                        230
+                                    ],
+                                    "score": 1.0,
+                                    "content": "discriminators for the final imputer architecture.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 6
+                        }
+                    ],
+                    "index": 45,
+                    "bbox_fs": [
+                        105,
+                        594,
+                        506,
+                        694
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "image",
+                    "bbox": [
+                        129,
+                        82,
+                        484,
+                        147
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                129,
+                                82,
+                                484,
+                                147
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        129,
+                                        82,
+                                        484,
+                                        147
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                129,
+                                                82,
+                                                484,
+                                                147
+                                            ],
+                                            "score": 0.953,
+                                            "type": "image",
+                                            "image_path": "c5713bc074e138a2a20d050db7edf84f541cc471e67603e24e6900e610f80f08.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 1,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        129,
+                                        82,
+                                        484,
+                                        103.66666666666667
+                                    ],
+                                    "spans": [],
+                                    "index": 0
+                                },
+                                {
+                                    "bbox": [
+                                        129,
+                                        103.66666666666667,
+                                        484,
+                                        125.33333333333334
+                                    ],
+                                    "spans": [],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        129,
+                                        125.33333333333334,
+                                        484,
+                                        147.0
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                106,
+                                161,
+                                504,
+                                184
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        159,
+                                        506,
+                                        175
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                159,
+                                                401,
+                                                175
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 1: Block diagram of the proposed adversarial imputation method.",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                401,
+                                                162,
+                                                408,
+                                                172
+                                            ],
+                                            "score": 0.63,
+                                            "content": "h",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                409,
+                                                159,
+                                                506,
+                                                175
+                                            ],
+                                            "score": 1.0,
+                                            "content": "represents the blending",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 3
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        172,
+                                        345,
+                                        183
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                172,
+                                                186,
+                                                183
+                                            ],
+                                            "score": 1.0,
+                                            "content": "function of (1), and",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                187,
+                                                173,
+                                                195,
+                                                182
+                                            ],
+                                            "score": 0.81,
+                                            "content": "L",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                195,
+                                                172,
+                                                345,
+                                                183
+                                            ],
+                                            "score": 1.0,
+                                            "content": "is the adversarial loss function of (2).",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 4
+                                }
+                            ],
+                            "index": 3.5
+                        }
+                    ],
+                    "index": 2.25
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        206,
+                        502,
+                        228
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                205,
+                                505,
+                                220
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        205,
+                                        505,
+                                        220
+                                    ],
+                                    "score": 1.0,
+                                    "content": "discriminator hint vectors. Also as another example, Li et al. (2019) uses three generators and three",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                217,
+                                300,
+                                230
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        217,
+                                        300,
+                                        230
+                                    ],
+                                    "score": 1.0,
+                                    "content": "discriminators for the final imputer architecture.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        }
+                    ],
+                    "index": 5.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        234,
+                        505,
+                        388
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                233,
+                                505,
+                                247
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        233,
+                                        505,
+                                        247
+                                    ],
+                                    "score": 1.0,
+                                    "content": "From the perspective of supervised analysis, imputation and handling missing values are usually",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                245,
+                                506,
+                                258
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        245,
+                                        506,
+                                        258
+                                    ],
+                                    "score": 1.0,
+                                    "content": "considered as a preprocessing step. A few exceptions exist such as Bayesian models and decision",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                256,
+                                506,
+                                268
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        256,
+                                        506,
+                                        268
+                                    ],
+                                    "score": 1.0,
+                                    "content": "trees that permit direct analysis on incomplete data (Nielsen & Jensen, 2009; Zhang et al., 2005).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                266,
+                                506,
+                                280
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        266,
+                                        506,
+                                        280
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Note that while certain Bayesian methods such as probabilistic Bayesian networks allow handling of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                278,
+                                506,
+                                291
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        278,
+                                        506,
+                                        291
+                                    ],
+                                    "score": 1.0,
+                                    "content": "missing values as unobserved variables. However, given an incomplete training dataset and without",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                289,
+                                506,
+                                302
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        289,
+                                        506,
+                                        302
+                                    ],
+                                    "score": 1.0,
+                                    "content": "any known causal structure as a priori, learning such models is a very challenging problem with the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                300,
+                                506,
+                                312
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        300,
+                                        506,
+                                        312
+                                    ],
+                                    "score": 1.0,
+                                    "content": "complexity of at least NP-complete to learn the network architecture in addition to an iterative EM",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                311,
+                                506,
+                                324
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        311,
+                                        506,
+                                        324
+                                    ],
+                                    "score": 1.0,
+                                    "content": "optimization to learn model parameters (Darwiche, 2009; Neapolitan et al., 2004). We argue that",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                322,
+                                506,
+                                335
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        322,
+                                        506,
+                                        335
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the simplistic approach of imputing missing values as a preprocessing step discards uncertainties",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                333,
+                                504,
+                                344
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        333,
+                                        504,
+                                        344
+                                    ],
+                                    "score": 1.0,
+                                    "content": "that exist in original incomplete data samples. Instead, there is a need for methods that reflect these",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                344,
+                                506,
+                                357
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        344,
+                                        506,
+                                        357
+                                    ],
+                                    "score": 1.0,
+                                    "content": "uncertainties on the final predicted target distribution. This work suggests the idea of training a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                355,
+                                506,
+                                367
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        355,
+                                        506,
+                                        367
+                                    ],
+                                    "score": 1.0,
+                                    "content": "predictor on different imputed samples to capture the uncertainties over class assignments. Compared",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                365,
+                                506,
+                                378
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        365,
+                                        506,
+                                        378
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to MI, the suggested method interleaves imputation and training a downstream prediction model,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                377,
+                                369,
+                                388
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        377,
+                                        369,
+                                        388
+                                    ],
+                                    "score": 1.0,
+                                    "content": "enabling to estimate classification uncertainties for each instance.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        }
+                    ],
+                    "index": 13.5
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        404,
+                        230,
+                        417
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                403,
+                                232,
+                                420
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        403,
+                                        232,
+                                        420
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3 PROPOSED METHOD",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        }
+                    ],
+                    "index": 21
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        430,
+                        227,
+                        442
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                430,
+                                228,
+                                443
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        430,
+                                        228,
+                                        443
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3.1 PROBLEM DEFINITION",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        }
+                    ],
+                    "index": 22
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        451,
+                        506,
+                        531
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                450,
+                                506,
+                                464
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        450,
+                                        452,
+                                        464
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In this paper, we make the general assumption of having access to an incomplete dataset",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        453,
+                                        452,
+                                        462,
+                                        461
+                                    ],
+                                    "score": 0.81,
+                                    "content": "\\mathcal { D }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        462,
+                                        450,
+                                        506,
+                                        464
+                                    ],
+                                    "score": 1.0,
+                                    "content": "consisting",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                462,
+                                506,
+                                475
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        462,
+                                        356,
+                                        475
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of a set of feature vector, mask vector, and target class pairs",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        357,
+                                        462,
+                                        401,
+                                        474
+                                    ],
+                                    "score": 0.93,
+                                    "content": "( { \\pmb x } _ { i } , { \\pmb k } _ { i } , y _ { i } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        401,
+                                        462,
+                                        506,
+                                        475
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". For each feature vector,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                474,
+                                506,
+                                489
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        474,
+                                        141,
+                                        487
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\pmb { x } _ { i } \\in \\mathbb { R } ^ { d }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        142,
+                                        474,
+                                        378,
+                                        489
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", only a subset of the features is available. The mask vector",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        378,
+                                        474,
+                                        430,
+                                        488
+                                    ],
+                                    "score": 0.94,
+                                    "content": "\\pmb { k } _ { i } \\in \\{ 0 , 1 \\} ^ { d }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        430,
+                                        474,
+                                        506,
+                                        489
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is used to indicate",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                486,
+                                506,
+                                500
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        486,
+                                        506,
+                                        500
+                                    ],
+                                    "score": 1.0,
+                                    "content": "available features and missing features by ones and zeros, respectively. Here, to represent features as",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                497,
+                                506,
+                                510
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        497,
+                                        506,
+                                        510
+                                    ],
+                                    "score": 1.0,
+                                    "content": "fixed-width vectors, arbitrary (or NaN) values are used to fill missing values. Also, for convenience,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                101,
+                                504,
+                                509,
+                                526
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        101,
+                                        504,
+                                        158,
+                                        526
+                                    ],
+                                    "score": 1.0,
+                                    "content": "we often use",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        158,
+                                        508,
+                                        177,
+                                        520
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\pmb { x } _ { i } ^ { o b s }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        177,
+                                        504,
+                                        195,
+                                        526
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        195,
+                                        508,
+                                        220,
+                                        521
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\pmb { x } _ { i } ^ { m i s s }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        221,
+                                        504,
+                                        509,
+                                        526
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to refer to the set of observed and missing features for the feature vector",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                107,
+                                521,
+                                122,
+                                533
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        522,
+                                        117,
+                                        531
+                                    ],
+                                    "score": 0.82,
+                                    "content": "\\mathbf { \\Delta } _ { \\mathbf { \\mathcal { X } } _ { i } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        118,
+                                        521,
+                                        122,
+                                        533
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        }
+                    ],
+                    "index": 26
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        536,
+                        505,
+                        613
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                535,
+                                506,
+                                550
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        535,
+                                        255,
+                                        550
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We define our objective in two steps:",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        255,
+                                        536,
+                                        268,
+                                        548
+                                    ],
+                                    "score": 0.56,
+                                    "content": "( i )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        268,
+                                        535,
+                                        506,
+                                        550
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Imputing missing values via sampling from the conditional",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                103,
+                                543,
+                                507,
+                                563
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        103,
+                                        543,
+                                        355,
+                                        563
+                                    ],
+                                    "score": 1.0,
+                                    "content": "distribution of missing features given observed features i.e.,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        356,
+                                        547,
+                                        417,
+                                        559
+                                    ],
+                                    "score": 0.91,
+                                    "content": "P ( \\pmb { x } _ { i } ^ { m i s s } | \\pmb { x } _ { i } ^ { o b s } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        417,
+                                        543,
+                                        423,
+                                        563
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        424,
+                                        547,
+                                        441,
+                                        559
+                                    ],
+                                    "score": 0.64,
+                                    "content": "( i i )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        441,
+                                        543,
+                                        507,
+                                        563
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Estimating the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                557,
+                                506,
+                                572
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        557,
+                                        506,
+                                        572
+                                    ],
+                                    "score": 1.0,
+                                    "content": "distribution of target classes given the observed features and the distribution of missing features i.e.,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                107,
+                                564,
+                                509,
+                                587
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        569,
+                                        178,
+                                        581
+                                    ],
+                                    "score": 0.92,
+                                    "content": "P ( y | \\mathbf { x } _ { i } ^ { o b s } , \\mathbf { x } _ { i } ^ { m i s s } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        179,
+                                        564,
+                                        509,
+                                        587
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". For the first part, we are interested in sampling from the conditional distribution",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                579,
+                                505,
+                                593
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        579,
+                                        505,
+                                        593
+                                    ],
+                                    "score": 1.0,
+                                    "content": "rather than finding the mode of the distribution as the most probable imputation. Similarly, for the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                591,
+                                505,
+                                604
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        591,
+                                        505,
+                                        604
+                                    ],
+                                    "score": 1.0,
+                                    "content": "second part, we are interested in obtaining a distribution over the possible target assignments and the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                602,
+                                412,
+                                614
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        602,
+                                        412,
+                                        614
+                                    ],
+                                    "score": 1.0,
+                                    "content": "confidence of each class rather than maximum likelihood class assignments.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        }
+                    ],
+                    "index": 33
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        628,
+                        244,
+                        639
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                627,
+                                245,
+                                641
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        627,
+                                        245,
+                                        641
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3.2 GENERATIVE IMPUTATION",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        }
+                    ],
+                    "index": 37
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        649,
+                        505,
+                        693
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                648,
+                                507,
+                                662
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        648,
+                                        507,
+                                        662
+                                    ],
+                                    "score": 1.0,
+                                    "content": "To generate samples from the distribution of missing features conditioned on the observed features,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                659,
+                                506,
+                                673
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        659,
+                                        506,
+                                        673
+                                    ],
+                                    "score": 1.0,
+                                    "content": "we follow the idea first suggested by Yoon et al. (2018). In this paradigm, a generator network is",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                671,
+                                505,
+                                684
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        671,
+                                        505,
+                                        684
+                                    ],
+                                    "score": 1.0,
+                                    "content": "responsible for generating imputations while a discriminator is trying to distinguish imputed features",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                681,
+                                261,
+                                694
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        681,
+                                        261,
+                                        694
+                                    ],
+                                    "score": 1.0,
+                                    "content": "from observed features (see Figure 1).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        }
+                    ],
+                    "index": 39.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        698,
+                        504,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                697,
+                                506,
+                                712
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        697,
+                                        247,
+                                        712
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Specifically, the generator function",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        247,
+                                        698,
+                                        322,
+                                        711
+                                    ],
+                                    "score": 0.92,
+                                    "content": "G ( \\pmb { x } _ { i } , \\pmb { k } _ { i } , z ) \\in \\mathbb { R } ^ { d }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        322,
+                                        697,
+                                        506,
+                                        712
+                                    ],
+                                    "score": 1.0,
+                                    "content": "generates an imputed feature vector, based on",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                710,
+                                505,
+                                722
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        710,
+                                        399,
+                                        722
+                                    ],
+                                    "score": 1.0,
+                                    "content": "observed features, the corresponding mask, and a Gaussian noise vector",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        400,
+                                        710,
+                                        414,
+                                        721
+                                    ],
+                                    "score": 0.68,
+                                    "content": "( z )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        414,
+                                        710,
+                                        505,
+                                        722
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Note that, in order to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                720,
+                                505,
+                                734
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        720,
+                                        238,
+                                        734
+                                    ],
+                                    "score": 1.0,
+                                    "content": "achieve the final imputed vector,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        239,
+                                        721,
+                                        249,
+                                        732
+                                    ],
+                                    "score": 0.88,
+                                    "content": "\\widehat { \\pmb { x } } _ { i }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        250,
+                                        720,
+                                        505,
+                                        734
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", we blend (or, merge) the output of the generator with the input",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        }
+                    ],
+                    "index": 43
+                }
+            ],
+            "page_idx": 2,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        302,
+                        751,
+                        309,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                301,
+                                750,
+                                310,
+                                762
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        301,
+                                        750,
+                                        310,
+                                        762
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        106,
+                        27,
+                        308,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                26,
+                                308,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        26,
+                                        308,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Under review as a conference paper at ICLR 2020",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "image",
+                    "bbox": [
+                        129,
+                        82,
+                        484,
+                        147
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                129,
+                                82,
+                                484,
+                                147
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        129,
+                                        82,
+                                        484,
+                                        147
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                129,
+                                                82,
+                                                484,
+                                                147
+                                            ],
+                                            "score": 0.953,
+                                            "type": "image",
+                                            "image_path": "c5713bc074e138a2a20d050db7edf84f541cc471e67603e24e6900e610f80f08.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 1,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        129,
+                                        82,
+                                        484,
+                                        103.66666666666667
+                                    ],
+                                    "spans": [],
+                                    "index": 0
+                                },
+                                {
+                                    "bbox": [
+                                        129,
+                                        103.66666666666667,
+                                        484,
+                                        125.33333333333334
+                                    ],
+                                    "spans": [],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        129,
+                                        125.33333333333334,
+                                        484,
+                                        147.0
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                106,
+                                161,
+                                504,
+                                184
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        159,
+                                        506,
+                                        175
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                159,
+                                                401,
+                                                175
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 1: Block diagram of the proposed adversarial imputation method.",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                401,
+                                                162,
+                                                408,
+                                                172
+                                            ],
+                                            "score": 0.63,
+                                            "content": "h",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                409,
+                                                159,
+                                                506,
+                                                175
+                                            ],
+                                            "score": 1.0,
+                                            "content": "represents the blending",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 3
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        172,
+                                        345,
+                                        183
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                172,
+                                                186,
+                                                183
+                                            ],
+                                            "score": 1.0,
+                                            "content": "function of (1), and",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                187,
+                                                173,
+                                                195,
+                                                182
+                                            ],
+                                            "score": 0.81,
+                                            "content": "L",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                195,
+                                                172,
+                                                345,
+                                                183
+                                            ],
+                                            "score": 1.0,
+                                            "content": "is the adversarial loss function of (2).",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 4
+                                }
+                            ],
+                            "index": 3.5
+                        }
+                    ],
+                    "index": 2.25
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        206,
+                        502,
+                        228
+                    ],
+                    "lines": [],
+                    "index": 5.5,
+                    "bbox_fs": [
+                        106,
+                        205,
+                        505,
+                        230
+                    ],
+                    "lines_deleted": true
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        234,
+                        505,
+                        388
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                233,
+                                505,
+                                247
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        233,
+                                        505,
+                                        247
+                                    ],
+                                    "score": 1.0,
+                                    "content": "From the perspective of supervised analysis, imputation and handling missing values are usually",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                245,
+                                506,
+                                258
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        245,
+                                        506,
+                                        258
+                                    ],
+                                    "score": 1.0,
+                                    "content": "considered as a preprocessing step. A few exceptions exist such as Bayesian models and decision",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                256,
+                                506,
+                                268
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        256,
+                                        506,
+                                        268
+                                    ],
+                                    "score": 1.0,
+                                    "content": "trees that permit direct analysis on incomplete data (Nielsen & Jensen, 2009; Zhang et al., 2005).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                266,
+                                506,
+                                280
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        266,
+                                        506,
+                                        280
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Note that while certain Bayesian methods such as probabilistic Bayesian networks allow handling of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                278,
+                                506,
+                                291
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        278,
+                                        506,
+                                        291
+                                    ],
+                                    "score": 1.0,
+                                    "content": "missing values as unobserved variables. However, given an incomplete training dataset and without",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                289,
+                                506,
+                                302
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        289,
+                                        506,
+                                        302
+                                    ],
+                                    "score": 1.0,
+                                    "content": "any known causal structure as a priori, learning such models is a very challenging problem with the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                300,
+                                506,
+                                312
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        300,
+                                        506,
+                                        312
+                                    ],
+                                    "score": 1.0,
+                                    "content": "complexity of at least NP-complete to learn the network architecture in addition to an iterative EM",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                311,
+                                506,
+                                324
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        311,
+                                        506,
+                                        324
+                                    ],
+                                    "score": 1.0,
+                                    "content": "optimization to learn model parameters (Darwiche, 2009; Neapolitan et al., 2004). We argue that",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                322,
+                                506,
+                                335
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        322,
+                                        506,
+                                        335
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the simplistic approach of imputing missing values as a preprocessing step discards uncertainties",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                333,
+                                504,
+                                344
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        333,
+                                        504,
+                                        344
+                                    ],
+                                    "score": 1.0,
+                                    "content": "that exist in original incomplete data samples. Instead, there is a need for methods that reflect these",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                344,
+                                506,
+                                357
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        344,
+                                        506,
+                                        357
+                                    ],
+                                    "score": 1.0,
+                                    "content": "uncertainties on the final predicted target distribution. This work suggests the idea of training a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                355,
+                                506,
+                                367
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        355,
+                                        506,
+                                        367
+                                    ],
+                                    "score": 1.0,
+                                    "content": "predictor on different imputed samples to capture the uncertainties over class assignments. Compared",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                365,
+                                506,
+                                378
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        365,
+                                        506,
+                                        378
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to MI, the suggested method interleaves imputation and training a downstream prediction model,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                377,
+                                369,
+                                388
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        377,
+                                        369,
+                                        388
+                                    ],
+                                    "score": 1.0,
+                                    "content": "enabling to estimate classification uncertainties for each instance.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        }
+                    ],
+                    "index": 13.5,
+                    "bbox_fs": [
+                        105,
+                        233,
+                        506,
+                        388
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        404,
+                        230,
+                        417
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                403,
+                                232,
+                                420
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        403,
+                                        232,
+                                        420
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3 PROPOSED METHOD",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        }
+                    ],
+                    "index": 21
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        430,
+                        227,
+                        442
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                430,
+                                228,
+                                443
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        430,
+                                        228,
+                                        443
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3.1 PROBLEM DEFINITION",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        }
+                    ],
+                    "index": 22
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        451,
+                        506,
+                        531
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                450,
+                                506,
+                                464
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        450,
+                                        452,
+                                        464
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In this paper, we make the general assumption of having access to an incomplete dataset",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        453,
+                                        452,
+                                        462,
+                                        461
+                                    ],
+                                    "score": 0.81,
+                                    "content": "\\mathcal { D }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        462,
+                                        450,
+                                        506,
+                                        464
+                                    ],
+                                    "score": 1.0,
+                                    "content": "consisting",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                462,
+                                506,
+                                475
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        462,
+                                        356,
+                                        475
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of a set of feature vector, mask vector, and target class pairs",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        357,
+                                        462,
+                                        401,
+                                        474
+                                    ],
+                                    "score": 0.93,
+                                    "content": "( { \\pmb x } _ { i } , { \\pmb k } _ { i } , y _ { i } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        401,
+                                        462,
+                                        506,
+                                        475
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". For each feature vector,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                474,
+                                506,
+                                489
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        474,
+                                        141,
+                                        487
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\pmb { x } _ { i } \\in \\mathbb { R } ^ { d }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        142,
+                                        474,
+                                        378,
+                                        489
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", only a subset of the features is available. The mask vector",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        378,
+                                        474,
+                                        430,
+                                        488
+                                    ],
+                                    "score": 0.94,
+                                    "content": "\\pmb { k } _ { i } \\in \\{ 0 , 1 \\} ^ { d }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        430,
+                                        474,
+                                        506,
+                                        489
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is used to indicate",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                486,
+                                506,
+                                500
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        486,
+                                        506,
+                                        500
+                                    ],
+                                    "score": 1.0,
+                                    "content": "available features and missing features by ones and zeros, respectively. Here, to represent features as",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                497,
+                                506,
+                                510
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        497,
+                                        506,
+                                        510
+                                    ],
+                                    "score": 1.0,
+                                    "content": "fixed-width vectors, arbitrary (or NaN) values are used to fill missing values. Also, for convenience,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                101,
+                                504,
+                                509,
+                                526
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        101,
+                                        504,
+                                        158,
+                                        526
+                                    ],
+                                    "score": 1.0,
+                                    "content": "we often use",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        158,
+                                        508,
+                                        177,
+                                        520
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\pmb { x } _ { i } ^ { o b s }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        177,
+                                        504,
+                                        195,
+                                        526
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        195,
+                                        508,
+                                        220,
+                                        521
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\pmb { x } _ { i } ^ { m i s s }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        221,
+                                        504,
+                                        509,
+                                        526
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to refer to the set of observed and missing features for the feature vector",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                107,
+                                521,
+                                122,
+                                533
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        522,
+                                        117,
+                                        531
+                                    ],
+                                    "score": 0.82,
+                                    "content": "\\mathbf { \\Delta } _ { \\mathbf { \\mathcal { X } } _ { i } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        118,
+                                        521,
+                                        122,
+                                        533
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        }
+                    ],
+                    "index": 26,
+                    "bbox_fs": [
+                        101,
+                        450,
+                        509,
+                        533
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        536,
+                        505,
+                        613
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                535,
+                                506,
+                                550
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        535,
+                                        255,
+                                        550
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We define our objective in two steps:",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        255,
+                                        536,
+                                        268,
+                                        548
+                                    ],
+                                    "score": 0.56,
+                                    "content": "( i )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        268,
+                                        535,
+                                        506,
+                                        550
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Imputing missing values via sampling from the conditional",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                103,
+                                543,
+                                507,
+                                563
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        103,
+                                        543,
+                                        355,
+                                        563
+                                    ],
+                                    "score": 1.0,
+                                    "content": "distribution of missing features given observed features i.e.,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        356,
+                                        547,
+                                        417,
+                                        559
+                                    ],
+                                    "score": 0.91,
+                                    "content": "P ( \\pmb { x } _ { i } ^ { m i s s } | \\pmb { x } _ { i } ^ { o b s } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        417,
+                                        543,
+                                        423,
+                                        563
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        424,
+                                        547,
+                                        441,
+                                        559
+                                    ],
+                                    "score": 0.64,
+                                    "content": "( i i )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        441,
+                                        543,
+                                        507,
+                                        563
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Estimating the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                557,
+                                506,
+                                572
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        557,
+                                        506,
+                                        572
+                                    ],
+                                    "score": 1.0,
+                                    "content": "distribution of target classes given the observed features and the distribution of missing features i.e.,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                107,
+                                564,
+                                509,
+                                587
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        569,
+                                        178,
+                                        581
+                                    ],
+                                    "score": 0.92,
+                                    "content": "P ( y | \\mathbf { x } _ { i } ^ { o b s } , \\mathbf { x } _ { i } ^ { m i s s } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        179,
+                                        564,
+                                        509,
+                                        587
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". For the first part, we are interested in sampling from the conditional distribution",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                579,
+                                505,
+                                593
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        579,
+                                        505,
+                                        593
+                                    ],
+                                    "score": 1.0,
+                                    "content": "rather than finding the mode of the distribution as the most probable imputation. Similarly, for the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                591,
+                                505,
+                                604
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        591,
+                                        505,
+                                        604
+                                    ],
+                                    "score": 1.0,
+                                    "content": "second part, we are interested in obtaining a distribution over the possible target assignments and the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                602,
+                                412,
+                                614
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        602,
+                                        412,
+                                        614
+                                    ],
+                                    "score": 1.0,
+                                    "content": "confidence of each class rather than maximum likelihood class assignments.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        }
+                    ],
+                    "index": 33,
+                    "bbox_fs": [
+                        103,
+                        535,
+                        509,
+                        614
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        628,
+                        244,
+                        639
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                627,
+                                245,
+                                641
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        627,
+                                        245,
+                                        641
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3.2 GENERATIVE IMPUTATION",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        }
+                    ],
+                    "index": 37
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        649,
+                        505,
+                        693
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                648,
+                                507,
+                                662
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        648,
+                                        507,
+                                        662
+                                    ],
+                                    "score": 1.0,
+                                    "content": "To generate samples from the distribution of missing features conditioned on the observed features,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                659,
+                                506,
+                                673
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        659,
+                                        506,
+                                        673
+                                    ],
+                                    "score": 1.0,
+                                    "content": "we follow the idea first suggested by Yoon et al. (2018). In this paradigm, a generator network is",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                671,
+                                505,
+                                684
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        671,
+                                        505,
+                                        684
+                                    ],
+                                    "score": 1.0,
+                                    "content": "responsible for generating imputations while a discriminator is trying to distinguish imputed features",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                681,
+                                261,
+                                694
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        681,
+                                        261,
+                                        694
+                                    ],
+                                    "score": 1.0,
+                                    "content": "from observed features (see Figure 1).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        }
+                    ],
+                    "index": 39.5,
+                    "bbox_fs": [
+                        105,
+                        648,
+                        507,
+                        694
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        698,
+                        504,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                697,
+                                506,
+                                712
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        697,
+                                        247,
+                                        712
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Specifically, the generator function",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        247,
+                                        698,
+                                        322,
+                                        711
+                                    ],
+                                    "score": 0.92,
+                                    "content": "G ( \\pmb { x } _ { i } , \\pmb { k } _ { i } , z ) \\in \\mathbb { R } ^ { d }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        322,
+                                        697,
+                                        506,
+                                        712
+                                    ],
+                                    "score": 1.0,
+                                    "content": "generates an imputed feature vector, based on",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                710,
+                                505,
+                                722
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        710,
+                                        399,
+                                        722
+                                    ],
+                                    "score": 1.0,
+                                    "content": "observed features, the corresponding mask, and a Gaussian noise vector",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        400,
+                                        710,
+                                        414,
+                                        721
+                                    ],
+                                    "score": 0.68,
+                                    "content": "( z )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        414,
+                                        710,
+                                        505,
+                                        722
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Note that, in order to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                720,
+                                505,
+                                734
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        720,
+                                        238,
+                                        734
+                                    ],
+                                    "score": 1.0,
+                                    "content": "achieve the final imputed vector,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        239,
+                                        721,
+                                        249,
+                                        732
+                                    ],
+                                    "score": 0.88,
+                                    "content": "\\widehat { \\pmb { x } } _ { i }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        250,
+                                        720,
+                                        505,
+                                        734
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", we blend (or, merge) the output of the generator with the input",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                81,
+                                423,
+                                96
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        81,
+                                        423,
+                                        96
+                                    ],
+                                    "score": 1.0,
+                                    "content": "features to replace generated values with the exact values of observed features:",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 0
+                        }
+                    ],
+                    "index": 43,
+                    "bbox_fs": [
+                        105,
+                        697,
+                        506,
+                        734
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        82,
+                        422,
+                        94
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                81,
+                                423,
+                                96
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        81,
+                                        423,
+                                        96
+                                    ],
+                                    "score": 1.0,
+                                    "content": "features to replace generated values with the exact values of observed features:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        }
+                    ],
+                    "index": 0
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        229,
+                        100,
+                        381,
+                        128
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                229,
+                                100,
+                                381,
+                                128
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        229,
+                                        100,
+                                        381,
+                                        128
+                                    ],
+                                    "score": 0.94,
+                                    "content": "\\begin{array} { r } { \\widehat { \\pmb { x } } _ { i , j } = \\left\\{ \\begin{array} { l l } { \\pmb { x } _ { i , j } } & { \\mathrm { i f } ~ \\pmb { k } _ { i , j } = 1 } \\\\ { G ( \\pmb { x } _ { i } , \\pmb { k } _ { i } , z ) _ { j } } & { \\mathrm { i f } ~ \\pmb { k } _ { i , j } = 0 } \\end{array} \\right. , } \\end{array}",
+                                    "type": "interline_equation",
+                                    "image_path": "316b443fc9a2d734705d8ec9fb97c77cc89aef805263dbe96b8ab476ecdb014c.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 1.5,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                229,
+                                100,
+                                381,
+                                114.0
+                            ],
+                            "spans": [],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                229,
+                                114.0,
+                                381,
+                                128.0
+                            ],
+                            "spans": [],
+                            "index": 2
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        133,
+                        504,
+                        168
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                132,
+                                505,
+                                147
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        132,
+                                        133,
+                                        147
+                                    ],
+                                    "score": 1.0,
+                                    "content": "where",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        133,
+                                        136,
+                                        150,
+                                        146
+                                    ],
+                                    "score": 0.89,
+                                    "content": "\\mathbf { \\delta } _ { \\mathbf { x } _ { i , j } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        151,
+                                        132,
+                                        187,
+                                        147
+                                    ],
+                                    "score": 1.0,
+                                    "content": "refers to",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        188,
+                                        135,
+                                        195,
+                                        145
+                                    ],
+                                    "score": 0.81,
+                                    "content": "j ^ { : }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        195,
+                                        132,
+                                        279,
+                                        147
+                                    ],
+                                    "score": 1.0,
+                                    "content": "’th feature of sample",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        279,
+                                        135,
+                                        284,
+                                        144
+                                    ],
+                                    "score": 0.72,
+                                    "content": "i",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        284,
+                                        132,
+                                        402,
+                                        147
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Also, note that by sampling",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        402,
+                                        136,
+                                        410,
+                                        144
+                                    ],
+                                    "score": 0.75,
+                                    "content": "_ { z }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        410,
+                                        132,
+                                        505,
+                                        147
+                                    ],
+                                    "score": 1.0,
+                                    "content": "multiple times, we can",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                145,
+                                505,
+                                158
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        145,
+                                        436,
+                                        158
+                                    ],
+                                    "score": 1.0,
+                                    "content": "obtain different imputation samples from the conditional distribution indicated by",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        436,
+                                        145,
+                                        447,
+                                        158
+                                    ],
+                                    "score": 0.9,
+                                    "content": "\\widehat { \\pmb { x } } _ { i } ^ { l }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        448,
+                                        145,
+                                        475,
+                                        158
+                                    ],
+                                    "score": 1.0,
+                                    "content": "where",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        475,
+                                        146,
+                                        480,
+                                        156
+                                    ],
+                                    "score": 0.76,
+                                    "content": "l",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        480,
+                                        145,
+                                        505,
+                                        158
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                156,
+                                172,
+                                169
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        156,
+                                        172,
+                                        169
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sample number.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        }
+                    ],
+                    "index": 4
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        173,
+                        505,
+                        209
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                172,
+                                506,
+                                187
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        172,
+                                        209,
+                                        187
+                                    ],
+                                    "score": 1.0,
+                                    "content": "A discriminator network,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        210,
+                                        174,
+                                        236,
+                                        186
+                                    ],
+                                    "score": 0.91,
+                                    "content": "D ( \\widehat { \\mathbf { x } } _ { i } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        237,
+                                        172,
+                                        506,
+                                        187
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", is trained to distinguish real and imputed features by generating a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                185,
+                                505,
+                                200
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        187,
+                                        233,
+                                        200
+                                    ],
+                                    "score": 1.0,
+                                    "content": "predicted softmax mask output,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        233,
+                                        185,
+                                        244,
+                                        198
+                                    ],
+                                    "score": 0.88,
+                                    "content": "\\widehat { \\pmb { k } } _ { i }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        244,
+                                        187,
+                                        505,
+                                        200
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Here a binary cross-entropy loss per mask element is used as the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                198,
+                                230,
+                                211
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        198,
+                                        230,
+                                        211
+                                    ],
+                                    "score": 1.0,
+                                    "content": "adversarial objective function:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        }
+                    ],
+                    "index": 7
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        141,
+                        214,
+                        469,
+                        235
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                141,
+                                214,
+                                469,
+                                235
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        214,
+                                        469,
+                                        235
+                                    ],
+                                    "score": 0.89,
+                                    "content": "\\operatorname* { m a x } _ { G } \\operatorname* { m i n } _ { D } L ( G , D ) = \\mathbb { E } _ { k \\sim \\mathcal { D } , \\widehat { k } \\sim D ( G ( x , k , z ) ) } \\ [ k ^ { T } \\log ( \\widehat { k } ) \\ + \\ ( 1 - k ) ^ { T } \\log ( 1 - \\widehat { k } ) ] .",
+                                    "type": "interline_equation",
+                                    "image_path": "05f2884f3e6458cf6cd8b12ada277081d36c3ce2f6721f115066e191fe65151b.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 9,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                141,
+                                214,
+                                469,
+                                235
+                            ],
+                            "spans": [],
+                            "index": 9
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        245,
+                        505,
+                        280
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                245,
+                                505,
+                                259
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        245,
+                                        505,
+                                        259
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The intuition behind this adversarial loss function is that given a generator function which captures",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                256,
+                                505,
+                                270
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        256,
+                                        505,
+                                        270
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the data distribution successfully, the discriminator would not be able to distinguish the parts of the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                267,
+                                280,
+                                281
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        267,
+                                        280,
+                                        281
+                                    ],
+                                    "score": 1.0,
+                                    "content": "feature vector that were originally missing.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        }
+                    ],
+                    "index": 11
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        284,
+                        505,
+                        362
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                285,
+                                506,
+                                297
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        285,
+                                        506,
+                                        297
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Compared to Yoon et al. (2018), the objective function of (2) does not have an MSE loss term. Instead,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                295,
+                                506,
+                                309
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        295,
+                                        506,
+                                        309
+                                    ],
+                                    "score": 1.0,
+                                    "content": "we use recent advances in GAN stabilization and training to improve the training process (see Section",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                307,
+                                505,
+                                319
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        307,
+                                        505,
+                                        319
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3.4). While it is quite prevalent in the adversarial learning literature to use additional loss terms such",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                318,
+                                506,
+                                330
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        318,
+                                        506,
+                                        330
+                                    ],
+                                    "score": 1.0,
+                                    "content": "as mean squared error (MSE) to enhance the quality of generated samples, we decided to keep our",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                329,
+                                505,
+                                342
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        329,
+                                        505,
+                                        342
+                                    ],
+                                    "score": 1.0,
+                                    "content": "solution as simple as possible. Additionally, in our experiments, we provide supporting evidence that",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                340,
+                                506,
+                                352
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        340,
+                                        506,
+                                        352
+                                    ],
+                                    "score": 1.0,
+                                    "content": "this simple loss function enables us to sample from the conditional distribution and prevents biased",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                351,
+                                264,
+                                362
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        351,
+                                        264,
+                                        362
+                                    ],
+                                    "score": 1.0,
+                                    "content": "inclinations toward distribution modes.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        }
+                    ],
+                    "index": 16
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        375,
+                        243,
+                        387
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                375,
+                                244,
+                                388
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        375,
+                                        244,
+                                        388
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3.3 STOCHASTIC PREDICTION",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        }
+                    ],
+                    "index": 20
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        395,
+                        505,
+                        463
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                396,
+                                505,
+                                408
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        396,
+                                        505,
+                                        408
+                                    ],
+                                    "score": 1.0,
+                                    "content": "To capture the distribution of target classes given incomplete data, we suggest the idea of stochastic",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                407,
+                                505,
+                                420
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        407,
+                                        505,
+                                        420
+                                    ],
+                                    "score": 1.0,
+                                    "content": "prediction. As indicated in the previous section, the generator can be used to sample from the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                417,
+                                505,
+                                430
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        417,
+                                        505,
+                                        430
+                                    ],
+                                    "score": 1.0,
+                                    "content": "conditional distribution. Here, a predictor is trained based on the imputed samples to predict class",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                429,
+                                506,
+                                442
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        429,
+                                        506,
+                                        442
+                                    ],
+                                    "score": 1.0,
+                                    "content": "assignments and to calculate the confidence of these assignments. For instance, for a specific test",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                441,
+                                505,
+                                452
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        441,
+                                        505,
+                                        452
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sample at hand, if a certain missing feature is a strong indicator of the target class, we would like to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                451,
+                                437,
+                                463
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        451,
+                                        437,
+                                        463
+                                    ],
+                                    "score": 1.0,
+                                    "content": "observe the impact of different imputations for that feature on the final hypothesis.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        }
+                    ],
+                    "index": 23.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        467,
+                        492,
+                        479
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                466,
+                                493,
+                                480
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        466,
+                                        493,
+                                        480
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Formally, we are interested in finding the certainty of class assignments given observed features:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        }
+                    ],
+                    "index": 27
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        271,
+                        484,
+                        340,
+                        500
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                271,
+                                484,
+                                340,
+                                500
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        271,
+                                        484,
+                                        340,
+                                        500
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\Psi = P ( y | \\mathbf { x } _ { i } ^ { o b s } ) .",
+                                    "type": "interline_equation",
+                                    "image_path": "ac9223ff756409c368c037770a52636ea42d523d930f2c3fec7e95ff161a5778.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 28,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                271,
+                                484,
+                                340,
+                                500
+                            ],
+                            "spans": [],
+                            "index": 28
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        505,
+                        507,
+                        525
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                503,
+                                505,
+                                518
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        503,
+                                        131,
+                                        518
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Here,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        131,
+                                        505,
+                                        140,
+                                        515
+                                    ],
+                                    "score": 0.84,
+                                    "content": "\\Psi",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        141,
+                                        503,
+                                        505,
+                                        518
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is a vector where each element is representing a certain class. Rewriting (3) as a marginal",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                515,
+                                145,
+                                528
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        515,
+                                        145,
+                                        528
+                                    ],
+                                    "score": 1.0,
+                                    "content": "we have:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        }
+                    ],
+                    "index": 29.5
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        215,
+                        523,
+                        396,
+                        549
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                215,
+                                523,
+                                396,
+                                549
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        215,
+                                        523,
+                                        396,
+                                        549
+                                    ],
+                                    "score": 0.94,
+                                    "content": "\\Psi = \\int P ( \\pmb { x } _ { i } ^ { m i s s } ) P ( y | \\pmb { x } _ { i } ^ { o b s } , \\pmb { x } _ { i } ^ { m i s s } ) d \\pmb { x } _ { i } ^ { m i s s } .",
+                                    "type": "interline_equation",
+                                    "image_path": "470d6e7a29f360f57c1cae6febf98824385510b7dba8545cf78581e9577b11bb.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 31,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                215,
+                                523,
+                                396,
+                                549
+                            ],
+                            "spans": [],
+                            "index": 31
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        551,
+                        494,
+                        563
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                549,
+                                496,
+                                565
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        549,
+                                        403,
+                                        565
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Approximating the integration using a summation, given enough samples,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        403,
+                                        551,
+                                        412,
+                                        561
+                                    ],
+                                    "score": 0.85,
+                                    "content": "\\Psi",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        412,
+                                        549,
+                                        496,
+                                        565
+                                    ],
+                                    "score": 1.0,
+                                    "content": "can be estimated by:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        }
+                    ],
+                    "index": 32
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        242,
+                        567,
+                        368,
+                        591
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                242,
+                                567,
+                                368,
+                                591
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        242,
+                                        567,
+                                        368,
+                                        591
+                                    ],
+                                    "score": 0.95,
+                                    "content": "\\Psi \\approx \\frac { 1 } { N } \\sum P ( y | \\pmb { x } _ { i } ^ { o b s } , \\widehat { \\pmb { x } } _ { i } ^ { m i s s } ) ,",
+                                    "type": "interline_equation",
+                                    "image_path": "cc396e8a45febe2930d813ed68f2f15e7c1080af486e7d6e38ae1739a25905e1.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 33,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                242,
+                                567,
+                                368,
+                                591
+                            ],
+                            "spans": [],
+                            "index": 33
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        596,
+                        505,
+                        631
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                596,
+                                505,
+                                609
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        596,
+                                        133,
+                                        609
+                                    ],
+                                    "score": 1.0,
+                                    "content": "where",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        133,
+                                        597,
+                                        158,
+                                        609
+                                    ],
+                                    "score": 0.9,
+                                    "content": "\\widehat { \\pmb { x } } _ { i } ^ { m i s s }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        158,
+                                        596,
+                                        505,
+                                        609
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are samples taken from the conditional distribution of missing features given observed",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                608,
+                                505,
+                                621
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        608,
+                                        505,
+                                        621
+                                    ],
+                                    "score": 1.0,
+                                    "content": "bones. We use the suggested generative imputation method to generate samples required for this",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                619,
+                                470,
+                                631
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        619,
+                                        470,
+                                        631
+                                    ],
+                                    "score": 1.0,
+                                    "content": "approximation. Rewriting (1) using Hadamard product and as function of the noise vector:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        }
+                    ],
+                    "index": 35
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        223,
+                        636,
+                        387,
+                        650
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                223,
+                                636,
+                                387,
+                                650
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        223,
+                                        636,
+                                        387,
+                                        650
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\widehat { \\mathbf { x } } _ { i } = \\pmb { k } _ { i } \\odot \\pmb { x } _ { i } + \\left( 1 - \\pmb { k } _ { i } \\right) \\odot G ( \\pmb { x } _ { i } , \\pmb { k } _ { i } , z )",
+                                    "type": "interline_equation",
+                                    "image_path": "ea12ac3bd2ae21bbd617c307f801f8487f98ff3bda3cc54317f9fad1e9bde676.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 37,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                223,
+                                636,
+                                387,
+                                650
+                            ],
+                            "spans": [],
+                            "index": 37
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        104,
+                        654,
+                        505,
+                        677
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                653,
+                                505,
+                                668
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        653,
+                                        215,
+                                        668
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Assuming that a predictor,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        215,
+                                        655,
+                                        227,
+                                        666
+                                    ],
+                                    "score": 0.89,
+                                    "content": "F _ { \\theta }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        227,
+                                        653,
+                                        505,
+                                        668
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", is available which predicts class assignments for a complete feature",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                666,
+                                228,
+                                677
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        666,
+                                        136,
+                                        677
+                                    ],
+                                    "score": 1.0,
+                                    "content": "vector,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        136,
+                                        667,
+                                        145,
+                                        675
+                                    ],
+                                    "score": 0.82,
+                                    "content": "\\Psi",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        145,
+                                        666,
+                                        228,
+                                        677
+                                    ],
+                                    "score": 1.0,
+                                    "content": "can be estimated as:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        }
+                    ],
+                    "index": 38.5
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        231,
+                        681,
+                        380,
+                        705
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                231,
+                                681,
+                                380,
+                                705
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        231,
+                                        681,
+                                        380,
+                                        705
+                                    ],
+                                    "score": 0.95,
+                                    "content": "\\Psi = \\mathbb { E } _ { \\mathbf { z } } [ F _ { \\boldsymbol { \\theta } } ( \\widehat { \\mathbf { x } } _ { i } ) ] \\approx \\frac { 1 } { N } { \\sum _ { l = 1 } ^ { N } } F _ { \\boldsymbol { \\theta } } ( \\widehat { \\mathbf { x } } _ { i } ^ { l } ) ~ .",
+                                    "type": "interline_equation",
+                                    "image_path": "bfb16b641f19556475102ee95cb45ba3863f00d59720a9e9105fdc78b4be5e3d.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 40,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                231,
+                                681,
+                                380,
+                                705
+                            ],
+                            "spans": [],
+                            "index": 40
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        709,
+                        505,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                709,
+                                505,
+                                722
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        709,
+                                        505,
+                                        722
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Algorithm 1 presents the suggested algorithm for training the predictor. It consists of taking samples",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                720,
+                                505,
+                                733
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        720,
+                                        505,
+                                        733
+                                    ],
+                                    "score": 1.0,
+                                    "content": "from the incomplete dataset, then imputing them using our generator network, and using the imputed",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        }
+                    ],
+                    "index": 41.5
+                }
+            ],
+            "page_idx": 3,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        106,
+                        26,
+                        308,
+                        38
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                25,
+                                309,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        25,
+                                        309,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Under review as a conference paper at ICLR 2020",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        302,
+                        751,
+                        308,
+                        759
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                301,
+                                750,
+                                309,
+                                762
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        301,
+                                        750,
+                                        309,
+                                        762
+                                    ],
+                                    "score": 1.0,
+                                    "content": "",
+                                    "type": "text",
+                                    "height": 12,
+                                    "width": 8
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        82,
+                        422,
+                        94
+                    ],
+                    "lines": [],
+                    "index": 0,
+                    "bbox_fs": [
+                        105,
+                        81,
+                        423,
+                        96
+                    ],
+                    "lines_deleted": true
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        229,
+                        100,
+                        381,
+                        128
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                229,
+                                100,
+                                381,
+                                128
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        229,
+                                        100,
+                                        381,
+                                        128
+                                    ],
+                                    "score": 0.94,
+                                    "content": "\\begin{array} { r } { \\widehat { \\pmb { x } } _ { i , j } = \\left\\{ \\begin{array} { l l } { \\pmb { x } _ { i , j } } & { \\mathrm { i f } ~ \\pmb { k } _ { i , j } = 1 } \\\\ { G ( \\pmb { x } _ { i } , \\pmb { k } _ { i } , z ) _ { j } } & { \\mathrm { i f } ~ \\pmb { k } _ { i , j } = 0 } \\end{array} \\right. , } \\end{array}",
+                                    "type": "interline_equation",
+                                    "image_path": "316b443fc9a2d734705d8ec9fb97c77cc89aef805263dbe96b8ab476ecdb014c.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 1.5,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                229,
+                                100,
+                                381,
+                                114.0
+                            ],
+                            "spans": [],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                229,
+                                114.0,
+                                381,
+                                128.0
+                            ],
+                            "spans": [],
+                            "index": 2
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        133,
+                        504,
+                        168
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                132,
+                                505,
+                                147
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        132,
+                                        133,
+                                        147
+                                    ],
+                                    "score": 1.0,
+                                    "content": "where",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        133,
+                                        136,
+                                        150,
+                                        146
+                                    ],
+                                    "score": 0.89,
+                                    "content": "\\mathbf { \\delta } _ { \\mathbf { x } _ { i , j } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        151,
+                                        132,
+                                        187,
+                                        147
+                                    ],
+                                    "score": 1.0,
+                                    "content": "refers to",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        188,
+                                        135,
+                                        195,
+                                        145
+                                    ],
+                                    "score": 0.81,
+                                    "content": "j ^ { : }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        195,
+                                        132,
+                                        279,
+                                        147
+                                    ],
+                                    "score": 1.0,
+                                    "content": "’th feature of sample",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        279,
+                                        135,
+                                        284,
+                                        144
+                                    ],
+                                    "score": 0.72,
+                                    "content": "i",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        284,
+                                        132,
+                                        402,
+                                        147
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Also, note that by sampling",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        402,
+                                        136,
+                                        410,
+                                        144
+                                    ],
+                                    "score": 0.75,
+                                    "content": "_ { z }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        410,
+                                        132,
+                                        505,
+                                        147
+                                    ],
+                                    "score": 1.0,
+                                    "content": "multiple times, we can",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                145,
+                                505,
+                                158
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        145,
+                                        436,
+                                        158
+                                    ],
+                                    "score": 1.0,
+                                    "content": "obtain different imputation samples from the conditional distribution indicated by",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        436,
+                                        145,
+                                        447,
+                                        158
+                                    ],
+                                    "score": 0.9,
+                                    "content": "\\widehat { \\pmb { x } } _ { i } ^ { l }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        448,
+                                        145,
+                                        475,
+                                        158
+                                    ],
+                                    "score": 1.0,
+                                    "content": "where",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        475,
+                                        146,
+                                        480,
+                                        156
+                                    ],
+                                    "score": 0.76,
+                                    "content": "l",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        480,
+                                        145,
+                                        505,
+                                        158
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                156,
+                                172,
+                                169
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        156,
+                                        172,
+                                        169
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sample number.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        }
+                    ],
+                    "index": 4,
+                    "bbox_fs": [
+                        105,
+                        132,
+                        505,
+                        169
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        173,
+                        505,
+                        209
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                172,
+                                506,
+                                187
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        172,
+                                        209,
+                                        187
+                                    ],
+                                    "score": 1.0,
+                                    "content": "A discriminator network,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        210,
+                                        174,
+                                        236,
+                                        186
+                                    ],
+                                    "score": 0.91,
+                                    "content": "D ( \\widehat { \\mathbf { x } } _ { i } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        237,
+                                        172,
+                                        506,
+                                        187
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", is trained to distinguish real and imputed features by generating a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                185,
+                                505,
+                                200
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        187,
+                                        233,
+                                        200
+                                    ],
+                                    "score": 1.0,
+                                    "content": "predicted softmax mask output,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        233,
+                                        185,
+                                        244,
+                                        198
+                                    ],
+                                    "score": 0.88,
+                                    "content": "\\widehat { \\pmb { k } } _ { i }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        244,
+                                        187,
+                                        505,
+                                        200
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Here a binary cross-entropy loss per mask element is used as the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                198,
+                                230,
+                                211
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        198,
+                                        230,
+                                        211
+                                    ],
+                                    "score": 1.0,
+                                    "content": "adversarial objective function:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        }
+                    ],
+                    "index": 7,
+                    "bbox_fs": [
+                        104,
+                        172,
+                        506,
+                        211
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        141,
+                        214,
+                        469,
+                        235
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                141,
+                                214,
+                                469,
+                                235
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        214,
+                                        469,
+                                        235
+                                    ],
+                                    "score": 0.89,
+                                    "content": "\\operatorname* { m a x } _ { G } \\operatorname* { m i n } _ { D } L ( G , D ) = \\mathbb { E } _ { k \\sim \\mathcal { D } , \\widehat { k } \\sim D ( G ( x , k , z ) ) } \\ [ k ^ { T } \\log ( \\widehat { k } ) \\ + \\ ( 1 - k ) ^ { T } \\log ( 1 - \\widehat { k } ) ] .",
+                                    "type": "interline_equation",
+                                    "image_path": "05f2884f3e6458cf6cd8b12ada277081d36c3ce2f6721f115066e191fe65151b.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 9,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                141,
+                                214,
+                                469,
+                                235
+                            ],
+                            "spans": [],
+                            "index": 9
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        245,
+                        505,
+                        280
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                245,
+                                505,
+                                259
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        245,
+                                        505,
+                                        259
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The intuition behind this adversarial loss function is that given a generator function which captures",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                256,
+                                505,
+                                270
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        256,
+                                        505,
+                                        270
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the data distribution successfully, the discriminator would not be able to distinguish the parts of the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                267,
+                                280,
+                                281
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        267,
+                                        280,
+                                        281
+                                    ],
+                                    "score": 1.0,
+                                    "content": "feature vector that were originally missing.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        }
+                    ],
+                    "index": 11,
+                    "bbox_fs": [
+                        105,
+                        245,
+                        505,
+                        281
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        284,
+                        505,
+                        362
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                285,
+                                506,
+                                297
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        285,
+                                        506,
+                                        297
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Compared to Yoon et al. (2018), the objective function of (2) does not have an MSE loss term. Instead,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                295,
+                                506,
+                                309
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        295,
+                                        506,
+                                        309
+                                    ],
+                                    "score": 1.0,
+                                    "content": "we use recent advances in GAN stabilization and training to improve the training process (see Section",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                307,
+                                505,
+                                319
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        307,
+                                        505,
+                                        319
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3.4). While it is quite prevalent in the adversarial learning literature to use additional loss terms such",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                318,
+                                506,
+                                330
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        318,
+                                        506,
+                                        330
+                                    ],
+                                    "score": 1.0,
+                                    "content": "as mean squared error (MSE) to enhance the quality of generated samples, we decided to keep our",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                329,
+                                505,
+                                342
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        329,
+                                        505,
+                                        342
+                                    ],
+                                    "score": 1.0,
+                                    "content": "solution as simple as possible. Additionally, in our experiments, we provide supporting evidence that",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                340,
+                                506,
+                                352
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        340,
+                                        506,
+                                        352
+                                    ],
+                                    "score": 1.0,
+                                    "content": "this simple loss function enables us to sample from the conditional distribution and prevents biased",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                351,
+                                264,
+                                362
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        351,
+                                        264,
+                                        362
+                                    ],
+                                    "score": 1.0,
+                                    "content": "inclinations toward distribution modes.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        }
+                    ],
+                    "index": 16,
+                    "bbox_fs": [
+                        105,
+                        285,
+                        506,
+                        362
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        375,
+                        243,
+                        387
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                375,
+                                244,
+                                388
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        375,
+                                        244,
+                                        388
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3.3 STOCHASTIC PREDICTION",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        }
+                    ],
+                    "index": 20
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        395,
+                        505,
+                        463
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                396,
+                                505,
+                                408
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        396,
+                                        505,
+                                        408
+                                    ],
+                                    "score": 1.0,
+                                    "content": "To capture the distribution of target classes given incomplete data, we suggest the idea of stochastic",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                407,
+                                505,
+                                420
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        407,
+                                        505,
+                                        420
+                                    ],
+                                    "score": 1.0,
+                                    "content": "prediction. As indicated in the previous section, the generator can be used to sample from the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                417,
+                                505,
+                                430
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        417,
+                                        505,
+                                        430
+                                    ],
+                                    "score": 1.0,
+                                    "content": "conditional distribution. Here, a predictor is trained based on the imputed samples to predict class",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                429,
+                                506,
+                                442
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        429,
+                                        506,
+                                        442
+                                    ],
+                                    "score": 1.0,
+                                    "content": "assignments and to calculate the confidence of these assignments. For instance, for a specific test",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                441,
+                                505,
+                                452
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        441,
+                                        505,
+                                        452
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sample at hand, if a certain missing feature is a strong indicator of the target class, we would like to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                451,
+                                437,
+                                463
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        451,
+                                        437,
+                                        463
+                                    ],
+                                    "score": 1.0,
+                                    "content": "observe the impact of different imputations for that feature on the final hypothesis.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        }
+                    ],
+                    "index": 23.5,
+                    "bbox_fs": [
+                        105,
+                        396,
+                        506,
+                        463
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        467,
+                        492,
+                        479
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                466,
+                                493,
+                                480
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        466,
+                                        493,
+                                        480
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Formally, we are interested in finding the certainty of class assignments given observed features:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        }
+                    ],
+                    "index": 27,
+                    "bbox_fs": [
+                        106,
+                        466,
+                        493,
+                        480
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        271,
+                        484,
+                        340,
+                        500
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                271,
+                                484,
+                                340,
+                                500
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        271,
+                                        484,
+                                        340,
+                                        500
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\Psi = P ( y | \\mathbf { x } _ { i } ^ { o b s } ) .",
+                                    "type": "interline_equation",
+                                    "image_path": "ac9223ff756409c368c037770a52636ea42d523d930f2c3fec7e95ff161a5778.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 28,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                271,
+                                484,
+                                340,
+                                500
+                            ],
+                            "spans": [],
+                            "index": 28
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        505,
+                        507,
+                        525
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                503,
+                                505,
+                                518
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        503,
+                                        131,
+                                        518
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Here,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        131,
+                                        505,
+                                        140,
+                                        515
+                                    ],
+                                    "score": 0.84,
+                                    "content": "\\Psi",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        141,
+                                        503,
+                                        505,
+                                        518
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is a vector where each element is representing a certain class. Rewriting (3) as a marginal",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                515,
+                                145,
+                                528
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        515,
+                                        145,
+                                        528
+                                    ],
+                                    "score": 1.0,
+                                    "content": "we have:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        }
+                    ],
+                    "index": 29.5,
+                    "bbox_fs": [
+                        105,
+                        503,
+                        505,
+                        528
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        215,
+                        523,
+                        396,
+                        549
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                215,
+                                523,
+                                396,
+                                549
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        215,
+                                        523,
+                                        396,
+                                        549
+                                    ],
+                                    "score": 0.94,
+                                    "content": "\\Psi = \\int P ( \\pmb { x } _ { i } ^ { m i s s } ) P ( y | \\pmb { x } _ { i } ^ { o b s } , \\pmb { x } _ { i } ^ { m i s s } ) d \\pmb { x } _ { i } ^ { m i s s } .",
+                                    "type": "interline_equation",
+                                    "image_path": "470d6e7a29f360f57c1cae6febf98824385510b7dba8545cf78581e9577b11bb.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 31,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                215,
+                                523,
+                                396,
+                                549
+                            ],
+                            "spans": [],
+                            "index": 31
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        551,
+                        494,
+                        563
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                549,
+                                496,
+                                565
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        549,
+                                        403,
+                                        565
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Approximating the integration using a summation, given enough samples,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        403,
+                                        551,
+                                        412,
+                                        561
+                                    ],
+                                    "score": 0.85,
+                                    "content": "\\Psi",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        412,
+                                        549,
+                                        496,
+                                        565
+                                    ],
+                                    "score": 1.0,
+                                    "content": "can be estimated by:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        }
+                    ],
+                    "index": 32,
+                    "bbox_fs": [
+                        105,
+                        549,
+                        496,
+                        565
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        242,
+                        567,
+                        368,
+                        591
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                242,
+                                567,
+                                368,
+                                591
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        242,
+                                        567,
+                                        368,
+                                        591
+                                    ],
+                                    "score": 0.95,
+                                    "content": "\\Psi \\approx \\frac { 1 } { N } \\sum P ( y | \\pmb { x } _ { i } ^ { o b s } , \\widehat { \\pmb { x } } _ { i } ^ { m i s s } ) ,",
+                                    "type": "interline_equation",
+                                    "image_path": "cc396e8a45febe2930d813ed68f2f15e7c1080af486e7d6e38ae1739a25905e1.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 33,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                242,
+                                567,
+                                368,
+                                591
+                            ],
+                            "spans": [],
+                            "index": 33
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        596,
+                        505,
+                        631
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                596,
+                                505,
+                                609
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        596,
+                                        133,
+                                        609
+                                    ],
+                                    "score": 1.0,
+                                    "content": "where",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        133,
+                                        597,
+                                        158,
+                                        609
+                                    ],
+                                    "score": 0.9,
+                                    "content": "\\widehat { \\pmb { x } } _ { i } ^ { m i s s }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        158,
+                                        596,
+                                        505,
+                                        609
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are samples taken from the conditional distribution of missing features given observed",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                608,
+                                505,
+                                621
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        608,
+                                        505,
+                                        621
+                                    ],
+                                    "score": 1.0,
+                                    "content": "bones. We use the suggested generative imputation method to generate samples required for this",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                619,
+                                470,
+                                631
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        619,
+                                        470,
+                                        631
+                                    ],
+                                    "score": 1.0,
+                                    "content": "approximation. Rewriting (1) using Hadamard product and as function of the noise vector:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        }
+                    ],
+                    "index": 35,
+                    "bbox_fs": [
+                        104,
+                        596,
+                        505,
+                        631
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        223,
+                        636,
+                        387,
+                        650
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                223,
+                                636,
+                                387,
+                                650
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        223,
+                                        636,
+                                        387,
+                                        650
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\widehat { \\mathbf { x } } _ { i } = \\pmb { k } _ { i } \\odot \\pmb { x } _ { i } + \\left( 1 - \\pmb { k } _ { i } \\right) \\odot G ( \\pmb { x } _ { i } , \\pmb { k } _ { i } , z )",
+                                    "type": "interline_equation",
+                                    "image_path": "ea12ac3bd2ae21bbd617c307f801f8487f98ff3bda3cc54317f9fad1e9bde676.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 37,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                223,
+                                636,
+                                387,
+                                650
+                            ],
+                            "spans": [],
+                            "index": 37
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        104,
+                        654,
+                        505,
+                        677
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                653,
+                                505,
+                                668
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        653,
+                                        215,
+                                        668
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Assuming that a predictor,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        215,
+                                        655,
+                                        227,
+                                        666
+                                    ],
+                                    "score": 0.89,
+                                    "content": "F _ { \\theta }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        227,
+                                        653,
+                                        505,
+                                        668
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", is available which predicts class assignments for a complete feature",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                666,
+                                228,
+                                677
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        666,
+                                        136,
+                                        677
+                                    ],
+                                    "score": 1.0,
+                                    "content": "vector,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        136,
+                                        667,
+                                        145,
+                                        675
+                                    ],
+                                    "score": 0.82,
+                                    "content": "\\Psi",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        145,
+                                        666,
+                                        228,
+                                        677
+                                    ],
+                                    "score": 1.0,
+                                    "content": "can be estimated as:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        }
+                    ],
+                    "index": 38.5,
+                    "bbox_fs": [
+                        105,
+                        653,
+                        505,
+                        677
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        231,
+                        681,
+                        380,
+                        705
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                231,
+                                681,
+                                380,
+                                705
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        231,
+                                        681,
+                                        380,
+                                        705
+                                    ],
+                                    "score": 0.95,
+                                    "content": "\\Psi = \\mathbb { E } _ { \\mathbf { z } } [ F _ { \\boldsymbol { \\theta } } ( \\widehat { \\mathbf { x } } _ { i } ) ] \\approx \\frac { 1 } { N } { \\sum _ { l = 1 } ^ { N } } F _ { \\boldsymbol { \\theta } } ( \\widehat { \\mathbf { x } } _ { i } ^ { l } ) ~ .",
+                                    "type": "interline_equation",
+                                    "image_path": "bfb16b641f19556475102ee95cb45ba3863f00d59720a9e9105fdc78b4be5e3d.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 40,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                231,
+                                681,
+                                380,
+                                705
+                            ],
+                            "spans": [],
+                            "index": 40
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        709,
+                        505,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                709,
+                                505,
+                                722
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        709,
+                                        505,
+                                        722
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Algorithm 1 presents the suggested algorithm for training the predictor. It consists of taking samples",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                720,
+                                505,
+                                733
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        720,
+                                        505,
+                                        733
+                                    ],
+                                    "score": 1.0,
+                                    "content": "from the incomplete dataset, then imputing them using our generator network, and using the imputed",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        }
+                    ],
+                    "index": 41.5,
+                    "bbox_fs": [
+                        106,
+                        709,
+                        505,
+                        733
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "table",
+                    "bbox": [
+                        106,
+                        84,
+                        291,
+                        221
+                    ],
+                    "blocks": [
+                        {
+                            "type": "table_body",
+                            "bbox": [
+                                106,
+                                84,
+                                291,
+                                221
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        84,
+                                        291,
+                                        221
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                84,
+                                                291,
+                                                221
+                                            ],
+                                            "score": 0.497,
+                                            "html": "<table><tr><td>Algorithm1: Training the predictor.</td></tr><tr><td>Input: G (trained imputer), D (dataset) Output: Fθ (trained predictor) foreachTrainingEpoch do foreach(xi,ki,yi) in D do z ~ N(0,I) Xi ←ki①xi+(1-ki)OG(xi,ki,z) loss ← L(yi,yred) Backpropagate loss</td></tr></table>",
+                                            "type": "table",
+                                            "image_path": "209c06eee3400b4d45d16029079eb73985b40561bfa9e60facddc6458b4b6abc.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 4.5,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        84,
+                                        291,
+                                        97.7
+                                    ],
+                                    "spans": [],
+                                    "index": 0
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        97.7,
+                                        291,
+                                        111.4
+                                    ],
+                                    "spans": [],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        111.4,
+                                        291,
+                                        125.10000000000001
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        125.10000000000001,
+                                        291,
+                                        138.8
+                                    ],
+                                    "spans": [],
+                                    "index": 3
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        138.8,
+                                        291,
+                                        152.5
+                                    ],
+                                    "spans": [],
+                                    "index": 4
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        152.5,
+                                        291,
+                                        166.2
+                                    ],
+                                    "spans": [],
+                                    "index": 5
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        166.2,
+                                        291,
+                                        179.89999999999998
+                                    ],
+                                    "spans": [],
+                                    "index": 6
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        179.89999999999998,
+                                        291,
+                                        193.59999999999997
+                                    ],
+                                    "spans": [],
+                                    "index": 7
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        193.59999999999997,
+                                        291,
+                                        207.29999999999995
+                                    ],
+                                    "spans": [],
+                                    "index": 8
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        207.29999999999995,
+                                        291,
+                                        220.99999999999994
+                                    ],
+                                    "spans": [],
+                                    "index": 9
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 4.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        296,
+                        85,
+                        478,
+                        216
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                296,
+                                85,
+                                478,
+                                98
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        296,
+                                        85,
+                                        478,
+                                        98
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Algorithm 2: Estimating target distributions.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                297,
+                                99,
+                                465,
+                                112
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        297,
+                                        99,
+                                        326,
+                                        112
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Input:",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        326,
+                                        100,
+                                        338,
+                                        111
+                                    ],
+                                    "score": 0.75,
+                                    "content": "F _ { \\theta }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        339,
+                                        99,
+                                        418,
+                                        112
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(trained predictor),",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        418,
+                                        100,
+                                        444,
+                                        111
+                                    ],
+                                    "score": 0.89,
+                                    "content": "( x , k )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        444,
+                                        99,
+                                        465,
+                                        112
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(test",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                325,
+                                110,
+                                454,
+                                122
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        325,
+                                        110,
+                                        362,
+                                        122
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sample),",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        362,
+                                        111,
+                                        372,
+                                        121
+                                    ],
+                                    "score": 0.36,
+                                    "content": "\\mathbf { N }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        372,
+                                        110,
+                                        454,
+                                        122
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(ensemble samples)",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                295,
+                                120,
+                                474,
+                                134
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        295,
+                                        120,
+                                        333,
+                                        134
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Output:",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        334,
+                                        122,
+                                        343,
+                                        131
+                                    ],
+                                    "score": 0.8,
+                                    "content": "\\Psi",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        343,
+                                        120,
+                                        474,
+                                        134
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(distribution over target classes)",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                295,
+                                131,
+                                399,
+                                144
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        295,
+                                        131,
+                                        399,
+                                        144
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Ψ ← zeros ∈ R#classes",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                296,
+                                143,
+                                443,
+                                155
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        296,
+                                        143,
+                                        403,
+                                        155
+                                    ],
+                                    "score": 1.0,
+                                    "content": "foreach Ensemble Sample",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        403,
+                                        144,
+                                        410,
+                                        153
+                                    ],
+                                    "score": 0.41,
+                                    "content": "^ { l }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        410,
+                                        143,
+                                        420,
+                                        155
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        421,
+                                        144,
+                                        430,
+                                        153
+                                    ],
+                                    "score": 0.56,
+                                    "content": "N",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        430,
+                                        143,
+                                        443,
+                                        155
+                                    ],
+                                    "score": 1.0,
+                                    "content": "do",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                310,
+                                154,
+                                364,
+                                166
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        310,
+                                        154,
+                                        364,
+                                        166
+                                    ],
+                                    "score": 1.0,
+                                    "content": "z ∼ N (0, I)",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                308,
+                                165,
+                                457,
+                                177
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        308,
+                                        165,
+                                        457,
+                                        177
+                                    ],
+                                    "score": 0.53,
+                                    "content": "\\widehat { \\mathbf { x } }  k \\odot x + ( 1 - k ) \\odot G ( { \\pmb { x } } , { \\pmb { k } } , z )",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                309,
+                                174,
+                                375,
+                                189
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        309,
+                                        174,
+                                        375,
+                                        189
+                                    ],
+                                    "score": 1.0,
+                                    "content": "bypred ← Fθ(x)",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                308,
+                                185,
+                                399,
+                                202
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        308,
+                                        185,
+                                        399,
+                                        202
+                                    ],
+                                    "score": 1.0,
+                                    "content": "bj ← argmax(ypred)",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                310,
+                                195,
+                                375,
+                                214
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        310,
+                                        195,
+                                        375,
+                                        214
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Ψj ← Ψj + 1N",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        }
+                    ],
+                    "index": 15
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        247,
+                        505,
+                        302
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                246,
+                                505,
+                                260
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        246,
+                                        505,
+                                        260
+                                    ],
+                                    "score": 1.0,
+                                    "content": "samples to update the predictor. Note that, on each epoch and for each sample, the generator generates",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                258,
+                                506,
+                                270
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        258,
+                                        506,
+                                        270
+                                    ],
+                                    "score": 1.0,
+                                    "content": "a new sample from the conditional distribution. Intuitively, it means that the predictor observes and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                268,
+                                505,
+                                283
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        268,
+                                        505,
+                                        283
+                                    ],
+                                    "score": 1.0,
+                                    "content": "learns to operate under different imputations for a given sample. This is different from approaches",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                280,
+                                506,
+                                293
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        280,
+                                        506,
+                                        293
+                                    ],
+                                    "score": 1.0,
+                                    "content": "such as multiple imputation where several predictors are trained on different imputed versions of a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                291,
+                                139,
+                                303
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        291,
+                                        139,
+                                        303
+                                    ],
+                                    "score": 1.0,
+                                    "content": "dataset.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        }
+                    ],
+                    "index": 23
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        307,
+                        505,
+                        364
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                308,
+                                506,
+                                320
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        308,
+                                        506,
+                                        320
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Algorithm 2 presents the suggested algorithm for making predictions and estimating target distribu-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                319,
+                                505,
+                                331
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        319,
+                                        368,
+                                        331
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tions given a trained predictor model. Here, a sample is imputed",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        369,
+                                        320,
+                                        379,
+                                        329
+                                    ],
+                                    "score": 0.8,
+                                    "content": "N",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        379,
+                                        319,
+                                        505,
+                                        331
+                                    ],
+                                    "score": 1.0,
+                                    "content": "times and inference on this set",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                329,
+                                505,
+                                342
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        329,
+                                        505,
+                                        342
+                                    ],
+                                    "score": 1.0,
+                                    "content": "results in an ensemble of predictions over different imputations. The output of this algorithm can be",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                340,
+                                506,
+                                354
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        340,
+                                        506,
+                                        354
+                                    ],
+                                    "score": 1.0,
+                                    "content": "interpreted as a distribution over the confidence of class assignments given a partially observed test",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                351,
+                                433,
+                                365
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        351,
+                                        433,
+                                        365
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sample. The following claims justify the validity of Algorithm 1and Algorithm 2.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        }
+                    ],
+                    "index": 28
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        368,
+                        503,
+                        402
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                367,
+                                505,
+                                383
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        367,
+                                        363,
+                                        383
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Claim 1. (Generalization of the predictor). If we assume imputed",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        363,
+                                        369,
+                                        377,
+                                        380
+                                    ],
+                                    "score": 0.86,
+                                    "content": "\\widehat { x _ { i } s }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        378,
+                                        367,
+                                        505,
+                                        383
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are samples from the underlying",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                379,
+                                505,
+                                393
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        379,
+                                        505,
+                                        393
+                                    ],
+                                    "score": 1.0,
+                                    "content": "bfeature distribution, then the assigned training set labels can be modeled as labels generated from a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                390,
+                                200,
+                                404
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        390,
+                                        200,
+                                        404
+                                    ],
+                                    "score": 1.0,
+                                    "content": "noisy labeling process.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        }
+                    ],
+                    "index": 32
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        407,
+                        505,
+                        452
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                406,
+                                505,
+                                421
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        406,
+                                        505,
+                                        421
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Claim 1 permits the analysis of the generalization and convergence for the predictor trained using",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                419,
+                                506,
+                                431
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        419,
+                                        506,
+                                        431
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Algorithm 1 based on current literature in training models with noisy labels (Natarajan et al., 2013;",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                429,
+                                505,
+                                443
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        429,
+                                        505,
+                                        443
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Reed et al., 2014; Chen et al., 2019). From the analysis provided by Chen et al. (2019), test accuracy",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                441,
+                                420,
+                                453
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        441,
+                                        420,
+                                        453
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in asymmetric label noise conditions is a quadratic function of the label noise:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        }
+                    ],
+                    "index": 35.5
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        246,
+                        456,
+                        364,
+                        471
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                246,
+                                456,
+                                364,
+                                471
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        246,
+                                        456,
+                                        364,
+                                        471
+                                    ],
+                                    "score": 0.92,
+                                    "content": "P ( y _ { i } = \\widehat { y _ { i } } ) = ( 1 - \\epsilon ) ^ { 2 } + \\epsilon ^ { 2 } ,",
+                                    "type": "interline_equation",
+                                    "image_path": "9ada2a757bda1fa063129713d7bfe7a24fe80e1f7c59bea48bebc2158db6572e.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 38,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                246,
+                                456,
+                                364,
+                                471
+                            ],
+                            "spans": [],
+                            "index": 38
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        475,
+                        505,
+                        509
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                474,
+                                505,
+                                488
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        474,
+                                        133,
+                                        488
+                                    ],
+                                    "score": 1.0,
+                                    "content": "where",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        133,
+                                        476,
+                                        142,
+                                        487
+                                    ],
+                                    "score": 0.87,
+                                    "content": "\\widehat { y _ { i } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        142,
+                                        474,
+                                        361,
+                                        488
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is underlying true label for the imputed feature vector",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        361,
+                                        476,
+                                        377,
+                                        487
+                                    ],
+                                    "score": 0.85,
+                                    "content": "( \\widehat { x _ { i } } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        378,
+                                        474,
+                                        398,
+                                        488
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        398,
+                                        477,
+                                        408,
+                                        487
+                                    ],
+                                    "score": 0.85,
+                                    "content": "y _ { i }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        408,
+                                        474,
+                                        505,
+                                        488
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is the label provided by",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                487,
+                                505,
+                                499
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        487,
+                                        300,
+                                        499
+                                    ],
+                                    "score": 1.0,
+                                    "content": "bthe incomplete dataset. In (8), label noise ratio,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        301,
+                                        489,
+                                        306,
+                                        497
+                                    ],
+                                    "score": 0.67,
+                                    "content": "\\epsilon",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        307,
+                                        487,
+                                        505,
+                                        499
+                                    ],
+                                    "score": 1.0,
+                                    "content": "b, represents the probability of the label transition",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                498,
+                                257,
+                                510
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        498,
+                                        257,
+                                        510
+                                    ],
+                                    "score": 1.0,
+                                    "content": "from a certain target class to another:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        }
+                    ],
+                    "index": 40
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        250,
+                        514,
+                        361,
+                        528
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                250,
+                                514,
+                                361,
+                                528
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        250,
+                                        514,
+                                        361,
+                                        528
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\epsilon = 1 - P ( \\widehat { y _ { i } } = j | y _ { i } = j ) .",
+                                    "type": "interline_equation",
+                                    "image_path": "2d600ebdd741a923b07c08d88e148eb392dd20a835c1cff559210c8791c19559.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 42,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                250,
+                                514,
+                                361,
+                                528
+                            ],
+                            "spans": [],
+                            "index": 42
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        532,
+                        505,
+                        555
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                532,
+                                505,
+                                545
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        532,
+                                        157,
+                                        545
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In practice,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        157,
+                                        535,
+                                        163,
+                                        543
+                                    ],
+                                    "score": 0.7,
+                                    "content": "\\epsilon",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        163,
+                                        532,
+                                        505,
+                                        545
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is determined by the problem-specific underlying data distribution as well as the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                543,
+                                229,
+                                556
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        543,
+                                        229,
+                                        556
+                                    ],
+                                    "score": 1.0,
+                                    "content": "distribution of missing values.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        }
+                    ],
+                    "index": 43.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        560,
+                        506,
+                        644
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                559,
+                                507,
+                                576
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        559,
+                                        330,
+                                        576
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Justification for claim 1 is straightforward, assume that",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        330,
+                                        560,
+                                        384,
+                                        574
+                                    ],
+                                    "score": 0.94,
+                                    "content": "\\{ \\widehat { y _ { i } } ^ { 1 } \\ldots \\widehat { y _ { i } } ^ { N } \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        385,
+                                        559,
+                                        507,
+                                        576
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are underlying true labels for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                102,
+                                568,
+                                510,
+                                593
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        102,
+                                        568,
+                                        138,
+                                        593
+                                    ],
+                                    "score": 1.0,
+                                    "content": "each of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        138,
+                                        573,
+                                        194,
+                                        587
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\{ \\widehat { x _ { i } } ^ { 1 } \\ldots \\widehat { x _ { i } } ^ { N } \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        194,
+                                        568,
+                                        371,
+                                        593
+                                    ],
+                                    "score": 1.0,
+                                    "content": "b b. During training, for any imputed sample in",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        371,
+                                        574,
+                                        426,
+                                        587
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\{ \\widehat { x _ { i } } ^ { 1 } \\ldots \\widehat { x _ { i } } ^ { N } \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        427,
+                                        568,
+                                        510,
+                                        593
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", we use the dataset",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                586,
+                                506,
+                                600
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        586,
+                                        172,
+                                        600
+                                    ],
+                                    "score": 1.0,
+                                    "content": "bprovided label,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        172,
+                                        589,
+                                        178,
+                                        598
+                                    ],
+                                    "score": 0.78,
+                                    "content": "y",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        178,
+                                        586,
+                                        506,
+                                        600
+                                    ],
+                                    "score": 1.0,
+                                    "content": "b b, to calculate the loss and to update model parameters. In the case that any of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 47
+                        },
+                        {
+                            "bbox": [
+                                107,
+                                597,
+                                506,
+                                613
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        597,
+                                        161,
+                                        611
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\{ \\widehat { y _ { i } } ^ { 1 } \\ldots \\widehat { y _ { i } } ^ { N } \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        161,
+                                        597,
+                                        229,
+                                        613
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is different from",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        229,
+                                        601,
+                                        236,
+                                        611
+                                    ],
+                                    "score": 0.76,
+                                    "content": "y",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        236,
+                                        597,
+                                        506,
+                                        613
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", the loss term corresponding to that term would be calculated using",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 48
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                609,
+                                506,
+                                622
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        609,
+                                        506,
+                                        622
+                                    ],
+                                    "score": 1.0,
+                                    "content": "b ba wrong label. Here, if we consider the average impact on gradients for batches of samples rather",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 49
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                619,
+                                506,
+                                635
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        619,
+                                        506,
+                                        635
+                                    ],
+                                    "score": 1.0,
+                                    "content": "than individual cases, the overall impact on training would be very similar to the case of training",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 50
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                631,
+                                183,
+                                644
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        631,
+                                        183,
+                                        644
+                                    ],
+                                    "score": 1.0,
+                                    "content": "using noisy labels.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 51
+                        }
+                    ],
+                    "index": 48
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        648,
+                        402,
+                        704
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                649,
+                                373,
+                                661
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        649,
+                                        373,
+                                        661
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Claim 2. (Approximation of the target distribution). If we assume:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 52
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                660,
+                                402,
+                                673
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        660,
+                                        152,
+                                        673
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(i) imputed",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        153,
+                                        660,
+                                        167,
+                                        671
+                                    ],
+                                    "score": 0.88,
+                                    "content": "\\widehat { x _ { i } s }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        167,
+                                        660,
+                                        402,
+                                        673
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are valid samples from the underlying feature distribution,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 53
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                671,
+                                357,
+                                683
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        671,
+                                        357,
+                                        683
+                                    ],
+                                    "score": 1.0,
+                                    "content": "b(ii) a good predictor can be trained using the incomplete data,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 54
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                682,
+                                401,
+                                694
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        682,
+                                        401,
+                                        694
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(iii) enough samples are used and the Monte Carlo estimator is unbiased,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 55
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                693,
+                                345,
+                                705
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        693,
+                                        218,
+                                        705
+                                    ],
+                                    "score": 1.0,
+                                    "content": "then the target distribution,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        218,
+                                        694,
+                                        226,
+                                        703
+                                    ],
+                                    "score": 0.77,
+                                    "content": "\\Psi",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        227,
+                                        693,
+                                        345,
+                                        705
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", can be estimated accurately.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 56
+                        }
+                    ],
+                    "index": 54
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        709,
+                        503,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                708,
+                                505,
+                                723
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        708,
+                                        505,
+                                        723
+                                    ],
+                                    "score": 1.0,
+                                    "content": "This claim supports Algorithm 2 that is suggested to estimate the target distribution given a partially",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 57
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                720,
+                                204,
+                                732
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        720,
+                                        204,
+                                        732
+                                    ],
+                                    "score": 1.0,
+                                    "content": "observed feature vector.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 58
+                        }
+                    ],
+                    "index": 57.5
+                }
+            ],
+            "page_idx": 4,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        107,
+                        27,
+                        308,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                25,
+                                309,
+                                39
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        25,
+                                        309,
+                                        39
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Under review as a conference paper at ICLR 2020",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        302,
+                        751,
+                        308,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                302,
+                                750,
+                                309,
+                                763
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        302,
+                                        750,
+                                        309,
+                                        763
+                                    ],
+                                    "score": 1.0,
+                                    "content": "5",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "table",
+                    "bbox": [
+                        106,
+                        84,
+                        291,
+                        221
+                    ],
+                    "blocks": [
+                        {
+                            "type": "table_body",
+                            "bbox": [
+                                106,
+                                84,
+                                291,
+                                221
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        84,
+                                        291,
+                                        221
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                84,
+                                                291,
+                                                221
+                                            ],
+                                            "score": 0.497,
+                                            "html": "<table><tr><td>Algorithm1: Training the predictor.</td></tr><tr><td>Input: G (trained imputer), D (dataset) Output: Fθ (trained predictor) foreachTrainingEpoch do foreach(xi,ki,yi) in D do z ~ N(0,I) Xi ←ki①xi+(1-ki)OG(xi,ki,z) loss ← L(yi,yred) Backpropagate loss</td></tr></table>",
+                                            "type": "table",
+                                            "image_path": "209c06eee3400b4d45d16029079eb73985b40561bfa9e60facddc6458b4b6abc.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 4.5,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        84,
+                                        291,
+                                        97.7
+                                    ],
+                                    "spans": [],
+                                    "index": 0
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        97.7,
+                                        291,
+                                        111.4
+                                    ],
+                                    "spans": [],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        111.4,
+                                        291,
+                                        125.10000000000001
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        125.10000000000001,
+                                        291,
+                                        138.8
+                                    ],
+                                    "spans": [],
+                                    "index": 3
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        138.8,
+                                        291,
+                                        152.5
+                                    ],
+                                    "spans": [],
+                                    "index": 4
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        152.5,
+                                        291,
+                                        166.2
+                                    ],
+                                    "spans": [],
+                                    "index": 5
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        166.2,
+                                        291,
+                                        179.89999999999998
+                                    ],
+                                    "spans": [],
+                                    "index": 6
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        179.89999999999998,
+                                        291,
+                                        193.59999999999997
+                                    ],
+                                    "spans": [],
+                                    "index": 7
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        193.59999999999997,
+                                        291,
+                                        207.29999999999995
+                                    ],
+                                    "spans": [],
+                                    "index": 8
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        207.29999999999995,
+                                        291,
+                                        220.99999999999994
+                                    ],
+                                    "spans": [],
+                                    "index": 9
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 4.5
+                },
+                {
+                    "type": "list",
+                    "bbox": [
+                        296,
+                        85,
+                        478,
+                        216
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                296,
+                                85,
+                                478,
+                                98
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        296,
+                                        85,
+                                        478,
+                                        98
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Algorithm 2: Estimating target distributions.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                297,
+                                99,
+                                465,
+                                112
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        297,
+                                        99,
+                                        326,
+                                        112
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Input:",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        326,
+                                        100,
+                                        338,
+                                        111
+                                    ],
+                                    "score": 0.75,
+                                    "content": "F _ { \\theta }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        339,
+                                        99,
+                                        418,
+                                        112
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(trained predictor),",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        418,
+                                        100,
+                                        444,
+                                        111
+                                    ],
+                                    "score": 0.89,
+                                    "content": "( x , k )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        444,
+                                        99,
+                                        465,
+                                        112
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(test",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                325,
+                                110,
+                                454,
+                                122
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        325,
+                                        110,
+                                        362,
+                                        122
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sample),",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        362,
+                                        111,
+                                        372,
+                                        121
+                                    ],
+                                    "score": 0.36,
+                                    "content": "\\mathbf { N }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        372,
+                                        110,
+                                        454,
+                                        122
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(ensemble samples)",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                295,
+                                120,
+                                474,
+                                134
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        295,
+                                        120,
+                                        333,
+                                        134
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Output:",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        334,
+                                        122,
+                                        343,
+                                        131
+                                    ],
+                                    "score": 0.8,
+                                    "content": "\\Psi",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        343,
+                                        120,
+                                        474,
+                                        134
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(distribution over target classes)",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                295,
+                                131,
+                                399,
+                                144
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        295,
+                                        131,
+                                        399,
+                                        144
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Ψ ← zeros ∈ R#classes",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14,
+                            "is_list_start_line": true,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                296,
+                                143,
+                                443,
+                                155
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        296,
+                                        143,
+                                        403,
+                                        155
+                                    ],
+                                    "score": 1.0,
+                                    "content": "foreach Ensemble Sample",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        403,
+                                        144,
+                                        410,
+                                        153
+                                    ],
+                                    "score": 0.41,
+                                    "content": "^ { l }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        410,
+                                        143,
+                                        420,
+                                        155
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        421,
+                                        144,
+                                        430,
+                                        153
+                                    ],
+                                    "score": 0.56,
+                                    "content": "N",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        430,
+                                        143,
+                                        443,
+                                        155
+                                    ],
+                                    "score": 1.0,
+                                    "content": "do",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15,
+                            "is_list_start_line": true,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                310,
+                                154,
+                                364,
+                                166
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        310,
+                                        154,
+                                        364,
+                                        166
+                                    ],
+                                    "score": 1.0,
+                                    "content": "z ∼ N (0, I)",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                308,
+                                165,
+                                457,
+                                177
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        308,
+                                        165,
+                                        457,
+                                        177
+                                    ],
+                                    "score": 0.53,
+                                    "content": "\\widehat { \\mathbf { x } }  k \\odot x + ( 1 - k ) \\odot G ( { \\pmb { x } } , { \\pmb { k } } , z )",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 17,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                309,
+                                174,
+                                375,
+                                189
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        309,
+                                        174,
+                                        375,
+                                        189
+                                    ],
+                                    "score": 1.0,
+                                    "content": "bypred ← Fθ(x)",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                308,
+                                185,
+                                399,
+                                202
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        308,
+                                        185,
+                                        399,
+                                        202
+                                    ],
+                                    "score": 1.0,
+                                    "content": "bj ← argmax(ypred)",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                310,
+                                195,
+                                375,
+                                214
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        310,
+                                        195,
+                                        375,
+                                        214
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Ψj ← Ψj + 1N",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20,
+                            "is_list_end_line": true
+                        }
+                    ],
+                    "index": 15,
+                    "bbox_fs": [
+                        295,
+                        85,
+                        478,
+                        214
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        247,
+                        505,
+                        302
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                246,
+                                505,
+                                260
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        246,
+                                        505,
+                                        260
+                                    ],
+                                    "score": 1.0,
+                                    "content": "samples to update the predictor. Note that, on each epoch and for each sample, the generator generates",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                258,
+                                506,
+                                270
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        258,
+                                        506,
+                                        270
+                                    ],
+                                    "score": 1.0,
+                                    "content": "a new sample from the conditional distribution. Intuitively, it means that the predictor observes and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                268,
+                                505,
+                                283
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        268,
+                                        505,
+                                        283
+                                    ],
+                                    "score": 1.0,
+                                    "content": "learns to operate under different imputations for a given sample. This is different from approaches",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                280,
+                                506,
+                                293
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        280,
+                                        506,
+                                        293
+                                    ],
+                                    "score": 1.0,
+                                    "content": "such as multiple imputation where several predictors are trained on different imputed versions of a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                291,
+                                139,
+                                303
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        291,
+                                        139,
+                                        303
+                                    ],
+                                    "score": 1.0,
+                                    "content": "dataset.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        }
+                    ],
+                    "index": 23,
+                    "bbox_fs": [
+                        104,
+                        246,
+                        506,
+                        303
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        307,
+                        505,
+                        364
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                308,
+                                506,
+                                320
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        308,
+                                        506,
+                                        320
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Algorithm 2 presents the suggested algorithm for making predictions and estimating target distribu-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                319,
+                                505,
+                                331
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        319,
+                                        368,
+                                        331
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tions given a trained predictor model. Here, a sample is imputed",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        369,
+                                        320,
+                                        379,
+                                        329
+                                    ],
+                                    "score": 0.8,
+                                    "content": "N",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        379,
+                                        319,
+                                        505,
+                                        331
+                                    ],
+                                    "score": 1.0,
+                                    "content": "times and inference on this set",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                329,
+                                505,
+                                342
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        329,
+                                        505,
+                                        342
+                                    ],
+                                    "score": 1.0,
+                                    "content": "results in an ensemble of predictions over different imputations. The output of this algorithm can be",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                340,
+                                506,
+                                354
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        340,
+                                        506,
+                                        354
+                                    ],
+                                    "score": 1.0,
+                                    "content": "interpreted as a distribution over the confidence of class assignments given a partially observed test",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                351,
+                                433,
+                                365
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        351,
+                                        433,
+                                        365
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sample. The following claims justify the validity of Algorithm 1and Algorithm 2.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        }
+                    ],
+                    "index": 28,
+                    "bbox_fs": [
+                        105,
+                        308,
+                        506,
+                        365
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        368,
+                        503,
+                        402
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                367,
+                                505,
+                                383
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        367,
+                                        363,
+                                        383
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Claim 1. (Generalization of the predictor). If we assume imputed",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        363,
+                                        369,
+                                        377,
+                                        380
+                                    ],
+                                    "score": 0.86,
+                                    "content": "\\widehat { x _ { i } s }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        378,
+                                        367,
+                                        505,
+                                        383
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are samples from the underlying",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                379,
+                                505,
+                                393
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        379,
+                                        505,
+                                        393
+                                    ],
+                                    "score": 1.0,
+                                    "content": "bfeature distribution, then the assigned training set labels can be modeled as labels generated from a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                390,
+                                200,
+                                404
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        390,
+                                        200,
+                                        404
+                                    ],
+                                    "score": 1.0,
+                                    "content": "noisy labeling process.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        }
+                    ],
+                    "index": 32,
+                    "bbox_fs": [
+                        105,
+                        367,
+                        505,
+                        404
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        407,
+                        505,
+                        452
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                406,
+                                505,
+                                421
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        406,
+                                        505,
+                                        421
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Claim 1 permits the analysis of the generalization and convergence for the predictor trained using",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                419,
+                                506,
+                                431
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        419,
+                                        506,
+                                        431
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Algorithm 1 based on current literature in training models with noisy labels (Natarajan et al., 2013;",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                429,
+                                505,
+                                443
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        429,
+                                        505,
+                                        443
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Reed et al., 2014; Chen et al., 2019). From the analysis provided by Chen et al. (2019), test accuracy",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                441,
+                                420,
+                                453
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        441,
+                                        420,
+                                        453
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in asymmetric label noise conditions is a quadratic function of the label noise:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        }
+                    ],
+                    "index": 35.5,
+                    "bbox_fs": [
+                        105,
+                        406,
+                        506,
+                        453
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        246,
+                        456,
+                        364,
+                        471
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                246,
+                                456,
+                                364,
+                                471
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        246,
+                                        456,
+                                        364,
+                                        471
+                                    ],
+                                    "score": 0.92,
+                                    "content": "P ( y _ { i } = \\widehat { y _ { i } } ) = ( 1 - \\epsilon ) ^ { 2 } + \\epsilon ^ { 2 } ,",
+                                    "type": "interline_equation",
+                                    "image_path": "9ada2a757bda1fa063129713d7bfe7a24fe80e1f7c59bea48bebc2158db6572e.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 38,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                246,
+                                456,
+                                364,
+                                471
+                            ],
+                            "spans": [],
+                            "index": 38
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        475,
+                        505,
+                        509
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                474,
+                                505,
+                                488
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        474,
+                                        133,
+                                        488
+                                    ],
+                                    "score": 1.0,
+                                    "content": "where",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        133,
+                                        476,
+                                        142,
+                                        487
+                                    ],
+                                    "score": 0.87,
+                                    "content": "\\widehat { y _ { i } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        142,
+                                        474,
+                                        361,
+                                        488
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is underlying true label for the imputed feature vector",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        361,
+                                        476,
+                                        377,
+                                        487
+                                    ],
+                                    "score": 0.85,
+                                    "content": "( \\widehat { x _ { i } } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        378,
+                                        474,
+                                        398,
+                                        488
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        398,
+                                        477,
+                                        408,
+                                        487
+                                    ],
+                                    "score": 0.85,
+                                    "content": "y _ { i }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        408,
+                                        474,
+                                        505,
+                                        488
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is the label provided by",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                487,
+                                505,
+                                499
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        487,
+                                        300,
+                                        499
+                                    ],
+                                    "score": 1.0,
+                                    "content": "bthe incomplete dataset. In (8), label noise ratio,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        301,
+                                        489,
+                                        306,
+                                        497
+                                    ],
+                                    "score": 0.67,
+                                    "content": "\\epsilon",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        307,
+                                        487,
+                                        505,
+                                        499
+                                    ],
+                                    "score": 1.0,
+                                    "content": "b, represents the probability of the label transition",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                498,
+                                257,
+                                510
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        498,
+                                        257,
+                                        510
+                                    ],
+                                    "score": 1.0,
+                                    "content": "from a certain target class to another:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        }
+                    ],
+                    "index": 40,
+                    "bbox_fs": [
+                        105,
+                        474,
+                        505,
+                        510
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        250,
+                        514,
+                        361,
+                        528
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                250,
+                                514,
+                                361,
+                                528
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        250,
+                                        514,
+                                        361,
+                                        528
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\epsilon = 1 - P ( \\widehat { y _ { i } } = j | y _ { i } = j ) .",
+                                    "type": "interline_equation",
+                                    "image_path": "2d600ebdd741a923b07c08d88e148eb392dd20a835c1cff559210c8791c19559.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 42,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                250,
+                                514,
+                                361,
+                                528
+                            ],
+                            "spans": [],
+                            "index": 42
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        532,
+                        505,
+                        555
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                532,
+                                505,
+                                545
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        532,
+                                        157,
+                                        545
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In practice,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        157,
+                                        535,
+                                        163,
+                                        543
+                                    ],
+                                    "score": 0.7,
+                                    "content": "\\epsilon",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        163,
+                                        532,
+                                        505,
+                                        545
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is determined by the problem-specific underlying data distribution as well as the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                543,
+                                229,
+                                556
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        543,
+                                        229,
+                                        556
+                                    ],
+                                    "score": 1.0,
+                                    "content": "distribution of missing values.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        }
+                    ],
+                    "index": 43.5,
+                    "bbox_fs": [
+                        106,
+                        532,
+                        505,
+                        556
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        560,
+                        506,
+                        644
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                559,
+                                507,
+                                576
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        559,
+                                        330,
+                                        576
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Justification for claim 1 is straightforward, assume that",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        330,
+                                        560,
+                                        384,
+                                        574
+                                    ],
+                                    "score": 0.94,
+                                    "content": "\\{ \\widehat { y _ { i } } ^ { 1 } \\ldots \\widehat { y _ { i } } ^ { N } \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        385,
+                                        559,
+                                        507,
+                                        576
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are underlying true labels for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                102,
+                                568,
+                                510,
+                                593
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        102,
+                                        568,
+                                        138,
+                                        593
+                                    ],
+                                    "score": 1.0,
+                                    "content": "each of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        138,
+                                        573,
+                                        194,
+                                        587
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\{ \\widehat { x _ { i } } ^ { 1 } \\ldots \\widehat { x _ { i } } ^ { N } \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        194,
+                                        568,
+                                        371,
+                                        593
+                                    ],
+                                    "score": 1.0,
+                                    "content": "b b. During training, for any imputed sample in",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        371,
+                                        574,
+                                        426,
+                                        587
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\{ \\widehat { x _ { i } } ^ { 1 } \\ldots \\widehat { x _ { i } } ^ { N } \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        427,
+                                        568,
+                                        510,
+                                        593
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", we use the dataset",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                586,
+                                506,
+                                600
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        586,
+                                        172,
+                                        600
+                                    ],
+                                    "score": 1.0,
+                                    "content": "bprovided label,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        172,
+                                        589,
+                                        178,
+                                        598
+                                    ],
+                                    "score": 0.78,
+                                    "content": "y",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        178,
+                                        586,
+                                        506,
+                                        600
+                                    ],
+                                    "score": 1.0,
+                                    "content": "b b, to calculate the loss and to update model parameters. In the case that any of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 47
+                        },
+                        {
+                            "bbox": [
+                                107,
+                                597,
+                                506,
+                                613
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        597,
+                                        161,
+                                        611
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\{ \\widehat { y _ { i } } ^ { 1 } \\ldots \\widehat { y _ { i } } ^ { N } \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        161,
+                                        597,
+                                        229,
+                                        613
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is different from",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        229,
+                                        601,
+                                        236,
+                                        611
+                                    ],
+                                    "score": 0.76,
+                                    "content": "y",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        236,
+                                        597,
+                                        506,
+                                        613
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", the loss term corresponding to that term would be calculated using",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 48
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                609,
+                                506,
+                                622
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        609,
+                                        506,
+                                        622
+                                    ],
+                                    "score": 1.0,
+                                    "content": "b ba wrong label. Here, if we consider the average impact on gradients for batches of samples rather",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 49
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                619,
+                                506,
+                                635
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        619,
+                                        506,
+                                        635
+                                    ],
+                                    "score": 1.0,
+                                    "content": "than individual cases, the overall impact on training would be very similar to the case of training",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 50
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                631,
+                                183,
+                                644
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        631,
+                                        183,
+                                        644
+                                    ],
+                                    "score": 1.0,
+                                    "content": "using noisy labels.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 51
+                        }
+                    ],
+                    "index": 48,
+                    "bbox_fs": [
+                        102,
+                        559,
+                        510,
+                        644
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        648,
+                        402,
+                        704
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                649,
+                                373,
+                                661
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        649,
+                                        373,
+                                        661
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Claim 2. (Approximation of the target distribution). If we assume:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 52
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                660,
+                                402,
+                                673
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        660,
+                                        152,
+                                        673
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(i) imputed",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        153,
+                                        660,
+                                        167,
+                                        671
+                                    ],
+                                    "score": 0.88,
+                                    "content": "\\widehat { x _ { i } s }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        167,
+                                        660,
+                                        402,
+                                        673
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are valid samples from the underlying feature distribution,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 53
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                671,
+                                357,
+                                683
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        671,
+                                        357,
+                                        683
+                                    ],
+                                    "score": 1.0,
+                                    "content": "b(ii) a good predictor can be trained using the incomplete data,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 54
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                682,
+                                401,
+                                694
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        682,
+                                        401,
+                                        694
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(iii) enough samples are used and the Monte Carlo estimator is unbiased,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 55
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                693,
+                                345,
+                                705
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        693,
+                                        218,
+                                        705
+                                    ],
+                                    "score": 1.0,
+                                    "content": "then the target distribution,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        218,
+                                        694,
+                                        226,
+                                        703
+                                    ],
+                                    "score": 0.77,
+                                    "content": "\\Psi",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        227,
+                                        693,
+                                        345,
+                                        705
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", can be estimated accurately.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 56
+                        }
+                    ],
+                    "index": 54,
+                    "bbox_fs": [
+                        105,
+                        649,
+                        402,
+                        705
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        709,
+                        503,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                708,
+                                505,
+                                723
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        708,
+                                        505,
+                                        723
+                                    ],
+                                    "score": 1.0,
+                                    "content": "This claim supports Algorithm 2 that is suggested to estimate the target distribution given a partially",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 57
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                720,
+                                204,
+                                732
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        720,
+                                        204,
+                                        732
+                                    ],
+                                    "score": 1.0,
+                                    "content": "observed feature vector.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 58
+                        }
+                    ],
+                    "index": 57.5,
+                    "bbox_fs": [
+                        105,
+                        708,
+                        505,
+                        732
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        82,
+                        505,
+                        148
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                83,
+                                505,
+                                94
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        83,
+                                        505,
+                                        94
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The first assumption is consistent with the theoretical analysis of generative adversarial networks",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                93,
+                                505,
+                                106
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        93,
+                                        505,
+                                        106
+                                    ],
+                                    "score": 1.0,
+                                    "content": "that they can converge to the true underlying distribution (Arora et al., 2018; Liu et al., 2017). The",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                105,
+                                505,
+                                117
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        105,
+                                        505,
+                                        117
+                                    ],
+                                    "score": 1.0,
+                                    "content": "second assumption is supported by Claim 1. Regarding the last assumption, each sample requires",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                115,
+                                505,
+                                128
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        115,
+                                        505,
+                                        128
+                                    ],
+                                    "score": 1.0,
+                                    "content": "one forward computation of the generator and predictor networks which, based on the scalability",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                127,
+                                505,
+                                138
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        127,
+                                        505,
+                                        138
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of current network architectures, usually permits thousands of samples to be taken at a reasonable",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                138,
+                                187,
+                                150
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        138,
+                                        187,
+                                        150
+                                    ],
+                                    "score": 1.0,
+                                    "content": "computational cost.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        }
+                    ],
+                    "index": 2.5
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        162,
+                        248,
+                        173
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                162,
+                                250,
+                                174
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        162,
+                                        250,
+                                        174
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3.4 IMPLEMENTATION DETAILS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        }
+                    ],
+                    "index": 6
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        183,
+                        505,
+                        304
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                183,
+                                506,
+                                195
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        183,
+                                        506,
+                                        195
+                                    ],
+                                    "score": 1.0,
+                                    "content": "As we conduct experiments on image and tabular datasets, we use different architectures for each.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                194,
+                                506,
+                                207
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        194,
+                                        506,
+                                        207
+                                    ],
+                                    "score": 1.0,
+                                    "content": "For image datasets, we used a generator and discriminator architectures similar to the ones suggested",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                204,
+                                506,
+                                219
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        204,
+                                        506,
+                                        219
+                                    ],
+                                    "score": 1.0,
+                                    "content": "by Wang et al. (2018). However, we improved these architectures using self-attention layers (Zhang",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                215,
+                                505,
+                                228
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        215,
+                                        505,
+                                        228
+                                    ],
+                                    "score": 1.0,
+                                    "content": "et al., 2018). It should be noted that, while Zhang et al. (2018) suggests using a single self-attention",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                226,
+                                505,
+                                239
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        226,
+                                        505,
+                                        239
+                                    ],
+                                    "score": 1.0,
+                                    "content": "layer in the middle of the network, we observed consistent improvements by inserting multiple",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                238,
+                                505,
+                                250
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        238,
+                                        505,
+                                        250
+                                    ],
+                                    "score": 1.0,
+                                    "content": "self-attention layers before each residual block within the network. Furthermore, as input to the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                249,
+                                505,
+                                261
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        249,
+                                        348,
+                                        261
+                                    ],
+                                    "score": 1.0,
+                                    "content": "generator, we concatenate input image, mask, and a random",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        349,
+                                        250,
+                                        357,
+                                        259
+                                    ],
+                                    "score": 0.75,
+                                    "content": "_ z",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        357,
+                                        249,
+                                        505,
+                                        261
+                                    ],
+                                    "score": 1.0,
+                                    "content": "frame along the channels dimension",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                260,
+                                506,
+                                272
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        260,
+                                        506,
+                                        272
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and use it as input. For tabular datasets, we use a simple 4 layer network consisting of fully-connected",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                271,
+                                506,
+                                283
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        271,
+                                        506,
+                                        283
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and batch-norm layers. Also, the input to the generator is the concatenation of a feature vector, mask",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                281,
+                                506,
+                                295
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        281,
+                                        159,
+                                        295
+                                    ],
+                                    "score": 1.0,
+                                    "content": "vector, and a",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        159,
+                                        284,
+                                        167,
+                                        291
+                                    ],
+                                    "score": 0.64,
+                                    "content": "_ { z }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        167,
+                                        281,
+                                        223,
+                                        295
+                                    ],
+                                    "score": 1.0,
+                                    "content": "vector of size",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        223,
+                                        281,
+                                        231,
+                                        295
+                                    ],
+                                    "score": 0.85,
+                                    "content": "\\frac { 1 } { 8 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        231,
+                                        281,
+                                        465,
+                                        295
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of the input. For all experiments, we use an ensemble size",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        466,
+                                        282,
+                                        478,
+                                        292
+                                    ],
+                                    "score": 0.42,
+                                    "content": "( N )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        478,
+                                        281,
+                                        506,
+                                        295
+                                    ],
+                                    "score": 1.0,
+                                    "content": ") equal",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                293,
+                                137,
+                                304
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        293,
+                                        137,
+                                        304
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to 128.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        }
+                    ],
+                    "index": 12
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        309,
+                        505,
+                        420
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                309,
+                                505,
+                                322
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        309,
+                                        505,
+                                        322
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We used Adam (Kingma & Ba, 2014) for model optimization. Two time-scale update rule",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                320,
+                                506,
+                                333
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        320,
+                                        506,
+                                        333
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(TTUR) (Heusel et al., 2017) was used to balance training the generator and discriminator net-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                330,
+                                507,
+                                345
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        330,
+                                        507,
+                                        345
+                                    ],
+                                    "score": 1.0,
+                                    "content": "works. We explored best TTUR learning-rate settings from the set of {0.001, 0.0005, 0.0001,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                341,
+                                506,
+                                356
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        341,
+                                        249,
+                                        356
+                                    ],
+                                    "score": 1.0,
+                                    "content": "0.00005}. Here, Adam parameters",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        250,
+                                        343,
+                                        261,
+                                        354
+                                    ],
+                                    "score": 0.88,
+                                    "content": "\\beta _ { 1 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        262,
+                                        341,
+                                        280,
+                                        356
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        280,
+                                        343,
+                                        291,
+                                        354
+                                    ],
+                                    "score": 0.88,
+                                    "content": "\\beta _ { 2 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        291,
+                                        341,
+                                        506,
+                                        356
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are set to 0.5 and 0.999, respectively. Also, spectral",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                353,
+                                504,
+                                365
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        353,
+                                        504,
+                                        365
+                                    ],
+                                    "score": 1.0,
+                                    "content": "normalization was used to stabilize both the generator and discriminator network in our experiments",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                365,
+                                505,
+                                377
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        365,
+                                        505,
+                                        377
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with image data (Miyato et al., 2018). For the predictor network, we used the default Adam settings",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                375,
+                                506,
+                                387
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        375,
+                                        506,
+                                        387
+                                    ],
+                                    "score": 1.0,
+                                    "content": "as suggested by Kingma & Ba (2014). In all training procedures, we decay learning rate by a factor of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                387,
+                                506,
+                                399
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        387,
+                                        506,
+                                        399
+                                    ],
+                                    "score": 1.0,
+                                    "content": "5 after reaching a plateau. For all experiments, we use a batch size of 64. Based on our experiments,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                397,
+                                506,
+                                409
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        397,
+                                        464,
+                                        409
+                                    ],
+                                    "score": 1.0,
+                                    "content": "we found that pretraining the discriminator while fixing the generator network for the first",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        464,
+                                        397,
+                                        479,
+                                        407
+                                    ],
+                                    "score": 0.86,
+                                    "content": "5 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        479,
+                                        397,
+                                        506,
+                                        409
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                408,
+                                289,
+                                422
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        408,
+                                        289,
+                                        422
+                                    ],
+                                    "score": 1.0,
+                                    "content": "training epochs helps the stability of training.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        }
+                    ],
+                    "index": 22.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        425,
+                        504,
+                        447
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                424,
+                                505,
+                                438
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        424,
+                                        505,
+                                        438
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Further detail on exact architectures, experiments, software dependencies, etc. as well as ablation",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                437,
+                                257,
+                                448
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        437,
+                                        257,
+                                        448
+                                    ],
+                                    "score": 1.0,
+                                    "content": "studies is provided in the appendices.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        }
+                    ],
+                    "index": 28.5
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        464,
+                        200,
+                        476
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                464,
+                                201,
+                                478
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        464,
+                                        201,
+                                        478
+                                    ],
+                                    "score": 1.0,
+                                    "content": "4 EXPERIMENTS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        }
+                    ],
+                    "index": 30
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        488,
+                        176,
+                        500
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                487,
+                                177,
+                                502
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        487,
+                                        177,
+                                        502
+                                    ],
+                                    "score": 1.0,
+                                    "content": "4.1 DATASETS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        }
+                    ],
+                    "index": 31
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        509,
+                        505,
+                        586
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                508,
+                                506,
+                                523
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        508,
+                                        506,
+                                        523
+                                    ],
+                                    "score": 1.0,
+                                    "content": "To evaluate the proposed method we use CIFAR-10 (Krizhevsky & Hinton, 2009) as an image",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                519,
+                                506,
+                                532
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        519,
+                                        506,
+                                        532
+                                    ],
+                                    "score": 1.0,
+                                    "content": "classification dataset as well as three non-image datasets: UCI Landsat (Dua & Graff, 2017)2,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                530,
+                                507,
+                                544
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        530,
+                                        507,
+                                        544
+                                    ],
+                                    "score": 1.0,
+                                    "content": "MIT-BIH arrhythmia (Moody & Mark, 2001), and Diabetes classification (Kachuee et al., 2019) 3.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                541,
+                                506,
+                                555
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        541,
+                                        227,
+                                        555
+                                    ],
+                                    "score": 1.0,
+                                    "content": "CIFAR-10 dataset consists of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        227,
+                                        542,
+                                        284,
+                                        553
+                                    ],
+                                    "score": 0.43,
+                                    "content": "6 0 , 0 0 0 \\ 3 2 \\mathrm { x } \\ 3 2",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        284,
+                                        541,
+                                        506,
+                                        555
+                                    ],
+                                    "score": 1.0,
+                                    "content": "images from 10 different classes. For this task, we use",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                552,
+                                506,
+                                566
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        552,
+                                        506,
+                                        566
+                                    ],
+                                    "score": 1.0,
+                                    "content": "train and test sets as provided by the dataset. As a preprocessing step, we normalize pixel values to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                564,
+                                506,
+                                577
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        564,
+                                        506,
+                                        577
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the range of [0,1] and subtract the mean image. The only data augmentation we use for this task is to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                575,
+                                289,
+                                587
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        575,
+                                        289,
+                                        587
+                                    ],
+                                    "score": 1.0,
+                                    "content": "randomly flip training images for each batch.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        }
+                    ],
+                    "index": 35
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        591,
+                        505,
+                        691
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                592,
+                                505,
+                                604
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        592,
+                                        505,
+                                        604
+                                    ],
+                                    "score": 1.0,
+                                    "content": "UCI Landsat consists of 6435 samples of 36 features from 6 different categories. We follow the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                603,
+                                506,
+                                615
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        603,
+                                        506,
+                                        615
+                                    ],
+                                    "score": 1.0,
+                                    "content": "same train and test split as provided by the dataset. MIT-BIH dataset consists of annotated heartbeat",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                613,
+                                505,
+                                627
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        613,
+                                        505,
+                                        627
+                                    ],
+                                    "score": 1.0,
+                                    "content": "signals from which we used the preprocessed version available online4 consisting of 92062 samples",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                625,
+                                505,
+                                637
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        625,
+                                        505,
+                                        637
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of 5 different arrhythmia classes. Diabetes dataset is a real-world health dataset of 92,062 samples",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                635,
+                                506,
+                                649
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        635,
+                                        506,
+                                        649
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and 45 features from different categories such as questionnaire, demographics, medical examination,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                646,
+                                506,
+                                660
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        646,
+                                        506,
+                                        660
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and lab results. The objective is to classify between three different diabetes conditions i.e., normal,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                657,
+                                506,
+                                671
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        657,
+                                        506,
+                                        671
+                                    ],
+                                    "score": 1.0,
+                                    "content": "pre-diabetes, and diabetes. As MIT-BIH and Diabetes datasets do not provide explicit train and test",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                668,
+                                506,
+                                682
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        668,
+                                        209,
+                                        682
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sets, we randomly select",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        209,
+                                        669,
+                                        229,
+                                        679
+                                    ],
+                                    "score": 0.87,
+                                    "content": "80 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        229,
+                                        668,
+                                        506,
+                                        682
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of samples as a training set and the rest as a test set. To preprocess",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                681,
+                                505,
+                                691
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        681,
+                                        505,
+                                        691
+                                    ],
+                                    "score": 1.0,
+                                    "content": "our tabular datasets, statistical and unity based normalization are used to balance the variance of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 47
+                        }
+                    ],
+                    "index": 43
+                }
+            ],
+            "page_idx": 5,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        118,
+                        700,
+                        483,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                118,
+                                698,
+                                484,
+                                711
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        118,
+                                        698,
+                                        376,
+                                        711
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2https://archive.ics.uci.edu/ml/datasets/Statlog",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        376,
+                                        701,
+                                        382,
+                                        709
+                                    ],
+                                    "score": 0.42,
+                                    "content": "^ +",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        383,
+                                        698,
+                                        424,
+                                        711
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(Landsat",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        424,
+                                        701,
+                                        430,
+                                        708
+                                    ],
+                                    "score": 0.33,
+                                    "content": "^ +",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        431,
+                                        698,
+                                        484,
+                                        711
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Satellite)",
+                                    "type": "text"
+                                }
+                            ]
+                        },
+                        {
+                            "bbox": [
+                                118,
+                                708,
+                                346,
+                                723
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        118,
+                                        708,
+                                        346,
+                                        723
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3https://github.com/mkachuee/Opportunistic",
+                                    "type": "text"
+                                }
+                            ]
+                        },
+                        {
+                            "bbox": [
+                                117,
+                                719,
+                                367,
+                                734
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        117,
+                                        719,
+                                        367,
+                                        734
+                                    ],
+                                    "score": 1.0,
+                                    "content": "4https://www.kaggle.com/shayanfazeli/heartbeat",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        107,
+                        27,
+                        308,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                25,
+                                308,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        25,
+                                        308,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Under review as a conference paper at ICLR 2020",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        302,
+                        752,
+                        308,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                302,
+                                751,
+                                309,
+                                762
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        302,
+                                        751,
+                                        309,
+                                        762
+                                    ],
+                                    "score": 1.0,
+                                    "content": "6",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        82,
+                        505,
+                        148
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                83,
+                                505,
+                                94
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        83,
+                                        505,
+                                        94
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The first assumption is consistent with the theoretical analysis of generative adversarial networks",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                93,
+                                505,
+                                106
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        93,
+                                        505,
+                                        106
+                                    ],
+                                    "score": 1.0,
+                                    "content": "that they can converge to the true underlying distribution (Arora et al., 2018; Liu et al., 2017). The",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                105,
+                                505,
+                                117
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        105,
+                                        505,
+                                        117
+                                    ],
+                                    "score": 1.0,
+                                    "content": "second assumption is supported by Claim 1. Regarding the last assumption, each sample requires",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                115,
+                                505,
+                                128
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        115,
+                                        505,
+                                        128
+                                    ],
+                                    "score": 1.0,
+                                    "content": "one forward computation of the generator and predictor networks which, based on the scalability",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                127,
+                                505,
+                                138
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        127,
+                                        505,
+                                        138
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of current network architectures, usually permits thousands of samples to be taken at a reasonable",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                138,
+                                187,
+                                150
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        138,
+                                        187,
+                                        150
+                                    ],
+                                    "score": 1.0,
+                                    "content": "computational cost.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        }
+                    ],
+                    "index": 2.5,
+                    "bbox_fs": [
+                        105,
+                        83,
+                        505,
+                        150
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        162,
+                        248,
+                        173
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                162,
+                                250,
+                                174
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        162,
+                                        250,
+                                        174
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3.4 IMPLEMENTATION DETAILS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        }
+                    ],
+                    "index": 6
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        183,
+                        505,
+                        304
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                183,
+                                506,
+                                195
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        183,
+                                        506,
+                                        195
+                                    ],
+                                    "score": 1.0,
+                                    "content": "As we conduct experiments on image and tabular datasets, we use different architectures for each.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                194,
+                                506,
+                                207
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        194,
+                                        506,
+                                        207
+                                    ],
+                                    "score": 1.0,
+                                    "content": "For image datasets, we used a generator and discriminator architectures similar to the ones suggested",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                204,
+                                506,
+                                219
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        204,
+                                        506,
+                                        219
+                                    ],
+                                    "score": 1.0,
+                                    "content": "by Wang et al. (2018). However, we improved these architectures using self-attention layers (Zhang",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                215,
+                                505,
+                                228
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        215,
+                                        505,
+                                        228
+                                    ],
+                                    "score": 1.0,
+                                    "content": "et al., 2018). It should be noted that, while Zhang et al. (2018) suggests using a single self-attention",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                226,
+                                505,
+                                239
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        226,
+                                        505,
+                                        239
+                                    ],
+                                    "score": 1.0,
+                                    "content": "layer in the middle of the network, we observed consistent improvements by inserting multiple",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                238,
+                                505,
+                                250
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        238,
+                                        505,
+                                        250
+                                    ],
+                                    "score": 1.0,
+                                    "content": "self-attention layers before each residual block within the network. Furthermore, as input to the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                249,
+                                505,
+                                261
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        249,
+                                        348,
+                                        261
+                                    ],
+                                    "score": 1.0,
+                                    "content": "generator, we concatenate input image, mask, and a random",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        349,
+                                        250,
+                                        357,
+                                        259
+                                    ],
+                                    "score": 0.75,
+                                    "content": "_ z",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        357,
+                                        249,
+                                        505,
+                                        261
+                                    ],
+                                    "score": 1.0,
+                                    "content": "frame along the channels dimension",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                260,
+                                506,
+                                272
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        260,
+                                        506,
+                                        272
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and use it as input. For tabular datasets, we use a simple 4 layer network consisting of fully-connected",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                271,
+                                506,
+                                283
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        271,
+                                        506,
+                                        283
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and batch-norm layers. Also, the input to the generator is the concatenation of a feature vector, mask",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                281,
+                                506,
+                                295
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        281,
+                                        159,
+                                        295
+                                    ],
+                                    "score": 1.0,
+                                    "content": "vector, and a",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        159,
+                                        284,
+                                        167,
+                                        291
+                                    ],
+                                    "score": 0.64,
+                                    "content": "_ { z }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        167,
+                                        281,
+                                        223,
+                                        295
+                                    ],
+                                    "score": 1.0,
+                                    "content": "vector of size",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        223,
+                                        281,
+                                        231,
+                                        295
+                                    ],
+                                    "score": 0.85,
+                                    "content": "\\frac { 1 } { 8 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        231,
+                                        281,
+                                        465,
+                                        295
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of the input. For all experiments, we use an ensemble size",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        466,
+                                        282,
+                                        478,
+                                        292
+                                    ],
+                                    "score": 0.42,
+                                    "content": "( N )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        478,
+                                        281,
+                                        506,
+                                        295
+                                    ],
+                                    "score": 1.0,
+                                    "content": ") equal",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                293,
+                                137,
+                                304
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        293,
+                                        137,
+                                        304
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to 128.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        }
+                    ],
+                    "index": 12,
+                    "bbox_fs": [
+                        104,
+                        183,
+                        506,
+                        304
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        309,
+                        505,
+                        420
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                309,
+                                505,
+                                322
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        309,
+                                        505,
+                                        322
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We used Adam (Kingma & Ba, 2014) for model optimization. Two time-scale update rule",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                320,
+                                506,
+                                333
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        320,
+                                        506,
+                                        333
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(TTUR) (Heusel et al., 2017) was used to balance training the generator and discriminator net-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                330,
+                                507,
+                                345
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        330,
+                                        507,
+                                        345
+                                    ],
+                                    "score": 1.0,
+                                    "content": "works. We explored best TTUR learning-rate settings from the set of {0.001, 0.0005, 0.0001,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                341,
+                                506,
+                                356
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        341,
+                                        249,
+                                        356
+                                    ],
+                                    "score": 1.0,
+                                    "content": "0.00005}. Here, Adam parameters",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        250,
+                                        343,
+                                        261,
+                                        354
+                                    ],
+                                    "score": 0.88,
+                                    "content": "\\beta _ { 1 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        262,
+                                        341,
+                                        280,
+                                        356
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        280,
+                                        343,
+                                        291,
+                                        354
+                                    ],
+                                    "score": 0.88,
+                                    "content": "\\beta _ { 2 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        291,
+                                        341,
+                                        506,
+                                        356
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are set to 0.5 and 0.999, respectively. Also, spectral",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                353,
+                                504,
+                                365
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        353,
+                                        504,
+                                        365
+                                    ],
+                                    "score": 1.0,
+                                    "content": "normalization was used to stabilize both the generator and discriminator network in our experiments",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                365,
+                                505,
+                                377
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        365,
+                                        505,
+                                        377
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with image data (Miyato et al., 2018). For the predictor network, we used the default Adam settings",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                375,
+                                506,
+                                387
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        375,
+                                        506,
+                                        387
+                                    ],
+                                    "score": 1.0,
+                                    "content": "as suggested by Kingma & Ba (2014). In all training procedures, we decay learning rate by a factor of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                387,
+                                506,
+                                399
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        387,
+                                        506,
+                                        399
+                                    ],
+                                    "score": 1.0,
+                                    "content": "5 after reaching a plateau. For all experiments, we use a batch size of 64. Based on our experiments,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                397,
+                                506,
+                                409
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        397,
+                                        464,
+                                        409
+                                    ],
+                                    "score": 1.0,
+                                    "content": "we found that pretraining the discriminator while fixing the generator network for the first",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        464,
+                                        397,
+                                        479,
+                                        407
+                                    ],
+                                    "score": 0.86,
+                                    "content": "5 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        479,
+                                        397,
+                                        506,
+                                        409
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                408,
+                                289,
+                                422
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        408,
+                                        289,
+                                        422
+                                    ],
+                                    "score": 1.0,
+                                    "content": "training epochs helps the stability of training.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        }
+                    ],
+                    "index": 22.5,
+                    "bbox_fs": [
+                        104,
+                        309,
+                        507,
+                        422
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        425,
+                        504,
+                        447
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                424,
+                                505,
+                                438
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        424,
+                                        505,
+                                        438
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Further detail on exact architectures, experiments, software dependencies, etc. as well as ablation",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                437,
+                                257,
+                                448
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        437,
+                                        257,
+                                        448
+                                    ],
+                                    "score": 1.0,
+                                    "content": "studies is provided in the appendices.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        }
+                    ],
+                    "index": 28.5,
+                    "bbox_fs": [
+                        105,
+                        424,
+                        505,
+                        448
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        464,
+                        200,
+                        476
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                464,
+                                201,
+                                478
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        464,
+                                        201,
+                                        478
+                                    ],
+                                    "score": 1.0,
+                                    "content": "4 EXPERIMENTS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        }
+                    ],
+                    "index": 30
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        488,
+                        176,
+                        500
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                487,
+                                177,
+                                502
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        487,
+                                        177,
+                                        502
+                                    ],
+                                    "score": 1.0,
+                                    "content": "4.1 DATASETS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        }
+                    ],
+                    "index": 31
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        509,
+                        505,
+                        586
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                508,
+                                506,
+                                523
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        508,
+                                        506,
+                                        523
+                                    ],
+                                    "score": 1.0,
+                                    "content": "To evaluate the proposed method we use CIFAR-10 (Krizhevsky & Hinton, 2009) as an image",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                519,
+                                506,
+                                532
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        519,
+                                        506,
+                                        532
+                                    ],
+                                    "score": 1.0,
+                                    "content": "classification dataset as well as three non-image datasets: UCI Landsat (Dua & Graff, 2017)2,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                530,
+                                507,
+                                544
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        530,
+                                        507,
+                                        544
+                                    ],
+                                    "score": 1.0,
+                                    "content": "MIT-BIH arrhythmia (Moody & Mark, 2001), and Diabetes classification (Kachuee et al., 2019) 3.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                541,
+                                506,
+                                555
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        541,
+                                        227,
+                                        555
+                                    ],
+                                    "score": 1.0,
+                                    "content": "CIFAR-10 dataset consists of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        227,
+                                        542,
+                                        284,
+                                        553
+                                    ],
+                                    "score": 0.43,
+                                    "content": "6 0 , 0 0 0 \\ 3 2 \\mathrm { x } \\ 3 2",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        284,
+                                        541,
+                                        506,
+                                        555
+                                    ],
+                                    "score": 1.0,
+                                    "content": "images from 10 different classes. For this task, we use",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                552,
+                                506,
+                                566
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        552,
+                                        506,
+                                        566
+                                    ],
+                                    "score": 1.0,
+                                    "content": "train and test sets as provided by the dataset. As a preprocessing step, we normalize pixel values to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                564,
+                                506,
+                                577
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        564,
+                                        506,
+                                        577
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the range of [0,1] and subtract the mean image. The only data augmentation we use for this task is to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                575,
+                                289,
+                                587
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        575,
+                                        289,
+                                        587
+                                    ],
+                                    "score": 1.0,
+                                    "content": "randomly flip training images for each batch.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        }
+                    ],
+                    "index": 35,
+                    "bbox_fs": [
+                        104,
+                        508,
+                        507,
+                        587
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        591,
+                        505,
+                        691
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                592,
+                                505,
+                                604
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        592,
+                                        505,
+                                        604
+                                    ],
+                                    "score": 1.0,
+                                    "content": "UCI Landsat consists of 6435 samples of 36 features from 6 different categories. We follow the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                603,
+                                506,
+                                615
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        603,
+                                        506,
+                                        615
+                                    ],
+                                    "score": 1.0,
+                                    "content": "same train and test split as provided by the dataset. MIT-BIH dataset consists of annotated heartbeat",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                613,
+                                505,
+                                627
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        613,
+                                        505,
+                                        627
+                                    ],
+                                    "score": 1.0,
+                                    "content": "signals from which we used the preprocessed version available online4 consisting of 92062 samples",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                625,
+                                505,
+                                637
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        625,
+                                        505,
+                                        637
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of 5 different arrhythmia classes. Diabetes dataset is a real-world health dataset of 92,062 samples",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                635,
+                                506,
+                                649
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        635,
+                                        506,
+                                        649
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and 45 features from different categories such as questionnaire, demographics, medical examination,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                646,
+                                506,
+                                660
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        646,
+                                        506,
+                                        660
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and lab results. The objective is to classify between three different diabetes conditions i.e., normal,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                657,
+                                506,
+                                671
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        657,
+                                        506,
+                                        671
+                                    ],
+                                    "score": 1.0,
+                                    "content": "pre-diabetes, and diabetes. As MIT-BIH and Diabetes datasets do not provide explicit train and test",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                668,
+                                506,
+                                682
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        668,
+                                        209,
+                                        682
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sets, we randomly select",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        209,
+                                        669,
+                                        229,
+                                        679
+                                    ],
+                                    "score": 0.87,
+                                    "content": "80 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        229,
+                                        668,
+                                        506,
+                                        682
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of samples as a training set and the rest as a test set. To preprocess",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                681,
+                                505,
+                                691
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        681,
+                                        505,
+                                        691
+                                    ],
+                                    "score": 1.0,
+                                    "content": "our tabular datasets, statistical and unity based normalization are used to balance the variance of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 47
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                82,
+                                505,
+                                95
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        82,
+                                        505,
+                                        95
+                                    ],
+                                    "score": 1.0,
+                                    "content": "different features and center them around zero. Also, while different encoding and representation",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                94,
+                                506,
+                                106
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        94,
+                                        506,
+                                        106
+                                    ],
+                                    "score": 1.0,
+                                    "content": "methods are suggested in the literature to handle categorical features (Jang et al., 2016; Nazabal et al.,",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                104,
+                                506,
+                                117
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        104,
+                                        506,
+                                        117
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2018), in this paper, we take the simple approach of encoding categorical variables using one-hot",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                115,
+                                506,
+                                128
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        115,
+                                        506,
+                                        128
+                                    ],
+                                    "score": 1.0,
+                                    "content": "representation and smoothing them by adding Gaussian noise with zero mean and variance equal",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                127,
+                                505,
+                                139
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        127,
+                                        117,
+                                        139
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to",
+                                    "type": "text",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        117,
+                                        127,
+                                        132,
+                                        137
+                                    ],
+                                    "score": 0.86,
+                                    "content": "5 \\%",
+                                    "type": "inline_equation",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        132,
+                                        127,
+                                        505,
+                                        139
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of feature variances. In our experiments, we observed a reasonable performance using the",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                138,
+                                505,
+                                149
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        138,
+                                        505,
+                                        149
+                                    ],
+                                    "score": 1.0,
+                                    "content": "suggested simple smoothing; however, more advanced encoding methods are also applicable in this",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                149,
+                                376,
+                                161
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        149,
+                                        376,
+                                        161
+                                    ],
+                                    "score": 1.0,
+                                    "content": "setup and can be applied to enhance the performances even further.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 6
+                        }
+                    ],
+                    "index": 43,
+                    "bbox_fs": [
+                        105,
+                        592,
+                        506,
+                        691
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        82,
+                        505,
+                        160
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                82,
+                                505,
+                                95
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        82,
+                                        505,
+                                        95
+                                    ],
+                                    "score": 1.0,
+                                    "content": "different features and center them around zero. Also, while different encoding and representation",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                94,
+                                506,
+                                106
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        94,
+                                        506,
+                                        106
+                                    ],
+                                    "score": 1.0,
+                                    "content": "methods are suggested in the literature to handle categorical features (Jang et al., 2016; Nazabal et al.,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                104,
+                                506,
+                                117
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        104,
+                                        506,
+                                        117
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2018), in this paper, we take the simple approach of encoding categorical variables using one-hot",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                115,
+                                506,
+                                128
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        115,
+                                        506,
+                                        128
+                                    ],
+                                    "score": 1.0,
+                                    "content": "representation and smoothing them by adding Gaussian noise with zero mean and variance equal",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                127,
+                                505,
+                                139
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        127,
+                                        117,
+                                        139
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        117,
+                                        127,
+                                        132,
+                                        137
+                                    ],
+                                    "score": 0.86,
+                                    "content": "5 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        132,
+                                        127,
+                                        505,
+                                        139
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of feature variances. In our experiments, we observed a reasonable performance using the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                138,
+                                505,
+                                149
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        138,
+                                        505,
+                                        149
+                                    ],
+                                    "score": 1.0,
+                                    "content": "suggested simple smoothing; however, more advanced encoding methods are also applicable in this",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                149,
+                                376,
+                                161
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        149,
+                                        376,
+                                        161
+                                    ],
+                                    "score": 1.0,
+                                    "content": "setup and can be applied to enhance the performances even further.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        }
+                    ],
+                    "index": 3
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        200,
+                        253,
+                        211
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                199,
+                                254,
+                                212
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        199,
+                                        254,
+                                        212
+                                    ],
+                                    "score": 1.0,
+                                    "content": "4.2 MISSINGNESS MECHANISMS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        }
+                    ],
+                    "index": 7
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        231,
+                        505,
+                        341
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                231,
+                                507,
+                                244
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        231,
+                                        507,
+                                        244
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In our experiments, we consider MCAR uniform and MCAR rectangular missingness structures.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                243,
+                                505,
+                                254
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        243,
+                                        505,
+                                        254
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In MCAR uniform, each feature of each sample is missing based on a Bernoulli distribution with",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                253,
+                                505,
+                                265
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        253,
+                                        505,
+                                        265
+                                    ],
+                                    "score": 1.0,
+                                    "content": "a certain missingness probability (i.e., missing rate) independent of other features. In addition",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                264,
+                                506,
+                                277
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        264,
+                                        506,
+                                        277
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to the case of uniform missingness, for image tasks, we use rectangular missingness/observation",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                275,
+                                506,
+                                287
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        275,
+                                        506,
+                                        287
+                                    ],
+                                    "score": 1.0,
+                                    "content": "structure where rectangular regions of dataset images are missing/observed. To control the rate of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                286,
+                                506,
+                                299
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        286,
+                                        506,
+                                        299
+                                    ],
+                                    "score": 1.0,
+                                    "content": "missingness and decide on the regions that are missing for each case, we use a latent beta distribution",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                297,
+                                506,
+                                310
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        297,
+                                        506,
+                                        310
+                                    ],
+                                    "score": 1.0,
+                                    "content": "that samples rectangular region’s width and height such that the average missing rate is maintained.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                308,
+                                505,
+                                321
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        308,
+                                        212,
+                                        321
+                                    ],
+                                    "score": 1.0,
+                                    "content": "For missing rates less than",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        212,
+                                        308,
+                                        232,
+                                        318
+                                    ],
+                                    "score": 0.86,
+                                    "content": "50 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        232,
+                                        308,
+                                        505,
+                                        321
+                                    ],
+                                    "score": 1.0,
+                                    "content": "we make the assumption of having a random rectangular region to be",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                318,
+                                505,
+                                332
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        318,
+                                        287,
+                                        332
+                                    ],
+                                    "score": 1.0,
+                                    "content": "missing, whereas for missing rates more than",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        288,
+                                        319,
+                                        307,
+                                        330
+                                    ],
+                                    "score": 0.86,
+                                    "content": "50 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        307,
+                                        318,
+                                        505,
+                                        332
+                                    ],
+                                    "score": 1.0,
+                                    "content": "we assume that only a random rectangular region",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                329,
+                                300,
+                                344
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        329,
+                                        300,
+                                        344
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is observed and the rest of the image is missing.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        }
+                    ],
+                    "index": 12.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        347,
+                        505,
+                        413
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                346,
+                                505,
+                                359
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        346,
+                                        505,
+                                        359
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We would like to note that while the suggested solution in this paper is readily compatible with MAR",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                358,
+                                505,
+                                369
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        358,
+                                        505,
+                                        369
+                                    ],
+                                    "score": 1.0,
+                                    "content": "structures, in our experiments, to simplify the presentation of results and to have a fair comparison",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                369,
+                                505,
+                                381
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        369,
+                                        505,
+                                        381
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with other work that does not support the MAR assumption, we limited the scope of our experiments",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                379,
+                                505,
+                                392
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        379,
+                                        505,
+                                        392
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to MCAR. Furthermore, to simulate incomplete datasets and to make sure the same features are",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                391,
+                                505,
+                                402
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        391,
+                                        505,
+                                        402
+                                    ],
+                                    "score": 1.0,
+                                    "content": "missing without explicitly storing masks, we use hashed feature vectors to seed random number",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                402,
+                                441,
+                                414
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        402,
+                                        441,
+                                        414
+                                    ],
+                                    "score": 1.0,
+                                    "content": "generators used to sample missing features. More detail is provided in Appendix C.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        }
+                    ],
+                    "index": 20.5
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        453,
+                        238,
+                        464
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                453,
+                                239,
+                                466
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        453,
+                                        239,
+                                        466
+                                    ],
+                                    "score": 1.0,
+                                    "content": "4.3 EVALUATION MEASURES",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        }
+                    ],
+                    "index": 24
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        484,
+                        505,
+                        594
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                483,
+                                506,
+                                498
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        483,
+                                        506,
+                                        498
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Fréchet inception distance (FID) (Heusel et al., 2017) score is used to measure the quality of missing",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                495,
+                                506,
+                                508
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        495,
+                                        506,
+                                        508
+                                    ],
+                                    "score": 1.0,
+                                    "content": "data imputation in experiments with images5. We also considered using root means squared error",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                504,
+                                506,
+                                520
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        504,
+                                        506,
+                                        520
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(RMSE); however, we decided not to use this measure as we observed an inconsistent behavior using",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                518,
+                                505,
+                                529
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        518,
+                                        505,
+                                        529
+                                    ],
+                                    "score": 1.0,
+                                    "content": "RMSE in our comparisons as RMSE favors methods that show less variance rather than realistic",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                528,
+                                505,
+                                541
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        528,
+                                        505,
+                                        541
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and sharp samples from the distribution. Also, for each dataset and each missingness scenario, we",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                540,
+                                506,
+                                552
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        540,
+                                        506,
+                                        552
+                                    ],
+                                    "score": 1.0,
+                                    "content": "report top-1 classification accuracy based on the majority vote estimated using Algorithm 2. Another",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                551,
+                                505,
+                                562
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        551,
+                                        505,
+                                        562
+                                    ],
+                                    "score": 1.0,
+                                    "content": "measure that we use in this paper is the comparison between the estimated target certainties and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                561,
+                                507,
+                                574
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        561,
+                                        507,
+                                        574
+                                    ],
+                                    "score": 1.0,
+                                    "content": "average accuracies achieved for each confidence assignment. We run each experiment multiple times:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                572,
+                                506,
+                                585
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        572,
+                                        506,
+                                        585
+                                    ],
+                                    "score": 1.0,
+                                    "content": "4 times for CIFAR-10 and 8 times for tabular datasets. We report the mean and standard deviation of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                583,
+                                192,
+                                595
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        583,
+                                        192,
+                                        595
+                                    ],
+                                    "score": 1.0,
+                                    "content": "results for each case.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        }
+                    ],
+                    "index": 29.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        600,
+                        506,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                600,
+                                506,
+                                612
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        600,
+                                        506,
+                                        612
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We compare our results with MisGAN (Li et al., 2019) and GAIN (Yoon et al., 2018) as the state of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                611,
+                                506,
+                                623
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        611,
+                                        506,
+                                        623
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the art imputation algorithms based on GANs as well as basic denoising autoencoder (DAE) (Vincent",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                622,
+                                506,
+                                634
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        622,
+                                        506,
+                                        634
+                                    ],
+                                    "score": 1.0,
+                                    "content": "et al., 2008) and multiple imputation by chained equations (MICE) (Buuren & Groothuis-Oudshoorn,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                632,
+                                506,
+                                646
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        632,
+                                        506,
+                                        646
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2010) as baselines. For experiments using MisGAN, we used the same architectures and hyper-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                643,
+                                506,
+                                657
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        643,
+                                        506,
+                                        657
+                                    ],
+                                    "score": 1.0,
+                                    "content": "parameters as suggested by the MisGAN authors6. The only modification was to adapt the last",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                655,
+                                506,
+                                667
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        655,
+                                        506,
+                                        667
+                                    ],
+                                    "score": 1.0,
+                                    "content": "generator layer to generate images with resolutions as we use. Regarding GAIN, we used the same",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                666,
+                                506,
+                                678
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        666,
+                                        506,
+                                        678
+                                    ],
+                                    "score": 1.0,
+                                    "content": "network architecture as our implementation of GI and hyper-parameters as used by the GAIN authors7.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                676,
+                                506,
+                                690
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        676,
+                                        506,
+                                        690
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In the DAE implementation, due to the incomplete data assumption, only observed features appear in",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                688,
+                                505,
+                                700
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        688,
+                                        505,
+                                        700
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the loss function, ignoring reconstruction terms corresponding to missing features. Due to scalability",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                699,
+                                506,
+                                711
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        699,
+                                        506,
+                                        711
+                                    ],
+                                    "score": 1.0,
+                                    "content": "issues, we were only able to use MICE for the smaller non-image datasets. For these methods, to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                709,
+                                506,
+                                723
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        709,
+                                        506,
+                                        723
+                                    ],
+                                    "score": 1.0,
+                                    "content": "train and evaluate classifiers, we use predictors trained on imputed datasets rather than the stochastic",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                721,
+                                252,
+                                733
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        721,
+                                        252,
+                                        733
+                                    ],
+                                    "score": 1.0,
+                                    "content": "predictor suggested in Algorithm 1.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        }
+                    ],
+                    "index": 40.5
+                }
+            ],
+            "page_idx": 6,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        107,
+                        27,
+                        308,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                25,
+                                308,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        25,
+                                        308,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Under review as a conference paper at ICLR 2020",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        303,
+                        751,
+                        308,
+                        759
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                302,
+                                750,
+                                309,
+                                763
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        302,
+                                        750,
+                                        309,
+                                        763
+                                    ],
+                                    "score": 1.0,
+                                    "content": "",
+                                    "type": "text",
+                                    "height": 13,
+                                    "width": 7
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        82,
+                        505,
+                        160
+                    ],
+                    "lines": [],
+                    "index": 3,
+                    "bbox_fs": [
+                        105,
+                        82,
+                        506,
+                        161
+                    ],
+                    "lines_deleted": true
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        200,
+                        253,
+                        211
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                199,
+                                254,
+                                212
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        199,
+                                        254,
+                                        212
+                                    ],
+                                    "score": 1.0,
+                                    "content": "4.2 MISSINGNESS MECHANISMS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        }
+                    ],
+                    "index": 7
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        231,
+                        505,
+                        341
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                231,
+                                507,
+                                244
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        231,
+                                        507,
+                                        244
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In our experiments, we consider MCAR uniform and MCAR rectangular missingness structures.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                243,
+                                505,
+                                254
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        243,
+                                        505,
+                                        254
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In MCAR uniform, each feature of each sample is missing based on a Bernoulli distribution with",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                253,
+                                505,
+                                265
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        253,
+                                        505,
+                                        265
+                                    ],
+                                    "score": 1.0,
+                                    "content": "a certain missingness probability (i.e., missing rate) independent of other features. In addition",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                264,
+                                506,
+                                277
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        264,
+                                        506,
+                                        277
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to the case of uniform missingness, for image tasks, we use rectangular missingness/observation",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                275,
+                                506,
+                                287
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        275,
+                                        506,
+                                        287
+                                    ],
+                                    "score": 1.0,
+                                    "content": "structure where rectangular regions of dataset images are missing/observed. To control the rate of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                286,
+                                506,
+                                299
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        286,
+                                        506,
+                                        299
+                                    ],
+                                    "score": 1.0,
+                                    "content": "missingness and decide on the regions that are missing for each case, we use a latent beta distribution",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                297,
+                                506,
+                                310
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        297,
+                                        506,
+                                        310
+                                    ],
+                                    "score": 1.0,
+                                    "content": "that samples rectangular region’s width and height such that the average missing rate is maintained.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                308,
+                                505,
+                                321
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        308,
+                                        212,
+                                        321
+                                    ],
+                                    "score": 1.0,
+                                    "content": "For missing rates less than",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        212,
+                                        308,
+                                        232,
+                                        318
+                                    ],
+                                    "score": 0.86,
+                                    "content": "50 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        232,
+                                        308,
+                                        505,
+                                        321
+                                    ],
+                                    "score": 1.0,
+                                    "content": "we make the assumption of having a random rectangular region to be",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                318,
+                                505,
+                                332
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        318,
+                                        287,
+                                        332
+                                    ],
+                                    "score": 1.0,
+                                    "content": "missing, whereas for missing rates more than",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        288,
+                                        319,
+                                        307,
+                                        330
+                                    ],
+                                    "score": 0.86,
+                                    "content": "50 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        307,
+                                        318,
+                                        505,
+                                        332
+                                    ],
+                                    "score": 1.0,
+                                    "content": "we assume that only a random rectangular region",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                329,
+                                300,
+                                344
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        329,
+                                        300,
+                                        344
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is observed and the rest of the image is missing.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        }
+                    ],
+                    "index": 12.5,
+                    "bbox_fs": [
+                        105,
+                        231,
+                        507,
+                        344
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        347,
+                        505,
+                        413
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                346,
+                                505,
+                                359
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        346,
+                                        505,
+                                        359
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We would like to note that while the suggested solution in this paper is readily compatible with MAR",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                358,
+                                505,
+                                369
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        358,
+                                        505,
+                                        369
+                                    ],
+                                    "score": 1.0,
+                                    "content": "structures, in our experiments, to simplify the presentation of results and to have a fair comparison",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                369,
+                                505,
+                                381
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        369,
+                                        505,
+                                        381
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with other work that does not support the MAR assumption, we limited the scope of our experiments",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                379,
+                                505,
+                                392
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        379,
+                                        505,
+                                        392
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to MCAR. Furthermore, to simulate incomplete datasets and to make sure the same features are",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                391,
+                                505,
+                                402
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        391,
+                                        505,
+                                        402
+                                    ],
+                                    "score": 1.0,
+                                    "content": "missing without explicitly storing masks, we use hashed feature vectors to seed random number",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                402,
+                                441,
+                                414
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        402,
+                                        441,
+                                        414
+                                    ],
+                                    "score": 1.0,
+                                    "content": "generators used to sample missing features. More detail is provided in Appendix C.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        }
+                    ],
+                    "index": 20.5,
+                    "bbox_fs": [
+                        105,
+                        346,
+                        505,
+                        414
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        453,
+                        238,
+                        464
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                453,
+                                239,
+                                466
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        453,
+                                        239,
+                                        466
+                                    ],
+                                    "score": 1.0,
+                                    "content": "4.3 EVALUATION MEASURES",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        }
+                    ],
+                    "index": 24
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        484,
+                        505,
+                        594
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                483,
+                                506,
+                                498
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        483,
+                                        506,
+                                        498
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Fréchet inception distance (FID) (Heusel et al., 2017) score is used to measure the quality of missing",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                495,
+                                506,
+                                508
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        495,
+                                        506,
+                                        508
+                                    ],
+                                    "score": 1.0,
+                                    "content": "data imputation in experiments with images5. We also considered using root means squared error",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                504,
+                                506,
+                                520
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        504,
+                                        506,
+                                        520
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(RMSE); however, we decided not to use this measure as we observed an inconsistent behavior using",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                518,
+                                505,
+                                529
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        518,
+                                        505,
+                                        529
+                                    ],
+                                    "score": 1.0,
+                                    "content": "RMSE in our comparisons as RMSE favors methods that show less variance rather than realistic",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                528,
+                                505,
+                                541
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        528,
+                                        505,
+                                        541
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and sharp samples from the distribution. Also, for each dataset and each missingness scenario, we",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                540,
+                                506,
+                                552
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        540,
+                                        506,
+                                        552
+                                    ],
+                                    "score": 1.0,
+                                    "content": "report top-1 classification accuracy based on the majority vote estimated using Algorithm 2. Another",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                551,
+                                505,
+                                562
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        551,
+                                        505,
+                                        562
+                                    ],
+                                    "score": 1.0,
+                                    "content": "measure that we use in this paper is the comparison between the estimated target certainties and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                561,
+                                507,
+                                574
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        561,
+                                        507,
+                                        574
+                                    ],
+                                    "score": 1.0,
+                                    "content": "average accuracies achieved for each confidence assignment. We run each experiment multiple times:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                572,
+                                506,
+                                585
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        572,
+                                        506,
+                                        585
+                                    ],
+                                    "score": 1.0,
+                                    "content": "4 times for CIFAR-10 and 8 times for tabular datasets. We report the mean and standard deviation of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                583,
+                                192,
+                                595
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        583,
+                                        192,
+                                        595
+                                    ],
+                                    "score": 1.0,
+                                    "content": "results for each case.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        }
+                    ],
+                    "index": 29.5,
+                    "bbox_fs": [
+                        104,
+                        483,
+                        507,
+                        595
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        600,
+                        506,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                600,
+                                506,
+                                612
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        600,
+                                        506,
+                                        612
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We compare our results with MisGAN (Li et al., 2019) and GAIN (Yoon et al., 2018) as the state of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                611,
+                                506,
+                                623
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        611,
+                                        506,
+                                        623
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the art imputation algorithms based on GANs as well as basic denoising autoencoder (DAE) (Vincent",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                622,
+                                506,
+                                634
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        622,
+                                        506,
+                                        634
+                                    ],
+                                    "score": 1.0,
+                                    "content": "et al., 2008) and multiple imputation by chained equations (MICE) (Buuren & Groothuis-Oudshoorn,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                632,
+                                506,
+                                646
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        632,
+                                        506,
+                                        646
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2010) as baselines. For experiments using MisGAN, we used the same architectures and hyper-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                643,
+                                506,
+                                657
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        643,
+                                        506,
+                                        657
+                                    ],
+                                    "score": 1.0,
+                                    "content": "parameters as suggested by the MisGAN authors6. The only modification was to adapt the last",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                655,
+                                506,
+                                667
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        655,
+                                        506,
+                                        667
+                                    ],
+                                    "score": 1.0,
+                                    "content": "generator layer to generate images with resolutions as we use. Regarding GAIN, we used the same",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                666,
+                                506,
+                                678
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        666,
+                                        506,
+                                        678
+                                    ],
+                                    "score": 1.0,
+                                    "content": "network architecture as our implementation of GI and hyper-parameters as used by the GAIN authors7.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                676,
+                                506,
+                                690
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        676,
+                                        506,
+                                        690
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In the DAE implementation, due to the incomplete data assumption, only observed features appear in",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                688,
+                                505,
+                                700
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        688,
+                                        505,
+                                        700
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the loss function, ignoring reconstruction terms corresponding to missing features. Due to scalability",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                699,
+                                506,
+                                711
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        699,
+                                        506,
+                                        711
+                                    ],
+                                    "score": 1.0,
+                                    "content": "issues, we were only able to use MICE for the smaller non-image datasets. For these methods, to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                709,
+                                506,
+                                723
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        709,
+                                        506,
+                                        723
+                                    ],
+                                    "score": 1.0,
+                                    "content": "train and evaluate classifiers, we use predictors trained on imputed datasets rather than the stochastic",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                721,
+                                252,
+                                733
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        721,
+                                        252,
+                                        733
+                                    ],
+                                    "score": 1.0,
+                                    "content": "predictor suggested in Algorithm 1.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        }
+                    ],
+                    "index": 40.5,
+                    "bbox_fs": [
+                        104,
+                        600,
+                        506,
+                        733
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "image",
+                    "bbox": [
+                        109,
+                        83,
+                        500,
+                        242
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                109,
+                                83,
+                                500,
+                                242
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        109,
+                                        83,
+                                        500,
+                                        242
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                109,
+                                                83,
+                                                500,
+                                                242
+                                            ],
+                                            "score": 0.95,
+                                            "type": "image",
+                                            "image_path": "7dee5514f122f68c96a6de74132a186a9e0266bbd52176e7a7dd6f46d002cb83.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 1,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        109,
+                                        83,
+                                        500,
+                                        136.0
+                                    ],
+                                    "spans": [],
+                                    "index": 0
+                                },
+                                {
+                                    "bbox": [
+                                        109,
+                                        136.0,
+                                        500,
+                                        189.0
+                                    ],
+                                    "spans": [],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        109,
+                                        189.0,
+                                        500,
+                                        242.0
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                106,
+                                251,
+                                505,
+                                285
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        251,
+                                        506,
+                                        264
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                251,
+                                                506,
+                                                264
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 2: Comparison of FID scores on CIFAR-10 dataset for (a) uniform and (b) rectangular",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 3
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        263,
+                                        505,
+                                        275
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                263,
+                                                505,
+                                                275
+                                            ],
+                                            "score": 1.0,
+                                            "content": "missingness. Lower FID score is better. In many cases, variance values are very small and only",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 4
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        273,
+                                        260,
+                                        286
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                273,
+                                                260,
+                                                286
+                                            ],
+                                            "score": 1.0,
+                                            "content": "observable by magnifying the figures.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 5
+                                }
+                            ],
+                            "index": 4
+                        }
+                    ],
+                    "index": 2.5
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        307,
+                        170,
+                        318
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                307,
+                                171,
+                                319
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        307,
+                                        171,
+                                        319
+                                    ],
+                                    "score": 1.0,
+                                    "content": "4.4 RESULTS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        }
+                    ],
+                    "index": 6
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        327,
+                        505,
+                        448
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                326,
+                                505,
+                                341
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        326,
+                                        505,
+                                        341
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Figure 2 presents the comparison of FID scores on the CIFAR-10 dataset at different missing rates",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                338,
+                                505,
+                                352
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        338,
+                                        505,
+                                        352
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for uniform and rectangular missingness. As it can be inferred from these plots, GI outperforms",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                350,
+                                505,
+                                361
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        350,
+                                        505,
+                                        361
+                                    ],
+                                    "score": 1.0,
+                                    "content": "other alternatives in all cases. Also, it can be seen that GAIN is able to provide more reasonable",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                360,
+                                505,
+                                373
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        360,
+                                        505,
+                                        373
+                                    ],
+                                    "score": 1.0,
+                                    "content": "results for uniform missing data structure compared to MisGAN which is mainly effective in the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                371,
+                                506,
+                                384
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        371,
+                                        506,
+                                        384
+                                    ],
+                                    "score": 1.0,
+                                    "content": "rectangular missing data structure. One possible explanation for this behavior might be the fact that",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                382,
+                                506,
+                                395
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        382,
+                                        506,
+                                        395
+                                    ],
+                                    "score": 1.0,
+                                    "content": "GAIN has an MSE loss term acting similar to an autoencoder loss smoothing noisy missing pixels.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                393,
+                                506,
+                                406
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        393,
+                                        506,
+                                        406
+                                    ],
+                                    "score": 1.0,
+                                    "content": "On the other hand, MisGAN tries to explicitly model missingness structure and is more successful in",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                405,
+                                505,
+                                416
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        405,
+                                        505,
+                                        416
+                                    ],
+                                    "score": 1.0,
+                                    "content": "capturing a more structured missingness such as the case of a rectangular structure. Table 1 provides",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                416,
+                                505,
+                                428
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        416,
+                                        505,
+                                        428
+                                    ],
+                                    "score": 1.0,
+                                    "content": "a comparison between the top-1 classification accuracy achieved using each method at different",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                426,
+                                505,
+                                439
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        426,
+                                        505,
+                                        439
+                                    ],
+                                    "score": 1.0,
+                                    "content": "missing rates and structures. From this table, GI outperforms other work by achieving the best results",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                436,
+                                188,
+                                449
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        436,
+                                        188,
+                                        449
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in 5 out of 6 cases8.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        }
+                    ],
+                    "index": 12
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        454,
+                        505,
+                        509
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                455,
+                                505,
+                                466
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        455,
+                                        505,
+                                        466
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Table 2 presents a comparison of classification accuracies for Landsat, MIT-BIH, and Diabetes",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                465,
+                                506,
+                                477
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        465,
+                                        506,
+                                        477
+                                    ],
+                                    "score": 1.0,
+                                    "content": "datasets at different missing rates. In the Landsat benchmark, GI outperforms other work in all cases.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                476,
+                                505,
+                                488
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        476,
+                                        484,
+                                        488
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Regarding the MIT-BIH experiemts, GI outperforms other work for missing rates more than",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        484,
+                                        476,
+                                        505,
+                                        487
+                                    ],
+                                    "score": 0.85,
+                                    "content": "30 \\%",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                487,
+                                505,
+                                499
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        487,
+                                        505,
+                                        499
+                                    ],
+                                    "score": 1.0,
+                                    "content": "while achieving similar accuracies to GAIN for lower missing rates. In the diabetes classification",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                498,
+                                413,
+                                511
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        498,
+                                        389,
+                                        511
+                                    ],
+                                    "score": 1.0,
+                                    "content": "task, GI appears to be most effective imputing missing rates more than",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        390,
+                                        498,
+                                        409,
+                                        508
+                                    ],
+                                    "score": 0.85,
+                                    "content": "20 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        410,
+                                        498,
+                                        413,
+                                        511
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        }
+                    ],
+                    "index": 20
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        515,
+                        505,
+                        582
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                514,
+                                506,
+                                528
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        514,
+                                        506,
+                                        528
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Figure 3 shows a comparison of accuracy versus certainty plots for GI, MisGAN, and GAIN on",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                526,
+                                505,
+                                538
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        527,
+                                        261,
+                                        538
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Landsat dataset at the missing rate of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        262,
+                                        526,
+                                        281,
+                                        537
+                                    ],
+                                    "score": 0.88,
+                                    "content": "40 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        282,
+                                        527,
+                                        505,
+                                        538
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". To generate these figures we trained each imputation",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                537,
+                                505,
+                                550
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        537,
+                                        505,
+                                        550
+                                    ],
+                                    "score": 1.0,
+                                    "content": "method and then used Algorithm 1 to train predictors on imputed samples. Finally, Algorithm 2 used",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                548,
+                                505,
+                                560
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        548,
+                                        505,
+                                        560
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to measure the average accuracy at different prediction confidence levels based on a sample of 128",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                559,
+                                505,
+                                572
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        559,
+                                        505,
+                                        572
+                                    ],
+                                    "score": 1.0,
+                                    "content": "imputations for each test example. As it can be seen from the plots, GI provides results closest to the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                569,
+                                408,
+                                583
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        569,
+                                        408,
+                                        583
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ideal case of having average confidence values equal to average accuracies.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        }
+                    ],
+                    "index": 25.5
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        595,
+                        318,
+                        606
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                594,
+                                320,
+                                608
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        594,
+                                        320,
+                                        608
+                                    ],
+                                    "score": 1.0,
+                                    "content": "4.5 VISUALIZATION USING SYNTHESIZED DATA",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        }
+                    ],
+                    "index": 29
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        615,
+                        505,
+                        660
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                615,
+                                505,
+                                628
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        615,
+                                        505,
+                                        628
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In order to provide further insight into the operation of GI and how imputations can potentially",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                627,
+                                506,
+                                639
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        627,
+                                        506,
+                                        639
+                                    ],
+                                    "score": 1.0,
+                                    "content": "influence the outcomes of predictions, we conduct experiments on a synthesized dataset. The original",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                638,
+                                505,
+                                649
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        638,
+                                        505,
+                                        649
+                                    ],
+                                    "score": 1.0,
+                                    "content": "underlying data distribution is generated by sampling 5000 samples from 4 Gaussians of standard",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                649,
+                                506,
+                                661
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        649,
+                                        506,
+                                        661
+                                    ],
+                                    "score": 1.0,
+                                    "content": "deviation 0.1 centered on the vertices of a unit square. We assign two different classes to each cluster",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        }
+                    ],
+                    "index": 31.5
+                }
+            ],
+            "page_idx": 7,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        107,
+                        668,
+                        506,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                118,
+                                667,
+                                472,
+                                681
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        118,
+                                        667,
+                                        472,
+                                        681
+                                    ],
+                                    "score": 1.0,
+                                    "content": "5https://github.com/mseitzer/pytorch-fid is adapted to measure the FID scores.",
+                                    "type": "text"
+                                }
+                            ]
+                        },
+                        {
+                            "bbox": [
+                                118,
+                                678,
+                                303,
+                                693
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        118,
+                                        678,
+                                        303,
+                                        693
+                                    ],
+                                    "score": 1.0,
+                                    "content": "6https://github.com/steveli/misgan",
+                                    "type": "text"
+                                }
+                            ]
+                        },
+                        {
+                            "bbox": [
+                                118,
+                                689,
+                                308,
+                                703
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        118,
+                                        689,
+                                        308,
+                                        703
+                                    ],
+                                    "score": 1.0,
+                                    "content": "7https://github.com/jsyoon0823/GAIN",
+                                    "type": "text"
+                                }
+                            ]
+                        },
+                        {
+                            "bbox": [
+                                118,
+                                699,
+                                506,
+                                715
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        118,
+                                        699,
+                                        506,
+                                        715
+                                    ],
+                                    "score": 1.0,
+                                    "content": "8An earlier version of this paper reported results that are different from the current manuscript. The current",
+                                    "type": "text"
+                                }
+                            ]
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                711,
+                                506,
+                                723
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        711,
+                                        506,
+                                        723
+                                    ],
+                                    "score": 1.0,
+                                    "content": "version is using the stochastic predictor exclusively on the suggested imputation method and trained using more",
+                                    "type": "text"
+                                }
+                            ]
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                721,
+                                227,
+                                733
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        721,
+                                        227,
+                                        733
+                                    ],
+                                    "score": 1.0,
+                                    "content": "precise hyper-parameter settings.",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        302,
+                        751,
+                        308,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                300,
+                                750,
+                                309,
+                                761
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        300,
+                                        750,
+                                        309,
+                                        761
+                                    ],
+                                    "score": 1.0,
+                                    "content": "8",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        107,
+                        27,
+                        308,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                25,
+                                309,
+                                39
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        25,
+                                        309,
+                                        39
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Under review as a conference paper at ICLR 2020",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "image",
+                    "bbox": [
+                        109,
+                        83,
+                        500,
+                        242
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                109,
+                                83,
+                                500,
+                                242
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        109,
+                                        83,
+                                        500,
+                                        242
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                109,
+                                                83,
+                                                500,
+                                                242
+                                            ],
+                                            "score": 0.95,
+                                            "type": "image",
+                                            "image_path": "7dee5514f122f68c96a6de74132a186a9e0266bbd52176e7a7dd6f46d002cb83.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 1,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        109,
+                                        83,
+                                        500,
+                                        136.0
+                                    ],
+                                    "spans": [],
+                                    "index": 0
+                                },
+                                {
+                                    "bbox": [
+                                        109,
+                                        136.0,
+                                        500,
+                                        189.0
+                                    ],
+                                    "spans": [],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        109,
+                                        189.0,
+                                        500,
+                                        242.0
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                106,
+                                251,
+                                505,
+                                285
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        251,
+                                        506,
+                                        264
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                251,
+                                                506,
+                                                264
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 2: Comparison of FID scores on CIFAR-10 dataset for (a) uniform and (b) rectangular",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 3
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        263,
+                                        505,
+                                        275
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                263,
+                                                505,
+                                                275
+                                            ],
+                                            "score": 1.0,
+                                            "content": "missingness. Lower FID score is better. In many cases, variance values are very small and only",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 4
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        273,
+                                        260,
+                                        286
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                273,
+                                                260,
+                                                286
+                                            ],
+                                            "score": 1.0,
+                                            "content": "observable by magnifying the figures.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 5
+                                }
+                            ],
+                            "index": 4
+                        }
+                    ],
+                    "index": 2.5
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        307,
+                        170,
+                        318
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                307,
+                                171,
+                                319
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        307,
+                                        171,
+                                        319
+                                    ],
+                                    "score": 1.0,
+                                    "content": "4.4 RESULTS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        }
+                    ],
+                    "index": 6
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        327,
+                        505,
+                        448
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                326,
+                                505,
+                                341
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        326,
+                                        505,
+                                        341
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Figure 2 presents the comparison of FID scores on the CIFAR-10 dataset at different missing rates",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                338,
+                                505,
+                                352
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        338,
+                                        505,
+                                        352
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for uniform and rectangular missingness. As it can be inferred from these plots, GI outperforms",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                350,
+                                505,
+                                361
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        350,
+                                        505,
+                                        361
+                                    ],
+                                    "score": 1.0,
+                                    "content": "other alternatives in all cases. Also, it can be seen that GAIN is able to provide more reasonable",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                360,
+                                505,
+                                373
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        360,
+                                        505,
+                                        373
+                                    ],
+                                    "score": 1.0,
+                                    "content": "results for uniform missing data structure compared to MisGAN which is mainly effective in the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                371,
+                                506,
+                                384
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        371,
+                                        506,
+                                        384
+                                    ],
+                                    "score": 1.0,
+                                    "content": "rectangular missing data structure. One possible explanation for this behavior might be the fact that",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                382,
+                                506,
+                                395
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        382,
+                                        506,
+                                        395
+                                    ],
+                                    "score": 1.0,
+                                    "content": "GAIN has an MSE loss term acting similar to an autoencoder loss smoothing noisy missing pixels.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                393,
+                                506,
+                                406
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        393,
+                                        506,
+                                        406
+                                    ],
+                                    "score": 1.0,
+                                    "content": "On the other hand, MisGAN tries to explicitly model missingness structure and is more successful in",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                405,
+                                505,
+                                416
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        405,
+                                        505,
+                                        416
+                                    ],
+                                    "score": 1.0,
+                                    "content": "capturing a more structured missingness such as the case of a rectangular structure. Table 1 provides",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                416,
+                                505,
+                                428
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        416,
+                                        505,
+                                        428
+                                    ],
+                                    "score": 1.0,
+                                    "content": "a comparison between the top-1 classification accuracy achieved using each method at different",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                426,
+                                505,
+                                439
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        426,
+                                        505,
+                                        439
+                                    ],
+                                    "score": 1.0,
+                                    "content": "missing rates and structures. From this table, GI outperforms other work by achieving the best results",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                436,
+                                188,
+                                449
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        436,
+                                        188,
+                                        449
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in 5 out of 6 cases8.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        }
+                    ],
+                    "index": 12,
+                    "bbox_fs": [
+                        105,
+                        326,
+                        506,
+                        449
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        454,
+                        505,
+                        509
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                455,
+                                505,
+                                466
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        455,
+                                        505,
+                                        466
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Table 2 presents a comparison of classification accuracies for Landsat, MIT-BIH, and Diabetes",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                465,
+                                506,
+                                477
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        465,
+                                        506,
+                                        477
+                                    ],
+                                    "score": 1.0,
+                                    "content": "datasets at different missing rates. In the Landsat benchmark, GI outperforms other work in all cases.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                476,
+                                505,
+                                488
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        476,
+                                        484,
+                                        488
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Regarding the MIT-BIH experiemts, GI outperforms other work for missing rates more than",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        484,
+                                        476,
+                                        505,
+                                        487
+                                    ],
+                                    "score": 0.85,
+                                    "content": "30 \\%",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                487,
+                                505,
+                                499
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        487,
+                                        505,
+                                        499
+                                    ],
+                                    "score": 1.0,
+                                    "content": "while achieving similar accuracies to GAIN for lower missing rates. In the diabetes classification",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                498,
+                                413,
+                                511
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        498,
+                                        389,
+                                        511
+                                    ],
+                                    "score": 1.0,
+                                    "content": "task, GI appears to be most effective imputing missing rates more than",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        390,
+                                        498,
+                                        409,
+                                        508
+                                    ],
+                                    "score": 0.85,
+                                    "content": "20 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        410,
+                                        498,
+                                        413,
+                                        511
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        }
+                    ],
+                    "index": 20,
+                    "bbox_fs": [
+                        105,
+                        455,
+                        506,
+                        511
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        515,
+                        505,
+                        582
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                514,
+                                506,
+                                528
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        514,
+                                        506,
+                                        528
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Figure 3 shows a comparison of accuracy versus certainty plots for GI, MisGAN, and GAIN on",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                526,
+                                505,
+                                538
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        527,
+                                        261,
+                                        538
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Landsat dataset at the missing rate of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        262,
+                                        526,
+                                        281,
+                                        537
+                                    ],
+                                    "score": 0.88,
+                                    "content": "40 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        282,
+                                        527,
+                                        505,
+                                        538
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". To generate these figures we trained each imputation",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                537,
+                                505,
+                                550
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        537,
+                                        505,
+                                        550
+                                    ],
+                                    "score": 1.0,
+                                    "content": "method and then used Algorithm 1 to train predictors on imputed samples. Finally, Algorithm 2 used",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                548,
+                                505,
+                                560
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        548,
+                                        505,
+                                        560
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to measure the average accuracy at different prediction confidence levels based on a sample of 128",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                559,
+                                505,
+                                572
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        559,
+                                        505,
+                                        572
+                                    ],
+                                    "score": 1.0,
+                                    "content": "imputations for each test example. As it can be seen from the plots, GI provides results closest to the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                569,
+                                408,
+                                583
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        569,
+                                        408,
+                                        583
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ideal case of having average confidence values equal to average accuracies.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        }
+                    ],
+                    "index": 25.5,
+                    "bbox_fs": [
+                        105,
+                        514,
+                        506,
+                        583
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        595,
+                        318,
+                        606
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                594,
+                                320,
+                                608
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        594,
+                                        320,
+                                        608
+                                    ],
+                                    "score": 1.0,
+                                    "content": "4.5 VISUALIZATION USING SYNTHESIZED DATA",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        }
+                    ],
+                    "index": 29
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        615,
+                        505,
+                        660
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                615,
+                                505,
+                                628
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        615,
+                                        505,
+                                        628
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In order to provide further insight into the operation of GI and how imputations can potentially",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                627,
+                                506,
+                                639
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        627,
+                                        506,
+                                        639
+                                    ],
+                                    "score": 1.0,
+                                    "content": "influence the outcomes of predictions, we conduct experiments on a synthesized dataset. The original",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                638,
+                                505,
+                                649
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        638,
+                                        505,
+                                        649
+                                    ],
+                                    "score": 1.0,
+                                    "content": "underlying data distribution is generated by sampling 5000 samples from 4 Gaussians of standard",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                649,
+                                506,
+                                661
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        649,
+                                        506,
+                                        661
+                                    ],
+                                    "score": 1.0,
+                                    "content": "deviation 0.1 centered on the vertices of a unit square. We assign two different classes to each cluster",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        }
+                    ],
+                    "index": 31.5,
+                    "bbox_fs": [
+                        105,
+                        615,
+                        506,
+                        661
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "table",
+                    "bbox": [
+                        108,
+                        101,
+                        503,
+                        232
+                    ],
+                    "blocks": [
+                        {
+                            "type": "table_caption",
+                            "bbox": [
+                                123,
+                                80,
+                                486,
+                                92
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        122,
+                                        79,
+                                        488,
+                                        94
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                122,
+                                                79,
+                                                488,
+                                                94
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Table 1: Top-1 CIFAR-10 classification accuracy for different missing rates and structures.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 0
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "type": "table_body",
+                            "bbox": [
+                                108,
+                                101,
+                                503,
+                                232
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        108,
+                                        101,
+                                        503,
+                                        232
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                108,
+                                                101,
+                                                503,
+                                                232
+                                            ],
+                                            "score": 0.983,
+                                            "html": "<table><tr><td rowspan=\"2\"></td><td colspan=\"5\">Accuracy at Missing Rate (%) MCAR Rect.</td><td rowspan=\"2\"></td></tr><tr><td colspan=\"2\">MCARUniform</td><td colspan=\"3\"></td></tr><tr><td>Method</td><td>20%</td><td>40%</td><td>60%</td><td>20%</td><td>40%</td><td>60%</td></tr><tr><td>GI</td><td>89.5 (±0.45)</td><td>87.1 (±0.54)</td><td>80.3 (±0.26)</td><td>84.0 (±0.03)</td><td>76.9 (±0.03)</td><td>66.1 (±0.16)</td></tr><tr><td>MisGAN</td><td>86.5 (±0.31)</td><td>83.7 (±0.40)</td><td>78.7 (±0.26)</td><td>82.9 (±0.44)</td><td>75.6 (±0.20)</td><td>65.0 (±0.31)</td></tr><tr><td>GAIN</td><td>88.7 (±0.45)</td><td>86.0 (±0.86)</td><td>81.8 (±0.03)</td><td>81.7 (±0.03)</td><td>73.6 (±0.35)</td><td>58.4 (±1.66)</td></tr><tr><td>DAE</td><td>88.0 (±0.22)</td><td>84.0 (±0.50)</td><td>79.8 (±0.71)</td><td>83.3 (±0.64)</td><td>75.5 (±0.44)</td><td>63.8 (±0.24)</td></tr><tr><td>Mean</td><td>85.7 (±0.02)</td><td>83.4 (±0.38)</td><td>79.2 (±0.16)</td><td>82.7 (±0.15)</td><td>75.3 (±0.16)</td><td>64.0 (±0.32)</td></tr></table>",
+                                            "type": "table",
+                                            "image_path": "542a5e4e556ade099fa91694dab860b1113629d1e9a5da54d4590b4438731691.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 2,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        108,
+                                        101,
+                                        503,
+                                        144.66666666666666
+                                    ],
+                                    "spans": [],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        108,
+                                        144.66666666666666,
+                                        503,
+                                        188.33333333333331
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                },
+                                {
+                                    "bbox": [
+                                        108,
+                                        188.33333333333331,
+                                        503,
+                                        231.99999999999997
+                                    ],
+                                    "spans": [],
+                                    "index": 3
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 1.0
+                },
+                {
+                    "type": "table",
+                    "bbox": [
+                        108,
+                        280,
+                        504,
+                        518
+                    ],
+                    "blocks": [
+                        {
+                            "type": "table_caption",
+                            "bbox": [
+                                106,
+                                248,
+                                504,
+                                272
+                            ],
+                            "group_id": 1,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        248,
+                                        506,
+                                        262
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                248,
+                                                506,
+                                                262
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Table 2: Comparison of classification accuracies for Landsat, MIT-BIH, and Diabetes datasets at",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 4
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        259,
+                                        200,
+                                        273
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                259,
+                                                200,
+                                                273
+                                            ],
+                                            "score": 1.0,
+                                            "content": "different missing rates.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 5
+                                }
+                            ],
+                            "index": 4.5
+                        },
+                        {
+                            "type": "table_body",
+                            "bbox": [
+                                108,
+                                280,
+                                504,
+                                518
+                            ],
+                            "group_id": 1,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        108,
+                                        280,
+                                        504,
+                                        518
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                108,
+                                                280,
+                                                504,
+                                                518
+                                            ],
+                                            "score": 0.985,
+                                            "html": "<table><tr><td></td><td></td><td colspan=\"4\"> Accuracy at Missing Rate (%)a</td></tr><tr><td>Dataset</td><td>Method</td><td>10%</td><td>20%</td><td>30%</td><td>40%</td></tr><tr><td rowspan=\"5\">Landsat (Dua &amp; Graff, 2017)</td><td>GI</td><td>89.9 (±0.36)</td><td>89.6 (±0.36)</td><td>89.0 (±0.03)</td><td>88.0 (±0.22)</td></tr><tr><td>MisGAN</td><td>87.2 (±0.01)</td><td>85.7 (±0.19)</td><td>84.0 (±0.61)</td><td>82.9 (±0.75)</td></tr><tr><td>GAIN</td><td>89.7 (±0.42)</td><td>89.4 (±0.56)</td><td>88.4 (±0.71)</td><td>87.7 (±0.10)</td></tr><tr><td>DAE</td><td>89.4 (±0.10)</td><td>88.6 (±0.54)</td><td>87.5 (±0.14)</td><td>86.6 (±0.21)</td></tr><tr><td>MICE</td><td>89.5 (±0.16)</td><td>89.3 (±0.10)</td><td>88.1 (±0.49)</td><td>87.5 (±0.03)</td></tr><tr><td rowspan=\"5\">MIT-BIH(Moody &amp; Mark,2001)</td><td>GI</td><td>98.5 (±0.02)</td><td>98.4 (±0.03)</td><td>98.2 (±0.07)</td><td>97.7 (±0.03)</td></tr><tr><td>MisGAN</td><td>97.8 (±0.13)</td><td>97.4 (±0.07)</td><td>96.7 (±0.07)</td><td>96.2 (±0.09)</td></tr><tr><td>GAIN</td><td>98.5 (±0.02)</td><td>98.4 (±0.06)</td><td>98.0 (±0.09)</td><td>97.5 (±0.18)</td></tr><tr><td>DAE</td><td>98.4 (±0.02)</td><td>98.2 (±0.11)</td><td>97.9 (±0.09)</td><td>97.4 (±0.02)</td></tr><tr><td>MICE</td><td>98.4 (±0.01)</td><td>98.3 (±0.01)</td><td>98.1 (±0.01)</td><td>97.5 (±0.12)</td></tr><tr><td rowspan=\"5\">Diabetes (Kachuee et al., 2019)</td><td>GI</td><td>89.6 (±0.13)</td><td>89.0 (±0.03)</td><td>88.2 (±0.62)</td><td>86.8 (±0.38)</td></tr><tr><td>MisGAN</td><td>89.7 (±0.01)</td><td>88.9 (±0.30)</td><td>87.6 (±0.02)</td><td>86.4 (±0.68)</td></tr><tr><td>GAIN</td><td>89.2 (±0.09)</td><td>88.3 (±0.02)</td><td>86.9 (±0.09)</td><td>83.8 (±1.44)</td></tr><tr><td>DAE</td><td>89.3 (±0.05)</td><td>88.2 (±0.19)</td><td>86.9 (±0.09)</td><td>84.8 (±0.03)</td></tr><tr><td>MICE</td><td>89.8 (±0.08)</td><td>88.8 (±0.01)</td><td>88.0 (±0.08)</td><td>86.1 (±0.02)</td></tr></table>",
+                                            "type": "table",
+                                            "image_path": "20ad6cf606f031aa541edca374af1e1c4f0f167e153165882ca7b2dd13238666.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 7,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        108,
+                                        280,
+                                        504,
+                                        359.3333333333333
+                                    ],
+                                    "spans": [],
+                                    "index": 6
+                                },
+                                {
+                                    "bbox": [
+                                        108,
+                                        359.3333333333333,
+                                        504,
+                                        438.66666666666663
+                                    ],
+                                    "spans": [],
+                                    "index": 7
+                                },
+                                {
+                                    "bbox": [
+                                        108,
+                                        438.66666666666663,
+                                        504,
+                                        518.0
+                                    ],
+                                    "spans": [],
+                                    "index": 8
+                                }
+                            ]
+                        },
+                        {
+                            "type": "table_footnote",
+                            "bbox": [
+                                109,
+                                524,
+                                504,
+                                544
+                            ],
+                            "group_id": 1,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        118,
+                                        522,
+                                        506,
+                                        535
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                118,
+                                                522,
+                                                393,
+                                                535
+                                            ],
+                                            "score": 1.0,
+                                            "content": "aBaseline accuracies for complete datasets (zero missing rate) are equal to",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                394,
+                                                524,
+                                                419,
+                                                533
+                                            ],
+                                            "score": 0.83,
+                                            "content": "9 0 . 9 \\%",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                419,
+                                                522,
+                                                421,
+                                                535
+                                            ],
+                                            "score": 1.0,
+                                            "content": ",",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                422,
+                                                524,
+                                                447,
+                                                533
+                                            ],
+                                            "score": 0.82,
+                                            "content": "9 8 . 6 \\%",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                447,
+                                                522,
+                                                466,
+                                                535
+                                            ],
+                                            "score": 1.0,
+                                            "content": ", and",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                466,
+                                                524,
+                                                491,
+                                                533
+                                            ],
+                                            "score": 0.86,
+                                            "content": "9 0 . 7 \\%",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                491,
+                                                522,
+                                                506,
+                                                535
+                                            ],
+                                            "score": 1.0,
+                                            "content": "for",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 9
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        532,
+                                        273,
+                                        545
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                532,
+                                                273,
+                                                545
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Landsat, MIT-BIH and Diabetes, respectively.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 10
+                                }
+                            ],
+                            "index": 9.5
+                        }
+                    ],
+                    "index": 7
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        572,
+                        505,
+                        594
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                572,
+                                506,
+                                584
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        572,
+                                        506,
+                                        584
+                                    ],
+                                    "score": 1.0,
+                                    "content": "such that diagonal vertices are of the same class (see Figure 4a, classes are represented with colors).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                582,
+                                487,
+                                597
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        582,
+                                        390,
+                                        597
+                                    ],
+                                    "score": 1.0,
+                                    "content": "From this underlying distribution, we make an incomplete dataset with",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        391,
+                                        583,
+                                        411,
+                                        593
+                                    ],
+                                    "score": 0.86,
+                                    "content": "50 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        411,
+                                        582,
+                                        487,
+                                        597
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of values missing.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        }
+                    ],
+                    "index": 11.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        600,
+                        505,
+                        731
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                601,
+                                506,
+                                612
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        601,
+                                        506,
+                                        612
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The incomplete synthesized dataset is used to train GI and other imputation methods. We take a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                611,
+                                506,
+                                623
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        611,
+                                        506,
+                                        623
+                                    ],
+                                    "score": 1.0,
+                                    "content": "random test sample in which the second feature has a value of about 0.1 and the other feature is",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                622,
+                                505,
+                                634
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        622,
+                                        505,
+                                        634
+                                    ],
+                                    "score": 1.0,
+                                    "content": "missing. Ideally, in the imputation phase, we would like to sample from the condition distribution",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                632,
+                                505,
+                                646
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        632,
+                                        122,
+                                        646
+                                    ],
+                                    "score": 1.0,
+                                    "content": "i.e.",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        122,
+                                        633,
+                                        186,
+                                        645
+                                    ],
+                                    "score": 0.9,
+                                    "content": "P ( x _ { 1 } | x _ { 2 } = 0 . 1 )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        186,
+                                        632,
+                                        505,
+                                        646
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(see Figure 4b). Here, in the prediction phase, an ideal method would decide on",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                644,
+                                505,
+                                656
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        644,
+                                        505,
+                                        656
+                                    ],
+                                    "score": 1.0,
+                                    "content": "not making a confident classification and report the uncertainty. Note that solely observing the value",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                654,
+                                506,
+                                668
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        654,
+                                        506,
+                                        668
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of 0.1 for the second feature does not provide any useful evidence for the prediction. Figure 4c-f",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                666,
+                                506,
+                                677
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        666,
+                                        506,
+                                        677
+                                    ],
+                                    "score": 1.0,
+                                    "content": "provide samples and classification results for GI, MisGAN, GAIN, and DAE. As it can be inferred",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                677,
+                                505,
+                                689
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        677,
+                                        505,
+                                        689
+                                    ],
+                                    "score": 1.0,
+                                    "content": "from these figures, GI generates reasonable samples from the conditional distribution and also reflects",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                688,
+                                506,
+                                700
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        688,
+                                        506,
+                                        700
+                                    ],
+                                    "score": 1.0,
+                                    "content": "this uncertainty over the prediction. On the other hand MisGAN, probably due to its complexity of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                698,
+                                506,
+                                711
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        698,
+                                        506,
+                                        711
+                                    ],
+                                    "score": 1.0,
+                                    "content": "using three different generators and discriminator pairs, is suffering from mode collapse and is unable",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                709,
+                                505,
+                                723
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        709,
+                                        505,
+                                        723
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to generate samples from the other class, resulting in over-confident assignments. GAIN, perhaps",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                720,
+                                506,
+                                733
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        720,
+                                        506,
+                                        733
+                                    ],
+                                    "score": 1.0,
+                                    "content": "due to the MSE loss terms, is inclined towards the mean of the conditional distribution at the origin.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        }
+                    ],
+                    "index": 18.5
+                }
+            ],
+            "page_idx": 8,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        107,
+                        27,
+                        308,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                107,
+                                26,
+                                308,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        26,
+                                        308,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Under review as a conference paper at ICLR 2020",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        302,
+                        751,
+                        309,
+                        759
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                302,
+                                751,
+                                309,
+                                762
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        302,
+                                        751,
+                                        309,
+                                        762
+                                    ],
+                                    "score": 1.0,
+                                    "content": "9",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "table",
+                    "bbox": [
+                        108,
+                        101,
+                        503,
+                        232
+                    ],
+                    "blocks": [
+                        {
+                            "type": "table_caption",
+                            "bbox": [
+                                123,
+                                80,
+                                486,
+                                92
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        122,
+                                        79,
+                                        488,
+                                        94
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                122,
+                                                79,
+                                                488,
+                                                94
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Table 1: Top-1 CIFAR-10 classification accuracy for different missing rates and structures.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 0
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "type": "table_body",
+                            "bbox": [
+                                108,
+                                101,
+                                503,
+                                232
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        108,
+                                        101,
+                                        503,
+                                        232
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                108,
+                                                101,
+                                                503,
+                                                232
+                                            ],
+                                            "score": 0.983,
+                                            "html": "<table><tr><td rowspan=\"2\"></td><td colspan=\"5\">Accuracy at Missing Rate (%) MCAR Rect.</td><td rowspan=\"2\"></td></tr><tr><td colspan=\"2\">MCARUniform</td><td colspan=\"3\"></td></tr><tr><td>Method</td><td>20%</td><td>40%</td><td>60%</td><td>20%</td><td>40%</td><td>60%</td></tr><tr><td>GI</td><td>89.5 (±0.45)</td><td>87.1 (±0.54)</td><td>80.3 (±0.26)</td><td>84.0 (±0.03)</td><td>76.9 (±0.03)</td><td>66.1 (±0.16)</td></tr><tr><td>MisGAN</td><td>86.5 (±0.31)</td><td>83.7 (±0.40)</td><td>78.7 (±0.26)</td><td>82.9 (±0.44)</td><td>75.6 (±0.20)</td><td>65.0 (±0.31)</td></tr><tr><td>GAIN</td><td>88.7 (±0.45)</td><td>86.0 (±0.86)</td><td>81.8 (±0.03)</td><td>81.7 (±0.03)</td><td>73.6 (±0.35)</td><td>58.4 (±1.66)</td></tr><tr><td>DAE</td><td>88.0 (±0.22)</td><td>84.0 (±0.50)</td><td>79.8 (±0.71)</td><td>83.3 (±0.64)</td><td>75.5 (±0.44)</td><td>63.8 (±0.24)</td></tr><tr><td>Mean</td><td>85.7 (±0.02)</td><td>83.4 (±0.38)</td><td>79.2 (±0.16)</td><td>82.7 (±0.15)</td><td>75.3 (±0.16)</td><td>64.0 (±0.32)</td></tr></table>",
+                                            "type": "table",
+                                            "image_path": "542a5e4e556ade099fa91694dab860b1113629d1e9a5da54d4590b4438731691.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 2,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        108,
+                                        101,
+                                        503,
+                                        144.66666666666666
+                                    ],
+                                    "spans": [],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        108,
+                                        144.66666666666666,
+                                        503,
+                                        188.33333333333331
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                },
+                                {
+                                    "bbox": [
+                                        108,
+                                        188.33333333333331,
+                                        503,
+                                        231.99999999999997
+                                    ],
+                                    "spans": [],
+                                    "index": 3
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 1.0
+                },
+                {
+                    "type": "table",
+                    "bbox": [
+                        108,
+                        280,
+                        504,
+                        518
+                    ],
+                    "blocks": [
+                        {
+                            "type": "table_caption",
+                            "bbox": [
+                                106,
+                                248,
+                                504,
+                                272
+                            ],
+                            "group_id": 1,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        248,
+                                        506,
+                                        262
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                248,
+                                                506,
+                                                262
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Table 2: Comparison of classification accuracies for Landsat, MIT-BIH, and Diabetes datasets at",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 4
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        259,
+                                        200,
+                                        273
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                259,
+                                                200,
+                                                273
+                                            ],
+                                            "score": 1.0,
+                                            "content": "different missing rates.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 5
+                                }
+                            ],
+                            "index": 4.5
+                        },
+                        {
+                            "type": "table_body",
+                            "bbox": [
+                                108,
+                                280,
+                                504,
+                                518
+                            ],
+                            "group_id": 1,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        108,
+                                        280,
+                                        504,
+                                        518
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                108,
+                                                280,
+                                                504,
+                                                518
+                                            ],
+                                            "score": 0.985,
+                                            "html": "<table><tr><td></td><td></td><td colspan=\"4\"> Accuracy at Missing Rate (%)a</td></tr><tr><td>Dataset</td><td>Method</td><td>10%</td><td>20%</td><td>30%</td><td>40%</td></tr><tr><td rowspan=\"5\">Landsat (Dua &amp; Graff, 2017)</td><td>GI</td><td>89.9 (±0.36)</td><td>89.6 (±0.36)</td><td>89.0 (±0.03)</td><td>88.0 (±0.22)</td></tr><tr><td>MisGAN</td><td>87.2 (±0.01)</td><td>85.7 (±0.19)</td><td>84.0 (±0.61)</td><td>82.9 (±0.75)</td></tr><tr><td>GAIN</td><td>89.7 (±0.42)</td><td>89.4 (±0.56)</td><td>88.4 (±0.71)</td><td>87.7 (±0.10)</td></tr><tr><td>DAE</td><td>89.4 (±0.10)</td><td>88.6 (±0.54)</td><td>87.5 (±0.14)</td><td>86.6 (±0.21)</td></tr><tr><td>MICE</td><td>89.5 (±0.16)</td><td>89.3 (±0.10)</td><td>88.1 (±0.49)</td><td>87.5 (±0.03)</td></tr><tr><td rowspan=\"5\">MIT-BIH(Moody &amp; Mark,2001)</td><td>GI</td><td>98.5 (±0.02)</td><td>98.4 (±0.03)</td><td>98.2 (±0.07)</td><td>97.7 (±0.03)</td></tr><tr><td>MisGAN</td><td>97.8 (±0.13)</td><td>97.4 (±0.07)</td><td>96.7 (±0.07)</td><td>96.2 (±0.09)</td></tr><tr><td>GAIN</td><td>98.5 (±0.02)</td><td>98.4 (±0.06)</td><td>98.0 (±0.09)</td><td>97.5 (±0.18)</td></tr><tr><td>DAE</td><td>98.4 (±0.02)</td><td>98.2 (±0.11)</td><td>97.9 (±0.09)</td><td>97.4 (±0.02)</td></tr><tr><td>MICE</td><td>98.4 (±0.01)</td><td>98.3 (±0.01)</td><td>98.1 (±0.01)</td><td>97.5 (±0.12)</td></tr><tr><td rowspan=\"5\">Diabetes (Kachuee et al., 2019)</td><td>GI</td><td>89.6 (±0.13)</td><td>89.0 (±0.03)</td><td>88.2 (±0.62)</td><td>86.8 (±0.38)</td></tr><tr><td>MisGAN</td><td>89.7 (±0.01)</td><td>88.9 (±0.30)</td><td>87.6 (±0.02)</td><td>86.4 (±0.68)</td></tr><tr><td>GAIN</td><td>89.2 (±0.09)</td><td>88.3 (±0.02)</td><td>86.9 (±0.09)</td><td>83.8 (±1.44)</td></tr><tr><td>DAE</td><td>89.3 (±0.05)</td><td>88.2 (±0.19)</td><td>86.9 (±0.09)</td><td>84.8 (±0.03)</td></tr><tr><td>MICE</td><td>89.8 (±0.08)</td><td>88.8 (±0.01)</td><td>88.0 (±0.08)</td><td>86.1 (±0.02)</td></tr></table>",
+                                            "type": "table",
+                                            "image_path": "20ad6cf606f031aa541edca374af1e1c4f0f167e153165882ca7b2dd13238666.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 7,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        108,
+                                        280,
+                                        504,
+                                        359.3333333333333
+                                    ],
+                                    "spans": [],
+                                    "index": 6
+                                },
+                                {
+                                    "bbox": [
+                                        108,
+                                        359.3333333333333,
+                                        504,
+                                        438.66666666666663
+                                    ],
+                                    "spans": [],
+                                    "index": 7
+                                },
+                                {
+                                    "bbox": [
+                                        108,
+                                        438.66666666666663,
+                                        504,
+                                        518.0
+                                    ],
+                                    "spans": [],
+                                    "index": 8
+                                }
+                            ]
+                        },
+                        {
+                            "type": "table_footnote",
+                            "bbox": [
+                                109,
+                                524,
+                                504,
+                                544
+                            ],
+                            "group_id": 1,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        118,
+                                        522,
+                                        506,
+                                        535
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                118,
+                                                522,
+                                                393,
+                                                535
+                                            ],
+                                            "score": 1.0,
+                                            "content": "aBaseline accuracies for complete datasets (zero missing rate) are equal to",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                394,
+                                                524,
+                                                419,
+                                                533
+                                            ],
+                                            "score": 0.83,
+                                            "content": "9 0 . 9 \\%",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                419,
+                                                522,
+                                                421,
+                                                535
+                                            ],
+                                            "score": 1.0,
+                                            "content": ",",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                422,
+                                                524,
+                                                447,
+                                                533
+                                            ],
+                                            "score": 0.82,
+                                            "content": "9 8 . 6 \\%",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                447,
+                                                522,
+                                                466,
+                                                535
+                                            ],
+                                            "score": 1.0,
+                                            "content": ", and",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                466,
+                                                524,
+                                                491,
+                                                533
+                                            ],
+                                            "score": 0.86,
+                                            "content": "9 0 . 7 \\%",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                491,
+                                                522,
+                                                506,
+                                                535
+                                            ],
+                                            "score": 1.0,
+                                            "content": "for",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 9
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        532,
+                                        273,
+                                        545
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                532,
+                                                273,
+                                                545
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Landsat, MIT-BIH and Diabetes, respectively.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 10
+                                }
+                            ],
+                            "index": 9.5
+                        }
+                    ],
+                    "index": 7
+                },
+                {
+                    "type": "list",
+                    "bbox": [
+                        107,
+                        572,
+                        505,
+                        594
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                572,
+                                506,
+                                584
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        572,
+                                        506,
+                                        584
+                                    ],
+                                    "score": 1.0,
+                                    "content": "such that diagonal vertices are of the same class (see Figure 4a, classes are represented with colors).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                582,
+                                487,
+                                597
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        582,
+                                        390,
+                                        597
+                                    ],
+                                    "score": 1.0,
+                                    "content": "From this underlying distribution, we make an incomplete dataset with",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        391,
+                                        583,
+                                        411,
+                                        593
+                                    ],
+                                    "score": 0.86,
+                                    "content": "50 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        411,
+                                        582,
+                                        487,
+                                        597
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of values missing.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12,
+                            "is_list_start_line": true,
+                            "is_list_end_line": true
+                        }
+                    ],
+                    "index": 11.5,
+                    "bbox_fs": [
+                        105,
+                        572,
+                        506,
+                        597
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        600,
+                        505,
+                        731
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                601,
+                                506,
+                                612
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        601,
+                                        506,
+                                        612
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The incomplete synthesized dataset is used to train GI and other imputation methods. We take a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                611,
+                                506,
+                                623
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        611,
+                                        506,
+                                        623
+                                    ],
+                                    "score": 1.0,
+                                    "content": "random test sample in which the second feature has a value of about 0.1 and the other feature is",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                622,
+                                505,
+                                634
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        622,
+                                        505,
+                                        634
+                                    ],
+                                    "score": 1.0,
+                                    "content": "missing. Ideally, in the imputation phase, we would like to sample from the condition distribution",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                632,
+                                505,
+                                646
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        632,
+                                        122,
+                                        646
+                                    ],
+                                    "score": 1.0,
+                                    "content": "i.e.",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        122,
+                                        633,
+                                        186,
+                                        645
+                                    ],
+                                    "score": 0.9,
+                                    "content": "P ( x _ { 1 } | x _ { 2 } = 0 . 1 )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        186,
+                                        632,
+                                        505,
+                                        646
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(see Figure 4b). Here, in the prediction phase, an ideal method would decide on",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                644,
+                                505,
+                                656
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        644,
+                                        505,
+                                        656
+                                    ],
+                                    "score": 1.0,
+                                    "content": "not making a confident classification and report the uncertainty. Note that solely observing the value",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                654,
+                                506,
+                                668
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        654,
+                                        506,
+                                        668
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of 0.1 for the second feature does not provide any useful evidence for the prediction. Figure 4c-f",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                666,
+                                506,
+                                677
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        666,
+                                        506,
+                                        677
+                                    ],
+                                    "score": 1.0,
+                                    "content": "provide samples and classification results for GI, MisGAN, GAIN, and DAE. As it can be inferred",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                677,
+                                505,
+                                689
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        677,
+                                        505,
+                                        689
+                                    ],
+                                    "score": 1.0,
+                                    "content": "from these figures, GI generates reasonable samples from the conditional distribution and also reflects",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                688,
+                                506,
+                                700
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        688,
+                                        506,
+                                        700
+                                    ],
+                                    "score": 1.0,
+                                    "content": "this uncertainty over the prediction. On the other hand MisGAN, probably due to its complexity of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                698,
+                                506,
+                                711
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        698,
+                                        506,
+                                        711
+                                    ],
+                                    "score": 1.0,
+                                    "content": "using three different generators and discriminator pairs, is suffering from mode collapse and is unable",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                709,
+                                505,
+                                723
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        709,
+                                        505,
+                                        723
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to generate samples from the other class, resulting in over-confident assignments. GAIN, perhaps",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                720,
+                                506,
+                                733
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        720,
+                                        506,
+                                        733
+                                    ],
+                                    "score": 1.0,
+                                    "content": "due to the MSE loss terms, is inclined towards the mean of the conditional distribution at the origin.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        }
+                    ],
+                    "index": 18.5,
+                    "bbox_fs": [
+                        104,
+                        601,
+                        506,
+                        733
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "image",
+                    "bbox": [
+                        131,
+                        85,
+                        479,
+                        196
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                131,
+                                85,
+                                479,
+                                196
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        131,
+                                        85,
+                                        479,
+                                        196
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                131,
+                                                85,
+                                                479,
+                                                196
+                                            ],
+                                            "score": 0.971,
+                                            "type": "image",
+                                            "image_path": "99ff87491c12b80a83f83e95b0f6597c20ef1cd8b2470028e19f5844c61fce0b.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 1,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        131,
+                                        85,
+                                        479,
+                                        122.0
+                                    ],
+                                    "spans": [],
+                                    "index": 0
+                                },
+                                {
+                                    "bbox": [
+                                        131,
+                                        122.0,
+                                        479,
+                                        159.0
+                                    ],
+                                    "spans": [],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        131,
+                                        159.0,
+                                        479,
+                                        196.0
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                107,
+                                208,
+                                503,
+                                231
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        208,
+                                        506,
+                                        222
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                208,
+                                                506,
+                                                222
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 3: Accuracy versus certainty plots for (a) GI, (b) MisGAN, and (c) GAIN on Landsat dataset",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 3
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        220,
+                                        216,
+                                        232
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                220,
+                                                192,
+                                                232
+                                            ],
+                                            "score": 1.0,
+                                            "content": "at the missing rate of",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                192,
+                                                220,
+                                                212,
+                                                231
+                                            ],
+                                            "score": 0.88,
+                                            "content": "40 \\%",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                212,
+                                                220,
+                                                216,
+                                                232
+                                            ],
+                                            "score": 1.0,
+                                            "content": ".",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 4
+                                }
+                            ],
+                            "index": 3.5
+                        }
+                    ],
+                    "index": 2.25
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        258,
+                        505,
+                        282
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                259,
+                                505,
+                                272
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        259,
+                                        505,
+                                        272
+                                    ],
+                                    "score": 1.0,
+                                    "content": "DAE, as expected, due to its MSE loss term, only captures the expected value of the distribution",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                271,
+                                424,
+                                282
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        271,
+                                        424,
+                                        282
+                                    ],
+                                    "score": 1.0,
+                                    "content": "mean hence reducing the MSE error and generates over-smoothed imputations.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        }
+                    ],
+                    "index": 5.5
+                },
+                {
+                    "type": "image",
+                    "bbox": [
+                        127,
+                        300,
+                        480,
+                        532
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                127,
+                                300,
+                                480,
+                                532
+                            ],
+                            "group_id": 1,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        300,
+                                        480,
+                                        532
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                127,
+                                                300,
+                                                480,
+                                                532
+                                            ],
+                                            "score": 0.974,
+                                            "type": "image",
+                                            "image_path": "efa63b3362bf0eec014da42a29ee10be1a5bc69514aec9b20546466115db1b16.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 8,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        300,
+                                        480,
+                                        377.3333333333333
+                                    ],
+                                    "spans": [],
+                                    "index": 7
+                                },
+                                {
+                                    "bbox": [
+                                        127,
+                                        377.3333333333333,
+                                        480,
+                                        454.66666666666663
+                                    ],
+                                    "spans": [],
+                                    "index": 8
+                                },
+                                {
+                                    "bbox": [
+                                        127,
+                                        454.66666666666663,
+                                        480,
+                                        532.0
+                                    ],
+                                    "spans": [],
+                                    "index": 9
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                106,
+                                545,
+                                505,
+                                579
+                            ],
+                            "group_id": 1,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        545,
+                                        505,
+                                        558
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                545,
+                                                505,
+                                                558
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 4: Evaluation using synthesized data: (a) samples from the underlying distribution, (b) samples",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 10
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        556,
+                                        505,
+                                        569
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                556,
+                                                505,
+                                                569
+                                            ],
+                                            "score": 1.0,
+                                            "content": "from the conditional underlying distribution, (c-f) samples from the conditional distribution generate",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 11
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        568,
+                                        249,
+                                        579
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                568,
+                                                249,
+                                                579
+                                            ],
+                                            "score": 1.0,
+                                            "content": "by GI, MisGAN, GAIN, and DAE.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 12
+                                }
+                            ],
+                            "index": 11
+                        }
+                    ],
+                    "index": 9.5
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        614,
+                        195,
+                        627
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                613,
+                                197,
+                                630
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        613,
+                                        197,
+                                        630
+                                    ],
+                                    "score": 1.0,
+                                    "content": "5 CONCLUSION",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        }
+                    ],
+                    "index": 13
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        643,
+                        505,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                644,
+                                506,
+                                657
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        644,
+                                        506,
+                                        657
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In this paper, we proposed a novel method to generate imputations and measure uncertainties over",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                655,
+                                505,
+                                666
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        655,
+                                        505,
+                                        666
+                                    ],
+                                    "score": 1.0,
+                                    "content": "target class assignments based on incomplete feature vectors. We evaluated the effectiveness of the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                666,
+                                506,
+                                678
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        666,
+                                        506,
+                                        678
+                                    ],
+                                    "score": 1.0,
+                                    "content": "suggested approach on image and tabular data via using different measures such as FID distance,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                676,
+                                505,
+                                689
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        676,
+                                        505,
+                                        689
+                                    ],
+                                    "score": 1.0,
+                                    "content": "classification accuracy, and confidence versus accuracy plots. According to the experiments, the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                688,
+                                505,
+                                700
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        688,
+                                        505,
+                                        700
+                                    ],
+                                    "score": 1.0,
+                                    "content": "proposed method not only can generate accurate imputations but also is able to model prediction",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                699,
+                                506,
+                                711
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        699,
+                                        506,
+                                        711
+                                    ],
+                                    "score": 1.0,
+                                    "content": "uncertainties arising from missing values. The proposed method is applicable to many real-world",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                710,
+                                505,
+                                722
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        710,
+                                        505,
+                                        722
+                                    ],
+                                    "score": 1.0,
+                                    "content": "applications where only an incomplete dataset is available, and modeling classification uncertainties",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                721,
+                                163,
+                                734
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        721,
+                                        163,
+                                        734
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is a necessity.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        }
+                    ],
+                    "index": 17.5
+                }
+            ],
+            "page_idx": 9,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        106,
+                        27,
+                        308,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                26,
+                                308,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        26,
+                                        308,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Under review as a conference paper at ICLR 2020",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        300,
+                        751,
+                        311,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                299,
+                                750,
+                                312,
+                                764
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        299,
+                                        750,
+                                        312,
+                                        764
+                                    ],
+                                    "score": 1.0,
+                                    "content": "10",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "image",
+                    "bbox": [
+                        131,
+                        85,
+                        479,
+                        196
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                131,
+                                85,
+                                479,
+                                196
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        131,
+                                        85,
+                                        479,
+                                        196
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                131,
+                                                85,
+                                                479,
+                                                196
+                                            ],
+                                            "score": 0.971,
+                                            "type": "image",
+                                            "image_path": "99ff87491c12b80a83f83e95b0f6597c20ef1cd8b2470028e19f5844c61fce0b.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 1,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        131,
+                                        85,
+                                        479,
+                                        122.0
+                                    ],
+                                    "spans": [],
+                                    "index": 0
+                                },
+                                {
+                                    "bbox": [
+                                        131,
+                                        122.0,
+                                        479,
+                                        159.0
+                                    ],
+                                    "spans": [],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        131,
+                                        159.0,
+                                        479,
+                                        196.0
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                107,
+                                208,
+                                503,
+                                231
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        208,
+                                        506,
+                                        222
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                208,
+                                                506,
+                                                222
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 3: Accuracy versus certainty plots for (a) GI, (b) MisGAN, and (c) GAIN on Landsat dataset",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 3
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        220,
+                                        216,
+                                        232
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                220,
+                                                192,
+                                                232
+                                            ],
+                                            "score": 1.0,
+                                            "content": "at the missing rate of",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                192,
+                                                220,
+                                                212,
+                                                231
+                                            ],
+                                            "score": 0.88,
+                                            "content": "40 \\%",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                212,
+                                                220,
+                                                216,
+                                                232
+                                            ],
+                                            "score": 1.0,
+                                            "content": ".",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 4
+                                }
+                            ],
+                            "index": 3.5
+                        }
+                    ],
+                    "index": 2.25
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        258,
+                        505,
+                        282
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                259,
+                                505,
+                                272
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        259,
+                                        505,
+                                        272
+                                    ],
+                                    "score": 1.0,
+                                    "content": "DAE, as expected, due to its MSE loss term, only captures the expected value of the distribution",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                271,
+                                424,
+                                282
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        271,
+                                        424,
+                                        282
+                                    ],
+                                    "score": 1.0,
+                                    "content": "mean hence reducing the MSE error and generates over-smoothed imputations.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        }
+                    ],
+                    "index": 5.5,
+                    "bbox_fs": [
+                        106,
+                        259,
+                        505,
+                        282
+                    ]
+                },
+                {
+                    "type": "image",
+                    "bbox": [
+                        127,
+                        300,
+                        480,
+                        532
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                127,
+                                300,
+                                480,
+                                532
+                            ],
+                            "group_id": 1,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        300,
+                                        480,
+                                        532
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                127,
+                                                300,
+                                                480,
+                                                532
+                                            ],
+                                            "score": 0.974,
+                                            "type": "image",
+                                            "image_path": "efa63b3362bf0eec014da42a29ee10be1a5bc69514aec9b20546466115db1b16.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 8,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        300,
+                                        480,
+                                        377.3333333333333
+                                    ],
+                                    "spans": [],
+                                    "index": 7
+                                },
+                                {
+                                    "bbox": [
+                                        127,
+                                        377.3333333333333,
+                                        480,
+                                        454.66666666666663
+                                    ],
+                                    "spans": [],
+                                    "index": 8
+                                },
+                                {
+                                    "bbox": [
+                                        127,
+                                        454.66666666666663,
+                                        480,
+                                        532.0
+                                    ],
+                                    "spans": [],
+                                    "index": 9
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                106,
+                                545,
+                                505,
+                                579
+                            ],
+                            "group_id": 1,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        545,
+                                        505,
+                                        558
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                545,
+                                                505,
+                                                558
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 4: Evaluation using synthesized data: (a) samples from the underlying distribution, (b) samples",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 10
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        556,
+                                        505,
+                                        569
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                556,
+                                                505,
+                                                569
+                                            ],
+                                            "score": 1.0,
+                                            "content": "from the conditional underlying distribution, (c-f) samples from the conditional distribution generate",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 11
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        568,
+                                        249,
+                                        579
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                568,
+                                                249,
+                                                579
+                                            ],
+                                            "score": 1.0,
+                                            "content": "by GI, MisGAN, GAIN, and DAE.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 12
+                                }
+                            ],
+                            "index": 11
+                        }
+                    ],
+                    "index": 9.5
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        614,
+                        195,
+                        627
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                613,
+                                197,
+                                630
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        613,
+                                        197,
+                                        630
+                                    ],
+                                    "score": 1.0,
+                                    "content": "5 CONCLUSION",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        }
+                    ],
+                    "index": 13
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        643,
+                        505,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                644,
+                                506,
+                                657
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        644,
+                                        506,
+                                        657
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In this paper, we proposed a novel method to generate imputations and measure uncertainties over",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                655,
+                                505,
+                                666
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        655,
+                                        505,
+                                        666
+                                    ],
+                                    "score": 1.0,
+                                    "content": "target class assignments based on incomplete feature vectors. We evaluated the effectiveness of the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                666,
+                                506,
+                                678
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        666,
+                                        506,
+                                        678
+                                    ],
+                                    "score": 1.0,
+                                    "content": "suggested approach on image and tabular data via using different measures such as FID distance,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                676,
+                                505,
+                                689
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        676,
+                                        505,
+                                        689
+                                    ],
+                                    "score": 1.0,
+                                    "content": "classification accuracy, and confidence versus accuracy plots. According to the experiments, the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                688,
+                                505,
+                                700
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        688,
+                                        505,
+                                        700
+                                    ],
+                                    "score": 1.0,
+                                    "content": "proposed method not only can generate accurate imputations but also is able to model prediction",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                699,
+                                506,
+                                711
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        699,
+                                        506,
+                                        711
+                                    ],
+                                    "score": 1.0,
+                                    "content": "uncertainties arising from missing values. The proposed method is applicable to many real-world",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                710,
+                                505,
+                                722
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        710,
+                                        505,
+                                        722
+                                    ],
+                                    "score": 1.0,
+                                    "content": "applications where only an incomplete dataset is available, and modeling classification uncertainties",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                721,
+                                163,
+                                734
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        721,
+                                        163,
+                                        734
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is a necessity.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        }
+                    ],
+                    "index": 17.5,
+                    "bbox_fs": [
+                        105,
+                        644,
+                        506,
+                        734
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        81,
+                        176,
+                        93
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                82,
+                                176,
+                                94
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        82,
+                                        176,
+                                        94
+                                    ],
+                                    "score": 1.0,
+                                    "content": "REFERENCES",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        }
+                    ],
+                    "index": 0
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        100,
+                        506,
+                        133
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                100,
+                                505,
+                                112
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        100,
+                                        505,
+                                        112
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Theodore W Anderson. Maximum likelihood estimates for a multivariate normal distribution when",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                111,
+                                506,
+                                123
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        111,
+                                        506,
+                                        123
+                                    ],
+                                    "score": 1.0,
+                                    "content": "some observations are missing. Journal of the american Statistical Association, 52(278):200–203,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                121,
+                                142,
+                                133
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        121,
+                                        142,
+                                        133
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1957.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        }
+                    ],
+                    "index": 2
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        141,
+                        504,
+                        164
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                140,
+                                505,
+                                155
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        140,
+                                        505,
+                                        155
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Sanjeev Arora, Andrej Risteski, and Yi Zhang. Do gans learn the distribution? some theory and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                152,
+                                182,
+                                164
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        152,
+                                        182,
+                                        164
+                                    ],
+                                    "score": 1.0,
+                                    "content": "empirics. 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        }
+                    ],
+                    "index": 4.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        171,
+                        505,
+                        195
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                170,
+                                505,
+                                185
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        170,
+                                        505,
+                                        185
+                                    ],
+                                    "score": 1.0,
+                                    "content": "S van Buuren and Karin Groothuis-Oudshoorn. mice: Multivariate imputation by chained equations",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                182,
+                                324,
+                                195
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        182,
+                                        324,
+                                        195
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in r. Journal of statistical software, pp. 1–68, 2010.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        }
+                    ],
+                    "index": 6.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        201,
+                        504,
+                        225
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                200,
+                                506,
+                                216
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        200,
+                                        506,
+                                        216
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Pengfei Chen, Benben Liao, Guangyong Chen, and Shengyu Zhang. Understanding and utilizing",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                213,
+                                470,
+                                226
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        213,
+                                        470,
+                                        226
+                                    ],
+                                    "score": 1.0,
+                                    "content": "deep neural networks trained with noisy labels. arXiv preprint arXiv:1905.05040, 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        }
+                    ],
+                    "index": 8.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        231,
+                        505,
+                        254
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                230,
+                                506,
+                                246
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        230,
+                                        506,
+                                        246
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Adnan Darwiche. Modeling and reasoning with Bayesian networks. Cambridge university press,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                114,
+                                241,
+                                143,
+                                255
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        114,
+                                        241,
+                                        143,
+                                        255
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2009.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        }
+                    ],
+                    "index": 10.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        261,
+                        506,
+                        285
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                260,
+                                508,
+                                276
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        260,
+                                        508,
+                                        276
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Dheeru Dua and Casey Graff. UCI machine learning repository, 2017. URL http://archive.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                273,
+                                205,
+                                285
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        273,
+                                        205,
+                                        285
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ics.uci.edu/ml.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        }
+                    ],
+                    "index": 12.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        291,
+                        505,
+                        316
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                291,
+                                506,
+                                306
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        291,
+                                        506,
+                                        306
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Vincent Dumoulin, Ishmael Belghazi, Ben Poole, Olivier Mastropietro, Alex Lamb, Martin Arjovsky,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                303,
+                                499,
+                                316
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        303,
+                                        499,
+                                        316
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and Aaron Courville. Adversarially learned inference. arXiv preprint arXiv:1606.00704, 2016.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        }
+                    ],
+                    "index": 14.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        322,
+                        505,
+                        356
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                321,
+                                507,
+                                336
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        321,
+                                        507,
+                                        336
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Ian Goodfellow, Jean Pouget-Abadie, Mehdi Mirza, Bing Xu, David Warde-Farley, Sherjil Ozair,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                333,
+                                506,
+                                347
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        333,
+                                        506,
+                                        347
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Aaron Courville, and Yoshua Bengio. Generative adversarial nets. In Advances in neural informa-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                344,
+                                305,
+                                358
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        344,
+                                        305,
+                                        358
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tion processing systems, pp. 2672–2680, 2014.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        }
+                    ],
+                    "index": 17
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        363,
+                        504,
+                        387
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                363,
+                                506,
+                                376
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        363,
+                                        506,
+                                        376
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Trevor Hastie, Robert Tibshirani, Gavin Sherlock, Michael Eisen, Patrick Brown, and David Botstein.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                374,
+                                343,
+                                387
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        374,
+                                        343,
+                                        387
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Imputing missing data for gene expression arrays, 1999.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        }
+                    ],
+                    "index": 19.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        393,
+                        505,
+                        428
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                393,
+                                505,
+                                406
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        393,
+                                        505,
+                                        406
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Martin Heusel, Hubert Ramsauer, Thomas Unterthiner, Bernhard Nessler, and Sepp Hochreiter. Gans",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                405,
+                                505,
+                                417
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        405,
+                                        505,
+                                        417
+                                    ],
+                                    "score": 1.0,
+                                    "content": "trained by a two time-scale update rule converge to a local nash equilibrium. In Advances in Neural",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                415,
+                                339,
+                                429
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        415,
+                                        339,
+                                        429
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Information Processing Systems, pp. 6626–6637, 2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        }
+                    ],
+                    "index": 22
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        434,
+                        505,
+                        458
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                434,
+                                505,
+                                448
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        434,
+                                        505,
+                                        448
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Eric Jang, Shixiang Gu, and Ben Poole. Categorical reparameterization with gumbel-softmax. arXiv",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                446,
+                                254,
+                                458
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        446,
+                                        254,
+                                        458
+                                    ],
+                                    "score": 1.0,
+                                    "content": "preprint arXiv:1611.01144, 2016.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        }
+                    ],
+                    "index": 24.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        464,
+                        506,
+                        498
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                464,
+                                506,
+                                478
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        464,
+                                        506,
+                                        478
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Mohammad Kachuee, Kimmo Karkkainen, Orpaz Goldstein, Davina Zamanzadeh, and Majid Sar-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                475,
+                                507,
+                                489
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        475,
+                                        507,
+                                        489
+                                    ],
+                                    "score": 1.0,
+                                    "content": "rafzadeh. Nutrition and health data for cost-sensitive learning. arXiv preprint arXiv:1902.07102,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                486,
+                                144,
+                                499
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        486,
+                                        144,
+                                        499
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        }
+                    ],
+                    "index": 27
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        506,
+                        505,
+                        529
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                504,
+                                505,
+                                520
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        504,
+                                        505,
+                                        520
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Diederik P Kingma and Jimmy Ba. Adam: A method for stochastic optimization. arXiv preprint",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                516,
+                                214,
+                                529
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        516,
+                                        214,
+                                        529
+                                    ],
+                                    "score": 1.0,
+                                    "content": "arXiv:1412.6980, 2014.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        }
+                    ],
+                    "index": 29.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        536,
+                        505,
+                        559
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                534,
+                                505,
+                                550
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        534,
+                                        505,
+                                        550
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Diederik P Kingma and Max Welling. Auto-encoding variational bayes. arXiv preprint",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                547,
+                                215,
+                                559
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        547,
+                                        215,
+                                        559
+                                    ],
+                                    "score": 1.0,
+                                    "content": "arXiv:1312.6114, 2013.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        }
+                    ],
+                    "index": 31.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        566,
+                        505,
+                        590
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                565,
+                                506,
+                                581
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        565,
+                                        506,
+                                        581
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Alex Krizhevsky and Geoffrey Hinton. Learning multiple layers of features from tiny images.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                578,
+                                248,
+                                590
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        578,
+                                        248,
+                                        590
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Technical report, Citeseer, 2009.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        }
+                    ],
+                    "index": 33.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        596,
+                        505,
+                        631
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                596,
+                                505,
+                                610
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        596,
+                                        505,
+                                        610
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Steven Cheng-Xian Li, Bo Jiang, and Benjamin Marlin. Learning from incomplete data with",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                608,
+                                507,
+                                621
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        608,
+                                        507,
+                                        621
+                                    ],
+                                    "score": 1.0,
+                                    "content": "generative adversarial networks. In International Conference on Learning Representations, 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                618,
+                                396,
+                                632
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        618,
+                                        325,
+                                        632
+                                    ],
+                                    "score": 1.0,
+                                    "content": "URL https://openreview.net/forum?id",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        325,
+                                        621,
+                                        331,
+                                        628
+                                    ],
+                                    "score": 0.45,
+                                    "content": "=",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        332,
+                                        618,
+                                        396,
+                                        632
+                                    ],
+                                    "score": 1.0,
+                                    "content": "S1lDV3RcKm.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        }
+                    ],
+                    "index": 36
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        637,
+                        505,
+                        660
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                636,
+                                506,
+                                652
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        636,
+                                        506,
+                                        652
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Roderick JA Little and Donald B Rubin. Statistical analysis with missing data, volume 793. Wiley,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                114,
+                                648,
+                                144,
+                                662
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        114,
+                                        648,
+                                        144,
+                                        662
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        }
+                    ],
+                    "index": 38.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        668,
+                        504,
+                        702
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                667,
+                                506,
+                                681
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        667,
+                                        506,
+                                        681
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Shuang Liu, Olivier Bousquet, and Kamalika Chaudhuri. Approximation and convergence properties",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                678,
+                                506,
+                                693
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        678,
+                                        506,
+                                        693
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of generative adversarial learning. In Advances in Neural Information Processing Systems, pp.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                691,
+                                191,
+                                702
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        691,
+                                        191,
+                                        702
+                                    ],
+                                    "score": 1.0,
+                                    "content": "5545–5553, 2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        }
+                    ],
+                    "index": 41
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        709,
+                        503,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                709,
+                                505,
+                                723
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        709,
+                                        505,
+                                        723
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Pierre-Alexandre Mattei and Jes Frellsen. missiwae: Deep generative modelling and imputation of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                720,
+                                349,
+                                732
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        720,
+                                        349,
+                                        732
+                                    ],
+                                    "score": 1.0,
+                                    "content": "incomplete data. arXiv preprint arXiv:1812.02633, 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        }
+                    ],
+                    "index": 43.5
+                }
+            ],
+            "page_idx": 10,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        107,
+                        27,
+                        308,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                107,
+                                26,
+                                308,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        26,
+                                        308,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Under review as a conference paper at ICLR 2020",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        300,
+                        751,
+                        309,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                299,
+                                750,
+                                312,
+                                765
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        299,
+                                        750,
+                                        312,
+                                        765
+                                    ],
+                                    "score": 1.0,
+                                    "content": "",
+                                    "type": "text",
+                                    "height": 15,
+                                    "width": 13
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        81,
+                        176,
+                        93
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                82,
+                                176,
+                                94
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        82,
+                                        176,
+                                        94
+                                    ],
+                                    "score": 1.0,
+                                    "content": "REFERENCES",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        }
+                    ],
+                    "index": 0
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        100,
+                        506,
+                        133
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                100,
+                                505,
+                                112
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        100,
+                                        505,
+                                        112
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Theodore W Anderson. Maximum likelihood estimates for a multivariate normal distribution when",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                111,
+                                506,
+                                123
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        111,
+                                        506,
+                                        123
+                                    ],
+                                    "score": 1.0,
+                                    "content": "some observations are missing. Journal of the american Statistical Association, 52(278):200–203,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                121,
+                                142,
+                                133
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        121,
+                                        142,
+                                        133
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1957.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        }
+                    ],
+                    "index": 2,
+                    "bbox_fs": [
+                        106,
+                        100,
+                        506,
+                        133
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        141,
+                        504,
+                        164
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                140,
+                                505,
+                                155
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        140,
+                                        505,
+                                        155
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Sanjeev Arora, Andrej Risteski, and Yi Zhang. Do gans learn the distribution? some theory and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                152,
+                                182,
+                                164
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        152,
+                                        182,
+                                        164
+                                    ],
+                                    "score": 1.0,
+                                    "content": "empirics. 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        }
+                    ],
+                    "index": 4.5,
+                    "bbox_fs": [
+                        105,
+                        140,
+                        505,
+                        164
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        171,
+                        505,
+                        195
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                170,
+                                505,
+                                185
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        170,
+                                        505,
+                                        185
+                                    ],
+                                    "score": 1.0,
+                                    "content": "S van Buuren and Karin Groothuis-Oudshoorn. mice: Multivariate imputation by chained equations",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                182,
+                                324,
+                                195
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        182,
+                                        324,
+                                        195
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in r. Journal of statistical software, pp. 1–68, 2010.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        }
+                    ],
+                    "index": 6.5,
+                    "bbox_fs": [
+                        105,
+                        170,
+                        505,
+                        195
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        201,
+                        504,
+                        225
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                200,
+                                506,
+                                216
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        200,
+                                        506,
+                                        216
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Pengfei Chen, Benben Liao, Guangyong Chen, and Shengyu Zhang. Understanding and utilizing",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                213,
+                                470,
+                                226
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        213,
+                                        470,
+                                        226
+                                    ],
+                                    "score": 1.0,
+                                    "content": "deep neural networks trained with noisy labels. arXiv preprint arXiv:1905.05040, 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        }
+                    ],
+                    "index": 8.5,
+                    "bbox_fs": [
+                        105,
+                        200,
+                        506,
+                        226
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        231,
+                        505,
+                        254
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                230,
+                                506,
+                                246
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        230,
+                                        506,
+                                        246
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Adnan Darwiche. Modeling and reasoning with Bayesian networks. Cambridge university press,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                114,
+                                241,
+                                143,
+                                255
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        114,
+                                        241,
+                                        143,
+                                        255
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2009.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        }
+                    ],
+                    "index": 10.5,
+                    "bbox_fs": [
+                        105,
+                        230,
+                        506,
+                        255
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        261,
+                        506,
+                        285
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                260,
+                                508,
+                                276
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        260,
+                                        508,
+                                        276
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Dheeru Dua and Casey Graff. UCI machine learning repository, 2017. URL http://archive.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                273,
+                                205,
+                                285
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        273,
+                                        205,
+                                        285
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ics.uci.edu/ml.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        }
+                    ],
+                    "index": 12.5,
+                    "bbox_fs": [
+                        105,
+                        260,
+                        508,
+                        285
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        291,
+                        505,
+                        316
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                291,
+                                506,
+                                306
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        291,
+                                        506,
+                                        306
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Vincent Dumoulin, Ishmael Belghazi, Ben Poole, Olivier Mastropietro, Alex Lamb, Martin Arjovsky,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                303,
+                                499,
+                                316
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        303,
+                                        499,
+                                        316
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and Aaron Courville. Adversarially learned inference. arXiv preprint arXiv:1606.00704, 2016.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        }
+                    ],
+                    "index": 14.5,
+                    "bbox_fs": [
+                        106,
+                        291,
+                        506,
+                        316
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        322,
+                        505,
+                        356
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                321,
+                                507,
+                                336
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        321,
+                                        507,
+                                        336
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Ian Goodfellow, Jean Pouget-Abadie, Mehdi Mirza, Bing Xu, David Warde-Farley, Sherjil Ozair,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                333,
+                                506,
+                                347
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        333,
+                                        506,
+                                        347
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Aaron Courville, and Yoshua Bengio. Generative adversarial nets. In Advances in neural informa-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                344,
+                                305,
+                                358
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        344,
+                                        305,
+                                        358
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tion processing systems, pp. 2672–2680, 2014.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        }
+                    ],
+                    "index": 17,
+                    "bbox_fs": [
+                        105,
+                        321,
+                        507,
+                        358
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        363,
+                        504,
+                        387
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                363,
+                                506,
+                                376
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        363,
+                                        506,
+                                        376
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Trevor Hastie, Robert Tibshirani, Gavin Sherlock, Michael Eisen, Patrick Brown, and David Botstein.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                374,
+                                343,
+                                387
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        374,
+                                        343,
+                                        387
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Imputing missing data for gene expression arrays, 1999.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        }
+                    ],
+                    "index": 19.5,
+                    "bbox_fs": [
+                        105,
+                        363,
+                        506,
+                        387
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        393,
+                        505,
+                        428
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                393,
+                                505,
+                                406
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        393,
+                                        505,
+                                        406
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Martin Heusel, Hubert Ramsauer, Thomas Unterthiner, Bernhard Nessler, and Sepp Hochreiter. Gans",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                405,
+                                505,
+                                417
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        405,
+                                        505,
+                                        417
+                                    ],
+                                    "score": 1.0,
+                                    "content": "trained by a two time-scale update rule converge to a local nash equilibrium. In Advances in Neural",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                415,
+                                339,
+                                429
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        415,
+                                        339,
+                                        429
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Information Processing Systems, pp. 6626–6637, 2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        }
+                    ],
+                    "index": 22,
+                    "bbox_fs": [
+                        105,
+                        393,
+                        505,
+                        429
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        434,
+                        505,
+                        458
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                434,
+                                505,
+                                448
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        434,
+                                        505,
+                                        448
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Eric Jang, Shixiang Gu, and Ben Poole. Categorical reparameterization with gumbel-softmax. arXiv",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                446,
+                                254,
+                                458
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        446,
+                                        254,
+                                        458
+                                    ],
+                                    "score": 1.0,
+                                    "content": "preprint arXiv:1611.01144, 2016.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        }
+                    ],
+                    "index": 24.5,
+                    "bbox_fs": [
+                        105,
+                        434,
+                        505,
+                        458
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        464,
+                        506,
+                        498
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                464,
+                                506,
+                                478
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        464,
+                                        506,
+                                        478
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Mohammad Kachuee, Kimmo Karkkainen, Orpaz Goldstein, Davina Zamanzadeh, and Majid Sar-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                475,
+                                507,
+                                489
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        475,
+                                        507,
+                                        489
+                                    ],
+                                    "score": 1.0,
+                                    "content": "rafzadeh. Nutrition and health data for cost-sensitive learning. arXiv preprint arXiv:1902.07102,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                486,
+                                144,
+                                499
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        486,
+                                        144,
+                                        499
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        }
+                    ],
+                    "index": 27,
+                    "bbox_fs": [
+                        105,
+                        464,
+                        507,
+                        499
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        506,
+                        505,
+                        529
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                504,
+                                505,
+                                520
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        504,
+                                        505,
+                                        520
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Diederik P Kingma and Jimmy Ba. Adam: A method for stochastic optimization. arXiv preprint",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                516,
+                                214,
+                                529
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        516,
+                                        214,
+                                        529
+                                    ],
+                                    "score": 1.0,
+                                    "content": "arXiv:1412.6980, 2014.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        }
+                    ],
+                    "index": 29.5,
+                    "bbox_fs": [
+                        105,
+                        504,
+                        505,
+                        529
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        536,
+                        505,
+                        559
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                534,
+                                505,
+                                550
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        534,
+                                        505,
+                                        550
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Diederik P Kingma and Max Welling. Auto-encoding variational bayes. arXiv preprint",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                547,
+                                215,
+                                559
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        547,
+                                        215,
+                                        559
+                                    ],
+                                    "score": 1.0,
+                                    "content": "arXiv:1312.6114, 2013.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        }
+                    ],
+                    "index": 31.5,
+                    "bbox_fs": [
+                        105,
+                        534,
+                        505,
+                        559
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        566,
+                        505,
+                        590
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                565,
+                                506,
+                                581
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        565,
+                                        506,
+                                        581
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Alex Krizhevsky and Geoffrey Hinton. Learning multiple layers of features from tiny images.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                578,
+                                248,
+                                590
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        578,
+                                        248,
+                                        590
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Technical report, Citeseer, 2009.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        }
+                    ],
+                    "index": 33.5,
+                    "bbox_fs": [
+                        105,
+                        565,
+                        506,
+                        590
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        596,
+                        505,
+                        631
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                596,
+                                505,
+                                610
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        596,
+                                        505,
+                                        610
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Steven Cheng-Xian Li, Bo Jiang, and Benjamin Marlin. Learning from incomplete data with",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                608,
+                                507,
+                                621
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        608,
+                                        507,
+                                        621
+                                    ],
+                                    "score": 1.0,
+                                    "content": "generative adversarial networks. In International Conference on Learning Representations, 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                618,
+                                396,
+                                632
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        618,
+                                        325,
+                                        632
+                                    ],
+                                    "score": 1.0,
+                                    "content": "URL https://openreview.net/forum?id",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        325,
+                                        621,
+                                        331,
+                                        628
+                                    ],
+                                    "score": 0.45,
+                                    "content": "=",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        332,
+                                        618,
+                                        396,
+                                        632
+                                    ],
+                                    "score": 1.0,
+                                    "content": "S1lDV3RcKm.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        }
+                    ],
+                    "index": 36,
+                    "bbox_fs": [
+                        105,
+                        596,
+                        507,
+                        632
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        637,
+                        505,
+                        660
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                636,
+                                506,
+                                652
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        636,
+                                        506,
+                                        652
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Roderick JA Little and Donald B Rubin. Statistical analysis with missing data, volume 793. Wiley,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                114,
+                                648,
+                                144,
+                                662
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        114,
+                                        648,
+                                        144,
+                                        662
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        }
+                    ],
+                    "index": 38.5,
+                    "bbox_fs": [
+                        105,
+                        636,
+                        506,
+                        662
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        668,
+                        504,
+                        702
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                667,
+                                506,
+                                681
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        667,
+                                        506,
+                                        681
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Shuang Liu, Olivier Bousquet, and Kamalika Chaudhuri. Approximation and convergence properties",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                678,
+                                506,
+                                693
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        678,
+                                        506,
+                                        693
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of generative adversarial learning. In Advances in Neural Information Processing Systems, pp.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                691,
+                                191,
+                                702
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        691,
+                                        191,
+                                        702
+                                    ],
+                                    "score": 1.0,
+                                    "content": "5545–5553, 2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        }
+                    ],
+                    "index": 41,
+                    "bbox_fs": [
+                        105,
+                        667,
+                        506,
+                        702
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        709,
+                        503,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                709,
+                                505,
+                                723
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        709,
+                                        505,
+                                        723
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Pierre-Alexandre Mattei and Jes Frellsen. missiwae: Deep generative modelling and imputation of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                720,
+                                349,
+                                732
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        720,
+                                        349,
+                                        732
+                                    ],
+                                    "score": 1.0,
+                                    "content": "incomplete data. arXiv preprint arXiv:1812.02633, 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        }
+                    ],
+                    "index": 43.5,
+                    "bbox_fs": [
+                        106,
+                        709,
+                        505,
+                        732
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        82,
+                        505,
+                        106
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                82,
+                                505,
+                                95
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        82,
+                                        505,
+                                        95
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Takeru Miyato, Toshiki Kataoka, Masanori Koyama, and Yuichi Yoshida. Spectral normalization for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                94,
+                                411,
+                                106
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        94,
+                                        411,
+                                        106
+                                    ],
+                                    "score": 1.0,
+                                    "content": "generative adversarial networks. arXiv preprint arXiv:1802.05957, 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        }
+                    ],
+                    "index": 0.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        111,
+                        504,
+                        136
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                112,
+                                505,
+                                125
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        112,
+                                        505,
+                                        125
+                                    ],
+                                    "score": 1.0,
+                                    "content": "George B Moody and Roger G Mark. The impact of the mit-bih arrhythmia database. IEEE",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                123,
+                                393,
+                                136
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        123,
+                                        393,
+                                        136
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Engineering in Medicine and Biology Magazine, 20(3):45–50, 2001.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        }
+                    ],
+                    "index": 2.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        142,
+                        505,
+                        165
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                142,
+                                505,
+                                155
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        142,
+                                        505,
+                                        155
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Jared S Murray et al. Multiple imputation: a review of practical and theoretical findings. Statistical",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                153,
+                                241,
+                                165
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        153,
+                                        241,
+                                        165
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Science, 33(2):142–159, 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        }
+                    ],
+                    "index": 4.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        171,
+                        504,
+                        195
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                172,
+                                505,
+                                185
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        172,
+                                        505,
+                                        185
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Nagarajan Natarajan, Inderjit S Dhillon, Pradeep K Ravikumar, and Ambuj Tewari. Learning with",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                183,
+                                480,
+                                195
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        183,
+                                        480,
+                                        195
+                                    ],
+                                    "score": 1.0,
+                                    "content": "noisy labels. In Advances in neural information processing systems, pp. 1196–1204, 2013.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        }
+                    ],
+                    "index": 6.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        201,
+                        504,
+                        225
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                200,
+                                505,
+                                215
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        200,
+                                        505,
+                                        215
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Alfredo Nazabal, Pablo M Olmos, Zoubin Ghahramani, and Isabel Valera. Handling incomplete",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                213,
+                                405,
+                                225
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        213,
+                                        405,
+                                        225
+                                    ],
+                                    "score": 1.0,
+                                    "content": "heterogeneous data using vaes. arXiv preprint arXiv:1807.03653, 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        }
+                    ],
+                    "index": 8.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        231,
+                        505,
+                        254
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                230,
+                                506,
+                                245
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        230,
+                                        506,
+                                        245
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Richard E Neapolitan et al. Learning bayesian networks, volume 38. Pearson Prentice Hall Upper",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                243,
+                                214,
+                                255
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        243,
+                                        214,
+                                        255
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Saddle River, NJ, 2004.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        }
+                    ],
+                    "index": 10.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        261,
+                        504,
+                        285
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                259,
+                                506,
+                                276
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        259,
+                                        506,
+                                        276
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Thomas Dyhre Nielsen and Finn Verner Jensen. Bayesian networks and decision graphs. Springer",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                117,
+                                273,
+                                253,
+                                285
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        117,
+                                        273,
+                                        253,
+                                        285
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Science & Business Media, 2009.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        }
+                    ],
+                    "index": 12.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        291,
+                        504,
+                        325
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                291,
+                                506,
+                                304
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        291,
+                                        506,
+                                        304
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Scott Reed, Honglak Lee, Dragomir Anguelov, Christian Szegedy, Dumitru Erhan, and Andrew",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                302,
+                                505,
+                                316
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        302,
+                                        505,
+                                        316
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Rabinovich. Training deep neural networks on noisy labels with bootstrapping. arXiv preprint",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                314,
+                                214,
+                                324
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        314,
+                                        214,
+                                        324
+                                    ],
+                                    "score": 1.0,
+                                    "content": "arXiv:1412.6596, 2014.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        }
+                    ],
+                    "index": 15
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        332,
+                        429,
+                        345
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                331,
+                                429,
+                                345
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        331,
+                                        429,
+                                        345
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Donald B Rubin. Inference and missing data. Biometrika, 63(3):581–592, 1976.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        }
+                    ],
+                    "index": 17
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        351,
+                        504,
+                        374
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                350,
+                                506,
+                                365
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        350,
+                                        506,
+                                        365
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Joseph L Schafer and John W Graham. Missing data: our view of the state of the art. Psychological",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                362,
+                                219,
+                                374
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        362,
+                                        219,
+                                        374
+                                    ],
+                                    "score": 1.0,
+                                    "content": "methods, 7(2):147, 2002.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        }
+                    ],
+                    "index": 18.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        381,
+                        505,
+                        415
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                380,
+                                505,
+                                395
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        380,
+                                        505,
+                                        395
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Kihyuk Sohn, Honglak Lee, and Xinchen Yan. Learning structured output representation using",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                391,
+                                506,
+                                406
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        391,
+                                        506,
+                                        406
+                                    ],
+                                    "score": 1.0,
+                                    "content": "deep conditional generative models. In Advances in neural information processing systems, pp.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                403,
+                                192,
+                                414
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        403,
+                                        192,
+                                        414
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3483–3491, 2015.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        }
+                    ],
+                    "index": 21
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        421,
+                        506,
+                        456
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                421,
+                                506,
+                                435
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        421,
+                                        506,
+                                        435
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Pascal Vincent, Hugo Larochelle, Yoshua Bengio, and Pierre-Antoine Manzagol. Extracting and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                433,
+                                505,
+                                446
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        433,
+                                        505,
+                                        446
+                                    ],
+                                    "score": 1.0,
+                                    "content": "composing robust features with denoising autoencoders. In Proceedings of the 25th international",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                444,
+                                367,
+                                457
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        444,
+                                        367,
+                                        457
+                                    ],
+                                    "score": 1.0,
+                                    "content": "conference on Machine learning, pp. 1096–1103. ACM, 2008.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        }
+                    ],
+                    "index": 24
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        462,
+                        506,
+                        497
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                462,
+                                506,
+                                476
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        462,
+                                        506,
+                                        476
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Ting-Chun Wang, Ming-Yu Liu, Jun-Yan Zhu, Andrew Tao, Jan Kautz, and Bryan Catanzaro. High-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                474,
+                                506,
+                                488
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        474,
+                                        506,
+                                        488
+                                    ],
+                                    "score": 1.0,
+                                    "content": "resolution image synthesis and semantic manipulation with conditional gans. In Proceedings of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                485,
+                                477,
+                                497
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        485,
+                                        477,
+                                        497
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the IEEE Conference on Computer Vision and Pattern Recognition, pp. 8798–8807, 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        }
+                    ],
+                    "index": 27
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        503,
+                        504,
+                        527
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                501,
+                                506,
+                                519
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        501,
+                                        506,
+                                        519
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Jinsung Yoon, James Jordon, and Mihaela Van Der Schaar. Gain: Missing data imputation using",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                515,
+                                390,
+                                527
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        515,
+                                        390,
+                                        527
+                                    ],
+                                    "score": 1.0,
+                                    "content": "generative adversarial nets. arXiv preprint arXiv:1806.02920, 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        }
+                    ],
+                    "index": 29.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        533,
+                        504,
+                        556
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                533,
+                                505,
+                                547
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        533,
+                                        505,
+                                        547
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Han Zhang, Ian Goodfellow, Dimitris Metaxas, and Augustus Odena. Self-attention generative",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                544,
+                                367,
+                                557
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        544,
+                                        367,
+                                        557
+                                    ],
+                                    "score": 1.0,
+                                    "content": "adversarial networks. arXiv preprint arXiv:1805.08318, 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        }
+                    ],
+                    "index": 31.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        563,
+                        505,
+                        597
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                561,
+                                506,
+                                578
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        561,
+                                        506,
+                                        578
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Shichao Zhang, Zhenxing Qin, Charles X Ling, and Shengli Sheng. \" missing is useful\": missing",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                575,
+                                505,
+                                587
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        575,
+                                        505,
+                                        587
+                                    ],
+                                    "score": 1.0,
+                                    "content": "values in cost-sensitive decision trees. IEEE transactions on knowledge and data engineering, 17",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                586,
+                                210,
+                                597
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        586,
+                                        210,
+                                        597
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(12):1689–1693, 2005.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        }
+                    ],
+                    "index": 34
+                }
+            ],
+            "page_idx": 11,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        107,
+                        27,
+                        308,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                107,
+                                26,
+                                308,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        26,
+                                        308,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Under review as a conference paper at ICLR 2020",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        300,
+                        751,
+                        311,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                299,
+                                750,
+                                312,
+                                764
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        299,
+                                        750,
+                                        312,
+                                        764
+                                    ],
+                                    "score": 1.0,
+                                    "content": "12",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        82,
+                        505,
+                        106
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                82,
+                                505,
+                                95
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        82,
+                                        505,
+                                        95
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Takeru Miyato, Toshiki Kataoka, Masanori Koyama, and Yuichi Yoshida. Spectral normalization for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                94,
+                                411,
+                                106
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        94,
+                                        411,
+                                        106
+                                    ],
+                                    "score": 1.0,
+                                    "content": "generative adversarial networks. arXiv preprint arXiv:1802.05957, 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        }
+                    ],
+                    "index": 0.5,
+                    "bbox_fs": [
+                        106,
+                        82,
+                        505,
+                        106
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        111,
+                        504,
+                        136
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                112,
+                                505,
+                                125
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        112,
+                                        505,
+                                        125
+                                    ],
+                                    "score": 1.0,
+                                    "content": "George B Moody and Roger G Mark. The impact of the mit-bih arrhythmia database. IEEE",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                123,
+                                393,
+                                136
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        123,
+                                        393,
+                                        136
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Engineering in Medicine and Biology Magazine, 20(3):45–50, 2001.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        }
+                    ],
+                    "index": 2.5,
+                    "bbox_fs": [
+                        106,
+                        112,
+                        505,
+                        136
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        142,
+                        505,
+                        165
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                142,
+                                505,
+                                155
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        142,
+                                        505,
+                                        155
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Jared S Murray et al. Multiple imputation: a review of practical and theoretical findings. Statistical",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                153,
+                                241,
+                                165
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        153,
+                                        241,
+                                        165
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Science, 33(2):142–159, 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        }
+                    ],
+                    "index": 4.5,
+                    "bbox_fs": [
+                        105,
+                        142,
+                        505,
+                        165
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        171,
+                        504,
+                        195
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                172,
+                                505,
+                                185
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        172,
+                                        505,
+                                        185
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Nagarajan Natarajan, Inderjit S Dhillon, Pradeep K Ravikumar, and Ambuj Tewari. Learning with",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                183,
+                                480,
+                                195
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        183,
+                                        480,
+                                        195
+                                    ],
+                                    "score": 1.0,
+                                    "content": "noisy labels. In Advances in neural information processing systems, pp. 1196–1204, 2013.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        }
+                    ],
+                    "index": 6.5,
+                    "bbox_fs": [
+                        106,
+                        172,
+                        505,
+                        195
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        201,
+                        504,
+                        225
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                200,
+                                505,
+                                215
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        200,
+                                        505,
+                                        215
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Alfredo Nazabal, Pablo M Olmos, Zoubin Ghahramani, and Isabel Valera. Handling incomplete",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                213,
+                                405,
+                                225
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        213,
+                                        405,
+                                        225
+                                    ],
+                                    "score": 1.0,
+                                    "content": "heterogeneous data using vaes. arXiv preprint arXiv:1807.03653, 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        }
+                    ],
+                    "index": 8.5,
+                    "bbox_fs": [
+                        106,
+                        200,
+                        505,
+                        225
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        231,
+                        505,
+                        254
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                230,
+                                506,
+                                245
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        230,
+                                        506,
+                                        245
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Richard E Neapolitan et al. Learning bayesian networks, volume 38. Pearson Prentice Hall Upper",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                243,
+                                214,
+                                255
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        243,
+                                        214,
+                                        255
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Saddle River, NJ, 2004.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        }
+                    ],
+                    "index": 10.5,
+                    "bbox_fs": [
+                        105,
+                        230,
+                        506,
+                        255
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        261,
+                        504,
+                        285
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                259,
+                                506,
+                                276
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        259,
+                                        506,
+                                        276
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Thomas Dyhre Nielsen and Finn Verner Jensen. Bayesian networks and decision graphs. Springer",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                117,
+                                273,
+                                253,
+                                285
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        117,
+                                        273,
+                                        253,
+                                        285
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Science & Business Media, 2009.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        }
+                    ],
+                    "index": 12.5,
+                    "bbox_fs": [
+                        105,
+                        259,
+                        506,
+                        285
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        291,
+                        504,
+                        325
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                291,
+                                506,
+                                304
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        291,
+                                        506,
+                                        304
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Scott Reed, Honglak Lee, Dragomir Anguelov, Christian Szegedy, Dumitru Erhan, and Andrew",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                302,
+                                505,
+                                316
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        302,
+                                        505,
+                                        316
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Rabinovich. Training deep neural networks on noisy labels with bootstrapping. arXiv preprint",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                314,
+                                214,
+                                324
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        314,
+                                        214,
+                                        324
+                                    ],
+                                    "score": 1.0,
+                                    "content": "arXiv:1412.6596, 2014.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        }
+                    ],
+                    "index": 15,
+                    "bbox_fs": [
+                        105,
+                        291,
+                        506,
+                        324
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        332,
+                        429,
+                        345
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                331,
+                                429,
+                                345
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        331,
+                                        429,
+                                        345
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Donald B Rubin. Inference and missing data. Biometrika, 63(3):581–592, 1976.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        }
+                    ],
+                    "index": 17,
+                    "bbox_fs": [
+                        106,
+                        331,
+                        429,
+                        345
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        351,
+                        504,
+                        374
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                350,
+                                506,
+                                365
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        350,
+                                        506,
+                                        365
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Joseph L Schafer and John W Graham. Missing data: our view of the state of the art. Psychological",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                362,
+                                219,
+                                374
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        362,
+                                        219,
+                                        374
+                                    ],
+                                    "score": 1.0,
+                                    "content": "methods, 7(2):147, 2002.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        }
+                    ],
+                    "index": 18.5,
+                    "bbox_fs": [
+                        105,
+                        350,
+                        506,
+                        374
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        381,
+                        505,
+                        415
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                380,
+                                505,
+                                395
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        380,
+                                        505,
+                                        395
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Kihyuk Sohn, Honglak Lee, and Xinchen Yan. Learning structured output representation using",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                391,
+                                506,
+                                406
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        391,
+                                        506,
+                                        406
+                                    ],
+                                    "score": 1.0,
+                                    "content": "deep conditional generative models. In Advances in neural information processing systems, pp.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                403,
+                                192,
+                                414
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        403,
+                                        192,
+                                        414
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3483–3491, 2015.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        }
+                    ],
+                    "index": 21,
+                    "bbox_fs": [
+                        105,
+                        380,
+                        506,
+                        414
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        421,
+                        506,
+                        456
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                421,
+                                506,
+                                435
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        421,
+                                        506,
+                                        435
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Pascal Vincent, Hugo Larochelle, Yoshua Bengio, and Pierre-Antoine Manzagol. Extracting and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                433,
+                                505,
+                                446
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        433,
+                                        505,
+                                        446
+                                    ],
+                                    "score": 1.0,
+                                    "content": "composing robust features with denoising autoencoders. In Proceedings of the 25th international",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                444,
+                                367,
+                                457
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        444,
+                                        367,
+                                        457
+                                    ],
+                                    "score": 1.0,
+                                    "content": "conference on Machine learning, pp. 1096–1103. ACM, 2008.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        }
+                    ],
+                    "index": 24,
+                    "bbox_fs": [
+                        105,
+                        421,
+                        506,
+                        457
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        462,
+                        506,
+                        497
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                462,
+                                506,
+                                476
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        462,
+                                        506,
+                                        476
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Ting-Chun Wang, Ming-Yu Liu, Jun-Yan Zhu, Andrew Tao, Jan Kautz, and Bryan Catanzaro. High-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                474,
+                                506,
+                                488
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        474,
+                                        506,
+                                        488
+                                    ],
+                                    "score": 1.0,
+                                    "content": "resolution image synthesis and semantic manipulation with conditional gans. In Proceedings of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                485,
+                                477,
+                                497
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        485,
+                                        477,
+                                        497
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the IEEE Conference on Computer Vision and Pattern Recognition, pp. 8798–8807, 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        }
+                    ],
+                    "index": 27,
+                    "bbox_fs": [
+                        106,
+                        462,
+                        506,
+                        497
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        503,
+                        504,
+                        527
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                501,
+                                506,
+                                519
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        501,
+                                        506,
+                                        519
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Jinsung Yoon, James Jordon, and Mihaela Van Der Schaar. Gain: Missing data imputation using",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                515,
+                                390,
+                                527
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        515,
+                                        390,
+                                        527
+                                    ],
+                                    "score": 1.0,
+                                    "content": "generative adversarial nets. arXiv preprint arXiv:1806.02920, 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        }
+                    ],
+                    "index": 29.5,
+                    "bbox_fs": [
+                        104,
+                        501,
+                        506,
+                        527
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        533,
+                        504,
+                        556
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                533,
+                                505,
+                                547
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        533,
+                                        505,
+                                        547
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Han Zhang, Ian Goodfellow, Dimitris Metaxas, and Augustus Odena. Self-attention generative",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                544,
+                                367,
+                                557
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        544,
+                                        367,
+                                        557
+                                    ],
+                                    "score": 1.0,
+                                    "content": "adversarial networks. arXiv preprint arXiv:1805.08318, 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        }
+                    ],
+                    "index": 31.5,
+                    "bbox_fs": [
+                        105,
+                        533,
+                        505,
+                        557
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        563,
+                        505,
+                        597
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                561,
+                                506,
+                                578
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        561,
+                                        506,
+                                        578
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Shichao Zhang, Zhenxing Qin, Charles X Ling, and Shengli Sheng. \" missing is useful\": missing",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                575,
+                                505,
+                                587
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        575,
+                                        505,
+                                        587
+                                    ],
+                                    "score": 1.0,
+                                    "content": "values in cost-sensitive decision trees. IEEE transactions on knowledge and data engineering, 17",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                586,
+                                210,
+                                597
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        586,
+                                        210,
+                                        597
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(12):1689–1693, 2005.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        }
+                    ],
+                    "index": 34,
+                    "bbox_fs": [
+                        105,
+                        561,
+                        506,
+                        597
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        81,
+                        325,
+                        94
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                81,
+                                325,
+                                95
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        81,
+                                        325,
+                                        95
+                                    ],
+                                    "score": 1.0,
+                                    "content": "A IMPLEMENTATION AND EXPERIMENTS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        }
+                    ],
+                    "index": 0
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        108,
+                        505,
+                        153
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                108,
+                                506,
+                                121
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        108,
+                                        506,
+                                        121
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Table 3 presents the list of software dependencies and versions used in our implementation. To",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                120,
+                                505,
+                                131
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        120,
+                                        505,
+                                        131
+                                    ],
+                                    "score": 1.0,
+                                    "content": "produce results related to this paper, we used a workstation with 4 NVIDIA GeForce RTX-2080Ti",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                130,
+                                506,
+                                144
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        130,
+                                        506,
+                                        144
+                                    ],
+                                    "score": 1.0,
+                                    "content": "GPUs, a 12 core Intel Core i9-7920X processor, and 128 GB memory. Each experiment took between",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                141,
+                                385,
+                                154
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        141,
+                                        385,
+                                        154
+                                    ],
+                                    "score": 1.0,
+                                    "content": "about 4 hours to 48 hours, based on the task and method being tested.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        }
+                    ],
+                    "index": 2.5
+                },
+                {
+                    "type": "table",
+                    "bbox": [
+                        237,
+                        188,
+                        378,
+                        381
+                    ],
+                    "blocks": [
+                        {
+                            "type": "table_caption",
+                            "bbox": [
+                                240,
+                                167,
+                                371,
+                                179
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        239,
+                                        165,
+                                        372,
+                                        181
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                239,
+                                                165,
+                                                372,
+                                                181
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Table 3: Software dependencies.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 5
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "type": "table_body",
+                            "bbox": [
+                                237,
+                                188,
+                                378,
+                                381
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        237,
+                                        188,
+                                        378,
+                                        381
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                237,
+                                                188,
+                                                378,
+                                                381
+                                            ],
+                                            "score": 0.977,
+                                            "html": "<table><tr><td>Dependency</td><td>Version</td></tr><tr><td>python</td><td>3.7.1</td></tr><tr><td>pytorch</td><td>1.0.0</td></tr><tr><td>cuda100</td><td>1.0</td></tr><tr><td>ipython</td><td>7.2.0</td></tr><tr><td>jupyter</td><td>1.0.0</td></tr><tr><td>numpy</td><td>1.15.4</td></tr><tr><td>pandas</td><td>0.24.1</td></tr><tr><td>scikit-learn</td><td>0.20.1</td></tr><tr><td>scipy</td><td>1.1.0</td></tr><tr><td>torchvision</td><td>0.2.1</td></tr><tr><td>tqdm</td><td>4.28.1</td></tr><tr><td>matplotlib</td><td>3.0.1</td></tr></table>",
+                                            "type": "table",
+                                            "image_path": "c34dd7ed960160ceeeb9080e4ac592a923c1d401524807a6a19321bc3e36daa6.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 6,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        237,
+                                        188,
+                                        378,
+                                        381
+                                    ],
+                                    "spans": [],
+                                    "index": 6
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 5.5
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        412,
+                        272,
+                        425
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                411,
+                                273,
+                                426
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        411,
+                                        273,
+                                        426
+                                    ],
+                                    "score": 1.0,
+                                    "content": "B NETWORK ARCHITECTURES",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        }
+                    ],
+                    "index": 7
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        439,
+                        506,
+                        528
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                439,
+                                505,
+                                452
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        439,
+                                        505,
+                                        452
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Table 4 shows the exact architectures used in this paper. To show each layer or block we used the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                450,
+                                505,
+                                464
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        450,
+                                        184,
+                                        464
+                                    ],
+                                    "score": 1.0,
+                                    "content": "following notation.",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        185,
+                                        452,
+                                        233,
+                                        462
+                                    ],
+                                    "score": 0.72,
+                                    "content": "\\mathtt { C x S y P z - t }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        234,
+                                        450,
+                                        424,
+                                        464
+                                    ],
+                                    "score": 1.0,
+                                    "content": "represents a 2-d convolution layer of kernel size",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        425,
+                                        452,
+                                        432,
+                                        461
+                                    ],
+                                    "score": 0.43,
+                                    "content": "_ \\textrm { x }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        433,
+                                        450,
+                                        505,
+                                        464
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", stride y, padding",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                107,
+                                462,
+                                505,
+                                474
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        464,
+                                        113,
+                                        472
+                                    ],
+                                    "score": 0.31,
+                                    "content": "_ { \\textrm { Z } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        114,
+                                        462,
+                                        243,
+                                        474
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", and number of output channels",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        243,
+                                        463,
+                                        251,
+                                        472
+                                    ],
+                                    "score": 0.41,
+                                    "content": "\\scriptstyle \\pm",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        251,
+                                        462,
+                                        505,
+                                        474
+                                    ],
+                                    "score": 1.0,
+                                    "content": "followed by ReLU activation. Attn represents a self-attention",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                473,
+                                505,
+                                485
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        473,
+                                        229,
+                                        485
+                                    ],
+                                    "score": 1.0,
+                                    "content": "layer similar to Zhang et al.9.",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        230,
+                                        473,
+                                        249,
+                                        483
+                                    ],
+                                    "score": 0.74,
+                                    "content": "_ \\mathrm { R - x }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        250,
+                                        473,
+                                        505,
+                                        485
+                                    ],
+                                    "score": 1.0,
+                                    "content": "represents a residual block consisting of two 2-d convolutions",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                484,
+                                505,
+                                496
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        484,
+                                        425,
+                                        496
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with kernel size 3 (padding size 1), batch normalization, and ReLU activation.",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        425,
+                                        484,
+                                        479,
+                                        495
+                                    ],
+                                    "score": 0.87,
+                                    "content": "\\mathsf { C T x S y P z - t }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        480,
+                                        484,
+                                        505,
+                                        496
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                494,
+                                506,
+                                507
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        494,
+                                        266,
+                                        507
+                                    ],
+                                    "score": 1.0,
+                                    "content": "convolution transpose corresponding to",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        267,
+                                        496,
+                                        316,
+                                        506
+                                    ],
+                                    "score": 0.64,
+                                    "content": "\\mathtt { C x S y P z - t }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        316,
+                                        494,
+                                        319,
+                                        507
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        320,
+                                        495,
+                                        345,
+                                        505
+                                    ],
+                                    "score": 0.65,
+                                    "content": "\\operatorname { F C - x }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        346,
+                                        494,
+                                        506,
+                                        507
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is representing a linear fully-connected",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                505,
+                                506,
+                                518
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        505,
+                                        139,
+                                        518
+                                    ],
+                                    "score": 1.0,
+                                    "content": "layer of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        139,
+                                        507,
+                                        147,
+                                        516
+                                    ],
+                                    "score": 0.4,
+                                    "content": "_ \\textrm { x }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        147,
+                                        505,
+                                        506,
+                                        518
+                                    ],
+                                    "score": 1.0,
+                                    "content": "output neurons with biases. We use spectral normalization as suggested by Miyato et al.10",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                517,
+                                400,
+                                528
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        517,
+                                        400,
+                                        528
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for all convolutional layers in both generator and discriminator networks.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        }
+                    ],
+                    "index": 11.5
+                },
+                {
+                    "type": "table",
+                    "bbox": [
+                        107,
+                        564,
+                        503,
+                        684
+                    ],
+                    "blocks": [
+                        {
+                            "type": "table_caption",
+                            "bbox": [
+                                192,
+                                542,
+                                418,
+                                554
+                            ],
+                            "group_id": 1,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        191,
+                                        540,
+                                        419,
+                                        555
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                191,
+                                                540,
+                                                419,
+                                                555
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Table 4: Network architectures used in our experiments.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 16
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "type": "table_body",
+                            "bbox": [
+                                107,
+                                564,
+                                503,
+                                684
+                            ],
+                            "group_id": 1,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        564,
+                                        503,
+                                        684
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                107,
+                                                564,
+                                                503,
+                                                684
+                                            ],
+                                            "score": 0.982,
+                                            "html": "<table><tr><td rowspan=1 colspan=1>Dataset</td><td rowspan=1 colspan=1>Generator/Discriminator Architecture</td><td rowspan=1 colspan=1>Predictor Architecture</td></tr><tr><td rowspan=1 colspan=1>CIFAR-10</td><td rowspan=1 colspan=1>C7S1P3-64，C3S2P1-128，Attn，R-128,Attn，R-128，Attn，R-128，Attn,R-128，CT3S2P1-128，CT7S1P3-3，Tanh/Sigmoid</td><td rowspan=1 colspan=1>ResNet-18 11,12</td></tr><tr><td rowspan=1 colspan=1>Landsat</td><td rowspan=1 colspan=1>FC-64，Sigmoid，BNorm，FC-64，Sigmoid,BNorm,FC-64，Sigmoid，BNorm，FC-36，Tanh/Sigmoid</td><td rowspan=1 colspan=1>FC-64，ReLU，BNorm，FC-64,ReLU，BNorm，FC-6，Softmax</td></tr><tr><td rowspan=1 colspan=1>MIT-BIH</td><td rowspan=1 colspan=1>FC-1860，ReLU，BNorm，FC-1860，ReLU，BNorm,FC-1860，ReLU，BNorm, FC-186,Tanh/Sigmoid</td><td rowspan=1 colspan=1>FC-1860，ReLU，BNorm，FC-1860,ReLU，BNorm，FC-5，Softmax</td></tr><tr><td rowspan=1 colspan=1>Diabetes</td><td rowspan=1 colspan=1>FC-45，ReLU，BNorm，FC-45，ReLU，BNorm,FC-45，ReLU，BNorm，FC-45，Tanh/Sigmoid</td><td rowspan=1 colspan=1>FC-22，ReLU，BNorm，FC-22,ReLU，BNorm，FC-3，Softmax</td></tr></table>",
+                                            "type": "table",
+                                            "image_path": "1dc4ede59e46fd6fe85ec060a8aeed53fa3365bbe7e3af9f8c634bf3d02c9b6d.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 18,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        564,
+                                        503,
+                                        604.0
+                                    ],
+                                    "spans": [],
+                                    "index": 17
+                                },
+                                {
+                                    "bbox": [
+                                        107,
+                                        604.0,
+                                        503,
+                                        644.0
+                                    ],
+                                    "spans": [],
+                                    "index": 18
+                                },
+                                {
+                                    "bbox": [
+                                        107,
+                                        644.0,
+                                        503,
+                                        684.0
+                                    ],
+                                    "spans": [],
+                                    "index": 19
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 17.0
+                }
+            ],
+            "page_idx": 12,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        107,
+                        26,
+                        308,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                25,
+                                308,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        25,
+                                        308,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Under review as a conference paper at ICLR 2020",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        300,
+                        751,
+                        310,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                299,
+                                750,
+                                312,
+                                764
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        299,
+                                        750,
+                                        312,
+                                        764
+                                    ],
+                                    "score": 1.0,
+                                    "content": "",
+                                    "type": "text",
+                                    "height": 14,
+                                    "width": 13
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        107,
+                        700,
+                        505,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                117,
+                                698,
+                                505,
+                                714
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        117,
+                                        698,
+                                        505,
+                                        714
+                                    ],
+                                    "score": 1.0,
+                                    "content": "9Zhang, Han, et al. \"Self-attention generative adversarial networks.\" arXiv preprint arXiv:1805.08318 (2018).",
+                                    "type": "text"
+                                }
+                            ]
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                708,
+                                506,
+                                725
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        708,
+                                        506,
+                                        725
+                                    ],
+                                    "score": 1.0,
+                                    "content": "10Miyato, Takeru, et al. \"Spectral normalization for generative adversarial networks.\" arXiv preprint",
+                                    "type": "text"
+                                }
+                            ]
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                721,
+                                203,
+                                732
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        721,
+                                        203,
+                                        732
+                                    ],
+                                    "score": 1.0,
+                                    "content": "arXiv:1802.05957 (2018).",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        81,
+                        325,
+                        94
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                81,
+                                325,
+                                95
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        81,
+                                        325,
+                                        95
+                                    ],
+                                    "score": 1.0,
+                                    "content": "A IMPLEMENTATION AND EXPERIMENTS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        }
+                    ],
+                    "index": 0
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        108,
+                        505,
+                        153
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                108,
+                                506,
+                                121
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        108,
+                                        506,
+                                        121
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Table 3 presents the list of software dependencies and versions used in our implementation. To",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                120,
+                                505,
+                                131
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        120,
+                                        505,
+                                        131
+                                    ],
+                                    "score": 1.0,
+                                    "content": "produce results related to this paper, we used a workstation with 4 NVIDIA GeForce RTX-2080Ti",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                130,
+                                506,
+                                144
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        130,
+                                        506,
+                                        144
+                                    ],
+                                    "score": 1.0,
+                                    "content": "GPUs, a 12 core Intel Core i9-7920X processor, and 128 GB memory. Each experiment took between",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                141,
+                                385,
+                                154
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        141,
+                                        385,
+                                        154
+                                    ],
+                                    "score": 1.0,
+                                    "content": "about 4 hours to 48 hours, based on the task and method being tested.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        }
+                    ],
+                    "index": 2.5,
+                    "bbox_fs": [
+                        105,
+                        108,
+                        506,
+                        154
+                    ]
+                },
+                {
+                    "type": "table",
+                    "bbox": [
+                        237,
+                        188,
+                        378,
+                        381
+                    ],
+                    "blocks": [
+                        {
+                            "type": "table_caption",
+                            "bbox": [
+                                240,
+                                167,
+                                371,
+                                179
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        239,
+                                        165,
+                                        372,
+                                        181
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                239,
+                                                165,
+                                                372,
+                                                181
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Table 3: Software dependencies.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 5
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "type": "table_body",
+                            "bbox": [
+                                237,
+                                188,
+                                378,
+                                381
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        237,
+                                        188,
+                                        378,
+                                        381
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                237,
+                                                188,
+                                                378,
+                                                381
+                                            ],
+                                            "score": 0.977,
+                                            "html": "<table><tr><td>Dependency</td><td>Version</td></tr><tr><td>python</td><td>3.7.1</td></tr><tr><td>pytorch</td><td>1.0.0</td></tr><tr><td>cuda100</td><td>1.0</td></tr><tr><td>ipython</td><td>7.2.0</td></tr><tr><td>jupyter</td><td>1.0.0</td></tr><tr><td>numpy</td><td>1.15.4</td></tr><tr><td>pandas</td><td>0.24.1</td></tr><tr><td>scikit-learn</td><td>0.20.1</td></tr><tr><td>scipy</td><td>1.1.0</td></tr><tr><td>torchvision</td><td>0.2.1</td></tr><tr><td>tqdm</td><td>4.28.1</td></tr><tr><td>matplotlib</td><td>3.0.1</td></tr></table>",
+                                            "type": "table",
+                                            "image_path": "c34dd7ed960160ceeeb9080e4ac592a923c1d401524807a6a19321bc3e36daa6.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 6,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        237,
+                                        188,
+                                        378,
+                                        381
+                                    ],
+                                    "spans": [],
+                                    "index": 6
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 5.5
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        412,
+                        272,
+                        425
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                411,
+                                273,
+                                426
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        411,
+                                        273,
+                                        426
+                                    ],
+                                    "score": 1.0,
+                                    "content": "B NETWORK ARCHITECTURES",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        }
+                    ],
+                    "index": 7
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        439,
+                        506,
+                        528
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                439,
+                                505,
+                                452
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        439,
+                                        505,
+                                        452
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Table 4 shows the exact architectures used in this paper. To show each layer or block we used the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                450,
+                                505,
+                                464
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        450,
+                                        184,
+                                        464
+                                    ],
+                                    "score": 1.0,
+                                    "content": "following notation.",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        185,
+                                        452,
+                                        233,
+                                        462
+                                    ],
+                                    "score": 0.72,
+                                    "content": "\\mathtt { C x S y P z - t }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        234,
+                                        450,
+                                        424,
+                                        464
+                                    ],
+                                    "score": 1.0,
+                                    "content": "represents a 2-d convolution layer of kernel size",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        425,
+                                        452,
+                                        432,
+                                        461
+                                    ],
+                                    "score": 0.43,
+                                    "content": "_ \\textrm { x }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        433,
+                                        450,
+                                        505,
+                                        464
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", stride y, padding",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                107,
+                                462,
+                                505,
+                                474
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        464,
+                                        113,
+                                        472
+                                    ],
+                                    "score": 0.31,
+                                    "content": "_ { \\textrm { Z } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        114,
+                                        462,
+                                        243,
+                                        474
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", and number of output channels",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        243,
+                                        463,
+                                        251,
+                                        472
+                                    ],
+                                    "score": 0.41,
+                                    "content": "\\scriptstyle \\pm",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        251,
+                                        462,
+                                        505,
+                                        474
+                                    ],
+                                    "score": 1.0,
+                                    "content": "followed by ReLU activation. Attn represents a self-attention",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                473,
+                                505,
+                                485
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        473,
+                                        229,
+                                        485
+                                    ],
+                                    "score": 1.0,
+                                    "content": "layer similar to Zhang et al.9.",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        230,
+                                        473,
+                                        249,
+                                        483
+                                    ],
+                                    "score": 0.74,
+                                    "content": "_ \\mathrm { R - x }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        250,
+                                        473,
+                                        505,
+                                        485
+                                    ],
+                                    "score": 1.0,
+                                    "content": "represents a residual block consisting of two 2-d convolutions",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                484,
+                                505,
+                                496
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        484,
+                                        425,
+                                        496
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with kernel size 3 (padding size 1), batch normalization, and ReLU activation.",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        425,
+                                        484,
+                                        479,
+                                        495
+                                    ],
+                                    "score": 0.87,
+                                    "content": "\\mathsf { C T x S y P z - t }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        480,
+                                        484,
+                                        505,
+                                        496
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                494,
+                                506,
+                                507
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        494,
+                                        266,
+                                        507
+                                    ],
+                                    "score": 1.0,
+                                    "content": "convolution transpose corresponding to",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        267,
+                                        496,
+                                        316,
+                                        506
+                                    ],
+                                    "score": 0.64,
+                                    "content": "\\mathtt { C x S y P z - t }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        316,
+                                        494,
+                                        319,
+                                        507
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        320,
+                                        495,
+                                        345,
+                                        505
+                                    ],
+                                    "score": 0.65,
+                                    "content": "\\operatorname { F C - x }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        346,
+                                        494,
+                                        506,
+                                        507
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is representing a linear fully-connected",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                505,
+                                506,
+                                518
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        505,
+                                        139,
+                                        518
+                                    ],
+                                    "score": 1.0,
+                                    "content": "layer of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        139,
+                                        507,
+                                        147,
+                                        516
+                                    ],
+                                    "score": 0.4,
+                                    "content": "_ \\textrm { x }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        147,
+                                        505,
+                                        506,
+                                        518
+                                    ],
+                                    "score": 1.0,
+                                    "content": "output neurons with biases. We use spectral normalization as suggested by Miyato et al.10",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                517,
+                                400,
+                                528
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        517,
+                                        400,
+                                        528
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for all convolutional layers in both generator and discriminator networks.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        }
+                    ],
+                    "index": 11.5,
+                    "bbox_fs": [
+                        104,
+                        439,
+                        506,
+                        528
+                    ]
+                },
+                {
+                    "type": "table",
+                    "bbox": [
+                        107,
+                        564,
+                        503,
+                        684
+                    ],
+                    "blocks": [
+                        {
+                            "type": "table_caption",
+                            "bbox": [
+                                192,
+                                542,
+                                418,
+                                554
+                            ],
+                            "group_id": 1,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        191,
+                                        540,
+                                        419,
+                                        555
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                191,
+                                                540,
+                                                419,
+                                                555
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Table 4: Network architectures used in our experiments.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 16
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "type": "table_body",
+                            "bbox": [
+                                107,
+                                564,
+                                503,
+                                684
+                            ],
+                            "group_id": 1,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        564,
+                                        503,
+                                        684
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                107,
+                                                564,
+                                                503,
+                                                684
+                                            ],
+                                            "score": 0.982,
+                                            "html": "<table><tr><td rowspan=1 colspan=1>Dataset</td><td rowspan=1 colspan=1>Generator/Discriminator Architecture</td><td rowspan=1 colspan=1>Predictor Architecture</td></tr><tr><td rowspan=1 colspan=1>CIFAR-10</td><td rowspan=1 colspan=1>C7S1P3-64，C3S2P1-128，Attn，R-128,Attn，R-128，Attn，R-128，Attn,R-128，CT3S2P1-128，CT7S1P3-3，Tanh/Sigmoid</td><td rowspan=1 colspan=1>ResNet-18 11,12</td></tr><tr><td rowspan=1 colspan=1>Landsat</td><td rowspan=1 colspan=1>FC-64，Sigmoid，BNorm，FC-64，Sigmoid,BNorm,FC-64，Sigmoid，BNorm，FC-36，Tanh/Sigmoid</td><td rowspan=1 colspan=1>FC-64，ReLU，BNorm，FC-64,ReLU，BNorm，FC-6，Softmax</td></tr><tr><td rowspan=1 colspan=1>MIT-BIH</td><td rowspan=1 colspan=1>FC-1860，ReLU，BNorm，FC-1860，ReLU，BNorm,FC-1860，ReLU，BNorm, FC-186,Tanh/Sigmoid</td><td rowspan=1 colspan=1>FC-1860，ReLU，BNorm，FC-1860,ReLU，BNorm，FC-5，Softmax</td></tr><tr><td rowspan=1 colspan=1>Diabetes</td><td rowspan=1 colspan=1>FC-45，ReLU，BNorm，FC-45，ReLU，BNorm,FC-45，ReLU，BNorm，FC-45，Tanh/Sigmoid</td><td rowspan=1 colspan=1>FC-22，ReLU，BNorm，FC-22,ReLU，BNorm，FC-3，Softmax</td></tr></table>",
+                                            "type": "table",
+                                            "image_path": "1dc4ede59e46fd6fe85ec060a8aeed53fa3365bbe7e3af9f8c634bf3d02c9b6d.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 18,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        564,
+                                        503,
+                                        604.0
+                                    ],
+                                    "spans": [],
+                                    "index": 17
+                                },
+                                {
+                                    "bbox": [
+                                        107,
+                                        604.0,
+                                        503,
+                                        644.0
+                                    ],
+                                    "spans": [],
+                                    "index": 18
+                                },
+                                {
+                                    "bbox": [
+                                        107,
+                                        644.0,
+                                        503,
+                                        684.0
+                                    ],
+                                    "spans": [],
+                                    "index": 19
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 17.0
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        81,
+                        279,
+                        94
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                79,
+                                281,
+                                96
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        79,
+                                        281,
+                                        96
+                                    ],
+                                    "score": 1.0,
+                                    "content": "C MISSING DATA MECHANISMS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        }
+                    ],
+                    "index": 0
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        106,
+                        505,
+                        194
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                106,
+                                506,
+                                119
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        106,
+                                        506,
+                                        119
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In this paper, we conduct experiments on two mechanisms for missing values: MCAR uniform and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                117,
+                                505,
+                                129
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        117,
+                                        505,
+                                        129
+                                    ],
+                                    "score": 1.0,
+                                    "content": "MCAR rectangular. As in our experiments and comparisons, we consider the case where only an",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                128,
+                                505,
+                                141
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        128,
+                                        505,
+                                        141
+                                    ],
+                                    "score": 1.0,
+                                    "content": "incomplete dataset is available for training. It is crucial to guarantee that each method has only access",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                140,
+                                505,
+                                151
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        140,
+                                        505,
+                                        151
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to a unique incomplete version of each sample. However, it is relatively expensive to load and store",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                150,
+                                505,
+                                163
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        150,
+                                        505,
+                                        163
+                                    ],
+                                    "score": 1.0,
+                                    "content": "feature masks for each sample in the dataset. Instead, we generate missing values during the data",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                160,
+                                505,
+                                174
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        160,
+                                        505,
+                                        174
+                                    ],
+                                    "score": 1.0,
+                                    "content": "load for each batch. A hashing mechanism is used to ensure that the same parts are missing for each",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                172,
+                                505,
+                                184
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        172,
+                                        505,
+                                        184
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sample throughout the training. Note that we set system, python, and external library hash seeds to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                182,
+                                353,
+                                195
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        182,
+                                        353,
+                                        195
+                                    ],
+                                    "score": 1.0,
+                                    "content": "fixed values to ensure the consistency between different runs.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        }
+                    ],
+                    "index": 4.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        200,
+                        505,
+                        321
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                200,
+                                505,
+                                212
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        200,
+                                        505,
+                                        212
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Algorithm 3 presents the procedure used for generating missing values with uniform structure. This",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                210,
+                                506,
+                                223
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        210,
+                                        506,
+                                        223
+                                    ],
+                                    "score": 1.0,
+                                    "content": "algorithm is sampling independent Bernoulli distributions with probabilities equal to the missing rate.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                221,
+                                506,
+                                235
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        221,
+                                        506,
+                                        235
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Algorithm 4 shows the outline for the rectangular missing structure used in image experiments. It",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                233,
+                                506,
+                                245
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        233,
+                                        506,
+                                        245
+                                    ],
+                                    "score": 1.0,
+                                    "content": "consists of selecting a random point as the center of the rectangle and then deciding on parameters",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                244,
+                                506,
+                                256
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        244,
+                                        506,
+                                        256
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to be used for the beta distribution based on the missing rate. Finally, the width and height of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                255,
+                                506,
+                                267
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        255,
+                                        506,
+                                        267
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the rectangular region are sampled from the latent beta distribution. In other words, we generate",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                266,
+                                505,
+                                278
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        266,
+                                        505,
+                                        278
+                                    ],
+                                    "score": 1.0,
+                                    "content": "rectangular regions centered at random locations within the image which have width and height",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                277,
+                                506,
+                                289
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        277,
+                                        469,
+                                        289
+                                    ],
+                                    "score": 1.0,
+                                    "content": "values determined by samples from a latent beta distribution. Here, distribution parameters,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        469,
+                                        278,
+                                        478,
+                                        286
+                                    ],
+                                    "score": 0.78,
+                                    "content": "\\alpha",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        478,
+                                        277,
+                                        495,
+                                        289
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        495,
+                                        277,
+                                        502,
+                                        288
+                                    ],
+                                    "score": 0.82,
+                                    "content": "\\beta",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        503,
+                                        277,
+                                        506,
+                                        289
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                288,
+                                505,
+                                300
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        288,
+                                        505,
+                                        300
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are used to control the average missing rate. The outcome would be rectangular regions of different",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                299,
+                                505,
+                                311
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        299,
+                                        505,
+                                        311
+                                    ],
+                                    "score": 1.0,
+                                    "content": "shape at different locations within the frame with the expected portion of missing area equal to the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                309,
+                                160,
+                                323
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        309,
+                                        160,
+                                        323
+                                    ],
+                                    "score": 1.0,
+                                    "content": "missing rate.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        }
+                    ],
+                    "index": 14
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        326,
+                        505,
+                        436
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                325,
+                                507,
+                                339
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        325,
+                                        344,
+                                        339
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In order to decide on the beta distribution parameters i.e.",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        344,
+                                        328,
+                                        352,
+                                        336
+                                    ],
+                                    "score": 0.73,
+                                    "content": "\\alpha",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        353,
+                                        325,
+                                        371,
+                                        339
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        371,
+                                        327,
+                                        379,
+                                        338
+                                    ],
+                                    "score": 0.85,
+                                    "content": "\\beta",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        379,
+                                        325,
+                                        507,
+                                        339
+                                    ],
+                                    "score": 1.0,
+                                    "content": "we use numerical simulations.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                336,
+                                507,
+                                351
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        336,
+                                        507,
+                                        351
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Specifically, we fix one of the parameters to 1 and change the other parameter in the range of [1,10],",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                348,
+                                506,
+                                361
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        348,
+                                        506,
+                                        361
+                                    ],
+                                    "score": 1.0,
+                                    "content": "while measuring the average missing rate caused by each case. Figure 5 shows the missing rates",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                359,
+                                505,
+                                372
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        359,
+                                        505,
+                                        372
+                                    ],
+                                    "score": 1.0,
+                                    "content": "caused by different beta distribution parameters. The first half of Figure 5 (missing rates less than",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                369,
+                                506,
+                                383
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        369,
+                                        249,
+                                        383
+                                    ],
+                                    "score": 1.0,
+                                    "content": "about 0.18) corresponds to setting",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        249,
+                                        371,
+                                        257,
+                                        382
+                                    ],
+                                    "score": 0.84,
+                                    "content": "\\beta",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        257,
+                                        369,
+                                        334,
+                                        383
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to 1 and changing",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        335,
+                                        372,
+                                        343,
+                                        380
+                                    ],
+                                    "score": 0.74,
+                                    "content": "\\alpha",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        343,
+                                        369,
+                                        478,
+                                        383
+                                    ],
+                                    "score": 1.0,
+                                    "content": "values; and the other half fixing",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        478,
+                                        372,
+                                        486,
+                                        380
+                                    ],
+                                    "score": 0.75,
+                                    "content": "\\alpha",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        486,
+                                        369,
+                                        506,
+                                        383
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to 1",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                381,
+                                506,
+                                394
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        381,
+                                        162,
+                                        394
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and changing",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        162,
+                                        382,
+                                        170,
+                                        393
+                                    ],
+                                    "score": 0.82,
+                                    "content": "\\beta",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        171,
+                                        381,
+                                        348,
+                                        394
+                                    ],
+                                    "score": 1.0,
+                                    "content": "values. To generate missing rates more than",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        348,
+                                        381,
+                                        369,
+                                        392
+                                    ],
+                                    "score": 0.87,
+                                    "content": "50 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        369,
+                                        381,
+                                        506,
+                                        394
+                                    ],
+                                    "score": 1.0,
+                                    "content": "we invert our masks and limit the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                391,
+                                505,
+                                405
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        391,
+                                        505,
+                                        405
+                                    ],
+                                    "score": 1.0,
+                                    "content": "observation to the rectangular region while the rest of the image is missing. Note that missing rates",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                402,
+                                506,
+                                415
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        402,
+                                        506,
+                                        415
+                                    ],
+                                    "score": 1.0,
+                                    "content": "indicate the ratio of features that are missing on the average case. As we are using a latent model for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                415,
+                                505,
+                                426
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        415,
+                                        505,
+                                        426
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sampling width and height for the rectangles, the actual missing ratios for each specific sample differs",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                425,
+                                505,
+                                438
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        425,
+                                        505,
+                                        438
+                                    ],
+                                    "score": 1.0,
+                                    "content": "between samples. See Table 5 for visual examples of different missing rates and missing structures.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        }
+                    ],
+                    "index": 24.5
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        167,
+                        452,
+                        334,
+                        463
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                166,
+                                452,
+                                335,
+                                463
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        166,
+                                        452,
+                                        335,
+                                        463
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Algorithm 3: MCAR uniform generation.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        }
+                    ],
+                    "index": 30
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        165,
+                        466,
+                        367,
+                        523
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                165,
+                                465,
+                                348,
+                                478
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        165,
+                                        465,
+                                        196,
+                                        478
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Input:",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        196,
+                                        468,
+                                        204,
+                                        476
+                                    ],
+                                    "score": 0.53,
+                                    "content": "_ { \\textbf { \\em x } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        204,
+                                        465,
+                                        283,
+                                        478
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(complete feature),",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        284,
+                                        468,
+                                        290,
+                                        475
+                                    ],
+                                    "score": 0.53,
+                                    "content": "r",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        290,
+                                        465,
+                                        348,
+                                        478
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(missing rate)",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                166,
+                                477,
+                                304,
+                                488
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        166,
+                                        477,
+                                        203,
+                                        488
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Output:",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        203,
+                                        478,
+                                        219,
+                                        488
+                                    ],
+                                    "score": 0.86,
+                                    "content": "{ \\pmb x } _ { m }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        220,
+                                        477,
+                                        304,
+                                        488
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(incomplete feature)",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                165,
+                                488,
+                                367,
+                                523
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        165,
+                                        488,
+                                        367,
+                                        523
+                                    ],
+                                    "score": 0.44,
+                                    "content": "\\begin{array} { r l } & { s e e d _ { x }  h a s h ( \\pmb { x } ) } \\\\ & { \\pmb { k }  1 - B e r n o u l l i ( s e e d _ { x } , s h a p e ( x ) , p r o b = r ) } \\\\ & { \\pmb { x } _ { m }  \\pmb { k } \\odot \\pmb { x } + ( 1 - \\pmb { k } ) \\odot N a N } \\end{array}",
+                                    "type": "inline_equation",
+                                    "image_path": "06a26e654993e552790911ee6abfdab00e2873558b2fa34c3a8a3dd454621dc6.jpg"
+                                }
+                            ],
+                            "index": 33
+                        }
+                    ],
+                    "index": 32
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        167,
+                        532,
+                        321,
+                        543
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                165,
+                                529,
+                                322,
+                                545
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        165,
+                                        529,
+                                        322,
+                                        545
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Algorithm 4: MCAR rect. generation.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        }
+                    ],
+                    "index": 34
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        164,
+                        545,
+                        437,
+                        669
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                165,
+                                544,
+                                348,
+                                557
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        165,
+                                        544,
+                                        196,
+                                        557
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Input:",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        196,
+                                        547,
+                                        204,
+                                        555
+                                    ],
+                                    "score": 0.51,
+                                    "content": "_ { \\textbf { \\em x } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        204,
+                                        544,
+                                        283,
+                                        557
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(complete feature),",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        283,
+                                        547,
+                                        290,
+                                        555
+                                    ],
+                                    "score": 0.49,
+                                    "content": "r",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        290,
+                                        544,
+                                        348,
+                                        557
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(missing rate)",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                165,
+                                555,
+                                304,
+                                568
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        165,
+                                        555,
+                                        203,
+                                        568
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Output:",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        203,
+                                        557,
+                                        220,
+                                        567
+                                    ],
+                                    "score": 0.86,
+                                    "content": "{ \\pmb x } _ { m }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        220,
+                                        555,
+                                        304,
+                                        568
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(incomplete feature)",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                166,
+                                568,
+                                242,
+                                579
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        166,
+                                        568,
+                                        242,
+                                        579
+                                    ],
+                                    "score": 0.75,
+                                    "content": "s e e d _ { x } \\gets h a s h ( { \\pmb x } )",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                164,
+                                578,
+                                249,
+                                590
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        164,
+                                        578,
+                                        171,
+                                        590
+                                    ],
+                                    "score": 1.0,
+                                    "content": "n",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        171,
+                                        579,
+                                        249,
+                                        590
+                                    ],
+                                    "score": 0.48,
+                                    "content": "\\iota _ { x } , n _ { y } \\gets s h a p e ( { \\pmb x } )",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                165,
+                                589,
+                                358,
+                                601
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        165,
+                                        589,
+                                        358,
+                                        601
+                                    ],
+                                    "score": 0.39,
+                                    "content": "( p _ { x } , \\bar { p _ { y } } ) \\sim ( u n i f o r m ( 0 , n _ { x } ) , u n i f o r m ( 0 , n _ { y } ) )",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                165,
+                                600,
+                                430,
+                                612
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        165,
+                                        600,
+                                        386,
+                                        612
+                                    ],
+                                    "score": 1.0,
+                                    "content": "α, β ← beta_params(r) // beta_params gives",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        387,
+                                        600,
+                                        406,
+                                        612
+                                    ],
+                                    "score": 0.81,
+                                    "content": "\\alpha , \\beta",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        406,
+                                        600,
+                                        430,
+                                        612
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                183,
+                                611,
+                                401,
+                                622
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        183,
+                                        611,
+                                        401,
+                                        622
+                                    ],
+                                    "score": 1.0,
+                                    "content": "each missing rate based on numerical",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                184,
+                                622,
+                                253,
+                                633
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        184,
+                                        622,
+                                        253,
+                                        633
+                                    ],
+                                    "score": 1.0,
+                                    "content": "simulations",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                164,
+                                632,
+                                363,
+                                668
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        164,
+                                        632,
+                                        363,
+                                        668
+                                    ],
+                                    "score": 0.41,
+                                    "content": "\\begin{array} { r l } & { ( \\boldsymbol { w } , h ) \\sim ( B e t a ( \\alpha , \\beta ) \\times n _ { x } ) , B e t a ( \\alpha , \\beta ) \\times n _ { y } ) ) } \\\\ & { \\boldsymbol { k }  r e c t \\_ m a s k ( p _ { x } , p _ { y } , w , h ) } \\\\ & { \\mathbf { \\boldsymbol { x } } _ { m }  \\boldsymbol { k } \\odot \\mathbf { \\boldsymbol { x } } + ( 1 - \\boldsymbol { k } ) \\odot N a N } \\end{array}",
+                                    "type": "inline_equation",
+                                    "image_path": "7d273e767eb2dc99b0474f6566df8e4c26eafa4cdaa362a676cb7d4550bb992d.jpg"
+                                }
+                            ],
+                            "index": 43
+                        }
+                    ],
+                    "index": 39
+                }
+            ],
+            "page_idx": 13,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        107,
+                        700,
+                        505,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                115,
+                                698,
+                                506,
+                                713
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        698,
+                                        506,
+                                        713
+                                    ],
+                                    "score": 1.0,
+                                    "content": "11He, Kaiming, et al. \"Deep residual learning for image recognition.\" Proceedings of the IEEE conference on",
+                                    "type": "text"
+                                }
+                            ]
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                711,
+                                277,
+                                721
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        711,
+                                        277,
+                                        721
+                                    ],
+                                    "score": 1.0,
+                                    "content": "computer vision and pattern recognition. 2016.",
+                                    "type": "text"
+                                }
+                            ]
+                        },
+                        {
+                            "bbox": [
+                                114,
+                                718,
+                                347,
+                                735
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        114,
+                                        718,
+                                        347,
+                                        735
+                                    ],
+                                    "score": 1.0,
+                                    "content": "12https://github.com/kuangliu/pytorch-cifar",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        300,
+                        751,
+                        310,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                299,
+                                750,
+                                312,
+                                764
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        299,
+                                        750,
+                                        312,
+                                        764
+                                    ],
+                                    "score": 1.0,
+                                    "content": "",
+                                    "type": "text",
+                                    "height": 14,
+                                    "width": 13
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        107,
+                        27,
+                        308,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                107,
+                                26,
+                                308,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        26,
+                                        308,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Under review as a conference paper at ICLR 2020",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        81,
+                        279,
+                        94
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                79,
+                                281,
+                                96
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        79,
+                                        281,
+                                        96
+                                    ],
+                                    "score": 1.0,
+                                    "content": "C MISSING DATA MECHANISMS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        }
+                    ],
+                    "index": 0
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        106,
+                        505,
+                        194
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                106,
+                                506,
+                                119
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        106,
+                                        506,
+                                        119
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In this paper, we conduct experiments on two mechanisms for missing values: MCAR uniform and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                117,
+                                505,
+                                129
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        117,
+                                        505,
+                                        129
+                                    ],
+                                    "score": 1.0,
+                                    "content": "MCAR rectangular. As in our experiments and comparisons, we consider the case where only an",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                128,
+                                505,
+                                141
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        128,
+                                        505,
+                                        141
+                                    ],
+                                    "score": 1.0,
+                                    "content": "incomplete dataset is available for training. It is crucial to guarantee that each method has only access",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                140,
+                                505,
+                                151
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        140,
+                                        505,
+                                        151
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to a unique incomplete version of each sample. However, it is relatively expensive to load and store",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                150,
+                                505,
+                                163
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        150,
+                                        505,
+                                        163
+                                    ],
+                                    "score": 1.0,
+                                    "content": "feature masks for each sample in the dataset. Instead, we generate missing values during the data",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                160,
+                                505,
+                                174
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        160,
+                                        505,
+                                        174
+                                    ],
+                                    "score": 1.0,
+                                    "content": "load for each batch. A hashing mechanism is used to ensure that the same parts are missing for each",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                172,
+                                505,
+                                184
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        172,
+                                        505,
+                                        184
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sample throughout the training. Note that we set system, python, and external library hash seeds to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                182,
+                                353,
+                                195
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        182,
+                                        353,
+                                        195
+                                    ],
+                                    "score": 1.0,
+                                    "content": "fixed values to ensure the consistency between different runs.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        }
+                    ],
+                    "index": 4.5,
+                    "bbox_fs": [
+                        105,
+                        106,
+                        506,
+                        195
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        200,
+                        505,
+                        321
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                200,
+                                505,
+                                212
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        200,
+                                        505,
+                                        212
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Algorithm 3 presents the procedure used for generating missing values with uniform structure. This",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                210,
+                                506,
+                                223
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        210,
+                                        506,
+                                        223
+                                    ],
+                                    "score": 1.0,
+                                    "content": "algorithm is sampling independent Bernoulli distributions with probabilities equal to the missing rate.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                221,
+                                506,
+                                235
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        221,
+                                        506,
+                                        235
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Algorithm 4 shows the outline for the rectangular missing structure used in image experiments. It",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                233,
+                                506,
+                                245
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        233,
+                                        506,
+                                        245
+                                    ],
+                                    "score": 1.0,
+                                    "content": "consists of selecting a random point as the center of the rectangle and then deciding on parameters",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                244,
+                                506,
+                                256
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        244,
+                                        506,
+                                        256
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to be used for the beta distribution based on the missing rate. Finally, the width and height of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                255,
+                                506,
+                                267
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        255,
+                                        506,
+                                        267
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the rectangular region are sampled from the latent beta distribution. In other words, we generate",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                266,
+                                505,
+                                278
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        266,
+                                        505,
+                                        278
+                                    ],
+                                    "score": 1.0,
+                                    "content": "rectangular regions centered at random locations within the image which have width and height",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                277,
+                                506,
+                                289
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        277,
+                                        469,
+                                        289
+                                    ],
+                                    "score": 1.0,
+                                    "content": "values determined by samples from a latent beta distribution. Here, distribution parameters,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        469,
+                                        278,
+                                        478,
+                                        286
+                                    ],
+                                    "score": 0.78,
+                                    "content": "\\alpha",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        478,
+                                        277,
+                                        495,
+                                        289
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        495,
+                                        277,
+                                        502,
+                                        288
+                                    ],
+                                    "score": 0.82,
+                                    "content": "\\beta",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        503,
+                                        277,
+                                        506,
+                                        289
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                288,
+                                505,
+                                300
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        288,
+                                        505,
+                                        300
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are used to control the average missing rate. The outcome would be rectangular regions of different",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                299,
+                                505,
+                                311
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        299,
+                                        505,
+                                        311
+                                    ],
+                                    "score": 1.0,
+                                    "content": "shape at different locations within the frame with the expected portion of missing area equal to the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                309,
+                                160,
+                                323
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        309,
+                                        160,
+                                        323
+                                    ],
+                                    "score": 1.0,
+                                    "content": "missing rate.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        }
+                    ],
+                    "index": 14,
+                    "bbox_fs": [
+                        105,
+                        200,
+                        506,
+                        323
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        326,
+                        505,
+                        436
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                325,
+                                507,
+                                339
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        325,
+                                        344,
+                                        339
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In order to decide on the beta distribution parameters i.e.",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        344,
+                                        328,
+                                        352,
+                                        336
+                                    ],
+                                    "score": 0.73,
+                                    "content": "\\alpha",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        353,
+                                        325,
+                                        371,
+                                        339
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        371,
+                                        327,
+                                        379,
+                                        338
+                                    ],
+                                    "score": 0.85,
+                                    "content": "\\beta",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        379,
+                                        325,
+                                        507,
+                                        339
+                                    ],
+                                    "score": 1.0,
+                                    "content": "we use numerical simulations.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                336,
+                                507,
+                                351
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        336,
+                                        507,
+                                        351
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Specifically, we fix one of the parameters to 1 and change the other parameter in the range of [1,10],",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                348,
+                                506,
+                                361
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        348,
+                                        506,
+                                        361
+                                    ],
+                                    "score": 1.0,
+                                    "content": "while measuring the average missing rate caused by each case. Figure 5 shows the missing rates",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                359,
+                                505,
+                                372
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        359,
+                                        505,
+                                        372
+                                    ],
+                                    "score": 1.0,
+                                    "content": "caused by different beta distribution parameters. The first half of Figure 5 (missing rates less than",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                369,
+                                506,
+                                383
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        369,
+                                        249,
+                                        383
+                                    ],
+                                    "score": 1.0,
+                                    "content": "about 0.18) corresponds to setting",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        249,
+                                        371,
+                                        257,
+                                        382
+                                    ],
+                                    "score": 0.84,
+                                    "content": "\\beta",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        257,
+                                        369,
+                                        334,
+                                        383
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to 1 and changing",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        335,
+                                        372,
+                                        343,
+                                        380
+                                    ],
+                                    "score": 0.74,
+                                    "content": "\\alpha",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        343,
+                                        369,
+                                        478,
+                                        383
+                                    ],
+                                    "score": 1.0,
+                                    "content": "values; and the other half fixing",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        478,
+                                        372,
+                                        486,
+                                        380
+                                    ],
+                                    "score": 0.75,
+                                    "content": "\\alpha",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        486,
+                                        369,
+                                        506,
+                                        383
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to 1",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                381,
+                                506,
+                                394
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        381,
+                                        162,
+                                        394
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and changing",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        162,
+                                        382,
+                                        170,
+                                        393
+                                    ],
+                                    "score": 0.82,
+                                    "content": "\\beta",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        171,
+                                        381,
+                                        348,
+                                        394
+                                    ],
+                                    "score": 1.0,
+                                    "content": "values. To generate missing rates more than",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        348,
+                                        381,
+                                        369,
+                                        392
+                                    ],
+                                    "score": 0.87,
+                                    "content": "50 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        369,
+                                        381,
+                                        506,
+                                        394
+                                    ],
+                                    "score": 1.0,
+                                    "content": "we invert our masks and limit the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                391,
+                                505,
+                                405
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        391,
+                                        505,
+                                        405
+                                    ],
+                                    "score": 1.0,
+                                    "content": "observation to the rectangular region while the rest of the image is missing. Note that missing rates",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                402,
+                                506,
+                                415
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        402,
+                                        506,
+                                        415
+                                    ],
+                                    "score": 1.0,
+                                    "content": "indicate the ratio of features that are missing on the average case. As we are using a latent model for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                415,
+                                505,
+                                426
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        415,
+                                        505,
+                                        426
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sampling width and height for the rectangles, the actual missing ratios for each specific sample differs",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                425,
+                                505,
+                                438
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        425,
+                                        505,
+                                        438
+                                    ],
+                                    "score": 1.0,
+                                    "content": "between samples. See Table 5 for visual examples of different missing rates and missing structures.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        }
+                    ],
+                    "index": 24.5,
+                    "bbox_fs": [
+                        105,
+                        325,
+                        507,
+                        438
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        167,
+                        452,
+                        334,
+                        463
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                166,
+                                452,
+                                335,
+                                463
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        166,
+                                        452,
+                                        335,
+                                        463
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Algorithm 3: MCAR uniform generation.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        }
+                    ],
+                    "index": 30
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        165,
+                        466,
+                        367,
+                        523
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                165,
+                                465,
+                                348,
+                                478
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        165,
+                                        465,
+                                        196,
+                                        478
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Input:",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        196,
+                                        468,
+                                        204,
+                                        476
+                                    ],
+                                    "score": 0.53,
+                                    "content": "_ { \\textbf { \\em x } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        204,
+                                        465,
+                                        283,
+                                        478
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(complete feature),",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        284,
+                                        468,
+                                        290,
+                                        475
+                                    ],
+                                    "score": 0.53,
+                                    "content": "r",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        290,
+                                        465,
+                                        348,
+                                        478
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(missing rate)",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                166,
+                                477,
+                                304,
+                                488
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        166,
+                                        477,
+                                        203,
+                                        488
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Output:",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        203,
+                                        478,
+                                        219,
+                                        488
+                                    ],
+                                    "score": 0.86,
+                                    "content": "{ \\pmb x } _ { m }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        220,
+                                        477,
+                                        304,
+                                        488
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(incomplete feature)",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                165,
+                                488,
+                                367,
+                                523
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        165,
+                                        488,
+                                        367,
+                                        523
+                                    ],
+                                    "score": 0.44,
+                                    "content": "\\begin{array} { r l } & { s e e d _ { x }  h a s h ( \\pmb { x } ) } \\\\ & { \\pmb { k }  1 - B e r n o u l l i ( s e e d _ { x } , s h a p e ( x ) , p r o b = r ) } \\\\ & { \\pmb { x } _ { m }  \\pmb { k } \\odot \\pmb { x } + ( 1 - \\pmb { k } ) \\odot N a N } \\end{array}",
+                                    "type": "inline_equation",
+                                    "image_path": "06a26e654993e552790911ee6abfdab00e2873558b2fa34c3a8a3dd454621dc6.jpg"
+                                }
+                            ],
+                            "index": 33
+                        }
+                    ],
+                    "index": 32,
+                    "bbox_fs": [
+                        165,
+                        465,
+                        367,
+                        523
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        167,
+                        532,
+                        321,
+                        543
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                165,
+                                529,
+                                322,
+                                545
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        165,
+                                        529,
+                                        322,
+                                        545
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Algorithm 4: MCAR rect. generation.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        }
+                    ],
+                    "index": 34
+                },
+                {
+                    "type": "list",
+                    "bbox": [
+                        164,
+                        545,
+                        437,
+                        669
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                165,
+                                544,
+                                348,
+                                557
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        165,
+                                        544,
+                                        196,
+                                        557
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Input:",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        196,
+                                        547,
+                                        204,
+                                        555
+                                    ],
+                                    "score": 0.51,
+                                    "content": "_ { \\textbf { \\em x } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        204,
+                                        544,
+                                        283,
+                                        557
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(complete feature),",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        283,
+                                        547,
+                                        290,
+                                        555
+                                    ],
+                                    "score": 0.49,
+                                    "content": "r",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        290,
+                                        544,
+                                        348,
+                                        557
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(missing rate)",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35,
+                            "is_list_start_line": true,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                165,
+                                555,
+                                304,
+                                568
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        165,
+                                        555,
+                                        203,
+                                        568
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Output:",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        203,
+                                        557,
+                                        220,
+                                        567
+                                    ],
+                                    "score": 0.86,
+                                    "content": "{ \\pmb x } _ { m }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        220,
+                                        555,
+                                        304,
+                                        568
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(incomplete feature)",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36,
+                            "is_list_start_line": true,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                166,
+                                568,
+                                242,
+                                579
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        166,
+                                        568,
+                                        242,
+                                        579
+                                    ],
+                                    "score": 0.75,
+                                    "content": "s e e d _ { x } \\gets h a s h ( { \\pmb x } )",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 37,
+                            "is_list_start_line": true,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                164,
+                                578,
+                                249,
+                                590
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        164,
+                                        578,
+                                        171,
+                                        590
+                                    ],
+                                    "score": 1.0,
+                                    "content": "n",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        171,
+                                        579,
+                                        249,
+                                        590
+                                    ],
+                                    "score": 0.48,
+                                    "content": "\\iota _ { x } , n _ { y } \\gets s h a p e ( { \\pmb x } )",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 38,
+                            "is_list_start_line": true,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                165,
+                                589,
+                                358,
+                                601
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        165,
+                                        589,
+                                        358,
+                                        601
+                                    ],
+                                    "score": 0.39,
+                                    "content": "( p _ { x } , \\bar { p _ { y } } ) \\sim ( u n i f o r m ( 0 , n _ { x } ) , u n i f o r m ( 0 , n _ { y } ) )",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 39,
+                            "is_list_start_line": true,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                165,
+                                600,
+                                430,
+                                612
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        165,
+                                        600,
+                                        386,
+                                        612
+                                    ],
+                                    "score": 1.0,
+                                    "content": "α, β ← beta_params(r) // beta_params gives",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        387,
+                                        600,
+                                        406,
+                                        612
+                                    ],
+                                    "score": 0.81,
+                                    "content": "\\alpha , \\beta",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        406,
+                                        600,
+                                        430,
+                                        612
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                183,
+                                611,
+                                401,
+                                622
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        183,
+                                        611,
+                                        401,
+                                        622
+                                    ],
+                                    "score": 1.0,
+                                    "content": "each missing rate based on numerical",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                184,
+                                622,
+                                253,
+                                633
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        184,
+                                        622,
+                                        253,
+                                        633
+                                    ],
+                                    "score": 1.0,
+                                    "content": "simulations",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                164,
+                                632,
+                                363,
+                                668
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        164,
+                                        632,
+                                        363,
+                                        668
+                                    ],
+                                    "score": 0.41,
+                                    "content": "\\begin{array} { r l } & { ( \\boldsymbol { w } , h ) \\sim ( B e t a ( \\alpha , \\beta ) \\times n _ { x } ) , B e t a ( \\alpha , \\beta ) \\times n _ { y } ) ) } \\\\ & { \\boldsymbol { k }  r e c t \\_ m a s k ( p _ { x } , p _ { y } , w , h ) } \\\\ & { \\mathbf { \\boldsymbol { x } } _ { m }  \\boldsymbol { k } \\odot \\mathbf { \\boldsymbol { x } } + ( 1 - \\boldsymbol { k } ) \\odot N a N } \\end{array}",
+                                    "type": "inline_equation",
+                                    "image_path": "7d273e767eb2dc99b0474f6566df8e4c26eafa4cdaa362a676cb7d4550bb992d.jpg"
+                                }
+                            ],
+                            "index": 43,
+                            "is_list_start_line": true,
+                            "is_list_end_line": true
+                        }
+                    ],
+                    "index": 39,
+                    "bbox_fs": [
+                        164,
+                        544,
+                        430,
+                        668
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "image",
+                    "bbox": [
+                        173,
+                        150,
+                        436,
+                        320
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                173,
+                                150,
+                                436,
+                                320
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        173,
+                                        150,
+                                        436,
+                                        320
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                173,
+                                                150,
+                                                436,
+                                                320
+                                            ],
+                                            "score": 0.966,
+                                            "type": "image",
+                                            "image_path": "29b1ccc7a90bf84d0c4b18de35c4df32b98f105fff00abdb71b3fde7ca644a3b.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 1,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        173,
+                                        150,
+                                        436,
+                                        206.66666666666666
+                                    ],
+                                    "spans": [],
+                                    "index": 0
+                                },
+                                {
+                                    "bbox": [
+                                        173,
+                                        206.66666666666666,
+                                        436,
+                                        263.3333333333333
+                                    ],
+                                    "spans": [],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        173,
+                                        263.3333333333333,
+                                        436,
+                                        320.0
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                107,
+                                336,
+                                504,
+                                360
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        336,
+                                        505,
+                                        349
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                336,
+                                                505,
+                                                349
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 5: Simulation results for measuring average missing rate given different beta distribution",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 3
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        349,
+                                        155,
+                                        360
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                349,
+                                                155,
+                                                360
+                                            ],
+                                            "score": 1.0,
+                                            "content": "parameters.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 4
+                                }
+                            ],
+                            "index": 3.5
+                        }
+                    ],
+                    "index": 2.25
+                },
+                {
+                    "type": "image",
+                    "bbox": [
+                        109,
+                        512,
+                        499,
+                        666
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                109,
+                                512,
+                                499,
+                                666
+                            ],
+                            "group_id": 1,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        109,
+                                        512,
+                                        499,
+                                        666
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                109,
+                                                512,
+                                                499,
+                                                666
+                                            ],
+                                            "score": 0.904,
+                                            "type": "image",
+                                            "image_path": "b85dc27446c318885a97542d42c9e6453ece8284078aca51ca167c500d3ddc30.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 6,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        109,
+                                        512,
+                                        499,
+                                        563.3333333333334
+                                    ],
+                                    "spans": [],
+                                    "index": 5
+                                },
+                                {
+                                    "bbox": [
+                                        109,
+                                        563.3333333333334,
+                                        499,
+                                        614.6666666666667
+                                    ],
+                                    "spans": [],
+                                    "index": 6
+                                },
+                                {
+                                    "bbox": [
+                                        109,
+                                        614.6666666666667,
+                                        499,
+                                        666.0000000000001
+                                    ],
+                                    "spans": [],
+                                    "index": 7
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 6
+                }
+            ],
+            "page_idx": 14,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        106,
+                        27,
+                        309,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                26,
+                                308,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        26,
+                                        308,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Under review as a conference paper at ICLR 2020",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        300,
+                        751,
+                        310,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                299,
+                                750,
+                                312,
+                                764
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        299,
+                                        750,
+                                        312,
+                                        764
+                                    ],
+                                    "score": 1.0,
+                                    "content": "",
+                                    "type": "text",
+                                    "height": 14,
+                                    "width": 13
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "image",
+                    "bbox": [
+                        173,
+                        150,
+                        436,
+                        320
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                173,
+                                150,
+                                436,
+                                320
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        173,
+                                        150,
+                                        436,
+                                        320
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                173,
+                                                150,
+                                                436,
+                                                320
+                                            ],
+                                            "score": 0.966,
+                                            "type": "image",
+                                            "image_path": "29b1ccc7a90bf84d0c4b18de35c4df32b98f105fff00abdb71b3fde7ca644a3b.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 1,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        173,
+                                        150,
+                                        436,
+                                        206.66666666666666
+                                    ],
+                                    "spans": [],
+                                    "index": 0
+                                },
+                                {
+                                    "bbox": [
+                                        173,
+                                        206.66666666666666,
+                                        436,
+                                        263.3333333333333
+                                    ],
+                                    "spans": [],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        173,
+                                        263.3333333333333,
+                                        436,
+                                        320.0
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                107,
+                                336,
+                                504,
+                                360
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        336,
+                                        505,
+                                        349
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                336,
+                                                505,
+                                                349
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 5: Simulation results for measuring average missing rate given different beta distribution",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 3
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        349,
+                                        155,
+                                        360
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                349,
+                                                155,
+                                                360
+                                            ],
+                                            "score": 1.0,
+                                            "content": "parameters.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 4
+                                }
+                            ],
+                            "index": 3.5
+                        }
+                    ],
+                    "index": 2.25
+                },
+                {
+                    "type": "image",
+                    "bbox": [
+                        109,
+                        512,
+                        499,
+                        666
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                109,
+                                512,
+                                499,
+                                666
+                            ],
+                            "group_id": 1,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        109,
+                                        512,
+                                        499,
+                                        666
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                109,
+                                                512,
+                                                499,
+                                                666
+                                            ],
+                                            "score": 0.904,
+                                            "type": "image",
+                                            "image_path": "b85dc27446c318885a97542d42c9e6453ece8284078aca51ca167c500d3ddc30.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 6,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        109,
+                                        512,
+                                        499,
+                                        563.3333333333334
+                                    ],
+                                    "spans": [],
+                                    "index": 5
+                                },
+                                {
+                                    "bbox": [
+                                        109,
+                                        563.3333333333334,
+                                        499,
+                                        614.6666666666667
+                                    ],
+                                    "spans": [],
+                                    "index": 6
+                                },
+                                {
+                                    "bbox": [
+                                        109,
+                                        614.6666666666667,
+                                        499,
+                                        666.0000000000001
+                                    ],
+                                    "spans": [],
+                                    "index": 7
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 6
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        81,
+                        221,
+                        93
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                80,
+                                223,
+                                96
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        80,
+                                        223,
+                                        96
+                                    ],
+                                    "score": 1.0,
+                                    "content": "D ABLATION STUDY",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        }
+                    ],
+                    "index": 0
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        107,
+                        506,
+                        196
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                108,
+                                506,
+                                120
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        108,
+                                        506,
+                                        120
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Figure 6 presents a comparison between using (GI W/ Atten.) and not using (GI W/O Atten.)",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                119,
+                                506,
+                                131
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        119,
+                                        506,
+                                        131
+                                    ],
+                                    "score": 1.0,
+                                    "content": "self-attention layers before each residual block in the proposed architecture. We report FID scores",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                129,
+                                506,
+                                143
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        129,
+                                        506,
+                                        143
+                                    ],
+                                    "score": 1.0,
+                                    "content": "on CIFAR-10 with rectangular missingness. As it can be inferred from this comparison, using",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                141,
+                                506,
+                                153
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        141,
+                                        506,
+                                        153
+                                    ],
+                                    "score": 1.0,
+                                    "content": "self-attention achieves a consistent improvement over the baseline. We also examined the case of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                152,
+                                505,
+                                164
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        152,
+                                        505,
+                                        164
+                                    ],
+                                    "score": 1.0,
+                                    "content": "uniform missingness; however, we did not observe any significant improvement for this case. One",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                163,
+                                505,
+                                175
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        163,
+                                        505,
+                                        175
+                                    ],
+                                    "score": 1.0,
+                                    "content": "possible explanation could be the fact that imputing missing data with a uniform structure can be",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                173,
+                                506,
+                                187
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        173,
+                                        506,
+                                        187
+                                    ],
+                                    "score": 1.0,
+                                    "content": "done by processing local regions and does not require attending to different distant regions across the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                184,
+                                137,
+                                199
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        184,
+                                        137,
+                                        199
+                                    ],
+                                    "score": 1.0,
+                                    "content": "image.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        }
+                    ],
+                    "index": 4.5
+                },
+                {
+                    "type": "image",
+                    "bbox": [
+                        170,
+                        215,
+                        438,
+                        413
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                170,
+                                215,
+                                438,
+                                413
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        170,
+                                        215,
+                                        438,
+                                        413
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                170,
+                                                215,
+                                                438,
+                                                413
+                                            ],
+                                            "score": 0.973,
+                                            "type": "image",
+                                            "image_path": "555f58484ccab468a30f31b8a1a2e7e6951c3a282f1c5c77ff9c483cf318fc49.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 10,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        170,
+                                        215,
+                                        438,
+                                        281.0
+                                    ],
+                                    "spans": [],
+                                    "index": 9
+                                },
+                                {
+                                    "bbox": [
+                                        170,
+                                        281.0,
+                                        438,
+                                        347.0
+                                    ],
+                                    "spans": [],
+                                    "index": 10
+                                },
+                                {
+                                    "bbox": [
+                                        170,
+                                        347.0,
+                                        438,
+                                        413.0
+                                    ],
+                                    "spans": [],
+                                    "index": 11
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                107,
+                                427,
+                                504,
+                                450
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        427,
+                                        506,
+                                        440
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                427,
+                                                506,
+                                                440
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 6: Comparison of FID scores achieved with (GI W/ Atten.) and without (GI W/O Atten.)",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 12
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        438,
+                                        499,
+                                        451
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                438,
+                                                499,
+                                                451
+                                            ],
+                                            "score": 1.0,
+                                            "content": "self-attention layers on CIFAR-10 dataset and rectangular missingness. Lower FID score is better.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 13
+                                }
+                            ],
+                            "index": 12.5
+                        }
+                    ],
+                    "index": 11.25
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        465,
+                        505,
+                        509
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                464,
+                                506,
+                                478
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        464,
+                                        506,
+                                        478
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Figure 7 shows a comparison of classification accuracies for the Landsat dataset achieved using",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                476,
+                                505,
+                                488
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        476,
+                                        203,
+                                        488
+                                    ],
+                                    "score": 1.0,
+                                    "content": "different ensemble sizes",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        203,
+                                        477,
+                                        219,
+                                        487
+                                    ],
+                                    "score": 0.6,
+                                    "content": "( N )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        219,
+                                        476,
+                                        420,
+                                        488
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". As it can be seen from this figure, higher values of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        421,
+                                        477,
+                                        431,
+                                        486
+                                    ],
+                                    "score": 0.8,
+                                    "content": "N",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        431,
+                                        476,
+                                        505,
+                                        488
+                                    ],
+                                    "score": 1.0,
+                                    "content": "result in improved",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                488,
+                                505,
+                                499
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        488,
+                                        505,
+                                        499
+                                    ],
+                                    "score": 1.0,
+                                    "content": "accuracies, especially for higher missing rates. Also, it can be observed that for N values more than",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                498,
+                                230,
+                                510
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        498,
+                                        230,
+                                        510
+                                    ],
+                                    "score": 1.0,
+                                    "content": "64 the difference is negligible.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        }
+                    ],
+                    "index": 15.5
+                },
+                {
+                    "type": "image",
+                    "bbox": [
+                        172,
+                        528,
+                        437,
+                        702
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                172,
+                                528,
+                                437,
+                                702
+                            ],
+                            "group_id": 1,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        172,
+                                        528,
+                                        437,
+                                        702
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                172,
+                                                528,
+                                                437,
+                                                702
+                                            ],
+                                            "score": 0.972,
+                                            "type": "image",
+                                            "image_path": "12c219ac2d3a48f6cc52d091728437b57a1e9af830f5bd30babc7198f2bac84e.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 19,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        172,
+                                        528,
+                                        437,
+                                        586.0
+                                    ],
+                                    "spans": [],
+                                    "index": 18
+                                },
+                                {
+                                    "bbox": [
+                                        172,
+                                        586.0,
+                                        437,
+                                        644.0
+                                    ],
+                                    "spans": [],
+                                    "index": 19
+                                },
+                                {
+                                    "bbox": [
+                                        172,
+                                        644.0,
+                                        437,
+                                        702.0
+                                    ],
+                                    "spans": [],
+                                    "index": 20
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                117,
+                                717,
+                                491,
+                                729
+                            ],
+                            "group_id": 1,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        118,
+                                        716,
+                                        492,
+                                        730
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                118,
+                                                716,
+                                                472,
+                                                730
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 7: Comparison of classification accuracies achieved with different ensemble size",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                473,
+                                                717,
+                                                488,
+                                                729
+                                            ],
+                                            "score": 0.67,
+                                            "content": "( N )",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                489,
+                                                716,
+                                                492,
+                                                730
+                                            ],
+                                            "score": 1.0,
+                                            "content": ".",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 21
+                                }
+                            ],
+                            "index": 21
+                        }
+                    ],
+                    "index": 20.0
+                }
+            ],
+            "page_idx": 15,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        107,
+                        27,
+                        308,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                107,
+                                26,
+                                308,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        26,
+                                        308,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Under review as a conference paper at ICLR 2020",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        300,
+                        751,
+                        311,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                299,
+                                750,
+                                312,
+                                764
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        299,
+                                        750,
+                                        312,
+                                        764
+                                    ],
+                                    "score": 1.0,
+                                    "content": "16",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        81,
+                        221,
+                        93
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                80,
+                                223,
+                                96
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        80,
+                                        223,
+                                        96
+                                    ],
+                                    "score": 1.0,
+                                    "content": "D ABLATION STUDY",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        }
+                    ],
+                    "index": 0
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        107,
+                        506,
+                        196
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                108,
+                                506,
+                                120
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        108,
+                                        506,
+                                        120
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Figure 6 presents a comparison between using (GI W/ Atten.) and not using (GI W/O Atten.)",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                119,
+                                506,
+                                131
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        119,
+                                        506,
+                                        131
+                                    ],
+                                    "score": 1.0,
+                                    "content": "self-attention layers before each residual block in the proposed architecture. We report FID scores",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                129,
+                                506,
+                                143
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        129,
+                                        506,
+                                        143
+                                    ],
+                                    "score": 1.0,
+                                    "content": "on CIFAR-10 with rectangular missingness. As it can be inferred from this comparison, using",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                141,
+                                506,
+                                153
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        141,
+                                        506,
+                                        153
+                                    ],
+                                    "score": 1.0,
+                                    "content": "self-attention achieves a consistent improvement over the baseline. We also examined the case of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                152,
+                                505,
+                                164
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        152,
+                                        505,
+                                        164
+                                    ],
+                                    "score": 1.0,
+                                    "content": "uniform missingness; however, we did not observe any significant improvement for this case. One",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                163,
+                                505,
+                                175
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        163,
+                                        505,
+                                        175
+                                    ],
+                                    "score": 1.0,
+                                    "content": "possible explanation could be the fact that imputing missing data with a uniform structure can be",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                173,
+                                506,
+                                187
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        173,
+                                        506,
+                                        187
+                                    ],
+                                    "score": 1.0,
+                                    "content": "done by processing local regions and does not require attending to different distant regions across the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                184,
+                                137,
+                                199
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        184,
+                                        137,
+                                        199
+                                    ],
+                                    "score": 1.0,
+                                    "content": "image.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        }
+                    ],
+                    "index": 4.5,
+                    "bbox_fs": [
+                        105,
+                        108,
+                        506,
+                        199
+                    ]
+                },
+                {
+                    "type": "image",
+                    "bbox": [
+                        170,
+                        215,
+                        438,
+                        413
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                170,
+                                215,
+                                438,
+                                413
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        170,
+                                        215,
+                                        438,
+                                        413
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                170,
+                                                215,
+                                                438,
+                                                413
+                                            ],
+                                            "score": 0.973,
+                                            "type": "image",
+                                            "image_path": "555f58484ccab468a30f31b8a1a2e7e6951c3a282f1c5c77ff9c483cf318fc49.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 10,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        170,
+                                        215,
+                                        438,
+                                        281.0
+                                    ],
+                                    "spans": [],
+                                    "index": 9
+                                },
+                                {
+                                    "bbox": [
+                                        170,
+                                        281.0,
+                                        438,
+                                        347.0
+                                    ],
+                                    "spans": [],
+                                    "index": 10
+                                },
+                                {
+                                    "bbox": [
+                                        170,
+                                        347.0,
+                                        438,
+                                        413.0
+                                    ],
+                                    "spans": [],
+                                    "index": 11
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                107,
+                                427,
+                                504,
+                                450
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        427,
+                                        506,
+                                        440
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                427,
+                                                506,
+                                                440
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 6: Comparison of FID scores achieved with (GI W/ Atten.) and without (GI W/O Atten.)",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 12
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        438,
+                                        499,
+                                        451
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                438,
+                                                499,
+                                                451
+                                            ],
+                                            "score": 1.0,
+                                            "content": "self-attention layers on CIFAR-10 dataset and rectangular missingness. Lower FID score is better.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 13
+                                }
+                            ],
+                            "index": 12.5
+                        }
+                    ],
+                    "index": 11.25
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        465,
+                        505,
+                        509
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                464,
+                                506,
+                                478
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        464,
+                                        506,
+                                        478
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Figure 7 shows a comparison of classification accuracies for the Landsat dataset achieved using",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                476,
+                                505,
+                                488
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        476,
+                                        203,
+                                        488
+                                    ],
+                                    "score": 1.0,
+                                    "content": "different ensemble sizes",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        203,
+                                        477,
+                                        219,
+                                        487
+                                    ],
+                                    "score": 0.6,
+                                    "content": "( N )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        219,
+                                        476,
+                                        420,
+                                        488
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". As it can be seen from this figure, higher values of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        421,
+                                        477,
+                                        431,
+                                        486
+                                    ],
+                                    "score": 0.8,
+                                    "content": "N",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        431,
+                                        476,
+                                        505,
+                                        488
+                                    ],
+                                    "score": 1.0,
+                                    "content": "result in improved",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                488,
+                                505,
+                                499
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        488,
+                                        505,
+                                        499
+                                    ],
+                                    "score": 1.0,
+                                    "content": "accuracies, especially for higher missing rates. Also, it can be observed that for N values more than",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                498,
+                                230,
+                                510
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        498,
+                                        230,
+                                        510
+                                    ],
+                                    "score": 1.0,
+                                    "content": "64 the difference is negligible.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        }
+                    ],
+                    "index": 15.5,
+                    "bbox_fs": [
+                        105,
+                        464,
+                        506,
+                        510
+                    ]
+                },
+                {
+                    "type": "image",
+                    "bbox": [
+                        172,
+                        528,
+                        437,
+                        702
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                172,
+                                528,
+                                437,
+                                702
+                            ],
+                            "group_id": 1,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        172,
+                                        528,
+                                        437,
+                                        702
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                172,
+                                                528,
+                                                437,
+                                                702
+                                            ],
+                                            "score": 0.972,
+                                            "type": "image",
+                                            "image_path": "12c219ac2d3a48f6cc52d091728437b57a1e9af830f5bd30babc7198f2bac84e.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 19,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        172,
+                                        528,
+                                        437,
+                                        586.0
+                                    ],
+                                    "spans": [],
+                                    "index": 18
+                                },
+                                {
+                                    "bbox": [
+                                        172,
+                                        586.0,
+                                        437,
+                                        644.0
+                                    ],
+                                    "spans": [],
+                                    "index": 19
+                                },
+                                {
+                                    "bbox": [
+                                        172,
+                                        644.0,
+                                        437,
+                                        702.0
+                                    ],
+                                    "spans": [],
+                                    "index": 20
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                117,
+                                717,
+                                491,
+                                729
+                            ],
+                            "group_id": 1,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        118,
+                                        716,
+                                        492,
+                                        730
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                118,
+                                                716,
+                                                472,
+                                                730
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 7: Comparison of classification accuracies achieved with different ensemble size",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                473,
+                                                717,
+                                                488,
+                                                729
+                                            ],
+                                            "score": 0.67,
+                                            "content": "( N )",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                489,
+                                                716,
+                                                492,
+                                                730
+                                            ],
+                                            "score": 1.0,
+                                            "content": ".",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 21
+                                }
+                            ],
+                            "index": 21
+                        }
+                    ],
+                    "index": 20.0
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        82,
+                        505,
+                        171
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                83,
+                                505,
+                                95
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        83,
+                                        505,
+                                        95
+                                    ],
+                                    "score": 1.0,
+                                    "content": "To study the benefits of the suggested stochastic predictor, we conducted experiments comparing GI",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                92,
+                                506,
+                                107
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        92,
+                                        240,
+                                        107
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with its non-stochastic variation",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        241,
+                                        94,
+                                        263,
+                                        104
+                                    ],
+                                    "score": 0.73,
+                                    "content": "( \\mathrm { N } { = } 1 )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        263,
+                                        92,
+                                        506,
+                                        107
+                                    ],
+                                    "score": 1.0,
+                                    "content": "). Here, the CIFAR-10 dataset with the rectangular missing",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                105,
+                                505,
+                                117
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        105,
+                                        241,
+                                        117
+                                    ],
+                                    "score": 1.0,
+                                    "content": "structure and missing rates from",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        241,
+                                        105,
+                                        262,
+                                        115
+                                    ],
+                                    "score": 0.85,
+                                    "content": "20 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        262,
+                                        105,
+                                        317,
+                                        117
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to as high as",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        317,
+                                        105,
+                                        338,
+                                        115
+                                    ],
+                                    "score": 0.87,
+                                    "content": "90 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        338,
+                                        105,
+                                        505,
+                                        117
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is used. From Table 6 it can be inferred",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                114,
+                                506,
+                                128
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        114,
+                                        506,
+                                        128
+                                    ],
+                                    "score": 1.0,
+                                    "content": "that as the rate of missingness increases, the benefits of the suggested predictor algorithm increase",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                126,
+                                506,
+                                139
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        126,
+                                        506,
+                                        139
+                                    ],
+                                    "score": 1.0,
+                                    "content": "significantly. We hypothesize that at higher rates of missingness, the conditional distribution of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                137,
+                                505,
+                                150
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        137,
+                                        505,
+                                        150
+                                    ],
+                                    "score": 1.0,
+                                    "content": "missing features becomes multimodal. In such a scenario, the suggested method captures the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                149,
+                                505,
+                                161
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        149,
+                                        505,
+                                        161
+                                    ],
+                                    "score": 1.0,
+                                    "content": "uncertainties over the target distribution resulting in the predictor to make more reliable class",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                160,
+                                160,
+                                172
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        160,
+                                        160,
+                                        172
+                                    ],
+                                    "score": 1.0,
+                                    "content": "assignments.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        }
+                    ],
+                    "index": 3.5
+                },
+                {
+                    "type": "table",
+                    "bbox": [
+                        146,
+                        212,
+                        464,
+                        298
+                    ],
+                    "blocks": [
+                        {
+                            "type": "table_caption",
+                            "bbox": [
+                                104,
+                                181,
+                                504,
+                                204
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        179,
+                                        505,
+                                        194
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                179,
+                                                360,
+                                                194
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Table 6: Comparison of CIFAR-10 accuracies for the stochastic",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                360,
+                                                181,
+                                                391,
+                                                192
+                                            ],
+                                            "score": 0.78,
+                                            "content": "\\mathrm { ( N = 1 2 8 ) }",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                391,
+                                                179,
+                                                480,
+                                                194
+                                            ],
+                                            "score": 1.0,
+                                            "content": ") and the deterministic",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                480,
+                                                181,
+                                                502,
+                                                192
+                                            ],
+                                            "score": 0.64,
+                                            "content": "\\left( \\mathrm { N } { = } 1 \\right)",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                502,
+                                                179,
+                                                505,
+                                                194
+                                            ],
+                                            "score": 1.0,
+                                            "content": ")",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 8
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        192,
+                                        270,
+                                        205
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                192,
+                                                270,
+                                                205
+                                            ],
+                                            "score": 1.0,
+                                            "content": "predictor under rectangular missingness.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 9
+                                }
+                            ],
+                            "index": 8.5
+                        },
+                        {
+                            "type": "table_body",
+                            "bbox": [
+                                146,
+                                212,
+                                464,
+                                298
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        146,
+                                        212,
+                                        464,
+                                        298
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                146,
+                                                212,
+                                                464,
+                                                298
+                                            ],
+                                            "score": 0.98,
+                                            "html": "<table><tr><td></td><td colspan=\"6\">Accuracy at Missing Rate (%)</td></tr><tr><td>Method</td><td>20%</td><td>40%</td><td>60%</td><td>70%</td><td>80%</td><td>90%</td></tr><tr><td>GI (N=128)</td><td>84.0</td><td>76.9</td><td>66.1</td><td>59.1</td><td>46.0</td><td>32.1</td></tr><tr><td>GI (N=1)</td><td>83.6</td><td>75.7</td><td>65.1</td><td>56.7</td><td>42.8</td><td>29.4</td></tr><tr><td>% difference (normalized)</td><td>0.5</td><td>1.6</td><td>1.5</td><td>4.1</td><td>6.9</td><td>8.4</td></tr></table>",
+                                            "type": "table",
+                                            "image_path": "ae7388387b96894d1aab2c0bea5caf9ef83886b0f974a93634b3532bca3a445a.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 11,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        146,
+                                        212,
+                                        464,
+                                        240.66666666666666
+                                    ],
+                                    "spans": [],
+                                    "index": 10
+                                },
+                                {
+                                    "bbox": [
+                                        146,
+                                        240.66666666666666,
+                                        464,
+                                        269.3333333333333
+                                    ],
+                                    "spans": [],
+                                    "index": 11
+                                },
+                                {
+                                    "bbox": [
+                                        146,
+                                        269.3333333333333,
+                                        464,
+                                        298.0
+                                    ],
+                                    "spans": [],
+                                    "index": 12
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 9.75
+                }
+            ],
+            "page_idx": 16,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        300,
+                        751,
+                        310,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                299,
+                                750,
+                                312,
+                                764
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        299,
+                                        750,
+                                        312,
+                                        764
+                                    ],
+                                    "score": 1.0,
+                                    "content": "",
+                                    "type": "text",
+                                    "height": 14,
+                                    "width": 13
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        107,
+                        27,
+                        308,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                25,
+                                308,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        25,
+                                        308,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Under review as a conference paper at ICLR 2020",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        82,
+                        505,
+                        171
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                83,
+                                505,
+                                95
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        83,
+                                        505,
+                                        95
+                                    ],
+                                    "score": 1.0,
+                                    "content": "To study the benefits of the suggested stochastic predictor, we conducted experiments comparing GI",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                92,
+                                506,
+                                107
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        92,
+                                        240,
+                                        107
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with its non-stochastic variation",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        241,
+                                        94,
+                                        263,
+                                        104
+                                    ],
+                                    "score": 0.73,
+                                    "content": "( \\mathrm { N } { = } 1 )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        263,
+                                        92,
+                                        506,
+                                        107
+                                    ],
+                                    "score": 1.0,
+                                    "content": "). Here, the CIFAR-10 dataset with the rectangular missing",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                105,
+                                505,
+                                117
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        105,
+                                        241,
+                                        117
+                                    ],
+                                    "score": 1.0,
+                                    "content": "structure and missing rates from",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        241,
+                                        105,
+                                        262,
+                                        115
+                                    ],
+                                    "score": 0.85,
+                                    "content": "20 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        262,
+                                        105,
+                                        317,
+                                        117
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to as high as",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        317,
+                                        105,
+                                        338,
+                                        115
+                                    ],
+                                    "score": 0.87,
+                                    "content": "90 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        338,
+                                        105,
+                                        505,
+                                        117
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is used. From Table 6 it can be inferred",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                114,
+                                506,
+                                128
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        114,
+                                        506,
+                                        128
+                                    ],
+                                    "score": 1.0,
+                                    "content": "that as the rate of missingness increases, the benefits of the suggested predictor algorithm increase",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                126,
+                                506,
+                                139
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        126,
+                                        506,
+                                        139
+                                    ],
+                                    "score": 1.0,
+                                    "content": "significantly. We hypothesize that at higher rates of missingness, the conditional distribution of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                137,
+                                505,
+                                150
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        137,
+                                        505,
+                                        150
+                                    ],
+                                    "score": 1.0,
+                                    "content": "missing features becomes multimodal. In such a scenario, the suggested method captures the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                149,
+                                505,
+                                161
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        149,
+                                        505,
+                                        161
+                                    ],
+                                    "score": 1.0,
+                                    "content": "uncertainties over the target distribution resulting in the predictor to make more reliable class",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                160,
+                                160,
+                                172
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        160,
+                                        160,
+                                        172
+                                    ],
+                                    "score": 1.0,
+                                    "content": "assignments.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        }
+                    ],
+                    "index": 3.5,
+                    "bbox_fs": [
+                        105,
+                        83,
+                        506,
+                        172
+                    ]
+                },
+                {
+                    "type": "table",
+                    "bbox": [
+                        146,
+                        212,
+                        464,
+                        298
+                    ],
+                    "blocks": [
+                        {
+                            "type": "table_caption",
+                            "bbox": [
+                                104,
+                                181,
+                                504,
+                                204
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        179,
+                                        505,
+                                        194
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                179,
+                                                360,
+                                                194
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Table 6: Comparison of CIFAR-10 accuracies for the stochastic",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                360,
+                                                181,
+                                                391,
+                                                192
+                                            ],
+                                            "score": 0.78,
+                                            "content": "\\mathrm { ( N = 1 2 8 ) }",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                391,
+                                                179,
+                                                480,
+                                                194
+                                            ],
+                                            "score": 1.0,
+                                            "content": ") and the deterministic",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                480,
+                                                181,
+                                                502,
+                                                192
+                                            ],
+                                            "score": 0.64,
+                                            "content": "\\left( \\mathrm { N } { = } 1 \\right)",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                502,
+                                                179,
+                                                505,
+                                                194
+                                            ],
+                                            "score": 1.0,
+                                            "content": ")",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 8
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        192,
+                                        270,
+                                        205
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                192,
+                                                270,
+                                                205
+                                            ],
+                                            "score": 1.0,
+                                            "content": "predictor under rectangular missingness.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 9
+                                }
+                            ],
+                            "index": 8.5
+                        },
+                        {
+                            "type": "table_body",
+                            "bbox": [
+                                146,
+                                212,
+                                464,
+                                298
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        146,
+                                        212,
+                                        464,
+                                        298
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                146,
+                                                212,
+                                                464,
+                                                298
+                                            ],
+                                            "score": 0.98,
+                                            "html": "<table><tr><td></td><td colspan=\"6\">Accuracy at Missing Rate (%)</td></tr><tr><td>Method</td><td>20%</td><td>40%</td><td>60%</td><td>70%</td><td>80%</td><td>90%</td></tr><tr><td>GI (N=128)</td><td>84.0</td><td>76.9</td><td>66.1</td><td>59.1</td><td>46.0</td><td>32.1</td></tr><tr><td>GI (N=1)</td><td>83.6</td><td>75.7</td><td>65.1</td><td>56.7</td><td>42.8</td><td>29.4</td></tr><tr><td>% difference (normalized)</td><td>0.5</td><td>1.6</td><td>1.5</td><td>4.1</td><td>6.9</td><td>8.4</td></tr></table>",
+                                            "type": "table",
+                                            "image_path": "ae7388387b96894d1aab2c0bea5caf9ef83886b0f974a93634b3532bca3a445a.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 11,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        146,
+                                        212,
+                                        464,
+                                        240.66666666666666
+                                    ],
+                                    "spans": [],
+                                    "index": 10
+                                },
+                                {
+                                    "bbox": [
+                                        146,
+                                        240.66666666666666,
+                                        464,
+                                        269.3333333333333
+                                    ],
+                                    "spans": [],
+                                    "index": 11
+                                },
+                                {
+                                    "bbox": [
+                                        146,
+                                        269.3333333333333,
+                                        464,
+                                        298.0
+                                    ],
+                                    "spans": [],
+                                    "index": 12
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 9.75
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        81,
+                        240,
+                        94
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                79,
+                                241,
+                                96
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        79,
+                                        241,
+                                        96
+                                    ],
+                                    "score": 1.0,
+                                    "content": "E VISUAL COMPARISON",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        }
+                    ],
+                    "index": 0
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        106,
+                        505,
+                        183
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                105,
+                                506,
+                                119
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        105,
+                                        506,
+                                        119
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Table 7 and 8 provide a visual comparison of GI, MisGAN, and GAIN. For each missingness structure,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                117,
+                                506,
+                                129
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        117,
+                                        506,
+                                        129
+                                    ],
+                                    "score": 1.0,
+                                    "content": "we compare the best two imputation methods based on FID scores in Figure 2 i.e., GI versus MisGAN",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                129,
+                                505,
+                                140
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        129,
+                                        505,
+                                        140
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for rectangular missingness and GI versus GAIN for uniform missingness. From Table 7 it can be",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                139,
+                                506,
+                                151
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        139,
+                                        506,
+                                        151
+                                    ],
+                                    "score": 1.0,
+                                    "content": "seen that GI is more capable in the reconstruction of fine details such as horse legs, car wheels, or",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                150,
+                                505,
+                                163
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        150,
+                                        505,
+                                        163
+                                    ],
+                                    "score": 1.0,
+                                    "content": "plane wings. Regarding the results provided in Table 8, GI imputed samples are generally sharper",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                162,
+                                505,
+                                173
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        162,
+                                        505,
+                                        173
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and more realistic, which is consistent with our hypothesis about the drawbacks of the MSE term in",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                171,
+                                226,
+                                185
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        171,
+                                        226,
+                                        185
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the GAIN objective function.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        }
+                    ],
+                    "index": 4
+                },
+                {
+                    "type": "image",
+                    "bbox": [
+                        109,
+                        221,
+                        498,
+                        644
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                109,
+                                221,
+                                498,
+                                644
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        109,
+                                        221,
+                                        498,
+                                        644
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                109,
+                                                221,
+                                                498,
+                                                644
+                                            ],
+                                            "score": 0.729,
+                                            "type": "image",
+                                            "image_path": "61024c4babfabb7e7dd65df430bff40654a51f9e986c3d422c05dfe4e0963228.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 9,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        109,
+                                        221,
+                                        498,
+                                        362.0
+                                    ],
+                                    "spans": [],
+                                    "index": 8
+                                },
+                                {
+                                    "bbox": [
+                                        109,
+                                        362.0,
+                                        498,
+                                        503.0
+                                    ],
+                                    "spans": [],
+                                    "index": 9
+                                },
+                                {
+                                    "bbox": [
+                                        109,
+                                        503.0,
+                                        498,
+                                        644.0
+                                    ],
+                                    "spans": [],
+                                    "index": 10
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 9
+                }
+            ],
+            "page_idx": 17,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        107,
+                        27,
+                        308,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                107,
+                                26,
+                                308,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        26,
+                                        308,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Under review as a conference paper at ICLR 2020",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        300,
+                        751,
+                        311,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                299,
+                                750,
+                                312,
+                                763
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        299,
+                                        750,
+                                        312,
+                                        763
+                                    ],
+                                    "score": 1.0,
+                                    "content": "18",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        81,
+                        240,
+                        94
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                79,
+                                241,
+                                96
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        79,
+                                        241,
+                                        96
+                                    ],
+                                    "score": 1.0,
+                                    "content": "E VISUAL COMPARISON",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        }
+                    ],
+                    "index": 0
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        106,
+                        505,
+                        183
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                105,
+                                506,
+                                119
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        105,
+                                        506,
+                                        119
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Table 7 and 8 provide a visual comparison of GI, MisGAN, and GAIN. For each missingness structure,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                117,
+                                506,
+                                129
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        117,
+                                        506,
+                                        129
+                                    ],
+                                    "score": 1.0,
+                                    "content": "we compare the best two imputation methods based on FID scores in Figure 2 i.e., GI versus MisGAN",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                129,
+                                505,
+                                140
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        129,
+                                        505,
+                                        140
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for rectangular missingness and GI versus GAIN for uniform missingness. From Table 7 it can be",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                139,
+                                506,
+                                151
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        139,
+                                        506,
+                                        151
+                                    ],
+                                    "score": 1.0,
+                                    "content": "seen that GI is more capable in the reconstruction of fine details such as horse legs, car wheels, or",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                150,
+                                505,
+                                163
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        150,
+                                        505,
+                                        163
+                                    ],
+                                    "score": 1.0,
+                                    "content": "plane wings. Regarding the results provided in Table 8, GI imputed samples are generally sharper",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                162,
+                                505,
+                                173
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        162,
+                                        505,
+                                        173
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and more realistic, which is consistent with our hypothesis about the drawbacks of the MSE term in",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                171,
+                                226,
+                                185
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        171,
+                                        226,
+                                        185
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the GAIN objective function.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        }
+                    ],
+                    "index": 4,
+                    "bbox_fs": [
+                        105,
+                        105,
+                        506,
+                        185
+                    ]
+                },
+                {
+                    "type": "image",
+                    "bbox": [
+                        109,
+                        221,
+                        498,
+                        644
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                109,
+                                221,
+                                498,
+                                644
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        109,
+                                        221,
+                                        498,
+                                        644
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                109,
+                                                221,
+                                                498,
+                                                644
+                                            ],
+                                            "score": 0.729,
+                                            "type": "image",
+                                            "image_path": "61024c4babfabb7e7dd65df430bff40654a51f9e986c3d422c05dfe4e0963228.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 9,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        109,
+                                        221,
+                                        498,
+                                        362.0
+                                    ],
+                                    "spans": [],
+                                    "index": 8
+                                },
+                                {
+                                    "bbox": [
+                                        109,
+                                        362.0,
+                                        498,
+                                        503.0
+                                    ],
+                                    "spans": [],
+                                    "index": 9
+                                },
+                                {
+                                    "bbox": [
+                                        109,
+                                        503.0,
+                                        498,
+                                        644.0
+                                    ],
+                                    "spans": [],
+                                    "index": 10
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 9
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "image",
+                    "bbox": [
+                        109,
+                        200,
+                        498,
+                        635
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                109,
+                                200,
+                                498,
+                                635
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        109,
+                                        200,
+                                        498,
+                                        635
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                109,
+                                                200,
+                                                498,
+                                                635
+                                            ],
+                                            "score": 0.789,
+                                            "type": "image",
+                                            "image_path": "3958733293402574782f6e1c021c4a754f6dbed20e2753d3f34209f404938a06.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 1,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        109,
+                                        200,
+                                        498,
+                                        345.0
+                                    ],
+                                    "spans": [],
+                                    "index": 0
+                                },
+                                {
+                                    "bbox": [
+                                        109,
+                                        345.0,
+                                        498,
+                                        490.0
+                                    ],
+                                    "spans": [],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        109,
+                                        490.0,
+                                        498,
+                                        635.0
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 1
+                }
+            ],
+            "page_idx": 18,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        106,
+                        27,
+                        309,
+                        38
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                25,
+                                308,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        25,
+                                        308,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Under review as a conference paper at ICLR 2020",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        300,
+                        751,
+                        311,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                299,
+                                750,
+                                312,
+                                765
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        299,
+                                        750,
+                                        312,
+                                        765
+                                    ],
+                                    "score": 1.0,
+                                    "content": "19",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "image",
+                    "bbox": [
+                        109,
+                        200,
+                        498,
+                        635
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                109,
+                                200,
+                                498,
+                                635
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        109,
+                                        200,
+                                        498,
+                                        635
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                109,
+                                                200,
+                                                498,
+                                                635
+                                            ],
+                                            "score": 0.789,
+                                            "type": "image",
+                                            "image_path": "3958733293402574782f6e1c021c4a754f6dbed20e2753d3f34209f404938a06.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 1,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        109,
+                                        200,
+                                        498,
+                                        345.0
+                                    ],
+                                    "spans": [],
+                                    "index": 0
+                                },
+                                {
+                                    "bbox": [
+                                        109,
+                                        345.0,
+                                        498,
+                                        490.0
+                                    ],
+                                    "spans": [],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        109,
+                                        490.0,
+                                        498,
+                                        635.0
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 1
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        81,
+                        312,
+                        94
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                80,
+                                314,
+                                96
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        80,
+                                        314,
+                                        96
+                                    ],
+                                    "score": 1.0,
+                                    "content": "F ANALYSIS OF THE RMSE MEASURE",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        }
+                    ],
+                    "index": 0
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        106,
+                        505,
+                        195
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                105,
+                                506,
+                                119
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        105,
+                                        506,
+                                        119
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Table 9 presents the comparison of different imputation methods using the RMSE measure on",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                117,
+                                505,
+                                129
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        117,
+                                        505,
+                                        129
+                                    ],
+                                    "score": 1.0,
+                                    "content": "CIFAR-10 for different missing structures and rates. Generally, RMSE values for the uniform",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                128,
+                                506,
+                                140
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        128,
+                                        506,
+                                        140
+                                    ],
+                                    "score": 1.0,
+                                    "content": "missing structure are lower than their rectangular counterparts. It is consistent with our intuition that",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                139,
+                                506,
+                                151
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        139,
+                                        506,
+                                        151
+                                    ],
+                                    "score": 1.0,
+                                    "content": "imputing uniform missingness is most similar to denoising problems where the RMSE measure is",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                149,
+                                506,
+                                164
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        149,
+                                        506,
+                                        164
+                                    ],
+                                    "score": 1.0,
+                                    "content": "frequently used. Additionally, comparing the performance of different imputation methods using",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                161,
+                                506,
+                                173
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        161,
+                                        506,
+                                        173
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the FID measure (Section 4.4) does not demonstrate a clear correlation to results shown in Table 9.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                172,
+                                506,
+                                184
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        172,
+                                        506,
+                                        184
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Nonetheless, it is well-known that the FID measure is more suited to measuring the performance of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                183,
+                                395,
+                                195
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        183,
+                                        395,
+                                        195
+                                    ],
+                                    "score": 1.0,
+                                    "content": "generated images from the underlying distribution (Heusel et al., 2017).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        }
+                    ],
+                    "index": 4.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        199,
+                        505,
+                        255
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                199,
+                                507,
+                                213
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        199,
+                                        507,
+                                        213
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Similarly, in Table 10, we provide RMSE values corresponding to experiments on the tabular datasets.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                209,
+                                506,
+                                223
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        209,
+                                        506,
+                                        223
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Here, GAIN and DAE provide very similar results that are generally better than GI or MisGAN. This",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                222,
+                                505,
+                                234
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        222,
+                                        505,
+                                        234
+                                    ],
+                                    "score": 1.0,
+                                    "content": "signifies our hypothesis that the MSE loss term may skew generated samples toward the mean of the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                231,
+                                506,
+                                245
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        231,
+                                        506,
+                                        245
+                                    ],
+                                    "score": 1.0,
+                                    "content": "distribution, resulting in better RMSE values but not necessarily higher final classification accuracies",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                242,
+                                162,
+                                257
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        242,
+                                        162,
+                                        257
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(see Table 2).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        }
+                    ],
+                    "index": 11
+                },
+                {
+                    "type": "table",
+                    "bbox": [
+                        108,
+                        298,
+                        503,
+                        400
+                    ],
+                    "blocks": [
+                        {
+                            "type": "table_caption",
+                            "bbox": [
+                                107,
+                                264,
+                                503,
+                                288
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        264,
+                                        505,
+                                        277
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                264,
+                                                505,
+                                                277
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Table 9: Comparison of imputation RMSE values for CIFAR-10 at different missing structures and",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 14
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        276,
+                                        131,
+                                        289
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                276,
+                                                131,
+                                                289
+                                            ],
+                                            "score": 1.0,
+                                            "content": "rates.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 15
+                                }
+                            ],
+                            "index": 14.5
+                        },
+                        {
+                            "type": "table_body",
+                            "bbox": [
+                                108,
+                                298,
+                                503,
+                                400
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        108,
+                                        298,
+                                        503,
+                                        400
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                108,
+                                                298,
+                                                503,
+                                                400
+                                            ],
+                                            "score": 0.982,
+                                            "html": "<table><tr><td></td><td colspan=\"5\">RMSE at Missing Rate (%)</td><td></td></tr><tr><td></td><td colspan=\"5\">MCAR Uniform</td><td>MCAR Rect.</td></tr><tr><td>Method</td><td>20%</td><td>40%</td><td>60%</td><td>20%</td><td>40%</td><td>60%</td></tr><tr><td>GI</td><td>0.026 (±0.003)</td><td>0.057 (±0.008)</td><td>0.090 (±0.006)</td><td>0.097 (±0.02)</td><td>0.148 (±0.001)</td><td>0.660 (±0.010)</td></tr><tr><td>MisGAN</td><td>0.079 (±0.001)</td><td>0.161 (±0.001)</td><td>0.257 (±0.002)</td><td>0.106 (±0.005)</td><td>0.158 (±0.004)</td><td>0.250 (±0.001)</td></tr><tr><td>GAIN</td><td>0.027 (±0.003)</td><td>0.045 (±0.001)</td><td>0.072 (±0.005)</td><td>0.340 (±0.047)</td><td>0.511 (±0.001)</td><td>0.660 (±0.010)</td></tr><tr><td>DAE</td><td>0.036 (±0.001)</td><td>0.075 (±0.002)</td><td>0.121 (±0.005)</td><td>0.116 (±0.007)</td><td>0.160 (±0.001)</td><td>0.233 (±0.029)</td></tr></table>",
+                                            "type": "table",
+                                            "image_path": "529c815376d6c496fdf132f191539c1f32c20d0235e9256fc3ebe56ba40b8e1f.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 17,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        108,
+                                        298,
+                                        503,
+                                        332.0
+                                    ],
+                                    "spans": [],
+                                    "index": 16
+                                },
+                                {
+                                    "bbox": [
+                                        108,
+                                        332.0,
+                                        503,
+                                        366.0
+                                    ],
+                                    "spans": [],
+                                    "index": 17
+                                },
+                                {
+                                    "bbox": [
+                                        108,
+                                        366.0,
+                                        503,
+                                        400.0
+                                    ],
+                                    "spans": [],
+                                    "index": 18
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 15.75
+                },
+                {
+                    "type": "table",
+                    "bbox": [
+                        108,
+                        450,
+                        504,
+                        630
+                    ],
+                    "blocks": [
+                        {
+                            "type": "table_caption",
+                            "bbox": [
+                                105,
+                                417,
+                                504,
+                                440
+                            ],
+                            "group_id": 1,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        416,
+                                        506,
+                                        430
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                416,
+                                                506,
+                                                430
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Table 10: Comparison of imputation RMSE values for Landsat, MIT-BIH, and Diabetes datasets at",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 19
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        428,
+                                        200,
+                                        441
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                428,
+                                                200,
+                                                441
+                                            ],
+                                            "score": 1.0,
+                                            "content": "different missing rates.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 20
+                                }
+                            ],
+                            "index": 19.5
+                        },
+                        {
+                            "type": "table_body",
+                            "bbox": [
+                                108,
+                                450,
+                                504,
+                                630
+                            ],
+                            "group_id": 1,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        108,
+                                        450,
+                                        504,
+                                        630
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                108,
+                                                450,
+                                                504,
+                                                630
+                                            ],
+                                            "score": 0.985,
+                                            "html": "<table><tr><td colspan=\"2\"></td><td colspan=\"4\">RMSE at Missing Rate (%)</td></tr><tr><td>Dataset</td><td>Method</td><td>10%</td><td>20%</td><td>30%</td><td>40%</td></tr><tr><td rowspan=\"4\">Landsat (Dua &amp; Graff,2017)</td><td>GI</td><td>0.040 (±0.005)</td><td>0.067 (±0.007)</td><td>0.076 (±0.020)</td><td>0.136 (±0.002)</td></tr><tr><td>MisGAN</td><td>0.068 (±0.001)</td><td>0.096 (±0.001)</td><td>0.118 (±0.001)</td><td>0.136 (±0.001)</td></tr><tr><td>GAIN</td><td>0.018 (±0.001)</td><td>0.024 (±0.001)</td><td>0.030 (±0.001)</td><td>0.037 (±0.001)</td></tr><tr><td>DAE</td><td>0.020 (±0.001)</td><td>0.031 (±0.001)</td><td>0.041 (±0.001)</td><td>0.052 (±0.001)</td></tr><tr><td rowspan=\"4\">MIT-BIH(Moody &amp; Mark,2001)</td><td>GI</td><td>0.038 (±0.001)</td><td>0.060 (±0.004)</td><td>0.071 (±0.002)</td><td>0.095 (±0.002)</td></tr><tr><td>MisGAN</td><td>0.073 (±0.007)</td><td>0.092 (±0.002)</td><td>0.115 (±0.003)</td><td>0.111 (±0.001)</td></tr><tr><td>GAIN</td><td>0.032 (±0.008)</td><td>0.046 (±0.001)</td><td>0.055 (±0.004)</td><td>0.067 (±0.007)</td></tr><tr><td>DAE</td><td>0.029 (±0.001)</td><td>0.048 (±0.008)</td><td>0.061 (±0.009)</td><td>0.068 (±0.003)</td></tr><tr><td rowspan=\"4\">Diabetes (Kachuee et al., 2019)</td><td>GI</td><td>0.080 (±0.002)</td><td>0.118 (±0.008)</td><td>0.149 (±0.020)</td><td>0.189 (±0.009)</td></tr><tr><td>MisGAN</td><td>0.082 (±0.004)</td><td>0.111 (±0.002)</td><td>0.133 (±0.001)</td><td>0.151 (±0.001)</td></tr><tr><td>GAIN</td><td>0.064 (±0.001)</td><td>0.092 (±0.001)</td><td>0.119 (±0.001)</td><td>0.140 (±0.001)</td></tr><tr><td>DAE</td><td>0.065 (±0.001)</td><td>0.093 (±0.001)</td><td>0.118 (±0.001)</td><td>0.143 (±0.001)</td></tr></table>",
+                                            "type": "table",
+                                            "image_path": "5a52e3e2e067bb0f4bbb49239b9639bda2bb046e5e27aa435b8404de869a0a9b.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 22,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        108,
+                                        450,
+                                        504,
+                                        510.0
+                                    ],
+                                    "spans": [],
+                                    "index": 21
+                                },
+                                {
+                                    "bbox": [
+                                        108,
+                                        510.0,
+                                        504,
+                                        570.0
+                                    ],
+                                    "spans": [],
+                                    "index": 22
+                                },
+                                {
+                                    "bbox": [
+                                        108,
+                                        570.0,
+                                        504,
+                                        630.0
+                                    ],
+                                    "spans": [],
+                                    "index": 23
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 20.75
+                }
+            ],
+            "page_idx": 19,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        107,
+                        27,
+                        308,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                107,
+                                26,
+                                308,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        26,
+                                        308,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Under review as a conference paper at ICLR 2020",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        300,
+                        751,
+                        311,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                298,
+                                749,
+                                313,
+                                764
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        298,
+                                        749,
+                                        313,
+                                        764
+                                    ],
+                                    "score": 1.0,
+                                    "content": "",
+                                    "type": "text",
+                                    "height": 15,
+                                    "width": 15
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        81,
+                        312,
+                        94
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                80,
+                                314,
+                                96
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        80,
+                                        314,
+                                        96
+                                    ],
+                                    "score": 1.0,
+                                    "content": "F ANALYSIS OF THE RMSE MEASURE",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        }
+                    ],
+                    "index": 0
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        106,
+                        505,
+                        195
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                105,
+                                506,
+                                119
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        105,
+                                        506,
+                                        119
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Table 9 presents the comparison of different imputation methods using the RMSE measure on",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                117,
+                                505,
+                                129
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        117,
+                                        505,
+                                        129
+                                    ],
+                                    "score": 1.0,
+                                    "content": "CIFAR-10 for different missing structures and rates. Generally, RMSE values for the uniform",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                128,
+                                506,
+                                140
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        128,
+                                        506,
+                                        140
+                                    ],
+                                    "score": 1.0,
+                                    "content": "missing structure are lower than their rectangular counterparts. It is consistent with our intuition that",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                139,
+                                506,
+                                151
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        139,
+                                        506,
+                                        151
+                                    ],
+                                    "score": 1.0,
+                                    "content": "imputing uniform missingness is most similar to denoising problems where the RMSE measure is",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                149,
+                                506,
+                                164
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        149,
+                                        506,
+                                        164
+                                    ],
+                                    "score": 1.0,
+                                    "content": "frequently used. Additionally, comparing the performance of different imputation methods using",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                161,
+                                506,
+                                173
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        161,
+                                        506,
+                                        173
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the FID measure (Section 4.4) does not demonstrate a clear correlation to results shown in Table 9.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                172,
+                                506,
+                                184
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        172,
+                                        506,
+                                        184
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Nonetheless, it is well-known that the FID measure is more suited to measuring the performance of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                183,
+                                395,
+                                195
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        183,
+                                        395,
+                                        195
+                                    ],
+                                    "score": 1.0,
+                                    "content": "generated images from the underlying distribution (Heusel et al., 2017).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        }
+                    ],
+                    "index": 4.5,
+                    "bbox_fs": [
+                        105,
+                        105,
+                        506,
+                        195
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        199,
+                        505,
+                        255
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                199,
+                                507,
+                                213
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        199,
+                                        507,
+                                        213
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Similarly, in Table 10, we provide RMSE values corresponding to experiments on the tabular datasets.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                209,
+                                506,
+                                223
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        209,
+                                        506,
+                                        223
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Here, GAIN and DAE provide very similar results that are generally better than GI or MisGAN. This",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                222,
+                                505,
+                                234
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        222,
+                                        505,
+                                        234
+                                    ],
+                                    "score": 1.0,
+                                    "content": "signifies our hypothesis that the MSE loss term may skew generated samples toward the mean of the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                231,
+                                506,
+                                245
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        231,
+                                        506,
+                                        245
+                                    ],
+                                    "score": 1.0,
+                                    "content": "distribution, resulting in better RMSE values but not necessarily higher final classification accuracies",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                242,
+                                162,
+                                257
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        242,
+                                        162,
+                                        257
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(see Table 2).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        }
+                    ],
+                    "index": 11,
+                    "bbox_fs": [
+                        105,
+                        199,
+                        507,
+                        257
+                    ]
+                },
+                {
+                    "type": "table",
+                    "bbox": [
+                        108,
+                        298,
+                        503,
+                        400
+                    ],
+                    "blocks": [
+                        {
+                            "type": "table_caption",
+                            "bbox": [
+                                107,
+                                264,
+                                503,
+                                288
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        264,
+                                        505,
+                                        277
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                264,
+                                                505,
+                                                277
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Table 9: Comparison of imputation RMSE values for CIFAR-10 at different missing structures and",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 14
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        276,
+                                        131,
+                                        289
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                276,
+                                                131,
+                                                289
+                                            ],
+                                            "score": 1.0,
+                                            "content": "rates.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 15
+                                }
+                            ],
+                            "index": 14.5
+                        },
+                        {
+                            "type": "table_body",
+                            "bbox": [
+                                108,
+                                298,
+                                503,
+                                400
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        108,
+                                        298,
+                                        503,
+                                        400
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                108,
+                                                298,
+                                                503,
+                                                400
+                                            ],
+                                            "score": 0.982,
+                                            "html": "<table><tr><td></td><td colspan=\"5\">RMSE at Missing Rate (%)</td><td></td></tr><tr><td></td><td colspan=\"5\">MCAR Uniform</td><td>MCAR Rect.</td></tr><tr><td>Method</td><td>20%</td><td>40%</td><td>60%</td><td>20%</td><td>40%</td><td>60%</td></tr><tr><td>GI</td><td>0.026 (±0.003)</td><td>0.057 (±0.008)</td><td>0.090 (±0.006)</td><td>0.097 (±0.02)</td><td>0.148 (±0.001)</td><td>0.660 (±0.010)</td></tr><tr><td>MisGAN</td><td>0.079 (±0.001)</td><td>0.161 (±0.001)</td><td>0.257 (±0.002)</td><td>0.106 (±0.005)</td><td>0.158 (±0.004)</td><td>0.250 (±0.001)</td></tr><tr><td>GAIN</td><td>0.027 (±0.003)</td><td>0.045 (±0.001)</td><td>0.072 (±0.005)</td><td>0.340 (±0.047)</td><td>0.511 (±0.001)</td><td>0.660 (±0.010)</td></tr><tr><td>DAE</td><td>0.036 (±0.001)</td><td>0.075 (±0.002)</td><td>0.121 (±0.005)</td><td>0.116 (±0.007)</td><td>0.160 (±0.001)</td><td>0.233 (±0.029)</td></tr></table>",
+                                            "type": "table",
+                                            "image_path": "529c815376d6c496fdf132f191539c1f32c20d0235e9256fc3ebe56ba40b8e1f.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 17,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        108,
+                                        298,
+                                        503,
+                                        332.0
+                                    ],
+                                    "spans": [],
+                                    "index": 16
+                                },
+                                {
+                                    "bbox": [
+                                        108,
+                                        332.0,
+                                        503,
+                                        366.0
+                                    ],
+                                    "spans": [],
+                                    "index": 17
+                                },
+                                {
+                                    "bbox": [
+                                        108,
+                                        366.0,
+                                        503,
+                                        400.0
+                                    ],
+                                    "spans": [],
+                                    "index": 18
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 15.75
+                },
+                {
+                    "type": "table",
+                    "bbox": [
+                        108,
+                        450,
+                        504,
+                        630
+                    ],
+                    "blocks": [
+                        {
+                            "type": "table_caption",
+                            "bbox": [
+                                105,
+                                417,
+                                504,
+                                440
+                            ],
+                            "group_id": 1,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        416,
+                                        506,
+                                        430
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                416,
+                                                506,
+                                                430
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Table 10: Comparison of imputation RMSE values for Landsat, MIT-BIH, and Diabetes datasets at",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 19
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        428,
+                                        200,
+                                        441
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                428,
+                                                200,
+                                                441
+                                            ],
+                                            "score": 1.0,
+                                            "content": "different missing rates.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 20
+                                }
+                            ],
+                            "index": 19.5
+                        },
+                        {
+                            "type": "table_body",
+                            "bbox": [
+                                108,
+                                450,
+                                504,
+                                630
+                            ],
+                            "group_id": 1,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        108,
+                                        450,
+                                        504,
+                                        630
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                108,
+                                                450,
+                                                504,
+                                                630
+                                            ],
+                                            "score": 0.985,
+                                            "html": "<table><tr><td colspan=\"2\"></td><td colspan=\"4\">RMSE at Missing Rate (%)</td></tr><tr><td>Dataset</td><td>Method</td><td>10%</td><td>20%</td><td>30%</td><td>40%</td></tr><tr><td rowspan=\"4\">Landsat (Dua &amp; Graff,2017)</td><td>GI</td><td>0.040 (±0.005)</td><td>0.067 (±0.007)</td><td>0.076 (±0.020)</td><td>0.136 (±0.002)</td></tr><tr><td>MisGAN</td><td>0.068 (±0.001)</td><td>0.096 (±0.001)</td><td>0.118 (±0.001)</td><td>0.136 (±0.001)</td></tr><tr><td>GAIN</td><td>0.018 (±0.001)</td><td>0.024 (±0.001)</td><td>0.030 (±0.001)</td><td>0.037 (±0.001)</td></tr><tr><td>DAE</td><td>0.020 (±0.001)</td><td>0.031 (±0.001)</td><td>0.041 (±0.001)</td><td>0.052 (±0.001)</td></tr><tr><td rowspan=\"4\">MIT-BIH(Moody &amp; Mark,2001)</td><td>GI</td><td>0.038 (±0.001)</td><td>0.060 (±0.004)</td><td>0.071 (±0.002)</td><td>0.095 (±0.002)</td></tr><tr><td>MisGAN</td><td>0.073 (±0.007)</td><td>0.092 (±0.002)</td><td>0.115 (±0.003)</td><td>0.111 (±0.001)</td></tr><tr><td>GAIN</td><td>0.032 (±0.008)</td><td>0.046 (±0.001)</td><td>0.055 (±0.004)</td><td>0.067 (±0.007)</td></tr><tr><td>DAE</td><td>0.029 (±0.001)</td><td>0.048 (±0.008)</td><td>0.061 (±0.009)</td><td>0.068 (±0.003)</td></tr><tr><td rowspan=\"4\">Diabetes (Kachuee et al., 2019)</td><td>GI</td><td>0.080 (±0.002)</td><td>0.118 (±0.008)</td><td>0.149 (±0.020)</td><td>0.189 (±0.009)</td></tr><tr><td>MisGAN</td><td>0.082 (±0.004)</td><td>0.111 (±0.002)</td><td>0.133 (±0.001)</td><td>0.151 (±0.001)</td></tr><tr><td>GAIN</td><td>0.064 (±0.001)</td><td>0.092 (±0.001)</td><td>0.119 (±0.001)</td><td>0.140 (±0.001)</td></tr><tr><td>DAE</td><td>0.065 (±0.001)</td><td>0.093 (±0.001)</td><td>0.118 (±0.001)</td><td>0.143 (±0.001)</td></tr></table>",
+                                            "type": "table",
+                                            "image_path": "5a52e3e2e067bb0f4bbb49239b9639bda2bb046e5e27aa435b8404de869a0a9b.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 22,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        108,
+                                        450,
+                                        504,
+                                        510.0
+                                    ],
+                                    "spans": [],
+                                    "index": 21
+                                },
+                                {
+                                    "bbox": [
+                                        108,
+                                        510.0,
+                                        504,
+                                        570.0
+                                    ],
+                                    "spans": [],
+                                    "index": 22
+                                },
+                                {
+                                    "bbox": [
+                                        108,
+                                        570.0,
+                                        504,
+                                        630.0
+                                    ],
+                                    "spans": [],
+                                    "index": 23
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 20.75
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        81,
+                        276,
+                        94
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                80,
+                                277,
+                                96
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        80,
+                                        277,
+                                        96
+                                    ],
+                                    "score": 1.0,
+                                    "content": "G IMPACT OF TRAINING NOISE",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        }
+                    ],
+                    "index": 0
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        106,
+                        505,
+                        172
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                106,
+                                505,
+                                119
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        106,
+                                        505,
+                                        119
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Addition of noise to input vectors often serves as an input augmentation and results in improved",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                117,
+                                505,
+                                129
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        117,
+                                        505,
+                                        129
+                                    ],
+                                    "score": 1.0,
+                                    "content": "generalization accuracies. In order to verify that the improved GI performance is not merely due",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                126,
+                                506,
+                                142
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        126,
+                                        506,
+                                        142
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to the introduction of noise in the suggested architecture, we conducted an experiment by adding",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                138,
+                                505,
+                                153
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        138,
+                                        505,
+                                        153
+                                    ],
+                                    "score": 1.0,
+                                    "content": "different amounts of Gaussian noise during the training process for GAIN and GI. Specifically, we",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                150,
+                                505,
+                                163
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        150,
+                                        505,
+                                        163
+                                    ],
+                                    "score": 1.0,
+                                    "content": "compared how the CIFAR-10 test accuracies change at different degrees of training noise for uniform",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                160,
+                                401,
+                                174
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        160,
+                                        376,
+                                        174
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and rectangular missingess structures at the average missing rate of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        377,
+                                        161,
+                                        396,
+                                        171
+                                    ],
+                                    "score": 0.87,
+                                    "content": "40 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        397,
+                                        160,
+                                        401,
+                                        174
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        }
+                    ],
+                    "index": 3.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        177,
+                        505,
+                        233
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                177,
+                                505,
+                                191
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        177,
+                                        505,
+                                        191
+                                    ],
+                                    "score": 1.0,
+                                    "content": "According to Table 11, adding small amounts of Gaussian noise (e.g., std=0.0125) improves the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                188,
+                                506,
+                                201
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        188,
+                                        506,
+                                        201
+                                    ],
+                                    "score": 1.0,
+                                    "content": "generalization under uniform missingness for both GI and GAIN. Even in this case, GI is still",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                200,
+                                506,
+                                212
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        200,
+                                        506,
+                                        212
+                                    ],
+                                    "score": 1.0,
+                                    "content": "outperforming GAIN in terms of final classification performance. It is also interesting to point out",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                209,
+                                506,
+                                223
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        209,
+                                        506,
+                                        223
+                                    ],
+                                    "score": 1.0,
+                                    "content": "that for the case of rectangular missingness adding Gaussian noise results in a consistent reduction in",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                222,
+                                286,
+                                234
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        222,
+                                        286,
+                                        234
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the classification accuracy for both methods.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        }
+                    ],
+                    "index": 9
+                },
+                {
+                    "type": "table",
+                    "bbox": [
+                        172,
+                        264,
+                        439,
+                        396
+                    ],
+                    "blocks": [
+                        {
+                            "type": "table_caption",
+                            "bbox": [
+                                105,
+                                243,
+                                501,
+                                255
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        241,
+                                        502,
+                                        257
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                107,
+                                                241,
+                                                321,
+                                                257
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Table 11: Top-1 CIFAR-10 classification accuracy at",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                321,
+                                                244,
+                                                341,
+                                                254
+                                            ],
+                                            "score": 0.86,
+                                            "content": "40 \\%",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                341,
+                                                241,
+                                                502,
+                                                257
+                                            ],
+                                            "score": 1.0,
+                                            "content": "missing rate using added training noise.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 12
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "type": "table_body",
+                            "bbox": [
+                                172,
+                                264,
+                                439,
+                                396
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        172,
+                                        264,
+                                        439,
+                                        396
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                172,
+                                                264,
+                                                439,
+                                                396
+                                            ],
+                                            "score": 0.983,
+                                            "html": "<table><tr><td></td><td colspan=\"2\">Accuracy (%)</td></tr><tr><td>Noise STD</td><td>MCAR Uniform (40%) GI GAIN</td><td>MCAR Rect. (40%) GI GAIN</td></tr><tr><td>0.0</td><td>87.1 86.0</td><td>76.9 73.6</td></tr><tr><td>0.0125</td><td>87.3 86.3</td><td>76.8 73.3</td></tr><tr><td>0.025</td><td>86.5 86.6</td><td>76.7 73.2</td></tr><tr><td>0.05</td><td>85.6 84.7</td><td>73.7 72.4</td></tr><tr><td>0.1</td><td>82.0 80.6</td><td>68.7 67.0</td></tr></table>",
+                                            "type": "table",
+                                            "image_path": "d046f8a6d59738370e36a4b7922d58dd7a1f1787bb18517afbe2533c0193e688.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 14,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        172,
+                                        264,
+                                        439,
+                                        308.0
+                                    ],
+                                    "spans": [],
+                                    "index": 13
+                                },
+                                {
+                                    "bbox": [
+                                        172,
+                                        308.0,
+                                        439,
+                                        352.0
+                                    ],
+                                    "spans": [],
+                                    "index": 14
+                                },
+                                {
+                                    "bbox": [
+                                        172,
+                                        352.0,
+                                        439,
+                                        396.0
+                                    ],
+                                    "spans": [],
+                                    "index": 15
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 13.0
+                }
+            ],
+            "page_idx": 20,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        107,
+                        27,
+                        308,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                107,
+                                26,
+                                308,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        26,
+                                        308,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Under review as a conference paper at ICLR 2020",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        300,
+                        751,
+                        310,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                298,
+                                750,
+                                312,
+                                764
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        298,
+                                        750,
+                                        312,
+                                        764
+                                    ],
+                                    "score": 1.0,
+                                    "content": "",
+                                    "type": "text",
+                                    "height": 14,
+                                    "width": 14
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        81,
+                        276,
+                        94
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                80,
+                                277,
+                                96
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        80,
+                                        277,
+                                        96
+                                    ],
+                                    "score": 1.0,
+                                    "content": "G IMPACT OF TRAINING NOISE",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        }
+                    ],
+                    "index": 0
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        106,
+                        505,
+                        172
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                106,
+                                505,
+                                119
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        106,
+                                        505,
+                                        119
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Addition of noise to input vectors often serves as an input augmentation and results in improved",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                117,
+                                505,
+                                129
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        117,
+                                        505,
+                                        129
+                                    ],
+                                    "score": 1.0,
+                                    "content": "generalization accuracies. In order to verify that the improved GI performance is not merely due",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                126,
+                                506,
+                                142
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        126,
+                                        506,
+                                        142
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to the introduction of noise in the suggested architecture, we conducted an experiment by adding",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                138,
+                                505,
+                                153
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        138,
+                                        505,
+                                        153
+                                    ],
+                                    "score": 1.0,
+                                    "content": "different amounts of Gaussian noise during the training process for GAIN and GI. Specifically, we",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                150,
+                                505,
+                                163
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        150,
+                                        505,
+                                        163
+                                    ],
+                                    "score": 1.0,
+                                    "content": "compared how the CIFAR-10 test accuracies change at different degrees of training noise for uniform",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                160,
+                                401,
+                                174
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        160,
+                                        376,
+                                        174
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and rectangular missingess structures at the average missing rate of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        377,
+                                        161,
+                                        396,
+                                        171
+                                    ],
+                                    "score": 0.87,
+                                    "content": "40 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        397,
+                                        160,
+                                        401,
+                                        174
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        }
+                    ],
+                    "index": 3.5,
+                    "bbox_fs": [
+                        104,
+                        106,
+                        506,
+                        174
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        177,
+                        505,
+                        233
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                177,
+                                505,
+                                191
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        177,
+                                        505,
+                                        191
+                                    ],
+                                    "score": 1.0,
+                                    "content": "According to Table 11, adding small amounts of Gaussian noise (e.g., std=0.0125) improves the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                188,
+                                506,
+                                201
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        188,
+                                        506,
+                                        201
+                                    ],
+                                    "score": 1.0,
+                                    "content": "generalization under uniform missingness for both GI and GAIN. Even in this case, GI is still",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                200,
+                                506,
+                                212
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        200,
+                                        506,
+                                        212
+                                    ],
+                                    "score": 1.0,
+                                    "content": "outperforming GAIN in terms of final classification performance. It is also interesting to point out",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                209,
+                                506,
+                                223
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        209,
+                                        506,
+                                        223
+                                    ],
+                                    "score": 1.0,
+                                    "content": "that for the case of rectangular missingness adding Gaussian noise results in a consistent reduction in",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                222,
+                                286,
+                                234
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        222,
+                                        286,
+                                        234
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the classification accuracy for both methods.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        }
+                    ],
+                    "index": 9,
+                    "bbox_fs": [
+                        105,
+                        177,
+                        506,
+                        234
+                    ]
+                },
+                {
+                    "type": "table",
+                    "bbox": [
+                        172,
+                        264,
+                        439,
+                        396
+                    ],
+                    "blocks": [
+                        {
+                            "type": "table_caption",
+                            "bbox": [
+                                105,
+                                243,
+                                501,
+                                255
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        241,
+                                        502,
+                                        257
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                107,
+                                                241,
+                                                321,
+                                                257
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Table 11: Top-1 CIFAR-10 classification accuracy at",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                321,
+                                                244,
+                                                341,
+                                                254
+                                            ],
+                                            "score": 0.86,
+                                            "content": "40 \\%",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                341,
+                                                241,
+                                                502,
+                                                257
+                                            ],
+                                            "score": 1.0,
+                                            "content": "missing rate using added training noise.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 12
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "type": "table_body",
+                            "bbox": [
+                                172,
+                                264,
+                                439,
+                                396
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        172,
+                                        264,
+                                        439,
+                                        396
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                172,
+                                                264,
+                                                439,
+                                                396
+                                            ],
+                                            "score": 0.983,
+                                            "html": "<table><tr><td></td><td colspan=\"2\">Accuracy (%)</td></tr><tr><td>Noise STD</td><td>MCAR Uniform (40%) GI GAIN</td><td>MCAR Rect. (40%) GI GAIN</td></tr><tr><td>0.0</td><td>87.1 86.0</td><td>76.9 73.6</td></tr><tr><td>0.0125</td><td>87.3 86.3</td><td>76.8 73.3</td></tr><tr><td>0.025</td><td>86.5 86.6</td><td>76.7 73.2</td></tr><tr><td>0.05</td><td>85.6 84.7</td><td>73.7 72.4</td></tr><tr><td>0.1</td><td>82.0 80.6</td><td>68.7 67.0</td></tr></table>",
+                                            "type": "table",
+                                            "image_path": "d046f8a6d59738370e36a4b7922d58dd7a1f1787bb18517afbe2533c0193e688.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 14,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        172,
+                                        264,
+                                        439,
+                                        308.0
+                                    ],
+                                    "spans": [],
+                                    "index": 13
+                                },
+                                {
+                                    "bbox": [
+                                        172,
+                                        308.0,
+                                        439,
+                                        352.0
+                                    ],
+                                    "spans": [],
+                                    "index": 14
+                                },
+                                {
+                                    "bbox": [
+                                        172,
+                                        352.0,
+                                        439,
+                                        396.0
+                                    ],
+                                    "spans": [],
+                                    "index": 15
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 13.0
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        81,
+                        301,
+                        94
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                81,
+                                302,
+                                95
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        81,
+                                        302,
+                                        95
+                                    ],
+                                    "score": 1.0,
+                                    "content": "H IMPACT OF THE MSE LOSS TERM",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        }
+                    ],
+                    "index": 0
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        106,
+                        506,
+                        172
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                106,
+                                505,
+                                118
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        106,
+                                        505,
+                                        118
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In our earlier discussions, we stated that the MSE loss term used in GAIN would bias the distribution",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                118,
+                                505,
+                                129
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        118,
+                                        505,
+                                        129
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of generated samples toward the mean of the distribution. Here, a synthesized dataset is used to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                127,
+                                506,
+                                141
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        127,
+                                        506,
+                                        141
+                                    ],
+                                    "score": 1.0,
+                                    "content": "illustrate the impact of MSE loss term on the distribution of generated samples. A hyperparameter,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                107,
+                                140,
+                                505,
+                                151
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        140,
+                                        114,
+                                        149
+                                    ],
+                                    "score": 0.75,
+                                    "content": "\\lambda",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        114,
+                                        140,
+                                        505,
+                                        151
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", controls the weight of the MSE term in the final objective function. As it can be observed from",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                150,
+                                505,
+                                162
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        150,
+                                        201,
+                                        162
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Figure 8, the higher the",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        201,
+                                        151,
+                                        209,
+                                        160
+                                    ],
+                                    "score": 0.62,
+                                    "content": "\\lambda",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        209,
+                                        150,
+                                        505,
+                                        162
+                                    ],
+                                    "score": 1.0,
+                                    "content": "parameter, the lower the variance of the generated samples (i.e., more bias",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                161,
+                                254,
+                                172
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        161,
+                                        254,
+                                        172
+                                    ],
+                                    "score": 1.0,
+                                    "content": "toward the mean of the distribution).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        }
+                    ],
+                    "index": 3.5
+                },
+                {
+                    "type": "image",
+                    "bbox": [
+                        130,
+                        184,
+                        488,
+                        627
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                130,
+                                184,
+                                488,
+                                627
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        130,
+                                        184,
+                                        488,
+                                        627
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                130,
+                                                184,
+                                                488,
+                                                627
+                                            ],
+                                            "score": 0.975,
+                                            "type": "image",
+                                            "image_path": "e6f566e62f5c08205f49f8240037709b45a8d8f533c27ee98af701aa8fdcd47f.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 8,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        130,
+                                        184,
+                                        488,
+                                        331.66666666666663
+                                    ],
+                                    "spans": [],
+                                    "index": 7
+                                },
+                                {
+                                    "bbox": [
+                                        130,
+                                        331.66666666666663,
+                                        488,
+                                        479.33333333333326
+                                    ],
+                                    "spans": [],
+                                    "index": 8
+                                },
+                                {
+                                    "bbox": [
+                                        130,
+                                        479.33333333333326,
+                                        488,
+                                        626.9999999999999
+                                    ],
+                                    "spans": [],
+                                    "index": 9
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                106,
+                                634,
+                                506,
+                                668
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        634,
+                                        505,
+                                        647
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                634,
+                                                505,
+                                                647
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 8: Comparison of generating samples from a Gaussian distribution (a) samples from the",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 10
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        646,
+                                        505,
+                                        658
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                646,
+                                                505,
+                                                658
+                                            ],
+                                            "score": 1.0,
+                                            "content": "original distribution, (b) samples generated using GAIN imputers with different significance of the",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 11
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        655,
+                                        223,
+                                        669
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                655,
+                                                209,
+                                                669
+                                            ],
+                                            "score": 1.0,
+                                            "content": "MSE term (controlled by",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                209,
+                                                657,
+                                                216,
+                                                667
+                                            ],
+                                            "score": 0.65,
+                                            "content": "\\lambda",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                216,
+                                                655,
+                                                223,
+                                                669
+                                            ],
+                                            "score": 1.0,
+                                            "content": ").",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 12
+                                }
+                            ],
+                            "index": 11
+                        }
+                    ],
+                    "index": 9.5
+                }
+            ],
+            "page_idx": 21,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        107,
+                        27,
+                        308,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                107,
+                                26,
+                                308,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        26,
+                                        308,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Under review as a conference paper at ICLR 2020",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        300,
+                        751,
+                        311,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                298,
+                                750,
+                                313,
+                                765
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        298,
+                                        750,
+                                        313,
+                                        765
+                                    ],
+                                    "score": 1.0,
+                                    "content": "",
+                                    "type": "text",
+                                    "height": 15,
+                                    "width": 15
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        81,
+                        301,
+                        94
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                81,
+                                302,
+                                95
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        81,
+                                        302,
+                                        95
+                                    ],
+                                    "score": 1.0,
+                                    "content": "H IMPACT OF THE MSE LOSS TERM",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        }
+                    ],
+                    "index": 0
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        106,
+                        506,
+                        172
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                106,
+                                505,
+                                118
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        106,
+                                        505,
+                                        118
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In our earlier discussions, we stated that the MSE loss term used in GAIN would bias the distribution",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                118,
+                                505,
+                                129
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        118,
+                                        505,
+                                        129
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of generated samples toward the mean of the distribution. Here, a synthesized dataset is used to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                127,
+                                506,
+                                141
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        127,
+                                        506,
+                                        141
+                                    ],
+                                    "score": 1.0,
+                                    "content": "illustrate the impact of MSE loss term on the distribution of generated samples. A hyperparameter,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                107,
+                                140,
+                                505,
+                                151
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        140,
+                                        114,
+                                        149
+                                    ],
+                                    "score": 0.75,
+                                    "content": "\\lambda",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        114,
+                                        140,
+                                        505,
+                                        151
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", controls the weight of the MSE term in the final objective function. As it can be observed from",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                150,
+                                505,
+                                162
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        150,
+                                        201,
+                                        162
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Figure 8, the higher the",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        201,
+                                        151,
+                                        209,
+                                        160
+                                    ],
+                                    "score": 0.62,
+                                    "content": "\\lambda",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        209,
+                                        150,
+                                        505,
+                                        162
+                                    ],
+                                    "score": 1.0,
+                                    "content": "parameter, the lower the variance of the generated samples (i.e., more bias",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                161,
+                                254,
+                                172
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        161,
+                                        254,
+                                        172
+                                    ],
+                                    "score": 1.0,
+                                    "content": "toward the mean of the distribution).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        }
+                    ],
+                    "index": 3.5,
+                    "bbox_fs": [
+                        105,
+                        106,
+                        506,
+                        172
+                    ]
+                },
+                {
+                    "type": "image",
+                    "bbox": [
+                        130,
+                        184,
+                        488,
+                        627
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                130,
+                                184,
+                                488,
+                                627
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        130,
+                                        184,
+                                        488,
+                                        627
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                130,
+                                                184,
+                                                488,
+                                                627
+                                            ],
+                                            "score": 0.975,
+                                            "type": "image",
+                                            "image_path": "e6f566e62f5c08205f49f8240037709b45a8d8f533c27ee98af701aa8fdcd47f.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 8,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        130,
+                                        184,
+                                        488,
+                                        331.66666666666663
+                                    ],
+                                    "spans": [],
+                                    "index": 7
+                                },
+                                {
+                                    "bbox": [
+                                        130,
+                                        331.66666666666663,
+                                        488,
+                                        479.33333333333326
+                                    ],
+                                    "spans": [],
+                                    "index": 8
+                                },
+                                {
+                                    "bbox": [
+                                        130,
+                                        479.33333333333326,
+                                        488,
+                                        626.9999999999999
+                                    ],
+                                    "spans": [],
+                                    "index": 9
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                106,
+                                634,
+                                506,
+                                668
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        634,
+                                        505,
+                                        647
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                634,
+                                                505,
+                                                647
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 8: Comparison of generating samples from a Gaussian distribution (a) samples from the",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 10
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        646,
+                                        505,
+                                        658
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                646,
+                                                505,
+                                                658
+                                            ],
+                                            "score": 1.0,
+                                            "content": "original distribution, (b) samples generated using GAIN imputers with different significance of the",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 11
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        655,
+                                        223,
+                                        669
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                655,
+                                                209,
+                                                669
+                                            ],
+                                            "score": 1.0,
+                                            "content": "MSE term (controlled by",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                209,
+                                                657,
+                                                216,
+                                                667
+                                            ],
+                                            "score": 0.65,
+                                            "content": "\\lambda",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                216,
+                                                655,
+                                                223,
+                                                669
+                                            ],
+                                            "score": 1.0,
+                                            "content": ").",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 12
+                                }
+                            ],
+                            "index": 11
+                        }
+                    ],
+                    "index": 9.5
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "title",
+                    "bbox": [
+                        106,
+                        81,
+                        366,
+                        94
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                79,
+                                368,
+                                96
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        79,
+                                        368,
+                                        96
+                                    ],
+                                    "score": 1.0,
+                                    "content": "I IMPACT OF THE DISCRIMINATOR HINT VECTOR",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        }
+                    ],
+                    "index": 0
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        106,
+                        506,
+                        216
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                105,
+                                507,
+                                119
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        105,
+                                        507,
+                                        119
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Yoon et al. (2018) suggested the idea of guiding the discriminator network using a hint mechanism.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                116,
+                                505,
+                                129
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        116,
+                                        505,
+                                        129
+                                    ],
+                                    "score": 1.0,
+                                    "content": "A hint vector reveals a subset of features that are missing to the discriminator. In Figure 9 and 10 we",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                128,
+                                506,
+                                141
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        128,
+                                        506,
+                                        141
+                                    ],
+                                    "score": 1.0,
+                                    "content": "provide a comparison of learning curves for GI implemented using different hint rates. From Figure 9,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                139,
+                                506,
+                                151
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        139,
+                                        506,
+                                        151
+                                    ],
+                                    "score": 1.0,
+                                    "content": "using the hint mechanism does not result in any noticeable improvement in the final imputation",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                150,
+                                506,
+                                163
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        150,
+                                        506,
+                                        163
+                                    ],
+                                    "score": 1.0,
+                                    "content": "quality justifying the added complexity. For the case of the rectangular missing structure in Figure 10;",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                161,
+                                507,
+                                174
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        161,
+                                        507,
+                                        174
+                                    ],
+                                    "score": 1.0,
+                                    "content": "however, using the hint vector causes instabilities in the training process. One possible explanation is:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                172,
+                                506,
+                                184
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        172,
+                                        506,
+                                        184
+                                    ],
+                                    "score": 1.0,
+                                    "content": "providing even a small portion of the mask as a hint, due to the deterministic nature of the rectangular",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                183,
+                                506,
+                                196
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        183,
+                                        506,
+                                        196
+                                    ],
+                                    "score": 1.0,
+                                    "content": "shape it is equivalent to providing region boundaries to the discriminator making it obvious for the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                192,
+                                506,
+                                207
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        192,
+                                        506,
+                                        207
+                                    ],
+                                    "score": 1.0,
+                                    "content": "discriminator. In GAN training we generally want to have equal competition between the generator",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                204,
+                                181,
+                                217
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        204,
+                                        181,
+                                        217
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and discriminator.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        }
+                    ],
+                    "index": 5.5
+                },
+                {
+                    "type": "image",
+                    "bbox": [
+                        172,
+                        231,
+                        437,
+                        403
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                172,
+                                231,
+                                437,
+                                403
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        172,
+                                        231,
+                                        437,
+                                        403
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                172,
+                                                231,
+                                                437,
+                                                403
+                                            ],
+                                            "score": 0.968,
+                                            "type": "image",
+                                            "image_path": "d1773a0973e8e20e8a1e95585cf0570e2f70c7ae32a782150bd1926ef6c94fdd.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 12,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        172,
+                                        231,
+                                        437,
+                                        288.3333333333333
+                                    ],
+                                    "spans": [],
+                                    "index": 11
+                                },
+                                {
+                                    "bbox": [
+                                        172,
+                                        288.3333333333333,
+                                        437,
+                                        345.66666666666663
+                                    ],
+                                    "spans": [],
+                                    "index": 12
+                                },
+                                {
+                                    "bbox": [
+                                        172,
+                                        345.66666666666663,
+                                        437,
+                                        402.99999999999994
+                                    ],
+                                    "spans": [],
+                                    "index": 13
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                104,
+                                419,
+                                505,
+                                442
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        419,
+                                        506,
+                                        432
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                419,
+                                                506,
+                                                432
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 9: Learning curves for CIFAR-10 with uniform missing structure at different discriminator",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 14
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        429,
+                                        149,
+                                        443
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                429,
+                                                149,
+                                                443
+                                            ],
+                                            "score": 1.0,
+                                            "content": "hint rates.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 15
+                                }
+                            ],
+                            "index": 14.5
+                        }
+                    ],
+                    "index": 13.25
+                },
+                {
+                    "type": "image",
+                    "bbox": [
+                        172,
+                        451,
+                        437,
+                        623
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                172,
+                                451,
+                                437,
+                                623
+                            ],
+                            "group_id": 1,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        172,
+                                        451,
+                                        437,
+                                        623
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                172,
+                                                451,
+                                                437,
+                                                623
+                                            ],
+                                            "score": 0.968,
+                                            "type": "image",
+                                            "image_path": "1d860d6685302d772bd786609a2406e9a1cc659ad5aaf397983d57ed6fba2242.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 17,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        172,
+                                        451,
+                                        437,
+                                        508.3333333333333
+                                    ],
+                                    "spans": [],
+                                    "index": 16
+                                },
+                                {
+                                    "bbox": [
+                                        172,
+                                        508.3333333333333,
+                                        437,
+                                        565.6666666666666
+                                    ],
+                                    "spans": [],
+                                    "index": 17
+                                },
+                                {
+                                    "bbox": [
+                                        172,
+                                        565.6666666666666,
+                                        437,
+                                        623.0
+                                    ],
+                                    "spans": [],
+                                    "index": 18
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                106,
+                                639,
+                                504,
+                                662
+                            ],
+                            "group_id": 1,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        639,
+                                        505,
+                                        652
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                639,
+                                                505,
+                                                652
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 10: Learning curves for CIFAR-10 with rectangular missing structure at different discriminator",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 19
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        650,
+                                        149,
+                                        663
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                650,
+                                                149,
+                                                663
+                                            ],
+                                            "score": 1.0,
+                                            "content": "hint rates.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 20
+                                }
+                            ],
+                            "index": 19.5
+                        }
+                    ],
+                    "index": 18.25
+                }
+            ],
+            "page_idx": 22,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        107,
+                        27,
+                        308,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                107,
+                                26,
+                                308,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        26,
+                                        308,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Under review as a conference paper at ICLR 2020",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        300,
+                        751,
+                        311,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                298,
+                                750,
+                                312,
+                                763
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        298,
+                                        750,
+                                        312,
+                                        763
+                                    ],
+                                    "score": 1.0,
+                                    "content": "23",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "title",
+                    "bbox": [
+                        106,
+                        81,
+                        366,
+                        94
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                79,
+                                368,
+                                96
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        79,
+                                        368,
+                                        96
+                                    ],
+                                    "score": 1.0,
+                                    "content": "I IMPACT OF THE DISCRIMINATOR HINT VECTOR",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        }
+                    ],
+                    "index": 0
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        106,
+                        506,
+                        216
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                105,
+                                507,
+                                119
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        105,
+                                        507,
+                                        119
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Yoon et al. (2018) suggested the idea of guiding the discriminator network using a hint mechanism.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                116,
+                                505,
+                                129
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        116,
+                                        505,
+                                        129
+                                    ],
+                                    "score": 1.0,
+                                    "content": "A hint vector reveals a subset of features that are missing to the discriminator. In Figure 9 and 10 we",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                128,
+                                506,
+                                141
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        128,
+                                        506,
+                                        141
+                                    ],
+                                    "score": 1.0,
+                                    "content": "provide a comparison of learning curves for GI implemented using different hint rates. From Figure 9,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                139,
+                                506,
+                                151
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        139,
+                                        506,
+                                        151
+                                    ],
+                                    "score": 1.0,
+                                    "content": "using the hint mechanism does not result in any noticeable improvement in the final imputation",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                150,
+                                506,
+                                163
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        150,
+                                        506,
+                                        163
+                                    ],
+                                    "score": 1.0,
+                                    "content": "quality justifying the added complexity. For the case of the rectangular missing structure in Figure 10;",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                161,
+                                507,
+                                174
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        161,
+                                        507,
+                                        174
+                                    ],
+                                    "score": 1.0,
+                                    "content": "however, using the hint vector causes instabilities in the training process. One possible explanation is:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                172,
+                                506,
+                                184
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        172,
+                                        506,
+                                        184
+                                    ],
+                                    "score": 1.0,
+                                    "content": "providing even a small portion of the mask as a hint, due to the deterministic nature of the rectangular",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                183,
+                                506,
+                                196
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        183,
+                                        506,
+                                        196
+                                    ],
+                                    "score": 1.0,
+                                    "content": "shape it is equivalent to providing region boundaries to the discriminator making it obvious for the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                192,
+                                506,
+                                207
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        192,
+                                        506,
+                                        207
+                                    ],
+                                    "score": 1.0,
+                                    "content": "discriminator. In GAN training we generally want to have equal competition between the generator",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                204,
+                                181,
+                                217
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        204,
+                                        181,
+                                        217
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and discriminator.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        }
+                    ],
+                    "index": 5.5,
+                    "bbox_fs": [
+                        105,
+                        105,
+                        507,
+                        217
+                    ]
+                },
+                {
+                    "type": "image",
+                    "bbox": [
+                        172,
+                        231,
+                        437,
+                        403
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                172,
+                                231,
+                                437,
+                                403
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        172,
+                                        231,
+                                        437,
+                                        403
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                172,
+                                                231,
+                                                437,
+                                                403
+                                            ],
+                                            "score": 0.968,
+                                            "type": "image",
+                                            "image_path": "d1773a0973e8e20e8a1e95585cf0570e2f70c7ae32a782150bd1926ef6c94fdd.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 12,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        172,
+                                        231,
+                                        437,
+                                        288.3333333333333
+                                    ],
+                                    "spans": [],
+                                    "index": 11
+                                },
+                                {
+                                    "bbox": [
+                                        172,
+                                        288.3333333333333,
+                                        437,
+                                        345.66666666666663
+                                    ],
+                                    "spans": [],
+                                    "index": 12
+                                },
+                                {
+                                    "bbox": [
+                                        172,
+                                        345.66666666666663,
+                                        437,
+                                        402.99999999999994
+                                    ],
+                                    "spans": [],
+                                    "index": 13
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                104,
+                                419,
+                                505,
+                                442
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        419,
+                                        506,
+                                        432
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                419,
+                                                506,
+                                                432
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 9: Learning curves for CIFAR-10 with uniform missing structure at different discriminator",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 14
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        429,
+                                        149,
+                                        443
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                429,
+                                                149,
+                                                443
+                                            ],
+                                            "score": 1.0,
+                                            "content": "hint rates.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 15
+                                }
+                            ],
+                            "index": 14.5
+                        }
+                    ],
+                    "index": 13.25
+                },
+                {
+                    "type": "image",
+                    "bbox": [
+                        172,
+                        451,
+                        437,
+                        623
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                172,
+                                451,
+                                437,
+                                623
+                            ],
+                            "group_id": 1,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        172,
+                                        451,
+                                        437,
+                                        623
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                172,
+                                                451,
+                                                437,
+                                                623
+                                            ],
+                                            "score": 0.968,
+                                            "type": "image",
+                                            "image_path": "1d860d6685302d772bd786609a2406e9a1cc659ad5aaf397983d57ed6fba2242.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 17,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        172,
+                                        451,
+                                        437,
+                                        508.3333333333333
+                                    ],
+                                    "spans": [],
+                                    "index": 16
+                                },
+                                {
+                                    "bbox": [
+                                        172,
+                                        508.3333333333333,
+                                        437,
+                                        565.6666666666666
+                                    ],
+                                    "spans": [],
+                                    "index": 17
+                                },
+                                {
+                                    "bbox": [
+                                        172,
+                                        565.6666666666666,
+                                        437,
+                                        623.0
+                                    ],
+                                    "spans": [],
+                                    "index": 18
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                106,
+                                639,
+                                504,
+                                662
+                            ],
+                            "group_id": 1,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        639,
+                                        505,
+                                        652
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                639,
+                                                505,
+                                                652
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 10: Learning curves for CIFAR-10 with rectangular missing structure at different discriminator",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 19
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        650,
+                                        149,
+                                        663
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                650,
+                                                149,
+                                                663
+                                            ],
+                                            "score": 1.0,
+                                            "content": "hint rates.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 20
+                                }
+                            ],
+                            "index": 19.5
+                        }
+                    ],
+                    "index": 18.25
+                }
+            ]
+        }
+    ],
+    "_backend": "pipeline",
+    "_version_name": "2.2.2"
+}
\ No newline at end of file
diff --git a/parse/train/B1em9h4KDS/B1em9h4KDS_model.json b/parse/train/B1em9h4KDS/B1em9h4KDS_model.json
new file mode 100644
index 0000000000000000000000000000000000000000..0a86e260b766e0b0e69390e95d3e0d4d693d7641
--- /dev/null
+++ b/parse/train/B1em9h4KDS/B1em9h4KDS_model.json
@@ -0,0 +1,24721 @@
+[
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    398,
+                    536,
+                    1304,
+                    536,
+                    1304,
+                    1025,
+                    398,
+                    1025
+                ],
+                "score": 0.984
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1590,
+                    1405,
+                    1590,
+                    1405,
+                    1835,
+                    298,
+                    1835
+                ],
+                "score": 0.98
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1360,
+                    1405,
+                    1360,
+                    1405,
+                    1574,
+                    298,
+                    1574
+                ],
+                "score": 0.979
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    1851,
+                    1404,
+                    1851,
+                    1404,
+                    2034,
+                    299,
+                    2034
+                ],
+                "score": 0.979
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1161,
+                    1404,
+                    1161,
+                    1404,
+                    1345,
+                    298,
+                    1345
+                ],
+                "score": 0.978
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    298,
+                    219,
+                    1480,
+                    219,
+                    1480,
+                    270,
+                    298,
+                    270
+                ],
+                "score": 0.905
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    298,
+                    75,
+                    858,
+                    75,
+                    858,
+                    105,
+                    298,
+                    105
+                ],
+                "score": 0.899
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    302,
+                    1089,
+                    573,
+                    1089,
+                    573,
+                    1124,
+                    302,
+                    1124
+                ],
+                "score": 0.893
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    773,
+                    464,
+                    927,
+                    464,
+                    927,
+                    497,
+                    773,
+                    497
+                ],
+                "score": 0.878
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    313,
+                    322,
+                    681,
+                    322,
+                    681,
+                    382,
+                    313,
+                    382
+                ],
+                "score": 0.877
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    841,
+                    2088,
+                    856,
+                    2088,
+                    856,
+                    2112,
+                    841,
+                    2112
+                ],
+                "score": 0.721
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    217.0,
+                    1491.0,
+                    217.0,
+                    1491.0,
+                    276.0,
+                    293.0,
+                    276.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    298.0,
+                    73.0,
+                    858.0,
+                    73.0,
+                    858.0,
+                    108.0,
+                    298.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1086.0,
+                    579.0,
+                    1086.0,
+                    579.0,
+                    1133.0,
+                    294.0,
+                    1133.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    768.0,
+                    460.0,
+                    934.0,
+                    460.0,
+                    934.0,
+                    503.0,
+                    768.0,
+                    503.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    840.0,
+                    2088.0,
+                    857.0,
+                    2088.0,
+                    857.0,
+                    2116.0,
+                    840.0,
+                    2116.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    392.0,
+                    536.0,
+                    1305.0,
+                    536.0,
+                    1305.0,
+                    571.0,
+                    392.0,
+                    571.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    394.0,
+                    567.0,
+                    1304.0,
+                    567.0,
+                    1304.0,
+                    602.0,
+                    394.0,
+                    602.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    393.0,
+                    595.0,
+                    1305.0,
+                    595.0,
+                    1305.0,
+                    633.0,
+                    393.0,
+                    633.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    394.0,
+                    627.0,
+                    1306.0,
+                    627.0,
+                    1306.0,
+                    661.0,
+                    394.0,
+                    661.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    393.0,
+                    658.0,
+                    1308.0,
+                    658.0,
+                    1308.0,
+                    694.0,
+                    393.0,
+                    694.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    393.0,
+                    689.0,
+                    1306.0,
+                    689.0,
+                    1306.0,
+                    724.0,
+                    393.0,
+                    724.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    392.0,
+                    717.0,
+                    1306.0,
+                    717.0,
+                    1306.0,
+                    754.0,
+                    392.0,
+                    754.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    394.0,
+                    752.0,
+                    1308.0,
+                    752.0,
+                    1308.0,
+                    784.0,
+                    394.0,
+                    784.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    393.0,
+                    779.0,
+                    1308.0,
+                    779.0,
+                    1308.0,
+                    815.0,
+                    393.0,
+                    815.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    393.0,
+                    811.0,
+                    1310.0,
+                    811.0,
+                    1310.0,
+                    846.0,
+                    393.0,
+                    846.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    393.0,
+                    841.0,
+                    1308.0,
+                    841.0,
+                    1308.0,
+                    875.0,
+                    393.0,
+                    875.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    394.0,
+                    870.0,
+                    1306.0,
+                    870.0,
+                    1306.0,
+                    905.0,
+                    394.0,
+                    905.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    393.0,
+                    901.0,
+                    1305.0,
+                    901.0,
+                    1305.0,
+                    935.0,
+                    393.0,
+                    935.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    395.0,
+                    933.0,
+                    1305.0,
+                    933.0,
+                    1305.0,
+                    965.0,
+                    395.0,
+                    965.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    394.0,
+                    964.0,
+                    1304.0,
+                    964.0,
+                    1304.0,
+                    995.0,
+                    394.0,
+                    995.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    392.0,
+                    990.0,
+                    1015.0,
+                    990.0,
+                    1015.0,
+                    1028.0,
+                    392.0,
+                    1028.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1588.0,
+                    1406.0,
+                    1588.0,
+                    1406.0,
+                    1627.0,
+                    292.0,
+                    1627.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1621.0,
+                    1406.0,
+                    1621.0,
+                    1406.0,
+                    1655.0,
+                    295.0,
+                    1655.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1654.0,
+                    1405.0,
+                    1654.0,
+                    1405.0,
+                    1684.0,
+                    295.0,
+                    1684.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1679.0,
+                    1409.0,
+                    1679.0,
+                    1409.0,
+                    1719.0,
+                    295.0,
+                    1719.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1714.0,
+                    1403.0,
+                    1714.0,
+                    1403.0,
+                    1747.0,
+                    295.0,
+                    1747.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1742.0,
+                    1405.0,
+                    1742.0,
+                    1405.0,
+                    1776.0,
+                    295.0,
+                    1776.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1772.0,
+                    1406.0,
+                    1772.0,
+                    1406.0,
+                    1808.0,
+                    293.0,
+                    1808.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1805.0,
+                    1133.0,
+                    1805.0,
+                    1133.0,
+                    1838.0,
+                    296.0,
+                    1838.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1360.0,
+                    1408.0,
+                    1360.0,
+                    1408.0,
+                    1395.0,
+                    295.0,
+                    1395.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1392.0,
+                    1407.0,
+                    1392.0,
+                    1407.0,
+                    1427.0,
+                    296.0,
+                    1427.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1420.0,
+                    1408.0,
+                    1420.0,
+                    1408.0,
+                    1456.0,
+                    293.0,
+                    1456.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1451.0,
+                    1406.0,
+                    1451.0,
+                    1406.0,
+                    1487.0,
+                    293.0,
+                    1487.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1480.0,
+                    1405.0,
+                    1480.0,
+                    1405.0,
+                    1517.0,
+                    293.0,
+                    1517.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1512.0,
+                    1406.0,
+                    1512.0,
+                    1406.0,
+                    1552.0,
+                    291.0,
+                    1552.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1546.0,
+                    637.0,
+                    1546.0,
+                    637.0,
+                    1576.0,
+                    296.0,
+                    1576.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1850.0,
+                    1405.0,
+                    1850.0,
+                    1405.0,
+                    1886.0,
+                    295.0,
+                    1886.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1882.0,
+                    1406.0,
+                    1882.0,
+                    1406.0,
+                    1917.0,
+                    295.0,
+                    1917.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1910.0,
+                    1405.0,
+                    1910.0,
+                    1405.0,
+                    1946.0,
+                    294.0,
+                    1946.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1943.0,
+                    1406.0,
+                    1943.0,
+                    1406.0,
+                    1978.0,
+                    295.0,
+                    1978.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1970.0,
+                    1408.0,
+                    1970.0,
+                    1408.0,
+                    2009.0,
+                    293.0,
+                    2009.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1999.0,
+                    1404.0,
+                    1999.0,
+                    1404.0,
+                    2039.0,
+                    294.0,
+                    2039.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1159.0,
+                    1406.0,
+                    1159.0,
+                    1406.0,
+                    1197.0,
+                    293.0,
+                    1197.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1193.0,
+                    1402.0,
+                    1193.0,
+                    1402.0,
+                    1225.0,
+                    296.0,
+                    1225.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1221.0,
+                    1405.0,
+                    1221.0,
+                    1405.0,
+                    1258.0,
+                    293.0,
+                    1258.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1252.0,
+                    1407.0,
+                    1252.0,
+                    1407.0,
+                    1287.0,
+                    294.0,
+                    1287.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1283.0,
+                    1405.0,
+                    1283.0,
+                    1405.0,
+                    1319.0,
+                    294.0,
+                    1319.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1314.0,
+                    1126.0,
+                    1314.0,
+                    1126.0,
+                    1349.0,
+                    296.0,
+                    1349.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    315.0,
+                    323.0,
+                    559.0,
+                    323.0,
+                    559.0,
+                    355.0,
+                    315.0,
+                    355.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    311.0,
+                    351.0,
+                    681.0,
+                    351.0,
+                    681.0,
+                    384.0,
+                    311.0,
+                    384.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 0,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    889,
+                    1404,
+                    889,
+                    1404,
+                    1255,
+                    298,
+                    1255
+                ],
+                "score": 0.983
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1272,
+                    1405,
+                    1272,
+                    1405,
+                    1637,
+                    298,
+                    1637
+                ],
+                "score": 0.982
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    307,
+                    1404,
+                    307,
+                    1404,
+                    552,
+                    298,
+                    552
+                ],
+                "score": 0.979
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    689,
+                    1403,
+                    689,
+                    1403,
+                    873,
+                    298,
+                    873
+                ],
+                "score": 0.979
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1652,
+                    1404,
+                    1652,
+                    1404,
+                    1928,
+                    298,
+                    1928
+                ],
+                "score": 0.977
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    229,
+                    1399,
+                    229,
+                    1399,
+                    292,
+                    297,
+                    292
+                ],
+                "score": 0.951
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    302,
+                    1977,
+                    1404,
+                    1977,
+                    1404,
+                    2033,
+                    302,
+                    2033
+                ],
+                "score": 0.926
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    301,
+                    610,
+                    586,
+                    610,
+                    586,
+                    646,
+                    301,
+                    646
+                ],
+                "score": 0.903
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    298,
+                    76,
+                    858,
+                    76,
+                    858,
+                    104,
+                    298,
+                    104
+                ],
+                "score": 0.884
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    841,
+                    2088,
+                    859,
+                    2088,
+                    859,
+                    2112,
+                    841,
+                    2112
+                ],
+                "score": 0.744
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    712,
+                    753,
+                    731,
+                    753,
+                    731,
+                    780,
+                    712,
+                    780
+                ],
+                "score": 0.37,
+                "latex": "\\mathbf { k }"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    330.0,
+                    1970.0,
+                    1404.0,
+                    1970.0,
+                    1404.0,
+                    2012.0,
+                    330.0,
+                    2012.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    2001.0,
+                    385.0,
+                    2001.0,
+                    385.0,
+                    2037.0,
+                    294.0,
+                    2037.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    607.0,
+                    593.0,
+                    607.0,
+                    593.0,
+                    654.0,
+                    291.0,
+                    654.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    70.0,
+                    859.0,
+                    70.0,
+                    859.0,
+                    109.0,
+                    296.0,
+                    109.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    838.0,
+                    2085.0,
+                    862.0,
+                    2085.0,
+                    862.0,
+                    2120.0,
+                    838.0,
+                    2120.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    890.0,
+                    1404.0,
+                    890.0,
+                    1404.0,
+                    924.0,
+                    294.0,
+                    924.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    919.0,
+                    1402.0,
+                    919.0,
+                    1402.0,
+                    953.0,
+                    294.0,
+                    953.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    949.0,
+                    1408.0,
+                    949.0,
+                    1408.0,
+                    987.0,
+                    293.0,
+                    987.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    980.0,
+                    1404.0,
+                    980.0,
+                    1404.0,
+                    1016.0,
+                    293.0,
+                    1016.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1009.0,
+                    1406.0,
+                    1009.0,
+                    1406.0,
+                    1048.0,
+                    293.0,
+                    1048.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1042.0,
+                    1405.0,
+                    1042.0,
+                    1405.0,
+                    1076.0,
+                    294.0,
+                    1076.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1072.0,
+                    1406.0,
+                    1072.0,
+                    1406.0,
+                    1107.0,
+                    293.0,
+                    1107.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1102.0,
+                    1406.0,
+                    1102.0,
+                    1406.0,
+                    1136.0,
+                    294.0,
+                    1136.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1134.0,
+                    1406.0,
+                    1134.0,
+                    1406.0,
+                    1167.0,
+                    294.0,
+                    1167.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1163.0,
+                    1404.0,
+                    1163.0,
+                    1404.0,
+                    1197.0,
+                    294.0,
+                    1197.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1191.0,
+                    1406.0,
+                    1191.0,
+                    1406.0,
+                    1231.0,
+                    292.0,
+                    1231.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1224.0,
+                    970.0,
+                    1224.0,
+                    970.0,
+                    1258.0,
+                    294.0,
+                    1258.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1272.0,
+                    1405.0,
+                    1272.0,
+                    1405.0,
+                    1306.0,
+                    295.0,
+                    1306.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1302.0,
+                    1403.0,
+                    1302.0,
+                    1403.0,
+                    1336.0,
+                    293.0,
+                    1336.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1333.0,
+                    1405.0,
+                    1333.0,
+                    1405.0,
+                    1367.0,
+                    295.0,
+                    1367.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1361.0,
+                    1406.0,
+                    1361.0,
+                    1406.0,
+                    1399.0,
+                    293.0,
+                    1399.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1390.0,
+                    1409.0,
+                    1390.0,
+                    1409.0,
+                    1430.0,
+                    292.0,
+                    1430.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1422.0,
+                    1407.0,
+                    1422.0,
+                    1407.0,
+                    1461.0,
+                    292.0,
+                    1461.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1456.0,
+                    1405.0,
+                    1456.0,
+                    1405.0,
+                    1490.0,
+                    293.0,
+                    1490.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1485.0,
+                    1406.0,
+                    1485.0,
+                    1406.0,
+                    1519.0,
+                    295.0,
+                    1519.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1515.0,
+                    1403.0,
+                    1515.0,
+                    1403.0,
+                    1549.0,
+                    295.0,
+                    1549.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1545.0,
+                    1406.0,
+                    1545.0,
+                    1406.0,
+                    1579.0,
+                    292.0,
+                    1579.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1573.0,
+                    1406.0,
+                    1573.0,
+                    1406.0,
+                    1612.0,
+                    292.0,
+                    1612.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1605.0,
+                    443.0,
+                    1605.0,
+                    443.0,
+                    1640.0,
+                    292.0,
+                    1640.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    307.0,
+                    1408.0,
+                    307.0,
+                    1408.0,
+                    344.0,
+                    293.0,
+                    344.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    338.0,
+                    1407.0,
+                    338.0,
+                    1407.0,
+                    372.0,
+                    293.0,
+                    372.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    368.0,
+                    1405.0,
+                    368.0,
+                    1405.0,
+                    402.0,
+                    293.0,
+                    402.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    398.0,
+                    1406.0,
+                    398.0,
+                    1406.0,
+                    434.0,
+                    295.0,
+                    434.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    431.0,
+                    1405.0,
+                    431.0,
+                    1405.0,
+                    464.0,
+                    294.0,
+                    464.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    459.0,
+                    1402.0,
+                    459.0,
+                    1402.0,
+                    493.0,
+                    294.0,
+                    493.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    492.0,
+                    1405.0,
+                    492.0,
+                    1405.0,
+                    525.0,
+                    294.0,
+                    525.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    524.0,
+                    658.0,
+                    524.0,
+                    658.0,
+                    554.0,
+                    296.0,
+                    554.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    688.0,
+                    1407.0,
+                    688.0,
+                    1407.0,
+                    725.0,
+                    294.0,
+                    725.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    719.0,
+                    1405.0,
+                    719.0,
+                    1405.0,
+                    756.0,
+                    293.0,
+                    756.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    751.0,
+                    711.0,
+                    751.0,
+                    711.0,
+                    786.0,
+                    294.0,
+                    786.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    732.0,
+                    751.0,
+                    1407.0,
+                    751.0,
+                    1407.0,
+                    786.0,
+                    732.0,
+                    786.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    781.0,
+                    1404.0,
+                    781.0,
+                    1404.0,
+                    817.0,
+                    294.0,
+                    817.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    810.0,
+                    1408.0,
+                    810.0,
+                    1408.0,
+                    847.0,
+                    292.0,
+                    847.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    843.0,
+                    1296.0,
+                    843.0,
+                    1296.0,
+                    877.0,
+                    294.0,
+                    877.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1652.0,
+                    1406.0,
+                    1652.0,
+                    1406.0,
+                    1689.0,
+                    293.0,
+                    1689.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1683.0,
+                    1405.0,
+                    1683.0,
+                    1405.0,
+                    1720.0,
+                    294.0,
+                    1720.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1713.0,
+                    1406.0,
+                    1713.0,
+                    1406.0,
+                    1750.0,
+                    293.0,
+                    1750.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1744.0,
+                    1406.0,
+                    1744.0,
+                    1406.0,
+                    1781.0,
+                    293.0,
+                    1781.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1773.0,
+                    1406.0,
+                    1773.0,
+                    1406.0,
+                    1811.0,
+                    293.0,
+                    1811.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1804.0,
+                    1405.0,
+                    1804.0,
+                    1405.0,
+                    1841.0,
+                    294.0,
+                    1841.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1834.0,
+                    1407.0,
+                    1834.0,
+                    1407.0,
+                    1872.0,
+                    292.0,
+                    1872.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1866.0,
+                    1408.0,
+                    1866.0,
+                    1408.0,
+                    1902.0,
+                    294.0,
+                    1902.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1897.0,
+                    1407.0,
+                    1897.0,
+                    1407.0,
+                    1930.0,
+                    293.0,
+                    1930.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    227.0,
+                    1403.0,
+                    227.0,
+                    1403.0,
+                    266.0,
+                    293.0,
+                    266.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    263.0,
+                    760.0,
+                    263.0,
+                    760.0,
+                    295.0,
+                    297.0,
+                    295.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 1,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    650,
+                    1405,
+                    650,
+                    1405,
+                    1078,
+                    298,
+                    1078
+                ],
+                "score": 0.983
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1490,
+                    1404,
+                    1490,
+                    1404,
+                    1705,
+                    298,
+                    1705
+                ],
+                "score": 0.98
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1253,
+                    1406,
+                    1253,
+                    1406,
+                    1476,
+                    297,
+                    1476
+                ],
+                "score": 0.98
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    1803,
+                    1404,
+                    1803,
+                    1404,
+                    1927,
+                    299,
+                    1927
+                ],
+                "score": 0.975
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    300,
+                    1940,
+                    1401,
+                    1940,
+                    1401,
+                    2035,
+                    300,
+                    2035
+                ],
+                "score": 0.971
+            },
+            {
+                "category_id": 3,
+                "poly": [
+                    359,
+                    230,
+                    1345,
+                    230,
+                    1345,
+                    411,
+                    359,
+                    411
+                ],
+                "score": 0.953
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    300,
+                    573,
+                    1397,
+                    573,
+                    1397,
+                    635,
+                    300,
+                    635
+                ],
+                "score": 0.946
+            },
+            {
+                "category_id": 4,
+                "poly": [
+                    295,
+                    448,
+                    1402,
+                    448,
+                    1402,
+                    512,
+                    295,
+                    512
+                ],
+                "score": 0.923
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    300,
+                    1196,
+                    632,
+                    1196,
+                    632,
+                    1228,
+                    300,
+                    1228
+                ],
+                "score": 0.915
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    301,
+                    1124,
+                    641,
+                    1124,
+                    641,
+                    1160,
+                    301,
+                    1160
+                ],
+                "score": 0.903
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    300,
+                    1745,
+                    678,
+                    1745,
+                    678,
+                    1776,
+                    300,
+                    1776
+                ],
+                "score": 0.898
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    297,
+                    75,
+                    858,
+                    75,
+                    858,
+                    105,
+                    297,
+                    105
+                ],
+                "score": 0.825
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    841,
+                    2088,
+                    858,
+                    2088,
+                    858,
+                    2112,
+                    841,
+                    2112
+                ],
+                "score": 0.608
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    841,
+                    2088,
+                    859,
+                    2088,
+                    859,
+                    2112,
+                    841,
+                    2112
+                ],
+                "score": 0.408
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    297,
+                    75,
+                    858,
+                    75,
+                    858,
+                    104,
+                    297,
+                    104
+                ],
+                "score": 0.102
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1052,
+                    1317,
+                    1195,
+                    1317,
+                    1195,
+                    1356,
+                    1052,
+                    1356
+                ],
+                "score": 0.94,
+                "latex": "\\pmb { k } _ { i } \\in \\{ 0 , 1 \\} ^ { d }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    992,
+                    1285,
+                    1115,
+                    1285,
+                    1115,
+                    1319,
+                    992,
+                    1319
+                ],
+                "score": 0.93,
+                "latex": "( { \\pmb x } _ { i } , { \\pmb k } _ { i } , y _ { i } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    688,
+                    1939,
+                    895,
+                    1939,
+                    895,
+                    1976,
+                    688,
+                    1976
+                ],
+                "score": 0.92,
+                "latex": "G ( \\pmb { x } _ { i } , \\pmb { k } _ { i } , z ) \\in \\mathbb { R } ^ { d }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    543,
+                    1413,
+                    613,
+                    1413,
+                    613,
+                    1448,
+                    543,
+                    1448
+                ],
+                "score": 0.92,
+                "latex": "\\pmb { x } _ { i } ^ { m i s s }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    440,
+                    1412,
+                    493,
+                    1412,
+                    493,
+                    1447,
+                    440,
+                    1447
+                ],
+                "score": 0.92,
+                "latex": "\\pmb { x } _ { i } ^ { o b s }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    298,
+                    1581,
+                    497,
+                    1581,
+                    497,
+                    1616,
+                    298,
+                    1616
+                ],
+                "score": 0.92,
+                "latex": "P ( y | \\mathbf { x } _ { i } ^ { o b s } , \\mathbf { x } _ { i } ^ { m i s s } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    297,
+                    1319,
+                    394,
+                    1319,
+                    394,
+                    1353,
+                    297,
+                    1353
+                ],
+                "score": 0.91,
+                "latex": "\\pmb { x } _ { i } \\in \\mathbb { R } ^ { d }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    989,
+                    1521,
+                    1159,
+                    1521,
+                    1159,
+                    1555,
+                    989,
+                    1555
+                ],
+                "score": 0.91,
+                "latex": "P ( \\pmb { x } _ { i } ^ { m i s s } | \\pmb { x } _ { i } ^ { o b s } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    664,
+                    2004,
+                    694,
+                    2004,
+                    694,
+                    2034,
+                    664,
+                    2034
+                ],
+                "score": 0.88,
+                "latex": "\\widehat { \\pmb { x } } _ { i }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    298,
+                    1451,
+                    327,
+                    1451,
+                    327,
+                    1475,
+                    298,
+                    1475
+                ],
+                "score": 0.82,
+                "latex": "\\mathbf { \\Delta } _ { \\mathbf { \\mathcal { X } } _ { i } }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    520,
+                    481,
+                    542,
+                    481,
+                    542,
+                    507,
+                    520,
+                    507
+                ],
+                "score": 0.81,
+                "latex": "L"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1259,
+                    1256,
+                    1285,
+                    1256,
+                    1285,
+                    1282,
+                    1259,
+                    1282
+                ],
+                "score": 0.81,
+                "latex": "\\mathcal { D }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1112,
+                    1974,
+                    1150,
+                    1974,
+                    1150,
+                    2004,
+                    1112,
+                    2004
+                ],
+                "score": 0.68,
+                "latex": "( z )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1178,
+                    1522,
+                    1226,
+                    1522,
+                    1226,
+                    1555,
+                    1178,
+                    1555
+                ],
+                "score": 0.64,
+                "latex": "( i i )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1115,
+                    451,
+                    1136,
+                    451,
+                    1136,
+                    478,
+                    1115,
+                    478
+                ],
+                "score": 0.63,
+                "latex": "h"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    509,
+                    324,
+                    629,
+                    324,
+                    629,
+                    355,
+                    509,
+                    355
+                ],
+                "score": 0.58,
+                "latex": "G ( x _ { i } , k _ { i } , z )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    362,
+                    326,
+                    386,
+                    326,
+                    386,
+                    354,
+                    362,
+                    354
+                ],
+                "score": 0.58,
+                "latex": "k _ { i }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    710,
+                    1491,
+                    746,
+                    1491,
+                    746,
+                    1524,
+                    710,
+                    1524
+                ],
+                "score": 0.56,
+                "latex": "( i )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1001,
+                    323,
+                    1071,
+                    323,
+                    1071,
+                    356,
+                    1001,
+                    356
+                ],
+                "score": 0.56,
+                "latex": "D ( \\widehat { x } _ { i } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    362,
+                    298,
+                    386,
+                    298,
+                    386,
+                    318,
+                    362,
+                    318
+                ],
+                "score": 0.33,
+                "latex": "x _ { i }"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    354.0,
+                    282.0,
+                    361.0,
+                    282.0,
+                    361.0,
+                    397.0,
+                    354.0,
+                    397.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    387.0,
+                    282.0,
+                    396.0,
+                    282.0,
+                    396.0,
+                    397.0,
+                    387.0,
+                    397.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    504.0,
+                    316.0,
+                    508.0,
+                    316.0,
+                    508.0,
+                    363.0,
+                    504.0,
+                    363.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    630.0,
+                    316.0,
+                    638.0,
+                    316.0,
+                    638.0,
+                    363.0,
+                    630.0,
+                    363.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    736.0,
+                    309.0,
+                    931.0,
+                    309.0,
+                    931.0,
+                    362.0,
+                    736.0,
+                    362.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    992.0,
+                    311.0,
+                    1000.0,
+                    311.0,
+                    1000.0,
+                    367.0,
+                    992.0,
+                    367.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1072.0,
+                    311.0,
+                    1084.0,
+                    311.0,
+                    1084.0,
+                    367.0,
+                    1072.0,
+                    367.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1180.0,
+                    301.0,
+                    1265.0,
+                    301.0,
+                    1265.0,
+                    369.0,
+                    1180.0,
+                    369.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1206.0,
+                    363.0,
+                    1314.0,
+                    363.0,
+                    1314.0,
+                    412.0,
+                    1206.0,
+                    412.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    444.0,
+                    1114.0,
+                    444.0,
+                    1114.0,
+                    488.0,
+                    293.0,
+                    488.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1137.0,
+                    444.0,
+                    1407.0,
+                    444.0,
+                    1407.0,
+                    488.0,
+                    1137.0,
+                    488.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    479.0,
+                    519.0,
+                    479.0,
+                    519.0,
+                    511.0,
+                    297.0,
+                    511.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    543.0,
+                    479.0,
+                    960.0,
+                    479.0,
+                    960.0,
+                    511.0,
+                    543.0,
+                    511.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1196.0,
+                    636.0,
+                    1196.0,
+                    636.0,
+                    1232.0,
+                    294.0,
+                    1232.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1121.0,
+                    645.0,
+                    1121.0,
+                    645.0,
+                    1167.0,
+                    291.0,
+                    1167.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1742.0,
+                    682.0,
+                    1742.0,
+                    682.0,
+                    1781.0,
+                    294.0,
+                    1781.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    73.0,
+                    858.0,
+                    73.0,
+                    858.0,
+                    108.0,
+                    297.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    839.0,
+                    2085.0,
+                    860.0,
+                    2085.0,
+                    860.0,
+                    2116.0,
+                    839.0,
+                    2116.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    838.0,
+                    2085.0,
+                    862.0,
+                    2085.0,
+                    862.0,
+                    2118.0,
+                    838.0,
+                    2118.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    72.0,
+                    858.0,
+                    72.0,
+                    858.0,
+                    108.0,
+                    297.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    649.0,
+                    1405.0,
+                    649.0,
+                    1405.0,
+                    688.0,
+                    293.0,
+                    688.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    683.0,
+                    1406.0,
+                    683.0,
+                    1406.0,
+                    717.0,
+                    293.0,
+                    717.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    712.0,
+                    1408.0,
+                    712.0,
+                    1408.0,
+                    747.0,
+                    295.0,
+                    747.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    741.0,
+                    1407.0,
+                    741.0,
+                    1407.0,
+                    778.0,
+                    293.0,
+                    778.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    774.0,
+                    1406.0,
+                    774.0,
+                    1406.0,
+                    809.0,
+                    295.0,
+                    809.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    804.0,
+                    1406.0,
+                    804.0,
+                    1406.0,
+                    839.0,
+                    295.0,
+                    839.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    835.0,
+                    1406.0,
+                    835.0,
+                    1406.0,
+                    868.0,
+                    293.0,
+                    868.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    865.0,
+                    1406.0,
+                    865.0,
+                    1406.0,
+                    900.0,
+                    293.0,
+                    900.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    896.0,
+                    1406.0,
+                    896.0,
+                    1406.0,
+                    931.0,
+                    295.0,
+                    931.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    926.0,
+                    1402.0,
+                    926.0,
+                    1402.0,
+                    958.0,
+                    296.0,
+                    958.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    957.0,
+                    1407.0,
+                    957.0,
+                    1407.0,
+                    992.0,
+                    295.0,
+                    992.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    987.0,
+                    1407.0,
+                    987.0,
+                    1407.0,
+                    1021.0,
+                    293.0,
+                    1021.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1016.0,
+                    1408.0,
+                    1016.0,
+                    1408.0,
+                    1051.0,
+                    293.0,
+                    1051.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1048.0,
+                    1026.0,
+                    1048.0,
+                    1026.0,
+                    1080.0,
+                    296.0,
+                    1080.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1488.0,
+                    709.0,
+                    1488.0,
+                    709.0,
+                    1529.0,
+                    293.0,
+                    1529.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    747.0,
+                    1488.0,
+                    1406.0,
+                    1488.0,
+                    1406.0,
+                    1529.0,
+                    747.0,
+                    1529.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    288.0,
+                    1511.0,
+                    988.0,
+                    1511.0,
+                    988.0,
+                    1566.0,
+                    288.0,
+                    1566.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1160.0,
+                    1511.0,
+                    1177.0,
+                    1511.0,
+                    1177.0,
+                    1566.0,
+                    1160.0,
+                    1566.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1227.0,
+                    1511.0,
+                    1410.0,
+                    1511.0,
+                    1410.0,
+                    1566.0,
+                    1227.0,
+                    1566.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1549.0,
+                    1408.0,
+                    1549.0,
+                    1408.0,
+                    1589.0,
+                    293.0,
+                    1589.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    286.0,
+                    1569.0,
+                    297.0,
+                    1569.0,
+                    297.0,
+                    1631.0,
+                    286.0,
+                    1631.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    498.0,
+                    1569.0,
+                    1414.0,
+                    1569.0,
+                    1414.0,
+                    1631.0,
+                    498.0,
+                    1631.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1611.0,
+                    1404.0,
+                    1611.0,
+                    1404.0,
+                    1648.0,
+                    293.0,
+                    1648.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1644.0,
+                    1404.0,
+                    1644.0,
+                    1404.0,
+                    1679.0,
+                    294.0,
+                    1679.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1673.0,
+                    1146.0,
+                    1673.0,
+                    1146.0,
+                    1708.0,
+                    293.0,
+                    1708.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1252.0,
+                    1258.0,
+                    1252.0,
+                    1258.0,
+                    1290.0,
+                    292.0,
+                    1290.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1286.0,
+                    1252.0,
+                    1406.0,
+                    1252.0,
+                    1406.0,
+                    1290.0,
+                    1286.0,
+                    1290.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1284.0,
+                    991.0,
+                    1284.0,
+                    991.0,
+                    1320.0,
+                    294.0,
+                    1320.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1116.0,
+                    1284.0,
+                    1408.0,
+                    1284.0,
+                    1408.0,
+                    1320.0,
+                    1116.0,
+                    1320.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1317.0,
+                    296.0,
+                    1317.0,
+                    296.0,
+                    1359.0,
+                    292.0,
+                    1359.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    395.0,
+                    1317.0,
+                    1051.0,
+                    1317.0,
+                    1051.0,
+                    1359.0,
+                    395.0,
+                    1359.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1196.0,
+                    1317.0,
+                    1408.0,
+                    1317.0,
+                    1408.0,
+                    1359.0,
+                    1196.0,
+                    1359.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1351.0,
+                    1406.0,
+                    1351.0,
+                    1406.0,
+                    1389.0,
+                    292.0,
+                    1389.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1381.0,
+                    1408.0,
+                    1381.0,
+                    1408.0,
+                    1419.0,
+                    294.0,
+                    1419.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    283.0,
+                    1401.0,
+                    439.0,
+                    1401.0,
+                    439.0,
+                    1462.0,
+                    283.0,
+                    1462.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    494.0,
+                    1401.0,
+                    542.0,
+                    1401.0,
+                    542.0,
+                    1462.0,
+                    494.0,
+                    1462.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    614.0,
+                    1401.0,
+                    1414.0,
+                    1401.0,
+                    1414.0,
+                    1462.0,
+                    614.0,
+                    1462.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1448.0,
+                    297.0,
+                    1448.0,
+                    297.0,
+                    1481.0,
+                    292.0,
+                    1481.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    328.0,
+                    1448.0,
+                    341.0,
+                    1448.0,
+                    341.0,
+                    1481.0,
+                    328.0,
+                    1481.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1802.0,
+                    1410.0,
+                    1802.0,
+                    1410.0,
+                    1839.0,
+                    293.0,
+                    1839.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1833.0,
+                    1406.0,
+                    1833.0,
+                    1406.0,
+                    1870.0,
+                    294.0,
+                    1870.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1865.0,
+                    1405.0,
+                    1865.0,
+                    1405.0,
+                    1901.0,
+                    294.0,
+                    1901.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1893.0,
+                    725.0,
+                    1893.0,
+                    725.0,
+                    1929.0,
+                    294.0,
+                    1929.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1938.0,
+                    687.0,
+                    1938.0,
+                    687.0,
+                    1979.0,
+                    294.0,
+                    1979.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    896.0,
+                    1938.0,
+                    1407.0,
+                    1938.0,
+                    1407.0,
+                    1979.0,
+                    896.0,
+                    1979.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1973.0,
+                    1111.0,
+                    1973.0,
+                    1111.0,
+                    2007.0,
+                    296.0,
+                    2007.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1151.0,
+                    1973.0,
+                    1405.0,
+                    1973.0,
+                    1405.0,
+                    2007.0,
+                    1151.0,
+                    2007.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    2000.0,
+                    663.0,
+                    2000.0,
+                    663.0,
+                    2039.0,
+                    295.0,
+                    2039.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    695.0,
+                    2000.0,
+                    1403.0,
+                    2000.0,
+                    1403.0,
+                    2039.0,
+                    695.0,
+                    2039.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    570.0,
+                    1403.0,
+                    570.0,
+                    1403.0,
+                    612.0,
+                    295.0,
+                    612.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    603.0,
+                    836.0,
+                    603.0,
+                    836.0,
+                    639.0,
+                    295.0,
+                    639.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 2,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    790,
+                    1405,
+                    790,
+                    1405,
+                    1008,
+                    297,
+                    1008
+                ],
+                "score": 0.983
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    296,
+                    1099,
+                    1405,
+                    1099,
+                    1405,
+                    1287,
+                    296,
+                    1287
+                ],
+                "score": 0.981
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    482,
+                    1404,
+                    482,
+                    1404,
+                    583,
+                    298,
+                    583
+                ],
+                "score": 0.977
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    371,
+                    1402,
+                    371,
+                    1402,
+                    469,
+                    298,
+                    469
+                ],
+                "score": 0.976
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    683,
+                    1403,
+                    683,
+                    1403,
+                    778,
+                    297,
+                    778
+                ],
+                "score": 0.976
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1658,
+                    1405,
+                    1658,
+                    1405,
+                    1753,
+                    298,
+                    1753
+                ],
+                "score": 0.964
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    645,
+                    1896,
+                    1051,
+                    1896,
+                    1051,
+                    1959,
+                    645,
+                    1959
+                ],
+                "score": 0.956
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    676,
+                    1579,
+                    1023,
+                    1579,
+                    1023,
+                    1643,
+                    676,
+                    1643
+                ],
+                "score": 0.955
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    638,
+                    280,
+                    1060,
+                    280,
+                    1060,
+                    358,
+                    638,
+                    358
+                ],
+                "score": 0.954
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1971,
+                    1404,
+                    1971,
+                    1404,
+                    2036,
+                    297,
+                    2036
+                ],
+                "score": 0.952
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    291,
+                    1818,
+                    1403,
+                    1818,
+                    1403,
+                    1882,
+                    291,
+                    1882
+                ],
+                "score": 0.948
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    621,
+                    1767,
+                    1076,
+                    1767,
+                    1076,
+                    1806,
+                    621,
+                    1806
+                ],
+                "score": 0.942
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1403,
+                    1411,
+                    1403,
+                    1411,
+                    1461,
+                    297,
+                    1461
+                ],
+                "score": 0.941
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    599,
+                    1460,
+                    1097,
+                    1460,
+                    1097,
+                    1523,
+                    599,
+                    1523
+                ],
+                "score": 0.939
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    394,
+                    598,
+                    1302,
+                    598,
+                    1302,
+                    650,
+                    394,
+                    650
+                ],
+                "score": 0.932
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    756,
+                    1349,
+                    940,
+                    1349,
+                    940,
+                    1389,
+                    756,
+                    1389
+                ],
+                "score": 0.929
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    294,
+                    228,
+                    1174,
+                    228,
+                    1174,
+                    263,
+                    294,
+                    263
+                ],
+                "score": 0.925
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    297,
+                    73,
+                    858,
+                    73,
+                    858,
+                    106,
+                    297,
+                    106
+                ],
+                "score": 0.923
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1531,
+                    1374,
+                    1531,
+                    1374,
+                    1564,
+                    297,
+                    1564
+                ],
+                "score": 0.922
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    299,
+                    1043,
+                    675,
+                    1043,
+                    675,
+                    1075,
+                    299,
+                    1075
+                ],
+                "score": 0.919
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    292,
+                    1299,
+                    1368,
+                    1299,
+                    1368,
+                    1333,
+                    292,
+                    1333
+                ],
+                "score": 0.919
+            },
+            {
+                "category_id": 9,
+                "poly": [
+                    1366,
+                    302,
+                    1400,
+                    302,
+                    1400,
+                    333,
+                    1366,
+                    333
+                ],
+                "score": 0.895
+            },
+            {
+                "category_id": 9,
+                "poly": [
+                    1366,
+                    1772,
+                    1399,
+                    1772,
+                    1399,
+                    1801,
+                    1366,
+                    1801
+                ],
+                "score": 0.891
+            },
+            {
+                "category_id": 9,
+                "poly": [
+                    1366,
+                    1596,
+                    1400,
+                    1596,
+                    1400,
+                    1626,
+                    1366,
+                    1626
+                ],
+                "score": 0.89
+            },
+            {
+                "category_id": 9,
+                "poly": [
+                    1366,
+                    1354,
+                    1400,
+                    1354,
+                    1400,
+                    1384,
+                    1366,
+                    1384
+                ],
+                "score": 0.886
+            },
+            {
+                "category_id": 9,
+                "poly": [
+                    1366,
+                    1913,
+                    1400,
+                    1913,
+                    1400,
+                    1944,
+                    1366,
+                    1944
+                ],
+                "score": 0.881
+            },
+            {
+                "category_id": 9,
+                "poly": [
+                    1366,
+                    1476,
+                    1400,
+                    1476,
+                    1400,
+                    1505,
+                    1366,
+                    1505
+                ],
+                "score": 0.872
+            },
+            {
+                "category_id": 9,
+                "poly": [
+                    1366,
+                    607,
+                    1400,
+                    607,
+                    1400,
+                    637,
+                    1366,
+                    637
+                ],
+                "score": 0.865
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    840,
+                    2088,
+                    858,
+                    2088,
+                    858,
+                    2111,
+                    840,
+                    2111
+                ],
+                "score": 0.804
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    644,
+                    1893,
+                    1057,
+                    1893,
+                    1057,
+                    1960,
+                    644,
+                    1960
+                ],
+                "score": 0.95,
+                "latex": "\\Psi = \\mathbb { E } _ { \\mathbf { z } } [ F _ { \\boldsymbol { \\theta } } ( \\widehat { \\mathbf { x } } _ { i } ) ] \\approx \\frac { 1 } { N } { \\sum _ { l = 1 } ^ { N } } F _ { \\boldsymbol { \\theta } } ( \\widehat { \\mathbf { x } } _ { i } ^ { l } ) ~ ."
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    674,
+                    1576,
+                    1023,
+                    1576,
+                    1023,
+                    1644,
+                    674,
+                    1644
+                ],
+                "score": 0.95,
+                "latex": "\\Psi \\approx \\frac { 1 } { N } \\sum P ( y | \\pmb { x } _ { i } ^ { o b s } , \\widehat { \\pmb { x } } _ { i } ^ { m i s s } ) ,"
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    638,
+                    278,
+                    1059,
+                    278,
+                    1059,
+                    357,
+                    638,
+                    357
+                ],
+                "score": 0.94,
+                "latex": "\\begin{array} { r } { \\widehat { \\pmb { x } } _ { i , j } = \\left\\{ \\begin{array} { l l } { \\pmb { x } _ { i , j } } & { \\mathrm { i f } ~ \\pmb { k } _ { i , j } = 1 } \\\\ { G ( \\pmb { x } _ { i } , \\pmb { k } _ { i } , z ) _ { j } } & { \\mathrm { i f } ~ \\pmb { k } _ { i , j } = 0 } \\end{array} \\right. , } \\end{array}"
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    599,
+                    1454,
+                    1101,
+                    1454,
+                    1101,
+                    1525,
+                    599,
+                    1525
+                ],
+                "score": 0.94,
+                "latex": "\\Psi = \\int P ( \\pmb { x } _ { i } ^ { m i s s } ) P ( y | \\pmb { x } _ { i } ^ { o b s } , \\pmb { x } _ { i } ^ { m i s s } ) d \\pmb { x } _ { i } ^ { m i s s } ."
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    755,
+                    1347,
+                    945,
+                    1347,
+                    945,
+                    1389,
+                    755,
+                    1389
+                ],
+                "score": 0.92,
+                "latex": "\\Psi = P ( y | \\mathbf { x } _ { i } ^ { o b s } ) ."
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    621,
+                    1767,
+                    1077,
+                    1767,
+                    1077,
+                    1806,
+                    621,
+                    1806
+                ],
+                "score": 0.91,
+                "latex": "\\widehat { \\mathbf { x } } _ { i } = \\pmb { k } _ { i } \\odot \\pmb { x } _ { i } + \\left( 1 - \\pmb { k } _ { i } \\right) \\odot G ( \\pmb { x } _ { i } , \\pmb { k } _ { i } , z )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    584,
+                    484,
+                    658,
+                    484,
+                    658,
+                    517,
+                    584,
+                    517
+                ],
+                "score": 0.91,
+                "latex": "D ( \\widehat { \\mathbf { x } } _ { i } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    371,
+                    1659,
+                    440,
+                    1659,
+                    440,
+                    1692,
+                    371,
+                    1692
+                ],
+                "score": 0.9,
+                "latex": "\\widehat { \\pmb { x } } _ { i } ^ { m i s s }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1213,
+                    404,
+                    1244,
+                    404,
+                    1244,
+                    440,
+                    1213,
+                    440
+                ],
+                "score": 0.9,
+                "latex": "\\widehat { \\pmb { x } } _ { i } ^ { l }"
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    393,
+                    596,
+                    1303,
+                    596,
+                    1303,
+                    653,
+                    393,
+                    653
+                ],
+                "score": 0.89,
+                "latex": "\\operatorname* { m a x } _ { G } \\operatorname* { m i n } _ { D } L ( G , D ) = \\mathbb { E } _ { k \\sim \\mathcal { D } , \\widehat { k } \\sim D ( G ( x , k , z ) ) } \\ [ k ^ { T } \\log ( \\widehat { k } ) \\ + \\ ( 1 - k ) ^ { T } \\log ( 1 - \\widehat { k } ) ] ."
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    372,
+                    378,
+                    419,
+                    378,
+                    419,
+                    407,
+                    372,
+                    407
+                ],
+                "score": 0.89,
+                "latex": "\\mathbf { \\delta } _ { \\mathbf { x } _ { i , j } }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    599,
+                    1821,
+                    631,
+                    1821,
+                    631,
+                    1850,
+                    599,
+                    1850
+                ],
+                "score": 0.89,
+                "latex": "F _ { \\theta }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    649,
+                    516,
+                    678,
+                    516,
+                    678,
+                    552,
+                    649,
+                    552
+                ],
+                "score": 0.88,
+                "latex": "\\widehat { \\pmb { k } } _ { i }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1122,
+                    1533,
+                    1146,
+                    1533,
+                    1146,
+                    1559,
+                    1122,
+                    1559
+                ],
+                "score": 0.85,
+                "latex": "\\Psi"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    366,
+                    1405,
+                    391,
+                    1405,
+                    391,
+                    1431,
+                    366,
+                    1431
+                ],
+                "score": 0.84,
+                "latex": "\\Psi"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    379,
+                    1853,
+                    403,
+                    1853,
+                    403,
+                    1877,
+                    379,
+                    1877
+                ],
+                "score": 0.82,
+                "latex": "\\Psi"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    523,
+                    375,
+                    542,
+                    375,
+                    542,
+                    405,
+                    523,
+                    405
+                ],
+                "score": 0.81,
+                "latex": "j ^ { : }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1322,
+                    408,
+                    1335,
+                    408,
+                    1335,
+                    434,
+                    1322,
+                    434
+                ],
+                "score": 0.76,
+                "latex": "l"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1119,
+                    378,
+                    1139,
+                    378,
+                    1139,
+                    401,
+                    1119,
+                    401
+                ],
+                "score": 0.75,
+                "latex": "_ { z }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    776,
+                    375,
+                    789,
+                    375,
+                    789,
+                    400,
+                    776,
+                    400
+                ],
+                "score": 0.72,
+                "latex": "i"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    72.0,
+                    859.0,
+                    72.0,
+                    859.0,
+                    108.0,
+                    297.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1043.0,
+                    678.0,
+                    1043.0,
+                    678.0,
+                    1079.0,
+                    295.0,
+                    1079.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    837.0,
+                    2086.0,
+                    861.0,
+                    2086.0,
+                    861.0,
+                    2118.0,
+                    837.0,
+                    2118.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    792.0,
+                    1408.0,
+                    792.0,
+                    1408.0,
+                    827.0,
+                    296.0,
+                    827.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    822.0,
+                    1406.0,
+                    822.0,
+                    1406.0,
+                    859.0,
+                    292.0,
+                    859.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    853.0,
+                    1405.0,
+                    853.0,
+                    1405.0,
+                    888.0,
+                    294.0,
+                    888.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    884.0,
+                    1406.0,
+                    884.0,
+                    1406.0,
+                    919.0,
+                    294.0,
+                    919.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    915.0,
+                    1405.0,
+                    915.0,
+                    1405.0,
+                    950.0,
+                    295.0,
+                    950.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    945.0,
+                    1406.0,
+                    945.0,
+                    1406.0,
+                    980.0,
+                    295.0,
+                    980.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    977.0,
+                    735.0,
+                    977.0,
+                    735.0,
+                    1008.0,
+                    295.0,
+                    1008.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1102.0,
+                    1403.0,
+                    1102.0,
+                    1403.0,
+                    1134.0,
+                    295.0,
+                    1134.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1133.0,
+                    1405.0,
+                    1133.0,
+                    1405.0,
+                    1167.0,
+                    294.0,
+                    1167.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1160.0,
+                    1405.0,
+                    1160.0,
+                    1405.0,
+                    1197.0,
+                    293.0,
+                    1197.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1193.0,
+                    1406.0,
+                    1193.0,
+                    1406.0,
+                    1228.0,
+                    294.0,
+                    1228.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1225.0,
+                    1405.0,
+                    1225.0,
+                    1405.0,
+                    1257.0,
+                    295.0,
+                    1257.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1255.0,
+                    1216.0,
+                    1255.0,
+                    1216.0,
+                    1287.0,
+                    296.0,
+                    1287.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    480.0,
+                    583.0,
+                    480.0,
+                    583.0,
+                    522.0,
+                    291.0,
+                    522.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    659.0,
+                    480.0,
+                    1408.0,
+                    480.0,
+                    1408.0,
+                    522.0,
+                    659.0,
+                    522.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    521.0,
+                    648.0,
+                    521.0,
+                    648.0,
+                    556.0,
+                    294.0,
+                    556.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    679.0,
+                    521.0,
+                    1404.0,
+                    521.0,
+                    1404.0,
+                    556.0,
+                    679.0,
+                    556.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    551.0,
+                    640.0,
+                    551.0,
+                    640.0,
+                    587.0,
+                    294.0,
+                    587.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    369.0,
+                    371.0,
+                    369.0,
+                    371.0,
+                    410.0,
+                    293.0,
+                    410.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    420.0,
+                    369.0,
+                    522.0,
+                    369.0,
+                    522.0,
+                    410.0,
+                    420.0,
+                    410.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    543.0,
+                    369.0,
+                    775.0,
+                    369.0,
+                    775.0,
+                    410.0,
+                    543.0,
+                    410.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    790.0,
+                    369.0,
+                    1118.0,
+                    369.0,
+                    1118.0,
+                    410.0,
+                    790.0,
+                    410.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1140.0,
+                    369.0,
+                    1404.0,
+                    369.0,
+                    1404.0,
+                    410.0,
+                    1140.0,
+                    410.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    404.0,
+                    1212.0,
+                    404.0,
+                    1212.0,
+                    439.0,
+                    296.0,
+                    439.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1245.0,
+                    404.0,
+                    1321.0,
+                    404.0,
+                    1321.0,
+                    439.0,
+                    1245.0,
+                    439.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1336.0,
+                    404.0,
+                    1403.0,
+                    404.0,
+                    1403.0,
+                    439.0,
+                    1336.0,
+                    439.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    436.0,
+                    479.0,
+                    436.0,
+                    479.0,
+                    472.0,
+                    293.0,
+                    472.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    682.0,
+                    1404.0,
+                    682.0,
+                    1404.0,
+                    720.0,
+                    294.0,
+                    720.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    713.0,
+                    1404.0,
+                    713.0,
+                    1404.0,
+                    750.0,
+                    294.0,
+                    750.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    744.0,
+                    780.0,
+                    744.0,
+                    780.0,
+                    782.0,
+                    294.0,
+                    782.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    290.0,
+                    1657.0,
+                    370.0,
+                    1657.0,
+                    370.0,
+                    1694.0,
+                    290.0,
+                    1694.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    441.0,
+                    1657.0,
+                    1405.0,
+                    1657.0,
+                    1405.0,
+                    1694.0,
+                    441.0,
+                    1694.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1691.0,
+                    1405.0,
+                    1691.0,
+                    1405.0,
+                    1725.0,
+                    295.0,
+                    1725.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1721.0,
+                    1307.0,
+                    1721.0,
+                    1307.0,
+                    1755.0,
+                    294.0,
+                    1755.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1971.0,
+                    1404.0,
+                    1971.0,
+                    1404.0,
+                    2007.0,
+                    295.0,
+                    2007.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    2001.0,
+                    1404.0,
+                    2001.0,
+                    1404.0,
+                    2037.0,
+                    295.0,
+                    2037.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1815.0,
+                    598.0,
+                    1815.0,
+                    598.0,
+                    1856.0,
+                    293.0,
+                    1856.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    632.0,
+                    1815.0,
+                    1405.0,
+                    1815.0,
+                    1405.0,
+                    1856.0,
+                    632.0,
+                    1856.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1851.0,
+                    378.0,
+                    1851.0,
+                    378.0,
+                    1883.0,
+                    295.0,
+                    1883.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    404.0,
+                    1851.0,
+                    634.0,
+                    1851.0,
+                    634.0,
+                    1883.0,
+                    404.0,
+                    1883.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1399.0,
+                    365.0,
+                    1399.0,
+                    365.0,
+                    1439.0,
+                    293.0,
+                    1439.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    392.0,
+                    1399.0,
+                    1404.0,
+                    1399.0,
+                    1404.0,
+                    1439.0,
+                    392.0,
+                    1439.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1432.0,
+                    405.0,
+                    1432.0,
+                    405.0,
+                    1468.0,
+                    293.0,
+                    1468.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    227.0,
+                    1177.0,
+                    227.0,
+                    1177.0,
+                    268.0,
+                    292.0,
+                    268.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1525.0,
+                    1121.0,
+                    1525.0,
+                    1121.0,
+                    1572.0,
+                    294.0,
+                    1572.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1147.0,
+                    1525.0,
+                    1380.0,
+                    1525.0,
+                    1380.0,
+                    1572.0,
+                    1147.0,
+                    1572.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1296.0,
+                    1370.0,
+                    1296.0,
+                    1370.0,
+                    1336.0,
+                    296.0,
+                    1336.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 3,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1558,
+                    1406,
+                    1558,
+                    1406,
+                    1789,
+                    297,
+                    1789
+                ],
+                "score": 0.98
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1132,
+                    1404,
+                    1132,
+                    1404,
+                    1257,
+                    297,
+                    1257
+                ],
+                "score": 0.979
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    855,
+                    1405,
+                    855,
+                    1405,
+                    1012,
+                    297,
+                    1012
+                ],
+                "score": 0.977
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    687,
+                    1405,
+                    687,
+                    1405,
+                    841,
+                    297,
+                    841
+                ],
+                "score": 0.976
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1322,
+                    1404,
+                    1322,
+                    1404,
+                    1414,
+                    297,
+                    1414
+                ],
+                "score": 0.973
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    1024,
+                    1399,
+                    1024,
+                    1399,
+                    1118,
+                    299,
+                    1118
+                ],
+                "score": 0.958
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    1480,
+                    1405,
+                    1480,
+                    1405,
+                    1544,
+                    299,
+                    1544
+                ],
+                "score": 0.952
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    697,
+                    1429,
+                    1001,
+                    1429,
+                    1001,
+                    1469,
+                    697,
+                    1469
+                ],
+                "score": 0.947
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    296,
+                    1802,
+                    1118,
+                    1802,
+                    1118,
+                    1958,
+                    296,
+                    1958
+                ],
+                "score": 0.946
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    688,
+                    1268,
+                    1011,
+                    1268,
+                    1011,
+                    1309,
+                    688,
+                    1309
+                ],
+                "score": 0.946
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    296,
+                    1971,
+                    1399,
+                    1971,
+                    1399,
+                    2034,
+                    296,
+                    2034
+                ],
+                "score": 0.929
+            },
+            {
+                "category_id": 9,
+                "poly": [
+                    1366,
+                    1275,
+                    1399,
+                    1275,
+                    1399,
+                    1304,
+                    1366,
+                    1304
+                ],
+                "score": 0.896
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    298,
+                    76,
+                    858,
+                    76,
+                    858,
+                    104,
+                    298,
+                    104
+                ],
+                "score": 0.888
+            },
+            {
+                "category_id": 9,
+                "poly": [
+                    1366,
+                    1434,
+                    1399,
+                    1434,
+                    1399,
+                    1463,
+                    1366,
+                    1463
+                ],
+                "score": 0.875
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    841,
+                    2088,
+                    858,
+                    2088,
+                    858,
+                    2112,
+                    841,
+                    2112
+                ],
+                "score": 0.743
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    823,
+                    238,
+                    1328,
+                    238,
+                    1328,
+                    602,
+                    823,
+                    602
+                ],
+                "score": 0.499
+            },
+            {
+                "category_id": 5,
+                "poly": [
+                    297,
+                    236,
+                    809,
+                    236,
+                    809,
+                    615,
+                    297,
+                    615
+                ],
+                "score": 0.497,
+                "html": "<table><tr><td>Algorithm1: Training the predictor.</td></tr><tr><td>Input: G (trained imputer), D (dataset) Output: Fθ (trained predictor) foreachTrainingEpoch do foreach(xi,ki,yi) in D do z ~ N(0,I) Xi ←ki①xi+(1-ki)OG(xi,ki,z) loss ← L(yi,yred) Backpropagate loss</td></tr></table>"
+            },
+            {
+                "category_id": 6,
+                "poly": [
+                    298,
+                    238,
+                    706,
+                    238,
+                    706,
+                    270,
+                    298,
+                    270
+                ],
+                "score": 0.175
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    919,
+                    1558,
+                    1069,
+                    1558,
+                    1069,
+                    1597,
+                    919,
+                    1597
+                ],
+                "score": 0.94,
+                "latex": "\\{ \\widehat { y _ { i } } ^ { 1 } \\ldots \\widehat { y _ { i } } ^ { N } \\}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1033,
+                    1595,
+                    1186,
+                    1595,
+                    1186,
+                    1633,
+                    1033,
+                    1633
+                ],
+                "score": 0.93,
+                "latex": "\\{ \\widehat { x _ { i } } ^ { 1 } \\ldots \\widehat { x _ { i } } ^ { N } \\}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    386,
+                    1594,
+                    540,
+                    1594,
+                    540,
+                    1633,
+                    386,
+                    1633
+                ],
+                "score": 0.93,
+                "latex": "\\{ \\widehat { x _ { i } } ^ { 1 } \\ldots \\widehat { x _ { i } } ^ { N } \\}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    299,
+                    1661,
+                    448,
+                    1661,
+                    448,
+                    1699,
+                    299,
+                    1699
+                ],
+                "score": 0.92,
+                "latex": "\\{ \\widehat { y _ { i } } ^ { 1 } \\ldots \\widehat { y _ { i } } ^ { N } \\}"
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    685,
+                    1267,
+                    1012,
+                    1267,
+                    1012,
+                    1309,
+                    685,
+                    1309
+                ],
+                "score": 0.92,
+                "latex": "P ( y _ { i } = \\widehat { y _ { i } } ) = ( 1 - \\epsilon ) ^ { 2 } + \\epsilon ^ { 2 } ,"
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    696,
+                    1429,
+                    1003,
+                    1429,
+                    1003,
+                    1468,
+                    696,
+                    1468
+                ],
+                "score": 0.92,
+                "latex": "\\epsilon = 1 - P ( \\widehat { y _ { i } } = j | y _ { i } = j ) ."
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1163,
+                    278,
+                    1234,
+                    278,
+                    1234,
+                    311,
+                    1163,
+                    311
+                ],
+                "score": 0.89,
+                "latex": "( x , k )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    425,
+                    1836,
+                    464,
+                    1836,
+                    464,
+                    1865,
+                    425,
+                    1865
+                ],
+                "score": 0.88,
+                "latex": "\\widehat { x _ { i } s }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    371,
+                    1323,
+                    396,
+                    1323,
+                    396,
+                    1354,
+                    371,
+                    1354
+                ],
+                "score": 0.87,
+                "latex": "\\widehat { y _ { i } }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1010,
+                    1026,
+                    1049,
+                    1026,
+                    1049,
+                    1056,
+                    1010,
+                    1056
+                ],
+                "score": 0.86,
+                "latex": "\\widehat { x _ { i } s }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    435,
+                    369,
+                    560,
+                    369,
+                    560,
+                    399,
+                    435,
+                    399
+                ],
+                "score": 0.86,
+                "latex": "( \\boldsymbol { x } _ { i } , \\boldsymbol { k } _ { i } , y _ { i } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    401,
+                    308,
+                    434,
+                    308,
+                    434,
+                    338,
+                    401,
+                    338
+                ],
+                "score": 0.85,
+                "latex": "F _ { \\theta }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1108,
+                    1327,
+                    1134,
+                    1327,
+                    1134,
+                    1354,
+                    1108,
+                    1354
+                ],
+                "score": 0.85,
+                "latex": "y _ { i }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1005,
+                    1323,
+                    1049,
+                    1323,
+                    1049,
+                    1354,
+                    1005,
+                    1354
+                ],
+                "score": 0.85,
+                "latex": "( \\widehat { x _ { i } } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    375,
+                    428,
+                    808,
+                    428,
+                    808,
+                    462,
+                    375,
+                    462
+                ],
+                "score": 0.81,
+                "latex": "\\widehat { \\mathbf { x } } _ { i } \\gets k _ { i } \\odot \\dot { { \\mathbf { x } } _ { i } } + ( 1 - k _ { i } ) \\odot G ( { \\mathbf { x } } _ { i } , k _ { i } , z )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1025,
+                    889,
+                    1053,
+                    889,
+                    1053,
+                    915,
+                    1025,
+                    915
+                ],
+                "score": 0.8,
+                "latex": "N"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    928,
+                    340,
+                    953,
+                    340,
+                    953,
+                    366,
+                    928,
+                    366
+                ],
+                "score": 0.8,
+                "latex": "\\Psi"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    479,
+                    1638,
+                    496,
+                    1638,
+                    496,
+                    1663,
+                    479,
+                    1663
+                ],
+                "score": 0.78,
+                "latex": "y"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    608,
+                    1930,
+                    630,
+                    1930,
+                    630,
+                    1953,
+                    608,
+                    1953
+                ],
+                "score": 0.77,
+                "latex": "\\Psi"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    638,
+                    1671,
+                    656,
+                    1671,
+                    656,
+                    1698,
+                    638,
+                    1698
+                ],
+                "score": 0.76,
+                "latex": "y"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    908,
+                    280,
+                    941,
+                    280,
+                    941,
+                    309,
+                    908,
+                    309
+                ],
+                "score": 0.75,
+                "latex": "F _ { \\theta }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    592,
+                    369,
+                    618,
+                    369,
+                    618,
+                    395,
+                    592,
+                    395
+                ],
+                "score": 0.74,
+                "latex": "\\mathcal { D }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    379,
+                    399,
+                    526,
+                    399,
+                    526,
+                    430,
+                    379,
+                    430
+                ],
+                "score": 0.71,
+                "latex": "z \\sim N ( 0 , I )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    438,
+                    1488,
+                    453,
+                    1488,
+                    453,
+                    1509,
+                    438,
+                    1509
+                ],
+                "score": 0.7,
+                "latex": "\\epsilon"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    837,
+                    1360,
+                    852,
+                    1360,
+                    852,
+                    1382,
+                    837,
+                    1382
+                ],
+                "score": 0.67,
+                "latex": "\\epsilon"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    381,
+                    278,
+                    407,
+                    278,
+                    407,
+                    305,
+                    381,
+                    305
+                ],
+                "score": 0.64,
+                "latex": "G"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    382,
+                    462,
+                    564,
+                    462,
+                    564,
+                    498,
+                    382,
+                    498
+                ],
+                "score": 0.56,
+                "latex": "y _ { i } ^ { p r e d }  F _ { \\theta } ( \\widehat { \\mathbf { x } } _ { i } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1170,
+                    402,
+                    1195,
+                    402,
+                    1195,
+                    427,
+                    1170,
+                    427
+                ],
+                "score": 0.56,
+                "latex": "N"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    612,
+                    278,
+                    638,
+                    278,
+                    638,
+                    305,
+                    612,
+                    305
+                ],
+                "score": 0.54,
+                "latex": "\\mathcal { D }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    857,
+                    461,
+                    1271,
+                    461,
+                    1271,
+                    494,
+                    857,
+                    494
+                ],
+                "score": 0.53,
+                "latex": "\\widehat { \\mathbf { x } }  k \\odot x + ( 1 - k ) \\odot G ( { \\pmb { x } } , { \\pmb { k } } , z )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1122,
+                    402,
+                    1139,
+                    402,
+                    1139,
+                    427,
+                    1122,
+                    427
+                ],
+                "score": 0.41,
+                "latex": "^ { l }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    382,
+                    500,
+                    614,
+                    500,
+                    614,
+                    536,
+                    382,
+                    536
+                ],
+                "score": 0.37,
+                "latex": "l o s s \\gets L ( y _ { i } , y _ { i } ^ { p r e d } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    824,
+                    368,
+                    1102,
+                    368,
+                    1102,
+                    399,
+                    824,
+                    399
+                ],
+                "score": 0.37,
+                "latex": "\\Psi  z e r o s \\in R ^ { \\# c l a s s e s }"
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    823,
+                    369,
+                    1101,
+                    369,
+                    1101,
+                    399,
+                    823,
+                    399
+                ],
+                "score": 0.37,
+                "latex": "\\Psi  z e r o s \\in R ^ { \\# c l a s s e s }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1008,
+                    310,
+                    1034,
+                    310,
+                    1034,
+                    337,
+                    1008,
+                    337
+                ],
+                "score": 0.36,
+                "latex": "\\mathbf { N }"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    70.0,
+                    859.0,
+                    70.0,
+                    859.0,
+                    109.0,
+                    296.0,
+                    109.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    839.0,
+                    2085.0,
+                    861.0,
+                    2085.0,
+                    861.0,
+                    2120.0,
+                    839.0,
+                    2120.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    232.0,
+                    710.0,
+                    232.0,
+                    710.0,
+                    278.0,
+                    294.0,
+                    278.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    290.0,
+                    1555.0,
+                    918.0,
+                    1555.0,
+                    918.0,
+                    1602.0,
+                    290.0,
+                    1602.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1070.0,
+                    1555.0,
+                    1409.0,
+                    1555.0,
+                    1409.0,
+                    1602.0,
+                    1070.0,
+                    1602.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    285.0,
+                    1579.0,
+                    385.0,
+                    1579.0,
+                    385.0,
+                    1648.0,
+                    285.0,
+                    1648.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    541.0,
+                    1579.0,
+                    1032.0,
+                    1579.0,
+                    1032.0,
+                    1648.0,
+                    541.0,
+                    1648.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1187.0,
+                    1579.0,
+                    1417.0,
+                    1579.0,
+                    1417.0,
+                    1648.0,
+                    1187.0,
+                    1648.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1628.0,
+                    478.0,
+                    1628.0,
+                    478.0,
+                    1668.0,
+                    294.0,
+                    1668.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    497.0,
+                    1628.0,
+                    1407.0,
+                    1628.0,
+                    1407.0,
+                    1668.0,
+                    497.0,
+                    1668.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1661.0,
+                    298.0,
+                    1661.0,
+                    298.0,
+                    1705.0,
+                    295.0,
+                    1705.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    449.0,
+                    1661.0,
+                    637.0,
+                    1661.0,
+                    637.0,
+                    1705.0,
+                    449.0,
+                    1705.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    657.0,
+                    1661.0,
+                    1408.0,
+                    1661.0,
+                    1408.0,
+                    1705.0,
+                    657.0,
+                    1705.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1694.0,
+                    1406.0,
+                    1694.0,
+                    1406.0,
+                    1730.0,
+                    292.0,
+                    1730.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1721.0,
+                    1407.0,
+                    1721.0,
+                    1407.0,
+                    1765.0,
+                    292.0,
+                    1765.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1755.0,
+                    511.0,
+                    1755.0,
+                    511.0,
+                    1791.0,
+                    294.0,
+                    1791.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1130.0,
+                    1405.0,
+                    1130.0,
+                    1405.0,
+                    1171.0,
+                    294.0,
+                    1171.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1165.0,
+                    1407.0,
+                    1165.0,
+                    1407.0,
+                    1198.0,
+                    295.0,
+                    1198.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1193.0,
+                    1405.0,
+                    1193.0,
+                    1405.0,
+                    1231.0,
+                    292.0,
+                    1231.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1227.0,
+                    1169.0,
+                    1227.0,
+                    1169.0,
+                    1259.0,
+                    292.0,
+                    1259.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    857.0,
+                    1406.0,
+                    857.0,
+                    1406.0,
+                    891.0,
+                    296.0,
+                    891.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    888.0,
+                    1024.0,
+                    888.0,
+                    1024.0,
+                    921.0,
+                    296.0,
+                    921.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1054.0,
+                    888.0,
+                    1405.0,
+                    888.0,
+                    1405.0,
+                    921.0,
+                    1054.0,
+                    921.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    915.0,
+                    1403.0,
+                    915.0,
+                    1403.0,
+                    952.0,
+                    294.0,
+                    952.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    947.0,
+                    1406.0,
+                    947.0,
+                    1406.0,
+                    984.0,
+                    295.0,
+                    984.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    976.0,
+                    1205.0,
+                    976.0,
+                    1205.0,
+                    1014.0,
+                    295.0,
+                    1014.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    686.0,
+                    1405.0,
+                    686.0,
+                    1405.0,
+                    723.0,
+                    295.0,
+                    723.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    718.0,
+                    1406.0,
+                    718.0,
+                    1406.0,
+                    752.0,
+                    291.0,
+                    752.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    745.0,
+                    1405.0,
+                    745.0,
+                    1405.0,
+                    788.0,
+                    294.0,
+                    788.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    779.0,
+                    1406.0,
+                    779.0,
+                    1406.0,
+                    816.0,
+                    294.0,
+                    816.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    809.0,
+                    388.0,
+                    809.0,
+                    388.0,
+                    843.0,
+                    294.0,
+                    843.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1319.0,
+                    370.0,
+                    1319.0,
+                    370.0,
+                    1358.0,
+                    292.0,
+                    1358.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    397.0,
+                    1319.0,
+                    1004.0,
+                    1319.0,
+                    1004.0,
+                    1358.0,
+                    397.0,
+                    1358.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1050.0,
+                    1319.0,
+                    1107.0,
+                    1319.0,
+                    1107.0,
+                    1358.0,
+                    1050.0,
+                    1358.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1135.0,
+                    1319.0,
+                    1404.0,
+                    1319.0,
+                    1404.0,
+                    1358.0,
+                    1135.0,
+                    1358.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1354.0,
+                    836.0,
+                    1354.0,
+                    836.0,
+                    1387.0,
+                    295.0,
+                    1387.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    853.0,
+                    1354.0,
+                    1404.0,
+                    1354.0,
+                    1404.0,
+                    1387.0,
+                    853.0,
+                    1387.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1384.0,
+                    715.0,
+                    1384.0,
+                    715.0,
+                    1417.0,
+                    295.0,
+                    1417.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1020.0,
+                    1009.0,
+                    1020.0,
+                    1009.0,
+                    1064.0,
+                    294.0,
+                    1064.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1050.0,
+                    1020.0,
+                    1404.0,
+                    1020.0,
+                    1404.0,
+                    1064.0,
+                    1050.0,
+                    1064.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1055.0,
+                    1405.0,
+                    1055.0,
+                    1405.0,
+                    1092.0,
+                    294.0,
+                    1092.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1085.0,
+                    557.0,
+                    1085.0,
+                    557.0,
+                    1123.0,
+                    294.0,
+                    1123.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1480.0,
+                    437.0,
+                    1480.0,
+                    437.0,
+                    1516.0,
+                    295.0,
+                    1516.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    454.0,
+                    1480.0,
+                    1403.0,
+                    1480.0,
+                    1403.0,
+                    1516.0,
+                    454.0,
+                    1516.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1511.0,
+                    638.0,
+                    1511.0,
+                    638.0,
+                    1545.0,
+                    296.0,
+                    1545.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    1803.0,
+                    1038.0,
+                    1803.0,
+                    1038.0,
+                    1837.0,
+                    297.0,
+                    1837.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1834.0,
+                    424.0,
+                    1834.0,
+                    424.0,
+                    1870.0,
+                    293.0,
+                    1870.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    465.0,
+                    1834.0,
+                    1118.0,
+                    1834.0,
+                    1118.0,
+                    1870.0,
+                    465.0,
+                    1870.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1865.0,
+                    993.0,
+                    1865.0,
+                    993.0,
+                    1899.0,
+                    293.0,
+                    1899.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1895.0,
+                    1116.0,
+                    1895.0,
+                    1116.0,
+                    1928.0,
+                    293.0,
+                    1928.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1926.0,
+                    607.0,
+                    1926.0,
+                    607.0,
+                    1960.0,
+                    295.0,
+                    1960.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    631.0,
+                    1926.0,
+                    961.0,
+                    1926.0,
+                    961.0,
+                    1960.0,
+                    631.0,
+                    1960.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1968.0,
+                    1403.0,
+                    1968.0,
+                    1403.0,
+                    2009.0,
+                    292.0,
+                    2009.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    2000.0,
+                    568.0,
+                    2000.0,
+                    568.0,
+                    2036.0,
+                    295.0,
+                    2036.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    823.0,
+                    237.0,
+                    1330.0,
+                    237.0,
+                    1330.0,
+                    273.0,
+                    823.0,
+                    273.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    825.0,
+                    277.0,
+                    907.0,
+                    277.0,
+                    907.0,
+                    312.0,
+                    825.0,
+                    312.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    942.0,
+                    277.0,
+                    1162.0,
+                    277.0,
+                    1162.0,
+                    312.0,
+                    942.0,
+                    312.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1235.0,
+                    277.0,
+                    1292.0,
+                    277.0,
+                    1292.0,
+                    312.0,
+                    1235.0,
+                    312.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    904.0,
+                    307.0,
+                    1007.0,
+                    307.0,
+                    1007.0,
+                    341.0,
+                    904.0,
+                    341.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1035.0,
+                    307.0,
+                    1262.0,
+                    307.0,
+                    1262.0,
+                    341.0,
+                    1035.0,
+                    341.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    822.0,
+                    336.0,
+                    927.0,
+                    336.0,
+                    927.0,
+                    373.0,
+                    822.0,
+                    373.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    954.0,
+                    336.0,
+                    1317.0,
+                    336.0,
+                    1317.0,
+                    373.0,
+                    954.0,
+                    373.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    823.0,
+                    398.0,
+                    1121.0,
+                    398.0,
+                    1121.0,
+                    431.0,
+                    823.0,
+                    431.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1140.0,
+                    398.0,
+                    1169.0,
+                    398.0,
+                    1169.0,
+                    431.0,
+                    1140.0,
+                    431.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1196.0,
+                    398.0,
+                    1233.0,
+                    398.0,
+                    1233.0,
+                    431.0,
+                    1196.0,
+                    431.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    863.0,
+                    429.0,
+                    1013.0,
+                    429.0,
+                    1013.0,
+                    462.0,
+                    863.0,
+                    462.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    860.0,
+                    484.0,
+                    1042.0,
+                    484.0,
+                    1042.0,
+                    525.0,
+                    860.0,
+                    525.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    862.0,
+                    544.0,
+                    1042.0,
+                    544.0,
+                    1042.0,
+                    596.0,
+                    862.0,
+                    596.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    820.0,
+                    365.0,
+                    1109.0,
+                    365.0,
+                    1109.0,
+                    401.0,
+                    820.0,
+                    401.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    858.0,
+                    514.5,
+                    1111.0,
+                    514.5,
+                    1111.0,
+                    562.5,
+                    858.0,
+                    562.5
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 4,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    860,
+                    1405,
+                    860,
+                    1405,
+                    1167,
+                    297,
+                    1167
+                ],
+                "score": 0.982
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    509,
+                    1405,
+                    509,
+                    1405,
+                    845,
+                    297,
+                    845
+                ],
+                "score": 0.982
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1644,
+                    1405,
+                    1644,
+                    1405,
+                    1921,
+                    298,
+                    1921
+                ],
+                "score": 0.979
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1415,
+                    1405,
+                    1415,
+                    1405,
+                    1630,
+                    298,
+                    1630
+                ],
+                "score": 0.978
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    229,
+                    1403,
+                    229,
+                    1403,
+                    413,
+                    299,
+                    413
+                ],
+                "score": 0.976
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    330,
+                    1945,
+                    1343,
+                    1945,
+                    1343,
+                    2034,
+                    330,
+                    2034
+                ],
+                "score": 0.95
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    1182,
+                    1400,
+                    1182,
+                    1400,
+                    1243,
+                    299,
+                    1243
+                ],
+                "score": 0.948
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    301,
+                    451,
+                    691,
+                    451,
+                    691,
+                    483,
+                    301,
+                    483
+                ],
+                "score": 0.896
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    299,
+                    1358,
+                    489,
+                    1358,
+                    489,
+                    1390,
+                    299,
+                    1390
+                ],
+                "score": 0.894
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    301,
+                    1289,
+                    557,
+                    1289,
+                    557,
+                    1324,
+                    301,
+                    1324
+                ],
+                "score": 0.891
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    298,
+                    75,
+                    857,
+                    75,
+                    857,
+                    104,
+                    298,
+                    104
+                ],
+                "score": 0.88
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    840,
+                    2089,
+                    858,
+                    2089,
+                    858,
+                    2112,
+                    840,
+                    2112
+                ],
+                "score": 0.791
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    695,
+                    953,
+                    727,
+                    953,
+                    727,
+                    984,
+                    695,
+                    984
+                ],
+                "score": 0.88,
+                "latex": "\\beta _ { 1 }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    779,
+                    953,
+                    810,
+                    953,
+                    810,
+                    984,
+                    779,
+                    984
+                ],
+                "score": 0.88,
+                "latex": "\\beta _ { 2 }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    582,
+                    1859,
+                    637,
+                    1859,
+                    637,
+                    1888,
+                    582,
+                    1888
+                ],
+                "score": 0.87,
+                "latex": "80 \\%"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1290,
+                    1105,
+                    1331,
+                    1105,
+                    1331,
+                    1133,
+                    1290,
+                    1133
+                ],
+                "score": 0.86,
+                "latex": "5 \\%"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    621,
+                    782,
+                    642,
+                    782,
+                    642,
+                    820,
+                    621,
+                    820
+                ],
+                "score": 0.85,
+                "latex": "\\frac { 1 } { 8 }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    970,
+                    696,
+                    992,
+                    696,
+                    992,
+                    721,
+                    970,
+                    721
+                ],
+                "score": 0.75,
+                "latex": "_ z"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    443,
+                    789,
+                    464,
+                    789,
+                    464,
+                    811,
+                    443,
+                    811
+                ],
+                "score": 0.64,
+                "latex": "_ { z }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    632,
+                    1508,
+                    789,
+                    1508,
+                    789,
+                    1537,
+                    632,
+                    1537
+                ],
+                "score": 0.43,
+                "latex": "6 0 , 0 0 0 \\ 3 2 \\mathrm { x } \\ 3 2"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1046,
+                    1949,
+                    1063,
+                    1949,
+                    1063,
+                    1970,
+                    1046,
+                    1970
+                ],
+                "score": 0.42,
+                "latex": "^ +"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1295,
+                    785,
+                    1329,
+                    785,
+                    1329,
+                    812,
+                    1295,
+                    812
+                ],
+                "score": 0.42,
+                "latex": "( N )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1180,
+                    1949,
+                    1197,
+                    1949,
+                    1197,
+                    1969,
+                    1180,
+                    1969
+                ],
+                "score": 0.33,
+                "latex": "^ +"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    330.0,
+                    1940.0,
+                    1045.0,
+                    1940.0,
+                    1045.0,
+                    1977.0,
+                    330.0,
+                    1977.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1064.0,
+                    1940.0,
+                    1179.0,
+                    1940.0,
+                    1179.0,
+                    1977.0,
+                    1064.0,
+                    1977.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1198.0,
+                    1940.0,
+                    1347.0,
+                    1940.0,
+                    1347.0,
+                    1977.0,
+                    1198.0,
+                    1977.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    328.0,
+                    1968.0,
+                    963.0,
+                    1968.0,
+                    963.0,
+                    2011.0,
+                    328.0,
+                    2011.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    327.0,
+                    1999.0,
+                    1021.0,
+                    1999.0,
+                    1021.0,
+                    2040.0,
+                    327.0,
+                    2040.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    452.0,
+                    695.0,
+                    452.0,
+                    695.0,
+                    486.0,
+                    295.0,
+                    486.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1355.0,
+                    494.0,
+                    1355.0,
+                    494.0,
+                    1395.0,
+                    294.0,
+                    1395.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1289.0,
+                    560.0,
+                    1289.0,
+                    560.0,
+                    1329.0,
+                    293.0,
+                    1329.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    72.0,
+                    858.0,
+                    72.0,
+                    858.0,
+                    108.0,
+                    297.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    840.0,
+                    2087.0,
+                    861.0,
+                    2087.0,
+                    861.0,
+                    2118.0,
+                    840.0,
+                    2118.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    861.0,
+                    1405.0,
+                    861.0,
+                    1405.0,
+                    896.0,
+                    295.0,
+                    896.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    891.0,
+                    1408.0,
+                    891.0,
+                    1408.0,
+                    927.0,
+                    295.0,
+                    927.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    919.0,
+                    1410.0,
+                    919.0,
+                    1410.0,
+                    959.0,
+                    292.0,
+                    959.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    949.0,
+                    694.0,
+                    949.0,
+                    694.0,
+                    989.0,
+                    291.0,
+                    989.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    728.0,
+                    949.0,
+                    778.0,
+                    949.0,
+                    778.0,
+                    989.0,
+                    728.0,
+                    989.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    811.0,
+                    949.0,
+                    1407.0,
+                    949.0,
+                    1407.0,
+                    989.0,
+                    811.0,
+                    989.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    983.0,
+                    1402.0,
+                    983.0,
+                    1402.0,
+                    1015.0,
+                    295.0,
+                    1015.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1014.0,
+                    1405.0,
+                    1014.0,
+                    1405.0,
+                    1049.0,
+                    295.0,
+                    1049.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1044.0,
+                    1406.0,
+                    1044.0,
+                    1406.0,
+                    1077.0,
+                    292.0,
+                    1077.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1075.0,
+                    1408.0,
+                    1075.0,
+                    1408.0,
+                    1110.0,
+                    294.0,
+                    1110.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1104.0,
+                    1289.0,
+                    1104.0,
+                    1289.0,
+                    1138.0,
+                    292.0,
+                    1138.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1332.0,
+                    1104.0,
+                    1406.0,
+                    1104.0,
+                    1406.0,
+                    1138.0,
+                    1332.0,
+                    1138.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1134.0,
+                    805.0,
+                    1134.0,
+                    805.0,
+                    1174.0,
+                    294.0,
+                    1174.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    510.0,
+                    1407.0,
+                    510.0,
+                    1407.0,
+                    544.0,
+                    295.0,
+                    544.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    539.0,
+                    1406.0,
+                    539.0,
+                    1406.0,
+                    577.0,
+                    292.0,
+                    577.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    568.0,
+                    1406.0,
+                    568.0,
+                    1406.0,
+                    609.0,
+                    291.0,
+                    609.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    599.0,
+                    1405.0,
+                    599.0,
+                    1405.0,
+                    636.0,
+                    292.0,
+                    636.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    630.0,
+                    1403.0,
+                    630.0,
+                    1403.0,
+                    665.0,
+                    294.0,
+                    665.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    662.0,
+                    1405.0,
+                    662.0,
+                    1405.0,
+                    697.0,
+                    294.0,
+                    697.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    693.0,
+                    969.0,
+                    693.0,
+                    969.0,
+                    727.0,
+                    294.0,
+                    727.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    993.0,
+                    693.0,
+                    1405.0,
+                    693.0,
+                    1405.0,
+                    727.0,
+                    993.0,
+                    727.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    723.0,
+                    1406.0,
+                    723.0,
+                    1406.0,
+                    758.0,
+                    295.0,
+                    758.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    754.0,
+                    1406.0,
+                    754.0,
+                    1406.0,
+                    788.0,
+                    295.0,
+                    788.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    782.0,
+                    442.0,
+                    782.0,
+                    442.0,
+                    821.0,
+                    291.0,
+                    821.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    465.0,
+                    782.0,
+                    620.0,
+                    782.0,
+                    620.0,
+                    821.0,
+                    465.0,
+                    821.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    643.0,
+                    782.0,
+                    1294.0,
+                    782.0,
+                    1294.0,
+                    821.0,
+                    643.0,
+                    821.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1330.0,
+                    782.0,
+                    1407.0,
+                    782.0,
+                    1407.0,
+                    821.0,
+                    1330.0,
+                    821.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    814.0,
+                    382.0,
+                    814.0,
+                    382.0,
+                    847.0,
+                    294.0,
+                    847.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    1645.0,
+                    1405.0,
+                    1645.0,
+                    1405.0,
+                    1678.0,
+                    297.0,
+                    1678.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1675.0,
+                    1406.0,
+                    1675.0,
+                    1406.0,
+                    1710.0,
+                    293.0,
+                    1710.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1705.0,
+                    1405.0,
+                    1705.0,
+                    1405.0,
+                    1742.0,
+                    295.0,
+                    1742.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1738.0,
+                    1405.0,
+                    1738.0,
+                    1405.0,
+                    1771.0,
+                    295.0,
+                    1771.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1765.0,
+                    1408.0,
+                    1765.0,
+                    1408.0,
+                    1803.0,
+                    292.0,
+                    1803.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1797.0,
+                    1407.0,
+                    1797.0,
+                    1407.0,
+                    1834.0,
+                    295.0,
+                    1834.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1827.0,
+                    1408.0,
+                    1827.0,
+                    1408.0,
+                    1865.0,
+                    292.0,
+                    1865.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1858.0,
+                    581.0,
+                    1858.0,
+                    581.0,
+                    1895.0,
+                    293.0,
+                    1895.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    638.0,
+                    1858.0,
+                    1406.0,
+                    1858.0,
+                    1406.0,
+                    1895.0,
+                    638.0,
+                    1895.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1892.0,
+                    1405.0,
+                    1892.0,
+                    1405.0,
+                    1921.0,
+                    296.0,
+                    1921.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1412.0,
+                    1406.0,
+                    1412.0,
+                    1406.0,
+                    1454.0,
+                    292.0,
+                    1454.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1444.0,
+                    1406.0,
+                    1444.0,
+                    1406.0,
+                    1479.0,
+                    292.0,
+                    1479.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1473.0,
+                    1410.0,
+                    1473.0,
+                    1410.0,
+                    1513.0,
+                    291.0,
+                    1513.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1505.0,
+                    631.0,
+                    1505.0,
+                    631.0,
+                    1542.0,
+                    293.0,
+                    1542.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    790.0,
+                    1505.0,
+                    1406.0,
+                    1505.0,
+                    1406.0,
+                    1542.0,
+                    790.0,
+                    1542.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1536.0,
+                    1406.0,
+                    1536.0,
+                    1406.0,
+                    1573.0,
+                    293.0,
+                    1573.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1569.0,
+                    1406.0,
+                    1569.0,
+                    1406.0,
+                    1604.0,
+                    295.0,
+                    1604.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1599.0,
+                    804.0,
+                    1599.0,
+                    804.0,
+                    1632.0,
+                    293.0,
+                    1632.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    231.0,
+                    1404.0,
+                    231.0,
+                    1404.0,
+                    263.0,
+                    295.0,
+                    263.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    261.0,
+                    1405.0,
+                    261.0,
+                    1405.0,
+                    297.0,
+                    294.0,
+                    297.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    292.0,
+                    1405.0,
+                    292.0,
+                    1405.0,
+                    327.0,
+                    294.0,
+                    327.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    321.0,
+                    1403.0,
+                    321.0,
+                    1403.0,
+                    357.0,
+                    294.0,
+                    357.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    354.0,
+                    1404.0,
+                    354.0,
+                    1404.0,
+                    386.0,
+                    295.0,
+                    386.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    384.0,
+                    520.0,
+                    384.0,
+                    520.0,
+                    417.0,
+                    295.0,
+                    417.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1180.0,
+                    1405.0,
+                    1180.0,
+                    1405.0,
+                    1219.0,
+                    294.0,
+                    1219.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    1214.0,
+                    716.0,
+                    1214.0,
+                    716.0,
+                    1245.0,
+                    297.0,
+                    1245.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 5,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1667,
+                    1406,
+                    1667,
+                    1406,
+                    2034,
+                    298,
+                    2034
+                ],
+                "score": 0.983
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    642,
+                    1405,
+                    642,
+                    1405,
+                    949,
+                    298,
+                    949
+                ],
+                "score": 0.983
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1346,
+                    1404,
+                    1346,
+                    1404,
+                    1651,
+                    298,
+                    1651
+                ],
+                "score": 0.983
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    964,
+                    1403,
+                    964,
+                    1403,
+                    1148,
+                    298,
+                    1148
+                ],
+                "score": 0.98
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    229,
+                    1404,
+                    229,
+                    1404,
+                    445,
+                    298,
+                    445
+                ],
+                "score": 0.98
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    299,
+                    1259,
+                    662,
+                    1259,
+                    662,
+                    1291,
+                    299,
+                    1291
+                ],
+                "score": 0.894
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    298,
+                    75,
+                    857,
+                    75,
+                    857,
+                    104,
+                    298,
+                    104
+                ],
+                "score": 0.871
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    300,
+                    556,
+                    704,
+                    556,
+                    704,
+                    587,
+                    300,
+                    587
+                ],
+                "score": 0.869
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    842,
+                    2088,
+                    857,
+                    2088,
+                    857,
+                    2111,
+                    842,
+                    2111
+                ],
+                "score": 0.736
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    326,
+                    353,
+                    368,
+                    353,
+                    368,
+                    382,
+                    326,
+                    382
+                ],
+                "score": 0.86,
+                "latex": "5 \\%"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    591,
+                    857,
+                    645,
+                    857,
+                    645,
+                    886,
+                    591,
+                    886
+                ],
+                "score": 0.86,
+                "latex": "50 \\%"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    800,
+                    888,
+                    854,
+                    888,
+                    854,
+                    917,
+                    800,
+                    917
+                ],
+                "score": 0.86,
+                "latex": "50 \\%"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1259.0,
+                    665.0,
+                    1259.0,
+                    665.0,
+                    1295.0,
+                    294.0,
+                    1295.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    72.0,
+                    858.0,
+                    72.0,
+                    858.0,
+                    108.0,
+                    297.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    554.0,
+                    708.0,
+                    554.0,
+                    708.0,
+                    590.0,
+                    295.0,
+                    590.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    839.0,
+                    2085.0,
+                    861.0,
+                    2085.0,
+                    861.0,
+                    2120.0,
+                    839.0,
+                    2120.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1668.0,
+                    1407.0,
+                    1668.0,
+                    1407.0,
+                    1702.0,
+                    295.0,
+                    1702.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1698.0,
+                    1406.0,
+                    1698.0,
+                    1406.0,
+                    1733.0,
+                    295.0,
+                    1733.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1729.0,
+                    1408.0,
+                    1729.0,
+                    1408.0,
+                    1763.0,
+                    295.0,
+                    1763.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1758.0,
+                    1408.0,
+                    1758.0,
+                    1408.0,
+                    1795.0,
+                    293.0,
+                    1795.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    289.0,
+                    1787.0,
+                    1408.0,
+                    1787.0,
+                    1408.0,
+                    1825.0,
+                    289.0,
+                    1825.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1821.0,
+                    1407.0,
+                    1821.0,
+                    1407.0,
+                    1855.0,
+                    292.0,
+                    1855.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1851.0,
+                    1408.0,
+                    1851.0,
+                    1408.0,
+                    1885.0,
+                    292.0,
+                    1885.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1878.0,
+                    1407.0,
+                    1878.0,
+                    1407.0,
+                    1919.0,
+                    292.0,
+                    1919.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1913.0,
+                    1404.0,
+                    1913.0,
+                    1404.0,
+                    1947.0,
+                    295.0,
+                    1947.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1943.0,
+                    1406.0,
+                    1943.0,
+                    1406.0,
+                    1977.0,
+                    293.0,
+                    1977.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1972.0,
+                    1406.0,
+                    1972.0,
+                    1406.0,
+                    2010.0,
+                    293.0,
+                    2010.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    2003.0,
+                    700.0,
+                    2003.0,
+                    700.0,
+                    2037.0,
+                    295.0,
+                    2037.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    642.0,
+                    1410.0,
+                    642.0,
+                    1410.0,
+                    679.0,
+                    293.0,
+                    679.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    675.0,
+                    1403.0,
+                    675.0,
+                    1403.0,
+                    706.0,
+                    295.0,
+                    706.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    704.0,
+                    1405.0,
+                    704.0,
+                    1405.0,
+                    738.0,
+                    292.0,
+                    738.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    734.0,
+                    1406.0,
+                    734.0,
+                    1406.0,
+                    770.0,
+                    295.0,
+                    770.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    765.0,
+                    1407.0,
+                    765.0,
+                    1407.0,
+                    799.0,
+                    293.0,
+                    799.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    796.0,
+                    1406.0,
+                    796.0,
+                    1406.0,
+                    831.0,
+                    295.0,
+                    831.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    825.0,
+                    1408.0,
+                    825.0,
+                    1408.0,
+                    862.0,
+                    293.0,
+                    862.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    857.0,
+                    590.0,
+                    857.0,
+                    590.0,
+                    892.0,
+                    293.0,
+                    892.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    646.0,
+                    857.0,
+                    1405.0,
+                    857.0,
+                    1405.0,
+                    892.0,
+                    646.0,
+                    892.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    886.0,
+                    799.0,
+                    886.0,
+                    799.0,
+                    924.0,
+                    293.0,
+                    924.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    855.0,
+                    886.0,
+                    1405.0,
+                    886.0,
+                    1405.0,
+                    924.0,
+                    855.0,
+                    924.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    914.0,
+                    834.0,
+                    914.0,
+                    834.0,
+                    956.0,
+                    292.0,
+                    956.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1343.0,
+                    1406.0,
+                    1343.0,
+                    1406.0,
+                    1386.0,
+                    291.0,
+                    1386.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1377.0,
+                    1406.0,
+                    1377.0,
+                    1406.0,
+                    1412.0,
+                    294.0,
+                    1412.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1402.0,
+                    1408.0,
+                    1402.0,
+                    1408.0,
+                    1447.0,
+                    292.0,
+                    1447.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1439.0,
+                    1404.0,
+                    1439.0,
+                    1404.0,
+                    1472.0,
+                    295.0,
+                    1472.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1467.0,
+                    1405.0,
+                    1467.0,
+                    1405.0,
+                    1504.0,
+                    295.0,
+                    1504.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1500.0,
+                    1406.0,
+                    1500.0,
+                    1406.0,
+                    1534.0,
+                    293.0,
+                    1534.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1531.0,
+                    1405.0,
+                    1531.0,
+                    1405.0,
+                    1563.0,
+                    294.0,
+                    1563.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1561.0,
+                    1409.0,
+                    1561.0,
+                    1409.0,
+                    1595.0,
+                    292.0,
+                    1595.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1590.0,
+                    1406.0,
+                    1590.0,
+                    1406.0,
+                    1625.0,
+                    292.0,
+                    1625.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1621.0,
+                    534.0,
+                    1621.0,
+                    534.0,
+                    1653.0,
+                    294.0,
+                    1653.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    962.0,
+                    1404.0,
+                    962.0,
+                    1404.0,
+                    998.0,
+                    294.0,
+                    998.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    995.0,
+                    1403.0,
+                    995.0,
+                    1403.0,
+                    1027.0,
+                    296.0,
+                    1027.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1027.0,
+                    1404.0,
+                    1027.0,
+                    1404.0,
+                    1059.0,
+                    296.0,
+                    1059.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1055.0,
+                    1405.0,
+                    1055.0,
+                    1405.0,
+                    1090.0,
+                    294.0,
+                    1090.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1088.0,
+                    1405.0,
+                    1088.0,
+                    1405.0,
+                    1119.0,
+                    294.0,
+                    1119.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1117.0,
+                    1227.0,
+                    1117.0,
+                    1227.0,
+                    1151.0,
+                    293.0,
+                    1151.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    228.0,
+                    1404.0,
+                    228.0,
+                    1404.0,
+                    265.0,
+                    295.0,
+                    265.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    262.0,
+                    1408.0,
+                    262.0,
+                    1408.0,
+                    296.0,
+                    294.0,
+                    296.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    290.0,
+                    1406.0,
+                    290.0,
+                    1406.0,
+                    327.0,
+                    293.0,
+                    327.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    322.0,
+                    1406.0,
+                    322.0,
+                    1406.0,
+                    357.0,
+                    293.0,
+                    357.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    353.0,
+                    325.0,
+                    353.0,
+                    325.0,
+                    388.0,
+                    294.0,
+                    388.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    369.0,
+                    353.0,
+                    1404.0,
+                    353.0,
+                    1404.0,
+                    388.0,
+                    369.0,
+                    388.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    384.0,
+                    1404.0,
+                    384.0,
+                    1404.0,
+                    416.0,
+                    293.0,
+                    416.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    415.0,
+                    1046.0,
+                    415.0,
+                    1046.0,
+                    448.0,
+                    293.0,
+                    448.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 6,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    910,
+                    1405,
+                    910,
+                    1405,
+                    1246,
+                    297,
+                    1246
+                ],
+                "score": 0.98
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1431,
+                    1403,
+                    1431,
+                    1403,
+                    1617,
+                    297,
+                    1617
+                ],
+                "score": 0.976
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1262,
+                    1404,
+                    1262,
+                    1404,
+                    1416,
+                    298,
+                    1416
+                ],
+                "score": 0.976
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1710,
+                    1404,
+                    1710,
+                    1404,
+                    1835,
+                    298,
+                    1835
+                ],
+                "score": 0.974
+            },
+            {
+                "category_id": 4,
+                "poly": [
+                    297,
+                    699,
+                    1405,
+                    699,
+                    1405,
+                    794,
+                    297,
+                    794
+                ],
+                "score": 0.959
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    299,
+                    1858,
+                    1406,
+                    1858,
+                    1406,
+                    2034,
+                    299,
+                    2034
+                ],
+                "score": 0.952
+            },
+            {
+                "category_id": 3,
+                "poly": [
+                    305,
+                    231,
+                    1391,
+                    231,
+                    1391,
+                    674,
+                    305,
+                    674
+                ],
+                "score": 0.95
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    298,
+                    1654,
+                    885,
+                    1654,
+                    885,
+                    1685,
+                    298,
+                    1685
+                ],
+                "score": 0.91
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    299,
+                    854,
+                    473,
+                    854,
+                    473,
+                    884,
+                    299,
+                    884
+                ],
+                "score": 0.825
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    840,
+                    2088,
+                    858,
+                    2088,
+                    858,
+                    2112,
+                    840,
+                    2112
+                ],
+                "score": 0.795
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    298,
+                    76,
+                    858,
+                    76,
+                    858,
+                    104,
+                    298,
+                    104
+                ],
+                "score": 0.791
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    300,
+                    76,
+                    855,
+                    76,
+                    855,
+                    104,
+                    300,
+                    104
+                ],
+                "score": 0.38
+            },
+            {
+                "category_id": 3,
+                "poly": [
+                    306,
+                    230,
+                    1391,
+                    230,
+                    1391,
+                    675,
+                    306,
+                    675
+                ],
+                "score": 0.359
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    728,
+                    1463,
+                    783,
+                    1463,
+                    783,
+                    1492,
+                    728,
+                    1492
+                ],
+                "score": 0.88,
+                "latex": "40 \\%"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1347,
+                    1323,
+                    1403,
+                    1323,
+                    1403,
+                    1353,
+                    1347,
+                    1353
+                ],
+                "score": 0.85,
+                "latex": "30 \\%"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1084,
+                    1385,
+                    1138,
+                    1385,
+                    1138,
+                    1413,
+                    1084,
+                    1413
+                ],
+                "score": 0.85,
+                "latex": "20 \\%"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    698.0,
+                    1406.0,
+                    698.0,
+                    1406.0,
+                    736.0,
+                    295.0,
+                    736.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    731.0,
+                    1403.0,
+                    731.0,
+                    1403.0,
+                    765.0,
+                    295.0,
+                    765.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    760.0,
+                    723.0,
+                    760.0,
+                    723.0,
+                    797.0,
+                    295.0,
+                    797.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    328.0,
+                    1855.0,
+                    1313.0,
+                    1855.0,
+                    1313.0,
+                    1893.0,
+                    328.0,
+                    1893.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    328.0,
+                    1884.0,
+                    844.0,
+                    1884.0,
+                    844.0,
+                    1925.0,
+                    328.0,
+                    1925.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    330.0,
+                    1914.0,
+                    858.0,
+                    1914.0,
+                    858.0,
+                    1953.0,
+                    330.0,
+                    1953.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    328.0,
+                    1942.0,
+                    1407.0,
+                    1942.0,
+                    1407.0,
+                    1987.0,
+                    328.0,
+                    1987.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1976.0,
+                    1406.0,
+                    1976.0,
+                    1406.0,
+                    2011.0,
+                    293.0,
+                    2011.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    2004.0,
+                    632.0,
+                    2004.0,
+                    632.0,
+                    2038.0,
+                    294.0,
+                    2038.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    502.0,
+                    229.0,
+                    683.0,
+                    229.0,
+                    683.0,
+                    256.0,
+                    502.0,
+                    256.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1072.0,
+                    229.0,
+                    1231.0,
+                    229.0,
+                    1231.0,
+                    256.0,
+                    1072.0,
+                    256.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    367.0,
+                    257.0,
+                    427.0,
+                    257.0,
+                    427.0,
+                    286.0,
+                    367.0,
+                    286.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    273.0,
+                    365.0,
+                    273.0,
+                    365.0,
+                    300.0,
+                    322.0,
+                    300.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    396.0,
+                    277.0,
+                    456.0,
+                    277.0,
+                    456.0,
+                    301.0,
+                    396.0,
+                    301.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    883.0,
+                    279.0,
+                    923.0,
+                    279.0,
+                    923.0,
+                    305.0,
+                    883.0,
+                    305.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    957.0,
+                    277.0,
+                    1016.0,
+                    277.0,
+                    1016.0,
+                    301.0,
+                    957.0,
+                    301.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    382.0,
+                    295.0,
+                    439.0,
+                    295.0,
+                    439.0,
+                    317.0,
+                    382.0,
+                    317.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    957.0,
+                    295.0,
+                    1000.0,
+                    295.0,
+                    1000.0,
+                    317.0,
+                    957.0,
+                    317.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    376.0,
+                    311.0,
+                    438.0,
+                    311.0,
+                    438.0,
+                    336.0,
+                    376.0,
+                    336.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    939.0,
+                    319.0,
+                    949.0,
+                    319.0,
+                    949.0,
+                    329.0,
+                    939.0,
+                    329.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    955.0,
+                    311.0,
+                    997.0,
+                    311.0,
+                    997.0,
+                    336.0,
+                    955.0,
+                    336.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    329.0,
+                    364.0,
+                    329.0,
+                    364.0,
+                    354.0,
+                    323.0,
+                    354.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    884.0,
+                    348.0,
+                    923.0,
+                    348.0,
+                    923.0,
+                    374.0,
+                    884.0,
+                    374.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    304.0,
+                    377.0,
+                    360.0,
+                    377.0,
+                    360.0,
+                    461.0,
+                    304.0,
+                    461.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    864.0,
+                    377.0,
+                    892.0,
+                    377.0,
+                    892.0,
+                    460.0,
+                    864.0,
+                    460.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    886.0,
+                    420.0,
+                    920.0,
+                    420.0,
+                    920.0,
+                    442.0,
+                    886.0,
+                    442.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    442.0,
+                    362.0,
+                    442.0,
+                    362.0,
+                    467.0,
+                    323.0,
+                    467.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    891.0,
+                    489.0,
+                    925.0,
+                    489.0,
+                    925.0,
+                    512.0,
+                    891.0,
+                    512.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    332.0,
+                    500.0,
+                    361.0,
+                    500.0,
+                    361.0,
+                    523.0,
+                    332.0,
+                    523.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    658.0,
+                    501.0,
+                    668.0,
+                    501.0,
+                    668.0,
+                    509.0,
+                    658.0,
+                    509.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    339.0,
+                    557.0,
+                    359.0,
+                    557.0,
+                    359.0,
+                    577.0,
+                    339.0,
+                    577.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    376.0,
+                    560.0,
+                    387.0,
+                    560.0,
+                    387.0,
+                    571.0,
+                    376.0,
+                    571.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    901.0,
+                    560.0,
+                    921.0,
+                    560.0,
+                    921.0,
+                    580.0,
+                    901.0,
+                    580.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1074.0,
+                    552.0,
+                    1089.0,
+                    552.0,
+                    1089.0,
+                    566.0,
+                    1074.0,
+                    566.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    366.0,
+                    580.0,
+                    398.0,
+                    580.0,
+                    398.0,
+                    608.0,
+                    366.0,
+                    608.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    438.0,
+                    580.0,
+                    466.0,
+                    580.0,
+                    466.0,
+                    605.0,
+                    438.0,
+                    605.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    507.0,
+                    579.0,
+                    539.0,
+                    579.0,
+                    539.0,
+                    606.0,
+                    507.0,
+                    606.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    579.0,
+                    581.0,
+                    607.0,
+                    581.0,
+                    607.0,
+                    605.0,
+                    579.0,
+                    605.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    649.0,
+                    580.0,
+                    676.0,
+                    580.0,
+                    676.0,
+                    605.0,
+                    649.0,
+                    605.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    720.0,
+                    580.0,
+                    748.0,
+                    580.0,
+                    748.0,
+                    605.0,
+                    720.0,
+                    605.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    789.0,
+                    581.0,
+                    817.0,
+                    581.0,
+                    817.0,
+                    605.0,
+                    789.0,
+                    605.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    927.0,
+                    581.0,
+                    957.0,
+                    581.0,
+                    957.0,
+                    605.0,
+                    927.0,
+                    605.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    996.0,
+                    580.0,
+                    1028.0,
+                    580.0,
+                    1028.0,
+                    608.0,
+                    996.0,
+                    608.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1068.0,
+                    581.0,
+                    1096.0,
+                    581.0,
+                    1096.0,
+                    605.0,
+                    1068.0,
+                    605.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1137.0,
+                    577.0,
+                    1168.0,
+                    577.0,
+                    1168.0,
+                    606.0,
+                    1137.0,
+                    606.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1209.0,
+                    581.0,
+                    1237.0,
+                    581.0,
+                    1237.0,
+                    605.0,
+                    1209.0,
+                    605.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1279.0,
+                    581.0,
+                    1308.0,
+                    581.0,
+                    1308.0,
+                    605.0,
+                    1279.0,
+                    605.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1347.0,
+                    580.0,
+                    1379.0,
+                    580.0,
+                    1379.0,
+                    608.0,
+                    1347.0,
+                    608.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    527.0,
+                    598.0,
+                    658.0,
+                    598.0,
+                    658.0,
+                    625.0,
+                    527.0,
+                    625.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1086.0,
+                    598.0,
+                    1216.0,
+                    598.0,
+                    1216.0,
+                    625.0,
+                    1086.0,
+                    625.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    547.0,
+                    642.0,
+                    594.0,
+                    642.0,
+                    594.0,
+                    680.0,
+                    547.0,
+                    680.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1106.0,
+                    642.0,
+                    1154.0,
+                    642.0,
+                    1154.0,
+                    680.0,
+                    1106.0,
+                    680.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    926.0,
+                    258.0,
+                    987.0,
+                    258.0,
+                    987.0,
+                    287.0,
+                    926.0,
+                    287.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1650.0,
+                    889.0,
+                    1650.0,
+                    889.0,
+                    1690.0,
+                    293.0,
+                    1690.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    853.0,
+                    477.0,
+                    853.0,
+                    477.0,
+                    888.0,
+                    295.0,
+                    888.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    836.0,
+                    2085.0,
+                    859.0,
+                    2085.0,
+                    859.0,
+                    2116.0,
+                    836.0,
+                    2116.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    70.0,
+                    859.0,
+                    70.0,
+                    859.0,
+                    109.0,
+                    296.0,
+                    109.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    73.0,
+                    858.0,
+                    73.0,
+                    858.0,
+                    108.0,
+                    297.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    504.0,
+                    232.0,
+                    683.0,
+                    232.0,
+                    683.0,
+                    255.0,
+                    504.0,
+                    255.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1072.0,
+                    232.0,
+                    1230.0,
+                    232.0,
+                    1230.0,
+                    255.0,
+                    1072.0,
+                    255.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    375.0,
+                    260.0,
+                    425.0,
+                    260.0,
+                    425.0,
+                    283.0,
+                    375.0,
+                    283.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    798.0,
+                    268.0,
+                    808.0,
+                    268.0,
+                    808.0,
+                    277.0,
+                    798.0,
+                    277.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    930.0,
+                    259.0,
+                    984.0,
+                    259.0,
+                    984.0,
+                    285.0,
+                    930.0,
+                    285.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    274.0,
+                    366.0,
+                    274.0,
+                    366.0,
+                    299.0,
+                    323.0,
+                    299.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    397.0,
+                    276.0,
+                    456.0,
+                    276.0,
+                    456.0,
+                    300.0,
+                    397.0,
+                    300.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    882.0,
+                    279.0,
+                    923.0,
+                    279.0,
+                    923.0,
+                    304.0,
+                    882.0,
+                    304.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    934.0,
+                    282.0,
+                    952.0,
+                    282.0,
+                    952.0,
+                    296.0,
+                    934.0,
+                    296.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    956.0,
+                    276.0,
+                    1015.0,
+                    276.0,
+                    1015.0,
+                    300.0,
+                    956.0,
+                    300.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    382.0,
+                    293.0,
+                    441.0,
+                    293.0,
+                    441.0,
+                    319.0,
+                    382.0,
+                    319.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    941.0,
+                    302.0,
+                    955.0,
+                    302.0,
+                    955.0,
+                    310.0,
+                    941.0,
+                    310.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    957.0,
+                    296.0,
+                    1000.0,
+                    296.0,
+                    1000.0,
+                    317.0,
+                    957.0,
+                    317.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    375.0,
+                    310.0,
+                    437.0,
+                    310.0,
+                    437.0,
+                    336.0,
+                    375.0,
+                    336.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    932.0,
+                    310.0,
+                    997.0,
+                    310.0,
+                    997.0,
+                    336.0,
+                    932.0,
+                    336.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    328.0,
+                    363.0,
+                    328.0,
+                    363.0,
+                    355.0,
+                    323.0,
+                    355.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    883.0,
+                    348.0,
+                    923.0,
+                    348.0,
+                    923.0,
+                    373.0,
+                    883.0,
+                    373.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    304.0,
+                    377.0,
+                    361.0,
+                    377.0,
+                    361.0,
+                    461.0,
+                    304.0,
+                    461.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    862.0,
+                    376.0,
+                    894.0,
+                    376.0,
+                    894.0,
+                    462.0,
+                    862.0,
+                    462.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    886.0,
+                    420.0,
+                    920.0,
+                    420.0,
+                    920.0,
+                    442.0,
+                    886.0,
+                    442.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    442.0,
+                    363.0,
+                    442.0,
+                    363.0,
+                    467.0,
+                    323.0,
+                    467.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    377.0,
+                    460.0,
+                    386.0,
+                    460.0,
+                    386.0,
+                    470.0,
+                    377.0,
+                    470.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    892.0,
+                    489.0,
+                    925.0,
+                    489.0,
+                    925.0,
+                    512.0,
+                    892.0,
+                    512.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    331.0,
+                    499.0,
+                    362.0,
+                    499.0,
+                    362.0,
+                    522.0,
+                    331.0,
+                    522.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    658.0,
+                    500.0,
+                    668.0,
+                    500.0,
+                    668.0,
+                    510.0,
+                    658.0,
+                    510.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    340.0,
+                    557.0,
+                    360.0,
+                    557.0,
+                    360.0,
+                    577.0,
+                    340.0,
+                    577.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    377.0,
+                    560.0,
+                    388.0,
+                    560.0,
+                    388.0,
+                    572.0,
+                    377.0,
+                    572.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1077.0,
+                    555.0,
+                    1088.0,
+                    555.0,
+                    1088.0,
+                    566.0,
+                    1077.0,
+                    566.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    368.0,
+                    580.0,
+                    397.0,
+                    580.0,
+                    397.0,
+                    606.0,
+                    368.0,
+                    606.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    436.0,
+                    579.0,
+                    468.0,
+                    579.0,
+                    468.0,
+                    607.0,
+                    436.0,
+                    607.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    508.0,
+                    580.0,
+                    536.0,
+                    580.0,
+                    536.0,
+                    606.0,
+                    508.0,
+                    606.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    577.0,
+                    578.0,
+                    609.0,
+                    578.0,
+                    609.0,
+                    606.0,
+                    577.0,
+                    606.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    650.0,
+                    580.0,
+                    677.0,
+                    580.0,
+                    677.0,
+                    606.0,
+                    650.0,
+                    606.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    718.0,
+                    578.0,
+                    749.0,
+                    578.0,
+                    749.0,
+                    607.0,
+                    718.0,
+                    607.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    788.0,
+                    581.0,
+                    818.0,
+                    581.0,
+                    818.0,
+                    606.0,
+                    788.0,
+                    606.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    926.0,
+                    579.0,
+                    958.0,
+                    579.0,
+                    958.0,
+                    607.0,
+                    926.0,
+                    607.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    998.0,
+                    580.0,
+                    1026.0,
+                    580.0,
+                    1026.0,
+                    606.0,
+                    998.0,
+                    606.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1068.0,
+                    581.0,
+                    1097.0,
+                    581.0,
+                    1097.0,
+                    606.0,
+                    1068.0,
+                    606.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1137.0,
+                    577.0,
+                    1168.0,
+                    577.0,
+                    1168.0,
+                    607.0,
+                    1137.0,
+                    607.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1206.0,
+                    579.0,
+                    1239.0,
+                    579.0,
+                    1239.0,
+                    607.0,
+                    1206.0,
+                    607.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1279.0,
+                    580.0,
+                    1308.0,
+                    580.0,
+                    1308.0,
+                    606.0,
+                    1279.0,
+                    606.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1348.0,
+                    581.0,
+                    1378.0,
+                    581.0,
+                    1378.0,
+                    606.0,
+                    1348.0,
+                    606.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    526.0,
+                    596.0,
+                    657.0,
+                    596.0,
+                    657.0,
+                    627.0,
+                    526.0,
+                    627.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1086.0,
+                    596.0,
+                    1218.0,
+                    596.0,
+                    1218.0,
+                    627.0,
+                    1086.0,
+                    627.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    547.0,
+                    642.0,
+                    593.0,
+                    642.0,
+                    593.0,
+                    680.0,
+                    547.0,
+                    680.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1106.0,
+                    641.0,
+                    1153.0,
+                    641.0,
+                    1153.0,
+                    680.0,
+                    1106.0,
+                    680.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    908.0,
+                    1405.0,
+                    908.0,
+                    1405.0,
+                    948.0,
+                    294.0,
+                    948.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    940.0,
+                    1405.0,
+                    940.0,
+                    1405.0,
+                    978.0,
+                    294.0,
+                    978.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    974.0,
+                    1403.0,
+                    974.0,
+                    1403.0,
+                    1005.0,
+                    296.0,
+                    1005.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1002.0,
+                    1405.0,
+                    1002.0,
+                    1405.0,
+                    1037.0,
+                    294.0,
+                    1037.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1033.0,
+                    1406.0,
+                    1033.0,
+                    1406.0,
+                    1068.0,
+                    294.0,
+                    1068.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1062.0,
+                    1408.0,
+                    1062.0,
+                    1408.0,
+                    1099.0,
+                    294.0,
+                    1099.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1094.0,
+                    1406.0,
+                    1094.0,
+                    1406.0,
+                    1129.0,
+                    295.0,
+                    1129.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1125.0,
+                    1403.0,
+                    1125.0,
+                    1403.0,
+                    1158.0,
+                    295.0,
+                    1158.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1156.0,
+                    1405.0,
+                    1156.0,
+                    1405.0,
+                    1189.0,
+                    292.0,
+                    1189.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1185.0,
+                    1405.0,
+                    1185.0,
+                    1405.0,
+                    1220.0,
+                    292.0,
+                    1220.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1213.0,
+                    523.0,
+                    1213.0,
+                    523.0,
+                    1249.0,
+                    294.0,
+                    1249.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1430.0,
+                    1407.0,
+                    1430.0,
+                    1407.0,
+                    1469.0,
+                    293.0,
+                    1469.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1464.0,
+                    727.0,
+                    1464.0,
+                    727.0,
+                    1496.0,
+                    295.0,
+                    1496.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    784.0,
+                    1464.0,
+                    1405.0,
+                    1464.0,
+                    1405.0,
+                    1496.0,
+                    784.0,
+                    1496.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1492.0,
+                    1405.0,
+                    1492.0,
+                    1405.0,
+                    1528.0,
+                    295.0,
+                    1528.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1523.0,
+                    1405.0,
+                    1523.0,
+                    1405.0,
+                    1558.0,
+                    292.0,
+                    1558.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1553.0,
+                    1405.0,
+                    1553.0,
+                    1405.0,
+                    1589.0,
+                    293.0,
+                    1589.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1583.0,
+                    1136.0,
+                    1583.0,
+                    1136.0,
+                    1621.0,
+                    294.0,
+                    1621.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1264.0,
+                    1405.0,
+                    1264.0,
+                    1405.0,
+                    1297.0,
+                    296.0,
+                    1297.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1294.0,
+                    1406.0,
+                    1294.0,
+                    1406.0,
+                    1327.0,
+                    296.0,
+                    1327.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1324.0,
+                    1346.0,
+                    1324.0,
+                    1346.0,
+                    1358.0,
+                    296.0,
+                    1358.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1355.0,
+                    1404.0,
+                    1355.0,
+                    1404.0,
+                    1388.0,
+                    296.0,
+                    1388.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1384.0,
+                    1083.0,
+                    1384.0,
+                    1083.0,
+                    1421.0,
+                    294.0,
+                    1421.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1139.0,
+                    1384.0,
+                    1149.0,
+                    1384.0,
+                    1149.0,
+                    1421.0,
+                    1139.0,
+                    1421.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1710.0,
+                    1405.0,
+                    1710.0,
+                    1405.0,
+                    1747.0,
+                    293.0,
+                    1747.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1743.0,
+                    1406.0,
+                    1743.0,
+                    1406.0,
+                    1776.0,
+                    294.0,
+                    1776.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1773.0,
+                    1405.0,
+                    1773.0,
+                    1405.0,
+                    1805.0,
+                    294.0,
+                    1805.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1804.0,
+                    1406.0,
+                    1804.0,
+                    1406.0,
+                    1837.0,
+                    294.0,
+                    1837.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 7,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 5,
+                "poly": [
+                    300,
+                    780,
+                    1402,
+                    780,
+                    1402,
+                    1439,
+                    300,
+                    1439
+                ],
+                "score": 0.985,
+                "html": "<table><tr><td></td><td></td><td colspan=\"4\"> Accuracy at Missing Rate (%)a</td></tr><tr><td>Dataset</td><td>Method</td><td>10%</td><td>20%</td><td>30%</td><td>40%</td></tr><tr><td rowspan=\"5\">Landsat (Dua &amp; Graff, 2017)</td><td>GI</td><td>89.9 (±0.36)</td><td>89.6 (±0.36)</td><td>89.0 (±0.03)</td><td>88.0 (±0.22)</td></tr><tr><td>MisGAN</td><td>87.2 (±0.01)</td><td>85.7 (±0.19)</td><td>84.0 (±0.61)</td><td>82.9 (±0.75)</td></tr><tr><td>GAIN</td><td>89.7 (±0.42)</td><td>89.4 (±0.56)</td><td>88.4 (±0.71)</td><td>87.7 (±0.10)</td></tr><tr><td>DAE</td><td>89.4 (±0.10)</td><td>88.6 (±0.54)</td><td>87.5 (±0.14)</td><td>86.6 (±0.21)</td></tr><tr><td>MICE</td><td>89.5 (±0.16)</td><td>89.3 (±0.10)</td><td>88.1 (±0.49)</td><td>87.5 (±0.03)</td></tr><tr><td rowspan=\"5\">MIT-BIH(Moody &amp; Mark,2001)</td><td>GI</td><td>98.5 (±0.02)</td><td>98.4 (±0.03)</td><td>98.2 (±0.07)</td><td>97.7 (±0.03)</td></tr><tr><td>MisGAN</td><td>97.8 (±0.13)</td><td>97.4 (±0.07)</td><td>96.7 (±0.07)</td><td>96.2 (±0.09)</td></tr><tr><td>GAIN</td><td>98.5 (±0.02)</td><td>98.4 (±0.06)</td><td>98.0 (±0.09)</td><td>97.5 (±0.18)</td></tr><tr><td>DAE</td><td>98.4 (±0.02)</td><td>98.2 (±0.11)</td><td>97.9 (±0.09)</td><td>97.4 (±0.02)</td></tr><tr><td>MICE</td><td>98.4 (±0.01)</td><td>98.3 (±0.01)</td><td>98.1 (±0.01)</td><td>97.5 (±0.12)</td></tr><tr><td rowspan=\"5\">Diabetes (Kachuee et al., 2019)</td><td>GI</td><td>89.6 (±0.13)</td><td>89.0 (±0.03)</td><td>88.2 (±0.62)</td><td>86.8 (±0.38)</td></tr><tr><td>MisGAN</td><td>89.7 (±0.01)</td><td>88.9 (±0.30)</td><td>87.6 (±0.02)</td><td>86.4 (±0.68)</td></tr><tr><td>GAIN</td><td>89.2 (±0.09)</td><td>88.3 (±0.02)</td><td>86.9 (±0.09)</td><td>83.8 (±1.44)</td></tr><tr><td>DAE</td><td>89.3 (±0.05)</td><td>88.2 (±0.19)</td><td>86.9 (±0.09)</td><td>84.8 (±0.03)</td></tr><tr><td>MICE</td><td>89.8 (±0.08)</td><td>88.8 (±0.01)</td><td>88.0 (±0.08)</td><td>86.1 (±0.02)</td></tr></table>"
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1668,
+                    1405,
+                    1668,
+                    1405,
+                    2033,
+                    297,
+                    2033
+                ],
+                "score": 0.983
+            },
+            {
+                "category_id": 5,
+                "poly": [
+                    301,
+                    282,
+                    1398,
+                    282,
+                    1398,
+                    645,
+                    301,
+                    645
+                ],
+                "score": 0.983,
+                "html": "<table><tr><td rowspan=\"2\"></td><td colspan=\"5\">Accuracy at Missing Rate (%) MCAR Rect.</td><td rowspan=\"2\"></td></tr><tr><td colspan=\"2\">MCARUniform</td><td colspan=\"3\"></td></tr><tr><td>Method</td><td>20%</td><td>40%</td><td>60%</td><td>20%</td><td>40%</td><td>60%</td></tr><tr><td>GI</td><td>89.5 (±0.45)</td><td>87.1 (±0.54)</td><td>80.3 (±0.26)</td><td>84.0 (±0.03)</td><td>76.9 (±0.03)</td><td>66.1 (±0.16)</td></tr><tr><td>MisGAN</td><td>86.5 (±0.31)</td><td>83.7 (±0.40)</td><td>78.7 (±0.26)</td><td>82.9 (±0.44)</td><td>75.6 (±0.20)</td><td>65.0 (±0.31)</td></tr><tr><td>GAIN</td><td>88.7 (±0.45)</td><td>86.0 (±0.86)</td><td>81.8 (±0.03)</td><td>81.7 (±0.03)</td><td>73.6 (±0.35)</td><td>58.4 (±1.66)</td></tr><tr><td>DAE</td><td>88.0 (±0.22)</td><td>84.0 (±0.50)</td><td>79.8 (±0.71)</td><td>83.3 (±0.64)</td><td>75.5 (±0.44)</td><td>63.8 (±0.24)</td></tr><tr><td>Mean</td><td>85.7 (±0.02)</td><td>83.4 (±0.38)</td><td>79.2 (±0.16)</td><td>82.7 (±0.15)</td><td>75.3 (±0.16)</td><td>64.0 (±0.32)</td></tr></table>"
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1589,
+                    1403,
+                    1589,
+                    1403,
+                    1652,
+                    298,
+                    1652
+                ],
+                "score": 0.947
+            },
+            {
+                "category_id": 7,
+                "poly": [
+                    303,
+                    1456,
+                    1401,
+                    1456,
+                    1401,
+                    1512,
+                    303,
+                    1512
+                ],
+                "score": 0.935
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    298,
+                    75,
+                    858,
+                    75,
+                    858,
+                    105,
+                    298,
+                    105
+                ],
+                "score": 0.912
+            },
+            {
+                "category_id": 6,
+                "poly": [
+                    297,
+                    691,
+                    1402,
+                    691,
+                    1402,
+                    756,
+                    297,
+                    756
+                ],
+                "score": 0.908
+            },
+            {
+                "category_id": 6,
+                "poly": [
+                    342,
+                    224,
+                    1351,
+                    224,
+                    1351,
+                    258,
+                    342,
+                    258
+                ],
+                "score": 0.885
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    840,
+                    2088,
+                    859,
+                    2088,
+                    859,
+                    2111,
+                    840,
+                    2111
+                ],
+                "score": 0.796
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    340,
+                    1759,
+                    518,
+                    1759,
+                    518,
+                    1793,
+                    340,
+                    1793
+                ],
+                "score": 0.9,
+                "latex": "P ( x _ { 1 } | x _ { 2 } = 0 . 1 )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1087,
+                    1621,
+                    1142,
+                    1621,
+                    1142,
+                    1649,
+                    1087,
+                    1649
+                ],
+                "score": 0.86,
+                "latex": "50 \\%"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1296,
+                    1456,
+                    1365,
+                    1456,
+                    1365,
+                    1483,
+                    1296,
+                    1483
+                ],
+                "score": 0.86,
+                "latex": "9 0 . 7 \\%"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1095,
+                    1456,
+                    1164,
+                    1456,
+                    1164,
+                    1483,
+                    1095,
+                    1483
+                ],
+                "score": 0.83,
+                "latex": "9 0 . 9 \\%"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1173,
+                    1456,
+                    1242,
+                    1456,
+                    1242,
+                    1483,
+                    1173,
+                    1483
+                ],
+                "score": 0.82,
+                "latex": "9 8 . 6 \\%"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1055,
+                    302,
+                    1089,
+                    302,
+                    1089,
+                    330,
+                    1055,
+                    330
+                ],
+                "score": 0.28,
+                "latex": "( \\%"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    329.0,
+                    1451.0,
+                    1094.0,
+                    1451.0,
+                    1094.0,
+                    1488.0,
+                    329.0,
+                    1488.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1165.0,
+                    1451.0,
+                    1172.0,
+                    1451.0,
+                    1172.0,
+                    1488.0,
+                    1165.0,
+                    1488.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1243.0,
+                    1451.0,
+                    1295.0,
+                    1451.0,
+                    1295.0,
+                    1488.0,
+                    1243.0,
+                    1488.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1366.0,
+                    1451.0,
+                    1406.0,
+                    1451.0,
+                    1406.0,
+                    1488.0,
+                    1366.0,
+                    1488.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1480.0,
+                    760.0,
+                    1480.0,
+                    760.0,
+                    1515.0,
+                    296.0,
+                    1515.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    298.0,
+                    73.0,
+                    858.0,
+                    73.0,
+                    858.0,
+                    108.0,
+                    298.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    689.0,
+                    1407.0,
+                    689.0,
+                    1407.0,
+                    729.0,
+                    292.0,
+                    729.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    722.0,
+                    557.0,
+                    722.0,
+                    557.0,
+                    760.0,
+                    295.0,
+                    760.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    341.0,
+                    220.0,
+                    1356.0,
+                    220.0,
+                    1356.0,
+                    262.0,
+                    341.0,
+                    262.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    839.0,
+                    2087.0,
+                    861.0,
+                    2087.0,
+                    861.0,
+                    2117.0,
+                    839.0,
+                    2117.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1670.0,
+                    1406.0,
+                    1670.0,
+                    1406.0,
+                    1700.0,
+                    296.0,
+                    1700.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1699.0,
+                    1406.0,
+                    1699.0,
+                    1406.0,
+                    1732.0,
+                    291.0,
+                    1732.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1729.0,
+                    1405.0,
+                    1729.0,
+                    1405.0,
+                    1763.0,
+                    295.0,
+                    1763.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1758.0,
+                    339.0,
+                    1758.0,
+                    339.0,
+                    1795.0,
+                    294.0,
+                    1795.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    519.0,
+                    1758.0,
+                    1405.0,
+                    1758.0,
+                    1405.0,
+                    1795.0,
+                    519.0,
+                    1795.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1790.0,
+                    1405.0,
+                    1790.0,
+                    1405.0,
+                    1824.0,
+                    294.0,
+                    1824.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1819.0,
+                    1408.0,
+                    1819.0,
+                    1408.0,
+                    1857.0,
+                    294.0,
+                    1857.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1852.0,
+                    1406.0,
+                    1852.0,
+                    1406.0,
+                    1883.0,
+                    294.0,
+                    1883.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1882.0,
+                    1403.0,
+                    1882.0,
+                    1403.0,
+                    1916.0,
+                    295.0,
+                    1916.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1912.0,
+                    1406.0,
+                    1912.0,
+                    1406.0,
+                    1946.0,
+                    295.0,
+                    1946.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1941.0,
+                    1406.0,
+                    1941.0,
+                    1406.0,
+                    1976.0,
+                    292.0,
+                    1976.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1972.0,
+                    1405.0,
+                    1972.0,
+                    1405.0,
+                    2009.0,
+                    294.0,
+                    2009.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    2001.0,
+                    1407.0,
+                    2001.0,
+                    1407.0,
+                    2038.0,
+                    294.0,
+                    2038.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1590.0,
+                    1408.0,
+                    1590.0,
+                    1408.0,
+                    1624.0,
+                    294.0,
+                    1624.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1617.0,
+                    1086.0,
+                    1617.0,
+                    1086.0,
+                    1659.0,
+                    292.0,
+                    1659.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1143.0,
+                    1617.0,
+                    1353.0,
+                    1617.0,
+                    1353.0,
+                    1659.0,
+                    1143.0,
+                    1659.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 8,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1788,
+                    1405,
+                    1788,
+                    1405,
+                    2035,
+                    298,
+                    2035
+                ],
+                "score": 0.978
+            },
+            {
+                "category_id": 3,
+                "poly": [
+                    353,
+                    835,
+                    1335,
+                    835,
+                    1335,
+                    1480,
+                    353,
+                    1480
+                ],
+                "score": 0.974
+            },
+            {
+                "category_id": 3,
+                "poly": [
+                    365,
+                    237,
+                    1333,
+                    237,
+                    1333,
+                    546,
+                    365,
+                    546
+                ],
+                "score": 0.971
+            },
+            {
+                "category_id": 4,
+                "poly": [
+                    296,
+                    1515,
+                    1404,
+                    1515,
+                    1404,
+                    1609,
+                    296,
+                    1609
+                ],
+                "score": 0.956
+            },
+            {
+                "category_id": 4,
+                "poly": [
+                    298,
+                    579,
+                    1399,
+                    579,
+                    1399,
+                    644,
+                    298,
+                    644
+                ],
+                "score": 0.947
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    719,
+                    1403,
+                    719,
+                    1403,
+                    784,
+                    297,
+                    784
+                ],
+                "score": 0.946
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    300,
+                    1708,
+                    544,
+                    1708,
+                    544,
+                    1743,
+                    300,
+                    1743
+                ],
+                "score": 0.896
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    297,
+                    76,
+                    858,
+                    76,
+                    858,
+                    104,
+                    297,
+                    104
+                ],
+                "score": 0.888
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    836,
+                    2088,
+                    864,
+                    2088,
+                    864,
+                    2112,
+                    836,
+                    2112
+                ],
+                "score": 0.852
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    536,
+                    613,
+                    590,
+                    613,
+                    590,
+                    642,
+                    536,
+                    642
+                ],
+                "score": 0.88,
+                "latex": "40 \\%"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    537,
+                    1165,
+                    658,
+                    1165,
+                    658,
+                    1191,
+                    537,
+                    1191
+                ],
+                "score": 0.36,
+                "latex": "P ( x _ { 1 } | x _ { 2 } = 0 . 1 ;"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    462.0,
+                    837.0,
+                    642.0,
+                    837.0,
+                    642.0,
+                    873.0,
+                    462.0,
+                    873.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    789.0,
+                    839.0,
+                    965.0,
+                    839.0,
+                    965.0,
+                    871.0,
+                    789.0,
+                    871.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1099.0,
+                    838.0,
+                    1303.0,
+                    838.0,
+                    1303.0,
+                    873.0,
+                    1099.0,
+                    873.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    376.0,
+                    874.0,
+                    430.0,
+                    874.0,
+                    430.0,
+                    906.0,
+                    376.0,
+                    906.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    700.0,
+                    873.0,
+                    756.0,
+                    873.0,
+                    756.0,
+                    908.0,
+                    700.0,
+                    908.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1024.0,
+                    873.0,
+                    1082.0,
+                    873.0,
+                    1082.0,
+                    908.0,
+                    1024.0,
+                    908.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    376.0,
+                    918.0,
+                    433.0,
+                    918.0,
+                    433.0,
+                    950.0,
+                    376.0,
+                    950.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    594.0,
+                    887.0,
+                    660.0,
+                    887.0,
+                    660.0,
+                    946.0,
+                    594.0,
+                    946.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    701.0,
+                    918.0,
+                    757.0,
+                    918.0,
+                    757.0,
+                    950.0,
+                    701.0,
+                    950.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1024.0,
+                    917.0,
+                    1084.0,
+                    917.0,
+                    1084.0,
+                    952.0,
+                    1024.0,
+                    952.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    354.0,
+                    959.0,
+                    433.0,
+                    959.0,
+                    433.0,
+                    997.0,
+                    354.0,
+                    997.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    678.0,
+                    959.0,
+                    757.0,
+                    959.0,
+                    757.0,
+                    997.0,
+                    678.0,
+                    997.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1003.0,
+                    959.0,
+                    1082.0,
+                    959.0,
+                    1082.0,
+                    997.0,
+                    1003.0,
+                    997.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    376.0,
+                    1003.0,
+                    433.0,
+                    1003.0,
+                    433.0,
+                    1039.0,
+                    376.0,
+                    1039.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    700.0,
+                    1003.0,
+                    757.0,
+                    1003.0,
+                    757.0,
+                    1039.0,
+                    700.0,
+                    1039.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1024.0,
+                    1003.0,
+                    1083.0,
+                    1003.0,
+                    1083.0,
+                    1040.0,
+                    1024.0,
+                    1040.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    789.0,
+                    1036.0,
+                    817.0,
+                    1036.0,
+                    817.0,
+                    1052.0,
+                    789.0,
+                    1052.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    921.0,
+                    1034.0,
+                    989.0,
+                    1034.0,
+                    989.0,
+                    1052.0,
+                    921.0,
+                    1052.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    375.0,
+                    1048.0,
+                    431.0,
+                    1048.0,
+                    431.0,
+                    1083.0,
+                    375.0,
+                    1083.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    700.0,
+                    1048.0,
+                    757.0,
+                    1048.0,
+                    757.0,
+                    1083.0,
+                    700.0,
+                    1083.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1024.0,
+                    1048.0,
+                    1083.0,
+                    1048.0,
+                    1083.0,
+                    1083.0,
+                    1024.0,
+                    1083.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1216.0,
+                    1048.0,
+                    1247.0,
+                    1048.0,
+                    1247.0,
+                    1062.0,
+                    1216.0,
+                    1062.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    426.0,
+                    1082.0,
+                    679.0,
+                    1082.0,
+                    679.0,
+                    1114.0,
+                    426.0,
+                    1114.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    750.0,
+                    1082.0,
+                    1004.0,
+                    1082.0,
+                    1004.0,
+                    1114.0,
+                    750.0,
+                    1114.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1075.0,
+                    1082.0,
+                    1328.0,
+                    1082.0,
+                    1328.0,
+                    1114.0,
+                    1075.0,
+                    1114.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    528.0,
+                    1104.0,
+                    577.0,
+                    1104.0,
+                    577.0,
+                    1163.0,
+                    528.0,
+                    1163.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    859.0,
+                    1109.0,
+                    895.0,
+                    1109.0,
+                    895.0,
+                    1140.0,
+                    859.0,
+                    1140.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1183.0,
+                    1109.0,
+                    1219.0,
+                    1109.0,
+                    1219.0,
+                    1140.0,
+                    1183.0,
+                    1140.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    856.0,
+                    1125.0,
+                    897.0,
+                    1125.0,
+                    897.0,
+                    1160.0,
+                    856.0,
+                    1160.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1180.0,
+                    1125.0,
+                    1223.0,
+                    1125.0,
+                    1223.0,
+                    1160.0,
+                    1180.0,
+                    1160.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    436.0,
+                    1160.0,
+                    536.0,
+                    1160.0,
+                    536.0,
+                    1194.0,
+                    436.0,
+                    1194.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    659.0,
+                    1160.0,
+                    668.0,
+                    1160.0,
+                    668.0,
+                    1194.0,
+                    659.0,
+                    1194.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    763.0,
+                    1160.0,
+                    991.0,
+                    1160.0,
+                    991.0,
+                    1194.0,
+                    763.0,
+                    1194.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1105.0,
+                    1163.0,
+                    1299.0,
+                    1163.0,
+                    1299.0,
+                    1192.0,
+                    1105.0,
+                    1192.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    373.0,
+                    1195.0,
+                    432.0,
+                    1195.0,
+                    432.0,
+                    1227.0,
+                    373.0,
+                    1227.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    698.0,
+                    1195.0,
+                    757.0,
+                    1195.0,
+                    757.0,
+                    1227.0,
+                    698.0,
+                    1227.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1024.0,
+                    1194.0,
+                    1082.0,
+                    1194.0,
+                    1082.0,
+                    1230.0,
+                    1024.0,
+                    1230.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    375.0,
+                    1238.0,
+                    435.0,
+                    1238.0,
+                    435.0,
+                    1274.0,
+                    375.0,
+                    1274.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    700.0,
+                    1238.0,
+                    759.0,
+                    1238.0,
+                    759.0,
+                    1274.0,
+                    700.0,
+                    1274.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1024.0,
+                    1238.0,
+                    1084.0,
+                    1238.0,
+                    1084.0,
+                    1274.0,
+                    1024.0,
+                    1274.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    354.0,
+                    1281.0,
+                    435.0,
+                    1281.0,
+                    435.0,
+                    1319.0,
+                    354.0,
+                    1319.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    678.0,
+                    1281.0,
+                    758.0,
+                    1281.0,
+                    758.0,
+                    1319.0,
+                    678.0,
+                    1319.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1003.0,
+                    1281.0,
+                    1083.0,
+                    1281.0,
+                    1083.0,
+                    1319.0,
+                    1003.0,
+                    1319.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    375.0,
+                    1326.0,
+                    436.0,
+                    1326.0,
+                    436.0,
+                    1361.0,
+                    375.0,
+                    1361.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    700.0,
+                    1326.0,
+                    758.0,
+                    1326.0,
+                    758.0,
+                    1361.0,
+                    700.0,
+                    1361.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1024.0,
+                    1326.0,
+                    1083.0,
+                    1326.0,
+                    1083.0,
+                    1361.0,
+                    1024.0,
+                    1361.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1192.0,
+                    1359.0,
+                    1208.0,
+                    1359.0,
+                    1208.0,
+                    1373.0,
+                    1192.0,
+                    1373.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    375.0,
+                    1369.0,
+                    437.0,
+                    1369.0,
+                    437.0,
+                    1405.0,
+                    375.0,
+                    1405.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    700.0,
+                    1370.0,
+                    761.0,
+                    1370.0,
+                    761.0,
+                    1405.0,
+                    700.0,
+                    1405.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    939.0,
+                    1368.0,
+                    961.0,
+                    1368.0,
+                    961.0,
+                    1385.0,
+                    939.0,
+                    1385.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1024.0,
+                    1370.0,
+                    1085.0,
+                    1370.0,
+                    1085.0,
+                    1405.0,
+                    1024.0,
+                    1405.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    426.0,
+                    1405.0,
+                    681.0,
+                    1405.0,
+                    681.0,
+                    1436.0,
+                    426.0,
+                    1436.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    750.0,
+                    1405.0,
+                    1005.0,
+                    1405.0,
+                    1005.0,
+                    1436.0,
+                    750.0,
+                    1436.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1075.0,
+                    1405.0,
+                    1329.0,
+                    1405.0,
+                    1329.0,
+                    1436.0,
+                    1075.0,
+                    1436.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    528.0,
+                    1426.0,
+                    577.0,
+                    1426.0,
+                    577.0,
+                    1485.0,
+                    528.0,
+                    1485.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    853.0,
+                    1426.0,
+                    901.0,
+                    1426.0,
+                    901.0,
+                    1486.0,
+                    853.0,
+                    1486.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1180.0,
+                    1427.0,
+                    1222.0,
+                    1427.0,
+                    1222.0,
+                    1485.0,
+                    1180.0,
+                    1485.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    454.0,
+                    884.5,
+                    500.0,
+                    884.5,
+                    500.0,
+                    945.0,
+                    454.0,
+                    945.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    433.75,
+                    1012.5,
+                    523.75,
+                    1012.5,
+                    523.75,
+                    1068.5,
+                    433.75,
+                    1068.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    387.0,
+                    243.0,
+                    431.0,
+                    243.0,
+                    431.0,
+                    273.0,
+                    387.0,
+                    273.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    549.0,
+                    253.0,
+                    579.0,
+                    253.0,
+                    579.0,
+                    280.0,
+                    549.0,
+                    280.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    712.0,
+                    243.0,
+                    752.0,
+                    243.0,
+                    752.0,
+                    273.0,
+                    712.0,
+                    273.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    862.0,
+                    253.0,
+                    935.0,
+                    253.0,
+                    935.0,
+                    280.0,
+                    862.0,
+                    280.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1036.0,
+                    243.0,
+                    1077.0,
+                    243.0,
+                    1077.0,
+                    273.0,
+                    1036.0,
+                    273.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1197.0,
+                    252.0,
+                    1248.0,
+                    252.0,
+                    1248.0,
+                    281.0,
+                    1197.0,
+                    281.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1291.0,
+                    258.0,
+                    1303.0,
+                    258.0,
+                    1303.0,
+                    273.0,
+                    1291.0,
+                    273.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    631.0,
+                    268.0,
+                    644.0,
+                    268.0,
+                    644.0,
+                    283.0,
+                    631.0,
+                    283.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    955.0,
+                    267.0,
+                    967.0,
+                    267.0,
+                    967.0,
+                    283.0,
+                    955.0,
+                    283.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    387.0,
+                    286.0,
+                    427.0,
+                    286.0,
+                    427.0,
+                    316.0,
+                    387.0,
+                    316.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    548.0,
+                    274.0,
+                    599.0,
+                    274.0,
+                    599.0,
+                    305.0,
+                    548.0,
+                    305.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    712.0,
+                    287.0,
+                    753.0,
+                    287.0,
+                    753.0,
+                    316.0,
+                    712.0,
+                    316.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    861.0,
+                    275.0,
+                    914.0,
+                    275.0,
+                    914.0,
+                    303.0,
+                    861.0,
+                    303.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    936.0,
+                    296.0,
+                    944.0,
+                    296.0,
+                    944.0,
+                    306.0,
+                    936.0,
+                    306.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1036.0,
+                    286.0,
+                    1076.0,
+                    286.0,
+                    1076.0,
+                    316.0,
+                    1036.0,
+                    316.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1196.0,
+                    275.0,
+                    1246.0,
+                    275.0,
+                    1246.0,
+                    303.0,
+                    1196.0,
+                    303.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    364.0,
+                    310.0,
+                    425.0,
+                    310.0,
+                    425.0,
+                    409.0,
+                    364.0,
+                    409.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    575.0,
+                    331.0,
+                    584.0,
+                    331.0,
+                    584.0,
+                    343.0,
+                    575.0,
+                    343.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    595.0,
+                    302.0,
+                    613.0,
+                    302.0,
+                    613.0,
+                    321.0,
+                    595.0,
+                    321.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    689.0,
+                    308.0,
+                    751.0,
+                    308.0,
+                    751.0,
+                    410.0,
+                    689.0,
+                    410.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1013.0,
+                    310.0,
+                    1074.0,
+                    310.0,
+                    1074.0,
+                    409.0,
+                    1013.0,
+                    409.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    566.0,
+                    340.0,
+                    575.0,
+                    340.0,
+                    575.0,
+                    352.0,
+                    566.0,
+                    352.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    880.0,
+                    352.0,
+                    890.0,
+                    352.0,
+                    890.0,
+                    361.0,
+                    880.0,
+                    361.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    866.0,
+                    367.0,
+                    876.0,
+                    367.0,
+                    876.0,
+                    379.0,
+                    866.0,
+                    379.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    388.0,
+                    373.0,
+                    425.0,
+                    373.0,
+                    425.0,
+                    400.0,
+                    388.0,
+                    400.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    713.0,
+                    374.0,
+                    750.0,
+                    374.0,
+                    750.0,
+                    401.0,
+                    713.0,
+                    401.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1038.0,
+                    373.0,
+                    1074.0,
+                    373.0,
+                    1074.0,
+                    400.0,
+                    1038.0,
+                    400.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    507.0,
+                    401.0,
+                    517.0,
+                    401.0,
+                    517.0,
+                    412.0,
+                    507.0,
+                    412.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    387.0,
+                    415.0,
+                    427.0,
+                    415.0,
+                    427.0,
+                    445.0,
+                    387.0,
+                    445.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    500.0,
+                    415.0,
+                    505.0,
+                    415.0,
+                    505.0,
+                    421.0,
+                    500.0,
+                    421.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    711.0,
+                    415.0,
+                    751.0,
+                    415.0,
+                    751.0,
+                    445.0,
+                    711.0,
+                    445.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1036.0,
+                    415.0,
+                    1076.0,
+                    415.0,
+                    1076.0,
+                    445.0,
+                    1036.0,
+                    445.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    797.0,
+                    437.0,
+                    810.0,
+                    437.0,
+                    810.0,
+                    451.0,
+                    797.0,
+                    451.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    387.0,
+                    457.0,
+                    426.0,
+                    457.0,
+                    426.0,
+                    487.0,
+                    387.0,
+                    487.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    711.0,
+                    458.0,
+                    752.0,
+                    458.0,
+                    752.0,
+                    487.0,
+                    711.0,
+                    487.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1037.0,
+                    459.0,
+                    1074.0,
+                    459.0,
+                    1074.0,
+                    486.0,
+                    1037.0,
+                    486.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    426.0,
+                    472.0,
+                    679.0,
+                    472.0,
+                    679.0,
+                    504.0,
+                    426.0,
+                    504.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    750.0,
+                    472.0,
+                    1003.0,
+                    472.0,
+                    1003.0,
+                    504.0,
+                    750.0,
+                    504.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1075.0,
+                    472.0,
+                    1327.0,
+                    472.0,
+                    1327.0,
+                    504.0,
+                    1075.0,
+                    504.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    496.0,
+                    494.0,
+                    611.0,
+                    494.0,
+                    611.0,
+                    527.0,
+                    496.0,
+                    527.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    819.0,
+                    494.0,
+                    936.0,
+                    494.0,
+                    936.0,
+                    527.0,
+                    819.0,
+                    527.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1144.0,
+                    494.0,
+                    1261.0,
+                    494.0,
+                    1261.0,
+                    527.0,
+                    1144.0,
+                    527.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    531.0,
+                    516.0,
+                    573.0,
+                    516.0,
+                    573.0,
+                    551.0,
+                    531.0,
+                    551.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    857.0,
+                    517.0,
+                    897.0,
+                    517.0,
+                    897.0,
+                    548.0,
+                    857.0,
+                    548.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1181.0,
+                    516.0,
+                    1223.0,
+                    516.0,
+                    1223.0,
+                    551.0,
+                    1181.0,
+                    551.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    645.0,
+                    258.0,
+                    648.0,
+                    258.0,
+                    648.0,
+                    273.0,
+                    645.0,
+                    273.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    614.0,
+                    279.0,
+                    638.0,
+                    279.0,
+                    638.0,
+                    293.0,
+                    614.0,
+                    293.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1276.0,
+                    270.5,
+                    1297.0,
+                    270.5,
+                    1297.0,
+                    281.0,
+                    1276.0,
+                    281.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    546.0,
+                    354.0,
+                    568.0,
+                    354.0,
+                    568.0,
+                    362.0,
+                    546.0,
+                    362.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    842.75,
+                    379.0,
+                    871.75,
+                    379.0,
+                    871.75,
+                    390.0,
+                    842.75,
+                    390.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    824.0,
+                    395.0,
+                    858.0,
+                    395.0,
+                    858.0,
+                    408.5,
+                    824.0,
+                    408.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    799.0,
+                    419.5,
+                    839.0,
+                    419.5,
+                    839.0,
+                    431.5,
+                    799.0,
+                    431.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    465.0,
+                    430.0,
+                    505.0,
+                    430.0,
+                    505.0,
+                    441.5,
+                    465.0,
+                    441.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1111.75,
+                    356.5,
+                    1291.75,
+                    356.5,
+                    1291.75,
+                    372.0,
+                    1111.75,
+                    372.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    451.0,
+                    448.5,
+                    482.0,
+                    448.5,
+                    482.0,
+                    460.5,
+                    451.0,
+                    460.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    776.25,
+                    450.5,
+                    803.25,
+                    450.5,
+                    803.25,
+                    460.5,
+                    776.25,
+                    460.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1104.0,
+                    451.0,
+                    1127.0,
+                    451.0,
+                    1127.0,
+                    459.0,
+                    1104.0,
+                    459.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1515.0,
+                    1405.0,
+                    1515.0,
+                    1405.0,
+                    1552.0,
+                    294.0,
+                    1552.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1545.0,
+                    1404.0,
+                    1545.0,
+                    1404.0,
+                    1581.0,
+                    294.0,
+                    1581.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1578.0,
+                    693.0,
+                    1578.0,
+                    693.0,
+                    1609.0,
+                    294.0,
+                    1609.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    578.0,
+                    1406.0,
+                    578.0,
+                    1406.0,
+                    617.0,
+                    293.0,
+                    617.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    613.0,
+                    535.0,
+                    613.0,
+                    535.0,
+                    645.0,
+                    294.0,
+                    645.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    591.0,
+                    613.0,
+                    601.0,
+                    613.0,
+                    601.0,
+                    645.0,
+                    591.0,
+                    645.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1704.0,
+                    548.0,
+                    1704.0,
+                    548.0,
+                    1752.0,
+                    291.0,
+                    1752.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    73.0,
+                    858.0,
+                    73.0,
+                    858.0,
+                    108.0,
+                    297.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    831.0,
+                    2084.0,
+                    868.0,
+                    2084.0,
+                    868.0,
+                    2124.0,
+                    831.0,
+                    2124.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1789.0,
+                    1407.0,
+                    1789.0,
+                    1407.0,
+                    1826.0,
+                    293.0,
+                    1826.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1821.0,
+                    1405.0,
+                    1821.0,
+                    1405.0,
+                    1852.0,
+                    293.0,
+                    1852.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1851.0,
+                    1407.0,
+                    1851.0,
+                    1407.0,
+                    1885.0,
+                    293.0,
+                    1885.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1880.0,
+                    1405.0,
+                    1880.0,
+                    1405.0,
+                    1916.0,
+                    295.0,
+                    1916.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1912.0,
+                    1405.0,
+                    1912.0,
+                    1405.0,
+                    1946.0,
+                    292.0,
+                    1946.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1943.0,
+                    1406.0,
+                    1943.0,
+                    1406.0,
+                    1977.0,
+                    292.0,
+                    1977.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1973.0,
+                    1405.0,
+                    1973.0,
+                    1405.0,
+                    2007.0,
+                    293.0,
+                    2007.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    2004.0,
+                    455.0,
+                    2004.0,
+                    455.0,
+                    2039.0,
+                    293.0,
+                    2039.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    721.0,
+                    1405.0,
+                    721.0,
+                    1405.0,
+                    757.0,
+                    295.0,
+                    757.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    753.0,
+                    1178.0,
+                    753.0,
+                    1178.0,
+                    785.0,
+                    296.0,
+                    785.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 9,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 2,
+                "poly": [
+                    298,
+                    75,
+                    858,
+                    75,
+                    858,
+                    105,
+                    298,
+                    105
+                ],
+                "score": 0.895
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    295,
+                    392,
+                    1402,
+                    392,
+                    1402,
+                    457,
+                    295,
+                    457
+                ],
+                "score": 0.823
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    835,
+                    2088,
+                    861,
+                    2088,
+                    861,
+                    2113,
+                    835,
+                    2113
+                ],
+                "score": 0.818
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    294,
+                    1574,
+                    1403,
+                    1574,
+                    1403,
+                    1639,
+                    294,
+                    1639
+                ],
+                "score": 0.813
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    296,
+                    727,
+                    1406,
+                    727,
+                    1406,
+                    793,
+                    296,
+                    793
+                ],
+                "score": 0.813
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    295,
+                    477,
+                    1403,
+                    477,
+                    1403,
+                    542,
+                    295,
+                    542
+                ],
+                "score": 0.812
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    293,
+                    644,
+                    1405,
+                    644,
+                    1405,
+                    708,
+                    293,
+                    708
+                ],
+                "score": 0.805
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    302,
+                    1657,
+                    1404,
+                    1657,
+                    1404,
+                    1753,
+                    302,
+                    1753
+                ],
+                "score": 0.802
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1407,
+                    1404,
+                    1407,
+                    1404,
+                    1471,
+                    297,
+                    1471
+                ],
+                "score": 0.788
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    295,
+                    1490,
+                    1405,
+                    1490,
+                    1405,
+                    1555,
+                    295,
+                    1555
+                ],
+                "score": 0.786
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    296,
+                    560,
+                    1402,
+                    560,
+                    1402,
+                    626,
+                    296,
+                    626
+                ],
+                "score": 0.778
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    295,
+                    1772,
+                    1405,
+                    1772,
+                    1405,
+                    1836,
+                    295,
+                    1836
+                ],
+                "score": 0.77
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    300,
+                    227,
+                    489,
+                    227,
+                    489,
+                    261,
+                    300,
+                    261
+                ],
+                "score": 0.758
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    294,
+                    811,
+                    1404,
+                    811,
+                    1404,
+                    878,
+                    294,
+                    878
+                ],
+                "score": 0.751
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    279,
+                    1407,
+                    279,
+                    1407,
+                    371,
+                    299,
+                    371
+                ],
+                "score": 0.727
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1972,
+                    1398,
+                    1972,
+                    1398,
+                    2034,
+                    297,
+                    2034
+                ],
+                "score": 0.727
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1009,
+                    1402,
+                    1009,
+                    1402,
+                    1076,
+                    298,
+                    1076
+                ],
+                "score": 0.715
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1208,
+                    1403,
+                    1208,
+                    1403,
+                    1273,
+                    297,
+                    1273
+                ],
+                "score": 0.708
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1857,
+                    1402,
+                    1857,
+                    1402,
+                    1951,
+                    297,
+                    1951
+                ],
+                "score": 0.66
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    296,
+                    1093,
+                    1403,
+                    1093,
+                    1403,
+                    1190,
+                    296,
+                    1190
+                ],
+                "score": 0.639
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1291,
+                    1407,
+                    1291,
+                    1407,
+                    1386,
+                    297,
+                    1386
+                ],
+                "score": 0.626
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    895,
+                    1405,
+                    895,
+                    1405,
+                    991,
+                    299,
+                    991
+                ],
+                "score": 0.608
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    300,
+                    227,
+                    489,
+                    227,
+                    489,
+                    261,
+                    300,
+                    261
+                ],
+                "score": 0.121
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    904,
+                    1725,
+                    922,
+                    1725,
+                    922,
+                    1747,
+                    904,
+                    1747
+                ],
+                "score": 0.45,
+                "latex": "="
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    298.0,
+                    73.0,
+                    858.0,
+                    73.0,
+                    858.0,
+                    108.0,
+                    298.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    831.0,
+                    2085.0,
+                    867.0,
+                    2085.0,
+                    867.0,
+                    2125.0,
+                    831.0,
+                    2125.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    229.0,
+                    490.0,
+                    229.0,
+                    490.0,
+                    263.0,
+                    297.0,
+                    263.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    391.0,
+                    1405.0,
+                    391.0,
+                    1405.0,
+                    431.0,
+                    294.0,
+                    431.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    321.0,
+                    423.0,
+                    506.0,
+                    423.0,
+                    506.0,
+                    456.0,
+                    321.0,
+                    456.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1570.0,
+                    1408.0,
+                    1570.0,
+                    1408.0,
+                    1614.0,
+                    293.0,
+                    1614.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    1606.0,
+                    690.0,
+                    1606.0,
+                    690.0,
+                    1639.0,
+                    323.0,
+                    1639.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    724.0,
+                    1412.0,
+                    724.0,
+                    1412.0,
+                    767.0,
+                    293.0,
+                    767.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    324.0,
+                    761.0,
+                    571.0,
+                    761.0,
+                    571.0,
+                    792.0,
+                    324.0,
+                    792.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    473.0,
+                    1404.0,
+                    473.0,
+                    1404.0,
+                    516.0,
+                    293.0,
+                    516.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    320.0,
+                    507.0,
+                    901.0,
+                    507.0,
+                    901.0,
+                    542.0,
+                    320.0,
+                    542.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    639.0,
+                    1408.0,
+                    639.0,
+                    1408.0,
+                    684.0,
+                    292.0,
+                    684.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    319.0,
+                    671.0,
+                    398.0,
+                    671.0,
+                    398.0,
+                    709.0,
+                    319.0,
+                    709.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1656.0,
+                    1405.0,
+                    1656.0,
+                    1405.0,
+                    1695.0,
+                    293.0,
+                    1695.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    321.0,
+                    1689.0,
+                    1409.0,
+                    1689.0,
+                    1409.0,
+                    1726.0,
+                    321.0,
+                    1726.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    1717.0,
+                    903.0,
+                    1717.0,
+                    903.0,
+                    1758.0,
+                    323.0,
+                    1758.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    923.0,
+                    1717.0,
+                    1100.0,
+                    1717.0,
+                    1100.0,
+                    1758.0,
+                    923.0,
+                    1758.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1402.0,
+                    1405.0,
+                    1402.0,
+                    1405.0,
+                    1445.0,
+                    292.0,
+                    1445.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    321.0,
+                    1435.0,
+                    597.0,
+                    1435.0,
+                    597.0,
+                    1471.0,
+                    321.0,
+                    1471.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1486.0,
+                    1405.0,
+                    1486.0,
+                    1405.0,
+                    1530.0,
+                    292.0,
+                    1530.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    320.0,
+                    1520.0,
+                    598.0,
+                    1520.0,
+                    598.0,
+                    1553.0,
+                    320.0,
+                    1553.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    558.0,
+                    1407.0,
+                    558.0,
+                    1407.0,
+                    601.0,
+                    293.0,
+                    601.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    592.0,
+                    1306.0,
+                    592.0,
+                    1306.0,
+                    628.0,
+                    323.0,
+                    628.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1768.0,
+                    1408.0,
+                    1768.0,
+                    1408.0,
+                    1812.0,
+                    292.0,
+                    1812.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    319.0,
+                    1800.0,
+                    400.0,
+                    1800.0,
+                    400.0,
+                    1839.0,
+                    319.0,
+                    1839.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    809.0,
+                    1407.0,
+                    809.0,
+                    1407.0,
+                    850.0,
+                    296.0,
+                    850.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    843.0,
+                    1387.0,
+                    843.0,
+                    1387.0,
+                    880.0,
+                    323.0,
+                    880.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    278.0,
+                    1405.0,
+                    278.0,
+                    1405.0,
+                    312.0,
+                    296.0,
+                    312.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    311.0,
+                    1408.0,
+                    311.0,
+                    1408.0,
+                    344.0,
+                    322.0,
+                    344.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    321.0,
+                    338.0,
+                    395.0,
+                    338.0,
+                    395.0,
+                    372.0,
+                    321.0,
+                    372.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1970.0,
+                    1405.0,
+                    1970.0,
+                    1405.0,
+                    2009.0,
+                    295.0,
+                    2009.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    2002.0,
+                    970.0,
+                    2002.0,
+                    970.0,
+                    2036.0,
+                    322.0,
+                    2036.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1009.0,
+                    1407.0,
+                    1009.0,
+                    1407.0,
+                    1046.0,
+                    294.0,
+                    1046.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    1041.0,
+                    953.0,
+                    1041.0,
+                    953.0,
+                    1076.0,
+                    323.0,
+                    1076.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1207.0,
+                    1405.0,
+                    1207.0,
+                    1405.0,
+                    1246.0,
+                    294.0,
+                    1246.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    320.0,
+                    1241.0,
+                    707.0,
+                    1241.0,
+                    707.0,
+                    1274.0,
+                    320.0,
+                    1274.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1854.0,
+                    1406.0,
+                    1854.0,
+                    1406.0,
+                    1894.0,
+                    294.0,
+                    1894.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    321.0,
+                    1886.0,
+                    1408.0,
+                    1886.0,
+                    1408.0,
+                    1927.0,
+                    321.0,
+                    1927.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    324.0,
+                    1920.0,
+                    533.0,
+                    1920.0,
+                    533.0,
+                    1950.0,
+                    324.0,
+                    1950.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1092.0,
+                    1404.0,
+                    1092.0,
+                    1404.0,
+                    1130.0,
+                    294.0,
+                    1130.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    324.0,
+                    1127.0,
+                    1405.0,
+                    1127.0,
+                    1405.0,
+                    1161.0,
+                    324.0,
+                    1161.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    1155.0,
+                    944.0,
+                    1155.0,
+                    944.0,
+                    1193.0,
+                    323.0,
+                    1193.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1290.0,
+                    1407.0,
+                    1290.0,
+                    1407.0,
+                    1328.0,
+                    294.0,
+                    1328.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    320.0,
+                    1320.0,
+                    1410.0,
+                    1320.0,
+                    1410.0,
+                    1360.0,
+                    320.0,
+                    1360.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    320.0,
+                    1351.0,
+                    400.0,
+                    1351.0,
+                    400.0,
+                    1387.0,
+                    320.0,
+                    1387.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    893.0,
+                    1409.0,
+                    893.0,
+                    1409.0,
+                    934.0,
+                    293.0,
+                    934.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    927.0,
+                    1407.0,
+                    927.0,
+                    1407.0,
+                    964.0,
+                    322.0,
+                    964.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    958.0,
+                    849.0,
+                    958.0,
+                    849.0,
+                    996.0,
+                    323.0,
+                    996.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    229.0,
+                    490.0,
+                    229.0,
+                    490.0,
+                    263.0,
+                    297.0,
+                    263.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 10,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 2,
+                "poly": [
+                    298,
+                    75,
+                    858,
+                    75,
+                    858,
+                    105,
+                    298,
+                    105
+                ],
+                "score": 0.893
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    835,
+                    2088,
+                    864,
+                    2088,
+                    864,
+                    2113,
+                    835,
+                    2113
+                ],
+                "score": 0.836
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1399,
+                    1401,
+                    1399,
+                    1401,
+                    1464,
+                    298,
+                    1464
+                ],
+                "score": 0.83
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    296,
+                    727,
+                    1402,
+                    727,
+                    1402,
+                    792,
+                    296,
+                    792
+                ],
+                "score": 0.821
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    293,
+                    976,
+                    1402,
+                    976,
+                    1402,
+                    1041,
+                    293,
+                    1041
+                ],
+                "score": 0.816
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    296,
+                    1482,
+                    1402,
+                    1482,
+                    1402,
+                    1547,
+                    296,
+                    1547
+                ],
+                "score": 0.813
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    301,
+                    643,
+                    1403,
+                    643,
+                    1403,
+                    708,
+                    301,
+                    708
+                ],
+                "score": 0.811
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    809,
+                    1402,
+                    809,
+                    1402,
+                    905,
+                    298,
+                    905
+                ],
+                "score": 0.802
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1286,
+                    1407,
+                    1286,
+                    1407,
+                    1382,
+                    298,
+                    1382
+                ],
+                "score": 0.786
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1059,
+                    1405,
+                    1059,
+                    1405,
+                    1153,
+                    297,
+                    1153
+                ],
+                "score": 0.785
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    1565,
+                    1405,
+                    1565,
+                    1405,
+                    1661,
+                    299,
+                    1661
+                ],
+                "score": 0.775
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1172,
+                    1407,
+                    1172,
+                    1407,
+                    1269,
+                    297,
+                    1269
+                ],
+                "score": 0.772
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    296,
+                    923,
+                    1193,
+                    923,
+                    1193,
+                    959,
+                    296,
+                    959
+                ],
+                "score": 0.753
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    295,
+                    560,
+                    1401,
+                    560,
+                    1401,
+                    627,
+                    295,
+                    627
+                ],
+                "score": 0.745
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    477,
+                    1402,
+                    477,
+                    1402,
+                    544,
+                    297,
+                    544
+                ],
+                "score": 0.71
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    294,
+                    395,
+                    1403,
+                    395,
+                    1403,
+                    460,
+                    294,
+                    460
+                ],
+                "score": 0.682
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    292,
+                    228,
+                    1403,
+                    228,
+                    1403,
+                    295,
+                    292,
+                    295
+                ],
+                "score": 0.623
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    294,
+                    311,
+                    1402,
+                    311,
+                    1402,
+                    378,
+                    294,
+                    378
+                ],
+                "score": 0.594
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    298.0,
+                    73.0,
+                    858.0,
+                    73.0,
+                    858.0,
+                    108.0,
+                    298.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    831.0,
+                    2085.0,
+                    869.0,
+                    2085.0,
+                    869.0,
+                    2124.0,
+                    831.0,
+                    2124.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1393.0,
+                    1407.0,
+                    1393.0,
+                    1407.0,
+                    1443.0,
+                    291.0,
+                    1443.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    1432.0,
+                    1085.0,
+                    1432.0,
+                    1085.0,
+                    1465.0,
+                    322.0,
+                    1465.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    722.0,
+                    1406.0,
+                    722.0,
+                    1406.0,
+                    767.0,
+                    294.0,
+                    767.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    325.0,
+                    759.0,
+                    705.0,
+                    759.0,
+                    705.0,
+                    792.0,
+                    325.0,
+                    792.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    974.0,
+                    1407.0,
+                    974.0,
+                    1407.0,
+                    1014.0,
+                    293.0,
+                    1014.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    320.0,
+                    1007.0,
+                    609.0,
+                    1007.0,
+                    609.0,
+                    1041.0,
+                    320.0,
+                    1041.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1481.0,
+                    1404.0,
+                    1481.0,
+                    1404.0,
+                    1520.0,
+                    294.0,
+                    1520.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    1513.0,
+                    1022.0,
+                    1513.0,
+                    1022.0,
+                    1548.0,
+                    323.0,
+                    1548.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    639.0,
+                    1407.0,
+                    639.0,
+                    1407.0,
+                    683.0,
+                    292.0,
+                    683.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    324.0,
+                    677.0,
+                    595.0,
+                    677.0,
+                    595.0,
+                    709.0,
+                    324.0,
+                    709.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    809.0,
+                    1406.0,
+                    809.0,
+                    1406.0,
+                    847.0,
+                    294.0,
+                    847.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    841.0,
+                    1404.0,
+                    841.0,
+                    1404.0,
+                    878.0,
+                    322.0,
+                    878.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    873.0,
+                    596.0,
+                    873.0,
+                    596.0,
+                    902.0,
+                    323.0,
+                    902.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1284.0,
+                    1408.0,
+                    1284.0,
+                    1408.0,
+                    1324.0,
+                    296.0,
+                    1324.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    320.0,
+                    1317.0,
+                    1408.0,
+                    1317.0,
+                    1408.0,
+                    1356.0,
+                    320.0,
+                    1356.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    324.0,
+                    1349.0,
+                    1325.0,
+                    1349.0,
+                    1325.0,
+                    1383.0,
+                    324.0,
+                    1383.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1056.0,
+                    1405.0,
+                    1056.0,
+                    1405.0,
+                    1099.0,
+                    294.0,
+                    1099.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    320.0,
+                    1088.0,
+                    1408.0,
+                    1088.0,
+                    1408.0,
+                    1130.0,
+                    320.0,
+                    1130.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    324.0,
+                    1122.0,
+                    534.0,
+                    1122.0,
+                    534.0,
+                    1152.0,
+                    324.0,
+                    1152.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1560.0,
+                    1406.0,
+                    1560.0,
+                    1406.0,
+                    1607.0,
+                    292.0,
+                    1607.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    1598.0,
+                    1405.0,
+                    1598.0,
+                    1405.0,
+                    1633.0,
+                    323.0,
+                    1633.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    320.0,
+                    1628.0,
+                    584.0,
+                    1628.0,
+                    584.0,
+                    1659.0,
+                    320.0,
+                    1659.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1172.0,
+                    1407.0,
+                    1172.0,
+                    1407.0,
+                    1211.0,
+                    294.0,
+                    1211.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    1205.0,
+                    1405.0,
+                    1205.0,
+                    1405.0,
+                    1239.0,
+                    323.0,
+                    1239.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    1235.0,
+                    1022.0,
+                    1235.0,
+                    1022.0,
+                    1270.0,
+                    323.0,
+                    1270.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    922.0,
+                    1192.0,
+                    922.0,
+                    1192.0,
+                    961.0,
+                    295.0,
+                    961.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    557.0,
+                    1403.0,
+                    557.0,
+                    1403.0,
+                    599.0,
+                    295.0,
+                    599.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    592.0,
+                    1127.0,
+                    592.0,
+                    1127.0,
+                    627.0,
+                    322.0,
+                    627.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    478.0,
+                    1403.0,
+                    478.0,
+                    1403.0,
+                    514.0,
+                    296.0,
+                    514.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    511.0,
+                    1336.0,
+                    511.0,
+                    1336.0,
+                    544.0,
+                    322.0,
+                    544.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    395.0,
+                    1405.0,
+                    395.0,
+                    1405.0,
+                    431.0,
+                    294.0,
+                    431.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    427.0,
+                    671.0,
+                    427.0,
+                    671.0,
+                    460.0,
+                    323.0,
+                    460.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    229.0,
+                    1405.0,
+                    229.0,
+                    1405.0,
+                    265.0,
+                    295.0,
+                    265.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    321.0,
+                    262.0,
+                    1144.0,
+                    262.0,
+                    1144.0,
+                    295.0,
+                    321.0,
+                    295.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    312.0,
+                    1405.0,
+                    312.0,
+                    1405.0,
+                    348.0,
+                    296.0,
+                    348.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    321.0,
+                    343.0,
+                    1092.0,
+                    343.0,
+                    1092.0,
+                    378.0,
+                    321.0,
+                    378.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 11,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1220,
+                    1406,
+                    1220,
+                    1406,
+                    1467,
+                    297,
+                    1467
+                ],
+                "score": 0.983
+            },
+            {
+                "category_id": 5,
+                "poly": [
+                    299,
+                    1567,
+                    1399,
+                    1567,
+                    1399,
+                    1902,
+                    299,
+                    1902
+                ],
+                "score": 0.982,
+                "html": "<table><tr><td rowspan=1 colspan=1>Dataset</td><td rowspan=1 colspan=1>Generator/Discriminator Architecture</td><td rowspan=1 colspan=1>Predictor Architecture</td></tr><tr><td rowspan=1 colspan=1>CIFAR-10</td><td rowspan=1 colspan=1>C7S1P3-64，C3S2P1-128，Attn，R-128,Attn，R-128，Attn，R-128，Attn,R-128，CT3S2P1-128，CT7S1P3-3，Tanh/Sigmoid</td><td rowspan=1 colspan=1>ResNet-18 11,12</td></tr><tr><td rowspan=1 colspan=1>Landsat</td><td rowspan=1 colspan=1>FC-64，Sigmoid，BNorm，FC-64，Sigmoid,BNorm,FC-64，Sigmoid，BNorm，FC-36，Tanh/Sigmoid</td><td rowspan=1 colspan=1>FC-64，ReLU，BNorm，FC-64,ReLU，BNorm，FC-6，Softmax</td></tr><tr><td rowspan=1 colspan=1>MIT-BIH</td><td rowspan=1 colspan=1>FC-1860，ReLU，BNorm，FC-1860，ReLU，BNorm,FC-1860，ReLU，BNorm, FC-186,Tanh/Sigmoid</td><td rowspan=1 colspan=1>FC-1860，ReLU，BNorm，FC-1860,ReLU，BNorm，FC-5，Softmax</td></tr><tr><td rowspan=1 colspan=1>Diabetes</td><td rowspan=1 colspan=1>FC-45，ReLU，BNorm，FC-45，ReLU，BNorm,FC-45，ReLU，BNorm，FC-45，Tanh/Sigmoid</td><td rowspan=1 colspan=1>FC-22，ReLU，BNorm，FC-22,ReLU，BNorm，FC-3，Softmax</td></tr></table>"
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    302,
+                    1405,
+                    302,
+                    1405,
+                    426,
+                    297,
+                    426
+                ],
+                "score": 0.977
+            },
+            {
+                "category_id": 5,
+                "poly": [
+                    660,
+                    523,
+                    1051,
+                    523,
+                    1051,
+                    1060,
+                    660,
+                    1060
+                ],
+                "score": 0.977,
+                "html": "<table><tr><td>Dependency</td><td>Version</td></tr><tr><td>python</td><td>3.7.1</td></tr><tr><td>pytorch</td><td>1.0.0</td></tr><tr><td>cuda100</td><td>1.0</td></tr><tr><td>ipython</td><td>7.2.0</td></tr><tr><td>jupyter</td><td>1.0.0</td></tr><tr><td>numpy</td><td>1.15.4</td></tr><tr><td>pandas</td><td>0.24.1</td></tr><tr><td>scikit-learn</td><td>0.20.1</td></tr><tr><td>scipy</td><td>1.1.0</td></tr><tr><td>torchvision</td><td>0.2.1</td></tr><tr><td>tqdm</td><td>4.28.1</td></tr><tr><td>matplotlib</td><td>3.0.1</td></tr></table>"
+            },
+            {
+                "category_id": 6,
+                "poly": [
+                    667,
+                    465,
+                    1033,
+                    465,
+                    1033,
+                    498,
+                    667,
+                    498
+                ],
+                "score": 0.917
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    300,
+                    1146,
+                    757,
+                    1146,
+                    757,
+                    1182,
+                    300,
+                    1182
+                ],
+                "score": 0.904
+            },
+            {
+                "category_id": 6,
+                "poly": [
+                    534,
+                    1506,
+                    1162,
+                    1506,
+                    1162,
+                    1539,
+                    534,
+                    1539
+                ],
+                "score": 0.902
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    300,
+                    226,
+                    903,
+                    226,
+                    903,
+                    262,
+                    300,
+                    262
+                ],
+                "score": 0.89
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    298,
+                    74,
+                    858,
+                    74,
+                    858,
+                    105,
+                    298,
+                    105
+                ],
+                "score": 0.876
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    835,
+                    2088,
+                    863,
+                    2088,
+                    863,
+                    2113,
+                    835,
+                    2113
+                ],
+                "score": 0.863
+            },
+            {
+                "category_id": 7,
+                "poly": [
+                    298,
+                    1947,
+                    1404,
+                    1947,
+                    1404,
+                    2034,
+                    298,
+                    2034
+                ],
+                "score": 0.588
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    298,
+                    1947,
+                    1404,
+                    1947,
+                    1404,
+                    2034,
+                    298,
+                    2034
+                ],
+                "score": 0.31
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1182,
+                    1347,
+                    1333,
+                    1347,
+                    1333,
+                    1376,
+                    1182,
+                    1376
+                ],
+                "score": 0.87,
+                "latex": "\\mathsf { C T x S y P z - t }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    639,
+                    1316,
+                    694,
+                    1316,
+                    694,
+                    1343,
+                    639,
+                    1343
+                ],
+                "score": 0.74,
+                "latex": "_ \\mathrm { R - x }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    514,
+                    1256,
+                    649,
+                    1256,
+                    649,
+                    1285,
+                    514,
+                    1285
+                ],
+                "score": 0.72,
+                "latex": "\\mathtt { C x S y P z - t }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    889,
+                    1377,
+                    961,
+                    1377,
+                    961,
+                    1404,
+                    889,
+                    1404
+                ],
+                "score": 0.65,
+                "latex": "\\operatorname { F C - x }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    742,
+                    1378,
+                    878,
+                    1378,
+                    878,
+                    1406,
+                    742,
+                    1406
+                ],
+                "score": 0.64,
+                "latex": "\\mathtt { C x S y P z - t }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1181,
+                    1258,
+                    1202,
+                    1258,
+                    1202,
+                    1282,
+                    1181,
+                    1282
+                ],
+                "score": 0.43,
+                "latex": "_ \\textrm { x }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    677,
+                    1287,
+                    698,
+                    1287,
+                    698,
+                    1312,
+                    677,
+                    1312
+                ],
+                "score": 0.41,
+                "latex": "\\scriptstyle \\pm"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    388,
+                    1410,
+                    410,
+                    1410,
+                    410,
+                    1434,
+                    388,
+                    1434
+                ],
+                "score": 0.4,
+                "latex": "_ \\textrm { x }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    299,
+                    1290,
+                    316,
+                    1290,
+                    316,
+                    1312,
+                    299,
+                    1312
+                ],
+                "score": 0.31,
+                "latex": "_ { \\textrm { Z } }"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    664.0,
+                    461.0,
+                    1034.0,
+                    461.0,
+                    1034.0,
+                    503.0,
+                    664.0,
+                    503.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1143.0,
+                    761.0,
+                    1143.0,
+                    761.0,
+                    1186.0,
+                    292.0,
+                    1186.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    533.0,
+                    1502.0,
+                    1164.0,
+                    1502.0,
+                    1164.0,
+                    1543.0,
+                    533.0,
+                    1543.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    225.0,
+                    904.0,
+                    225.0,
+                    904.0,
+                    266.0,
+                    294.0,
+                    266.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    72.0,
+                    857.0,
+                    72.0,
+                    857.0,
+                    108.0,
+                    297.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    831.0,
+                    2085.0,
+                    868.0,
+                    2085.0,
+                    868.0,
+                    2123.0,
+                    831.0,
+                    2123.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    327.0,
+                    1939.0,
+                    1405.0,
+                    1939.0,
+                    1405.0,
+                    1985.0,
+                    327.0,
+                    1985.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    320.0,
+                    1968.0,
+                    1408.0,
+                    1968.0,
+                    1408.0,
+                    2015.0,
+                    320.0,
+                    2015.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    2003.0,
+                    565.0,
+                    2003.0,
+                    565.0,
+                    2036.0,
+                    294.0,
+                    2036.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    327.0,
+                    1939.0,
+                    1405.0,
+                    1939.0,
+                    1405.0,
+                    1985.0,
+                    327.0,
+                    1985.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    320.0,
+                    1968.0,
+                    1408.0,
+                    1968.0,
+                    1408.0,
+                    2015.0,
+                    320.0,
+                    2015.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    2003.0,
+                    565.0,
+                    2003.0,
+                    565.0,
+                    2036.0,
+                    294.0,
+                    2036.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1221.0,
+                    1404.0,
+                    1221.0,
+                    1404.0,
+                    1257.0,
+                    292.0,
+                    1257.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1252.0,
+                    513.0,
+                    1252.0,
+                    513.0,
+                    1289.0,
+                    294.0,
+                    1289.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    650.0,
+                    1252.0,
+                    1180.0,
+                    1252.0,
+                    1180.0,
+                    1289.0,
+                    650.0,
+                    1289.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1203.0,
+                    1252.0,
+                    1404.0,
+                    1252.0,
+                    1404.0,
+                    1289.0,
+                    1203.0,
+                    1289.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1284.0,
+                    298.0,
+                    1284.0,
+                    298.0,
+                    1318.0,
+                    294.0,
+                    1318.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    317.0,
+                    1284.0,
+                    676.0,
+                    1284.0,
+                    676.0,
+                    1318.0,
+                    317.0,
+                    1318.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    699.0,
+                    1284.0,
+                    1404.0,
+                    1284.0,
+                    1404.0,
+                    1318.0,
+                    699.0,
+                    1318.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1315.0,
+                    638.0,
+                    1315.0,
+                    638.0,
+                    1348.0,
+                    294.0,
+                    1348.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    695.0,
+                    1315.0,
+                    1404.0,
+                    1315.0,
+                    1404.0,
+                    1348.0,
+                    695.0,
+                    1348.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1345.0,
+                    1181.0,
+                    1345.0,
+                    1181.0,
+                    1378.0,
+                    295.0,
+                    1378.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1334.0,
+                    1345.0,
+                    1404.0,
+                    1345.0,
+                    1404.0,
+                    1378.0,
+                    1334.0,
+                    1378.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1374.0,
+                    741.0,
+                    1374.0,
+                    741.0,
+                    1410.0,
+                    292.0,
+                    1410.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    879.0,
+                    1374.0,
+                    888.0,
+                    1374.0,
+                    888.0,
+                    1410.0,
+                    879.0,
+                    1410.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    962.0,
+                    1374.0,
+                    1406.0,
+                    1374.0,
+                    1406.0,
+                    1410.0,
+                    962.0,
+                    1410.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1403.0,
+                    387.0,
+                    1403.0,
+                    387.0,
+                    1441.0,
+                    291.0,
+                    1441.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    411.0,
+                    1403.0,
+                    1407.0,
+                    1403.0,
+                    1407.0,
+                    1441.0,
+                    411.0,
+                    1441.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1437.0,
+                    1113.0,
+                    1437.0,
+                    1113.0,
+                    1468.0,
+                    296.0,
+                    1468.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    302.0,
+                    1406.0,
+                    302.0,
+                    1406.0,
+                    338.0,
+                    292.0,
+                    338.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    335.0,
+                    1403.0,
+                    335.0,
+                    1403.0,
+                    365.0,
+                    292.0,
+                    365.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    363.0,
+                    1406.0,
+                    363.0,
+                    1406.0,
+                    401.0,
+                    292.0,
+                    401.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    393.0,
+                    1072.0,
+                    393.0,
+                    1072.0,
+                    428.0,
+                    294.0,
+                    428.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 12,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    556,
+                    1405,
+                    556,
+                    1405,
+                    892,
+                    298,
+                    892
+                ],
+                "score": 0.982
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    906,
+                    1405,
+                    906,
+                    1405,
+                    1212,
+                    298,
+                    1212
+                ],
+                "score": 0.979
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    299,
+                    1947,
+                    1404,
+                    1947,
+                    1404,
+                    2035,
+                    299,
+                    2035
+                ],
+                "score": 0.953
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    296,
+                    1403,
+                    296,
+                    1403,
+                    541,
+                    298,
+                    541
+                ],
+                "score": 0.935
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    302,
+                    225,
+                    777,
+                    225,
+                    777,
+                    262,
+                    302,
+                    262
+                ],
+                "score": 0.868
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    835,
+                    2088,
+                    863,
+                    2088,
+                    863,
+                    2112,
+                    835,
+                    2112
+                ],
+                "score": 0.859
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    298,
+                    75,
+                    858,
+                    75,
+                    858,
+                    105,
+                    298,
+                    105
+                ],
+                "score": 0.747
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    461,
+                    1296,
+                    1021,
+                    1296,
+                    1021,
+                    1452,
+                    461,
+                    1452
+                ],
+                "score": 0.596
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    465,
+                    1256,
+                    928,
+                    1256,
+                    928,
+                    1288,
+                    465,
+                    1288
+                ],
+                "score": 0.567
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    460,
+                    1516,
+                    1215,
+                    1516,
+                    1215,
+                    1859,
+                    460,
+                    1859
+                ],
+                "score": 0.466
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    464,
+                    1478,
+                    893,
+                    1478,
+                    893,
+                    1509,
+                    464,
+                    1509
+                ],
+                "score": 0.439
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    298,
+                    75,
+                    858,
+                    75,
+                    858,
+                    104,
+                    298,
+                    104
+                ],
+                "score": 0.191
+            },
+            {
+                "category_id": 6,
+                "poly": [
+                    464,
+                    1478,
+                    893,
+                    1478,
+                    893,
+                    1509,
+                    464,
+                    1509
+                ],
+                "score": 0.145
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    969,
+                    1061,
+                    1025,
+                    1061,
+                    1025,
+                    1090,
+                    969,
+                    1090
+                ],
+                "score": 0.87,
+                "latex": "50 \\%"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    566,
+                    1328,
+                    611,
+                    1328,
+                    611,
+                    1356,
+                    566,
+                    1356
+                ],
+                "score": 0.86,
+                "latex": "{ \\pmb x } _ { m }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    566,
+                    1549,
+                    612,
+                    1549,
+                    612,
+                    1577,
+                    566,
+                    1577
+                ],
+                "score": 0.86,
+                "latex": "{ \\pmb x } _ { m }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1032,
+                    909,
+                    1054,
+                    909,
+                    1054,
+                    939,
+                    1032,
+                    939
+                ],
+                "score": 0.85,
+                "latex": "\\beta"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    693,
+                    1031,
+                    714,
+                    1031,
+                    714,
+                    1062,
+                    693,
+                    1062
+                ],
+                "score": 0.84,
+                "latex": "\\beta"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1377,
+                    771,
+                    1397,
+                    771,
+                    1397,
+                    801,
+                    1377,
+                    801
+                ],
+                "score": 0.82,
+                "latex": "\\beta"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    452,
+                    1062,
+                    474,
+                    1062,
+                    474,
+                    1092,
+                    452,
+                    1092
+                ],
+                "score": 0.82,
+                "latex": "\\beta"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1075,
+                    1668,
+                    1129,
+                    1668,
+                    1129,
+                    1700,
+                    1075,
+                    1700
+                ],
+                "score": 0.81,
+                "latex": "\\alpha , \\beta"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1305,
+                    774,
+                    1328,
+                    774,
+                    1328,
+                    797,
+                    1305,
+                    797
+                ],
+                "score": 0.78,
+                "latex": "\\alpha"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    463,
+                    1578,
+                    674,
+                    1578,
+                    674,
+                    1609,
+                    463,
+                    1609
+                ],
+                "score": 0.75,
+                "latex": "s e e d _ { x } \\gets h a s h ( { \\pmb x } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1329,
+                    1034,
+                    1351,
+                    1034,
+                    1351,
+                    1058,
+                    1329,
+                    1058
+                ],
+                "score": 0.75,
+                "latex": "\\alpha"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    931,
+                    1035,
+                    953,
+                    1035,
+                    953,
+                    1058,
+                    931,
+                    1058
+                ],
+                "score": 0.74,
+                "latex": "\\alpha"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    957,
+                    913,
+                    980,
+                    913,
+                    980,
+                    936,
+                    957,
+                    936
+                ],
+                "score": 0.73,
+                "latex": "\\alpha"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    463,
+                    1357,
+                    674,
+                    1357,
+                    674,
+                    1386,
+                    463,
+                    1386
+                ],
+                "score": 0.71,
+                "latex": "s e e d _ { x } \\gets h a s h ( { \\pmb x } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    546,
+                    1301,
+                    568,
+                    1301,
+                    568,
+                    1323,
+                    546,
+                    1323
+                ],
+                "score": 0.53,
+                "latex": "_ { \\textbf { \\em x } }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    789,
+                    1302,
+                    806,
+                    1302,
+                    806,
+                    1322,
+                    789,
+                    1322
+                ],
+                "score": 0.53,
+                "latex": "r"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    546,
+                    1522,
+                    568,
+                    1522,
+                    568,
+                    1543,
+                    546,
+                    1543
+                ],
+                "score": 0.51,
+                "latex": "_ { \\textbf { \\em x } }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    788,
+                    1522,
+                    806,
+                    1522,
+                    806,
+                    1543,
+                    788,
+                    1543
+                ],
+                "score": 0.49,
+                "latex": "r"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    476,
+                    1609,
+                    692,
+                    1609,
+                    692,
+                    1639,
+                    476,
+                    1639
+                ],
+                "score": 0.48,
+                "latex": "\\iota _ { x } , n _ { y } \\gets s h a p e ( { \\pmb x } )"
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    460,
+                    1356,
+                    1020,
+                    1356,
+                    1020,
+                    1453,
+                    460,
+                    1453
+                ],
+                "score": 0.44,
+                "latex": "\\begin{array} { r l } & { s e e d _ { x }  h a s h ( \\pmb { x } ) } \\\\ & { \\pmb { k }  1 - B e r n o u l l i ( s e e d _ { x } , s h a p e ( x ) , p r o b = r ) } \\\\ & { \\pmb { x } _ { m }  \\pmb { k } \\odot \\pmb { x } + ( 1 - \\pmb { k } ) \\odot N a N } \\end{array}"
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    456,
+                    1757,
+                    1011,
+                    1757,
+                    1011,
+                    1856,
+                    456,
+                    1856
+                ],
+                "score": 0.41,
+                "latex": "\\begin{array} { r l } & { ( \\boldsymbol { w } , h ) \\sim ( B e t a ( \\alpha , \\beta ) \\times n _ { x } ) , B e t a ( \\alpha , \\beta ) \\times n _ { y } ) ) } \\\\ & { \\boldsymbol { k }  r e c t \\_ m a s k ( p _ { x } , p _ { y } , w , h ) } \\\\ & { \\mathbf { \\boldsymbol { x } } _ { m }  \\boldsymbol { k } \\odot \\mathbf { \\boldsymbol { x } } + ( 1 - \\boldsymbol { k } ) \\odot N a N } \\end{array}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    459,
+                    1638,
+                    996,
+                    1638,
+                    996,
+                    1670,
+                    459,
+                    1670
+                ],
+                "score": 0.39,
+                "latex": "( p _ { x } , \\bar { p _ { y } } ) \\sim ( u n i f o r m ( 0 , n _ { x } ) , u n i f o r m ( 0 , n _ { y } ) )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    461,
+                    1640,
+                    583,
+                    1640,
+                    583,
+                    1670,
+                    461,
+                    1670
+                ],
+                "score": 0.29,
+                "latex": "( p _ { x } , p _ { y } ) \\sim"
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    458,
+                    1759,
+                    1009,
+                    1759,
+                    1009,
+                    1792,
+                    458,
+                    1792
+                ],
+                "score": 0.28,
+                "latex": "( w , h ) \\sim ( B e t a ( \\alpha , \\beta ) \\times n _ { x } ) , B e t a ( \\alpha , \\beta ) \\times n _ { y } ) )"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    321.0,
+                    1940.0,
+                    1406.0,
+                    1940.0,
+                    1406.0,
+                    1983.0,
+                    321.0,
+                    1983.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1975.0,
+                    772.0,
+                    1975.0,
+                    772.0,
+                    2005.0,
+                    294.0,
+                    2005.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    318.0,
+                    1997.0,
+                    965.0,
+                    1997.0,
+                    965.0,
+                    2044.0,
+                    318.0,
+                    2044.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    222.0,
+                    783.0,
+                    222.0,
+                    783.0,
+                    268.0,
+                    294.0,
+                    268.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    831.0,
+                    2084.0,
+                    869.0,
+                    2084.0,
+                    869.0,
+                    2123.0,
+                    831.0,
+                    2123.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    298.0,
+                    73.0,
+                    858.0,
+                    73.0,
+                    858.0,
+                    108.0,
+                    298.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    462.0,
+                    1256.0,
+                    932.0,
+                    1256.0,
+                    932.0,
+                    1287.0,
+                    462.0,
+                    1287.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    459.0,
+                    1472.0,
+                    896.0,
+                    1472.0,
+                    896.0,
+                    1516.0,
+                    459.0,
+                    1516.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    72.0,
+                    858.0,
+                    72.0,
+                    858.0,
+                    108.0,
+                    297.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    459.0,
+                    1472.0,
+                    896.0,
+                    1472.0,
+                    896.0,
+                    1516.0,
+                    459.0,
+                    1516.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    557.0,
+                    1405.0,
+                    557.0,
+                    1405.0,
+                    591.0,
+                    295.0,
+                    591.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    585.0,
+                    1408.0,
+                    585.0,
+                    1408.0,
+                    622.0,
+                    293.0,
+                    622.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    616.0,
+                    1406.0,
+                    616.0,
+                    1406.0,
+                    653.0,
+                    293.0,
+                    653.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    648.0,
+                    1406.0,
+                    648.0,
+                    1406.0,
+                    683.0,
+                    295.0,
+                    683.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    678.0,
+                    1406.0,
+                    678.0,
+                    1406.0,
+                    713.0,
+                    295.0,
+                    713.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    709.0,
+                    1407.0,
+                    709.0,
+                    1407.0,
+                    744.0,
+                    295.0,
+                    744.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    740.0,
+                    1405.0,
+                    740.0,
+                    1405.0,
+                    774.0,
+                    293.0,
+                    774.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    770.0,
+                    1304.0,
+                    770.0,
+                    1304.0,
+                    805.0,
+                    295.0,
+                    805.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1329.0,
+                    770.0,
+                    1376.0,
+                    770.0,
+                    1376.0,
+                    805.0,
+                    1329.0,
+                    805.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1398.0,
+                    770.0,
+                    1407.0,
+                    770.0,
+                    1407.0,
+                    805.0,
+                    1398.0,
+                    805.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    800.0,
+                    1403.0,
+                    800.0,
+                    1403.0,
+                    834.0,
+                    295.0,
+                    834.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    831.0,
+                    1403.0,
+                    831.0,
+                    1403.0,
+                    865.0,
+                    296.0,
+                    865.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    860.0,
+                    447.0,
+                    860.0,
+                    447.0,
+                    899.0,
+                    292.0,
+                    899.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    904.0,
+                    956.0,
+                    904.0,
+                    956.0,
+                    942.0,
+                    293.0,
+                    942.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    981.0,
+                    904.0,
+                    1031.0,
+                    904.0,
+                    1031.0,
+                    942.0,
+                    981.0,
+                    942.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1055.0,
+                    904.0,
+                    1409.0,
+                    904.0,
+                    1409.0,
+                    942.0,
+                    1055.0,
+                    942.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    935.0,
+                    1409.0,
+                    935.0,
+                    1409.0,
+                    977.0,
+                    292.0,
+                    977.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    968.0,
+                    1406.0,
+                    968.0,
+                    1406.0,
+                    1004.0,
+                    293.0,
+                    1004.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    999.0,
+                    1405.0,
+                    999.0,
+                    1405.0,
+                    1035.0,
+                    295.0,
+                    1035.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1027.0,
+                    692.0,
+                    1027.0,
+                    692.0,
+                    1065.0,
+                    293.0,
+                    1065.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    715.0,
+                    1027.0,
+                    930.0,
+                    1027.0,
+                    930.0,
+                    1065.0,
+                    715.0,
+                    1065.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    954.0,
+                    1027.0,
+                    1328.0,
+                    1027.0,
+                    1328.0,
+                    1065.0,
+                    954.0,
+                    1065.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1352.0,
+                    1027.0,
+                    1406.0,
+                    1027.0,
+                    1406.0,
+                    1065.0,
+                    1352.0,
+                    1065.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1060.0,
+                    451.0,
+                    1060.0,
+                    451.0,
+                    1096.0,
+                    295.0,
+                    1096.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    475.0,
+                    1060.0,
+                    968.0,
+                    1060.0,
+                    968.0,
+                    1096.0,
+                    475.0,
+                    1096.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1026.0,
+                    1060.0,
+                    1406.0,
+                    1060.0,
+                    1406.0,
+                    1096.0,
+                    1026.0,
+                    1096.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1088.0,
+                    1405.0,
+                    1088.0,
+                    1405.0,
+                    1126.0,
+                    293.0,
+                    1126.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1119.0,
+                    1406.0,
+                    1119.0,
+                    1406.0,
+                    1155.0,
+                    293.0,
+                    1155.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1153.0,
+                    1403.0,
+                    1153.0,
+                    1403.0,
+                    1185.0,
+                    296.0,
+                    1185.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1181.0,
+                    1405.0,
+                    1181.0,
+                    1405.0,
+                    1218.0,
+                    293.0,
+                    1218.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    295.0,
+                    1407.0,
+                    295.0,
+                    1407.0,
+                    332.0,
+                    293.0,
+                    332.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    326.0,
+                    1405.0,
+                    326.0,
+                    1405.0,
+                    361.0,
+                    293.0,
+                    361.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    357.0,
+                    1405.0,
+                    357.0,
+                    1405.0,
+                    392.0,
+                    293.0,
+                    392.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    389.0,
+                    1404.0,
+                    389.0,
+                    1404.0,
+                    422.0,
+                    294.0,
+                    422.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    417.0,
+                    1405.0,
+                    417.0,
+                    1405.0,
+                    453.0,
+                    293.0,
+                    453.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    446.0,
+                    1405.0,
+                    446.0,
+                    1405.0,
+                    484.0,
+                    293.0,
+                    484.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    480.0,
+                    1405.0,
+                    480.0,
+                    1405.0,
+                    513.0,
+                    294.0,
+                    513.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    508.0,
+                    982.0,
+                    508.0,
+                    982.0,
+                    544.0,
+                    293.0,
+                    544.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    460.0,
+                    1293.0,
+                    545.0,
+                    1293.0,
+                    545.0,
+                    1329.0,
+                    460.0,
+                    1329.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    569.0,
+                    1293.0,
+                    788.0,
+                    1293.0,
+                    788.0,
+                    1329.0,
+                    569.0,
+                    1329.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    807.0,
+                    1293.0,
+                    968.0,
+                    1293.0,
+                    968.0,
+                    1329.0,
+                    807.0,
+                    1329.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    462.0,
+                    1326.0,
+                    565.0,
+                    1326.0,
+                    565.0,
+                    1358.0,
+                    462.0,
+                    1358.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    612.0,
+                    1326.0,
+                    847.0,
+                    1326.0,
+                    847.0,
+                    1358.0,
+                    612.0,
+                    1358.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    461.0,
+                    1513.0,
+                    545.0,
+                    1513.0,
+                    545.0,
+                    1549.0,
+                    461.0,
+                    1549.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    569.0,
+                    1513.0,
+                    787.0,
+                    1513.0,
+                    787.0,
+                    1549.0,
+                    569.0,
+                    1549.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    807.0,
+                    1513.0,
+                    968.0,
+                    1513.0,
+                    968.0,
+                    1549.0,
+                    807.0,
+                    1549.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    461.0,
+                    1544.0,
+                    565.0,
+                    1544.0,
+                    565.0,
+                    1579.0,
+                    461.0,
+                    1579.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    613.0,
+                    1544.0,
+                    847.0,
+                    1544.0,
+                    847.0,
+                    1579.0,
+                    613.0,
+                    1579.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    459.0,
+                    1572.0,
+                    462.0,
+                    1572.0,
+                    462.0,
+                    1611.0,
+                    459.0,
+                    1611.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    675.0,
+                    1572.0,
+                    679.0,
+                    1572.0,
+                    679.0,
+                    1611.0,
+                    675.0,
+                    1611.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    458.0,
+                    1607.0,
+                    475.0,
+                    1607.0,
+                    475.0,
+                    1639.0,
+                    458.0,
+                    1639.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    997.0,
+                    1634.0,
+                    1001.0,
+                    1634.0,
+                    1001.0,
+                    1673.0,
+                    997.0,
+                    1673.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    459.0,
+                    1668.0,
+                    1074.0,
+                    1668.0,
+                    1074.0,
+                    1700.0,
+                    459.0,
+                    1700.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1130.0,
+                    1668.0,
+                    1197.0,
+                    1668.0,
+                    1197.0,
+                    1700.0,
+                    1130.0,
+                    1700.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    511.0,
+                    1699.0,
+                    1115.0,
+                    1699.0,
+                    1115.0,
+                    1730.0,
+                    511.0,
+                    1730.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    513.0,
+                    1730.0,
+                    705.0,
+                    1730.0,
+                    705.0,
+                    1759.0,
+                    513.0,
+                    1759.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 13,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 3,
+                "poly": [
+                    482,
+                    419,
+                    1213,
+                    419,
+                    1213,
+                    891,
+                    482,
+                    891
+                ],
+                "score": 0.966
+            },
+            {
+                "category_id": 4,
+                "poly": [
+                    298,
+                    936,
+                    1401,
+                    936,
+                    1401,
+                    1000,
+                    298,
+                    1000
+                ],
+                "score": 0.945
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    297,
+                    75,
+                    859,
+                    75,
+                    859,
+                    105,
+                    297,
+                    105
+                ],
+                "score": 0.904
+            },
+            {
+                "category_id": 3,
+                "poly": [
+                    305,
+                    1423,
+                    1388,
+                    1423,
+                    1388,
+                    1851,
+                    305,
+                    1851
+                ],
+                "score": 0.904
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    835,
+                    2088,
+                    863,
+                    2088,
+                    863,
+                    2113,
+                    835,
+                    2113
+                ],
+                "score": 0.852
+            },
+            {
+                "category_id": 6,
+                "poly": [
+                    331,
+                    1363,
+                    1355,
+                    1363,
+                    1355,
+                    1396,
+                    331,
+                    1396
+                ],
+                "score": 0.821
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    513.0,
+                    426.0,
+                    554.0,
+                    426.0,
+                    554.0,
+                    460.0,
+                    513.0,
+                    460.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    564.0,
+                    435.0,
+                    580.0,
+                    435.0,
+                    580.0,
+                    451.0,
+                    564.0,
+                    451.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    546.0,
+                    439.0,
+                    555.0,
+                    439.0,
+                    555.0,
+                    447.0,
+                    546.0,
+                    447.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    596.0,
+                    450.0,
+                    786.0,
+                    450.0,
+                    786.0,
+                    489.0,
+                    596.0,
+                    489.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    802.0,
+                    465.0,
+                    810.0,
+                    465.0,
+                    810.0,
+                    474.0,
+                    802.0,
+                    474.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    863.0,
+                    451.0,
+                    1052.0,
+                    451.0,
+                    1052.0,
+                    490.0,
+                    863.0,
+                    490.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1170.0,
+                    466.0,
+                    1177.0,
+                    466.0,
+                    1177.0,
+                    479.0,
+                    1170.0,
+                    479.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    525.0,
+                    502.0,
+                    553.0,
+                    502.0,
+                    553.0,
+                    533.0,
+                    525.0,
+                    533.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    546.0,
+                    512.0,
+                    554.0,
+                    512.0,
+                    554.0,
+                    522.0,
+                    546.0,
+                    522.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    524.0,
+                    577.0,
+                    560.0,
+                    577.0,
+                    560.0,
+                    609.0,
+                    524.0,
+                    609.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    481.0,
+                    602.0,
+                    518.0,
+                    602.0,
+                    518.0,
+                    655.0,
+                    481.0,
+                    655.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    525.0,
+                    651.0,
+                    552.0,
+                    651.0,
+                    552.0,
+                    682.0,
+                    525.0,
+                    682.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    525.0,
+                    724.0,
+                    552.0,
+                    724.0,
+                    552.0,
+                    757.0,
+                    525.0,
+                    757.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    802.0,
+                    776.0,
+                    810.0,
+                    776.0,
+                    810.0,
+                    788.0,
+                    802.0,
+                    788.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1075.0,
+                    765.0,
+                    1180.0,
+                    765.0,
+                    1180.0,
+                    794.0,
+                    1075.0,
+                    794.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    526.0,
+                    801.0,
+                    550.0,
+                    801.0,
+                    550.0,
+                    829.0,
+                    526.0,
+                    829.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    793.0,
+                    781.0,
+                    819.0,
+                    781.0,
+                    819.0,
+                    821.0,
+                    793.0,
+                    821.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1076.0,
+                    790.0,
+                    1201.0,
+                    790.0,
+                    1201.0,
+                    823.0,
+                    1076.0,
+                    823.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    554.0,
+                    834.0,
+                    598.0,
+                    834.0,
+                    598.0,
+                    869.0,
+                    554.0,
+                    869.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    672.0,
+                    834.0,
+                    717.0,
+                    834.0,
+                    717.0,
+                    868.0,
+                    672.0,
+                    868.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    791.0,
+                    833.0,
+                    835.0,
+                    833.0,
+                    835.0,
+                    868.0,
+                    791.0,
+                    868.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    907.0,
+                    834.0,
+                    953.0,
+                    834.0,
+                    953.0,
+                    868.0,
+                    907.0,
+                    868.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1026.0,
+                    835.0,
+                    1071.0,
+                    835.0,
+                    1071.0,
+                    868.0,
+                    1026.0,
+                    868.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1144.0,
+                    833.0,
+                    1189.0,
+                    833.0,
+                    1189.0,
+                    868.0,
+                    1144.0,
+                    868.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    806.0,
+                    861.0,
+                    957.0,
+                    861.0,
+                    957.0,
+                    896.0,
+                    806.0,
+                    896.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    936.0,
+                    1405.0,
+                    936.0,
+                    1405.0,
+                    972.0,
+                    294.0,
+                    972.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    970.0,
+                    433.0,
+                    970.0,
+                    433.0,
+                    1001.0,
+                    293.0,
+                    1001.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    73.0,
+                    858.0,
+                    73.0,
+                    858.0,
+                    108.0,
+                    297.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    476.0,
+                    1430.0,
+                    583.0,
+                    1430.0,
+                    583.0,
+                    1468.0,
+                    476.0,
+                    1468.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    686.0,
+                    1428.0,
+                    754.0,
+                    1428.0,
+                    754.0,
+                    1467.0,
+                    686.0,
+                    1467.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    881.0,
+                    1430.0,
+                    945.0,
+                    1430.0,
+                    945.0,
+                    1466.0,
+                    881.0,
+                    1466.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1072.0,
+                    1430.0,
+                    1137.0,
+                    1430.0,
+                    1137.0,
+                    1466.0,
+                    1072.0,
+                    1466.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1264.0,
+                    1430.0,
+                    1328.0,
+                    1430.0,
+                    1328.0,
+                    1466.0,
+                    1264.0,
+                    1466.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    1558.0,
+                    422.0,
+                    1558.0,
+                    422.0,
+                    1590.0,
+                    322.0,
+                    1590.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    317.0,
+                    1746.0,
+                    386.0,
+                    1746.0,
+                    386.0,
+                    1780.0,
+                    317.0,
+                    1780.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    831.0,
+                    2085.0,
+                    868.0,
+                    2085.0,
+                    868.0,
+                    2124.0,
+                    831.0,
+                    2124.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    340.0,
+                    1360.0,
+                    1357.0,
+                    1360.0,
+                    1357.0,
+                    1400.0,
+                    340.0,
+                    1400.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 14,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 3,
+                "poly": [
+                    474,
+                    598,
+                    1217,
+                    598,
+                    1217,
+                    1148,
+                    474,
+                    1148
+                ],
+                "score": 0.973
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    295,
+                    299,
+                    1406,
+                    299,
+                    1406,
+                    547,
+                    295,
+                    547
+                ],
+                "score": 0.973
+            },
+            {
+                "category_id": 3,
+                "poly": [
+                    480,
+                    1469,
+                    1215,
+                    1469,
+                    1215,
+                    1950,
+                    480,
+                    1950
+                ],
+                "score": 0.972
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    296,
+                    1292,
+                    1405,
+                    1292,
+                    1405,
+                    1416,
+                    296,
+                    1416
+                ],
+                "score": 0.949
+            },
+            {
+                "category_id": 4,
+                "poly": [
+                    298,
+                    1187,
+                    1401,
+                    1187,
+                    1401,
+                    1251,
+                    298,
+                    1251
+                ],
+                "score": 0.942
+            },
+            {
+                "category_id": 4,
+                "poly": [
+                    327,
+                    1992,
+                    1366,
+                    1992,
+                    1366,
+                    2026,
+                    327,
+                    2026
+                ],
+                "score": 0.925
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    298,
+                    75,
+                    858,
+                    75,
+                    858,
+                    105,
+                    298,
+                    105
+                ],
+                "score": 0.908
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    836,
+                    2088,
+                    864,
+                    2088,
+                    864,
+                    2113,
+                    836,
+                    2113
+                ],
+                "score": 0.867
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    300,
+                    226,
+                    615,
+                    226,
+                    615,
+                    261,
+                    300,
+                    261
+                ],
+                "score": 0.803
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1170,
+                    1325,
+                    1198,
+                    1325,
+                    1198,
+                    1351,
+                    1170,
+                    1351
+                ],
+                "score": 0.8,
+                "latex": "N"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1314,
+                    1994,
+                    1358,
+                    1994,
+                    1358,
+                    2025,
+                    1314,
+                    2025
+                ],
+                "score": 0.67,
+                "latex": "( N )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    566,
+                    1325,
+                    609,
+                    1325,
+                    609,
+                    1354,
+                    566,
+                    1354
+                ],
+                "score": 0.6,
+                "latex": "( N )"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    765.0,
+                    596.0,
+                    988.0,
+                    596.0,
+                    988.0,
+                    625.0,
+                    765.0,
+                    625.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    594.0,
+                    635.0,
+                    711.0,
+                    635.0,
+                    711.0,
+                    665.0,
+                    594.0,
+                    665.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    499.0,
+                    655.0,
+                    547.0,
+                    655.0,
+                    547.0,
+                    688.0,
+                    499.0,
+                    688.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    564.0,
+                    667.0,
+                    581.0,
+                    667.0,
+                    581.0,
+                    684.0,
+                    564.0,
+                    684.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    594.0,
+                    658.0,
+                    724.0,
+                    658.0,
+                    724.0,
+                    690.0,
+                    594.0,
+                    690.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    499.0,
+                    728.0,
+                    547.0,
+                    728.0,
+                    547.0,
+                    761.0,
+                    499.0,
+                    761.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    472.0,
+                    799.0,
+                    513.0,
+                    799.0,
+                    513.0,
+                    918.0,
+                    472.0,
+                    918.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    499.0,
+                    802.0,
+                    542.0,
+                    802.0,
+                    542.0,
+                    832.0,
+                    499.0,
+                    832.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    499.0,
+                    872.0,
+                    541.0,
+                    872.0,
+                    541.0,
+                    908.0,
+                    499.0,
+                    908.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    499.0,
+                    947.0,
+                    547.0,
+                    947.0,
+                    547.0,
+                    981.0,
+                    499.0,
+                    981.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    772.0,
+                    984.0,
+                    780.0,
+                    984.0,
+                    780.0,
+                    992.0,
+                    772.0,
+                    992.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    513.0,
+                    1023.0,
+                    538.0,
+                    1023.0,
+                    538.0,
+                    1051.0,
+                    513.0,
+                    1051.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    563.0,
+                    1061.0,
+                    580.0,
+                    1061.0,
+                    580.0,
+                    1078.0,
+                    563.0,
+                    1078.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    552.0,
+                    1090.0,
+                    591.0,
+                    1090.0,
+                    591.0,
+                    1127.0,
+                    552.0,
+                    1127.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    655.0,
+                    1092.0,
+                    691.0,
+                    1092.0,
+                    691.0,
+                    1126.0,
+                    655.0,
+                    1126.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    756.0,
+                    1091.0,
+                    794.0,
+                    1091.0,
+                    794.0,
+                    1125.0,
+                    756.0,
+                    1125.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    857.0,
+                    1090.0,
+                    897.0,
+                    1090.0,
+                    897.0,
+                    1127.0,
+                    857.0,
+                    1127.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    959.0,
+                    1089.0,
+                    998.0,
+                    1089.0,
+                    998.0,
+                    1126.0,
+                    959.0,
+                    1126.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1061.0,
+                    1089.0,
+                    1100.0,
+                    1089.0,
+                    1100.0,
+                    1127.0,
+                    1061.0,
+                    1127.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1164.0,
+                    1092.0,
+                    1201.0,
+                    1092.0,
+                    1201.0,
+                    1126.0,
+                    1164.0,
+                    1126.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    784.0,
+                    1115.0,
+                    969.0,
+                    1115.0,
+                    969.0,
+                    1153.0,
+                    784.0,
+                    1153.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1173.0,
+                    653.5,
+                    1177.0,
+                    653.5,
+                    1177.0,
+                    689.5,
+                    1173.0,
+                    689.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    752.0,
+                    1468.0,
+                    1008.0,
+                    1468.0,
+                    1008.0,
+                    1501.0,
+                    752.0,
+                    1501.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    506.0,
+                    1511.0,
+                    553.0,
+                    1511.0,
+                    553.0,
+                    1545.0,
+                    506.0,
+                    1545.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    506.0,
+                    1564.0,
+                    555.0,
+                    1564.0,
+                    555.0,
+                    1599.0,
+                    506.0,
+                    1599.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    501.0,
+                    1612.0,
+                    557.0,
+                    1612.0,
+                    557.0,
+                    1688.0,
+                    501.0,
+                    1688.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    474.0,
+                    1617.0,
+                    516.0,
+                    1617.0,
+                    516.0,
+                    1775.0,
+                    474.0,
+                    1775.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    507.0,
+                    1721.0,
+                    547.0,
+                    1721.0,
+                    547.0,
+                    1759.0,
+                    507.0,
+                    1759.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    503.0,
+                    1777.0,
+                    555.0,
+                    1777.0,
+                    555.0,
+                    1811.0,
+                    503.0,
+                    1811.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    607.0,
+                    1768.0,
+                    724.0,
+                    1768.0,
+                    724.0,
+                    1800.0,
+                    607.0,
+                    1800.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    607.0,
+                    1792.0,
+                    714.0,
+                    1792.0,
+                    714.0,
+                    1828.0,
+                    607.0,
+                    1828.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    608.0,
+                    1821.0,
+                    712.0,
+                    1821.0,
+                    712.0,
+                    1852.0,
+                    608.0,
+                    1852.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    505.0,
+                    1831.0,
+                    557.0,
+                    1831.0,
+                    557.0,
+                    1867.0,
+                    505.0,
+                    1867.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    608.0,
+                    1847.0,
+                    703.0,
+                    1847.0,
+                    703.0,
+                    1880.0,
+                    608.0,
+                    1880.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    561.0,
+                    1892.0,
+                    599.0,
+                    1892.0,
+                    599.0,
+                    1923.0,
+                    561.0,
+                    1923.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    635.0,
+                    1890.0,
+                    676.0,
+                    1890.0,
+                    676.0,
+                    1925.0,
+                    635.0,
+                    1925.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    710.0,
+                    1892.0,
+                    748.0,
+                    1892.0,
+                    748.0,
+                    1924.0,
+                    710.0,
+                    1924.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    785.0,
+                    1892.0,
+                    823.0,
+                    1892.0,
+                    823.0,
+                    1923.0,
+                    785.0,
+                    1923.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    861.0,
+                    1892.0,
+                    900.0,
+                    1892.0,
+                    900.0,
+                    1923.0,
+                    861.0,
+                    1923.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    935.0,
+                    1892.0,
+                    973.0,
+                    1892.0,
+                    973.0,
+                    1923.0,
+                    935.0,
+                    1923.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1011.0,
+                    1893.0,
+                    1048.0,
+                    1893.0,
+                    1048.0,
+                    1923.0,
+                    1011.0,
+                    1923.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1085.0,
+                    1893.0,
+                    1123.0,
+                    1893.0,
+                    1123.0,
+                    1923.0,
+                    1085.0,
+                    1923.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1161.0,
+                    1892.0,
+                    1197.0,
+                    1892.0,
+                    1197.0,
+                    1924.0,
+                    1161.0,
+                    1924.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    781.0,
+                    1914.0,
+                    977.0,
+                    1914.0,
+                    977.0,
+                    1955.0,
+                    781.0,
+                    1955.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1165.25,
+                    1837.0,
+                    1174.25,
+                    1837.0,
+                    1174.25,
+                    1846.0,
+                    1165.25,
+                    1846.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1187.0,
+                    1406.0,
+                    1187.0,
+                    1406.0,
+                    1223.0,
+                    296.0,
+                    1223.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1218.0,
+                    1388.0,
+                    1218.0,
+                    1388.0,
+                    1254.0,
+                    294.0,
+                    1254.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    329.0,
+                    1990.0,
+                    1313.0,
+                    1990.0,
+                    1313.0,
+                    2030.0,
+                    329.0,
+                    2030.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1359.0,
+                    1990.0,
+                    1369.0,
+                    1990.0,
+                    1369.0,
+                    2030.0,
+                    1359.0,
+                    2030.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    298.0,
+                    73.0,
+                    858.0,
+                    73.0,
+                    858.0,
+                    108.0,
+                    298.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    831.0,
+                    2085.0,
+                    868.0,
+                    2085.0,
+                    868.0,
+                    2124.0,
+                    831.0,
+                    2124.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    223.0,
+                    621.0,
+                    223.0,
+                    621.0,
+                    267.0,
+                    294.0,
+                    267.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    302.0,
+                    1406.0,
+                    302.0,
+                    1406.0,
+                    336.0,
+                    295.0,
+                    336.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    332.0,
+                    1406.0,
+                    332.0,
+                    1406.0,
+                    366.0,
+                    294.0,
+                    366.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    360.0,
+                    1406.0,
+                    360.0,
+                    1406.0,
+                    399.0,
+                    294.0,
+                    399.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    393.0,
+                    1406.0,
+                    393.0,
+                    1406.0,
+                    427.0,
+                    295.0,
+                    427.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    424.0,
+                    1404.0,
+                    424.0,
+                    1404.0,
+                    458.0,
+                    297.0,
+                    458.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    453.0,
+                    1405.0,
+                    453.0,
+                    1405.0,
+                    488.0,
+                    293.0,
+                    488.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    482.0,
+                    1406.0,
+                    482.0,
+                    1406.0,
+                    521.0,
+                    292.0,
+                    521.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    513.0,
+                    382.0,
+                    513.0,
+                    382.0,
+                    554.0,
+                    292.0,
+                    554.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1290.0,
+                    1406.0,
+                    1290.0,
+                    1406.0,
+                    1330.0,
+                    293.0,
+                    1330.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1324.0,
+                    565.0,
+                    1324.0,
+                    565.0,
+                    1356.0,
+                    296.0,
+                    1356.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    610.0,
+                    1324.0,
+                    1169.0,
+                    1324.0,
+                    1169.0,
+                    1356.0,
+                    610.0,
+                    1356.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1199.0,
+                    1324.0,
+                    1405.0,
+                    1324.0,
+                    1405.0,
+                    1356.0,
+                    1199.0,
+                    1356.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1356.0,
+                    1405.0,
+                    1356.0,
+                    1405.0,
+                    1387.0,
+                    293.0,
+                    1387.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1386.0,
+                    640.0,
+                    1386.0,
+                    640.0,
+                    1418.0,
+                    295.0,
+                    1418.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 15,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    229,
+                    1404,
+                    229,
+                    1404,
+                    475,
+                    297,
+                    475
+                ],
+                "score": 0.981
+            },
+            {
+                "category_id": 5,
+                "poly": [
+                    406,
+                    591,
+                    1291,
+                    591,
+                    1291,
+                    830,
+                    406,
+                    830
+                ],
+                "score": 0.98,
+                "html": "<table><tr><td></td><td colspan=\"6\">Accuracy at Missing Rate (%)</td></tr><tr><td>Method</td><td>20%</td><td>40%</td><td>60%</td><td>70%</td><td>80%</td><td>90%</td></tr><tr><td>GI (N=128)</td><td>84.0</td><td>76.9</td><td>66.1</td><td>59.1</td><td>46.0</td><td>32.1</td></tr><tr><td>GI (N=1)</td><td>83.6</td><td>75.7</td><td>65.1</td><td>56.7</td><td>42.8</td><td>29.4</td></tr><tr><td>% difference (normalized)</td><td>0.5</td><td>1.6</td><td>1.5</td><td>4.1</td><td>6.9</td><td>8.4</td></tr></table>"
+            },
+            {
+                "category_id": 6,
+                "poly": [
+                    291,
+                    503,
+                    1402,
+                    503,
+                    1402,
+                    568,
+                    291,
+                    568
+                ],
+                "score": 0.886
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    835,
+                    2088,
+                    863,
+                    2088,
+                    863,
+                    2112,
+                    835,
+                    2112
+                ],
+                "score": 0.844
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    298,
+                    75,
+                    858,
+                    75,
+                    858,
+                    104,
+                    298,
+                    104
+                ],
+                "score": 0.812
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    298,
+                    75,
+                    858,
+                    75,
+                    858,
+                    105,
+                    298,
+                    105
+                ],
+                "score": 0.112
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    883,
+                    292,
+                    939,
+                    292,
+                    939,
+                    321,
+                    883,
+                    321
+                ],
+                "score": 0.87,
+                "latex": "90 \\%"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    672,
+                    293,
+                    728,
+                    293,
+                    728,
+                    321,
+                    672,
+                    321
+                ],
+                "score": 0.85,
+                "latex": "20 \\%"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1002,
+                    505,
+                    1087,
+                    505,
+                    1087,
+                    534,
+                    1002,
+                    534
+                ],
+                "score": 0.78,
+                "latex": "\\mathrm { ( N = 1 2 8 ) }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    670,
+                    263,
+                    732,
+                    263,
+                    732,
+                    291,
+                    670,
+                    291
+                ],
+                "score": 0.73,
+                "latex": "( \\mathrm { N } { = } 1 )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1336,
+                    505,
+                    1396,
+                    505,
+                    1396,
+                    534,
+                    1336,
+                    534
+                ],
+                "score": 0.64,
+                "latex": "\\left( \\mathrm { N } { = } 1 \\right)"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    471,
+                    697,
+                    563,
+                    697,
+                    563,
+                    729,
+                    471,
+                    729
+                ],
+                "score": 0.61,
+                "latex": "\\bf ( N { = } 1 2 8 )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1155,
+                    612,
+                    1200,
+                    612,
+                    1200,
+                    642,
+                    1155,
+                    642
+                ],
+                "score": 0.48,
+                "latex": "( \\% )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    468,
+                    740,
+                    536,
+                    740,
+                    536,
+                    771,
+                    468,
+                    771
+                ],
+                "score": 0.35,
+                "latex": "\\mathbf { \\rho } ( \\mathbf { N } { = } \\mathbf { 1 } )"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    499.0,
+                    1001.0,
+                    499.0,
+                    1001.0,
+                    541.0,
+                    293.0,
+                    541.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1088.0,
+                    499.0,
+                    1335.0,
+                    499.0,
+                    1335.0,
+                    541.0,
+                    1088.0,
+                    541.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1397.0,
+                    499.0,
+                    1405.0,
+                    499.0,
+                    1405.0,
+                    541.0,
+                    1397.0,
+                    541.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    534.0,
+                    752.0,
+                    534.0,
+                    752.0,
+                    570.0,
+                    293.0,
+                    570.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    831.0,
+                    2085.0,
+                    868.0,
+                    2085.0,
+                    868.0,
+                    2124.0,
+                    831.0,
+                    2124.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    72.0,
+                    858.0,
+                    72.0,
+                    858.0,
+                    108.0,
+                    297.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    298.0,
+                    73.0,
+                    858.0,
+                    73.0,
+                    858.0,
+                    108.0,
+                    298.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    231.0,
+                    1405.0,
+                    231.0,
+                    1405.0,
+                    264.0,
+                    295.0,
+                    264.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    258.0,
+                    669.0,
+                    258.0,
+                    669.0,
+                    298.0,
+                    292.0,
+                    298.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    733.0,
+                    258.0,
+                    1406.0,
+                    258.0,
+                    1406.0,
+                    298.0,
+                    733.0,
+                    298.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    292.0,
+                    671.0,
+                    292.0,
+                    671.0,
+                    326.0,
+                    294.0,
+                    326.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    729.0,
+                    292.0,
+                    882.0,
+                    292.0,
+                    882.0,
+                    326.0,
+                    729.0,
+                    326.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    940.0,
+                    292.0,
+                    1405.0,
+                    292.0,
+                    1405.0,
+                    326.0,
+                    940.0,
+                    326.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    318.0,
+                    1408.0,
+                    318.0,
+                    1408.0,
+                    358.0,
+                    292.0,
+                    358.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    352.0,
+                    1407.0,
+                    352.0,
+                    1407.0,
+                    387.0,
+                    292.0,
+                    387.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    382.0,
+                    1405.0,
+                    382.0,
+                    1405.0,
+                    417.0,
+                    292.0,
+                    417.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    414.0,
+                    1405.0,
+                    414.0,
+                    1405.0,
+                    448.0,
+                    295.0,
+                    448.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    447.0,
+                    447.0,
+                    447.0,
+                    447.0,
+                    478.0,
+                    296.0,
+                    478.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 16,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    295,
+                    1405,
+                    295,
+                    1405,
+                    510,
+                    297,
+                    510
+                ],
+                "score": 0.98
+            },
+            {
+                "category_id": 6,
+                "poly": [
+                    298,
+                    538,
+                    1401,
+                    538,
+                    1401,
+                    602,
+                    298,
+                    602
+                ],
+                "score": 0.923
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    298,
+                    75,
+                    858,
+                    75,
+                    858,
+                    105,
+                    298,
+                    105
+                ],
+                "score": 0.906
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    836,
+                    2088,
+                    864,
+                    2088,
+                    864,
+                    2113,
+                    836,
+                    2113
+                ],
+                "score": 0.851
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    299,
+                    226,
+                    668,
+                    226,
+                    668,
+                    262,
+                    299,
+                    262
+                ],
+                "score": 0.834
+            },
+            {
+                "category_id": 3,
+                "poly": [
+                    305,
+                    616,
+                    1386,
+                    616,
+                    1386,
+                    1791,
+                    305,
+                    1791
+                ],
+                "score": 0.729
+            },
+            {
+                "category_id": 5,
+                "poly": [
+                    305,
+                    616,
+                    1386,
+                    616,
+                    1386,
+                    1791,
+                    305,
+                    1791
+                ],
+                "score": 0.353,
+                "html": "<table><tr><td></td><td>Original</td><td>Mask</td><td colspan=\"3\"> Samples</td></tr><tr><td>GI</td><td></td><td></td><td># （20</td><td>#</td><td>中</td></tr><tr><td>MisG</td><td>开</td><td></td><td></td><td>S</td><td></td></tr><tr><td>GI</td><td>S</td><td></td><td></td><td></td><td>国</td></tr><tr><td></td><td></td><td></td><td></td><td></td><td></td></tr><tr><td>GI</td><td>正</td><td></td><td>水</td><td>K</td><td>正</td></tr><tr><td></td><td></td><td></td><td>下</td><td>水</td><td>上</td></tr></table>"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    537.0,
+                    1405.0,
+                    537.0,
+                    1405.0,
+                    574.0,
+                    293.0,
+                    574.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    570.0,
+                    1297.0,
+                    570.0,
+                    1297.0,
+                    603.0,
+                    293.0,
+                    603.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    298.0,
+                    73.0,
+                    858.0,
+                    73.0,
+                    858.0,
+                    108.0,
+                    298.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    831.0,
+                    2085.0,
+                    868.0,
+                    2085.0,
+                    868.0,
+                    2122.0,
+                    831.0,
+                    2122.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    221.0,
+                    672.0,
+                    221.0,
+                    672.0,
+                    267.0,
+                    292.0,
+                    267.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    476.0,
+                    637.0,
+                    588.0,
+                    637.0,
+                    588.0,
+                    677.0,
+                    476.0,
+                    677.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    686.0,
+                    638.0,
+                    762.0,
+                    638.0,
+                    762.0,
+                    675.0,
+                    686.0,
+                    675.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1052.0,
+                    635.0,
+                    1161.0,
+                    635.0,
+                    1161.0,
+                    679.0,
+                    1052.0,
+                    679.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    430.0,
+                    676.0,
+                    639.0,
+                    676.0,
+                    639.0,
+                    1048.0,
+                    430.0,
+                    1048.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1189.0,
+                    682.0,
+                    1397.0,
+                    682.0,
+                    1397.0,
+                    1053.0,
+                    1189.0,
+                    1053.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1029.0,
+                    709.0,
+                    1182.0,
+                    709.0,
+                    1182.0,
+                    850.0,
+                    1029.0,
+                    850.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    842.0,
+                    711.0,
+                    996.0,
+                    711.0,
+                    996.0,
+                    850.0,
+                    842.0,
+                    850.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    320.0,
+                    764.0,
+                    356.0,
+                    764.0,
+                    356.0,
+                    794.0,
+                    320.0,
+                    794.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1188.0,
+                    1047.0,
+                    1404.0,
+                    1047.0,
+                    1404.0,
+                    1506.0,
+                    1188.0,
+                    1506.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    422.0,
+                    1050.0,
+                    643.0,
+                    1050.0,
+                    643.0,
+                    1440.0,
+                    422.0,
+                    1440.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    1146.0,
+                    353.0,
+                    1146.0,
+                    353.0,
+                    1172.0,
+                    322.0,
+                    1172.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1182.0,
+                    1425.0,
+                    1400.0,
+                    1425.0,
+                    1400.0,
+                    1805.0,
+                    1182.0,
+                    1805.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    433.0,
+                    1432.0,
+                    643.0,
+                    1432.0,
+                    643.0,
+                    1809.0,
+                    433.0,
+                    1809.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    820.0,
+                    1447.0,
+                    1018.0,
+                    1447.0,
+                    1018.0,
+                    1675.0,
+                    820.0,
+                    1675.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1030.0,
+                    1455.0,
+                    1205.0,
+                    1455.0,
+                    1205.0,
+                    1639.0,
+                    1030.0,
+                    1639.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    1525.0,
+                    351.0,
+                    1525.0,
+                    351.0,
+                    1553.0,
+                    322.0,
+                    1553.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    823.0,
+                    1588.0,
+                    1034.0,
+                    1588.0,
+                    1034.0,
+                    1787.0,
+                    823.0,
+                    1787.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1021.0,
+                    1598.0,
+                    1216.0,
+                    1598.0,
+                    1216.0,
+                    1783.0,
+                    1021.0,
+                    1783.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    876.0,
+                    1749.0,
+                    939.0,
+                    1749.0,
+                    939.0,
+                    1772.0,
+                    876.0,
+                    1772.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    294.0,
+                    1408.0,
+                    294.0,
+                    1408.0,
+                    332.0,
+                    294.0,
+                    332.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    326.0,
+                    1406.0,
+                    326.0,
+                    1406.0,
+                    361.0,
+                    292.0,
+                    361.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    359.0,
+                    1403.0,
+                    359.0,
+                    1403.0,
+                    390.0,
+                    296.0,
+                    390.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    388.0,
+                    1406.0,
+                    388.0,
+                    1406.0,
+                    422.0,
+                    294.0,
+                    422.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    418.0,
+                    1405.0,
+                    418.0,
+                    1405.0,
+                    453.0,
+                    294.0,
+                    453.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    450.0,
+                    1405.0,
+                    450.0,
+                    1405.0,
+                    481.0,
+                    296.0,
+                    481.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    477.0,
+                    628.0,
+                    477.0,
+                    628.0,
+                    515.0,
+                    294.0,
+                    515.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 17,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 2,
+                "poly": [
+                    297,
+                    75,
+                    859,
+                    75,
+                    859,
+                    106,
+                    297,
+                    106
+                ],
+                "score": 0.906
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    835,
+                    2088,
+                    864,
+                    2088,
+                    864,
+                    2113,
+                    835,
+                    2113
+                ],
+                "score": 0.849
+            },
+            {
+                "category_id": 6,
+                "poly": [
+                    296,
+                    480,
+                    1402,
+                    480,
+                    1402,
+                    544,
+                    296,
+                    544
+                ],
+                "score": 0.813
+            },
+            {
+                "category_id": 3,
+                "poly": [
+                    304,
+                    558,
+                    1386,
+                    558,
+                    1386,
+                    1765,
+                    304,
+                    1765
+                ],
+                "score": 0.789
+            },
+            {
+                "category_id": 5,
+                "poly": [
+                    304,
+                    558,
+                    1386,
+                    558,
+                    1386,
+                    1765,
+                    304,
+                    1765
+                ],
+                "score": 0.277,
+                "html": "<table><tr><td>Original</td><td>Mask</td><td></td><td></td><td>Samples</td><td></td></tr><tr><td></td><td>B</td><td>. .</td><td></td><td></td><td>毛</td></tr><tr><td>GI</td><td></td><td></td><td></td><td></td><td>#</td></tr><tr><td></td><td>2</td><td>　</td><td>3</td><td>R</td><td>R</td></tr><tr><td></td><td></td><td></td><td>诊</td><td>Z</td><td></td></tr><tr><td></td><td>Z</td><td></td><td>沃</td><td>诊</td><td>多</td></tr></table>"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    72.0,
+                    858.0,
+                    72.0,
+                    858.0,
+                    108.0,
+                    297.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    831.0,
+                    2085.0,
+                    869.0,
+                    2085.0,
+                    869.0,
+                    2125.0,
+                    831.0,
+                    2125.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    480.0,
+                    1406.0,
+                    480.0,
+                    1406.0,
+                    516.0,
+                    295.0,
+                    516.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    511.0,
+                    1228.0,
+                    511.0,
+                    1228.0,
+                    547.0,
+                    295.0,
+                    547.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    452.0,
+                    579.0,
+                    563.0,
+                    579.0,
+                    563.0,
+                    624.0,
+                    452.0,
+                    624.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    664.0,
+                    580.0,
+                    744.0,
+                    580.0,
+                    744.0,
+                    617.0,
+                    664.0,
+                    617.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1042.0,
+                    577.0,
+                    1154.0,
+                    577.0,
+                    1154.0,
+                    622.0,
+                    1042.0,
+                    622.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    597.0,
+                    626.0,
+                    818.0,
+                    626.0,
+                    818.0,
+                    955.0,
+                    597.0,
+                    955.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1198.0,
+                    628.0,
+                    1391.0,
+                    628.0,
+                    1391.0,
+                    999.0,
+                    1198.0,
+                    999.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    409.0,
+                    640.0,
+                    611.0,
+                    640.0,
+                    611.0,
+                    875.0,
+                    409.0,
+                    875.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    325.0,
+                    715.0,
+                    347.0,
+                    715.0,
+                    347.0,
+                    737.0,
+                    325.0,
+                    737.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    595.0,
+                    999.0,
+                    847.0,
+                    999.0,
+                    847.0,
+                    1403.0,
+                    595.0,
+                    1403.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    402.0,
+                    1009.0,
+                    634.0,
+                    1009.0,
+                    634.0,
+                    1397.0,
+                    402.0,
+                    1397.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1200.0,
+                    1018.0,
+                    1390.0,
+                    1018.0,
+                    1390.0,
+                    1387.0,
+                    1200.0,
+                    1387.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    817.0,
+                    1044.0,
+                    1009.0,
+                    1044.0,
+                    1009.0,
+                    1224.0,
+                    817.0,
+                    1224.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1012.0,
+                    1062.0,
+                    1191.0,
+                    1062.0,
+                    1191.0,
+                    1197.0,
+                    1012.0,
+                    1197.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    829.0,
+                    1175.0,
+                    1004.0,
+                    1175.0,
+                    1004.0,
+                    1374.0,
+                    829.0,
+                    1374.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1032.0,
+                    1181.0,
+                    1209.0,
+                    1181.0,
+                    1209.0,
+                    1387.0,
+                    1032.0,
+                    1387.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    890.0,
+                    1190.0,
+                    937.0,
+                    1190.0,
+                    937.0,
+                    1204.0,
+                    890.0,
+                    1204.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    484.0,
+                    1311.0,
+                    1435.0,
+                    1311.0,
+                    1435.0,
+                    1814.0,
+                    484.0,
+                    1814.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    402.0,
+                    1401.0,
+                    641.0,
+                    1401.0,
+                    641.0,
+                    1781.0,
+                    402.0,
+                    1781.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    325.0,
+                    1498.0,
+                    351.0,
+                    1498.0,
+                    351.0,
+                    1518.0,
+                    325.0,
+                    1518.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    622.0,
+                    1556.0,
+                    831.0,
+                    1556.0,
+                    831.0,
+                    1763.0,
+                    622.0,
+                    1763.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    787.0,
+                    1575.0,
+                    1218.0,
+                    1575.0,
+                    1218.0,
+                    1748.0,
+                    787.0,
+                    1748.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    783.0,
+                    1635.0,
+                    827.0,
+                    1635.0,
+                    827.0,
+                    1684.0,
+                    783.0,
+                    1684.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    434.0,
+                    813.5,
+                    605.0,
+                    813.5,
+                    605.0,
+                    947.5,
+                    434.0,
+                    947.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1133.75,
+                    1620.0,
+                    1259.75,
+                    1620.0,
+                    1259.75,
+                    1684.0,
+                    1133.75,
+                    1684.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 18,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 5,
+                "poly": [
+                    301,
+                    1251,
+                    1400,
+                    1251,
+                    1400,
+                    1752,
+                    301,
+                    1752
+                ],
+                "score": 0.985,
+                "html": "<table><tr><td colspan=\"2\"></td><td colspan=\"4\">RMSE at Missing Rate (%)</td></tr><tr><td>Dataset</td><td>Method</td><td>10%</td><td>20%</td><td>30%</td><td>40%</td></tr><tr><td rowspan=\"4\">Landsat (Dua &amp; Graff,2017)</td><td>GI</td><td>0.040 (±0.005)</td><td>0.067 (±0.007)</td><td>0.076 (±0.020)</td><td>0.136 (±0.002)</td></tr><tr><td>MisGAN</td><td>0.068 (±0.001)</td><td>0.096 (±0.001)</td><td>0.118 (±0.001)</td><td>0.136 (±0.001)</td></tr><tr><td>GAIN</td><td>0.018 (±0.001)</td><td>0.024 (±0.001)</td><td>0.030 (±0.001)</td><td>0.037 (±0.001)</td></tr><tr><td>DAE</td><td>0.020 (±0.001)</td><td>0.031 (±0.001)</td><td>0.041 (±0.001)</td><td>0.052 (±0.001)</td></tr><tr><td rowspan=\"4\">MIT-BIH(Moody &amp; Mark,2001)</td><td>GI</td><td>0.038 (±0.001)</td><td>0.060 (±0.004)</td><td>0.071 (±0.002)</td><td>0.095 (±0.002)</td></tr><tr><td>MisGAN</td><td>0.073 (±0.007)</td><td>0.092 (±0.002)</td><td>0.115 (±0.003)</td><td>0.111 (±0.001)</td></tr><tr><td>GAIN</td><td>0.032 (±0.008)</td><td>0.046 (±0.001)</td><td>0.055 (±0.004)</td><td>0.067 (±0.007)</td></tr><tr><td>DAE</td><td>0.029 (±0.001)</td><td>0.048 (±0.008)</td><td>0.061 (±0.009)</td><td>0.068 (±0.003)</td></tr><tr><td rowspan=\"4\">Diabetes (Kachuee et al., 2019)</td><td>GI</td><td>0.080 (±0.002)</td><td>0.118 (±0.008)</td><td>0.149 (±0.020)</td><td>0.189 (±0.009)</td></tr><tr><td>MisGAN</td><td>0.082 (±0.004)</td><td>0.111 (±0.002)</td><td>0.133 (±0.001)</td><td>0.151 (±0.001)</td></tr><tr><td>GAIN</td><td>0.064 (±0.001)</td><td>0.092 (±0.001)</td><td>0.119 (±0.001)</td><td>0.140 (±0.001)</td></tr><tr><td>DAE</td><td>0.065 (±0.001)</td><td>0.093 (±0.001)</td><td>0.118 (±0.001)</td><td>0.143 (±0.001)</td></tr></table>"
+            },
+            {
+                "category_id": 5,
+                "poly": [
+                    301,
+                    828,
+                    1399,
+                    828,
+                    1399,
+                    1112,
+                    301,
+                    1112
+                ],
+                "score": 0.982,
+                "html": "<table><tr><td></td><td colspan=\"5\">RMSE at Missing Rate (%)</td><td></td></tr><tr><td></td><td colspan=\"5\">MCAR Uniform</td><td>MCAR Rect.</td></tr><tr><td>Method</td><td>20%</td><td>40%</td><td>60%</td><td>20%</td><td>40%</td><td>60%</td></tr><tr><td>GI</td><td>0.026 (±0.003)</td><td>0.057 (±0.008)</td><td>0.090 (±0.006)</td><td>0.097 (±0.02)</td><td>0.148 (±0.001)</td><td>0.660 (±0.010)</td></tr><tr><td>MisGAN</td><td>0.079 (±0.001)</td><td>0.161 (±0.001)</td><td>0.257 (±0.002)</td><td>0.106 (±0.005)</td><td>0.158 (±0.004)</td><td>0.250 (±0.001)</td></tr><tr><td>GAIN</td><td>0.027 (±0.003)</td><td>0.045 (±0.001)</td><td>0.072 (±0.005)</td><td>0.340 (±0.047)</td><td>0.511 (±0.001)</td><td>0.660 (±0.010)</td></tr><tr><td>DAE</td><td>0.036 (±0.001)</td><td>0.075 (±0.002)</td><td>0.121 (±0.005)</td><td>0.116 (±0.007)</td><td>0.160 (±0.001)</td><td>0.233 (±0.029)</td></tr></table>"
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    295,
+                    1405,
+                    295,
+                    1405,
+                    542,
+                    297,
+                    542
+                ],
+                "score": 0.979
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    555,
+                    1404,
+                    555,
+                    1404,
+                    710,
+                    298,
+                    710
+                ],
+                "score": 0.968
+            },
+            {
+                "category_id": 6,
+                "poly": [
+                    298,
+                    736,
+                    1398,
+                    736,
+                    1398,
+                    801,
+                    298,
+                    801
+                ],
+                "score": 0.911
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    298,
+                    75,
+                    858,
+                    75,
+                    858,
+                    105,
+                    298,
+                    105
+                ],
+                "score": 0.906
+            },
+            {
+                "category_id": 6,
+                "poly": [
+                    294,
+                    1159,
+                    1401,
+                    1159,
+                    1401,
+                    1223,
+                    294,
+                    1223
+                ],
+                "score": 0.902
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    834,
+                    2088,
+                    866,
+                    2088,
+                    866,
+                    2113,
+                    834,
+                    2113
+                ],
+                "score": 0.86
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    298,
+                    226,
+                    869,
+                    226,
+                    869,
+                    262,
+                    298,
+                    262
+                ],
+                "score": 0.851
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    734.0,
+                    1404.0,
+                    734.0,
+                    1404.0,
+                    772.0,
+                    293.0,
+                    772.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    768.0,
+                    366.0,
+                    768.0,
+                    366.0,
+                    803.0,
+                    292.0,
+                    803.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    298.0,
+                    73.0,
+                    858.0,
+                    73.0,
+                    858.0,
+                    108.0,
+                    298.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1156.0,
+                    1407.0,
+                    1156.0,
+                    1407.0,
+                    1196.0,
+                    293.0,
+                    1196.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1190.0,
+                    558.0,
+                    1190.0,
+                    558.0,
+                    1226.0,
+                    295.0,
+                    1226.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    828.0,
+                    2083.0,
+                    872.0,
+                    2083.0,
+                    872.0,
+                    2123.0,
+                    828.0,
+                    2123.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    223.0,
+                    873.0,
+                    223.0,
+                    873.0,
+                    267.0,
+                    291.0,
+                    267.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    294.0,
+                    1406.0,
+                    294.0,
+                    1406.0,
+                    331.0,
+                    292.0,
+                    331.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    327.0,
+                    1405.0,
+                    327.0,
+                    1405.0,
+                    361.0,
+                    296.0,
+                    361.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    358.0,
+                    1406.0,
+                    358.0,
+                    1406.0,
+                    391.0,
+                    294.0,
+                    391.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    388.0,
+                    1406.0,
+                    388.0,
+                    1406.0,
+                    422.0,
+                    295.0,
+                    422.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    416.0,
+                    1406.0,
+                    416.0,
+                    1406.0,
+                    456.0,
+                    292.0,
+                    456.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    449.0,
+                    1408.0,
+                    449.0,
+                    1408.0,
+                    482.0,
+                    295.0,
+                    482.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    478.0,
+                    1407.0,
+                    478.0,
+                    1407.0,
+                    513.0,
+                    294.0,
+                    513.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    510.0,
+                    1099.0,
+                    510.0,
+                    1099.0,
+                    544.0,
+                    292.0,
+                    544.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    555.0,
+                    1410.0,
+                    555.0,
+                    1410.0,
+                    593.0,
+                    294.0,
+                    593.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    583.0,
+                    1406.0,
+                    583.0,
+                    1406.0,
+                    622.0,
+                    293.0,
+                    622.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    617.0,
+                    1404.0,
+                    617.0,
+                    1404.0,
+                    652.0,
+                    293.0,
+                    652.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    644.0,
+                    1406.0,
+                    644.0,
+                    1406.0,
+                    683.0,
+                    293.0,
+                    683.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    673.0,
+                    452.0,
+                    673.0,
+                    452.0,
+                    715.0,
+                    292.0,
+                    715.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 19,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 5,
+                "poly": [
+                    478,
+                    735,
+                    1221,
+                    735,
+                    1221,
+                    1100,
+                    478,
+                    1100
+                ],
+                "score": 0.983,
+                "html": "<table><tr><td></td><td colspan=\"2\">Accuracy (%)</td></tr><tr><td>Noise STD</td><td>MCAR Uniform (40%) GI GAIN</td><td>MCAR Rect. (40%) GI GAIN</td></tr><tr><td>0.0</td><td>87.1 86.0</td><td>76.9 73.6</td></tr><tr><td>0.0125</td><td>87.3 86.3</td><td>76.8 73.3</td></tr><tr><td>0.025</td><td>86.5 86.6</td><td>76.7 73.2</td></tr><tr><td>0.05</td><td>85.6 84.7</td><td>73.7 72.4</td></tr><tr><td>0.1</td><td>82.0 80.6</td><td>68.7 67.0</td></tr></table>"
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    296,
+                    1403,
+                    296,
+                    1403,
+                    480,
+                    298,
+                    480
+                ],
+                "score": 0.979
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    494,
+                    1404,
+                    494,
+                    1404,
+                    649,
+                    299,
+                    649
+                ],
+                "score": 0.979
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    298,
+                    75,
+                    858,
+                    75,
+                    858,
+                    105,
+                    298,
+                    105
+                ],
+                "score": 0.906
+            },
+            {
+                "category_id": 6,
+                "poly": [
+                    293,
+                    676,
+                    1393,
+                    676,
+                    1393,
+                    711,
+                    293,
+                    711
+                ],
+                "score": 0.894
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    301,
+                    225,
+                    767,
+                    225,
+                    767,
+                    262,
+                    301,
+                    262
+                ],
+                "score": 0.894
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    834,
+                    2088,
+                    863,
+                    2088,
+                    863,
+                    2112,
+                    834,
+                    2112
+                ],
+                "score": 0.839
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1048,
+                    449,
+                    1102,
+                    449,
+                    1102,
+                    477,
+                    1048,
+                    477
+                ],
+                "score": 0.87,
+                "latex": "40 \\%"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    893,
+                    678,
+                    948,
+                    678,
+                    948,
+                    707,
+                    893,
+                    707
+                ],
+                "score": 0.86,
+                "latex": "40 \\%"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    298.0,
+                    73.0,
+                    858.0,
+                    73.0,
+                    858.0,
+                    108.0,
+                    298.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    299.0,
+                    672.0,
+                    892.0,
+                    672.0,
+                    892.0,
+                    715.0,
+                    299.0,
+                    715.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    949.0,
+                    672.0,
+                    1397.0,
+                    672.0,
+                    1397.0,
+                    715.0,
+                    949.0,
+                    715.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    223.0,
+                    772.0,
+                    223.0,
+                    772.0,
+                    267.0,
+                    295.0,
+                    267.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    830.0,
+                    2084.0,
+                    868.0,
+                    2084.0,
+                    868.0,
+                    2124.0,
+                    830.0,
+                    2124.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    295.0,
+                    1405.0,
+                    295.0,
+                    1405.0,
+                    331.0,
+                    294.0,
+                    331.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    326.0,
+                    1405.0,
+                    326.0,
+                    1405.0,
+                    360.0,
+                    293.0,
+                    360.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    352.0,
+                    1407.0,
+                    352.0,
+                    1407.0,
+                    396.0,
+                    291.0,
+                    396.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    385.0,
+                    1405.0,
+                    385.0,
+                    1405.0,
+                    425.0,
+                    293.0,
+                    425.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    418.0,
+                    1405.0,
+                    418.0,
+                    1405.0,
+                    454.0,
+                    294.0,
+                    454.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    446.0,
+                    1047.0,
+                    446.0,
+                    1047.0,
+                    484.0,
+                    294.0,
+                    484.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1103.0,
+                    446.0,
+                    1115.0,
+                    446.0,
+                    1115.0,
+                    484.0,
+                    1103.0,
+                    484.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    494.0,
+                    1404.0,
+                    494.0,
+                    1404.0,
+                    532.0,
+                    295.0,
+                    532.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    524.0,
+                    1406.0,
+                    524.0,
+                    1406.0,
+                    561.0,
+                    292.0,
+                    561.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    556.0,
+                    1406.0,
+                    556.0,
+                    1406.0,
+                    591.0,
+                    292.0,
+                    591.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    583.0,
+                    1406.0,
+                    583.0,
+                    1406.0,
+                    622.0,
+                    293.0,
+                    622.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    618.0,
+                    795.0,
+                    618.0,
+                    795.0,
+                    651.0,
+                    295.0,
+                    651.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 20,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    296,
+                    295,
+                    1407,
+                    295,
+                    1407,
+                    480,
+                    296,
+                    480
+                ],
+                "score": 0.979
+            },
+            {
+                "category_id": 3,
+                "poly": [
+                    363,
+                    513,
+                    1357,
+                    513,
+                    1357,
+                    1742,
+                    363,
+                    1742
+                ],
+                "score": 0.975
+            },
+            {
+                "category_id": 4,
+                "poly": [
+                    296,
+                    1762,
+                    1407,
+                    1762,
+                    1407,
+                    1858,
+                    296,
+                    1858
+                ],
+                "score": 0.962
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    298,
+                    225,
+                    838,
+                    225,
+                    838,
+                    263,
+                    298,
+                    263
+                ],
+                "score": 0.919
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    298,
+                    75,
+                    858,
+                    75,
+                    858,
+                    105,
+                    298,
+                    105
+                ],
+                "score": 0.91
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    834,
+                    2088,
+                    865,
+                    2088,
+                    865,
+                    2113,
+                    834,
+                    2113
+                ],
+                "score": 0.862
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    298,
+                    389,
+                    317,
+                    389,
+                    317,
+                    416,
+                    298,
+                    416
+                ],
+                "score": 0.75,
+                "latex": "\\lambda"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    582,
+                    1827,
+                    601,
+                    1827,
+                    601,
+                    1853,
+                    582,
+                    1853
+                ],
+                "score": 0.65,
+                "latex": "\\lambda"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    561,
+                    420,
+                    581,
+                    420,
+                    581,
+                    446,
+                    561,
+                    446
+                ],
+                "score": 0.62,
+                "latex": "\\lambda"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    879,
+                    526,
+                    970,
+                    526,
+                    970,
+                    552,
+                    879,
+                    552
+                ],
+                "score": 0.44,
+                "latex": "_ { ( \\sigma = 0 . 1 3 ) }"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    779.0,
+                    520.0,
+                    878.0,
+                    520.0,
+                    878.0,
+                    556.0,
+                    779.0,
+                    556.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    971.0,
+                    520.0,
+                    976.0,
+                    520.0,
+                    976.0,
+                    556.0,
+                    971.0,
+                    556.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    552.0,
+                    549.0,
+                    618.0,
+                    549.0,
+                    618.0,
+                    583.0,
+                    552.0,
+                    583.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    552.0,
+                    590.0,
+                    609.0,
+                    590.0,
+                    609.0,
+                    621.0,
+                    552.0,
+                    621.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    554.0,
+                    629.0,
+                    610.0,
+                    629.0,
+                    610.0,
+                    662.0,
+                    554.0,
+                    662.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    554.0,
+                    668.0,
+                    610.0,
+                    668.0,
+                    610.0,
+                    701.0,
+                    554.0,
+                    701.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    530.0,
+                    689.0,
+                    563.0,
+                    689.0,
+                    563.0,
+                    750.0,
+                    530.0,
+                    750.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    554.0,
+                    707.0,
+                    608.0,
+                    707.0,
+                    608.0,
+                    741.0,
+                    554.0,
+                    741.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    555.0,
+                    748.0,
+                    608.0,
+                    748.0,
+                    608.0,
+                    780.0,
+                    555.0,
+                    780.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    552.0,
+                    787.0,
+                    610.0,
+                    787.0,
+                    610.0,
+                    822.0,
+                    552.0,
+                    822.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    555.0,
+                    828.0,
+                    608.0,
+                    828.0,
+                    608.0,
+                    860.0,
+                    555.0,
+                    860.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    554.0,
+                    866.0,
+                    615.0,
+                    866.0,
+                    615.0,
+                    899.0,
+                    554.0,
+                    899.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    585.0,
+                    884.0,
+                    628.0,
+                    884.0,
+                    628.0,
+                    914.0,
+                    585.0,
+                    914.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    696.0,
+                    882.0,
+                    737.0,
+                    882.0,
+                    737.0,
+                    914.0,
+                    696.0,
+                    914.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    804.0,
+                    882.0,
+                    846.0,
+                    882.0,
+                    846.0,
+                    914.0,
+                    804.0,
+                    914.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    913.0,
+                    881.0,
+                    954.0,
+                    881.0,
+                    954.0,
+                    914.0,
+                    913.0,
+                    914.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1021.0,
+                    884.0,
+                    1062.0,
+                    884.0,
+                    1062.0,
+                    914.0,
+                    1021.0,
+                    914.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1131.0,
+                    884.0,
+                    1172.0,
+                    884.0,
+                    1172.0,
+                    914.0,
+                    1131.0,
+                    914.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    865.0,
+                    906.0,
+                    893.0,
+                    906.0,
+                    893.0,
+                    936.0,
+                    865.0,
+                    936.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    826.0,
+                    959.0,
+                    875.0,
+                    959.0,
+                    875.0,
+                    1003.0,
+                    826.0,
+                    1003.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    511.0,
+                    1021.0,
+                    775.0,
+                    1021.0,
+                    775.0,
+                    1074.0,
+                    511.0,
+                    1074.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    982.0,
+                    1017.0,
+                    1305.0,
+                    1017.0,
+                    1305.0,
+                    1074.0,
+                    982.0,
+                    1074.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    902.0,
+                    1087.0,
+                    967.0,
+                    1087.0,
+                    967.0,
+                    1134.0,
+                    902.0,
+                    1134.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    365.0,
+                    1111.0,
+                    472.0,
+                    1111.0,
+                    472.0,
+                    1215.0,
+                    365.0,
+                    1215.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    868.0,
+                    1111.0,
+                    916.0,
+                    1111.0,
+                    916.0,
+                    1212.0,
+                    868.0,
+                    1212.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    902.0,
+                    1161.0,
+                    965.0,
+                    1161.0,
+                    965.0,
+                    1201.0,
+                    902.0,
+                    1201.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    401.0,
+                    1230.0,
+                    463.0,
+                    1230.0,
+                    463.0,
+                    1270.0,
+                    401.0,
+                    1270.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    901.0,
+                    1229.0,
+                    975.0,
+                    1229.0,
+                    975.0,
+                    1271.0,
+                    901.0,
+                    1271.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    427.0,
+                    1255.0,
+                    504.0,
+                    1255.0,
+                    504.0,
+                    1298.0,
+                    427.0,
+                    1298.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    517.0,
+                    1253.0,
+                    593.0,
+                    1253.0,
+                    593.0,
+                    1298.0,
+                    517.0,
+                    1298.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    607.0,
+                    1253.0,
+                    685.0,
+                    1253.0,
+                    685.0,
+                    1298.0,
+                    607.0,
+                    1298.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    697.0,
+                    1255.0,
+                    774.0,
+                    1255.0,
+                    774.0,
+                    1298.0,
+                    697.0,
+                    1298.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    785.0,
+                    1255.0,
+                    861.0,
+                    1255.0,
+                    861.0,
+                    1298.0,
+                    785.0,
+                    1298.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    926.0,
+                    1257.0,
+                    1003.0,
+                    1257.0,
+                    1003.0,
+                    1297.0,
+                    926.0,
+                    1297.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1020.0,
+                    1253.0,
+                    1095.0,
+                    1253.0,
+                    1095.0,
+                    1298.0,
+                    1020.0,
+                    1298.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1110.0,
+                    1253.0,
+                    1187.0,
+                    1253.0,
+                    1187.0,
+                    1298.0,
+                    1110.0,
+                    1298.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1198.0,
+                    1253.0,
+                    1276.0,
+                    1253.0,
+                    1276.0,
+                    1298.0,
+                    1198.0,
+                    1298.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1288.0,
+                    1255.0,
+                    1362.0,
+                    1255.0,
+                    1362.0,
+                    1298.0,
+                    1288.0,
+                    1298.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    623.0,
+                    1289.0,
+                    669.0,
+                    1289.0,
+                    669.0,
+                    1337.0,
+                    623.0,
+                    1337.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1125.0,
+                    1288.0,
+                    1170.0,
+                    1288.0,
+                    1170.0,
+                    1337.0,
+                    1125.0,
+                    1337.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    480.0,
+                    1352.0,
+                    804.0,
+                    1352.0,
+                    804.0,
+                    1407.0,
+                    480.0,
+                    1407.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    982.0,
+                    1353.0,
+                    1306.0,
+                    1353.0,
+                    1306.0,
+                    1409.0,
+                    982.0,
+                    1409.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    398.0,
+                    1393.0,
+                    480.0,
+                    1393.0,
+                    480.0,
+                    1442.0,
+                    398.0,
+                    1442.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    363.0,
+                    1446.0,
+                    470.0,
+                    1446.0,
+                    470.0,
+                    1550.0,
+                    363.0,
+                    1550.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    867.0,
+                    1430.0,
+                    967.0,
+                    1430.0,
+                    967.0,
+                    1550.0,
+                    867.0,
+                    1550.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    397.0,
+                    1564.0,
+                    467.0,
+                    1564.0,
+                    467.0,
+                    1604.0,
+                    397.0,
+                    1604.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    901.0,
+                    1561.0,
+                    968.0,
+                    1561.0,
+                    968.0,
+                    1606.0,
+                    901.0,
+                    1606.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    427.0,
+                    1590.0,
+                    506.0,
+                    1590.0,
+                    506.0,
+                    1633.0,
+                    427.0,
+                    1633.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    515.0,
+                    1588.0,
+                    595.0,
+                    1588.0,
+                    595.0,
+                    1631.0,
+                    515.0,
+                    1631.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    604.0,
+                    1588.0,
+                    863.0,
+                    1588.0,
+                    863.0,
+                    1633.0,
+                    604.0,
+                    1633.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    930.0,
+                    1590.0,
+                    1008.0,
+                    1590.0,
+                    1008.0,
+                    1633.0,
+                    930.0,
+                    1633.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1019.0,
+                    1588.0,
+                    1098.0,
+                    1588.0,
+                    1098.0,
+                    1631.0,
+                    1019.0,
+                    1631.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1109.0,
+                    1588.0,
+                    1187.0,
+                    1588.0,
+                    1187.0,
+                    1631.0,
+                    1109.0,
+                    1631.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1193.0,
+                    1584.0,
+                    1365.0,
+                    1584.0,
+                    1365.0,
+                    1637.0,
+                    1193.0,
+                    1637.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    625.0,
+                    1626.0,
+                    666.0,
+                    1626.0,
+                    666.0,
+                    1669.0,
+                    625.0,
+                    1669.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1128.0,
+                    1626.0,
+                    1169.0,
+                    1626.0,
+                    1169.0,
+                    1669.0,
+                    1128.0,
+                    1669.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    824.0,
+                    1701.0,
+                    875.0,
+                    1701.0,
+                    875.0,
+                    1746.0,
+                    824.0,
+                    1746.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1763.0,
+                    1404.0,
+                    1763.0,
+                    1404.0,
+                    1798.0,
+                    295.0,
+                    1798.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1795.0,
+                    1404.0,
+                    1795.0,
+                    1404.0,
+                    1830.0,
+                    295.0,
+                    1830.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1822.0,
+                    581.0,
+                    1822.0,
+                    581.0,
+                    1860.0,
+                    293.0,
+                    1860.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    602.0,
+                    1822.0,
+                    621.0,
+                    1822.0,
+                    621.0,
+                    1860.0,
+                    602.0,
+                    1860.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    225.0,
+                    841.0,
+                    225.0,
+                    841.0,
+                    266.0,
+                    295.0,
+                    266.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    298.0,
+                    73.0,
+                    858.0,
+                    73.0,
+                    858.0,
+                    108.0,
+                    298.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    829.0,
+                    2084.0,
+                    871.0,
+                    2084.0,
+                    871.0,
+                    2125.0,
+                    829.0,
+                    2125.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    297.0,
+                    1404.0,
+                    297.0,
+                    1404.0,
+                    329.0,
+                    295.0,
+                    329.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    328.0,
+                    1404.0,
+                    328.0,
+                    1404.0,
+                    360.0,
+                    295.0,
+                    360.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    353.0,
+                    1408.0,
+                    353.0,
+                    1408.0,
+                    394.0,
+                    293.0,
+                    394.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    318.0,
+                    389.0,
+                    1404.0,
+                    389.0,
+                    1404.0,
+                    421.0,
+                    318.0,
+                    421.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    419.0,
+                    560.0,
+                    419.0,
+                    560.0,
+                    451.0,
+                    295.0,
+                    451.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    582.0,
+                    419.0,
+                    1404.0,
+                    419.0,
+                    1404.0,
+                    451.0,
+                    582.0,
+                    451.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    449.0,
+                    706.0,
+                    449.0,
+                    706.0,
+                    480.0,
+                    295.0,
+                    480.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 21,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    295,
+                    1406,
+                    295,
+                    1406,
+                    602,
+                    297,
+                    602
+                ],
+                "score": 0.983
+            },
+            {
+                "category_id": 3,
+                "poly": [
+                    480,
+                    1253,
+                    1214,
+                    1253,
+                    1214,
+                    1733,
+                    480,
+                    1733
+                ],
+                "score": 0.968
+            },
+            {
+                "category_id": 3,
+                "poly": [
+                    479,
+                    643,
+                    1214,
+                    643,
+                    1214,
+                    1122,
+                    479,
+                    1122
+                ],
+                "score": 0.968
+            },
+            {
+                "category_id": 4,
+                "poly": [
+                    289,
+                    1164,
+                    1404,
+                    1164,
+                    1404,
+                    1230,
+                    289,
+                    1230
+                ],
+                "score": 0.939
+            },
+            {
+                "category_id": 4,
+                "poly": [
+                    295,
+                    1776,
+                    1402,
+                    1776,
+                    1402,
+                    1840,
+                    295,
+                    1840
+                ],
+                "score": 0.937
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    298,
+                    75,
+                    858,
+                    75,
+                    858,
+                    105,
+                    298,
+                    105
+                ],
+                "score": 0.905
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    297,
+                    225,
+                    1018,
+                    225,
+                    1018,
+                    262,
+                    297,
+                    262
+                ],
+                "score": 0.904
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    835,
+                    2088,
+                    865,
+                    2088,
+                    865,
+                    2112,
+                    835,
+                    2112
+                ],
+                "score": 0.864
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    767.0,
+                    1252.0,
+                    1004.0,
+                    1252.0,
+                    1004.0,
+                    1285.0,
+                    767.0,
+                    1285.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    505.0,
+                    1307.0,
+                    559.0,
+                    1307.0,
+                    559.0,
+                    1343.0,
+                    505.0,
+                    1343.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1011.0,
+                    1298.0,
+                    1184.0,
+                    1298.0,
+                    1184.0,
+                    1330.0,
+                    1011.0,
+                    1330.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1010.0,
+                    1328.0,
+                    1198.0,
+                    1328.0,
+                    1198.0,
+                    1363.0,
+                    1010.0,
+                    1363.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    506.0,
+                    1355.0,
+                    561.0,
+                    1355.0,
+                    561.0,
+                    1392.0,
+                    506.0,
+                    1392.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1010.0,
+                    1360.0,
+                    1186.0,
+                    1360.0,
+                    1186.0,
+                    1395.0,
+                    1010.0,
+                    1395.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    506.0,
+                    1405.0,
+                    559.0,
+                    1405.0,
+                    559.0,
+                    1444.0,
+                    506.0,
+                    1444.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    478.0,
+                    1417.0,
+                    514.0,
+                    1417.0,
+                    514.0,
+                    1541.0,
+                    478.0,
+                    1541.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    517.0,
+                    1455.0,
+                    559.0,
+                    1455.0,
+                    559.0,
+                    1499.0,
+                    517.0,
+                    1499.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    518.0,
+                    1556.0,
+                    558.0,
+                    1556.0,
+                    558.0,
+                    1595.0,
+                    518.0,
+                    1595.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    518.0,
+                    1606.0,
+                    559.0,
+                    1606.0,
+                    559.0,
+                    1643.0,
+                    518.0,
+                    1643.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    532.0,
+                    1659.0,
+                    556.0,
+                    1659.0,
+                    556.0,
+                    1685.0,
+                    532.0,
+                    1685.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    595.0,
+                    1674.0,
+                    631.0,
+                    1674.0,
+                    631.0,
+                    1707.0,
+                    595.0,
+                    1707.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    683.0,
+                    1676.0,
+                    721.0,
+                    1676.0,
+                    721.0,
+                    1708.0,
+                    683.0,
+                    1708.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    771.0,
+                    1675.0,
+                    807.0,
+                    1675.0,
+                    807.0,
+                    1708.0,
+                    771.0,
+                    1708.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    850.0,
+                    1672.0,
+                    903.0,
+                    1672.0,
+                    903.0,
+                    1708.0,
+                    850.0,
+                    1708.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    938.0,
+                    1672.0,
+                    990.0,
+                    1672.0,
+                    990.0,
+                    1709.0,
+                    938.0,
+                    1709.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1026.0,
+                    1673.0,
+                    1079.0,
+                    1673.0,
+                    1079.0,
+                    1709.0,
+                    1026.0,
+                    1709.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1114.0,
+                    1672.0,
+                    1165.0,
+                    1672.0,
+                    1165.0,
+                    1709.0,
+                    1114.0,
+                    1709.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    832.0,
+                    1700.0,
+                    939.0,
+                    1700.0,
+                    939.0,
+                    1738.0,
+                    832.0,
+                    1738.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    528.75,
+                    1502.0,
+                    547.75,
+                    1502.0,
+                    547.75,
+                    1540.5,
+                    528.75,
+                    1540.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    749.0,
+                    637.0,
+                    1022.0,
+                    637.0,
+                    1022.0,
+                    675.0,
+                    749.0,
+                    675.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1021.0,
+                    682.0,
+                    1201.0,
+                    682.0,
+                    1201.0,
+                    722.0,
+                    1021.0,
+                    722.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    503.0,
+                    710.0,
+                    562.0,
+                    710.0,
+                    562.0,
+                    750.0,
+                    503.0,
+                    750.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1021.0,
+                    715.0,
+                    1201.0,
+                    715.0,
+                    1201.0,
+                    754.0,
+                    1021.0,
+                    754.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1022.0,
+                    747.0,
+                    1201.0,
+                    747.0,
+                    1201.0,
+                    786.0,
+                    1022.0,
+                    786.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    502.0,
+                    775.0,
+                    571.0,
+                    775.0,
+                    571.0,
+                    813.0,
+                    502.0,
+                    813.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    475.0,
+                    803.0,
+                    519.0,
+                    803.0,
+                    519.0,
+                    935.0,
+                    475.0,
+                    935.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    502.0,
+                    839.0,
+                    567.0,
+                    839.0,
+                    567.0,
+                    878.0,
+                    502.0,
+                    878.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    502.0,
+                    904.0,
+                    566.0,
+                    904.0,
+                    566.0,
+                    943.0,
+                    502.0,
+                    943.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    515.0,
+                    970.0,
+                    569.0,
+                    970.0,
+                    569.0,
+                    1007.0,
+                    515.0,
+                    1007.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    529.0,
+                    1036.0,
+                    569.0,
+                    1036.0,
+                    569.0,
+                    1068.0,
+                    529.0,
+                    1068.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    593.0,
+                    1062.0,
+                    634.0,
+                    1062.0,
+                    634.0,
+                    1099.0,
+                    593.0,
+                    1099.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    681.0,
+                    1062.0,
+                    722.0,
+                    1062.0,
+                    722.0,
+                    1098.0,
+                    681.0,
+                    1098.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    769.0,
+                    1062.0,
+                    809.0,
+                    1062.0,
+                    809.0,
+                    1098.0,
+                    769.0,
+                    1098.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    850.0,
+                    1061.0,
+                    904.0,
+                    1061.0,
+                    904.0,
+                    1098.0,
+                    850.0,
+                    1098.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    937.0,
+                    1059.0,
+                    992.0,
+                    1059.0,
+                    992.0,
+                    1099.0,
+                    937.0,
+                    1099.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1025.0,
+                    1061.0,
+                    1078.0,
+                    1061.0,
+                    1078.0,
+                    1098.0,
+                    1025.0,
+                    1098.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1112.0,
+                    1061.0,
+                    1166.0,
+                    1061.0,
+                    1166.0,
+                    1099.0,
+                    1112.0,
+                    1099.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    831.0,
+                    1088.0,
+                    941.0,
+                    1088.0,
+                    941.0,
+                    1128.0,
+                    831.0,
+                    1128.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1165.0,
+                    1406.0,
+                    1165.0,
+                    1406.0,
+                    1201.0,
+                    295.0,
+                    1201.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1194.0,
+                    415.0,
+                    1194.0,
+                    415.0,
+                    1231.0,
+                    293.0,
+                    1231.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1776.0,
+                    1405.0,
+                    1776.0,
+                    1405.0,
+                    1812.0,
+                    294.0,
+                    1812.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1806.0,
+                    415.0,
+                    1806.0,
+                    415.0,
+                    1843.0,
+                    293.0,
+                    1843.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    298.0,
+                    73.0,
+                    858.0,
+                    73.0,
+                    858.0,
+                    108.0,
+                    298.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    290.0,
+                    222.0,
+                    1024.0,
+                    222.0,
+                    1024.0,
+                    268.0,
+                    290.0,
+                    268.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    830.0,
+                    2084.0,
+                    869.0,
+                    2084.0,
+                    869.0,
+                    2122.0,
+                    830.0,
+                    2122.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    293.0,
+                    1409.0,
+                    293.0,
+                    1409.0,
+                    332.0,
+                    294.0,
+                    332.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    324.0,
+                    1404.0,
+                    324.0,
+                    1404.0,
+                    361.0,
+                    294.0,
+                    361.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    357.0,
+                    1408.0,
+                    357.0,
+                    1408.0,
+                    393.0,
+                    292.0,
+                    393.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    387.0,
+                    1406.0,
+                    387.0,
+                    1406.0,
+                    422.0,
+                    294.0,
+                    422.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    419.0,
+                    1408.0,
+                    419.0,
+                    1408.0,
+                    455.0,
+                    295.0,
+                    455.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    449.0,
+                    1409.0,
+                    449.0,
+                    1409.0,
+                    484.0,
+                    294.0,
+                    484.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    479.0,
+                    1406.0,
+                    479.0,
+                    1406.0,
+                    513.0,
+                    294.0,
+                    513.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    510.0,
+                    1406.0,
+                    510.0,
+                    1406.0,
+                    545.0,
+                    295.0,
+                    545.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    536.0,
+                    1407.0,
+                    536.0,
+                    1407.0,
+                    577.0,
+                    294.0,
+                    577.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    568.0,
+                    504.0,
+                    568.0,
+                    504.0,
+                    604.0,
+                    294.0,
+                    604.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 22,
+            "width": 1700,
+            "height": 2200
+        }
+    }
+]
\ No newline at end of file
diff --git a/parse/train/Bklfsi0cKm/Bklfsi0cKm.md b/parse/train/Bklfsi0cKm/Bklfsi0cKm.md
new file mode 100644
index 0000000000000000000000000000000000000000..8486fb7b41b7145b4a74f004331d205ef2264926
--- /dev/null
+++ b/parse/train/Bklfsi0cKm/Bklfsi0cKm.md
@@ -0,0 +1,436 @@
+# DEEP CONVOLUTIONAL NETWORKS AS SHALLOW GAUSSIAN PROCESSES
+
+Adrià Garriga-Alonso department of Engineering University of Cambridge ag919@cam.ac.uk
+
+Laurence Aitchison   
+Department of Engineering   
+University of Cambridge   
+laurence.aitchison@gmail.com
+
+# ABSTRACT
+
+We show that the output of a (residual) convolutional neural network (CNN) with an appropriate prior over the weights and biases is a Gaussian process (GP) in the limit of infinitely many convolutional filters, extending similar results for dense networks. For a CNN, the equivalent kernel can be computed exactly and, unlike “deep kernels”, has very few parameters: only the hyperparameters of the original CNN. Further, we show that this kernel has two properties that allow it to be computed efficiently; the cost of evaluating the kernel for a pair of images is similar to a single forward pass through the original CNN with only one filter per layer. The kernel equivalent to a 32-layer ResNet obtains $0 . 8 4 \%$ classification error on MNIST, a new record for GPs with a comparable number of parameters.
+
+# 1 INTRODUCTION
+
+Convolutional Neural Networks (CNNs) have powerful pattern-recognition capabilities that have recently given dramatic improvements in important tasks such as image classification (Krizhevsky et al., 2012). However, as CNNs are increasingly being applied in real-world, safety-critical domains, their vulnerability to adversarial examples (Szegedy et al., 2013; Kurakin et al., 2016), and their poor uncertainty estimates are becoming increasingly problematic. Bayesian inference is a theoretically principled and demonstrably successful (Snoek et al., 2012; Deisenroth & Rasmussen, 2011) framework for learning in the face of uncertainty, which may also help to address the problems of adversarial examples (Gal & Smith, 2018). Unfortunately, Bayesian inference in CNNs is extremely difficult due to the very large number of parameters, requiring highly approximate factorised variational approximations (Blundell et al., 2015; Gal & Ghahramani, 2015), or requiring the storage (Lakshminarayanan et al., 2017) of large numbers of posterior samples (Welling & Teh, 2011; Mandt et al., 2017).
+
+Other methods such as those based on Gaussian Processes (GPs) are more amenable to Bayesian inference, allowing us to compute the posterior uncertainty exactly (Rasmussen & Williams, 2006). This raises the question of whether it might be possible to combine the pattern-recognition capabilities of CNNs with exact probabilistic computations in GPs. Two such approaches exist in the literature. First, deep convolutional kernels (Wilson et al., 2016) parameterise a GP kernel using the weights and biases of a CNN, which is used to embed the input images into some latent space before computing their similarity. The CNN parameters of the resulting kernel then have to be optimised by gradient descent. However, the large number of kernel parameters in the CNN reintroduces the risk of overconfidence and overfitting. To avoid this risk, we need to infer a posterior over the CNN kernel parameters, which is as difficult as directly inferring a posterior over the parameters of the original CNN. Second, it is possible to define a convolutional GP (van der Wilk et al., 2017) or a deep convolutional GP (Kumar et al., 2018) by defining a GP that takes an image patch as input, and using that GP as a component in a larger CNN-like system. However, inference in such systems is very computationally expensive, at least without the use of potentially severe variational approximations (van der Wilk et al., 2017).
+
+An alternative approach is suggested by the underlying connection between Bayesian neural networks (NNs) and GPs. In particular, Neal (1996) showed that the function defined by a single-layer fully-connected NN with infinitely many hidden units, and random independent zero-mean weights and biases is equivalent to a GP, implying that we can do exact Bayesian inference in such a NN by working with the equivalent GP. Recently, this result was extended to arbitrarily deep fullyconnected NNs with infinitely many hidden units at each layer (Lee et al., 2017; Matthews et al., 2018a). However, these fully-connected networks are rarely used in practice, as they are unable to exploit important properties of images such as translational invariance, raising the question of whether state-of-the-art architectures such as CNNs (LeCun et al., 1990) and ResNets (He et al., 2016a) have equivalent GP representations. Here, we answer in the affirmative, giving the GP kernel corresponding to arbitrarily deep CNNs and to (convolutional) residual neural networks (He et al., 2016a). In this case, if each hidden layer has an infinite number of convolutional filters, the network prior is equivalent to a GP.
+
+Furthermore, we show that two properties of the GP kernel induced by a CNN allow it to be computed very efficiently. First, in previous work it was necessary to compute the covariance matrix for the output of a single convolutional filter applied at all possible locations within a single image (van der Wilk et al., 2017), which was prohibitively computationally expensive. In contrast, under our prior, the downstream weights are independent with zero-mean, which decorrelates the contribution from each location, and implies that it is necessary only to track the patch variances, and not their covariances. Second, while it is still necessary to compute the variance of the output of a convolutional filter applied at all locations within the image, the specific structure of the kernel induced by the CNN means that the variance at every location can be computed simultaneously and efficiently as a convolution.
+
+Finally, we empirically demonstrate the performance increase coming from adding translationinvariant structure to the GP prior. Without computing any gradients, and without augmenting the training set (e.g. using translations), we obtain $0 . 8 4 \%$ error rate on the MNIST classification benchmark, setting a new record for nonparametric GP-based methods.
+
+# 2 GP BEHAVIOUR IN A CNN
+
+For clarity of exposition, we will treat the case of a 2D convolutional NN. The result applies straightforwardly to $n \mathbf { D }$ convolutions, dilated convolutions and upconvolutions (“deconvolutions”), since they can be represented as linear transformations with tied coefficients (see Fig. 1).
+
+# 2.1 A 2D CONVOLUTIONAL NETWORK PRIOR
+
+The network takes an arbitrary input image $\mathbf { X }$ of height $H ^ { ( 0 ) }$ and width $D ^ { ( 0 ) }$ , as a $C ^ { ( 0 ) } \times ( H ^ { ( 0 ) } D ^ { ( 0 ) } )$ real matrix. Each row, which we denote $\mathbf { x } _ { 1 } , \mathbf { x } _ { 2 } , \ldots , \mathbf { x } _ { C ^ { ( 0 ) } }$ , corresponds to a channel of the image (e.g. $C ^ { ( 0 ) } = 3$ for RGB), flattened to form a vector. The first activations $\mathbf { A } ^ { ( 1 ) } ( \mathbf { X } )$ are a linear transformation of the inputs. For $i \in \{ 1 , \ldots , C ^ { ( 1 ) } \}$ :
+
+$$
+\mathbf { a } _ { i } ^ { ( 1 ) } ( \mathbf { X } ) : = b _ { i } ^ { ( 1 ) } \mathbf { 1 } + \sum _ { j = 1 } ^ { C ^ { ( 0 ) } } \mathbf { W } _ { i , j } ^ { ( 1 ) } \mathbf { x } _ { j } \ .
+$$
+
+We consider a network with $L$ hidden layers. The other activations of the network, from $\mathbf { A } ^ { ( 2 ) } ( \mathbf { X } )$ up to $\mathbf { A } ^ { ( L + 1 ) } ( \mathbf { X } )$ , are defined recursively:
+
+$$
+\mathbf { a } _ { i } ^ { ( \ell + 1 ) } ( \mathbf { X } ) : = b _ { i } ^ { ( \ell + 1 ) } \mathbf { 1 } + \sum _ { j = 1 } ^ { C ^ { ( \ell ) } } \mathbf { W } _ { i , j } ^ { ( \ell + 1 ) } \phi \left( \mathbf { a } _ { j } ^ { ( \ell ) } ( \mathbf { X } ) \right) .
+$$
+
+![](images/b76e3f5dce59ce270069488ab3ea5f08d6e69a5c74ce89a25358ac556f83a901.jpg)  
+Figure 1: The 2D convolution U(i, $\mathbf { U } _ { i , j } ^ { ( 0 ) } * \mathbf { x } _ { j }$ 0)j ∗ xj as the dot product W(0)i,j xj . The blank elements of W(0)i,j are zeros. The µth row of W(0)i,j corresponds to applying the filter to the $\mu$ th convolutional patch of the channel $\mathbf { x } _ { j }$ .
+
+The activations $\mathbf { A } ^ { ( \ell ) } ( \mathbf { X } )$ are $C ^ { ( \ell ) } \times ( H ^ { ( \ell ) } D ^ { ( \ell ) } )$ matrices. Each row $\mathbf { a } _ { i } ^ { ( \ell + 1 ) }$ represents the flattened $j$ th channel of the image that results from applying a convolutional filter to $\phi ( \mathbf { A } ^ { ( \ell ) } ( \mathbf { X } ) )$ .
+
+and The structure of the pseudo-weight matrices W(\`+1)i,j , depends on the archite and biases or a convolu $b _ { i } ^ { ( \ell + 1 ) }$ , for layer, $i \in \{ 1 , \ldots , C ^ { ( \ell + 1 ) } \}$ $j \in \{ 1 , \ldots , C ^ { ( \ell ) } \}$ $\mathbf { W } _ { i , j } ^ { ( \ell + 1 ) }$ xj represents applying the convolutional filter U(\`+1)i,j to the $j$ th channel. Thus, the elements of each row of where it do W(\`+1) are: 0 where the filter does not apply and the corresponding element of strated in Fig. 1. $\mathbf { U } _ { i , j } ^ { ( \ell + 1 ) }$
+
+The outputs of the network are the last activations, $\mathbf { A } ^ { ( L + 1 ) } ( \mathbf { X } )$ . In the classification or regression setting, the outputs are not spatially extended, so we have $H ^ { ( L + 1 ) } = D ^ { ( L + 1 ) } = 1$ , which is equivalent to a fully-connected output layer. In this case, the pseudo-weights W(L+1)i,j only have one row, and the activations $\mathbf { a } _ { i } ^ { ( L + 1 ) }$ are single-element vectors.
+
+Finally, we define the prior distribution over functions by making the filters $\mathbf { U } _ { i , j } ^ { ( \ell ) }$ and biases $b _ { i } ^ { ( \ell ) }$ be independent Gaussian random variables (RVs). For each layer $\ell$ , channels $i , j$ and locations within the filter $x , y$ :
+
+$$
+U _ { i , j , x , y } ^ { ( \ell ) } \sim \mathcal { N } \left( 0 , \sigma _ { w } ^ { 2 } / C ^ { ( \ell ) } \right) , \qquad b _ { i } ^ { ( \ell ) } \sim \mathcal { N } \left( 0 , \sigma _ { b } ^ { 2 } \right) .
+$$
+
+Note that, to keep the activation variance constant, the weight variance is divided by the number of input channels. The weight variance can also be divided by the number of elements of the filter, which makes it equivalent to the NN weight initialisation scheme introduced by He et al. (2016a).
+
+# 2.2 ARGUMENT FOR GP BEHAVIOUR
+
+We follow the proofs by Lee et al. (2017) and Matthews et al. (2018a) to show that the output of the CNN described in the previous section, $\mathbf { A } ^ { ( L + 1 ) }$ , defines a GP indexed by the inputs, $\mathbf { X }$ . Their proof (Lee et al., 2017) proceeds by applying the multivariate Central Limit Theorem (CLT) to each layer in sequence, i.e. taking the limit as $N ^ { ( 1 ) }  \infty$ , then $N ^ { ( 2 ) }  \infty$ etc, where $N ^ { ( \ell ) }$ is the number of hidden units in layer $\ell$ . By analogy, we sequentially apply the multivariate CLT by taking the limit as the number of channels goes to infinity, i.e. $\dot { C ^ { ( 1 ) } } \stackrel { - } {  } \infty$ , then $C ^ { ( 2 ) }  \infty$ etc. While this is the simplest approach to taking the limits, other potentially more realistic approaches also exist (Matthews et al., 2018a).
+
+The fundamental quantity we consider is a vector formed by concatenating the feature maps (or equivalently channels), ${ \bf a } _ { j } ^ { ( \ell ) } ( { \bf X } )$ and $\mathbf { a } _ { j } ^ { ( \ell ) } ( \mathbf { X } ^ { \prime } )$ from data points $\mathbf { X }$ and $\mathbf { X } ^ { \prime }$ ,
+
+$$
+\begin{array} { r } { { \bf a } _ { i } ^ { ( \ell ) } ( { \bf X } , { \bf X } ^ { \prime } ) = \left( { \bf a } _ { i } ^ { ( \ell ) } ( { \bf X } ) \right) . } \end{array}
+$$
+
+This quantity (and the following arguments) can all be extended to the case of countably finite numbers of input points.
+
+Induction base case. For any pair of data points, $\mathbf { X }$ and $\mathbf { X } ^ { \prime }$ the feature-maps corresponding to the $j$ th channel, $\mathbf { a } _ { j } ^ { ( 1 ) } ( \mathbf { X } , \mathbf { X } ^ { \prime } )$ have a multivariate Gaussian joint distribution. This is because each element is a linear combination of shared Gaussian random variables: the biases, ${ \bf b } _ { j } ^ { ( 0 ) }$ and the filters, $\mathbf { U } _ { j , : } ^ { ( 0 ) }$ . Following Eq. (1),
+
+$$
+\mathbf { a } _ { i } ^ { ( 1 ) } ( \mathbf { X } , \mathbf { X } ^ { \prime } ) = b _ { i } ^ { ( 1 ) } \mathbf { 1 } + \sum _ { i = 1 } ^ { C ^ { ( 0 ) } } \left( \begin{array} { c c } { \mathbf { W } _ { i , j } ^ { ( 1 ) } } & { \mathbf { 0 } } \\ { \mathbf { 0 } } & { \mathbf { W } _ { i , j } ^ { ( 1 ) } } \end{array} \right) \left( \begin{array} { c } { \mathbf { x } _ { i } } \\ { \mathbf { x } _ { i } ^ { \prime } } \end{array} \right) ,
+$$
+
+where 1 is a vector of all-ones. While the elements within a feature map display strong correlations, different feature maps are independent and identically distributed (iid) conditioned on the data (i.e. $\mathbf { a } _ { i } ^ { ( 1 ) } ( \mathbf { X } , \mathbf { X } ^ { \prime } )$ and $\mathbf { a } _ { i ^ { \prime } } ^ { ( \mathrm { i } ) } ( \mathbf { X } , \mathbf { X ^ { \prime } } )$ are iid for $i \neq i ^ { \prime }$ ), because the parameters for different feature-maps (i.e. the biases, $b _ { i } ^ { ( 1 ) }$ and the filters, $\mathbf { W } _ { i , : } ^ { ( 1 ) }$ ) are themselves iid.
+
+Induction step. Consider the feature maps at the \`th layer, $\mathbf { a } _ { j } ^ { ( \ell ) } ( \mathbf { X } , \mathbf { X } ^ { \prime } )$ , to be iid multivariate Gaussian RVs (i.e. for $j \neq j ^ { \prime }$ , $\mathbf { a } _ { j } ^ { ( \ell ) } ( \mathbf { X } , \mathbf { X } ^ { \prime } )$ and $\mathbf { a } _ { j ^ { \prime } } ^ { ( \ell ) } ( \mathbf { X } , \mathbf { X } ^ { \prime } )$ are iid). Our goal is to show that, taking the number of channels at layer $\ell$ to infinity (i.e. $C ^ { ( \ell ) } \to \infty ^ { \cdot }$ ), the same properties hold at the next layer (i.e. all feature maps, a(\`+1)i (X, X0), are iid multivariate Gaussian RVs). Writing eq. (2) for two training examples, $\mathbf { X }$ and $\mathbf { X } ^ { \prime }$ , we obtain,
+
+$$
+\mathbf { a } _ { i } ^ { ( \ell + 1 ) } ( \mathbf { X } , \mathbf { X } ^ { \prime } ) = b _ { i } ^ { ( \ell + 1 ) } \mathbf { 1 } + \sum _ { j = 1 } ^ { C ^ { ( \ell ) } } \left( \begin{array} { c c } { \mathbf { W } _ { i , j } ^ { ( \ell + 1 ) } } & { \mathbf { 0 } } \\ { \mathbf { 0 } } & { \mathbf { W } _ { i , j } ^ { ( \ell + 1 ) } } \end{array} \right) \phi ( \mathbf { a } _ { j } ^ { ( \ell ) } ( \mathbf { X } , \mathbf { X } ^ { \prime } ) )
+$$
+
+We begin by showing that $\mathbf { a } _ { i } ^ { ( \ell + 1 ) } ( \mathbf { X } , \mathbf { X } ^ { \prime } )$ is a multivariate Gaussian RV. The first term is multivariate Gaussian, as it is a linear function of $b _ { i } ^ { ( \ell + 1 ) }$ , which is itself iid Gaussian. We can apply the multivariate CLT to show that the second term is also Gaussian, because, in the limit as $C ^ { ( \ell ) } \to \infty$ , it is the sum of infinitely many iid terms: $\mathbf { a } _ { j } ^ { ( \ell ) } ( \mathbf { X } , \mathbf { X } ^ { \prime } )$ are iid by assumption, and $\mathbf { W } _ { i , j } ^ { ( \ell + 1 ) }$ are iid by definition. Note that the same argument applies to all feature maps jointly, so all elements of $\mathbf { A } ^ { ( \ell + 1 ) } ( \mathbf { X } , \mathbf { X } ^ { \prime } )$ (defined by analogy with eq. 4) are jointly multivariate Gaussian.
+
+Following Lee et al. (2017), to complete the argument, we need to show that the output feature maps are iid, i.e. $\mathbf { a } _ { i } ^ { ( \ell + 1 ) } ( \mathbf { X } , \mathbf { X } ^ { \prime } )$ )(X, X0) and a(\`i0 $\mathbf { a } _ { i ^ { \prime } } ^ { ( \ell + 1 ) } ( \mathbf { X } , \mathbf { X } ^ { \prime } )$ are iid for $i \neq i ^ { \prime }$ . They are identically distributed, re as b(\`+1) a nd r tha $\mathbf { W } _ { i , j } ^ { ( \ell + 1 ) }$ and  and $\phi ( \mathbf { a } _ { j } ^ { ( \ell ) } ( \mathbf { X } , \mathbf { X } ^ { \prime } ) )$ is shared. To show that they are independent,are jointly Gaussian, so it is sufficient to show $\mathbf { a } _ { i } ^ { ( \ell + 1 ) } ( \mathbf { X } , \mathbf { X } ^ { \prime } )$ $\mathbf { a } _ { i ^ { \prime } } ^ { ( \ell + 1 ) } ( \mathbf { X } , \mathbf { X } ^ { \prime } )$   
+that they are uncorrelated, and we can show that they are uncorrelated because the weights, $\mathbf { W } _ { i , j } ^ { ( \ell + 1 ) }$ are independent with zero-mean, eliminating any correlations that might arise through the shared RV, $\phi ( \mathbf { a } _ { j } ^ { ( \ell ) } ( \mathbf { X } , \mathbf { X } ^ { \prime } ) )$ . In the appendix, we consider the more complex case where we take limits simultaneously.
+
+# 3 THE CONVNET AND RESNET KERNELS
+
+Here we derive a computationally efficient kernel corresponding to the CNN described in the previous section. It is surprising that we can compute the kernel efficiently because the feature maps, ${ \bf a } _ { i } ^ { ( \ell ) } ( { \bf X } )$ , display rich covariance structure due to the shared convolutional filter. Computing and representing these covariances would be prohibitively computationally expensive. However, in many cases we only need the variance of the output, e.g. in the case of classification or regression with a final dense layer. It turns out that this propagates backwards through the convolutional network, implying that for every layer, we only need the “diagonal covariance” of the activations: the covariance between the corresponding elements of ${ \bf a } _ { j } ^ { ( \ell ) } ( { \bf X } )$ and $\mathbf { a } _ { j } ^ { ( \ell ) } ( \mathbf { X } ^ { \prime } )$ (i.e. $\mathrm { d i a g } \left( \mathbb { C } \left[ \mathbf { a } _ { j } ^ { ( \ell ) } ( \mathbf { X } ) , \mathbf { a } _ { j } ^ { ( \ell ) } ( \mathbf { X } ^ { \prime } ) \right] \right) )$ .
+
+# 3.1 GP MEAN AND COVARIANCE
+
+A GP is completely specified by its mean and covariance (kernel) functions. These give the parameters of the joint Gaussian distribution of the RVs indexed by any two inputs, $\mathbf { X }$ and $\mathbf { X } ^ { \prime }$ . For the purposes of computing the mean and covariance, it is easiest to consider the network as being written entirely in index notation,
+
+$$
+A _ { i , \mu } ^ { ( \ell + 1 ) } ( \mathbf { X } ) = b _ { i } ^ { ( \ell + 1 ) } + \sum _ { j = 1 } ^ { C ^ { ( \ell ) } } \sum _ { \nu = 1 } ^ { H ^ { ( \ell ) } D ^ { ( \ell ) } } W _ { i , j , \mu , \nu } ^ { ( \ell + 1 ) } \phi ( A _ { j , \nu } ^ { ( \ell ) } ( \mathbf { X } ) ) .
+$$
+
+where $\ell$ and $\ell + 1$ denote the input and output layers respectively, $j$ and $i \in \{ 1 , \ldots , C ^ { ( \ell + 1 ) } \}$ denote the input and output channels, and $\nu$ and $\mu \in \{ 1 , \ldots , H ^ { ( \ell + 1 ) } D ^ { ( \ell + 1 ) } \}$ denote the location within the input and output channel or feature-maps.
+
+The mean function is thus easy to compute
+
+$$
+\mathbb { E } \left[ A _ { i , \mu } ^ { ( \ell + 1 ) } ( \mathbf { X } ) \right] = \mathbb { E } \left[ b _ { i } ^ { ( \ell + 1 ) } \right] + \sum _ { j = 1 } ^ { C ^ { ( \ell ) } } \sum _ { \nu = 1 } ^ { H ^ { ( \ell ) } D ^ { ( \ell ) } } \mathbb { E } \left[ W _ { i , j , \mu , \nu } ^ { ( \ell + 1 ) } \phi ( A _ { j , \nu } ^ { ( \ell ) } ( \mathbf { X } ) ) \right] = 0 .
+$$
+
+as b(\`+1)i a nd W (\`+1)i,j,µ,ν have zero mean, and W (\`+1)i,j,ν,µ are independent of the activations at the previous layer, $\phi ( A _ { j , \nu } ^ { ( \ell ) } ( \mathbf { X } ) )$ .
+
+Now we show that it is possible to efficiently compute the covariance function. This is surprising because for many networks, we nlocations in the feature map (i.e. $\mathbb { C } \left[ A _ { i , \mu } ^ { ( \ell + 1 ) } ( \mathbf { \bar { X } } ) , A _ { i , \mu ^ { \prime } } ^ { ( \ell + 1 ) } ( \mathbf { X } ^ { \prime } ) \right]$ ce ofor $\mu , \mu ^ { \prime } \in \{ 1 , \ldots , H ^ { ( \ell + 1 ) } D ^ { ( \ell + 1 ) } \} )$ and this object is extremely high-dimensional, $N ^ { 2 } ( H ^ { ( \ell + 1 ) } D ^ { \overline { { ( \ell + 1 ) } } } ) ^ { 2 }$ . However, it turns out that we only need to consider the “diagonal” covariance, (i.e. we only need $\mathbb { C } \left[ A _ { i , \mu } ^ { ( \ell + 1 ) } ( \mathbf { X } ) , A _ { i , \mu } ^ { ( \ell + 1 ) } ( \mathbf { X } ^ { \prime } ) \right]$ fo r $\mu \in \{ 1 , \ldots , H ^ { ( \ell + 1 ) } D ^ { ( \ell + 1 ) } \} )$ , which is a more manageable quantity of size $N ^ { 2 } ( H ^ { ( \ell + 1 ) } D ^ { ( \ell + 1 ) } )$ .
+
+This is true at the output layer $( L + 1 )$ : in order to achieve an output suitable for classification or regression, we use only a single output location $H ^ { ( L + 1 ) } = D ^ { ( L + 1 ) } = 1$ , with a number of “channels” equal to the number of of outputs/classes, so it is only possible to compute the covariance at that single location. We now show that, if we only need the covariance at corresponding locations in the outputs, we only need the covariance at corresponding locations in the inputs, and this requirement propagates backwards through the network.
+
+Formally, as the activations are composed of a sum of terms, their covariance is the sum of the covariances of all those underlying terms,
+
+$$
+\begin{array} { r l } { \mathbb { C } \left[ A _ { i , \mu } ^ { ( \ell + 1 ) } ( \mathbf { X } ) , A _ { i , \mu } ^ { ( \ell + 1 ) } ( \mathbf { X } ^ { \prime } ) \right] = \mathbb { V } \left[ b _ { i } ^ { ( \ell ) } \right] + } & { } \\ { \displaystyle \sum _ { j = 1 } ^ { C ^ { ( \ell ) } } \sum _ { j ^ { \prime } = 1 } ^ { C ^ { ( \ell ) } } \sum _ { \nu = 1 } ^ { H ^ { ( \ell ) } D ^ { ( \ell ) } } \sum _ { \nu ^ { \prime } = 1 } ^ { H ^ { ( \ell ) } D ^ { ( \ell ) } } \mathbb { C } \left[ W _ { i , j , \mu , \nu } ^ { ( \ell + 1 ) } \phi ( A _ { j , \nu } ^ { ( \ell ) } ( \mathbf { X } ) ) , W _ { i , j ^ { \prime } , \mu , \nu ^ { \prime } } ^ { ( \ell + 1 ) } \phi ( A _ { j ^ { \prime } , \nu ^ { \prime } } ^ { ( \ell ) } ( \mathbf { X } ^ { \prime } ) ) \right] . } \end{array}
+$$
+
+As the terms in the covariance have mean zero, and as the weights and activations from the previous layer are independent,
+
+$$
+\begin{array} { r l } { \mathbb { C } \left[ A _ { i , \mu } ^ { ( \ell + 1 ) } ( \mathbf { X } ) , A _ { i , \mu } ^ { ( \ell + 1 ) } ( \mathbf { X } ^ { \prime } ) \right] = \sigma _ { \mathrm { b } } ^ { 2 } + } & { } \\ { \quad } & { \qquad \sum _ { j = 1 } ^ { { \ell ^ { ( \ell ) } } } \displaystyle \sum _ { j ^ { \prime } = 1 } ^ { { \ell ^ { ( \ell ) } } } \sum _ { \nu = 1 } ^ { { H ^ { ( \ell ) } } D ^ { ( \ell ) } } \sum _ { \nu ^ { \prime } = 1 } ^ { H ^ { ( \ell ) } D ^ { ( \ell ) } } \mathbb { E } \left[ W _ { i , j , \mu , \nu } ^ { ( \ell + 1 ) } W _ { i , j ^ { \prime } , \mu , \nu ^ { \prime } } ^ { ( \ell + 1 ) } \right] \mathbb { E } \left[ \phi ( A _ { j , \nu } ^ { ( \ell ) } ( \mathbf { X } ) ) \phi ( A _ { j ^ { \prime } , \nu ^ { \prime } } ^ { ( \ell ) } ( \mathbf { X } ^ { \prime } ) ) \right] . } \end{array}
+$$
+
+# Algorithm 1 The ConvNet kernel $k ( \mathbf { X } , \mathbf { X } ^ { \prime } )$
+
+1: Input: two images, X, X0 ∈ RC(0)×(H(0)W (0)).   
+2: Compute $K _ { \mu } ^ { ( 1 ) } ( { \bf X } , { \bf X } )$ , $K _ { \mu } ^ { ( 1 ) } ( { \bf X } , { \bf X ^ { \prime } } )$ , and $K _ { \mu } ^ { ( 1 ) } ( \mathbf { X } ^ { \prime } , \mathbf { X } ^ { \prime } )$ for $\mu \in \{ 1 , \ldots , H ^ { ( 1 ) } D ^ { ( 1 ) } \}$ ; using Eq. (10).   
+3: for $\ell = 1 , 2 , \dots , L$ do   
+4: Compute $V _ { \mu } ^ { ( \ell ) } ( { \bf X } , { \bf X ^ { \prime } } )$ , $V _ { \mu } ^ { ( \ell ) } ( { \bf X } , { \bf X ^ { \prime } } )$ and $V _ { \mu } ^ { ( \ell ) } ( { \bf X } , { \bf X ^ { \prime } } )$ for $\mu \in \{ 1 , \ldots , H ^ { ( \ell ) } D ^ { ( \ell ) } \}$ ; using   
+5: Compute Eq. (13), or some other nonlinearity. $K _ { \mu } ^ { ( \ell + 1 ) } ( { \mathbf { X } } , { \mathbf { X } } )$ , $K _ { \mu } ^ { ( \ell + 1 ) } ( { \bf X } , { \bf X ^ { \prime } } )$ , and $K _ { \mu } ^ { ( \ell + 1 ) } ( { \mathbf { X } } ^ { \prime } , { \mathbf { X } } ^ { \prime } )$ for $\mu$ ∈ $\{ 1 , \ldots , H ^ { ( \ell + 1 ) } D ^ { ( \ell + 1 ) } \}$ ; using Eq. (11).
+
+6: end for   
+7: Output the scalar $K _ { 1 } ^ { ( L + 1 ) } ( { \bf X } , { \bf X ^ { \prime } } )$ .
+
+nt ffor ent channels: Further, each $\mathbf { W } _ { i , j } ^ { ( \ell + 1 ) }$ and of the $\mathbf { W } _ { i , j ^ { \prime } } ^ { ( \ell + 1 ) }$ are iid f matrices $j \neq j ^ { \prime }$ , sonly contains independent variables or zeros (Fig. 1), so $\mathbb { E } \left[ W _ { i , j , \mu , \nu } ^ { ( \ell + 1 ) } W _ { i , j ^ { \prime } , \mu , \nu ^ { \prime } } ^ { ( \ell + 1 ) } \right] = 0$ $j \neq j ^ { \prime }$ $\mathbb { E } \left[ W _ { i , j , \mu , \nu } ^ { ( \ell + 1 ) } W _ { i , j ^ { \prime } , \mu , \nu ^ { \prime } } ^ { ( \ell + 1 ) } \right] = 0$ $\mu$ for $\nu \neq \nu ^ { \prime }$ $\mathbf { W } _ { i , j } ^ { ( \ell + 1 ) }$ . Thus, we can eliminate the sums over $j ^ { \prime }$ and $\nu ^ { \prime }$ :
+
+$$
+\Sigma \left[ A _ { i , \mu } ^ { ( \ell + 1 ) } ( \mathbf { X } ) , A _ { i , \mu } ^ { ( \ell + 1 ) } ( \mathbf { X ^ { \prime } } ) \right] = \sigma _ { \mathrm { b } } ^ { ( \ell ) } + \sum _ { j = 1 } ^ { Q ^ { ( \ell ) } } \sum _ { \nu = 1 } ^ { H ^ { ( \ell ) } D ^ { ( \ell ) } } \mathbb { E } \left[ W _ { i , j , \mu , \nu } ^ { ( \ell + 1 ) } W _ { i , j , \mu , \nu } ^ { ( \ell + 1 ) } \right] \mathbb { E } \left[ \phi ( A _ { j , \nu } ^ { ( \ell ) } ( \mathbf { X } ) ) \phi ( A _ { j , \nu } ^ { ( \ell ) } ( \mathbf { X ^ { \prime } } ) ) \right] .
+$$
+
+The µth row of W(\`+1)i,j i s zero for indices $\nu$ that do not belong to its convolutional patch, so we can restrict the sum over $\nu$ to that region. We also define $v _ { g } ^ { ( 1 ) } ( \mathbf { X } , \mathbf { X } ^ { \prime } )$ , to emphasise that the covariances are independent of the output channel, $j$ . The variance of the first layer is
+
+$$
+K _ { \mu } ^ { ( 1 ) } ( \mathbf { X } , \mathbf { X ^ { \prime } } ) = \mathbb { C } \left[ A _ { i , \mu } ^ { ( 1 ) } ( \mathbf { X } ) , A _ { i , \mu } ^ { ( 1 ) } ( \mathbf { X ^ { \prime } } ) \right] = \sigma _ { \mathrm { b } } ^ { 2 } + \frac { \sigma _ { \mathrm { w } } ^ { 2 } } { C ^ { ( 0 ) } } \sum _ { i = 1 } ^ { C ^ { ( 0 ) } } \sum _ { \nu \in \mu \mathrm { t h } \mathrm { p a t c h } } X _ { i , \nu } X _ { i , \nu } ^ { \prime } .
+$$
+
+And we do the same for the other layers,
+
+$$
+K _ { \mu } ^ { ( \ell + 1 ) } ( \mathbf { X } , \mathbf { X } ^ { \prime } ) = \mathbb { C } \left[ A _ { i , \mu } ^ { ( \ell + 1 ) } ( \mathbf { X } ) , A _ { i , \mu } ^ { ( \ell + 1 ) } ( \mathbf { X } ^ { \prime } ) \right] = \sigma _ { \mathrm { b } } ^ { 2 } + \sigma _ { \mathrm { w } } ^ { 2 } \sum _ { \nu \in \mu \mathrm { t h p a t c h } } V _ { \nu } ^ { ( \ell ) } ( \mathbf { X } , \mathbf { X } ^ { \prime } ) ,
+$$
+
+where
+
+$$
+V _ { \nu } ^ { ( \ell ) } ( \mathbf { X } , \mathbf { X } ^ { \prime } ) = \mathbb { E } \left[ \phi ( A _ { j , \nu } ^ { ( \ell ) } ( \mathbf { X } ) ) \phi ( A _ { j , \nu } ^ { ( \ell ) } ( \mathbf { X } ^ { \prime } ) ) \right]
+$$
+
+is the covariance of the activations, which is again independent of the channel.
+
+# 3.2 COVARIANCE OF THE ACTIVITIES
+
+The elementwise covariance in the right-hand side of Eq. (11) can be computed in closed form for many choices of $\phi$ if the activations are Gaussian. For each element of the activations, one needs to keep track of the 3 distinct entries of the bivariate covariance matrix between the inputs, $K _ { \mu } ^ { ( \ell + 1 ) } ( { \mathbf { X } } , { \mathbf { X } } ) , K _ { \mu } ^ { ( \ell + 1 ) } ( { \mathbf { X } } , { \mathbf { X } } ^ { \prime } )$ and $K _ { \mu } ^ { ( \ell + 1 ) } ( { \mathbf { X } } ^ { \prime } , { \mathbf { X } } ^ { \prime } )$ .
+
+For example, for the ReLU nonlinearity $( \phi ( x ) = \operatorname* { m a x } ( 0 , x ) )$ , one can adapt Cho & Saul (2009) in the same way as Matthews et al. (2018a, section 3) to obtain
+
+$$
+V _ { \nu } ^ { ( \ell ) } ( { \mathbf { X } } , { \mathbf { X ^ { \prime } } } ) = \frac { \sqrt { K _ { \nu } ^ { ( \ell ) } ( { \mathbf { X } } , { \mathbf { X } } ) K _ { \nu } ^ { ( \ell ) } ( { \mathbf { X ^ { \prime } } } , { \mathbf { X ^ { \prime } } } ) } } { \pi } \left( \sin \theta _ { \nu } ^ { ( \ell ) } + ( \pi - \theta _ { \nu } ^ { ( \ell ) } ) \cos \theta _ { \nu } ^ { ( \ell ) } \right)
+$$
+
+where $\theta _ { \nu } ^ { ( \ell ) } = \cos ^ { - 1 } \bigg ( K _ { \nu } ^ { ( \ell ) } ( { \mathbf X } , { \mathbf X ^ { \prime } } ) / \sqrt { K _ { \nu } ^ { ( \ell ) } ( { \mathbf X } , { \mathbf X } ) K _ { \nu } ^ { ( \ell ) } ( { \mathbf X ^ { \prime } } , { \mathbf X ^ { \prime } } ) } \bigg ) .$
+
+# 3.3 EFFICIENCY OF THE CONVNET KERNEL
+
+We now have all the pieces for computing the kernel, as written in Algorithm 1.
+
+Putting together Eq. (11) and Eq. (13) gives us the surprising result that the diagonal covariances of the activations at layer $\ell + 1$ only depend on the diagonal covariances of the activations at layer $\ell$ . This is very important, because it makes the computational cost of the kernel be within a constant factor of the cost of a forward pass for the equivalent CNN with 1 filter per layer.
+
+Thus, the algorithm is more efficient that one would naively think. A priori, one needs to compute the covariance between all the elements of ${ \bf a } _ { j } ^ { ( \ell ) } ( { \bf X } )$ and $\mathbf { a } _ { j } ^ { ( \ell ) } ( \mathbf { X } ^ { \prime } )$ combined, yielding a $2 H ^ { ( \ell ) } D ^ { ( \ell ) } \times$ $2 H ^ { ( \ell ) } D ^ { ( \ell ) }$ covariance matrix for every pair of points. Instead, we only need to keep track of a $H ^ { ( \ell ) } D ^ { ( \ell ) }$ -dimensional vector per layer and pair of points.
+
+Furthermore, the particular form for the kernel (eq. 1 and eq. 2) implies that the required variances and covariances at all required locations can be computed efficiently as a convolution.
+
+# 3.4 KERNEL FOR A RESIDUAL CNN
+
+The induction step in the argument for GP behaviour from Sec. 2.2 depends only on the previous activations being iid Gaussian. Since all the activations are iid Gaussian, we can add skip connections between the activations of different layers while preserving GP behaviour, e.g. $\mathbf { A } ^ { ( \ell + 1 ) }$ and $\mathbf { A } ^ { ( \ell - s ) }$ where $s$ is the number of layers that the skip connection spans. If we change the NN recursion (Eq. 2) to
+
+$$
+\mathbf { a } _ { i } ^ { ( \ell + 1 ) } ( \mathbf { X } ) : = \mathbf { a } _ { i } ^ { ( \ell - s ) } ( \mathbf { X } ) + \mathbf { b } _ { i } ^ { ( \ell + 1 ) } + \sum _ { j = 1 } ^ { C ^ { ( \ell ) } } \mathbf { W } _ { i , j } ^ { ( \ell ) } \phi \left( \mathbf { a } _ { j } ^ { ( \ell ) } ( \mathbf { X } ) \right) ,
+$$
+
+then the kernel recursion (Eq. 11) becomes
+
+$$
+K _ { \mu } ^ { ( \ell + 1 ) } ( \mathbf { X } , \mathbf { X } ^ { \prime } ) = K _ { \mu } ^ { ( \ell - s ) } ( \mathbf { X } , \mathbf { X } ^ { \prime } ) + \sigma _ { b } ^ { 2 } + \sigma _ { w } ^ { 2 } \sum _ { \nu \in \mu \mathrm { t h } \mathrm { p a t c h } } V _ { \nu } ^ { ( \ell ) } ( \mathbf { X } , \mathbf { X } ^ { \prime } ) .
+$$
+
+This way of adding skip connections is equivalent to the “pre-activation” shortcuts described by He et al. (2016b). Remarkably, the natural way of adding residual connections to NNs is the one that performed best in their empirical evaluations.
+
+# 4 EXPERIMENTS
+
+We evaluate our kernel on the MNIST handwritten digit classification task. Classification likelihoods are not conjugate for GPs, so we must make an approximation, and we follow Lee et al. (2017), in re-framing classification as multi-output regression.
+
+The training set is split into $N = 5 0 0 0 0$ training and 10000 validation examples. The regression targets $\mathbf { Y } \in \{ - 1 , 1 \} ^ { \overline { { N } } \times 1 0 }$ are a one-hot encoding of the example’s class: $y _ { n , c } = 1$ if the $n$ th example belongs to class $c$ , and $- 1$ otherwise.
+
+Training is exact conjugate likelihood GP regression with noiseless targets $\mathbf { Y }$ (Rasmussen & Williams, 2006). First we compute the $N \times N$ kernel matrix ${ \bf K } _ { x x }$ , which contains the kernel between every pair of examples. Then we compute ${ \mathbf { K } } _ { x x } ^ { - 1 } { \mathbf { Y } }$ using a linear system solver.
+
+The test set has $N _ { T } = 1 0 0 0 0$ examples. We compute the $N _ { T } \times N$ matrix $\mathbf { K } _ { x ^ { * } x }$ , the kernel between each test example and all the training examples. The predictions are given by the row-wise maximum of $\mathbf { K } _ { x ^ { * } x } \mathbf { K } _ { x x } ^ { - 1 } \bar { \mathbf { Y } }$ .
+
+For the “ConvNet GP” and “Residual CNN GP”, (Table 1) we optimise the kernel hyperparameters by random search. We draw $M$ random hyperparameter samples, compute the resulting kernel’s performance in the validation set, and pick the highest performing run. The kernel hyperparameters are: $\sigma _ { b } ^ { 2 }$ , $\sigma _ { w } ^ { 2 }$ ; the number of layers; the convolution stride, filter sizes and edge behaviour; the nonlinearity (we consider the error function and ReLU); and the frequency of residual skip connections (for Residual CNN GPs). We do not retrain the model on the validation set after choosing hyperparameters.
+
+Table 1: MNIST classification results. #samples gives the number of kernels that were randomly sampled for the hyperparameter search. “ConvNet GP” and “Residual CNN GP” are random CNN architectures with a fixed filter size, whereas “ResNet GP” is a slight modification of the architecture by He et al. (2016b). Entries labelled “SGD” used stochastic gradient descent for tuning hyperparameters, by maximising the likelihood of the training set. The last two methods use parametric neural networks. The hyperparameters of the ResNet GP were not optimised (they were fixed based on the architecture from He et al., 2016b).   
+
+<table><tr><td>Method</td><td>#samples</td><td>Validation error</td><td>Test error</td></tr><tr><td>NNGP (Lee et al., 2017)</td><td>~250</td><td>1</td><td>1.21%</td></tr><tr><td>Convolutional GP (van der Wilk et al., 2017)</td><td>SGD</td><td></td><td>1.17%</td></tr><tr><td>Deep Conv. GP (Kumar et al., 2018)</td><td>SGD</td><td>1</td><td>1.34%</td></tr><tr><td>ConvNetGP</td><td>27</td><td>0.71%</td><td>1.03%</td></tr><tr><td>Residual CNN GP</td><td>27</td><td>0.72%</td><td>0.96%</td></tr><tr><td>ResNet GP</td><td>1</td><td>0.68%</td><td>0.84%</td></tr><tr><td>GP + parametric deep kernel (Bradshaw et al., 2017)</td><td>SGD</td><td>1</td><td>0.60%</td></tr><tr><td>ResNet (Chen et al.,2018)</td><td>一</td><td></td><td>0.41%</td></tr></table>
+
+The “ResNet GP” (Table 1) is the kernel equivalent to a 32-layer version of the basic residual architecture by He et al. (2016a). The differences are: an initial $3 \times 3$ convolutional layer and a final dense layer instead of average pooling. We chose to remove the pooling because computing its output variance requires the off-diagonal elements of the filter covariance, in which case we could not exploit the efficiency gains described in Sec. 3.3.
+
+We found that the despite it not being optimised, the 32-layer ResNet GP outperformed all other comparable architectures (Table 1), including the NNGP in Lee et al. (2017), which is state-ofthe-art for non-convolutional networks, and convolutional GPs (van der Wilk et al., 2017; Kumar et al., 2018). That said, our results have not reached state-of-the-art for methods that incorporate a parametric neural network, such as a standard ResNet (Chen et al., 2018) and a Gaussian process with a deep neural network kernel Bradshaw et al. (2017).
+
+To check whether the GP limit is applicable to relatively small networks used practically (with of the order of 100 channels in the first layers), we randomly sampled 10, 000 32-layer ResNets, with 3, 10, 30 and 100 channels in the first layers, and, following the usual practice for ResNets we increase the number the number of hidden units when we downsample the feature maps. The probability density plots show a good match around 100 channels (Fig. 2A), which matches a more sensitive graphical procedure based on quantile-quantile plots (Fig. 2B). Notably, even for only 30 channels, the moments match closely (Fig. 2C). For comparison, typical ResNets use from 64 (He et al., 2016a) to 192 (Zagoruyko & Komodakis, 2016) channels in their first layers. We believe that this is because the moment propagation equations only require the Gaussianity assumption for propagation through the relu, and presumably this is robust to non-Gaussian input activations.
+
+Computational efficiency. Asymptotically, computing the kernel matrix takes $O ( N ^ { 2 } L D )$ time, where $L$ is the number of layers in the network and $D$ is the dimensionality of the input, and inverting the kernel matrix takes $O ( \bar { N } ^ { 3 } )$ . As such, we expect that for very large datasets, inverting the kernel matrix will dominate the computation time. However, on MNIST, $N ^ { 3 }$ is only around a factor of 10 larger than $N ^ { 2 } L D$ . In practice, we found that it was more expensive to compute the kernel matrix than to invert it. For the ResNet kernel, the most expensive, computing ${ \bf K } _ { x x }$ , and ${ \bf K } _ { x x * }$ for validation and test took 3h 40min on two Tesla P100 GPUs. In contrast, inverting ${ \bf K } _ { x x }$ and computing validation and test performance took $4 3 . 2 5 \pm 8 . 8$ seconds on a single Tesla P100 GPU.
+
+# 5 RELATED WORK
+
+Van der Wilk et al. (van der Wilk et al., 2017) also adapted GPs to image classification. They defined a prior on functions $f$ that takes an image and outputs a scalar. First, draw a function $g \sim \mathcal { G P } ( 0 , k _ { p } ( \mathbf { X } , \mathbf { X } ^ { \prime } ) )$ . Then, $f$ is the sum of the output of $g$ applied to each of the convolutional patches. Their approach is also inspired by convolutional NNs, but their kernel $k _ { p }$ is applied to all pairs of patches of $\mathbf { X }$ and $\mathbf { X } ^ { \prime }$ . This makes their convolutional kernel expensive to evaluate, requiring inter-domain inducing point approximations to remain tractable. The kernels in this work, directly motivated by the infinite-filter limit of a CNN, only apply something like $k _ { p }$ to the corresponding pairs of patches within $\mathbf { X }$ and $\mathbf { X } ^ { \prime }$ (Eq. 10). As such, the CNN kernels are cheaper to compute and exhibit superior performance (Table 1), despite the use of an approximate likelihood function.
+
+![](images/855e86413bd7de48d01d9cca88bd8f2c7466cd7a38d25eebae9541aa8c975eaf.jpg)  
+Figure 2: Comparison of the infinite limit, and outputs from finite 32-layer ResNets with 3, 10, 30, and 100 channels in their first layers. A Comparison of the empirical and limiting probability densities. B A more sensitive test of Gaussianity is a quantile-quantile plot, which shows converges with 100 channels. C The moments (variances and covariances) for 100 training inputs shows gives a good match for all numbers of channels.
+
+Kumar et al. (2018) define a prior over functions by stacking several GPs with van der Wilk’s convolutional kernel, forming a “Deep GP” (Damianou & Lawrence, 2013). In contrast, the kernel in this paper confines all hierarchy to the definition of the kernel, and the resulting GPs is shallow.
+
+Wilson et al. (2016) introduced and Bradshaw et al. (2017) improved deep kernel learning. The inputs to a classic GP kernel $k$ (e.g. RBF) are preprocessed by applying a feature extractor $g$ (a deep NN) prior to computing the kernel: $k _ { \mathrm { d e e p } } ( \mathbf { X } , \bar { \mathbf { X } } ^ { \prime } ) : = k ( g ( \mathbf { X } ; \theta ) , \bar { g ( \mathbf { X } ^ { \prime } , \theta ) } )$ . The NN parameters are optimised by gradient ascent using the likelihood as the objective, as in standard GP kernel learning (Rasmussen & Williams, 2006, Chapter 5). Since deep kernel learning incorporates a state-of-the-art NN with over $1 0 ^ { 6 }$ parameters, we expect it to perform similarly to a NN applied directly to the task of image classification. At present both CNNs and deep kernel learning display superior performance to the GP kernels in this work. However, the kernels defined here have far fewer parameters (around 10, compared to their $1 0 ^ { 6 }$ ).
+
+Borovykh (2018) also suggests that a CNN exhibits GP behaviour. However, they take the infinite limit with respect to the filter size, not the number of filters. Thus, their infinite network is inapplicable to real data which is always of finite dimension.
+
+Finally, there is a series of papers analysing the mean-field behaviour of deep NNs and CNNs which aims to find good random initializations, i.e. those that do not exhibit vanishing or exploding gradients or activations (Schoenholz et al., 2016; Yang & Schoenholz, 2017). Apart from their very different focus, the key difference to our work is that they compute the variance for a single trainingexample, whereas to obtain the GPs kernel, we additionally need to compute the output covariances for different training/test examples (Xiao et al., 2018).
+
+# 6 CONCLUSIONS AND FUTURE WORK
+
+We have shown that deep Bayesian CNNs with infinitely many filters are equivalent to a GP with a recursive kernel. We also derived the kernel for the GP equivalent to a CNN, and showed that, in handwritten digit classification, it outperforms all previous GP approaches that do not incorporate a parametric NN into the kernel. Given that most state-of-the-art neural networks incorporate structure (convolutional or otherwise) into their architecture, the equivalence between CNNs and GPs is potentially of considerable practical relevance. In particular, we hope to apply GP CNNs in domains as widespread as adversarial examples, lifelong learning and $\mathbf { k }$ -shot learning, and we hope to improve them by developing efficient multi-layered inducing point approximation schemes.
+
+# REFERENCES
+
+JR Blum, H Chernoff, M Rosenblatt, and H Teicher. Central limit theorems for interchangeable processes. Canad. J. Math, 10:222–229, 1958.
+
+Charles Blundell, Julien Cornebise, Koray Kavukcuoglu, and Daan Wierstra. Weight uncertainty in neural network. In International Conference on Machine Learning, pp. 1613–1622, 2015.
+
+Anastasia Borovykh. A gaussian process perspective on convolutional neural networks. ResearchGate, 05 2018.
+
+John Bradshaw, Alexander G de G Matthews, and Zoubin Ghahramani. Adversarial examples, uncertainty, and transfer testing robustness in gaussian process hybrid deep networks. arXiv preprint arXiv:1707.02476, 2017.
+
+Tian Qi Chen, Yulia Rubanova, Jesse Bettencourt, and David Duvenaud. Neural ordinary differential equations. arXiv preprint arXiv:1806.07366, 2018.
+
+Youngmin Cho and Lawrence K Saul. Kernel methods for deep learning. In Advances in neural information processing systems, pp. 342–350, 2009. URL http://papers.nips.cc/ paper/3628-kernel-methods-for-deep-learning.pdf.
+
+Andreas Damianou and Neil Lawrence. Deep gaussian processes. In Carlos M. Carvalho and Pradeep Ravikumar (eds.), Proceedings of the Sixteenth International Conference on Artificial Intelligence and Statistics, volume 31 of Proceedings of Machine Learning Research, pp. 207– 215, Scottsdale, Arizona, USA, 29 Apr–01 May 2013. PMLR. URL http://proceedings. mlr.press/v31/damianou13a.html.
+
+Marc Deisenroth and Carl E Rasmussen. PILCO: A model-based and data-efficient approach to policy search. In Proceedings of the 28th International Conference on machine learning (ICML11), pp. 465–472, 2011.
+
+Yarin Gal and Zoubin Ghahramani. Dropout as a bayesian approximation: Representing model uncertainty in deep learning. arXiv preprint arXiv:1506.02142, 2015.
+
+Yarin Gal and Lewis Smith. Idealised bayesian neural networks cannot have adversarial examples: Theoretical and empirical study. arXiv preprint arXiv:1806.00667, 2018.
+
+Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image recognition. In Proceedings of the IEEE conference on computer vision and pattern recognition, pp. 770–778, 2016a. URL https://arxiv.org/pdf/1512.03385.pdf.
+
+Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Identity mappings in deep residual networks. In European conference on computer vision, pp. 630–645. Springer, 2016b.
+
+Daniel Hernández-lobato, Jose M. Hernández-lobato, and Pierre Dupont. Robust multiclass gaussian process classification. In J. Shawe-Taylor, R. S. Zemel, P. L. Bartlett, F. Pereira, and K. Q. Weinberger (eds.), Advances in Neural Information Processing Systems 24, pp. 280–288. 2011. URL http://papers.nips.cc/paper/ 4241-robust-multi-class-gaussian-process-classification.pdf.
+
+Alex Krizhevsky, Ilya Sutskever, and Geoffrey E Hinton. Imagenet classification with deep convolutional neural networks. In Advances in neural information processing systems, pp. 1097–1105, 2012.
+
+Vinayak Kumar, Vaibhav Singh, PK Srijith, and Andreas Damianou. Deep gaussian processes with convolutional kernels. arXiv preprint arXiv:1806.01655, 2018.
+
+Alexey Kurakin, Ian Goodfellow, and Samy Bengio. Adversarial examples in the physical world. arXiv preprint arXiv:1607.02533, 2016.
+
+Balaji Lakshminarayanan, Alexander Pritzel, and Charles Blundell. Simple and scalable predictive uncertainty estimation using deep ensembles. In Advances in Neural Information Processing Systems, pp. 6402–6413, 2017.
+
+Yann LeCun, Bernhard E Boser, John S Denker, Donnie Henderson, Richard E Howard, Wayne E Hubbard, and Lawrence D Jackel. Handwritten digit recognition with a back-propagation network. In Advances in neural information processing systems, pp. 396–404, 1990.
+
+Jaehoon Lee, Yasaman Bahri, Roman Novak, Samuel S Schoenholz, Jeffrey Pennington, and Jascha Sohl-Dickstein. Deep neural networks as gaussian processes. arXiv preprint arXiv:1711.00165, 2017.
+
+Stephan Mandt, Matthew D Hoffman, and David M Blei. Stochastic gradient descent as approximate bayesian inference. The Journal of Machine Learning Research, 18(1):4873–4907, 2017.
+
+Alexander G. de G. Matthews, Jiri Hron, Mark Rowland, Richard E. Turner, and Zoubin Ghahramani. Gaussian process behaviour in wide deep neural networks. In International Conference on Learning Representations, 2018a. URL https://openreview.net/forum?id $=$ H1-nGgWC-.
+
+Alexander G de G Matthews, Mark Rowland, Jiri Hron, Richard E Turner, and Zoubin Ghahramani. Gaussian process behaviour in wide deep neural networks. arXiv preprint arXiv:1804.11271, 2018b.
+
+De G Matthews, G Alexander, Mark Van Der Wilk, Tom Nickson, Keisuke Fujii, Alexis Boukouvalas, Pablo León-Villagrá, Zoubin Ghahramani, and James Hensman. GPflow: A gaussian process library using TensorFlow. The Journal of Machine Learning Research, 18(1):1299–1304, 2017.
+
+Radford M. Neal. Bayesian Learning for Neural Networks. Springer-Verlag, Berlin, Heidelberg, 1996. ISBN 0387947248.
+
+Carl Edward Rasmussen and Christopher KI Williams. Gaussian processes for machine learning, volume 1. MIT press Cambridge, 2006.
+
+Samuel S Schoenholz, Justin Gilmer, Surya Ganguli, and Jascha Sohl-Dickstein. Deep information propagation. arXiv preprint arXiv:1611.01232, 2016.
+
+Jasper Snoek, Hugo Larochelle, and Ryan P Adams. Practical bayesian optimization of machine learning algorithms. In Advances in neural information processing systems, pp. 2951–2959, 2012.
+
+Christian Szegedy, Wojciech Zaremba, Ilya Sutskever, Joan Bruna, Dumitru Erhan, Ian Goodfellow, and Rob Fergus. Intriguing properties of neural networks. arXiv preprint arXiv:1312.6199, 2013.
+
+Mark van der Wilk, Carl Edward Rasmussen, and James Hensman. Convolutional gaussian processes. In Advances in Neural Information Processing Systems, pp. 2845–2854, 2017.
+
+Max Welling and Yee W Teh. Bayesian learning via stochastic gradient langevin dynamics. In Proceedings of the 28th International Conference on Machine Learning (ICML-11), pp. 681–688, 2011.
+
+Andrew Gordon Wilson, Zhiting Hu, Ruslan Salakhutdinov, and Eric P. Xing. Deep kernel learning. In Arthur Gretton and Christian C. Robert (eds.), Proceedings of the 19th International Conference on Artificial Intelligence and Statistics, volume 51 of Proceedings of Machine Learning Research, pp. 370–378, Cadiz, Spain, 09–11 May 2016. PMLR. URL http: //proceedings.mlr.press/v51/wilson16.html.
+
+Lechao Xiao, Yasaman Bahri, Jascha Sohl-Dickstein, Samuel S Schoenholz, and Jeffrey Pennington. Dynamical isometry and a mean field theory of cnns: How to train 10,000-layer vanilla convolutional neural networks. arXiv preprint arXiv:1806.05393, 2018.
+
+Ge Yang and Samuel Schoenholz. Mean field residual networks: On the edge of chaos. In Advances in neural information processing systems, pp. 7103–7114, 2017.
+
+Sergey Zagoruyko and Nikos Komodakis. Wide residual networks. arXiv preprint arXiv:1605.07146, 2016.
+
+# 7 APPENDIX
+
+# 7.1 TECHNICAL NOTES ON LIMITS
+
+The key technical issues in the proof (and the key differences between Lee et al. 2017 and Matthews et al. 2018b) arise from exactly how and where we take limits. In particular, consider the activations as being functions of the activities at the previous layer,
+
+$$
+{ \bf A } ^ { ( 4 ) } = { \bf A } ^ { ( 4 ) } ( { \bf A } ^ { ( 3 ) } ( { \bf A } ^ { ( 2 ) } ( { \bf A } ^ { ( 1 ) } ( { \bf X } ) ) ) )
+$$
+
+Now, there are two approaches to taking limits. First, both our argument in the main text, and the argument in Lee et al. (2017) is valid if we are able to take limits “inside” the network,
+
+$$
+\mathbf { A } _ { \mathrm { L } } ^ { ( 4 ) } = \operatorname* { l i m } _ { C ^ { ( 3 ) } \to \infty } \mathbf { A } ^ { ( 4 ) } \left( \operatorname* { l i m } _ { C ^ { ( 2 ) } \to \infty } \mathbf { A } ^ { ( 3 ) } \left( \operatorname* { l i m } _ { C ^ { ( 1 ) } \to \infty } \mathbf { A } ^ { ( 2 ) } \left( \mathbf { A } ^ { ( 1 ) } ( \mathbf { X } ) \right) \right) \right) .
+$$
+
+However, Matthews et al. (2018a;b) argue that is preferable to take limits “outside” the network. In particular, Matthews et al. (2018b) take the limit with all layers simultaneously,
+
+$$
+{ \bf A } _ { \mathrm { M } } ^ { ( 4 ) } = \operatorname* { l i m } _ { n  \infty } { \bf A } ^ { ( 4 ) } ( { \bf A } ^ { ( 3 ) } ( { \bf A } ^ { ( 2 ) } ( { \bf A } ^ { ( 1 ) } ( { \bf X } ) ) ) ) ,
+$$
+
+where $C ^ { ( \ell ) } = C ^ { ( \ell ) } ( n )$ goes to infinity as $n \to \infty$ . That said, similar technical issues arise if we take limits in sequence, but outside the network.
+
+# 7.2 EXTENDING THE DERIVATIONS OF MATTHEWS ET AL. (2018B) TO THE CONVOLUTIONAL CASE
+
+In the main text, we follow Lee et al. (2017) in sequentially taking the limit of each layer to infinity (i.e. $C ^ { ( 1 ) } \to \infty$ , then $C ^ { ( 2 ) }  \infty$ etc.). This dramatically simplified the argument, because taking the number of units in the previous layer to infinity means that the inputs from that layer are exactly Gaussian distributed. However, Matthews et al. (2018b) argue that the more practically relevant limit is where we take all layers to infinity simultaneously. This raises considerable additional difficulties, because we must reason about convergence in the case where the previous layer is finite. Note that this section is not intended to stand independently: it is intended to be read alongside Matthews et al. (2018b), and we use several of their results without proof.
+
+Mirroring Definition 3 in Matthews et al. (2018b), we begin by choosing a set of “width” functions, $C ^ { ( \ell ) } ( n )$ , for $\ell \in \{ 1 , \ldots , L \}$ which all approach infinity as $n  \infty$ . In Matthews et al. (2018b), these functions described the number of hidden units in each layer, whereas here they describe the number of channels. Our goal is then to extend the proofs in Matthews et al. (2018b) (in particular, of theorem 4), to show that the output of our convolutional networks converge in distribution to a Gaussian process as $n  \infty$ , with mean zero and covariance given by the recursion in Eqs. (10 – 12).
+
+The proof in Matthews et al. (2018b) has three main steps. First, they use the Cramér-Wold device, to reduce the full problem to that of proving convergence of scalar random variables to a Gaussian with specified variance. Second, if the previous layers have finite numbers of channels, then the channels ${ \bf a } _ { j } ^ { ( \ell ) } ( { \bf X } )$ and $\mathbf { a } _ { j } ^ { ( \ell ) } ( \mathbf { X } ^ { \prime } )$ are uncorrelated but no longer independent, so we cannot apply the CLT directly, as we did in the main text. Instead, they write the activations as a sum of exchangeable random variables, and derive an adapted CLT for exchangeable (rather than independent) random variables (Blum et al., 1958). Third, they show that moment conditions required by their exchangeable CLT are satisfied.
+
+To extend their proofs to the convolutional case, we begin by defining our networks in a form that is easier to manipulate and as close as possible to Eq. (21-23) in Matthews et al. (2018b),
+
+$$
+\begin{array} { c } { { A _ { i , \mu } ^ { ( 1 ) } = f _ { i , \mu } ^ { ( 1 ) } ( x ) = \displaystyle \frac { \sigma _ { \mathbf { w } } } { \sqrt { C ^ { ( 0 ) } } } \sum _ { j = 1 } ^ { C ^ { ( 0 ) } } \sum _ { \nu \in \mu \mathbb { h } \mathrm { p a t e h } } \epsilon _ { i , j , \mu , \nu } ^ { ( 1 ) } x _ { j , \nu } + b _ { i } ^ { ( 1 ) } , \quad i \in \mathbb { N } } } \\ { { g _ { i , \mu } ^ { ( \ell ) } ( x ) = \phi \left( f _ { i , \mu } ^ { ( \ell ) } ( x ) \right) } } \\ { { A _ { i , \mu } ^ { ( \ell + 1 ) } = f _ { i , \mu } ^ { ( \ell + 1 ) } ( x ) = \displaystyle \frac { \sigma _ { \mathbf { w } } } { \sqrt { C ^ { ( \ell ) } ( n ) } } \sum _ { j = 1 } ^ { C ^ { ( \ell ) } } \sum _ { \nu \in \mu \mathrm { t h } \mathrm { p a t e h } } \epsilon _ { i , j , \mu , \nu } ^ { ( \ell + 1 ) } g _ { j , \nu } ^ { ( \ell ) } ( x ) + b _ { i } ^ { ( \ell + 1 ) } , \quad i \in \mathbb { N } } } \end{array}
+$$
+
+where,
+
+$$
+\epsilon _ { i , j , \mu , \nu } \sim \mathcal { N } ( 0 , 1 ) .
+$$
+
+The first step is to use the Cramér-Wold device (Lemma 6 in Matthews et al., 2018b), which indicates that convergence in distribution of a sequence of finite-dimensional vectors is equivalent to convergence on all possible linear projections to the corresponding real-valued random variable. Mirroring Eq. 25 in Matthews et al. (2018b), we consider convergence of random vectors, $f _ { i , \mu } ^ { ( \ell ) } ( x ) [ n ] - b _ { i } ^ { ( \ell ) }$ , projected onto α(x,i,µ),
+
+$$
+\mathcal { T } ^ { ( \ell ) } \left( \mathcal { L } , \alpha \right) \left[ n \right] = \sum _ { ( x , i , \mu ) \in \mathcal { L } } \alpha ^ { ( x , i , \mu ) } \left[ f _ { i , \mu } ^ { ( \ell ) } ( x ) [ n ] - b _ { i } ^ { ( \ell ) } \right] .
+$$
+
+where $\mathcal { L } \subset \mathcal { X } \times \mathbb { N } \times \{ 1 , \ldots , H ^ { ( \ell ) } D ^ { ( \ell ) } \}$ is a finite set of tuples of data points and channel indicies, $i$ , and indicies of elements within channels/feature maps, $\mu$ . The suffix $[ n ]$ indicates width functions that are instantiated with input, $n$ .
+
+Now, we must prove that these projections converge in distribution a Gaussian. We begin by defining summands, as in Eq. 26 in Matthews et al. (2018b),
+
+$$
+\gamma _ { j } ^ { ( \ell ) } \left( \mathcal { L } , \alpha \right) [ n ] : = \sigma _ { \mathrm { w } } \sum _ { ( x , i , \mu ) \in \mathcal { L } } \alpha ^ { ( x , i , \mu ) } \sum _ { \nu \in \mu \mathrm { t h p a t c h } } \epsilon _ { i , j , \mu , \nu } ^ { ( \ell ) } g _ { j , \nu } ^ { ( \ell - 1 ) } ( x ) [ n ] ,
+$$
+
+such that the projections can be written as a sum of the summands, exactly as in Eq. 27 in Matthews et al. (2018b),
+
+$$
+\mathcal { T } ^ { ( \ell ) } \left( \mathcal { L } , \alpha \right) [ n ] = \frac { 1 } { \sqrt { C ^ { ( \ell - 1 ) } ( n ) } } \sum _ { j = 1 } ^ { C ^ { ( \ell - 1 ) } ( n ) } \gamma _ { j } ^ { ( \ell ) } \left( \mathcal { L } , \alpha \right) [ n ] .
+$$
+
+Now we can apply the exchangeable CLT to prove that ${ \mathcal { T } } ^ { ( \ell ) } \left( { \mathcal { L } } , \alpha \right) [ n ]$ converges to the limiting Gaussian implied by the recursions in the main text. To apply the exchangeable CLT, the first step is to mirror Lemma 8 in Matthews et al. (2018b), in showing that for each fixed $n$ and $\ell \in \{ 2 , \ldots , L +$ $1 \}$ , the summands, $\gamma _ { j } ^ { ( \ell ) } \left( \mathcal { L } , \alpha \right) [ n ]$ are exchangeable with respect to the index $j$ . In particular, we apply de Finetti’s theorem, which states that a sequence of random variables is exchangeable if and only if they are i.i.d. conditional on some set of random variables, so it is sufficient to exhibit such a set of random variables. Mirroring Eq. 29 in Matthews et al. (2018b), we apply the recursion,
+
+$$
+ \overset { ( \ell ) } { j } ( \mathcal { L } , \alpha ) [ n ] : = \sigma _ { \mathrm { w } } \sum _ { ( x , i , \mu ) \in \mathcal { L } } \alpha ^ { ( x , i , \mu ) } \sum _ { \nu \in \mu \mathrm { t h } \mathrm { p a t e h } } \epsilon _ { i , j , \mu , \nu } ^ { ( \ell ) } \phi ( \frac { \sigma _ { \mathrm { w } } } { \sqrt { C ^ { ( \ell - 2 ) } ( n ) } } \sum _ { k = 1 } ^ { C ^ { ( \ell - 2 ) } ( n ) } \sum _ { \xi \in \nu \mathrm { t h } \mathrm { p a t e h } } \epsilon _ { j , k , \nu ; \xi } ^ { ( \ell - 1 ) } g _ { k , \xi } ^ { ( \ell - 2 ) } ) ,
+$$
+
+As such, the summands are iid conditional on the finite set of random variables $\left\{ g _ { k , \xi } ^ { ( \ell - 2 ) } ( x ) [ n ] : k \in \{ 1 , \ldots , C ^ { ( \ell - 2 ) } \} , \xi \in \left\{ 1 , \ldots , H ^ { ( \ell - 2 ) } D ^ { ( \ell - 2 ) } \right\} , x \in \mathcal { L } _ { \mathcal { X } } \right\}$ , where $\mathcal { L } _ { \mathcal { X } }$ is the set of input points in $\mathcal { L }$ .
+
+The exchangeable CLT in Lemma 10 in Matthews et al. (2018b) indicates that ${ \mathcal { T } } ^ { ( \ell ) } \left( { \mathcal { L } } , \alpha \right) [ n ]$ converges in distribution to $\mathcal { N } \left( 0 , \sigma _ { * } ^ { 2 } \right)$ if the summands are exchangeable (which we showed above), and if three conditions hold,
+
+$$
+\begin{array} { r l } & { \mathrm { a ) ~ \mathbb { E } } _ { n } [ \gamma _ { j } ^ { ( \ell ) } \gamma _ { j ^ { \prime } } ^ { ( \ell ) } ] = 0 } \\ & { \mathrm { b ) ~ \operatorname* { l i m } _ { n  \infty } \mathbb { E } } _ { n } [ ( \gamma _ { j } ^ { ( \ell ) } ) ^ { 2 } ( \gamma _ { j ^ { \prime } } ^ { ( \ell ) } ) ^ { 2 } ] = \sigma _ { * } ^ { 4 } } \\ & { \mathrm { c ) ~ \mathbb { E } } _ { n } [ | \gamma _ { j } ^ { ( \ell ) } | ^ { 3 } ] = o ( \sqrt { C ^ { ( \ell ) } ( n ) } ) } \end{array}
+$$
+
+Condition a) follows immediately as the summands are uncorrelated and zero-mean. Conditions b) and c) are more involved as convergence in distribution in the previous layers does not imply convergence in moments for our activation functions.
+
+We begin by considering the extension of Lemma 20 in Matthews et al. (2018b), which allow us to show conditions b) and c) above, even in the case of unbounded but linearly enveloped nonlinearities (Definition 1 in Matthews et al., 2018b). Lembounded by a finite constant independent of states that the eighth moments of . We prove this by induction. Th $f _ { i , \mu } ^ { ( t ) } ( x ) [ n ]$ aree is $n \in \mathbb N$   
+trivial, as $f _ { j , \mu } ^ { ( 1 ) } ( x ) [ n ]$ is Gaussian. Following Matthews et al. (2018b), assume the condition holds up to $\ell - 1$ , and show that the condition holds for layer $\ell$ . Using Eq. (21), we can bound the activations at layer $\ell$ ,
+
+$$
+\mathbb { E } \left[ | f _ { i , \mu } ^ { ( \ell ) } ( x ) [ n ] | ^ { 8 } \right] \leq 2 ^ { 8 - 1 } \mathbb { E } \left[ | b _ { i } ^ { ( \ell ) } | ^ { 8 } + \left| \frac { \sigma _ { \mathbf { w } } } { \sqrt { C ^ { ( \ell - 1 ) } } } \sum _ { j = 1 } ^ { C ^ { ( \ell - 1 ) } ( n ) } \sum _ { \nu \in \mu \mathrm { t h } \mathrm { p a t e h } } \epsilon _ { i , j , \mu , \nu } ^ { ( \ell ) } g _ { j , \nu } ^ { ( \ell - 1 ) } ( x ) [ n ] \right| ^ { 8 } \right]
+$$
+
+Following Eq. 48 in Matthews et al. (2018b), which uses Lemma 19 in Matthews et al. (2018b), we have,
+
+$$
+\begin{array} { r l } { \mathbb { E } \left[ \left| \frac { \sigma _ { \mathrm { w } } } { \sqrt { C ^ { ( \ell - 1 ) } } } \sum _ { j = 1 } ^ { C ^ { ( \ell - 1 ) } ( n ) } \displaystyle \sum _ { \nu \in \mu \mathrm { t h } \mathrm { p a t e h } } \epsilon _ { i , j , \mu , \nu } ^ { ( \ell ) } g _ { j , \nu } ^ { ( \ell - 1 ) } ( x ) [ n ] \right| ^ { 8 } \right] } & { } \\ { = \frac { 2 ^ { 4 } \Gamma ( 4 + 1 / 2 ) } { \Gamma ( 1 / 2 ) } \mathbb { E } \left[ \left| \frac { \sigma _ { \mathrm { w } } ^ { 2 } } { C ^ { ( \ell - 1 ) } ( n ) } \| g _ { j \in \{ 1 , \dots , C ^ { ( \ell - 1 ) } ( n ) \} , \nu \in \mu \mathrm { t h } \mathrm { p a t c h } } ^ { ( \ell - 1 ) } ( x ) [ n ] \| _ { 2 } ^ { 2 } \right| ^ { 4 } \right] . } \end{array}
+$$
+
+wher e g(\`−1)j∈{1,...,C(\`−1)(n)},ν∈µth patch(x)[n] is the set of post-nonlinearities corresponding to j ∈ $\{ 1 , \ldots , C ^ { ( \ell - 1 ) } ( n ) \}$ and $\nu \in \mu \mathrm { t h }$ patch. Following Matthews et al. (2018b), observe that,
+
+$$
+\begin{array} { r } { \frac { 1 } { \sum ^ { ( \ell - 1 ) } \left( n \right) } \| g _ { j \in \{ 1 , \dots , C ^ { ( \ell - 1 ) } ( n ) \} , \nu \in \mu \mathrm { t h p a t e h } } ^ { ( \ell - 1 ) } ( x ) [ n ] \| _ { 2 } ^ { 2 } = \frac { 1 } { C ^ { ( \ell - 1 ) } ( n ) } \displaystyle \sum _ { j = 1 } ^ { C ^ { ( \ell - 1 ) } ( n ) } \sum _ { \nu \in \mu \mathrm { t h p a t e h } } \left( g _ { j , \nu } ^ { ( \ell - 1 ) } ( x ) [ n ] \right) ^ { 2 } } \\ { \leq \frac { 1 } { C ^ { ( \ell - 1 ) } ( n ) } \displaystyle \sum _ { j = 1 } ^ { C ^ { ( \ell - 1 ) } ( n ) } \sum _ { \nu \in \mu \mathrm { t h p a t e h } } ^ { C ^ { ( \ell - 1 ) } ( n ) } \left( c + m \big | f _ { j , \nu } ^ { ( \ell - 1 ) } ( x ) \big | \right. } \end{array}
+$$
+
+by the linear envelope property, $| \phi ( u ) | \leq c + m | u |$ . Following Matthews et al. (2018b), we substitute this bound back into Eq. (28) and suppress a multiplicative constant independent of $x$ and $n$ ,
+
+$$
+\begin{array} { r l } & { \mathbb { E } \left[ \left| \frac { \sigma _ { \mathbf { w } } } { \sqrt { C ^ { ( \ell - 1 ) } ( n ) } } \sum _ { j = 1 } ^ { C ^ { ( \ell - 1 ) } ( n ) } \sum _ { \nu \in \mu \mathrm { t h p a t c h } } \epsilon _ { i , j , \mu , \nu } ^ { ( \ell ) } g _ { j , \nu } ^ { ( \ell - 1 ) } ( x ) [ n ] \right| ^ { 8 } \right] } \\ & { \leq \frac { 1 } { \left( C ^ { ( \ell - 1 ) } ( n ) \right) ^ { 4 } } \mathbb { E } \left[ \left| \sum _ { j = 1 } ^ { C ^ { ( \ell - 1 ) } ( n ) } \sum _ { \nu \in \mu \mathrm { t h p a c h } } c ^ { 2 } + 2 c m | f _ { j , \mu } ^ { ( \ell - 1 ) } ( x ) [ n ] | + m ^ { 2 } | f _ { j , \mu } ^ { ( \ell - 1 ) } ( x ) [ n ] | ^ { 2 } \right| ^ { 4 } \right] } \end{array}
+$$
+
+This can be multiplied out, yielding a weighted sum of expectations of the form,
+
+$$
+\mathbb { E } \left[ | f _ { k , \nu } ^ { ( \ell - 1 ) } ( x ) [ n ] | ^ { p _ { 1 } } | f _ { l , \xi } ^ { ( \ell - 1 ) } ( x ) [ n ] | ^ { p _ { 2 } } | f _ { r , \pi } ^ { ( \ell - 1 ) } ( x ) [ n ] | ^ { p _ { 3 } } | f _ { q , \rho } ^ { ( \ell - 1 ) } ( x ) [ n ] | ^ { p _ { 4 } } \right]
+$$
+
+with $p _ { i } \in \{ 0 , 1 , 2 \}$ for $i = { 1 , 2 , 3 , 4 }$ , and $k , l , r , q \in \{ 1 , \dots , C ^ { ( \ell - 1 ) } ( n ) \}$ , and $\nu , \xi , \pi , \rho \in \mu \mathrm { t h }$ patch where the weights of these terms are independent of (2018b), each of these terms is bounded if the eighth mo $n$ . Usinents of in Matthews et al.are bounded, which $f _ { k , \mu } ^ { ( \ell - 1 ) } ( x ) [ n ]$ $\left( 2 C ^ { ( \ell - 1 ) } ( n ) | \mu \mathrm { t h } \mathrm { p a t c h } | \right) ^ { 4 }$ , where $| \mu \mathrm { t h } \ p a \mathrm { t c h } |$ is the number of elements in a convolutional patch. Thus, we can use the same constant for any due to the $1 / \left( C ^ { ( \ell - 1 ) } ( n ) \right) ^ { 4 }$ scaling. As in Matthews et al. (2018b), noting that $f _ { j , \mu } ^ { ( \ell - 1 ) } ( x ) [ n ]$ are exchangeable over $j$ for any $x$ and $n$ concludes the proof. Using this result, we can obtain a straightforward adaptation of Lemmas 15, 16 and 21 in Matthews et al. (2018b). Lemma 15 gives condition b), Lemma 16 gives condition c); Lemma 15 requires uniform integrability, which is established by Lemma 21.
+
+# 7.3 CALIBRATION OF GAUSSIAN PROCESS UNCERTAINTY
+
+It is important to check that the estimates of uncertainty produced by our Gaussian process are reasonable. However, to make this assessment, we needed to use a proper likelihood, and not the squared-error loss in the main text. We therefore used our kernel to perform the full, multi-class classification problem in GPflow Matthews et al. (2017), with a RobustMax likelihood (Hernándezlobato et al., 2011). The more difficult non-conjugate inference problem forced us to use 1000 inducing points, randomly chosen from the training inputs. Both our kernel and an RBF kernel have similar calibration curves, that closely track the diagonal, indicating accurate uncertainty estimation. However, even in the inducing point setting, our convolutional kernel gave considerably better performance than the RBF kernel $2 . 4 \%$ error vs $3 . 4 \%$ error).
+
+![](images/8864ab9d76c33870f50470fe5185392a7878cda9c47764b2ab0ca5045ac895b3.jpg)  
+Figure 3: Calibration plots for an RBF kernel (left) and the ResNet kernel (right). The $\mathbf { X }$ -axis gives GP prediction for the label probability. The points give corresponding proportion of test points with that label, and the bars give the proportion of training examples in each bin.
\ No newline at end of file
diff --git a/parse/train/Bklfsi0cKm/Bklfsi0cKm_content_list.json b/parse/train/Bklfsi0cKm/Bklfsi0cKm_content_list.json
new file mode 100644
index 0000000000000000000000000000000000000000..73af5de35b368831978beaaaa08f4df8a71c3feb
--- /dev/null
+++ b/parse/train/Bklfsi0cKm/Bklfsi0cKm_content_list.json
@@ -0,0 +1,2138 @@
+[
+    {
+        "type": "text",
+        "text": "DEEP CONVOLUTIONAL NETWORKS AS SHALLOW GAUSSIAN PROCESSES ",
+        "text_level": 1,
+        "bbox": [
+            173,
+            99,
+            648,
+            146
+        ],
+        "page_idx": 0
+    },
+    {
+        "type": "text",
+        "text": "Adrià Garriga-Alonso department of Engineering University of Cambridge ag919@cam.ac.uk ",
+        "bbox": [
+            184,
+            170,
+            361,
+            226
+        ],
+        "page_idx": 0
+    },
+    {
+        "type": "text",
+        "text": "Laurence Aitchison   \nDepartment of Engineering   \nUniversity of Cambridge   \nlaurence.aitchison@gmail.com ",
+        "bbox": [
+            184,
+            247,
+            460,
+            303
+        ],
+        "page_idx": 0
+    },
+    {
+        "type": "text",
+        "text": "ABSTRACT ",
+        "text_level": 1,
+        "bbox": [
+            454,
+            340,
+            544,
+            354
+        ],
+        "page_idx": 0
+    },
+    {
+        "type": "text",
+        "text": "We show that the output of a (residual) convolutional neural network (CNN) with an appropriate prior over the weights and biases is a Gaussian process (GP) in the limit of infinitely many convolutional filters, extending similar results for dense networks. For a CNN, the equivalent kernel can be computed exactly and, unlike “deep kernels”, has very few parameters: only the hyperparameters of the original CNN. Further, we show that this kernel has two properties that allow it to be computed efficiently; the cost of evaluating the kernel for a pair of images is similar to a single forward pass through the original CNN with only one filter per layer. The kernel equivalent to a 32-layer ResNet obtains $0 . 8 4 \\%$ classification error on MNIST, a new record for GPs with a comparable number of parameters. ",
+        "bbox": [
+            233,
+            371,
+            764,
+            511
+        ],
+        "page_idx": 0
+    },
+    {
+        "type": "text",
+        "text": "1 INTRODUCTION ",
+        "text_level": 1,
+        "bbox": [
+            176,
+            539,
+            336,
+            555
+        ],
+        "page_idx": 0
+    },
+    {
+        "type": "text",
+        "text": "Convolutional Neural Networks (CNNs) have powerful pattern-recognition capabilities that have recently given dramatic improvements in important tasks such as image classification (Krizhevsky et al., 2012). However, as CNNs are increasingly being applied in real-world, safety-critical domains, their vulnerability to adversarial examples (Szegedy et al., 2013; Kurakin et al., 2016), and their poor uncertainty estimates are becoming increasingly problematic. Bayesian inference is a theoretically principled and demonstrably successful (Snoek et al., 2012; Deisenroth & Rasmussen, 2011) framework for learning in the face of uncertainty, which may also help to address the problems of adversarial examples (Gal & Smith, 2018). Unfortunately, Bayesian inference in CNNs is extremely difficult due to the very large number of parameters, requiring highly approximate factorised variational approximations (Blundell et al., 2015; Gal & Ghahramani, 2015), or requiring the storage (Lakshminarayanan et al., 2017) of large numbers of posterior samples (Welling & Teh, 2011; Mandt et al., 2017). ",
+        "bbox": [
+            174,
+            571,
+            825,
+            738
+        ],
+        "page_idx": 0
+    },
+    {
+        "type": "text",
+        "text": "Other methods such as those based on Gaussian Processes (GPs) are more amenable to Bayesian inference, allowing us to compute the posterior uncertainty exactly (Rasmussen & Williams, 2006). This raises the question of whether it might be possible to combine the pattern-recognition capabilities of CNNs with exact probabilistic computations in GPs. Two such approaches exist in the literature. First, deep convolutional kernels (Wilson et al., 2016) parameterise a GP kernel using the weights and biases of a CNN, which is used to embed the input images into some latent space before computing their similarity. The CNN parameters of the resulting kernel then have to be optimised by gradient descent. However, the large number of kernel parameters in the CNN reintroduces the risk of overconfidence and overfitting. To avoid this risk, we need to infer a posterior over the CNN kernel parameters, which is as difficult as directly inferring a posterior over the parameters of the original CNN. Second, it is possible to define a convolutional GP (van der Wilk et al., 2017) or a deep convolutional GP (Kumar et al., 2018) by defining a GP that takes an image patch as input, and using that GP as a component in a larger CNN-like system. However, inference in such systems is very computationally expensive, at least without the use of potentially severe variational approximations (van der Wilk et al., 2017). ",
+        "bbox": [
+            174,
+            744,
+            825,
+            897
+        ],
+        "page_idx": 0
+    },
+    {
+        "type": "text",
+        "text": "",
+        "bbox": [
+            174,
+            103,
+            823,
+            159
+        ],
+        "page_idx": 1
+    },
+    {
+        "type": "text",
+        "text": "An alternative approach is suggested by the underlying connection between Bayesian neural networks (NNs) and GPs. In particular, Neal (1996) showed that the function defined by a single-layer fully-connected NN with infinitely many hidden units, and random independent zero-mean weights and biases is equivalent to a GP, implying that we can do exact Bayesian inference in such a NN by working with the equivalent GP. Recently, this result was extended to arbitrarily deep fullyconnected NNs with infinitely many hidden units at each layer (Lee et al., 2017; Matthews et al., 2018a). However, these fully-connected networks are rarely used in practice, as they are unable to exploit important properties of images such as translational invariance, raising the question of whether state-of-the-art architectures such as CNNs (LeCun et al., 1990) and ResNets (He et al., 2016a) have equivalent GP representations. Here, we answer in the affirmative, giving the GP kernel corresponding to arbitrarily deep CNNs and to (convolutional) residual neural networks (He et al., 2016a). In this case, if each hidden layer has an infinite number of convolutional filters, the network prior is equivalent to a GP. ",
+        "bbox": [
+            174,
+            166,
+            825,
+            347
+        ],
+        "page_idx": 1
+    },
+    {
+        "type": "text",
+        "text": "Furthermore, we show that two properties of the GP kernel induced by a CNN allow it to be computed very efficiently. First, in previous work it was necessary to compute the covariance matrix for the output of a single convolutional filter applied at all possible locations within a single image (van der Wilk et al., 2017), which was prohibitively computationally expensive. In contrast, under our prior, the downstream weights are independent with zero-mean, which decorrelates the contribution from each location, and implies that it is necessary only to track the patch variances, and not their covariances. Second, while it is still necessary to compute the variance of the output of a convolutional filter applied at all locations within the image, the specific structure of the kernel induced by the CNN means that the variance at every location can be computed simultaneously and efficiently as a convolution. ",
+        "bbox": [
+            174,
+            354,
+            825,
+            492
+        ],
+        "page_idx": 1
+    },
+    {
+        "type": "text",
+        "text": "Finally, we empirically demonstrate the performance increase coming from adding translationinvariant structure to the GP prior. Without computing any gradients, and without augmenting the training set (e.g. using translations), we obtain $0 . 8 4 \\%$ error rate on the MNIST classification benchmark, setting a new record for nonparametric GP-based methods. ",
+        "bbox": [
+            174,
+            500,
+            823,
+            556
+        ],
+        "page_idx": 1
+    },
+    {
+        "type": "text",
+        "text": "2 GP BEHAVIOUR IN A CNN ",
+        "text_level": 1,
+        "bbox": [
+            176,
+            579,
+            424,
+            595
+        ],
+        "page_idx": 1
+    },
+    {
+        "type": "text",
+        "text": "For clarity of exposition, we will treat the case of a 2D convolutional NN. The result applies straightforwardly to $n \\mathbf { D }$ convolutions, dilated convolutions and upconvolutions (“deconvolutions”), since they can be represented as linear transformations with tied coefficients (see Fig. 1). ",
+        "bbox": [
+            174,
+            613,
+            825,
+            656
+        ],
+        "page_idx": 1
+    },
+    {
+        "type": "text",
+        "text": "2.1 A 2D CONVOLUTIONAL NETWORK PRIOR ",
+        "text_level": 1,
+        "bbox": [
+            176,
+            676,
+            500,
+            690
+        ],
+        "page_idx": 1
+    },
+    {
+        "type": "text",
+        "text": "The network takes an arbitrary input image $\\mathbf { X }$ of height $H ^ { ( 0 ) }$ and width $D ^ { ( 0 ) }$ , as a $C ^ { ( 0 ) } \\times ( H ^ { ( 0 ) } D ^ { ( 0 ) } )$ real matrix. Each row, which we denote $\\mathbf { x } _ { 1 } , \\mathbf { x } _ { 2 } , \\ldots , \\mathbf { x } _ { C ^ { ( 0 ) } }$ , corresponds to a channel of the image (e.g. $C ^ { ( 0 ) } = 3$ for RGB), flattened to form a vector. The first activations $\\mathbf { A } ^ { ( 1 ) } ( \\mathbf { X } )$ are a linear transformation of the inputs. For $i \\in \\{ 1 , \\ldots , C ^ { ( 1 ) } \\}$ : ",
+        "bbox": [
+            173,
+            702,
+            825,
+            763
+        ],
+        "page_idx": 1
+    },
+    {
+        "type": "equation",
+        "img_path": "images/78d2577cfa8c3c860a766e9edaf7965be6e069473e6fa7fd0bb74ab7b7d9c1af.jpg",
+        "text": "$$\n\\mathbf { a } _ { i } ^ { ( 1 ) } ( \\mathbf { X } ) : = b _ { i } ^ { ( 1 ) } \\mathbf { 1 } + \\sum _ { j = 1 } ^ { C ^ { ( 0 ) } } \\mathbf { W } _ { i , j } ^ { ( 1 ) } \\mathbf { x } _ { j } \\ .\n$$",
+        "text_format": "latex",
+        "bbox": [
+            383,
+            773,
+            612,
+            821
+        ],
+        "page_idx": 1
+    },
+    {
+        "type": "text",
+        "text": "We consider a network with $L$ hidden layers. The other activations of the network, from $\\mathbf { A } ^ { ( 2 ) } ( \\mathbf { X } )$ up to $\\mathbf { A } ^ { ( L + 1 ) } ( \\mathbf { X } )$ , are defined recursively: ",
+        "bbox": [
+            173,
+            839,
+            823,
+            871
+        ],
+        "page_idx": 1
+    },
+    {
+        "type": "equation",
+        "img_path": "images/eb19bdb6a9577e101e45b0eeeda139f9ceee459688f7506d7d09bdf291760691.jpg",
+        "text": "$$\n\\mathbf { a } _ { i } ^ { ( \\ell + 1 ) } ( \\mathbf { X } ) : = b _ { i } ^ { ( \\ell + 1 ) } \\mathbf { 1 } + \\sum _ { j = 1 } ^ { C ^ { ( \\ell ) } } \\mathbf { W } _ { i , j } ^ { ( \\ell + 1 ) } \\phi \\left( \\mathbf { a } _ { j } ^ { ( \\ell ) } ( \\mathbf { X } ) \\right) .\n$$",
+        "text_format": "latex",
+        "bbox": [
+            328,
+            881,
+            669,
+            928
+        ],
+        "page_idx": 1
+    },
+    {
+        "type": "image",
+        "img_path": "images/b76e3f5dce59ce270069488ab3ea5f08d6e69a5c74ce89a25358ac556f83a901.jpg",
+        "image_caption": [
+            "Figure 1: The 2D convolution U(i, $\\mathbf { U } _ { i , j } ^ { ( 0 ) } * \\mathbf { x } _ { j }$ 0)j ∗ xj as the dot product W(0)i,j xj . The blank elements of W(0)i,j are zeros. The µth row of W(0)i,j corresponds to applying the filter to the $\\mu$ th convolutional patch of the channel $\\mathbf { x } _ { j }$ . "
+        ],
+        "image_footnote": [],
+        "bbox": [
+            310,
+            98,
+            661,
+            310
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "text",
+        "text": "The activations $\\mathbf { A } ^ { ( \\ell ) } ( \\mathbf { X } )$ are $C ^ { ( \\ell ) } \\times ( H ^ { ( \\ell ) } D ^ { ( \\ell ) } )$ matrices. Each row $\\mathbf { a } _ { i } ^ { ( \\ell + 1 ) }$ represents the flattened $j$ th channel of the image that results from applying a convolutional filter to $\\phi ( \\mathbf { A } ^ { ( \\ell ) } ( \\mathbf { X } ) )$ . ",
+        "bbox": [
+            173,
+            400,
+            823,
+            434
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "text",
+        "text": "and The structure of the pseudo-weight matrices W(\\`+1)i,j , depends on the archite and biases or a convolu $b _ { i } ^ { ( \\ell + 1 ) }$ , for layer, $i \\in \\{ 1 , \\ldots , C ^ { ( \\ell + 1 ) } \\}$ $j \\in \\{ 1 , \\ldots , C ^ { ( \\ell ) } \\}$ $\\mathbf { W } _ { i , j } ^ { ( \\ell + 1 ) }$ xj represents applying the convolutional filter U(\\`+1)i,j to the $j$ th channel. Thus, the elements of each row of where it do W(\\`+1) are: 0 where the filter does not apply and the corresponding element of strated in Fig. 1. $\\mathbf { U } _ { i , j } ^ { ( \\ell + 1 ) }$ ",
+        "bbox": [
+            173,
+            439,
+            825,
+            544
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "text",
+        "text": "The outputs of the network are the last activations, $\\mathbf { A } ^ { ( L + 1 ) } ( \\mathbf { X } )$ . In the classification or regression setting, the outputs are not spatially extended, so we have $H ^ { ( L + 1 ) } = D ^ { ( L + 1 ) } = 1$ , which is equivalent to a fully-connected output layer. In this case, the pseudo-weights W(L+1)i,j only have one row, and the activations $\\mathbf { a } _ { i } ^ { ( L + 1 ) }$ are single-element vectors. ",
+        "bbox": [
+            173,
+            549,
+            825,
+            618
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "text",
+        "text": "Finally, we define the prior distribution over functions by making the filters $\\mathbf { U } _ { i , j } ^ { ( \\ell ) }$ and biases $b _ { i } ^ { ( \\ell ) }$ be independent Gaussian random variables (RVs). For each layer $\\ell$ , channels $i , j$ and locations within the filter $x , y$ : ",
+        "bbox": [
+            173,
+            626,
+            825,
+            672
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "equation",
+        "img_path": "images/2e64ca37f686411e1352f20b7f11f8072dd889ca1dc3f9acbeb4054fe8bcca67.jpg",
+        "text": "$$\nU _ { i , j , x , y } ^ { ( \\ell ) } \\sim \\mathcal { N } \\left( 0 , \\sigma _ { w } ^ { 2 } / C ^ { ( \\ell ) } \\right) , \\qquad b _ { i } ^ { ( \\ell ) } \\sim \\mathcal { N } \\left( 0 , \\sigma _ { b } ^ { 2 } \\right) .\n$$",
+        "text_format": "latex",
+        "bbox": [
+            287,
+            681,
+            712,
+            708
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "text",
+        "text": "Note that, to keep the activation variance constant, the weight variance is divided by the number of input channels. The weight variance can also be divided by the number of elements of the filter, which makes it equivalent to the NN weight initialisation scheme introduced by He et al. (2016a). ",
+        "bbox": [
+            174,
+            717,
+            825,
+            761
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "text",
+        "text": "2.2 ARGUMENT FOR GP BEHAVIOUR ",
+        "text_level": 1,
+        "bbox": [
+            176,
+            780,
+            442,
+            795
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "text",
+        "text": "We follow the proofs by Lee et al. (2017) and Matthews et al. (2018a) to show that the output of the CNN described in the previous section, $\\mathbf { A } ^ { ( L + 1 ) }$ , defines a GP indexed by the inputs, $\\mathbf { X }$ . Their proof (Lee et al., 2017) proceeds by applying the multivariate Central Limit Theorem (CLT) to each layer in sequence, i.e. taking the limit as $N ^ { ( 1 ) }  \\infty$ , then $N ^ { ( 2 ) }  \\infty$ etc, where $N ^ { ( \\ell ) }$ is the number of hidden units in layer $\\ell$ . By analogy, we sequentially apply the multivariate CLT by taking the limit as the number of channels goes to infinity, i.e. $\\dot { C ^ { ( 1 ) } } \\stackrel { - } {  } \\infty$ , then $C ^ { ( 2 ) }  \\infty$ etc. While this is the simplest approach to taking the limits, other potentially more realistic approaches also exist (Matthews et al., 2018a). ",
+        "bbox": [
+            173,
+            808,
+            825,
+            924
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "text",
+        "text": "The fundamental quantity we consider is a vector formed by concatenating the feature maps (or equivalently channels), ${ \\bf a } _ { j } ^ { ( \\ell ) } ( { \\bf X } )$ and $\\mathbf { a } _ { j } ^ { ( \\ell ) } ( \\mathbf { X } ^ { \\prime } )$ from data points $\\mathbf { X }$ and $\\mathbf { X } ^ { \\prime }$ , ",
+        "bbox": [
+            174,
+            102,
+            825,
+            137
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "equation",
+        "img_path": "images/d9800bcb592548e273050ba34b1727b8824d0e237f31f59f087f18e26c388d11.jpg",
+        "text": "$$\n\\begin{array} { r } { { \\bf a } _ { i } ^ { ( \\ell ) } ( { \\bf X } , { \\bf X } ^ { \\prime } ) = \\left( { \\bf a } _ { i } ^ { ( \\ell ) } ( { \\bf X } ) \\right) . } \\end{array}\n$$",
+        "text_format": "latex",
+        "bbox": [
+            401,
+            145,
+            596,
+            188
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "This quantity (and the following arguments) can all be extended to the case of countably finite numbers of input points. ",
+        "bbox": [
+            174,
+            194,
+            823,
+            223
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "Induction base case. For any pair of data points, $\\mathbf { X }$ and $\\mathbf { X } ^ { \\prime }$ the feature-maps corresponding to the $j$ th channel, $\\mathbf { a } _ { j } ^ { ( 1 ) } ( \\mathbf { X } , \\mathbf { X } ^ { \\prime } )$ have a multivariate Gaussian joint distribution. This is because each element is a linear combination of shared Gaussian random variables: the biases, ${ \\bf b } _ { j } ^ { ( 0 ) }$ and the filters, $\\mathbf { U } _ { j , : } ^ { ( 0 ) }$ . Following Eq. (1), ",
+        "bbox": [
+            173,
+            238,
+            825,
+            311
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "equation",
+        "img_path": "images/7c1f702cca405e1c3cc5e8b9a4e3e3432322df0dbd23cd4a844fff3c942657a6.jpg",
+        "text": "$$\n\\mathbf { a } _ { i } ^ { ( 1 ) } ( \\mathbf { X } , \\mathbf { X } ^ { \\prime } ) = b _ { i } ^ { ( 1 ) } \\mathbf { 1 } + \\sum _ { i = 1 } ^ { C ^ { ( 0 ) } } \\left( \\begin{array} { c c } { \\mathbf { W } _ { i , j } ^ { ( 1 ) } } & { \\mathbf { 0 } } \\\\ { \\mathbf { 0 } } & { \\mathbf { W } _ { i , j } ^ { ( 1 ) } } \\end{array} \\right) \\left( \\begin{array} { c } { \\mathbf { x } _ { i } } \\\\ { \\mathbf { x } _ { i } ^ { \\prime } } \\end{array} \\right) ,\n$$",
+        "text_format": "latex",
+        "bbox": [
+            321,
+            320,
+            674,
+            367
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "where 1 is a vector of all-ones. While the elements within a feature map display strong correlations, different feature maps are independent and identically distributed (iid) conditioned on the data (i.e. $\\mathbf { a } _ { i } ^ { ( 1 ) } ( \\mathbf { X } , \\mathbf { X } ^ { \\prime } )$ and $\\mathbf { a } _ { i ^ { \\prime } } ^ { ( \\mathrm { i } ) } ( \\mathbf { X } , \\mathbf { X ^ { \\prime } } )$ are iid for $i \\neq i ^ { \\prime }$ ), because the parameters for different feature-maps (i.e. the biases, $b _ { i } ^ { ( 1 ) }$ and the filters, $\\mathbf { W } _ { i , : } ^ { ( 1 ) }$ ) are themselves iid. ",
+        "bbox": [
+            173,
+            372,
+            825,
+            439
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "Induction step. Consider the feature maps at the \\`th layer, $\\mathbf { a } _ { j } ^ { ( \\ell ) } ( \\mathbf { X } , \\mathbf { X } ^ { \\prime } )$ , to be iid multivariate Gaussian RVs (i.e. for $j \\neq j ^ { \\prime }$ , $\\mathbf { a } _ { j } ^ { ( \\ell ) } ( \\mathbf { X } , \\mathbf { X } ^ { \\prime } )$ and $\\mathbf { a } _ { j ^ { \\prime } } ^ { ( \\ell ) } ( \\mathbf { X } , \\mathbf { X } ^ { \\prime } )$ are iid). Our goal is to show that, taking the number of channels at layer $\\ell$ to infinity (i.e. $C ^ { ( \\ell ) } \\to \\infty ^ { \\cdot }$ ), the same properties hold at the next layer (i.e. all feature maps, a(\\`+1)i (X, X0), are iid multivariate Gaussian RVs). Writing eq. (2) for two training examples, $\\mathbf { X }$ and $\\mathbf { X } ^ { \\prime }$ , we obtain, ",
+        "bbox": [
+            174,
+            455,
+            826,
+            542
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "equation",
+        "img_path": "images/3cc3207e1f6d7550a22043ec02cfccb2c0052e0bfb6d7fedaae056ea3c1ef617.jpg",
+        "text": "$$\n\\mathbf { a } _ { i } ^ { ( \\ell + 1 ) } ( \\mathbf { X } , \\mathbf { X } ^ { \\prime } ) = b _ { i } ^ { ( \\ell + 1 ) } \\mathbf { 1 } + \\sum _ { j = 1 } ^ { C ^ { ( \\ell ) } } \\left( \\begin{array} { c c } { \\mathbf { W } _ { i , j } ^ { ( \\ell + 1 ) } } & { \\mathbf { 0 } } \\\\ { \\mathbf { 0 } } & { \\mathbf { W } _ { i , j } ^ { ( \\ell + 1 ) } } \\end{array} \\right) \\phi ( \\mathbf { a } _ { j } ^ { ( \\ell ) } ( \\mathbf { X } , \\mathbf { X } ^ { \\prime } ) )\n$$",
+        "text_format": "latex",
+        "bbox": [
+            264,
+            550,
+            733,
+            597
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "We begin by showing that $\\mathbf { a } _ { i } ^ { ( \\ell + 1 ) } ( \\mathbf { X } , \\mathbf { X } ^ { \\prime } )$ is a multivariate Gaussian RV. The first term is multivariate Gaussian, as it is a linear function of $b _ { i } ^ { ( \\ell + 1 ) }$ , which is itself iid Gaussian. We can apply the multivariate CLT to show that the second term is also Gaussian, because, in the limit as $C ^ { ( \\ell ) } \\to \\infty$ , it is the sum of infinitely many iid terms: $\\mathbf { a } _ { j } ^ { ( \\ell ) } ( \\mathbf { X } , \\mathbf { X } ^ { \\prime } )$ are iid by assumption, and $\\mathbf { W } _ { i , j } ^ { ( \\ell + 1 ) }$ are iid by definition. Note that the same argument applies to all feature maps jointly, so all elements of $\\mathbf { A } ^ { ( \\ell + 1 ) } ( \\mathbf { X } , \\mathbf { X } ^ { \\prime } )$ (defined by analogy with eq. 4) are jointly multivariate Gaussian. ",
+        "bbox": [
+            173,
+            606,
+            826,
+            704
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "Following Lee et al. (2017), to complete the argument, we need to show that the output feature maps are iid, i.e. $\\mathbf { a } _ { i } ^ { ( \\ell + 1 ) } ( \\mathbf { X } , \\mathbf { X } ^ { \\prime } )$ )(X, X0) and a(\\`i0 $\\mathbf { a } _ { i ^ { \\prime } } ^ { ( \\ell + 1 ) } ( \\mathbf { X } , \\mathbf { X } ^ { \\prime } )$ are iid for $i \\neq i ^ { \\prime }$ . They are identically distributed, re as b(\\`+1) a nd r tha $\\mathbf { W } _ { i , j } ^ { ( \\ell + 1 ) }$ and  and $\\phi ( \\mathbf { a } _ { j } ^ { ( \\ell ) } ( \\mathbf { X } , \\mathbf { X } ^ { \\prime } ) )$ is shared. To show that they are independent,are jointly Gaussian, so it is sufficient to show $\\mathbf { a } _ { i } ^ { ( \\ell + 1 ) } ( \\mathbf { X } , \\mathbf { X } ^ { \\prime } )$ $\\mathbf { a } _ { i ^ { \\prime } } ^ { ( \\ell + 1 ) } ( \\mathbf { X } , \\mathbf { X } ^ { \\prime } )$   \nthat they are uncorrelated, and we can show that they are uncorrelated because the weights, $\\mathbf { W } _ { i , j } ^ { ( \\ell + 1 ) }$ are independent with zero-mean, eliminating any correlations that might arise through the shared RV, $\\phi ( \\mathbf { a } _ { j } ^ { ( \\ell ) } ( \\mathbf { X } , \\mathbf { X } ^ { \\prime } ) )$ . In the appendix, we consider the more complex case where we take limits simultaneously. ",
+        "bbox": [
+            173,
+            708,
+            825,
+            843
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "3 THE CONVNET AND RESNET KERNELS ",
+        "text_level": 1,
+        "bbox": [
+            174,
+            863,
+            532,
+            880
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "Here we derive a computationally efficient kernel corresponding to the CNN described in the previous section. It is surprising that we can compute the kernel efficiently because the feature maps, ${ \\bf a } _ { i } ^ { ( \\ell ) } ( { \\bf X } )$ , display rich covariance structure due to the shared convolutional filter. Computing and representing these covariances would be prohibitively computationally expensive. However, in many cases we only need the variance of the output, e.g. in the case of classification or regression with a final dense layer. It turns out that this propagates backwards through the convolutional network, implying that for every layer, we only need the “diagonal covariance” of the activations: the covariance between the corresponding elements of ${ \\bf a } _ { j } ^ { ( \\ell ) } ( { \\bf X } )$ and $\\mathbf { a } _ { j } ^ { ( \\ell ) } ( \\mathbf { X } ^ { \\prime } )$ (i.e. $\\mathrm { d i a g } \\left( \\mathbb { C } \\left[ \\mathbf { a } _ { j } ^ { ( \\ell ) } ( \\mathbf { X } ) , \\mathbf { a } _ { j } ^ { ( \\ell ) } ( \\mathbf { X } ^ { \\prime } ) \\right] \\right) )$ . ",
+        "bbox": [
+            174,
+            895,
+            823,
+            924
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "",
+        "bbox": [
+            173,
+            102,
+            825,
+            196
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "text",
+        "text": "3.1 GP MEAN AND COVARIANCE ",
+        "text_level": 1,
+        "bbox": [
+            176,
+            212,
+            413,
+            227
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "text",
+        "text": "A GP is completely specified by its mean and covariance (kernel) functions. These give the parameters of the joint Gaussian distribution of the RVs indexed by any two inputs, $\\mathbf { X }$ and $\\mathbf { X } ^ { \\prime }$ . For the purposes of computing the mean and covariance, it is easiest to consider the network as being written entirely in index notation, ",
+        "bbox": [
+            174,
+            238,
+            825,
+            295
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "equation",
+        "img_path": "images/80a3003fe9e92f02bddb811082b168b48e11c15ca923cf028b88f1306e789b49.jpg",
+        "text": "$$\nA _ { i , \\mu } ^ { ( \\ell + 1 ) } ( \\mathbf { X } ) = b _ { i } ^ { ( \\ell + 1 ) } + \\sum _ { j = 1 } ^ { C ^ { ( \\ell ) } } \\sum _ { \\nu = 1 } ^ { H ^ { ( \\ell ) } D ^ { ( \\ell ) } } W _ { i , j , \\mu , \\nu } ^ { ( \\ell + 1 ) } \\phi ( A _ { j , \\nu } ^ { ( \\ell ) } ( \\mathbf { X } ) ) .\n$$",
+        "text_format": "latex",
+        "bbox": [
+            308,
+            296,
+            689,
+            344
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "text",
+        "text": "where $\\ell$ and $\\ell + 1$ denote the input and output layers respectively, $j$ and $i \\in \\{ 1 , \\ldots , C ^ { ( \\ell + 1 ) } \\}$ denote the input and output channels, and $\\nu$ and $\\mu \\in \\{ 1 , \\ldots , H ^ { ( \\ell + 1 ) } D ^ { ( \\ell + 1 ) } \\}$ denote the location within the input and output channel or feature-maps. ",
+        "bbox": [
+            174,
+            348,
+            826,
+            392
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "text",
+        "text": "The mean function is thus easy to compute ",
+        "bbox": [
+            174,
+            398,
+            457,
+            412
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "equation",
+        "img_path": "images/e747a94ec49cda03ffab5ab9caa3a170304fdc5e31eb4da72587b300ad8e8aee.jpg",
+        "text": "$$\n\\mathbb { E } \\left[ A _ { i , \\mu } ^ { ( \\ell + 1 ) } ( \\mathbf { X } ) \\right] = \\mathbb { E } \\left[ b _ { i } ^ { ( \\ell + 1 ) } \\right] + \\sum _ { j = 1 } ^ { C ^ { ( \\ell ) } } \\sum _ { \\nu = 1 } ^ { H ^ { ( \\ell ) } D ^ { ( \\ell ) } } \\mathbb { E } \\left[ W _ { i , j , \\mu , \\nu } ^ { ( \\ell + 1 ) } \\phi ( A _ { j , \\nu } ^ { ( \\ell ) } ( \\mathbf { X } ) ) \\right] = 0 .\n$$",
+        "text_format": "latex",
+        "bbox": [
+            250,
+            415,
+            746,
+            463
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "text",
+        "text": "as b(\\`+1)i a nd W (\\`+1)i,j,µ,ν have zero mean, and W (\\`+1)i,j,ν,µ are independent of the activations at the previous layer, $\\phi ( A _ { j , \\nu } ^ { ( \\ell ) } ( \\mathbf { X } ) )$ . ",
+        "bbox": [
+            176,
+            465,
+            821,
+            505
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "text",
+        "text": "Now we show that it is possible to efficiently compute the covariance function. This is surprising because for many networks, we nlocations in the feature map (i.e. $\\mathbb { C } \\left[ A _ { i , \\mu } ^ { ( \\ell + 1 ) } ( \\mathbf { \\bar { X } } ) , A _ { i , \\mu ^ { \\prime } } ^ { ( \\ell + 1 ) } ( \\mathbf { X } ^ { \\prime } ) \\right]$ ce ofor $\\mu , \\mu ^ { \\prime } \\in \\{ 1 , \\ldots , H ^ { ( \\ell + 1 ) } D ^ { ( \\ell + 1 ) } \\} )$ and this object is extremely high-dimensional, $N ^ { 2 } ( H ^ { ( \\ell + 1 ) } D ^ { \\overline { { ( \\ell + 1 ) } } } ) ^ { 2 }$ . However, it turns out that we only need to consider the “diagonal” covariance, (i.e. we only need $\\mathbb { C } \\left[ A _ { i , \\mu } ^ { ( \\ell + 1 ) } ( \\mathbf { X } ) , A _ { i , \\mu } ^ { ( \\ell + 1 ) } ( \\mathbf { X } ^ { \\prime } ) \\right]$ fo r $\\mu \\in \\{ 1 , \\ldots , H ^ { ( \\ell + 1 ) } D ^ { ( \\ell + 1 ) } \\} )$ , which is a more manageable quantity of size $N ^ { 2 } ( H ^ { ( \\ell + 1 ) } D ^ { ( \\ell + 1 ) } )$ . ",
+        "bbox": [
+            173,
+            508,
+            826,
+            617
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "text",
+        "text": "This is true at the output layer $( L + 1 )$ : in order to achieve an output suitable for classification or regression, we use only a single output location $H ^ { ( L + 1 ) } = D ^ { ( L + 1 ) } = 1$ , with a number of “channels” equal to the number of of outputs/classes, so it is only possible to compute the covariance at that single location. We now show that, if we only need the covariance at corresponding locations in the outputs, we only need the covariance at corresponding locations in the inputs, and this requirement propagates backwards through the network. ",
+        "bbox": [
+            173,
+            622,
+            825,
+            709
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "text",
+        "text": "Formally, as the activations are composed of a sum of terms, their covariance is the sum of the covariances of all those underlying terms, ",
+        "bbox": [
+            171,
+            715,
+            823,
+            744
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "equation",
+        "img_path": "images/69d31cc97d2fc0067ded1890ef2527e60cbaecb75da932e44d6be86c09e39c9e.jpg",
+        "text": "$$\n\\begin{array} { r l } { \\mathbb { C } \\left[ A _ { i , \\mu } ^ { ( \\ell + 1 ) } ( \\mathbf { X } ) , A _ { i , \\mu } ^ { ( \\ell + 1 ) } ( \\mathbf { X } ^ { \\prime } ) \\right] = \\mathbb { V } \\left[ b _ { i } ^ { ( \\ell ) } \\right] + } & { } \\\\ { \\displaystyle \\sum _ { j = 1 } ^ { C ^ { ( \\ell ) } } \\sum _ { j ^ { \\prime } = 1 } ^ { C ^ { ( \\ell ) } } \\sum _ { \\nu = 1 } ^ { H ^ { ( \\ell ) } D ^ { ( \\ell ) } } \\sum _ { \\nu ^ { \\prime } = 1 } ^ { H ^ { ( \\ell ) } D ^ { ( \\ell ) } } \\mathbb { C } \\left[ W _ { i , j , \\mu , \\nu } ^ { ( \\ell + 1 ) } \\phi ( A _ { j , \\nu } ^ { ( \\ell ) } ( \\mathbf { X } ) ) , W _ { i , j ^ { \\prime } , \\mu , \\nu ^ { \\prime } } ^ { ( \\ell + 1 ) } \\phi ( A _ { j ^ { \\prime } , \\nu ^ { \\prime } } ^ { ( \\ell ) } ( \\mathbf { X } ^ { \\prime } ) ) \\right] . } \\end{array}\n$$",
+        "text_format": "latex",
+        "bbox": [
+            222,
+            744,
+            779,
+            821
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "text",
+        "text": "As the terms in the covariance have mean zero, and as the weights and activations from the previous layer are independent, ",
+        "bbox": [
+            174,
+            823,
+            828,
+            851
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "equation",
+        "img_path": "images/3654c365438b76a29a2541724e15f6857f74e810a656fd4cf794da4018587cad.jpg",
+        "text": "$$\n\\begin{array} { r l } { \\mathbb { C } \\left[ A _ { i , \\mu } ^ { ( \\ell + 1 ) } ( \\mathbf { X } ) , A _ { i , \\mu } ^ { ( \\ell + 1 ) } ( \\mathbf { X } ^ { \\prime } ) \\right] = \\sigma _ { \\mathrm { b } } ^ { 2 } + } & { } \\\\ { \\quad } & { \\qquad \\sum _ { j = 1 } ^ { { \\ell ^ { ( \\ell ) } } } \\displaystyle \\sum _ { j ^ { \\prime } = 1 } ^ { { \\ell ^ { ( \\ell ) } } } \\sum _ { \\nu = 1 } ^ { { H ^ { ( \\ell ) } } D ^ { ( \\ell ) } } \\sum _ { \\nu ^ { \\prime } = 1 } ^ { H ^ { ( \\ell ) } D ^ { ( \\ell ) } } \\mathbb { E } \\left[ W _ { i , j , \\mu , \\nu } ^ { ( \\ell + 1 ) } W _ { i , j ^ { \\prime } , \\mu , \\nu ^ { \\prime } } ^ { ( \\ell + 1 ) } \\right] \\mathbb { E } \\left[ \\phi ( A _ { j , \\nu } ^ { ( \\ell ) } ( \\mathbf { X } ) ) \\phi ( A _ { j ^ { \\prime } , \\nu ^ { \\prime } } ^ { ( \\ell ) } ( \\mathbf { X } ^ { \\prime } ) ) \\right] . } \\end{array}\n$$",
+        "text_format": "latex",
+        "bbox": [
+            199,
+            852,
+            781,
+            929
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "text",
+        "text": "Algorithm 1 The ConvNet kernel $k ( \\mathbf { X } , \\mathbf { X } ^ { \\prime } )$ ",
+        "text_level": 1,
+        "bbox": [
+            176,
+            103,
+            462,
+            119
+        ],
+        "page_idx": 5
+    },
+    {
+        "type": "text",
+        "text": "1: Input: two images, X, X0 ∈ RC(0)×(H(0)W (0)).   \n2: Compute $K _ { \\mu } ^ { ( 1 ) } ( { \\bf X } , { \\bf X } )$ , $K _ { \\mu } ^ { ( 1 ) } ( { \\bf X } , { \\bf X ^ { \\prime } } )$ , and $K _ { \\mu } ^ { ( 1 ) } ( \\mathbf { X } ^ { \\prime } , \\mathbf { X } ^ { \\prime } )$ for $\\mu \\in \\{ 1 , \\ldots , H ^ { ( 1 ) } D ^ { ( 1 ) } \\}$ ; using Eq. (10).   \n3: for $\\ell = 1 , 2 , \\dots , L$ do   \n4: Compute $V _ { \\mu } ^ { ( \\ell ) } ( { \\bf X } , { \\bf X ^ { \\prime } } )$ , $V _ { \\mu } ^ { ( \\ell ) } ( { \\bf X } , { \\bf X ^ { \\prime } } )$ and $V _ { \\mu } ^ { ( \\ell ) } ( { \\bf X } , { \\bf X ^ { \\prime } } )$ for $\\mu \\in \\{ 1 , \\ldots , H ^ { ( \\ell ) } D ^ { ( \\ell ) } \\}$ ; using   \n5: Compute Eq. (13), or some other nonlinearity. $K _ { \\mu } ^ { ( \\ell + 1 ) } ( { \\mathbf { X } } , { \\mathbf { X } } )$ , $K _ { \\mu } ^ { ( \\ell + 1 ) } ( { \\bf X } , { \\bf X ^ { \\prime } } )$ , and $K _ { \\mu } ^ { ( \\ell + 1 ) } ( { \\mathbf { X } } ^ { \\prime } , { \\mathbf { X } } ^ { \\prime } )$ for $\\mu$ ∈ $\\{ 1 , \\ldots , H ^ { ( \\ell + 1 ) } D ^ { ( \\ell + 1 ) } \\}$ ; using Eq. (11). ",
+        "bbox": [
+            179,
+            122,
+            826,
+            251
+        ],
+        "page_idx": 5
+    },
+    {
+        "type": "text",
+        "text": "6: end for   \n7: Output the scalar $K _ { 1 } ^ { ( L + 1 ) } ( { \\bf X } , { \\bf X ^ { \\prime } } )$ . ",
+        "bbox": [
+            179,
+            250,
+            428,
+            277
+        ],
+        "page_idx": 5
+    },
+    {
+        "type": "text",
+        "text": "nt ffor ent channels: Further, each $\\mathbf { W } _ { i , j } ^ { ( \\ell + 1 ) }$ and of the $\\mathbf { W } _ { i , j ^ { \\prime } } ^ { ( \\ell + 1 ) }$ are iid f matrices $j \\neq j ^ { \\prime }$ , sonly contains independent variables or zeros (Fig. 1), so $\\mathbb { E } \\left[ W _ { i , j , \\mu , \\nu } ^ { ( \\ell + 1 ) } W _ { i , j ^ { \\prime } , \\mu , \\nu ^ { \\prime } } ^ { ( \\ell + 1 ) } \\right] = 0$ $j \\neq j ^ { \\prime }$ $\\mathbb { E } \\left[ W _ { i , j , \\mu , \\nu } ^ { ( \\ell + 1 ) } W _ { i , j ^ { \\prime } , \\mu , \\nu ^ { \\prime } } ^ { ( \\ell + 1 ) } \\right] = 0$ $\\mu$ for $\\nu \\neq \\nu ^ { \\prime }$ $\\mathbf { W } _ { i , j } ^ { ( \\ell + 1 ) }$ . Thus, we can eliminate the sums over $j ^ { \\prime }$ and $\\nu ^ { \\prime }$ : ",
+        "bbox": [
+            173,
+            305,
+            826,
+            388
+        ],
+        "page_idx": 5
+    },
+    {
+        "type": "equation",
+        "img_path": "images/24bda7afff24b06f4b1c3e39fbc4a4f4d5ba9a86f7c69a2de5b9bc08f7b317b8.jpg",
+        "text": "$$\n\\Sigma \\left[ A _ { i , \\mu } ^ { ( \\ell + 1 ) } ( \\mathbf { X } ) , A _ { i , \\mu } ^ { ( \\ell + 1 ) } ( \\mathbf { X ^ { \\prime } } ) \\right] = \\sigma _ { \\mathrm { b } } ^ { ( \\ell ) } + \\sum _ { j = 1 } ^ { Q ^ { ( \\ell ) } } \\sum _ { \\nu = 1 } ^ { H ^ { ( \\ell ) } D ^ { ( \\ell ) } } \\mathbb { E } \\left[ W _ { i , j , \\mu , \\nu } ^ { ( \\ell + 1 ) } W _ { i , j , \\mu , \\nu } ^ { ( \\ell + 1 ) } \\right] \\mathbb { E } \\left[ \\phi ( A _ { j , \\nu } ^ { ( \\ell ) } ( \\mathbf { X } ) ) \\phi ( A _ { j , \\nu } ^ { ( \\ell ) } ( \\mathbf { X ^ { \\prime } } ) ) \\right] .\n$$",
+        "text_format": "latex",
+        "bbox": [
+            181,
+            395,
+            833,
+            443
+        ],
+        "page_idx": 5
+    },
+    {
+        "type": "text",
+        "text": "The µth row of W(\\`+1)i,j i s zero for indices $\\nu$ that do not belong to its convolutional patch, so we can restrict the sum over $\\nu$ to that region. We also define $v _ { g } ^ { ( 1 ) } ( \\mathbf { X } , \\mathbf { X } ^ { \\prime } )$ , to emphasise that the covariances are independent of the output channel, $j$ . The variance of the first layer is ",
+        "bbox": [
+            174,
+            460,
+            825,
+            512
+        ],
+        "page_idx": 5
+    },
+    {
+        "type": "equation",
+        "img_path": "images/a04654dd8ea70692c62a0660905bb02e1a41cd6a0d55334a6d1907128435ddcf.jpg",
+        "text": "$$\nK _ { \\mu } ^ { ( 1 ) } ( \\mathbf { X } , \\mathbf { X ^ { \\prime } } ) = \\mathbb { C } \\left[ A _ { i , \\mu } ^ { ( 1 ) } ( \\mathbf { X } ) , A _ { i , \\mu } ^ { ( 1 ) } ( \\mathbf { X ^ { \\prime } } ) \\right] = \\sigma _ { \\mathrm { b } } ^ { 2 } + \\frac { \\sigma _ { \\mathrm { w } } ^ { 2 } } { C ^ { ( 0 ) } } \\sum _ { i = 1 } ^ { C ^ { ( 0 ) } } \\sum _ { \\nu \\in \\mu \\mathrm { t h } \\mathrm { p a t c h } } X _ { i , \\nu } X _ { i , \\nu } ^ { \\prime } .\n$$",
+        "text_format": "latex",
+        "bbox": [
+            246,
+            518,
+            769,
+            566
+        ],
+        "page_idx": 5
+    },
+    {
+        "type": "text",
+        "text": "And we do the same for the other layers, ",
+        "bbox": [
+            174,
+            577,
+            441,
+            592
+        ],
+        "page_idx": 5
+    },
+    {
+        "type": "equation",
+        "img_path": "images/549433f5673785abcc4cef3a361b3221d508b66c2586e63790f0e4e64aa1c2fe.jpg",
+        "text": "$$\nK _ { \\mu } ^ { ( \\ell + 1 ) } ( \\mathbf { X } , \\mathbf { X } ^ { \\prime } ) = \\mathbb { C } \\left[ A _ { i , \\mu } ^ { ( \\ell + 1 ) } ( \\mathbf { X } ) , A _ { i , \\mu } ^ { ( \\ell + 1 ) } ( \\mathbf { X } ^ { \\prime } ) \\right] = \\sigma _ { \\mathrm { b } } ^ { 2 } + \\sigma _ { \\mathrm { w } } ^ { 2 } \\sum _ { \\nu \\in \\mu \\mathrm { t h p a t c h } } V _ { \\nu } ^ { ( \\ell ) } ( \\mathbf { X } , \\mathbf { X } ^ { \\prime } ) ,\n$$",
+        "text_format": "latex",
+        "bbox": [
+            200,
+            602,
+            743,
+            638
+        ],
+        "page_idx": 5
+    },
+    {
+        "type": "text",
+        "text": "where ",
+        "bbox": [
+            173,
+            646,
+            217,
+            660
+        ],
+        "page_idx": 5
+    },
+    {
+        "type": "equation",
+        "img_path": "images/f76318a4bc7d12e24edd6391255fdb84c2a51867f3242d20b9177e93df8867a9.jpg",
+        "text": "$$\nV _ { \\nu } ^ { ( \\ell ) } ( \\mathbf { X } , \\mathbf { X } ^ { \\prime } ) = \\mathbb { E } \\left[ \\phi ( A _ { j , \\nu } ^ { ( \\ell ) } ( \\mathbf { X } ) ) \\phi ( A _ { j , \\nu } ^ { ( \\ell ) } ( \\mathbf { X } ^ { \\prime } ) ) \\right]\n$$",
+        "text_format": "latex",
+        "bbox": [
+            349,
+            656,
+            648,
+            684
+        ],
+        "page_idx": 5
+    },
+    {
+        "type": "text",
+        "text": "is the covariance of the activations, which is again independent of the channel. ",
+        "bbox": [
+            174,
+            685,
+            686,
+            702
+        ],
+        "page_idx": 5
+    },
+    {
+        "type": "text",
+        "text": "3.2 COVARIANCE OF THE ACTIVITIES ",
+        "text_level": 1,
+        "bbox": [
+            176,
+            718,
+            446,
+            733
+        ],
+        "page_idx": 5
+    },
+    {
+        "type": "text",
+        "text": "The elementwise covariance in the right-hand side of Eq. (11) can be computed in closed form for many choices of $\\phi$ if the activations are Gaussian. For each element of the activations, one needs to keep track of the 3 distinct entries of the bivariate covariance matrix between the inputs, $K _ { \\mu } ^ { ( \\ell + 1 ) } ( { \\mathbf { X } } , { \\mathbf { X } } ) , K _ { \\mu } ^ { ( \\ell + 1 ) } ( { \\mathbf { X } } , { \\mathbf { X } } ^ { \\prime } )$ and $K _ { \\mu } ^ { ( \\ell + 1 ) } ( { \\mathbf { X } } ^ { \\prime } , { \\mathbf { X } } ^ { \\prime } )$ . ",
+        "bbox": [
+            173,
+            743,
+            826,
+            805
+        ],
+        "page_idx": 5
+    },
+    {
+        "type": "text",
+        "text": "For example, for the ReLU nonlinearity $( \\phi ( x ) = \\operatorname* { m a x } ( 0 , x ) )$ , one can adapt Cho & Saul (2009) in the same way as Matthews et al. (2018a, section 3) to obtain ",
+        "bbox": [
+            171,
+            810,
+            826,
+            839
+        ],
+        "page_idx": 5
+    },
+    {
+        "type": "equation",
+        "img_path": "images/f0970a595d7eac2c5d1aee2436da987cea6ee155b47b22a44fa6ced9b03dab1d.jpg",
+        "text": "$$\nV _ { \\nu } ^ { ( \\ell ) } ( { \\mathbf { X } } , { \\mathbf { X ^ { \\prime } } } ) = \\frac { \\sqrt { K _ { \\nu } ^ { ( \\ell ) } ( { \\mathbf { X } } , { \\mathbf { X } } ) K _ { \\nu } ^ { ( \\ell ) } ( { \\mathbf { X ^ { \\prime } } } , { \\mathbf { X ^ { \\prime } } } ) } } { \\pi } \\left( \\sin \\theta _ { \\nu } ^ { ( \\ell ) } + ( \\pi - \\theta _ { \\nu } ^ { ( \\ell ) } ) \\cos \\theta _ { \\nu } ^ { ( \\ell ) } \\right)\n$$",
+        "text_format": "latex",
+        "bbox": [
+            246,
+            847,
+            750,
+            888
+        ],
+        "page_idx": 5
+    },
+    {
+        "type": "text",
+        "text": "where $\\theta _ { \\nu } ^ { ( \\ell ) } = \\cos ^ { - 1 } \\bigg ( K _ { \\nu } ^ { ( \\ell ) } ( { \\mathbf X } , { \\mathbf X ^ { \\prime } } ) / \\sqrt { K _ { \\nu } ^ { ( \\ell ) } ( { \\mathbf X } , { \\mathbf X } ) K _ { \\nu } ^ { ( \\ell ) } ( { \\mathbf X ^ { \\prime } } , { \\mathbf X ^ { \\prime } } ) } \\bigg ) .$ ",
+        "bbox": [
+            173,
+            895,
+            614,
+            928
+        ],
+        "page_idx": 5
+    },
+    {
+        "type": "text",
+        "text": "3.3 EFFICIENCY OF THE CONVNET KERNEL ",
+        "text_level": 1,
+        "bbox": [
+            174,
+            103,
+            488,
+            118
+        ],
+        "page_idx": 6
+    },
+    {
+        "type": "text",
+        "text": "We now have all the pieces for computing the kernel, as written in Algorithm 1. ",
+        "bbox": [
+            176,
+            128,
+            692,
+            145
+        ],
+        "page_idx": 6
+    },
+    {
+        "type": "text",
+        "text": "Putting together Eq. (11) and Eq. (13) gives us the surprising result that the diagonal covariances of the activations at layer $\\ell + 1$ only depend on the diagonal covariances of the activations at layer $\\ell$ . This is very important, because it makes the computational cost of the kernel be within a constant factor of the cost of a forward pass for the equivalent CNN with 1 filter per layer. ",
+        "bbox": [
+            174,
+            150,
+            825,
+            207
+        ],
+        "page_idx": 6
+    },
+    {
+        "type": "text",
+        "text": "Thus, the algorithm is more efficient that one would naively think. A priori, one needs to compute the covariance between all the elements of ${ \\bf a } _ { j } ^ { ( \\ell ) } ( { \\bf X } )$ and $\\mathbf { a } _ { j } ^ { ( \\ell ) } ( \\mathbf { X } ^ { \\prime } )$ combined, yielding a $2 H ^ { ( \\ell ) } D ^ { ( \\ell ) } \\times$ $2 H ^ { ( \\ell ) } D ^ { ( \\ell ) }$ covariance matrix for every pair of points. Instead, we only need to keep track of a $H ^ { ( \\ell ) } D ^ { ( \\ell ) }$ -dimensional vector per layer and pair of points. ",
+        "bbox": [
+            174,
+            213,
+            825,
+            279
+        ],
+        "page_idx": 6
+    },
+    {
+        "type": "text",
+        "text": "Furthermore, the particular form for the kernel (eq. 1 and eq. 2) implies that the required variances and covariances at all required locations can be computed efficiently as a convolution. ",
+        "bbox": [
+            174,
+            285,
+            823,
+            314
+        ],
+        "page_idx": 6
+    },
+    {
+        "type": "text",
+        "text": "3.4 KERNEL FOR A RESIDUAL CNN ",
+        "text_level": 1,
+        "bbox": [
+            176,
+            330,
+            434,
+            344
+        ],
+        "page_idx": 6
+    },
+    {
+        "type": "text",
+        "text": "The induction step in the argument for GP behaviour from Sec. 2.2 depends only on the previous activations being iid Gaussian. Since all the activations are iid Gaussian, we can add skip connections between the activations of different layers while preserving GP behaviour, e.g. $\\mathbf { A } ^ { ( \\ell + 1 ) }$ and $\\mathbf { A } ^ { ( \\ell - s ) }$ where $s$ is the number of layers that the skip connection spans. If we change the NN recursion (Eq. 2) to ",
+        "bbox": [
+            173,
+            356,
+            825,
+            428
+        ],
+        "page_idx": 6
+    },
+    {
+        "type": "equation",
+        "img_path": "images/bf1c605df69c210dd72d42c616d1e888a45f018936a2f58c5328d55704234552.jpg",
+        "text": "$$\n\\mathbf { a } _ { i } ^ { ( \\ell + 1 ) } ( \\mathbf { X } ) : = \\mathbf { a } _ { i } ^ { ( \\ell - s ) } ( \\mathbf { X } ) + \\mathbf { b } _ { i } ^ { ( \\ell + 1 ) } + \\sum _ { j = 1 } ^ { C ^ { ( \\ell ) } } \\mathbf { W } _ { i , j } ^ { ( \\ell ) } \\phi \\left( \\mathbf { a } _ { j } ^ { ( \\ell ) } ( \\mathbf { X } ) \\right) ,\n$$",
+        "text_format": "latex",
+        "bbox": [
+            295,
+            425,
+            700,
+            473
+        ],
+        "page_idx": 6
+    },
+    {
+        "type": "text",
+        "text": "then the kernel recursion (Eq. 11) becomes ",
+        "bbox": [
+            174,
+            474,
+            457,
+            489
+        ],
+        "page_idx": 6
+    },
+    {
+        "type": "equation",
+        "img_path": "images/6bb7fa338ac056fae2ed317c9d8fe2d9b4ea34d60134d81a69cef145b31d3492.jpg",
+        "text": "$$\nK _ { \\mu } ^ { ( \\ell + 1 ) } ( \\mathbf { X } , \\mathbf { X } ^ { \\prime } ) = K _ { \\mu } ^ { ( \\ell - s ) } ( \\mathbf { X } , \\mathbf { X } ^ { \\prime } ) + \\sigma _ { b } ^ { 2 } + \\sigma _ { w } ^ { 2 } \\sum _ { \\nu \\in \\mu \\mathrm { t h } \\mathrm { p a t c h } } V _ { \\nu } ^ { ( \\ell ) } ( \\mathbf { X } , \\mathbf { X } ^ { \\prime } ) .\n$$",
+        "text_format": "latex",
+        "bbox": [
+            269,
+            494,
+            727,
+            530
+        ],
+        "page_idx": 6
+    },
+    {
+        "type": "text",
+        "text": "This way of adding skip connections is equivalent to the “pre-activation” shortcuts described by He et al. (2016b). Remarkably, the natural way of adding residual connections to NNs is the one that performed best in their empirical evaluations. ",
+        "bbox": [
+            174,
+            536,
+            825,
+            579
+        ],
+        "page_idx": 6
+    },
+    {
+        "type": "text",
+        "text": "4 EXPERIMENTS ",
+        "text_level": 1,
+        "bbox": [
+            174,
+            598,
+            326,
+            614
+        ],
+        "page_idx": 6
+    },
+    {
+        "type": "text",
+        "text": "We evaluate our kernel on the MNIST handwritten digit classification task. Classification likelihoods are not conjugate for GPs, so we must make an approximation, and we follow Lee et al. (2017), in re-framing classification as multi-output regression. ",
+        "bbox": [
+            176,
+            630,
+            823,
+            672
+        ],
+        "page_idx": 6
+    },
+    {
+        "type": "text",
+        "text": "The training set is split into $N = 5 0 0 0 0$ training and 10000 validation examples. The regression targets $\\mathbf { Y } \\in \\{ - 1 , 1 \\} ^ { \\overline { { N } } \\times 1 0 }$ are a one-hot encoding of the example’s class: $y _ { n , c } = 1$ if the $n$ th example belongs to class $c$ , and $- 1$ otherwise. ",
+        "bbox": [
+            176,
+            679,
+            821,
+            720
+        ],
+        "page_idx": 6
+    },
+    {
+        "type": "text",
+        "text": "Training is exact conjugate likelihood GP regression with noiseless targets $\\mathbf { Y }$ (Rasmussen & Williams, 2006). First we compute the $N \\times N$ kernel matrix ${ \\bf K } _ { x x }$ , which contains the kernel between every pair of examples. Then we compute ${ \\mathbf { K } } _ { x x } ^ { - 1 } { \\mathbf { Y } }$ using a linear system solver. ",
+        "bbox": [
+            174,
+            728,
+            820,
+            771
+        ],
+        "page_idx": 6
+    },
+    {
+        "type": "text",
+        "text": "The test set has $N _ { T } = 1 0 0 0 0$ examples. We compute the $N _ { T } \\times N$ matrix $\\mathbf { K } _ { x ^ { * } x }$ , the kernel between each test example and all the training examples. The predictions are given by the row-wise maximum of $\\mathbf { K } _ { x ^ { * } x } \\mathbf { K } _ { x x } ^ { - 1 } \\bar { \\mathbf { Y } }$ . ",
+        "bbox": [
+            176,
+            776,
+            820,
+            820
+        ],
+        "page_idx": 6
+    },
+    {
+        "type": "text",
+        "text": "For the “ConvNet GP” and “Residual CNN GP”, (Table 1) we optimise the kernel hyperparameters by random search. We draw $M$ random hyperparameter samples, compute the resulting kernel’s performance in the validation set, and pick the highest performing run. The kernel hyperparameters are: $\\sigma _ { b } ^ { 2 }$ , $\\sigma _ { w } ^ { 2 }$ ; the number of layers; the convolution stride, filter sizes and edge behaviour; the nonlinearity (we consider the error function and ReLU); and the frequency of residual skip connections (for Residual CNN GPs). We do not retrain the model on the validation set after choosing hyperparameters. ",
+        "bbox": [
+            173,
+            825,
+            825,
+            924
+        ],
+        "page_idx": 6
+    },
+    {
+        "type": "table",
+        "img_path": "images/f762ab560f768795c1ec5b70d1189c780fac2c18790fd6deb0de130f54ee163d.jpg",
+        "table_caption": [
+            "Table 1: MNIST classification results. #samples gives the number of kernels that were randomly sampled for the hyperparameter search. “ConvNet GP” and “Residual CNN GP” are random CNN architectures with a fixed filter size, whereas “ResNet GP” is a slight modification of the architecture by He et al. (2016b). Entries labelled “SGD” used stochastic gradient descent for tuning hyperparameters, by maximising the likelihood of the training set. The last two methods use parametric neural networks. The hyperparameters of the ResNet GP were not optimised (they were fixed based on the architecture from He et al., 2016b). "
+        ],
+        "table_footnote": [],
+        "table_body": "<table><tr><td>Method</td><td>#samples</td><td>Validation error</td><td>Test error</td></tr><tr><td>NNGP (Lee et al., 2017)</td><td>~250</td><td>1</td><td>1.21%</td></tr><tr><td>Convolutional GP (van der Wilk et al., 2017)</td><td>SGD</td><td></td><td>1.17%</td></tr><tr><td>Deep Conv. GP (Kumar et al., 2018)</td><td>SGD</td><td>1</td><td>1.34%</td></tr><tr><td>ConvNetGP</td><td>27</td><td>0.71%</td><td>1.03%</td></tr><tr><td>Residual CNN GP</td><td>27</td><td>0.72%</td><td>0.96%</td></tr><tr><td>ResNet GP</td><td>1</td><td>0.68%</td><td>0.84%</td></tr><tr><td>GP + parametric deep kernel (Bradshaw et al., 2017)</td><td>SGD</td><td>1</td><td>0.60%</td></tr><tr><td>ResNet (Chen et al.,2018)</td><td>一</td><td></td><td>0.41%</td></tr></table>",
+        "bbox": [
+            173,
+            101,
+            841,
+            229
+        ],
+        "page_idx": 7
+    },
+    {
+        "type": "text",
+        "text": "The “ResNet GP” (Table 1) is the kernel equivalent to a 32-layer version of the basic residual architecture by He et al. (2016a). The differences are: an initial $3 \\times 3$ convolutional layer and a final dense layer instead of average pooling. We chose to remove the pooling because computing its output variance requires the off-diagonal elements of the filter covariance, in which case we could not exploit the efficiency gains described in Sec. 3.3. ",
+        "bbox": [
+            174,
+            368,
+            823,
+            439
+        ],
+        "page_idx": 7
+    },
+    {
+        "type": "text",
+        "text": "We found that the despite it not being optimised, the 32-layer ResNet GP outperformed all other comparable architectures (Table 1), including the NNGP in Lee et al. (2017), which is state-ofthe-art for non-convolutional networks, and convolutional GPs (van der Wilk et al., 2017; Kumar et al., 2018). That said, our results have not reached state-of-the-art for methods that incorporate a parametric neural network, such as a standard ResNet (Chen et al., 2018) and a Gaussian process with a deep neural network kernel Bradshaw et al. (2017). ",
+        "bbox": [
+            174,
+            445,
+            825,
+            529
+        ],
+        "page_idx": 7
+    },
+    {
+        "type": "text",
+        "text": "To check whether the GP limit is applicable to relatively small networks used practically (with of the order of 100 channels in the first layers), we randomly sampled 10, 000 32-layer ResNets, with 3, 10, 30 and 100 channels in the first layers, and, following the usual practice for ResNets we increase the number the number of hidden units when we downsample the feature maps. The probability density plots show a good match around 100 channels (Fig. 2A), which matches a more sensitive graphical procedure based on quantile-quantile plots (Fig. 2B). Notably, even for only 30 channels, the moments match closely (Fig. 2C). For comparison, typical ResNets use from 64 (He et al., 2016a) to 192 (Zagoruyko & Komodakis, 2016) channels in their first layers. We believe that this is because the moment propagation equations only require the Gaussianity assumption for propagation through the relu, and presumably this is robust to non-Gaussian input activations. ",
+        "bbox": [
+            174,
+            536,
+            825,
+            675
+        ],
+        "page_idx": 7
+    },
+    {
+        "type": "text",
+        "text": "Computational efficiency. Asymptotically, computing the kernel matrix takes $O ( N ^ { 2 } L D )$ time, where $L$ is the number of layers in the network and $D$ is the dimensionality of the input, and inverting the kernel matrix takes $O ( \\bar { N } ^ { 3 } )$ . As such, we expect that for very large datasets, inverting the kernel matrix will dominate the computation time. However, on MNIST, $N ^ { 3 }$ is only around a factor of 10 larger than $N ^ { 2 } L D$ . In practice, we found that it was more expensive to compute the kernel matrix than to invert it. For the ResNet kernel, the most expensive, computing ${ \\bf K } _ { x x }$ , and ${ \\bf K } _ { x x * }$ for validation and test took 3h 40min on two Tesla P100 GPUs. In contrast, inverting ${ \\bf K } _ { x x }$ and computing validation and test performance took $4 3 . 2 5 \\pm 8 . 8$ seconds on a single Tesla P100 GPU. ",
+        "bbox": [
+            174,
+            690,
+            825,
+            803
+        ],
+        "page_idx": 7
+    },
+    {
+        "type": "text",
+        "text": "5 RELATED WORK ",
+        "text_level": 1,
+        "bbox": [
+            176,
+            823,
+            341,
+            838
+        ],
+        "page_idx": 7
+    },
+    {
+        "type": "text",
+        "text": "Van der Wilk et al. (van der Wilk et al., 2017) also adapted GPs to image classification. They defined a prior on functions $f$ that takes an image and outputs a scalar. First, draw a function $g \\sim \\mathcal { G P } ( 0 , k _ { p } ( \\mathbf { X } , \\mathbf { X } ^ { \\prime } ) )$ . Then, $f$ is the sum of the output of $g$ applied to each of the convolutional patches. Their approach is also inspired by convolutional NNs, but their kernel $k _ { p }$ is applied to all pairs of patches of $\\mathbf { X }$ and $\\mathbf { X } ^ { \\prime }$ . This makes their convolutional kernel expensive to evaluate, requiring inter-domain inducing point approximations to remain tractable. The kernels in this work, directly motivated by the infinite-filter limit of a CNN, only apply something like $k _ { p }$ to the corresponding pairs of patches within $\\mathbf { X }$ and $\\mathbf { X } ^ { \\prime }$ (Eq. 10). As such, the CNN kernels are cheaper to compute and exhibit superior performance (Table 1), despite the use of an approximate likelihood function. ",
+        "bbox": [
+            174,
+            853,
+            823,
+            924
+        ],
+        "page_idx": 7
+    },
+    {
+        "type": "image",
+        "img_path": "images/855e86413bd7de48d01d9cca88bd8f2c7466cd7a38d25eebae9541aa8c975eaf.jpg",
+        "image_caption": [
+            "Figure 2: Comparison of the infinite limit, and outputs from finite 32-layer ResNets with 3, 10, 30, and 100 channels in their first layers. A Comparison of the empirical and limiting probability densities. B A more sensitive test of Gaussianity is a quantile-quantile plot, which shows converges with 100 channels. C The moments (variances and covariances) for 100 training inputs shows gives a good match for all numbers of channels. "
+        ],
+        "image_footnote": [],
+        "bbox": [
+            179,
+            99,
+            826,
+            440
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "",
+        "bbox": [
+            174,
+            574,
+            825,
+            631
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "Kumar et al. (2018) define a prior over functions by stacking several GPs with van der Wilk’s convolutional kernel, forming a “Deep GP” (Damianou & Lawrence, 2013). In contrast, the kernel in this paper confines all hierarchy to the definition of the kernel, and the resulting GPs is shallow. ",
+        "bbox": [
+            174,
+            637,
+            825,
+            680
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "Wilson et al. (2016) introduced and Bradshaw et al. (2017) improved deep kernel learning. The inputs to a classic GP kernel $k$ (e.g. RBF) are preprocessed by applying a feature extractor $g$ (a deep NN) prior to computing the kernel: $k _ { \\mathrm { d e e p } } ( \\mathbf { X } , \\bar { \\mathbf { X } } ^ { \\prime } ) : = k ( g ( \\mathbf { X } ; \\theta ) , \\bar { g ( \\mathbf { X } ^ { \\prime } , \\theta ) } )$ . The NN parameters are optimised by gradient ascent using the likelihood as the objective, as in standard GP kernel learning (Rasmussen & Williams, 2006, Chapter 5). Since deep kernel learning incorporates a state-of-the-art NN with over $1 0 ^ { 6 }$ parameters, we expect it to perform similarly to a NN applied directly to the task of image classification. At present both CNNs and deep kernel learning display superior performance to the GP kernels in this work. However, the kernels defined here have far fewer parameters (around 10, compared to their $1 0 ^ { 6 }$ ). ",
+        "bbox": [
+            174,
+            686,
+            825,
+            811
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "Borovykh (2018) also suggests that a CNN exhibits GP behaviour. However, they take the infinite limit with respect to the filter size, not the number of filters. Thus, their infinite network is inapplicable to real data which is always of finite dimension. ",
+        "bbox": [
+            174,
+            819,
+            825,
+            861
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "Finally, there is a series of papers analysing the mean-field behaviour of deep NNs and CNNs which aims to find good random initializations, i.e. those that do not exhibit vanishing or exploding gradients or activations (Schoenholz et al., 2016; Yang & Schoenholz, 2017). Apart from their very different focus, the key difference to our work is that they compute the variance for a single trainingexample, whereas to obtain the GPs kernel, we additionally need to compute the output covariances for different training/test examples (Xiao et al., 2018). ",
+        "bbox": [
+            174,
+            867,
+            823,
+            924
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "",
+        "bbox": [
+            173,
+            103,
+            823,
+            132
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "6 CONCLUSIONS AND FUTURE WORK ",
+        "text_level": 1,
+        "bbox": [
+            176,
+            155,
+            495,
+            170
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "We have shown that deep Bayesian CNNs with infinitely many filters are equivalent to a GP with a recursive kernel. We also derived the kernel for the GP equivalent to a CNN, and showed that, in handwritten digit classification, it outperforms all previous GP approaches that do not incorporate a parametric NN into the kernel. Given that most state-of-the-art neural networks incorporate structure (convolutional or otherwise) into their architecture, the equivalence between CNNs and GPs is potentially of considerable practical relevance. In particular, we hope to apply GP CNNs in domains as widespread as adversarial examples, lifelong learning and $\\mathbf { k }$ -shot learning, and we hope to improve them by developing efficient multi-layered inducing point approximation schemes. ",
+        "bbox": [
+            174,
+            186,
+            825,
+            299
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "REFERENCES ",
+        "text_level": 1,
+        "bbox": [
+            176,
+            321,
+            285,
+            335
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "JR Blum, H Chernoff, M Rosenblatt, and H Teicher. Central limit theorems for interchangeable processes. Canad. J. Math, 10:222–229, 1958. ",
+        "bbox": [
+            176,
+            345,
+            823,
+            375
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "Charles Blundell, Julien Cornebise, Koray Kavukcuoglu, and Daan Wierstra. Weight uncertainty in neural network. In International Conference on Machine Learning, pp. 1613–1622, 2015. ",
+        "bbox": [
+            173,
+            386,
+            823,
+            416
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "Anastasia Borovykh. A gaussian process perspective on convolutional neural networks. ResearchGate, 05 2018. ",
+        "bbox": [
+            173,
+            428,
+            821,
+            457
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "John Bradshaw, Alexander G de G Matthews, and Zoubin Ghahramani. Adversarial examples, uncertainty, and transfer testing robustness in gaussian process hybrid deep networks. arXiv preprint arXiv:1707.02476, 2017. ",
+        "bbox": [
+            174,
+            468,
+            826,
+            511
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "Tian Qi Chen, Yulia Rubanova, Jesse Bettencourt, and David Duvenaud. Neural ordinary differential equations. arXiv preprint arXiv:1806.07366, 2018. ",
+        "bbox": [
+            173,
+            523,
+            823,
+            553
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "Youngmin Cho and Lawrence K Saul. Kernel methods for deep learning. In Advances in neural information processing systems, pp. 342–350, 2009. URL http://papers.nips.cc/ paper/3628-kernel-methods-for-deep-learning.pdf. ",
+        "bbox": [
+            174,
+            565,
+            823,
+            608
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "Andreas Damianou and Neil Lawrence. Deep gaussian processes. In Carlos M. Carvalho and Pradeep Ravikumar (eds.), Proceedings of the Sixteenth International Conference on Artificial Intelligence and Statistics, volume 31 of Proceedings of Machine Learning Research, pp. 207– 215, Scottsdale, Arizona, USA, 29 Apr–01 May 2013. PMLR. URL http://proceedings. mlr.press/v31/damianou13a.html. ",
+        "bbox": [
+            173,
+            619,
+            825,
+            690
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "Marc Deisenroth and Carl E Rasmussen. PILCO: A model-based and data-efficient approach to policy search. In Proceedings of the 28th International Conference on machine learning (ICML11), pp. 465–472, 2011. ",
+        "bbox": [
+            174,
+            702,
+            823,
+            746
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "Yarin Gal and Zoubin Ghahramani. Dropout as a bayesian approximation: Representing model uncertainty in deep learning. arXiv preprint arXiv:1506.02142, 2015. ",
+        "bbox": [
+            173,
+            757,
+            823,
+            786
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "Yarin Gal and Lewis Smith. Idealised bayesian neural networks cannot have adversarial examples: Theoretical and empirical study. arXiv preprint arXiv:1806.00667, 2018. ",
+        "bbox": [
+            174,
+            799,
+            823,
+            828
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image recognition. In Proceedings of the IEEE conference on computer vision and pattern recognition, pp. 770–778, 2016a. URL https://arxiv.org/pdf/1512.03385.pdf. ",
+        "bbox": [
+            176,
+            840,
+            823,
+            883
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Identity mappings in deep residual networks. In European conference on computer vision, pp. 630–645. Springer, 2016b. ",
+        "bbox": [
+            174,
+            895,
+            823,
+            924
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "Daniel Hernández-lobato, Jose M. Hernández-lobato, and Pierre Dupont. Robust multiclass gaussian process classification. In J. Shawe-Taylor, R. S. Zemel, P. L. Bartlett, F. Pereira, and K. Q. Weinberger (eds.), Advances in Neural Information Processing Systems 24, pp. 280–288. 2011. URL http://papers.nips.cc/paper/ 4241-robust-multi-class-gaussian-process-classification.pdf. ",
+        "bbox": [
+            174,
+            103,
+            825,
+            174
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Alex Krizhevsky, Ilya Sutskever, and Geoffrey E Hinton. Imagenet classification with deep convolutional neural networks. In Advances in neural information processing systems, pp. 1097–1105, 2012. ",
+        "bbox": [
+            174,
+            184,
+            825,
+            227
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Vinayak Kumar, Vaibhav Singh, PK Srijith, and Andreas Damianou. Deep gaussian processes with convolutional kernels. arXiv preprint arXiv:1806.01655, 2018. ",
+        "bbox": [
+            173,
+            237,
+            825,
+            266
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Alexey Kurakin, Ian Goodfellow, and Samy Bengio. Adversarial examples in the physical world. arXiv preprint arXiv:1607.02533, 2016. ",
+        "bbox": [
+            173,
+            276,
+            823,
+            305
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Balaji Lakshminarayanan, Alexander Pritzel, and Charles Blundell. Simple and scalable predictive uncertainty estimation using deep ensembles. In Advances in Neural Information Processing Systems, pp. 6402–6413, 2017. ",
+        "bbox": [
+            173,
+            314,
+            825,
+            358
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Yann LeCun, Bernhard E Boser, John S Denker, Donnie Henderson, Richard E Howard, Wayne E Hubbard, and Lawrence D Jackel. Handwritten digit recognition with a back-propagation network. In Advances in neural information processing systems, pp. 396–404, 1990. ",
+        "bbox": [
+            176,
+            367,
+            821,
+            411
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Jaehoon Lee, Yasaman Bahri, Roman Novak, Samuel S Schoenholz, Jeffrey Pennington, and Jascha Sohl-Dickstein. Deep neural networks as gaussian processes. arXiv preprint arXiv:1711.00165, 2017. ",
+        "bbox": [
+            173,
+            420,
+            825,
+            463
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Stephan Mandt, Matthew D Hoffman, and David M Blei. Stochastic gradient descent as approximate bayesian inference. The Journal of Machine Learning Research, 18(1):4873–4907, 2017. ",
+        "bbox": [
+            173,
+            473,
+            823,
+            503
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Alexander G. de G. Matthews, Jiri Hron, Mark Rowland, Richard E. Turner, and Zoubin Ghahramani. Gaussian process behaviour in wide deep neural networks. In International Conference on Learning Representations, 2018a. URL https://openreview.net/forum?id $=$ H1-nGgWC-. ",
+        "bbox": [
+            174,
+            512,
+            825,
+            569
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Alexander G de G Matthews, Mark Rowland, Jiri Hron, Richard E Turner, and Zoubin Ghahramani. Gaussian process behaviour in wide deep neural networks. arXiv preprint arXiv:1804.11271, 2018b. ",
+        "bbox": [
+            176,
+            579,
+            823,
+            622
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "De G Matthews, G Alexander, Mark Van Der Wilk, Tom Nickson, Keisuke Fujii, Alexis Boukouvalas, Pablo León-Villagrá, Zoubin Ghahramani, and James Hensman. GPflow: A gaussian process library using TensorFlow. The Journal of Machine Learning Research, 18(1):1299–1304, 2017. ",
+        "bbox": [
+            174,
+            632,
+            825,
+            689
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Radford M. Neal. Bayesian Learning for Neural Networks. Springer-Verlag, Berlin, Heidelberg, 1996. ISBN 0387947248. ",
+        "bbox": [
+            173,
+            699,
+            823,
+            728
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Carl Edward Rasmussen and Christopher KI Williams. Gaussian processes for machine learning, volume 1. MIT press Cambridge, 2006. ",
+        "bbox": [
+            173,
+            738,
+            825,
+            767
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Samuel S Schoenholz, Justin Gilmer, Surya Ganguli, and Jascha Sohl-Dickstein. Deep information propagation. arXiv preprint arXiv:1611.01232, 2016. ",
+        "bbox": [
+            171,
+            777,
+            825,
+            806
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Jasper Snoek, Hugo Larochelle, and Ryan P Adams. Practical bayesian optimization of machine learning algorithms. In Advances in neural information processing systems, pp. 2951–2959, 2012. ",
+        "bbox": [
+            171,
+            816,
+            823,
+            845
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Christian Szegedy, Wojciech Zaremba, Ilya Sutskever, Joan Bruna, Dumitru Erhan, Ian Goodfellow, and Rob Fergus. Intriguing properties of neural networks. arXiv preprint arXiv:1312.6199, 2013. ",
+        "bbox": [
+            171,
+            856,
+            823,
+            885
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Mark van der Wilk, Carl Edward Rasmussen, and James Hensman. Convolutional gaussian processes. In Advances in Neural Information Processing Systems, pp. 2845–2854, 2017. ",
+        "bbox": [
+            173,
+            895,
+            821,
+            924
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Max Welling and Yee W Teh. Bayesian learning via stochastic gradient langevin dynamics. In Proceedings of the 28th International Conference on Machine Learning (ICML-11), pp. 681–688, 2011. ",
+        "bbox": [
+            174,
+            103,
+            823,
+            146
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "Andrew Gordon Wilson, Zhiting Hu, Ruslan Salakhutdinov, and Eric P. Xing. Deep kernel learning. In Arthur Gretton and Christian C. Robert (eds.), Proceedings of the 19th International Conference on Artificial Intelligence and Statistics, volume 51 of Proceedings of Machine Learning Research, pp. 370–378, Cadiz, Spain, 09–11 May 2016. PMLR. URL http: //proceedings.mlr.press/v51/wilson16.html. ",
+        "bbox": [
+            174,
+            155,
+            825,
+            226
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "Lechao Xiao, Yasaman Bahri, Jascha Sohl-Dickstein, Samuel S Schoenholz, and Jeffrey Pennington. Dynamical isometry and a mean field theory of cnns: How to train 10,000-layer vanilla convolutional neural networks. arXiv preprint arXiv:1806.05393, 2018. ",
+        "bbox": [
+            174,
+            234,
+            825,
+            277
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "Ge Yang and Samuel Schoenholz. Mean field residual networks: On the edge of chaos. In Advances in neural information processing systems, pp. 7103–7114, 2017. ",
+        "bbox": [
+            174,
+            286,
+            821,
+            316
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "Sergey Zagoruyko and Nikos Komodakis. Wide residual networks. arXiv preprint arXiv:1605.07146, 2016. ",
+        "bbox": [
+            174,
+            325,
+            825,
+            353
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "7 APPENDIX ",
+        "text_level": 1,
+        "bbox": [
+            174,
+            381,
+            294,
+            397
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "7.1 TECHNICAL NOTES ON LIMITS ",
+        "text_level": 1,
+        "bbox": [
+            176,
+            412,
+            423,
+            428
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "The key technical issues in the proof (and the key differences between Lee et al. 2017 and Matthews et al. 2018b) arise from exactly how and where we take limits. In particular, consider the activations as being functions of the activities at the previous layer, ",
+        "bbox": [
+            174,
+            438,
+            825,
+            481
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "equation",
+        "img_path": "images/b3df92ef3a1e1a9f6a55bfd23b526811785c6d6ebffe0128f92602b89618fcb5.jpg",
+        "text": "$$\n{ \\bf A } ^ { ( 4 ) } = { \\bf A } ^ { ( 4 ) } ( { \\bf A } ^ { ( 3 ) } ( { \\bf A } ^ { ( 2 ) } ( { \\bf A } ^ { ( 1 ) } ( { \\bf X } ) ) ) )\n$$",
+        "text_format": "latex",
+        "bbox": [
+            374,
+            487,
+            622,
+            507
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "Now, there are two approaches to taking limits. First, both our argument in the main text, and the argument in Lee et al. (2017) is valid if we are able to take limits “inside” the network, ",
+        "bbox": [
+            173,
+            513,
+            823,
+            542
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "equation",
+        "img_path": "images/ecda43d5472ce6796fa65efa1ae828841615270ab00a72e6b4c5b375739ab9de.jpg",
+        "text": "$$\n\\mathbf { A } _ { \\mathrm { L } } ^ { ( 4 ) } = \\operatorname* { l i m } _ { C ^ { ( 3 ) } \\to \\infty } \\mathbf { A } ^ { ( 4 ) } \\left( \\operatorname* { l i m } _ { C ^ { ( 2 ) } \\to \\infty } \\mathbf { A } ^ { ( 3 ) } \\left( \\operatorname* { l i m } _ { C ^ { ( 1 ) } \\to \\infty } \\mathbf { A } ^ { ( 2 ) } \\left( \\mathbf { A } ^ { ( 1 ) } ( \\mathbf { X } ) \\right) \\right) \\right) .\n$$",
+        "text_format": "latex",
+        "bbox": [
+            269,
+            547,
+            725,
+            584
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "However, Matthews et al. (2018a;b) argue that is preferable to take limits “outside” the network. In particular, Matthews et al. (2018b) take the limit with all layers simultaneously, ",
+        "bbox": [
+            173,
+            590,
+            821,
+            618
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "equation",
+        "img_path": "images/60aa0f685aef6c4f814cd4e1691b1779363228de038f37492e5c1d5ec39199bd.jpg",
+        "text": "$$\n{ \\bf A } _ { \\mathrm { M } } ^ { ( 4 ) } = \\operatorname* { l i m } _ { n  \\infty } { \\bf A } ^ { ( 4 ) } ( { \\bf A } ^ { ( 3 ) } ( { \\bf A } ^ { ( 2 ) } ( { \\bf A } ^ { ( 1 ) } ( { \\bf X } ) ) ) ) ,\n$$",
+        "text_format": "latex",
+        "bbox": [
+            336,
+            625,
+            658,
+            652
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "where $C ^ { ( \\ell ) } = C ^ { ( \\ell ) } ( n )$ goes to infinity as $n \\to \\infty$ . That said, similar technical issues arise if we take limits in sequence, but outside the network. ",
+        "bbox": [
+            174,
+            660,
+            825,
+            690
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "7.2 EXTENDING THE DERIVATIONS OF MATTHEWS ET AL. (2018B) TO THE CONVOLUTIONAL CASE ",
+        "text_level": 1,
+        "bbox": [
+            174,
+            707,
+            704,
+            736
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "In the main text, we follow Lee et al. (2017) in sequentially taking the limit of each layer to infinity (i.e. $C ^ { ( 1 ) } \\to \\infty$ , then $C ^ { ( 2 ) }  \\infty$ etc.). This dramatically simplified the argument, because taking the number of units in the previous layer to infinity means that the inputs from that layer are exactly Gaussian distributed. However, Matthews et al. (2018b) argue that the more practically relevant limit is where we take all layers to infinity simultaneously. This raises considerable additional difficulties, because we must reason about convergence in the case where the previous layer is finite. Note that this section is not intended to stand independently: it is intended to be read alongside Matthews et al. (2018b), and we use several of their results without proof. ",
+        "bbox": [
+            173,
+            746,
+            825,
+            859
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "Mirroring Definition 3 in Matthews et al. (2018b), we begin by choosing a set of “width” functions, $C ^ { ( \\ell ) } ( n )$ , for $\\ell \\in \\{ 1 , \\ldots , L \\}$ which all approach infinity as $n  \\infty$ . In Matthews et al. (2018b), these functions described the number of hidden units in each layer, whereas here they describe the number of channels. Our goal is then to extend the proofs in Matthews et al. (2018b) (in particular, of theorem 4), to show that the output of our convolutional networks converge in distribution to a Gaussian process as $n  \\infty$ , with mean zero and covariance given by the recursion in Eqs. (10 – 12). ",
+        "bbox": [
+            174,
+            866,
+            825,
+            924
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "",
+        "bbox": [
+            174,
+            103,
+            825,
+            146
+        ],
+        "page_idx": 12
+    },
+    {
+        "type": "text",
+        "text": "The proof in Matthews et al. (2018b) has three main steps. First, they use the Cramér-Wold device, to reduce the full problem to that of proving convergence of scalar random variables to a Gaussian with specified variance. Second, if the previous layers have finite numbers of channels, then the channels ${ \\bf a } _ { j } ^ { ( \\ell ) } ( { \\bf X } )$ and $\\mathbf { a } _ { j } ^ { ( \\ell ) } ( \\mathbf { X } ^ { \\prime } )$ are uncorrelated but no longer independent, so we cannot apply the CLT directly, as we did in the main text. Instead, they write the activations as a sum of exchangeable random variables, and derive an adapted CLT for exchangeable (rather than independent) random variables (Blum et al., 1958). Third, they show that moment conditions required by their exchangeable CLT are satisfied. ",
+        "bbox": [
+            173,
+            151,
+            825,
+            267
+        ],
+        "page_idx": 12
+    },
+    {
+        "type": "text",
+        "text": "To extend their proofs to the convolutional case, we begin by defining our networks in a form that is easier to manipulate and as close as possible to Eq. (21-23) in Matthews et al. (2018b), ",
+        "bbox": [
+            173,
+            273,
+            823,
+            303
+        ],
+        "page_idx": 12
+    },
+    {
+        "type": "equation",
+        "img_path": "images/5ac0c6fe0b60b6f942651d07a0b16b9896617fb2361fb2788fbba4e2dc4d936a.jpg",
+        "text": "$$\n\\begin{array} { c } { { A _ { i , \\mu } ^ { ( 1 ) } = f _ { i , \\mu } ^ { ( 1 ) } ( x ) = \\displaystyle \\frac { \\sigma _ { \\mathbf { w } } } { \\sqrt { C ^ { ( 0 ) } } } \\sum _ { j = 1 } ^ { C ^ { ( 0 ) } } \\sum _ { \\nu \\in \\mu \\mathbb { h } \\mathrm { p a t e h } } \\epsilon _ { i , j , \\mu , \\nu } ^ { ( 1 ) } x _ { j , \\nu } + b _ { i } ^ { ( 1 ) } , \\quad i \\in \\mathbb { N } } } \\\\ { { g _ { i , \\mu } ^ { ( \\ell ) } ( x ) = \\phi \\left( f _ { i , \\mu } ^ { ( \\ell ) } ( x ) \\right) } } \\\\ { { A _ { i , \\mu } ^ { ( \\ell + 1 ) } = f _ { i , \\mu } ^ { ( \\ell + 1 ) } ( x ) = \\displaystyle \\frac { \\sigma _ { \\mathbf { w } } } { \\sqrt { C ^ { ( \\ell ) } ( n ) } } \\sum _ { j = 1 } ^ { C ^ { ( \\ell ) } } \\sum _ { \\nu \\in \\mu \\mathrm { t h } \\mathrm { p a t e h } } \\epsilon _ { i , j , \\mu , \\nu } ^ { ( \\ell + 1 ) } g _ { j , \\nu } ^ { ( \\ell ) } ( x ) + b _ { i } ^ { ( \\ell + 1 ) } , \\quad i \\in \\mathbb { N } } } \\end{array}\n$$",
+        "text_format": "latex",
+        "bbox": [
+            207,
+            309,
+            764,
+            435
+        ],
+        "page_idx": 12
+    },
+    {
+        "type": "text",
+        "text": "where, ",
+        "bbox": [
+            174,
+            445,
+            220,
+            458
+        ],
+        "page_idx": 12
+    },
+    {
+        "type": "equation",
+        "img_path": "images/377587ad330efcc50a95e3c6182081ba16330c814e44797308546f20e4d79e77.jpg",
+        "text": "$$\n\\epsilon _ { i , j , \\mu , \\nu } \\sim \\mathcal { N } ( 0 , 1 ) .\n$$",
+        "text_format": "latex",
+        "bbox": [
+            297,
+            469,
+            423,
+            487
+        ],
+        "page_idx": 12
+    },
+    {
+        "type": "text",
+        "text": "The first step is to use the Cramér-Wold device (Lemma 6 in Matthews et al., 2018b), which indicates that convergence in distribution of a sequence of finite-dimensional vectors is equivalent to convergence on all possible linear projections to the corresponding real-valued random variable. Mirroring Eq. 25 in Matthews et al. (2018b), we consider convergence of random vectors, $f _ { i , \\mu } ^ { ( \\ell ) } ( x ) [ n ] - b _ { i } ^ { ( \\ell ) }$ , projected onto α(x,i,µ), ",
+        "bbox": [
+            173,
+            501,
+            825,
+            578
+        ],
+        "page_idx": 12
+    },
+    {
+        "type": "equation",
+        "img_path": "images/cd4997d6c67a70841410e50fd66d16ca4c261c7e83d0dac73d11e7138fa29fb7.jpg",
+        "text": "$$\n\\mathcal { T } ^ { ( \\ell ) } \\left( \\mathcal { L } , \\alpha \\right) \\left[ n \\right] = \\sum _ { ( x , i , \\mu ) \\in \\mathcal { L } } \\alpha ^ { ( x , i , \\mu ) } \\left[ f _ { i , \\mu } ^ { ( \\ell ) } ( x ) [ n ] - b _ { i } ^ { ( \\ell ) } \\right] .\n$$",
+        "text_format": "latex",
+        "bbox": [
+            315,
+            584,
+            681,
+            622
+        ],
+        "page_idx": 12
+    },
+    {
+        "type": "text",
+        "text": "where $\\mathcal { L } \\subset \\mathcal { X } \\times \\mathbb { N } \\times \\{ 1 , \\ldots , H ^ { ( \\ell ) } D ^ { ( \\ell ) } \\}$ is a finite set of tuples of data points and channel indicies, $i$ , and indicies of elements within channels/feature maps, $\\mu$ . The suffix $[ n ]$ indicates width functions that are instantiated with input, $n$ . ",
+        "bbox": [
+            174,
+            630,
+            826,
+            674
+        ],
+        "page_idx": 12
+    },
+    {
+        "type": "text",
+        "text": "Now, we must prove that these projections converge in distribution a Gaussian. We begin by defining summands, as in Eq. 26 in Matthews et al. (2018b), ",
+        "bbox": [
+            171,
+            679,
+            823,
+            709
+        ],
+        "page_idx": 12
+    },
+    {
+        "type": "equation",
+        "img_path": "images/3ed295d2d79e28d39103a42271becb9cebc58bde3a117ebe29ab944b80dffd4b.jpg",
+        "text": "$$\n\\gamma _ { j } ^ { ( \\ell ) } \\left( \\mathcal { L } , \\alpha \\right) [ n ] : = \\sigma _ { \\mathrm { w } } \\sum _ { ( x , i , \\mu ) \\in \\mathcal { L } } \\alpha ^ { ( x , i , \\mu ) } \\sum _ { \\nu \\in \\mu \\mathrm { t h p a t c h } } \\epsilon _ { i , j , \\mu , \\nu } ^ { ( \\ell ) } g _ { j , \\nu } ^ { ( \\ell - 1 ) } ( x ) [ n ] ,\n$$",
+        "text_format": "latex",
+        "bbox": [
+            274,
+            715,
+            722,
+            752
+        ],
+        "page_idx": 12
+    },
+    {
+        "type": "text",
+        "text": "such that the projections can be written as a sum of the summands, exactly as in Eq. 27 in Matthews et al. (2018b), ",
+        "bbox": [
+            173,
+            757,
+            823,
+            787
+        ],
+        "page_idx": 12
+    },
+    {
+        "type": "equation",
+        "img_path": "images/699567a0d3051b276ae2add7f5ce075f543504cbdfeacc1ea663050f2b30baca.jpg",
+        "text": "$$\n\\mathcal { T } ^ { ( \\ell ) } \\left( \\mathcal { L } , \\alpha \\right) [ n ] = \\frac { 1 } { \\sqrt { C ^ { ( \\ell - 1 ) } ( n ) } } \\sum _ { j = 1 } ^ { C ^ { ( \\ell - 1 ) } ( n ) } \\gamma _ { j } ^ { ( \\ell ) } \\left( \\mathcal { L } , \\alpha \\right) [ n ] .\n$$",
+        "text_format": "latex",
+        "bbox": [
+            313,
+            792,
+            683,
+            842
+        ],
+        "page_idx": 12
+    },
+    {
+        "type": "text",
+        "text": "Now we can apply the exchangeable CLT to prove that ${ \\mathcal { T } } ^ { ( \\ell ) } \\left( { \\mathcal { L } } , \\alpha \\right) [ n ]$ converges to the limiting Gaussian implied by the recursions in the main text. To apply the exchangeable CLT, the first step is to mirror Lemma 8 in Matthews et al. (2018b), in showing that for each fixed $n$ and $\\ell \\in \\{ 2 , \\ldots , L +$ $1 \\}$ , the summands, $\\gamma _ { j } ^ { ( \\ell ) } \\left( \\mathcal { L } , \\alpha \\right) [ n ]$ are exchangeable with respect to the index $j$ . In particular, we apply de Finetti’s theorem, which states that a sequence of random variables is exchangeable if and only if they are i.i.d. conditional on some set of random variables, so it is sufficient to exhibit such a set of random variables. Mirroring Eq. 29 in Matthews et al. (2018b), we apply the recursion, ",
+        "bbox": [
+            173,
+            848,
+            825,
+            924
+        ],
+        "page_idx": 12
+    },
+    {
+        "type": "text",
+        "text": "",
+        "bbox": [
+            169,
+            102,
+            825,
+            133
+        ],
+        "page_idx": 13
+    },
+    {
+        "type": "equation",
+        "img_path": "images/17c0e7e933133d259cd80eda41b96e3b779964d857d42527541a14a92629a63e.jpg",
+        "text": "$$\n \\overset { ( \\ell ) } { j } ( \\mathcal { L } , \\alpha ) [ n ] : = \\sigma _ { \\mathrm { w } } \\sum _ { ( x , i , \\mu ) \\in \\mathcal { L } } \\alpha ^ { ( x , i , \\mu ) } \\sum _ { \\nu \\in \\mu \\mathrm { t h } \\mathrm { p a t e h } } \\epsilon _ { i , j , \\mu , \\nu } ^ { ( \\ell ) } \\phi ( \\frac { \\sigma _ { \\mathrm { w } } } { \\sqrt { C ^ { ( \\ell - 2 ) } ( n ) } } \\sum _ { k = 1 } ^ { C ^ { ( \\ell - 2 ) } ( n ) } \\sum _ { \\xi \\in \\nu \\mathrm { t h } \\mathrm { p a t e h } } \\epsilon _ { j , k , \\nu ; \\xi } ^ { ( \\ell - 1 ) } g _ { k , \\xi } ^ { ( \\ell - 2 ) } ) ,\n$$",
+        "text_format": "latex",
+        "bbox": [
+            181,
+            133,
+            839,
+            184
+        ],
+        "page_idx": 13
+    },
+    {
+        "type": "text",
+        "text": "As such, the summands are iid conditional on the finite set of random variables $\\left\\{ g _ { k , \\xi } ^ { ( \\ell - 2 ) } ( x ) [ n ] : k \\in \\{ 1 , \\ldots , C ^ { ( \\ell - 2 ) } \\} , \\xi \\in \\left\\{ 1 , \\ldots , H ^ { ( \\ell - 2 ) } D ^ { ( \\ell - 2 ) } \\right\\} , x \\in \\mathcal { L } _ { \\mathcal { X } } \\right\\}$ , where $\\mathcal { L } _ { \\mathcal { X } }$ is the set of input points in $\\mathcal { L }$ . ",
+        "bbox": [
+            173,
+            199,
+            826,
+            251
+        ],
+        "page_idx": 13
+    },
+    {
+        "type": "text",
+        "text": "The exchangeable CLT in Lemma 10 in Matthews et al. (2018b) indicates that ${ \\mathcal { T } } ^ { ( \\ell ) } \\left( { \\mathcal { L } } , \\alpha \\right) [ n ]$ converges in distribution to $\\mathcal { N } \\left( 0 , \\sigma _ { * } ^ { 2 } \\right)$ if the summands are exchangeable (which we showed above), and if three conditions hold, ",
+        "bbox": [
+            173,
+            257,
+            825,
+            303
+        ],
+        "page_idx": 13
+    },
+    {
+        "type": "equation",
+        "img_path": "images/d2bd2b8a709e0411be37c1c704aac55b083a7ab7c52b160eaccbcf65054d7360.jpg",
+        "text": "$$\n\\begin{array} { r l } & { \\mathrm { a ) ~ \\mathbb { E } } _ { n } [ \\gamma _ { j } ^ { ( \\ell ) } \\gamma _ { j ^ { \\prime } } ^ { ( \\ell ) } ] = 0 } \\\\ & { \\mathrm { b ) ~ \\operatorname* { l i m } _ { n  \\infty } \\mathbb { E } } _ { n } [ ( \\gamma _ { j } ^ { ( \\ell ) } ) ^ { 2 } ( \\gamma _ { j ^ { \\prime } } ^ { ( \\ell ) } ) ^ { 2 } ] = \\sigma _ { * } ^ { 4 } } \\\\ & { \\mathrm { c ) ~ \\mathbb { E } } _ { n } [ | \\gamma _ { j } ^ { ( \\ell ) } | ^ { 3 } ] = o ( \\sqrt { C ^ { ( \\ell ) } ( n ) } ) } \\end{array}\n$$",
+        "text_format": "latex",
+        "bbox": [
+            207,
+            308,
+            482,
+            402
+        ],
+        "page_idx": 13
+    },
+    {
+        "type": "text",
+        "text": "Condition a) follows immediately as the summands are uncorrelated and zero-mean. Conditions b) and c) are more involved as convergence in distribution in the previous layers does not imply convergence in moments for our activation functions. ",
+        "bbox": [
+            174,
+            406,
+            825,
+            450
+        ],
+        "page_idx": 13
+    },
+    {
+        "type": "text",
+        "text": "We begin by considering the extension of Lemma 20 in Matthews et al. (2018b), which allow us to show conditions b) and c) above, even in the case of unbounded but linearly enveloped nonlinearities (Definition 1 in Matthews et al., 2018b). Lembounded by a finite constant independent of states that the eighth moments of . We prove this by induction. Th $f _ { i , \\mu } ^ { ( t ) } ( x ) [ n ]$ aree is $n \\in \\mathbb N$   \ntrivial, as $f _ { j , \\mu } ^ { ( 1 ) } ( x ) [ n ]$ is Gaussian. Following Matthews et al. (2018b), assume the condition holds up to $\\ell - 1$ , and show that the condition holds for layer $\\ell$ . Using Eq. (21), we can bound the activations at layer $\\ell$ , ",
+        "bbox": [
+            173,
+            455,
+            826,
+            564
+        ],
+        "page_idx": 13
+    },
+    {
+        "type": "equation",
+        "img_path": "images/e1525aadc195dba6d3e8ef0ab2a26a0be8b65495eb06427ba51a80150e64f40d.jpg",
+        "text": "$$\n\\mathbb { E } \\left[ | f _ { i , \\mu } ^ { ( \\ell ) } ( x ) [ n ] | ^ { 8 } \\right] \\leq 2 ^ { 8 - 1 } \\mathbb { E } \\left[ | b _ { i } ^ { ( \\ell ) } | ^ { 8 } + \\left| \\frac { \\sigma _ { \\mathbf { w } } } { \\sqrt { C ^ { ( \\ell - 1 ) } } } \\sum _ { j = 1 } ^ { C ^ { ( \\ell - 1 ) } ( n ) } \\sum _ { \\nu \\in \\mu \\mathrm { t h } \\mathrm { p a t e h } } \\epsilon _ { i , j , \\mu , \\nu } ^ { ( \\ell ) } g _ { j , \\nu } ^ { ( \\ell - 1 ) } ( x ) [ n ] \\right| ^ { 8 } \\right]\n$$",
+        "text_format": "latex",
+        "bbox": [
+            183,
+            564,
+            784,
+            622
+        ],
+        "page_idx": 13
+    },
+    {
+        "type": "text",
+        "text": "Following Eq. 48 in Matthews et al. (2018b), which uses Lemma 19 in Matthews et al. (2018b), we have, ",
+        "bbox": [
+            173,
+            622,
+            830,
+            651
+        ],
+        "page_idx": 13
+    },
+    {
+        "type": "equation",
+        "img_path": "images/d5c365ee7ab68b781b018199bf3460179ad3b31af3693fa570d752d714ff7103.jpg",
+        "text": "$$\n\\begin{array} { r l } { \\mathbb { E } \\left[ \\left| \\frac { \\sigma _ { \\mathrm { w } } } { \\sqrt { C ^ { ( \\ell - 1 ) } } } \\sum _ { j = 1 } ^ { C ^ { ( \\ell - 1 ) } ( n ) } \\displaystyle \\sum _ { \\nu \\in \\mu \\mathrm { t h } \\mathrm { p a t e h } } \\epsilon _ { i , j , \\mu , \\nu } ^ { ( \\ell ) } g _ { j , \\nu } ^ { ( \\ell - 1 ) } ( x ) [ n ] \\right| ^ { 8 } \\right] } & { } \\\\ { = \\frac { 2 ^ { 4 } \\Gamma ( 4 + 1 / 2 ) } { \\Gamma ( 1 / 2 ) } \\mathbb { E } \\left[ \\left| \\frac { \\sigma _ { \\mathrm { w } } ^ { 2 } } { C ^ { ( \\ell - 1 ) } ( n ) } \\| g _ { j \\in \\{ 1 , \\dots , C ^ { ( \\ell - 1 ) } ( n ) \\} , \\nu \\in \\mu \\mathrm { t h } \\mathrm { p a t c h } } ^ { ( \\ell - 1 ) } ( x ) [ n ] \\| _ { 2 } ^ { 2 } \\right| ^ { 4 } \\right] . } \\end{array}\n$$",
+        "text_format": "latex",
+        "bbox": [
+            191,
+            655,
+            781,
+            756
+        ],
+        "page_idx": 13
+    },
+    {
+        "type": "text",
+        "text": "wher e g(\\`−1)j∈{1,...,C(\\`−1)(n)},ν∈µth patch(x)[n] is the set of post-nonlinearities corresponding to j ∈ $\\{ 1 , \\ldots , C ^ { ( \\ell - 1 ) } ( n ) \\}$ and $\\nu \\in \\mu \\mathrm { t h }$ patch. Following Matthews et al. (2018b), observe that, ",
+        "bbox": [
+            173,
+            760,
+            828,
+            796
+        ],
+        "page_idx": 13
+    },
+    {
+        "type": "equation",
+        "img_path": "images/acefb7cbee5fe7c46f897dadaf5a1e645b463a9143b940c5edeeb1ace37f3784.jpg",
+        "text": "$$\n\\begin{array} { r } { \\frac { 1 } { \\sum ^ { ( \\ell - 1 ) } \\left( n \\right) } \\| g _ { j \\in \\{ 1 , \\dots , C ^ { ( \\ell - 1 ) } ( n ) \\} , \\nu \\in \\mu \\mathrm { t h p a t e h } } ^ { ( \\ell - 1 ) } ( x ) [ n ] \\| _ { 2 } ^ { 2 } = \\frac { 1 } { C ^ { ( \\ell - 1 ) } ( n ) } \\displaystyle \\sum _ { j = 1 } ^ { C ^ { ( \\ell - 1 ) } ( n ) } \\sum _ { \\nu \\in \\mu \\mathrm { t h p a t e h } } \\left( g _ { j , \\nu } ^ { ( \\ell - 1 ) } ( x ) [ n ] \\right) ^ { 2 } } \\\\ { \\leq \\frac { 1 } { C ^ { ( \\ell - 1 ) } ( n ) } \\displaystyle \\sum _ { j = 1 } ^ { C ^ { ( \\ell - 1 ) } ( n ) } \\sum _ { \\nu \\in \\mu \\mathrm { t h p a t e h } } ^ { C ^ { ( \\ell - 1 ) } ( n ) } \\left( c + m \\big | f _ { j , \\nu } ^ { ( \\ell - 1 ) } ( x ) \\big | \\right. } \\end{array}\n$$",
+        "text_format": "latex",
+        "bbox": [
+            181,
+            797,
+            838,
+            912
+        ],
+        "page_idx": 13
+    },
+    {
+        "type": "text",
+        "text": "by the linear envelope property, $| \\phi ( u ) | \\leq c + m | u |$ . Following Matthews et al. (2018b), we substitute this bound back into Eq. (28) and suppress a multiplicative constant independent of $x$ and $n$ , ",
+        "bbox": [
+            169,
+            102,
+            825,
+            133
+        ],
+        "page_idx": 14
+    },
+    {
+        "type": "equation",
+        "img_path": "images/763d53037e2ed6fdbd5080681e93bf38d156c39f3e7ef9b04ba81650c480c4d2.jpg",
+        "text": "$$\n\\begin{array} { r l } & { \\mathbb { E } \\left[ \\left| \\frac { \\sigma _ { \\mathbf { w } } } { \\sqrt { C ^ { ( \\ell - 1 ) } ( n ) } } \\sum _ { j = 1 } ^ { C ^ { ( \\ell - 1 ) } ( n ) } \\sum _ { \\nu \\in \\mu \\mathrm { t h p a t c h } } \\epsilon _ { i , j , \\mu , \\nu } ^ { ( \\ell ) } g _ { j , \\nu } ^ { ( \\ell - 1 ) } ( x ) [ n ] \\right| ^ { 8 } \\right] } \\\\ & { \\leq \\frac { 1 } { \\left( C ^ { ( \\ell - 1 ) } ( n ) \\right) ^ { 4 } } \\mathbb { E } \\left[ \\left| \\sum _ { j = 1 } ^ { C ^ { ( \\ell - 1 ) } ( n ) } \\sum _ { \\nu \\in \\mu \\mathrm { t h p a c h } } c ^ { 2 } + 2 c m | f _ { j , \\mu } ^ { ( \\ell - 1 ) } ( x ) [ n ] | + m ^ { 2 } | f _ { j , \\mu } ^ { ( \\ell - 1 ) } ( x ) [ n ] | ^ { 2 } \\right| ^ { 4 } \\right] } \\end{array}\n$$",
+        "text_format": "latex",
+        "bbox": [
+            187,
+            142,
+            781,
+            258
+        ],
+        "page_idx": 14
+    },
+    {
+        "type": "text",
+        "text": "This can be multiplied out, yielding a weighted sum of expectations of the form, ",
+        "bbox": [
+            174,
+            262,
+            700,
+            279
+        ],
+        "page_idx": 14
+    },
+    {
+        "type": "equation",
+        "img_path": "images/c53f65198268464e7dfdb89ac275223b8f600644dd163b731320170af9d4f07e.jpg",
+        "text": "$$\n\\mathbb { E } \\left[ | f _ { k , \\nu } ^ { ( \\ell - 1 ) } ( x ) [ n ] | ^ { p _ { 1 } } | f _ { l , \\xi } ^ { ( \\ell - 1 ) } ( x ) [ n ] | ^ { p _ { 2 } } | f _ { r , \\pi } ^ { ( \\ell - 1 ) } ( x ) [ n ] | ^ { p _ { 3 } } | f _ { q , \\rho } ^ { ( \\ell - 1 ) } ( x ) [ n ] | ^ { p _ { 4 } } \\right]\n$$",
+        "text_format": "latex",
+        "bbox": [
+            267,
+            284,
+            728,
+            313
+        ],
+        "page_idx": 14
+    },
+    {
+        "type": "text",
+        "text": "with $p _ { i } \\in \\{ 0 , 1 , 2 \\}$ for $i = { 1 , 2 , 3 , 4 }$ , and $k , l , r , q \\in \\{ 1 , \\dots , C ^ { ( \\ell - 1 ) } ( n ) \\}$ , and $\\nu , \\xi , \\pi , \\rho \\in \\mu \\mathrm { t h }$ patch where the weights of these terms are independent of (2018b), each of these terms is bounded if the eighth mo $n$ . Usinents of in Matthews et al.are bounded, which $f _ { k , \\mu } ^ { ( \\ell - 1 ) } ( x ) [ n ]$ $\\left( 2 C ^ { ( \\ell - 1 ) } ( n ) | \\mu \\mathrm { t h } \\mathrm { p a t c h } | \\right) ^ { 4 }$ , where $| \\mu \\mathrm { t h } \\ p a \\mathrm { t c h } |$ is the number of elements in a convolutional patch. Thus, we can use the same constant for any due to the $1 / \\left( C ^ { ( \\ell - 1 ) } ( n ) \\right) ^ { 4 }$ scaling. As in Matthews et al. (2018b), noting that $f _ { j , \\mu } ^ { ( \\ell - 1 ) } ( x ) [ n ]$ are exchangeable over $j$ for any $x$ and $n$ concludes the proof. Using this result, we can obtain a straightforward adaptation of Lemmas 15, 16 and 21 in Matthews et al. (2018b). Lemma 15 gives condition b), Lemma 16 gives condition c); Lemma 15 requires uniform integrability, which is established by Lemma 21. ",
+        "bbox": [
+            173,
+            318,
+            826,
+            486
+        ],
+        "page_idx": 14
+    },
+    {
+        "type": "text",
+        "text": "7.3 CALIBRATION OF GAUSSIAN PROCESS UNCERTAINTY ",
+        "text_level": 1,
+        "bbox": [
+            173,
+            502,
+            581,
+            517
+        ],
+        "page_idx": 14
+    },
+    {
+        "type": "text",
+        "text": "It is important to check that the estimates of uncertainty produced by our Gaussian process are reasonable. However, to make this assessment, we needed to use a proper likelihood, and not the squared-error loss in the main text. We therefore used our kernel to perform the full, multi-class classification problem in GPflow Matthews et al. (2017), with a RobustMax likelihood (Hernándezlobato et al., 2011). The more difficult non-conjugate inference problem forced us to use 1000 inducing points, randomly chosen from the training inputs. Both our kernel and an RBF kernel have similar calibration curves, that closely track the diagonal, indicating accurate uncertainty estimation. However, even in the inducing point setting, our convolutional kernel gave considerably better performance than the RBF kernel $2 . 4 \\%$ error vs $3 . 4 \\%$ error). ",
+        "bbox": [
+            173,
+            529,
+            825,
+            655
+        ],
+        "page_idx": 14
+    },
+    {
+        "type": "image",
+        "img_path": "images/8864ab9d76c33870f50470fe5185392a7878cda9c47764b2ab0ca5045ac895b3.jpg",
+        "image_caption": [
+            "Figure 3: Calibration plots for an RBF kernel (left) and the ResNet kernel (right). The $\\mathbf { X }$ -axis gives GP prediction for the label probability. The points give corresponding proportion of test points with that label, and the bars give the proportion of training examples in each bin. "
+        ],
+        "image_footnote": [],
+        "bbox": [
+            189,
+            369,
+            807,
+            595
+        ],
+        "page_idx": 15
+    }
+]
\ No newline at end of file
diff --git a/parse/train/Bklfsi0cKm/Bklfsi0cKm_middle.json b/parse/train/Bklfsi0cKm/Bklfsi0cKm_middle.json
new file mode 100644
index 0000000000000000000000000000000000000000..32afebf43b325a9976483e6968d43669a6f018e5
--- /dev/null
+++ b/parse/train/Bklfsi0cKm/Bklfsi0cKm_middle.json
@@ -0,0 +1,44154 @@
+{
+    "pdf_info": [
+        {
+            "preproc_blocks": [
+                {
+                    "type": "title",
+                    "bbox": [
+                        106,
+                        79,
+                        397,
+                        116
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                77,
+                                398,
+                                97
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        77,
+                                        398,
+                                        97
+                                    ],
+                                    "score": 1.0,
+                                    "content": "DEEP CONVOLUTIONAL NETWORKS AS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                99,
+                                354,
+                                117
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        99,
+                                        354,
+                                        117
+                                    ],
+                                    "score": 1.0,
+                                    "content": "SHALLOW GAUSSIAN PROCESSES",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        }
+                    ],
+                    "index": 0.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        113,
+                        135,
+                        221,
+                        179
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                114,
+                                134,
+                                212,
+                                147
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        114,
+                                        134,
+                                        212,
+                                        147
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Adrià Garriga-Alonso",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                112,
+                                144,
+                                222,
+                                160
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        112,
+                                        144,
+                                        222,
+                                        160
+                                    ],
+                                    "score": 1.0,
+                                    "content": "department of Engineering",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                112,
+                                156,
+                                214,
+                                170
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        112,
+                                        156,
+                                        214,
+                                        170
+                                    ],
+                                    "score": 1.0,
+                                    "content": "University of Cambridge",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                112,
+                                168,
+                                205,
+                                180
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        112,
+                                        168,
+                                        205,
+                                        180
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ag919@cam.ac.uk",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        }
+                    ],
+                    "index": 3.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        113,
+                        196,
+                        282,
+                        240
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                111,
+                                196,
+                                198,
+                                209
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        111,
+                                        196,
+                                        198,
+                                        209
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Laurence Aitchison",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                111,
+                                205,
+                                225,
+                                221
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        111,
+                                        205,
+                                        225,
+                                        221
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Department of Engineering",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                112,
+                                217,
+                                214,
+                                231
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        112,
+                                        217,
+                                        214,
+                                        231
+                                    ],
+                                    "score": 1.0,
+                                    "content": "University of Cambridge",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                112,
+                                229,
+                                282,
+                                242
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        112,
+                                        229,
+                                        282,
+                                        242
+                                    ],
+                                    "score": 1.0,
+                                    "content": "laurence.aitchison@gmail.com",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        }
+                    ],
+                    "index": 7.5
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        278,
+                        270,
+                        333,
+                        281
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                276,
+                                268,
+                                336,
+                                284
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        276,
+                                        268,
+                                        336,
+                                        284
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ABSTRACT",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        }
+                    ],
+                    "index": 10
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        143,
+                        294,
+                        468,
+                        405
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                142,
+                                294,
+                                469,
+                                308
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        142,
+                                        294,
+                                        469,
+                                        308
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We show that the output of a (residual) convolutional neural network (CNN) with",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                306,
+                                470,
+                                319
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        306,
+                                        470,
+                                        319
+                                    ],
+                                    "score": 1.0,
+                                    "content": "an appropriate prior over the weights and biases is a Gaussian process (GP) in the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                317,
+                                470,
+                                329
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        317,
+                                        470,
+                                        329
+                                    ],
+                                    "score": 1.0,
+                                    "content": "limit of infinitely many convolutional filters, extending similar results for dense",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                328,
+                                470,
+                                341
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        328,
+                                        470,
+                                        341
+                                    ],
+                                    "score": 1.0,
+                                    "content": "networks. For a CNN, the equivalent kernel can be computed exactly and, unlike",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                338,
+                                470,
+                                352
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        338,
+                                        470,
+                                        352
+                                    ],
+                                    "score": 1.0,
+                                    "content": "“deep kernels”, has very few parameters: only the hyperparameters of the original",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                349,
+                                469,
+                                362
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        349,
+                                        469,
+                                        362
+                                    ],
+                                    "score": 1.0,
+                                    "content": "CNN. Further, we show that this kernel has two properties that allow it to be com-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                361,
+                                470,
+                                373
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        361,
+                                        470,
+                                        373
+                                    ],
+                                    "score": 1.0,
+                                    "content": "puted efficiently; the cost of evaluating the kernel for a pair of images is similar",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                371,
+                                469,
+                                385
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        371,
+                                        469,
+                                        385
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to a single forward pass through the original CNN with only one filter per layer.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                382,
+                                469,
+                                396
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        382,
+                                        350,
+                                        396
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The kernel equivalent to a 32-layer ResNet obtains",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        351,
+                                        383,
+                                        378,
+                                        393
+                                    ],
+                                    "score": 0.87,
+                                    "content": "0 . 8 4 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        378,
+                                        382,
+                                        469,
+                                        396
+                                    ],
+                                    "score": 1.0,
+                                    "content": "classification error on",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                394,
+                                434,
+                                406
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        394,
+                                        434,
+                                        406
+                                    ],
+                                    "score": 1.0,
+                                    "content": "MNIST, a new record for GPs with a comparable number of parameters.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        }
+                    ],
+                    "index": 15.5
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        427,
+                        206,
+                        440
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                426,
+                                208,
+                                443
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        426,
+                                        208,
+                                        443
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1 INTRODUCTION",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        }
+                    ],
+                    "index": 21
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        453,
+                        505,
+                        585
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                453,
+                                505,
+                                465
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        453,
+                                        505,
+                                        465
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Convolutional Neural Networks (CNNs) have powerful pattern-recognition capabilities that have",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                465,
+                                504,
+                                477
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        465,
+                                        504,
+                                        477
+                                    ],
+                                    "score": 1.0,
+                                    "content": "recently given dramatic improvements in important tasks such as image classification (Krizhevsky",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                475,
+                                504,
+                                487
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        475,
+                                        504,
+                                        487
+                                    ],
+                                    "score": 1.0,
+                                    "content": "et al., 2012). However, as CNNs are increasingly being applied in real-world, safety-critical do-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                486,
+                                506,
+                                498
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        486,
+                                        506,
+                                        498
+                                    ],
+                                    "score": 1.0,
+                                    "content": "mains, their vulnerability to adversarial examples (Szegedy et al., 2013; Kurakin et al., 2016), and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                497,
+                                506,
+                                510
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        497,
+                                        506,
+                                        510
+                                    ],
+                                    "score": 1.0,
+                                    "content": "their poor uncertainty estimates are becoming increasingly problematic. Bayesian inference is a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                507,
+                                506,
+                                521
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        507,
+                                        506,
+                                        521
+                                    ],
+                                    "score": 1.0,
+                                    "content": "theoretically principled and demonstrably successful (Snoek et al., 2012; Deisenroth & Rasmussen,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                518,
+                                505,
+                                532
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        518,
+                                        505,
+                                        532
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2011) framework for learning in the face of uncertainty, which may also help to address the prob-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                529,
+                                505,
+                                542
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        529,
+                                        505,
+                                        542
+                                    ],
+                                    "score": 1.0,
+                                    "content": "lems of adversarial examples (Gal & Smith, 2018). Unfortunately, Bayesian inference in CNNs is",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                541,
+                                505,
+                                554
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        541,
+                                        505,
+                                        554
+                                    ],
+                                    "score": 1.0,
+                                    "content": "extremely difficult due to the very large number of parameters, requiring highly approximate fac-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                550,
+                                505,
+                                565
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        550,
+                                        505,
+                                        565
+                                    ],
+                                    "score": 1.0,
+                                    "content": "torised variational approximations (Blundell et al., 2015; Gal & Ghahramani, 2015), or requiring",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                562,
+                                505,
+                                576
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        562,
+                                        505,
+                                        576
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the storage (Lakshminarayanan et al., 2017) of large numbers of posterior samples (Welling & Teh,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                574,
+                                212,
+                                585
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        574,
+                                        212,
+                                        585
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2011; Mandt et al., 2017).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        }
+                    ],
+                    "index": 27.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        590,
+                        505,
+                        711
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                590,
+                                505,
+                                603
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        590,
+                                        505,
+                                        603
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Other methods such as those based on Gaussian Processes (GPs) are more amenable to Bayesian",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                602,
+                                504,
+                                614
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        602,
+                                        504,
+                                        614
+                                    ],
+                                    "score": 1.0,
+                                    "content": "inference, allowing us to compute the posterior uncertainty exactly (Rasmussen & Williams, 2006).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                612,
+                                505,
+                                626
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        612,
+                                        505,
+                                        626
+                                    ],
+                                    "score": 1.0,
+                                    "content": "This raises the question of whether it might be possible to combine the pattern-recognition capa-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                623,
+                                505,
+                                636
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        623,
+                                        505,
+                                        636
+                                    ],
+                                    "score": 1.0,
+                                    "content": "bilities of CNNs with exact probabilistic computations in GPs. Two such approaches exist in the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                633,
+                                505,
+                                648
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        633,
+                                        505,
+                                        648
+                                    ],
+                                    "score": 1.0,
+                                    "content": "literature. First, deep convolutional kernels (Wilson et al., 2016) parameterise a GP kernel using the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                645,
+                                506,
+                                659
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        645,
+                                        506,
+                                        659
+                                    ],
+                                    "score": 1.0,
+                                    "content": "weights and biases of a CNN, which is used to embed the input images into some latent space before",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                656,
+                                506,
+                                669
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        656,
+                                        506,
+                                        669
+                                    ],
+                                    "score": 1.0,
+                                    "content": "computing their similarity. The CNN parameters of the resulting kernel then have to be optimised",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                667,
+                                506,
+                                680
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        667,
+                                        506,
+                                        680
+                                    ],
+                                    "score": 1.0,
+                                    "content": "by gradient descent. However, the large number of kernel parameters in the CNN reintroduces the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                678,
+                                505,
+                                691
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        678,
+                                        505,
+                                        691
+                                    ],
+                                    "score": 1.0,
+                                    "content": "risk of overconfidence and overfitting. To avoid this risk, we need to infer a posterior over the CNN",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                690,
+                                506,
+                                702
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        690,
+                                        506,
+                                        702
+                                    ],
+                                    "score": 1.0,
+                                    "content": "kernel parameters, which is as difficult as directly inferring a posterior over the parameters of the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                700,
+                                506,
+                                713
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        700,
+                                        506,
+                                        713
+                                    ],
+                                    "score": 1.0,
+                                    "content": "original CNN. Second, it is possible to define a convolutional GP (van der Wilk et al., 2017) or a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        }
+                    ],
+                    "index": 39
+                }
+            ],
+            "page_idx": 0,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        118,
+                        721,
+                        461,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                119,
+                                719,
+                                463,
+                                735
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        119,
+                                        719,
+                                        463,
+                                        735
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1Code to replicate this paper is available at https://github.com/convnets-as-gps/convnets-as-gps",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        107,
+                        27,
+                        293,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                25,
+                                294,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        25,
+                                        294,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Published as a conference paper at ICLR 2019",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        387,
+                        135,
+                        498,
+                        179
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                387,
+                                134,
+                                496,
+                                146
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        387,
+                                        134,
+                                        496,
+                                        146
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Carl Edward Rasmussen",
+                                    "type": "text"
+                                }
+                            ]
+                        },
+                        {
+                            "bbox": [
+                                387,
+                                145,
+                                499,
+                                159
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        387,
+                                        145,
+                                        499,
+                                        159
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Department of Engineering",
+                                    "type": "text"
+                                }
+                            ]
+                        },
+                        {
+                            "bbox": [
+                                387,
+                                156,
+                                489,
+                                170
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        387,
+                                        156,
+                                        489,
+                                        170
+                                    ],
+                                    "score": 1.0,
+                                    "content": "University of Cambridge",
+                                    "type": "text"
+                                }
+                            ]
+                        },
+                        {
+                            "bbox": [
+                                388,
+                                168,
+                                479,
+                                180
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        388,
+                                        168,
+                                        479,
+                                        180
+                                    ],
+                                    "score": 1.0,
+                                    "content": "cer54@cam.ac.uk",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        302,
+                        751,
+                        308,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                302,
+                                751,
+                                309,
+                                762
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        302,
+                                        751,
+                                        309,
+                                        762
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "title",
+                    "bbox": [
+                        106,
+                        79,
+                        397,
+                        116
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                77,
+                                398,
+                                97
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        77,
+                                        398,
+                                        97
+                                    ],
+                                    "score": 1.0,
+                                    "content": "DEEP CONVOLUTIONAL NETWORKS AS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                99,
+                                354,
+                                117
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        99,
+                                        354,
+                                        117
+                                    ],
+                                    "score": 1.0,
+                                    "content": "SHALLOW GAUSSIAN PROCESSES",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        }
+                    ],
+                    "index": 0.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        113,
+                        135,
+                        221,
+                        179
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                114,
+                                134,
+                                212,
+                                147
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        114,
+                                        134,
+                                        212,
+                                        147
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Adrià Garriga-Alonso",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                112,
+                                144,
+                                222,
+                                160
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        112,
+                                        144,
+                                        222,
+                                        160
+                                    ],
+                                    "score": 1.0,
+                                    "content": "department of Engineering",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                112,
+                                156,
+                                214,
+                                170
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        112,
+                                        156,
+                                        214,
+                                        170
+                                    ],
+                                    "score": 1.0,
+                                    "content": "University of Cambridge",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                112,
+                                168,
+                                205,
+                                180
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        112,
+                                        168,
+                                        205,
+                                        180
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ag919@cam.ac.uk",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        }
+                    ],
+                    "index": 3.5,
+                    "bbox_fs": [
+                        112,
+                        134,
+                        222,
+                        180
+                    ]
+                },
+                {
+                    "type": "list",
+                    "bbox": [
+                        113,
+                        196,
+                        282,
+                        240
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                111,
+                                196,
+                                198,
+                                209
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        111,
+                                        196,
+                                        198,
+                                        209
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Laurence Aitchison",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                111,
+                                205,
+                                225,
+                                221
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        111,
+                                        205,
+                                        225,
+                                        221
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Department of Engineering",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                112,
+                                217,
+                                214,
+                                231
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        112,
+                                        217,
+                                        214,
+                                        231
+                                    ],
+                                    "score": 1.0,
+                                    "content": "University of Cambridge",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                112,
+                                229,
+                                282,
+                                242
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        112,
+                                        229,
+                                        282,
+                                        242
+                                    ],
+                                    "score": 1.0,
+                                    "content": "laurence.aitchison@gmail.com",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9,
+                            "is_list_start_line": true
+                        }
+                    ],
+                    "index": 7.5,
+                    "bbox_fs": [
+                        111,
+                        196,
+                        282,
+                        242
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        278,
+                        270,
+                        333,
+                        281
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                276,
+                                268,
+                                336,
+                                284
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        276,
+                                        268,
+                                        336,
+                                        284
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ABSTRACT",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        }
+                    ],
+                    "index": 10
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        143,
+                        294,
+                        468,
+                        405
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                142,
+                                294,
+                                469,
+                                308
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        142,
+                                        294,
+                                        469,
+                                        308
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We show that the output of a (residual) convolutional neural network (CNN) with",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                306,
+                                470,
+                                319
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        306,
+                                        470,
+                                        319
+                                    ],
+                                    "score": 1.0,
+                                    "content": "an appropriate prior over the weights and biases is a Gaussian process (GP) in the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                317,
+                                470,
+                                329
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        317,
+                                        470,
+                                        329
+                                    ],
+                                    "score": 1.0,
+                                    "content": "limit of infinitely many convolutional filters, extending similar results for dense",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                328,
+                                470,
+                                341
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        328,
+                                        470,
+                                        341
+                                    ],
+                                    "score": 1.0,
+                                    "content": "networks. For a CNN, the equivalent kernel can be computed exactly and, unlike",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                338,
+                                470,
+                                352
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        338,
+                                        470,
+                                        352
+                                    ],
+                                    "score": 1.0,
+                                    "content": "“deep kernels”, has very few parameters: only the hyperparameters of the original",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                349,
+                                469,
+                                362
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        349,
+                                        469,
+                                        362
+                                    ],
+                                    "score": 1.0,
+                                    "content": "CNN. Further, we show that this kernel has two properties that allow it to be com-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                361,
+                                470,
+                                373
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        361,
+                                        470,
+                                        373
+                                    ],
+                                    "score": 1.0,
+                                    "content": "puted efficiently; the cost of evaluating the kernel for a pair of images is similar",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                371,
+                                469,
+                                385
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        371,
+                                        469,
+                                        385
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to a single forward pass through the original CNN with only one filter per layer.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                382,
+                                469,
+                                396
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        382,
+                                        350,
+                                        396
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The kernel equivalent to a 32-layer ResNet obtains",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        351,
+                                        383,
+                                        378,
+                                        393
+                                    ],
+                                    "score": 0.87,
+                                    "content": "0 . 8 4 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        378,
+                                        382,
+                                        469,
+                                        396
+                                    ],
+                                    "score": 1.0,
+                                    "content": "classification error on",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                394,
+                                434,
+                                406
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        394,
+                                        434,
+                                        406
+                                    ],
+                                    "score": 1.0,
+                                    "content": "MNIST, a new record for GPs with a comparable number of parameters.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        }
+                    ],
+                    "index": 15.5,
+                    "bbox_fs": [
+                        141,
+                        294,
+                        470,
+                        406
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        427,
+                        206,
+                        440
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                426,
+                                208,
+                                443
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        426,
+                                        208,
+                                        443
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1 INTRODUCTION",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        }
+                    ],
+                    "index": 21
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        453,
+                        505,
+                        585
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                453,
+                                505,
+                                465
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        453,
+                                        505,
+                                        465
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Convolutional Neural Networks (CNNs) have powerful pattern-recognition capabilities that have",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                465,
+                                504,
+                                477
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        465,
+                                        504,
+                                        477
+                                    ],
+                                    "score": 1.0,
+                                    "content": "recently given dramatic improvements in important tasks such as image classification (Krizhevsky",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                475,
+                                504,
+                                487
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        475,
+                                        504,
+                                        487
+                                    ],
+                                    "score": 1.0,
+                                    "content": "et al., 2012). However, as CNNs are increasingly being applied in real-world, safety-critical do-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                486,
+                                506,
+                                498
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        486,
+                                        506,
+                                        498
+                                    ],
+                                    "score": 1.0,
+                                    "content": "mains, their vulnerability to adversarial examples (Szegedy et al., 2013; Kurakin et al., 2016), and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                497,
+                                506,
+                                510
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        497,
+                                        506,
+                                        510
+                                    ],
+                                    "score": 1.0,
+                                    "content": "their poor uncertainty estimates are becoming increasingly problematic. Bayesian inference is a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                507,
+                                506,
+                                521
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        507,
+                                        506,
+                                        521
+                                    ],
+                                    "score": 1.0,
+                                    "content": "theoretically principled and demonstrably successful (Snoek et al., 2012; Deisenroth & Rasmussen,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                518,
+                                505,
+                                532
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        518,
+                                        505,
+                                        532
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2011) framework for learning in the face of uncertainty, which may also help to address the prob-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                529,
+                                505,
+                                542
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        529,
+                                        505,
+                                        542
+                                    ],
+                                    "score": 1.0,
+                                    "content": "lems of adversarial examples (Gal & Smith, 2018). Unfortunately, Bayesian inference in CNNs is",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                541,
+                                505,
+                                554
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        541,
+                                        505,
+                                        554
+                                    ],
+                                    "score": 1.0,
+                                    "content": "extremely difficult due to the very large number of parameters, requiring highly approximate fac-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                550,
+                                505,
+                                565
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        550,
+                                        505,
+                                        565
+                                    ],
+                                    "score": 1.0,
+                                    "content": "torised variational approximations (Blundell et al., 2015; Gal & Ghahramani, 2015), or requiring",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                562,
+                                505,
+                                576
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        562,
+                                        505,
+                                        576
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the storage (Lakshminarayanan et al., 2017) of large numbers of posterior samples (Welling & Teh,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                574,
+                                212,
+                                585
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        574,
+                                        212,
+                                        585
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2011; Mandt et al., 2017).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        }
+                    ],
+                    "index": 27.5,
+                    "bbox_fs": [
+                        105,
+                        453,
+                        506,
+                        585
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        590,
+                        505,
+                        711
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                590,
+                                505,
+                                603
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        590,
+                                        505,
+                                        603
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Other methods such as those based on Gaussian Processes (GPs) are more amenable to Bayesian",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                602,
+                                504,
+                                614
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        602,
+                                        504,
+                                        614
+                                    ],
+                                    "score": 1.0,
+                                    "content": "inference, allowing us to compute the posterior uncertainty exactly (Rasmussen & Williams, 2006).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                612,
+                                505,
+                                626
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        612,
+                                        505,
+                                        626
+                                    ],
+                                    "score": 1.0,
+                                    "content": "This raises the question of whether it might be possible to combine the pattern-recognition capa-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                623,
+                                505,
+                                636
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        623,
+                                        505,
+                                        636
+                                    ],
+                                    "score": 1.0,
+                                    "content": "bilities of CNNs with exact probabilistic computations in GPs. Two such approaches exist in the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                633,
+                                505,
+                                648
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        633,
+                                        505,
+                                        648
+                                    ],
+                                    "score": 1.0,
+                                    "content": "literature. First, deep convolutional kernels (Wilson et al., 2016) parameterise a GP kernel using the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                645,
+                                506,
+                                659
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        645,
+                                        506,
+                                        659
+                                    ],
+                                    "score": 1.0,
+                                    "content": "weights and biases of a CNN, which is used to embed the input images into some latent space before",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                656,
+                                506,
+                                669
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        656,
+                                        506,
+                                        669
+                                    ],
+                                    "score": 1.0,
+                                    "content": "computing their similarity. The CNN parameters of the resulting kernel then have to be optimised",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                667,
+                                506,
+                                680
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        667,
+                                        506,
+                                        680
+                                    ],
+                                    "score": 1.0,
+                                    "content": "by gradient descent. However, the large number of kernel parameters in the CNN reintroduces the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                678,
+                                505,
+                                691
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        678,
+                                        505,
+                                        691
+                                    ],
+                                    "score": 1.0,
+                                    "content": "risk of overconfidence and overfitting. To avoid this risk, we need to infer a posterior over the CNN",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                690,
+                                506,
+                                702
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        690,
+                                        506,
+                                        702
+                                    ],
+                                    "score": 1.0,
+                                    "content": "kernel parameters, which is as difficult as directly inferring a posterior over the parameters of the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                700,
+                                506,
+                                713
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        700,
+                                        506,
+                                        713
+                                    ],
+                                    "score": 1.0,
+                                    "content": "original CNN. Second, it is possible to define a convolutional GP (van der Wilk et al., 2017) or a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                82,
+                                505,
+                                96
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        82,
+                                        505,
+                                        96
+                                    ],
+                                    "score": 1.0,
+                                    "content": "deep convolutional GP (Kumar et al., 2018) by defining a GP that takes an image patch as input,",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                93,
+                                505,
+                                106
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        93,
+                                        505,
+                                        106
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and using that GP as a component in a larger CNN-like system. However, inference in such sys-",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                105,
+                                506,
+                                118
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        105,
+                                        506,
+                                        118
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tems is very computationally expensive, at least without the use of potentially severe variational",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                116,
+                                281,
+                                127
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        116,
+                                        281,
+                                        127
+                                    ],
+                                    "score": 1.0,
+                                    "content": "approximations (van der Wilk et al., 2017).",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 3
+                        }
+                    ],
+                    "index": 39,
+                    "bbox_fs": [
+                        105,
+                        590,
+                        506,
+                        713
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        82,
+                        504,
+                        126
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                82,
+                                505,
+                                96
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        82,
+                                        505,
+                                        96
+                                    ],
+                                    "score": 1.0,
+                                    "content": "deep convolutional GP (Kumar et al., 2018) by defining a GP that takes an image patch as input,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                93,
+                                505,
+                                106
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        93,
+                                        505,
+                                        106
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and using that GP as a component in a larger CNN-like system. However, inference in such sys-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                105,
+                                506,
+                                118
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        105,
+                                        506,
+                                        118
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tems is very computationally expensive, at least without the use of potentially severe variational",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                116,
+                                281,
+                                127
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        116,
+                                        281,
+                                        127
+                                    ],
+                                    "score": 1.0,
+                                    "content": "approximations (van der Wilk et al., 2017).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        }
+                    ],
+                    "index": 1.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        132,
+                        505,
+                        275
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                133,
+                                504,
+                                144
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        133,
+                                        504,
+                                        144
+                                    ],
+                                    "score": 1.0,
+                                    "content": "An alternative approach is suggested by the underlying connection between Bayesian neural net-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                142,
+                                505,
+                                156
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        142,
+                                        505,
+                                        156
+                                    ],
+                                    "score": 1.0,
+                                    "content": "works (NNs) and GPs. In particular, Neal (1996) showed that the function defined by a single-layer",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                155,
+                                505,
+                                166
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        155,
+                                        505,
+                                        166
+                                    ],
+                                    "score": 1.0,
+                                    "content": "fully-connected NN with infinitely many hidden units, and random independent zero-mean weights",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                165,
+                                505,
+                                178
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        165,
+                                        505,
+                                        178
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and biases is equivalent to a GP, implying that we can do exact Bayesian inference in such a NN",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                176,
+                                504,
+                                189
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        176,
+                                        504,
+                                        189
+                                    ],
+                                    "score": 1.0,
+                                    "content": "by working with the equivalent GP. Recently, this result was extended to arbitrarily deep fully-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                187,
+                                504,
+                                199
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        187,
+                                        504,
+                                        199
+                                    ],
+                                    "score": 1.0,
+                                    "content": "connected NNs with infinitely many hidden units at each layer (Lee et al., 2017; Matthews et al.,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                198,
+                                505,
+                                210
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        198,
+                                        505,
+                                        210
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2018a). However, these fully-connected networks are rarely used in practice, as they are unable",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                209,
+                                506,
+                                222
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        209,
+                                        506,
+                                        222
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to exploit important properties of images such as translational invariance, raising the question of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                220,
+                                505,
+                                233
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        220,
+                                        505,
+                                        233
+                                    ],
+                                    "score": 1.0,
+                                    "content": "whether state-of-the-art architectures such as CNNs (LeCun et al., 1990) and ResNets (He et al.,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                231,
+                                506,
+                                244
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        231,
+                                        506,
+                                        244
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2016a) have equivalent GP representations. Here, we answer in the affirmative, giving the GP ker-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                243,
+                                505,
+                                254
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        243,
+                                        505,
+                                        254
+                                    ],
+                                    "score": 1.0,
+                                    "content": "nel corresponding to arbitrarily deep CNNs and to (convolutional) residual neural networks (He",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                252,
+                                506,
+                                266
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        252,
+                                        506,
+                                        266
+                                    ],
+                                    "score": 1.0,
+                                    "content": "et al., 2016a). In this case, if each hidden layer has an infinite number of convolutional filters, the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                264,
+                                250,
+                                276
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        264,
+                                        250,
+                                        276
+                                    ],
+                                    "score": 1.0,
+                                    "content": "network prior is equivalent to a GP.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        }
+                    ],
+                    "index": 10
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        281,
+                        505,
+                        390
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                281,
+                                505,
+                                293
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        281,
+                                        505,
+                                        293
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Furthermore, we show that two properties of the GP kernel induced by a CNN allow it to be com-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                291,
+                                506,
+                                304
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        291,
+                                        506,
+                                        304
+                                    ],
+                                    "score": 1.0,
+                                    "content": "puted very efficiently. First, in previous work it was necessary to compute the covariance matrix",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                302,
+                                505,
+                                316
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        302,
+                                        505,
+                                        316
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for the output of a single convolutional filter applied at all possible locations within a single image",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                313,
+                                505,
+                                326
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        313,
+                                        505,
+                                        326
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(van der Wilk et al., 2017), which was prohibitively computationally expensive. In contrast, under",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                325,
+                                505,
+                                337
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        325,
+                                        505,
+                                        337
+                                    ],
+                                    "score": 1.0,
+                                    "content": "our prior, the downstream weights are independent with zero-mean, which decorrelates the contri-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                335,
+                                506,
+                                348
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        335,
+                                        506,
+                                        348
+                                    ],
+                                    "score": 1.0,
+                                    "content": "bution from each location, and implies that it is necessary only to track the patch variances, and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                347,
+                                506,
+                                358
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        347,
+                                        506,
+                                        358
+                                    ],
+                                    "score": 1.0,
+                                    "content": "not their covariances. Second, while it is still necessary to compute the variance of the output of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                357,
+                                506,
+                                369
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        357,
+                                        506,
+                                        369
+                                    ],
+                                    "score": 1.0,
+                                    "content": "a convolutional filter applied at all locations within the image, the specific structure of the kernel",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                367,
+                                505,
+                                381
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        367,
+                                        505,
+                                        381
+                                    ],
+                                    "score": 1.0,
+                                    "content": "induced by the CNN means that the variance at every location can be computed simultaneously and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                379,
+                                219,
+                                392
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        379,
+                                        219,
+                                        392
+                                    ],
+                                    "score": 1.0,
+                                    "content": "efficiently as a convolution.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        }
+                    ],
+                    "index": 21.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        396,
+                        504,
+                        441
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                396,
+                                504,
+                                409
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        396,
+                                        504,
+                                        409
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Finally, we empirically demonstrate the performance increase coming from adding translation-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                407,
+                                506,
+                                420
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        407,
+                                        506,
+                                        420
+                                    ],
+                                    "score": 1.0,
+                                    "content": "invariant structure to the GP prior. Without computing any gradients, and without augmenting",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                419,
+                                506,
+                                430
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        419,
+                                        316,
+                                        430
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the training set (e.g. using translations), we obtain",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        316,
+                                        419,
+                                        344,
+                                        429
+                                    ],
+                                    "score": 0.87,
+                                    "content": "0 . 8 4 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        344,
+                                        419,
+                                        506,
+                                        430
+                                    ],
+                                    "score": 1.0,
+                                    "content": "error rate on the MNIST classification",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                430,
+                                393,
+                                441
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        430,
+                                        393,
+                                        441
+                                    ],
+                                    "score": 1.0,
+                                    "content": "benchmark, setting a new record for nonparametric GP-based methods.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        }
+                    ],
+                    "index": 28.5
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        459,
+                        260,
+                        472
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                458,
+                                262,
+                                474
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        458,
+                                        262,
+                                        474
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2 GP BEHAVIOUR IN A CNN",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        }
+                    ],
+                    "index": 31
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        486,
+                        505,
+                        520
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                486,
+                                505,
+                                499
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        486,
+                                        505,
+                                        499
+                                    ],
+                                    "score": 1.0,
+                                    "content": "For clarity of exposition, we will treat the case of a 2D convolutional NN. The result applies straight-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                498,
+                                505,
+                                509
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        498,
+                                        160,
+                                        509
+                                    ],
+                                    "score": 1.0,
+                                    "content": "forwardly to",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        160,
+                                        498,
+                                        175,
+                                        507
+                                    ],
+                                    "score": 0.41,
+                                    "content": "n \\mathbf { D }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        175,
+                                        498,
+                                        505,
+                                        509
+                                    ],
+                                    "score": 1.0,
+                                    "content": "convolutions, dilated convolutions and upconvolutions (“deconvolutions”), since",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                508,
+                                439,
+                                522
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        508,
+                                        439,
+                                        522
+                                    ],
+                                    "score": 1.0,
+                                    "content": "they can be represented as linear transformations with tied coefficients (see Fig. 1).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        }
+                    ],
+                    "index": 33
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        536,
+                        306,
+                        547
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                536,
+                                308,
+                                549
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        536,
+                                        308,
+                                        549
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2.1 A 2D CONVOLUTIONAL NETWORK PRIOR",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        }
+                    ],
+                    "index": 35
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        556,
+                        505,
+                        605
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                556,
+                                503,
+                                570
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        556,
+                                        277,
+                                        570
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The network takes an arbitrary input image",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        278,
+                                        558,
+                                        288,
+                                        568
+                                    ],
+                                    "score": 0.67,
+                                    "content": "\\mathbf { X }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        288,
+                                        556,
+                                        325,
+                                        570
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of height",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        326,
+                                        556,
+                                        346,
+                                        568
+                                    ],
+                                    "score": 0.89,
+                                    "content": "H ^ { ( 0 ) }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        346,
+                                        556,
+                                        388,
+                                        570
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and width",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        388,
+                                        556,
+                                        408,
+                                        568
+                                    ],
+                                    "score": 0.9,
+                                    "content": "D ^ { ( 0 ) }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        408,
+                                        556,
+                                        429,
+                                        570
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", as a",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        429,
+                                        556,
+                                        503,
+                                        569
+                                    ],
+                                    "score": 0.92,
+                                    "content": "C ^ { ( 0 ) } \\times ( H ^ { ( 0 ) } D ^ { ( 0 ) } )",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                568,
+                                506,
+                                583
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        568,
+                                        272,
+                                        583
+                                    ],
+                                    "score": 1.0,
+                                    "content": "real matrix. Each row, which we denote",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        273,
+                                        571,
+                                        343,
+                                        581
+                                    ],
+                                    "score": 0.88,
+                                    "content": "\\mathbf { x } _ { 1 } , \\mathbf { x } _ { 2 } , \\ldots , \\mathbf { x } _ { C ^ { ( 0 ) } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        343,
+                                        568,
+                                        506,
+                                        583
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", corresponds to a channel of the image",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                579,
+                                506,
+                                594
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        579,
+                                        130,
+                                        594
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(e.g.",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        131,
+                                        579,
+                                        173,
+                                        592
+                                    ],
+                                    "score": 0.91,
+                                    "content": "C ^ { ( 0 ) } = 3",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        173,
+                                        579,
+                                        417,
+                                        594
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for RGB), flattened to form a vector. The first activations",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        417,
+                                        579,
+                                        453,
+                                        593
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\mathbf { A } ^ { ( 1 ) } ( \\mathbf { X } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        454,
+                                        579,
+                                        506,
+                                        594
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are a linear",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                591,
+                                317,
+                                606
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        591,
+                                        239,
+                                        606
+                                    ],
+                                    "score": 1.0,
+                                    "content": "transformation of the inputs. For",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        240,
+                                        592,
+                                        312,
+                                        605
+                                    ],
+                                    "score": 0.93,
+                                    "content": "i \\in \\{ 1 , \\ldots , C ^ { ( 1 ) } \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        312,
+                                        591,
+                                        317,
+                                        606
+                                    ],
+                                    "score": 1.0,
+                                    "content": ":",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        }
+                    ],
+                    "index": 37.5
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        235,
+                        613,
+                        375,
+                        651
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                235,
+                                613,
+                                375,
+                                651
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        235,
+                                        613,
+                                        375,
+                                        651
+                                    ],
+                                    "score": 0.94,
+                                    "content": "\\mathbf { a } _ { i } ^ { ( 1 ) } ( \\mathbf { X } ) : = b _ { i } ^ { ( 1 ) } \\mathbf { 1 } + \\sum _ { j = 1 } ^ { C ^ { ( 0 ) } } \\mathbf { W } _ { i , j } ^ { ( 1 ) } \\mathbf { x } _ { j } \\ .",
+                                    "type": "interline_equation",
+                                    "image_path": "78d2577cfa8c3c860a766e9edaf7965be6e069473e6fa7fd0bb74ab7b7d9c1af.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 40.5,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                235,
+                                613,
+                                375,
+                                632.0
+                            ],
+                            "spans": [],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                235,
+                                632.0,
+                                375,
+                                651.0
+                            ],
+                            "spans": [],
+                            "index": 41
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        665,
+                        504,
+                        690
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                663,
+                                504,
+                                679
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        663,
+                                        222,
+                                        679
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We consider a network with",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        223,
+                                        667,
+                                        230,
+                                        676
+                                    ],
+                                    "score": 0.83,
+                                    "content": "L",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        231,
+                                        663,
+                                        467,
+                                        679
+                                    ],
+                                    "score": 1.0,
+                                    "content": "hidden layers. The other activations of the network, from",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        467,
+                                        665,
+                                        504,
+                                        678
+                                    ],
+                                    "score": 0.9,
+                                    "content": "\\mathbf { A } ^ { ( 2 ) } ( \\mathbf { X } )",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                676,
+                                278,
+                                691
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        676,
+                                        129,
+                                        691
+                                    ],
+                                    "score": 1.0,
+                                    "content": "up to",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        130,
+                                        677,
+                                        177,
+                                        690
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\mathbf { A } ^ { ( L + 1 ) } ( \\mathbf { X } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        178,
+                                        676,
+                                        278,
+                                        691
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", are defined recursively:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        }
+                    ],
+                    "index": 42.5
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        201,
+                        698,
+                        410,
+                        735
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                201,
+                                698,
+                                410,
+                                735
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        201,
+                                        698,
+                                        410,
+                                        735
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\mathbf { a } _ { i } ^ { ( \\ell + 1 ) } ( \\mathbf { X } ) : = b _ { i } ^ { ( \\ell + 1 ) } \\mathbf { 1 } + \\sum _ { j = 1 } ^ { C ^ { ( \\ell ) } } \\mathbf { W } _ { i , j } ^ { ( \\ell + 1 ) } \\phi \\left( \\mathbf { a } _ { j } ^ { ( \\ell ) } ( \\mathbf { X } ) \\right) .",
+                                    "type": "interline_equation",
+                                    "image_path": "eb19bdb6a9577e101e45b0eeeda139f9ceee459688f7506d7d09bdf291760691.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 44.5,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                201,
+                                698,
+                                410,
+                                716.5
+                            ],
+                            "spans": [],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                201,
+                                716.5,
+                                410,
+                                735.0
+                            ],
+                            "spans": [],
+                            "index": 45
+                        }
+                    ]
+                }
+            ],
+            "page_idx": 1,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        108,
+                        27,
+                        293,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                26,
+                                294,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        26,
+                                        294,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Published as a conference paper at ICLR 2019",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        302,
+                        751,
+                        309,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                301,
+                                750,
+                                310,
+                                763
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        301,
+                                        750,
+                                        310,
+                                        763
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        82,
+                        504,
+                        126
+                    ],
+                    "lines": [],
+                    "index": 1.5,
+                    "bbox_fs": [
+                        105,
+                        82,
+                        506,
+                        127
+                    ],
+                    "lines_deleted": true
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        132,
+                        505,
+                        275
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                133,
+                                504,
+                                144
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        133,
+                                        504,
+                                        144
+                                    ],
+                                    "score": 1.0,
+                                    "content": "An alternative approach is suggested by the underlying connection between Bayesian neural net-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                142,
+                                505,
+                                156
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        142,
+                                        505,
+                                        156
+                                    ],
+                                    "score": 1.0,
+                                    "content": "works (NNs) and GPs. In particular, Neal (1996) showed that the function defined by a single-layer",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                155,
+                                505,
+                                166
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        155,
+                                        505,
+                                        166
+                                    ],
+                                    "score": 1.0,
+                                    "content": "fully-connected NN with infinitely many hidden units, and random independent zero-mean weights",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                165,
+                                505,
+                                178
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        165,
+                                        505,
+                                        178
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and biases is equivalent to a GP, implying that we can do exact Bayesian inference in such a NN",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                176,
+                                504,
+                                189
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        176,
+                                        504,
+                                        189
+                                    ],
+                                    "score": 1.0,
+                                    "content": "by working with the equivalent GP. Recently, this result was extended to arbitrarily deep fully-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                187,
+                                504,
+                                199
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        187,
+                                        504,
+                                        199
+                                    ],
+                                    "score": 1.0,
+                                    "content": "connected NNs with infinitely many hidden units at each layer (Lee et al., 2017; Matthews et al.,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                198,
+                                505,
+                                210
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        198,
+                                        505,
+                                        210
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2018a). However, these fully-connected networks are rarely used in practice, as they are unable",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                209,
+                                506,
+                                222
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        209,
+                                        506,
+                                        222
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to exploit important properties of images such as translational invariance, raising the question of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                220,
+                                505,
+                                233
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        220,
+                                        505,
+                                        233
+                                    ],
+                                    "score": 1.0,
+                                    "content": "whether state-of-the-art architectures such as CNNs (LeCun et al., 1990) and ResNets (He et al.,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                231,
+                                506,
+                                244
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        231,
+                                        506,
+                                        244
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2016a) have equivalent GP representations. Here, we answer in the affirmative, giving the GP ker-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                243,
+                                505,
+                                254
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        243,
+                                        505,
+                                        254
+                                    ],
+                                    "score": 1.0,
+                                    "content": "nel corresponding to arbitrarily deep CNNs and to (convolutional) residual neural networks (He",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                252,
+                                506,
+                                266
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        252,
+                                        506,
+                                        266
+                                    ],
+                                    "score": 1.0,
+                                    "content": "et al., 2016a). In this case, if each hidden layer has an infinite number of convolutional filters, the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                264,
+                                250,
+                                276
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        264,
+                                        250,
+                                        276
+                                    ],
+                                    "score": 1.0,
+                                    "content": "network prior is equivalent to a GP.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        }
+                    ],
+                    "index": 10,
+                    "bbox_fs": [
+                        104,
+                        133,
+                        506,
+                        276
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        281,
+                        505,
+                        390
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                281,
+                                505,
+                                293
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        281,
+                                        505,
+                                        293
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Furthermore, we show that two properties of the GP kernel induced by a CNN allow it to be com-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                291,
+                                506,
+                                304
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        291,
+                                        506,
+                                        304
+                                    ],
+                                    "score": 1.0,
+                                    "content": "puted very efficiently. First, in previous work it was necessary to compute the covariance matrix",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                302,
+                                505,
+                                316
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        302,
+                                        505,
+                                        316
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for the output of a single convolutional filter applied at all possible locations within a single image",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                313,
+                                505,
+                                326
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        313,
+                                        505,
+                                        326
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(van der Wilk et al., 2017), which was prohibitively computationally expensive. In contrast, under",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                325,
+                                505,
+                                337
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        325,
+                                        505,
+                                        337
+                                    ],
+                                    "score": 1.0,
+                                    "content": "our prior, the downstream weights are independent with zero-mean, which decorrelates the contri-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                335,
+                                506,
+                                348
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        335,
+                                        506,
+                                        348
+                                    ],
+                                    "score": 1.0,
+                                    "content": "bution from each location, and implies that it is necessary only to track the patch variances, and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                347,
+                                506,
+                                358
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        347,
+                                        506,
+                                        358
+                                    ],
+                                    "score": 1.0,
+                                    "content": "not their covariances. Second, while it is still necessary to compute the variance of the output of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                357,
+                                506,
+                                369
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        357,
+                                        506,
+                                        369
+                                    ],
+                                    "score": 1.0,
+                                    "content": "a convolutional filter applied at all locations within the image, the specific structure of the kernel",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                367,
+                                505,
+                                381
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        367,
+                                        505,
+                                        381
+                                    ],
+                                    "score": 1.0,
+                                    "content": "induced by the CNN means that the variance at every location can be computed simultaneously and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                379,
+                                219,
+                                392
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        379,
+                                        219,
+                                        392
+                                    ],
+                                    "score": 1.0,
+                                    "content": "efficiently as a convolution.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        }
+                    ],
+                    "index": 21.5,
+                    "bbox_fs": [
+                        104,
+                        281,
+                        506,
+                        392
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        396,
+                        504,
+                        441
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                396,
+                                504,
+                                409
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        396,
+                                        504,
+                                        409
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Finally, we empirically demonstrate the performance increase coming from adding translation-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                407,
+                                506,
+                                420
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        407,
+                                        506,
+                                        420
+                                    ],
+                                    "score": 1.0,
+                                    "content": "invariant structure to the GP prior. Without computing any gradients, and without augmenting",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                419,
+                                506,
+                                430
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        419,
+                                        316,
+                                        430
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the training set (e.g. using translations), we obtain",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        316,
+                                        419,
+                                        344,
+                                        429
+                                    ],
+                                    "score": 0.87,
+                                    "content": "0 . 8 4 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        344,
+                                        419,
+                                        506,
+                                        430
+                                    ],
+                                    "score": 1.0,
+                                    "content": "error rate on the MNIST classification",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                430,
+                                393,
+                                441
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        430,
+                                        393,
+                                        441
+                                    ],
+                                    "score": 1.0,
+                                    "content": "benchmark, setting a new record for nonparametric GP-based methods.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        }
+                    ],
+                    "index": 28.5,
+                    "bbox_fs": [
+                        105,
+                        396,
+                        506,
+                        441
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        459,
+                        260,
+                        472
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                458,
+                                262,
+                                474
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        458,
+                                        262,
+                                        474
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2 GP BEHAVIOUR IN A CNN",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        }
+                    ],
+                    "index": 31
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        486,
+                        505,
+                        520
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                486,
+                                505,
+                                499
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        486,
+                                        505,
+                                        499
+                                    ],
+                                    "score": 1.0,
+                                    "content": "For clarity of exposition, we will treat the case of a 2D convolutional NN. The result applies straight-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                498,
+                                505,
+                                509
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        498,
+                                        160,
+                                        509
+                                    ],
+                                    "score": 1.0,
+                                    "content": "forwardly to",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        160,
+                                        498,
+                                        175,
+                                        507
+                                    ],
+                                    "score": 0.41,
+                                    "content": "n \\mathbf { D }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        175,
+                                        498,
+                                        505,
+                                        509
+                                    ],
+                                    "score": 1.0,
+                                    "content": "convolutions, dilated convolutions and upconvolutions (“deconvolutions”), since",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                508,
+                                439,
+                                522
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        508,
+                                        439,
+                                        522
+                                    ],
+                                    "score": 1.0,
+                                    "content": "they can be represented as linear transformations with tied coefficients (see Fig. 1).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        }
+                    ],
+                    "index": 33,
+                    "bbox_fs": [
+                        105,
+                        486,
+                        505,
+                        522
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        536,
+                        306,
+                        547
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                536,
+                                308,
+                                549
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        536,
+                                        308,
+                                        549
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2.1 A 2D CONVOLUTIONAL NETWORK PRIOR",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        }
+                    ],
+                    "index": 35
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        556,
+                        505,
+                        605
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                556,
+                                503,
+                                570
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        556,
+                                        277,
+                                        570
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The network takes an arbitrary input image",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        278,
+                                        558,
+                                        288,
+                                        568
+                                    ],
+                                    "score": 0.67,
+                                    "content": "\\mathbf { X }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        288,
+                                        556,
+                                        325,
+                                        570
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of height",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        326,
+                                        556,
+                                        346,
+                                        568
+                                    ],
+                                    "score": 0.89,
+                                    "content": "H ^ { ( 0 ) }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        346,
+                                        556,
+                                        388,
+                                        570
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and width",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        388,
+                                        556,
+                                        408,
+                                        568
+                                    ],
+                                    "score": 0.9,
+                                    "content": "D ^ { ( 0 ) }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        408,
+                                        556,
+                                        429,
+                                        570
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", as a",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        429,
+                                        556,
+                                        503,
+                                        569
+                                    ],
+                                    "score": 0.92,
+                                    "content": "C ^ { ( 0 ) } \\times ( H ^ { ( 0 ) } D ^ { ( 0 ) } )",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                568,
+                                506,
+                                583
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        568,
+                                        272,
+                                        583
+                                    ],
+                                    "score": 1.0,
+                                    "content": "real matrix. Each row, which we denote",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        273,
+                                        571,
+                                        343,
+                                        581
+                                    ],
+                                    "score": 0.88,
+                                    "content": "\\mathbf { x } _ { 1 } , \\mathbf { x } _ { 2 } , \\ldots , \\mathbf { x } _ { C ^ { ( 0 ) } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        343,
+                                        568,
+                                        506,
+                                        583
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", corresponds to a channel of the image",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                579,
+                                506,
+                                594
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        579,
+                                        130,
+                                        594
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(e.g.",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        131,
+                                        579,
+                                        173,
+                                        592
+                                    ],
+                                    "score": 0.91,
+                                    "content": "C ^ { ( 0 ) } = 3",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        173,
+                                        579,
+                                        417,
+                                        594
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for RGB), flattened to form a vector. The first activations",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        417,
+                                        579,
+                                        453,
+                                        593
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\mathbf { A } ^ { ( 1 ) } ( \\mathbf { X } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        454,
+                                        579,
+                                        506,
+                                        594
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are a linear",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                591,
+                                317,
+                                606
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        591,
+                                        239,
+                                        606
+                                    ],
+                                    "score": 1.0,
+                                    "content": "transformation of the inputs. For",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        240,
+                                        592,
+                                        312,
+                                        605
+                                    ],
+                                    "score": 0.93,
+                                    "content": "i \\in \\{ 1 , \\ldots , C ^ { ( 1 ) } \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        312,
+                                        591,
+                                        317,
+                                        606
+                                    ],
+                                    "score": 1.0,
+                                    "content": ":",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        }
+                    ],
+                    "index": 37.5,
+                    "bbox_fs": [
+                        104,
+                        556,
+                        506,
+                        606
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        235,
+                        613,
+                        375,
+                        651
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                235,
+                                613,
+                                375,
+                                651
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        235,
+                                        613,
+                                        375,
+                                        651
+                                    ],
+                                    "score": 0.94,
+                                    "content": "\\mathbf { a } _ { i } ^ { ( 1 ) } ( \\mathbf { X } ) : = b _ { i } ^ { ( 1 ) } \\mathbf { 1 } + \\sum _ { j = 1 } ^ { C ^ { ( 0 ) } } \\mathbf { W } _ { i , j } ^ { ( 1 ) } \\mathbf { x } _ { j } \\ .",
+                                    "type": "interline_equation",
+                                    "image_path": "78d2577cfa8c3c860a766e9edaf7965be6e069473e6fa7fd0bb74ab7b7d9c1af.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 40.5,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                235,
+                                613,
+                                375,
+                                632.0
+                            ],
+                            "spans": [],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                235,
+                                632.0,
+                                375,
+                                651.0
+                            ],
+                            "spans": [],
+                            "index": 41
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        665,
+                        504,
+                        690
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                663,
+                                504,
+                                679
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        663,
+                                        222,
+                                        679
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We consider a network with",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        223,
+                                        667,
+                                        230,
+                                        676
+                                    ],
+                                    "score": 0.83,
+                                    "content": "L",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        231,
+                                        663,
+                                        467,
+                                        679
+                                    ],
+                                    "score": 1.0,
+                                    "content": "hidden layers. The other activations of the network, from",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        467,
+                                        665,
+                                        504,
+                                        678
+                                    ],
+                                    "score": 0.9,
+                                    "content": "\\mathbf { A } ^ { ( 2 ) } ( \\mathbf { X } )",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                676,
+                                278,
+                                691
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        676,
+                                        129,
+                                        691
+                                    ],
+                                    "score": 1.0,
+                                    "content": "up to",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        130,
+                                        677,
+                                        177,
+                                        690
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\mathbf { A } ^ { ( L + 1 ) } ( \\mathbf { X } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        178,
+                                        676,
+                                        278,
+                                        691
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", are defined recursively:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        }
+                    ],
+                    "index": 42.5,
+                    "bbox_fs": [
+                        105,
+                        663,
+                        504,
+                        691
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        201,
+                        698,
+                        410,
+                        735
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                201,
+                                698,
+                                410,
+                                735
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        201,
+                                        698,
+                                        410,
+                                        735
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\mathbf { a } _ { i } ^ { ( \\ell + 1 ) } ( \\mathbf { X } ) : = b _ { i } ^ { ( \\ell + 1 ) } \\mathbf { 1 } + \\sum _ { j = 1 } ^ { C ^ { ( \\ell ) } } \\mathbf { W } _ { i , j } ^ { ( \\ell + 1 ) } \\phi \\left( \\mathbf { a } _ { j } ^ { ( \\ell ) } ( \\mathbf { X } ) \\right) .",
+                                    "type": "interline_equation",
+                                    "image_path": "eb19bdb6a9577e101e45b0eeeda139f9ceee459688f7506d7d09bdf291760691.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 44.5,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                201,
+                                698,
+                                410,
+                                716.5
+                            ],
+                            "spans": [],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                201,
+                                716.5,
+                                410,
+                                735.0
+                            ],
+                            "spans": [],
+                            "index": 45
+                        }
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "image",
+                    "bbox": [
+                        190,
+                        78,
+                        405,
+                        246
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                190,
+                                78,
+                                405,
+                                246
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        190,
+                                        78,
+                                        405,
+                                        246
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                190,
+                                                78,
+                                                405,
+                                                246
+                                            ],
+                                            "score": 0.962,
+                                            "type": "image",
+                                            "image_path": "b76e3f5dce59ce270069488ab3ea5f08d6e69a5c74ce89a25358ac556f83a901.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 4.5,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        190,
+                                        78,
+                                        405,
+                                        94.8
+                                    ],
+                                    "spans": [],
+                                    "index": 0
+                                },
+                                {
+                                    "bbox": [
+                                        190,
+                                        94.8,
+                                        405,
+                                        111.6
+                                    ],
+                                    "spans": [],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        190,
+                                        111.6,
+                                        405,
+                                        128.4
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                },
+                                {
+                                    "bbox": [
+                                        190,
+                                        128.4,
+                                        405,
+                                        145.20000000000002
+                                    ],
+                                    "spans": [],
+                                    "index": 3
+                                },
+                                {
+                                    "bbox": [
+                                        190,
+                                        145.20000000000002,
+                                        405,
+                                        162.00000000000003
+                                    ],
+                                    "spans": [],
+                                    "index": 4
+                                },
+                                {
+                                    "bbox": [
+                                        190,
+                                        162.00000000000003,
+                                        405,
+                                        178.80000000000004
+                                    ],
+                                    "spans": [],
+                                    "index": 5
+                                },
+                                {
+                                    "bbox": [
+                                        190,
+                                        178.80000000000004,
+                                        405,
+                                        195.60000000000005
+                                    ],
+                                    "spans": [],
+                                    "index": 6
+                                },
+                                {
+                                    "bbox": [
+                                        190,
+                                        195.60000000000005,
+                                        405,
+                                        212.40000000000006
+                                    ],
+                                    "spans": [],
+                                    "index": 7
+                                },
+                                {
+                                    "bbox": [
+                                        190,
+                                        212.40000000000006,
+                                        405,
+                                        229.20000000000007
+                                    ],
+                                    "spans": [],
+                                    "index": 8
+                                },
+                                {
+                                    "bbox": [
+                                        190,
+                                        229.20000000000007,
+                                        405,
+                                        246.00000000000009
+                                    ],
+                                    "spans": [],
+                                    "index": 9
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                105,
+                                254,
+                                506,
+                                296
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        249,
+                                        509,
+                                        276
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                104,
+                                                253,
+                                                254,
+                                                272
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 1: The 2D convolution U(i,",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                230,
+                                                254,
+                                                271,
+                                                271
+                                            ],
+                                            "score": 0.93,
+                                            "content": "\\mathbf { U } _ { i , j } ^ { ( 0 ) } * \\mathbf { x } _ { j }",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                245,
+                                                249,
+                                                509,
+                                                276
+                                            ],
+                                            "score": 1.0,
+                                            "content": "0)j ∗ xj as the dot product W(0)i,j xj . The blank elements of W(0)i,j",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 10
+                                },
+                                {
+                                    "bbox": [
+                                        102,
+                                        268,
+                                        507,
+                                        289
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                102,
+                                                268,
+                                                239,
+                                                289
+                                            ],
+                                            "score": 1.0,
+                                            "content": "are zeros. The µth row of W(0)i,j",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                235,
+                                                271,
+                                                396,
+                                                288
+                                            ],
+                                            "score": 1.0,
+                                            "content": "corresponds to applying the filter to the",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                397,
+                                                274,
+                                                404,
+                                                284
+                                            ],
+                                            "score": 0.61,
+                                            "content": "\\mu",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                405,
+                                                271,
+                                                507,
+                                                288
+                                            ],
+                                            "score": 1.0,
+                                            "content": "th convolutional patch of",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 11
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        281,
+                                        170,
+                                        298
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                281,
+                                                154,
+                                                298
+                                            ],
+                                            "score": 1.0,
+                                            "content": "the channel",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                154,
+                                                285,
+                                                166,
+                                                296
+                                            ],
+                                            "score": 0.87,
+                                            "content": "\\mathbf { x } _ { j }",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                166,
+                                                281,
+                                                170,
+                                                298
+                                            ],
+                                            "score": 1.0,
+                                            "content": ".",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 12
+                                }
+                            ],
+                            "index": 11
+                        }
+                    ],
+                    "index": 7.75
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        317,
+                        504,
+                        344
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                315,
+                                506,
+                                332
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        315,
+                                        170,
+                                        332
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The activations",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        170,
+                                        318,
+                                        206,
+                                        331
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\mathbf { A } ^ { ( \\ell ) } ( \\mathbf { X } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        207,
+                                        315,
+                                        223,
+                                        332
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        223,
+                                        318,
+                                        300,
+                                        331
+                                    ],
+                                    "score": 0.93,
+                                    "content": "C ^ { ( \\ell ) } \\times ( H ^ { ( \\ell ) } D ^ { ( \\ell ) } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        300,
+                                        315,
+                                        381,
+                                        332
+                                    ],
+                                    "score": 1.0,
+                                    "content": "matrices. Each row",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        382,
+                                        317,
+                                        408,
+                                        332
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\mathbf { a } _ { i } ^ { ( \\ell + 1 ) }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        409,
+                                        315,
+                                        506,
+                                        332
+                                    ],
+                                    "score": 1.0,
+                                    "content": "represents the flattened",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                107,
+                                329,
+                                462,
+                                346
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        333,
+                                        112,
+                                        344
+                                    ],
+                                    "score": 0.7,
+                                    "content": "j",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        112,
+                                        329,
+                                        408,
+                                        346
+                                    ],
+                                    "score": 1.0,
+                                    "content": "th channel of the image that results from applying a convolutional filter to",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        408,
+                                        331,
+                                        457,
+                                        344
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\phi ( \\mathbf { A } ^ { ( \\ell ) } ( \\mathbf { X } ) )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        457,
+                                        329,
+                                        462,
+                                        346
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        }
+                    ],
+                    "index": 13.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        348,
+                        505,
+                        431
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                99,
+                                346,
+                                504,
+                                392
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        99,
+                                        348,
+                                        124,
+                                        392
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        102,
+                                        346,
+                                        329,
+                                        369
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The structure of the pseudo-weight matrices W(`+1)i,j",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        197,
+                                        348,
+                                        290,
+                                        392
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", depends on the archite",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        324,
+                                        348,
+                                        372,
+                                        392
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and biases or a convolu",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        372,
+                                        349,
+                                        397,
+                                        364
+                                    ],
+                                    "score": 0.92,
+                                    "content": "b _ { i } ^ { ( \\ell + 1 ) }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        398,
+                                        348,
+                                        418,
+                                        392
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", for layer,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        418,
+                                        351,
+                                        504,
+                                        364
+                                    ],
+                                    "score": 0.9,
+                                    "content": "i \\in \\{ 1 , \\ldots , C ^ { ( \\ell + 1 ) } \\}",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                124,
+                                365,
+                                503,
+                                381
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        124,
+                                        366,
+                                        197,
+                                        380
+                                    ],
+                                    "score": 0.93,
+                                    "content": "j \\in \\{ 1 , \\ldots , C ^ { ( \\ell ) } \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        471,
+                                        365,
+                                        503,
+                                        381
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\mathbf { W } _ { i , j } ^ { ( \\ell + 1 ) }",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                102,
+                                388,
+                                507,
+                                408
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        102,
+                                        388,
+                                        333,
+                                        408
+                                    ],
+                                    "score": 1.0,
+                                    "content": "xj represents applying the convolutional filter U(`+1)i,j",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        329,
+                                        389,
+                                        357,
+                                        406
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to the",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        358,
+                                        393,
+                                        363,
+                                        404
+                                    ],
+                                    "score": 0.73,
+                                    "content": "j",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        363,
+                                        389,
+                                        507,
+                                        406
+                                    ],
+                                    "score": 1.0,
+                                    "content": "th channel. Thus, the elements of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                403,
+                                503,
+                                431
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        403,
+                                        155,
+                                        431
+                                    ],
+                                    "score": 1.0,
+                                    "content": "each row of where it do",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        405,
+                                        192,
+                                        419
+                                    ],
+                                    "score": 1.0,
+                                    "content": "W(`+1)",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        189,
+                                        403,
+                                        474,
+                                        431
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are: 0 where the filter does not apply and the corresponding element of strated in Fig. 1.",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        474,
+                                        405,
+                                        503,
+                                        421
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\mathbf { U } _ { i , j } ^ { ( \\ell + 1 ) }",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 18
+                        }
+                    ],
+                    "index": 16.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        435,
+                        505,
+                        490
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                433,
+                                506,
+                                450
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        433,
+                                        315,
+                                        450
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The outputs of the network are the last activations,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        315,
+                                        435,
+                                        363,
+                                        448
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\mathbf { A } ^ { ( L + 1 ) } ( \\mathbf { X } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        363,
+                                        433,
+                                        506,
+                                        450
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". In the classification or regression",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                445,
+                                506,
+                                463
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        445,
+                                        338,
+                                        463
+                                    ],
+                                    "score": 1.0,
+                                    "content": "setting, the outputs are not spatially extended, so we have",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        339,
+                                        448,
+                                        434,
+                                        459
+                                    ],
+                                    "score": 0.91,
+                                    "content": "H ^ { ( L + 1 ) } = D ^ { ( L + 1 ) } = 1",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        434,
+                                        445,
+                                        506,
+                                        463
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", which is equiva-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                102,
+                                457,
+                                506,
+                                479
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        102,
+                                        457,
+                                        431,
+                                        479
+                                    ],
+                                    "score": 1.0,
+                                    "content": "lent to a fully-connected output layer. In this case, the pseudo-weights W(L+1)i,j",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        426,
+                                        460,
+                                        506,
+                                        476
+                                    ],
+                                    "score": 1.0,
+                                    "content": "only have one row,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                474,
+                                322,
+                                493
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        474,
+                                        183,
+                                        493
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and the activations",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        183,
+                                        475,
+                                        212,
+                                        490
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\mathbf { a } _ { i } ^ { ( L + 1 ) }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        212,
+                                        474,
+                                        322,
+                                        493
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are single-element vectors.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        }
+                    ],
+                    "index": 20.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        496,
+                        505,
+                        533
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                103,
+                                495,
+                                506,
+                                511
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        103,
+                                        495,
+                                        410,
+                                        511
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Finally, we define the prior distribution over functions by making the filters",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        411,
+                                        495,
+                                        430,
+                                        511
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\mathbf { U } _ { i , j } ^ { ( \\ell ) }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        431,
+                                        495,
+                                        476,
+                                        511
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and biases",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        477,
+                                        496,
+                                        491,
+                                        510
+                                    ],
+                                    "score": 0.92,
+                                    "content": "b _ { i } ^ { ( \\ell ) }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        492,
+                                        495,
+                                        506,
+                                        511
+                                    ],
+                                    "score": 1.0,
+                                    "content": "be",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                510,
+                                505,
+                                522
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        510,
+                                        358,
+                                        522
+                                    ],
+                                    "score": 1.0,
+                                    "content": "independent Gaussian random variables (RVs). For each layer",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        359,
+                                        511,
+                                        364,
+                                        520
+                                    ],
+                                    "score": 0.69,
+                                    "content": "\\ell",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        365,
+                                        510,
+                                        406,
+                                        522
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", channels",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        406,
+                                        511,
+                                        420,
+                                        522
+                                    ],
+                                    "score": 0.88,
+                                    "content": "i , j",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        420,
+                                        510,
+                                        505,
+                                        522
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and locations within",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                519,
+                                164,
+                                535
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        519,
+                                        142,
+                                        535
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the filter",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        142,
+                                        523,
+                                        159,
+                                        533
+                                    ],
+                                    "score": 0.85,
+                                    "content": "x , y",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        159,
+                                        519,
+                                        164,
+                                        535
+                                    ],
+                                    "score": 1.0,
+                                    "content": ":",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        }
+                    ],
+                    "index": 24
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        176,
+                        540,
+                        436,
+                        561
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                176,
+                                540,
+                                436,
+                                561
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        176,
+                                        540,
+                                        436,
+                                        561
+                                    ],
+                                    "score": 0.91,
+                                    "content": "U _ { i , j , x , y } ^ { ( \\ell ) } \\sim \\mathcal { N } \\left( 0 , \\sigma _ { w } ^ { 2 } / C ^ { ( \\ell ) } \\right) , \\qquad b _ { i } ^ { ( \\ell ) } \\sim \\mathcal { N } \\left( 0 , \\sigma _ { b } ^ { 2 } \\right) .",
+                                    "type": "interline_equation",
+                                    "image_path": "2e64ca37f686411e1352f20b7f11f8072dd889ca1dc3f9acbeb4054fe8bcca67.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 26,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                176,
+                                540,
+                                436,
+                                561
+                            ],
+                            "spans": [],
+                            "index": 26
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        568,
+                        505,
+                        603
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                568,
+                                506,
+                                582
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        568,
+                                        506,
+                                        582
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Note that, to keep the activation variance constant, the weight variance is divided by the number of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                580,
+                                505,
+                                592
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        580,
+                                        505,
+                                        592
+                                    ],
+                                    "score": 1.0,
+                                    "content": "input channels. The weight variance can also be divided by the number of elements of the filter,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                591,
+                                499,
+                                603
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        591,
+                                        499,
+                                        603
+                                    ],
+                                    "score": 1.0,
+                                    "content": "which makes it equivalent to the NN weight initialisation scheme introduced by He et al. (2016a).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        }
+                    ],
+                    "index": 28
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        618,
+                        271,
+                        630
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                618,
+                                272,
+                                631
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        618,
+                                        272,
+                                        631
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2.2 ARGUMENT FOR GP BEHAVIOUR",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        }
+                    ],
+                    "index": 30
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        640,
+                        505,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                641,
+                                505,
+                                653
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        641,
+                                        505,
+                                        653
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We follow the proofs by Lee et al. (2017) and Matthews et al. (2018a) to show that the output of the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                650,
+                                506,
+                                666
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        650,
+                                        264,
+                                        666
+                                    ],
+                                    "score": 1.0,
+                                    "content": "CNN described in the previous section,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        265,
+                                        651,
+                                        296,
+                                        663
+                                    ],
+                                    "score": 0.89,
+                                    "content": "\\mathbf { A } ^ { ( L + 1 ) }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        297,
+                                        650,
+                                        442,
+                                        666
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", defines a GP indexed by the inputs,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        443,
+                                        653,
+                                        453,
+                                        663
+                                    ],
+                                    "score": 0.49,
+                                    "content": "\\mathbf { X }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        453,
+                                        650,
+                                        506,
+                                        666
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Their proof",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                663,
+                                506,
+                                677
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        663,
+                                        506,
+                                        677
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(Lee et al., 2017) proceeds by applying the multivariate Central Limit Theorem (CLT) to each layer",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                674,
+                                506,
+                                689
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        674,
+                                        254,
+                                        689
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in sequence, i.e. taking the limit as",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        254,
+                                        675,
+                                        303,
+                                        686
+                                    ],
+                                    "score": 0.91,
+                                    "content": "N ^ { ( 1 ) }  \\infty",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        304,
+                                        674,
+                                        328,
+                                        689
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", then",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        329,
+                                        675,
+                                        378,
+                                        686
+                                    ],
+                                    "score": 0.91,
+                                    "content": "N ^ { ( 2 ) }  \\infty",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        378,
+                                        674,
+                                        424,
+                                        689
+                                    ],
+                                    "score": 1.0,
+                                    "content": "etc, where",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        424,
+                                        675,
+                                        444,
+                                        686
+                                    ],
+                                    "score": 0.92,
+                                    "content": "N ^ { ( \\ell ) }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        445,
+                                        674,
+                                        506,
+                                        689
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is the number",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                687,
+                                505,
+                                699
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        687,
+                                        205,
+                                        699
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of hidden units in layer",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        205,
+                                        687,
+                                        211,
+                                        696
+                                    ],
+                                    "score": 0.37,
+                                    "content": "\\ell",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        211,
+                                        687,
+                                        505,
+                                        699
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". By analogy, we sequentially apply the multivariate CLT by taking the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                696,
+                                506,
+                                713
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        696,
+                                        320,
+                                        713
+                                    ],
+                                    "score": 1.0,
+                                    "content": "limit as the number of channels goes to infinity, i.e.",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        321,
+                                        697,
+                                        368,
+                                        709
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\dot { C ^ { ( 1 ) } } \\stackrel { - } {  } \\infty",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        368,
+                                        696,
+                                        393,
+                                        713
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", then",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        393,
+                                        697,
+                                        440,
+                                        709
+                                    ],
+                                    "score": 0.93,
+                                    "content": "C ^ { ( 2 ) }  \\infty",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        441,
+                                        696,
+                                        506,
+                                        713
+                                    ],
+                                    "score": 1.0,
+                                    "content": "etc. While this",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                709,
+                                506,
+                                723
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        709,
+                                        506,
+                                        723
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is the simplest approach to taking the limits, other potentially more realistic approaches also exist",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                721,
+                                208,
+                                732
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        721,
+                                        208,
+                                        732
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(Matthews et al., 2018a).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        }
+                    ],
+                    "index": 34.5
+                }
+            ],
+            "page_idx": 2,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        107,
+                        27,
+                        293,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                26,
+                                294,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        26,
+                                        294,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Published as a conference paper at ICLR 2019",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        302,
+                        751,
+                        309,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                301,
+                                750,
+                                310,
+                                762
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        301,
+                                        750,
+                                        310,
+                                        762
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "image",
+                    "bbox": [
+                        190,
+                        78,
+                        405,
+                        246
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                190,
+                                78,
+                                405,
+                                246
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        190,
+                                        78,
+                                        405,
+                                        246
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                190,
+                                                78,
+                                                405,
+                                                246
+                                            ],
+                                            "score": 0.962,
+                                            "type": "image",
+                                            "image_path": "b76e3f5dce59ce270069488ab3ea5f08d6e69a5c74ce89a25358ac556f83a901.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 4.5,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        190,
+                                        78,
+                                        405,
+                                        94.8
+                                    ],
+                                    "spans": [],
+                                    "index": 0
+                                },
+                                {
+                                    "bbox": [
+                                        190,
+                                        94.8,
+                                        405,
+                                        111.6
+                                    ],
+                                    "spans": [],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        190,
+                                        111.6,
+                                        405,
+                                        128.4
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                },
+                                {
+                                    "bbox": [
+                                        190,
+                                        128.4,
+                                        405,
+                                        145.20000000000002
+                                    ],
+                                    "spans": [],
+                                    "index": 3
+                                },
+                                {
+                                    "bbox": [
+                                        190,
+                                        145.20000000000002,
+                                        405,
+                                        162.00000000000003
+                                    ],
+                                    "spans": [],
+                                    "index": 4
+                                },
+                                {
+                                    "bbox": [
+                                        190,
+                                        162.00000000000003,
+                                        405,
+                                        178.80000000000004
+                                    ],
+                                    "spans": [],
+                                    "index": 5
+                                },
+                                {
+                                    "bbox": [
+                                        190,
+                                        178.80000000000004,
+                                        405,
+                                        195.60000000000005
+                                    ],
+                                    "spans": [],
+                                    "index": 6
+                                },
+                                {
+                                    "bbox": [
+                                        190,
+                                        195.60000000000005,
+                                        405,
+                                        212.40000000000006
+                                    ],
+                                    "spans": [],
+                                    "index": 7
+                                },
+                                {
+                                    "bbox": [
+                                        190,
+                                        212.40000000000006,
+                                        405,
+                                        229.20000000000007
+                                    ],
+                                    "spans": [],
+                                    "index": 8
+                                },
+                                {
+                                    "bbox": [
+                                        190,
+                                        229.20000000000007,
+                                        405,
+                                        246.00000000000009
+                                    ],
+                                    "spans": [],
+                                    "index": 9
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                105,
+                                254,
+                                506,
+                                296
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        249,
+                                        509,
+                                        276
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                104,
+                                                253,
+                                                254,
+                                                272
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 1: The 2D convolution U(i,",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                230,
+                                                254,
+                                                271,
+                                                271
+                                            ],
+                                            "score": 0.93,
+                                            "content": "\\mathbf { U } _ { i , j } ^ { ( 0 ) } * \\mathbf { x } _ { j }",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                245,
+                                                249,
+                                                509,
+                                                276
+                                            ],
+                                            "score": 1.0,
+                                            "content": "0)j ∗ xj as the dot product W(0)i,j xj . The blank elements of W(0)i,j",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 10
+                                },
+                                {
+                                    "bbox": [
+                                        102,
+                                        268,
+                                        507,
+                                        289
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                102,
+                                                268,
+                                                239,
+                                                289
+                                            ],
+                                            "score": 1.0,
+                                            "content": "are zeros. The µth row of W(0)i,j",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                235,
+                                                271,
+                                                396,
+                                                288
+                                            ],
+                                            "score": 1.0,
+                                            "content": "corresponds to applying the filter to the",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                397,
+                                                274,
+                                                404,
+                                                284
+                                            ],
+                                            "score": 0.61,
+                                            "content": "\\mu",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                405,
+                                                271,
+                                                507,
+                                                288
+                                            ],
+                                            "score": 1.0,
+                                            "content": "th convolutional patch of",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 11
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        281,
+                                        170,
+                                        298
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                281,
+                                                154,
+                                                298
+                                            ],
+                                            "score": 1.0,
+                                            "content": "the channel",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                154,
+                                                285,
+                                                166,
+                                                296
+                                            ],
+                                            "score": 0.87,
+                                            "content": "\\mathbf { x } _ { j }",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                166,
+                                                281,
+                                                170,
+                                                298
+                                            ],
+                                            "score": 1.0,
+                                            "content": ".",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 12
+                                }
+                            ],
+                            "index": 11
+                        }
+                    ],
+                    "index": 7.75
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        317,
+                        504,
+                        344
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                315,
+                                506,
+                                332
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        315,
+                                        170,
+                                        332
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The activations",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        170,
+                                        318,
+                                        206,
+                                        331
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\mathbf { A } ^ { ( \\ell ) } ( \\mathbf { X } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        207,
+                                        315,
+                                        223,
+                                        332
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        223,
+                                        318,
+                                        300,
+                                        331
+                                    ],
+                                    "score": 0.93,
+                                    "content": "C ^ { ( \\ell ) } \\times ( H ^ { ( \\ell ) } D ^ { ( \\ell ) } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        300,
+                                        315,
+                                        381,
+                                        332
+                                    ],
+                                    "score": 1.0,
+                                    "content": "matrices. Each row",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        382,
+                                        317,
+                                        408,
+                                        332
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\mathbf { a } _ { i } ^ { ( \\ell + 1 ) }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        409,
+                                        315,
+                                        506,
+                                        332
+                                    ],
+                                    "score": 1.0,
+                                    "content": "represents the flattened",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                107,
+                                329,
+                                462,
+                                346
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        333,
+                                        112,
+                                        344
+                                    ],
+                                    "score": 0.7,
+                                    "content": "j",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        112,
+                                        329,
+                                        408,
+                                        346
+                                    ],
+                                    "score": 1.0,
+                                    "content": "th channel of the image that results from applying a convolutional filter to",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        408,
+                                        331,
+                                        457,
+                                        344
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\phi ( \\mathbf { A } ^ { ( \\ell ) } ( \\mathbf { X } ) )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        457,
+                                        329,
+                                        462,
+                                        346
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        }
+                    ],
+                    "index": 13.5,
+                    "bbox_fs": [
+                        106,
+                        315,
+                        506,
+                        346
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        348,
+                        505,
+                        431
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                99,
+                                346,
+                                504,
+                                392
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        99,
+                                        348,
+                                        124,
+                                        392
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        102,
+                                        346,
+                                        329,
+                                        369
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The structure of the pseudo-weight matrices W(`+1)i,j",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        197,
+                                        348,
+                                        290,
+                                        392
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", depends on the archite",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        324,
+                                        348,
+                                        372,
+                                        392
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and biases or a convolu",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        372,
+                                        349,
+                                        397,
+                                        364
+                                    ],
+                                    "score": 0.92,
+                                    "content": "b _ { i } ^ { ( \\ell + 1 ) }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        398,
+                                        348,
+                                        418,
+                                        392
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", for layer,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        418,
+                                        351,
+                                        504,
+                                        364
+                                    ],
+                                    "score": 0.9,
+                                    "content": "i \\in \\{ 1 , \\ldots , C ^ { ( \\ell + 1 ) } \\}",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                124,
+                                365,
+                                503,
+                                381
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        124,
+                                        366,
+                                        197,
+                                        380
+                                    ],
+                                    "score": 0.93,
+                                    "content": "j \\in \\{ 1 , \\ldots , C ^ { ( \\ell ) } \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        471,
+                                        365,
+                                        503,
+                                        381
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\mathbf { W } _ { i , j } ^ { ( \\ell + 1 ) }",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                102,
+                                388,
+                                507,
+                                408
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        102,
+                                        388,
+                                        333,
+                                        408
+                                    ],
+                                    "score": 1.0,
+                                    "content": "xj represents applying the convolutional filter U(`+1)i,j",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        329,
+                                        389,
+                                        357,
+                                        406
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to the",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        358,
+                                        393,
+                                        363,
+                                        404
+                                    ],
+                                    "score": 0.73,
+                                    "content": "j",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        363,
+                                        389,
+                                        507,
+                                        406
+                                    ],
+                                    "score": 1.0,
+                                    "content": "th channel. Thus, the elements of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                403,
+                                503,
+                                431
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        403,
+                                        155,
+                                        431
+                                    ],
+                                    "score": 1.0,
+                                    "content": "each row of where it do",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        405,
+                                        192,
+                                        419
+                                    ],
+                                    "score": 1.0,
+                                    "content": "W(`+1)",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        189,
+                                        403,
+                                        474,
+                                        431
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are: 0 where the filter does not apply and the corresponding element of strated in Fig. 1.",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        474,
+                                        405,
+                                        503,
+                                        421
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\mathbf { U } _ { i , j } ^ { ( \\ell + 1 ) }",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 18
+                        }
+                    ],
+                    "index": 16.5,
+                    "bbox_fs": [
+                        99,
+                        346,
+                        507,
+                        431
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        435,
+                        505,
+                        490
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                433,
+                                506,
+                                450
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        433,
+                                        315,
+                                        450
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The outputs of the network are the last activations,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        315,
+                                        435,
+                                        363,
+                                        448
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\mathbf { A } ^ { ( L + 1 ) } ( \\mathbf { X } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        363,
+                                        433,
+                                        506,
+                                        450
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". In the classification or regression",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                445,
+                                506,
+                                463
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        445,
+                                        338,
+                                        463
+                                    ],
+                                    "score": 1.0,
+                                    "content": "setting, the outputs are not spatially extended, so we have",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        339,
+                                        448,
+                                        434,
+                                        459
+                                    ],
+                                    "score": 0.91,
+                                    "content": "H ^ { ( L + 1 ) } = D ^ { ( L + 1 ) } = 1",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        434,
+                                        445,
+                                        506,
+                                        463
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", which is equiva-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                102,
+                                457,
+                                506,
+                                479
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        102,
+                                        457,
+                                        431,
+                                        479
+                                    ],
+                                    "score": 1.0,
+                                    "content": "lent to a fully-connected output layer. In this case, the pseudo-weights W(L+1)i,j",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        426,
+                                        460,
+                                        506,
+                                        476
+                                    ],
+                                    "score": 1.0,
+                                    "content": "only have one row,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                474,
+                                322,
+                                493
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        474,
+                                        183,
+                                        493
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and the activations",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        183,
+                                        475,
+                                        212,
+                                        490
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\mathbf { a } _ { i } ^ { ( L + 1 ) }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        212,
+                                        474,
+                                        322,
+                                        493
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are single-element vectors.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        }
+                    ],
+                    "index": 20.5,
+                    "bbox_fs": [
+                        102,
+                        433,
+                        506,
+                        493
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        496,
+                        505,
+                        533
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                103,
+                                495,
+                                506,
+                                511
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        103,
+                                        495,
+                                        410,
+                                        511
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Finally, we define the prior distribution over functions by making the filters",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        411,
+                                        495,
+                                        430,
+                                        511
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\mathbf { U } _ { i , j } ^ { ( \\ell ) }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        431,
+                                        495,
+                                        476,
+                                        511
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and biases",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        477,
+                                        496,
+                                        491,
+                                        510
+                                    ],
+                                    "score": 0.92,
+                                    "content": "b _ { i } ^ { ( \\ell ) }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        492,
+                                        495,
+                                        506,
+                                        511
+                                    ],
+                                    "score": 1.0,
+                                    "content": "be",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                510,
+                                505,
+                                522
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        510,
+                                        358,
+                                        522
+                                    ],
+                                    "score": 1.0,
+                                    "content": "independent Gaussian random variables (RVs). For each layer",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        359,
+                                        511,
+                                        364,
+                                        520
+                                    ],
+                                    "score": 0.69,
+                                    "content": "\\ell",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        365,
+                                        510,
+                                        406,
+                                        522
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", channels",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        406,
+                                        511,
+                                        420,
+                                        522
+                                    ],
+                                    "score": 0.88,
+                                    "content": "i , j",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        420,
+                                        510,
+                                        505,
+                                        522
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and locations within",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                519,
+                                164,
+                                535
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        519,
+                                        142,
+                                        535
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the filter",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        142,
+                                        523,
+                                        159,
+                                        533
+                                    ],
+                                    "score": 0.85,
+                                    "content": "x , y",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        159,
+                                        519,
+                                        164,
+                                        535
+                                    ],
+                                    "score": 1.0,
+                                    "content": ":",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        }
+                    ],
+                    "index": 24,
+                    "bbox_fs": [
+                        103,
+                        495,
+                        506,
+                        535
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        176,
+                        540,
+                        436,
+                        561
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                176,
+                                540,
+                                436,
+                                561
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        176,
+                                        540,
+                                        436,
+                                        561
+                                    ],
+                                    "score": 0.91,
+                                    "content": "U _ { i , j , x , y } ^ { ( \\ell ) } \\sim \\mathcal { N } \\left( 0 , \\sigma _ { w } ^ { 2 } / C ^ { ( \\ell ) } \\right) , \\qquad b _ { i } ^ { ( \\ell ) } \\sim \\mathcal { N } \\left( 0 , \\sigma _ { b } ^ { 2 } \\right) .",
+                                    "type": "interline_equation",
+                                    "image_path": "2e64ca37f686411e1352f20b7f11f8072dd889ca1dc3f9acbeb4054fe8bcca67.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 26,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                176,
+                                540,
+                                436,
+                                561
+                            ],
+                            "spans": [],
+                            "index": 26
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        568,
+                        505,
+                        603
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                568,
+                                506,
+                                582
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        568,
+                                        506,
+                                        582
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Note that, to keep the activation variance constant, the weight variance is divided by the number of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                580,
+                                505,
+                                592
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        580,
+                                        505,
+                                        592
+                                    ],
+                                    "score": 1.0,
+                                    "content": "input channels. The weight variance can also be divided by the number of elements of the filter,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                591,
+                                499,
+                                603
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        591,
+                                        499,
+                                        603
+                                    ],
+                                    "score": 1.0,
+                                    "content": "which makes it equivalent to the NN weight initialisation scheme introduced by He et al. (2016a).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        }
+                    ],
+                    "index": 28,
+                    "bbox_fs": [
+                        105,
+                        568,
+                        506,
+                        603
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        618,
+                        271,
+                        630
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                618,
+                                272,
+                                631
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        618,
+                                        272,
+                                        631
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2.2 ARGUMENT FOR GP BEHAVIOUR",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        }
+                    ],
+                    "index": 30
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        640,
+                        505,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                641,
+                                505,
+                                653
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        641,
+                                        505,
+                                        653
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We follow the proofs by Lee et al. (2017) and Matthews et al. (2018a) to show that the output of the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                650,
+                                506,
+                                666
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        650,
+                                        264,
+                                        666
+                                    ],
+                                    "score": 1.0,
+                                    "content": "CNN described in the previous section,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        265,
+                                        651,
+                                        296,
+                                        663
+                                    ],
+                                    "score": 0.89,
+                                    "content": "\\mathbf { A } ^ { ( L + 1 ) }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        297,
+                                        650,
+                                        442,
+                                        666
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", defines a GP indexed by the inputs,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        443,
+                                        653,
+                                        453,
+                                        663
+                                    ],
+                                    "score": 0.49,
+                                    "content": "\\mathbf { X }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        453,
+                                        650,
+                                        506,
+                                        666
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Their proof",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                663,
+                                506,
+                                677
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        663,
+                                        506,
+                                        677
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(Lee et al., 2017) proceeds by applying the multivariate Central Limit Theorem (CLT) to each layer",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                674,
+                                506,
+                                689
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        674,
+                                        254,
+                                        689
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in sequence, i.e. taking the limit as",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        254,
+                                        675,
+                                        303,
+                                        686
+                                    ],
+                                    "score": 0.91,
+                                    "content": "N ^ { ( 1 ) }  \\infty",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        304,
+                                        674,
+                                        328,
+                                        689
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", then",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        329,
+                                        675,
+                                        378,
+                                        686
+                                    ],
+                                    "score": 0.91,
+                                    "content": "N ^ { ( 2 ) }  \\infty",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        378,
+                                        674,
+                                        424,
+                                        689
+                                    ],
+                                    "score": 1.0,
+                                    "content": "etc, where",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        424,
+                                        675,
+                                        444,
+                                        686
+                                    ],
+                                    "score": 0.92,
+                                    "content": "N ^ { ( \\ell ) }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        445,
+                                        674,
+                                        506,
+                                        689
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is the number",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                687,
+                                505,
+                                699
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        687,
+                                        205,
+                                        699
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of hidden units in layer",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        205,
+                                        687,
+                                        211,
+                                        696
+                                    ],
+                                    "score": 0.37,
+                                    "content": "\\ell",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        211,
+                                        687,
+                                        505,
+                                        699
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". By analogy, we sequentially apply the multivariate CLT by taking the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                696,
+                                506,
+                                713
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        696,
+                                        320,
+                                        713
+                                    ],
+                                    "score": 1.0,
+                                    "content": "limit as the number of channels goes to infinity, i.e.",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        321,
+                                        697,
+                                        368,
+                                        709
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\dot { C ^ { ( 1 ) } } \\stackrel { - } {  } \\infty",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        368,
+                                        696,
+                                        393,
+                                        713
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", then",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        393,
+                                        697,
+                                        440,
+                                        709
+                                    ],
+                                    "score": 0.93,
+                                    "content": "C ^ { ( 2 ) }  \\infty",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        441,
+                                        696,
+                                        506,
+                                        713
+                                    ],
+                                    "score": 1.0,
+                                    "content": "etc. While this",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                709,
+                                506,
+                                723
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        709,
+                                        506,
+                                        723
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is the simplest approach to taking the limits, other potentially more realistic approaches also exist",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                721,
+                                208,
+                                732
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        721,
+                                        208,
+                                        732
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(Matthews et al., 2018a).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        }
+                    ],
+                    "index": 34.5,
+                    "bbox_fs": [
+                        104,
+                        641,
+                        506,
+                        732
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        81,
+                        505,
+                        109
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                80,
+                                505,
+                                96
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        80,
+                                        505,
+                                        96
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The fundamental quantity we consider is a vector formed by concatenating the feature maps (or",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                93,
+                                403,
+                                110
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        93,
+                                        200,
+                                        110
+                                    ],
+                                    "score": 1.0,
+                                    "content": "equivalently channels),",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        201,
+                                        93,
+                                        234,
+                                        110
+                                    ],
+                                    "score": 0.93,
+                                    "content": "{ \\bf a } _ { j } ^ { ( \\ell ) } ( { \\bf X } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        234,
+                                        93,
+                                        252,
+                                        110
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        252,
+                                        93,
+                                        288,
+                                        110
+                                    ],
+                                    "score": 0.94,
+                                    "content": "\\mathbf { a } _ { j } ^ { ( \\ell ) } ( \\mathbf { X } ^ { \\prime } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        289,
+                                        93,
+                                        357,
+                                        110
+                                    ],
+                                    "score": 1.0,
+                                    "content": "from data points",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        358,
+                                        96,
+                                        367,
+                                        106
+                                    ],
+                                    "score": 0.58,
+                                    "content": "\\mathbf { X }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        368,
+                                        93,
+                                        385,
+                                        110
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        385,
+                                        96,
+                                        398,
+                                        106
+                                    ],
+                                    "score": 0.85,
+                                    "content": "\\mathbf { X } ^ { \\prime }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        398,
+                                        93,
+                                        403,
+                                        110
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        }
+                    ],
+                    "index": 0.5
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        246,
+                        115,
+                        365,
+                        149
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                246,
+                                115,
+                                365,
+                                149
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        246,
+                                        115,
+                                        365,
+                                        149
+                                    ],
+                                    "score": 0.95,
+                                    "content": "\\begin{array} { r } { { \\bf a } _ { i } ^ { ( \\ell ) } ( { \\bf X } , { \\bf X } ^ { \\prime } ) = \\left( { \\bf a } _ { i } ^ { ( \\ell ) } ( { \\bf X } ) \\right) . } \\end{array}",
+                                    "type": "interline_equation",
+                                    "image_path": "d9800bcb592548e273050ba34b1727b8824d0e237f31f59f087f18e26c388d11.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 2,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                246,
+                                115,
+                                365,
+                                149
+                            ],
+                            "spans": [],
+                            "index": 2
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        154,
+                        504,
+                        177
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                153,
+                                506,
+                                168
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        153,
+                                        506,
+                                        168
+                                    ],
+                                    "score": 1.0,
+                                    "content": "This quantity (and the following arguments) can all be extended to the case of countably finite",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                165,
+                                207,
+                                178
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        165,
+                                        207,
+                                        178
+                                    ],
+                                    "score": 1.0,
+                                    "content": "numbers of input points.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        }
+                    ],
+                    "index": 3.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        189,
+                        505,
+                        247
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                188,
+                                505,
+                                203
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        188,
+                                        319,
+                                        203
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Induction base case. For any pair of data points,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        319,
+                                        190,
+                                        329,
+                                        200
+                                    ],
+                                    "score": 0.61,
+                                    "content": "\\mathbf { X }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        330,
+                                        188,
+                                        348,
+                                        203
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        349,
+                                        190,
+                                        362,
+                                        200
+                                    ],
+                                    "score": 0.86,
+                                    "content": "\\mathbf { X } ^ { \\prime }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        362,
+                                        188,
+                                        505,
+                                        203
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the feature-maps corresponding to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                199,
+                                506,
+                                218
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        199,
+                                        121,
+                                        218
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        122,
+                                        204,
+                                        127,
+                                        215
+                                    ],
+                                    "score": 0.75,
+                                    "content": "j",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        128,
+                                        199,
+                                        173,
+                                        218
+                                    ],
+                                    "score": 1.0,
+                                    "content": "th channel,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        174,
+                                        201,
+                                        223,
+                                        217
+                                    ],
+                                    "score": 0.94,
+                                    "content": "\\mathbf { a } _ { j } ^ { ( 1 ) } ( \\mathbf { X } , \\mathbf { X } ^ { \\prime } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        224,
+                                        199,
+                                        506,
+                                        218
+                                    ],
+                                    "score": 1.0,
+                                    "content": "have a multivariate Gaussian joint distribution. This is because each",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                216,
+                                505,
+                                233
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        217,
+                                        427,
+                                        233
+                                    ],
+                                    "score": 1.0,
+                                    "content": "element is a linear combination of shared Gaussian random variables: the biases,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        427,
+                                        216,
+                                        445,
+                                        233
+                                    ],
+                                    "score": 0.91,
+                                    "content": "{ \\bf b } _ { j } ^ { ( 0 ) }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        446,
+                                        217,
+                                        505,
+                                        233
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and the filters,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                229,
+                                209,
+                                250
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        231,
+                                        128,
+                                        248
+                                    ],
+                                    "score": 0.88,
+                                    "content": "\\mathbf { U } _ { j , : } ^ { ( 0 ) }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        128,
+                                        229,
+                                        209,
+                                        250
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Following Eq. (1),",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        }
+                    ],
+                    "index": 6.5
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        197,
+                        254,
+                        413,
+                        291
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                197,
+                                254,
+                                413,
+                                291
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        197,
+                                        254,
+                                        413,
+                                        291
+                                    ],
+                                    "score": 0.94,
+                                    "content": "\\mathbf { a } _ { i } ^ { ( 1 ) } ( \\mathbf { X } , \\mathbf { X } ^ { \\prime } ) = b _ { i } ^ { ( 1 ) } \\mathbf { 1 } + \\sum _ { i = 1 } ^ { C ^ { ( 0 ) } } \\left( \\begin{array} { c c } { \\mathbf { W } _ { i , j } ^ { ( 1 ) } } & { \\mathbf { 0 } } \\\\ { \\mathbf { 0 } } & { \\mathbf { W } _ { i , j } ^ { ( 1 ) } } \\end{array} \\right) \\left( \\begin{array} { c } { \\mathbf { x } _ { i } } \\\\ { \\mathbf { x } _ { i } ^ { \\prime } } \\end{array} \\right) ,",
+                                    "type": "interline_equation",
+                                    "image_path": "7c1f702cca405e1c3cc5e8b9a4e3e3432322df0dbd23cd4a844fff3c942657a6.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 9.5,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                197,
+                                254,
+                                413,
+                                272.5
+                            ],
+                            "spans": [],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                197,
+                                272.5,
+                                413,
+                                291.0
+                            ],
+                            "spans": [],
+                            "index": 10
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        295,
+                        505,
+                        348
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                295,
+                                505,
+                                309
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        295,
+                                        505,
+                                        309
+                                    ],
+                                    "score": 1.0,
+                                    "content": "where 1 is a vector of all-ones. While the elements within a feature map display strong correlations,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                306,
+                                505,
+                                320
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        306,
+                                        505,
+                                        320
+                                    ],
+                                    "score": 1.0,
+                                    "content": "different feature maps are independent and identically distributed (iid) conditioned on the data (i.e.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                315,
+                                505,
+                                336
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        318,
+                                        156,
+                                        333
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\mathbf { a } _ { i } ^ { ( 1 ) } ( \\mathbf { X } , \\mathbf { X } ^ { \\prime } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        156,
+                                        315,
+                                        175,
+                                        336
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        176,
+                                        318,
+                                        225,
+                                        333
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\mathbf { a } _ { i ^ { \\prime } } ^ { ( \\mathrm { i } ) } ( \\mathbf { X } , \\mathbf { X ^ { \\prime } } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        225,
+                                        315,
+                                        270,
+                                        336
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are iid for",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        270,
+                                        320,
+                                        295,
+                                        332
+                                    ],
+                                    "score": 0.92,
+                                    "content": "i \\neq i ^ { \\prime }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        296,
+                                        315,
+                                        505,
+                                        336
+                                    ],
+                                    "score": 1.0,
+                                    "content": "), because the parameters for different feature-maps",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                331,
+                                352,
+                                349
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        331,
+                                        169,
+                                        349
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(i.e. the biases,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        169,
+                                        333,
+                                        185,
+                                        347
+                                    ],
+                                    "score": 0.91,
+                                    "content": "b _ { i } ^ { ( 1 ) }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        185,
+                                        331,
+                                        245,
+                                        349
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and the filters,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        246,
+                                        332,
+                                        270,
+                                        348
+                                    ],
+                                    "score": 0.86,
+                                    "content": "\\mathbf { W } _ { i , : } ^ { ( 1 ) }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        270,
+                                        331,
+                                        352,
+                                        349
+                                    ],
+                                    "score": 1.0,
+                                    "content": ") are themselves iid.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        }
+                    ],
+                    "index": 12.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        361,
+                        506,
+                        430
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                361,
+                                506,
+                                378
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        361,
+                                        361,
+                                        377
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Induction step. Consider the feature maps at the `th layer,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        362,
+                                        361,
+                                        410,
+                                        378
+                                    ],
+                                    "score": 0.94,
+                                    "content": "\\mathbf { a } _ { j } ^ { ( \\ell ) } ( \\mathbf { X } , \\mathbf { X } ^ { \\prime } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        411,
+                                        361,
+                                        506,
+                                        377
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", to be iid multivariate",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                375,
+                                506,
+                                394
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        375,
+                                        197,
+                                        394
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Gaussian RVs (i.e. for",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        198,
+                                        378,
+                                        223,
+                                        392
+                                    ],
+                                    "score": 0.79,
+                                    "content": "j \\neq j ^ { \\prime }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        224,
+                                        375,
+                                        227,
+                                        394
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        227,
+                                        376,
+                                        276,
+                                        393
+                                    ],
+                                    "score": 0.79,
+                                    "content": "\\mathbf { a } _ { j } ^ { ( \\ell ) } ( \\mathbf { X } , \\mathbf { X } ^ { \\prime } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        276,
+                                        375,
+                                        294,
+                                        394
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        294,
+                                        376,
+                                        343,
+                                        393
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\mathbf { a } _ { j ^ { \\prime } } ^ { ( \\ell ) } ( \\mathbf { X } , \\mathbf { X } ^ { \\prime } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        343,
+                                        375,
+                                        506,
+                                        394
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are iid). Our goal is to show that, taking",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                390,
+                                506,
+                                407
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        390,
+                                        236,
+                                        407
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the number of channels at layer",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        236,
+                                        395,
+                                        242,
+                                        404
+                                    ],
+                                    "score": 0.8,
+                                    "content": "\\ell",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        243,
+                                        390,
+                                        305,
+                                        407
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to infinity (i.e.",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        306,
+                                        393,
+                                        352,
+                                        405
+                                    ],
+                                    "score": 0.86,
+                                    "content": "C ^ { ( \\ell ) } \\to \\infty ^ { \\cdot }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        352,
+                                        390,
+                                        506,
+                                        407
+                                    ],
+                                    "score": 1.0,
+                                    "content": "), the same properties hold at the next",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                403,
+                                507,
+                                422
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        405,
+                                        219,
+                                        421
+                                    ],
+                                    "score": 1.0,
+                                    "content": "layer (i.e. all feature maps,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        221,
+                                        403,
+                                        507,
+                                        422
+                                    ],
+                                    "score": 1.0,
+                                    "content": "a(`+1)i (X, X0), are iid multivariate Gaussian RVs). Writing eq. (2) for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                418,
+                                290,
+                                431
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        418,
+                                        200,
+                                        431
+                                    ],
+                                    "score": 1.0,
+                                    "content": "two training examples,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        200,
+                                        419,
+                                        209,
+                                        428
+                                    ],
+                                    "score": 0.66,
+                                    "content": "\\mathbf { X }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        210,
+                                        418,
+                                        228,
+                                        431
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        228,
+                                        419,
+                                        240,
+                                        428
+                                    ],
+                                    "score": 0.82,
+                                    "content": "\\mathbf { X } ^ { \\prime }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        240,
+                                        418,
+                                        290,
+                                        431
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", we obtain,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        }
+                    ],
+                    "index": 17
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        162,
+                        436,
+                        449,
+                        473
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                162,
+                                436,
+                                449,
+                                473
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        162,
+                                        436,
+                                        449,
+                                        473
+                                    ],
+                                    "score": 0.95,
+                                    "content": "\\mathbf { a } _ { i } ^ { ( \\ell + 1 ) } ( \\mathbf { X } , \\mathbf { X } ^ { \\prime } ) = b _ { i } ^ { ( \\ell + 1 ) } \\mathbf { 1 } + \\sum _ { j = 1 } ^ { C ^ { ( \\ell ) } } \\left( \\begin{array} { c c } { \\mathbf { W } _ { i , j } ^ { ( \\ell + 1 ) } } & { \\mathbf { 0 } } \\\\ { \\mathbf { 0 } } & { \\mathbf { W } _ { i , j } ^ { ( \\ell + 1 ) } } \\end{array} \\right) \\phi ( \\mathbf { a } _ { j } ^ { ( \\ell ) } ( \\mathbf { X } , \\mathbf { X } ^ { \\prime } ) )",
+                                    "type": "interline_equation",
+                                    "image_path": "3cc3207e1f6d7550a22043ec02cfccb2c0052e0bfb6d7fedaae056ea3c1ef617.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 21,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                162,
+                                436,
+                                449,
+                                448.3333333333333
+                            ],
+                            "spans": [],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                162,
+                                448.3333333333333,
+                                449,
+                                460.66666666666663
+                            ],
+                            "spans": [],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                162,
+                                460.66666666666663,
+                                449,
+                                472.99999999999994
+                            ],
+                            "spans": [],
+                            "index": 22
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        480,
+                        506,
+                        558
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                477,
+                                508,
+                                498
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        477,
+                                        218,
+                                        498
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We begin by showing that",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        218,
+                                        480,
+                                        277,
+                                        495
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\mathbf { a } _ { i } ^ { ( \\ell + 1 ) } ( \\mathbf { X } , \\mathbf { X } ^ { \\prime } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        277,
+                                        477,
+                                        508,
+                                        498
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is a multivariate Gaussian RV. The first term is multi-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                491,
+                                507,
+                                511
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        491,
+                                        287,
+                                        511
+                                    ],
+                                    "score": 1.0,
+                                    "content": "variate Gaussian, as it is a linear function of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        287,
+                                        494,
+                                        312,
+                                        509
+                                    ],
+                                    "score": 0.93,
+                                    "content": "b _ { i } ^ { ( \\ell + 1 ) }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        313,
+                                        491,
+                                        507,
+                                        511
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", which is itself iid Gaussian. We can apply the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                505,
+                                508,
+                                524
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        505,
+                                        457,
+                                        524
+                                    ],
+                                    "score": 1.0,
+                                    "content": "multivariate CLT to show that the second term is also Gaussian, because, in the limit as",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        457,
+                                        508,
+                                        501,
+                                        519
+                                    ],
+                                    "score": 0.9,
+                                    "content": "C ^ { ( \\ell ) } \\to \\infty",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        502,
+                                        505,
+                                        508,
+                                        524
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                102,
+                                519,
+                                506,
+                                537
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        102,
+                                        519,
+                                        279,
+                                        537
+                                    ],
+                                    "score": 1.0,
+                                    "content": "it is the sum of infinitely many iid terms:",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        279,
+                                        520,
+                                        328,
+                                        536
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\mathbf { a } _ { j } ^ { ( \\ell ) } ( \\mathbf { X } , \\mathbf { X } ^ { \\prime } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        329,
+                                        519,
+                                        441,
+                                        537
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are iid by assumption, and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        441,
+                                        520,
+                                        474,
+                                        536
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\mathbf { W } _ { i , j } ^ { ( \\ell + 1 ) }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        475,
+                                        519,
+                                        506,
+                                        537
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are iid",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                532,
+                                507,
+                                546
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        532,
+                                        507,
+                                        546
+                                    ],
+                                    "score": 1.0,
+                                    "content": "by definition. Note that the same argument applies to all feature maps jointly, so all elements of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                107,
+                                543,
+                                432,
+                                559
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        544,
+                                        168,
+                                        558
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\mathbf { A } ^ { ( \\ell + 1 ) } ( \\mathbf { X } , \\mathbf { X } ^ { \\prime } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        169,
+                                        543,
+                                        432,
+                                        559
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(defined by analogy with eq. 4) are jointly multivariate Gaussian.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        }
+                    ],
+                    "index": 25.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        561,
+                        505,
+                        668
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                561,
+                                505,
+                                576
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        561,
+                                        505,
+                                        576
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Following Lee et al. (2017), to complete the argument, we need to show that the output feature maps",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                571,
+                                506,
+                                590
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        571,
+                                        155,
+                                        589
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are iid, i.e.",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        156,
+                                        573,
+                                        215,
+                                        588
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\mathbf { a } _ { i } ^ { ( \\ell + 1 ) } ( \\mathbf { X } , \\mathbf { X } ^ { \\prime } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        180,
+                                        572,
+                                        250,
+                                        590
+                                    ],
+                                    "score": 1.0,
+                                    "content": ")(X, X0) and a(`i0",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        235,
+                                        573,
+                                        294,
+                                        588
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\mathbf { a } _ { i ^ { \\prime } } ^ { ( \\ell + 1 ) } ( \\mathbf { X } , \\mathbf { X } ^ { \\prime } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        294,
+                                        571,
+                                        339,
+                                        590
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are iid for",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        340,
+                                        575,
+                                        366,
+                                        588
+                                    ],
+                                    "score": 0.91,
+                                    "content": "i \\neq i ^ { \\prime }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        367,
+                                        571,
+                                        506,
+                                        590
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". They are identically distributed,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                583,
+                                509,
+                                622
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        585,
+                                        118,
+                                        622
+                                    ],
+                                    "score": 1.0,
+                                    "content": "re",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        110,
+                                        583,
+                                        151,
+                                        603
+                                    ],
+                                    "score": 1.0,
+                                    "content": "as b(`+1) a",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        144,
+                                        585,
+                                        164,
+                                        622
+                                    ],
+                                    "score": 1.0,
+                                    "content": "nd r tha",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        164,
+                                        588,
+                                        198,
+                                        604
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\mathbf { W } _ { i , j } ^ { ( \\ell + 1 ) }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        228,
+                                        585,
+                                        248,
+                                        622
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and  and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        249,
+                                        588,
+                                        309,
+                                        603
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\phi ( \\mathbf { a } _ { j } ^ { ( \\ell ) } ( \\mathbf { X } , \\mathbf { X } ^ { \\prime } ) )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        310,
+                                        585,
+                                        509,
+                                        622
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is shared. To show that they are independent,are jointly Gaussian, so it is sufficient to show",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                169,
+                                603,
+                                308,
+                                618
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        169,
+                                        603,
+                                        228,
+                                        618
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\mathbf { a } _ { i } ^ { ( \\ell + 1 ) } ( \\mathbf { X } , \\mathbf { X } ^ { \\prime } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        249,
+                                        604,
+                                        308,
+                                        618
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\mathbf { a } _ { i ^ { \\prime } } ^ { ( \\ell + 1 ) } ( \\mathbf { X } , \\mathbf { X } ^ { \\prime } )",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                102,
+                                617,
+                                503,
+                                634
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        102,
+                                        617,
+                                        470,
+                                        634
+                                    ],
+                                    "score": 1.0,
+                                    "content": "that they are uncorrelated, and we can show that they are uncorrelated because the weights,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        470,
+                                        617,
+                                        503,
+                                        634
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\mathbf { W } _ { i , j } ^ { ( \\ell + 1 ) }",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                632,
+                                505,
+                                643
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        632,
+                                        505,
+                                        643
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are independent with zero-mean, eliminating any correlations that might arise through the shared",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                642,
+                                506,
+                                658
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        642,
+                                        126,
+                                        658
+                                    ],
+                                    "score": 1.0,
+                                    "content": "RV,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        126,
+                                        642,
+                                        187,
+                                        658
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\phi ( \\mathbf { a } _ { j } ^ { ( \\ell ) } ( \\mathbf { X } , \\mathbf { X } ^ { \\prime } ) )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        188,
+                                        642,
+                                        506,
+                                        658
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". In the appendix, we consider the more complex case where we take limits",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                654,
+                                171,
+                                668
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        654,
+                                        171,
+                                        668
+                                    ],
+                                    "score": 1.0,
+                                    "content": "simultaneously.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        }
+                    ],
+                    "index": 32.5
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        684,
+                        326,
+                        697
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                682,
+                                327,
+                                699
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        682,
+                                        327,
+                                        699
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3 THE CONVNET AND RESNET KERNELS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        }
+                    ],
+                    "index": 37
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        709,
+                        504,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                710,
+                                504,
+                                722
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        710,
+                                        504,
+                                        722
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Here we derive a computationally efficient kernel corresponding to the CNN described in the pre-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                719,
+                                506,
+                                734
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        719,
+                                        506,
+                                        734
+                                    ],
+                                    "score": 1.0,
+                                    "content": "vious section. It is surprising that we can compute the kernel efficiently because the feature maps,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        }
+                    ],
+                    "index": 38.5
+                }
+            ],
+            "page_idx": 3,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        107,
+                        26,
+                        293,
+                        38
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                25,
+                                294,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        25,
+                                        294,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Published as a conference paper at ICLR 2019",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        302,
+                        751,
+                        309,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                301,
+                                750,
+                                310,
+                                762
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        301,
+                                        750,
+                                        310,
+                                        762
+                                    ],
+                                    "score": 1.0,
+                                    "content": "4",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        81,
+                        505,
+                        109
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                80,
+                                505,
+                                96
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        80,
+                                        505,
+                                        96
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The fundamental quantity we consider is a vector formed by concatenating the feature maps (or",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                93,
+                                403,
+                                110
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        93,
+                                        200,
+                                        110
+                                    ],
+                                    "score": 1.0,
+                                    "content": "equivalently channels),",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        201,
+                                        93,
+                                        234,
+                                        110
+                                    ],
+                                    "score": 0.93,
+                                    "content": "{ \\bf a } _ { j } ^ { ( \\ell ) } ( { \\bf X } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        234,
+                                        93,
+                                        252,
+                                        110
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        252,
+                                        93,
+                                        288,
+                                        110
+                                    ],
+                                    "score": 0.94,
+                                    "content": "\\mathbf { a } _ { j } ^ { ( \\ell ) } ( \\mathbf { X } ^ { \\prime } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        289,
+                                        93,
+                                        357,
+                                        110
+                                    ],
+                                    "score": 1.0,
+                                    "content": "from data points",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        358,
+                                        96,
+                                        367,
+                                        106
+                                    ],
+                                    "score": 0.58,
+                                    "content": "\\mathbf { X }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        368,
+                                        93,
+                                        385,
+                                        110
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        385,
+                                        96,
+                                        398,
+                                        106
+                                    ],
+                                    "score": 0.85,
+                                    "content": "\\mathbf { X } ^ { \\prime }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        398,
+                                        93,
+                                        403,
+                                        110
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        }
+                    ],
+                    "index": 0.5,
+                    "bbox_fs": [
+                        105,
+                        80,
+                        505,
+                        110
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        246,
+                        115,
+                        365,
+                        149
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                246,
+                                115,
+                                365,
+                                149
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        246,
+                                        115,
+                                        365,
+                                        149
+                                    ],
+                                    "score": 0.95,
+                                    "content": "\\begin{array} { r } { { \\bf a } _ { i } ^ { ( \\ell ) } ( { \\bf X } , { \\bf X } ^ { \\prime } ) = \\left( { \\bf a } _ { i } ^ { ( \\ell ) } ( { \\bf X } ) \\right) . } \\end{array}",
+                                    "type": "interline_equation",
+                                    "image_path": "d9800bcb592548e273050ba34b1727b8824d0e237f31f59f087f18e26c388d11.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 2,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                246,
+                                115,
+                                365,
+                                149
+                            ],
+                            "spans": [],
+                            "index": 2
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        154,
+                        504,
+                        177
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                153,
+                                506,
+                                168
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        153,
+                                        506,
+                                        168
+                                    ],
+                                    "score": 1.0,
+                                    "content": "This quantity (and the following arguments) can all be extended to the case of countably finite",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                165,
+                                207,
+                                178
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        165,
+                                        207,
+                                        178
+                                    ],
+                                    "score": 1.0,
+                                    "content": "numbers of input points.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        }
+                    ],
+                    "index": 3.5,
+                    "bbox_fs": [
+                        105,
+                        153,
+                        506,
+                        178
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        189,
+                        505,
+                        247
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                188,
+                                505,
+                                203
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        188,
+                                        319,
+                                        203
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Induction base case. For any pair of data points,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        319,
+                                        190,
+                                        329,
+                                        200
+                                    ],
+                                    "score": 0.61,
+                                    "content": "\\mathbf { X }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        330,
+                                        188,
+                                        348,
+                                        203
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        349,
+                                        190,
+                                        362,
+                                        200
+                                    ],
+                                    "score": 0.86,
+                                    "content": "\\mathbf { X } ^ { \\prime }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        362,
+                                        188,
+                                        505,
+                                        203
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the feature-maps corresponding to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                199,
+                                506,
+                                218
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        199,
+                                        121,
+                                        218
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        122,
+                                        204,
+                                        127,
+                                        215
+                                    ],
+                                    "score": 0.75,
+                                    "content": "j",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        128,
+                                        199,
+                                        173,
+                                        218
+                                    ],
+                                    "score": 1.0,
+                                    "content": "th channel,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        174,
+                                        201,
+                                        223,
+                                        217
+                                    ],
+                                    "score": 0.94,
+                                    "content": "\\mathbf { a } _ { j } ^ { ( 1 ) } ( \\mathbf { X } , \\mathbf { X } ^ { \\prime } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        224,
+                                        199,
+                                        506,
+                                        218
+                                    ],
+                                    "score": 1.0,
+                                    "content": "have a multivariate Gaussian joint distribution. This is because each",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                216,
+                                505,
+                                233
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        217,
+                                        427,
+                                        233
+                                    ],
+                                    "score": 1.0,
+                                    "content": "element is a linear combination of shared Gaussian random variables: the biases,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        427,
+                                        216,
+                                        445,
+                                        233
+                                    ],
+                                    "score": 0.91,
+                                    "content": "{ \\bf b } _ { j } ^ { ( 0 ) }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        446,
+                                        217,
+                                        505,
+                                        233
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and the filters,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                229,
+                                209,
+                                250
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        231,
+                                        128,
+                                        248
+                                    ],
+                                    "score": 0.88,
+                                    "content": "\\mathbf { U } _ { j , : } ^ { ( 0 ) }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        128,
+                                        229,
+                                        209,
+                                        250
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Following Eq. (1),",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        }
+                    ],
+                    "index": 6.5,
+                    "bbox_fs": [
+                        104,
+                        188,
+                        506,
+                        250
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        197,
+                        254,
+                        413,
+                        291
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                197,
+                                254,
+                                413,
+                                291
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        197,
+                                        254,
+                                        413,
+                                        291
+                                    ],
+                                    "score": 0.94,
+                                    "content": "\\mathbf { a } _ { i } ^ { ( 1 ) } ( \\mathbf { X } , \\mathbf { X } ^ { \\prime } ) = b _ { i } ^ { ( 1 ) } \\mathbf { 1 } + \\sum _ { i = 1 } ^ { C ^ { ( 0 ) } } \\left( \\begin{array} { c c } { \\mathbf { W } _ { i , j } ^ { ( 1 ) } } & { \\mathbf { 0 } } \\\\ { \\mathbf { 0 } } & { \\mathbf { W } _ { i , j } ^ { ( 1 ) } } \\end{array} \\right) \\left( \\begin{array} { c } { \\mathbf { x } _ { i } } \\\\ { \\mathbf { x } _ { i } ^ { \\prime } } \\end{array} \\right) ,",
+                                    "type": "interline_equation",
+                                    "image_path": "7c1f702cca405e1c3cc5e8b9a4e3e3432322df0dbd23cd4a844fff3c942657a6.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 9.5,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                197,
+                                254,
+                                413,
+                                272.5
+                            ],
+                            "spans": [],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                197,
+                                272.5,
+                                413,
+                                291.0
+                            ],
+                            "spans": [],
+                            "index": 10
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        295,
+                        505,
+                        348
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                295,
+                                505,
+                                309
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        295,
+                                        505,
+                                        309
+                                    ],
+                                    "score": 1.0,
+                                    "content": "where 1 is a vector of all-ones. While the elements within a feature map display strong correlations,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                306,
+                                505,
+                                320
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        306,
+                                        505,
+                                        320
+                                    ],
+                                    "score": 1.0,
+                                    "content": "different feature maps are independent and identically distributed (iid) conditioned on the data (i.e.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                315,
+                                505,
+                                336
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        318,
+                                        156,
+                                        333
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\mathbf { a } _ { i } ^ { ( 1 ) } ( \\mathbf { X } , \\mathbf { X } ^ { \\prime } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        156,
+                                        315,
+                                        175,
+                                        336
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        176,
+                                        318,
+                                        225,
+                                        333
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\mathbf { a } _ { i ^ { \\prime } } ^ { ( \\mathrm { i } ) } ( \\mathbf { X } , \\mathbf { X ^ { \\prime } } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        225,
+                                        315,
+                                        270,
+                                        336
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are iid for",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        270,
+                                        320,
+                                        295,
+                                        332
+                                    ],
+                                    "score": 0.92,
+                                    "content": "i \\neq i ^ { \\prime }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        296,
+                                        315,
+                                        505,
+                                        336
+                                    ],
+                                    "score": 1.0,
+                                    "content": "), because the parameters for different feature-maps",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                331,
+                                352,
+                                349
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        331,
+                                        169,
+                                        349
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(i.e. the biases,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        169,
+                                        333,
+                                        185,
+                                        347
+                                    ],
+                                    "score": 0.91,
+                                    "content": "b _ { i } ^ { ( 1 ) }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        185,
+                                        331,
+                                        245,
+                                        349
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and the filters,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        246,
+                                        332,
+                                        270,
+                                        348
+                                    ],
+                                    "score": 0.86,
+                                    "content": "\\mathbf { W } _ { i , : } ^ { ( 1 ) }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        270,
+                                        331,
+                                        352,
+                                        349
+                                    ],
+                                    "score": 1.0,
+                                    "content": ") are themselves iid.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        }
+                    ],
+                    "index": 12.5,
+                    "bbox_fs": [
+                        104,
+                        295,
+                        505,
+                        349
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        361,
+                        506,
+                        430
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                361,
+                                506,
+                                378
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        361,
+                                        361,
+                                        377
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Induction step. Consider the feature maps at the `th layer,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        362,
+                                        361,
+                                        410,
+                                        378
+                                    ],
+                                    "score": 0.94,
+                                    "content": "\\mathbf { a } _ { j } ^ { ( \\ell ) } ( \\mathbf { X } , \\mathbf { X } ^ { \\prime } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        411,
+                                        361,
+                                        506,
+                                        377
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", to be iid multivariate",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                375,
+                                506,
+                                394
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        375,
+                                        197,
+                                        394
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Gaussian RVs (i.e. for",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        198,
+                                        378,
+                                        223,
+                                        392
+                                    ],
+                                    "score": 0.79,
+                                    "content": "j \\neq j ^ { \\prime }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        224,
+                                        375,
+                                        227,
+                                        394
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        227,
+                                        376,
+                                        276,
+                                        393
+                                    ],
+                                    "score": 0.79,
+                                    "content": "\\mathbf { a } _ { j } ^ { ( \\ell ) } ( \\mathbf { X } , \\mathbf { X } ^ { \\prime } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        276,
+                                        375,
+                                        294,
+                                        394
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        294,
+                                        376,
+                                        343,
+                                        393
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\mathbf { a } _ { j ^ { \\prime } } ^ { ( \\ell ) } ( \\mathbf { X } , \\mathbf { X } ^ { \\prime } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        343,
+                                        375,
+                                        506,
+                                        394
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are iid). Our goal is to show that, taking",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                390,
+                                506,
+                                407
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        390,
+                                        236,
+                                        407
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the number of channels at layer",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        236,
+                                        395,
+                                        242,
+                                        404
+                                    ],
+                                    "score": 0.8,
+                                    "content": "\\ell",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        243,
+                                        390,
+                                        305,
+                                        407
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to infinity (i.e.",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        306,
+                                        393,
+                                        352,
+                                        405
+                                    ],
+                                    "score": 0.86,
+                                    "content": "C ^ { ( \\ell ) } \\to \\infty ^ { \\cdot }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        352,
+                                        390,
+                                        506,
+                                        407
+                                    ],
+                                    "score": 1.0,
+                                    "content": "), the same properties hold at the next",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                403,
+                                507,
+                                422
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        405,
+                                        219,
+                                        421
+                                    ],
+                                    "score": 1.0,
+                                    "content": "layer (i.e. all feature maps,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        221,
+                                        403,
+                                        507,
+                                        422
+                                    ],
+                                    "score": 1.0,
+                                    "content": "a(`+1)i (X, X0), are iid multivariate Gaussian RVs). Writing eq. (2) for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                418,
+                                290,
+                                431
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        418,
+                                        200,
+                                        431
+                                    ],
+                                    "score": 1.0,
+                                    "content": "two training examples,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        200,
+                                        419,
+                                        209,
+                                        428
+                                    ],
+                                    "score": 0.66,
+                                    "content": "\\mathbf { X }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        210,
+                                        418,
+                                        228,
+                                        431
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        228,
+                                        419,
+                                        240,
+                                        428
+                                    ],
+                                    "score": 0.82,
+                                    "content": "\\mathbf { X } ^ { \\prime }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        240,
+                                        418,
+                                        290,
+                                        431
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", we obtain,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        }
+                    ],
+                    "index": 17,
+                    "bbox_fs": [
+                        104,
+                        361,
+                        507,
+                        431
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        162,
+                        436,
+                        449,
+                        473
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                162,
+                                436,
+                                449,
+                                473
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        162,
+                                        436,
+                                        449,
+                                        473
+                                    ],
+                                    "score": 0.95,
+                                    "content": "\\mathbf { a } _ { i } ^ { ( \\ell + 1 ) } ( \\mathbf { X } , \\mathbf { X } ^ { \\prime } ) = b _ { i } ^ { ( \\ell + 1 ) } \\mathbf { 1 } + \\sum _ { j = 1 } ^ { C ^ { ( \\ell ) } } \\left( \\begin{array} { c c } { \\mathbf { W } _ { i , j } ^ { ( \\ell + 1 ) } } & { \\mathbf { 0 } } \\\\ { \\mathbf { 0 } } & { \\mathbf { W } _ { i , j } ^ { ( \\ell + 1 ) } } \\end{array} \\right) \\phi ( \\mathbf { a } _ { j } ^ { ( \\ell ) } ( \\mathbf { X } , \\mathbf { X } ^ { \\prime } ) )",
+                                    "type": "interline_equation",
+                                    "image_path": "3cc3207e1f6d7550a22043ec02cfccb2c0052e0bfb6d7fedaae056ea3c1ef617.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 21,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                162,
+                                436,
+                                449,
+                                448.3333333333333
+                            ],
+                            "spans": [],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                162,
+                                448.3333333333333,
+                                449,
+                                460.66666666666663
+                            ],
+                            "spans": [],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                162,
+                                460.66666666666663,
+                                449,
+                                472.99999999999994
+                            ],
+                            "spans": [],
+                            "index": 22
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        480,
+                        506,
+                        558
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                477,
+                                508,
+                                498
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        477,
+                                        218,
+                                        498
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We begin by showing that",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        218,
+                                        480,
+                                        277,
+                                        495
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\mathbf { a } _ { i } ^ { ( \\ell + 1 ) } ( \\mathbf { X } , \\mathbf { X } ^ { \\prime } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        277,
+                                        477,
+                                        508,
+                                        498
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is a multivariate Gaussian RV. The first term is multi-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                491,
+                                507,
+                                511
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        491,
+                                        287,
+                                        511
+                                    ],
+                                    "score": 1.0,
+                                    "content": "variate Gaussian, as it is a linear function of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        287,
+                                        494,
+                                        312,
+                                        509
+                                    ],
+                                    "score": 0.93,
+                                    "content": "b _ { i } ^ { ( \\ell + 1 ) }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        313,
+                                        491,
+                                        507,
+                                        511
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", which is itself iid Gaussian. We can apply the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                505,
+                                508,
+                                524
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        505,
+                                        457,
+                                        524
+                                    ],
+                                    "score": 1.0,
+                                    "content": "multivariate CLT to show that the second term is also Gaussian, because, in the limit as",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        457,
+                                        508,
+                                        501,
+                                        519
+                                    ],
+                                    "score": 0.9,
+                                    "content": "C ^ { ( \\ell ) } \\to \\infty",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        502,
+                                        505,
+                                        508,
+                                        524
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                102,
+                                519,
+                                506,
+                                537
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        102,
+                                        519,
+                                        279,
+                                        537
+                                    ],
+                                    "score": 1.0,
+                                    "content": "it is the sum of infinitely many iid terms:",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        279,
+                                        520,
+                                        328,
+                                        536
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\mathbf { a } _ { j } ^ { ( \\ell ) } ( \\mathbf { X } , \\mathbf { X } ^ { \\prime } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        329,
+                                        519,
+                                        441,
+                                        537
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are iid by assumption, and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        441,
+                                        520,
+                                        474,
+                                        536
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\mathbf { W } _ { i , j } ^ { ( \\ell + 1 ) }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        475,
+                                        519,
+                                        506,
+                                        537
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are iid",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                532,
+                                507,
+                                546
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        532,
+                                        507,
+                                        546
+                                    ],
+                                    "score": 1.0,
+                                    "content": "by definition. Note that the same argument applies to all feature maps jointly, so all elements of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                107,
+                                543,
+                                432,
+                                559
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        544,
+                                        168,
+                                        558
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\mathbf { A } ^ { ( \\ell + 1 ) } ( \\mathbf { X } , \\mathbf { X } ^ { \\prime } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        169,
+                                        543,
+                                        432,
+                                        559
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(defined by analogy with eq. 4) are jointly multivariate Gaussian.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        }
+                    ],
+                    "index": 25.5,
+                    "bbox_fs": [
+                        102,
+                        477,
+                        508,
+                        559
+                    ]
+                },
+                {
+                    "type": "list",
+                    "bbox": [
+                        106,
+                        561,
+                        505,
+                        668
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                561,
+                                505,
+                                576
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        561,
+                                        505,
+                                        576
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Following Lee et al. (2017), to complete the argument, we need to show that the output feature maps",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                571,
+                                506,
+                                590
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        571,
+                                        155,
+                                        589
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are iid, i.e.",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        156,
+                                        573,
+                                        215,
+                                        588
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\mathbf { a } _ { i } ^ { ( \\ell + 1 ) } ( \\mathbf { X } , \\mathbf { X } ^ { \\prime } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        180,
+                                        572,
+                                        250,
+                                        590
+                                    ],
+                                    "score": 1.0,
+                                    "content": ")(X, X0) and a(`i0",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        235,
+                                        573,
+                                        294,
+                                        588
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\mathbf { a } _ { i ^ { \\prime } } ^ { ( \\ell + 1 ) } ( \\mathbf { X } , \\mathbf { X } ^ { \\prime } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        294,
+                                        571,
+                                        339,
+                                        590
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are iid for",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        340,
+                                        575,
+                                        366,
+                                        588
+                                    ],
+                                    "score": 0.91,
+                                    "content": "i \\neq i ^ { \\prime }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        367,
+                                        571,
+                                        506,
+                                        590
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". They are identically distributed,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                583,
+                                509,
+                                622
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        585,
+                                        118,
+                                        622
+                                    ],
+                                    "score": 1.0,
+                                    "content": "re",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        110,
+                                        583,
+                                        151,
+                                        603
+                                    ],
+                                    "score": 1.0,
+                                    "content": "as b(`+1) a",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        144,
+                                        585,
+                                        164,
+                                        622
+                                    ],
+                                    "score": 1.0,
+                                    "content": "nd r tha",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        164,
+                                        588,
+                                        198,
+                                        604
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\mathbf { W } _ { i , j } ^ { ( \\ell + 1 ) }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        228,
+                                        585,
+                                        248,
+                                        622
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and  and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        249,
+                                        588,
+                                        309,
+                                        603
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\phi ( \\mathbf { a } _ { j } ^ { ( \\ell ) } ( \\mathbf { X } , \\mathbf { X } ^ { \\prime } ) )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        310,
+                                        585,
+                                        509,
+                                        622
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is shared. To show that they are independent,are jointly Gaussian, so it is sufficient to show",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                169,
+                                603,
+                                308,
+                                618
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        169,
+                                        603,
+                                        228,
+                                        618
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\mathbf { a } _ { i } ^ { ( \\ell + 1 ) } ( \\mathbf { X } , \\mathbf { X } ^ { \\prime } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        249,
+                                        604,
+                                        308,
+                                        618
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\mathbf { a } _ { i ^ { \\prime } } ^ { ( \\ell + 1 ) } ( \\mathbf { X } , \\mathbf { X } ^ { \\prime } )",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 32,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                102,
+                                617,
+                                503,
+                                634
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        102,
+                                        617,
+                                        470,
+                                        634
+                                    ],
+                                    "score": 1.0,
+                                    "content": "that they are uncorrelated, and we can show that they are uncorrelated because the weights,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        470,
+                                        617,
+                                        503,
+                                        634
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\mathbf { W } _ { i , j } ^ { ( \\ell + 1 ) }",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 33,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                632,
+                                505,
+                                643
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        632,
+                                        505,
+                                        643
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are independent with zero-mean, eliminating any correlations that might arise through the shared",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                642,
+                                506,
+                                658
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        642,
+                                        126,
+                                        658
+                                    ],
+                                    "score": 1.0,
+                                    "content": "RV,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        126,
+                                        642,
+                                        187,
+                                        658
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\phi ( \\mathbf { a } _ { j } ^ { ( \\ell ) } ( \\mathbf { X } , \\mathbf { X } ^ { \\prime } ) )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        188,
+                                        642,
+                                        506,
+                                        658
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". In the appendix, we consider the more complex case where we take limits",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                654,
+                                171,
+                                668
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        654,
+                                        171,
+                                        668
+                                    ],
+                                    "score": 1.0,
+                                    "content": "simultaneously.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36,
+                            "is_list_end_line": true
+                        }
+                    ],
+                    "index": 32.5,
+                    "bbox_fs": [
+                        102,
+                        561,
+                        509,
+                        668
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        684,
+                        326,
+                        697
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                682,
+                                327,
+                                699
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        682,
+                                        327,
+                                        699
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3 THE CONVNET AND RESNET KERNELS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        }
+                    ],
+                    "index": 37
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        709,
+                        504,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                710,
+                                504,
+                                722
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        710,
+                                        504,
+                                        722
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Here we derive a computationally efficient kernel corresponding to the CNN described in the pre-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                719,
+                                506,
+                                734
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        719,
+                                        506,
+                                        734
+                                    ],
+                                    "score": 1.0,
+                                    "content": "vious section. It is surprising that we can compute the kernel efficiently because the feature maps,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                79,
+                                506,
+                                97
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        80,
+                                        139,
+                                        95
+                                    ],
+                                    "score": 0.92,
+                                    "content": "{ \\bf a } _ { i } ^ { ( \\ell ) } ( { \\bf X } )",
+                                    "type": "inline_equation",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        140,
+                                        79,
+                                        506,
+                                        97
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", display rich covariance structure due to the shared convolutional filter. Computing and rep-",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                93,
+                                505,
+                                107
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        93,
+                                        505,
+                                        107
+                                    ],
+                                    "score": 1.0,
+                                    "content": "resenting these covariances would be prohibitively computationally expensive. However, in many",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                105,
+                                506,
+                                118
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        105,
+                                        506,
+                                        118
+                                    ],
+                                    "score": 1.0,
+                                    "content": "cases we only need the variance of the output, e.g. in the case of classification or regression with a",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                115,
+                                505,
+                                128
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        115,
+                                        505,
+                                        128
+                                    ],
+                                    "score": 1.0,
+                                    "content": "final dense layer. It turns out that this propagates backwards through the convolutional network, im-",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                127,
+                                505,
+                                138
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        127,
+                                        505,
+                                        138
+                                    ],
+                                    "score": 1.0,
+                                    "content": "plying that for every layer, we only need the “diagonal covariance” of the activations: the covariance",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                102,
+                                136,
+                                503,
+                                159
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        102,
+                                        136,
+                                        265,
+                                        159
+                                    ],
+                                    "score": 1.0,
+                                    "content": "between the corresponding elements of",
+                                    "type": "text",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        265,
+                                        138,
+                                        298,
+                                        155
+                                    ],
+                                    "score": 0.93,
+                                    "content": "{ \\bf a } _ { j } ^ { ( \\ell ) } ( { \\bf X } )",
+                                    "type": "inline_equation",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        299,
+                                        136,
+                                        316,
+                                        159
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        316,
+                                        138,
+                                        352,
+                                        155
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\mathbf { a } _ { j } ^ { ( \\ell ) } ( \\mathbf { X } ^ { \\prime } )",
+                                    "type": "inline_equation",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        353,
+                                        136,
+                                        372,
+                                        159
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(i.e.",
+                                    "type": "text",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        372,
+                                        137,
+                                        497,
+                                        158
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\mathrm { d i a g } \\left( \\mathbb { C } \\left[ \\mathbf { a } _ { j } ^ { ( \\ell ) } ( \\mathbf { X } ) , \\mathbf { a } _ { j } ^ { ( \\ell ) } ( \\mathbf { X } ^ { \\prime } ) \\right] \\right) )",
+                                    "type": "inline_equation",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        498,
+                                        136,
+                                        503,
+                                        159
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 5
+                        }
+                    ],
+                    "index": 38.5,
+                    "bbox_fs": [
+                        105,
+                        710,
+                        506,
+                        734
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        81,
+                        505,
+                        156
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                79,
+                                506,
+                                97
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        80,
+                                        139,
+                                        95
+                                    ],
+                                    "score": 0.92,
+                                    "content": "{ \\bf a } _ { i } ^ { ( \\ell ) } ( { \\bf X } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        140,
+                                        79,
+                                        506,
+                                        97
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", display rich covariance structure due to the shared convolutional filter. Computing and rep-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                93,
+                                505,
+                                107
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        93,
+                                        505,
+                                        107
+                                    ],
+                                    "score": 1.0,
+                                    "content": "resenting these covariances would be prohibitively computationally expensive. However, in many",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                105,
+                                506,
+                                118
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        105,
+                                        506,
+                                        118
+                                    ],
+                                    "score": 1.0,
+                                    "content": "cases we only need the variance of the output, e.g. in the case of classification or regression with a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                115,
+                                505,
+                                128
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        115,
+                                        505,
+                                        128
+                                    ],
+                                    "score": 1.0,
+                                    "content": "final dense layer. It turns out that this propagates backwards through the convolutional network, im-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                127,
+                                505,
+                                138
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        127,
+                                        505,
+                                        138
+                                    ],
+                                    "score": 1.0,
+                                    "content": "plying that for every layer, we only need the “diagonal covariance” of the activations: the covariance",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                102,
+                                136,
+                                503,
+                                159
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        102,
+                                        136,
+                                        265,
+                                        159
+                                    ],
+                                    "score": 1.0,
+                                    "content": "between the corresponding elements of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        265,
+                                        138,
+                                        298,
+                                        155
+                                    ],
+                                    "score": 0.93,
+                                    "content": "{ \\bf a } _ { j } ^ { ( \\ell ) } ( { \\bf X } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        299,
+                                        136,
+                                        316,
+                                        159
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        316,
+                                        138,
+                                        352,
+                                        155
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\mathbf { a } _ { j } ^ { ( \\ell ) } ( \\mathbf { X } ^ { \\prime } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        353,
+                                        136,
+                                        372,
+                                        159
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(i.e.",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        372,
+                                        137,
+                                        497,
+                                        158
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\mathrm { d i a g } \\left( \\mathbb { C } \\left[ \\mathbf { a } _ { j } ^ { ( \\ell ) } ( \\mathbf { X } ) , \\mathbf { a } _ { j } ^ { ( \\ell ) } ( \\mathbf { X } ^ { \\prime } ) \\right] \\right) )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        498,
+                                        136,
+                                        503,
+                                        159
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        }
+                    ],
+                    "index": 2.5
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        168,
+                        253,
+                        180
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                168,
+                                254,
+                                181
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        168,
+                                        254,
+                                        181
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3.1 GP MEAN AND COVARIANCE",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        }
+                    ],
+                    "index": 6
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        189,
+                        505,
+                        234
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                188,
+                                505,
+                                203
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        188,
+                                        505,
+                                        203
+                                    ],
+                                    "score": 1.0,
+                                    "content": "A GP is completely specified by its mean and covariance (kernel) functions. These give the pa-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                200,
+                                506,
+                                213
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        200,
+                                        440,
+                                        213
+                                    ],
+                                    "score": 1.0,
+                                    "content": "rameters of the joint Gaussian distribution of the RVs indexed by any two inputs,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        441,
+                                        201,
+                                        451,
+                                        211
+                                    ],
+                                    "score": 0.55,
+                                    "content": "\\mathbf { X }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        451,
+                                        200,
+                                        470,
+                                        213
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        470,
+                                        200,
+                                        482,
+                                        211
+                                    ],
+                                    "score": 0.85,
+                                    "content": "\\mathbf { X } ^ { \\prime }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        483,
+                                        200,
+                                        506,
+                                        213
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". For",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                211,
+                                505,
+                                225
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        211,
+                                        505,
+                                        225
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the purposes of computing the mean and covariance, it is easiest to consider the network as being",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                222,
+                                242,
+                                235
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        222,
+                                        242,
+                                        235
+                                    ],
+                                    "score": 1.0,
+                                    "content": "written entirely in index notation,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        }
+                    ],
+                    "index": 8.5
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        189,
+                        235,
+                        422,
+                        273
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                189,
+                                235,
+                                422,
+                                273
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        189,
+                                        235,
+                                        422,
+                                        273
+                                    ],
+                                    "score": 0.95,
+                                    "content": "A _ { i , \\mu } ^ { ( \\ell + 1 ) } ( \\mathbf { X } ) = b _ { i } ^ { ( \\ell + 1 ) } + \\sum _ { j = 1 } ^ { C ^ { ( \\ell ) } } \\sum _ { \\nu = 1 } ^ { H ^ { ( \\ell ) } D ^ { ( \\ell ) } } W _ { i , j , \\mu , \\nu } ^ { ( \\ell + 1 ) } \\phi ( A _ { j , \\nu } ^ { ( \\ell ) } ( \\mathbf { X } ) ) .",
+                                    "type": "interline_equation",
+                                    "image_path": "80a3003fe9e92f02bddb811082b168b48e11c15ca923cf028b88f1306e789b49.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 11.5,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                189,
+                                235,
+                                422,
+                                254.0
+                            ],
+                            "spans": [],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                189,
+                                254.0,
+                                422,
+                                273.0
+                            ],
+                            "spans": [],
+                            "index": 12
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        276,
+                        506,
+                        311
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                275,
+                                506,
+                                289
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        275,
+                                        133,
+                                        289
+                                    ],
+                                    "score": 1.0,
+                                    "content": "where",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        133,
+                                        277,
+                                        139,
+                                        286
+                                    ],
+                                    "score": 0.77,
+                                    "content": "\\ell",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        139,
+                                        275,
+                                        156,
+                                        289
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        157,
+                                        276,
+                                        179,
+                                        287
+                                    ],
+                                    "score": 0.9,
+                                    "content": "\\ell + 1",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        179,
+                                        275,
+                                        369,
+                                        289
+                                    ],
+                                    "score": 1.0,
+                                    "content": "denote the input and output layers respectively,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        369,
+                                        277,
+                                        375,
+                                        288
+                                    ],
+                                    "score": 0.79,
+                                    "content": "j",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        375,
+                                        275,
+                                        393,
+                                        289
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        393,
+                                        275,
+                                        475,
+                                        288
+                                    ],
+                                    "score": 0.92,
+                                    "content": "i \\in \\{ 1 , \\ldots , C ^ { ( \\ell + 1 ) } \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        475,
+                                        275,
+                                        506,
+                                        289
+                                    ],
+                                    "score": 1.0,
+                                    "content": "denote",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                285,
+                                506,
+                                302
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        285,
+                                        244,
+                                        302
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the input and output channels, and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        245,
+                                        290,
+                                        252,
+                                        298
+                                    ],
+                                    "score": 0.76,
+                                    "content": "\\nu",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        252,
+                                        285,
+                                        269,
+                                        302
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        270,
+                                        287,
+                                        384,
+                                        300
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\mu \\in \\{ 1 , \\ldots , H ^ { ( \\ell + 1 ) } D ^ { ( \\ell + 1 ) } \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        384,
+                                        285,
+                                        506,
+                                        302
+                                    ],
+                                    "score": 1.0,
+                                    "content": "denote the location within the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                299,
+                                276,
+                                313
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        299,
+                                        276,
+                                        313
+                                    ],
+                                    "score": 1.0,
+                                    "content": "input and output channel or feature-maps.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        }
+                    ],
+                    "index": 14
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        316,
+                        280,
+                        327
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                314,
+                                280,
+                                330
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        314,
+                                        280,
+                                        330
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The mean function is thus easy to compute",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        }
+                    ],
+                    "index": 16
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        153,
+                        329,
+                        457,
+                        367
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                153,
+                                329,
+                                457,
+                                367
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        153,
+                                        329,
+                                        457,
+                                        367
+                                    ],
+                                    "score": 0.94,
+                                    "content": "\\mathbb { E } \\left[ A _ { i , \\mu } ^ { ( \\ell + 1 ) } ( \\mathbf { X } ) \\right] = \\mathbb { E } \\left[ b _ { i } ^ { ( \\ell + 1 ) } \\right] + \\sum _ { j = 1 } ^ { C ^ { ( \\ell ) } } \\sum _ { \\nu = 1 } ^ { H ^ { ( \\ell ) } D ^ { ( \\ell ) } } \\mathbb { E } \\left[ W _ { i , j , \\mu , \\nu } ^ { ( \\ell + 1 ) } \\phi ( A _ { j , \\nu } ^ { ( \\ell ) } ( \\mathbf { X } ) ) \\right] = 0 .",
+                                    "type": "interline_equation",
+                                    "image_path": "e747a94ec49cda03ffab5ab9caa3a170304fdc5e31eb4da72587b300ad8e8aee.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 18,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                153,
+                                329,
+                                457,
+                                341.6666666666667
+                            ],
+                            "spans": [],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                153,
+                                341.6666666666667,
+                                457,
+                                354.33333333333337
+                            ],
+                            "spans": [],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                153,
+                                354.33333333333337,
+                                457,
+                                367.00000000000006
+                            ],
+                            "spans": [],
+                            "index": 19
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        369,
+                        503,
+                        400
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                100,
+                                366,
+                                505,
+                                387
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        100,
+                                        366,
+                                        149,
+                                        386
+                                    ],
+                                    "score": 1.0,
+                                    "content": "as b(`+1)i a",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        138,
+                                        367,
+                                        197,
+                                        385
+                                    ],
+                                    "score": 1.0,
+                                    "content": "nd W (`+1)i,j,µ,ν",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        190,
+                                        367,
+                                        313,
+                                        386
+                                    ],
+                                    "score": 1.0,
+                                    "content": "have zero mean, and W (`+1)i,j,ν,µ",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        309,
+                                        371,
+                                        505,
+                                        387
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are independent of the activations at the previous",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                102,
+                                381,
+                                187,
+                                405
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        102,
+                                        381,
+                                        131,
+                                        405
+                                    ],
+                                    "score": 1.0,
+                                    "content": "layer,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        131,
+                                        385,
+                                        179,
+                                        401
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\phi ( A _ { j , \\nu } ^ { ( \\ell ) } ( \\mathbf { X } ) )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        180,
+                                        381,
+                                        187,
+                                        405
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        }
+                    ],
+                    "index": 20.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        403,
+                        506,
+                        489
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                402,
+                                506,
+                                418
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        402,
+                                        506,
+                                        418
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Now we show that it is possible to efficiently compute the covariance function. This is surprising",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                415,
+                                504,
+                                452
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        415,
+                                        241,
+                                        452
+                                    ],
+                                    "score": 1.0,
+                                    "content": "because for many networks, we nlocations in the feature map (i.e.",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        241,
+                                        426,
+                                        356,
+                                        446
+                                    ],
+                                    "score": 0.94,
+                                    "content": "\\mathbb { C } \\left[ A _ { i , \\mu } ^ { ( \\ell + 1 ) } ( \\mathbf { \\bar { X } } ) , A _ { i , \\mu ^ { \\prime } } ^ { ( \\ell + 1 ) } ( \\mathbf { X } ^ { \\prime } ) \\right]",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        356,
+                                        415,
+                                        372,
+                                        452
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ce ofor",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        372,
+                                        428,
+                                        504,
+                                        442
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\mu , \\mu ^ { \\prime } \\in \\{ 1 , \\ldots , H ^ { ( \\ell + 1 ) } D ^ { ( \\ell + 1 ) } \\} )",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                442,
+                                507,
+                                460
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        442,
+                                        295,
+                                        460
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and this object is extremely high-dimensional,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        295,
+                                        444,
+                                        380,
+                                        459
+                                    ],
+                                    "score": 0.89,
+                                    "content": "N ^ { 2 } ( H ^ { ( \\ell + 1 ) } D ^ { \\overline { { ( \\ell + 1 ) } } } ) ^ { 2 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        380,
+                                        442,
+                                        507,
+                                        460
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". However, it turns out that we",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                451,
+                                506,
+                                483
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        451,
+                                        375,
+                                        483
+                                    ],
+                                    "score": 1.0,
+                                    "content": "only need to consider the “diagonal” covariance, (i.e. we only need",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        376,
+                                        457,
+                                        489,
+                                        477
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\mathbb { C } \\left[ A _ { i , \\mu } ^ { ( \\ell + 1 ) } ( \\mathbf { X } ) , A _ { i , \\mu } ^ { ( \\ell + 1 ) } ( \\mathbf { X } ^ { \\prime } ) \\right]",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        490,
+                                        451,
+                                        506,
+                                        483
+                                    ],
+                                    "score": 1.0,
+                                    "content": "fo r",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                474,
+                                493,
+                                491
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        475,
+                                        223,
+                                        489
+                                    ],
+                                    "score": 0.89,
+                                    "content": "\\mu \\in \\{ 1 , \\ldots , H ^ { ( \\ell + 1 ) } D ^ { ( \\ell + 1 ) } \\} )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        224,
+                                        474,
+                                        408,
+                                        491
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", which is a more manageable quantity of size",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        408,
+                                        476,
+                                        488,
+                                        489
+                                    ],
+                                    "score": 0.88,
+                                    "content": "N ^ { 2 } ( H ^ { ( \\ell + 1 ) } D ^ { ( \\ell + 1 ) } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        488,
+                                        474,
+                                        493,
+                                        491
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        }
+                    ],
+                    "index": 24
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        493,
+                        505,
+                        562
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                493,
+                                505,
+                                507
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        493,
+                                        228,
+                                        507
+                                    ],
+                                    "score": 1.0,
+                                    "content": "This is true at the output layer",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        228,
+                                        494,
+                                        259,
+                                        506
+                                    ],
+                                    "score": 0.91,
+                                    "content": "( L + 1 )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        259,
+                                        493,
+                                        505,
+                                        507
+                                    ],
+                                    "score": 1.0,
+                                    "content": ": in order to achieve an output suitable for classification or re-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                504,
+                                506,
+                                519
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        504,
+                                        290,
+                                        519
+                                    ],
+                                    "score": 1.0,
+                                    "content": "gression, we use only a single output location",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        290,
+                                        505,
+                                        385,
+                                        517
+                                    ],
+                                    "score": 0.92,
+                                    "content": "H ^ { ( L + 1 ) } = D ^ { ( L + 1 ) } = 1",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        385,
+                                        504,
+                                        506,
+                                        519
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", with a number of “channels”",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                517,
+                                506,
+                                530
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        517,
+                                        506,
+                                        530
+                                    ],
+                                    "score": 1.0,
+                                    "content": "equal to the number of of outputs/classes, so it is only possible to compute the covariance at that",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                529,
+                                505,
+                                541
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        529,
+                                        505,
+                                        541
+                                    ],
+                                    "score": 1.0,
+                                    "content": "single location. We now show that, if we only need the covariance at corresponding locations in the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                540,
+                                505,
+                                551
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        540,
+                                        505,
+                                        551
+                                    ],
+                                    "score": 1.0,
+                                    "content": "outputs, we only need the covariance at corresponding locations in the inputs, and this requirement",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                551,
+                                283,
+                                562
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        551,
+                                        283,
+                                        562
+                                    ],
+                                    "score": 1.0,
+                                    "content": "propagates backwards through the network.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        }
+                    ],
+                    "index": 29.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        567,
+                        504,
+                        590
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                567,
+                                505,
+                                579
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        567,
+                                        505,
+                                        579
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Formally, as the activations are composed of a sum of terms, their covariance is the sum of the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                577,
+                                275,
+                                591
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        577,
+                                        275,
+                                        591
+                                    ],
+                                    "score": 1.0,
+                                    "content": "covariances of all those underlying terms,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        }
+                    ],
+                    "index": 33.5
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        136,
+                        590,
+                        477,
+                        651
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                136,
+                                590,
+                                477,
+                                651
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        136,
+                                        590,
+                                        477,
+                                        651
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\begin{array} { r l } { \\mathbb { C } \\left[ A _ { i , \\mu } ^ { ( \\ell + 1 ) } ( \\mathbf { X } ) , A _ { i , \\mu } ^ { ( \\ell + 1 ) } ( \\mathbf { X } ^ { \\prime } ) \\right] = \\mathbb { V } \\left[ b _ { i } ^ { ( \\ell ) } \\right] + } & { } \\\\ { \\displaystyle \\sum _ { j = 1 } ^ { C ^ { ( \\ell ) } } \\sum _ { j ^ { \\prime } = 1 } ^ { C ^ { ( \\ell ) } } \\sum _ { \\nu = 1 } ^ { H ^ { ( \\ell ) } D ^ { ( \\ell ) } } \\sum _ { \\nu ^ { \\prime } = 1 } ^ { H ^ { ( \\ell ) } D ^ { ( \\ell ) } } \\mathbb { C } \\left[ W _ { i , j , \\mu , \\nu } ^ { ( \\ell + 1 ) } \\phi ( A _ { j , \\nu } ^ { ( \\ell ) } ( \\mathbf { X } ) ) , W _ { i , j ^ { \\prime } , \\mu , \\nu ^ { \\prime } } ^ { ( \\ell + 1 ) } \\phi ( A _ { j ^ { \\prime } , \\nu ^ { \\prime } } ^ { ( \\ell ) } ( \\mathbf { X } ^ { \\prime } ) ) \\right] . } \\end{array}",
+                                    "type": "interline_equation",
+                                    "image_path": "69d31cc97d2fc0067ded1890ef2527e60cbaecb75da932e44d6be86c09e39c9e.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 36,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                136,
+                                590,
+                                477,
+                                610.3333333333334
+                            ],
+                            "spans": [],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                136,
+                                610.3333333333334,
+                                477,
+                                630.6666666666667
+                            ],
+                            "spans": [],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                136,
+                                630.6666666666667,
+                                477,
+                                651.0000000000001
+                            ],
+                            "spans": [],
+                            "index": 37
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        652,
+                        507,
+                        674
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                651,
+                                505,
+                                664
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        651,
+                                        505,
+                                        664
+                                    ],
+                                    "score": 1.0,
+                                    "content": "As the terms in the covariance have mean zero, and as the weights and activations from the previous",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                663,
+                                197,
+                                676
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        663,
+                                        197,
+                                        676
+                                    ],
+                                    "score": 1.0,
+                                    "content": "layer are independent,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        }
+                    ],
+                    "index": 38.5
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        122,
+                        675,
+                        478,
+                        736
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                122,
+                                675,
+                                478,
+                                736
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        122,
+                                        675,
+                                        478,
+                                        736
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\begin{array} { r l } { \\mathbb { C } \\left[ A _ { i , \\mu } ^ { ( \\ell + 1 ) } ( \\mathbf { X } ) , A _ { i , \\mu } ^ { ( \\ell + 1 ) } ( \\mathbf { X } ^ { \\prime } ) \\right] = \\sigma _ { \\mathrm { b } } ^ { 2 } + } & { } \\\\ { \\quad } & { \\qquad \\sum _ { j = 1 } ^ { { \\ell ^ { ( \\ell ) } } } \\displaystyle \\sum _ { j ^ { \\prime } = 1 } ^ { { \\ell ^ { ( \\ell ) } } } \\sum _ { \\nu = 1 } ^ { { H ^ { ( \\ell ) } } D ^ { ( \\ell ) } } \\sum _ { \\nu ^ { \\prime } = 1 } ^ { H ^ { ( \\ell ) } D ^ { ( \\ell ) } } \\mathbb { E } \\left[ W _ { i , j , \\mu , \\nu } ^ { ( \\ell + 1 ) } W _ { i , j ^ { \\prime } , \\mu , \\nu ^ { \\prime } } ^ { ( \\ell + 1 ) } \\right] \\mathbb { E } \\left[ \\phi ( A _ { j , \\nu } ^ { ( \\ell ) } ( \\mathbf { X } ) ) \\phi ( A _ { j ^ { \\prime } , \\nu ^ { \\prime } } ^ { ( \\ell ) } ( \\mathbf { X } ^ { \\prime } ) ) \\right] . } \\end{array}",
+                                    "type": "interline_equation",
+                                    "image_path": "3654c365438b76a29a2541724e15f6857f74e810a656fd4cf794da4018587cad.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 41,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                122,
+                                675,
+                                478,
+                                695.3333333333334
+                            ],
+                            "spans": [],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                122,
+                                695.3333333333334,
+                                478,
+                                715.6666666666667
+                            ],
+                            "spans": [],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                122,
+                                715.6666666666667,
+                                478,
+                                736.0000000000001
+                            ],
+                            "spans": [],
+                            "index": 42
+                        }
+                    ]
+                }
+            ],
+            "page_idx": 4,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        107,
+                        27,
+                        293,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                26,
+                                294,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        26,
+                                        294,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Published as a conference paper at ICLR 2019",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        302,
+                        751,
+                        309,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                302,
+                                750,
+                                309,
+                                762
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        302,
+                                        750,
+                                        309,
+                                        762
+                                    ],
+                                    "score": 1.0,
+                                    "content": "5",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        81,
+                        505,
+                        156
+                    ],
+                    "lines": [],
+                    "index": 2.5,
+                    "bbox_fs": [
+                        102,
+                        79,
+                        506,
+                        159
+                    ],
+                    "lines_deleted": true
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        168,
+                        253,
+                        180
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                168,
+                                254,
+                                181
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        168,
+                                        254,
+                                        181
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3.1 GP MEAN AND COVARIANCE",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        }
+                    ],
+                    "index": 6
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        189,
+                        505,
+                        234
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                188,
+                                505,
+                                203
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        188,
+                                        505,
+                                        203
+                                    ],
+                                    "score": 1.0,
+                                    "content": "A GP is completely specified by its mean and covariance (kernel) functions. These give the pa-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                200,
+                                506,
+                                213
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        200,
+                                        440,
+                                        213
+                                    ],
+                                    "score": 1.0,
+                                    "content": "rameters of the joint Gaussian distribution of the RVs indexed by any two inputs,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        441,
+                                        201,
+                                        451,
+                                        211
+                                    ],
+                                    "score": 0.55,
+                                    "content": "\\mathbf { X }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        451,
+                                        200,
+                                        470,
+                                        213
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        470,
+                                        200,
+                                        482,
+                                        211
+                                    ],
+                                    "score": 0.85,
+                                    "content": "\\mathbf { X } ^ { \\prime }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        483,
+                                        200,
+                                        506,
+                                        213
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". For",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                211,
+                                505,
+                                225
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        211,
+                                        505,
+                                        225
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the purposes of computing the mean and covariance, it is easiest to consider the network as being",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                222,
+                                242,
+                                235
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        222,
+                                        242,
+                                        235
+                                    ],
+                                    "score": 1.0,
+                                    "content": "written entirely in index notation,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        }
+                    ],
+                    "index": 8.5,
+                    "bbox_fs": [
+                        104,
+                        188,
+                        506,
+                        235
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        189,
+                        235,
+                        422,
+                        273
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                189,
+                                235,
+                                422,
+                                273
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        189,
+                                        235,
+                                        422,
+                                        273
+                                    ],
+                                    "score": 0.95,
+                                    "content": "A _ { i , \\mu } ^ { ( \\ell + 1 ) } ( \\mathbf { X } ) = b _ { i } ^ { ( \\ell + 1 ) } + \\sum _ { j = 1 } ^ { C ^ { ( \\ell ) } } \\sum _ { \\nu = 1 } ^ { H ^ { ( \\ell ) } D ^ { ( \\ell ) } } W _ { i , j , \\mu , \\nu } ^ { ( \\ell + 1 ) } \\phi ( A _ { j , \\nu } ^ { ( \\ell ) } ( \\mathbf { X } ) ) .",
+                                    "type": "interline_equation",
+                                    "image_path": "80a3003fe9e92f02bddb811082b168b48e11c15ca923cf028b88f1306e789b49.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 11.5,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                189,
+                                235,
+                                422,
+                                254.0
+                            ],
+                            "spans": [],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                189,
+                                254.0,
+                                422,
+                                273.0
+                            ],
+                            "spans": [],
+                            "index": 12
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        276,
+                        506,
+                        311
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                275,
+                                506,
+                                289
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        275,
+                                        133,
+                                        289
+                                    ],
+                                    "score": 1.0,
+                                    "content": "where",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        133,
+                                        277,
+                                        139,
+                                        286
+                                    ],
+                                    "score": 0.77,
+                                    "content": "\\ell",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        139,
+                                        275,
+                                        156,
+                                        289
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        157,
+                                        276,
+                                        179,
+                                        287
+                                    ],
+                                    "score": 0.9,
+                                    "content": "\\ell + 1",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        179,
+                                        275,
+                                        369,
+                                        289
+                                    ],
+                                    "score": 1.0,
+                                    "content": "denote the input and output layers respectively,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        369,
+                                        277,
+                                        375,
+                                        288
+                                    ],
+                                    "score": 0.79,
+                                    "content": "j",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        375,
+                                        275,
+                                        393,
+                                        289
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        393,
+                                        275,
+                                        475,
+                                        288
+                                    ],
+                                    "score": 0.92,
+                                    "content": "i \\in \\{ 1 , \\ldots , C ^ { ( \\ell + 1 ) } \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        475,
+                                        275,
+                                        506,
+                                        289
+                                    ],
+                                    "score": 1.0,
+                                    "content": "denote",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                285,
+                                506,
+                                302
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        285,
+                                        244,
+                                        302
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the input and output channels, and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        245,
+                                        290,
+                                        252,
+                                        298
+                                    ],
+                                    "score": 0.76,
+                                    "content": "\\nu",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        252,
+                                        285,
+                                        269,
+                                        302
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        270,
+                                        287,
+                                        384,
+                                        300
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\mu \\in \\{ 1 , \\ldots , H ^ { ( \\ell + 1 ) } D ^ { ( \\ell + 1 ) } \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        384,
+                                        285,
+                                        506,
+                                        302
+                                    ],
+                                    "score": 1.0,
+                                    "content": "denote the location within the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                299,
+                                276,
+                                313
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        299,
+                                        276,
+                                        313
+                                    ],
+                                    "score": 1.0,
+                                    "content": "input and output channel or feature-maps.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        }
+                    ],
+                    "index": 14,
+                    "bbox_fs": [
+                        105,
+                        275,
+                        506,
+                        313
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        316,
+                        280,
+                        327
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                314,
+                                280,
+                                330
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        314,
+                                        280,
+                                        330
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The mean function is thus easy to compute",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        }
+                    ],
+                    "index": 16,
+                    "bbox_fs": [
+                        106,
+                        314,
+                        280,
+                        330
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        153,
+                        329,
+                        457,
+                        367
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                153,
+                                329,
+                                457,
+                                367
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        153,
+                                        329,
+                                        457,
+                                        367
+                                    ],
+                                    "score": 0.94,
+                                    "content": "\\mathbb { E } \\left[ A _ { i , \\mu } ^ { ( \\ell + 1 ) } ( \\mathbf { X } ) \\right] = \\mathbb { E } \\left[ b _ { i } ^ { ( \\ell + 1 ) } \\right] + \\sum _ { j = 1 } ^ { C ^ { ( \\ell ) } } \\sum _ { \\nu = 1 } ^ { H ^ { ( \\ell ) } D ^ { ( \\ell ) } } \\mathbb { E } \\left[ W _ { i , j , \\mu , \\nu } ^ { ( \\ell + 1 ) } \\phi ( A _ { j , \\nu } ^ { ( \\ell ) } ( \\mathbf { X } ) ) \\right] = 0 .",
+                                    "type": "interline_equation",
+                                    "image_path": "e747a94ec49cda03ffab5ab9caa3a170304fdc5e31eb4da72587b300ad8e8aee.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 18,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                153,
+                                329,
+                                457,
+                                341.6666666666667
+                            ],
+                            "spans": [],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                153,
+                                341.6666666666667,
+                                457,
+                                354.33333333333337
+                            ],
+                            "spans": [],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                153,
+                                354.33333333333337,
+                                457,
+                                367.00000000000006
+                            ],
+                            "spans": [],
+                            "index": 19
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        369,
+                        503,
+                        400
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                100,
+                                366,
+                                505,
+                                387
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        100,
+                                        366,
+                                        149,
+                                        386
+                                    ],
+                                    "score": 1.0,
+                                    "content": "as b(`+1)i a",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        138,
+                                        367,
+                                        197,
+                                        385
+                                    ],
+                                    "score": 1.0,
+                                    "content": "nd W (`+1)i,j,µ,ν",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        190,
+                                        367,
+                                        313,
+                                        386
+                                    ],
+                                    "score": 1.0,
+                                    "content": "have zero mean, and W (`+1)i,j,ν,µ",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        309,
+                                        371,
+                                        505,
+                                        387
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are independent of the activations at the previous",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                102,
+                                381,
+                                187,
+                                405
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        102,
+                                        381,
+                                        131,
+                                        405
+                                    ],
+                                    "score": 1.0,
+                                    "content": "layer,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        131,
+                                        385,
+                                        179,
+                                        401
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\phi ( A _ { j , \\nu } ^ { ( \\ell ) } ( \\mathbf { X } ) )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        180,
+                                        381,
+                                        187,
+                                        405
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        }
+                    ],
+                    "index": 20.5,
+                    "bbox_fs": [
+                        100,
+                        366,
+                        505,
+                        405
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        403,
+                        506,
+                        489
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                402,
+                                506,
+                                418
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        402,
+                                        506,
+                                        418
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Now we show that it is possible to efficiently compute the covariance function. This is surprising",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                415,
+                                504,
+                                452
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        415,
+                                        241,
+                                        452
+                                    ],
+                                    "score": 1.0,
+                                    "content": "because for many networks, we nlocations in the feature map (i.e.",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        241,
+                                        426,
+                                        356,
+                                        446
+                                    ],
+                                    "score": 0.94,
+                                    "content": "\\mathbb { C } \\left[ A _ { i , \\mu } ^ { ( \\ell + 1 ) } ( \\mathbf { \\bar { X } } ) , A _ { i , \\mu ^ { \\prime } } ^ { ( \\ell + 1 ) } ( \\mathbf { X } ^ { \\prime } ) \\right]",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        356,
+                                        415,
+                                        372,
+                                        452
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ce ofor",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        372,
+                                        428,
+                                        504,
+                                        442
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\mu , \\mu ^ { \\prime } \\in \\{ 1 , \\ldots , H ^ { ( \\ell + 1 ) } D ^ { ( \\ell + 1 ) } \\} )",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                442,
+                                507,
+                                460
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        442,
+                                        295,
+                                        460
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and this object is extremely high-dimensional,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        295,
+                                        444,
+                                        380,
+                                        459
+                                    ],
+                                    "score": 0.89,
+                                    "content": "N ^ { 2 } ( H ^ { ( \\ell + 1 ) } D ^ { \\overline { { ( \\ell + 1 ) } } } ) ^ { 2 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        380,
+                                        442,
+                                        507,
+                                        460
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". However, it turns out that we",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                451,
+                                506,
+                                483
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        451,
+                                        375,
+                                        483
+                                    ],
+                                    "score": 1.0,
+                                    "content": "only need to consider the “diagonal” covariance, (i.e. we only need",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        376,
+                                        457,
+                                        489,
+                                        477
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\mathbb { C } \\left[ A _ { i , \\mu } ^ { ( \\ell + 1 ) } ( \\mathbf { X } ) , A _ { i , \\mu } ^ { ( \\ell + 1 ) } ( \\mathbf { X } ^ { \\prime } ) \\right]",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        490,
+                                        451,
+                                        506,
+                                        483
+                                    ],
+                                    "score": 1.0,
+                                    "content": "fo r",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                474,
+                                493,
+                                491
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        475,
+                                        223,
+                                        489
+                                    ],
+                                    "score": 0.89,
+                                    "content": "\\mu \\in \\{ 1 , \\ldots , H ^ { ( \\ell + 1 ) } D ^ { ( \\ell + 1 ) } \\} )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        224,
+                                        474,
+                                        408,
+                                        491
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", which is a more manageable quantity of size",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        408,
+                                        476,
+                                        488,
+                                        489
+                                    ],
+                                    "score": 0.88,
+                                    "content": "N ^ { 2 } ( H ^ { ( \\ell + 1 ) } D ^ { ( \\ell + 1 ) } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        488,
+                                        474,
+                                        493,
+                                        491
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        }
+                    ],
+                    "index": 24,
+                    "bbox_fs": [
+                        104,
+                        402,
+                        507,
+                        491
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        493,
+                        505,
+                        562
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                493,
+                                505,
+                                507
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        493,
+                                        228,
+                                        507
+                                    ],
+                                    "score": 1.0,
+                                    "content": "This is true at the output layer",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        228,
+                                        494,
+                                        259,
+                                        506
+                                    ],
+                                    "score": 0.91,
+                                    "content": "( L + 1 )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        259,
+                                        493,
+                                        505,
+                                        507
+                                    ],
+                                    "score": 1.0,
+                                    "content": ": in order to achieve an output suitable for classification or re-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                504,
+                                506,
+                                519
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        504,
+                                        290,
+                                        519
+                                    ],
+                                    "score": 1.0,
+                                    "content": "gression, we use only a single output location",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        290,
+                                        505,
+                                        385,
+                                        517
+                                    ],
+                                    "score": 0.92,
+                                    "content": "H ^ { ( L + 1 ) } = D ^ { ( L + 1 ) } = 1",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        385,
+                                        504,
+                                        506,
+                                        519
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", with a number of “channels”",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                517,
+                                506,
+                                530
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        517,
+                                        506,
+                                        530
+                                    ],
+                                    "score": 1.0,
+                                    "content": "equal to the number of of outputs/classes, so it is only possible to compute the covariance at that",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                529,
+                                505,
+                                541
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        529,
+                                        505,
+                                        541
+                                    ],
+                                    "score": 1.0,
+                                    "content": "single location. We now show that, if we only need the covariance at corresponding locations in the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                540,
+                                505,
+                                551
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        540,
+                                        505,
+                                        551
+                                    ],
+                                    "score": 1.0,
+                                    "content": "outputs, we only need the covariance at corresponding locations in the inputs, and this requirement",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                551,
+                                283,
+                                562
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        551,
+                                        283,
+                                        562
+                                    ],
+                                    "score": 1.0,
+                                    "content": "propagates backwards through the network.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        }
+                    ],
+                    "index": 29.5,
+                    "bbox_fs": [
+                        105,
+                        493,
+                        506,
+                        562
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        567,
+                        504,
+                        590
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                567,
+                                505,
+                                579
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        567,
+                                        505,
+                                        579
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Formally, as the activations are composed of a sum of terms, their covariance is the sum of the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                577,
+                                275,
+                                591
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        577,
+                                        275,
+                                        591
+                                    ],
+                                    "score": 1.0,
+                                    "content": "covariances of all those underlying terms,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        }
+                    ],
+                    "index": 33.5,
+                    "bbox_fs": [
+                        105,
+                        567,
+                        505,
+                        591
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        136,
+                        590,
+                        477,
+                        651
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                136,
+                                590,
+                                477,
+                                651
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        136,
+                                        590,
+                                        477,
+                                        651
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\begin{array} { r l } { \\mathbb { C } \\left[ A _ { i , \\mu } ^ { ( \\ell + 1 ) } ( \\mathbf { X } ) , A _ { i , \\mu } ^ { ( \\ell + 1 ) } ( \\mathbf { X } ^ { \\prime } ) \\right] = \\mathbb { V } \\left[ b _ { i } ^ { ( \\ell ) } \\right] + } & { } \\\\ { \\displaystyle \\sum _ { j = 1 } ^ { C ^ { ( \\ell ) } } \\sum _ { j ^ { \\prime } = 1 } ^ { C ^ { ( \\ell ) } } \\sum _ { \\nu = 1 } ^ { H ^ { ( \\ell ) } D ^ { ( \\ell ) } } \\sum _ { \\nu ^ { \\prime } = 1 } ^ { H ^ { ( \\ell ) } D ^ { ( \\ell ) } } \\mathbb { C } \\left[ W _ { i , j , \\mu , \\nu } ^ { ( \\ell + 1 ) } \\phi ( A _ { j , \\nu } ^ { ( \\ell ) } ( \\mathbf { X } ) ) , W _ { i , j ^ { \\prime } , \\mu , \\nu ^ { \\prime } } ^ { ( \\ell + 1 ) } \\phi ( A _ { j ^ { \\prime } , \\nu ^ { \\prime } } ^ { ( \\ell ) } ( \\mathbf { X } ^ { \\prime } ) ) \\right] . } \\end{array}",
+                                    "type": "interline_equation",
+                                    "image_path": "69d31cc97d2fc0067ded1890ef2527e60cbaecb75da932e44d6be86c09e39c9e.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 36,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                136,
+                                590,
+                                477,
+                                610.3333333333334
+                            ],
+                            "spans": [],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                136,
+                                610.3333333333334,
+                                477,
+                                630.6666666666667
+                            ],
+                            "spans": [],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                136,
+                                630.6666666666667,
+                                477,
+                                651.0000000000001
+                            ],
+                            "spans": [],
+                            "index": 37
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        652,
+                        507,
+                        674
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                651,
+                                505,
+                                664
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        651,
+                                        505,
+                                        664
+                                    ],
+                                    "score": 1.0,
+                                    "content": "As the terms in the covariance have mean zero, and as the weights and activations from the previous",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                663,
+                                197,
+                                676
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        663,
+                                        197,
+                                        676
+                                    ],
+                                    "score": 1.0,
+                                    "content": "layer are independent,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        }
+                    ],
+                    "index": 38.5,
+                    "bbox_fs": [
+                        105,
+                        651,
+                        505,
+                        676
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        122,
+                        675,
+                        478,
+                        736
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                122,
+                                675,
+                                478,
+                                736
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        122,
+                                        675,
+                                        478,
+                                        736
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\begin{array} { r l } { \\mathbb { C } \\left[ A _ { i , \\mu } ^ { ( \\ell + 1 ) } ( \\mathbf { X } ) , A _ { i , \\mu } ^ { ( \\ell + 1 ) } ( \\mathbf { X } ^ { \\prime } ) \\right] = \\sigma _ { \\mathrm { b } } ^ { 2 } + } & { } \\\\ { \\quad } & { \\qquad \\sum _ { j = 1 } ^ { { \\ell ^ { ( \\ell ) } } } \\displaystyle \\sum _ { j ^ { \\prime } = 1 } ^ { { \\ell ^ { ( \\ell ) } } } \\sum _ { \\nu = 1 } ^ { { H ^ { ( \\ell ) } } D ^ { ( \\ell ) } } \\sum _ { \\nu ^ { \\prime } = 1 } ^ { H ^ { ( \\ell ) } D ^ { ( \\ell ) } } \\mathbb { E } \\left[ W _ { i , j , \\mu , \\nu } ^ { ( \\ell + 1 ) } W _ { i , j ^ { \\prime } , \\mu , \\nu ^ { \\prime } } ^ { ( \\ell + 1 ) } \\right] \\mathbb { E } \\left[ \\phi ( A _ { j , \\nu } ^ { ( \\ell ) } ( \\mathbf { X } ) ) \\phi ( A _ { j ^ { \\prime } , \\nu ^ { \\prime } } ^ { ( \\ell ) } ( \\mathbf { X } ^ { \\prime } ) ) \\right] . } \\end{array}",
+                                    "type": "interline_equation",
+                                    "image_path": "3654c365438b76a29a2541724e15f6857f74e810a656fd4cf794da4018587cad.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 41,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                122,
+                                675,
+                                478,
+                                695.3333333333334
+                            ],
+                            "spans": [],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                122,
+                                695.3333333333334,
+                                478,
+                                715.6666666666667
+                            ],
+                            "spans": [],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                122,
+                                715.6666666666667,
+                                478,
+                                736.0000000000001
+                            ],
+                            "spans": [],
+                            "index": 42
+                        }
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        82,
+                        283,
+                        95
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                107,
+                                81,
+                                283,
+                                97
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        81,
+                                        244,
+                                        97
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Algorithm 1 The ConvNet kernel",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        244,
+                                        83,
+                                        283,
+                                        95
+                                    ],
+                                    "score": 0.91,
+                                    "content": "k ( \\mathbf { X } , \\mathbf { X } ^ { \\prime } )",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 0
+                        }
+                    ],
+                    "index": 0
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        110,
+                        97,
+                        506,
+                        199
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                109,
+                                95,
+                                311,
+                                113
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        109,
+                                        95,
+                                        311,
+                                        113
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1: Input: two images, X, X0 ∈ RC(0)×(H(0)W (0)).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                107,
+                                109,
+                                507,
+                                128
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        109,
+                                        164,
+                                        128
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2: Compute",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        164,
+                                        111,
+                                        214,
+                                        126
+                                    ],
+                                    "score": 0.61,
+                                    "content": "K _ { \\mu } ^ { ( 1 ) } ( { \\bf X } , { \\bf X } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        215,
+                                        109,
+                                        220,
+                                        128
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        221,
+                                        111,
+                                        273,
+                                        126
+                                    ],
+                                    "score": 0.82,
+                                    "content": "K _ { \\mu } ^ { ( 1 ) } ( { \\bf X } , { \\bf X ^ { \\prime } } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        274,
+                                        109,
+                                        298,
+                                        128
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        298,
+                                        111,
+                                        354,
+                                        126
+                                    ],
+                                    "score": 0.92,
+                                    "content": "K _ { \\mu } ^ { ( 1 ) } ( \\mathbf { X } ^ { \\prime } , \\mathbf { X } ^ { \\prime } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        354,
+                                        109,
+                                        373,
+                                        128
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        373,
+                                        112,
+                                        474,
+                                        126
+                                    ],
+                                    "score": 0.9,
+                                    "content": "\\mu \\in \\{ 1 , \\ldots , H ^ { ( 1 ) } D ^ { ( 1 ) } \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        475,
+                                        109,
+                                        507,
+                                        128
+                                    ],
+                                    "score": 1.0,
+                                    "content": "; using",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                121,
+                                123,
+                                163,
+                                137
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        121,
+                                        123,
+                                        163,
+                                        137
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Eq. (10).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                109,
+                                133,
+                                215,
+                                147
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        109,
+                                        133,
+                                        138,
+                                        147
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3: for",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        139,
+                                        136,
+                                        200,
+                                        147
+                                    ],
+                                    "score": 0.88,
+                                    "content": "\\ell = 1 , 2 , \\dots , L",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        201,
+                                        133,
+                                        215,
+                                        147
+                                    ],
+                                    "score": 1.0,
+                                    "content": "do",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                110,
+                                143,
+                                507,
+                                164
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        110,
+                                        148,
+                                        123,
+                                        161
+                                    ],
+                                    "score": 1.0,
+                                    "content": "4:",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        128,
+                                        143,
+                                        174,
+                                        164
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Compute",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        174,
+                                        146,
+                                        225,
+                                        161
+                                    ],
+                                    "score": 0.87,
+                                    "content": "V _ { \\mu } ^ { ( \\ell ) } ( { \\bf X } , { \\bf X ^ { \\prime } } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        226,
+                                        143,
+                                        231,
+                                        164
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        231,
+                                        146,
+                                        282,
+                                        161
+                                    ],
+                                    "score": 0.83,
+                                    "content": "V _ { \\mu } ^ { ( \\ell ) } ( { \\bf X } , { \\bf X ^ { \\prime } } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        283,
+                                        143,
+                                        304,
+                                        164
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        304,
+                                        146,
+                                        356,
+                                        161
+                                    ],
+                                    "score": 0.94,
+                                    "content": "V _ { \\mu } ^ { ( \\ell ) } ( { \\bf X } , { \\bf X ^ { \\prime } } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        356,
+                                        143,
+                                        374,
+                                        164
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        375,
+                                        147,
+                                        474,
+                                        161
+                                    ],
+                                    "score": 0.9,
+                                    "content": "\\mu \\in \\{ 1 , \\ldots , H ^ { ( \\ell ) } D ^ { ( \\ell ) } \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        475,
+                                        143,
+                                        507,
+                                        164
+                                    ],
+                                    "score": 1.0,
+                                    "content": "; using",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                110,
+                                159,
+                                511,
+                                192
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        110,
+                                        173,
+                                        123,
+                                        185
+                                    ],
+                                    "score": 1.0,
+                                    "content": "5:",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        128,
+                                        166,
+                                        182,
+                                        191
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Compute",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        132,
+                                        159,
+                                        281,
+                                        172
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Eq. (13), or some other nonlinearity.",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        183,
+                                        171,
+                                        242,
+                                        185
+                                    ],
+                                    "score": 0.84,
+                                    "content": "K _ { \\mu } ^ { ( \\ell + 1 ) } ( { \\mathbf { X } } , { \\mathbf { X } } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        243,
+                                        166,
+                                        252,
+                                        191
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        252,
+                                        164,
+                                        258,
+                                        192
+                                    ],
+                                    "score": 0.0,
+                                    "content": "",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        259,
+                                        170,
+                                        322,
+                                        186
+                                    ],
+                                    "score": 0.67,
+                                    "content": "K _ { \\mu } ^ { ( \\ell + 1 ) } ( { \\bf X } , { \\bf X ^ { \\prime } } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        322,
+                                        164,
+                                        365,
+                                        192
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        366,
+                                        170,
+                                        432,
+                                        186
+                                    ],
+                                    "score": 0.31,
+                                    "content": "K _ { \\mu } ^ { ( \\ell + 1 ) } ( { \\mathbf { X } } ^ { \\prime } , { \\mathbf { X } } ^ { \\prime } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        433,
+                                        164,
+                                        467,
+                                        192
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        468,
+                                        175,
+                                        476,
+                                        186
+                                    ],
+                                    "score": 0.29,
+                                    "content": "\\mu",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        476,
+                                        164,
+                                        511,
+                                        192
+                                    ],
+                                    "score": 1.0,
+                                    "content": "∈",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                134,
+                                182,
+                                297,
+                                201
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        134,
+                                        185,
+                                        230,
+                                        198
+                                    ],
+                                    "score": 0.89,
+                                    "content": "\\{ 1 , \\ldots , H ^ { ( \\ell + 1 ) } D ^ { ( \\ell + 1 ) } \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        230,
+                                        182,
+                                        297,
+                                        201
+                                    ],
+                                    "score": 1.0,
+                                    "content": "; using Eq. (11).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        }
+                    ],
+                    "index": 4
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        110,
+                        198,
+                        262,
+                        220
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                110,
+                                196,
+                                157,
+                                208
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        110,
+                                        196,
+                                        157,
+                                        208
+                                    ],
+                                    "score": 1.0,
+                                    "content": "6: end for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                108,
+                                202,
+                                264,
+                                222
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        108,
+                                        202,
+                                        194,
+                                        222
+                                    ],
+                                    "score": 1.0,
+                                    "content": "7: Output the scalar",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        194,
+                                        205,
+                                        259,
+                                        220
+                                    ],
+                                    "score": 0.91,
+                                    "content": "K _ { 1 } ^ { ( L + 1 ) } ( { \\bf X } , { \\bf X ^ { \\prime } } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        259,
+                                        202,
+                                        264,
+                                        222
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        }
+                    ],
+                    "index": 8.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        242,
+                        506,
+                        308
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                219,
+                                241,
+                                506,
+                                284
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        219,
+                                        241,
+                                        235,
+                                        284
+                                    ],
+                                    "score": 1.0,
+                                    "content": "nt ffor",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        267,
+                                        241,
+                                        323,
+                                        284
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ent channels: Further, each",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        323,
+                                        242,
+                                        357,
+                                        259
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\mathbf { W } _ { i , j } ^ { ( \\ell + 1 ) }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        357,
+                                        241,
+                                        377,
+                                        284
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and of the",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        377,
+                                        242,
+                                        411,
+                                        260
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\mathbf { W } _ { i , j ^ { \\prime } } ^ { ( \\ell + 1 ) }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        411,
+                                        241,
+                                        448,
+                                        284
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are iid f matrices",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        458,
+                                        245,
+                                        488,
+                                        258
+                                    ],
+                                    "score": 0.91,
+                                    "content": "j \\neq j ^ { \\prime }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        489,
+                                        241,
+                                        506,
+                                        284
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", sonly",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                258,
+                                506,
+                                304
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        279,
+                                        311,
+                                        294
+                                    ],
+                                    "score": 1.0,
+                                    "content": "contains independent variables or zeros (Fig. 1), so",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        258,
+                                        218,
+                                        279
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\mathbb { E } \\left[ W _ { i , j , \\mu , \\nu } ^ { ( \\ell + 1 ) } W _ { i , j ^ { \\prime } , \\mu , \\nu ^ { \\prime } } ^ { ( \\ell + 1 ) } \\right] = 0",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        236,
+                                        262,
+                                        266,
+                                        274
+                                    ],
+                                    "score": 0.9,
+                                    "content": "j \\neq j ^ { \\prime }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        311,
+                                        277,
+                                        419,
+                                        297
+                                    ],
+                                    "score": 0.94,
+                                    "content": "\\mathbb { E } \\left[ W _ { i , j , \\mu , \\nu } ^ { ( \\ell + 1 ) } W _ { i , j ^ { \\prime } , \\mu , \\nu ^ { \\prime } } ^ { ( \\ell + 1 ) } \\right] = 0",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        344,
+                                        264,
+                                        352,
+                                        274
+                                    ],
+                                    "score": 0.79,
+                                    "content": "\\mu",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        419,
+                                        267,
+                                        433,
+                                        304
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        434,
+                                        281,
+                                        462,
+                                        293
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\nu \\neq \\nu ^ { \\prime }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        449,
+                                        259,
+                                        482,
+                                        276
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\mathbf { W } _ { i , j } ^ { ( \\ell + 1 ) }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        462,
+                                        267,
+                                        506,
+                                        304
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Thus, we",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                296,
+                                260,
+                                308
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        296,
+                                        219,
+                                        307
+                                    ],
+                                    "score": 1.0,
+                                    "content": "can eliminate the sums over",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        220,
+                                        296,
+                                        229,
+                                        308
+                                    ],
+                                    "score": 0.87,
+                                    "content": "j ^ { \\prime }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        229,
+                                        296,
+                                        246,
+                                        307
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        247,
+                                        296,
+                                        256,
+                                        306
+                                    ],
+                                    "score": 0.85,
+                                    "content": "\\nu ^ { \\prime }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        256,
+                                        296,
+                                        260,
+                                        307
+                                    ],
+                                    "score": 1.0,
+                                    "content": ":",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        }
+                    ],
+                    "index": 11
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        111,
+                        313,
+                        510,
+                        351
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                111,
+                                313,
+                                510,
+                                351
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        111,
+                                        313,
+                                        510,
+                                        351
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\Sigma \\left[ A _ { i , \\mu } ^ { ( \\ell + 1 ) } ( \\mathbf { X } ) , A _ { i , \\mu } ^ { ( \\ell + 1 ) } ( \\mathbf { X ^ { \\prime } } ) \\right] = \\sigma _ { \\mathrm { b } } ^ { ( \\ell ) } + \\sum _ { j = 1 } ^ { Q ^ { ( \\ell ) } } \\sum _ { \\nu = 1 } ^ { H ^ { ( \\ell ) } D ^ { ( \\ell ) } } \\mathbb { E } \\left[ W _ { i , j , \\mu , \\nu } ^ { ( \\ell + 1 ) } W _ { i , j , \\mu , \\nu } ^ { ( \\ell + 1 ) } \\right] \\mathbb { E } \\left[ \\phi ( A _ { j , \\nu } ^ { ( \\ell ) } ( \\mathbf { X } ) ) \\phi ( A _ { j , \\nu } ^ { ( \\ell ) } ( \\mathbf { X ^ { \\prime } } ) ) \\right] .",
+                                    "type": "interline_equation",
+                                    "image_path": "24bda7afff24b06f4b1c3e39fbc4a4f4d5ba9a86f7c69a2de5b9bc08f7b317b8.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 14,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                111,
+                                313,
+                                510,
+                                325.6666666666667
+                            ],
+                            "spans": [],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                111,
+                                325.6666666666667,
+                                510,
+                                338.33333333333337
+                            ],
+                            "spans": [],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                111,
+                                338.33333333333337,
+                                510,
+                                351.00000000000006
+                            ],
+                            "spans": [],
+                            "index": 15
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        365,
+                        505,
+                        406
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                103,
+                                363,
+                                505,
+                                383
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        103,
+                                        363,
+                                        207,
+                                        383
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The µth row of W(`+1)i,j i",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        202,
+                                        366,
+                                        276,
+                                        381
+                                    ],
+                                    "score": 1.0,
+                                    "content": "s zero for indices",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        276,
+                                        370,
+                                        283,
+                                        377
+                                    ],
+                                    "score": 0.78,
+                                    "content": "\\nu",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        284,
+                                        366,
+                                        505,
+                                        381
+                                    ],
+                                    "score": 1.0,
+                                    "content": "that do not belong to its convolutional patch, so we can",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                380,
+                                506,
+                                398
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        381,
+                                        190,
+                                        398
+                                    ],
+                                    "score": 1.0,
+                                    "content": "restrict the sum over",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        190,
+                                        385,
+                                        197,
+                                        393
+                                    ],
+                                    "score": 0.74,
+                                    "content": "\\nu",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        197,
+                                        381,
+                                        317,
+                                        398
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to that region. We also define",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        317,
+                                        380,
+                                        366,
+                                        396
+                                    ],
+                                    "score": 0.93,
+                                    "content": "v _ { g } ^ { ( 1 ) } ( \\mathbf { X } , \\mathbf { X } ^ { \\prime } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        366,
+                                        381,
+                                        506,
+                                        398
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", to emphasise that the covariances",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                394,
+                                401,
+                                407
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        394,
+                                        262,
+                                        407
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are independent of the output channel,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        262,
+                                        396,
+                                        267,
+                                        406
+                                    ],
+                                    "score": 0.8,
+                                    "content": "j",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        268,
+                                        394,
+                                        401,
+                                        407
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". The variance of the first layer is",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        }
+                    ],
+                    "index": 17
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        151,
+                        411,
+                        471,
+                        449
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                151,
+                                411,
+                                471,
+                                449
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        151,
+                                        411,
+                                        471,
+                                        449
+                                    ],
+                                    "score": 0.94,
+                                    "content": "K _ { \\mu } ^ { ( 1 ) } ( \\mathbf { X } , \\mathbf { X ^ { \\prime } } ) = \\mathbb { C } \\left[ A _ { i , \\mu } ^ { ( 1 ) } ( \\mathbf { X } ) , A _ { i , \\mu } ^ { ( 1 ) } ( \\mathbf { X ^ { \\prime } } ) \\right] = \\sigma _ { \\mathrm { b } } ^ { 2 } + \\frac { \\sigma _ { \\mathrm { w } } ^ { 2 } } { C ^ { ( 0 ) } } \\sum _ { i = 1 } ^ { C ^ { ( 0 ) } } \\sum _ { \\nu \\in \\mu \\mathrm { t h } \\mathrm { p a t c h } } X _ { i , \\nu } X _ { i , \\nu } ^ { \\prime } .",
+                                    "type": "interline_equation",
+                                    "image_path": "a04654dd8ea70692c62a0660905bb02e1a41cd6a0d55334a6d1907128435ddcf.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 20,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                151,
+                                411,
+                                471,
+                                423.6666666666667
+                            ],
+                            "spans": [],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                151,
+                                423.6666666666667,
+                                471,
+                                436.33333333333337
+                            ],
+                            "spans": [],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                151,
+                                436.33333333333337,
+                                471,
+                                449.00000000000006
+                            ],
+                            "spans": [],
+                            "index": 21
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        457,
+                        270,
+                        469
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                455,
+                                271,
+                                471
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        455,
+                                        271,
+                                        471
+                                    ],
+                                    "score": 1.0,
+                                    "content": "And we do the same for the other layers,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        }
+                    ],
+                    "index": 22
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        123,
+                        477,
+                        455,
+                        506
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                123,
+                                477,
+                                455,
+                                506
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        123,
+                                        477,
+                                        455,
+                                        506
+                                    ],
+                                    "score": 0.92,
+                                    "content": "K _ { \\mu } ^ { ( \\ell + 1 ) } ( \\mathbf { X } , \\mathbf { X } ^ { \\prime } ) = \\mathbb { C } \\left[ A _ { i , \\mu } ^ { ( \\ell + 1 ) } ( \\mathbf { X } ) , A _ { i , \\mu } ^ { ( \\ell + 1 ) } ( \\mathbf { X } ^ { \\prime } ) \\right] = \\sigma _ { \\mathrm { b } } ^ { 2 } + \\sigma _ { \\mathrm { w } } ^ { 2 } \\sum _ { \\nu \\in \\mu \\mathrm { t h p a t c h } } V _ { \\nu } ^ { ( \\ell ) } ( \\mathbf { X } , \\mathbf { X } ^ { \\prime } ) ,",
+                                    "type": "interline_equation",
+                                    "image_path": "549433f5673785abcc4cef3a361b3221d508b66c2586e63790f0e4e64aa1c2fe.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 23,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                123,
+                                477,
+                                455,
+                                506
+                            ],
+                            "spans": [],
+                            "index": 23
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        512,
+                        133,
+                        523
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                510,
+                                135,
+                                524
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        510,
+                                        135,
+                                        524
+                                    ],
+                                    "score": 1.0,
+                                    "content": "where",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        }
+                    ],
+                    "index": 24
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        214,
+                        520,
+                        397,
+                        542
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                214,
+                                520,
+                                397,
+                                542
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        214,
+                                        520,
+                                        397,
+                                        542
+                                    ],
+                                    "score": 0.94,
+                                    "content": "V _ { \\nu } ^ { ( \\ell ) } ( \\mathbf { X } , \\mathbf { X } ^ { \\prime } ) = \\mathbb { E } \\left[ \\phi ( A _ { j , \\nu } ^ { ( \\ell ) } ( \\mathbf { X } ) ) \\phi ( A _ { j , \\nu } ^ { ( \\ell ) } ( \\mathbf { X } ^ { \\prime } ) ) \\right]",
+                                    "type": "interline_equation",
+                                    "image_path": "f76318a4bc7d12e24edd6391255fdb84c2a51867f3242d20b9177e93df8867a9.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 25,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                214,
+                                520,
+                                397,
+                                542
+                            ],
+                            "spans": [],
+                            "index": 25
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        543,
+                        420,
+                        556
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                543,
+                                421,
+                                557
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        543,
+                                        421,
+                                        557
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is the covariance of the activations, which is again independent of the channel.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        }
+                    ],
+                    "index": 26
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        569,
+                        273,
+                        581
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                569,
+                                274,
+                                582
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        569,
+                                        274,
+                                        582
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3.2 COVARIANCE OF THE ACTIVITIES",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        }
+                    ],
+                    "index": 27
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        589,
+                        506,
+                        638
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                590,
+                                505,
+                                602
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        590,
+                                        505,
+                                        602
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The elementwise covariance in the right-hand side of Eq. (11) can be computed in closed form",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                602,
+                                505,
+                                613
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        602,
+                                        193,
+                                        613
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for many choices of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        193,
+                                        602,
+                                        200,
+                                        613
+                                    ],
+                                    "score": 0.84,
+                                    "content": "\\phi",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        201,
+                                        602,
+                                        505,
+                                        613
+                                    ],
+                                    "score": 1.0,
+                                    "content": "if the activations are Gaussian. For each element of the activations, one",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                612,
+                                505,
+                                625
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        612,
+                                        505,
+                                        625
+                                    ],
+                                    "score": 1.0,
+                                    "content": "needs to keep track of the 3 distinct entries of the bivariate covariance matrix between the inputs,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                619,
+                                324,
+                                642
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        623,
+                                        233,
+                                        638
+                                    ],
+                                    "score": 0.73,
+                                    "content": "K _ { \\mu } ^ { ( \\ell + 1 ) } ( { \\mathbf { X } } , { \\mathbf { X } } ) , K _ { \\mu } ^ { ( \\ell + 1 ) } ( { \\mathbf { X } } , { \\mathbf { X } } ^ { \\prime } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        234,
+                                        619,
+                                        252,
+                                        642
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        252,
+                                        623,
+                                        317,
+                                        638
+                                    ],
+                                    "score": 0.93,
+                                    "content": "K _ { \\mu } ^ { ( \\ell + 1 ) } ( { \\mathbf { X } } ^ { \\prime } , { \\mathbf { X } } ^ { \\prime } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        317,
+                                        619,
+                                        324,
+                                        642
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        }
+                    ],
+                    "index": 29.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        642,
+                        506,
+                        665
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                642,
+                                505,
+                                655
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        642,
+                                        271,
+                                        655
+                                    ],
+                                    "score": 1.0,
+                                    "content": "For example, for the ReLU nonlinearity",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        271,
+                                        642,
+                                        352,
+                                        655
+                                    ],
+                                    "score": 0.9,
+                                    "content": "( \\phi ( x ) = \\operatorname* { m a x } ( 0 , x ) )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        352,
+                                        642,
+                                        505,
+                                        655
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", one can adapt Cho & Saul (2009) in",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                653,
+                                350,
+                                666
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        653,
+                                        350,
+                                        666
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the same way as Matthews et al. (2018a, section 3) to obtain",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        }
+                    ],
+                    "index": 32.5
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        151,
+                        671,
+                        459,
+                        704
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                151,
+                                671,
+                                459,
+                                704
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        151,
+                                        671,
+                                        459,
+                                        704
+                                    ],
+                                    "score": 0.87,
+                                    "content": "V _ { \\nu } ^ { ( \\ell ) } ( { \\mathbf { X } } , { \\mathbf { X ^ { \\prime } } } ) = \\frac { \\sqrt { K _ { \\nu } ^ { ( \\ell ) } ( { \\mathbf { X } } , { \\mathbf { X } } ) K _ { \\nu } ^ { ( \\ell ) } ( { \\mathbf { X ^ { \\prime } } } , { \\mathbf { X ^ { \\prime } } } ) } } { \\pi } \\left( \\sin \\theta _ { \\nu } ^ { ( \\ell ) } + ( \\pi - \\theta _ { \\nu } ^ { ( \\ell ) } ) \\cos \\theta _ { \\nu } ^ { ( \\ell ) } \\right)",
+                                    "type": "interline_equation",
+                                    "image_path": "f0970a595d7eac2c5d1aee2436da987cea6ee155b47b22a44fa6ced9b03dab1d.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 35,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                151,
+                                671,
+                                459,
+                                682.0
+                            ],
+                            "spans": [],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                151,
+                                682.0,
+                                459,
+                                693.0
+                            ],
+                            "spans": [],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                151,
+                                693.0,
+                                459,
+                                704.0
+                            ],
+                            "spans": [],
+                            "index": 36
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        709,
+                        376,
+                        735
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                100,
+                                703,
+                                404,
+                                739
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        100,
+                                        703,
+                                        133,
+                                        739
+                                    ],
+                                    "score": 1.0,
+                                    "content": "where",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        133,
+                                        709,
+                                        404,
+                                        736
+                                    ],
+                                    "score": 0.69,
+                                    "content": "\\theta _ { \\nu } ^ { ( \\ell ) } = \\cos ^ { - 1 } \\bigg ( K _ { \\nu } ^ { ( \\ell ) } ( { \\mathbf X } , { \\mathbf X ^ { \\prime } } ) / \\sqrt { K _ { \\nu } ^ { ( \\ell ) } ( { \\mathbf X } , { \\mathbf X } ) K _ { \\nu } ^ { ( \\ell ) } ( { \\mathbf X ^ { \\prime } } , { \\mathbf X ^ { \\prime } } ) } \\bigg ) .",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 37
+                        }
+                    ],
+                    "index": 37
+                }
+            ],
+            "page_idx": 5,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        107,
+                        26,
+                        293,
+                        38
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                25,
+                                294,
+                                39
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        25,
+                                        294,
+                                        39
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Published as a conference paper at ICLR 2019",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        302,
+                        751,
+                        309,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                302,
+                                751,
+                                309,
+                                762
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        302,
+                                        751,
+                                        309,
+                                        762
+                                    ],
+                                    "score": 1.0,
+                                    "content": "6",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        82,
+                        283,
+                        95
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                107,
+                                81,
+                                283,
+                                97
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        81,
+                                        244,
+                                        97
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Algorithm 1 The ConvNet kernel",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        244,
+                                        83,
+                                        283,
+                                        95
+                                    ],
+                                    "score": 0.91,
+                                    "content": "k ( \\mathbf { X } , \\mathbf { X } ^ { \\prime } )",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 0
+                        }
+                    ],
+                    "index": 0
+                },
+                {
+                    "type": "list",
+                    "bbox": [
+                        110,
+                        97,
+                        506,
+                        199
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                109,
+                                95,
+                                311,
+                                113
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        109,
+                                        95,
+                                        311,
+                                        113
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1: Input: two images, X, X0 ∈ RC(0)×(H(0)W (0)).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1,
+                            "is_list_start_line": true,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                107,
+                                109,
+                                507,
+                                128
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        109,
+                                        164,
+                                        128
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2: Compute",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        164,
+                                        111,
+                                        214,
+                                        126
+                                    ],
+                                    "score": 0.61,
+                                    "content": "K _ { \\mu } ^ { ( 1 ) } ( { \\bf X } , { \\bf X } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        215,
+                                        109,
+                                        220,
+                                        128
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        221,
+                                        111,
+                                        273,
+                                        126
+                                    ],
+                                    "score": 0.82,
+                                    "content": "K _ { \\mu } ^ { ( 1 ) } ( { \\bf X } , { \\bf X ^ { \\prime } } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        274,
+                                        109,
+                                        298,
+                                        128
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        298,
+                                        111,
+                                        354,
+                                        126
+                                    ],
+                                    "score": 0.92,
+                                    "content": "K _ { \\mu } ^ { ( 1 ) } ( \\mathbf { X } ^ { \\prime } , \\mathbf { X } ^ { \\prime } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        354,
+                                        109,
+                                        373,
+                                        128
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        373,
+                                        112,
+                                        474,
+                                        126
+                                    ],
+                                    "score": 0.9,
+                                    "content": "\\mu \\in \\{ 1 , \\ldots , H ^ { ( 1 ) } D ^ { ( 1 ) } \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        475,
+                                        109,
+                                        507,
+                                        128
+                                    ],
+                                    "score": 1.0,
+                                    "content": "; using",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                121,
+                                123,
+                                163,
+                                137
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        121,
+                                        123,
+                                        163,
+                                        137
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Eq. (10).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                109,
+                                133,
+                                215,
+                                147
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        109,
+                                        133,
+                                        138,
+                                        147
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3: for",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        139,
+                                        136,
+                                        200,
+                                        147
+                                    ],
+                                    "score": 0.88,
+                                    "content": "\\ell = 1 , 2 , \\dots , L",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        201,
+                                        133,
+                                        215,
+                                        147
+                                    ],
+                                    "score": 1.0,
+                                    "content": "do",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4,
+                            "is_list_start_line": true,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                110,
+                                143,
+                                507,
+                                164
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        110,
+                                        148,
+                                        123,
+                                        161
+                                    ],
+                                    "score": 1.0,
+                                    "content": "4:",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        128,
+                                        143,
+                                        174,
+                                        164
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Compute",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        174,
+                                        146,
+                                        225,
+                                        161
+                                    ],
+                                    "score": 0.87,
+                                    "content": "V _ { \\mu } ^ { ( \\ell ) } ( { \\bf X } , { \\bf X ^ { \\prime } } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        226,
+                                        143,
+                                        231,
+                                        164
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        231,
+                                        146,
+                                        282,
+                                        161
+                                    ],
+                                    "score": 0.83,
+                                    "content": "V _ { \\mu } ^ { ( \\ell ) } ( { \\bf X } , { \\bf X ^ { \\prime } } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        283,
+                                        143,
+                                        304,
+                                        164
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        304,
+                                        146,
+                                        356,
+                                        161
+                                    ],
+                                    "score": 0.94,
+                                    "content": "V _ { \\mu } ^ { ( \\ell ) } ( { \\bf X } , { \\bf X ^ { \\prime } } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        356,
+                                        143,
+                                        374,
+                                        164
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        375,
+                                        147,
+                                        474,
+                                        161
+                                    ],
+                                    "score": 0.9,
+                                    "content": "\\mu \\in \\{ 1 , \\ldots , H ^ { ( \\ell ) } D ^ { ( \\ell ) } \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        475,
+                                        143,
+                                        507,
+                                        164
+                                    ],
+                                    "score": 1.0,
+                                    "content": "; using",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                110,
+                                159,
+                                511,
+                                192
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        110,
+                                        173,
+                                        123,
+                                        185
+                                    ],
+                                    "score": 1.0,
+                                    "content": "5:",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        128,
+                                        166,
+                                        182,
+                                        191
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Compute",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        132,
+                                        159,
+                                        281,
+                                        172
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Eq. (13), or some other nonlinearity.",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        183,
+                                        171,
+                                        242,
+                                        185
+                                    ],
+                                    "score": 0.84,
+                                    "content": "K _ { \\mu } ^ { ( \\ell + 1 ) } ( { \\mathbf { X } } , { \\mathbf { X } } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        243,
+                                        166,
+                                        252,
+                                        191
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        252,
+                                        164,
+                                        258,
+                                        192
+                                    ],
+                                    "score": 0.0,
+                                    "content": "",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        259,
+                                        170,
+                                        322,
+                                        186
+                                    ],
+                                    "score": 0.67,
+                                    "content": "K _ { \\mu } ^ { ( \\ell + 1 ) } ( { \\bf X } , { \\bf X ^ { \\prime } } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        322,
+                                        164,
+                                        365,
+                                        192
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        366,
+                                        170,
+                                        432,
+                                        186
+                                    ],
+                                    "score": 0.31,
+                                    "content": "K _ { \\mu } ^ { ( \\ell + 1 ) } ( { \\mathbf { X } } ^ { \\prime } , { \\mathbf { X } } ^ { \\prime } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        433,
+                                        164,
+                                        467,
+                                        192
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        468,
+                                        175,
+                                        476,
+                                        186
+                                    ],
+                                    "score": 0.29,
+                                    "content": "\\mu",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        476,
+                                        164,
+                                        511,
+                                        192
+                                    ],
+                                    "score": 1.0,
+                                    "content": "∈",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                134,
+                                182,
+                                297,
+                                201
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        134,
+                                        185,
+                                        230,
+                                        198
+                                    ],
+                                    "score": 0.89,
+                                    "content": "\\{ 1 , \\ldots , H ^ { ( \\ell + 1 ) } D ^ { ( \\ell + 1 ) } \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        230,
+                                        182,
+                                        297,
+                                        201
+                                    ],
+                                    "score": 1.0,
+                                    "content": "; using Eq. (11).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7,
+                            "is_list_end_line": true
+                        }
+                    ],
+                    "index": 4,
+                    "bbox_fs": [
+                        107,
+                        95,
+                        511,
+                        201
+                    ]
+                },
+                {
+                    "type": "index",
+                    "bbox": [
+                        110,
+                        198,
+                        262,
+                        220
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                110,
+                                196,
+                                157,
+                                208
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        110,
+                                        196,
+                                        157,
+                                        208
+                                    ],
+                                    "score": 1.0,
+                                    "content": "6: end for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                108,
+                                202,
+                                264,
+                                222
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        108,
+                                        202,
+                                        194,
+                                        222
+                                    ],
+                                    "score": 1.0,
+                                    "content": "7: Output the scalar",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        194,
+                                        205,
+                                        259,
+                                        220
+                                    ],
+                                    "score": 0.91,
+                                    "content": "K _ { 1 } ^ { ( L + 1 ) } ( { \\bf X } , { \\bf X ^ { \\prime } } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        259,
+                                        202,
+                                        264,
+                                        222
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9,
+                            "is_list_start_line": true
+                        }
+                    ],
+                    "index": 8.5,
+                    "bbox_fs": [
+                        108,
+                        196,
+                        264,
+                        222
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        242,
+                        506,
+                        308
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                219,
+                                241,
+                                506,
+                                284
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        219,
+                                        241,
+                                        235,
+                                        284
+                                    ],
+                                    "score": 1.0,
+                                    "content": "nt ffor",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        267,
+                                        241,
+                                        323,
+                                        284
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ent channels: Further, each",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        323,
+                                        242,
+                                        357,
+                                        259
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\mathbf { W } _ { i , j } ^ { ( \\ell + 1 ) }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        357,
+                                        241,
+                                        377,
+                                        284
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and of the",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        377,
+                                        242,
+                                        411,
+                                        260
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\mathbf { W } _ { i , j ^ { \\prime } } ^ { ( \\ell + 1 ) }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        411,
+                                        241,
+                                        448,
+                                        284
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are iid f matrices",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        458,
+                                        245,
+                                        488,
+                                        258
+                                    ],
+                                    "score": 0.91,
+                                    "content": "j \\neq j ^ { \\prime }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        489,
+                                        241,
+                                        506,
+                                        284
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", sonly",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                258,
+                                506,
+                                304
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        279,
+                                        311,
+                                        294
+                                    ],
+                                    "score": 1.0,
+                                    "content": "contains independent variables or zeros (Fig. 1), so",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        258,
+                                        218,
+                                        279
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\mathbb { E } \\left[ W _ { i , j , \\mu , \\nu } ^ { ( \\ell + 1 ) } W _ { i , j ^ { \\prime } , \\mu , \\nu ^ { \\prime } } ^ { ( \\ell + 1 ) } \\right] = 0",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        236,
+                                        262,
+                                        266,
+                                        274
+                                    ],
+                                    "score": 0.9,
+                                    "content": "j \\neq j ^ { \\prime }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        311,
+                                        277,
+                                        419,
+                                        297
+                                    ],
+                                    "score": 0.94,
+                                    "content": "\\mathbb { E } \\left[ W _ { i , j , \\mu , \\nu } ^ { ( \\ell + 1 ) } W _ { i , j ^ { \\prime } , \\mu , \\nu ^ { \\prime } } ^ { ( \\ell + 1 ) } \\right] = 0",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        344,
+                                        264,
+                                        352,
+                                        274
+                                    ],
+                                    "score": 0.79,
+                                    "content": "\\mu",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        419,
+                                        267,
+                                        433,
+                                        304
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        434,
+                                        281,
+                                        462,
+                                        293
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\nu \\neq \\nu ^ { \\prime }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        449,
+                                        259,
+                                        482,
+                                        276
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\mathbf { W } _ { i , j } ^ { ( \\ell + 1 ) }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        462,
+                                        267,
+                                        506,
+                                        304
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Thus, we",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                296,
+                                260,
+                                308
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        296,
+                                        219,
+                                        307
+                                    ],
+                                    "score": 1.0,
+                                    "content": "can eliminate the sums over",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        220,
+                                        296,
+                                        229,
+                                        308
+                                    ],
+                                    "score": 0.87,
+                                    "content": "j ^ { \\prime }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        229,
+                                        296,
+                                        246,
+                                        307
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        247,
+                                        296,
+                                        256,
+                                        306
+                                    ],
+                                    "score": 0.85,
+                                    "content": "\\nu ^ { \\prime }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        256,
+                                        296,
+                                        260,
+                                        307
+                                    ],
+                                    "score": 1.0,
+                                    "content": ":",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        }
+                    ],
+                    "index": 11,
+                    "bbox_fs": [
+                        105,
+                        241,
+                        506,
+                        308
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        111,
+                        313,
+                        510,
+                        351
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                111,
+                                313,
+                                510,
+                                351
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        111,
+                                        313,
+                                        510,
+                                        351
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\Sigma \\left[ A _ { i , \\mu } ^ { ( \\ell + 1 ) } ( \\mathbf { X } ) , A _ { i , \\mu } ^ { ( \\ell + 1 ) } ( \\mathbf { X ^ { \\prime } } ) \\right] = \\sigma _ { \\mathrm { b } } ^ { ( \\ell ) } + \\sum _ { j = 1 } ^ { Q ^ { ( \\ell ) } } \\sum _ { \\nu = 1 } ^ { H ^ { ( \\ell ) } D ^ { ( \\ell ) } } \\mathbb { E } \\left[ W _ { i , j , \\mu , \\nu } ^ { ( \\ell + 1 ) } W _ { i , j , \\mu , \\nu } ^ { ( \\ell + 1 ) } \\right] \\mathbb { E } \\left[ \\phi ( A _ { j , \\nu } ^ { ( \\ell ) } ( \\mathbf { X } ) ) \\phi ( A _ { j , \\nu } ^ { ( \\ell ) } ( \\mathbf { X ^ { \\prime } } ) ) \\right] .",
+                                    "type": "interline_equation",
+                                    "image_path": "24bda7afff24b06f4b1c3e39fbc4a4f4d5ba9a86f7c69a2de5b9bc08f7b317b8.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 14,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                111,
+                                313,
+                                510,
+                                325.6666666666667
+                            ],
+                            "spans": [],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                111,
+                                325.6666666666667,
+                                510,
+                                338.33333333333337
+                            ],
+                            "spans": [],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                111,
+                                338.33333333333337,
+                                510,
+                                351.00000000000006
+                            ],
+                            "spans": [],
+                            "index": 15
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        365,
+                        505,
+                        406
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                103,
+                                363,
+                                505,
+                                383
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        103,
+                                        363,
+                                        207,
+                                        383
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The µth row of W(`+1)i,j i",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        202,
+                                        366,
+                                        276,
+                                        381
+                                    ],
+                                    "score": 1.0,
+                                    "content": "s zero for indices",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        276,
+                                        370,
+                                        283,
+                                        377
+                                    ],
+                                    "score": 0.78,
+                                    "content": "\\nu",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        284,
+                                        366,
+                                        505,
+                                        381
+                                    ],
+                                    "score": 1.0,
+                                    "content": "that do not belong to its convolutional patch, so we can",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                380,
+                                506,
+                                398
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        381,
+                                        190,
+                                        398
+                                    ],
+                                    "score": 1.0,
+                                    "content": "restrict the sum over",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        190,
+                                        385,
+                                        197,
+                                        393
+                                    ],
+                                    "score": 0.74,
+                                    "content": "\\nu",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        197,
+                                        381,
+                                        317,
+                                        398
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to that region. We also define",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        317,
+                                        380,
+                                        366,
+                                        396
+                                    ],
+                                    "score": 0.93,
+                                    "content": "v _ { g } ^ { ( 1 ) } ( \\mathbf { X } , \\mathbf { X } ^ { \\prime } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        366,
+                                        381,
+                                        506,
+                                        398
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", to emphasise that the covariances",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                394,
+                                401,
+                                407
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        394,
+                                        262,
+                                        407
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are independent of the output channel,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        262,
+                                        396,
+                                        267,
+                                        406
+                                    ],
+                                    "score": 0.8,
+                                    "content": "j",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        268,
+                                        394,
+                                        401,
+                                        407
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". The variance of the first layer is",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        }
+                    ],
+                    "index": 17,
+                    "bbox_fs": [
+                        103,
+                        363,
+                        506,
+                        407
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        151,
+                        411,
+                        471,
+                        449
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                151,
+                                411,
+                                471,
+                                449
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        151,
+                                        411,
+                                        471,
+                                        449
+                                    ],
+                                    "score": 0.94,
+                                    "content": "K _ { \\mu } ^ { ( 1 ) } ( \\mathbf { X } , \\mathbf { X ^ { \\prime } } ) = \\mathbb { C } \\left[ A _ { i , \\mu } ^ { ( 1 ) } ( \\mathbf { X } ) , A _ { i , \\mu } ^ { ( 1 ) } ( \\mathbf { X ^ { \\prime } } ) \\right] = \\sigma _ { \\mathrm { b } } ^ { 2 } + \\frac { \\sigma _ { \\mathrm { w } } ^ { 2 } } { C ^ { ( 0 ) } } \\sum _ { i = 1 } ^ { C ^ { ( 0 ) } } \\sum _ { \\nu \\in \\mu \\mathrm { t h } \\mathrm { p a t c h } } X _ { i , \\nu } X _ { i , \\nu } ^ { \\prime } .",
+                                    "type": "interline_equation",
+                                    "image_path": "a04654dd8ea70692c62a0660905bb02e1a41cd6a0d55334a6d1907128435ddcf.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 20,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                151,
+                                411,
+                                471,
+                                423.6666666666667
+                            ],
+                            "spans": [],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                151,
+                                423.6666666666667,
+                                471,
+                                436.33333333333337
+                            ],
+                            "spans": [],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                151,
+                                436.33333333333337,
+                                471,
+                                449.00000000000006
+                            ],
+                            "spans": [],
+                            "index": 21
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        457,
+                        270,
+                        469
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                455,
+                                271,
+                                471
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        455,
+                                        271,
+                                        471
+                                    ],
+                                    "score": 1.0,
+                                    "content": "And we do the same for the other layers,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        }
+                    ],
+                    "index": 22,
+                    "bbox_fs": [
+                        105,
+                        455,
+                        271,
+                        471
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        123,
+                        477,
+                        455,
+                        506
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                123,
+                                477,
+                                455,
+                                506
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        123,
+                                        477,
+                                        455,
+                                        506
+                                    ],
+                                    "score": 0.92,
+                                    "content": "K _ { \\mu } ^ { ( \\ell + 1 ) } ( \\mathbf { X } , \\mathbf { X } ^ { \\prime } ) = \\mathbb { C } \\left[ A _ { i , \\mu } ^ { ( \\ell + 1 ) } ( \\mathbf { X } ) , A _ { i , \\mu } ^ { ( \\ell + 1 ) } ( \\mathbf { X } ^ { \\prime } ) \\right] = \\sigma _ { \\mathrm { b } } ^ { 2 } + \\sigma _ { \\mathrm { w } } ^ { 2 } \\sum _ { \\nu \\in \\mu \\mathrm { t h p a t c h } } V _ { \\nu } ^ { ( \\ell ) } ( \\mathbf { X } , \\mathbf { X } ^ { \\prime } ) ,",
+                                    "type": "interline_equation",
+                                    "image_path": "549433f5673785abcc4cef3a361b3221d508b66c2586e63790f0e4e64aa1c2fe.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 23,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                123,
+                                477,
+                                455,
+                                506
+                            ],
+                            "spans": [],
+                            "index": 23
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        512,
+                        133,
+                        523
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                510,
+                                135,
+                                524
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        510,
+                                        135,
+                                        524
+                                    ],
+                                    "score": 1.0,
+                                    "content": "where",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        }
+                    ],
+                    "index": 24,
+                    "bbox_fs": [
+                        106,
+                        510,
+                        135,
+                        524
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        214,
+                        520,
+                        397,
+                        542
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                214,
+                                520,
+                                397,
+                                542
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        214,
+                                        520,
+                                        397,
+                                        542
+                                    ],
+                                    "score": 0.94,
+                                    "content": "V _ { \\nu } ^ { ( \\ell ) } ( \\mathbf { X } , \\mathbf { X } ^ { \\prime } ) = \\mathbb { E } \\left[ \\phi ( A _ { j , \\nu } ^ { ( \\ell ) } ( \\mathbf { X } ) ) \\phi ( A _ { j , \\nu } ^ { ( \\ell ) } ( \\mathbf { X } ^ { \\prime } ) ) \\right]",
+                                    "type": "interline_equation",
+                                    "image_path": "f76318a4bc7d12e24edd6391255fdb84c2a51867f3242d20b9177e93df8867a9.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 25,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                214,
+                                520,
+                                397,
+                                542
+                            ],
+                            "spans": [],
+                            "index": 25
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        543,
+                        420,
+                        556
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                543,
+                                421,
+                                557
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        543,
+                                        421,
+                                        557
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is the covariance of the activations, which is again independent of the channel.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        }
+                    ],
+                    "index": 26,
+                    "bbox_fs": [
+                        105,
+                        543,
+                        421,
+                        557
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        569,
+                        273,
+                        581
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                569,
+                                274,
+                                582
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        569,
+                                        274,
+                                        582
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3.2 COVARIANCE OF THE ACTIVITIES",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        }
+                    ],
+                    "index": 27
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        589,
+                        506,
+                        638
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                590,
+                                505,
+                                602
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        590,
+                                        505,
+                                        602
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The elementwise covariance in the right-hand side of Eq. (11) can be computed in closed form",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                602,
+                                505,
+                                613
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        602,
+                                        193,
+                                        613
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for many choices of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        193,
+                                        602,
+                                        200,
+                                        613
+                                    ],
+                                    "score": 0.84,
+                                    "content": "\\phi",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        201,
+                                        602,
+                                        505,
+                                        613
+                                    ],
+                                    "score": 1.0,
+                                    "content": "if the activations are Gaussian. For each element of the activations, one",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                612,
+                                505,
+                                625
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        612,
+                                        505,
+                                        625
+                                    ],
+                                    "score": 1.0,
+                                    "content": "needs to keep track of the 3 distinct entries of the bivariate covariance matrix between the inputs,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                619,
+                                324,
+                                642
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        623,
+                                        233,
+                                        638
+                                    ],
+                                    "score": 0.73,
+                                    "content": "K _ { \\mu } ^ { ( \\ell + 1 ) } ( { \\mathbf { X } } , { \\mathbf { X } } ) , K _ { \\mu } ^ { ( \\ell + 1 ) } ( { \\mathbf { X } } , { \\mathbf { X } } ^ { \\prime } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        234,
+                                        619,
+                                        252,
+                                        642
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        252,
+                                        623,
+                                        317,
+                                        638
+                                    ],
+                                    "score": 0.93,
+                                    "content": "K _ { \\mu } ^ { ( \\ell + 1 ) } ( { \\mathbf { X } } ^ { \\prime } , { \\mathbf { X } } ^ { \\prime } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        317,
+                                        619,
+                                        324,
+                                        642
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        }
+                    ],
+                    "index": 29.5,
+                    "bbox_fs": [
+                        105,
+                        590,
+                        505,
+                        642
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        642,
+                        506,
+                        665
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                642,
+                                505,
+                                655
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        642,
+                                        271,
+                                        655
+                                    ],
+                                    "score": 1.0,
+                                    "content": "For example, for the ReLU nonlinearity",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        271,
+                                        642,
+                                        352,
+                                        655
+                                    ],
+                                    "score": 0.9,
+                                    "content": "( \\phi ( x ) = \\operatorname* { m a x } ( 0 , x ) )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        352,
+                                        642,
+                                        505,
+                                        655
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", one can adapt Cho & Saul (2009) in",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                653,
+                                350,
+                                666
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        653,
+                                        350,
+                                        666
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the same way as Matthews et al. (2018a, section 3) to obtain",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        }
+                    ],
+                    "index": 32.5,
+                    "bbox_fs": [
+                        106,
+                        642,
+                        505,
+                        666
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        151,
+                        671,
+                        459,
+                        704
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                151,
+                                671,
+                                459,
+                                704
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        151,
+                                        671,
+                                        459,
+                                        704
+                                    ],
+                                    "score": 0.87,
+                                    "content": "V _ { \\nu } ^ { ( \\ell ) } ( { \\mathbf { X } } , { \\mathbf { X ^ { \\prime } } } ) = \\frac { \\sqrt { K _ { \\nu } ^ { ( \\ell ) } ( { \\mathbf { X } } , { \\mathbf { X } } ) K _ { \\nu } ^ { ( \\ell ) } ( { \\mathbf { X ^ { \\prime } } } , { \\mathbf { X ^ { \\prime } } } ) } } { \\pi } \\left( \\sin \\theta _ { \\nu } ^ { ( \\ell ) } + ( \\pi - \\theta _ { \\nu } ^ { ( \\ell ) } ) \\cos \\theta _ { \\nu } ^ { ( \\ell ) } \\right)",
+                                    "type": "interline_equation",
+                                    "image_path": "f0970a595d7eac2c5d1aee2436da987cea6ee155b47b22a44fa6ced9b03dab1d.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 35,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                151,
+                                671,
+                                459,
+                                682.0
+                            ],
+                            "spans": [],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                151,
+                                682.0,
+                                459,
+                                693.0
+                            ],
+                            "spans": [],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                151,
+                                693.0,
+                                459,
+                                704.0
+                            ],
+                            "spans": [],
+                            "index": 36
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        709,
+                        376,
+                        735
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                100,
+                                703,
+                                404,
+                                739
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        100,
+                                        703,
+                                        133,
+                                        739
+                                    ],
+                                    "score": 1.0,
+                                    "content": "where",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        133,
+                                        709,
+                                        404,
+                                        736
+                                    ],
+                                    "score": 0.69,
+                                    "content": "\\theta _ { \\nu } ^ { ( \\ell ) } = \\cos ^ { - 1 } \\bigg ( K _ { \\nu } ^ { ( \\ell ) } ( { \\mathbf X } , { \\mathbf X ^ { \\prime } } ) / \\sqrt { K _ { \\nu } ^ { ( \\ell ) } ( { \\mathbf X } , { \\mathbf X } ) K _ { \\nu } ^ { ( \\ell ) } ( { \\mathbf X ^ { \\prime } } , { \\mathbf X ^ { \\prime } } ) } \\bigg ) .",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 37
+                        }
+                    ],
+                    "index": 37,
+                    "bbox_fs": [
+                        100,
+                        703,
+                        404,
+                        739
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        82,
+                        299,
+                        94
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                81,
+                                301,
+                                95
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        81,
+                                        301,
+                                        95
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3.3 EFFICIENCY OF THE CONVNET KERNEL",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        }
+                    ],
+                    "index": 0
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        102,
+                        424,
+                        115
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                102,
+                                426,
+                                116
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        102,
+                                        426,
+                                        116
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We now have all the pieces for computing the kernel, as written in Algorithm 1.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        }
+                    ],
+                    "index": 1
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        119,
+                        505,
+                        164
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                119,
+                                506,
+                                132
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        119,
+                                        506,
+                                        132
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Putting together Eq. (11) and Eq. (13) gives us the surprising result that the diagonal covariances of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                130,
+                                505,
+                                144
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        130,
+                                        200,
+                                        144
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the activations at layer",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        200,
+                                        132,
+                                        223,
+                                        142
+                                    ],
+                                    "score": 0.9,
+                                    "content": "\\ell + 1",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        223,
+                                        130,
+                                        496,
+                                        144
+                                    ],
+                                    "score": 1.0,
+                                    "content": "only depend on the diagonal covariances of the activations at layer",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        496,
+                                        131,
+                                        501,
+                                        141
+                                    ],
+                                    "score": 0.62,
+                                    "content": "\\ell",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        502,
+                                        130,
+                                        505,
+                                        144
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                142,
+                                506,
+                                155
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        142,
+                                        506,
+                                        155
+                                    ],
+                                    "score": 1.0,
+                                    "content": "This is very important, because it makes the computational cost of the kernel be within a constant",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                152,
+                                432,
+                                166
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        152,
+                                        432,
+                                        166
+                                    ],
+                                    "score": 1.0,
+                                    "content": "factor of the cost of a forward pass for the equivalent CNN with 1 filter per layer.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        }
+                    ],
+                    "index": 3.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        169,
+                        505,
+                        221
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                168,
+                                505,
+                                183
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        168,
+                                        505,
+                                        183
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Thus, the algorithm is more efficient that one would naively think. A priori, one needs to compute",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                180,
+                                505,
+                                197
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        180,
+                                        276,
+                                        197
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the covariance between all the elements of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        277,
+                                        181,
+                                        309,
+                                        197
+                                    ],
+                                    "score": 0.94,
+                                    "content": "{ \\bf a } _ { j } ^ { ( \\ell ) } ( { \\bf X } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        310,
+                                        180,
+                                        327,
+                                        197
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        327,
+                                        181,
+                                        363,
+                                        197
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\mathbf { a } _ { j } ^ { ( \\ell ) } ( \\mathbf { X } ^ { \\prime } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        364,
+                                        180,
+                                        450,
+                                        197
+                                    ],
+                                    "score": 1.0,
+                                    "content": "combined, yielding a",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        451,
+                                        181,
+                                        505,
+                                        194
+                                    ],
+                                    "score": 0.91,
+                                    "content": "2 H ^ { ( \\ell ) } D ^ { ( \\ell ) } \\times",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                194,
+                                506,
+                                212
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        195,
+                                        150,
+                                        207
+                                    ],
+                                    "score": 0.9,
+                                    "content": "2 H ^ { ( \\ell ) } D ^ { ( \\ell ) }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        151,
+                                        194,
+                                        506,
+                                        212
+                                    ],
+                                    "score": 1.0,
+                                    "content": "covariance matrix for every pair of points. Instead, we only need to keep track of a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                107,
+                                207,
+                                339,
+                                223
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        208,
+                                        145,
+                                        219
+                                    ],
+                                    "score": 0.89,
+                                    "content": "H ^ { ( \\ell ) } D ^ { ( \\ell ) }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        145,
+                                        207,
+                                        339,
+                                        223
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-dimensional vector per layer and pair of points.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        }
+                    ],
+                    "index": 7.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        226,
+                        504,
+                        249
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                225,
+                                505,
+                                239
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        225,
+                                        505,
+                                        239
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Furthermore, the particular form for the kernel (eq. 1 and eq. 2) implies that the required variances",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                238,
+                                451,
+                                249
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        238,
+                                        451,
+                                        249
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and covariances at all required locations can be computed efficiently as a convolution.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        }
+                    ],
+                    "index": 10.5
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        262,
+                        266,
+                        273
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                262,
+                                267,
+                                275
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        262,
+                                        267,
+                                        275
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3.4 KERNEL FOR A RESIDUAL CNN",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        }
+                    ],
+                    "index": 12
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        282,
+                        505,
+                        339
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                281,
+                                505,
+                                296
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        281,
+                                        505,
+                                        296
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The induction step in the argument for GP behaviour from Sec. 2.2 depends only on the previous ac-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                294,
+                                505,
+                                306
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        294,
+                                        505,
+                                        306
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tivations being iid Gaussian. Since all the activations are iid Gaussian, we can add skip connections",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                304,
+                                503,
+                                319
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        304,
+                                        425,
+                                        319
+                                    ],
+                                    "score": 1.0,
+                                    "content": "between the activations of different layers while preserving GP behaviour, e.g.",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        425,
+                                        304,
+                                        455,
+                                        316
+                                    ],
+                                    "score": 0.87,
+                                    "content": "\\mathbf { A } ^ { ( \\ell + 1 ) }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        455,
+                                        304,
+                                        474,
+                                        319
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        474,
+                                        304,
+                                        503,
+                                        316
+                                    ],
+                                    "score": 0.89,
+                                    "content": "\\mathbf { A } ^ { ( \\ell - s ) }",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                315,
+                                505,
+                                330
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        315,
+                                        134,
+                                        330
+                                    ],
+                                    "score": 1.0,
+                                    "content": "where",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        134,
+                                        318,
+                                        141,
+                                        326
+                                    ],
+                                    "score": 0.77,
+                                    "content": "s",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        141,
+                                        315,
+                                        505,
+                                        330
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is the number of layers that the skip connection spans. If we change the NN recursion",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                327,
+                                148,
+                                341
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        327,
+                                        148,
+                                        341
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(Eq. 2) to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        }
+                    ],
+                    "index": 15
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        181,
+                        337,
+                        429,
+                        375
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                181,
+                                337,
+                                429,
+                                375
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        181,
+                                        337,
+                                        429,
+                                        375
+                                    ],
+                                    "score": 0.95,
+                                    "content": "\\mathbf { a } _ { i } ^ { ( \\ell + 1 ) } ( \\mathbf { X } ) : = \\mathbf { a } _ { i } ^ { ( \\ell - s ) } ( \\mathbf { X } ) + \\mathbf { b } _ { i } ^ { ( \\ell + 1 ) } + \\sum _ { j = 1 } ^ { C ^ { ( \\ell ) } } \\mathbf { W } _ { i , j } ^ { ( \\ell ) } \\phi \\left( \\mathbf { a } _ { j } ^ { ( \\ell ) } ( \\mathbf { X } ) \\right) ,",
+                                    "type": "interline_equation",
+                                    "image_path": "bf1c605df69c210dd72d42c616d1e888a45f018936a2f58c5328d55704234552.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 19,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                181,
+                                337,
+                                429,
+                                349.6666666666667
+                            ],
+                            "spans": [],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                181,
+                                349.6666666666667,
+                                429,
+                                362.33333333333337
+                            ],
+                            "spans": [],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                181,
+                                362.33333333333337,
+                                429,
+                                375.00000000000006
+                            ],
+                            "spans": [],
+                            "index": 20
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        376,
+                        280,
+                        388
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                375,
+                                280,
+                                389
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        375,
+                                        280,
+                                        389
+                                    ],
+                                    "score": 1.0,
+                                    "content": "then the kernel recursion (Eq. 11) becomes",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        }
+                    ],
+                    "index": 21
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        165,
+                        392,
+                        445,
+                        420
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                165,
+                                392,
+                                445,
+                                420
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        165,
+                                        392,
+                                        445,
+                                        420
+                                    ],
+                                    "score": 0.92,
+                                    "content": "K _ { \\mu } ^ { ( \\ell + 1 ) } ( \\mathbf { X } , \\mathbf { X } ^ { \\prime } ) = K _ { \\mu } ^ { ( \\ell - s ) } ( \\mathbf { X } , \\mathbf { X } ^ { \\prime } ) + \\sigma _ { b } ^ { 2 } + \\sigma _ { w } ^ { 2 } \\sum _ { \\nu \\in \\mu \\mathrm { t h } \\mathrm { p a t c h } } V _ { \\nu } ^ { ( \\ell ) } ( \\mathbf { X } , \\mathbf { X } ^ { \\prime } ) .",
+                                    "type": "interline_equation",
+                                    "image_path": "6bb7fa338ac056fae2ed317c9d8fe2d9b4ea34d60134d81a69cef145b31d3492.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 22,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                165,
+                                392,
+                                445,
+                                420
+                            ],
+                            "spans": [],
+                            "index": 22
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        425,
+                        505,
+                        459
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                424,
+                                506,
+                                438
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        424,
+                                        506,
+                                        438
+                                    ],
+                                    "score": 1.0,
+                                    "content": "This way of adding skip connections is equivalent to the “pre-activation” shortcuts described by He",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                436,
+                                505,
+                                448
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        436,
+                                        505,
+                                        448
+                                    ],
+                                    "score": 1.0,
+                                    "content": "et al. (2016b). Remarkably, the natural way of adding residual connections to NNs is the one that",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                447,
+                                290,
+                                460
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        447,
+                                        290,
+                                        460
+                                    ],
+                                    "score": 1.0,
+                                    "content": "performed best in their empirical evaluations.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        }
+                    ],
+                    "index": 24
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        474,
+                        200,
+                        487
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                474,
+                                201,
+                                489
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        474,
+                                        201,
+                                        489
+                                    ],
+                                    "score": 1.0,
+                                    "content": "4 EXPERIMENTS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        }
+                    ],
+                    "index": 26
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        499,
+                        504,
+                        533
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                499,
+                                505,
+                                511
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        499,
+                                        505,
+                                        511
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We evaluate our kernel on the MNIST handwritten digit classification task. Classification likelihoods",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                510,
+                                505,
+                                522
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        510,
+                                        505,
+                                        522
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are not conjugate for GPs, so we must make an approximation, and we follow Lee et al. (2017), in",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                522,
+                                315,
+                                534
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        522,
+                                        315,
+                                        534
+                                    ],
+                                    "score": 1.0,
+                                    "content": "re-framing classification as multi-output regression.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        }
+                    ],
+                    "index": 28
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        538,
+                        503,
+                        571
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                538,
+                                505,
+                                550
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        538,
+                                        222,
+                                        550
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The training set is split into",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        222,
+                                        538,
+                                        274,
+                                        549
+                                    ],
+                                    "score": 0.89,
+                                    "content": "N = 5 0 0 0 0",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        274,
+                                        538,
+                                        505,
+                                        550
+                                    ],
+                                    "score": 1.0,
+                                    "content": "training and 10000 validation examples. The regression",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                545,
+                                506,
+                                564
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        545,
+                                        135,
+                                        564
+                                    ],
+                                    "score": 1.0,
+                                    "content": "targets",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        135,
+                                        548,
+                                        210,
+                                        561
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\mathbf { Y } \\in \\{ - 1 , 1 \\} ^ { \\overline { { N } } \\times 1 0 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        211,
+                                        545,
+                                        394,
+                                        564
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are a one-hot encoding of the example’s class:",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        394,
+                                        550,
+                                        430,
+                                        562
+                                    ],
+                                    "score": 0.92,
+                                    "content": "y _ { n , c } = 1",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        430,
+                                        545,
+                                        453,
+                                        564
+                                    ],
+                                    "score": 1.0,
+                                    "content": "if the",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        453,
+                                        551,
+                                        460,
+                                        559
+                                    ],
+                                    "score": 0.64,
+                                    "content": "n",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        460,
+                                        545,
+                                        506,
+                                        564
+                                    ],
+                                    "score": 1.0,
+                                    "content": "th example",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                560,
+                                257,
+                                573
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        560,
+                                        172,
+                                        573
+                                    ],
+                                    "score": 1.0,
+                                    "content": "belongs to class",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        172,
+                                        563,
+                                        178,
+                                        570
+                                    ],
+                                    "score": 0.71,
+                                    "content": "c",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        178,
+                                        560,
+                                        198,
+                                        573
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        199,
+                                        561,
+                                        212,
+                                        571
+                                    ],
+                                    "score": 0.69,
+                                    "content": "- 1",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        213,
+                                        560,
+                                        257,
+                                        573
+                                    ],
+                                    "score": 1.0,
+                                    "content": "otherwise.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        }
+                    ],
+                    "index": 31
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        577,
+                        502,
+                        611
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                577,
+                                505,
+                                590
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        577,
+                                        428,
+                                        590
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Training is exact conjugate likelihood GP regression with noiseless targets",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        428,
+                                        577,
+                                        439,
+                                        587
+                                    ],
+                                    "score": 0.61,
+                                    "content": "\\mathbf { Y }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        439,
+                                        577,
+                                        505,
+                                        590
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(Rasmussen &",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                588,
+                                505,
+                                600
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        588,
+                                        260,
+                                        600
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Williams, 2006). First we compute the",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        260,
+                                        588,
+                                        288,
+                                        598
+                                    ],
+                                    "score": 0.9,
+                                    "content": "N \\times N",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        289,
+                                        588,
+                                        343,
+                                        600
+                                    ],
+                                    "score": 1.0,
+                                    "content": "kernel matrix",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        344,
+                                        588,
+                                        363,
+                                        599
+                                    ],
+                                    "score": 0.91,
+                                    "content": "{ \\bf K } _ { x x }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        363,
+                                        588,
+                                        505,
+                                        600
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", which contains the kernel between",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                597,
+                                423,
+                                614
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        597,
+                                        276,
+                                        614
+                                    ],
+                                    "score": 1.0,
+                                    "content": "every pair of examples. Then we compute",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        276,
+                                        599,
+                                        306,
+                                        611
+                                    ],
+                                    "score": 0.93,
+                                    "content": "{ \\mathbf { K } } _ { x x } ^ { - 1 } { \\mathbf { Y } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        307,
+                                        597,
+                                        423,
+                                        614
+                                    ],
+                                    "score": 1.0,
+                                    "content": "using a linear system solver.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        }
+                    ],
+                    "index": 34
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        615,
+                        502,
+                        650
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                614,
+                                505,
+                                629
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        614,
+                                        170,
+                                        629
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The test set has",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        170,
+                                        616,
+                                        224,
+                                        627
+                                    ],
+                                    "score": 0.9,
+                                    "content": "N _ { T } = 1 0 0 0 0",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        224,
+                                        614,
+                                        334,
+                                        629
+                                    ],
+                                    "score": 1.0,
+                                    "content": "examples. We compute the",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        334,
+                                        616,
+                                        370,
+                                        627
+                                    ],
+                                    "score": 0.91,
+                                    "content": "N _ { T } \\times N",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        371,
+                                        614,
+                                        399,
+                                        629
+                                    ],
+                                    "score": 1.0,
+                                    "content": "matrix",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        400,
+                                        616,
+                                        424,
+                                        627
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\mathbf { K } _ { x ^ { * } x }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        424,
+                                        614,
+                                        505,
+                                        629
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", the kernel between",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                627,
+                                504,
+                                639
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        627,
+                                        504,
+                                        639
+                                    ],
+                                    "score": 1.0,
+                                    "content": "each test example and all the training examples. The predictions are given by the row-wise maximum",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                636,
+                                174,
+                                652
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        636,
+                                        117,
+                                        652
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        118,
+                                        637,
+                                        169,
+                                        650
+                                    ],
+                                    "score": 0.94,
+                                    "content": "\\mathbf { K } _ { x ^ { * } x } \\mathbf { K } _ { x x } ^ { - 1 } \\bar { \\mathbf { Y } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        170,
+                                        636,
+                                        174,
+                                        652
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        }
+                    ],
+                    "index": 37
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        654,
+                        505,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                653,
+                                505,
+                                668
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        653,
+                                        505,
+                                        668
+                                    ],
+                                    "score": 1.0,
+                                    "content": "For the “ConvNet GP” and “Residual CNN GP”, (Table 1) we optimise the kernel hyperparameters",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                666,
+                                505,
+                                678
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        666,
+                                        227,
+                                        678
+                                    ],
+                                    "score": 1.0,
+                                    "content": "by random search. We draw",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        227,
+                                        666,
+                                        239,
+                                        676
+                                    ],
+                                    "score": 0.71,
+                                    "content": "M",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        240,
+                                        666,
+                                        505,
+                                        678
+                                    ],
+                                    "score": 1.0,
+                                    "content": "random hyperparameter samples, compute the resulting kernel’s",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                676,
+                                506,
+                                690
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        676,
+                                        506,
+                                        690
+                                    ],
+                                    "score": 1.0,
+                                    "content": "performance in the validation set, and pick the highest performing run. The kernel hyperparame-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                102,
+                                683,
+                                508,
+                                705
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        102,
+                                        683,
+                                        143,
+                                        705
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ters are:",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        143,
+                                        687,
+                                        155,
+                                        700
+                                    ],
+                                    "score": 0.8,
+                                    "content": "\\sigma _ { b } ^ { 2 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        155,
+                                        683,
+                                        159,
+                                        705
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        159,
+                                        687,
+                                        172,
+                                        700
+                                    ],
+                                    "score": 0.74,
+                                    "content": "\\sigma _ { w } ^ { 2 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        173,
+                                        683,
+                                        508,
+                                        705
+                                    ],
+                                    "score": 1.0,
+                                    "content": "; the number of layers; the convolution stride, filter sizes and edge behaviour; the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                698,
+                                505,
+                                711
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        698,
+                                        505,
+                                        711
+                                    ],
+                                    "score": 1.0,
+                                    "content": "nonlinearity (we consider the error function and ReLU); and the frequency of residual skip connec-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                709,
+                                505,
+                                722
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        709,
+                                        505,
+                                        722
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tions (for Residual CNN GPs). We do not retrain the model on the validation set after choosing",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                722,
+                                178,
+                                733
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        722,
+                                        178,
+                                        733
+                                    ],
+                                    "score": 1.0,
+                                    "content": "hyperparameters.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        }
+                    ],
+                    "index": 42
+                }
+            ],
+            "page_idx": 6,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        107,
+                        26,
+                        293,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                25,
+                                294,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        25,
+                                        294,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Published as a conference paper at ICLR 2019",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        302,
+                        751,
+                        309,
+                        759
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                302,
+                                750,
+                                309,
+                                762
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        302,
+                                        750,
+                                        309,
+                                        762
+                                    ],
+                                    "score": 1.0,
+                                    "content": "7",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        82,
+                        299,
+                        94
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                81,
+                                301,
+                                95
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        81,
+                                        301,
+                                        95
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3.3 EFFICIENCY OF THE CONVNET KERNEL",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        }
+                    ],
+                    "index": 0
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        102,
+                        424,
+                        115
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                102,
+                                426,
+                                116
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        102,
+                                        426,
+                                        116
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We now have all the pieces for computing the kernel, as written in Algorithm 1.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        }
+                    ],
+                    "index": 1,
+                    "bbox_fs": [
+                        106,
+                        102,
+                        426,
+                        116
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        119,
+                        505,
+                        164
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                119,
+                                506,
+                                132
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        119,
+                                        506,
+                                        132
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Putting together Eq. (11) and Eq. (13) gives us the surprising result that the diagonal covariances of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                130,
+                                505,
+                                144
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        130,
+                                        200,
+                                        144
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the activations at layer",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        200,
+                                        132,
+                                        223,
+                                        142
+                                    ],
+                                    "score": 0.9,
+                                    "content": "\\ell + 1",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        223,
+                                        130,
+                                        496,
+                                        144
+                                    ],
+                                    "score": 1.0,
+                                    "content": "only depend on the diagonal covariances of the activations at layer",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        496,
+                                        131,
+                                        501,
+                                        141
+                                    ],
+                                    "score": 0.62,
+                                    "content": "\\ell",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        502,
+                                        130,
+                                        505,
+                                        144
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                142,
+                                506,
+                                155
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        142,
+                                        506,
+                                        155
+                                    ],
+                                    "score": 1.0,
+                                    "content": "This is very important, because it makes the computational cost of the kernel be within a constant",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                152,
+                                432,
+                                166
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        152,
+                                        432,
+                                        166
+                                    ],
+                                    "score": 1.0,
+                                    "content": "factor of the cost of a forward pass for the equivalent CNN with 1 filter per layer.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        }
+                    ],
+                    "index": 3.5,
+                    "bbox_fs": [
+                        104,
+                        119,
+                        506,
+                        166
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        169,
+                        505,
+                        221
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                168,
+                                505,
+                                183
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        168,
+                                        505,
+                                        183
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Thus, the algorithm is more efficient that one would naively think. A priori, one needs to compute",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                180,
+                                505,
+                                197
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        180,
+                                        276,
+                                        197
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the covariance between all the elements of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        277,
+                                        181,
+                                        309,
+                                        197
+                                    ],
+                                    "score": 0.94,
+                                    "content": "{ \\bf a } _ { j } ^ { ( \\ell ) } ( { \\bf X } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        310,
+                                        180,
+                                        327,
+                                        197
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        327,
+                                        181,
+                                        363,
+                                        197
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\mathbf { a } _ { j } ^ { ( \\ell ) } ( \\mathbf { X } ^ { \\prime } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        364,
+                                        180,
+                                        450,
+                                        197
+                                    ],
+                                    "score": 1.0,
+                                    "content": "combined, yielding a",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        451,
+                                        181,
+                                        505,
+                                        194
+                                    ],
+                                    "score": 0.91,
+                                    "content": "2 H ^ { ( \\ell ) } D ^ { ( \\ell ) } \\times",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                194,
+                                506,
+                                212
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        195,
+                                        150,
+                                        207
+                                    ],
+                                    "score": 0.9,
+                                    "content": "2 H ^ { ( \\ell ) } D ^ { ( \\ell ) }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        151,
+                                        194,
+                                        506,
+                                        212
+                                    ],
+                                    "score": 1.0,
+                                    "content": "covariance matrix for every pair of points. Instead, we only need to keep track of a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                107,
+                                207,
+                                339,
+                                223
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        208,
+                                        145,
+                                        219
+                                    ],
+                                    "score": 0.89,
+                                    "content": "H ^ { ( \\ell ) } D ^ { ( \\ell ) }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        145,
+                                        207,
+                                        339,
+                                        223
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-dimensional vector per layer and pair of points.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        }
+                    ],
+                    "index": 7.5,
+                    "bbox_fs": [
+                        104,
+                        168,
+                        506,
+                        223
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        226,
+                        504,
+                        249
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                225,
+                                505,
+                                239
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        225,
+                                        505,
+                                        239
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Furthermore, the particular form for the kernel (eq. 1 and eq. 2) implies that the required variances",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                238,
+                                451,
+                                249
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        238,
+                                        451,
+                                        249
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and covariances at all required locations can be computed efficiently as a convolution.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        }
+                    ],
+                    "index": 10.5,
+                    "bbox_fs": [
+                        105,
+                        225,
+                        505,
+                        249
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        262,
+                        266,
+                        273
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                262,
+                                267,
+                                275
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        262,
+                                        267,
+                                        275
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3.4 KERNEL FOR A RESIDUAL CNN",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        }
+                    ],
+                    "index": 12
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        282,
+                        505,
+                        339
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                281,
+                                505,
+                                296
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        281,
+                                        505,
+                                        296
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The induction step in the argument for GP behaviour from Sec. 2.2 depends only on the previous ac-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                294,
+                                505,
+                                306
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        294,
+                                        505,
+                                        306
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tivations being iid Gaussian. Since all the activations are iid Gaussian, we can add skip connections",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                304,
+                                503,
+                                319
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        304,
+                                        425,
+                                        319
+                                    ],
+                                    "score": 1.0,
+                                    "content": "between the activations of different layers while preserving GP behaviour, e.g.",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        425,
+                                        304,
+                                        455,
+                                        316
+                                    ],
+                                    "score": 0.87,
+                                    "content": "\\mathbf { A } ^ { ( \\ell + 1 ) }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        455,
+                                        304,
+                                        474,
+                                        319
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        474,
+                                        304,
+                                        503,
+                                        316
+                                    ],
+                                    "score": 0.89,
+                                    "content": "\\mathbf { A } ^ { ( \\ell - s ) }",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                315,
+                                505,
+                                330
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        315,
+                                        134,
+                                        330
+                                    ],
+                                    "score": 1.0,
+                                    "content": "where",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        134,
+                                        318,
+                                        141,
+                                        326
+                                    ],
+                                    "score": 0.77,
+                                    "content": "s",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        141,
+                                        315,
+                                        505,
+                                        330
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is the number of layers that the skip connection spans. If we change the NN recursion",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                327,
+                                148,
+                                341
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        327,
+                                        148,
+                                        341
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(Eq. 2) to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        }
+                    ],
+                    "index": 15,
+                    "bbox_fs": [
+                        104,
+                        281,
+                        505,
+                        341
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        181,
+                        337,
+                        429,
+                        375
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                181,
+                                337,
+                                429,
+                                375
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        181,
+                                        337,
+                                        429,
+                                        375
+                                    ],
+                                    "score": 0.95,
+                                    "content": "\\mathbf { a } _ { i } ^ { ( \\ell + 1 ) } ( \\mathbf { X } ) : = \\mathbf { a } _ { i } ^ { ( \\ell - s ) } ( \\mathbf { X } ) + \\mathbf { b } _ { i } ^ { ( \\ell + 1 ) } + \\sum _ { j = 1 } ^ { C ^ { ( \\ell ) } } \\mathbf { W } _ { i , j } ^ { ( \\ell ) } \\phi \\left( \\mathbf { a } _ { j } ^ { ( \\ell ) } ( \\mathbf { X } ) \\right) ,",
+                                    "type": "interline_equation",
+                                    "image_path": "bf1c605df69c210dd72d42c616d1e888a45f018936a2f58c5328d55704234552.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 19,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                181,
+                                337,
+                                429,
+                                349.6666666666667
+                            ],
+                            "spans": [],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                181,
+                                349.6666666666667,
+                                429,
+                                362.33333333333337
+                            ],
+                            "spans": [],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                181,
+                                362.33333333333337,
+                                429,
+                                375.00000000000006
+                            ],
+                            "spans": [],
+                            "index": 20
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        376,
+                        280,
+                        388
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                375,
+                                280,
+                                389
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        375,
+                                        280,
+                                        389
+                                    ],
+                                    "score": 1.0,
+                                    "content": "then the kernel recursion (Eq. 11) becomes",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        }
+                    ],
+                    "index": 21,
+                    "bbox_fs": [
+                        106,
+                        375,
+                        280,
+                        389
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        165,
+                        392,
+                        445,
+                        420
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                165,
+                                392,
+                                445,
+                                420
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        165,
+                                        392,
+                                        445,
+                                        420
+                                    ],
+                                    "score": 0.92,
+                                    "content": "K _ { \\mu } ^ { ( \\ell + 1 ) } ( \\mathbf { X } , \\mathbf { X } ^ { \\prime } ) = K _ { \\mu } ^ { ( \\ell - s ) } ( \\mathbf { X } , \\mathbf { X } ^ { \\prime } ) + \\sigma _ { b } ^ { 2 } + \\sigma _ { w } ^ { 2 } \\sum _ { \\nu \\in \\mu \\mathrm { t h } \\mathrm { p a t c h } } V _ { \\nu } ^ { ( \\ell ) } ( \\mathbf { X } , \\mathbf { X } ^ { \\prime } ) .",
+                                    "type": "interline_equation",
+                                    "image_path": "6bb7fa338ac056fae2ed317c9d8fe2d9b4ea34d60134d81a69cef145b31d3492.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 22,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                165,
+                                392,
+                                445,
+                                420
+                            ],
+                            "spans": [],
+                            "index": 22
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        425,
+                        505,
+                        459
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                424,
+                                506,
+                                438
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        424,
+                                        506,
+                                        438
+                                    ],
+                                    "score": 1.0,
+                                    "content": "This way of adding skip connections is equivalent to the “pre-activation” shortcuts described by He",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                436,
+                                505,
+                                448
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        436,
+                                        505,
+                                        448
+                                    ],
+                                    "score": 1.0,
+                                    "content": "et al. (2016b). Remarkably, the natural way of adding residual connections to NNs is the one that",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                447,
+                                290,
+                                460
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        447,
+                                        290,
+                                        460
+                                    ],
+                                    "score": 1.0,
+                                    "content": "performed best in their empirical evaluations.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        }
+                    ],
+                    "index": 24,
+                    "bbox_fs": [
+                        105,
+                        424,
+                        506,
+                        460
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        474,
+                        200,
+                        487
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                474,
+                                201,
+                                489
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        474,
+                                        201,
+                                        489
+                                    ],
+                                    "score": 1.0,
+                                    "content": "4 EXPERIMENTS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        }
+                    ],
+                    "index": 26
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        499,
+                        504,
+                        533
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                499,
+                                505,
+                                511
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        499,
+                                        505,
+                                        511
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We evaluate our kernel on the MNIST handwritten digit classification task. Classification likelihoods",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                510,
+                                505,
+                                522
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        510,
+                                        505,
+                                        522
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are not conjugate for GPs, so we must make an approximation, and we follow Lee et al. (2017), in",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                522,
+                                315,
+                                534
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        522,
+                                        315,
+                                        534
+                                    ],
+                                    "score": 1.0,
+                                    "content": "re-framing classification as multi-output regression.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        }
+                    ],
+                    "index": 28,
+                    "bbox_fs": [
+                        105,
+                        499,
+                        505,
+                        534
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        538,
+                        503,
+                        571
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                538,
+                                505,
+                                550
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        538,
+                                        222,
+                                        550
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The training set is split into",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        222,
+                                        538,
+                                        274,
+                                        549
+                                    ],
+                                    "score": 0.89,
+                                    "content": "N = 5 0 0 0 0",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        274,
+                                        538,
+                                        505,
+                                        550
+                                    ],
+                                    "score": 1.0,
+                                    "content": "training and 10000 validation examples. The regression",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                545,
+                                506,
+                                564
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        545,
+                                        135,
+                                        564
+                                    ],
+                                    "score": 1.0,
+                                    "content": "targets",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        135,
+                                        548,
+                                        210,
+                                        561
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\mathbf { Y } \\in \\{ - 1 , 1 \\} ^ { \\overline { { N } } \\times 1 0 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        211,
+                                        545,
+                                        394,
+                                        564
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are a one-hot encoding of the example’s class:",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        394,
+                                        550,
+                                        430,
+                                        562
+                                    ],
+                                    "score": 0.92,
+                                    "content": "y _ { n , c } = 1",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        430,
+                                        545,
+                                        453,
+                                        564
+                                    ],
+                                    "score": 1.0,
+                                    "content": "if the",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        453,
+                                        551,
+                                        460,
+                                        559
+                                    ],
+                                    "score": 0.64,
+                                    "content": "n",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        460,
+                                        545,
+                                        506,
+                                        564
+                                    ],
+                                    "score": 1.0,
+                                    "content": "th example",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                560,
+                                257,
+                                573
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        560,
+                                        172,
+                                        573
+                                    ],
+                                    "score": 1.0,
+                                    "content": "belongs to class",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        172,
+                                        563,
+                                        178,
+                                        570
+                                    ],
+                                    "score": 0.71,
+                                    "content": "c",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        178,
+                                        560,
+                                        198,
+                                        573
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        199,
+                                        561,
+                                        212,
+                                        571
+                                    ],
+                                    "score": 0.69,
+                                    "content": "- 1",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        213,
+                                        560,
+                                        257,
+                                        573
+                                    ],
+                                    "score": 1.0,
+                                    "content": "otherwise.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        }
+                    ],
+                    "index": 31,
+                    "bbox_fs": [
+                        104,
+                        538,
+                        506,
+                        573
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        577,
+                        502,
+                        611
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                577,
+                                505,
+                                590
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        577,
+                                        428,
+                                        590
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Training is exact conjugate likelihood GP regression with noiseless targets",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        428,
+                                        577,
+                                        439,
+                                        587
+                                    ],
+                                    "score": 0.61,
+                                    "content": "\\mathbf { Y }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        439,
+                                        577,
+                                        505,
+                                        590
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(Rasmussen &",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                588,
+                                505,
+                                600
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        588,
+                                        260,
+                                        600
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Williams, 2006). First we compute the",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        260,
+                                        588,
+                                        288,
+                                        598
+                                    ],
+                                    "score": 0.9,
+                                    "content": "N \\times N",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        289,
+                                        588,
+                                        343,
+                                        600
+                                    ],
+                                    "score": 1.0,
+                                    "content": "kernel matrix",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        344,
+                                        588,
+                                        363,
+                                        599
+                                    ],
+                                    "score": 0.91,
+                                    "content": "{ \\bf K } _ { x x }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        363,
+                                        588,
+                                        505,
+                                        600
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", which contains the kernel between",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                597,
+                                423,
+                                614
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        597,
+                                        276,
+                                        614
+                                    ],
+                                    "score": 1.0,
+                                    "content": "every pair of examples. Then we compute",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        276,
+                                        599,
+                                        306,
+                                        611
+                                    ],
+                                    "score": 0.93,
+                                    "content": "{ \\mathbf { K } } _ { x x } ^ { - 1 } { \\mathbf { Y } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        307,
+                                        597,
+                                        423,
+                                        614
+                                    ],
+                                    "score": 1.0,
+                                    "content": "using a linear system solver.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        }
+                    ],
+                    "index": 34,
+                    "bbox_fs": [
+                        104,
+                        577,
+                        505,
+                        614
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        615,
+                        502,
+                        650
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                614,
+                                505,
+                                629
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        614,
+                                        170,
+                                        629
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The test set has",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        170,
+                                        616,
+                                        224,
+                                        627
+                                    ],
+                                    "score": 0.9,
+                                    "content": "N _ { T } = 1 0 0 0 0",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        224,
+                                        614,
+                                        334,
+                                        629
+                                    ],
+                                    "score": 1.0,
+                                    "content": "examples. We compute the",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        334,
+                                        616,
+                                        370,
+                                        627
+                                    ],
+                                    "score": 0.91,
+                                    "content": "N _ { T } \\times N",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        371,
+                                        614,
+                                        399,
+                                        629
+                                    ],
+                                    "score": 1.0,
+                                    "content": "matrix",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        400,
+                                        616,
+                                        424,
+                                        627
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\mathbf { K } _ { x ^ { * } x }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        424,
+                                        614,
+                                        505,
+                                        629
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", the kernel between",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                627,
+                                504,
+                                639
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        627,
+                                        504,
+                                        639
+                                    ],
+                                    "score": 1.0,
+                                    "content": "each test example and all the training examples. The predictions are given by the row-wise maximum",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                636,
+                                174,
+                                652
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        636,
+                                        117,
+                                        652
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        118,
+                                        637,
+                                        169,
+                                        650
+                                    ],
+                                    "score": 0.94,
+                                    "content": "\\mathbf { K } _ { x ^ { * } x } \\mathbf { K } _ { x x } ^ { - 1 } \\bar { \\mathbf { Y } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        170,
+                                        636,
+                                        174,
+                                        652
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        }
+                    ],
+                    "index": 37,
+                    "bbox_fs": [
+                        105,
+                        614,
+                        505,
+                        652
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        654,
+                        505,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                653,
+                                505,
+                                668
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        653,
+                                        505,
+                                        668
+                                    ],
+                                    "score": 1.0,
+                                    "content": "For the “ConvNet GP” and “Residual CNN GP”, (Table 1) we optimise the kernel hyperparameters",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                666,
+                                505,
+                                678
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        666,
+                                        227,
+                                        678
+                                    ],
+                                    "score": 1.0,
+                                    "content": "by random search. We draw",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        227,
+                                        666,
+                                        239,
+                                        676
+                                    ],
+                                    "score": 0.71,
+                                    "content": "M",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        240,
+                                        666,
+                                        505,
+                                        678
+                                    ],
+                                    "score": 1.0,
+                                    "content": "random hyperparameter samples, compute the resulting kernel’s",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                676,
+                                506,
+                                690
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        676,
+                                        506,
+                                        690
+                                    ],
+                                    "score": 1.0,
+                                    "content": "performance in the validation set, and pick the highest performing run. The kernel hyperparame-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                102,
+                                683,
+                                508,
+                                705
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        102,
+                                        683,
+                                        143,
+                                        705
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ters are:",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        143,
+                                        687,
+                                        155,
+                                        700
+                                    ],
+                                    "score": 0.8,
+                                    "content": "\\sigma _ { b } ^ { 2 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        155,
+                                        683,
+                                        159,
+                                        705
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        159,
+                                        687,
+                                        172,
+                                        700
+                                    ],
+                                    "score": 0.74,
+                                    "content": "\\sigma _ { w } ^ { 2 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        173,
+                                        683,
+                                        508,
+                                        705
+                                    ],
+                                    "score": 1.0,
+                                    "content": "; the number of layers; the convolution stride, filter sizes and edge behaviour; the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                698,
+                                505,
+                                711
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        698,
+                                        505,
+                                        711
+                                    ],
+                                    "score": 1.0,
+                                    "content": "nonlinearity (we consider the error function and ReLU); and the frequency of residual skip connec-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                709,
+                                505,
+                                722
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        709,
+                                        505,
+                                        722
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tions (for Residual CNN GPs). We do not retrain the model on the validation set after choosing",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                722,
+                                178,
+                                733
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        722,
+                                        178,
+                                        733
+                                    ],
+                                    "score": 1.0,
+                                    "content": "hyperparameters.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        }
+                    ],
+                    "index": 42,
+                    "bbox_fs": [
+                        102,
+                        653,
+                        508,
+                        733
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "table",
+                    "bbox": [
+                        106,
+                        80,
+                        515,
+                        182
+                    ],
+                    "blocks": [
+                        {
+                            "type": "table_body",
+                            "bbox": [
+                                106,
+                                80,
+                                515,
+                                182
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        80,
+                                        515,
+                                        182
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                80,
+                                                515,
+                                                182
+                                            ],
+                                            "score": 0.98,
+                                            "html": "<table><tr><td>Method</td><td>#samples</td><td>Validation error</td><td>Test error</td></tr><tr><td>NNGP (Lee et al., 2017)</td><td>~250</td><td>1</td><td>1.21%</td></tr><tr><td>Convolutional GP (van der Wilk et al., 2017)</td><td>SGD</td><td></td><td>1.17%</td></tr><tr><td>Deep Conv. GP (Kumar et al., 2018)</td><td>SGD</td><td>1</td><td>1.34%</td></tr><tr><td>ConvNetGP</td><td>27</td><td>0.71%</td><td>1.03%</td></tr><tr><td>Residual CNN GP</td><td>27</td><td>0.72%</td><td>0.96%</td></tr><tr><td>ResNet GP</td><td>1</td><td>0.68%</td><td>0.84%</td></tr><tr><td>GP + parametric deep kernel (Bradshaw et al., 2017)</td><td>SGD</td><td>1</td><td>0.60%</td></tr><tr><td>ResNet (Chen et al.,2018)</td><td>一</td><td></td><td>0.41%</td></tr></table>",
+                                            "type": "table",
+                                            "image_path": "f762ab560f768795c1ec5b70d1189c780fac2c18790fd6deb0de130f54ee163d.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 1,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        80,
+                                        515,
+                                        114.0
+                                    ],
+                                    "spans": [],
+                                    "index": 0
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        114.0,
+                                        515,
+                                        148.0
+                                    ],
+                                    "spans": [],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        148.0,
+                                        515,
+                                        182.0
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                }
+                            ]
+                        },
+                        {
+                            "type": "table_caption",
+                            "bbox": [
+                                106,
+                                195,
+                                505,
+                                273
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        195,
+                                        505,
+                                        208
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                195,
+                                                505,
+                                                208
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Table 1: MNIST classification results. #samples gives the number of kernels that were randomly",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 3
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        207,
+                                        505,
+                                        218
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                207,
+                                                505,
+                                                218
+                                            ],
+                                            "score": 1.0,
+                                            "content": "sampled for the hyperparameter search. “ConvNet GP” and “Residual CNN GP” are random CNN",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 4
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        217,
+                                        505,
+                                        230
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                217,
+                                                505,
+                                                230
+                                            ],
+                                            "score": 1.0,
+                                            "content": "architectures with a fixed filter size, whereas “ResNet GP” is a slight modification of the architecture",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 5
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        228,
+                                        504,
+                                        241
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                228,
+                                                504,
+                                                241
+                                            ],
+                                            "score": 1.0,
+                                            "content": "by He et al. (2016b). Entries labelled “SGD” used stochastic gradient descent for tuning hyperpa-",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 6
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        240,
+                                        505,
+                                        252
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                240,
+                                                505,
+                                                252
+                                            ],
+                                            "score": 1.0,
+                                            "content": "rameters, by maximising the likelihood of the training set. The last two methods use parametric",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 7
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        251,
+                                        505,
+                                        263
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                251,
+                                                505,
+                                                263
+                                            ],
+                                            "score": 1.0,
+                                            "content": "neural networks. The hyperparameters of the ResNet GP were not optimised (they were fixed based",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 8
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        262,
+                                        276,
+                                        273
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                262,
+                                                276,
+                                                273
+                                            ],
+                                            "score": 1.0,
+                                            "content": "on the architecture from He et al., 2016b).",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 9
+                                }
+                            ],
+                            "index": 6
+                        }
+                    ],
+                    "index": 3.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        292,
+                        504,
+                        348
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                291,
+                                506,
+                                305
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        291,
+                                        506,
+                                        305
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The “ResNet GP” (Table 1) is the kernel equivalent to a 32-layer version of the basic residual ar-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                304,
+                                505,
+                                316
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        304,
+                                        356,
+                                        316
+                                    ],
+                                    "score": 1.0,
+                                    "content": "chitecture by He et al. (2016a). The differences are: an initial",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        356,
+                                        304,
+                                        380,
+                                        315
+                                    ],
+                                    "score": 0.9,
+                                    "content": "3 \\times 3",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        380,
+                                        304,
+                                        505,
+                                        316
+                                    ],
+                                    "score": 1.0,
+                                    "content": "convolutional layer and a final",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                314,
+                                506,
+                                328
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        314,
+                                        506,
+                                        328
+                                    ],
+                                    "score": 1.0,
+                                    "content": "dense layer instead of average pooling. We chose to remove the pooling because computing its out-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                325,
+                                506,
+                                339
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        325,
+                                        506,
+                                        339
+                                    ],
+                                    "score": 1.0,
+                                    "content": "put variance requires the off-diagonal elements of the filter covariance, in which case we could not",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                337,
+                                304,
+                                349
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        337,
+                                        304,
+                                        349
+                                    ],
+                                    "score": 1.0,
+                                    "content": "exploit the efficiency gains described in Sec. 3.3.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        }
+                    ],
+                    "index": 12
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        353,
+                        505,
+                        419
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                353,
+                                505,
+                                366
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        353,
+                                        505,
+                                        366
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We found that the despite it not being optimised, the 32-layer ResNet GP outperformed all other",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                365,
+                                505,
+                                376
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        365,
+                                        505,
+                                        376
+                                    ],
+                                    "score": 1.0,
+                                    "content": "comparable architectures (Table 1), including the NNGP in Lee et al. (2017), which is state-of-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                375,
+                                505,
+                                388
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        375,
+                                        505,
+                                        388
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the-art for non-convolutional networks, and convolutional GPs (van der Wilk et al., 2017; Kumar",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                386,
+                                506,
+                                399
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        386,
+                                        506,
+                                        399
+                                    ],
+                                    "score": 1.0,
+                                    "content": "et al., 2018). That said, our results have not reached state-of-the-art for methods that incorporate a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                397,
+                                506,
+                                411
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        397,
+                                        506,
+                                        411
+                                    ],
+                                    "score": 1.0,
+                                    "content": "parametric neural network, such as a standard ResNet (Chen et al., 2018) and a Gaussian process",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                408,
+                                340,
+                                420
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        408,
+                                        340,
+                                        420
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with a deep neural network kernel Bradshaw et al. (2017).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        }
+                    ],
+                    "index": 17.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        425,
+                        505,
+                        535
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                424,
+                                505,
+                                438
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        424,
+                                        505,
+                                        438
+                                    ],
+                                    "score": 1.0,
+                                    "content": "To check whether the GP limit is applicable to relatively small networks used practically (with of the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                436,
+                                505,
+                                449
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        436,
+                                        505,
+                                        449
+                                    ],
+                                    "score": 1.0,
+                                    "content": "order of 100 channels in the first layers), we randomly sampled 10, 000 32-layer ResNets, with 3,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                446,
+                                506,
+                                460
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        446,
+                                        506,
+                                        460
+                                    ],
+                                    "score": 1.0,
+                                    "content": "10, 30 and 100 channels in the first layers, and, following the usual practice for ResNets we increase",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                458,
+                                505,
+                                471
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        458,
+                                        505,
+                                        471
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the number the number of hidden units when we downsample the feature maps. The probability",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                469,
+                                505,
+                                482
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        469,
+                                        505,
+                                        482
+                                    ],
+                                    "score": 1.0,
+                                    "content": "density plots show a good match around 100 channels (Fig. 2A), which matches a more sensitive",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                480,
+                                506,
+                                493
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        480,
+                                        506,
+                                        493
+                                    ],
+                                    "score": 1.0,
+                                    "content": "graphical procedure based on quantile-quantile plots (Fig. 2B). Notably, even for only 30 channels,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                490,
+                                506,
+                                505
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        490,
+                                        506,
+                                        505
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the moments match closely (Fig. 2C). For comparison, typical ResNets use from 64 (He et al.,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                501,
+                                505,
+                                514
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        501,
+                                        505,
+                                        514
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2016a) to 192 (Zagoruyko & Komodakis, 2016) channels in their first layers. We believe that this is",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                513,
+                                505,
+                                526
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        513,
+                                        505,
+                                        526
+                                    ],
+                                    "score": 1.0,
+                                    "content": "because the moment propagation equations only require the Gaussianity assumption for propagation",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                524,
+                                432,
+                                537
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        524,
+                                        432,
+                                        537
+                                    ],
+                                    "score": 1.0,
+                                    "content": "through the relu, and presumably this is robust to non-Gaussian input activations.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        }
+                    ],
+                    "index": 25.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        547,
+                        505,
+                        636
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                545,
+                                506,
+                                561
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        545,
+                                        434,
+                                        561
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Computational efficiency. Asymptotically, computing the kernel matrix takes",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        435,
+                                        547,
+                                        480,
+                                        560
+                                    ],
+                                    "score": 0.92,
+                                    "content": "O ( N ^ { 2 } L D )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        480,
+                                        545,
+                                        506,
+                                        561
+                                    ],
+                                    "score": 1.0,
+                                    "content": "time,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                558,
+                                505,
+                                571
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        558,
+                                        132,
+                                        571
+                                    ],
+                                    "score": 1.0,
+                                    "content": "where",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        133,
+                                        559,
+                                        141,
+                                        568
+                                    ],
+                                    "score": 0.79,
+                                    "content": "L",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        141,
+                                        558,
+                                        307,
+                                        571
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is the number of layers in the network and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        307,
+                                        559,
+                                        317,
+                                        568
+                                    ],
+                                    "score": 0.82,
+                                    "content": "D",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        317,
+                                        558,
+                                        505,
+                                        571
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is the dimensionality of the input, and inverting",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                569,
+                                506,
+                                582
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        569,
+                                        200,
+                                        582
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the kernel matrix takes",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        200,
+                                        569,
+                                        230,
+                                        581
+                                    ],
+                                    "score": 0.92,
+                                    "content": "O ( \\bar { N } ^ { 3 } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        230,
+                                        569,
+                                        506,
+                                        582
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". As such, we expect that for very large datasets, inverting the kernel",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                579,
+                                506,
+                                593
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        579,
+                                        380,
+                                        593
+                                    ],
+                                    "score": 1.0,
+                                    "content": "matrix will dominate the computation time. However, on MNIST,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        381,
+                                        580,
+                                        396,
+                                        591
+                                    ],
+                                    "score": 0.88,
+                                    "content": "N ^ { 3 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        396,
+                                        579,
+                                        506,
+                                        593
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is only around a factor of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                591,
+                                505,
+                                604
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        591,
+                                        168,
+                                        604
+                                    ],
+                                    "score": 1.0,
+                                    "content": "10 larger than",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        168,
+                                        591,
+                                        198,
+                                        602
+                                    ],
+                                    "score": 0.9,
+                                    "content": "N ^ { 2 } L D",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        198,
+                                        591,
+                                        505,
+                                        604
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". In practice, we found that it was more expensive to compute the kernel",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                601,
+                                504,
+                                616
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        601,
+                                        437,
+                                        616
+                                    ],
+                                    "score": 1.0,
+                                    "content": "matrix than to invert it. For the ResNet kernel, the most expensive, computing",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        437,
+                                        603,
+                                        456,
+                                        614
+                                    ],
+                                    "score": 0.9,
+                                    "content": "{ \\bf K } _ { x x }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        457,
+                                        601,
+                                        479,
+                                        616
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        480,
+                                        603,
+                                        504,
+                                        613
+                                    ],
+                                    "score": 0.9,
+                                    "content": "{ \\bf K } _ { x x * }",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                613,
+                                505,
+                                626
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        613,
+                                        466,
+                                        626
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for validation and test took 3h 40min on two Tesla P100 GPUs. In contrast, inverting",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        466,
+                                        614,
+                                        486,
+                                        624
+                                    ],
+                                    "score": 0.89,
+                                    "content": "{ \\bf K } _ { x x }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        486,
+                                        613,
+                                        505,
+                                        626
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                624,
+                                501,
+                                636
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        624,
+                                        300,
+                                        636
+                                    ],
+                                    "score": 1.0,
+                                    "content": "computing validation and test performance took",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        300,
+                                        624,
+                                        349,
+                                        635
+                                    ],
+                                    "score": 0.87,
+                                    "content": "4 3 . 2 5 \\pm 8 . 8",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        350,
+                                        624,
+                                        501,
+                                        636
+                                    ],
+                                    "score": 1.0,
+                                    "content": "seconds on a single Tesla P100 GPU.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        }
+                    ],
+                    "index": 34.5
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        652,
+                        209,
+                        664
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                650,
+                                211,
+                                666
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        650,
+                                        211,
+                                        666
+                                    ],
+                                    "score": 1.0,
+                                    "content": "5 RELATED WORK",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        }
+                    ],
+                    "index": 39
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        676,
+                        504,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                676,
+                                505,
+                                690
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        676,
+                                        505,
+                                        690
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Van der Wilk et al. (van der Wilk et al., 2017) also adapted GPs to image classification. They",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                688,
+                                505,
+                                700
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        688,
+                                        226,
+                                        700
+                                    ],
+                                    "score": 1.0,
+                                    "content": "defined a prior on functions",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        226,
+                                        688,
+                                        233,
+                                        699
+                                    ],
+                                    "score": 0.87,
+                                    "content": "f",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        234,
+                                        688,
+                                        505,
+                                        700
+                                    ],
+                                    "score": 1.0,
+                                    "content": "that takes an image and outputs a scalar. First, draw a function",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                107,
+                                698,
+                                506,
+                                712
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        699,
+                                        200,
+                                        711
+                                    ],
+                                    "score": 0.91,
+                                    "content": "g \\sim \\mathcal { G P } ( 0 , k _ { p } ( \\mathbf { X } , \\mathbf { X } ^ { \\prime } ) )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        201,
+                                        698,
+                                        232,
+                                        712
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Then,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        232,
+                                        699,
+                                        239,
+                                        711
+                                    ],
+                                    "score": 0.85,
+                                    "content": "f",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        240,
+                                        698,
+                                        350,
+                                        712
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is the sum of the output of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        351,
+                                        700,
+                                        357,
+                                        710
+                                    ],
+                                    "score": 0.81,
+                                    "content": "g",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        357,
+                                        698,
+                                        506,
+                                        712
+                                    ],
+                                    "score": 1.0,
+                                    "content": "applied to each of the convolutional",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                709,
+                                506,
+                                723
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        709,
+                                        427,
+                                        723
+                                    ],
+                                    "score": 1.0,
+                                    "content": "patches. Their approach is also inspired by convolutional NNs, but their kernel",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        428,
+                                        710,
+                                        439,
+                                        722
+                                    ],
+                                    "score": 0.88,
+                                    "content": "k _ { p }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        439,
+                                        709,
+                                        506,
+                                        723
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is applied to all",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                718,
+                                506,
+                                735
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        718,
+                                        181,
+                                        735
+                                    ],
+                                    "score": 1.0,
+                                    "content": "pairs of patches of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        182,
+                                        721,
+                                        191,
+                                        730
+                                    ],
+                                    "score": 0.68,
+                                    "content": "\\mathbf { X }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        191,
+                                        718,
+                                        209,
+                                        735
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        209,
+                                        721,
+                                        221,
+                                        730
+                                    ],
+                                    "score": 0.86,
+                                    "content": "\\mathbf { X } ^ { \\prime }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        222,
+                                        718,
+                                        506,
+                                        735
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". This makes their convolutional kernel expensive to evaluate, requiring",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        }
+                    ],
+                    "index": 42
+                }
+            ],
+            "page_idx": 7,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        108,
+                        27,
+                        293,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                26,
+                                294,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        26,
+                                        294,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Published as a conference paper at ICLR 2019",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        302,
+                        752,
+                        308,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                302,
+                                750,
+                                309,
+                                761
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        302,
+                                        750,
+                                        309,
+                                        761
+                                    ],
+                                    "score": 1.0,
+                                    "content": "8",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "table",
+                    "bbox": [
+                        106,
+                        80,
+                        515,
+                        182
+                    ],
+                    "blocks": [
+                        {
+                            "type": "table_body",
+                            "bbox": [
+                                106,
+                                80,
+                                515,
+                                182
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        80,
+                                        515,
+                                        182
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                80,
+                                                515,
+                                                182
+                                            ],
+                                            "score": 0.98,
+                                            "html": "<table><tr><td>Method</td><td>#samples</td><td>Validation error</td><td>Test error</td></tr><tr><td>NNGP (Lee et al., 2017)</td><td>~250</td><td>1</td><td>1.21%</td></tr><tr><td>Convolutional GP (van der Wilk et al., 2017)</td><td>SGD</td><td></td><td>1.17%</td></tr><tr><td>Deep Conv. GP (Kumar et al., 2018)</td><td>SGD</td><td>1</td><td>1.34%</td></tr><tr><td>ConvNetGP</td><td>27</td><td>0.71%</td><td>1.03%</td></tr><tr><td>Residual CNN GP</td><td>27</td><td>0.72%</td><td>0.96%</td></tr><tr><td>ResNet GP</td><td>1</td><td>0.68%</td><td>0.84%</td></tr><tr><td>GP + parametric deep kernel (Bradshaw et al., 2017)</td><td>SGD</td><td>1</td><td>0.60%</td></tr><tr><td>ResNet (Chen et al.,2018)</td><td>一</td><td></td><td>0.41%</td></tr></table>",
+                                            "type": "table",
+                                            "image_path": "f762ab560f768795c1ec5b70d1189c780fac2c18790fd6deb0de130f54ee163d.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 1,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        80,
+                                        515,
+                                        114.0
+                                    ],
+                                    "spans": [],
+                                    "index": 0
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        114.0,
+                                        515,
+                                        148.0
+                                    ],
+                                    "spans": [],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        148.0,
+                                        515,
+                                        182.0
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                }
+                            ]
+                        },
+                        {
+                            "type": "table_caption",
+                            "bbox": [
+                                106,
+                                195,
+                                505,
+                                273
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        195,
+                                        505,
+                                        208
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                195,
+                                                505,
+                                                208
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Table 1: MNIST classification results. #samples gives the number of kernels that were randomly",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 3
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        207,
+                                        505,
+                                        218
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                207,
+                                                505,
+                                                218
+                                            ],
+                                            "score": 1.0,
+                                            "content": "sampled for the hyperparameter search. “ConvNet GP” and “Residual CNN GP” are random CNN",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 4
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        217,
+                                        505,
+                                        230
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                217,
+                                                505,
+                                                230
+                                            ],
+                                            "score": 1.0,
+                                            "content": "architectures with a fixed filter size, whereas “ResNet GP” is a slight modification of the architecture",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 5
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        228,
+                                        504,
+                                        241
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                228,
+                                                504,
+                                                241
+                                            ],
+                                            "score": 1.0,
+                                            "content": "by He et al. (2016b). Entries labelled “SGD” used stochastic gradient descent for tuning hyperpa-",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 6
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        240,
+                                        505,
+                                        252
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                240,
+                                                505,
+                                                252
+                                            ],
+                                            "score": 1.0,
+                                            "content": "rameters, by maximising the likelihood of the training set. The last two methods use parametric",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 7
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        251,
+                                        505,
+                                        263
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                251,
+                                                505,
+                                                263
+                                            ],
+                                            "score": 1.0,
+                                            "content": "neural networks. The hyperparameters of the ResNet GP were not optimised (they were fixed based",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 8
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        262,
+                                        276,
+                                        273
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                262,
+                                                276,
+                                                273
+                                            ],
+                                            "score": 1.0,
+                                            "content": "on the architecture from He et al., 2016b).",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 9
+                                }
+                            ],
+                            "index": 6
+                        }
+                    ],
+                    "index": 3.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        292,
+                        504,
+                        348
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                291,
+                                506,
+                                305
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        291,
+                                        506,
+                                        305
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The “ResNet GP” (Table 1) is the kernel equivalent to a 32-layer version of the basic residual ar-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                304,
+                                505,
+                                316
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        304,
+                                        356,
+                                        316
+                                    ],
+                                    "score": 1.0,
+                                    "content": "chitecture by He et al. (2016a). The differences are: an initial",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        356,
+                                        304,
+                                        380,
+                                        315
+                                    ],
+                                    "score": 0.9,
+                                    "content": "3 \\times 3",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        380,
+                                        304,
+                                        505,
+                                        316
+                                    ],
+                                    "score": 1.0,
+                                    "content": "convolutional layer and a final",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                314,
+                                506,
+                                328
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        314,
+                                        506,
+                                        328
+                                    ],
+                                    "score": 1.0,
+                                    "content": "dense layer instead of average pooling. We chose to remove the pooling because computing its out-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                325,
+                                506,
+                                339
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        325,
+                                        506,
+                                        339
+                                    ],
+                                    "score": 1.0,
+                                    "content": "put variance requires the off-diagonal elements of the filter covariance, in which case we could not",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                337,
+                                304,
+                                349
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        337,
+                                        304,
+                                        349
+                                    ],
+                                    "score": 1.0,
+                                    "content": "exploit the efficiency gains described in Sec. 3.3.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        }
+                    ],
+                    "index": 12,
+                    "bbox_fs": [
+                        104,
+                        291,
+                        506,
+                        349
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        353,
+                        505,
+                        419
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                353,
+                                505,
+                                366
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        353,
+                                        505,
+                                        366
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We found that the despite it not being optimised, the 32-layer ResNet GP outperformed all other",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                365,
+                                505,
+                                376
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        365,
+                                        505,
+                                        376
+                                    ],
+                                    "score": 1.0,
+                                    "content": "comparable architectures (Table 1), including the NNGP in Lee et al. (2017), which is state-of-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                375,
+                                505,
+                                388
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        375,
+                                        505,
+                                        388
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the-art for non-convolutional networks, and convolutional GPs (van der Wilk et al., 2017; Kumar",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                386,
+                                506,
+                                399
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        386,
+                                        506,
+                                        399
+                                    ],
+                                    "score": 1.0,
+                                    "content": "et al., 2018). That said, our results have not reached state-of-the-art for methods that incorporate a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                397,
+                                506,
+                                411
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        397,
+                                        506,
+                                        411
+                                    ],
+                                    "score": 1.0,
+                                    "content": "parametric neural network, such as a standard ResNet (Chen et al., 2018) and a Gaussian process",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                408,
+                                340,
+                                420
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        408,
+                                        340,
+                                        420
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with a deep neural network kernel Bradshaw et al. (2017).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        }
+                    ],
+                    "index": 17.5,
+                    "bbox_fs": [
+                        104,
+                        353,
+                        506,
+                        420
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        425,
+                        505,
+                        535
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                424,
+                                505,
+                                438
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        424,
+                                        505,
+                                        438
+                                    ],
+                                    "score": 1.0,
+                                    "content": "To check whether the GP limit is applicable to relatively small networks used practically (with of the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                436,
+                                505,
+                                449
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        436,
+                                        505,
+                                        449
+                                    ],
+                                    "score": 1.0,
+                                    "content": "order of 100 channels in the first layers), we randomly sampled 10, 000 32-layer ResNets, with 3,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                446,
+                                506,
+                                460
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        446,
+                                        506,
+                                        460
+                                    ],
+                                    "score": 1.0,
+                                    "content": "10, 30 and 100 channels in the first layers, and, following the usual practice for ResNets we increase",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                458,
+                                505,
+                                471
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        458,
+                                        505,
+                                        471
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the number the number of hidden units when we downsample the feature maps. The probability",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                469,
+                                505,
+                                482
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        469,
+                                        505,
+                                        482
+                                    ],
+                                    "score": 1.0,
+                                    "content": "density plots show a good match around 100 channels (Fig. 2A), which matches a more sensitive",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                480,
+                                506,
+                                493
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        480,
+                                        506,
+                                        493
+                                    ],
+                                    "score": 1.0,
+                                    "content": "graphical procedure based on quantile-quantile plots (Fig. 2B). Notably, even for only 30 channels,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                490,
+                                506,
+                                505
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        490,
+                                        506,
+                                        505
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the moments match closely (Fig. 2C). For comparison, typical ResNets use from 64 (He et al.,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                501,
+                                505,
+                                514
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        501,
+                                        505,
+                                        514
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2016a) to 192 (Zagoruyko & Komodakis, 2016) channels in their first layers. We believe that this is",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                513,
+                                505,
+                                526
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        513,
+                                        505,
+                                        526
+                                    ],
+                                    "score": 1.0,
+                                    "content": "because the moment propagation equations only require the Gaussianity assumption for propagation",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                524,
+                                432,
+                                537
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        524,
+                                        432,
+                                        537
+                                    ],
+                                    "score": 1.0,
+                                    "content": "through the relu, and presumably this is robust to non-Gaussian input activations.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        }
+                    ],
+                    "index": 25.5,
+                    "bbox_fs": [
+                        105,
+                        424,
+                        506,
+                        537
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        547,
+                        505,
+                        636
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                545,
+                                506,
+                                561
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        545,
+                                        434,
+                                        561
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Computational efficiency. Asymptotically, computing the kernel matrix takes",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        435,
+                                        547,
+                                        480,
+                                        560
+                                    ],
+                                    "score": 0.92,
+                                    "content": "O ( N ^ { 2 } L D )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        480,
+                                        545,
+                                        506,
+                                        561
+                                    ],
+                                    "score": 1.0,
+                                    "content": "time,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                558,
+                                505,
+                                571
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        558,
+                                        132,
+                                        571
+                                    ],
+                                    "score": 1.0,
+                                    "content": "where",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        133,
+                                        559,
+                                        141,
+                                        568
+                                    ],
+                                    "score": 0.79,
+                                    "content": "L",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        141,
+                                        558,
+                                        307,
+                                        571
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is the number of layers in the network and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        307,
+                                        559,
+                                        317,
+                                        568
+                                    ],
+                                    "score": 0.82,
+                                    "content": "D",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        317,
+                                        558,
+                                        505,
+                                        571
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is the dimensionality of the input, and inverting",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                569,
+                                506,
+                                582
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        569,
+                                        200,
+                                        582
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the kernel matrix takes",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        200,
+                                        569,
+                                        230,
+                                        581
+                                    ],
+                                    "score": 0.92,
+                                    "content": "O ( \\bar { N } ^ { 3 } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        230,
+                                        569,
+                                        506,
+                                        582
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". As such, we expect that for very large datasets, inverting the kernel",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                579,
+                                506,
+                                593
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        579,
+                                        380,
+                                        593
+                                    ],
+                                    "score": 1.0,
+                                    "content": "matrix will dominate the computation time. However, on MNIST,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        381,
+                                        580,
+                                        396,
+                                        591
+                                    ],
+                                    "score": 0.88,
+                                    "content": "N ^ { 3 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        396,
+                                        579,
+                                        506,
+                                        593
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is only around a factor of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                591,
+                                505,
+                                604
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        591,
+                                        168,
+                                        604
+                                    ],
+                                    "score": 1.0,
+                                    "content": "10 larger than",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        168,
+                                        591,
+                                        198,
+                                        602
+                                    ],
+                                    "score": 0.9,
+                                    "content": "N ^ { 2 } L D",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        198,
+                                        591,
+                                        505,
+                                        604
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". In practice, we found that it was more expensive to compute the kernel",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                601,
+                                504,
+                                616
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        601,
+                                        437,
+                                        616
+                                    ],
+                                    "score": 1.0,
+                                    "content": "matrix than to invert it. For the ResNet kernel, the most expensive, computing",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        437,
+                                        603,
+                                        456,
+                                        614
+                                    ],
+                                    "score": 0.9,
+                                    "content": "{ \\bf K } _ { x x }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        457,
+                                        601,
+                                        479,
+                                        616
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        480,
+                                        603,
+                                        504,
+                                        613
+                                    ],
+                                    "score": 0.9,
+                                    "content": "{ \\bf K } _ { x x * }",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                613,
+                                505,
+                                626
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        613,
+                                        466,
+                                        626
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for validation and test took 3h 40min on two Tesla P100 GPUs. In contrast, inverting",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        466,
+                                        614,
+                                        486,
+                                        624
+                                    ],
+                                    "score": 0.89,
+                                    "content": "{ \\bf K } _ { x x }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        486,
+                                        613,
+                                        505,
+                                        626
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                624,
+                                501,
+                                636
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        624,
+                                        300,
+                                        636
+                                    ],
+                                    "score": 1.0,
+                                    "content": "computing validation and test performance took",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        300,
+                                        624,
+                                        349,
+                                        635
+                                    ],
+                                    "score": 0.87,
+                                    "content": "4 3 . 2 5 \\pm 8 . 8",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        350,
+                                        624,
+                                        501,
+                                        636
+                                    ],
+                                    "score": 1.0,
+                                    "content": "seconds on a single Tesla P100 GPU.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        }
+                    ],
+                    "index": 34.5,
+                    "bbox_fs": [
+                        105,
+                        545,
+                        506,
+                        636
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        652,
+                        209,
+                        664
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                650,
+                                211,
+                                666
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        650,
+                                        211,
+                                        666
+                                    ],
+                                    "score": 1.0,
+                                    "content": "5 RELATED WORK",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        }
+                    ],
+                    "index": 39
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        676,
+                        504,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                676,
+                                505,
+                                690
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        676,
+                                        505,
+                                        690
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Van der Wilk et al. (van der Wilk et al., 2017) also adapted GPs to image classification. They",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                688,
+                                505,
+                                700
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        688,
+                                        226,
+                                        700
+                                    ],
+                                    "score": 1.0,
+                                    "content": "defined a prior on functions",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        226,
+                                        688,
+                                        233,
+                                        699
+                                    ],
+                                    "score": 0.87,
+                                    "content": "f",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        234,
+                                        688,
+                                        505,
+                                        700
+                                    ],
+                                    "score": 1.0,
+                                    "content": "that takes an image and outputs a scalar. First, draw a function",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                107,
+                                698,
+                                506,
+                                712
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        699,
+                                        200,
+                                        711
+                                    ],
+                                    "score": 0.91,
+                                    "content": "g \\sim \\mathcal { G P } ( 0 , k _ { p } ( \\mathbf { X } , \\mathbf { X } ^ { \\prime } ) )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        201,
+                                        698,
+                                        232,
+                                        712
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Then,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        232,
+                                        699,
+                                        239,
+                                        711
+                                    ],
+                                    "score": 0.85,
+                                    "content": "f",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        240,
+                                        698,
+                                        350,
+                                        712
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is the sum of the output of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        351,
+                                        700,
+                                        357,
+                                        710
+                                    ],
+                                    "score": 0.81,
+                                    "content": "g",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        357,
+                                        698,
+                                        506,
+                                        712
+                                    ],
+                                    "score": 1.0,
+                                    "content": "applied to each of the convolutional",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                709,
+                                506,
+                                723
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        709,
+                                        427,
+                                        723
+                                    ],
+                                    "score": 1.0,
+                                    "content": "patches. Their approach is also inspired by convolutional NNs, but their kernel",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        428,
+                                        710,
+                                        439,
+                                        722
+                                    ],
+                                    "score": 0.88,
+                                    "content": "k _ { p }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        439,
+                                        709,
+                                        506,
+                                        723
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is applied to all",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                718,
+                                506,
+                                735
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        718,
+                                        181,
+                                        735
+                                    ],
+                                    "score": 1.0,
+                                    "content": "pairs of patches of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        182,
+                                        721,
+                                        191,
+                                        730
+                                    ],
+                                    "score": 0.68,
+                                    "content": "\\mathbf { X }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        191,
+                                        718,
+                                        209,
+                                        735
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        209,
+                                        721,
+                                        221,
+                                        730
+                                    ],
+                                    "score": 0.86,
+                                    "content": "\\mathbf { X } ^ { \\prime }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        222,
+                                        718,
+                                        506,
+                                        735
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". This makes their convolutional kernel expensive to evaluate, requiring",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                455,
+                                505,
+                                468
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        455,
+                                        505,
+                                        468
+                                    ],
+                                    "score": 1.0,
+                                    "content": "inter-domain inducing point approximations to remain tractable. The kernels in this work, directly",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                466,
+                                505,
+                                479
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        466,
+                                        405,
+                                        479
+                                    ],
+                                    "score": 1.0,
+                                    "content": "motivated by the infinite-filter limit of a CNN, only apply something like",
+                                    "type": "text",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        406,
+                                        467,
+                                        417,
+                                        479
+                                    ],
+                                    "score": 0.88,
+                                    "content": "k _ { p }",
+                                    "type": "inline_equation",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        417,
+                                        466,
+                                        505,
+                                        479
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to the corresponding",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                477,
+                                506,
+                                490
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        478,
+                                        200,
+                                        490
+                                    ],
+                                    "score": 1.0,
+                                    "content": "pairs of patches within",
+                                    "type": "text",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        201,
+                                        478,
+                                        211,
+                                        488
+                                    ],
+                                    "score": 0.33,
+                                    "content": "\\mathbf { X }",
+                                    "type": "inline_equation",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        212,
+                                        478,
+                                        229,
+                                        490
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        230,
+                                        477,
+                                        243,
+                                        488
+                                    ],
+                                    "score": 0.83,
+                                    "content": "\\mathbf { X } ^ { \\prime }",
+                                    "type": "inline_equation",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        243,
+                                        478,
+                                        506,
+                                        490
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(Eq. 10). As such, the CNN kernels are cheaper to compute and",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                489,
+                                482,
+                                501
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        489,
+                                        482,
+                                        501
+                                    ],
+                                    "score": 1.0,
+                                    "content": "exhibit superior performance (Table 1), despite the use of an approximate likelihood function.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 11
+                        }
+                    ],
+                    "index": 42,
+                    "bbox_fs": [
+                        105,
+                        676,
+                        506,
+                        735
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "image",
+                    "bbox": [
+                        110,
+                        79,
+                        506,
+                        349
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                110,
+                                79,
+                                506,
+                                349
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        110,
+                                        79,
+                                        506,
+                                        349
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                110,
+                                                79,
+                                                506,
+                                                349
+                                            ],
+                                            "score": 0.974,
+                                            "type": "image",
+                                            "image_path": "855e86413bd7de48d01d9cca88bd8f2c7466cd7a38d25eebae9541aa8c975eaf.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 1,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        110,
+                                        79,
+                                        506,
+                                        169.0
+                                    ],
+                                    "spans": [],
+                                    "index": 0
+                                },
+                                {
+                                    "bbox": [
+                                        110,
+                                        169.0,
+                                        506,
+                                        259.0
+                                    ],
+                                    "spans": [],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        110,
+                                        259.0,
+                                        506,
+                                        349.0
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                106,
+                                371,
+                                505,
+                                427
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        371,
+                                        505,
+                                        384
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                371,
+                                                505,
+                                                384
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 2: Comparison of the infinite limit, and outputs from finite 32-layer ResNets with 3, 10,",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 3
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        381,
+                                        505,
+                                        396
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                381,
+                                                505,
+                                                396
+                                            ],
+                                            "score": 1.0,
+                                            "content": "30, and 100 channels in their first layers. A Comparison of the empirical and limiting probability",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 4
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        393,
+                                        506,
+                                        407
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                393,
+                                                506,
+                                                407
+                                            ],
+                                            "score": 1.0,
+                                            "content": "densities. B A more sensitive test of Gaussianity is a quantile-quantile plot, which shows converges",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 5
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        403,
+                                        505,
+                                        418
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                403,
+                                                505,
+                                                418
+                                            ],
+                                            "score": 1.0,
+                                            "content": "with 100 channels. C The moments (variances and covariances) for 100 training inputs shows gives",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 6
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        415,
+                                        276,
+                                        428
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                415,
+                                                276,
+                                                428
+                                            ],
+                                            "score": 1.0,
+                                            "content": "a good match for all numbers of channels.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 7
+                                }
+                            ],
+                            "index": 5
+                        }
+                    ],
+                    "index": 3.0
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        455,
+                        505,
+                        500
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                455,
+                                505,
+                                468
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        455,
+                                        505,
+                                        468
+                                    ],
+                                    "score": 1.0,
+                                    "content": "inter-domain inducing point approximations to remain tractable. The kernels in this work, directly",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                466,
+                                505,
+                                479
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        466,
+                                        405,
+                                        479
+                                    ],
+                                    "score": 1.0,
+                                    "content": "motivated by the infinite-filter limit of a CNN, only apply something like",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        406,
+                                        467,
+                                        417,
+                                        479
+                                    ],
+                                    "score": 0.88,
+                                    "content": "k _ { p }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        417,
+                                        466,
+                                        505,
+                                        479
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to the corresponding",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                477,
+                                506,
+                                490
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        478,
+                                        200,
+                                        490
+                                    ],
+                                    "score": 1.0,
+                                    "content": "pairs of patches within",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        201,
+                                        478,
+                                        211,
+                                        488
+                                    ],
+                                    "score": 0.33,
+                                    "content": "\\mathbf { X }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        212,
+                                        478,
+                                        229,
+                                        490
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        230,
+                                        477,
+                                        243,
+                                        488
+                                    ],
+                                    "score": 0.83,
+                                    "content": "\\mathbf { X } ^ { \\prime }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        243,
+                                        478,
+                                        506,
+                                        490
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(Eq. 10). As such, the CNN kernels are cheaper to compute and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                489,
+                                482,
+                                501
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        489,
+                                        482,
+                                        501
+                                    ],
+                                    "score": 1.0,
+                                    "content": "exhibit superior performance (Table 1), despite the use of an approximate likelihood function.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        }
+                    ],
+                    "index": 9.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        505,
+                        505,
+                        539
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                505,
+                                505,
+                                519
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        505,
+                                        505,
+                                        519
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Kumar et al. (2018) define a prior over functions by stacking several GPs with van der Wilk’s con-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                516,
+                                505,
+                                529
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        516,
+                                        505,
+                                        529
+                                    ],
+                                    "score": 1.0,
+                                    "content": "volutional kernel, forming a “Deep GP” (Damianou & Lawrence, 2013). In contrast, the kernel in",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                527,
+                                489,
+                                541
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        527,
+                                        489,
+                                        541
+                                    ],
+                                    "score": 1.0,
+                                    "content": "this paper confines all hierarchy to the definition of the kernel, and the resulting GPs is shallow.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        }
+                    ],
+                    "index": 13
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        544,
+                        505,
+                        643
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                543,
+                                506,
+                                557
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        543,
+                                        506,
+                                        557
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Wilson et al. (2016) introduced and Bradshaw et al. (2017) improved deep kernel learning. The",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                555,
+                                505,
+                                569
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        555,
+                                        221,
+                                        569
+                                    ],
+                                    "score": 1.0,
+                                    "content": "inputs to a classic GP kernel",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        221,
+                                        556,
+                                        228,
+                                        565
+                                    ],
+                                    "score": 0.76,
+                                    "content": "k",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        228,
+                                        555,
+                                        466,
+                                        569
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(e.g. RBF) are preprocessed by applying a feature extractor",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        466,
+                                        557,
+                                        473,
+                                        567
+                                    ],
+                                    "score": 0.79,
+                                    "content": "g",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        473,
+                                        555,
+                                        505,
+                                        569
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(a deep",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                565,
+                                506,
+                                580
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        565,
+                                        251,
+                                        580
+                                    ],
+                                    "score": 1.0,
+                                    "content": "NN) prior to computing the kernel:",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        252,
+                                        566,
+                                        402,
+                                        578
+                                    ],
+                                    "score": 0.92,
+                                    "content": "k _ { \\mathrm { d e e p } } ( \\mathbf { X } , \\bar { \\mathbf { X } } ^ { \\prime } ) : = k ( g ( \\mathbf { X } ; \\theta ) , \\bar { g ( \\mathbf { X } ^ { \\prime } , \\theta ) } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        403,
+                                        565,
+                                        506,
+                                        580
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". The NN parameters are",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                575,
+                                506,
+                                592
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        575,
+                                        506,
+                                        592
+                                    ],
+                                    "score": 1.0,
+                                    "content": "optimised by gradient ascent using the likelihood as the objective, as in standard GP kernel learning",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                588,
+                                506,
+                                601
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        588,
+                                        506,
+                                        601
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(Rasmussen & Williams, 2006, Chapter 5). Since deep kernel learning incorporates a state-of-the-art",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                598,
+                                506,
+                                613
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        598,
+                                        162,
+                                        613
+                                    ],
+                                    "score": 1.0,
+                                    "content": "NN with over",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        162,
+                                        599,
+                                        178,
+                                        609
+                                    ],
+                                    "score": 0.87,
+                                    "content": "1 0 ^ { 6 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        178,
+                                        598,
+                                        506,
+                                        613
+                                    ],
+                                    "score": 1.0,
+                                    "content": "parameters, we expect it to perform similarly to a NN applied directly to the task of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                609,
+                                505,
+                                623
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        609,
+                                        505,
+                                        623
+                                    ],
+                                    "score": 1.0,
+                                    "content": "image classification. At present both CNNs and deep kernel learning display superior performance",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                621,
+                                505,
+                                634
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        621,
+                                        505,
+                                        634
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to the GP kernels in this work. However, the kernels defined here have far fewer parameters (around",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                631,
+                                217,
+                                644
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        631,
+                                        194,
+                                        644
+                                    ],
+                                    "score": 1.0,
+                                    "content": "10, compared to their",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        194,
+                                        632,
+                                        210,
+                                        642
+                                    ],
+                                    "score": 0.85,
+                                    "content": "1 0 ^ { 6 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        210,
+                                        631,
+                                        217,
+                                        644
+                                    ],
+                                    "score": 1.0,
+                                    "content": ").",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        }
+                    ],
+                    "index": 19
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        649,
+                        505,
+                        682
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                648,
+                                505,
+                                661
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        648,
+                                        505,
+                                        661
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Borovykh (2018) also suggests that a CNN exhibits GP behaviour. However, they take the infi-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                660,
+                                505,
+                                671
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        660,
+                                        505,
+                                        671
+                                    ],
+                                    "score": 1.0,
+                                    "content": "nite limit with respect to the filter size, not the number of filters. Thus, their infinite network is",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                671,
+                                350,
+                                683
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        671,
+                                        350,
+                                        683
+                                    ],
+                                    "score": 1.0,
+                                    "content": "inapplicable to real data which is always of finite dimension.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        }
+                    ],
+                    "index": 25
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        687,
+                        504,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                687,
+                                505,
+                                700
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        687,
+                                        505,
+                                        700
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Finally, there is a series of papers analysing the mean-field behaviour of deep NNs and CNNs which",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                698,
+                                506,
+                                712
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        698,
+                                        506,
+                                        712
+                                    ],
+                                    "score": 1.0,
+                                    "content": "aims to find good random initializations, i.e. those that do not exhibit vanishing or exploding gra-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                709,
+                                505,
+                                722
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        709,
+                                        505,
+                                        722
+                                    ],
+                                    "score": 1.0,
+                                    "content": "dients or activations (Schoenholz et al., 2016; Yang & Schoenholz, 2017). Apart from their very",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                720,
+                                505,
+                                734
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        720,
+                                        505,
+                                        734
+                                    ],
+                                    "score": 1.0,
+                                    "content": "different focus, the key difference to our work is that they compute the variance for a single training-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        }
+                    ],
+                    "index": 28.5
+                }
+            ],
+            "page_idx": 8,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        107,
+                        27,
+                        293,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                26,
+                                293,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        26,
+                                        293,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Published as a conference paper at ICLR 2019",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        302,
+                        751,
+                        308,
+                        759
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                302,
+                                751,
+                                309,
+                                762
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        302,
+                                        751,
+                                        309,
+                                        762
+                                    ],
+                                    "score": 1.0,
+                                    "content": "9",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "image",
+                    "bbox": [
+                        110,
+                        79,
+                        506,
+                        349
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                110,
+                                79,
+                                506,
+                                349
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        110,
+                                        79,
+                                        506,
+                                        349
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                110,
+                                                79,
+                                                506,
+                                                349
+                                            ],
+                                            "score": 0.974,
+                                            "type": "image",
+                                            "image_path": "855e86413bd7de48d01d9cca88bd8f2c7466cd7a38d25eebae9541aa8c975eaf.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 1,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        110,
+                                        79,
+                                        506,
+                                        169.0
+                                    ],
+                                    "spans": [],
+                                    "index": 0
+                                },
+                                {
+                                    "bbox": [
+                                        110,
+                                        169.0,
+                                        506,
+                                        259.0
+                                    ],
+                                    "spans": [],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        110,
+                                        259.0,
+                                        506,
+                                        349.0
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                106,
+                                371,
+                                505,
+                                427
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        371,
+                                        505,
+                                        384
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                371,
+                                                505,
+                                                384
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 2: Comparison of the infinite limit, and outputs from finite 32-layer ResNets with 3, 10,",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 3
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        381,
+                                        505,
+                                        396
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                381,
+                                                505,
+                                                396
+                                            ],
+                                            "score": 1.0,
+                                            "content": "30, and 100 channels in their first layers. A Comparison of the empirical and limiting probability",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 4
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        393,
+                                        506,
+                                        407
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                393,
+                                                506,
+                                                407
+                                            ],
+                                            "score": 1.0,
+                                            "content": "densities. B A more sensitive test of Gaussianity is a quantile-quantile plot, which shows converges",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 5
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        403,
+                                        505,
+                                        418
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                403,
+                                                505,
+                                                418
+                                            ],
+                                            "score": 1.0,
+                                            "content": "with 100 channels. C The moments (variances and covariances) for 100 training inputs shows gives",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 6
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        415,
+                                        276,
+                                        428
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                415,
+                                                276,
+                                                428
+                                            ],
+                                            "score": 1.0,
+                                            "content": "a good match for all numbers of channels.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 7
+                                }
+                            ],
+                            "index": 5
+                        }
+                    ],
+                    "index": 3.0
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        455,
+                        505,
+                        500
+                    ],
+                    "lines": [],
+                    "index": 9.5,
+                    "bbox_fs": [
+                        105,
+                        455,
+                        506,
+                        501
+                    ],
+                    "lines_deleted": true
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        505,
+                        505,
+                        539
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                505,
+                                505,
+                                519
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        505,
+                                        505,
+                                        519
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Kumar et al. (2018) define a prior over functions by stacking several GPs with van der Wilk’s con-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                516,
+                                505,
+                                529
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        516,
+                                        505,
+                                        529
+                                    ],
+                                    "score": 1.0,
+                                    "content": "volutional kernel, forming a “Deep GP” (Damianou & Lawrence, 2013). In contrast, the kernel in",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                527,
+                                489,
+                                541
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        527,
+                                        489,
+                                        541
+                                    ],
+                                    "score": 1.0,
+                                    "content": "this paper confines all hierarchy to the definition of the kernel, and the resulting GPs is shallow.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        }
+                    ],
+                    "index": 13,
+                    "bbox_fs": [
+                        105,
+                        505,
+                        505,
+                        541
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        544,
+                        505,
+                        643
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                543,
+                                506,
+                                557
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        543,
+                                        506,
+                                        557
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Wilson et al. (2016) introduced and Bradshaw et al. (2017) improved deep kernel learning. The",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                555,
+                                505,
+                                569
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        555,
+                                        221,
+                                        569
+                                    ],
+                                    "score": 1.0,
+                                    "content": "inputs to a classic GP kernel",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        221,
+                                        556,
+                                        228,
+                                        565
+                                    ],
+                                    "score": 0.76,
+                                    "content": "k",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        228,
+                                        555,
+                                        466,
+                                        569
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(e.g. RBF) are preprocessed by applying a feature extractor",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        466,
+                                        557,
+                                        473,
+                                        567
+                                    ],
+                                    "score": 0.79,
+                                    "content": "g",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        473,
+                                        555,
+                                        505,
+                                        569
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(a deep",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                565,
+                                506,
+                                580
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        565,
+                                        251,
+                                        580
+                                    ],
+                                    "score": 1.0,
+                                    "content": "NN) prior to computing the kernel:",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        252,
+                                        566,
+                                        402,
+                                        578
+                                    ],
+                                    "score": 0.92,
+                                    "content": "k _ { \\mathrm { d e e p } } ( \\mathbf { X } , \\bar { \\mathbf { X } } ^ { \\prime } ) : = k ( g ( \\mathbf { X } ; \\theta ) , \\bar { g ( \\mathbf { X } ^ { \\prime } , \\theta ) } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        403,
+                                        565,
+                                        506,
+                                        580
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". The NN parameters are",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                575,
+                                506,
+                                592
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        575,
+                                        506,
+                                        592
+                                    ],
+                                    "score": 1.0,
+                                    "content": "optimised by gradient ascent using the likelihood as the objective, as in standard GP kernel learning",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                588,
+                                506,
+                                601
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        588,
+                                        506,
+                                        601
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(Rasmussen & Williams, 2006, Chapter 5). Since deep kernel learning incorporates a state-of-the-art",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                598,
+                                506,
+                                613
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        598,
+                                        162,
+                                        613
+                                    ],
+                                    "score": 1.0,
+                                    "content": "NN with over",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        162,
+                                        599,
+                                        178,
+                                        609
+                                    ],
+                                    "score": 0.87,
+                                    "content": "1 0 ^ { 6 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        178,
+                                        598,
+                                        506,
+                                        613
+                                    ],
+                                    "score": 1.0,
+                                    "content": "parameters, we expect it to perform similarly to a NN applied directly to the task of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                609,
+                                505,
+                                623
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        609,
+                                        505,
+                                        623
+                                    ],
+                                    "score": 1.0,
+                                    "content": "image classification. At present both CNNs and deep kernel learning display superior performance",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                621,
+                                505,
+                                634
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        621,
+                                        505,
+                                        634
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to the GP kernels in this work. However, the kernels defined here have far fewer parameters (around",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                631,
+                                217,
+                                644
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        631,
+                                        194,
+                                        644
+                                    ],
+                                    "score": 1.0,
+                                    "content": "10, compared to their",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        194,
+                                        632,
+                                        210,
+                                        642
+                                    ],
+                                    "score": 0.85,
+                                    "content": "1 0 ^ { 6 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        210,
+                                        631,
+                                        217,
+                                        644
+                                    ],
+                                    "score": 1.0,
+                                    "content": ").",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        }
+                    ],
+                    "index": 19,
+                    "bbox_fs": [
+                        104,
+                        543,
+                        506,
+                        644
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        649,
+                        505,
+                        682
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                648,
+                                505,
+                                661
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        648,
+                                        505,
+                                        661
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Borovykh (2018) also suggests that a CNN exhibits GP behaviour. However, they take the infi-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                660,
+                                505,
+                                671
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        660,
+                                        505,
+                                        671
+                                    ],
+                                    "score": 1.0,
+                                    "content": "nite limit with respect to the filter size, not the number of filters. Thus, their infinite network is",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                671,
+                                350,
+                                683
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        671,
+                                        350,
+                                        683
+                                    ],
+                                    "score": 1.0,
+                                    "content": "inapplicable to real data which is always of finite dimension.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        }
+                    ],
+                    "index": 25,
+                    "bbox_fs": [
+                        105,
+                        648,
+                        505,
+                        683
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        687,
+                        504,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                687,
+                                505,
+                                700
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        687,
+                                        505,
+                                        700
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Finally, there is a series of papers analysing the mean-field behaviour of deep NNs and CNNs which",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                698,
+                                506,
+                                712
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        698,
+                                        506,
+                                        712
+                                    ],
+                                    "score": 1.0,
+                                    "content": "aims to find good random initializations, i.e. those that do not exhibit vanishing or exploding gra-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                709,
+                                505,
+                                722
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        709,
+                                        505,
+                                        722
+                                    ],
+                                    "score": 1.0,
+                                    "content": "dients or activations (Schoenholz et al., 2016; Yang & Schoenholz, 2017). Apart from their very",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                720,
+                                505,
+                                734
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        720,
+                                        505,
+                                        734
+                                    ],
+                                    "score": 1.0,
+                                    "content": "different focus, the key difference to our work is that they compute the variance for a single training-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                81,
+                                505,
+                                96
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        81,
+                                        505,
+                                        96
+                                    ],
+                                    "score": 1.0,
+                                    "content": "example, whereas to obtain the GPs kernel, we additionally need to compute the output covariances",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                93,
+                                325,
+                                106
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        93,
+                                        325,
+                                        106
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for different training/test examples (Xiao et al., 2018).",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 1
+                        }
+                    ],
+                    "index": 28.5,
+                    "bbox_fs": [
+                        105,
+                        687,
+                        506,
+                        734
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        82,
+                        504,
+                        105
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                81,
+                                505,
+                                96
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        81,
+                                        505,
+                                        96
+                                    ],
+                                    "score": 1.0,
+                                    "content": "example, whereas to obtain the GPs kernel, we additionally need to compute the output covariances",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                93,
+                                325,
+                                106
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        93,
+                                        325,
+                                        106
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for different training/test examples (Xiao et al., 2018).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        }
+                    ],
+                    "index": 0.5
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        123,
+                        303,
+                        135
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                122,
+                                306,
+                                137
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        122,
+                                        306,
+                                        137
+                                    ],
+                                    "score": 1.0,
+                                    "content": "6 CONCLUSIONS AND FUTURE WORK",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        }
+                    ],
+                    "index": 2
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        148,
+                        505,
+                        237
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                149,
+                                505,
+                                161
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        149,
+                                        505,
+                                        161
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We have shown that deep Bayesian CNNs with infinitely many filters are equivalent to a GP with",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                160,
+                                505,
+                                171
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        160,
+                                        505,
+                                        171
+                                    ],
+                                    "score": 1.0,
+                                    "content": "a recursive kernel. We also derived the kernel for the GP equivalent to a CNN, and showed that,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                170,
+                                505,
+                                183
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        170,
+                                        505,
+                                        183
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in handwritten digit classification, it outperforms all previous GP approaches that do not incorpo-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                182,
+                                505,
+                                193
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        182,
+                                        505,
+                                        193
+                                    ],
+                                    "score": 1.0,
+                                    "content": "rate a parametric NN into the kernel. Given that most state-of-the-art neural networks incorporate",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                192,
+                                505,
+                                204
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        192,
+                                        505,
+                                        204
+                                    ],
+                                    "score": 1.0,
+                                    "content": "structure (convolutional or otherwise) into their architecture, the equivalence between CNNs and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                203,
+                                505,
+                                216
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        203,
+                                        505,
+                                        216
+                                    ],
+                                    "score": 1.0,
+                                    "content": "GPs is potentially of considerable practical relevance. In particular, we hope to apply GP CNNs in",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                214,
+                                505,
+                                227
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        214,
+                                        386,
+                                        227
+                                    ],
+                                    "score": 1.0,
+                                    "content": "domains as widespread as adversarial examples, lifelong learning and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        387,
+                                        215,
+                                        393,
+                                        225
+                                    ],
+                                    "score": 0.31,
+                                    "content": "\\mathbf { k }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        394,
+                                        214,
+                                        505,
+                                        227
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-shot learning, and we hope",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                226,
+                                484,
+                                238
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        226,
+                                        484,
+                                        238
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to improve them by developing efficient multi-layered inducing point approximation schemes.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        }
+                    ],
+                    "index": 6.5
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        255,
+                        175,
+                        266
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                255,
+                                176,
+                                268
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        255,
+                                        176,
+                                        268
+                                    ],
+                                    "score": 1.0,
+                                    "content": "REFERENCES",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        }
+                    ],
+                    "index": 11
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        274,
+                        504,
+                        297
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                274,
+                                505,
+                                287
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        274,
+                                        505,
+                                        287
+                                    ],
+                                    "score": 1.0,
+                                    "content": "JR Blum, H Chernoff, M Rosenblatt, and H Teicher. Central limit theorems for interchangeable",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                285,
+                                305,
+                                297
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        285,
+                                        305,
+                                        297
+                                    ],
+                                    "score": 1.0,
+                                    "content": "processes. Canad. J. Math, 10:222–229, 1958.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        }
+                    ],
+                    "index": 12.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        306,
+                        504,
+                        330
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                306,
+                                505,
+                                320
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        306,
+                                        505,
+                                        320
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Charles Blundell, Julien Cornebise, Koray Kavukcuoglu, and Daan Wierstra. Weight uncertainty in",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                317,
+                                477,
+                                330
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        317,
+                                        477,
+                                        330
+                                    ],
+                                    "score": 1.0,
+                                    "content": "neural network. In International Conference on Machine Learning, pp. 1613–1622, 2015.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        }
+                    ],
+                    "index": 14.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        339,
+                        503,
+                        362
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                338,
+                                505,
+                                353
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        338,
+                                        505,
+                                        353
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Anastasia Borovykh. A gaussian process perspective on convolutional neural networks. Research-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                350,
+                                178,
+                                362
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        350,
+                                        178,
+                                        362
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Gate, 05 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        }
+                    ],
+                    "index": 16.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        371,
+                        506,
+                        405
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                370,
+                                505,
+                                385
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        370,
+                                        505,
+                                        385
+                                    ],
+                                    "score": 1.0,
+                                    "content": "John Bradshaw, Alexander G de G Matthews, and Zoubin Ghahramani. Adversarial examples,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                383,
+                                505,
+                                396
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        383,
+                                        505,
+                                        396
+                                    ],
+                                    "score": 1.0,
+                                    "content": "uncertainty, and transfer testing robustness in gaussian process hybrid deep networks. arXiv",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                394,
+                                254,
+                                405
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        394,
+                                        254,
+                                        405
+                                    ],
+                                    "score": 1.0,
+                                    "content": "preprint arXiv:1707.02476, 2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        }
+                    ],
+                    "index": 19
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        415,
+                        504,
+                        438
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                415,
+                                505,
+                                428
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        415,
+                                        505,
+                                        428
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Tian Qi Chen, Yulia Rubanova, Jesse Bettencourt, and David Duvenaud. Neural ordinary differential",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                427,
+                                322,
+                                439
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        427,
+                                        322,
+                                        439
+                                    ],
+                                    "score": 1.0,
+                                    "content": "equations. arXiv preprint arXiv:1806.07366, 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        }
+                    ],
+                    "index": 21.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        448,
+                        504,
+                        482
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                107,
+                                447,
+                                504,
+                                460
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        447,
+                                        504,
+                                        460
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Youngmin Cho and Lawrence K Saul. Kernel methods for deep learning. In Advances in neu-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                459,
+                                504,
+                                472
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        459,
+                                        504,
+                                        472
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ral information processing systems, pp. 342–350, 2009. URL http://papers.nips.cc/",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                470,
+                                405,
+                                482
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        470,
+                                        405,
+                                        482
+                                    ],
+                                    "score": 1.0,
+                                    "content": "paper/3628-kernel-methods-for-deep-learning.pdf.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        }
+                    ],
+                    "index": 24
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        491,
+                        505,
+                        547
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                491,
+                                505,
+                                504
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        491,
+                                        505,
+                                        504
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Andreas Damianou and Neil Lawrence. Deep gaussian processes. In Carlos M. Carvalho and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                501,
+                                505,
+                                515
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        501,
+                                        505,
+                                        515
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Pradeep Ravikumar (eds.), Proceedings of the Sixteenth International Conference on Artificial",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                513,
+                                505,
+                                526
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        513,
+                                        505,
+                                        526
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Intelligence and Statistics, volume 31 of Proceedings of Machine Learning Research, pp. 207–",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                114,
+                                523,
+                                505,
+                                539
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        114,
+                                        523,
+                                        505,
+                                        539
+                                    ],
+                                    "score": 1.0,
+                                    "content": "215, Scottsdale, Arizona, USA, 29 Apr–01 May 2013. PMLR. URL http://proceedings.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                536,
+                                300,
+                                547
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        536,
+                                        300,
+                                        547
+                                    ],
+                                    "score": 1.0,
+                                    "content": "mlr.press/v31/damianou13a.html.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        }
+                    ],
+                    "index": 28
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        556,
+                        504,
+                        591
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                556,
+                                505,
+                                570
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        556,
+                                        505,
+                                        570
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Marc Deisenroth and Carl E Rasmussen. PILCO: A model-based and data-efficient approach to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                568,
+                                505,
+                                581
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        568,
+                                        505,
+                                        581
+                                    ],
+                                    "score": 1.0,
+                                    "content": "policy search. In Proceedings of the 28th International Conference on machine learning (ICML-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                579,
+                                215,
+                                591
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        579,
+                                        215,
+                                        591
+                                    ],
+                                    "score": 1.0,
+                                    "content": "11), pp. 465–472, 2011.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        }
+                    ],
+                    "index": 32
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        600,
+                        504,
+                        623
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                600,
+                                504,
+                                614
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        600,
+                                        504,
+                                        614
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Yarin Gal and Zoubin Ghahramani. Dropout as a bayesian approximation: Representing model",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                612,
+                                396,
+                                623
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        612,
+                                        396,
+                                        623
+                                    ],
+                                    "score": 1.0,
+                                    "content": "uncertainty in deep learning. arXiv preprint arXiv:1506.02142, 2015.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        }
+                    ],
+                    "index": 34.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        633,
+                        504,
+                        656
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                633,
+                                505,
+                                646
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        633,
+                                        505,
+                                        646
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Yarin Gal and Lewis Smith. Idealised bayesian neural networks cannot have adversarial examples:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                644,
+                                411,
+                                657
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        644,
+                                        411,
+                                        657
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Theoretical and empirical study. arXiv preprint arXiv:1806.00667, 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        }
+                    ],
+                    "index": 36.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        666,
+                        504,
+                        700
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                664,
+                                505,
+                                681
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        664,
+                                        505,
+                                        681
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image recog-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                676,
+                                505,
+                                691
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        676,
+                                        505,
+                                        691
+                                    ],
+                                    "score": 1.0,
+                                    "content": "nition. In Proceedings of the IEEE conference on computer vision and pattern recognition, pp.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                687,
+                                430,
+                                701
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        687,
+                                        430,
+                                        701
+                                    ],
+                                    "score": 1.0,
+                                    "content": "770–778, 2016a. URL https://arxiv.org/pdf/1512.03385.pdf.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        }
+                    ],
+                    "index": 39
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        709,
+                        504,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                709,
+                                505,
+                                722
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        709,
+                                        505,
+                                        722
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Identity mappings in deep residual",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                720,
+                                462,
+                                733
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        720,
+                                        462,
+                                        733
+                                    ],
+                                    "score": 1.0,
+                                    "content": "networks. In European conference on computer vision, pp. 630–645. Springer, 2016b.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        }
+                    ],
+                    "index": 41.5
+                }
+            ],
+            "page_idx": 9,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        108,
+                        27,
+                        293,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                26,
+                                294,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        26,
+                                        294,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Published as a conference paper at ICLR 2019",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        301,
+                        751,
+                        311,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                299,
+                                750,
+                                313,
+                                765
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        299,
+                                        750,
+                                        313,
+                                        765
+                                    ],
+                                    "score": 1.0,
+                                    "content": "",
+                                    "type": "text",
+                                    "height": 15,
+                                    "width": 14
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        82,
+                        504,
+                        105
+                    ],
+                    "lines": [],
+                    "index": 0.5,
+                    "bbox_fs": [
+                        105,
+                        81,
+                        505,
+                        106
+                    ],
+                    "lines_deleted": true
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        123,
+                        303,
+                        135
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                122,
+                                306,
+                                137
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        122,
+                                        306,
+                                        137
+                                    ],
+                                    "score": 1.0,
+                                    "content": "6 CONCLUSIONS AND FUTURE WORK",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        }
+                    ],
+                    "index": 2
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        148,
+                        505,
+                        237
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                149,
+                                505,
+                                161
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        149,
+                                        505,
+                                        161
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We have shown that deep Bayesian CNNs with infinitely many filters are equivalent to a GP with",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                160,
+                                505,
+                                171
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        160,
+                                        505,
+                                        171
+                                    ],
+                                    "score": 1.0,
+                                    "content": "a recursive kernel. We also derived the kernel for the GP equivalent to a CNN, and showed that,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                170,
+                                505,
+                                183
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        170,
+                                        505,
+                                        183
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in handwritten digit classification, it outperforms all previous GP approaches that do not incorpo-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                182,
+                                505,
+                                193
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        182,
+                                        505,
+                                        193
+                                    ],
+                                    "score": 1.0,
+                                    "content": "rate a parametric NN into the kernel. Given that most state-of-the-art neural networks incorporate",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                192,
+                                505,
+                                204
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        192,
+                                        505,
+                                        204
+                                    ],
+                                    "score": 1.0,
+                                    "content": "structure (convolutional or otherwise) into their architecture, the equivalence between CNNs and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                203,
+                                505,
+                                216
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        203,
+                                        505,
+                                        216
+                                    ],
+                                    "score": 1.0,
+                                    "content": "GPs is potentially of considerable practical relevance. In particular, we hope to apply GP CNNs in",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                214,
+                                505,
+                                227
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        214,
+                                        386,
+                                        227
+                                    ],
+                                    "score": 1.0,
+                                    "content": "domains as widespread as adversarial examples, lifelong learning and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        387,
+                                        215,
+                                        393,
+                                        225
+                                    ],
+                                    "score": 0.31,
+                                    "content": "\\mathbf { k }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        394,
+                                        214,
+                                        505,
+                                        227
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-shot learning, and we hope",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                226,
+                                484,
+                                238
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        226,
+                                        484,
+                                        238
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to improve them by developing efficient multi-layered inducing point approximation schemes.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        }
+                    ],
+                    "index": 6.5,
+                    "bbox_fs": [
+                        105,
+                        149,
+                        505,
+                        238
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        255,
+                        175,
+                        266
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                255,
+                                176,
+                                268
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        255,
+                                        176,
+                                        268
+                                    ],
+                                    "score": 1.0,
+                                    "content": "REFERENCES",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        }
+                    ],
+                    "index": 11
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        274,
+                        504,
+                        297
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                274,
+                                505,
+                                287
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        274,
+                                        505,
+                                        287
+                                    ],
+                                    "score": 1.0,
+                                    "content": "JR Blum, H Chernoff, M Rosenblatt, and H Teicher. Central limit theorems for interchangeable",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                285,
+                                305,
+                                297
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        285,
+                                        305,
+                                        297
+                                    ],
+                                    "score": 1.0,
+                                    "content": "processes. Canad. J. Math, 10:222–229, 1958.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        }
+                    ],
+                    "index": 12.5,
+                    "bbox_fs": [
+                        106,
+                        274,
+                        505,
+                        297
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        306,
+                        504,
+                        330
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                306,
+                                505,
+                                320
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        306,
+                                        505,
+                                        320
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Charles Blundell, Julien Cornebise, Koray Kavukcuoglu, and Daan Wierstra. Weight uncertainty in",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                317,
+                                477,
+                                330
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        317,
+                                        477,
+                                        330
+                                    ],
+                                    "score": 1.0,
+                                    "content": "neural network. In International Conference on Machine Learning, pp. 1613–1622, 2015.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        }
+                    ],
+                    "index": 14.5,
+                    "bbox_fs": [
+                        106,
+                        306,
+                        505,
+                        330
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        339,
+                        503,
+                        362
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                338,
+                                505,
+                                353
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        338,
+                                        505,
+                                        353
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Anastasia Borovykh. A gaussian process perspective on convolutional neural networks. Research-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                350,
+                                178,
+                                362
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        350,
+                                        178,
+                                        362
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Gate, 05 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        }
+                    ],
+                    "index": 16.5,
+                    "bbox_fs": [
+                        106,
+                        338,
+                        505,
+                        362
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        371,
+                        506,
+                        405
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                370,
+                                505,
+                                385
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        370,
+                                        505,
+                                        385
+                                    ],
+                                    "score": 1.0,
+                                    "content": "John Bradshaw, Alexander G de G Matthews, and Zoubin Ghahramani. Adversarial examples,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                383,
+                                505,
+                                396
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        383,
+                                        505,
+                                        396
+                                    ],
+                                    "score": 1.0,
+                                    "content": "uncertainty, and transfer testing robustness in gaussian process hybrid deep networks. arXiv",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                394,
+                                254,
+                                405
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        394,
+                                        254,
+                                        405
+                                    ],
+                                    "score": 1.0,
+                                    "content": "preprint arXiv:1707.02476, 2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        }
+                    ],
+                    "index": 19,
+                    "bbox_fs": [
+                        105,
+                        370,
+                        505,
+                        405
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        415,
+                        504,
+                        438
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                415,
+                                505,
+                                428
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        415,
+                                        505,
+                                        428
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Tian Qi Chen, Yulia Rubanova, Jesse Bettencourt, and David Duvenaud. Neural ordinary differential",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                427,
+                                322,
+                                439
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        427,
+                                        322,
+                                        439
+                                    ],
+                                    "score": 1.0,
+                                    "content": "equations. arXiv preprint arXiv:1806.07366, 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        }
+                    ],
+                    "index": 21.5,
+                    "bbox_fs": [
+                        106,
+                        415,
+                        505,
+                        439
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        448,
+                        504,
+                        482
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                107,
+                                447,
+                                504,
+                                460
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        447,
+                                        504,
+                                        460
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Youngmin Cho and Lawrence K Saul. Kernel methods for deep learning. In Advances in neu-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                459,
+                                504,
+                                472
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        459,
+                                        504,
+                                        472
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ral information processing systems, pp. 342–350, 2009. URL http://papers.nips.cc/",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                470,
+                                405,
+                                482
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        470,
+                                        405,
+                                        482
+                                    ],
+                                    "score": 1.0,
+                                    "content": "paper/3628-kernel-methods-for-deep-learning.pdf.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        }
+                    ],
+                    "index": 24,
+                    "bbox_fs": [
+                        107,
+                        447,
+                        504,
+                        482
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        491,
+                        505,
+                        547
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                491,
+                                505,
+                                504
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        491,
+                                        505,
+                                        504
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Andreas Damianou and Neil Lawrence. Deep gaussian processes. In Carlos M. Carvalho and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                501,
+                                505,
+                                515
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        501,
+                                        505,
+                                        515
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Pradeep Ravikumar (eds.), Proceedings of the Sixteenth International Conference on Artificial",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                513,
+                                505,
+                                526
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        513,
+                                        505,
+                                        526
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Intelligence and Statistics, volume 31 of Proceedings of Machine Learning Research, pp. 207–",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                114,
+                                523,
+                                505,
+                                539
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        114,
+                                        523,
+                                        505,
+                                        539
+                                    ],
+                                    "score": 1.0,
+                                    "content": "215, Scottsdale, Arizona, USA, 29 Apr–01 May 2013. PMLR. URL http://proceedings.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                536,
+                                300,
+                                547
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        536,
+                                        300,
+                                        547
+                                    ],
+                                    "score": 1.0,
+                                    "content": "mlr.press/v31/damianou13a.html.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        }
+                    ],
+                    "index": 28,
+                    "bbox_fs": [
+                        106,
+                        491,
+                        505,
+                        547
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        556,
+                        504,
+                        591
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                556,
+                                505,
+                                570
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        556,
+                                        505,
+                                        570
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Marc Deisenroth and Carl E Rasmussen. PILCO: A model-based and data-efficient approach to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                568,
+                                505,
+                                581
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        568,
+                                        505,
+                                        581
+                                    ],
+                                    "score": 1.0,
+                                    "content": "policy search. In Proceedings of the 28th International Conference on machine learning (ICML-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                579,
+                                215,
+                                591
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        579,
+                                        215,
+                                        591
+                                    ],
+                                    "score": 1.0,
+                                    "content": "11), pp. 465–472, 2011.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        }
+                    ],
+                    "index": 32,
+                    "bbox_fs": [
+                        105,
+                        556,
+                        505,
+                        591
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        600,
+                        504,
+                        623
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                600,
+                                504,
+                                614
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        600,
+                                        504,
+                                        614
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Yarin Gal and Zoubin Ghahramani. Dropout as a bayesian approximation: Representing model",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                612,
+                                396,
+                                623
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        612,
+                                        396,
+                                        623
+                                    ],
+                                    "score": 1.0,
+                                    "content": "uncertainty in deep learning. arXiv preprint arXiv:1506.02142, 2015.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        }
+                    ],
+                    "index": 34.5,
+                    "bbox_fs": [
+                        106,
+                        600,
+                        504,
+                        623
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        633,
+                        504,
+                        656
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                633,
+                                505,
+                                646
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        633,
+                                        505,
+                                        646
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Yarin Gal and Lewis Smith. Idealised bayesian neural networks cannot have adversarial examples:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                644,
+                                411,
+                                657
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        644,
+                                        411,
+                                        657
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Theoretical and empirical study. arXiv preprint arXiv:1806.00667, 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        }
+                    ],
+                    "index": 36.5,
+                    "bbox_fs": [
+                        106,
+                        633,
+                        505,
+                        657
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        666,
+                        504,
+                        700
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                664,
+                                505,
+                                681
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        664,
+                                        505,
+                                        681
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image recog-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                676,
+                                505,
+                                691
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        676,
+                                        505,
+                                        691
+                                    ],
+                                    "score": 1.0,
+                                    "content": "nition. In Proceedings of the IEEE conference on computer vision and pattern recognition, pp.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                687,
+                                430,
+                                701
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        687,
+                                        430,
+                                        701
+                                    ],
+                                    "score": 1.0,
+                                    "content": "770–778, 2016a. URL https://arxiv.org/pdf/1512.03385.pdf.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        }
+                    ],
+                    "index": 39,
+                    "bbox_fs": [
+                        105,
+                        664,
+                        505,
+                        701
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        709,
+                        504,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                709,
+                                505,
+                                722
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        709,
+                                        505,
+                                        722
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Identity mappings in deep residual",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                720,
+                                462,
+                                733
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        720,
+                                        462,
+                                        733
+                                    ],
+                                    "score": 1.0,
+                                    "content": "networks. In European conference on computer vision, pp. 630–645. Springer, 2016b.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        }
+                    ],
+                    "index": 41.5,
+                    "bbox_fs": [
+                        106,
+                        709,
+                        505,
+                        733
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        82,
+                        505,
+                        138
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                82,
+                                504,
+                                95
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        82,
+                                        504,
+                                        95
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Daniel Hernández-lobato, Jose M. Hernández-lobato, and Pierre Dupont. Robust multi-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                93,
+                                504,
+                                105
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        93,
+                                        504,
+                                        105
+                                    ],
+                                    "score": 1.0,
+                                    "content": "class gaussian process classification. In J. Shawe-Taylor, R. S. Zemel, P. L. Bartlett,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                103,
+                                505,
+                                117
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        103,
+                                        505,
+                                        117
+                                    ],
+                                    "score": 1.0,
+                                    "content": "F. Pereira, and K. Q. Weinberger (eds.), Advances in Neural Information Process-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                114,
+                                505,
+                                129
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        114,
+                                        298,
+                                        127
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ing Systems 24, pp. 280–288. 2011.",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        300,
+                                        114,
+                                        505,
+                                        129
+                                    ],
+                                    "score": 1.0,
+                                    "content": "URL http://papers.nips.cc/paper/",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                125,
+                                478,
+                                139
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        125,
+                                        478,
+                                        139
+                                    ],
+                                    "score": 1.0,
+                                    "content": "4241-robust-multi-class-gaussian-process-classification.pdf.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        }
+                    ],
+                    "index": 2
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        146,
+                        505,
+                        180
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                145,
+                                505,
+                                159
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        145,
+                                        505,
+                                        159
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Alex Krizhevsky, Ilya Sutskever, and Geoffrey E Hinton. Imagenet classification with deep convo-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                157,
+                                505,
+                                171
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        157,
+                                        505,
+                                        171
+                                    ],
+                                    "score": 1.0,
+                                    "content": "lutional neural networks. In Advances in neural information processing systems, pp. 1097–1105,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                114,
+                                167,
+                                143,
+                                180
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        114,
+                                        167,
+                                        143,
+                                        180
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2012.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        }
+                    ],
+                    "index": 6
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        188,
+                        505,
+                        211
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                187,
+                                505,
+                                201
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        187,
+                                        505,
+                                        201
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Vinayak Kumar, Vaibhav Singh, PK Srijith, and Andreas Damianou. Deep gaussian processes with",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                199,
+                                370,
+                                211
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        199,
+                                        370,
+                                        211
+                                    ],
+                                    "score": 1.0,
+                                    "content": "convolutional kernels. arXiv preprint arXiv:1806.01655, 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        }
+                    ],
+                    "index": 8.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        219,
+                        504,
+                        242
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                219,
+                                504,
+                                232
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        219,
+                                        504,
+                                        232
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Alexey Kurakin, Ian Goodfellow, and Samy Bengio. Adversarial examples in the physical world.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                230,
+                                279,
+                                242
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        230,
+                                        279,
+                                        242
+                                    ],
+                                    "score": 1.0,
+                                    "content": "arXiv preprint arXiv:1607.02533, 2016.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        }
+                    ],
+                    "index": 10.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        249,
+                        505,
+                        284
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                249,
+                                505,
+                                263
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        249,
+                                        505,
+                                        263
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Balaji Lakshminarayanan, Alexander Pritzel, and Charles Blundell. Simple and scalable predictive",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                259,
+                                505,
+                                275
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        259,
+                                        505,
+                                        275
+                                    ],
+                                    "score": 1.0,
+                                    "content": "uncertainty estimation using deep ensembles. In Advances in Neural Information Processing",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                272,
+                                243,
+                                284
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        272,
+                                        243,
+                                        284
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Systems, pp. 6402–6413, 2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        }
+                    ],
+                    "index": 13
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        291,
+                        503,
+                        326
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                292,
+                                505,
+                                304
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        292,
+                                        505,
+                                        304
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Yann LeCun, Bernhard E Boser, John S Denker, Donnie Henderson, Richard E Howard, Wayne E",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                302,
+                                505,
+                                316
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        302,
+                                        505,
+                                        316
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Hubbard, and Lawrence D Jackel. Handwritten digit recognition with a back-propagation net-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                313,
+                                443,
+                                327
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        313,
+                                        443,
+                                        327
+                                    ],
+                                    "score": 1.0,
+                                    "content": "work. In Advances in neural information processing systems, pp. 396–404, 1990.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        }
+                    ],
+                    "index": 16
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        333,
+                        505,
+                        367
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                334,
+                                505,
+                                346
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        334,
+                                        505,
+                                        346
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Jaehoon Lee, Yasaman Bahri, Roman Novak, Samuel S Schoenholz, Jeffrey Pennington, and Jascha",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                345,
+                                505,
+                                357
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        345,
+                                        505,
+                                        357
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Sohl-Dickstein. Deep neural networks as gaussian processes. arXiv preprint arXiv:1711.00165,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                355,
+                                143,
+                                368
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        355,
+                                        143,
+                                        368
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        }
+                    ],
+                    "index": 19
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        375,
+                        504,
+                        399
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                375,
+                                505,
+                                389
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        375,
+                                        505,
+                                        389
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Stephan Mandt, Matthew D Hoffman, and David M Blei. Stochastic gradient descent as approximate",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                387,
+                                474,
+                                399
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        387,
+                                        474,
+                                        399
+                                    ],
+                                    "score": 1.0,
+                                    "content": "bayesian inference. The Journal of Machine Learning Research, 18(1):4873–4907, 2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        }
+                    ],
+                    "index": 21.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        406,
+                        505,
+                        451
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                406,
+                                505,
+                                419
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        406,
+                                        505,
+                                        419
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Alexander G. de G. Matthews, Jiri Hron, Mark Rowland, Richard E. Turner, and Zoubin Ghahra-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                418,
+                                505,
+                                430
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        418,
+                                        505,
+                                        430
+                                    ],
+                                    "score": 1.0,
+                                    "content": "mani. Gaussian process behaviour in wide deep neural networks. In International Confer-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                114,
+                                429,
+                                505,
+                                442
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        114,
+                                        429,
+                                        497,
+                                        442
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ence on Learning Representations, 2018a. URL https://openreview.net/forum?id",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        497,
+                                        430,
+                                        505,
+                                        438
+                                    ],
+                                    "score": 0.52,
+                                    "content": "=",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                439,
+                                177,
+                                452
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        439,
+                                        177,
+                                        452
+                                    ],
+                                    "score": 1.0,
+                                    "content": "H1-nGgWC-.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        }
+                    ],
+                    "index": 24.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        459,
+                        504,
+                        493
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                459,
+                                505,
+                                471
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        459,
+                                        505,
+                                        471
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Alexander G de G Matthews, Mark Rowland, Jiri Hron, Richard E Turner, and Zoubin Ghahramani.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                470,
+                                505,
+                                483
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        470,
+                                        505,
+                                        483
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Gaussian process behaviour in wide deep neural networks. arXiv preprint arXiv:1804.11271,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                479,
+                                147,
+                                494
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        479,
+                                        147,
+                                        494
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2018b.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        }
+                    ],
+                    "index": 28
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        501,
+                        505,
+                        546
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                501,
+                                505,
+                                514
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        501,
+                                        505,
+                                        514
+                                    ],
+                                    "score": 1.0,
+                                    "content": "De G Matthews, G Alexander, Mark Van Der Wilk, Tom Nickson, Keisuke Fujii, Alexis Bouk-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                512,
+                                505,
+                                525
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        512,
+                                        505,
+                                        525
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ouvalas, Pablo León-Villagrá, Zoubin Ghahramani, and James Hensman. GPflow: A gaussian",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                114,
+                                524,
+                                506,
+                                536
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        114,
+                                        524,
+                                        506,
+                                        536
+                                    ],
+                                    "score": 1.0,
+                                    "content": "process library using TensorFlow. The Journal of Machine Learning Research, 18(1):1299–1304,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                114,
+                                533,
+                                143,
+                                546
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        114,
+                                        533,
+                                        143,
+                                        546
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        }
+                    ],
+                    "index": 31.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        554,
+                        504,
+                        577
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                551,
+                                506,
+                                570
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        551,
+                                        506,
+                                        570
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Radford M. Neal. Bayesian Learning for Neural Networks. Springer-Verlag, Berlin, Heidelberg,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                117,
+                                565,
+                                223,
+                                577
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        117,
+                                        565,
+                                        223,
+                                        577
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1996. ISBN 0387947248.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        }
+                    ],
+                    "index": 34.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        585,
+                        505,
+                        608
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                583,
+                                506,
+                                600
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        583,
+                                        506,
+                                        600
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Carl Edward Rasmussen and Christopher KI Williams. Gaussian processes for machine learning,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                597,
+                                277,
+                                609
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        597,
+                                        277,
+                                        609
+                                    ],
+                                    "score": 1.0,
+                                    "content": "volume 1. MIT press Cambridge, 2006.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        }
+                    ],
+                    "index": 36.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        616,
+                        505,
+                        639
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                617,
+                                505,
+                                630
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        617,
+                                        505,
+                                        630
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Samuel S Schoenholz, Justin Gilmer, Surya Ganguli, and Jascha Sohl-Dickstein. Deep information",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                628,
+                                331,
+                                640
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        628,
+                                        331,
+                                        640
+                                    ],
+                                    "score": 1.0,
+                                    "content": "propagation. arXiv preprint arXiv:1611.01232, 2016.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        }
+                    ],
+                    "index": 38.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        647,
+                        504,
+                        670
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                647,
+                                505,
+                                660
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        647,
+                                        505,
+                                        660
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Jasper Snoek, Hugo Larochelle, and Ryan P Adams. Practical bayesian optimization of machine",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                658,
+                                505,
+                                672
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        658,
+                                        505,
+                                        672
+                                    ],
+                                    "score": 1.0,
+                                    "content": "learning algorithms. In Advances in neural information processing systems, pp. 2951–2959, 2012.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        }
+                    ],
+                    "index": 40.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        678,
+                        504,
+                        701
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                678,
+                                505,
+                                691
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        678,
+                                        505,
+                                        691
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Christian Szegedy, Wojciech Zaremba, Ilya Sutskever, Joan Bruna, Dumitru Erhan, Ian Goodfellow,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                690,
+                                505,
+                                702
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        690,
+                                        505,
+                                        702
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and Rob Fergus. Intriguing properties of neural networks. arXiv preprint arXiv:1312.6199, 2013.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        }
+                    ],
+                    "index": 42.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        709,
+                        503,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                708,
+                                505,
+                                723
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        708,
+                                        505,
+                                        723
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Mark van der Wilk, Carl Edward Rasmussen, and James Hensman. Convolutional gaussian pro-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                720,
+                                462,
+                                732
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        720,
+                                        462,
+                                        732
+                                    ],
+                                    "score": 1.0,
+                                    "content": "cesses. In Advances in Neural Information Processing Systems, pp. 2845–2854, 2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        }
+                    ],
+                    "index": 44.5
+                }
+            ],
+            "page_idx": 10,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        107,
+                        27,
+                        293,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                26,
+                                294,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        26,
+                                        294,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Published as a conference paper at ICLR 2019",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        300,
+                        751,
+                        310,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                299,
+                                750,
+                                312,
+                                765
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        299,
+                                        750,
+                                        312,
+                                        765
+                                    ],
+                                    "score": 1.0,
+                                    "content": "",
+                                    "type": "text",
+                                    "height": 15,
+                                    "width": 13
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        82,
+                        505,
+                        138
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                82,
+                                504,
+                                95
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        82,
+                                        504,
+                                        95
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Daniel Hernández-lobato, Jose M. Hernández-lobato, and Pierre Dupont. Robust multi-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                93,
+                                504,
+                                105
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        93,
+                                        504,
+                                        105
+                                    ],
+                                    "score": 1.0,
+                                    "content": "class gaussian process classification. In J. Shawe-Taylor, R. S. Zemel, P. L. Bartlett,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                103,
+                                505,
+                                117
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        103,
+                                        505,
+                                        117
+                                    ],
+                                    "score": 1.0,
+                                    "content": "F. Pereira, and K. Q. Weinberger (eds.), Advances in Neural Information Process-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                114,
+                                505,
+                                129
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        114,
+                                        298,
+                                        127
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ing Systems 24, pp. 280–288. 2011.",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        300,
+                                        114,
+                                        505,
+                                        129
+                                    ],
+                                    "score": 1.0,
+                                    "content": "URL http://papers.nips.cc/paper/",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                125,
+                                478,
+                                139
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        125,
+                                        478,
+                                        139
+                                    ],
+                                    "score": 1.0,
+                                    "content": "4241-robust-multi-class-gaussian-process-classification.pdf.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        }
+                    ],
+                    "index": 2,
+                    "bbox_fs": [
+                        106,
+                        82,
+                        505,
+                        139
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        146,
+                        505,
+                        180
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                145,
+                                505,
+                                159
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        145,
+                                        505,
+                                        159
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Alex Krizhevsky, Ilya Sutskever, and Geoffrey E Hinton. Imagenet classification with deep convo-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                157,
+                                505,
+                                171
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        157,
+                                        505,
+                                        171
+                                    ],
+                                    "score": 1.0,
+                                    "content": "lutional neural networks. In Advances in neural information processing systems, pp. 1097–1105,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                114,
+                                167,
+                                143,
+                                180
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        114,
+                                        167,
+                                        143,
+                                        180
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2012.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        }
+                    ],
+                    "index": 6,
+                    "bbox_fs": [
+                        105,
+                        145,
+                        505,
+                        180
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        188,
+                        505,
+                        211
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                187,
+                                505,
+                                201
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        187,
+                                        505,
+                                        201
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Vinayak Kumar, Vaibhav Singh, PK Srijith, and Andreas Damianou. Deep gaussian processes with",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                199,
+                                370,
+                                211
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        199,
+                                        370,
+                                        211
+                                    ],
+                                    "score": 1.0,
+                                    "content": "convolutional kernels. arXiv preprint arXiv:1806.01655, 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        }
+                    ],
+                    "index": 8.5,
+                    "bbox_fs": [
+                        106,
+                        187,
+                        505,
+                        211
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        219,
+                        504,
+                        242
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                219,
+                                504,
+                                232
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        219,
+                                        504,
+                                        232
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Alexey Kurakin, Ian Goodfellow, and Samy Bengio. Adversarial examples in the physical world.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                230,
+                                279,
+                                242
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        230,
+                                        279,
+                                        242
+                                    ],
+                                    "score": 1.0,
+                                    "content": "arXiv preprint arXiv:1607.02533, 2016.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        }
+                    ],
+                    "index": 10.5,
+                    "bbox_fs": [
+                        106,
+                        219,
+                        504,
+                        242
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        249,
+                        505,
+                        284
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                249,
+                                505,
+                                263
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        249,
+                                        505,
+                                        263
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Balaji Lakshminarayanan, Alexander Pritzel, and Charles Blundell. Simple and scalable predictive",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                259,
+                                505,
+                                275
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        259,
+                                        505,
+                                        275
+                                    ],
+                                    "score": 1.0,
+                                    "content": "uncertainty estimation using deep ensembles. In Advances in Neural Information Processing",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                272,
+                                243,
+                                284
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        272,
+                                        243,
+                                        284
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Systems, pp. 6402–6413, 2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        }
+                    ],
+                    "index": 13,
+                    "bbox_fs": [
+                        106,
+                        249,
+                        505,
+                        284
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        291,
+                        503,
+                        326
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                292,
+                                505,
+                                304
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        292,
+                                        505,
+                                        304
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Yann LeCun, Bernhard E Boser, John S Denker, Donnie Henderson, Richard E Howard, Wayne E",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                302,
+                                505,
+                                316
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        302,
+                                        505,
+                                        316
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Hubbard, and Lawrence D Jackel. Handwritten digit recognition with a back-propagation net-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                313,
+                                443,
+                                327
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        313,
+                                        443,
+                                        327
+                                    ],
+                                    "score": 1.0,
+                                    "content": "work. In Advances in neural information processing systems, pp. 396–404, 1990.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        }
+                    ],
+                    "index": 16,
+                    "bbox_fs": [
+                        106,
+                        292,
+                        505,
+                        327
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        333,
+                        505,
+                        367
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                334,
+                                505,
+                                346
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        334,
+                                        505,
+                                        346
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Jaehoon Lee, Yasaman Bahri, Roman Novak, Samuel S Schoenholz, Jeffrey Pennington, and Jascha",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                345,
+                                505,
+                                357
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        345,
+                                        505,
+                                        357
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Sohl-Dickstein. Deep neural networks as gaussian processes. arXiv preprint arXiv:1711.00165,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                355,
+                                143,
+                                368
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        355,
+                                        143,
+                                        368
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        }
+                    ],
+                    "index": 19,
+                    "bbox_fs": [
+                        106,
+                        334,
+                        505,
+                        368
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        375,
+                        504,
+                        399
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                375,
+                                505,
+                                389
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        375,
+                                        505,
+                                        389
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Stephan Mandt, Matthew D Hoffman, and David M Blei. Stochastic gradient descent as approximate",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                387,
+                                474,
+                                399
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        387,
+                                        474,
+                                        399
+                                    ],
+                                    "score": 1.0,
+                                    "content": "bayesian inference. The Journal of Machine Learning Research, 18(1):4873–4907, 2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        }
+                    ],
+                    "index": 21.5,
+                    "bbox_fs": [
+                        105,
+                        375,
+                        505,
+                        399
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        406,
+                        505,
+                        451
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                406,
+                                505,
+                                419
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        406,
+                                        505,
+                                        419
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Alexander G. de G. Matthews, Jiri Hron, Mark Rowland, Richard E. Turner, and Zoubin Ghahra-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                418,
+                                505,
+                                430
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        418,
+                                        505,
+                                        430
+                                    ],
+                                    "score": 1.0,
+                                    "content": "mani. Gaussian process behaviour in wide deep neural networks. In International Confer-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                114,
+                                429,
+                                505,
+                                442
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        114,
+                                        429,
+                                        497,
+                                        442
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ence on Learning Representations, 2018a. URL https://openreview.net/forum?id",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        497,
+                                        430,
+                                        505,
+                                        438
+                                    ],
+                                    "score": 0.52,
+                                    "content": "=",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                439,
+                                177,
+                                452
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        439,
+                                        177,
+                                        452
+                                    ],
+                                    "score": 1.0,
+                                    "content": "H1-nGgWC-.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        }
+                    ],
+                    "index": 24.5,
+                    "bbox_fs": [
+                        105,
+                        406,
+                        505,
+                        452
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        459,
+                        504,
+                        493
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                459,
+                                505,
+                                471
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        459,
+                                        505,
+                                        471
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Alexander G de G Matthews, Mark Rowland, Jiri Hron, Richard E Turner, and Zoubin Ghahramani.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                470,
+                                505,
+                                483
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        470,
+                                        505,
+                                        483
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Gaussian process behaviour in wide deep neural networks. arXiv preprint arXiv:1804.11271,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                479,
+                                147,
+                                494
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        479,
+                                        147,
+                                        494
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2018b.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        }
+                    ],
+                    "index": 28,
+                    "bbox_fs": [
+                        106,
+                        459,
+                        505,
+                        494
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        501,
+                        505,
+                        546
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                501,
+                                505,
+                                514
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        501,
+                                        505,
+                                        514
+                                    ],
+                                    "score": 1.0,
+                                    "content": "De G Matthews, G Alexander, Mark Van Der Wilk, Tom Nickson, Keisuke Fujii, Alexis Bouk-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                512,
+                                505,
+                                525
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        512,
+                                        505,
+                                        525
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ouvalas, Pablo León-Villagrá, Zoubin Ghahramani, and James Hensman. GPflow: A gaussian",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                114,
+                                524,
+                                506,
+                                536
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        114,
+                                        524,
+                                        506,
+                                        536
+                                    ],
+                                    "score": 1.0,
+                                    "content": "process library using TensorFlow. The Journal of Machine Learning Research, 18(1):1299–1304,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                114,
+                                533,
+                                143,
+                                546
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        114,
+                                        533,
+                                        143,
+                                        546
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        }
+                    ],
+                    "index": 31.5,
+                    "bbox_fs": [
+                        105,
+                        501,
+                        506,
+                        546
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        554,
+                        504,
+                        577
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                551,
+                                506,
+                                570
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        551,
+                                        506,
+                                        570
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Radford M. Neal. Bayesian Learning for Neural Networks. Springer-Verlag, Berlin, Heidelberg,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                117,
+                                565,
+                                223,
+                                577
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        117,
+                                        565,
+                                        223,
+                                        577
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1996. ISBN 0387947248.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        }
+                    ],
+                    "index": 34.5,
+                    "bbox_fs": [
+                        104,
+                        551,
+                        506,
+                        577
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        585,
+                        505,
+                        608
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                583,
+                                506,
+                                600
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        583,
+                                        506,
+                                        600
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Carl Edward Rasmussen and Christopher KI Williams. Gaussian processes for machine learning,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                597,
+                                277,
+                                609
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        597,
+                                        277,
+                                        609
+                                    ],
+                                    "score": 1.0,
+                                    "content": "volume 1. MIT press Cambridge, 2006.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        }
+                    ],
+                    "index": 36.5,
+                    "bbox_fs": [
+                        105,
+                        583,
+                        506,
+                        609
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        616,
+                        505,
+                        639
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                617,
+                                505,
+                                630
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        617,
+                                        505,
+                                        630
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Samuel S Schoenholz, Justin Gilmer, Surya Ganguli, and Jascha Sohl-Dickstein. Deep information",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                628,
+                                331,
+                                640
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        628,
+                                        331,
+                                        640
+                                    ],
+                                    "score": 1.0,
+                                    "content": "propagation. arXiv preprint arXiv:1611.01232, 2016.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        }
+                    ],
+                    "index": 38.5,
+                    "bbox_fs": [
+                        106,
+                        617,
+                        505,
+                        640
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        647,
+                        504,
+                        670
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                647,
+                                505,
+                                660
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        647,
+                                        505,
+                                        660
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Jasper Snoek, Hugo Larochelle, and Ryan P Adams. Practical bayesian optimization of machine",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                658,
+                                505,
+                                672
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        658,
+                                        505,
+                                        672
+                                    ],
+                                    "score": 1.0,
+                                    "content": "learning algorithms. In Advances in neural information processing systems, pp. 2951–2959, 2012.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        }
+                    ],
+                    "index": 40.5,
+                    "bbox_fs": [
+                        105,
+                        647,
+                        505,
+                        672
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        678,
+                        504,
+                        701
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                678,
+                                505,
+                                691
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        678,
+                                        505,
+                                        691
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Christian Szegedy, Wojciech Zaremba, Ilya Sutskever, Joan Bruna, Dumitru Erhan, Ian Goodfellow,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                690,
+                                505,
+                                702
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        690,
+                                        505,
+                                        702
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and Rob Fergus. Intriguing properties of neural networks. arXiv preprint arXiv:1312.6199, 2013.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        }
+                    ],
+                    "index": 42.5,
+                    "bbox_fs": [
+                        106,
+                        678,
+                        505,
+                        702
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        709,
+                        503,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                708,
+                                505,
+                                723
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        708,
+                                        505,
+                                        723
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Mark van der Wilk, Carl Edward Rasmussen, and James Hensman. Convolutional gaussian pro-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                720,
+                                462,
+                                732
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        720,
+                                        462,
+                                        732
+                                    ],
+                                    "score": 1.0,
+                                    "content": "cesses. In Advances in Neural Information Processing Systems, pp. 2845–2854, 2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        }
+                    ],
+                    "index": 44.5,
+                    "bbox_fs": [
+                        105,
+                        708,
+                        505,
+                        732
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        82,
+                        504,
+                        116
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                81,
+                                506,
+                                95
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        81,
+                                        506,
+                                        95
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Max Welling and Yee W Teh. Bayesian learning via stochastic gradient langevin dynamics. In",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                93,
+                                505,
+                                106
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        93,
+                                        505,
+                                        106
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Proceedings of the 28th International Conference on Machine Learning (ICML-11), pp. 681–688,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                114,
+                                102,
+                                143,
+                                117
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        114,
+                                        102,
+                                        143,
+                                        117
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2011.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        }
+                    ],
+                    "index": 1
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        123,
+                        505,
+                        179
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                124,
+                                505,
+                                136
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        124,
+                                        505,
+                                        136
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Andrew Gordon Wilson, Zhiting Hu, Ruslan Salakhutdinov, and Eric P. Xing. Deep kernel",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                135,
+                                504,
+                                147
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        135,
+                                        504,
+                                        147
+                                    ],
+                                    "score": 1.0,
+                                    "content": "learning. In Arthur Gretton and Christian C. Robert (eds.), Proceedings of the 19th Interna-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                145,
+                                505,
+                                158
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        145,
+                                        505,
+                                        158
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tional Conference on Artificial Intelligence and Statistics, volume 51 of Proceedings of Ma-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                154,
+                                505,
+                                171
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        154,
+                                        505,
+                                        171
+                                    ],
+                                    "score": 1.0,
+                                    "content": "chine Learning Research, pp. 370–378, Cadiz, Spain, 09–11 May 2016. PMLR. URL http:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                167,
+                                367,
+                                180
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        167,
+                                        367,
+                                        180
+                                    ],
+                                    "score": 1.0,
+                                    "content": "//proceedings.mlr.press/v51/wilson16.html.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        }
+                    ],
+                    "index": 5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        186,
+                        505,
+                        220
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                185,
+                                505,
+                                201
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        185,
+                                        505,
+                                        201
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Lechao Xiao, Yasaman Bahri, Jascha Sohl-Dickstein, Samuel S Schoenholz, and Jeffrey Penning-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                198,
+                                505,
+                                210
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        198,
+                                        505,
+                                        210
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ton. Dynamical isometry and a mean field theory of cnns: How to train 10,000-layer vanilla",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                209,
+                                405,
+                                221
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        209,
+                                        405,
+                                        221
+                                    ],
+                                    "score": 1.0,
+                                    "content": "convolutional neural networks. arXiv preprint arXiv:1806.05393, 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        }
+                    ],
+                    "index": 9
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        227,
+                        503,
+                        251
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                226,
+                                505,
+                                241
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        226,
+                                        505,
+                                        241
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Ge Yang and Samuel Schoenholz. Mean field residual networks: On the edge of chaos. In Advances",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                239,
+                                375,
+                                251
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        239,
+                                        375,
+                                        251
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in neural information processing systems, pp. 7103–7114, 2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        }
+                    ],
+                    "index": 11.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        258,
+                        505,
+                        280
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                256,
+                                505,
+                                271
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        257,
+                                        297,
+                                        270
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Sergey Zagoruyko and Nikos Komodakis.",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        313,
+                                        256,
+                                        425,
+                                        270
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Wide residual networks.",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        438,
+                                        258,
+                                        505,
+                                        271
+                                    ],
+                                    "score": 1.0,
+                                    "content": "arXiv preprint",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                269,
+                                219,
+                                280
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        269,
+                                        219,
+                                        280
+                                    ],
+                                    "score": 1.0,
+                                    "content": "arXiv:1605.07146, 2016.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        }
+                    ],
+                    "index": 13.5
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        302,
+                        180,
+                        315
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                300,
+                                182,
+                                317
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        300,
+                                        182,
+                                        317
+                                    ],
+                                    "score": 1.0,
+                                    "content": "7 APPENDIX",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        }
+                    ],
+                    "index": 15
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        327,
+                        259,
+                        339
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                326,
+                                261,
+                                340
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        326,
+                                        261,
+                                        340
+                                    ],
+                                    "score": 1.0,
+                                    "content": "7.1 TECHNICAL NOTES ON LIMITS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        }
+                    ],
+                    "index": 16
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        347,
+                        505,
+                        381
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                348,
+                                505,
+                                360
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        348,
+                                        505,
+                                        360
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The key technical issues in the proof (and the key differences between Lee et al. 2017 and Matthews",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                358,
+                                505,
+                                371
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        358,
+                                        505,
+                                        371
+                                    ],
+                                    "score": 1.0,
+                                    "content": "et al. 2018b) arise from exactly how and where we take limits. In particular, consider the activations",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                369,
+                                331,
+                                383
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        369,
+                                        331,
+                                        383
+                                    ],
+                                    "score": 1.0,
+                                    "content": "as being functions of the activities at the previous layer,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        }
+                    ],
+                    "index": 18
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        229,
+                        386,
+                        381,
+                        402
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                229,
+                                386,
+                                381,
+                                402
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        229,
+                                        386,
+                                        381,
+                                        402
+                                    ],
+                                    "score": 0.93,
+                                    "content": "{ \\bf A } ^ { ( 4 ) } = { \\bf A } ^ { ( 4 ) } ( { \\bf A } ^ { ( 3 ) } ( { \\bf A } ^ { ( 2 ) } ( { \\bf A } ^ { ( 1 ) } ( { \\bf X } ) ) ) )",
+                                    "type": "interline_equation",
+                                    "image_path": "b3df92ef3a1e1a9f6a55bfd23b526811785c6d6ebffe0128f92602b89618fcb5.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 20,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                229,
+                                386,
+                                381,
+                                402
+                            ],
+                            "spans": [],
+                            "index": 20
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        407,
+                        504,
+                        430
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                408,
+                                505,
+                                419
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        408,
+                                        505,
+                                        419
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Now, there are two approaches to taking limits. First, both our argument in the main text, and the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                418,
+                                455,
+                                430
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        418,
+                                        455,
+                                        430
+                                    ],
+                                    "score": 1.0,
+                                    "content": "argument in Lee et al. (2017) is valid if we are able to take limits “inside” the network,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        }
+                    ],
+                    "index": 21.5
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        165,
+                        434,
+                        444,
+                        463
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                165,
+                                434,
+                                444,
+                                463
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        165,
+                                        434,
+                                        444,
+                                        463
+                                    ],
+                                    "score": 0.94,
+                                    "content": "\\mathbf { A } _ { \\mathrm { L } } ^ { ( 4 ) } = \\operatorname* { l i m } _ { C ^ { ( 3 ) } \\to \\infty } \\mathbf { A } ^ { ( 4 ) } \\left( \\operatorname* { l i m } _ { C ^ { ( 2 ) } \\to \\infty } \\mathbf { A } ^ { ( 3 ) } \\left( \\operatorname* { l i m } _ { C ^ { ( 1 ) } \\to \\infty } \\mathbf { A } ^ { ( 2 ) } \\left( \\mathbf { A } ^ { ( 1 ) } ( \\mathbf { X } ) \\right) \\right) \\right) .",
+                                    "type": "interline_equation",
+                                    "image_path": "ecda43d5472ce6796fa65efa1ae828841615270ab00a72e6b4c5b375739ab9de.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 24,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                165,
+                                434,
+                                444,
+                                443.6666666666667
+                            ],
+                            "spans": [],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                165,
+                                443.6666666666667,
+                                444,
+                                453.33333333333337
+                            ],
+                            "spans": [],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                165,
+                                453.33333333333337,
+                                444,
+                                463.00000000000006
+                            ],
+                            "spans": [],
+                            "index": 25
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        468,
+                        503,
+                        490
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                468,
+                                505,
+                                480
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        468,
+                                        505,
+                                        480
+                                    ],
+                                    "score": 1.0,
+                                    "content": "However, Matthews et al. (2018a;b) argue that is preferable to take limits “outside” the network. In",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                478,
+                                425,
+                                491
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        478,
+                                        425,
+                                        491
+                                    ],
+                                    "score": 1.0,
+                                    "content": "particular, Matthews et al. (2018b) take the limit with all layers simultaneously,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        }
+                    ],
+                    "index": 26.5
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        206,
+                        495,
+                        403,
+                        517
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                206,
+                                495,
+                                403,
+                                517
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        206,
+                                        495,
+                                        403,
+                                        517
+                                    ],
+                                    "score": 0.93,
+                                    "content": "{ \\bf A } _ { \\mathrm { M } } ^ { ( 4 ) } = \\operatorname* { l i m } _ { n  \\infty } { \\bf A } ^ { ( 4 ) } ( { \\bf A } ^ { ( 3 ) } ( { \\bf A } ^ { ( 2 ) } ( { \\bf A } ^ { ( 1 ) } ( { \\bf X } ) ) ) ) ,",
+                                    "type": "interline_equation",
+                                    "image_path": "60aa0f685aef6c4f814cd4e1691b1779363228de038f37492e5c1d5ec39199bd.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 28,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                206,
+                                495,
+                                403,
+                                517
+                            ],
+                            "spans": [],
+                            "index": 28
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        523,
+                        505,
+                        547
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                522,
+                                505,
+                                537
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        522,
+                                        133,
+                                        537
+                                    ],
+                                    "score": 1.0,
+                                    "content": "where",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        133,
+                                        523,
+                                        197,
+                                        536
+                                    ],
+                                    "score": 0.93,
+                                    "content": "C ^ { ( \\ell ) } = C ^ { ( \\ell ) } ( n )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        198,
+                                        522,
+                                        270,
+                                        537
+                                    ],
+                                    "score": 1.0,
+                                    "content": "goes to infinity as",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        270,
+                                        526,
+                                        303,
+                                        534
+                                    ],
+                                    "score": 0.87,
+                                    "content": "n \\to \\infty",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        303,
+                                        522,
+                                        505,
+                                        537
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". That said, similar technical issues arise if we take",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                534,
+                                281,
+                                546
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        534,
+                                        281,
+                                        546
+                                    ],
+                                    "score": 1.0,
+                                    "content": "limits in sequence, but outside the network.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        }
+                    ],
+                    "index": 29.5
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        560,
+                        431,
+                        583
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                558,
+                                432,
+                                573
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        558,
+                                        432,
+                                        573
+                                    ],
+                                    "score": 1.0,
+                                    "content": "7.2 EXTENDING THE DERIVATIONS OF MATTHEWS ET AL. (2018B) TO THE",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                130,
+                                572,
+                                232,
+                                583
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        130,
+                                        572,
+                                        232,
+                                        583
+                                    ],
+                                    "score": 1.0,
+                                    "content": "CONVOLUTIONAL CASE",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        }
+                    ],
+                    "index": 31.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        591,
+                        505,
+                        681
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                591,
+                                505,
+                                605
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        591,
+                                        505,
+                                        605
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In the main text, we follow Lee et al. (2017) in sequentially taking the limit of each layer to infinity",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                601,
+                                506,
+                                618
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        601,
+                                        126,
+                                        618
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(i.e.",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        126,
+                                        603,
+                                        173,
+                                        614
+                                    ],
+                                    "score": 0.91,
+                                    "content": "C ^ { ( 1 ) } \\to \\infty",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        173,
+                                        601,
+                                        197,
+                                        618
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", then",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        197,
+                                        603,
+                                        244,
+                                        614
+                                    ],
+                                    "score": 0.91,
+                                    "content": "C ^ { ( 2 ) }  \\infty",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        244,
+                                        601,
+                                        506,
+                                        618
+                                    ],
+                                    "score": 1.0,
+                                    "content": "etc.). This dramatically simplified the argument, because taking",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                614,
+                                505,
+                                628
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        614,
+                                        505,
+                                        628
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the number of units in the previous layer to infinity means that the inputs from that layer are exactly",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                626,
+                                505,
+                                638
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        626,
+                                        505,
+                                        638
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Gaussian distributed. However, Matthews et al. (2018b) argue that the more practically relevant limit",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                637,
+                                505,
+                                649
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        637,
+                                        505,
+                                        649
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is where we take all layers to infinity simultaneously. This raises considerable additional difficulties,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                648,
+                                505,
+                                660
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        648,
+                                        505,
+                                        660
+                                    ],
+                                    "score": 1.0,
+                                    "content": "because we must reason about convergence in the case where the previous layer is finite. Note that",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                659,
+                                505,
+                                671
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        659,
+                                        505,
+                                        671
+                                    ],
+                                    "score": 1.0,
+                                    "content": "this section is not intended to stand independently: it is intended to be read alongside Matthews et al.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                670,
+                                339,
+                                681
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        670,
+                                        339,
+                                        681
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(2018b), and we use several of their results without proof.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        }
+                    ],
+                    "index": 36.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        686,
+                        505,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                686,
+                                506,
+                                699
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        686,
+                                        506,
+                                        699
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Mirroring Definition 3 in Matthews et al. (2018b), we begin by choosing a set of “width” functions,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                696,
+                                507,
+                                714
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        697,
+                                        139,
+                                        711
+                                    ],
+                                    "score": 0.92,
+                                    "content": "C ^ { ( \\ell ) } ( n )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        140,
+                                        696,
+                                        159,
+                                        714
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", for",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        159,
+                                        698,
+                                        223,
+                                        711
+                                    ],
+                                    "score": 0.94,
+                                    "content": "\\ell \\in \\{ 1 , \\ldots , L \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        223,
+                                        696,
+                                        349,
+                                        714
+                                    ],
+                                    "score": 1.0,
+                                    "content": "which all approach infinity as",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        349,
+                                        700,
+                                        385,
+                                        709
+                                    ],
+                                    "score": 0.86,
+                                    "content": "n  \\infty",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        385,
+                                        696,
+                                        507,
+                                        714
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". In Matthews et al. (2018b),",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                710,
+                                505,
+                                722
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        710,
+                                        505,
+                                        722
+                                    ],
+                                    "score": 1.0,
+                                    "content": "these functions described the number of hidden units in each layer, whereas here they describe the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                720,
+                                505,
+                                734
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        720,
+                                        505,
+                                        734
+                                    ],
+                                    "score": 1.0,
+                                    "content": "number of channels. Our goal is then to extend the proofs in Matthews et al. (2018b) (in particular,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        }
+                    ],
+                    "index": 42.5
+                }
+            ],
+            "page_idx": 11,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        108,
+                        27,
+                        293,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                26,
+                                294,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        26,
+                                        294,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Published as a conference paper at ICLR 2019",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        300,
+                        751,
+                        311,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                299,
+                                750,
+                                313,
+                                764
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        299,
+                                        750,
+                                        313,
+                                        764
+                                    ],
+                                    "score": 1.0,
+                                    "content": "12",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        82,
+                        504,
+                        116
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                81,
+                                506,
+                                95
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        81,
+                                        506,
+                                        95
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Max Welling and Yee W Teh. Bayesian learning via stochastic gradient langevin dynamics. In",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                93,
+                                505,
+                                106
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        93,
+                                        505,
+                                        106
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Proceedings of the 28th International Conference on Machine Learning (ICML-11), pp. 681–688,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                114,
+                                102,
+                                143,
+                                117
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        114,
+                                        102,
+                                        143,
+                                        117
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2011.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        }
+                    ],
+                    "index": 1,
+                    "bbox_fs": [
+                        105,
+                        81,
+                        506,
+                        117
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        123,
+                        505,
+                        179
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                124,
+                                505,
+                                136
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        124,
+                                        505,
+                                        136
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Andrew Gordon Wilson, Zhiting Hu, Ruslan Salakhutdinov, and Eric P. Xing. Deep kernel",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                135,
+                                504,
+                                147
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        135,
+                                        504,
+                                        147
+                                    ],
+                                    "score": 1.0,
+                                    "content": "learning. In Arthur Gretton and Christian C. Robert (eds.), Proceedings of the 19th Interna-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                145,
+                                505,
+                                158
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        145,
+                                        505,
+                                        158
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tional Conference on Artificial Intelligence and Statistics, volume 51 of Proceedings of Ma-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                154,
+                                505,
+                                171
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        154,
+                                        505,
+                                        171
+                                    ],
+                                    "score": 1.0,
+                                    "content": "chine Learning Research, pp. 370–378, Cadiz, Spain, 09–11 May 2016. PMLR. URL http:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                167,
+                                367,
+                                180
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        167,
+                                        367,
+                                        180
+                                    ],
+                                    "score": 1.0,
+                                    "content": "//proceedings.mlr.press/v51/wilson16.html.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        }
+                    ],
+                    "index": 5,
+                    "bbox_fs": [
+                        106,
+                        124,
+                        505,
+                        180
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        186,
+                        505,
+                        220
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                185,
+                                505,
+                                201
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        185,
+                                        505,
+                                        201
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Lechao Xiao, Yasaman Bahri, Jascha Sohl-Dickstein, Samuel S Schoenholz, and Jeffrey Penning-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                198,
+                                505,
+                                210
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        198,
+                                        505,
+                                        210
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ton. Dynamical isometry and a mean field theory of cnns: How to train 10,000-layer vanilla",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                209,
+                                405,
+                                221
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        209,
+                                        405,
+                                        221
+                                    ],
+                                    "score": 1.0,
+                                    "content": "convolutional neural networks. arXiv preprint arXiv:1806.05393, 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        }
+                    ],
+                    "index": 9,
+                    "bbox_fs": [
+                        104,
+                        185,
+                        505,
+                        221
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        227,
+                        503,
+                        251
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                226,
+                                505,
+                                241
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        226,
+                                        505,
+                                        241
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Ge Yang and Samuel Schoenholz. Mean field residual networks: On the edge of chaos. In Advances",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                239,
+                                375,
+                                251
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        239,
+                                        375,
+                                        251
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in neural information processing systems, pp. 7103–7114, 2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        }
+                    ],
+                    "index": 11.5,
+                    "bbox_fs": [
+                        105,
+                        226,
+                        505,
+                        251
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        258,
+                        505,
+                        280
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                256,
+                                505,
+                                271
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        257,
+                                        297,
+                                        270
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Sergey Zagoruyko and Nikos Komodakis.",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        313,
+                                        256,
+                                        425,
+                                        270
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Wide residual networks.",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        438,
+                                        258,
+                                        505,
+                                        271
+                                    ],
+                                    "score": 1.0,
+                                    "content": "arXiv preprint",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                269,
+                                219,
+                                280
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        269,
+                                        219,
+                                        280
+                                    ],
+                                    "score": 1.0,
+                                    "content": "arXiv:1605.07146, 2016.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        }
+                    ],
+                    "index": 13.5,
+                    "bbox_fs": [
+                        105,
+                        256,
+                        505,
+                        280
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        302,
+                        180,
+                        315
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                300,
+                                182,
+                                317
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        300,
+                                        182,
+                                        317
+                                    ],
+                                    "score": 1.0,
+                                    "content": "7 APPENDIX",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        }
+                    ],
+                    "index": 15
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        327,
+                        259,
+                        339
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                326,
+                                261,
+                                340
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        326,
+                                        261,
+                                        340
+                                    ],
+                                    "score": 1.0,
+                                    "content": "7.1 TECHNICAL NOTES ON LIMITS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        }
+                    ],
+                    "index": 16
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        347,
+                        505,
+                        381
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                348,
+                                505,
+                                360
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        348,
+                                        505,
+                                        360
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The key technical issues in the proof (and the key differences between Lee et al. 2017 and Matthews",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                358,
+                                505,
+                                371
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        358,
+                                        505,
+                                        371
+                                    ],
+                                    "score": 1.0,
+                                    "content": "et al. 2018b) arise from exactly how and where we take limits. In particular, consider the activations",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                369,
+                                331,
+                                383
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        369,
+                                        331,
+                                        383
+                                    ],
+                                    "score": 1.0,
+                                    "content": "as being functions of the activities at the previous layer,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        }
+                    ],
+                    "index": 18,
+                    "bbox_fs": [
+                        105,
+                        348,
+                        505,
+                        383
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        229,
+                        386,
+                        381,
+                        402
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                229,
+                                386,
+                                381,
+                                402
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        229,
+                                        386,
+                                        381,
+                                        402
+                                    ],
+                                    "score": 0.93,
+                                    "content": "{ \\bf A } ^ { ( 4 ) } = { \\bf A } ^ { ( 4 ) } ( { \\bf A } ^ { ( 3 ) } ( { \\bf A } ^ { ( 2 ) } ( { \\bf A } ^ { ( 1 ) } ( { \\bf X } ) ) ) )",
+                                    "type": "interline_equation",
+                                    "image_path": "b3df92ef3a1e1a9f6a55bfd23b526811785c6d6ebffe0128f92602b89618fcb5.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 20,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                229,
+                                386,
+                                381,
+                                402
+                            ],
+                            "spans": [],
+                            "index": 20
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        407,
+                        504,
+                        430
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                408,
+                                505,
+                                419
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        408,
+                                        505,
+                                        419
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Now, there are two approaches to taking limits. First, both our argument in the main text, and the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                418,
+                                455,
+                                430
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        418,
+                                        455,
+                                        430
+                                    ],
+                                    "score": 1.0,
+                                    "content": "argument in Lee et al. (2017) is valid if we are able to take limits “inside” the network,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        }
+                    ],
+                    "index": 21.5,
+                    "bbox_fs": [
+                        105,
+                        408,
+                        505,
+                        430
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        165,
+                        434,
+                        444,
+                        463
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                165,
+                                434,
+                                444,
+                                463
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        165,
+                                        434,
+                                        444,
+                                        463
+                                    ],
+                                    "score": 0.94,
+                                    "content": "\\mathbf { A } _ { \\mathrm { L } } ^ { ( 4 ) } = \\operatorname* { l i m } _ { C ^ { ( 3 ) } \\to \\infty } \\mathbf { A } ^ { ( 4 ) } \\left( \\operatorname* { l i m } _ { C ^ { ( 2 ) } \\to \\infty } \\mathbf { A } ^ { ( 3 ) } \\left( \\operatorname* { l i m } _ { C ^ { ( 1 ) } \\to \\infty } \\mathbf { A } ^ { ( 2 ) } \\left( \\mathbf { A } ^ { ( 1 ) } ( \\mathbf { X } ) \\right) \\right) \\right) .",
+                                    "type": "interline_equation",
+                                    "image_path": "ecda43d5472ce6796fa65efa1ae828841615270ab00a72e6b4c5b375739ab9de.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 24,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                165,
+                                434,
+                                444,
+                                443.6666666666667
+                            ],
+                            "spans": [],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                165,
+                                443.6666666666667,
+                                444,
+                                453.33333333333337
+                            ],
+                            "spans": [],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                165,
+                                453.33333333333337,
+                                444,
+                                463.00000000000006
+                            ],
+                            "spans": [],
+                            "index": 25
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        468,
+                        503,
+                        490
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                468,
+                                505,
+                                480
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        468,
+                                        505,
+                                        480
+                                    ],
+                                    "score": 1.0,
+                                    "content": "However, Matthews et al. (2018a;b) argue that is preferable to take limits “outside” the network. In",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                478,
+                                425,
+                                491
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        478,
+                                        425,
+                                        491
+                                    ],
+                                    "score": 1.0,
+                                    "content": "particular, Matthews et al. (2018b) take the limit with all layers simultaneously,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        }
+                    ],
+                    "index": 26.5,
+                    "bbox_fs": [
+                        106,
+                        468,
+                        505,
+                        491
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        206,
+                        495,
+                        403,
+                        517
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                206,
+                                495,
+                                403,
+                                517
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        206,
+                                        495,
+                                        403,
+                                        517
+                                    ],
+                                    "score": 0.93,
+                                    "content": "{ \\bf A } _ { \\mathrm { M } } ^ { ( 4 ) } = \\operatorname* { l i m } _ { n  \\infty } { \\bf A } ^ { ( 4 ) } ( { \\bf A } ^ { ( 3 ) } ( { \\bf A } ^ { ( 2 ) } ( { \\bf A } ^ { ( 1 ) } ( { \\bf X } ) ) ) ) ,",
+                                    "type": "interline_equation",
+                                    "image_path": "60aa0f685aef6c4f814cd4e1691b1779363228de038f37492e5c1d5ec39199bd.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 28,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                206,
+                                495,
+                                403,
+                                517
+                            ],
+                            "spans": [],
+                            "index": 28
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        523,
+                        505,
+                        547
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                522,
+                                505,
+                                537
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        522,
+                                        133,
+                                        537
+                                    ],
+                                    "score": 1.0,
+                                    "content": "where",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        133,
+                                        523,
+                                        197,
+                                        536
+                                    ],
+                                    "score": 0.93,
+                                    "content": "C ^ { ( \\ell ) } = C ^ { ( \\ell ) } ( n )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        198,
+                                        522,
+                                        270,
+                                        537
+                                    ],
+                                    "score": 1.0,
+                                    "content": "goes to infinity as",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        270,
+                                        526,
+                                        303,
+                                        534
+                                    ],
+                                    "score": 0.87,
+                                    "content": "n \\to \\infty",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        303,
+                                        522,
+                                        505,
+                                        537
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". That said, similar technical issues arise if we take",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                534,
+                                281,
+                                546
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        534,
+                                        281,
+                                        546
+                                    ],
+                                    "score": 1.0,
+                                    "content": "limits in sequence, but outside the network.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        }
+                    ],
+                    "index": 29.5,
+                    "bbox_fs": [
+                        105,
+                        522,
+                        505,
+                        546
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        560,
+                        431,
+                        583
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                558,
+                                432,
+                                573
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        558,
+                                        432,
+                                        573
+                                    ],
+                                    "score": 1.0,
+                                    "content": "7.2 EXTENDING THE DERIVATIONS OF MATTHEWS ET AL. (2018B) TO THE",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                130,
+                                572,
+                                232,
+                                583
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        130,
+                                        572,
+                                        232,
+                                        583
+                                    ],
+                                    "score": 1.0,
+                                    "content": "CONVOLUTIONAL CASE",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        }
+                    ],
+                    "index": 31.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        591,
+                        505,
+                        681
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                591,
+                                505,
+                                605
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        591,
+                                        505,
+                                        605
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In the main text, we follow Lee et al. (2017) in sequentially taking the limit of each layer to infinity",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                601,
+                                506,
+                                618
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        601,
+                                        126,
+                                        618
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(i.e.",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        126,
+                                        603,
+                                        173,
+                                        614
+                                    ],
+                                    "score": 0.91,
+                                    "content": "C ^ { ( 1 ) } \\to \\infty",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        173,
+                                        601,
+                                        197,
+                                        618
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", then",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        197,
+                                        603,
+                                        244,
+                                        614
+                                    ],
+                                    "score": 0.91,
+                                    "content": "C ^ { ( 2 ) }  \\infty",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        244,
+                                        601,
+                                        506,
+                                        618
+                                    ],
+                                    "score": 1.0,
+                                    "content": "etc.). This dramatically simplified the argument, because taking",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                614,
+                                505,
+                                628
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        614,
+                                        505,
+                                        628
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the number of units in the previous layer to infinity means that the inputs from that layer are exactly",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                626,
+                                505,
+                                638
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        626,
+                                        505,
+                                        638
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Gaussian distributed. However, Matthews et al. (2018b) argue that the more practically relevant limit",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                637,
+                                505,
+                                649
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        637,
+                                        505,
+                                        649
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is where we take all layers to infinity simultaneously. This raises considerable additional difficulties,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                648,
+                                505,
+                                660
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        648,
+                                        505,
+                                        660
+                                    ],
+                                    "score": 1.0,
+                                    "content": "because we must reason about convergence in the case where the previous layer is finite. Note that",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                659,
+                                505,
+                                671
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        659,
+                                        505,
+                                        671
+                                    ],
+                                    "score": 1.0,
+                                    "content": "this section is not intended to stand independently: it is intended to be read alongside Matthews et al.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                670,
+                                339,
+                                681
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        670,
+                                        339,
+                                        681
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(2018b), and we use several of their results without proof.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        }
+                    ],
+                    "index": 36.5,
+                    "bbox_fs": [
+                        104,
+                        591,
+                        506,
+                        681
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        686,
+                        505,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                686,
+                                506,
+                                699
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        686,
+                                        506,
+                                        699
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Mirroring Definition 3 in Matthews et al. (2018b), we begin by choosing a set of “width” functions,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                696,
+                                507,
+                                714
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        697,
+                                        139,
+                                        711
+                                    ],
+                                    "score": 0.92,
+                                    "content": "C ^ { ( \\ell ) } ( n )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        140,
+                                        696,
+                                        159,
+                                        714
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", for",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        159,
+                                        698,
+                                        223,
+                                        711
+                                    ],
+                                    "score": 0.94,
+                                    "content": "\\ell \\in \\{ 1 , \\ldots , L \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        223,
+                                        696,
+                                        349,
+                                        714
+                                    ],
+                                    "score": 1.0,
+                                    "content": "which all approach infinity as",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        349,
+                                        700,
+                                        385,
+                                        709
+                                    ],
+                                    "score": 0.86,
+                                    "content": "n  \\infty",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        385,
+                                        696,
+                                        507,
+                                        714
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". In Matthews et al. (2018b),",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                710,
+                                505,
+                                722
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        710,
+                                        505,
+                                        722
+                                    ],
+                                    "score": 1.0,
+                                    "content": "these functions described the number of hidden units in each layer, whereas here they describe the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                720,
+                                505,
+                                734
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        720,
+                                        505,
+                                        734
+                                    ],
+                                    "score": 1.0,
+                                    "content": "number of channels. Our goal is then to extend the proofs in Matthews et al. (2018b) (in particular,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                82,
+                                506,
+                                95
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        82,
+                                        506,
+                                        95
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of theorem 4), to show that the output of our convolutional networks converge in distribution to a",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                93,
+                                506,
+                                106
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        93,
+                                        190,
+                                        106
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Gaussian process as",
+                                    "type": "text",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        190,
+                                        95,
+                                        225,
+                                        104
+                                    ],
+                                    "score": 0.87,
+                                    "content": "n  \\infty",
+                                    "type": "inline_equation",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        225,
+                                        93,
+                                        506,
+                                        106
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", with mean zero and covariance given by the recursion in Eqs. (10 –",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                103,
+                                126,
+                                118
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        103,
+                                        126,
+                                        118
+                                    ],
+                                    "score": 1.0,
+                                    "content": "12).",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 2
+                        }
+                    ],
+                    "index": 42.5,
+                    "bbox_fs": [
+                        105,
+                        686,
+                        507,
+                        734
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        82,
+                        505,
+                        116
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                82,
+                                506,
+                                95
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        82,
+                                        506,
+                                        95
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of theorem 4), to show that the output of our convolutional networks converge in distribution to a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                93,
+                                506,
+                                106
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        93,
+                                        190,
+                                        106
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Gaussian process as",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        190,
+                                        95,
+                                        225,
+                                        104
+                                    ],
+                                    "score": 0.87,
+                                    "content": "n  \\infty",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        225,
+                                        93,
+                                        506,
+                                        106
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", with mean zero and covariance given by the recursion in Eqs. (10 –",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                103,
+                                126,
+                                118
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        103,
+                                        126,
+                                        118
+                                    ],
+                                    "score": 1.0,
+                                    "content": "12).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        }
+                    ],
+                    "index": 1
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        120,
+                        505,
+                        212
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                120,
+                                506,
+                                135
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        120,
+                                        506,
+                                        135
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The proof in Matthews et al. (2018b) has three main steps. First, they use the Cramér-Wold device,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                132,
+                                506,
+                                145
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        132,
+                                        506,
+                                        145
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to reduce the full problem to that of proving convergence of scalar random variables to a Gaussian",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                143,
+                                505,
+                                156
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        143,
+                                        505,
+                                        156
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with specified variance. Second, if the previous layers have finite numbers of channels, then the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                154,
+                                506,
+                                171
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        154,
+                                        144,
+                                        171
+                                    ],
+                                    "score": 1.0,
+                                    "content": "channels",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        144,
+                                        154,
+                                        177,
+                                        170
+                                    ],
+                                    "score": 0.92,
+                                    "content": "{ \\bf a } _ { j } ^ { ( \\ell ) } ( { \\bf X } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        178,
+                                        154,
+                                        195,
+                                        171
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        196,
+                                        154,
+                                        232,
+                                        171
+                                    ],
+                                    "score": 0.95,
+                                    "content": "\\mathbf { a } _ { j } ^ { ( \\ell ) } ( \\mathbf { X } ^ { \\prime } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        232,
+                                        154,
+                                        506,
+                                        171
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are uncorrelated but no longer independent, so we cannot apply the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                166,
+                                506,
+                                181
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        166,
+                                        506,
+                                        181
+                                    ],
+                                    "score": 1.0,
+                                    "content": "CLT directly, as we did in the main text. Instead, they write the activations as a sum of exchangeable",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                178,
+                                505,
+                                191
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        178,
+                                        505,
+                                        191
+                                    ],
+                                    "score": 1.0,
+                                    "content": "random variables, and derive an adapted CLT for exchangeable (rather than independent) random",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                189,
+                                504,
+                                203
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        189,
+                                        504,
+                                        203
+                                    ],
+                                    "score": 1.0,
+                                    "content": "variables (Blum et al., 1958). Third, they show that moment conditions required by their exchange-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                200,
+                                199,
+                                212
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        200,
+                                        199,
+                                        212
+                                    ],
+                                    "score": 1.0,
+                                    "content": "able CLT are satisfied.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        }
+                    ],
+                    "index": 6.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        217,
+                        504,
+                        240
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                217,
+                                506,
+                                230
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        217,
+                                        506,
+                                        230
+                                    ],
+                                    "score": 1.0,
+                                    "content": "To extend their proofs to the convolutional case, we begin by defining our networks in a form that is",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                228,
+                                456,
+                                241
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        228,
+                                        456,
+                                        241
+                                    ],
+                                    "score": 1.0,
+                                    "content": "easier to manipulate and as close as possible to Eq. (21-23) in Matthews et al. (2018b),",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        }
+                    ],
+                    "index": 11.5
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        127,
+                        245,
+                        468,
+                        345
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                127,
+                                245,
+                                468,
+                                345
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        245,
+                                        468,
+                                        345
+                                    ],
+                                    "score": 0.95,
+                                    "content": "\\begin{array} { c } { { A _ { i , \\mu } ^ { ( 1 ) } = f _ { i , \\mu } ^ { ( 1 ) } ( x ) = \\displaystyle \\frac { \\sigma _ { \\mathbf { w } } } { \\sqrt { C ^ { ( 0 ) } } } \\sum _ { j = 1 } ^ { C ^ { ( 0 ) } } \\sum _ { \\nu \\in \\mu \\mathbb { h } \\mathrm { p a t e h } } \\epsilon _ { i , j , \\mu , \\nu } ^ { ( 1 ) } x _ { j , \\nu } + b _ { i } ^ { ( 1 ) } , \\quad i \\in \\mathbb { N } } } \\\\ { { g _ { i , \\mu } ^ { ( \\ell ) } ( x ) = \\phi \\left( f _ { i , \\mu } ^ { ( \\ell ) } ( x ) \\right) } } \\\\ { { A _ { i , \\mu } ^ { ( \\ell + 1 ) } = f _ { i , \\mu } ^ { ( \\ell + 1 ) } ( x ) = \\displaystyle \\frac { \\sigma _ { \\mathbf { w } } } { \\sqrt { C ^ { ( \\ell ) } ( n ) } } \\sum _ { j = 1 } ^ { C ^ { ( \\ell ) } } \\sum _ { \\nu \\in \\mu \\mathrm { t h } \\mathrm { p a t e h } } \\epsilon _ { i , j , \\mu , \\nu } ^ { ( \\ell + 1 ) } g _ { j , \\nu } ^ { ( \\ell ) } ( x ) + b _ { i } ^ { ( \\ell + 1 ) } , \\quad i \\in \\mathbb { N } } } \\end{array}",
+                                    "type": "interline_equation",
+                                    "image_path": "5ac0c6fe0b60b6f942651d07a0b16b9896617fb2361fb2788fbba4e2dc4d936a.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 14,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                127,
+                                245,
+                                468,
+                                278.3333333333333
+                            ],
+                            "spans": [],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                278.3333333333333,
+                                468,
+                                311.66666666666663
+                            ],
+                            "spans": [],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                311.66666666666663,
+                                468,
+                                344.99999999999994
+                            ],
+                            "spans": [],
+                            "index": 15
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        353,
+                        135,
+                        363
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                350,
+                                137,
+                                366
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        350,
+                                        137,
+                                        366
+                                    ],
+                                    "score": 1.0,
+                                    "content": "where,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        }
+                    ],
+                    "index": 16
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        182,
+                        372,
+                        259,
+                        386
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                182,
+                                372,
+                                259,
+                                386
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        182,
+                                        372,
+                                        259,
+                                        386
+                                    ],
+                                    "score": 0.84,
+                                    "content": "\\epsilon _ { i , j , \\mu , \\nu } \\sim \\mathcal { N } ( 0 , 1 ) .",
+                                    "type": "interline_equation",
+                                    "image_path": "377587ad330efcc50a95e3c6182081ba16330c814e44797308546f20e4d79e77.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 17,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                182,
+                                372,
+                                259,
+                                386
+                            ],
+                            "spans": [],
+                            "index": 17
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        397,
+                        505,
+                        458
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                397,
+                                505,
+                                410
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        397,
+                                        505,
+                                        410
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The first step is to use the Cramér-Wold device (Lemma 6 in Matthews et al., 2018b), which indicates",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                409,
+                                505,
+                                420
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        409,
+                                        505,
+                                        420
+                                    ],
+                                    "score": 1.0,
+                                    "content": "that convergence in distribution of a sequence of finite-dimensional vectors is equivalent to conver-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                420,
+                                505,
+                                432
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        420,
+                                        505,
+                                        432
+                                    ],
+                                    "score": 1.0,
+                                    "content": "gence on all possible linear projections to the corresponding real-valued random variable. Mirroring",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                102,
+                                430,
+                                506,
+                                449
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        102,
+                                        430,
+                                        432,
+                                        449
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Eq. 25 in Matthews et al. (2018b), we consider convergence of random vectors,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        433,
+                                        430,
+                                        501,
+                                        447
+                                    ],
+                                    "score": 0.92,
+                                    "content": "f _ { i , \\mu } ^ { ( \\ell ) } ( x ) [ n ] - b _ { i } ^ { ( \\ell ) }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        501,
+                                        430,
+                                        506,
+                                        449
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                445,
+                                201,
+                                459
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        445,
+                                        201,
+                                        459
+                                    ],
+                                    "score": 1.0,
+                                    "content": "projected onto α(x,i,µ),",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        }
+                    ],
+                    "index": 20
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        193,
+                        463,
+                        417,
+                        493
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                193,
+                                463,
+                                417,
+                                493
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        193,
+                                        463,
+                                        417,
+                                        493
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\mathcal { T } ^ { ( \\ell ) } \\left( \\mathcal { L } , \\alpha \\right) \\left[ n \\right] = \\sum _ { ( x , i , \\mu ) \\in \\mathcal { L } } \\alpha ^ { ( x , i , \\mu ) } \\left[ f _ { i , \\mu } ^ { ( \\ell ) } ( x ) [ n ] - b _ { i } ^ { ( \\ell ) } \\right] .",
+                                    "type": "interline_equation",
+                                    "image_path": "cd4997d6c67a70841410e50fd66d16ca4c261c7e83d0dac73d11e7138fa29fb7.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 23.5,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                193,
+                                463,
+                                417,
+                                478.0
+                            ],
+                            "spans": [],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                193,
+                                478.0,
+                                417,
+                                493.0
+                            ],
+                            "spans": [],
+                            "index": 24
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        499,
+                        506,
+                        534
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                498,
+                                505,
+                                513
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        498,
+                                        133,
+                                        513
+                                    ],
+                                    "score": 1.0,
+                                    "content": "where",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        133,
+                                        499,
+                                        270,
+                                        513
+                                    ],
+                                    "score": 0.94,
+                                    "content": "\\mathcal { L } \\subset \\mathcal { X } \\times \\mathbb { N } \\times \\{ 1 , \\ldots , H ^ { ( \\ell ) } D ^ { ( \\ell ) } \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        271,
+                                        498,
+                                        505,
+                                        513
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is a finite set of tuples of data points and channel indicies,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                107,
+                                511,
+                                505,
+                                524
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        513,
+                                        111,
+                                        522
+                                    ],
+                                    "score": 0.73,
+                                    "content": "i",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        111,
+                                        511,
+                                        334,
+                                        523
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", and indicies of elements within channels/feature maps,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        335,
+                                        513,
+                                        342,
+                                        523
+                                    ],
+                                    "score": 0.79,
+                                    "content": "\\mu",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        342,
+                                        511,
+                                        389,
+                                        523
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". The suffix",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        389,
+                                        512,
+                                        401,
+                                        524
+                                    ],
+                                    "score": 0.89,
+                                    "content": "[ n ]",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        402,
+                                        511,
+                                        505,
+                                        523
+                                    ],
+                                    "score": 1.0,
+                                    "content": "indicates width functions",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                521,
+                                244,
+                                535
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        521,
+                                        233,
+                                        535
+                                    ],
+                                    "score": 1.0,
+                                    "content": "that are instantiated with input,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        233,
+                                        525,
+                                        240,
+                                        532
+                                    ],
+                                    "score": 0.72,
+                                    "content": "n",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        240,
+                                        521,
+                                        244,
+                                        535
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        }
+                    ],
+                    "index": 26
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        538,
+                        504,
+                        562
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                537,
+                                505,
+                                553
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        537,
+                                        505,
+                                        553
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Now, we must prove that these projections converge in distribution a Gaussian. We begin by defining",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                551,
+                                314,
+                                562
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        551,
+                                        314,
+                                        562
+                                    ],
+                                    "score": 1.0,
+                                    "content": "summands, as in Eq. 26 in Matthews et al. (2018b),",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        }
+                    ],
+                    "index": 28.5
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        168,
+                        567,
+                        442,
+                        596
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                168,
+                                567,
+                                442,
+                                596
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        168,
+                                        567,
+                                        442,
+                                        596
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\gamma _ { j } ^ { ( \\ell ) } \\left( \\mathcal { L } , \\alpha \\right) [ n ] : = \\sigma _ { \\mathrm { w } } \\sum _ { ( x , i , \\mu ) \\in \\mathcal { L } } \\alpha ^ { ( x , i , \\mu ) } \\sum _ { \\nu \\in \\mu \\mathrm { t h p a t c h } } \\epsilon _ { i , j , \\mu , \\nu } ^ { ( \\ell ) } g _ { j , \\nu } ^ { ( \\ell - 1 ) } ( x ) [ n ] ,",
+                                    "type": "interline_equation",
+                                    "image_path": "3ed295d2d79e28d39103a42271becb9cebc58bde3a117ebe29ab944b80dffd4b.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 31,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                168,
+                                567,
+                                442,
+                                576.6666666666666
+                            ],
+                            "spans": [],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                168,
+                                576.6666666666666,
+                                442,
+                                586.3333333333333
+                            ],
+                            "spans": [],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                168,
+                                586.3333333333333,
+                                442,
+                                595.9999999999999
+                            ],
+                            "spans": [],
+                            "index": 32
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        600,
+                        504,
+                        624
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                600,
+                                505,
+                                614
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        600,
+                                        505,
+                                        614
+                                    ],
+                                    "score": 1.0,
+                                    "content": "such that the projections can be written as a sum of the summands, exactly as in Eq. 27 in Matthews",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                612,
+                                165,
+                                624
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        612,
+                                        165,
+                                        624
+                                    ],
+                                    "score": 1.0,
+                                    "content": "et al. (2018b),",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        }
+                    ],
+                    "index": 33.5
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        192,
+                        628,
+                        418,
+                        667
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                192,
+                                628,
+                                418,
+                                667
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        192,
+                                        628,
+                                        418,
+                                        667
+                                    ],
+                                    "score": 0.94,
+                                    "content": "\\mathcal { T } ^ { ( \\ell ) } \\left( \\mathcal { L } , \\alpha \\right) [ n ] = \\frac { 1 } { \\sqrt { C ^ { ( \\ell - 1 ) } ( n ) } } \\sum _ { j = 1 } ^ { C ^ { ( \\ell - 1 ) } ( n ) } \\gamma _ { j } ^ { ( \\ell ) } \\left( \\mathcal { L } , \\alpha \\right) [ n ] .",
+                                    "type": "interline_equation",
+                                    "image_path": "699567a0d3051b276ae2add7f5ce075f543504cbdfeacc1ea663050f2b30baca.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 36,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                192,
+                                628,
+                                418,
+                                641.0
+                            ],
+                            "spans": [],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                192,
+                                641.0,
+                                418,
+                                654.0
+                            ],
+                            "spans": [],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                192,
+                                654.0,
+                                418,
+                                667.0
+                            ],
+                            "spans": [],
+                            "index": 37
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        672,
+                        505,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                672,
+                                506,
+                                687
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        672,
+                                        339,
+                                        687
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Now we can apply the exchangeable CLT to prove that",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        339,
+                                        672,
+                                        398,
+                                        686
+                                    ],
+                                    "score": 0.93,
+                                    "content": "{ \\mathcal { T } } ^ { ( \\ell ) } \\left( { \\mathcal { L } } , \\alpha \\right) [ n ]",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        398,
+                                        672,
+                                        506,
+                                        687
+                                    ],
+                                    "score": 1.0,
+                                    "content": "converges to the limiting",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                685,
+                                505,
+                                697
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        685,
+                                        505,
+                                        697
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Gaussian implied by the recursions in the main text. To apply the exchangeable CLT, the first step is",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                695,
+                                505,
+                                708
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        695,
+                                        413,
+                                        708
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to mirror Lemma 8 in Matthews et al. (2018b), in showing that for each fixed",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        414,
+                                        697,
+                                        420,
+                                        705
+                                    ],
+                                    "score": 0.79,
+                                    "content": "n",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        421,
+                                        695,
+                                        438,
+                                        708
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        438,
+                                        696,
+                                        505,
+                                        708
+                                    ],
+                                    "score": 0.9,
+                                    "content": "\\ell \\in \\{ 2 , \\ldots , L +",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                705,
+                                507,
+                                724
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        709,
+                                        117,
+                                        722
+                                    ],
+                                    "score": 0.58,
+                                    "content": "1 \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        118,
+                                        705,
+                                        186,
+                                        724
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", the summands,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        187,
+                                        707,
+                                        244,
+                                        723
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\gamma _ { j } ^ { ( \\ell ) } \\left( \\mathcal { L } , \\alpha \\right) [ n ]",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        244,
+                                        705,
+                                        423,
+                                        724
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are exchangeable with respect to the index",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        423,
+                                        711,
+                                        429,
+                                        721
+                                    ],
+                                    "score": 0.79,
+                                    "content": "j",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        429,
+                                        705,
+                                        507,
+                                        724
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". In particular, we",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                720,
+                                506,
+                                733
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        720,
+                                        506,
+                                        733
+                                    ],
+                                    "score": 1.0,
+                                    "content": "apply de Finetti’s theorem, which states that a sequence of random variables is exchangeable if and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        }
+                    ],
+                    "index": 40
+                }
+            ],
+            "page_idx": 12,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        107,
+                        27,
+                        293,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                26,
+                                294,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        26,
+                                        294,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Published as a conference paper at ICLR 2019",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        300,
+                        751,
+                        311,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                299,
+                                750,
+                                313,
+                                764
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        299,
+                                        750,
+                                        313,
+                                        764
+                                    ],
+                                    "score": 1.0,
+                                    "content": "13",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        82,
+                        505,
+                        116
+                    ],
+                    "lines": [],
+                    "index": 1,
+                    "bbox_fs": [
+                        105,
+                        82,
+                        506,
+                        118
+                    ],
+                    "lines_deleted": true
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        120,
+                        505,
+                        212
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                120,
+                                506,
+                                135
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        120,
+                                        506,
+                                        135
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The proof in Matthews et al. (2018b) has three main steps. First, they use the Cramér-Wold device,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                132,
+                                506,
+                                145
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        132,
+                                        506,
+                                        145
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to reduce the full problem to that of proving convergence of scalar random variables to a Gaussian",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                143,
+                                505,
+                                156
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        143,
+                                        505,
+                                        156
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with specified variance. Second, if the previous layers have finite numbers of channels, then the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                154,
+                                506,
+                                171
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        154,
+                                        144,
+                                        171
+                                    ],
+                                    "score": 1.0,
+                                    "content": "channels",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        144,
+                                        154,
+                                        177,
+                                        170
+                                    ],
+                                    "score": 0.92,
+                                    "content": "{ \\bf a } _ { j } ^ { ( \\ell ) } ( { \\bf X } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        178,
+                                        154,
+                                        195,
+                                        171
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        196,
+                                        154,
+                                        232,
+                                        171
+                                    ],
+                                    "score": 0.95,
+                                    "content": "\\mathbf { a } _ { j } ^ { ( \\ell ) } ( \\mathbf { X } ^ { \\prime } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        232,
+                                        154,
+                                        506,
+                                        171
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are uncorrelated but no longer independent, so we cannot apply the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                166,
+                                506,
+                                181
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        166,
+                                        506,
+                                        181
+                                    ],
+                                    "score": 1.0,
+                                    "content": "CLT directly, as we did in the main text. Instead, they write the activations as a sum of exchangeable",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                178,
+                                505,
+                                191
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        178,
+                                        505,
+                                        191
+                                    ],
+                                    "score": 1.0,
+                                    "content": "random variables, and derive an adapted CLT for exchangeable (rather than independent) random",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                189,
+                                504,
+                                203
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        189,
+                                        504,
+                                        203
+                                    ],
+                                    "score": 1.0,
+                                    "content": "variables (Blum et al., 1958). Third, they show that moment conditions required by their exchange-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                200,
+                                199,
+                                212
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        200,
+                                        199,
+                                        212
+                                    ],
+                                    "score": 1.0,
+                                    "content": "able CLT are satisfied.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        }
+                    ],
+                    "index": 6.5,
+                    "bbox_fs": [
+                        104,
+                        120,
+                        506,
+                        212
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        217,
+                        504,
+                        240
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                217,
+                                506,
+                                230
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        217,
+                                        506,
+                                        230
+                                    ],
+                                    "score": 1.0,
+                                    "content": "To extend their proofs to the convolutional case, we begin by defining our networks in a form that is",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                228,
+                                456,
+                                241
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        228,
+                                        456,
+                                        241
+                                    ],
+                                    "score": 1.0,
+                                    "content": "easier to manipulate and as close as possible to Eq. (21-23) in Matthews et al. (2018b),",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        }
+                    ],
+                    "index": 11.5,
+                    "bbox_fs": [
+                        106,
+                        217,
+                        506,
+                        241
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        127,
+                        245,
+                        468,
+                        345
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                127,
+                                245,
+                                468,
+                                345
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        245,
+                                        468,
+                                        345
+                                    ],
+                                    "score": 0.95,
+                                    "content": "\\begin{array} { c } { { A _ { i , \\mu } ^ { ( 1 ) } = f _ { i , \\mu } ^ { ( 1 ) } ( x ) = \\displaystyle \\frac { \\sigma _ { \\mathbf { w } } } { \\sqrt { C ^ { ( 0 ) } } } \\sum _ { j = 1 } ^ { C ^ { ( 0 ) } } \\sum _ { \\nu \\in \\mu \\mathbb { h } \\mathrm { p a t e h } } \\epsilon _ { i , j , \\mu , \\nu } ^ { ( 1 ) } x _ { j , \\nu } + b _ { i } ^ { ( 1 ) } , \\quad i \\in \\mathbb { N } } } \\\\ { { g _ { i , \\mu } ^ { ( \\ell ) } ( x ) = \\phi \\left( f _ { i , \\mu } ^ { ( \\ell ) } ( x ) \\right) } } \\\\ { { A _ { i , \\mu } ^ { ( \\ell + 1 ) } = f _ { i , \\mu } ^ { ( \\ell + 1 ) } ( x ) = \\displaystyle \\frac { \\sigma _ { \\mathbf { w } } } { \\sqrt { C ^ { ( \\ell ) } ( n ) } } \\sum _ { j = 1 } ^ { C ^ { ( \\ell ) } } \\sum _ { \\nu \\in \\mu \\mathrm { t h } \\mathrm { p a t e h } } \\epsilon _ { i , j , \\mu , \\nu } ^ { ( \\ell + 1 ) } g _ { j , \\nu } ^ { ( \\ell ) } ( x ) + b _ { i } ^ { ( \\ell + 1 ) } , \\quad i \\in \\mathbb { N } } } \\end{array}",
+                                    "type": "interline_equation",
+                                    "image_path": "5ac0c6fe0b60b6f942651d07a0b16b9896617fb2361fb2788fbba4e2dc4d936a.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 14,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                127,
+                                245,
+                                468,
+                                278.3333333333333
+                            ],
+                            "spans": [],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                278.3333333333333,
+                                468,
+                                311.66666666666663
+                            ],
+                            "spans": [],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                311.66666666666663,
+                                468,
+                                344.99999999999994
+                            ],
+                            "spans": [],
+                            "index": 15
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        353,
+                        135,
+                        363
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                350,
+                                137,
+                                366
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        350,
+                                        137,
+                                        366
+                                    ],
+                                    "score": 1.0,
+                                    "content": "where,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        }
+                    ],
+                    "index": 16,
+                    "bbox_fs": [
+                        106,
+                        350,
+                        137,
+                        366
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        182,
+                        372,
+                        259,
+                        386
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                182,
+                                372,
+                                259,
+                                386
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        182,
+                                        372,
+                                        259,
+                                        386
+                                    ],
+                                    "score": 0.84,
+                                    "content": "\\epsilon _ { i , j , \\mu , \\nu } \\sim \\mathcal { N } ( 0 , 1 ) .",
+                                    "type": "interline_equation",
+                                    "image_path": "377587ad330efcc50a95e3c6182081ba16330c814e44797308546f20e4d79e77.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 17,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                182,
+                                372,
+                                259,
+                                386
+                            ],
+                            "spans": [],
+                            "index": 17
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        397,
+                        505,
+                        458
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                397,
+                                505,
+                                410
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        397,
+                                        505,
+                                        410
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The first step is to use the Cramér-Wold device (Lemma 6 in Matthews et al., 2018b), which indicates",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                409,
+                                505,
+                                420
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        409,
+                                        505,
+                                        420
+                                    ],
+                                    "score": 1.0,
+                                    "content": "that convergence in distribution of a sequence of finite-dimensional vectors is equivalent to conver-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                420,
+                                505,
+                                432
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        420,
+                                        505,
+                                        432
+                                    ],
+                                    "score": 1.0,
+                                    "content": "gence on all possible linear projections to the corresponding real-valued random variable. Mirroring",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                102,
+                                430,
+                                506,
+                                449
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        102,
+                                        430,
+                                        432,
+                                        449
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Eq. 25 in Matthews et al. (2018b), we consider convergence of random vectors,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        433,
+                                        430,
+                                        501,
+                                        447
+                                    ],
+                                    "score": 0.92,
+                                    "content": "f _ { i , \\mu } ^ { ( \\ell ) } ( x ) [ n ] - b _ { i } ^ { ( \\ell ) }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        501,
+                                        430,
+                                        506,
+                                        449
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                445,
+                                201,
+                                459
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        445,
+                                        201,
+                                        459
+                                    ],
+                                    "score": 1.0,
+                                    "content": "projected onto α(x,i,µ),",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        }
+                    ],
+                    "index": 20,
+                    "bbox_fs": [
+                        102,
+                        397,
+                        506,
+                        459
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        193,
+                        463,
+                        417,
+                        493
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                193,
+                                463,
+                                417,
+                                493
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        193,
+                                        463,
+                                        417,
+                                        493
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\mathcal { T } ^ { ( \\ell ) } \\left( \\mathcal { L } , \\alpha \\right) \\left[ n \\right] = \\sum _ { ( x , i , \\mu ) \\in \\mathcal { L } } \\alpha ^ { ( x , i , \\mu ) } \\left[ f _ { i , \\mu } ^ { ( \\ell ) } ( x ) [ n ] - b _ { i } ^ { ( \\ell ) } \\right] .",
+                                    "type": "interline_equation",
+                                    "image_path": "cd4997d6c67a70841410e50fd66d16ca4c261c7e83d0dac73d11e7138fa29fb7.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 23.5,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                193,
+                                463,
+                                417,
+                                478.0
+                            ],
+                            "spans": [],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                193,
+                                478.0,
+                                417,
+                                493.0
+                            ],
+                            "spans": [],
+                            "index": 24
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        499,
+                        506,
+                        534
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                498,
+                                505,
+                                513
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        498,
+                                        133,
+                                        513
+                                    ],
+                                    "score": 1.0,
+                                    "content": "where",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        133,
+                                        499,
+                                        270,
+                                        513
+                                    ],
+                                    "score": 0.94,
+                                    "content": "\\mathcal { L } \\subset \\mathcal { X } \\times \\mathbb { N } \\times \\{ 1 , \\ldots , H ^ { ( \\ell ) } D ^ { ( \\ell ) } \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        271,
+                                        498,
+                                        505,
+                                        513
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is a finite set of tuples of data points and channel indicies,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                107,
+                                511,
+                                505,
+                                524
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        513,
+                                        111,
+                                        522
+                                    ],
+                                    "score": 0.73,
+                                    "content": "i",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        111,
+                                        511,
+                                        334,
+                                        523
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", and indicies of elements within channels/feature maps,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        335,
+                                        513,
+                                        342,
+                                        523
+                                    ],
+                                    "score": 0.79,
+                                    "content": "\\mu",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        342,
+                                        511,
+                                        389,
+                                        523
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". The suffix",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        389,
+                                        512,
+                                        401,
+                                        524
+                                    ],
+                                    "score": 0.89,
+                                    "content": "[ n ]",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        402,
+                                        511,
+                                        505,
+                                        523
+                                    ],
+                                    "score": 1.0,
+                                    "content": "indicates width functions",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                521,
+                                244,
+                                535
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        521,
+                                        233,
+                                        535
+                                    ],
+                                    "score": 1.0,
+                                    "content": "that are instantiated with input,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        233,
+                                        525,
+                                        240,
+                                        532
+                                    ],
+                                    "score": 0.72,
+                                    "content": "n",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        240,
+                                        521,
+                                        244,
+                                        535
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        }
+                    ],
+                    "index": 26,
+                    "bbox_fs": [
+                        105,
+                        498,
+                        505,
+                        535
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        538,
+                        504,
+                        562
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                537,
+                                505,
+                                553
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        537,
+                                        505,
+                                        553
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Now, we must prove that these projections converge in distribution a Gaussian. We begin by defining",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                551,
+                                314,
+                                562
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        551,
+                                        314,
+                                        562
+                                    ],
+                                    "score": 1.0,
+                                    "content": "summands, as in Eq. 26 in Matthews et al. (2018b),",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        }
+                    ],
+                    "index": 28.5,
+                    "bbox_fs": [
+                        105,
+                        537,
+                        505,
+                        562
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        168,
+                        567,
+                        442,
+                        596
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                168,
+                                567,
+                                442,
+                                596
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        168,
+                                        567,
+                                        442,
+                                        596
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\gamma _ { j } ^ { ( \\ell ) } \\left( \\mathcal { L } , \\alpha \\right) [ n ] : = \\sigma _ { \\mathrm { w } } \\sum _ { ( x , i , \\mu ) \\in \\mathcal { L } } \\alpha ^ { ( x , i , \\mu ) } \\sum _ { \\nu \\in \\mu \\mathrm { t h p a t c h } } \\epsilon _ { i , j , \\mu , \\nu } ^ { ( \\ell ) } g _ { j , \\nu } ^ { ( \\ell - 1 ) } ( x ) [ n ] ,",
+                                    "type": "interline_equation",
+                                    "image_path": "3ed295d2d79e28d39103a42271becb9cebc58bde3a117ebe29ab944b80dffd4b.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 31,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                168,
+                                567,
+                                442,
+                                576.6666666666666
+                            ],
+                            "spans": [],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                168,
+                                576.6666666666666,
+                                442,
+                                586.3333333333333
+                            ],
+                            "spans": [],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                168,
+                                586.3333333333333,
+                                442,
+                                595.9999999999999
+                            ],
+                            "spans": [],
+                            "index": 32
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        600,
+                        504,
+                        624
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                600,
+                                505,
+                                614
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        600,
+                                        505,
+                                        614
+                                    ],
+                                    "score": 1.0,
+                                    "content": "such that the projections can be written as a sum of the summands, exactly as in Eq. 27 in Matthews",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                612,
+                                165,
+                                624
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        612,
+                                        165,
+                                        624
+                                    ],
+                                    "score": 1.0,
+                                    "content": "et al. (2018b),",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        }
+                    ],
+                    "index": 33.5,
+                    "bbox_fs": [
+                        105,
+                        600,
+                        505,
+                        624
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        192,
+                        628,
+                        418,
+                        667
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                192,
+                                628,
+                                418,
+                                667
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        192,
+                                        628,
+                                        418,
+                                        667
+                                    ],
+                                    "score": 0.94,
+                                    "content": "\\mathcal { T } ^ { ( \\ell ) } \\left( \\mathcal { L } , \\alpha \\right) [ n ] = \\frac { 1 } { \\sqrt { C ^ { ( \\ell - 1 ) } ( n ) } } \\sum _ { j = 1 } ^ { C ^ { ( \\ell - 1 ) } ( n ) } \\gamma _ { j } ^ { ( \\ell ) } \\left( \\mathcal { L } , \\alpha \\right) [ n ] .",
+                                    "type": "interline_equation",
+                                    "image_path": "699567a0d3051b276ae2add7f5ce075f543504cbdfeacc1ea663050f2b30baca.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 36,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                192,
+                                628,
+                                418,
+                                641.0
+                            ],
+                            "spans": [],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                192,
+                                641.0,
+                                418,
+                                654.0
+                            ],
+                            "spans": [],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                192,
+                                654.0,
+                                418,
+                                667.0
+                            ],
+                            "spans": [],
+                            "index": 37
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        672,
+                        505,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                672,
+                                506,
+                                687
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        672,
+                                        339,
+                                        687
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Now we can apply the exchangeable CLT to prove that",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        339,
+                                        672,
+                                        398,
+                                        686
+                                    ],
+                                    "score": 0.93,
+                                    "content": "{ \\mathcal { T } } ^ { ( \\ell ) } \\left( { \\mathcal { L } } , \\alpha \\right) [ n ]",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        398,
+                                        672,
+                                        506,
+                                        687
+                                    ],
+                                    "score": 1.0,
+                                    "content": "converges to the limiting",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                685,
+                                505,
+                                697
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        685,
+                                        505,
+                                        697
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Gaussian implied by the recursions in the main text. To apply the exchangeable CLT, the first step is",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                695,
+                                505,
+                                708
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        695,
+                                        413,
+                                        708
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to mirror Lemma 8 in Matthews et al. (2018b), in showing that for each fixed",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        414,
+                                        697,
+                                        420,
+                                        705
+                                    ],
+                                    "score": 0.79,
+                                    "content": "n",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        421,
+                                        695,
+                                        438,
+                                        708
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        438,
+                                        696,
+                                        505,
+                                        708
+                                    ],
+                                    "score": 0.9,
+                                    "content": "\\ell \\in \\{ 2 , \\ldots , L +",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                705,
+                                507,
+                                724
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        709,
+                                        117,
+                                        722
+                                    ],
+                                    "score": 0.58,
+                                    "content": "1 \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        118,
+                                        705,
+                                        186,
+                                        724
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", the summands,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        187,
+                                        707,
+                                        244,
+                                        723
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\gamma _ { j } ^ { ( \\ell ) } \\left( \\mathcal { L } , \\alpha \\right) [ n ]",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        244,
+                                        705,
+                                        423,
+                                        724
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are exchangeable with respect to the index",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        423,
+                                        711,
+                                        429,
+                                        721
+                                    ],
+                                    "score": 0.79,
+                                    "content": "j",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        429,
+                                        705,
+                                        507,
+                                        724
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". In particular, we",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                720,
+                                506,
+                                733
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        720,
+                                        506,
+                                        733
+                                    ],
+                                    "score": 1.0,
+                                    "content": "apply de Finetti’s theorem, which states that a sequence of random variables is exchangeable if and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                83,
+                                505,
+                                95
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        83,
+                                        505,
+                                        95
+                                    ],
+                                    "score": 1.0,
+                                    "content": "only if they are i.i.d. conditional on some set of random variables, so it is sufficient to exhibit such",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                93,
+                                488,
+                                107
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        93,
+                                        488,
+                                        107
+                                    ],
+                                    "score": 1.0,
+                                    "content": "a set of random variables. Mirroring Eq. 29 in Matthews et al. (2018b), we apply the recursion,",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 1
+                        }
+                    ],
+                    "index": 40,
+                    "bbox_fs": [
+                        104,
+                        672,
+                        507,
+                        733
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        104,
+                        81,
+                        505,
+                        106
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                83,
+                                505,
+                                95
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        83,
+                                        505,
+                                        95
+                                    ],
+                                    "score": 1.0,
+                                    "content": "only if they are i.i.d. conditional on some set of random variables, so it is sufficient to exhibit such",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                93,
+                                488,
+                                107
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        93,
+                                        488,
+                                        107
+                                    ],
+                                    "score": 1.0,
+                                    "content": "a set of random variables. Mirroring Eq. 29 in Matthews et al. (2018b), we apply the recursion,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        }
+                    ],
+                    "index": 0.5
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        111,
+                        106,
+                        514,
+                        146
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                111,
+                                106,
+                                514,
+                                146
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        111,
+                                        106,
+                                        514,
+                                        146
+                                    ],
+                                    "score": 0.86,
+                                    "content": " \\overset { ( \\ell ) } { j } ( \\mathcal { L } , \\alpha ) [ n ] : = \\sigma _ { \\mathrm { w } } \\sum _ { ( x , i , \\mu ) \\in \\mathcal { L } } \\alpha ^ { ( x , i , \\mu ) } \\sum _ { \\nu \\in \\mu \\mathrm { t h } \\mathrm { p a t e h } } \\epsilon _ { i , j , \\mu , \\nu } ^ { ( \\ell ) } \\phi ( \\frac { \\sigma _ { \\mathrm { w } } } { \\sqrt { C ^ { ( \\ell - 2 ) } ( n ) } } \\sum _ { k = 1 } ^ { C ^ { ( \\ell - 2 ) } ( n ) } \\sum _ { \\xi \\in \\nu \\mathrm { t h } \\mathrm { p a t e h } } \\epsilon _ { j , k , \\nu ; \\xi } ^ { ( \\ell - 1 ) } g _ { k , \\xi } ^ { ( \\ell - 2 ) } ) ,",
+                                    "type": "interline_equation",
+                                    "image_path": "17c0e7e933133d259cd80eda41b96e3b779964d857d42527541a14a92629a63e.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 3,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                111,
+                                106,
+                                514,
+                                119.33333333333333
+                            ],
+                            "spans": [],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                111,
+                                119.33333333333333,
+                                514,
+                                132.66666666666666
+                            ],
+                            "spans": [],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                111,
+                                132.66666666666666,
+                                514,
+                                146.0
+                            ],
+                            "spans": [],
+                            "index": 4
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        158,
+                        506,
+                        199
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                158,
+                                509,
+                                194
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        158,
+                                        505,
+                                        171
+                                    ],
+                                    "score": 1.0,
+                                    "content": "As such, the summands are iid conditional on the finite set of random variables",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        170,
+                                        415,
+                                        189
+                                    ],
+                                    "score": 0.71,
+                                    "content": "\\left\\{ g _ { k , \\xi } ^ { ( \\ell - 2 ) } ( x ) [ n ] : k \\in \\{ 1 , \\ldots , C ^ { ( \\ell - 2 ) } \\} , \\xi \\in \\left\\{ 1 , \\ldots , H ^ { ( \\ell - 2 ) } D ^ { ( \\ell - 2 ) } \\right\\} , x \\in \\mathcal { L } _ { \\mathcal { X } } \\right\\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        415,
+                                        163,
+                                        448,
+                                        194
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", where",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        448,
+                                        173,
+                                        464,
+                                        184
+                                    ],
+                                    "score": 0.88,
+                                    "content": "\\mathcal { L } _ { \\mathcal { X } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        464,
+                                        163,
+                                        509,
+                                        194
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is the set",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                186,
+                                189,
+                                199
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        186,
+                                        177,
+                                        199
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of input points in",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        178,
+                                        187,
+                                        185,
+                                        197
+                                    ],
+                                    "score": 0.72,
+                                    "content": "\\mathcal { L }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        186,
+                                        186,
+                                        189,
+                                        199
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        }
+                    ],
+                    "index": 5.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        204,
+                        505,
+                        240
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                202,
+                                505,
+                                218
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        202,
+                                        424,
+                                        218
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The exchangeable CLT in Lemma 10 in Matthews et al. (2018b) indicates that",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        424,
+                                        204,
+                                        483,
+                                        217
+                                    ],
+                                    "score": 0.93,
+                                    "content": "{ \\mathcal { T } } ^ { ( \\ell ) } \\left( { \\mathcal { L } } , \\alpha \\right) [ n ]",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        483,
+                                        202,
+                                        505,
+                                        218
+                                    ],
+                                    "score": 1.0,
+                                    "content": "con-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                216,
+                                505,
+                                230
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        216,
+                                        207,
+                                        229
+                                    ],
+                                    "score": 1.0,
+                                    "content": "verges in distribution to",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        207,
+                                        216,
+                                        248,
+                                        230
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\mathcal { N } \\left( 0 , \\sigma _ { * } ^ { 2 } \\right)",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        248,
+                                        216,
+                                        505,
+                                        229
+                                    ],
+                                    "score": 1.0,
+                                    "content": "if the summands are exchangeable (which we showed above),",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                227,
+                                222,
+                                240
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        227,
+                                        222,
+                                        240
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and if three conditions hold,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        }
+                    ],
+                    "index": 8
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        127,
+                        244,
+                        295,
+                        319
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                127,
+                                244,
+                                295,
+                                319
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        244,
+                                        295,
+                                        319
+                                    ],
+                                    "score": 0.89,
+                                    "content": "\\begin{array} { r l } & { \\mathrm { a ) ~ \\mathbb { E } } _ { n } [ \\gamma _ { j } ^ { ( \\ell ) } \\gamma _ { j ^ { \\prime } } ^ { ( \\ell ) } ] = 0 } \\\\ & { \\mathrm { b ) ~ \\operatorname* { l i m } _ { n  \\infty } \\mathbb { E } } _ { n } [ ( \\gamma _ { j } ^ { ( \\ell ) } ) ^ { 2 } ( \\gamma _ { j ^ { \\prime } } ^ { ( \\ell ) } ) ^ { 2 } ] = \\sigma _ { * } ^ { 4 } } \\\\ & { \\mathrm { c ) ~ \\mathbb { E } } _ { n } [ | \\gamma _ { j } ^ { ( \\ell ) } | ^ { 3 } ] = o ( \\sqrt { C ^ { ( \\ell ) } ( n ) } ) } \\end{array}",
+                                    "type": "interline_equation",
+                                    "image_path": "d2bd2b8a709e0411be37c1c704aac55b083a7ab7c52b160eaccbcf65054d7360.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 12,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                127,
+                                244,
+                                295,
+                                259.0
+                            ],
+                            "spans": [],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                259.0,
+                                295,
+                                274.0
+                            ],
+                            "spans": [],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                274.0,
+                                295,
+                                289.0
+                            ],
+                            "spans": [],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                289.0,
+                                295,
+                                304.0
+                            ],
+                            "spans": [],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                304.0,
+                                295,
+                                319.0
+                            ],
+                            "spans": [],
+                            "index": 14
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        322,
+                        505,
+                        357
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                322,
+                                505,
+                                335
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        322,
+                                        505,
+                                        335
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Condition a) follows immediately as the summands are uncorrelated and zero-mean. Conditions",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                333,
+                                505,
+                                347
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        333,
+                                        505,
+                                        347
+                                    ],
+                                    "score": 1.0,
+                                    "content": "b) and c) are more involved as convergence in distribution in the previous layers does not imply",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                345,
+                                321,
+                                357
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        345,
+                                        321,
+                                        357
+                                    ],
+                                    "score": 1.0,
+                                    "content": "convergence in moments for our activation functions.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        }
+                    ],
+                    "index": 16
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        361,
+                        506,
+                        447
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                362,
+                                506,
+                                374
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        362,
+                                        506,
+                                        374
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We begin by considering the extension of Lemma 20 in Matthews et al. (2018b), which allow us to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                373,
+                                505,
+                                385
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        373,
+                                        505,
+                                        385
+                                    ],
+                                    "score": 1.0,
+                                    "content": "show conditions b) and c) above, even in the case of unbounded but linearly enveloped nonlinearities",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                102,
+                                379,
+                                510,
+                                411
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        102,
+                                        379,
+                                        286,
+                                        411
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(Definition 1 in Matthews et al., 2018b). Lembounded by a finite constant independent of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        315,
+                                        379,
+                                        448,
+                                        411
+                                    ],
+                                    "score": 1.0,
+                                    "content": "states that the eighth moments of . We prove this by induction. Th",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        448,
+                                        384,
+                                        489,
+                                        400
+                                    ],
+                                    "score": 0.94,
+                                    "content": "f _ { i , \\mu } ^ { ( t ) } ( x ) [ n ]",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        489,
+                                        379,
+                                        510,
+                                        411
+                                    ],
+                                    "score": 1.0,
+                                    "content": "aree is",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                286,
+                                398,
+                                315,
+                                409
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        286,
+                                        398,
+                                        315,
+                                        409
+                                    ],
+                                    "score": 0.9,
+                                    "content": "n \\in \\mathbb N",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                102,
+                                406,
+                                508,
+                                430
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        102,
+                                        406,
+                                        146,
+                                        430
+                                    ],
+                                    "score": 1.0,
+                                    "content": "trivial, as",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        146,
+                                        409,
+                                        188,
+                                        426
+                                    ],
+                                    "score": 0.93,
+                                    "content": "f _ { j , \\mu } ^ { ( 1 ) } ( x ) [ n ]",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        189,
+                                        406,
+                                        508,
+                                        430
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is Gaussian. Following Matthews et al. (2018b), assume the condition holds up",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                423,
+                                505,
+                                437
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        423,
+                                        117,
+                                        437
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        117,
+                                        424,
+                                        138,
+                                        434
+                                    ],
+                                    "score": 0.86,
+                                    "content": "\\ell - 1",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        138,
+                                        423,
+                                        314,
+                                        437
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", and show that the condition holds for layer",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        315,
+                                        424,
+                                        320,
+                                        434
+                                    ],
+                                    "score": 0.44,
+                                    "content": "\\ell",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        320,
+                                        423,
+                                        505,
+                                        437
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Using Eq. (21), we can bound the activations",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                434,
+                                149,
+                                448
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        434,
+                                        138,
+                                        448
+                                    ],
+                                    "score": 1.0,
+                                    "content": "at layer",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        139,
+                                        435,
+                                        144,
+                                        445
+                                    ],
+                                    "score": 0.67,
+                                    "content": "\\ell",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        145,
+                                        434,
+                                        149,
+                                        448
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        }
+                    ],
+                    "index": 21
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        112,
+                        447,
+                        480,
+                        493
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                112,
+                                447,
+                                480,
+                                493
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        112,
+                                        447,
+                                        480,
+                                        493
+                                    ],
+                                    "score": 0.94,
+                                    "content": "\\mathbb { E } \\left[ | f _ { i , \\mu } ^ { ( \\ell ) } ( x ) [ n ] | ^ { 8 } \\right] \\leq 2 ^ { 8 - 1 } \\mathbb { E } \\left[ | b _ { i } ^ { ( \\ell ) } | ^ { 8 } + \\left| \\frac { \\sigma _ { \\mathbf { w } } } { \\sqrt { C ^ { ( \\ell - 1 ) } } } \\sum _ { j = 1 } ^ { C ^ { ( \\ell - 1 ) } ( n ) } \\sum _ { \\nu \\in \\mu \\mathrm { t h } \\mathrm { p a t e h } } \\epsilon _ { i , j , \\mu , \\nu } ^ { ( \\ell ) } g _ { j , \\nu } ^ { ( \\ell - 1 ) } ( x ) [ n ] \\right| ^ { 8 } \\right]",
+                                    "type": "interline_equation",
+                                    "image_path": "e1525aadc195dba6d3e8ef0ab2a26a0be8b65495eb06427ba51a80150e64f40d.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 26,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                112,
+                                447,
+                                480,
+                                462.3333333333333
+                            ],
+                            "spans": [],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                112,
+                                462.3333333333333,
+                                480,
+                                477.66666666666663
+                            ],
+                            "spans": [],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                112,
+                                477.66666666666663,
+                                480,
+                                492.99999999999994
+                            ],
+                            "spans": [],
+                            "index": 27
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        493,
+                        508,
+                        516
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                492,
+                                505,
+                                507
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        492,
+                                        505,
+                                        507
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Following Eq. 48 in Matthews et al. (2018b), which uses Lemma 19 in Matthews et al. (2018b), we",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                502,
+                                132,
+                                519
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        502,
+                                        132,
+                                        519
+                                    ],
+                                    "score": 1.0,
+                                    "content": "have,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        }
+                    ],
+                    "index": 28.5
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        117,
+                        519,
+                        478,
+                        599
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                117,
+                                519,
+                                478,
+                                599
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        117,
+                                        519,
+                                        478,
+                                        599
+                                    ],
+                                    "score": 0.94,
+                                    "content": "\\begin{array} { r l } { \\mathbb { E } \\left[ \\left| \\frac { \\sigma _ { \\mathrm { w } } } { \\sqrt { C ^ { ( \\ell - 1 ) } } } \\sum _ { j = 1 } ^ { C ^ { ( \\ell - 1 ) } ( n ) } \\displaystyle \\sum _ { \\nu \\in \\mu \\mathrm { t h } \\mathrm { p a t e h } } \\epsilon _ { i , j , \\mu , \\nu } ^ { ( \\ell ) } g _ { j , \\nu } ^ { ( \\ell - 1 ) } ( x ) [ n ] \\right| ^ { 8 } \\right] } & { } \\\\ { = \\frac { 2 ^ { 4 } \\Gamma ( 4 + 1 / 2 ) } { \\Gamma ( 1 / 2 ) } \\mathbb { E } \\left[ \\left| \\frac { \\sigma _ { \\mathrm { w } } ^ { 2 } } { C ^ { ( \\ell - 1 ) } ( n ) } \\| g _ { j \\in \\{ 1 , \\dots , C ^ { ( \\ell - 1 ) } ( n ) \\} , \\nu \\in \\mu \\mathrm { t h } \\mathrm { p a t c h } } ^ { ( \\ell - 1 ) } ( x ) [ n ] \\| _ { 2 } ^ { 2 } \\right| ^ { 4 } \\right] . } \\end{array}",
+                                    "type": "interline_equation",
+                                    "image_path": "d5c365ee7ab68b781b018199bf3460179ad3b31af3693fa570d752d714ff7103.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 31,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                117,
+                                519,
+                                478,
+                                545.6666666666666
+                            ],
+                            "spans": [],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                117,
+                                545.6666666666666,
+                                478,
+                                572.3333333333333
+                            ],
+                            "spans": [],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                117,
+                                572.3333333333333,
+                                478,
+                                598.9999999999999
+                            ],
+                            "spans": [],
+                            "index": 32
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        602,
+                        507,
+                        631
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                599,
+                                507,
+                                620
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        601,
+                                        135,
+                                        615
+                                    ],
+                                    "score": 1.0,
+                                    "content": "wher",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        130,
+                                        599,
+                                        507,
+                                        620
+                                    ],
+                                    "score": 1.0,
+                                    "content": "e g(`−1)j∈{1,...,C(`−1)(n)},ν∈µth patch(x)[n] is the set of post-nonlinearities corresponding to j ∈",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                107,
+                                615,
+                                464,
+                                632
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        618,
+                                        186,
+                                        631
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\{ 1 , \\ldots , C ^ { ( \\ell - 1 ) } ( n ) \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        187,
+                                        615,
+                                        204,
+                                        632
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        205,
+                                        619,
+                                        237,
+                                        630
+                                    ],
+                                    "score": 0.88,
+                                    "content": "\\nu \\in \\mu \\mathrm { t h }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        238,
+                                        615,
+                                        464,
+                                        632
+                                    ],
+                                    "score": 1.0,
+                                    "content": "patch. Following Matthews et al. (2018b), observe that,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        }
+                    ],
+                    "index": 33.5
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        111,
+                        632,
+                        513,
+                        723
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                111,
+                                632,
+                                513,
+                                723
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        111,
+                                        632,
+                                        513,
+                                        723
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\begin{array} { r } { \\frac { 1 } { \\sum ^ { ( \\ell - 1 ) } \\left( n \\right) } \\| g _ { j \\in \\{ 1 , \\dots , C ^ { ( \\ell - 1 ) } ( n ) \\} , \\nu \\in \\mu \\mathrm { t h p a t e h } } ^ { ( \\ell - 1 ) } ( x ) [ n ] \\| _ { 2 } ^ { 2 } = \\frac { 1 } { C ^ { ( \\ell - 1 ) } ( n ) } \\displaystyle \\sum _ { j = 1 } ^ { C ^ { ( \\ell - 1 ) } ( n ) } \\sum _ { \\nu \\in \\mu \\mathrm { t h p a t e h } } \\left( g _ { j , \\nu } ^ { ( \\ell - 1 ) } ( x ) [ n ] \\right) ^ { 2 } } \\\\ { \\leq \\frac { 1 } { C ^ { ( \\ell - 1 ) } ( n ) } \\displaystyle \\sum _ { j = 1 } ^ { C ^ { ( \\ell - 1 ) } ( n ) } \\sum _ { \\nu \\in \\mu \\mathrm { t h p a t e h } } ^ { C ^ { ( \\ell - 1 ) } ( n ) } \\left( c + m \\big | f _ { j , \\nu } ^ { ( \\ell - 1 ) } ( x ) \\big | \\right. } \\end{array}",
+                                    "type": "interline_equation",
+                                    "image_path": "acefb7cbee5fe7c46f897dadaf5a1e645b463a9143b940c5edeeb1ace37f3784.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 36,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                111,
+                                632,
+                                513,
+                                662.3333333333334
+                            ],
+                            "spans": [],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                111,
+                                662.3333333333334,
+                                513,
+                                692.6666666666667
+                            ],
+                            "spans": [],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                111,
+                                692.6666666666667,
+                                513,
+                                723.0000000000001
+                            ],
+                            "spans": [],
+                            "index": 37
+                        }
+                    ]
+                }
+            ],
+            "page_idx": 13,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        107,
+                        26,
+                        293,
+                        38
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                25,
+                                294,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        25,
+                                        294,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Published as a conference paper at ICLR 2019",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        300,
+                        751,
+                        311,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                299,
+                                750,
+                                313,
+                                764
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        299,
+                                        750,
+                                        313,
+                                        764
+                                    ],
+                                    "score": 1.0,
+                                    "content": "14",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        104,
+                        81,
+                        505,
+                        106
+                    ],
+                    "lines": [],
+                    "index": 0.5,
+                    "bbox_fs": [
+                        105,
+                        83,
+                        505,
+                        107
+                    ],
+                    "lines_deleted": true
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        111,
+                        106,
+                        514,
+                        146
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                111,
+                                106,
+                                514,
+                                146
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        111,
+                                        106,
+                                        514,
+                                        146
+                                    ],
+                                    "score": 0.86,
+                                    "content": " \\overset { ( \\ell ) } { j } ( \\mathcal { L } , \\alpha ) [ n ] : = \\sigma _ { \\mathrm { w } } \\sum _ { ( x , i , \\mu ) \\in \\mathcal { L } } \\alpha ^ { ( x , i , \\mu ) } \\sum _ { \\nu \\in \\mu \\mathrm { t h } \\mathrm { p a t e h } } \\epsilon _ { i , j , \\mu , \\nu } ^ { ( \\ell ) } \\phi ( \\frac { \\sigma _ { \\mathrm { w } } } { \\sqrt { C ^ { ( \\ell - 2 ) } ( n ) } } \\sum _ { k = 1 } ^ { C ^ { ( \\ell - 2 ) } ( n ) } \\sum _ { \\xi \\in \\nu \\mathrm { t h } \\mathrm { p a t e h } } \\epsilon _ { j , k , \\nu ; \\xi } ^ { ( \\ell - 1 ) } g _ { k , \\xi } ^ { ( \\ell - 2 ) } ) ,",
+                                    "type": "interline_equation",
+                                    "image_path": "17c0e7e933133d259cd80eda41b96e3b779964d857d42527541a14a92629a63e.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 3,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                111,
+                                106,
+                                514,
+                                119.33333333333333
+                            ],
+                            "spans": [],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                111,
+                                119.33333333333333,
+                                514,
+                                132.66666666666666
+                            ],
+                            "spans": [],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                111,
+                                132.66666666666666,
+                                514,
+                                146.0
+                            ],
+                            "spans": [],
+                            "index": 4
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        158,
+                        506,
+                        199
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                158,
+                                509,
+                                194
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        158,
+                                        505,
+                                        171
+                                    ],
+                                    "score": 1.0,
+                                    "content": "As such, the summands are iid conditional on the finite set of random variables",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        170,
+                                        415,
+                                        189
+                                    ],
+                                    "score": 0.71,
+                                    "content": "\\left\\{ g _ { k , \\xi } ^ { ( \\ell - 2 ) } ( x ) [ n ] : k \\in \\{ 1 , \\ldots , C ^ { ( \\ell - 2 ) } \\} , \\xi \\in \\left\\{ 1 , \\ldots , H ^ { ( \\ell - 2 ) } D ^ { ( \\ell - 2 ) } \\right\\} , x \\in \\mathcal { L } _ { \\mathcal { X } } \\right\\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        415,
+                                        163,
+                                        448,
+                                        194
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", where",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        448,
+                                        173,
+                                        464,
+                                        184
+                                    ],
+                                    "score": 0.88,
+                                    "content": "\\mathcal { L } _ { \\mathcal { X } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        464,
+                                        163,
+                                        509,
+                                        194
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is the set",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                186,
+                                189,
+                                199
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        186,
+                                        177,
+                                        199
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of input points in",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        178,
+                                        187,
+                                        185,
+                                        197
+                                    ],
+                                    "score": 0.72,
+                                    "content": "\\mathcal { L }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        186,
+                                        186,
+                                        189,
+                                        199
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        }
+                    ],
+                    "index": 5.5,
+                    "bbox_fs": [
+                        105,
+                        158,
+                        509,
+                        199
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        204,
+                        505,
+                        240
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                202,
+                                505,
+                                218
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        202,
+                                        424,
+                                        218
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The exchangeable CLT in Lemma 10 in Matthews et al. (2018b) indicates that",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        424,
+                                        204,
+                                        483,
+                                        217
+                                    ],
+                                    "score": 0.93,
+                                    "content": "{ \\mathcal { T } } ^ { ( \\ell ) } \\left( { \\mathcal { L } } , \\alpha \\right) [ n ]",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        483,
+                                        202,
+                                        505,
+                                        218
+                                    ],
+                                    "score": 1.0,
+                                    "content": "con-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                216,
+                                505,
+                                230
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        216,
+                                        207,
+                                        229
+                                    ],
+                                    "score": 1.0,
+                                    "content": "verges in distribution to",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        207,
+                                        216,
+                                        248,
+                                        230
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\mathcal { N } \\left( 0 , \\sigma _ { * } ^ { 2 } \\right)",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        248,
+                                        216,
+                                        505,
+                                        229
+                                    ],
+                                    "score": 1.0,
+                                    "content": "if the summands are exchangeable (which we showed above),",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                227,
+                                222,
+                                240
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        227,
+                                        222,
+                                        240
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and if three conditions hold,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        }
+                    ],
+                    "index": 8,
+                    "bbox_fs": [
+                        105,
+                        202,
+                        505,
+                        240
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        127,
+                        244,
+                        295,
+                        319
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                127,
+                                244,
+                                295,
+                                319
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        244,
+                                        295,
+                                        319
+                                    ],
+                                    "score": 0.89,
+                                    "content": "\\begin{array} { r l } & { \\mathrm { a ) ~ \\mathbb { E } } _ { n } [ \\gamma _ { j } ^ { ( \\ell ) } \\gamma _ { j ^ { \\prime } } ^ { ( \\ell ) } ] = 0 } \\\\ & { \\mathrm { b ) ~ \\operatorname* { l i m } _ { n  \\infty } \\mathbb { E } } _ { n } [ ( \\gamma _ { j } ^ { ( \\ell ) } ) ^ { 2 } ( \\gamma _ { j ^ { \\prime } } ^ { ( \\ell ) } ) ^ { 2 } ] = \\sigma _ { * } ^ { 4 } } \\\\ & { \\mathrm { c ) ~ \\mathbb { E } } _ { n } [ | \\gamma _ { j } ^ { ( \\ell ) } | ^ { 3 } ] = o ( \\sqrt { C ^ { ( \\ell ) } ( n ) } ) } \\end{array}",
+                                    "type": "interline_equation",
+                                    "image_path": "d2bd2b8a709e0411be37c1c704aac55b083a7ab7c52b160eaccbcf65054d7360.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 12,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                127,
+                                244,
+                                295,
+                                259.0
+                            ],
+                            "spans": [],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                259.0,
+                                295,
+                                274.0
+                            ],
+                            "spans": [],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                274.0,
+                                295,
+                                289.0
+                            ],
+                            "spans": [],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                289.0,
+                                295,
+                                304.0
+                            ],
+                            "spans": [],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                304.0,
+                                295,
+                                319.0
+                            ],
+                            "spans": [],
+                            "index": 14
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        322,
+                        505,
+                        357
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                322,
+                                505,
+                                335
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        322,
+                                        505,
+                                        335
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Condition a) follows immediately as the summands are uncorrelated and zero-mean. Conditions",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                333,
+                                505,
+                                347
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        333,
+                                        505,
+                                        347
+                                    ],
+                                    "score": 1.0,
+                                    "content": "b) and c) are more involved as convergence in distribution in the previous layers does not imply",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                345,
+                                321,
+                                357
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        345,
+                                        321,
+                                        357
+                                    ],
+                                    "score": 1.0,
+                                    "content": "convergence in moments for our activation functions.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        }
+                    ],
+                    "index": 16,
+                    "bbox_fs": [
+                        105,
+                        322,
+                        505,
+                        357
+                    ]
+                },
+                {
+                    "type": "list",
+                    "bbox": [
+                        106,
+                        361,
+                        506,
+                        447
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                362,
+                                506,
+                                374
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        362,
+                                        506,
+                                        374
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We begin by considering the extension of Lemma 20 in Matthews et al. (2018b), which allow us to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                373,
+                                505,
+                                385
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        373,
+                                        505,
+                                        385
+                                    ],
+                                    "score": 1.0,
+                                    "content": "show conditions b) and c) above, even in the case of unbounded but linearly enveloped nonlinearities",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                102,
+                                379,
+                                510,
+                                411
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        102,
+                                        379,
+                                        286,
+                                        411
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(Definition 1 in Matthews et al., 2018b). Lembounded by a finite constant independent of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        315,
+                                        379,
+                                        448,
+                                        411
+                                    ],
+                                    "score": 1.0,
+                                    "content": "states that the eighth moments of . We prove this by induction. Th",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        448,
+                                        384,
+                                        489,
+                                        400
+                                    ],
+                                    "score": 0.94,
+                                    "content": "f _ { i , \\mu } ^ { ( t ) } ( x ) [ n ]",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        489,
+                                        379,
+                                        510,
+                                        411
+                                    ],
+                                    "score": 1.0,
+                                    "content": "aree is",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                286,
+                                398,
+                                315,
+                                409
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        286,
+                                        398,
+                                        315,
+                                        409
+                                    ],
+                                    "score": 0.9,
+                                    "content": "n \\in \\mathbb N",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 21,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                102,
+                                406,
+                                508,
+                                430
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        102,
+                                        406,
+                                        146,
+                                        430
+                                    ],
+                                    "score": 1.0,
+                                    "content": "trivial, as",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        146,
+                                        409,
+                                        188,
+                                        426
+                                    ],
+                                    "score": 0.93,
+                                    "content": "f _ { j , \\mu } ^ { ( 1 ) } ( x ) [ n ]",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        189,
+                                        406,
+                                        508,
+                                        430
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is Gaussian. Following Matthews et al. (2018b), assume the condition holds up",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                423,
+                                505,
+                                437
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        423,
+                                        117,
+                                        437
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        117,
+                                        424,
+                                        138,
+                                        434
+                                    ],
+                                    "score": 0.86,
+                                    "content": "\\ell - 1",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        138,
+                                        423,
+                                        314,
+                                        437
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", and show that the condition holds for layer",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        315,
+                                        424,
+                                        320,
+                                        434
+                                    ],
+                                    "score": 0.44,
+                                    "content": "\\ell",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        320,
+                                        423,
+                                        505,
+                                        437
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Using Eq. (21), we can bound the activations",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                434,
+                                149,
+                                448
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        434,
+                                        138,
+                                        448
+                                    ],
+                                    "score": 1.0,
+                                    "content": "at layer",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        139,
+                                        435,
+                                        144,
+                                        445
+                                    ],
+                                    "score": 0.67,
+                                    "content": "\\ell",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        145,
+                                        434,
+                                        149,
+                                        448
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24,
+                            "is_list_end_line": true
+                        }
+                    ],
+                    "index": 21,
+                    "bbox_fs": [
+                        102,
+                        362,
+                        510,
+                        448
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        112,
+                        447,
+                        480,
+                        493
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                112,
+                                447,
+                                480,
+                                493
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        112,
+                                        447,
+                                        480,
+                                        493
+                                    ],
+                                    "score": 0.94,
+                                    "content": "\\mathbb { E } \\left[ | f _ { i , \\mu } ^ { ( \\ell ) } ( x ) [ n ] | ^ { 8 } \\right] \\leq 2 ^ { 8 - 1 } \\mathbb { E } \\left[ | b _ { i } ^ { ( \\ell ) } | ^ { 8 } + \\left| \\frac { \\sigma _ { \\mathbf { w } } } { \\sqrt { C ^ { ( \\ell - 1 ) } } } \\sum _ { j = 1 } ^ { C ^ { ( \\ell - 1 ) } ( n ) } \\sum _ { \\nu \\in \\mu \\mathrm { t h } \\mathrm { p a t e h } } \\epsilon _ { i , j , \\mu , \\nu } ^ { ( \\ell ) } g _ { j , \\nu } ^ { ( \\ell - 1 ) } ( x ) [ n ] \\right| ^ { 8 } \\right]",
+                                    "type": "interline_equation",
+                                    "image_path": "e1525aadc195dba6d3e8ef0ab2a26a0be8b65495eb06427ba51a80150e64f40d.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 26,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                112,
+                                447,
+                                480,
+                                462.3333333333333
+                            ],
+                            "spans": [],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                112,
+                                462.3333333333333,
+                                480,
+                                477.66666666666663
+                            ],
+                            "spans": [],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                112,
+                                477.66666666666663,
+                                480,
+                                492.99999999999994
+                            ],
+                            "spans": [],
+                            "index": 27
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        493,
+                        508,
+                        516
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                492,
+                                505,
+                                507
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        492,
+                                        505,
+                                        507
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Following Eq. 48 in Matthews et al. (2018b), which uses Lemma 19 in Matthews et al. (2018b), we",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                502,
+                                132,
+                                519
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        502,
+                                        132,
+                                        519
+                                    ],
+                                    "score": 1.0,
+                                    "content": "have,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        }
+                    ],
+                    "index": 28.5,
+                    "bbox_fs": [
+                        104,
+                        492,
+                        505,
+                        519
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        117,
+                        519,
+                        478,
+                        599
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                117,
+                                519,
+                                478,
+                                599
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        117,
+                                        519,
+                                        478,
+                                        599
+                                    ],
+                                    "score": 0.94,
+                                    "content": "\\begin{array} { r l } { \\mathbb { E } \\left[ \\left| \\frac { \\sigma _ { \\mathrm { w } } } { \\sqrt { C ^ { ( \\ell - 1 ) } } } \\sum _ { j = 1 } ^ { C ^ { ( \\ell - 1 ) } ( n ) } \\displaystyle \\sum _ { \\nu \\in \\mu \\mathrm { t h } \\mathrm { p a t e h } } \\epsilon _ { i , j , \\mu , \\nu } ^ { ( \\ell ) } g _ { j , \\nu } ^ { ( \\ell - 1 ) } ( x ) [ n ] \\right| ^ { 8 } \\right] } & { } \\\\ { = \\frac { 2 ^ { 4 } \\Gamma ( 4 + 1 / 2 ) } { \\Gamma ( 1 / 2 ) } \\mathbb { E } \\left[ \\left| \\frac { \\sigma _ { \\mathrm { w } } ^ { 2 } } { C ^ { ( \\ell - 1 ) } ( n ) } \\| g _ { j \\in \\{ 1 , \\dots , C ^ { ( \\ell - 1 ) } ( n ) \\} , \\nu \\in \\mu \\mathrm { t h } \\mathrm { p a t c h } } ^ { ( \\ell - 1 ) } ( x ) [ n ] \\| _ { 2 } ^ { 2 } \\right| ^ { 4 } \\right] . } \\end{array}",
+                                    "type": "interline_equation",
+                                    "image_path": "d5c365ee7ab68b781b018199bf3460179ad3b31af3693fa570d752d714ff7103.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 31,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                117,
+                                519,
+                                478,
+                                545.6666666666666
+                            ],
+                            "spans": [],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                117,
+                                545.6666666666666,
+                                478,
+                                572.3333333333333
+                            ],
+                            "spans": [],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                117,
+                                572.3333333333333,
+                                478,
+                                598.9999999999999
+                            ],
+                            "spans": [],
+                            "index": 32
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        602,
+                        507,
+                        631
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                599,
+                                507,
+                                620
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        601,
+                                        135,
+                                        615
+                                    ],
+                                    "score": 1.0,
+                                    "content": "wher",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        130,
+                                        599,
+                                        507,
+                                        620
+                                    ],
+                                    "score": 1.0,
+                                    "content": "e g(`−1)j∈{1,...,C(`−1)(n)},ν∈µth patch(x)[n] is the set of post-nonlinearities corresponding to j ∈",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                107,
+                                615,
+                                464,
+                                632
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        618,
+                                        186,
+                                        631
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\{ 1 , \\ldots , C ^ { ( \\ell - 1 ) } ( n ) \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        187,
+                                        615,
+                                        204,
+                                        632
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        205,
+                                        619,
+                                        237,
+                                        630
+                                    ],
+                                    "score": 0.88,
+                                    "content": "\\nu \\in \\mu \\mathrm { t h }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        238,
+                                        615,
+                                        464,
+                                        632
+                                    ],
+                                    "score": 1.0,
+                                    "content": "patch. Following Matthews et al. (2018b), observe that,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        }
+                    ],
+                    "index": 33.5,
+                    "bbox_fs": [
+                        106,
+                        599,
+                        507,
+                        632
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        111,
+                        632,
+                        513,
+                        723
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                111,
+                                632,
+                                513,
+                                723
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        111,
+                                        632,
+                                        513,
+                                        723
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\begin{array} { r } { \\frac { 1 } { \\sum ^ { ( \\ell - 1 ) } \\left( n \\right) } \\| g _ { j \\in \\{ 1 , \\dots , C ^ { ( \\ell - 1 ) } ( n ) \\} , \\nu \\in \\mu \\mathrm { t h p a t e h } } ^ { ( \\ell - 1 ) } ( x ) [ n ] \\| _ { 2 } ^ { 2 } = \\frac { 1 } { C ^ { ( \\ell - 1 ) } ( n ) } \\displaystyle \\sum _ { j = 1 } ^ { C ^ { ( \\ell - 1 ) } ( n ) } \\sum _ { \\nu \\in \\mu \\mathrm { t h p a t e h } } \\left( g _ { j , \\nu } ^ { ( \\ell - 1 ) } ( x ) [ n ] \\right) ^ { 2 } } \\\\ { \\leq \\frac { 1 } { C ^ { ( \\ell - 1 ) } ( n ) } \\displaystyle \\sum _ { j = 1 } ^ { C ^ { ( \\ell - 1 ) } ( n ) } \\sum _ { \\nu \\in \\mu \\mathrm { t h p a t e h } } ^ { C ^ { ( \\ell - 1 ) } ( n ) } \\left( c + m \\big | f _ { j , \\nu } ^ { ( \\ell - 1 ) } ( x ) \\big | \\right. } \\end{array}",
+                                    "type": "interline_equation",
+                                    "image_path": "acefb7cbee5fe7c46f897dadaf5a1e645b463a9143b940c5edeeb1ace37f3784.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 36,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                111,
+                                632,
+                                513,
+                                662.3333333333334
+                            ],
+                            "spans": [],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                111,
+                                662.3333333333334,
+                                513,
+                                692.6666666666667
+                            ],
+                            "spans": [],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                111,
+                                692.6666666666667,
+                                513,
+                                723.0000000000001
+                            ],
+                            "spans": [],
+                            "index": 37
+                        }
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        104,
+                        81,
+                        505,
+                        106
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                82,
+                                505,
+                                95
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        82,
+                                        233,
+                                        95
+                                    ],
+                                    "score": 1.0,
+                                    "content": "by the linear envelope property,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        233,
+                                        82,
+                                        306,
+                                        95
+                                    ],
+                                    "score": 0.94,
+                                    "content": "| \\phi ( u ) | \\leq c + m | u |",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        306,
+                                        82,
+                                        505,
+                                        95
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Following Matthews et al. (2018b), we substitute",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                92,
+                                477,
+                                107
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        92,
+                                        441,
+                                        107
+                                    ],
+                                    "score": 1.0,
+                                    "content": "this bound back into Eq. (28) and suppress a multiplicative constant independent of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        441,
+                                        96,
+                                        448,
+                                        104
+                                    ],
+                                    "score": 0.8,
+                                    "content": "x",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        448,
+                                        92,
+                                        466,
+                                        107
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        466,
+                                        96,
+                                        473,
+                                        104
+                                    ],
+                                    "score": 0.75,
+                                    "content": "n",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        473,
+                                        92,
+                                        477,
+                                        107
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        }
+                    ],
+                    "index": 0.5
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        115,
+                        113,
+                        478,
+                        205
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                115,
+                                113,
+                                478,
+                                205
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        113,
+                                        478,
+                                        205
+                                    ],
+                                    "score": 0.94,
+                                    "content": "\\begin{array} { r l } & { \\mathbb { E } \\left[ \\left| \\frac { \\sigma _ { \\mathbf { w } } } { \\sqrt { C ^ { ( \\ell - 1 ) } ( n ) } } \\sum _ { j = 1 } ^ { C ^ { ( \\ell - 1 ) } ( n ) } \\sum _ { \\nu \\in \\mu \\mathrm { t h p a t c h } } \\epsilon _ { i , j , \\mu , \\nu } ^ { ( \\ell ) } g _ { j , \\nu } ^ { ( \\ell - 1 ) } ( x ) [ n ] \\right| ^ { 8 } \\right] } \\\\ & { \\leq \\frac { 1 } { \\left( C ^ { ( \\ell - 1 ) } ( n ) \\right) ^ { 4 } } \\mathbb { E } \\left[ \\left| \\sum _ { j = 1 } ^ { C ^ { ( \\ell - 1 ) } ( n ) } \\sum _ { \\nu \\in \\mu \\mathrm { t h p a c h } } c ^ { 2 } + 2 c m | f _ { j , \\mu } ^ { ( \\ell - 1 ) } ( x ) [ n ] | + m ^ { 2 } | f _ { j , \\mu } ^ { ( \\ell - 1 ) } ( x ) [ n ] | ^ { 2 } \\right| ^ { 4 } \\right] } \\end{array}",
+                                    "type": "interline_equation",
+                                    "image_path": "763d53037e2ed6fdbd5080681e93bf38d156c39f3e7ef9b04ba81650c480c4d2.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 3,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                115,
+                                113,
+                                478,
+                                143.66666666666666
+                            ],
+                            "spans": [],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                143.66666666666666,
+                                478,
+                                174.33333333333331
+                            ],
+                            "spans": [],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                174.33333333333331,
+                                478,
+                                204.99999999999997
+                            ],
+                            "spans": [],
+                            "index": 4
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        208,
+                        429,
+                        221
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                207,
+                                429,
+                                222
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        207,
+                                        429,
+                                        222
+                                    ],
+                                    "score": 1.0,
+                                    "content": "This can be multiplied out, yielding a weighted sum of expectations of the form,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        }
+                    ],
+                    "index": 5
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        164,
+                        225,
+                        446,
+                        248
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                164,
+                                225,
+                                446,
+                                248
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        164,
+                                        225,
+                                        446,
+                                        248
+                                    ],
+                                    "score": 0.89,
+                                    "content": "\\mathbb { E } \\left[ | f _ { k , \\nu } ^ { ( \\ell - 1 ) } ( x ) [ n ] | ^ { p _ { 1 } } | f _ { l , \\xi } ^ { ( \\ell - 1 ) } ( x ) [ n ] | ^ { p _ { 2 } } | f _ { r , \\pi } ^ { ( \\ell - 1 ) } ( x ) [ n ] | ^ { p _ { 3 } } | f _ { q , \\rho } ^ { ( \\ell - 1 ) } ( x ) [ n ] | ^ { p _ { 4 } } \\right]",
+                                    "type": "interline_equation",
+                                    "image_path": "c53f65198268464e7dfdb89ac275223b8f600644dd163b731320170af9d4f07e.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 6,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                164,
+                                225,
+                                446,
+                                248
+                            ],
+                            "spans": [],
+                            "index": 6
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        252,
+                        506,
+                        385
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                251,
+                                506,
+                                267
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        251,
+                                        127,
+                                        267
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        127,
+                                        254,
+                                        183,
+                                        266
+                                    ],
+                                    "score": 0.93,
+                                    "content": "p _ { i } \\in \\{ 0 , 1 , 2 \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        183,
+                                        251,
+                                        199,
+                                        267
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        200,
+                                        254,
+                                        251,
+                                        265
+                                    ],
+                                    "score": 0.9,
+                                    "content": "i = { 1 , 2 , 3 , 4 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        252,
+                                        251,
+                                        272,
+                                        267
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        272,
+                                        252,
+                                        396,
+                                        266
+                                    ],
+                                    "score": 0.91,
+                                    "content": "k , l , r , q \\in \\{ 1 , \\dots , C ^ { ( \\ell - 1 ) } ( n ) \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        397,
+                                        251,
+                                        417,
+                                        267
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        417,
+                                        254,
+                                        480,
+                                        266
+                                    ],
+                                    "score": 0.88,
+                                    "content": "\\nu , \\xi , \\pi , \\rho \\in \\mu \\mathrm { t h }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        480,
+                                        251,
+                                        506,
+                                        267
+                                    ],
+                                    "score": 1.0,
+                                    "content": "patch",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                100,
+                                264,
+                                511,
+                                304
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        100,
+                                        264,
+                                        332,
+                                        304
+                                    ],
+                                    "score": 1.0,
+                                    "content": "where the weights of these terms are independent of (2018b), each of these terms is bounded if the eighth mo",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        333,
+                                        267,
+                                        339,
+                                        275
+                                    ],
+                                    "score": 0.76,
+                                    "content": "n",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        340,
+                                        264,
+                                        370,
+                                        304
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Usinents of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        423,
+                                        264,
+                                        511,
+                                        304
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in Matthews et al.are bounded, which",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                371,
+                                276,
+                                423,
+                                292
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        371,
+                                        276,
+                                        423,
+                                        292
+                                    ],
+                                    "score": 0.94,
+                                    "content": "f _ { k , \\mu } ^ { ( \\ell - 1 ) } ( x ) [ n ]",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                107,
+                                301,
+                                505,
+                                317
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        301,
+                                        211,
+                                        317
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\left( 2 C ^ { ( \\ell - 1 ) } ( n ) | \\mu \\mathrm { t h } \\mathrm { p a t c h } | \\right) ^ { 4 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        211,
+                                        302,
+                                        244,
+                                        317
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", where",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        244,
+                                        304,
+                                        289,
+                                        317
+                                    ],
+                                    "score": 0.75,
+                                    "content": "| \\mu \\mathrm { t h } \\ p a \\mathrm { t c h } |",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        289,
+                                        302,
+                                        505,
+                                        317
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is the number of elements in a convolutional patch.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                317,
+                                506,
+                                333
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        318,
+                                        284,
+                                        333
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Thus, we can use the same constant for any",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        291,
+                                        318,
+                                        335,
+                                        333
+                                    ],
+                                    "score": 1.0,
+                                    "content": "due to the",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        335,
+                                        317,
+                                        403,
+                                        333
+                                    ],
+                                    "score": 0.92,
+                                    "content": "1 / \\left( C ^ { ( \\ell - 1 ) } ( n ) \\right) ^ { 4 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        403,
+                                        318,
+                                        506,
+                                        333
+                                    ],
+                                    "score": 1.0,
+                                    "content": "scaling. As in Matthews",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                324,
+                                511,
+                                356
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        324,
+                                        209,
+                                        356
+                                    ],
+                                    "score": 1.0,
+                                    "content": "et al. (2018b), noting that",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        209,
+                                        331,
+                                        261,
+                                        349
+                                    ],
+                                    "score": 0.94,
+                                    "content": "f _ { j , \\mu } ^ { ( \\ell - 1 ) } ( x ) [ n ]",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        262,
+                                        324,
+                                        352,
+                                        356
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are exchangeable over",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        352,
+                                        336,
+                                        358,
+                                        347
+                                    ],
+                                    "score": 0.84,
+                                    "content": "j",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        358,
+                                        324,
+                                        389,
+                                        356
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for any",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        390,
+                                        337,
+                                        396,
+                                        345
+                                    ],
+                                    "score": 0.76,
+                                    "content": "x",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        397,
+                                        324,
+                                        414,
+                                        356
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        414,
+                                        337,
+                                        421,
+                                        345
+                                    ],
+                                    "score": 0.75,
+                                    "content": "n",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        421,
+                                        324,
+                                        511,
+                                        356
+                                    ],
+                                    "score": 1.0,
+                                    "content": "concludes the proof.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                352,
+                                505,
+                                364
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        352,
+                                        505,
+                                        364
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Using this result, we can obtain a straightforward adaptation of Lemmas 15, 16 and 21 in Matthews",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                362,
+                                505,
+                                375
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        362,
+                                        505,
+                                        375
+                                    ],
+                                    "score": 1.0,
+                                    "content": "et al. (2018b). Lemma 15 gives condition b), Lemma 16 gives condition c); Lemma 15 requires",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                374,
+                                338,
+                                386
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        374,
+                                        338,
+                                        386
+                                    ],
+                                    "score": 1.0,
+                                    "content": "uniform integrability, which is established by Lemma 21.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        }
+                    ],
+                    "index": 11
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        106,
+                        398,
+                        356,
+                        410
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                398,
+                                357,
+                                411
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        398,
+                                        357,
+                                        411
+                                    ],
+                                    "score": 1.0,
+                                    "content": "7.3 CALIBRATION OF GAUSSIAN PROCESS UNCERTAINTY",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        }
+                    ],
+                    "index": 16
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        419,
+                        505,
+                        519
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                420,
+                                506,
+                                431
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        420,
+                                        506,
+                                        431
+                                    ],
+                                    "score": 1.0,
+                                    "content": "It is important to check that the estimates of uncertainty produced by our Gaussian process are",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                430,
+                                505,
+                                442
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        430,
+                                        505,
+                                        442
+                                    ],
+                                    "score": 1.0,
+                                    "content": "reasonable. However, to make this assessment, we needed to use a proper likelihood, and not the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                441,
+                                506,
+                                453
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        441,
+                                        506,
+                                        453
+                                    ],
+                                    "score": 1.0,
+                                    "content": "squared-error loss in the main text. We therefore used our kernel to perform the full, multi-class",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                452,
+                                504,
+                                464
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        452,
+                                        504,
+                                        464
+                                    ],
+                                    "score": 1.0,
+                                    "content": "classification problem in GPflow Matthews et al. (2017), with a RobustMax likelihood (Hernández-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                462,
+                                505,
+                                476
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        462,
+                                        505,
+                                        476
+                                    ],
+                                    "score": 1.0,
+                                    "content": "lobato et al., 2011). The more difficult non-conjugate inference problem forced us to use 1000",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                474,
+                                504,
+                                486
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        474,
+                                        504,
+                                        486
+                                    ],
+                                    "score": 1.0,
+                                    "content": "inducing points, randomly chosen from the training inputs. Both our kernel and an RBF kernel have",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                484,
+                                505,
+                                498
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        484,
+                                        505,
+                                        498
+                                    ],
+                                    "score": 1.0,
+                                    "content": "similar calibration curves, that closely track the diagonal, indicating accurate uncertainty estima-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                496,
+                                505,
+                                509
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        496,
+                                        505,
+                                        509
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tion. However, even in the inducing point setting, our convolutional kernel gave considerably better",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                507,
+                                352,
+                                519
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        507,
+                                        245,
+                                        519
+                                    ],
+                                    "score": 1.0,
+                                    "content": "performance than the RBF kernel",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        246,
+                                        507,
+                                        268,
+                                        518
+                                    ],
+                                    "score": 0.85,
+                                    "content": "2 . 4 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        268,
+                                        507,
+                                        302,
+                                        519
+                                    ],
+                                    "score": 1.0,
+                                    "content": "error vs",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        302,
+                                        507,
+                                        324,
+                                        518
+                                    ],
+                                    "score": 0.85,
+                                    "content": "3 . 4 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        325,
+                                        507,
+                                        352,
+                                        519
+                                    ],
+                                    "score": 1.0,
+                                    "content": "error).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        }
+                    ],
+                    "index": 21
+                }
+            ],
+            "page_idx": 14,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        107,
+                        26,
+                        293,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                25,
+                                294,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        25,
+                                        294,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Published as a conference paper at ICLR 2019",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        300,
+                        751,
+                        311,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                299,
+                                750,
+                                313,
+                                764
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        299,
+                                        750,
+                                        313,
+                                        764
+                                    ],
+                                    "score": 1.0,
+                                    "content": "15",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        104,
+                        81,
+                        505,
+                        106
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                82,
+                                505,
+                                95
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        82,
+                                        233,
+                                        95
+                                    ],
+                                    "score": 1.0,
+                                    "content": "by the linear envelope property,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        233,
+                                        82,
+                                        306,
+                                        95
+                                    ],
+                                    "score": 0.94,
+                                    "content": "| \\phi ( u ) | \\leq c + m | u |",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        306,
+                                        82,
+                                        505,
+                                        95
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Following Matthews et al. (2018b), we substitute",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                92,
+                                477,
+                                107
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        92,
+                                        441,
+                                        107
+                                    ],
+                                    "score": 1.0,
+                                    "content": "this bound back into Eq. (28) and suppress a multiplicative constant independent of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        441,
+                                        96,
+                                        448,
+                                        104
+                                    ],
+                                    "score": 0.8,
+                                    "content": "x",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        448,
+                                        92,
+                                        466,
+                                        107
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        466,
+                                        96,
+                                        473,
+                                        104
+                                    ],
+                                    "score": 0.75,
+                                    "content": "n",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        473,
+                                        92,
+                                        477,
+                                        107
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        }
+                    ],
+                    "index": 0.5,
+                    "bbox_fs": [
+                        105,
+                        82,
+                        505,
+                        107
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        115,
+                        113,
+                        478,
+                        205
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                115,
+                                113,
+                                478,
+                                205
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        113,
+                                        478,
+                                        205
+                                    ],
+                                    "score": 0.94,
+                                    "content": "\\begin{array} { r l } & { \\mathbb { E } \\left[ \\left| \\frac { \\sigma _ { \\mathbf { w } } } { \\sqrt { C ^ { ( \\ell - 1 ) } ( n ) } } \\sum _ { j = 1 } ^ { C ^ { ( \\ell - 1 ) } ( n ) } \\sum _ { \\nu \\in \\mu \\mathrm { t h p a t c h } } \\epsilon _ { i , j , \\mu , \\nu } ^ { ( \\ell ) } g _ { j , \\nu } ^ { ( \\ell - 1 ) } ( x ) [ n ] \\right| ^ { 8 } \\right] } \\\\ & { \\leq \\frac { 1 } { \\left( C ^ { ( \\ell - 1 ) } ( n ) \\right) ^ { 4 } } \\mathbb { E } \\left[ \\left| \\sum _ { j = 1 } ^ { C ^ { ( \\ell - 1 ) } ( n ) } \\sum _ { \\nu \\in \\mu \\mathrm { t h p a c h } } c ^ { 2 } + 2 c m | f _ { j , \\mu } ^ { ( \\ell - 1 ) } ( x ) [ n ] | + m ^ { 2 } | f _ { j , \\mu } ^ { ( \\ell - 1 ) } ( x ) [ n ] | ^ { 2 } \\right| ^ { 4 } \\right] } \\end{array}",
+                                    "type": "interline_equation",
+                                    "image_path": "763d53037e2ed6fdbd5080681e93bf38d156c39f3e7ef9b04ba81650c480c4d2.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 3,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                115,
+                                113,
+                                478,
+                                143.66666666666666
+                            ],
+                            "spans": [],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                143.66666666666666,
+                                478,
+                                174.33333333333331
+                            ],
+                            "spans": [],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                174.33333333333331,
+                                478,
+                                204.99999999999997
+                            ],
+                            "spans": [],
+                            "index": 4
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        208,
+                        429,
+                        221
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                207,
+                                429,
+                                222
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        207,
+                                        429,
+                                        222
+                                    ],
+                                    "score": 1.0,
+                                    "content": "This can be multiplied out, yielding a weighted sum of expectations of the form,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        }
+                    ],
+                    "index": 5,
+                    "bbox_fs": [
+                        106,
+                        207,
+                        429,
+                        222
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        164,
+                        225,
+                        446,
+                        248
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                164,
+                                225,
+                                446,
+                                248
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        164,
+                                        225,
+                                        446,
+                                        248
+                                    ],
+                                    "score": 0.89,
+                                    "content": "\\mathbb { E } \\left[ | f _ { k , \\nu } ^ { ( \\ell - 1 ) } ( x ) [ n ] | ^ { p _ { 1 } } | f _ { l , \\xi } ^ { ( \\ell - 1 ) } ( x ) [ n ] | ^ { p _ { 2 } } | f _ { r , \\pi } ^ { ( \\ell - 1 ) } ( x ) [ n ] | ^ { p _ { 3 } } | f _ { q , \\rho } ^ { ( \\ell - 1 ) } ( x ) [ n ] | ^ { p _ { 4 } } \\right]",
+                                    "type": "interline_equation",
+                                    "image_path": "c53f65198268464e7dfdb89ac275223b8f600644dd163b731320170af9d4f07e.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 6,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                164,
+                                225,
+                                446,
+                                248
+                            ],
+                            "spans": [],
+                            "index": 6
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        252,
+                        506,
+                        385
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                251,
+                                506,
+                                267
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        251,
+                                        127,
+                                        267
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        127,
+                                        254,
+                                        183,
+                                        266
+                                    ],
+                                    "score": 0.93,
+                                    "content": "p _ { i } \\in \\{ 0 , 1 , 2 \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        183,
+                                        251,
+                                        199,
+                                        267
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        200,
+                                        254,
+                                        251,
+                                        265
+                                    ],
+                                    "score": 0.9,
+                                    "content": "i = { 1 , 2 , 3 , 4 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        252,
+                                        251,
+                                        272,
+                                        267
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        272,
+                                        252,
+                                        396,
+                                        266
+                                    ],
+                                    "score": 0.91,
+                                    "content": "k , l , r , q \\in \\{ 1 , \\dots , C ^ { ( \\ell - 1 ) } ( n ) \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        397,
+                                        251,
+                                        417,
+                                        267
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        417,
+                                        254,
+                                        480,
+                                        266
+                                    ],
+                                    "score": 0.88,
+                                    "content": "\\nu , \\xi , \\pi , \\rho \\in \\mu \\mathrm { t h }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        480,
+                                        251,
+                                        506,
+                                        267
+                                    ],
+                                    "score": 1.0,
+                                    "content": "patch",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                100,
+                                264,
+                                511,
+                                304
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        100,
+                                        264,
+                                        332,
+                                        304
+                                    ],
+                                    "score": 1.0,
+                                    "content": "where the weights of these terms are independent of (2018b), each of these terms is bounded if the eighth mo",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        333,
+                                        267,
+                                        339,
+                                        275
+                                    ],
+                                    "score": 0.76,
+                                    "content": "n",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        340,
+                                        264,
+                                        370,
+                                        304
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Usinents of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        423,
+                                        264,
+                                        511,
+                                        304
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in Matthews et al.are bounded, which",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                371,
+                                276,
+                                423,
+                                292
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        371,
+                                        276,
+                                        423,
+                                        292
+                                    ],
+                                    "score": 0.94,
+                                    "content": "f _ { k , \\mu } ^ { ( \\ell - 1 ) } ( x ) [ n ]",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                107,
+                                301,
+                                505,
+                                317
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        301,
+                                        211,
+                                        317
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\left( 2 C ^ { ( \\ell - 1 ) } ( n ) | \\mu \\mathrm { t h } \\mathrm { p a t c h } | \\right) ^ { 4 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        211,
+                                        302,
+                                        244,
+                                        317
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", where",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        244,
+                                        304,
+                                        289,
+                                        317
+                                    ],
+                                    "score": 0.75,
+                                    "content": "| \\mu \\mathrm { t h } \\ p a \\mathrm { t c h } |",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        289,
+                                        302,
+                                        505,
+                                        317
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is the number of elements in a convolutional patch.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                317,
+                                506,
+                                333
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        318,
+                                        284,
+                                        333
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Thus, we can use the same constant for any",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        291,
+                                        318,
+                                        335,
+                                        333
+                                    ],
+                                    "score": 1.0,
+                                    "content": "due to the",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        335,
+                                        317,
+                                        403,
+                                        333
+                                    ],
+                                    "score": 0.92,
+                                    "content": "1 / \\left( C ^ { ( \\ell - 1 ) } ( n ) \\right) ^ { 4 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        403,
+                                        318,
+                                        506,
+                                        333
+                                    ],
+                                    "score": 1.0,
+                                    "content": "scaling. As in Matthews",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                324,
+                                511,
+                                356
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        324,
+                                        209,
+                                        356
+                                    ],
+                                    "score": 1.0,
+                                    "content": "et al. (2018b), noting that",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        209,
+                                        331,
+                                        261,
+                                        349
+                                    ],
+                                    "score": 0.94,
+                                    "content": "f _ { j , \\mu } ^ { ( \\ell - 1 ) } ( x ) [ n ]",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        262,
+                                        324,
+                                        352,
+                                        356
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are exchangeable over",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        352,
+                                        336,
+                                        358,
+                                        347
+                                    ],
+                                    "score": 0.84,
+                                    "content": "j",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        358,
+                                        324,
+                                        389,
+                                        356
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for any",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        390,
+                                        337,
+                                        396,
+                                        345
+                                    ],
+                                    "score": 0.76,
+                                    "content": "x",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        397,
+                                        324,
+                                        414,
+                                        356
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        414,
+                                        337,
+                                        421,
+                                        345
+                                    ],
+                                    "score": 0.75,
+                                    "content": "n",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        421,
+                                        324,
+                                        511,
+                                        356
+                                    ],
+                                    "score": 1.0,
+                                    "content": "concludes the proof.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                352,
+                                505,
+                                364
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        352,
+                                        505,
+                                        364
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Using this result, we can obtain a straightforward adaptation of Lemmas 15, 16 and 21 in Matthews",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                362,
+                                505,
+                                375
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        362,
+                                        505,
+                                        375
+                                    ],
+                                    "score": 1.0,
+                                    "content": "et al. (2018b). Lemma 15 gives condition b), Lemma 16 gives condition c); Lemma 15 requires",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                374,
+                                338,
+                                386
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        374,
+                                        338,
+                                        386
+                                    ],
+                                    "score": 1.0,
+                                    "content": "uniform integrability, which is established by Lemma 21.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        }
+                    ],
+                    "index": 11,
+                    "bbox_fs": [
+                        100,
+                        251,
+                        511,
+                        386
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        106,
+                        398,
+                        356,
+                        410
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                398,
+                                357,
+                                411
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        398,
+                                        357,
+                                        411
+                                    ],
+                                    "score": 1.0,
+                                    "content": "7.3 CALIBRATION OF GAUSSIAN PROCESS UNCERTAINTY",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        }
+                    ],
+                    "index": 16
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        419,
+                        505,
+                        519
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                420,
+                                506,
+                                431
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        420,
+                                        506,
+                                        431
+                                    ],
+                                    "score": 1.0,
+                                    "content": "It is important to check that the estimates of uncertainty produced by our Gaussian process are",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                430,
+                                505,
+                                442
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        430,
+                                        505,
+                                        442
+                                    ],
+                                    "score": 1.0,
+                                    "content": "reasonable. However, to make this assessment, we needed to use a proper likelihood, and not the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                441,
+                                506,
+                                453
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        441,
+                                        506,
+                                        453
+                                    ],
+                                    "score": 1.0,
+                                    "content": "squared-error loss in the main text. We therefore used our kernel to perform the full, multi-class",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                452,
+                                504,
+                                464
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        452,
+                                        504,
+                                        464
+                                    ],
+                                    "score": 1.0,
+                                    "content": "classification problem in GPflow Matthews et al. (2017), with a RobustMax likelihood (Hernández-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                462,
+                                505,
+                                476
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        462,
+                                        505,
+                                        476
+                                    ],
+                                    "score": 1.0,
+                                    "content": "lobato et al., 2011). The more difficult non-conjugate inference problem forced us to use 1000",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                474,
+                                504,
+                                486
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        474,
+                                        504,
+                                        486
+                                    ],
+                                    "score": 1.0,
+                                    "content": "inducing points, randomly chosen from the training inputs. Both our kernel and an RBF kernel have",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                484,
+                                505,
+                                498
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        484,
+                                        505,
+                                        498
+                                    ],
+                                    "score": 1.0,
+                                    "content": "similar calibration curves, that closely track the diagonal, indicating accurate uncertainty estima-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                496,
+                                505,
+                                509
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        496,
+                                        505,
+                                        509
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tion. However, even in the inducing point setting, our convolutional kernel gave considerably better",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                507,
+                                352,
+                                519
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        507,
+                                        245,
+                                        519
+                                    ],
+                                    "score": 1.0,
+                                    "content": "performance than the RBF kernel",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        246,
+                                        507,
+                                        268,
+                                        518
+                                    ],
+                                    "score": 0.85,
+                                    "content": "2 . 4 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        268,
+                                        507,
+                                        302,
+                                        519
+                                    ],
+                                    "score": 1.0,
+                                    "content": "error vs",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        302,
+                                        507,
+                                        324,
+                                        518
+                                    ],
+                                    "score": 0.85,
+                                    "content": "3 . 4 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        325,
+                                        507,
+                                        352,
+                                        519
+                                    ],
+                                    "score": 1.0,
+                                    "content": "error).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        }
+                    ],
+                    "index": 21,
+                    "bbox_fs": [
+                        105,
+                        420,
+                        506,
+                        519
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "image",
+                    "bbox": [
+                        116,
+                        293,
+                        494,
+                        472
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                116,
+                                293,
+                                494,
+                                472
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        293,
+                                        494,
+                                        472
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                116,
+                                                293,
+                                                494,
+                                                472
+                                            ],
+                                            "score": 0.967,
+                                            "type": "image",
+                                            "image_path": "8864ab9d76c33870f50470fe5185392a7878cda9c47764b2ab0ca5045ac895b3.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 1,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        293,
+                                        494,
+                                        352.6666666666667
+                                    ],
+                                    "spans": [],
+                                    "index": 0
+                                },
+                                {
+                                    "bbox": [
+                                        116,
+                                        352.6666666666667,
+                                        494,
+                                        412.33333333333337
+                                    ],
+                                    "spans": [],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        116,
+                                        412.33333333333337,
+                                        494,
+                                        472.00000000000006
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                106,
+                                493,
+                                506,
+                                528
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        493,
+                                        506,
+                                        507
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                493,
+                                                455,
+                                                507
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 3: Calibration plots for an RBF kernel (left) and the ResNet kernel (right). The",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                455,
+                                                496,
+                                                461,
+                                                504
+                                            ],
+                                            "score": 0.26,
+                                            "content": "\\mathbf { X }",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                462,
+                                                493,
+                                                506,
+                                                507
+                                            ],
+                                            "score": 1.0,
+                                            "content": "-axis gives",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 3
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        504,
+                                        505,
+                                        517
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                504,
+                                                505,
+                                                517
+                                            ],
+                                            "score": 1.0,
+                                            "content": "GP prediction for the label probability. The points give corresponding proportion of test points with",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 4
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        515,
+                                        410,
+                                        529
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                515,
+                                                410,
+                                                529
+                                            ],
+                                            "score": 1.0,
+                                            "content": "that label, and the bars give the proportion of training examples in each bin.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 5
+                                }
+                            ],
+                            "index": 4
+                        }
+                    ],
+                    "index": 2.5
+                }
+            ],
+            "page_idx": 15,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        107,
+                        27,
+                        293,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                26,
+                                294,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        26,
+                                        294,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Published as a conference paper at ICLR 2019",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        300,
+                        751,
+                        311,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                299,
+                                750,
+                                313,
+                                764
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        299,
+                                        750,
+                                        313,
+                                        764
+                                    ],
+                                    "score": 1.0,
+                                    "content": "16",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "image",
+                    "bbox": [
+                        116,
+                        293,
+                        494,
+                        472
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                116,
+                                293,
+                                494,
+                                472
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        293,
+                                        494,
+                                        472
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                116,
+                                                293,
+                                                494,
+                                                472
+                                            ],
+                                            "score": 0.967,
+                                            "type": "image",
+                                            "image_path": "8864ab9d76c33870f50470fe5185392a7878cda9c47764b2ab0ca5045ac895b3.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 1,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        293,
+                                        494,
+                                        352.6666666666667
+                                    ],
+                                    "spans": [],
+                                    "index": 0
+                                },
+                                {
+                                    "bbox": [
+                                        116,
+                                        352.6666666666667,
+                                        494,
+                                        412.33333333333337
+                                    ],
+                                    "spans": [],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        116,
+                                        412.33333333333337,
+                                        494,
+                                        472.00000000000006
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                106,
+                                493,
+                                506,
+                                528
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        493,
+                                        506,
+                                        507
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                493,
+                                                455,
+                                                507
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 3: Calibration plots for an RBF kernel (left) and the ResNet kernel (right). The",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                455,
+                                                496,
+                                                461,
+                                                504
+                                            ],
+                                            "score": 0.26,
+                                            "content": "\\mathbf { X }",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                462,
+                                                493,
+                                                506,
+                                                507
+                                            ],
+                                            "score": 1.0,
+                                            "content": "-axis gives",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 3
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        504,
+                                        505,
+                                        517
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                504,
+                                                505,
+                                                517
+                                            ],
+                                            "score": 1.0,
+                                            "content": "GP prediction for the label probability. The points give corresponding proportion of test points with",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 4
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        515,
+                                        410,
+                                        529
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                515,
+                                                410,
+                                                529
+                                            ],
+                                            "score": 1.0,
+                                            "content": "that label, and the bars give the proportion of training examples in each bin.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 5
+                                }
+                            ],
+                            "index": 4
+                        }
+                    ],
+                    "index": 2.5
+                }
+            ]
+        }
+    ],
+    "_backend": "pipeline",
+    "_version_name": "2.2.2"
+}
\ No newline at end of file
diff --git a/parse/train/Bklfsi0cKm/Bklfsi0cKm_model.json b/parse/train/Bklfsi0cKm/Bklfsi0cKm_model.json
new file mode 100644
index 0000000000000000000000000000000000000000..38e00fb8798cb727a58ba3c912ad00b317328184
--- /dev/null
+++ b/parse/train/Bklfsi0cKm/Bklfsi0cKm_model.json
@@ -0,0 +1,23591 @@
+[
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    398,
+                    819,
+                    1302,
+                    819,
+                    1302,
+                    1126,
+                    398,
+                    1126
+                ],
+                "score": 0.982
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1259,
+                    1404,
+                    1259,
+                    1404,
+                    1625,
+                    298,
+                    1625
+                ],
+                "score": 0.98
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1641,
+                    1404,
+                    1641,
+                    1404,
+                    1977,
+                    298,
+                    1977
+                ],
+                "score": 0.975
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    315,
+                    376,
+                    616,
+                    376,
+                    616,
+                    498,
+                    315,
+                    498
+                ],
+                "score": 0.965
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    297,
+                    220,
+                    1105,
+                    220,
+                    1105,
+                    324,
+                    297,
+                    324
+                ],
+                "score": 0.962
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    314,
+                    546,
+                    784,
+                    546,
+                    784,
+                    668,
+                    314,
+                    668
+                ],
+                "score": 0.921
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    328,
+                    2005,
+                    1281,
+                    2005,
+                    1281,
+                    2034,
+                    328,
+                    2034
+                ],
+                "score": 0.903
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    302,
+                    1188,
+                    573,
+                    1188,
+                    573,
+                    1223,
+                    302,
+                    1223
+                ],
+                "score": 0.894
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    299,
+                    75,
+                    816,
+                    75,
+                    816,
+                    104,
+                    299,
+                    104
+                ],
+                "score": 0.87
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    773,
+                    750,
+                    927,
+                    750,
+                    927,
+                    783,
+                    773,
+                    783
+                ],
+                "score": 0.861
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    1077,
+                    376,
+                    1385,
+                    376,
+                    1385,
+                    498,
+                    1077,
+                    498
+                ],
+                "score": 0.726
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    841,
+                    2088,
+                    857,
+                    2088,
+                    857,
+                    2112,
+                    841,
+                    2112
+                ],
+                "score": 0.707
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    1077,
+                    376,
+                    1385,
+                    376,
+                    1385,
+                    498,
+                    1077,
+                    498
+                ],
+                "score": 0.322
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    315,
+                    547,
+                    550,
+                    547,
+                    550,
+                    576,
+                    315,
+                    576
+                ],
+                "score": 0.213
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    975,
+                    1065,
+                    1051,
+                    1065,
+                    1051,
+                    1094,
+                    975,
+                    1094
+                ],
+                "score": 0.87,
+                "latex": "0 . 8 4 \\%"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    216.0,
+                    1108.0,
+                    216.0,
+                    1108.0,
+                    272.0,
+                    295.0,
+                    272.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    275.0,
+                    984.0,
+                    275.0,
+                    984.0,
+                    326.0,
+                    294.0,
+                    326.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    332.0,
+                    1998.0,
+                    1287.0,
+                    1998.0,
+                    1287.0,
+                    2042.0,
+                    332.0,
+                    2042.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1185.0,
+                    579.0,
+                    1185.0,
+                    579.0,
+                    1232.0,
+                    294.0,
+                    1232.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    72.0,
+                    817.0,
+                    72.0,
+                    817.0,
+                    108.0,
+                    295.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    768.0,
+                    746.0,
+                    934.0,
+                    746.0,
+                    934.0,
+                    789.0,
+                    768.0,
+                    789.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1077.0,
+                    374.0,
+                    1379.0,
+                    374.0,
+                    1379.0,
+                    408.0,
+                    1077.0,
+                    408.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1075.0,
+                    404.0,
+                    1388.0,
+                    404.0,
+                    1388.0,
+                    443.0,
+                    1075.0,
+                    443.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1075.0,
+                    435.0,
+                    1361.0,
+                    435.0,
+                    1361.0,
+                    473.0,
+                    1075.0,
+                    473.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1078.0,
+                    469.0,
+                    1333.0,
+                    469.0,
+                    1333.0,
+                    500.0,
+                    1078.0,
+                    500.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    840.0,
+                    2087.0,
+                    860.0,
+                    2087.0,
+                    860.0,
+                    2117.0,
+                    840.0,
+                    2117.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    311.0,
+                    545.0,
+                    552.0,
+                    545.0,
+                    552.0,
+                    581.0,
+                    311.0,
+                    581.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    395.0,
+                    819.0,
+                    1305.0,
+                    819.0,
+                    1305.0,
+                    856.0,
+                    395.0,
+                    856.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    393.0,
+                    852.0,
+                    1306.0,
+                    852.0,
+                    1306.0,
+                    887.0,
+                    393.0,
+                    887.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    394.0,
+                    882.0,
+                    1306.0,
+                    882.0,
+                    1306.0,
+                    915.0,
+                    394.0,
+                    915.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    393.0,
+                    912.0,
+                    1307.0,
+                    912.0,
+                    1307.0,
+                    948.0,
+                    393.0,
+                    948.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    393.0,
+                    941.0,
+                    1308.0,
+                    941.0,
+                    1308.0,
+                    980.0,
+                    393.0,
+                    980.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    394.0,
+                    972.0,
+                    1305.0,
+                    972.0,
+                    1305.0,
+                    1007.0,
+                    394.0,
+                    1007.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    392.0,
+                    1004.0,
+                    1306.0,
+                    1004.0,
+                    1306.0,
+                    1038.0,
+                    392.0,
+                    1038.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    392.0,
+                    1033.0,
+                    1304.0,
+                    1033.0,
+                    1304.0,
+                    1070.0,
+                    392.0,
+                    1070.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    394.0,
+                    1063.0,
+                    974.0,
+                    1063.0,
+                    974.0,
+                    1100.0,
+                    394.0,
+                    1100.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1052.0,
+                    1063.0,
+                    1305.0,
+                    1063.0,
+                    1305.0,
+                    1100.0,
+                    1052.0,
+                    1100.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    394.0,
+                    1095.0,
+                    1207.0,
+                    1095.0,
+                    1207.0,
+                    1128.0,
+                    394.0,
+                    1128.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1260.0,
+                    1404.0,
+                    1260.0,
+                    1404.0,
+                    1294.0,
+                    296.0,
+                    1294.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1292.0,
+                    1402.0,
+                    1292.0,
+                    1402.0,
+                    1325.0,
+                    294.0,
+                    1325.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1321.0,
+                    1402.0,
+                    1321.0,
+                    1402.0,
+                    1355.0,
+                    294.0,
+                    1355.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1352.0,
+                    1406.0,
+                    1352.0,
+                    1406.0,
+                    1386.0,
+                    294.0,
+                    1386.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1383.0,
+                    1408.0,
+                    1383.0,
+                    1408.0,
+                    1417.0,
+                    294.0,
+                    1417.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1410.0,
+                    1406.0,
+                    1410.0,
+                    1406.0,
+                    1448.0,
+                    293.0,
+                    1448.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1440.0,
+                    1404.0,
+                    1440.0,
+                    1404.0,
+                    1478.0,
+                    293.0,
+                    1478.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1472.0,
+                    1405.0,
+                    1472.0,
+                    1405.0,
+                    1506.0,
+                    294.0,
+                    1506.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1503.0,
+                    1404.0,
+                    1503.0,
+                    1404.0,
+                    1539.0,
+                    293.0,
+                    1539.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1529.0,
+                    1405.0,
+                    1529.0,
+                    1405.0,
+                    1572.0,
+                    292.0,
+                    1572.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1563.0,
+                    1405.0,
+                    1563.0,
+                    1405.0,
+                    1601.0,
+                    293.0,
+                    1601.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1596.0,
+                    591.0,
+                    1596.0,
+                    591.0,
+                    1626.0,
+                    296.0,
+                    1626.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1639.0,
+                    1404.0,
+                    1639.0,
+                    1404.0,
+                    1676.0,
+                    295.0,
+                    1676.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1673.0,
+                    1402.0,
+                    1673.0,
+                    1402.0,
+                    1707.0,
+                    294.0,
+                    1707.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1700.0,
+                    1404.0,
+                    1700.0,
+                    1404.0,
+                    1740.0,
+                    293.0,
+                    1740.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1731.0,
+                    1405.0,
+                    1731.0,
+                    1405.0,
+                    1769.0,
+                    293.0,
+                    1769.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1759.0,
+                    1405.0,
+                    1759.0,
+                    1405.0,
+                    1800.0,
+                    292.0,
+                    1800.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1793.0,
+                    1406.0,
+                    1793.0,
+                    1406.0,
+                    1831.0,
+                    294.0,
+                    1831.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1823.0,
+                    1406.0,
+                    1823.0,
+                    1406.0,
+                    1860.0,
+                    292.0,
+                    1860.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1854.0,
+                    1406.0,
+                    1854.0,
+                    1406.0,
+                    1889.0,
+                    292.0,
+                    1889.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1886.0,
+                    1405.0,
+                    1886.0,
+                    1405.0,
+                    1920.0,
+                    294.0,
+                    1920.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1917.0,
+                    1406.0,
+                    1917.0,
+                    1406.0,
+                    1951.0,
+                    294.0,
+                    1951.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1946.0,
+                    1407.0,
+                    1946.0,
+                    1407.0,
+                    1981.0,
+                    293.0,
+                    1981.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    319.0,
+                    373.0,
+                    590.0,
+                    373.0,
+                    590.0,
+                    410.0,
+                    319.0,
+                    410.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    312.0,
+                    402.0,
+                    619.0,
+                    402.0,
+                    619.0,
+                    445.0,
+                    312.0,
+                    445.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    312.0,
+                    435.0,
+                    597.0,
+                    435.0,
+                    597.0,
+                    473.0,
+                    312.0,
+                    473.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    312.0,
+                    469.0,
+                    570.0,
+                    469.0,
+                    570.0,
+                    501.0,
+                    312.0,
+                    501.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    313.0,
+                    548.0,
+                    550.0,
+                    548.0,
+                    550.0,
+                    576.0,
+                    313.0,
+                    576.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    310.0,
+                    572.0,
+                    625.0,
+                    572.0,
+                    625.0,
+                    614.0,
+                    310.0,
+                    614.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    312.0,
+                    605.0,
+                    597.0,
+                    605.0,
+                    597.0,
+                    643.0,
+                    312.0,
+                    643.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    312.0,
+                    637.0,
+                    786.0,
+                    637.0,
+                    786.0,
+                    673.0,
+                    312.0,
+                    673.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1077.0,
+                    374.0,
+                    1379.0,
+                    374.0,
+                    1379.0,
+                    408.0,
+                    1077.0,
+                    408.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1075.0,
+                    404.0,
+                    1388.0,
+                    404.0,
+                    1388.0,
+                    443.0,
+                    1075.0,
+                    443.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1075.0,
+                    435.0,
+                    1361.0,
+                    435.0,
+                    1361.0,
+                    473.0,
+                    1075.0,
+                    473.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1078.0,
+                    469.0,
+                    1333.0,
+                    469.0,
+                    1333.0,
+                    500.0,
+                    1078.0,
+                    500.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 0,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    368,
+                    1404,
+                    368,
+                    1404,
+                    765,
+                    298,
+                    765
+                ],
+                "score": 0.983
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    781,
+                    1404,
+                    781,
+                    1404,
+                    1086,
+                    298,
+                    1086
+                ],
+                "score": 0.982
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1546,
+                    1404,
+                    1546,
+                    1404,
+                    1683,
+                    297,
+                    1683
+                ],
+                "score": 0.979
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1102,
+                    1402,
+                    1102,
+                    1402,
+                    1225,
+                    298,
+                    1225
+                ],
+                "score": 0.977
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    1351,
+                    1403,
+                    1351,
+                    1403,
+                    1445,
+                    299,
+                    1445
+                ],
+                "score": 0.975
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    228,
+                    1401,
+                    228,
+                    1401,
+                    352,
+                    299,
+                    352
+                ],
+                "score": 0.975
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    558,
+                    1944,
+                    1136,
+                    1944,
+                    1136,
+                    2042,
+                    558,
+                    2042
+                ],
+                "score": 0.963
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    653,
+                    1708,
+                    1039,
+                    1708,
+                    1039,
+                    1807,
+                    653,
+                    1807
+                ],
+                "score": 0.962
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    296,
+                    1849,
+                    1401,
+                    1849,
+                    1401,
+                    1919,
+                    296,
+                    1919
+                ],
+                "score": 0.958
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    300,
+                    76,
+                    815,
+                    76,
+                    815,
+                    104,
+                    300,
+                    104
+                ],
+                "score": 0.898
+            },
+            {
+                "category_id": 9,
+                "poly": [
+                    1366,
+                    1742,
+                    1400,
+                    1742,
+                    1400,
+                    1774,
+                    1366,
+                    1774
+                ],
+                "score": 0.895
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    302,
+                    1276,
+                    724,
+                    1276,
+                    724,
+                    1312,
+                    302,
+                    1312
+                ],
+                "score": 0.894
+            },
+            {
+                "category_id": 9,
+                "poly": [
+                    1365,
+                    1978,
+                    1401,
+                    1978,
+                    1401,
+                    2009,
+                    1365,
+                    2009
+                ],
+                "score": 0.874
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    301,
+                    1490,
+                    851,
+                    1490,
+                    851,
+                    1521,
+                    301,
+                    1521
+                ],
+                "score": 0.821
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    840,
+                    2088,
+                    859,
+                    2088,
+                    859,
+                    2113,
+                    840,
+                    2113
+                ],
+                "score": 0.725
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    840,
+                    2087,
+                    860,
+                    2087,
+                    860,
+                    2113,
+                    840,
+                    2113
+                ],
+                "score": 0.404
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    298,
+                    1490,
+                    853,
+                    1490,
+                    853,
+                    1522,
+                    298,
+                    1522
+                ],
+                "score": 0.109
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    655,
+                    1703,
+                    1044,
+                    1703,
+                    1044,
+                    1809,
+                    655,
+                    1809
+                ],
+                "score": 0.94,
+                "latex": "\\mathbf { a } _ { i } ^ { ( 1 ) } ( \\mathbf { X } ) : = b _ { i } ^ { ( 1 ) } \\mathbf { 1 } + \\sum _ { j = 1 } ^ { C ^ { ( 0 ) } } \\mathbf { W } _ { i , j } ^ { ( 1 ) } \\mathbf { x } _ { j } \\ ."
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    667,
+                    1646,
+                    867,
+                    1646,
+                    867,
+                    1683,
+                    667,
+                    1683
+                ],
+                "score": 0.93,
+                "latex": "i \\in \\{ 1 , \\ldots , C ^ { ( 1 ) } \\}"
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    560,
+                    1940,
+                    1139,
+                    1940,
+                    1139,
+                    2044,
+                    560,
+                    2044
+                ],
+                "score": 0.93,
+                "latex": "\\mathbf { a } _ { i } ^ { ( \\ell + 1 ) } ( \\mathbf { X } ) : = b _ { i } ^ { ( \\ell + 1 ) } \\mathbf { 1 } + \\sum _ { j = 1 } ^ { C ^ { ( \\ell ) } } \\mathbf { W } _ { i , j } ^ { ( \\ell + 1 ) } \\phi \\left( \\mathbf { a } _ { j } ^ { ( \\ell ) } ( \\mathbf { X } ) \\right) ."
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    362,
+                    1882,
+                    494,
+                    1882,
+                    494,
+                    1919,
+                    362,
+                    1919
+                ],
+                "score": 0.92,
+                "latex": "\\mathbf { A } ^ { ( L + 1 ) } ( \\mathbf { X } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1160,
+                    1611,
+                    1261,
+                    1611,
+                    1261,
+                    1649,
+                    1160,
+                    1649
+                ],
+                "score": 0.92,
+                "latex": "\\mathbf { A } ^ { ( 1 ) } ( \\mathbf { X } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1193,
+                    1546,
+                    1399,
+                    1546,
+                    1399,
+                    1583,
+                    1193,
+                    1583
+                ],
+                "score": 0.92,
+                "latex": "C ^ { ( 0 ) } \\times ( H ^ { ( 0 ) } D ^ { ( 0 ) } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    364,
+                    1611,
+                    481,
+                    1611,
+                    481,
+                    1645,
+                    364,
+                    1645
+                ],
+                "score": 0.91,
+                "latex": "C ^ { ( 0 ) } = 3"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1299,
+                    1848,
+                    1400,
+                    1848,
+                    1400,
+                    1885,
+                    1299,
+                    1885
+                ],
+                "score": 0.9,
+                "latex": "\\mathbf { A } ^ { ( 2 ) } ( \\mathbf { X } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1080,
+                    1547,
+                    1135,
+                    1547,
+                    1135,
+                    1579,
+                    1080,
+                    1579
+                ],
+                "score": 0.9,
+                "latex": "D ^ { ( 0 ) }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    906,
+                    1547,
+                    962,
+                    1547,
+                    962,
+                    1579,
+                    906,
+                    1579
+                ],
+                "score": 0.89,
+                "latex": "H ^ { ( 0 ) }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    759,
+                    1587,
+                    953,
+                    1587,
+                    953,
+                    1614,
+                    759,
+                    1614
+                ],
+                "score": 0.88,
+                "latex": "\\mathbf { x } _ { 1 } , \\mathbf { x } _ { 2 } , \\ldots , \\mathbf { x } _ { C ^ { ( 0 ) } }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    880,
+                    1164,
+                    956,
+                    1164,
+                    956,
+                    1193,
+                    880,
+                    1193
+                ],
+                "score": 0.87,
+                "latex": "0 . 8 4 \\%"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    620,
+                    1854,
+                    641,
+                    1854,
+                    641,
+                    1879,
+                    620,
+                    1879
+                ],
+                "score": 0.83,
+                "latex": "L"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    773,
+                    1551,
+                    800,
+                    1551,
+                    800,
+                    1578,
+                    773,
+                    1578
+                ],
+                "score": 0.67,
+                "latex": "\\mathbf { X }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    446,
+                    1384,
+                    487,
+                    1384,
+                    487,
+                    1411,
+                    446,
+                    1411
+                ],
+                "score": 0.41,
+                "latex": "n \\mathbf { D }"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    73.0,
+                    817.0,
+                    73.0,
+                    817.0,
+                    108.0,
+                    296.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1274.0,
+                    729.0,
+                    1274.0,
+                    729.0,
+                    1317.0,
+                    292.0,
+                    1317.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1490.0,
+                    856.0,
+                    1490.0,
+                    856.0,
+                    1525.0,
+                    296.0,
+                    1525.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    838.0,
+                    2086.0,
+                    863.0,
+                    2086.0,
+                    863.0,
+                    2120.0,
+                    838.0,
+                    2120.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    838.0,
+                    2085.0,
+                    862.0,
+                    2085.0,
+                    862.0,
+                    2121.0,
+                    838.0,
+                    2121.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1490.0,
+                    856.0,
+                    1490.0,
+                    856.0,
+                    1526.0,
+                    295.0,
+                    1526.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    370.0,
+                    1402.0,
+                    370.0,
+                    1402.0,
+                    402.0,
+                    297.0,
+                    402.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    397.0,
+                    1405.0,
+                    397.0,
+                    1405.0,
+                    436.0,
+                    293.0,
+                    436.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    431.0,
+                    1405.0,
+                    431.0,
+                    1405.0,
+                    463.0,
+                    296.0,
+                    463.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    459.0,
+                    1405.0,
+                    459.0,
+                    1405.0,
+                    495.0,
+                    294.0,
+                    495.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    490.0,
+                    1402.0,
+                    490.0,
+                    1402.0,
+                    526.0,
+                    294.0,
+                    526.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    522.0,
+                    1402.0,
+                    522.0,
+                    1402.0,
+                    555.0,
+                    294.0,
+                    555.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    551.0,
+                    1405.0,
+                    551.0,
+                    1405.0,
+                    586.0,
+                    295.0,
+                    586.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    582.0,
+                    1406.0,
+                    582.0,
+                    1406.0,
+                    618.0,
+                    294.0,
+                    618.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    612.0,
+                    1405.0,
+                    612.0,
+                    1405.0,
+                    648.0,
+                    294.0,
+                    648.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    642.0,
+                    1406.0,
+                    642.0,
+                    1406.0,
+                    678.0,
+                    294.0,
+                    678.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    675.0,
+                    1405.0,
+                    675.0,
+                    1405.0,
+                    708.0,
+                    294.0,
+                    708.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    701.0,
+                    1406.0,
+                    701.0,
+                    1406.0,
+                    740.0,
+                    291.0,
+                    740.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    736.0,
+                    696.0,
+                    736.0,
+                    696.0,
+                    768.0,
+                    294.0,
+                    768.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    783.0,
+                    1404.0,
+                    783.0,
+                    1404.0,
+                    815.0,
+                    294.0,
+                    815.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    811.0,
+                    1407.0,
+                    811.0,
+                    1407.0,
+                    847.0,
+                    291.0,
+                    847.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    840.0,
+                    1404.0,
+                    840.0,
+                    1404.0,
+                    879.0,
+                    292.0,
+                    879.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    870.0,
+                    1405.0,
+                    870.0,
+                    1405.0,
+                    908.0,
+                    293.0,
+                    908.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    903.0,
+                    1405.0,
+                    903.0,
+                    1405.0,
+                    937.0,
+                    292.0,
+                    937.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    933.0,
+                    1406.0,
+                    933.0,
+                    1406.0,
+                    969.0,
+                    293.0,
+                    969.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    965.0,
+                    1406.0,
+                    965.0,
+                    1406.0,
+                    997.0,
+                    294.0,
+                    997.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    994.0,
+                    1406.0,
+                    994.0,
+                    1406.0,
+                    1027.0,
+                    291.0,
+                    1027.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1022.0,
+                    1405.0,
+                    1022.0,
+                    1405.0,
+                    1060.0,
+                    293.0,
+                    1060.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1055.0,
+                    609.0,
+                    1055.0,
+                    609.0,
+                    1089.0,
+                    295.0,
+                    1089.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1545.0,
+                    772.0,
+                    1545.0,
+                    772.0,
+                    1584.0,
+                    292.0,
+                    1584.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    801.0,
+                    1545.0,
+                    905.0,
+                    1545.0,
+                    905.0,
+                    1584.0,
+                    801.0,
+                    1584.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    963.0,
+                    1545.0,
+                    1079.0,
+                    1545.0,
+                    1079.0,
+                    1584.0,
+                    963.0,
+                    1584.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1136.0,
+                    1545.0,
+                    1192.0,
+                    1545.0,
+                    1192.0,
+                    1584.0,
+                    1136.0,
+                    1584.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1400.0,
+                    1545.0,
+                    1405.0,
+                    1545.0,
+                    1405.0,
+                    1584.0,
+                    1400.0,
+                    1584.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1579.0,
+                    758.0,
+                    1579.0,
+                    758.0,
+                    1621.0,
+                    291.0,
+                    1621.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    954.0,
+                    1579.0,
+                    1406.0,
+                    1579.0,
+                    1406.0,
+                    1621.0,
+                    954.0,
+                    1621.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1610.0,
+                    363.0,
+                    1610.0,
+                    363.0,
+                    1652.0,
+                    291.0,
+                    1652.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    482.0,
+                    1610.0,
+                    1159.0,
+                    1610.0,
+                    1159.0,
+                    1652.0,
+                    482.0,
+                    1652.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1262.0,
+                    1610.0,
+                    1406.0,
+                    1610.0,
+                    1406.0,
+                    1652.0,
+                    1262.0,
+                    1652.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1644.0,
+                    666.0,
+                    1644.0,
+                    666.0,
+                    1686.0,
+                    294.0,
+                    1686.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    868.0,
+                    1644.0,
+                    881.0,
+                    1644.0,
+                    881.0,
+                    1686.0,
+                    868.0,
+                    1686.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1102.0,
+                    1402.0,
+                    1102.0,
+                    1402.0,
+                    1137.0,
+                    293.0,
+                    1137.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1131.0,
+                    1406.0,
+                    1131.0,
+                    1406.0,
+                    1169.0,
+                    292.0,
+                    1169.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1165.0,
+                    879.0,
+                    1165.0,
+                    879.0,
+                    1197.0,
+                    293.0,
+                    1197.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    957.0,
+                    1165.0,
+                    1406.0,
+                    1165.0,
+                    1406.0,
+                    1197.0,
+                    957.0,
+                    1197.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1195.0,
+                    1092.0,
+                    1195.0,
+                    1092.0,
+                    1227.0,
+                    296.0,
+                    1227.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1351.0,
+                    1403.0,
+                    1351.0,
+                    1403.0,
+                    1388.0,
+                    294.0,
+                    1388.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    1385.0,
+                    445.0,
+                    1385.0,
+                    445.0,
+                    1415.0,
+                    297.0,
+                    1415.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    488.0,
+                    1385.0,
+                    1404.0,
+                    1385.0,
+                    1404.0,
+                    1415.0,
+                    488.0,
+                    1415.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1413.0,
+                    1222.0,
+                    1413.0,
+                    1222.0,
+                    1450.0,
+                    294.0,
+                    1450.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    229.0,
+                    1405.0,
+                    229.0,
+                    1405.0,
+                    267.0,
+                    293.0,
+                    267.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    261.0,
+                    1405.0,
+                    261.0,
+                    1405.0,
+                    297.0,
+                    294.0,
+                    297.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    292.0,
+                    1406.0,
+                    292.0,
+                    1406.0,
+                    328.0,
+                    293.0,
+                    328.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    324.0,
+                    781.0,
+                    324.0,
+                    781.0,
+                    355.0,
+                    294.0,
+                    355.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1844.0,
+                    619.0,
+                    1844.0,
+                    619.0,
+                    1888.0,
+                    294.0,
+                    1888.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    642.0,
+                    1844.0,
+                    1298.0,
+                    1844.0,
+                    1298.0,
+                    1888.0,
+                    642.0,
+                    1888.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1879.0,
+                    361.0,
+                    1879.0,
+                    361.0,
+                    1922.0,
+                    292.0,
+                    1922.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    495.0,
+                    1879.0,
+                    773.0,
+                    1879.0,
+                    773.0,
+                    1922.0,
+                    495.0,
+                    1922.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 1,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1778,
+                    1405,
+                    1778,
+                    1405,
+                    2035,
+                    297,
+                    2035
+                ],
+                "score": 0.981
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    296,
+                    969,
+                    1405,
+                    969,
+                    1405,
+                    1198,
+                    296,
+                    1198
+                ],
+                "score": 0.979
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1209,
+                    1404,
+                    1209,
+                    1404,
+                    1362,
+                    297,
+                    1362
+                ],
+                "score": 0.977
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1580,
+                    1404,
+                    1580,
+                    1404,
+                    1675,
+                    298,
+                    1675
+                ],
+                "score": 0.976
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1380,
+                    1405,
+                    1380,
+                    1405,
+                    1482,
+                    297,
+                    1482
+                ],
+                "score": 0.973
+            },
+            {
+                "category_id": 3,
+                "poly": [
+                    528,
+                    219,
+                    1126,
+                    219,
+                    1126,
+                    684,
+                    528,
+                    684
+                ],
+                "score": 0.962
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    296,
+                    881,
+                    1402,
+                    881,
+                    1402,
+                    957,
+                    296,
+                    957
+                ],
+                "score": 0.951
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    483,
+                    1501,
+                    1210,
+                    1501,
+                    1210,
+                    1560,
+                    483,
+                    1560
+                ],
+                "score": 0.933
+            },
+            {
+                "category_id": 4,
+                "poly": [
+                    294,
+                    706,
+                    1406,
+                    706,
+                    1406,
+                    824,
+                    294,
+                    824
+                ],
+                "score": 0.932
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    301,
+                    1719,
+                    753,
+                    1719,
+                    753,
+                    1751,
+                    301,
+                    1751
+                ],
+                "score": 0.911
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    298,
+                    75,
+                    815,
+                    75,
+                    815,
+                    105,
+                    298,
+                    105
+                ],
+                "score": 0.893
+            },
+            {
+                "category_id": 9,
+                "poly": [
+                    1366,
+                    1516,
+                    1400,
+                    1516,
+                    1400,
+                    1547,
+                    1366,
+                    1547
+                ],
+                "score": 0.87
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    490,
+                    1501,
+                    813,
+                    1501,
+                    813,
+                    1561,
+                    490,
+                    1561
+                ],
+                "score": 0.761
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    841,
+                    2088,
+                    859,
+                    2088,
+                    859,
+                    2112,
+                    841,
+                    2112
+                ],
+                "score": 0.67
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    841,
+                    2088,
+                    859,
+                    2088,
+                    859,
+                    2112,
+                    841,
+                    2112
+                ],
+                "score": 0.378
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    967,
+                    707,
+                    1061,
+                    707,
+                    1061,
+                    754,
+                    967,
+                    754
+                ],
+                "score": 0.94,
+                "latex": "\\mathbf { W } _ { i , j } ^ { ( 0 ) } \\mathbf { x } _ { j }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    830,
+                    1082,
+                    913,
+                    1082,
+                    913,
+                    1128,
+                    830,
+                    1128
+                ],
+                "score": 0.93,
+                "latex": "\\mathbf { U } _ { i , j } ^ { ( \\ell + 1 ) }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    809,
+                    970,
+                    901,
+                    970,
+                    901,
+                    1017,
+                    809,
+                    1017
+                ],
+                "score": 0.93,
+                "latex": "\\mathbf { W } _ { i , j } ^ { ( \\ell + 1 ) }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    641,
+                    708,
+                    754,
+                    708,
+                    754,
+                    753,
+                    641,
+                    753
+                ],
+                "score": 0.93,
+                "latex": "\\mathbf { U } _ { i , j } ^ { ( 0 ) } * \\mathbf { x } _ { j }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    434,
+                    1125,
+                    525,
+                    1125,
+                    525,
+                    1171,
+                    434,
+                    1171
+                ],
+                "score": 0.93,
+                "latex": "\\mathbf { W } _ { i , j \\dots } ^ { ( \\ell + 1 ) }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1086,
+                    1278,
+                    1183,
+                    1278,
+                    1183,
+                    1324,
+                    1086,
+                    1324
+                ],
+                "score": 0.93,
+                "latex": "\\mathbf { W } _ { i , j } ^ { ( L + 1 ) }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    346,
+                    1018,
+                    548,
+                    1018,
+                    548,
+                    1056,
+                    346,
+                    1056
+                ],
+                "score": 0.93,
+                "latex": "j \\in \\{ 1 , \\ldots , C ^ { ( \\ell ) } \\}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    474,
+                    885,
+                    574,
+                    885,
+                    574,
+                    922,
+                    474,
+                    922
+                ],
+                "score": 0.93,
+                "latex": "\\mathbf { A } ^ { ( \\ell ) } ( \\mathbf { X } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    510,
+                    1320,
+                    590,
+                    1320,
+                    590,
+                    1363,
+                    510,
+                    1363
+                ],
+                "score": 0.93,
+                "latex": "\\mathbf { a } _ { i } ^ { ( L + 1 ) }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1062,
+                    881,
+                    1136,
+                    881,
+                    1136,
+                    923,
+                    1062,
+                    923
+                ],
+                "score": 0.93,
+                "latex": "\\mathbf { a } _ { i } ^ { ( \\ell + 1 ) }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    621,
+                    884,
+                    834,
+                    884,
+                    834,
+                    922,
+                    621,
+                    922
+                ],
+                "score": 0.93,
+                "latex": "C ^ { ( \\ell ) } \\times ( H ^ { ( \\ell ) } D ^ { ( \\ell ) } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1093,
+                    1938,
+                    1224,
+                    1938,
+                    1224,
+                    1971,
+                    1093,
+                    1971
+                ],
+                "score": 0.93,
+                "latex": "C ^ { ( 2 ) }  \\infty"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1180,
+                    1875,
+                    1236,
+                    1875,
+                    1236,
+                    1906,
+                    1180,
+                    1906
+                ],
+                "score": 0.92,
+                "latex": "N ^ { ( \\ell ) }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1035,
+                    970,
+                    1105,
+                    970,
+                    1105,
+                    1013,
+                    1035,
+                    1013
+                ],
+                "score": 0.92,
+                "latex": "b _ { i } ^ { ( \\ell + 1 ) }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    892,
+                    1938,
+                    1023,
+                    1938,
+                    1023,
+                    1971,
+                    892,
+                    1971
+                ],
+                "score": 0.92,
+                "latex": "\\dot { C ^ { ( 1 ) } } \\stackrel { - } {  } \\infty"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1142,
+                    1377,
+                    1197,
+                    1377,
+                    1197,
+                    1422,
+                    1142,
+                    1422
+                ],
+                "score": 0.92,
+                "latex": "\\mathbf { U } _ { i , j } ^ { ( \\ell ) }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    590,
+                    751,
+                    654,
+                    751,
+                    654,
+                    797,
+                    590,
+                    797
+                ],
+                "score": 0.92,
+                "latex": "\\mathbf { W } _ { i , j } ^ { ( 0 ) }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1135,
+                    920,
+                    1270,
+                    920,
+                    1270,
+                    958,
+                    1135,
+                    958
+                ],
+                "score": 0.92,
+                "latex": "\\phi ( \\mathbf { A } ^ { ( \\ell ) } ( \\mathbf { X } ) )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    876,
+                    1210,
+                    1009,
+                    1210,
+                    1009,
+                    1246,
+                    876,
+                    1246
+                ],
+                "score": 0.92,
+                "latex": "\\mathbf { A } ^ { ( L + 1 ) } ( \\mathbf { X } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1325,
+                    1378,
+                    1366,
+                    1378,
+                    1366,
+                    1419,
+                    1325,
+                    1419
+                ],
+                "score": 0.92,
+                "latex": "b _ { i } ^ { ( \\ell ) }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1318,
+                    1125,
+                    1399,
+                    1125,
+                    1399,
+                    1172,
+                    1318,
+                    1172
+                ],
+                "score": 0.92,
+                "latex": "\\mathbf { U } _ { i , j } ^ { ( \\ell + 1 ) }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    708,
+                    1875,
+                    844,
+                    1875,
+                    844,
+                    1907,
+                    708,
+                    1907
+                ],
+                "score": 0.91,
+                "latex": "N ^ { ( 1 ) }  \\infty"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    914,
+                    1875,
+                    1051,
+                    1875,
+                    1051,
+                    1906,
+                    914,
+                    1906
+                ],
+                "score": 0.91,
+                "latex": "N ^ { ( 2 ) }  \\infty"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1309,
+                    1015,
+                    1399,
+                    1015,
+                    1399,
+                    1059,
+                    1309,
+                    1059
+                ],
+                "score": 0.91,
+                "latex": "\\mathbf { W } _ { i , j } ^ { ( \\ell + 1 ) }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1335,
+                    708,
+                    1399,
+                    708,
+                    1399,
+                    754,
+                    1335,
+                    754
+                ],
+                "score": 0.91,
+                "latex": "\\mathbf { W } _ { i , j } ^ { ( 0 ) }"
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    489,
+                    1501,
+                    1212,
+                    1501,
+                    1212,
+                    1561,
+                    489,
+                    1561
+                ],
+                "score": 0.91,
+                "latex": "U _ { i , j , x , y } ^ { ( \\ell ) } \\sim \\mathcal { N } \\left( 0 , \\sigma _ { w } ^ { 2 } / C ^ { ( \\ell ) } \\right) , \\qquad b _ { i } ^ { ( \\ell ) } \\sim \\mathcal { N } \\left( 0 , \\sigma _ { b } ^ { 2 } \\right) ."
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    942,
+                    1245,
+                    1206,
+                    1245,
+                    1206,
+                    1276,
+                    942,
+                    1276
+                ],
+                "score": 0.91,
+                "latex": "H ^ { ( L + 1 ) } = D ^ { ( L + 1 ) } = 1"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1163,
+                    975,
+                    1402,
+                    975,
+                    1402,
+                    1013,
+                    1163,
+                    1013
+                ],
+                "score": 0.9,
+                "latex": "i \\in \\{ 1 , \\ldots , C ^ { ( \\ell + 1 ) } \\}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    737,
+                    1811,
+                    824,
+                    1811,
+                    824,
+                    1843,
+                    737,
+                    1843
+                ],
+                "score": 0.89,
+                "latex": "\\mathbf { A } ^ { ( L + 1 ) }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1129,
+                    1421,
+                    1167,
+                    1421,
+                    1167,
+                    1451,
+                    1129,
+                    1451
+                ],
+                "score": 0.88,
+                "latex": "i , j"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    430,
+                    794,
+                    462,
+                    794,
+                    462,
+                    824,
+                    430,
+                    824
+                ],
+                "score": 0.87,
+                "latex": "\\mathbf { x } _ { j }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    397,
+                    1453,
+                    443,
+                    1453,
+                    443,
+                    1481,
+                    397,
+                    1481
+                ],
+                "score": 0.85,
+                "latex": "x , y"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    297,
+                    1096,
+                    329,
+                    1096,
+                    329,
+                    1125,
+                    297,
+                    1125
+                ],
+                "score": 0.84,
+                "latex": "\\mathbf { x } _ { j }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    599,
+                    334,
+                    655,
+                    334,
+                    655,
+                    383,
+                    599,
+                    383
+                ],
+                "score": 0.73,
+                "latex": "\\mathbf { U } _ { i , j } ^ { ( 0 ) }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    995,
+                    1092,
+                    1010,
+                    1092,
+                    1010,
+                    1123,
+                    995,
+                    1123
+                ],
+                "score": 0.73,
+                "latex": "j"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    298,
+                    925,
+                    312,
+                    925,
+                    312,
+                    956,
+                    298,
+                    956
+                ],
+                "score": 0.7,
+                "latex": "j"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    998,
+                    1420,
+                    1013,
+                    1420,
+                    1013,
+                    1446,
+                    998,
+                    1446
+                ],
+                "score": 0.69,
+                "latex": "\\ell"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    463,
+                    763,
+                    483,
+                    763,
+                    483,
+                    790,
+                    463,
+                    790
+                ],
+                "score": 0.63,
+                "latex": "\\mu"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1103,
+                    763,
+                    1124,
+                    763,
+                    1124,
+                    791,
+                    1103,
+                    791
+                ],
+                "score": 0.61,
+                "latex": "\\mu"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1231,
+                    1815,
+                    1259,
+                    1815,
+                    1259,
+                    1842,
+                    1231,
+                    1842
+                ],
+                "score": 0.49,
+                "latex": "\\mathbf { X }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    572,
+                    1910,
+                    587,
+                    1910,
+                    587,
+                    1936,
+                    572,
+                    1936
+                ],
+                "score": 0.37,
+                "latex": "\\ell"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    930.0,
+                    230.0,
+                    954.0,
+                    230.0,
+                    954.0,
+                    256.0,
+                    930.0,
+                    256.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    900.0,
+                    231.0,
+                    924.0,
+                    231.0,
+                    924.0,
+                    256.0,
+                    900.0,
+                    256.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    875.0,
+                    234.0,
+                    890.0,
+                    234.0,
+                    890.0,
+                    251.0,
+                    875.0,
+                    251.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    525.0,
+                    255.0,
+                    829.0,
+                    255.0,
+                    829.0,
+                    289.0,
+                    525.0,
+                    289.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    872.0,
+                    256.0,
+                    929.0,
+                    256.0,
+                    929.0,
+                    287.0,
+                    872.0,
+                    287.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    930.0,
+                    259.0,
+                    955.0,
+                    259.0,
+                    955.0,
+                    286.0,
+                    930.0,
+                    286.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    875.0,
+                    295.0,
+                    890.0,
+                    295.0,
+                    890.0,
+                    311.0,
+                    875.0,
+                    311.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    899.0,
+                    288.0,
+                    928.0,
+                    288.0,
+                    928.0,
+                    317.0,
+                    899.0,
+                    317.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    929.0,
+                    290.0,
+                    954.0,
+                    290.0,
+                    954.0,
+                    316.0,
+                    929.0,
+                    316.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    997.0,
+                    279.0,
+                    1118.0,
+                    279.0,
+                    1118.0,
+                    325.0,
+                    997.0,
+                    325.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    689.0,
+                    325.0,
+                    753.0,
+                    325.0,
+                    753.0,
+                    361.0,
+                    689.0,
+                    361.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    984.0,
+                    322.0,
+                    1126.0,
+                    322.0,
+                    1126.0,
+                    354.0,
+                    984.0,
+                    354.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    689.0,
+                    354.0,
+                    755.0,
+                    354.0,
+                    755.0,
+                    392.0,
+                    689.0,
+                    392.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    656.0,
+                    357.0,
+                    659.0,
+                    357.0,
+                    659.0,
+                    386.0,
+                    656.0,
+                    386.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    892.0,
+                    364.0,
+                    933.0,
+                    364.0,
+                    933.0,
+                    649.0,
+                    892.0,
+                    649.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1082.0,
+                    397.0,
+                    1120.0,
+                    397.0,
+                    1120.0,
+                    584.0,
+                    1082.0,
+                    584.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    586.0,
+                    442.0,
+                    651.0,
+                    442.0,
+                    651.0,
+                    480.0,
+                    586.0,
+                    480.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    548.0,
+                    449.0,
+                    585.0,
+                    449.0,
+                    585.0,
+                    550.0,
+                    548.0,
+                    550.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    682.0,
+                    450.0,
+                    708.0,
+                    450.0,
+                    708.0,
+                    473.0,
+                    682.0,
+                    473.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    710.0,
+                    450.0,
+                    736.0,
+                    450.0,
+                    736.0,
+                    473.0,
+                    710.0,
+                    473.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    619.0,
+                    473.0,
+                    679.0,
+                    473.0,
+                    679.0,
+                    508.0,
+                    619.0,
+                    508.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    708.0,
+                    473.0,
+                    769.0,
+                    473.0,
+                    769.0,
+                    507.0,
+                    708.0,
+                    507.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    676.0,
+                    501.0,
+                    741.0,
+                    501.0,
+                    741.0,
+                    540.0,
+                    676.0,
+                    540.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    771.0,
+                    508.0,
+                    797.0,
+                    508.0,
+                    797.0,
+                    534.0,
+                    771.0,
+                    534.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    799.0,
+                    508.0,
+                    825.0,
+                    508.0,
+                    825.0,
+                    532.0,
+                    799.0,
+                    532.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    707.0,
+                    532.0,
+                    769.0,
+                    532.0,
+                    769.0,
+                    567.0,
+                    707.0,
+                    567.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    800.0,
+                    537.0,
+                    854.0,
+                    537.0,
+                    854.0,
+                    563.0,
+                    800.0,
+                    563.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    731.0,
+                    623.0,
+                    772.0,
+                    623.0,
+                    772.0,
+                    657.0,
+                    731.0,
+                    657.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    896.0,
+                    652.0,
+                    935.0,
+                    652.0,
+                    935.0,
+                    690.0,
+                    896.0,
+                    690.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    522.25,
+                    331.0,
+                    672.25,
+                    331.0,
+                    672.25,
+                    376.0,
+                    522.25,
+                    376.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    719.0,
+                    399.5,
+                    726.0,
+                    399.5,
+                    726.0,
+                    425.0,
+                    719.0,
+                    425.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    692.25,
+                    599.5,
+                    784.25,
+                    599.5,
+                    784.25,
+                    649.5,
+                    692.25,
+                    649.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    655.0,
+                    753.0,
+                    1102.0,
+                    753.0,
+                    1102.0,
+                    800.0,
+                    655.0,
+                    800.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1125.0,
+                    753.0,
+                    1409.0,
+                    753.0,
+                    1409.0,
+                    800.0,
+                    1125.0,
+                    800.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    782.0,
+                    429.0,
+                    782.0,
+                    429.0,
+                    830.0,
+                    293.0,
+                    830.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    463.0,
+                    782.0,
+                    474.0,
+                    782.0,
+                    474.0,
+                    830.0,
+                    463.0,
+                    830.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    290.0,
+                    704.5,
+                    708.0,
+                    704.5,
+                    708.0,
+                    756.5,
+                    290.0,
+                    756.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    682.75,
+                    694.0,
+                    1414.75,
+                    694.0,
+                    1414.75,
+                    767.5,
+                    682.75,
+                    767.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    286.0,
+                    745.5,
+                    665.0,
+                    745.5,
+                    665.0,
+                    803.5,
+                    286.0,
+                    803.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1718.0,
+                    756.0,
+                    1718.0,
+                    756.0,
+                    1755.0,
+                    295.0,
+                    1755.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    73.0,
+                    817.0,
+                    73.0,
+                    817.0,
+                    108.0,
+                    296.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    838.0,
+                    2085.0,
+                    862.0,
+                    2085.0,
+                    862.0,
+                    2118.0,
+                    838.0,
+                    2118.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    838.0,
+                    2085.0,
+                    862.0,
+                    2085.0,
+                    862.0,
+                    2118.0,
+                    838.0,
+                    2118.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1781.0,
+                    1405.0,
+                    1781.0,
+                    1405.0,
+                    1816.0,
+                    296.0,
+                    1816.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1807.0,
+                    736.0,
+                    1807.0,
+                    736.0,
+                    1852.0,
+                    291.0,
+                    1852.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    825.0,
+                    1807.0,
+                    1230.0,
+                    1807.0,
+                    1230.0,
+                    1852.0,
+                    825.0,
+                    1852.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1260.0,
+                    1807.0,
+                    1408.0,
+                    1807.0,
+                    1408.0,
+                    1852.0,
+                    1260.0,
+                    1852.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1843.0,
+                    1406.0,
+                    1843.0,
+                    1406.0,
+                    1882.0,
+                    294.0,
+                    1882.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1873.0,
+                    707.0,
+                    1873.0,
+                    707.0,
+                    1914.0,
+                    291.0,
+                    1914.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    845.0,
+                    1873.0,
+                    913.0,
+                    1873.0,
+                    913.0,
+                    1914.0,
+                    845.0,
+                    1914.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1052.0,
+                    1873.0,
+                    1179.0,
+                    1873.0,
+                    1179.0,
+                    1914.0,
+                    1052.0,
+                    1914.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1237.0,
+                    1873.0,
+                    1407.0,
+                    1873.0,
+                    1407.0,
+                    1914.0,
+                    1237.0,
+                    1914.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1909.0,
+                    571.0,
+                    1909.0,
+                    571.0,
+                    1944.0,
+                    295.0,
+                    1944.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    588.0,
+                    1909.0,
+                    1405.0,
+                    1909.0,
+                    1405.0,
+                    1944.0,
+                    588.0,
+                    1944.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    290.0,
+                    1935.0,
+                    891.0,
+                    1935.0,
+                    891.0,
+                    1981.0,
+                    290.0,
+                    1981.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1024.0,
+                    1935.0,
+                    1092.0,
+                    1935.0,
+                    1092.0,
+                    1981.0,
+                    1024.0,
+                    1981.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1225.0,
+                    1935.0,
+                    1407.0,
+                    1935.0,
+                    1407.0,
+                    1981.0,
+                    1225.0,
+                    1981.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    290.0,
+                    1970.0,
+                    1406.0,
+                    1970.0,
+                    1406.0,
+                    2011.0,
+                    290.0,
+                    2011.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    2004.0,
+                    579.0,
+                    2004.0,
+                    579.0,
+                    2035.0,
+                    296.0,
+                    2035.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    276.0,
+                    969.0,
+                    345.0,
+                    969.0,
+                    345.0,
+                    1089.0,
+                    276.0,
+                    1089.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    549.0,
+                    969.0,
+                    808.0,
+                    969.0,
+                    808.0,
+                    1089.0,
+                    549.0,
+                    1089.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    902.0,
+                    969.0,
+                    1034.0,
+                    969.0,
+                    1034.0,
+                    1089.0,
+                    902.0,
+                    1089.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1106.0,
+                    969.0,
+                    1162.0,
+                    969.0,
+                    1162.0,
+                    1089.0,
+                    1106.0,
+                    1089.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1403.0,
+                    969.0,
+                    1422.0,
+                    969.0,
+                    1422.0,
+                    1089.0,
+                    1403.0,
+                    1089.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    914.0,
+                    1082.0,
+                    994.0,
+                    1082.0,
+                    994.0,
+                    1129.0,
+                    914.0,
+                    1129.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1011.0,
+                    1082.0,
+                    1409.0,
+                    1082.0,
+                    1409.0,
+                    1129.0,
+                    1011.0,
+                    1129.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1122.0,
+                    433.0,
+                    1122.0,
+                    433.0,
+                    1199.0,
+                    294.0,
+                    1199.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    526.0,
+                    1122.0,
+                    1317.0,
+                    1122.0,
+                    1317.0,
+                    1199.0,
+                    526.0,
+                    1199.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1400.0,
+                    1122.0,
+                    1411.0,
+                    1122.0,
+                    1411.0,
+                    1199.0,
+                    1400.0,
+                    1199.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    286.0,
+                    962.0,
+                    916.0,
+                    962.0,
+                    916.0,
+                    1026.5,
+                    286.0,
+                    1026.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1026.0,
+                    980.5,
+                    1067.0,
+                    980.5,
+                    1067.0,
+                    1017.0,
+                    1026.0,
+                    1017.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    286.0,
+                    1078.0,
+                    925.0,
+                    1078.0,
+                    925.0,
+                    1134.0,
+                    286.0,
+                    1134.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1125.5,
+                    534.0,
+                    1125.5,
+                    534.0,
+                    1166.5,
+                    292.0,
+                    1166.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1204.0,
+                    875.0,
+                    1204.0,
+                    875.0,
+                    1251.0,
+                    292.0,
+                    1251.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1010.0,
+                    1204.0,
+                    1406.0,
+                    1204.0,
+                    1406.0,
+                    1251.0,
+                    1010.0,
+                    1251.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    290.0,
+                    1237.0,
+                    941.0,
+                    1237.0,
+                    941.0,
+                    1288.0,
+                    290.0,
+                    1288.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1207.0,
+                    1237.0,
+                    1407.0,
+                    1237.0,
+                    1407.0,
+                    1288.0,
+                    1207.0,
+                    1288.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1184.0,
+                    1280.0,
+                    1406.0,
+                    1280.0,
+                    1406.0,
+                    1323.0,
+                    1184.0,
+                    1323.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1317.0,
+                    509.0,
+                    1317.0,
+                    509.0,
+                    1372.0,
+                    292.0,
+                    1372.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    591.0,
+                    1317.0,
+                    897.0,
+                    1317.0,
+                    897.0,
+                    1372.0,
+                    591.0,
+                    1372.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    285.75,
+                    1270.5,
+                    1197.75,
+                    1270.5,
+                    1197.75,
+                    1332.0,
+                    285.75,
+                    1332.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1580.0,
+                    1408.0,
+                    1580.0,
+                    1408.0,
+                    1618.0,
+                    293.0,
+                    1618.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1613.0,
+                    1405.0,
+                    1613.0,
+                    1405.0,
+                    1647.0,
+                    294.0,
+                    1647.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1643.0,
+                    1387.0,
+                    1643.0,
+                    1387.0,
+                    1677.0,
+                    296.0,
+                    1677.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    288.0,
+                    1376.0,
+                    1141.0,
+                    1376.0,
+                    1141.0,
+                    1421.0,
+                    288.0,
+                    1421.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1198.0,
+                    1376.0,
+                    1324.0,
+                    1376.0,
+                    1324.0,
+                    1421.0,
+                    1198.0,
+                    1421.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1367.0,
+                    1376.0,
+                    1406.0,
+                    1376.0,
+                    1406.0,
+                    1421.0,
+                    1367.0,
+                    1421.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1417.0,
+                    997.0,
+                    1417.0,
+                    997.0,
+                    1452.0,
+                    295.0,
+                    1452.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1014.0,
+                    1417.0,
+                    1128.0,
+                    1417.0,
+                    1128.0,
+                    1452.0,
+                    1014.0,
+                    1452.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1168.0,
+                    1417.0,
+                    1403.0,
+                    1417.0,
+                    1403.0,
+                    1452.0,
+                    1168.0,
+                    1452.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1443.0,
+                    396.0,
+                    1443.0,
+                    396.0,
+                    1487.0,
+                    295.0,
+                    1487.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    444.0,
+                    1443.0,
+                    456.0,
+                    1443.0,
+                    456.0,
+                    1487.0,
+                    444.0,
+                    1487.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    876.0,
+                    473.0,
+                    876.0,
+                    473.0,
+                    923.0,
+                    295.0,
+                    923.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    575.0,
+                    876.0,
+                    620.0,
+                    876.0,
+                    620.0,
+                    923.0,
+                    575.0,
+                    923.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    835.0,
+                    876.0,
+                    1061.0,
+                    876.0,
+                    1061.0,
+                    923.0,
+                    835.0,
+                    923.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1137.0,
+                    876.0,
+                    1406.0,
+                    876.0,
+                    1406.0,
+                    923.0,
+                    1137.0,
+                    923.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    916.0,
+                    297.0,
+                    916.0,
+                    297.0,
+                    963.0,
+                    292.0,
+                    963.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    313.0,
+                    916.0,
+                    1134.0,
+                    916.0,
+                    1134.0,
+                    963.0,
+                    313.0,
+                    963.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1271.0,
+                    916.0,
+                    1285.0,
+                    916.0,
+                    1285.0,
+                    963.0,
+                    1271.0,
+                    963.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 2,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    296,
+                    1561,
+                    1405,
+                    1561,
+                    1405,
+                    1857,
+                    296,
+                    1857
+                ],
+                "score": 0.984
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1004,
+                    1407,
+                    1004,
+                    1407,
+                    1197,
+                    298,
+                    1197
+                ],
+                "score": 0.982
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1334,
+                    1406,
+                    1334,
+                    1406,
+                    1551,
+                    297,
+                    1551
+                ],
+                "score": 0.982
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    296,
+                    526,
+                    1405,
+                    526,
+                    1405,
+                    688,
+                    296,
+                    688
+                ],
+                "score": 0.98
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    821,
+                    1405,
+                    821,
+                    1405,
+                    967,
+                    297,
+                    967
+                ],
+                "score": 0.978
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    684,
+                    323,
+                    1010,
+                    323,
+                    1010,
+                    416,
+                    684,
+                    416
+                ],
+                "score": 0.96
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    548,
+                    710,
+                    1148,
+                    710,
+                    1148,
+                    809,
+                    548,
+                    809
+                ],
+                "score": 0.96
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    450,
+                    1214,
+                    1250,
+                    1214,
+                    1250,
+                    1316,
+                    450,
+                    1316
+                ],
+                "score": 0.955
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    227,
+                    1404,
+                    227,
+                    1404,
+                    304,
+                    298,
+                    304
+                ],
+                "score": 0.952
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    429,
+                    1402,
+                    429,
+                    1402,
+                    494,
+                    299,
+                    494
+                ],
+                "score": 0.952
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    1970,
+                    1401,
+                    1970,
+                    1401,
+                    2036,
+                    299,
+                    2036
+                ],
+                "score": 0.948
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    299,
+                    1900,
+                    908,
+                    1900,
+                    908,
+                    1938,
+                    299,
+                    1938
+                ],
+                "score": 0.936
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    298,
+                    73,
+                    816,
+                    73,
+                    816,
+                    106,
+                    298,
+                    106
+                ],
+                "score": 0.922
+            },
+            {
+                "category_id": 9,
+                "poly": [
+                    1365,
+                    353,
+                    1401,
+                    353,
+                    1401,
+                    384,
+                    1365,
+                    384
+                ],
+                "score": 0.891
+            },
+            {
+                "category_id": 9,
+                "poly": [
+                    1365,
+                    746,
+                    1401,
+                    746,
+                    1401,
+                    777,
+                    1365,
+                    777
+                ],
+                "score": 0.881
+            },
+            {
+                "category_id": 9,
+                "poly": [
+                    1365,
+                    1250,
+                    1401,
+                    1250,
+                    1401,
+                    1281,
+                    1365,
+                    1281
+                ],
+                "score": 0.879
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    841,
+                    2088,
+                    859,
+                    2088,
+                    859,
+                    2112,
+                    841,
+                    2112
+                ],
+                "score": 0.773
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    451,
+                    1212,
+                    1249,
+                    1212,
+                    1249,
+                    1316,
+                    451,
+                    1316
+                ],
+                "score": 0.95,
+                "latex": "\\mathbf { a } _ { i } ^ { ( \\ell + 1 ) } ( \\mathbf { X } , \\mathbf { X } ^ { \\prime } ) = b _ { i } ^ { ( \\ell + 1 ) } \\mathbf { 1 } + \\sum _ { j = 1 } ^ { C ^ { ( \\ell ) } } \\left( \\begin{array} { c c } { \\mathbf { W } _ { i , j } ^ { ( \\ell + 1 ) } } & { \\mathbf { 0 } } \\\\ { \\mathbf { 0 } } & { \\mathbf { W } _ { i , j } ^ { ( \\ell + 1 ) } } \\end{array} \\right) \\phi ( \\mathbf { a } _ { j } ^ { ( \\ell ) } ( \\mathbf { X } , \\mathbf { X } ^ { \\prime } ) )"
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    684,
+                    321,
+                    1016,
+                    321,
+                    1016,
+                    416,
+                    684,
+                    416
+                ],
+                "score": 0.95,
+                "latex": "\\begin{array} { r } { { \\bf a } _ { i } ^ { ( \\ell ) } ( { \\bf X } , { \\bf X } ^ { \\prime } ) = \\left( { \\bf a } _ { i } ^ { ( \\ell ) } ( { \\bf X } ) \\right) . } \\end{array}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1006,
+                    1005,
+                    1141,
+                    1005,
+                    1141,
+                    1051,
+                    1006,
+                    1051
+                ],
+                "score": 0.94,
+                "latex": "\\mathbf { a } _ { j } ^ { ( \\ell ) } ( \\mathbf { X } , \\mathbf { X } ^ { \\prime } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    702,
+                    260,
+                    802,
+                    260,
+                    802,
+                    306,
+                    702,
+                    306
+                ],
+                "score": 0.94,
+                "latex": "\\mathbf { a } _ { j } ^ { ( \\ell ) } ( \\mathbf { X } ^ { \\prime } )"
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    549,
+                    708,
+                    1149,
+                    708,
+                    1149,
+                    809,
+                    549,
+                    809
+                ],
+                "score": 0.94,
+                "latex": "\\mathbf { a } _ { i } ^ { ( 1 ) } ( \\mathbf { X } , \\mathbf { X } ^ { \\prime } ) = b _ { i } ^ { ( 1 ) } \\mathbf { 1 } + \\sum _ { i = 1 } ^ { C ^ { ( 0 ) } } \\left( \\begin{array} { c c } { \\mathbf { W } _ { i , j } ^ { ( 1 ) } } & { \\mathbf { 0 } } \\\\ { \\mathbf { 0 } } & { \\mathbf { W } _ { i , j } ^ { ( 1 ) } } \\end{array} \\right) \\left( \\begin{array} { c } { \\mathbf { x } _ { i } } \\\\ { \\mathbf { x } _ { i } ^ { \\prime } } \\end{array} \\right) ,"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    484,
+                    559,
+                    622,
+                    559,
+                    622,
+                    605,
+                    484,
+                    605
+                ],
+                "score": 0.94,
+                "latex": "\\mathbf { a } _ { j } ^ { ( 1 ) } ( \\mathbf { X } , \\mathbf { X } ^ { \\prime } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    559,
+                    260,
+                    651,
+                    260,
+                    651,
+                    306,
+                    559,
+                    306
+                ],
+                "score": 0.93,
+                "latex": "{ \\bf a } _ { j } ^ { ( \\ell ) } ( { \\bf X } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    607,
+                    1335,
+                    770,
+                    1335,
+                    770,
+                    1377,
+                    607,
+                    1377
+                ],
+                "score": 0.93,
+                "latex": "\\mathbf { a } _ { i } ^ { ( \\ell + 1 ) } ( \\mathbf { X } , \\mathbf { X } ^ { \\prime } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    799,
+                    1374,
+                    869,
+                    1374,
+                    869,
+                    1416,
+                    799,
+                    1416
+                ],
+                "score": 0.93,
+                "latex": "b _ { i } ^ { ( \\ell + 1 ) }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    471,
+                    1677,
+                    635,
+                    1677,
+                    635,
+                    1718,
+                    471,
+                    1718
+                ],
+                "score": 0.93,
+                "latex": "\\mathbf { a } _ { i } ^ { ( \\ell + 1 ) } ( \\mathbf { X } , \\mathbf { X } ^ { \\prime } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1227,
+                    1446,
+                    1319,
+                    1446,
+                    1319,
+                    1491,
+                    1227,
+                    1491
+                ],
+                "score": 0.93,
+                "latex": "\\mathbf { W } _ { i , j } ^ { ( \\ell + 1 ) }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    352,
+                    1784,
+                    522,
+                    1784,
+                    522,
+                    1829,
+                    352,
+                    1829
+                ],
+                "score": 0.93,
+                "latex": "\\phi ( \\mathbf { a } _ { j } ^ { ( \\ell ) } ( \\mathbf { X } , \\mathbf { X } ^ { \\prime } ) )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    777,
+                    1446,
+                    913,
+                    1446,
+                    913,
+                    1490,
+                    777,
+                    1490
+                ],
+                "score": 0.93,
+                "latex": "\\mathbf { a } _ { j } ^ { ( \\ell ) } ( \\mathbf { X } , \\mathbf { X } ^ { \\prime } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    297,
+                    884,
+                    435,
+                    884,
+                    435,
+                    926,
+                    297,
+                    926
+                ],
+                "score": 0.92,
+                "latex": "\\mathbf { a } _ { i } ^ { ( 1 ) } ( \\mathbf { X } , \\mathbf { X } ^ { \\prime } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    330,
+                    1633,
+                    401,
+                    1633,
+                    401,
+                    1675,
+                    330,
+                    1675
+                ],
+                "score": 0.92,
+                "latex": "b _ { i } ^ { ( \\ell + 1 ) }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    819,
+                    1047,
+                    954,
+                    1047,
+                    954,
+                    1094,
+                    819,
+                    1094
+                ],
+                "score": 0.92,
+                "latex": "\\mathbf { a } _ { j ^ { \\prime } } ^ { ( \\ell ) } ( \\mathbf { X } , \\mathbf { X } ^ { \\prime } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    434,
+                    1593,
+                    598,
+                    1593,
+                    598,
+                    1634,
+                    434,
+                    1634
+                ],
+                "score": 0.92,
+                "latex": "\\mathbf { a } _ { i } ^ { ( \\ell + 1 ) } ( \\mathbf { X } , \\mathbf { X } ^ { \\prime } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    298,
+                    1513,
+                    469,
+                    1513,
+                    469,
+                    1550,
+                    298,
+                    1550
+                ],
+                "score": 0.92,
+                "latex": "\\mathbf { A } ^ { ( \\ell + 1 ) } ( \\mathbf { X } , \\mathbf { X } ^ { \\prime } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    489,
+                    884,
+                    626,
+                    884,
+                    626,
+                    926,
+                    489,
+                    926
+                ],
+                "score": 0.92,
+                "latex": "\\mathbf { a } _ { i ^ { \\prime } } ^ { ( \\mathrm { i } ) } ( \\mathbf { X } , \\mathbf { X ^ { \\prime } } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    610,
+                    1126,
+                    773,
+                    1126,
+                    773,
+                    1166,
+                    610,
+                    1166
+                ],
+                "score": 0.92,
+                "latex": "\\mathbf { a } _ { i } ^ { ( \\ell + 1 ) } ( \\mathbf { X } , \\mathbf { X } ^ { \\prime } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    654,
+                    1593,
+                    817,
+                    1593,
+                    817,
+                    1635,
+                    654,
+                    1635
+                ],
+                "score": 0.92,
+                "latex": "\\mathbf { a } _ { i ^ { \\prime } } ^ { ( \\ell + 1 ) } ( \\mathbf { X } , \\mathbf { X } ^ { \\prime } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1308,
+                    1715,
+                    1399,
+                    1715,
+                    1399,
+                    1762,
+                    1308,
+                    1762
+                ],
+                "score": 0.92,
+                "latex": "\\mathbf { W } _ { i , j } ^ { ( \\ell + 1 ) }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    458,
+                    1635,
+                    550,
+                    1635,
+                    550,
+                    1678,
+                    458,
+                    1678
+                ],
+                "score": 0.92,
+                "latex": "\\mathbf { W } _ { i , j } ^ { ( \\ell + 1 ) }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    751,
+                    891,
+                    822,
+                    891,
+                    822,
+                    924,
+                    751,
+                    924
+                ],
+                "score": 0.92,
+                "latex": "i \\neq i ^ { \\prime }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    945,
+                    1599,
+                    1019,
+                    1599,
+                    1019,
+                    1634,
+                    945,
+                    1634
+                ],
+                "score": 0.91,
+                "latex": "i \\neq i ^ { \\prime }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1188,
+                    601,
+                    1238,
+                    601,
+                    1238,
+                    648,
+                    1188,
+                    648
+                ],
+                "score": 0.91,
+                "latex": "{ \\bf b } _ { j } ^ { ( 0 ) }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    472,
+                    926,
+                    515,
+                    926,
+                    515,
+                    966,
+                    472,
+                    966
+                ],
+                "score": 0.91,
+                "latex": "b _ { i } ^ { ( 1 ) }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    692,
+                    1636,
+                    861,
+                    1636,
+                    861,
+                    1677,
+                    692,
+                    1677
+                ],
+                "score": 0.91,
+                "latex": "\\phi ( \\mathbf { a } _ { j } ^ { ( \\ell ) } ( \\mathbf { X } , \\mathbf { X } ^ { \\prime } ) )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    692,
+                    1678,
+                    856,
+                    1678,
+                    856,
+                    1719,
+                    692,
+                    1719
+                ],
+                "score": 0.91,
+                "latex": "\\mathbf { a } _ { i ^ { \\prime } } ^ { ( \\ell + 1 ) } ( \\mathbf { X } , \\mathbf { X } ^ { \\prime } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1271,
+                    1413,
+                    1394,
+                    1413,
+                    1394,
+                    1444,
+                    1271,
+                    1444
+                ],
+                "score": 0.9,
+                "latex": "C ^ { ( \\ell ) } \\to \\infty"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    296,
+                    644,
+                    356,
+                    644,
+                    356,
+                    691,
+                    296,
+                    691
+                ],
+                "score": 0.88,
+                "latex": "\\mathbf { U } _ { j , : } ^ { ( 0 ) }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    684,
+                    924,
+                    750,
+                    924,
+                    750,
+                    969,
+                    684,
+                    969
+                ],
+                "score": 0.86,
+                "latex": "\\mathbf { W } _ { i , : } ^ { ( 1 ) }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    850,
+                    1094,
+                    979,
+                    1094,
+                    979,
+                    1125,
+                    850,
+                    1125
+                ],
+                "score": 0.86,
+                "latex": "C ^ { ( \\ell ) } \\to \\infty ^ { \\cdot }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    970,
+                    529,
+                    1006,
+                    529,
+                    1006,
+                    558,
+                    970,
+                    558
+                ],
+                "score": 0.86,
+                "latex": "\\mathbf { X } ^ { \\prime }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1072,
+                    268,
+                    1106,
+                    268,
+                    1106,
+                    297,
+                    1072,
+                    297
+                ],
+                "score": 0.85,
+                "latex": "\\mathbf { X } ^ { \\prime }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    635,
+                    1166,
+                    668,
+                    1166,
+                    668,
+                    1191,
+                    635,
+                    1191
+                ],
+                "score": 0.82,
+                "latex": "\\mathbf { X } ^ { \\prime }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    658,
+                    1098,
+                    674,
+                    1098,
+                    674,
+                    1123,
+                    658,
+                    1123
+                ],
+                "score": 0.8,
+                "latex": "\\ell"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    632,
+                    1047,
+                    768,
+                    1047,
+                    768,
+                    1094,
+                    632,
+                    1094
+                ],
+                "score": 0.79,
+                "latex": "\\mathbf { a } _ { j } ^ { ( \\ell ) } ( \\mathbf { X } , \\mathbf { X } ^ { \\prime } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    550,
+                    1051,
+                    622,
+                    1051,
+                    622,
+                    1091,
+                    550,
+                    1091
+                ],
+                "score": 0.79,
+                "latex": "j \\neq j ^ { \\prime }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    339,
+                    568,
+                    355,
+                    568,
+                    355,
+                    599,
+                    339,
+                    599
+                ],
+                "score": 0.75,
+                "latex": "j"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    557,
+                    1164,
+                    583,
+                    1164,
+                    583,
+                    1191,
+                    557,
+                    1191
+                ],
+                "score": 0.66,
+                "latex": "\\mathbf { X }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    888,
+                    530,
+                    916,
+                    530,
+                    916,
+                    557,
+                    888,
+                    557
+                ],
+                "score": 0.61,
+                "latex": "\\mathbf { X }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    995,
+                    268,
+                    1022,
+                    268,
+                    1022,
+                    296,
+                    995,
+                    296
+                ],
+                "score": 0.58,
+                "latex": "\\mathbf { X }"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1896.0,
+                    911.0,
+                    1896.0,
+                    911.0,
+                    1943.0,
+                    291.0,
+                    1943.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    72.0,
+                    818.0,
+                    72.0,
+                    818.0,
+                    108.0,
+                    295.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    838.0,
+                    2085.0,
+                    863.0,
+                    2085.0,
+                    863.0,
+                    2118.0,
+                    838.0,
+                    2118.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1560.0,
+                    1403.0,
+                    1560.0,
+                    1403.0,
+                    1600.0,
+                    293.0,
+                    1600.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1588.0,
+                    433.0,
+                    1588.0,
+                    433.0,
+                    1637.0,
+                    292.0,
+                    1637.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    818.0,
+                    1588.0,
+                    944.0,
+                    1588.0,
+                    944.0,
+                    1641.0,
+                    818.0,
+                    1641.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1020.0,
+                    1588.0,
+                    1406.0,
+                    1588.0,
+                    1406.0,
+                    1641.0,
+                    1020.0,
+                    1641.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1626.0,
+                    329.0,
+                    1626.0,
+                    329.0,
+                    1730.0,
+                    292.0,
+                    1730.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    402.0,
+                    1626.0,
+                    457.0,
+                    1626.0,
+                    457.0,
+                    1730.0,
+                    402.0,
+                    1730.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    636.0,
+                    1626.0,
+                    691.0,
+                    1626.0,
+                    691.0,
+                    1730.0,
+                    636.0,
+                    1730.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    862.0,
+                    1626.0,
+                    1415.0,
+                    1626.0,
+                    1415.0,
+                    1730.0,
+                    862.0,
+                    1730.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    286.0,
+                    1715.0,
+                    1307.0,
+                    1715.0,
+                    1307.0,
+                    1762.0,
+                    286.0,
+                    1762.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1400.0,
+                    1715.0,
+                    1412.0,
+                    1715.0,
+                    1412.0,
+                    1762.0,
+                    1400.0,
+                    1762.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1756.0,
+                    1405.0,
+                    1756.0,
+                    1405.0,
+                    1787.0,
+                    296.0,
+                    1787.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1784.0,
+                    351.0,
+                    1784.0,
+                    351.0,
+                    1830.0,
+                    291.0,
+                    1830.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    523.0,
+                    1784.0,
+                    1407.0,
+                    1784.0,
+                    1407.0,
+                    1830.0,
+                    523.0,
+                    1830.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1819.0,
+                    477.0,
+                    1819.0,
+                    477.0,
+                    1858.0,
+                    294.0,
+                    1858.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    500.0,
+                    1590.0,
+                    695.0,
+                    1590.0,
+                    695.0,
+                    1640.0,
+                    500.0,
+                    1640.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    307.0,
+                    1621.5,
+                    420.0,
+                    1621.5,
+                    420.0,
+                    1676.0,
+                    307.0,
+                    1676.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    485.25,
+                    1673.5,
+                    552.25,
+                    1673.5,
+                    552.25,
+                    1706.5,
+                    485.25,
+                    1706.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1005.0,
+                    1005.0,
+                    1005.0,
+                    1005.0,
+                    1049.0,
+                    293.0,
+                    1049.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1142.0,
+                    1005.0,
+                    1407.0,
+                    1005.0,
+                    1407.0,
+                    1049.0,
+                    1142.0,
+                    1049.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1043.0,
+                    549.0,
+                    1043.0,
+                    549.0,
+                    1097.0,
+                    291.0,
+                    1097.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    623.0,
+                    1043.0,
+                    631.0,
+                    1043.0,
+                    631.0,
+                    1097.0,
+                    623.0,
+                    1097.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    769.0,
+                    1043.0,
+                    818.0,
+                    1043.0,
+                    818.0,
+                    1097.0,
+                    769.0,
+                    1097.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    955.0,
+                    1043.0,
+                    1408.0,
+                    1043.0,
+                    1408.0,
+                    1097.0,
+                    955.0,
+                    1097.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1086.0,
+                    657.0,
+                    1086.0,
+                    657.0,
+                    1133.0,
+                    291.0,
+                    1133.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    675.0,
+                    1086.0,
+                    849.0,
+                    1086.0,
+                    849.0,
+                    1133.0,
+                    675.0,
+                    1133.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    980.0,
+                    1086.0,
+                    1408.0,
+                    1086.0,
+                    1408.0,
+                    1133.0,
+                    980.0,
+                    1133.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1127.0,
+                    609.0,
+                    1127.0,
+                    609.0,
+                    1172.0,
+                    292.0,
+                    1172.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1162.0,
+                    556.0,
+                    1162.0,
+                    556.0,
+                    1199.0,
+                    295.0,
+                    1199.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    584.0,
+                    1162.0,
+                    634.0,
+                    1162.0,
+                    634.0,
+                    1199.0,
+                    584.0,
+                    1199.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    669.0,
+                    1162.0,
+                    806.0,
+                    1162.0,
+                    806.0,
+                    1199.0,
+                    669.0,
+                    1199.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    614.0,
+                    1121.0,
+                    1410.0,
+                    1121.0,
+                    1410.0,
+                    1174.5,
+                    614.0,
+                    1174.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    290.0,
+                    1326.0,
+                    606.0,
+                    1326.0,
+                    606.0,
+                    1384.0,
+                    290.0,
+                    1384.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    771.0,
+                    1326.0,
+                    1412.0,
+                    1326.0,
+                    1412.0,
+                    1384.0,
+                    771.0,
+                    1384.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    289.0,
+                    1366.0,
+                    798.0,
+                    1366.0,
+                    798.0,
+                    1421.0,
+                    289.0,
+                    1421.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    870.0,
+                    1366.0,
+                    1411.0,
+                    1366.0,
+                    1411.0,
+                    1421.0,
+                    870.0,
+                    1421.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    289.0,
+                    1403.0,
+                    1270.0,
+                    1403.0,
+                    1270.0,
+                    1457.0,
+                    289.0,
+                    1457.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1395.0,
+                    1403.0,
+                    1412.0,
+                    1403.0,
+                    1412.0,
+                    1457.0,
+                    1395.0,
+                    1457.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    285.0,
+                    1443.0,
+                    776.0,
+                    1443.0,
+                    776.0,
+                    1492.0,
+                    285.0,
+                    1492.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    914.0,
+                    1443.0,
+                    1226.0,
+                    1443.0,
+                    1226.0,
+                    1492.0,
+                    914.0,
+                    1492.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1320.0,
+                    1443.0,
+                    1407.0,
+                    1443.0,
+                    1407.0,
+                    1492.0,
+                    1320.0,
+                    1492.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1480.0,
+                    1409.0,
+                    1480.0,
+                    1409.0,
+                    1518.0,
+                    292.0,
+                    1518.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1511.0,
+                    297.0,
+                    1511.0,
+                    297.0,
+                    1555.0,
+                    294.0,
+                    1555.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    470.0,
+                    1511.0,
+                    1202.0,
+                    1511.0,
+                    1202.0,
+                    1555.0,
+                    470.0,
+                    1555.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    524.0,
+                    887.0,
+                    524.0,
+                    887.0,
+                    566.0,
+                    294.0,
+                    566.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    917.0,
+                    524.0,
+                    969.0,
+                    524.0,
+                    969.0,
+                    566.0,
+                    917.0,
+                    566.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1007.0,
+                    524.0,
+                    1405.0,
+                    524.0,
+                    1405.0,
+                    566.0,
+                    1007.0,
+                    566.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    555.0,
+                    338.0,
+                    555.0,
+                    338.0,
+                    606.0,
+                    293.0,
+                    606.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    356.0,
+                    555.0,
+                    483.0,
+                    555.0,
+                    483.0,
+                    606.0,
+                    356.0,
+                    606.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    623.0,
+                    555.0,
+                    1408.0,
+                    555.0,
+                    1408.0,
+                    606.0,
+                    623.0,
+                    606.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    289.0,
+                    603.0,
+                    1187.0,
+                    603.0,
+                    1187.0,
+                    648.0,
+                    289.0,
+                    648.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1239.0,
+                    603.0,
+                    1405.0,
+                    603.0,
+                    1405.0,
+                    648.0,
+                    1239.0,
+                    648.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    637.0,
+                    295.0,
+                    637.0,
+                    295.0,
+                    695.0,
+                    291.0,
+                    695.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    357.0,
+                    637.0,
+                    581.0,
+                    637.0,
+                    581.0,
+                    695.0,
+                    357.0,
+                    695.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    820.0,
+                    1405.0,
+                    820.0,
+                    1405.0,
+                    859.0,
+                    295.0,
+                    859.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    852.0,
+                    1403.0,
+                    852.0,
+                    1403.0,
+                    889.0,
+                    294.0,
+                    889.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    286.0,
+                    875.0,
+                    296.0,
+                    875.0,
+                    296.0,
+                    935.0,
+                    286.0,
+                    935.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    436.0,
+                    875.0,
+                    488.0,
+                    875.0,
+                    488.0,
+                    935.0,
+                    436.0,
+                    935.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    627.0,
+                    875.0,
+                    750.0,
+                    875.0,
+                    750.0,
+                    935.0,
+                    627.0,
+                    935.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    823.0,
+                    875.0,
+                    1405.0,
+                    875.0,
+                    1405.0,
+                    935.0,
+                    823.0,
+                    935.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    921.0,
+                    471.0,
+                    921.0,
+                    471.0,
+                    970.0,
+                    291.0,
+                    970.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    516.0,
+                    921.0,
+                    683.0,
+                    921.0,
+                    683.0,
+                    970.0,
+                    516.0,
+                    970.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    751.0,
+                    921.0,
+                    979.0,
+                    921.0,
+                    979.0,
+                    970.0,
+                    751.0,
+                    970.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    224.0,
+                    1404.0,
+                    224.0,
+                    1404.0,
+                    268.0,
+                    295.0,
+                    268.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    260.0,
+                    558.0,
+                    260.0,
+                    558.0,
+                    307.0,
+                    292.0,
+                    307.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    652.0,
+                    260.0,
+                    701.0,
+                    260.0,
+                    701.0,
+                    307.0,
+                    652.0,
+                    307.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    803.0,
+                    260.0,
+                    994.0,
+                    260.0,
+                    994.0,
+                    307.0,
+                    803.0,
+                    307.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1023.0,
+                    260.0,
+                    1071.0,
+                    260.0,
+                    1071.0,
+                    307.0,
+                    1023.0,
+                    307.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1107.0,
+                    260.0,
+                    1120.0,
+                    260.0,
+                    1120.0,
+                    307.0,
+                    1107.0,
+                    307.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    427.0,
+                    1406.0,
+                    427.0,
+                    1406.0,
+                    467.0,
+                    294.0,
+                    467.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    460.0,
+                    575.0,
+                    460.0,
+                    575.0,
+                    496.0,
+                    295.0,
+                    496.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    1973.0,
+                    1402.0,
+                    1973.0,
+                    1402.0,
+                    2006.0,
+                    297.0,
+                    2006.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1999.0,
+                    1406.0,
+                    1999.0,
+                    1406.0,
+                    2041.0,
+                    293.0,
+                    2041.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 3,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    525,
+                    1405,
+                    525,
+                    1405,
+                    650,
+                    298,
+                    650
+                ],
+                "score": 0.977
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    296,
+                    225,
+                    1405,
+                    225,
+                    1405,
+                    436,
+                    296,
+                    436
+                ],
+                "score": 0.977
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    767,
+                    1406,
+                    767,
+                    1406,
+                    865,
+                    299,
+                    865
+                ],
+                "score": 0.975
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1122,
+                    1406,
+                    1122,
+                    1406,
+                    1361,
+                    297,
+                    1361
+                ],
+                "score": 0.974
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    295,
+                    1371,
+                    1405,
+                    1371,
+                    1405,
+                    1563,
+                    295,
+                    1563
+                ],
+                "score": 0.97
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    346,
+                    1881,
+                    1325,
+                    1881,
+                    1325,
+                    2043,
+                    346,
+                    2043
+                ],
+                "score": 0.963
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    380,
+                    1644,
+                    1321,
+                    1644,
+                    1321,
+                    1805,
+                    380,
+                    1805
+                ],
+                "score": 0.961
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    526,
+                    657,
+                    1171,
+                    657,
+                    1171,
+                    757,
+                    526,
+                    757
+                ],
+                "score": 0.959
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    423,
+                    918,
+                    1272,
+                    918,
+                    1272,
+                    1017,
+                    423,
+                    1017
+                ],
+                "score": 0.958
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    301,
+                    1025,
+                    1398,
+                    1025,
+                    1398,
+                    1112,
+                    301,
+                    1112
+                ],
+                "score": 0.952
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    292,
+                    1575,
+                    1402,
+                    1575,
+                    1402,
+                    1640,
+                    292,
+                    1640
+                ],
+                "score": 0.946
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1812,
+                    1409,
+                    1812,
+                    1409,
+                    1874,
+                    298,
+                    1874
+                ],
+                "score": 0.946
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    879,
+                    779,
+                    879,
+                    779,
+                    911,
+                    299,
+                    911
+                ],
+                "score": 0.924
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    301,
+                    469,
+                    704,
+                    469,
+                    704,
+                    501,
+                    301,
+                    501
+                ],
+                "score": 0.914
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    299,
+                    75,
+                    816,
+                    75,
+                    816,
+                    105,
+                    299,
+                    105
+                ],
+                "score": 0.911
+            },
+            {
+                "category_id": 9,
+                "poly": [
+                    1366,
+                    1712,
+                    1401,
+                    1712,
+                    1401,
+                    1741,
+                    1366,
+                    1741
+                ],
+                "score": 0.851
+            },
+            {
+                "category_id": 9,
+                "poly": [
+                    1365,
+                    1947,
+                    1401,
+                    1947,
+                    1401,
+                    1976,
+                    1365,
+                    1976
+                ],
+                "score": 0.845
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    840,
+                    2087,
+                    859,
+                    2087,
+                    859,
+                    2112,
+                    840,
+                    2112
+                ],
+                "score": 0.761
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    526,
+                    655,
+                    1174,
+                    655,
+                    1174,
+                    760,
+                    526,
+                    760
+                ],
+                "score": 0.95,
+                "latex": "A _ { i , \\mu } ^ { ( \\ell + 1 ) } ( \\mathbf { X } ) = b _ { i } ^ { ( \\ell + 1 ) } + \\sum _ { j = 1 } ^ { C ^ { ( \\ell ) } } \\sum _ { \\nu = 1 } ^ { H ^ { ( \\ell ) } D ^ { ( \\ell ) } } W _ { i , j , \\mu , \\nu } ^ { ( \\ell + 1 ) } \\phi ( A _ { j , \\nu } ^ { ( \\ell ) } ( \\mathbf { X } ) ) ."
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    426,
+                    916,
+                    1270,
+                    916,
+                    1270,
+                    1020,
+                    426,
+                    1020
+                ],
+                "score": 0.94,
+                "latex": "\\mathbb { E } \\left[ A _ { i , \\mu } ^ { ( \\ell + 1 ) } ( \\mathbf { X } ) \\right] = \\mathbb { E } \\left[ b _ { i } ^ { ( \\ell + 1 ) } \\right] + \\sum _ { j = 1 } ^ { C ^ { ( \\ell ) } } \\sum _ { \\nu = 1 } ^ { H ^ { ( \\ell ) } D ^ { ( \\ell ) } } \\mathbb { E } \\left[ W _ { i , j , \\mu , \\nu } ^ { ( \\ell + 1 ) } \\phi ( A _ { j , \\nu } ^ { ( \\ell ) } ( \\mathbf { X } ) ) \\right] = 0 ."
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    671,
+                    1184,
+                    989,
+                    1184,
+                    989,
+                    1239,
+                    671,
+                    1239
+                ],
+                "score": 0.94,
+                "latex": "\\mathbb { C } \\left[ A _ { i , \\mu } ^ { ( \\ell + 1 ) } ( \\mathbf { \\bar { X } } ) , A _ { i , \\mu ^ { \\prime } } ^ { ( \\ell + 1 ) } ( \\mathbf { X } ^ { \\prime } ) \\right]"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    771,
+                    1026,
+                    860,
+                    1026,
+                    860,
+                    1072,
+                    771,
+                    1072
+                ],
+                "score": 0.94,
+                "latex": "W _ { i , j , \\nu , \\mu } ^ { ( \\ell + 1 ) }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    738,
+                    385,
+                    830,
+                    385,
+                    830,
+                    432,
+                    738,
+                    432
+                ],
+                "score": 0.93,
+                "latex": "{ \\bf a } _ { j } ^ { ( \\ell ) } ( { \\bf X } )"
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    340,
+                    1876,
+                    1328,
+                    1876,
+                    1328,
+                    2047,
+                    340,
+                    2047
+                ],
+                "score": 0.93,
+                "latex": "\\begin{array} { r l } { \\mathbb { C } \\left[ A _ { i , \\mu } ^ { ( \\ell + 1 ) } ( \\mathbf { X } ) , A _ { i , \\mu } ^ { ( \\ell + 1 ) } ( \\mathbf { X } ^ { \\prime } ) \\right] = \\sigma _ { \\mathrm { b } } ^ { 2 } + } & { } \\\\ { \\quad } & { \\qquad \\sum _ { j = 1 } ^ { { \\ell ^ { ( \\ell ) } } } \\displaystyle \\sum _ { j ^ { \\prime } = 1 } ^ { { \\ell ^ { ( \\ell ) } } } \\sum _ { \\nu = 1 } ^ { { H ^ { ( \\ell ) } } D ^ { ( \\ell ) } } \\sum _ { \\nu ^ { \\prime } = 1 } ^ { H ^ { ( \\ell ) } D ^ { ( \\ell ) } } \\mathbb { E } \\left[ W _ { i , j , \\mu , \\nu } ^ { ( \\ell + 1 ) } W _ { i , j ^ { \\prime } , \\mu , \\nu ^ { \\prime } } ^ { ( \\ell + 1 ) } \\right] \\mathbb { E } \\left[ \\phi ( A _ { j , \\nu } ^ { ( \\ell ) } ( \\mathbf { X } ) ) \\phi ( A _ { j ^ { \\prime } , \\nu ^ { \\prime } } ^ { ( \\ell ) } ( \\mathbf { X } ^ { \\prime } ) ) \\right] . } \\end{array}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1045,
+                    1270,
+                    1361,
+                    1270,
+                    1361,
+                    1326,
+                    1045,
+                    1326
+                ],
+                "score": 0.93,
+                "latex": "\\mathbb { C } \\left[ A _ { i , \\mu } ^ { ( \\ell + 1 ) } ( \\mathbf { X } ) , A _ { i , \\mu } ^ { ( \\ell + 1 ) } ( \\mathbf { X } ^ { \\prime } ) \\right]"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    880,
+                    385,
+                    980,
+                    385,
+                    980,
+                    432,
+                    880,
+                    432
+                ],
+                "score": 0.93,
+                "latex": "\\mathbf { a } _ { j } ^ { ( \\ell ) } ( \\mathbf { X } ^ { \\prime } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    327,
+                    1025,
+                    396,
+                    1025,
+                    396,
+                    1068,
+                    327,
+                    1068
+                ],
+                "score": 0.93,
+                "latex": "b _ { i } ^ { ( \\ell + 1 ) }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    750,
+                    799,
+                    1067,
+                    799,
+                    1067,
+                    836,
+                    750,
+                    836
+                ],
+                "score": 0.93,
+                "latex": "\\mu \\in \\{ 1 , \\ldots , H ^ { ( \\ell + 1 ) } D ^ { ( \\ell + 1 ) } \\}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    366,
+                    1070,
+                    499,
+                    1070,
+                    499,
+                    1115,
+                    366,
+                    1115
+                ],
+                "score": 0.93,
+                "latex": "\\phi ( A _ { j , \\nu } ^ { ( \\ell ) } ( \\mathbf { X } ) )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1036,
+                    382,
+                    1383,
+                    382,
+                    1383,
+                    440,
+                    1036,
+                    440
+                ],
+                "score": 0.92,
+                "latex": "\\mathrm { d i a g } \\left( \\mathbb { C } \\left[ \\mathbf { a } _ { j } ^ { ( \\ell ) } ( \\mathbf { X } ) , \\mathbf { a } _ { j } ^ { ( \\ell ) } ( \\mathbf { X } ^ { \\prime } ) \\right] \\right) )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    446,
+                    1026,
+                    537,
+                    1026,
+                    537,
+                    1071,
+                    446,
+                    1071
+                ],
+                "score": 0.92,
+                "latex": "W _ { i , j , \\mu , \\nu } ^ { ( \\ell + 1 ) }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1094,
+                    765,
+                    1320,
+                    765,
+                    1320,
+                    802,
+                    1094,
+                    802
+                ],
+                "score": 0.92,
+                "latex": "i \\in \\{ 1 , \\ldots , C ^ { ( \\ell + 1 ) } \\}"
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    378,
+                    1641,
+                    1326,
+                    1641,
+                    1326,
+                    1810,
+                    378,
+                    1810
+                ],
+                "score": 0.92,
+                "latex": "\\begin{array} { r l } { \\mathbb { C } \\left[ A _ { i , \\mu } ^ { ( \\ell + 1 ) } ( \\mathbf { X } ) , A _ { i , \\mu } ^ { ( \\ell + 1 ) } ( \\mathbf { X } ^ { \\prime } ) \\right] = \\mathbb { V } \\left[ b _ { i } ^ { ( \\ell ) } \\right] + } & { } \\\\ { \\displaystyle \\sum _ { j = 1 } ^ { C ^ { ( \\ell ) } } \\sum _ { j ^ { \\prime } = 1 } ^ { C ^ { ( \\ell ) } } \\sum _ { \\nu = 1 } ^ { H ^ { ( \\ell ) } D ^ { ( \\ell ) } } \\sum _ { \\nu ^ { \\prime } = 1 } ^ { H ^ { ( \\ell ) } D ^ { ( \\ell ) } } \\mathbb { C } \\left[ W _ { i , j , \\mu , \\nu } ^ { ( \\ell + 1 ) } \\phi ( A _ { j , \\nu } ^ { ( \\ell ) } ( \\mathbf { X } ) ) , W _ { i , j ^ { \\prime } , \\mu , \\nu ^ { \\prime } } ^ { ( \\ell + 1 ) } \\phi ( A _ { j ^ { \\prime } , \\nu ^ { \\prime } } ^ { ( \\ell ) } ( \\mathbf { X } ^ { \\prime } ) ) \\right] . } \\end{array}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    807,
+                    1404,
+                    1070,
+                    1404,
+                    1070,
+                    1437,
+                    807,
+                    1437
+                ],
+                "score": 0.92,
+                "latex": "H ^ { ( L + 1 ) } = D ^ { ( L + 1 ) } = 1"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    297,
+                    223,
+                    388,
+                    223,
+                    388,
+                    266,
+                    297,
+                    266
+                ],
+                "score": 0.92,
+                "latex": "{ \\bf a } _ { i } ^ { ( \\ell ) } ( { \\bf X } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    635,
+                    1373,
+                    720,
+                    1373,
+                    720,
+                    1407,
+                    635,
+                    1407
+                ],
+                "score": 0.91,
+                "latex": "( L + 1 )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1035,
+                    1189,
+                    1400,
+                    1189,
+                    1400,
+                    1229,
+                    1035,
+                    1229
+                ],
+                "score": 0.91,
+                "latex": "\\mu , \\mu ^ { \\prime } \\in \\{ 1 , \\ldots , H ^ { ( \\ell + 1 ) } D ^ { ( \\ell + 1 ) } \\} )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    437,
+                    769,
+                    498,
+                    769,
+                    498,
+                    798,
+                    437,
+                    798
+                ],
+                "score": 0.9,
+                "latex": "\\ell + 1"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    297,
+                    1321,
+                    622,
+                    1321,
+                    622,
+                    1361,
+                    297,
+                    1361
+                ],
+                "score": 0.89,
+                "latex": "\\mu \\in \\{ 1 , \\ldots , H ^ { ( \\ell + 1 ) } D ^ { ( \\ell + 1 ) } \\} )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    822,
+                    1236,
+                    1056,
+                    1236,
+                    1056,
+                    1275,
+                    822,
+                    1275
+                ],
+                "score": 0.89,
+                "latex": "N ^ { 2 } ( H ^ { ( \\ell + 1 ) } D ^ { \\overline { { ( \\ell + 1 ) } } } ) ^ { 2 }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1135,
+                    1323,
+                    1357,
+                    1323,
+                    1357,
+                    1360,
+                    1135,
+                    1360
+                ],
+                "score": 0.88,
+                "latex": "N ^ { 2 } ( H ^ { ( \\ell + 1 ) } D ^ { ( \\ell + 1 ) } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1307,
+                    558,
+                    1341,
+                    558,
+                    1341,
+                    587,
+                    1307,
+                    587
+                ],
+                "score": 0.85,
+                "latex": "\\mathbf { X } ^ { \\prime }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1027,
+                    771,
+                    1042,
+                    771,
+                    1042,
+                    800,
+                    1027,
+                    800
+                ],
+                "score": 0.79,
+                "latex": "j"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    371,
+                    770,
+                    387,
+                    770,
+                    387,
+                    796,
+                    371,
+                    796
+                ],
+                "score": 0.77,
+                "latex": "\\ell"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    681,
+                    807,
+                    700,
+                    807,
+                    700,
+                    830,
+                    681,
+                    830
+                ],
+                "score": 0.76,
+                "latex": "\\nu"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1225,
+                    559,
+                    1253,
+                    559,
+                    1253,
+                    587,
+                    1225,
+                    587
+                ],
+                "score": 0.55,
+                "latex": "\\mathbf { X }"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    469.0,
+                    708.0,
+                    469.0,
+                    708.0,
+                    505.0,
+                    294.0,
+                    505.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    73.0,
+                    817.0,
+                    73.0,
+                    817.0,
+                    108.0,
+                    295.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    839.0,
+                    2086.0,
+                    861.0,
+                    2086.0,
+                    861.0,
+                    2118.0,
+                    839.0,
+                    2118.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    524.0,
+                    1405.0,
+                    524.0,
+                    1405.0,
+                    564.0,
+                    292.0,
+                    564.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    557.0,
+                    1224.0,
+                    557.0,
+                    1224.0,
+                    592.0,
+                    291.0,
+                    592.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1254.0,
+                    557.0,
+                    1306.0,
+                    557.0,
+                    1306.0,
+                    592.0,
+                    1254.0,
+                    592.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1342.0,
+                    557.0,
+                    1407.0,
+                    557.0,
+                    1407.0,
+                    592.0,
+                    1342.0,
+                    592.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    587.0,
+                    1405.0,
+                    587.0,
+                    1405.0,
+                    625.0,
+                    292.0,
+                    625.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    619.0,
+                    674.0,
+                    619.0,
+                    674.0,
+                    653.0,
+                    295.0,
+                    653.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    221.0,
+                    296.0,
+                    221.0,
+                    296.0,
+                    271.0,
+                    293.0,
+                    271.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    389.0,
+                    221.0,
+                    1406.0,
+                    221.0,
+                    1406.0,
+                    271.0,
+                    389.0,
+                    271.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    261.0,
+                    1403.0,
+                    261.0,
+                    1403.0,
+                    298.0,
+                    293.0,
+                    298.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    292.0,
+                    1408.0,
+                    292.0,
+                    1408.0,
+                    329.0,
+                    293.0,
+                    329.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    321.0,
+                    1405.0,
+                    321.0,
+                    1405.0,
+                    357.0,
+                    291.0,
+                    357.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    354.0,
+                    1403.0,
+                    354.0,
+                    1403.0,
+                    386.0,
+                    294.0,
+                    386.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    285.0,
+                    379.0,
+                    737.0,
+                    379.0,
+                    737.0,
+                    443.0,
+                    285.0,
+                    443.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    831.0,
+                    379.0,
+                    879.0,
+                    379.0,
+                    879.0,
+                    443.0,
+                    831.0,
+                    443.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    981.0,
+                    379.0,
+                    1035.0,
+                    379.0,
+                    1035.0,
+                    443.0,
+                    981.0,
+                    443.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1384.0,
+                    379.0,
+                    1398.0,
+                    379.0,
+                    1398.0,
+                    443.0,
+                    1384.0,
+                    443.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    764.0,
+                    370.0,
+                    764.0,
+                    370.0,
+                    805.0,
+                    293.0,
+                    805.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    388.0,
+                    764.0,
+                    436.0,
+                    764.0,
+                    436.0,
+                    805.0,
+                    388.0,
+                    805.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    499.0,
+                    764.0,
+                    1026.0,
+                    764.0,
+                    1026.0,
+                    805.0,
+                    499.0,
+                    805.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1043.0,
+                    764.0,
+                    1093.0,
+                    764.0,
+                    1093.0,
+                    805.0,
+                    1043.0,
+                    805.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1321.0,
+                    764.0,
+                    1406.0,
+                    764.0,
+                    1406.0,
+                    805.0,
+                    1321.0,
+                    805.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    794.0,
+                    680.0,
+                    794.0,
+                    680.0,
+                    840.0,
+                    292.0,
+                    840.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    701.0,
+                    794.0,
+                    749.0,
+                    794.0,
+                    749.0,
+                    840.0,
+                    701.0,
+                    840.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1068.0,
+                    794.0,
+                    1407.0,
+                    794.0,
+                    1407.0,
+                    840.0,
+                    1068.0,
+                    840.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    831.0,
+                    767.0,
+                    831.0,
+                    767.0,
+                    870.0,
+                    294.0,
+                    870.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1118.0,
+                    1407.0,
+                    1118.0,
+                    1407.0,
+                    1162.0,
+                    292.0,
+                    1162.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1153.0,
+                    670.0,
+                    1153.0,
+                    670.0,
+                    1257.0,
+                    294.0,
+                    1257.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    990.0,
+                    1153.0,
+                    1034.0,
+                    1153.0,
+                    1034.0,
+                    1257.0,
+                    990.0,
+                    1257.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1401.0,
+                    1153.0,
+                    1421.0,
+                    1153.0,
+                    1421.0,
+                    1257.0,
+                    1401.0,
+                    1257.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1230.0,
+                    821.0,
+                    1230.0,
+                    821.0,
+                    1280.0,
+                    291.0,
+                    1280.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1057.0,
+                    1230.0,
+                    1409.0,
+                    1230.0,
+                    1409.0,
+                    1280.0,
+                    1057.0,
+                    1280.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1255.0,
+                    1044.0,
+                    1255.0,
+                    1044.0,
+                    1342.0,
+                    294.0,
+                    1342.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1362.0,
+                    1255.0,
+                    1407.0,
+                    1255.0,
+                    1407.0,
+                    1342.0,
+                    1362.0,
+                    1342.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    289.0,
+                    1318.0,
+                    296.0,
+                    1318.0,
+                    296.0,
+                    1365.0,
+                    289.0,
+                    1365.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    623.0,
+                    1318.0,
+                    1134.0,
+                    1318.0,
+                    1134.0,
+                    1365.0,
+                    623.0,
+                    1365.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1358.0,
+                    1318.0,
+                    1370.0,
+                    1318.0,
+                    1370.0,
+                    1365.0,
+                    1358.0,
+                    1365.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1371.0,
+                    634.0,
+                    1371.0,
+                    634.0,
+                    1410.0,
+                    294.0,
+                    1410.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    721.0,
+                    1371.0,
+                    1405.0,
+                    1371.0,
+                    1405.0,
+                    1410.0,
+                    721.0,
+                    1410.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1402.0,
+                    806.0,
+                    1402.0,
+                    806.0,
+                    1443.0,
+                    292.0,
+                    1443.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1071.0,
+                    1402.0,
+                    1407.0,
+                    1402.0,
+                    1407.0,
+                    1443.0,
+                    1071.0,
+                    1443.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1438.0,
+                    1406.0,
+                    1438.0,
+                    1406.0,
+                    1474.0,
+                    294.0,
+                    1474.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    1471.0,
+                    1403.0,
+                    1471.0,
+                    1403.0,
+                    1503.0,
+                    297.0,
+                    1503.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1501.0,
+                    1405.0,
+                    1501.0,
+                    1405.0,
+                    1533.0,
+                    295.0,
+                    1533.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1531.0,
+                    787.0,
+                    1531.0,
+                    787.0,
+                    1562.0,
+                    293.0,
+                    1562.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    465.0,
+                    1044.0,
+                    546.0,
+                    1044.0,
+                    546.0,
+                    1078.0,
+                    465.0,
+                    1078.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    861.0,
+                    1031.0,
+                    1404.0,
+                    1031.0,
+                    1404.0,
+                    1077.0,
+                    861.0,
+                    1077.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    286.0,
+                    1059.0,
+                    365.0,
+                    1059.0,
+                    365.0,
+                    1125.0,
+                    286.0,
+                    1125.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    500.0,
+                    1059.0,
+                    522.0,
+                    1059.0,
+                    522.0,
+                    1125.0,
+                    500.0,
+                    1125.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    279.0,
+                    1017.0,
+                    415.0,
+                    1017.0,
+                    415.0,
+                    1072.5,
+                    279.0,
+                    1072.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    384.25,
+                    1020.0,
+                    548.25,
+                    1020.0,
+                    548.25,
+                    1070.0,
+                    384.25,
+                    1070.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    528.0,
+                    1020.0,
+                    871.0,
+                    1020.0,
+                    871.0,
+                    1072.5,
+                    528.0,
+                    1072.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1575.0,
+                    1405.0,
+                    1575.0,
+                    1405.0,
+                    1611.0,
+                    295.0,
+                    1611.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1605.0,
+                    766.0,
+                    1605.0,
+                    766.0,
+                    1644.0,
+                    294.0,
+                    1644.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1811.0,
+                    1403.0,
+                    1811.0,
+                    1403.0,
+                    1847.0,
+                    295.0,
+                    1847.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1842.0,
+                    549.0,
+                    1842.0,
+                    549.0,
+                    1878.0,
+                    293.0,
+                    1878.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    874.0,
+                    779.0,
+                    874.0,
+                    779.0,
+                    918.0,
+                    295.0,
+                    918.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 4,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    295,
+                    674,
+                    1407,
+                    674,
+                    1407,
+                    858,
+                    295,
+                    858
+                ],
+                "score": 0.977
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    296,
+                    1638,
+                    1406,
+                    1638,
+                    1406,
+                    1774,
+                    296,
+                    1774
+                ],
+                "score": 0.975
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    1015,
+                    1405,
+                    1015,
+                    1405,
+                    1130,
+                    299,
+                    1130
+                ],
+                "score": 0.974
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    421,
+                    1148,
+                    1309,
+                    1148,
+                    1309,
+                    1247,
+                    421,
+                    1247
+                ],
+                "score": 0.953
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    294,
+                    1784,
+                    1407,
+                    1784,
+                    1407,
+                    1849,
+                    294,
+                    1849
+                ],
+                "score": 0.941
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    596,
+                    1446,
+                    1102,
+                    1446,
+                    1102,
+                    1505,
+                    596,
+                    1505
+                ],
+                "score": 0.938
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    1511,
+                    1167,
+                    1511,
+                    1167,
+                    1545,
+                    299,
+                    1545
+                ],
+                "score": 0.933
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1270,
+                    752,
+                    1270,
+                    752,
+                    1304,
+                    298,
+                    1304
+                ],
+                "score": 0.927
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    345,
+                    1325,
+                    1268,
+                    1325,
+                    1268,
+                    1407,
+                    345,
+                    1407
+                ],
+                "score": 0.917
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    299,
+                    74,
+                    816,
+                    74,
+                    816,
+                    106,
+                    299,
+                    106
+                ],
+                "score": 0.915
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1424,
+                    371,
+                    1424,
+                    371,
+                    1454,
+                    297,
+                    1454
+                ],
+                "score": 0.913
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    307,
+                    270,
+                    1406,
+                    270,
+                    1406,
+                    554,
+                    307,
+                    554
+                ],
+                "score": 0.911
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    301,
+                    1582,
+                    759,
+                    1582,
+                    759,
+                    1615,
+                    301,
+                    1615
+                ],
+                "score": 0.909
+            },
+            {
+                "category_id": 9,
+                "poly": [
+                    1351,
+                    1341,
+                    1401,
+                    1341,
+                    1401,
+                    1373,
+                    1351,
+                    1373
+                ],
+                "score": 0.896
+            },
+            {
+                "category_id": 9,
+                "poly": [
+                    1351,
+                    1460,
+                    1401,
+                    1460,
+                    1401,
+                    1492,
+                    1351,
+                    1492
+                ],
+                "score": 0.891
+            },
+            {
+                "category_id": 9,
+                "poly": [
+                    1351,
+                    1181,
+                    1401,
+                    1181,
+                    1401,
+                    1213,
+                    1351,
+                    1213
+                ],
+                "score": 0.877
+            },
+            {
+                "category_id": 9,
+                "poly": [
+                    1351,
+                    1907,
+                    1401,
+                    1907,
+                    1401,
+                    1939,
+                    1351,
+                    1939
+                ],
+                "score": 0.876
+            },
+            {
+                "category_id": 9,
+                "poly": [
+                    1365,
+                    970,
+                    1401,
+                    970,
+                    1401,
+                    1000,
+                    1365,
+                    1000
+                ],
+                "score": 0.853
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    839,
+                    2088,
+                    859,
+                    2088,
+                    859,
+                    2113,
+                    839,
+                    2113
+                ],
+                "score": 0.815
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    302,
+                    230,
+                    787,
+                    230,
+                    787,
+                    264,
+                    302,
+                    264
+                ],
+                "score": 0.719
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    308,
+                    551,
+                    729,
+                    551,
+                    729,
+                    613,
+                    308,
+                    613
+                ],
+                "score": 0.651
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    296,
+                    873,
+                    1411,
+                    873,
+                    1411,
+                    973,
+                    296,
+                    973
+                ],
+                "score": 0.573
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    366,
+                    1866,
+                    1279,
+                    1866,
+                    1279,
+                    2044,
+                    366,
+                    2044
+                ],
+                "score": 0.497
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    417,
+                    1867,
+                    1279,
+                    1867,
+                    1279,
+                    1956,
+                    417,
+                    1956
+                ],
+                "score": 0.476
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    296,
+                    1971,
+                    1045,
+                    1971,
+                    1045,
+                    2043,
+                    296,
+                    2043
+                ],
+                "score": 0.47
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    296,
+                    873,
+                    1411,
+                    873,
+                    1411,
+                    973,
+                    296,
+                    973
+                ],
+                "score": 0.388
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    302,
+                    230,
+                    787,
+                    230,
+                    787,
+                    264,
+                    302,
+                    264
+                ],
+                "score": 0.213
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    847,
+                    407,
+                    990,
+                    407,
+                    990,
+                    449,
+                    847,
+                    449
+                ],
+                "score": 0.94,
+                "latex": "V _ { \\mu } ^ { ( \\ell ) } ( { \\bf X } , { \\bf X ^ { \\prime } } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    471,
+                    1015,
+                    562,
+                    1015,
+                    562,
+                    1061,
+                    471,
+                    1061
+                ],
+                "score": 0.94,
+                "latex": "\\mathbf { W } _ { i , j } ^ { ( \\ell + 1 ) }"
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    421,
+                    1144,
+                    1309,
+                    1144,
+                    1309,
+                    1248,
+                    421,
+                    1248
+                ],
+                "score": 0.94,
+                "latex": "K _ { \\mu } ^ { ( 1 ) } ( \\mathbf { X } , \\mathbf { X ^ { \\prime } } ) = \\mathbb { C } \\left[ A _ { i , \\mu } ^ { ( 1 ) } ( \\mathbf { X } ) , A _ { i , \\mu } ^ { ( 1 ) } ( \\mathbf { X ^ { \\prime } } ) \\right] = \\sigma _ { \\mathrm { b } } ^ { 2 } + \\frac { \\sigma _ { \\mathrm { w } } ^ { 2 } } { C ^ { ( 0 ) } } \\sum _ { i = 1 } ^ { C ^ { ( 0 ) } } \\sum _ { \\nu \\in \\mu \\mathrm { t h } \\mathrm { p a t c h } } X _ { i , \\nu } X _ { i , \\nu } ^ { \\prime } ."
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    596,
+                    1446,
+                    1104,
+                    1446,
+                    1104,
+                    1506,
+                    596,
+                    1506
+                ],
+                "score": 0.94,
+                "latex": "V _ { \\nu } ^ { ( \\ell ) } ( \\mathbf { X } , \\mathbf { X } ^ { \\prime } ) = \\mathbb { E } \\left[ \\phi ( A _ { j , \\nu } ^ { ( \\ell ) } ( \\mathbf { X } ) ) \\phi ( A _ { j , \\nu } ^ { ( \\ell ) } ( \\mathbf { X } ^ { \\prime } ) ) \\right]"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    866,
+                    770,
+                    1164,
+                    770,
+                    1164,
+                    827,
+                    866,
+                    827
+                ],
+                "score": 0.94,
+                "latex": "\\mathbb { E } \\left[ W _ { i , j , \\mu , \\nu } ^ { ( \\ell + 1 ) } W _ { i , j ^ { \\prime } , \\mu , \\nu ^ { \\prime } } ^ { ( \\ell + 1 ) } \\right] = 0"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    899,
+                    674,
+                    993,
+                    674,
+                    993,
+                    721,
+                    899,
+                    721
+                ],
+                "score": 0.93,
+                "latex": "\\mathbf { W } _ { i , j } ^ { ( \\ell + 1 ) }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1049,
+                    674,
+                    1142,
+                    674,
+                    1142,
+                    723,
+                    1049,
+                    723
+                ],
+                "score": 0.93,
+                "latex": "\\mathbf { W } _ { i , j ^ { \\prime } } ^ { ( \\ell + 1 ) }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    701,
+                    1732,
+                    881,
+                    1732,
+                    881,
+                    1773,
+                    701,
+                    1773
+                ],
+                "score": 0.93,
+                "latex": "K _ { \\mu } ^ { ( \\ell + 1 ) } ( { \\mathbf { X } } ^ { \\prime } , { \\mathbf { X } } ^ { \\prime } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1248,
+                    721,
+                    1341,
+                    721,
+                    1341,
+                    768,
+                    1248,
+                    768
+                ],
+                "score": 0.93,
+                "latex": "\\mathbf { W } _ { i , j } ^ { ( \\ell + 1 ) }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    295,
+                    718,
+                    608,
+                    718,
+                    608,
+                    776,
+                    295,
+                    776
+                ],
+                "score": 0.93,
+                "latex": "\\mathbb { E } \\left[ W _ { i , j , \\mu , \\nu } ^ { ( \\ell + 1 ) } W _ { i , j ^ { \\prime } , \\mu , \\nu ^ { \\prime } } ^ { ( \\ell + 1 ) } \\right] = 0"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    882,
+                    1058,
+                    1018,
+                    1058,
+                    1018,
+                    1100,
+                    882,
+                    1100
+                ],
+                "score": 0.93,
+                "latex": "v _ { g } ^ { ( 1 ) } ( \\mathbf { X } , \\mathbf { X } ^ { \\prime } )"
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    310,
+                    870,
+                    1418,
+                    870,
+                    1418,
+                    975,
+                    310,
+                    975
+                ],
+                "score": 0.93,
+                "latex": "\\Sigma \\left[ A _ { i , \\mu } ^ { ( \\ell + 1 ) } ( \\mathbf { X } ) , A _ { i , \\mu } ^ { ( \\ell + 1 ) } ( \\mathbf { X ^ { \\prime } } ) \\right] = \\sigma _ { \\mathrm { b } } ^ { ( \\ell ) } + \\sum _ { j = 1 } ^ { Q ^ { ( \\ell ) } } \\sum _ { \\nu = 1 } ^ { H ^ { ( \\ell ) } D ^ { ( \\ell ) } } \\mathbb { E } \\left[ W _ { i , j , \\mu , \\nu } ^ { ( \\ell + 1 ) } W _ { i , j , \\mu , \\nu } ^ { ( \\ell + 1 ) } \\right] \\mathbb { E } \\left[ \\phi ( A _ { j , \\nu } ^ { ( \\ell ) } ( \\mathbf { X } ) ) \\phi ( A _ { j , \\nu } ^ { ( \\ell ) } ( \\mathbf { X ^ { \\prime } } ) ) \\right] ."
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    343,
+                    1326,
+                    1264,
+                    1326,
+                    1264,
+                    1407,
+                    343,
+                    1407
+                ],
+                "score": 0.92,
+                "latex": "K _ { \\mu } ^ { ( \\ell + 1 ) } ( \\mathbf { X } , \\mathbf { X } ^ { \\prime } ) = \\mathbb { C } \\left[ A _ { i , \\mu } ^ { ( \\ell + 1 ) } ( \\mathbf { X } ) , A _ { i , \\mu } ^ { ( \\ell + 1 ) } ( \\mathbf { X } ^ { \\prime } ) \\right] = \\sigma _ { \\mathrm { b } } ^ { 2 } + \\sigma _ { \\mathrm { w } } ^ { 2 } \\sum _ { \\nu \\in \\mu \\mathrm { t h p a t c h } } V _ { \\nu } ^ { ( \\ell ) } ( \\mathbf { X } , \\mathbf { X } ^ { \\prime } ) ,"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    830,
+                    309,
+                    985,
+                    309,
+                    985,
+                    351,
+                    830,
+                    351
+                ],
+                "score": 0.92,
+                "latex": "K _ { \\mu } ^ { ( 1 ) } ( \\mathbf { X } ^ { \\prime } , \\mathbf { X } ^ { \\prime } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    541,
+                    570,
+                    720,
+                    570,
+                    720,
+                    613,
+                    541,
+                    613
+                ],
+                "score": 0.91,
+                "latex": "K _ { 1 } ^ { ( L + 1 ) } ( { \\bf X } , { \\bf X ^ { \\prime } } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    680,
+                    232,
+                    787,
+                    232,
+                    787,
+                    266,
+                    680,
+                    266
+                ],
+                "score": 0.91,
+                "latex": "k ( \\mathbf { X } , \\mathbf { X } ^ { \\prime } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1273,
+                    682,
+                    1358,
+                    682,
+                    1358,
+                    717,
+                    1273,
+                    717
+                ],
+                "score": 0.91,
+                "latex": "j \\neq j ^ { \\prime }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1206,
+                    781,
+                    1284,
+                    781,
+                    1284,
+                    815,
+                    1206,
+                    815
+                ],
+                "score": 0.91,
+                "latex": "\\nu \\neq \\nu ^ { \\prime }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    656,
+                    728,
+                    741,
+                    728,
+                    741,
+                    762,
+                    656,
+                    762
+                ],
+                "score": 0.9,
+                "latex": "j \\neq j ^ { \\prime }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    755,
+                    1785,
+                    979,
+                    1785,
+                    979,
+                    1820,
+                    755,
+                    1820
+                ],
+                "score": 0.9,
+                "latex": "( \\phi ( x ) = \\operatorname* { m a x } ( 0 , x ) )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1042,
+                    411,
+                    1319,
+                    411,
+                    1319,
+                    449,
+                    1042,
+                    449
+                ],
+                "score": 0.9,
+                "latex": "\\mu \\in \\{ 1 , \\ldots , H ^ { ( \\ell ) } D ^ { ( \\ell ) } \\}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1038,
+                    313,
+                    1319,
+                    313,
+                    1319,
+                    351,
+                    1038,
+                    351
+                ],
+                "score": 0.9,
+                "latex": "\\mu \\in \\{ 1 , \\ldots , H ^ { ( 1 ) } D ^ { ( 1 ) } \\}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    373,
+                    516,
+                    639,
+                    516,
+                    639,
+                    551,
+                    373,
+                    551
+                ],
+                "score": 0.89,
+                "latex": "\\{ 1 , \\ldots , H ^ { ( \\ell + 1 ) } D ^ { ( \\ell + 1 ) } \\}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    387,
+                    379,
+                    558,
+                    379,
+                    558,
+                    409,
+                    387,
+                    409
+                ],
+                "score": 0.88,
+                "latex": "\\ell = 1 , 2 , \\dots , L"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    486,
+                    408,
+                    627,
+                    408,
+                    627,
+                    449,
+                    486,
+                    449
+                ],
+                "score": 0.87,
+                "latex": "V _ { \\mu } ^ { ( \\ell ) } ( { \\bf X } , { \\bf X ^ { \\prime } } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    612,
+                    823,
+                    637,
+                    823,
+                    637,
+                    856,
+                    612,
+                    856
+                ],
+                "score": 0.87,
+                "latex": "j ^ { \\prime }"
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    420,
+                    1864,
+                    1276,
+                    1864,
+                    1276,
+                    1956,
+                    420,
+                    1956
+                ],
+                "score": 0.87,
+                "latex": "V _ { \\nu } ^ { ( \\ell ) } ( { \\mathbf { X } } , { \\mathbf { X ^ { \\prime } } } ) = \\frac { \\sqrt { K _ { \\nu } ^ { ( \\ell ) } ( { \\mathbf { X } } , { \\mathbf { X } } ) K _ { \\nu } ^ { ( \\ell ) } ( { \\mathbf { X ^ { \\prime } } } , { \\mathbf { X ^ { \\prime } } } ) } } { \\pi } \\left( \\sin \\theta _ { \\nu } ^ { ( \\ell ) } + ( \\pi - \\theta _ { \\nu } ^ { ( \\ell ) } ) \\cos \\theta _ { \\nu } ^ { ( \\ell ) } \\right)"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    560,
+                    271,
+                    858,
+                    271,
+                    858,
+                    308,
+                    560,
+                    308
+                ],
+                "score": 0.86,
+                "latex": "\\mathbf { X } , \\mathbf { X } ^ { \\prime } \\in \\mathbb { R } ^ { C ^ { ( 0 ) } \\times ( H ^ { ( 0 ) } W ^ { ( 0 ) } ) }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    687,
+                    823,
+                    712,
+                    823,
+                    712,
+                    851,
+                    687,
+                    851
+                ],
+                "score": 0.85,
+                "latex": "\\nu ^ { \\prime }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    538,
+                    1673,
+                    558,
+                    1673,
+                    558,
+                    1703,
+                    538,
+                    1703
+                ],
+                "score": 0.84,
+                "latex": "\\phi"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    509,
+                    475,
+                    674,
+                    475,
+                    674,
+                    516,
+                    509,
+                    516
+                ],
+                "score": 0.84,
+                "latex": "K _ { \\mu } ^ { ( \\ell + 1 ) } ( { \\mathbf { X } } , { \\mathbf { X } } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    644,
+                    407,
+                    786,
+                    407,
+                    786,
+                    449,
+                    644,
+                    449
+                ],
+                "score": 0.83,
+                "latex": "V _ { \\mu } ^ { ( \\ell ) } ( { \\bf X } , { \\bf X ^ { \\prime } } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    614,
+                    311,
+                    761,
+                    311,
+                    761,
+                    351,
+                    614,
+                    351
+                ],
+                "score": 0.82,
+                "latex": "K _ { \\mu } ^ { ( 1 ) } ( { \\bf X } , { \\bf X ^ { \\prime } } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    729,
+                    1100,
+                    744,
+                    1100,
+                    744,
+                    1129,
+                    729,
+                    1129
+                ],
+                "score": 0.8,
+                "latex": "j"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    957,
+                    736,
+                    978,
+                    736,
+                    978,
+                    762,
+                    957,
+                    762
+                ],
+                "score": 0.79,
+                "latex": "\\mu"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    769,
+                    1028,
+                    788,
+                    1028,
+                    788,
+                    1049,
+                    769,
+                    1049
+                ],
+                "score": 0.78,
+                "latex": "\\nu"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    529,
+                    1072,
+                    548,
+                    1072,
+                    548,
+                    1093,
+                    529,
+                    1093
+                ],
+                "score": 0.74,
+                "latex": "\\nu"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    297,
+                    1732,
+                    649,
+                    1732,
+                    649,
+                    1774,
+                    297,
+                    1774
+                ],
+                "score": 0.73,
+                "latex": "K _ { \\mu } ^ { ( \\ell + 1 ) } ( { \\mathbf { X } } , { \\mathbf { X } } ) , K _ { \\mu } ^ { ( \\ell + 1 ) } ( { \\mathbf { X } } , { \\mathbf { X } } ^ { \\prime } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    371,
+                    1971,
+                    1123,
+                    1971,
+                    1123,
+                    2045,
+                    371,
+                    2045
+                ],
+                "score": 0.69,
+                "latex": "\\theta _ { \\nu } ^ { ( \\ell ) } = \\cos ^ { - 1 } \\bigg ( K _ { \\nu } ^ { ( \\ell ) } ( { \\mathbf X } , { \\mathbf X ^ { \\prime } } ) / \\sqrt { K _ { \\nu } ^ { ( \\ell ) } ( { \\mathbf X } , { \\mathbf X } ) K _ { \\nu } ^ { ( \\ell ) } ( { \\mathbf X ^ { \\prime } } , { \\mathbf X ^ { \\prime } } ) } \\bigg ) ."
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    720,
+                    474,
+                    896,
+                    474,
+                    896,
+                    518,
+                    720,
+                    518
+                ],
+                "score": 0.67,
+                "latex": "K _ { \\mu } ^ { ( \\ell + 1 ) } ( { \\bf X } , { \\bf X ^ { \\prime } } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    347,
+                    1028,
+                    365,
+                    1028,
+                    365,
+                    1055,
+                    347,
+                    1055
+                ],
+                "score": 0.64,
+                "latex": "\\mu"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    458,
+                    310,
+                    597,
+                    310,
+                    597,
+                    351,
+                    458,
+                    351
+                ],
+                "score": 0.61,
+                "latex": "K _ { \\mu } ^ { ( 1 ) } ( { \\bf X } , { \\bf X } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1017,
+                    474,
+                    1202,
+                    474,
+                    1202,
+                    518,
+                    1017,
+                    518
+                ],
+                "score": 0.31,
+                "latex": "K _ { \\mu } ^ { ( \\ell + 1 ) } ( { \\mathbf { X } } ^ { \\prime } , { \\mathbf { X } } ^ { \\prime } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1300,
+                    488,
+                    1323,
+                    488,
+                    1323,
+                    517,
+                    1300,
+                    517
+                ],
+                "score": 0.29,
+                "latex": "\\mu"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    72.0,
+                    817.0,
+                    72.0,
+                    817.0,
+                    109.0,
+                    295.0,
+                    109.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1582.0,
+                    763.0,
+                    1582.0,
+                    763.0,
+                    1618.0,
+                    294.0,
+                    1618.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    840.0,
+                    2087.0,
+                    861.0,
+                    2087.0,
+                    861.0,
+                    2117.0,
+                    840.0,
+                    2117.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    298.0,
+                    227.0,
+                    679.0,
+                    227.0,
+                    679.0,
+                    270.0,
+                    298.0,
+                    270.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    290.0,
+                    670.0,
+                    294.0,
+                    670.0,
+                    294.0,
+                    790.0,
+                    290.0,
+                    790.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    609.0,
+                    670.0,
+                    655.0,
+                    670.0,
+                    655.0,
+                    790.0,
+                    609.0,
+                    790.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    742.0,
+                    670.0,
+                    898.0,
+                    670.0,
+                    898.0,
+                    790.0,
+                    742.0,
+                    790.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    994.0,
+                    670.0,
+                    1048.0,
+                    670.0,
+                    1048.0,
+                    790.0,
+                    994.0,
+                    790.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1143.0,
+                    670.0,
+                    1247.0,
+                    670.0,
+                    1247.0,
+                    790.0,
+                    1143.0,
+                    790.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1359.0,
+                    670.0,
+                    1407.0,
+                    670.0,
+                    1407.0,
+                    790.0,
+                    1359.0,
+                    790.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    776.0,
+                    865.0,
+                    776.0,
+                    865.0,
+                    817.0,
+                    293.0,
+                    817.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1165.0,
+                    744.0,
+                    1205.0,
+                    744.0,
+                    1205.0,
+                    845.0,
+                    1165.0,
+                    845.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1285.0,
+                    744.0,
+                    1407.0,
+                    744.0,
+                    1407.0,
+                    845.0,
+                    1285.0,
+                    845.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    823.0,
+                    611.0,
+                    823.0,
+                    611.0,
+                    855.0,
+                    294.0,
+                    855.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    638.0,
+                    823.0,
+                    686.0,
+                    823.0,
+                    686.0,
+                    855.0,
+                    638.0,
+                    855.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    713.0,
+                    823.0,
+                    723.0,
+                    823.0,
+                    723.0,
+                    855.0,
+                    713.0,
+                    855.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    882.25,
+                    761.5,
+                    1009.25,
+                    761.5,
+                    1009.25,
+                    817.5,
+                    882.25,
+                    817.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1640.0,
+                    1404.0,
+                    1640.0,
+                    1404.0,
+                    1674.0,
+                    295.0,
+                    1674.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1673.0,
+                    537.0,
+                    1673.0,
+                    537.0,
+                    1704.0,
+                    295.0,
+                    1704.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    559.0,
+                    1673.0,
+                    1404.0,
+                    1673.0,
+                    1404.0,
+                    1704.0,
+                    559.0,
+                    1704.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1700.0,
+                    1404.0,
+                    1700.0,
+                    1404.0,
+                    1737.0,
+                    293.0,
+                    1737.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    650.0,
+                    1722.0,
+                    700.0,
+                    1722.0,
+                    700.0,
+                    1785.0,
+                    650.0,
+                    1785.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    882.0,
+                    1722.0,
+                    901.0,
+                    1722.0,
+                    901.0,
+                    1785.0,
+                    882.0,
+                    1785.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    282.0,
+                    1723.5,
+                    400.0,
+                    1723.5,
+                    400.0,
+                    1778.5,
+                    282.0,
+                    1778.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    563.0,
+                    1017.0,
+                    768.0,
+                    1017.0,
+                    768.0,
+                    1060.0,
+                    563.0,
+                    1060.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    789.0,
+                    1017.0,
+                    1404.0,
+                    1017.0,
+                    1404.0,
+                    1060.0,
+                    789.0,
+                    1060.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    289.0,
+                    1060.0,
+                    528.0,
+                    1060.0,
+                    528.0,
+                    1106.0,
+                    289.0,
+                    1106.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    549.0,
+                    1060.0,
+                    881.0,
+                    1060.0,
+                    881.0,
+                    1106.0,
+                    549.0,
+                    1106.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1019.0,
+                    1060.0,
+                    1408.0,
+                    1060.0,
+                    1408.0,
+                    1106.0,
+                    1019.0,
+                    1106.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1095.0,
+                    728.0,
+                    1095.0,
+                    728.0,
+                    1133.0,
+                    295.0,
+                    1133.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    745.0,
+                    1095.0,
+                    1116.0,
+                    1095.0,
+                    1116.0,
+                    1133.0,
+                    745.0,
+                    1133.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    288.25,
+                    1009.5,
+                    576.25,
+                    1009.5,
+                    576.25,
+                    1065.5,
+                    288.25,
+                    1065.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1786.0,
+                    754.0,
+                    1786.0,
+                    754.0,
+                    1822.0,
+                    295.0,
+                    1822.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    980.0,
+                    1786.0,
+                    1403.0,
+                    1786.0,
+                    1403.0,
+                    1822.0,
+                    980.0,
+                    1822.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1815.0,
+                    973.0,
+                    1815.0,
+                    973.0,
+                    1851.0,
+                    295.0,
+                    1851.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1509.0,
+                    1171.0,
+                    1509.0,
+                    1171.0,
+                    1548.0,
+                    293.0,
+                    1548.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1265.0,
+                    755.0,
+                    1265.0,
+                    755.0,
+                    1309.0,
+                    294.0,
+                    1309.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1419.0,
+                    376.0,
+                    1419.0,
+                    376.0,
+                    1458.0,
+                    295.0,
+                    1458.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    298.0,
+                    303.0,
+                    457.0,
+                    303.0,
+                    457.0,
+                    358.0,
+                    298.0,
+                    358.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    598.0,
+                    303.0,
+                    613.0,
+                    303.0,
+                    613.0,
+                    358.0,
+                    598.0,
+                    358.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    762.0,
+                    303.0,
+                    829.0,
+                    303.0,
+                    829.0,
+                    358.0,
+                    762.0,
+                    358.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    986.0,
+                    303.0,
+                    1037.0,
+                    303.0,
+                    1037.0,
+                    358.0,
+                    986.0,
+                    358.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1320.0,
+                    303.0,
+                    1409.0,
+                    303.0,
+                    1409.0,
+                    358.0,
+                    1320.0,
+                    358.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    338.0,
+                    344.0,
+                    453.0,
+                    344.0,
+                    453.0,
+                    383.0,
+                    338.0,
+                    383.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    304.0,
+                    371.0,
+                    386.0,
+                    371.0,
+                    386.0,
+                    410.0,
+                    304.0,
+                    410.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    559.0,
+                    371.0,
+                    599.0,
+                    371.0,
+                    599.0,
+                    410.0,
+                    559.0,
+                    410.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    307.0,
+                    413.0,
+                    342.0,
+                    413.0,
+                    342.0,
+                    448.0,
+                    307.0,
+                    448.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    356.0,
+                    399.0,
+                    485.0,
+                    399.0,
+                    485.0,
+                    458.0,
+                    356.0,
+                    458.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    628.0,
+                    399.0,
+                    643.0,
+                    399.0,
+                    643.0,
+                    458.0,
+                    628.0,
+                    458.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    787.0,
+                    399.0,
+                    846.0,
+                    399.0,
+                    846.0,
+                    458.0,
+                    787.0,
+                    458.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    991.0,
+                    399.0,
+                    1041.0,
+                    399.0,
+                    1041.0,
+                    458.0,
+                    991.0,
+                    458.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1320.0,
+                    399.0,
+                    1410.0,
+                    399.0,
+                    1410.0,
+                    458.0,
+                    1320.0,
+                    458.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    369.0,
+                    443.0,
+                    783.0,
+                    443.0,
+                    783.0,
+                    480.0,
+                    369.0,
+                    480.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    306.0,
+                    483.0,
+                    342.0,
+                    483.0,
+                    342.0,
+                    516.0,
+                    306.0,
+                    516.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    358.0,
+                    462.0,
+                    508.0,
+                    462.0,
+                    508.0,
+                    532.0,
+                    358.0,
+                    532.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    675.0,
+                    462.0,
+                    700.0,
+                    462.0,
+                    700.0,
+                    532.0,
+                    675.0,
+                    532.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    702.0,
+                    458.0,
+                    719.0,
+                    458.0,
+                    719.0,
+                    536.0,
+                    702.0,
+                    536.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    897.0,
+                    458.0,
+                    1016.0,
+                    458.0,
+                    1016.0,
+                    536.0,
+                    897.0,
+                    536.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1203.0,
+                    458.0,
+                    1299.0,
+                    458.0,
+                    1299.0,
+                    536.0,
+                    1203.0,
+                    536.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1324.0,
+                    458.0,
+                    1421.0,
+                    458.0,
+                    1421.0,
+                    536.0,
+                    1324.0,
+                    536.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    367.0,
+                    507.0,
+                    372.0,
+                    507.0,
+                    372.0,
+                    560.0,
+                    367.0,
+                    560.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    640.0,
+                    507.0,
+                    827.0,
+                    507.0,
+                    827.0,
+                    560.0,
+                    640.0,
+                    560.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    303.75,
+                    266.0,
+                    865.75,
+                    266.0,
+                    865.75,
+                    314.0,
+                    303.75,
+                    314.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    308.0,
+                    546.0,
+                    437.0,
+                    546.0,
+                    437.0,
+                    579.0,
+                    308.0,
+                    579.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    301.0,
+                    562.0,
+                    540.0,
+                    562.0,
+                    540.0,
+                    619.0,
+                    301.0,
+                    619.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    721.0,
+                    562.0,
+                    735.0,
+                    562.0,
+                    735.0,
+                    619.0,
+                    721.0,
+                    619.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    280.0,
+                    1954.0,
+                    370.0,
+                    1954.0,
+                    370.0,
+                    2053.0,
+                    280.0,
+                    2053.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    289.0,
+                    885.0,
+                    309.0,
+                    885.0,
+                    309.0,
+                    962.0,
+                    289.0,
+                    962.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    862.25,
+                    885.5,
+                    1071.25,
+                    885.5,
+                    1071.25,
+                    943.0,
+                    862.25,
+                    943.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    298.0,
+                    227.0,
+                    679.0,
+                    227.0,
+                    679.0,
+                    270.0,
+                    298.0,
+                    270.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 5,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    784,
+                    1403,
+                    784,
+                    1403,
+                    942,
+                    297,
+                    942
+                ],
+                "score": 0.979
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1818,
+                    1403,
+                    1818,
+                    1403,
+                    2035,
+                    297,
+                    2035
+                ],
+                "score": 0.978
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    471,
+                    1404,
+                    471,
+                    1404,
+                    615,
+                    298,
+                    615
+                ],
+                "score": 0.976
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    1182,
+                    1403,
+                    1182,
+                    1403,
+                    1275,
+                    299,
+                    1275
+                ],
+                "score": 0.974
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    333,
+                    1405,
+                    333,
+                    1405,
+                    457,
+                    298,
+                    457
+                ],
+                "score": 0.972
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    300,
+                    1387,
+                    1401,
+                    1387,
+                    1401,
+                    1481,
+                    300,
+                    1481
+                ],
+                "score": 0.97
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    1603,
+                    1397,
+                    1603,
+                    1397,
+                    1699,
+                    299,
+                    1699
+                ],
+                "score": 0.962
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    301,
+                    1711,
+                    1397,
+                    1711,
+                    1397,
+                    1806,
+                    301,
+                    1806
+                ],
+                "score": 0.962
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    301,
+                    1495,
+                    1399,
+                    1495,
+                    1399,
+                    1588,
+                    301,
+                    1588
+                ],
+                "score": 0.961
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    628,
+                    1402,
+                    628,
+                    1402,
+                    693,
+                    298,
+                    693
+                ],
+                "score": 0.952
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    502,
+                    946,
+                    1193,
+                    946,
+                    1193,
+                    1039,
+                    502,
+                    1039
+                ],
+                "score": 0.946
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    461,
+                    1091,
+                    1238,
+                    1091,
+                    1238,
+                    1167,
+                    461,
+                    1167
+                ],
+                "score": 0.944
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    1045,
+                    778,
+                    1045,
+                    778,
+                    1078,
+                    299,
+                    1078
+                ],
+                "score": 0.933
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    299,
+                    1319,
+                    558,
+                    1319,
+                    558,
+                    1354,
+                    299,
+                    1354
+                ],
+                "score": 0.911
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    301,
+                    728,
+                    741,
+                    728,
+                    741,
+                    760,
+                    301,
+                    760
+                ],
+                "score": 0.911
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    299,
+                    74,
+                    816,
+                    74,
+                    816,
+                    105,
+                    299,
+                    105
+                ],
+                "score": 0.904
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    300,
+                    286,
+                    1180,
+                    286,
+                    1180,
+                    320,
+                    300,
+                    320
+                ],
+                "score": 0.902
+            },
+            {
+                "category_id": 9,
+                "poly": [
+                    1352,
+                    1102,
+                    1400,
+                    1102,
+                    1400,
+                    1134,
+                    1352,
+                    1134
+                ],
+                "score": 0.894
+            },
+            {
+                "category_id": 9,
+                "poly": [
+                    1351,
+                    977,
+                    1400,
+                    977,
+                    1400,
+                    1007,
+                    1351,
+                    1007
+                ],
+                "score": 0.886
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    299,
+                    228,
+                    833,
+                    228,
+                    833,
+                    262,
+                    299,
+                    262
+                ],
+                "score": 0.841
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    842,
+                    2087,
+                    858,
+                    2087,
+                    858,
+                    2111,
+                    842,
+                    2111
+                ],
+                "score": 0.673
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    841,
+                    2087,
+                    859,
+                    2087,
+                    859,
+                    2111,
+                    841,
+                    2111
+                ],
+                "score": 0.231
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    503,
+                    938,
+                    1192,
+                    938,
+                    1192,
+                    1042,
+                    503,
+                    1042
+                ],
+                "score": 0.95,
+                "latex": "\\mathbf { a } _ { i } ^ { ( \\ell + 1 ) } ( \\mathbf { X } ) : = \\mathbf { a } _ { i } ^ { ( \\ell - s ) } ( \\mathbf { X } ) + \\mathbf { b } _ { i } ^ { ( \\ell + 1 ) } + \\sum _ { j = 1 } ^ { C ^ { ( \\ell ) } } \\mathbf { W } _ { i , j } ^ { ( \\ell ) } \\phi \\left( \\mathbf { a } _ { j } ^ { ( \\ell ) } ( \\mathbf { X } ) \\right) ,"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    328,
+                    1772,
+                    472,
+                    1772,
+                    472,
+                    1807,
+                    328,
+                    1807
+                ],
+                "score": 0.94,
+                "latex": "\\mathbf { K } _ { x ^ { * } x } \\mathbf { K } _ { x x } ^ { - 1 } \\bar { \\mathbf { Y } }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    770,
+                    503,
+                    861,
+                    503,
+                    861,
+                    549,
+                    770,
+                    549
+                ],
+                "score": 0.94,
+                "latex": "{ \\bf a } _ { j } ^ { ( \\ell ) } ( { \\bf X } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    377,
+                    1524,
+                    586,
+                    1524,
+                    586,
+                    1559,
+                    377,
+                    1559
+                ],
+                "score": 0.93,
+                "latex": "\\mathbf { Y } \\in \\{ - 1 , 1 \\} ^ { \\overline { { N } } \\times 1 0 }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    911,
+                    503,
+                    1011,
+                    503,
+                    1011,
+                    549,
+                    911,
+                    549
+                ],
+                "score": 0.93,
+                "latex": "\\mathbf { a } _ { j } ^ { ( \\ell ) } ( \\mathbf { X } ^ { \\prime } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    769,
+                    1664,
+                    852,
+                    1664,
+                    852,
+                    1699,
+                    769,
+                    1699
+                ],
+                "score": 0.93,
+                "latex": "{ \\mathbf { K } } _ { x x } ^ { - 1 } { \\mathbf { Y } }"
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    461,
+                    1091,
+                    1238,
+                    1091,
+                    1238,
+                    1169,
+                    461,
+                    1169
+                ],
+                "score": 0.92,
+                "latex": "K _ { \\mu } ^ { ( \\ell + 1 ) } ( \\mathbf { X } , \\mathbf { X } ^ { \\prime } ) = K _ { \\mu } ^ { ( \\ell - s ) } ( \\mathbf { X } , \\mathbf { X } ^ { \\prime } ) + \\sigma _ { b } ^ { 2 } + \\sigma _ { w } ^ { 2 } \\sum _ { \\nu \\in \\mu \\mathrm { t h } \\mathrm { p a t c h } } V _ { \\nu } ^ { ( \\ell ) } ( \\mathbf { X } , \\mathbf { X } ^ { \\prime } ) ."
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1097,
+                    1528,
+                    1196,
+                    1528,
+                    1196,
+                    1562,
+                    1097,
+                    1562
+                ],
+                "score": 0.92,
+                "latex": "y _ { n , c } = 1"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    930,
+                    1713,
+                    1030,
+                    1713,
+                    1030,
+                    1743,
+                    930,
+                    1743
+                ],
+                "score": 0.91,
+                "latex": "N _ { T } \\times N"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1112,
+                    1713,
+                    1178,
+                    1713,
+                    1178,
+                    1743,
+                    1112,
+                    1743
+                ],
+                "score": 0.91,
+                "latex": "\\mathbf { K } _ { x ^ { * } x }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    956,
+                    1635,
+                    1010,
+                    1635,
+                    1010,
+                    1665,
+                    956,
+                    1665
+                ],
+                "score": 0.91,
+                "latex": "{ \\bf K } _ { x x }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1253,
+                    505,
+                    1405,
+                    505,
+                    1405,
+                    541,
+                    1253,
+                    541
+                ],
+                "score": 0.91,
+                "latex": "2 H ^ { ( \\ell ) } D ^ { ( \\ell ) } \\times"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    297,
+                    544,
+                    419,
+                    544,
+                    419,
+                    577,
+                    297,
+                    577
+                ],
+                "score": 0.9,
+                "latex": "2 H ^ { ( \\ell ) } D ^ { ( \\ell ) }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    474,
+                    1713,
+                    623,
+                    1713,
+                    623,
+                    1743,
+                    474,
+                    1743
+                ],
+                "score": 0.9,
+                "latex": "N _ { T } = 1 0 0 0 0"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    724,
+                    1635,
+                    802,
+                    1635,
+                    802,
+                    1663,
+                    724,
+                    1663
+                ],
+                "score": 0.9,
+                "latex": "N \\times N"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    557,
+                    367,
+                    621,
+                    367,
+                    621,
+                    395,
+                    557,
+                    395
+                ],
+                "score": 0.9,
+                "latex": "\\ell + 1"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1318,
+                    847,
+                    1399,
+                    847,
+                    1399,
+                    879,
+                    1318,
+                    879
+                ],
+                "score": 0.89,
+                "latex": "\\mathbf { A } ^ { ( \\ell - s ) }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    298,
+                    579,
+                    404,
+                    579,
+                    404,
+                    611,
+                    298,
+                    611
+                ],
+                "score": 0.89,
+                "latex": "H ^ { ( \\ell ) } D ^ { ( \\ell ) }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    619,
+                    1497,
+                    762,
+                    1497,
+                    762,
+                    1525,
+                    619,
+                    1525
+                ],
+                "score": 0.89,
+                "latex": "N = 5 0 0 0 0"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1182,
+                    847,
+                    1264,
+                    847,
+                    1264,
+                    879,
+                    1182,
+                    879
+                ],
+                "score": 0.87,
+                "latex": "\\mathbf { A } ^ { ( \\ell + 1 ) }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    399,
+                    1910,
+                    431,
+                    1910,
+                    431,
+                    1946,
+                    399,
+                    1946
+                ],
+                "score": 0.8,
+                "latex": "\\sigma _ { b } ^ { 2 }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    374,
+                    886,
+                    392,
+                    886,
+                    392,
+                    908,
+                    374,
+                    908
+                ],
+                "score": 0.77,
+                "latex": "s"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    444,
+                    1910,
+                    480,
+                    1910,
+                    480,
+                    1945,
+                    444,
+                    1945
+                ],
+                "score": 0.74,
+                "latex": "\\sigma _ { w } ^ { 2 }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    479,
+                    1565,
+                    495,
+                    1565,
+                    495,
+                    1585,
+                    479,
+                    1585
+                ],
+                "score": 0.71,
+                "latex": "c"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    633,
+                    1852,
+                    666,
+                    1852,
+                    666,
+                    1878,
+                    633,
+                    1878
+                ],
+                "score": 0.71,
+                "latex": "M"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    553,
+                    1561,
+                    591,
+                    1561,
+                    591,
+                    1587,
+                    553,
+                    1587
+                ],
+                "score": 0.69,
+                "latex": "- 1"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1260,
+                    1532,
+                    1279,
+                    1532,
+                    1279,
+                    1554,
+                    1260,
+                    1554
+                ],
+                "score": 0.64,
+                "latex": "n"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1379,
+                    366,
+                    1394,
+                    366,
+                    1394,
+                    393,
+                    1379,
+                    393
+                ],
+                "score": 0.62,
+                "latex": "\\ell"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1190,
+                    1605,
+                    1220,
+                    1605,
+                    1220,
+                    1633,
+                    1190,
+                    1633
+                ],
+                "score": 0.61,
+                "latex": "\\mathbf { Y }"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1319.0,
+                    561.0,
+                    1319.0,
+                    561.0,
+                    1359.0,
+                    294.0,
+                    1359.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    728.0,
+                    744.0,
+                    728.0,
+                    744.0,
+                    764.0,
+                    295.0,
+                    764.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    72.0,
+                    817.0,
+                    72.0,
+                    817.0,
+                    108.0,
+                    295.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    225.0,
+                    838.0,
+                    225.0,
+                    838.0,
+                    266.0,
+                    293.0,
+                    266.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    839.0,
+                    2086.0,
+                    860.0,
+                    2086.0,
+                    860.0,
+                    2118.0,
+                    839.0,
+                    2118.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    839.0,
+                    2086.0,
+                    860.0,
+                    2086.0,
+                    860.0,
+                    2118.0,
+                    839.0,
+                    2118.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    782.0,
+                    1404.0,
+                    782.0,
+                    1404.0,
+                    823.0,
+                    294.0,
+                    823.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    817.0,
+                    1403.0,
+                    817.0,
+                    1403.0,
+                    851.0,
+                    296.0,
+                    851.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    846.0,
+                    1181.0,
+                    846.0,
+                    1181.0,
+                    887.0,
+                    291.0,
+                    887.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1265.0,
+                    846.0,
+                    1317.0,
+                    846.0,
+                    1317.0,
+                    887.0,
+                    1265.0,
+                    887.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1400.0,
+                    846.0,
+                    1405.0,
+                    846.0,
+                    1405.0,
+                    887.0,
+                    1400.0,
+                    887.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    876.0,
+                    373.0,
+                    876.0,
+                    373.0,
+                    917.0,
+                    294.0,
+                    917.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    393.0,
+                    876.0,
+                    1404.0,
+                    876.0,
+                    1404.0,
+                    917.0,
+                    393.0,
+                    917.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    910.0,
+                    413.0,
+                    910.0,
+                    413.0,
+                    949.0,
+                    293.0,
+                    949.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1815.0,
+                    1405.0,
+                    1815.0,
+                    1405.0,
+                    1857.0,
+                    292.0,
+                    1857.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1850.0,
+                    632.0,
+                    1850.0,
+                    632.0,
+                    1884.0,
+                    295.0,
+                    1884.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    667.0,
+                    1850.0,
+                    1404.0,
+                    1850.0,
+                    1404.0,
+                    1884.0,
+                    667.0,
+                    1884.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1879.0,
+                    1407.0,
+                    1879.0,
+                    1407.0,
+                    1919.0,
+                    292.0,
+                    1919.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    286.0,
+                    1898.0,
+                    398.0,
+                    1898.0,
+                    398.0,
+                    1960.0,
+                    286.0,
+                    1960.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    432.0,
+                    1898.0,
+                    443.0,
+                    1898.0,
+                    443.0,
+                    1960.0,
+                    432.0,
+                    1960.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    481.0,
+                    1898.0,
+                    1413.0,
+                    1898.0,
+                    1413.0,
+                    1960.0,
+                    481.0,
+                    1960.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1941.0,
+                    1405.0,
+                    1941.0,
+                    1405.0,
+                    1977.0,
+                    292.0,
+                    1977.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1970.0,
+                    1405.0,
+                    1970.0,
+                    1405.0,
+                    2008.0,
+                    294.0,
+                    2008.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    2006.0,
+                    496.0,
+                    2006.0,
+                    496.0,
+                    2038.0,
+                    296.0,
+                    2038.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    467.0,
+                    1405.0,
+                    467.0,
+                    1405.0,
+                    509.0,
+                    293.0,
+                    509.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    501.0,
+                    769.0,
+                    501.0,
+                    769.0,
+                    549.0,
+                    291.0,
+                    549.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    862.0,
+                    501.0,
+                    910.0,
+                    501.0,
+                    910.0,
+                    549.0,
+                    862.0,
+                    549.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1012.0,
+                    501.0,
+                    1252.0,
+                    501.0,
+                    1252.0,
+                    549.0,
+                    1012.0,
+                    549.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    540.0,
+                    296.0,
+                    540.0,
+                    296.0,
+                    589.0,
+                    291.0,
+                    589.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    420.0,
+                    540.0,
+                    1408.0,
+                    540.0,
+                    1408.0,
+                    589.0,
+                    420.0,
+                    589.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    405.0,
+                    575.0,
+                    943.0,
+                    575.0,
+                    943.0,
+                    621.0,
+                    405.0,
+                    621.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1180.0,
+                    1407.0,
+                    1180.0,
+                    1407.0,
+                    1218.0,
+                    294.0,
+                    1218.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1212.0,
+                    1404.0,
+                    1212.0,
+                    1404.0,
+                    1245.0,
+                    294.0,
+                    1245.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1243.0,
+                    807.0,
+                    1243.0,
+                    807.0,
+                    1279.0,
+                    293.0,
+                    1279.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    333.0,
+                    1407.0,
+                    333.0,
+                    1407.0,
+                    369.0,
+                    293.0,
+                    369.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    363.0,
+                    556.0,
+                    363.0,
+                    556.0,
+                    400.0,
+                    295.0,
+                    400.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    622.0,
+                    363.0,
+                    1378.0,
+                    363.0,
+                    1378.0,
+                    400.0,
+                    622.0,
+                    400.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1395.0,
+                    363.0,
+                    1405.0,
+                    363.0,
+                    1405.0,
+                    400.0,
+                    1395.0,
+                    400.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    395.0,
+                    1407.0,
+                    395.0,
+                    1407.0,
+                    431.0,
+                    295.0,
+                    431.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    423.0,
+                    1201.0,
+                    423.0,
+                    1201.0,
+                    462.0,
+                    291.0,
+                    462.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1388.0,
+                    1405.0,
+                    1388.0,
+                    1405.0,
+                    1422.0,
+                    296.0,
+                    1422.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1418.0,
+                    1405.0,
+                    1418.0,
+                    1405.0,
+                    1452.0,
+                    294.0,
+                    1452.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1450.0,
+                    876.0,
+                    1450.0,
+                    876.0,
+                    1484.0,
+                    295.0,
+                    1484.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1604.0,
+                    1189.0,
+                    1604.0,
+                    1189.0,
+                    1639.0,
+                    295.0,
+                    1639.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1221.0,
+                    1604.0,
+                    1403.0,
+                    1604.0,
+                    1403.0,
+                    1639.0,
+                    1221.0,
+                    1639.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1634.0,
+                    723.0,
+                    1634.0,
+                    723.0,
+                    1668.0,
+                    296.0,
+                    1668.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    803.0,
+                    1634.0,
+                    955.0,
+                    1634.0,
+                    955.0,
+                    1668.0,
+                    803.0,
+                    1668.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1011.0,
+                    1634.0,
+                    1405.0,
+                    1634.0,
+                    1405.0,
+                    1668.0,
+                    1011.0,
+                    1668.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1661.0,
+                    768.0,
+                    1661.0,
+                    768.0,
+                    1707.0,
+                    291.0,
+                    1707.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    853.0,
+                    1661.0,
+                    1177.0,
+                    1661.0,
+                    1177.0,
+                    1707.0,
+                    853.0,
+                    1707.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1707.0,
+                    473.0,
+                    1707.0,
+                    473.0,
+                    1749.0,
+                    295.0,
+                    1749.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    624.0,
+                    1707.0,
+                    929.0,
+                    1707.0,
+                    929.0,
+                    1749.0,
+                    624.0,
+                    1749.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1031.0,
+                    1707.0,
+                    1111.0,
+                    1707.0,
+                    1111.0,
+                    1749.0,
+                    1031.0,
+                    1749.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1179.0,
+                    1707.0,
+                    1403.0,
+                    1707.0,
+                    1403.0,
+                    1749.0,
+                    1179.0,
+                    1749.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1743.0,
+                    1402.0,
+                    1743.0,
+                    1402.0,
+                    1777.0,
+                    296.0,
+                    1777.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1768.0,
+                    327.0,
+                    1768.0,
+                    327.0,
+                    1812.0,
+                    293.0,
+                    1812.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    473.0,
+                    1768.0,
+                    485.0,
+                    1768.0,
+                    485.0,
+                    1812.0,
+                    473.0,
+                    1812.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    1496.0,
+                    618.0,
+                    1496.0,
+                    618.0,
+                    1529.0,
+                    297.0,
+                    1529.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    763.0,
+                    1496.0,
+                    1403.0,
+                    1496.0,
+                    1403.0,
+                    1529.0,
+                    763.0,
+                    1529.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    290.0,
+                    1516.0,
+                    376.0,
+                    1516.0,
+                    376.0,
+                    1567.0,
+                    290.0,
+                    1567.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    587.0,
+                    1516.0,
+                    1096.0,
+                    1516.0,
+                    1096.0,
+                    1567.0,
+                    587.0,
+                    1567.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1197.0,
+                    1516.0,
+                    1259.0,
+                    1516.0,
+                    1259.0,
+                    1567.0,
+                    1197.0,
+                    1567.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1280.0,
+                    1516.0,
+                    1408.0,
+                    1516.0,
+                    1408.0,
+                    1567.0,
+                    1280.0,
+                    1567.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1556.0,
+                    478.0,
+                    1556.0,
+                    478.0,
+                    1592.0,
+                    296.0,
+                    1592.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    496.0,
+                    1556.0,
+                    552.0,
+                    1556.0,
+                    552.0,
+                    1592.0,
+                    496.0,
+                    1592.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    592.0,
+                    1556.0,
+                    715.0,
+                    1556.0,
+                    715.0,
+                    1592.0,
+                    592.0,
+                    1592.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    627.0,
+                    1404.0,
+                    627.0,
+                    1404.0,
+                    666.0,
+                    294.0,
+                    666.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    662.0,
+                    1253.0,
+                    662.0,
+                    1253.0,
+                    694.0,
+                    297.0,
+                    694.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1043.0,
+                    780.0,
+                    1043.0,
+                    780.0,
+                    1082.0,
+                    295.0,
+                    1082.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    284.0,
+                    1186.0,
+                    284.0,
+                    1186.0,
+                    324.0,
+                    297.0,
+                    324.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 6,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1181,
+                    1404,
+                    1181,
+                    1404,
+                    1487,
+                    298,
+                    1487
+                ],
+                "score": 0.983
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1522,
+                    1404,
+                    1522,
+                    1404,
+                    1767,
+                    298,
+                    1767
+                ],
+                "score": 0.98
+            },
+            {
+                "category_id": 5,
+                "poly": [
+                    295,
+                    223,
+                    1432,
+                    223,
+                    1432,
+                    508,
+                    295,
+                    508
+                ],
+                "score": 0.98,
+                "html": "<table><tr><td>Method</td><td>#samples</td><td>Validation error</td><td>Test error</td></tr><tr><td>NNGP (Lee et al., 2017)</td><td>~250</td><td>1</td><td>1.21%</td></tr><tr><td>Convolutional GP (van der Wilk et al., 2017)</td><td>SGD</td><td></td><td>1.17%</td></tr><tr><td>Deep Conv. GP (Kumar et al., 2018)</td><td>SGD</td><td>1</td><td>1.34%</td></tr><tr><td>ConvNetGP</td><td>27</td><td>0.71%</td><td>1.03%</td></tr><tr><td>Residual CNN GP</td><td>27</td><td>0.72%</td><td>0.96%</td></tr><tr><td>ResNet GP</td><td>1</td><td>0.68%</td><td>0.84%</td></tr><tr><td>GP + parametric deep kernel (Bradshaw et al., 2017)</td><td>SGD</td><td>1</td><td>0.60%</td></tr><tr><td>ResNet (Chen et al.,2018)</td><td>一</td><td></td><td>0.41%</td></tr></table>"
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    982,
+                    1403,
+                    982,
+                    1403,
+                    1166,
+                    298,
+                    1166
+                ],
+                "score": 0.978
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    813,
+                    1402,
+                    813,
+                    1402,
+                    968,
+                    298,
+                    968
+                ],
+                "score": 0.976
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    1880,
+                    1401,
+                    1880,
+                    1401,
+                    2034,
+                    299,
+                    2034
+                ],
+                "score": 0.975
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    301,
+                    1812,
+                    581,
+                    1812,
+                    581,
+                    1846,
+                    301,
+                    1846
+                ],
+                "score": 0.892
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    300,
+                    76,
+                    815,
+                    76,
+                    815,
+                    104,
+                    300,
+                    104
+                ],
+                "score": 0.889
+            },
+            {
+                "category_id": 6,
+                "poly": [
+                    297,
+                    543,
+                    1404,
+                    543,
+                    1404,
+                    759,
+                    297,
+                    759
+                ],
+                "score": 0.8
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    841,
+                    2089,
+                    858,
+                    2089,
+                    858,
+                    2112,
+                    841,
+                    2112
+                ],
+                "score": 0.789
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    543,
+                    1404,
+                    543,
+                    1404,
+                    759,
+                    297,
+                    759
+                ],
+                "score": 0.275
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    557,
+                    1582,
+                    640,
+                    1582,
+                    640,
+                    1616,
+                    557,
+                    1616
+                ],
+                "score": 0.92,
+                "latex": "O ( \\bar { N } ^ { 3 } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1209,
+                    1521,
+                    1335,
+                    1521,
+                    1335,
+                    1556,
+                    1209,
+                    1556
+                ],
+                "score": 0.92,
+                "latex": "O ( N ^ { 2 } L D )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    298,
+                    1942,
+                    558,
+                    1942,
+                    558,
+                    1976,
+                    298,
+                    1976
+                ],
+                "score": 0.91,
+                "latex": "g \\sim \\mathcal { G P } ( 0 , k _ { p } ( \\mathbf { X } , \\mathbf { X } ^ { \\prime } ) )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    468,
+                    1643,
+                    551,
+                    1643,
+                    551,
+                    1673,
+                    468,
+                    1673
+                ],
+                "score": 0.9,
+                "latex": "N ^ { 2 } L D"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1334,
+                    1675,
+                    1400,
+                    1675,
+                    1400,
+                    1705,
+                    1334,
+                    1705
+                ],
+                "score": 0.9,
+                "latex": "{ \\bf K } _ { x x * }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1215,
+                    1675,
+                    1269,
+                    1675,
+                    1269,
+                    1706,
+                    1215,
+                    1706
+                ],
+                "score": 0.9,
+                "latex": "{ \\bf K } _ { x x }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    990,
+                    846,
+                    1057,
+                    846,
+                    1057,
+                    875,
+                    990,
+                    875
+                ],
+                "score": 0.9,
+                "latex": "3 \\times 3"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1296,
+                    1706,
+                    1350,
+                    1706,
+                    1350,
+                    1736,
+                    1296,
+                    1736
+                ],
+                "score": 0.89,
+                "latex": "{ \\bf K } _ { x x }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1189,
+                    1974,
+                    1220,
+                    1974,
+                    1220,
+                    2007,
+                    1189,
+                    2007
+                ],
+                "score": 0.88,
+                "latex": "k _ { p }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1059,
+                    1612,
+                    1101,
+                    1612,
+                    1101,
+                    1642,
+                    1059,
+                    1642
+                ],
+                "score": 0.88,
+                "latex": "N ^ { 3 }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    836,
+                    1736,
+                    972,
+                    1736,
+                    972,
+                    1766,
+                    836,
+                    1766
+                ],
+                "score": 0.87,
+                "latex": "4 3 . 2 5 \\pm 8 . 8"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    629,
+                    1913,
+                    649,
+                    1913,
+                    649,
+                    1944,
+                    629,
+                    1944
+                ],
+                "score": 0.87,
+                "latex": "f"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    582,
+                    2003,
+                    616,
+                    2003,
+                    616,
+                    2030,
+                    582,
+                    2030
+                ],
+                "score": 0.86,
+                "latex": "\\mathbf { X } ^ { \\prime }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    647,
+                    1944,
+                    666,
+                    1944,
+                    666,
+                    1975,
+                    647,
+                    1975
+                ],
+                "score": 0.85,
+                "latex": "f"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    855,
+                    1554,
+                    881,
+                    1554,
+                    881,
+                    1580,
+                    855,
+                    1580
+                ],
+                "score": 0.82,
+                "latex": "D"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    975,
+                    1947,
+                    993,
+                    1947,
+                    993,
+                    1974,
+                    975,
+                    1974
+                ],
+                "score": 0.81,
+                "latex": "g"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    370,
+                    1554,
+                    392,
+                    1554,
+                    392,
+                    1580,
+                    370,
+                    1580
+                ],
+                "score": 0.79,
+                "latex": "L"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    506,
+                    2004,
+                    532,
+                    2004,
+                    532,
+                    2030,
+                    506,
+                    2030
+                ],
+                "score": 0.68,
+                "latex": "\\mathbf { X }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    355,
+                    446,
+                    378,
+                    446,
+                    378,
+                    470,
+                    355,
+                    470
+                ],
+                "score": 0.61,
+                "latex": "^ +"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1807.0,
+                    587.0,
+                    1807.0,
+                    587.0,
+                    1852.0,
+                    292.0,
+                    1852.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    73.0,
+                    817.0,
+                    73.0,
+                    817.0,
+                    108.0,
+                    296.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    544.0,
+                    1404.0,
+                    544.0,
+                    1404.0,
+                    578.0,
+                    295.0,
+                    578.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    576.0,
+                    1405.0,
+                    576.0,
+                    1405.0,
+                    607.0,
+                    293.0,
+                    607.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    604.0,
+                    1405.0,
+                    604.0,
+                    1405.0,
+                    640.0,
+                    294.0,
+                    640.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    636.0,
+                    1402.0,
+                    636.0,
+                    1402.0,
+                    671.0,
+                    294.0,
+                    671.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    667.0,
+                    1405.0,
+                    667.0,
+                    1405.0,
+                    700.0,
+                    292.0,
+                    700.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    698.0,
+                    1405.0,
+                    698.0,
+                    1405.0,
+                    732.0,
+                    294.0,
+                    732.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    729.0,
+                    769.0,
+                    729.0,
+                    769.0,
+                    760.0,
+                    296.0,
+                    760.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    839.0,
+                    2086.0,
+                    860.0,
+                    2086.0,
+                    860.0,
+                    2116.0,
+                    839.0,
+                    2116.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1180.0,
+                    1405.0,
+                    1180.0,
+                    1405.0,
+                    1217.0,
+                    293.0,
+                    1217.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1213.0,
+                    1405.0,
+                    1213.0,
+                    1405.0,
+                    1249.0,
+                    294.0,
+                    1249.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1241.0,
+                    1406.0,
+                    1241.0,
+                    1406.0,
+                    1279.0,
+                    295.0,
+                    1279.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1274.0,
+                    1404.0,
+                    1274.0,
+                    1404.0,
+                    1310.0,
+                    294.0,
+                    1310.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1305.0,
+                    1405.0,
+                    1305.0,
+                    1405.0,
+                    1340.0,
+                    294.0,
+                    1340.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1335.0,
+                    1406.0,
+                    1335.0,
+                    1406.0,
+                    1372.0,
+                    292.0,
+                    1372.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1363.0,
+                    1406.0,
+                    1363.0,
+                    1406.0,
+                    1403.0,
+                    293.0,
+                    1403.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1394.0,
+                    1405.0,
+                    1394.0,
+                    1405.0,
+                    1430.0,
+                    293.0,
+                    1430.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1425.0,
+                    1404.0,
+                    1425.0,
+                    1404.0,
+                    1462.0,
+                    293.0,
+                    1462.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1457.0,
+                    1201.0,
+                    1457.0,
+                    1201.0,
+                    1493.0,
+                    294.0,
+                    1493.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1516.0,
+                    1208.0,
+                    1516.0,
+                    1208.0,
+                    1559.0,
+                    292.0,
+                    1559.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1336.0,
+                    1516.0,
+                    1406.0,
+                    1516.0,
+                    1406.0,
+                    1559.0,
+                    1336.0,
+                    1559.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1550.0,
+                    369.0,
+                    1550.0,
+                    369.0,
+                    1588.0,
+                    293.0,
+                    1588.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    393.0,
+                    1550.0,
+                    854.0,
+                    1550.0,
+                    854.0,
+                    1588.0,
+                    393.0,
+                    1588.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    882.0,
+                    1550.0,
+                    1404.0,
+                    1550.0,
+                    1404.0,
+                    1588.0,
+                    882.0,
+                    1588.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1581.0,
+                    556.0,
+                    1581.0,
+                    556.0,
+                    1618.0,
+                    293.0,
+                    1618.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    641.0,
+                    1581.0,
+                    1406.0,
+                    1581.0,
+                    1406.0,
+                    1618.0,
+                    641.0,
+                    1618.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1611.0,
+                    1058.0,
+                    1611.0,
+                    1058.0,
+                    1648.0,
+                    293.0,
+                    1648.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1102.0,
+                    1611.0,
+                    1408.0,
+                    1611.0,
+                    1408.0,
+                    1648.0,
+                    1102.0,
+                    1648.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1643.0,
+                    467.0,
+                    1643.0,
+                    467.0,
+                    1679.0,
+                    295.0,
+                    1679.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    552.0,
+                    1643.0,
+                    1405.0,
+                    1643.0,
+                    1405.0,
+                    1679.0,
+                    552.0,
+                    1679.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1671.0,
+                    1214.0,
+                    1671.0,
+                    1214.0,
+                    1712.0,
+                    292.0,
+                    1712.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1270.0,
+                    1671.0,
+                    1333.0,
+                    1671.0,
+                    1333.0,
+                    1712.0,
+                    1270.0,
+                    1712.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1401.0,
+                    1671.0,
+                    1404.0,
+                    1671.0,
+                    1404.0,
+                    1712.0,
+                    1401.0,
+                    1712.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1703.0,
+                    1295.0,
+                    1703.0,
+                    1295.0,
+                    1740.0,
+                    293.0,
+                    1740.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1351.0,
+                    1703.0,
+                    1405.0,
+                    1703.0,
+                    1405.0,
+                    1740.0,
+                    1351.0,
+                    1740.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1736.0,
+                    835.0,
+                    1736.0,
+                    835.0,
+                    1769.0,
+                    294.0,
+                    1769.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    973.0,
+                    1736.0,
+                    1394.0,
+                    1736.0,
+                    1394.0,
+                    1769.0,
+                    973.0,
+                    1769.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    983.0,
+                    1405.0,
+                    983.0,
+                    1405.0,
+                    1018.0,
+                    296.0,
+                    1018.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1014.0,
+                    1405.0,
+                    1014.0,
+                    1405.0,
+                    1047.0,
+                    293.0,
+                    1047.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1042.0,
+                    1405.0,
+                    1042.0,
+                    1405.0,
+                    1079.0,
+                    293.0,
+                    1079.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1073.0,
+                    1407.0,
+                    1073.0,
+                    1407.0,
+                    1110.0,
+                    292.0,
+                    1110.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1104.0,
+                    1407.0,
+                    1104.0,
+                    1407.0,
+                    1142.0,
+                    291.0,
+                    1142.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1135.0,
+                    946.0,
+                    1135.0,
+                    946.0,
+                    1169.0,
+                    293.0,
+                    1169.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    811.0,
+                    1406.0,
+                    811.0,
+                    1406.0,
+                    849.0,
+                    293.0,
+                    849.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    845.0,
+                    989.0,
+                    845.0,
+                    989.0,
+                    879.0,
+                    296.0,
+                    879.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1058.0,
+                    845.0,
+                    1404.0,
+                    845.0,
+                    1404.0,
+                    879.0,
+                    1058.0,
+                    879.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    873.0,
+                    1406.0,
+                    873.0,
+                    1406.0,
+                    912.0,
+                    293.0,
+                    912.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    904.0,
+                    1407.0,
+                    904.0,
+                    1407.0,
+                    942.0,
+                    291.0,
+                    942.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    937.0,
+                    846.0,
+                    937.0,
+                    846.0,
+                    971.0,
+                    293.0,
+                    971.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1878.0,
+                    1405.0,
+                    1878.0,
+                    1405.0,
+                    1917.0,
+                    294.0,
+                    1917.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1912.0,
+                    628.0,
+                    1912.0,
+                    628.0,
+                    1945.0,
+                    295.0,
+                    1945.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    650.0,
+                    1912.0,
+                    1405.0,
+                    1912.0,
+                    1405.0,
+                    1945.0,
+                    650.0,
+                    1945.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    290.0,
+                    1940.0,
+                    297.0,
+                    1940.0,
+                    297.0,
+                    1978.0,
+                    290.0,
+                    1978.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    559.0,
+                    1940.0,
+                    646.0,
+                    1940.0,
+                    646.0,
+                    1978.0,
+                    559.0,
+                    1978.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    667.0,
+                    1940.0,
+                    974.0,
+                    1940.0,
+                    974.0,
+                    1978.0,
+                    667.0,
+                    1978.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    994.0,
+                    1940.0,
+                    1406.0,
+                    1940.0,
+                    1406.0,
+                    1978.0,
+                    994.0,
+                    1978.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1970.0,
+                    1188.0,
+                    1970.0,
+                    1188.0,
+                    2009.0,
+                    292.0,
+                    2009.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1221.0,
+                    1970.0,
+                    1406.0,
+                    1970.0,
+                    1406.0,
+                    2009.0,
+                    1221.0,
+                    2009.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1997.0,
+                    505.0,
+                    1997.0,
+                    505.0,
+                    2042.0,
+                    293.0,
+                    2042.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    533.0,
+                    1997.0,
+                    581.0,
+                    1997.0,
+                    581.0,
+                    2042.0,
+                    533.0,
+                    2042.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    617.0,
+                    1997.0,
+                    1406.0,
+                    1997.0,
+                    1406.0,
+                    2042.0,
+                    617.0,
+                    2042.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    544.0,
+                    1404.0,
+                    544.0,
+                    1404.0,
+                    578.0,
+                    295.0,
+                    578.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    576.0,
+                    1405.0,
+                    576.0,
+                    1405.0,
+                    607.0,
+                    293.0,
+                    607.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    604.0,
+                    1405.0,
+                    604.0,
+                    1405.0,
+                    640.0,
+                    294.0,
+                    640.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    636.0,
+                    1402.0,
+                    636.0,
+                    1402.0,
+                    671.0,
+                    294.0,
+                    671.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    667.0,
+                    1405.0,
+                    667.0,
+                    1405.0,
+                    700.0,
+                    292.0,
+                    700.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    698.0,
+                    1405.0,
+                    698.0,
+                    1405.0,
+                    732.0,
+                    294.0,
+                    732.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    729.0,
+                    769.0,
+                    729.0,
+                    769.0,
+                    760.0,
+                    296.0,
+                    760.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 7,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1512,
+                    1404,
+                    1512,
+                    1404,
+                    1788,
+                    298,
+                    1788
+                ],
+                "score": 0.979
+            },
+            {
+                "category_id": 3,
+                "poly": [
+                    306,
+                    222,
+                    1406,
+                    222,
+                    1406,
+                    970,
+                    306,
+                    970
+                ],
+                "score": 0.974
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    1265,
+                    1403,
+                    1265,
+                    1403,
+                    1390,
+                    299,
+                    1390
+                ],
+                "score": 0.972
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    1911,
+                    1401,
+                    1911,
+                    1401,
+                    2034,
+                    299,
+                    2034
+                ],
+                "score": 0.971
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    1803,
+                    1403,
+                    1803,
+                    1403,
+                    1896,
+                    299,
+                    1896
+                ],
+                "score": 0.969
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    1405,
+                    1403,
+                    1405,
+                    1403,
+                    1498,
+                    299,
+                    1498
+                ],
+                "score": 0.969
+            },
+            {
+                "category_id": 4,
+                "poly": [
+                    296,
+                    1031,
+                    1404,
+                    1031,
+                    1404,
+                    1187,
+                    296,
+                    1187
+                ],
+                "score": 0.968
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    299,
+                    76,
+                    815,
+                    76,
+                    815,
+                    104,
+                    299,
+                    104
+                ],
+                "score": 0.89
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    840,
+                    2088,
+                    858,
+                    2088,
+                    858,
+                    2111,
+                    840,
+                    2111
+                ],
+                "score": 0.789
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    700,
+                    1573,
+                    1119,
+                    1573,
+                    1119,
+                    1607,
+                    700,
+                    1607
+                ],
+                "score": 0.92,
+                "latex": "k _ { \\mathrm { d e e p } } ( \\mathbf { X } , \\bar { \\mathbf { X } } ^ { \\prime } ) : = k ( g ( \\mathbf { X } ; \\theta ) , \\bar { g ( \\mathbf { X } ^ { \\prime } , \\theta ) } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1128,
+                    1298,
+                    1160,
+                    1298,
+                    1160,
+                    1331,
+                    1128,
+                    1331
+                ],
+                "score": 0.88,
+                "latex": "k _ { p }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    452,
+                    1664,
+                    495,
+                    1664,
+                    495,
+                    1694,
+                    452,
+                    1694
+                ],
+                "score": 0.87,
+                "latex": "1 0 ^ { 6 }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    541,
+                    1756,
+                    584,
+                    1756,
+                    584,
+                    1785,
+                    541,
+                    1785
+                ],
+                "score": 0.85,
+                "latex": "1 0 ^ { 6 }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    639,
+                    1327,
+                    675,
+                    1327,
+                    675,
+                    1356,
+                    639,
+                    1356
+                ],
+                "score": 0.83,
+                "latex": "\\mathbf { X } ^ { \\prime }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1297,
+                    1548,
+                    1315,
+                    1548,
+                    1315,
+                    1576,
+                    1297,
+                    1576
+                ],
+                "score": 0.79,
+                "latex": "g"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    615,
+                    1545,
+                    634,
+                    1545,
+                    634,
+                    1572,
+                    615,
+                    1572
+                ],
+                "score": 0.76,
+                "latex": "k"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    559,
+                    1328,
+                    588,
+                    1328,
+                    588,
+                    1356,
+                    559,
+                    1356
+                ],
+                "score": 0.33,
+                "latex": "\\mathbf { X }"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    311.0,
+                    220.0,
+                    345.0,
+                    220.0,
+                    345.0,
+                    259.0,
+                    311.0,
+                    259.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    447.0,
+                    248.0,
+                    526.0,
+                    248.0,
+                    526.0,
+                    290.0,
+                    447.0,
+                    290.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    756.0,
+                    248.0,
+                    809.0,
+                    248.0,
+                    809.0,
+                    286.0,
+                    756.0,
+                    286.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    987.0,
+                    241.0,
+                    1087.0,
+                    241.0,
+                    1087.0,
+                    293.0,
+                    987.0,
+                    293.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1252.0,
+                    243.0,
+                    1371.0,
+                    243.0,
+                    1371.0,
+                    293.0,
+                    1252.0,
+                    293.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    302.0,
+                    277.0,
+                    410.0,
+                    277.0,
+                    410.0,
+                    354.0,
+                    302.0,
+                    354.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    616.0,
+                    277.0,
+                    687.0,
+                    277.0,
+                    687.0,
+                    322.0,
+                    616.0,
+                    322.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    892.0,
+                    277.0,
+                    960.0,
+                    277.0,
+                    960.0,
+                    322.0,
+                    892.0,
+                    322.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1166.0,
+                    274.0,
+                    1240.0,
+                    274.0,
+                    1240.0,
+                    325.0,
+                    1166.0,
+                    325.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1306.0,
+                    306.0,
+                    1317.0,
+                    306.0,
+                    1317.0,
+                    318.0,
+                    1306.0,
+                    318.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    343.0,
+                    364.0,
+                    416.0,
+                    364.0,
+                    416.0,
+                    404.0,
+                    343.0,
+                    404.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    558.0,
+                    384.0,
+                    570.0,
+                    384.0,
+                    570.0,
+                    397.0,
+                    558.0,
+                    397.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    619.0,
+                    364.0,
+                    682.0,
+                    364.0,
+                    682.0,
+                    404.0,
+                    619.0,
+                    404.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    894.0,
+                    365.0,
+                    956.0,
+                    365.0,
+                    956.0,
+                    405.0,
+                    894.0,
+                    405.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1169.0,
+                    364.0,
+                    1231.0,
+                    364.0,
+                    1231.0,
+                    404.0,
+                    1169.0,
+                    404.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    386.0,
+                    395.0,
+                    440.0,
+                    395.0,
+                    440.0,
+                    432.0,
+                    386.0,
+                    432.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    470.0,
+                    394.0,
+                    505.0,
+                    394.0,
+                    505.0,
+                    432.0,
+                    470.0,
+                    432.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    549.0,
+                    395.0,
+                    579.0,
+                    395.0,
+                    579.0,
+                    431.0,
+                    549.0,
+                    431.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    661.0,
+                    395.0,
+                    715.0,
+                    395.0,
+                    715.0,
+                    432.0,
+                    661.0,
+                    432.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    745.0,
+                    394.0,
+                    780.0,
+                    394.0,
+                    780.0,
+                    432.0,
+                    745.0,
+                    432.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    823.0,
+                    397.0,
+                    854.0,
+                    397.0,
+                    854.0,
+                    431.0,
+                    823.0,
+                    431.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    936.0,
+                    394.0,
+                    990.0,
+                    394.0,
+                    990.0,
+                    432.0,
+                    936.0,
+                    432.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1020.0,
+                    394.0,
+                    1055.0,
+                    394.0,
+                    1055.0,
+                    432.0,
+                    1020.0,
+                    432.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1099.0,
+                    395.0,
+                    1128.0,
+                    395.0,
+                    1128.0,
+                    431.0,
+                    1099.0,
+                    431.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1210.0,
+                    393.0,
+                    1266.0,
+                    393.0,
+                    1266.0,
+                    434.0,
+                    1210.0,
+                    434.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1293.0,
+                    394.0,
+                    1330.0,
+                    394.0,
+                    1330.0,
+                    432.0,
+                    1293.0,
+                    432.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1373.0,
+                    397.0,
+                    1404.0,
+                    397.0,
+                    1404.0,
+                    431.0,
+                    1373.0,
+                    431.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    437.0,
+                    431.0,
+                    536.0,
+                    431.0,
+                    536.0,
+                    474.0,
+                    437.0,
+                    474.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    715.0,
+                    432.0,
+                    811.0,
+                    432.0,
+                    811.0,
+                    475.0,
+                    715.0,
+                    475.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    991.0,
+                    432.0,
+                    1086.0,
+                    432.0,
+                    1086.0,
+                    475.0,
+                    991.0,
+                    475.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1266.0,
+                    432.0,
+                    1361.0,
+                    432.0,
+                    1361.0,
+                    475.0,
+                    1266.0,
+                    475.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    306.0,
+                    461.0,
+                    351.0,
+                    461.0,
+                    351.0,
+                    651.0,
+                    306.0,
+                    651.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    366.0,
+                    498.0,
+                    416.0,
+                    498.0,
+                    416.0,
+                    535.0,
+                    366.0,
+                    535.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    640.0,
+                    495.0,
+                    692.0,
+                    495.0,
+                    692.0,
+                    538.0,
+                    640.0,
+                    538.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    915.0,
+                    495.0,
+                    970.0,
+                    495.0,
+                    970.0,
+                    539.0,
+                    915.0,
+                    539.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1190.0,
+                    494.0,
+                    1247.0,
+                    494.0,
+                    1247.0,
+                    543.0,
+                    1190.0,
+                    543.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1378.0,
+                    513.0,
+                    1389.0,
+                    513.0,
+                    1389.0,
+                    522.0,
+                    1378.0,
+                    522.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    643.0,
+                    554.0,
+                    678.0,
+                    554.0,
+                    678.0,
+                    593.0,
+                    643.0,
+                    593.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    917.0,
+                    553.0,
+                    964.0,
+                    553.0,
+                    964.0,
+                    595.0,
+                    917.0,
+                    595.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1191.0,
+                    552.0,
+                    1232.0,
+                    552.0,
+                    1232.0,
+                    595.0,
+                    1191.0,
+                    595.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    368.0,
+                    614.0,
+                    400.0,
+                    614.0,
+                    400.0,
+                    644.0,
+                    368.0,
+                    644.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    558.0,
+                    633.0,
+                    568.0,
+                    633.0,
+                    568.0,
+                    643.0,
+                    558.0,
+                    643.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    627.0,
+                    612.0,
+                    677.0,
+                    612.0,
+                    677.0,
+                    649.0,
+                    627.0,
+                    649.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    835.0,
+                    634.0,
+                    845.0,
+                    634.0,
+                    845.0,
+                    643.0,
+                    835.0,
+                    643.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    906.0,
+                    614.0,
+                    955.0,
+                    614.0,
+                    955.0,
+                    646.0,
+                    906.0,
+                    646.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1176.0,
+                    613.0,
+                    1226.0,
+                    613.0,
+                    1226.0,
+                    646.0,
+                    1176.0,
+                    646.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    390.0,
+                    641.0,
+                    440.0,
+                    641.0,
+                    440.0,
+                    678.0,
+                    390.0,
+                    678.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    441.0,
+                    645.0,
+                    502.0,
+                    645.0,
+                    502.0,
+                    677.0,
+                    441.0,
+                    677.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    549.0,
+                    645.0,
+                    579.0,
+                    645.0,
+                    579.0,
+                    677.0,
+                    549.0,
+                    677.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    666.0,
+                    643.0,
+                    776.0,
+                    643.0,
+                    776.0,
+                    679.0,
+                    666.0,
+                    679.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    825.0,
+                    645.0,
+                    854.0,
+                    645.0,
+                    854.0,
+                    678.0,
+                    825.0,
+                    678.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    942.0,
+                    644.0,
+                    988.0,
+                    644.0,
+                    988.0,
+                    678.0,
+                    942.0,
+                    678.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1022.0,
+                    645.0,
+                    1052.0,
+                    645.0,
+                    1052.0,
+                    677.0,
+                    1022.0,
+                    677.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1101.0,
+                    646.0,
+                    1127.0,
+                    646.0,
+                    1127.0,
+                    676.0,
+                    1101.0,
+                    676.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1216.0,
+                    641.0,
+                    1265.0,
+                    641.0,
+                    1265.0,
+                    679.0,
+                    1216.0,
+                    679.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1266.0,
+                    645.0,
+                    1327.0,
+                    645.0,
+                    1327.0,
+                    677.0,
+                    1266.0,
+                    677.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1376.0,
+                    648.0,
+                    1402.0,
+                    648.0,
+                    1402.0,
+                    676.0,
+                    1376.0,
+                    676.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    422.0,
+                    676.0,
+                    553.0,
+                    676.0,
+                    553.0,
+                    724.0,
+                    422.0,
+                    724.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    697.0,
+                    676.0,
+                    828.0,
+                    676.0,
+                    828.0,
+                    724.0,
+                    697.0,
+                    724.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    973.0,
+                    675.0,
+                    1103.0,
+                    675.0,
+                    1103.0,
+                    725.0,
+                    973.0,
+                    725.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1248.0,
+                    675.0,
+                    1378.0,
+                    675.0,
+                    1378.0,
+                    724.0,
+                    1248.0,
+                    724.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    305.0,
+                    711.0,
+                    350.0,
+                    711.0,
+                    350.0,
+                    917.0,
+                    305.0,
+                    917.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    368.0,
+                    744.0,
+                    401.0,
+                    744.0,
+                    401.0,
+                    785.0,
+                    368.0,
+                    785.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    645.0,
+                    745.0,
+                    675.0,
+                    745.0,
+                    675.0,
+                    780.0,
+                    645.0,
+                    780.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    915.0,
+                    741.0,
+                    966.0,
+                    741.0,
+                    966.0,
+                    785.0,
+                    915.0,
+                    785.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1190.0,
+                    740.0,
+                    1247.0,
+                    740.0,
+                    1247.0,
+                    789.0,
+                    1190.0,
+                    789.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    374.0,
+                    805.0,
+                    397.0,
+                    805.0,
+                    397.0,
+                    833.0,
+                    374.0,
+                    833.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    647.0,
+                    804.0,
+                    672.0,
+                    804.0,
+                    672.0,
+                    834.0,
+                    647.0,
+                    834.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    918.0,
+                    800.0,
+                    964.0,
+                    800.0,
+                    964.0,
+                    841.0,
+                    918.0,
+                    841.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1195.0,
+                    802.0,
+                    1225.0,
+                    802.0,
+                    1225.0,
+                    837.0,
+                    1195.0,
+                    837.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    370.0,
+                    861.0,
+                    400.0,
+                    861.0,
+                    400.0,
+                    895.0,
+                    370.0,
+                    895.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    645.0,
+                    861.0,
+                    673.0,
+                    861.0,
+                    673.0,
+                    895.0,
+                    645.0,
+                    895.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    917.0,
+                    860.0,
+                    955.0,
+                    860.0,
+                    955.0,
+                    896.0,
+                    917.0,
+                    896.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1194.0,
+                    861.0,
+                    1225.0,
+                    861.0,
+                    1225.0,
+                    895.0,
+                    1194.0,
+                    895.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    397.0,
+                    893.0,
+                    422.0,
+                    893.0,
+                    422.0,
+                    924.0,
+                    397.0,
+                    924.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    476.0,
+                    895.0,
+                    497.0,
+                    895.0,
+                    497.0,
+                    919.0,
+                    476.0,
+                    919.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    550.0,
+                    890.0,
+                    579.0,
+                    890.0,
+                    579.0,
+                    925.0,
+                    550.0,
+                    925.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    672.0,
+                    893.0,
+                    697.0,
+                    893.0,
+                    697.0,
+                    924.0,
+                    672.0,
+                    924.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    751.0,
+                    895.0,
+                    772.0,
+                    895.0,
+                    772.0,
+                    919.0,
+                    751.0,
+                    919.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    825.0,
+                    890.0,
+                    854.0,
+                    890.0,
+                    854.0,
+                    925.0,
+                    825.0,
+                    925.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    945.0,
+                    890.0,
+                    973.0,
+                    890.0,
+                    973.0,
+                    927.0,
+                    945.0,
+                    927.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1026.0,
+                    894.0,
+                    1046.0,
+                    894.0,
+                    1046.0,
+                    919.0,
+                    1026.0,
+                    919.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1099.0,
+                    889.0,
+                    1128.0,
+                    889.0,
+                    1128.0,
+                    927.0,
+                    1099.0,
+                    927.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1221.0,
+                    890.0,
+                    1249.0,
+                    890.0,
+                    1249.0,
+                    925.0,
+                    1221.0,
+                    925.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1300.0,
+                    894.0,
+                    1325.0,
+                    894.0,
+                    1325.0,
+                    920.0,
+                    1300.0,
+                    920.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1375.0,
+                    890.0,
+                    1404.0,
+                    890.0,
+                    1404.0,
+                    924.0,
+                    1375.0,
+                    924.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    408.0,
+                    922.0,
+                    566.0,
+                    922.0,
+                    566.0,
+                    972.0,
+                    408.0,
+                    972.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    683.0,
+                    922.0,
+                    841.0,
+                    922.0,
+                    841.0,
+                    972.0,
+                    683.0,
+                    972.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    957.0,
+                    922.0,
+                    1116.0,
+                    922.0,
+                    1116.0,
+                    972.0,
+                    957.0,
+                    972.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1233.0,
+                    924.0,
+                    1390.0,
+                    924.0,
+                    1390.0,
+                    970.0,
+                    1233.0,
+                    970.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    725.0,
+                    252.5,
+                    755.0,
+                    252.5,
+                    755.0,
+                    281.5,
+                    725.0,
+                    281.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    362.0,
+                    553.0,
+                    410.0,
+                    553.0,
+                    410.0,
+                    591.0,
+                    362.0,
+                    591.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    958.75,
+                    610.5,
+                    1000.75,
+                    610.5,
+                    1000.75,
+                    632.5,
+                    958.75,
+                    632.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    423.0,
+                    794.5,
+                    557.0,
+                    794.5,
+                    557.0,
+                    820.0,
+                    423.0,
+                    820.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    706.0,
+                    797.5,
+                    832.0,
+                    797.5,
+                    832.0,
+                    818.5,
+                    706.0,
+                    818.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1032.0,
+                    1405.0,
+                    1032.0,
+                    1405.0,
+                    1069.0,
+                    294.0,
+                    1069.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1061.0,
+                    1404.0,
+                    1061.0,
+                    1404.0,
+                    1101.0,
+                    293.0,
+                    1101.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1092.0,
+                    1406.0,
+                    1092.0,
+                    1406.0,
+                    1132.0,
+                    293.0,
+                    1132.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1122.0,
+                    1405.0,
+                    1122.0,
+                    1405.0,
+                    1162.0,
+                    293.0,
+                    1162.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1154.0,
+                    769.0,
+                    1154.0,
+                    769.0,
+                    1190.0,
+                    294.0,
+                    1190.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    73.0,
+                    816.0,
+                    73.0,
+                    816.0,
+                    108.0,
+                    295.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    839.0,
+                    2087.0,
+                    860.0,
+                    2087.0,
+                    860.0,
+                    2117.0,
+                    839.0,
+                    2117.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1510.0,
+                    1406.0,
+                    1510.0,
+                    1406.0,
+                    1549.0,
+                    295.0,
+                    1549.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1542.0,
+                    614.0,
+                    1542.0,
+                    614.0,
+                    1581.0,
+                    293.0,
+                    1581.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    635.0,
+                    1542.0,
+                    1296.0,
+                    1542.0,
+                    1296.0,
+                    1581.0,
+                    635.0,
+                    1581.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1316.0,
+                    1542.0,
+                    1404.0,
+                    1542.0,
+                    1404.0,
+                    1581.0,
+                    1316.0,
+                    1581.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1572.0,
+                    699.0,
+                    1572.0,
+                    699.0,
+                    1612.0,
+                    293.0,
+                    1612.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1120.0,
+                    1572.0,
+                    1408.0,
+                    1572.0,
+                    1408.0,
+                    1612.0,
+                    1120.0,
+                    1612.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1599.0,
+                    1408.0,
+                    1599.0,
+                    1408.0,
+                    1645.0,
+                    291.0,
+                    1645.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1634.0,
+                    1408.0,
+                    1634.0,
+                    1408.0,
+                    1671.0,
+                    296.0,
+                    1671.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1663.0,
+                    451.0,
+                    1663.0,
+                    451.0,
+                    1704.0,
+                    293.0,
+                    1704.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    496.0,
+                    1663.0,
+                    1408.0,
+                    1663.0,
+                    1408.0,
+                    1704.0,
+                    496.0,
+                    1704.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1693.0,
+                    1405.0,
+                    1693.0,
+                    1405.0,
+                    1732.0,
+                    293.0,
+                    1732.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1726.0,
+                    1405.0,
+                    1726.0,
+                    1405.0,
+                    1762.0,
+                    294.0,
+                    1762.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1753.0,
+                    540.0,
+                    1753.0,
+                    540.0,
+                    1791.0,
+                    293.0,
+                    1791.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    585.0,
+                    1753.0,
+                    605.0,
+                    1753.0,
+                    605.0,
+                    1791.0,
+                    585.0,
+                    1791.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1265.0,
+                    1404.0,
+                    1265.0,
+                    1404.0,
+                    1301.0,
+                    293.0,
+                    1301.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1297.0,
+                    1127.0,
+                    1297.0,
+                    1127.0,
+                    1333.0,
+                    294.0,
+                    1333.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1161.0,
+                    1297.0,
+                    1404.0,
+                    1297.0,
+                    1404.0,
+                    1333.0,
+                    1161.0,
+                    1333.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1328.0,
+                    558.0,
+                    1328.0,
+                    558.0,
+                    1363.0,
+                    293.0,
+                    1363.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    589.0,
+                    1328.0,
+                    638.0,
+                    1328.0,
+                    638.0,
+                    1363.0,
+                    589.0,
+                    1363.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    676.0,
+                    1328.0,
+                    1407.0,
+                    1328.0,
+                    1407.0,
+                    1363.0,
+                    676.0,
+                    1363.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1360.0,
+                    1340.0,
+                    1360.0,
+                    1340.0,
+                    1393.0,
+                    295.0,
+                    1393.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1910.0,
+                    1405.0,
+                    1910.0,
+                    1405.0,
+                    1945.0,
+                    293.0,
+                    1945.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1940.0,
+                    1406.0,
+                    1940.0,
+                    1406.0,
+                    1978.0,
+                    293.0,
+                    1978.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1972.0,
+                    1403.0,
+                    1972.0,
+                    1403.0,
+                    2007.0,
+                    294.0,
+                    2007.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    2002.0,
+                    1403.0,
+                    2002.0,
+                    1403.0,
+                    2040.0,
+                    293.0,
+                    2040.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1801.0,
+                    1404.0,
+                    1801.0,
+                    1404.0,
+                    1837.0,
+                    293.0,
+                    1837.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    1836.0,
+                    1404.0,
+                    1836.0,
+                    1404.0,
+                    1866.0,
+                    297.0,
+                    1866.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    1865.0,
+                    974.0,
+                    1865.0,
+                    974.0,
+                    1899.0,
+                    297.0,
+                    1899.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1404.0,
+                    1404.0,
+                    1404.0,
+                    1404.0,
+                    1442.0,
+                    294.0,
+                    1442.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    1436.0,
+                    1404.0,
+                    1436.0,
+                    1404.0,
+                    1470.0,
+                    297.0,
+                    1470.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1466.0,
+                    1360.0,
+                    1466.0,
+                    1360.0,
+                    1503.0,
+                    295.0,
+                    1503.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 8,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    412,
+                    1404,
+                    412,
+                    1404,
+                    659,
+                    298,
+                    659
+                ],
+                "score": 0.975
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    295,
+                    230,
+                    1401,
+                    230,
+                    1401,
+                    293,
+                    295,
+                    293
+                ],
+                "score": 0.94
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    300,
+                    75,
+                    816,
+                    75,
+                    816,
+                    105,
+                    300,
+                    105
+                ],
+                "score": 0.892
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    300,
+                    342,
+                    844,
+                    342,
+                    844,
+                    376,
+                    300,
+                    376
+                ],
+                "score": 0.874
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    300,
+                    709,
+                    488,
+                    709,
+                    488,
+                    741,
+                    300,
+                    741
+                ],
+                "score": 0.839
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    837,
+                    2088,
+                    865,
+                    2088,
+                    865,
+                    2112,
+                    837,
+                    2112
+                ],
+                "score": 0.815
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    301,
+                    763,
+                    1402,
+                    763,
+                    1402,
+                    825,
+                    301,
+                    825
+                ],
+                "score": 0.811
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    1245,
+                    1402,
+                    1245,
+                    1402,
+                    1340,
+                    299,
+                    1340
+                ],
+                "score": 0.795
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    295,
+                    1154,
+                    1401,
+                    1154,
+                    1401,
+                    1219,
+                    295,
+                    1219
+                ],
+                "score": 0.767
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    296,
+                    852,
+                    1400,
+                    852,
+                    1400,
+                    917,
+                    296,
+                    917
+                ],
+                "score": 0.731
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1366,
+                    1403,
+                    1366,
+                    1403,
+                    1521,
+                    297,
+                    1521
+                ],
+                "score": 0.729
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    296,
+                    943,
+                    1399,
+                    943,
+                    1399,
+                    1006,
+                    296,
+                    1006
+                ],
+                "score": 0.723
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    1547,
+                    1401,
+                    1547,
+                    1401,
+                    1642,
+                    299,
+                    1642
+                ],
+                "score": 0.722
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1669,
+                    1400,
+                    1669,
+                    1400,
+                    1733,
+                    297,
+                    1733
+                ],
+                "score": 0.663
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    1759,
+                    1401,
+                    1759,
+                    1401,
+                    1824,
+                    299,
+                    1824
+                ],
+                "score": 0.635
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1971,
+                    1401,
+                    1971,
+                    1401,
+                    2035,
+                    298,
+                    2035
+                ],
+                "score": 0.63
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    300,
+                    1850,
+                    1401,
+                    1850,
+                    1401,
+                    1946,
+                    300,
+                    1946
+                ],
+                "score": 0.58
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1032,
+                    1406,
+                    1032,
+                    1406,
+                    1127,
+                    298,
+                    1127
+                ],
+                "score": 0.495
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1075,
+                    598,
+                    1094,
+                    598,
+                    1094,
+                    625,
+                    1075,
+                    625
+                ],
+                "score": 0.31,
+                "latex": "\\mathbf { k }"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    73.0,
+                    817.0,
+                    73.0,
+                    817.0,
+                    108.0,
+                    296.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    339.0,
+                    852.0,
+                    339.0,
+                    852.0,
+                    381.0,
+                    293.0,
+                    381.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    709.0,
+                    490.0,
+                    709.0,
+                    490.0,
+                    745.0,
+                    296.0,
+                    745.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    831.0,
+                    2084.0,
+                    871.0,
+                    2084.0,
+                    871.0,
+                    2125.0,
+                    831.0,
+                    2125.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    414.0,
+                    1404.0,
+                    414.0,
+                    1404.0,
+                    448.0,
+                    296.0,
+                    448.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    446.0,
+                    1404.0,
+                    446.0,
+                    1404.0,
+                    476.0,
+                    293.0,
+                    476.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    474.0,
+                    1403.0,
+                    474.0,
+                    1403.0,
+                    510.0,
+                    293.0,
+                    510.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    508.0,
+                    1404.0,
+                    508.0,
+                    1404.0,
+                    538.0,
+                    294.0,
+                    538.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    536.0,
+                    1405.0,
+                    536.0,
+                    1405.0,
+                    568.0,
+                    293.0,
+                    568.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    564.0,
+                    1405.0,
+                    564.0,
+                    1405.0,
+                    601.0,
+                    293.0,
+                    601.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    596.0,
+                    1074.0,
+                    596.0,
+                    1074.0,
+                    632.0,
+                    293.0,
+                    632.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1095.0,
+                    596.0,
+                    1404.0,
+                    596.0,
+                    1404.0,
+                    632.0,
+                    1095.0,
+                    632.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    628.0,
+                    1346.0,
+                    628.0,
+                    1346.0,
+                    662.0,
+                    294.0,
+                    662.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    226.0,
+                    1405.0,
+                    226.0,
+                    1405.0,
+                    269.0,
+                    294.0,
+                    269.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    259.0,
+                    903.0,
+                    259.0,
+                    903.0,
+                    296.0,
+                    295.0,
+                    296.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    762.0,
+                    1404.0,
+                    762.0,
+                    1404.0,
+                    798.0,
+                    295.0,
+                    798.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    321.0,
+                    793.0,
+                    849.0,
+                    793.0,
+                    849.0,
+                    827.0,
+                    321.0,
+                    827.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    298.0,
+                    1244.0,
+                    1402.0,
+                    1244.0,
+                    1402.0,
+                    1280.0,
+                    298.0,
+                    1280.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    1275.0,
+                    1402.0,
+                    1275.0,
+                    1402.0,
+                    1313.0,
+                    323.0,
+                    1313.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    1308.0,
+                    1126.0,
+                    1308.0,
+                    1126.0,
+                    1341.0,
+                    322.0,
+                    1341.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    1154.0,
+                    1403.0,
+                    1154.0,
+                    1403.0,
+                    1190.0,
+                    297.0,
+                    1190.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    1187.0,
+                    897.0,
+                    1187.0,
+                    897.0,
+                    1220.0,
+                    322.0,
+                    1220.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    850.0,
+                    1405.0,
+                    850.0,
+                    1405.0,
+                    890.0,
+                    295.0,
+                    890.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    321.0,
+                    883.0,
+                    1326.0,
+                    883.0,
+                    1326.0,
+                    918.0,
+                    321.0,
+                    918.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1365.0,
+                    1405.0,
+                    1365.0,
+                    1405.0,
+                    1401.0,
+                    295.0,
+                    1401.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    321.0,
+                    1394.0,
+                    1405.0,
+                    1394.0,
+                    1405.0,
+                    1433.0,
+                    321.0,
+                    1433.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    321.0,
+                    1425.0,
+                    1405.0,
+                    1425.0,
+                    1405.0,
+                    1463.0,
+                    321.0,
+                    1463.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    319.0,
+                    1453.0,
+                    1403.0,
+                    1453.0,
+                    1403.0,
+                    1498.0,
+                    319.0,
+                    1498.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    321.0,
+                    1490.0,
+                    836.0,
+                    1490.0,
+                    836.0,
+                    1520.0,
+                    321.0,
+                    1520.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    941.0,
+                    1404.0,
+                    941.0,
+                    1404.0,
+                    981.0,
+                    295.0,
+                    981.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    324.0,
+                    973.0,
+                    495.0,
+                    973.0,
+                    495.0,
+                    1006.0,
+                    324.0,
+                    1006.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1546.0,
+                    1403.0,
+                    1546.0,
+                    1403.0,
+                    1584.0,
+                    294.0,
+                    1584.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    320.0,
+                    1579.0,
+                    1405.0,
+                    1579.0,
+                    1405.0,
+                    1615.0,
+                    320.0,
+                    1615.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    1610.0,
+                    598.0,
+                    1610.0,
+                    598.0,
+                    1643.0,
+                    323.0,
+                    1643.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1668.0,
+                    1402.0,
+                    1668.0,
+                    1402.0,
+                    1706.0,
+                    296.0,
+                    1706.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    1701.0,
+                    1102.0,
+                    1701.0,
+                    1102.0,
+                    1733.0,
+                    322.0,
+                    1733.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    1759.0,
+                    1403.0,
+                    1759.0,
+                    1403.0,
+                    1795.0,
+                    297.0,
+                    1795.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    324.0,
+                    1790.0,
+                    1143.0,
+                    1790.0,
+                    1143.0,
+                    1825.0,
+                    324.0,
+                    1825.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1971.0,
+                    1405.0,
+                    1971.0,
+                    1405.0,
+                    2007.0,
+                    296.0,
+                    2007.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    2002.0,
+                    1284.0,
+                    2002.0,
+                    1284.0,
+                    2038.0,
+                    322.0,
+                    2038.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1845.0,
+                    1404.0,
+                    1845.0,
+                    1404.0,
+                    1892.0,
+                    293.0,
+                    1892.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    320.0,
+                    1878.0,
+                    1405.0,
+                    1878.0,
+                    1405.0,
+                    1921.0,
+                    320.0,
+                    1921.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    320.0,
+                    1909.0,
+                    1197.0,
+                    1909.0,
+                    1197.0,
+                    1949.0,
+                    320.0,
+                    1949.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1030.0,
+                    1404.0,
+                    1030.0,
+                    1404.0,
+                    1071.0,
+                    292.0,
+                    1071.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    324.0,
+                    1066.0,
+                    1404.0,
+                    1066.0,
+                    1404.0,
+                    1100.0,
+                    324.0,
+                    1100.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    320.0,
+                    1097.0,
+                    707.0,
+                    1097.0,
+                    707.0,
+                    1127.0,
+                    320.0,
+                    1127.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 9,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 2,
+                "poly": [
+                    299,
+                    75,
+                    815,
+                    75,
+                    815,
+                    105,
+                    299,
+                    105
+                ],
+                "score": 0.881
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    300,
+                    1276,
+                    1401,
+                    1276,
+                    1401,
+                    1371,
+                    300,
+                    1371
+                ],
+                "score": 0.858
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    1129,
+                    1403,
+                    1129,
+                    1403,
+                    1254,
+                    299,
+                    1254
+                ],
+                "score": 0.821
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    230,
+                    1405,
+                    230,
+                    1405,
+                    384,
+                    298,
+                    384
+                ],
+                "score": 0.815
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    836,
+                    2088,
+                    862,
+                    2088,
+                    862,
+                    2113,
+                    836,
+                    2113
+                ],
+                "score": 0.795
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    295,
+                    1043,
+                    1400,
+                    1043,
+                    1400,
+                    1109,
+                    295,
+                    1109
+                ],
+                "score": 0.786
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1392,
+                    1405,
+                    1392,
+                    1405,
+                    1517,
+                    298,
+                    1517
+                ],
+                "score": 0.785
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    301,
+                    810,
+                    1398,
+                    810,
+                    1398,
+                    907,
+                    301,
+                    907
+                ],
+                "score": 0.738
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    296,
+                    1540,
+                    1401,
+                    1540,
+                    1401,
+                    1604,
+                    296,
+                    1604
+                ],
+                "score": 0.733
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    295,
+                    1626,
+                    1403,
+                    1626,
+                    1403,
+                    1691,
+                    295,
+                    1691
+                ],
+                "score": 0.728
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    292,
+                    1712,
+                    1404,
+                    1712,
+                    1404,
+                    1777,
+                    292,
+                    1777
+                ],
+                "score": 0.72
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    694,
+                    1403,
+                    694,
+                    1403,
+                    789,
+                    297,
+                    789
+                ],
+                "score": 0.716
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    927,
+                    1405,
+                    927,
+                    1405,
+                    1022,
+                    297,
+                    1022
+                ],
+                "score": 0.714
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    609,
+                    1401,
+                    609,
+                    1401,
+                    673,
+                    297,
+                    673
+                ],
+                "score": 0.65
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    523,
+                    1403,
+                    523,
+                    1403,
+                    587,
+                    297,
+                    587
+                ],
+                "score": 0.623
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    296,
+                    1971,
+                    1399,
+                    1971,
+                    1399,
+                    2035,
+                    296,
+                    2035
+                ],
+                "score": 0.57
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    292,
+                    1798,
+                    1402,
+                    1798,
+                    1402,
+                    1863,
+                    292,
+                    1863
+                ],
+                "score": 0.561
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    293,
+                    1885,
+                    1401,
+                    1885,
+                    1401,
+                    1949,
+                    293,
+                    1949
+                ],
+                "score": 0.545
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    406,
+                    1404,
+                    406,
+                    1404,
+                    500,
+                    299,
+                    500
+                ],
+                "score": 0.505
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1383,
+                    1196,
+                    1403,
+                    1196,
+                    1403,
+                    1219,
+                    1383,
+                    1219
+                ],
+                "score": 0.52,
+                "latex": "="
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    73.0,
+                    817.0,
+                    73.0,
+                    817.0,
+                    108.0,
+                    295.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    832.0,
+                    2085.0,
+                    868.0,
+                    2085.0,
+                    868.0,
+                    2125.0,
+                    832.0,
+                    2125.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1277.0,
+                    1403.0,
+                    1277.0,
+                    1403.0,
+                    1311.0,
+                    296.0,
+                    1311.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    321.0,
+                    1308.0,
+                    1405.0,
+                    1308.0,
+                    1405.0,
+                    1343.0,
+                    321.0,
+                    1343.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    320.0,
+                    1333.0,
+                    411.0,
+                    1333.0,
+                    411.0,
+                    1374.0,
+                    320.0,
+                    1374.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1129.0,
+                    1404.0,
+                    1129.0,
+                    1404.0,
+                    1165.0,
+                    294.0,
+                    1165.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    1163.0,
+                    1405.0,
+                    1163.0,
+                    1405.0,
+                    1196.0,
+                    322.0,
+                    1196.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    319.0,
+                    1192.0,
+                    1382.0,
+                    1192.0,
+                    1382.0,
+                    1228.0,
+                    319.0,
+                    1228.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1404.0,
+                    1192.0,
+                    1407.0,
+                    1192.0,
+                    1407.0,
+                    1228.0,
+                    1404.0,
+                    1228.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    1222.0,
+                    493.0,
+                    1222.0,
+                    493.0,
+                    1258.0,
+                    322.0,
+                    1258.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    228.0,
+                    1402.0,
+                    228.0,
+                    1402.0,
+                    265.0,
+                    295.0,
+                    265.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    261.0,
+                    1402.0,
+                    261.0,
+                    1402.0,
+                    294.0,
+                    323.0,
+                    294.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    320.0,
+                    287.0,
+                    1405.0,
+                    287.0,
+                    1405.0,
+                    327.0,
+                    320.0,
+                    327.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    319.0,
+                    829.0,
+                    319.0,
+                    829.0,
+                    355.0,
+                    322.0,
+                    355.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    834.0,
+                    318.0,
+                    1404.0,
+                    318.0,
+                    1404.0,
+                    359.0,
+                    834.0,
+                    359.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    349.0,
+                    1328.0,
+                    349.0,
+                    1328.0,
+                    388.0,
+                    322.0,
+                    388.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1042.0,
+                    1405.0,
+                    1042.0,
+                    1405.0,
+                    1081.0,
+                    294.0,
+                    1081.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    1075.0,
+                    1319.0,
+                    1075.0,
+                    1319.0,
+                    1110.0,
+                    322.0,
+                    1110.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1392.0,
+                    1405.0,
+                    1392.0,
+                    1405.0,
+                    1430.0,
+                    292.0,
+                    1430.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    321.0,
+                    1424.0,
+                    1405.0,
+                    1424.0,
+                    1405.0,
+                    1459.0,
+                    321.0,
+                    1459.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    318.0,
+                    1457.0,
+                    1406.0,
+                    1457.0,
+                    1406.0,
+                    1490.0,
+                    318.0,
+                    1490.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    318.0,
+                    1483.0,
+                    398.0,
+                    1483.0,
+                    398.0,
+                    1518.0,
+                    318.0,
+                    1518.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    812.0,
+                    1404.0,
+                    812.0,
+                    1404.0,
+                    846.0,
+                    297.0,
+                    846.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    840.0,
+                    1404.0,
+                    840.0,
+                    1404.0,
+                    879.0,
+                    322.0,
+                    879.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    872.0,
+                    1232.0,
+                    872.0,
+                    1232.0,
+                    910.0,
+                    322.0,
+                    910.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    290.0,
+                    1532.0,
+                    1407.0,
+                    1532.0,
+                    1407.0,
+                    1584.0,
+                    290.0,
+                    1584.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    325.0,
+                    1572.0,
+                    620.0,
+                    1572.0,
+                    620.0,
+                    1604.0,
+                    325.0,
+                    1604.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1622.0,
+                    1406.0,
+                    1622.0,
+                    1406.0,
+                    1668.0,
+                    294.0,
+                    1668.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    1659.0,
+                    771.0,
+                    1659.0,
+                    771.0,
+                    1692.0,
+                    323.0,
+                    1692.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1714.0,
+                    1403.0,
+                    1714.0,
+                    1403.0,
+                    1750.0,
+                    295.0,
+                    1750.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    321.0,
+                    1746.0,
+                    922.0,
+                    1746.0,
+                    922.0,
+                    1778.0,
+                    321.0,
+                    1778.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    694.0,
+                    1404.0,
+                    694.0,
+                    1404.0,
+                    732.0,
+                    295.0,
+                    732.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    321.0,
+                    722.0,
+                    1404.0,
+                    722.0,
+                    1404.0,
+                    765.0,
+                    321.0,
+                    765.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    757.0,
+                    676.0,
+                    757.0,
+                    676.0,
+                    790.0,
+                    322.0,
+                    790.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    928.0,
+                    1403.0,
+                    928.0,
+                    1403.0,
+                    962.0,
+                    295.0,
+                    962.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    960.0,
+                    1405.0,
+                    960.0,
+                    1405.0,
+                    994.0,
+                    323.0,
+                    994.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    320.0,
+                    987.0,
+                    399.0,
+                    987.0,
+                    399.0,
+                    1023.0,
+                    320.0,
+                    1023.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    609.0,
+                    1402.0,
+                    609.0,
+                    1402.0,
+                    645.0,
+                    297.0,
+                    645.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    640.0,
+                    775.0,
+                    640.0,
+                    775.0,
+                    674.0,
+                    323.0,
+                    674.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    522.0,
+                    1403.0,
+                    522.0,
+                    1403.0,
+                    560.0,
+                    296.0,
+                    560.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    554.0,
+                    1030.0,
+                    554.0,
+                    1030.0,
+                    588.0,
+                    322.0,
+                    588.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1967.0,
+                    1404.0,
+                    1967.0,
+                    1404.0,
+                    2009.0,
+                    292.0,
+                    2009.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    321.0,
+                    2002.0,
+                    1285.0,
+                    2002.0,
+                    1285.0,
+                    2036.0,
+                    321.0,
+                    2036.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1798.0,
+                    1404.0,
+                    1798.0,
+                    1404.0,
+                    1836.0,
+                    294.0,
+                    1836.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    321.0,
+                    1828.0,
+                    1405.0,
+                    1828.0,
+                    1405.0,
+                    1867.0,
+                    321.0,
+                    1867.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1885.0,
+                    1404.0,
+                    1885.0,
+                    1404.0,
+                    1922.0,
+                    295.0,
+                    1922.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    1917.0,
+                    1404.0,
+                    1917.0,
+                    1404.0,
+                    1950.0,
+                    322.0,
+                    1950.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    403.0,
+                    1404.0,
+                    403.0,
+                    1404.0,
+                    443.0,
+                    293.0,
+                    443.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    321.0,
+                    437.0,
+                    1405.0,
+                    437.0,
+                    1405.0,
+                    475.0,
+                    321.0,
+                    475.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    319.0,
+                    465.0,
+                    398.0,
+                    465.0,
+                    398.0,
+                    502.0,
+                    319.0,
+                    502.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 10,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    296,
+                    1642,
+                    1405,
+                    1642,
+                    1405,
+                    1894,
+                    296,
+                    1894
+                ],
+                "score": 0.981
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1907,
+                    1403,
+                    1907,
+                    1403,
+                    2035,
+                    298,
+                    2035
+                ],
+                "score": 0.974
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    966,
+                    1404,
+                    966,
+                    1404,
+                    1060,
+                    298,
+                    1060
+                ],
+                "score": 0.974
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    1455,
+                    1404,
+                    1455,
+                    1404,
+                    1520,
+                    299,
+                    1520
+                ],
+                "score": 0.955
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    463,
+                    1210,
+                    1233,
+                    1210,
+                    1233,
+                    1287,
+                    463,
+                    1287
+                ],
+                "score": 0.951
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    577,
+                    1376,
+                    1122,
+                    1376,
+                    1122,
+                    1438,
+                    577,
+                    1438
+                ],
+                "score": 0.948
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    295,
+                    1132,
+                    1401,
+                    1132,
+                    1401,
+                    1195,
+                    295,
+                    1195
+                ],
+                "score": 0.947
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1300,
+                    1398,
+                    1300,
+                    1398,
+                    1363,
+                    297,
+                    1363
+                ],
+                "score": 0.947
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    344,
+                    1403,
+                    344,
+                    1403,
+                    498,
+                    298,
+                    498
+                ],
+                "score": 0.94
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    640,
+                    1074,
+                    1060,
+                    1074,
+                    1060,
+                    1117,
+                    640,
+                    1117
+                ],
+                "score": 0.939
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    228,
+                    1402,
+                    228,
+                    1402,
+                    323,
+                    299,
+                    323
+                ],
+                "score": 0.933
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    518,
+                    1404,
+                    518,
+                    1404,
+                    613,
+                    298,
+                    613
+                ],
+                "score": 0.928
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    717,
+                    1403,
+                    717,
+                    1403,
+                    780,
+                    299,
+                    780
+                ],
+                "score": 0.921
+            },
+            {
+                "category_id": 9,
+                "poly": [
+                    1352,
+                    1083,
+                    1399,
+                    1083,
+                    1399,
+                    1113,
+                    1352,
+                    1113
+                ],
+                "score": 0.909
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    633,
+                    1399,
+                    633,
+                    1399,
+                    698,
+                    298,
+                    698
+                ],
+                "score": 0.907
+            },
+            {
+                "category_id": 9,
+                "poly": [
+                    1352,
+                    1232,
+                    1400,
+                    1232,
+                    1400,
+                    1264,
+                    1352,
+                    1264
+                ],
+                "score": 0.901
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    299,
+                    840,
+                    502,
+                    840,
+                    502,
+                    876,
+                    299,
+                    876
+                ],
+                "score": 0.9
+            },
+            {
+                "category_id": 9,
+                "poly": [
+                    1352,
+                    1390,
+                    1400,
+                    1390,
+                    1400,
+                    1421,
+                    1352,
+                    1421
+                ],
+                "score": 0.899
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    298,
+                    1556,
+                    1198,
+                    1556,
+                    1198,
+                    1620,
+                    298,
+                    1620
+                ],
+                "score": 0.884
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    300,
+                    75,
+                    816,
+                    75,
+                    816,
+                    105,
+                    300,
+                    105
+                ],
+                "score": 0.883
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    301,
+                    909,
+                    722,
+                    909,
+                    722,
+                    942,
+                    301,
+                    942
+                ],
+                "score": 0.868
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    836,
+                    2088,
+                    865,
+                    2088,
+                    865,
+                    2113,
+                    836,
+                    2113
+                ],
+                "score": 0.855
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    461,
+                    1208,
+                    1236,
+                    1208,
+                    1236,
+                    1287,
+                    461,
+                    1287
+                ],
+                "score": 0.94,
+                "latex": "\\mathbf { A } _ { \\mathrm { L } } ^ { ( 4 ) } = \\operatorname* { l i m } _ { C ^ { ( 3 ) } \\to \\infty } \\mathbf { A } ^ { ( 4 ) } \\left( \\operatorname* { l i m } _ { C ^ { ( 2 ) } \\to \\infty } \\mathbf { A } ^ { ( 3 ) } \\left( \\operatorname* { l i m } _ { C ^ { ( 1 ) } \\to \\infty } \\mathbf { A } ^ { ( 2 ) } \\left( \\mathbf { A } ^ { ( 1 ) } ( \\mathbf { X } ) \\right) \\right) \\right) ."
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    443,
+                    1941,
+                    620,
+                    1941,
+                    620,
+                    1976,
+                    443,
+                    1976
+                ],
+                "score": 0.94,
+                "latex": "\\ell \\in \\{ 1 , \\ldots , L \\}"
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    573,
+                    1376,
+                    1120,
+                    1376,
+                    1120,
+                    1438,
+                    573,
+                    1438
+                ],
+                "score": 0.93,
+                "latex": "{ \\bf A } _ { \\mathrm { M } } ^ { ( 4 ) } = \\operatorname* { l i m } _ { n  \\infty } { \\bf A } ^ { ( 4 ) } ( { \\bf A } ^ { ( 3 ) } ( { \\bf A } ^ { ( 2 ) } ( { \\bf A } ^ { ( 1 ) } ( { \\bf X } ) ) ) ) ,"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    372,
+                    1453,
+                    549,
+                    1453,
+                    549,
+                    1490,
+                    372,
+                    1490
+                ],
+                "score": 0.93,
+                "latex": "C ^ { ( \\ell ) } = C ^ { ( \\ell ) } ( n )"
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    637,
+                    1074,
+                    1061,
+                    1074,
+                    1061,
+                    1118,
+                    637,
+                    1118
+                ],
+                "score": 0.93,
+                "latex": "{ \\bf A } ^ { ( 4 ) } = { \\bf A } ^ { ( 4 ) } ( { \\bf A } ^ { ( 3 ) } ( { \\bf A } ^ { ( 2 ) } ( { \\bf A } ^ { ( 1 ) } ( { \\bf X } ) ) ) )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    297,
+                    1938,
+                    388,
+                    1938,
+                    388,
+                    1975,
+                    297,
+                    1975
+                ],
+                "score": 0.92,
+                "latex": "C ^ { ( \\ell ) } ( n )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    549,
+                    1675,
+                    679,
+                    1675,
+                    679,
+                    1707,
+                    549,
+                    1707
+                ],
+                "score": 0.91,
+                "latex": "C ^ { ( 2 ) }  \\infty"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    352,
+                    1675,
+                    481,
+                    1675,
+                    481,
+                    1707,
+                    352,
+                    1707
+                ],
+                "score": 0.91,
+                "latex": "C ^ { ( 1 ) } \\to \\infty"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    752,
+                    1462,
+                    842,
+                    1462,
+                    842,
+                    1486,
+                    752,
+                    1486
+                ],
+                "score": 0.87,
+                "latex": "n \\to \\infty"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    972,
+                    1946,
+                    1070,
+                    1946,
+                    1070,
+                    1971,
+                    972,
+                    1971
+                ],
+                "score": 0.86,
+                "latex": "n  \\infty"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    834.0,
+                    508.0,
+                    834.0,
+                    508.0,
+                    883.0,
+                    292.0,
+                    883.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1552.0,
+                    1200.0,
+                    1552.0,
+                    1200.0,
+                    1593.0,
+                    294.0,
+                    1593.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    362.0,
+                    1591.0,
+                    646.0,
+                    1591.0,
+                    646.0,
+                    1620.0,
+                    362.0,
+                    1620.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    73.0,
+                    817.0,
+                    73.0,
+                    817.0,
+                    108.0,
+                    296.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    908.0,
+                    727.0,
+                    908.0,
+                    727.0,
+                    945.0,
+                    295.0,
+                    945.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    832.0,
+                    2085.0,
+                    870.0,
+                    2085.0,
+                    870.0,
+                    2124.0,
+                    832.0,
+                    2124.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1642.0,
+                    1405.0,
+                    1642.0,
+                    1405.0,
+                    1681.0,
+                    291.0,
+                    1681.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    290.0,
+                    1671.0,
+                    351.0,
+                    1671.0,
+                    351.0,
+                    1717.0,
+                    290.0,
+                    1717.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    482.0,
+                    1671.0,
+                    548.0,
+                    1671.0,
+                    548.0,
+                    1717.0,
+                    482.0,
+                    1717.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    680.0,
+                    1671.0,
+                    1407.0,
+                    1671.0,
+                    1407.0,
+                    1717.0,
+                    680.0,
+                    1717.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1708.0,
+                    1403.0,
+                    1708.0,
+                    1403.0,
+                    1745.0,
+                    293.0,
+                    1745.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1740.0,
+                    1405.0,
+                    1740.0,
+                    1405.0,
+                    1774.0,
+                    295.0,
+                    1774.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1770.0,
+                    1403.0,
+                    1770.0,
+                    1403.0,
+                    1805.0,
+                    293.0,
+                    1805.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1801.0,
+                    1405.0,
+                    1801.0,
+                    1405.0,
+                    1835.0,
+                    294.0,
+                    1835.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1832.0,
+                    1403.0,
+                    1832.0,
+                    1403.0,
+                    1866.0,
+                    294.0,
+                    1866.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1863.0,
+                    944.0,
+                    1863.0,
+                    944.0,
+                    1894.0,
+                    296.0,
+                    1894.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1906.0,
+                    1407.0,
+                    1906.0,
+                    1407.0,
+                    1944.0,
+                    293.0,
+                    1944.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1934.0,
+                    296.0,
+                    1934.0,
+                    296.0,
+                    1984.0,
+                    292.0,
+                    1984.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    389.0,
+                    1934.0,
+                    442.0,
+                    1934.0,
+                    442.0,
+                    1984.0,
+                    389.0,
+                    1984.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    621.0,
+                    1934.0,
+                    971.0,
+                    1934.0,
+                    971.0,
+                    1984.0,
+                    621.0,
+                    1984.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1071.0,
+                    1934.0,
+                    1409.0,
+                    1934.0,
+                    1409.0,
+                    1984.0,
+                    1071.0,
+                    1984.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1973.0,
+                    1405.0,
+                    1973.0,
+                    1405.0,
+                    2007.0,
+                    294.0,
+                    2007.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    2001.0,
+                    1405.0,
+                    2001.0,
+                    1405.0,
+                    2039.0,
+                    292.0,
+                    2039.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    967.0,
+                    1404.0,
+                    967.0,
+                    1404.0,
+                    1001.0,
+                    294.0,
+                    1001.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    997.0,
+                    1404.0,
+                    997.0,
+                    1404.0,
+                    1031.0,
+                    294.0,
+                    1031.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1026.0,
+                    920.0,
+                    1026.0,
+                    920.0,
+                    1065.0,
+                    292.0,
+                    1065.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1450.0,
+                    371.0,
+                    1450.0,
+                    371.0,
+                    1494.0,
+                    293.0,
+                    1494.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    550.0,
+                    1450.0,
+                    751.0,
+                    1450.0,
+                    751.0,
+                    1494.0,
+                    550.0,
+                    1494.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    843.0,
+                    1450.0,
+                    1405.0,
+                    1450.0,
+                    1405.0,
+                    1494.0,
+                    843.0,
+                    1494.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    1486.0,
+                    781.0,
+                    1486.0,
+                    781.0,
+                    1518.0,
+                    297.0,
+                    1518.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1134.0,
+                    1403.0,
+                    1134.0,
+                    1403.0,
+                    1166.0,
+                    295.0,
+                    1166.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1162.0,
+                    1264.0,
+                    1162.0,
+                    1264.0,
+                    1197.0,
+                    293.0,
+                    1197.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1300.0,
+                    1404.0,
+                    1300.0,
+                    1404.0,
+                    1335.0,
+                    295.0,
+                    1335.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1330.0,
+                    1182.0,
+                    1330.0,
+                    1182.0,
+                    1366.0,
+                    295.0,
+                    1366.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    346.0,
+                    1404.0,
+                    346.0,
+                    1404.0,
+                    379.0,
+                    296.0,
+                    379.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    376.0,
+                    1402.0,
+                    376.0,
+                    1402.0,
+                    409.0,
+                    323.0,
+                    409.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    404.0,
+                    1404.0,
+                    404.0,
+                    1404.0,
+                    441.0,
+                    323.0,
+                    441.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    321.0,
+                    430.0,
+                    1403.0,
+                    430.0,
+                    1403.0,
+                    475.0,
+                    321.0,
+                    475.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    466.0,
+                    1020.0,
+                    466.0,
+                    1020.0,
+                    500.0,
+                    323.0,
+                    500.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    227.0,
+                    1406.0,
+                    227.0,
+                    1406.0,
+                    266.0,
+                    293.0,
+                    266.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    260.0,
+                    1404.0,
+                    260.0,
+                    1404.0,
+                    296.0,
+                    322.0,
+                    296.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    319.0,
+                    285.0,
+                    399.0,
+                    285.0,
+                    399.0,
+                    326.0,
+                    319.0,
+                    326.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    514.0,
+                    1404.0,
+                    514.0,
+                    1404.0,
+                    559.0,
+                    291.0,
+                    559.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    551.0,
+                    1404.0,
+                    551.0,
+                    1404.0,
+                    585.0,
+                    323.0,
+                    585.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    581.0,
+                    1127.0,
+                    581.0,
+                    1127.0,
+                    614.0,
+                    323.0,
+                    614.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    715.0,
+                    827.0,
+                    715.0,
+                    827.0,
+                    752.0,
+                    293.0,
+                    752.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    870.0,
+                    713.0,
+                    1181.0,
+                    713.0,
+                    1181.0,
+                    752.0,
+                    870.0,
+                    752.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1218.0,
+                    717.0,
+                    1405.0,
+                    717.0,
+                    1405.0,
+                    754.0,
+                    1218.0,
+                    754.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    749.0,
+                    609.0,
+                    749.0,
+                    609.0,
+                    779.0,
+                    322.0,
+                    779.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    629.0,
+                    1404.0,
+                    629.0,
+                    1404.0,
+                    672.0,
+                    294.0,
+                    672.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    666.0,
+                    1044.0,
+                    666.0,
+                    1044.0,
+                    699.0,
+                    322.0,
+                    699.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 11,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    336,
+                    1405,
+                    336,
+                    1405,
+                    591,
+                    297,
+                    591
+                ],
+                "score": 0.984
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1103,
+                    1404,
+                    1103,
+                    1404,
+                    1274,
+                    297,
+                    1274
+                ],
+                "score": 0.981
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    296,
+                    1869,
+                    1405,
+                    1869,
+                    1405,
+                    2036,
+                    296,
+                    2036
+                ],
+                "score": 0.978
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    1387,
+                    1406,
+                    1387,
+                    1406,
+                    1485,
+                    299,
+                    1485
+                ],
+                "score": 0.974
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    228,
+                    1403,
+                    228,
+                    1403,
+                    323,
+                    299,
+                    323
+                ],
+                "score": 0.965
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    535,
+                    1750,
+                    1162,
+                    1750,
+                    1162,
+                    1851,
+                    535,
+                    1851
+                ],
+                "score": 0.959
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    293,
+                    1497,
+                    1402,
+                    1497,
+                    1402,
+                    1563,
+                    293,
+                    1563
+                ],
+                "score": 0.956
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    539,
+                    1288,
+                    1158,
+                    1288,
+                    1158,
+                    1371,
+                    539,
+                    1371
+                ],
+                "score": 0.955
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    468,
+                    1575,
+                    1232,
+                    1575,
+                    1232,
+                    1655,
+                    468,
+                    1655
+                ],
+                "score": 0.951
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    296,
+                    1669,
+                    1400,
+                    1669,
+                    1400,
+                    1734,
+                    296,
+                    1734
+                ],
+                "score": 0.951
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    604,
+                    1401,
+                    604,
+                    1401,
+                    669,
+                    297,
+                    669
+                ],
+                "score": 0.95
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    357,
+                    856,
+                    1301,
+                    856,
+                    1301,
+                    957,
+                    357,
+                    957
+                ],
+                "score": 0.943
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    410,
+                    687,
+                    1169,
+                    687,
+                    1169,
+                    786,
+                    410,
+                    786
+                ],
+                "score": 0.941
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    500,
+                    792,
+                    757,
+                    792,
+                    757,
+                    848,
+                    500,
+                    848
+                ],
+                "score": 0.931
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    982,
+                    376,
+                    982,
+                    376,
+                    1011,
+                    298,
+                    1011
+                ],
+                "score": 0.922
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    505,
+                    1035,
+                    717,
+                    1035,
+                    717,
+                    1075,
+                    505,
+                    1075
+                ],
+                "score": 0.919
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    299,
+                    75,
+                    816,
+                    75,
+                    816,
+                    105,
+                    299,
+                    105
+                ],
+                "score": 0.911
+            },
+            {
+                "category_id": 9,
+                "poly": [
+                    1352,
+                    1038,
+                    1400,
+                    1038,
+                    1400,
+                    1069,
+                    1352,
+                    1069
+                ],
+                "score": 0.9
+            },
+            {
+                "category_id": 9,
+                "poly": [
+                    1351,
+                    1786,
+                    1400,
+                    1786,
+                    1400,
+                    1818,
+                    1351,
+                    1818
+                ],
+                "score": 0.896
+            },
+            {
+                "category_id": 9,
+                "poly": [
+                    1351,
+                    1302,
+                    1400,
+                    1302,
+                    1400,
+                    1334,
+                    1351,
+                    1334
+                ],
+                "score": 0.896
+            },
+            {
+                "category_id": 9,
+                "poly": [
+                    1351,
+                    721,
+                    1400,
+                    721,
+                    1400,
+                    751,
+                    1351,
+                    751
+                ],
+                "score": 0.884
+            },
+            {
+                "category_id": 9,
+                "poly": [
+                    1351,
+                    1588,
+                    1400,
+                    1588,
+                    1400,
+                    1618,
+                    1351,
+                    1618
+                ],
+                "score": 0.88
+            },
+            {
+                "category_id": 9,
+                "poly": [
+                    1351,
+                    804,
+                    1400,
+                    804,
+                    1400,
+                    835,
+                    1351,
+                    835
+                ],
+                "score": 0.878
+            },
+            {
+                "category_id": 9,
+                "poly": [
+                    1351,
+                    892,
+                    1401,
+                    892,
+                    1401,
+                    923,
+                    1351,
+                    923
+                ],
+                "score": 0.872
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    836,
+                    2087,
+                    865,
+                    2087,
+                    865,
+                    2113,
+                    836,
+                    2113
+                ],
+                "score": 0.868
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    545,
+                    429,
+                    645,
+                    429,
+                    645,
+                    476,
+                    545,
+                    476
+                ],
+                "score": 0.95,
+                "latex": "\\mathbf { a } _ { j } ^ { ( \\ell ) } ( \\mathbf { X } ^ { \\prime } )"
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    354,
+                    683,
+                    1301,
+                    683,
+                    1301,
+                    961,
+                    354,
+                    961
+                ],
+                "score": 0.95,
+                "latex": "\\begin{array} { c } { { A _ { i , \\mu } ^ { ( 1 ) } = f _ { i , \\mu } ^ { ( 1 ) } ( x ) = \\displaystyle \\frac { \\sigma _ { \\mathbf { w } } } { \\sqrt { C ^ { ( 0 ) } } } \\sum _ { j = 1 } ^ { C ^ { ( 0 ) } } \\sum _ { \\nu \\in \\mu \\mathbb { h } \\mathrm { p a t e h } } \\epsilon _ { i , j , \\mu , \\nu } ^ { ( 1 ) } x _ { j , \\nu } + b _ { i } ^ { ( 1 ) } , \\quad i \\in \\mathbb { N } } } \\\\ { { g _ { i , \\mu } ^ { ( \\ell ) } ( x ) = \\phi \\left( f _ { i , \\mu } ^ { ( \\ell ) } ( x ) \\right) } } \\\\ { { A _ { i , \\mu } ^ { ( \\ell + 1 ) } = f _ { i , \\mu } ^ { ( \\ell + 1 ) } ( x ) = \\displaystyle \\frac { \\sigma _ { \\mathbf { w } } } { \\sqrt { C ^ { ( \\ell ) } ( n ) } } \\sum _ { j = 1 } ^ { C ^ { ( \\ell ) } } \\sum _ { \\nu \\in \\mu \\mathrm { t h } \\mathrm { p a t e h } } \\epsilon _ { i , j , \\mu , \\nu } ^ { ( \\ell + 1 ) } g _ { j , \\nu } ^ { ( \\ell ) } ( x ) + b _ { i } ^ { ( \\ell + 1 ) } , \\quad i \\in \\mathbb { N } } } \\end{array}"
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    535,
+                    1747,
+                    1163,
+                    1747,
+                    1163,
+                    1853,
+                    535,
+                    1853
+                ],
+                "score": 0.94,
+                "latex": "\\mathcal { T } ^ { ( \\ell ) } \\left( \\mathcal { L } , \\alpha \\right) [ n ] = \\frac { 1 } { \\sqrt { C ^ { ( \\ell - 1 ) } ( n ) } } \\sum _ { j = 1 } ^ { C ^ { ( \\ell - 1 ) } ( n ) } \\gamma _ { j } ^ { ( \\ell ) } \\left( \\mathcal { L } , \\alpha \\right) [ n ] ."
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    372,
+                    1388,
+                    752,
+                    1388,
+                    752,
+                    1425,
+                    372,
+                    1425
+                ],
+                "score": 0.94,
+                "latex": "\\mathcal { L } \\subset \\mathcal { X } \\times \\mathbb { N } \\times \\{ 1 , \\ldots , H ^ { ( \\ell ) } D ^ { ( \\ell ) } \\}"
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    469,
+                    1575,
+                    1228,
+                    1575,
+                    1228,
+                    1656,
+                    469,
+                    1656
+                ],
+                "score": 0.93,
+                "latex": "\\gamma _ { j } ^ { ( \\ell ) } \\left( \\mathcal { L } , \\alpha \\right) [ n ] : = \\sigma _ { \\mathrm { w } } \\sum _ { ( x , i , \\mu ) \\in \\mathcal { L } } \\alpha ^ { ( x , i , \\mu ) } \\sum _ { \\nu \\in \\mu \\mathrm { t h p a t c h } } \\epsilon _ { i , j , \\mu , \\nu } ^ { ( \\ell ) } g _ { j , \\nu } ^ { ( \\ell - 1 ) } ( x ) [ n ] ,"
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    538,
+                    1288,
+                    1159,
+                    1288,
+                    1159,
+                    1372,
+                    538,
+                    1372
+                ],
+                "score": 0.93,
+                "latex": "\\mathcal { T } ^ { ( \\ell ) } \\left( \\mathcal { L } , \\alpha \\right) \\left[ n \\right] = \\sum _ { ( x , i , \\mu ) \\in \\mathcal { L } } \\alpha ^ { ( x , i , \\mu ) } \\left[ f _ { i , \\mu } ^ { ( \\ell ) } ( x ) [ n ] - b _ { i } ^ { ( \\ell ) } \\right] ."
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    943,
+                    1869,
+                    1107,
+                    1869,
+                    1107,
+                    1906,
+                    943,
+                    1906
+                ],
+                "score": 0.93,
+                "latex": "{ \\mathcal { T } } ^ { ( \\ell ) } \\left( { \\mathcal { L } } , \\alpha \\right) [ n ]"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    401,
+                    429,
+                    494,
+                    429,
+                    494,
+                    474,
+                    401,
+                    474
+                ],
+                "score": 0.92,
+                "latex": "{ \\bf a } _ { j } ^ { ( \\ell ) } ( { \\bf X } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1203,
+                    1196,
+                    1393,
+                    1196,
+                    1393,
+                    1242,
+                    1203,
+                    1242
+                ],
+                "score": 0.92,
+                "latex": "f _ { i , \\mu } ^ { ( \\ell ) } ( x ) [ n ] - b _ { i } ^ { ( \\ell ) }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    520,
+                    1965,
+                    678,
+                    1965,
+                    678,
+                    2009,
+                    520,
+                    2009
+                ],
+                "score": 0.92,
+                "latex": "\\gamma _ { j } ^ { ( \\ell ) } \\left( \\mathcal { L } , \\alpha \\right) [ n ]"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    464,
+                    1238,
+                    549,
+                    1238,
+                    549,
+                    1270,
+                    464,
+                    1270
+                ],
+                "score": 0.9,
+                "latex": "\\alpha ^ { ( x , i , \\mu ) }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1219,
+                    1934,
+                    1404,
+                    1934,
+                    1404,
+                    1967,
+                    1219,
+                    1967
+                ],
+                "score": 0.9,
+                "latex": "\\ell \\in \\{ 2 , \\ldots , L +"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1083,
+                    1423,
+                    1116,
+                    1423,
+                    1116,
+                    1456,
+                    1083,
+                    1456
+                ],
+                "score": 0.89,
+                "latex": "[ n ]"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    530,
+                    266,
+                    625,
+                    266,
+                    625,
+                    290,
+                    530,
+                    290
+                ],
+                "score": 0.87,
+                "latex": "n  \\infty"
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    506,
+                    1035,
+                    720,
+                    1035,
+                    720,
+                    1074,
+                    506,
+                    1074
+                ],
+                "score": 0.84,
+                "latex": "\\epsilon _ { i , j , \\mu , \\nu } \\sim \\mathcal { N } ( 0 , 1 ) ."
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1150,
+                    1938,
+                    1169,
+                    1938,
+                    1169,
+                    1961,
+                    1150,
+                    1961
+                ],
+                "score": 0.79,
+                "latex": "n"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1176,
+                    1975,
+                    1192,
+                    1975,
+                    1192,
+                    2005,
+                    1176,
+                    2005
+                ],
+                "score": 0.79,
+                "latex": "j"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    931,
+                    1427,
+                    951,
+                    1427,
+                    951,
+                    1455,
+                    931,
+                    1455
+                ],
+                "score": 0.79,
+                "latex": "\\mu"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    298,
+                    1427,
+                    309,
+                    1427,
+                    309,
+                    1450,
+                    298,
+                    1450
+                ],
+                "score": 0.73,
+                "latex": "i"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    649,
+                    1460,
+                    667,
+                    1460,
+                    667,
+                    1480,
+                    649,
+                    1480
+                ],
+                "score": 0.72,
+                "latex": "n"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    297,
+                    1971,
+                    327,
+                    1971,
+                    327,
+                    2006,
+                    297,
+                    2006
+                ],
+                "score": 0.58,
+                "latex": "1 \\}"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    73.0,
+                    817.0,
+                    73.0,
+                    817.0,
+                    108.0,
+                    295.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    831.0,
+                    2085.0,
+                    870.0,
+                    2085.0,
+                    870.0,
+                    2123.0,
+                    831.0,
+                    2123.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    334.0,
+                    1406.0,
+                    334.0,
+                    1406.0,
+                    375.0,
+                    294.0,
+                    375.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    369.0,
+                    1406.0,
+                    369.0,
+                    1406.0,
+                    403.0,
+                    295.0,
+                    403.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    399.0,
+                    1405.0,
+                    399.0,
+                    1405.0,
+                    434.0,
+                    296.0,
+                    434.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    430.0,
+                    400.0,
+                    430.0,
+                    400.0,
+                    475.0,
+                    291.0,
+                    475.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    495.0,
+                    430.0,
+                    544.0,
+                    430.0,
+                    544.0,
+                    475.0,
+                    495.0,
+                    475.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    646.0,
+                    430.0,
+                    1407.0,
+                    430.0,
+                    1407.0,
+                    475.0,
+                    646.0,
+                    475.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    463.0,
+                    1406.0,
+                    463.0,
+                    1406.0,
+                    503.0,
+                    294.0,
+                    503.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    497.0,
+                    1403.0,
+                    497.0,
+                    1403.0,
+                    531.0,
+                    294.0,
+                    531.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    526.0,
+                    1402.0,
+                    526.0,
+                    1402.0,
+                    565.0,
+                    294.0,
+                    565.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    558.0,
+                    553.0,
+                    558.0,
+                    553.0,
+                    589.0,
+                    296.0,
+                    589.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1105.0,
+                    1404.0,
+                    1105.0,
+                    1404.0,
+                    1139.0,
+                    296.0,
+                    1139.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1138.0,
+                    1404.0,
+                    1138.0,
+                    1404.0,
+                    1168.0,
+                    296.0,
+                    1168.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1167.0,
+                    1405.0,
+                    1167.0,
+                    1405.0,
+                    1202.0,
+                    292.0,
+                    1202.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    286.0,
+                    1195.0,
+                    1202.0,
+                    1195.0,
+                    1202.0,
+                    1248.0,
+                    286.0,
+                    1248.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1394.0,
+                    1195.0,
+                    1407.0,
+                    1195.0,
+                    1407.0,
+                    1248.0,
+                    1394.0,
+                    1248.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.75,
+                    1237.5,
+                    560.75,
+                    1237.5,
+                    560.75,
+                    1276.5,
+                    291.75,
+                    1276.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1867.0,
+                    942.0,
+                    1867.0,
+                    942.0,
+                    1910.0,
+                    291.0,
+                    1910.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1108.0,
+                    1867.0,
+                    1407.0,
+                    1867.0,
+                    1407.0,
+                    1910.0,
+                    1108.0,
+                    1910.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1904.0,
+                    1405.0,
+                    1904.0,
+                    1405.0,
+                    1937.0,
+                    295.0,
+                    1937.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1932.0,
+                    1149.0,
+                    1932.0,
+                    1149.0,
+                    1969.0,
+                    293.0,
+                    1969.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1170.0,
+                    1932.0,
+                    1218.0,
+                    1932.0,
+                    1218.0,
+                    1969.0,
+                    1170.0,
+                    1969.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1961.0,
+                    296.0,
+                    1961.0,
+                    296.0,
+                    2013.0,
+                    291.0,
+                    2013.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    328.0,
+                    1961.0,
+                    519.0,
+                    1961.0,
+                    519.0,
+                    2013.0,
+                    328.0,
+                    2013.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    679.0,
+                    1961.0,
+                    1175.0,
+                    1961.0,
+                    1175.0,
+                    2013.0,
+                    679.0,
+                    2013.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1193.0,
+                    1961.0,
+                    1411.0,
+                    1961.0,
+                    1411.0,
+                    2013.0,
+                    1193.0,
+                    2013.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    2001.0,
+                    1406.0,
+                    2001.0,
+                    1406.0,
+                    2037.0,
+                    294.0,
+                    2037.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1385.0,
+                    371.0,
+                    1385.0,
+                    371.0,
+                    1427.0,
+                    293.0,
+                    1427.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    753.0,
+                    1385.0,
+                    1404.0,
+                    1385.0,
+                    1404.0,
+                    1427.0,
+                    753.0,
+                    1427.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1420.0,
+                    297.0,
+                    1420.0,
+                    297.0,
+                    1455.0,
+                    294.0,
+                    1455.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    310.0,
+                    1420.0,
+                    930.0,
+                    1420.0,
+                    930.0,
+                    1455.0,
+                    310.0,
+                    1455.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    952.0,
+                    1420.0,
+                    1082.0,
+                    1420.0,
+                    1082.0,
+                    1455.0,
+                    952.0,
+                    1455.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1117.0,
+                    1420.0,
+                    1403.0,
+                    1420.0,
+                    1403.0,
+                    1455.0,
+                    1117.0,
+                    1455.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1448.0,
+                    648.0,
+                    1448.0,
+                    648.0,
+                    1488.0,
+                    295.0,
+                    1488.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    668.0,
+                    1448.0,
+                    680.0,
+                    1448.0,
+                    680.0,
+                    1488.0,
+                    668.0,
+                    1488.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    228.0,
+                    1407.0,
+                    228.0,
+                    1407.0,
+                    265.0,
+                    293.0,
+                    265.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    261.0,
+                    529.0,
+                    261.0,
+                    529.0,
+                    295.0,
+                    294.0,
+                    295.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    626.0,
+                    261.0,
+                    1407.0,
+                    261.0,
+                    1407.0,
+                    295.0,
+                    626.0,
+                    295.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    288.0,
+                    352.0,
+                    288.0,
+                    352.0,
+                    328.0,
+                    292.0,
+                    328.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1494.0,
+                    1405.0,
+                    1494.0,
+                    1405.0,
+                    1538.0,
+                    292.0,
+                    1538.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1531.0,
+                    873.0,
+                    1531.0,
+                    873.0,
+                    1562.0,
+                    295.0,
+                    1562.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1667.0,
+                    1405.0,
+                    1667.0,
+                    1405.0,
+                    1707.0,
+                    294.0,
+                    1707.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1701.0,
+                    460.0,
+                    1701.0,
+                    460.0,
+                    1735.0,
+                    295.0,
+                    1735.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    604.0,
+                    1406.0,
+                    604.0,
+                    1406.0,
+                    640.0,
+                    296.0,
+                    640.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    635.0,
+                    1267.0,
+                    635.0,
+                    1267.0,
+                    671.0,
+                    295.0,
+                    671.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    974.0,
+                    383.0,
+                    974.0,
+                    383.0,
+                    1018.0,
+                    295.0,
+                    1018.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 12,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    295,
+                    1004,
+                    1406,
+                    1004,
+                    1406,
+                    1243,
+                    295,
+                    1243
+                ],
+                "score": 0.979
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    325,
+                    1444,
+                    1325,
+                    1444,
+                    1325,
+                    1664,
+                    325,
+                    1664
+                ],
+                "score": 0.97
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    296,
+                    441,
+                    1406,
+                    441,
+                    1406,
+                    553,
+                    296,
+                    553
+                ],
+                "score": 0.969
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    897,
+                    1404,
+                    897,
+                    1404,
+                    992,
+                    299,
+                    992
+                ],
+                "score": 0.968
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    567,
+                    1403,
+                    567,
+                    1403,
+                    668,
+                    297,
+                    668
+                ],
+                "score": 0.966
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    310,
+                    1248,
+                    1329,
+                    1248,
+                    1329,
+                    1365,
+                    310,
+                    1365
+                ],
+                "score": 0.949
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    290,
+                    227,
+                    1404,
+                    227,
+                    1404,
+                    295,
+                    290,
+                    295
+                ],
+                "score": 0.947
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1372,
+                    1412,
+                    1372,
+                    1412,
+                    1435,
+                    297,
+                    1435
+                ],
+                "score": 0.941
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    293,
+                    300,
+                    1643,
+                    300,
+                    1643,
+                    405,
+                    293,
+                    405
+                ],
+                "score": 0.938
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1674,
+                    1409,
+                    1674,
+                    1409,
+                    1753,
+                    297,
+                    1753
+                ],
+                "score": 0.935
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    848,
+                    1902,
+                    1493,
+                    1902,
+                    1493,
+                    2005,
+                    848,
+                    2005
+                ],
+                "score": 0.931
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    298,
+                    73,
+                    816,
+                    73,
+                    816,
+                    106,
+                    298,
+                    106
+                ],
+                "score": 0.922
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    297,
+                    1761,
+                    1402,
+                    1761,
+                    1402,
+                    1861,
+                    297,
+                    1861
+                ],
+                "score": 0.894
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    835,
+                    2087,
+                    865,
+                    2087,
+                    865,
+                    2113,
+                    835,
+                    2113
+                ],
+                "score": 0.862
+            },
+            {
+                "category_id": 9,
+                "poly": [
+                    1351,
+                    1291,
+                    1401,
+                    1291,
+                    1401,
+                    1323,
+                    1351,
+                    1323
+                ],
+                "score": 0.855
+            },
+            {
+                "category_id": 9,
+                "poly": [
+                    1351,
+                    405,
+                    1401,
+                    405,
+                    1401,
+                    434,
+                    1351,
+                    434
+                ],
+                "score": 0.85
+            },
+            {
+                "category_id": 9,
+                "poly": [
+                    1351,
+                    2004,
+                    1402,
+                    2004,
+                    1402,
+                    2034,
+                    1351,
+                    2034
+                ],
+                "score": 0.835
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    355,
+                    682,
+                    822,
+                    682,
+                    822,
+                    883,
+                    355,
+                    883
+                ],
+                "score": 0.811
+            },
+            {
+                "category_id": 9,
+                "poly": [
+                    1351,
+                    1862,
+                    1402,
+                    1862,
+                    1402,
+                    1892,
+                    1351,
+                    1892
+                ],
+                "score": 0.748
+            },
+            {
+                "category_id": 9,
+                "poly": [
+                    1352,
+                    1603,
+                    1401,
+                    1603,
+                    1401,
+                    1634,
+                    1352,
+                    1634
+                ],
+                "score": 0.396
+            },
+            {
+                "category_id": 9,
+                "poly": [
+                    1352,
+                    1603,
+                    1401,
+                    1603,
+                    1401,
+                    1634,
+                    1352,
+                    1634
+                ],
+                "score": 0.333
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    355,
+                    682,
+                    822,
+                    682,
+                    822,
+                    883,
+                    355,
+                    883
+                ],
+                "score": 0.234
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    325,
+                    1443,
+                    1330,
+                    1443,
+                    1330,
+                    1666,
+                    325,
+                    1666
+                ],
+                "score": 0.94,
+                "latex": "\\begin{array} { r l } { \\mathbb { E } \\left[ \\left| \\frac { \\sigma _ { \\mathrm { w } } } { \\sqrt { C ^ { ( \\ell - 1 ) } } } \\sum _ { j = 1 } ^ { C ^ { ( \\ell - 1 ) } ( n ) } \\displaystyle \\sum _ { \\nu \\in \\mu \\mathrm { t h } \\mathrm { p a t e h } } \\epsilon _ { i , j , \\mu , \\nu } ^ { ( \\ell ) } g _ { j , \\nu } ^ { ( \\ell - 1 ) } ( x ) [ n ] \\right| ^ { 8 } \\right] } & { } \\\\ { = \\frac { 2 ^ { 4 } \\Gamma ( 4 + 1 / 2 ) } { \\Gamma ( 1 / 2 ) } \\mathbb { E } \\left[ \\left| \\frac { \\sigma _ { \\mathrm { w } } ^ { 2 } } { C ^ { ( \\ell - 1 ) } ( n ) } \\| g _ { j \\in \\{ 1 , \\dots , C ^ { ( \\ell - 1 ) } ( n ) \\} , \\nu \\in \\mu \\mathrm { t h } \\mathrm { p a t c h } } ^ { ( \\ell - 1 ) } ( x ) [ n ] \\| _ { 2 } ^ { 2 } \\right| ^ { 4 } \\right] . } \\end{array}"
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    313,
+                    1243,
+                    1334,
+                    1243,
+                    1334,
+                    1371,
+                    313,
+                    1371
+                ],
+                "score": 0.94,
+                "latex": "\\mathbb { E } \\left[ | f _ { i , \\mu } ^ { ( \\ell ) } ( x ) [ n ] | ^ { 8 } \\right] \\leq 2 ^ { 8 - 1 } \\mathbb { E } \\left[ | b _ { i } ^ { ( \\ell ) } | ^ { 8 } + \\left| \\frac { \\sigma _ { \\mathbf { w } } } { \\sqrt { C ^ { ( \\ell - 1 ) } } } \\sum _ { j = 1 } ^ { C ^ { ( \\ell - 1 ) } ( n ) } \\sum _ { \\nu \\in \\mu \\mathrm { t h } \\mathrm { p a t e h } } \\epsilon _ { i , j , \\mu , \\nu } ^ { ( \\ell ) } g _ { j , \\nu } ^ { ( \\ell - 1 ) } ( x ) [ n ] \\right| ^ { 8 } \\right]"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1246,
+                    1067,
+                    1360,
+                    1067,
+                    1360,
+                    1113,
+                    1246,
+                    1113
+                ],
+                "score": 0.94,
+                "latex": "f _ { i , \\mu } ^ { ( t ) } ( x ) [ n ]"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    576,
+                    601,
+                    689,
+                    601,
+                    689,
+                    640,
+                    576,
+                    640
+                ],
+                "score": 0.93,
+                "latex": "\\mathcal { N } \\left( 0 , \\sigma _ { * } ^ { 2 } \\right)"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    407,
+                    1138,
+                    524,
+                    1138,
+                    524,
+                    1184,
+                    407,
+                    1184
+                ],
+                "score": 0.93,
+                "latex": "f _ { j , \\mu } ^ { ( 1 ) } ( x ) [ n ]"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1179,
+                    567,
+                    1343,
+                    567,
+                    1343,
+                    604,
+                    1179,
+                    604
+                ],
+                "score": 0.93,
+                "latex": "{ \\mathcal { T } } ^ { ( \\ell ) } \\left( { \\mathcal { L } } , \\alpha \\right) [ n ]"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    377,
+                    1670,
+                    762,
+                    1670,
+                    762,
+                    1719,
+                    377,
+                    1719
+                ],
+                "score": 0.92,
+                "latex": "g _ { j \\in \\{ 1 , \\dots , C ^ { ( \\ell - 1 ) } ( n ) \\} , \\nu \\in \\mu \\mathrm { t h p a t c h } } ^ { ( \\ell - 1 ) } ( x ) [ n ]"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    298,
+                    1718,
+                    519,
+                    1718,
+                    519,
+                    1753,
+                    298,
+                    1753
+                ],
+                "score": 0.91,
+                "latex": "\\{ 1 , \\ldots , C ^ { ( \\ell - 1 ) } ( n ) \\}"
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    310,
+                    1757,
+                    1426,
+                    1757,
+                    1426,
+                    2010,
+                    310,
+                    2010
+                ],
+                "score": 0.91,
+                "latex": "\\begin{array} { r } { \\frac { 1 } { \\sum ^ { ( \\ell - 1 ) } \\left( n \\right) } \\| g _ { j \\in \\{ 1 , \\dots , C ^ { ( \\ell - 1 ) } ( n ) \\} , \\nu \\in \\mu \\mathrm { t h p a t e h } } ^ { ( \\ell - 1 ) } ( x ) [ n ] \\| _ { 2 } ^ { 2 } = \\frac { 1 } { C ^ { ( \\ell - 1 ) } ( n ) } \\displaystyle \\sum _ { j = 1 } ^ { C ^ { ( \\ell - 1 ) } ( n ) } \\sum _ { \\nu \\in \\mu \\mathrm { t h p a t e h } } \\left( g _ { j , \\nu } ^ { ( \\ell - 1 ) } ( x ) [ n ] \\right) ^ { 2 } } \\\\ { \\leq \\frac { 1 } { C ^ { ( \\ell - 1 ) } ( n ) } \\displaystyle \\sum _ { j = 1 } ^ { C ^ { ( \\ell - 1 ) } ( n ) } \\sum _ { \\nu \\in \\mu \\mathrm { t h p a t e h } } ^ { C ^ { ( \\ell - 1 ) } ( n ) } \\left( c + m \\big | f _ { j , \\nu } ^ { ( \\ell - 1 ) } ( x ) \\big | \\right. } \\end{array}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    797,
+                    1107,
+                    876,
+                    1107,
+                    876,
+                    1137,
+                    797,
+                    1137
+                ],
+                "score": 0.9,
+                "latex": "n \\in \\mathbb N"
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    355,
+                    679,
+                    821,
+                    679,
+                    821,
+                    887,
+                    355,
+                    887
+                ],
+                "score": 0.89,
+                "latex": "\\begin{array} { r l } & { \\mathrm { a ) ~ \\mathbb { E } } _ { n } [ \\gamma _ { j } ^ { ( \\ell ) } \\gamma _ { j ^ { \\prime } } ^ { ( \\ell ) } ] = 0 } \\\\ & { \\mathrm { b ) ~ \\operatorname* { l i m } _ { n  \\infty } \\mathbb { E } } _ { n } [ ( \\gamma _ { j } ^ { ( \\ell ) } ) ^ { 2 } ( \\gamma _ { j ^ { \\prime } } ^ { ( \\ell ) } ) ^ { 2 } ] = \\sigma _ { * } ^ { 4 } } \\\\ & { \\mathrm { c ) ~ \\mathbb { E } } _ { n } [ | \\gamma _ { j } ^ { ( \\ell ) } | ^ { 3 } ] = o ( \\sqrt { C ^ { ( \\ell ) } ( n ) } ) } \\end{array}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1246,
+                    481,
+                    1289,
+                    481,
+                    1289,
+                    512,
+                    1246,
+                    512
+                ],
+                "score": 0.88,
+                "latex": "\\mathcal { L } _ { \\mathcal { X } }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    570,
+                    1722,
+                    661,
+                    1722,
+                    661,
+                    1752,
+                    570,
+                    1752
+                ],
+                "score": 0.88,
+                "latex": "\\nu \\in \\mu \\mathrm { t h }"
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    310,
+                    296,
+                    1428,
+                    296,
+                    1428,
+                    408,
+                    310,
+                    408
+                ],
+                "score": 0.86,
+                "latex": " \\overset { ( \\ell ) } { j } ( \\mathcal { L } , \\alpha ) [ n ] : = \\sigma _ { \\mathrm { w } } \\sum _ { ( x , i , \\mu ) \\in \\mathcal { L } } \\alpha ^ { ( x , i , \\mu ) } \\sum _ { \\nu \\in \\mu \\mathrm { t h } \\mathrm { p a t e h } } \\epsilon _ { i , j , \\mu , \\nu } ^ { ( \\ell ) } \\phi ( \\frac { \\sigma _ { \\mathrm { w } } } { \\sqrt { C ^ { ( \\ell - 2 ) } ( n ) } } \\sum _ { k = 1 } ^ { C ^ { ( \\ell - 2 ) } ( n ) } \\sum _ { \\xi \\in \\nu \\mathrm { t h } \\mathrm { p a t e h } } \\epsilon _ { j , k , \\nu ; \\xi } ^ { ( \\ell - 1 ) } g _ { k , \\xi } ^ { ( \\ell - 2 ) } ) ,"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    326,
+                    1179,
+                    385,
+                    1179,
+                    385,
+                    1208,
+                    326,
+                    1208
+                ],
+                "score": 0.86,
+                "latex": "\\ell - 1"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1349,
+                    1677,
+                    1404,
+                    1677,
+                    1404,
+                    1710,
+                    1349,
+                    1710
+                ],
+                "score": 0.86,
+                "latex": "j \\in"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    495,
+                    522,
+                    516,
+                    522,
+                    516,
+                    548,
+                    495,
+                    548
+                ],
+                "score": 0.72,
+                "latex": "\\mathcal { L }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    297,
+                    474,
+                    1154,
+                    474,
+                    1154,
+                    525,
+                    297,
+                    525
+                ],
+                "score": 0.71,
+                "latex": "\\left\\{ g _ { k , \\xi } ^ { ( \\ell - 2 ) } ( x ) [ n ] : k \\in \\{ 1 , \\ldots , C ^ { ( \\ell - 2 ) } \\} , \\xi \\in \\left\\{ 1 , \\ldots , H ^ { ( \\ell - 2 ) } D ^ { ( \\ell - 2 ) } \\right\\} , x \\in \\mathcal { L } _ { \\mathcal { X } } \\right\\}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    387,
+                    1211,
+                    402,
+                    1211,
+                    402,
+                    1237,
+                    387,
+                    1237
+                ],
+                "score": 0.67,
+                "latex": "\\ell"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    875,
+                    1180,
+                    890,
+                    1180,
+                    890,
+                    1206,
+                    875,
+                    1206
+                ],
+                "score": 0.44,
+                "latex": "\\ell"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    72.0,
+                    818.0,
+                    72.0,
+                    818.0,
+                    108.0,
+                    295.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    832.0,
+                    2085.0,
+                    870.0,
+                    2085.0,
+                    870.0,
+                    2124.0,
+                    832.0,
+                    2124.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1006.0,
+                    1406.0,
+                    1006.0,
+                    1406.0,
+                    1040.0,
+                    295.0,
+                    1040.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1037.0,
+                    1404.0,
+                    1037.0,
+                    1404.0,
+                    1071.0,
+                    295.0,
+                    1071.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    284.0,
+                    1055.0,
+                    796.0,
+                    1055.0,
+                    796.0,
+                    1143.0,
+                    284.0,
+                    1143.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    877.0,
+                    1055.0,
+                    1245.0,
+                    1055.0,
+                    1245.0,
+                    1143.0,
+                    877.0,
+                    1143.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1361.0,
+                    1055.0,
+                    1417.0,
+                    1055.0,
+                    1417.0,
+                    1143.0,
+                    1361.0,
+                    1143.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    285.0,
+                    1130.0,
+                    406.0,
+                    1130.0,
+                    406.0,
+                    1196.0,
+                    285.0,
+                    1196.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    525.0,
+                    1130.0,
+                    1413.0,
+                    1130.0,
+                    1413.0,
+                    1196.0,
+                    525.0,
+                    1196.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1175.0,
+                    325.0,
+                    1175.0,
+                    325.0,
+                    1215.0,
+                    292.0,
+                    1215.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    386.0,
+                    1175.0,
+                    874.0,
+                    1175.0,
+                    874.0,
+                    1215.0,
+                    386.0,
+                    1215.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    891.0,
+                    1175.0,
+                    1404.0,
+                    1175.0,
+                    1404.0,
+                    1215.0,
+                    891.0,
+                    1215.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1207.0,
+                    386.0,
+                    1207.0,
+                    386.0,
+                    1247.0,
+                    293.0,
+                    1247.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    403.0,
+                    1207.0,
+                    415.0,
+                    1207.0,
+                    415.0,
+                    1247.0,
+                    403.0,
+                    1247.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    439.0,
+                    1404.0,
+                    439.0,
+                    1404.0,
+                    476.0,
+                    295.0,
+                    476.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1155.0,
+                    454.0,
+                    1245.0,
+                    454.0,
+                    1245.0,
+                    540.0,
+                    1155.0,
+                    540.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1290.0,
+                    454.0,
+                    1415.0,
+                    454.0,
+                    1415.0,
+                    540.0,
+                    1290.0,
+                    540.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    517.0,
+                    494.0,
+                    517.0,
+                    494.0,
+                    555.0,
+                    294.0,
+                    555.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    517.0,
+                    517.0,
+                    527.0,
+                    517.0,
+                    527.0,
+                    555.0,
+                    517.0,
+                    555.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    896.0,
+                    1404.0,
+                    896.0,
+                    1404.0,
+                    932.0,
+                    294.0,
+                    932.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    925.0,
+                    1403.0,
+                    925.0,
+                    1403.0,
+                    966.0,
+                    292.0,
+                    966.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    960.0,
+                    894.0,
+                    960.0,
+                    894.0,
+                    993.0,
+                    294.0,
+                    993.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    563.0,
+                    1178.0,
+                    563.0,
+                    1178.0,
+                    607.0,
+                    292.0,
+                    607.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1344.0,
+                    563.0,
+                    1404.0,
+                    563.0,
+                    1404.0,
+                    607.0,
+                    1344.0,
+                    607.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    602.0,
+                    575.0,
+                    602.0,
+                    575.0,
+                    638.0,
+                    296.0,
+                    638.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    690.0,
+                    602.0,
+                    1405.0,
+                    602.0,
+                    1405.0,
+                    638.0,
+                    690.0,
+                    638.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    633.0,
+                    618.0,
+                    633.0,
+                    618.0,
+                    668.0,
+                    295.0,
+                    668.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    231.0,
+                    1403.0,
+                    231.0,
+                    1403.0,
+                    264.0,
+                    296.0,
+                    264.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    260.0,
+                    1358.0,
+                    260.0,
+                    1358.0,
+                    298.0,
+                    292.0,
+                    298.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1368.0,
+                    1405.0,
+                    1368.0,
+                    1405.0,
+                    1409.0,
+                    294.0,
+                    1409.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1396.0,
+                    369.0,
+                    1396.0,
+                    369.0,
+                    1443.0,
+                    291.0,
+                    1443.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    386.0,
+                    1663.0,
+                    454.0,
+                    1663.0,
+                    454.0,
+                    1697.0,
+                    386.0,
+                    1697.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1672.0,
+                    376.0,
+                    1672.0,
+                    376.0,
+                    1711.0,
+                    296.0,
+                    1711.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1711.0,
+                    297.0,
+                    1711.0,
+                    297.0,
+                    1758.0,
+                    292.0,
+                    1758.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    520.0,
+                    1711.0,
+                    569.0,
+                    1711.0,
+                    569.0,
+                    1758.0,
+                    520.0,
+                    1758.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    662.0,
+                    1711.0,
+                    1289.0,
+                    1711.0,
+                    1289.0,
+                    1758.0,
+                    662.0,
+                    1758.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    361.75,
+                    1666.5,
+                    1408.75,
+                    1666.5,
+                    1408.75,
+                    1723.5,
+                    361.75,
+                    1723.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    347.0,
+                    670.0,
+                    354.0,
+                    670.0,
+                    354.0,
+                    751.0,
+                    347.0,
+                    751.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    344.0,
+                    739.0,
+                    354.0,
+                    739.0,
+                    354.0,
+                    823.0,
+                    344.0,
+                    823.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    822.0,
+                    739.0,
+                    840.0,
+                    739.0,
+                    840.0,
+                    823.0,
+                    822.0,
+                    823.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    351.0,
+                    815.0,
+                    354.0,
+                    815.0,
+                    354.0,
+                    888.0,
+                    351.0,
+                    888.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 13,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1164,
+                    1405,
+                    1164,
+                    1405,
+                    1442,
+                    297,
+                    1442
+                ],
+                "score": 0.981
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    321,
+                    312,
+                    1325,
+                    312,
+                    1325,
+                    567,
+                    321,
+                    567
+                ],
+                "score": 0.967
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    289,
+                    227,
+                    1403,
+                    227,
+                    1403,
+                    295,
+                    289,
+                    295
+                ],
+                "score": 0.951
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    295,
+                    702,
+                    1406,
+                    702,
+                    1406,
+                    1072,
+                    295,
+                    1072
+                ],
+                "score": 0.941
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    455,
+                    628,
+                    1240,
+                    628,
+                    1240,
+                    688,
+                    455,
+                    688
+                ],
+                "score": 0.931
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    298,
+                    74,
+                    816,
+                    74,
+                    816,
+                    105,
+                    298,
+                    105
+                ],
+                "score": 0.917
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    580,
+                    1192,
+                    580,
+                    1192,
+                    614,
+                    299,
+                    614
+                ],
+                "score": 0.913
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    297,
+                    1107,
+                    991,
+                    1107,
+                    991,
+                    1140,
+                    297,
+                    1140
+                ],
+                "score": 0.906
+            },
+            {
+                "category_id": 9,
+                "poly": [
+                    1351,
+                    641,
+                    1400,
+                    641,
+                    1400,
+                    673,
+                    1351,
+                    673
+                ],
+                "score": 0.9
+            },
+            {
+                "category_id": 9,
+                "poly": [
+                    1351,
+                    490,
+                    1401,
+                    490,
+                    1401,
+                    522,
+                    1351,
+                    522
+                ],
+                "score": 0.873
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    836,
+                    2088,
+                    865,
+                    2088,
+                    865,
+                    2112,
+                    836,
+                    2112
+                ],
+                "score": 0.842
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    322,
+                    314,
+                    1328,
+                    314,
+                    1328,
+                    571,
+                    322,
+                    571
+                ],
+                "score": 0.94,
+                "latex": "\\begin{array} { r l } & { \\mathbb { E } \\left[ \\left| \\frac { \\sigma _ { \\mathbf { w } } } { \\sqrt { C ^ { ( \\ell - 1 ) } ( n ) } } \\sum _ { j = 1 } ^ { C ^ { ( \\ell - 1 ) } ( n ) } \\sum _ { \\nu \\in \\mu \\mathrm { t h p a t c h } } \\epsilon _ { i , j , \\mu , \\nu } ^ { ( \\ell ) } g _ { j , \\nu } ^ { ( \\ell - 1 ) } ( x ) [ n ] \\right| ^ { 8 } \\right] } \\\\ & { \\leq \\frac { 1 } { \\left( C ^ { ( \\ell - 1 ) } ( n ) \\right) ^ { 4 } } \\mathbb { E } \\left[ \\left| \\sum _ { j = 1 } ^ { C ^ { ( \\ell - 1 ) } ( n ) } \\sum _ { \\nu \\in \\mu \\mathrm { t h p a c h } } c ^ { 2 } + 2 c m | f _ { j , \\mu } ^ { ( \\ell - 1 ) } ( x ) [ n ] | + m ^ { 2 } | f _ { j , \\mu } ^ { ( \\ell - 1 ) } ( x ) [ n ] | ^ { 2 } \\right| ^ { 4 } \\right] } \\end{array}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    583,
+                    922,
+                    727,
+                    922,
+                    727,
+                    970,
+                    583,
+                    970
+                ],
+                "score": 0.94,
+                "latex": "f _ { j , \\mu } ^ { ( \\ell - 1 ) } ( x ) [ n ]"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    649,
+                    230,
+                    850,
+                    230,
+                    850,
+                    264,
+                    649,
+                    264
+                ],
+                "score": 0.94,
+                "latex": "| \\phi ( u ) | \\leq c + m | u |"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1031,
+                    767,
+                    1175,
+                    767,
+                    1175,
+                    813,
+                    1031,
+                    813
+                ],
+                "score": 0.94,
+                "latex": "f _ { k , \\mu } ^ { ( \\ell - 1 ) } ( x ) [ n ]"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    354,
+                    706,
+                    510,
+                    706,
+                    510,
+                    740,
+                    354,
+                    740
+                ],
+                "score": 0.93,
+                "latex": "p _ { i } \\in \\{ 0 , 1 , 2 \\}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    933,
+                    881,
+                    1121,
+                    881,
+                    1121,
+                    926,
+                    933,
+                    926
+                ],
+                "score": 0.92,
+                "latex": "1 / \\left( C ^ { ( \\ell - 1 ) } ( n ) \\right) ^ { 4 }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    758,
+                    702,
+                    1102,
+                    702,
+                    1102,
+                    740,
+                    758,
+                    740
+                ],
+                "score": 0.91,
+                "latex": "k , l , r , q \\in \\{ 1 , \\dots , C ^ { ( \\ell - 1 ) } ( n ) \\}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    299,
+                    838,
+                    587,
+                    838,
+                    587,
+                    883,
+                    299,
+                    883
+                ],
+                "score": 0.91,
+                "latex": "\\left( 2 C ^ { ( \\ell - 1 ) } ( n ) | \\mu \\mathrm { t h } \\mathrm { p a t c h } | \\right) ^ { 4 }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    556,
+                    707,
+                    699,
+                    707,
+                    699,
+                    738,
+                    556,
+                    738
+                ],
+                "score": 0.9,
+                "latex": "i = { 1 , 2 , 3 , 4 }"
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    457,
+                    627,
+                    1239,
+                    627,
+                    1239,
+                    689,
+                    457,
+                    689
+                ],
+                "score": 0.89,
+                "latex": "\\mathbb { E } \\left[ | f _ { k , \\nu } ^ { ( \\ell - 1 ) } ( x ) [ n ] | ^ { p _ { 1 } } | f _ { l , \\xi } ^ { ( \\ell - 1 ) } ( x ) [ n ] | ^ { p _ { 2 } } | f _ { r , \\pi } ^ { ( \\ell - 1 ) } ( x ) [ n ] | ^ { p _ { 3 } } | f _ { q , \\rho } ^ { ( \\ell - 1 ) } ( x ) [ n ] | ^ { p _ { 4 } } \\right]"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1161,
+                    707,
+                    1334,
+                    707,
+                    1334,
+                    739,
+                    1161,
+                    739
+                ],
+                "score": 0.88,
+                "latex": "\\nu , \\xi , \\pi , \\rho \\in \\mu \\mathrm { t h }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    840,
+                    1410,
+                    902,
+                    1410,
+                    902,
+                    1439,
+                    840,
+                    1439
+                ],
+                "score": 0.85,
+                "latex": "3 . 4 \\%"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    684,
+                    1410,
+                    745,
+                    1410,
+                    745,
+                    1440,
+                    684,
+                    1440
+                ],
+                "score": 0.85,
+                "latex": "2 . 4 \\%"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    980,
+                    934,
+                    996,
+                    934,
+                    996,
+                    964,
+                    980,
+                    964
+                ],
+                "score": 0.84,
+                "latex": "j"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1226,
+                    267,
+                    1246,
+                    267,
+                    1246,
+                    289,
+                    1226,
+                    289
+                ],
+                "score": 0.8,
+                "latex": "x"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1084,
+                    938,
+                    1102,
+                    938,
+                    1102,
+                    959,
+                    1084,
+                    959
+                ],
+                "score": 0.76,
+                "latex": "x"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    925,
+                    744,
+                    944,
+                    744,
+                    944,
+                    764,
+                    925,
+                    764
+                ],
+                "score": 0.76,
+                "latex": "n"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    680,
+                    845,
+                    804,
+                    845,
+                    804,
+                    881,
+                    680,
+                    881
+                ],
+                "score": 0.75,
+                "latex": "| \\mu \\mathrm { t h } \\ p a \\mathrm { t c h } |"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1151,
+                    938,
+                    1171,
+                    938,
+                    1171,
+                    959,
+                    1151,
+                    959
+                ],
+                "score": 0.75,
+                "latex": "n"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1296,
+                    268,
+                    1315,
+                    268,
+                    1315,
+                    289,
+                    1296,
+                    289
+                ],
+                "score": 0.75,
+                "latex": "n"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    790,
+                    895,
+                    810,
+                    895,
+                    810,
+                    917,
+                    790,
+                    917
+                ],
+                "score": 0.73,
+                "latex": "n"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    72.0,
+                    818.0,
+                    72.0,
+                    818.0,
+                    108.0,
+                    296.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1107.0,
+                    994.0,
+                    1107.0,
+                    994.0,
+                    1143.0,
+                    296.0,
+                    1143.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    832.0,
+                    2085.0,
+                    870.0,
+                    2085.0,
+                    870.0,
+                    2124.0,
+                    832.0,
+                    2124.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1167.0,
+                    1406.0,
+                    1167.0,
+                    1406.0,
+                    1199.0,
+                    294.0,
+                    1199.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1197.0,
+                    1405.0,
+                    1197.0,
+                    1405.0,
+                    1229.0,
+                    294.0,
+                    1229.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1226.0,
+                    1406.0,
+                    1226.0,
+                    1406.0,
+                    1261.0,
+                    292.0,
+                    1261.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1257.0,
+                    1402.0,
+                    1257.0,
+                    1402.0,
+                    1290.0,
+                    295.0,
+                    1290.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1286.0,
+                    1405.0,
+                    1286.0,
+                    1405.0,
+                    1323.0,
+                    294.0,
+                    1323.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1317.0,
+                    1402.0,
+                    1317.0,
+                    1402.0,
+                    1350.0,
+                    295.0,
+                    1350.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1346.0,
+                    1403.0,
+                    1346.0,
+                    1403.0,
+                    1384.0,
+                    292.0,
+                    1384.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1378.0,
+                    1405.0,
+                    1378.0,
+                    1405.0,
+                    1415.0,
+                    294.0,
+                    1415.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1411.0,
+                    683.0,
+                    1411.0,
+                    683.0,
+                    1442.0,
+                    294.0,
+                    1442.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    746.0,
+                    1411.0,
+                    839.0,
+                    1411.0,
+                    839.0,
+                    1442.0,
+                    746.0,
+                    1442.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    903.0,
+                    1411.0,
+                    979.0,
+                    1411.0,
+                    979.0,
+                    1442.0,
+                    903.0,
+                    1442.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    229.0,
+                    648.0,
+                    229.0,
+                    648.0,
+                    266.0,
+                    296.0,
+                    266.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    851.0,
+                    229.0,
+                    1404.0,
+                    229.0,
+                    1404.0,
+                    266.0,
+                    851.0,
+                    266.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    258.0,
+                    1225.0,
+                    258.0,
+                    1225.0,
+                    298.0,
+                    293.0,
+                    298.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1247.0,
+                    258.0,
+                    1295.0,
+                    258.0,
+                    1295.0,
+                    298.0,
+                    1247.0,
+                    298.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1316.0,
+                    258.0,
+                    1327.0,
+                    258.0,
+                    1327.0,
+                    298.0,
+                    1316.0,
+                    298.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    698.0,
+                    353.0,
+                    698.0,
+                    353.0,
+                    743.0,
+                    293.0,
+                    743.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    511.0,
+                    698.0,
+                    555.0,
+                    698.0,
+                    555.0,
+                    743.0,
+                    511.0,
+                    743.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    700.0,
+                    698.0,
+                    757.0,
+                    698.0,
+                    757.0,
+                    743.0,
+                    700.0,
+                    743.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1103.0,
+                    698.0,
+                    1160.0,
+                    698.0,
+                    1160.0,
+                    743.0,
+                    1103.0,
+                    743.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1335.0,
+                    698.0,
+                    1407.0,
+                    698.0,
+                    1407.0,
+                    743.0,
+                    1335.0,
+                    743.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    278.0,
+                    735.0,
+                    924.0,
+                    735.0,
+                    924.0,
+                    845.0,
+                    278.0,
+                    845.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    945.0,
+                    735.0,
+                    1030.0,
+                    735.0,
+                    1030.0,
+                    845.0,
+                    945.0,
+                    845.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1176.0,
+                    735.0,
+                    1422.0,
+                    735.0,
+                    1422.0,
+                    845.0,
+                    1176.0,
+                    845.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    588.0,
+                    840.0,
+                    679.0,
+                    840.0,
+                    679.0,
+                    882.0,
+                    588.0,
+                    882.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    805.0,
+                    840.0,
+                    1404.0,
+                    840.0,
+                    1404.0,
+                    882.0,
+                    805.0,
+                    882.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    885.0,
+                    789.0,
+                    885.0,
+                    789.0,
+                    926.0,
+                    293.0,
+                    926.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    811.0,
+                    885.0,
+                    932.0,
+                    885.0,
+                    932.0,
+                    926.0,
+                    811.0,
+                    926.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1122.0,
+                    885.0,
+                    1407.0,
+                    885.0,
+                    1407.0,
+                    926.0,
+                    1122.0,
+                    926.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    902.0,
+                    582.0,
+                    902.0,
+                    582.0,
+                    991.0,
+                    291.0,
+                    991.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    728.0,
+                    902.0,
+                    979.0,
+                    902.0,
+                    979.0,
+                    991.0,
+                    728.0,
+                    991.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    997.0,
+                    902.0,
+                    1083.0,
+                    902.0,
+                    1083.0,
+                    991.0,
+                    997.0,
+                    991.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1103.0,
+                    902.0,
+                    1150.0,
+                    902.0,
+                    1150.0,
+                    991.0,
+                    1103.0,
+                    991.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1172.0,
+                    902.0,
+                    1421.0,
+                    902.0,
+                    1421.0,
+                    991.0,
+                    1172.0,
+                    991.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    978.0,
+                    1404.0,
+                    978.0,
+                    1404.0,
+                    1012.0,
+                    295.0,
+                    1012.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1008.0,
+                    1404.0,
+                    1008.0,
+                    1404.0,
+                    1042.0,
+                    294.0,
+                    1042.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1039.0,
+                    939.0,
+                    1039.0,
+                    939.0,
+                    1073.0,
+                    295.0,
+                    1073.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    577.0,
+                    1193.0,
+                    577.0,
+                    1193.0,
+                    619.0,
+                    295.0,
+                    619.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 14,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 3,
+                "poly": [
+                    323,
+                    814,
+                    1374,
+                    814,
+                    1374,
+                    1313,
+                    323,
+                    1313
+                ],
+                "score": 0.967
+            },
+            {
+                "category_id": 4,
+                "poly": [
+                    296,
+                    1372,
+                    1406,
+                    1372,
+                    1406,
+                    1468,
+                    296,
+                    1468
+                ],
+                "score": 0.95
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    299,
+                    75,
+                    816,
+                    75,
+                    816,
+                    105,
+                    299,
+                    105
+                ],
+                "score": 0.886
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    836,
+                    2088,
+                    865,
+                    2088,
+                    865,
+                    2113,
+                    836,
+                    2113
+                ],
+                "score": 0.844
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1265,
+                    1378,
+                    1283,
+                    1378,
+                    1283,
+                    1402,
+                    1265,
+                    1402
+                ],
+                "score": 0.26,
+                "latex": "\\mathbf { X }"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    352.0,
+                    825.0,
+                    419.0,
+                    825.0,
+                    419.0,
+                    868.0,
+                    352.0,
+                    868.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    797.0,
+                    834.0,
+                    835.0,
+                    834.0,
+                    835.0,
+                    867.0,
+                    797.0,
+                    867.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    869.0,
+                    822.0,
+                    938.0,
+                    822.0,
+                    938.0,
+                    869.0,
+                    869.0,
+                    869.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1315.0,
+                    833.0,
+                    1353.0,
+                    833.0,
+                    1353.0,
+                    867.0,
+                    1315.0,
+                    867.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1279.0,
+                    866.0,
+                    1305.0,
+                    866.0,
+                    1305.0,
+                    892.0,
+                    1279.0,
+                    892.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    351.0,
+                    895.0,
+                    422.0,
+                    895.0,
+                    422.0,
+                    942.0,
+                    351.0,
+                    942.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    870.0,
+                    897.0,
+                    938.0,
+                    897.0,
+                    938.0,
+                    941.0,
+                    870.0,
+                    941.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1242.0,
+                    893.0,
+                    1261.0,
+                    893.0,
+                    1261.0,
+                    911.0,
+                    1242.0,
+                    911.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1206.0,
+                    954.0,
+                    1215.0,
+                    954.0,
+                    1215.0,
+                    964.0,
+                    1206.0,
+                    964.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    321.0,
+                    961.0,
+                    417.0,
+                    961.0,
+                    417.0,
+                    1092.0,
+                    321.0,
+                    1092.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    871.0,
+                    971.0,
+                    937.0,
+                    971.0,
+                    937.0,
+                    1013.0,
+                    871.0,
+                    1013.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    352.0,
+                    1044.0,
+                    419.0,
+                    1044.0,
+                    419.0,
+                    1085.0,
+                    352.0,
+                    1085.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    870.0,
+                    1044.0,
+                    938.0,
+                    1044.0,
+                    938.0,
+                    1086.0,
+                    870.0,
+                    1086.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    352.0,
+                    1116.0,
+                    417.0,
+                    1116.0,
+                    417.0,
+                    1158.0,
+                    352.0,
+                    1158.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    870.0,
+                    1115.0,
+                    929.0,
+                    1115.0,
+                    929.0,
+                    1158.0,
+                    870.0,
+                    1158.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1042.0,
+                    1124.0,
+                    1062.0,
+                    1124.0,
+                    1062.0,
+                    1144.0,
+                    1042.0,
+                    1144.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    483.0,
+                    1158.0,
+                    502.0,
+                    1158.0,
+                    502.0,
+                    1179.0,
+                    483.0,
+                    1179.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    988.0,
+                    1153.0,
+                    1021.0,
+                    1153.0,
+                    1021.0,
+                    1175.0,
+                    988.0,
+                    1175.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    352.0,
+                    1189.0,
+                    420.0,
+                    1189.0,
+                    420.0,
+                    1230.0,
+                    352.0,
+                    1230.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    869.0,
+                    1186.0,
+                    942.0,
+                    1186.0,
+                    942.0,
+                    1232.0,
+                    869.0,
+                    1232.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    948.0,
+                    1193.0,
+                    971.0,
+                    1193.0,
+                    971.0,
+                    1220.0,
+                    948.0,
+                    1220.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    405.0,
+                    1237.0,
+                    471.0,
+                    1237.0,
+                    471.0,
+                    1280.0,
+                    405.0,
+                    1280.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    502.0,
+                    1238.0,
+                    567.0,
+                    1238.0,
+                    567.0,
+                    1280.0,
+                    502.0,
+                    1280.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    600.0,
+                    1237.0,
+                    666.0,
+                    1237.0,
+                    666.0,
+                    1280.0,
+                    600.0,
+                    1280.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    696.0,
+                    1234.0,
+                    765.0,
+                    1234.0,
+                    765.0,
+                    1281.0,
+                    696.0,
+                    1281.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    796.0,
+                    1237.0,
+                    860.0,
+                    1237.0,
+                    860.0,
+                    1280.0,
+                    796.0,
+                    1280.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    923.0,
+                    1237.0,
+                    989.0,
+                    1237.0,
+                    989.0,
+                    1280.0,
+                    923.0,
+                    1280.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1020.0,
+                    1237.0,
+                    1087.0,
+                    1237.0,
+                    1087.0,
+                    1280.0,
+                    1020.0,
+                    1280.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1118.0,
+                    1237.0,
+                    1184.0,
+                    1237.0,
+                    1184.0,
+                    1280.0,
+                    1118.0,
+                    1280.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1215.0,
+                    1234.0,
+                    1283.0,
+                    1234.0,
+                    1283.0,
+                    1281.0,
+                    1215.0,
+                    1281.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1315.0,
+                    1237.0,
+                    1380.0,
+                    1237.0,
+                    1380.0,
+                    1280.0,
+                    1315.0,
+                    1280.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    486.0,
+                    1268.0,
+                    776.0,
+                    1268.0,
+                    776.0,
+                    1318.0,
+                    486.0,
+                    1318.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    990.0,
+                    1270.0,
+                    1305.0,
+                    1270.0,
+                    1305.0,
+                    1316.0,
+                    990.0,
+                    1316.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    438.0,
+                    1189.0,
+                    444.0,
+                    1189.0,
+                    444.0,
+                    1219.0,
+                    438.0,
+                    1219.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1371.0,
+                    1264.0,
+                    1371.0,
+                    1264.0,
+                    1409.0,
+                    294.0,
+                    1409.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1284.0,
+                    1371.0,
+                    1406.0,
+                    1371.0,
+                    1406.0,
+                    1409.0,
+                    1284.0,
+                    1409.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1400.0,
+                    1404.0,
+                    1400.0,
+                    1404.0,
+                    1438.0,
+                    295.0,
+                    1438.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1431.0,
+                    1141.0,
+                    1431.0,
+                    1141.0,
+                    1470.0,
+                    295.0,
+                    1470.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    73.0,
+                    817.0,
+                    73.0,
+                    817.0,
+                    108.0,
+                    295.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    832.0,
+                    2085.0,
+                    871.0,
+                    2085.0,
+                    871.0,
+                    2123.0,
+                    832.0,
+                    2123.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 15,
+            "width": 1700,
+            "height": 2200
+        }
+    }
+]
\ No newline at end of file
diff --git a/parse/train/CuQoImkKkIj/CuQoImkKkIj.md b/parse/train/CuQoImkKkIj/CuQoImkKkIj.md
new file mode 100644
index 0000000000000000000000000000000000000000..3155f6277c86f0d4b9b32f044bb31a6f857a2b23
--- /dev/null
+++ b/parse/train/CuQoImkKkIj/CuQoImkKkIj.md
@@ -0,0 +1,336 @@
+# Robust and differentially private mean estimation
+
+Xiyang Liu, Weihao Kong, Sham Kakade, Sewoong Oh
+
+Paul G. Allen School of Computer Science and Engineering, University of Washington {xiyangl,whkong,sham,sewoong}@cs.washington.edu
+
+# Abstract
+
+In statistical learning and analysis from shared data, which is increasingly widely adopted in platforms such as federated learning and meta-learning, there are two major concerns: privacy and robustness. Each participating individual should be able to contribute without the fear of leaking one’s sensitive information. At the same time, the system should be robust in the presence of malicious participants inserting corrupted data. Recent algorithmic advances in learning from shared data focus on either one of these threats, leaving the system vulnerable to the other. We bridge this gap for the canonical problem of estimating the mean from i.i.d. samples. We introduce PRIME, which is the first efficient algorithm that achieves both privacy and robustness for a wide range of distributions. We further complement this result with a novel exponential time algorithm that improves the sample complexity of PRIME, achieving a near-optimal guarantee and matching a known lower bound for (non-robust) private mean estimation. This proves that there is no extra statistical cost to simultaneously guaranteeing privacy and robustness.
+
+# 1 Introduction
+
+When releasing database statistics on a collection of entries from individuals, we would ideally like to make it impossible to reverse-engineer each individual’s potentially sensitive information. Privacy-preserving techniques add just enough randomness tailored to the statistical task to guarantee protection. At the same time, it is becoming increasingly common to apply such techniques to databases collected from multiple sources, not all of which can be trusted. Emerging data access frameworks, such as federated analyses across users’ devices or data silos [50], make it easier to temper with such collected datasets, leaving private statistical analyses vulnerable to a malicious corruption of a fraction of the data.
+
+Differential privacy has emerged as a widely accepted de facto measure of privacy, which is now a standard in releasing the statistics of the U.S. Census data [2] statistics and also deployed in real-world commercial systems [74, 40, 41]. A statistical analysis is said to be differentially private (DP) if the likelihood of the (randomized) outcome does not change significantly when a single arbitrary entry is added/removed (formally defined in $\ S 1 . 2 )$ . This provides a strong privacy guarantee: even a powerful adversary who knows all the other entries in the database cannot confidently identify whether a particular individual is participating in the database based on the outcome of the analysis. This ensures plausible deniability, central to protecting an individual’s privacy.
+
+In this paper, we focus on one of the most canonical problems in statistics: estimating the mean of a distribution from i.i.d. samples. For distributions with unbounded support, such as sub-Gaussian and heavy-tailed distributions, fundamental trade-offs between accuracy, sample size, and privacy have only recently been identified [58, 52, 54, 3] and efficient private estimators proposed. However, these approaches are brittle when a fraction of the data is corrupted, posing a real threat, referred to as data poisoning attacks [19, 79]. In defense of such attacks, robust (but not necessarily private) statistics has emerged as a popular setting of recent algorithmic and mathematical breakthroughs [73, 30].
+
+One might be misled into thinking that privacy ensures robustness since DP guarantees that a single outlier cannot change the estimation too much. This intuition is true only in a low dimension; each sample has to be an obvious outlier to significantly change the mean. However, in a high dimension, each corrupted data point can look perfectly uncorrupted but still shift the mean significant when colluding together (e.g., see Fig. 1). Focusing on the canonical problem of mean estimation, we introduce novel algorithms that achieve robustness and privacy simultaneously even when a fraction of data is corrupted arbitrarily. For such algorithms, there is a fundamental question of interest: do we need more samples to make private mean estimation also robust against adversarial corruption?
+
+Sub-Gaussian distributions. If we can afford exponential run-time in the dimension, robustness can be achieved without extra cost in sample complexity. We introduce a novel estimator that (i) satisfies $( \varepsilon , \delta )$ -DP, $( i i )$ achieves near-optimal robustness under $\alpha$ -fraction of corrupted data, achieving accuracy of $O ( \alpha \sqrt { \log ( 1 / \alpha ) } )$ nearly matching the fundamental lower bound of $\Omega ( \alpha )$ that holds even for a (non-private) robust mean estimation with infinite samples, and $( i i i )$ achieves near-optimal sample complexity matching that of a fundamental lower bound for a (non-robust) private mean estimation as shown in Table 1.
+
+Theorem 1 (Informal Theorem 7, exponential time). Algorithm 2 is $( \varepsilon , \delta )$ -DP. When $\alpha$ fraction of the data is arbitrarily corrupted from $n$ samples from a $d$ -dimensional sub-Gaussian distribution with mean $\mu$ and an identity sub-Gaussian parameter, if $\begin{array} { r } { n = \widetilde \Omega ( d / \alpha ^ { 2 } + ( d + d ^ { 1 / 2 } \log ( 1 / \delta ) ) / ( \alpha \varepsilon ) ) } \end{array}$ then Algorithm 2 achieves $\| \hat { \mu } - \mu \| _ { 2 } = O ( \alpha \sqrt { \log ( 1 / \alpha ) } ) \ : w . h . p$ .
+
+We introduce PRIME (PRIvate and robust Mean Estimation) in $\ S 2 . 3$ with details in Algorithm 9 in Appendix E.1, to achieve computational efficiency. It requires a run-time of only $\widetilde { \cal O } ( d ^ { 3 } + n d ^ { 2 } )$ , but at the cost of requiring extra $d ^ { 1 / 2 }$ factor larger number of samples. This cannot be improved upon with current techniques since efficient robust estimators rely on the top PCA directions of the covariance matrix to detect outliers. [78] showed that $\widetilde \Omega ( d ^ { 3 / 2 } )$ samples are necessary to compute PCA directions while preserving $( \varepsilon , \delta )$ -DP when $\| x _ { i } \| _ { 2 } = O ( { \sqrt { d } } )$ . It remains an open question if this $\widetilde \Omega ( d ^ { 3 / 2 } / ( \alpha \varepsilon ) )$ bottleneck is fundamental; no matching lower bound is currently known.
+
+Theorem 2 (Informal Theorem 6, polynomial time). PRIME is $( \varepsilon , \delta )$ -DP and under the assumption of Thm.1, if $\begin{array} { r } { \dot { n } = \widetilde \Omega ( d / \alpha ^ { 2 } + ( d ^ { 3 / 2 } \log ( 1 / \delta ) ) / ( \alpha \varepsilon ) ) . } \end{array}$ , achieves $\begin{array} { r } { \| \hat { \mu } - \mu \| _ { 2 } = O ( \alpha \sqrt { \log ( 1 / \alpha ) } ) ~ w . h . p . } \end{array}$ .
+
+<table><tr><td rowspan=1 colspan=1></td><td rowspan=1 colspan=1>Upper bound (poly-time)</td><td rowspan=1 colspan=1>Upper bound (exp-time)</td><td rowspan=1 colspan=1>Lower bound</td></tr><tr><td rowspan=1 colspan=1>(ε,δ)-DP[52]</td><td rowspan=1 colspan=1>0+dlog1/2(1/8））Qε</td><td rowspan=1 colspan=1>+）</td><td rowspan=1 colspan=1>（+）</td></tr><tr><td rowspan=1 colspan=1>α-corruption [36]</td><td rowspan=1 colspan=1>0（</td><td rowspan=1 colspan=1>（</td><td rowspan=1 colspan=1>（</td></tr><tr><td rowspan=1 colspan=1>α-corruption and(ε,δ)-DP (this paper)</td><td rowspan=1 colspan=1>（+/210g(1/8）)αε[Theorem6]</td><td rowspan=1 colspan=1>0（+d+d1/210g(1/8))Q[Theorem7]</td><td rowspan=1 colspan=1>(）[52]</td></tr></table>
+
+Table 1: For estimating the mean $\mu \in \mathbb { R } ^ { d }$ of a sub-Gaussian distribution with a known covariance, we list the sufficient or necessary conditions on the sample sizes to achieve an error $\| \hat { \mu } - \mu \| _ { 2 } = \widetilde { O } ( \alpha )$ under $( \varepsilon , \delta )$ -DP, corruption of an $\alpha$ -fraction of samples, and both. √ $\clubsuit$ requires the distribution to be a Gaussian [14] and $\spadesuit$ requires $\delta \leq \sqrt { d } / n$ .
+
+Heavy-tailed distributions. When samples are drawn from a distribution with a bounded covariance, parameters of Algorithm 2 can be modified to nearly match the optimal sample complexity of (nonrobust) private mean estimation in Table 2. This algorithm also matches the fundamental limit on the accuracy of (non-private) robust estimation, which in this case is $\Omega ( \alpha ^ { 1 / 2 } )$ .
+
+Theorem 3 (Informal Theorem 8, exponential time). From a distribution with mean $\mu \in \mathbb { R } ^ { d }$ and covariance $\Sigma \preceq \mathbf { I }$ , $n$ samples are drawn and $\alpha$ -fraction is corrupted. Algorithm 2 is $( \varepsilon , \delta )$ -DP and $i f$ $n = \widetilde \Omega ( ( d + d ^ { 1 / 2 } \log ( 1 / \delta ) ) / ( \alpha \varepsilon ) + d ^ { 1 / 2 } \log ^ { 3 / 2 } ( 1 / \delta ) / \varepsilon )$ achieves $\| \hat { \mu } - \mu \| _ { 2 } = O ( \alpha ^ { 1 / 2 } ) \ : w . h . p .$ .
+
+The proposed PRIME-HT for covariance bounded distributions achieve computational efficiency at the cost of an extra factor of $d ^ { 1 / 2 }$ in sample size. This bottleneck is also due to DP PCA, and it remains open whether this gap can be closed by an efficient estimator.
+
+Theorem 4 (Informal Theorem 9, polynomial time). PRIME-HT is $( \varepsilon , \delta ) – D P$ and if $n \_ =$ $\widetilde \Omega ( ( d ^ { 3 / 2 } \log ( 1 / \delta ) ) / ( \alpha \varepsilon ) )$ achieves $\| \hat { \mu } - \mu \| _ { 2 } = O ( \alpha ^ { 1 / 2 } ) \ : w . h . p$ . under the assumptions of Thm. 3.
+
+<table><tr><td rowspan=1 colspan=1></td><td rowspan=1 colspan=1>Upper bound (poly-time)</td><td rowspan=1 colspan=1>Upper bound (exp-time)</td><td rowspan=1 colspan=1>Lower bound</td></tr><tr><td rowspan=1 colspan=1>(ε,δ)-DP [54]</td><td rowspan=1 colspan=1>O（d10g1/2（1/）Qε</td><td rowspan=1 colspan=1>O（d10g1/2（1/0））Qε</td><td rowspan=1 colspan=1>2（）</td></tr><tr><td rowspan=1 colspan=1>α-corruption [36]</td><td rowspan=1 colspan=1>）</td><td rowspan=1 colspan=1></td><td rowspan=1 colspan=1>(）</td></tr><tr><td rowspan=1 colspan=1>α-corruption and(ε,δ)-DP (this paper)</td><td rowspan=1 colspan=1>0（(/210g(1/8))αε[Theorem 9]</td><td rowspan=1 colspan=1>（+d1721g/2(18）αε[Theorem 8]</td><td rowspan=1 colspan=1>(([54])</td></tr></table>
+
+Table 2: For estimating the mean $\mu \in \mathbb { R } ^ { d }$ of a covariance bounded distribution, we list the sufficient or necessary conditions on the sample size to achieve an error $\| \hat { \mu } - \mu \| _ { 2 } = O ( \alpha ^ { 1 / 2 } )$ under $( \varepsilon , \delta )$ -DP, corruption of an $\alpha$ -fraction of samples, and both.
+
+# 1.1 Technical contributions
+
+We introduce PRIME which simultaneously achieves $( \varepsilon , \delta )$ -DP and robustness against $\alpha$ -fraction of corruption. A major challenge in making a standard filter-based robust estimation algorithm (e.g., [30]) private is the high sensitivity of the filtered set that we pass from one iteration to the next. We propose a new framework which makes private only the statistics of the set, hence significantly reducing the sensitivity. Our major innovation is a tight analysis of the end-to-end sensitivity of this multiple interactive accesses to the database. This is critical in achieving robustness while preserving privacy and is also of independent interest in making general iterative filtering algorithms private.
+
+The classical filter approach (see, e.g. [30]) needs to access the database √ $O ( d )$ times, which brings an extra $O ( { \sqrt { d } } )$ factor in the sample complexity due to DP composition. In order to reduce the iteration complexity, following the approach in [36], we propose filtering multiple directions simultaneously using a new score based on the matrix multiplicative weights (MMW). In order to privatize the MMW filter, our major innovation is a novel adaptive filtering algorithm DPTHRESHOLD(·) that outputs a single private threshold which guarantees sufficient progress at every iteration. This brings the number of database accesses from $O ( d )$ to $O ( ( \log d ) ^ { 2 } )$ .
+
+One downside of PRIME is that it requires an extra $d ^ { 1 / 2 }$ factor in the sample complexity, compared to known lower bounds for (non-robust) DP mean estimation. To investigate whether this is also necessary, we propose a sample optimal exponential time robust mean estimation algorithm in $\ S 4$ and prove that there is no extra statistical cost to jointly requiring privacy and robustness. Our major technical innovations is in using resilience property of the dataset to not only find robust mean (which is the typical use case of resilience) but also bound sensitivity of that robust mean.
+
+# 1.2 Preliminary on differential privacy (DP)
+
+DP is a formal metric for measuring privacy leakage when a dataset is accessed with a query [37].
+
+Definition 1.1. Given two datasets $S = \{ x _ { i } \} _ { i = 1 } ^ { n }$ and $S ^ { \prime } = \{ x _ { i } ^ { \prime } \} _ { i = 1 } ^ { n ^ { \prime } }$ , we say $S$ and $S ^ { \prime }$ are neighboring $i f d _ { \triangle } ( S , S ^ { \prime } ) \leq 1$ where $d _ { \triangle } ( S , S ^ { \prime } ) \triangleq \operatorname* { m a x } \{ | S \setminus S ^ { \prime } | , | S ^ { \prime } \setminus S | \}$ , which is denoted by $S \sim S ^ { \prime }$ . For an output of a stochastic query q on a database, we say q satisfies $( \varepsilon , \delta )$ -differential privacy for some $\varepsilon > 0$ and $\delta \in ( 0 , 1 )$ i $\begin{array} { r } { \textstyle f \mathbb { P } ( q ( S ) \in A ) \le e ^ { \varepsilon } \mathbb { P } ( q ( S ^ { \prime } ) \in A ) + \delta } \end{array}$ for all $S \sim S ^ { \prime }$ and all subset $A$ .
+
+Let $z \sim \mathrm { L a p } ( b )$ be a random vector with entries i.i.d. sampled from Laplace distribution with pdf $( 1 / 2 b ) e ^ { - | z | / b }$ . Let $z \sim \mathcal { N } ( \mu , \Sigma )$ denote a Gaussian random vector with mean $\mu$ and covariance $\Sigma$ .
+
+Definition 1.2. The sensitivity of a query $f ( S ) ~ \in ~ \mathbb { R } ^ { k }$ is defined as $\Delta _ { p } ~ = ~ \operatorname* { s u p } _ { S \sim S ^ { \prime } } \| f ( S ) ~ -$ $f ( S ^ { \prime } ) \| _ { p }$ for a norm $\begin{array} { r } { \| { \boldsymbol x } \| _ { p } ~ = ~ ( \sum _ { i \in [ k ] } | x _ { i } | ^ { p } ) ^ { 1 / p } } \end{array}$ . For $p ~ = ~ 1$ , the Laplace mechanism outputs $f ( S ) + \mathrm { L a p } ( { \Delta _ { 1 } } / { \varepsilon } )$ and achieves $( \varepsilon , 0 ) – D P \ : I 3 7 J$ . For $p = 2$ , the Gaussian mechanism outputs $f ( S ) + \mathcal { N } ( 0 , ( \Delta _ { 2 } ( \sqrt { 2 \log ( 1 . 2 5 / \delta ) } ) / \varepsilon ) ^ { 2 } \mathbf { I } )$ and achieves $( \varepsilon , \delta ) – D P / 3 8 J$ .
+
+We use these output perturbation mechanisms along with the exponential mechanism [69] as building blocks. Appendix A provides detailed survey of privacy and robust estimation.
+
+# 1.3 Problem formulation
+
+We are given $n$ samples from a sub-Gaussian distribution with a known covariance but unknown mean, and $\alpha$ fraction of the samples are corrupted by an adversary. Our goal is to estimate the unknown mean. We follow the standard definition of adversary in [30], which can adaptively choose which samples to corrupt and arbitrarily replace them with any points.
+
+Assumption 1. An uncorrupted dataset $S _ { \mathrm { g o o d } }$ consists of $n$ i.i.d. samples from a $d$ -dimensional sub-Gaussian distribution with mean $\mu \in \mathbb { R } ^ { d }$ and covariance $\mathbb { E } [ x x ^ { \top } ] = \mathbf { I } _ { d } ,$ , which is 1-sub-Gaussian, i.e., $\mathbb { E } [ \exp ( v ^ { \top } x ) ] \leq \exp ( \| v \| _ { 2 } ^ { 2 } / 2 )$ for all $v \in \mathbb { R } ^ { d }$ . For some $\alpha \in ( 0 , 1 / 2 )$ , we are given a corrupted dataset $S = \{ x _ { i } \in \mathbb { R } ^ { d } \} _ { i = 1 } ^ { n }$ where an adversary adaptively inspects all the samples in $S _ { \mathrm { g o o d } }$ , removes αn of them, and replaces them with $S _ { \mathrm { b a d } }$ which are αn arbitrary points in $\mathbb { R } ^ { d }$ .
+
+Similarly, we consider the same problem for heavy-tailed distributions with a bounded covariance.   
+We present the assumption and main results for covariance bounded distributions in Appendix B.
+
+Notations. Let $[ n ] = \{ 1 , 2 , \dots , n \}$ . For $x \in \mathbb { R } ^ { d }$ , we use $\textstyle \| x \| _ { 2 } = ( \sum _ { i \in [ d ] } ( x _ { i } ) ^ { 2 } ) ^ { 1 / 2 }$ to denote the Euclidean norm. For $X \in \mathbb { R } ^ { d \times d }$ , we use $\begin{array} { r } { \| X \| _ { 2 } = \operatorname* { m a x } _ { \| v \| _ { 2 } = 1 } \| X v \| _ { 2 } } \end{array}$ to denote the spectral norm. The $d \times d$ identity matrix is $\mathbf { I } _ { d \times d }$ . Whenever it is clear from context, we use $S$ to denote both a set of data points and also the set of indices of those data points. $\widetilde O$ and $\widetilde \Omega$ hide poly-logarithmic factors in $d , n , 1 / \alpha$ , and the failure probability.
+
+Outline. We present PRIME for sub-Gaussian distribution in $\ S 2$ , and present theoretical analysis in $\ S 3$ . We then introduce an exponential time algorithm with near optimal guarantee in $\ S 4$ . Due to space constraints, analogous results for heavy-tailed distributions are presented in Appendix B.
+
+# 2 PRIME: efficient algorithm for robust and DP mean estimation
+
+In order to describe the proposed algorithm PRIME, we need to first describe a standard (non-private) iterative filtering algorithm for robust mean estimation.
+
+# 2.1 Background on (non-private) iterative filtering for robust mean estimation
+
+Non-private robust mean estimation approaches recursively apply the following filter, whose framework is first proposed in [28]. Given a dataset $S = \{ x _ { i } \} _ { i = 1 } ^ { n }$ , the current set $S _ { 0 } \subseteq [ n ]$ of data points is updated starting with $S _ { 1 } = [ n ]$ . At each step, the following filter (Algorithm 1 in [63]) attempts to detect the corrupted data points and remove them.
+
+1. Compute the top eigenvector $v _ { t } \gets \arg \operatorname* { m a x } _ { v : \| v \| _ { 2 } = 1 } v ^ { \top } \mathrm { C o v } ( S _ { t - 1 } ) v$ of the covariance of the current data set $\{ x _ { i } \} _ { i \in S _ { t - 1 } }$ ;   
+2. Compute scores for all data points $j \in S _ { t - 1 } \colon \tau _ { j } \gets \left( v _ { t } ^ { \top } \left( x _ { j } - \mathrm { M e a n } ( S _ { t - 1 } ) \right) \right) ^ { 2 } \ ;$   
+3. Draw a random threshold: $Z _ { t } \gets \mathrm { U n i f } ( [ 0 , 1 ] )$ ;   
+4. Remove outliers from $S _ { t - 1 }$ defined as $\{ i \in S _ { t - 1 } : \tau _ { i }$ is in the largest $2 \alpha$ -tail of $\{ \tau _ { j } \} _ { j \in S _ { t - 1 } }$ and $\tau _ { i } \geq Z _ { t } \tau _ { \operatorname* { m a x } } \}$ , where $\tau _ { \mathrm { m a x } } = \operatorname* { m a x } _ { j \in S _ { t - 1 } } \tau _ { j }$
+
+This is repeated until the empirical covariance is sufficiently small and the empirical mean $\hat { \mu }$ is output. At a high level, the correctness of this algorithm relies on the key observation that the $\alpha$ -fraction of adversarial corruption can not significantly change the mean of the dataset without introducing large eigenvalues in the empirical covariance. Therefore, the algorithm finds top eigenvector of the empirical covariance in step 1, and tries to correct the empirical covariance by removing corrupted data points. Each data point is assigned a score in step 2 which indicates the “badness” of the data points, and a threshold $Z _ { t }$ in step 3 is carefully designed such that step 4 guarantees to remove more corrupted data points than good data points (in expectation). This guarantees the following bound achieving the near-optimal sample complexity shown in the second row of Table 1. A formal description of this algorithm is in Algorithm 4 in Appendix C.
+
+Proposition 2.1 (Corollary of [63, Theorem 2.1]). Under assumption $^ { l }$ , the above filtering algorithm achieves accuracy $\| \hat { \mu } - \mu \| _ { 2 } \le O ( \alpha \sqrt { \log ( 1 / \alpha ) } ) \ w . p . \ 0 . 9 i f n \ge \widetilde { \Omega } ( d / \alpha ^ { 2 } ) \ / .$ .
+
+Challenges in making robust mean estimation private. To get a DP and robust mean, a naive attempt is to apply a standard output perturbation mechanism to $\hat { \mu }$ . However, this is obviously challenging since the end-to-end sensitivity is intractable. The standard recipe to circumvent this is to make the current “state” $S _ { t }$ private at every iteration. Once $S _ { t - 1 }$ is private (hence, public knowledge), making the next “state” $S _ { t }$ private is simpler. We only need to analyze the sensitivity of a single step and apply some output perturbation mechanism with $\left( \varepsilon _ { t } , \delta _ { t } \right)$ . End-to-end privacy is guaranteed by accounting for all these $\left( \varepsilon _ { t } , \delta _ { t } \right)$ ’s using the advanced composition [51]. This recipe has been quite successful, for example, in training neural networks with (stochastic) gradient descent [1], where the current state can be the optimization variable $\mathbf { x } _ { t }$ . However, for the above (non-private) filtering algorithm, this standard recipe fails, since the state $S _ { t }$ is a set and has large sensitivity. Changing a single data point in $S _ { t }$ can significantly alter which (and how many) samples are filtered out.
+
+# 2.2 A new framework for private iterative filtering
+
+Instead of making the (highly sensitive) $S _ { t }$ itself private, we propose a new framework which makes private only the statistics of $S _ { t }$ : the mean $\mu _ { t }$ and the top principal direction $v _ { t }$ . There are two versions of this algorithm, which output the exactly same $\hat { \mu }$ with the exactly same privacy guarantees, but are written from two different perspectives. We present here the interactive version from the perspective of an analyst accessing the dataset via DP queries $\cdot q _ { \mathrm { r a n g e } }$ , $q _ { \mathrm { s i z e } }$ , $q _ { \mathrm { m e a n } }$ , $q _ { \mathrm { n o r m } }$ and $q _ { \mathrm { P C A } } ,$ ), because this version makes clear the inner operations of each private mechanisms, hence making $( i )$ the sensitivity analysis transparent, $( i i )$ checking the correctness of privacy guarantees easy, and $( i i i )$ tracking privacy accountant simple. In practice, one should implement the centralized version (Algorithm 7 in Appendix D), which is significantly more efficient.
+
+# Algorithm 1: Private iterative filtering (interactive version)
+
+Input: $S = \{ x _ { i } \} _ { i \in [ n ] }$ , $\alpha \in ( 0 , 1 / 2 )$ , probability $\zeta \in ( 0 , 1 )$ , # of iterations $T = \Theta ( d ) , ( \varepsilon , \delta )$   
+1 $( \bar { x } , B ) \gets q _ { \mathrm { r a n g e } } ( S , 0 . 0 1 \varepsilon , 0 . 0 1 \delta )$   
+2 $\varepsilon _ { 1 } \gets \operatorname* { m i n } \{ 0 . 9 9 \varepsilon , 0 . 9 \} / ( 4 \sqrt { 2 T \log ( 2 / \delta ) } )$ ), δ1 ← 0.99δ/(8T )   
+3 if $n < ( 4 / \varepsilon _ { 1 } ) \log ( 1 / ( 2 \delta _ { 1 } ) )$ then Output: $\varnothing$   
+4 for $t = 1 , \dots , T$ do   
+5 $n _ { t }  q _ { \mathrm { s i z e } } ( \{ ( \mu _ { \ell } , v _ { \ell } , Z _ { \ell } ) \} _ { \ell \in [ t - 1 ] } , \varepsilon _ { 1 } , \bar { x } , B )$ , if $n _ { t } < 3 n / 4$ then Output: $\varnothing$   
+6 $\mu _ { t }  q _ { \mathrm { m e a n } } ( \{ ( \mu _ { \ell } , v _ { \ell } , Z _ { \ell } ) \} _ { \ell \in [ t - 1 ] } , \varepsilon _ { 1 } , \bar { x } , B )$   
+7 $\lambda _ { t } \gets q _ { \mathrm { n o r m } } ( \{ ( \mu _ { \ell } , v _ { \ell } , Z _ { \ell } ) \} _ { \ell \in [ t - 1 ] } , \mu _ { t } , \varepsilon _ { 1 } , \bar { x } , B )$   
+8 if $\lambda _ { t } \le ( C - 0 . 0 1 ) \alpha \log { 1 / \alpha }$ then Output: $\mu _ { t }$   
+9 $\boldsymbol { v } _ { t }  q _ { \mathrm { P C A } } ( \{ ( \mu _ { \ell } , v _ { \ell } , Z _ { \ell } ) \} _ { \ell \in [ t - 1 ] } , \mu _ { t } , \varepsilon _ { 1 } , \delta _ { 1 } , \bar { x } , B ) )$   
+10 $Z _ { t } \gets \mathrm { U n i f } ( [ 0 , 1 ] )$
+
+# Output: $\mu _ { t }$
+
+We give a high-level explanation of each step of Algorithm 1 here and give the formal definitions of all the queries in Appendix D. First, $q _ { \mathrm { r a n g e } }$ returns (the parameters of) a hypercube $\bar { x } + [ - B / 2 , B / 2 ] ^ { d }$ that is guaranteed to include all uncorrupted samples while preserving privacy. This is achieved by running $d$ coordinate-wise private histograms and selecting $\bar { x } _ { j }$ as the center of the largest bin for the $j$ -th coordinate. Since covariance is I, $q _ { \mathrm { r a n g e } }$ returns a fixed $B = 8 \sigma \sqrt { \log ( d n / \zeta ) }$ . Such an adaptive estimate of the support is critical in tightly bounding the sensitivity of all subsequent queries, which operate on the clipped dataset; all data points are projected as $\mathcal { P } _ { \bar { x } + [ - B / 2 , B / 2 ] ^ { d } } ( x ) =$ arg $\begin{array} { r } { \operatorname* { m i n } _ { y \in \bar { x } + [ - B / 2 , B / 2 ] ^ { c } } } \end{array}$ $\lVert y - x \rVert _ { 2 }$ in all the queries that follow. With clipping, a single data point can now change at most by $B \sqrt { d }$ .
+
+The subsequent steps perform the non-private filtering algorithm of $\ S 2 . 1$ , but with private statistics $\mu _ { t }$ and $v _ { t }$ . As the set $S _ { t }$ changes over time, we lower bound its size (which we choose to be $| S _ { t } | > n / 2 )$ ) to upper bound the sensitivity of other queries $q _ { \mathrm { m e a n } } , q _ { \mathrm { n o r m } }$ and $q _ { \mathrm { P C A } }$ .
+
+At the $t$ -th iterations, every time a query is called the data curator $( i )$ uses $( { \bar { x } } , B )$ to clip the data, $( i i )$ computes $S _ { t }$ by running $t - 1$ steps of the non-private filtering algorithm of $\ S 2 . 1$ but with a given fixed set of parameters $\{ ( \mu _ { \ell } , v _ { \ell } ) \} _ { \ell \in [ t - 1 ] }$ (and the given randomness $\{ Z _ { \ell } \} _ { \ell \in [ t - 1 ] } )$ , and $( i i i )$ computes the queried private statistics of $S _ { t }$ . If the private spectral norm of the covariance of $S _ { t }$ (i.e., $\lambda _ { t } ,$ ) is sufficiently small, we output the private and robust mean $\hat { \mu } = \mu _ { t }$ (line 8). Otherwise, we compute the private top PCA direction $v _ { t }$ and draw an randomness $Z _ { t }$ to be used in the next step of filtering, as in the non-private filtering algorithm. We emphasize that $\{ S _ { \ell } \}$ are not private, and hence never returned to the analyst. We also note that this interactive version is redundant as every query is re-computing $S _ { t }$ . In our setting, the analyst has the dataset and there is no need to separate them. This leads to a centralized version we provide in Algorithm 7 in the appendix, which avoids redundant computations and hence is significantly more efficient.
+
+The main challenge in this framework is the privacy analysis. Because $\{ S _ { \ell } \} _ { \ell \in [ t - 1 ] }$ is not private, each query runs $t - 1$ steps of filtering whose end-to-end sensitivity could blow-up. Algorithmically, $( i )$ we start with a specific choice of a non-private iterative filtering algorithm (among several variations that are equivalent in non-private setting but widely differ in its sensitivity), and $( i i )$ make appropriate changes in the private queries (Algorithm 1) to keep the sensitivity small. Analytically, the following key technical lemma allows a sharp analysis of the end-to-end sensitivity of iterative filtering.
+
+Lemma 2.2. Let $S _ { t } ( S )$ denote the resulting subset of samples after $t$ iterations of the filtering in the queries $( q _ { \mathrm { s i z e } }$ , $q _ { \mathrm { m e a n } }$ , $q _ { \mathrm { n o r m } }$ , and $q _ { \mathrm { P C A } } )$ are applied to a dataset $s$ using fixed parameters $\{ ( \mu _ { \ell } , v _ { \ell } , Z _ { \ell } ) \} _ { \ell = 1 } ^ { t }$ . Then, we have $d _ { \triangle } ( S _ { t } ( S ) , S _ { t } ( S ^ { \prime } ) ) \leq d _ { \triangle } ( S , S ^ { \prime } )$ , where $d _ { \triangle } ( S , S ^ { \prime } ) \triangleq \operatorname* { m a x } \{ | S \ \backslash$ $S ^ { \prime } | , | S ^ { \prime } \backslash S | \}$ .
+
+Recall that two datasets are neighboring, i.e., $\boldsymbol { S } \sim \boldsymbol { S } ^ { \prime }$ , iff $d _ { \triangle } ( S , S ^ { \prime } ) \leq 1$ . This lemma implies that if two datasets are neighboring, then they are still neighboring after filtering with the same parameters, no matter how many times we filter them. Hence, this lemma allows us to use the standard outputperturbation mechanisms with $( \varepsilon _ { 1 } , \delta _ { 1 } )$ -DP. Advanced composition ensures that end-to-end guarantee of $4 T$ such queries is $( 0 . 9 9 \varepsilon , 0 . 9 9 \delta )$ -DP. Together with $( 0 . 0 1 \varepsilon , 0 . 0 1 \delta )$ -DP budget used in $q _ { \mathrm { r a n g e } }$ , this satisfied the target privacy. Analyzing the utility of this algorithm, we get the following guarantee.
+
+Theorem 5. Algorithm $^ { l }$ is $( \varepsilon , \delta )$ -DP. Under Assumption $^ { l }$ , there exists a universal constant $c \in$ $( 0 , 0 . 1 )$ such that if $\alpha \leq c$ and $n = \widetilde \Omega \left( ( d / \alpha ^ { 2 } ) + d ^ { 2 } ( \log ( 1 / \delta ) ) ^ { 3 / 2 } / ( \varepsilon \alpha ) \right)$ then Algorithm 1 achieves $\| \hat { \mu } - \mu \| _ { 2 } \leq O ( \alpha \sqrt { \log ( 1 / \alpha ) } )$ with probability 0.9.
+
+The first term $O ( d / \alpha ^ { 2 } )$ in the sample complexity is optimal (cf. Table 1), but there is a factor of $d$ gap in the second term. This is due to the fact that we need to run $O ( d )$ iterations in the worst-case. Such numerous accesses to the database result in large noise to be added at each iteration, requiring large sample size to combat that extra noise. We introduce PRIME to reduce the number of iterations to ${ \bar { O } } ( ( \log { d } ) ^ { 2 } )$ and significantly reduce the sample complexity.
+
+# 2.3 PRIME: novel robust and private mean estimator
+
+Algorithm 1 (specifically Filter(·) in Algorithm 1) accesses the database $O ( d )$ times. This is necessary for two reasons. First, the filter checks only one direction $v _ { t }$ at each iteration. In the worst case, the corrupted samples can be scattered in $\Omega ( d )$ orthogonal directions such that the filter needs to be repeated $O ( d )$ times. Secondly, even if the corrupted samples are clustered together in one direction, the filter still needs to be repeated $O ( d )$ times. This is because we had to use a large (random) threshold of $d B ^ { 2 } Z _ { t } = { \cal { O } } ( d )$ to make the threshold data-independent so that we can keep the sensitivity of Filter(·) low, which results in slow progress. We propose filtering multiple directions simultaneously using a new score $\{ \tau _ { i } \}$ based on the matrix multiplicative weights. Central to this approach is a novel adaptive filtering algorithm DPTHRESHOLD $( \cdot )$ that guarantees sufficient decrease in the total score at every iteration.
+
+# 2.3.1 Matrix Multiplicative Weight (MMW) scoring
+
+The MMW-based approach, pioneered in [36] for non-private robust mean estimation, filters out multiple directions simultaneously. It runs over $O ( \log d )$ epochs and every epoch consists of $O ( \log d )$ iterations. At every epoch $s$ and iteration $t$ , step 2 of the iterative filtering in $\ S 2 . 1$ is replaced by a new score $\tau _ { i } = ( x _ { i } - \mathrm { M e a n } ( S _ { t } ^ { ( s ) } ) ) ^ { T } U _ { t } ^ { ( s ) } ( x _ { i } - \mathrm { M e a n } ( S _ { t } ^ { ( s ) } ) )$ where $U _ { t } ^ { ( s ) }$ now accounts for all directions in $\mathbb { R } ^ { d }$ but appropriately weighted. Precisely, it is defined via the matrix multiplicative update:
+
+$$
+U _ { t } ^ { ( s ) } = \frac { \exp \left( \alpha ^ { ( s ) } \sum _ { r \in [ t ] } ( \mathrm { C o v } ( S _ { r } ^ { ( s ) } ) - { \bf I } ) \right) } { \mathrm { T r } \big ( \exp ( \alpha ^ { ( s ) } \sum _ { r \in [ t ] } ( \mathrm { C o v } ( S _ { r } ^ { ( s ) } ) - { \bf I } ) ) \big ) } ,
+$$
+
+for some choice of $\alpha ^ { ( s ) } > 0$ . If we set the number of iterations to one, a choice of $\alpha ^ { ( s ) } = \infty$ recovers the previous score that relied on the top singular vector from $\ S 2 . 1$ and a choice of $\alpha ^ { ( s ) } = 0$ gives a simple norm based score $\tau _ { i } = \| x _ { i } \| _ { 2 } ^ { 2 }$ . An appropriate choice of $\alpha ^ { ( s ) }$ smoothly interpolates between these two extremes, which ensures that $O ( \log d )$ iterations are sufficient for the spectral norm of the covariance to decrease strictly by a constant factor. This guarantees that after $O ( \log d )$ epochs, we sufficiently decrease the covariance to ensure that the empirical mean is accurate enough. Critical in achieving this gain is our carefully designed filtering algorithm DPTHRESHOLD that uses the privately computed MMW-based scores using Gaussian mechanism on the covariance matrices as shown in Algorithm 11 in Appendix E.
+
+# 2.3.2 Adaptive filtering with DPTHRESHOLD
+
+Novelty. The corresponding non-private filtering of [36, Algorithm 9] for robust mean estimation takes advantage of an adaptive threshold, but filters out each sample independently resulting in a prohibitively large sensitivity; the coupling between each sample and the randomness used to filter it can change widely between two neighboring datasets. On the other hand, Algorithm 1 (i.e., Filter(·) in Algorithm 6) takes advantage of jointly filtering all points above a single threshold $B ^ { 2 } d Z _ { t }$ with a single randomness $Z _ { t } \sim \mathrm { U n i f } [ 0 , 1 ]$ , but the non-adaptive (and hence large) choice of the range $B ^ { 2 } d$ results in a large number of iterations because each filtering only decrease the score by little. To sufficiently reduce the total score while maintaining a small sensitivity, we introduce a filter with a single and adaptive threshold.
+
+Algorithm. Our goal here is to privately find a single scalar $\rho$ such that when a randomized filter is applied on the scores $\{ \tau _ { i } \}$ with a (random) threshold $\rho Z$ (with $Z$ drawn uniform in $[ 0 , 1 ] )$ , we filter out enough samples to make progress in each iteration while ensuring that we do not remove too many uncorrupted samples. This is a slight generalization of the non-private algorithm in Section 2.1, which simply set $\rho = \operatorname* { m a x } _ { j \in S _ { t } } \tau _ { j }$ . While this guarantees the filter removes more corrupted samples than good samples, it does not make sufficient progress in reducing the total score of the samples.
+
+Ideally, we want the thresholding to decrease the total score by a constant multiplicative factor, which will in the end allow the algorithm to terminate within logarithmic iterations. To this end, we propose a new scheme of using the largest $\rho$ such that the following inequality holds:
+
+$$
+\sum _ { \tau _ { i } > \rho } ( \tau _ { i } - \rho ) \geq 0 . 3 1 \sum _ { \tau _ { i } \in S _ { t } } ( \tau _ { i } - 1 ) .
+$$
+
+We use a private histogram of the scores to approximate this threshold. Similar to [55, 58], we use geometrically increasing bin sizes such that we use only $O ( \log B ^ { 2 } d )$ bins while achieving a preferred multiplicative error in our quantization. At each epoch $s$ and iteration $t$ , we run DPTHRESHOLD sketched in the following to approximate $\rho$ followed by a random filter. Step 3 replaces the non-private condition in Eq. (1). A complete description is provided in Algorithm 11.
+
+1. Privately compute scores for all data points $i \in S _ { t } ^ { ( s ) } : \tau _ { i } \gets ( x _ { i } - \mu _ { t } ) ^ { \top } U _ { t } ^ { ( s ) } ( x _ { i } - \mu _ { t } ) ;$ ;   
+2. Compute a private histogram $\{ \tilde { h } _ { j } \} _ { j = 1 } ^ { 2 + \log ( B ^ { 2 } d ) }$ t i  i  t  t i  t of the scores over geometrically sized bins   
+$I _ { 1 } = [ 1 / 4 , 1 / 2 )$ , $I _ { 2 } = [ 1 / 2 , 1 ) , \ldots , \bar { I } _ { 2 + \log ( B ^ { 2 } d ) } = [ 2 ^ { \log ( B ^ { 2 } d ) - 1 } , 2 ^ { \log ( B ^ { 2 } d ) } ]$ ;   
+3. Privately find the largest $\ell$ satisfying $\begin{array} { r } { \sum _ { j \ge \ell } ( 2 ^ { j } - 2 ^ { \ell } ) \tilde { h } _ { j } \ge 0 . 3 1 \sum _ { i \in S _ { t } ^ { ( s ) } } ( \tau _ { i } - 1 ) ; } \end{array}$ ;   
+4. Output $\rho = 2 ^ { \ell }$ .
+
+# 3 Analyses of PRIME
+
+Building on the framework of Algorithm 1, PRIME (Algorithm 9) replaces the score with the MMWbased score presented in $\ S 2 . 3 . 1$ and the filter with the adaptive DPTHRESHOLD. This reduces the number of iterations to $T = O ( ( \log d ) ^ { 2 } )$ achieving the following bound.
+
+Theorem 6. PRIME is $( \varepsilon , \delta )$ -differentially private. Under Assumption $I$ there exists a universal constant $c \in ( 0 , 0 . 1 )$ such that if $\alpha \leq c$ and $n = \widetilde \Omega ( ( d / \alpha ^ { 2 } ) + ( d ^ { 3 / 2 } / ( \varepsilon \alpha ) ) \log ( 1 / \delta ) )$ , then PRIME achieves $\| \hat { \mu } - \mu \| _ { 2 } = O ( \alpha \sqrt { \log ( 1 / \alpha ) } )$ with probability 0.9.
+
+A proof is provided in Appendix F. The notation $\widetilde { \Omega } ( \cdot )$ hides logarithmic terms in $d , R$ , and $1 / \alpha$ . To achieve an error of $O ( \alpha \sqrt { \log ( 1 / \alpha ) } )$ , the first term $\widetilde \Omega ( d / \alpha ^ { 2 } \log ( 1 / \alpha ) )$ is necessary even if there is no corruption. The accuracy of $O ( \alpha \sqrt { \log ( 1 / \alpha ) } )$ matches the lower bound shown in [33] for any polynomial time statistical query algorithm, and it nearly matches the information theoretical lower bound on robust estimation of $\Omega ( \alpha )$ . On the other hand, the second term of $\widetilde \Omega ( d ^ { 3 / 2 } / ( \varepsilon \alpha \log ( 1 / \alpha ) ) )$
+
+has an extra factor of $d ^ { 1 / 2 }$ compared to the optimal one achieved by exponential time Algorithm 2. It is an open question if this gap can be closed by a polynomial time algorithm.
+
+The bottleneck is the private matrix multiplicative weights. Such spectral analyses are crucial in filter-based robust estimators. Even for a special case of privately computing the top principal component, the best polynomial time algorithm requires $\bar { O } ( d ^ { 3 / 2 } )$ samples [39, 18, 78], and this sample complexity is also necessary as shown in [39, Corollary 25].
+
+To boost the success probability to $1 - \zeta$ for some small $\zeta > 0$ , we need an extra $\log ( 1 / \zeta )$ factor in the sample complexity to make sure the dataset satisfies the regularity condition with probability $\zeta / 2$ . Then we can run PRIME $\log ( 1 / \zeta )$ times and choose the output of a run that satisfies $n ^ { ( s ) } > n ( 1 { - } 1 0 \alpha )$ and $\lambda ^ { ( s ) } \le C \alpha \log ( 1 / \alpha )$ at termination.
+
+![](images/46a53c55cd9bcc36cd60657375546149f2ff820d7058a98083ffc81ea44ac394.jpg)  
+Figure 1: Private mean estimators (e.g., DP mean [52]) are vulnerable to adversarial corruption especially in high dimensions, while the proposed PRIME achieves robustness (and privacy) regardless of the dimension of the samples.
+
+Numerical experiments support our theoretical claims. The left figure with $( \alpha , \varepsilon , \delta , n ) \ =$ $( 0 . 0 5 , 2 0 , 0 . 0 1 , 1 0 ^ { 6 } )$ is in the large $\alpha$ regime where the DP Mean error is dominates by $\alpha \sqrt { d }$ and PRIME error by $\alpha \sqrt { \log ( 1 / \alpha ) }$ . Hence, PRIME error is constant whereas DP Mean error increases with the dimension $d$ . The second figure with √ $( \alpha , \varepsilon , \delta , n ) = ( 0 . 0 0 1 , 2 0 , 0 . 0 1 , 1 0 ^ { 6 } )$ is in the small $\alpha$ regime when DP Mean error consists of $\alpha \sqrt { d } + \sqrt { d / n }$ and PRIME is dominated by $\sqrt { d / n }$ . Both increase with the dimension $d$ , and the gap can be made large by increasing $\alpha$ . The right figure√ with $( \alpha , \delta , d , n ) = ( 0 . 1 , 0 . 0 1 , 1 0 , 1 0 ^ { 6 } )$ is when DP Mean error is dominated by $\alpha \sqrt { d }$ and PRIME by $\alpha \sqrt { \log ( 1 / \alpha ) }$ when $\varepsilon > c d ^ { 1 . 5 } / ( \alpha n )$ . Below this threshold, which happens in this example around $\varepsilon = 0 . 0 5$ , the added noise in the private mechanism starts to dominate with decreasing $\varepsilon$ . Both algorithms have respective thresholds below which the error increases with decreasing $\varepsilon$ . This threshold is larger for PRIME because it uses the privacy budget to perform multiple operations and hence the noise added to the final output is larger compared to DP Mean. Below this threshold, which can be easily determined based on the known parameters $( \varepsilon , \delta , n , \alpha )$ , we should either collect more data (which will decrease the threshold) or give up filtering and spend all privacy budget on $q _ { \mathrm { r a n g e } }$ and the empirical mean (which will reduce the error). Details of the experiments are in Appendix L.
+
+# 4 Exponential time algorithm with near-optimal sample complexity
+
+Novelty. An existing exponential time algorithm for robust and private mean estimation in [14] strictly requires the uncorrupted samples to be drawn from a Gaussian distribution. We also provide a similar algorithm based on private Tukey median in Appendix I and its analysis in Appendix J. In this section, we introduce a novel estimator that achieves near-optimal guarantees for more general sub-Gaussian distributions (and also covariance bounded distributions) but takes an exponential run-time. Its innovation is in leveraging on the resilience property of well-behaved distributions not only to estimate the mean robustly (which is the standard use of the property) but also to adaptively bound the sensitivity of the estimator, thus achieving optimal privacy-accuracy tradeoff.
+
+Definition 4.1 (Resilience from Definition 1 in [73]). $A$ set of points $\{ x _ { i } \} _ { i \in S }$ lying in $\mathbb { R } ^ { d }$ is $( \sigma , \alpha )$ - resilient around a point $\mu$ $\begin{array} { r } { . i f \| ( 1 / | T | ) \sum _ { i \in T } ( x _ { i } - \mu ) \| _ { 2 } \leq \sigma } \end{array}$ for all subsets $T \subset S$ of size $( 1 - \alpha ) | S |$ .
+
+Algorithm. As data is corrupted, we define $R ( S )$ as a surrogate for resilience of the uncorrupted part of the set. If $S$ indeed consists of a $1 - \alpha$ fraction of independent samples from the promised class of distributions, the goodness score $R ( S )$ will be close to the resilience property of the good data.
+
+Definition 4.2 (Goodness of a set). For $\mu ( S ) = ( 1 / | S | ) \sum _ { i \in S } x _ { i }$ , let us define
+
+$$
+R ( S ) \stackrel { \Delta } { = } \operatorname* { m i n } _ { \substack { S ^ { \prime } \subset S , | S ^ { \prime } | = ( 1 - 2 \alpha ) | S | . } } \operatorname* { m a x } _ { \substack { T \subset S ^ { \prime } , | T | = ( 1 - \alpha ) | S ^ { \prime } | . } } \| \mu ( T ) - \mu ( S ^ { \prime } ) \| _ { 2 } .
+$$
+
+Algorithm 2 first checks if the resilience matches that of the promised distribution. The data is pre-processed with $q _ { \mathrm { r a n g e } }$ to ensure we can check $R ( S )$ privately. Once resilience is cleared, we can safely use the exponential mechanism based on the score function $d ( \hat { \mu } , S )$ in Definition 4.3 to select an approximate robust mean $\hat { \mu }$ privately. The choice of the sensitivity critically relies on the fact that resilient datasets have small sensitivity of ${ \cal O } ( ( 1 / n ) \sqrt { \log ( 1 / \alpha ) } )$ . Without the resilience check, the sensitivity is $O ( d ^ { 1 / 2 } / n )$ resulting in an extra factor of $\sqrt { d }$ in the sample complexity.
+
+Input: $S = \{ x _ { i } \} _ { i \in [ n ] } , \alpha \in ( 0 , 1 / 2 ) , ( \varepsilon , \delta )$   
+1 if $n < c d ^ { 1 / 2 } \log ( 1 / \delta ) / \left( \varepsilon \alpha \sqrt { \log ( 1 / \alpha ) } \right)$ then Output: $\emptyset \ [ \ c d ^ { 1 / 2 } \log ( 1 / \delta ) / \ ( \varepsilon \alpha )$ for hevay-tail]   
+2 $( \bar { x } , B ) \gets q _ { \mathrm { r a n g e } } ( S , ( 1 / 3 ) \varepsilon , ( 1 / 3 ) \delta )$ $[ \ q _ { \mathrm { r a n g e - h t } } ( \cdot )$ for hevay-tail]   
+3 Project the data points onto the ball: $x _ { i }  \mathcal { P } _ { B _ { \sqrt { d } B / 2 } ( \bar { x } ) } ( x _ { i } )$ , for all $i \in [ n ]$   
+4 $\widehat { R } ( S ) \gets R ( S ) + \mathrm { L a p } ( 3 B d ^ { 1 / 2 } / ( n \varepsilon ) )$   
+5 if $\widehat { R } ( S ) > 2 \alpha \sqrt { \log ( 1 / \alpha ) }$ then Output: $\varnothing$ $[ \widehat { R } ( S ) > 2 c _ { \zeta } \sqrt { \alpha }$ for hevay-tail]   
+6 else Output: a randomly drawn point $\hat { \mu } \in \mathcal { B } _ { \sqrt { d } B / 2 } ( \bar { x } )$ sampled from a density   
+7 r(ˆµ) ∝ e−(1/(24 log(1/α)))ε n d(ˆµ,S) $[ e ^ { - ( \varepsilon n \sqrt { \alpha } / ( 2 4 c _ { \zeta } ) ) d ( \hat { \mu } , S ) }$ for heavy-tail]
+
+We propose the score function $d ( \hat { \mu } , S )$ in the following definition, which is a robust estimator of the distance between the mean and the candidate $\hat { \mu }$ .
+
+Definition 4.3. For a set of data $\{ x _ { i } \} _ { i \in S }$ lying in $\mathbb { R } ^ { d }$ , for any $v \in \mathbb { S } ^ { d - 1 }$ , define $\mathcal { T } ^ { v }$ to be the $3 \alpha | S |$ points with the largest $v ^ { \top } x _ { i }$ value, $B ^ { v }$ to be the $3 \alpha | S |$ points with the smallest $v ^ { \top } x _ { i }$ value, and $\mathcal { M } ^ { v } = S \setminus ( \mathcal { T } ^ { v } \cup B ^ { v } )$ . Define $\begin{array} { r l r } { d ( \hat { \mu } , S ) } & { \triangleq } & { \operatorname* { m a x } _ { v \in \mathbb { S } ^ { d - 1 } } \left. v ^ { \top } \left( \mu ( \mathcal { M } ^ { v } ) - \hat { \mu } \right) \right. } \end{array}$ .
+
+Analysis. For any direction $v$ , the truncated mean estimator $\mu ( \mathcal { M } ^ { v } )$ provides a robust estimation of the true mean along the direction $v$ , thus the distance can be simply defined by taking the maximum over all directions $v$ . We show the sensitivity of this simple estimator is bounded by the resilience property $\sigma$ divided by $n$ , which is ${ \cal O } ( ( 1 / n ) \sqrt { \log ( 1 / \alpha ) } )$ once the resilience check is passed. This leads to the following near-optimal sample complexity. We provide a proof in Appendix H.2.
+
+Theorem 7 (Exponential time algorithm for sub-Gaussian distributions). Algorithm 2 is $( \varepsilon , \delta )$ -DP. Under Assumption $^ { l }$ , this algorithm achieves $\| \hat { \mu } - \mu \| _ { 2 } = O ( \alpha \sqrt { \log ( 1 / \alpha ) } )$ with probability $1 - \zeta i f$
+
+$$
+n = \widetilde \Omega \Big ( \frac { d + \log \frac { 1 } { \xi } } { \alpha ^ { 2 } \log \frac { 1 } { \alpha } } + \frac { d \log \Big ( d \sqrt { \log ( d n / \zeta ) } / \alpha \Big ) + d ^ { 1 / 2 } \log \frac { 1 } { \delta } + \log \frac { 1 } { \zeta } } { \varepsilon \alpha } + \frac { \sqrt { d \log \frac { 1 } { \delta } } \log \frac { d } { \zeta \delta } } { \varepsilon } \Big ) ~ .
+$$
+
+Run-time. Computing $R ( S )$ exactly can take $O ( d e ^ { \Theta ( n ) } )$ operations. The exponential mechanism implemented with $\alpha$ -covering for $\hat { \mu }$ and a constant covering for $v$ can take $O ( n d ( \sqrt { \log ( d n / \zeta ) } / \alpha ) ^ { d } )$ operations.
+
+# 5 Conclusion
+
+Differentially private mean estimation is brittle against a small fraction of the samples being corrupted by an adversary. We show that robustness can be achieved without any increase in the sample complexity by introducing a novel DP mean estimator, which requires run-time exponential in the dimension of the samples. The technical contribution is in leveraging the resilience property of well-behaved distributions in an innovative way to not only find robust mean (which is the typical use case of resilience) but also bound sensitivity for optimal privacy guarantee. To cope with the computational challenge, we propose an efficient algorithm, which we call PRIME, that achieves the optimal target accuracy at the cost of an increased sample complexity. The technical contributions are $( i )$ a novel framework for private iterative filtering and its tight analysis of the end-to-end sensitivity and $( i i )$ novel filtering algorithm of DPTHRESHOLD which is critical in privately running matrix multiplicative weights and hence significantly reducing the number of accesses to the database. With appropriately chosen parameters, we show that our exponential time approach achieves near-optimal guarantees for both sub-Gaussian and covariance bounded distributions and PRIME achieves the same accuracy efficiently but at the cost of an increased sample complexity by a $d ^ { 1 / 2 }$ factor.
+
+There are several directions for improving our results further and applying the framework to solve other problems. PRIME provides a new design principle for private and robust estimation. This can be more broadly applied to fundamental statistical analyses such as robust covariance estimation [28, 30, 64] robust PCA [60, 48], and robust linear regression [59, 35].
+
+PRIME could be improved in a few directions. First, the sample complexity of $\widetilde \Omega ( ( d / ( \alpha ^ { 2 } \log ( 1 / \alpha ) ) ) +$ $( d ^ { 3 / 2 } / ( \varepsilon \alpha \log ( 1 / \alpha ) ) ) \log ( 1 / \delta ) )$ in Theorem 6 is suboptimal in the second term. Improving the $d ^ { 3 / 2 }$ factor requires bypassing differentially private singular value decomposition, which seems to be a challenging task. However, it might be possible to separate the $\bar { \log ( 1 / \delta ) }$ factor from the rest of the terms and get an additive error of the form $\widetilde \Omega ( ( d / ( \alpha ^ { 2 } \log ( 1 / \alpha ) ) ) + ( d ^ { 3 / 2 } / ( \varepsilon \alpha \log ( 1 / \alpha ) ) ) +$ $( 1 / \varepsilon ) \log ( 1 / \delta ) )$ . This requires using Laplace mechanism in private MMW (line 16 Algortihm 10). Secondly, the time complexity of PRIME is dominated by computation time of the matrix exponential in (line 16 Algortihm 10). Total number of operations scale as ${ \widetilde { O } } ( d ^ { 3 } + n d ^ { 2 } )$ . One might hope to achieve $\widetilde O ( n d )$ time complexity using approximate computations of $\tau _ { j }$ ’s using techniques from [36]. This does not improve the sample complexity, as the number of times the dataset is accessed remains the same. Finally, for (non-robust) private mean estimation, COINPRESS provides a practical improvement in the small sample regime by progressively refining the search space [12]. The same principle could be applied to PRIME to design a robust version of COINPRESS. One important question remains open; how are differential privacy and robust statistics fundamentally related? We believe our exponential time algorithm hints on a fundamental connection between robust statistics of a data projected onto one-dimensional subspace and sensitivity of resulting score function for the exponential mechanism. It is an interesting direction to pursue this connection further to design novel algorithms that bridge privacy and robustness.
+
+# Acknowledgement
+
+Sham Kakade acknowledges funding from the National Science Foundation under award CCF1703574. Sewoong Oh acknowledges funding from Google faculty research award, NSF grants IIS-1929955, CCF-1705007, CNS-2002664, CCF 2019844 as a part of Institute for Foundation of Machine Learning, and CNS-2112471 as a part of Institute for Future Edge Networks and Distributed Intelligence.
+
+# References
+
+[1] Martin Abadi, Andy Chu, Ian Goodfellow, H Brendan McMahan, Ilya Mironov, Kunal Talwar, and Li Zhang. Deep learning with differential privacy. In Proceedings of the 2016 ACM SIGSAC Conference on Computer and Communications Security, pages 308–318, 2016.   
+[2] John M Abowd. The us census bureau adopts differential privacy. In Proceedings of the 24th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining, pages 2867–2867, 2018.   
+[3] Ishaq Aden-Ali, Hassan Ashtiani, and Gautam Kamath. On the sample complexity of privately learning unbounded high-dimensional gaussians. arXiv preprint arXiv:2010.09929, 2020.   
+[4] Zeyuan Allen-Zhu, Zhenyu Liao, and Lorenzo Orecchia. Spectral sparsification and regret minimization beyond matrix multiplicative updates. In Proceedings of the forty-seventh annual ACM symposium on Theory of computing, pages 237–245, 2015.   
+[5] Edoardo Amaldi and Viggo Kann. The complexity and approximability of finding maximum feasible subsystems of linear relations. Theoretical computer science, 147(1-2):181–210, 1995.   
+[6] Frank J Anscombe. Rejection of outliers. Technometrics, 2(2):123–146, 1960. [7] Ainesh Bakshi and Pravesh Kothari. List-decodable subspace recovery via sum-of-squares. arXiv preprint arXiv:2002.05139, 2020.   
+[8] S. Balakrishnan, S. S. Du, J. Li, and A. Singh. Computationally efficient robust sparse estimation in high dimensions. In Proceedings of the 30th Conference on Learning Theory, COLT 2017, pages 169–212, 2017.   
+[9] Amos Beimel, Shay Moran, Kobbi Nissim, and Uri Stemmer. Private center points and learning of halfspaces. arXiv preprint arXiv:1902.10731, 2019.   
+[10] K. Bhatia, P. Jain, P. Kamalaruban, and P. Kar. Consistent robust regression. In Advances in Neural Information Processing Systems 30: Annual Conference on Neural Information Processing Systems 2017, pages 2107–2116, 2017.   
+[11] Kush Bhatia, Prateek Jain, and Purushottam Kar. Robust regression via hard thresholding. In Advances in Neural Information Processing Systems, pages 721–729, 2015.   
+[12] Sourav Biswas, Yihe Dong, Gautam Kamath, and Jonathan Ullman. Coinpress: Practical private mean and covariance estimation. arXiv preprint arXiv:2006.06618, 2020.   
+[13] Avrim Blum, Cynthia Dwork, Frank McSherry, and Kobbi Nissim. Practical privacy: the sulq framework. In Proceedings of the twenty-fourth ACM SIGMOD-SIGACT-SIGART symposium on Principles of database systems, pages 128–138, 2005.   
+[14] Mark Bun, Gautam Kamath, Thomas Steinke, and Steven Z Wu. Private hypothesis selection. In Advances in Neural Information Processing Systems, pages 156–167, 2019.   
+[15] T Tony Cai, Yichen Wang, and Linjun Zhang. The cost of privacy: Optimal rates of convergence for parameter estimation with differential privacy. arXiv preprint arXiv:1902.04495, 2019.   
+[16] Clément L Canonne, Gautam Kamath, Audra McMillan, Jonathan Ullman, and Lydia Zakynthinou. Private identity testing for high-dimensional distributions. arXiv preprint arXiv:1905.11947, 2019.   
+[17] Moses Charikar, Jacob Steinhardt, and Gregory Valiant. Learning from untrusted data. In Proceedings of the 49th Annual ACM SIGACT Symposium on Theory of Computing, pages 47–60, 2017.   
+[18] Kamalika Chaudhuri, Anand D Sarwate, and Kaushik Sinha. A near-optimal algorithm for differentially-private principal components. The Journal of Machine Learning Research, 14(1):2905–2943, 2013.   
+[19] Xinyun Chen, Chang Liu, Bo Li, Kimberly Lu, and Dawn Song. Targeted backdoor attacks on deep learning systems using data poisoning. arXiv preprint arXiv:1712.05526, 2017.   
+[20] Yu Cheng, Ilias Diakonikolas, and Rong Ge. High-dimensional robust mean estimation in nearly-linear time. In Proceedings of the Thirtieth Annual ACM-SIAM Symposium on Discrete Algorithms, pages 2755–2771. SIAM, 2019.   
+[21] Yu Cheng, Ilias Diakonikolas, Rong Ge, and David P Woodruff. Faster algorithms for highdimensional robust covariance estimation. In Conference on Learning Theory, pages 727–757. PMLR, 2019.   
+[22] Yeshwanth Cherapanamjeri, Sidhanth Mohanty, and Morris Yau. List decodable mean estimation in nearly linear time. arXiv preprint arXiv:2005.09796, 2020.   
+[23] Arnak Dalalyan and Philip Thompson. Outlier-robust estimation of a sparse linear model using $\ell _ { 1 }$ -penalized huber’s $m$ -estimator. In Advances in Neural Information Processing Systems, pages 13188–13198, 2019.   
+[24] Jules Depersin and Guillaume Lecué. Robust subgaussian estimation of a mean vector in nearly linear time. arXiv preprint arXiv:1906.03058, 2019.   
+[25] Luc Devroye and Gábor Lugosi. Combinatorial methods in density estimation. Springer Science & Business Media, 2012.
+
+[26] Aditya Dhar and Jason Huang. Designing differentially private estimators in high dimensions. arXiv preprint arXiv:2006.01944, 2020.
+
+[27] Ilias Diakonikolas, Samuel B Hopkins, Daniel Kane, and Sushrut Karmalkar. Robustly learning any clusterable mixture of gaussians. arXiv preprint arXiv:2005.06417, 2020.
+
+[28] Ilias Diakonikolas, Gautam Kamath, Daniel Kane, Jerry Li, Ankur Moitra, and Alistair Stewart. Robust estimators in high-dimensions without the computational intractability. SIAM Journal on Computing, 48(2):742–864, 2019.
+
+[29] Ilias Diakonikolas, Gautam Kamath, Daniel Kane, Jerry Li, Jacob Steinhardt, and Alistair Stewart. Sever: A robust meta-algorithm for stochastic optimization. In International Conference on Machine Learning, pages 1596–1606, 2019.
+
+[30] Ilias Diakonikolas, Gautam Kamath, Daniel M. Kane, Jerry Li, Ankur Moitra, and Alistair Stewart. Being Robust (in High Dimensions) Can Be Practical. arXiv e-prints, page arXiv:1703.00893, March 2017.
+
+[31] Ilias Diakonikolas, Gautam Kamath, Daniel M Kane, Jerry Li, Ankur Moitra, and Alistair Stewart. Robustly learning a gaussian: Getting optimal error, efficiently. In Proceedings of the Twenty-Ninth Annual ACM-SIAM Symposium on Discrete Algorithms, pages 2683–2702. SIAM, 2018.
+
+[32] Ilias Diakonikolas and Daniel M Kane. Recent advances in algorithmic high-dimensional robust statistics. arXiv preprint arXiv:1911.05911, 2019.
+
+[33] Ilias Diakonikolas, Daniel M Kane, and Alistair Stewart. Statistical query lower bounds for robust estimation of high-dimensional gaussians and gaussian mixtures. In 2017 IEEE 58th Annual Symposium on Foundations of Computer Science (FOCS), pages 73–84. IEEE, 2017.
+
+[34] Ilias Diakonikolas, Daniel M Kane, and Alistair Stewart. List-decodable robust mean estimation and learning mixtures of spherical gaussians. In Proceedings of the 50th Annual ACM SIGACT Symposium on Theory of Computing, pages 1047–1060, 2018.
+
+[35] Ilias Diakonikolas, Weihao Kong, and Alistair Stewart. Efficient algorithms and lower bounds for robust linear regression. In Proceedings of the Thirtieth Annual ACM-SIAM Symposium on Discrete Algorithms, pages 2745–2754. SIAM, 2019.
+
+[36] Yihe Dong, Samuel Hopkins, and Jerry Li. Quantum entropy scoring for fast robust mean estimation and improved outlier detection. In Advances in Neural Information Processing Systems, pages 6067–6077, 2019.
+
+[37] Cynthia Dwork, Frank McSherry, Kobbi Nissim, and Adam Smith. Calibrating noise to sensitivity in private data analysis. In Theory of cryptography conference, pages 265–284. Springer, 2006.
+
+[38] Cynthia Dwork and Aaron Roth. The algorithmic foundations of differential privacy. Foundations and Trends in Theoretical Computer Science, 9(3-4):211–407, 2014.
+
+[39] Cynthia Dwork, Kunal Talwar, Abhradeep Thakurta, and Li Zhang. Analyze gauss: optimal bounds for privacy-preserving principal component analysis. In Proceedings of the forty-sixth annual ACM symposium on Theory of computing, pages 11–20, 2014.
+
+[40] Úlfar Erlingsson, Vasyl Pihur, and Aleksandra Korolova. Rappor: Randomized aggregatable privacy-preserving ordinal response. In Proceedings of the 2014 ACM SIGSAC conference on computer and communications security, pages 1054–1067, 2014.
+
+[41] Giulia Fanti, Vasyl Pihur, and Úlfar Erlingsson. Building a rappor with the unknown: Privacypreserving learning of associations and data dictionaries. Proceedings on Privacy Enhancing Technologies, 2016(3):41–61, 2016.
+
+[42] Chao Gao et al. Robust regression via mutivariate regression depth. Bernoulli, 26(2):1139–1170, 2020.
+
+[43] Sam Hopkins, Jerry Li, and Fred Zhang. Robust and heavy-tailed mean estimation made simple, via regret minimization. Advances in Neural Information Processing Systems, 33, 2020.
+
+[44] Samuel B Hopkins. Mean estimation with sub-gaussian rates in polynomial time. Annals of Statistics, 48(2):1193–1213, 2020.   
+[45] Samuel B Hopkins and Jerry Li. Mixture models, robustness, and sum of squares proofs. In Proceedings of the 50th Annual ACM SIGACT Symposium on Theory of Computing, pages 1021–1034, 2018.   
+[46] Samuel B Hopkins and Jerry Li. How hard is robust mean estimation? In Conference on Learning Theory, pages 1649–1682. PMLR, 2019.   
+[47] Peter J. Huber. Robust Estimation of a Location Parameter. The Annals of Mathematical Statistics, 35(1):73 – 101, 1964.   
+[48] Arun Jambulapati, Jerry Li, and Kevin Tian. Robust sub-gaussian principal component analysis and width-independent schatten packing. Advances in Neural Information Processing Systems, 33, 2020.   
+[49] He Jia and Santosh Vempala. Robustly clustering a mixture of gaussians. arXiv preprint arXiv:1911.11838, 2019.   
+[50] Peter Kairouz, H Brendan McMahan, Brendan Avent, Aurélien Bellet, Mehdi Bennis, Arjun Nitin Bhagoji, Keith Bonawitz, Zachary Charles, Graham Cormode, Rachel Cummings, et al. Advances and open problems in federated learning. arXiv preprint arXiv:1912.04977, 2019.   
+[51] Peter Kairouz, Sewoong Oh, and Pramod Viswanath. The composition theorem for differential privacy. In International conference on machine learning, pages 1376–1385, 2015.   
+[52] Gautam Kamath, Jerry Li, Vikrant Singhal, and Jonathan Ullman. Privately learning highdimensional distributions. In Conference on Learning Theory, pages 1853–1902, 2019.   
+[53] Gautam Kamath, Or Sheffet, Vikrant Singhal, and Jonathan Ullman. Differentially private algorithms for learning mixtures of separated gaussians. In 2020 Information Theory and Applications Workshop (ITA), pages 1–62. IEEE, 2020.   
+[54] Gautam Kamath, Vikrant Singhal, and Jonathan Ullman. Private mean estimation of heavy-tailed distributions. arXiv preprint arXiv:2002.09464, 2020.   
+[55] Haim Kaplan, Katrina Ligett, Yishay Mansour, Moni Naor, and Uri Stemmer. Privately learning thresholds: Closing the exponential gap. In Conference on Learning Theory, pages 2263–2285. PMLR, 2020.   
+[56] Sushrut Karmalkar, Adam Klivans, and Pravesh Kothari. List-decodable linear regression. In Advances in Neural Information Processing Systems, pages 7423–7432, 2019.   
+[57] Sushrut Karmalkar and Eric Price. Compressed sensing with adversarial sparse noise via l1 regression. In 2nd Symposium on Simplicity in Algorithms, 2019.   
+[58] Vishesh Karwa and Salil Vadhan. Finite sample differentially private confidence intervals. arXiv preprint arXiv:1711.03908, 2017.   
+[59] Adam Klivans, Pravesh K Kothari, and Raghu Meka. Efficient algorithms for outlier-robust regression. In Conference On Learning Theory, pages 1420–1430, 2018.   
+[60] Weihao Kong, Raghav Somani, Sham Kakade, and Sewoong Oh. Robust meta-learning for mixed linear regression with small batches. Advances in Neural Information Processing Systems, 33, 2020.   
+[61] Pravesh K Kothari, Jacob Steinhardt, and David Steurer. Robust moment estimation and improved clustering via sum of squares. In Proceedings of the 50th Annual ACM SIGACT Symposium on Theory of Computing, pages 1035–1046, 2018.   
+[62] Kevin A Lai, Anup B Rao, and Santosh Vempala. Agnostic estimation of mean and covariance. In 2016 IEEE 57th Annual Symposium on Foundations of Computer Science (FOCS), pages 665–674. IEEE, 2016.   
+[63] Jerry Li. CSE 599-M, Lecture Notes: Robustness in Machine Learning , 2019. URL: https: //jerryzli.github.io/robust-ml-fall19/lec7.pdf.   
+[64] Jerry Li and Guanghao Ye. Robust gaussian covariance estimation in nearly-matrix multiplication time. Advances in Neural Information Processing Systems, 33, 2020.   
+[65] Liu Liu, Yanyao Shen, Tianyang Li, and Constantine Caramanis. High dimensional robust sparse regression. arXiv preprint arXiv:1805.11643, 2018.   
+[66] Xiaohui Liu. Fast implementation of the tukey depth. Computational Statistics, 32(4):1395– 1410, 2017.   
+[67] Xiaohui Liu, Karl Mosler, and Pavlo Mozharovskyi. Fast computation of tukey trimmed regions and median in dimension p> 2. Journal of Computational and Graphical Statistics, 28(3):682–697, 2019.   
+[68] Gábor Lugosi, Shahar Mendelson, et al. Sub-gaussian estimators of the mean of a random vector. Annals of Statistics, 47(2):783–794, 2019.   
+[69] Frank McSherry and Kunal Talwar. Mechanism design via differential privacy. In 48th Annual IEEE Symposium on Foundations of Computer Science (FOCS’07), pages 94–103. IEEE, 2007.   
+[70] Bhaskar Mukhoty, Govind Gopakumar, Prateek Jain, and Purushottam Kar. Globally-convergent iteratively reweighted least squares for robust regression problems. In The 22nd International Conference on Artificial Intelligence and Statistics, pages 313–322, 2019.   
+[71] A. Prasad, A. S. Suggala, S. Balakrishnan, and P. Ravikumar. Robust estimation via robust gradient estimation. arXiv preprint arXiv:1802.06485, 2018.   
+[72] Prasad Raghavendra and Morris Yau. List decodable learning via sum of squares. In Proceedings of the Fourteenth Annual ACM-SIAM Symposium on Discrete Algorithms, pages 161–180. SIAM, 2020.   
+[73] Jacob Steinhardt, Moses Charikar, and Gregory Valiant. Resilience: A criterion for learning in the presence of arbitrary outliers. In 9th Innovations in Theoretical Computer Science Conference (ITCS 2018). Schloss Dagstuhl-Leibniz-Zentrum fuer Informatik, 2018.   
+[74] Jun Tang, Aleksandra Korolova, Xiaolong Bai, Xueqiang Wang, and Xiaofeng Wang. Privacy loss in apple’s implementation of differential privacy on macos 10.12. arXiv preprint arXiv:1709.02753, 2017.   
+[75] Terence Tao. Topics in random matrix theory, volume 132. American Mathematical Soc., 2012.   
+[76] John W Tukey. A survey of sampling from contaminated distributions. Contributions to probability and statistics, pages 448–485, 1960.   
+[77] Martin J Wainwright. High-dimensional statistics: A non-asymptotic viewpoint, volume 48. Cambridge University Press, 2019.   
+[78] Lu Wei, Anand D Sarwate, Jukka Corander, Alfred Hero, and Vahid Tarokh. Analysis of a privacy-preserving pca algorithm using random matrix theory. In 2016 IEEE Global Conference on Signal and Information Processing (GlobalSIP), pages 1335–1339. IEEE, 2016.   
+[79] Huang Xiao, Battista Biggio, Gavin Brown, Giorgio Fumera, Claudia Eckert, and Fabio Roli. Is feature selection secure against training data poisoning? In International Conference on Machine Learning, pages 1689–1698. PMLR, 2015.   
+[80] Huanyu Zhang, Gautam Kamath, Janardhan Kulkarni, and Zhiwei Steven Wu. Privately learning markov random fields. arXiv preprint arXiv:2002.09463, 2020.
\ No newline at end of file
diff --git a/parse/train/CuQoImkKkIj/CuQoImkKkIj_content_list.json b/parse/train/CuQoImkKkIj/CuQoImkKkIj_content_list.json
new file mode 100644
index 0000000000000000000000000000000000000000..3503e1488897b9e6ff384092f514b118f9dc5dc9
--- /dev/null
+++ b/parse/train/CuQoImkKkIj/CuQoImkKkIj_content_list.json
@@ -0,0 +1,1458 @@
+[
+    {
+        "type": "text",
+        "text": "Robust and differentially private mean estimation ",
+        "text_level": 1,
+        "bbox": [
+            196,
+            122,
+            802,
+            148
+        ],
+        "page_idx": 0
+    },
+    {
+        "type": "text",
+        "text": "Xiyang Liu, Weihao Kong, Sham Kakade, Sewoong Oh ",
+        "bbox": [
+            285,
+            200,
+            710,
+            215
+        ],
+        "page_idx": 0
+    },
+    {
+        "type": "text",
+        "text": "Paul G. Allen School of Computer Science and Engineering, University of Washington {xiyangl,whkong,sham,sewoong}@cs.washington.edu ",
+        "bbox": [
+            294,
+            215,
+            699,
+            257
+        ],
+        "page_idx": 0
+    },
+    {
+        "type": "text",
+        "text": "Abstract ",
+        "text_level": 1,
+        "bbox": [
+            462,
+            292,
+            535,
+            309
+        ],
+        "page_idx": 0
+    },
+    {
+        "type": "text",
+        "text": "In statistical learning and analysis from shared data, which is increasingly widely adopted in platforms such as federated learning and meta-learning, there are two major concerns: privacy and robustness. Each participating individual should be able to contribute without the fear of leaking one’s sensitive information. At the same time, the system should be robust in the presence of malicious participants inserting corrupted data. Recent algorithmic advances in learning from shared data focus on either one of these threats, leaving the system vulnerable to the other. We bridge this gap for the canonical problem of estimating the mean from i.i.d. samples. We introduce PRIME, which is the first efficient algorithm that achieves both privacy and robustness for a wide range of distributions. We further complement this result with a novel exponential time algorithm that improves the sample complexity of PRIME, achieving a near-optimal guarantee and matching a known lower bound for (non-robust) private mean estimation. This proves that there is no extra statistical cost to simultaneously guaranteeing privacy and robustness. ",
+        "bbox": [
+            233,
+            321,
+            766,
+            516
+        ],
+        "page_idx": 0
+    },
+    {
+        "type": "text",
+        "text": "1 Introduction ",
+        "text_level": 1,
+        "bbox": [
+            176,
+            540,
+            310,
+            558
+        ],
+        "page_idx": 0
+    },
+    {
+        "type": "text",
+        "text": "When releasing database statistics on a collection of entries from individuals, we would ideally like to make it impossible to reverse-engineer each individual’s potentially sensitive information. Privacy-preserving techniques add just enough randomness tailored to the statistical task to guarantee protection. At the same time, it is becoming increasingly common to apply such techniques to databases collected from multiple sources, not all of which can be trusted. Emerging data access frameworks, such as federated analyses across users’ devices or data silos [50], make it easier to temper with such collected datasets, leaving private statistical analyses vulnerable to a malicious corruption of a fraction of the data. ",
+        "bbox": [
+            174,
+            570,
+            825,
+            681
+        ],
+        "page_idx": 0
+    },
+    {
+        "type": "text",
+        "text": "Differential privacy has emerged as a widely accepted de facto measure of privacy, which is now a standard in releasing the statistics of the U.S. Census data [2] statistics and also deployed in real-world commercial systems [74, 40, 41]. A statistical analysis is said to be differentially private (DP) if the likelihood of the (randomized) outcome does not change significantly when a single arbitrary entry is added/removed (formally defined in $\\ S 1 . 2 )$ . This provides a strong privacy guarantee: even a powerful adversary who knows all the other entries in the database cannot confidently identify whether a particular individual is participating in the database based on the outcome of the analysis. This ensures plausible deniability, central to protecting an individual’s privacy. ",
+        "bbox": [
+            174,
+            689,
+            825,
+            799
+        ],
+        "page_idx": 0
+    },
+    {
+        "type": "text",
+        "text": "In this paper, we focus on one of the most canonical problems in statistics: estimating the mean of a distribution from i.i.d. samples. For distributions with unbounded support, such as sub-Gaussian and heavy-tailed distributions, fundamental trade-offs between accuracy, sample size, and privacy have only recently been identified [58, 52, 54, 3] and efficient private estimators proposed. However, these approaches are brittle when a fraction of the data is corrupted, posing a real threat, referred to as data poisoning attacks [19, 79]. In defense of such attacks, robust (but not necessarily private) statistics has emerged as a popular setting of recent algorithmic and mathematical breakthroughs [73, 30]. ",
+        "bbox": [
+            174,
+            805,
+            825,
+            902
+        ],
+        "page_idx": 0
+    },
+    {
+        "type": "text",
+        "text": "One might be misled into thinking that privacy ensures robustness since DP guarantees that a single outlier cannot change the estimation too much. This intuition is true only in a low dimension; each sample has to be an obvious outlier to significantly change the mean. However, in a high dimension, each corrupted data point can look perfectly uncorrupted but still shift the mean significant when colluding together (e.g., see Fig. 1). Focusing on the canonical problem of mean estimation, we introduce novel algorithms that achieve robustness and privacy simultaneously even when a fraction of data is corrupted arbitrarily. For such algorithms, there is a fundamental question of interest: do we need more samples to make private mean estimation also robust against adversarial corruption? ",
+        "bbox": [
+            173,
+            90,
+            825,
+            203
+        ],
+        "page_idx": 1
+    },
+    {
+        "type": "text",
+        "text": "Sub-Gaussian distributions. If we can afford exponential run-time in the dimension, robustness can be achieved without extra cost in sample complexity. We introduce a novel estimator that (i) satisfies $( \\varepsilon , \\delta )$ -DP, $( i i )$ achieves near-optimal robustness under $\\alpha$ -fraction of corrupted data, achieving accuracy of $O ( \\alpha \\sqrt { \\log ( 1 / \\alpha ) } )$ nearly matching the fundamental lower bound of $\\Omega ( \\alpha )$ that holds even for a (non-private) robust mean estimation with infinite samples, and $( i i i )$ achieves near-optimal sample complexity matching that of a fundamental lower bound for a (non-robust) private mean estimation as shown in Table 1. ",
+        "bbox": [
+            173,
+            208,
+            825,
+            308
+        ],
+        "page_idx": 1
+    },
+    {
+        "type": "text",
+        "text": "Theorem 1 (Informal Theorem 7, exponential time). Algorithm 2 is $( \\varepsilon , \\delta )$ -DP. When $\\alpha$ fraction of the data is arbitrarily corrupted from $n$ samples from a $d$ -dimensional sub-Gaussian distribution with mean $\\mu$ and an identity sub-Gaussian parameter, if $\\begin{array} { r } { n = \\widetilde \\Omega ( d / \\alpha ^ { 2 } + ( d + d ^ { 1 / 2 } \\log ( 1 / \\delta ) ) / ( \\alpha \\varepsilon ) ) } \\end{array}$ then Algorithm 2 achieves $\\| \\hat { \\mu } - \\mu \\| _ { 2 } = O ( \\alpha \\sqrt { \\log ( 1 / \\alpha ) } ) \\ : w . h . p$ . ",
+        "bbox": [
+            173,
+            310,
+            826,
+            372
+        ],
+        "page_idx": 1
+    },
+    {
+        "type": "text",
+        "text": "We introduce PRIME (PRIvate and robust Mean Estimation) in $\\ S 2 . 3$ with details in Algorithm 9 in Appendix E.1, to achieve computational efficiency. It requires a run-time of only $\\widetilde { \\cal O } ( d ^ { 3 } + n d ^ { 2 } )$ , but at the cost of requiring extra $d ^ { 1 / 2 }$ factor larger number of samples. This cannot be improved upon with current techniques since efficient robust estimators rely on the top PCA directions of the covariance matrix to detect outliers. [78] showed that $\\widetilde \\Omega ( d ^ { 3 / 2 } )$ samples are necessary to compute PCA directions while preserving $( \\varepsilon , \\delta )$ -DP when $\\| x _ { i } \\| _ { 2 } = O ( { \\sqrt { d } } )$ . It remains an open question if this $\\widetilde \\Omega ( d ^ { 3 / 2 } / ( \\alpha \\varepsilon ) )$ bottleneck is fundamental; no matching lower bound is currently known. ",
+        "bbox": [
+            173,
+            380,
+            826,
+            488
+        ],
+        "page_idx": 1
+    },
+    {
+        "type": "text",
+        "text": "Theorem 2 (Informal Theorem 6, polynomial time). PRIME is $( \\varepsilon , \\delta )$ -DP and under the assumption of Thm.1, if $\\begin{array} { r } { \\dot { n } = \\widetilde \\Omega ( d / \\alpha ^ { 2 } + ( d ^ { 3 / 2 } \\log ( 1 / \\delta ) ) / ( \\alpha \\varepsilon ) ) . } \\end{array}$ , achieves $\\begin{array} { r } { \\| \\hat { \\mu } - \\mu \\| _ { 2 } = O ( \\alpha \\sqrt { \\log ( 1 / \\alpha ) } ) ~ w . h . p . } \\end{array}$ . ",
+        "bbox": [
+            174,
+            489,
+            823,
+            522
+        ],
+        "page_idx": 1
+    },
+    {
+        "type": "table",
+        "img_path": "images/1e64b991060acd268af93ff0942da309a127755f1575377d2afa3e229b23eccd.jpg",
+        "table_caption": [],
+        "table_footnote": [],
+        "table_body": "<table><tr><td rowspan=1 colspan=1></td><td rowspan=1 colspan=1>Upper bound (poly-time)</td><td rowspan=1 colspan=1>Upper bound (exp-time)</td><td rowspan=1 colspan=1>Lower bound</td></tr><tr><td rowspan=1 colspan=1>(ε,δ)-DP[52]</td><td rowspan=1 colspan=1>0+dlog1/2(1/8））Qε</td><td rowspan=1 colspan=1>+）</td><td rowspan=1 colspan=1>（+）</td></tr><tr><td rowspan=1 colspan=1>α-corruption [36]</td><td rowspan=1 colspan=1>0（</td><td rowspan=1 colspan=1>（</td><td rowspan=1 colspan=1>（</td></tr><tr><td rowspan=1 colspan=1>α-corruption and(ε,δ)-DP (this paper)</td><td rowspan=1 colspan=1>（+/210g(1/8）)αε[Theorem6]</td><td rowspan=1 colspan=1>0（+d+d1/210g(1/8))Q[Theorem7]</td><td rowspan=1 colspan=1>(）[52]</td></tr></table>",
+        "bbox": [
+            179,
+            536,
+            818,
+            625
+        ],
+        "page_idx": 1
+    },
+    {
+        "type": "text",
+        "text": "Table 1: For estimating the mean $\\mu \\in \\mathbb { R } ^ { d }$ of a sub-Gaussian distribution with a known covariance, we list the sufficient or necessary conditions on the sample sizes to achieve an error $\\| \\hat { \\mu } - \\mu \\| _ { 2 } = \\widetilde { O } ( \\alpha )$ under $( \\varepsilon , \\delta )$ -DP, corruption of an $\\alpha$ -fraction of samples, and both. √ $\\clubsuit$ requires the distribution to be a Gaussian [14] and $\\spadesuit$ requires $\\delta \\leq \\sqrt { d } / n$ . ",
+        "bbox": [
+            173,
+            637,
+            825,
+            702
+        ],
+        "page_idx": 1
+    },
+    {
+        "type": "text",
+        "text": "Heavy-tailed distributions. When samples are drawn from a distribution with a bounded covariance, parameters of Algorithm 2 can be modified to nearly match the optimal sample complexity of (nonrobust) private mean estimation in Table 2. This algorithm also matches the fundamental limit on the accuracy of (non-private) robust estimation, which in this case is $\\Omega ( \\alpha ^ { 1 / 2 } )$ . ",
+        "bbox": [
+            174,
+            712,
+            826,
+            770
+        ],
+        "page_idx": 1
+    },
+    {
+        "type": "text",
+        "text": "Theorem 3 (Informal Theorem 8, exponential time). From a distribution with mean $\\mu \\in \\mathbb { R } ^ { d }$ and covariance $\\Sigma \\preceq \\mathbf { I }$ , $n$ samples are drawn and $\\alpha$ -fraction is corrupted. Algorithm 2 is $( \\varepsilon , \\delta )$ -DP and $i f$ $n = \\widetilde \\Omega ( ( d + d ^ { 1 / 2 } \\log ( 1 / \\delta ) ) / ( \\alpha \\varepsilon ) + d ^ { 1 / 2 } \\log ^ { 3 / 2 } ( 1 / \\delta ) / \\varepsilon )$ achieves $\\| \\hat { \\mu } - \\mu \\| _ { 2 } = O ( \\alpha ^ { 1 / 2 } ) \\ : w . h . p .$ . ",
+        "bbox": [
+            173,
+            772,
+            826,
+            820
+        ],
+        "page_idx": 1
+    },
+    {
+        "type": "text",
+        "text": "The proposed PRIME-HT for covariance bounded distributions achieve computational efficiency at the cost of an extra factor of $d ^ { 1 / 2 }$ in sample size. This bottleneck is also due to DP PCA, and it remains open whether this gap can be closed by an efficient estimator. ",
+        "bbox": [
+            174,
+            827,
+            821,
+            872
+        ],
+        "page_idx": 1
+    },
+    {
+        "type": "text",
+        "text": "Theorem 4 (Informal Theorem 9, polynomial time). PRIME-HT is $( \\varepsilon , \\delta ) – D P$ and if $n \\_ =$ $\\widetilde \\Omega ( ( d ^ { 3 / 2 } \\log ( 1 / \\delta ) ) / ( \\alpha \\varepsilon ) )$ achieves $\\| \\hat { \\mu } - \\mu \\| _ { 2 } = O ( \\alpha ^ { 1 / 2 } ) \\ : w . h . p$ . under the assumptions of Thm. 3. ",
+        "bbox": [
+            171,
+            873,
+            823,
+            906
+        ],
+        "page_idx": 1
+    },
+    {
+        "type": "table",
+        "img_path": "images/191d63f4ee021bb7fbfe8805bb187077f698af2e6e124ee5962d88fbe956348f.jpg",
+        "table_caption": [],
+        "table_footnote": [],
+        "table_body": "<table><tr><td rowspan=1 colspan=1></td><td rowspan=1 colspan=1>Upper bound (poly-time)</td><td rowspan=1 colspan=1>Upper bound (exp-time)</td><td rowspan=1 colspan=1>Lower bound</td></tr><tr><td rowspan=1 colspan=1>(ε,δ)-DP [54]</td><td rowspan=1 colspan=1>O（d10g1/2（1/）Qε</td><td rowspan=1 colspan=1>O（d10g1/2（1/0））Qε</td><td rowspan=1 colspan=1>2（）</td></tr><tr><td rowspan=1 colspan=1>α-corruption [36]</td><td rowspan=1 colspan=1>）</td><td rowspan=1 colspan=1></td><td rowspan=1 colspan=1>(）</td></tr><tr><td rowspan=1 colspan=1>α-corruption and(ε,δ)-DP (this paper)</td><td rowspan=1 colspan=1>0（(/210g(1/8))αε[Theorem 9]</td><td rowspan=1 colspan=1>（+d1721g/2(18）αε[Theorem 8]</td><td rowspan=1 colspan=1>(([54])</td></tr></table>",
+        "bbox": [
+            184,
+            88,
+            813,
+            176
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "text",
+        "text": "Table 2: For estimating the mean $\\mu \\in \\mathbb { R } ^ { d }$ of a covariance bounded distribution, we list the sufficient or necessary conditions on the sample size to achieve an error $\\| \\hat { \\mu } - \\mu \\| _ { 2 } = O ( \\alpha ^ { 1 / 2 } )$ under $( \\varepsilon , \\delta )$ -DP, corruption of an $\\alpha$ -fraction of samples, and both. ",
+        "bbox": [
+            174,
+            189,
+            825,
+            234
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "text",
+        "text": "1.1 Technical contributions ",
+        "text_level": 1,
+        "bbox": [
+            174,
+            257,
+            377,
+            272
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "text",
+        "text": "We introduce PRIME which simultaneously achieves $( \\varepsilon , \\delta )$ -DP and robustness against $\\alpha$ -fraction of corruption. A major challenge in making a standard filter-based robust estimation algorithm (e.g., [30]) private is the high sensitivity of the filtered set that we pass from one iteration to the next. We propose a new framework which makes private only the statistics of the set, hence significantly reducing the sensitivity. Our major innovation is a tight analysis of the end-to-end sensitivity of this multiple interactive accesses to the database. This is critical in achieving robustness while preserving privacy and is also of independent interest in making general iterative filtering algorithms private. ",
+        "bbox": [
+            173,
+            281,
+            826,
+            381
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "text",
+        "text": "The classical filter approach (see, e.g. [30]) needs to access the database √ $O ( d )$ times, which brings an extra $O ( { \\sqrt { d } } )$ factor in the sample complexity due to DP composition. In order to reduce the iteration complexity, following the approach in [36], we propose filtering multiple directions simultaneously using a new score based on the matrix multiplicative weights (MMW). In order to privatize the MMW filter, our major innovation is a novel adaptive filtering algorithm DPTHRESHOLD(·) that outputs a single private threshold which guarantees sufficient progress at every iteration. This brings the number of database accesses from $O ( d )$ to $O ( ( \\log d ) ^ { 2 } )$ . ",
+        "bbox": [
+            174,
+            386,
+            825,
+            486
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "text",
+        "text": "One downside of PRIME is that it requires an extra $d ^ { 1 / 2 }$ factor in the sample complexity, compared to known lower bounds for (non-robust) DP mean estimation. To investigate whether this is also necessary, we propose a sample optimal exponential time robust mean estimation algorithm in $\\ S 4$ and prove that there is no extra statistical cost to jointly requiring privacy and robustness. Our major technical innovations is in using resilience property of the dataset to not only find robust mean (which is the typical use case of resilience) but also bound sensitivity of that robust mean. ",
+        "bbox": [
+            173,
+            492,
+            825,
+            577
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "text",
+        "text": "1.2 Preliminary on differential privacy (DP) ",
+        "text_level": 1,
+        "bbox": [
+            174,
+            592,
+            493,
+            608
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "text",
+        "text": "DP is a formal metric for measuring privacy leakage when a dataset is accessed with a query [37]. ",
+        "bbox": [
+            173,
+            617,
+            813,
+            632
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "text",
+        "text": "Definition 1.1. Given two datasets $S = \\{ x _ { i } \\} _ { i = 1 } ^ { n }$ and $S ^ { \\prime } = \\{ x _ { i } ^ { \\prime } \\} _ { i = 1 } ^ { n ^ { \\prime } }$ , we say $S$ and $S ^ { \\prime }$ are neighboring $i f d _ { \\triangle } ( S , S ^ { \\prime } ) \\leq 1$ where $d _ { \\triangle } ( S , S ^ { \\prime } ) \\triangleq \\operatorname* { m a x } \\{ | S \\setminus S ^ { \\prime } | , | S ^ { \\prime } \\setminus S | \\}$ , which is denoted by $S \\sim S ^ { \\prime }$ . For an output of a stochastic query q on a database, we say q satisfies $( \\varepsilon , \\delta )$ -differential privacy for some $\\varepsilon > 0$ and $\\delta \\in ( 0 , 1 )$ i $\\begin{array} { r } { \\textstyle f \\mathbb { P } ( q ( S ) \\in A ) \\le e ^ { \\varepsilon } \\mathbb { P } ( q ( S ^ { \\prime } ) \\in A ) + \\delta } \\end{array}$ for all $S \\sim S ^ { \\prime }$ and all subset $A$ . ",
+        "bbox": [
+            174,
+            636,
+            825,
+            696
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "text",
+        "text": "Let $z \\sim \\mathrm { L a p } ( b )$ be a random vector with entries i.i.d. sampled from Laplace distribution with pdf $( 1 / 2 b ) e ^ { - | z | / b }$ . Let $z \\sim \\mathcal { N } ( \\mu , \\Sigma )$ denote a Gaussian random vector with mean $\\mu$ and covariance $\\Sigma$ . ",
+        "bbox": [
+            173,
+            705,
+            823,
+            736
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "text",
+        "text": "Definition 1.2. The sensitivity of a query $f ( S ) ~ \\in ~ \\mathbb { R } ^ { k }$ is defined as $\\Delta _ { p } ~ = ~ \\operatorname* { s u p } _ { S \\sim S ^ { \\prime } } \\| f ( S ) ~ -$ $f ( S ^ { \\prime } ) \\| _ { p }$ for a norm $\\begin{array} { r } { \\| { \\boldsymbol x } \\| _ { p } ~ = ~ ( \\sum _ { i \\in [ k ] } | x _ { i } | ^ { p } ) ^ { 1 / p } } \\end{array}$ . For $p ~ = ~ 1$ , the Laplace mechanism outputs $f ( S ) + \\mathrm { L a p } ( { \\Delta _ { 1 } } / { \\varepsilon } )$ and achieves $( \\varepsilon , 0 ) – D P \\ : I 3 7 J$ . For $p = 2$ , the Gaussian mechanism outputs $f ( S ) + \\mathcal { N } ( 0 , ( \\Delta _ { 2 } ( \\sqrt { 2 \\log ( 1 . 2 5 / \\delta ) } ) / \\varepsilon ) ^ { 2 } \\mathbf { I } )$ and achieves $( \\varepsilon , \\delta ) – D P / 3 8 J$ . ",
+        "bbox": [
+            173,
+            739,
+            825,
+            805
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "text",
+        "text": "We use these output perturbation mechanisms along with the exponential mechanism [69] as building blocks. Appendix A provides detailed survey of privacy and robust estimation. ",
+        "bbox": [
+            173,
+            813,
+            823,
+            842
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "text",
+        "text": "1.3 Problem formulation ",
+        "text_level": 1,
+        "bbox": [
+            176,
+            857,
+            359,
+            872
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "text",
+        "text": "We are given $n$ samples from a sub-Gaussian distribution with a known covariance but unknown mean, and $\\alpha$ fraction of the samples are corrupted by an adversary. Our goal is to estimate the unknown mean. We follow the standard definition of adversary in [30], which can adaptively choose which samples to corrupt and arbitrarily replace them with any points. ",
+        "bbox": [
+            174,
+            882,
+            821,
+            911
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "text",
+        "text": "",
+        "bbox": [
+            171,
+            90,
+            823,
+            119
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "Assumption 1. An uncorrupted dataset $S _ { \\mathrm { g o o d } }$ consists of $n$ i.i.d. samples from a $d$ -dimensional sub-Gaussian distribution with mean $\\mu \\in \\mathbb { R } ^ { d }$ and covariance $\\mathbb { E } [ x x ^ { \\top } ] = \\mathbf { I } _ { d } ,$ , which is 1-sub-Gaussian, i.e., $\\mathbb { E } [ \\exp ( v ^ { \\top } x ) ] \\leq \\exp ( \\| v \\| _ { 2 } ^ { 2 } / 2 )$ for all $v \\in \\mathbb { R } ^ { d }$ . For some $\\alpha \\in ( 0 , 1 / 2 )$ , we are given a corrupted dataset $S = \\{ x _ { i } \\in \\mathbb { R } ^ { d } \\} _ { i = 1 } ^ { n }$ where an adversary adaptively inspects all the samples in $S _ { \\mathrm { g o o d } }$ , removes αn of them, and replaces them with $S _ { \\mathrm { b a d } }$ which are αn arbitrary points in $\\mathbb { R } ^ { d }$ . ",
+        "bbox": [
+            173,
+            122,
+            825,
+            199
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "Similarly, we consider the same problem for heavy-tailed distributions with a bounded covariance.   \nWe present the assumption and main results for covariance bounded distributions in Appendix B. ",
+        "bbox": [
+            174,
+            209,
+            823,
+            238
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "Notations. Let $[ n ] = \\{ 1 , 2 , \\dots , n \\}$ . For $x \\in \\mathbb { R } ^ { d }$ , we use $\\textstyle \\| x \\| _ { 2 } = ( \\sum _ { i \\in [ d ] } ( x _ { i } ) ^ { 2 } ) ^ { 1 / 2 }$ to denote the Euclidean norm. For $X \\in \\mathbb { R } ^ { d \\times d }$ , we use $\\begin{array} { r } { \\| X \\| _ { 2 } = \\operatorname* { m a x } _ { \\| v \\| _ { 2 } = 1 } \\| X v \\| _ { 2 } } \\end{array}$ to denote the spectral norm. The $d \\times d$ identity matrix is $\\mathbf { I } _ { d \\times d }$ . Whenever it is clear from context, we use $S$ to denote both a set of data points and also the set of indices of those data points. $\\widetilde O$ and $\\widetilde \\Omega$ hide poly-logarithmic factors in $d , n , 1 / \\alpha$ , and the failure probability. ",
+        "bbox": [
+            173,
+            244,
+            826,
+            323
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "Outline. We present PRIME for sub-Gaussian distribution in $\\ S 2$ , and present theoretical analysis in $\\ S 3$ . We then introduce an exponential time algorithm with near optimal guarantee in $\\ S 4$ . Due to space constraints, analogous results for heavy-tailed distributions are presented in Appendix B. ",
+        "bbox": [
+            173,
+            328,
+            825,
+            371
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "2 PRIME: efficient algorithm for robust and DP mean estimation ",
+        "text_level": 1,
+        "bbox": [
+            173,
+            388,
+            733,
+            406
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "In order to describe the proposed algorithm PRIME, we need to first describe a standard (non-private) iterative filtering algorithm for robust mean estimation. ",
+        "bbox": [
+            174,
+            420,
+            825,
+            449
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "2.1 Background on (non-private) iterative filtering for robust mean estimation ",
+        "text_level": 1,
+        "bbox": [
+            174,
+            464,
+            728,
+            479
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "Non-private robust mean estimation approaches recursively apply the following filter, whose framework is first proposed in [28]. Given a dataset $S = \\{ x _ { i } \\} _ { i = 1 } ^ { n }$ , the current set $S _ { 0 } \\subseteq [ n ]$ of data points is updated starting with $S _ { 1 } = [ n ]$ . At each step, the following filter (Algorithm 1 in [63]) attempts to detect the corrupted data points and remove them. ",
+        "bbox": [
+            174,
+            489,
+            826,
+            546
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "1. Compute the top eigenvector $v _ { t } \\gets \\arg \\operatorname* { m a x } _ { v : \\| v \\| _ { 2 } = 1 } v ^ { \\top } \\mathrm { C o v } ( S _ { t - 1 } ) v$ of the covariance of the current data set $\\{ x _ { i } \\} _ { i \\in S _ { t - 1 } }$ ;   \n2. Compute scores for all data points $j \\in S _ { t - 1 } \\colon \\tau _ { j } \\gets \\left( v _ { t } ^ { \\top } \\left( x _ { j } - \\mathrm { M e a n } ( S _ { t - 1 } ) \\right) \\right) ^ { 2 } \\ ;$   \n3. Draw a random threshold: $Z _ { t } \\gets \\mathrm { U n i f } ( [ 0 , 1 ] )$ ;   \n4. Remove outliers from $S _ { t - 1 }$ defined as $\\{ i \\in S _ { t - 1 } : \\tau _ { i }$ is in the largest $2 \\alpha$ -tail of $\\{ \\tau _ { j } \\} _ { j \\in S _ { t - 1 } }$ and $\\tau _ { i } \\geq Z _ { t } \\tau _ { \\operatorname* { m a x } } \\}$ , where $\\tau _ { \\mathrm { m a x } } = \\operatorname* { m a x } _ { j \\in S _ { t - 1 } } \\tau _ { j }$ ",
+        "bbox": [
+            210,
+            549,
+            825,
+            650
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "This is repeated until the empirical covariance is sufficiently small and the empirical mean $\\hat { \\mu }$ is output. At a high level, the correctness of this algorithm relies on the key observation that the $\\alpha$ -fraction of adversarial corruption can not significantly change the mean of the dataset without introducing large eigenvalues in the empirical covariance. Therefore, the algorithm finds top eigenvector of the empirical covariance in step 1, and tries to correct the empirical covariance by removing corrupted data points. Each data point is assigned a score in step 2 which indicates the “badness” of the data points, and a threshold $Z _ { t }$ in step 3 is carefully designed such that step 4 guarantees to remove more corrupted data points than good data points (in expectation). This guarantees the following bound achieving the near-optimal sample complexity shown in the second row of Table 1. A formal description of this algorithm is in Algorithm 4 in Appendix C. ",
+        "bbox": [
+            173,
+            657,
+            825,
+            796
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "Proposition 2.1 (Corollary of [63, Theorem 2.1]). Under assumption $^ { l }$ , the above filtering algorithm achieves accuracy $\\| \\hat { \\mu } - \\mu \\| _ { 2 } \\le O ( \\alpha \\sqrt { \\log ( 1 / \\alpha ) } ) \\ w . p . \\ 0 . 9 i f n \\ge \\widetilde { \\Omega } ( d / \\alpha ^ { 2 } ) \\ / .$ . ",
+        "bbox": [
+            171,
+            799,
+            823,
+            833
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "Challenges in making robust mean estimation private. To get a DP and robust mean, a naive attempt is to apply a standard output perturbation mechanism to $\\hat { \\mu }$ . However, this is obviously challenging since the end-to-end sensitivity is intractable. The standard recipe to circumvent this is to make the current “state” $S _ { t }$ private at every iteration. Once $S _ { t - 1 }$ is private (hence, public knowledge), making the next “state” $S _ { t }$ private is simpler. We only need to analyze the sensitivity of a single step and apply some output perturbation mechanism with $\\left( \\varepsilon _ { t } , \\delta _ { t } \\right)$ . End-to-end privacy is guaranteed by accounting for all these $\\left( \\varepsilon _ { t } , \\delta _ { t } \\right)$ ’s using the advanced composition [51]. This recipe has been quite successful, for example, in training neural networks with (stochastic) gradient descent [1], where the current state can be the optimization variable $\\mathbf { x } _ { t }$ . However, for the above (non-private) filtering algorithm, this standard recipe fails, since the state $S _ { t }$ is a set and has large sensitivity. Changing a single data point in $S _ { t }$ can significantly alter which (and how many) samples are filtered out. ",
+        "bbox": [
+            174,
+            842,
+            825,
+            911
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "",
+        "bbox": [
+            174,
+            90,
+            825,
+            175
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "text",
+        "text": "2.2 A new framework for private iterative filtering ",
+        "text_level": 1,
+        "bbox": [
+            174,
+            196,
+            534,
+            212
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "text",
+        "text": "Instead of making the (highly sensitive) $S _ { t }$ itself private, we propose a new framework which makes private only the statistics of $S _ { t }$ : the mean $\\mu _ { t }$ and the top principal direction $v _ { t }$ . There are two versions of this algorithm, which output the exactly same $\\hat { \\mu }$ with the exactly same privacy guarantees, but are written from two different perspectives. We present here the interactive version from the perspective of an analyst accessing the dataset via DP queries $\\cdot q _ { \\mathrm { r a n g e } }$ , $q _ { \\mathrm { s i z e } }$ , $q _ { \\mathrm { m e a n } }$ , $q _ { \\mathrm { n o r m } }$ and $q _ { \\mathrm { P C A } } ,$ ), because this version makes clear the inner operations of each private mechanisms, hence making $( i )$ the sensitivity analysis transparent, $( i i )$ checking the correctness of privacy guarantees easy, and $( i i i )$ tracking privacy accountant simple. In practice, one should implement the centralized version (Algorithm 7 in Appendix D), which is significantly more efficient. ",
+        "bbox": [
+            173,
+            223,
+            825,
+            348
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "text",
+        "text": "Algorithm 1: Private iterative filtering (interactive version) ",
+        "text_level": 1,
+        "bbox": [
+            173,
+            373,
+            563,
+            387
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "text",
+        "text": "Input: $S = \\{ x _ { i } \\} _ { i \\in [ n ] }$ , $\\alpha \\in ( 0 , 1 / 2 )$ , probability $\\zeta \\in ( 0 , 1 )$ , # of iterations $T = \\Theta ( d ) , ( \\varepsilon , \\delta )$   \n1 $( \\bar { x } , B ) \\gets q _ { \\mathrm { r a n g e } } ( S , 0 . 0 1 \\varepsilon , 0 . 0 1 \\delta )$   \n2 $\\varepsilon _ { 1 } \\gets \\operatorname* { m i n } \\{ 0 . 9 9 \\varepsilon , 0 . 9 \\} / ( 4 \\sqrt { 2 T \\log ( 2 / \\delta ) } )$ ), δ1 ← 0.99δ/(8T )   \n3 if $n < ( 4 / \\varepsilon _ { 1 } ) \\log ( 1 / ( 2 \\delta _ { 1 } ) )$ then Output: $\\varnothing$   \n4 for $t = 1 , \\dots , T$ do   \n5 $n _ { t }  q _ { \\mathrm { s i z e } } ( \\{ ( \\mu _ { \\ell } , v _ { \\ell } , Z _ { \\ell } ) \\} _ { \\ell \\in [ t - 1 ] } , \\varepsilon _ { 1 } , \\bar { x } , B )$ , if $n _ { t } < 3 n / 4$ then Output: $\\varnothing$   \n6 $\\mu _ { t }  q _ { \\mathrm { m e a n } } ( \\{ ( \\mu _ { \\ell } , v _ { \\ell } , Z _ { \\ell } ) \\} _ { \\ell \\in [ t - 1 ] } , \\varepsilon _ { 1 } , \\bar { x } , B )$   \n7 $\\lambda _ { t } \\gets q _ { \\mathrm { n o r m } } ( \\{ ( \\mu _ { \\ell } , v _ { \\ell } , Z _ { \\ell } ) \\} _ { \\ell \\in [ t - 1 ] } , \\mu _ { t } , \\varepsilon _ { 1 } , \\bar { x } , B )$   \n8 if $\\lambda _ { t } \\le ( C - 0 . 0 1 ) \\alpha \\log { 1 / \\alpha }$ then Output: $\\mu _ { t }$   \n9 $\\boldsymbol { v } _ { t }  q _ { \\mathrm { P C A } } ( \\{ ( \\mu _ { \\ell } , v _ { \\ell } , Z _ { \\ell } ) \\} _ { \\ell \\in [ t - 1 ] } , \\mu _ { t } , \\varepsilon _ { 1 } , \\delta _ { 1 } , \\bar { x } , B ) )$   \n10 $Z _ { t } \\gets \\mathrm { U n i f } ( [ 0 , 1 ] )$ ",
+        "bbox": [
+            160,
+            390,
+            776,
+            558
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "text",
+        "text": "Output: $\\mu _ { t }$ ",
+        "text_level": 1,
+        "bbox": [
+            174,
+            559,
+            253,
+            573
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "text",
+        "text": "We give a high-level explanation of each step of Algorithm 1 here and give the formal definitions of all the queries in Appendix D. First, $q _ { \\mathrm { r a n g e } }$ returns (the parameters of) a hypercube $\\bar { x } + [ - B / 2 , B / 2 ] ^ { d }$ that is guaranteed to include all uncorrupted samples while preserving privacy. This is achieved by running $d$ coordinate-wise private histograms and selecting $\\bar { x } _ { j }$ as the center of the largest bin for the $j$ -th coordinate. Since covariance is I, $q _ { \\mathrm { r a n g e } }$ returns a fixed $B = 8 \\sigma \\sqrt { \\log ( d n / \\zeta ) }$ . Such an adaptive estimate of the support is critical in tightly bounding the sensitivity of all subsequent queries, which operate on the clipped dataset; all data points are projected as $\\mathcal { P } _ { \\bar { x } + [ - B / 2 , B / 2 ] ^ { d } } ( x ) =$ arg $\\begin{array} { r } { \\operatorname* { m i n } _ { y \\in \\bar { x } + [ - B / 2 , B / 2 ] ^ { c } } } \\end{array}$ $\\lVert y - x \\rVert _ { 2 }$ in all the queries that follow. With clipping, a single data point can now change at most by $B \\sqrt { d }$ . ",
+        "bbox": [
+            173,
+            598,
+            825,
+            732
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "text",
+        "text": "The subsequent steps perform the non-private filtering algorithm of $\\ S 2 . 1$ , but with private statistics $\\mu _ { t }$ and $v _ { t }$ . As the set $S _ { t }$ changes over time, we lower bound its size (which we choose to be $| S _ { t } | > n / 2 )$ ) to upper bound the sensitivity of other queries $q _ { \\mathrm { m e a n } } , q _ { \\mathrm { n o r m } }$ and $q _ { \\mathrm { P C A } }$ . ",
+        "bbox": [
+            174,
+            738,
+            825,
+            781
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "text",
+        "text": "At the $t$ -th iterations, every time a query is called the data curator $( i )$ uses $( { \\bar { x } } , B )$ to clip the data, $( i i )$ computes $S _ { t }$ by running $t - 1$ steps of the non-private filtering algorithm of $\\ S 2 . 1$ but with a given fixed set of parameters $\\{ ( \\mu _ { \\ell } , v _ { \\ell } ) \\} _ { \\ell \\in [ t - 1 ] }$ (and the given randomness $\\{ Z _ { \\ell } \\} _ { \\ell \\in [ t - 1 ] } )$ , and $( i i i )$ computes the queried private statistics of $S _ { t }$ . If the private spectral norm of the covariance of $S _ { t }$ (i.e., $\\lambda _ { t } ,$ ) is sufficiently small, we output the private and robust mean $\\hat { \\mu } = \\mu _ { t }$ (line 8). Otherwise, we compute the private top PCA direction $v _ { t }$ and draw an randomness $Z _ { t }$ to be used in the next step of filtering, as in the non-private filtering algorithm. We emphasize that $\\{ S _ { \\ell } \\}$ are not private, and hence never returned to the analyst. We also note that this interactive version is redundant as every query is re-computing $S _ { t }$ . In our setting, the analyst has the dataset and there is no need to separate them. This leads to a centralized version we provide in Algorithm 7 in the appendix, which avoids redundant computations and hence is significantly more efficient. ",
+        "bbox": [
+            173,
+            786,
+            825,
+            911
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "text",
+        "text": "",
+        "bbox": [
+            173,
+            90,
+            823,
+            119
+        ],
+        "page_idx": 5
+    },
+    {
+        "type": "text",
+        "text": "The main challenge in this framework is the privacy analysis. Because $\\{ S _ { \\ell } \\} _ { \\ell \\in [ t - 1 ] }$ is not private, each query runs $t - 1$ steps of filtering whose end-to-end sensitivity could blow-up. Algorithmically, $( i )$ we start with a specific choice of a non-private iterative filtering algorithm (among several variations that are equivalent in non-private setting but widely differ in its sensitivity), and $( i i )$ make appropriate changes in the private queries (Algorithm 1) to keep the sensitivity small. Analytically, the following key technical lemma allows a sharp analysis of the end-to-end sensitivity of iterative filtering. ",
+        "bbox": [
+            173,
+            126,
+            825,
+            210
+        ],
+        "page_idx": 5
+    },
+    {
+        "type": "text",
+        "text": "Lemma 2.2. Let $S _ { t } ( S )$ denote the resulting subset of samples after $t$ iterations of the filtering in the queries $( q _ { \\mathrm { s i z e } }$ , $q _ { \\mathrm { m e a n } }$ , $q _ { \\mathrm { n o r m } }$ , and $q _ { \\mathrm { P C A } } )$ are applied to a dataset $s$ using fixed parameters $\\{ ( \\mu _ { \\ell } , v _ { \\ell } , Z _ { \\ell } ) \\} _ { \\ell = 1 } ^ { t }$ . Then, we have $d _ { \\triangle } ( S _ { t } ( S ) , S _ { t } ( S ^ { \\prime } ) ) \\leq d _ { \\triangle } ( S , S ^ { \\prime } )$ , where $d _ { \\triangle } ( S , S ^ { \\prime } ) \\triangleq \\operatorname* { m a x } \\{ | S \\ \\backslash$ $S ^ { \\prime } | , | S ^ { \\prime } \\backslash S | \\}$ . ",
+        "bbox": [
+            173,
+            213,
+            825,
+            273
+        ],
+        "page_idx": 5
+    },
+    {
+        "type": "text",
+        "text": "Recall that two datasets are neighboring, i.e., $\\boldsymbol { S } \\sim \\boldsymbol { S } ^ { \\prime }$ , iff $d _ { \\triangle } ( S , S ^ { \\prime } ) \\leq 1$ . This lemma implies that if two datasets are neighboring, then they are still neighboring after filtering with the same parameters, no matter how many times we filter them. Hence, this lemma allows us to use the standard outputperturbation mechanisms with $( \\varepsilon _ { 1 } , \\delta _ { 1 } )$ -DP. Advanced composition ensures that end-to-end guarantee of $4 T$ such queries is $( 0 . 9 9 \\varepsilon , 0 . 9 9 \\delta )$ -DP. Together with $( 0 . 0 1 \\varepsilon , 0 . 0 1 \\delta )$ -DP budget used in $q _ { \\mathrm { r a n g e } }$ , this satisfied the target privacy. Analyzing the utility of this algorithm, we get the following guarantee. ",
+        "bbox": [
+            173,
+            281,
+            825,
+            367
+        ],
+        "page_idx": 5
+    },
+    {
+        "type": "text",
+        "text": "Theorem 5. Algorithm $^ { l }$ is $( \\varepsilon , \\delta )$ -DP. Under Assumption $^ { l }$ , there exists a universal constant $c \\in$ $( 0 , 0 . 1 )$ such that if $\\alpha \\leq c$ and $n = \\widetilde \\Omega \\left( ( d / \\alpha ^ { 2 } ) + d ^ { 2 } ( \\log ( 1 / \\delta ) ) ^ { 3 / 2 } / ( \\varepsilon \\alpha ) \\right)$ then Algorithm 1 achieves $\\| \\hat { \\mu } - \\mu \\| _ { 2 } \\leq O ( \\alpha \\sqrt { \\log ( 1 / \\alpha ) } )$ with probability 0.9. ",
+        "bbox": [
+            173,
+            369,
+            825,
+            420
+        ],
+        "page_idx": 5
+    },
+    {
+        "type": "text",
+        "text": "The first term $O ( d / \\alpha ^ { 2 } )$ in the sample complexity is optimal (cf. Table 1), but there is a factor of $d$ gap in the second term. This is due to the fact that we need to run $O ( d )$ iterations in the worst-case. Such numerous accesses to the database result in large noise to be added at each iteration, requiring large sample size to combat that extra noise. We introduce PRIME to reduce the number of iterations to ${ \\bar { O } } ( ( \\log { d } ) ^ { 2 } )$ and significantly reduce the sample complexity. ",
+        "bbox": [
+            173,
+            429,
+            825,
+            501
+        ],
+        "page_idx": 5
+    },
+    {
+        "type": "text",
+        "text": "2.3 PRIME: novel robust and private mean estimator ",
+        "text_level": 1,
+        "bbox": [
+            173,
+            515,
+            558,
+            531
+        ],
+        "page_idx": 5
+    },
+    {
+        "type": "text",
+        "text": "Algorithm 1 (specifically Filter(·) in Algorithm 1) accesses the database $O ( d )$ times. This is necessary for two reasons. First, the filter checks only one direction $v _ { t }$ at each iteration. In the worst case, the corrupted samples can be scattered in $\\Omega ( d )$ orthogonal directions such that the filter needs to be repeated $O ( d )$ times. Secondly, even if the corrupted samples are clustered together in one direction, the filter still needs to be repeated $O ( d )$ times. This is because we had to use a large (random) threshold of $d B ^ { 2 } Z _ { t } = { \\cal { O } } ( d )$ to make the threshold data-independent so that we can keep the sensitivity of Filter(·) low, which results in slow progress. We propose filtering multiple directions simultaneously using a new score $\\{ \\tau _ { i } \\}$ based on the matrix multiplicative weights. Central to this approach is a novel adaptive filtering algorithm DPTHRESHOLD $( \\cdot )$ that guarantees sufficient decrease in the total score at every iteration. ",
+        "bbox": [
+            173,
+            540,
+            826,
+            680
+        ],
+        "page_idx": 5
+    },
+    {
+        "type": "text",
+        "text": "2.3.1 Matrix Multiplicative Weight (MMW) scoring ",
+        "text_level": 1,
+        "bbox": [
+            173,
+            693,
+            547,
+            709
+        ],
+        "page_idx": 5
+    },
+    {
+        "type": "text",
+        "text": "The MMW-based approach, pioneered in [36] for non-private robust mean estimation, filters out multiple directions simultaneously. It runs over $O ( \\log d )$ epochs and every epoch consists of $O ( \\log d )$ iterations. At every epoch $s$ and iteration $t$ , step 2 of the iterative filtering in $\\ S 2 . 1$ is replaced by a new score $\\tau _ { i } = ( x _ { i } - \\mathrm { M e a n } ( S _ { t } ^ { ( s ) } ) ) ^ { T } U _ { t } ^ { ( s ) } ( x _ { i } - \\mathrm { M e a n } ( S _ { t } ^ { ( s ) } ) )$ where $U _ { t } ^ { ( s ) }$ now accounts for all directions in $\\mathbb { R } ^ { d }$ but appropriately weighted. Precisely, it is defined via the matrix multiplicative update: ",
+        "bbox": [
+            174,
+            717,
+            826,
+            792
+        ],
+        "page_idx": 5
+    },
+    {
+        "type": "equation",
+        "img_path": "images/53bb8d5442b2b4afe0bd0508483ccca717b6e5178e5593b665b58a74b0e0e679.jpg",
+        "text": "$$\nU _ { t } ^ { ( s ) } = \\frac { \\exp \\left( \\alpha ^ { ( s ) } \\sum _ { r \\in [ t ] } ( \\mathrm { C o v } ( S _ { r } ^ { ( s ) } ) - { \\bf I } ) \\right) } { \\mathrm { T r } \\big ( \\exp ( \\alpha ^ { ( s ) } \\sum _ { r \\in [ t ] } ( \\mathrm { C o v } ( S _ { r } ^ { ( s ) } ) - { \\bf I } ) ) \\big ) } ,\n$$",
+        "text_format": "latex",
+        "bbox": [
+            312,
+            797,
+            651,
+            847
+        ],
+        "page_idx": 5
+    },
+    {
+        "type": "text",
+        "text": "for some choice of $\\alpha ^ { ( s ) } > 0$ . If we set the number of iterations to one, a choice of $\\alpha ^ { ( s ) } = \\infty$ recovers the previous score that relied on the top singular vector from $\\ S 2 . 1$ and a choice of $\\alpha ^ { ( s ) } = 0$ gives a simple norm based score $\\tau _ { i } = \\| x _ { i } \\| _ { 2 } ^ { 2 }$ . An appropriate choice of $\\alpha ^ { ( s ) }$ smoothly interpolates between these two extremes, which ensures that $O ( \\log d )$ iterations are sufficient for the spectral norm of the covariance to decrease strictly by a constant factor. This guarantees that after $O ( \\log d )$ epochs, we sufficiently decrease the covariance to ensure that the empirical mean is accurate enough. Critical in achieving this gain is our carefully designed filtering algorithm DPTHRESHOLD that uses the privately computed MMW-based scores using Gaussian mechanism on the covariance matrices as shown in Algorithm 11 in Appendix E. ",
+        "bbox": [
+            174,
+            853,
+            825,
+            911
+        ],
+        "page_idx": 5
+    },
+    {
+        "type": "text",
+        "text": "",
+        "bbox": [
+            174,
+            90,
+            825,
+            161
+        ],
+        "page_idx": 6
+    },
+    {
+        "type": "text",
+        "text": "2.3.2 Adaptive filtering with DPTHRESHOLD ",
+        "text_level": 1,
+        "bbox": [
+            173,
+            175,
+            495,
+            190
+        ],
+        "page_idx": 6
+    },
+    {
+        "type": "text",
+        "text": "Novelty. The corresponding non-private filtering of [36, Algorithm 9] for robust mean estimation takes advantage of an adaptive threshold, but filters out each sample independently resulting in a prohibitively large sensitivity; the coupling between each sample and the randomness used to filter it can change widely between two neighboring datasets. On the other hand, Algorithm 1 (i.e., Filter(·) in Algorithm 6) takes advantage of jointly filtering all points above a single threshold $B ^ { 2 } d Z _ { t }$ with a single randomness $Z _ { t } \\sim \\mathrm { U n i f } [ 0 , 1 ]$ , but the non-adaptive (and hence large) choice of the range $B ^ { 2 } d$ results in a large number of iterations because each filtering only decrease the score by little. To sufficiently reduce the total score while maintaining a small sensitivity, we introduce a filter with a single and adaptive threshold. ",
+        "bbox": [
+            173,
+            198,
+            825,
+            324
+        ],
+        "page_idx": 6
+    },
+    {
+        "type": "text",
+        "text": "Algorithm. Our goal here is to privately find a single scalar $\\rho$ such that when a randomized filter is applied on the scores $\\{ \\tau _ { i } \\}$ with a (random) threshold $\\rho Z$ (with $Z$ drawn uniform in $[ 0 , 1 ] )$ , we filter out enough samples to make progress in each iteration while ensuring that we do not remove too many uncorrupted samples. This is a slight generalization of the non-private algorithm in Section 2.1, which simply set $\\rho = \\operatorname* { m a x } _ { j \\in S _ { t } } \\tau _ { j }$ . While this guarantees the filter removes more corrupted samples than good samples, it does not make sufficient progress in reducing the total score of the samples. ",
+        "bbox": [
+            173,
+            329,
+            825,
+            414
+        ],
+        "page_idx": 6
+    },
+    {
+        "type": "text",
+        "text": "Ideally, we want the thresholding to decrease the total score by a constant multiplicative factor, which will in the end allow the algorithm to terminate within logarithmic iterations. To this end, we propose a new scheme of using the largest $\\rho$ such that the following inequality holds: ",
+        "bbox": [
+            174,
+            420,
+            825,
+            462
+        ],
+        "page_idx": 6
+    },
+    {
+        "type": "equation",
+        "img_path": "images/6e717cdd2620c553eeae0f0bac341635344a945d930dd30b1b3e271d36a171e1.jpg",
+        "text": "$$\n\\sum _ { \\tau _ { i } > \\rho } ( \\tau _ { i } - \\rho ) \\geq 0 . 3 1 \\sum _ { \\tau _ { i } \\in S _ { t } } ( \\tau _ { i } - 1 ) .\n$$",
+        "text_format": "latex",
+        "bbox": [
+            364,
+            467,
+            599,
+            502
+        ],
+        "page_idx": 6
+    },
+    {
+        "type": "text",
+        "text": "We use a private histogram of the scores to approximate this threshold. Similar to [55, 58], we use geometrically increasing bin sizes such that we use only $O ( \\log B ^ { 2 } d )$ bins while achieving a preferred multiplicative error in our quantization. At each epoch $s$ and iteration $t$ , we run DPTHRESHOLD sketched in the following to approximate $\\rho$ followed by a random filter. Step 3 replaces the non-private condition in Eq. (1). A complete description is provided in Algorithm 11. ",
+        "bbox": [
+            174,
+            507,
+            825,
+            579
+        ],
+        "page_idx": 6
+    },
+    {
+        "type": "text",
+        "text": "1. Privately compute scores for all data points $i \\in S _ { t } ^ { ( s ) } : \\tau _ { i } \\gets ( x _ { i } - \\mu _ { t } ) ^ { \\top } U _ { t } ^ { ( s ) } ( x _ { i } - \\mu _ { t } ) ;$ ;   \n2. Compute a private histogram $\\{ \\tilde { h } _ { j } \\} _ { j = 1 } ^ { 2 + \\log ( B ^ { 2 } d ) }$ t i  i  t  t i  t of the scores over geometrically sized bins   \n$I _ { 1 } = [ 1 / 4 , 1 / 2 )$ , $I _ { 2 } = [ 1 / 2 , 1 ) , \\ldots , \\bar { I } _ { 2 + \\log ( B ^ { 2 } d ) } = [ 2 ^ { \\log ( B ^ { 2 } d ) - 1 } , 2 ^ { \\log ( B ^ { 2 } d ) } ]$ ;   \n3. Privately find the largest $\\ell$ satisfying $\\begin{array} { r } { \\sum _ { j \\ge \\ell } ( 2 ^ { j } - 2 ^ { \\ell } ) \\tilde { h } _ { j } \\ge 0 . 3 1 \\sum _ { i \\in S _ { t } ^ { ( s ) } } ( \\tau _ { i } - 1 ) ; } \\end{array}$ ;   \n4. Output $\\rho = 2 ^ { \\ell }$ . ",
+        "bbox": [
+            210,
+            587,
+            826,
+            676
+        ],
+        "page_idx": 6
+    },
+    {
+        "type": "text",
+        "text": "3 Analyses of PRIME ",
+        "text_level": 1,
+        "bbox": [
+            174,
+            693,
+            370,
+            710
+        ],
+        "page_idx": 6
+    },
+    {
+        "type": "text",
+        "text": "Building on the framework of Algorithm 1, PRIME (Algorithm 9) replaces the score with the MMWbased score presented in $\\ S 2 . 3 . 1$ and the filter with the adaptive DPTHRESHOLD. This reduces the number of iterations to $T = O ( ( \\log d ) ^ { 2 } )$ achieving the following bound. ",
+        "bbox": [
+            174,
+            724,
+            826,
+            767
+        ],
+        "page_idx": 6
+    },
+    {
+        "type": "text",
+        "text": "Theorem 6. PRIME is $( \\varepsilon , \\delta )$ -differentially private. Under Assumption $I$ there exists a universal constant $c \\in ( 0 , 0 . 1 )$ such that if $\\alpha \\leq c$ and $n = \\widetilde \\Omega ( ( d / \\alpha ^ { 2 } ) + ( d ^ { 3 / 2 } / ( \\varepsilon \\alpha ) ) \\log ( 1 / \\delta ) )$ , then PRIME achieves $\\| \\hat { \\mu } - \\mu \\| _ { 2 } = O ( \\alpha \\sqrt { \\log ( 1 / \\alpha ) } )$ with probability 0.9. ",
+        "bbox": [
+            174,
+            771,
+            825,
+            820
+        ],
+        "page_idx": 6
+    },
+    {
+        "type": "text",
+        "text": "A proof is provided in Appendix F. The notation $\\widetilde { \\Omega } ( \\cdot )$ hides logarithmic terms in $d , R$ , and $1 / \\alpha$ . To achieve an error of $O ( \\alpha \\sqrt { \\log ( 1 / \\alpha ) } )$ , the first term $\\widetilde \\Omega ( d / \\alpha ^ { 2 } \\log ( 1 / \\alpha ) )$ is necessary even if there is no corruption. The accuracy of $O ( \\alpha \\sqrt { \\log ( 1 / \\alpha ) } )$ matches the lower bound shown in [33] for any polynomial time statistical query algorithm, and it nearly matches the information theoretical lower bound on robust estimation of $\\Omega ( \\alpha )$ . On the other hand, the second term of $\\widetilde \\Omega ( d ^ { 3 / 2 } / ( \\varepsilon \\alpha \\log ( 1 / \\alpha ) ) )$ ",
+        "bbox": [
+            174,
+            830,
+            825,
+            912
+        ],
+        "page_idx": 6
+    },
+    {
+        "type": "text",
+        "text": "has an extra factor of $d ^ { 1 / 2 }$ compared to the optimal one achieved by exponential time Algorithm 2. It is an open question if this gap can be closed by a polynomial time algorithm. ",
+        "bbox": [
+            171,
+            90,
+            823,
+            119
+        ],
+        "page_idx": 7
+    },
+    {
+        "type": "text",
+        "text": "The bottleneck is the private matrix multiplicative weights. Such spectral analyses are crucial in filter-based robust estimators. Even for a special case of privately computing the top principal component, the best polynomial time algorithm requires $\\bar { O } ( d ^ { 3 / 2 } )$ samples [39, 18, 78], and this sample complexity is also necessary as shown in [39, Corollary 25]. ",
+        "bbox": [
+            174,
+            126,
+            825,
+            183
+        ],
+        "page_idx": 7
+    },
+    {
+        "type": "text",
+        "text": "To boost the success probability to $1 - \\zeta$ for some small $\\zeta > 0$ , we need an extra $\\log ( 1 / \\zeta )$ factor in the sample complexity to make sure the dataset satisfies the regularity condition with probability $\\zeta / 2$ . Then we can run PRIME $\\log ( 1 / \\zeta )$ times and choose the output of a run that satisfies $n ^ { ( s ) } > n ( 1 { - } 1 0 \\alpha )$ and $\\lambda ^ { ( s ) } \\le C \\alpha \\log ( 1 / \\alpha )$ at termination. ",
+        "bbox": [
+            174,
+            189,
+            826,
+            250
+        ],
+        "page_idx": 7
+    },
+    {
+        "type": "image",
+        "img_path": "images/46a53c55cd9bcc36cd60657375546149f2ff820d7058a98083ffc81ea44ac394.jpg",
+        "image_caption": [
+            "Figure 1: Private mean estimators (e.g., DP mean [52]) are vulnerable to adversarial corruption especially in high dimensions, while the proposed PRIME achieves robustness (and privacy) regardless of the dimension of the samples. "
+        ],
+        "image_footnote": [],
+        "bbox": [
+            189,
+            266,
+            826,
+            363
+        ],
+        "page_idx": 7
+    },
+    {
+        "type": "text",
+        "text": "Numerical experiments support our theoretical claims. The left figure with $( \\alpha , \\varepsilon , \\delta , n ) \\ =$ $( 0 . 0 5 , 2 0 , 0 . 0 1 , 1 0 ^ { 6 } )$ is in the large $\\alpha$ regime where the DP Mean error is dominates by $\\alpha \\sqrt { d }$ and PRIME error by $\\alpha \\sqrt { \\log ( 1 / \\alpha ) }$ . Hence, PRIME error is constant whereas DP Mean error increases with the dimension $d$ . The second figure with √ $( \\alpha , \\varepsilon , \\delta , n ) = ( 0 . 0 0 1 , 2 0 , 0 . 0 1 , 1 0 ^ { 6 } )$ is in the small $\\alpha$ regime when DP Mean error consists of $\\alpha \\sqrt { d } + \\sqrt { d / n }$ and PRIME is dominated by $\\sqrt { d / n }$ . Both increase with the dimension $d$ , and the gap can be made large by increasing $\\alpha$ . The right figure√ with $( \\alpha , \\delta , d , n ) = ( 0 . 1 , 0 . 0 1 , 1 0 , 1 0 ^ { 6 } )$ is when DP Mean error is dominated by $\\alpha \\sqrt { d }$ and PRIME by $\\alpha \\sqrt { \\log ( 1 / \\alpha ) }$ when $\\varepsilon > c d ^ { 1 . 5 } / ( \\alpha n )$ . Below this threshold, which happens in this example around $\\varepsilon = 0 . 0 5$ , the added noise in the private mechanism starts to dominate with decreasing $\\varepsilon$ . Both algorithms have respective thresholds below which the error increases with decreasing $\\varepsilon$ . This threshold is larger for PRIME because it uses the privacy budget to perform multiple operations and hence the noise added to the final output is larger compared to DP Mean. Below this threshold, which can be easily determined based on the known parameters $( \\varepsilon , \\delta , n , \\alpha )$ , we should either collect more data (which will decrease the threshold) or give up filtering and spend all privacy budget on $q _ { \\mathrm { r a n g e } }$ and the empirical mean (which will reduce the error). Details of the experiments are in Appendix L. ",
+        "bbox": [
+            173,
+            430,
+            825,
+            650
+        ],
+        "page_idx": 7
+    },
+    {
+        "type": "text",
+        "text": "4 Exponential time algorithm with near-optimal sample complexity ",
+        "text_level": 1,
+        "bbox": [
+            171,
+            674,
+            750,
+            693
+        ],
+        "page_idx": 7
+    },
+    {
+        "type": "text",
+        "text": "Novelty. An existing exponential time algorithm for robust and private mean estimation in [14] strictly requires the uncorrupted samples to be drawn from a Gaussian distribution. We also provide a similar algorithm based on private Tukey median in Appendix I and its analysis in Appendix J. In this section, we introduce a novel estimator that achieves near-optimal guarantees for more general sub-Gaussian distributions (and also covariance bounded distributions) but takes an exponential run-time. Its innovation is in leveraging on the resilience property of well-behaved distributions not only to estimate the mean robustly (which is the standard use of the property) but also to adaptively bound the sensitivity of the estimator, thus achieving optimal privacy-accuracy tradeoff. ",
+        "bbox": [
+            173,
+            708,
+            825,
+            820
+        ],
+        "page_idx": 7
+    },
+    {
+        "type": "text",
+        "text": "Definition 4.1 (Resilience from Definition 1 in [73]). $A$ set of points $\\{ x _ { i } \\} _ { i \\in S }$ lying in $\\mathbb { R } ^ { d }$ is $( \\sigma , \\alpha )$ - resilient around a point $\\mu$ $\\begin{array} { r } { . i f \\| ( 1 / | T | ) \\sum _ { i \\in T } ( x _ { i } - \\mu ) \\| _ { 2 } \\leq \\sigma } \\end{array}$ for all subsets $T \\subset S$ of size $( 1 - \\alpha ) | S |$ . ",
+        "bbox": [
+            171,
+            825,
+            825,
+            857
+        ],
+        "page_idx": 7
+    },
+    {
+        "type": "text",
+        "text": "Algorithm. As data is corrupted, we define $R ( S )$ as a surrogate for resilience of the uncorrupted part of the set. If $S$ indeed consists of a $1 - \\alpha$ fraction of independent samples from the promised class of distributions, the goodness score $R ( S )$ will be close to the resilience property of the good data. ",
+        "bbox": [
+            174,
+            868,
+            825,
+            911
+        ],
+        "page_idx": 7
+    },
+    {
+        "type": "text",
+        "text": "Definition 4.2 (Goodness of a set). For $\\mu ( S ) = ( 1 / | S | ) \\sum _ { i \\in S } x _ { i }$ , let us define ",
+        "bbox": [
+            171,
+            89,
+            691,
+            108
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "equation",
+        "img_path": "images/a63fff882c56dd4d22cd7a5e3f73c597dc62e918ef47ae2bcdaf9517671e805d.jpg",
+        "text": "$$\nR ( S ) \\stackrel { \\Delta } { = } \\operatorname* { m i n } _ { \\substack { S ^ { \\prime } \\subset S , | S ^ { \\prime } | = ( 1 - 2 \\alpha ) | S | . } } \\operatorname* { m a x } _ { \\substack { T \\subset S ^ { \\prime } , | T | = ( 1 - \\alpha ) | S ^ { \\prime } | . } } \\| \\mu ( T ) - \\mu ( S ^ { \\prime } ) \\| _ { 2 } .\n$$",
+        "text_format": "latex",
+        "bbox": [
+            261,
+            113,
+            738,
+            140
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "Algorithm 2 first checks if the resilience matches that of the promised distribution. The data is pre-processed with $q _ { \\mathrm { r a n g e } }$ to ensure we can check $R ( S )$ privately. Once resilience is cleared, we can safely use the exponential mechanism based on the score function $d ( \\hat { \\mu } , S )$ in Definition 4.3 to select an approximate robust mean $\\hat { \\mu }$ privately. The choice of the sensitivity critically relies on the fact that resilient datasets have small sensitivity of ${ \\cal O } ( ( 1 / n ) \\sqrt { \\log ( 1 / \\alpha ) } )$ . Without the resilience check, the sensitivity is $O ( d ^ { 1 / 2 } / n )$ resulting in an extra factor of $\\sqrt { d }$ in the sample complexity. ",
+        "bbox": [
+            173,
+            161,
+            826,
+            252
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "Input: $S = \\{ x _ { i } \\} _ { i \\in [ n ] } , \\alpha \\in ( 0 , 1 / 2 ) , ( \\varepsilon , \\delta )$   \n1 if $n < c d ^ { 1 / 2 } \\log ( 1 / \\delta ) / \\left( \\varepsilon \\alpha \\sqrt { \\log ( 1 / \\alpha ) } \\right)$ then Output: $\\emptyset \\ [ \\ c d ^ { 1 / 2 } \\log ( 1 / \\delta ) / \\ ( \\varepsilon \\alpha )$ for hevay-tail]   \n2 $( \\bar { x } , B ) \\gets q _ { \\mathrm { r a n g e } } ( S , ( 1 / 3 ) \\varepsilon , ( 1 / 3 ) \\delta )$ $[ \\ q _ { \\mathrm { r a n g e - h t } } ( \\cdot )$ for hevay-tail]   \n3 Project the data points onto the ball: $x _ { i }  \\mathcal { P } _ { B _ { \\sqrt { d } B / 2 } ( \\bar { x } ) } ( x _ { i } )$ , for all $i \\in [ n ]$   \n4 $\\widehat { R } ( S ) \\gets R ( S ) + \\mathrm { L a p } ( 3 B d ^ { 1 / 2 } / ( n \\varepsilon ) )$   \n5 if $\\widehat { R } ( S ) > 2 \\alpha \\sqrt { \\log ( 1 / \\alpha ) }$ then Output: $\\varnothing$ $[ \\widehat { R } ( S ) > 2 c _ { \\zeta } \\sqrt { \\alpha }$ for hevay-tail]   \n6 else Output: a randomly drawn point $\\hat { \\mu } \\in \\mathcal { B } _ { \\sqrt { d } B / 2 } ( \\bar { x } )$ sampled from a density   \n7 r(ˆµ) ∝ e−(1/(24 log(1/α)))ε n d(ˆµ,S) $[ e ^ { - ( \\varepsilon n \\sqrt { \\alpha } / ( 2 4 c _ { \\zeta } ) ) d ( \\hat { \\mu } , S ) }$ for heavy-tail] ",
+        "bbox": [
+            158,
+            287,
+            815,
+            425
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "We propose the score function $d ( \\hat { \\mu } , S )$ in the following definition, which is a robust estimator of the distance between the mean and the candidate $\\hat { \\mu }$ . ",
+        "bbox": [
+            171,
+            443,
+            823,
+            472
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "Definition 4.3. For a set of data $\\{ x _ { i } \\} _ { i \\in S }$ lying in $\\mathbb { R } ^ { d }$ , for any $v \\in \\mathbb { S } ^ { d - 1 }$ , define $\\mathcal { T } ^ { v }$ to be the $3 \\alpha | S |$ points with the largest $v ^ { \\top } x _ { i }$ value, $B ^ { v }$ to be the $3 \\alpha | S |$ points with the smallest $v ^ { \\top } x _ { i }$ value, and $\\mathcal { M } ^ { v } = S \\setminus ( \\mathcal { T } ^ { v } \\cup B ^ { v } )$ . Define $\\begin{array} { r l r } { d ( \\hat { \\mu } , S ) } & { \\triangleq } & { \\operatorname* { m a x } _ { v \\in \\mathbb { S } ^ { d - 1 } } \\left. v ^ { \\top } \\left( \\mu ( \\mathcal { M } ^ { v } ) - \\hat { \\mu } \\right) \\right. } \\end{array}$ . ",
+        "bbox": [
+            173,
+            474,
+            825,
+            523
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "Analysis. For any direction $v$ , the truncated mean estimator $\\mu ( \\mathcal { M } ^ { v } )$ provides a robust estimation of the true mean along the direction $v$ , thus the distance can be simply defined by taking the maximum over all directions $v$ . We show the sensitivity of this simple estimator is bounded by the resilience property $\\sigma$ divided by $n$ , which is ${ \\cal O } ( ( 1 / n ) \\sqrt { \\log ( 1 / \\alpha ) } )$ once the resilience check is passed. This leads to the following near-optimal sample complexity. We provide a proof in Appendix H.2. ",
+        "bbox": [
+            173,
+            532,
+            825,
+            606
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "Theorem 7 (Exponential time algorithm for sub-Gaussian distributions). Algorithm 2 is $( \\varepsilon , \\delta )$ -DP. Under Assumption $^ { l }$ , this algorithm achieves $\\| \\hat { \\mu } - \\mu \\| _ { 2 } = O ( \\alpha \\sqrt { \\log ( 1 / \\alpha ) } )$ with probability $1 - \\zeta i f$ ",
+        "bbox": [
+            173,
+            608,
+            826,
+            641
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "equation",
+        "img_path": "images/9aa30506fb06fc1067dfbb7018804b830bc0f54e20e433fb8eb13e7b665ff047.jpg",
+        "text": "$$\nn = \\widetilde \\Omega \\Big ( \\frac { d + \\log \\frac { 1 } { \\xi } } { \\alpha ^ { 2 } \\log \\frac { 1 } { \\alpha } } + \\frac { d \\log \\Big ( d \\sqrt { \\log ( d n / \\zeta ) } / \\alpha \\Big ) + d ^ { 1 / 2 } \\log \\frac { 1 } { \\delta } + \\log \\frac { 1 } { \\zeta } } { \\varepsilon \\alpha } + \\frac { \\sqrt { d \\log \\frac { 1 } { \\delta } } \\log \\frac { d } { \\zeta \\delta } } { \\varepsilon } \\Big ) ~ .\n$$",
+        "text_format": "latex",
+        "bbox": [
+            194,
+            646,
+            803,
+            693
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "Run-time. Computing $R ( S )$ exactly can take $O ( d e ^ { \\Theta ( n ) } )$ operations. The exponential mechanism implemented with $\\alpha$ -covering for $\\hat { \\mu }$ and a constant covering for $v$ can take $O ( n d ( \\sqrt { \\log ( d n / \\zeta ) } / \\alpha ) ^ { d } )$ operations. ",
+        "bbox": [
+            174,
+            704,
+            826,
+            751
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "5 Conclusion ",
+        "text_level": 1,
+        "bbox": [
+            173,
+            768,
+            299,
+            786
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "Differentially private mean estimation is brittle against a small fraction of the samples being corrupted by an adversary. We show that robustness can be achieved without any increase in the sample complexity by introducing a novel DP mean estimator, which requires run-time exponential in the dimension of the samples. The technical contribution is in leveraging the resilience property of well-behaved distributions in an innovative way to not only find robust mean (which is the typical use case of resilience) but also bound sensitivity for optimal privacy guarantee. To cope with the computational challenge, we propose an efficient algorithm, which we call PRIME, that achieves the optimal target accuracy at the cost of an increased sample complexity. The technical contributions are $( i )$ a novel framework for private iterative filtering and its tight analysis of the end-to-end sensitivity and $( i i )$ novel filtering algorithm of DPTHRESHOLD which is critical in privately running matrix multiplicative weights and hence significantly reducing the number of accesses to the database. With appropriately chosen parameters, we show that our exponential time approach achieves near-optimal guarantees for both sub-Gaussian and covariance bounded distributions and PRIME achieves the same accuracy efficiently but at the cost of an increased sample complexity by a $d ^ { 1 / 2 }$ factor. ",
+        "bbox": [
+            173,
+            799,
+            825,
+            911
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "",
+        "bbox": [
+            174,
+            92,
+            825,
+            176
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "There are several directions for improving our results further and applying the framework to solve other problems. PRIME provides a new design principle for private and robust estimation. This can be more broadly applied to fundamental statistical analyses such as robust covariance estimation [28, 30, 64] robust PCA [60, 48], and robust linear regression [59, 35]. ",
+        "bbox": [
+            174,
+            183,
+            825,
+            238
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "PRIME could be improved in a few directions. First, the sample complexity of $\\widetilde \\Omega ( ( d / ( \\alpha ^ { 2 } \\log ( 1 / \\alpha ) ) ) +$ $( d ^ { 3 / 2 } / ( \\varepsilon \\alpha \\log ( 1 / \\alpha ) ) ) \\log ( 1 / \\delta ) )$ in Theorem 6 is suboptimal in the second term. Improving the $d ^ { 3 / 2 }$ factor requires bypassing differentially private singular value decomposition, which seems to be a challenging task. However, it might be possible to separate the $\\bar { \\log ( 1 / \\delta ) }$ factor from the rest of the terms and get an additive error of the form $\\widetilde \\Omega ( ( d / ( \\alpha ^ { 2 } \\log ( 1 / \\alpha ) ) ) + ( d ^ { 3 / 2 } / ( \\varepsilon \\alpha \\log ( 1 / \\alpha ) ) ) +$ $( 1 / \\varepsilon ) \\log ( 1 / \\delta ) )$ . This requires using Laplace mechanism in private MMW (line 16 Algortihm 10). Secondly, the time complexity of PRIME is dominated by computation time of the matrix exponential in (line 16 Algortihm 10). Total number of operations scale as ${ \\widetilde { O } } ( d ^ { 3 } + n d ^ { 2 } )$ . One might hope to achieve $\\widetilde O ( n d )$ time complexity using approximate computations of $\\tau _ { j }$ ’s using techniques from [36]. This does not improve the sample complexity, as the number of times the dataset is accessed remains the same. Finally, for (non-robust) private mean estimation, COINPRESS provides a practical improvement in the small sample regime by progressively refining the search space [12]. The same principle could be applied to PRIME to design a robust version of COINPRESS. One important question remains open; how are differential privacy and robust statistics fundamentally related? We believe our exponential time algorithm hints on a fundamental connection between robust statistics of a data projected onto one-dimensional subspace and sensitivity of resulting score function for the exponential mechanism. It is an interesting direction to pursue this connection further to design novel algorithms that bridge privacy and robustness. ",
+        "bbox": [
+            173,
+            246,
+            826,
+            505
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "Acknowledgement ",
+        "text_level": 1,
+        "bbox": [
+            176,
+            525,
+            330,
+            541
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "Sham Kakade acknowledges funding from the National Science Foundation under award CCF1703574. Sewoong Oh acknowledges funding from Google faculty research award, NSF grants IIS-1929955, CCF-1705007, CNS-2002664, CCF 2019844 as a part of Institute for Foundation of Machine Learning, and CNS-2112471 as a part of Institute for Future Edge Networks and Distributed Intelligence. ",
+        "bbox": [
+            174,
+            555,
+            825,
+            626
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "References ",
+        "text_level": 1,
+        "bbox": [
+            176,
+            645,
+            266,
+            660
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "[1] Martin Abadi, Andy Chu, Ian Goodfellow, H Brendan McMahan, Ilya Mironov, Kunal Talwar, and Li Zhang. Deep learning with differential privacy. In Proceedings of the 2016 ACM SIGSAC Conference on Computer and Communications Security, pages 308–318, 2016.   \n[2] John M Abowd. The us census bureau adopts differential privacy. In Proceedings of the 24th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining, pages 2867–2867, 2018.   \n[3] Ishaq Aden-Ali, Hassan Ashtiani, and Gautam Kamath. On the sample complexity of privately learning unbounded high-dimensional gaussians. arXiv preprint arXiv:2010.09929, 2020.   \n[4] Zeyuan Allen-Zhu, Zhenyu Liao, and Lorenzo Orecchia. Spectral sparsification and regret minimization beyond matrix multiplicative updates. In Proceedings of the forty-seventh annual ACM symposium on Theory of computing, pages 237–245, 2015.   \n[5] Edoardo Amaldi and Viggo Kann. The complexity and approximability of finding maximum feasible subsystems of linear relations. Theoretical computer science, 147(1-2):181–210, 1995.   \n[6] Frank J Anscombe. Rejection of outliers. Technometrics, 2(2):123–146, 1960. [7] Ainesh Bakshi and Pravesh Kothari. List-decodable subspace recovery via sum-of-squares. arXiv preprint arXiv:2002.05139, 2020.   \n[8] S. Balakrishnan, S. S. Du, J. Li, and A. Singh. Computationally efficient robust sparse estimation in high dimensions. In Proceedings of the 30th Conference on Learning Theory, COLT 2017, pages 169–212, 2017.   \n[9] Amos Beimel, Shay Moran, Kobbi Nissim, and Uri Stemmer. Private center points and learning of halfspaces. arXiv preprint arXiv:1902.10731, 2019.   \n[10] K. Bhatia, P. Jain, P. Kamalaruban, and P. Kar. Consistent robust regression. In Advances in Neural Information Processing Systems 30: Annual Conference on Neural Information Processing Systems 2017, pages 2107–2116, 2017.   \n[11] Kush Bhatia, Prateek Jain, and Purushottam Kar. Robust regression via hard thresholding. In Advances in Neural Information Processing Systems, pages 721–729, 2015.   \n[12] Sourav Biswas, Yihe Dong, Gautam Kamath, and Jonathan Ullman. Coinpress: Practical private mean and covariance estimation. arXiv preprint arXiv:2006.06618, 2020.   \n[13] Avrim Blum, Cynthia Dwork, Frank McSherry, and Kobbi Nissim. Practical privacy: the sulq framework. In Proceedings of the twenty-fourth ACM SIGMOD-SIGACT-SIGART symposium on Principles of database systems, pages 128–138, 2005.   \n[14] Mark Bun, Gautam Kamath, Thomas Steinke, and Steven Z Wu. Private hypothesis selection. In Advances in Neural Information Processing Systems, pages 156–167, 2019.   \n[15] T Tony Cai, Yichen Wang, and Linjun Zhang. The cost of privacy: Optimal rates of convergence for parameter estimation with differential privacy. arXiv preprint arXiv:1902.04495, 2019.   \n[16] Clément L Canonne, Gautam Kamath, Audra McMillan, Jonathan Ullman, and Lydia Zakynthinou. Private identity testing for high-dimensional distributions. arXiv preprint arXiv:1905.11947, 2019.   \n[17] Moses Charikar, Jacob Steinhardt, and Gregory Valiant. Learning from untrusted data. In Proceedings of the 49th Annual ACM SIGACT Symposium on Theory of Computing, pages 47–60, 2017.   \n[18] Kamalika Chaudhuri, Anand D Sarwate, and Kaushik Sinha. A near-optimal algorithm for differentially-private principal components. The Journal of Machine Learning Research, 14(1):2905–2943, 2013.   \n[19] Xinyun Chen, Chang Liu, Bo Li, Kimberly Lu, and Dawn Song. Targeted backdoor attacks on deep learning systems using data poisoning. arXiv preprint arXiv:1712.05526, 2017.   \n[20] Yu Cheng, Ilias Diakonikolas, and Rong Ge. High-dimensional robust mean estimation in nearly-linear time. In Proceedings of the Thirtieth Annual ACM-SIAM Symposium on Discrete Algorithms, pages 2755–2771. SIAM, 2019.   \n[21] Yu Cheng, Ilias Diakonikolas, Rong Ge, and David P Woodruff. Faster algorithms for highdimensional robust covariance estimation. In Conference on Learning Theory, pages 727–757. PMLR, 2019.   \n[22] Yeshwanth Cherapanamjeri, Sidhanth Mohanty, and Morris Yau. List decodable mean estimation in nearly linear time. arXiv preprint arXiv:2005.09796, 2020.   \n[23] Arnak Dalalyan and Philip Thompson. Outlier-robust estimation of a sparse linear model using $\\ell _ { 1 }$ -penalized huber’s $m$ -estimator. In Advances in Neural Information Processing Systems, pages 13188–13198, 2019.   \n[24] Jules Depersin and Guillaume Lecué. Robust subgaussian estimation of a mean vector in nearly linear time. arXiv preprint arXiv:1906.03058, 2019.   \n[25] Luc Devroye and Gábor Lugosi. Combinatorial methods in density estimation. Springer Science & Business Media, 2012. ",
+        "bbox": [
+            179,
+            667,
+            826,
+            914
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "",
+        "bbox": [
+            171,
+            56,
+            828,
+            917
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "[26] Aditya Dhar and Jason Huang. Designing differentially private estimators in high dimensions. arXiv preprint arXiv:2006.01944, 2020. ",
+        "bbox": [
+            171,
+            90,
+            825,
+            119
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "[27] Ilias Diakonikolas, Samuel B Hopkins, Daniel Kane, and Sushrut Karmalkar. Robustly learning any clusterable mixture of gaussians. arXiv preprint arXiv:2005.06417, 2020. ",
+        "bbox": [
+            171,
+            128,
+            823,
+            159
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "[28] Ilias Diakonikolas, Gautam Kamath, Daniel Kane, Jerry Li, Ankur Moitra, and Alistair Stewart. Robust estimators in high-dimensions without the computational intractability. SIAM Journal on Computing, 48(2):742–864, 2019. ",
+        "bbox": [
+            174,
+            166,
+            825,
+            210
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "[29] Ilias Diakonikolas, Gautam Kamath, Daniel Kane, Jerry Li, Jacob Steinhardt, and Alistair Stewart. Sever: A robust meta-algorithm for stochastic optimization. In International Conference on Machine Learning, pages 1596–1606, 2019. ",
+        "bbox": [
+            171,
+            218,
+            823,
+            262
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "[30] Ilias Diakonikolas, Gautam Kamath, Daniel M. Kane, Jerry Li, Ankur Moitra, and Alistair Stewart. Being Robust (in High Dimensions) Can Be Practical. arXiv e-prints, page arXiv:1703.00893, March 2017. ",
+        "bbox": [
+            173,
+            271,
+            825,
+            314
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "[31] Ilias Diakonikolas, Gautam Kamath, Daniel M Kane, Jerry Li, Ankur Moitra, and Alistair Stewart. Robustly learning a gaussian: Getting optimal error, efficiently. In Proceedings of the Twenty-Ninth Annual ACM-SIAM Symposium on Discrete Algorithms, pages 2683–2702. SIAM, 2018. ",
+        "bbox": [
+            173,
+            321,
+            825,
+            378
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "[32] Ilias Diakonikolas and Daniel M Kane. Recent advances in algorithmic high-dimensional robust statistics. arXiv preprint arXiv:1911.05911, 2019. ",
+        "bbox": [
+            173,
+            387,
+            823,
+            417
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "[33] Ilias Diakonikolas, Daniel M Kane, and Alistair Stewart. Statistical query lower bounds for robust estimation of high-dimensional gaussians and gaussian mixtures. In 2017 IEEE 58th Annual Symposium on Foundations of Computer Science (FOCS), pages 73–84. IEEE, 2017. ",
+        "bbox": [
+            176,
+            426,
+            823,
+            469
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "[34] Ilias Diakonikolas, Daniel M Kane, and Alistair Stewart. List-decodable robust mean estimation and learning mixtures of spherical gaussians. In Proceedings of the 50th Annual ACM SIGACT Symposium on Theory of Computing, pages 1047–1060, 2018. ",
+        "bbox": [
+            174,
+            478,
+            823,
+            522
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "[35] Ilias Diakonikolas, Weihao Kong, and Alistair Stewart. Efficient algorithms and lower bounds for robust linear regression. In Proceedings of the Thirtieth Annual ACM-SIAM Symposium on Discrete Algorithms, pages 2745–2754. SIAM, 2019. ",
+        "bbox": [
+            174,
+            530,
+            823,
+            573
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "[36] Yihe Dong, Samuel Hopkins, and Jerry Li. Quantum entropy scoring for fast robust mean estimation and improved outlier detection. In Advances in Neural Information Processing Systems, pages 6067–6077, 2019. ",
+        "bbox": [
+            173,
+            582,
+            823,
+            625
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "[37] Cynthia Dwork, Frank McSherry, Kobbi Nissim, and Adam Smith. Calibrating noise to sensitivity in private data analysis. In Theory of cryptography conference, pages 265–284. Springer, 2006. ",
+        "bbox": [
+            173,
+            633,
+            825,
+            676
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "[38] Cynthia Dwork and Aaron Roth. The algorithmic foundations of differential privacy. Foundations and Trends in Theoretical Computer Science, 9(3-4):211–407, 2014. ",
+        "bbox": [
+            173,
+            685,
+            825,
+            715
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "[39] Cynthia Dwork, Kunal Talwar, Abhradeep Thakurta, and Li Zhang. Analyze gauss: optimal bounds for privacy-preserving principal component analysis. In Proceedings of the forty-sixth annual ACM symposium on Theory of computing, pages 11–20, 2014. ",
+        "bbox": [
+            173,
+            724,
+            825,
+            767
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "[40] Úlfar Erlingsson, Vasyl Pihur, and Aleksandra Korolova. Rappor: Randomized aggregatable privacy-preserving ordinal response. In Proceedings of the 2014 ACM SIGSAC conference on computer and communications security, pages 1054–1067, 2014. ",
+        "bbox": [
+            173,
+            776,
+            823,
+            820
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "[41] Giulia Fanti, Vasyl Pihur, and Úlfar Erlingsson. Building a rappor with the unknown: Privacypreserving learning of associations and data dictionaries. Proceedings on Privacy Enhancing Technologies, 2016(3):41–61, 2016. ",
+        "bbox": [
+            173,
+            830,
+            821,
+            873
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "[42] Chao Gao et al. Robust regression via mutivariate regression depth. Bernoulli, 26(2):1139–1170, 2020. ",
+        "bbox": [
+            173,
+            882,
+            821,
+            911
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "[43] Sam Hopkins, Jerry Li, and Fred Zhang. Robust and heavy-tailed mean estimation made simple, via regret minimization. Advances in Neural Information Processing Systems, 33, 2020. ",
+        "bbox": [
+            171,
+            90,
+            826,
+            121
+        ],
+        "page_idx": 12
+    },
+    {
+        "type": "text",
+        "text": "[44] Samuel B Hopkins. Mean estimation with sub-gaussian rates in polynomial time. Annals of Statistics, 48(2):1193–1213, 2020.   \n[45] Samuel B Hopkins and Jerry Li. Mixture models, robustness, and sum of squares proofs. In Proceedings of the 50th Annual ACM SIGACT Symposium on Theory of Computing, pages 1021–1034, 2018.   \n[46] Samuel B Hopkins and Jerry Li. How hard is robust mean estimation? In Conference on Learning Theory, pages 1649–1682. PMLR, 2019.   \n[47] Peter J. Huber. Robust Estimation of a Location Parameter. The Annals of Mathematical Statistics, 35(1):73 – 101, 1964.   \n[48] Arun Jambulapati, Jerry Li, and Kevin Tian. Robust sub-gaussian principal component analysis and width-independent schatten packing. Advances in Neural Information Processing Systems, 33, 2020.   \n[49] He Jia and Santosh Vempala. Robustly clustering a mixture of gaussians. arXiv preprint arXiv:1911.11838, 2019.   \n[50] Peter Kairouz, H Brendan McMahan, Brendan Avent, Aurélien Bellet, Mehdi Bennis, Arjun Nitin Bhagoji, Keith Bonawitz, Zachary Charles, Graham Cormode, Rachel Cummings, et al. Advances and open problems in federated learning. arXiv preprint arXiv:1912.04977, 2019.   \n[51] Peter Kairouz, Sewoong Oh, and Pramod Viswanath. The composition theorem for differential privacy. In International conference on machine learning, pages 1376–1385, 2015.   \n[52] Gautam Kamath, Jerry Li, Vikrant Singhal, and Jonathan Ullman. Privately learning highdimensional distributions. In Conference on Learning Theory, pages 1853–1902, 2019.   \n[53] Gautam Kamath, Or Sheffet, Vikrant Singhal, and Jonathan Ullman. Differentially private algorithms for learning mixtures of separated gaussians. In 2020 Information Theory and Applications Workshop (ITA), pages 1–62. IEEE, 2020.   \n[54] Gautam Kamath, Vikrant Singhal, and Jonathan Ullman. Private mean estimation of heavy-tailed distributions. arXiv preprint arXiv:2002.09464, 2020.   \n[55] Haim Kaplan, Katrina Ligett, Yishay Mansour, Moni Naor, and Uri Stemmer. Privately learning thresholds: Closing the exponential gap. In Conference on Learning Theory, pages 2263–2285. PMLR, 2020.   \n[56] Sushrut Karmalkar, Adam Klivans, and Pravesh Kothari. List-decodable linear regression. In Advances in Neural Information Processing Systems, pages 7423–7432, 2019.   \n[57] Sushrut Karmalkar and Eric Price. Compressed sensing with adversarial sparse noise via l1 regression. In 2nd Symposium on Simplicity in Algorithms, 2019.   \n[58] Vishesh Karwa and Salil Vadhan. Finite sample differentially private confidence intervals. arXiv preprint arXiv:1711.03908, 2017.   \n[59] Adam Klivans, Pravesh K Kothari, and Raghu Meka. Efficient algorithms for outlier-robust regression. In Conference On Learning Theory, pages 1420–1430, 2018.   \n[60] Weihao Kong, Raghav Somani, Sham Kakade, and Sewoong Oh. Robust meta-learning for mixed linear regression with small batches. Advances in Neural Information Processing Systems, 33, 2020.   \n[61] Pravesh K Kothari, Jacob Steinhardt, and David Steurer. Robust moment estimation and improved clustering via sum of squares. In Proceedings of the 50th Annual ACM SIGACT Symposium on Theory of Computing, pages 1035–1046, 2018.   \n[62] Kevin A Lai, Anup B Rao, and Santosh Vempala. Agnostic estimation of mean and covariance. In 2016 IEEE 57th Annual Symposium on Foundations of Computer Science (FOCS), pages 665–674. IEEE, 2016.   \n[63] Jerry Li. CSE 599-M, Lecture Notes: Robustness in Machine Learning , 2019. URL: https: //jerryzli.github.io/robust-ml-fall19/lec7.pdf.   \n[64] Jerry Li and Guanghao Ye. Robust gaussian covariance estimation in nearly-matrix multiplication time. Advances in Neural Information Processing Systems, 33, 2020.   \n[65] Liu Liu, Yanyao Shen, Tianyang Li, and Constantine Caramanis. High dimensional robust sparse regression. arXiv preprint arXiv:1805.11643, 2018.   \n[66] Xiaohui Liu. Fast implementation of the tukey depth. Computational Statistics, 32(4):1395– 1410, 2017.   \n[67] Xiaohui Liu, Karl Mosler, and Pavlo Mozharovskyi. Fast computation of tukey trimmed regions and median in dimension p> 2. Journal of Computational and Graphical Statistics, 28(3):682–697, 2019.   \n[68] Gábor Lugosi, Shahar Mendelson, et al. Sub-gaussian estimators of the mean of a random vector. Annals of Statistics, 47(2):783–794, 2019.   \n[69] Frank McSherry and Kunal Talwar. Mechanism design via differential privacy. In 48th Annual IEEE Symposium on Foundations of Computer Science (FOCS’07), pages 94–103. IEEE, 2007.   \n[70] Bhaskar Mukhoty, Govind Gopakumar, Prateek Jain, and Purushottam Kar. Globally-convergent iteratively reweighted least squares for robust regression problems. In The 22nd International Conference on Artificial Intelligence and Statistics, pages 313–322, 2019.   \n[71] A. Prasad, A. S. Suggala, S. Balakrishnan, and P. Ravikumar. Robust estimation via robust gradient estimation. arXiv preprint arXiv:1802.06485, 2018.   \n[72] Prasad Raghavendra and Morris Yau. List decodable learning via sum of squares. In Proceedings of the Fourteenth Annual ACM-SIAM Symposium on Discrete Algorithms, pages 161–180. SIAM, 2020.   \n[73] Jacob Steinhardt, Moses Charikar, and Gregory Valiant. Resilience: A criterion for learning in the presence of arbitrary outliers. In 9th Innovations in Theoretical Computer Science Conference (ITCS 2018). Schloss Dagstuhl-Leibniz-Zentrum fuer Informatik, 2018.   \n[74] Jun Tang, Aleksandra Korolova, Xiaolong Bai, Xueqiang Wang, and Xiaofeng Wang. Privacy loss in apple’s implementation of differential privacy on macos 10.12. arXiv preprint arXiv:1709.02753, 2017.   \n[75] Terence Tao. Topics in random matrix theory, volume 132. American Mathematical Soc., 2012.   \n[76] John W Tukey. A survey of sampling from contaminated distributions. Contributions to probability and statistics, pages 448–485, 1960.   \n[77] Martin J Wainwright. High-dimensional statistics: A non-asymptotic viewpoint, volume 48. Cambridge University Press, 2019.   \n[78] Lu Wei, Anand D Sarwate, Jukka Corander, Alfred Hero, and Vahid Tarokh. Analysis of a privacy-preserving pca algorithm using random matrix theory. In 2016 IEEE Global Conference on Signal and Information Processing (GlobalSIP), pages 1335–1339. IEEE, 2016.   \n[79] Huang Xiao, Battista Biggio, Gavin Brown, Giorgio Fumera, Claudia Eckert, and Fabio Roli. Is feature selection secure against training data poisoning? In International Conference on Machine Learning, pages 1689–1698. PMLR, 2015.   \n[80] Huanyu Zhang, Gautam Kamath, Janardhan Kulkarni, and Zhiwei Steven Wu. Privately learning markov random fields. arXiv preprint arXiv:2002.09463, 2020. ",
+        "bbox": [
+            169,
+            128,
+            828,
+            917
+        ],
+        "page_idx": 12
+    },
+    {
+        "type": "text",
+        "text": "",
+        "bbox": [
+            171,
+            71,
+            828,
+            914
+        ],
+        "page_idx": 13
+    }
+]
\ No newline at end of file
diff --git a/parse/train/CuQoImkKkIj/CuQoImkKkIj_middle.json b/parse/train/CuQoImkKkIj/CuQoImkKkIj_middle.json
new file mode 100644
index 0000000000000000000000000000000000000000..4b384e4cdd0bbd3e374282d44b495fa2413760d2
--- /dev/null
+++ b/parse/train/CuQoImkKkIj/CuQoImkKkIj_middle.json
@@ -0,0 +1,49894 @@
+{
+    "pdf_info": [
+        {
+            "preproc_blocks": [
+                {
+                    "type": "title",
+                    "bbox": [
+                        120,
+                        97,
+                        491,
+                        118
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                119,
+                                96,
+                                491,
+                                119
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        119,
+                                        96,
+                                        491,
+                                        119
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Robust and differentially private mean estimation",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        }
+                    ],
+                    "index": 0
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        175,
+                        159,
+                        435,
+                        171
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                174,
+                                158,
+                                436,
+                                174
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        174,
+                                        158,
+                                        436,
+                                        174
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Xiyang Liu, Weihao Kong, Sham Kakade, Sewoong Oh",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        }
+                    ],
+                    "index": 1
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        180,
+                        171,
+                        428,
+                        204
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                182,
+                                167,
+                                428,
+                                184
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        182,
+                                        167,
+                                        428,
+                                        184
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Paul G. Allen School of Computer Science and Engineering,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                253,
+                                180,
+                                358,
+                                194
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        253,
+                                        180,
+                                        358,
+                                        194
+                                    ],
+                                    "score": 1.0,
+                                    "content": "University of Washington",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                179,
+                                192,
+                                427,
+                                205
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        179,
+                                        192,
+                                        427,
+                                        205
+                                    ],
+                                    "score": 1.0,
+                                    "content": "{xiyangl,whkong,sham,sewoong}@cs.washington.edu",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        }
+                    ],
+                    "index": 3
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        283,
+                        232,
+                        328,
+                        245
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                281,
+                                231,
+                                330,
+                                246
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        281,
+                                        231,
+                                        330,
+                                        246
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Abstract",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        }
+                    ],
+                    "index": 5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        143,
+                        255,
+                        469,
+                        409
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                141,
+                                255,
+                                469,
+                                269
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        255,
+                                        469,
+                                        269
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In statistical learning and analysis from shared data, which is increasingly widely",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                267,
+                                470,
+                                280
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        267,
+                                        470,
+                                        280
+                                    ],
+                                    "score": 1.0,
+                                    "content": "adopted in platforms such as federated learning and meta-learning, there are two",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                279,
+                                469,
+                                290
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        279,
+                                        469,
+                                        290
+                                    ],
+                                    "score": 1.0,
+                                    "content": "major concerns: privacy and robustness. Each participating individual should be",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                289,
+                                469,
+                                301
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        289,
+                                        469,
+                                        301
+                                    ],
+                                    "score": 1.0,
+                                    "content": "able to contribute without the fear of leaking one’s sensitive information. At the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                300,
+                                469,
+                                312
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        300,
+                                        469,
+                                        312
+                                    ],
+                                    "score": 1.0,
+                                    "content": "same time, the system should be robust in the presence of malicious participants",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                311,
+                                470,
+                                323
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        311,
+                                        470,
+                                        323
+                                    ],
+                                    "score": 1.0,
+                                    "content": "inserting corrupted data. Recent algorithmic advances in learning from shared",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                322,
+                                470,
+                                334
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        322,
+                                        470,
+                                        334
+                                    ],
+                                    "score": 1.0,
+                                    "content": "data focus on either one of these threats, leaving the system vulnerable to the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                142,
+                                333,
+                                469,
+                                344
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        142,
+                                        333,
+                                        469,
+                                        344
+                                    ],
+                                    "score": 1.0,
+                                    "content": "other. We bridge this gap for the canonical problem of estimating the mean from",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                344,
+                                470,
+                                355
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        344,
+                                        470,
+                                        355
+                                    ],
+                                    "score": 1.0,
+                                    "content": "i.i.d. samples. We introduce PRIME, which is the first efficient algorithm that",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                142,
+                                355,
+                                470,
+                                366
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        142,
+                                        355,
+                                        470,
+                                        366
+                                    ],
+                                    "score": 1.0,
+                                    "content": "achieves both privacy and robustness for a wide range of distributions. We further",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                142,
+                                366,
+                                469,
+                                377
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        142,
+                                        366,
+                                        469,
+                                        377
+                                    ],
+                                    "score": 1.0,
+                                    "content": "complement this result with a novel exponential time algorithm that improves the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                376,
+                                470,
+                                389
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        376,
+                                        470,
+                                        389
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sample complexity of PRIME, achieving a near-optimal guarantee and matching a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                388,
+                                469,
+                                399
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        388,
+                                        469,
+                                        399
+                                    ],
+                                    "score": 1.0,
+                                    "content": "known lower bound for (non-robust) private mean estimation. This proves that there",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                398,
+                                467,
+                                411
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        398,
+                                        467,
+                                        411
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is no extra statistical cost to simultaneously guaranteeing privacy and robustness.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        }
+                    ],
+                    "index": 12.5
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        428,
+                        190,
+                        442
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                427,
+                                192,
+                                444
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        427,
+                                        192,
+                                        444
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1 Introduction",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        }
+                    ],
+                    "index": 20
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        452,
+                        505,
+                        540
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                453,
+                                505,
+                                465
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        453,
+                                        505,
+                                        465
+                                    ],
+                                    "score": 1.0,
+                                    "content": "When releasing database statistics on a collection of entries from individuals, we would ideally",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                463,
+                                506,
+                                476
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        463,
+                                        506,
+                                        476
+                                    ],
+                                    "score": 1.0,
+                                    "content": "like to make it impossible to reverse-engineer each individual’s potentially sensitive information.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                474,
+                                505,
+                                487
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        474,
+                                        505,
+                                        487
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Privacy-preserving techniques add just enough randomness tailored to the statistical task to guarantee",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                486,
+                                506,
+                                498
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        486,
+                                        506,
+                                        498
+                                    ],
+                                    "score": 1.0,
+                                    "content": "protection. At the same time, it is becoming increasingly common to apply such techniques to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                496,
+                                505,
+                                509
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        496,
+                                        505,
+                                        509
+                                    ],
+                                    "score": 1.0,
+                                    "content": "databases collected from multiple sources, not all of which can be trusted. Emerging data access",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                507,
+                                505,
+                                520
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        507,
+                                        505,
+                                        520
+                                    ],
+                                    "score": 1.0,
+                                    "content": "frameworks, such as federated analyses across users’ devices or data silos [50], make it easier to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                519,
+                                505,
+                                530
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        519,
+                                        505,
+                                        530
+                                    ],
+                                    "score": 1.0,
+                                    "content": "temper with such collected datasets, leaving private statistical analyses vulnerable to a malicious",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                529,
+                                249,
+                                541
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        529,
+                                        249,
+                                        541
+                                    ],
+                                    "score": 1.0,
+                                    "content": "corruption of a fraction of the data.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        }
+                    ],
+                    "index": 24.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        546,
+                        505,
+                        633
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                545,
+                                506,
+                                559
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        545,
+                                        506,
+                                        559
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Differential privacy has emerged as a widely accepted de facto measure of privacy, which is now a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                557,
+                                506,
+                                569
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        557,
+                                        506,
+                                        569
+                                    ],
+                                    "score": 1.0,
+                                    "content": "standard in releasing the statistics of the U.S. Census data [2] statistics and also deployed in real-world",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                568,
+                                506,
+                                580
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        568,
+                                        506,
+                                        580
+                                    ],
+                                    "score": 1.0,
+                                    "content": "commercial systems [74, 40, 41]. A statistical analysis is said to be differentially private (DP) if",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                578,
+                                505,
+                                592
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        578,
+                                        505,
+                                        592
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the likelihood of the (randomized) outcome does not change significantly when a single arbitrary",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                589,
+                                505,
+                                602
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        589,
+                                        287,
+                                        602
+                                    ],
+                                    "score": 1.0,
+                                    "content": "entry is added/removed (formally defined in",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        287,
+                                        590,
+                                        308,
+                                        601
+                                    ],
+                                    "score": 0.84,
+                                    "content": "\\ S 1 . 2 )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        309,
+                                        589,
+                                        505,
+                                        602
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". This provides a strong privacy guarantee: even",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                600,
+                                505,
+                                613
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        600,
+                                        505,
+                                        613
+                                    ],
+                                    "score": 1.0,
+                                    "content": "a powerful adversary who knows all the other entries in the database cannot confidently identify",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                612,
+                                506,
+                                623
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        612,
+                                        506,
+                                        623
+                                    ],
+                                    "score": 1.0,
+                                    "content": "whether a particular individual is participating in the database based on the outcome of the analysis.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                621,
+                                423,
+                                636
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        621,
+                                        423,
+                                        636
+                                    ],
+                                    "score": 1.0,
+                                    "content": "This ensures plausible deniability, central to protecting an individual’s privacy.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        }
+                    ],
+                    "index": 32.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        638,
+                        505,
+                        715
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                638,
+                                506,
+                                651
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        638,
+                                        506,
+                                        651
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In this paper, we focus on one of the most canonical problems in statistics: estimating the mean of a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                648,
+                                505,
+                                662
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        648,
+                                        505,
+                                        662
+                                    ],
+                                    "score": 1.0,
+                                    "content": "distribution from i.i.d. samples. For distributions with unbounded support, such as sub-Gaussian and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                660,
+                                505,
+                                673
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        660,
+                                        505,
+                                        673
+                                    ],
+                                    "score": 1.0,
+                                    "content": "heavy-tailed distributions, fundamental trade-offs between accuracy, sample size, and privacy have",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                671,
+                                505,
+                                684
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        671,
+                                        505,
+                                        684
+                                    ],
+                                    "score": 1.0,
+                                    "content": "only recently been identified [58, 52, 54, 3] and efficient private estimators proposed. However, these",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                682,
+                                505,
+                                695
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        682,
+                                        505,
+                                        695
+                                    ],
+                                    "score": 1.0,
+                                    "content": "approaches are brittle when a fraction of the data is corrupted, posing a real threat, referred to as data",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                693,
+                                506,
+                                706
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        693,
+                                        506,
+                                        706
+                                    ],
+                                    "score": 1.0,
+                                    "content": "poisoning attacks [19, 79]. In defense of such attacks, robust (but not necessarily private) statistics",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                704,
+                                493,
+                                716
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        704,
+                                        493,
+                                        716
+                                    ],
+                                    "score": 1.0,
+                                    "content": "has emerged as a popular setting of recent algorithmic and mathematical breakthroughs [73, 30].",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        }
+                    ],
+                    "index": 40
+                }
+            ],
+            "page_idx": 0,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        107,
+                        731,
+                        385,
+                        741
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                730,
+                                387,
+                                744
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        730,
+                                        387,
+                                        744
+                                    ],
+                                    "score": 1.0,
+                                    "content": "35th Conference on Neural Information Processing Systems (NeurIPS 2021).",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "title",
+                    "bbox": [
+                        120,
+                        97,
+                        491,
+                        118
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                119,
+                                96,
+                                491,
+                                119
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        119,
+                                        96,
+                                        491,
+                                        119
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Robust and differentially private mean estimation",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        }
+                    ],
+                    "index": 0
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        175,
+                        159,
+                        435,
+                        171
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                174,
+                                158,
+                                436,
+                                174
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        174,
+                                        158,
+                                        436,
+                                        174
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Xiyang Liu, Weihao Kong, Sham Kakade, Sewoong Oh",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        }
+                    ],
+                    "index": 1,
+                    "bbox_fs": [
+                        174,
+                        158,
+                        436,
+                        174
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        180,
+                        171,
+                        428,
+                        204
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                182,
+                                167,
+                                428,
+                                184
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        182,
+                                        167,
+                                        428,
+                                        184
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Paul G. Allen School of Computer Science and Engineering,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                253,
+                                180,
+                                358,
+                                194
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        253,
+                                        180,
+                                        358,
+                                        194
+                                    ],
+                                    "score": 1.0,
+                                    "content": "University of Washington",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                179,
+                                192,
+                                427,
+                                205
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        179,
+                                        192,
+                                        427,
+                                        205
+                                    ],
+                                    "score": 1.0,
+                                    "content": "{xiyangl,whkong,sham,sewoong}@cs.washington.edu",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        }
+                    ],
+                    "index": 3,
+                    "bbox_fs": [
+                        179,
+                        167,
+                        428,
+                        205
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        283,
+                        232,
+                        328,
+                        245
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                281,
+                                231,
+                                330,
+                                246
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        281,
+                                        231,
+                                        330,
+                                        246
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Abstract",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        }
+                    ],
+                    "index": 5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        143,
+                        255,
+                        469,
+                        409
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                141,
+                                255,
+                                469,
+                                269
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        255,
+                                        469,
+                                        269
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In statistical learning and analysis from shared data, which is increasingly widely",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                267,
+                                470,
+                                280
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        267,
+                                        470,
+                                        280
+                                    ],
+                                    "score": 1.0,
+                                    "content": "adopted in platforms such as federated learning and meta-learning, there are two",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                279,
+                                469,
+                                290
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        279,
+                                        469,
+                                        290
+                                    ],
+                                    "score": 1.0,
+                                    "content": "major concerns: privacy and robustness. Each participating individual should be",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                289,
+                                469,
+                                301
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        289,
+                                        469,
+                                        301
+                                    ],
+                                    "score": 1.0,
+                                    "content": "able to contribute without the fear of leaking one’s sensitive information. At the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                300,
+                                469,
+                                312
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        300,
+                                        469,
+                                        312
+                                    ],
+                                    "score": 1.0,
+                                    "content": "same time, the system should be robust in the presence of malicious participants",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                311,
+                                470,
+                                323
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        311,
+                                        470,
+                                        323
+                                    ],
+                                    "score": 1.0,
+                                    "content": "inserting corrupted data. Recent algorithmic advances in learning from shared",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                322,
+                                470,
+                                334
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        322,
+                                        470,
+                                        334
+                                    ],
+                                    "score": 1.0,
+                                    "content": "data focus on either one of these threats, leaving the system vulnerable to the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                142,
+                                333,
+                                469,
+                                344
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        142,
+                                        333,
+                                        469,
+                                        344
+                                    ],
+                                    "score": 1.0,
+                                    "content": "other. We bridge this gap for the canonical problem of estimating the mean from",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                344,
+                                470,
+                                355
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        344,
+                                        470,
+                                        355
+                                    ],
+                                    "score": 1.0,
+                                    "content": "i.i.d. samples. We introduce PRIME, which is the first efficient algorithm that",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                142,
+                                355,
+                                470,
+                                366
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        142,
+                                        355,
+                                        470,
+                                        366
+                                    ],
+                                    "score": 1.0,
+                                    "content": "achieves both privacy and robustness for a wide range of distributions. We further",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                142,
+                                366,
+                                469,
+                                377
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        142,
+                                        366,
+                                        469,
+                                        377
+                                    ],
+                                    "score": 1.0,
+                                    "content": "complement this result with a novel exponential time algorithm that improves the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                376,
+                                470,
+                                389
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        376,
+                                        470,
+                                        389
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sample complexity of PRIME, achieving a near-optimal guarantee and matching a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                388,
+                                469,
+                                399
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        388,
+                                        469,
+                                        399
+                                    ],
+                                    "score": 1.0,
+                                    "content": "known lower bound for (non-robust) private mean estimation. This proves that there",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                398,
+                                467,
+                                411
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        398,
+                                        467,
+                                        411
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is no extra statistical cost to simultaneously guaranteeing privacy and robustness.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        }
+                    ],
+                    "index": 12.5,
+                    "bbox_fs": [
+                        141,
+                        255,
+                        470,
+                        411
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        428,
+                        190,
+                        442
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                427,
+                                192,
+                                444
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        427,
+                                        192,
+                                        444
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1 Introduction",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        }
+                    ],
+                    "index": 20
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        452,
+                        505,
+                        540
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                453,
+                                505,
+                                465
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        453,
+                                        505,
+                                        465
+                                    ],
+                                    "score": 1.0,
+                                    "content": "When releasing database statistics on a collection of entries from individuals, we would ideally",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                463,
+                                506,
+                                476
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        463,
+                                        506,
+                                        476
+                                    ],
+                                    "score": 1.0,
+                                    "content": "like to make it impossible to reverse-engineer each individual’s potentially sensitive information.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                474,
+                                505,
+                                487
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        474,
+                                        505,
+                                        487
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Privacy-preserving techniques add just enough randomness tailored to the statistical task to guarantee",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                486,
+                                506,
+                                498
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        486,
+                                        506,
+                                        498
+                                    ],
+                                    "score": 1.0,
+                                    "content": "protection. At the same time, it is becoming increasingly common to apply such techniques to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                496,
+                                505,
+                                509
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        496,
+                                        505,
+                                        509
+                                    ],
+                                    "score": 1.0,
+                                    "content": "databases collected from multiple sources, not all of which can be trusted. Emerging data access",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                507,
+                                505,
+                                520
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        507,
+                                        505,
+                                        520
+                                    ],
+                                    "score": 1.0,
+                                    "content": "frameworks, such as federated analyses across users’ devices or data silos [50], make it easier to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                519,
+                                505,
+                                530
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        519,
+                                        505,
+                                        530
+                                    ],
+                                    "score": 1.0,
+                                    "content": "temper with such collected datasets, leaving private statistical analyses vulnerable to a malicious",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                529,
+                                249,
+                                541
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        529,
+                                        249,
+                                        541
+                                    ],
+                                    "score": 1.0,
+                                    "content": "corruption of a fraction of the data.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        }
+                    ],
+                    "index": 24.5,
+                    "bbox_fs": [
+                        105,
+                        453,
+                        506,
+                        541
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        546,
+                        505,
+                        633
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                545,
+                                506,
+                                559
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        545,
+                                        506,
+                                        559
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Differential privacy has emerged as a widely accepted de facto measure of privacy, which is now a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                557,
+                                506,
+                                569
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        557,
+                                        506,
+                                        569
+                                    ],
+                                    "score": 1.0,
+                                    "content": "standard in releasing the statistics of the U.S. Census data [2] statistics and also deployed in real-world",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                568,
+                                506,
+                                580
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        568,
+                                        506,
+                                        580
+                                    ],
+                                    "score": 1.0,
+                                    "content": "commercial systems [74, 40, 41]. A statistical analysis is said to be differentially private (DP) if",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                578,
+                                505,
+                                592
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        578,
+                                        505,
+                                        592
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the likelihood of the (randomized) outcome does not change significantly when a single arbitrary",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                589,
+                                505,
+                                602
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        589,
+                                        287,
+                                        602
+                                    ],
+                                    "score": 1.0,
+                                    "content": "entry is added/removed (formally defined in",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        287,
+                                        590,
+                                        308,
+                                        601
+                                    ],
+                                    "score": 0.84,
+                                    "content": "\\ S 1 . 2 )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        309,
+                                        589,
+                                        505,
+                                        602
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". This provides a strong privacy guarantee: even",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                600,
+                                505,
+                                613
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        600,
+                                        505,
+                                        613
+                                    ],
+                                    "score": 1.0,
+                                    "content": "a powerful adversary who knows all the other entries in the database cannot confidently identify",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                612,
+                                506,
+                                623
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        612,
+                                        506,
+                                        623
+                                    ],
+                                    "score": 1.0,
+                                    "content": "whether a particular individual is participating in the database based on the outcome of the analysis.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                621,
+                                423,
+                                636
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        621,
+                                        423,
+                                        636
+                                    ],
+                                    "score": 1.0,
+                                    "content": "This ensures plausible deniability, central to protecting an individual’s privacy.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        }
+                    ],
+                    "index": 32.5,
+                    "bbox_fs": [
+                        104,
+                        545,
+                        506,
+                        636
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        638,
+                        505,
+                        715
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                638,
+                                506,
+                                651
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        638,
+                                        506,
+                                        651
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In this paper, we focus on one of the most canonical problems in statistics: estimating the mean of a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                648,
+                                505,
+                                662
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        648,
+                                        505,
+                                        662
+                                    ],
+                                    "score": 1.0,
+                                    "content": "distribution from i.i.d. samples. For distributions with unbounded support, such as sub-Gaussian and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                660,
+                                505,
+                                673
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        660,
+                                        505,
+                                        673
+                                    ],
+                                    "score": 1.0,
+                                    "content": "heavy-tailed distributions, fundamental trade-offs between accuracy, sample size, and privacy have",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                671,
+                                505,
+                                684
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        671,
+                                        505,
+                                        684
+                                    ],
+                                    "score": 1.0,
+                                    "content": "only recently been identified [58, 52, 54, 3] and efficient private estimators proposed. However, these",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                682,
+                                505,
+                                695
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        682,
+                                        505,
+                                        695
+                                    ],
+                                    "score": 1.0,
+                                    "content": "approaches are brittle when a fraction of the data is corrupted, posing a real threat, referred to as data",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                693,
+                                506,
+                                706
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        693,
+                                        506,
+                                        706
+                                    ],
+                                    "score": 1.0,
+                                    "content": "poisoning attacks [19, 79]. In defense of such attacks, robust (but not necessarily private) statistics",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                704,
+                                493,
+                                716
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        704,
+                                        493,
+                                        716
+                                    ],
+                                    "score": 1.0,
+                                    "content": "has emerged as a popular setting of recent algorithmic and mathematical breakthroughs [73, 30].",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        }
+                    ],
+                    "index": 40,
+                    "bbox_fs": [
+                        105,
+                        638,
+                        506,
+                        716
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        72,
+                        505,
+                        161
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                73,
+                                505,
+                                85
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        73,
+                                        505,
+                                        85
+                                    ],
+                                    "score": 1.0,
+                                    "content": "One might be misled into thinking that privacy ensures robustness since DP guarantees that a single",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                84,
+                                505,
+                                95
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        84,
+                                        505,
+                                        95
+                                    ],
+                                    "score": 1.0,
+                                    "content": "outlier cannot change the estimation too much. This intuition is true only in a low dimension; each",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                94,
+                                506,
+                                107
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        94,
+                                        506,
+                                        107
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sample has to be an obvious outlier to significantly change the mean. However, in a high dimension,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                106,
+                                505,
+                                118
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        106,
+                                        505,
+                                        118
+                                    ],
+                                    "score": 1.0,
+                                    "content": "each corrupted data point can look perfectly uncorrupted but still shift the mean significant when",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                117,
+                                506,
+                                130
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        117,
+                                        506,
+                                        130
+                                    ],
+                                    "score": 1.0,
+                                    "content": "colluding together (e.g., see Fig. 1). Focusing on the canonical problem of mean estimation, we",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                127,
+                                505,
+                                140
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        127,
+                                        505,
+                                        140
+                                    ],
+                                    "score": 1.0,
+                                    "content": "introduce novel algorithms that achieve robustness and privacy simultaneously even when a fraction",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                138,
+                                505,
+                                150
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        138,
+                                        505,
+                                        150
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of data is corrupted arbitrarily. For such algorithms, there is a fundamental question of interest: do",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                149,
+                                502,
+                                161
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        149,
+                                        502,
+                                        161
+                                    ],
+                                    "score": 1.0,
+                                    "content": "we need more samples to make private mean estimation also robust against adversarial corruption?",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        }
+                    ],
+                    "index": 3.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        165,
+                        505,
+                        244
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                164,
+                                506,
+                                178
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        164,
+                                        506,
+                                        178
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Sub-Gaussian distributions. If we can afford exponential run-time in the dimension, robustness",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                176,
+                                505,
+                                189
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        176,
+                                        505,
+                                        189
+                                    ],
+                                    "score": 1.0,
+                                    "content": "can be achieved without extra cost in sample complexity. We introduce a novel estimator that (i)",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                186,
+                                506,
+                                201
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        186,
+                                        140,
+                                        201
+                                    ],
+                                    "score": 1.0,
+                                    "content": "satisfies",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        140,
+                                        187,
+                                        162,
+                                        199
+                                    ],
+                                    "score": 0.91,
+                                    "content": "( \\varepsilon , \\delta )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        163,
+                                        186,
+                                        181,
+                                        201
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-DP,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        181,
+                                        187,
+                                        196,
+                                        199
+                                    ],
+                                    "score": 0.78,
+                                    "content": "( i i )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        196,
+                                        186,
+                                        352,
+                                        201
+                                    ],
+                                    "score": 1.0,
+                                    "content": "achieves near-optimal robustness under",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        352,
+                                        189,
+                                        360,
+                                        198
+                                    ],
+                                    "score": 0.79,
+                                    "content": "\\alpha",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        360,
+                                        186,
+                                        506,
+                                        201
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-fraction of corrupted data, achieving",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                199,
+                                505,
+                                213
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        200,
+                                        154,
+                                        212
+                                    ],
+                                    "score": 1.0,
+                                    "content": "accuracy of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        155,
+                                        199,
+                                        223,
+                                        213
+                                    ],
+                                    "score": 0.93,
+                                    "content": "O ( \\alpha \\sqrt { \\log ( 1 / \\alpha ) } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        223,
+                                        200,
+                                        420,
+                                        212
+                                    ],
+                                    "score": 1.0,
+                                    "content": "nearly matching the fundamental lower bound of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        420,
+                                        199,
+                                        442,
+                                        212
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\Omega ( \\alpha )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        442,
+                                        200,
+                                        505,
+                                        212
+                                    ],
+                                    "score": 1.0,
+                                    "content": "that holds even",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                211,
+                                505,
+                                224
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        211,
+                                        392,
+                                        224
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for a (non-private) robust mean estimation with infinite samples, and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        393,
+                                        211,
+                                        411,
+                                        223
+                                    ],
+                                    "score": 0.85,
+                                    "content": "( i i i )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        411,
+                                        211,
+                                        505,
+                                        224
+                                    ],
+                                    "score": 1.0,
+                                    "content": "achieves near-optimal",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                222,
+                                505,
+                                234
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        222,
+                                        505,
+                                        234
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sample complexity matching that of a fundamental lower bound for a (non-robust) private mean",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                233,
+                                235,
+                                244
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        233,
+                                        235,
+                                        244
+                                    ],
+                                    "score": 1.0,
+                                    "content": "estimation as shown in Table 1.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        }
+                    ],
+                    "index": 11
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        246,
+                        506,
+                        295
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                245,
+                                506,
+                                259
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        245,
+                                        384,
+                                        259
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Theorem 1 (Informal Theorem 7, exponential time). Algorithm 2 is",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        384,
+                                        246,
+                                        406,
+                                        258
+                                    ],
+                                    "score": 0.9,
+                                    "content": "( \\varepsilon , \\delta )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        406,
+                                        245,
+                                        451,
+                                        259
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-DP. When",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        451,
+                                        248,
+                                        459,
+                                        257
+                                    ],
+                                    "score": 0.66,
+                                    "content": "\\alpha",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        460,
+                                        245,
+                                        506,
+                                        259
+                                    ],
+                                    "score": 1.0,
+                                    "content": "fraction of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                257,
+                                505,
+                                269
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        257,
+                                        262,
+                                        269
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the data is arbitrarily corrupted from",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        262,
+                                        259,
+                                        269,
+                                        267
+                                    ],
+                                    "score": 0.3,
+                                    "content": "n",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        270,
+                                        257,
+                                        335,
+                                        269
+                                    ],
+                                    "score": 1.0,
+                                    "content": "samples from a",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        336,
+                                        258,
+                                        342,
+                                        267
+                                    ],
+                                    "score": 0.71,
+                                    "content": "d",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        343,
+                                        257,
+                                        505,
+                                        269
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-dimensional sub-Gaussian distribution",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                268,
+                                504,
+                                284
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        268,
+                                        150,
+                                        284
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with mean",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        150,
+                                        272,
+                                        158,
+                                        282
+                                    ],
+                                    "score": 0.63,
+                                    "content": "\\mu",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        158,
+                                        268,
+                                        332,
+                                        284
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and an identity sub-Gaussian parameter, if",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        332,
+                                        268,
+                                        504,
+                                        282
+                                    ],
+                                    "score": 0.9,
+                                    "content": "\\begin{array} { r } { n = \\widetilde \\Omega ( d / \\alpha ^ { 2 } + ( d + d ^ { 1 / 2 } \\log ( 1 / \\delta ) ) / ( \\alpha \\varepsilon ) ) } \\end{array}",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                281,
+                                362,
+                                296
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        282,
+                                        213,
+                                        296
+                                    ],
+                                    "score": 1.0,
+                                    "content": "then Algorithm 2 achieves",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        214,
+                                        281,
+                                        358,
+                                        295
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\| \\hat { \\mu } - \\mu \\| _ { 2 } = O ( \\alpha \\sqrt { \\log ( 1 / \\alpha ) } ) \\ : w . h . p",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        359,
+                                        282,
+                                        362,
+                                        296
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        }
+                    ],
+                    "index": 16.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        301,
+                        506,
+                        387
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                300,
+                                506,
+                                314
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        300,
+                                        363,
+                                        314
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We introduce PRIME (PRIvate and robust Mean Estimation) in",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        363,
+                                        302,
+                                        382,
+                                        313
+                                    ],
+                                    "score": 0.85,
+                                    "content": "\\ S 2 . 3",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        383,
+                                        300,
+                                        506,
+                                        314
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with details in Algorithm 9 in",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                312,
+                                506,
+                                327
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        314,
+                                        424,
+                                        327
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Appendix E.1, to achieve computational efficiency. It requires a run-time of only",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        424,
+                                        312,
+                                        477,
+                                        327
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\widetilde { \\cal O } ( d ^ { 3 } + n d ^ { 2 } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        477,
+                                        314,
+                                        506,
+                                        327
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", but at",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                325,
+                                506,
+                                341
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        325,
+                                        210,
+                                        341
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the cost of requiring extra",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        211,
+                                        326,
+                                        230,
+                                        337
+                                    ],
+                                    "score": 0.9,
+                                    "content": "d ^ { 1 / 2 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        230,
+                                        325,
+                                        506,
+                                        341
+                                    ],
+                                    "score": 1.0,
+                                    "content": "factor larger number of samples. This cannot be improved upon with",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                338,
+                                506,
+                                351
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        338,
+                                        506,
+                                        351
+                                    ],
+                                    "score": 1.0,
+                                    "content": "current techniques since efficient robust estimators rely on the top PCA directions of the covariance",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                348,
+                                506,
+                                366
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        348,
+                                        273,
+                                        366
+                                    ],
+                                    "score": 1.0,
+                                    "content": "matrix to detect outliers. [78] showed that",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        273,
+                                        348,
+                                        307,
+                                        363
+                                    ],
+                                    "score": 0.94,
+                                    "content": "\\widetilde \\Omega ( d ^ { 3 / 2 } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        307,
+                                        348,
+                                        506,
+                                        366
+                                    ],
+                                    "score": 1.0,
+                                    "content": "samples are necessary to compute PCA directions",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                361,
+                                505,
+                                378
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        361,
+                                        175,
+                                        378
+                                    ],
+                                    "score": 1.0,
+                                    "content": "while preserving",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        175,
+                                        364,
+                                        198,
+                                        376
+                                    ],
+                                    "score": 0.89,
+                                    "content": "( \\varepsilon , \\delta )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        198,
+                                        361,
+                                        239,
+                                        378
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-DP when",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        239,
+                                        363,
+                                        306,
+                                        376
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\| x _ { i } \\| _ { 2 } = O ( { \\sqrt { d } } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        306,
+                                        361,
+                                        447,
+                                        378
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". It remains an open question if this",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        448,
+                                        362,
+                                        505,
+                                        376
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\widetilde \\Omega ( d ^ { 3 / 2 } / ( \\alpha \\varepsilon ) )",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                374,
+                                398,
+                                388
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        374,
+                                        398,
+                                        388
+                                    ],
+                                    "score": 1.0,
+                                    "content": "bottleneck is fundamental; no matching lower bound is currently known.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        }
+                    ],
+                    "index": 22
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        388,
+                        504,
+                        414
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                387,
+                                505,
+                                401
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        387,
+                                        360,
+                                        401
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Theorem 2 (Informal Theorem 6, polynomial time). PRIME is",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        360,
+                                        388,
+                                        382,
+                                        400
+                                    ],
+                                    "score": 0.89,
+                                    "content": "( \\varepsilon , \\delta )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        383,
+                                        387,
+                                        505,
+                                        401
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-DP and under the assumption",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                399,
+                                499,
+                                416
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        399,
+                                        154,
+                                        416
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of Thm.1, if",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        154,
+                                        399,
+                                        310,
+                                        414
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\begin{array} { r } { \\dot { n } = \\widetilde \\Omega ( d / \\alpha ^ { 2 } + ( d ^ { 3 / 2 } \\log ( 1 / \\delta ) ) / ( \\alpha \\varepsilon ) ) . } \\end{array}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        310,
+                                        399,
+                                        351,
+                                        416
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", achieves",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        351,
+                                        400,
+                                        496,
+                                        414
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\begin{array} { r } { \\| \\hat { \\mu } - \\mu \\| _ { 2 } = O ( \\alpha \\sqrt { \\log ( 1 / \\alpha ) } ) ~ w . h . p . } \\end{array}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        496,
+                                        399,
+                                        499,
+                                        416
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        }
+                    ],
+                    "index": 26.5
+                },
+                {
+                    "type": "table",
+                    "bbox": [
+                        110,
+                        425,
+                        501,
+                        495
+                    ],
+                    "blocks": [
+                        {
+                            "type": "table_body",
+                            "bbox": [
+                                110,
+                                425,
+                                501,
+                                495
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        110,
+                                        425,
+                                        501,
+                                        495
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                110,
+                                                425,
+                                                501,
+                                                495
+                                            ],
+                                            "score": 0.98,
+                                            "html": "<table><tr><td rowspan=1 colspan=1></td><td rowspan=1 colspan=1>Upper bound (poly-time)</td><td rowspan=1 colspan=1>Upper bound (exp-time)</td><td rowspan=1 colspan=1>Lower bound</td></tr><tr><td rowspan=1 colspan=1>(ε,δ)-DP[52]</td><td rowspan=1 colspan=1>0+dlog1/2(1/8））Qε</td><td rowspan=1 colspan=1>+）</td><td rowspan=1 colspan=1>（+）</td></tr><tr><td rowspan=1 colspan=1>α-corruption [36]</td><td rowspan=1 colspan=1>0（</td><td rowspan=1 colspan=1>（</td><td rowspan=1 colspan=1>（</td></tr><tr><td rowspan=1 colspan=1>α-corruption and(ε,δ)-DP (this paper)</td><td rowspan=1 colspan=1>（+/210g(1/8）)αε[Theorem6]</td><td rowspan=1 colspan=1>0（+d+d1/210g(1/8))Q[Theorem7]</td><td rowspan=1 colspan=1>(）[52]</td></tr></table>",
+                                            "type": "table",
+                                            "image_path": "1e64b991060acd268af93ff0942da309a127755f1575377d2afa3e229b23eccd.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 29,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        110,
+                                        425,
+                                        501,
+                                        448.3333333333333
+                                    ],
+                                    "spans": [],
+                                    "index": 28
+                                },
+                                {
+                                    "bbox": [
+                                        110,
+                                        448.3333333333333,
+                                        501,
+                                        471.66666666666663
+                                    ],
+                                    "spans": [],
+                                    "index": 29
+                                },
+                                {
+                                    "bbox": [
+                                        110,
+                                        471.66666666666663,
+                                        501,
+                                        494.99999999999994
+                                    ],
+                                    "spans": [],
+                                    "index": 30
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 29
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        505,
+                        505,
+                        556
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                505,
+                                506,
+                                519
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        505,
+                                        237,
+                                        519
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Table 1: For estimating the mean",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        237,
+                                        505,
+                                        268,
+                                        518
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\mu \\in \\mathbb { R } ^ { d }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        269,
+                                        505,
+                                        506,
+                                        519
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of a sub-Gaussian distribution with a known covariance, we",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                518,
+                                505,
+                                532
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        518,
+                                        430,
+                                        532
+                                    ],
+                                    "score": 1.0,
+                                    "content": "list the sufficient or necessary conditions on the sample sizes to achieve an error",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        430,
+                                        518,
+                                        505,
+                                        531
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\| \\hat { \\mu } - \\mu \\| _ { 2 } = \\widetilde { O } ( \\alpha )",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                530,
+                                506,
+                                544
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        530,
+                                        132,
+                                        544
+                                    ],
+                                    "score": 1.0,
+                                    "content": "under",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        132,
+                                        531,
+                                        154,
+                                        543
+                                    ],
+                                    "score": 0.91,
+                                    "content": "( \\varepsilon , \\delta )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        154,
+                                        530,
+                                        239,
+                                        544
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-DP, corruption of an",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        239,
+                                        533,
+                                        247,
+                                        541
+                                    ],
+                                    "score": 0.81,
+                                    "content": "\\alpha",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        247,
+                                        530,
+                                        369,
+                                        544
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-fraction of samples, and both. √",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        370,
+                                        530,
+                                        378,
+                                        541
+                                    ],
+                                    "score": 0.3,
+                                    "content": "\\clubsuit",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        378,
+                                        530,
+                                        506,
+                                        544
+                                    ],
+                                    "score": 1.0,
+                                    "content": "requires the distribution to be a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                542,
+                                273,
+                                557
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        542,
+                                        181,
+                                        557
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Gaussian [14] and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        181,
+                                        543,
+                                        190,
+                                        554
+                                    ],
+                                    "score": 0.4,
+                                    "content": "\\spadesuit",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        190,
+                                        542,
+                                        225,
+                                        557
+                                    ],
+                                    "score": 1.0,
+                                    "content": "requires",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        225,
+                                        542,
+                                        269,
+                                        556
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\delta \\leq \\sqrt { d } / n",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        269,
+                                        542,
+                                        273,
+                                        557
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        }
+                    ],
+                    "index": 32.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        564,
+                        506,
+                        610
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                563,
+                                507,
+                                578
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        563,
+                                        507,
+                                        578
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Heavy-tailed distributions. When samples are drawn from a distribution with a bounded covariance,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                576,
+                                507,
+                                588
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        576,
+                                        507,
+                                        588
+                                    ],
+                                    "score": 1.0,
+                                    "content": "parameters of Algorithm 2 can be modified to nearly match the optimal sample complexity of (non-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                587,
+                                505,
+                                598
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        587,
+                                        505,
+                                        598
+                                    ],
+                                    "score": 1.0,
+                                    "content": "robust) private mean estimation in Table 2. This algorithm also matches the fundamental limit on the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                597,
+                                405,
+                                611
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        597,
+                                        366,
+                                        611
+                                    ],
+                                    "score": 1.0,
+                                    "content": "accuracy of (non-private) robust estimation, which in this case is",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        366,
+                                        597,
+                                        401,
+                                        611
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\Omega ( \\alpha ^ { 1 / 2 } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        401,
+                                        597,
+                                        405,
+                                        611
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        }
+                    ],
+                    "index": 36.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        612,
+                        506,
+                        650
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                611,
+                                506,
+                                625
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        611,
+                                        453,
+                                        625
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Theorem 3 (Informal Theorem 8, exponential time). From a distribution with mean",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        453,
+                                        612,
+                                        486,
+                                        624
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\mu \\in \\mathbb { R } ^ { d }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        486,
+                                        611,
+                                        506,
+                                        625
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                624,
+                                506,
+                                636
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        624,
+                                        152,
+                                        636
+                                    ],
+                                    "score": 1.0,
+                                    "content": "covariance",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        153,
+                                        624,
+                                        178,
+                                        635
+                                    ],
+                                    "score": 0.87,
+                                    "content": "\\Sigma \\preceq \\mathbf { I }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        179,
+                                        624,
+                                        182,
+                                        636
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        182,
+                                        626,
+                                        190,
+                                        634
+                                    ],
+                                    "score": 0.26,
+                                    "content": "n",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        190,
+                                        624,
+                                        284,
+                                        636
+                                    ],
+                                    "score": 1.0,
+                                    "content": "samples are drawn and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        285,
+                                        625,
+                                        292,
+                                        634
+                                    ],
+                                    "score": 0.78,
+                                    "content": "\\alpha",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        293,
+                                        624,
+                                        440,
+                                        636
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-fraction is corrupted. Algorithm 2 is",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        440,
+                                        624,
+                                        462,
+                                        636
+                                    ],
+                                    "score": 0.89,
+                                    "content": "( \\varepsilon , \\delta )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        462,
+                                        624,
+                                        497,
+                                        636
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-DP and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        497,
+                                        624,
+                                        506,
+                                        636
+                                    ],
+                                    "score": 0.6,
+                                    "content": "i f",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                634,
+                                488,
+                                651
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        635,
+                                        334,
+                                        649
+                                    ],
+                                    "score": 0.89,
+                                    "content": "n = \\widetilde \\Omega ( ( d + d ^ { 1 / 2 } \\log ( 1 / \\delta ) ) / ( \\alpha \\varepsilon ) + d ^ { 1 / 2 } \\log ^ { 3 / 2 } ( 1 / \\delta ) / \\varepsilon )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        335,
+                                        634,
+                                        373,
+                                        651
+                                    ],
+                                    "score": 1.0,
+                                    "content": "achieves",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        373,
+                                        636,
+                                        484,
+                                        649
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\| \\hat { \\mu } - \\mu \\| _ { 2 } = O ( \\alpha ^ { 1 / 2 } ) \\ : w . h . p .",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        484,
+                                        634,
+                                        488,
+                                        651
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        }
+                    ],
+                    "index": 40
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        655,
+                        503,
+                        691
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                654,
+                                505,
+                                670
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        654,
+                                        505,
+                                        670
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The proposed PRIME-HT for covariance bounded distributions achieve computational efficiency",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                666,
+                                506,
+                                680
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        666,
+                                        234,
+                                        680
+                                    ],
+                                    "score": 1.0,
+                                    "content": "at the cost of an extra factor of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        234,
+                                        667,
+                                        252,
+                                        678
+                                    ],
+                                    "score": 0.9,
+                                    "content": "d ^ { 1 / 2 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        253,
+                                        666,
+                                        506,
+                                        680
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in sample size. This bottleneck is also due to DP PCA, and it",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                679,
+                                387,
+                                692
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        679,
+                                        387,
+                                        692
+                                    ],
+                                    "score": 1.0,
+                                    "content": "remains open whether this gap can be closed by an efficient estimator.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        }
+                    ],
+                    "index": 43
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        692,
+                        504,
+                        718
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                691,
+                                505,
+                                705
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        691,
+                                        407,
+                                        705
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Theorem 4 (Informal Theorem 9, polynomial time). PRIME-HT is",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        407,
+                                        692,
+                                        447,
+                                        705
+                                    ],
+                                    "score": 0.66,
+                                    "content": "( \\varepsilon , \\delta ) – D P",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        448,
+                                        691,
+                                        480,
+                                        705
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and if",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        481,
+                                        693,
+                                        505,
+                                        704
+                                    ],
+                                    "score": 0.84,
+                                    "content": "n \\_ =",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                703,
+                                497,
+                                720
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        703,
+                                        208,
+                                        718
+                                    ],
+                                    "score": 0.9,
+                                    "content": "\\widetilde \\Omega ( ( d ^ { 3 / 2 } \\log ( 1 / \\delta ) ) / ( \\alpha \\varepsilon ) )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        209,
+                                        703,
+                                        247,
+                                        720
+                                    ],
+                                    "score": 1.0,
+                                    "content": "achieves",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        247,
+                                        704,
+                                        358,
+                                        718
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\| \\hat { \\mu } - \\mu \\| _ { 2 } = O ( \\alpha ^ { 1 / 2 } ) \\ : w . h . p",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        358,
+                                        703,
+                                        497,
+                                        720
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". under the assumptions of Thm. 3.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        }
+                    ],
+                    "index": 45.5
+                }
+            ],
+            "page_idx": 1,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        302,
+                        741,
+                        309,
+                        750
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                301,
+                                740,
+                                310,
+                                753
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        301,
+                                        740,
+                                        310,
+                                        753
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        72,
+                        505,
+                        161
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                73,
+                                505,
+                                85
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        73,
+                                        505,
+                                        85
+                                    ],
+                                    "score": 1.0,
+                                    "content": "One might be misled into thinking that privacy ensures robustness since DP guarantees that a single",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                84,
+                                505,
+                                95
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        84,
+                                        505,
+                                        95
+                                    ],
+                                    "score": 1.0,
+                                    "content": "outlier cannot change the estimation too much. This intuition is true only in a low dimension; each",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                94,
+                                506,
+                                107
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        94,
+                                        506,
+                                        107
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sample has to be an obvious outlier to significantly change the mean. However, in a high dimension,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                106,
+                                505,
+                                118
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        106,
+                                        505,
+                                        118
+                                    ],
+                                    "score": 1.0,
+                                    "content": "each corrupted data point can look perfectly uncorrupted but still shift the mean significant when",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                117,
+                                506,
+                                130
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        117,
+                                        506,
+                                        130
+                                    ],
+                                    "score": 1.0,
+                                    "content": "colluding together (e.g., see Fig. 1). Focusing on the canonical problem of mean estimation, we",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                127,
+                                505,
+                                140
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        127,
+                                        505,
+                                        140
+                                    ],
+                                    "score": 1.0,
+                                    "content": "introduce novel algorithms that achieve robustness and privacy simultaneously even when a fraction",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                138,
+                                505,
+                                150
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        138,
+                                        505,
+                                        150
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of data is corrupted arbitrarily. For such algorithms, there is a fundamental question of interest: do",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                149,
+                                502,
+                                161
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        149,
+                                        502,
+                                        161
+                                    ],
+                                    "score": 1.0,
+                                    "content": "we need more samples to make private mean estimation also robust against adversarial corruption?",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        }
+                    ],
+                    "index": 3.5,
+                    "bbox_fs": [
+                        105,
+                        73,
+                        506,
+                        161
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        165,
+                        505,
+                        244
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                164,
+                                506,
+                                178
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        164,
+                                        506,
+                                        178
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Sub-Gaussian distributions. If we can afford exponential run-time in the dimension, robustness",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                176,
+                                505,
+                                189
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        176,
+                                        505,
+                                        189
+                                    ],
+                                    "score": 1.0,
+                                    "content": "can be achieved without extra cost in sample complexity. We introduce a novel estimator that (i)",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                186,
+                                506,
+                                201
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        186,
+                                        140,
+                                        201
+                                    ],
+                                    "score": 1.0,
+                                    "content": "satisfies",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        140,
+                                        187,
+                                        162,
+                                        199
+                                    ],
+                                    "score": 0.91,
+                                    "content": "( \\varepsilon , \\delta )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        163,
+                                        186,
+                                        181,
+                                        201
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-DP,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        181,
+                                        187,
+                                        196,
+                                        199
+                                    ],
+                                    "score": 0.78,
+                                    "content": "( i i )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        196,
+                                        186,
+                                        352,
+                                        201
+                                    ],
+                                    "score": 1.0,
+                                    "content": "achieves near-optimal robustness under",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        352,
+                                        189,
+                                        360,
+                                        198
+                                    ],
+                                    "score": 0.79,
+                                    "content": "\\alpha",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        360,
+                                        186,
+                                        506,
+                                        201
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-fraction of corrupted data, achieving",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                199,
+                                505,
+                                213
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        200,
+                                        154,
+                                        212
+                                    ],
+                                    "score": 1.0,
+                                    "content": "accuracy of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        155,
+                                        199,
+                                        223,
+                                        213
+                                    ],
+                                    "score": 0.93,
+                                    "content": "O ( \\alpha \\sqrt { \\log ( 1 / \\alpha ) } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        223,
+                                        200,
+                                        420,
+                                        212
+                                    ],
+                                    "score": 1.0,
+                                    "content": "nearly matching the fundamental lower bound of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        420,
+                                        199,
+                                        442,
+                                        212
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\Omega ( \\alpha )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        442,
+                                        200,
+                                        505,
+                                        212
+                                    ],
+                                    "score": 1.0,
+                                    "content": "that holds even",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                211,
+                                505,
+                                224
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        211,
+                                        392,
+                                        224
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for a (non-private) robust mean estimation with infinite samples, and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        393,
+                                        211,
+                                        411,
+                                        223
+                                    ],
+                                    "score": 0.85,
+                                    "content": "( i i i )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        411,
+                                        211,
+                                        505,
+                                        224
+                                    ],
+                                    "score": 1.0,
+                                    "content": "achieves near-optimal",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                222,
+                                505,
+                                234
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        222,
+                                        505,
+                                        234
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sample complexity matching that of a fundamental lower bound for a (non-robust) private mean",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                233,
+                                235,
+                                244
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        233,
+                                        235,
+                                        244
+                                    ],
+                                    "score": 1.0,
+                                    "content": "estimation as shown in Table 1.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        }
+                    ],
+                    "index": 11,
+                    "bbox_fs": [
+                        105,
+                        164,
+                        506,
+                        244
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        246,
+                        506,
+                        295
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                245,
+                                506,
+                                259
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        245,
+                                        384,
+                                        259
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Theorem 1 (Informal Theorem 7, exponential time). Algorithm 2 is",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        384,
+                                        246,
+                                        406,
+                                        258
+                                    ],
+                                    "score": 0.9,
+                                    "content": "( \\varepsilon , \\delta )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        406,
+                                        245,
+                                        451,
+                                        259
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-DP. When",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        451,
+                                        248,
+                                        459,
+                                        257
+                                    ],
+                                    "score": 0.66,
+                                    "content": "\\alpha",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        460,
+                                        245,
+                                        506,
+                                        259
+                                    ],
+                                    "score": 1.0,
+                                    "content": "fraction of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                257,
+                                505,
+                                269
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        257,
+                                        262,
+                                        269
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the data is arbitrarily corrupted from",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        262,
+                                        259,
+                                        269,
+                                        267
+                                    ],
+                                    "score": 0.3,
+                                    "content": "n",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        270,
+                                        257,
+                                        335,
+                                        269
+                                    ],
+                                    "score": 1.0,
+                                    "content": "samples from a",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        336,
+                                        258,
+                                        342,
+                                        267
+                                    ],
+                                    "score": 0.71,
+                                    "content": "d",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        343,
+                                        257,
+                                        505,
+                                        269
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-dimensional sub-Gaussian distribution",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                268,
+                                504,
+                                284
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        268,
+                                        150,
+                                        284
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with mean",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        150,
+                                        272,
+                                        158,
+                                        282
+                                    ],
+                                    "score": 0.63,
+                                    "content": "\\mu",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        158,
+                                        268,
+                                        332,
+                                        284
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and an identity sub-Gaussian parameter, if",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        332,
+                                        268,
+                                        504,
+                                        282
+                                    ],
+                                    "score": 0.9,
+                                    "content": "\\begin{array} { r } { n = \\widetilde \\Omega ( d / \\alpha ^ { 2 } + ( d + d ^ { 1 / 2 } \\log ( 1 / \\delta ) ) / ( \\alpha \\varepsilon ) ) } \\end{array}",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                281,
+                                362,
+                                296
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        282,
+                                        213,
+                                        296
+                                    ],
+                                    "score": 1.0,
+                                    "content": "then Algorithm 2 achieves",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        214,
+                                        281,
+                                        358,
+                                        295
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\| \\hat { \\mu } - \\mu \\| _ { 2 } = O ( \\alpha \\sqrt { \\log ( 1 / \\alpha ) } ) \\ : w . h . p",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        359,
+                                        282,
+                                        362,
+                                        296
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        }
+                    ],
+                    "index": 16.5,
+                    "bbox_fs": [
+                        105,
+                        245,
+                        506,
+                        296
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        301,
+                        506,
+                        387
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                300,
+                                506,
+                                314
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        300,
+                                        363,
+                                        314
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We introduce PRIME (PRIvate and robust Mean Estimation) in",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        363,
+                                        302,
+                                        382,
+                                        313
+                                    ],
+                                    "score": 0.85,
+                                    "content": "\\ S 2 . 3",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        383,
+                                        300,
+                                        506,
+                                        314
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with details in Algorithm 9 in",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                312,
+                                506,
+                                327
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        314,
+                                        424,
+                                        327
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Appendix E.1, to achieve computational efficiency. It requires a run-time of only",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        424,
+                                        312,
+                                        477,
+                                        327
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\widetilde { \\cal O } ( d ^ { 3 } + n d ^ { 2 } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        477,
+                                        314,
+                                        506,
+                                        327
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", but at",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                325,
+                                506,
+                                341
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        325,
+                                        210,
+                                        341
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the cost of requiring extra",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        211,
+                                        326,
+                                        230,
+                                        337
+                                    ],
+                                    "score": 0.9,
+                                    "content": "d ^ { 1 / 2 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        230,
+                                        325,
+                                        506,
+                                        341
+                                    ],
+                                    "score": 1.0,
+                                    "content": "factor larger number of samples. This cannot be improved upon with",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                338,
+                                506,
+                                351
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        338,
+                                        506,
+                                        351
+                                    ],
+                                    "score": 1.0,
+                                    "content": "current techniques since efficient robust estimators rely on the top PCA directions of the covariance",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                348,
+                                506,
+                                366
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        348,
+                                        273,
+                                        366
+                                    ],
+                                    "score": 1.0,
+                                    "content": "matrix to detect outliers. [78] showed that",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        273,
+                                        348,
+                                        307,
+                                        363
+                                    ],
+                                    "score": 0.94,
+                                    "content": "\\widetilde \\Omega ( d ^ { 3 / 2 } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        307,
+                                        348,
+                                        506,
+                                        366
+                                    ],
+                                    "score": 1.0,
+                                    "content": "samples are necessary to compute PCA directions",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                361,
+                                505,
+                                378
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        361,
+                                        175,
+                                        378
+                                    ],
+                                    "score": 1.0,
+                                    "content": "while preserving",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        175,
+                                        364,
+                                        198,
+                                        376
+                                    ],
+                                    "score": 0.89,
+                                    "content": "( \\varepsilon , \\delta )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        198,
+                                        361,
+                                        239,
+                                        378
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-DP when",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        239,
+                                        363,
+                                        306,
+                                        376
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\| x _ { i } \\| _ { 2 } = O ( { \\sqrt { d } } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        306,
+                                        361,
+                                        447,
+                                        378
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". It remains an open question if this",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        448,
+                                        362,
+                                        505,
+                                        376
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\widetilde \\Omega ( d ^ { 3 / 2 } / ( \\alpha \\varepsilon ) )",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                374,
+                                398,
+                                388
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        374,
+                                        398,
+                                        388
+                                    ],
+                                    "score": 1.0,
+                                    "content": "bottleneck is fundamental; no matching lower bound is currently known.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        }
+                    ],
+                    "index": 22,
+                    "bbox_fs": [
+                        104,
+                        300,
+                        506,
+                        388
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        388,
+                        504,
+                        414
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                387,
+                                505,
+                                401
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        387,
+                                        360,
+                                        401
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Theorem 2 (Informal Theorem 6, polynomial time). PRIME is",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        360,
+                                        388,
+                                        382,
+                                        400
+                                    ],
+                                    "score": 0.89,
+                                    "content": "( \\varepsilon , \\delta )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        383,
+                                        387,
+                                        505,
+                                        401
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-DP and under the assumption",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                399,
+                                499,
+                                416
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        399,
+                                        154,
+                                        416
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of Thm.1, if",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        154,
+                                        399,
+                                        310,
+                                        414
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\begin{array} { r } { \\dot { n } = \\widetilde \\Omega ( d / \\alpha ^ { 2 } + ( d ^ { 3 / 2 } \\log ( 1 / \\delta ) ) / ( \\alpha \\varepsilon ) ) . } \\end{array}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        310,
+                                        399,
+                                        351,
+                                        416
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", achieves",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        351,
+                                        400,
+                                        496,
+                                        414
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\begin{array} { r } { \\| \\hat { \\mu } - \\mu \\| _ { 2 } = O ( \\alpha \\sqrt { \\log ( 1 / \\alpha ) } ) ~ w . h . p . } \\end{array}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        496,
+                                        399,
+                                        499,
+                                        416
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        }
+                    ],
+                    "index": 26.5,
+                    "bbox_fs": [
+                        105,
+                        387,
+                        505,
+                        416
+                    ]
+                },
+                {
+                    "type": "table",
+                    "bbox": [
+                        110,
+                        425,
+                        501,
+                        495
+                    ],
+                    "blocks": [
+                        {
+                            "type": "table_body",
+                            "bbox": [
+                                110,
+                                425,
+                                501,
+                                495
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        110,
+                                        425,
+                                        501,
+                                        495
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                110,
+                                                425,
+                                                501,
+                                                495
+                                            ],
+                                            "score": 0.98,
+                                            "html": "<table><tr><td rowspan=1 colspan=1></td><td rowspan=1 colspan=1>Upper bound (poly-time)</td><td rowspan=1 colspan=1>Upper bound (exp-time)</td><td rowspan=1 colspan=1>Lower bound</td></tr><tr><td rowspan=1 colspan=1>(ε,δ)-DP[52]</td><td rowspan=1 colspan=1>0+dlog1/2(1/8））Qε</td><td rowspan=1 colspan=1>+）</td><td rowspan=1 colspan=1>（+）</td></tr><tr><td rowspan=1 colspan=1>α-corruption [36]</td><td rowspan=1 colspan=1>0（</td><td rowspan=1 colspan=1>（</td><td rowspan=1 colspan=1>（</td></tr><tr><td rowspan=1 colspan=1>α-corruption and(ε,δ)-DP (this paper)</td><td rowspan=1 colspan=1>（+/210g(1/8）)αε[Theorem6]</td><td rowspan=1 colspan=1>0（+d+d1/210g(1/8))Q[Theorem7]</td><td rowspan=1 colspan=1>(）[52]</td></tr></table>",
+                                            "type": "table",
+                                            "image_path": "1e64b991060acd268af93ff0942da309a127755f1575377d2afa3e229b23eccd.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 29,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        110,
+                                        425,
+                                        501,
+                                        448.3333333333333
+                                    ],
+                                    "spans": [],
+                                    "index": 28
+                                },
+                                {
+                                    "bbox": [
+                                        110,
+                                        448.3333333333333,
+                                        501,
+                                        471.66666666666663
+                                    ],
+                                    "spans": [],
+                                    "index": 29
+                                },
+                                {
+                                    "bbox": [
+                                        110,
+                                        471.66666666666663,
+                                        501,
+                                        494.99999999999994
+                                    ],
+                                    "spans": [],
+                                    "index": 30
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 29
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        505,
+                        505,
+                        556
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                505,
+                                506,
+                                519
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        505,
+                                        237,
+                                        519
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Table 1: For estimating the mean",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        237,
+                                        505,
+                                        268,
+                                        518
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\mu \\in \\mathbb { R } ^ { d }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        269,
+                                        505,
+                                        506,
+                                        519
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of a sub-Gaussian distribution with a known covariance, we",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                518,
+                                505,
+                                532
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        518,
+                                        430,
+                                        532
+                                    ],
+                                    "score": 1.0,
+                                    "content": "list the sufficient or necessary conditions on the sample sizes to achieve an error",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        430,
+                                        518,
+                                        505,
+                                        531
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\| \\hat { \\mu } - \\mu \\| _ { 2 } = \\widetilde { O } ( \\alpha )",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                530,
+                                506,
+                                544
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        530,
+                                        132,
+                                        544
+                                    ],
+                                    "score": 1.0,
+                                    "content": "under",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        132,
+                                        531,
+                                        154,
+                                        543
+                                    ],
+                                    "score": 0.91,
+                                    "content": "( \\varepsilon , \\delta )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        154,
+                                        530,
+                                        239,
+                                        544
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-DP, corruption of an",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        239,
+                                        533,
+                                        247,
+                                        541
+                                    ],
+                                    "score": 0.81,
+                                    "content": "\\alpha",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        247,
+                                        530,
+                                        369,
+                                        544
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-fraction of samples, and both. √",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        370,
+                                        530,
+                                        378,
+                                        541
+                                    ],
+                                    "score": 0.3,
+                                    "content": "\\clubsuit",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        378,
+                                        530,
+                                        506,
+                                        544
+                                    ],
+                                    "score": 1.0,
+                                    "content": "requires the distribution to be a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                542,
+                                273,
+                                557
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        542,
+                                        181,
+                                        557
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Gaussian [14] and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        181,
+                                        543,
+                                        190,
+                                        554
+                                    ],
+                                    "score": 0.4,
+                                    "content": "\\spadesuit",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        190,
+                                        542,
+                                        225,
+                                        557
+                                    ],
+                                    "score": 1.0,
+                                    "content": "requires",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        225,
+                                        542,
+                                        269,
+                                        556
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\delta \\leq \\sqrt { d } / n",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        269,
+                                        542,
+                                        273,
+                                        557
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        }
+                    ],
+                    "index": 32.5,
+                    "bbox_fs": [
+                        105,
+                        505,
+                        506,
+                        557
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        564,
+                        506,
+                        610
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                563,
+                                507,
+                                578
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        563,
+                                        507,
+                                        578
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Heavy-tailed distributions. When samples are drawn from a distribution with a bounded covariance,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                576,
+                                507,
+                                588
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        576,
+                                        507,
+                                        588
+                                    ],
+                                    "score": 1.0,
+                                    "content": "parameters of Algorithm 2 can be modified to nearly match the optimal sample complexity of (non-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                587,
+                                505,
+                                598
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        587,
+                                        505,
+                                        598
+                                    ],
+                                    "score": 1.0,
+                                    "content": "robust) private mean estimation in Table 2. This algorithm also matches the fundamental limit on the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                597,
+                                405,
+                                611
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        597,
+                                        366,
+                                        611
+                                    ],
+                                    "score": 1.0,
+                                    "content": "accuracy of (non-private) robust estimation, which in this case is",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        366,
+                                        597,
+                                        401,
+                                        611
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\Omega ( \\alpha ^ { 1 / 2 } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        401,
+                                        597,
+                                        405,
+                                        611
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        }
+                    ],
+                    "index": 36.5,
+                    "bbox_fs": [
+                        105,
+                        563,
+                        507,
+                        611
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        612,
+                        506,
+                        650
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                611,
+                                506,
+                                625
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        611,
+                                        453,
+                                        625
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Theorem 3 (Informal Theorem 8, exponential time). From a distribution with mean",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        453,
+                                        612,
+                                        486,
+                                        624
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\mu \\in \\mathbb { R } ^ { d }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        486,
+                                        611,
+                                        506,
+                                        625
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                624,
+                                506,
+                                636
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        624,
+                                        152,
+                                        636
+                                    ],
+                                    "score": 1.0,
+                                    "content": "covariance",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        153,
+                                        624,
+                                        178,
+                                        635
+                                    ],
+                                    "score": 0.87,
+                                    "content": "\\Sigma \\preceq \\mathbf { I }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        179,
+                                        624,
+                                        182,
+                                        636
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        182,
+                                        626,
+                                        190,
+                                        634
+                                    ],
+                                    "score": 0.26,
+                                    "content": "n",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        190,
+                                        624,
+                                        284,
+                                        636
+                                    ],
+                                    "score": 1.0,
+                                    "content": "samples are drawn and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        285,
+                                        625,
+                                        292,
+                                        634
+                                    ],
+                                    "score": 0.78,
+                                    "content": "\\alpha",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        293,
+                                        624,
+                                        440,
+                                        636
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-fraction is corrupted. Algorithm 2 is",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        440,
+                                        624,
+                                        462,
+                                        636
+                                    ],
+                                    "score": 0.89,
+                                    "content": "( \\varepsilon , \\delta )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        462,
+                                        624,
+                                        497,
+                                        636
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-DP and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        497,
+                                        624,
+                                        506,
+                                        636
+                                    ],
+                                    "score": 0.6,
+                                    "content": "i f",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                634,
+                                488,
+                                651
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        635,
+                                        334,
+                                        649
+                                    ],
+                                    "score": 0.89,
+                                    "content": "n = \\widetilde \\Omega ( ( d + d ^ { 1 / 2 } \\log ( 1 / \\delta ) ) / ( \\alpha \\varepsilon ) + d ^ { 1 / 2 } \\log ^ { 3 / 2 } ( 1 / \\delta ) / \\varepsilon )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        335,
+                                        634,
+                                        373,
+                                        651
+                                    ],
+                                    "score": 1.0,
+                                    "content": "achieves",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        373,
+                                        636,
+                                        484,
+                                        649
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\| \\hat { \\mu } - \\mu \\| _ { 2 } = O ( \\alpha ^ { 1 / 2 } ) \\ : w . h . p .",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        484,
+                                        634,
+                                        488,
+                                        651
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        }
+                    ],
+                    "index": 40,
+                    "bbox_fs": [
+                        105,
+                        611,
+                        506,
+                        651
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        655,
+                        503,
+                        691
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                654,
+                                505,
+                                670
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        654,
+                                        505,
+                                        670
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The proposed PRIME-HT for covariance bounded distributions achieve computational efficiency",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                666,
+                                506,
+                                680
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        666,
+                                        234,
+                                        680
+                                    ],
+                                    "score": 1.0,
+                                    "content": "at the cost of an extra factor of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        234,
+                                        667,
+                                        252,
+                                        678
+                                    ],
+                                    "score": 0.9,
+                                    "content": "d ^ { 1 / 2 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        253,
+                                        666,
+                                        506,
+                                        680
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in sample size. This bottleneck is also due to DP PCA, and it",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                679,
+                                387,
+                                692
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        679,
+                                        387,
+                                        692
+                                    ],
+                                    "score": 1.0,
+                                    "content": "remains open whether this gap can be closed by an efficient estimator.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        }
+                    ],
+                    "index": 43,
+                    "bbox_fs": [
+                        105,
+                        654,
+                        506,
+                        692
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        692,
+                        504,
+                        718
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                691,
+                                505,
+                                705
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        691,
+                                        407,
+                                        705
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Theorem 4 (Informal Theorem 9, polynomial time). PRIME-HT is",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        407,
+                                        692,
+                                        447,
+                                        705
+                                    ],
+                                    "score": 0.66,
+                                    "content": "( \\varepsilon , \\delta ) – D P",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        448,
+                                        691,
+                                        480,
+                                        705
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and if",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        481,
+                                        693,
+                                        505,
+                                        704
+                                    ],
+                                    "score": 0.84,
+                                    "content": "n \\_ =",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                703,
+                                497,
+                                720
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        703,
+                                        208,
+                                        718
+                                    ],
+                                    "score": 0.9,
+                                    "content": "\\widetilde \\Omega ( ( d ^ { 3 / 2 } \\log ( 1 / \\delta ) ) / ( \\alpha \\varepsilon ) )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        209,
+                                        703,
+                                        247,
+                                        720
+                                    ],
+                                    "score": 1.0,
+                                    "content": "achieves",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        247,
+                                        704,
+                                        358,
+                                        718
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\| \\hat { \\mu } - \\mu \\| _ { 2 } = O ( \\alpha ^ { 1 / 2 } ) \\ : w . h . p",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        358,
+                                        703,
+                                        497,
+                                        720
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". under the assumptions of Thm. 3.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        }
+                    ],
+                    "index": 45.5,
+                    "bbox_fs": [
+                        106,
+                        691,
+                        505,
+                        720
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "table",
+                    "bbox": [
+                        113,
+                        70,
+                        498,
+                        140
+                    ],
+                    "blocks": [
+                        {
+                            "type": "table_body",
+                            "bbox": [
+                                113,
+                                70,
+                                498,
+                                140
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        113,
+                                        70,
+                                        498,
+                                        140
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                113,
+                                                70,
+                                                498,
+                                                140
+                                            ],
+                                            "score": 0.98,
+                                            "html": "<table><tr><td rowspan=1 colspan=1></td><td rowspan=1 colspan=1>Upper bound (poly-time)</td><td rowspan=1 colspan=1>Upper bound (exp-time)</td><td rowspan=1 colspan=1>Lower bound</td></tr><tr><td rowspan=1 colspan=1>(ε,δ)-DP [54]</td><td rowspan=1 colspan=1>O（d10g1/2（1/）Qε</td><td rowspan=1 colspan=1>O（d10g1/2（1/0））Qε</td><td rowspan=1 colspan=1>2（）</td></tr><tr><td rowspan=1 colspan=1>α-corruption [36]</td><td rowspan=1 colspan=1>）</td><td rowspan=1 colspan=1></td><td rowspan=1 colspan=1>(）</td></tr><tr><td rowspan=1 colspan=1>α-corruption and(ε,δ)-DP (this paper)</td><td rowspan=1 colspan=1>0（(/210g(1/8))αε[Theorem 9]</td><td rowspan=1 colspan=1>（+d1721g/2(18）αε[Theorem 8]</td><td rowspan=1 colspan=1>(([54])</td></tr></table>",
+                                            "type": "table",
+                                            "image_path": "191d63f4ee021bb7fbfe8805bb187077f698af2e6e124ee5962d88fbe956348f.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 1,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        113,
+                                        70,
+                                        498,
+                                        93.33333333333333
+                                    ],
+                                    "spans": [],
+                                    "index": 0
+                                },
+                                {
+                                    "bbox": [
+                                        113,
+                                        93.33333333333333,
+                                        498,
+                                        116.66666666666666
+                                    ],
+                                    "spans": [],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        113,
+                                        116.66666666666666,
+                                        498,
+                                        140.0
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 1
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        150,
+                        505,
+                        186
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                150,
+                                505,
+                                163
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        151,
+                                        240,
+                                        163
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Table 2: For estimating the mean",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        240,
+                                        150,
+                                        271,
+                                        163
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\mu \\in \\mathbb { R } ^ { d }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        272,
+                                        151,
+                                        505,
+                                        163
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of a covariance bounded distribution, we list the sufficient",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                161,
+                                507,
+                                177
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        161,
+                                        352,
+                                        177
+                                    ],
+                                    "score": 1.0,
+                                    "content": "or necessary conditions on the sample size to achieve an error",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        352,
+                                        162,
+                                        439,
+                                        175
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\| \\hat { \\mu } - \\mu \\| _ { 2 } = O ( \\alpha ^ { 1 / 2 } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        439,
+                                        161,
+                                        465,
+                                        177
+                                    ],
+                                    "score": 1.0,
+                                    "content": "under",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        466,
+                                        163,
+                                        487,
+                                        175
+                                    ],
+                                    "score": 0.91,
+                                    "content": "( \\varepsilon , \\delta )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        488,
+                                        161,
+                                        507,
+                                        177
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-DP,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                174,
+                                303,
+                                186
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        174,
+                                        173,
+                                        186
+                                    ],
+                                    "score": 1.0,
+                                    "content": "corruption of an",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        173,
+                                        176,
+                                        181,
+                                        184
+                                    ],
+                                    "score": 0.78,
+                                    "content": "\\alpha",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        181,
+                                        174,
+                                        303,
+                                        186
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-fraction of samples, and both.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        }
+                    ],
+                    "index": 4
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        204,
+                        231,
+                        216
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                203,
+                                231,
+                                217
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        203,
+                                        231,
+                                        217
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1.1 Technical contributions",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        }
+                    ],
+                    "index": 6
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        223,
+                        506,
+                        302
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                224,
+                                506,
+                                237
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        224,
+                                        320,
+                                        237
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We introduce PRIME which simultaneously achieves",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        320,
+                                        225,
+                                        342,
+                                        236
+                                    ],
+                                    "score": 0.91,
+                                    "content": "( \\varepsilon , \\delta )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        343,
+                                        224,
+                                        451,
+                                        237
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-DP and robustness against",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        451,
+                                        226,
+                                        459,
+                                        235
+                                    ],
+                                    "score": 0.79,
+                                    "content": "\\alpha",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        459,
+                                        224,
+                                        506,
+                                        237
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-fraction of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                234,
+                                507,
+                                249
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        234,
+                                        507,
+                                        249
+                                    ],
+                                    "score": 1.0,
+                                    "content": "corruption. A major challenge in making a standard filter-based robust estimation algorithm (e.g.,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                246,
+                                506,
+                                258
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        246,
+                                        506,
+                                        258
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[30]) private is the high sensitivity of the filtered set that we pass from one iteration to the next.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                257,
+                                505,
+                                270
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        257,
+                                        505,
+                                        270
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We propose a new framework which makes private only the statistics of the set, hence significantly",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                267,
+                                506,
+                                281
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        267,
+                                        506,
+                                        281
+                                    ],
+                                    "score": 1.0,
+                                    "content": "reducing the sensitivity. Our major innovation is a tight analysis of the end-to-end sensitivity of this",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                278,
+                                506,
+                                292
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        278,
+                                        506,
+                                        292
+                                    ],
+                                    "score": 1.0,
+                                    "content": "multiple interactive accesses to the database. This is critical in achieving robustness while preserving",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                290,
+                                497,
+                                303
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        290,
+                                        497,
+                                        303
+                                    ],
+                                    "score": 1.0,
+                                    "content": "privacy and is also of independent interest in making general iterative filtering algorithms private.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        }
+                    ],
+                    "index": 10
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        306,
+                        505,
+                        385
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                305,
+                                505,
+                                320
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        305,
+                                        390,
+                                        320
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The classical filter approach (see, e.g. [30]) needs to access the database √",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        391,
+                                        306,
+                                        413,
+                                        318
+                                    ],
+                                    "score": 0.91,
+                                    "content": "O ( d )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        413,
+                                        305,
+                                        505,
+                                        320
+                                    ],
+                                    "score": 1.0,
+                                    "content": "times, which brings an",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                317,
+                                506,
+                                332
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        317,
+                                        128,
+                                        332
+                                    ],
+                                    "score": 1.0,
+                                    "content": "extra",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        129,
+                                        317,
+                                        159,
+                                        331
+                                    ],
+                                    "score": 0.93,
+                                    "content": "O ( { \\sqrt { d } } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        159,
+                                        317,
+                                        506,
+                                        332
+                                    ],
+                                    "score": 1.0,
+                                    "content": "factor in the sample complexity due to DP composition. In order to reduce the iteration",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                330,
+                                504,
+                                343
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        330,
+                                        504,
+                                        343
+                                    ],
+                                    "score": 1.0,
+                                    "content": "complexity, following the approach in [36], we propose filtering multiple directions simultaneously",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                340,
+                                505,
+                                353
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        340,
+                                        505,
+                                        353
+                                    ],
+                                    "score": 1.0,
+                                    "content": "using a new score based on the matrix multiplicative weights (MMW). In order to privatize the MMW",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                351,
+                                504,
+                                363
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        351,
+                                        504,
+                                        363
+                                    ],
+                                    "score": 1.0,
+                                    "content": "filter, our major innovation is a novel adaptive filtering algorithm DPTHRESHOLD(·) that outputs",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                363,
+                                505,
+                                375
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        363,
+                                        505,
+                                        375
+                                    ],
+                                    "score": 1.0,
+                                    "content": "a single private threshold which guarantees sufficient progress at every iteration. This brings the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                372,
+                                331,
+                                387
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        372,
+                                        245,
+                                        387
+                                    ],
+                                    "score": 1.0,
+                                    "content": "number of database accesses from",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        245,
+                                        373,
+                                        267,
+                                        385
+                                    ],
+                                    "score": 0.92,
+                                    "content": "O ( d )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        267,
+                                        372,
+                                        279,
+                                        387
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        279,
+                                        373,
+                                        325,
+                                        385
+                                    ],
+                                    "score": 0.92,
+                                    "content": "O ( ( \\log d ) ^ { 2 } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        326,
+                                        372,
+                                        331,
+                                        387
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        }
+                    ],
+                    "index": 17
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        390,
+                        505,
+                        457
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                389,
+                                506,
+                                404
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        389,
+                                        313,
+                                        404
+                                    ],
+                                    "score": 1.0,
+                                    "content": "One downside of PRIME is that it requires an extra",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        313,
+                                        389,
+                                        332,
+                                        401
+                                    ],
+                                    "score": 0.9,
+                                    "content": "d ^ { 1 / 2 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        332,
+                                        389,
+                                        506,
+                                        404
+                                    ],
+                                    "score": 1.0,
+                                    "content": "factor in the sample complexity, compared",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                402,
+                                505,
+                                414
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        402,
+                                        505,
+                                        414
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to known lower bounds for (non-robust) DP mean estimation. To investigate whether this is also",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                413,
+                                504,
+                                425
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        413,
+                                        492,
+                                        425
+                                    ],
+                                    "score": 1.0,
+                                    "content": "necessary, we propose a sample optimal exponential time robust mean estimation algorithm in",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        492,
+                                        413,
+                                        504,
+                                        424
+                                    ],
+                                    "score": 0.8,
+                                    "content": "\\ S 4",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                424,
+                                505,
+                                437
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        424,
+                                        505,
+                                        437
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and prove that there is no extra statistical cost to jointly requiring privacy and robustness. Our major",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                434,
+                                506,
+                                448
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        434,
+                                        506,
+                                        448
+                                    ],
+                                    "score": 1.0,
+                                    "content": "technical innovations is in using resilience property of the dataset to not only find robust mean (which",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                446,
+                                436,
+                                458
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        446,
+                                        436,
+                                        458
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is the typical use case of resilience) but also bound sensitivity of that robust mean.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        }
+                    ],
+                    "index": 23.5
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        469,
+                        302,
+                        482
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                467,
+                                303,
+                                484
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        467,
+                                        303,
+                                        484
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1.2 Preliminary on differential privacy (DP)",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        }
+                    ],
+                    "index": 27
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        489,
+                        498,
+                        501
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                487,
+                                499,
+                                504
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        487,
+                                        499,
+                                        504
+                                    ],
+                                    "score": 1.0,
+                                    "content": "DP is a formal metric for measuring privacy leakage when a dataset is accessed with a query [37].",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        }
+                    ],
+                    "index": 28
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        504,
+                        505,
+                        552
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                503,
+                                506,
+                                521
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        503,
+                                        245,
+                                        521
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Definition 1.1. Given two datasets",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        246,
+                                        505,
+                                        298,
+                                        518
+                                    ],
+                                    "score": 0.91,
+                                    "content": "S = \\{ x _ { i } \\} _ { i = 1 } ^ { n }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        299,
+                                        503,
+                                        316,
+                                        521
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        316,
+                                        503,
+                                        372,
+                                        518
+                                    ],
+                                    "score": 0.92,
+                                    "content": "S ^ { \\prime } = \\{ x _ { i } ^ { \\prime } \\} _ { i = 1 } ^ { n ^ { \\prime } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        372,
+                                        503,
+                                        404,
+                                        521
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", we say",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        404,
+                                        506,
+                                        412,
+                                        516
+                                    ],
+                                    "score": 0.73,
+                                    "content": "S",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        412,
+                                        503,
+                                        429,
+                                        521
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        429,
+                                        506,
+                                        439,
+                                        515
+                                    ],
+                                    "score": 0.85,
+                                    "content": "S ^ { \\prime }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        440,
+                                        503,
+                                        506,
+                                        521
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are neighboring",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                517,
+                                505,
+                                531
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        518,
+                                        175,
+                                        530
+                                    ],
+                                    "score": 0.91,
+                                    "content": "i f d _ { \\triangle } ( S , S ^ { \\prime } ) \\leq 1",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        175,
+                                        517,
+                                        203,
+                                        531
+                                    ],
+                                    "score": 1.0,
+                                    "content": "where",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        203,
+                                        518,
+                                        353,
+                                        530
+                                    ],
+                                    "score": 0.89,
+                                    "content": "d _ { \\triangle } ( S , S ^ { \\prime } ) \\triangleq \\operatorname* { m a x } \\{ | S \\setminus S ^ { \\prime } | , | S ^ { \\prime } \\setminus S | \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        353,
+                                        517,
+                                        440,
+                                        531
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", which is denoted by",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        440,
+                                        518,
+                                        470,
+                                        528
+                                    ],
+                                    "score": 0.89,
+                                    "content": "S \\sim S ^ { \\prime }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        471,
+                                        517,
+                                        505,
+                                        531
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". For an",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                529,
+                                505,
+                                542
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        529,
+                                        363,
+                                        542
+                                    ],
+                                    "score": 1.0,
+                                    "content": "output of a stochastic query q on a database, we say q satisfies",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        363,
+                                        529,
+                                        385,
+                                        541
+                                    ],
+                                    "score": 0.91,
+                                    "content": "( \\varepsilon , \\delta )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        385,
+                                        529,
+                                        505,
+                                        542
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-differential privacy for some",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                540,
+                                478,
+                                552
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        541,
+                                        131,
+                                        551
+                                    ],
+                                    "score": 0.88,
+                                    "content": "\\varepsilon > 0",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        131,
+                                        540,
+                                        149,
+                                        552
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        150,
+                                        540,
+                                        189,
+                                        552
+                                    ],
+                                    "score": 0.88,
+                                    "content": "\\delta \\in ( 0 , 1 )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        190,
+                                        540,
+                                        195,
+                                        552
+                                    ],
+                                    "score": 1.0,
+                                    "content": "i",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        196,
+                                        540,
+                                        348,
+                                        552
+                                    ],
+                                    "score": 0.87,
+                                    "content": "\\begin{array} { r } { \\textstyle f \\mathbb { P } ( q ( S ) \\in A ) \\le e ^ { \\varepsilon } \\mathbb { P } ( q ( S ^ { \\prime } ) \\in A ) + \\delta } \\end{array}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        348,
+                                        540,
+                                        376,
+                                        552
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for all",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        376,
+                                        540,
+                                        406,
+                                        550
+                                    ],
+                                    "score": 0.92,
+                                    "content": "S \\sim S ^ { \\prime }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        407,
+                                        540,
+                                        466,
+                                        552
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and all subset",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        466,
+                                        541,
+                                        474,
+                                        550
+                                    ],
+                                    "score": 0.78,
+                                    "content": "A",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        474,
+                                        540,
+                                        478,
+                                        552
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        }
+                    ],
+                    "index": 30.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        559,
+                        504,
+                        583
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                558,
+                                506,
+                                573
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        558,
+                                        122,
+                                        573
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Let",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        122,
+                                        559,
+                                        171,
+                                        571
+                                    ],
+                                    "score": 0.91,
+                                    "content": "z \\sim \\mathrm { L a p } ( b )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        172,
+                                        558,
+                                        506,
+                                        573
+                                    ],
+                                    "score": 1.0,
+                                    "content": "be a random vector with entries i.i.d. sampled from Laplace distribution with pdf",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                569,
+                                500,
+                                585
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        571,
+                                        161,
+                                        584
+                                    ],
+                                    "score": 0.91,
+                                    "content": "( 1 / 2 b ) e ^ { - | z | / b }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        161,
+                                        569,
+                                        181,
+                                        585
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Let",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        182,
+                                        571,
+                                        236,
+                                        584
+                                    ],
+                                    "score": 0.93,
+                                    "content": "z \\sim \\mathcal { N } ( \\mu , \\Sigma )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        236,
+                                        569,
+                                        416,
+                                        585
+                                    ],
+                                    "score": 1.0,
+                                    "content": "denote a Gaussian random vector with mean",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        417,
+                                        574,
+                                        424,
+                                        583
+                                    ],
+                                    "score": 0.79,
+                                    "content": "\\mu",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        424,
+                                        569,
+                                        487,
+                                        585
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and covariance",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        488,
+                                        572,
+                                        496,
+                                        582
+                                    ],
+                                    "score": 0.8,
+                                    "content": "\\Sigma",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        496,
+                                        569,
+                                        500,
+                                        585
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        }
+                    ],
+                    "index": 33.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        586,
+                        505,
+                        638
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                584,
+                                505,
+                                601
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        584,
+                                        286,
+                                        601
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Definition 1.2. The sensitivity of a query",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        286,
+                                        586,
+                                        338,
+                                        599
+                                    ],
+                                    "score": 0.92,
+                                    "content": "f ( S ) ~ \\in ~ \\mathbb { R } ^ { k }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        338,
+                                        584,
+                                        397,
+                                        601
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is defined as",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        398,
+                                        586,
+                                        505,
+                                        599
+                                    ],
+                                    "score": 0.88,
+                                    "content": "\\Delta _ { p } ~ = ~ \\operatorname* { s u p } _ { S \\sim S ^ { \\prime } } \\| f ( S ) ~ -",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                597,
+                                506,
+                                613
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        599,
+                                        140,
+                                        612
+                                    ],
+                                    "score": 0.92,
+                                    "content": "f ( S ^ { \\prime } ) \\| _ { p }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        141,
+                                        597,
+                                        194,
+                                        613
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for a norm",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        195,
+                                        599,
+                                        304,
+                                        613
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\begin{array} { r } { \\| { \\boldsymbol x } \\| _ { p } ~ = ~ ( \\sum _ { i \\in [ k ] } | x _ { i } | ^ { p } ) ^ { 1 / p } } \\end{array}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        304,
+                                        597,
+                                        332,
+                                        613
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". For",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        332,
+                                        600,
+                                        363,
+                                        611
+                                    ],
+                                    "score": 0.89,
+                                    "content": "p ~ = ~ 1",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        363,
+                                        597,
+                                        506,
+                                        613
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", the Laplace mechanism outputs",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                107,
+                                612,
+                                505,
+                                626
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        612,
+                                        189,
+                                        625
+                                    ],
+                                    "score": 0.9,
+                                    "content": "f ( S ) + \\mathrm { L a p } ( { \\Delta _ { 1 } } / { \\varepsilon } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        189,
+                                        612,
+                                        248,
+                                        626
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and achieves",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        248,
+                                        613,
+                                        309,
+                                        624
+                                    ],
+                                    "score": 0.41,
+                                    "content": "( \\varepsilon , 0 ) – D P \\ : I 3 7 J",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        309,
+                                        612,
+                                        333,
+                                        626
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". For",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        333,
+                                        613,
+                                        361,
+                                        624
+                                    ],
+                                    "score": 0.89,
+                                    "content": "p = 2",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        361,
+                                        612,
+                                        505,
+                                        626
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", the Gaussian mechanism outputs",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                623,
+                                398,
+                                639
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        624,
+                                        277,
+                                        638
+                                    ],
+                                    "score": 0.88,
+                                    "content": "f ( S ) + \\mathcal { N } ( 0 , ( \\Delta _ { 2 } ( \\sqrt { 2 \\log ( 1 . 2 5 / \\delta ) } ) / \\varepsilon ) ^ { 2 } \\mathbf { I } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        277,
+                                        623,
+                                        334,
+                                        639
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and achieves",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        334,
+                                        625,
+                                        393,
+                                        637
+                                    ],
+                                    "score": 0.77,
+                                    "content": "( \\varepsilon , \\delta ) – D P / 3 8 J",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        394,
+                                        623,
+                                        398,
+                                        639
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        }
+                    ],
+                    "index": 36.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        644,
+                        504,
+                        667
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                643,
+                                506,
+                                659
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        643,
+                                        506,
+                                        659
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We use these output perturbation mechanisms along with the exponential mechanism [69] as building",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                655,
+                                422,
+                                668
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        655,
+                                        422,
+                                        668
+                                    ],
+                                    "score": 1.0,
+                                    "content": "blocks. Appendix A provides detailed survey of privacy and robust estimation.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        }
+                    ],
+                    "index": 39.5
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        679,
+                        220,
+                        691
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                679,
+                                221,
+                                692
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        679,
+                                        221,
+                                        692
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1.3 Problem formulation",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        }
+                    ],
+                    "index": 41
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        699,
+                        503,
+                        722
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                700,
+                                505,
+                                712
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        700,
+                                        162,
+                                        712
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We are given",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        163,
+                                        703,
+                                        170,
+                                        710
+                                    ],
+                                    "score": 0.73,
+                                    "content": "n",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        170,
+                                        700,
+                                        505,
+                                        712
+                                    ],
+                                    "score": 1.0,
+                                    "content": "samples from a sub-Gaussian distribution with a known covariance but unknown",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                711,
+                                504,
+                                723
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        711,
+                                        152,
+                                        723
+                                    ],
+                                    "score": 1.0,
+                                    "content": "mean, and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        153,
+                                        713,
+                                        160,
+                                        721
+                                    ],
+                                    "score": 0.78,
+                                    "content": "\\alpha",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        161,
+                                        711,
+                                        504,
+                                        723
+                                    ],
+                                    "score": 1.0,
+                                    "content": "fraction of the samples are corrupted by an adversary. Our goal is to estimate the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        }
+                    ],
+                    "index": 42.5
+                }
+            ],
+            "page_idx": 2,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        302,
+                        741,
+                        309,
+                        750
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                301,
+                                740,
+                                310,
+                                752
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        301,
+                                        740,
+                                        310,
+                                        752
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "table",
+                    "bbox": [
+                        113,
+                        70,
+                        498,
+                        140
+                    ],
+                    "blocks": [
+                        {
+                            "type": "table_body",
+                            "bbox": [
+                                113,
+                                70,
+                                498,
+                                140
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        113,
+                                        70,
+                                        498,
+                                        140
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                113,
+                                                70,
+                                                498,
+                                                140
+                                            ],
+                                            "score": 0.98,
+                                            "html": "<table><tr><td rowspan=1 colspan=1></td><td rowspan=1 colspan=1>Upper bound (poly-time)</td><td rowspan=1 colspan=1>Upper bound (exp-time)</td><td rowspan=1 colspan=1>Lower bound</td></tr><tr><td rowspan=1 colspan=1>(ε,δ)-DP [54]</td><td rowspan=1 colspan=1>O（d10g1/2（1/）Qε</td><td rowspan=1 colspan=1>O（d10g1/2（1/0））Qε</td><td rowspan=1 colspan=1>2（）</td></tr><tr><td rowspan=1 colspan=1>α-corruption [36]</td><td rowspan=1 colspan=1>）</td><td rowspan=1 colspan=1></td><td rowspan=1 colspan=1>(）</td></tr><tr><td rowspan=1 colspan=1>α-corruption and(ε,δ)-DP (this paper)</td><td rowspan=1 colspan=1>0（(/210g(1/8))αε[Theorem 9]</td><td rowspan=1 colspan=1>（+d1721g/2(18）αε[Theorem 8]</td><td rowspan=1 colspan=1>(([54])</td></tr></table>",
+                                            "type": "table",
+                                            "image_path": "191d63f4ee021bb7fbfe8805bb187077f698af2e6e124ee5962d88fbe956348f.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 1,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        113,
+                                        70,
+                                        498,
+                                        93.33333333333333
+                                    ],
+                                    "spans": [],
+                                    "index": 0
+                                },
+                                {
+                                    "bbox": [
+                                        113,
+                                        93.33333333333333,
+                                        498,
+                                        116.66666666666666
+                                    ],
+                                    "spans": [],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        113,
+                                        116.66666666666666,
+                                        498,
+                                        140.0
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 1
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        150,
+                        505,
+                        186
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                150,
+                                505,
+                                163
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        151,
+                                        240,
+                                        163
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Table 2: For estimating the mean",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        240,
+                                        150,
+                                        271,
+                                        163
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\mu \\in \\mathbb { R } ^ { d }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        272,
+                                        151,
+                                        505,
+                                        163
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of a covariance bounded distribution, we list the sufficient",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                161,
+                                507,
+                                177
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        161,
+                                        352,
+                                        177
+                                    ],
+                                    "score": 1.0,
+                                    "content": "or necessary conditions on the sample size to achieve an error",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        352,
+                                        162,
+                                        439,
+                                        175
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\| \\hat { \\mu } - \\mu \\| _ { 2 } = O ( \\alpha ^ { 1 / 2 } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        439,
+                                        161,
+                                        465,
+                                        177
+                                    ],
+                                    "score": 1.0,
+                                    "content": "under",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        466,
+                                        163,
+                                        487,
+                                        175
+                                    ],
+                                    "score": 0.91,
+                                    "content": "( \\varepsilon , \\delta )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        488,
+                                        161,
+                                        507,
+                                        177
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-DP,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                174,
+                                303,
+                                186
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        174,
+                                        173,
+                                        186
+                                    ],
+                                    "score": 1.0,
+                                    "content": "corruption of an",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        173,
+                                        176,
+                                        181,
+                                        184
+                                    ],
+                                    "score": 0.78,
+                                    "content": "\\alpha",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        181,
+                                        174,
+                                        303,
+                                        186
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-fraction of samples, and both.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        }
+                    ],
+                    "index": 4,
+                    "bbox_fs": [
+                        105,
+                        150,
+                        507,
+                        186
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        204,
+                        231,
+                        216
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                203,
+                                231,
+                                217
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        203,
+                                        231,
+                                        217
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1.1 Technical contributions",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        }
+                    ],
+                    "index": 6
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        223,
+                        506,
+                        302
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                224,
+                                506,
+                                237
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        224,
+                                        320,
+                                        237
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We introduce PRIME which simultaneously achieves",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        320,
+                                        225,
+                                        342,
+                                        236
+                                    ],
+                                    "score": 0.91,
+                                    "content": "( \\varepsilon , \\delta )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        343,
+                                        224,
+                                        451,
+                                        237
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-DP and robustness against",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        451,
+                                        226,
+                                        459,
+                                        235
+                                    ],
+                                    "score": 0.79,
+                                    "content": "\\alpha",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        459,
+                                        224,
+                                        506,
+                                        237
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-fraction of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                234,
+                                507,
+                                249
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        234,
+                                        507,
+                                        249
+                                    ],
+                                    "score": 1.0,
+                                    "content": "corruption. A major challenge in making a standard filter-based robust estimation algorithm (e.g.,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                246,
+                                506,
+                                258
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        246,
+                                        506,
+                                        258
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[30]) private is the high sensitivity of the filtered set that we pass from one iteration to the next.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                257,
+                                505,
+                                270
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        257,
+                                        505,
+                                        270
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We propose a new framework which makes private only the statistics of the set, hence significantly",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                267,
+                                506,
+                                281
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        267,
+                                        506,
+                                        281
+                                    ],
+                                    "score": 1.0,
+                                    "content": "reducing the sensitivity. Our major innovation is a tight analysis of the end-to-end sensitivity of this",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                278,
+                                506,
+                                292
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        278,
+                                        506,
+                                        292
+                                    ],
+                                    "score": 1.0,
+                                    "content": "multiple interactive accesses to the database. This is critical in achieving robustness while preserving",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                290,
+                                497,
+                                303
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        290,
+                                        497,
+                                        303
+                                    ],
+                                    "score": 1.0,
+                                    "content": "privacy and is also of independent interest in making general iterative filtering algorithms private.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        }
+                    ],
+                    "index": 10,
+                    "bbox_fs": [
+                        104,
+                        224,
+                        507,
+                        303
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        306,
+                        505,
+                        385
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                305,
+                                505,
+                                320
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        305,
+                                        390,
+                                        320
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The classical filter approach (see, e.g. [30]) needs to access the database √",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        391,
+                                        306,
+                                        413,
+                                        318
+                                    ],
+                                    "score": 0.91,
+                                    "content": "O ( d )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        413,
+                                        305,
+                                        505,
+                                        320
+                                    ],
+                                    "score": 1.0,
+                                    "content": "times, which brings an",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                317,
+                                506,
+                                332
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        317,
+                                        128,
+                                        332
+                                    ],
+                                    "score": 1.0,
+                                    "content": "extra",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        129,
+                                        317,
+                                        159,
+                                        331
+                                    ],
+                                    "score": 0.93,
+                                    "content": "O ( { \\sqrt { d } } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        159,
+                                        317,
+                                        506,
+                                        332
+                                    ],
+                                    "score": 1.0,
+                                    "content": "factor in the sample complexity due to DP composition. In order to reduce the iteration",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                330,
+                                504,
+                                343
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        330,
+                                        504,
+                                        343
+                                    ],
+                                    "score": 1.0,
+                                    "content": "complexity, following the approach in [36], we propose filtering multiple directions simultaneously",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                340,
+                                505,
+                                353
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        340,
+                                        505,
+                                        353
+                                    ],
+                                    "score": 1.0,
+                                    "content": "using a new score based on the matrix multiplicative weights (MMW). In order to privatize the MMW",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                351,
+                                504,
+                                363
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        351,
+                                        504,
+                                        363
+                                    ],
+                                    "score": 1.0,
+                                    "content": "filter, our major innovation is a novel adaptive filtering algorithm DPTHRESHOLD(·) that outputs",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                363,
+                                505,
+                                375
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        363,
+                                        505,
+                                        375
+                                    ],
+                                    "score": 1.0,
+                                    "content": "a single private threshold which guarantees sufficient progress at every iteration. This brings the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                372,
+                                331,
+                                387
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        372,
+                                        245,
+                                        387
+                                    ],
+                                    "score": 1.0,
+                                    "content": "number of database accesses from",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        245,
+                                        373,
+                                        267,
+                                        385
+                                    ],
+                                    "score": 0.92,
+                                    "content": "O ( d )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        267,
+                                        372,
+                                        279,
+                                        387
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        279,
+                                        373,
+                                        325,
+                                        385
+                                    ],
+                                    "score": 0.92,
+                                    "content": "O ( ( \\log d ) ^ { 2 } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        326,
+                                        372,
+                                        331,
+                                        387
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        }
+                    ],
+                    "index": 17,
+                    "bbox_fs": [
+                        105,
+                        305,
+                        506,
+                        387
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        390,
+                        505,
+                        457
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                389,
+                                506,
+                                404
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        389,
+                                        313,
+                                        404
+                                    ],
+                                    "score": 1.0,
+                                    "content": "One downside of PRIME is that it requires an extra",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        313,
+                                        389,
+                                        332,
+                                        401
+                                    ],
+                                    "score": 0.9,
+                                    "content": "d ^ { 1 / 2 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        332,
+                                        389,
+                                        506,
+                                        404
+                                    ],
+                                    "score": 1.0,
+                                    "content": "factor in the sample complexity, compared",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                402,
+                                505,
+                                414
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        402,
+                                        505,
+                                        414
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to known lower bounds for (non-robust) DP mean estimation. To investigate whether this is also",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                413,
+                                504,
+                                425
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        413,
+                                        492,
+                                        425
+                                    ],
+                                    "score": 1.0,
+                                    "content": "necessary, we propose a sample optimal exponential time robust mean estimation algorithm in",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        492,
+                                        413,
+                                        504,
+                                        424
+                                    ],
+                                    "score": 0.8,
+                                    "content": "\\ S 4",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                424,
+                                505,
+                                437
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        424,
+                                        505,
+                                        437
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and prove that there is no extra statistical cost to jointly requiring privacy and robustness. Our major",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                434,
+                                506,
+                                448
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        434,
+                                        506,
+                                        448
+                                    ],
+                                    "score": 1.0,
+                                    "content": "technical innovations is in using resilience property of the dataset to not only find robust mean (which",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                446,
+                                436,
+                                458
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        446,
+                                        436,
+                                        458
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is the typical use case of resilience) but also bound sensitivity of that robust mean.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        }
+                    ],
+                    "index": 23.5,
+                    "bbox_fs": [
+                        105,
+                        389,
+                        506,
+                        458
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        469,
+                        302,
+                        482
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                467,
+                                303,
+                                484
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        467,
+                                        303,
+                                        484
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1.2 Preliminary on differential privacy (DP)",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        }
+                    ],
+                    "index": 27
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        489,
+                        498,
+                        501
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                487,
+                                499,
+                                504
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        487,
+                                        499,
+                                        504
+                                    ],
+                                    "score": 1.0,
+                                    "content": "DP is a formal metric for measuring privacy leakage when a dataset is accessed with a query [37].",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        }
+                    ],
+                    "index": 28,
+                    "bbox_fs": [
+                        105,
+                        487,
+                        499,
+                        504
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        504,
+                        505,
+                        552
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                503,
+                                506,
+                                521
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        503,
+                                        245,
+                                        521
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Definition 1.1. Given two datasets",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        246,
+                                        505,
+                                        298,
+                                        518
+                                    ],
+                                    "score": 0.91,
+                                    "content": "S = \\{ x _ { i } \\} _ { i = 1 } ^ { n }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        299,
+                                        503,
+                                        316,
+                                        521
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        316,
+                                        503,
+                                        372,
+                                        518
+                                    ],
+                                    "score": 0.92,
+                                    "content": "S ^ { \\prime } = \\{ x _ { i } ^ { \\prime } \\} _ { i = 1 } ^ { n ^ { \\prime } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        372,
+                                        503,
+                                        404,
+                                        521
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", we say",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        404,
+                                        506,
+                                        412,
+                                        516
+                                    ],
+                                    "score": 0.73,
+                                    "content": "S",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        412,
+                                        503,
+                                        429,
+                                        521
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        429,
+                                        506,
+                                        439,
+                                        515
+                                    ],
+                                    "score": 0.85,
+                                    "content": "S ^ { \\prime }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        440,
+                                        503,
+                                        506,
+                                        521
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are neighboring",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                517,
+                                505,
+                                531
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        518,
+                                        175,
+                                        530
+                                    ],
+                                    "score": 0.91,
+                                    "content": "i f d _ { \\triangle } ( S , S ^ { \\prime } ) \\leq 1",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        175,
+                                        517,
+                                        203,
+                                        531
+                                    ],
+                                    "score": 1.0,
+                                    "content": "where",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        203,
+                                        518,
+                                        353,
+                                        530
+                                    ],
+                                    "score": 0.89,
+                                    "content": "d _ { \\triangle } ( S , S ^ { \\prime } ) \\triangleq \\operatorname* { m a x } \\{ | S \\setminus S ^ { \\prime } | , | S ^ { \\prime } \\setminus S | \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        353,
+                                        517,
+                                        440,
+                                        531
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", which is denoted by",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        440,
+                                        518,
+                                        470,
+                                        528
+                                    ],
+                                    "score": 0.89,
+                                    "content": "S \\sim S ^ { \\prime }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        471,
+                                        517,
+                                        505,
+                                        531
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". For an",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                529,
+                                505,
+                                542
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        529,
+                                        363,
+                                        542
+                                    ],
+                                    "score": 1.0,
+                                    "content": "output of a stochastic query q on a database, we say q satisfies",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        363,
+                                        529,
+                                        385,
+                                        541
+                                    ],
+                                    "score": 0.91,
+                                    "content": "( \\varepsilon , \\delta )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        385,
+                                        529,
+                                        505,
+                                        542
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-differential privacy for some",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                540,
+                                478,
+                                552
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        541,
+                                        131,
+                                        551
+                                    ],
+                                    "score": 0.88,
+                                    "content": "\\varepsilon > 0",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        131,
+                                        540,
+                                        149,
+                                        552
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        150,
+                                        540,
+                                        189,
+                                        552
+                                    ],
+                                    "score": 0.88,
+                                    "content": "\\delta \\in ( 0 , 1 )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        190,
+                                        540,
+                                        195,
+                                        552
+                                    ],
+                                    "score": 1.0,
+                                    "content": "i",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        196,
+                                        540,
+                                        348,
+                                        552
+                                    ],
+                                    "score": 0.87,
+                                    "content": "\\begin{array} { r } { \\textstyle f \\mathbb { P } ( q ( S ) \\in A ) \\le e ^ { \\varepsilon } \\mathbb { P } ( q ( S ^ { \\prime } ) \\in A ) + \\delta } \\end{array}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        348,
+                                        540,
+                                        376,
+                                        552
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for all",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        376,
+                                        540,
+                                        406,
+                                        550
+                                    ],
+                                    "score": 0.92,
+                                    "content": "S \\sim S ^ { \\prime }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        407,
+                                        540,
+                                        466,
+                                        552
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and all subset",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        466,
+                                        541,
+                                        474,
+                                        550
+                                    ],
+                                    "score": 0.78,
+                                    "content": "A",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        474,
+                                        540,
+                                        478,
+                                        552
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        }
+                    ],
+                    "index": 30.5,
+                    "bbox_fs": [
+                        104,
+                        503,
+                        506,
+                        552
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        559,
+                        504,
+                        583
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                558,
+                                506,
+                                573
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        558,
+                                        122,
+                                        573
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Let",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        122,
+                                        559,
+                                        171,
+                                        571
+                                    ],
+                                    "score": 0.91,
+                                    "content": "z \\sim \\mathrm { L a p } ( b )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        172,
+                                        558,
+                                        506,
+                                        573
+                                    ],
+                                    "score": 1.0,
+                                    "content": "be a random vector with entries i.i.d. sampled from Laplace distribution with pdf",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                569,
+                                500,
+                                585
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        571,
+                                        161,
+                                        584
+                                    ],
+                                    "score": 0.91,
+                                    "content": "( 1 / 2 b ) e ^ { - | z | / b }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        161,
+                                        569,
+                                        181,
+                                        585
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Let",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        182,
+                                        571,
+                                        236,
+                                        584
+                                    ],
+                                    "score": 0.93,
+                                    "content": "z \\sim \\mathcal { N } ( \\mu , \\Sigma )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        236,
+                                        569,
+                                        416,
+                                        585
+                                    ],
+                                    "score": 1.0,
+                                    "content": "denote a Gaussian random vector with mean",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        417,
+                                        574,
+                                        424,
+                                        583
+                                    ],
+                                    "score": 0.79,
+                                    "content": "\\mu",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        424,
+                                        569,
+                                        487,
+                                        585
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and covariance",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        488,
+                                        572,
+                                        496,
+                                        582
+                                    ],
+                                    "score": 0.8,
+                                    "content": "\\Sigma",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        496,
+                                        569,
+                                        500,
+                                        585
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        }
+                    ],
+                    "index": 33.5,
+                    "bbox_fs": [
+                        104,
+                        558,
+                        506,
+                        585
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        586,
+                        505,
+                        638
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                584,
+                                505,
+                                601
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        584,
+                                        286,
+                                        601
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Definition 1.2. The sensitivity of a query",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        286,
+                                        586,
+                                        338,
+                                        599
+                                    ],
+                                    "score": 0.92,
+                                    "content": "f ( S ) ~ \\in ~ \\mathbb { R } ^ { k }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        338,
+                                        584,
+                                        397,
+                                        601
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is defined as",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        398,
+                                        586,
+                                        505,
+                                        599
+                                    ],
+                                    "score": 0.88,
+                                    "content": "\\Delta _ { p } ~ = ~ \\operatorname* { s u p } _ { S \\sim S ^ { \\prime } } \\| f ( S ) ~ -",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                597,
+                                506,
+                                613
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        599,
+                                        140,
+                                        612
+                                    ],
+                                    "score": 0.92,
+                                    "content": "f ( S ^ { \\prime } ) \\| _ { p }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        141,
+                                        597,
+                                        194,
+                                        613
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for a norm",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        195,
+                                        599,
+                                        304,
+                                        613
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\begin{array} { r } { \\| { \\boldsymbol x } \\| _ { p } ~ = ~ ( \\sum _ { i \\in [ k ] } | x _ { i } | ^ { p } ) ^ { 1 / p } } \\end{array}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        304,
+                                        597,
+                                        332,
+                                        613
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". For",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        332,
+                                        600,
+                                        363,
+                                        611
+                                    ],
+                                    "score": 0.89,
+                                    "content": "p ~ = ~ 1",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        363,
+                                        597,
+                                        506,
+                                        613
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", the Laplace mechanism outputs",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                107,
+                                612,
+                                505,
+                                626
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        612,
+                                        189,
+                                        625
+                                    ],
+                                    "score": 0.9,
+                                    "content": "f ( S ) + \\mathrm { L a p } ( { \\Delta _ { 1 } } / { \\varepsilon } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        189,
+                                        612,
+                                        248,
+                                        626
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and achieves",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        248,
+                                        613,
+                                        309,
+                                        624
+                                    ],
+                                    "score": 0.41,
+                                    "content": "( \\varepsilon , 0 ) – D P \\ : I 3 7 J",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        309,
+                                        612,
+                                        333,
+                                        626
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". For",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        333,
+                                        613,
+                                        361,
+                                        624
+                                    ],
+                                    "score": 0.89,
+                                    "content": "p = 2",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        361,
+                                        612,
+                                        505,
+                                        626
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", the Gaussian mechanism outputs",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                623,
+                                398,
+                                639
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        624,
+                                        277,
+                                        638
+                                    ],
+                                    "score": 0.88,
+                                    "content": "f ( S ) + \\mathcal { N } ( 0 , ( \\Delta _ { 2 } ( \\sqrt { 2 \\log ( 1 . 2 5 / \\delta ) } ) / \\varepsilon ) ^ { 2 } \\mathbf { I } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        277,
+                                        623,
+                                        334,
+                                        639
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and achieves",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        334,
+                                        625,
+                                        393,
+                                        637
+                                    ],
+                                    "score": 0.77,
+                                    "content": "( \\varepsilon , \\delta ) – D P / 3 8 J",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        394,
+                                        623,
+                                        398,
+                                        639
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        }
+                    ],
+                    "index": 36.5,
+                    "bbox_fs": [
+                        104,
+                        584,
+                        506,
+                        639
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        644,
+                        504,
+                        667
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                643,
+                                506,
+                                659
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        643,
+                                        506,
+                                        659
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We use these output perturbation mechanisms along with the exponential mechanism [69] as building",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                655,
+                                422,
+                                668
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        655,
+                                        422,
+                                        668
+                                    ],
+                                    "score": 1.0,
+                                    "content": "blocks. Appendix A provides detailed survey of privacy and robust estimation.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        }
+                    ],
+                    "index": 39.5,
+                    "bbox_fs": [
+                        105,
+                        643,
+                        506,
+                        668
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        679,
+                        220,
+                        691
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                679,
+                                221,
+                                692
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        679,
+                                        221,
+                                        692
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1.3 Problem formulation",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        }
+                    ],
+                    "index": 41
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        699,
+                        503,
+                        722
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                700,
+                                505,
+                                712
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        700,
+                                        162,
+                                        712
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We are given",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        163,
+                                        703,
+                                        170,
+                                        710
+                                    ],
+                                    "score": 0.73,
+                                    "content": "n",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        170,
+                                        700,
+                                        505,
+                                        712
+                                    ],
+                                    "score": 1.0,
+                                    "content": "samples from a sub-Gaussian distribution with a known covariance but unknown",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                711,
+                                504,
+                                723
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        711,
+                                        152,
+                                        723
+                                    ],
+                                    "score": 1.0,
+                                    "content": "mean, and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        153,
+                                        713,
+                                        160,
+                                        721
+                                    ],
+                                    "score": 0.78,
+                                    "content": "\\alpha",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        161,
+                                        711,
+                                        504,
+                                        723
+                                    ],
+                                    "score": 1.0,
+                                    "content": "fraction of the samples are corrupted by an adversary. Our goal is to estimate the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                71,
+                                505,
+                                86
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        71,
+                                        505,
+                                        86
+                                    ],
+                                    "score": 1.0,
+                                    "content": "unknown mean. We follow the standard definition of adversary in [30], which can adaptively choose",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                84,
+                                389,
+                                96
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        84,
+                                        389,
+                                        96
+                                    ],
+                                    "score": 1.0,
+                                    "content": "which samples to corrupt and arbitrarily replace them with any points.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 1
+                        }
+                    ],
+                    "index": 42.5,
+                    "bbox_fs": [
+                        106,
+                        700,
+                        505,
+                        723
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        72,
+                        504,
+                        95
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                71,
+                                505,
+                                86
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        71,
+                                        505,
+                                        86
+                                    ],
+                                    "score": 1.0,
+                                    "content": "unknown mean. We follow the standard definition of adversary in [30], which can adaptively choose",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                84,
+                                389,
+                                96
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        84,
+                                        389,
+                                        96
+                                    ],
+                                    "score": 1.0,
+                                    "content": "which samples to corrupt and arbitrarily replace them with any points.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        }
+                    ],
+                    "index": 0.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        97,
+                        505,
+                        158
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                97,
+                                506,
+                                112
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        97,
+                                        274,
+                                        112
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Assumption 1. An uncorrupted dataset",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        274,
+                                        98,
+                                        298,
+                                        110
+                                    ],
+                                    "score": 0.9,
+                                    "content": "S _ { \\mathrm { g o o d } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        299,
+                                        97,
+                                        347,
+                                        112
+                                    ],
+                                    "score": 1.0,
+                                    "content": "consists of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        348,
+                                        100,
+                                        354,
+                                        108
+                                    ],
+                                    "score": 0.26,
+                                    "content": "n",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        355,
+                                        97,
+                                        444,
+                                        112
+                                    ],
+                                    "score": 1.0,
+                                    "content": "i.i.d. samples from a",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        444,
+                                        99,
+                                        451,
+                                        108
+                                    ],
+                                    "score": 0.69,
+                                    "content": "d",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        451,
+                                        97,
+                                        506,
+                                        112
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-dimensional",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                109,
+                                506,
+                                123
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        110,
+                                        253,
+                                        123
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sub-Gaussian distribution with mean",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        253,
+                                        110,
+                                        284,
+                                        122
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\mu \\in \\mathbb { R } ^ { d }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        285,
+                                        110,
+                                        348,
+                                        123
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and covariance",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        349,
+                                        109,
+                                        403,
+                                        122
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\mathbb { E } [ x x ^ { \\top } ] = \\mathbf { I } _ { d } ,",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        403,
+                                        110,
+                                        506,
+                                        123
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", which is 1-sub-Gaussian,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                121,
+                                506,
+                                136
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        121,
+                                        124,
+                                        136
+                                    ],
+                                    "score": 1.0,
+                                    "content": "i.e.,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        124,
+                                        122,
+                                        243,
+                                        134
+                                    ],
+                                    "score": 0.89,
+                                    "content": "\\mathbb { E } [ \\exp ( v ^ { \\top } x ) ] \\leq \\exp ( \\| v \\| _ { 2 } ^ { 2 } / 2 )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        244,
+                                        121,
+                                        272,
+                                        136
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for all",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        272,
+                                        122,
+                                        302,
+                                        133
+                                    ],
+                                    "score": 0.9,
+                                    "content": "v \\in \\mathbb { R } ^ { d }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        302,
+                                        121,
+                                        346,
+                                        136
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". For some",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        347,
+                                        123,
+                                        397,
+                                        135
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\alpha \\in ( 0 , 1 / 2 )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        398,
+                                        121,
+                                        506,
+                                        136
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", we are given a corrupted",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                133,
+                                506,
+                                148
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        133,
+                                        137,
+                                        148
+                                    ],
+                                    "score": 1.0,
+                                    "content": "dataset",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        137,
+                                        134,
+                                        214,
+                                        147
+                                    ],
+                                    "score": 0.92,
+                                    "content": "S = \\{ x _ { i } \\in \\mathbb { R } ^ { d } \\} _ { i = 1 } ^ { n }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        215,
+                                        133,
+                                        442,
+                                        148
+                                    ],
+                                    "score": 1.0,
+                                    "content": "where an adversary adaptively inspects all the samples in",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        442,
+                                        135,
+                                        466,
+                                        146
+                                    ],
+                                    "score": 0.91,
+                                    "content": "S _ { \\mathrm { g o o d } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        467,
+                                        133,
+                                        506,
+                                        148
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", removes",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                146,
+                                422,
+                                159
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        146,
+                                        251,
+                                        159
+                                    ],
+                                    "score": 1.0,
+                                    "content": "αn of them, and replaces them with",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        251,
+                                        147,
+                                        271,
+                                        158
+                                    ],
+                                    "score": 0.9,
+                                    "content": "S _ { \\mathrm { b a d } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        272,
+                                        146,
+                                        405,
+                                        159
+                                    ],
+                                    "score": 1.0,
+                                    "content": "which are αn arbitrary points in",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        405,
+                                        146,
+                                        417,
+                                        156
+                                    ],
+                                    "score": 0.88,
+                                    "content": "\\mathbb { R } ^ { d }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        418,
+                                        146,
+                                        422,
+                                        159
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        }
+                    ],
+                    "index": 4
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        166,
+                        504,
+                        189
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                167,
+                                504,
+                                178
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        167,
+                                        504,
+                                        178
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Similarly, we consider the same problem for heavy-tailed distributions with a bounded covariance.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                177,
+                                495,
+                                190
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        177,
+                                        495,
+                                        190
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We present the assumption and main results for covariance bounded distributions in Appendix B.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        }
+                    ],
+                    "index": 7.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        194,
+                        506,
+                        256
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                193,
+                                506,
+                                210
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        193,
+                                        171,
+                                        210
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Notations. Let",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        171,
+                                        195,
+                                        251,
+                                        207
+                                    ],
+                                    "score": 0.92,
+                                    "content": "[ n ] = \\{ 1 , 2 , \\dots , n \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        251,
+                                        193,
+                                        273,
+                                        210
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". For",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        273,
+                                        194,
+                                        306,
+                                        206
+                                    ],
+                                    "score": 0.9,
+                                    "content": "x \\in \\mathbb { R } ^ { d }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        306,
+                                        193,
+                                        342,
+                                        210
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", we use",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        342,
+                                        194,
+                                        447,
+                                        209
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\textstyle \\| x \\| _ { 2 } = ( \\sum _ { i \\in [ d ] } ( x _ { i } ) ^ { 2 } ) ^ { 1 / 2 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        448,
+                                        193,
+                                        506,
+                                        210
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to denote the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                205,
+                                507,
+                                225
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        205,
+                                        194,
+                                        225
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Euclidean norm. For",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        194,
+                                        208,
+                                        239,
+                                        219
+                                    ],
+                                    "score": 0.91,
+                                    "content": "X \\in \\mathbb { R } ^ { d \\times d }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        240,
+                                        205,
+                                        275,
+                                        225
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", we use",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        275,
+                                        209,
+                                        389,
+                                        222
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\begin{array} { r } { \\| X \\| _ { 2 } = \\operatorname* { m a x } _ { \\| v \\| _ { 2 } = 1 } \\| X v \\| _ { 2 } } \\end{array}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        389,
+                                        205,
+                                        507,
+                                        225
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to denote the spectral norm.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                219,
+                                506,
+                                233
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        219,
+                                        123,
+                                        233
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        124,
+                                        220,
+                                        147,
+                                        231
+                                    ],
+                                    "score": 0.91,
+                                    "content": "d \\times d",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        148,
+                                        219,
+                                        217,
+                                        233
+                                    ],
+                                    "score": 1.0,
+                                    "content": "identity matrix is",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        217,
+                                        221,
+                                        238,
+                                        232
+                                    ],
+                                    "score": 0.9,
+                                    "content": "\\mathbf { I } _ { d \\times d }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        238,
+                                        219,
+                                        407,
+                                        233
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Whenever it is clear from context, we use",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        407,
+                                        221,
+                                        415,
+                                        230
+                                    ],
+                                    "score": 0.82,
+                                    "content": "S",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        415,
+                                        219,
+                                        506,
+                                        233
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to denote both a set of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                231,
+                                506,
+                                246
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        232,
+                                        340,
+                                        246
+                                    ],
+                                    "score": 1.0,
+                                    "content": "data points and also the set of indices of those data points.",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        340,
+                                        231,
+                                        349,
+                                        244
+                                    ],
+                                    "score": 0.85,
+                                    "content": "\\widetilde O",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        349,
+                                        232,
+                                        367,
+                                        246
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        367,
+                                        231,
+                                        376,
+                                        243
+                                    ],
+                                    "score": 0.86,
+                                    "content": "\\widetilde \\Omega",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        376,
+                                        232,
+                                        506,
+                                        246
+                                    ],
+                                    "score": 1.0,
+                                    "content": "hide poly-logarithmic factors in",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                107,
+                                244,
+                                256,
+                                257
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        244,
+                                        144,
+                                        256
+                                    ],
+                                    "score": 0.93,
+                                    "content": "d , n , 1 / \\alpha",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        145,
+                                        244,
+                                        256,
+                                        257
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", and the failure probability.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        }
+                    ],
+                    "index": 11
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        260,
+                        505,
+                        294
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                260,
+                                505,
+                                273
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        260,
+                                        352,
+                                        273
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Outline. We present PRIME for sub-Gaussian distribution in",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        353,
+                                        261,
+                                        364,
+                                        271
+                                    ],
+                                    "score": 0.76,
+                                    "content": "\\ S 2",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        364,
+                                        260,
+                                        505,
+                                        273
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", and present theoretical analysis in",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                107,
+                                271,
+                                505,
+                                284
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        272,
+                                        118,
+                                        283
+                                    ],
+                                    "score": 0.81,
+                                    "content": "\\ S 3",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        118,
+                                        271,
+                                        437,
+                                        284
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". We then introduce an exponential time algorithm with near optimal guarantee in",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        437,
+                                        272,
+                                        448,
+                                        282
+                                    ],
+                                    "score": 0.83,
+                                    "content": "\\ S 4",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        448,
+                                        271,
+                                        505,
+                                        284
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Due to space",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                282,
+                                462,
+                                295
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        282,
+                                        462,
+                                        295
+                                    ],
+                                    "score": 1.0,
+                                    "content": "constraints, analogous results for heavy-tailed distributions are presented in Appendix B.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        }
+                    ],
+                    "index": 15
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        106,
+                        308,
+                        449,
+                        322
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                308,
+                                450,
+                                324
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        308,
+                                        450,
+                                        324
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2 PRIME: efficient algorithm for robust and DP mean estimation",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        }
+                    ],
+                    "index": 17
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        333,
+                        505,
+                        356
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                333,
+                                505,
+                                346
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        333,
+                                        505,
+                                        346
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In order to describe the proposed algorithm PRIME, we need to first describe a standard (non-private)",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                345,
+                                328,
+                                357
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        345,
+                                        328,
+                                        357
+                                    ],
+                                    "score": 1.0,
+                                    "content": "iterative filtering algorithm for robust mean estimation.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        }
+                    ],
+                    "index": 18.5
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        368,
+                        446,
+                        380
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                367,
+                                447,
+                                382
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        367,
+                                        447,
+                                        382
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2.1 Background on (non-private) iterative filtering for robust mean estimation",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        }
+                    ],
+                    "index": 20
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        388,
+                        506,
+                        433
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                389,
+                                507,
+                                402
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        389,
+                                        507,
+                                        402
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Non-private robust mean estimation approaches recursively apply the following filter, whose frame-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                398,
+                                507,
+                                414
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        398,
+                                        288,
+                                        414
+                                    ],
+                                    "score": 1.0,
+                                    "content": "work is first proposed in [28]. Given a dataset",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        288,
+                                        399,
+                                        341,
+                                        412
+                                    ],
+                                    "score": 0.92,
+                                    "content": "S = \\{ x _ { i } \\} _ { i = 1 } ^ { n }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        341,
+                                        398,
+                                        403,
+                                        414
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", the current set",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        403,
+                                        400,
+                                        439,
+                                        411
+                                    ],
+                                    "score": 0.92,
+                                    "content": "S _ { 0 } \\subseteq [ n ]",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        439,
+                                        398,
+                                        507,
+                                        414
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of data points is",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                410,
+                                506,
+                                424
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        410,
+                                        194,
+                                        424
+                                    ],
+                                    "score": 1.0,
+                                    "content": "updated starting with",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        195,
+                                        411,
+                                        231,
+                                        423
+                                    ],
+                                    "score": 0.93,
+                                    "content": "S _ { 1 } = [ n ]",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        231,
+                                        410,
+                                        506,
+                                        424
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". At each step, the following filter (Algorithm 1 in [63]) attempts to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                422,
+                                308,
+                                433
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        422,
+                                        308,
+                                        433
+                                    ],
+                                    "score": 1.0,
+                                    "content": "detect the corrupted data points and remove them.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        }
+                    ],
+                    "index": 22.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        129,
+                        435,
+                        505,
+                        515
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                129,
+                                435,
+                                506,
+                                451
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        129,
+                                        435,
+                                        263,
+                                        451
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1. Compute the top eigenvector",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        263,
+                                        436,
+                                        420,
+                                        451
+                                    ],
+                                    "score": 0.88,
+                                    "content": "v _ { t } \\gets \\arg \\operatorname* { m a x } _ { v : \\| v \\| _ { 2 } = 1 } v ^ { \\top } \\mathrm { C o v } ( S _ { t - 1 } ) v",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        420,
+                                        435,
+                                        506,
+                                        451
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of the covariance of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                448,
+                                274,
+                                464
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        448,
+                                        221,
+                                        464
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the current data set",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        221,
+                                        449,
+                                        266,
+                                        462
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\{ x _ { i } \\} _ { i \\in S _ { t - 1 } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        266,
+                                        448,
+                                        274,
+                                        464
+                                    ],
+                                    "score": 1.0,
+                                    "content": ";",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                462,
+                                465,
+                                479
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        462,
+                                        281,
+                                        479
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2. Compute scores for all data points",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        281,
+                                        463,
+                                        465,
+                                        479
+                                    ],
+                                    "score": 0.83,
+                                    "content": "j \\in S _ { t - 1 } \\colon \\tau _ { j } \\gets \\left( v _ { t } ^ { \\top } \\left( x _ { j } - \\mathrm { M e a n } ( S _ { t - 1 } ) \\right) \\right) ^ { 2 } \\ ;",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                476,
+                                333,
+                                492
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        476,
+                                        250,
+                                        492
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3. Draw a random threshold:",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        250,
+                                        478,
+                                        325,
+                                        491
+                                    ],
+                                    "score": 0.9,
+                                    "content": "Z _ { t } \\gets \\mathrm { U n i f } ( [ 0 , 1 ] )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        325,
+                                        476,
+                                        333,
+                                        492
+                                    ],
+                                    "score": 1.0,
+                                    "content": ";",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                486,
+                                504,
+                                506
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        486,
+                                        230,
+                                        506
+                                    ],
+                                    "score": 1.0,
+                                    "content": "4. Remove outliers from",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        230,
+                                        491,
+                                        251,
+                                        502
+                                    ],
+                                    "score": 0.91,
+                                    "content": "S _ { t - 1 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        252,
+                                        486,
+                                        295,
+                                        506
+                                    ],
+                                    "score": 1.0,
+                                    "content": "defined as",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        295,
+                                        491,
+                                        355,
+                                        502
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\{ i \\in S _ { t - 1 } : \\tau _ { i }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        356,
+                                        486,
+                                        418,
+                                        506
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is in the largest",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        418,
+                                        491,
+                                        431,
+                                        501
+                                    ],
+                                    "score": 0.8,
+                                    "content": "2 \\alpha",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        431,
+                                        486,
+                                        458,
+                                        506
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-tail of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        459,
+                                        490,
+                                        504,
+                                        503
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\{ \\tau _ { j } \\} _ { j \\in S _ { t - 1 } }",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                499,
+                                338,
+                                516
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        499,
+                                        159,
+                                        516
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        159,
+                                        502,
+                                        218,
+                                        514
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\tau _ { i } \\geq Z _ { t } \\tau _ { \\operatorname* { m a x } } \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        218,
+                                        499,
+                                        248,
+                                        516
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", where",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        248,
+                                        503,
+                                        338,
+                                        514
+                                    ],
+                                    "score": 0.89,
+                                    "content": "\\tau _ { \\mathrm { m a x } } = \\operatorname* { m a x } _ { j \\in S _ { t - 1 } } \\tau _ { j }",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 30
+                        }
+                    ],
+                    "index": 27.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        521,
+                        505,
+                        631
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                521,
+                                506,
+                                534
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        521,
+                                        460,
+                                        534
+                                    ],
+                                    "score": 1.0,
+                                    "content": "This is repeated until the empirical covariance is sufficiently small and the empirical mean",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        460,
+                                        522,
+                                        467,
+                                        533
+                                    ],
+                                    "score": 0.85,
+                                    "content": "\\hat { \\mu }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        467,
+                                        521,
+                                        506,
+                                        534
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is output.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                533,
+                                505,
+                                545
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        533,
+                                        461,
+                                        545
+                                    ],
+                                    "score": 1.0,
+                                    "content": "At a high level, the correctness of this algorithm relies on the key observation that the",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        461,
+                                        534,
+                                        469,
+                                        542
+                                    ],
+                                    "score": 0.79,
+                                    "content": "\\alpha",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        469,
+                                        533,
+                                        505,
+                                        545
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-fraction",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                541,
+                                506,
+                                557
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        541,
+                                        506,
+                                        557
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of adversarial corruption can not significantly change the mean of the dataset without introducing",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                554,
+                                505,
+                                567
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        554,
+                                        505,
+                                        567
+                                    ],
+                                    "score": 1.0,
+                                    "content": "large eigenvalues in the empirical covariance. Therefore, the algorithm finds top eigenvector of the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                565,
+                                506,
+                                577
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        565,
+                                        506,
+                                        577
+                                    ],
+                                    "score": 1.0,
+                                    "content": "empirical covariance in step 1, and tries to correct the empirical covariance by removing corrupted",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                576,
+                                506,
+                                589
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        576,
+                                        506,
+                                        589
+                                    ],
+                                    "score": 1.0,
+                                    "content": "data points. Each data point is assigned a score in step 2 which indicates the “badness” of the data",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                587,
+                                505,
+                                600
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        587,
+                                        204,
+                                        600
+                                    ],
+                                    "score": 1.0,
+                                    "content": "points, and a threshold",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        204,
+                                        587,
+                                        216,
+                                        598
+                                    ],
+                                    "score": 0.88,
+                                    "content": "Z _ { t }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        216,
+                                        587,
+                                        505,
+                                        600
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in step 3 is carefully designed such that step 4 guarantees to remove",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                597,
+                                506,
+                                611
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        597,
+                                        506,
+                                        611
+                                    ],
+                                    "score": 1.0,
+                                    "content": "more corrupted data points than good data points (in expectation). This guarantees the following",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                609,
+                                505,
+                                621
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        609,
+                                        505,
+                                        621
+                                    ],
+                                    "score": 1.0,
+                                    "content": "bound achieving the near-optimal sample complexity shown in the second row of Table 1. A formal",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                619,
+                                357,
+                                632
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        619,
+                                        357,
+                                        632
+                                    ],
+                                    "score": 1.0,
+                                    "content": "description of this algorithm is in Algorithm 4 in Appendix C.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        }
+                    ],
+                    "index": 35.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        633,
+                        504,
+                        660
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                632,
+                                505,
+                                647
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        632,
+                                        381,
+                                        647
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Proposition 2.1 (Corollary of [63, Theorem 2.1]). Under assumption",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        382,
+                                        635,
+                                        387,
+                                        644
+                                    ],
+                                    "score": 0.27,
+                                    "content": "^ { l }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        388,
+                                        632,
+                                        505,
+                                        647
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", the above filtering algorithm",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                645,
+                                409,
+                                661
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        646,
+                                        182,
+                                        661
+                                    ],
+                                    "score": 1.0,
+                                    "content": "achieves accuracy",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        182,
+                                        645,
+                                        406,
+                                        660
+                                    ],
+                                    "score": 0.75,
+                                    "content": "\\| \\hat { \\mu } - \\mu \\| _ { 2 } \\le O ( \\alpha \\sqrt { \\log ( 1 / \\alpha ) } ) \\ w . p . \\ 0 . 9 i f n \\ge \\widetilde { \\Omega } ( d / \\alpha ^ { 2 } ) \\ / .",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        407,
+                                        646,
+                                        409,
+                                        661
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        }
+                    ],
+                    "index": 41.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        667,
+                        505,
+                        722
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                666,
+                                505,
+                                680
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        666,
+                                        505,
+                                        680
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Challenges in making robust mean estimation private. To get a DP and robust mean, a naive",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                678,
+                                505,
+                                690
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        678,
+                                        377,
+                                        690
+                                    ],
+                                    "score": 1.0,
+                                    "content": "attempt is to apply a standard output perturbation mechanism to",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        378,
+                                        678,
+                                        385,
+                                        690
+                                    ],
+                                    "score": 0.83,
+                                    "content": "\\hat { \\mu }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        385,
+                                        678,
+                                        505,
+                                        690
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". However, this is obviously",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                689,
+                                506,
+                                701
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        689,
+                                        506,
+                                        701
+                                    ],
+                                    "score": 1.0,
+                                    "content": "challenging since the end-to-end sensitivity is intractable. The standard recipe to circumvent this is to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                699,
+                                506,
+                                714
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        699,
+                                        203,
+                                        714
+                                    ],
+                                    "score": 1.0,
+                                    "content": "make the current “state”",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        203,
+                                        700,
+                                        214,
+                                        711
+                                    ],
+                                    "score": 0.87,
+                                    "content": "S _ { t }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        214,
+                                        699,
+                                        338,
+                                        714
+                                    ],
+                                    "score": 1.0,
+                                    "content": "private at every iteration. Once",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        338,
+                                        700,
+                                        359,
+                                        712
+                                    ],
+                                    "score": 0.91,
+                                    "content": "S _ { t - 1 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        359,
+                                        699,
+                                        506,
+                                        714
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is private (hence, public knowledge),",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                709,
+                                506,
+                                725
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        709,
+                                        201,
+                                        725
+                                    ],
+                                    "score": 1.0,
+                                    "content": "making the next “state”",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        201,
+                                        711,
+                                        212,
+                                        722
+                                    ],
+                                    "score": 0.86,
+                                    "content": "S _ { t }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        213,
+                                        709,
+                                        506,
+                                        725
+                                    ],
+                                    "score": 1.0,
+                                    "content": "private is simpler. We only need to analyze the sensitivity of a single step",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 47
+                        }
+                    ],
+                    "index": 45
+                }
+            ],
+            "page_idx": 3,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        302,
+                        741,
+                        308,
+                        750
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                301,
+                                740,
+                                310,
+                                752
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        301,
+                                        740,
+                                        310,
+                                        752
+                                    ],
+                                    "score": 1.0,
+                                    "content": "",
+                                    "type": "text",
+                                    "height": 12,
+                                    "width": 9
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        72,
+                        504,
+                        95
+                    ],
+                    "lines": [],
+                    "index": 0.5,
+                    "bbox_fs": [
+                        105,
+                        71,
+                        505,
+                        96
+                    ],
+                    "lines_deleted": true
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        97,
+                        505,
+                        158
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                97,
+                                506,
+                                112
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        97,
+                                        274,
+                                        112
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Assumption 1. An uncorrupted dataset",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        274,
+                                        98,
+                                        298,
+                                        110
+                                    ],
+                                    "score": 0.9,
+                                    "content": "S _ { \\mathrm { g o o d } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        299,
+                                        97,
+                                        347,
+                                        112
+                                    ],
+                                    "score": 1.0,
+                                    "content": "consists of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        348,
+                                        100,
+                                        354,
+                                        108
+                                    ],
+                                    "score": 0.26,
+                                    "content": "n",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        355,
+                                        97,
+                                        444,
+                                        112
+                                    ],
+                                    "score": 1.0,
+                                    "content": "i.i.d. samples from a",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        444,
+                                        99,
+                                        451,
+                                        108
+                                    ],
+                                    "score": 0.69,
+                                    "content": "d",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        451,
+                                        97,
+                                        506,
+                                        112
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-dimensional",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                109,
+                                506,
+                                123
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        110,
+                                        253,
+                                        123
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sub-Gaussian distribution with mean",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        253,
+                                        110,
+                                        284,
+                                        122
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\mu \\in \\mathbb { R } ^ { d }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        285,
+                                        110,
+                                        348,
+                                        123
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and covariance",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        349,
+                                        109,
+                                        403,
+                                        122
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\mathbb { E } [ x x ^ { \\top } ] = \\mathbf { I } _ { d } ,",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        403,
+                                        110,
+                                        506,
+                                        123
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", which is 1-sub-Gaussian,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                121,
+                                506,
+                                136
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        121,
+                                        124,
+                                        136
+                                    ],
+                                    "score": 1.0,
+                                    "content": "i.e.,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        124,
+                                        122,
+                                        243,
+                                        134
+                                    ],
+                                    "score": 0.89,
+                                    "content": "\\mathbb { E } [ \\exp ( v ^ { \\top } x ) ] \\leq \\exp ( \\| v \\| _ { 2 } ^ { 2 } / 2 )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        244,
+                                        121,
+                                        272,
+                                        136
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for all",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        272,
+                                        122,
+                                        302,
+                                        133
+                                    ],
+                                    "score": 0.9,
+                                    "content": "v \\in \\mathbb { R } ^ { d }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        302,
+                                        121,
+                                        346,
+                                        136
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". For some",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        347,
+                                        123,
+                                        397,
+                                        135
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\alpha \\in ( 0 , 1 / 2 )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        398,
+                                        121,
+                                        506,
+                                        136
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", we are given a corrupted",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                133,
+                                506,
+                                148
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        133,
+                                        137,
+                                        148
+                                    ],
+                                    "score": 1.0,
+                                    "content": "dataset",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        137,
+                                        134,
+                                        214,
+                                        147
+                                    ],
+                                    "score": 0.92,
+                                    "content": "S = \\{ x _ { i } \\in \\mathbb { R } ^ { d } \\} _ { i = 1 } ^ { n }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        215,
+                                        133,
+                                        442,
+                                        148
+                                    ],
+                                    "score": 1.0,
+                                    "content": "where an adversary adaptively inspects all the samples in",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        442,
+                                        135,
+                                        466,
+                                        146
+                                    ],
+                                    "score": 0.91,
+                                    "content": "S _ { \\mathrm { g o o d } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        467,
+                                        133,
+                                        506,
+                                        148
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", removes",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                146,
+                                422,
+                                159
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        146,
+                                        251,
+                                        159
+                                    ],
+                                    "score": 1.0,
+                                    "content": "αn of them, and replaces them with",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        251,
+                                        147,
+                                        271,
+                                        158
+                                    ],
+                                    "score": 0.9,
+                                    "content": "S _ { \\mathrm { b a d } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        272,
+                                        146,
+                                        405,
+                                        159
+                                    ],
+                                    "score": 1.0,
+                                    "content": "which are αn arbitrary points in",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        405,
+                                        146,
+                                        417,
+                                        156
+                                    ],
+                                    "score": 0.88,
+                                    "content": "\\mathbb { R } ^ { d }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        418,
+                                        146,
+                                        422,
+                                        159
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        }
+                    ],
+                    "index": 4,
+                    "bbox_fs": [
+                        105,
+                        97,
+                        506,
+                        159
+                    ]
+                },
+                {
+                    "type": "list",
+                    "bbox": [
+                        107,
+                        166,
+                        504,
+                        189
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                167,
+                                504,
+                                178
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        167,
+                                        504,
+                                        178
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Similarly, we consider the same problem for heavy-tailed distributions with a bounded covariance.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                177,
+                                495,
+                                190
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        177,
+                                        495,
+                                        190
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We present the assumption and main results for covariance bounded distributions in Appendix B.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8,
+                            "is_list_start_line": true,
+                            "is_list_end_line": true
+                        }
+                    ],
+                    "index": 7.5,
+                    "bbox_fs": [
+                        106,
+                        167,
+                        504,
+                        190
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        194,
+                        506,
+                        256
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                193,
+                                506,
+                                210
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        193,
+                                        171,
+                                        210
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Notations. Let",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        171,
+                                        195,
+                                        251,
+                                        207
+                                    ],
+                                    "score": 0.92,
+                                    "content": "[ n ] = \\{ 1 , 2 , \\dots , n \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        251,
+                                        193,
+                                        273,
+                                        210
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". For",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        273,
+                                        194,
+                                        306,
+                                        206
+                                    ],
+                                    "score": 0.9,
+                                    "content": "x \\in \\mathbb { R } ^ { d }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        306,
+                                        193,
+                                        342,
+                                        210
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", we use",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        342,
+                                        194,
+                                        447,
+                                        209
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\textstyle \\| x \\| _ { 2 } = ( \\sum _ { i \\in [ d ] } ( x _ { i } ) ^ { 2 } ) ^ { 1 / 2 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        448,
+                                        193,
+                                        506,
+                                        210
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to denote the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                205,
+                                507,
+                                225
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        205,
+                                        194,
+                                        225
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Euclidean norm. For",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        194,
+                                        208,
+                                        239,
+                                        219
+                                    ],
+                                    "score": 0.91,
+                                    "content": "X \\in \\mathbb { R } ^ { d \\times d }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        240,
+                                        205,
+                                        275,
+                                        225
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", we use",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        275,
+                                        209,
+                                        389,
+                                        222
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\begin{array} { r } { \\| X \\| _ { 2 } = \\operatorname* { m a x } _ { \\| v \\| _ { 2 } = 1 } \\| X v \\| _ { 2 } } \\end{array}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        389,
+                                        205,
+                                        507,
+                                        225
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to denote the spectral norm.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                219,
+                                506,
+                                233
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        219,
+                                        123,
+                                        233
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        124,
+                                        220,
+                                        147,
+                                        231
+                                    ],
+                                    "score": 0.91,
+                                    "content": "d \\times d",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        148,
+                                        219,
+                                        217,
+                                        233
+                                    ],
+                                    "score": 1.0,
+                                    "content": "identity matrix is",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        217,
+                                        221,
+                                        238,
+                                        232
+                                    ],
+                                    "score": 0.9,
+                                    "content": "\\mathbf { I } _ { d \\times d }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        238,
+                                        219,
+                                        407,
+                                        233
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Whenever it is clear from context, we use",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        407,
+                                        221,
+                                        415,
+                                        230
+                                    ],
+                                    "score": 0.82,
+                                    "content": "S",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        415,
+                                        219,
+                                        506,
+                                        233
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to denote both a set of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                231,
+                                506,
+                                246
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        232,
+                                        340,
+                                        246
+                                    ],
+                                    "score": 1.0,
+                                    "content": "data points and also the set of indices of those data points.",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        340,
+                                        231,
+                                        349,
+                                        244
+                                    ],
+                                    "score": 0.85,
+                                    "content": "\\widetilde O",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        349,
+                                        232,
+                                        367,
+                                        246
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        367,
+                                        231,
+                                        376,
+                                        243
+                                    ],
+                                    "score": 0.86,
+                                    "content": "\\widetilde \\Omega",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        376,
+                                        232,
+                                        506,
+                                        246
+                                    ],
+                                    "score": 1.0,
+                                    "content": "hide poly-logarithmic factors in",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                107,
+                                244,
+                                256,
+                                257
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        244,
+                                        144,
+                                        256
+                                    ],
+                                    "score": 0.93,
+                                    "content": "d , n , 1 / \\alpha",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        145,
+                                        244,
+                                        256,
+                                        257
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", and the failure probability.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        }
+                    ],
+                    "index": 11,
+                    "bbox_fs": [
+                        104,
+                        193,
+                        507,
+                        257
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        260,
+                        505,
+                        294
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                260,
+                                505,
+                                273
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        260,
+                                        352,
+                                        273
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Outline. We present PRIME for sub-Gaussian distribution in",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        353,
+                                        261,
+                                        364,
+                                        271
+                                    ],
+                                    "score": 0.76,
+                                    "content": "\\ S 2",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        364,
+                                        260,
+                                        505,
+                                        273
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", and present theoretical analysis in",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                107,
+                                271,
+                                505,
+                                284
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        272,
+                                        118,
+                                        283
+                                    ],
+                                    "score": 0.81,
+                                    "content": "\\ S 3",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        118,
+                                        271,
+                                        437,
+                                        284
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". We then introduce an exponential time algorithm with near optimal guarantee in",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        437,
+                                        272,
+                                        448,
+                                        282
+                                    ],
+                                    "score": 0.83,
+                                    "content": "\\ S 4",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        448,
+                                        271,
+                                        505,
+                                        284
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Due to space",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                282,
+                                462,
+                                295
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        282,
+                                        462,
+                                        295
+                                    ],
+                                    "score": 1.0,
+                                    "content": "constraints, analogous results for heavy-tailed distributions are presented in Appendix B.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        }
+                    ],
+                    "index": 15,
+                    "bbox_fs": [
+                        105,
+                        260,
+                        505,
+                        295
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        106,
+                        308,
+                        449,
+                        322
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                308,
+                                450,
+                                324
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        308,
+                                        450,
+                                        324
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2 PRIME: efficient algorithm for robust and DP mean estimation",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        }
+                    ],
+                    "index": 17
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        333,
+                        505,
+                        356
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                333,
+                                505,
+                                346
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        333,
+                                        505,
+                                        346
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In order to describe the proposed algorithm PRIME, we need to first describe a standard (non-private)",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                345,
+                                328,
+                                357
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        345,
+                                        328,
+                                        357
+                                    ],
+                                    "score": 1.0,
+                                    "content": "iterative filtering algorithm for robust mean estimation.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        }
+                    ],
+                    "index": 18.5,
+                    "bbox_fs": [
+                        105,
+                        333,
+                        505,
+                        357
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        368,
+                        446,
+                        380
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                367,
+                                447,
+                                382
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        367,
+                                        447,
+                                        382
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2.1 Background on (non-private) iterative filtering for robust mean estimation",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        }
+                    ],
+                    "index": 20
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        388,
+                        506,
+                        433
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                389,
+                                507,
+                                402
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        389,
+                                        507,
+                                        402
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Non-private robust mean estimation approaches recursively apply the following filter, whose frame-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                398,
+                                507,
+                                414
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        398,
+                                        288,
+                                        414
+                                    ],
+                                    "score": 1.0,
+                                    "content": "work is first proposed in [28]. Given a dataset",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        288,
+                                        399,
+                                        341,
+                                        412
+                                    ],
+                                    "score": 0.92,
+                                    "content": "S = \\{ x _ { i } \\} _ { i = 1 } ^ { n }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        341,
+                                        398,
+                                        403,
+                                        414
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", the current set",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        403,
+                                        400,
+                                        439,
+                                        411
+                                    ],
+                                    "score": 0.92,
+                                    "content": "S _ { 0 } \\subseteq [ n ]",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        439,
+                                        398,
+                                        507,
+                                        414
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of data points is",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                410,
+                                506,
+                                424
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        410,
+                                        194,
+                                        424
+                                    ],
+                                    "score": 1.0,
+                                    "content": "updated starting with",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        195,
+                                        411,
+                                        231,
+                                        423
+                                    ],
+                                    "score": 0.93,
+                                    "content": "S _ { 1 } = [ n ]",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        231,
+                                        410,
+                                        506,
+                                        424
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". At each step, the following filter (Algorithm 1 in [63]) attempts to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                422,
+                                308,
+                                433
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        422,
+                                        308,
+                                        433
+                                    ],
+                                    "score": 1.0,
+                                    "content": "detect the corrupted data points and remove them.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        }
+                    ],
+                    "index": 22.5,
+                    "bbox_fs": [
+                        104,
+                        389,
+                        507,
+                        433
+                    ]
+                },
+                {
+                    "type": "list",
+                    "bbox": [
+                        129,
+                        435,
+                        505,
+                        515
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                129,
+                                435,
+                                506,
+                                451
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        129,
+                                        435,
+                                        263,
+                                        451
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1. Compute the top eigenvector",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        263,
+                                        436,
+                                        420,
+                                        451
+                                    ],
+                                    "score": 0.88,
+                                    "content": "v _ { t } \\gets \\arg \\operatorname* { m a x } _ { v : \\| v \\| _ { 2 } = 1 } v ^ { \\top } \\mathrm { C o v } ( S _ { t - 1 } ) v",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        420,
+                                        435,
+                                        506,
+                                        451
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of the covariance of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                448,
+                                274,
+                                464
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        448,
+                                        221,
+                                        464
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the current data set",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        221,
+                                        449,
+                                        266,
+                                        462
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\{ x _ { i } \\} _ { i \\in S _ { t - 1 } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        266,
+                                        448,
+                                        274,
+                                        464
+                                    ],
+                                    "score": 1.0,
+                                    "content": ";",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                462,
+                                465,
+                                479
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        462,
+                                        281,
+                                        479
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2. Compute scores for all data points",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        281,
+                                        463,
+                                        465,
+                                        479
+                                    ],
+                                    "score": 0.83,
+                                    "content": "j \\in S _ { t - 1 } \\colon \\tau _ { j } \\gets \\left( v _ { t } ^ { \\top } \\left( x _ { j } - \\mathrm { M e a n } ( S _ { t - 1 } ) \\right) \\right) ^ { 2 } \\ ;",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 27,
+                            "is_list_start_line": true,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                476,
+                                333,
+                                492
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        476,
+                                        250,
+                                        492
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3. Draw a random threshold:",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        250,
+                                        478,
+                                        325,
+                                        491
+                                    ],
+                                    "score": 0.9,
+                                    "content": "Z _ { t } \\gets \\mathrm { U n i f } ( [ 0 , 1 ] )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        325,
+                                        476,
+                                        333,
+                                        492
+                                    ],
+                                    "score": 1.0,
+                                    "content": ";",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28,
+                            "is_list_start_line": true,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                486,
+                                504,
+                                506
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        486,
+                                        230,
+                                        506
+                                    ],
+                                    "score": 1.0,
+                                    "content": "4. Remove outliers from",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        230,
+                                        491,
+                                        251,
+                                        502
+                                    ],
+                                    "score": 0.91,
+                                    "content": "S _ { t - 1 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        252,
+                                        486,
+                                        295,
+                                        506
+                                    ],
+                                    "score": 1.0,
+                                    "content": "defined as",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        295,
+                                        491,
+                                        355,
+                                        502
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\{ i \\in S _ { t - 1 } : \\tau _ { i }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        356,
+                                        486,
+                                        418,
+                                        506
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is in the largest",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        418,
+                                        491,
+                                        431,
+                                        501
+                                    ],
+                                    "score": 0.8,
+                                    "content": "2 \\alpha",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        431,
+                                        486,
+                                        458,
+                                        506
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-tail of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        459,
+                                        490,
+                                        504,
+                                        503
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\{ \\tau _ { j } \\} _ { j \\in S _ { t - 1 } }",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 29,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                499,
+                                338,
+                                516
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        499,
+                                        159,
+                                        516
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        159,
+                                        502,
+                                        218,
+                                        514
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\tau _ { i } \\geq Z _ { t } \\tau _ { \\operatorname* { m a x } } \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        218,
+                                        499,
+                                        248,
+                                        516
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", where",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        248,
+                                        503,
+                                        338,
+                                        514
+                                    ],
+                                    "score": 0.89,
+                                    "content": "\\tau _ { \\mathrm { m a x } } = \\operatorname* { m a x } _ { j \\in S _ { t - 1 } } \\tau _ { j }",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 30,
+                            "is_list_end_line": true
+                        }
+                    ],
+                    "index": 27.5,
+                    "bbox_fs": [
+                        127,
+                        435,
+                        506,
+                        516
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        521,
+                        505,
+                        631
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                521,
+                                506,
+                                534
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        521,
+                                        460,
+                                        534
+                                    ],
+                                    "score": 1.0,
+                                    "content": "This is repeated until the empirical covariance is sufficiently small and the empirical mean",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        460,
+                                        522,
+                                        467,
+                                        533
+                                    ],
+                                    "score": 0.85,
+                                    "content": "\\hat { \\mu }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        467,
+                                        521,
+                                        506,
+                                        534
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is output.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                533,
+                                505,
+                                545
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        533,
+                                        461,
+                                        545
+                                    ],
+                                    "score": 1.0,
+                                    "content": "At a high level, the correctness of this algorithm relies on the key observation that the",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        461,
+                                        534,
+                                        469,
+                                        542
+                                    ],
+                                    "score": 0.79,
+                                    "content": "\\alpha",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        469,
+                                        533,
+                                        505,
+                                        545
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-fraction",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                541,
+                                506,
+                                557
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        541,
+                                        506,
+                                        557
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of adversarial corruption can not significantly change the mean of the dataset without introducing",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                554,
+                                505,
+                                567
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        554,
+                                        505,
+                                        567
+                                    ],
+                                    "score": 1.0,
+                                    "content": "large eigenvalues in the empirical covariance. Therefore, the algorithm finds top eigenvector of the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                565,
+                                506,
+                                577
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        565,
+                                        506,
+                                        577
+                                    ],
+                                    "score": 1.0,
+                                    "content": "empirical covariance in step 1, and tries to correct the empirical covariance by removing corrupted",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                576,
+                                506,
+                                589
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        576,
+                                        506,
+                                        589
+                                    ],
+                                    "score": 1.0,
+                                    "content": "data points. Each data point is assigned a score in step 2 which indicates the “badness” of the data",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                587,
+                                505,
+                                600
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        587,
+                                        204,
+                                        600
+                                    ],
+                                    "score": 1.0,
+                                    "content": "points, and a threshold",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        204,
+                                        587,
+                                        216,
+                                        598
+                                    ],
+                                    "score": 0.88,
+                                    "content": "Z _ { t }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        216,
+                                        587,
+                                        505,
+                                        600
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in step 3 is carefully designed such that step 4 guarantees to remove",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                597,
+                                506,
+                                611
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        597,
+                                        506,
+                                        611
+                                    ],
+                                    "score": 1.0,
+                                    "content": "more corrupted data points than good data points (in expectation). This guarantees the following",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                609,
+                                505,
+                                621
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        609,
+                                        505,
+                                        621
+                                    ],
+                                    "score": 1.0,
+                                    "content": "bound achieving the near-optimal sample complexity shown in the second row of Table 1. A formal",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                619,
+                                357,
+                                632
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        619,
+                                        357,
+                                        632
+                                    ],
+                                    "score": 1.0,
+                                    "content": "description of this algorithm is in Algorithm 4 in Appendix C.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        }
+                    ],
+                    "index": 35.5,
+                    "bbox_fs": [
+                        104,
+                        521,
+                        506,
+                        632
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        633,
+                        504,
+                        660
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                632,
+                                505,
+                                647
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        632,
+                                        381,
+                                        647
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Proposition 2.1 (Corollary of [63, Theorem 2.1]). Under assumption",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        382,
+                                        635,
+                                        387,
+                                        644
+                                    ],
+                                    "score": 0.27,
+                                    "content": "^ { l }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        388,
+                                        632,
+                                        505,
+                                        647
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", the above filtering algorithm",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                645,
+                                409,
+                                661
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        646,
+                                        182,
+                                        661
+                                    ],
+                                    "score": 1.0,
+                                    "content": "achieves accuracy",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        182,
+                                        645,
+                                        406,
+                                        660
+                                    ],
+                                    "score": 0.75,
+                                    "content": "\\| \\hat { \\mu } - \\mu \\| _ { 2 } \\le O ( \\alpha \\sqrt { \\log ( 1 / \\alpha ) } ) \\ w . p . \\ 0 . 9 i f n \\ge \\widetilde { \\Omega } ( d / \\alpha ^ { 2 } ) \\ / .",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        407,
+                                        646,
+                                        409,
+                                        661
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        }
+                    ],
+                    "index": 41.5,
+                    "bbox_fs": [
+                        105,
+                        632,
+                        505,
+                        661
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        667,
+                        505,
+                        722
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                666,
+                                505,
+                                680
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        666,
+                                        505,
+                                        680
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Challenges in making robust mean estimation private. To get a DP and robust mean, a naive",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                678,
+                                505,
+                                690
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        678,
+                                        377,
+                                        690
+                                    ],
+                                    "score": 1.0,
+                                    "content": "attempt is to apply a standard output perturbation mechanism to",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        378,
+                                        678,
+                                        385,
+                                        690
+                                    ],
+                                    "score": 0.83,
+                                    "content": "\\hat { \\mu }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        385,
+                                        678,
+                                        505,
+                                        690
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". However, this is obviously",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                689,
+                                506,
+                                701
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        689,
+                                        506,
+                                        701
+                                    ],
+                                    "score": 1.0,
+                                    "content": "challenging since the end-to-end sensitivity is intractable. The standard recipe to circumvent this is to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                699,
+                                506,
+                                714
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        699,
+                                        203,
+                                        714
+                                    ],
+                                    "score": 1.0,
+                                    "content": "make the current “state”",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        203,
+                                        700,
+                                        214,
+                                        711
+                                    ],
+                                    "score": 0.87,
+                                    "content": "S _ { t }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        214,
+                                        699,
+                                        338,
+                                        714
+                                    ],
+                                    "score": 1.0,
+                                    "content": "private at every iteration. Once",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        338,
+                                        700,
+                                        359,
+                                        712
+                                    ],
+                                    "score": 0.91,
+                                    "content": "S _ { t - 1 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        359,
+                                        699,
+                                        506,
+                                        714
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is private (hence, public knowledge),",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                709,
+                                506,
+                                725
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        709,
+                                        201,
+                                        725
+                                    ],
+                                    "score": 1.0,
+                                    "content": "making the next “state”",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        201,
+                                        711,
+                                        212,
+                                        722
+                                    ],
+                                    "score": 0.86,
+                                    "content": "S _ { t }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        213,
+                                        709,
+                                        506,
+                                        725
+                                    ],
+                                    "score": 1.0,
+                                    "content": "private is simpler. We only need to analyze the sensitivity of a single step",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 47
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                73,
+                                505,
+                                86
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        73,
+                                        322,
+                                        86
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and apply some output perturbation mechanism with",
+                                    "type": "text",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        323,
+                                        73,
+                                        351,
+                                        85
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\left( \\varepsilon _ { t } , \\delta _ { t } \\right)",
+                                    "type": "inline_equation",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        352,
+                                        73,
+                                        505,
+                                        86
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". End-to-end privacy is guaranteed by",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                83,
+                                505,
+                                96
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        83,
+                                        204,
+                                        96
+                                    ],
+                                    "score": 1.0,
+                                    "content": "accounting for all these",
+                                    "type": "text",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        204,
+                                        84,
+                                        233,
+                                        96
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\left( \\varepsilon _ { t } , \\delta _ { t } \\right)",
+                                    "type": "inline_equation",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        234,
+                                        83,
+                                        505,
+                                        96
+                                    ],
+                                    "score": 1.0,
+                                    "content": "’s using the advanced composition [51]. This recipe has been quite",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                93,
+                                505,
+                                108
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        93,
+                                        505,
+                                        108
+                                    ],
+                                    "score": 1.0,
+                                    "content": "successful, for example, in training neural networks with (stochastic) gradient descent [1], where",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                104,
+                                506,
+                                120
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        104,
+                                        304,
+                                        120
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the current state can be the optimization variable",
+                                    "type": "text",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        304,
+                                        106,
+                                        315,
+                                        117
+                                    ],
+                                    "score": 0.85,
+                                    "content": "\\mathbf { x } _ { t }",
+                                    "type": "inline_equation",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        315,
+                                        104,
+                                        506,
+                                        120
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". However, for the above (non-private) filtering",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                115,
+                                507,
+                                130
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        115,
+                                        312,
+                                        130
+                                    ],
+                                    "score": 1.0,
+                                    "content": "algorithm, this standard recipe fails, since the state",
+                                    "type": "text",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        313,
+                                        117,
+                                        324,
+                                        127
+                                    ],
+                                    "score": 0.88,
+                                    "content": "S _ { t }",
+                                    "type": "inline_equation",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        324,
+                                        115,
+                                        507,
+                                        130
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is a set and has large sensitivity. Changing a",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                127,
+                                478,
+                                140
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        127,
+                                        185,
+                                        140
+                                    ],
+                                    "score": 1.0,
+                                    "content": "single data point in",
+                                    "type": "text",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        185,
+                                        128,
+                                        196,
+                                        138
+                                    ],
+                                    "score": 0.88,
+                                    "content": "S _ { t }",
+                                    "type": "inline_equation",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        196,
+                                        127,
+                                        478,
+                                        140
+                                    ],
+                                    "score": 1.0,
+                                    "content": "can significantly alter which (and how many) samples are filtered out.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 5
+                        }
+                    ],
+                    "index": 45,
+                    "bbox_fs": [
+                        105,
+                        666,
+                        506,
+                        725
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        72,
+                        505,
+                        139
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                73,
+                                505,
+                                86
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        73,
+                                        322,
+                                        86
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and apply some output perturbation mechanism with",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        323,
+                                        73,
+                                        351,
+                                        85
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\left( \\varepsilon _ { t } , \\delta _ { t } \\right)",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        352,
+                                        73,
+                                        505,
+                                        86
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". End-to-end privacy is guaranteed by",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                83,
+                                505,
+                                96
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        83,
+                                        204,
+                                        96
+                                    ],
+                                    "score": 1.0,
+                                    "content": "accounting for all these",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        204,
+                                        84,
+                                        233,
+                                        96
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\left( \\varepsilon _ { t } , \\delta _ { t } \\right)",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        234,
+                                        83,
+                                        505,
+                                        96
+                                    ],
+                                    "score": 1.0,
+                                    "content": "’s using the advanced composition [51]. This recipe has been quite",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                93,
+                                505,
+                                108
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        93,
+                                        505,
+                                        108
+                                    ],
+                                    "score": 1.0,
+                                    "content": "successful, for example, in training neural networks with (stochastic) gradient descent [1], where",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                104,
+                                506,
+                                120
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        104,
+                                        304,
+                                        120
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the current state can be the optimization variable",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        304,
+                                        106,
+                                        315,
+                                        117
+                                    ],
+                                    "score": 0.85,
+                                    "content": "\\mathbf { x } _ { t }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        315,
+                                        104,
+                                        506,
+                                        120
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". However, for the above (non-private) filtering",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                115,
+                                507,
+                                130
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        115,
+                                        312,
+                                        130
+                                    ],
+                                    "score": 1.0,
+                                    "content": "algorithm, this standard recipe fails, since the state",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        313,
+                                        117,
+                                        324,
+                                        127
+                                    ],
+                                    "score": 0.88,
+                                    "content": "S _ { t }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        324,
+                                        115,
+                                        507,
+                                        130
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is a set and has large sensitivity. Changing a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                127,
+                                478,
+                                140
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        127,
+                                        185,
+                                        140
+                                    ],
+                                    "score": 1.0,
+                                    "content": "single data point in",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        185,
+                                        128,
+                                        196,
+                                        138
+                                    ],
+                                    "score": 0.88,
+                                    "content": "S _ { t }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        196,
+                                        127,
+                                        478,
+                                        140
+                                    ],
+                                    "score": 1.0,
+                                    "content": "can significantly alter which (and how many) samples are filtered out.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        }
+                    ],
+                    "index": 2.5
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        156,
+                        327,
+                        168
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                154,
+                                329,
+                                171
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        154,
+                                        329,
+                                        171
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2.2 A new framework for private iterative filtering",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        }
+                    ],
+                    "index": 6
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        177,
+                        505,
+                        276
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                177,
+                                505,
+                                190
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        177,
+                                        266,
+                                        190
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Instead of making the (highly sensitive)",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        266,
+                                        178,
+                                        276,
+                                        189
+                                    ],
+                                    "score": 0.88,
+                                    "content": "S _ { t }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        277,
+                                        177,
+                                        505,
+                                        190
+                                    ],
+                                    "score": 1.0,
+                                    "content": "itself private, we propose a new framework which makes",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                189,
+                                505,
+                                202
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        189,
+                                        216,
+                                        202
+                                    ],
+                                    "score": 1.0,
+                                    "content": "private only the statistics of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        216,
+                                        189,
+                                        226,
+                                        200
+                                    ],
+                                    "score": 0.88,
+                                    "content": "S _ { t }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        227,
+                                        189,
+                                        268,
+                                        202
+                                    ],
+                                    "score": 1.0,
+                                    "content": ": the mean",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        269,
+                                        191,
+                                        280,
+                                        200
+                                    ],
+                                    "score": 0.85,
+                                    "content": "\\mu _ { t }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        280,
+                                        189,
+                                        400,
+                                        202
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and the top principal direction",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        400,
+                                        190,
+                                        410,
+                                        200
+                                    ],
+                                    "score": 0.83,
+                                    "content": "v _ { t }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        410,
+                                        189,
+                                        505,
+                                        202
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". There are two versions",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                199,
+                                505,
+                                213
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        199,
+                                        300,
+                                        213
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of this algorithm, which output the exactly same",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        300,
+                                        200,
+                                        307,
+                                        212
+                                    ],
+                                    "score": 0.84,
+                                    "content": "\\hat { \\mu }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        308,
+                                        199,
+                                        505,
+                                        213
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with the exactly same privacy guarantees, but are",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                210,
+                                505,
+                                223
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        210,
+                                        505,
+                                        223
+                                    ],
+                                    "score": 1.0,
+                                    "content": "written from two different perspectives. We present here the interactive version from the perspective",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                221,
+                                506,
+                                236
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        221,
+                                        305,
+                                        236
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of an analyst accessing the dataset via DP queries",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        305,
+                                        223,
+                                        330,
+                                        234
+                                    ],
+                                    "score": 0.25,
+                                    "content": "\\cdot q _ { \\mathrm { r a n g e } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        330,
+                                        221,
+                                        333,
+                                        236
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        334,
+                                        223,
+                                        352,
+                                        233
+                                    ],
+                                    "score": 0.43,
+                                    "content": "q _ { \\mathrm { s i z e } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        352,
+                                        221,
+                                        355,
+                                        236
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        356,
+                                        223,
+                                        380,
+                                        233
+                                    ],
+                                    "score": 0.55,
+                                    "content": "q _ { \\mathrm { m e a n } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        380,
+                                        221,
+                                        383,
+                                        236
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        383,
+                                        223,
+                                        408,
+                                        233
+                                    ],
+                                    "score": 0.61,
+                                    "content": "q _ { \\mathrm { n o r m } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        408,
+                                        221,
+                                        425,
+                                        236
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        425,
+                                        222,
+                                        449,
+                                        233
+                                    ],
+                                    "score": 0.81,
+                                    "content": "q _ { \\mathrm { P C A } } ,",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        450,
+                                        221,
+                                        506,
+                                        236
+                                    ],
+                                    "score": 1.0,
+                                    "content": "), because this",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                232,
+                                505,
+                                246
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        232,
+                                        435,
+                                        246
+                                    ],
+                                    "score": 1.0,
+                                    "content": "version makes clear the inner operations of each private mechanisms, hence making",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        436,
+                                        233,
+                                        447,
+                                        245
+                                    ],
+                                    "score": 0.68,
+                                    "content": "( i )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        447,
+                                        232,
+                                        505,
+                                        246
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the sensitivity",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                241,
+                                506,
+                                258
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        241,
+                                        194,
+                                        258
+                                    ],
+                                    "score": 1.0,
+                                    "content": "analysis transparent,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        194,
+                                        244,
+                                        209,
+                                        255
+                                    ],
+                                    "score": 0.71,
+                                    "content": "( i i )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        209,
+                                        241,
+                                        448,
+                                        258
+                                    ],
+                                    "score": 1.0,
+                                    "content": "checking the correctness of privacy guarantees easy, and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        449,
+                                        244,
+                                        467,
+                                        255
+                                    ],
+                                    "score": 0.74,
+                                    "content": "( i i i )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        467,
+                                        241,
+                                        506,
+                                        258
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tracking",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                254,
+                                505,
+                                267
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        254,
+                                        505,
+                                        267
+                                    ],
+                                    "score": 1.0,
+                                    "content": "privacy accountant simple. In practice, one should implement the centralized version (Algorithm 7 in",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                265,
+                                311,
+                                278
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        265,
+                                        311,
+                                        278
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Appendix D), which is significantly more efficient.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        }
+                    ],
+                    "index": 11
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        106,
+                        296,
+                        345,
+                        307
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                295,
+                                345,
+                                308
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        295,
+                                        345,
+                                        308
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Algorithm 1: Private iterative filtering (interactive version)",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        }
+                    ],
+                    "index": 16
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        98,
+                        309,
+                        475,
+                        442
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                308,
+                                474,
+                                324
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        308,
+                                        139,
+                                        324
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Input:",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        140,
+                                        310,
+                                        197,
+                                        323
+                                    ],
+                                    "score": 0.87,
+                                    "content": "S = \\{ x _ { i } \\} _ { i \\in [ n ] }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        198,
+                                        308,
+                                        201,
+                                        324
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        201,
+                                        310,
+                                        253,
+                                        322
+                                    ],
+                                    "score": 0.87,
+                                    "content": "\\alpha \\in ( 0 , 1 / 2 )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        253,
+                                        308,
+                                        303,
+                                        324
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", probability",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        303,
+                                        310,
+                                        343,
+                                        322
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\zeta \\in ( 0 , 1 )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        344,
+                                        308,
+                                        405,
+                                        324
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", # of iterations",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        406,
+                                        310,
+                                        474,
+                                        322
+                                    ],
+                                    "score": 0.66,
+                                    "content": "T = \\Theta ( d ) , ( \\varepsilon , \\delta )",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                97,
+                                321,
+                                240,
+                                335
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        97,
+                                        321,
+                                        107,
+                                        335
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        107,
+                                        323,
+                                        240,
+                                        335
+                                    ],
+                                    "score": 0.88,
+                                    "content": "( \\bar { x } , B ) \\gets q _ { \\mathrm { r a n g e } } ( S , 0 . 0 1 \\varepsilon , 0 . 0 1 \\delta )",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                96,
+                                334,
+                                356,
+                                349
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        96,
+                                        334,
+                                        106,
+                                        349
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        335,
+                                        273,
+                                        348
+                                    ],
+                                    "score": 0.62,
+                                    "content": "\\varepsilon _ { 1 } \\gets \\operatorname* { m i n } \\{ 0 . 9 9 \\varepsilon , 0 . 9 \\} / ( 4 \\sqrt { 2 T \\log ( 2 / \\delta ) } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        273,
+                                        334,
+                                        356,
+                                        349
+                                    ],
+                                    "score": 1.0,
+                                    "content": "), δ1 ← 0.99δ/(8T )",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                96,
+                                345,
+                                282,
+                                360
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        96,
+                                        345,
+                                        114,
+                                        360
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3 if",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        115,
+                                        348,
+                                        215,
+                                        359
+                                    ],
+                                    "score": 0.83,
+                                    "content": "n < ( 4 / \\varepsilon _ { 1 } ) \\log ( 1 / ( 2 \\delta _ { 1 } ) )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        215,
+                                        345,
+                                        276,
+                                        360
+                                    ],
+                                    "score": 1.0,
+                                    "content": "then Output:",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        276,
+                                        347,
+                                        282,
+                                        357
+                                    ],
+                                    "score": 0.32,
+                                    "content": "\\varnothing",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                97,
+                                358,
+                                189,
+                                369
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        97,
+                                        358,
+                                        121,
+                                        369
+                                    ],
+                                    "score": 1.0,
+                                    "content": "4 for",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        122,
+                                        359,
+                                        174,
+                                        369
+                                    ],
+                                    "score": 0.85,
+                                    "content": "t = 1 , \\dots , T",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        174,
+                                        358,
+                                        189,
+                                        369
+                                    ],
+                                    "score": 1.0,
+                                    "content": "do",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                98,
+                                368,
+                                416,
+                                381
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        98,
+                                        370,
+                                        105,
+                                        379
+                                    ],
+                                    "score": 1.0,
+                                    "content": "5",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        122,
+                                        369,
+                                        292,
+                                        381
+                                    ],
+                                    "score": 0.89,
+                                    "content": "n _ { t }  q _ { \\mathrm { s i z e } } ( \\{ ( \\mu _ { \\ell } , v _ { \\ell } , Z _ { \\ell } ) \\} _ { \\ell \\in [ t - 1 ] } , \\varepsilon _ { 1 } , \\bar { x } , B )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        293,
+                                        368,
+                                        304,
+                                        381
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", if",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        304,
+                                        369,
+                                        349,
+                                        380
+                                    ],
+                                    "score": 0.89,
+                                    "content": "n _ { t } < 3 n / 4",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        350,
+                                        368,
+                                        409,
+                                        381
+                                    ],
+                                    "score": 1.0,
+                                    "content": "then Output:",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        409,
+                                        369,
+                                        416,
+                                        379
+                                    ],
+                                    "score": 0.69,
+                                    "content": "\\varnothing",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                98,
+                                382,
+                                298,
+                                393
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        98,
+                                        383,
+                                        105,
+                                        392
+                                    ],
+                                    "score": 1.0,
+                                    "content": "6",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        121,
+                                        382,
+                                        298,
+                                        393
+                                    ],
+                                    "score": 0.85,
+                                    "content": "\\mu _ { t }  q _ { \\mathrm { m e a n } } ( \\{ ( \\mu _ { \\ell } , v _ { \\ell } , Z _ { \\ell } ) \\} _ { \\ell \\in [ t - 1 ] } , \\varepsilon _ { 1 } , \\bar { x } , B )",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                98,
+                                394,
+                                312,
+                                405
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        98,
+                                        395,
+                                        105,
+                                        404
+                                    ],
+                                    "score": 1.0,
+                                    "content": "7",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        122,
+                                        394,
+                                        312,
+                                        405
+                                    ],
+                                    "score": 0.84,
+                                    "content": "\\lambda _ { t } \\gets q _ { \\mathrm { n o r m } } ( \\{ ( \\mu _ { \\ell } , v _ { \\ell } , Z _ { \\ell } ) \\} _ { \\ell \\in [ t - 1 ] } , \\mu _ { t } , \\varepsilon _ { 1 } , \\bar { x } , B )",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                98,
+                                403,
+                                309,
+                                419
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        98,
+                                        407,
+                                        105,
+                                        416
+                                    ],
+                                    "score": 1.0,
+                                    "content": "8",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        120,
+                                        403,
+                                        130,
+                                        419
+                                    ],
+                                    "score": 1.0,
+                                    "content": "if",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        131,
+                                        406,
+                                        239,
+                                        417
+                                    ],
+                                    "score": 0.84,
+                                    "content": "\\lambda _ { t } \\le ( C - 0 . 0 1 ) \\alpha \\log { 1 / \\alpha }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        239,
+                                        403,
+                                        298,
+                                        419
+                                    ],
+                                    "score": 1.0,
+                                    "content": "then Output:",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        299,
+                                        407,
+                                        309,
+                                        417
+                                    ],
+                                    "score": 0.36,
+                                    "content": "\\mu _ { t }",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                97,
+                                417,
+                                328,
+                                429
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        97,
+                                        418,
+                                        105,
+                                        426
+                                    ],
+                                    "score": 1.0,
+                                    "content": "9",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        122,
+                                        417,
+                                        328,
+                                        429
+                                    ],
+                                    "score": 0.71,
+                                    "content": "\\boldsymbol { v } _ { t }  q _ { \\mathrm { P C A } } ( \\{ ( \\mu _ { \\ell } , v _ { \\ell } , Z _ { \\ell } ) \\} _ { \\ell \\in [ t - 1 ] } , \\mu _ { t } , \\varepsilon _ { 1 } , \\delta _ { 1 } , \\bar { x } , B ) )",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                96,
+                                429,
+                                195,
+                                441
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        96,
+                                        430,
+                                        105,
+                                        439
+                                    ],
+                                    "score": 1.0,
+                                    "content": "10",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        122,
+                                        429,
+                                        195,
+                                        441
+                                    ],
+                                    "score": 0.83,
+                                    "content": "Z _ { t } \\gets \\mathrm { U n i f } ( [ 0 , 1 ] )",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 27
+                        }
+                    ],
+                    "index": 22
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        443,
+                        155,
+                        454
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                440,
+                                154,
+                                457
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        440,
+                                        143,
+                                        457
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Output:",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        144,
+                                        444,
+                                        154,
+                                        454
+                                    ],
+                                    "score": 0.64,
+                                    "content": "\\mu _ { t }",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 28
+                        }
+                    ],
+                    "index": 28
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        474,
+                        505,
+                        580
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                474,
+                                506,
+                                488
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        474,
+                                        506,
+                                        488
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We give a high-level explanation of each step of Algorithm 1 here and give the formal definitions of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                484,
+                                504,
+                                500
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        484,
+                                        247,
+                                        500
+                                    ],
+                                    "score": 1.0,
+                                    "content": "all the queries in Appendix D. First,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        248,
+                                        487,
+                                        273,
+                                        498
+                                    ],
+                                    "score": 0.84,
+                                    "content": "q _ { \\mathrm { r a n g e } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        273,
+                                        484,
+                                        428,
+                                        500
+                                    ],
+                                    "score": 1.0,
+                                    "content": "returns (the parameters of) a hypercube",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        429,
+                                        485,
+                                        504,
+                                        498
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\bar { x } + [ - B / 2 , B / 2 ] ^ { d }",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                496,
+                                506,
+                                509
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        496,
+                                        506,
+                                        509
+                                    ],
+                                    "score": 1.0,
+                                    "content": "that is guaranteed to include all uncorrupted samples while preserving privacy. This is achieved",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                507,
+                                505,
+                                521
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        507,
+                                        154,
+                                        521
+                                    ],
+                                    "score": 1.0,
+                                    "content": "by running",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        154,
+                                        508,
+                                        161,
+                                        518
+                                    ],
+                                    "score": 0.72,
+                                    "content": "d",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        162,
+                                        507,
+                                        365,
+                                        521
+                                    ],
+                                    "score": 1.0,
+                                    "content": "coordinate-wise private histograms and selecting",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        365,
+                                        509,
+                                        376,
+                                        520
+                                    ],
+                                    "score": 0.87,
+                                    "content": "\\bar { x } _ { j }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        377,
+                                        507,
+                                        505,
+                                        521
+                                    ],
+                                    "score": 1.0,
+                                    "content": "as the center of the largest bin",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                519,
+                                506,
+                                534
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        519,
+                                        136,
+                                        534
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for the",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        137,
+                                        521,
+                                        143,
+                                        532
+                                    ],
+                                    "score": 0.82,
+                                    "content": "j",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        143,
+                                        519,
+                                        298,
+                                        534
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-th coordinate. Since covariance is I,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        299,
+                                        522,
+                                        324,
+                                        532
+                                    ],
+                                    "score": 0.88,
+                                    "content": "q _ { \\mathrm { r a n g e } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        325,
+                                        519,
+                                        388,
+                                        534
+                                    ],
+                                    "score": 1.0,
+                                    "content": "returns a fixed",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        388,
+                                        519,
+                                        476,
+                                        533
+                                    ],
+                                    "score": 0.94,
+                                    "content": "B = 8 \\sigma \\sqrt { \\log ( d n / \\zeta ) }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        476,
+                                        519,
+                                        506,
+                                        534
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Such",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                531,
+                                506,
+                                546
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        531,
+                                        506,
+                                        546
+                                    ],
+                                    "score": 1.0,
+                                    "content": "an adaptive estimate of the support is critical in tightly bounding the sensitivity of all subsequent",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                541,
+                                505,
+                                558
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        541,
+                                        414,
+                                        558
+                                    ],
+                                    "score": 1.0,
+                                    "content": "queries, which operate on the clipped dataset; all data points are projected as",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        415,
+                                        543,
+                                        505,
+                                        556
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\mathcal { P } _ { \\bar { x } + [ - B / 2 , B / 2 ] ^ { d } } ( x ) =",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                554,
+                                506,
+                                568
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        554,
+                                        122,
+                                        568
+                                    ],
+                                    "score": 1.0,
+                                    "content": "arg",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        123,
+                                        555,
+                                        204,
+                                        568
+                                    ],
+                                    "score": 0.66,
+                                    "content": "\\begin{array} { r } { \\operatorname* { m i n } _ { y \\in \\bar { x } + [ - B / 2 , B / 2 ] ^ { c } } } \\end{array}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        208,
+                                        555,
+                                        246,
+                                        567
+                                    ],
+                                    "score": 0.84,
+                                    "content": "\\lVert y - x \\rVert _ { 2 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        246,
+                                        554,
+                                        506,
+                                        568
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in all the queries that follow. With clipping, a single data point",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                568,
+                                244,
+                                581
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        568,
+                                        217,
+                                        581
+                                    ],
+                                    "score": 1.0,
+                                    "content": "can now change at most by",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        217,
+                                        568,
+                                        240,
+                                        580
+                                    ],
+                                    "score": 0.9,
+                                    "content": "B \\sqrt { d }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        240,
+                                        568,
+                                        244,
+                                        581
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        }
+                    ],
+                    "index": 33
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        585,
+                        505,
+                        619
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                583,
+                                504,
+                                599
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        583,
+                                        371,
+                                        599
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The subsequent steps perform the non-private filtering algorithm of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        372,
+                                        586,
+                                        389,
+                                        596
+                                    ],
+                                    "score": 0.88,
+                                    "content": "\\ S 2 . 1",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        390,
+                                        583,
+                                        493,
+                                        599
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", but with private statistics",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        493,
+                                        587,
+                                        504,
+                                        596
+                                    ],
+                                    "score": 0.8,
+                                    "content": "\\mu _ { t }",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                595,
+                                506,
+                                609
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        595,
+                                        123,
+                                        609
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        123,
+                                        597,
+                                        132,
+                                        607
+                                    ],
+                                    "score": 0.84,
+                                    "content": "v _ { t }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        133,
+                                        595,
+                                        178,
+                                        609
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". As the set",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        178,
+                                        597,
+                                        189,
+                                        607
+                                    ],
+                                    "score": 0.88,
+                                    "content": "S _ { t }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        189,
+                                        595,
+                                        456,
+                                        609
+                                    ],
+                                    "score": 1.0,
+                                    "content": "changes over time, we lower bound its size (which we choose to be",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        456,
+                                        596,
+                                        503,
+                                        608
+                                    ],
+                                    "score": 0.9,
+                                    "content": "| S _ { t } | > n / 2 )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        503,
+                                        595,
+                                        506,
+                                        609
+                                    ],
+                                    "score": 1.0,
+                                    "content": ")",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                606,
+                                390,
+                                621
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        606,
+                                        292,
+                                        621
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to upper bound the sensitivity of other queries",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        292,
+                                        609,
+                                        344,
+                                        619
+                                    ],
+                                    "score": 0.87,
+                                    "content": "q _ { \\mathrm { m e a n } } , q _ { \\mathrm { n o r m } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        345,
+                                        606,
+                                        362,
+                                        621
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        362,
+                                        609,
+                                        385,
+                                        619
+                                    ],
+                                    "score": 0.81,
+                                    "content": "q _ { \\mathrm { P C A } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        385,
+                                        606,
+                                        390,
+                                        621
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        }
+                    ],
+                    "index": 39
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        623,
+                        505,
+                        722
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                623,
+                                505,
+                                636
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        623,
+                                        132,
+                                        636
+                                    ],
+                                    "score": 1.0,
+                                    "content": "At the",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        133,
+                                        624,
+                                        138,
+                                        633
+                                    ],
+                                    "score": 0.76,
+                                    "content": "t",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        138,
+                                        623,
+                                        365,
+                                        636
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-th iterations, every time a query is called the data curator",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        365,
+                                        623,
+                                        377,
+                                        635
+                                    ],
+                                    "score": 0.78,
+                                    "content": "( i )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        377,
+                                        623,
+                                        398,
+                                        636
+                                    ],
+                                    "score": 1.0,
+                                    "content": "uses",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        398,
+                                        624,
+                                        424,
+                                        635
+                                    ],
+                                    "score": 0.92,
+                                    "content": "( { \\bar { x } } , B )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        425,
+                                        623,
+                                        489,
+                                        636
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to clip the data,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        489,
+                                        623,
+                                        505,
+                                        636
+                                    ],
+                                    "score": 0.81,
+                                    "content": "( i i )",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                634,
+                                506,
+                                648
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        634,
+                                        148,
+                                        648
+                                    ],
+                                    "score": 1.0,
+                                    "content": "computes",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        148,
+                                        635,
+                                        159,
+                                        645
+                                    ],
+                                    "score": 0.88,
+                                    "content": "S _ { t }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        159,
+                                        634,
+                                        207,
+                                        648
+                                    ],
+                                    "score": 1.0,
+                                    "content": "by running",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        207,
+                                        635,
+                                        230,
+                                        645
+                                    ],
+                                    "score": 0.86,
+                                    "content": "t - 1",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        230,
+                                        634,
+                                        417,
+                                        648
+                                    ],
+                                    "score": 1.0,
+                                    "content": "steps of the non-private filtering algorithm of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        417,
+                                        635,
+                                        435,
+                                        645
+                                    ],
+                                    "score": 0.86,
+                                    "content": "\\ S 2 . 1",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        436,
+                                        634,
+                                        506,
+                                        648
+                                    ],
+                                    "score": 1.0,
+                                    "content": "but with a given",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                644,
+                                506,
+                                660
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        644,
+                                        196,
+                                        660
+                                    ],
+                                    "score": 1.0,
+                                    "content": "fixed set of parameters",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        197,
+                                        645,
+                                        264,
+                                        658
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\{ ( \\mu _ { \\ell } , v _ { \\ell } ) \\} _ { \\ell \\in [ t - 1 ] }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        265,
+                                        644,
+                                        373,
+                                        660
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(and the given randomness",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        373,
+                                        645,
+                                        424,
+                                        658
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\{ Z _ { \\ell } \\} _ { \\ell \\in [ t - 1 ] } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        425,
+                                        644,
+                                        445,
+                                        660
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        446,
+                                        645,
+                                        464,
+                                        657
+                                    ],
+                                    "score": 0.88,
+                                    "content": "( i i i )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        464,
+                                        644,
+                                        506,
+                                        660
+                                    ],
+                                    "score": 1.0,
+                                    "content": "computes",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                656,
+                                506,
+                                669
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        656,
+                                        234,
+                                        669
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the queried private statistics of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        234,
+                                        658,
+                                        245,
+                                        667
+                                    ],
+                                    "score": 0.87,
+                                    "content": "S _ { t }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        245,
+                                        656,
+                                        447,
+                                        669
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". If the private spectral norm of the covariance of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        447,
+                                        657,
+                                        458,
+                                        667
+                                    ],
+                                    "score": 0.87,
+                                    "content": "S _ { t }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        458,
+                                        656,
+                                        480,
+                                        669
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(i.e.,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        480,
+                                        657,
+                                        492,
+                                        667
+                                    ],
+                                    "score": 0.81,
+                                    "content": "\\lambda _ { t } ,",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        492,
+                                        656,
+                                        506,
+                                        669
+                                    ],
+                                    "score": 1.0,
+                                    "content": ") is",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                667,
+                                505,
+                                680
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        667,
+                                        330,
+                                        680
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sufficiently small, we output the private and robust mean",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        331,
+                                        668,
+                                        360,
+                                        679
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\hat { \\mu } = \\mu _ { t }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        361,
+                                        667,
+                                        505,
+                                        680
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(line 8). Otherwise, we compute the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                678,
+                                505,
+                                691
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        678,
+                                        209,
+                                        691
+                                    ],
+                                    "score": 1.0,
+                                    "content": "private top PCA direction",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        210,
+                                        680,
+                                        219,
+                                        689
+                                    ],
+                                    "score": 0.83,
+                                    "content": "v _ { t }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        219,
+                                        678,
+                                        320,
+                                        691
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and draw an randomness",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        321,
+                                        679,
+                                        331,
+                                        689
+                                    ],
+                                    "score": 0.89,
+                                    "content": "Z _ { t }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        332,
+                                        678,
+                                        505,
+                                        691
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to be used in the next step of filtering, as in",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                689,
+                                505,
+                                701
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        689,
+                                        321,
+                                        701
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the non-private filtering algorithm. We emphasize that",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        322,
+                                        689,
+                                        342,
+                                        701
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\{ S _ { \\ell } \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        342,
+                                        689,
+                                        505,
+                                        701
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are not private, and hence never returned",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 47
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                699,
+                                505,
+                                714
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        699,
+                                        505,
+                                        714
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to the analyst. We also note that this interactive version is redundant as every query is re-computing",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 48
+                        },
+                        {
+                            "bbox": [
+                                107,
+                                711,
+                                506,
+                                723
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        711,
+                                        117,
+                                        722
+                                    ],
+                                    "score": 0.86,
+                                    "content": "S _ { t }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        117,
+                                        711,
+                                        506,
+                                        723
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". In our setting, the analyst has the dataset and there is no need to separate them. This leads to a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 49
+                        }
+                    ],
+                    "index": 45
+                }
+            ],
+            "page_idx": 4,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        302,
+                        741,
+                        308,
+                        750
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                302,
+                                740,
+                                309,
+                                753
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        302,
+                                        740,
+                                        309,
+                                        753
+                                    ],
+                                    "score": 1.0,
+                                    "content": "5",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        72,
+                        505,
+                        139
+                    ],
+                    "lines": [],
+                    "index": 2.5,
+                    "bbox_fs": [
+                        105,
+                        73,
+                        507,
+                        140
+                    ],
+                    "lines_deleted": true
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        156,
+                        327,
+                        168
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                154,
+                                329,
+                                171
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        154,
+                                        329,
+                                        171
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2.2 A new framework for private iterative filtering",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        }
+                    ],
+                    "index": 6
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        177,
+                        505,
+                        276
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                177,
+                                505,
+                                190
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        177,
+                                        266,
+                                        190
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Instead of making the (highly sensitive)",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        266,
+                                        178,
+                                        276,
+                                        189
+                                    ],
+                                    "score": 0.88,
+                                    "content": "S _ { t }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        277,
+                                        177,
+                                        505,
+                                        190
+                                    ],
+                                    "score": 1.0,
+                                    "content": "itself private, we propose a new framework which makes",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                189,
+                                505,
+                                202
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        189,
+                                        216,
+                                        202
+                                    ],
+                                    "score": 1.0,
+                                    "content": "private only the statistics of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        216,
+                                        189,
+                                        226,
+                                        200
+                                    ],
+                                    "score": 0.88,
+                                    "content": "S _ { t }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        227,
+                                        189,
+                                        268,
+                                        202
+                                    ],
+                                    "score": 1.0,
+                                    "content": ": the mean",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        269,
+                                        191,
+                                        280,
+                                        200
+                                    ],
+                                    "score": 0.85,
+                                    "content": "\\mu _ { t }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        280,
+                                        189,
+                                        400,
+                                        202
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and the top principal direction",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        400,
+                                        190,
+                                        410,
+                                        200
+                                    ],
+                                    "score": 0.83,
+                                    "content": "v _ { t }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        410,
+                                        189,
+                                        505,
+                                        202
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". There are two versions",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                199,
+                                505,
+                                213
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        199,
+                                        300,
+                                        213
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of this algorithm, which output the exactly same",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        300,
+                                        200,
+                                        307,
+                                        212
+                                    ],
+                                    "score": 0.84,
+                                    "content": "\\hat { \\mu }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        308,
+                                        199,
+                                        505,
+                                        213
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with the exactly same privacy guarantees, but are",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                210,
+                                505,
+                                223
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        210,
+                                        505,
+                                        223
+                                    ],
+                                    "score": 1.0,
+                                    "content": "written from two different perspectives. We present here the interactive version from the perspective",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                221,
+                                506,
+                                236
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        221,
+                                        305,
+                                        236
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of an analyst accessing the dataset via DP queries",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        305,
+                                        223,
+                                        330,
+                                        234
+                                    ],
+                                    "score": 0.25,
+                                    "content": "\\cdot q _ { \\mathrm { r a n g e } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        330,
+                                        221,
+                                        333,
+                                        236
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        334,
+                                        223,
+                                        352,
+                                        233
+                                    ],
+                                    "score": 0.43,
+                                    "content": "q _ { \\mathrm { s i z e } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        352,
+                                        221,
+                                        355,
+                                        236
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        356,
+                                        223,
+                                        380,
+                                        233
+                                    ],
+                                    "score": 0.55,
+                                    "content": "q _ { \\mathrm { m e a n } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        380,
+                                        221,
+                                        383,
+                                        236
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        383,
+                                        223,
+                                        408,
+                                        233
+                                    ],
+                                    "score": 0.61,
+                                    "content": "q _ { \\mathrm { n o r m } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        408,
+                                        221,
+                                        425,
+                                        236
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        425,
+                                        222,
+                                        449,
+                                        233
+                                    ],
+                                    "score": 0.81,
+                                    "content": "q _ { \\mathrm { P C A } } ,",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        450,
+                                        221,
+                                        506,
+                                        236
+                                    ],
+                                    "score": 1.0,
+                                    "content": "), because this",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                232,
+                                505,
+                                246
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        232,
+                                        435,
+                                        246
+                                    ],
+                                    "score": 1.0,
+                                    "content": "version makes clear the inner operations of each private mechanisms, hence making",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        436,
+                                        233,
+                                        447,
+                                        245
+                                    ],
+                                    "score": 0.68,
+                                    "content": "( i )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        447,
+                                        232,
+                                        505,
+                                        246
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the sensitivity",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                241,
+                                506,
+                                258
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        241,
+                                        194,
+                                        258
+                                    ],
+                                    "score": 1.0,
+                                    "content": "analysis transparent,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        194,
+                                        244,
+                                        209,
+                                        255
+                                    ],
+                                    "score": 0.71,
+                                    "content": "( i i )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        209,
+                                        241,
+                                        448,
+                                        258
+                                    ],
+                                    "score": 1.0,
+                                    "content": "checking the correctness of privacy guarantees easy, and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        449,
+                                        244,
+                                        467,
+                                        255
+                                    ],
+                                    "score": 0.74,
+                                    "content": "( i i i )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        467,
+                                        241,
+                                        506,
+                                        258
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tracking",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                254,
+                                505,
+                                267
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        254,
+                                        505,
+                                        267
+                                    ],
+                                    "score": 1.0,
+                                    "content": "privacy accountant simple. In practice, one should implement the centralized version (Algorithm 7 in",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                265,
+                                311,
+                                278
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        265,
+                                        311,
+                                        278
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Appendix D), which is significantly more efficient.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        }
+                    ],
+                    "index": 11,
+                    "bbox_fs": [
+                        104,
+                        177,
+                        506,
+                        278
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        106,
+                        296,
+                        345,
+                        307
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                295,
+                                345,
+                                308
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        295,
+                                        345,
+                                        308
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Algorithm 1: Private iterative filtering (interactive version)",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        }
+                    ],
+                    "index": 16
+                },
+                {
+                    "type": "index",
+                    "bbox": [
+                        98,
+                        309,
+                        475,
+                        442
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                308,
+                                474,
+                                324
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        308,
+                                        139,
+                                        324
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Input:",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        140,
+                                        310,
+                                        197,
+                                        323
+                                    ],
+                                    "score": 0.87,
+                                    "content": "S = \\{ x _ { i } \\} _ { i \\in [ n ] }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        198,
+                                        308,
+                                        201,
+                                        324
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        201,
+                                        310,
+                                        253,
+                                        322
+                                    ],
+                                    "score": 0.87,
+                                    "content": "\\alpha \\in ( 0 , 1 / 2 )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        253,
+                                        308,
+                                        303,
+                                        324
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", probability",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        303,
+                                        310,
+                                        343,
+                                        322
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\zeta \\in ( 0 , 1 )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        344,
+                                        308,
+                                        405,
+                                        324
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", # of iterations",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        406,
+                                        310,
+                                        474,
+                                        322
+                                    ],
+                                    "score": 0.66,
+                                    "content": "T = \\Theta ( d ) , ( \\varepsilon , \\delta )",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 17,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                97,
+                                321,
+                                240,
+                                335
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        97,
+                                        321,
+                                        107,
+                                        335
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        107,
+                                        323,
+                                        240,
+                                        335
+                                    ],
+                                    "score": 0.88,
+                                    "content": "( \\bar { x } , B ) \\gets q _ { \\mathrm { r a n g e } } ( S , 0 . 0 1 \\varepsilon , 0 . 0 1 \\delta )",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 18,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                96,
+                                334,
+                                356,
+                                349
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        96,
+                                        334,
+                                        106,
+                                        349
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        335,
+                                        273,
+                                        348
+                                    ],
+                                    "score": 0.62,
+                                    "content": "\\varepsilon _ { 1 } \\gets \\operatorname* { m i n } \\{ 0 . 9 9 \\varepsilon , 0 . 9 \\} / ( 4 \\sqrt { 2 T \\log ( 2 / \\delta ) } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        273,
+                                        334,
+                                        356,
+                                        349
+                                    ],
+                                    "score": 1.0,
+                                    "content": "), δ1 ← 0.99δ/(8T )",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                96,
+                                345,
+                                282,
+                                360
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        96,
+                                        345,
+                                        114,
+                                        360
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3 if",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        115,
+                                        348,
+                                        215,
+                                        359
+                                    ],
+                                    "score": 0.83,
+                                    "content": "n < ( 4 / \\varepsilon _ { 1 } ) \\log ( 1 / ( 2 \\delta _ { 1 } ) )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        215,
+                                        345,
+                                        276,
+                                        360
+                                    ],
+                                    "score": 1.0,
+                                    "content": "then Output:",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        276,
+                                        347,
+                                        282,
+                                        357
+                                    ],
+                                    "score": 0.32,
+                                    "content": "\\varnothing",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 20,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                97,
+                                358,
+                                189,
+                                369
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        97,
+                                        358,
+                                        121,
+                                        369
+                                    ],
+                                    "score": 1.0,
+                                    "content": "4 for",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        122,
+                                        359,
+                                        174,
+                                        369
+                                    ],
+                                    "score": 0.85,
+                                    "content": "t = 1 , \\dots , T",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        174,
+                                        358,
+                                        189,
+                                        369
+                                    ],
+                                    "score": 1.0,
+                                    "content": "do",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                98,
+                                368,
+                                416,
+                                381
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        98,
+                                        370,
+                                        105,
+                                        379
+                                    ],
+                                    "score": 1.0,
+                                    "content": "5",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        122,
+                                        369,
+                                        292,
+                                        381
+                                    ],
+                                    "score": 0.89,
+                                    "content": "n _ { t }  q _ { \\mathrm { s i z e } } ( \\{ ( \\mu _ { \\ell } , v _ { \\ell } , Z _ { \\ell } ) \\} _ { \\ell \\in [ t - 1 ] } , \\varepsilon _ { 1 } , \\bar { x } , B )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        293,
+                                        368,
+                                        304,
+                                        381
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", if",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        304,
+                                        369,
+                                        349,
+                                        380
+                                    ],
+                                    "score": 0.89,
+                                    "content": "n _ { t } < 3 n / 4",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        350,
+                                        368,
+                                        409,
+                                        381
+                                    ],
+                                    "score": 1.0,
+                                    "content": "then Output:",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        409,
+                                        369,
+                                        416,
+                                        379
+                                    ],
+                                    "score": 0.69,
+                                    "content": "\\varnothing",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 22,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                98,
+                                382,
+                                298,
+                                393
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        98,
+                                        383,
+                                        105,
+                                        392
+                                    ],
+                                    "score": 1.0,
+                                    "content": "6",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        121,
+                                        382,
+                                        298,
+                                        393
+                                    ],
+                                    "score": 0.85,
+                                    "content": "\\mu _ { t }  q _ { \\mathrm { m e a n } } ( \\{ ( \\mu _ { \\ell } , v _ { \\ell } , Z _ { \\ell } ) \\} _ { \\ell \\in [ t - 1 ] } , \\varepsilon _ { 1 } , \\bar { x } , B )",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 23,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                98,
+                                394,
+                                312,
+                                405
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        98,
+                                        395,
+                                        105,
+                                        404
+                                    ],
+                                    "score": 1.0,
+                                    "content": "7",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        122,
+                                        394,
+                                        312,
+                                        405
+                                    ],
+                                    "score": 0.84,
+                                    "content": "\\lambda _ { t } \\gets q _ { \\mathrm { n o r m } } ( \\{ ( \\mu _ { \\ell } , v _ { \\ell } , Z _ { \\ell } ) \\} _ { \\ell \\in [ t - 1 ] } , \\mu _ { t } , \\varepsilon _ { 1 } , \\bar { x } , B )",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 24,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                98,
+                                403,
+                                309,
+                                419
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        98,
+                                        407,
+                                        105,
+                                        416
+                                    ],
+                                    "score": 1.0,
+                                    "content": "8",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        120,
+                                        403,
+                                        130,
+                                        419
+                                    ],
+                                    "score": 1.0,
+                                    "content": "if",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        131,
+                                        406,
+                                        239,
+                                        417
+                                    ],
+                                    "score": 0.84,
+                                    "content": "\\lambda _ { t } \\le ( C - 0 . 0 1 ) \\alpha \\log { 1 / \\alpha }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        239,
+                                        403,
+                                        298,
+                                        419
+                                    ],
+                                    "score": 1.0,
+                                    "content": "then Output:",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        299,
+                                        407,
+                                        309,
+                                        417
+                                    ],
+                                    "score": 0.36,
+                                    "content": "\\mu _ { t }",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 25,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                97,
+                                417,
+                                328,
+                                429
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        97,
+                                        418,
+                                        105,
+                                        426
+                                    ],
+                                    "score": 1.0,
+                                    "content": "9",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        122,
+                                        417,
+                                        328,
+                                        429
+                                    ],
+                                    "score": 0.71,
+                                    "content": "\\boldsymbol { v } _ { t }  q _ { \\mathrm { P C A } } ( \\{ ( \\mu _ { \\ell } , v _ { \\ell } , Z _ { \\ell } ) \\} _ { \\ell \\in [ t - 1 ] } , \\mu _ { t } , \\varepsilon _ { 1 } , \\delta _ { 1 } , \\bar { x } , B ) )",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 26,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                96,
+                                429,
+                                195,
+                                441
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        96,
+                                        430,
+                                        105,
+                                        439
+                                    ],
+                                    "score": 1.0,
+                                    "content": "10",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        122,
+                                        429,
+                                        195,
+                                        441
+                                    ],
+                                    "score": 0.83,
+                                    "content": "Z _ { t } \\gets \\mathrm { U n i f } ( [ 0 , 1 ] )",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 27,
+                            "is_list_start_line": true
+                        }
+                    ],
+                    "index": 22,
+                    "bbox_fs": [
+                        96,
+                        308,
+                        474,
+                        441
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        443,
+                        155,
+                        454
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                440,
+                                154,
+                                457
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        440,
+                                        143,
+                                        457
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Output:",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        144,
+                                        444,
+                                        154,
+                                        454
+                                    ],
+                                    "score": 0.64,
+                                    "content": "\\mu _ { t }",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 28
+                        }
+                    ],
+                    "index": 28
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        474,
+                        505,
+                        580
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                474,
+                                506,
+                                488
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        474,
+                                        506,
+                                        488
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We give a high-level explanation of each step of Algorithm 1 here and give the formal definitions of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                484,
+                                504,
+                                500
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        484,
+                                        247,
+                                        500
+                                    ],
+                                    "score": 1.0,
+                                    "content": "all the queries in Appendix D. First,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        248,
+                                        487,
+                                        273,
+                                        498
+                                    ],
+                                    "score": 0.84,
+                                    "content": "q _ { \\mathrm { r a n g e } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        273,
+                                        484,
+                                        428,
+                                        500
+                                    ],
+                                    "score": 1.0,
+                                    "content": "returns (the parameters of) a hypercube",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        429,
+                                        485,
+                                        504,
+                                        498
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\bar { x } + [ - B / 2 , B / 2 ] ^ { d }",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                496,
+                                506,
+                                509
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        496,
+                                        506,
+                                        509
+                                    ],
+                                    "score": 1.0,
+                                    "content": "that is guaranteed to include all uncorrupted samples while preserving privacy. This is achieved",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                507,
+                                505,
+                                521
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        507,
+                                        154,
+                                        521
+                                    ],
+                                    "score": 1.0,
+                                    "content": "by running",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        154,
+                                        508,
+                                        161,
+                                        518
+                                    ],
+                                    "score": 0.72,
+                                    "content": "d",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        162,
+                                        507,
+                                        365,
+                                        521
+                                    ],
+                                    "score": 1.0,
+                                    "content": "coordinate-wise private histograms and selecting",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        365,
+                                        509,
+                                        376,
+                                        520
+                                    ],
+                                    "score": 0.87,
+                                    "content": "\\bar { x } _ { j }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        377,
+                                        507,
+                                        505,
+                                        521
+                                    ],
+                                    "score": 1.0,
+                                    "content": "as the center of the largest bin",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                519,
+                                506,
+                                534
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        519,
+                                        136,
+                                        534
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for the",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        137,
+                                        521,
+                                        143,
+                                        532
+                                    ],
+                                    "score": 0.82,
+                                    "content": "j",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        143,
+                                        519,
+                                        298,
+                                        534
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-th coordinate. Since covariance is I,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        299,
+                                        522,
+                                        324,
+                                        532
+                                    ],
+                                    "score": 0.88,
+                                    "content": "q _ { \\mathrm { r a n g e } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        325,
+                                        519,
+                                        388,
+                                        534
+                                    ],
+                                    "score": 1.0,
+                                    "content": "returns a fixed",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        388,
+                                        519,
+                                        476,
+                                        533
+                                    ],
+                                    "score": 0.94,
+                                    "content": "B = 8 \\sigma \\sqrt { \\log ( d n / \\zeta ) }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        476,
+                                        519,
+                                        506,
+                                        534
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Such",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                531,
+                                506,
+                                546
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        531,
+                                        506,
+                                        546
+                                    ],
+                                    "score": 1.0,
+                                    "content": "an adaptive estimate of the support is critical in tightly bounding the sensitivity of all subsequent",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                541,
+                                505,
+                                558
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        541,
+                                        414,
+                                        558
+                                    ],
+                                    "score": 1.0,
+                                    "content": "queries, which operate on the clipped dataset; all data points are projected as",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        415,
+                                        543,
+                                        505,
+                                        556
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\mathcal { P } _ { \\bar { x } + [ - B / 2 , B / 2 ] ^ { d } } ( x ) =",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                554,
+                                506,
+                                568
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        554,
+                                        122,
+                                        568
+                                    ],
+                                    "score": 1.0,
+                                    "content": "arg",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        123,
+                                        555,
+                                        204,
+                                        568
+                                    ],
+                                    "score": 0.66,
+                                    "content": "\\begin{array} { r } { \\operatorname* { m i n } _ { y \\in \\bar { x } + [ - B / 2 , B / 2 ] ^ { c } } } \\end{array}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        208,
+                                        555,
+                                        246,
+                                        567
+                                    ],
+                                    "score": 0.84,
+                                    "content": "\\lVert y - x \\rVert _ { 2 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        246,
+                                        554,
+                                        506,
+                                        568
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in all the queries that follow. With clipping, a single data point",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                568,
+                                244,
+                                581
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        568,
+                                        217,
+                                        581
+                                    ],
+                                    "score": 1.0,
+                                    "content": "can now change at most by",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        217,
+                                        568,
+                                        240,
+                                        580
+                                    ],
+                                    "score": 0.9,
+                                    "content": "B \\sqrt { d }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        240,
+                                        568,
+                                        244,
+                                        581
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        }
+                    ],
+                    "index": 33,
+                    "bbox_fs": [
+                        104,
+                        474,
+                        506,
+                        581
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        585,
+                        505,
+                        619
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                583,
+                                504,
+                                599
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        583,
+                                        371,
+                                        599
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The subsequent steps perform the non-private filtering algorithm of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        372,
+                                        586,
+                                        389,
+                                        596
+                                    ],
+                                    "score": 0.88,
+                                    "content": "\\ S 2 . 1",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        390,
+                                        583,
+                                        493,
+                                        599
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", but with private statistics",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        493,
+                                        587,
+                                        504,
+                                        596
+                                    ],
+                                    "score": 0.8,
+                                    "content": "\\mu _ { t }",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                595,
+                                506,
+                                609
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        595,
+                                        123,
+                                        609
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        123,
+                                        597,
+                                        132,
+                                        607
+                                    ],
+                                    "score": 0.84,
+                                    "content": "v _ { t }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        133,
+                                        595,
+                                        178,
+                                        609
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". As the set",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        178,
+                                        597,
+                                        189,
+                                        607
+                                    ],
+                                    "score": 0.88,
+                                    "content": "S _ { t }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        189,
+                                        595,
+                                        456,
+                                        609
+                                    ],
+                                    "score": 1.0,
+                                    "content": "changes over time, we lower bound its size (which we choose to be",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        456,
+                                        596,
+                                        503,
+                                        608
+                                    ],
+                                    "score": 0.9,
+                                    "content": "| S _ { t } | > n / 2 )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        503,
+                                        595,
+                                        506,
+                                        609
+                                    ],
+                                    "score": 1.0,
+                                    "content": ")",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                606,
+                                390,
+                                621
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        606,
+                                        292,
+                                        621
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to upper bound the sensitivity of other queries",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        292,
+                                        609,
+                                        344,
+                                        619
+                                    ],
+                                    "score": 0.87,
+                                    "content": "q _ { \\mathrm { m e a n } } , q _ { \\mathrm { n o r m } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        345,
+                                        606,
+                                        362,
+                                        621
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        362,
+                                        609,
+                                        385,
+                                        619
+                                    ],
+                                    "score": 0.81,
+                                    "content": "q _ { \\mathrm { P C A } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        385,
+                                        606,
+                                        390,
+                                        621
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        }
+                    ],
+                    "index": 39,
+                    "bbox_fs": [
+                        105,
+                        583,
+                        506,
+                        621
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        623,
+                        505,
+                        722
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                623,
+                                505,
+                                636
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        623,
+                                        132,
+                                        636
+                                    ],
+                                    "score": 1.0,
+                                    "content": "At the",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        133,
+                                        624,
+                                        138,
+                                        633
+                                    ],
+                                    "score": 0.76,
+                                    "content": "t",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        138,
+                                        623,
+                                        365,
+                                        636
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-th iterations, every time a query is called the data curator",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        365,
+                                        623,
+                                        377,
+                                        635
+                                    ],
+                                    "score": 0.78,
+                                    "content": "( i )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        377,
+                                        623,
+                                        398,
+                                        636
+                                    ],
+                                    "score": 1.0,
+                                    "content": "uses",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        398,
+                                        624,
+                                        424,
+                                        635
+                                    ],
+                                    "score": 0.92,
+                                    "content": "( { \\bar { x } } , B )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        425,
+                                        623,
+                                        489,
+                                        636
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to clip the data,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        489,
+                                        623,
+                                        505,
+                                        636
+                                    ],
+                                    "score": 0.81,
+                                    "content": "( i i )",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                634,
+                                506,
+                                648
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        634,
+                                        148,
+                                        648
+                                    ],
+                                    "score": 1.0,
+                                    "content": "computes",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        148,
+                                        635,
+                                        159,
+                                        645
+                                    ],
+                                    "score": 0.88,
+                                    "content": "S _ { t }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        159,
+                                        634,
+                                        207,
+                                        648
+                                    ],
+                                    "score": 1.0,
+                                    "content": "by running",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        207,
+                                        635,
+                                        230,
+                                        645
+                                    ],
+                                    "score": 0.86,
+                                    "content": "t - 1",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        230,
+                                        634,
+                                        417,
+                                        648
+                                    ],
+                                    "score": 1.0,
+                                    "content": "steps of the non-private filtering algorithm of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        417,
+                                        635,
+                                        435,
+                                        645
+                                    ],
+                                    "score": 0.86,
+                                    "content": "\\ S 2 . 1",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        436,
+                                        634,
+                                        506,
+                                        648
+                                    ],
+                                    "score": 1.0,
+                                    "content": "but with a given",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                644,
+                                506,
+                                660
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        644,
+                                        196,
+                                        660
+                                    ],
+                                    "score": 1.0,
+                                    "content": "fixed set of parameters",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        197,
+                                        645,
+                                        264,
+                                        658
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\{ ( \\mu _ { \\ell } , v _ { \\ell } ) \\} _ { \\ell \\in [ t - 1 ] }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        265,
+                                        644,
+                                        373,
+                                        660
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(and the given randomness",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        373,
+                                        645,
+                                        424,
+                                        658
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\{ Z _ { \\ell } \\} _ { \\ell \\in [ t - 1 ] } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        425,
+                                        644,
+                                        445,
+                                        660
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        446,
+                                        645,
+                                        464,
+                                        657
+                                    ],
+                                    "score": 0.88,
+                                    "content": "( i i i )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        464,
+                                        644,
+                                        506,
+                                        660
+                                    ],
+                                    "score": 1.0,
+                                    "content": "computes",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                656,
+                                506,
+                                669
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        656,
+                                        234,
+                                        669
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the queried private statistics of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        234,
+                                        658,
+                                        245,
+                                        667
+                                    ],
+                                    "score": 0.87,
+                                    "content": "S _ { t }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        245,
+                                        656,
+                                        447,
+                                        669
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". If the private spectral norm of the covariance of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        447,
+                                        657,
+                                        458,
+                                        667
+                                    ],
+                                    "score": 0.87,
+                                    "content": "S _ { t }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        458,
+                                        656,
+                                        480,
+                                        669
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(i.e.,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        480,
+                                        657,
+                                        492,
+                                        667
+                                    ],
+                                    "score": 0.81,
+                                    "content": "\\lambda _ { t } ,",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        492,
+                                        656,
+                                        506,
+                                        669
+                                    ],
+                                    "score": 1.0,
+                                    "content": ") is",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                667,
+                                505,
+                                680
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        667,
+                                        330,
+                                        680
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sufficiently small, we output the private and robust mean",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        331,
+                                        668,
+                                        360,
+                                        679
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\hat { \\mu } = \\mu _ { t }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        361,
+                                        667,
+                                        505,
+                                        680
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(line 8). Otherwise, we compute the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                678,
+                                505,
+                                691
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        678,
+                                        209,
+                                        691
+                                    ],
+                                    "score": 1.0,
+                                    "content": "private top PCA direction",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        210,
+                                        680,
+                                        219,
+                                        689
+                                    ],
+                                    "score": 0.83,
+                                    "content": "v _ { t }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        219,
+                                        678,
+                                        320,
+                                        691
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and draw an randomness",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        321,
+                                        679,
+                                        331,
+                                        689
+                                    ],
+                                    "score": 0.89,
+                                    "content": "Z _ { t }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        332,
+                                        678,
+                                        505,
+                                        691
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to be used in the next step of filtering, as in",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                689,
+                                505,
+                                701
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        689,
+                                        321,
+                                        701
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the non-private filtering algorithm. We emphasize that",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        322,
+                                        689,
+                                        342,
+                                        701
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\{ S _ { \\ell } \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        342,
+                                        689,
+                                        505,
+                                        701
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are not private, and hence never returned",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 47
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                699,
+                                505,
+                                714
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        699,
+                                        505,
+                                        714
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to the analyst. We also note that this interactive version is redundant as every query is re-computing",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 48
+                        },
+                        {
+                            "bbox": [
+                                107,
+                                711,
+                                506,
+                                723
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        711,
+                                        117,
+                                        722
+                                    ],
+                                    "score": 0.86,
+                                    "content": "S _ { t }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        117,
+                                        711,
+                                        506,
+                                        723
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". In our setting, the analyst has the dataset and there is no need to separate them. This leads to a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 49
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                72,
+                                506,
+                                86
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        72,
+                                        506,
+                                        86
+                                    ],
+                                    "score": 1.0,
+                                    "content": "centralized version we provide in Algorithm 7 in the appendix, which avoids redundant computations",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                84,
+                                270,
+                                96
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        84,
+                                        270,
+                                        96
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and hence is significantly more efficient.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 1
+                        }
+                    ],
+                    "index": 45,
+                    "bbox_fs": [
+                        104,
+                        623,
+                        506,
+                        723
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        72,
+                        504,
+                        95
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                72,
+                                506,
+                                86
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        72,
+                                        506,
+                                        86
+                                    ],
+                                    "score": 1.0,
+                                    "content": "centralized version we provide in Algorithm 7 in the appendix, which avoids redundant computations",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                84,
+                                270,
+                                96
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        84,
+                                        270,
+                                        96
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and hence is significantly more efficient.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        }
+                    ],
+                    "index": 0.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        100,
+                        505,
+                        167
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                99,
+                                505,
+                                113
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        99,
+                                        381,
+                                        113
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The main challenge in this framework is the privacy analysis. Because",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        381,
+                                        100,
+                                        429,
+                                        113
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\{ S _ { \\ell } \\} _ { \\ell \\in [ t - 1 ] }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        429,
+                                        99,
+                                        505,
+                                        113
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is not private, each",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                112,
+                                505,
+                                125
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        112,
+                                        151,
+                                        125
+                                    ],
+                                    "score": 1.0,
+                                    "content": "query runs",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        151,
+                                        112,
+                                        174,
+                                        123
+                                    ],
+                                    "score": 0.85,
+                                    "content": "t - 1",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        174,
+                                        112,
+                                        493,
+                                        125
+                                    ],
+                                    "score": 1.0,
+                                    "content": "steps of filtering whose end-to-end sensitivity could blow-up. Algorithmically,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        493,
+                                        112,
+                                        505,
+                                        124
+                                    ],
+                                    "score": 0.43,
+                                    "content": "( i )",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                123,
+                                504,
+                                135
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        123,
+                                        504,
+                                        135
+                                    ],
+                                    "score": 1.0,
+                                    "content": "we start with a specific choice of a non-private iterative filtering algorithm (among several variations",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                133,
+                                505,
+                                147
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        133,
+                                        418,
+                                        147
+                                    ],
+                                    "score": 1.0,
+                                    "content": "that are equivalent in non-private setting but widely differ in its sensitivity), and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        419,
+                                        134,
+                                        433,
+                                        145
+                                    ],
+                                    "score": 0.82,
+                                    "content": "( i i )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        434,
+                                        133,
+                                        505,
+                                        147
+                                    ],
+                                    "score": 1.0,
+                                    "content": "make appropriate",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                143,
+                                506,
+                                159
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        143,
+                                        506,
+                                        159
+                                    ],
+                                    "score": 1.0,
+                                    "content": "changes in the private queries (Algorithm 1) to keep the sensitivity small. Analytically, the following",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                154,
+                                481,
+                                169
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        154,
+                                        481,
+                                        169
+                                    ],
+                                    "score": 1.0,
+                                    "content": "key technical lemma allows a sharp analysis of the end-to-end sensitivity of iterative filtering.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        }
+                    ],
+                    "index": 4.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        169,
+                        505,
+                        217
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                168,
+                                505,
+                                183
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        168,
+                                        180,
+                                        183
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Lemma 2.2. Let",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        180,
+                                        170,
+                                        205,
+                                        182
+                                    ],
+                                    "score": 0.92,
+                                    "content": "S _ { t } ( S )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        205,
+                                        168,
+                                        393,
+                                        183
+                                    ],
+                                    "score": 1.0,
+                                    "content": "denote the resulting subset of samples after",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        393,
+                                        171,
+                                        398,
+                                        180
+                                    ],
+                                    "score": 0.28,
+                                    "content": "t",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        399,
+                                        168,
+                                        505,
+                                        183
+                                    ],
+                                    "score": 1.0,
+                                    "content": "iterations of the filtering",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                180,
+                                506,
+                                195
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        180,
+                                        169,
+                                        195
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in the queries",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        169,
+                                        182,
+                                        187,
+                                        192
+                                    ],
+                                    "score": 0.63,
+                                    "content": "( q _ { \\mathrm { s i z e } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        188,
+                                        180,
+                                        192,
+                                        195
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        192,
+                                        183,
+                                        216,
+                                        193
+                                    ],
+                                    "score": 0.43,
+                                    "content": "q _ { \\mathrm { m e a n } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        217,
+                                        180,
+                                        221,
+                                        195
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        222,
+                                        183,
+                                        245,
+                                        193
+                                    ],
+                                    "score": 0.66,
+                                    "content": "q _ { \\mathrm { n o r m } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        246,
+                                        180,
+                                        268,
+                                        195
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        269,
+                                        182,
+                                        294,
+                                        192
+                                    ],
+                                    "score": 0.76,
+                                    "content": "q _ { \\mathrm { P C A } } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        294,
+                                        180,
+                                        398,
+                                        195
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are applied to a dataset",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        398,
+                                        182,
+                                        406,
+                                        191
+                                    ],
+                                    "score": 0.76,
+                                    "content": "s",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        407,
+                                        180,
+                                        506,
+                                        195
+                                    ],
+                                    "score": 1.0,
+                                    "content": "using fixed parameters",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                107,
+                                192,
+                                505,
+                                207
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        192,
+                                        177,
+                                        205
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\{ ( \\mu _ { \\ell } , v _ { \\ell } , Z _ { \\ell } ) \\} _ { \\ell = 1 } ^ { t }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        177,
+                                        192,
+                                        244,
+                                        207
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Then, we have",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        244,
+                                        193,
+                                        376,
+                                        205
+                                    ],
+                                    "score": 0.92,
+                                    "content": "d _ { \\triangle } ( S _ { t } ( S ) , S _ { t } ( S ^ { \\prime } ) ) \\leq d _ { \\triangle } ( S , S ^ { \\prime } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        376,
+                                        192,
+                                        407,
+                                        207
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", where",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        407,
+                                        192,
+                                        505,
+                                        205
+                                    ],
+                                    "score": 0.9,
+                                    "content": "d _ { \\triangle } ( S , S ^ { \\prime } ) \\triangleq \\operatorname* { m a x } \\{ | S \\ \\backslash",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                202,
+                                165,
+                                218
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        205,
+                                        160,
+                                        217
+                                    ],
+                                    "score": 0.9,
+                                    "content": "S ^ { \\prime } | , | S ^ { \\prime } \\backslash S | \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        161,
+                                        202,
+                                        165,
+                                        218
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        }
+                    ],
+                    "index": 9.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        223,
+                        505,
+                        291
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                223,
+                                506,
+                                236
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        223,
+                                        286,
+                                        236
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Recall that two datasets are neighboring, i.e.,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        287,
+                                        224,
+                                        317,
+                                        235
+                                    ],
+                                    "score": 0.9,
+                                    "content": "\\boldsymbol { S } \\sim \\boldsymbol { S } ^ { \\prime }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        318,
+                                        223,
+                                        333,
+                                        236
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", iff",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        333,
+                                        224,
+                                        394,
+                                        236
+                                    ],
+                                    "score": 0.93,
+                                    "content": "d _ { \\triangle } ( S , S ^ { \\prime } ) \\leq 1",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        394,
+                                        223,
+                                        506,
+                                        236
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". This lemma implies that if",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                234,
+                                506,
+                                248
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        234,
+                                        506,
+                                        248
+                                    ],
+                                    "score": 1.0,
+                                    "content": "two datasets are neighboring, then they are still neighboring after filtering with the same parameters,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                246,
+                                506,
+                                259
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        246,
+                                        506,
+                                        259
+                                    ],
+                                    "score": 1.0,
+                                    "content": "no matter how many times we filter them. Hence, this lemma allows us to use the standard output-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                256,
+                                506,
+                                270
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        256,
+                                        228,
+                                        270
+                                    ],
+                                    "score": 1.0,
+                                    "content": "perturbation mechanisms with",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        228,
+                                        257,
+                                        259,
+                                        268
+                                    ],
+                                    "score": 0.91,
+                                    "content": "( \\varepsilon _ { 1 } , \\delta _ { 1 } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        259,
+                                        256,
+                                        506,
+                                        270
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-DP. Advanced composition ensures that end-to-end guarantee",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                267,
+                                506,
+                                281
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        267,
+                                        117,
+                                        281
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        117,
+                                        268,
+                                        131,
+                                        278
+                                    ],
+                                    "score": 0.74,
+                                    "content": "4 T",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        131,
+                                        267,
+                                        192,
+                                        281
+                                    ],
+                                    "score": 1.0,
+                                    "content": "such queries is",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        193,
+                                        268,
+                                        249,
+                                        279
+                                    ],
+                                    "score": 0.82,
+                                    "content": "( 0 . 9 9 \\varepsilon , 0 . 9 9 \\delta )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        250,
+                                        267,
+                                        326,
+                                        281
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-DP. Together with",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        326,
+                                        268,
+                                        383,
+                                        280
+                                    ],
+                                    "score": 0.75,
+                                    "content": "( 0 . 0 1 \\varepsilon , 0 . 0 1 \\delta )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        383,
+                                        267,
+                                        459,
+                                        281
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-DP budget used in",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        460,
+                                        268,
+                                        484,
+                                        280
+                                    ],
+                                    "score": 0.87,
+                                    "content": "q _ { \\mathrm { r a n g e } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        485,
+                                        267,
+                                        506,
+                                        281
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", this",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                279,
+                                500,
+                                291
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        279,
+                                        500,
+                                        291
+                                    ],
+                                    "score": 1.0,
+                                    "content": "satisfied the target privacy. Analyzing the utility of this algorithm, we get the following guarantee.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        }
+                    ],
+                    "index": 14.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        293,
+                        505,
+                        333
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                291,
+                                505,
+                                306
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        291,
+                                        204,
+                                        306
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Theorem 5. Algorithm",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        205,
+                                        294,
+                                        210,
+                                        303
+                                    ],
+                                    "score": 0.42,
+                                    "content": "^ { l }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        211,
+                                        291,
+                                        222,
+                                        306
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        222,
+                                        293,
+                                        244,
+                                        305
+                                    ],
+                                    "score": 0.9,
+                                    "content": "( \\varepsilon , \\delta )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        245,
+                                        291,
+                                        343,
+                                        306
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-DP. Under Assumption",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        344,
+                                        294,
+                                        349,
+                                        303
+                                    ],
+                                    "score": 0.36,
+                                    "content": "^ { l }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        350,
+                                        291,
+                                        488,
+                                        306
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", there exists a universal constant",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        488,
+                                        294,
+                                        505,
+                                        304
+                                    ],
+                                    "score": 0.79,
+                                    "content": "c \\in",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                303,
+                                506,
+                                321
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        306,
+                                        136,
+                                        318
+                                    ],
+                                    "score": 0.62,
+                                    "content": "( 0 , 0 . 1 )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        136,
+                                        303,
+                                        184,
+                                        321
+                                    ],
+                                    "score": 1.0,
+                                    "content": "such that if",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        185,
+                                        307,
+                                        210,
+                                        318
+                                    ],
+                                    "score": 0.86,
+                                    "content": "\\alpha \\leq c",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        210,
+                                        303,
+                                        228,
+                                        321
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        229,
+                                        304,
+                                        396,
+                                        319
+                                    ],
+                                    "score": 0.93,
+                                    "content": "n = \\widetilde \\Omega \\left( ( d / \\alpha ^ { 2 } ) + d ^ { 2 } ( \\log ( 1 / \\delta ) ) ^ { 3 / 2 } / ( \\varepsilon \\alpha ) \\right)",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        396,
+                                        303,
+                                        506,
+                                        321
+                                    ],
+                                    "score": 1.0,
+                                    "content": "then Algorithm 1 achieves",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                107,
+                                319,
+                                314,
+                                333
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        319,
+                                        228,
+                                        333
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\| \\hat { \\mu } - \\mu \\| _ { 2 } \\leq O ( \\alpha \\sqrt { \\log ( 1 / \\alpha ) } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        228,
+                                        319,
+                                        314,
+                                        333
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with probability 0.9.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        }
+                    ],
+                    "index": 19
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        340,
+                        505,
+                        397
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                341,
+                                504,
+                                353
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        341,
+                                        164,
+                                        353
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The first term",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        164,
+                                        341,
+                                        201,
+                                        353
+                                    ],
+                                    "score": 0.93,
+                                    "content": "O ( d / \\alpha ^ { 2 } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        202,
+                                        341,
+                                        497,
+                                        353
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in the sample complexity is optimal (cf. Table 1), but there is a factor of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        497,
+                                        342,
+                                        504,
+                                        351
+                                    ],
+                                    "score": 0.76,
+                                    "content": "d",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                352,
+                                506,
+                                365
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        352,
+                                        371,
+                                        365
+                                    ],
+                                    "score": 1.0,
+                                    "content": "gap in the second term. This is due to the fact that we need to run",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        371,
+                                        352,
+                                        393,
+                                        364
+                                    ],
+                                    "score": 0.91,
+                                    "content": "O ( d )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        393,
+                                        352,
+                                        506,
+                                        365
+                                    ],
+                                    "score": 1.0,
+                                    "content": "iterations in the worst-case.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                360,
+                                506,
+                                377
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        360,
+                                        506,
+                                        377
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Such numerous accesses to the database result in large noise to be added at each iteration, requiring",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                374,
+                                505,
+                                386
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        374,
+                                        505,
+                                        386
+                                    ],
+                                    "score": 1.0,
+                                    "content": "large sample size to combat that extra noise. We introduce PRIME to reduce the number of iterations",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                384,
+                                358,
+                                398
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        384,
+                                        117,
+                                        398
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        117,
+                                        385,
+                                        165,
+                                        397
+                                    ],
+                                    "score": 0.93,
+                                    "content": "{ \\bar { O } } ( ( \\log { d } ) ^ { 2 } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        165,
+                                        384,
+                                        358,
+                                        398
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and significantly reduce the sample complexity.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        }
+                    ],
+                    "index": 23
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        106,
+                        408,
+                        342,
+                        421
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                407,
+                                343,
+                                422
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        407,
+                                        343,
+                                        422
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2.3 PRIME: novel robust and private mean estimator",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        }
+                    ],
+                    "index": 26
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        428,
+                        506,
+                        539
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                428,
+                                505,
+                                442
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        428,
+                                        387,
+                                        442
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Algorithm 1 (specifically Filter(·) in Algorithm 1) accesses the database",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        387,
+                                        429,
+                                        409,
+                                        441
+                                    ],
+                                    "score": 0.92,
+                                    "content": "O ( d )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        409,
+                                        428,
+                                        505,
+                                        442
+                                    ],
+                                    "score": 1.0,
+                                    "content": "times. This is necessary",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                439,
+                                506,
+                                453
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        439,
+                                        347,
+                                        453
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for two reasons. First, the filter checks only one direction",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        347,
+                                        442,
+                                        357,
+                                        451
+                                    ],
+                                    "score": 0.85,
+                                    "content": "v _ { t }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        357,
+                                        439,
+                                        506,
+                                        453
+                                    ],
+                                    "score": 1.0,
+                                    "content": "at each iteration. In the worst case,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                451,
+                                506,
+                                464
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        451,
+                                        279,
+                                        464
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the corrupted samples can be scattered in",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        279,
+                                        451,
+                                        300,
+                                        463
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\Omega ( d )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        300,
+                                        451,
+                                        506,
+                                        464
+                                    ],
+                                    "score": 1.0,
+                                    "content": "orthogonal directions such that the filter needs to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                462,
+                                505,
+                                474
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        462,
+                                        158,
+                                        474
+                                    ],
+                                    "score": 1.0,
+                                    "content": "be repeated",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        158,
+                                        462,
+                                        180,
+                                        474
+                                    ],
+                                    "score": 0.92,
+                                    "content": "O ( d )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        180,
+                                        462,
+                                        505,
+                                        474
+                                    ],
+                                    "score": 1.0,
+                                    "content": "times. Secondly, even if the corrupted samples are clustered together in one",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                471,
+                                506,
+                                487
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        471,
+                                        293,
+                                        487
+                                    ],
+                                    "score": 1.0,
+                                    "content": "direction, the filter still needs to be repeated",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        293,
+                                        473,
+                                        315,
+                                        484
+                                    ],
+                                    "score": 0.92,
+                                    "content": "O ( d )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        315,
+                                        471,
+                                        506,
+                                        487
+                                    ],
+                                    "score": 1.0,
+                                    "content": "times. This is because we had to use a large",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                483,
+                                505,
+                                497
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        483,
+                                        198,
+                                        497
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(random) threshold of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        198,
+                                        483,
+                                        261,
+                                        496
+                                    ],
+                                    "score": 0.93,
+                                    "content": "d B ^ { 2 } Z _ { t } = { \\cal { O } } ( d )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        261,
+                                        483,
+                                        505,
+                                        497
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to make the threshold data-independent so that we can keep",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                494,
+                                505,
+                                507
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        494,
+                                        505,
+                                        507
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the sensitivity of Filter(·) low, which results in slow progress. We propose filtering multiple directions",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                505,
+                                505,
+                                518
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        506,
+                                        246,
+                                        518
+                                    ],
+                                    "score": 1.0,
+                                    "content": "simultaneously using a new score",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        246,
+                                        505,
+                                        264,
+                                        518
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\{ \\tau _ { i } \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        265,
+                                        506,
+                                        505,
+                                        518
+                                    ],
+                                    "score": 1.0,
+                                    "content": "based on the matrix multiplicative weights. Central to this",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                516,
+                                506,
+                                529
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        516,
+                                        358,
+                                        529
+                                    ],
+                                    "score": 1.0,
+                                    "content": "approach is a novel adaptive filtering algorithm DPTHRESHOLD",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        358,
+                                        517,
+                                        368,
+                                        528
+                                    ],
+                                    "score": 0.31,
+                                    "content": "( \\cdot )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        369,
+                                        516,
+                                        506,
+                                        529
+                                    ],
+                                    "score": 1.0,
+                                    "content": "that guarantees sufficient decrease",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                528,
+                                247,
+                                539
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        528,
+                                        247,
+                                        539
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in the total score at every iteration.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        }
+                    ],
+                    "index": 31.5
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        106,
+                        549,
+                        335,
+                        562
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                547,
+                                336,
+                                565
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        547,
+                                        336,
+                                        565
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2.3.1 Matrix Multiplicative Weight (MMW) scoring",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        }
+                    ],
+                    "index": 37
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        568,
+                        506,
+                        628
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                568,
+                                506,
+                                581
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        568,
+                                        506,
+                                        581
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The MMW-based approach, pioneered in [36] for non-private robust mean estimation, filters out",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                579,
+                                505,
+                                592
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        579,
+                                        293,
+                                        592
+                                    ],
+                                    "score": 1.0,
+                                    "content": "multiple directions simultaneously. It runs over",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        293,
+                                        580,
+                                        329,
+                                        591
+                                    ],
+                                    "score": 0.92,
+                                    "content": "O ( \\log d )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        329,
+                                        579,
+                                        469,
+                                        592
+                                    ],
+                                    "score": 1.0,
+                                    "content": "epochs and every epoch consists of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        469,
+                                        579,
+                                        505,
+                                        592
+                                    ],
+                                    "score": 0.93,
+                                    "content": "O ( \\log d )",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                590,
+                                506,
+                                604
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        590,
+                                        210,
+                                        604
+                                    ],
+                                    "score": 1.0,
+                                    "content": "iterations. At every epoch",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        210,
+                                        593,
+                                        216,
+                                        600
+                                    ],
+                                    "score": 0.74,
+                                    "content": "s",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        216,
+                                        590,
+                                        268,
+                                        604
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and iteration",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        269,
+                                        592,
+                                        273,
+                                        601
+                                    ],
+                                    "score": 0.75,
+                                    "content": "t",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        273,
+                                        590,
+                                        404,
+                                        604
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", step 2 of the iterative filtering in",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        404,
+                                        591,
+                                        423,
+                                        602
+                                    ],
+                                    "score": 0.86,
+                                    "content": "\\ S 2 . 1",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        423,
+                                        590,
+                                        506,
+                                        604
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is replaced by a new",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                103,
+                                601,
+                                505,
+                                618
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        103,
+                                        601,
+                                        129,
+                                        618
+                                    ],
+                                    "score": 1.0,
+                                    "content": "score",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        130,
+                                        602,
+                                        332,
+                                        617
+                                    ],
+                                    "score": 0.89,
+                                    "content": "\\tau _ { i } = ( x _ { i } - \\mathrm { M e a n } ( S _ { t } ^ { ( s ) } ) ) ^ { T } U _ { t } ^ { ( s ) } ( x _ { i } - \\mathrm { M e a n } ( S _ { t } ^ { ( s ) } ) )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        332,
+                                        601,
+                                        360,
+                                        618
+                                    ],
+                                    "score": 1.0,
+                                    "content": "where",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        361,
+                                        602,
+                                        380,
+                                        616
+                                    ],
+                                    "score": 0.92,
+                                    "content": "U _ { t } ^ { ( s ) }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        380,
+                                        601,
+                                        505,
+                                        618
+                                    ],
+                                    "score": 1.0,
+                                    "content": "now accounts for all directions",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                615,
+                                481,
+                                629
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        615,
+                                        117,
+                                        629
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        117,
+                                        615,
+                                        130,
+                                        626
+                                    ],
+                                    "score": 0.89,
+                                    "content": "\\mathbb { R } ^ { d }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        130,
+                                        615,
+                                        481,
+                                        629
+                                    ],
+                                    "score": 1.0,
+                                    "content": "but appropriately weighted. Precisely, it is defined via the matrix multiplicative update:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        }
+                    ],
+                    "index": 40
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        191,
+                        632,
+                        399,
+                        671
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                191,
+                                632,
+                                399,
+                                671
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        191,
+                                        632,
+                                        399,
+                                        671
+                                    ],
+                                    "score": 0.94,
+                                    "content": "U _ { t } ^ { ( s ) } = \\frac { \\exp \\left( \\alpha ^ { ( s ) } \\sum _ { r \\in [ t ] } ( \\mathrm { C o v } ( S _ { r } ^ { ( s ) } ) - { \\bf I } ) \\right) } { \\mathrm { T r } \\big ( \\exp ( \\alpha ^ { ( s ) } \\sum _ { r \\in [ t ] } ( \\mathrm { C o v } ( S _ { r } ^ { ( s ) } ) - { \\bf I } ) ) \\big ) } ,",
+                                    "type": "interline_equation",
+                                    "image_path": "53bb8d5442b2b4afe0bd0508483ccca717b6e5178e5593b665b58a74b0e0e679.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 43.5,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                191,
+                                632,
+                                399,
+                                651.5
+                            ],
+                            "spans": [],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                191,
+                                651.5,
+                                399,
+                                671.0
+                            ],
+                            "spans": [],
+                            "index": 44
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        676,
+                        505,
+                        722
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                674,
+                                506,
+                                690
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        674,
+                                        181,
+                                        690
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for some choice of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        182,
+                                        676,
+                                        218,
+                                        687
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\alpha ^ { ( s ) } > 0",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        218,
+                                        674,
+                                        427,
+                                        690
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". If we set the number of iterations to one, a choice of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        428,
+                                        676,
+                                        469,
+                                        687
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\alpha ^ { ( s ) } = \\infty",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        469,
+                                        674,
+                                        506,
+                                        690
+                                    ],
+                                    "score": 1.0,
+                                    "content": "recovers",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                686,
+                                507,
+                                700
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        686,
+                                        353,
+                                        700
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the previous score that relied on the top singular vector from",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        353,
+                                        688,
+                                        371,
+                                        699
+                                    ],
+                                    "score": 0.75,
+                                    "content": "\\ S 2 . 1",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        372,
+                                        686,
+                                        437,
+                                        700
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and a choice of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        437,
+                                        687,
+                                        474,
+                                        698
+                                    ],
+                                    "score": 0.89,
+                                    "content": "\\alpha ^ { ( s ) } = 0",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        474,
+                                        686,
+                                        507,
+                                        700
+                                    ],
+                                    "score": 1.0,
+                                    "content": "gives a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                698,
+                                506,
+                                713
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        698,
+                                        208,
+                                        713
+                                    ],
+                                    "score": 1.0,
+                                    "content": "simple norm based score",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        209,
+                                        701,
+                                        254,
+                                        712
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\tau _ { i } = \\| x _ { i } \\| _ { 2 } ^ { 2 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        254,
+                                        698,
+                                        361,
+                                        713
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". An appropriate choice of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        361,
+                                        700,
+                                        378,
+                                        710
+                                    ],
+                                    "score": 0.9,
+                                    "content": "\\alpha ^ { ( s ) }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        379,
+                                        698,
+                                        506,
+                                        713
+                                    ],
+                                    "score": 1.0,
+                                    "content": "smoothly interpolates between",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 47
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                711,
+                                505,
+                                723
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        711,
+                                        264,
+                                        723
+                                    ],
+                                    "score": 1.0,
+                                    "content": "these two extremes, which ensures that",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        264,
+                                        712,
+                                        299,
+                                        723
+                                    ],
+                                    "score": 0.92,
+                                    "content": "O ( \\log d )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        300,
+                                        711,
+                                        505,
+                                        723
+                                    ],
+                                    "score": 1.0,
+                                    "content": "iterations are sufficient for the spectral norm of the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 48
+                        }
+                    ],
+                    "index": 46.5
+                }
+            ],
+            "page_idx": 5,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        302,
+                        741,
+                        309,
+                        750
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                302,
+                                741,
+                                310,
+                                752
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        302,
+                                        741,
+                                        310,
+                                        752
+                                    ],
+                                    "score": 1.0,
+                                    "content": "6",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        72,
+                        504,
+                        95
+                    ],
+                    "lines": [],
+                    "index": 0.5,
+                    "bbox_fs": [
+                        105,
+                        72,
+                        506,
+                        96
+                    ],
+                    "lines_deleted": true
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        100,
+                        505,
+                        167
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                99,
+                                505,
+                                113
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        99,
+                                        381,
+                                        113
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The main challenge in this framework is the privacy analysis. Because",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        381,
+                                        100,
+                                        429,
+                                        113
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\{ S _ { \\ell } \\} _ { \\ell \\in [ t - 1 ] }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        429,
+                                        99,
+                                        505,
+                                        113
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is not private, each",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                112,
+                                505,
+                                125
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        112,
+                                        151,
+                                        125
+                                    ],
+                                    "score": 1.0,
+                                    "content": "query runs",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        151,
+                                        112,
+                                        174,
+                                        123
+                                    ],
+                                    "score": 0.85,
+                                    "content": "t - 1",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        174,
+                                        112,
+                                        493,
+                                        125
+                                    ],
+                                    "score": 1.0,
+                                    "content": "steps of filtering whose end-to-end sensitivity could blow-up. Algorithmically,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        493,
+                                        112,
+                                        505,
+                                        124
+                                    ],
+                                    "score": 0.43,
+                                    "content": "( i )",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                123,
+                                504,
+                                135
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        123,
+                                        504,
+                                        135
+                                    ],
+                                    "score": 1.0,
+                                    "content": "we start with a specific choice of a non-private iterative filtering algorithm (among several variations",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                133,
+                                505,
+                                147
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        133,
+                                        418,
+                                        147
+                                    ],
+                                    "score": 1.0,
+                                    "content": "that are equivalent in non-private setting but widely differ in its sensitivity), and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        419,
+                                        134,
+                                        433,
+                                        145
+                                    ],
+                                    "score": 0.82,
+                                    "content": "( i i )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        434,
+                                        133,
+                                        505,
+                                        147
+                                    ],
+                                    "score": 1.0,
+                                    "content": "make appropriate",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                143,
+                                506,
+                                159
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        143,
+                                        506,
+                                        159
+                                    ],
+                                    "score": 1.0,
+                                    "content": "changes in the private queries (Algorithm 1) to keep the sensitivity small. Analytically, the following",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                154,
+                                481,
+                                169
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        154,
+                                        481,
+                                        169
+                                    ],
+                                    "score": 1.0,
+                                    "content": "key technical lemma allows a sharp analysis of the end-to-end sensitivity of iterative filtering.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        }
+                    ],
+                    "index": 4.5,
+                    "bbox_fs": [
+                        104,
+                        99,
+                        506,
+                        169
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        169,
+                        505,
+                        217
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                168,
+                                505,
+                                183
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        168,
+                                        180,
+                                        183
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Lemma 2.2. Let",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        180,
+                                        170,
+                                        205,
+                                        182
+                                    ],
+                                    "score": 0.92,
+                                    "content": "S _ { t } ( S )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        205,
+                                        168,
+                                        393,
+                                        183
+                                    ],
+                                    "score": 1.0,
+                                    "content": "denote the resulting subset of samples after",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        393,
+                                        171,
+                                        398,
+                                        180
+                                    ],
+                                    "score": 0.28,
+                                    "content": "t",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        399,
+                                        168,
+                                        505,
+                                        183
+                                    ],
+                                    "score": 1.0,
+                                    "content": "iterations of the filtering",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                180,
+                                506,
+                                195
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        180,
+                                        169,
+                                        195
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in the queries",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        169,
+                                        182,
+                                        187,
+                                        192
+                                    ],
+                                    "score": 0.63,
+                                    "content": "( q _ { \\mathrm { s i z e } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        188,
+                                        180,
+                                        192,
+                                        195
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        192,
+                                        183,
+                                        216,
+                                        193
+                                    ],
+                                    "score": 0.43,
+                                    "content": "q _ { \\mathrm { m e a n } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        217,
+                                        180,
+                                        221,
+                                        195
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        222,
+                                        183,
+                                        245,
+                                        193
+                                    ],
+                                    "score": 0.66,
+                                    "content": "q _ { \\mathrm { n o r m } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        246,
+                                        180,
+                                        268,
+                                        195
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        269,
+                                        182,
+                                        294,
+                                        192
+                                    ],
+                                    "score": 0.76,
+                                    "content": "q _ { \\mathrm { P C A } } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        294,
+                                        180,
+                                        398,
+                                        195
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are applied to a dataset",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        398,
+                                        182,
+                                        406,
+                                        191
+                                    ],
+                                    "score": 0.76,
+                                    "content": "s",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        407,
+                                        180,
+                                        506,
+                                        195
+                                    ],
+                                    "score": 1.0,
+                                    "content": "using fixed parameters",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                107,
+                                192,
+                                505,
+                                207
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        192,
+                                        177,
+                                        205
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\{ ( \\mu _ { \\ell } , v _ { \\ell } , Z _ { \\ell } ) \\} _ { \\ell = 1 } ^ { t }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        177,
+                                        192,
+                                        244,
+                                        207
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Then, we have",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        244,
+                                        193,
+                                        376,
+                                        205
+                                    ],
+                                    "score": 0.92,
+                                    "content": "d _ { \\triangle } ( S _ { t } ( S ) , S _ { t } ( S ^ { \\prime } ) ) \\leq d _ { \\triangle } ( S , S ^ { \\prime } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        376,
+                                        192,
+                                        407,
+                                        207
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", where",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        407,
+                                        192,
+                                        505,
+                                        205
+                                    ],
+                                    "score": 0.9,
+                                    "content": "d _ { \\triangle } ( S , S ^ { \\prime } ) \\triangleq \\operatorname* { m a x } \\{ | S \\ \\backslash",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                202,
+                                165,
+                                218
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        205,
+                                        160,
+                                        217
+                                    ],
+                                    "score": 0.9,
+                                    "content": "S ^ { \\prime } | , | S ^ { \\prime } \\backslash S | \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        161,
+                                        202,
+                                        165,
+                                        218
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        }
+                    ],
+                    "index": 9.5,
+                    "bbox_fs": [
+                        104,
+                        168,
+                        506,
+                        218
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        223,
+                        505,
+                        291
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                223,
+                                506,
+                                236
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        223,
+                                        286,
+                                        236
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Recall that two datasets are neighboring, i.e.,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        287,
+                                        224,
+                                        317,
+                                        235
+                                    ],
+                                    "score": 0.9,
+                                    "content": "\\boldsymbol { S } \\sim \\boldsymbol { S } ^ { \\prime }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        318,
+                                        223,
+                                        333,
+                                        236
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", iff",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        333,
+                                        224,
+                                        394,
+                                        236
+                                    ],
+                                    "score": 0.93,
+                                    "content": "d _ { \\triangle } ( S , S ^ { \\prime } ) \\leq 1",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        394,
+                                        223,
+                                        506,
+                                        236
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". This lemma implies that if",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                234,
+                                506,
+                                248
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        234,
+                                        506,
+                                        248
+                                    ],
+                                    "score": 1.0,
+                                    "content": "two datasets are neighboring, then they are still neighboring after filtering with the same parameters,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                246,
+                                506,
+                                259
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        246,
+                                        506,
+                                        259
+                                    ],
+                                    "score": 1.0,
+                                    "content": "no matter how many times we filter them. Hence, this lemma allows us to use the standard output-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                256,
+                                506,
+                                270
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        256,
+                                        228,
+                                        270
+                                    ],
+                                    "score": 1.0,
+                                    "content": "perturbation mechanisms with",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        228,
+                                        257,
+                                        259,
+                                        268
+                                    ],
+                                    "score": 0.91,
+                                    "content": "( \\varepsilon _ { 1 } , \\delta _ { 1 } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        259,
+                                        256,
+                                        506,
+                                        270
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-DP. Advanced composition ensures that end-to-end guarantee",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                267,
+                                506,
+                                281
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        267,
+                                        117,
+                                        281
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        117,
+                                        268,
+                                        131,
+                                        278
+                                    ],
+                                    "score": 0.74,
+                                    "content": "4 T",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        131,
+                                        267,
+                                        192,
+                                        281
+                                    ],
+                                    "score": 1.0,
+                                    "content": "such queries is",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        193,
+                                        268,
+                                        249,
+                                        279
+                                    ],
+                                    "score": 0.82,
+                                    "content": "( 0 . 9 9 \\varepsilon , 0 . 9 9 \\delta )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        250,
+                                        267,
+                                        326,
+                                        281
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-DP. Together with",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        326,
+                                        268,
+                                        383,
+                                        280
+                                    ],
+                                    "score": 0.75,
+                                    "content": "( 0 . 0 1 \\varepsilon , 0 . 0 1 \\delta )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        383,
+                                        267,
+                                        459,
+                                        281
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-DP budget used in",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        460,
+                                        268,
+                                        484,
+                                        280
+                                    ],
+                                    "score": 0.87,
+                                    "content": "q _ { \\mathrm { r a n g e } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        485,
+                                        267,
+                                        506,
+                                        281
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", this",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                279,
+                                500,
+                                291
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        279,
+                                        500,
+                                        291
+                                    ],
+                                    "score": 1.0,
+                                    "content": "satisfied the target privacy. Analyzing the utility of this algorithm, we get the following guarantee.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        }
+                    ],
+                    "index": 14.5,
+                    "bbox_fs": [
+                        104,
+                        223,
+                        506,
+                        291
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        293,
+                        505,
+                        333
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                291,
+                                505,
+                                306
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        291,
+                                        204,
+                                        306
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Theorem 5. Algorithm",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        205,
+                                        294,
+                                        210,
+                                        303
+                                    ],
+                                    "score": 0.42,
+                                    "content": "^ { l }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        211,
+                                        291,
+                                        222,
+                                        306
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        222,
+                                        293,
+                                        244,
+                                        305
+                                    ],
+                                    "score": 0.9,
+                                    "content": "( \\varepsilon , \\delta )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        245,
+                                        291,
+                                        343,
+                                        306
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-DP. Under Assumption",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        344,
+                                        294,
+                                        349,
+                                        303
+                                    ],
+                                    "score": 0.36,
+                                    "content": "^ { l }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        350,
+                                        291,
+                                        488,
+                                        306
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", there exists a universal constant",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        488,
+                                        294,
+                                        505,
+                                        304
+                                    ],
+                                    "score": 0.79,
+                                    "content": "c \\in",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                303,
+                                506,
+                                321
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        306,
+                                        136,
+                                        318
+                                    ],
+                                    "score": 0.62,
+                                    "content": "( 0 , 0 . 1 )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        136,
+                                        303,
+                                        184,
+                                        321
+                                    ],
+                                    "score": 1.0,
+                                    "content": "such that if",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        185,
+                                        307,
+                                        210,
+                                        318
+                                    ],
+                                    "score": 0.86,
+                                    "content": "\\alpha \\leq c",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        210,
+                                        303,
+                                        228,
+                                        321
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        229,
+                                        304,
+                                        396,
+                                        319
+                                    ],
+                                    "score": 0.93,
+                                    "content": "n = \\widetilde \\Omega \\left( ( d / \\alpha ^ { 2 } ) + d ^ { 2 } ( \\log ( 1 / \\delta ) ) ^ { 3 / 2 } / ( \\varepsilon \\alpha ) \\right)",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        396,
+                                        303,
+                                        506,
+                                        321
+                                    ],
+                                    "score": 1.0,
+                                    "content": "then Algorithm 1 achieves",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                107,
+                                319,
+                                314,
+                                333
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        319,
+                                        228,
+                                        333
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\| \\hat { \\mu } - \\mu \\| _ { 2 } \\leq O ( \\alpha \\sqrt { \\log ( 1 / \\alpha ) } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        228,
+                                        319,
+                                        314,
+                                        333
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with probability 0.9.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        }
+                    ],
+                    "index": 19,
+                    "bbox_fs": [
+                        106,
+                        291,
+                        506,
+                        333
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        340,
+                        505,
+                        397
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                341,
+                                504,
+                                353
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        341,
+                                        164,
+                                        353
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The first term",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        164,
+                                        341,
+                                        201,
+                                        353
+                                    ],
+                                    "score": 0.93,
+                                    "content": "O ( d / \\alpha ^ { 2 } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        202,
+                                        341,
+                                        497,
+                                        353
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in the sample complexity is optimal (cf. Table 1), but there is a factor of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        497,
+                                        342,
+                                        504,
+                                        351
+                                    ],
+                                    "score": 0.76,
+                                    "content": "d",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                352,
+                                506,
+                                365
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        352,
+                                        371,
+                                        365
+                                    ],
+                                    "score": 1.0,
+                                    "content": "gap in the second term. This is due to the fact that we need to run",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        371,
+                                        352,
+                                        393,
+                                        364
+                                    ],
+                                    "score": 0.91,
+                                    "content": "O ( d )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        393,
+                                        352,
+                                        506,
+                                        365
+                                    ],
+                                    "score": 1.0,
+                                    "content": "iterations in the worst-case.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                360,
+                                506,
+                                377
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        360,
+                                        506,
+                                        377
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Such numerous accesses to the database result in large noise to be added at each iteration, requiring",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                374,
+                                505,
+                                386
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        374,
+                                        505,
+                                        386
+                                    ],
+                                    "score": 1.0,
+                                    "content": "large sample size to combat that extra noise. We introduce PRIME to reduce the number of iterations",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                384,
+                                358,
+                                398
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        384,
+                                        117,
+                                        398
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        117,
+                                        385,
+                                        165,
+                                        397
+                                    ],
+                                    "score": 0.93,
+                                    "content": "{ \\bar { O } } ( ( \\log { d } ) ^ { 2 } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        165,
+                                        384,
+                                        358,
+                                        398
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and significantly reduce the sample complexity.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        }
+                    ],
+                    "index": 23,
+                    "bbox_fs": [
+                        104,
+                        341,
+                        506,
+                        398
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        106,
+                        408,
+                        342,
+                        421
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                407,
+                                343,
+                                422
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        407,
+                                        343,
+                                        422
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2.3 PRIME: novel robust and private mean estimator",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        }
+                    ],
+                    "index": 26
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        428,
+                        506,
+                        539
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                428,
+                                505,
+                                442
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        428,
+                                        387,
+                                        442
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Algorithm 1 (specifically Filter(·) in Algorithm 1) accesses the database",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        387,
+                                        429,
+                                        409,
+                                        441
+                                    ],
+                                    "score": 0.92,
+                                    "content": "O ( d )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        409,
+                                        428,
+                                        505,
+                                        442
+                                    ],
+                                    "score": 1.0,
+                                    "content": "times. This is necessary",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                439,
+                                506,
+                                453
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        439,
+                                        347,
+                                        453
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for two reasons. First, the filter checks only one direction",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        347,
+                                        442,
+                                        357,
+                                        451
+                                    ],
+                                    "score": 0.85,
+                                    "content": "v _ { t }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        357,
+                                        439,
+                                        506,
+                                        453
+                                    ],
+                                    "score": 1.0,
+                                    "content": "at each iteration. In the worst case,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                451,
+                                506,
+                                464
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        451,
+                                        279,
+                                        464
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the corrupted samples can be scattered in",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        279,
+                                        451,
+                                        300,
+                                        463
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\Omega ( d )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        300,
+                                        451,
+                                        506,
+                                        464
+                                    ],
+                                    "score": 1.0,
+                                    "content": "orthogonal directions such that the filter needs to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                462,
+                                505,
+                                474
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        462,
+                                        158,
+                                        474
+                                    ],
+                                    "score": 1.0,
+                                    "content": "be repeated",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        158,
+                                        462,
+                                        180,
+                                        474
+                                    ],
+                                    "score": 0.92,
+                                    "content": "O ( d )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        180,
+                                        462,
+                                        505,
+                                        474
+                                    ],
+                                    "score": 1.0,
+                                    "content": "times. Secondly, even if the corrupted samples are clustered together in one",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                471,
+                                506,
+                                487
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        471,
+                                        293,
+                                        487
+                                    ],
+                                    "score": 1.0,
+                                    "content": "direction, the filter still needs to be repeated",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        293,
+                                        473,
+                                        315,
+                                        484
+                                    ],
+                                    "score": 0.92,
+                                    "content": "O ( d )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        315,
+                                        471,
+                                        506,
+                                        487
+                                    ],
+                                    "score": 1.0,
+                                    "content": "times. This is because we had to use a large",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                483,
+                                505,
+                                497
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        483,
+                                        198,
+                                        497
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(random) threshold of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        198,
+                                        483,
+                                        261,
+                                        496
+                                    ],
+                                    "score": 0.93,
+                                    "content": "d B ^ { 2 } Z _ { t } = { \\cal { O } } ( d )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        261,
+                                        483,
+                                        505,
+                                        497
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to make the threshold data-independent so that we can keep",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                494,
+                                505,
+                                507
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        494,
+                                        505,
+                                        507
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the sensitivity of Filter(·) low, which results in slow progress. We propose filtering multiple directions",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                505,
+                                505,
+                                518
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        506,
+                                        246,
+                                        518
+                                    ],
+                                    "score": 1.0,
+                                    "content": "simultaneously using a new score",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        246,
+                                        505,
+                                        264,
+                                        518
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\{ \\tau _ { i } \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        265,
+                                        506,
+                                        505,
+                                        518
+                                    ],
+                                    "score": 1.0,
+                                    "content": "based on the matrix multiplicative weights. Central to this",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                516,
+                                506,
+                                529
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        516,
+                                        358,
+                                        529
+                                    ],
+                                    "score": 1.0,
+                                    "content": "approach is a novel adaptive filtering algorithm DPTHRESHOLD",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        358,
+                                        517,
+                                        368,
+                                        528
+                                    ],
+                                    "score": 0.31,
+                                    "content": "( \\cdot )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        369,
+                                        516,
+                                        506,
+                                        529
+                                    ],
+                                    "score": 1.0,
+                                    "content": "that guarantees sufficient decrease",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                528,
+                                247,
+                                539
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        528,
+                                        247,
+                                        539
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in the total score at every iteration.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        }
+                    ],
+                    "index": 31.5,
+                    "bbox_fs": [
+                        105,
+                        428,
+                        506,
+                        539
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        106,
+                        549,
+                        335,
+                        562
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                547,
+                                336,
+                                565
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        547,
+                                        336,
+                                        565
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2.3.1 Matrix Multiplicative Weight (MMW) scoring",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        }
+                    ],
+                    "index": 37
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        568,
+                        506,
+                        628
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                568,
+                                506,
+                                581
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        568,
+                                        506,
+                                        581
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The MMW-based approach, pioneered in [36] for non-private robust mean estimation, filters out",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                579,
+                                505,
+                                592
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        579,
+                                        293,
+                                        592
+                                    ],
+                                    "score": 1.0,
+                                    "content": "multiple directions simultaneously. It runs over",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        293,
+                                        580,
+                                        329,
+                                        591
+                                    ],
+                                    "score": 0.92,
+                                    "content": "O ( \\log d )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        329,
+                                        579,
+                                        469,
+                                        592
+                                    ],
+                                    "score": 1.0,
+                                    "content": "epochs and every epoch consists of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        469,
+                                        579,
+                                        505,
+                                        592
+                                    ],
+                                    "score": 0.93,
+                                    "content": "O ( \\log d )",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                590,
+                                506,
+                                604
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        590,
+                                        210,
+                                        604
+                                    ],
+                                    "score": 1.0,
+                                    "content": "iterations. At every epoch",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        210,
+                                        593,
+                                        216,
+                                        600
+                                    ],
+                                    "score": 0.74,
+                                    "content": "s",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        216,
+                                        590,
+                                        268,
+                                        604
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and iteration",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        269,
+                                        592,
+                                        273,
+                                        601
+                                    ],
+                                    "score": 0.75,
+                                    "content": "t",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        273,
+                                        590,
+                                        404,
+                                        604
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", step 2 of the iterative filtering in",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        404,
+                                        591,
+                                        423,
+                                        602
+                                    ],
+                                    "score": 0.86,
+                                    "content": "\\ S 2 . 1",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        423,
+                                        590,
+                                        506,
+                                        604
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is replaced by a new",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                103,
+                                601,
+                                505,
+                                618
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        103,
+                                        601,
+                                        129,
+                                        618
+                                    ],
+                                    "score": 1.0,
+                                    "content": "score",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        130,
+                                        602,
+                                        332,
+                                        617
+                                    ],
+                                    "score": 0.89,
+                                    "content": "\\tau _ { i } = ( x _ { i } - \\mathrm { M e a n } ( S _ { t } ^ { ( s ) } ) ) ^ { T } U _ { t } ^ { ( s ) } ( x _ { i } - \\mathrm { M e a n } ( S _ { t } ^ { ( s ) } ) )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        332,
+                                        601,
+                                        360,
+                                        618
+                                    ],
+                                    "score": 1.0,
+                                    "content": "where",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        361,
+                                        602,
+                                        380,
+                                        616
+                                    ],
+                                    "score": 0.92,
+                                    "content": "U _ { t } ^ { ( s ) }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        380,
+                                        601,
+                                        505,
+                                        618
+                                    ],
+                                    "score": 1.0,
+                                    "content": "now accounts for all directions",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                615,
+                                481,
+                                629
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        615,
+                                        117,
+                                        629
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        117,
+                                        615,
+                                        130,
+                                        626
+                                    ],
+                                    "score": 0.89,
+                                    "content": "\\mathbb { R } ^ { d }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        130,
+                                        615,
+                                        481,
+                                        629
+                                    ],
+                                    "score": 1.0,
+                                    "content": "but appropriately weighted. Precisely, it is defined via the matrix multiplicative update:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        }
+                    ],
+                    "index": 40,
+                    "bbox_fs": [
+                        103,
+                        568,
+                        506,
+                        629
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        191,
+                        632,
+                        399,
+                        671
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                191,
+                                632,
+                                399,
+                                671
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        191,
+                                        632,
+                                        399,
+                                        671
+                                    ],
+                                    "score": 0.94,
+                                    "content": "U _ { t } ^ { ( s ) } = \\frac { \\exp \\left( \\alpha ^ { ( s ) } \\sum _ { r \\in [ t ] } ( \\mathrm { C o v } ( S _ { r } ^ { ( s ) } ) - { \\bf I } ) \\right) } { \\mathrm { T r } \\big ( \\exp ( \\alpha ^ { ( s ) } \\sum _ { r \\in [ t ] } ( \\mathrm { C o v } ( S _ { r } ^ { ( s ) } ) - { \\bf I } ) ) \\big ) } ,",
+                                    "type": "interline_equation",
+                                    "image_path": "53bb8d5442b2b4afe0bd0508483ccca717b6e5178e5593b665b58a74b0e0e679.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 43.5,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                191,
+                                632,
+                                399,
+                                651.5
+                            ],
+                            "spans": [],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                191,
+                                651.5,
+                                399,
+                                671.0
+                            ],
+                            "spans": [],
+                            "index": 44
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        676,
+                        505,
+                        722
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                674,
+                                506,
+                                690
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        674,
+                                        181,
+                                        690
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for some choice of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        182,
+                                        676,
+                                        218,
+                                        687
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\alpha ^ { ( s ) } > 0",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        218,
+                                        674,
+                                        427,
+                                        690
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". If we set the number of iterations to one, a choice of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        428,
+                                        676,
+                                        469,
+                                        687
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\alpha ^ { ( s ) } = \\infty",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        469,
+                                        674,
+                                        506,
+                                        690
+                                    ],
+                                    "score": 1.0,
+                                    "content": "recovers",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                686,
+                                507,
+                                700
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        686,
+                                        353,
+                                        700
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the previous score that relied on the top singular vector from",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        353,
+                                        688,
+                                        371,
+                                        699
+                                    ],
+                                    "score": 0.75,
+                                    "content": "\\ S 2 . 1",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        372,
+                                        686,
+                                        437,
+                                        700
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and a choice of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        437,
+                                        687,
+                                        474,
+                                        698
+                                    ],
+                                    "score": 0.89,
+                                    "content": "\\alpha ^ { ( s ) } = 0",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        474,
+                                        686,
+                                        507,
+                                        700
+                                    ],
+                                    "score": 1.0,
+                                    "content": "gives a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                698,
+                                506,
+                                713
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        698,
+                                        208,
+                                        713
+                                    ],
+                                    "score": 1.0,
+                                    "content": "simple norm based score",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        209,
+                                        701,
+                                        254,
+                                        712
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\tau _ { i } = \\| x _ { i } \\| _ { 2 } ^ { 2 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        254,
+                                        698,
+                                        361,
+                                        713
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". An appropriate choice of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        361,
+                                        700,
+                                        378,
+                                        710
+                                    ],
+                                    "score": 0.9,
+                                    "content": "\\alpha ^ { ( s ) }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        379,
+                                        698,
+                                        506,
+                                        713
+                                    ],
+                                    "score": 1.0,
+                                    "content": "smoothly interpolates between",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 47
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                711,
+                                505,
+                                723
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        711,
+                                        264,
+                                        723
+                                    ],
+                                    "score": 1.0,
+                                    "content": "these two extremes, which ensures that",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        264,
+                                        712,
+                                        299,
+                                        723
+                                    ],
+                                    "score": 0.92,
+                                    "content": "O ( \\log d )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        300,
+                                        711,
+                                        505,
+                                        723
+                                    ],
+                                    "score": 1.0,
+                                    "content": "iterations are sufficient for the spectral norm of the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 48
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                73,
+                                505,
+                                85
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        73,
+                                        420,
+                                        85
+                                    ],
+                                    "score": 1.0,
+                                    "content": "covariance to decrease strictly by a constant factor. This guarantees that after",
+                                    "type": "text",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        420,
+                                        73,
+                                        456,
+                                        85
+                                    ],
+                                    "score": 0.92,
+                                    "content": "O ( \\log d )",
+                                    "type": "inline_equation",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        457,
+                                        73,
+                                        505,
+                                        85
+                                    ],
+                                    "score": 1.0,
+                                    "content": "epochs, we",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                83,
+                                505,
+                                96
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        83,
+                                        505,
+                                        96
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sufficiently decrease the covariance to ensure that the empirical mean is accurate enough. Critical",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                94,
+                                505,
+                                106
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        94,
+                                        505,
+                                        106
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in achieving this gain is our carefully designed filtering algorithm DPTHRESHOLD that uses the",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                105,
+                                506,
+                                118
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        105,
+                                        506,
+                                        118
+                                    ],
+                                    "score": 1.0,
+                                    "content": "privately computed MMW-based scores using Gaussian mechanism on the covariance matrices as",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                116,
+                                264,
+                                129
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        116,
+                                        264,
+                                        129
+                                    ],
+                                    "score": 1.0,
+                                    "content": "shown in Algorithm 11 in Appendix E.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 4
+                        }
+                    ],
+                    "index": 46.5,
+                    "bbox_fs": [
+                        105,
+                        674,
+                        507,
+                        723
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        72,
+                        505,
+                        128
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                73,
+                                505,
+                                85
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        73,
+                                        420,
+                                        85
+                                    ],
+                                    "score": 1.0,
+                                    "content": "covariance to decrease strictly by a constant factor. This guarantees that after",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        420,
+                                        73,
+                                        456,
+                                        85
+                                    ],
+                                    "score": 0.92,
+                                    "content": "O ( \\log d )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        457,
+                                        73,
+                                        505,
+                                        85
+                                    ],
+                                    "score": 1.0,
+                                    "content": "epochs, we",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                83,
+                                505,
+                                96
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        83,
+                                        505,
+                                        96
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sufficiently decrease the covariance to ensure that the empirical mean is accurate enough. Critical",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                94,
+                                505,
+                                106
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        94,
+                                        505,
+                                        106
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in achieving this gain is our carefully designed filtering algorithm DPTHRESHOLD that uses the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                105,
+                                506,
+                                118
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        105,
+                                        506,
+                                        118
+                                    ],
+                                    "score": 1.0,
+                                    "content": "privately computed MMW-based scores using Gaussian mechanism on the covariance matrices as",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                116,
+                                264,
+                                129
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        116,
+                                        264,
+                                        129
+                                    ],
+                                    "score": 1.0,
+                                    "content": "shown in Algorithm 11 in Appendix E.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        }
+                    ],
+                    "index": 2
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        106,
+                        139,
+                        303,
+                        151
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                138,
+                                304,
+                                153
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        138,
+                                        304,
+                                        153
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2.3.2 Adaptive filtering with DPTHRESHOLD",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        }
+                    ],
+                    "index": 5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        157,
+                        505,
+                        257
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                157,
+                                506,
+                                171
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        157,
+                                        506,
+                                        171
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Novelty. The corresponding non-private filtering of [36, Algorithm 9] for robust mean estimation",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                169,
+                                506,
+                                181
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        169,
+                                        506,
+                                        181
+                                    ],
+                                    "score": 1.0,
+                                    "content": "takes advantage of an adaptive threshold, but filters out each sample independently resulting in a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                179,
+                                506,
+                                192
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        179,
+                                        506,
+                                        192
+                                    ],
+                                    "score": 1.0,
+                                    "content": "prohibitively large sensitivity; the coupling between each sample and the randomness used to filter it",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                191,
+                                506,
+                                204
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        191,
+                                        506,
+                                        204
+                                    ],
+                                    "score": 1.0,
+                                    "content": "can change widely between two neighboring datasets. On the other hand, Algorithm 1 (i.e., Filter(·)",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                200,
+                                507,
+                                215
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        200,
+                                        448,
+                                        215
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in Algorithm 6) takes advantage of jointly filtering all points above a single threshold",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        448,
+                                        201,
+                                        477,
+                                        213
+                                    ],
+                                    "score": 0.92,
+                                    "content": "B ^ { 2 } d Z _ { t }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        477,
+                                        200,
+                                        507,
+                                        215
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                211,
+                                504,
+                                226
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        211,
+                                        183,
+                                        226
+                                    ],
+                                    "score": 1.0,
+                                    "content": "single randomness",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        183,
+                                        213,
+                                        247,
+                                        224
+                                    ],
+                                    "score": 0.91,
+                                    "content": "Z _ { t } \\sim \\mathrm { U n i f } [ 0 , 1 ]",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        247,
+                                        211,
+                                        484,
+                                        226
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", but the non-adaptive (and hence large) choice of the range",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        485,
+                                        212,
+                                        504,
+                                        223
+                                    ],
+                                    "score": 0.89,
+                                    "content": "B ^ { 2 } d",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                223,
+                                505,
+                                236
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        223,
+                                        505,
+                                        236
+                                    ],
+                                    "score": 1.0,
+                                    "content": "results in a large number of iterations because each filtering only decrease the score by little. To",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                235,
+                                506,
+                                246
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        235,
+                                        506,
+                                        246
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sufficiently reduce the total score while maintaining a small sensitivity, we introduce a filter with a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                245,
+                                227,
+                                258
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        245,
+                                        227,
+                                        258
+                                    ],
+                                    "score": 1.0,
+                                    "content": "single and adaptive threshold.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        }
+                    ],
+                    "index": 10
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        261,
+                        505,
+                        328
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                261,
+                                506,
+                                274
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        261,
+                                        347,
+                                        274
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Algorithm. Our goal here is to privately find a single scalar",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        348,
+                                        264,
+                                        354,
+                                        273
+                                    ],
+                                    "score": 0.83,
+                                    "content": "\\rho",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        355,
+                                        261,
+                                        506,
+                                        274
+                                    ],
+                                    "score": 1.0,
+                                    "content": "such that when a randomized filter is",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                272,
+                                506,
+                                285
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        272,
+                                        193,
+                                        285
+                                    ],
+                                    "score": 1.0,
+                                    "content": "applied on the scores",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        194,
+                                        272,
+                                        212,
+                                        285
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\{ \\tau _ { i } \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        213,
+                                        272,
+                                        320,
+                                        285
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with a (random) threshold",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        320,
+                                        273,
+                                        334,
+                                        284
+                                    ],
+                                    "score": 0.88,
+                                    "content": "\\rho Z",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        335,
+                                        272,
+                                        359,
+                                        285
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(with",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        359,
+                                        273,
+                                        368,
+                                        283
+                                    ],
+                                    "score": 0.73,
+                                    "content": "Z",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        368,
+                                        272,
+                                        442,
+                                        285
+                                    ],
+                                    "score": 1.0,
+                                    "content": "drawn uniform in",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        442,
+                                        273,
+                                        465,
+                                        285
+                                    ],
+                                    "score": 0.25,
+                                    "content": "[ 0 , 1 ] )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        465,
+                                        272,
+                                        506,
+                                        285
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", we filter",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                284,
+                                505,
+                                295
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        284,
+                                        505,
+                                        295
+                                    ],
+                                    "score": 1.0,
+                                    "content": "out enough samples to make progress in each iteration while ensuring that we do not remove too",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                295,
+                                506,
+                                307
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        295,
+                                        506,
+                                        307
+                                    ],
+                                    "score": 1.0,
+                                    "content": "many uncorrupted samples. This is a slight generalization of the non-private algorithm in Section 2.1,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                304,
+                                506,
+                                320
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        304,
+                                        176,
+                                        320
+                                    ],
+                                    "score": 1.0,
+                                    "content": "which simply set",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        176,
+                                        307,
+                                        242,
+                                        318
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\rho = \\operatorname* { m a x } _ { j \\in S _ { t } } \\tau _ { j }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        242,
+                                        304,
+                                        506,
+                                        320
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". While this guarantees the filter removes more corrupted samples",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                316,
+                                498,
+                                329
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        316,
+                                        498,
+                                        329
+                                    ],
+                                    "score": 1.0,
+                                    "content": "than good samples, it does not make sufficient progress in reducing the total score of the samples.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        }
+                    ],
+                    "index": 17.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        333,
+                        505,
+                        366
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                332,
+                                505,
+                                344
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        332,
+                                        505,
+                                        344
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Ideally, we want the thresholding to decrease the total score by a constant multiplicative factor, which",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                342,
+                                506,
+                                357
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        342,
+                                        506,
+                                        357
+                                    ],
+                                    "score": 1.0,
+                                    "content": "will in the end allow the algorithm to terminate within logarithmic iterations. To this end, we propose",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                354,
+                                414,
+                                367
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        354,
+                                        243,
+                                        367
+                                    ],
+                                    "score": 1.0,
+                                    "content": "a new scheme of using the largest",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        243,
+                                        356,
+                                        250,
+                                        366
+                                    ],
+                                    "score": 0.82,
+                                    "content": "\\rho",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        250,
+                                        354,
+                                        414,
+                                        367
+                                    ],
+                                    "score": 1.0,
+                                    "content": "such that the following inequality holds:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        }
+                    ],
+                    "index": 22
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        223,
+                        370,
+                        367,
+                        398
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                223,
+                                370,
+                                367,
+                                398
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        223,
+                                        370,
+                                        367,
+                                        398
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\sum _ { \\tau _ { i } > \\rho } ( \\tau _ { i } - \\rho ) \\geq 0 . 3 1 \\sum _ { \\tau _ { i } \\in S _ { t } } ( \\tau _ { i } - 1 ) .",
+                                    "type": "interline_equation",
+                                    "image_path": "6e717cdd2620c553eeae0f0bac341635344a945d930dd30b1b3e271d36a171e1.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 24.5,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                223,
+                                370,
+                                367,
+                                384.0
+                            ],
+                            "spans": [],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                223,
+                                384.0,
+                                367,
+                                398.0
+                            ],
+                            "spans": [],
+                            "index": 25
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        402,
+                        505,
+                        459
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                403,
+                                505,
+                                416
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        403,
+                                        505,
+                                        416
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We use a private histogram of the scores to approximate this threshold. Similar to [55, 58], we use",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                413,
+                                506,
+                                428
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        413,
+                                        327,
+                                        428
+                                    ],
+                                    "score": 1.0,
+                                    "content": "geometrically increasing bin sizes such that we use only",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        327,
+                                        414,
+                                        376,
+                                        426
+                                    ],
+                                    "score": 0.92,
+                                    "content": "O ( \\log B ^ { 2 } d )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        376,
+                                        413,
+                                        506,
+                                        428
+                                    ],
+                                    "score": 1.0,
+                                    "content": "bins while achieving a preferred",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                425,
+                                505,
+                                437
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        425,
+                                        333,
+                                        437
+                                    ],
+                                    "score": 1.0,
+                                    "content": "multiplicative error in our quantization. At each epoch",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        334,
+                                        427,
+                                        340,
+                                        435
+                                    ],
+                                    "score": 0.75,
+                                    "content": "s",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        340,
+                                        425,
+                                        395,
+                                        437
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and iteration",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        396,
+                                        426,
+                                        400,
+                                        435
+                                    ],
+                                    "score": 0.76,
+                                    "content": "t",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        401,
+                                        425,
+                                        505,
+                                        437
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", we run DPTHRESHOLD",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                435,
+                                505,
+                                449
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        435,
+                                        266,
+                                        449
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sketched in the following to approximate",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        267,
+                                        438,
+                                        273,
+                                        447
+                                    ],
+                                    "score": 0.82,
+                                    "content": "\\rho",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        273,
+                                        435,
+                                        505,
+                                        449
+                                    ],
+                                    "score": 1.0,
+                                    "content": "followed by a random filter. Step 3 replaces the non-private",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                446,
+                                402,
+                                460
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        446,
+                                        402,
+                                        460
+                                    ],
+                                    "score": 1.0,
+                                    "content": "condition in Eq. (1). A complete description is provided in Algorithm 11.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        }
+                    ],
+                    "index": 28
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        129,
+                        465,
+                        506,
+                        536
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                128,
+                                461,
+                                494,
+                                483
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        128,
+                                        461,
+                                        317,
+                                        483
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1. Privately compute scores for all data points",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        317,
+                                        465,
+                                        489,
+                                        480
+                                    ],
+                                    "score": 0.89,
+                                    "content": "i \\in S _ { t } ^ { ( s ) } : \\tau _ { i } \\gets ( x _ { i } - \\mu _ { t } ) ^ { \\top } U _ { t } ^ { ( s ) } ( x _ { i } - \\mu _ { t } ) ;",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        489,
+                                        461,
+                                        494,
+                                        483
+                                    ],
+                                    "score": 1.0,
+                                    "content": ";",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                123,
+                                471,
+                                510,
+                                503
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        123,
+                                        471,
+                                        264,
+                                        503
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2. Compute a private histogram",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        264,
+                                        478,
+                                        326,
+                                        496
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\{ \\tilde { h } _ { j } \\} _ { j = 1 } ^ { 2 + \\log ( B ^ { 2 } d ) }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        326,
+                                        471,
+                                        510,
+                                        503
+                                    ],
+                                    "score": 1.0,
+                                    "content": "t i  i  t  t i  t of the scores over geometrically sized bins",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                142,
+                                493,
+                                447,
+                                511
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        142,
+                                        496,
+                                        206,
+                                        509
+                                    ],
+                                    "score": 0.86,
+                                    "content": "I _ { 1 } = [ 1 / 4 , 1 / 2 )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        206,
+                                        493,
+                                        209,
+                                        511
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        209,
+                                        494,
+                                        441,
+                                        509
+                                    ],
+                                    "score": 0.84,
+                                    "content": "I _ { 2 } = [ 1 / 2 , 1 ) , \\ldots , \\bar { I } _ { 2 + \\log ( B ^ { 2 } d ) } = [ 2 ^ { \\log ( B ^ { 2 } d ) - 1 } , 2 ^ { \\log ( B ^ { 2 } d ) } ]",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        442,
+                                        493,
+                                        447,
+                                        511
+                                    ],
+                                    "score": 1.0,
+                                    "content": ";",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                504,
+                                472,
+                                527
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        504,
+                                        241,
+                                        527
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3. Privately find the largest",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        242,
+                                        510,
+                                        248,
+                                        519
+                                    ],
+                                    "score": 0.77,
+                                    "content": "\\ell",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        248,
+                                        504,
+                                        290,
+                                        527
+                                    ],
+                                    "score": 1.0,
+                                    "content": "satisfying",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        290,
+                                        509,
+                                        468,
+                                        525
+                                    ],
+                                    "score": 0.85,
+                                    "content": "\\begin{array} { r } { \\sum _ { j \\ge \\ell } ( 2 ^ { j } - 2 ^ { \\ell } ) \\tilde { h } _ { j } \\ge 0 . 3 1 \\sum _ { i \\in S _ { t } ^ { ( s ) } } ( \\tau _ { i } - 1 ) ; } \\end{array}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        468,
+                                        504,
+                                        472,
+                                        527
+                                    ],
+                                    "score": 1.0,
+                                    "content": ";",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                523,
+                                210,
+                                536
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        523,
+                                        172,
+                                        536
+                                    ],
+                                    "score": 1.0,
+                                    "content": "4. Output",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        173,
+                                        523,
+                                        201,
+                                        536
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\rho = 2 ^ { \\ell }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        201,
+                                        523,
+                                        210,
+                                        536
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        }
+                    ],
+                    "index": 33
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        549,
+                        227,
+                        563
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                549,
+                                228,
+                                566
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        549,
+                                        228,
+                                        566
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3 Analyses of PRIME",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        }
+                    ],
+                    "index": 36
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        574,
+                        506,
+                        608
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                574,
+                                506,
+                                586
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        574,
+                                        506,
+                                        586
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Building on the framework of Algorithm 1, PRIME (Algorithm 9) replaces the score with the MMW-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                586,
+                                505,
+                                597
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        586,
+                                        208,
+                                        597
+                                    ],
+                                    "score": 1.0,
+                                    "content": "based score presented in",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        209,
+                                        586,
+                                        235,
+                                        596
+                                    ],
+                                    "score": 0.82,
+                                    "content": "\\ S 2 . 3 . 1",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        235,
+                                        586,
+                                        505,
+                                        597
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and the filter with the adaptive DPTHRESHOLD. This reduces the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                596,
+                                397,
+                                609
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        596,
+                                        200,
+                                        609
+                                    ],
+                                    "score": 1.0,
+                                    "content": "number of iterations to",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        200,
+                                        596,
+                                        268,
+                                        609
+                                    ],
+                                    "score": 0.93,
+                                    "content": "T = O ( ( \\log d ) ^ { 2 } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        269,
+                                        596,
+                                        397,
+                                        609
+                                    ],
+                                    "score": 1.0,
+                                    "content": "achieving the following bound.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        }
+                    ],
+                    "index": 38
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        611,
+                        505,
+                        650
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                610,
+                                506,
+                                624
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        610,
+                                        204,
+                                        624
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Theorem 6. PRIME is",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        205,
+                                        611,
+                                        226,
+                                        623
+                                    ],
+                                    "score": 0.91,
+                                    "content": "( \\varepsilon , \\delta )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        227,
+                                        610,
+                                        399,
+                                        624
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-differentially private. Under Assumption",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        400,
+                                        612,
+                                        406,
+                                        621
+                                    ],
+                                    "score": 0.4,
+                                    "content": "I",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        406,
+                                        610,
+                                        506,
+                                        624
+                                    ],
+                                    "score": 1.0,
+                                    "content": "there exists a universal",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                622,
+                                506,
+                                638
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        623,
+                                        144,
+                                        638
+                                    ],
+                                    "score": 1.0,
+                                    "content": "constant",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        144,
+                                        624,
+                                        191,
+                                        636
+                                    ],
+                                    "score": 0.89,
+                                    "content": "c \\in ( 0 , 0 . 1 )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        191,
+                                        623,
+                                        240,
+                                        638
+                                    ],
+                                    "score": 1.0,
+                                    "content": "such that if",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        241,
+                                        625,
+                                        266,
+                                        636
+                                    ],
+                                    "score": 0.85,
+                                    "content": "\\alpha \\leq c",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        267,
+                                        623,
+                                        285,
+                                        638
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        286,
+                                        622,
+                                        447,
+                                        637
+                                    ],
+                                    "score": 0.92,
+                                    "content": "n = \\widetilde \\Omega ( ( d / \\alpha ^ { 2 } ) + ( d ^ { 3 / 2 } / ( \\varepsilon \\alpha ) ) \\log ( 1 / \\delta ) )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        447,
+                                        623,
+                                        506,
+                                        638
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", then PRIME",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                636,
+                                350,
+                                651
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        636,
+                                        144,
+                                        651
+                                    ],
+                                    "score": 1.0,
+                                    "content": "achieves",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        144,
+                                        636,
+                                        264,
+                                        650
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\| \\hat { \\mu } - \\mu \\| _ { 2 } = O ( \\alpha \\sqrt { \\log ( 1 / \\alpha ) } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        265,
+                                        636,
+                                        350,
+                                        651
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with probability 0.9.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        }
+                    ],
+                    "index": 41
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        658,
+                        505,
+                        723
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                658,
+                                505,
+                                673
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        659,
+                                        303,
+                                        673
+                                    ],
+                                    "score": 1.0,
+                                    "content": "A proof is provided in Appendix F. The notation",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        303,
+                                        658,
+                                        322,
+                                        672
+                                    ],
+                                    "score": 0.9,
+                                    "content": "\\widetilde { \\Omega } ( \\cdot )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        322,
+                                        659,
+                                        430,
+                                        673
+                                    ],
+                                    "score": 1.0,
+                                    "content": "hides logarithmic terms in",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        430,
+                                        660,
+                                        450,
+                                        671
+                                    ],
+                                    "score": 0.56,
+                                    "content": "d , R",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        450,
+                                        659,
+                                        470,
+                                        673
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        470,
+                                        660,
+                                        488,
+                                        672
+                                    ],
+                                    "score": 0.9,
+                                    "content": "1 / \\alpha",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        488,
+                                        659,
+                                        505,
+                                        673
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". To",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                672,
+                                507,
+                                687
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        672,
+                                        185,
+                                        687
+                                    ],
+                                    "score": 1.0,
+                                    "content": "achieve an error of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        186,
+                                        672,
+                                        254,
+                                        686
+                                    ],
+                                    "score": 0.92,
+                                    "content": "O ( \\alpha \\sqrt { \\log ( 1 / \\alpha ) } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        255,
+                                        672,
+                                        314,
+                                        687
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", the first term",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        315,
+                                        672,
+                                        389,
+                                        686
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\widetilde \\Omega ( d / \\alpha ^ { 2 } \\log ( 1 / \\alpha ) )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        389,
+                                        672,
+                                        507,
+                                        687
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is necessary even if there is",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                686,
+                                506,
+                                700
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        686,
+                                        237,
+                                        700
+                                    ],
+                                    "score": 1.0,
+                                    "content": "no corruption. The accuracy of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        238,
+                                        686,
+                                        307,
+                                        699
+                                    ],
+                                    "score": 0.91,
+                                    "content": "O ( \\alpha \\sqrt { \\log ( 1 / \\alpha ) } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        307,
+                                        686,
+                                        506,
+                                        700
+                                    ],
+                                    "score": 1.0,
+                                    "content": "matches the lower bound shown in [33] for any",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                698,
+                                506,
+                                710
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        698,
+                                        506,
+                                        710
+                                    ],
+                                    "score": 1.0,
+                                    "content": "polynomial time statistical query algorithm, and it nearly matches the information theoretical lower",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                708,
+                                504,
+                                724
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        708,
+                                        228,
+                                        724
+                                    ],
+                                    "score": 1.0,
+                                    "content": "bound on robust estimation of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        229,
+                                        711,
+                                        251,
+                                        723
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\Omega ( \\alpha )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        251,
+                                        708,
+                                        409,
+                                        724
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". On the other hand, the second term of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        409,
+                                        708,
+                                        504,
+                                        723
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\widetilde \\Omega ( d ^ { 3 / 2 } / ( \\varepsilon \\alpha \\log ( 1 / \\alpha ) ) )",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 47
+                        }
+                    ],
+                    "index": 45
+                }
+            ],
+            "page_idx": 6,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        302,
+                        741,
+                        308,
+                        750
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                302,
+                                741,
+                                309,
+                                752
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        302,
+                                        741,
+                                        309,
+                                        752
+                                    ],
+                                    "score": 1.0,
+                                    "content": "7",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        72,
+                        505,
+                        128
+                    ],
+                    "lines": [],
+                    "index": 2,
+                    "bbox_fs": [
+                        105,
+                        73,
+                        506,
+                        129
+                    ],
+                    "lines_deleted": true
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        106,
+                        139,
+                        303,
+                        151
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                138,
+                                304,
+                                153
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        138,
+                                        304,
+                                        153
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2.3.2 Adaptive filtering with DPTHRESHOLD",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        }
+                    ],
+                    "index": 5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        157,
+                        505,
+                        257
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                157,
+                                506,
+                                171
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        157,
+                                        506,
+                                        171
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Novelty. The corresponding non-private filtering of [36, Algorithm 9] for robust mean estimation",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                169,
+                                506,
+                                181
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        169,
+                                        506,
+                                        181
+                                    ],
+                                    "score": 1.0,
+                                    "content": "takes advantage of an adaptive threshold, but filters out each sample independently resulting in a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                179,
+                                506,
+                                192
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        179,
+                                        506,
+                                        192
+                                    ],
+                                    "score": 1.0,
+                                    "content": "prohibitively large sensitivity; the coupling between each sample and the randomness used to filter it",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                191,
+                                506,
+                                204
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        191,
+                                        506,
+                                        204
+                                    ],
+                                    "score": 1.0,
+                                    "content": "can change widely between two neighboring datasets. On the other hand, Algorithm 1 (i.e., Filter(·)",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                200,
+                                507,
+                                215
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        200,
+                                        448,
+                                        215
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in Algorithm 6) takes advantage of jointly filtering all points above a single threshold",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        448,
+                                        201,
+                                        477,
+                                        213
+                                    ],
+                                    "score": 0.92,
+                                    "content": "B ^ { 2 } d Z _ { t }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        477,
+                                        200,
+                                        507,
+                                        215
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                211,
+                                504,
+                                226
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        211,
+                                        183,
+                                        226
+                                    ],
+                                    "score": 1.0,
+                                    "content": "single randomness",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        183,
+                                        213,
+                                        247,
+                                        224
+                                    ],
+                                    "score": 0.91,
+                                    "content": "Z _ { t } \\sim \\mathrm { U n i f } [ 0 , 1 ]",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        247,
+                                        211,
+                                        484,
+                                        226
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", but the non-adaptive (and hence large) choice of the range",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        485,
+                                        212,
+                                        504,
+                                        223
+                                    ],
+                                    "score": 0.89,
+                                    "content": "B ^ { 2 } d",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                223,
+                                505,
+                                236
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        223,
+                                        505,
+                                        236
+                                    ],
+                                    "score": 1.0,
+                                    "content": "results in a large number of iterations because each filtering only decrease the score by little. To",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                235,
+                                506,
+                                246
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        235,
+                                        506,
+                                        246
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sufficiently reduce the total score while maintaining a small sensitivity, we introduce a filter with a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                245,
+                                227,
+                                258
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        245,
+                                        227,
+                                        258
+                                    ],
+                                    "score": 1.0,
+                                    "content": "single and adaptive threshold.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        }
+                    ],
+                    "index": 10,
+                    "bbox_fs": [
+                        105,
+                        157,
+                        507,
+                        258
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        261,
+                        505,
+                        328
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                261,
+                                506,
+                                274
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        261,
+                                        347,
+                                        274
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Algorithm. Our goal here is to privately find a single scalar",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        348,
+                                        264,
+                                        354,
+                                        273
+                                    ],
+                                    "score": 0.83,
+                                    "content": "\\rho",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        355,
+                                        261,
+                                        506,
+                                        274
+                                    ],
+                                    "score": 1.0,
+                                    "content": "such that when a randomized filter is",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                272,
+                                506,
+                                285
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        272,
+                                        193,
+                                        285
+                                    ],
+                                    "score": 1.0,
+                                    "content": "applied on the scores",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        194,
+                                        272,
+                                        212,
+                                        285
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\{ \\tau _ { i } \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        213,
+                                        272,
+                                        320,
+                                        285
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with a (random) threshold",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        320,
+                                        273,
+                                        334,
+                                        284
+                                    ],
+                                    "score": 0.88,
+                                    "content": "\\rho Z",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        335,
+                                        272,
+                                        359,
+                                        285
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(with",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        359,
+                                        273,
+                                        368,
+                                        283
+                                    ],
+                                    "score": 0.73,
+                                    "content": "Z",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        368,
+                                        272,
+                                        442,
+                                        285
+                                    ],
+                                    "score": 1.0,
+                                    "content": "drawn uniform in",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        442,
+                                        273,
+                                        465,
+                                        285
+                                    ],
+                                    "score": 0.25,
+                                    "content": "[ 0 , 1 ] )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        465,
+                                        272,
+                                        506,
+                                        285
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", we filter",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                284,
+                                505,
+                                295
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        284,
+                                        505,
+                                        295
+                                    ],
+                                    "score": 1.0,
+                                    "content": "out enough samples to make progress in each iteration while ensuring that we do not remove too",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                295,
+                                506,
+                                307
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        295,
+                                        506,
+                                        307
+                                    ],
+                                    "score": 1.0,
+                                    "content": "many uncorrupted samples. This is a slight generalization of the non-private algorithm in Section 2.1,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                304,
+                                506,
+                                320
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        304,
+                                        176,
+                                        320
+                                    ],
+                                    "score": 1.0,
+                                    "content": "which simply set",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        176,
+                                        307,
+                                        242,
+                                        318
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\rho = \\operatorname* { m a x } _ { j \\in S _ { t } } \\tau _ { j }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        242,
+                                        304,
+                                        506,
+                                        320
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". While this guarantees the filter removes more corrupted samples",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                316,
+                                498,
+                                329
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        316,
+                                        498,
+                                        329
+                                    ],
+                                    "score": 1.0,
+                                    "content": "than good samples, it does not make sufficient progress in reducing the total score of the samples.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        }
+                    ],
+                    "index": 17.5,
+                    "bbox_fs": [
+                        105,
+                        261,
+                        506,
+                        329
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        333,
+                        505,
+                        366
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                332,
+                                505,
+                                344
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        332,
+                                        505,
+                                        344
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Ideally, we want the thresholding to decrease the total score by a constant multiplicative factor, which",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                342,
+                                506,
+                                357
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        342,
+                                        506,
+                                        357
+                                    ],
+                                    "score": 1.0,
+                                    "content": "will in the end allow the algorithm to terminate within logarithmic iterations. To this end, we propose",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                354,
+                                414,
+                                367
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        354,
+                                        243,
+                                        367
+                                    ],
+                                    "score": 1.0,
+                                    "content": "a new scheme of using the largest",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        243,
+                                        356,
+                                        250,
+                                        366
+                                    ],
+                                    "score": 0.82,
+                                    "content": "\\rho",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        250,
+                                        354,
+                                        414,
+                                        367
+                                    ],
+                                    "score": 1.0,
+                                    "content": "such that the following inequality holds:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        }
+                    ],
+                    "index": 22,
+                    "bbox_fs": [
+                        105,
+                        332,
+                        506,
+                        367
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        223,
+                        370,
+                        367,
+                        398
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                223,
+                                370,
+                                367,
+                                398
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        223,
+                                        370,
+                                        367,
+                                        398
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\sum _ { \\tau _ { i } > \\rho } ( \\tau _ { i } - \\rho ) \\geq 0 . 3 1 \\sum _ { \\tau _ { i } \\in S _ { t } } ( \\tau _ { i } - 1 ) .",
+                                    "type": "interline_equation",
+                                    "image_path": "6e717cdd2620c553eeae0f0bac341635344a945d930dd30b1b3e271d36a171e1.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 24.5,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                223,
+                                370,
+                                367,
+                                384.0
+                            ],
+                            "spans": [],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                223,
+                                384.0,
+                                367,
+                                398.0
+                            ],
+                            "spans": [],
+                            "index": 25
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        402,
+                        505,
+                        459
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                403,
+                                505,
+                                416
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        403,
+                                        505,
+                                        416
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We use a private histogram of the scores to approximate this threshold. Similar to [55, 58], we use",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                413,
+                                506,
+                                428
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        413,
+                                        327,
+                                        428
+                                    ],
+                                    "score": 1.0,
+                                    "content": "geometrically increasing bin sizes such that we use only",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        327,
+                                        414,
+                                        376,
+                                        426
+                                    ],
+                                    "score": 0.92,
+                                    "content": "O ( \\log B ^ { 2 } d )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        376,
+                                        413,
+                                        506,
+                                        428
+                                    ],
+                                    "score": 1.0,
+                                    "content": "bins while achieving a preferred",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                425,
+                                505,
+                                437
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        425,
+                                        333,
+                                        437
+                                    ],
+                                    "score": 1.0,
+                                    "content": "multiplicative error in our quantization. At each epoch",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        334,
+                                        427,
+                                        340,
+                                        435
+                                    ],
+                                    "score": 0.75,
+                                    "content": "s",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        340,
+                                        425,
+                                        395,
+                                        437
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and iteration",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        396,
+                                        426,
+                                        400,
+                                        435
+                                    ],
+                                    "score": 0.76,
+                                    "content": "t",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        401,
+                                        425,
+                                        505,
+                                        437
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", we run DPTHRESHOLD",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                435,
+                                505,
+                                449
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        435,
+                                        266,
+                                        449
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sketched in the following to approximate",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        267,
+                                        438,
+                                        273,
+                                        447
+                                    ],
+                                    "score": 0.82,
+                                    "content": "\\rho",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        273,
+                                        435,
+                                        505,
+                                        449
+                                    ],
+                                    "score": 1.0,
+                                    "content": "followed by a random filter. Step 3 replaces the non-private",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                446,
+                                402,
+                                460
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        446,
+                                        402,
+                                        460
+                                    ],
+                                    "score": 1.0,
+                                    "content": "condition in Eq. (1). A complete description is provided in Algorithm 11.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        }
+                    ],
+                    "index": 28,
+                    "bbox_fs": [
+                        105,
+                        403,
+                        506,
+                        460
+                    ]
+                },
+                {
+                    "type": "index",
+                    "bbox": [
+                        129,
+                        465,
+                        506,
+                        536
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                128,
+                                461,
+                                494,
+                                483
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        128,
+                                        461,
+                                        317,
+                                        483
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1. Privately compute scores for all data points",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        317,
+                                        465,
+                                        489,
+                                        480
+                                    ],
+                                    "score": 0.89,
+                                    "content": "i \\in S _ { t } ^ { ( s ) } : \\tau _ { i } \\gets ( x _ { i } - \\mu _ { t } ) ^ { \\top } U _ { t } ^ { ( s ) } ( x _ { i } - \\mu _ { t } ) ;",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        489,
+                                        461,
+                                        494,
+                                        483
+                                    ],
+                                    "score": 1.0,
+                                    "content": ";",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                123,
+                                471,
+                                510,
+                                503
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        123,
+                                        471,
+                                        264,
+                                        503
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2. Compute a private histogram",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        264,
+                                        478,
+                                        326,
+                                        496
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\{ \\tilde { h } _ { j } \\} _ { j = 1 } ^ { 2 + \\log ( B ^ { 2 } d ) }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        326,
+                                        471,
+                                        510,
+                                        503
+                                    ],
+                                    "score": 1.0,
+                                    "content": "t i  i  t  t i  t of the scores over geometrically sized bins",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                142,
+                                493,
+                                447,
+                                511
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        142,
+                                        496,
+                                        206,
+                                        509
+                                    ],
+                                    "score": 0.86,
+                                    "content": "I _ { 1 } = [ 1 / 4 , 1 / 2 )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        206,
+                                        493,
+                                        209,
+                                        511
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        209,
+                                        494,
+                                        441,
+                                        509
+                                    ],
+                                    "score": 0.84,
+                                    "content": "I _ { 2 } = [ 1 / 2 , 1 ) , \\ldots , \\bar { I } _ { 2 + \\log ( B ^ { 2 } d ) } = [ 2 ^ { \\log ( B ^ { 2 } d ) - 1 } , 2 ^ { \\log ( B ^ { 2 } d ) } ]",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        442,
+                                        493,
+                                        447,
+                                        511
+                                    ],
+                                    "score": 1.0,
+                                    "content": ";",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                504,
+                                472,
+                                527
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        504,
+                                        241,
+                                        527
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3. Privately find the largest",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        242,
+                                        510,
+                                        248,
+                                        519
+                                    ],
+                                    "score": 0.77,
+                                    "content": "\\ell",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        248,
+                                        504,
+                                        290,
+                                        527
+                                    ],
+                                    "score": 1.0,
+                                    "content": "satisfying",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        290,
+                                        509,
+                                        468,
+                                        525
+                                    ],
+                                    "score": 0.85,
+                                    "content": "\\begin{array} { r } { \\sum _ { j \\ge \\ell } ( 2 ^ { j } - 2 ^ { \\ell } ) \\tilde { h } _ { j } \\ge 0 . 3 1 \\sum _ { i \\in S _ { t } ^ { ( s ) } } ( \\tau _ { i } - 1 ) ; } \\end{array}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        468,
+                                        504,
+                                        472,
+                                        527
+                                    ],
+                                    "score": 1.0,
+                                    "content": ";",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                523,
+                                210,
+                                536
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        523,
+                                        172,
+                                        536
+                                    ],
+                                    "score": 1.0,
+                                    "content": "4. Output",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        173,
+                                        523,
+                                        201,
+                                        536
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\rho = 2 ^ { \\ell }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        201,
+                                        523,
+                                        210,
+                                        536
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35,
+                            "is_list_start_line": true
+                        }
+                    ],
+                    "index": 33,
+                    "bbox_fs": [
+                        123,
+                        461,
+                        510,
+                        536
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        549,
+                        227,
+                        563
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                549,
+                                228,
+                                566
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        549,
+                                        228,
+                                        566
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3 Analyses of PRIME",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        }
+                    ],
+                    "index": 36
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        574,
+                        506,
+                        608
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                574,
+                                506,
+                                586
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        574,
+                                        506,
+                                        586
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Building on the framework of Algorithm 1, PRIME (Algorithm 9) replaces the score with the MMW-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                586,
+                                505,
+                                597
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        586,
+                                        208,
+                                        597
+                                    ],
+                                    "score": 1.0,
+                                    "content": "based score presented in",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        209,
+                                        586,
+                                        235,
+                                        596
+                                    ],
+                                    "score": 0.82,
+                                    "content": "\\ S 2 . 3 . 1",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        235,
+                                        586,
+                                        505,
+                                        597
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and the filter with the adaptive DPTHRESHOLD. This reduces the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                596,
+                                397,
+                                609
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        596,
+                                        200,
+                                        609
+                                    ],
+                                    "score": 1.0,
+                                    "content": "number of iterations to",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        200,
+                                        596,
+                                        268,
+                                        609
+                                    ],
+                                    "score": 0.93,
+                                    "content": "T = O ( ( \\log d ) ^ { 2 } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        269,
+                                        596,
+                                        397,
+                                        609
+                                    ],
+                                    "score": 1.0,
+                                    "content": "achieving the following bound.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        }
+                    ],
+                    "index": 38,
+                    "bbox_fs": [
+                        105,
+                        574,
+                        506,
+                        609
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        611,
+                        505,
+                        650
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                610,
+                                506,
+                                624
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        610,
+                                        204,
+                                        624
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Theorem 6. PRIME is",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        205,
+                                        611,
+                                        226,
+                                        623
+                                    ],
+                                    "score": 0.91,
+                                    "content": "( \\varepsilon , \\delta )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        227,
+                                        610,
+                                        399,
+                                        624
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-differentially private. Under Assumption",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        400,
+                                        612,
+                                        406,
+                                        621
+                                    ],
+                                    "score": 0.4,
+                                    "content": "I",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        406,
+                                        610,
+                                        506,
+                                        624
+                                    ],
+                                    "score": 1.0,
+                                    "content": "there exists a universal",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                622,
+                                506,
+                                638
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        623,
+                                        144,
+                                        638
+                                    ],
+                                    "score": 1.0,
+                                    "content": "constant",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        144,
+                                        624,
+                                        191,
+                                        636
+                                    ],
+                                    "score": 0.89,
+                                    "content": "c \\in ( 0 , 0 . 1 )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        191,
+                                        623,
+                                        240,
+                                        638
+                                    ],
+                                    "score": 1.0,
+                                    "content": "such that if",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        241,
+                                        625,
+                                        266,
+                                        636
+                                    ],
+                                    "score": 0.85,
+                                    "content": "\\alpha \\leq c",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        267,
+                                        623,
+                                        285,
+                                        638
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        286,
+                                        622,
+                                        447,
+                                        637
+                                    ],
+                                    "score": 0.92,
+                                    "content": "n = \\widetilde \\Omega ( ( d / \\alpha ^ { 2 } ) + ( d ^ { 3 / 2 } / ( \\varepsilon \\alpha ) ) \\log ( 1 / \\delta ) )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        447,
+                                        623,
+                                        506,
+                                        638
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", then PRIME",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                636,
+                                350,
+                                651
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        636,
+                                        144,
+                                        651
+                                    ],
+                                    "score": 1.0,
+                                    "content": "achieves",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        144,
+                                        636,
+                                        264,
+                                        650
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\| \\hat { \\mu } - \\mu \\| _ { 2 } = O ( \\alpha \\sqrt { \\log ( 1 / \\alpha ) } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        265,
+                                        636,
+                                        350,
+                                        651
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with probability 0.9.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        }
+                    ],
+                    "index": 41,
+                    "bbox_fs": [
+                        105,
+                        610,
+                        506,
+                        651
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        658,
+                        505,
+                        723
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                658,
+                                505,
+                                673
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        659,
+                                        303,
+                                        673
+                                    ],
+                                    "score": 1.0,
+                                    "content": "A proof is provided in Appendix F. The notation",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        303,
+                                        658,
+                                        322,
+                                        672
+                                    ],
+                                    "score": 0.9,
+                                    "content": "\\widetilde { \\Omega } ( \\cdot )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        322,
+                                        659,
+                                        430,
+                                        673
+                                    ],
+                                    "score": 1.0,
+                                    "content": "hides logarithmic terms in",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        430,
+                                        660,
+                                        450,
+                                        671
+                                    ],
+                                    "score": 0.56,
+                                    "content": "d , R",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        450,
+                                        659,
+                                        470,
+                                        673
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        470,
+                                        660,
+                                        488,
+                                        672
+                                    ],
+                                    "score": 0.9,
+                                    "content": "1 / \\alpha",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        488,
+                                        659,
+                                        505,
+                                        673
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". To",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                672,
+                                507,
+                                687
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        672,
+                                        185,
+                                        687
+                                    ],
+                                    "score": 1.0,
+                                    "content": "achieve an error of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        186,
+                                        672,
+                                        254,
+                                        686
+                                    ],
+                                    "score": 0.92,
+                                    "content": "O ( \\alpha \\sqrt { \\log ( 1 / \\alpha ) } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        255,
+                                        672,
+                                        314,
+                                        687
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", the first term",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        315,
+                                        672,
+                                        389,
+                                        686
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\widetilde \\Omega ( d / \\alpha ^ { 2 } \\log ( 1 / \\alpha ) )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        389,
+                                        672,
+                                        507,
+                                        687
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is necessary even if there is",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                686,
+                                506,
+                                700
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        686,
+                                        237,
+                                        700
+                                    ],
+                                    "score": 1.0,
+                                    "content": "no corruption. The accuracy of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        238,
+                                        686,
+                                        307,
+                                        699
+                                    ],
+                                    "score": 0.91,
+                                    "content": "O ( \\alpha \\sqrt { \\log ( 1 / \\alpha ) } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        307,
+                                        686,
+                                        506,
+                                        700
+                                    ],
+                                    "score": 1.0,
+                                    "content": "matches the lower bound shown in [33] for any",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                698,
+                                506,
+                                710
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        698,
+                                        506,
+                                        710
+                                    ],
+                                    "score": 1.0,
+                                    "content": "polynomial time statistical query algorithm, and it nearly matches the information theoretical lower",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                708,
+                                504,
+                                724
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        708,
+                                        228,
+                                        724
+                                    ],
+                                    "score": 1.0,
+                                    "content": "bound on robust estimation of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        229,
+                                        711,
+                                        251,
+                                        723
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\Omega ( \\alpha )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        251,
+                                        708,
+                                        409,
+                                        724
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". On the other hand, the second term of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        409,
+                                        708,
+                                        504,
+                                        723
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\widetilde \\Omega ( d ^ { 3 / 2 } / ( \\varepsilon \\alpha \\log ( 1 / \\alpha ) ) )",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 47
+                        }
+                    ],
+                    "index": 45,
+                    "bbox_fs": [
+                        105,
+                        658,
+                        507,
+                        724
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        72,
+                        504,
+                        95
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                70,
+                                506,
+                                86
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        70,
+                                        191,
+                                        86
+                                    ],
+                                    "score": 1.0,
+                                    "content": "has an extra factor of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        192,
+                                        72,
+                                        210,
+                                        83
+                                    ],
+                                    "score": 0.91,
+                                    "content": "d ^ { 1 / 2 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        210,
+                                        70,
+                                        506,
+                                        86
+                                    ],
+                                    "score": 1.0,
+                                    "content": "compared to the optimal one achieved by exponential time Algorithm 2. It",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                83,
+                                415,
+                                96
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        83,
+                                        415,
+                                        96
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is an open question if this gap can be closed by a polynomial time algorithm.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        }
+                    ],
+                    "index": 0.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        100,
+                        505,
+                        145
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                100,
+                                504,
+                                112
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        100,
+                                        504,
+                                        112
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The bottleneck is the private matrix multiplicative weights. Such spectral analyses are crucial in",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                111,
+                                505,
+                                123
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        111,
+                                        505,
+                                        123
+                                    ],
+                                    "score": 1.0,
+                                    "content": "filter-based robust estimators. Even for a special case of privately computing the top principal",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                122,
+                                505,
+                                135
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        122,
+                                        343,
+                                        135
+                                    ],
+                                    "score": 1.0,
+                                    "content": "component, the best polynomial time algorithm requires",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        343,
+                                        122,
+                                        377,
+                                        135
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\bar { O } ( d ^ { 3 / 2 } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        378,
+                                        122,
+                                        505,
+                                        135
+                                    ],
+                                    "score": 1.0,
+                                    "content": "samples [39, 18, 78], and this",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                133,
+                                380,
+                                146
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        133,
+                                        380,
+                                        146
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sample complexity is also necessary as shown in [39, Corollary 25].",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        }
+                    ],
+                    "index": 3.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        150,
+                        506,
+                        198
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                148,
+                                506,
+                                164
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        148,
+                                        246,
+                                        164
+                                    ],
+                                    "score": 1.0,
+                                    "content": "To boost the success probability to",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        247,
+                                        150,
+                                        270,
+                                        162
+                                    ],
+                                    "score": 0.91,
+                                    "content": "1 - \\zeta",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        271,
+                                        148,
+                                        333,
+                                        164
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for some small",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        333,
+                                        150,
+                                        357,
+                                        162
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\zeta > 0",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        358,
+                                        148,
+                                        430,
+                                        164
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", we need an extra",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        431,
+                                        150,
+                                        468,
+                                        162
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\log ( 1 / \\zeta )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        468,
+                                        148,
+                                        506,
+                                        164
+                                    ],
+                                    "score": 1.0,
+                                    "content": "factor in",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                160,
+                                507,
+                                174
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        160,
+                                        487,
+                                        174
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the sample complexity to make sure the dataset satisfies the regularity condition with probability",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        487,
+                                        161,
+                                        503,
+                                        173
+                                    ],
+                                    "score": 0.88,
+                                    "content": "\\zeta / 2",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        503,
+                                        160,
+                                        507,
+                                        174
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                172,
+                                505,
+                                187
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        172,
+                                        203,
+                                        187
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Then we can run PRIME",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        204,
+                                        173,
+                                        240,
+                                        186
+                                    ],
+                                    "score": 0.9,
+                                    "content": "\\log ( 1 / \\zeta )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        240,
+                                        172,
+                                        430,
+                                        187
+                                    ],
+                                    "score": 1.0,
+                                    "content": "times and choose the output of a run that satisfies",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        431,
+                                        173,
+                                        505,
+                                        186
+                                    ],
+                                    "score": 0.92,
+                                    "content": "n ^ { ( s ) } > n ( 1 { - } 1 0 \\alpha )",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                184,
+                                270,
+                                199
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        184,
+                                        123,
+                                        199
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        123,
+                                        185,
+                                        207,
+                                        198
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\lambda ^ { ( s ) } \\le C \\alpha \\log ( 1 / \\alpha )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        207,
+                                        184,
+                                        270,
+                                        199
+                                    ],
+                                    "score": 1.0,
+                                    "content": "at termination.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        }
+                    ],
+                    "index": 7.5
+                },
+                {
+                    "type": "image",
+                    "bbox": [
+                        116,
+                        211,
+                        506,
+                        288
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                116,
+                                211,
+                                506,
+                                288
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        211,
+                                        506,
+                                        288
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                116,
+                                                211,
+                                                506,
+                                                288
+                                            ],
+                                            "score": 0.964,
+                                            "type": "image",
+                                            "image_path": "46a53c55cd9bcc36cd60657375546149f2ff820d7058a98083ffc81ea44ac394.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 11,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        211,
+                                        506,
+                                        236.66666666666666
+                                    ],
+                                    "spans": [],
+                                    "index": 10
+                                },
+                                {
+                                    "bbox": [
+                                        116,
+                                        236.66666666666666,
+                                        506,
+                                        262.3333333333333
+                                    ],
+                                    "spans": [],
+                                    "index": 11
+                                },
+                                {
+                                    "bbox": [
+                                        116,
+                                        262.3333333333333,
+                                        506,
+                                        288.0
+                                    ],
+                                    "spans": [],
+                                    "index": 12
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                108,
+                                298,
+                                504,
+                                332
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        299,
+                                        505,
+                                        311
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                299,
+                                                505,
+                                                311
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 1: Private mean estimators (e.g., DP mean [52]) are vulnerable to adversarial corruption",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 13
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        309,
+                                        505,
+                                        323
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                309,
+                                                505,
+                                                323
+                                            ],
+                                            "score": 1.0,
+                                            "content": "especially in high dimensions, while the proposed PRIME achieves robustness (and privacy) regardless",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 14
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        320,
+                                        238,
+                                        333
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                320,
+                                                238,
+                                                333
+                                            ],
+                                            "score": 1.0,
+                                            "content": "of the dimension of the samples.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 15
+                                }
+                            ],
+                            "index": 14
+                        }
+                    ],
+                    "index": 12.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        341,
+                        505,
+                        515
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                340,
+                                504,
+                                354
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        340,
+                                        445,
+                                        354
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Numerical experiments support our theoretical claims. The left figure with",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        445,
+                                        341,
+                                        504,
+                                        353
+                                    ],
+                                    "score": 0.88,
+                                    "content": "( \\alpha , \\varepsilon , \\delta , n ) \\ =",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                353,
+                                505,
+                                366
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        353,
+                                        187,
+                                        366
+                                    ],
+                                    "score": 0.85,
+                                    "content": "( 0 . 0 5 , 2 0 , 0 . 0 1 , 1 0 ^ { 6 } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        188,
+                                        353,
+                                        248,
+                                        366
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is in the large",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        248,
+                                        356,
+                                        257,
+                                        364
+                                    ],
+                                    "score": 0.74,
+                                    "content": "\\alpha",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        257,
+                                        353,
+                                        465,
+                                        366
+                                    ],
+                                    "score": 1.0,
+                                    "content": "regime where the DP Mean error is dominates by",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        465,
+                                        353,
+                                        486,
+                                        365
+                                    ],
+                                    "score": 0.9,
+                                    "content": "\\alpha \\sqrt { d }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        487,
+                                        353,
+                                        505,
+                                        366
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                365,
+                                506,
+                                380
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        365,
+                                        175,
+                                        380
+                                    ],
+                                    "score": 1.0,
+                                    "content": "PRIME error by",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        175,
+                                        365,
+                                        229,
+                                        379
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\alpha \\sqrt { \\log ( 1 / \\alpha ) }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        230,
+                                        365,
+                                        506,
+                                        380
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Hence, PRIME error is constant whereas DP Mean error increases",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                378,
+                                504,
+                                392
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        378,
+                                        186,
+                                        392
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with the dimension",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        187,
+                                        380,
+                                        193,
+                                        389
+                                    ],
+                                    "score": 0.72,
+                                    "content": "d",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        194,
+                                        378,
+                                        293,
+                                        392
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". The second figure with √",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        294,
+                                        378,
+                                        435,
+                                        391
+                                    ],
+                                    "score": 0.92,
+                                    "content": "( \\alpha , \\varepsilon , \\delta , n ) = ( 0 . 0 0 1 , 2 0 , 0 . 0 1 , 1 0 ^ { 6 } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        436,
+                                        378,
+                                        496,
+                                        392
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is in the small",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        496,
+                                        381,
+                                        504,
+                                        389
+                                    ],
+                                    "score": 0.75,
+                                    "content": "\\alpha",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                390,
+                                506,
+                                405
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        390,
+                                        271,
+                                        405
+                                    ],
+                                    "score": 1.0,
+                                    "content": "regime when DP Mean error consists of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        272,
+                                        390,
+                                        331,
+                                        404
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\alpha \\sqrt { d } + \\sqrt { d / n }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        331,
+                                        390,
+                                        451,
+                                        405
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and PRIME is dominated by",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        451,
+                                        390,
+                                        478,
+                                        404
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\sqrt { d / n }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        478,
+                                        390,
+                                        506,
+                                        405
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Both",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                402,
+                                505,
+                                416
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        402,
+                                        225,
+                                        416
+                                    ],
+                                    "score": 1.0,
+                                    "content": "increase with the dimension",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        225,
+                                        403,
+                                        232,
+                                        412
+                                    ],
+                                    "score": 0.77,
+                                    "content": "d",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        232,
+                                        402,
+                                        423,
+                                        416
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", and the gap can be made large by increasing",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        424,
+                                        405,
+                                        431,
+                                        412
+                                    ],
+                                    "score": 0.75,
+                                    "content": "\\alpha",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        432,
+                                        402,
+                                        505,
+                                        416
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". The right figure√",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                413,
+                                505,
+                                428
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        413,
+                                        126,
+                                        428
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        127,
+                                        414,
+                                        259,
+                                        427
+                                    ],
+                                    "score": 0.91,
+                                    "content": "( \\alpha , \\delta , d , n ) = ( 0 . 1 , 0 . 0 1 , 1 0 , 1 0 ^ { 6 } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        260,
+                                        413,
+                                        421,
+                                        428
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is when DP Mean error is dominated by",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        421,
+                                        414,
+                                        442,
+                                        426
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\alpha \\sqrt { d }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        443,
+                                        413,
+                                        505,
+                                        428
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and PRIME by",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                107,
+                                425,
+                                506,
+                                441
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        426,
+                                        161,
+                                        440
+                                    ],
+                                    "score": 0.9,
+                                    "content": "\\alpha \\sqrt { \\log ( 1 / \\alpha ) }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        161,
+                                        425,
+                                        187,
+                                        441
+                                    ],
+                                    "score": 1.0,
+                                    "content": "when",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        187,
+                                        427,
+                                        252,
+                                        440
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\varepsilon > c d ^ { 1 . 5 } / ( \\alpha n )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        252,
+                                        425,
+                                        506,
+                                        441
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Below this threshold, which happens in this example around",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                438,
+                                506,
+                                451
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        440,
+                                        146,
+                                        449
+                                    ],
+                                    "score": 0.84,
+                                    "content": "\\varepsilon = 0 . 0 5",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        146,
+                                        438,
+                                        470,
+                                        451
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", the added noise in the private mechanism starts to dominate with decreasing",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        470,
+                                        441,
+                                        476,
+                                        448
+                                    ],
+                                    "score": 0.68,
+                                    "content": "\\varepsilon",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        477,
+                                        438,
+                                        506,
+                                        451
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Both",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                450,
+                                505,
+                                462
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        450,
+                                        470,
+                                        462
+                                    ],
+                                    "score": 1.0,
+                                    "content": "algorithms have respective thresholds below which the error increases with decreasing",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        471,
+                                        452,
+                                        477,
+                                        460
+                                    ],
+                                    "score": 0.7,
+                                    "content": "\\varepsilon",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        477,
+                                        450,
+                                        505,
+                                        462
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". This",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                459,
+                                506,
+                                474
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        459,
+                                        506,
+                                        474
+                                    ],
+                                    "score": 1.0,
+                                    "content": "threshold is larger for PRIME because it uses the privacy budget to perform multiple operations and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                471,
+                                505,
+                                484
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        471,
+                                        505,
+                                        484
+                                    ],
+                                    "score": 1.0,
+                                    "content": "hence the noise added to the final output is larger compared to DP Mean. Below this threshold, which",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                482,
+                                505,
+                                495
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        482,
+                                        337,
+                                        495
+                                    ],
+                                    "score": 1.0,
+                                    "content": "can be easily determined based on the known parameters",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        337,
+                                        482,
+                                        380,
+                                        494
+                                    ],
+                                    "score": 0.93,
+                                    "content": "( \\varepsilon , \\delta , n , \\alpha )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        380,
+                                        482,
+                                        505,
+                                        495
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", we should either collect more",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                489,
+                                504,
+                                510
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        489,
+                                        478,
+                                        510
+                                    ],
+                                    "score": 1.0,
+                                    "content": "data (which will decrease the threshold) or give up filtering and spend all privacy budget on",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        479,
+                                        495,
+                                        504,
+                                        505
+                                    ],
+                                    "score": 0.3,
+                                    "content": "q _ { \\mathrm { r a n g e } }",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                504,
+                                506,
+                                517
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        504,
+                                        506,
+                                        517
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and the empirical mean (which will reduce the error). Details of the experiments are in Appendix L.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        }
+                    ],
+                    "index": 23
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        105,
+                        534,
+                        459,
+                        549
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                103,
+                                533,
+                                460,
+                                552
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        103,
+                                        533,
+                                        460,
+                                        552
+                                    ],
+                                    "score": 1.0,
+                                    "content": "4 Exponential time algorithm with near-optimal sample complexity",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        }
+                    ],
+                    "index": 31
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        561,
+                        505,
+                        650
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                562,
+                                505,
+                                574
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        562,
+                                        505,
+                                        574
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Novelty. An existing exponential time algorithm for robust and private mean estimation in [14]",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                573,
+                                506,
+                                585
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        573,
+                                        506,
+                                        585
+                                    ],
+                                    "score": 1.0,
+                                    "content": "strictly requires the uncorrupted samples to be drawn from a Gaussian distribution. We also provide a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                583,
+                                505,
+                                596
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        583,
+                                        505,
+                                        596
+                                    ],
+                                    "score": 1.0,
+                                    "content": "similar algorithm based on private Tukey median in Appendix I and its analysis in Appendix J. In",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                594,
+                                506,
+                                607
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        594,
+                                        506,
+                                        607
+                                    ],
+                                    "score": 1.0,
+                                    "content": "this section, we introduce a novel estimator that achieves near-optimal guarantees for more general",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                605,
+                                506,
+                                618
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        605,
+                                        506,
+                                        618
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sub-Gaussian distributions (and also covariance bounded distributions) but takes an exponential",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                616,
+                                506,
+                                629
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        616,
+                                        506,
+                                        629
+                                    ],
+                                    "score": 1.0,
+                                    "content": "run-time. Its innovation is in leveraging on the resilience property of well-behaved distributions not",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                628,
+                                505,
+                                640
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        628,
+                                        505,
+                                        640
+                                    ],
+                                    "score": 1.0,
+                                    "content": "only to estimate the mean robustly (which is the standard use of the property) but also to adaptively",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                638,
+                                459,
+                                651
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        638,
+                                        459,
+                                        651
+                                    ],
+                                    "score": 1.0,
+                                    "content": "bound the sensitivity of the estimator, thus achieving optimal privacy-accuracy tradeoff.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        }
+                    ],
+                    "index": 35.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        654,
+                        505,
+                        679
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                653,
+                                507,
+                                668
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        653,
+                                        324,
+                                        668
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Definition 4.1 (Resilience from Definition 1 in [73]).",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        324,
+                                        656,
+                                        332,
+                                        665
+                                    ],
+                                    "score": 0.35,
+                                    "content": "A",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        332,
+                                        653,
+                                        385,
+                                        668
+                                    ],
+                                    "score": 1.0,
+                                    "content": "set of points",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        385,
+                                        655,
+                                        419,
+                                        667
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\{ x _ { i } \\} _ { i \\in S }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        419,
+                                        653,
+                                        453,
+                                        668
+                                    ],
+                                    "score": 1.0,
+                                    "content": "lying in",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        453,
+                                        654,
+                                        466,
+                                        665
+                                    ],
+                                    "score": 0.87,
+                                    "content": "\\mathbb { R } ^ { d }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        467,
+                                        653,
+                                        477,
+                                        668
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        477,
+                                        654,
+                                        502,
+                                        666
+                                    ],
+                                    "score": 0.91,
+                                    "content": "( \\sigma , \\alpha )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        502,
+                                        653,
+                                        507,
+                                        668
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                665,
+                                506,
+                                679
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        665,
+                                        200,
+                                        679
+                                    ],
+                                    "score": 1.0,
+                                    "content": "resilient around a point",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        201,
+                                        667,
+                                        208,
+                                        677
+                                    ],
+                                    "score": 0.38,
+                                    "content": "\\mu",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        207,
+                                        666,
+                                        343,
+                                        679
+                                    ],
+                                    "score": 0.86,
+                                    "content": "\\begin{array} { r } { . i f \\| ( 1 / | T | ) \\sum _ { i \\in T } ( x _ { i } - \\mu ) \\| _ { 2 } \\leq \\sigma } \\end{array}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        344,
+                                        665,
+                                        402,
+                                        679
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for all subsets",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        402,
+                                        667,
+                                        430,
+                                        676
+                                    ],
+                                    "score": 0.89,
+                                    "content": "T \\subset S",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        431,
+                                        665,
+                                        459,
+                                        679
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of size",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        460,
+                                        666,
+                                        503,
+                                        678
+                                    ],
+                                    "score": 0.92,
+                                    "content": "( 1 - \\alpha ) | S |",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        503,
+                                        665,
+                                        506,
+                                        679
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        }
+                    ],
+                    "index": 40.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        688,
+                        505,
+                        722
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                688,
+                                506,
+                                702
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        688,
+                                        279,
+                                        702
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Algorithm. As data is corrupted, we define",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        279,
+                                        689,
+                                        302,
+                                        700
+                                    ],
+                                    "score": 0.92,
+                                    "content": "R ( S )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        303,
+                                        688,
+                                        506,
+                                        702
+                                    ],
+                                    "score": 1.0,
+                                    "content": "as a surrogate for resilience of the uncorrupted part",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                699,
+                                505,
+                                712
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        699,
+                                        157,
+                                        712
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of the set. If",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        157,
+                                        700,
+                                        165,
+                                        710
+                                    ],
+                                    "score": 0.8,
+                                    "content": "S",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        165,
+                                        699,
+                                        245,
+                                        712
+                                    ],
+                                    "score": 1.0,
+                                    "content": "indeed consists of a",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        245,
+                                        700,
+                                        270,
+                                        710
+                                    ],
+                                    "score": 0.87,
+                                    "content": "1 - \\alpha",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        270,
+                                        699,
+                                        505,
+                                        712
+                                    ],
+                                    "score": 1.0,
+                                    "content": "fraction of independent samples from the promised class of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                711,
+                                488,
+                                724
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        711,
+                                        239,
+                                        724
+                                    ],
+                                    "score": 1.0,
+                                    "content": "distributions, the goodness score",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        239,
+                                        711,
+                                        262,
+                                        723
+                                    ],
+                                    "score": 0.91,
+                                    "content": "R ( S )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        262,
+                                        711,
+                                        488,
+                                        724
+                                    ],
+                                    "score": 1.0,
+                                    "content": "will be close to the resilience property of the good data.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        }
+                    ],
+                    "index": 43
+                }
+            ],
+            "page_idx": 7,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        302,
+                        742,
+                        308,
+                        750
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                301,
+                                740,
+                                310,
+                                752
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        301,
+                                        740,
+                                        310,
+                                        752
+                                    ],
+                                    "score": 1.0,
+                                    "content": "",
+                                    "type": "text",
+                                    "height": 12,
+                                    "width": 9
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        72,
+                        504,
+                        95
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                70,
+                                506,
+                                86
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        70,
+                                        191,
+                                        86
+                                    ],
+                                    "score": 1.0,
+                                    "content": "has an extra factor of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        192,
+                                        72,
+                                        210,
+                                        83
+                                    ],
+                                    "score": 0.91,
+                                    "content": "d ^ { 1 / 2 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        210,
+                                        70,
+                                        506,
+                                        86
+                                    ],
+                                    "score": 1.0,
+                                    "content": "compared to the optimal one achieved by exponential time Algorithm 2. It",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                83,
+                                415,
+                                96
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        83,
+                                        415,
+                                        96
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is an open question if this gap can be closed by a polynomial time algorithm.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        }
+                    ],
+                    "index": 0.5,
+                    "bbox_fs": [
+                        105,
+                        70,
+                        506,
+                        96
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        100,
+                        505,
+                        145
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                100,
+                                504,
+                                112
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        100,
+                                        504,
+                                        112
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The bottleneck is the private matrix multiplicative weights. Such spectral analyses are crucial in",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                111,
+                                505,
+                                123
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        111,
+                                        505,
+                                        123
+                                    ],
+                                    "score": 1.0,
+                                    "content": "filter-based robust estimators. Even for a special case of privately computing the top principal",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                122,
+                                505,
+                                135
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        122,
+                                        343,
+                                        135
+                                    ],
+                                    "score": 1.0,
+                                    "content": "component, the best polynomial time algorithm requires",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        343,
+                                        122,
+                                        377,
+                                        135
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\bar { O } ( d ^ { 3 / 2 } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        378,
+                                        122,
+                                        505,
+                                        135
+                                    ],
+                                    "score": 1.0,
+                                    "content": "samples [39, 18, 78], and this",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                133,
+                                380,
+                                146
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        133,
+                                        380,
+                                        146
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sample complexity is also necessary as shown in [39, Corollary 25].",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        }
+                    ],
+                    "index": 3.5,
+                    "bbox_fs": [
+                        105,
+                        100,
+                        505,
+                        146
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        150,
+                        506,
+                        198
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                148,
+                                506,
+                                164
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        148,
+                                        246,
+                                        164
+                                    ],
+                                    "score": 1.0,
+                                    "content": "To boost the success probability to",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        247,
+                                        150,
+                                        270,
+                                        162
+                                    ],
+                                    "score": 0.91,
+                                    "content": "1 - \\zeta",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        271,
+                                        148,
+                                        333,
+                                        164
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for some small",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        333,
+                                        150,
+                                        357,
+                                        162
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\zeta > 0",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        358,
+                                        148,
+                                        430,
+                                        164
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", we need an extra",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        431,
+                                        150,
+                                        468,
+                                        162
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\log ( 1 / \\zeta )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        468,
+                                        148,
+                                        506,
+                                        164
+                                    ],
+                                    "score": 1.0,
+                                    "content": "factor in",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                160,
+                                507,
+                                174
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        160,
+                                        487,
+                                        174
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the sample complexity to make sure the dataset satisfies the regularity condition with probability",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        487,
+                                        161,
+                                        503,
+                                        173
+                                    ],
+                                    "score": 0.88,
+                                    "content": "\\zeta / 2",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        503,
+                                        160,
+                                        507,
+                                        174
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                172,
+                                505,
+                                187
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        172,
+                                        203,
+                                        187
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Then we can run PRIME",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        204,
+                                        173,
+                                        240,
+                                        186
+                                    ],
+                                    "score": 0.9,
+                                    "content": "\\log ( 1 / \\zeta )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        240,
+                                        172,
+                                        430,
+                                        187
+                                    ],
+                                    "score": 1.0,
+                                    "content": "times and choose the output of a run that satisfies",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        431,
+                                        173,
+                                        505,
+                                        186
+                                    ],
+                                    "score": 0.92,
+                                    "content": "n ^ { ( s ) } > n ( 1 { - } 1 0 \\alpha )",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                184,
+                                270,
+                                199
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        184,
+                                        123,
+                                        199
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        123,
+                                        185,
+                                        207,
+                                        198
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\lambda ^ { ( s ) } \\le C \\alpha \\log ( 1 / \\alpha )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        207,
+                                        184,
+                                        270,
+                                        199
+                                    ],
+                                    "score": 1.0,
+                                    "content": "at termination.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        }
+                    ],
+                    "index": 7.5,
+                    "bbox_fs": [
+                        104,
+                        148,
+                        507,
+                        199
+                    ]
+                },
+                {
+                    "type": "image",
+                    "bbox": [
+                        116,
+                        211,
+                        506,
+                        288
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                116,
+                                211,
+                                506,
+                                288
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        211,
+                                        506,
+                                        288
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                116,
+                                                211,
+                                                506,
+                                                288
+                                            ],
+                                            "score": 0.964,
+                                            "type": "image",
+                                            "image_path": "46a53c55cd9bcc36cd60657375546149f2ff820d7058a98083ffc81ea44ac394.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 11,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        211,
+                                        506,
+                                        236.66666666666666
+                                    ],
+                                    "spans": [],
+                                    "index": 10
+                                },
+                                {
+                                    "bbox": [
+                                        116,
+                                        236.66666666666666,
+                                        506,
+                                        262.3333333333333
+                                    ],
+                                    "spans": [],
+                                    "index": 11
+                                },
+                                {
+                                    "bbox": [
+                                        116,
+                                        262.3333333333333,
+                                        506,
+                                        288.0
+                                    ],
+                                    "spans": [],
+                                    "index": 12
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                108,
+                                298,
+                                504,
+                                332
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        299,
+                                        505,
+                                        311
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                299,
+                                                505,
+                                                311
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 1: Private mean estimators (e.g., DP mean [52]) are vulnerable to adversarial corruption",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 13
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        309,
+                                        505,
+                                        323
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                309,
+                                                505,
+                                                323
+                                            ],
+                                            "score": 1.0,
+                                            "content": "especially in high dimensions, while the proposed PRIME achieves robustness (and privacy) regardless",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 14
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        320,
+                                        238,
+                                        333
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                320,
+                                                238,
+                                                333
+                                            ],
+                                            "score": 1.0,
+                                            "content": "of the dimension of the samples.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 15
+                                }
+                            ],
+                            "index": 14
+                        }
+                    ],
+                    "index": 12.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        341,
+                        505,
+                        515
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                340,
+                                504,
+                                354
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        340,
+                                        445,
+                                        354
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Numerical experiments support our theoretical claims. The left figure with",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        445,
+                                        341,
+                                        504,
+                                        353
+                                    ],
+                                    "score": 0.88,
+                                    "content": "( \\alpha , \\varepsilon , \\delta , n ) \\ =",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                353,
+                                505,
+                                366
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        353,
+                                        187,
+                                        366
+                                    ],
+                                    "score": 0.85,
+                                    "content": "( 0 . 0 5 , 2 0 , 0 . 0 1 , 1 0 ^ { 6 } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        188,
+                                        353,
+                                        248,
+                                        366
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is in the large",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        248,
+                                        356,
+                                        257,
+                                        364
+                                    ],
+                                    "score": 0.74,
+                                    "content": "\\alpha",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        257,
+                                        353,
+                                        465,
+                                        366
+                                    ],
+                                    "score": 1.0,
+                                    "content": "regime where the DP Mean error is dominates by",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        465,
+                                        353,
+                                        486,
+                                        365
+                                    ],
+                                    "score": 0.9,
+                                    "content": "\\alpha \\sqrt { d }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        487,
+                                        353,
+                                        505,
+                                        366
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                365,
+                                506,
+                                380
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        365,
+                                        175,
+                                        380
+                                    ],
+                                    "score": 1.0,
+                                    "content": "PRIME error by",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        175,
+                                        365,
+                                        229,
+                                        379
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\alpha \\sqrt { \\log ( 1 / \\alpha ) }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        230,
+                                        365,
+                                        506,
+                                        380
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Hence, PRIME error is constant whereas DP Mean error increases",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                378,
+                                504,
+                                392
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        378,
+                                        186,
+                                        392
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with the dimension",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        187,
+                                        380,
+                                        193,
+                                        389
+                                    ],
+                                    "score": 0.72,
+                                    "content": "d",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        194,
+                                        378,
+                                        293,
+                                        392
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". The second figure with √",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        294,
+                                        378,
+                                        435,
+                                        391
+                                    ],
+                                    "score": 0.92,
+                                    "content": "( \\alpha , \\varepsilon , \\delta , n ) = ( 0 . 0 0 1 , 2 0 , 0 . 0 1 , 1 0 ^ { 6 } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        436,
+                                        378,
+                                        496,
+                                        392
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is in the small",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        496,
+                                        381,
+                                        504,
+                                        389
+                                    ],
+                                    "score": 0.75,
+                                    "content": "\\alpha",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                390,
+                                506,
+                                405
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        390,
+                                        271,
+                                        405
+                                    ],
+                                    "score": 1.0,
+                                    "content": "regime when DP Mean error consists of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        272,
+                                        390,
+                                        331,
+                                        404
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\alpha \\sqrt { d } + \\sqrt { d / n }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        331,
+                                        390,
+                                        451,
+                                        405
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and PRIME is dominated by",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        451,
+                                        390,
+                                        478,
+                                        404
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\sqrt { d / n }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        478,
+                                        390,
+                                        506,
+                                        405
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Both",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                402,
+                                505,
+                                416
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        402,
+                                        225,
+                                        416
+                                    ],
+                                    "score": 1.0,
+                                    "content": "increase with the dimension",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        225,
+                                        403,
+                                        232,
+                                        412
+                                    ],
+                                    "score": 0.77,
+                                    "content": "d",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        232,
+                                        402,
+                                        423,
+                                        416
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", and the gap can be made large by increasing",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        424,
+                                        405,
+                                        431,
+                                        412
+                                    ],
+                                    "score": 0.75,
+                                    "content": "\\alpha",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        432,
+                                        402,
+                                        505,
+                                        416
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". The right figure√",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                413,
+                                505,
+                                428
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        413,
+                                        126,
+                                        428
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        127,
+                                        414,
+                                        259,
+                                        427
+                                    ],
+                                    "score": 0.91,
+                                    "content": "( \\alpha , \\delta , d , n ) = ( 0 . 1 , 0 . 0 1 , 1 0 , 1 0 ^ { 6 } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        260,
+                                        413,
+                                        421,
+                                        428
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is when DP Mean error is dominated by",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        421,
+                                        414,
+                                        442,
+                                        426
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\alpha \\sqrt { d }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        443,
+                                        413,
+                                        505,
+                                        428
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and PRIME by",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                107,
+                                425,
+                                506,
+                                441
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        426,
+                                        161,
+                                        440
+                                    ],
+                                    "score": 0.9,
+                                    "content": "\\alpha \\sqrt { \\log ( 1 / \\alpha ) }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        161,
+                                        425,
+                                        187,
+                                        441
+                                    ],
+                                    "score": 1.0,
+                                    "content": "when",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        187,
+                                        427,
+                                        252,
+                                        440
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\varepsilon > c d ^ { 1 . 5 } / ( \\alpha n )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        252,
+                                        425,
+                                        506,
+                                        441
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Below this threshold, which happens in this example around",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                438,
+                                506,
+                                451
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        440,
+                                        146,
+                                        449
+                                    ],
+                                    "score": 0.84,
+                                    "content": "\\varepsilon = 0 . 0 5",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        146,
+                                        438,
+                                        470,
+                                        451
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", the added noise in the private mechanism starts to dominate with decreasing",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        470,
+                                        441,
+                                        476,
+                                        448
+                                    ],
+                                    "score": 0.68,
+                                    "content": "\\varepsilon",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        477,
+                                        438,
+                                        506,
+                                        451
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Both",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                450,
+                                505,
+                                462
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        450,
+                                        470,
+                                        462
+                                    ],
+                                    "score": 1.0,
+                                    "content": "algorithms have respective thresholds below which the error increases with decreasing",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        471,
+                                        452,
+                                        477,
+                                        460
+                                    ],
+                                    "score": 0.7,
+                                    "content": "\\varepsilon",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        477,
+                                        450,
+                                        505,
+                                        462
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". This",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                459,
+                                506,
+                                474
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        459,
+                                        506,
+                                        474
+                                    ],
+                                    "score": 1.0,
+                                    "content": "threshold is larger for PRIME because it uses the privacy budget to perform multiple operations and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                471,
+                                505,
+                                484
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        471,
+                                        505,
+                                        484
+                                    ],
+                                    "score": 1.0,
+                                    "content": "hence the noise added to the final output is larger compared to DP Mean. Below this threshold, which",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                482,
+                                505,
+                                495
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        482,
+                                        337,
+                                        495
+                                    ],
+                                    "score": 1.0,
+                                    "content": "can be easily determined based on the known parameters",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        337,
+                                        482,
+                                        380,
+                                        494
+                                    ],
+                                    "score": 0.93,
+                                    "content": "( \\varepsilon , \\delta , n , \\alpha )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        380,
+                                        482,
+                                        505,
+                                        495
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", we should either collect more",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                489,
+                                504,
+                                510
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        489,
+                                        478,
+                                        510
+                                    ],
+                                    "score": 1.0,
+                                    "content": "data (which will decrease the threshold) or give up filtering and spend all privacy budget on",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        479,
+                                        495,
+                                        504,
+                                        505
+                                    ],
+                                    "score": 0.3,
+                                    "content": "q _ { \\mathrm { r a n g e } }",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                504,
+                                506,
+                                517
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        504,
+                                        506,
+                                        517
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and the empirical mean (which will reduce the error). Details of the experiments are in Appendix L.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        }
+                    ],
+                    "index": 23,
+                    "bbox_fs": [
+                        104,
+                        340,
+                        506,
+                        517
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        105,
+                        534,
+                        459,
+                        549
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                103,
+                                533,
+                                460,
+                                552
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        103,
+                                        533,
+                                        460,
+                                        552
+                                    ],
+                                    "score": 1.0,
+                                    "content": "4 Exponential time algorithm with near-optimal sample complexity",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        }
+                    ],
+                    "index": 31
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        561,
+                        505,
+                        650
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                562,
+                                505,
+                                574
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        562,
+                                        505,
+                                        574
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Novelty. An existing exponential time algorithm for robust and private mean estimation in [14]",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                573,
+                                506,
+                                585
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        573,
+                                        506,
+                                        585
+                                    ],
+                                    "score": 1.0,
+                                    "content": "strictly requires the uncorrupted samples to be drawn from a Gaussian distribution. We also provide a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                583,
+                                505,
+                                596
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        583,
+                                        505,
+                                        596
+                                    ],
+                                    "score": 1.0,
+                                    "content": "similar algorithm based on private Tukey median in Appendix I and its analysis in Appendix J. In",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                594,
+                                506,
+                                607
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        594,
+                                        506,
+                                        607
+                                    ],
+                                    "score": 1.0,
+                                    "content": "this section, we introduce a novel estimator that achieves near-optimal guarantees for more general",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                605,
+                                506,
+                                618
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        605,
+                                        506,
+                                        618
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sub-Gaussian distributions (and also covariance bounded distributions) but takes an exponential",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                616,
+                                506,
+                                629
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        616,
+                                        506,
+                                        629
+                                    ],
+                                    "score": 1.0,
+                                    "content": "run-time. Its innovation is in leveraging on the resilience property of well-behaved distributions not",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                628,
+                                505,
+                                640
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        628,
+                                        505,
+                                        640
+                                    ],
+                                    "score": 1.0,
+                                    "content": "only to estimate the mean robustly (which is the standard use of the property) but also to adaptively",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                638,
+                                459,
+                                651
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        638,
+                                        459,
+                                        651
+                                    ],
+                                    "score": 1.0,
+                                    "content": "bound the sensitivity of the estimator, thus achieving optimal privacy-accuracy tradeoff.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        }
+                    ],
+                    "index": 35.5,
+                    "bbox_fs": [
+                        104,
+                        562,
+                        506,
+                        651
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        654,
+                        505,
+                        679
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                653,
+                                507,
+                                668
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        653,
+                                        324,
+                                        668
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Definition 4.1 (Resilience from Definition 1 in [73]).",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        324,
+                                        656,
+                                        332,
+                                        665
+                                    ],
+                                    "score": 0.35,
+                                    "content": "A",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        332,
+                                        653,
+                                        385,
+                                        668
+                                    ],
+                                    "score": 1.0,
+                                    "content": "set of points",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        385,
+                                        655,
+                                        419,
+                                        667
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\{ x _ { i } \\} _ { i \\in S }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        419,
+                                        653,
+                                        453,
+                                        668
+                                    ],
+                                    "score": 1.0,
+                                    "content": "lying in",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        453,
+                                        654,
+                                        466,
+                                        665
+                                    ],
+                                    "score": 0.87,
+                                    "content": "\\mathbb { R } ^ { d }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        467,
+                                        653,
+                                        477,
+                                        668
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        477,
+                                        654,
+                                        502,
+                                        666
+                                    ],
+                                    "score": 0.91,
+                                    "content": "( \\sigma , \\alpha )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        502,
+                                        653,
+                                        507,
+                                        668
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                665,
+                                506,
+                                679
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        665,
+                                        200,
+                                        679
+                                    ],
+                                    "score": 1.0,
+                                    "content": "resilient around a point",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        201,
+                                        667,
+                                        208,
+                                        677
+                                    ],
+                                    "score": 0.38,
+                                    "content": "\\mu",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        207,
+                                        666,
+                                        343,
+                                        679
+                                    ],
+                                    "score": 0.86,
+                                    "content": "\\begin{array} { r } { . i f \\| ( 1 / | T | ) \\sum _ { i \\in T } ( x _ { i } - \\mu ) \\| _ { 2 } \\leq \\sigma } \\end{array}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        344,
+                                        665,
+                                        402,
+                                        679
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for all subsets",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        402,
+                                        667,
+                                        430,
+                                        676
+                                    ],
+                                    "score": 0.89,
+                                    "content": "T \\subset S",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        431,
+                                        665,
+                                        459,
+                                        679
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of size",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        460,
+                                        666,
+                                        503,
+                                        678
+                                    ],
+                                    "score": 0.92,
+                                    "content": "( 1 - \\alpha ) | S |",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        503,
+                                        665,
+                                        506,
+                                        679
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        }
+                    ],
+                    "index": 40.5,
+                    "bbox_fs": [
+                        105,
+                        653,
+                        507,
+                        679
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        688,
+                        505,
+                        722
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                688,
+                                506,
+                                702
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        688,
+                                        279,
+                                        702
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Algorithm. As data is corrupted, we define",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        279,
+                                        689,
+                                        302,
+                                        700
+                                    ],
+                                    "score": 0.92,
+                                    "content": "R ( S )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        303,
+                                        688,
+                                        506,
+                                        702
+                                    ],
+                                    "score": 1.0,
+                                    "content": "as a surrogate for resilience of the uncorrupted part",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                699,
+                                505,
+                                712
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        699,
+                                        157,
+                                        712
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of the set. If",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        157,
+                                        700,
+                                        165,
+                                        710
+                                    ],
+                                    "score": 0.8,
+                                    "content": "S",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        165,
+                                        699,
+                                        245,
+                                        712
+                                    ],
+                                    "score": 1.0,
+                                    "content": "indeed consists of a",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        245,
+                                        700,
+                                        270,
+                                        710
+                                    ],
+                                    "score": 0.87,
+                                    "content": "1 - \\alpha",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        270,
+                                        699,
+                                        505,
+                                        712
+                                    ],
+                                    "score": 1.0,
+                                    "content": "fraction of independent samples from the promised class of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                711,
+                                488,
+                                724
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        711,
+                                        239,
+                                        724
+                                    ],
+                                    "score": 1.0,
+                                    "content": "distributions, the goodness score",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        239,
+                                        711,
+                                        262,
+                                        723
+                                    ],
+                                    "score": 0.91,
+                                    "content": "R ( S )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        262,
+                                        711,
+                                        488,
+                                        724
+                                    ],
+                                    "score": 1.0,
+                                    "content": "will be close to the resilience property of the good data.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        }
+                    ],
+                    "index": 43,
+                    "bbox_fs": [
+                        105,
+                        688,
+                        506,
+                        724
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        71,
+                        423,
+                        86
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                69,
+                                423,
+                                88
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        69,
+                                        267,
+                                        88
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Definition 4.2 (Goodness of a set). For",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        267,
+                                        72,
+                                        369,
+                                        86
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\mu ( S ) = ( 1 / | S | ) \\sum _ { i \\in S } x _ { i }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        369,
+                                        69,
+                                        423,
+                                        88
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", let us define",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        }
+                    ],
+                    "index": 0
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        160,
+                        90,
+                        452,
+                        111
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                160,
+                                90,
+                                452,
+                                111
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        160,
+                                        90,
+                                        452,
+                                        111
+                                    ],
+                                    "score": 0.87,
+                                    "content": "R ( S ) \\stackrel { \\Delta } { = } \\operatorname* { m i n } _ { \\substack { S ^ { \\prime } \\subset S , | S ^ { \\prime } | = ( 1 - 2 \\alpha ) | S | . } } \\operatorname* { m a x } _ { \\substack { T \\subset S ^ { \\prime } , | T | = ( 1 - \\alpha ) | S ^ { \\prime } | . } } \\| \\mu ( T ) - \\mu ( S ^ { \\prime } ) \\| _ { 2 } .",
+                                    "type": "interline_equation",
+                                    "image_path": "a63fff882c56dd4d22cd7a5e3f73c597dc62e918ef47ae2bcdaf9517671e805d.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 1,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                160,
+                                90,
+                                452,
+                                111
+                            ],
+                            "spans": [],
+                            "index": 1
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        128,
+                        506,
+                        200
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                129,
+                                505,
+                                141
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        129,
+                                        505,
+                                        141
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Algorithm 2 first checks if the resilience matches that of the promised distribution. The data is",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                139,
+                                506,
+                                153
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        139,
+                                        183,
+                                        153
+                                    ],
+                                    "score": 1.0,
+                                    "content": "pre-processed with",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        184,
+                                        141,
+                                        209,
+                                        152
+                                    ],
+                                    "score": 0.88,
+                                    "content": "q _ { \\mathrm { r a n g e } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        209,
+                                        139,
+                                        304,
+                                        153
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to ensure we can check",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        305,
+                                        140,
+                                        327,
+                                        152
+                                    ],
+                                    "score": 0.92,
+                                    "content": "R ( S )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        328,
+                                        139,
+                                        506,
+                                        153
+                                    ],
+                                    "score": 1.0,
+                                    "content": "privately. Once resilience is cleared, we can",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                150,
+                                506,
+                                164
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        150,
+                                        370,
+                                        164
+                                    ],
+                                    "score": 1.0,
+                                    "content": "safely use the exponential mechanism based on the score function",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        370,
+                                        151,
+                                        401,
+                                        163
+                                    ],
+                                    "score": 0.93,
+                                    "content": "d ( \\hat { \\mu } , S )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        401,
+                                        150,
+                                        506,
+                                        164
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in Definition 4.3 to select",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                161,
+                                506,
+                                174
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        161,
+                                        220,
+                                        174
+                                    ],
+                                    "score": 1.0,
+                                    "content": "an approximate robust mean",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        221,
+                                        162,
+                                        228,
+                                        173
+                                    ],
+                                    "score": 0.84,
+                                    "content": "\\hat { \\mu }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        228,
+                                        161,
+                                        506,
+                                        174
+                                    ],
+                                    "score": 1.0,
+                                    "content": "privately. The choice of the sensitivity critically relies on the fact that",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                173,
+                                506,
+                                187
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        173,
+                                        278,
+                                        187
+                                    ],
+                                    "score": 1.0,
+                                    "content": "resilient datasets have small sensitivity of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        278,
+                                        173,
+                                        365,
+                                        187
+                                    ],
+                                    "score": 0.93,
+                                    "content": "{ \\cal O } ( ( 1 / n ) \\sqrt { \\log ( 1 / \\alpha ) } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        365,
+                                        173,
+                                        506,
+                                        187
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Without the resilience check, the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                186,
+                                445,
+                                201
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        186,
+                                        159,
+                                        201
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sensitivity is",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        159,
+                                        186,
+                                        204,
+                                        200
+                                    ],
+                                    "score": 0.93,
+                                    "content": "O ( d ^ { 1 / 2 } / n )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        204,
+                                        186,
+                                        324,
+                                        201
+                                    ],
+                                    "score": 1.0,
+                                    "content": "resulting in an extra factor of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        324,
+                                        187,
+                                        339,
+                                        198
+                                    ],
+                                    "score": 0.89,
+                                    "content": "\\sqrt { d }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        339,
+                                        186,
+                                        445,
+                                        201
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in the sample complexity.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        }
+                    ],
+                    "index": 4.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        97,
+                        228,
+                        499,
+                        337
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                227,
+                                279,
+                                241
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        227,
+                                        136,
+                                        241
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Input:",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        137,
+                                        228,
+                                        279,
+                                        241
+                                    ],
+                                    "score": 0.27,
+                                    "content": "S = \\{ x _ { i } \\} _ { i \\in [ n ] } , \\alpha \\in ( 0 , 1 / 2 ) , ( \\varepsilon , \\delta )",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                96,
+                                240,
+                                490,
+                                256
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        96,
+                                        240,
+                                        115,
+                                        256
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1 if",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        115,
+                                        242,
+                                        267,
+                                        254
+                                    ],
+                                    "score": 0.77,
+                                    "content": "n < c d ^ { 1 / 2 } \\log ( 1 / \\delta ) / \\left( \\varepsilon \\alpha \\sqrt { \\log ( 1 / \\alpha ) } \\right)",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        267,
+                                        240,
+                                        328,
+                                        256
+                                    ],
+                                    "score": 1.0,
+                                    "content": "then Output:",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        328,
+                                        241,
+                                        429,
+                                        254
+                                    ],
+                                    "score": 0.89,
+                                    "content": "\\emptyset \\ [ \\ c d ^ { 1 / 2 } \\log ( 1 / \\delta ) / \\ ( \\varepsilon \\alpha )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        430,
+                                        240,
+                                        490,
+                                        256
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for hevay-tail]",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                96,
+                                253,
+                                490,
+                                267
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        96,
+                                        253,
+                                        106,
+                                        267
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        254,
+                                        258,
+                                        266
+                                    ],
+                                    "score": 0.79,
+                                    "content": "( \\bar { x } , B ) \\gets q _ { \\mathrm { r a n g e } } ( S , ( 1 / 3 ) \\varepsilon , ( 1 / 3 ) \\delta )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        375,
+                                        254,
+                                        429,
+                                        266
+                                    ],
+                                    "score": 0.73,
+                                    "content": "[ \\ q _ { \\mathrm { r a n g e - h t } } ( \\cdot )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        430,
+                                        253,
+                                        490,
+                                        266
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for hevay-tail]",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                95,
+                                262,
+                                403,
+                                280
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        95,
+                                        262,
+                                        253,
+                                        280
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3 Project the data points onto the ball:",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        254,
+                                        264,
+                                        342,
+                                        279
+                                    ],
+                                    "score": 0.91,
+                                    "content": "x _ { i }  \\mathcal { P } _ { B _ { \\sqrt { d } B / 2 } ( \\bar { x } ) } ( x _ { i } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        342,
+                                        262,
+                                        372,
+                                        280
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", for all",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        372,
+                                        265,
+                                        400,
+                                        276
+                                    ],
+                                    "score": 0.77,
+                                    "content": "i \\in [ n ]",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        401,
+                                        262,
+                                        403,
+                                        280
+                                    ],
+                                    "score": 0.0,
+                                    "content": "",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                96,
+                                276,
+                                258,
+                                294
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        96,
+                                        276,
+                                        105,
+                                        294
+                                    ],
+                                    "score": 1.0,
+                                    "content": "4",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        278,
+                                        258,
+                                        293
+                                    ],
+                                    "score": 0.85,
+                                    "content": "\\widehat { R } ( S ) \\gets R ( S ) + \\mathrm { L a p } ( 3 B d ^ { 1 / 2 } / ( n \\varepsilon ) )",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                96,
+                                291,
+                                490,
+                                308
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        96,
+                                        293,
+                                        115,
+                                        307
+                                    ],
+                                    "score": 1.0,
+                                    "content": "5 if",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        115,
+                                        293,
+                                        212,
+                                        307
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\widehat { R } ( S ) > 2 \\alpha \\sqrt { \\log ( 1 / \\alpha ) }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        212,
+                                        293,
+                                        271,
+                                        307
+                                    ],
+                                    "score": 1.0,
+                                    "content": "then Output:",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        272,
+                                        294,
+                                        279,
+                                        305
+                                    ],
+                                    "score": 0.69,
+                                    "content": "\\varnothing",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        361,
+                                        293,
+                                        429,
+                                        307
+                                    ],
+                                    "score": 0.91,
+                                    "content": "[ \\widehat { R } ( S ) > 2 c _ { \\zeta } \\sqrt { \\alpha }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        430,
+                                        291,
+                                        490,
+                                        308
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for hevay-tail]",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                96,
+                                304,
+                                423,
+                                320
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        96,
+                                        304,
+                                        260,
+                                        320
+                                    ],
+                                    "score": 1.0,
+                                    "content": "6 else Output: a randomly drawn point",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        260,
+                                        305,
+                                        325,
+                                        320
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\hat { \\mu } \\in \\mathcal { B } _ { \\sqrt { d } B / 2 } ( \\bar { x } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        325,
+                                        304,
+                                        423,
+                                        320
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sampled from a density",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                97,
+                                317,
+                                489,
+                                337
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        97,
+                                        324,
+                                        104,
+                                        333
+                                    ],
+                                    "score": 1.0,
+                                    "content": "7",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        136,
+                                        318,
+                                        284,
+                                        334
+                                    ],
+                                    "score": 1.0,
+                                    "content": "r(ˆµ) ∝ e−(1/(24 log(1/α)))ε n d(ˆµ,S)",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        335,
+                                        321,
+                                        426,
+                                        333
+                                    ],
+                                    "score": 0.8,
+                                    "content": "[ e ^ { - ( \\varepsilon n \\sqrt { \\alpha } / ( 2 4 c _ { \\zeta } ) ) d ( \\hat { \\mu } , S ) }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        426,
+                                        317,
+                                        489,
+                                        337
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for heavy-tail]",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        }
+                    ],
+                    "index": 11.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        351,
+                        504,
+                        374
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                351,
+                                506,
+                                365
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        351,
+                                        229,
+                                        365
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We propose the score function",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        229,
+                                        352,
+                                        260,
+                                        364
+                                    ],
+                                    "score": 0.93,
+                                    "content": "d ( \\hat { \\mu } , S )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        261,
+                                        351,
+                                        506,
+                                        365
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in the following definition, which is a robust estimator of the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                361,
+                                299,
+                                375
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        361,
+                                        288,
+                                        375
+                                    ],
+                                    "score": 1.0,
+                                    "content": "distance between the mean and the candidate",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        288,
+                                        363,
+                                        295,
+                                        374
+                                    ],
+                                    "score": 0.84,
+                                    "content": "\\hat { \\mu }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        296,
+                                        361,
+                                        299,
+                                        375
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        }
+                    ],
+                    "index": 16.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        376,
+                        505,
+                        415
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                375,
+                                504,
+                                390
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        375,
+                                        240,
+                                        390
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Definition 4.3. For a set of data",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        240,
+                                        377,
+                                        274,
+                                        389
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\{ x _ { i } \\} _ { i \\in S }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        274,
+                                        375,
+                                        308,
+                                        390
+                                    ],
+                                    "score": 1.0,
+                                    "content": "lying in",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        308,
+                                        376,
+                                        321,
+                                        388
+                                    ],
+                                    "score": 0.85,
+                                    "content": "\\mathbb { R } ^ { d }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        322,
+                                        375,
+                                        357,
+                                        390
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", for any",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        357,
+                                        376,
+                                        396,
+                                        388
+                                    ],
+                                    "score": 0.91,
+                                    "content": "v \\in \\mathbb { S } ^ { d - 1 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        396,
+                                        375,
+                                        426,
+                                        390
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", define",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        426,
+                                        378,
+                                        440,
+                                        388
+                                    ],
+                                    "score": 0.86,
+                                    "content": "\\mathcal { T } ^ { v }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        441,
+                                        375,
+                                        479,
+                                        390
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to be the",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        479,
+                                        378,
+                                        504,
+                                        389
+                                    ],
+                                    "score": 0.9,
+                                    "content": "3 \\alpha | S |",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                388,
+                                506,
+                                402
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        388,
+                                        201,
+                                        402
+                                    ],
+                                    "score": 1.0,
+                                    "content": "points with the largest",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        202,
+                                        388,
+                                        224,
+                                        400
+                                    ],
+                                    "score": 0.92,
+                                    "content": "v ^ { \\top } x _ { i }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        224,
+                                        388,
+                                        253,
+                                        402
+                                    ],
+                                    "score": 1.0,
+                                    "content": "value,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        253,
+                                        390,
+                                        266,
+                                        399
+                                    ],
+                                    "score": 0.86,
+                                    "content": "B ^ { v }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        267,
+                                        388,
+                                        307,
+                                        402
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to be the",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        308,
+                                        389,
+                                        333,
+                                        401
+                                    ],
+                                    "score": 0.91,
+                                    "content": "3 \\alpha | S |",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        333,
+                                        388,
+                                        435,
+                                        402
+                                    ],
+                                    "score": 1.0,
+                                    "content": "points with the smallest",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        435,
+                                        388,
+                                        457,
+                                        400
+                                    ],
+                                    "score": 0.91,
+                                    "content": "v ^ { \\top } x _ { i }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        458,
+                                        388,
+                                        506,
+                                        402
+                                    ],
+                                    "score": 1.0,
+                                    "content": "value, and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                107,
+                                398,
+                                417,
+                                416
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        401,
+                                        197,
+                                        414
+                                    ],
+                                    "score": 0.9,
+                                    "content": "\\mathcal { M } ^ { v } = S \\setminus ( \\mathcal { T } ^ { v } \\cup B ^ { v } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        197,
+                                        398,
+                                        230,
+                                        416
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Define",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        230,
+                                        401,
+                                        411,
+                                        415
+                                    ],
+                                    "score": 0.78,
+                                    "content": "\\begin{array} { r l r } { d ( \\hat { \\mu } , S ) } & { \\triangleq } & { \\operatorname* { m a x } _ { v \\in \\mathbb { S } ^ { d - 1 } } \\left. v ^ { \\top } \\left( \\mu ( \\mathcal { M } ^ { v } ) - \\hat { \\mu } \\right) \\right. } \\end{array}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        411,
+                                        398,
+                                        417,
+                                        416
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        }
+                    ],
+                    "index": 19
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        422,
+                        505,
+                        480
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                422,
+                                506,
+                                436
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        422,
+                                        218,
+                                        436
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Analysis. For any direction",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        219,
+                                        425,
+                                        225,
+                                        433
+                                    ],
+                                    "score": 0.74,
+                                    "content": "v",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        225,
+                                        422,
+                                        347,
+                                        436
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", the truncated mean estimator",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        347,
+                                        423,
+                                        378,
+                                        435
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\mu ( \\mathcal { M } ^ { v } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        378,
+                                        422,
+                                        506,
+                                        436
+                                    ],
+                                    "score": 1.0,
+                                    "content": "provides a robust estimation of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                433,
+                                505,
+                                446
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        433,
+                                        240,
+                                        446
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the true mean along the direction",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        240,
+                                        436,
+                                        246,
+                                        443
+                                    ],
+                                    "score": 0.71,
+                                    "content": "v",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        246,
+                                        433,
+                                        505,
+                                        446
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", thus the distance can be simply defined by taking the maximum",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                444,
+                                505,
+                                458
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        444,
+                                        182,
+                                        458
+                                    ],
+                                    "score": 1.0,
+                                    "content": "over all directions",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        182,
+                                        447,
+                                        188,
+                                        455
+                                    ],
+                                    "score": 0.7,
+                                    "content": "v",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        189,
+                                        444,
+                                        505,
+                                        458
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". We show the sensitivity of this simple estimator is bounded by the resilience",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                456,
+                                506,
+                                470
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        456,
+                                        143,
+                                        470
+                                    ],
+                                    "score": 1.0,
+                                    "content": "property",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        144,
+                                        459,
+                                        151,
+                                        467
+                                    ],
+                                    "score": 0.76,
+                                    "content": "\\sigma",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        151,
+                                        456,
+                                        198,
+                                        470
+                                    ],
+                                    "score": 1.0,
+                                    "content": "divided by",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        199,
+                                        459,
+                                        205,
+                                        467
+                                    ],
+                                    "score": 0.74,
+                                    "content": "n",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        206,
+                                        456,
+                                        247,
+                                        470
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", which is",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        248,
+                                        456,
+                                        334,
+                                        470
+                                    ],
+                                    "score": 0.93,
+                                    "content": "{ \\cal O } ( ( 1 / n ) \\sqrt { \\log ( 1 / \\alpha ) } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        334,
+                                        456,
+                                        506,
+                                        470
+                                    ],
+                                    "score": 1.0,
+                                    "content": "once the resilience check is passed. This",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                468,
+                                479,
+                                481
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        468,
+                                        479,
+                                        481
+                                    ],
+                                    "score": 1.0,
+                                    "content": "leads to the following near-optimal sample complexity. We provide a proof in Appendix H.2.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        }
+                    ],
+                    "index": 23
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        482,
+                        506,
+                        508
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                482,
+                                506,
+                                495
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        482,
+                                        465,
+                                        495
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Theorem 7 (Exponential time algorithm for sub-Gaussian distributions). Algorithm 2 is",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        465,
+                                        482,
+                                        487,
+                                        495
+                                    ],
+                                    "score": 0.83,
+                                    "content": "( \\varepsilon , \\delta )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        488,
+                                        482,
+                                        506,
+                                        495
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-DP.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                493,
+                                506,
+                                509
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        493,
+                                        182,
+                                        509
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Under Assumption",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        182,
+                                        496,
+                                        187,
+                                        505
+                                    ],
+                                    "score": 0.48,
+                                    "content": "^ { l }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        187,
+                                        493,
+                                        285,
+                                        509
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", this algorithm achieves",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        285,
+                                        494,
+                                        406,
+                                        508
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\| \\hat { \\mu } - \\mu \\| _ { 2 } = O ( \\alpha \\sqrt { \\log ( 1 / \\alpha ) } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        406,
+                                        493,
+                                        472,
+                                        509
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with probability",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        473,
+                                        495,
+                                        506,
+                                        507
+                                    ],
+                                    "score": 0.87,
+                                    "content": "1 - \\zeta i f",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 27
+                        }
+                    ],
+                    "index": 26.5
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        119,
+                        512,
+                        492,
+                        549
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                119,
+                                512,
+                                492,
+                                549
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        119,
+                                        512,
+                                        492,
+                                        549
+                                    ],
+                                    "score": 0.92,
+                                    "content": "n = \\widetilde \\Omega \\Big ( \\frac { d + \\log \\frac { 1 } { \\xi } } { \\alpha ^ { 2 } \\log \\frac { 1 } { \\alpha } } + \\frac { d \\log \\Big ( d \\sqrt { \\log ( d n / \\zeta ) } / \\alpha \\Big ) + d ^ { 1 / 2 } \\log \\frac { 1 } { \\delta } + \\log \\frac { 1 } { \\zeta } } { \\varepsilon \\alpha } + \\frac { \\sqrt { d \\log \\frac { 1 } { \\delta } } \\log \\frac { d } { \\zeta \\delta } } { \\varepsilon } \\Big ) ~ .",
+                                    "type": "interline_equation",
+                                    "image_path": "9aa30506fb06fc1067dfbb7018804b830bc0f54e20e433fb8eb13e7b665ff047.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 29,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                119,
+                                512,
+                                492,
+                                524.3333333333334
+                            ],
+                            "spans": [],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                119,
+                                524.3333333333334,
+                                492,
+                                536.6666666666667
+                            ],
+                            "spans": [],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                119,
+                                536.6666666666667,
+                                492,
+                                549.0000000000001
+                            ],
+                            "spans": [],
+                            "index": 30
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        558,
+                        506,
+                        595
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                558,
+                                506,
+                                573
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        558,
+                                        201,
+                                        573
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Run-time. Computing",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        201,
+                                        560,
+                                        224,
+                                        572
+                                    ],
+                                    "score": 0.92,
+                                    "content": "R ( S )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        225,
+                                        558,
+                                        294,
+                                        573
+                                    ],
+                                    "score": 1.0,
+                                    "content": "exactly can take",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        294,
+                                        558,
+                                        338,
+                                        572
+                                    ],
+                                    "score": 0.93,
+                                    "content": "O ( d e ^ { \\Theta ( n ) } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        338,
+                                        558,
+                                        506,
+                                        573
+                                    ],
+                                    "score": 1.0,
+                                    "content": "operations. The exponential mechanism",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                570,
+                                504,
+                                587
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        570,
+                                        180,
+                                        587
+                                    ],
+                                    "score": 1.0,
+                                    "content": "implemented with",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        181,
+                                        574,
+                                        188,
+                                        582
+                                    ],
+                                    "score": 0.77,
+                                    "content": "\\alpha",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        189,
+                                        570,
+                                        240,
+                                        587
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-covering for",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        241,
+                                        573,
+                                        248,
+                                        584
+                                    ],
+                                    "score": 0.85,
+                                    "content": "\\hat { \\mu }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        248,
+                                        570,
+                                        358,
+                                        587
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and a constant covering for",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        359,
+                                        575,
+                                        365,
+                                        582
+                                    ],
+                                    "score": 0.72,
+                                    "content": "v",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        366,
+                                        570,
+                                        401,
+                                        587
+                                    ],
+                                    "score": 1.0,
+                                    "content": "can take",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        401,
+                                        570,
+                                        504,
+                                        585
+                                    ],
+                                    "score": 0.92,
+                                    "content": "O ( n d ( \\sqrt { \\log ( d n / \\zeta ) } / \\alpha ) ^ { d } )",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                583,
+                                153,
+                                596
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        583,
+                                        153,
+                                        596
+                                    ],
+                                    "score": 1.0,
+                                    "content": "operations.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        }
+                    ],
+                    "index": 32
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        106,
+                        609,
+                        183,
+                        623
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                607,
+                                185,
+                                626
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        607,
+                                        185,
+                                        626
+                                    ],
+                                    "score": 1.0,
+                                    "content": "5 Conclusion",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        }
+                    ],
+                    "index": 34
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        633,
+                        505,
+                        722
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                634,
+                                505,
+                                647
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        634,
+                                        505,
+                                        647
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Differentially private mean estimation is brittle against a small fraction of the samples being corrupted",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                645,
+                                505,
+                                657
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        645,
+                                        505,
+                                        657
+                                    ],
+                                    "score": 1.0,
+                                    "content": "by an adversary. We show that robustness can be achieved without any increase in the sample",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                656,
+                                505,
+                                668
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        656,
+                                        505,
+                                        668
+                                    ],
+                                    "score": 1.0,
+                                    "content": "complexity by introducing a novel DP mean estimator, which requires run-time exponential in the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                668,
+                                506,
+                                680
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        668,
+                                        506,
+                                        680
+                                    ],
+                                    "score": 1.0,
+                                    "content": "dimension of the samples. The technical contribution is in leveraging the resilience property of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                677,
+                                505,
+                                690
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        677,
+                                        505,
+                                        690
+                                    ],
+                                    "score": 1.0,
+                                    "content": "well-behaved distributions in an innovative way to not only find robust mean (which is the typical",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                688,
+                                506,
+                                702
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        688,
+                                        506,
+                                        702
+                                    ],
+                                    "score": 1.0,
+                                    "content": "use case of resilience) but also bound sensitivity for optimal privacy guarantee. To cope with the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                700,
+                                505,
+                                712
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        700,
+                                        505,
+                                        712
+                                    ],
+                                    "score": 1.0,
+                                    "content": "computational challenge, we propose an efficient algorithm, which we call PRIME, that achieves the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                711,
+                                506,
+                                723
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        711,
+                                        506,
+                                        723
+                                    ],
+                                    "score": 1.0,
+                                    "content": "optimal target accuracy at the cost of an increased sample complexity. The technical contributions are",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        }
+                    ],
+                    "index": 38.5
+                }
+            ],
+            "page_idx": 8,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        302,
+                        741,
+                        309,
+                        750
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                302,
+                                741,
+                                309,
+                                752
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        302,
+                                        741,
+                                        309,
+                                        752
+                                    ],
+                                    "score": 1.0,
+                                    "content": "9",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        71,
+                        423,
+                        86
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                69,
+                                423,
+                                88
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        69,
+                                        267,
+                                        88
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Definition 4.2 (Goodness of a set). For",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        267,
+                                        72,
+                                        369,
+                                        86
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\mu ( S ) = ( 1 / | S | ) \\sum _ { i \\in S } x _ { i }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        369,
+                                        69,
+                                        423,
+                                        88
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", let us define",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        }
+                    ],
+                    "index": 0,
+                    "bbox_fs": [
+                        105,
+                        69,
+                        423,
+                        88
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        160,
+                        90,
+                        452,
+                        111
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                160,
+                                90,
+                                452,
+                                111
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        160,
+                                        90,
+                                        452,
+                                        111
+                                    ],
+                                    "score": 0.87,
+                                    "content": "R ( S ) \\stackrel { \\Delta } { = } \\operatorname* { m i n } _ { \\substack { S ^ { \\prime } \\subset S , | S ^ { \\prime } | = ( 1 - 2 \\alpha ) | S | . } } \\operatorname* { m a x } _ { \\substack { T \\subset S ^ { \\prime } , | T | = ( 1 - \\alpha ) | S ^ { \\prime } | . } } \\| \\mu ( T ) - \\mu ( S ^ { \\prime } ) \\| _ { 2 } .",
+                                    "type": "interline_equation",
+                                    "image_path": "a63fff882c56dd4d22cd7a5e3f73c597dc62e918ef47ae2bcdaf9517671e805d.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 1,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                160,
+                                90,
+                                452,
+                                111
+                            ],
+                            "spans": [],
+                            "index": 1
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        128,
+                        506,
+                        200
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                129,
+                                505,
+                                141
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        129,
+                                        505,
+                                        141
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Algorithm 2 first checks if the resilience matches that of the promised distribution. The data is",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                139,
+                                506,
+                                153
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        139,
+                                        183,
+                                        153
+                                    ],
+                                    "score": 1.0,
+                                    "content": "pre-processed with",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        184,
+                                        141,
+                                        209,
+                                        152
+                                    ],
+                                    "score": 0.88,
+                                    "content": "q _ { \\mathrm { r a n g e } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        209,
+                                        139,
+                                        304,
+                                        153
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to ensure we can check",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        305,
+                                        140,
+                                        327,
+                                        152
+                                    ],
+                                    "score": 0.92,
+                                    "content": "R ( S )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        328,
+                                        139,
+                                        506,
+                                        153
+                                    ],
+                                    "score": 1.0,
+                                    "content": "privately. Once resilience is cleared, we can",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                150,
+                                506,
+                                164
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        150,
+                                        370,
+                                        164
+                                    ],
+                                    "score": 1.0,
+                                    "content": "safely use the exponential mechanism based on the score function",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        370,
+                                        151,
+                                        401,
+                                        163
+                                    ],
+                                    "score": 0.93,
+                                    "content": "d ( \\hat { \\mu } , S )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        401,
+                                        150,
+                                        506,
+                                        164
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in Definition 4.3 to select",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                161,
+                                506,
+                                174
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        161,
+                                        220,
+                                        174
+                                    ],
+                                    "score": 1.0,
+                                    "content": "an approximate robust mean",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        221,
+                                        162,
+                                        228,
+                                        173
+                                    ],
+                                    "score": 0.84,
+                                    "content": "\\hat { \\mu }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        228,
+                                        161,
+                                        506,
+                                        174
+                                    ],
+                                    "score": 1.0,
+                                    "content": "privately. The choice of the sensitivity critically relies on the fact that",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                173,
+                                506,
+                                187
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        173,
+                                        278,
+                                        187
+                                    ],
+                                    "score": 1.0,
+                                    "content": "resilient datasets have small sensitivity of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        278,
+                                        173,
+                                        365,
+                                        187
+                                    ],
+                                    "score": 0.93,
+                                    "content": "{ \\cal O } ( ( 1 / n ) \\sqrt { \\log ( 1 / \\alpha ) } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        365,
+                                        173,
+                                        506,
+                                        187
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Without the resilience check, the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                186,
+                                445,
+                                201
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        186,
+                                        159,
+                                        201
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sensitivity is",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        159,
+                                        186,
+                                        204,
+                                        200
+                                    ],
+                                    "score": 0.93,
+                                    "content": "O ( d ^ { 1 / 2 } / n )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        204,
+                                        186,
+                                        324,
+                                        201
+                                    ],
+                                    "score": 1.0,
+                                    "content": "resulting in an extra factor of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        324,
+                                        187,
+                                        339,
+                                        198
+                                    ],
+                                    "score": 0.89,
+                                    "content": "\\sqrt { d }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        339,
+                                        186,
+                                        445,
+                                        201
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in the sample complexity.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        }
+                    ],
+                    "index": 4.5,
+                    "bbox_fs": [
+                        104,
+                        129,
+                        506,
+                        201
+                    ]
+                },
+                {
+                    "type": "index",
+                    "bbox": [
+                        97,
+                        228,
+                        499,
+                        337
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                227,
+                                279,
+                                241
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        227,
+                                        136,
+                                        241
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Input:",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        137,
+                                        228,
+                                        279,
+                                        241
+                                    ],
+                                    "score": 0.27,
+                                    "content": "S = \\{ x _ { i } \\} _ { i \\in [ n ] } , \\alpha \\in ( 0 , 1 / 2 ) , ( \\varepsilon , \\delta )",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 8,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                96,
+                                240,
+                                490,
+                                256
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        96,
+                                        240,
+                                        115,
+                                        256
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1 if",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        115,
+                                        242,
+                                        267,
+                                        254
+                                    ],
+                                    "score": 0.77,
+                                    "content": "n < c d ^ { 1 / 2 } \\log ( 1 / \\delta ) / \\left( \\varepsilon \\alpha \\sqrt { \\log ( 1 / \\alpha ) } \\right)",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        267,
+                                        240,
+                                        328,
+                                        256
+                                    ],
+                                    "score": 1.0,
+                                    "content": "then Output:",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        328,
+                                        241,
+                                        429,
+                                        254
+                                    ],
+                                    "score": 0.89,
+                                    "content": "\\emptyset \\ [ \\ c d ^ { 1 / 2 } \\log ( 1 / \\delta ) / \\ ( \\varepsilon \\alpha )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        430,
+                                        240,
+                                        490,
+                                        256
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for hevay-tail]",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                96,
+                                253,
+                                490,
+                                267
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        96,
+                                        253,
+                                        106,
+                                        267
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        254,
+                                        258,
+                                        266
+                                    ],
+                                    "score": 0.79,
+                                    "content": "( \\bar { x } , B ) \\gets q _ { \\mathrm { r a n g e } } ( S , ( 1 / 3 ) \\varepsilon , ( 1 / 3 ) \\delta )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        375,
+                                        254,
+                                        429,
+                                        266
+                                    ],
+                                    "score": 0.73,
+                                    "content": "[ \\ q _ { \\mathrm { r a n g e - h t } } ( \\cdot )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        430,
+                                        253,
+                                        490,
+                                        266
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for hevay-tail]",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                95,
+                                262,
+                                403,
+                                280
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        95,
+                                        262,
+                                        253,
+                                        280
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3 Project the data points onto the ball:",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        254,
+                                        264,
+                                        342,
+                                        279
+                                    ],
+                                    "score": 0.91,
+                                    "content": "x _ { i }  \\mathcal { P } _ { B _ { \\sqrt { d } B / 2 } ( \\bar { x } ) } ( x _ { i } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        342,
+                                        262,
+                                        372,
+                                        280
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", for all",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        372,
+                                        265,
+                                        400,
+                                        276
+                                    ],
+                                    "score": 0.77,
+                                    "content": "i \\in [ n ]",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        401,
+                                        262,
+                                        403,
+                                        280
+                                    ],
+                                    "score": 0.0,
+                                    "content": "",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                96,
+                                276,
+                                258,
+                                294
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        96,
+                                        276,
+                                        105,
+                                        294
+                                    ],
+                                    "score": 1.0,
+                                    "content": "4",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        278,
+                                        258,
+                                        293
+                                    ],
+                                    "score": 0.85,
+                                    "content": "\\widehat { R } ( S ) \\gets R ( S ) + \\mathrm { L a p } ( 3 B d ^ { 1 / 2 } / ( n \\varepsilon ) )",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 12,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                96,
+                                291,
+                                490,
+                                308
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        96,
+                                        293,
+                                        115,
+                                        307
+                                    ],
+                                    "score": 1.0,
+                                    "content": "5 if",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        115,
+                                        293,
+                                        212,
+                                        307
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\widehat { R } ( S ) > 2 \\alpha \\sqrt { \\log ( 1 / \\alpha ) }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        212,
+                                        293,
+                                        271,
+                                        307
+                                    ],
+                                    "score": 1.0,
+                                    "content": "then Output:",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        272,
+                                        294,
+                                        279,
+                                        305
+                                    ],
+                                    "score": 0.69,
+                                    "content": "\\varnothing",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        361,
+                                        293,
+                                        429,
+                                        307
+                                    ],
+                                    "score": 0.91,
+                                    "content": "[ \\widehat { R } ( S ) > 2 c _ { \\zeta } \\sqrt { \\alpha }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        430,
+                                        291,
+                                        490,
+                                        308
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for hevay-tail]",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                96,
+                                304,
+                                423,
+                                320
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        96,
+                                        304,
+                                        260,
+                                        320
+                                    ],
+                                    "score": 1.0,
+                                    "content": "6 else Output: a randomly drawn point",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        260,
+                                        305,
+                                        325,
+                                        320
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\hat { \\mu } \\in \\mathcal { B } _ { \\sqrt { d } B / 2 } ( \\bar { x } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        325,
+                                        304,
+                                        423,
+                                        320
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sampled from a density",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                97,
+                                317,
+                                489,
+                                337
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        97,
+                                        324,
+                                        104,
+                                        333
+                                    ],
+                                    "score": 1.0,
+                                    "content": "7",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        136,
+                                        318,
+                                        284,
+                                        334
+                                    ],
+                                    "score": 1.0,
+                                    "content": "r(ˆµ) ∝ e−(1/(24 log(1/α)))ε n d(ˆµ,S)",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        335,
+                                        321,
+                                        426,
+                                        333
+                                    ],
+                                    "score": 0.8,
+                                    "content": "[ e ^ { - ( \\varepsilon n \\sqrt { \\alpha } / ( 2 4 c _ { \\zeta } ) ) d ( \\hat { \\mu } , S ) }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        426,
+                                        317,
+                                        489,
+                                        337
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for heavy-tail]",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15,
+                            "is_list_start_line": true
+                        }
+                    ],
+                    "index": 11.5,
+                    "bbox_fs": [
+                        95,
+                        227,
+                        490,
+                        337
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        351,
+                        504,
+                        374
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                351,
+                                506,
+                                365
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        351,
+                                        229,
+                                        365
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We propose the score function",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        229,
+                                        352,
+                                        260,
+                                        364
+                                    ],
+                                    "score": 0.93,
+                                    "content": "d ( \\hat { \\mu } , S )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        261,
+                                        351,
+                                        506,
+                                        365
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in the following definition, which is a robust estimator of the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                361,
+                                299,
+                                375
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        361,
+                                        288,
+                                        375
+                                    ],
+                                    "score": 1.0,
+                                    "content": "distance between the mean and the candidate",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        288,
+                                        363,
+                                        295,
+                                        374
+                                    ],
+                                    "score": 0.84,
+                                    "content": "\\hat { \\mu }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        296,
+                                        361,
+                                        299,
+                                        375
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        }
+                    ],
+                    "index": 16.5,
+                    "bbox_fs": [
+                        105,
+                        351,
+                        506,
+                        375
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        376,
+                        505,
+                        415
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                375,
+                                504,
+                                390
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        375,
+                                        240,
+                                        390
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Definition 4.3. For a set of data",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        240,
+                                        377,
+                                        274,
+                                        389
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\{ x _ { i } \\} _ { i \\in S }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        274,
+                                        375,
+                                        308,
+                                        390
+                                    ],
+                                    "score": 1.0,
+                                    "content": "lying in",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        308,
+                                        376,
+                                        321,
+                                        388
+                                    ],
+                                    "score": 0.85,
+                                    "content": "\\mathbb { R } ^ { d }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        322,
+                                        375,
+                                        357,
+                                        390
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", for any",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        357,
+                                        376,
+                                        396,
+                                        388
+                                    ],
+                                    "score": 0.91,
+                                    "content": "v \\in \\mathbb { S } ^ { d - 1 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        396,
+                                        375,
+                                        426,
+                                        390
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", define",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        426,
+                                        378,
+                                        440,
+                                        388
+                                    ],
+                                    "score": 0.86,
+                                    "content": "\\mathcal { T } ^ { v }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        441,
+                                        375,
+                                        479,
+                                        390
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to be the",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        479,
+                                        378,
+                                        504,
+                                        389
+                                    ],
+                                    "score": 0.9,
+                                    "content": "3 \\alpha | S |",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                388,
+                                506,
+                                402
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        388,
+                                        201,
+                                        402
+                                    ],
+                                    "score": 1.0,
+                                    "content": "points with the largest",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        202,
+                                        388,
+                                        224,
+                                        400
+                                    ],
+                                    "score": 0.92,
+                                    "content": "v ^ { \\top } x _ { i }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        224,
+                                        388,
+                                        253,
+                                        402
+                                    ],
+                                    "score": 1.0,
+                                    "content": "value,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        253,
+                                        390,
+                                        266,
+                                        399
+                                    ],
+                                    "score": 0.86,
+                                    "content": "B ^ { v }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        267,
+                                        388,
+                                        307,
+                                        402
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to be the",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        308,
+                                        389,
+                                        333,
+                                        401
+                                    ],
+                                    "score": 0.91,
+                                    "content": "3 \\alpha | S |",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        333,
+                                        388,
+                                        435,
+                                        402
+                                    ],
+                                    "score": 1.0,
+                                    "content": "points with the smallest",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        435,
+                                        388,
+                                        457,
+                                        400
+                                    ],
+                                    "score": 0.91,
+                                    "content": "v ^ { \\top } x _ { i }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        458,
+                                        388,
+                                        506,
+                                        402
+                                    ],
+                                    "score": 1.0,
+                                    "content": "value, and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                107,
+                                398,
+                                417,
+                                416
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        401,
+                                        197,
+                                        414
+                                    ],
+                                    "score": 0.9,
+                                    "content": "\\mathcal { M } ^ { v } = S \\setminus ( \\mathcal { T } ^ { v } \\cup B ^ { v } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        197,
+                                        398,
+                                        230,
+                                        416
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Define",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        230,
+                                        401,
+                                        411,
+                                        415
+                                    ],
+                                    "score": 0.78,
+                                    "content": "\\begin{array} { r l r } { d ( \\hat { \\mu } , S ) } & { \\triangleq } & { \\operatorname* { m a x } _ { v \\in \\mathbb { S } ^ { d - 1 } } \\left. v ^ { \\top } \\left( \\mu ( \\mathcal { M } ^ { v } ) - \\hat { \\mu } \\right) \\right. } \\end{array}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        411,
+                                        398,
+                                        417,
+                                        416
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        }
+                    ],
+                    "index": 19,
+                    "bbox_fs": [
+                        104,
+                        375,
+                        506,
+                        416
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        422,
+                        505,
+                        480
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                422,
+                                506,
+                                436
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        422,
+                                        218,
+                                        436
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Analysis. For any direction",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        219,
+                                        425,
+                                        225,
+                                        433
+                                    ],
+                                    "score": 0.74,
+                                    "content": "v",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        225,
+                                        422,
+                                        347,
+                                        436
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", the truncated mean estimator",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        347,
+                                        423,
+                                        378,
+                                        435
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\mu ( \\mathcal { M } ^ { v } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        378,
+                                        422,
+                                        506,
+                                        436
+                                    ],
+                                    "score": 1.0,
+                                    "content": "provides a robust estimation of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                433,
+                                505,
+                                446
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        433,
+                                        240,
+                                        446
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the true mean along the direction",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        240,
+                                        436,
+                                        246,
+                                        443
+                                    ],
+                                    "score": 0.71,
+                                    "content": "v",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        246,
+                                        433,
+                                        505,
+                                        446
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", thus the distance can be simply defined by taking the maximum",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                444,
+                                505,
+                                458
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        444,
+                                        182,
+                                        458
+                                    ],
+                                    "score": 1.0,
+                                    "content": "over all directions",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        182,
+                                        447,
+                                        188,
+                                        455
+                                    ],
+                                    "score": 0.7,
+                                    "content": "v",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        189,
+                                        444,
+                                        505,
+                                        458
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". We show the sensitivity of this simple estimator is bounded by the resilience",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                456,
+                                506,
+                                470
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        456,
+                                        143,
+                                        470
+                                    ],
+                                    "score": 1.0,
+                                    "content": "property",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        144,
+                                        459,
+                                        151,
+                                        467
+                                    ],
+                                    "score": 0.76,
+                                    "content": "\\sigma",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        151,
+                                        456,
+                                        198,
+                                        470
+                                    ],
+                                    "score": 1.0,
+                                    "content": "divided by",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        199,
+                                        459,
+                                        205,
+                                        467
+                                    ],
+                                    "score": 0.74,
+                                    "content": "n",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        206,
+                                        456,
+                                        247,
+                                        470
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", which is",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        248,
+                                        456,
+                                        334,
+                                        470
+                                    ],
+                                    "score": 0.93,
+                                    "content": "{ \\cal O } ( ( 1 / n ) \\sqrt { \\log ( 1 / \\alpha ) } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        334,
+                                        456,
+                                        506,
+                                        470
+                                    ],
+                                    "score": 1.0,
+                                    "content": "once the resilience check is passed. This",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                468,
+                                479,
+                                481
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        468,
+                                        479,
+                                        481
+                                    ],
+                                    "score": 1.0,
+                                    "content": "leads to the following near-optimal sample complexity. We provide a proof in Appendix H.2.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        }
+                    ],
+                    "index": 23,
+                    "bbox_fs": [
+                        105,
+                        422,
+                        506,
+                        481
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        482,
+                        506,
+                        508
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                482,
+                                506,
+                                495
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        482,
+                                        465,
+                                        495
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Theorem 7 (Exponential time algorithm for sub-Gaussian distributions). Algorithm 2 is",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        465,
+                                        482,
+                                        487,
+                                        495
+                                    ],
+                                    "score": 0.83,
+                                    "content": "( \\varepsilon , \\delta )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        488,
+                                        482,
+                                        506,
+                                        495
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-DP.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                493,
+                                506,
+                                509
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        493,
+                                        182,
+                                        509
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Under Assumption",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        182,
+                                        496,
+                                        187,
+                                        505
+                                    ],
+                                    "score": 0.48,
+                                    "content": "^ { l }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        187,
+                                        493,
+                                        285,
+                                        509
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", this algorithm achieves",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        285,
+                                        494,
+                                        406,
+                                        508
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\| \\hat { \\mu } - \\mu \\| _ { 2 } = O ( \\alpha \\sqrt { \\log ( 1 / \\alpha ) } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        406,
+                                        493,
+                                        472,
+                                        509
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with probability",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        473,
+                                        495,
+                                        506,
+                                        507
+                                    ],
+                                    "score": 0.87,
+                                    "content": "1 - \\zeta i f",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 27
+                        }
+                    ],
+                    "index": 26.5,
+                    "bbox_fs": [
+                        105,
+                        482,
+                        506,
+                        509
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        119,
+                        512,
+                        492,
+                        549
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                119,
+                                512,
+                                492,
+                                549
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        119,
+                                        512,
+                                        492,
+                                        549
+                                    ],
+                                    "score": 0.92,
+                                    "content": "n = \\widetilde \\Omega \\Big ( \\frac { d + \\log \\frac { 1 } { \\xi } } { \\alpha ^ { 2 } \\log \\frac { 1 } { \\alpha } } + \\frac { d \\log \\Big ( d \\sqrt { \\log ( d n / \\zeta ) } / \\alpha \\Big ) + d ^ { 1 / 2 } \\log \\frac { 1 } { \\delta } + \\log \\frac { 1 } { \\zeta } } { \\varepsilon \\alpha } + \\frac { \\sqrt { d \\log \\frac { 1 } { \\delta } } \\log \\frac { d } { \\zeta \\delta } } { \\varepsilon } \\Big ) ~ .",
+                                    "type": "interline_equation",
+                                    "image_path": "9aa30506fb06fc1067dfbb7018804b830bc0f54e20e433fb8eb13e7b665ff047.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 29,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                119,
+                                512,
+                                492,
+                                524.3333333333334
+                            ],
+                            "spans": [],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                119,
+                                524.3333333333334,
+                                492,
+                                536.6666666666667
+                            ],
+                            "spans": [],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                119,
+                                536.6666666666667,
+                                492,
+                                549.0000000000001
+                            ],
+                            "spans": [],
+                            "index": 30
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        558,
+                        506,
+                        595
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                558,
+                                506,
+                                573
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        558,
+                                        201,
+                                        573
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Run-time. Computing",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        201,
+                                        560,
+                                        224,
+                                        572
+                                    ],
+                                    "score": 0.92,
+                                    "content": "R ( S )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        225,
+                                        558,
+                                        294,
+                                        573
+                                    ],
+                                    "score": 1.0,
+                                    "content": "exactly can take",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        294,
+                                        558,
+                                        338,
+                                        572
+                                    ],
+                                    "score": 0.93,
+                                    "content": "O ( d e ^ { \\Theta ( n ) } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        338,
+                                        558,
+                                        506,
+                                        573
+                                    ],
+                                    "score": 1.0,
+                                    "content": "operations. The exponential mechanism",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                570,
+                                504,
+                                587
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        570,
+                                        180,
+                                        587
+                                    ],
+                                    "score": 1.0,
+                                    "content": "implemented with",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        181,
+                                        574,
+                                        188,
+                                        582
+                                    ],
+                                    "score": 0.77,
+                                    "content": "\\alpha",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        189,
+                                        570,
+                                        240,
+                                        587
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-covering for",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        241,
+                                        573,
+                                        248,
+                                        584
+                                    ],
+                                    "score": 0.85,
+                                    "content": "\\hat { \\mu }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        248,
+                                        570,
+                                        358,
+                                        587
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and a constant covering for",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        359,
+                                        575,
+                                        365,
+                                        582
+                                    ],
+                                    "score": 0.72,
+                                    "content": "v",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        366,
+                                        570,
+                                        401,
+                                        587
+                                    ],
+                                    "score": 1.0,
+                                    "content": "can take",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        401,
+                                        570,
+                                        504,
+                                        585
+                                    ],
+                                    "score": 0.92,
+                                    "content": "O ( n d ( \\sqrt { \\log ( d n / \\zeta ) } / \\alpha ) ^ { d } )",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                583,
+                                153,
+                                596
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        583,
+                                        153,
+                                        596
+                                    ],
+                                    "score": 1.0,
+                                    "content": "operations.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        }
+                    ],
+                    "index": 32,
+                    "bbox_fs": [
+                        104,
+                        558,
+                        506,
+                        596
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        106,
+                        609,
+                        183,
+                        623
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                607,
+                                185,
+                                626
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        607,
+                                        185,
+                                        626
+                                    ],
+                                    "score": 1.0,
+                                    "content": "5 Conclusion",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        }
+                    ],
+                    "index": 34
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        633,
+                        505,
+                        722
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                634,
+                                505,
+                                647
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        634,
+                                        505,
+                                        647
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Differentially private mean estimation is brittle against a small fraction of the samples being corrupted",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                645,
+                                505,
+                                657
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        645,
+                                        505,
+                                        657
+                                    ],
+                                    "score": 1.0,
+                                    "content": "by an adversary. We show that robustness can be achieved without any increase in the sample",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                656,
+                                505,
+                                668
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        656,
+                                        505,
+                                        668
+                                    ],
+                                    "score": 1.0,
+                                    "content": "complexity by introducing a novel DP mean estimator, which requires run-time exponential in the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                668,
+                                506,
+                                680
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        668,
+                                        506,
+                                        680
+                                    ],
+                                    "score": 1.0,
+                                    "content": "dimension of the samples. The technical contribution is in leveraging the resilience property of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                677,
+                                505,
+                                690
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        677,
+                                        505,
+                                        690
+                                    ],
+                                    "score": 1.0,
+                                    "content": "well-behaved distributions in an innovative way to not only find robust mean (which is the typical",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                688,
+                                506,
+                                702
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        688,
+                                        506,
+                                        702
+                                    ],
+                                    "score": 1.0,
+                                    "content": "use case of resilience) but also bound sensitivity for optimal privacy guarantee. To cope with the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                700,
+                                505,
+                                712
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        700,
+                                        505,
+                                        712
+                                    ],
+                                    "score": 1.0,
+                                    "content": "computational challenge, we propose an efficient algorithm, which we call PRIME, that achieves the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                711,
+                                506,
+                                723
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        711,
+                                        506,
+                                        723
+                                    ],
+                                    "score": 1.0,
+                                    "content": "optimal target accuracy at the cost of an increased sample complexity. The technical contributions are",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                73,
+                                505,
+                                86
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        73,
+                                        118,
+                                        84
+                                    ],
+                                    "score": 0.68,
+                                    "content": "( i )",
+                                    "type": "inline_equation",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        118,
+                                        73,
+                                        505,
+                                        86
+                                    ],
+                                    "score": 1.0,
+                                    "content": "a novel framework for private iterative filtering and its tight analysis of the end-to-end sensitivity",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                83,
+                                506,
+                                96
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        83,
+                                        124,
+                                        96
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        125,
+                                        84,
+                                        140,
+                                        95
+                                    ],
+                                    "score": 0.72,
+                                    "content": "( i i )",
+                                    "type": "inline_equation",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        140,
+                                        83,
+                                        506,
+                                        96
+                                    ],
+                                    "score": 1.0,
+                                    "content": "novel filtering algorithm of DPTHRESHOLD which is critical in privately running matrix",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                94,
+                                506,
+                                106
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        94,
+                                        506,
+                                        106
+                                    ],
+                                    "score": 1.0,
+                                    "content": "multiplicative weights and hence significantly reducing the number of accesses to the database. With",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                106,
+                                505,
+                                118
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        106,
+                                        505,
+                                        118
+                                    ],
+                                    "score": 1.0,
+                                    "content": "appropriately chosen parameters, we show that our exponential time approach achieves near-optimal",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                116,
+                                506,
+                                129
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        116,
+                                        506,
+                                        129
+                                    ],
+                                    "score": 1.0,
+                                    "content": "guarantees for both sub-Gaussian and covariance bounded distributions and PRIME achieves the",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                127,
+                                476,
+                                141
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        127,
+                                        428,
+                                        141
+                                    ],
+                                    "score": 1.0,
+                                    "content": "same accuracy efficiently but at the cost of an increased sample complexity by a",
+                                    "type": "text",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        428,
+                                        127,
+                                        447,
+                                        138
+                                    ],
+                                    "score": 0.89,
+                                    "content": "d ^ { 1 / 2 }",
+                                    "type": "inline_equation",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        447,
+                                        127,
+                                        476,
+                                        141
+                                    ],
+                                    "score": 1.0,
+                                    "content": "factor.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 5
+                        }
+                    ],
+                    "index": 38.5,
+                    "bbox_fs": [
+                        105,
+                        634,
+                        506,
+                        723
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        73,
+                        505,
+                        140
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                73,
+                                505,
+                                86
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        73,
+                                        118,
+                                        84
+                                    ],
+                                    "score": 0.68,
+                                    "content": "( i )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        118,
+                                        73,
+                                        505,
+                                        86
+                                    ],
+                                    "score": 1.0,
+                                    "content": "a novel framework for private iterative filtering and its tight analysis of the end-to-end sensitivity",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                83,
+                                506,
+                                96
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        83,
+                                        124,
+                                        96
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        125,
+                                        84,
+                                        140,
+                                        95
+                                    ],
+                                    "score": 0.72,
+                                    "content": "( i i )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        140,
+                                        83,
+                                        506,
+                                        96
+                                    ],
+                                    "score": 1.0,
+                                    "content": "novel filtering algorithm of DPTHRESHOLD which is critical in privately running matrix",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                94,
+                                506,
+                                106
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        94,
+                                        506,
+                                        106
+                                    ],
+                                    "score": 1.0,
+                                    "content": "multiplicative weights and hence significantly reducing the number of accesses to the database. With",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                106,
+                                505,
+                                118
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        106,
+                                        505,
+                                        118
+                                    ],
+                                    "score": 1.0,
+                                    "content": "appropriately chosen parameters, we show that our exponential time approach achieves near-optimal",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                116,
+                                506,
+                                129
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        116,
+                                        506,
+                                        129
+                                    ],
+                                    "score": 1.0,
+                                    "content": "guarantees for both sub-Gaussian and covariance bounded distributions and PRIME achieves the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                127,
+                                476,
+                                141
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        127,
+                                        428,
+                                        141
+                                    ],
+                                    "score": 1.0,
+                                    "content": "same accuracy efficiently but at the cost of an increased sample complexity by a",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        428,
+                                        127,
+                                        447,
+                                        138
+                                    ],
+                                    "score": 0.89,
+                                    "content": "d ^ { 1 / 2 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        447,
+                                        127,
+                                        476,
+                                        141
+                                    ],
+                                    "score": 1.0,
+                                    "content": "factor.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        }
+                    ],
+                    "index": 2.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        145,
+                        505,
+                        189
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                145,
+                                505,
+                                158
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        145,
+                                        505,
+                                        158
+                                    ],
+                                    "score": 1.0,
+                                    "content": "There are several directions for improving our results further and applying the framework to solve",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                155,
+                                506,
+                                168
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        155,
+                                        506,
+                                        168
+                                    ],
+                                    "score": 1.0,
+                                    "content": "other problems. PRIME provides a new design principle for private and robust estimation. This can",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                167,
+                                505,
+                                179
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        167,
+                                        505,
+                                        179
+                                    ],
+                                    "score": 1.0,
+                                    "content": "be more broadly applied to fundamental statistical analyses such as robust covariance estimation",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                177,
+                                391,
+                                190
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        177,
+                                        391,
+                                        190
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[28, 30, 64] robust PCA [60, 48], and robust linear regression [59, 35].",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        }
+                    ],
+                    "index": 7.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        195,
+                        506,
+                        400
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                194,
+                                506,
+                                210
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        194,
+                                        410,
+                                        210
+                                    ],
+                                    "score": 1.0,
+                                    "content": "PRIME could be improved in a few directions. First, the sample complexity of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        411,
+                                        194,
+                                        506,
+                                        208
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\widetilde \\Omega ( ( d / ( \\alpha ^ { 2 } \\log ( 1 / \\alpha ) ) ) +",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                205,
+                                504,
+                                221
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        207,
+                                        235,
+                                        220
+                                    ],
+                                    "score": 0.88,
+                                    "content": "( d ^ { 3 / 2 } / ( \\varepsilon \\alpha \\log ( 1 / \\alpha ) ) ) \\log ( 1 / \\delta ) )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        235,
+                                        205,
+                                        485,
+                                        221
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in Theorem 6 is suboptimal in the second term. Improving the",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        485,
+                                        208,
+                                        504,
+                                        218
+                                    ],
+                                    "score": 0.86,
+                                    "content": "d ^ { 3 / 2 }",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                218,
+                                506,
+                                232
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        218,
+                                        506,
+                                        232
+                                    ],
+                                    "score": 1.0,
+                                    "content": "factor requires bypassing differentially private singular value decomposition, which seems to be",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                230,
+                                505,
+                                243
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        230,
+                                        382,
+                                        243
+                                    ],
+                                    "score": 1.0,
+                                    "content": "a challenging task. However, it might be possible to separate the",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        383,
+                                        230,
+                                        419,
+                                        242
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\bar { \\log ( 1 / \\delta ) }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        420,
+                                        230,
+                                        505,
+                                        243
+                                    ],
+                                    "score": 1.0,
+                                    "content": "factor from the rest",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                241,
+                                507,
+                                257
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        241,
+                                        309,
+                                        257
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of the terms and get an additive error of the form",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        309,
+                                        241,
+                                        507,
+                                        255
+                                    ],
+                                    "score": 0.87,
+                                    "content": "\\widetilde \\Omega ( ( d / ( \\alpha ^ { 2 } \\log ( 1 / \\alpha ) ) ) + ( d ^ { 3 / 2 } / ( \\varepsilon \\alpha \\log ( 1 / \\alpha ) ) ) +",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                254,
+                                506,
+                                267
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        254,
+                                        168,
+                                        266
+                                    ],
+                                    "score": 0.91,
+                                    "content": "( 1 / \\varepsilon ) \\log ( 1 / \\delta ) )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        169,
+                                        254,
+                                        506,
+                                        267
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". This requires using Laplace mechanism in private MMW (line 16 Algortihm 10).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                264,
+                                506,
+                                278
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        264,
+                                        506,
+                                        278
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Secondly, the time complexity of PRIME is dominated by computation time of the matrix exponential",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                276,
+                                506,
+                                291
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        276,
+                                        372,
+                                        291
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in (line 16 Algortihm 10). Total number of operations scale as",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        373,
+                                        276,
+                                        428,
+                                        290
+                                    ],
+                                    "score": 0.93,
+                                    "content": "{ \\widetilde { O } } ( d ^ { 3 } + n d ^ { 2 } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        428,
+                                        276,
+                                        506,
+                                        291
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". One might hope",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                289,
+                                505,
+                                303
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        291,
+                                        150,
+                                        303
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to achieve",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        150,
+                                        289,
+                                        178,
+                                        303
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\widetilde O ( n d )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        178,
+                                        291,
+                                        394,
+                                        303
+                                    ],
+                                    "score": 1.0,
+                                    "content": "time complexity using approximate computations of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        394,
+                                        292,
+                                        404,
+                                        303
+                                    ],
+                                    "score": 0.86,
+                                    "content": "\\tau _ { j }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        404,
+                                        291,
+                                        505,
+                                        303
+                                    ],
+                                    "score": 1.0,
+                                    "content": "’s using techniques from",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                302,
+                                505,
+                                314
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        302,
+                                        505,
+                                        314
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[36]. This does not improve the sample complexity, as the number of times the dataset is accessed",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                313,
+                                505,
+                                326
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        313,
+                                        505,
+                                        326
+                                    ],
+                                    "score": 1.0,
+                                    "content": "remains the same. Finally, for (non-robust) private mean estimation, COINPRESS provides a practical",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                323,
+                                506,
+                                337
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        323,
+                                        506,
+                                        337
+                                    ],
+                                    "score": 1.0,
+                                    "content": "improvement in the small sample regime by progressively refining the search space [12]. The same",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                335,
+                                506,
+                                347
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        335,
+                                        506,
+                                        347
+                                    ],
+                                    "score": 1.0,
+                                    "content": "principle could be applied to PRIME to design a robust version of COINPRESS. One important",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                345,
+                                506,
+                                358
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        345,
+                                        506,
+                                        358
+                                    ],
+                                    "score": 1.0,
+                                    "content": "question remains open; how are differential privacy and robust statistics fundamentally related? We",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                357,
+                                506,
+                                369
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        357,
+                                        506,
+                                        369
+                                    ],
+                                    "score": 1.0,
+                                    "content": "believe our exponential time algorithm hints on a fundamental connection between robust statistics of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                367,
+                                505,
+                                380
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        367,
+                                        505,
+                                        380
+                                    ],
+                                    "score": 1.0,
+                                    "content": "a data projected onto one-dimensional subspace and sensitivity of resulting score function for the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                379,
+                                506,
+                                391
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        379,
+                                        506,
+                                        391
+                                    ],
+                                    "score": 1.0,
+                                    "content": "exponential mechanism. It is an interesting direction to pursue this connection further to design novel",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                390,
+                                293,
+                                402
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        390,
+                                        293,
+                                        402
+                                    ],
+                                    "score": 1.0,
+                                    "content": "algorithms that bridge privacy and robustness.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        }
+                    ],
+                    "index": 18.5
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        416,
+                        202,
+                        429
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                415,
+                                204,
+                                432
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        415,
+                                        204,
+                                        432
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Acknowledgement",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        }
+                    ],
+                    "index": 28
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        440,
+                        505,
+                        496
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                440,
+                                507,
+                                453
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        440,
+                                        507,
+                                        453
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Sham Kakade acknowledges funding from the National Science Foundation under award CCF-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                452,
+                                505,
+                                464
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        452,
+                                        505,
+                                        464
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1703574. Sewoong Oh acknowledges funding from Google faculty research award, NSF grants",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                462,
+                                505,
+                                474
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        462,
+                                        505,
+                                        474
+                                    ],
+                                    "score": 1.0,
+                                    "content": "IIS-1929955, CCF-1705007, CNS-2002664, CCF 2019844 as a part of Institute for Foundation of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                472,
+                                506,
+                                486
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        472,
+                                        506,
+                                        486
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Machine Learning, and CNS-2112471 as a part of Institute for Future Edge Networks and Distributed",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                483,
+                                159,
+                                498
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        483,
+                                        159,
+                                        498
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Intelligence.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        }
+                    ],
+                    "index": 31
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        511,
+                        163,
+                        523
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                510,
+                                165,
+                                525
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        510,
+                                        165,
+                                        525
+                                    ],
+                                    "score": 1.0,
+                                    "content": "References",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        }
+                    ],
+                    "index": 34
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        110,
+                        529,
+                        506,
+                        724
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                110,
+                                530,
+                                506,
+                                544
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        110,
+                                        530,
+                                        506,
+                                        544
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[1] Martin Abadi, Andy Chu, Ian Goodfellow, H Brendan McMahan, Ilya Mironov, Kunal Talwar,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                541,
+                                506,
+                                555
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        541,
+                                        506,
+                                        555
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and Li Zhang. Deep learning with differential privacy. In Proceedings of the 2016 ACM SIGSAC",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                552,
+                                446,
+                                566
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        552,
+                                        446,
+                                        566
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Conference on Computer and Communications Security, pages 308–318, 2016.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                110,
+                                570,
+                                506,
+                                585
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        110,
+                                        570,
+                                        506,
+                                        585
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[2] John M Abowd. The us census bureau adopts differential privacy. In Proceedings of the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                580,
+                                506,
+                                596
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        580,
+                                        506,
+                                        596
+                                    ],
+                                    "score": 1.0,
+                                    "content": "24th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining, pages",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                593,
+                                203,
+                                605
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        593,
+                                        203,
+                                        605
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2867–2867, 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                110,
+                                611,
+                                505,
+                                625
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        110,
+                                        611,
+                                        505,
+                                        625
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[3] Ishaq Aden-Ali, Hassan Ashtiani, and Gautam Kamath. On the sample complexity of privately",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                622,
+                                488,
+                                635
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        622,
+                                        488,
+                                        635
+                                    ],
+                                    "score": 1.0,
+                                    "content": "learning unbounded high-dimensional gaussians. arXiv preprint arXiv:2010.09929, 2020.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                110,
+                                640,
+                                506,
+                                654
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        110,
+                                        640,
+                                        506,
+                                        654
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[4] Zeyuan Allen-Zhu, Zhenyu Liao, and Lorenzo Orecchia. Spectral sparsification and regret",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                652,
+                                505,
+                                665
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        652,
+                                        505,
+                                        665
+                                    ],
+                                    "score": 1.0,
+                                    "content": "minimization beyond matrix multiplicative updates. In Proceedings of the forty-seventh annual",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                662,
+                                388,
+                                676
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        662,
+                                        388,
+                                        676
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ACM symposium on Theory of computing, pages 237–245, 2015.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                110,
+                                681,
+                                506,
+                                695
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        110,
+                                        681,
+                                        506,
+                                        695
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[5] Edoardo Amaldi and Viggo Kann. The complexity and approximability of finding maximum",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                692,
+                                507,
+                                705
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        692,
+                                        507,
+                                        705
+                                    ],
+                                    "score": 1.0,
+                                    "content": "feasible subsystems of linear relations. Theoretical computer science, 147(1-2):181–210, 1995.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 47
+                        },
+                        {
+                            "bbox": [
+                                110,
+                                711,
+                                442,
+                                723
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        110,
+                                        711,
+                                        442,
+                                        723
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[6] Frank J Anscombe. Rejection of outliers. Technometrics, 2(2):123–146, 1960.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 48
+                        }
+                    ],
+                    "index": 41.5
+                }
+            ],
+            "page_idx": 9,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        300,
+                        741,
+                        311,
+                        750
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                299,
+                                740,
+                                313,
+                                754
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        299,
+                                        740,
+                                        313,
+                                        754
+                                    ],
+                                    "score": 1.0,
+                                    "content": "10",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        73,
+                        505,
+                        140
+                    ],
+                    "lines": [],
+                    "index": 2.5,
+                    "bbox_fs": [
+                        105,
+                        73,
+                        506,
+                        141
+                    ],
+                    "lines_deleted": true
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        145,
+                        505,
+                        189
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                145,
+                                505,
+                                158
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        145,
+                                        505,
+                                        158
+                                    ],
+                                    "score": 1.0,
+                                    "content": "There are several directions for improving our results further and applying the framework to solve",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                155,
+                                506,
+                                168
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        155,
+                                        506,
+                                        168
+                                    ],
+                                    "score": 1.0,
+                                    "content": "other problems. PRIME provides a new design principle for private and robust estimation. This can",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                167,
+                                505,
+                                179
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        167,
+                                        505,
+                                        179
+                                    ],
+                                    "score": 1.0,
+                                    "content": "be more broadly applied to fundamental statistical analyses such as robust covariance estimation",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                177,
+                                391,
+                                190
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        177,
+                                        391,
+                                        190
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[28, 30, 64] robust PCA [60, 48], and robust linear regression [59, 35].",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        }
+                    ],
+                    "index": 7.5,
+                    "bbox_fs": [
+                        105,
+                        145,
+                        506,
+                        190
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        195,
+                        506,
+                        400
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                194,
+                                506,
+                                210
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        194,
+                                        410,
+                                        210
+                                    ],
+                                    "score": 1.0,
+                                    "content": "PRIME could be improved in a few directions. First, the sample complexity of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        411,
+                                        194,
+                                        506,
+                                        208
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\widetilde \\Omega ( ( d / ( \\alpha ^ { 2 } \\log ( 1 / \\alpha ) ) ) +",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                205,
+                                504,
+                                221
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        207,
+                                        235,
+                                        220
+                                    ],
+                                    "score": 0.88,
+                                    "content": "( d ^ { 3 / 2 } / ( \\varepsilon \\alpha \\log ( 1 / \\alpha ) ) ) \\log ( 1 / \\delta ) )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        235,
+                                        205,
+                                        485,
+                                        221
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in Theorem 6 is suboptimal in the second term. Improving the",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        485,
+                                        208,
+                                        504,
+                                        218
+                                    ],
+                                    "score": 0.86,
+                                    "content": "d ^ { 3 / 2 }",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                218,
+                                506,
+                                232
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        218,
+                                        506,
+                                        232
+                                    ],
+                                    "score": 1.0,
+                                    "content": "factor requires bypassing differentially private singular value decomposition, which seems to be",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                230,
+                                505,
+                                243
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        230,
+                                        382,
+                                        243
+                                    ],
+                                    "score": 1.0,
+                                    "content": "a challenging task. However, it might be possible to separate the",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        383,
+                                        230,
+                                        419,
+                                        242
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\bar { \\log ( 1 / \\delta ) }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        420,
+                                        230,
+                                        505,
+                                        243
+                                    ],
+                                    "score": 1.0,
+                                    "content": "factor from the rest",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                241,
+                                507,
+                                257
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        241,
+                                        309,
+                                        257
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of the terms and get an additive error of the form",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        309,
+                                        241,
+                                        507,
+                                        255
+                                    ],
+                                    "score": 0.87,
+                                    "content": "\\widetilde \\Omega ( ( d / ( \\alpha ^ { 2 } \\log ( 1 / \\alpha ) ) ) + ( d ^ { 3 / 2 } / ( \\varepsilon \\alpha \\log ( 1 / \\alpha ) ) ) +",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                254,
+                                506,
+                                267
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        254,
+                                        168,
+                                        266
+                                    ],
+                                    "score": 0.91,
+                                    "content": "( 1 / \\varepsilon ) \\log ( 1 / \\delta ) )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        169,
+                                        254,
+                                        506,
+                                        267
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". This requires using Laplace mechanism in private MMW (line 16 Algortihm 10).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                264,
+                                506,
+                                278
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        264,
+                                        506,
+                                        278
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Secondly, the time complexity of PRIME is dominated by computation time of the matrix exponential",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                276,
+                                506,
+                                291
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        276,
+                                        372,
+                                        291
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in (line 16 Algortihm 10). Total number of operations scale as",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        373,
+                                        276,
+                                        428,
+                                        290
+                                    ],
+                                    "score": 0.93,
+                                    "content": "{ \\widetilde { O } } ( d ^ { 3 } + n d ^ { 2 } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        428,
+                                        276,
+                                        506,
+                                        291
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". One might hope",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                289,
+                                505,
+                                303
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        291,
+                                        150,
+                                        303
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to achieve",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        150,
+                                        289,
+                                        178,
+                                        303
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\widetilde O ( n d )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        178,
+                                        291,
+                                        394,
+                                        303
+                                    ],
+                                    "score": 1.0,
+                                    "content": "time complexity using approximate computations of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        394,
+                                        292,
+                                        404,
+                                        303
+                                    ],
+                                    "score": 0.86,
+                                    "content": "\\tau _ { j }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        404,
+                                        291,
+                                        505,
+                                        303
+                                    ],
+                                    "score": 1.0,
+                                    "content": "’s using techniques from",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                302,
+                                505,
+                                314
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        302,
+                                        505,
+                                        314
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[36]. This does not improve the sample complexity, as the number of times the dataset is accessed",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                313,
+                                505,
+                                326
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        313,
+                                        505,
+                                        326
+                                    ],
+                                    "score": 1.0,
+                                    "content": "remains the same. Finally, for (non-robust) private mean estimation, COINPRESS provides a practical",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                323,
+                                506,
+                                337
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        323,
+                                        506,
+                                        337
+                                    ],
+                                    "score": 1.0,
+                                    "content": "improvement in the small sample regime by progressively refining the search space [12]. The same",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                335,
+                                506,
+                                347
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        335,
+                                        506,
+                                        347
+                                    ],
+                                    "score": 1.0,
+                                    "content": "principle could be applied to PRIME to design a robust version of COINPRESS. One important",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                345,
+                                506,
+                                358
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        345,
+                                        506,
+                                        358
+                                    ],
+                                    "score": 1.0,
+                                    "content": "question remains open; how are differential privacy and robust statistics fundamentally related? We",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                357,
+                                506,
+                                369
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        357,
+                                        506,
+                                        369
+                                    ],
+                                    "score": 1.0,
+                                    "content": "believe our exponential time algorithm hints on a fundamental connection between robust statistics of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                367,
+                                505,
+                                380
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        367,
+                                        505,
+                                        380
+                                    ],
+                                    "score": 1.0,
+                                    "content": "a data projected onto one-dimensional subspace and sensitivity of resulting score function for the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                379,
+                                506,
+                                391
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        379,
+                                        506,
+                                        391
+                                    ],
+                                    "score": 1.0,
+                                    "content": "exponential mechanism. It is an interesting direction to pursue this connection further to design novel",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                390,
+                                293,
+                                402
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        390,
+                                        293,
+                                        402
+                                    ],
+                                    "score": 1.0,
+                                    "content": "algorithms that bridge privacy and robustness.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        }
+                    ],
+                    "index": 18.5,
+                    "bbox_fs": [
+                        104,
+                        194,
+                        507,
+                        402
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        416,
+                        202,
+                        429
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                415,
+                                204,
+                                432
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        415,
+                                        204,
+                                        432
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Acknowledgement",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        }
+                    ],
+                    "index": 28
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        440,
+                        505,
+                        496
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                440,
+                                507,
+                                453
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        440,
+                                        507,
+                                        453
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Sham Kakade acknowledges funding from the National Science Foundation under award CCF-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                452,
+                                505,
+                                464
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        452,
+                                        505,
+                                        464
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1703574. Sewoong Oh acknowledges funding from Google faculty research award, NSF grants",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                462,
+                                505,
+                                474
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        462,
+                                        505,
+                                        474
+                                    ],
+                                    "score": 1.0,
+                                    "content": "IIS-1929955, CCF-1705007, CNS-2002664, CCF 2019844 as a part of Institute for Foundation of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                472,
+                                506,
+                                486
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        472,
+                                        506,
+                                        486
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Machine Learning, and CNS-2112471 as a part of Institute for Future Edge Networks and Distributed",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                483,
+                                159,
+                                498
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        483,
+                                        159,
+                                        498
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Intelligence.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        }
+                    ],
+                    "index": 31,
+                    "bbox_fs": [
+                        105,
+                        440,
+                        507,
+                        498
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        511,
+                        163,
+                        523
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                510,
+                                165,
+                                525
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        510,
+                                        165,
+                                        525
+                                    ],
+                                    "score": 1.0,
+                                    "content": "References",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        }
+                    ],
+                    "index": 34
+                },
+                {
+                    "type": "list",
+                    "bbox": [
+                        110,
+                        529,
+                        506,
+                        724
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                110,
+                                530,
+                                506,
+                                544
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        110,
+                                        530,
+                                        506,
+                                        544
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[1] Martin Abadi, Andy Chu, Ian Goodfellow, H Brendan McMahan, Ilya Mironov, Kunal Talwar,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                541,
+                                506,
+                                555
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        541,
+                                        506,
+                                        555
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and Li Zhang. Deep learning with differential privacy. In Proceedings of the 2016 ACM SIGSAC",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                552,
+                                446,
+                                566
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        552,
+                                        446,
+                                        566
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Conference on Computer and Communications Security, pages 308–318, 2016.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                110,
+                                570,
+                                506,
+                                585
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        110,
+                                        570,
+                                        506,
+                                        585
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[2] John M Abowd. The us census bureau adopts differential privacy. In Proceedings of the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                580,
+                                506,
+                                596
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        580,
+                                        506,
+                                        596
+                                    ],
+                                    "score": 1.0,
+                                    "content": "24th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining, pages",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                593,
+                                203,
+                                605
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        593,
+                                        203,
+                                        605
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2867–2867, 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                110,
+                                611,
+                                505,
+                                625
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        110,
+                                        611,
+                                        505,
+                                        625
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[3] Ishaq Aden-Ali, Hassan Ashtiani, and Gautam Kamath. On the sample complexity of privately",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                622,
+                                488,
+                                635
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        622,
+                                        488,
+                                        635
+                                    ],
+                                    "score": 1.0,
+                                    "content": "learning unbounded high-dimensional gaussians. arXiv preprint arXiv:2010.09929, 2020.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                110,
+                                640,
+                                506,
+                                654
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        110,
+                                        640,
+                                        506,
+                                        654
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[4] Zeyuan Allen-Zhu, Zhenyu Liao, and Lorenzo Orecchia. Spectral sparsification and regret",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                652,
+                                505,
+                                665
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        652,
+                                        505,
+                                        665
+                                    ],
+                                    "score": 1.0,
+                                    "content": "minimization beyond matrix multiplicative updates. In Proceedings of the forty-seventh annual",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                662,
+                                388,
+                                676
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        662,
+                                        388,
+                                        676
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ACM symposium on Theory of computing, pages 237–245, 2015.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                110,
+                                681,
+                                506,
+                                695
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        110,
+                                        681,
+                                        506,
+                                        695
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[5] Edoardo Amaldi and Viggo Kann. The complexity and approximability of finding maximum",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                692,
+                                507,
+                                705
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        692,
+                                        507,
+                                        705
+                                    ],
+                                    "score": 1.0,
+                                    "content": "feasible subsystems of linear relations. Theoretical computer science, 147(1-2):181–210, 1995.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 47
+                        },
+                        {
+                            "bbox": [
+                                110,
+                                711,
+                                442,
+                                723
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        110,
+                                        711,
+                                        442,
+                                        723
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[6] Frank J Anscombe. Rejection of outliers. Technometrics, 2(2):123–146, 1960.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 48,
+                            "is_list_start_line": true,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                110,
+                                72,
+                                507,
+                                86
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        110,
+                                        72,
+                                        507,
+                                        86
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[7] Ainesh Bakshi and Pravesh Kothari. List-decodable subspace recovery via sum-of-squares.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                83,
+                                291,
+                                97
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        83,
+                                        291,
+                                        97
+                                    ],
+                                    "score": 1.0,
+                                    "content": "arXiv preprint arXiv:2002.05139, 2020.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 1,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                109,
+                                101,
+                                506,
+                                116
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        109,
+                                        101,
+                                        506,
+                                        116
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[8] S. Balakrishnan, S. S. Du, J. Li, and A. Singh. Computationally efficient robust sparse estimation",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 2,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                113,
+                                507,
+                                127
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        113,
+                                        507,
+                                        127
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in high dimensions. In Proceedings of the 30th Conference on Learning Theory, COLT 2017,",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                125,
+                                124,
+                                219,
+                                136
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        125,
+                                        124,
+                                        219,
+                                        136
+                                    ],
+                                    "score": 1.0,
+                                    "content": "pages 169–212, 2017.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 4,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                109,
+                                142,
+                                506,
+                                156
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        109,
+                                        142,
+                                        506,
+                                        156
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[9] Amos Beimel, Shay Moran, Kobbi Nissim, and Uri Stemmer. Private center points and learning",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 5,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                153,
+                                348,
+                                167
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        153,
+                                        348,
+                                        167
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of halfspaces. arXiv preprint arXiv:1902.10731, 2019.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 6,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                173,
+                                505,
+                                185
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        173,
+                                        505,
+                                        185
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[10] K. Bhatia, P. Jain, P. Kamalaruban, and P. Kar. Consistent robust regression. In Advances",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 7,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                183,
+                                505,
+                                195
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        183,
+                                        505,
+                                        195
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in Neural Information Processing Systems 30: Annual Conference on Neural Information",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                194,
+                                333,
+                                206
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        194,
+                                        333,
+                                        206
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Processing Systems 2017, pages 2107–2116, 2017.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 9,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                212,
+                                506,
+                                226
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        212,
+                                        506,
+                                        226
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[11] Kush Bhatia, Prateek Jain, and Purushottam Kar. Robust regression via hard thresholding. In",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 10,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                222,
+                                432,
+                                236
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        222,
+                                        432,
+                                        236
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Advances in Neural Information Processing Systems, pages 721–729, 2015.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 11,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                241,
+                                506,
+                                255
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        241,
+                                        506,
+                                        255
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[12] Sourav Biswas, Yihe Dong, Gautam Kamath, and Jonathan Ullman. Coinpress: Practical private",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 12,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                252,
+                                424,
+                                266
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        252,
+                                        424,
+                                        266
+                                    ],
+                                    "score": 1.0,
+                                    "content": "mean and covariance estimation. arXiv preprint arXiv:2006.06618, 2020.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 13,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                103,
+                                269,
+                                507,
+                                285
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        103,
+                                        269,
+                                        507,
+                                        285
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[13] Avrim Blum, Cynthia Dwork, Frank McSherry, and Kobbi Nissim. Practical privacy: the sulq",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 14,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                281,
+                                506,
+                                295
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        281,
+                                        506,
+                                        295
+                                    ],
+                                    "score": 1.0,
+                                    "content": "framework. In Proceedings of the twenty-fourth ACM SIGMOD-SIGACT-SIGART symposium",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                293,
+                                358,
+                                306
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        293,
+                                        358,
+                                        306
+                                    ],
+                                    "score": 1.0,
+                                    "content": "on Principles of database systems, pages 128–138, 2005.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 16,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                312,
+                                506,
+                                324
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        312,
+                                        506,
+                                        324
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[14] Mark Bun, Gautam Kamath, Thomas Steinke, and Steven Z Wu. Private hypothesis selection.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 17,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                322,
+                                442,
+                                336
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        322,
+                                        442,
+                                        336
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In Advances in Neural Information Processing Systems, pages 156–167, 2019.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 18,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                340,
+                                506,
+                                354
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        340,
+                                        506,
+                                        354
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[15] T Tony Cai, Yichen Wang, and Linjun Zhang. The cost of privacy: Optimal rates of convergence",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 19,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                352,
+                                492,
+                                366
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        352,
+                                        492,
+                                        366
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for parameter estimation with differential privacy. arXiv preprint arXiv:1902.04495, 2019.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 20,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                370,
+                                506,
+                                382
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        370,
+                                        506,
+                                        382
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[16] Clément L Canonne, Gautam Kamath, Audra McMillan, Jonathan Ullman, and Lydia Za-",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 21,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                381,
+                                507,
+                                395
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        381,
+                                        507,
+                                        395
+                                    ],
+                                    "score": 1.0,
+                                    "content": "kynthinou. Private identity testing for high-dimensional distributions. arXiv preprint",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                392,
+                                230,
+                                404
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        392,
+                                        230,
+                                        404
+                                    ],
+                                    "score": 1.0,
+                                    "content": "arXiv:1905.11947, 2019.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 23,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                411,
+                                505,
+                                423
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        411,
+                                        505,
+                                        423
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[17] Moses Charikar, Jacob Steinhardt, and Gregory Valiant. Learning from untrusted data. In",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 24,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                420,
+                                506,
+                                435
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        420,
+                                        506,
+                                        435
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Proceedings of the 49th Annual ACM SIGACT Symposium on Theory of Computing, pages",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                432,
+                                184,
+                                446
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        432,
+                                        184,
+                                        446
+                                    ],
+                                    "score": 1.0,
+                                    "content": "47–60, 2017.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 26,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                450,
+                                506,
+                                465
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        450,
+                                        506,
+                                        465
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[18] Kamalika Chaudhuri, Anand D Sarwate, and Kaushik Sinha. A near-optimal algorithm for",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 27,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                461,
+                                507,
+                                475
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        461,
+                                        507,
+                                        475
+                                    ],
+                                    "score": 1.0,
+                                    "content": "differentially-private principal components. The Journal of Machine Learning Research,",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                473,
+                                227,
+                                485
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        473,
+                                        227,
+                                        485
+                                    ],
+                                    "score": 1.0,
+                                    "content": "14(1):2905–2943, 2013.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 29,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                492,
+                                505,
+                                504
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        492,
+                                        505,
+                                        504
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[19] Xinyun Chen, Chang Liu, Bo Li, Kimberly Lu, and Dawn Song. Targeted backdoor attacks on",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 30,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                501,
+                                470,
+                                515
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        501,
+                                        470,
+                                        515
+                                    ],
+                                    "score": 1.0,
+                                    "content": "deep learning systems using data poisoning. arXiv preprint arXiv:1712.05526, 2017.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 31,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                520,
+                                506,
+                                534
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        520,
+                                        506,
+                                        534
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[20] Yu Cheng, Ilias Diakonikolas, and Rong Ge. High-dimensional robust mean estimation in",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 32,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                128,
+                                531,
+                                505,
+                                543
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        128,
+                                        531,
+                                        505,
+                                        543
+                                    ],
+                                    "score": 1.0,
+                                    "content": "nearly-linear time. In Proceedings of the Thirtieth Annual ACM-SIAM Symposium on Discrete",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                542,
+                                307,
+                                556
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        542,
+                                        307,
+                                        556
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Algorithms, pages 2755–2771. SIAM, 2019.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 34,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                561,
+                                506,
+                                573
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        561,
+                                        506,
+                                        573
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[21] Yu Cheng, Ilias Diakonikolas, Rong Ge, and David P Woodruff. Faster algorithms for high-",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 35,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                571,
+                                507,
+                                585
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        571,
+                                        507,
+                                        585
+                                    ],
+                                    "score": 1.0,
+                                    "content": "dimensional robust covariance estimation. In Conference on Learning Theory, pages 727–757.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                581,
+                                187,
+                                596
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        581,
+                                        187,
+                                        596
+                                    ],
+                                    "score": 1.0,
+                                    "content": "PMLR, 2019.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 37,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                602,
+                                505,
+                                614
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        602,
+                                        505,
+                                        614
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[22] Yeshwanth Cherapanamjeri, Sidhanth Mohanty, and Morris Yau. List decodable mean estimation",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 38,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                612,
+                                377,
+                                624
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        612,
+                                        377,
+                                        624
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in nearly linear time. arXiv preprint arXiv:2005.09796, 2020.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 39,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                628,
+                                507,
+                                644
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        628,
+                                        507,
+                                        644
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[23] Arnak Dalalyan and Philip Thompson. Outlier-robust estimation of a sparse linear model using",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 40,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                128,
+                                641,
+                                507,
+                                655
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        128,
+                                        641,
+                                        138,
+                                        652
+                                    ],
+                                    "score": 0.86,
+                                    "content": "\\ell _ { 1 }",
+                                    "type": "inline_equation",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        138,
+                                        641,
+                                        215,
+                                        655
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-penalized huber’s",
+                                    "type": "text",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        216,
+                                        642,
+                                        225,
+                                        651
+                                    ],
+                                    "score": 0.72,
+                                    "content": "m",
+                                    "type": "inline_equation",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        226,
+                                        641,
+                                        507,
+                                        655
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-estimator. In Advances in Neural Information Processing Systems,",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                652,
+                                239,
+                                664
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        652,
+                                        239,
+                                        664
+                                    ],
+                                    "score": 1.0,
+                                    "content": "pages 13188–13198, 2019.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 42,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                670,
+                                505,
+                                684
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        670,
+                                        505,
+                                        684
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[24] Jules Depersin and Guillaume Lecué. Robust subgaussian estimation of a mean vector in nearly",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 43,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                682,
+                                339,
+                                694
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        682,
+                                        339,
+                                        694
+                                    ],
+                                    "score": 1.0,
+                                    "content": "linear time. arXiv preprint arXiv:1906.03058, 2019.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 44,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                700,
+                                505,
+                                712
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        700,
+                                        505,
+                                        712
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[25] Luc Devroye and Gábor Lugosi. Combinatorial methods in density estimation. Springer Science",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 45,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                711,
+                                231,
+                                723
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        711,
+                                        231,
+                                        723
+                                    ],
+                                    "score": 1.0,
+                                    "content": "& Business Media, 2012.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 46,
+                            "is_list_end_line": true
+                        }
+                    ],
+                    "index": 41.5,
+                    "bbox_fs": [
+                        110,
+                        530,
+                        507,
+                        723
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        45,
+                        507,
+                        727
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                110,
+                                72,
+                                507,
+                                86
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        110,
+                                        72,
+                                        507,
+                                        86
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[7] Ainesh Bakshi and Pravesh Kothari. List-decodable subspace recovery via sum-of-squares.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                83,
+                                291,
+                                97
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        83,
+                                        291,
+                                        97
+                                    ],
+                                    "score": 1.0,
+                                    "content": "arXiv preprint arXiv:2002.05139, 2020.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                109,
+                                101,
+                                506,
+                                116
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        109,
+                                        101,
+                                        506,
+                                        116
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[8] S. Balakrishnan, S. S. Du, J. Li, and A. Singh. Computationally efficient robust sparse estimation",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                113,
+                                507,
+                                127
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        113,
+                                        507,
+                                        127
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in high dimensions. In Proceedings of the 30th Conference on Learning Theory, COLT 2017,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                125,
+                                124,
+                                219,
+                                136
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        125,
+                                        124,
+                                        219,
+                                        136
+                                    ],
+                                    "score": 1.0,
+                                    "content": "pages 169–212, 2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                109,
+                                142,
+                                506,
+                                156
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        109,
+                                        142,
+                                        506,
+                                        156
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[9] Amos Beimel, Shay Moran, Kobbi Nissim, and Uri Stemmer. Private center points and learning",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                153,
+                                348,
+                                167
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        153,
+                                        348,
+                                        167
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of halfspaces. arXiv preprint arXiv:1902.10731, 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                173,
+                                505,
+                                185
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        173,
+                                        505,
+                                        185
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[10] K. Bhatia, P. Jain, P. Kamalaruban, and P. Kar. Consistent robust regression. In Advances",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                183,
+                                505,
+                                195
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        183,
+                                        505,
+                                        195
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in Neural Information Processing Systems 30: Annual Conference on Neural Information",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                194,
+                                333,
+                                206
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        194,
+                                        333,
+                                        206
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Processing Systems 2017, pages 2107–2116, 2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                212,
+                                506,
+                                226
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        212,
+                                        506,
+                                        226
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[11] Kush Bhatia, Prateek Jain, and Purushottam Kar. Robust regression via hard thresholding. In",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                222,
+                                432,
+                                236
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        222,
+                                        432,
+                                        236
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Advances in Neural Information Processing Systems, pages 721–729, 2015.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                241,
+                                506,
+                                255
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        241,
+                                        506,
+                                        255
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[12] Sourav Biswas, Yihe Dong, Gautam Kamath, and Jonathan Ullman. Coinpress: Practical private",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                252,
+                                424,
+                                266
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        252,
+                                        424,
+                                        266
+                                    ],
+                                    "score": 1.0,
+                                    "content": "mean and covariance estimation. arXiv preprint arXiv:2006.06618, 2020.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                103,
+                                269,
+                                507,
+                                285
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        103,
+                                        269,
+                                        507,
+                                        285
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[13] Avrim Blum, Cynthia Dwork, Frank McSherry, and Kobbi Nissim. Practical privacy: the sulq",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                281,
+                                506,
+                                295
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        281,
+                                        506,
+                                        295
+                                    ],
+                                    "score": 1.0,
+                                    "content": "framework. In Proceedings of the twenty-fourth ACM SIGMOD-SIGACT-SIGART symposium",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                293,
+                                358,
+                                306
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        293,
+                                        358,
+                                        306
+                                    ],
+                                    "score": 1.0,
+                                    "content": "on Principles of database systems, pages 128–138, 2005.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                312,
+                                506,
+                                324
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        312,
+                                        506,
+                                        324
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[14] Mark Bun, Gautam Kamath, Thomas Steinke, and Steven Z Wu. Private hypothesis selection.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                322,
+                                442,
+                                336
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        322,
+                                        442,
+                                        336
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In Advances in Neural Information Processing Systems, pages 156–167, 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                340,
+                                506,
+                                354
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        340,
+                                        506,
+                                        354
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[15] T Tony Cai, Yichen Wang, and Linjun Zhang. The cost of privacy: Optimal rates of convergence",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                352,
+                                492,
+                                366
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        352,
+                                        492,
+                                        366
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for parameter estimation with differential privacy. arXiv preprint arXiv:1902.04495, 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                370,
+                                506,
+                                382
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        370,
+                                        506,
+                                        382
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[16] Clément L Canonne, Gautam Kamath, Audra McMillan, Jonathan Ullman, and Lydia Za-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                381,
+                                507,
+                                395
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        381,
+                                        507,
+                                        395
+                                    ],
+                                    "score": 1.0,
+                                    "content": "kynthinou. Private identity testing for high-dimensional distributions. arXiv preprint",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                392,
+                                230,
+                                404
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        392,
+                                        230,
+                                        404
+                                    ],
+                                    "score": 1.0,
+                                    "content": "arXiv:1905.11947, 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                411,
+                                505,
+                                423
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        411,
+                                        505,
+                                        423
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[17] Moses Charikar, Jacob Steinhardt, and Gregory Valiant. Learning from untrusted data. In",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                420,
+                                506,
+                                435
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        420,
+                                        506,
+                                        435
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Proceedings of the 49th Annual ACM SIGACT Symposium on Theory of Computing, pages",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                432,
+                                184,
+                                446
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        432,
+                                        184,
+                                        446
+                                    ],
+                                    "score": 1.0,
+                                    "content": "47–60, 2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                450,
+                                506,
+                                465
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        450,
+                                        506,
+                                        465
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[18] Kamalika Chaudhuri, Anand D Sarwate, and Kaushik Sinha. A near-optimal algorithm for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                461,
+                                507,
+                                475
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        461,
+                                        507,
+                                        475
+                                    ],
+                                    "score": 1.0,
+                                    "content": "differentially-private principal components. The Journal of Machine Learning Research,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                473,
+                                227,
+                                485
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        473,
+                                        227,
+                                        485
+                                    ],
+                                    "score": 1.0,
+                                    "content": "14(1):2905–2943, 2013.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                492,
+                                505,
+                                504
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        492,
+                                        505,
+                                        504
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[19] Xinyun Chen, Chang Liu, Bo Li, Kimberly Lu, and Dawn Song. Targeted backdoor attacks on",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                501,
+                                470,
+                                515
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        501,
+                                        470,
+                                        515
+                                    ],
+                                    "score": 1.0,
+                                    "content": "deep learning systems using data poisoning. arXiv preprint arXiv:1712.05526, 2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                520,
+                                506,
+                                534
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        520,
+                                        506,
+                                        534
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[20] Yu Cheng, Ilias Diakonikolas, and Rong Ge. High-dimensional robust mean estimation in",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                128,
+                                531,
+                                505,
+                                543
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        128,
+                                        531,
+                                        505,
+                                        543
+                                    ],
+                                    "score": 1.0,
+                                    "content": "nearly-linear time. In Proceedings of the Thirtieth Annual ACM-SIAM Symposium on Discrete",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                542,
+                                307,
+                                556
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        542,
+                                        307,
+                                        556
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Algorithms, pages 2755–2771. SIAM, 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                561,
+                                506,
+                                573
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        561,
+                                        506,
+                                        573
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[21] Yu Cheng, Ilias Diakonikolas, Rong Ge, and David P Woodruff. Faster algorithms for high-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                571,
+                                507,
+                                585
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        571,
+                                        507,
+                                        585
+                                    ],
+                                    "score": 1.0,
+                                    "content": "dimensional robust covariance estimation. In Conference on Learning Theory, pages 727–757.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                581,
+                                187,
+                                596
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        581,
+                                        187,
+                                        596
+                                    ],
+                                    "score": 1.0,
+                                    "content": "PMLR, 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                602,
+                                505,
+                                614
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        602,
+                                        505,
+                                        614
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[22] Yeshwanth Cherapanamjeri, Sidhanth Mohanty, and Morris Yau. List decodable mean estimation",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                612,
+                                377,
+                                624
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        612,
+                                        377,
+                                        624
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in nearly linear time. arXiv preprint arXiv:2005.09796, 2020.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                628,
+                                507,
+                                644
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        628,
+                                        507,
+                                        644
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[23] Arnak Dalalyan and Philip Thompson. Outlier-robust estimation of a sparse linear model using",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                128,
+                                641,
+                                507,
+                                655
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        128,
+                                        641,
+                                        138,
+                                        652
+                                    ],
+                                    "score": 0.86,
+                                    "content": "\\ell _ { 1 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        138,
+                                        641,
+                                        215,
+                                        655
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-penalized huber’s",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        216,
+                                        642,
+                                        225,
+                                        651
+                                    ],
+                                    "score": 0.72,
+                                    "content": "m",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        226,
+                                        641,
+                                        507,
+                                        655
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-estimator. In Advances in Neural Information Processing Systems,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                652,
+                                239,
+                                664
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        652,
+                                        239,
+                                        664
+                                    ],
+                                    "score": 1.0,
+                                    "content": "pages 13188–13198, 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                670,
+                                505,
+                                684
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        670,
+                                        505,
+                                        684
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[24] Jules Depersin and Guillaume Lecué. Robust subgaussian estimation of a mean vector in nearly",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                682,
+                                339,
+                                694
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        682,
+                                        339,
+                                        694
+                                    ],
+                                    "score": 1.0,
+                                    "content": "linear time. arXiv preprint arXiv:1906.03058, 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                700,
+                                505,
+                                712
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        700,
+                                        505,
+                                        712
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[25] Luc Devroye and Gábor Lugosi. Combinatorial methods in density estimation. Springer Science",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                711,
+                                231,
+                                723
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        711,
+                                        231,
+                                        723
+                                    ],
+                                    "score": 1.0,
+                                    "content": "& Business Media, 2012.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        }
+                    ],
+                    "index": 23
+                }
+            ],
+            "page_idx": 10,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        300,
+                        741,
+                        310,
+                        750
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                299,
+                                740,
+                                312,
+                                755
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        299,
+                                        740,
+                                        312,
+                                        755
+                                    ],
+                                    "score": 1.0,
+                                    "content": "",
+                                    "type": "text",
+                                    "height": 15,
+                                    "width": 13
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "list",
+                    "bbox": [
+                        105,
+                        45,
+                        507,
+                        727
+                    ],
+                    "lines": [],
+                    "index": 23,
+                    "bbox_fs": [
+                        103,
+                        72,
+                        507,
+                        723
+                    ],
+                    "lines_deleted": true
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        72,
+                        505,
+                        95
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                72,
+                                506,
+                                85
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        72,
+                                        506,
+                                        85
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[26] Aditya Dhar and Jason Huang. Designing differentially private estimators in high dimensions.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                83,
+                                290,
+                                95
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        83,
+                                        290,
+                                        95
+                                    ],
+                                    "score": 1.0,
+                                    "content": "arXiv preprint arXiv:2006.01944, 2020.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        }
+                    ],
+                    "index": 0.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        102,
+                        504,
+                        126
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                101,
+                                505,
+                                117
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        101,
+                                        505,
+                                        117
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[27] Ilias Diakonikolas, Samuel B Hopkins, Daniel Kane, and Sushrut Karmalkar. Robustly learning",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                114,
+                                441,
+                                126
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        114,
+                                        441,
+                                        126
+                                    ],
+                                    "score": 1.0,
+                                    "content": "any clusterable mixture of gaussians. arXiv preprint arXiv:2005.06417, 2020.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        }
+                    ],
+                    "index": 2.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        132,
+                        505,
+                        167
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                132,
+                                506,
+                                146
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        132,
+                                        506,
+                                        146
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[28] Ilias Diakonikolas, Gautam Kamath, Daniel Kane, Jerry Li, Ankur Moitra, and Alistair Stewart.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                144,
+                                505,
+                                156
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        144,
+                                        505,
+                                        156
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Robust estimators in high-dimensions without the computational intractability. SIAM Journal",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                155,
+                                279,
+                                167
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        155,
+                                        279,
+                                        167
+                                    ],
+                                    "score": 1.0,
+                                    "content": "on Computing, 48(2):742–864, 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        }
+                    ],
+                    "index": 5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        173,
+                        504,
+                        208
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                173,
+                                506,
+                                187
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        173,
+                                        506,
+                                        187
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[29] Ilias Diakonikolas, Gautam Kamath, Daniel Kane, Jerry Li, Jacob Steinhardt, and Alistair Stew-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                185,
+                                505,
+                                198
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        185,
+                                        505,
+                                        198
+                                    ],
+                                    "score": 1.0,
+                                    "content": "art. Sever: A robust meta-algorithm for stochastic optimization. In International Conference on",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                195,
+                                307,
+                                209
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        195,
+                                        307,
+                                        209
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Machine Learning, pages 1596–1606, 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        }
+                    ],
+                    "index": 8
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        215,
+                        505,
+                        249
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                214,
+                                506,
+                                227
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        214,
+                                        506,
+                                        227
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[30] Ilias Diakonikolas, Gautam Kamath, Daniel M. Kane, Jerry Li, Ankur Moitra, and Alis-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                225,
+                                506,
+                                240
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        225,
+                                        506,
+                                        240
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tair Stewart. Being Robust (in High Dimensions) Can Be Practical. arXiv e-prints, page",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                128,
+                                236,
+                                259,
+                                248
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        128,
+                                        236,
+                                        259,
+                                        248
+                                    ],
+                                    "score": 1.0,
+                                    "content": "arXiv:1703.00893, March 2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        }
+                    ],
+                    "index": 11
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        255,
+                        505,
+                        300
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                255,
+                                506,
+                                269
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        255,
+                                        506,
+                                        269
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[31] Ilias Diakonikolas, Gautam Kamath, Daniel M Kane, Jerry Li, Ankur Moitra, and Alistair",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                267,
+                                506,
+                                280
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        267,
+                                        506,
+                                        280
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Stewart. Robustly learning a gaussian: Getting optimal error, efficiently. In Proceedings of the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                277,
+                                507,
+                                291
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        277,
+                                        507,
+                                        291
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Twenty-Ninth Annual ACM-SIAM Symposium on Discrete Algorithms, pages 2683–2702. SIAM,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                288,
+                                155,
+                                301
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        288,
+                                        155,
+                                        301
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        }
+                    ],
+                    "index": 14.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        307,
+                        504,
+                        331
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                307,
+                                506,
+                                322
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        307,
+                                        506,
+                                        322
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[32] Ilias Diakonikolas and Daniel M Kane. Recent advances in algorithmic high-dimensional robust",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                320,
+                                330,
+                                331
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        320,
+                                        330,
+                                        331
+                                    ],
+                                    "score": 1.0,
+                                    "content": "statistics. arXiv preprint arXiv:1911.05911, 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        }
+                    ],
+                    "index": 17.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        338,
+                        504,
+                        372
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                338,
+                                505,
+                                351
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        338,
+                                        505,
+                                        351
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[33] Ilias Diakonikolas, Daniel M Kane, and Alistair Stewart. Statistical query lower bounds for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                349,
+                                505,
+                                362
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        349,
+                                        505,
+                                        362
+                                    ],
+                                    "score": 1.0,
+                                    "content": "robust estimation of high-dimensional gaussians and gaussian mixtures. In 2017 IEEE 58th",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                360,
+                                500,
+                                374
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        360,
+                                        500,
+                                        374
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Annual Symposium on Foundations of Computer Science (FOCS), pages 73–84. IEEE, 2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        }
+                    ],
+                    "index": 20
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        379,
+                        504,
+                        414
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                379,
+                                506,
+                                392
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        379,
+                                        506,
+                                        392
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[34] Ilias Diakonikolas, Daniel M Kane, and Alistair Stewart. List-decodable robust mean estimation",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                391,
+                                506,
+                                403
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        391,
+                                        506,
+                                        403
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and learning mixtures of spherical gaussians. In Proceedings of the 50th Annual ACM SIGACT",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                128,
+                                401,
+                                379,
+                                415
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        128,
+                                        401,
+                                        379,
+                                        415
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Symposium on Theory of Computing, pages 1047–1060, 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        }
+                    ],
+                    "index": 23
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        420,
+                        504,
+                        454
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                419,
+                                506,
+                                434
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        419,
+                                        506,
+                                        434
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[35] Ilias Diakonikolas, Weihao Kong, and Alistair Stewart. Efficient algorithms and lower bounds",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                432,
+                                505,
+                                444
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        432,
+                                        505,
+                                        444
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for robust linear regression. In Proceedings of the Thirtieth Annual ACM-SIAM Symposium on",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                128,
+                                443,
+                                343,
+                                455
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        128,
+                                        443,
+                                        343,
+                                        455
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Discrete Algorithms, pages 2745–2754. SIAM, 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        }
+                    ],
+                    "index": 26
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        461,
+                        504,
+                        495
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                461,
+                                505,
+                                475
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        461,
+                                        505,
+                                        475
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[36] Yihe Dong, Samuel Hopkins, and Jerry Li. Quantum entropy scoring for fast robust mean",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                471,
+                                505,
+                                487
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        471,
+                                        505,
+                                        487
+                                    ],
+                                    "score": 1.0,
+                                    "content": "estimation and improved outlier detection. In Advances in Neural Information Processing",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                484,
+                                264,
+                                496
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        484,
+                                        264,
+                                        496
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Systems, pages 6067–6077, 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        }
+                    ],
+                    "index": 29
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        502,
+                        505,
+                        536
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                502,
+                                506,
+                                516
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        502,
+                                        506,
+                                        516
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[37] Cynthia Dwork, Frank McSherry, Kobbi Nissim, and Adam Smith. Calibrating noise to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                513,
+                                508,
+                                528
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        513,
+                                        508,
+                                        528
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sensitivity in private data analysis. In Theory of cryptography conference, pages 265–284.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                524,
+                                192,
+                                537
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        524,
+                                        192,
+                                        537
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Springer, 2006.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        }
+                    ],
+                    "index": 32
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        543,
+                        505,
+                        567
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                544,
+                                506,
+                                557
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        544,
+                                        506,
+                                        557
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[38] Cynthia Dwork and Aaron Roth. The algorithmic foundations of differential privacy. Founda-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                555,
+                                425,
+                                567
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        555,
+                                        425,
+                                        567
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tions and Trends in Theoretical Computer Science, 9(3-4):211–407, 2014.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        }
+                    ],
+                    "index": 34.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        574,
+                        505,
+                        608
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                574,
+                                505,
+                                587
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        574,
+                                        505,
+                                        587
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[39] Cynthia Dwork, Kunal Talwar, Abhradeep Thakurta, and Li Zhang. Analyze gauss: optimal",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                585,
+                                505,
+                                598
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        585,
+                                        505,
+                                        598
+                                    ],
+                                    "score": 1.0,
+                                    "content": "bounds for privacy-preserving principal component analysis. In Proceedings of the forty-sixth",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                596,
+                                409,
+                                609
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        596,
+                                        409,
+                                        609
+                                    ],
+                                    "score": 1.0,
+                                    "content": "annual ACM symposium on Theory of computing, pages 11–20, 2014.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        }
+                    ],
+                    "index": 37
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        615,
+                        504,
+                        650
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                615,
+                                506,
+                                630
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        615,
+                                        506,
+                                        630
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[40] Úlfar Erlingsson, Vasyl Pihur, and Aleksandra Korolova. Rappor: Randomized aggregatable",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                627,
+                                506,
+                                640
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        627,
+                                        506,
+                                        640
+                                    ],
+                                    "score": 1.0,
+                                    "content": "privacy-preserving ordinal response. In Proceedings of the 2014 ACM SIGSAC conference on",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                128,
+                                638,
+                                390,
+                                651
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        128,
+                                        638,
+                                        390,
+                                        651
+                                    ],
+                                    "score": 1.0,
+                                    "content": "computer and communications security, pages 1054–1067, 2014.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        }
+                    ],
+                    "index": 40
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        658,
+                        503,
+                        692
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                657,
+                                505,
+                                673
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        657,
+                                        505,
+                                        673
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[41] Giulia Fanti, Vasyl Pihur, and Úlfar Erlingsson. Building a rappor with the unknown: Privacy-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                669,
+                                506,
+                                684
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        669,
+                                        506,
+                                        684
+                                    ],
+                                    "score": 1.0,
+                                    "content": "preserving learning of associations and data dictionaries. Proceedings on Privacy Enhancing",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                680,
+                                273,
+                                693
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        680,
+                                        273,
+                                        693
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Technologies, 2016(3):41–61, 2016.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        }
+                    ],
+                    "index": 43
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        699,
+                        503,
+                        722
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                698,
+                                505,
+                                713
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        698,
+                                        505,
+                                        713
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[42] Chao Gao et al. Robust regression via mutivariate regression depth. Bernoulli, 26(2):1139–1170,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                708,
+                                154,
+                                723
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        708,
+                                        154,
+                                        723
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2020.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        }
+                    ],
+                    "index": 45.5
+                }
+            ],
+            "page_idx": 11,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        300,
+                        741,
+                        311,
+                        750
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                299,
+                                740,
+                                313,
+                                754
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        299,
+                                        740,
+                                        313,
+                                        754
+                                    ],
+                                    "score": 1.0,
+                                    "content": "12",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        72,
+                        505,
+                        95
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                72,
+                                506,
+                                85
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        72,
+                                        506,
+                                        85
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[26] Aditya Dhar and Jason Huang. Designing differentially private estimators in high dimensions.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                83,
+                                290,
+                                95
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        83,
+                                        290,
+                                        95
+                                    ],
+                                    "score": 1.0,
+                                    "content": "arXiv preprint arXiv:2006.01944, 2020.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        }
+                    ],
+                    "index": 0.5,
+                    "bbox_fs": [
+                        106,
+                        72,
+                        506,
+                        95
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        102,
+                        504,
+                        126
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                101,
+                                505,
+                                117
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        101,
+                                        505,
+                                        117
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[27] Ilias Diakonikolas, Samuel B Hopkins, Daniel Kane, and Sushrut Karmalkar. Robustly learning",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                114,
+                                441,
+                                126
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        114,
+                                        441,
+                                        126
+                                    ],
+                                    "score": 1.0,
+                                    "content": "any clusterable mixture of gaussians. arXiv preprint arXiv:2005.06417, 2020.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        }
+                    ],
+                    "index": 2.5,
+                    "bbox_fs": [
+                        105,
+                        101,
+                        505,
+                        126
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        132,
+                        505,
+                        167
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                132,
+                                506,
+                                146
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        132,
+                                        506,
+                                        146
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[28] Ilias Diakonikolas, Gautam Kamath, Daniel Kane, Jerry Li, Ankur Moitra, and Alistair Stewart.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                144,
+                                505,
+                                156
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        144,
+                                        505,
+                                        156
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Robust estimators in high-dimensions without the computational intractability. SIAM Journal",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                155,
+                                279,
+                                167
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        155,
+                                        279,
+                                        167
+                                    ],
+                                    "score": 1.0,
+                                    "content": "on Computing, 48(2):742–864, 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        }
+                    ],
+                    "index": 5,
+                    "bbox_fs": [
+                        105,
+                        132,
+                        506,
+                        167
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        173,
+                        504,
+                        208
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                173,
+                                506,
+                                187
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        173,
+                                        506,
+                                        187
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[29] Ilias Diakonikolas, Gautam Kamath, Daniel Kane, Jerry Li, Jacob Steinhardt, and Alistair Stew-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                185,
+                                505,
+                                198
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        185,
+                                        505,
+                                        198
+                                    ],
+                                    "score": 1.0,
+                                    "content": "art. Sever: A robust meta-algorithm for stochastic optimization. In International Conference on",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                195,
+                                307,
+                                209
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        195,
+                                        307,
+                                        209
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Machine Learning, pages 1596–1606, 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        }
+                    ],
+                    "index": 8,
+                    "bbox_fs": [
+                        106,
+                        173,
+                        506,
+                        209
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        215,
+                        505,
+                        249
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                214,
+                                506,
+                                227
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        214,
+                                        506,
+                                        227
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[30] Ilias Diakonikolas, Gautam Kamath, Daniel M. Kane, Jerry Li, Ankur Moitra, and Alis-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                225,
+                                506,
+                                240
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        225,
+                                        506,
+                                        240
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tair Stewart. Being Robust (in High Dimensions) Can Be Practical. arXiv e-prints, page",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                128,
+                                236,
+                                259,
+                                248
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        128,
+                                        236,
+                                        259,
+                                        248
+                                    ],
+                                    "score": 1.0,
+                                    "content": "arXiv:1703.00893, March 2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        }
+                    ],
+                    "index": 11,
+                    "bbox_fs": [
+                        105,
+                        214,
+                        506,
+                        248
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        255,
+                        505,
+                        300
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                255,
+                                506,
+                                269
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        255,
+                                        506,
+                                        269
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[31] Ilias Diakonikolas, Gautam Kamath, Daniel M Kane, Jerry Li, Ankur Moitra, and Alistair",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                267,
+                                506,
+                                280
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        267,
+                                        506,
+                                        280
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Stewart. Robustly learning a gaussian: Getting optimal error, efficiently. In Proceedings of the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                277,
+                                507,
+                                291
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        277,
+                                        507,
+                                        291
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Twenty-Ninth Annual ACM-SIAM Symposium on Discrete Algorithms, pages 2683–2702. SIAM,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                288,
+                                155,
+                                301
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        288,
+                                        155,
+                                        301
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        }
+                    ],
+                    "index": 14.5,
+                    "bbox_fs": [
+                        105,
+                        255,
+                        507,
+                        301
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        307,
+                        504,
+                        331
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                307,
+                                506,
+                                322
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        307,
+                                        506,
+                                        322
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[32] Ilias Diakonikolas and Daniel M Kane. Recent advances in algorithmic high-dimensional robust",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                320,
+                                330,
+                                331
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        320,
+                                        330,
+                                        331
+                                    ],
+                                    "score": 1.0,
+                                    "content": "statistics. arXiv preprint arXiv:1911.05911, 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        }
+                    ],
+                    "index": 17.5,
+                    "bbox_fs": [
+                        105,
+                        307,
+                        506,
+                        331
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        338,
+                        504,
+                        372
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                338,
+                                505,
+                                351
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        338,
+                                        505,
+                                        351
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[33] Ilias Diakonikolas, Daniel M Kane, and Alistair Stewart. Statistical query lower bounds for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                349,
+                                505,
+                                362
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        349,
+                                        505,
+                                        362
+                                    ],
+                                    "score": 1.0,
+                                    "content": "robust estimation of high-dimensional gaussians and gaussian mixtures. In 2017 IEEE 58th",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                360,
+                                500,
+                                374
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        360,
+                                        500,
+                                        374
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Annual Symposium on Foundations of Computer Science (FOCS), pages 73–84. IEEE, 2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        }
+                    ],
+                    "index": 20,
+                    "bbox_fs": [
+                        105,
+                        338,
+                        505,
+                        374
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        379,
+                        504,
+                        414
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                379,
+                                506,
+                                392
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        379,
+                                        506,
+                                        392
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[34] Ilias Diakonikolas, Daniel M Kane, and Alistair Stewart. List-decodable robust mean estimation",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                391,
+                                506,
+                                403
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        391,
+                                        506,
+                                        403
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and learning mixtures of spherical gaussians. In Proceedings of the 50th Annual ACM SIGACT",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                128,
+                                401,
+                                379,
+                                415
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        128,
+                                        401,
+                                        379,
+                                        415
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Symposium on Theory of Computing, pages 1047–1060, 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        }
+                    ],
+                    "index": 23,
+                    "bbox_fs": [
+                        105,
+                        379,
+                        506,
+                        415
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        420,
+                        504,
+                        454
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                419,
+                                506,
+                                434
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        419,
+                                        506,
+                                        434
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[35] Ilias Diakonikolas, Weihao Kong, and Alistair Stewart. Efficient algorithms and lower bounds",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                432,
+                                505,
+                                444
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        432,
+                                        505,
+                                        444
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for robust linear regression. In Proceedings of the Thirtieth Annual ACM-SIAM Symposium on",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                128,
+                                443,
+                                343,
+                                455
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        128,
+                                        443,
+                                        343,
+                                        455
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Discrete Algorithms, pages 2745–2754. SIAM, 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        }
+                    ],
+                    "index": 26,
+                    "bbox_fs": [
+                        105,
+                        419,
+                        506,
+                        455
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        461,
+                        504,
+                        495
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                461,
+                                505,
+                                475
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        461,
+                                        505,
+                                        475
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[36] Yihe Dong, Samuel Hopkins, and Jerry Li. Quantum entropy scoring for fast robust mean",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                471,
+                                505,
+                                487
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        471,
+                                        505,
+                                        487
+                                    ],
+                                    "score": 1.0,
+                                    "content": "estimation and improved outlier detection. In Advances in Neural Information Processing",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                484,
+                                264,
+                                496
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        484,
+                                        264,
+                                        496
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Systems, pages 6067–6077, 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        }
+                    ],
+                    "index": 29,
+                    "bbox_fs": [
+                        106,
+                        461,
+                        505,
+                        496
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        502,
+                        505,
+                        536
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                502,
+                                506,
+                                516
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        502,
+                                        506,
+                                        516
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[37] Cynthia Dwork, Frank McSherry, Kobbi Nissim, and Adam Smith. Calibrating noise to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                513,
+                                508,
+                                528
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        513,
+                                        508,
+                                        528
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sensitivity in private data analysis. In Theory of cryptography conference, pages 265–284.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                524,
+                                192,
+                                537
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        524,
+                                        192,
+                                        537
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Springer, 2006.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        }
+                    ],
+                    "index": 32,
+                    "bbox_fs": [
+                        105,
+                        502,
+                        508,
+                        537
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        543,
+                        505,
+                        567
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                544,
+                                506,
+                                557
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        544,
+                                        506,
+                                        557
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[38] Cynthia Dwork and Aaron Roth. The algorithmic foundations of differential privacy. Founda-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                555,
+                                425,
+                                567
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        555,
+                                        425,
+                                        567
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tions and Trends in Theoretical Computer Science, 9(3-4):211–407, 2014.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        }
+                    ],
+                    "index": 34.5,
+                    "bbox_fs": [
+                        106,
+                        544,
+                        506,
+                        567
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        574,
+                        505,
+                        608
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                574,
+                                505,
+                                587
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        574,
+                                        505,
+                                        587
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[39] Cynthia Dwork, Kunal Talwar, Abhradeep Thakurta, and Li Zhang. Analyze gauss: optimal",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                585,
+                                505,
+                                598
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        585,
+                                        505,
+                                        598
+                                    ],
+                                    "score": 1.0,
+                                    "content": "bounds for privacy-preserving principal component analysis. In Proceedings of the forty-sixth",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                596,
+                                409,
+                                609
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        596,
+                                        409,
+                                        609
+                                    ],
+                                    "score": 1.0,
+                                    "content": "annual ACM symposium on Theory of computing, pages 11–20, 2014.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        }
+                    ],
+                    "index": 37,
+                    "bbox_fs": [
+                        106,
+                        574,
+                        505,
+                        609
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        615,
+                        504,
+                        650
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                615,
+                                506,
+                                630
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        615,
+                                        506,
+                                        630
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[40] Úlfar Erlingsson, Vasyl Pihur, and Aleksandra Korolova. Rappor: Randomized aggregatable",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                627,
+                                506,
+                                640
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        627,
+                                        506,
+                                        640
+                                    ],
+                                    "score": 1.0,
+                                    "content": "privacy-preserving ordinal response. In Proceedings of the 2014 ACM SIGSAC conference on",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                128,
+                                638,
+                                390,
+                                651
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        128,
+                                        638,
+                                        390,
+                                        651
+                                    ],
+                                    "score": 1.0,
+                                    "content": "computer and communications security, pages 1054–1067, 2014.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        }
+                    ],
+                    "index": 40,
+                    "bbox_fs": [
+                        106,
+                        615,
+                        506,
+                        651
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        658,
+                        503,
+                        692
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                657,
+                                505,
+                                673
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        657,
+                                        505,
+                                        673
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[41] Giulia Fanti, Vasyl Pihur, and Úlfar Erlingsson. Building a rappor with the unknown: Privacy-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                669,
+                                506,
+                                684
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        669,
+                                        506,
+                                        684
+                                    ],
+                                    "score": 1.0,
+                                    "content": "preserving learning of associations and data dictionaries. Proceedings on Privacy Enhancing",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                680,
+                                273,
+                                693
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        680,
+                                        273,
+                                        693
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Technologies, 2016(3):41–61, 2016.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        }
+                    ],
+                    "index": 43,
+                    "bbox_fs": [
+                        105,
+                        657,
+                        506,
+                        693
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        699,
+                        503,
+                        722
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                698,
+                                505,
+                                713
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        698,
+                                        505,
+                                        713
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[42] Chao Gao et al. Robust regression via mutivariate regression depth. Bernoulli, 26(2):1139–1170,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                708,
+                                154,
+                                723
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        708,
+                                        154,
+                                        723
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2020.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        }
+                    ],
+                    "index": 45.5,
+                    "bbox_fs": [
+                        106,
+                        698,
+                        505,
+                        723
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        72,
+                        506,
+                        96
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                72,
+                                506,
+                                86
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        72,
+                                        506,
+                                        86
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[43] Sam Hopkins, Jerry Li, and Fred Zhang. Robust and heavy-tailed mean estimation made simple,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                83,
+                                480,
+                                96
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        83,
+                                        480,
+                                        96
+                                    ],
+                                    "score": 1.0,
+                                    "content": "via regret minimization. Advances in Neural Information Processing Systems, 33, 2020.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        }
+                    ],
+                    "index": 0.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        104,
+                        102,
+                        507,
+                        727
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                102,
+                                506,
+                                116
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        102,
+                                        506,
+                                        116
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[44] Samuel B Hopkins. Mean estimation with sub-gaussian rates in polynomial time. Annals of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                113,
+                                268,
+                                127
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        113,
+                                        268,
+                                        127
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Statistics, 48(2):1193–1213, 2020.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                133,
+                                506,
+                                146
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        133,
+                                        506,
+                                        146
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[45] Samuel B Hopkins and Jerry Li. Mixture models, robustness, and sum of squares proofs. In",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                142,
+                                506,
+                                158
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        142,
+                                        506,
+                                        158
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Proceedings of the 50th Annual ACM SIGACT Symposium on Theory of Computing, pages",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                155,
+                                203,
+                                168
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        155,
+                                        203,
+                                        168
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1021–1034, 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                173,
+                                506,
+                                188
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        173,
+                                        506,
+                                        188
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[46] Samuel B Hopkins and Jerry Li. How hard is robust mean estimation? In Conference on",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                125,
+                                184,
+                                334,
+                                197
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        125,
+                                        184,
+                                        334,
+                                        197
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Learning Theory, pages 1649–1682. PMLR, 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                204,
+                                506,
+                                217
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        204,
+                                        506,
+                                        217
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[47] Peter J. Huber. Robust Estimation of a Location Parameter. The Annals of Mathematical",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                214,
+                                258,
+                                228
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        214,
+                                        258,
+                                        228
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Statistics, 35(1):73 – 101, 1964.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                233,
+                                506,
+                                248
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        233,
+                                        506,
+                                        248
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[48] Arun Jambulapati, Jerry Li, and Kevin Tian. Robust sub-gaussian principal component analysis",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                244,
+                                507,
+                                259
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        244,
+                                        507,
+                                        259
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and width-independent schatten packing. Advances in Neural Information Processing Systems,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                255,
+                                169,
+                                268
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        255,
+                                        169,
+                                        268
+                                    ],
+                                    "score": 1.0,
+                                    "content": "33, 2020.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                273,
+                                506,
+                                289
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        273,
+                                        506,
+                                        289
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[49] He Jia and Santosh Vempala. Robustly clustering a mixture of gaussians. arXiv preprint",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                285,
+                                232,
+                                299
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        285,
+                                        232,
+                                        299
+                                    ],
+                                    "score": 1.0,
+                                    "content": "arXiv:1911.11838, 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                303,
+                                507,
+                                318
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        303,
+                                        507,
+                                        318
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[50] Peter Kairouz, H Brendan McMahan, Brendan Avent, Aurélien Bellet, Mehdi Bennis, Ar-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                315,
+                                507,
+                                330
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        315,
+                                        507,
+                                        330
+                                    ],
+                                    "score": 1.0,
+                                    "content": "jun Nitin Bhagoji, Keith Bonawitz, Zachary Charles, Graham Cormode, Rachel Cummings,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                325,
+                                507,
+                                340
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        325,
+                                        507,
+                                        340
+                                    ],
+                                    "score": 1.0,
+                                    "content": "et al. Advances and open problems in federated learning. arXiv preprint arXiv:1912.04977,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                336,
+                                154,
+                                351
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        336,
+                                        154,
+                                        351
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                356,
+                                505,
+                                370
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        356,
+                                        505,
+                                        370
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[51] Peter Kairouz, Sewoong Oh, and Pramod Viswanath. The composition theorem for differential",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                367,
+                                462,
+                                381
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        367,
+                                        462,
+                                        381
+                                    ],
+                                    "score": 1.0,
+                                    "content": "privacy. In International conference on machine learning, pages 1376–1385, 2015.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                387,
+                                505,
+                                400
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        387,
+                                        505,
+                                        400
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[52] Gautam Kamath, Jerry Li, Vikrant Singhal, and Jonathan Ullman. Privately learning high-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                398,
+                                478,
+                                411
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        398,
+                                        478,
+                                        411
+                                    ],
+                                    "score": 1.0,
+                                    "content": "dimensional distributions. In Conference on Learning Theory, pages 1853–1902, 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                416,
+                                506,
+                                431
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        416,
+                                        506,
+                                        431
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[53] Gautam Kamath, Or Sheffet, Vikrant Singhal, and Jonathan Ullman. Differentially private",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                427,
+                                506,
+                                441
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        427,
+                                        506,
+                                        441
+                                    ],
+                                    "score": 1.0,
+                                    "content": "algorithms for learning mixtures of separated gaussians. In 2020 Information Theory and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                438,
+                                350,
+                                453
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        438,
+                                        350,
+                                        453
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Applications Workshop (ITA), pages 1–62. IEEE, 2020.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                456,
+                                506,
+                                471
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        456,
+                                        506,
+                                        471
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[54] Gautam Kamath, Vikrant Singhal, and Jonathan Ullman. Private mean estimation of heavy-tailed",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                468,
+                                346,
+                                481
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        468,
+                                        346,
+                                        481
+                                    ],
+                                    "score": 1.0,
+                                    "content": "distributions. arXiv preprint arXiv:2002.09464, 2020.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                487,
+                                506,
+                                502
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        487,
+                                        506,
+                                        502
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[55] Haim Kaplan, Katrina Ligett, Yishay Mansour, Moni Naor, and Uri Stemmer. Privately learning",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                498,
+                                506,
+                                511
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        498,
+                                        506,
+                                        511
+                                    ],
+                                    "score": 1.0,
+                                    "content": "thresholds: Closing the exponential gap. In Conference on Learning Theory, pages 2263–2285.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                509,
+                                185,
+                                522
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        509,
+                                        185,
+                                        522
+                                    ],
+                                    "score": 1.0,
+                                    "content": "PMLR, 2020.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                527,
+                                507,
+                                542
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        527,
+                                        507,
+                                        542
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[56] Sushrut Karmalkar, Adam Klivans, and Pravesh Kothari. List-decodable linear regression. In",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                125,
+                                538,
+                                443,
+                                552
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        125,
+                                        538,
+                                        443,
+                                        552
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Advances in Neural Information Processing Systems, pages 7423–7432, 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                558,
+                                506,
+                                572
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        558,
+                                        506,
+                                        572
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[57] Sushrut Karmalkar and Eric Price. Compressed sensing with adversarial sparse noise via l1",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                125,
+                                568,
+                                391,
+                                582
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        125,
+                                        568,
+                                        391,
+                                        582
+                                    ],
+                                    "score": 1.0,
+                                    "content": "regression. In 2nd Symposium on Simplicity in Algorithms, 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                587,
+                                506,
+                                601
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        587,
+                                        506,
+                                        601
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[58] Vishesh Karwa and Salil Vadhan. Finite sample differentially private confidence intervals. arXiv",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                125,
+                                599,
+                                267,
+                                613
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        125,
+                                        599,
+                                        267,
+                                        613
+                                    ],
+                                    "score": 1.0,
+                                    "content": "preprint arXiv:1711.03908, 2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                618,
+                                506,
+                                631
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        618,
+                                        506,
+                                        631
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[59] Adam Klivans, Pravesh K Kothari, and Raghu Meka. Efficient algorithms for outlier-robust",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                629,
+                                420,
+                                642
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        629,
+                                        420,
+                                        642
+                                    ],
+                                    "score": 1.0,
+                                    "content": "regression. In Conference On Learning Theory, pages 1420–1430, 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                647,
+                                506,
+                                662
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        647,
+                                        506,
+                                        662
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[60] Weihao Kong, Raghav Somani, Sham Kakade, and Sewoong Oh. Robust meta-learning for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                659,
+                                507,
+                                672
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        659,
+                                        507,
+                                        672
+                                    ],
+                                    "score": 1.0,
+                                    "content": "mixed linear regression with small batches. Advances in Neural Information Processing Systems,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                670,
+                                169,
+                                683
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        670,
+                                        169,
+                                        683
+                                    ],
+                                    "score": 1.0,
+                                    "content": "33, 2020.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                689,
+                                506,
+                                702
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        689,
+                                        506,
+                                        702
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[61] Pravesh K Kothari, Jacob Steinhardt, and David Steurer. Robust moment estimation and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                699,
+                                506,
+                                714
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        699,
+                                        506,
+                                        714
+                                    ],
+                                    "score": 1.0,
+                                    "content": "improved clustering via sum of squares. In Proceedings of the 50th Annual ACM SIGACT",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                710,
+                                379,
+                                725
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        710,
+                                        379,
+                                        725
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Symposium on Theory of Computing, pages 1035–1046, 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        }
+                    ],
+                    "index": 23.5
+                }
+            ],
+            "page_idx": 12,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        300,
+                        741,
+                        311,
+                        750
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                299,
+                                740,
+                                313,
+                                754
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        299,
+                                        740,
+                                        313,
+                                        754
+                                    ],
+                                    "score": 1.0,
+                                    "content": "13",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        72,
+                        506,
+                        96
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                72,
+                                506,
+                                86
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        72,
+                                        506,
+                                        86
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[43] Sam Hopkins, Jerry Li, and Fred Zhang. Robust and heavy-tailed mean estimation made simple,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                83,
+                                480,
+                                96
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        83,
+                                        480,
+                                        96
+                                    ],
+                                    "score": 1.0,
+                                    "content": "via regret minimization. Advances in Neural Information Processing Systems, 33, 2020.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        }
+                    ],
+                    "index": 0.5,
+                    "bbox_fs": [
+                        106,
+                        72,
+                        506,
+                        96
+                    ]
+                },
+                {
+                    "type": "list",
+                    "bbox": [
+                        104,
+                        102,
+                        507,
+                        727
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                102,
+                                506,
+                                116
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        102,
+                                        506,
+                                        116
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[44] Samuel B Hopkins. Mean estimation with sub-gaussian rates in polynomial time. Annals of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                113,
+                                268,
+                                127
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        113,
+                                        268,
+                                        127
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Statistics, 48(2):1193–1213, 2020.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                133,
+                                506,
+                                146
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        133,
+                                        506,
+                                        146
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[45] Samuel B Hopkins and Jerry Li. Mixture models, robustness, and sum of squares proofs. In",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                142,
+                                506,
+                                158
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        142,
+                                        506,
+                                        158
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Proceedings of the 50th Annual ACM SIGACT Symposium on Theory of Computing, pages",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                155,
+                                203,
+                                168
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        155,
+                                        203,
+                                        168
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1021–1034, 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                173,
+                                506,
+                                188
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        173,
+                                        506,
+                                        188
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[46] Samuel B Hopkins and Jerry Li. How hard is robust mean estimation? In Conference on",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                125,
+                                184,
+                                334,
+                                197
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        125,
+                                        184,
+                                        334,
+                                        197
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Learning Theory, pages 1649–1682. PMLR, 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                204,
+                                506,
+                                217
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        204,
+                                        506,
+                                        217
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[47] Peter J. Huber. Robust Estimation of a Location Parameter. The Annals of Mathematical",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                214,
+                                258,
+                                228
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        214,
+                                        258,
+                                        228
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Statistics, 35(1):73 – 101, 1964.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                233,
+                                506,
+                                248
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        233,
+                                        506,
+                                        248
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[48] Arun Jambulapati, Jerry Li, and Kevin Tian. Robust sub-gaussian principal component analysis",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                244,
+                                507,
+                                259
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        244,
+                                        507,
+                                        259
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and width-independent schatten packing. Advances in Neural Information Processing Systems,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                255,
+                                169,
+                                268
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        255,
+                                        169,
+                                        268
+                                    ],
+                                    "score": 1.0,
+                                    "content": "33, 2020.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                273,
+                                506,
+                                289
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        273,
+                                        506,
+                                        289
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[49] He Jia and Santosh Vempala. Robustly clustering a mixture of gaussians. arXiv preprint",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                285,
+                                232,
+                                299
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        285,
+                                        232,
+                                        299
+                                    ],
+                                    "score": 1.0,
+                                    "content": "arXiv:1911.11838, 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                303,
+                                507,
+                                318
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        303,
+                                        507,
+                                        318
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[50] Peter Kairouz, H Brendan McMahan, Brendan Avent, Aurélien Bellet, Mehdi Bennis, Ar-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                315,
+                                507,
+                                330
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        315,
+                                        507,
+                                        330
+                                    ],
+                                    "score": 1.0,
+                                    "content": "jun Nitin Bhagoji, Keith Bonawitz, Zachary Charles, Graham Cormode, Rachel Cummings,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                325,
+                                507,
+                                340
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        325,
+                                        507,
+                                        340
+                                    ],
+                                    "score": 1.0,
+                                    "content": "et al. Advances and open problems in federated learning. arXiv preprint arXiv:1912.04977,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                336,
+                                154,
+                                351
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        336,
+                                        154,
+                                        351
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                356,
+                                505,
+                                370
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        356,
+                                        505,
+                                        370
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[51] Peter Kairouz, Sewoong Oh, and Pramod Viswanath. The composition theorem for differential",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                367,
+                                462,
+                                381
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        367,
+                                        462,
+                                        381
+                                    ],
+                                    "score": 1.0,
+                                    "content": "privacy. In International conference on machine learning, pages 1376–1385, 2015.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                387,
+                                505,
+                                400
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        387,
+                                        505,
+                                        400
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[52] Gautam Kamath, Jerry Li, Vikrant Singhal, and Jonathan Ullman. Privately learning high-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                398,
+                                478,
+                                411
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        398,
+                                        478,
+                                        411
+                                    ],
+                                    "score": 1.0,
+                                    "content": "dimensional distributions. In Conference on Learning Theory, pages 1853–1902, 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                416,
+                                506,
+                                431
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        416,
+                                        506,
+                                        431
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[53] Gautam Kamath, Or Sheffet, Vikrant Singhal, and Jonathan Ullman. Differentially private",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                427,
+                                506,
+                                441
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        427,
+                                        506,
+                                        441
+                                    ],
+                                    "score": 1.0,
+                                    "content": "algorithms for learning mixtures of separated gaussians. In 2020 Information Theory and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                438,
+                                350,
+                                453
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        438,
+                                        350,
+                                        453
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Applications Workshop (ITA), pages 1–62. IEEE, 2020.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                456,
+                                506,
+                                471
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        456,
+                                        506,
+                                        471
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[54] Gautam Kamath, Vikrant Singhal, and Jonathan Ullman. Private mean estimation of heavy-tailed",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                468,
+                                346,
+                                481
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        468,
+                                        346,
+                                        481
+                                    ],
+                                    "score": 1.0,
+                                    "content": "distributions. arXiv preprint arXiv:2002.09464, 2020.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                487,
+                                506,
+                                502
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        487,
+                                        506,
+                                        502
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[55] Haim Kaplan, Katrina Ligett, Yishay Mansour, Moni Naor, and Uri Stemmer. Privately learning",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                498,
+                                506,
+                                511
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        498,
+                                        506,
+                                        511
+                                    ],
+                                    "score": 1.0,
+                                    "content": "thresholds: Closing the exponential gap. In Conference on Learning Theory, pages 2263–2285.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                509,
+                                185,
+                                522
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        509,
+                                        185,
+                                        522
+                                    ],
+                                    "score": 1.0,
+                                    "content": "PMLR, 2020.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                527,
+                                507,
+                                542
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        527,
+                                        507,
+                                        542
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[56] Sushrut Karmalkar, Adam Klivans, and Pravesh Kothari. List-decodable linear regression. In",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                125,
+                                538,
+                                443,
+                                552
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        125,
+                                        538,
+                                        443,
+                                        552
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Advances in Neural Information Processing Systems, pages 7423–7432, 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                558,
+                                506,
+                                572
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        558,
+                                        506,
+                                        572
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[57] Sushrut Karmalkar and Eric Price. Compressed sensing with adversarial sparse noise via l1",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                125,
+                                568,
+                                391,
+                                582
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        125,
+                                        568,
+                                        391,
+                                        582
+                                    ],
+                                    "score": 1.0,
+                                    "content": "regression. In 2nd Symposium on Simplicity in Algorithms, 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                587,
+                                506,
+                                601
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        587,
+                                        506,
+                                        601
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[58] Vishesh Karwa and Salil Vadhan. Finite sample differentially private confidence intervals. arXiv",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                125,
+                                599,
+                                267,
+                                613
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        125,
+                                        599,
+                                        267,
+                                        613
+                                    ],
+                                    "score": 1.0,
+                                    "content": "preprint arXiv:1711.03908, 2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                618,
+                                506,
+                                631
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        618,
+                                        506,
+                                        631
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[59] Adam Klivans, Pravesh K Kothari, and Raghu Meka. Efficient algorithms for outlier-robust",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                629,
+                                420,
+                                642
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        629,
+                                        420,
+                                        642
+                                    ],
+                                    "score": 1.0,
+                                    "content": "regression. In Conference On Learning Theory, pages 1420–1430, 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                647,
+                                506,
+                                662
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        647,
+                                        506,
+                                        662
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[60] Weihao Kong, Raghav Somani, Sham Kakade, and Sewoong Oh. Robust meta-learning for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                659,
+                                507,
+                                672
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        659,
+                                        507,
+                                        672
+                                    ],
+                                    "score": 1.0,
+                                    "content": "mixed linear regression with small batches. Advances in Neural Information Processing Systems,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                670,
+                                169,
+                                683
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        670,
+                                        169,
+                                        683
+                                    ],
+                                    "score": 1.0,
+                                    "content": "33, 2020.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                689,
+                                506,
+                                702
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        689,
+                                        506,
+                                        702
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[61] Pravesh K Kothari, Jacob Steinhardt, and David Steurer. Robust moment estimation and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                699,
+                                506,
+                                714
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        699,
+                                        506,
+                                        714
+                                    ],
+                                    "score": 1.0,
+                                    "content": "improved clustering via sum of squares. In Proceedings of the 50th Annual ACM SIGACT",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                710,
+                                379,
+                                725
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        710,
+                                        379,
+                                        725
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Symposium on Theory of Computing, pages 1035–1046, 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                73,
+                                507,
+                                86
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        73,
+                                        507,
+                                        86
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[62] Kevin A Lai, Anup B Rao, and Santosh Vempala. Agnostic estimation of mean and covariance.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 0,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                125,
+                                81,
+                                507,
+                                99
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        125,
+                                        81,
+                                        507,
+                                        99
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In 2016 IEEE 57th Annual Symposium on Foundations of Computer Science (FOCS), pages",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                128,
+                                95,
+                                219,
+                                107
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        128,
+                                        95,
+                                        219,
+                                        107
+                                    ],
+                                    "score": 1.0,
+                                    "content": "665–674. IEEE, 2016.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 2,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                114,
+                                507,
+                                128
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        114,
+                                        507,
+                                        128
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[63] Jerry Li. CSE 599-M, Lecture Notes: Robustness in Machine Learning , 2019. URL: https:",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 3,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                125,
+                                374,
+                                139
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        125,
+                                        374,
+                                        139
+                                    ],
+                                    "score": 1.0,
+                                    "content": "//jerryzli.github.io/robust-ml-fall19/lec7.pdf.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 4,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                146,
+                                506,
+                                157
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        146,
+                                        506,
+                                        157
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[64] Jerry Li and Guanghao Ye. Robust gaussian covariance estimation in nearly-matrix multiplica-",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 5,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                156,
+                                423,
+                                170
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        156,
+                                        423,
+                                        170
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tion time. Advances in Neural Information Processing Systems, 33, 2020.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 6,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                175,
+                                506,
+                                189
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        175,
+                                        506,
+                                        189
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[65] Liu Liu, Yanyao Shen, Tianyang Li, and Constantine Caramanis. High dimensional robust",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 7,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                187,
+                                364,
+                                200
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        187,
+                                        364,
+                                        200
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sparse regression. arXiv preprint arXiv:1805.11643, 2018.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 8,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                206,
+                                507,
+                                219
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        206,
+                                        507,
+                                        219
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[66] Xiaohui Liu. Fast implementation of the tukey depth. Computational Statistics, 32(4):1395–",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 9,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                217,
+                                180,
+                                231
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        217,
+                                        180,
+                                        231
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1410, 2017.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 10,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                236,
+                                506,
+                                250
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        236,
+                                        506,
+                                        250
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[67] Xiaohui Liu, Karl Mosler, and Pavlo Mozharovskyi. Fast computation of tukey trimmed",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 11,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                247,
+                                507,
+                                261
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        247,
+                                        507,
+                                        261
+                                    ],
+                                    "score": 1.0,
+                                    "content": "regions and median in dimension p> 2. Journal of Computational and Graphical Statistics,",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                258,
+                                218,
+                                272
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        258,
+                                        218,
+                                        272
+                                    ],
+                                    "score": 1.0,
+                                    "content": "28(3):682–697, 2019.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 13,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                277,
+                                507,
+                                292
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        277,
+                                        507,
+                                        292
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[68] Gábor Lugosi, Shahar Mendelson, et al. Sub-gaussian estimators of the mean of a random",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 14,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                289,
+                                330,
+                                303
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        289,
+                                        330,
+                                        303
+                                    ],
+                                    "score": 1.0,
+                                    "content": "vector. Annals of Statistics, 47(2):783–794, 2019.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 15,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                308,
+                                506,
+                                322
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        308,
+                                        506,
+                                        322
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[69] Frank McSherry and Kunal Talwar. Mechanism design via differential privacy. In 48th Annual",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 16,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                320,
+                                507,
+                                334
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        320,
+                                        507,
+                                        334
+                                    ],
+                                    "score": 1.0,
+                                    "content": "IEEE Symposium on Foundations of Computer Science (FOCS’07), pages 94–103. IEEE, 2007.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                338,
+                                507,
+                                354
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        338,
+                                        507,
+                                        354
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[70] Bhaskar Mukhoty, Govind Gopakumar, Prateek Jain, and Purushottam Kar. Globally-convergent",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 18,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                351,
+                                506,
+                                365
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        351,
+                                        506,
+                                        365
+                                    ],
+                                    "score": 1.0,
+                                    "content": "iteratively reweighted least squares for robust regression problems. In The 22nd International",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                362,
+                                424,
+                                375
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        362,
+                                        424,
+                                        375
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Conference on Artificial Intelligence and Statistics, pages 313–322, 2019.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 20,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                381,
+                                506,
+                                394
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        381,
+                                        506,
+                                        394
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[71] A. Prasad, A. S. Suggala, S. Balakrishnan, and P. Ravikumar. Robust estimation via robust",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 21,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                392,
+                                375,
+                                406
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        392,
+                                        375,
+                                        406
+                                    ],
+                                    "score": 1.0,
+                                    "content": "gradient estimation. arXiv preprint arXiv:1802.06485, 2018.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 22,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                411,
+                                506,
+                                425
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        411,
+                                        506,
+                                        425
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[72] Prasad Raghavendra and Morris Yau. List decodable learning via sum of squares. In Proceedings",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 23,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                422,
+                                507,
+                                436
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        422,
+                                        507,
+                                        436
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of the Fourteenth Annual ACM-SIAM Symposium on Discrete Algorithms, pages 161–180. SIAM,",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                433,
+                                154,
+                                447
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        433,
+                                        154,
+                                        447
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2020.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 25,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                452,
+                                507,
+                                469
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        452,
+                                        507,
+                                        469
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[73] Jacob Steinhardt, Moses Charikar, and Gregory Valiant. Resilience: A criterion for learning",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 26,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                462,
+                                507,
+                                479
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        462,
+                                        507,
+                                        479
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in the presence of arbitrary outliers. In 9th Innovations in Theoretical Computer Science",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                474,
+                                465,
+                                488
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        474,
+                                        465,
+                                        488
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Conference (ITCS 2018). Schloss Dagstuhl-Leibniz-Zentrum fuer Informatik, 2018.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 28,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                494,
+                                507,
+                                508
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        494,
+                                        507,
+                                        508
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[74] Jun Tang, Aleksandra Korolova, Xiaolong Bai, Xueqiang Wang, and Xiaofeng Wang. Pri-",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 29,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                504,
+                                507,
+                                520
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        504,
+                                        507,
+                                        520
+                                    ],
+                                    "score": 1.0,
+                                    "content": "vacy loss in apple’s implementation of differential privacy on macos 10.12. arXiv preprint",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                516,
+                                231,
+                                530
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        516,
+                                        231,
+                                        530
+                                    ],
+                                    "score": 1.0,
+                                    "content": "arXiv:1709.02753, 2017.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 31,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                536,
+                                507,
+                                550
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        536,
+                                        507,
+                                        550
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[75] Terence Tao. Topics in random matrix theory, volume 132. American Mathematical Soc., 2012.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 32,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                556,
+                                507,
+                                569
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        556,
+                                        507,
+                                        569
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[76] John W Tukey. A survey of sampling from contaminated distributions. Contributions to",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 33,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                124,
+                                566,
+                                322,
+                                581
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        124,
+                                        566,
+                                        322,
+                                        581
+                                    ],
+                                    "score": 1.0,
+                                    "content": "probability and statistics, pages 448–485, 1960.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 34,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                584,
+                                507,
+                                600
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        584,
+                                        507,
+                                        600
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[77] Martin J Wainwright. High-dimensional statistics: A non-asymptotic viewpoint, volume 48.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 35,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                128,
+                                598,
+                                270,
+                                610
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        128,
+                                        598,
+                                        270,
+                                        610
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Cambridge University Press, 2019.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 36,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                615,
+                                507,
+                                631
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        615,
+                                        507,
+                                        631
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[78] Lu Wei, Anand D Sarwate, Jukka Corander, Alfred Hero, and Vahid Tarokh. Analysis of a",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 37,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                627,
+                                506,
+                                642
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        627,
+                                        506,
+                                        642
+                                    ],
+                                    "score": 1.0,
+                                    "content": "privacy-preserving pca algorithm using random matrix theory. In 2016 IEEE Global Conference",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                637,
+                                464,
+                                653
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        637,
+                                        464,
+                                        653
+                                    ],
+                                    "score": 1.0,
+                                    "content": "on Signal and Information Processing (GlobalSIP), pages 1335–1339. IEEE, 2016.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 39,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                658,
+                                507,
+                                672
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        658,
+                                        507,
+                                        672
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[79] Huang Xiao, Battista Biggio, Gavin Brown, Giorgio Fumera, Claudia Eckert, and Fabio Roli.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 40,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                670,
+                                506,
+                                684
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        670,
+                                        506,
+                                        684
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Is feature selection secure against training data poisoning? In International Conference on",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                679,
+                                340,
+                                695
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        679,
+                                        340,
+                                        695
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Machine Learning, pages 1689–1698. PMLR, 2015.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 42,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                698,
+                                506,
+                                715
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        698,
+                                        506,
+                                        715
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[80] Huanyu Zhang, Gautam Kamath, Janardhan Kulkarni, and Zhiwei Steven Wu. Privately learning",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 43,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                711,
+                                384,
+                                725
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        711,
+                                        384,
+                                        725
+                                    ],
+                                    "score": 1.0,
+                                    "content": "markov random fields. arXiv preprint arXiv:2002.09463, 2020.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 44,
+                            "is_list_end_line": true
+                        }
+                    ],
+                    "index": 23.5,
+                    "bbox_fs": [
+                        104,
+                        102,
+                        507,
+                        725
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        57,
+                        507,
+                        724
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                73,
+                                507,
+                                86
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        73,
+                                        507,
+                                        86
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[62] Kevin A Lai, Anup B Rao, and Santosh Vempala. Agnostic estimation of mean and covariance.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                125,
+                                81,
+                                507,
+                                99
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        125,
+                                        81,
+                                        507,
+                                        99
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In 2016 IEEE 57th Annual Symposium on Foundations of Computer Science (FOCS), pages",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                128,
+                                95,
+                                219,
+                                107
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        128,
+                                        95,
+                                        219,
+                                        107
+                                    ],
+                                    "score": 1.0,
+                                    "content": "665–674. IEEE, 2016.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                114,
+                                507,
+                                128
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        114,
+                                        507,
+                                        128
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[63] Jerry Li. CSE 599-M, Lecture Notes: Robustness in Machine Learning , 2019. URL: https:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                125,
+                                374,
+                                139
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        125,
+                                        374,
+                                        139
+                                    ],
+                                    "score": 1.0,
+                                    "content": "//jerryzli.github.io/robust-ml-fall19/lec7.pdf.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                146,
+                                506,
+                                157
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        146,
+                                        506,
+                                        157
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[64] Jerry Li and Guanghao Ye. Robust gaussian covariance estimation in nearly-matrix multiplica-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                156,
+                                423,
+                                170
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        156,
+                                        423,
+                                        170
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tion time. Advances in Neural Information Processing Systems, 33, 2020.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                175,
+                                506,
+                                189
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        175,
+                                        506,
+                                        189
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[65] Liu Liu, Yanyao Shen, Tianyang Li, and Constantine Caramanis. High dimensional robust",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                187,
+                                364,
+                                200
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        187,
+                                        364,
+                                        200
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sparse regression. arXiv preprint arXiv:1805.11643, 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                206,
+                                507,
+                                219
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        206,
+                                        507,
+                                        219
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[66] Xiaohui Liu. Fast implementation of the tukey depth. Computational Statistics, 32(4):1395–",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                217,
+                                180,
+                                231
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        217,
+                                        180,
+                                        231
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1410, 2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                236,
+                                506,
+                                250
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        236,
+                                        506,
+                                        250
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[67] Xiaohui Liu, Karl Mosler, and Pavlo Mozharovskyi. Fast computation of tukey trimmed",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                247,
+                                507,
+                                261
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        247,
+                                        507,
+                                        261
+                                    ],
+                                    "score": 1.0,
+                                    "content": "regions and median in dimension p> 2. Journal of Computational and Graphical Statistics,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                258,
+                                218,
+                                272
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        258,
+                                        218,
+                                        272
+                                    ],
+                                    "score": 1.0,
+                                    "content": "28(3):682–697, 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                277,
+                                507,
+                                292
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        277,
+                                        507,
+                                        292
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[68] Gábor Lugosi, Shahar Mendelson, et al. Sub-gaussian estimators of the mean of a random",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                289,
+                                330,
+                                303
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        289,
+                                        330,
+                                        303
+                                    ],
+                                    "score": 1.0,
+                                    "content": "vector. Annals of Statistics, 47(2):783–794, 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                308,
+                                506,
+                                322
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        308,
+                                        506,
+                                        322
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[69] Frank McSherry and Kunal Talwar. Mechanism design via differential privacy. In 48th Annual",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                320,
+                                507,
+                                334
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        320,
+                                        507,
+                                        334
+                                    ],
+                                    "score": 1.0,
+                                    "content": "IEEE Symposium on Foundations of Computer Science (FOCS’07), pages 94–103. IEEE, 2007.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                338,
+                                507,
+                                354
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        338,
+                                        507,
+                                        354
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[70] Bhaskar Mukhoty, Govind Gopakumar, Prateek Jain, and Purushottam Kar. Globally-convergent",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                351,
+                                506,
+                                365
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        351,
+                                        506,
+                                        365
+                                    ],
+                                    "score": 1.0,
+                                    "content": "iteratively reweighted least squares for robust regression problems. In The 22nd International",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                362,
+                                424,
+                                375
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        362,
+                                        424,
+                                        375
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Conference on Artificial Intelligence and Statistics, pages 313–322, 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                381,
+                                506,
+                                394
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        381,
+                                        506,
+                                        394
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[71] A. Prasad, A. S. Suggala, S. Balakrishnan, and P. Ravikumar. Robust estimation via robust",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                392,
+                                375,
+                                406
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        392,
+                                        375,
+                                        406
+                                    ],
+                                    "score": 1.0,
+                                    "content": "gradient estimation. arXiv preprint arXiv:1802.06485, 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                411,
+                                506,
+                                425
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        411,
+                                        506,
+                                        425
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[72] Prasad Raghavendra and Morris Yau. List decodable learning via sum of squares. In Proceedings",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                422,
+                                507,
+                                436
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        422,
+                                        507,
+                                        436
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of the Fourteenth Annual ACM-SIAM Symposium on Discrete Algorithms, pages 161–180. SIAM,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                433,
+                                154,
+                                447
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        433,
+                                        154,
+                                        447
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2020.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                452,
+                                507,
+                                469
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        452,
+                                        507,
+                                        469
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[73] Jacob Steinhardt, Moses Charikar, and Gregory Valiant. Resilience: A criterion for learning",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                462,
+                                507,
+                                479
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        462,
+                                        507,
+                                        479
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in the presence of arbitrary outliers. In 9th Innovations in Theoretical Computer Science",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                474,
+                                465,
+                                488
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        474,
+                                        465,
+                                        488
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Conference (ITCS 2018). Schloss Dagstuhl-Leibniz-Zentrum fuer Informatik, 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                494,
+                                507,
+                                508
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        494,
+                                        507,
+                                        508
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[74] Jun Tang, Aleksandra Korolova, Xiaolong Bai, Xueqiang Wang, and Xiaofeng Wang. Pri-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                504,
+                                507,
+                                520
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        504,
+                                        507,
+                                        520
+                                    ],
+                                    "score": 1.0,
+                                    "content": "vacy loss in apple’s implementation of differential privacy on macos 10.12. arXiv preprint",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                516,
+                                231,
+                                530
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        516,
+                                        231,
+                                        530
+                                    ],
+                                    "score": 1.0,
+                                    "content": "arXiv:1709.02753, 2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                536,
+                                507,
+                                550
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        536,
+                                        507,
+                                        550
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[75] Terence Tao. Topics in random matrix theory, volume 132. American Mathematical Soc., 2012.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                556,
+                                507,
+                                569
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        556,
+                                        507,
+                                        569
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[76] John W Tukey. A survey of sampling from contaminated distributions. Contributions to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                124,
+                                566,
+                                322,
+                                581
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        124,
+                                        566,
+                                        322,
+                                        581
+                                    ],
+                                    "score": 1.0,
+                                    "content": "probability and statistics, pages 448–485, 1960.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                584,
+                                507,
+                                600
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        584,
+                                        507,
+                                        600
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[77] Martin J Wainwright. High-dimensional statistics: A non-asymptotic viewpoint, volume 48.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                128,
+                                598,
+                                270,
+                                610
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        128,
+                                        598,
+                                        270,
+                                        610
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Cambridge University Press, 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                615,
+                                507,
+                                631
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        615,
+                                        507,
+                                        631
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[78] Lu Wei, Anand D Sarwate, Jukka Corander, Alfred Hero, and Vahid Tarokh. Analysis of a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                627,
+                                506,
+                                642
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        627,
+                                        506,
+                                        642
+                                    ],
+                                    "score": 1.0,
+                                    "content": "privacy-preserving pca algorithm using random matrix theory. In 2016 IEEE Global Conference",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                637,
+                                464,
+                                653
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        637,
+                                        464,
+                                        653
+                                    ],
+                                    "score": 1.0,
+                                    "content": "on Signal and Information Processing (GlobalSIP), pages 1335–1339. IEEE, 2016.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                658,
+                                507,
+                                672
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        658,
+                                        507,
+                                        672
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[79] Huang Xiao, Battista Biggio, Gavin Brown, Giorgio Fumera, Claudia Eckert, and Fabio Roli.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                670,
+                                506,
+                                684
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        670,
+                                        506,
+                                        684
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Is feature selection secure against training data poisoning? In International Conference on",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                679,
+                                340,
+                                695
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        679,
+                                        340,
+                                        695
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Machine Learning, pages 1689–1698. PMLR, 2015.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                698,
+                                506,
+                                715
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        698,
+                                        506,
+                                        715
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[80] Huanyu Zhang, Gautam Kamath, Janardhan Kulkarni, and Zhiwei Steven Wu. Privately learning",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                711,
+                                384,
+                                725
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        711,
+                                        384,
+                                        725
+                                    ],
+                                    "score": 1.0,
+                                    "content": "markov random fields. arXiv preprint arXiv:2002.09463, 2020.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        }
+                    ],
+                    "index": 22
+                }
+            ],
+            "page_idx": 13,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        300,
+                        741,
+                        311,
+                        750
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                299,
+                                740,
+                                313,
+                                754
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        299,
+                                        740,
+                                        313,
+                                        754
+                                    ],
+                                    "score": 1.0,
+                                    "content": "14",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "list",
+                    "bbox": [
+                        105,
+                        57,
+                        507,
+                        724
+                    ],
+                    "lines": [],
+                    "index": 22,
+                    "bbox_fs": [
+                        104,
+                        73,
+                        507,
+                        725
+                    ],
+                    "lines_deleted": true
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [],
+            "page_idx": 14,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [],
+            "para_blocks": []
+        }
+    ],
+    "_backend": "pipeline",
+    "_version_name": "2.2.2"
+}
\ No newline at end of file
diff --git a/parse/train/CuQoImkKkIj/CuQoImkKkIj_model.json b/parse/train/CuQoImkKkIj/CuQoImkKkIj_model.json
new file mode 100644
index 0000000000000000000000000000000000000000..5096cc1676647eb9e8e9618a102112b92cea2b95
--- /dev/null
+++ b/parse/train/CuQoImkKkIj/CuQoImkKkIj_model.json
@@ -0,0 +1,25508 @@
+[
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    398,
+                    711,
+                    1303,
+                    711,
+                    1303,
+                    1138,
+                    398,
+                    1138
+                ],
+                "score": 0.984
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    1258,
+                    1404,
+                    1258,
+                    1404,
+                    1502,
+                    299,
+                    1502
+                ],
+                "score": 0.98
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1517,
+                    1404,
+                    1517,
+                    1404,
+                    1761,
+                    298,
+                    1761
+                ],
+                "score": 0.979
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1774,
+                    1404,
+                    1774,
+                    1404,
+                    1988,
+                    298,
+                    1988
+                ],
+                "score": 0.978
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    300,
+                    1190,
+                    530,
+                    1190,
+                    530,
+                    1228,
+                    300,
+                    1228
+                ],
+                "score": 0.905
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    335,
+                    270,
+                    1366,
+                    270,
+                    1366,
+                    328,
+                    335,
+                    328
+                ],
+                "score": 0.896
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    788,
+                    645,
+                    912,
+                    645,
+                    912,
+                    681,
+                    788,
+                    681
+                ],
+                "score": 0.891
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    500,
+                    477,
+                    1189,
+                    477,
+                    1189,
+                    567,
+                    500,
+                    567
+                ],
+                "score": 0.779
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    487,
+                    443,
+                    1211,
+                    443,
+                    1211,
+                    476,
+                    487,
+                    476
+                ],
+                "score": 0.731
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    298,
+                    2032,
+                    1070,
+                    2032,
+                    1070,
+                    2061,
+                    298,
+                    2061
+                ],
+                "score": 0.681
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    2032,
+                    1070,
+                    2032,
+                    1070,
+                    2061,
+                    298,
+                    2061
+                ],
+                "score": 0.206
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    799,
+                    1639,
+                    858,
+                    1639,
+                    858,
+                    1670,
+                    799,
+                    1670
+                ],
+                "score": 0.84,
+                "latex": "\\ S 1 . 2 )"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1187.0,
+                    536.0,
+                    1187.0,
+                    536.0,
+                    1235.0,
+                    292.0,
+                    1235.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    333.0,
+                    269.0,
+                    1365.0,
+                    269.0,
+                    1365.0,
+                    332.0,
+                    333.0,
+                    332.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    782.0,
+                    642.0,
+                    919.0,
+                    642.0,
+                    919.0,
+                    686.0,
+                    782.0,
+                    686.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    2029.0,
+                    1075.0,
+                    2029.0,
+                    1075.0,
+                    2067.0,
+                    294.0,
+                    2067.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    393.0,
+                    710.0,
+                    1305.0,
+                    710.0,
+                    1305.0,
+                    749.0,
+                    393.0,
+                    749.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    394.0,
+                    743.0,
+                    1306.0,
+                    743.0,
+                    1306.0,
+                    778.0,
+                    394.0,
+                    778.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    394.0,
+                    775.0,
+                    1304.0,
+                    775.0,
+                    1304.0,
+                    806.0,
+                    394.0,
+                    806.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    394.0,
+                    803.0,
+                    1304.0,
+                    803.0,
+                    1304.0,
+                    837.0,
+                    394.0,
+                    837.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    394.0,
+                    836.0,
+                    1304.0,
+                    836.0,
+                    1304.0,
+                    868.0,
+                    394.0,
+                    868.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    393.0,
+                    866.0,
+                    1306.0,
+                    866.0,
+                    1306.0,
+                    898.0,
+                    393.0,
+                    898.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    393.0,
+                    895.0,
+                    1306.0,
+                    895.0,
+                    1306.0,
+                    930.0,
+                    393.0,
+                    930.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    395.0,
+                    927.0,
+                    1305.0,
+                    927.0,
+                    1305.0,
+                    958.0,
+                    395.0,
+                    958.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    394.0,
+                    956.0,
+                    1306.0,
+                    956.0,
+                    1306.0,
+                    988.0,
+                    394.0,
+                    988.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    395.0,
+                    987.0,
+                    1306.0,
+                    987.0,
+                    1306.0,
+                    1019.0,
+                    395.0,
+                    1019.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    395.0,
+                    1017.0,
+                    1304.0,
+                    1017.0,
+                    1304.0,
+                    1049.0,
+                    395.0,
+                    1049.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    393.0,
+                    1046.0,
+                    1307.0,
+                    1046.0,
+                    1307.0,
+                    1083.0,
+                    393.0,
+                    1083.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    394.0,
+                    1078.0,
+                    1303.0,
+                    1078.0,
+                    1303.0,
+                    1110.0,
+                    394.0,
+                    1110.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    392.0,
+                    1107.0,
+                    1299.0,
+                    1107.0,
+                    1299.0,
+                    1142.0,
+                    392.0,
+                    1142.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1259.0,
+                    1404.0,
+                    1259.0,
+                    1404.0,
+                    1293.0,
+                    295.0,
+                    1293.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1288.0,
+                    1408.0,
+                    1288.0,
+                    1408.0,
+                    1324.0,
+                    293.0,
+                    1324.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1319.0,
+                    1404.0,
+                    1319.0,
+                    1404.0,
+                    1355.0,
+                    294.0,
+                    1355.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1350.0,
+                    1406.0,
+                    1350.0,
+                    1406.0,
+                    1386.0,
+                    292.0,
+                    1386.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1380.0,
+                    1405.0,
+                    1380.0,
+                    1405.0,
+                    1415.0,
+                    293.0,
+                    1415.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1410.0,
+                    1405.0,
+                    1410.0,
+                    1405.0,
+                    1445.0,
+                    293.0,
+                    1445.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1442.0,
+                    1405.0,
+                    1442.0,
+                    1405.0,
+                    1474.0,
+                    294.0,
+                    1474.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1472.0,
+                    693.0,
+                    1472.0,
+                    693.0,
+                    1503.0,
+                    294.0,
+                    1503.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1515.0,
+                    1408.0,
+                    1515.0,
+                    1408.0,
+                    1553.0,
+                    293.0,
+                    1553.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1548.0,
+                    1406.0,
+                    1548.0,
+                    1406.0,
+                    1582.0,
+                    294.0,
+                    1582.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1578.0,
+                    1406.0,
+                    1578.0,
+                    1406.0,
+                    1612.0,
+                    294.0,
+                    1612.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1606.0,
+                    1405.0,
+                    1606.0,
+                    1405.0,
+                    1645.0,
+                    292.0,
+                    1645.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1638.0,
+                    798.0,
+                    1638.0,
+                    798.0,
+                    1674.0,
+                    293.0,
+                    1674.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    859.0,
+                    1638.0,
+                    1405.0,
+                    1638.0,
+                    1405.0,
+                    1674.0,
+                    859.0,
+                    1674.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1669.0,
+                    1405.0,
+                    1669.0,
+                    1405.0,
+                    1703.0,
+                    291.0,
+                    1703.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1700.0,
+                    1406.0,
+                    1700.0,
+                    1406.0,
+                    1733.0,
+                    294.0,
+                    1733.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1726.0,
+                    1176.0,
+                    1726.0,
+                    1176.0,
+                    1768.0,
+                    292.0,
+                    1768.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1774.0,
+                    1406.0,
+                    1774.0,
+                    1406.0,
+                    1809.0,
+                    294.0,
+                    1809.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1802.0,
+                    1405.0,
+                    1802.0,
+                    1405.0,
+                    1839.0,
+                    293.0,
+                    1839.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1834.0,
+                    1404.0,
+                    1834.0,
+                    1404.0,
+                    1871.0,
+                    293.0,
+                    1871.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1866.0,
+                    1404.0,
+                    1866.0,
+                    1404.0,
+                    1901.0,
+                    294.0,
+                    1901.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1895.0,
+                    1405.0,
+                    1895.0,
+                    1405.0,
+                    1931.0,
+                    293.0,
+                    1931.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1925.0,
+                    1406.0,
+                    1925.0,
+                    1406.0,
+                    1962.0,
+                    292.0,
+                    1962.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1957.0,
+                    1372.0,
+                    1957.0,
+                    1372.0,
+                    1991.0,
+                    293.0,
+                    1991.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    508.0,
+                    466.0,
+                    1190.0,
+                    466.0,
+                    1190.0,
+                    513.0,
+                    508.0,
+                    513.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    703.0,
+                    502.0,
+                    997.0,
+                    502.0,
+                    997.0,
+                    540.0,
+                    703.0,
+                    540.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    498.0,
+                    534.0,
+                    1187.0,
+                    534.0,
+                    1187.0,
+                    572.0,
+                    498.0,
+                    572.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    486.0,
+                    440.0,
+                    1213.0,
+                    440.0,
+                    1213.0,
+                    485.0,
+                    486.0,
+                    485.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    2029.0,
+                    1075.0,
+                    2029.0,
+                    1075.0,
+                    2067.0,
+                    294.0,
+                    2067.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 0,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 5,
+                "poly": [
+                    308,
+                    1182,
+                    1392,
+                    1182,
+                    1392,
+                    1375,
+                    308,
+                    1375
+                ],
+                "score": 0.98,
+                "html": "<table><tr><td rowspan=1 colspan=1></td><td rowspan=1 colspan=1>Upper bound (poly-time)</td><td rowspan=1 colspan=1>Upper bound (exp-time)</td><td rowspan=1 colspan=1>Lower bound</td></tr><tr><td rowspan=1 colspan=1>(ε,δ)-DP[52]</td><td rowspan=1 colspan=1>0+dlog1/2(1/8））Qε</td><td rowspan=1 colspan=1>+）</td><td rowspan=1 colspan=1>（+）</td></tr><tr><td rowspan=1 colspan=1>α-corruption [36]</td><td rowspan=1 colspan=1>0（</td><td rowspan=1 colspan=1>（</td><td rowspan=1 colspan=1>（</td></tr><tr><td rowspan=1 colspan=1>α-corruption and(ε,δ)-DP (this paper)</td><td rowspan=1 colspan=1>（+/210g(1/8）)αε[Theorem6]</td><td rowspan=1 colspan=1>0（+d+d1/210g(1/8))Q[Theorem7]</td><td rowspan=1 colspan=1>(）[52]</td></tr></table>"
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    296,
+                    837,
+                    1406,
+                    837,
+                    1406,
+                    1075,
+                    296,
+                    1075
+                ],
+                "score": 0.98
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    459,
+                    1405,
+                    459,
+                    1405,
+                    679,
+                    297,
+                    679
+                ],
+                "score": 0.979
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    201,
+                    1405,
+                    201,
+                    1405,
+                    448,
+                    297,
+                    448
+                ],
+                "score": 0.978
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1568,
+                    1406,
+                    1568,
+                    1406,
+                    1695,
+                    298,
+                    1695
+                ],
+                "score": 0.975
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    684,
+                    1406,
+                    684,
+                    1406,
+                    822,
+                    297,
+                    822
+                ],
+                "score": 0.972
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1405,
+                    1405,
+                    1405,
+                    1405,
+                    1545,
+                    297,
+                    1545
+                ],
+                "score": 0.972
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1702,
+                    1407,
+                    1702,
+                    1407,
+                    1806,
+                    297,
+                    1806
+                ],
+                "score": 0.966
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1822,
+                    1399,
+                    1822,
+                    1399,
+                    1920,
+                    298,
+                    1920
+                ],
+                "score": 0.954
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    294,
+                    1923,
+                    1400,
+                    1923,
+                    1400,
+                    1997,
+                    294,
+                    1997
+                ],
+                "score": 0.945
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1078,
+                    1401,
+                    1078,
+                    1401,
+                    1152,
+                    298,
+                    1152
+                ],
+                "score": 0.935
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    840,
+                    2061,
+                    859,
+                    2061,
+                    859,
+                    2085,
+                    840,
+                    2085
+                ],
+                "score": 0.726
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    840,
+                    2061,
+                    860,
+                    2061,
+                    860,
+                    2085,
+                    840,
+                    2085
+                ],
+                "score": 0.216
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    760,
+                    969,
+                    853,
+                    969,
+                    853,
+                    1009,
+                    760,
+                    1009
+                ],
+                "score": 0.94,
+                "latex": "\\widetilde \\Omega ( d ^ { 3 / 2 } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1019,
+                    1660,
+                    1114,
+                    1660,
+                    1114,
+                    1698,
+                    1019,
+                    1698
+                ],
+                "score": 0.93,
+                "latex": "\\Omega ( \\alpha ^ { 1 / 2 } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1179,
+                    869,
+                    1326,
+                    869,
+                    1326,
+                    909,
+                    1179,
+                    909
+                ],
+                "score": 0.93,
+                "latex": "\\widetilde { \\cal O } ( d ^ { 3 } + n d ^ { 2 } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1196,
+                    1439,
+                    1404,
+                    1439,
+                    1404,
+                    1476,
+                    1196,
+                    1476
+                ],
+                "score": 0.93,
+                "latex": "\\| \\hat { \\mu } - \\mu \\| _ { 2 } = \\widetilde { O } ( \\alpha )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    627,
+                    1507,
+                    748,
+                    1507,
+                    748,
+                    1545,
+                    627,
+                    1545
+                ],
+                "score": 0.93,
+                "latex": "\\delta \\leq \\sqrt { d } / n"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1245,
+                    1006,
+                    1403,
+                    1006,
+                    1403,
+                    1046,
+                    1245,
+                    1046
+                ],
+                "score": 0.93,
+                "latex": "\\widetilde \\Omega ( d ^ { 3 / 2 } / ( \\alpha \\varepsilon ) )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    431,
+                    555,
+                    621,
+                    555,
+                    621,
+                    592,
+                    431,
+                    592
+                ],
+                "score": 0.93,
+                "latex": "O ( \\alpha \\sqrt { \\log ( 1 / \\alpha ) } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    666,
+                    1010,
+                    851,
+                    1010,
+                    851,
+                    1046,
+                    666,
+                    1046
+                ],
+                "score": 0.92,
+                "latex": "\\| x _ { i } \\| _ { 2 } = O ( { \\sqrt { d } } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1168,
+                    555,
+                    1229,
+                    555,
+                    1229,
+                    590,
+                    1168,
+                    590
+                ],
+                "score": 0.92,
+                "latex": "\\Omega ( \\alpha )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    429,
+                    1110,
+                    862,
+                    1110,
+                    862,
+                    1150,
+                    429,
+                    1150
+                ],
+                "score": 0.92,
+                "latex": "\\begin{array} { r } { \\dot { n } = \\widetilde \\Omega ( d / \\alpha ^ { 2 } + ( d ^ { 3 / 2 } \\log ( 1 / \\delta ) ) / ( \\alpha \\varepsilon ) ) . } \\end{array}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    660,
+                    1405,
+                    747,
+                    1405,
+                    747,
+                    1440,
+                    660,
+                    1440
+                ],
+                "score": 0.92,
+                "latex": "\\mu \\in \\mathbb { R } ^ { d }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1261,
+                    1701,
+                    1351,
+                    1701,
+                    1351,
+                    1734,
+                    1261,
+                    1734
+                ],
+                "score": 0.92,
+                "latex": "\\mu \\in \\mathbb { R } ^ { d }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    368,
+                    1476,
+                    429,
+                    1476,
+                    429,
+                    1510,
+                    368,
+                    1510
+                ],
+                "score": 0.91,
+                "latex": "( \\varepsilon , \\delta )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    595,
+                    783,
+                    997,
+                    783,
+                    997,
+                    822,
+                    595,
+                    822
+                ],
+                "score": 0.91,
+                "latex": "\\| \\hat { \\mu } - \\mu \\| _ { 2 } = O ( \\alpha \\sqrt { \\log ( 1 / \\alpha ) } ) \\ : w . h . p"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    391,
+                    521,
+                    452,
+                    521,
+                    452,
+                    555,
+                    391,
+                    555
+                ],
+                "score": 0.91,
+                "latex": "( \\varepsilon , \\delta )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    976,
+                    1112,
+                    1379,
+                    1112,
+                    1379,
+                    1151,
+                    976,
+                    1151
+                ],
+                "score": 0.91,
+                "latex": "\\begin{array} { r } { \\| \\hat { \\mu } - \\mu \\| _ { 2 } = O ( \\alpha \\sqrt { \\log ( 1 / \\alpha ) } ) ~ w . h . p . } \\end{array}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1038,
+                    1768,
+                    1345,
+                    1768,
+                    1345,
+                    1805,
+                    1038,
+                    1805
+                ],
+                "score": 0.91,
+                "latex": "\\| \\hat { \\mu } - \\mu \\| _ { 2 } = O ( \\alpha ^ { 1 / 2 } ) \\ : w . h . p ."
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    688,
+                    1958,
+                    995,
+                    1958,
+                    995,
+                    1996,
+                    688,
+                    1996
+                ],
+                "score": 0.91,
+                "latex": "\\| \\hat { \\mu } - \\mu \\| _ { 2 } = O ( \\alpha ^ { 1 / 2 } ) \\ : w . h . p"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1068,
+                    686,
+                    1129,
+                    686,
+                    1129,
+                    718,
+                    1068,
+                    718
+                ],
+                "score": 0.9,
+                "latex": "( \\varepsilon , \\delta )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    924,
+                    746,
+                    1402,
+                    746,
+                    1402,
+                    785,
+                    924,
+                    785
+                ],
+                "score": 0.9,
+                "latex": "\\begin{array} { r } { n = \\widetilde \\Omega ( d / \\alpha ^ { 2 } + ( d + d ^ { 1 / 2 } \\log ( 1 / \\delta ) ) / ( \\alpha \\varepsilon ) ) } \\end{array}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    297,
+                    1955,
+                    580,
+                    1955,
+                    580,
+                    1996,
+                    297,
+                    1996
+                ],
+                "score": 0.9,
+                "latex": "\\widetilde \\Omega ( ( d ^ { 3 / 2 } \\log ( 1 / \\delta ) ) / ( \\alpha \\varepsilon ) )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    651,
+                    1853,
+                    702,
+                    1853,
+                    702,
+                    1884,
+                    651,
+                    1884
+                ],
+                "score": 0.9,
+                "latex": "d ^ { 1 / 2 }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    587,
+                    906,
+                    639,
+                    906,
+                    639,
+                    937,
+                    587,
+                    937
+                ],
+                "score": 0.9,
+                "latex": "d ^ { 1 / 2 }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1002,
+                    1079,
+                    1063,
+                    1079,
+                    1063,
+                    1113,
+                    1002,
+                    1113
+                ],
+                "score": 0.89,
+                "latex": "( \\varepsilon , \\delta )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1224,
+                    1735,
+                    1284,
+                    1735,
+                    1284,
+                    1767,
+                    1224,
+                    1767
+                ],
+                "score": 0.89,
+                "latex": "( \\varepsilon , \\delta )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    488,
+                    1012,
+                    550,
+                    1012,
+                    550,
+                    1046,
+                    488,
+                    1046
+                ],
+                "score": 0.89,
+                "latex": "( \\varepsilon , \\delta )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1212,
+                    1304,
+                    1371,
+                    1304,
+                    1371,
+                    1343,
+                    1212,
+                    1343
+                ],
+                "score": 0.89,
+                "latex": "\\begin{array} { r } { \\widetilde \\Omega ( \\frac { d } { \\alpha ^ { 2 } } + \\frac { d } { \\alpha \\varepsilon } ) ^ { \\star } } \\end{array}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    297,
+                    1766,
+                    930,
+                    1766,
+                    930,
+                    1805,
+                    297,
+                    1805
+                ],
+                "score": 0.89,
+                "latex": "n = \\widetilde \\Omega ( ( d + d ^ { 1 / 2 } \\log ( 1 / \\delta ) ) / ( \\alpha \\varepsilon ) + d ^ { 1 / 2 } \\log ^ { 3 / 2 } ( 1 / \\delta ) / \\varepsilon )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    909,
+                    1298,
+                    1181,
+                    1298,
+                    1181,
+                    1344,
+                    909,
+                    1344
+                ],
+                "score": 0.88,
+                "latex": "\\begin{array} { r } { \\widetilde { O } ( \\frac { d } { \\alpha ^ { 2 } } + \\frac { d + d ^ { 1 / 2 } \\log ( 1 / \\delta ) } { \\alpha \\varepsilon } ) } \\end{array}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    425,
+                    1735,
+                    497,
+                    1735,
+                    497,
+                    1764,
+                    425,
+                    1764
+                ],
+                "score": 0.87,
+                "latex": "\\Sigma \\preceq \\mathbf { I }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1254,
+                    1264,
+                    1329,
+                    1264,
+                    1329,
+                    1299,
+                    1254,
+                    1299
+                ],
+                "score": 0.87,
+                "latex": "\\Omega ( \\frac { d } { \\alpha ^ { 2 } } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1212,
+                    1223,
+                    1370,
+                    1223,
+                    1370,
+                    1262,
+                    1212,
+                    1262
+                ],
+                "score": 0.85,
+                "latex": "\\begin{array} { r } { \\widetilde \\Omega ( \\frac { d } { \\alpha ^ { 2 } } + \\frac { d } { \\alpha \\varepsilon } ) ^ { \\star } } \\end{array}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1010,
+                    840,
+                    1063,
+                    840,
+                    1063,
+                    871,
+                    1010,
+                    871
+                ],
+                "score": 0.85,
+                "latex": "\\ S 2 . 3"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1092,
+                    588,
+                    1143,
+                    588,
+                    1143,
+                    620,
+                    1092,
+                    620
+                ],
+                "score": 0.85,
+                "latex": "( i i i )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    328,
+                    1338,
+                    388,
+                    1338,
+                    388,
+                    1371,
+                    328,
+                    1371
+                ],
+                "score": 0.84,
+                "latex": "( \\varepsilon , \\delta )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1337,
+                    1926,
+                    1403,
+                    1926,
+                    1403,
+                    1957,
+                    1337,
+                    1957
+                ],
+                "score": 0.84,
+                "latex": "n \\_ ="
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    666,
+                    1482,
+                    687,
+                    1482,
+                    687,
+                    1504,
+                    666,
+                    1504
+                ],
+                "score": 0.81,
+                "latex": "\\alpha"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    611,
+                    1298,
+                    863,
+                    1298,
+                    863,
+                    1344,
+                    611,
+                    1344
+                ],
+                "score": 0.8,
+                "latex": "\\begin{array} { r } { \\widetilde { O } \\big ( \\frac { d } { \\alpha ^ { 2 } } + \\frac { d ^ { 3 / 2 } \\log ( 1 / \\delta ) } { \\alpha \\varepsilon _ { - } } \\big ) } \\end{array}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    979,
+                    526,
+                    1000,
+                    526,
+                    1000,
+                    550,
+                    979,
+                    550
+                ],
+                "score": 0.79,
+                "latex": "\\alpha"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    792,
+                    1738,
+                    813,
+                    1738,
+                    813,
+                    1763,
+                    792,
+                    1763
+                ],
+                "score": 0.78,
+                "latex": "\\alpha"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    504,
+                    521,
+                    545,
+                    521,
+                    545,
+                    554,
+                    504,
+                    554
+                ],
+                "score": 0.78,
+                "latex": "( i i )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    349,
+                    1277,
+                    368,
+                    1277,
+                    368,
+                    1290,
+                    349,
+                    1290
+                ],
+                "score": 0.77,
+                "latex": "\\alpha"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    964,
+                    1220,
+                    1125,
+                    1220,
+                    1125,
+                    1263,
+                    964,
+                    1263
+                ],
+                "score": 0.75,
+                "latex": "\\begin{array} { r } { \\widetilde { O } ( \\frac { d } { \\alpha ^ { 2 } } + \\frac { d } { \\alpha \\varepsilon } ) ^ { \\bullet } } \\end{array}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1006,
+                    1263,
+                    1083,
+                    1263,
+                    1083,
+                    1297,
+                    1006,
+                    1297
+                ],
+                "score": 0.74,
+                "latex": "{ \\textstyle \\ddot { O } } ( { \\frac { d } { \\alpha ^ { 2 } } } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    351,
+                    1318,
+                    371,
+                    1318,
+                    371,
+                    1334,
+                    351,
+                    1334
+                ],
+                "score": 0.74,
+                "latex": "\\alpha"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    617,
+                    1216,
+                    859,
+                    1216,
+                    859,
+                    1262,
+                    617,
+                    1262
+                ],
+                "score": 0.74,
+                "latex": "\\begin{array} { r } { \\overline { { \\widetilde { \\cal O } ( \\frac { d } { \\alpha ^ { 2 } } + \\frac { d \\log ^ { 1 / 2 } ( 1 / \\delta ) } { \\alpha \\varepsilon } ) } } } \\end{array}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    934,
+                    718,
+                    952,
+                    718,
+                    952,
+                    743,
+                    934,
+                    743
+                ],
+                "score": 0.71,
+                "latex": "d"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1132,
+                    1924,
+                    1244,
+                    1924,
+                    1244,
+                    1959,
+                    1132,
+                    1959
+                ],
+                "score": 0.66,
+                "latex": "( \\varepsilon , \\delta ) – D P"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1255,
+                    690,
+                    1277,
+                    690,
+                    1277,
+                    715,
+                    1255,
+                    715
+                ],
+                "score": 0.66,
+                "latex": "\\alpha"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    699,
+                    1263,
+                    776,
+                    1263,
+                    776,
+                    1297,
+                    699,
+                    1297
+                ],
+                "score": 0.65,
+                "latex": "{ \\textstyle \\ddot { O } } ( { \\frac { d } { \\alpha ^ { 2 } } } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    419,
+                    758,
+                    440,
+                    758,
+                    440,
+                    784,
+                    419,
+                    784
+                ],
+                "score": 0.63,
+                "latex": "\\mu"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1382,
+                    1736,
+                    1406,
+                    1736,
+                    1406,
+                    1767,
+                    1382,
+                    1767
+                ],
+                "score": 0.6,
+                "latex": "i f"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    505,
+                    1509,
+                    528,
+                    1509,
+                    528,
+                    1539,
+                    505,
+                    1539
+                ],
+                "score": 0.4,
+                "latex": "\\spadesuit"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    368,
+                    1229,
+                    429,
+                    1229,
+                    429,
+                    1260,
+                    368,
+                    1260
+                ],
+                "score": 0.35,
+                "latex": "( \\varepsilon , \\delta )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1028,
+                    1474,
+                    1051,
+                    1474,
+                    1051,
+                    1503,
+                    1028,
+                    1503
+                ],
+                "score": 0.3,
+                "latex": "\\clubsuit"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    729,
+                    722,
+                    749,
+                    722,
+                    749,
+                    743,
+                    729,
+                    743
+                ],
+                "score": 0.3,
+                "latex": "n"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    507,
+                    1740,
+                    528,
+                    1740,
+                    528,
+                    1762,
+                    507,
+                    1762
+                ],
+                "score": 0.26,
+                "latex": "n"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    838.0,
+                    2058.0,
+                    863.0,
+                    2058.0,
+                    863.0,
+                    2093.0,
+                    838.0,
+                    2093.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    838.0,
+                    2058.0,
+                    862.0,
+                    2058.0,
+                    862.0,
+                    2093.0,
+                    838.0,
+                    2093.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    836.0,
+                    1009.0,
+                    836.0,
+                    1009.0,
+                    874.0,
+                    293.0,
+                    874.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1064.0,
+                    836.0,
+                    1406.0,
+                    836.0,
+                    1406.0,
+                    874.0,
+                    1064.0,
+                    874.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    874.0,
+                    1178.0,
+                    874.0,
+                    1178.0,
+                    910.0,
+                    294.0,
+                    910.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1327.0,
+                    874.0,
+                    1407.0,
+                    874.0,
+                    1407.0,
+                    910.0,
+                    1327.0,
+                    910.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    903.0,
+                    586.0,
+                    903.0,
+                    586.0,
+                    948.0,
+                    293.0,
+                    948.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    640.0,
+                    903.0,
+                    1406.0,
+                    903.0,
+                    1406.0,
+                    948.0,
+                    640.0,
+                    948.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    939.0,
+                    1406.0,
+                    939.0,
+                    1406.0,
+                    976.0,
+                    294.0,
+                    976.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    290.0,
+                    968.0,
+                    759.0,
+                    968.0,
+                    759.0,
+                    1017.0,
+                    290.0,
+                    1017.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    854.0,
+                    968.0,
+                    1407.0,
+                    968.0,
+                    1407.0,
+                    1017.0,
+                    854.0,
+                    1017.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1005.0,
+                    487.0,
+                    1005.0,
+                    487.0,
+                    1052.0,
+                    291.0,
+                    1052.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    551.0,
+                    1005.0,
+                    665.0,
+                    1005.0,
+                    665.0,
+                    1052.0,
+                    551.0,
+                    1052.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    852.0,
+                    1005.0,
+                    1244.0,
+                    1005.0,
+                    1244.0,
+                    1052.0,
+                    852.0,
+                    1052.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1404.0,
+                    1005.0,
+                    1408.0,
+                    1005.0,
+                    1408.0,
+                    1052.0,
+                    1404.0,
+                    1052.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1040.0,
+                    1106.0,
+                    1040.0,
+                    1106.0,
+                    1079.0,
+                    294.0,
+                    1079.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    458.0,
+                    1406.0,
+                    458.0,
+                    1406.0,
+                    495.0,
+                    294.0,
+                    495.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    491.0,
+                    1405.0,
+                    491.0,
+                    1405.0,
+                    527.0,
+                    295.0,
+                    527.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    518.0,
+                    390.0,
+                    518.0,
+                    390.0,
+                    560.0,
+                    292.0,
+                    560.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    453.0,
+                    518.0,
+                    503.0,
+                    518.0,
+                    503.0,
+                    560.0,
+                    453.0,
+                    560.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    546.0,
+                    518.0,
+                    978.0,
+                    518.0,
+                    978.0,
+                    560.0,
+                    546.0,
+                    560.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1001.0,
+                    518.0,
+                    1406.0,
+                    518.0,
+                    1406.0,
+                    560.0,
+                    1001.0,
+                    560.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    557.0,
+                    430.0,
+                    557.0,
+                    430.0,
+                    591.0,
+                    295.0,
+                    591.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    622.0,
+                    557.0,
+                    1167.0,
+                    557.0,
+                    1167.0,
+                    591.0,
+                    622.0,
+                    591.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1230.0,
+                    557.0,
+                    1405.0,
+                    557.0,
+                    1405.0,
+                    591.0,
+                    1230.0,
+                    591.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    588.0,
+                    1091.0,
+                    588.0,
+                    1091.0,
+                    623.0,
+                    295.0,
+                    623.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1144.0,
+                    588.0,
+                    1405.0,
+                    588.0,
+                    1405.0,
+                    623.0,
+                    1144.0,
+                    623.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    617.0,
+                    1403.0,
+                    617.0,
+                    1403.0,
+                    651.0,
+                    295.0,
+                    651.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    648.0,
+                    653.0,
+                    648.0,
+                    653.0,
+                    679.0,
+                    295.0,
+                    679.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    204.0,
+                    1405.0,
+                    204.0,
+                    1405.0,
+                    238.0,
+                    297.0,
+                    238.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    235.0,
+                    1403.0,
+                    235.0,
+                    1403.0,
+                    265.0,
+                    296.0,
+                    265.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    263.0,
+                    1407.0,
+                    263.0,
+                    1407.0,
+                    299.0,
+                    294.0,
+                    299.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    296.0,
+                    1405.0,
+                    296.0,
+                    1405.0,
+                    329.0,
+                    295.0,
+                    329.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    325.0,
+                    1406.0,
+                    325.0,
+                    1406.0,
+                    362.0,
+                    295.0,
+                    362.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    353.0,
+                    1405.0,
+                    353.0,
+                    1405.0,
+                    390.0,
+                    294.0,
+                    390.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    385.0,
+                    1403.0,
+                    385.0,
+                    1403.0,
+                    418.0,
+                    295.0,
+                    418.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    416.0,
+                    1397.0,
+                    416.0,
+                    1397.0,
+                    449.0,
+                    292.0,
+                    449.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1566.0,
+                    1409.0,
+                    1566.0,
+                    1409.0,
+                    1606.0,
+                    293.0,
+                    1606.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1600.0,
+                    1409.0,
+                    1600.0,
+                    1409.0,
+                    1635.0,
+                    292.0,
+                    1635.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1632.0,
+                    1404.0,
+                    1632.0,
+                    1404.0,
+                    1663.0,
+                    293.0,
+                    1663.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1661.0,
+                    1018.0,
+                    1661.0,
+                    1018.0,
+                    1698.0,
+                    292.0,
+                    1698.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1115.0,
+                    1661.0,
+                    1126.0,
+                    1661.0,
+                    1126.0,
+                    1698.0,
+                    1115.0,
+                    1698.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    681.0,
+                    1067.0,
+                    681.0,
+                    1067.0,
+                    722.0,
+                    292.0,
+                    722.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1130.0,
+                    681.0,
+                    1254.0,
+                    681.0,
+                    1254.0,
+                    722.0,
+                    1130.0,
+                    722.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1278.0,
+                    681.0,
+                    1408.0,
+                    681.0,
+                    1408.0,
+                    722.0,
+                    1278.0,
+                    722.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    715.0,
+                    728.0,
+                    715.0,
+                    728.0,
+                    749.0,
+                    295.0,
+                    749.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    750.0,
+                    715.0,
+                    933.0,
+                    715.0,
+                    933.0,
+                    749.0,
+                    750.0,
+                    749.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    953.0,
+                    715.0,
+                    1404.0,
+                    715.0,
+                    1404.0,
+                    749.0,
+                    953.0,
+                    749.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    747.0,
+                    418.0,
+                    747.0,
+                    418.0,
+                    789.0,
+                    294.0,
+                    789.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    441.0,
+                    747.0,
+                    923.0,
+                    747.0,
+                    923.0,
+                    789.0,
+                    441.0,
+                    789.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1403.0,
+                    747.0,
+                    1407.0,
+                    747.0,
+                    1407.0,
+                    789.0,
+                    1403.0,
+                    789.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    784.0,
+                    594.0,
+                    784.0,
+                    594.0,
+                    824.0,
+                    292.0,
+                    824.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    998.0,
+                    784.0,
+                    1006.0,
+                    784.0,
+                    1006.0,
+                    824.0,
+                    998.0,
+                    824.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1404.0,
+                    659.0,
+                    1404.0,
+                    659.0,
+                    1444.0,
+                    292.0,
+                    1444.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    748.0,
+                    1404.0,
+                    1406.0,
+                    1404.0,
+                    1406.0,
+                    1444.0,
+                    748.0,
+                    1444.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1440.0,
+                    1195.0,
+                    1440.0,
+                    1195.0,
+                    1479.0,
+                    294.0,
+                    1479.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1474.0,
+                    367.0,
+                    1474.0,
+                    367.0,
+                    1513.0,
+                    296.0,
+                    1513.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    430.0,
+                    1474.0,
+                    665.0,
+                    1474.0,
+                    665.0,
+                    1513.0,
+                    430.0,
+                    1513.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    688.0,
+                    1474.0,
+                    1027.0,
+                    1474.0,
+                    1027.0,
+                    1513.0,
+                    688.0,
+                    1513.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1052.0,
+                    1474.0,
+                    1408.0,
+                    1474.0,
+                    1408.0,
+                    1513.0,
+                    1052.0,
+                    1513.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1506.0,
+                    504.0,
+                    1506.0,
+                    504.0,
+                    1548.0,
+                    295.0,
+                    1548.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    529.0,
+                    1506.0,
+                    626.0,
+                    1506.0,
+                    626.0,
+                    1548.0,
+                    529.0,
+                    1548.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    749.0,
+                    1506.0,
+                    759.0,
+                    1506.0,
+                    759.0,
+                    1548.0,
+                    749.0,
+                    1548.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1699.0,
+                    1260.0,
+                    1699.0,
+                    1260.0,
+                    1737.0,
+                    294.0,
+                    1737.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1352.0,
+                    1699.0,
+                    1408.0,
+                    1699.0,
+                    1408.0,
+                    1737.0,
+                    1352.0,
+                    1737.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1734.0,
+                    424.0,
+                    1734.0,
+                    424.0,
+                    1769.0,
+                    295.0,
+                    1769.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    498.0,
+                    1734.0,
+                    506.0,
+                    1734.0,
+                    506.0,
+                    1769.0,
+                    498.0,
+                    1769.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    529.0,
+                    1734.0,
+                    791.0,
+                    1734.0,
+                    791.0,
+                    1769.0,
+                    529.0,
+                    1769.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    814.0,
+                    1734.0,
+                    1223.0,
+                    1734.0,
+                    1223.0,
+                    1769.0,
+                    814.0,
+                    1769.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1285.0,
+                    1734.0,
+                    1381.0,
+                    1734.0,
+                    1381.0,
+                    1769.0,
+                    1285.0,
+                    1769.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1763.0,
+                    296.0,
+                    1763.0,
+                    296.0,
+                    1810.0,
+                    291.0,
+                    1810.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    931.0,
+                    1763.0,
+                    1037.0,
+                    1763.0,
+                    1037.0,
+                    1810.0,
+                    931.0,
+                    1810.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1346.0,
+                    1763.0,
+                    1356.0,
+                    1763.0,
+                    1356.0,
+                    1810.0,
+                    1346.0,
+                    1810.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1819.0,
+                    1404.0,
+                    1819.0,
+                    1404.0,
+                    1862.0,
+                    292.0,
+                    1862.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1852.0,
+                    650.0,
+                    1852.0,
+                    650.0,
+                    1891.0,
+                    293.0,
+                    1891.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    703.0,
+                    1852.0,
+                    1406.0,
+                    1852.0,
+                    1406.0,
+                    1891.0,
+                    703.0,
+                    1891.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1888.0,
+                    1077.0,
+                    1888.0,
+                    1077.0,
+                    1923.0,
+                    294.0,
+                    1923.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1922.0,
+                    1131.0,
+                    1922.0,
+                    1131.0,
+                    1960.0,
+                    296.0,
+                    1960.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1245.0,
+                    1922.0,
+                    1336.0,
+                    1922.0,
+                    1336.0,
+                    1960.0,
+                    1245.0,
+                    1960.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1953.0,
+                    296.0,
+                    1953.0,
+                    296.0,
+                    2002.0,
+                    293.0,
+                    2002.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    581.0,
+                    1953.0,
+                    687.0,
+                    1953.0,
+                    687.0,
+                    2002.0,
+                    581.0,
+                    2002.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    996.0,
+                    1953.0,
+                    1381.0,
+                    1953.0,
+                    1381.0,
+                    2002.0,
+                    996.0,
+                    2002.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1076.0,
+                    1001.0,
+                    1076.0,
+                    1001.0,
+                    1116.0,
+                    294.0,
+                    1116.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1064.0,
+                    1076.0,
+                    1405.0,
+                    1076.0,
+                    1405.0,
+                    1116.0,
+                    1064.0,
+                    1116.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1111.0,
+                    428.0,
+                    1111.0,
+                    428.0,
+                    1156.0,
+                    294.0,
+                    1156.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    863.0,
+                    1111.0,
+                    975.0,
+                    1111.0,
+                    975.0,
+                    1156.0,
+                    863.0,
+                    1156.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1380.0,
+                    1111.0,
+                    1387.0,
+                    1111.0,
+                    1387.0,
+                    1156.0,
+                    1380.0,
+                    1156.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 1,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    850,
+                    1404,
+                    850,
+                    1404,
+                    1072,
+                    298,
+                    1072
+                ],
+                "score": 0.981
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    622,
+                    1406,
+                    622,
+                    1406,
+                    839,
+                    297,
+                    839
+                ],
+                "score": 0.98
+            },
+            {
+                "category_id": 5,
+                "poly": [
+                    314,
+                    197,
+                    1386,
+                    197,
+                    1386,
+                    389,
+                    314,
+                    389
+                ],
+                "score": 0.98,
+                "html": "<table><tr><td rowspan=1 colspan=1></td><td rowspan=1 colspan=1>Upper bound (poly-time)</td><td rowspan=1 colspan=1>Upper bound (exp-time)</td><td rowspan=1 colspan=1>Lower bound</td></tr><tr><td rowspan=1 colspan=1>(ε,δ)-DP [54]</td><td rowspan=1 colspan=1>O（d10g1/2（1/）Qε</td><td rowspan=1 colspan=1>O（d10g1/2（1/0））Qε</td><td rowspan=1 colspan=1>2（）</td></tr><tr><td rowspan=1 colspan=1>α-corruption [36]</td><td rowspan=1 colspan=1>）</td><td rowspan=1 colspan=1></td><td rowspan=1 colspan=1>(）</td></tr><tr><td rowspan=1 colspan=1>α-corruption and(ε,δ)-DP (this paper)</td><td rowspan=1 colspan=1>0（(/210g(1/8))αε[Theorem 9]</td><td rowspan=1 colspan=1>（+d1721g/2(18）αε[Theorem 8]</td><td rowspan=1 colspan=1>(([54])</td></tr></table>"
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1085,
+                    1404,
+                    1085,
+                    1404,
+                    1271,
+                    297,
+                    1271
+                ],
+                "score": 0.98
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1402,
+                    1404,
+                    1402,
+                    1404,
+                    1535,
+                    298,
+                    1535
+                ],
+                "score": 0.972
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1629,
+                    1404,
+                    1629,
+                    1404,
+                    1774,
+                    297,
+                    1774
+                ],
+                "score": 0.967
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    295,
+                    1791,
+                    1401,
+                    1791,
+                    1401,
+                    1855,
+                    295,
+                    1855
+                ],
+                "score": 0.95
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1944,
+                    1399,
+                    1944,
+                    1399,
+                    2008,
+                    298,
+                    2008
+                ],
+                "score": 0.949
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    296,
+                    1554,
+                    1400,
+                    1554,
+                    1400,
+                    1622,
+                    296,
+                    1622
+                ],
+                "score": 0.935
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    300,
+                    1888,
+                    612,
+                    1888,
+                    612,
+                    1921,
+                    300,
+                    1921
+                ],
+                "score": 0.925
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    299,
+                    1305,
+                    840,
+                    1305,
+                    840,
+                    1339,
+                    299,
+                    1339
+                ],
+                "score": 0.924
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    299,
+                    567,
+                    643,
+                    567,
+                    643,
+                    600,
+                    299,
+                    600
+                ],
+                "score": 0.918
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1361,
+                    1386,
+                    1361,
+                    1386,
+                    1394,
+                    297,
+                    1394
+                ],
+                "score": 0.891
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    419,
+                    1405,
+                    419,
+                    1405,
+                    517,
+                    299,
+                    517
+                ],
+                "score": 0.888
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    841,
+                    2061,
+                    859,
+                    2061,
+                    859,
+                    2085,
+                    841,
+                    2085
+                ],
+                "score": 0.637
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    841,
+                    2061,
+                    859,
+                    2061,
+                    859,
+                    2085,
+                    841,
+                    2085
+                ],
+                "score": 0.489
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    506,
+                    1588,
+                    656,
+                    1588,
+                    656,
+                    1623,
+                    506,
+                    1623
+                ],
+                "score": 0.93,
+                "latex": "z \\sim \\mathcal { N } ( \\mu , \\Sigma )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    359,
+                    882,
+                    442,
+                    882,
+                    442,
+                    920,
+                    359,
+                    920
+                ],
+                "score": 0.93,
+                "latex": "O ( { \\sqrt { d } } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    979,
+                    452,
+                    1221,
+                    452,
+                    1221,
+                    488,
+                    979,
+                    488
+                ],
+                "score": 0.93,
+                "latex": "\\| \\hat { \\mu } - \\mu \\| _ { 2 } = O ( \\alpha ^ { 1 / 2 } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    880,
+                    1399,
+                    1035,
+                    1399,
+                    1035,
+                    1439,
+                    880,
+                    1439
+                ],
+                "score": 0.92,
+                "latex": "S ^ { \\prime } = \\{ x _ { i } ^ { \\prime } \\} _ { i = 1 } ^ { n ^ { \\prime } }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    297,
+                    1665,
+                    391,
+                    1665,
+                    391,
+                    1701,
+                    297,
+                    1701
+                ],
+                "score": 0.92,
+                "latex": "f ( S ^ { \\prime } ) \\| _ { p }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    668,
+                    419,
+                    755,
+                    419,
+                    755,
+                    454,
+                    668,
+                    454
+                ],
+                "score": 0.92,
+                "latex": "\\mu \\in \\mathbb { R } ^ { d }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1046,
+                    1502,
+                    1130,
+                    1502,
+                    1130,
+                    1530,
+                    1046,
+                    1530
+                ],
+                "score": 0.92,
+                "latex": "S \\sim S ^ { \\prime }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    542,
+                    1664,
+                    846,
+                    1664,
+                    846,
+                    1704,
+                    542,
+                    1704
+                ],
+                "score": 0.92,
+                "latex": "\\begin{array} { r } { \\| { \\boldsymbol x } \\| _ { p } ~ = ~ ( \\sum _ { i \\in [ k ] } | x _ { i } | ^ { p } ) ^ { 1 / p } } \\end{array}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    683,
+                    1038,
+                    743,
+                    1038,
+                    743,
+                    1072,
+                    683,
+                    1072
+                ],
+                "score": 0.92,
+                "latex": "O ( d )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    797,
+                    1628,
+                    940,
+                    1628,
+                    940,
+                    1664,
+                    797,
+                    1664
+                ],
+                "score": 0.92,
+                "latex": "f ( S ) ~ \\in ~ \\mathbb { R } ^ { k }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    776,
+                    1037,
+                    905,
+                    1037,
+                    905,
+                    1072,
+                    776,
+                    1072
+                ],
+                "score": 0.92,
+                "latex": "O ( ( \\log d ) ^ { 2 } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1254,
+                    241,
+                    1328,
+                    241,
+                    1328,
+                    276,
+                    1254,
+                    276
+                ],
+                "score": 0.91,
+                "latex": "\\textstyle \\Omega ( { \\frac { d } { \\alpha \\varepsilon } } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    684,
+                    1404,
+                    830,
+                    1404,
+                    830,
+                    1439,
+                    684,
+                    1439
+                ],
+                "score": 0.91,
+                "latex": "S = \\{ x _ { i } \\} _ { i = 1 } ^ { n }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1087,
+                    851,
+                    1148,
+                    851,
+                    1148,
+                    885,
+                    1087,
+                    885
+                ],
+                "score": 0.91,
+                "latex": "O ( d )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1260,
+                    279,
+                    1323,
+                    279,
+                    1323,
+                    313,
+                    1260,
+                    313
+                ],
+                "score": 0.91,
+                "latex": "\\Omega ( \\frac { d } { \\alpha } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1010,
+                    1471,
+                    1071,
+                    1471,
+                    1071,
+                    1503,
+                    1010,
+                    1503
+                ],
+                "score": 0.91,
+                "latex": "( \\varepsilon , \\delta )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    891,
+                    625,
+                    952,
+                    625,
+                    952,
+                    657,
+                    891,
+                    657
+                ],
+                "score": 0.91,
+                "latex": "( \\varepsilon , \\delta )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    341,
+                    1555,
+                    477,
+                    1555,
+                    477,
+                    1587,
+                    341,
+                    1587
+                ],
+                "score": 0.91,
+                "latex": "z \\sim \\mathrm { L a p } ( b )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    297,
+                    1439,
+                    487,
+                    1439,
+                    487,
+                    1474,
+                    297,
+                    1474
+                ],
+                "score": 0.91,
+                "latex": "i f d _ { \\triangle } ( S , S ^ { \\prime } ) \\leq 1"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    656,
+                    232,
+                    832,
+                    232,
+                    832,
+                    278,
+                    656,
+                    278
+                ],
+                "score": 0.91,
+                "latex": "\\widetilde { O } ( \\frac { d \\log ^ { 1 / 2 } ( 1 / \\delta ) } { \\alpha \\varepsilon } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    295,
+                    1587,
+                    448,
+                    1587,
+                    448,
+                    1623,
+                    295,
+                    1623
+                ],
+                "score": 0.91,
+                "latex": "( 1 / 2 b ) e ^ { - | z | / b }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1295,
+                    454,
+                    1355,
+                    454,
+                    1355,
+                    488,
+                    1295,
+                    488
+                ],
+                "score": 0.91,
+                "latex": "( \\varepsilon , \\delta )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1255,
+                    321,
+                    1328,
+                    321,
+                    1328,
+                    357,
+                    1255,
+                    357
+                ],
+                "score": 0.9,
+                "latex": "\\Omega ( \\frac { d } { \\alpha \\varepsilon } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    871,
+                    1083,
+                    923,
+                    1083,
+                    923,
+                    1116,
+                    871,
+                    1116
+                ],
+                "score": 0.9,
+                "latex": "d ^ { 1 / 2 }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    962,
+                    232,
+                    1138,
+                    232,
+                    1138,
+                    278,
+                    962,
+                    278
+                ],
+                "score": 0.9,
+                "latex": "\\widetilde { O } ( \\frac { d \\log ^ { 1 / 2 } ( 1 / \\delta ) } { \\alpha \\varepsilon } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    298,
+                    1702,
+                    525,
+                    1702,
+                    525,
+                    1737,
+                    298,
+                    1737
+                ],
+                "score": 0.9,
+                "latex": "f ( S ) + \\mathrm { L a p } ( { \\Delta _ { 1 } } / { \\varepsilon } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    566,
+                    1439,
+                    982,
+                    1439,
+                    982,
+                    1474,
+                    566,
+                    1474
+                ],
+                "score": 0.89,
+                "latex": "d _ { \\triangle } ( S , S ^ { \\prime } ) \\triangleq \\operatorname* { m a x } \\{ | S \\setminus S ^ { \\prime } | , | S ^ { \\prime } \\setminus S | \\}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1224,
+                    1440,
+                    1308,
+                    1440,
+                    1308,
+                    1469,
+                    1224,
+                    1469
+                ],
+                "score": 0.89,
+                "latex": "S \\sim S ^ { \\prime }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    924,
+                    1667,
+                    1010,
+                    1667,
+                    1010,
+                    1699,
+                    924,
+                    1699
+                ],
+                "score": 0.89,
+                "latex": "p ~ = ~ 1"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    926,
+                    1704,
+                    1004,
+                    1704,
+                    1004,
+                    1736,
+                    926,
+                    1736
+                ],
+                "score": 0.89,
+                "latex": "p = 2"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1106,
+                    1630,
+                    1405,
+                    1630,
+                    1405,
+                    1666,
+                    1106,
+                    1666
+                ],
+                "score": 0.88,
+                "latex": "\\Delta _ { p } ~ = ~ \\operatorname* { s u p } _ { S \\sim S ^ { \\prime } } \\| f ( S ) ~ -"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    297,
+                    1735,
+                    771,
+                    1735,
+                    771,
+                    1773,
+                    297,
+                    1773
+                ],
+                "score": 0.88,
+                "latex": "f ( S ) + \\mathcal { N } ( 0 , ( \\Delta _ { 2 } ( \\sqrt { 2 \\log ( 1 . 2 5 / \\delta ) } ) / \\varepsilon ) ^ { 2 } \\mathbf { I } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    297,
+                    1503,
+                    365,
+                    1503,
+                    365,
+                    1531,
+                    297,
+                    1531
+                ],
+                "score": 0.88,
+                "latex": "\\varepsilon > 0"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    417,
+                    1501,
+                    527,
+                    1501,
+                    527,
+                    1535,
+                    417,
+                    1535
+                ],
+                "score": 0.88,
+                "latex": "\\delta \\in ( 0 , 1 )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    545,
+                    1501,
+                    968,
+                    1501,
+                    968,
+                    1536,
+                    545,
+                    1536
+                ],
+                "score": 0.87,
+                "latex": "\\begin{array} { r } { \\textstyle f \\mathbb { P } ( q ( S ) \\in A ) \\le e ^ { \\varepsilon } \\mathbb { P } ( q ( S ^ { \\prime } ) \\in A ) + \\delta } \\end{array}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1194,
+                    1406,
+                    1222,
+                    1406,
+                    1222,
+                    1433,
+                    1194,
+                    1433
+                ],
+                "score": 0.85,
+                "latex": "S ^ { \\prime }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    710,
+                    280,
+                    776,
+                    280,
+                    776,
+                    313,
+                    710,
+                    313
+                ],
+                "score": 0.84,
+                "latex": "\\underline { { O } } ( \\frac { d } { \\alpha } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    333,
+                    352,
+                    394,
+                    352,
+                    394,
+                    385,
+                    333,
+                    385
+                ],
+                "score": 0.84,
+                "latex": "( \\varepsilon , \\delta )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1017,
+                    279,
+                    1083,
+                    279,
+                    1083,
+                    312,
+                    1017,
+                    312
+                ],
+                "score": 0.81,
+                "latex": "O ( \\textstyle { \\frac { d } { \\alpha } } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    932,
+                    311,
+                    1169,
+                    311,
+                    1169,
+                    358,
+                    932,
+                    358
+                ],
+                "score": 0.81,
+                "latex": "\\begin{array} { r } { \\widetilde { O } ( \\frac { d + d ^ { 1 / 2 } \\log ^ { 3 / 2 } ( 1 / \\delta ) } { \\alpha \\varepsilon } ) } \\end{array}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1356,
+                    1591,
+                    1378,
+                    1591,
+                    1378,
+                    1617,
+                    1356,
+                    1617
+                ],
+                "score": 0.8,
+                "latex": "\\Sigma"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1369,
+                    1149,
+                    1402,
+                    1149,
+                    1402,
+                    1179,
+                    1369,
+                    1179
+                ],
+                "score": 0.8,
+                "latex": "\\ S 4"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1255,
+                    630,
+                    1276,
+                    630,
+                    1276,
+                    653,
+                    1255,
+                    653
+                ],
+                "score": 0.79,
+                "latex": "\\alpha"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1159,
+                    1595,
+                    1179,
+                    1595,
+                    1179,
+                    1622,
+                    1159,
+                    1622
+                ],
+                "score": 0.79,
+                "latex": "\\mu"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1296,
+                    1503,
+                    1318,
+                    1503,
+                    1318,
+                    1529,
+                    1296,
+                    1529
+                ],
+                "score": 0.78,
+                "latex": "A"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    425,
+                    1981,
+                    447,
+                    1981,
+                    447,
+                    2003,
+                    425,
+                    2003
+                ],
+                "score": 0.78,
+                "latex": "\\alpha"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    483,
+                    490,
+                    504,
+                    490,
+                    504,
+                    512,
+                    483,
+                    512
+                ],
+                "score": 0.78,
+                "latex": "\\alpha"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    930,
+                    1738,
+                    1094,
+                    1738,
+                    1094,
+                    1772,
+                    930,
+                    1772
+                ],
+                "score": 0.77,
+                "latex": "( \\varepsilon , \\delta ) – D P / 3 8 J"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    649,
+                    313,
+                    837,
+                    313,
+                    837,
+                    358,
+                    649,
+                    358
+                ],
+                "score": 0.74,
+                "latex": "\\displaystyle \\widetilde { \\cal O } \\Bigl ( \\frac { d ^ { 3 / 2 } \\log ( 1 / \\delta ) } { \\alpha \\varepsilon } \\Bigr )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1124,
+                    1406,
+                    1146,
+                    1406,
+                    1146,
+                    1434,
+                    1124,
+                    1434
+                ],
+                "score": 0.73,
+                "latex": "S"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    453,
+                    1953,
+                    473,
+                    1953,
+                    473,
+                    1973,
+                    453,
+                    1973
+                ],
+                "score": 0.73,
+                "latex": "n"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    357,
+                    333,
+                    376,
+                    333,
+                    376,
+                    348,
+                    357,
+                    348
+                ],
+                "score": 0.68,
+                "latex": "\\alpha"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    354,
+                    291,
+                    373,
+                    291,
+                    373,
+                    304,
+                    354,
+                    304
+                ],
+                "score": 0.66,
+                "latex": "\\alpha"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    373,
+                    244,
+                    434,
+                    244,
+                    434,
+                    274,
+                    373,
+                    274
+                ],
+                "score": 0.63,
+                "latex": "( \\varepsilon , \\delta )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    691,
+                    1705,
+                    751,
+                    1705,
+                    751,
+                    1736,
+                    691,
+                    1736
+                ],
+                "score": 0.59,
+                "latex": "( \\varepsilon , 0 )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    690,
+                    1704,
+                    859,
+                    1704,
+                    859,
+                    1736,
+                    690,
+                    1736
+                ],
+                "score": 0.41,
+                "latex": "( \\varepsilon , 0 ) – D P \\ : I 3 7 J"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    930,
+                    1738,
+                    991,
+                    1738,
+                    991,
+                    1772,
+                    930,
+                    1772
+                ],
+                "score": 0.26,
+                "latex": "( \\varepsilon , \\delta )"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1887.0,
+                    614.0,
+                    1887.0,
+                    614.0,
+                    1923.0,
+                    296.0,
+                    1923.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1299.0,
+                    843.0,
+                    1299.0,
+                    843.0,
+                    1347.0,
+                    292.0,
+                    1347.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    565.0,
+                    644.0,
+                    565.0,
+                    644.0,
+                    604.0,
+                    295.0,
+                    604.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    838.0,
+                    2058.0,
+                    862.0,
+                    2058.0,
+                    862.0,
+                    2091.0,
+                    838.0,
+                    2091.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    838.0,
+                    2058.0,
+                    862.0,
+                    2058.0,
+                    862.0,
+                    2091.0,
+                    838.0,
+                    2091.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    848.0,
+                    1086.0,
+                    848.0,
+                    1086.0,
+                    889.0,
+                    293.0,
+                    889.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1149.0,
+                    848.0,
+                    1405.0,
+                    848.0,
+                    1405.0,
+                    889.0,
+                    1149.0,
+                    889.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    883.0,
+                    358.0,
+                    883.0,
+                    358.0,
+                    923.0,
+                    293.0,
+                    923.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    443.0,
+                    883.0,
+                    1406.0,
+                    883.0,
+                    1406.0,
+                    923.0,
+                    443.0,
+                    923.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    918.0,
+                    1402.0,
+                    918.0,
+                    1402.0,
+                    953.0,
+                    294.0,
+                    953.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    947.0,
+                    1405.0,
+                    947.0,
+                    1405.0,
+                    981.0,
+                    293.0,
+                    981.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    976.0,
+                    1402.0,
+                    976.0,
+                    1402.0,
+                    1011.0,
+                    294.0,
+                    1011.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1009.0,
+                    1404.0,
+                    1009.0,
+                    1404.0,
+                    1044.0,
+                    293.0,
+                    1044.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1036.0,
+                    682.0,
+                    1036.0,
+                    682.0,
+                    1075.0,
+                    293.0,
+                    1075.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    744.0,
+                    1036.0,
+                    775.0,
+                    1036.0,
+                    775.0,
+                    1075.0,
+                    744.0,
+                    1075.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    906.0,
+                    1036.0,
+                    920.0,
+                    1036.0,
+                    920.0,
+                    1075.0,
+                    906.0,
+                    1075.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    623.0,
+                    890.0,
+                    623.0,
+                    890.0,
+                    661.0,
+                    294.0,
+                    661.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    953.0,
+                    623.0,
+                    1254.0,
+                    623.0,
+                    1254.0,
+                    661.0,
+                    953.0,
+                    661.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1277.0,
+                    623.0,
+                    1407.0,
+                    623.0,
+                    1407.0,
+                    661.0,
+                    1277.0,
+                    661.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    652.0,
+                    1409.0,
+                    652.0,
+                    1409.0,
+                    692.0,
+                    291.0,
+                    692.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    685.0,
+                    1408.0,
+                    685.0,
+                    1408.0,
+                    719.0,
+                    295.0,
+                    719.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    716.0,
+                    1404.0,
+                    716.0,
+                    1404.0,
+                    750.0,
+                    295.0,
+                    750.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    744.0,
+                    1406.0,
+                    744.0,
+                    1406.0,
+                    783.0,
+                    292.0,
+                    783.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    773.0,
+                    1406.0,
+                    773.0,
+                    1406.0,
+                    813.0,
+                    292.0,
+                    813.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    807.0,
+                    1383.0,
+                    807.0,
+                    1383.0,
+                    842.0,
+                    295.0,
+                    842.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1081.0,
+                    870.0,
+                    1081.0,
+                    870.0,
+                    1124.0,
+                    294.0,
+                    1124.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    924.0,
+                    1081.0,
+                    1406.0,
+                    1081.0,
+                    1406.0,
+                    1124.0,
+                    924.0,
+                    1124.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1119.0,
+                    1404.0,
+                    1119.0,
+                    1404.0,
+                    1151.0,
+                    295.0,
+                    1151.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1149.0,
+                    1368.0,
+                    1149.0,
+                    1368.0,
+                    1182.0,
+                    292.0,
+                    1182.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1179.0,
+                    1405.0,
+                    1179.0,
+                    1405.0,
+                    1215.0,
+                    295.0,
+                    1215.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1207.0,
+                    1407.0,
+                    1207.0,
+                    1407.0,
+                    1247.0,
+                    294.0,
+                    1247.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1239.0,
+                    1213.0,
+                    1239.0,
+                    1213.0,
+                    1274.0,
+                    294.0,
+                    1274.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    289.0,
+                    1398.0,
+                    683.0,
+                    1398.0,
+                    683.0,
+                    1448.0,
+                    289.0,
+                    1448.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    831.0,
+                    1398.0,
+                    879.0,
+                    1398.0,
+                    879.0,
+                    1448.0,
+                    831.0,
+                    1448.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1036.0,
+                    1398.0,
+                    1123.0,
+                    1398.0,
+                    1123.0,
+                    1448.0,
+                    1036.0,
+                    1448.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1147.0,
+                    1398.0,
+                    1193.0,
+                    1398.0,
+                    1193.0,
+                    1448.0,
+                    1147.0,
+                    1448.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1223.0,
+                    1398.0,
+                    1408.0,
+                    1398.0,
+                    1408.0,
+                    1448.0,
+                    1223.0,
+                    1448.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    488.0,
+                    1438.0,
+                    565.0,
+                    1438.0,
+                    565.0,
+                    1476.0,
+                    488.0,
+                    1476.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    983.0,
+                    1438.0,
+                    1223.0,
+                    1438.0,
+                    1223.0,
+                    1476.0,
+                    983.0,
+                    1476.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1309.0,
+                    1438.0,
+                    1405.0,
+                    1438.0,
+                    1405.0,
+                    1476.0,
+                    1309.0,
+                    1476.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1470.0,
+                    1009.0,
+                    1470.0,
+                    1009.0,
+                    1507.0,
+                    293.0,
+                    1507.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1072.0,
+                    1470.0,
+                    1405.0,
+                    1470.0,
+                    1405.0,
+                    1507.0,
+                    1072.0,
+                    1507.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1501.0,
+                    296.0,
+                    1501.0,
+                    296.0,
+                    1535.0,
+                    293.0,
+                    1535.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    366.0,
+                    1501.0,
+                    416.0,
+                    1501.0,
+                    416.0,
+                    1535.0,
+                    366.0,
+                    1535.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    528.0,
+                    1501.0,
+                    544.0,
+                    1501.0,
+                    544.0,
+                    1535.0,
+                    528.0,
+                    1535.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    969.0,
+                    1501.0,
+                    1045.0,
+                    1501.0,
+                    1045.0,
+                    1535.0,
+                    969.0,
+                    1535.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1131.0,
+                    1501.0,
+                    1295.0,
+                    1501.0,
+                    1295.0,
+                    1535.0,
+                    1131.0,
+                    1535.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1319.0,
+                    1501.0,
+                    1330.0,
+                    1501.0,
+                    1330.0,
+                    1535.0,
+                    1319.0,
+                    1535.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1624.0,
+                    796.0,
+                    1624.0,
+                    796.0,
+                    1670.0,
+                    291.0,
+                    1670.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    941.0,
+                    1624.0,
+                    1105.0,
+                    1624.0,
+                    1105.0,
+                    1670.0,
+                    941.0,
+                    1670.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1659.0,
+                    296.0,
+                    1659.0,
+                    296.0,
+                    1705.0,
+                    291.0,
+                    1705.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    392.0,
+                    1659.0,
+                    541.0,
+                    1659.0,
+                    541.0,
+                    1705.0,
+                    392.0,
+                    1705.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    847.0,
+                    1659.0,
+                    923.0,
+                    1659.0,
+                    923.0,
+                    1705.0,
+                    847.0,
+                    1705.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1011.0,
+                    1659.0,
+                    1407.0,
+                    1659.0,
+                    1407.0,
+                    1705.0,
+                    1011.0,
+                    1705.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    526.0,
+                    1700.0,
+                    689.0,
+                    1700.0,
+                    689.0,
+                    1740.0,
+                    526.0,
+                    1740.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    860.0,
+                    1700.0,
+                    925.0,
+                    1700.0,
+                    925.0,
+                    1740.0,
+                    860.0,
+                    1740.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1005.0,
+                    1700.0,
+                    1405.0,
+                    1700.0,
+                    1405.0,
+                    1740.0,
+                    1005.0,
+                    1740.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    772.0,
+                    1733.0,
+                    929.0,
+                    1733.0,
+                    929.0,
+                    1777.0,
+                    772.0,
+                    1777.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1095.0,
+                    1733.0,
+                    1106.0,
+                    1733.0,
+                    1106.0,
+                    1777.0,
+                    1095.0,
+                    1777.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1787.0,
+                    1406.0,
+                    1787.0,
+                    1406.0,
+                    1831.0,
+                    293.0,
+                    1831.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1822.0,
+                    1173.0,
+                    1822.0,
+                    1173.0,
+                    1858.0,
+                    295.0,
+                    1858.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    1947.0,
+                    452.0,
+                    1947.0,
+                    452.0,
+                    1979.0,
+                    297.0,
+                    1979.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    474.0,
+                    1947.0,
+                    1403.0,
+                    1947.0,
+                    1403.0,
+                    1979.0,
+                    474.0,
+                    1979.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1977.0,
+                    424.0,
+                    1977.0,
+                    424.0,
+                    2009.0,
+                    296.0,
+                    2009.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    448.0,
+                    1977.0,
+                    1401.0,
+                    1977.0,
+                    1401.0,
+                    2009.0,
+                    448.0,
+                    2009.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1550.0,
+                    340.0,
+                    1550.0,
+                    340.0,
+                    1592.0,
+                    291.0,
+                    1592.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    478.0,
+                    1550.0,
+                    1406.0,
+                    1550.0,
+                    1406.0,
+                    1592.0,
+                    478.0,
+                    1592.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    290.0,
+                    1583.0,
+                    294.0,
+                    1583.0,
+                    294.0,
+                    1626.0,
+                    290.0,
+                    1626.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    449.0,
+                    1583.0,
+                    505.0,
+                    1583.0,
+                    505.0,
+                    1626.0,
+                    449.0,
+                    1626.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    657.0,
+                    1583.0,
+                    1158.0,
+                    1583.0,
+                    1158.0,
+                    1626.0,
+                    657.0,
+                    1626.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1180.0,
+                    1583.0,
+                    1355.0,
+                    1583.0,
+                    1355.0,
+                    1626.0,
+                    1180.0,
+                    1626.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1379.0,
+                    1583.0,
+                    1391.0,
+                    1583.0,
+                    1391.0,
+                    1626.0,
+                    1379.0,
+                    1626.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1354.0,
+                    1388.0,
+                    1354.0,
+                    1388.0,
+                    1400.0,
+                    293.0,
+                    1400.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    420.0,
+                    667.0,
+                    420.0,
+                    667.0,
+                    454.0,
+                    294.0,
+                    454.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    756.0,
+                    420.0,
+                    1403.0,
+                    420.0,
+                    1403.0,
+                    454.0,
+                    756.0,
+                    454.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    449.0,
+                    978.0,
+                    449.0,
+                    978.0,
+                    492.0,
+                    293.0,
+                    492.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1222.0,
+                    449.0,
+                    1294.0,
+                    449.0,
+                    1294.0,
+                    492.0,
+                    1222.0,
+                    492.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1356.0,
+                    449.0,
+                    1409.0,
+                    449.0,
+                    1409.0,
+                    492.0,
+                    1356.0,
+                    492.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    485.0,
+                    482.0,
+                    485.0,
+                    482.0,
+                    517.0,
+                    294.0,
+                    517.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    505.0,
+                    485.0,
+                    844.0,
+                    485.0,
+                    844.0,
+                    517.0,
+                    505.0,
+                    517.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 2,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1080,
+                    1408,
+                    1080,
+                    1408,
+                    1203,
+                    298,
+                    1203
+                ],
+                "score": 0.979
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    296,
+                    1448,
+                    1405,
+                    1448,
+                    1405,
+                    1755,
+                    296,
+                    1755
+                ],
+                "score": 0.979
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    540,
+                    1406,
+                    540,
+                    1406,
+                    712,
+                    297,
+                    712
+                ],
+                "score": 0.976
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    272,
+                    1405,
+                    272,
+                    1405,
+                    441,
+                    297,
+                    441
+                ],
+                "score": 0.975
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1854,
+                    1404,
+                    1854,
+                    1404,
+                    2008,
+                    298,
+                    2008
+                ],
+                "score": 0.974
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    361,
+                    1211,
+                    1403,
+                    1211,
+                    1403,
+                    1431,
+                    361,
+                    1431
+                ],
+                "score": 0.963
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    296,
+                    723,
+                    1405,
+                    723,
+                    1405,
+                    819,
+                    296,
+                    819
+                ],
+                "score": 0.959
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    926,
+                    1403,
+                    926,
+                    1403,
+                    991,
+                    299,
+                    991
+                ],
+                "score": 0.956
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    294,
+                    202,
+                    1400,
+                    202,
+                    1400,
+                    266,
+                    294,
+                    266
+                ],
+                "score": 0.951
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    463,
+                    1401,
+                    463,
+                    1401,
+                    526,
+                    298,
+                    526
+                ],
+                "score": 0.948
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    294,
+                    1761,
+                    1402,
+                    1761,
+                    1402,
+                    1834,
+                    294,
+                    1834
+                ],
+                "score": 0.937
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    296,
+                    857,
+                    1248,
+                    857,
+                    1248,
+                    897,
+                    296,
+                    897
+                ],
+                "score": 0.934
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    299,
+                    1023,
+                    1240,
+                    1023,
+                    1240,
+                    1058,
+                    299,
+                    1058
+                ],
+                "score": 0.922
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    840,
+                    2061,
+                    858,
+                    2061,
+                    858,
+                    2084,
+                    840,
+                    2084
+                ],
+                "score": 0.767
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    444,
+                    1395,
+                    606,
+                    1395,
+                    606,
+                    1428,
+                    444,
+                    1428
+                ],
+                "score": 0.93,
+                "latex": "\\tau _ { i } \\geq Z _ { t } \\tau _ { \\operatorname* { m a x } } \\}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    952,
+                    539,
+                    1244,
+                    539,
+                    1244,
+                    583,
+                    952,
+                    583
+                ],
+                "score": 0.93,
+                "latex": "\\textstyle \\| x \\| _ { 2 } = ( \\sum _ { i \\in [ d ] } ( x _ { i } ) ^ { 2 } ) ^ { 1 / 2 }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    298,
+                    680,
+                    402,
+                    680,
+                    402,
+                    713,
+                    298,
+                    713
+                ],
+                "score": 0.93,
+                "latex": "d , n , 1 / \\alpha"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    542,
+                    1142,
+                    643,
+                    1142,
+                    643,
+                    1175,
+                    542,
+                    1175
+                ],
+                "score": 0.93,
+                "latex": "S _ { 1 } = [ n ]"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    964,
+                    343,
+                    1105,
+                    343,
+                    1105,
+                    375,
+                    964,
+                    375
+                ],
+                "score": 0.93,
+                "latex": "\\alpha \\in ( 0 , 1 / 2 )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    801,
+                    1111,
+                    948,
+                    1111,
+                    948,
+                    1145,
+                    801,
+                    1145
+                ],
+                "score": 0.92,
+                "latex": "S = \\{ x _ { i } \\} _ { i = 1 } ^ { n }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    615,
+                    1249,
+                    740,
+                    1249,
+                    740,
+                    1286,
+                    615,
+                    1286
+                ],
+                "score": 0.92,
+                "latex": "\\{ x _ { i } \\} _ { i \\in S _ { t - 1 } }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1121,
+                    1112,
+                    1221,
+                    1112,
+                    1221,
+                    1144,
+                    1121,
+                    1144
+                ],
+                "score": 0.92,
+                "latex": "S _ { 0 } \\subseteq [ n ]"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    383,
+                    374,
+                    597,
+                    374,
+                    597,
+                    409,
+                    383,
+                    409
+                ],
+                "score": 0.92,
+                "latex": "S = \\{ x _ { i } \\in \\mathbb { R } ^ { d } \\} _ { i = 1 } ^ { n }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    477,
+                    542,
+                    698,
+                    542,
+                    698,
+                    577,
+                    477,
+                    577
+                ],
+                "score": 0.92,
+                "latex": "[ n ] = \\{ 1 , 2 , \\dots , n \\}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1275,
+                    1363,
+                    1401,
+                    1363,
+                    1401,
+                    1399,
+                    1275,
+                    1399
+                ],
+                "score": 0.92,
+                "latex": "\\{ \\tau _ { j } \\} _ { j \\in S _ { t - 1 } }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    765,
+                    582,
+                    1081,
+                    582,
+                    1081,
+                    619,
+                    765,
+                    619
+                ],
+                "score": 0.92,
+                "latex": "\\begin{array} { r } { \\| X \\| _ { 2 } = \\operatorname* { m a x } _ { \\| v \\| _ { 2 } = 1 } \\| X v \\| _ { 2 } } \\end{array}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    705,
+                    307,
+                    791,
+                    307,
+                    791,
+                    340,
+                    705,
+                    340
+                ],
+                "score": 0.91,
+                "latex": "\\mu \\in \\mathbb { R } ^ { d }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    970,
+                    305,
+                    1120,
+                    305,
+                    1120,
+                    341,
+                    970,
+                    341
+                ],
+                "score": 0.91,
+                "latex": "\\mathbb { E } [ x x ^ { \\top } ] = \\mathbf { I } _ { d } ,"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1230,
+                    375,
+                    1297,
+                    375,
+                    1297,
+                    408,
+                    1230,
+                    408
+                ],
+                "score": 0.91,
+                "latex": "S _ { \\mathrm { g o o d } }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    641,
+                    1364,
+                    699,
+                    1364,
+                    699,
+                    1396,
+                    641,
+                    1396
+                ],
+                "score": 0.91,
+                "latex": "S _ { t - 1 }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    940,
+                    1946,
+                    998,
+                    1946,
+                    998,
+                    1978,
+                    940,
+                    1978
+                ],
+                "score": 0.91,
+                "latex": "S _ { t - 1 }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    821,
+                    1365,
+                    988,
+                    1365,
+                    988,
+                    1397,
+                    821,
+                    1397
+                ],
+                "score": 0.91,
+                "latex": "\\{ i \\in S _ { t - 1 } : \\tau _ { i }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    345,
+                    613,
+                    411,
+                    613,
+                    411,
+                    642,
+                    345,
+                    642
+                ],
+                "score": 0.91,
+                "latex": "d \\times d"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    541,
+                    580,
+                    666,
+                    580,
+                    666,
+                    611,
+                    541,
+                    611
+                ],
+                "score": 0.91,
+                "latex": "X \\in \\mathbb { R } ^ { d \\times d }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    699,
+                    409,
+                    755,
+                    409,
+                    755,
+                    439,
+                    699,
+                    439
+                ],
+                "score": 0.9,
+                "latex": "S _ { \\mathrm { b a d } }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    761,
+                    541,
+                    851,
+                    541,
+                    851,
+                    573,
+                    761,
+                    573
+                ],
+                "score": 0.9,
+                "latex": "x \\in \\mathbb { R } ^ { d }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    757,
+                    340,
+                    840,
+                    340,
+                    840,
+                    371,
+                    757,
+                    371
+                ],
+                "score": 0.9,
+                "latex": "v \\in \\mathbb { R } ^ { d }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    696,
+                    1330,
+                    904,
+                    1330,
+                    904,
+                    1364,
+                    696,
+                    1364
+                ],
+                "score": 0.9,
+                "latex": "Z _ { t } \\gets \\mathrm { U n i f } ( [ 0 , 1 ] )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    763,
+                    274,
+                    830,
+                    274,
+                    830,
+                    307,
+                    763,
+                    307
+                ],
+                "score": 0.9,
+                "latex": "S _ { \\mathrm { g o o d } }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    605,
+                    614,
+                    662,
+                    614,
+                    662,
+                    646,
+                    605,
+                    646
+                ],
+                "score": 0.9,
+                "latex": "\\mathbf { I } _ { d \\times d }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    691,
+                    1399,
+                    940,
+                    1399,
+                    940,
+                    1430,
+                    691,
+                    1430
+                ],
+                "score": 0.89,
+                "latex": "\\tau _ { \\mathrm { m a x } } = \\operatorname* { m a x } _ { j \\in S _ { t - 1 } } \\tau _ { j }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    346,
+                    339,
+                    677,
+                    339,
+                    677,
+                    374,
+                    346,
+                    374
+                ],
+                "score": 0.89,
+                "latex": "\\mathbb { E } [ \\exp ( v ^ { \\top } x ) ] \\leq \\exp ( \\| v \\| _ { 2 } ^ { 2 } / 2 )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    569,
+                    1632,
+                    601,
+                    1632,
+                    601,
+                    1662,
+                    569,
+                    1662
+                ],
+                "score": 0.88,
+                "latex": "Z _ { t }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1127,
+                    407,
+                    1161,
+                    407,
+                    1161,
+                    436,
+                    1127,
+                    436
+                ],
+                "score": 0.88,
+                "latex": "\\mathbb { R } ^ { d }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    732,
+                    1212,
+                    1168,
+                    1212,
+                    1168,
+                    1254,
+                    732,
+                    1254
+                ],
+                "score": 0.88,
+                "latex": "v _ { t } \\gets \\arg \\operatorname* { m a x } _ { v : \\| v \\| _ { 2 } = 1 } v ^ { \\top } \\mathrm { C o v } ( S _ { t - 1 } ) v"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    565,
+                    1946,
+                    595,
+                    1946,
+                    595,
+                    1976,
+                    565,
+                    1976
+                ],
+                "score": 0.87,
+                "latex": "S _ { t }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    561,
+                    1977,
+                    591,
+                    1977,
+                    591,
+                    2006,
+                    561,
+                    2006
+                ],
+                "score": 0.86,
+                "latex": "S _ { t }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1022,
+                    643,
+                    1045,
+                    643,
+                    1045,
+                    677,
+                    1022,
+                    677
+                ],
+                "score": 0.86,
+                "latex": "\\widetilde \\Omega"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    946,
+                    643,
+                    971,
+                    643,
+                    971,
+                    678,
+                    946,
+                    678
+                ],
+                "score": 0.85,
+                "latex": "\\widetilde O"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1279,
+                    1452,
+                    1298,
+                    1452,
+                    1298,
+                    1481,
+                    1279,
+                    1481
+                ],
+                "score": 0.85,
+                "latex": "\\hat { \\mu }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    782,
+                    1288,
+                    1292,
+                    1288,
+                    1292,
+                    1331,
+                    782,
+                    1331
+                ],
+                "score": 0.83,
+                "latex": "j \\in S _ { t - 1 } \\colon \\tau _ { j } \\gets \\left( v _ { t } ^ { \\top } \\left( x _ { j } - \\mathrm { M e a n } ( S _ { t - 1 } ) \\right) \\right) ^ { 2 } \\ ;"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1050,
+                    1886,
+                    1070,
+                    1886,
+                    1070,
+                    1917,
+                    1050,
+                    1917
+                ],
+                "score": 0.83,
+                "latex": "\\hat { \\mu }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1215,
+                    756,
+                    1246,
+                    756,
+                    1246,
+                    786,
+                    1215,
+                    786
+                ],
+                "score": 0.83,
+                "latex": "\\ S 4"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1132,
+                    614,
+                    1154,
+                    614,
+                    1154,
+                    640,
+                    1132,
+                    640
+                ],
+                "score": 0.82,
+                "latex": "S"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    298,
+                    756,
+                    328,
+                    756,
+                    328,
+                    787,
+                    298,
+                    787
+                ],
+                "score": 0.81,
+                "latex": "\\ S 3"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1163,
+                    1365,
+                    1198,
+                    1365,
+                    1198,
+                    1392,
+                    1163,
+                    1392
+                ],
+                "score": 0.8,
+                "latex": "2 \\alpha"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1282,
+                    1486,
+                    1304,
+                    1486,
+                    1304,
+                    1508,
+                    1282,
+                    1508
+                ],
+                "score": 0.79,
+                "latex": "\\alpha"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    981,
+                    726,
+                    1012,
+                    726,
+                    1012,
+                    755,
+                    981,
+                    755
+                ],
+                "score": 0.76,
+                "latex": "\\ S 2"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    508,
+                    1794,
+                    1130,
+                    1794,
+                    1130,
+                    1834,
+                    508,
+                    1834
+                ],
+                "score": 0.75,
+                "latex": "\\| \\hat { \\mu } - \\mu \\| _ { 2 } \\le O ( \\alpha \\sqrt { \\log ( 1 / \\alpha ) } ) \\ w . p . \\ 0 . 9 i f n \\ge \\widetilde { \\Omega } ( d / \\alpha ^ { 2 } ) \\ / ."
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1236,
+                    276,
+                    1253,
+                    276,
+                    1253,
+                    301,
+                    1236,
+                    301
+                ],
+                "score": 0.69,
+                "latex": "d"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    783,
+                    1294,
+                    887,
+                    1294,
+                    887,
+                    1328,
+                    783,
+                    1328
+                ],
+                "score": 0.38,
+                "latex": "j \\in S _ { t - 1 }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1062,
+                    1765,
+                    1077,
+                    1765,
+                    1077,
+                    1790,
+                    1062,
+                    1790
+                ],
+                "score": 0.27,
+                "latex": "^ { l }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    967,
+                    280,
+                    986,
+                    280,
+                    986,
+                    301,
+                    967,
+                    301
+                ],
+                "score": 0.26,
+                "latex": "n"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    856.0,
+                    1251.0,
+                    856.0,
+                    1251.0,
+                    901.0,
+                    291.0,
+                    901.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1022.0,
+                    1243.0,
+                    1022.0,
+                    1243.0,
+                    1063.0,
+                    292.0,
+                    1063.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    838.0,
+                    2058.0,
+                    863.0,
+                    2058.0,
+                    863.0,
+                    2091.0,
+                    838.0,
+                    2091.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1081.0,
+                    1409.0,
+                    1081.0,
+                    1409.0,
+                    1117.0,
+                    293.0,
+                    1117.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1107.0,
+                    800.0,
+                    1107.0,
+                    800.0,
+                    1151.0,
+                    291.0,
+                    1151.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    949.0,
+                    1107.0,
+                    1120.0,
+                    1107.0,
+                    1120.0,
+                    1151.0,
+                    949.0,
+                    1151.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1222.0,
+                    1107.0,
+                    1410.0,
+                    1107.0,
+                    1410.0,
+                    1151.0,
+                    1222.0,
+                    1151.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1141.0,
+                    541.0,
+                    1141.0,
+                    541.0,
+                    1178.0,
+                    295.0,
+                    1178.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    644.0,
+                    1141.0,
+                    1406.0,
+                    1141.0,
+                    1406.0,
+                    1178.0,
+                    644.0,
+                    1178.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1173.0,
+                    856.0,
+                    1173.0,
+                    856.0,
+                    1205.0,
+                    296.0,
+                    1205.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1448.0,
+                    1278.0,
+                    1448.0,
+                    1278.0,
+                    1485.0,
+                    293.0,
+                    1485.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1299.0,
+                    1448.0,
+                    1408.0,
+                    1448.0,
+                    1408.0,
+                    1485.0,
+                    1299.0,
+                    1485.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1482.0,
+                    1281.0,
+                    1482.0,
+                    1281.0,
+                    1514.0,
+                    295.0,
+                    1514.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1305.0,
+                    1482.0,
+                    1405.0,
+                    1482.0,
+                    1405.0,
+                    1514.0,
+                    1305.0,
+                    1514.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1505.0,
+                    1406.0,
+                    1505.0,
+                    1406.0,
+                    1549.0,
+                    291.0,
+                    1549.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1540.0,
+                    1405.0,
+                    1540.0,
+                    1405.0,
+                    1576.0,
+                    294.0,
+                    1576.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1571.0,
+                    1406.0,
+                    1571.0,
+                    1406.0,
+                    1605.0,
+                    293.0,
+                    1605.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1602.0,
+                    1406.0,
+                    1602.0,
+                    1406.0,
+                    1637.0,
+                    295.0,
+                    1637.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1632.0,
+                    568.0,
+                    1632.0,
+                    568.0,
+                    1668.0,
+                    294.0,
+                    1668.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    602.0,
+                    1632.0,
+                    1403.0,
+                    1632.0,
+                    1403.0,
+                    1668.0,
+                    602.0,
+                    1668.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1659.0,
+                    1406.0,
+                    1659.0,
+                    1406.0,
+                    1699.0,
+                    291.0,
+                    1699.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1692.0,
+                    1405.0,
+                    1692.0,
+                    1405.0,
+                    1727.0,
+                    294.0,
+                    1727.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1722.0,
+                    993.0,
+                    1722.0,
+                    993.0,
+                    1758.0,
+                    295.0,
+                    1758.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    537.0,
+                    476.0,
+                    537.0,
+                    476.0,
+                    584.0,
+                    292.0,
+                    584.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    699.0,
+                    537.0,
+                    760.0,
+                    537.0,
+                    760.0,
+                    584.0,
+                    699.0,
+                    584.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    852.0,
+                    537.0,
+                    951.0,
+                    537.0,
+                    951.0,
+                    584.0,
+                    852.0,
+                    584.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1245.0,
+                    537.0,
+                    1406.0,
+                    537.0,
+                    1406.0,
+                    584.0,
+                    1245.0,
+                    584.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    289.0,
+                    570.0,
+                    540.0,
+                    570.0,
+                    540.0,
+                    626.0,
+                    289.0,
+                    626.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    667.0,
+                    570.0,
+                    764.0,
+                    570.0,
+                    764.0,
+                    626.0,
+                    667.0,
+                    626.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1082.0,
+                    570.0,
+                    1410.0,
+                    570.0,
+                    1410.0,
+                    626.0,
+                    1082.0,
+                    626.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    610.0,
+                    344.0,
+                    610.0,
+                    344.0,
+                    648.0,
+                    294.0,
+                    648.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    412.0,
+                    610.0,
+                    604.0,
+                    610.0,
+                    604.0,
+                    648.0,
+                    412.0,
+                    648.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    663.0,
+                    610.0,
+                    1131.0,
+                    610.0,
+                    1131.0,
+                    648.0,
+                    663.0,
+                    648.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1155.0,
+                    610.0,
+                    1407.0,
+                    610.0,
+                    1407.0,
+                    648.0,
+                    1155.0,
+                    648.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    645.0,
+                    945.0,
+                    645.0,
+                    945.0,
+                    684.0,
+                    292.0,
+                    684.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    972.0,
+                    645.0,
+                    1021.0,
+                    645.0,
+                    1021.0,
+                    684.0,
+                    972.0,
+                    684.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1046.0,
+                    645.0,
+                    1406.0,
+                    645.0,
+                    1406.0,
+                    684.0,
+                    1046.0,
+                    684.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    678.0,
+                    297.0,
+                    678.0,
+                    297.0,
+                    715.0,
+                    294.0,
+                    715.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    403.0,
+                    678.0,
+                    712.0,
+                    678.0,
+                    712.0,
+                    715.0,
+                    403.0,
+                    715.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    271.0,
+                    762.0,
+                    271.0,
+                    762.0,
+                    312.0,
+                    292.0,
+                    312.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    831.0,
+                    271.0,
+                    966.0,
+                    271.0,
+                    966.0,
+                    312.0,
+                    831.0,
+                    312.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    987.0,
+                    271.0,
+                    1235.0,
+                    271.0,
+                    1235.0,
+                    312.0,
+                    987.0,
+                    312.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1254.0,
+                    271.0,
+                    1407.0,
+                    271.0,
+                    1407.0,
+                    312.0,
+                    1254.0,
+                    312.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    306.0,
+                    704.0,
+                    306.0,
+                    704.0,
+                    343.0,
+                    294.0,
+                    343.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    792.0,
+                    306.0,
+                    969.0,
+                    306.0,
+                    969.0,
+                    343.0,
+                    792.0,
+                    343.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1121.0,
+                    306.0,
+                    1408.0,
+                    306.0,
+                    1408.0,
+                    343.0,
+                    1121.0,
+                    343.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    338.0,
+                    345.0,
+                    338.0,
+                    345.0,
+                    378.0,
+                    292.0,
+                    378.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    678.0,
+                    338.0,
+                    756.0,
+                    338.0,
+                    756.0,
+                    378.0,
+                    678.0,
+                    378.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    841.0,
+                    338.0,
+                    963.0,
+                    338.0,
+                    963.0,
+                    378.0,
+                    841.0,
+                    378.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1106.0,
+                    338.0,
+                    1406.0,
+                    338.0,
+                    1406.0,
+                    378.0,
+                    1106.0,
+                    378.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    370.0,
+                    382.0,
+                    370.0,
+                    382.0,
+                    413.0,
+                    292.0,
+                    413.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    598.0,
+                    370.0,
+                    1229.0,
+                    370.0,
+                    1229.0,
+                    413.0,
+                    598.0,
+                    413.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1298.0,
+                    370.0,
+                    1406.0,
+                    370.0,
+                    1406.0,
+                    413.0,
+                    1298.0,
+                    413.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    407.0,
+                    698.0,
+                    407.0,
+                    698.0,
+                    443.0,
+                    292.0,
+                    443.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    756.0,
+                    407.0,
+                    1126.0,
+                    407.0,
+                    1126.0,
+                    443.0,
+                    756.0,
+                    443.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1162.0,
+                    407.0,
+                    1174.0,
+                    407.0,
+                    1174.0,
+                    443.0,
+                    1162.0,
+                    443.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1852.0,
+                    1405.0,
+                    1852.0,
+                    1405.0,
+                    1890.0,
+                    296.0,
+                    1890.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1886.0,
+                    1049.0,
+                    1886.0,
+                    1049.0,
+                    1919.0,
+                    296.0,
+                    1919.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1071.0,
+                    1886.0,
+                    1404.0,
+                    1886.0,
+                    1404.0,
+                    1919.0,
+                    1071.0,
+                    1919.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1915.0,
+                    1406.0,
+                    1915.0,
+                    1406.0,
+                    1948.0,
+                    296.0,
+                    1948.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1943.0,
+                    564.0,
+                    1943.0,
+                    564.0,
+                    1984.0,
+                    293.0,
+                    1984.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    596.0,
+                    1943.0,
+                    939.0,
+                    1943.0,
+                    939.0,
+                    1984.0,
+                    596.0,
+                    1984.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    999.0,
+                    1943.0,
+                    1406.0,
+                    1943.0,
+                    1406.0,
+                    1984.0,
+                    999.0,
+                    1984.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1971.0,
+                    560.0,
+                    1971.0,
+                    560.0,
+                    2014.0,
+                    292.0,
+                    2014.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    592.0,
+                    1971.0,
+                    1406.0,
+                    1971.0,
+                    1406.0,
+                    2014.0,
+                    592.0,
+                    2014.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    359.0,
+                    1209.0,
+                    731.0,
+                    1209.0,
+                    731.0,
+                    1255.0,
+                    359.0,
+                    1255.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1169.0,
+                    1209.0,
+                    1408.0,
+                    1209.0,
+                    1408.0,
+                    1255.0,
+                    1169.0,
+                    1255.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    392.0,
+                    1246.0,
+                    614.0,
+                    1246.0,
+                    614.0,
+                    1290.0,
+                    392.0,
+                    1290.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    741.0,
+                    1246.0,
+                    762.0,
+                    1246.0,
+                    762.0,
+                    1290.0,
+                    741.0,
+                    1290.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    353.0,
+                    1286.0,
+                    781.0,
+                    1286.0,
+                    781.0,
+                    1333.0,
+                    353.0,
+                    1333.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1293.0,
+                    1286.0,
+                    1298.0,
+                    1286.0,
+                    1298.0,
+                    1333.0,
+                    1293.0,
+                    1333.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    355.0,
+                    1323.0,
+                    695.0,
+                    1323.0,
+                    695.0,
+                    1367.0,
+                    355.0,
+                    1367.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    905.0,
+                    1323.0,
+                    925.0,
+                    1323.0,
+                    925.0,
+                    1367.0,
+                    905.0,
+                    1367.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    353.0,
+                    1352.0,
+                    640.0,
+                    1352.0,
+                    640.0,
+                    1406.0,
+                    353.0,
+                    1406.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    700.0,
+                    1352.0,
+                    820.0,
+                    1352.0,
+                    820.0,
+                    1406.0,
+                    700.0,
+                    1406.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    989.0,
+                    1352.0,
+                    1162.0,
+                    1352.0,
+                    1162.0,
+                    1406.0,
+                    989.0,
+                    1406.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1199.0,
+                    1352.0,
+                    1274.0,
+                    1352.0,
+                    1274.0,
+                    1406.0,
+                    1199.0,
+                    1406.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    392.0,
+                    1388.0,
+                    443.0,
+                    1388.0,
+                    443.0,
+                    1436.0,
+                    392.0,
+                    1436.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    607.0,
+                    1388.0,
+                    690.0,
+                    1388.0,
+                    690.0,
+                    1436.0,
+                    607.0,
+                    1436.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    941.0,
+                    1388.0,
+                    944.0,
+                    1388.0,
+                    944.0,
+                    1436.0,
+                    941.0,
+                    1436.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    724.0,
+                    980.0,
+                    724.0,
+                    980.0,
+                    759.0,
+                    295.0,
+                    759.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1013.0,
+                    724.0,
+                    1405.0,
+                    724.0,
+                    1405.0,
+                    759.0,
+                    1013.0,
+                    759.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    329.0,
+                    755.0,
+                    1214.0,
+                    755.0,
+                    1214.0,
+                    789.0,
+                    329.0,
+                    789.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1247.0,
+                    755.0,
+                    1403.0,
+                    755.0,
+                    1403.0,
+                    789.0,
+                    1247.0,
+                    789.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    786.0,
+                    1286.0,
+                    786.0,
+                    1286.0,
+                    820.0,
+                    294.0,
+                    820.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    925.0,
+                    1405.0,
+                    925.0,
+                    1405.0,
+                    962.0,
+                    294.0,
+                    962.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    960.0,
+                    912.0,
+                    960.0,
+                    912.0,
+                    992.0,
+                    297.0,
+                    992.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    199.0,
+                    1404.0,
+                    199.0,
+                    1404.0,
+                    239.0,
+                    294.0,
+                    239.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    235.0,
+                    1081.0,
+                    235.0,
+                    1081.0,
+                    267.0,
+                    297.0,
+                    267.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    465.0,
+                    1402.0,
+                    465.0,
+                    1402.0,
+                    497.0,
+                    297.0,
+                    497.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    493.0,
+                    1376.0,
+                    493.0,
+                    1376.0,
+                    529.0,
+                    296.0,
+                    529.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1758.0,
+                    1061.0,
+                    1758.0,
+                    1061.0,
+                    1799.0,
+                    294.0,
+                    1799.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1078.0,
+                    1758.0,
+                    1404.0,
+                    1758.0,
+                    1404.0,
+                    1799.0,
+                    1078.0,
+                    1799.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1795.0,
+                    507.0,
+                    1795.0,
+                    507.0,
+                    1837.0,
+                    294.0,
+                    1837.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1131.0,
+                    1795.0,
+                    1137.0,
+                    1795.0,
+                    1137.0,
+                    1837.0,
+                    1131.0,
+                    1837.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 3,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    296,
+                    493,
+                    1405,
+                    493,
+                    1405,
+                    769,
+                    296,
+                    769
+                ],
+                "score": 0.982
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1319,
+                    1405,
+                    1319,
+                    1405,
+                    1613,
+                    297,
+                    1613
+                ],
+                "score": 0.981
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    202,
+                    1405,
+                    202,
+                    1405,
+                    387,
+                    298,
+                    387
+                ],
+                "score": 0.979
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1732,
+                    1405,
+                    1732,
+                    1405,
+                    2008,
+                    297,
+                    2008
+                ],
+                "score": 0.979
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    1626,
+                    1405,
+                    1626,
+                    1405,
+                    1721,
+                    299,
+                    1721
+                ],
+                "score": 0.969
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    298,
+                    434,
+                    910,
+                    434,
+                    910,
+                    468,
+                    298,
+                    468
+                ],
+                "score": 0.929
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    273,
+                    861,
+                    1322,
+                    861,
+                    1322,
+                    1228,
+                    273,
+                    1228
+                ],
+                "score": 0.906
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    841,
+                    2061,
+                    858,
+                    2061,
+                    858,
+                    2085,
+                    841,
+                    2085
+                ],
+                "score": 0.721
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    295,
+                    823,
+                    959,
+                    823,
+                    959,
+                    855,
+                    295,
+                    855
+                ],
+                "score": 0.617
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    298,
+                    1233,
+                    431,
+                    1233,
+                    431,
+                    1263,
+                    298,
+                    1263
+                ],
+                "score": 0.391
+            },
+            {
+                "category_id": 6,
+                "poly": [
+                    295,
+                    823,
+                    959,
+                    823,
+                    959,
+                    855,
+                    295,
+                    855
+                ],
+                "score": 0.171
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1233,
+                    431,
+                    1233,
+                    431,
+                    1263,
+                    298,
+                    1263
+                ],
+                "score": 0.155
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1080,
+                    1443,
+                    1323,
+                    1443,
+                    1323,
+                    1483,
+                    1080,
+                    1483
+                ],
+                "score": 0.94,
+                "latex": "B = 8 \\sigma \\sqrt { \\log ( d n / \\zeta ) }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1192,
+                    1349,
+                    1401,
+                    1349,
+                    1401,
+                    1384,
+                    1192,
+                    1384
+                ],
+                "score": 0.92,
+                "latex": "\\bar { x } + [ - B / 2 , B / 2 ] ^ { d }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    898,
+                    203,
+                    977,
+                    203,
+                    977,
+                    237,
+                    898,
+                    237
+                ],
+                "score": 0.92,
+                "latex": "\\left( \\varepsilon _ { t } , \\delta _ { t } \\right)"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1108,
+                    1734,
+                    1180,
+                    1734,
+                    1180,
+                    1766,
+                    1108,
+                    1766
+                ],
+                "score": 0.92,
+                "latex": "( { \\bar { x } } , B )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    548,
+                    1793,
+                    736,
+                    1793,
+                    736,
+                    1828,
+                    548,
+                    1828
+                ],
+                "score": 0.92,
+                "latex": "\\{ ( \\mu _ { \\ell } , v _ { \\ell } ) \\} _ { \\ell \\in [ t - 1 ] }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    569,
+                    234,
+                    649,
+                    234,
+                    649,
+                    267,
+                    569,
+                    267
+                ],
+                "score": 0.92,
+                "latex": "\\left( \\varepsilon _ { t } , \\delta _ { t } \\right)"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    895,
+                    1916,
+                    951,
+                    1916,
+                    951,
+                    1948,
+                    895,
+                    1948
+                ],
+                "score": 0.91,
+                "latex": "\\{ S _ { \\ell } \\}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1038,
+                    1794,
+                    1180,
+                    1794,
+                    1180,
+                    1829,
+                    1038,
+                    1829
+                ],
+                "score": 0.91,
+                "latex": "\\{ Z _ { \\ell } \\} _ { \\ell \\in [ t - 1 ] } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    920,
+                    1856,
+                    1002,
+                    1856,
+                    1002,
+                    1887,
+                    920,
+                    1887
+                ],
+                "score": 0.91,
+                "latex": "\\hat { \\mu } = \\mu _ { t }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1153,
+                    1509,
+                    1405,
+                    1509,
+                    1405,
+                    1546,
+                    1153,
+                    1546
+                ],
+                "score": 0.91,
+                "latex": "\\mathcal { P } _ { \\bar { x } + [ - B / 2 , B / 2 ] ^ { d } } ( x ) ="
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    844,
+                    863,
+                    955,
+                    863,
+                    955,
+                    897,
+                    844,
+                    897
+                ],
+                "score": 0.91,
+                "latex": "\\zeta \\in ( 0 , 1 )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1268,
+                    1657,
+                    1398,
+                    1657,
+                    1398,
+                    1691,
+                    1268,
+                    1691
+                ],
+                "score": 0.9,
+                "latex": "| S _ { t } | > n / 2 )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    605,
+                    1578,
+                    667,
+                    1578,
+                    667,
+                    1612,
+                    605,
+                    1612
+                ],
+                "score": 0.9,
+                "latex": "B \\sqrt { d }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    847,
+                    1025,
+                    972,
+                    1025,
+                    972,
+                    1058,
+                    847,
+                    1058
+                ],
+                "score": 0.89,
+                "latex": "n _ { t } < 3 n / 4"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    892,
+                    1887,
+                    922,
+                    1887,
+                    922,
+                    1914,
+                    892,
+                    1914
+                ],
+                "score": 0.89,
+                "latex": "Z _ { t }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    339,
+                    1026,
+                    813,
+                    1026,
+                    813,
+                    1061,
+                    339,
+                    1061
+                ],
+                "score": 0.89,
+                "latex": "n _ { t }  q _ { \\mathrm { s i z e } } ( \\{ ( \\mu _ { \\ell } , v _ { \\ell } , Z _ { \\ell } ) \\} _ { \\ell \\in [ t - 1 ] } , \\varepsilon _ { 1 } , \\bar { x } , B )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1034,
+                    1628,
+                    1083,
+                    1628,
+                    1083,
+                    1657,
+                    1034,
+                    1657
+                ],
+                "score": 0.88,
+                "latex": "\\ S 2 . 1"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    870,
+                    326,
+                    900,
+                    326,
+                    900,
+                    355,
+                    870,
+                    355
+                ],
+                "score": 0.88,
+                "latex": "S _ { t }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    740,
+                    497,
+                    769,
+                    497,
+                    769,
+                    526,
+                    740,
+                    526
+                ],
+                "score": 0.88,
+                "latex": "S _ { t }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    299,
+                    898,
+                    669,
+                    898,
+                    669,
+                    931,
+                    299,
+                    931
+                ],
+                "score": 0.88,
+                "latex": "( \\bar { x } , B ) \\gets q _ { \\mathrm { r a n g e } } ( S , 0 . 0 1 \\varepsilon , 0 . 0 1 \\delta )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    413,
+                    1764,
+                    442,
+                    1764,
+                    442,
+                    1794,
+                    413,
+                    1794
+                ],
+                "score": 0.88,
+                "latex": "S _ { t }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    496,
+                    1659,
+                    526,
+                    1659,
+                    526,
+                    1688,
+                    496,
+                    1688
+                ],
+                "score": 0.88,
+                "latex": "S _ { t }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    831,
+                    1451,
+                    902,
+                    1451,
+                    902,
+                    1480,
+                    831,
+                    1480
+                ],
+                "score": 0.88,
+                "latex": "q _ { \\mathrm { r a n g e } }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    601,
+                    527,
+                    630,
+                    527,
+                    630,
+                    556,
+                    601,
+                    556
+                ],
+                "score": 0.88,
+                "latex": "S _ { t }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    516,
+                    356,
+                    545,
+                    356,
+                    545,
+                    385,
+                    516,
+                    385
+                ],
+                "score": 0.88,
+                "latex": "S _ { t }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1239,
+                    1794,
+                    1289,
+                    1794,
+                    1289,
+                    1826,
+                    1239,
+                    1826
+                ],
+                "score": 0.88,
+                "latex": "( i i i )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1243,
+                    1826,
+                    1273,
+                    1826,
+                    1273,
+                    1855,
+                    1243,
+                    1855
+                ],
+                "score": 0.87,
+                "latex": "S _ { t }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    652,
+                    1828,
+                    681,
+                    1828,
+                    681,
+                    1855,
+                    652,
+                    1855
+                ],
+                "score": 0.87,
+                "latex": "S _ { t }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    389,
+                    863,
+                    549,
+                    863,
+                    549,
+                    899,
+                    389,
+                    899
+                ],
+                "score": 0.87,
+                "latex": "S = \\{ x _ { i } \\} _ { i \\in [ n ] }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    561,
+                    863,
+                    703,
+                    863,
+                    703,
+                    897,
+                    561,
+                    897
+                ],
+                "score": 0.87,
+                "latex": "\\alpha \\in ( 0 , 1 / 2 )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1015,
+                    1414,
+                    1047,
+                    1414,
+                    1047,
+                    1446,
+                    1015,
+                    1446
+                ],
+                "score": 0.87,
+                "latex": "\\bar { x } _ { j }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    813,
+                    1692,
+                    958,
+                    1692,
+                    958,
+                    1720,
+                    813,
+                    1720
+                ],
+                "score": 0.87,
+                "latex": "q _ { \\mathrm { m e a n } } , q _ { \\mathrm { n o r m } }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1160,
+                    1765,
+                    1211,
+                    1765,
+                    1211,
+                    1794,
+                    1160,
+                    1794
+                ],
+                "score": 0.86,
+                "latex": "\\ S 2 . 1"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    298,
+                    1977,
+                    326,
+                    1977,
+                    326,
+                    2006,
+                    298,
+                    2006
+                ],
+                "score": 0.86,
+                "latex": "S _ { t }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    577,
+                    1765,
+                    639,
+                    1765,
+                    639,
+                    1792,
+                    577,
+                    1792
+                ],
+                "score": 0.86,
+                "latex": "t - 1"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    339,
+                    999,
+                    485,
+                    999,
+                    485,
+                    1027,
+                    339,
+                    1027
+                ],
+                "score": 0.85,
+                "latex": "t = 1 , \\dots , T"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    748,
+                    531,
+                    778,
+                    531,
+                    778,
+                    557,
+                    748,
+                    557
+                ],
+                "score": 0.85,
+                "latex": "\\mu _ { t }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    846,
+                    297,
+                    876,
+                    297,
+                    876,
+                    325,
+                    846,
+                    325
+                ],
+                "score": 0.85,
+                "latex": "\\mathbf { x } _ { t }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    338,
+                    1062,
+                    828,
+                    1062,
+                    828,
+                    1093,
+                    338,
+                    1093
+                ],
+                "score": 0.85,
+                "latex": "\\mu _ { t }  q _ { \\mathrm { m e a n } } ( \\{ ( \\mu _ { \\ell } , v _ { \\ell } , Z _ { \\ell } ) \\} _ { \\ell \\in [ t - 1 ] } , \\varepsilon _ { 1 } , \\bar { x } , B )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    835,
+                    558,
+                    855,
+                    558,
+                    855,
+                    589,
+                    835,
+                    589
+                ],
+                "score": 0.84,
+                "latex": "\\hat { \\mu }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    689,
+                    1354,
+                    759,
+                    1354,
+                    759,
+                    1384,
+                    689,
+                    1384
+                ],
+                "score": 0.84,
+                "latex": "q _ { \\mathrm { r a n g e } }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    340,
+                    1096,
+                    868,
+                    1096,
+                    868,
+                    1127,
+                    340,
+                    1127
+                ],
+                "score": 0.84,
+                "latex": "\\lambda _ { t } \\gets q _ { \\mathrm { n o r m } } ( \\{ ( \\mu _ { \\ell } , v _ { \\ell } , Z _ { \\ell } ) \\} _ { \\ell \\in [ t - 1 ] } , \\mu _ { t } , \\varepsilon _ { 1 } , \\bar { x } , B )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    343,
+                    1661,
+                    369,
+                    1661,
+                    369,
+                    1688,
+                    343,
+                    1688
+                ],
+                "score": 0.84,
+                "latex": "v _ { t }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    364,
+                    1128,
+                    664,
+                    1128,
+                    664,
+                    1160,
+                    364,
+                    1160
+                ],
+                "score": 0.84,
+                "latex": "\\lambda _ { t } \\le ( C - 0 . 0 1 ) \\alpha \\log { 1 / \\alpha }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    579,
+                    1542,
+                    685,
+                    1542,
+                    685,
+                    1577,
+                    579,
+                    1577
+                ],
+                "score": 0.84,
+                "latex": "\\lVert y - x \\rVert _ { 2 }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    584,
+                    1889,
+                    610,
+                    1889,
+                    610,
+                    1915,
+                    584,
+                    1915
+                ],
+                "score": 0.83,
+                "latex": "v _ { t }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    320,
+                    967,
+                    598,
+                    967,
+                    598,
+                    998,
+                    320,
+                    998
+                ],
+                "score": 0.83,
+                "latex": "n < ( 4 / \\varepsilon _ { 1 } ) \\log ( 1 / ( 2 \\delta _ { 1 } ) )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1113,
+                    530,
+                    1139,
+                    530,
+                    1139,
+                    556,
+                    1113,
+                    556
+                ],
+                "score": 0.83,
+                "latex": "v _ { t }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    339,
+                    1192,
+                    544,
+                    1192,
+                    544,
+                    1225,
+                    339,
+                    1225
+                ],
+                "score": 0.83,
+                "latex": "Z _ { t } \\gets \\mathrm { U n i f } ( [ 0 , 1 ] )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    381,
+                    1449,
+                    398,
+                    1449,
+                    398,
+                    1480,
+                    381,
+                    1480
+                ],
+                "score": 0.82,
+                "latex": "j"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1336,
+                    1825,
+                    1367,
+                    1825,
+                    1367,
+                    1855,
+                    1336,
+                    1855
+                ],
+                "score": 0.81,
+                "latex": "\\lambda _ { t } ,"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1182,
+                    619,
+                    1249,
+                    619,
+                    1249,
+                    648,
+                    1182,
+                    648
+                ],
+                "score": 0.81,
+                "latex": "q _ { \\mathrm { P C A } } ,"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1008,
+                    1693,
+                    1071,
+                    1693,
+                    1071,
+                    1720,
+                    1008,
+                    1720
+                ],
+                "score": 0.81,
+                "latex": "q _ { \\mathrm { P C A } }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1361,
+                    1733,
+                    1403,
+                    1733,
+                    1403,
+                    1767,
+                    1361,
+                    1767
+                ],
+                "score": 0.81,
+                "latex": "( i i )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1371,
+                    1633,
+                    1400,
+                    1633,
+                    1400,
+                    1658,
+                    1371,
+                    1658
+                ],
+                "score": 0.8,
+                "latex": "\\mu _ { t }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1016,
+                    1733,
+                    1048,
+                    1733,
+                    1048,
+                    1766,
+                    1016,
+                    1766
+                ],
+                "score": 0.78,
+                "latex": "( i )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    370,
+                    1736,
+                    384,
+                    1736,
+                    384,
+                    1761,
+                    370,
+                    1761
+                ],
+                "score": 0.76,
+                "latex": "t"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1248,
+                    678,
+                    1298,
+                    678,
+                    1298,
+                    710,
+                    1248,
+                    710
+                ],
+                "score": 0.74,
+                "latex": "( i i i )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    429,
+                    1413,
+                    449,
+                    1413,
+                    449,
+                    1440,
+                    429,
+                    1440
+                ],
+                "score": 0.72,
+                "latex": "d"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    540,
+                    678,
+                    582,
+                    678,
+                    582,
+                    711,
+                    540,
+                    711
+                ],
+                "score": 0.71,
+                "latex": "( i i )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    340,
+                    1160,
+                    913,
+                    1160,
+                    913,
+                    1194,
+                    340,
+                    1194
+                ],
+                "score": 0.71,
+                "latex": "\\boldsymbol { v } _ { t }  q _ { \\mathrm { P C A } } ( \\{ ( \\mu _ { \\ell } , v _ { \\ell } , Z _ { \\ell } ) \\} _ { \\ell \\in [ t - 1 ] } , \\mu _ { t } , \\varepsilon _ { 1 } , \\delta _ { 1 } , \\bar { x } , B ) )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1138,
+                    1025,
+                    1156,
+                    1025,
+                    1156,
+                    1055,
+                    1138,
+                    1055
+                ],
+                "score": 0.69,
+                "latex": "\\varnothing"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1212,
+                    649,
+                    1243,
+                    649,
+                    1243,
+                    681,
+                    1212,
+                    681
+                ],
+                "score": 0.68,
+                "latex": "( i )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1128,
+                    863,
+                    1318,
+                    863,
+                    1318,
+                    897,
+                    1128,
+                    897
+                ],
+                "score": 0.66,
+                "latex": "T = \\Theta ( d ) , ( \\varepsilon , \\delta )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    342,
+                    1543,
+                    569,
+                    1543,
+                    569,
+                    1580,
+                    342,
+                    1580
+                ],
+                "score": 0.66,
+                "latex": "\\begin{array} { r } { \\operatorname* { m i n } _ { y \\in \\bar { x } + [ - B / 2 , B / 2 ] ^ { c } } } \\end{array}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    400,
+                    1235,
+                    430,
+                    1235,
+                    430,
+                    1263,
+                    400,
+                    1263
+                ],
+                "score": 0.64,
+                "latex": "\\mu _ { t }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    296,
+                    933,
+                    759,
+                    933,
+                    759,
+                    967,
+                    296,
+                    967
+                ],
+                "score": 0.62,
+                "latex": "\\varepsilon _ { 1 } \\gets \\operatorname* { m i n } \\{ 0 . 9 9 \\varepsilon , 0 . 9 \\} / ( 4 \\sqrt { 2 T \\log ( 2 / \\delta ) } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1066,
+                    621,
+                    1134,
+                    621,
+                    1134,
+                    649,
+                    1066,
+                    649
+                ],
+                "score": 0.61,
+                "latex": "q _ { \\mathrm { n o r m } }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    989,
+                    622,
+                    1056,
+                    622,
+                    1056,
+                    649,
+                    989,
+                    649
+                ],
+                "score": 0.55,
+                "latex": "q _ { \\mathrm { m e a n } }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    928,
+                    622,
+                    979,
+                    622,
+                    979,
+                    649,
+                    928,
+                    649
+                ],
+                "score": 0.43,
+                "latex": "q _ { \\mathrm { s i z e } }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    831,
+                    1133,
+                    861,
+                    1133,
+                    861,
+                    1159,
+                    831,
+                    1159
+                ],
+                "score": 0.36,
+                "latex": "\\mu _ { t }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    769,
+                    966,
+                    786,
+                    966,
+                    786,
+                    994,
+                    769,
+                    994
+                ],
+                "score": 0.32,
+                "latex": "\\varnothing"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1257,
+                    864,
+                    1318,
+                    864,
+                    1318,
+                    897,
+                    1257,
+                    897
+                ],
+                "score": 0.28,
+                "latex": "( \\varepsilon , \\delta )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    849,
+                    620,
+                    918,
+                    620,
+                    918,
+                    650,
+                    849,
+                    650
+                ],
+                "score": 0.25,
+                "latex": "\\cdot q _ { \\mathrm { r a n g e } }"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    429.0,
+                    914.0,
+                    429.0,
+                    914.0,
+                    475.0,
+                    293.0,
+                    475.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    839.0,
+                    2058.0,
+                    861.0,
+                    2058.0,
+                    861.0,
+                    2093.0,
+                    839.0,
+                    2093.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    821.0,
+                    959.0,
+                    821.0,
+                    959.0,
+                    858.0,
+                    297.0,
+                    858.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1223.0,
+                    399.0,
+                    1223.0,
+                    399.0,
+                    1272.0,
+                    294.0,
+                    1272.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    431.0,
+                    1223.0,
+                    438.0,
+                    1223.0,
+                    438.0,
+                    1272.0,
+                    431.0,
+                    1272.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    821.0,
+                    959.0,
+                    821.0,
+                    959.0,
+                    858.0,
+                    297.0,
+                    858.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    493.0,
+                    739.0,
+                    493.0,
+                    739.0,
+                    530.0,
+                    293.0,
+                    530.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    770.0,
+                    493.0,
+                    1405.0,
+                    493.0,
+                    1405.0,
+                    530.0,
+                    770.0,
+                    530.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    525.0,
+                    600.0,
+                    525.0,
+                    600.0,
+                    562.0,
+                    294.0,
+                    562.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    631.0,
+                    525.0,
+                    747.0,
+                    525.0,
+                    747.0,
+                    562.0,
+                    631.0,
+                    562.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    779.0,
+                    525.0,
+                    1112.0,
+                    525.0,
+                    1112.0,
+                    562.0,
+                    779.0,
+                    562.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1140.0,
+                    525.0,
+                    1405.0,
+                    525.0,
+                    1405.0,
+                    562.0,
+                    1140.0,
+                    562.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    555.0,
+                    834.0,
+                    555.0,
+                    834.0,
+                    593.0,
+                    293.0,
+                    593.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    856.0,
+                    555.0,
+                    1405.0,
+                    555.0,
+                    1405.0,
+                    593.0,
+                    856.0,
+                    593.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    584.0,
+                    1405.0,
+                    584.0,
+                    1405.0,
+                    622.0,
+                    293.0,
+                    622.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    614.0,
+                    848.0,
+                    614.0,
+                    848.0,
+                    658.0,
+                    291.0,
+                    658.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    919.0,
+                    614.0,
+                    927.0,
+                    614.0,
+                    927.0,
+                    658.0,
+                    919.0,
+                    658.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    980.0,
+                    614.0,
+                    988.0,
+                    614.0,
+                    988.0,
+                    658.0,
+                    980.0,
+                    658.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1057.0,
+                    614.0,
+                    1065.0,
+                    614.0,
+                    1065.0,
+                    658.0,
+                    1057.0,
+                    658.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1135.0,
+                    614.0,
+                    1181.0,
+                    614.0,
+                    1181.0,
+                    658.0,
+                    1135.0,
+                    658.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1250.0,
+                    614.0,
+                    1408.0,
+                    614.0,
+                    1408.0,
+                    658.0,
+                    1250.0,
+                    658.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    647.0,
+                    1211.0,
+                    647.0,
+                    1211.0,
+                    685.0,
+                    293.0,
+                    685.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1244.0,
+                    647.0,
+                    1403.0,
+                    647.0,
+                    1403.0,
+                    685.0,
+                    1244.0,
+                    685.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    290.0,
+                    671.0,
+                    539.0,
+                    671.0,
+                    539.0,
+                    719.0,
+                    290.0,
+                    719.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    583.0,
+                    671.0,
+                    1247.0,
+                    671.0,
+                    1247.0,
+                    719.0,
+                    583.0,
+                    719.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1299.0,
+                    671.0,
+                    1407.0,
+                    671.0,
+                    1407.0,
+                    719.0,
+                    1299.0,
+                    719.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    707.0,
+                    1405.0,
+                    707.0,
+                    1405.0,
+                    743.0,
+                    294.0,
+                    743.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    738.0,
+                    864.0,
+                    738.0,
+                    864.0,
+                    773.0,
+                    294.0,
+                    773.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1318.0,
+                    1407.0,
+                    1318.0,
+                    1407.0,
+                    1356.0,
+                    294.0,
+                    1356.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    290.0,
+                    1345.0,
+                    688.0,
+                    1345.0,
+                    688.0,
+                    1391.0,
+                    290.0,
+                    1391.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    760.0,
+                    1345.0,
+                    1191.0,
+                    1345.0,
+                    1191.0,
+                    1391.0,
+                    760.0,
+                    1391.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1402.0,
+                    1345.0,
+                    1407.0,
+                    1345.0,
+                    1407.0,
+                    1391.0,
+                    1402.0,
+                    1391.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1379.0,
+                    1406.0,
+                    1379.0,
+                    1406.0,
+                    1416.0,
+                    292.0,
+                    1416.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1411.0,
+                    428.0,
+                    1411.0,
+                    428.0,
+                    1448.0,
+                    292.0,
+                    1448.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    450.0,
+                    1411.0,
+                    1014.0,
+                    1411.0,
+                    1014.0,
+                    1448.0,
+                    450.0,
+                    1448.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1048.0,
+                    1411.0,
+                    1405.0,
+                    1411.0,
+                    1405.0,
+                    1448.0,
+                    1048.0,
+                    1448.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1444.0,
+                    380.0,
+                    1444.0,
+                    380.0,
+                    1486.0,
+                    291.0,
+                    1486.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    399.0,
+                    1444.0,
+                    830.0,
+                    1444.0,
+                    830.0,
+                    1486.0,
+                    399.0,
+                    1486.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    903.0,
+                    1444.0,
+                    1079.0,
+                    1444.0,
+                    1079.0,
+                    1486.0,
+                    903.0,
+                    1486.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1324.0,
+                    1444.0,
+                    1406.0,
+                    1444.0,
+                    1406.0,
+                    1486.0,
+                    1324.0,
+                    1486.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1475.0,
+                    1407.0,
+                    1475.0,
+                    1407.0,
+                    1518.0,
+                    291.0,
+                    1518.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1505.0,
+                    1152.0,
+                    1505.0,
+                    1152.0,
+                    1552.0,
+                    292.0,
+                    1552.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1541.0,
+                    341.0,
+                    1541.0,
+                    341.0,
+                    1580.0,
+                    291.0,
+                    1580.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    570.0,
+                    1541.0,
+                    578.0,
+                    1541.0,
+                    578.0,
+                    1580.0,
+                    570.0,
+                    1580.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    686.0,
+                    1541.0,
+                    1406.0,
+                    1541.0,
+                    1406.0,
+                    1580.0,
+                    686.0,
+                    1580.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1579.0,
+                    604.0,
+                    1579.0,
+                    604.0,
+                    1615.0,
+                    293.0,
+                    1615.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    668.0,
+                    1579.0,
+                    678.0,
+                    1579.0,
+                    678.0,
+                    1615.0,
+                    668.0,
+                    1615.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    203.0,
+                    897.0,
+                    203.0,
+                    897.0,
+                    239.0,
+                    296.0,
+                    239.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    978.0,
+                    203.0,
+                    1405.0,
+                    203.0,
+                    1405.0,
+                    239.0,
+                    978.0,
+                    239.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    233.0,
+                    568.0,
+                    233.0,
+                    568.0,
+                    269.0,
+                    295.0,
+                    269.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    650.0,
+                    233.0,
+                    1405.0,
+                    233.0,
+                    1405.0,
+                    269.0,
+                    650.0,
+                    269.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    261.0,
+                    1405.0,
+                    261.0,
+                    1405.0,
+                    300.0,
+                    292.0,
+                    300.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    289.0,
+                    845.0,
+                    289.0,
+                    845.0,
+                    334.0,
+                    292.0,
+                    334.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    877.0,
+                    289.0,
+                    1406.0,
+                    289.0,
+                    1406.0,
+                    334.0,
+                    877.0,
+                    334.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    321.0,
+                    869.0,
+                    321.0,
+                    869.0,
+                    363.0,
+                    293.0,
+                    363.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    901.0,
+                    321.0,
+                    1409.0,
+                    321.0,
+                    1409.0,
+                    363.0,
+                    901.0,
+                    363.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    353.0,
+                    515.0,
+                    353.0,
+                    515.0,
+                    391.0,
+                    293.0,
+                    391.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    546.0,
+                    353.0,
+                    1328.0,
+                    353.0,
+                    1328.0,
+                    391.0,
+                    546.0,
+                    391.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1733.0,
+                    369.0,
+                    1733.0,
+                    369.0,
+                    1769.0,
+                    294.0,
+                    1769.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    385.0,
+                    1733.0,
+                    1015.0,
+                    1733.0,
+                    1015.0,
+                    1769.0,
+                    385.0,
+                    1769.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1049.0,
+                    1733.0,
+                    1107.0,
+                    1733.0,
+                    1107.0,
+                    1769.0,
+                    1049.0,
+                    1769.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1181.0,
+                    1733.0,
+                    1360.0,
+                    1733.0,
+                    1360.0,
+                    1769.0,
+                    1181.0,
+                    1769.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1762.0,
+                    412.0,
+                    1762.0,
+                    412.0,
+                    1800.0,
+                    292.0,
+                    1800.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    443.0,
+                    1762.0,
+                    576.0,
+                    1762.0,
+                    576.0,
+                    1800.0,
+                    443.0,
+                    1800.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    640.0,
+                    1762.0,
+                    1159.0,
+                    1762.0,
+                    1159.0,
+                    1800.0,
+                    640.0,
+                    1800.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1212.0,
+                    1762.0,
+                    1406.0,
+                    1762.0,
+                    1406.0,
+                    1800.0,
+                    1212.0,
+                    1800.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1790.0,
+                    547.0,
+                    1790.0,
+                    547.0,
+                    1835.0,
+                    291.0,
+                    1835.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    737.0,
+                    1790.0,
+                    1037.0,
+                    1790.0,
+                    1037.0,
+                    1835.0,
+                    737.0,
+                    1835.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1181.0,
+                    1790.0,
+                    1238.0,
+                    1790.0,
+                    1238.0,
+                    1835.0,
+                    1181.0,
+                    1835.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1290.0,
+                    1790.0,
+                    1408.0,
+                    1790.0,
+                    1408.0,
+                    1835.0,
+                    1290.0,
+                    1835.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1823.0,
+                    651.0,
+                    1823.0,
+                    651.0,
+                    1860.0,
+                    295.0,
+                    1860.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    682.0,
+                    1823.0,
+                    1242.0,
+                    1823.0,
+                    1242.0,
+                    1860.0,
+                    682.0,
+                    1860.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1274.0,
+                    1823.0,
+                    1335.0,
+                    1823.0,
+                    1335.0,
+                    1860.0,
+                    1274.0,
+                    1860.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1368.0,
+                    1823.0,
+                    1406.0,
+                    1823.0,
+                    1406.0,
+                    1860.0,
+                    1368.0,
+                    1860.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1854.0,
+                    919.0,
+                    1854.0,
+                    919.0,
+                    1891.0,
+                    294.0,
+                    1891.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1003.0,
+                    1854.0,
+                    1405.0,
+                    1854.0,
+                    1405.0,
+                    1891.0,
+                    1003.0,
+                    1891.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1884.0,
+                    583.0,
+                    1884.0,
+                    583.0,
+                    1920.0,
+                    294.0,
+                    1920.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    611.0,
+                    1884.0,
+                    891.0,
+                    1884.0,
+                    891.0,
+                    1920.0,
+                    611.0,
+                    1920.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    923.0,
+                    1884.0,
+                    1405.0,
+                    1884.0,
+                    1405.0,
+                    1920.0,
+                    923.0,
+                    1920.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1916.0,
+                    894.0,
+                    1916.0,
+                    894.0,
+                    1948.0,
+                    296.0,
+                    1948.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    952.0,
+                    1916.0,
+                    1405.0,
+                    1916.0,
+                    1405.0,
+                    1948.0,
+                    952.0,
+                    1948.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1942.0,
+                    1405.0,
+                    1942.0,
+                    1405.0,
+                    1984.0,
+                    292.0,
+                    1984.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    327.0,
+                    1977.0,
+                    1406.0,
+                    1977.0,
+                    1406.0,
+                    2010.0,
+                    327.0,
+                    2010.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1620.0,
+                    1033.0,
+                    1620.0,
+                    1033.0,
+                    1666.0,
+                    292.0,
+                    1666.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1084.0,
+                    1620.0,
+                    1370.0,
+                    1620.0,
+                    1370.0,
+                    1666.0,
+                    1084.0,
+                    1666.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1401.0,
+                    1620.0,
+                    1406.0,
+                    1620.0,
+                    1406.0,
+                    1666.0,
+                    1401.0,
+                    1666.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1655.0,
+                    342.0,
+                    1655.0,
+                    342.0,
+                    1693.0,
+                    294.0,
+                    1693.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    370.0,
+                    1655.0,
+                    495.0,
+                    1655.0,
+                    495.0,
+                    1693.0,
+                    370.0,
+                    1693.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    527.0,
+                    1655.0,
+                    1267.0,
+                    1655.0,
+                    1267.0,
+                    1693.0,
+                    527.0,
+                    1693.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1399.0,
+                    1655.0,
+                    1406.0,
+                    1655.0,
+                    1406.0,
+                    1693.0,
+                    1399.0,
+                    1693.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1686.0,
+                    812.0,
+                    1686.0,
+                    812.0,
+                    1726.0,
+                    293.0,
+                    1726.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    959.0,
+                    1686.0,
+                    1007.0,
+                    1686.0,
+                    1007.0,
+                    1726.0,
+                    959.0,
+                    1726.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1072.0,
+                    1686.0,
+                    1084.0,
+                    1686.0,
+                    1084.0,
+                    1726.0,
+                    1072.0,
+                    1726.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    856.0,
+                    388.0,
+                    856.0,
+                    388.0,
+                    900.0,
+                    294.0,
+                    900.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    550.0,
+                    856.0,
+                    560.0,
+                    856.0,
+                    560.0,
+                    900.0,
+                    550.0,
+                    900.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    704.0,
+                    856.0,
+                    843.0,
+                    856.0,
+                    843.0,
+                    900.0,
+                    704.0,
+                    900.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    956.0,
+                    856.0,
+                    1127.0,
+                    856.0,
+                    1127.0,
+                    900.0,
+                    956.0,
+                    900.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1319.0,
+                    856.0,
+                    1323.0,
+                    856.0,
+                    1323.0,
+                    900.0,
+                    1319.0,
+                    900.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    271.0,
+                    894.0,
+                    298.0,
+                    894.0,
+                    298.0,
+                    933.0,
+                    271.0,
+                    933.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    670.0,
+                    894.0,
+                    673.0,
+                    894.0,
+                    673.0,
+                    933.0,
+                    670.0,
+                    933.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    267.0,
+                    929.0,
+                    295.0,
+                    929.0,
+                    295.0,
+                    970.0,
+                    267.0,
+                    970.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    760.0,
+                    929.0,
+                    989.0,
+                    929.0,
+                    989.0,
+                    970.0,
+                    760.0,
+                    970.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    269.0,
+                    961.0,
+                    319.0,
+                    961.0,
+                    319.0,
+                    1000.0,
+                    269.0,
+                    1000.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    599.0,
+                    961.0,
+                    768.0,
+                    961.0,
+                    768.0,
+                    1000.0,
+                    599.0,
+                    1000.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    787.0,
+                    961.0,
+                    792.0,
+                    961.0,
+                    792.0,
+                    1000.0,
+                    787.0,
+                    1000.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    271.0,
+                    995.0,
+                    338.0,
+                    995.0,
+                    338.0,
+                    1025.0,
+                    271.0,
+                    1025.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    486.0,
+                    995.0,
+                    525.0,
+                    995.0,
+                    525.0,
+                    1025.0,
+                    486.0,
+                    1025.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    273.0,
+                    1030.0,
+                    292.0,
+                    1030.0,
+                    292.0,
+                    1055.0,
+                    273.0,
+                    1055.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    332.0,
+                    1023.0,
+                    338.0,
+                    1023.0,
+                    338.0,
+                    1060.0,
+                    332.0,
+                    1060.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    814.0,
+                    1023.0,
+                    846.0,
+                    1023.0,
+                    846.0,
+                    1060.0,
+                    814.0,
+                    1060.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    973.0,
+                    1023.0,
+                    1137.0,
+                    1023.0,
+                    1137.0,
+                    1060.0,
+                    973.0,
+                    1060.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1157.0,
+                    1023.0,
+                    1160.0,
+                    1023.0,
+                    1160.0,
+                    1060.0,
+                    1157.0,
+                    1060.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    273.0,
+                    1065.0,
+                    292.0,
+                    1065.0,
+                    292.0,
+                    1089.0,
+                    273.0,
+                    1089.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    334.0,
+                    1058.0,
+                    337.0,
+                    1058.0,
+                    337.0,
+                    1094.0,
+                    334.0,
+                    1094.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    829.0,
+                    1058.0,
+                    833.0,
+                    1058.0,
+                    833.0,
+                    1094.0,
+                    829.0,
+                    1094.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    273.0,
+                    1099.0,
+                    292.0,
+                    1099.0,
+                    292.0,
+                    1124.0,
+                    273.0,
+                    1124.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    334.0,
+                    1091.0,
+                    339.0,
+                    1091.0,
+                    339.0,
+                    1132.0,
+                    334.0,
+                    1132.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    273.0,
+                    1133.0,
+                    292.0,
+                    1133.0,
+                    292.0,
+                    1158.0,
+                    273.0,
+                    1158.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    334.0,
+                    1122.0,
+                    363.0,
+                    1122.0,
+                    363.0,
+                    1164.0,
+                    334.0,
+                    1164.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    665.0,
+                    1122.0,
+                    830.0,
+                    1122.0,
+                    830.0,
+                    1164.0,
+                    665.0,
+                    1164.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    862.0,
+                    1122.0,
+                    866.0,
+                    1122.0,
+                    866.0,
+                    1164.0,
+                    862.0,
+                    1164.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    272.0,
+                    1162.0,
+                    292.0,
+                    1162.0,
+                    292.0,
+                    1186.0,
+                    272.0,
+                    1186.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    333.0,
+                    1155.0,
+                    339.0,
+                    1155.0,
+                    339.0,
+                    1193.0,
+                    333.0,
+                    1193.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    267.0,
+                    1195.0,
+                    294.0,
+                    1195.0,
+                    294.0,
+                    1222.0,
+                    267.0,
+                    1222.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    334.0,
+                    1189.0,
+                    338.0,
+                    1189.0,
+                    338.0,
+                    1228.0,
+                    334.0,
+                    1228.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    545.0,
+                    1189.0,
+                    550.0,
+                    1189.0,
+                    550.0,
+                    1228.0,
+                    545.0,
+                    1228.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1223.0,
+                    399.0,
+                    1223.0,
+                    399.0,
+                    1272.0,
+                    294.0,
+                    1272.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    431.0,
+                    1223.0,
+                    438.0,
+                    1223.0,
+                    438.0,
+                    1272.0,
+                    431.0,
+                    1272.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 4,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1191,
+                    1406,
+                    1191,
+                    1406,
+                    1498,
+                    297,
+                    1498
+                ],
+                "score": 0.984
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1580,
+                    1406,
+                    1580,
+                    1406,
+                    1745,
+                    298,
+                    1745
+                ],
+                "score": 0.98
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    946,
+                    1405,
+                    946,
+                    1405,
+                    1104,
+                    297,
+                    1104
+                ],
+                "score": 0.977
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    621,
+                    1405,
+                    621,
+                    1405,
+                    809,
+                    297,
+                    809
+                ],
+                "score": 0.977
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    278,
+                    1404,
+                    278,
+                    1404,
+                    466,
+                    297,
+                    466
+                ],
+                "score": 0.977
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1879,
+                    1404,
+                    1879,
+                    1404,
+                    2008,
+                    298,
+                    2008
+                ],
+                "score": 0.976
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    814,
+                    1404,
+                    814,
+                    1404,
+                    925,
+                    297,
+                    925
+                ],
+                "score": 0.973
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    472,
+                    1404,
+                    472,
+                    1404,
+                    603,
+                    297,
+                    603
+                ],
+                "score": 0.97
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    530,
+                    1758,
+                    1110,
+                    1758,
+                    1110,
+                    1865,
+                    530,
+                    1865
+                ],
+                "score": 0.964
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    202,
+                    1402,
+                    202,
+                    1402,
+                    265,
+                    297,
+                    265
+                ],
+                "score": 0.95
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    297,
+                    1527,
+                    931,
+                    1527,
+                    931,
+                    1562,
+                    297,
+                    1562
+                ],
+                "score": 0.938
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    296,
+                    1136,
+                    951,
+                    1136,
+                    951,
+                    1170,
+                    296,
+                    1170
+                ],
+                "score": 0.928
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    840,
+                    2061,
+                    859,
+                    2061,
+                    859,
+                    2085,
+                    840,
+                    2085
+                ],
+                "score": 0.776
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    533,
+                    1756,
+                    1110,
+                    1756,
+                    1110,
+                    1865,
+                    533,
+                    1865
+                ],
+                "score": 0.94,
+                "latex": "U _ { t } ^ { ( s ) } = \\frac { \\exp \\left( \\alpha ^ { ( s ) } \\sum _ { r \\in [ t ] } ( \\mathrm { C o v } ( S _ { r } ^ { ( s ) } ) - { \\bf I } ) \\right) } { \\mathrm { T r } \\big ( \\exp ( \\alpha ^ { ( s ) } \\sum _ { r \\in [ t ] } ( \\mathrm { C o v } ( S _ { r } ^ { ( s ) } ) - { \\bf I } ) ) \\big ) } ,"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1304,
+                    1611,
+                    1403,
+                    1611,
+                    1403,
+                    1645,
+                    1304,
+                    1645
+                ],
+                "score": 0.93,
+                "latex": "O ( \\log d )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    927,
+                    623,
+                    1095,
+                    623,
+                    1095,
+                    656,
+                    927,
+                    656
+                ],
+                "score": 0.93,
+                "latex": "d _ { \\triangle } ( S , S ^ { \\prime } ) \\leq 1"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    637,
+                    847,
+                    1101,
+                    847,
+                    1101,
+                    888,
+                    637,
+                    888
+                ],
+                "score": 0.93,
+                "latex": "n = \\widetilde \\Omega \\left( ( d / \\alpha ^ { 2 } ) + d ^ { 2 } ( \\log ( 1 / \\delta ) ) ^ { 3 / 2 } / ( \\varepsilon \\alpha ) \\right)"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1061,
+                    279,
+                    1192,
+                    279,
+                    1192,
+                    315,
+                    1061,
+                    315
+                ],
+                "score": 0.93,
+                "latex": "\\{ S _ { \\ell } \\} _ { \\ell \\in [ t - 1 ] }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    581,
+                    1949,
+                    706,
+                    1949,
+                    706,
+                    1980,
+                    581,
+                    1980
+                ],
+                "score": 0.93,
+                "latex": "\\tau _ { i } = \\| x _ { i } \\| _ { 2 } ^ { 2 }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    327,
+                    1070,
+                    460,
+                    1070,
+                    460,
+                    1104,
+                    327,
+                    1104
+                ],
+                "score": 0.93,
+                "latex": "{ \\bar { O } } ( ( \\log { d } ) ^ { 2 } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    551,
+                    1343,
+                    725,
+                    1343,
+                    725,
+                    1378,
+                    551,
+                    1378
+                ],
+                "score": 0.93,
+                "latex": "d B ^ { 2 } Z _ { t } = { \\cal { O } } ( d )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    457,
+                    948,
+                    561,
+                    948,
+                    561,
+                    983,
+                    457,
+                    983
+                ],
+                "score": 0.93,
+                "latex": "O ( d / \\alpha ^ { 2 } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1003,
+                    1673,
+                    1056,
+                    1673,
+                    1056,
+                    1713,
+                    1003,
+                    1713
+                ],
+                "score": 0.92,
+                "latex": "U _ { t } ^ { ( s ) }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    815,
+                    1612,
+                    915,
+                    1612,
+                    915,
+                    1644,
+                    815,
+                    1644
+                ],
+                "score": 0.92,
+                "latex": "O ( \\log d )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    685,
+                    1405,
+                    736,
+                    1405,
+                    736,
+                    1439,
+                    685,
+                    1439
+                ],
+                "score": 0.92,
+                "latex": "\\{ \\tau _ { i } \\}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    679,
+                    537,
+                    1045,
+                    537,
+                    1045,
+                    571,
+                    679,
+                    571
+                ],
+                "score": 0.92,
+                "latex": "d _ { \\triangle } ( S _ { t } ( S ) , S _ { t } ( S ^ { \\prime } ) ) \\leq d _ { \\triangle } ( S , S ^ { \\prime } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    506,
+                    1878,
+                    607,
+                    1878,
+                    607,
+                    1910,
+                    506,
+                    1910
+                ],
+                "score": 0.92,
+                "latex": "\\alpha ^ { ( s ) } > 0"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    776,
+                    1254,
+                    834,
+                    1254,
+                    834,
+                    1287,
+                    776,
+                    1287
+                ],
+                "score": 0.92,
+                "latex": "\\Omega ( d )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    816,
+                    1314,
+                    876,
+                    1314,
+                    876,
+                    1347,
+                    816,
+                    1347
+                ],
+                "score": 0.92,
+                "latex": "O ( d )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    298,
+                    887,
+                    634,
+                    887,
+                    634,
+                    925,
+                    298,
+                    925
+                ],
+                "score": 0.92,
+                "latex": "\\| \\hat { \\mu } - \\mu \\| _ { 2 } \\leq O ( \\alpha \\sqrt { \\log ( 1 / \\alpha ) } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    440,
+                    1284,
+                    500,
+                    1284,
+                    500,
+                    1317,
+                    440,
+                    1317
+                ],
+                "score": 0.92,
+                "latex": "O ( d )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    735,
+                    1980,
+                    833,
+                    1980,
+                    833,
+                    2009,
+                    735,
+                    2009
+                ],
+                "score": 0.92,
+                "latex": "O ( \\log d )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    501,
+                    473,
+                    571,
+                    473,
+                    571,
+                    507,
+                    501,
+                    507
+                ],
+                "score": 0.92,
+                "latex": "S _ { t } ( S )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1076,
+                    1193,
+                    1137,
+                    1193,
+                    1137,
+                    1226,
+                    1076,
+                    1226
+                ],
+                "score": 0.92,
+                "latex": "O ( d )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1189,
+                    1878,
+                    1303,
+                    1878,
+                    1303,
+                    1909,
+                    1189,
+                    1909
+                ],
+                "score": 0.91,
+                "latex": "\\alpha ^ { ( s ) } = \\infty"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    298,
+                    536,
+                    492,
+                    536,
+                    492,
+                    571,
+                    298,
+                    571
+                ],
+                "score": 0.91,
+                "latex": "\\{ ( \\mu _ { \\ell } , v _ { \\ell } , Z _ { \\ell } ) \\} _ { \\ell = 1 } ^ { t }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1032,
+                    979,
+                    1092,
+                    979,
+                    1092,
+                    1012,
+                    1032,
+                    1012
+                ],
+                "score": 0.91,
+                "latex": "O ( d )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    636,
+                    715,
+                    720,
+                    715,
+                    720,
+                    747,
+                    636,
+                    747
+                ],
+                "score": 0.91,
+                "latex": "( \\varepsilon _ { 1 } , \\delta _ { 1 } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    619,
+                    815,
+                    680,
+                    815,
+                    680,
+                    848,
+                    619,
+                    848
+                ],
+                "score": 0.9,
+                "latex": "( \\varepsilon , \\delta )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    297,
+                    570,
+                    447,
+                    570,
+                    447,
+                    603,
+                    297,
+                    603
+                ],
+                "score": 0.9,
+                "latex": "S ^ { \\prime } | , | S ^ { \\prime } \\backslash S | \\}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    798,
+                    623,
+                    883,
+                    623,
+                    883,
+                    653,
+                    798,
+                    653
+                ],
+                "score": 0.9,
+                "latex": "\\boldsymbol { S } \\sim \\boldsymbol { S } ^ { \\prime }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1132,
+                    536,
+                    1404,
+                    536,
+                    1404,
+                    571,
+                    1132,
+                    571
+                ],
+                "score": 0.9,
+                "latex": "d _ { \\triangle } ( S , S ^ { \\prime } ) \\triangleq \\operatorname* { m a x } \\{ | S \\ \\backslash"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1005,
+                    1945,
+                    1052,
+                    1945,
+                    1052,
+                    1973,
+                    1005,
+                    1973
+                ],
+                "score": 0.9,
+                "latex": "\\alpha ^ { ( s ) }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    362,
+                    1674,
+                    923,
+                    1674,
+                    923,
+                    1714,
+                    362,
+                    1714
+                ],
+                "score": 0.89,
+                "latex": "\\tau _ { i } = ( x _ { i } - \\mathrm { M e a n } ( S _ { t } ^ { ( s ) } ) ) ^ { T } U _ { t } ^ { ( s ) } ( x _ { i } - \\mathrm { M e a n } ( S _ { t } ^ { ( s ) } ) )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1215,
+                    1910,
+                    1317,
+                    1910,
+                    1317,
+                    1940,
+                    1215,
+                    1940
+                ],
+                "score": 0.89,
+                "latex": "\\alpha ^ { ( s ) } = 0"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    326,
+                    1711,
+                    362,
+                    1711,
+                    362,
+                    1740,
+                    326,
+                    1740
+                ],
+                "score": 0.89,
+                "latex": "\\mathbb { R } ^ { d }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1278,
+                    747,
+                    1347,
+                    747,
+                    1347,
+                    778,
+                    1278,
+                    778
+                ],
+                "score": 0.87,
+                "latex": "q _ { \\mathrm { r a n g e } }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    514,
+                    853,
+                    585,
+                    853,
+                    585,
+                    884,
+                    514,
+                    884
+                ],
+                "score": 0.86,
+                "latex": "\\alpha \\leq c"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1124,
+                    1643,
+                    1175,
+                    1643,
+                    1175,
+                    1674,
+                    1124,
+                    1674
+                ],
+                "score": 0.86,
+                "latex": "\\ S 2 . 1"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    966,
+                    1229,
+                    993,
+                    1229,
+                    993,
+                    1254,
+                    966,
+                    1254
+                ],
+                "score": 0.85,
+                "latex": "v _ { t }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    422,
+                    313,
+                    484,
+                    313,
+                    484,
+                    342,
+                    422,
+                    342
+                ],
+                "score": 0.85,
+                "latex": "t - 1"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    537,
+                    746,
+                    694,
+                    746,
+                    694,
+                    777,
+                    537,
+                    777
+                ],
+                "score": 0.82,
+                "latex": "( 0 . 9 9 \\varepsilon , 0 . 9 9 \\delta )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1164,
+                    373,
+                    1205,
+                    373,
+                    1205,
+                    405,
+                    1164,
+                    405
+                ],
+                "score": 0.82,
+                "latex": "( i i )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1358,
+                    818,
+                    1404,
+                    818,
+                    1404,
+                    846,
+                    1358,
+                    846
+                ],
+                "score": 0.79,
+                "latex": "c \\in"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1108,
+                    506,
+                    1130,
+                    506,
+                    1130,
+                    533,
+                    1108,
+                    533
+                ],
+                "score": 0.76,
+                "latex": "s"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    748,
+                    508,
+                    817,
+                    508,
+                    817,
+                    536,
+                    748,
+                    536
+                ],
+                "score": 0.76,
+                "latex": "q _ { \\mathrm { P C A } } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1383,
+                    950,
+                    1402,
+                    950,
+                    1402,
+                    977,
+                    1383,
+                    977
+                ],
+                "score": 0.76,
+                "latex": "d"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    983,
+                    1912,
+                    1033,
+                    1912,
+                    1033,
+                    1942,
+                    983,
+                    1942
+                ],
+                "score": 0.75,
+                "latex": "\\ S 2 . 1"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    748,
+                    1647,
+                    759,
+                    1647,
+                    759,
+                    1670,
+                    748,
+                    1670
+                ],
+                "score": 0.75,
+                "latex": "t"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    907,
+                    745,
+                    1065,
+                    745,
+                    1065,
+                    778,
+                    907,
+                    778
+                ],
+                "score": 0.75,
+                "latex": "( 0 . 0 1 \\varepsilon , 0 . 0 1 \\delta )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    327,
+                    745,
+                    365,
+                    745,
+                    365,
+                    773,
+                    327,
+                    773
+                ],
+                "score": 0.74,
+                "latex": "4 T"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    585,
+                    1649,
+                    600,
+                    1649,
+                    600,
+                    1669,
+                    585,
+                    1669
+                ],
+                "score": 0.74,
+                "latex": "s"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    617,
+                    510,
+                    683,
+                    510,
+                    683,
+                    537,
+                    617,
+                    537
+                ],
+                "score": 0.66,
+                "latex": "q _ { \\mathrm { n o r m } }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    471,
+                    508,
+                    522,
+                    508,
+                    522,
+                    536,
+                    471,
+                    536
+                ],
+                "score": 0.63,
+                "latex": "( q _ { \\mathrm { s i z e } }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    296,
+                    851,
+                    379,
+                    851,
+                    379,
+                    886,
+                    296,
+                    886
+                ],
+                "score": 0.62,
+                "latex": "( 0 , 0 . 1 )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1371,
+                    313,
+                    1403,
+                    313,
+                    1403,
+                    346,
+                    1371,
+                    346
+                ],
+                "score": 0.43,
+                "latex": "( i )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    536,
+                    510,
+                    602,
+                    510,
+                    602,
+                    537,
+                    536,
+                    537
+                ],
+                "score": 0.43,
+                "latex": "q _ { \\mathrm { m e a n } }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    570,
+                    817,
+                    586,
+                    817,
+                    586,
+                    844,
+                    570,
+                    844
+                ],
+                "score": 0.42,
+                "latex": "^ { l }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    956,
+                    818,
+                    972,
+                    818,
+                    972,
+                    843,
+                    956,
+                    843
+                ],
+                "score": 0.36,
+                "latex": "^ { l }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    996,
+                    1437,
+                    1024,
+                    1437,
+                    1024,
+                    1467,
+                    996,
+                    1467
+                ],
+                "score": 0.31,
+                "latex": "( \\cdot )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1093,
+                    477,
+                    1108,
+                    477,
+                    1108,
+                    501,
+                    1093,
+                    501
+                ],
+                "score": 0.28,
+                "latex": "t"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1522.0,
+                    935.0,
+                    1522.0,
+                    935.0,
+                    1572.0,
+                    291.0,
+                    1572.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1133.0,
+                    954.0,
+                    1133.0,
+                    954.0,
+                    1174.0,
+                    293.0,
+                    1174.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    840.0,
+                    2059.0,
+                    862.0,
+                    2059.0,
+                    862.0,
+                    2089.0,
+                    840.0,
+                    2089.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1191.0,
+                    1075.0,
+                    1191.0,
+                    1075.0,
+                    1230.0,
+                    292.0,
+                    1230.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1138.0,
+                    1191.0,
+                    1404.0,
+                    1191.0,
+                    1404.0,
+                    1230.0,
+                    1138.0,
+                    1230.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1221.0,
+                    965.0,
+                    1221.0,
+                    965.0,
+                    1259.0,
+                    292.0,
+                    1259.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    994.0,
+                    1221.0,
+                    1408.0,
+                    1221.0,
+                    1408.0,
+                    1259.0,
+                    994.0,
+                    1259.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1253.0,
+                    775.0,
+                    1253.0,
+                    775.0,
+                    1289.0,
+                    295.0,
+                    1289.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    835.0,
+                    1253.0,
+                    1406.0,
+                    1253.0,
+                    1406.0,
+                    1289.0,
+                    835.0,
+                    1289.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1286.0,
+                    439.0,
+                    1286.0,
+                    439.0,
+                    1318.0,
+                    295.0,
+                    1318.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    501.0,
+                    1286.0,
+                    1404.0,
+                    1286.0,
+                    1404.0,
+                    1318.0,
+                    501.0,
+                    1318.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1309.0,
+                    815.0,
+                    1309.0,
+                    815.0,
+                    1353.0,
+                    292.0,
+                    1353.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    877.0,
+                    1309.0,
+                    1406.0,
+                    1309.0,
+                    1406.0,
+                    1353.0,
+                    877.0,
+                    1353.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1342.0,
+                    550.0,
+                    1342.0,
+                    550.0,
+                    1381.0,
+                    292.0,
+                    1381.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    726.0,
+                    1342.0,
+                    1404.0,
+                    1342.0,
+                    1404.0,
+                    1381.0,
+                    726.0,
+                    1381.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1374.0,
+                    1404.0,
+                    1374.0,
+                    1404.0,
+                    1411.0,
+                    294.0,
+                    1411.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1406.0,
+                    684.0,
+                    1406.0,
+                    684.0,
+                    1441.0,
+                    294.0,
+                    1441.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    737.0,
+                    1406.0,
+                    1404.0,
+                    1406.0,
+                    1404.0,
+                    1441.0,
+                    737.0,
+                    1441.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1435.0,
+                    995.0,
+                    1435.0,
+                    995.0,
+                    1472.0,
+                    292.0,
+                    1472.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1025.0,
+                    1435.0,
+                    1407.0,
+                    1435.0,
+                    1407.0,
+                    1472.0,
+                    1025.0,
+                    1472.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1467.0,
+                    687.0,
+                    1467.0,
+                    687.0,
+                    1498.0,
+                    295.0,
+                    1498.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1578.0,
+                    1407.0,
+                    1578.0,
+                    1407.0,
+                    1616.0,
+                    293.0,
+                    1616.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1610.0,
+                    814.0,
+                    1610.0,
+                    814.0,
+                    1647.0,
+                    295.0,
+                    1647.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    916.0,
+                    1610.0,
+                    1303.0,
+                    1610.0,
+                    1303.0,
+                    1647.0,
+                    916.0,
+                    1647.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1404.0,
+                    1610.0,
+                    1407.0,
+                    1610.0,
+                    1407.0,
+                    1647.0,
+                    1404.0,
+                    1647.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1641.0,
+                    584.0,
+                    1641.0,
+                    584.0,
+                    1678.0,
+                    295.0,
+                    1678.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    601.0,
+                    1641.0,
+                    747.0,
+                    1641.0,
+                    747.0,
+                    1678.0,
+                    601.0,
+                    1678.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    760.0,
+                    1641.0,
+                    1123.0,
+                    1641.0,
+                    1123.0,
+                    1678.0,
+                    760.0,
+                    1678.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1176.0,
+                    1641.0,
+                    1406.0,
+                    1641.0,
+                    1406.0,
+                    1678.0,
+                    1176.0,
+                    1678.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    288.0,
+                    1670.0,
+                    361.0,
+                    1670.0,
+                    361.0,
+                    1717.0,
+                    288.0,
+                    1717.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    924.0,
+                    1670.0,
+                    1002.0,
+                    1670.0,
+                    1002.0,
+                    1717.0,
+                    924.0,
+                    1717.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1057.0,
+                    1670.0,
+                    1405.0,
+                    1670.0,
+                    1405.0,
+                    1717.0,
+                    1057.0,
+                    1717.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1709.0,
+                    325.0,
+                    1709.0,
+                    325.0,
+                    1748.0,
+                    293.0,
+                    1748.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    363.0,
+                    1709.0,
+                    1338.0,
+                    1709.0,
+                    1338.0,
+                    1748.0,
+                    363.0,
+                    1748.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    948.0,
+                    456.0,
+                    948.0,
+                    456.0,
+                    982.0,
+                    296.0,
+                    982.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    562.0,
+                    948.0,
+                    1382.0,
+                    948.0,
+                    1382.0,
+                    982.0,
+                    562.0,
+                    982.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1403.0,
+                    948.0,
+                    1406.0,
+                    948.0,
+                    1406.0,
+                    982.0,
+                    1403.0,
+                    982.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    978.0,
+                    1031.0,
+                    978.0,
+                    1031.0,
+                    1016.0,
+                    294.0,
+                    1016.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1093.0,
+                    978.0,
+                    1408.0,
+                    978.0,
+                    1408.0,
+                    1016.0,
+                    1093.0,
+                    1016.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1002.0,
+                    1406.0,
+                    1002.0,
+                    1406.0,
+                    1049.0,
+                    291.0,
+                    1049.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1040.0,
+                    1403.0,
+                    1040.0,
+                    1403.0,
+                    1073.0,
+                    295.0,
+                    1073.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1068.0,
+                    326.0,
+                    1068.0,
+                    326.0,
+                    1108.0,
+                    294.0,
+                    1108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    461.0,
+                    1068.0,
+                    997.0,
+                    1068.0,
+                    997.0,
+                    1108.0,
+                    461.0,
+                    1108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    622.0,
+                    797.0,
+                    622.0,
+                    797.0,
+                    658.0,
+                    295.0,
+                    658.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    884.0,
+                    622.0,
+                    926.0,
+                    622.0,
+                    926.0,
+                    658.0,
+                    884.0,
+                    658.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1096.0,
+                    622.0,
+                    1407.0,
+                    622.0,
+                    1407.0,
+                    658.0,
+                    1096.0,
+                    658.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    652.0,
+                    1408.0,
+                    652.0,
+                    1408.0,
+                    689.0,
+                    294.0,
+                    689.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    684.0,
+                    1408.0,
+                    684.0,
+                    1408.0,
+                    720.0,
+                    294.0,
+                    720.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    712.0,
+                    635.0,
+                    712.0,
+                    635.0,
+                    751.0,
+                    291.0,
+                    751.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    721.0,
+                    712.0,
+                    1406.0,
+                    712.0,
+                    1406.0,
+                    751.0,
+                    721.0,
+                    751.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    743.0,
+                    326.0,
+                    743.0,
+                    326.0,
+                    783.0,
+                    294.0,
+                    783.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    366.0,
+                    743.0,
+                    536.0,
+                    743.0,
+                    536.0,
+                    783.0,
+                    366.0,
+                    783.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    695.0,
+                    743.0,
+                    906.0,
+                    743.0,
+                    906.0,
+                    783.0,
+                    695.0,
+                    783.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1066.0,
+                    743.0,
+                    1277.0,
+                    743.0,
+                    1277.0,
+                    783.0,
+                    1066.0,
+                    783.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1348.0,
+                    743.0,
+                    1406.0,
+                    743.0,
+                    1406.0,
+                    783.0,
+                    1348.0,
+                    783.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    775.0,
+                    1391.0,
+                    775.0,
+                    1391.0,
+                    811.0,
+                    295.0,
+                    811.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    276.0,
+                    1060.0,
+                    276.0,
+                    1060.0,
+                    315.0,
+                    294.0,
+                    315.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1193.0,
+                    276.0,
+                    1404.0,
+                    276.0,
+                    1404.0,
+                    315.0,
+                    1193.0,
+                    315.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    312.0,
+                    421.0,
+                    312.0,
+                    421.0,
+                    349.0,
+                    292.0,
+                    349.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    485.0,
+                    312.0,
+                    1370.0,
+                    312.0,
+                    1370.0,
+                    349.0,
+                    485.0,
+                    349.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    344.0,
+                    1401.0,
+                    344.0,
+                    1401.0,
+                    376.0,
+                    296.0,
+                    376.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    371.0,
+                    1163.0,
+                    371.0,
+                    1163.0,
+                    409.0,
+                    294.0,
+                    409.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1206.0,
+                    371.0,
+                    1404.0,
+                    371.0,
+                    1404.0,
+                    409.0,
+                    1206.0,
+                    409.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    399.0,
+                    1408.0,
+                    399.0,
+                    1408.0,
+                    443.0,
+                    292.0,
+                    443.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    430.0,
+                    1338.0,
+                    430.0,
+                    1338.0,
+                    472.0,
+                    291.0,
+                    472.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1873.0,
+                    505.0,
+                    1873.0,
+                    505.0,
+                    1917.0,
+                    292.0,
+                    1917.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    608.0,
+                    1873.0,
+                    1188.0,
+                    1873.0,
+                    1188.0,
+                    1917.0,
+                    608.0,
+                    1917.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1304.0,
+                    1873.0,
+                    1406.0,
+                    1873.0,
+                    1406.0,
+                    1917.0,
+                    1304.0,
+                    1917.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1907.0,
+                    982.0,
+                    1907.0,
+                    982.0,
+                    1947.0,
+                    293.0,
+                    1947.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1034.0,
+                    1907.0,
+                    1214.0,
+                    1907.0,
+                    1214.0,
+                    1947.0,
+                    1034.0,
+                    1947.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1318.0,
+                    1907.0,
+                    1409.0,
+                    1907.0,
+                    1409.0,
+                    1947.0,
+                    1318.0,
+                    1947.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1941.0,
+                    580.0,
+                    1941.0,
+                    580.0,
+                    1982.0,
+                    293.0,
+                    1982.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    707.0,
+                    1941.0,
+                    1004.0,
+                    1941.0,
+                    1004.0,
+                    1982.0,
+                    707.0,
+                    1982.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1053.0,
+                    1941.0,
+                    1406.0,
+                    1941.0,
+                    1406.0,
+                    1982.0,
+                    1053.0,
+                    1982.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1976.0,
+                    734.0,
+                    1976.0,
+                    734.0,
+                    2009.0,
+                    296.0,
+                    2009.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    834.0,
+                    1976.0,
+                    1404.0,
+                    1976.0,
+                    1404.0,
+                    2009.0,
+                    834.0,
+                    2009.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    811.0,
+                    569.0,
+                    811.0,
+                    569.0,
+                    852.0,
+                    295.0,
+                    852.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    587.0,
+                    811.0,
+                    618.0,
+                    811.0,
+                    618.0,
+                    852.0,
+                    587.0,
+                    852.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    681.0,
+                    811.0,
+                    955.0,
+                    811.0,
+                    955.0,
+                    852.0,
+                    681.0,
+                    852.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    973.0,
+                    811.0,
+                    1357.0,
+                    811.0,
+                    1357.0,
+                    852.0,
+                    973.0,
+                    852.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    844.0,
+                    295.0,
+                    844.0,
+                    295.0,
+                    893.0,
+                    291.0,
+                    893.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    380.0,
+                    844.0,
+                    513.0,
+                    844.0,
+                    513.0,
+                    893.0,
+                    380.0,
+                    893.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    586.0,
+                    844.0,
+                    636.0,
+                    844.0,
+                    636.0,
+                    893.0,
+                    586.0,
+                    893.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1102.0,
+                    844.0,
+                    1407.0,
+                    844.0,
+                    1407.0,
+                    893.0,
+                    1102.0,
+                    893.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    635.0,
+                    888.0,
+                    873.0,
+                    888.0,
+                    873.0,
+                    926.0,
+                    635.0,
+                    926.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    469.0,
+                    500.0,
+                    469.0,
+                    500.0,
+                    510.0,
+                    292.0,
+                    510.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    572.0,
+                    469.0,
+                    1092.0,
+                    469.0,
+                    1092.0,
+                    510.0,
+                    572.0,
+                    510.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1109.0,
+                    469.0,
+                    1405.0,
+                    469.0,
+                    1405.0,
+                    510.0,
+                    1109.0,
+                    510.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    500.0,
+                    470.0,
+                    500.0,
+                    470.0,
+                    544.0,
+                    291.0,
+                    544.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    523.0,
+                    500.0,
+                    535.0,
+                    500.0,
+                    535.0,
+                    544.0,
+                    523.0,
+                    544.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    603.0,
+                    500.0,
+                    616.0,
+                    500.0,
+                    616.0,
+                    544.0,
+                    603.0,
+                    544.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    684.0,
+                    500.0,
+                    747.0,
+                    500.0,
+                    747.0,
+                    544.0,
+                    684.0,
+                    544.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    818.0,
+                    500.0,
+                    1107.0,
+                    500.0,
+                    1107.0,
+                    544.0,
+                    818.0,
+                    544.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1131.0,
+                    500.0,
+                    1407.0,
+                    500.0,
+                    1407.0,
+                    544.0,
+                    1131.0,
+                    544.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    493.0,
+                    535.0,
+                    678.0,
+                    535.0,
+                    678.0,
+                    576.0,
+                    493.0,
+                    576.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1046.0,
+                    535.0,
+                    1131.0,
+                    535.0,
+                    1131.0,
+                    576.0,
+                    1046.0,
+                    576.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    562.0,
+                    296.0,
+                    562.0,
+                    296.0,
+                    606.0,
+                    293.0,
+                    606.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    448.0,
+                    562.0,
+                    460.0,
+                    562.0,
+                    460.0,
+                    606.0,
+                    448.0,
+                    606.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    200.0,
+                    1406.0,
+                    200.0,
+                    1406.0,
+                    240.0,
+                    293.0,
+                    240.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    234.0,
+                    752.0,
+                    234.0,
+                    752.0,
+                    268.0,
+                    295.0,
+                    268.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 5,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    438,
+                    1405,
+                    438,
+                    1405,
+                    715,
+                    297,
+                    715
+                ],
+                "score": 0.984
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    726,
+                    1405,
+                    726,
+                    1405,
+                    913,
+                    297,
+                    913
+                ],
+                "score": 0.978
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    202,
+                    1405,
+                    202,
+                    1405,
+                    357,
+                    298,
+                    357
+                ],
+                "score": 0.978
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1830,
+                    1404,
+                    1830,
+                    1404,
+                    2009,
+                    298,
+                    2009
+                ],
+                "score": 0.974
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    925,
+                    1404,
+                    925,
+                    1404,
+                    1018,
+                    298,
+                    1018
+                ],
+                "score": 0.974
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1119,
+                    1404,
+                    1119,
+                    1404,
+                    1275,
+                    298,
+                    1275
+                ],
+                "score": 0.973
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    1596,
+                    1406,
+                    1596,
+                    1406,
+                    1691,
+                    299,
+                    1691
+                ],
+                "score": 0.972
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1698,
+                    1404,
+                    1698,
+                    1404,
+                    1806,
+                    298,
+                    1806
+                ],
+                "score": 0.963
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    360,
+                    1293,
+                    1406,
+                    1293,
+                    1406,
+                    1491,
+                    360,
+                    1491
+                ],
+                "score": 0.962
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    622,
+                    1031,
+                    1020,
+                    1031,
+                    1020,
+                    1107,
+                    622,
+                    1107
+                ],
+                "score": 0.955
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    299,
+                    1527,
+                    633,
+                    1527,
+                    633,
+                    1566,
+                    299,
+                    1566
+                ],
+                "score": 0.921
+            },
+            {
+                "category_id": 9,
+                "poly": [
+                    1366,
+                    1043,
+                    1400,
+                    1043,
+                    1400,
+                    1074,
+                    1366,
+                    1074
+                ],
+                "score": 0.89
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    297,
+                    387,
+                    842,
+                    387,
+                    842,
+                    420,
+                    297,
+                    420
+                ],
+                "score": 0.765
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    841,
+                    2060,
+                    858,
+                    2060,
+                    858,
+                    2084,
+                    841,
+                    2084
+                ],
+                "score": 0.74
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    296,
+                    387,
+                    842,
+                    387,
+                    842,
+                    420,
+                    296,
+                    420
+                ],
+                "score": 0.126
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    621,
+                    1030,
+                    1022,
+                    1030,
+                    1022,
+                    1108,
+                    621,
+                    1108
+                ],
+                "score": 0.93,
+                "latex": "\\sum _ { \\tau _ { i } > \\rho } ( \\tau _ { i } - \\rho ) \\geq 0 . 3 1 \\sum _ { \\tau _ { i } \\in S _ { t } } ( \\tau _ { i } - 1 ) ."
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    736,
+                    1330,
+                    907,
+                    1330,
+                    907,
+                    1378,
+                    736,
+                    1378
+                ],
+                "score": 0.93,
+                "latex": "\\{ \\tilde { h } _ { j } \\} _ { j = 1 } ^ { 2 + \\log ( B ^ { 2 } d ) }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    557,
+                    1658,
+                    747,
+                    1658,
+                    747,
+                    1692,
+                    557,
+                    1692
+                ],
+                "score": 0.93,
+                "latex": "T = O ( ( \\log d ) ^ { 2 } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1138,
+                    1969,
+                    1401,
+                    1969,
+                    1401,
+                    2010,
+                    1138,
+                    2010
+                ],
+                "score": 0.93,
+                "latex": "\\widetilde \\Omega ( d ^ { 3 / 2 } / ( \\varepsilon \\alpha \\log ( 1 / \\alpha ) ) )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    911,
+                    1150,
+                    1046,
+                    1150,
+                    1046,
+                    1184,
+                    911,
+                    1184
+                ],
+                "score": 0.92,
+                "latex": "O ( \\log B ^ { 2 } d )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1168,
+                    203,
+                    1269,
+                    203,
+                    1269,
+                    237,
+                    1168,
+                    237
+                ],
+                "score": 0.92,
+                "latex": "O ( \\log d )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    875,
+                    1867,
+                    1082,
+                    1867,
+                    1082,
+                    1906,
+                    875,
+                    1906
+                ],
+                "score": 0.92,
+                "latex": "\\widetilde \\Omega ( d / \\alpha ^ { 2 } \\log ( 1 / \\alpha ) )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    539,
+                    758,
+                    591,
+                    758,
+                    591,
+                    792,
+                    539,
+                    792
+                ],
+                "score": 0.92,
+                "latex": "\\{ \\tau _ { i } \\}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1247,
+                    561,
+                    1325,
+                    561,
+                    1325,
+                    592,
+                    1247,
+                    592
+                ],
+                "score": 0.92,
+                "latex": "B ^ { 2 } d Z _ { t }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    517,
+                    1868,
+                    708,
+                    1868,
+                    708,
+                    1907,
+                    517,
+                    1907
+                ],
+                "score": 0.92,
+                "latex": "O ( \\alpha \\sqrt { \\log ( 1 / \\alpha ) } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    795,
+                    1730,
+                    1243,
+                    1730,
+                    1243,
+                    1771,
+                    795,
+                    1771
+                ],
+                "score": 0.92,
+                "latex": "n = \\widetilde \\Omega ( ( d / \\alpha ^ { 2 } ) + ( d ^ { 3 / 2 } / ( \\varepsilon \\alpha ) ) \\log ( 1 / \\delta ) )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    662,
+                    1906,
+                    853,
+                    1906,
+                    853,
+                    1944,
+                    662,
+                    1944
+                ],
+                "score": 0.91,
+                "latex": "O ( \\alpha \\sqrt { \\log ( 1 / \\alpha ) } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    637,
+                    1975,
+                    698,
+                    1975,
+                    698,
+                    2009,
+                    637,
+                    2009
+                ],
+                "score": 0.91,
+                "latex": "\\Omega ( \\alpha )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    510,
+                    592,
+                    687,
+                    592,
+                    687,
+                    624,
+                    510,
+                    624
+                ],
+                "score": 0.91,
+                "latex": "Z _ { t } \\sim \\mathrm { U n i f } [ 0 , 1 ]"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    570,
+                    1699,
+                    630,
+                    1699,
+                    630,
+                    1733,
+                    570,
+                    1733
+                ],
+                "score": 0.91,
+                "latex": "( \\varepsilon , \\delta )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    401,
+                    1769,
+                    736,
+                    1769,
+                    736,
+                    1806,
+                    401,
+                    1806
+                ],
+                "score": 0.91,
+                "latex": "\\| \\hat { \\mu } - \\mu \\| _ { 2 } = O ( \\alpha \\sqrt { \\log ( 1 / \\alpha ) } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    491,
+                    854,
+                    673,
+                    854,
+                    673,
+                    884,
+                    491,
+                    884
+                ],
+                "score": 0.91,
+                "latex": "\\rho = \\operatorname* { m a x } _ { j \\in S _ { t } } \\tau _ { j }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    481,
+                    1454,
+                    560,
+                    1454,
+                    560,
+                    1490,
+                    481,
+                    1490
+                ],
+                "score": 0.91,
+                "latex": "\\rho = 2 ^ { \\ell }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1308,
+                    1835,
+                    1356,
+                    1835,
+                    1356,
+                    1869,
+                    1308,
+                    1869
+                ],
+                "score": 0.9,
+                "latex": "1 / \\alpha"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    844,
+                    1829,
+                    895,
+                    1829,
+                    895,
+                    1868,
+                    844,
+                    1868
+                ],
+                "score": 0.9,
+                "latex": "\\widetilde { \\Omega } ( \\cdot )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    883,
+                    1292,
+                    1360,
+                    1292,
+                    1360,
+                    1335,
+                    883,
+                    1335
+                ],
+                "score": 0.89,
+                "latex": "i \\in S _ { t } ^ { ( s ) } : \\tau _ { i } \\gets ( x _ { i } - \\mu _ { t } ) ^ { \\top } U _ { t } ^ { ( s ) } ( x _ { i } - \\mu _ { t } ) ;"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    401,
+                    1736,
+                    532,
+                    1736,
+                    532,
+                    1769,
+                    401,
+                    1769
+                ],
+                "score": 0.89,
+                "latex": "c \\in ( 0 , 0 . 1 )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1348,
+                    590,
+                    1402,
+                    590,
+                    1402,
+                    620,
+                    1348,
+                    620
+                ],
+                "score": 0.89,
+                "latex": "B ^ { 2 } d"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    891,
+                    759,
+                    930,
+                    759,
+                    930,
+                    790,
+                    891,
+                    790
+                ],
+                "score": 0.88,
+                "latex": "\\rho Z"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    396,
+                    1378,
+                    573,
+                    1378,
+                    573,
+                    1414,
+                    396,
+                    1414
+                ],
+                "score": 0.86,
+                "latex": "I _ { 1 } = [ 1 / 4 , 1 / 2 )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    807,
+                    1414,
+                    1301,
+                    1414,
+                    1301,
+                    1459,
+                    807,
+                    1459
+                ],
+                "score": 0.85,
+                "latex": "\\begin{array} { r } { \\sum _ { j \\ge \\ell } ( 2 ^ { j } - 2 ^ { \\ell } ) \\tilde { h } _ { j } \\ge 0 . 3 1 \\sum _ { i \\in S _ { t } ^ { ( s ) } } ( \\tau _ { i } - 1 ) ; } \\end{array}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    670,
+                    1738,
+                    741,
+                    1738,
+                    741,
+                    1767,
+                    670,
+                    1767
+                ],
+                "score": 0.85,
+                "latex": "\\alpha \\leq c"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    582,
+                    1374,
+                    1227,
+                    1374,
+                    1227,
+                    1414,
+                    582,
+                    1414
+                ],
+                "score": 0.84,
+                "latex": "I _ { 2 } = [ 1 / 2 , 1 ) , \\ldots , \\bar { I } _ { 2 + \\log ( B ^ { 2 } d ) } = [ 2 ^ { \\log ( B ^ { 2 } d ) - 1 } , 2 ^ { \\log ( B ^ { 2 } d ) } ]"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    967,
+                    734,
+                    986,
+                    734,
+                    986,
+                    760,
+                    967,
+                    760
+                ],
+                "score": 0.83,
+                "latex": "\\rho"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    581,
+                    1628,
+                    654,
+                    1628,
+                    654,
+                    1658,
+                    581,
+                    1658
+                ],
+                "score": 0.82,
+                "latex": "\\ S 2 . 3 . 1"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    676,
+                    991,
+                    695,
+                    991,
+                    695,
+                    1018,
+                    676,
+                    1018
+                ],
+                "score": 0.82,
+                "latex": "\\rho"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    742,
+                    1217,
+                    760,
+                    1217,
+                    760,
+                    1244,
+                    742,
+                    1244
+                ],
+                "score": 0.82,
+                "latex": "\\rho"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    673,
+                    1418,
+                    689,
+                    1418,
+                    689,
+                    1444,
+                    673,
+                    1444
+                ],
+                "score": 0.77,
+                "latex": "\\ell"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1100,
+                    1185,
+                    1113,
+                    1185,
+                    1113,
+                    1210,
+                    1100,
+                    1210
+                ],
+                "score": 0.76,
+                "latex": "t"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    928,
+                    1188,
+                    945,
+                    1188,
+                    945,
+                    1209,
+                    928,
+                    1209
+                ],
+                "score": 0.75,
+                "latex": "s"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    999,
+                    760,
+                    1023,
+                    760,
+                    1023,
+                    787,
+                    999,
+                    787
+                ],
+                "score": 0.73,
+                "latex": "Z"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1197,
+                    1836,
+                    1251,
+                    1836,
+                    1251,
+                    1865,
+                    1197,
+                    1865
+                ],
+                "score": 0.56,
+                "latex": "d , R"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1112,
+                    1702,
+                    1129,
+                    1702,
+                    1129,
+                    1727,
+                    1112,
+                    1727
+                ],
+                "score": 0.4,
+                "latex": "I"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1229,
+                    759,
+                    1292,
+                    759,
+                    1292,
+                    792,
+                    1229,
+                    792
+                ],
+                "score": 0.25,
+                "latex": "[ 0 , 1 ] )"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1526.0,
+                    636.0,
+                    1526.0,
+                    636.0,
+                    1574.0,
+                    291.0,
+                    1574.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    385.0,
+                    845.0,
+                    385.0,
+                    845.0,
+                    427.0,
+                    293.0,
+                    427.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    839.0,
+                    2059.0,
+                    860.0,
+                    2059.0,
+                    860.0,
+                    2090.0,
+                    839.0,
+                    2090.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    385.0,
+                    845.0,
+                    385.0,
+                    845.0,
+                    427.0,
+                    293.0,
+                    427.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    438.0,
+                    1406.0,
+                    438.0,
+                    1406.0,
+                    476.0,
+                    294.0,
+                    476.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    472.0,
+                    1406.0,
+                    472.0,
+                    1406.0,
+                    505.0,
+                    296.0,
+                    505.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    499.0,
+                    1408.0,
+                    499.0,
+                    1408.0,
+                    536.0,
+                    292.0,
+                    536.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    531.0,
+                    1407.0,
+                    531.0,
+                    1407.0,
+                    567.0,
+                    295.0,
+                    567.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    558.0,
+                    1246.0,
+                    558.0,
+                    1246.0,
+                    599.0,
+                    292.0,
+                    599.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1326.0,
+                    558.0,
+                    1410.0,
+                    558.0,
+                    1410.0,
+                    599.0,
+                    1326.0,
+                    599.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    588.0,
+                    509.0,
+                    588.0,
+                    509.0,
+                    628.0,
+                    292.0,
+                    628.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    688.0,
+                    588.0,
+                    1347.0,
+                    588.0,
+                    1347.0,
+                    628.0,
+                    688.0,
+                    628.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1403.0,
+                    588.0,
+                    1408.0,
+                    588.0,
+                    1408.0,
+                    628.0,
+                    1403.0,
+                    628.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    621.0,
+                    1405.0,
+                    621.0,
+                    1405.0,
+                    656.0,
+                    292.0,
+                    656.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    654.0,
+                    1406.0,
+                    654.0,
+                    1406.0,
+                    686.0,
+                    296.0,
+                    686.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    683.0,
+                    633.0,
+                    683.0,
+                    633.0,
+                    718.0,
+                    293.0,
+                    718.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    727.0,
+                    966.0,
+                    727.0,
+                    966.0,
+                    763.0,
+                    295.0,
+                    763.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    987.0,
+                    727.0,
+                    1406.0,
+                    727.0,
+                    1406.0,
+                    763.0,
+                    987.0,
+                    763.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    758.0,
+                    538.0,
+                    758.0,
+                    538.0,
+                    793.0,
+                    295.0,
+                    793.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    592.0,
+                    758.0,
+                    890.0,
+                    758.0,
+                    890.0,
+                    793.0,
+                    592.0,
+                    793.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    931.0,
+                    758.0,
+                    998.0,
+                    758.0,
+                    998.0,
+                    793.0,
+                    931.0,
+                    793.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1024.0,
+                    758.0,
+                    1228.0,
+                    758.0,
+                    1228.0,
+                    793.0,
+                    1024.0,
+                    793.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1293.0,
+                    758.0,
+                    1407.0,
+                    758.0,
+                    1407.0,
+                    793.0,
+                    1293.0,
+                    793.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    790.0,
+                    1405.0,
+                    790.0,
+                    1405.0,
+                    822.0,
+                    296.0,
+                    822.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    820.0,
+                    1408.0,
+                    820.0,
+                    1408.0,
+                    855.0,
+                    294.0,
+                    855.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    846.0,
+                    490.0,
+                    846.0,
+                    490.0,
+                    890.0,
+                    292.0,
+                    890.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    674.0,
+                    846.0,
+                    1408.0,
+                    846.0,
+                    1408.0,
+                    890.0,
+                    674.0,
+                    890.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    880.0,
+                    1385.0,
+                    880.0,
+                    1385.0,
+                    916.0,
+                    295.0,
+                    916.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    204.0,
+                    1167.0,
+                    204.0,
+                    1167.0,
+                    237.0,
+                    296.0,
+                    237.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1270.0,
+                    204.0,
+                    1405.0,
+                    204.0,
+                    1405.0,
+                    237.0,
+                    1270.0,
+                    237.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    232.0,
+                    1405.0,
+                    232.0,
+                    1405.0,
+                    268.0,
+                    295.0,
+                    268.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    262.0,
+                    1405.0,
+                    262.0,
+                    1405.0,
+                    297.0,
+                    293.0,
+                    297.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    294.0,
+                    1406.0,
+                    294.0,
+                    1406.0,
+                    330.0,
+                    295.0,
+                    330.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    323.0,
+                    736.0,
+                    323.0,
+                    736.0,
+                    361.0,
+                    295.0,
+                    361.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1832.0,
+                    843.0,
+                    1832.0,
+                    843.0,
+                    1871.0,
+                    293.0,
+                    1871.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    896.0,
+                    1832.0,
+                    1196.0,
+                    1832.0,
+                    1196.0,
+                    1871.0,
+                    896.0,
+                    1871.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1252.0,
+                    1832.0,
+                    1307.0,
+                    1832.0,
+                    1307.0,
+                    1871.0,
+                    1252.0,
+                    1871.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1357.0,
+                    1832.0,
+                    1405.0,
+                    1832.0,
+                    1405.0,
+                    1871.0,
+                    1357.0,
+                    1871.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1868.0,
+                    516.0,
+                    1868.0,
+                    516.0,
+                    1911.0,
+                    293.0,
+                    1911.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    709.0,
+                    1868.0,
+                    874.0,
+                    1868.0,
+                    874.0,
+                    1911.0,
+                    709.0,
+                    1911.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1083.0,
+                    1868.0,
+                    1409.0,
+                    1868.0,
+                    1409.0,
+                    1911.0,
+                    1083.0,
+                    1911.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1907.0,
+                    661.0,
+                    1907.0,
+                    661.0,
+                    1946.0,
+                    293.0,
+                    1946.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    854.0,
+                    1907.0,
+                    1408.0,
+                    1907.0,
+                    1408.0,
+                    1946.0,
+                    854.0,
+                    1946.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1941.0,
+                    1406.0,
+                    1941.0,
+                    1406.0,
+                    1973.0,
+                    293.0,
+                    1973.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1969.0,
+                    636.0,
+                    1969.0,
+                    636.0,
+                    2013.0,
+                    292.0,
+                    2013.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    699.0,
+                    1969.0,
+                    1137.0,
+                    1969.0,
+                    1137.0,
+                    2013.0,
+                    699.0,
+                    2013.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1402.0,
+                    1969.0,
+                    1405.0,
+                    1969.0,
+                    1405.0,
+                    2013.0,
+                    1402.0,
+                    2013.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    924.0,
+                    1404.0,
+                    924.0,
+                    1404.0,
+                    958.0,
+                    293.0,
+                    958.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    952.0,
+                    1406.0,
+                    952.0,
+                    1406.0,
+                    992.0,
+                    292.0,
+                    992.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    986.0,
+                    675.0,
+                    986.0,
+                    675.0,
+                    1020.0,
+                    292.0,
+                    1020.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    696.0,
+                    986.0,
+                    1150.0,
+                    986.0,
+                    1150.0,
+                    1020.0,
+                    696.0,
+                    1020.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1120.0,
+                    1405.0,
+                    1120.0,
+                    1405.0,
+                    1157.0,
+                    296.0,
+                    1157.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1149.0,
+                    910.0,
+                    1149.0,
+                    910.0,
+                    1189.0,
+                    292.0,
+                    1189.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1047.0,
+                    1149.0,
+                    1406.0,
+                    1149.0,
+                    1406.0,
+                    1189.0,
+                    1047.0,
+                    1189.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1181.0,
+                    927.0,
+                    1181.0,
+                    927.0,
+                    1216.0,
+                    292.0,
+                    1216.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    946.0,
+                    1181.0,
+                    1099.0,
+                    1181.0,
+                    1099.0,
+                    1216.0,
+                    946.0,
+                    1216.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1114.0,
+                    1181.0,
+                    1405.0,
+                    1181.0,
+                    1405.0,
+                    1216.0,
+                    1114.0,
+                    1216.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1210.0,
+                    741.0,
+                    1210.0,
+                    741.0,
+                    1248.0,
+                    294.0,
+                    1248.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    761.0,
+                    1210.0,
+                    1405.0,
+                    1210.0,
+                    1405.0,
+                    1248.0,
+                    761.0,
+                    1248.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1241.0,
+                    1117.0,
+                    1241.0,
+                    1117.0,
+                    1278.0,
+                    294.0,
+                    1278.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1596.0,
+                    1408.0,
+                    1596.0,
+                    1408.0,
+                    1630.0,
+                    294.0,
+                    1630.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    1629.0,
+                    580.0,
+                    1629.0,
+                    580.0,
+                    1659.0,
+                    297.0,
+                    1659.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    655.0,
+                    1629.0,
+                    1403.0,
+                    1629.0,
+                    1403.0,
+                    1659.0,
+                    655.0,
+                    1659.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1658.0,
+                    556.0,
+                    1658.0,
+                    556.0,
+                    1692.0,
+                    296.0,
+                    1692.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    748.0,
+                    1658.0,
+                    1103.0,
+                    1658.0,
+                    1103.0,
+                    1692.0,
+                    748.0,
+                    1692.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1696.0,
+                    569.0,
+                    1696.0,
+                    569.0,
+                    1736.0,
+                    294.0,
+                    1736.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    631.0,
+                    1696.0,
+                    1111.0,
+                    1696.0,
+                    1111.0,
+                    1736.0,
+                    631.0,
+                    1736.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1130.0,
+                    1696.0,
+                    1406.0,
+                    1696.0,
+                    1406.0,
+                    1736.0,
+                    1130.0,
+                    1736.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1732.0,
+                    400.0,
+                    1732.0,
+                    400.0,
+                    1773.0,
+                    294.0,
+                    1773.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    533.0,
+                    1732.0,
+                    669.0,
+                    1732.0,
+                    669.0,
+                    1773.0,
+                    533.0,
+                    1773.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    742.0,
+                    1732.0,
+                    794.0,
+                    1732.0,
+                    794.0,
+                    1773.0,
+                    742.0,
+                    1773.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1244.0,
+                    1732.0,
+                    1408.0,
+                    1732.0,
+                    1408.0,
+                    1773.0,
+                    1244.0,
+                    1773.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1769.0,
+                    400.0,
+                    1769.0,
+                    400.0,
+                    1809.0,
+                    293.0,
+                    1809.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    737.0,
+                    1769.0,
+                    974.0,
+                    1769.0,
+                    974.0,
+                    1809.0,
+                    737.0,
+                    1809.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    356.0,
+                    1283.0,
+                    882.0,
+                    1283.0,
+                    882.0,
+                    1342.0,
+                    356.0,
+                    1342.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1361.0,
+                    1283.0,
+                    1374.0,
+                    1283.0,
+                    1374.0,
+                    1342.0,
+                    1361.0,
+                    1342.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    342.0,
+                    1311.0,
+                    735.0,
+                    1311.0,
+                    735.0,
+                    1398.0,
+                    342.0,
+                    1398.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    908.0,
+                    1311.0,
+                    1419.0,
+                    1311.0,
+                    1419.0,
+                    1398.0,
+                    908.0,
+                    1398.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    387.0,
+                    1372.0,
+                    395.0,
+                    1372.0,
+                    395.0,
+                    1420.0,
+                    387.0,
+                    1420.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    574.0,
+                    1372.0,
+                    581.0,
+                    1372.0,
+                    581.0,
+                    1420.0,
+                    574.0,
+                    1420.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1228.0,
+                    1372.0,
+                    1242.0,
+                    1372.0,
+                    1242.0,
+                    1420.0,
+                    1228.0,
+                    1420.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    352.0,
+                    1402.0,
+                    672.0,
+                    1402.0,
+                    672.0,
+                    1464.0,
+                    352.0,
+                    1464.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    690.0,
+                    1402.0,
+                    806.0,
+                    1402.0,
+                    806.0,
+                    1464.0,
+                    690.0,
+                    1464.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1302.0,
+                    1402.0,
+                    1313.0,
+                    1402.0,
+                    1313.0,
+                    1464.0,
+                    1302.0,
+                    1464.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    355.0,
+                    1454.0,
+                    480.0,
+                    1454.0,
+                    480.0,
+                    1491.0,
+                    355.0,
+                    1491.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    561.0,
+                    1454.0,
+                    585.0,
+                    1454.0,
+                    585.0,
+                    1491.0,
+                    561.0,
+                    1491.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 6,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1560,
+                    1404,
+                    1560,
+                    1404,
+                    1807,
+                    297,
+                    1807
+                ],
+                "score": 0.983
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    949,
+                    1405,
+                    949,
+                    1405,
+                    1433,
+                    297,
+                    1433
+                ],
+                "score": 0.981
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    418,
+                    1406,
+                    418,
+                    1406,
+                    550,
+                    298,
+                    550
+                ],
+                "score": 0.976
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    279,
+                    1404,
+                    279,
+                    1404,
+                    404,
+                    298,
+                    404
+                ],
+                "score": 0.973
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    1913,
+                    1405,
+                    1913,
+                    1405,
+                    2007,
+                    299,
+                    2007
+                ],
+                "score": 0.971
+            },
+            {
+                "category_id": 3,
+                "poly": [
+                    324,
+                    587,
+                    1407,
+                    587,
+                    1407,
+                    800,
+                    324,
+                    800
+                ],
+                "score": 0.964
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    293,
+                    202,
+                    1401,
+                    202,
+                    1401,
+                    266,
+                    293,
+                    266
+                ],
+                "score": 0.952
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    293,
+                    1819,
+                    1404,
+                    1819,
+                    1404,
+                    1887,
+                    293,
+                    1887
+                ],
+                "score": 0.942
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    294,
+                    1485,
+                    1276,
+                    1485,
+                    1276,
+                    1525,
+                    294,
+                    1525
+                ],
+                "score": 0.928
+            },
+            {
+                "category_id": 4,
+                "poly": [
+                    300,
+                    830,
+                    1400,
+                    830,
+                    1400,
+                    924,
+                    300,
+                    924
+                ],
+                "score": 0.822
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    840,
+                    2062,
+                    858,
+                    2062,
+                    858,
+                    2085,
+                    840,
+                    2085
+                ],
+                "score": 0.803
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    954,
+                    339,
+                    1049,
+                    339,
+                    1049,
+                    376,
+                    954,
+                    376
+                ],
+                "score": 0.93,
+                "latex": "\\bar { O } ( d ^ { 3 / 2 } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    756,
+                    1086,
+                    921,
+                    1086,
+                    921,
+                    1124,
+                    756,
+                    1124
+                ],
+                "score": 0.93,
+                "latex": "\\alpha \\sqrt { d } + \\sqrt { d / n }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    938,
+                    1341,
+                    1057,
+                    1341,
+                    1057,
+                    1374,
+                    938,
+                    1374
+                ],
+                "score": 0.93,
+                "latex": "( \\varepsilon , \\delta , n , \\alpha )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    488,
+                    1015,
+                    638,
+                    1015,
+                    638,
+                    1054,
+                    488,
+                    1054
+                ],
+                "score": 0.93,
+                "latex": "\\alpha \\sqrt { \\log ( 1 / \\alpha ) }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    344,
+                    515,
+                    576,
+                    515,
+                    576,
+                    551,
+                    344,
+                    551
+                ],
+                "score": 0.93,
+                "latex": "\\lambda ^ { ( s ) } \\le C \\alpha \\log ( 1 / \\alpha )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1198,
+                    481,
+                    1404,
+                    481,
+                    1404,
+                    517,
+                    1198,
+                    517
+                ],
+                "score": 0.92,
+                "latex": "n ^ { ( s ) } > n ( 1 { - } 1 0 \\alpha )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1198,
+                    418,
+                    1300,
+                    418,
+                    1300,
+                    451,
+                    1198,
+                    451
+                ],
+                "score": 0.92,
+                "latex": "\\log ( 1 / \\zeta )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1171,
+                    1150,
+                    1230,
+                    1150,
+                    1230,
+                    1185,
+                    1171,
+                    1185
+                ],
+                "score": 0.92,
+                "latex": "\\alpha \\sqrt { d }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    777,
+                    1915,
+                    841,
+                    1915,
+                    841,
+                    1947,
+                    777,
+                    1947
+                ],
+                "score": 0.92,
+                "latex": "R ( S )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1278,
+                    1852,
+                    1398,
+                    1852,
+                    1398,
+                    1884,
+                    1278,
+                    1884
+                ],
+                "score": 0.92,
+                "latex": "( 1 - \\alpha ) | S |"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1071,
+                    1820,
+                    1164,
+                    1820,
+                    1164,
+                    1853,
+                    1071,
+                    1853
+                ],
+                "score": 0.92,
+                "latex": "\\{ x _ { i } \\} _ { i \\in S }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1255,
+                    1084,
+                    1329,
+                    1084,
+                    1329,
+                    1124,
+                    1255,
+                    1124
+                ],
+                "score": 0.92,
+                "latex": "\\sqrt { d / n }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    817,
+                    1052,
+                    1211,
+                    1052,
+                    1211,
+                    1087,
+                    817,
+                    1087
+                ],
+                "score": 0.92,
+                "latex": "( \\alpha , \\varepsilon , \\delta , n ) = ( 0 . 0 0 1 , 2 0 , 0 . 0 1 , 1 0 ^ { 6 } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    353,
+                    1152,
+                    722,
+                    1152,
+                    722,
+                    1187,
+                    353,
+                    1187
+                ],
+                "score": 0.91,
+                "latex": "( \\alpha , \\delta , d , n ) = ( 0 . 1 , 0 . 0 1 , 1 0 , 1 0 ^ { 6 } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    666,
+                    1976,
+                    729,
+                    1976,
+                    729,
+                    2009,
+                    666,
+                    2009
+                ],
+                "score": 0.91,
+                "latex": "R ( S )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    927,
+                    419,
+                    994,
+                    419,
+                    994,
+                    450,
+                    927,
+                    450
+                ],
+                "score": 0.91,
+                "latex": "\\zeta > 0"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1327,
+                    1819,
+                    1396,
+                    1819,
+                    1396,
+                    1852,
+                    1327,
+                    1852
+                ],
+                "score": 0.91,
+                "latex": "( \\sigma , \\alpha )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    521,
+                    1188,
+                    701,
+                    1188,
+                    701,
+                    1223,
+                    521,
+                    1223
+                ],
+                "score": 0.91,
+                "latex": "\\varepsilon > c d ^ { 1 . 5 } / ( \\alpha n )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    687,
+                    419,
+                    752,
+                    419,
+                    752,
+                    451,
+                    687,
+                    451
+                ],
+                "score": 0.91,
+                "latex": "1 - \\zeta"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    534,
+                    200,
+                    585,
+                    200,
+                    585,
+                    231,
+                    534,
+                    231
+                ],
+                "score": 0.91,
+                "latex": "d ^ { 1 / 2 }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    567,
+                    483,
+                    668,
+                    483,
+                    668,
+                    517,
+                    567,
+                    517
+                ],
+                "score": 0.9,
+                "latex": "\\log ( 1 / \\zeta )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    299,
+                    1186,
+                    448,
+                    1186,
+                    448,
+                    1223,
+                    299,
+                    1223
+                ],
+                "score": 0.9,
+                "latex": "\\alpha \\sqrt { \\log ( 1 / \\alpha ) }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1293,
+                    982,
+                    1352,
+                    982,
+                    1352,
+                    1015,
+                    1293,
+                    1015
+                ],
+                "score": 0.9,
+                "latex": "\\alpha \\sqrt { d }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1119,
+                    1853,
+                    1197,
+                    1853,
+                    1197,
+                    1880,
+                    1119,
+                    1880
+                ],
+                "score": 0.89,
+                "latex": "T \\subset S"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1354,
+                    448,
+                    1398,
+                    448,
+                    1398,
+                    482,
+                    1354,
+                    482
+                ],
+                "score": 0.88,
+                "latex": "\\zeta / 2"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1238,
+                    948,
+                    1402,
+                    948,
+                    1402,
+                    981,
+                    1238,
+                    981
+                ],
+                "score": 0.88,
+                "latex": "( \\alpha , \\varepsilon , \\delta , n ) \\ ="
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    683,
+                    1947,
+                    750,
+                    1947,
+                    750,
+                    1974,
+                    683,
+                    1974
+                ],
+                "score": 0.87,
+                "latex": "1 - \\alpha"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1260,
+                    1818,
+                    1297,
+                    1818,
+                    1297,
+                    1849,
+                    1260,
+                    1849
+                ],
+                "score": 0.87,
+                "latex": "\\mathbb { R } ^ { d }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    576,
+                    1850,
+                    955,
+                    1850,
+                    955,
+                    1887,
+                    576,
+                    1887
+                ],
+                "score": 0.86,
+                "latex": "\\begin{array} { r } { . i f \\| ( 1 / | T | ) \\sum _ { i \\in T } ( x _ { i } - \\mu ) \\| _ { 2 } \\leq \\sigma } \\end{array}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    297,
+                    982,
+                    522,
+                    982,
+                    522,
+                    1017,
+                    297,
+                    1017
+                ],
+                "score": 0.85,
+                "latex": "( 0 . 0 5 , 2 0 , 0 . 0 1 , 1 0 ^ { 6 } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    297,
+                    1223,
+                    406,
+                    1223,
+                    406,
+                    1249,
+                    297,
+                    1249
+                ],
+                "score": 0.84,
+                "latex": "\\varepsilon = 0 . 0 5"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    438,
+                    1947,
+                    459,
+                    1947,
+                    459,
+                    1973,
+                    438,
+                    1973
+                ],
+                "score": 0.8,
+                "latex": "S"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    627,
+                    1121,
+                    645,
+                    1121,
+                    645,
+                    1147,
+                    627,
+                    1147
+                ],
+                "score": 0.77,
+                "latex": "d"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1380,
+                    1059,
+                    1401,
+                    1059,
+                    1401,
+                    1081,
+                    1380,
+                    1081
+                ],
+                "score": 0.75,
+                "latex": "\\alpha"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1178,
+                    1125,
+                    1199,
+                    1125,
+                    1199,
+                    1147,
+                    1178,
+                    1147
+                ],
+                "score": 0.75,
+                "latex": "\\alpha"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    691,
+                    990,
+                    714,
+                    990,
+                    714,
+                    1012,
+                    691,
+                    1012
+                ],
+                "score": 0.74,
+                "latex": "\\alpha"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    520,
+                    1056,
+                    538,
+                    1056,
+                    538,
+                    1081,
+                    520,
+                    1081
+                ],
+                "score": 0.72,
+                "latex": "d"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1309,
+                    1257,
+                    1326,
+                    1257,
+                    1326,
+                    1278,
+                    1309,
+                    1278
+                ],
+                "score": 0.7,
+                "latex": "\\varepsilon"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1307,
+                    1225,
+                    1324,
+                    1225,
+                    1324,
+                    1247,
+                    1307,
+                    1247
+                ],
+                "score": 0.68,
+                "latex": "\\varepsilon"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    559,
+                    1855,
+                    580,
+                    1855,
+                    580,
+                    1883,
+                    559,
+                    1883
+                ],
+                "score": 0.38,
+                "latex": "\\mu"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    902,
+                    1823,
+                    923,
+                    1823,
+                    923,
+                    1848,
+                    902,
+                    1848
+                ],
+                "score": 0.35,
+                "latex": "A"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1331,
+                    1377,
+                    1401,
+                    1377,
+                    1401,
+                    1404,
+                    1331,
+                    1404
+                ],
+                "score": 0.3,
+                "latex": "q _ { \\mathrm { r a n g e } }"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    710.0,
+                    587.0,
+                    748.0,
+                    587.0,
+                    748.0,
+                    604.0,
+                    710.0,
+                    604.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1112.0,
+                    588.0,
+                    1133.0,
+                    588.0,
+                    1133.0,
+                    603.0,
+                    1112.0,
+                    603.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    313.0,
+                    605.0,
+                    377.0,
+                    605.0,
+                    377.0,
+                    754.0,
+                    313.0,
+                    754.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    398.0,
+                    601.0,
+                    465.0,
+                    601.0,
+                    465.0,
+                    641.0,
+                    398.0,
+                    641.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    691.0,
+                    607.0,
+                    755.0,
+                    607.0,
+                    755.0,
+                    701.0,
+                    691.0,
+                    701.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    763.0,
+                    601.0,
+                    849.0,
+                    601.0,
+                    849.0,
+                    641.0,
+                    763.0,
+                    641.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1081.0,
+                    608.0,
+                    1149.0,
+                    608.0,
+                    1149.0,
+                    755.0,
+                    1081.0,
+                    755.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1331.0,
+                    597.0,
+                    1399.0,
+                    597.0,
+                    1399.0,
+                    641.0,
+                    1331.0,
+                    641.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    341.0,
+                    692.0,
+                    361.0,
+                    692.0,
+                    361.0,
+                    711.0,
+                    341.0,
+                    711.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    695.0,
+                    682.0,
+                    748.0,
+                    682.0,
+                    748.0,
+                    727.0,
+                    695.0,
+                    727.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    692.0,
+                    718.0,
+                    749.0,
+                    718.0,
+                    749.0,
+                    739.0,
+                    692.0,
+                    739.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1111.0,
+                    728.0,
+                    1134.0,
+                    728.0,
+                    1134.0,
+                    746.0,
+                    1111.0,
+                    746.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    342.0,
+                    739.0,
+                    360.0,
+                    739.0,
+                    360.0,
+                    753.0,
+                    342.0,
+                    753.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    709.0,
+                    740.0,
+                    750.0,
+                    740.0,
+                    750.0,
+                    761.0,
+                    709.0,
+                    761.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    339.0,
+                    758.0,
+                    363.0,
+                    758.0,
+                    363.0,
+                    777.0,
+                    339.0,
+                    777.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    365.0,
+                    773.0,
+                    376.0,
+                    773.0,
+                    376.0,
+                    785.0,
+                    365.0,
+                    785.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    411.0,
+                    772.0,
+                    431.0,
+                    772.0,
+                    431.0,
+                    786.0,
+                    411.0,
+                    786.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    453.0,
+                    772.0,
+                    543.0,
+                    772.0,
+                    543.0,
+                    801.0,
+                    453.0,
+                    801.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    560.0,
+                    771.0,
+                    582.0,
+                    771.0,
+                    582.0,
+                    789.0,
+                    560.0,
+                    789.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    709.0,
+                    762.0,
+                    762.0,
+                    762.0,
+                    762.0,
+                    785.0,
+                    709.0,
+                    785.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    795.0,
+                    771.0,
+                    817.0,
+                    771.0,
+                    817.0,
+                    789.0,
+                    795.0,
+                    789.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    837.0,
+                    772.0,
+                    927.0,
+                    772.0,
+                    927.0,
+                    801.0,
+                    837.0,
+                    801.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    946.0,
+                    771.0,
+                    967.0,
+                    771.0,
+                    967.0,
+                    789.0,
+                    946.0,
+                    789.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    994.0,
+                    771.0,
+                    1020.0,
+                    771.0,
+                    1020.0,
+                    788.0,
+                    994.0,
+                    788.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1191.0,
+                    771.0,
+                    1218.0,
+                    771.0,
+                    1218.0,
+                    789.0,
+                    1191.0,
+                    789.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1256.0,
+                    771.0,
+                    1277.0,
+                    771.0,
+                    1277.0,
+                    799.0,
+                    1256.0,
+                    799.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1319.0,
+                    772.0,
+                    1339.0,
+                    772.0,
+                    1339.0,
+                    786.0,
+                    1319.0,
+                    786.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1109.0,
+                    763.0,
+                    1153.0,
+                    763.0,
+                    1153.0,
+                    788.0,
+                    1109.0,
+                    788.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    607.0,
+                    769.0,
+                    636.0,
+                    769.0,
+                    636.0,
+                    788.0,
+                    607.0,
+                    788.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1377.0,
+                    769.0,
+                    1405.0,
+                    769.0,
+                    1405.0,
+                    788.0,
+                    1377.0,
+                    788.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    288.0,
+                    1482.0,
+                    1279.0,
+                    1482.0,
+                    1279.0,
+                    1535.0,
+                    288.0,
+                    1535.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    831.0,
+                    1404.0,
+                    831.0,
+                    1404.0,
+                    865.0,
+                    296.0,
+                    865.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    860.0,
+                    1405.0,
+                    860.0,
+                    1405.0,
+                    898.0,
+                    295.0,
+                    898.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    890.0,
+                    663.0,
+                    890.0,
+                    663.0,
+                    927.0,
+                    294.0,
+                    927.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    838.0,
+                    2058.0,
+                    862.0,
+                    2058.0,
+                    862.0,
+                    2090.0,
+                    838.0,
+                    2090.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1562.0,
+                    1405.0,
+                    1562.0,
+                    1405.0,
+                    1596.0,
+                    295.0,
+                    1596.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1593.0,
+                    1406.0,
+                    1593.0,
+                    1406.0,
+                    1627.0,
+                    295.0,
+                    1627.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1622.0,
+                    1404.0,
+                    1622.0,
+                    1404.0,
+                    1658.0,
+                    294.0,
+                    1658.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1652.0,
+                    1406.0,
+                    1652.0,
+                    1406.0,
+                    1687.0,
+                    294.0,
+                    1687.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1682.0,
+                    1406.0,
+                    1682.0,
+                    1406.0,
+                    1718.0,
+                    294.0,
+                    1718.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1712.0,
+                    1407.0,
+                    1712.0,
+                    1407.0,
+                    1749.0,
+                    291.0,
+                    1749.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1745.0,
+                    1404.0,
+                    1745.0,
+                    1404.0,
+                    1779.0,
+                    296.0,
+                    1779.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1774.0,
+                    1277.0,
+                    1774.0,
+                    1277.0,
+                    1811.0,
+                    294.0,
+                    1811.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    947.0,
+                    1237.0,
+                    947.0,
+                    1237.0,
+                    984.0,
+                    295.0,
+                    984.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    523.0,
+                    982.0,
+                    690.0,
+                    982.0,
+                    690.0,
+                    1019.0,
+                    523.0,
+                    1019.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    715.0,
+                    982.0,
+                    1292.0,
+                    982.0,
+                    1292.0,
+                    1019.0,
+                    715.0,
+                    1019.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1353.0,
+                    982.0,
+                    1405.0,
+                    982.0,
+                    1405.0,
+                    1019.0,
+                    1353.0,
+                    1019.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1015.0,
+                    487.0,
+                    1015.0,
+                    487.0,
+                    1057.0,
+                    292.0,
+                    1057.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    639.0,
+                    1015.0,
+                    1406.0,
+                    1015.0,
+                    1406.0,
+                    1057.0,
+                    639.0,
+                    1057.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1050.0,
+                    519.0,
+                    1050.0,
+                    519.0,
+                    1089.0,
+                    291.0,
+                    1089.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    539.0,
+                    1050.0,
+                    816.0,
+                    1050.0,
+                    816.0,
+                    1089.0,
+                    539.0,
+                    1089.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1212.0,
+                    1050.0,
+                    1379.0,
+                    1050.0,
+                    1379.0,
+                    1089.0,
+                    1212.0,
+                    1089.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1402.0,
+                    1050.0,
+                    1407.0,
+                    1050.0,
+                    1407.0,
+                    1089.0,
+                    1402.0,
+                    1089.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1085.0,
+                    755.0,
+                    1085.0,
+                    755.0,
+                    1126.0,
+                    292.0,
+                    1126.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    922.0,
+                    1085.0,
+                    1254.0,
+                    1085.0,
+                    1254.0,
+                    1126.0,
+                    922.0,
+                    1126.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1330.0,
+                    1085.0,
+                    1406.0,
+                    1085.0,
+                    1406.0,
+                    1126.0,
+                    1330.0,
+                    1126.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1119.0,
+                    626.0,
+                    1119.0,
+                    626.0,
+                    1156.0,
+                    294.0,
+                    1156.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    646.0,
+                    1119.0,
+                    1177.0,
+                    1119.0,
+                    1177.0,
+                    1156.0,
+                    646.0,
+                    1156.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1200.0,
+                    1119.0,
+                    1405.0,
+                    1119.0,
+                    1405.0,
+                    1156.0,
+                    1200.0,
+                    1156.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1149.0,
+                    352.0,
+                    1149.0,
+                    352.0,
+                    1191.0,
+                    291.0,
+                    1191.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    723.0,
+                    1149.0,
+                    1170.0,
+                    1149.0,
+                    1170.0,
+                    1191.0,
+                    723.0,
+                    1191.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1231.0,
+                    1149.0,
+                    1405.0,
+                    1149.0,
+                    1405.0,
+                    1191.0,
+                    1231.0,
+                    1191.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1183.0,
+                    298.0,
+                    1183.0,
+                    298.0,
+                    1227.0,
+                    292.0,
+                    1227.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    449.0,
+                    1183.0,
+                    520.0,
+                    1183.0,
+                    520.0,
+                    1227.0,
+                    449.0,
+                    1227.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    702.0,
+                    1183.0,
+                    1407.0,
+                    1183.0,
+                    1407.0,
+                    1227.0,
+                    702.0,
+                    1227.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1218.0,
+                    296.0,
+                    1218.0,
+                    296.0,
+                    1255.0,
+                    292.0,
+                    1255.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    407.0,
+                    1218.0,
+                    1306.0,
+                    1218.0,
+                    1306.0,
+                    1255.0,
+                    407.0,
+                    1255.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1325.0,
+                    1218.0,
+                    1406.0,
+                    1218.0,
+                    1406.0,
+                    1255.0,
+                    1325.0,
+                    1255.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1250.0,
+                    1308.0,
+                    1250.0,
+                    1308.0,
+                    1285.0,
+                    294.0,
+                    1285.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1327.0,
+                    1250.0,
+                    1405.0,
+                    1250.0,
+                    1405.0,
+                    1285.0,
+                    1327.0,
+                    1285.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1277.0,
+                    1408.0,
+                    1277.0,
+                    1408.0,
+                    1319.0,
+                    291.0,
+                    1319.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1310.0,
+                    1405.0,
+                    1310.0,
+                    1405.0,
+                    1346.0,
+                    294.0,
+                    1346.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1341.0,
+                    937.0,
+                    1341.0,
+                    937.0,
+                    1377.0,
+                    294.0,
+                    1377.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1058.0,
+                    1341.0,
+                    1405.0,
+                    1341.0,
+                    1405.0,
+                    1377.0,
+                    1058.0,
+                    1377.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    290.0,
+                    1360.0,
+                    1330.0,
+                    1360.0,
+                    1330.0,
+                    1418.0,
+                    290.0,
+                    1418.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1402.0,
+                    1360.0,
+                    1407.0,
+                    1360.0,
+                    1407.0,
+                    1418.0,
+                    1402.0,
+                    1418.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1401.0,
+                    1408.0,
+                    1401.0,
+                    1408.0,
+                    1437.0,
+                    295.0,
+                    1437.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    413.0,
+                    686.0,
+                    413.0,
+                    686.0,
+                    456.0,
+                    291.0,
+                    456.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    753.0,
+                    413.0,
+                    926.0,
+                    413.0,
+                    926.0,
+                    456.0,
+                    753.0,
+                    456.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    995.0,
+                    413.0,
+                    1197.0,
+                    413.0,
+                    1197.0,
+                    456.0,
+                    995.0,
+                    456.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1301.0,
+                    413.0,
+                    1407.0,
+                    413.0,
+                    1407.0,
+                    456.0,
+                    1301.0,
+                    456.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    447.0,
+                    1353.0,
+                    447.0,
+                    1353.0,
+                    484.0,
+                    295.0,
+                    484.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1399.0,
+                    447.0,
+                    1409.0,
+                    447.0,
+                    1409.0,
+                    484.0,
+                    1399.0,
+                    484.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    479.0,
+                    566.0,
+                    479.0,
+                    566.0,
+                    520.0,
+                    292.0,
+                    520.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    669.0,
+                    479.0,
+                    1197.0,
+                    479.0,
+                    1197.0,
+                    520.0,
+                    669.0,
+                    520.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1405.0,
+                    479.0,
+                    1408.0,
+                    479.0,
+                    1408.0,
+                    520.0,
+                    1405.0,
+                    520.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    513.0,
+                    343.0,
+                    513.0,
+                    343.0,
+                    553.0,
+                    293.0,
+                    553.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    577.0,
+                    513.0,
+                    750.0,
+                    513.0,
+                    750.0,
+                    553.0,
+                    577.0,
+                    553.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    279.0,
+                    1402.0,
+                    279.0,
+                    1402.0,
+                    312.0,
+                    296.0,
+                    312.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    309.0,
+                    1404.0,
+                    309.0,
+                    1404.0,
+                    344.0,
+                    295.0,
+                    344.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    340.0,
+                    953.0,
+                    340.0,
+                    953.0,
+                    377.0,
+                    293.0,
+                    377.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1050.0,
+                    340.0,
+                    1405.0,
+                    340.0,
+                    1405.0,
+                    377.0,
+                    1050.0,
+                    377.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    372.0,
+                    1057.0,
+                    372.0,
+                    1057.0,
+                    408.0,
+                    293.0,
+                    408.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1913.0,
+                    776.0,
+                    1913.0,
+                    776.0,
+                    1950.0,
+                    294.0,
+                    1950.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    842.0,
+                    1913.0,
+                    1406.0,
+                    1913.0,
+                    1406.0,
+                    1950.0,
+                    842.0,
+                    1950.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1944.0,
+                    437.0,
+                    1944.0,
+                    437.0,
+                    1978.0,
+                    295.0,
+                    1978.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    460.0,
+                    1944.0,
+                    682.0,
+                    1944.0,
+                    682.0,
+                    1978.0,
+                    460.0,
+                    1978.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    751.0,
+                    1944.0,
+                    1405.0,
+                    1944.0,
+                    1405.0,
+                    1978.0,
+                    751.0,
+                    1978.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1975.0,
+                    665.0,
+                    1975.0,
+                    665.0,
+                    2012.0,
+                    294.0,
+                    2012.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    730.0,
+                    1975.0,
+                    1356.0,
+                    1975.0,
+                    1356.0,
+                    2012.0,
+                    730.0,
+                    2012.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    197.0,
+                    533.0,
+                    197.0,
+                    533.0,
+                    240.0,
+                    292.0,
+                    240.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    586.0,
+                    197.0,
+                    1406.0,
+                    197.0,
+                    1406.0,
+                    240.0,
+                    586.0,
+                    240.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    233.0,
+                    1154.0,
+                    233.0,
+                    1154.0,
+                    269.0,
+                    293.0,
+                    269.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1816.0,
+                    901.0,
+                    1816.0,
+                    901.0,
+                    1857.0,
+                    295.0,
+                    1857.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    924.0,
+                    1816.0,
+                    1070.0,
+                    1816.0,
+                    1070.0,
+                    1857.0,
+                    924.0,
+                    1857.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1165.0,
+                    1816.0,
+                    1259.0,
+                    1816.0,
+                    1259.0,
+                    1857.0,
+                    1165.0,
+                    1857.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1298.0,
+                    1816.0,
+                    1326.0,
+                    1816.0,
+                    1326.0,
+                    1857.0,
+                    1298.0,
+                    1857.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1397.0,
+                    1816.0,
+                    1409.0,
+                    1816.0,
+                    1409.0,
+                    1857.0,
+                    1397.0,
+                    1857.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1848.0,
+                    558.0,
+                    1848.0,
+                    558.0,
+                    1888.0,
+                    293.0,
+                    1888.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    956.0,
+                    1848.0,
+                    1118.0,
+                    1848.0,
+                    1118.0,
+                    1888.0,
+                    956.0,
+                    1888.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1198.0,
+                    1848.0,
+                    1277.0,
+                    1848.0,
+                    1277.0,
+                    1888.0,
+                    1198.0,
+                    1888.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1399.0,
+                    1848.0,
+                    1407.0,
+                    1848.0,
+                    1407.0,
+                    1888.0,
+                    1399.0,
+                    1888.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 7,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1761,
+                    1405,
+                    1761,
+                    1405,
+                    2008,
+                    297,
+                    2008
+                ],
+                "score": 0.981
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    296,
+                    357,
+                    1406,
+                    357,
+                    1406,
+                    556,
+                    296,
+                    556
+                ],
+                "score": 0.978
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1552,
+                    1406,
+                    1552,
+                    1406,
+                    1654,
+                    298,
+                    1654
+                ],
+                "score": 0.972
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    296,
+                    1174,
+                    1405,
+                    1174,
+                    1405,
+                    1334,
+                    296,
+                    1334
+                ],
+                "score": 0.972
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    296,
+                    1047,
+                    1404,
+                    1047,
+                    1404,
+                    1154,
+                    296,
+                    1154
+                ],
+                "score": 0.966
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    294,
+                    977,
+                    1401,
+                    977,
+                    1401,
+                    1040,
+                    294,
+                    1040
+                ],
+                "score": 0.953
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    455,
+                    255,
+                    1250,
+                    255,
+                    1250,
+                    311,
+                    455,
+                    311
+                ],
+                "score": 0.933
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    326,
+                    1423,
+                    1378,
+                    1423,
+                    1378,
+                    1522,
+                    326,
+                    1522
+                ],
+                "score": 0.932
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    294,
+                    199,
+                    1175,
+                    199,
+                    1175,
+                    239,
+                    294,
+                    239
+                ],
+                "score": 0.929
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1340,
+                    1406,
+                    1340,
+                    1406,
+                    1412,
+                    297,
+                    1412
+                ],
+                "score": 0.925
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    297,
+                    1693,
+                    510,
+                    1693,
+                    510,
+                    1732,
+                    297,
+                    1732
+                ],
+                "score": 0.918
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    840,
+                    2061,
+                    859,
+                    2061,
+                    859,
+                    2084,
+                    840,
+                    2084
+                ],
+                "score": 0.799
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    271,
+                    634,
+                    1388,
+                    634,
+                    1388,
+                    938,
+                    271,
+                    938
+                ],
+                "score": 0.784
+            },
+            {
+                "category_id": 6,
+                "poly": [
+                    296,
+                    596,
+                    1047,
+                    596,
+                    1047,
+                    629,
+                    296,
+                    629
+                ],
+                "score": 0.412
+            },
+            {
+                "category_id": 5,
+                "poly": [
+                    271,
+                    634,
+                    1388,
+                    634,
+                    1388,
+                    938,
+                    271,
+                    938
+                ],
+                "score": 0.192,
+                "html": "<table><tr><td>Input: S={xi}i∈[n],α ∈ (0,1/2)，(ε,δ)</td></tr><tr><td>1 if n &lt;cd1/2 log(1/δ)/ (εα√log(1/α)) then 0utput: 0 [cd1/2 log(1/δ)/ (εα) for hevay-tail]</td></tr><tr><td>2 (x,B) ← qrange(S,(1/3)ε,(1/3)δ) [qrange-ht (·) for hevay-tail] 3 Project the data points onto the bal: xi ← PBaB/2(x)(xi), for alli ∈ [n]</td></tr><tr><td>4 R(S) ← R(S) + Lap(3Bd1/2/(nε))</td></tr><tr><td>5 if R(S) &gt; 2α√log(1/α) then Output: @ [R(S) &gt; 2cs √α for hevay-tail]</td></tr><tr><td>6 else Output: a randomly drawn point β ∈ B√dB/2(x) sampled from a density</td></tr><tr><td>r(μ) x e-(1/(24√/10g(1/a))ε n d(μ,s) [e-(en√α/(24c))d(@,S) for heavy-tail] 7</td></tr></table>"
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    389,
+                    889,
+                    786,
+                    889,
+                    786,
+                    931,
+                    389,
+                    931
+                ],
+                "score": 0.124
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    638,
+                    978,
+                    724,
+                    978,
+                    724,
+                    1012,
+                    638,
+                    1012
+                ],
+                "score": 0.93,
+                "latex": "d ( \\hat { \\mu } , S )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    689,
+                    1267,
+                    928,
+                    1267,
+                    928,
+                    1306,
+                    689,
+                    1306
+                ],
+                "score": 0.93,
+                "latex": "{ \\cal O } ( ( 1 / n ) \\sqrt { \\log ( 1 / \\alpha ) } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    443,
+                    519,
+                    568,
+                    519,
+                    568,
+                    556,
+                    443,
+                    556
+                ],
+                "score": 0.93,
+                "latex": "O ( d ^ { 1 / 2 } / n )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    793,
+                    1373,
+                    1129,
+                    1373,
+                    1129,
+                    1412,
+                    793,
+                    1412
+                ],
+                "score": 0.93,
+                "latex": "\\| \\hat { \\mu } - \\mu \\| _ { 2 } = O ( \\alpha \\sqrt { \\log ( 1 / \\alpha ) } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    818,
+                    1552,
+                    940,
+                    1552,
+                    940,
+                    1589,
+                    818,
+                    1589
+                ],
+                "score": 0.93,
+                "latex": "O ( d e ^ { \\Theta ( n ) } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    774,
+                    481,
+                    1014,
+                    481,
+                    1014,
+                    520,
+                    774,
+                    520
+                ],
+                "score": 0.93,
+                "latex": "{ \\cal O } ( ( 1 / n ) \\sqrt { \\log ( 1 / \\alpha ) } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1029,
+                    420,
+                    1114,
+                    420,
+                    1114,
+                    454,
+                    1029,
+                    454
+                ],
+                "score": 0.93,
+                "latex": "d ( \\hat { \\mu } , S )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    669,
+                    1049,
+                    762,
+                    1049,
+                    762,
+                    1083,
+                    669,
+                    1083
+                ],
+                "score": 0.92,
+                "latex": "\\{ x _ { i } \\} _ { i \\in S }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    561,
+                    1556,
+                    624,
+                    1556,
+                    624,
+                    1589,
+                    561,
+                    1589
+                ],
+                "score": 0.92,
+                "latex": "R ( S )"
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    333,
+                    1423,
+                    1367,
+                    1423,
+                    1367,
+                    1526,
+                    333,
+                    1526
+                ],
+                "score": 0.92,
+                "latex": "n = \\widetilde \\Omega \\Big ( \\frac { d + \\log \\frac { 1 } { \\xi } } { \\alpha ^ { 2 } \\log \\frac { 1 } { \\alpha } } + \\frac { d \\log \\Big ( d \\sqrt { \\log ( d n / \\zeta ) } / \\alpha \\Big ) + d ^ { 1 / 2 } \\log \\frac { 1 } { \\delta } + \\log \\frac { 1 } { \\zeta } } { \\varepsilon \\alpha } + \\frac { \\sqrt { d \\log \\frac { 1 } { \\delta } } \\log \\frac { d } { \\zeta \\delta } } { \\varepsilon } \\Big ) ~ ."
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    848,
+                    390,
+                    911,
+                    390,
+                    911,
+                    423,
+                    848,
+                    423
+                ],
+                "score": 0.92,
+                "latex": "R ( S )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1116,
+                    1586,
+                    1402,
+                    1586,
+                    1402,
+                    1626,
+                    1116,
+                    1626
+                ],
+                "score": 0.92,
+                "latex": "O ( n d ( \\sqrt { \\log ( d n / \\zeta ) } / \\alpha ) ^ { d } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    562,
+                    1080,
+                    623,
+                    1080,
+                    623,
+                    1113,
+                    562,
+                    1113
+                ],
+                "score": 0.92,
+                "latex": "v ^ { \\top } x _ { i }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    744,
+                    202,
+                    1025,
+                    202,
+                    1025,
+                    239,
+                    744,
+                    239
+                ],
+                "score": 0.92,
+                "latex": "\\mu ( S ) = ( 1 / | S | ) \\sum _ { i \\in S } x _ { i }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    724,
+                    849,
+                    903,
+                    849,
+                    903,
+                    890,
+                    724,
+                    890
+                ],
+                "score": 0.92,
+                "latex": "\\hat { \\mu } \\in \\mathcal { B } _ { \\sqrt { d } B / 2 } ( \\bar { x } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    965,
+                    1175,
+                    1051,
+                    1175,
+                    1051,
+                    1209,
+                    965,
+                    1209
+                ],
+                "score": 0.92,
+                "latex": "\\mu ( \\mathcal { M } ^ { v } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    856,
+                    1083,
+                    925,
+                    1083,
+                    925,
+                    1115,
+                    856,
+                    1115
+                ],
+                "score": 0.91,
+                "latex": "3 \\alpha | S |"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    706,
+                    735,
+                    950,
+                    735,
+                    950,
+                    777,
+                    706,
+                    777
+                ],
+                "score": 0.91,
+                "latex": "x _ { i }  \\mathcal { P } _ { B _ { \\sqrt { d } B / 2 } ( \\bar { x } ) } ( x _ { i } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1005,
+                    814,
+                    1194,
+                    814,
+                    1194,
+                    853,
+                    1005,
+                    853
+                ],
+                "score": 0.91,
+                "latex": "[ \\widehat { R } ( S ) > 2 c _ { \\zeta } \\sqrt { \\alpha }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    993,
+                    1047,
+                    1100,
+                    1047,
+                    1100,
+                    1079,
+                    993,
+                    1079
+                ],
+                "score": 0.91,
+                "latex": "v \\in \\mathbb { S } ^ { d - 1 }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    322,
+                    815,
+                    589,
+                    815,
+                    589,
+                    854,
+                    322,
+                    854
+                ],
+                "score": 0.91,
+                "latex": "\\widehat { R } ( S ) > 2 \\alpha \\sqrt { \\log ( 1 / \\alpha ) }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1211,
+                    1080,
+                    1272,
+                    1080,
+                    1272,
+                    1113,
+                    1211,
+                    1113
+                ],
+                "score": 0.91,
+                "latex": "v ^ { \\top } x _ { i }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    298,
+                    1116,
+                    548,
+                    1116,
+                    548,
+                    1151,
+                    298,
+                    1151
+                ],
+                "score": 0.9,
+                "latex": "\\mathcal { M } ^ { v } = S \\setminus ( \\mathcal { T } ^ { v } \\cup B ^ { v } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1332,
+                    1050,
+                    1400,
+                    1050,
+                    1400,
+                    1083,
+                    1332,
+                    1083
+                ],
+                "score": 0.9,
+                "latex": "3 \\alpha | S |"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    902,
+                    520,
+                    942,
+                    520,
+                    942,
+                    552,
+                    902,
+                    552
+                ],
+                "score": 0.89,
+                "latex": "\\sqrt { d }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    913,
+                    671,
+                    1194,
+                    671,
+                    1194,
+                    708,
+                    913,
+                    708
+                ],
+                "score": 0.89,
+                "latex": "\\emptyset \\ [ \\ c d ^ { 1 / 2 } \\log ( 1 / \\delta ) / \\ ( \\varepsilon \\alpha )"
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    381,
+                    886,
+                    788,
+                    886,
+                    788,
+                    932,
+                    381,
+                    932
+                ],
+                "score": 0.88,
+                "latex": "r ( \\hat { \\mu } ) \\propto e ^ { - ( 1 / ( 2 4 \\sqrt { \\log ( 1 / \\alpha ) } ) ) \\varepsilon n d ( \\hat { \\mu } , S ) }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    512,
+                    394,
+                    582,
+                    394,
+                    582,
+                    423,
+                    512,
+                    423
+                ],
+                "score": 0.88,
+                "latex": "q _ { \\mathrm { r a n g e } }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1314,
+                    1377,
+                    1406,
+                    1377,
+                    1406,
+                    1411,
+                    1314,
+                    1411
+                ],
+                "score": 0.87,
+                "latex": "1 - \\zeta i f"
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    445,
+                    252,
+                    1256,
+                    252,
+                    1256,
+                    310,
+                    445,
+                    310
+                ],
+                "score": 0.87,
+                "latex": "R ( S ) \\stackrel { \\Delta } { = } \\operatorname* { m i n } _ { \\substack { S ^ { \\prime } \\subset S , | S ^ { \\prime } | = ( 1 - 2 \\alpha ) | S | . } } \\operatorname* { m a x } _ { \\substack { T \\subset S ^ { \\prime } , | T | = ( 1 - \\alpha ) | S ^ { \\prime } | . } } \\| \\mu ( T ) - \\mu ( S ^ { \\prime } ) \\| _ { 2 } ."
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1186,
+                    1050,
+                    1224,
+                    1050,
+                    1224,
+                    1078,
+                    1186,
+                    1078
+                ],
+                "score": 0.86,
+                "latex": "\\mathcal { T } ^ { v }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    705,
+                    1084,
+                    741,
+                    1084,
+                    741,
+                    1111,
+                    705,
+                    1111
+                ],
+                "score": 0.86,
+                "latex": "B ^ { v }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    295,
+                    774,
+                    719,
+                    774,
+                    719,
+                    814,
+                    295,
+                    814
+                ],
+                "score": 0.85,
+                "latex": "\\widehat { R } ( S ) \\gets R ( S ) + \\mathrm { L a p } ( 3 B d ^ { 1 / 2 } / ( n \\varepsilon ) )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    858,
+                    1047,
+                    894,
+                    1047,
+                    894,
+                    1079,
+                    858,
+                    1079
+                ],
+                "score": 0.85,
+                "latex": "\\mathbb { R } ^ { d }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    670,
+                    1593,
+                    690,
+                    1593,
+                    690,
+                    1623,
+                    670,
+                    1623
+                ],
+                "score": 0.85,
+                "latex": "\\hat { \\mu }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    614,
+                    452,
+                    635,
+                    452,
+                    635,
+                    483,
+                    614,
+                    483
+                ],
+                "score": 0.84,
+                "latex": "\\hat { \\mu }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    802,
+                    1010,
+                    822,
+                    1010,
+                    822,
+                    1041,
+                    802,
+                    1041
+                ],
+                "score": 0.84,
+                "latex": "\\hat { \\mu }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1293,
+                    1340,
+                    1355,
+                    1340,
+                    1355,
+                    1375,
+                    1293,
+                    1375
+                ],
+                "score": 0.83,
+                "latex": "( \\varepsilon , \\delta )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    932,
+                    892,
+                    1185,
+                    892,
+                    1185,
+                    927,
+                    932,
+                    927
+                ],
+                "score": 0.8,
+                "latex": "[ e ^ { - ( \\varepsilon n \\sqrt { \\alpha } / ( 2 4 c _ { \\zeta } ) ) d ( \\hat { \\mu } , S ) }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    297,
+                    708,
+                    717,
+                    708,
+                    717,
+                    739,
+                    297,
+                    739
+                ],
+                "score": 0.79,
+                "latex": "( \\bar { x } , B ) \\gets q _ { \\mathrm { r a n g e } } ( S , ( 1 / 3 ) \\varepsilon , ( 1 / 3 ) \\delta )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    382,
+                    635,
+                    542,
+                    635,
+                    542,
+                    672,
+                    382,
+                    672
+                ],
+                "score": 0.79,
+                "latex": "S = \\{ x _ { i } \\} _ { i \\in [ n ] }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    641,
+                    1115,
+                    1142,
+                    1115,
+                    1142,
+                    1154,
+                    641,
+                    1154
+                ],
+                "score": 0.78,
+                "latex": "\\begin{array} { r l r } { d ( \\hat { \\mu } , S ) } & { \\triangleq } & { \\operatorname* { m a x } _ { v \\in \\mathbb { S } ^ { d - 1 } } \\left. v ^ { \\top } \\left( \\mu ( \\mathcal { M } ^ { v } ) - \\hat { \\mu } \\right) \\right. } \\end{array}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1035,
+                    737,
+                    1113,
+                    737,
+                    1113,
+                    769,
+                    1035,
+                    769
+                ],
+                "score": 0.77,
+                "latex": "i \\in [ n ]"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    503,
+                    1596,
+                    524,
+                    1596,
+                    524,
+                    1619,
+                    503,
+                    1619
+                ],
+                "score": 0.77,
+                "latex": "\\alpha"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    322,
+                    673,
+                    742,
+                    673,
+                    742,
+                    708,
+                    322,
+                    708
+                ],
+                "score": 0.77,
+                "latex": "n < c d ^ { 1 / 2 } \\log ( 1 / \\delta ) / \\left( \\varepsilon \\alpha \\sqrt { \\log ( 1 / \\alpha ) } \\right)"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    400,
+                    1275,
+                    421,
+                    1275,
+                    421,
+                    1298,
+                    400,
+                    1298
+                ],
+                "score": 0.76,
+                "latex": "\\sigma"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    553,
+                    1276,
+                    572,
+                    1276,
+                    572,
+                    1298,
+                    553,
+                    1298
+                ],
+                "score": 0.74,
+                "latex": "n"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    609,
+                    1182,
+                    626,
+                    1182,
+                    626,
+                    1203,
+                    609,
+                    1203
+                ],
+                "score": 0.74,
+                "latex": "v"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1043,
+                    708,
+                    1194,
+                    708,
+                    1194,
+                    739,
+                    1043,
+                    739
+                ],
+                "score": 0.73,
+                "latex": "[ \\ q _ { \\mathrm { r a n g e - h t } } ( \\cdot )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    998,
+                    1598,
+                    1016,
+                    1598,
+                    1016,
+                    1619,
+                    998,
+                    1619
+                ],
+                "score": 0.72,
+                "latex": "v"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    710,
+                    635,
+                    771,
+                    635,
+                    771,
+                    670,
+                    710,
+                    670
+                ],
+                "score": 0.72,
+                "latex": "( \\varepsilon , \\delta )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    554,
+                    635,
+                    697,
+                    635,
+                    697,
+                    670,
+                    554,
+                    670
+                ],
+                "score": 0.72,
+                "latex": "\\alpha \\in ( 0 , 1 / 2 )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    668,
+                    1212,
+                    685,
+                    1212,
+                    685,
+                    1233,
+                    668,
+                    1233
+                ],
+                "score": 0.71,
+                "latex": "v"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    507,
+                    1242,
+                    524,
+                    1242,
+                    524,
+                    1264,
+                    507,
+                    1264
+                ],
+                "score": 0.7,
+                "latex": "v"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    756,
+                    819,
+                    775,
+                    819,
+                    775,
+                    848,
+                    756,
+                    848
+                ],
+                "score": 0.69,
+                "latex": "\\varnothing"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    924,
+                    1114,
+                    1140,
+                    1114,
+                    1140,
+                    1155,
+                    924,
+                    1155
+                ],
+                "score": 0.55,
+                "latex": "\\left| \\boldsymbol v ^ { \\top } \\left( \\boldsymbol \\mu ( \\mathcal { M } ^ { v } ) - \\boldsymbol \\hat { \\mu } \\right) \\right|"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    507,
+                    1380,
+                    521,
+                    1380,
+                    521,
+                    1405,
+                    507,
+                    1405
+                ],
+                "score": 0.48,
+                "latex": "^ { l }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    381,
+                    634,
+                    775,
+                    634,
+                    775,
+                    670,
+                    381,
+                    670
+                ],
+                "score": 0.27,
+                "latex": "S = \\{ x _ { i } \\} _ { i \\in [ n ] } , \\alpha \\in ( 0 , 1 / 2 ) , ( \\varepsilon , \\delta )"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    289.0,
+                    1687.0,
+                    514.0,
+                    1687.0,
+                    514.0,
+                    1740.0,
+                    289.0,
+                    1740.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    839.0,
+                    2060.0,
+                    861.0,
+                    2060.0,
+                    861.0,
+                    2090.0,
+                    839.0,
+                    2090.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    594.0,
+                    1050.0,
+                    594.0,
+                    1050.0,
+                    633.0,
+                    297.0,
+                    633.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1762.0,
+                    1405.0,
+                    1762.0,
+                    1405.0,
+                    1798.0,
+                    294.0,
+                    1798.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1793.0,
+                    1403.0,
+                    1793.0,
+                    1403.0,
+                    1827.0,
+                    295.0,
+                    1827.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1824.0,
+                    1405.0,
+                    1824.0,
+                    1405.0,
+                    1857.0,
+                    294.0,
+                    1857.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1856.0,
+                    1406.0,
+                    1856.0,
+                    1406.0,
+                    1889.0,
+                    296.0,
+                    1889.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1883.0,
+                    1405.0,
+                    1883.0,
+                    1405.0,
+                    1919.0,
+                    294.0,
+                    1919.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1913.0,
+                    1406.0,
+                    1913.0,
+                    1406.0,
+                    1950.0,
+                    292.0,
+                    1950.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1946.0,
+                    1405.0,
+                    1946.0,
+                    1405.0,
+                    1980.0,
+                    294.0,
+                    1980.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1976.0,
+                    1407.0,
+                    1976.0,
+                    1407.0,
+                    2011.0,
+                    295.0,
+                    2011.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    359.0,
+                    1404.0,
+                    359.0,
+                    1404.0,
+                    392.0,
+                    296.0,
+                    392.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    388.0,
+                    511.0,
+                    388.0,
+                    511.0,
+                    427.0,
+                    291.0,
+                    427.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    583.0,
+                    388.0,
+                    847.0,
+                    388.0,
+                    847.0,
+                    427.0,
+                    583.0,
+                    427.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    912.0,
+                    388.0,
+                    1407.0,
+                    388.0,
+                    1407.0,
+                    427.0,
+                    912.0,
+                    427.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    419.0,
+                    1028.0,
+                    419.0,
+                    1028.0,
+                    457.0,
+                    294.0,
+                    457.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1115.0,
+                    419.0,
+                    1406.0,
+                    419.0,
+                    1406.0,
+                    457.0,
+                    1115.0,
+                    457.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    449.0,
+                    613.0,
+                    449.0,
+                    613.0,
+                    485.0,
+                    293.0,
+                    485.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    636.0,
+                    449.0,
+                    1406.0,
+                    449.0,
+                    1406.0,
+                    485.0,
+                    636.0,
+                    485.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    482.0,
+                    773.0,
+                    482.0,
+                    773.0,
+                    522.0,
+                    291.0,
+                    522.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1015.0,
+                    482.0,
+                    1407.0,
+                    482.0,
+                    1407.0,
+                    522.0,
+                    1015.0,
+                    522.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    517.0,
+                    442.0,
+                    517.0,
+                    442.0,
+                    561.0,
+                    293.0,
+                    561.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    569.0,
+                    517.0,
+                    901.0,
+                    517.0,
+                    901.0,
+                    561.0,
+                    569.0,
+                    561.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    943.0,
+                    517.0,
+                    1237.0,
+                    517.0,
+                    1237.0,
+                    561.0,
+                    943.0,
+                    561.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1550.0,
+                    560.0,
+                    1550.0,
+                    560.0,
+                    1593.0,
+                    293.0,
+                    1593.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    625.0,
+                    1550.0,
+                    817.0,
+                    1550.0,
+                    817.0,
+                    1593.0,
+                    625.0,
+                    1593.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    941.0,
+                    1550.0,
+                    1406.0,
+                    1550.0,
+                    1406.0,
+                    1593.0,
+                    941.0,
+                    1593.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1584.0,
+                    502.0,
+                    1584.0,
+                    502.0,
+                    1631.0,
+                    291.0,
+                    1631.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    525.0,
+                    1584.0,
+                    669.0,
+                    1584.0,
+                    669.0,
+                    1631.0,
+                    525.0,
+                    1631.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    691.0,
+                    1584.0,
+                    997.0,
+                    1584.0,
+                    997.0,
+                    1631.0,
+                    691.0,
+                    1631.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1017.0,
+                    1584.0,
+                    1115.0,
+                    1584.0,
+                    1115.0,
+                    1631.0,
+                    1017.0,
+                    1631.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1403.0,
+                    1584.0,
+                    1407.0,
+                    1584.0,
+                    1407.0,
+                    1631.0,
+                    1403.0,
+                    1631.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1621.0,
+                    427.0,
+                    1621.0,
+                    427.0,
+                    1658.0,
+                    293.0,
+                    1658.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1174.0,
+                    608.0,
+                    1174.0,
+                    608.0,
+                    1212.0,
+                    295.0,
+                    1212.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    627.0,
+                    1174.0,
+                    964.0,
+                    1174.0,
+                    964.0,
+                    1212.0,
+                    627.0,
+                    1212.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1052.0,
+                    1174.0,
+                    1406.0,
+                    1174.0,
+                    1406.0,
+                    1212.0,
+                    1052.0,
+                    1212.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1204.0,
+                    667.0,
+                    1204.0,
+                    667.0,
+                    1240.0,
+                    295.0,
+                    1240.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    686.0,
+                    1204.0,
+                    1403.0,
+                    1204.0,
+                    1403.0,
+                    1240.0,
+                    686.0,
+                    1240.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1235.0,
+                    506.0,
+                    1235.0,
+                    506.0,
+                    1273.0,
+                    294.0,
+                    1273.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    525.0,
+                    1235.0,
+                    1405.0,
+                    1235.0,
+                    1405.0,
+                    1273.0,
+                    525.0,
+                    1273.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1268.0,
+                    399.0,
+                    1268.0,
+                    399.0,
+                    1307.0,
+                    293.0,
+                    1307.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    422.0,
+                    1268.0,
+                    552.0,
+                    1268.0,
+                    552.0,
+                    1307.0,
+                    422.0,
+                    1307.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    573.0,
+                    1268.0,
+                    688.0,
+                    1268.0,
+                    688.0,
+                    1307.0,
+                    573.0,
+                    1307.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    929.0,
+                    1268.0,
+                    1406.0,
+                    1268.0,
+                    1406.0,
+                    1307.0,
+                    929.0,
+                    1307.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1300.0,
+                    1333.0,
+                    1300.0,
+                    1333.0,
+                    1338.0,
+                    294.0,
+                    1338.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1042.0,
+                    668.0,
+                    1042.0,
+                    668.0,
+                    1086.0,
+                    293.0,
+                    1086.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    763.0,
+                    1042.0,
+                    857.0,
+                    1042.0,
+                    857.0,
+                    1086.0,
+                    763.0,
+                    1086.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    895.0,
+                    1042.0,
+                    992.0,
+                    1042.0,
+                    992.0,
+                    1086.0,
+                    895.0,
+                    1086.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1101.0,
+                    1042.0,
+                    1185.0,
+                    1042.0,
+                    1185.0,
+                    1086.0,
+                    1101.0,
+                    1086.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1225.0,
+                    1042.0,
+                    1331.0,
+                    1042.0,
+                    1331.0,
+                    1086.0,
+                    1225.0,
+                    1086.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1401.0,
+                    1042.0,
+                    1406.0,
+                    1042.0,
+                    1406.0,
+                    1086.0,
+                    1401.0,
+                    1086.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    290.0,
+                    1079.0,
+                    561.0,
+                    1079.0,
+                    561.0,
+                    1117.0,
+                    290.0,
+                    1117.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    624.0,
+                    1079.0,
+                    704.0,
+                    1079.0,
+                    704.0,
+                    1117.0,
+                    624.0,
+                    1117.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    742.0,
+                    1079.0,
+                    855.0,
+                    1079.0,
+                    855.0,
+                    1117.0,
+                    742.0,
+                    1117.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    926.0,
+                    1079.0,
+                    1210.0,
+                    1079.0,
+                    1210.0,
+                    1117.0,
+                    926.0,
+                    1117.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1273.0,
+                    1079.0,
+                    1406.0,
+                    1079.0,
+                    1406.0,
+                    1117.0,
+                    1273.0,
+                    1117.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    290.0,
+                    1108.0,
+                    297.0,
+                    1108.0,
+                    297.0,
+                    1157.0,
+                    290.0,
+                    1157.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    549.0,
+                    1108.0,
+                    640.0,
+                    1108.0,
+                    640.0,
+                    1157.0,
+                    549.0,
+                    1157.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1143.0,
+                    1108.0,
+                    1160.0,
+                    1108.0,
+                    1160.0,
+                    1157.0,
+                    1143.0,
+                    1157.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    975.0,
+                    637.0,
+                    975.0,
+                    637.0,
+                    1015.0,
+                    294.0,
+                    1015.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    725.0,
+                    975.0,
+                    1406.0,
+                    975.0,
+                    1406.0,
+                    1015.0,
+                    725.0,
+                    1015.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1005.0,
+                    801.0,
+                    1005.0,
+                    801.0,
+                    1044.0,
+                    294.0,
+                    1044.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    823.0,
+                    1005.0,
+                    832.0,
+                    1005.0,
+                    832.0,
+                    1044.0,
+                    823.0,
+                    1044.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    194.0,
+                    743.0,
+                    194.0,
+                    743.0,
+                    245.0,
+                    292.0,
+                    245.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1026.0,
+                    194.0,
+                    1177.0,
+                    194.0,
+                    1177.0,
+                    245.0,
+                    1026.0,
+                    245.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    1341.0,
+                    1292.0,
+                    1341.0,
+                    1292.0,
+                    1375.0,
+                    297.0,
+                    1375.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1356.0,
+                    1341.0,
+                    1407.0,
+                    1341.0,
+                    1407.0,
+                    1375.0,
+                    1356.0,
+                    1375.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1372.0,
+                    506.0,
+                    1372.0,
+                    506.0,
+                    1416.0,
+                    294.0,
+                    1416.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    522.0,
+                    1372.0,
+                    792.0,
+                    1372.0,
+                    792.0,
+                    1416.0,
+                    522.0,
+                    1416.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1130.0,
+                    1372.0,
+                    1313.0,
+                    1372.0,
+                    1313.0,
+                    1416.0,
+                    1130.0,
+                    1416.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    631.0,
+                    380.0,
+                    631.0,
+                    380.0,
+                    672.0,
+                    293.0,
+                    672.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    269.0,
+                    669.0,
+                    321.0,
+                    669.0,
+                    321.0,
+                    712.0,
+                    269.0,
+                    712.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    743.0,
+                    669.0,
+                    912.0,
+                    669.0,
+                    912.0,
+                    712.0,
+                    743.0,
+                    712.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1195.0,
+                    669.0,
+                    1363.0,
+                    669.0,
+                    1363.0,
+                    712.0,
+                    1195.0,
+                    712.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    269.0,
+                    703.0,
+                    296.0,
+                    703.0,
+                    296.0,
+                    742.0,
+                    269.0,
+                    742.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1032.0,
+                    705.0,
+                    1042.0,
+                    705.0,
+                    1042.0,
+                    740.0,
+                    1032.0,
+                    740.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1195.0,
+                    705.0,
+                    1362.0,
+                    705.0,
+                    1362.0,
+                    740.0,
+                    1195.0,
+                    740.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    265.0,
+                    730.0,
+                    705.0,
+                    730.0,
+                    705.0,
+                    778.0,
+                    265.0,
+                    778.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    951.0,
+                    730.0,
+                    1034.0,
+                    730.0,
+                    1034.0,
+                    778.0,
+                    951.0,
+                    778.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1114.0,
+                    730.0,
+                    1120.0,
+                    730.0,
+                    1120.0,
+                    778.0,
+                    1114.0,
+                    778.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    267.0,
+                    769.0,
+                    294.0,
+                    769.0,
+                    294.0,
+                    818.0,
+                    267.0,
+                    818.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    720.0,
+                    769.0,
+                    724.0,
+                    769.0,
+                    724.0,
+                    818.0,
+                    720.0,
+                    818.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    267.0,
+                    814.0,
+                    321.0,
+                    814.0,
+                    321.0,
+                    853.0,
+                    267.0,
+                    853.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    590.0,
+                    814.0,
+                    755.0,
+                    814.0,
+                    755.0,
+                    853.0,
+                    590.0,
+                    853.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    776.0,
+                    814.0,
+                    779.0,
+                    814.0,
+                    779.0,
+                    853.0,
+                    776.0,
+                    853.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1001.0,
+                    810.0,
+                    1004.0,
+                    810.0,
+                    1004.0,
+                    858.0,
+                    1001.0,
+                    858.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1195.0,
+                    810.0,
+                    1363.0,
+                    810.0,
+                    1363.0,
+                    858.0,
+                    1195.0,
+                    858.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    267.0,
+                    845.0,
+                    723.0,
+                    845.0,
+                    723.0,
+                    890.0,
+                    267.0,
+                    890.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    904.0,
+                    845.0,
+                    1176.0,
+                    845.0,
+                    1176.0,
+                    890.0,
+                    904.0,
+                    890.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    272.0,
+                    902.0,
+                    291.0,
+                    902.0,
+                    291.0,
+                    925.0,
+                    272.0,
+                    925.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    926.0,
+                    882.0,
+                    931.0,
+                    882.0,
+                    931.0,
+                    938.0,
+                    926.0,
+                    938.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1186.0,
+                    882.0,
+                    1360.0,
+                    882.0,
+                    1360.0,
+                    938.0,
+                    1186.0,
+                    938.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    378.75,
+                    885.5,
+                    790.75,
+                    885.5,
+                    790.75,
+                    929.5,
+                    378.75,
+                    929.5
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 8,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    544,
+                    1406,
+                    544,
+                    1406,
+                    1113,
+                    297,
+                    1113
+                ],
+                "score": 0.982
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    203,
+                    1403,
+                    203,
+                    1403,
+                    389,
+                    298,
+                    389
+                ],
+                "score": 0.974
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    403,
+                    1403,
+                    403,
+                    1403,
+                    525,
+                    299,
+                    525
+                ],
+                "score": 0.967
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    1224,
+                    1405,
+                    1224,
+                    1405,
+                    1379,
+                    299,
+                    1379
+                ],
+                "score": 0.966
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    308,
+                    1470,
+                    1408,
+                    1470,
+                    1408,
+                    2012,
+                    308,
+                    2012
+                ],
+                "score": 0.959
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    300,
+                    1421,
+                    454,
+                    1421,
+                    454,
+                    1455,
+                    300,
+                    1455
+                ],
+                "score": 0.905
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    300,
+                    1157,
+                    562,
+                    1157,
+                    562,
+                    1194,
+                    300,
+                    1194
+                ],
+                "score": 0.898
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    836,
+                    2061,
+                    865,
+                    2061,
+                    865,
+                    2085,
+                    836,
+                    2085
+                ],
+                "score": 0.82
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    301,
+                    1156,
+                    562,
+                    1156,
+                    562,
+                    1194,
+                    301,
+                    1194
+                ],
+                "score": 0.354
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    419,
+                    803,
+                    496,
+                    803,
+                    496,
+                    843,
+                    419,
+                    843
+                ],
+                "score": 0.93,
+                "latex": "\\widetilde O ( n d )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1037,
+                    767,
+                    1190,
+                    767,
+                    1190,
+                    806,
+                    1037,
+                    806
+                ],
+                "score": 0.93,
+                "latex": "{ \\widetilde { O } } ( d ^ { 3 } + n d ^ { 2 } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1142,
+                    539,
+                    1408,
+                    539,
+                    1408,
+                    579,
+                    1142,
+                    579
+                ],
+                "score": 0.93,
+                "latex": "\\widetilde \\Omega ( ( d / ( \\alpha ^ { 2 } \\log ( 1 / \\alpha ) ) ) +"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1064,
+                    639,
+                    1166,
+                    639,
+                    1166,
+                    673,
+                    1064,
+                    673
+                ],
+                "score": 0.91,
+                "latex": "\\bar { \\log ( 1 / \\delta ) }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    295,
+                    706,
+                    469,
+                    706,
+                    469,
+                    740,
+                    295,
+                    740
+                ],
+                "score": 0.91,
+                "latex": "( 1 / \\varepsilon ) \\log ( 1 / \\delta ) )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1190,
+                    354,
+                    1242,
+                    354,
+                    1242,
+                    385,
+                    1190,
+                    385
+                ],
+                "score": 0.89,
+                "latex": "d ^ { 1 / 2 }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    296,
+                    576,
+                    653,
+                    576,
+                    653,
+                    613,
+                    296,
+                    613
+                ],
+                "score": 0.88,
+                "latex": "( d ^ { 3 / 2 } / ( \\varepsilon \\alpha \\log ( 1 / \\alpha ) ) ) \\log ( 1 / \\delta ) )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    860,
+                    672,
+                    1409,
+                    672,
+                    1409,
+                    710,
+                    860,
+                    710
+                ],
+                "score": 0.87,
+                "latex": "\\widetilde \\Omega ( ( d / ( \\alpha ^ { 2 } \\log ( 1 / \\alpha ) ) ) + ( d ^ { 3 / 2 } / ( \\varepsilon \\alpha \\log ( 1 / \\alpha ) ) ) +"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1097,
+                    813,
+                    1123,
+                    813,
+                    1123,
+                    843,
+                    1097,
+                    843
+                ],
+                "score": 0.86,
+                "latex": "\\tau _ { j }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1349,
+                    578,
+                    1400,
+                    578,
+                    1400,
+                    607,
+                    1349,
+                    607
+                ],
+                "score": 0.86,
+                "latex": "d ^ { 3 / 2 }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    348,
+                    234,
+                    389,
+                    234,
+                    389,
+                    266,
+                    348,
+                    266
+                ],
+                "score": 0.72,
+                "latex": "( i i )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    296,
+                    204,
+                    328,
+                    204,
+                    328,
+                    236,
+                    296,
+                    236
+                ],
+                "score": 0.68,
+                "latex": "( i )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    540,
+                    578,
+                    651,
+                    578,
+                    651,
+                    613,
+                    540,
+                    613
+                ],
+                "score": 0.25,
+                "latex": "\\log ( 1 / \\delta ) )"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1417.0,
+                    460.0,
+                    1417.0,
+                    460.0,
+                    1461.0,
+                    295.0,
+                    1461.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1153.0,
+                    568.0,
+                    1153.0,
+                    568.0,
+                    1200.0,
+                    295.0,
+                    1200.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    832.0,
+                    2057.0,
+                    871.0,
+                    2057.0,
+                    871.0,
+                    2096.0,
+                    832.0,
+                    2096.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1153.0,
+                    567.0,
+                    1153.0,
+                    567.0,
+                    1201.0,
+                    295.0,
+                    1201.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    539.0,
+                    1141.0,
+                    539.0,
+                    1141.0,
+                    584.0,
+                    291.0,
+                    584.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    290.0,
+                    572.0,
+                    295.0,
+                    572.0,
+                    295.0,
+                    616.0,
+                    290.0,
+                    616.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    654.0,
+                    572.0,
+                    1348.0,
+                    572.0,
+                    1348.0,
+                    616.0,
+                    654.0,
+                    616.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1401.0,
+                    572.0,
+                    1409.0,
+                    572.0,
+                    1409.0,
+                    616.0,
+                    1401.0,
+                    616.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    608.0,
+                    1406.0,
+                    608.0,
+                    1406.0,
+                    646.0,
+                    292.0,
+                    646.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    640.0,
+                    1063.0,
+                    640.0,
+                    1063.0,
+                    675.0,
+                    294.0,
+                    675.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1167.0,
+                    640.0,
+                    1404.0,
+                    640.0,
+                    1404.0,
+                    675.0,
+                    1167.0,
+                    675.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    672.0,
+                    859.0,
+                    672.0,
+                    859.0,
+                    714.0,
+                    292.0,
+                    714.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    470.0,
+                    706.0,
+                    1407.0,
+                    706.0,
+                    1407.0,
+                    744.0,
+                    470.0,
+                    744.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    735.0,
+                    1407.0,
+                    735.0,
+                    1407.0,
+                    774.0,
+                    292.0,
+                    774.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    768.0,
+                    1036.0,
+                    768.0,
+                    1036.0,
+                    811.0,
+                    291.0,
+                    811.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1191.0,
+                    768.0,
+                    1406.0,
+                    768.0,
+                    1406.0,
+                    811.0,
+                    1191.0,
+                    811.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    809.0,
+                    418.0,
+                    809.0,
+                    418.0,
+                    843.0,
+                    294.0,
+                    843.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    497.0,
+                    809.0,
+                    1096.0,
+                    809.0,
+                    1096.0,
+                    843.0,
+                    497.0,
+                    843.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1124.0,
+                    809.0,
+                    1403.0,
+                    809.0,
+                    1403.0,
+                    843.0,
+                    1124.0,
+                    843.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    840.0,
+                    1404.0,
+                    840.0,
+                    1404.0,
+                    874.0,
+                    295.0,
+                    874.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    872.0,
+                    1404.0,
+                    872.0,
+                    1404.0,
+                    906.0,
+                    292.0,
+                    906.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    899.0,
+                    1406.0,
+                    899.0,
+                    1406.0,
+                    938.0,
+                    292.0,
+                    938.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    932.0,
+                    1407.0,
+                    932.0,
+                    1407.0,
+                    966.0,
+                    295.0,
+                    966.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    961.0,
+                    1406.0,
+                    961.0,
+                    1406.0,
+                    995.0,
+                    294.0,
+                    995.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    992.0,
+                    1406.0,
+                    992.0,
+                    1406.0,
+                    1026.0,
+                    295.0,
+                    1026.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1022.0,
+                    1404.0,
+                    1022.0,
+                    1404.0,
+                    1056.0,
+                    294.0,
+                    1056.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1054.0,
+                    1406.0,
+                    1054.0,
+                    1406.0,
+                    1088.0,
+                    295.0,
+                    1088.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1084.0,
+                    814.0,
+                    1084.0,
+                    814.0,
+                    1119.0,
+                    296.0,
+                    1119.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    329.0,
+                    203.0,
+                    1404.0,
+                    203.0,
+                    1404.0,
+                    239.0,
+                    329.0,
+                    239.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    232.0,
+                    347.0,
+                    232.0,
+                    347.0,
+                    268.0,
+                    293.0,
+                    268.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    390.0,
+                    232.0,
+                    1407.0,
+                    232.0,
+                    1407.0,
+                    268.0,
+                    390.0,
+                    268.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    263.0,
+                    1407.0,
+                    263.0,
+                    1407.0,
+                    297.0,
+                    293.0,
+                    297.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    296.0,
+                    1405.0,
+                    296.0,
+                    1405.0,
+                    328.0,
+                    293.0,
+                    328.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    323.0,
+                    1406.0,
+                    323.0,
+                    1406.0,
+                    360.0,
+                    292.0,
+                    360.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    353.0,
+                    1189.0,
+                    353.0,
+                    1189.0,
+                    393.0,
+                    292.0,
+                    393.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1243.0,
+                    353.0,
+                    1324.0,
+                    353.0,
+                    1324.0,
+                    393.0,
+                    1243.0,
+                    393.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    403.0,
+                    1405.0,
+                    403.0,
+                    1405.0,
+                    439.0,
+                    293.0,
+                    439.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    433.0,
+                    1407.0,
+                    433.0,
+                    1407.0,
+                    469.0,
+                    294.0,
+                    469.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    465.0,
+                    1404.0,
+                    465.0,
+                    1404.0,
+                    498.0,
+                    295.0,
+                    498.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    494.0,
+                    1087.0,
+                    494.0,
+                    1087.0,
+                    530.0,
+                    294.0,
+                    530.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1223.0,
+                    1409.0,
+                    1223.0,
+                    1409.0,
+                    1260.0,
+                    295.0,
+                    1260.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1256.0,
+                    1403.0,
+                    1256.0,
+                    1403.0,
+                    1290.0,
+                    295.0,
+                    1290.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1286.0,
+                    1405.0,
+                    1286.0,
+                    1405.0,
+                    1319.0,
+                    295.0,
+                    1319.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1312.0,
+                    1406.0,
+                    1312.0,
+                    1406.0,
+                    1351.0,
+                    293.0,
+                    1351.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1342.0,
+                    443.0,
+                    1342.0,
+                    443.0,
+                    1385.0,
+                    293.0,
+                    1385.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    308.0,
+                    1473.0,
+                    1407.0,
+                    1473.0,
+                    1407.0,
+                    1512.0,
+                    308.0,
+                    1512.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    354.0,
+                    1504.0,
+                    1407.0,
+                    1504.0,
+                    1407.0,
+                    1543.0,
+                    354.0,
+                    1543.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    355.0,
+                    1534.0,
+                    1241.0,
+                    1534.0,
+                    1241.0,
+                    1573.0,
+                    355.0,
+                    1573.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    307.0,
+                    1584.0,
+                    1406.0,
+                    1584.0,
+                    1406.0,
+                    1626.0,
+                    307.0,
+                    1626.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    352.0,
+                    1613.0,
+                    1406.0,
+                    1613.0,
+                    1406.0,
+                    1657.0,
+                    352.0,
+                    1657.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    355.0,
+                    1648.0,
+                    564.0,
+                    1648.0,
+                    564.0,
+                    1681.0,
+                    355.0,
+                    1681.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    308.0,
+                    1698.0,
+                    1404.0,
+                    1698.0,
+                    1404.0,
+                    1737.0,
+                    308.0,
+                    1737.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    353.0,
+                    1730.0,
+                    1358.0,
+                    1730.0,
+                    1358.0,
+                    1765.0,
+                    353.0,
+                    1765.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    308.0,
+                    1779.0,
+                    1406.0,
+                    1779.0,
+                    1406.0,
+                    1817.0,
+                    308.0,
+                    1817.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    353.0,
+                    1812.0,
+                    1405.0,
+                    1812.0,
+                    1405.0,
+                    1848.0,
+                    353.0,
+                    1848.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    351.0,
+                    1841.0,
+                    1080.0,
+                    1841.0,
+                    1080.0,
+                    1880.0,
+                    351.0,
+                    1880.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    308.0,
+                    1893.0,
+                    1406.0,
+                    1893.0,
+                    1406.0,
+                    1932.0,
+                    308.0,
+                    1932.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    352.0,
+                    1923.0,
+                    1409.0,
+                    1923.0,
+                    1409.0,
+                    1960.0,
+                    352.0,
+                    1960.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    308.0,
+                    1975.0,
+                    1230.0,
+                    1975.0,
+                    1230.0,
+                    2011.0,
+                    308.0,
+                    2011.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 9,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 2,
+                "poly": [
+                    835,
+                    2061,
+                    863,
+                    2061,
+                    863,
+                    2086,
+                    835,
+                    2086
+                ],
+                "score": 0.813
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    292,
+                    125,
+                    1410,
+                    125,
+                    1410,
+                    2022,
+                    292,
+                    2022
+                ],
+                "score": 0.475
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    835,
+                    2061,
+                    863,
+                    2061,
+                    863,
+                    2086,
+                    835,
+                    2086
+                ],
+                "score": 0.098
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    357,
+                    1783,
+                    384,
+                    1783,
+                    384,
+                    1813,
+                    357,
+                    1813
+                ],
+                "score": 0.86,
+                "latex": "\\ell _ { 1 }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    600,
+                    1786,
+                    627,
+                    1786,
+                    627,
+                    1810,
+                    600,
+                    1810
+                ],
+                "score": 0.72,
+                "latex": "m"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    832.0,
+                    2058.0,
+                    868.0,
+                    2058.0,
+                    868.0,
+                    2098.0,
+                    832.0,
+                    2098.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    832.0,
+                    2058.0,
+                    868.0,
+                    2058.0,
+                    868.0,
+                    2098.0,
+                    832.0,
+                    2098.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    308.0,
+                    202.0,
+                    1409.0,
+                    202.0,
+                    1409.0,
+                    241.0,
+                    308.0,
+                    241.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    350.0,
+                    233.0,
+                    809.0,
+                    233.0,
+                    809.0,
+                    270.0,
+                    350.0,
+                    270.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    305.0,
+                    283.0,
+                    1407.0,
+                    283.0,
+                    1407.0,
+                    323.0,
+                    305.0,
+                    323.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    352.0,
+                    314.0,
+                    1409.0,
+                    314.0,
+                    1409.0,
+                    354.0,
+                    352.0,
+                    354.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    349.0,
+                    346.0,
+                    609.0,
+                    346.0,
+                    609.0,
+                    380.0,
+                    349.0,
+                    380.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    305.0,
+                    395.0,
+                    1407.0,
+                    395.0,
+                    1407.0,
+                    435.0,
+                    305.0,
+                    435.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    352.0,
+                    427.0,
+                    969.0,
+                    427.0,
+                    969.0,
+                    466.0,
+                    352.0,
+                    466.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    481.0,
+                    1405.0,
+                    481.0,
+                    1405.0,
+                    514.0,
+                    297.0,
+                    514.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    354.0,
+                    510.0,
+                    1405.0,
+                    510.0,
+                    1405.0,
+                    543.0,
+                    354.0,
+                    543.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    354.0,
+                    541.0,
+                    927.0,
+                    541.0,
+                    927.0,
+                    574.0,
+                    354.0,
+                    574.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    589.0,
+                    1407.0,
+                    589.0,
+                    1407.0,
+                    628.0,
+                    293.0,
+                    628.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    350.0,
+                    618.0,
+                    1200.0,
+                    618.0,
+                    1200.0,
+                    657.0,
+                    350.0,
+                    657.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    670.0,
+                    1407.0,
+                    670.0,
+                    1407.0,
+                    709.0,
+                    293.0,
+                    709.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    352.0,
+                    701.0,
+                    1179.0,
+                    701.0,
+                    1179.0,
+                    739.0,
+                    352.0,
+                    739.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    288.0,
+                    749.0,
+                    1409.0,
+                    749.0,
+                    1409.0,
+                    793.0,
+                    288.0,
+                    793.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    354.0,
+                    782.0,
+                    1407.0,
+                    782.0,
+                    1407.0,
+                    822.0,
+                    354.0,
+                    822.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    350.0,
+                    814.0,
+                    997.0,
+                    814.0,
+                    997.0,
+                    851.0,
+                    350.0,
+                    851.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    868.0,
+                    1407.0,
+                    868.0,
+                    1407.0,
+                    901.0,
+                    295.0,
+                    901.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    352.0,
+                    895.0,
+                    1230.0,
+                    895.0,
+                    1230.0,
+                    934.0,
+                    352.0,
+                    934.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    947.0,
+                    1407.0,
+                    947.0,
+                    1407.0,
+                    986.0,
+                    293.0,
+                    986.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    352.0,
+                    978.0,
+                    1369.0,
+                    978.0,
+                    1369.0,
+                    1017.0,
+                    352.0,
+                    1017.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1030.0,
+                    1407.0,
+                    1030.0,
+                    1407.0,
+                    1063.0,
+                    295.0,
+                    1063.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    352.0,
+                    1059.0,
+                    1409.0,
+                    1059.0,
+                    1409.0,
+                    1098.0,
+                    352.0,
+                    1098.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    352.0,
+                    1090.0,
+                    641.0,
+                    1090.0,
+                    641.0,
+                    1123.0,
+                    352.0,
+                    1123.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1142.0,
+                    1405.0,
+                    1142.0,
+                    1405.0,
+                    1175.0,
+                    295.0,
+                    1175.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    350.0,
+                    1167.0,
+                    1407.0,
+                    1167.0,
+                    1407.0,
+                    1211.0,
+                    350.0,
+                    1211.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    350.0,
+                    1200.0,
+                    513.0,
+                    1200.0,
+                    513.0,
+                    1240.0,
+                    350.0,
+                    1240.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1252.0,
+                    1407.0,
+                    1252.0,
+                    1407.0,
+                    1292.0,
+                    293.0,
+                    1292.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    354.0,
+                    1282.0,
+                    1409.0,
+                    1282.0,
+                    1409.0,
+                    1321.0,
+                    354.0,
+                    1321.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    354.0,
+                    1315.0,
+                    633.0,
+                    1315.0,
+                    633.0,
+                    1348.0,
+                    354.0,
+                    1348.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    1367.0,
+                    1405.0,
+                    1367.0,
+                    1405.0,
+                    1400.0,
+                    297.0,
+                    1400.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    354.0,
+                    1394.0,
+                    1306.0,
+                    1394.0,
+                    1306.0,
+                    1433.0,
+                    354.0,
+                    1433.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1446.0,
+                    1407.0,
+                    1446.0,
+                    1407.0,
+                    1485.0,
+                    295.0,
+                    1485.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    356.0,
+                    1477.0,
+                    1403.0,
+                    1477.0,
+                    1403.0,
+                    1510.0,
+                    356.0,
+                    1510.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    352.0,
+                    1506.0,
+                    853.0,
+                    1506.0,
+                    853.0,
+                    1546.0,
+                    352.0,
+                    1546.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1560.0,
+                    1407.0,
+                    1560.0,
+                    1407.0,
+                    1594.0,
+                    295.0,
+                    1594.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    354.0,
+                    1587.0,
+                    1409.0,
+                    1587.0,
+                    1409.0,
+                    1627.0,
+                    354.0,
+                    1627.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    353.0,
+                    1614.0,
+                    522.0,
+                    1614.0,
+                    522.0,
+                    1658.0,
+                    353.0,
+                    1658.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    1673.0,
+                    1405.0,
+                    1673.0,
+                    1405.0,
+                    1706.0,
+                    297.0,
+                    1706.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    354.0,
+                    1702.0,
+                    1048.0,
+                    1702.0,
+                    1048.0,
+                    1735.0,
+                    354.0,
+                    1735.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1747.0,
+                    1409.0,
+                    1747.0,
+                    1409.0,
+                    1791.0,
+                    293.0,
+                    1791.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    385.0,
+                    1781.0,
+                    599.0,
+                    1781.0,
+                    599.0,
+                    1820.0,
+                    385.0,
+                    1820.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    628.0,
+                    1781.0,
+                    1409.0,
+                    1781.0,
+                    1409.0,
+                    1820.0,
+                    628.0,
+                    1820.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    350.0,
+                    1812.0,
+                    666.0,
+                    1812.0,
+                    666.0,
+                    1847.0,
+                    350.0,
+                    1847.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1862.0,
+                    1405.0,
+                    1862.0,
+                    1405.0,
+                    1901.0,
+                    295.0,
+                    1901.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    354.0,
+                    1895.0,
+                    942.0,
+                    1895.0,
+                    942.0,
+                    1928.0,
+                    354.0,
+                    1928.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    1947.0,
+                    1405.0,
+                    1947.0,
+                    1405.0,
+                    1980.0,
+                    297.0,
+                    1980.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    354.0,
+                    1976.0,
+                    644.0,
+                    1976.0,
+                    644.0,
+                    2010.0,
+                    354.0,
+                    2010.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 10,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 2,
+                "poly": [
+                    836,
+                    2061,
+                    865,
+                    2061,
+                    865,
+                    2086,
+                    836,
+                    2086
+                ],
+                "score": 0.734
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    296,
+                    1510,
+                    1403,
+                    1510,
+                    1403,
+                    1576,
+                    296,
+                    1576
+                ],
+                "score": 0.581
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    292,
+                    202,
+                    1404,
+                    202,
+                    1404,
+                    266,
+                    292,
+                    266
+                ],
+                "score": 0.572
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    296,
+                    1396,
+                    1403,
+                    1396,
+                    1403,
+                    1491,
+                    296,
+                    1491
+                ],
+                "score": 0.555
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    296,
+                    1595,
+                    1404,
+                    1595,
+                    1404,
+                    1690,
+                    296,
+                    1690
+                ],
+                "score": 0.54
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    300,
+                    939,
+                    1400,
+                    939,
+                    1400,
+                    1036,
+                    300,
+                    1036
+                ],
+                "score": 0.536
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1053,
+                    1402,
+                    1053,
+                    1402,
+                    1150,
+                    298,
+                    1150
+                ],
+                "score": 0.526
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    295,
+                    598,
+                    1403,
+                    598,
+                    1403,
+                    692,
+                    295,
+                    692
+                ],
+                "score": 0.501
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    296,
+                    1283,
+                    1400,
+                    1283,
+                    1400,
+                    1377,
+                    296,
+                    1377
+                ],
+                "score": 0.499
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    855,
+                    1401,
+                    855,
+                    1401,
+                    922,
+                    297,
+                    922
+                ],
+                "score": 0.493
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    294,
+                    285,
+                    1402,
+                    285,
+                    1402,
+                    351,
+                    294,
+                    351
+                ],
+                "score": 0.482
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1167,
+                    1401,
+                    1167,
+                    1401,
+                    1263,
+                    298,
+                    1263
+                ],
+                "score": 0.481
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1711,
+                    1401,
+                    1711,
+                    1401,
+                    1808,
+                    297,
+                    1808
+                ],
+                "score": 0.471
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    294,
+                    483,
+                    1402,
+                    483,
+                    1402,
+                    578,
+                    294,
+                    578
+                ],
+                "score": 0.442
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    296,
+                    1944,
+                    1399,
+                    1944,
+                    1399,
+                    2007,
+                    296,
+                    2007
+                ],
+                "score": 0.427
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    369,
+                    1404,
+                    369,
+                    1404,
+                    464,
+                    298,
+                    464
+                ],
+                "score": 0.424
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1828,
+                    1399,
+                    1828,
+                    1399,
+                    1924,
+                    297,
+                    1924
+                ],
+                "score": 0.418
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    711,
+                    1405,
+                    711,
+                    1405,
+                    836,
+                    297,
+                    836
+                ],
+                "score": 0.313
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    836,
+                    2061,
+                    865,
+                    2061,
+                    865,
+                    2086,
+                    836,
+                    2086
+                ],
+                "score": 0.31
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    832.0,
+                    2058.0,
+                    870.0,
+                    2058.0,
+                    870.0,
+                    2097.0,
+                    832.0,
+                    2097.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    832.0,
+                    2058.0,
+                    870.0,
+                    2058.0,
+                    870.0,
+                    2097.0,
+                    832.0,
+                    2097.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1512.0,
+                    1406.0,
+                    1512.0,
+                    1406.0,
+                    1548.0,
+                    296.0,
+                    1548.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    355.0,
+                    1543.0,
+                    1183.0,
+                    1543.0,
+                    1183.0,
+                    1575.0,
+                    355.0,
+                    1575.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    202.0,
+                    1407.0,
+                    202.0,
+                    1407.0,
+                    238.0,
+                    296.0,
+                    238.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    353.0,
+                    233.0,
+                    807.0,
+                    233.0,
+                    807.0,
+                    266.0,
+                    353.0,
+                    266.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1395.0,
+                    1406.0,
+                    1395.0,
+                    1406.0,
+                    1435.0,
+                    294.0,
+                    1435.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    350.0,
+                    1426.0,
+                    1412.0,
+                    1426.0,
+                    1412.0,
+                    1468.0,
+                    350.0,
+                    1468.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    355.0,
+                    1458.0,
+                    535.0,
+                    1458.0,
+                    535.0,
+                    1493.0,
+                    355.0,
+                    1493.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1595.0,
+                    1405.0,
+                    1595.0,
+                    1405.0,
+                    1633.0,
+                    295.0,
+                    1633.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    353.0,
+                    1625.0,
+                    1405.0,
+                    1625.0,
+                    1405.0,
+                    1663.0,
+                    353.0,
+                    1663.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    355.0,
+                    1656.0,
+                    1138.0,
+                    1656.0,
+                    1138.0,
+                    1694.0,
+                    355.0,
+                    1694.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    939.0,
+                    1405.0,
+                    939.0,
+                    1405.0,
+                    976.0,
+                    294.0,
+                    976.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    354.0,
+                    971.0,
+                    1404.0,
+                    971.0,
+                    1404.0,
+                    1006.0,
+                    354.0,
+                    1006.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    351.0,
+                    1001.0,
+                    1390.0,
+                    1001.0,
+                    1390.0,
+                    1039.0,
+                    351.0,
+                    1039.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1054.0,
+                    1407.0,
+                    1054.0,
+                    1407.0,
+                    1089.0,
+                    293.0,
+                    1089.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    355.0,
+                    1087.0,
+                    1406.0,
+                    1087.0,
+                    1406.0,
+                    1121.0,
+                    355.0,
+                    1121.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    356.0,
+                    1115.0,
+                    1054.0,
+                    1115.0,
+                    1054.0,
+                    1153.0,
+                    356.0,
+                    1153.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    597.0,
+                    1406.0,
+                    597.0,
+                    1406.0,
+                    633.0,
+                    294.0,
+                    633.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    353.0,
+                    626.0,
+                    1406.0,
+                    626.0,
+                    1406.0,
+                    668.0,
+                    353.0,
+                    668.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    357.0,
+                    658.0,
+                    721.0,
+                    658.0,
+                    721.0,
+                    691.0,
+                    357.0,
+                    691.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1283.0,
+                    1405.0,
+                    1283.0,
+                    1405.0,
+                    1320.0,
+                    295.0,
+                    1320.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    353.0,
+                    1311.0,
+                    1405.0,
+                    1311.0,
+                    1405.0,
+                    1353.0,
+                    353.0,
+                    1353.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    355.0,
+                    1345.0,
+                    735.0,
+                    1345.0,
+                    735.0,
+                    1378.0,
+                    355.0,
+                    1378.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    853.0,
+                    1406.0,
+                    853.0,
+                    1406.0,
+                    895.0,
+                    293.0,
+                    895.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    354.0,
+                    889.0,
+                    919.0,
+                    889.0,
+                    919.0,
+                    922.0,
+                    354.0,
+                    922.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    283.0,
+                    1405.0,
+                    283.0,
+                    1405.0,
+                    326.0,
+                    294.0,
+                    326.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    353.0,
+                    318.0,
+                    1225.0,
+                    318.0,
+                    1225.0,
+                    352.0,
+                    353.0,
+                    352.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1166.0,
+                    1406.0,
+                    1166.0,
+                    1406.0,
+                    1206.0,
+                    292.0,
+                    1206.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    355.0,
+                    1202.0,
+                    1405.0,
+                    1202.0,
+                    1405.0,
+                    1236.0,
+                    355.0,
+                    1236.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    356.0,
+                    1231.0,
+                    954.0,
+                    1231.0,
+                    954.0,
+                    1265.0,
+                    356.0,
+                    1265.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1709.0,
+                    1407.0,
+                    1709.0,
+                    1407.0,
+                    1751.0,
+                    295.0,
+                    1751.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    354.0,
+                    1742.0,
+                    1406.0,
+                    1742.0,
+                    1406.0,
+                    1780.0,
+                    354.0,
+                    1780.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    356.0,
+                    1774.0,
+                    1085.0,
+                    1774.0,
+                    1085.0,
+                    1809.0,
+                    356.0,
+                    1809.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    482.0,
+                    1407.0,
+                    482.0,
+                    1407.0,
+                    520.0,
+                    296.0,
+                    520.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    355.0,
+                    516.0,
+                    1405.0,
+                    516.0,
+                    1405.0,
+                    550.0,
+                    355.0,
+                    550.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    353.0,
+                    544.0,
+                    854.0,
+                    544.0,
+                    854.0,
+                    582.0,
+                    353.0,
+                    582.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1941.0,
+                    1403.0,
+                    1941.0,
+                    1403.0,
+                    1981.0,
+                    295.0,
+                    1981.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    351.0,
+                    1969.0,
+                    430.0,
+                    1969.0,
+                    430.0,
+                    2010.0,
+                    351.0,
+                    2010.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    369.0,
+                    1407.0,
+                    369.0,
+                    1407.0,
+                    406.0,
+                    294.0,
+                    406.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    355.0,
+                    402.0,
+                    1405.0,
+                    402.0,
+                    1405.0,
+                    436.0,
+                    355.0,
+                    436.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    355.0,
+                    432.0,
+                    777.0,
+                    432.0,
+                    777.0,
+                    465.0,
+                    355.0,
+                    465.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1826.0,
+                    1405.0,
+                    1826.0,
+                    1405.0,
+                    1870.0,
+                    293.0,
+                    1870.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    353.0,
+                    1859.0,
+                    1406.0,
+                    1859.0,
+                    1406.0,
+                    1900.0,
+                    353.0,
+                    1900.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    353.0,
+                    1891.0,
+                    760.0,
+                    1891.0,
+                    760.0,
+                    1925.0,
+                    353.0,
+                    1925.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    710.0,
+                    1408.0,
+                    710.0,
+                    1408.0,
+                    749.0,
+                    292.0,
+                    749.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    354.0,
+                    743.0,
+                    1406.0,
+                    743.0,
+                    1406.0,
+                    779.0,
+                    354.0,
+                    779.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    353.0,
+                    772.0,
+                    1410.0,
+                    772.0,
+                    1410.0,
+                    809.0,
+                    353.0,
+                    809.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    351.0,
+                    802.0,
+                    431.0,
+                    802.0,
+                    431.0,
+                    838.0,
+                    351.0,
+                    838.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 11,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 2,
+                "poly": [
+                    836,
+                    2061,
+                    865,
+                    2061,
+                    865,
+                    2086,
+                    836,
+                    2086
+                ],
+                "score": 0.776
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    291,
+                    284,
+                    1410,
+                    284,
+                    1410,
+                    2020,
+                    291,
+                    2020
+                ],
+                "score": 0.562
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    294,
+                    202,
+                    1406,
+                    202,
+                    1406,
+                    267,
+                    294,
+                    267
+                ],
+                "score": 0.386
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    836,
+                    2061,
+                    865,
+                    2061,
+                    865,
+                    2086,
+                    836,
+                    2086
+                ],
+                "score": 0.213
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    832.0,
+                    2058.0,
+                    870.0,
+                    2058.0,
+                    870.0,
+                    2096.0,
+                    832.0,
+                    2096.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    832.0,
+                    2058.0,
+                    870.0,
+                    2058.0,
+                    870.0,
+                    2096.0,
+                    832.0,
+                    2096.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    284.0,
+                    1407.0,
+                    284.0,
+                    1407.0,
+                    324.0,
+                    294.0,
+                    324.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    353.0,
+                    316.0,
+                    746.0,
+                    316.0,
+                    746.0,
+                    353.0,
+                    353.0,
+                    353.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    370.0,
+                    1407.0,
+                    370.0,
+                    1407.0,
+                    406.0,
+                    296.0,
+                    406.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    350.0,
+                    397.0,
+                    1407.0,
+                    397.0,
+                    1407.0,
+                    441.0,
+                    350.0,
+                    441.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    355.0,
+                    431.0,
+                    565.0,
+                    431.0,
+                    565.0,
+                    467.0,
+                    355.0,
+                    467.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    481.0,
+                    1407.0,
+                    481.0,
+                    1407.0,
+                    523.0,
+                    292.0,
+                    523.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    349.0,
+                    513.0,
+                    928.0,
+                    513.0,
+                    928.0,
+                    549.0,
+                    349.0,
+                    549.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    567.0,
+                    1407.0,
+                    567.0,
+                    1407.0,
+                    603.0,
+                    296.0,
+                    603.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    353.0,
+                    597.0,
+                    717.0,
+                    597.0,
+                    717.0,
+                    634.0,
+                    353.0,
+                    634.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    649.0,
+                    1407.0,
+                    649.0,
+                    1407.0,
+                    691.0,
+                    294.0,
+                    691.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    350.0,
+                    678.0,
+                    1409.0,
+                    678.0,
+                    1409.0,
+                    722.0,
+                    350.0,
+                    722.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    351.0,
+                    710.0,
+                    471.0,
+                    710.0,
+                    471.0,
+                    747.0,
+                    351.0,
+                    747.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    760.0,
+                    1407.0,
+                    760.0,
+                    1407.0,
+                    804.0,
+                    291.0,
+                    804.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    353.0,
+                    794.0,
+                    645.0,
+                    794.0,
+                    645.0,
+                    831.0,
+                    353.0,
+                    831.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    844.0,
+                    1410.0,
+                    844.0,
+                    1410.0,
+                    884.0,
+                    292.0,
+                    884.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    350.0,
+                    877.0,
+                    1409.0,
+                    877.0,
+                    1409.0,
+                    919.0,
+                    350.0,
+                    919.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    350.0,
+                    905.0,
+                    1410.0,
+                    905.0,
+                    1410.0,
+                    947.0,
+                    350.0,
+                    947.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    351.0,
+                    936.0,
+                    428.0,
+                    936.0,
+                    428.0,
+                    976.0,
+                    351.0,
+                    976.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    991.0,
+                    1405.0,
+                    991.0,
+                    1405.0,
+                    1028.0,
+                    296.0,
+                    1028.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    351.0,
+                    1022.0,
+                    1285.0,
+                    1022.0,
+                    1285.0,
+                    1060.0,
+                    351.0,
+                    1060.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1076.0,
+                    1405.0,
+                    1076.0,
+                    1405.0,
+                    1112.0,
+                    296.0,
+                    1112.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    355.0,
+                    1106.0,
+                    1330.0,
+                    1106.0,
+                    1330.0,
+                    1142.0,
+                    355.0,
+                    1142.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1156.0,
+                    1407.0,
+                    1156.0,
+                    1407.0,
+                    1198.0,
+                    292.0,
+                    1198.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    353.0,
+                    1188.0,
+                    1407.0,
+                    1188.0,
+                    1407.0,
+                    1225.0,
+                    353.0,
+                    1225.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    351.0,
+                    1217.0,
+                    974.0,
+                    1217.0,
+                    974.0,
+                    1259.0,
+                    351.0,
+                    1259.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1269.0,
+                    1407.0,
+                    1269.0,
+                    1407.0,
+                    1309.0,
+                    292.0,
+                    1309.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    355.0,
+                    1301.0,
+                    963.0,
+                    1301.0,
+                    963.0,
+                    1338.0,
+                    355.0,
+                    1338.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1353.0,
+                    1407.0,
+                    1353.0,
+                    1407.0,
+                    1395.0,
+                    294.0,
+                    1395.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    355.0,
+                    1385.0,
+                    1407.0,
+                    1385.0,
+                    1407.0,
+                    1422.0,
+                    355.0,
+                    1422.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    353.0,
+                    1416.0,
+                    515.0,
+                    1416.0,
+                    515.0,
+                    1452.0,
+                    353.0,
+                    1452.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1466.0,
+                    1409.0,
+                    1466.0,
+                    1409.0,
+                    1508.0,
+                    294.0,
+                    1508.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    348.0,
+                    1496.0,
+                    1231.0,
+                    1496.0,
+                    1231.0,
+                    1536.0,
+                    348.0,
+                    1536.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1550.0,
+                    1407.0,
+                    1550.0,
+                    1407.0,
+                    1590.0,
+                    292.0,
+                    1590.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    349.0,
+                    1580.0,
+                    1088.0,
+                    1580.0,
+                    1088.0,
+                    1619.0,
+                    349.0,
+                    1619.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1632.0,
+                    1407.0,
+                    1632.0,
+                    1407.0,
+                    1672.0,
+                    292.0,
+                    1672.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    348.0,
+                    1665.0,
+                    742.0,
+                    1665.0,
+                    742.0,
+                    1703.0,
+                    348.0,
+                    1703.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1718.0,
+                    1407.0,
+                    1718.0,
+                    1407.0,
+                    1754.0,
+                    296.0,
+                    1754.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    351.0,
+                    1749.0,
+                    1169.0,
+                    1749.0,
+                    1169.0,
+                    1785.0,
+                    351.0,
+                    1785.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1798.0,
+                    1407.0,
+                    1798.0,
+                    1407.0,
+                    1840.0,
+                    294.0,
+                    1840.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    353.0,
+                    1833.0,
+                    1409.0,
+                    1833.0,
+                    1409.0,
+                    1869.0,
+                    353.0,
+                    1869.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    355.0,
+                    1862.0,
+                    471.0,
+                    1862.0,
+                    471.0,
+                    1898.0,
+                    355.0,
+                    1898.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1915.0,
+                    1407.0,
+                    1915.0,
+                    1407.0,
+                    1951.0,
+                    296.0,
+                    1951.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    350.0,
+                    1944.0,
+                    1407.0,
+                    1944.0,
+                    1407.0,
+                    1984.0,
+                    350.0,
+                    1984.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    351.0,
+                    1974.0,
+                    1054.0,
+                    1974.0,
+                    1054.0,
+                    2016.0,
+                    351.0,
+                    2016.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    200.0,
+                    1407.0,
+                    200.0,
+                    1407.0,
+                    241.0,
+                    296.0,
+                    241.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    354.0,
+                    233.0,
+                    1336.0,
+                    233.0,
+                    1336.0,
+                    267.0,
+                    354.0,
+                    267.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 12,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 2,
+                "poly": [
+                    836,
+                    2061,
+                    865,
+                    2061,
+                    865,
+                    2086,
+                    836,
+                    2086
+                ],
+                "score": 0.812
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    292,
+                    161,
+                    1409,
+                    161,
+                    1409,
+                    2013,
+                    292,
+                    2013
+                ],
+                "score": 0.543
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    832.0,
+                    2058.0,
+                    871.0,
+                    2058.0,
+                    871.0,
+                    2096.0,
+                    832.0,
+                    2096.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    203.0,
+                    1411.0,
+                    203.0,
+                    1411.0,
+                    241.0,
+                    296.0,
+                    241.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    348.0,
+                    227.0,
+                    1409.0,
+                    227.0,
+                    1409.0,
+                    276.0,
+                    348.0,
+                    276.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    356.0,
+                    266.0,
+                    610.0,
+                    266.0,
+                    610.0,
+                    298.0,
+                    356.0,
+                    298.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    318.0,
+                    1409.0,
+                    318.0,
+                    1409.0,
+                    357.0,
+                    296.0,
+                    357.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    352.0,
+                    349.0,
+                    1041.0,
+                    349.0,
+                    1041.0,
+                    388.0,
+                    352.0,
+                    388.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    406.0,
+                    1407.0,
+                    406.0,
+                    1407.0,
+                    438.0,
+                    296.0,
+                    438.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    354.0,
+                    434.0,
+                    1177.0,
+                    434.0,
+                    1177.0,
+                    473.0,
+                    354.0,
+                    473.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    487.0,
+                    1407.0,
+                    487.0,
+                    1407.0,
+                    526.0,
+                    294.0,
+                    526.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    354.0,
+                    520.0,
+                    1013.0,
+                    520.0,
+                    1013.0,
+                    558.0,
+                    354.0,
+                    558.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    573.0,
+                    1409.0,
+                    573.0,
+                    1409.0,
+                    611.0,
+                    294.0,
+                    611.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    354.0,
+                    603.0,
+                    500.0,
+                    603.0,
+                    500.0,
+                    642.0,
+                    354.0,
+                    642.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    658.0,
+                    1407.0,
+                    658.0,
+                    1407.0,
+                    697.0,
+                    294.0,
+                    697.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    352.0,
+                    688.0,
+                    1411.0,
+                    688.0,
+                    1411.0,
+                    727.0,
+                    352.0,
+                    727.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    354.0,
+                    719.0,
+                    606.0,
+                    719.0,
+                    606.0,
+                    758.0,
+                    354.0,
+                    758.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    290.0,
+                    770.0,
+                    1409.0,
+                    770.0,
+                    1409.0,
+                    812.0,
+                    290.0,
+                    812.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    354.0,
+                    804.0,
+                    919.0,
+                    804.0,
+                    919.0,
+                    843.0,
+                    354.0,
+                    843.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    857.0,
+                    1407.0,
+                    857.0,
+                    1407.0,
+                    896.0,
+                    296.0,
+                    896.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    354.0,
+                    890.0,
+                    1409.0,
+                    890.0,
+                    1409.0,
+                    928.0,
+                    354.0,
+                    928.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    290.0,
+                    939.0,
+                    1409.0,
+                    939.0,
+                    1409.0,
+                    985.0,
+                    290.0,
+                    985.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    354.0,
+                    975.0,
+                    1407.0,
+                    975.0,
+                    1407.0,
+                    1014.0,
+                    354.0,
+                    1014.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    354.0,
+                    1006.0,
+                    1179.0,
+                    1006.0,
+                    1179.0,
+                    1044.0,
+                    354.0,
+                    1044.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1059.0,
+                    1407.0,
+                    1059.0,
+                    1407.0,
+                    1097.0,
+                    296.0,
+                    1097.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    350.0,
+                    1089.0,
+                    1043.0,
+                    1089.0,
+                    1043.0,
+                    1130.0,
+                    350.0,
+                    1130.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1144.0,
+                    1407.0,
+                    1144.0,
+                    1407.0,
+                    1183.0,
+                    294.0,
+                    1183.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    354.0,
+                    1174.0,
+                    1409.0,
+                    1174.0,
+                    1409.0,
+                    1213.0,
+                    354.0,
+                    1213.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    352.0,
+                    1203.0,
+                    430.0,
+                    1203.0,
+                    430.0,
+                    1244.0,
+                    352.0,
+                    1244.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    290.0,
+                    1256.0,
+                    1409.0,
+                    1256.0,
+                    1409.0,
+                    1303.0,
+                    290.0,
+                    1303.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    350.0,
+                    1286.0,
+                    1409.0,
+                    1286.0,
+                    1409.0,
+                    1331.0,
+                    350.0,
+                    1331.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    354.0,
+                    1319.0,
+                    1293.0,
+                    1319.0,
+                    1293.0,
+                    1357.0,
+                    354.0,
+                    1357.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1374.0,
+                    1409.0,
+                    1374.0,
+                    1409.0,
+                    1412.0,
+                    294.0,
+                    1412.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    350.0,
+                    1400.0,
+                    1409.0,
+                    1400.0,
+                    1409.0,
+                    1447.0,
+                    350.0,
+                    1447.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    354.0,
+                    1435.0,
+                    642.0,
+                    1435.0,
+                    642.0,
+                    1473.0,
+                    354.0,
+                    1473.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1490.0,
+                    1411.0,
+                    1490.0,
+                    1411.0,
+                    1528.0,
+                    294.0,
+                    1528.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1545.0,
+                    1409.0,
+                    1545.0,
+                    1409.0,
+                    1583.0,
+                    294.0,
+                    1583.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    346.0,
+                    1573.0,
+                    896.0,
+                    1573.0,
+                    896.0,
+                    1614.0,
+                    346.0,
+                    1614.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1624.0,
+                    1409.0,
+                    1624.0,
+                    1409.0,
+                    1669.0,
+                    292.0,
+                    1669.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    356.0,
+                    1662.0,
+                    750.0,
+                    1662.0,
+                    750.0,
+                    1695.0,
+                    356.0,
+                    1695.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    290.0,
+                    1711.0,
+                    1409.0,
+                    1711.0,
+                    1409.0,
+                    1754.0,
+                    290.0,
+                    1754.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    352.0,
+                    1742.0,
+                    1407.0,
+                    1742.0,
+                    1407.0,
+                    1786.0,
+                    352.0,
+                    1786.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    350.0,
+                    1772.0,
+                    1289.0,
+                    1772.0,
+                    1289.0,
+                    1815.0,
+                    350.0,
+                    1815.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1829.0,
+                    1409.0,
+                    1829.0,
+                    1409.0,
+                    1868.0,
+                    294.0,
+                    1868.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    352.0,
+                    1862.0,
+                    1407.0,
+                    1862.0,
+                    1407.0,
+                    1900.0,
+                    352.0,
+                    1900.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    350.0,
+                    1888.0,
+                    945.0,
+                    1888.0,
+                    945.0,
+                    1933.0,
+                    350.0,
+                    1933.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1941.0,
+                    1407.0,
+                    1941.0,
+                    1407.0,
+                    1988.0,
+                    292.0,
+                    1988.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    352.0,
+                    1976.0,
+                    1069.0,
+                    1976.0,
+                    1069.0,
+                    2014.0,
+                    352.0,
+                    2014.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 13,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 2,
+                "poly": [
+                    836,
+                    2061,
+                    865,
+                    2061,
+                    865,
+                    2086,
+                    836,
+                    2086
+                ],
+                "score": 0.685
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    836,
+                    2061,
+                    865,
+                    2061,
+                    865,
+                    2086,
+                    836,
+                    2086
+                ],
+                "score": 0.395
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    290,
+                    202,
+                    1408,
+                    202,
+                    1408,
+                    349,
+                    290,
+                    349
+                ],
+                "score": 0.369
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    296,
+                    202,
+                    1408,
+                    202,
+                    1408,
+                    349,
+                    296,
+                    349
+                ],
+                "score": 0.2
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    832.0,
+                    2058.0,
+                    870.0,
+                    2058.0,
+                    870.0,
+                    2096.0,
+                    832.0,
+                    2096.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    832.0,
+                    2058.0,
+                    870.0,
+                    2058.0,
+                    870.0,
+                    2096.0,
+                    832.0,
+                    2096.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    202.0,
+                    1407.0,
+                    202.0,
+                    1407.0,
+                    238.0,
+                    295.0,
+                    238.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    353.0,
+                    233.0,
+                    808.0,
+                    233.0,
+                    808.0,
+                    268.0,
+                    353.0,
+                    268.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    283.0,
+                    1404.0,
+                    283.0,
+                    1404.0,
+                    323.0,
+                    294.0,
+                    323.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    352.0,
+                    318.0,
+                    738.0,
+                    318.0,
+                    738.0,
+                    351.0,
+                    352.0,
+                    351.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    202.0,
+                    1408.0,
+                    202.0,
+                    1408.0,
+                    238.0,
+                    295.0,
+                    238.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    353.0,
+                    233.0,
+                    807.0,
+                    233.0,
+                    807.0,
+                    268.0,
+                    353.0,
+                    268.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    283.0,
+                    1405.0,
+                    283.0,
+                    1405.0,
+                    323.0,
+                    295.0,
+                    323.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    352.0,
+                    318.0,
+                    738.0,
+                    318.0,
+                    738.0,
+                    351.0,
+                    352.0,
+                    351.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 14,
+            "width": 1700,
+            "height": 2200
+        }
+    }
+]
\ No newline at end of file
diff --git a/parse/train/HJgZrsC5t7/HJgZrsC5t7.md b/parse/train/HJgZrsC5t7/HJgZrsC5t7.md
new file mode 100644
index 0000000000000000000000000000000000000000..ab3887b2c6bfd46cfb02398e32badee2e27bb080
--- /dev/null
+++ b/parse/train/HJgZrsC5t7/HJgZrsC5t7.md
@@ -0,0 +1,327 @@
+# IMPROVING ON-POLICY LEARNING WITH STATISTICAL REWARD ACCUMULATION
+
+Anonymous authors Paper under double-blind review
+
+# ABSTRACT
+
+Deep reinforcement learning has obtained significant breakthroughs in recent years. Most methods in deep-RL achieve good results via the maximization of the reward signal provided by the environment, typically in the form of discounted cumulative returns. Such reward signals represent the immediate feedback of a particular action performed by an agent. However, tasks with sparse reward signals are still challenging to on-policy methods. In this paper, we introduce an effective characterization of past reward statistics (which can be seen as long-term feedback signals) to supplement this immediate reward feedback. In particular, value functions are learned with multi-critics supervision, enabling complex value functions to be more easily approximated in on-policy learning, even when the reward signals are sparse. We also introduce a novel exploration mechanism called “hot-wiring” that can give a boost to seemingly trapped agents. We demonstrate the effectiveness of our advantage actor multi-critic (A2MC) method across the discrete domains in Atari games as well as continuous domains in the MuJoCo environments. A video demo is provided at https://youtu.be/zBmpf3Yz8tc and source codes will be made available upon paper acceptance.
+
+# 1 INTRODUCTION
+
+Advances in deep learning have mobilized the research community to adopt deep reinforcement learning (RL) agents for challenging control problems, typically in complex environments with raw sensory state-spaces. Breakthroughs by Mnih et al. (2015) show RL-agents can reach abovehuman performance in Atari 2600 games, and AlphaGo Zero Silver et al. (2017) becomes the world champions on the game of Go. Still, training RL agents is non-trivial. Off-policy methods typically require days of training to obtain competitive performance, while on-policy methods could be trapped in local minima. Recent techniques featuring on-policy learning Mnih et al. (2016); Schulman et al. (2017); Wu et al. (2017) have shown promising results in stabilizing the learning processes, enabling an agent to solve a variety of tasks in much less time. In particular, the state-of-the-art on-policy ACKTR agent by Wu et al. (2017) shows improved sample efficiency with the help of Kronecker-factored (K-Fac) approximate curvature for natural gradient updates, resulting in stable and effective model updates towards a more promising direction.
+
+However, tasks with sparse rewards remain challenging to on-policy methods. An agent could take massive amount of exploration before reaching non-zero rewards; and as the agent learns on-policy, the sparseness of reward feedback (receiving all-zero rewards from most actions performed by the agent) could be malicious and render an agent to falsely predict all states in an environment towards a value of zero. As there does not exist a universal criterion for measuring “task sparseness”, we show an ad-hoc metric in Figure 1 in an attempt to provide intuition. For example, we observe that the ACKTR agent is unable to receive sufficient non-zero immediate rewards that can provide instructive agent updates in Atari games “Freeway” and “Enduro”, resulting in failures when solving these two games. Moreover, if ACKTR gets drawn to and trapped in unfavorable states (as in games like Boxing and WizardOfWor), it could again take long hours of exploration before the agent can get out of the local minima. Such evidence shows that on-policy agent could indeed suffer from the insufficiencies of guidance provided by the exclusive immediate reward signals from the environment.
+
+In this paper, we introduce an effective auxiliary reward signal in tasks with sparse rewards to remedy the deficiencies of learning purely from standard immediate reward feedbacks. As on-policy agents may take many explorations before reaching non-zero immediate rewards, we argue that we can leverage past reward statistics to provide more instructive feedbacks to agents in the same environment. To this end, we propose to characterize the past reward statistics in order to gauge the “long-term” performance of an agent (detailed in Section 4). After performing an action, an agent will receive a long-term reward signal describing its past performance upon this step, as well as the conventional immediate reward from the environment. To effectively characterize the long-term performance of the agent, we take into considerations both the crude amount of rewards and the volatility of rewards received in the past, where highly volatile distributions of long-term rewards are explicitly penalized. This enables complex value functions to be more easily approximated in multi-critics supervision. We find in practice that by explicitly penalizing highly volatile long-term rewards while maximizing the expectation of short-term rewards, the learning process and the overall performance are improved regarding both sample efficiency and final rewards. We further propose a “hot-wiring” exploration mechanism that can boost seemingly trapped agent in the earlier stage of learning. By leveraging the characterization of long/short-term reward statistics, our proposed advantage actor multi-critic model (A2MC) shows significantly improved performance on the Atari 2600 games and the MuJoCo tasks as compared to the state-of-the-art on-policy methods.
+
+![](images/cd753a39ce5a48461924628e1e3afe7243006032133b48f3e336125002681ed3.jpg)  
+Figure 1: Performance of A2MC on Atari games trained with 15 million timesteps. Our method has a winning rate of $5 5 . 3 \%$ among all the Atari games tested, as compared to the ACKTR. Our A2MC learns quickly in some of the hardest games for on-policy methods, such as “Boxing”, “Enduro”, “Freeway”, “Robotank” and “WizardOfWor”. The sparseness of a game is defined as the sparseness of average rewards $\mathbf { x }$ obtained by ACKTR within the first $n = 1 0 ^ { 6 }$ timesteps by $\begin{array} { r } { \varphi ( \mathbf { x } ) ^ { \mathbf { \tilde { \alpha } } } = \left( \sqrt { n } - \frac { \| \mathbf { x } \| _ { 1 } } { \| \mathbf { x } \| _ { 2 } } \right) \tilde { \mathbf { \alpha } } ( \sqrt { n } - 1 ) } \end{array}$ . A higher value of sparseness indicates sparser rewards. A relative performance margin (in terms of final reward) larger than $1 0 \%$ is deemed as winning / losing. The shaded region denotes the standard deviation over 2 random seeds.
+
+# 2 RELATED WORK
+
+Reward shaping and pseudo-rewards: To tackle the challenge in tasks with rarely observed rewards, pseudo-rewards maximization is adopted in earlier works Konidaris & Barto (2009); Silver & Ciosek (2012). Auxiliary vision tasks (e.g., learning pixel changes or network features) are adopted in the off-policy UNREAL agent Jaderberg et al. (2016) in order to facilitate learning better feature representations, particularly for sparse reward environments. Another direction of effort involves directly engineering a better reward function or shaping the reward signal. Andrychowicz et al. (2017) enhances off-policy learning by re-producing informative reward in hindsight for sequences of actions that do not lead to success previously. The HRA approach Van Seijen et al. (2017) exploits domain knowledge to define a set of environment-specific rewards based on reward categories. And the winning approach that learns playing “Doom” Lample & Chaplot (2017) shows promising success in the FPS game that carefully crafting the task rewards would indeed be beneficial. In contrast to heuristically defining vision-related auxiliary tasks, our proposed A2MC agent learns from the characterization of intrinsic past reward statistics obtainable from any environment; and different from the hybrid architecture pertaining to Ms. Pacman only and the reward shaping settings tailored specifically to ”Doom”, our proposed reward characterization mechanism is generic and our A2MC generalizes well to a variety of tasks without the need to engineer a decomposition of problemspecific environment rewards. Moreover, the capability of the proposed method to further boost reward shaping is evidenced in our case study on playing Doom (see Appendix F).
+
+Multi-agents: The multi-agent approaches Lanctot et al. (2017); Lowe et al. (2017); Jin et al. (2018) present another promising direction for learning. They propose to train multiple agents in parallel when solving a task, and to combine multiple action-value functions with a centralized action-value function. The multi-critics supervision in our proposed A2MC model can be seen as a form of joint-task or multi-task learning Teh et al. (2017) for both long-term and short-term rewards.
+
+On-policy v.s. Off-policy: Our empirical results based on learning the characterization of long/shortterm reward statistics also echo the effectiveness of a recently proposed off-policy reinforcement learning framework Bellemare et al. (2017) that features a distributional variant of Q-learning, wherein the value functions are learned to match the distribution of standard immediate returns. Also, Wang et al. (2016) shows that applying experience replay to on-policy methods can further enhance learning stability. Schulman et al. (2016) proposes a variant of advantage function using eligibility traces that provides both low-variance and low-bias gradient estimates. These works are orthogonal to our approach can potentially be combined with the proposed characterization of past reward statistics to further enhance learning performance. While our extensive experiments (see also Appendix E and Appendix F) show promising results of our approach in both on- and off-policy frameworks, we focus on “on-policy” methods (i.e., those that do not involve off-policy trajectories or experience replay) as in $\mathrm { W u }$ et al. (2017) in the main text in order to systematically evaluate the potential of our proposed reward mechanism within the scope of this work.
+
+# 3 PRELIMINARY
+
+Consider the standard reinforcement learning setting where an agent interacts with an environment over a number of discrete time step. At each time step $t$ , the agent receives an environment state $s _ { t }$ , then executes an action $a _ { t }$ based on policy $\pi _ { t }$ . The environment produces reward $r _ { t }$ and next state $s _ { t + 1 }$ , according to which the agent gets feedback of its immediate action and will decide its next action $a _ { t + 1 }$ . The process $< { \bf S } , { \bf A } , { \bf R } , { \bf S } >$ , typically considered as a Markov Decision Process, continues until a terminal state $s _ { T }$ upon which the environment resets itself and produces a new episode. Under conventional settings, the return is calculated as the discounted summation of rewards $r _ { t }$ accumulated from time step $t$ onwards $\begin{array} { r } { R _ { t } = \sum _ { k = 0 } ^ { \infty } \gamma ^ { k } r _ { t + k } } \end{array}$ . The goal of the agent is to maximize the expected return from each state $s _ { t }$ while following policy $\pi$ . Each policy has a corresponding action-value function defined as $Q ^ { \pi } ( s , a ) = \mathbb { E } [ R _ { t } | s _ { t } = s , a _ { t } = a ; \pi ]$ . Similarly, each state $s \in S$ under policy $\pi$ has a value function defined as: $V ^ { \pi } ( s ) = \mathbb { E } [ R _ { t } | s _ { t } = s ]$ . In value-based approaches (e.g., Q-learning Mnih et al. (2015)), function approximator $Q ( s , a ; \theta )$ can be used to approximate the optimal action value function $Q ^ { * } ( s , a )$ . This is conventionally learned by iteratively minimizing the below loss function:
+
+$$
+L ( \theta ) = \mathbb { E } [ ( y _ { t } ^ { t a r g e t } - Q ( s _ { t } , a _ { t } ; \theta ) ) ^ { 2 } ] ,
+$$
+
+where ytargett = rt + γ maxa0 Q(st+1, a0; θ) and st+1 is the next state following state st.
+
+In policy-based approaches (e.g., policy gradient methods), the optimal policy $\pi ^ { * } ( a | s )$ is approximated using the approximator $\pi ( a | s ; \theta )$ . The policy approximator is then learned by gradient ascent on $\nabla _ { \boldsymbol { \theta } } \mathbb { E } [ h _ { t } ] \approx \dot { \nabla _ { \boldsymbol { \theta } } } \log \pi ( a _ { t } | s _ { t } ; \boldsymbol { \theta } ) \dot { R } _ { t }$ . The REINFORCE method Williams (1992) further incorporates a baseline $b ( s _ { t } )$ to reduce the variance of the gradient estimator: $\nabla _ { \theta } \mathbb { E } [ R _ { t } ] _ { R E I N F O R C E } \approx$ $\nabla _ { \theta } \log \pi ( a _ { t } | s _ { t } ; \theta ) \dot { ( R _ { t } - b ( s _ { t } ) ) }$
+
+In actor-critic based approaches, the variance reduction further evolves into the advantage function $A ( s _ { t } , a _ { t } ) = Q ( s _ { t } , a _ { t } ) - V ( s _ { t } )$ in Mnih et al. (2016), where the action value $Q ^ { \pi } ( s _ { t } , a _ { t } )$ is approximated by $R _ { t }$ and $b ( s _ { t } )$ is replaced by $V ^ { \pi } ( s _ { t } )$ , deriving the advantage actor-critic architecture where actor-head $\pi ( \cdot | s )$ and the critic-head $V ( s )$ share low-level features. The gradient update rule w.r.t. the action-head is $\nabla _ { \theta } \log \pi ( a _ { t } | s _ { t } ; \theta ) ( R _ { t } - V ( s _ { t } ; \theta ) )$ . The gradient update w.r.t. the critic-head is: $\nabla _ { \boldsymbol { \theta } } ( R _ { t } - V ( s _ { t } ; \boldsymbol { \theta } ) ) ^ { 2 }$ , where $R _ { t } = r _ { t } + \gamma V ( s _ { t + 1 } )$ .
+
+![](images/691a77d1436be9adc6c816257080ddb1982e633dec02a7c34a27ed2085021986.jpg)  
+Figure 2: Illustration of the proposed variability-weighted reward (VWR). The first row shows the raw reward sequence (blue) while the second row presents the post-processed sequence $\vec { \mathcal { R } }$ (green) and the zero-variability reference $\vec { \mathcal { R } } ^ { z e r o }$ (orange), and $\mathcal { R } _ { H }$ is calculated as a reflection of how high the immediate reward is. The third row shows the volatility statistics of $\delta _ { \mathcal { R } }$ , representing how varied past rewards were. We curated 3 hypothetical reward sequences – (a) highly varied sequence with low immediate reward, resulting in the lowest VWR; (b) highly varied sequence with high immediate reward, leading to a relatively high VWR; (c) stable sequence with high immediate reward, achieving the best VWR. More examples can be found in the Appendix A.
+
+# 4 CHARACTERIZATION OF PAST REWARD STATISTICS
+
+The conventional reward $r _ { t }$ received from the environment at time step $t$ after an action $a _ { t }$ is performed represents the immediate reward regarding this particular action. This “immediacy” could be interpreted as a short-term horizon of how the agent is doing, i.e., evaluating the agent via judging its actions by immediate rewards. We argue that the deficiencies of learning solely from immediate rewards mainly come from this limitation that the agent is learning from one single type of exclusive short-term feedback.
+
+As the goal of providing reward feedback to an agent is to inform the agent of its performance, we seek to find an auxiliary performance metric that can measure whether the agent is performing consistently well. Inspired by the formulation of Sharpe Ratio $\begin{array} { r } { ( \mathbb { E } [ r ] \times \frac { 1 } { \sigma _ { r } } ) } \end{array}$ in evaluating the long-term performance of porfolio strategies where the return $\mathbb { E } [ r ]$ is inversely weighted by the risk $\sigma _ { r }$ , an effective characterization of historical reward statistics should take into account at least two factors, namely 1) how high the immediate reward is and 2) how varied past rewards were, bringing the desired notion of “risk-adjusted return” as in Sharpe (1994).
+
+# 4.1 VARIABILITY-WEIGHTED REWARD
+
+To this end, we follow insights behind Dowd (2000); Sharpe (1994) and define a variability-weighted characterization of past rewards. This is illustrated in Figure 2. We consider a historical sequence of $T$ rewards upon timestep $t$ (looking backward $T - 1$ timesteps): $\vec { \mathbf { r } } = \left[ r _ { t - ( T - 1 ) } . . . , r _ { t - 2 } , r _ { t - 1 } , r _ { t } \right]$ . In order to evaluate how high and varied the reward sequence is, a few steps of pre-processing $\mathcal { G }$ is applied, denoted as $\vec { \mathcal { R } } = \mathcal { G } ( \vec { \bf r } )$ . Specifically, we first derive the reward change at each timestep (similar to the “differential return” concept in Sharpe (1994)) with $d _ { n } = r _ { n } - r _ { n - 1 }$ :
+
+$$
+\vec { \mathbf { d } } = [ d _ { t - ( T - 1 ) } , d _ { t - ( T - 2 ) } , \dots , d _ { t } ] = [ r _ { t - ( T - 1 ) } , r _ { t - ( T - 2 ) } - r _ { t - ( T - 1 ) } , \dots , r _ { t } - r _ { t - 1 } ] .
+$$
+
+Then we re-order the sequence by flipping 1 with $f _ { n } = d _ { t + 1 - n }$ :
+
+$$
+\vec { \mathbf { f } } = [ f _ { 1 } , f _ { 2 } , \dotsc , f _ { T } ] = [ d _ { t } , d _ { t - 1 } , \dotsc , d _ { t - ( T - 1 ) } ] .
+$$
+
+Next we append $f _ { 0 } = 1$ to the head of sequence $\vec { \mathbf { f } }$ and take the normalized cumulative sum to obtain the post-processed reward sequence as $\begin{array} { r } { \vec { \mathcal { R } } = [ \mathcal { R } _ { 0 } , \mathcal { R } _ { 1 } , \ldots , \mathcal { R } _ { T } ] = \frac { 1 } { T + 1 } [ f _ { 0 } , f _ { 0 } + f _ { 1 } , \ldots , \sum _ { i = 0 } ^ { T } f _ { i } ] } \end{array}$ . Under such processing, numerical instability (see Eq. 4) when all rewards in the sequence are zero can be alleviated, while the averaging term $\\frac { 1 } { T + 1 }$ mitigates the effect of introducing the artificial $f _ { 0 }$ .
+
+The resulting $\vec { \mathcal { R } }$ is a reward sequence with $\begin{array} { r } { \mathcal { R } _ { T } - \mathcal { R } _ { 0 } = \frac { 1 } { T + 1 } r _ { t } } \end{array}$ , and $\begin{array} { r } { \mathcal { R } _ { n } - \mathcal { R } _ { n - 1 } = \frac { 1 } { T + 1 } ( r _ { t + 1 - n } - } \end{array}$ $r _ { t - n } )$ . Therefore, the difference between $\mathcal { R } _ { T }$ and $\mathcal { R } _ { 0 }$ represents the immediate reward and the whole sequence $\vec { \mathcal { R } }$ reflects the volatility of past rewards. In Figure 2, three examples of processed sequence are presented in the second row with the corresponding raw rewards shown in the first row. We account for how high the immediate reward is by defining the relative percentage log total return as:
+
+$$
+\mathcal { R } _ { H } = 1 0 0 \times \left( e ^ { \frac { 1 } { T } \ln \frac { \mathcal { R } _ { T } } { \mathcal { R } _ { 0 } } } - 1 \right) = \frac { { \mathcal { R } _ { T } } ^ { 1 / T } - { \mathcal { R } _ { 0 } } ^ { 1 / T } } { { \mathcal { R } _ { 0 } } ^ { 1 / T } } \times 1 0 0 .
+$$
+
+To account for how varied past rewards were, we first define a smooth zero-variability reference as: $\vec { \mathcal { R } } ^ { z e r o } = [ \mathcal { R } _ { 0 } ^ { z e r o } , \mathcal { R } _ { 1 } ^ { z e r o } , \ldots , \mathcal { R } _ { T } ^ { z e r o } ] = \mathcal { R } _ { 0 } [ e ^ { 0 \times \widetilde { \mathcal { R } } } , e ^ { 1 \times \widetilde { \mathcal { R } } } , \ldots , e ^ { T \widetilde { \mathcal { R } } } ]$ with $\begin{array} { r } { \widetilde { \mathcal { R } } = \frac { 1 } { T } \ln \frac { \mathcal { R } _ { T } } { \mathcal { R } _ { 0 } } } \end{array}$ , represent a smooth monotonic reference sequence from $\mathcal { R } _ { 0 }$ to $\mathcal { R } _ { T }$ . Then we define the reward differential $\delta _ { \mathcal { R } }$ as the differential reward versus its zero-variability reference as $\begin{array} { r } { \delta _ { \mathcal { R } } ( n ) = \frac { \mathcal { R } _ { n } - \mathcal { R } _ { n } ^ { z e r o } } { \mathcal { R } _ { n } ^ { z e r o } } } \end{array}$ , whose statistics are sketched in the third row of Figure 2. With maximally allowed volatility as $\sigma _ { m a x }$ , the variability weights can be defined as: $\begin{array} { r } { \omega = 1 - \big [ { \frac { \sigma ( \delta _ { \mathcal { R } } ) } { \sigma _ { m a x } } } \big ] ^ { \tau } } \end{array}$ , where $\sigma ( \cdot )$ is the standard deviation and $\tau$ controls the rate to penalize highly volatile reward distribution. Finally we can derive the variability-weighted past reward indicator $r ^ { v w r }$ for the characterization of past reward statistics:
+
+$$
+r ^ { v w r } = \left. \begin{array} { c c } { \mathcal { R } _ { H } ( 1 - [ \frac { \sigma ( \delta _ { \mathcal { R } } ) } { \sigma _ { m a x } } ] ^ { \tau } ) } & { \mathrm { i f } \sigma ( \delta _ { \mathcal { R } } ) < \sigma _ { m a x } , \mathcal { R } _ { T } > 0 } \\ { 0 } & { \mathrm { o t h e r w i s e } } \end{array} \right.
+$$
+
+The formulation of Equation 5 share principled themes as in Sharpe (1994) and Dowd (2000):
+
+1. Dowd (2000) compares the newly obtained $\mathbf { S R } ^ { n e w }$ with the previous $\mathbf { S R } ^ { o l d }$ in choosing new assets; we derive $\mathcal { R } _ { H }$ in Eq. 4 by comparing the latest reward $\mathcal { R } _ { T }$ with $\mathcal { R } _ { 0 }$ to explicitly encourage the agent to aim for reward improvements in “choosing new actions”; 2. Both the Sharpe Ratio (SR) and Eq. 5 involve “variability weights” to adjust for the unit risk of return $\mathbb { E } [ \mathcal { R } ]$ Sharpe (1994) (i.e., $\scriptstyle { \frac { 1 } { \sigma _ { r } } }$ for SR and $\begin{array} { r } { 1 - \big [ \frac { \sigma ( \delta \mathcal { R } ) } { \sigma _ { m a x } } \big ] ^ { \tau } } \end{array}$ [ σ(δR) ]τ for rvwr ); 3. Whereas Dowd (2000) introduces the concept of “minimum required return” based on the elasticity of value at risk (VaR), we consider the maximum tolerance level $\sigma _ { m a x }$ with elasticity controlled by $\tau$ for improved learning stability of $r ^ { v w r }$ (see also Appendix H).
+
+Example computed values of $r ^ { v w r }$ for the characterization of different reward statistics are shown in Figure 2 and we show strong empirical results (in Section 6) to confirm the validity and robustness of the proposed formulation in multiple reinforcement learning domains.
+
+# 4.2 MULTI-CRITIC ARCHITECTURE
+
+A higher value of $r ^ { v w r }$ indicates better agent performance as the result of the historical sequence of actions. The same set of optimization procedures for conventional value function (i.e., via maximization of immediate reward signal $r$ ) update can be applied accordingly. The actual returns computed from both the “long-term” and “short-term” rewards are discounted by the same factor $\gamma$ In particular, for standard $N$ -step look-ahead approaches, we have:
+
+$$
+R _ { t } ^ { \mathrm { { s h o r t . e r m } } } = \sum _ { n = 0 } ^ { N - 1 } \gamma ^ { n } r _ { t + n } + \gamma ^ { N } V ( s _ { t + N } ) , \ : \ : \ : R _ { t } ^ { \mathrm { l o n g . e r m } } = \sum _ { n = 0 } ^ { N - 1 } \gamma ^ { n } r _ { t + n } ^ { v w r } + \gamma ^ { N } V ^ { v w r } ( s _ { t + N } )
+$$
+
+Similar to the standard state value function $V ( s )$ , we further define $V ^ { v w r } ( s )$ as the value function w.r.t the variability-weighted reward $r ^ { v w r }$ . These value functions form multiple critics judging a given state $s$ . The gradients w.r.t. the critics then become:
+
+$$
+\nabla _ { \theta ^ { \mathrm { s h o r i c m } } } [ ( R _ { t } ^ { \mathrm { s h o r t - e r m } } - V ( s _ { t } ; \theta ^ { \mathrm { s h o r t - e r m } } ) ) ^ { 2 } ] + \nabla _ { \theta ^ { \mathrm { l o r g - t e r m } } } [ ( R _ { t } ^ { \mathrm { l o n g - t e r m } } - V ^ { v w r } ( s _ { t } ; \theta ^ { \mathrm { l o n g - t e r m } } ) ) ^ { 2 } ]
+$$
+
+![](images/fd2a20fd68d11f839066eb2b4cb98f37deca225e98926d444892f991d6494b10.jpg)  
+Figure 3: Performance of A2MC on Atari games. “Hot-Wiring” exploration makes the agent easier to figure out how to play challenging games like “Robotank” and “WizardOfWor”, and for most games, it provides a better initial state for the agent to start off at a game and hence to obtain better final results. The number in figure legend shows the average reward among the last 100 episodes and the percentage shows the performance margin as compared to ACKTR. The shaded region denotes the standard deviation over 2 random seeds.
+
+where the standard grading clipping approach can be applied in Eq. 7 for enhanced stability. More advanced methods for estimating $R _ { t } ^ { \mathrm { s h o r t - t e r m } }$ and $R _ { t } ^ { \mathrm { { l o n g - t e r m } } }$ above, such as the online variant of generalized advantage estimation (GAE) using eligibility traces Schulman et al. (2016) can be adopted in place of Eq. 6, as shown below (see also Appendix $\mathbf { G }$ ):
+
+$$
+\begin{array} { r l } & { A _ { t } ^ { \mathrm { s h o r t . t e r m } } = \displaystyle \sum _ { n = 0 } ^ { \infty } ( \gamma \lambda ) ^ { n } \delta _ { t + n } ^ { v w r } , \mathrm { w i t h } \delta _ { t } = r _ { t } + \gamma V ( s _ { t + 1 } ) - V ( s _ { t } ) } \\ & { A _ { t } ^ { \mathrm { l o n g . t e r m } } = \displaystyle \sum _ { n = 0 } ^ { \infty } ( \gamma \lambda ) ^ { n } \delta _ { t + n } ^ { v w r } , \mathrm { w i t h } \delta _ { t } ^ { v w r } = r _ { t } ^ { v w r } + \gamma V ^ { v w r } ( s _ { t + 1 } ) - V ^ { v w r } ( s _ { t } ) } \end{array}
+$$
+
+where the generalized estimator of the advantage function $A _ { t } ^ { \mathrm { s h o r t - t e r m } }$ and $A _ { t } ^ { \mathrm { l o n g - t e r m } }$ allows a trade-off of bias $\nu . s .$ . variance using the parameter $0 \leq \lambda \leq 1$ , similar to the $\mathrm { T D } ( \lambda )$ approach for eligibility traces. We show the effectiveness of the proposed characterization of past reward statistics in multiple advantage actor-critic frameworks (i.e., ACKTR and PPO), where the two different value functions can share the same low-level feature representation, enabling a single agent to learn multiple critics parameterized by $\theta ^ { j } , j \in \{ \mathrm { s h o r t - t e r m } , \mathrm { l o n g - t e r m } \}$ . (See also Appendix I for the full algorithm).
+
+# 5 HOT-WIRE $\epsilon$ -EXPLORATION
+
+Being handed a game-stick, a human most likely would try out all the available buttons on it to see which particular button entails whatever actions on the game screen, hence receiving useful feedbacks. Inspired by this, we propose to hot-wire the agent to perform an identical sequence of randomly chosen actions in the N-step look-ahead during the initial stage (randomly pressing down a game-stick button for a while):
+
+$$
+a _ { t + k } = \left\{ \begin{array} { l l } { { \mathrm { ~ a ~ r a n d o m ~ a c t i o n ~ i d e n t i c a l ~ f o r ~ a l l ~ k ~ } } } & { { \mathrm { w i t h ~ p r o b ~ } \epsilon } } \\ { { \pi ( a _ { t + k } | s _ { t + k } ) ~ \mathrm { f o r } ~ k = 0 , 1 , 2 , . . . , N - 1 } } & { { \mathrm { w i t h ~ p r o b ~ } 1 - \epsilon } } \end{array} \right.
+$$
+
+We show that by enabling the “hot-wiring” mechanism2, a seemingly trapped agent can be boosted to learn to quickly solve problems where rewards can only be triggered by particular action sequences, as shown in games like “Robotank” and “WizardOfWor” in Figure 3.
+
+# 6 EXPERIMENTS
+
+We use the same network architecture and natural gradient optimization method as in the ACKTR model Wu et al. (2017). We set $\sigma _ { m a x } = 1 . 0$ , $\tau = 2 . 0$ and $T = 2 0$ in the computation of variabilityweighted reward (see Appendix C for hyperparameter studies). For hot-wiring exploration, we choose $\epsilon = 0 . 2 0$ and initial stage to be first $\scriptstyle { \frac { 1 } { 4 0 } }$ of the total training period for all experiments. Other hyperparameters such as learning rate and gradient clipping remain the same as in the ACKTR model Wu et al. (2017), in addition to adopting GAE Schulman et al. (2016) for a stronger ACKTR baseline (see Sec 4.2). We first present results of evaluating the proposed A2MC model in two standard benchmarks, the discrete Atari experiments and the continuous MuJoCo domain. Then we show ablation studies on the robustness of the hyper-parameters involved as well as evaluating the extensibility of the proposed long/short-term reward characterizations to other on-policy methods. Further extensions to off-policy domains are presented in Appendix E and Appendix F.
+
+# 6.1 ATARI 2600 GAMES
+
+We follow standard evaluation protocol to evaluate A2MC in a variety of Atari game environments (starting with 30 no-op actions). We train our models for 15 million timesteps for each game environment and score each game based on the average episode rewards obtained among the last 100 episodes as in Wu et al. (2017). The learning results on 12 Atari games are shown in Figure 3 where we also included an ablation experiment of A2MC without hot-wiring. We observe that on average A2MC improves upon ACKTR in terms of final performance under the same training budget. Our A2MC is able to consistently improve agent performance based on the proposed characterization of reward statistics, hence the agent is able to get out of local minima in less time (higher sample efficiency) compared to ACKTR. The complete learning results on all games are attached in the Appendix B.
+
+Table 1: Comparison of average episode rewards at the end of 50 million timesteps in Atari experiments. The reward scores and the first episodes reaching human-level performance Mnih et al. (2015) are reported as in Wu et al. (2017). A2MC is able to solve games that are challenging to ACKTR and also retain comparable performance in the rest of games.   
+
+<table><tr><td colspan="2"></td><td colspan="2">ACKTR</td><td colspan="2">A2MC</td></tr><tr><td>Domain</td><td>Human Level</td><td>Rewards</td><td>Episode</td><td>Rewards</td><td>Episode</td></tr><tr><td>Asteroids</td><td>47388.7</td><td>34171.0</td><td>N/A</td><td>830232.5</td><td>11314</td></tr><tr><td>Beamrider</td><td>5775.0</td><td>13581.4</td><td>3279</td><td>13564.3</td><td>3012</td></tr><tr><td>Boxing</td><td>12.1</td><td>1.5</td><td>N/A</td><td>99.1</td><td>158</td></tr><tr><td>Breakout</td><td>31.8</td><td>735.7</td><td>4097</td><td>411.4</td><td>3664</td></tr><tr><td>Double Dunk</td><td>-16.4</td><td>-0.5</td><td>742</td><td>21.3</td><td>544</td></tr><tr><td>Enduro</td><td>860.5</td><td>0.0</td><td>N/A</td><td>3492.2</td><td>730</td></tr><tr><td>Freeway</td><td>29.6</td><td>0.0</td><td>N/A</td><td>32.7</td><td>1058</td></tr><tr><td>Pong</td><td>9.3</td><td>20.9</td><td>904</td><td>19.4</td><td>804</td></tr><tr><td>Q-bert</td><td>13455.0</td><td>21500.3</td><td>6422</td><td>25229.0</td><td>7259</td></tr><tr><td>Robotank</td><td>11.9</td><td>16.5</td><td>-</td><td>25.7</td><td>4158</td></tr><tr><td>Seaquest</td><td>20182.0</td><td>1776.0</td><td>N/A</td><td>1798.6</td><td>N/A</td></tr><tr><td>Space Invaders</td><td>1652.0</td><td>19723.0</td><td>14696</td><td>11774.0</td><td>11064</td></tr><tr><td>Wizard of Wor</td><td>4756.5</td><td>702</td><td>N/A</td><td>7471.0</td><td>8119</td></tr></table>
+
+We further expand the training budget and continue learning the games until 50 million timesteps as in Wu et al. (2017). As shown in Table 1, our A2MC model can solve games like “Boxing”, “Freeway” and “Enduro” that are challenging for the baseline ACKTR model. For a full picture of model performance in Atari games, A2MC has a human-level performance rate of $7 4 . 5 \%$ (38 out of 51 games) in the Atari benchmarks, compared to $6 3 . 6 \%$ reached by ACKTR. Individual game scores for all the Atari games are reported in the Appendix B.
+
+# 6.2 CONTINUOUS CONTROL
+
+For the evaluations on continuous control tasks simulated in MuJoCo environment, we first follow $\mathrm { W u }$ et al. (2017) and tune a different set of hyper-parameters from Atari experiments. Specifically, all
+
+![](images/3a06f904da805d4c7acb1d86247f9e70b0ea241d1cad9dbdc73b865f57ba7291.jpg)  
+Figure 4: Performance on the MuJoCo benchmark. A2MC is also competitive on MuJoCo continuous domain when compared to ACKTR. The shaded region denotes std over 3 random seeds.
+
+MuJoCo experiments are trained with a larger batch size of 2500. The results of eight MuJoCo environments trained for 1 million timesteps are shown in Figure 4. We observe that A2MC also performs well in all MuJoCo continuous control tasks. In particular, A2MC has brought significant improvement on the tasks of HalfCheetah, Swimmer and Walker2d (see Table 2).
+
+To test the robustness of A2MC, we perform another set of evaluations on MuJoCo tasks by keeping an identical set of hyper-parameters used in the Atari experiments. Figure 7 in Appendix C shows this ablation result. We observe that even under sub-optimal hyper-parameters, our A2MC model can still learn to solve the MuJoCo control tasks in the long run. Moreover, it is less prone to overfitting when compared to ACKTR under such “stress testing”. Additional hyper-parameter studies can be found in Appendix C.
+
+We also evaluate a multi-critics variant of the proximal policy optimization (PPO) model on the MuJoCo tasks with our proposed long/short-term rewards. In particular, we observe that our proposed variability-weighted reward generalizes well with the vanilla PPO, and our multi-critics PPO variant (MC-PPO) shows more favorable performance, as shown in Table 2. Specifically, MC-PPO shows the best performance on Hopper and Walker- $_ { 2 d }$ among all models under the 1-million timesteps training budget. Both of our multi-critics variants (A2MC and MC-PPO) have won 6 out of the 8 MuJoCo tasks with relative performance margins (percentages in parentheses) larger than $2 5 \%$ (see Table 2).
+
+Table 2: Average episode rewards obtained among the last 10 episodes upon 1 million timesteps of training in MuJoCo experiments.   
+
+<table><tr><td>GAMES</td><td>ACKTR</td><td colspan="2">Our A2MC</td><td>PPO</td><td colspan="2">Our MC-PPO</td></tr><tr><td>Ant</td><td>1671.6</td><td>2216.1</td><td>(32.5%)</td><td>411.4 (± 107.7)</td><td>618.9</td><td>(50.4%)</td></tr><tr><td>HalfCheetah</td><td>1676.2</td><td>2696.6</td><td>(60.8%)</td><td>1433.7 (± 83.9)</td><td>2473.4</td><td>(72.5%)</td></tr><tr><td>Hopper</td><td>2259.1</td><td>2835.7</td><td>(25.5%)</td><td>2055.8 (± 274.6)</td><td>3131.3</td><td>(52.3%)</td></tr><tr><td>InvertedDoublePendulum 6295.4</td><td></td><td>7872.6</td><td>(25.0%)</td><td>4454.1 (± 1098.1)</td><td>7648.7</td><td>(71.7%)</td></tr><tr><td>InvertedPendulum</td><td>1000.0</td><td>957.2</td><td>(-4.2%)</td><td>839.7 (± 127.1)</td><td>777.4</td><td>(-7.4%)</td></tr><tr><td>Reacher</td><td>-4.2</td><td>-3.9</td><td>(0.4%)</td><td>-5.47 (± 0.3)</td><td>-10.3</td><td>(-8.5%)</td></tr><tr><td>Swimmer</td><td>43.2</td><td></td><td>187.4 (333.7%)</td><td>79.1 (± 31.2)</td><td>102.9</td><td>(30.2%)</td></tr><tr><td>Walker2d</td><td>1090.8</td><td>2405.9 (120.5%)</td><td></td><td>2300.8 (± 397.6)</td><td>3718.1</td><td>(61.6%)</td></tr><tr><td>Win—Fair—Lose</td><td>N/A</td><td>6-2-0</td><td></td><td>N/A</td><td>6—2-0</td><td></td></tr></table>
+
+# 7 CONCLUSION
+
+In this work, we introduce an effective auxiliary reward signal to remedy the deficiencies of learning solely from the standard environment rewards. Our proposed characterization of past reward statistics results in improved learning and higher sample efficiencies for on-policy methods, especially in challenging tasks with sparse rewards. Experiments on both discrete tasks in Atari environment and MuJoCo continuous control tasks validate the effectiveness of utilizing the proposed long/short-term reward statistics for on-policy methods using multi-critic architectures. This suggests that expanding the form of reward feedbacks in reinforcement learning is a promising research direction.
+
+# REFERENCES
+
+Marcin Andrychowicz, Filip Wolski, Alex Ray, Jonas Schneider, Rachel Fong, Peter Welinder, Bob McGrew, Josh Tobin, OpenAI Pieter Abbeel, and Wojciech Zaremba. Hindsight experience replay. In Advances in Neural Information Processing Systems, pp. 5048–5058, 2017.
+
+Marc G Bellemare, Will Dabney, and Remi Munos. A distributional perspective on reinforcement ´ learning. arXiv preprint arXiv:1707.06887, 2017.
+
+Kevin Dowd. Adjusting for risk:: An improved sharpe ratio. International review of economics & finance, 9(3):209–222, 2000.
+
+Matteo Hessel, Joseph Modayil, Hado Van Hasselt, Tom Schaul, Georg Ostrovski, Will Dabney, Dan Horgan, Bilal Piot, Mohammad Azar, and David Silver. Rainbow: Combining improvements in deep reinforcement learning. 2018.
+
+Max Jaderberg, Volodymyr Mnih, Wojciech Marian Czarnecki, Tom Schaul, Joel Z Leibo, David Silver, and Koray Kavukcuoglu. Reinforcement learning with unsupervised auxiliary tasks. arXiv preprint arXiv:1611.05397, 2016.
+
+Junqi Jin, Chengru Song, Han Li, Kun Gai, Jun Wang, and Weinan Zhang. Real-time bidding with multi-agent reinforcement learning in display advertising. arXiv preprint arXiv:1802.09756, 2018.
+
+George Konidaris and Andrew G Barto. Skill discovery in continuous reinforcement learning domains using skill chaining. In Advances in neural information processing systems, pp. 1015–1023, 2009.
+
+Guillaume Lample and Devendra Singh Chaplot. Playing fps games with deep reinforcement learning. In AAAI, 2017.
+
+Marc Lanctot, Vinicius Zambaldi, Audrunas Gruslys, Angeliki Lazaridou, Julien Perolat, David Silver, Thore Graepel, et al. A unified game-theoretic approach to multiagent reinforcement learning. In Advances in Neural Information Processing Systems, pp. 4193–4206, 2017.
+
+Ryan Lowe, Yi Wu, Aviv Tamar, Jean Harb, OpenAI Pieter Abbeel, and Igor Mordatch. Multi-agent actor-critic for mixed cooperative-competitive environments. In Advances in Neural Information Processing Systems, pp. 6382–6393, 2017.
+
+Volodymyr Mnih, Koray Kavukcuoglu, David Silver, Andrei A Rusu, Joel Veness, Marc G Bellemare, Alex Graves, Martin Riedmiller, Andreas K Fidjeland, Georg Ostrovski, et al. Human-level control through deep reinforcement learning. Nature, 518(7540):529, 2015.
+
+Volodymyr Mnih, Adria Puigdomenech Badia, Mehdi Mirza, Alex Graves, Timothy Lillicrap, Tim Harley, David Silver, and Koray Kavukcuoglu. Asynchronous methods for deep reinforcement learning. In International Conference on Machine Learning, pp. 1928–1937, 2016.
+
+John Schulman, Philipp Moritz, Sergey Levine, Michael Jordan, and Pieter Abbeel. High-dimensional continuous control using generalized advantage estimation. ICLR, 2016.
+
+John Schulman, Filip Wolski, Prafulla Dhariwal, Alec Radford, and Oleg Klimov. Proximal policy optimization algorithms. arXiv preprint arXiv:1707.06347, 2017.
+
+William F Sharpe. The sharpe ratio. Journal of portfolio management, 21(1):49–58, 1994.
+
+David Silver and Kamil Ciosek. Compositional planning using optimal option models. arXiv preprint arXiv:1206.6473, 2012.
+
+David Silver, Julian Schrittwieser, Karen Simonyan, Ioannis Antonoglou, Aja Huang, Arthur Guez, Thomas Hubert, Lucas Baker, Matthew Lai, Adrian Bolton, et al. Mastering the game of go without human knowledge. Nature, 550(7676):354, 2017.
+
+Richard S Sutton and Andrew G Barto. Reinforcement learning: An introduction. MIT press, 2018.
+
+Yee Teh, Victor Bapst, Wojciech M Czarnecki, John Quan, James Kirkpatrick, Raia Hadsell, Nicolas Heess, and Razvan Pascanu. Distral: Robust multitask reinforcement learning. In Advances in Neural Information Processing Systems, pp. 4499–4509, 2017.
+
+Harm Van Seijen, Mehdi Fatemi, Joshua Romoff, Romain Laroche, Tavian Barnes, and Jeffrey Tsang. Hybrid reward architecture for reinforcement learning. In NIPS, pp. 5392–5402, 2017.
+
+Ziyu Wang, Victor Bapst, Nicolas Heess, Volodymyr Mnih, Remi Munos, Koray Kavukcuoglu, and Nando de Freitas. Sample efficient actor-critic with experience replay. arXiv preprint arXiv:1611.01224, 2016.
+
+Ronald J Williams. Simple statistical gradient-following algorithms for connectionist reinforcement learning. In Reinforcement Learning, pp. 5–32. Springer, 1992.
+
+Yuhuai Wu, Elman Mansimov, Roger B Grosse, Shun Liao, and Jimmy Ba. Scalable trust-region method for deep reinforcement learning using kronecker-factored approximation. In Advances in neural information processing systems, pp. 5285–5294, 2017.
+
+# APPENDIX
+
+# A EFFECTS OF FLIPPING
+
+While introducing the variability-weighted reward, a flipping operation is conducted in the preprocessing of the reward sequence as formulated in Eq. (3). In Figure 5 and 6, we construct 4 reward sequences to show that the flipping operation can further penalize the oscillation in the recent past rewards while encourage recent stable rewards. (a1, a2, b1, b2) share the same value of immediate reward at $t = 9$ and thus the $\mathcal { R } _ { H }$ of all reward sequences are the same. Therefore, the variability-weighted reward only depends on the volatility statistics of $\delta _ { R }$ , i.e., how varied past rewards were.
+
+![](images/b8c96c2bfb2986a8131744bff4f87fb64d6ad7d6ab6465c97ddc8d0c4bc5483f.jpg)  
+Figure 5: Calculation without flipping.
+
+![](images/03f318fa74dc2f857ea0e3562e3d4f2c6896c003afc641c8b8400191574d409c.jpg)  
+Figure 6: Calculation with flipping.
+
+Without flipping. In Figure 5, sequence $( a l )$ and $( a 2 )$ are mirror symmetrical to the $y$ -axis, and the only difference between them is that the recent past rewards $( t = 5 , 6 , 7 , 8 )$ ) of $( a 2 )$ are more stable than (a1). Intuitively, we want to encourage stable past rewards like $( a 2 )$ while penalizing oscillation in (a1). As presented in the third row of Figure 5, the $r ^ { v w r }$ difference of $( a I )$ and $( a 2 )$ is less than 1 without flipping in the pre-processing.
+
+With flipping. In Figure 6, (b1, b2) exactly have the same reward sequence as $( a l , a 2 )$ , respectively. However, flipping is performed as a step of pre-processing, largely increasing the $r ^ { v w r }$ gap (from less than 1 to nearly 4) between the two constructed sequences. Comparing $( b l , b 2 )$ with $( a l , a 2 )$ , the post-processed sequences $\vec { \mathcal { R } }$ (shown in green) become centrosymmetric to those without flipping. Specifically, the recent reward drops at $t = 6 , 7 , 8$ are reflected as high values at the beginning of $\vec { \mathcal { R } }$ as shown in $( b l )$ , while oscillations long ago are transformed into high values at the end of $\vec { \mathcal { R } }$ as presented in $( b 2 )$ . When compared to the zero-variability reference (shown in orange), which is designed as an exponential function, the flipping leads to a higher variability for the former sequence while a lower variability for the latter one, enlarging the $r ^ { v w r }$ gap between those two sequences.
+
+# B COMPLETE RESULTS IN ATARI 2600 GAMES
+
+We show the learning curves for 15 million timesteps on all Atari games in Figure 12 and in Table 3 we show the complete results of training til 50 million timesteps. report the mean episode reward as in Wu et al. (2017). Entries with $\sim$ indicates approximated value as retrieved from learning figures published by Wu et al. (2017). Results from other models are taken from Wu et al. (2017) and Mnih et al. (2015). We show that A2MC has reached a human-level performance rate of $7 4 . 5 \%$ (38 out of 51 games) as compared to $6 3 . 6 \%$ reached by ACKTR. The relative performance margin of A2MC as compared to ACKTR is also shown.
+
+# C HYPER-PARAMETER STUDIES
+
+The proposed variability-weighted reward mechanism considers the volatility of rewards by keeping a $T$ -step history of agent’s performance. The hyper-parameter $T = 2 0$ is empirically chosen to be the same as the look-ahead parameter $N$ in standard on-policy methods, so as to keep the same period $\textstyle T = N = 2 0$ ) in “T-step history” and “N-step forward”. And $\sigma _ { m a x } = 1$ is chosen as the maximum of the observed volatility based on statistics in the $\mathrm { T }$ history rewards of the ACKTR models. As parameter choices could be vital, we perform an additional ablation study shown below. The result shows that the performance of A2MC is robust across different parameters of choice and is not too sensitive to changes on either of the hyper-params.
+
+<table><tr><td>Games</td><td>ACKTR</td><td>A2MC w/</td><td>T=20 Omax=1</td><td>T=10 Omax=1</td><td>T=10 0max=2</td><td>T=40 Omax=1</td><td>T=40 Omax=2</td></tr><tr><td>Boxing</td><td>1.23</td><td></td><td>99.19</td><td>94.76</td><td>98.51</td><td>99.18</td><td>98.07</td></tr><tr><td>Jamesbond</td><td>409.50</td><td></td><td>453.50</td><td>438.50</td><td>470.00</td><td>442.25</td><td>457.75</td></tr><tr><td>Wizard of Wor</td><td>744.50</td><td></td><td>5448.00</td><td>5601.00</td><td>5363.50</td><td>2528.50</td><td>3287.50</td></tr></table>
+
+![](images/162bf3e0520ec61e2d4440c63fcb2a71cb0c9e75a49eabbe055eb3df783697fa.jpg)  
+Figure 7: “Stress testing” ablation study on the MuJoCo continuous benchmark using hyperparameters tuned in Atari discrete control. Although this set of hyperparameters is suboptimal for the MuJoCo continuous control tasks, A2MC still obtain reasonable performance in the long run and it is less prone to overfitting.
+
+Table 3: Raw scores across all games, starting with 30 no-op actions. Scores are reported by averaging the last 500 episodes upon 50 million timesteps of training as in Wu et al. (2017). A relative margin comparing A2MC to ACKTR is shown. Scores from other models are taken from Wu et al. (2017) and Mnih et al. (2015).   
+
+<table><tr><td>GAME</td><td>Human</td><td>DQN</td><td>DDQN Prior. Duel</td><td></td><td>ACKTR OurA2MC</td><td></td><td>(Margin)</td></tr><tr><td>Alien</td><td>7127.7</td><td>1620</td><td>3747.7</td><td>3941</td><td>3197.1</td><td>2986.3</td><td>-6.6%</td></tr><tr><td>Amidar</td><td>1719.5</td><td>978</td><td>1793.3</td><td>2296.8</td><td>1059.4</td><td>2040.1</td><td>92.6%</td></tr><tr><td>Assault</td><td>742.0</td><td>4280.4</td><td>5393.2</td><td>11477</td><td>10777.7</td><td>9892.4</td><td>-8.2%</td></tr><tr><td>Asterix</td><td>8503.3</td><td>4359</td><td>17356.5</td><td>375080</td><td>31583.0</td><td>32671.0</td><td>3.4%</td></tr><tr><td>Asteroids</td><td>47388.7</td><td>1364.5</td><td>734.7</td><td>1192.7</td><td>34171.6</td><td>828931.6</td><td>2325.8%</td></tr><tr><td>Atlantis</td><td>29028.1</td><td>279987</td><td>106056</td><td>395762</td><td>3433182.0</td><td>2886274.0</td><td> -15.9%</td></tr><tr><td>Bankheist</td><td>753.1</td><td>455</td><td>1030.6</td><td>1503.1</td><td>1289.7</td><td>1290.6</td><td>0.1%</td></tr><tr><td>Battlezone</td><td>37187.5</td><td>29900</td><td>31700</td><td>35520</td><td>8910.0</td><td>10570.0</td><td>18.6%</td></tr><tr><td>Beamrider</td><td>16926.5</td><td>8627.5</td><td>13772.8</td><td>30276.5</td><td>13581.4</td><td>13715.6</td><td>1.0%</td></tr><tr><td>Berzerk</td><td>2630.4</td><td>585.6</td><td>1225.4</td><td>3409</td><td>927.2</td><td>974.0</td><td>5.0%</td></tr><tr><td>Bowling</td><td>160.7</td><td>50.4</td><td>68.1</td><td>46.7</td><td>24.3</td><td>31.6</td><td>30.0%</td></tr><tr><td>Boxing</td><td>12.1</td><td>88</td><td>91.6</td><td>98.9</td><td>1.5</td><td>93.5</td><td>6344.8%</td></tr><tr><td>Breakout</td><td>30.5</td><td>385.5</td><td>418.5</td><td>366</td><td>735.7</td><td>420.6</td><td>-42.8%</td></tr><tr><td>Centipede</td><td>12017.0</td><td>4657.7</td><td>5409.4</td><td>7687.5</td><td>7125.3</td><td>12096.5</td><td>69.8%</td></tr><tr><td>Choppercommand</td><td>9882.0</td><td>N/A</td><td>N/A</td><td>N/A</td><td>~8000</td><td>12149.0</td><td>~42.5%</td></tr><tr><td>Crazyclimber</td><td>35829.4</td><td>110763</td><td>117282</td><td>162224</td><td>150444.0</td><td>152439.0</td><td>1.3%</td></tr><tr><td>Demonattack</td><td>1971.0</td><td>12149.4</td><td>58044.2</td><td>72878.6</td><td>274176.7</td><td>361807.1</td><td>32.0%</td></tr><tr><td>Doubledunk</td><td>-16.4</td><td>-6.6</td><td>-5.5</td><td>-12.5</td><td>-0.5</td><td>20.6</td><td>3907.5%</td></tr><tr><td>Enduro</td><td>860.5</td><td>729</td><td>1211.8</td><td>2306.4</td><td>0.0</td><td>3550.6</td><td>8%</td></tr><tr><td>Fishingderby</td><td>-38.7</td><td>-4.9</td><td>15.5</td><td>41.3</td><td>33.7</td><td>38.4</td><td>13.9%</td></tr><tr><td>Freeway</td><td>29.6</td><td>30.8</td><td>33.3</td><td>33</td><td>0.0</td><td>32.7</td><td>0%</td></tr><tr><td>Frostbite</td><td>4335.0</td><td>N/A</td><td>N/A</td><td>N/A</td><td>~280</td><td>293.7</td><td>~5.1%</td></tr><tr><td>Gopher</td><td>2412.5</td><td>8777.4</td><td>14840.8</td><td>104368.2</td><td>47730.8</td><td>86101.4</td><td>80.4%</td></tr><tr><td>Gravitar</td><td>2672.0</td><td>N/A</td><td>N/A</td><td>N/A</td><td>~300</td><td>995.0</td><td>-2.9%</td></tr><tr><td>Icehockey</td><td>0.9</td><td>-1.9</td><td>-2.7</td><td>-0.4</td><td>-4.2</td><td>-2.1</td><td>16.3%</td></tr><tr><td>Jamesbond</td><td>302.8</td><td>768.5</td><td>1358</td><td>812</td><td>490.0</td><td>545.0</td><td>11.2%</td></tr><tr><td>Kangaroo</td><td>3035.0</td><td>7259</td><td>12992</td><td>1792</td><td>3150.0</td><td>11269.0</td><td>257.7%</td></tr><tr><td>Krull</td><td>2665.5</td><td>8422.3</td><td>7920.5</td><td>10374.4</td><td>9686.9</td><td>10245.4</td><td>5.8%</td></tr><tr><td>Kungfumaster</td><td>22736.3</td><td>26059</td><td>29710</td><td>48375</td><td>34954.0</td><td>39773.0</td><td>13.8%</td></tr><tr><td>Mspacman</td><td>15693.0</td><td>N/A</td><td>N/A</td><td>N/A</td><td>~3500</td><td>5006.1</td><td>~34.5%</td></tr><tr><td>Namethisgame</td><td>4076.0</td><td>N/A</td><td>N/A</td><td>N/A</td><td>~12500</td><td>12569.9</td><td>~0.6%</td></tr><tr><td>Phoenix</td><td>7242.6</td><td>8485.2</td><td>12252.5</td><td>70324.3</td><td>133433.7</td><td>221288.3</td><td>65.8%</td></tr><tr><td>Pitfall</td><td>6463.7</td><td>-286.1</td><td>-29.9</td><td>0</td><td>-1.1</td><td>-2.5</td><td>-0.3%</td></tr><tr><td>Pong</td><td>14.6</td><td>20.9</td><td>21</td><td>20.9</td><td>20.9</td><td>19.7</td><td>-5.9%</td></tr><tr><td>Privateeye</td><td>69571.0</td><td>N/A</td><td>N/A</td><td>N/A</td><td>~560</td><td>507.0</td><td>-9.5%</td></tr><tr><td>Qbert</td><td>13455.0</td><td>13117.3</td><td>15088.5</td><td>18760.3</td><td>23151.5</td><td>24075.8</td><td>4.0%</td></tr><tr><td>Riverraid</td><td>17118.0</td><td>7377.6</td><td>14884.5</td><td>20607.6</td><td>17762.8</td><td>18671.9</td><td>5.1%</td></tr><tr><td>Roadrunner</td><td>7845.0</td><td>39544</td><td>44127</td><td>62151</td><td>53446.0</td><td>50071.0</td><td>-6.3%</td></tr><tr><td>Robotank</td><td>11.9</td><td>63.9</td><td>65.1</td><td>27.5</td><td>16.5</td><td>26.5</td><td>60.5%</td></tr><tr><td>Seaquest</td><td>42054.7</td><td>5860.6</td><td>16452.7</td><td>931.6</td><td>1776.0</td><td>1805.6</td><td>1.7%</td></tr><tr><td>Solaris</td><td>12326.7</td><td>3482.8</td><td>3067.8</td><td>133.4</td><td>2368.6</td><td>2277.2</td><td>-3.9%</td></tr><tr><td>Spaceinvaders</td><td>1668.7</td><td>1692.3</td><td>2525.5</td><td>15311.5</td><td>19723.0</td><td>13544.2</td><td>-31.3%</td></tr><tr><td>Stargunner</td><td>10250.0</td><td>54282</td><td>60142</td><td>125117</td><td>82920.0</td><td>89616.0</td><td>8.1%</td></tr><tr><td>Tennis</td><td>-8.9</td><td>N/A</td><td>N/A</td><td>N/A</td><td>~-12</td><td>-4.7</td><td>~20.4%</td></tr><tr><td>Timepilot</td><td>5229.2</td><td>4870</td><td>8339</td><td>7553</td><td>22286.0</td><td>21992.0</td><td>-1.3%</td></tr><tr><td>Tutankham</td><td>167.6</td><td>68.1</td><td>218.4</td><td>245.9</td><td>314.3</td><td>193.7</td><td>-38.4%</td></tr><tr><td>Upndown</td><td>11693.2</td><td>9989.9</td><td>22972.2</td><td>33879.1</td><td>436665.8</td><td>563659.3</td><td>29.1%</td></tr><tr><td>Videopinball</td><td>17667.9</td><td>196760.4</td><td>309941.9</td><td>479197</td><td>100496.0</td><td>127452.4</td><td>26.8%</td></tr><tr><td>Wizardofwor YarsRevenge</td><td>4756.5</td><td>2704</td><td>7492</td><td>12352</td><td>702.0 125169.0</td><td>7864.0 143141.5</td><td>1020.2% 14.4%</td></tr></table>
+
+# D EXTENSION TO MULTI-CRITIC PPO (MC-PPO)
+
+The learning results of the proposed MC-PPO model on the MuJoCo tasks are shown in Figure 8. MC-PPO shows the best performance on Hopper and Walker- $_ { 2 d }$ among all models under the 1-million timesteps training budget. Both of our multi-critics variants (A2MC and MC-PPO) have won 6 out of the 8 MuJoCo tasks with relative performance margins (percentages in parentheses) larger than $2 5 \%$ .
+
+![](images/47b1442f99fadd4e266ec71e7ffe602454c89c06ae82dbdc9f5bc465cc0f2b12.jpg)  
+Figure 8: Performance on the MuJoCo continuous control benchmarks using PPO-based methods. Our proposed long/short-term reward characterization can be extended to the PPO method, i.e., the proposed multi-critic variant of PPO (MC-PPO). The shaded region denotes the standard deviation over 3 random seeds.
+
+# E EXTENSION TO OFF-POLICY METHODS
+
+Methods involving experience replay belong to the family of off-policy methods as they learn from off-policy trajectories. They were considered to be beyond the scope of this work, as we set out to improve the family of “on-policy” methods and we try to present as complete the analyses as possible (on both Atari and MuJoCo) in the main text.
+
+Notwithstanding this, we have been actively exploring the potential of applying the proposed reward mechanism with off-policy methods (in particular, on the strong baseline Rainbow Hessel et al. (2018). For consistencies in comparisons, all hyperparameters (e.g., learning rate, distributional atoms, noisy net $\sigma _ { 0 }$ ) are kept identical as in Hessel et al. (2018) except that we used a smaller replay buffer size of 50,000 for both the baseline and our method (due to limited compute). Moreover, we use the same experiment settings as in Sec 6 and we have NOT further tuned any parameters in VWR. We show preliminary results at 10 million time steps on Atari games in Figure 9 and we observe it is promising that introducing the proposed characterization of variability-weighted reward mechanism improves off-policy methods as well.
+
+The robustness of our proposed reward mechanism across both on-policy and off-policy frameworks suggests that the concept of “risk-adjusted return” Sharpe (1994) should apply in reinforcement learning in general, as it brings the desired property in faciliating better sample efficiency and learning stability. Given limited time and computing resources we are not able to present a full analysis on all the off-policy frameworks as we did for the on-policy methods within this paper (since training off-policy models takes significantly longer time). Potentially we aim to have the complete results in an additional paper in our future works.
+
+![](images/49950388de336cc09e187e4bf97c4e8d8151b5317eb9c6ae2d692622fb39fa69.jpg)  
+Figure 9: Performance of applying the variability-weighted reward to the Rainbow model on the Atari benchmark. We observe that introducing the proposed reward characterizations significantly expediate the learning in games such as “Jamesbond” and “NameThisName”, while showing consistent improvement towards the rest. The shaded region denotes the standard deviation over 2 random seeds.
+
+# F CASE STUDY: PLAYING DOOM WITH REWARD SHAPING
+
+It is worth investigating whether the proposed auxiliary reward signal VWR can work “side-by-side” with carefully shaped rewards specific to some particular game scenario – for example, the FPS game Doom Lample & Chaplot (2017). As our proposed reward characterization is generic in design and orthogonal to reward shaping, we aim to validate that the concept of risk-adjusted return and variability weights can be equally applied under such shaping settings.
+
+To this end, we adopt the off-policy agent “Arnold” Lample & Chaplot (2017) with experience replay as our baseline and we calculate VWR (see Section 4]) based on the historical sequence of the shaped rewards defined in Lample & Chaplot (2017) (See the Table 4). For VWR parameters, we set $\sigma _ { m a x } = 5$ since the maximum (minimum) attainable reward is $5 . 0 \ : ( - 5 . 0 )$ under such reward shaping3. The rest of the game setup and bot numbers are defaulted to the code released by Lample & Chaplot (2017).
+
+Table 4: Reward shaping settings as in Arnold Lample & Chaplot (2017)   
+
+<table><tr><td>Type</td><td>Base /Dist Kill</td><td></td><td></td><td></td><td> Suicide Death Injured Use ammo Weapon /Ammo /Medkit /Armor</td></tr><tr><td>Value</td><td>0.0 5.0 -5.0</td><td>-5.0 -1.0</td><td></td><td>-0.2</td><td>1.0 / 1.0/ 1.0 /1.0</td></tr></table>
+
+We follow the evaluation criterion of Track-1 in ViZDoom AI Competition 2016 using “Frags per episode”, i.e., the number of kills minus the number of suicides for the agent in one round of game (higher is better). The result under 50 training hours is shown in Figure 10 and we consistently observe that the Arnold agent can be significantly boosted with the help of VWR. This confirms that our proposed reward characterization is able to bring further improvements on top of both reward shaping and experience replay methods across domains.
+
+<table><tr><td colspan="3">(a) Game statistics in 50 hours</td></tr><tr><td>After 24 hours</td><td>Arnold</td><td>Arnod + VWR</td></tr><tr><td>Kills</td><td>105</td><td>183</td></tr><tr><td>Frags</td><td>87</td><td>173</td></tr><tr><td>K/D ratio</td><td>1.48</td><td>2.08</td></tr><tr><td>After 50 hours</td><td></td><td></td></tr><tr><td>Kills</td><td>116</td><td>244</td></tr><tr><td>Frags</td><td>113</td><td>223</td></tr><tr><td>K/D ratio</td><td>2.00</td><td>2.65</td></tr></table>
+
+![](images/db55751aa06be9678b9b3cbbc236a6556fe7f569ecf88f6d40f94b718c84ff35.jpg)  
+Figure 10: Doom - Limited Deathmatch (Track-1)
+
+(b) Learning results averaged over 2 random seeds
+
+# G ABLATION STUDY: VWR V.S. ELIGIBILITY TRACE
+
+Eligibility traces $\mathrm { T D } ( \lambda )$ is widely used in bridging TD algorithms to Monte Carlo (MC) methods.   
+Essentially, the discounted cumulative return can be formulated by not just toward “any n-step” return (using n-step look ahead), but toward any average of n-step look-ahead returns Sutton & Barto (2018).   
+The online variant of generalized advantage estimation using eligibility traces (GAE) Schulman et al.   
+(2016) confirms that on-policy methods can benefit from $\mathrm { T D } ( \lambda )$ learning.
+
+For the proposed variability-weighted reward, the design theme is to look explicitly backward and to assess the past performance of the agent via the “risk-adjusted return” concept. These two mechanisms can be combined seamlessly via Eq. 8 and our empirical results suggest VWR brings further improvements on top of eligibility traces.
+
+As VWR and eligibility traces are thematically similar in some sense, we further perform an ablation study to contrast the contributions brought by VWR. As shown in Figure 11, we compare three different settings: (1) ACKTR $^ +$ GAE, (2) ACKTR $^ +$ vwr and (3) $\mathbf { A C K T R } + \mathbf { G A E } + \mathbf { v w r }$ (i.e., the proposed A2MC). We observe that on average VWR brings greater improvements compared to eligibility traces, and the combination of both (i.e., A2MC) results in consistently good performance across the Atari testbed.
+
+![](images/07967da42241a720cc821748d81b04ae5e04c2e52c7cff56b270943ae05ee01a.jpg)  
+Figure 11: Ablation study of separately applying the (1) the eligibility traces (GAE) and (2) variabilityweighted reward (VWR) to the ACKTR model on the Atari benchmark. We observe that the combination of both (i.e., A2MC) results in consistently good performance across the Atari testbed. The shaded region denotes the standard deviation over 2 random seeds.
+
+# H THE SHARPE RATIO ITSELF
+
+We have explored other forms of reward that fits the general idea of introducing variability weights to the reward shaping mechanism. One example is the “Sharpe Ratio” itself, which is defined as $\begin{array} { r } { r ^ { S R } \ = \ \frac { \mathbb { E } [ r ] } { \sigma ( r ) } } \end{array}$ . In our initial studies, we found it only improved upon the baseline marginally, as rSR could end up emphasizing on penalizing high-variations and it might discourage the agent too intensively (see Figure below). Thats why we have sought an alternative formulation using the proposed $r ^ { v w r }$ and found that $A 2 M C _ { V \bar { W } R } ~ > ~ A 2 M \bar { C _ { S R } } ~ > ~ A C K T R$ . An example highlighting the vwr benefit is provided in Appendix A and a more thorough survey on key components in reward designs/formulations will be included in our future works.
+
+# I ALGORITHM
+
+The learning algorithm of A2MC is shown in Algorithm 1.
+
+![](images/253d4905e0b1e4bda4a2e88e4a8861653c091bb27f95f978262ed8fdd7f93806.jpg)
+
+# Algorithm 1 Advantage Actor Multi-Critic Learning (A2MC)
+
+1: Initialize parameters: $\theta _ { a } , \theta _ { v } ^ { j } , j \in \{ \mathrm { s h o r t - t e r m } , \mathrm { l o n g - t e r m } \}$   
+2: Initialize look-ahead steps: $N$ , step counter: $T = 0$ , maximum step: $T _ { m a x }$   
+3: Initialize hot-wire probability: $\epsilon$   
+4: Initialize environment: Env   
+5: Initialize reward history: \~r   
+6: repeat   
+7: Reset gradients: $d \theta \gets 0$ and $d \theta _ { v } ^ { j } \gets 0 , j \in \{ \mathrm { s h o r t } \mathrm { - t e r m } , \mathrm { l o n g } \mathrm { - t e r m } \}$   
+8: Get state: $s _ { t } \gets E n v$   
+9: $f l a g = 1$ , $a _ { r a n d }$ is uniformly sampled in action space with probability $\epsilon$ , otherwise $f l a g = 0$   
+10: for $t = 0 : N - 1$ do   
+11: Perform $a _ { t }$ according to policy $\pi ( a _ { t } | s _ { t } ; \theta _ { a } )$ if not f lag else $a _ { t } = a _ { r a n d }$   
+12: Received reward $r _ { t }$ and new state $s _ { t + 1 }$ , append $r _ { t }$ to $\vec { \bf r }$   
+13: Calculate $r _ { t } ^ { v w r }$ from $\vec { \mathbf { r } }$ based on Eq. (2-7)   
+14: $T \gets T + \dot { 1 }$   
+15: end for   
+16: $R ^ { \mathrm { s h o r t - t e r m } } = V ( s _ { N } ; \theta _ { v } ^ { \mathrm { s h o r t - t e r m } } )$   
+17: $R ^ { \mathrm { { l o n g - t e r m } } } = V ( s _ { N } ; \theta _ { v } ^ { \mathrm { { l o n g - t e r m } } } )$   
+18: for $i = N - 1$ to 0 step $- 1$ do   
+19: $\begin{array} { l } { { R _ { } ^ { \mathrm { s h o r t - t e r m } }  r _ { i } + \bar { \gamma } R ^ { \mathrm { s h o r t - t e r m } } } } \\ { { R _ { } ^ { \mathrm { l o n g - t e r m } }  r _ { i } ^ { v w r } + \gamma R ^ { \mathrm { l o n g - t e r m } } } } \end{array}$   
+20:   
+21: Advantange gradients wrt $\begin{array} { r } { \theta _ { a } : d \theta _ { a } \gets d \theta _ { a } + \nabla _ { \theta _ { a } } \log \pi ( a _ { i } | s _ { i } ; \theta _ { a } ) \sum _ { j } ( R ^ { j } - V ( s _ { i } ; \theta _ { v } ^ { j } ) ) } \end{array}$   
+22: for $j \in$ {short-term, long-term} do   
+23: Accumulate gradients wrt $\dot { \theta } _ { v } ^ { j } : d \theta _ { v } ^ { j }  d \theta _ { v } ^ { j } + \partial ( R ^ { j } - V ( s _ { i } ; \theta _ { v } ^ { j } ) ) ^ { 2 } / \partial \theta _ { v } ^ { j }$   
+24: end for   
+25: end for   
+26: until $T \geq T _ { m a x }$
+
+![](images/9fd25f5ccbc9cf96954f6fdd9cbc44a1bd0277357795eda6b54b4793ba59308d.jpg)  
+Figure 12: Performance of A2MC on Atari games. The number in figure legend shows the average reward among the last 100 episodes upon 15 million timesteps and the percentage shows the performance margin as compared to ACKTR. The shaded region denotes the standard deviation over 2 random seeds.
\ No newline at end of file
diff --git a/parse/train/QpNz8r_Ri2Y/QpNz8r_Ri2Y.md b/parse/train/QpNz8r_Ri2Y/QpNz8r_Ri2Y.md
new file mode 100644
index 0000000000000000000000000000000000000000..2639fc0e5d4a28d3029b4d285c9a0804cb92d817
--- /dev/null
+++ b/parse/train/QpNz8r_Ri2Y/QpNz8r_Ri2Y.md
@@ -0,0 +1,582 @@
+# REPRESENTATION BALANCING OFFLINE MODEL-BASED REINFORCEMENT LEARNING
+
+Byung-Jun Lee1,3, Jongmin Lee1 & Kee-Eung $\mathbf { K i m } ^ { 1 , 2 }$
+
+1School of Computing, KAIST, Daejeon, Republic of Korea 2Graduate School of AI, KAIST, Daejeon, Republic of Korea 3Gauss Labs Inc., Seoul, Republic of Korea {bjlee,jmlee}@ai.kaist.ac.kr, kekim@kaist.ac.k
+
+# ABSTRACT
+
+One of the main challenges in offline and off-policy reinforcement learning is to cope with the distribution shift that arises from the mismatch between the target policy and the data collection policy. In this paper, we focus on a model-based approach, particularly on learning the representation for a robust model of the environment under the distribution shift, which has been first studied by Representation Balancing MDP (RepBM). Although this prior work has shown promising results, there are a number of shortcomings that still hinder its applicability to practical tasks. In particular, we address the curse of horizon exhibited by RepBM, rejecting most of the pre-collected data in long-term tasks. We present a new objective for model learning motivated by recent advances in the estimation of stationary distribution corrections. This effectively overcomes the aforementioned limitation of RepBM, as well as naturally extending to continuous action spaces and stochastic policies. We also present an offline model-based policy optimization using this new objective, yielding the state-of-the-art performance in a representative set of benchmark offline RL tasks.
+
+# 1 INTRODUCTION
+
+Reinforcement learning (RL) has accomplished remarkable results in a wide range of domains, but its successes were mostly based on a large number of online interactions with the environment. However, in many real-world tasks, exploratory online interactions are either very expensive or dangerous (e.g. robotics, autonomous driving, and healthcare), and applying a standard online RL would be impractical. Consequently, the ability to optimize RL agents reliably without online interactions has been considered as a key to practical deployment, which is the main goal of batch RL, also known as offline RL (Fujimoto et al., 2019; Levine et al., 2020).
+
+In an offline RL algorithm, accurate policy evaluation and reliable policy improvement are both crucial for the successful training of the agent. Evaluating policies in offline RL is essentially an off-policy evaluation (OPE) task, which aims to evaluate the target policy given the dataset collected from the behavior policy. The difference between the target and the behavior policies causes a distribution shift in the estimation, which needs to be adequately addressed for accurate policy evaluation. OPE itself is one of the long-standing hard problems in RL (Sutton et al., 1998; 2009; Thomas & Brunskill, 2016; Hallak & Mannor, 2017).
+
+However, recent offline RL studies mainly focus on how to improve the policy conservatively while using a common policy evaluation technique without much considerations for the distribution shift, e.g. mean squared temporal difference error minimization or maximum-likelihood training of environment model (Fujimoto et al., 2019; Kumar et al., 2019; Yu et al., 2020). While conservative policy improvement helps the policy evaluation by reducing the off-policyness, we hypothesize that addressing the distribution shift explicitly during the policy evaluation can further improve the overall performance, since it can provide a better foundation for policy improvement.
+
+To this end, we aim to explicitly address the distribution shift of the OPE estimator used in the offline RL algorithm. In particular, we focus on the model-based approach, where we train an environment model robust to the distribution shift. One of the notable prior works is Representation Balancing MDP (RepBM) (Liu et al., 2018b), which regularizes the representation learning of the model to be invariant between the distributions. However, despite the promising results by RepBM, its step-wise estimation of the distance between the distributions has a few drawbacks that limit the algorithm from being practical: not only it assumes a discrete-action task where the target policy is deterministic, but it also performs poorly in long-term tasks due to the curse of horizon of step-wise importance sampling (IS) estimators (Liu et al., 2018a).
+
+To address these limitations, we present the Representation Balancing with Stationary Distribution Estimation (RepB-SDE) framework, where we aim to learn a balanced representation by regularizing, in the representation space, the distance between the data distribution and the discounted stationary distribution induced by the target policy. Motivated by the recent advances in estimating stationary distribution corrections, we present a new representation balancing objective to train a model of the environment that no longer suffers from the curse of horizon. We empirically show that the model trained by the RepB-SDE objective is robust to the distribution shift for the OPE task, particularly when the difference between the target and the behavior is large. We also introduce a model-based offline RL algorithm based on the RepB-SDE framework and report its performance on the D4RL benchmark (Fu et al., 2020), showing the state-of-the-art performance in a representative set of tasks.
+
+# 2 RELATED WORK
+
+Learning balanced representation Learning a representation invariant to specific aspects of data is an established method for overcoming distribution shift that arises in unsupervised domain adaptation (Ben-David et al., 2007; Zemel et al., 2013) and in causal inference from observational data (Shalit et al., 2017; Johansson et al., 2018). They have shown that imposing a bound on the generalization error under the distribution shift leads to the objective that learns a balanced representation such that the training and the test distributions look similar. RepBM (Liu et al., 2018b) can be seen as a direct extension to the sequential case, which encourages the representation to be invariant under the target and behavior policies in each timestep.
+
+Stationary distribution correction estimation (DICE) Step-wise importance sampling (IS) estimators (Precup, 2000) compute importance weights by taking the product of per-step distribution ratios. Consequently, these methods suffer from exponentially high variance in the lengths of trajectories, which is a phenomenon called the curse of horizon (Liu et al., 2018a). Recently, techniques of computing a stationary DIstribution Correction Estimation (DICE) have made remarkable progress that effectively addresses the curse of horizon (Liu et al., 2018a; Nachum et al., $2 0 1 9 \mathrm { a }$ ; Tang et al., 2020; Zhang et al., 2020; Mousavi et al., 2020). DICE has been also used to explicitly address the distribution shift in online model-free RL, by directly applying IS on the policy and action-value objectives (Liu et al., 2019; Gelada & Bellemare, 2019). We adopt one of the estimation techniques, DualDICE (Nachum et al., 2019a), to measure the distance between the stationary distribution and the data distribution in the representation space.
+
+Offline reinforcement learning There are extensive studies on improving standard online modelfree RL algorithms (Mnih et al., 2015; Lillicrap et al., 2016; Haarnoja et al., 2018) for stable learning in the offline setting. The main idea behind them is to conservatively improve policy by (1) quantifying the uncertainty of value function estimate, e.g. using bootstrapped ensembles (Kumar et al., 2019; Agarwal et al., 2020), or/and (2) constraining the optimized target policy to be close to the behavior policy (i.e. behavior regularization approaches) (Fujimoto et al., 2019; Kumar et al., 2019; Wu et al., 2019; Lee et al., 2020). A notable exception is AlgaeDICE (Nachum et al., 2019b), which implicitly uses DICE to regularize the discounted stationary distribution induced by the target policy to be kept inside of the data support, similar to this work.
+
+On the other hand, Yu et al. (2020) argued that the model-based approach can be advantageous due to its ability to generalize predictions on the states outside of the data support. They introduce MOPO (Yu et al., 2020), which uses truncated rollouts and penalized rewards for conservative policy improvement. MOReL (Kidambi et al., 2020) trains a state-action novelty detector and use it to penalize rewards in the data-sparse region. Matsushima et al. (2020), MOOSE (Swazinna et al., 2020)
+
+and MBOP (Argenson & Dulac-Arnold, 2020) guide their policy optimization using the behavior policy, similar to the behavior regularization approaches.
+
+Note that these aforementioned offline RL methods build on the standard approximate dynamic programming algorithm for action-value estimation (model-free) or on a maximum-likelihood environment model (model-based), without explicitly addressing the distribution shift in the estimator. In contrast, we augment the objective for model learning to obtain a robust model under the distribution shift, which is the first attempt for offline RL to the best of our knowledge.
+
+# 3 PRELIMINARIES
+
+A Markov Decision Process (MDP) is specified by a tuple $M = \langle S , A , T , R , d _ { 0 } , \gamma \rangle$ , consisting of state space $s$ , action space $\mathcal { A }$ , transition function $T : \mathcal { S } \times \mathcal { A }  \Delta ( \mathcal { S } )$ , reward function $R : \mathcal { S } \times \overline { { \mathcal { A } } } \to \Delta ( [ 0 , r _ { \operatorname* { m a x } } ] )$ , initial state distribution $d _ { 0 }$ , and discount rate $\gamma$ . In this paper, we mainly focus on continuous state space $S \subseteq \mathbb { R } ^ { d _ { s } }$ and conduct experiments on both discrete action spaces ${ \cal A } = \{ a _ { 0 } , . . . a _ { n _ { a } } \}$ and continuous action spaces $\mathcal { A } \subseteq \mathbb { R } ^ { d _ { a } ^ { \star } }$ . Given MDP $M$ and policy $\pi$ , which is a (stochastic) mapping from state to action, the trajectory can be generated in the form of $s _ { 0 } , a _ { 0 } , r _ { 0 } , s _ { 1 } , a _ { 1 } , r _ { 1 } , \ldots$ , where $s _ { 0 } \sim d _ { 0 }$ and for each timestep $t \geq 0$ , $a _ { t } \sim \pi ( s _ { t } )$ , $\boldsymbol { r } _ { t } \sim R ( s _ { t } , a _ { t } )$ , and $s _ { t + 1 } \sim T ( s _ { t } , a _ { t } )$ . The goal of RL is to optimize or evaluate a policy, based on the normalized expected discounted return: $\begin{array} { r } { R ^ { \pi } \triangleq ( 1 - \gamma ) \mathbb { E } _ { M , \pi } \left[ \sum _ { t = 0 } ^ { \infty } \gamma ^ { t } r _ { t } \right] } \end{array}$ .
+
+A useful and important concept throughout the paper is the discounted stationary distribution, which represents the long-term occupancy of states:
+
+$$
+d ^ { \pi } ( s , a ) \triangleq ( 1 - \gamma ) \sum _ { t = 0 } ^ { \infty } \gamma ^ { t } \operatorname* { P r } ( s _ { t } = s , a _ { t } = a | M , \pi ) .
+$$
+
+From the definition, it can be observed that $R ^ { \pi }$ can be obtained by $R ^ { \pi } = \mathbb { E } _ { ( s , a ) \sim d ^ { \pi } } \left[ r ( s , a ) \right]$
+
+Offline RL and off-policy evaluation In this paper, we focus on the offline $R L$ problem where the agent can only access a static dataset $\mathcal { D } = \{ ( \bar { s } _ { i } , a _ { i } , r _ { i } , s _ { i } ^ { \prime } ) \} _ { i = 1 } ^ { N }$ for the maximization of $R ^ { \pi }$ . We consider a behavior-agnostic setting where we do not have any knowledge of the data collection process. We denote the empirical distribution of the dataset by $\dot { d } ^ { \mathcal { D } }$ .
+
+Before improving policy, we first aim to better evaluate $R ^ { \pi }$ given a target policy $\pi$ and a static dataset $\mathcal { D }$ , which corresponds to an off-policy evaluation (OPE) problem. We mainly focus on a modelbased approach where the algorithm first estimates the unknown dynamics $\widehat { T } , \widehat { R }$ using the dataset $\mathcal { D }$ . This defines an approximate MDP $\widehat { M } = \langle S , A , \widehat { T } , \widehat { R } , d _ { 0 } , \gamma \rangle$ , with the approximate expected discounted return $\begin{array} { r } { \widehat { R } ^ { \pi } \triangleq ( 1 - \gamma ) \mathbb { E } _ { \widehat { M } , \pi } \left[ \sum _ { t = 0 } ^ { \infty } \gamma ^ { t } r _ { t } \right] } \end{array}$ obtained from $\widehat { M }$ . In this paper, we are interested in the MDP estimate $\widehat { M }$ that can effectively reduce the error in the evaluation of policy $\pi$ , $| R ^ { \pi } - { \widehat R } ^ { \pi } |$ . In order to do so, we need to learn a good representation of a model that results in a small OPE error. We assume a bijective representation function $\phi : \mathcal { S } \times \mathcal { A }  \mathcal { Z }$ where $\mathcal { Z } \subseteq \mathbb { R } ^ { d _ { z } }$ is the representation space. We define the transition and the reward models in terms of the representation function $\phi$ , i.e. $\bar { \widehat { T } } = \widehat { T } _ { z } \circ \phi$ and $\widehat { R } = \widehat { R } _ { z } \circ \phi$ . In practice, where we use a neural network for $\widehat { T }$ and $\widehat { R }$ , $z$ can be chosen to be the output of an intermediate hidden layer, making $\phi$ represented by lower layers and $\widehat { T } _ { z } , \widehat { R } _ { z }$ by the remaining upper layers. We define $d _ { \phi } ^ { \pi } ( z )$ the discounted stationary distribution on $\mathcal { Z }$ induced by $d ^ { \pi } ( s , a )$ under the representation function $z = \phi ( s , a )$ , and similarly for $d _ { \phi } ^ { { \cal D } } ( z )$ .
+
+# 4 REPRESENTATION BALANCING OFFLINE MODEL-BASED RL
+
+# 4.1 GENERALIZATION ERROR BOUND FOR MODEL-BASED OFF-POLICY EVALUATION
+
+We aim to construct a model $\widehat { M }$ from the dataset $\mathcal { D }$ , which can accurately evaluate the policy $\pi$ , by minimizing a good upper bound of policy evaluation error $| R ^ { \pi } - { \widehat { R } } ^ { \pi } |$ . We define the following point-wise model loss for notational convenience:
+
+$$
+\begin{array} { r } { \mathcal { E } _ { \phi , \widehat { R } _ { z } , \widehat { T } _ { z } } ( s , a ) = c _ { R } D _ { T V } \Big ( R ( r | s , a ) \Big | \Big | \widehat { R } _ { z } \big ( r | \phi ( s , a ) \big ) \Big ) + c _ { T } D _ { T V } \Big ( T ( s ^ { \prime } | s , a ) \Big | \Big | \widehat { T } _ { z } \big ( s ^ { \prime } | \phi ( s , a ) \big ) \Big ) , } \end{array}
+$$
+
+where $c _ { R } = 2 ( 1 - \gamma )$ and $c _ { T } = 2 \gamma r _ { \operatorname* { m a x } }$ . Then, we start by restating the simulation lemma (Kearns & Singh, 2002) to bound the policy evaluation error in terms of the point-wise model loss. The proof is available in Appendix A.
+
+Lemma 4.1. Given an MDP $M$ and its estimate $\widehat { M }$ with a bijective representation function $\phi _ { : }$ , i.e. $( \widehat { T } , \widehat { R } ) = \langle \widehat { T } _ { z } \circ \phi , \widehat { R } _ { z } \circ \phi \rangle$ , the policy evaluation error of a policy $\pi$ can be bounded by:
+
+$$
+\begin{array} { r } { \left| R ^ { \pi } - \widehat { R } ^ { \pi } \right| \leq \mathbb { E } _ { ( s , a ) \sim d ^ { \pi } } \left[ \mathcal { E } _ { \phi , \widehat { R } _ { z } , \widehat { T } _ { z } } ( s , a ) \right] } \end{array}
+$$
+
+The Lemma 4.1 has a natural interpretation: if the model error is small in the states frequently visited by following the policy $\pi$ , the resultant policy evaluation error will also be small. However, minimizing the RHS of Eq. (1) in the off-policy evaluation (OPE) task is generally intractable since the distribution $d ^ { \pi }$ is not directly accessible. Therefore, the common practice has been to construct a maximum-likelihood MDP using $\mathcal { D }$ while ignoring the distribution shift, but its OPE performance is not guaranteed.
+
+Instead, we will derive a tractable upper bound on the policy evaluation error by eliminating the direct dependence on $d ^ { \pi }$ in Eq. (1). To this end, we adopt the distance metric between two distributions over representations $d _ { \phi } ^ { \pi }$ and $d _ { \phi } ^ { \mathcal { D } }$ that can bound their difference in expectations, which is the Integral Probability Metric (IPM) (Muller, 1997): ¨
+
+$$
+\mathrm { I P M } _ { \mathcal { G } } ( p , q ) = \operatorname* { s u p } _ { g \in \mathcal { G } } \big | \mathbb { E } _ { z \sim p } [ g ( z ) ] - \mathbb { E } _ { z \sim q } [ g ( z ) ] \big | .
+$$
+
+where particular choices of $\mathcal { G }$ make the IPM equivalent to different well-known distances of distributions, e.g. total variation distance or Wasserstein distance (Sriperumbudur et al., 2009).
+
+Theorem 4.2. Given an MDP $M$ and its estimate $\widehat { M }$ with a bijective representation function $\phi$ , i.e. ${ \mathcal { G } } \triangleq \{ g : { \mathcal { Z } } \to \mathbb { R } \}$ $( \widehat { T } , \widehat { R } ) = \langle \widehat { T } _ { z } \circ \phi , \widehat { R } _ { z } \circ \phi \rangle$ b such that $\begin{array} { r } { \frac { 1 } { B _ { \phi } } \mathcal { E } _ { \phi , \widehat { R } _ { z } , \widehat { T } _ { z } } \left( \phi ^ { - 1 } ( \cdot ) \right) \in \mathcal { G } } \end{array}$ , assume that there exists a constant . Then, for any policy $B _ { \phi } > 0$ $\pi$ and , $a$ function class
+
+$$
+\left| R ^ { \pi } - \widehat { R } ^ { \pi } \right| \leq \mathbb { E } _ { ( s , a ) \sim d ^ { \mathcal { D } } } \Big [ \mathcal { E } _ { \phi , \widehat { R } _ { z } , \widehat { T } _ { z } } ( s , a ) \Big ] + B _ { \phi } \mathrm { I P M } _ { \mathcal { G } } ( d _ { \phi } ^ { \pi } , d _ { \phi } ^ { \mathcal { D } } )
+$$
+
+This theorem is an adaptation of Lemma 1 of Shalit et al. (2017) to an infinite horizon model-based policy evaluation and can be derived by the definition of $\mathrm { I P M } _ { \mathcal { G } } ( d _ { \phi } ^ { \pi } , d _ { \phi } ^ { D } )$ since it serves as an upper bound of the difference in the expectations of any function in $\mathcal { G }$ . The first term in Eq. (3) corresponds to the fitness to the data following $d ^ { \mathcal { D } }$ , while the second term serves as a regularizer. To see this, minimizing the second term would yield a near-constant representation function, which would be bad for the first term since it cannot distinguish states and actions well enough. It shows a natural trade-off between optimizing the model that fits data better and learning the representation that is invariant with respect to $d _ { \phi } ^ { \pi }$ and $d _ { \phi } ^ { \mathcal { D } }$ .
+
+Nevertheless, RHS of Eq. (3) still cannot be evaluated naively due to its dependence on $d ^ { \pi }$ in estimating the IPM. We address this challenge via a change of variable, which is known as a DualDICE trick (Nachum et al., 2019a). Define $\nu : \mathcal { Z } \to \mathbb { R }$ as an arbitrary function of state-action pairs that satisfies:
+
+$$
+\nu \big ( \phi ( s , a ) \big ) \triangleq g \big ( \phi ( s , a ) \big ) + \gamma \mathbb { E } _ { s ^ { \prime } \sim T ( s , a ) } \Big [ \nu \big ( \phi ( s ^ { \prime } , a ^ { \prime } ) \big ) \Big ] , \forall ( s , a ) \in \mathcal { S } \times \mathcal { A } .
+$$
+
+Then we can rewrite the IPM as:
+
+$$
+\begin{array} { r l } & { \displaystyle \mathrm { I P M } _ { \mathcal { G } } ( d _ { \phi } ^ { \pi } , d _ { \phi } ^ { \mathcal { D } } ) = \underset { g \in \mathcal { G } } { \operatorname* { s u p } } \left| \mathbb { E } _ { ( s , a ) \sim d ^ { \pi } } \Big [ g \big ( \phi ( s , a ) \big ) \Big ] - \mathbb { E } _ { ( s , a ) \sim d ^ { \mathcal { D } } } \Big [ g \big ( \phi ( s , a ) \big ) \Big ] \right| } \\ & { \displaystyle \quad = \operatorname* { s u p } _ { \nu \in \mathcal { F } } \left| ( 1 - \gamma ) \mathbb { E } \qquad \Big [ \nu \big ( \phi ( s , a ) \big ) \Big ] - \mathbb { E } _ { ( s , a , s ^ { \prime } ) \sim d ^ { \mathcal { D } } } \Big [ \nu \big ( \phi ( s , a ) \big ) - \gamma \nu \big ( \phi ( s ^ { \prime } , a ^ { \prime } ) \big ) \Big ] \right| , } \\ & { \displaystyle \quad \quad \quad \quad \quad \quad \quad \quad \quad \quad \quad \quad \quad \quad \quad \quad \quad \quad \quad \quad \quad \quad \quad \quad \quad \quad \quad \quad \quad \quad \quad \quad a ^ { \prime } \sim \pi ^ { \prime } } \end{array}
+$$
+
+In other words, we are now taking a supremum over the new function class $\mathcal { F }$ , which captures a function that returns the expected discounted sum of $g ( \phi ( s _ { t } , a _ { t } ) )$ following the policy $\pi$ in an MDP
+
+$M$ given an initial representation $z$ . While it is now difficult to choose $\mathcal { F }$ from $\mathcal { G }$ , Eq. (3) still can be kept valid by using a sufficiently rich function class for $\mathcal { F }$ .
+
+In this work, we choose $\mathcal { F }$ to be the family of functions in the unit ball in a reproducing kernel Hilbert space (RKHS) $\mathcal { H } _ { k }$ with the kernel $k$ , which allows the following closed-form formula (see Lemma A.3 in Appendix for details):
+
+$$
+\begin{array} { r l r } { \mathrm { P M } _ { \mathcal { G } } ( d _ { \phi } ^ { \pi } , d _ { \phi } ^ { \mathcal { D } } ) ^ { 2 } = \mathbb { E } _ { s _ { 0 } \sim d _ { 0 } , a _ { 0 } \sim \pi ( s _ { 0 } ) , ( s , a , s ^ { \prime } ) \sim d ^ { \mathcal { D } } , a ^ { \prime } \sim \pi ( s ^ { \prime } ) } } & { } & { ( 5 ) } \\ { \bar { s } _ { 0 } \sim d _ { 0 } , \bar { u } _ { 0 } \sim d _ { 0 } ) , ( \bar { s } _ { 0 } , \bar { u } , s ^ { \prime } ) \sim d ^ { \mathcal { D } } , a ^ { \prime } \sim \pi ( \bar { s } ^ { \prime } ) } & { } & { } \\ { k \big ( \phi ( s , a ) , \phi ( \bar { s } , \bar { a } ) \big ) + ( 1 - \gamma ) ^ { 2 } k \big ( \phi ( s _ { 0 } , a _ { 0 } ) , \phi ( \bar { s } _ { 0 } , \bar { a } _ { 0 } ) \big ) + \gamma ^ { 2 } k \big ( \phi ( s ^ { \prime } , a ^ { \prime } ) , \phi ( \bar { s } ^ { \prime } , \bar { a } ^ { \prime } ) \big ) } & { } & { } \\ { - 2 ( 1 - \gamma ) k \big ( \phi ( s _ { 0 } , a _ { 0 } ) , \phi ( \bar { s } , \bar { a } ) \big ) - 2 \gamma k \big ( \phi ( s ^ { \prime } , a ^ { \prime } ) , \phi ( \bar { s } , \bar { a } ) \big ) + 2 \gamma ( 1 - \gamma ) k \big ( \phi ( s ^ { \prime } , a ^ { \prime } ) , \phi ( \bar { s } _ { 0 } , \bar { a } _ { 0 } ) \big ) \bigg ] . } & { } & \end{array}
+$$
+
+This completes our derivation for the tractable upper bound of policy evaluation error (Eq. (3)), whose direct dependence on $d ^ { \pi }$ is eliminated by Eq. (5). Finally, we can train a model by minimizing the upper bound that encourages us to learn balanced representation while improving data fitness, where the trained model can readily provide a model-based OPE.
+
+The $\mathrm { I P M } _ { \mathcal { G } } ( d _ { \phi } ^ { \pi } , d _ { \phi } ^ { D } ) ^ { 2 }$ in Eq. (5) can be estimated via finite random samples, and we denote its sampled-based estimator as $\widehat { \mathrm { I P M } } ( d _ { \phi } ^ { \pi } , d _ { \phi } ^ { D } ) ^ { 2 }$ . We show in the following that, a valid upper bound can be established based on the sample-based estimators instead of the exact terms in the RHS of Eq. (3) under certain conditions.
+
+Theorem 4.3. Given an MDP $M$ , its estimate $\widehat { M }$ with a bijective representation function $\phi$ , i.e. $( \widehat { T } , \widehat { R } ) = \langle \widehat { T } _ { z } \circ \phi , \widehat { R } _ { z } \circ \phi \rangle$ , and an RKHS $\begin{array} { r } { \mathcal { H } _ { k } \quad \subset \quad ( \mathcal { Z } \quad \to \quad \mathbb { R } ) } \end{array}$ induced by a universal kernel $k$ such that $\mathrm { s u p } _ { z \in \mathcal { Z } } k ( z , z ) ~ = ~ \bar { k }$ , assume that $f _ { \phi , \widehat { R } _ { z } , \widehat { T } _ { z } } \left( \widehat { z } \right) \ =$ $\begin{array} { r } { \mathbb { E } _ { T , \pi } \Big [ \sum _ { t = 0 } ^ { \infty } \gamma ^ { t } \mathcal { E } _ { \phi , \widehat { R } _ { z } , \widehat { T } _ { z } } \big ( s _ { t } , a _ { t } \big ) \Big | \big ( s _ { 0 } , a _ { 0 } \big ) = \phi ^ { - 1 } \big ( z ) \Big ] \in \mathcal { H } _ { k } } \end{array}$ with $B _ { \phi } = \| f _ { \phi , \widehat { R } _ { z } , \widehat { T } _ { z } } \| _ { \mathcal { H } _ { k } }$ and the loss is bounded by $\begin{array} { r } { \bar { \mathcal { E } } = \operatorname* { s u p } _ { s \in { \mathcal { S } } , a \in \mathcal { A } } \mathcal { E } _ { \phi , \widehat { R } _ { z } , \widehat { T } _ { z } } ( s , a ) } \end{array}$ . Let n be the number of data in $\mathcal { D }$ . With probability $1 - 2 \delta$ ,
+
+$$
+R ^ { \pi } - { \widehat R } ^ { \pi } \Big | \leq \frac { 1 } { n } \sum _ { ( s , a ) \in { \mathcal D } } \mathcal { E } _ { \phi , { \widehat R } _ { z } , { \widehat T } _ { z } } ( s , a ) + B _ { \phi } \widehat { \Gamma \mathrm { M } } \big ( d _ { \phi } ^ { \pi } , d _ { \phi } ^ { { D } } \big ) + \sqrt { \frac { \bar { \mathcal E } ^ { 2 } } { 2 n } \log \frac { 1 } { \delta } } + B _ { \phi } \sqrt { \frac { \bar { k } } { n } } \left( 4 + \sqrt { 8 \log \frac { 3 } { \delta } } \right) .
+$$
+
+This result can be proved by adapting the convergence results of the empirical estimate of the MMD (Gretton et al., 2012) and Hoeffding’s inequality (Hoeffding, 1963). With the choice of an RKHS $\mathcal { H } _ { k }$ , we can now interpret $B _ { \phi }$ as the RKHS norm $\| f _ { \phi , \widehat { R } _ { z } , \widehat { T } _ { z } } \| _ { \mathcal { H } _ { k } }$ , which captures the magbnitude and the smoothness of the expected cumulative model loss $f _ { \phi , \widehat { R } _ { z } , \widehat { T } _ { z } }$ . In general, assuming smooth underlying dynamics, we can expect $B _ { \phi }$ to be small when the model error is small. Furthermore, although $\bar { k }$ depends on the kernel function we use, we can always let √ $\bar { k } = 1$ and subsume it into $B _ { \phi }$ as long as it is bounded, i.e. using $\tilde { B } _ { \phi } \triangleq B _ { \phi } \sqrt { \bar { k } }$ . In the next section, we develop algorithms based on practical approximations of Eq. (3).
+
+Detailed comparison to RepBM As previously stated, RepBM (Liu et al., 2018b) is a modelbased finite-horizon OPE algorithm that trains the model to have balanced representation $\phi$ , which is encouraged to be invariant under the target and behavior policies. Specifically, given the deterministic target policy $\pi$ and the behavior policy $\mu$ , at each timestep $t$ , it defines the factual distribution on $\mathcal { Z }$ given that the actions until timestep $t$ have been executed according to the policy $\pi$ : $p _ { \phi , t } ^ { F } ( z ) = \bar { \mathrm { P r } } ( z _ { t } | M , \mu , a _ { 0 : t } = \pi ( s _ { 0 : t } ) )$ and the counterfactual distribution on $\mathcal { Z }$ given the same condition except the action at timestep $t$ : $p _ { \phi , t } ^ { C F } ( z ) = \mathrm { P r } ( z _ { t } | M , \mu , a _ { 0 : t - 1 } = \pi ( s _ { 0 : t - 1 } ) , a _ { t } \neq \pi ( s _ { t } ) )$ . Then, RepBM bounds the OPE error as,1
+
+$$
+R ^ { \pi } - \widehat { R } ^ { \pi } \Big | \leq ( 1 - \gamma ) \sum _ { t = 0 } ^ { \infty } \gamma ^ { t } \left( \mathbb { E } _ { p ( s _ { t } | M , \mu , a _ { 0 : t } = \pi ( s _ { 0 : t } ) ) } [ \mathcal { E } _ { \phi , \widehat { R } _ { z } , \widehat { T } _ { z } } ( s _ { t } , \pi ( s _ { t } ) ) ] + B _ { \phi , t } [ \mathrm { P M } _ { \mathcal { G } _ { t } } ( p _ { \phi , t } ^ { F } , p _ { \phi , t } ^ { C F } ) \right) .
+$$
+
+Although RepBM achieves performance improvement over other OPE algorithms, we found a number of practical challenges: from the definition of the $\mathrm { I P M } _ { \mathcal { G } _ { t } } ( p _ { \phi , t } ^ { F } , p _ { \phi , t } ^ { C F } )$ , it requires a discrete-action environment and a deterministic policy $\pi$ , which cannot be met by many practical RL settings. In addition, since the sample-based estimation of $\mathrm { I P M } _ { \mathcal { G } _ { t } } ( p _ { \phi , t } ^ { F } , p _ { \phi , t } ^ { C F } )$ requires samples consistent with the policy $\pi$ , i.e. $a _ { 0 : t - 1 } = \pi ( s _ { 0 : t - 1 } )$ , the algorithm would reject exponentially many samples with respect to $t$ , which is the curse of horizon (Liu et al., 2018a). When there is a large difference between the behavior and the target policies in long-term tasks, their implementation becomes close to using the maximum likelihood objective, which can also be observed empirically in our experiments. In contrast, our work is free from the abovementioned practical limitations by performing balancing between the discounted stationary distribution $d ^ { \pi }$ and the data distribution $d ^ { \bar { D } }$ , leveraging the recent advances in stationary distribution correction estimation (i.e. the DualDICE trick) to overcome the difficulties pertinent to the expectation concerning $d ^ { \pi }$ required to evaluate the IPM in the objective.
+
+4.2 REPRESENTATION BALANCING WITH STATIONARY DISTRIBUTION ESTIMATION
+
+In the following, we describe algorithms for OPE and offline RL based on the practical approximations to Eq. (3), which we call the RepB-SDE framework.
+
+Objective for off-policy evaluation As we mentioned earlier, we aim to minimize the upper bound of OPE error $| R ^ { \pi } - { \widehat { R } } ^ { \pi } |$ specified in Theorem 4.2. To make the RHS of Eq. (3) tractable for optimization, we replace the intractable total variation distance with KL-divergence, which can be easily minimized by maximizing the data log-likelihood. We also replace the IPM with its samplebased estimator to obtain the learning objective:
+
+$$
+\operatorname* { m i n } _ { \phi , \widehat { R } _ { z } , \widehat { T } _ { z } } \frac { 1 } { n } \sum _ { \substack { ( s , a , s ^ { \prime } , r ) \in \mathcal { D } } } \Big [ \underbrace { - \log \widehat { R } _ { z } \big ( r | \phi ( s , a ) \big ) } _ { D _ { \mathrm { K L } } ( R ( r | s , a ) | | \widehat { R } _ { z } ( r | \phi ( s , a ) ) ) D _ { \mathrm { K L } } ( T ( s ^ { \prime } | s , a ) | | \widehat { T } _ { z } ( s ^ { \prime } | \phi ( s , a ) ) ) } \Big ] + \alpha _ { M } \widehat { \mathrm { I P M } } ( d _ { \phi } ^ { \pi } , d _ { \phi } ^ { \mathcal { D } } )
+$$
+
+The constant $B _ { \phi }$ in Theorem 4.2 depends on the function classes and cannot be estimated, and thus, we replace it with a tunable hyperparameter $\alpha _ { M }$ that balances between data fitness and representation invariance. Remark that $\alpha _ { M } = 0$ recovers the simple maximum-likelihood objective.
+
+By simulating the target policy $\pi$ under the environment model $( \widehat { T } , \widehat { R } )$ obtained by minimizing Eq. (6), it is possible to perform a model-based OPE that approximately minimizes the upper bound of OPE error.
+
+Objectives for offline model-based RL By rearranging Eq. (3), we have,
+
+$$
+\begin{array} { r } { { \cal R } ^ { \pi } \geq \widehat { \cal R } ^ { \pi } - \mathbb { E } _ { ( s , a ) \sim \mathcal { D } } \left[ \mathcal { E } _ { \phi , \widehat { \cal R } _ { z } , \widehat { T } _ { z } } ( s , a ) \right] - { \cal B } _ { \phi } \mathrm { I P M } _ { \mathcal { G } } ( d _ { \phi } ^ { \pi } , d _ { \phi } ^ { \mathcal { D } } ) . } \end{array}
+$$
+
+Then, we can maximize the RHS of Eq. (7) to get the model and the policy that maximizes the lower bound of true return $R ^ { \pi }$ . Similar to the derivation of Eq. (6), we replace the total variation distance by KL-divergence to obtain the following learning objectives:
+
+$$
+\begin{array} { r l } & { \mathcal { L } _ { \widehat { M } } ( \widehat { M } , \pi , \alpha _ { M } ) = \mathbb { E } _ { d ^ { p } } \Big [ - \log \widehat { R } _ { z } \big ( r | \phi ( s , a ) \big ) - \log \widehat { T } _ { z } \big ( s ^ { \prime } | \phi ( s , a ) \big ) \Big ] + \alpha _ { M } \mathrm { I P M } _ { \mathcal { G } } \big ( d _ { \phi } ^ { \pi } , d _ { \phi } ^ { { D } } \big ) , } \\ & { \mathcal { I } _ { \pi } ( \pi , \widehat { M } , \alpha _ { \pi } ) = \mathbb { E } _ { \widehat { M } , \pi } \left[ \displaystyle \sum _ { t = 0 } ^ { \infty } \gamma ^ { t } r _ { t } \right] - \alpha _ { \pi } \mathrm { I P M } _ { \mathcal { G } } \big ( d _ { \phi } ^ { \pi } , d _ { \phi } ^ { { D } } \big ) . } \end{array}
+$$
+
+where the expectation in Eq. (9) can be optimized using various model-based RL algorithms, e.g. with planning (Chua et al., 2018) or using a model-free learner (Janner et al., 2019). By iterating between the minimization of ${ \mathcal { L } } _ { \widehat { M } }$ with respect to $\widehat { M }$ and the maximization of $\mathcal { I } _ { \pi }$ with respect to $\pi$ cby stochastic gradient method, it is possible to perform offline model-based RL that approximately maximizes the lower bound of the true return $R ^ { \pi }$ .
+
+Implementation details Following the recent practice (Chua et al., 2018; Janner et al., 2019; Yu et al., 2020), we model the dynamics $( \widehat { T } , \widehat { R } )$ using a bootstrap ensemble of neural networks. To optimize a policy based on the objective, we perform full rollouts (until reaching the terminal states or maximum timesteps) using learned dynamics $( \widehat { T } , \widehat { R } )$ . During obtaining the rollouts, we pessimistically augment the estimated reward function using the penalty proportional to the bootstrapped uncertainty, which helped the algorithm to perform robustly. We suspect the difficulty in calculating accurate IPM estimation is what makes the additional pessimism beneficial. We store the generated experiences to a separate dataset $\widehat { \mathcal { D } }$ and update the policy $\pi$ with IPM-regularized soft actor-critic (SAC) (Haarnoja et al., 2018) using samples from both datasets $\mathcal { D } \cup \widehat { \mathcal { D } }$ similar to MBPO (Janner et al., 2019).
+
+![](images/f7e55e839858979ed3073f247288e2993a18881c7140dc8bb252ba65b110e1f2.jpg)  
+Figure 1: The OPE results of different model learning algorithms with varying off-policyness on the $\mathbf { X }$ -axis. The y-axis plots the normalized individual MSE on test trajectories where the performance of the baseline model is set to 1. The tasks used are infinite-horizon discounted environments $\gamma = 0 . 9 8$ (HIV), $\gamma = 0 . 9 9$ (others)), where we truncated at $t = 1 0 0 0$ . The experiments are repeated 200 times, and the error bars indicate $9 5 \%$ confidence intervals.
+
+Since the presented model objective requires a policy $\pi$ to perform a balancing, we initially trained the model and the policy using $\alpha _ { M } = \alpha _ { \pi } = 0$ : by ${ \widehat { \widetilde M } } _ { 0 } = \arg \operatorname* { m i n } _ { { \widehat M } } { \mathcal { L } } _ { { \widehat M } } ^ { - } ( { \widehat M } , \cdot , 0 )$ and $\pi _ { 0 } = \arg \operatorname* { m i n } _ { \pi } \mathcal { L } _ { \pi } ( \pi , \widehat { M } _ { 0 } , 0 )$ . Then, we retrained the model and the policy using $\pi _ { 0 }$ : $\widehat { M } _ { 1 } \ =$ $\mathrm { a r g m i n } _ { \widehat { M } } \mathcal { L } _ { \widehat { M } } ( \widehat { M } , \pi _ { 0 } , \alpha _ { M } )$ and $\begin{array} { r } { \pi _ { 1 } = \arg \operatorname* { m i n } _ { \pi } \mathcal { L } _ { \pi } ( \pi , \widehat { M } _ { 1 } , \alpha _ { \pi } ) } \end{array}$ for some non-negative $\alpha _ { M }$ and $\alpha _ { \pi }$ . c cWhile it is desirable to repeat the optimization of the model and the policy until convergence, we did not observe significant improvement after the first iteration and reported the performance of the policy after the first iteration, $\pi _ { 1 }$ .
+
+# 5 EXPERIMENTS
+
+We demonstrate the effectiveness of the RepB-SDE framework, by comparing the OPE performance of Eq. (6) to that of RepBM and evaluating the presented model-based offline RL algorithm on the benchmarks. The code used to produce the results is available online.2 A detailed description of the experiments can be found in Appendix B.
+
+# 5.1 MODEL-BASED OFF-POLICY EVALUATION
+
+For the sake of comparison with RepBM, we test our OPE algorithm on three continuous-state discrete-action tasks where the goal is to evaluate a deterministic target policy. We trained a suboptimal deterministic target policy $\pi$ and used an $\epsilon$ -greedy policy with various values of $\epsilon$ as the data collection policy. In each experiment, we trained environment models for a fixed number of epochs concerning three different objectives: simple maximum-likelihood baseline, step-wise representation balancing objective used in RepBM (Liu et al., 2018b), and the presented OPE objective of RepB-SDE (Eq. (6)). We measured the individual mean squared error (Liu et al., 2018b).
+
+The normalized error of each algorithm, relative to the error of the baseline valued at 1, is presented in Figure 1. We can observe that the presented objective of RepB-SDE can reduce the OPE error from the baseline significantly, outperforming RepBM in most of the cases. As the off-policyness between the policies () increases, representation balancing algorithms should more benefit compared to the maximum-likelihood baseline in principle. However, the result shows that the performance of RepBM merely increases due to the increased sample rejection rate under large $\epsilon$ .
+
+Table 1: Normalized scores on D4RL MuJoCo benchmark datasets (Fu et al., 2020) where the score of 0 corresponds to a random policy and 100 corresponds to a converged SAC policy. All results (except MF, which is taken from Fu et al. (2020) and Kumar et al. (2020)) are averaged over 5 runs, where $\pm$ denotes the standard error. The highest scores are highlighted with boldface.   
+
+<table><tr><td>Dataset type</td><td>Environment</td><td>RepB-SDE (ours)</td><td>RP</td><td>Base</td><td>MOPO</td><td>MF</td><td>BC</td></tr><tr><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td></tr><tr><td>Random</td><td>Walker2d</td><td>21.1 ± 1.0</td><td>18.4</td><td>16.4</td><td>1.3</td><td>7.3 BEAR</td><td>0.0</td></tr><tr><td>Medium</td><td>Walker2d</td><td>72.1 ± 1.9</td><td>56.3</td><td>5.5</td><td>-0.1</td><td>81.1 BRAC</td><td>7.7</td></tr><tr><td>Med-Replay</td><td>Walker2d</td><td>49.8 ± 11.4</td><td>41.1</td><td>6.2</td><td>47.8</td><td>26.7 CQL</td><td>8.0</td></tr><tr><td>Med-Expert</td><td>Walker2d</td><td>88.8 ± 6.9</td><td>72.6</td><td>51.7</td><td>32.4</td><td>98.7 CQL</td><td>3.3</td></tr><tr><td>Random</td><td>Hopper</td><td>8.6 ±1.0</td><td>8.3</td><td>8.3</td><td>9.1</td><td>12.2 BRAC</td><td>9.0</td></tr><tr><td>Medium</td><td>Hopper</td><td>34.0 ± 2.8</td><td>27.5</td><td>19.8</td><td>19.2</td><td>58.0 CQL</td><td>34.5</td></tr><tr><td>Med-Replay</td><td>Hopper</td><td>62.2 ± 6.7</td><td>49.8</td><td>32.9</td><td>80.8</td><td>48.6 CQL</td><td>13.2</td></tr><tr><td>Med-Expert</td><td>Hopper</td><td>82.6 ± 7.0</td><td>74.0</td><td>19.1</td><td>23.2</td><td>111.0 CQL</td><td>41.9</td></tr><tr><td>Random</td><td>HalfCheetah</td><td>32.9 ± 1.1</td><td>31.3</td><td>26.1</td><td>29.9</td><td>35.4 CQL</td><td>-2.4</td></tr><tr><td>Medium</td><td>HalfCheetah</td><td>49.1 ± 0.3</td><td>47.3</td><td>22.1</td><td>45.0</td><td>46.3 BRAC</td><td>31.0</td></tr><tr><td>Med-Replay</td><td>HalfCheetah</td><td>57.5 ± 0.8</td><td>53.6</td><td>54.0</td><td>53.8</td><td>47.7 BRAC</td><td>14.7</td></tr><tr><td>Med-Expert</td><td>HalfCheetah</td><td>55.4 ± 8.3</td><td>36.7</td><td>31.9</td><td>91.0</td><td>64.7 BCQ</td><td>29.5</td></tr></table>
+
+5.2 OFFLINE REINFORCEMENT LEARNING WITH REPRESENTATION BALANCED MODEL
+
+We evaluate the offline model-based RL algorithm presented in Section 4.2 on a subset of datasets in the D4RL benchmark (Fu et al., 2020): using four types of datasets (Random, Medium, MediumReplay, and Medium-Expert) from three different MuJoCo environments (HalfCheetah-v2, Hopperv2, and Walker2d-v2) (Todorov et al., 2012). Random dataset contains $1 0 ^ { 6 }$ experience tuples from a random policy. Medium dataset contains $1 0 ^ { 6 }$ experience tuples from a policy trained to approximately $1 / 3$ the performance of the expert, which is an agent trained to completion with SAC. Med-Replay dataset contains $1 0 ^ { 5 }$ $2 \times \mathrm { i 0 ^ { 5 } }$ for Walker2d-v2) experience tuples, which are from the replay buffer of a policy trained up to the performance of the medium agent. Med-Expert dataset is a Medium dataset combined with $1 0 ^ { 6 }$ samples from the expert. This experimental setting exactly follows that of (Yu et al., 2020; Argenson & Dulac-Arnold, 2020).
+
+The normalized score of each algorithm is presented in Table 1. MF denotes the best score from offline model-free algorithms (taken from Fu et al. (2020) and Kumar et al. (2020)), including SAC (Haarnoja et al., 2018), BCQ (Fujimoto et al., 2019), BEAR (Kumar et al., 2019), BRAC (Wu et al., 2019), AWR (Peng et al., 2019), cREM (Agarwal et al., 2020), AlgaeDICE (Nachum et al., 2019b), and CQL (Kumar et al., 2020). The actual algorithm that achieves the reported score is presented next to the numbers. Base shows the performance of the most naive baseline, which attempts to maximize the estimated policy return under the maximum-likelihood model. RP denotes the performance of Base equipped with the appropriate reward penalty using the bootstrapped uncertainty of the model, which is equivalent to $\pi _ { 0 }$ described in Section 4.2. RepB-SDE denotes the performance after a single iteration of our algorithm, corresponding to $\pi _ { 1 }$ . We also provide BC, the performance of direct behavior cloning from the data, and MOPO (Yu et al., 2020), an offline model-based RL algorithm that optimizes policy based on truncated rollouts with the heuristic reward penalty.
+
+The significant gap between RepB-SDE and RP in the results shows the advantage brought by our framework that encourages balanced representation. While our approach was less successful on some of the datasets (mostly on the Hopper-v2 environment), we hypothesize that the conservative training techniques: the behavior regularization approaches exploited in the model-free algorithms, the rollout truncation technique in MOPO, and the pessimistic training based on the bootstrapped uncertainty estimates adopted in our algorithm exhibit their strengths in different datasets. For example, it may be the case that the ensemble models are overconfident especially in Hopper-v2, and should be regularized with more explicit methods. Nevertheless, we emphasize that the presented framework can be used jointly with any other conservative training technique to improve their performance.
+
+# 6 CONCLUSION AND FUTURE WORK
+
+In this paper, we presented RepB-SDE, a framework for balancing the model representation with stationary distribution estimation, aiming at obtaining a model robust to the distribution shift that arises in off-policy and offline RL. We started from the theoretical observation that the model-based policy evaluation error can be upper-bounded by the data fitness and the distance between two distributions in the representation space. Motivated by the bound, we presented a model learning objective for off-policy evaluation and model-based offline policy optimization. RepB-SDE can be seen as an extension of RepBM, which addresses the curse of horizon by leveraging the recent advances in stationary distribution correction estimation (i.e. the DualDICE trick). Using stationary distribution also frees us from other limitations of RepBM to be applied to more practical settings. To the best of our knowledge, it is the first attempt to introduce an augmented objective for the learning of model robust to a specific distribution shift in offline RL.
+
+In the experiments, we empirically demonstrated that we can significantly reduce the OPE error from the baseline, outperforming RepBM in most cases. We also showed that the robust model also helps in the offline model-based policy optimization, yielding the state-of-the-art performance in a representative set of D4RL benchmarks. We emphasize that our approach can be directly adopted in many other model-based offline RL algorithms.
+
+There are a number of promising directions for future work. Most importantly, we have not leveraged the learned representation in the policy when optimizing the policy, yet it is very natural to do so. We can easily incorporate the representation into the policy by assuming energy-based models, but this would make the computation of entropy intractable in entropy-regularized policy optimization algorithms. It would be also interesting to see if the proposed framework for learning balanced representation can benefit off-policy (and offline) model-free methods.
+
+# ACKNOWLEDGMENTS
+
+This work was supported by the National Research Foundation (NRF) of Korea (NRF2019M3F2A1072238 and NRF-2019R1A2C1087634), and the Ministry of Science and Information communication Technology (MSIT) of Korea (IITP No. 2019-0-00075, IITP No. 2020-0-00940 and IITP No. 2017-0-01779 XAI).
+
+# REFERENCES
+
+Rishabh Agarwal, Dale Schuurmans, and Mohammad Norouzi. An optimistic perspective on offline reinforcement learning. In International Conference on Machine Learning, 2020.
+
+Arthur Argenson and Gabriel Dulac-Arnold. Model-based offline planning. arXiv preprint arXiv:2008.05556, 2020.
+
+Peter L Bartlett and Shahar Mendelson. Rademacher and Gaussian complexities: Risk bounds and structural results. Journal of Machine Learning Research, 3(Nov):463–482, 2002.
+
+Shai Ben-David, John Blitzer, Koby Crammer, and Fernando Pereira. Analysis of representations for domain adaptation. In Advances in Neural Information Processing Systems, 2007.
+
+Greg Brockman, Vicki Cheung, Ludwig Pettersson, Jonas Schneider, John Schulman, Jie Tang, and Wojciech Zaremba. OpenAI Gym. arXiv preprint arXiv:1606.01540, 2016.
+
+Kurtland Chua, Roberto Calandra, Rowan McAllister, and Sergey Levine. Deep reinforcement learning in a handful of trials using probabilistic dynamics models. In Advances in Neural Information Processing Systems, 2018.
+
+Justin Fu, Aviral Kumar, Ofir Nachum, George Tucker, and Sergey Levine. D4RL: Datasets for deep data-driven reinforcement learning. arXiv preprint arXiv:2004.07219, 2020.
+
+Scott Fujimoto, David Meger, and Doina Precup. Off-policy deep reinforcement learning without exploration. In International Conference on Machine Learning, 2019.
+
+Carles Gelada and Marc G Bellemare. Off-policy deep reinforcement learning by bootstrapping the covariate shift. In AAAI Conference on Artificial Intelligence, 2019.
+
+Arthur Gretton, Karsten M Borgwardt, Malte J Rasch, Bernhard Scholkopf, and Alexander Smola.¨ A kernel two-sample test. Journal of Machine Learning Research, 13(Mar):723–773, 2012.
+
+Tuomas Haarnoja, Aurick Zhou, Pieter Abbeel, and Sergey Levine. Soft actor-critic: Off-policy maximum entropy deep reinforcement learning with a stochastic actor. In International Conference on Machine Learning, 2018.
+
+Assaf Hallak and Shie Mannor. Consistent on-line off-policy evaluation. In International Conference on Machine Learning, 2017.
+
+Wassily Hoeffding. Probability inequalities for sums of bounded random variables. Journal of the American Statistical Association, 58(301):13–30, 1963.
+
+Michael Janner, Justin Fu, Marvin Zhang, and Sergey Levine. When to trust your model: Modelbased policy optimization. In Advances in Neural Information Processing Systems, 2019.
+
+Nan Jiang and Lihong Li. Doubly robust off-policy value evaluation for reinforcement learning. In International Conference on Machine Learning, 2016.
+
+Fredrik D Johansson, Nathan Kallus, Uri Shalit, and David Sontag. Learning weighted representations for generalization across designs. arXiv preprint arXiv:1802.08598, 2018.
+
+Michael Kearns and Satinder Singh. Near-optimal reinforcement learning in polynomial time. Machine learning, 49(2-3):209–232, 2002.
+
+Rahul Kidambi, Aravind Rajeswaran, Praneeth Netrapalli, and Thorsten Joachims. MOReL: Modelbased offline reinforcement learning. In Advances in Neural Information Processing Systems, 2020.
+
+Diederik P Kingma and Jimmy Ba. Adam: A method for stochastic optimization. arXiv preprint arXiv:1412.6980, 2014.
+
+Aviral Kumar, Justin Fu, Matthew Soh, George Tucker, and Sergey Levine. Stabilizing off-policy q-learning via bootstrapping error reduction. In Advances in Neural Information Processing Systems, 2019.
+
+Aviral Kumar, Aurick Zhou, George Tucker, and Sergey Levine. Conservative Q-learning for offline reinforcement learning. In Advances in Neural Information Processing Systems, 2020.
+
+Byung-Jun Lee, Jongmin Lee, Peter Vrancx, Dongho Kim, and Kee-Eung Kim. Batch reinforcement learning with hyperparameter gradients. In International Conference on Machine Learning, 2020.
+
+Sergey Levine, Aviral Kumar, George Tucker, and Justin Fu. Offline reinforcement learning: Tutorial, review, and perspectives on open problems. arXiv preprint arXiv:2005.01643, 2020.
+
+Timothy P Lillicrap, Jonathan J Hunt, Alexander Pritzel, Nicolas Heess, Tom Erez, Yuval Tassa, David Silver, and Daan Wierstra. Continuous control with deep reinforcement learning. In International Conference on Learning Representations, 2016.
+
+Qiang Liu, Lihong Li, Ziyang Tang, and Dengyong Zhou. Breaking the curse of horizon: Infinitehorizon off-policy estimation. In Advances in Neural Information Processing Systems, 2018a.
+
+Yao Liu, Omer Gottesman, Aniruddh Raghu, Matthieu Komorowski, Aldo A Faisal, Finale DoshiVelez, and Emma Brunskill. Representation balancing MDPs for off-policy policy evaluation. In Advances in Neural Information Processing Systems, 2018b.
+
+Yao Liu, Adith Swaminathan, Alekh Agarwal, and Emma Brunskill. Off-policy policy gradient with state distribution correction. arXiv preprint arXiv:1904.08473, 2019.
+
+Tatsuya Matsushima, Hiroki Furuta, Yutaka Matsuo, Ofir Nachum, and Shixiang Gu. Deploymentefficient reinforcement learning via model-based offline optimization. arXiv preprint arXiv:2006.03647, 2020.
+
+Colin McDiarmid. On the method of bounded differences. Surveys in combinatorics, 141(1):148– 188, 1989.
+
+Volodymyr Mnih, Koray Kavukcuoglu, David Silver, Andrei A Rusu, Joel Veness, Marc G Bellemare, Alex Graves, Martin Riedmiller, Andreas K Fidjeland, Georg Ostrovski, et al. Human-level control through deep reinforcement learning. Nature, 518(7540):529–533, 2015.
+
+Ali Mousavi, Lihong Li, Qiang Liu, and Denny Zhou. Black-box off-policy estimation for infinitehorizon reinforcement learning. In International Conference on Learning Representations, 2020.
+
+Alfred Muller. Integral probability metrics and their generating classes of functions. ¨ Advances in Applied Probability, 29(2):429–443, 1997.
+
+Ofir Nachum, Yinlam Chow, Bo Dai, and Lihong Li. DualDICE: Behavior-agnostic estimation of discounted stationary distribution corrections. In Advances in Neural Information Processing Systems, 2019a.
+
+Ofir Nachum, Bo Dai, Ilya Kostrikov, Yinlam Chow, Lihong Li, and Dale Schuurmans. AlgaeDICE: Policy gradient from arbitrary experience. arXiv preprint arXiv:1912.02074, 2019b.
+
+Xue Bin Peng, Aviral Kumar, Grace Zhang, and Sergey Levine. Advantage-weighted regression: Simple and scalable off-policy reinforcement learning. arXiv preprint arXiv:1910.00177, 2019.
+
+Doina Precup. Eligibility traces for off-policy policy evaluation. Computer Science Department Faculty Publication Series, pp. 80, 2000.
+
+Prajit Ramachandran, Barret Zoph, and Quoc V Le. Swish: a self-gated activation function. arXiv preprint arXiv:1710.05941, 7, 2017.
+
+John Schulman, Filip Wolski, Prafulla Dhariwal, Alec Radford, and Oleg Klimov. Proximal policy optimization algorithms. arXiv preprint arXiv:1707.06347, 2017.
+
+Uri Shalit, Fredrik D Johansson, and David Sontag. Estimating individual treatment effect: generalization bounds and algorithms. In International Conference on Machine Learning, 2017.
+
+Bharath K Sriperumbudur, Kenji Fukumizu, Arthur Gretton, Bernhard Scholkopf, and Gert RG ¨ Lanckriet. On integral probability metrics, $\phi$ -divergences and binary classification. arXiv preprint arXiv:0901.2698, 2009.
+
+Richard S Sutton, Andrew G Barto, et al. Introduction to reinforcement learning, volume 135. MIT press Cambridge, 1998.
+
+Richard S Sutton, Hamid R Maei, and Csaba Szepesvari. A convergent ´ $o ( n )$ temporal-difference algorithm for off-policy learning with linear function approximation. In Advances in neural information processing systems, 2009.
+
+Phillip Swazinna, Steffen Udluft, and Thomas Runkler. Overcoming model bias for robust offline deep reinforcement learning. arXiv preprint arXiv:2008.05533, 2020.
+
+Ziyang Tang, Yihao Feng, Lihong Li, Dengyong Zhou, and Qiang Liu. Doubly robust bias reduction in infinite horizon off-policy estimation. In International Conference on Learning Representations, 2020.
+
+Philip Thomas and Emma Brunskill. Data-efficient off-policy policy evaluation for reinforcement learning. In International Conference on Machine Learning, 2016.
+
+Emanuel Todorov, Tom Erez, and Yuval Tassa. Mujoco: A physics engine for model-based control. In International Conference on Intelligent Robots and Systems, 2012.
+
+Yifan Wu, George Tucker, and Ofir Nachum. Behavior regularized offline reinforcement learning. arXiv preprint arXiv:1911.11361, 2019.
+
+Tianhe Yu, Garrett Thomas, Lantao Yu, Stefano Ermon, James Zou, Sergey Levine, Chelsea Finn, and Tengyu Ma. MOPO: Model-based offline policy optimization. In Advances in Neural Information Processing Systems, 2020.
+
+Rich Zemel, Yu Wu, Kevin Swersky, Toni Pitassi, and Cynthia Dwork. Learning fair representations. In International Conference on Machine Learning, 2013.
+
+Ruiyi Zhang, Bo Dai, Lihong Li, and Dale Schuurmans. GenDICE: Generalized offline estimation of stationary values. In International Conference on Learning Representations, 2020.
+
+# A PROOFS
+
+Lemma 4.1. Given an MDP $M$ and its estimate $\widehat { M }$ with a bijective representation function $\phi ,$ , i.e. $( \widehat { T } , \widehat { R } ) = \langle \widehat { T } _ { z } \circ \phi , \widehat { R } _ { z } \circ \phi \rangle$ , the policy evaluation error of a policy $\pi$ can be bounded by:
+
+$$
+\begin{array} { r } { \left| R ^ { \pi } - \widehat { R } ^ { \pi } \right| \leq \mathbb { E } _ { ( s , a ) \sim d ^ { \pi } } \left[ \mathcal { E } _ { \phi , \widehat { R } _ { z } , \widehat { T } _ { z } } ( s , a ) \right] } \end{array}
+$$
+
+Proof. We define a value function $V ^ { \pi } ( s ) = \mathbb { E } _ { M , \pi } [ \sum _ { t = 0 } ^ { \infty } \gamma ^ { t } r _ { t } | s _ { 0 } = s ]$ , the expected discounted return starting from the state $s$ , for the following proof. Due to its definition, we can write $R ^ { \pi } =$ $( 1 - \gamma ) \mathbb { E } _ { s \sim d _ { 0 } } \overline { { [ V ^ { \pi } ( s ) ] } }$ . The following recursive equation also holds:
+
+$$
+V ^ { \pi } ( s ) = \mathbb { E } _ { a \sim \pi ( s ) } \left[ r ( s , a ) + \gamma \mathbb { E } _ { s ^ { \prime } \sim T ( s , a ) } [ V ( s ^ { \prime } ) ] \right] ,
+$$
+
+and similarly with an approximate value function $\widehat { V } ^ { \pi } ( s ) = \mathbb { E } _ { \widehat { M } , \pi } [ \sum _ { t = 0 } ^ { \infty } \gamma ^ { t } r _ { t } | s _ { 0 } = s ]$ given an MDP estimate $\widehat { M }$ . Then,
+
+$$
+\begin{array} { r l } & { \displaystyle \frac { 1 } { 1 - \gamma } \Big | R ^ { \pi } - \widehat { R } ^ { \pi } \Big | = \left| \mathbb { E } _ { d _ { 0 } } \left[ V ^ { \pi } ( s _ { 0 } ) - \widehat { V } ^ { \pi } ( s _ { 0 } ) \right] \right| } \\ & { = \left| \mathbb { E } _ { d _ { 0 } , \pi } \left[ r _ { 0 } - \widehat { r } _ { 0 } \right] + \gamma \mathbb { E } _ { d _ { 0 } } \left[ \int \left( V ^ { \pi } ( s _ { 1 } ) T ( s _ { 1 } | s _ { 0 } , a _ { 0 } ) - \widehat { V } ^ { \pi } ( s _ { 1 } ) \widehat { T } ( s _ { 1 } | s _ { 0 } , a _ { 0 } ) \right) \pi ( a _ { 0 } | s _ { 0 } ) d s _ { 1 } d a _ { 0 } \right] \right| } \\ & { = \left| \mathbb { E } _ { d _ { 0 } , \pi } \left[ r _ { 0 } - \widehat { r } _ { 0 } \right] + \gamma \mathbb { E } _ { d _ { 0 } } \left[ \int \left( V ^ { \pi } ( s _ { 1 } ) T ( s _ { 1 } | s _ { 0 } , a _ { 0 } ) - \widehat { V } ^ { \pi } ( s _ { 1 } ) T ( s _ { 1 } | s _ { 0 } , a _ { 0 } ) \right. \right. } \\ & { \qquad \left. \left. + \widehat { V } ^ { \pi } ( s _ { 1 } ) T ( s _ { 1 } | s _ { 0 } , a _ { 0 } ) - \widehat { V } ^ { \pi } ( s _ { 1 } ) \widehat { T } ( s _ { 1 } | s _ { 0 } , a _ { 0 } ) \right) \pi ( a _ { 0 } | s _ { 0 } ) d s _ { 1 } d a _ { 0 } \right] \right| } \\ & { = \left| \mathbb { E } _ { d _ { 0 } , \pi } \left[ r _ { 0 } - \widehat { r } _ { 0 } \right] + \gamma \mathbb { E } _ { d _ { 0 } } \left[ \int \left( V ^ { \pi } ( s _ { 1 } ) - \widehat { V } ^ { \pi } ( s _ { 1 } ) \right) T ( s _ { 1 } | s _ { 0 } , a _ { 0 } ) \pi ( a _ { 0 } | s _ { 0 } ) d s _ { 1 } d a _ { 0 } \right] \right| } \end{array}
+$$
+
+$$
+\begin{array} { r l } & { \qquad + \gamma \mathbb { E } _ { d _ { 0 } } \bigg [ \int \hat { V } ^ { x } ( s _ { 1 } ) \bigg ( T ( s _ { 1 } | s _ { 0 } , a _ { 0 } ) - \hat { T } ( s _ { 1 } | s _ { 0 } , a _ { 0 } ) \bigg ) \pi ( a _ { 0 } | s _ { 0 } ) d s _ { 1 } d a _ { 0 } \bigg ] \bigg | } \\ & { = \left| \mathbb { E } _ { ( s , a ) \times d ^ { c } } \left[ \int \bigg ( R ( r | s , a ) - \hat { R } ( r | s , a ) \bigg ) d r + \gamma \int \hat { V } ^ { x } ( s ^ { \prime } ) \bigg ( T ( s ^ { \prime } | s , a ) - \hat { T } ( s ^ { \prime } | s , a ) \bigg ) d s ^ { \prime } \right] \right| } \\ & { \leq \mathbb { E } _ { ( s , a ) \sim d ^ { c } } \left[ \int \bigg | R ( r | s , a ) - \hat { R } ( r | s , a ) \bigg | d r + \gamma \int \hat { V } ^ { \tau } ( s ^ { \prime } ) \bigg | T ( s ^ { \prime } | s , a ) - \hat { T } ( s ^ { \prime } | s , a ) \bigg | d s ^ { \prime } \right] } \\ & { \leq \mathbb { E } _ { ( s , a ) \sim d ^ { c } } \left[ \int \bigg | R ( r | s , a ) - \hat { R } ( r | s , a ) \bigg | d r + \gamma \int \frac { r _ { \operatorname* { m a x } } } { 1 - \gamma } \bigg | T ( s ^ { \prime } | s , a ) - \hat { T } ( s ^ { \prime } | s , a ) \bigg | d s ^ { \prime } \right] } \\ & { = 2 \mathbb { E } _ { ( s , a ) \sim d ^ { c } } \left[ D _ { T V } \big ( R ( r | s , a ) \big | \hat { R } _ { s } ( r | \phi ( s , a ) ) \big ) \right] } \\ & { \qquad + \frac { 2 \gamma r _ { \operatorname* { m a x } } } { 1 - \gamma } \mathbb { E } _ { ( s , a ) \sim d ^ { c } } \left[ D _ { T V } \big ( T ( s ^ { \prime } | s , a ) \big | \hat { T } _ { 2 } ( s ^ { \prime } | \phi ( s , a ) ) \big ) \right] } \end{array}
+$$
+
+Theorem 4.2. Given an MDP $M$ and its estimate $\widehat { M }$ with a bijective representation function $\phi$ , i.e. $( \widehat { T } , \widehat { R } ) = \langle \widehat { T } _ { z } \circ \phi , \widehat { R } _ { z } \circ \phi \rangle$ , assume that there exists a constant $B _ { \phi } > 0$ and a function class ${ \mathcal { G } } \triangleq \{ g : { \mathcal { Z } } \to \mathbb { R } \}$ b such that $\begin{array} { r } { \frac { 1 } { B _ { \phi } } \mathcal { E } _ { \phi , \widehat { R } _ { z } , \widehat { T } _ { z } } \left( \phi ^ { - 1 } ( \cdot ) \right) \in \mathcal { G } . } \end{array}$ . Then, for any policy $\pi$ ,
+
+$$
+\left| R ^ { \pi } - \widehat { R } ^ { \pi } \right| \leq \mathbb { E } _ { ( s , a ) \sim d ^ { \mathcal { D } } } \Big [ \mathcal { E } _ { \phi , \widehat { R } _ { z } , \widehat { T } _ { z } } ( s , a ) \Big ] + B _ { \phi } \mathrm { I P M } _ { \mathcal { G } } ( d _ { \phi } ^ { \pi } , d _ { \phi } ^ { \mathcal { D } } )
+$$
+
+Proof. From Lemma 4.1, we directly have:
+
+$$
+\Big | R ^ { \pi } - \widehat { R } ^ { \pi } \Big | \leq \mathbb { E } _ { ( s , a ) \sim d ^ { D } } \Big [ \mathcal { E } _ { \phi , \widehat { R } _ { z } , \widehat { T } _ { z } } ( s , a ) \Big ] + \int \mathcal { E } _ { \phi , \widehat { R } _ { z } , \widehat { T } _ { z } } ( s , a ) \left( d ^ { \pi } ( s , a ) - d ^ { D } ( s , a ) \right) d s d a .
+$$
+
+Then,
+
+$$
+\begin{array} { r l } & { \displaystyle \int \mathcal { E } _ { \phi , \widehat { R } _ { + } , \widehat { T } _ { z } } ( s , a ) \left( d ^ { \pi } ( s , a ) - d ^ { D } ( s , a ) \right) d s d a } \\ & { \quad = \displaystyle \int \mathcal { E } _ { \phi , \widehat { R } _ { \epsilon } , \widehat { T } _ { z } } ( \phi ^ { - 1 } ( z ) ) \left( d ^ { \pi } ( z ) - d ^ { D } \left( \phi ^ { - 1 } ( \phi ^ { - 1 } ( z ) ) \right) \right) \left. \frac { d ( s , a ) } { d z } \right. d z } \\ & { \quad = \displaystyle \int \mathcal { E } _ { \phi , \widehat { R } _ { \epsilon } , \widehat { T } _ { z } } ( \phi ^ { - 1 } ( z ) ) \left( d _ { \phi } ^ { \pi } ( z ) - d _ { \phi } ^ { D } ( z ) \right) d z } \\ & { \quad \leq { B } _ { \phi } \displaystyle \left. \int \frac { 1 } { B _ { \phi } } \mathcal { E } _ { \phi , \widehat { R } _ { \epsilon } , \widehat { T } _ { z } } ( \phi ^ { - 1 } ( z ) ) \left( d _ { \phi } ^ { \pi } ( z ) - d _ { \phi } ^ { D } ( z ) \right) d z \right. \qquad \left( { B } _ { \phi } > 0 \right) } \\ & { \quad \leq { B } _ { \phi \in \nabla } \displaystyle \left. \int \mathcal { J } ( { \boldsymbol \jmath } ^ { \pi } ) \left( d _ { \phi } ^ { \pi } ( z ) - d _ { \phi } ^ { D } ( z ) \right) d z \right. \qquad \left( \frac { 1 } { B _ { \phi } } \mathcal { E } _ { \phi , \widehat { R } _ { \epsilon } , \widehat { T } _ { z } } ( \phi ^ { - 1 } ( z ) ) \in \mathcal { G } \right) } \\ & { \quad = { B } _ { \phi } \mathrm { F N M } _ { \phi } \left( a _ { \phi } ^ { \pi } , d _ { \phi } ^ { \theta } \right) } \end{array}
+$$
+
+We state some previous results, which are required for further proof.
+
+Theorem A.1 (McDiarmid’s inequality (McDiarmid, 1989)). Let $\{ X _ { i } \} _ { i = 1 } ^ { n }$ be independent random variables taking values in set $\mathcal { X }$ , and assume that $f : \mathcal { X } ^ { n }  \mathbb { R }$ satisfies
+
+$$
+\operatorname* { s u p } _ { \{ x _ { i } \} _ { i = 1 } ^ { n } \in \mathcal { X } ^ { n } , \tilde { x } \in \mathcal { X } } | f ( \{ x _ { i } \} _ { i = 1 } ^ { n } ) - f ( x _ { 1 } , . . . , x _ { i - 1 } , \tilde { x } , x _ { i + 1 } , . . . , x _ { n } ) | \leq c _ { i } .
+$$
+
+Then for every $\epsilon > 0$ ,
+
+$$
+{ \operatorname* { P r } } \{ f ( \{ X _ { i } \} _ { i = 1 } ^ { n } ) - \mathbb { E } _ { \{ X _ { i } \} _ { i = 1 } ^ { n } } [ f ( \{ X _ { i } \} _ { i = 1 } ^ { n } ) ] \geq \epsilon \} \leq \exp \left( - \frac { 2 \epsilon ^ { 2 } } { \sum _ { i = 1 } ^ { n } c _ { i } ^ { 2 } } \right) .
+$$
+
+Lemma A.2 (Rademacher complexity of RKHS (Bartlett & Mendelson, 2002)). Let $\mathcal { F }$ be a unit ball in a universal RKHS on the compact domain $\mathcal { X }$ , with kernel bounded according to $0 \leq k ( x , x ^ { \prime } ) \leq \bar { k }$ . Let $\{ x _ { i } \} _ { i = 1 } ^ { n }$ be an i.i.d. sample of size n drawn according to a probability measure $p$ on $\mathcal { X }$ , and let $\sigma _ { i }$ be i.i.d. and take values in $\{ - 1 , 1 \}$ with equal probability. The Rademacher complexity, which is defined as below, is upper bounded as:
+
+$$
+R _ { n } ( \mathcal { F } ) \triangleq \mathbb { E } _ { \{ x _ { i } \} _ { i = 1 } ^ { n } , \sigma } \operatorname* { s u p } _ { f \in \mathcal { F } } \left| \frac { 1 } { n } \sum _ { i = 1 } ^ { n } \sigma _ { i } f ( x _ { i } ) \right| \leq \sqrt { \frac { \bar { k } } { n } } .
+$$
+
+The upper bound is followed by Lemma 22 of (Bartlett & Mendelson, 2002).
+
+Now we prove the following using the results above.
+
+Lemma A.3. Let $\mathcal { H } _ { k }$ be a RKHS associated with universal kernel $k ( \cdot , \cdot )$ . Let $\langle \cdot , \cdot \rangle _ { \mathcal { H } _ { k } }$ be the inner product of $\mathcal { H } _ { k }$ , which satisfies the reproducing property $\nu ( z ) = \langle \nu , \dot { k } ( \cdot , z ) \rangle _ { \mathcal { H } _ { k } }$ . When $\mathcal { G }$ is chosen such that
+
+$$
+\begin{array} { r } { \boldsymbol { \operatorname { \mathrm { ? } } } = \biggl \{ g \in ( \mathcal { Z } \to \mathbb { R } ) : g ( z ) = \nu ( z ) - \gamma \mathbb { E } _ { s ^ { \prime } \sim T ( \phi ^ { - 1 } ( z ) ) } [ \nu ( \phi ( s ^ { \prime } , a ^ { \prime } ) ) ] , \nu \in ( \mathcal { Z } \to \mathbb { R } ) , \langle \nu , \nu \rangle _ { \mathcal { H } _ { k } } \leq 1 \biggr \} , } \end{array}
+$$
+
+the $\mathrm { I P M } _ { \mathcal { G } } ( d _ { \phi } ^ { \pi } , d _ { \phi } ^ { D } )$ has the following closed form definition:
+
+$$
+\begin{array} { r l } & { \mathrm { [ P M } _ { \mathcal { G } } ( d _ { \phi } ^ { \pi } , d _ { \phi } ^ { \mathcal { D } } ) ^ { 2 } = \mathbb { E } _ { s _ { 0 } \sim d _ { 0 } , a _ { 0 } \sim \pi ( s _ { 0 } ) , ( s , a , s ^ { \prime } ) \sim d ^ { \mathcal { D } } , a ^ { \prime } \sim \pi ( s ^ { \prime } ) } } \\ & { \quad \quad k ( \phi ( s , a ) , \phi ( \bar { s } , \bar { a } ) ) + ( 1 - \gamma ) ^ { 2 } k ( \phi ( s _ { 0 } , a _ { 0 } ) , \phi ( \bar { s } _ { 0 } , \bar { a } _ { 0 } ) ) + \gamma ^ { 2 } k ( \phi ( s ^ { \prime } , a ^ { \prime } ) , \phi ( \bar { s } ^ { \prime } , \bar { a } ^ { \prime } ) ) } \\ & { \quad - 2 ( 1 - \gamma ) k ( \phi ( s _ { 0 } , a _ { 0 } ) , \phi ( \bar { s } , \bar { a } ) ) - 2 \gamma k ( \phi ( s ^ { \prime } , a ^ { \prime } ) , \phi ( \bar { s } , \bar { a } ) ) + 2 \gamma ( 1 - \gamma ) k ( \phi ( s ^ { \prime } , a ^ { \prime } ) , \phi ( \bar { s } _ { 0 } , \bar { a } _ { 0 } ) ) \bigg ] . } \end{array}
+$$
+
+Furthermore, suppose that $\bar { k } \triangleq \operatorname* { s u p } _ { z \in \mathcal { Z } } k ( z , z )$ . The estimator $\widehat { \mathrm { I P M } } ( d _ { \phi } ^ { \pi } , d _ { \phi } ^ { D } ) ^ { 2 }$ , which is the samplebased estimation of $\mathrm { I P M } _ { \mathcal { G } } ( d _ { \phi } ^ { \pi } , d _ { \phi } ^ { \mathcal { D } } ) ^ { 2 }$ from $n$ samples, satisfies with probability at least $1 - \delta$ ,
+
+$$
+\left| \mathrm { I P M } _ { \mathcal { G } } ( d _ { \phi } ^ { \pi } , d _ { \phi } ^ { D } ) - \widehat { \mathrm { I P M } } ( d _ { \phi } ^ { \pi } , d _ { \phi } ^ { D } ) \right| \leq \sqrt { \frac { \bar { k } } { n } } \left( 4 + \sqrt { 8 \log \frac { 3 } { \delta } } \right) .
+$$
+
+Proof. In the below we write in shorthand, $\operatorname { I P M } _ { \mathcal { G } }$ to denote $\mathrm { I P M } _ { \mathcal { G } } ( d _ { \phi } ^ { \pi } , d _ { \phi } ^ { D } )$ . As in Eq. (4), we can rewrite the IPM term as:
+
+$$
+\mathrm { I P M } _ { \mathcal { G } } = \underset { \nu \in \mathcal { F } } { \operatorname* { s u p } } \left| ( 1 - \gamma ) \mathbb { E } _ { \underset { a \sim \pi ( s ) } { s \sim d _ { 0 } } } \left[ \nu ( \phi ( s , a ) ) \right] - \mathbb { E } _ { ( s , a , s ^ { \prime } ) \sim d ^ { \mathcal { D } } } \left[ \nu ( \phi ( s , a ) ) - \gamma \nu ( \phi ( s ^ { \prime } , a ^ { \prime } ) ) \right] \right| ,
+$$
+
+and $\mathcal { F }$ here becomes $\mathcal { F } = \{ \nu \in ( \mathcal { Z }  \mathbb { R } ) : \langle \nu , \nu \rangle _ { \mathcal { H } _ { k } } \leq 1 \}$ , a unit ball in RKHS $\mathcal { H } _ { k }$ . Using the reproducing property of $\mathcal { H } _ { k }$ :
+
+$$
+\begin{array} { r l } & { \mathrm { I P M } _ { \mathcal { G } } ^ { 2 } = \Bigg \{ \underset { s \in \mathcal { F } } { \operatorname* { s u p } } \Bigg \{ ( 1 - \gamma ) \mathbb { E } _ { \underset { \sigma \sim \infty } { s \operatorname* { s d } } } \left[ \nu ( \phi s , a ) \right] \big ] - \mathbb { E } _ { ( s , a , s ^ { \prime } ) \sim d } \mathbb { P } \big [ \nu ( \phi ( s , a ) ) - \gamma \nu ( \phi ( s ^ { \prime } , a ^ { \prime } ) ) \big ] \Bigg \} ^ { 2 } } \\ & { \qquad = \underset { s \in \mathcal { F } } { \operatorname* { s u p } } \Bigg \{ ( 1 - \gamma ) \mathbb { E } _ { \underset { \sigma \sim \infty } { s \operatorname* { s u p } } } \big [ \nu ( \phi ( s , a ) ) \big ] - \mathbb { E } _ { ( s , a , s ^ { \prime } ) \sim d } \mathbb { P } \big [ \nu ( \phi ( s , a ) ) - \gamma \nu ( \phi ( s ^ { \prime } , a ^ { \prime } ) ) \big ] \Bigg \} ^ { 2 } } \\ & { \qquad = \underset { s \in \mathcal { F } } { \operatorname* { s u p } } \Bigg \{ ( 1 - \gamma ) \mathbb { E } _ { \underset { \sigma \sim \infty } { s \operatorname* { s u p } } } \big [ ( \nu , k ( \cdot , \phi ( s , a ) ) ) u _ { k } \big ] } \\ & { \qquad \quad - \mathbb { E } _ { ( s , a , s ^ { \prime } ) \sim d } \mathbb { P } \big [ \nu , k ( \cdot , \cdot , \phi ( s , a ) ) \big ] u _ { k } - \gamma \big ( \nu , k ( \cdot , \phi ( s ^ { \prime } , a ^ { \prime } ) ) \big ) _ { k k } \bigg ] \Bigg \} ^ { 2 } } \\ & { \qquad - \mathbb { E } _ { ( s , a , s ^ { \prime } ) \sim d } \rho ^ { D } \left[ \big \langle \nu , k ( \cdot , \phi ( s , a ) ) \big \rangle _ { \mathcal { H } _ { k } } - \gamma \big \langle \nu , k ( \cdot , \phi ( s ^ { \prime } , a ^ { \prime } ) ) \big \rangle _ { \mathcal { H } _ { k } } \right] \Bigg \} ^ { 2 } } \\ & { \qquad = \underset { s \in \mathcal { F } } { \operatorname* { s u p } } \big \{ \nu , \nu ^ { * } \} _ { \mathcal { H } _ { k } } ^ { 2 } , } \end{array}
+$$
+
+where
+
+$$
+\begin{array} { r } { \nu ^ { * } ( \cdot ) = ( 1 - \gamma ) \mathbb { E } _ { \stackrel { s \sim d _ { 0 } } { a \sim \pi ( s ) } } \left[ k ( \cdot , \phi ( s , a ) ) \right] - \mathbb { E } _ { ( s , a , s ^ { \prime } ) \sim d ^ { \mathcal { D } } } \left[ k ( \cdot , \phi ( s , a ) ) - \gamma k ( \cdot , \phi ( s ^ { \prime } , a ^ { \prime } ) ) \right] . } \end{array}
+$$
+
+Due to the Cauchy-Schwarz inequality and $\langle \nu , \nu \rangle _ { \mathcal { H } _ { k } } \leq 1 , \forall \nu \in \mathcal { F }$ ,
+
+$$
+\langle \nu , \nu ^ { * } \rangle _ { \mathcal { H } _ { k } } ^ { 2 } \le \langle \nu , \nu \rangle _ { \mathcal { H } _ { k } } \langle \nu ^ { * } , \nu ^ { * } \rangle _ { \mathcal { H } _ { k } } \le \langle \nu ^ { * } , \nu ^ { * } \rangle _ { \mathcal { H } _ { k } }
+$$
+
+holds and $\begin{array} { r } { \operatorname* { s u p } _ { \nu \in \mathcal { F } } \langle \nu , \nu ^ { * } \rangle _ { \mathcal { H } _ { k } } ^ { 2 } = \langle \nu ^ { * } , \nu ^ { * } \rangle _ { \mathcal { H } _ { k } } } \end{array}$ . Using the property that $\langle k ( \cdot , z ) , k ( \cdot , \bar { z } ) \rangle _ { \mathcal { H } _ { k } } = k ( z , \bar { z } )$ we can derive the closed form expression in the lemma from $\langle \nu ^ { * } , \nu ^ { * } \rangle _ { \mathcal { H } _ { k } }$ .
+
+Now we prove the error bound of the estimator. First, we divide $\mathrm { I P M } _ { \mathcal { G } } ( d _ { \phi } ^ { \pi } , d _ { \phi } ^ { D } )$ into three parts:
+
+$$
+\begin{array} { r l } & { \mathrm { I P M } _ { \mathcal { G } } ( d _ { \phi } ^ { \pi } , d _ { \phi } ^ { D } ) = \underset { \nu \in \mathcal { F } } { \operatorname* { s u p } } | f _ { 1 } ( \nu ) + f _ { 2 } ( \nu ) + f _ { 3 } ( \nu ) | } \\ & { f _ { 1 } ( \nu ) = ( 1 - \gamma ) \mathbb { E } _ { s \sim d _ { 0 } , a \sim \pi ( s ) } [ \nu ( \phi ( s , a ) ) ] , \qquad f _ { 2 } ( \nu ) = - \mathbb { E } _ { ( s , a ) \sim d ^ { D } } \left[ \nu ( \phi ( s , a ) ) \right] , } \\ & { f _ { 3 } ( \nu ) = \mathbb { E } _ { ( s , a , s ^ { \prime } ) \sim d ^ { D } , a \cdot \sim \pi ( s ^ { \prime } ) } \left[ \gamma \nu ( \phi ( s ^ { \prime } , a ^ { \prime } ) ) \right] . } \end{array}
+$$
+
+Given $n$ samples $\{ s _ { 0 } ^ { ( i ) } , a _ { 0 } ^ { ( i ) } , s ^ { ( i ) } , a ^ { ( i ) } , s ^ { \prime ( i ) } , a ^ { \prime ( i ) } \} _ { i = 1 } ^ { n }$ from the generative process $s _ { 0 } ^ { ( i ) } \sim d _ { 0 } , a _ { 0 } ^ { ( i ) } \sim$ $\pi ( s _ { 0 } ^ { ( i ) } ) , ( s ^ { ( i ) } , a ^ { ( i ) } , s ^ { \prime ( i ) } ) ~ \sim ~ d ^ { D } , a ^ { \prime ( i ) } ~ \sim ~ \pi ( s ^ { \prime ( i ) } ) ~ \forall i ,$ , we define the sample-based estimator $\widehat { \mathrm { I P M } } ( d _ { \phi } ^ { \pi } , d _ { \phi } ^ { D } )$ :
+
+$$
+\begin{array} { l } { { \displaystyle \widehat { \mathrm { I P M } } ( d _ { \phi } ^ { \pi } , d _ { \phi } ^ { \mathcal { D } } ) ^ { 2 } = \frac { 1 } { n ^ { 2 } } \sum _ { i , j } \bigg [ k ( \phi ( s ^ { ( i ) } , a ^ { ( i ) } ) , \phi ( s ^ { ( j ) } , a ^ { ( j ) } ) ) + ( 1 - \gamma ) ^ { 2 } k ( \phi ( s _ { 0 } ^ { ( i ) } , a _ { 0 } ^ { ( i ) } ) , \phi ( s _ { 0 } ^ { ( j ) } , a _ { 0 } ^ { ( j ) } ) ) } } \\ { { \displaystyle \qquad + \gamma ^ { 2 } k ( \phi ( s ^ { \prime ( i ) } , a ^ { \prime ( i ) } ) , \phi ( s ^ { \prime ( j ) } , a ^ { \prime ( j ) } ) ) - 2 ( 1 - \gamma ) k ( \phi ( s _ { 0 } ^ { ( i ) } , a _ { 0 } ^ { ( i ) } ) , \phi ( s ^ { ( j ) } , a ^ { ( j ) } ) ) } } \\ { { \displaystyle \qquad - 2 \gamma k ( \phi ( s ^ { \prime ( i ) } , a ^ { \prime ( i ) } ) , \phi ( s ^ { ( j ) } , a ^ { ( j ) } ) ) + 2 \gamma ( 1 - \gamma ) k ( \phi ( s ^ { \prime ( i ) } , a ^ { \prime ( i ) } ) , \phi ( s _ { 0 } ^ { ( j ) } , a _ { 0 } ^ { ( j ) } ) ) \bigg ] . } } \end{array}
+$$
+
+By deriving in reverse order, we can recover its another definition in supremum, which can be divided into three parts:
+
+$$
+\begin{array} { r l } & { \widehat { \mathrm { I P M } } ( d _ { \phi } ^ { \pi } , d _ { \phi } ^ { 2 D } ) = \displaystyle \operatorname* { s u p } _ { \nu \in \mathcal { F } } \Big | \widehat { f } _ { 1 } ( \nu ) + \widehat { f } _ { 2 } ( \nu ) + \widehat { f } _ { 3 } ( \nu ) \Big | } \\ & { \mathrm { w h e r e } \widehat { f } _ { 1 } ( \nu ) = \displaystyle \frac { 1 - \gamma } { n } \sum _ { i = 1 } ^ { n } \nu \left( \phi \left( s _ { 0 } ^ { ( i ) } , a _ { 0 } ^ { ( i ) } \right) \right) , \widehat { f } _ { 2 } ( \nu ) = - \displaystyle \frac { 1 } { n } \sum _ { i = 1 } ^ { n } \nu \left( \phi \left( s ^ { ( i ) } , a ^ { ( i ) } \right) \right) , } \\ & { \widehat { f } _ { 3 } ( \nu ) = \displaystyle \frac { \gamma } { n } \sum _ { i = 1 } ^ { n } \nu \left( \phi \left( s ^ { \prime ( i ) } , a ^ { \prime ( i ) } \right) \right) . } \end{array}
+$$
+
+We can bound the error of sample-based estimator with individual errors as:
+
+$$
+\begin{array} { r l } { \mathrm { I P M } _ { \mathcal { G } } ( d _ { \phi } ^ { \pi } , d _ { \phi } ^ { 2 } ) - \widehat { \mathrm { I P M } } ( d _ { \phi } ^ { \pi } , d _ { \phi } ^ { \mathcal { D } } ) \Bigg | = \Bigg | \underset { \nu \in \mathcal { F } } { \operatorname* { s u p } } | f _ { 1 } ( \nu ) + f _ { 2 } ( \nu ) + f _ { 3 } ( \nu ) | - \underset { \nu \in \mathcal { F } } { \operatorname* { s u p } } | \widehat { f } _ { 1 } ( \nu ) + \widehat { f } _ { 2 } ( \nu ) + \widehat { f } _ { 3 } ( \nu ) | \Bigg | } & { } \\ { \leq \underset { \nu \in \mathcal { F } } { \operatorname* { s u p } } | | f _ { 1 } ( \nu ) + f _ { 2 } ( \nu ) + f _ { 3 } ( \nu ) | - | \widehat { f } _ { 1 } ( \nu ) + \widehat { f } _ { 2 } ( \nu ) + \widehat { f } _ { 3 } ( \nu ) | | } & { } \\ { \leq \underset { \nu \in \mathcal { F } } { \operatorname* { s u p } } | f _ { 1 } ( \nu ) + f _ { 2 } ( \nu ) + f _ { 3 } ( \nu ) - \widehat { f } _ { 1 } ( \nu ) - \widehat { f } _ { 2 } ( \nu ) - \widehat { f } _ { 3 } ( \nu ) | } & { } \\ { \leq \underset { \nu \in \mathcal { F } } { \operatorname* { s u p } } [ | f _ { 1 } ( \nu ) - \widehat { f } _ { 1 } ( \nu ) | + | f _ { 2 } ( \nu ) - \widehat { f } _ { 2 } ( \nu ) | + | f _ { 3 } ( \nu ) - \widehat { f } _ { 3 } ( \nu ) | ] } & { } \\ { \leq \underset { \nu \in \mathcal { F } } { \operatorname* { s u p } } [ | f _ { 1 } ( \nu ) - \widehat { f } _ { 1 } ( \nu ) | + | \widehat { f } _ { 2 } ( \nu ) - \widehat { f } _ { 2 } ( \nu ) | + | f _ { 3 } ( \nu ) - \widehat { f } _ { 3 } ( \nu ) | ] } & { } \\  \leq \underset { \nu \in \mathcal { F } } { \operatorname* { s u p } } | f _ { 1 } ( \nu ) - \widehat { f } _ { 1 } ( \nu ) | + \underset { \nu \in \mathcal { F } } { \operatorname* { s u p } } | f _   \end{array}
+$$
+
+We then observe that
+
+$$
+\begin{array} { l } { \displaystyle \underset { \nu \in \mathcal { F } } { \operatorname* { s u p } } \left| f _ { 1 } ( \nu ) - \widehat { f _ { 1 } } ( \nu ) \right| = ( 1 - \gamma ) \Biggl \{ \mathbb { E } _ { s \sim d _ { 0 } , a \sim \pi ( s ) } [ k ( \phi ( s , a ) , \phi ( \bar { s } , \bar { a } ) ) ] } \\ { \displaystyle - \frac { 2 } { n } \sum _ { i = 1 } ^ { n } \mathbb { E } _ { \bar { s } \sim d _ { 0 } , \bar { a } \sim \pi ( s ) } \left[ k \left( \phi \left( s ^ { ( i ) } , a ^ { ( i ) } \right) , \phi ( \bar { s } , \bar { a } ) \right) \right] } \\ { \displaystyle + \frac { 1 } { n ^ { 2 } } \sum _ { i = 1 } ^ { n } \sum _ { j = 1 } ^ { n } k \left( \phi \left( s ^ { ( i ) } , a ^ { ( i ) } \right) , \phi \left( s ^ { ( j ) } , a ^ { ( j ) } \right) \right) \Biggl \} ^ { 1 / 2 } , } \end{array}
+$$
+
+which shows that changing $s ^ { ( i ) } , a ^ { ( i ) }$ results in changes of ${ \mathrm { s u p } } _ { \nu \in { \mathcal { F } } } \left| f _ { 1 } ( \nu ) - { \widehat { f } } _ { 1 } ( \nu ) \right|$ in magnitude of at most $2 ( 1 - \gamma ) \bar { k } ^ { 1 / 2 } / n$ where ${ \bar { k } } = \operatorname* { s u p } _ { z \in { \mathcal { Z } } } k ( z , z )$ . Therefore, by McDiarmid’s inequality (Theorem A.1),
+
+$$
+\operatorname* { P r } \left\{ \operatorname* { s u p } _ { \nu \in \mathcal { F } } \left| f _ { 1 } ( \nu ) - \widehat f _ { 1 } ( \nu ) \right| - \mathbb { E } _ { s ^ { ( i ) } , a ^ { ( i ) } } \left[ \operatorname* { s u p } _ { \nu \in \mathcal { F } } \left| f _ { 1 } ( \nu ) - \widehat f _ { 1 } ( \nu ) \right| \right] \geq \epsilon \right\} \leq \exp \left( - \frac { n \epsilon ^ { 2 } } { 2 ( 1 - \gamma ) ^ { 2 } \bar { k } } \right) .
+$$
+
+Also,
+
+$$
+\begin{array} { r l } & { \mathbb { E } _ { s ^ { ( i ) } , \sigma ^ { ( k ) } } [ \underset { s ^ { ( r , F ) } } { \operatorname* { s u p } } [ f _ { 1 } ( \nu ) - \widehat { f } _ { 1 } ( \nu ) ] ] } \\ & { = \frac { 1 - \gamma } { n } \underset { s ^ { ( i ) } , a ^ { ( i ) } , b ^ { ( i ) } } { \operatorname* { s u p } } [ \underset { s ^ { ( r , F ) } } { \operatorname* { s u p } } [ \underset { s ^ { ( i ) } , a ^ { ( i ) } , b ^ { ( i ) } } { \operatorname* { s u p } } [ \underset { i = 1 } { \overset { n } { \sum } } \nu ( \phi ( \tilde { s } _ { 0 } ^ { ( i ) } , \bar { a } _ { 0 } ^ { ( i ) } ) ) ] - \underset { i = 1 } { \overset { n } { \sum } } \nu ( \phi ( s _ { 0 } ^ { ( i ) } , a _ { 0 } ^ { ( i ) } ) ) ) ] } \\ & { \leq \frac { 1 - \gamma } { n } \underset { s ^ { ( i ) } , a ^ { ( i ) } , b ^ { ( i ) } , a ^ { ( i ) } , b ^ { ( i ) } } { \operatorname* { s u p } } [ \underset { s ^ { ( r , F ) } } { \operatorname* { s u p } } [ \underset { i = 1 } { \overset { n } { \sum } } \nu ( \phi ( \frac { \sigma ^ { ( i ) } } { s } , \bar { a } _ { 0 } ^ { ( i ) } ) ) - \underset { i = 1 } { \overset { n } { \sum } } \nu ( \phi ( s _ { 0 } ^ { ( i ) } , a _ { 0 } ^ { ( i ) } ) ) ) ] } \\ & { = \frac { 1 - \gamma } { n } \underset { s ^ { ( i ) } , a ^ { ( i ) } , b ^ { ( i ) } , a ^ { ( i ) } , b ^ { ( i ) } } { \operatorname* { s u p } } [ \underset { s ^ { ( r , F ) } } { \operatorname* { s u p } } [ \underset { i = 1 } { \overset { n } { \sum } } \sigma ^ { ( i ) 1 } \{ \nu ( \phi ( \tilde { s } _ { 0 } ^ { ( i ) } , \bar { a } _ { 0 } ^ { ( i ) } ) ) - \nu ( \phi ( s _ { 0 } ^ { ( i ) } , a _ { 0 } ^ { ( i ) } ) ) \} ] } \\ &  \leq 2 ( 1 - \gamma )  \end{array}
+$$
+
+where the last inequality is from Lemma A.2. Combining the results, we get
+
+$$
+\operatorname* { P r } \left\{ \operatorname* { s u p } _ { \nu \in \mathcal { F } } \left| f _ { 1 } ( \nu ) - \widehat { f } _ { 1 } ( \nu ) \right| - 2 ( 1 - \gamma ) \sqrt { \frac { \bar { k } } { n } } \geq \epsilon \right\} \leq \exp \left( - \frac { n \epsilon ^ { 2 } } { 2 ( 1 - \gamma ) ^ { 2 } \bar { k } } \right) .
+$$
+
+Similarly, we derive bounds for $f _ { 2 }$ and $f _ { 3 }$ respectively:
+
+$$
+\begin{array} { l l } { { \displaystyle \operatorname* { P r } \left\{ \operatorname* { s u p } _ { \nu \in \mathcal { F } } \Big | f _ { 2 } ( \nu ) - \widehat { f } _ { 2 } ( \nu ) \Big | - 2 \sqrt \frac { \bar { k } } { n } \geq \epsilon \right\} \leq \exp \left( - \frac { n \epsilon ^ { 2 } } { 2 \bar { k } } \right) , } } & { { } } \\ { { \displaystyle \operatorname* { P r } \left\{ \operatorname* { s u p } _ { \nu \in \mathcal { F } } \Big | f _ { 3 } ( \nu ) - \widehat { f } _ { 3 } ( \nu ) \Big | - 2 \gamma \sqrt \frac { \bar { k } } { n } \geq \epsilon \right\} \leq \exp \left( - \frac { n \epsilon ^ { 2 } } { 2 \gamma ^ { 2 } \bar { k } } \right) . } } \end{array}
+$$
+
+By letting RHS of above bounds to be $\delta / 3$ and using union bound, we get, with probability $1 { - } \delta$ , we get
+
+$$
+\left| \mathrm { I P M } _ { \mathcal { G } } ( d _ { \phi } ^ { \pi } , d _ { \phi } ^ { D } ) - \widehat { \mathrm { I P M } } ( d _ { \phi } ^ { \pi } , d _ { \phi } ^ { D } ) \right| \leq \sqrt { \frac { \bar { k } } { n } } \left( 4 + \sqrt { 8 \log \frac { 3 } { \delta } } \right) .
+$$
+
+The relationship between $\mathcal { F }$ and $\mathcal { G }$
+
+$$
+\mathcal { J } = \left\{ g \in ( \mathcal { Z } \to \mathbb { R } ) : g ( z ) = \nu ( z ) - \gamma \mathbb { E } _ { s ^ { \prime } \sim T ( \phi ^ { - 1 } ( z ) ) } [ \nu ( \phi ( s ^ { \prime } , a ^ { \prime } ) ) ] , \nu \in ( \mathcal { Z } \to \mathbb { R } ) , \langle \nu , \nu \rangle _ { \mathcal { H } _ { k } } \leq 1 \right\}
+$$
+
+show that when the conditional expectation $\mathbb { E } _ { s ^ { \prime } \sim T ( \phi ^ { - 1 } ( \cdot ) ) , a ^ { \prime } \sim \pi ( s ^ { \prime } ) } [ \nu ( \phi ( s ^ { \prime } , a ^ { \prime } ) ) ] : \mathcal { Z } \to \mathbb { R }$ is a function in RKHS $\mathcal { H } _ { k }$ , $\mathcal { G }$ also becomes a subset of $\mathcal { H } _ { k }$ .
+
+Then we can prove the following Theorem.
+
+Theorem 4.3. Given an MDP $M$ , its estimate $\widehat { M }$ with a bijective representation function $\phi$ , i.e. $( \widehat { T } , \widehat { R } ) = \langle \widehat { T } _ { z } \circ \phi , \widehat { R } _ { z } \circ \phi \rangle$ , and an RKHS $\begin{array} { r } { \mathcal { H } _ { k } \quad \subset \quad ( \mathcal { Z } \quad \to \quad \mathbb { R } ) } \end{array}$ induced by a universal kernel $k$ such that $\mathrm { s u p } _ { z \in \mathcal { Z } } k ( z , z ) ~ = ~ \bar { k }$ , assume that $f _ { \phi , \widehat { R } _ { z } , \widehat { T } _ { z } } \left( \widehat { z } \right) \ =$ $\begin{array} { r } { \mathbb { E } _ { T , \pi } \Big [ \sum _ { t = 0 } ^ { \infty } \gamma ^ { t } \mathcal { E } _ { \phi , \widehat { R } _ { z } , \widehat { T } _ { z } } \big ( s _ { t } , a _ { t } \big ) \Big | \big ( s _ { 0 } , a _ { 0 } \big ) = \phi ^ { - 1 } \big ( z \big ) \Big ] \in \mathcal { H } _ { k } } \end{array}$ with $B _ { \phi } = \| f _ { \phi , \widehat { R } _ { z } , \widehat { T } _ { z } } \| _ { \mathcal { H } _ { k } }$ and the loss is bounded by $\begin{array} { r } { \bar { \mathcal { E } } = \operatorname* { s u p } _ { s \in { \mathcal { S } } , a \in \mathcal { A } } \mathcal { E } _ { \phi , \widehat { R } _ { z } , \widehat { T } _ { z } } ( s , a ) } \end{array}$ . Let n be the number of data in $\mathcal { D }$ . With probability $1 - 2 \delta$ ,
+
+$$
+R ^ { \pi } - { \widehat R } ^ { \pi } \Big | \leq \frac { 1 } { n } \sum _ { ( s , a ) \in { \mathcal D } } { \mathcal E } _ { \phi , { \widehat R } _ { z } , { \widehat T } _ { z } } ( s , a ) + B _ { \phi } \widehat \Pi \widehat M \big ( d _ { \phi } ^ { \pi } , d _ { \phi } ^ { { D } } \big ) + \sqrt { \frac { \bar { \mathcal E } ^ { 2 } } { 2 n } \log { \frac { 1 } { \delta } } } + B _ { \phi } \sqrt { \frac { \bar { k } } { n } } \left( 4 + \sqrt { 8 \log { \frac { 3 } { \delta } } } \right) .
+$$
+
+Proof. Applying the Hoeffding inequality (Hoeffding, 1963), with probability $1 - \delta$ , we get:
+
+$$
+\mathbb { E } _ { ( s , a ) \sim d ^ { \mathcal { D } } } \left[ \mathcal { E } _ { \phi , \widehat { R } _ { z } , \widehat { T } _ { z } } ( s , a ) \right] \leq \frac { 1 } { n } \sum _ { ( s , a ) \sim \mathcal { D } } \mathcal { E } _ { \phi , \widehat { R } _ { z } , \widehat { T } _ { z } } ( s , a ) + \sqrt { \frac { \bar { \mathcal { E } } ^ { 2 } } { 2 n } \log \frac { 1 } { \delta } } .
+$$
+
+By using an union bound with Eq. (13) and substituting terms in Eq. (3), we recover the result.
+
+# B EXPERIMENT DETAILS
+
+# B.1 COMPUTING INFRASTRUCTURE
+
+All experiments were conducted on the Google Cloud Platform. Specifically, we used computeoptimized machines (c2-standard-4) that provide 4 vCPUs and 16 GB memory for the evaluation experiment of Section 5.1, and we used high-memory machines (n1-highmem-4), which provide 4 vCPUs and 26GB memory, equipped with an Nvidia Tesla K80 GPU for the RL experiment of Section 5.2.
+
+# B.2 DETAILS OF THE OPE EXPERIMENT
+
+Task details We did not modify CartPole-v0 environment and Acrobot-v1 environment from the original implementation of OpenAI Gym (Brockman et al., 2016) except for the maximum trajectory length. We ran PPO (Schulman et al., 2017) to optimize policies to reach a certain performance level and set them as the target policies for CartPole-v0 and Acrobot-v1. For the HIV simulator, we used the code adapted by Liu et al. (2018b), which is originally from the implementation of $\mathsf { R L P y } ^ { 3 }$ . We modified the environment to have more randomness in the initial state (up to $1 0 \%$ perturbation from the baseline initial state) and to use the reward function that gives the logarithm of original reward values, as the original reward function scales up to $1 0 ^ { 1 0 }$ . We used a tree-based fitted ${ \bf q }$ -iteration algorithm implemented by Liu et al. (2018b) to optimize the target policy for the HIV simulator. All the other details are shown in Table 2. We assume that the termination conditions of tasks are known in prior.
+
+Table 2: Task settings of OPE experiments.   
+
+<table><tr><td></td><td>CartPole-vO</td><td>Acrobot-v1</td><td>HIV simulator</td></tr><tr><td></td><td></td><td></td><td></td></tr><tr><td>state space dimension</td><td>4</td><td>6</td><td>6</td></tr><tr><td># of actions</td><td>2</td><td>3</td><td>4</td></tr><tr><td>discount rate y</td><td>0.99</td><td>0.99</td><td>0.98</td></tr><tr><td># of trajectories</td><td>200</td><td>200</td><td>50</td></tr><tr><td>max length of training traj.</td><td>200</td><td>200</td><td>200</td></tr><tr><td>max length of rollouts for evaluation</td><td>1000</td><td>1000</td><td>1000</td></tr><tr><td>discounted return of target policy</td><td>89.4</td><td>-56.88</td><td>803.2</td></tr></table>
+
+Model and algorithm details The model we learn is composed of a representation module and a dynamics module. To be consistent with the experiment settings in Liu et al. (2018b), we use a representation module of a single hidden layer feed-forward network that takes the state as input and outputs representation. We squashed the representation between $( - 1 , 1 )$ using the tanh activation function. The dynamics module is also a single hidden layer feed-forward network that takes representation as input and outputs state difference and reward prediction for each action. We use the swish activation function (Ramachandran et al., 2017) for the hidden layers of two modules. As a whole, the model can also be seen as a feed-forward network with three hidden layers of varying activation functions, where the output of the second hidden layer is the representation we regularize.
+
+For the purpose of comparison, we minimize the $L 2$ distance between the model prediction and the desired outcome from data, which corresponds to using a model of Gaussian predictive distribution with fixed variance. We standardized the inputs and outputs of the neural network and used Adam (Kingma & Ba, 2014) with a learning rate of $3 \times 1 0 ^ { - 4 }$ for the optimization. When compared to the similar experiments conducted in Liu et al. (2018b), we used a larger and more expressive model with more optimization steps with a smaller learning rate for a more accurate comparison. While the derivation of the RepB-SDE objective was based on the state-action representation function, we use state representation in this experiment for direct comparison with RepBM, which uses state representation (it can be also understood as using action invariant kernel). We follow the choice of Liu et al. (2018b) and use dot product kernel $k ( \phi ( s ) , \phi ( \bar { s } ) ) = \phi ( s ) ^ { \top } \phi ( \bar { s } )$ for the OPE experiment, which is not universal but allows us to avoid search of kernel hyperparameters, such as length-scales.
+
+After the training, we generate another 200 trajectories (50 in case of HIV simulator), and rollout in both true and simulated (based on learned model) environments to evaluate models. We measure the individual MSE, which is
+
+$$
+\mathrm { I n d i v i d u a l ~ M S E } = \mathbb { E } _ { s _ { 0 } \sim d _ { 0 } } \bigg [ \bigg ( \mathbb { E } _ { \widehat { M } , \pi } \Big [ \sum _ { t = 0 } ^ { \infty } \gamma ^ { t } r _ { t } | s _ { 0 } \Big ] - \mathbb { E } _ { M , \pi } \Big [ \sum _ { t = 0 } ^ { \infty } \gamma ^ { t } r _ { t } | s _ { 0 } \Big ] \bigg ) ^ { 2 } \bigg ]
+$$
+
+for measuring the performance of each model. Whole experiment, from sampling data to learning and evaluating the model, is repeated 200 times with different random seeds.
+
+Choice and effect of hyperparameter $\alpha$ For choosing hyperparameter $\alpha$ for each algorithm, we searched over $\alpha \in \{ 0 . 0 0 1 , 0 . 0 1 , 0 . 1 , 1 , 1 0 \}$ for each off-policyness $\epsilon$ and for each environment. Chosen αs were mainly $\alpha \in$ $\lbrace 0 . 0 0 1 , 0 . 0 1 \rbrace$ for CartPole-v0, $\alpha \in \{ 1 , 1 0 \}$ for Acrobotv1, and $\alpha \in \{ 0 . 0 1 , 0 . 1 \}$ for HIV simulator for both algorithms. In general, large $\alpha$ was beneficial when high off-policyness () is present and/or the task is hard to generalize. On the right we show the example of effect of varying $\alpha$ in CartPole-v0.
+
+![](images/1c9ccb5c2c41ddb695510e4975c65cec88c2fe83b3762b2ff076037bc3277679.jpg)
+
+Comparison to other baselines In Figure 2, the OPE results with other model-free baselines are presented. FQE: Fitted Q-evaluation, IS: step-wise importance sampling, DR: doubly robust estimator based on step-wise importance sampling using the value function learned with fitted Qevaluation (Jiang & Li, 2016), DualDICE: stationary distribution correction algorithm (Nachum et al., 2019a). We used the implementation provided by the authors in case of DualDICE. All results are normalized to set the MSE of model-based baseline to be 1. Here, we used average MSEs instead of individual MSEs since importance sampling estimators are not suitable in computing individual MSEs. The results show that model-based methods are more robust to increasing off-policyness when compared to FQE. The results of model-based methods on Acrobot is relatively worse due to the difficult dynamics of Acrobot environment.
+
+![](images/1eb2b85496bae8da7d8f1105ae49dddbe789c2255747b8997f70703f372d78dd.jpg)  
+Figure 2: The OPE results compared to other model-free baselines. All experiments are repeated 200 times and the error bars denote $9 5 \%$ confidence interval.
+
+# B.3 DETAILS OF THE OFFLINE RL EXPERIMENTS
+
+Task details We used 12 datasets in D4RL $\mathrm { F u }$ et al., 2020) over four dataset types and three environments as specified in the main text. In Table 1, normalized scores suggested by (Fu et al., 2020) is used to report the result, where score of 0 corresponds to a random policy and 100 corresponds to a converged SAC policy. In HalfCheetah-v2, score of 0 means the undiscounted return of $- 2 8 0$ , and score of 1 means the undiscounted return of 12135. In Hopper-v2, score of 0 means the undiscounted return of $- 2 0$ , and score of 1 means the undiscounted return of 3234. In Walker2d-v2 score of 0 means the undiscounted return of 2, and score of 1 means the undiscounted return of 4592. We assume that the termination conditions of tasks are known in prior.
+
+Representation balancing maximizing undiscounted return As we report the undiscounted sum of rewards in the experiments, the maximization of lower bound of $R ^ { \pi }$ may result in an underutilization of experiences of later timesteps. One way to mitigate the mismatch is to optimize the policy by maximizing the returns starting from the states in the dataset. It corresponds to maximizing $\begin{array} { r } { \tilde { R } ^ { \pi } = ( 1 - \gamma ) \mathbb { E } , M , \pi \bigl [ \sum _ { t = 0 } ^ { \infty } \gamma ^ { t } r _ { t } \bigr ] } \end{array}$ instead of $R ^ { \pi }$ , where the expectation respect to the initial state distribution $d _ { 0 }$ is altered with the data distribution $d ^ { \mathcal { D } }$ .
+
+Consequently, to bound the error of ${ \tilde { R } } ^ { \pi }$ , the representation should be balanced with another dis  
+counted stationary distribution $\begin{array} { r } { \tilde { d } ^ { \pi } ( s , a ) \triangleq ( 1 - \gamma ) \sum _ { t = 0 } ^ { \infty } \gamma ^ { t } \operatorname* { P r } ( s _ { t } = s , a _ { t } = a | s _ { 0 } \sim d ^ { \mathcal { D } } , T , \pi ) } \end{array}$ , the $\pi$   
+derivations can be easily adapted by noting that:
+
+$$
+\mathrm { P M } _ { \mathcal { G } } ( \tilde { d } _ { \phi } ^ { \pi } , d _ { \phi } ^ { \mathcal { D } } ) = \operatorname* { s u p } _ { \nu \in \mathcal { F } } \left| ( 1 - \gamma ) \mathbb { E } _ { \alpha \sim \pi ( s ) } \left[ \nu \big ( \phi ( s , a ) \big ) \right] - \mathbb { E } _ { ( s , a , s ^ { \prime } ) \sim d ^ { \mathcal { D } } } \left[ \nu \big ( \phi ( s , a ) \big ) - \gamma \nu \big ( \phi ( s ^ { \prime } , a ^ { \prime } ) \big ) \right] \right| ,
+$$
+
+and changing the initial state sampling distributions to $d ^ { \mathcal { D } }$ during the estimation of IPM.
+
+Model and algorithm details Similar to the model used in the OPE experiment, the model we learn is composed of a representation module and a dynamics module. A representation module is a feed-forward network with two hidden layers that takes the state-action pair as input and outputs representation through the tanh activation function. The dynamics module is a single hidden layer network that takes representation as input and outputs parameters of diagonal Gaussian distribution predicting state difference and reward. We use 200 hidden units for all intermediate layers including the representation layer. Across all domains, we train an ensemble of 7 models and pick the best 5 models on their validation error on hold-out set of 1000 transitions in the dataset. The inputs and outputs of the neural network is normalized. We present the pseudo-code of the presented Representation Balancing Offline Model-based RL algorithm below.
+
+Input: Offline dataset $\mathcal { D }$ , previous policy $\pi$ Output: Optimized policy $\pi$   
+1: Sample $K$ independent datasets with replacement from $\mathcal { D }$ .   
+2: Train bootstrapped ensemble of $K$ models $\{ \widehat { T } _ { i } , \widehat { R } _ { i } \} _ { i = 0 } ^ { K }$ minimizing Eq. (8) (adapted with $\tilde { d } _ { \phi } ^ { \pi } ,$ ).   
+3: for rep ${ \mathcal { a } } t = 0 , 1 , \ldots$ do   
+4: for rollout $\mathbf { \Sigma } = 0 , 1 , \dots , B$ do   
+5: Sample initial rollout state $s _ { 0 }$ from $\mathcal { D }$ .   
+6: for $t = 0 , 1 , \ldots$ do   
+7: Sample an action $a _ { t } \sim \pi ( s _ { t } )$ .   
+8: Randomly pick $( \widehat { T } _ { i } , \widehat { R } _ { i } )$ and sample $( s _ { t + 1 } , r _ { t } ) \sim ( \widehat { T } _ { i } , \widehat { R } _ { i } ) ( s _ { t } , a _ { t } ) .$   
+9: Compute $ \tilde { r } _ { t } = r _ { t } - \gamma \lambda \sqrt { \mathbb { V } _ { K } [ \mu ( s _ { t } , a _ { t } ) ] } | | _ { 2 }$ and store $\left( { { s _ { t } } , { a _ { t } } , \tilde { r } _ { t } , { s _ { t + 1 } } } \right)$ in $\widehat { \mathcal { D } }$ .   
+10: end for   
+11: end for   
+12: Draw samples from $\mathcal { D }$ to compute $\widehat { \mathrm { I P M } } ( \tilde { d } _ { \phi } ^ { \pi } , d _ { \phi } ^ { D } )$ .   
+13: Draw samples from $\mathcal { D }$ and $\widehat { \mathcal { D } }$ to update critic $Q$ .   
+14: Maximize $\mathbb { E } _ { s \sim \mathcal { D } \cup \widehat { \mathcal { D } } , a \sim \pi ( s ) } [ Q ( s , a ) - \tau \log \pi ( a | s ) ] - \alpha _ { \pi } \widehat { \mathrm { I P M } } ( \tilde { d } _ { \phi } ^ { \pi } , d _ { \phi } ^ { \mathcal { D } } )$ to update $\pi$ .   
+15: end for
+
+For the result of MOPO (Yu et al., 2020), we ran the code kindly provided by the authors4 without any modification on the algorithm or the hyperparameters. All algorithms we experimented (Base, RP, RepB-SDE) share all the hyperparameters in common except the ones associated with changing objectives. We run SAC on the full rollouts from the trained ensemble models as shown in Algorithm 1. The common hyperparameters shared among algorithms are shown in Table 3. We simply tried the listed hyperparameters and not tuned them further. For RP and RepB-SDE, we penalized the reward from simulated environments with the standard deviation of prediction means of neural network ensembles. We used standardized output of all 7 neural networks to compute the reward penalty. We first search over the reward penalty coefficient $\lambda \in \{ 0 , 2 , 5 , 7 , 1 0 , 1 5 \}$ that grants RP the best performance. We shared same $\lambda$ for RepB-SDE and searched over $\alpha _ { M } \in \{ 0 . 0 1 , 0 . 1 , 1 \}$ , $\alpha _ { \pi } ~ \in ~ \{ 0 , 0 . 0 1 , 0 . 1 , 1 , 1 0 \}$ . We ran the algorithms for $1 . 5 \times 1 0 ^ { 6 }$ gradient updates, except for HalfCheetah-Medium-Expert where we ran for $5 . 0 \times 1 0 ^ { 6 }$ . In Table 4, we present the standard errors of results, which was omitted in the main text.
+
+Table 3: Common hyperparameters used in offline RL experiments.   
+
+<table><tr><td>Parameter Value</td></tr><tr><td>optimizer Adam (Kingma &amp; Ba, 2014)</td></tr><tr><td>learning rate 3×10-4</td></tr><tr><td>discount factor y 0.99</td></tr><tr><td>number of samples per minibatch 256</td></tr><tr><td>target smoothing coefficient T 5×10-3</td></tr><tr><td>[actor/critic] number of hidden layers 2</td></tr><tr><td>[actor/critic] number of hidden units per layer 256</td></tr><tr><td>[actor/critic] non-linearity ReLU</td></tr><tr><td># of rollouts 104</td></tr><tr><td>max length of rollouts 103</td></tr><tr><td>rollout buffer size 5×107</td></tr></table>
+
+Table 4: Normalized scores on D4RL MuJoCo benchmark datasets (Fu et al., 2020) with standard errors fully specified.   
+
+<table><tr><td>Dataset type</td><td>Environment</td><td>RepB-SDE (ours)</td><td>RP</td><td>Base</td><td>MOPO</td></tr><tr><td></td><td></td><td></td><td></td><td></td><td></td></tr><tr><td>Random</td><td>Walker2d-v2</td><td>21.1 ± 1.0</td><td>18.4 ±1.8</td><td>16.4 ± 2.2</td><td>1.3 ± 0.5</td></tr><tr><td>Medium</td><td>Walker2d-v2</td><td>72.1 ± 1.9</td><td>56.3 ± 4.3</td><td>5.5 ± 0.4</td><td>-0.1 ± 0.0</td></tr><tr><td>Med-Replay</td><td>Walker2d-v2</td><td>49.8 ± 11.4</td><td>41.1 ±10.3</td><td>6.2 ± 0.7</td><td>47.8 ± 4.8</td></tr><tr><td>Med-Expert</td><td>Walker2d-v2</td><td>88.8 ± 6.9</td><td>72.6 ± 5.1</td><td>51.7 ± 7.1</td><td>32.4 ± 8.8</td></tr><tr><td>Random</td><td>Hopper-v2</td><td>8.6 ±1.0</td><td>8.3 ±0.2</td><td>8.3 ±0.2</td><td>9.1 ± 0.3</td></tr><tr><td>Medium</td><td>Hopper-v2</td><td>34.0 ± 2.8</td><td>27.5 ± 3.1</td><td>19.8 ± 0.8</td><td>19.2 ± 4.5</td></tr><tr><td>Med-Replay</td><td>Hopper-v2</td><td>62.2 ± 6.7</td><td>49.8 ± 8.7</td><td>32.9 ± 5.5</td><td>80.8 ± 2.9</td></tr><tr><td>Med-Expert</td><td>Hopper-v2</td><td>82.6 ± 7.0</td><td>74.0 ± 7.2</td><td>19.1 ± 1.1</td><td>23.2 ± 1.3</td></tr><tr><td>Random</td><td>HalfCheetah-v2</td><td>32.9 ± 1.1</td><td>31.3 ±1.5</td><td>26.1 ± 6.2</td><td>29.9 ± 1.2</td></tr><tr><td>Medium</td><td>HalfCheetah-v2</td><td>49.1 ± 0.3</td><td>47.3 ± 0.3</td><td>22.1 ± 4.4</td><td>45.0 ± 0.4</td></tr><tr><td>Med-Replay</td><td>HalfCheetah-v2</td><td>57.5 ± 0.8</td><td>53.6 ± 0.8</td><td>54.0 ± 2.4</td><td>53.8 ± 1.1</td></tr><tr><td>Med-Expert</td><td>HalfCheetah-v2</td><td>55.4 ± 8.3</td><td>36.7 ± 12.0</td><td>31.9 ± 10.5</td><td>91.0 ± 2.5</td></tr></table>
+
+In addition to the performance after the final iteration presented in Table 1, we also present the learning curves of the experiment in Figure 3, and the effect of the choice of $\alpha _ { M }$ and $\alpha _ { \pi }$ in Figure 4. The Figure 4 shows that the presented regularization is robust to the choice of $\alpha _ { M } , \alpha _ { \pi }$ except for too large $\alpha _ { M }$ , consistently improving from RP, which corresponds to $\alpha _ { M } = 0 , \alpha _ { \pi } = 0$ .
+
+![](images/6da69fd1b89d4ec41b0f3dfa3d8233be66aedb17d085f5919e4c11e946c54cdc.jpg)  
+Figure 3: The learning curves of the D4RL experiment.
+
+![](images/bcf6e9d1e6e698ee7c20d46468f932a177986219b3341a07904605f54aa078ee.jpg)  
+Figure 4: The effect of varying hyperparameter in the D4RL experiment.
\ No newline at end of file
diff --git a/parse/train/QpNz8r_Ri2Y/QpNz8r_Ri2Y_content_list.json b/parse/train/QpNz8r_Ri2Y/QpNz8r_Ri2Y_content_list.json
new file mode 100644
index 0000000000000000000000000000000000000000..59aa793bf28e02567c4870a4263ba045b4e1a87d
--- /dev/null
+++ b/parse/train/QpNz8r_Ri2Y/QpNz8r_Ri2Y_content_list.json
@@ -0,0 +1,2754 @@
+[
+    {
+        "type": "text",
+        "text": "REPRESENTATION BALANCING OFFLINE MODEL-BASED REINFORCEMENT LEARNING ",
+        "text_level": 1,
+        "bbox": [
+            174,
+            99,
+            823,
+            146
+        ],
+        "page_idx": 0
+    },
+    {
+        "type": "text",
+        "text": "Byung-Jun Lee1,3, Jongmin Lee1 & Kee-Eung $\\mathbf { K i m } ^ { 1 , 2 }$ ",
+        "bbox": [
+            186,
+            167,
+            562,
+            184
+        ],
+        "page_idx": 0
+    },
+    {
+        "type": "text",
+        "text": "1School of Computing, KAIST, Daejeon, Republic of Korea 2Graduate School of AI, KAIST, Daejeon, Republic of Korea 3Gauss Labs Inc., Seoul, Republic of Korea {bjlee,jmlee}@ai.kaist.ac.kr, kekim@kaist.ac.k ",
+        "bbox": [
+            183,
+            185,
+            630,
+            242
+        ],
+        "page_idx": 0
+    },
+    {
+        "type": "text",
+        "text": "ABSTRACT ",
+        "text_level": 1,
+        "bbox": [
+            454,
+            279,
+            544,
+            294
+        ],
+        "page_idx": 0
+    },
+    {
+        "type": "text",
+        "text": "One of the main challenges in offline and off-policy reinforcement learning is to cope with the distribution shift that arises from the mismatch between the target policy and the data collection policy. In this paper, we focus on a model-based approach, particularly on learning the representation for a robust model of the environment under the distribution shift, which has been first studied by Representation Balancing MDP (RepBM). Although this prior work has shown promising results, there are a number of shortcomings that still hinder its applicability to practical tasks. In particular, we address the curse of horizon exhibited by RepBM, rejecting most of the pre-collected data in long-term tasks. We present a new objective for model learning motivated by recent advances in the estimation of stationary distribution corrections. This effectively overcomes the aforementioned limitation of RepBM, as well as naturally extending to continuous action spaces and stochastic policies. We also present an offline model-based policy optimization using this new objective, yielding the state-of-the-art performance in a representative set of benchmark offline RL tasks. ",
+        "bbox": [
+            233,
+            311,
+            764,
+            518
+        ],
+        "page_idx": 0
+    },
+    {
+        "type": "text",
+        "text": "1 INTRODUCTION ",
+        "text_level": 1,
+        "bbox": [
+            176,
+            549,
+            336,
+            565
+        ],
+        "page_idx": 0
+    },
+    {
+        "type": "text",
+        "text": "Reinforcement learning (RL) has accomplished remarkable results in a wide range of domains, but its successes were mostly based on a large number of online interactions with the environment. However, in many real-world tasks, exploratory online interactions are either very expensive or dangerous (e.g. robotics, autonomous driving, and healthcare), and applying a standard online RL would be impractical. Consequently, the ability to optimize RL agents reliably without online interactions has been considered as a key to practical deployment, which is the main goal of batch RL, also known as offline RL (Fujimoto et al., 2019; Levine et al., 2020). ",
+        "bbox": [
+            174,
+            582,
+            825,
+            680
+        ],
+        "page_idx": 0
+    },
+    {
+        "type": "text",
+        "text": "In an offline RL algorithm, accurate policy evaluation and reliable policy improvement are both crucial for the successful training of the agent. Evaluating policies in offline RL is essentially an off-policy evaluation (OPE) task, which aims to evaluate the target policy given the dataset collected from the behavior policy. The difference between the target and the behavior policies causes a distribution shift in the estimation, which needs to be adequately addressed for accurate policy evaluation. OPE itself is one of the long-standing hard problems in RL (Sutton et al., 1998; 2009; Thomas & Brunskill, 2016; Hallak & Mannor, 2017). ",
+        "bbox": [
+            174,
+            688,
+            825,
+            784
+        ],
+        "page_idx": 0
+    },
+    {
+        "type": "text",
+        "text": "However, recent offline RL studies mainly focus on how to improve the policy conservatively while using a common policy evaluation technique without much considerations for the distribution shift, e.g. mean squared temporal difference error minimization or maximum-likelihood training of environment model (Fujimoto et al., 2019; Kumar et al., 2019; Yu et al., 2020). While conservative policy improvement helps the policy evaluation by reducing the off-policyness, we hypothesize that addressing the distribution shift explicitly during the policy evaluation can further improve the overall performance, since it can provide a better foundation for policy improvement. ",
+        "bbox": [
+            174,
+            791,
+            823,
+            888
+        ],
+        "page_idx": 0
+    },
+    {
+        "type": "text",
+        "text": "To this end, we aim to explicitly address the distribution shift of the OPE estimator used in the offline RL algorithm. In particular, we focus on the model-based approach, where we train an environment model robust to the distribution shift. One of the notable prior works is Representation Balancing MDP (RepBM) (Liu et al., 2018b), which regularizes the representation learning of the model to be invariant between the distributions. However, despite the promising results by RepBM, its step-wise estimation of the distance between the distributions has a few drawbacks that limit the algorithm from being practical: not only it assumes a discrete-action task where the target policy is deterministic, but it also performs poorly in long-term tasks due to the curse of horizon of step-wise importance sampling (IS) estimators (Liu et al., 2018a). ",
+        "bbox": [
+            173,
+            895,
+            821,
+            924
+        ],
+        "page_idx": 0
+    },
+    {
+        "type": "text",
+        "text": "",
+        "bbox": [
+            174,
+            103,
+            825,
+            202
+        ],
+        "page_idx": 1
+    },
+    {
+        "type": "text",
+        "text": "To address these limitations, we present the Representation Balancing with Stationary Distribution Estimation (RepB-SDE) framework, where we aim to learn a balanced representation by regularizing, in the representation space, the distance between the data distribution and the discounted stationary distribution induced by the target policy. Motivated by the recent advances in estimating stationary distribution corrections, we present a new representation balancing objective to train a model of the environment that no longer suffers from the curse of horizon. We empirically show that the model trained by the RepB-SDE objective is robust to the distribution shift for the OPE task, particularly when the difference between the target and the behavior is large. We also introduce a model-based offline RL algorithm based on the RepB-SDE framework and report its performance on the D4RL benchmark (Fu et al., 2020), showing the state-of-the-art performance in a representative set of tasks. ",
+        "bbox": [
+            174,
+            208,
+            825,
+            361
+        ],
+        "page_idx": 1
+    },
+    {
+        "type": "text",
+        "text": "2 RELATED WORK ",
+        "text_level": 1,
+        "bbox": [
+            176,
+            385,
+            341,
+            401
+        ],
+        "page_idx": 1
+    },
+    {
+        "type": "text",
+        "text": "Learning balanced representation Learning a representation invariant to specific aspects of data is an established method for overcoming distribution shift that arises in unsupervised domain adaptation (Ben-David et al., 2007; Zemel et al., 2013) and in causal inference from observational data (Shalit et al., 2017; Johansson et al., 2018). They have shown that imposing a bound on the generalization error under the distribution shift leads to the objective that learns a balanced representation such that the training and the test distributions look similar. RepBM (Liu et al., 2018b) can be seen as a direct extension to the sequential case, which encourages the representation to be invariant under the target and behavior policies in each timestep. ",
+        "bbox": [
+            174,
+            419,
+            825,
+            530
+        ],
+        "page_idx": 1
+    },
+    {
+        "type": "text",
+        "text": "Stationary distribution correction estimation (DICE) Step-wise importance sampling (IS) estimators (Precup, 2000) compute importance weights by taking the product of per-step distribution ratios. Consequently, these methods suffer from exponentially high variance in the lengths of trajectories, which is a phenomenon called the curse of horizon (Liu et al., 2018a). Recently, techniques of computing a stationary DIstribution Correction Estimation (DICE) have made remarkable progress that effectively addresses the curse of horizon (Liu et al., 2018a; Nachum et al., $2 0 1 9 \\mathrm { a }$ ; Tang et al., 2020; Zhang et al., 2020; Mousavi et al., 2020). DICE has been also used to explicitly address the distribution shift in online model-free RL, by directly applying IS on the policy and action-value objectives (Liu et al., 2019; Gelada & Bellemare, 2019). We adopt one of the estimation techniques, DualDICE (Nachum et al., 2019a), to measure the distance between the stationary distribution and the data distribution in the representation space. ",
+        "bbox": [
+            174,
+            549,
+            825,
+            703
+        ],
+        "page_idx": 1
+    },
+    {
+        "type": "text",
+        "text": "Offline reinforcement learning There are extensive studies on improving standard online modelfree RL algorithms (Mnih et al., 2015; Lillicrap et al., 2016; Haarnoja et al., 2018) for stable learning in the offline setting. The main idea behind them is to conservatively improve policy by (1) quantifying the uncertainty of value function estimate, e.g. using bootstrapped ensembles (Kumar et al., 2019; Agarwal et al., 2020), or/and (2) constraining the optimized target policy to be close to the behavior policy (i.e. behavior regularization approaches) (Fujimoto et al., 2019; Kumar et al., 2019; Wu et al., 2019; Lee et al., 2020). A notable exception is AlgaeDICE (Nachum et al., 2019b), which implicitly uses DICE to regularize the discounted stationary distribution induced by the target policy to be kept inside of the data support, similar to this work. ",
+        "bbox": [
+            174,
+            722,
+            825,
+            847
+        ],
+        "page_idx": 1
+    },
+    {
+        "type": "text",
+        "text": "On the other hand, Yu et al. (2020) argued that the model-based approach can be advantageous due to its ability to generalize predictions on the states outside of the data support. They introduce MOPO (Yu et al., 2020), which uses truncated rollouts and penalized rewards for conservative policy improvement. MOReL (Kidambi et al., 2020) trains a state-action novelty detector and use it to penalize rewards in the data-sparse region. Matsushima et al. (2020), MOOSE (Swazinna et al., 2020) ",
+        "bbox": [
+            174,
+            854,
+            823,
+            924
+        ],
+        "page_idx": 1
+    },
+    {
+        "type": "text",
+        "text": "and MBOP (Argenson & Dulac-Arnold, 2020) guide their policy optimization using the behavior policy, similar to the behavior regularization approaches. ",
+        "bbox": [
+            173,
+            103,
+            823,
+            132
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "text",
+        "text": "Note that these aforementioned offline RL methods build on the standard approximate dynamic programming algorithm for action-value estimation (model-free) or on a maximum-likelihood environment model (model-based), without explicitly addressing the distribution shift in the estimator. In contrast, we augment the objective for model learning to obtain a robust model under the distribution shift, which is the first attempt for offline RL to the best of our knowledge. ",
+        "bbox": [
+            174,
+            138,
+            825,
+            209
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "text",
+        "text": "3 PRELIMINARIES ",
+        "text_level": 1,
+        "bbox": [
+            176,
+            228,
+            339,
+            244
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "text",
+        "text": "A Markov Decision Process (MDP) is specified by a tuple $M = \\langle S , A , T , R , d _ { 0 } , \\gamma \\rangle$ , consisting of state space $s$ , action space $\\mathcal { A }$ , transition function $T : \\mathcal { S } \\times \\mathcal { A }  \\Delta ( \\mathcal { S } )$ , reward function $R : \\mathcal { S } \\times \\overline { { \\mathcal { A } } } \\to \\Delta ( [ 0 , r _ { \\operatorname* { m a x } } ] )$ , initial state distribution $d _ { 0 }$ , and discount rate $\\gamma$ . In this paper, we mainly focus on continuous state space $S \\subseteq \\mathbb { R } ^ { d _ { s } }$ and conduct experiments on both discrete action spaces ${ \\cal A } = \\{ a _ { 0 } , . . . a _ { n _ { a } } \\}$ and continuous action spaces $\\mathcal { A } \\subseteq \\mathbb { R } ^ { d _ { a } ^ { \\star } }$ . Given MDP $M$ and policy $\\pi$ , which is a (stochastic) mapping from state to action, the trajectory can be generated in the form of $s _ { 0 } , a _ { 0 } , r _ { 0 } , s _ { 1 } , a _ { 1 } , r _ { 1 } , \\ldots$ , where $s _ { 0 } \\sim d _ { 0 }$ and for each timestep $t \\geq 0$ , $a _ { t } \\sim \\pi ( s _ { t } )$ , $\\boldsymbol { r } _ { t } \\sim R ( s _ { t } , a _ { t } )$ , and $s _ { t + 1 } \\sim T ( s _ { t } , a _ { t } )$ . The goal of RL is to optimize or evaluate a policy, based on the normalized expected discounted return: $\\begin{array} { r } { R ^ { \\pi } \\triangleq ( 1 - \\gamma ) \\mathbb { E } _ { M , \\pi } \\left[ \\sum _ { t = 0 } ^ { \\infty } \\gamma ^ { t } r _ { t } \\right] } \\end{array}$ . ",
+        "bbox": [
+            173,
+            258,
+            826,
+            390
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "text",
+        "text": "A useful and important concept throughout the paper is the discounted stationary distribution, which represents the long-term occupancy of states: ",
+        "bbox": [
+            173,
+            393,
+            823,
+            422
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "equation",
+        "img_path": "images/4e97c5ca6686900ac9b0f9bd4d01b84db14920e7a7632d71f5cbe853c1d7c3e0.jpg",
+        "text": "$$\nd ^ { \\pi } ( s , a ) \\triangleq ( 1 - \\gamma ) \\sum _ { t = 0 } ^ { \\infty } \\gamma ^ { t } \\operatorname* { P r } ( s _ { t } = s , a _ { t } = a | M , \\pi ) .\n$$",
+        "text_format": "latex",
+        "bbox": [
+            326,
+            426,
+            669,
+            468
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "text",
+        "text": "From the definition, it can be observed that $R ^ { \\pi }$ can be obtained by $R ^ { \\pi } = \\mathbb { E } _ { ( s , a ) \\sim d ^ { \\pi } } \\left[ r ( s , a ) \\right]$ ",
+        "bbox": [
+            169,
+            472,
+            774,
+            488
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "text",
+        "text": "Offline RL and off-policy evaluation In this paper, we focus on the offline $R L$ problem where the agent can only access a static dataset $\\mathcal { D } = \\{ ( \\bar { s } _ { i } , a _ { i } , r _ { i } , s _ { i } ^ { \\prime } ) \\} _ { i = 1 } ^ { N }$ for the maximization of $R ^ { \\pi }$ . We consider a behavior-agnostic setting where we do not have any knowledge of the data collection process. We denote the empirical distribution of the dataset by $\\dot { d } ^ { \\mathcal { D } }$ . ",
+        "bbox": [
+            173,
+            501,
+            825,
+            558
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "text",
+        "text": "Before improving policy, we first aim to better evaluate $R ^ { \\pi }$ given a target policy $\\pi$ and a static dataset $\\mathcal { D }$ , which corresponds to an off-policy evaluation (OPE) problem. We mainly focus on a modelbased approach where the algorithm first estimates the unknown dynamics $\\widehat { T } , \\widehat { R }$ using the dataset $\\mathcal { D }$ . This defines an approximate MDP $\\widehat { M } = \\langle S , A , \\widehat { T } , \\widehat { R } , d _ { 0 } , \\gamma \\rangle$ , with the approximate expected discounted return $\\begin{array} { r } { \\widehat { R } ^ { \\pi } \\triangleq ( 1 - \\gamma ) \\mathbb { E } _ { \\widehat { M } , \\pi } \\left[ \\sum _ { t = 0 } ^ { \\infty } \\gamma ^ { t } r _ { t } \\right] } \\end{array}$ obtained from $\\widehat { M }$ . In this paper, we are interested in the MDP estimate $\\widehat { M }$ that can effectively reduce the error in the evaluation of policy $\\pi$ , $| R ^ { \\pi } - { \\widehat R } ^ { \\pi } |$ . In order to do so, we need to learn a good representation of a model that results in a small OPE error. We assume a bijective representation function $\\phi : \\mathcal { S } \\times \\mathcal { A }  \\mathcal { Z }$ where $\\mathcal { Z } \\subseteq \\mathbb { R } ^ { d _ { z } }$ is the representation space. We define the transition and the reward models in terms of the representation function $\\phi$ , i.e. $\\bar { \\widehat { T } } = \\widehat { T } _ { z } \\circ \\phi$ and $\\widehat { R } = \\widehat { R } _ { z } \\circ \\phi$ . In practice, where we use a neural network for $\\widehat { T }$ and $\\widehat { R }$ , $z$ can be chosen to be the output of an intermediate hidden layer, making $\\phi$ represented by lower layers and $\\widehat { T } _ { z } , \\widehat { R } _ { z }$ by the remaining upper layers. We define $d _ { \\phi } ^ { \\pi } ( z )$ the discounted stationary distribution on $\\mathcal { Z }$ induced by $d ^ { \\pi } ( s , a )$ under the representation function $z = \\phi ( s , a )$ , and similarly for $d _ { \\phi } ^ { { \\cal D } } ( z )$ . ",
+        "bbox": [
+            171,
+            564,
+            826,
+            776
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "text",
+        "text": "4 REPRESENTATION BALANCING OFFLINE MODEL-BASED RL ",
+        "text_level": 1,
+        "bbox": [
+            174,
+            795,
+            694,
+            813
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "text",
+        "text": "4.1 GENERALIZATION ERROR BOUND FOR MODEL-BASED OFF-POLICY EVALUATION ",
+        "text_level": 1,
+        "bbox": [
+            169,
+            827,
+            767,
+            840
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "text",
+        "text": "We aim to construct a model $\\widehat { M }$ from the dataset $\\mathcal { D }$ , which can accurately evaluate the policy $\\pi$ , by minimizing a good upper bound of policy evaluation error $| R ^ { \\pi } - { \\widehat { R } } ^ { \\pi } |$ . We define the following point-wise model loss for notational convenience: ",
+        "bbox": [
+            174,
+            848,
+            825,
+            897
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "equation",
+        "img_path": "images/ad510db3114669b45b366d79bc47298250fb467c2c679abf0cf02629ee163245.jpg",
+        "text": "$$\n\\begin{array} { r } { \\mathcal { E } _ { \\phi , \\widehat { R } _ { z } , \\widehat { T } _ { z } } ( s , a ) = c _ { R } D _ { T V } \\Big ( R ( r | s , a ) \\Big | \\Big | \\widehat { R } _ { z } \\big ( r | \\phi ( s , a ) \\big ) \\Big ) + c _ { T } D _ { T V } \\Big ( T ( s ^ { \\prime } | s , a ) \\Big | \\Big | \\widehat { T } _ { z } \\big ( s ^ { \\prime } | \\phi ( s , a ) \\big ) \\Big ) , } \\end{array}\n$$",
+        "text_format": "latex",
+        "bbox": [
+            187,
+            901,
+            805,
+            929
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "text",
+        "text": "where $c _ { R } = 2 ( 1 - \\gamma )$ and $c _ { T } = 2 \\gamma r _ { \\operatorname* { m a x } }$ . Then, we start by restating the simulation lemma (Kearns & Singh, 2002) to bound the policy evaluation error in terms of the point-wise model loss. The proof is available in Appendix A. ",
+        "bbox": [
+            174,
+            103,
+            825,
+            146
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "Lemma 4.1. Given an MDP $M$ and its estimate $\\widehat { M }$ with a bijective representation function $\\phi _ { : }$ , i.e. $( \\widehat { T } , \\widehat { R } ) = \\langle \\widehat { T } _ { z } \\circ \\phi , \\widehat { R } _ { z } \\circ \\phi \\rangle$ , the policy evaluation error of a policy $\\pi$ can be bounded by: ",
+        "bbox": [
+            171,
+            150,
+            823,
+            184
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "equation",
+        "img_path": "images/be4430fb6e26dd6d13dc5a5eadf0c4ec02c96ae654aff21a36528c073cefe5c6.jpg",
+        "text": "$$\n\\begin{array} { r } { \\left| R ^ { \\pi } - \\widehat { R } ^ { \\pi } \\right| \\leq \\mathbb { E } _ { ( s , a ) \\sim d ^ { \\pi } } \\left[ \\mathcal { E } _ { \\phi , \\widehat { R } _ { z } , \\widehat { T } _ { z } } ( s , a ) \\right] } \\end{array}\n$$",
+        "text_format": "latex",
+        "bbox": [
+            362,
+            188,
+            633,
+            215
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "The Lemma 4.1 has a natural interpretation: if the model error is small in the states frequently visited by following the policy $\\pi$ , the resultant policy evaluation error will also be small. However, minimizing the RHS of Eq. (1) in the off-policy evaluation (OPE) task is generally intractable since the distribution $d ^ { \\pi }$ is not directly accessible. Therefore, the common practice has been to construct a maximum-likelihood MDP using $\\mathcal { D }$ while ignoring the distribution shift, but its OPE performance is not guaranteed. ",
+        "bbox": [
+            173,
+            224,
+            825,
+            310
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "Instead, we will derive a tractable upper bound on the policy evaluation error by eliminating the direct dependence on $d ^ { \\pi }$ in Eq. (1). To this end, we adopt the distance metric between two distributions over representations $d _ { \\phi } ^ { \\pi }$ and $d _ { \\phi } ^ { \\mathcal { D } }$ that can bound their difference in expectations, which is the Integral Probability Metric (IPM) (Muller, 1997): ¨ ",
+        "bbox": [
+            173,
+            315,
+            825,
+            375
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "equation",
+        "img_path": "images/b86cdd7b9a7b014a45536a084ffd03b37556682b8e9c250e180c7355c8694f81.jpg",
+        "text": "$$\n\\mathrm { I P M } _ { \\mathcal { G } } ( p , q ) = \\operatorname* { s u p } _ { g \\in \\mathcal { G } } \\big | \\mathbb { E } _ { z \\sim p } [ g ( z ) ] - \\mathbb { E } _ { z \\sim q } [ g ( z ) ] \\big | .\n$$",
+        "text_format": "latex",
+        "bbox": [
+            343,
+            377,
+            655,
+            405
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "where particular choices of $\\mathcal { G }$ make the IPM equivalent to different well-known distances of distributions, e.g. total variation distance or Wasserstein distance (Sriperumbudur et al., 2009). ",
+        "bbox": [
+            174,
+            410,
+            825,
+            439
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "Theorem 4.2. Given an MDP $M$ and its estimate $\\widehat { M }$ with a bijective representation function $\\phi$ , i.e. ${ \\mathcal { G } } \\triangleq \\{ g : { \\mathcal { Z } } \\to \\mathbb { R } \\}$ $( \\widehat { T } , \\widehat { R } ) = \\langle \\widehat { T } _ { z } \\circ \\phi , \\widehat { R } _ { z } \\circ \\phi \\rangle$ b such that $\\begin{array} { r } { \\frac { 1 } { B _ { \\phi } } \\mathcal { E } _ { \\phi , \\widehat { R } _ { z } , \\widehat { T } _ { z } } \\left( \\phi ^ { - 1 } ( \\cdot ) \\right) \\in \\mathcal { G } } \\end{array}$ , assume that there exists a constant . Then, for any policy $B _ { \\phi } > 0$ $\\pi$ and , $a$ function class ",
+        "bbox": [
+            173,
+            443,
+            825,
+            496
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "equation",
+        "img_path": "images/50332b075b998d81ff520371d10d176b4273f51b2bb0d2bb4ffa9954584b2c6d.jpg",
+        "text": "$$\n\\left| R ^ { \\pi } - \\widehat { R } ^ { \\pi } \\right| \\leq \\mathbb { E } _ { ( s , a ) \\sim d ^ { \\mathcal { D } } } \\Big [ \\mathcal { E } _ { \\phi , \\widehat { R } _ { z } , \\widehat { T } _ { z } } ( s , a ) \\Big ] + B _ { \\phi } \\mathrm { I P M } _ { \\mathcal { G } } ( d _ { \\phi } ^ { \\pi } , d _ { \\phi } ^ { \\mathcal { D } } )\n$$",
+        "text_format": "latex",
+        "bbox": [
+            292,
+            501,
+            704,
+            527
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "This theorem is an adaptation of Lemma 1 of Shalit et al. (2017) to an infinite horizon model-based policy evaluation and can be derived by the definition of $\\mathrm { I P M } _ { \\mathcal { G } } ( d _ { \\phi } ^ { \\pi } , d _ { \\phi } ^ { D } )$ since it serves as an upper bound of the difference in the expectations of any function in $\\mathcal { G }$ . The first term in Eq. (3) corresponds to the fitness to the data following $d ^ { \\mathcal { D } }$ , while the second term serves as a regularizer. To see this, minimizing the second term would yield a near-constant representation function, which would be bad for the first term since it cannot distinguish states and actions well enough. It shows a natural trade-off between optimizing the model that fits data better and learning the representation that is invariant with respect to $d _ { \\phi } ^ { \\pi }$ and $d _ { \\phi } ^ { \\mathcal { D } }$ . ",
+        "bbox": [
+            173,
+            537,
+            825,
+            654
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "Nevertheless, RHS of Eq. (3) still cannot be evaluated naively due to its dependence on $d ^ { \\pi }$ in estimating the IPM. We address this challenge via a change of variable, which is known as a DualDICE trick (Nachum et al., 2019a). Define $\\nu : \\mathcal { Z } \\to \\mathbb { R }$ as an arbitrary function of state-action pairs that satisfies: ",
+        "bbox": [
+            174,
+            659,
+            825,
+            715
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "equation",
+        "img_path": "images/18aa3fba76be38e90e93d6e50a0189bfe6bb43416fb2cfa8334b5cd0ee563228.jpg",
+        "text": "$$\n\\nu \\big ( \\phi ( s , a ) \\big ) \\triangleq g \\big ( \\phi ( s , a ) \\big ) + \\gamma \\mathbb { E } _ { s ^ { \\prime } \\sim T ( s , a ) } \\Big [ \\nu \\big ( \\phi ( s ^ { \\prime } , a ^ { \\prime } ) \\big ) \\Big ] , \\forall ( s , a ) \\in \\mathcal { S } \\times \\mathcal { A } .\n$$",
+        "text_format": "latex",
+        "bbox": [
+            254,
+            717,
+            743,
+            752
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "Then we can rewrite the IPM as: ",
+        "bbox": [
+            174,
+            756,
+            392,
+            770
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "equation",
+        "img_path": "images/70a7132ac2d1c0745a05676ac0d5bdbaa8b1b9e9f9212c8bae7665f886215dc3.jpg",
+        "text": "$$\n\\begin{array} { r l } & { \\displaystyle \\mathrm { I P M } _ { \\mathcal { G } } ( d _ { \\phi } ^ { \\pi } , d _ { \\phi } ^ { \\mathcal { D } } ) = \\underset { g \\in \\mathcal { G } } { \\operatorname* { s u p } } \\left| \\mathbb { E } _ { ( s , a ) \\sim d ^ { \\pi } } \\Big [ g \\big ( \\phi ( s , a ) \\big ) \\Big ] - \\mathbb { E } _ { ( s , a ) \\sim d ^ { \\mathcal { D } } } \\Big [ g \\big ( \\phi ( s , a ) \\big ) \\Big ] \\right| } \\\\ & { \\displaystyle \\quad = \\operatorname* { s u p } _ { \\nu \\in \\mathcal { F } } \\left| ( 1 - \\gamma ) \\mathbb { E } \\qquad \\Big [ \\nu \\big ( \\phi ( s , a ) \\big ) \\Big ] - \\mathbb { E } _ { ( s , a , s ^ { \\prime } ) \\sim d ^ { \\mathcal { D } } } \\Big [ \\nu \\big ( \\phi ( s , a ) \\big ) - \\gamma \\nu \\big ( \\phi ( s ^ { \\prime } , a ^ { \\prime } ) \\big ) \\Big ] \\right| , } \\\\ & { \\displaystyle \\quad \\quad \\quad \\quad \\quad \\quad \\quad \\quad \\quad \\quad \\quad \\quad \\quad \\quad \\quad \\quad \\quad \\quad \\quad \\quad \\quad \\quad \\quad \\quad \\quad \\quad \\quad \\quad \\quad \\quad \\quad \\quad a ^ { \\prime } \\sim \\pi ^ { \\prime } } \\end{array}\n$$",
+        "text_format": "latex",
+        "bbox": [
+            189,
+            772,
+            785,
+            893
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "In other words, we are now taking a supremum over the new function class $\\mathcal { F }$ , which captures a function that returns the expected discounted sum of $g ( \\phi ( s _ { t } , a _ { t } ) )$ following the policy $\\pi$ in an MDP ",
+        "bbox": [
+            173,
+            895,
+            823,
+            925
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "$M$ given an initial representation $z$ . While it is now difficult to choose $\\mathcal { F }$ from $\\mathcal { G }$ , Eq. (3) still can be kept valid by using a sufficiently rich function class for $\\mathcal { F }$ . ",
+        "bbox": [
+            171,
+            103,
+            823,
+            132
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "text",
+        "text": "In this work, we choose $\\mathcal { F }$ to be the family of functions in the unit ball in a reproducing kernel Hilbert space (RKHS) $\\mathcal { H } _ { k }$ with the kernel $k$ , which allows the following closed-form formula (see Lemma A.3 in Appendix for details): ",
+        "bbox": [
+            173,
+            138,
+            825,
+            181
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "equation",
+        "img_path": "images/1953b31a1cd60337242d69b916f242145a0e39f20e0603f1049a6db643290d90.jpg",
+        "text": "$$\n\\begin{array} { r l r } { \\mathrm { P M } _ { \\mathcal { G } } ( d _ { \\phi } ^ { \\pi } , d _ { \\phi } ^ { \\mathcal { D } } ) ^ { 2 } = \\mathbb { E } _ { s _ { 0 } \\sim d _ { 0 } , a _ { 0 } \\sim \\pi ( s _ { 0 } ) , ( s , a , s ^ { \\prime } ) \\sim d ^ { \\mathcal { D } } , a ^ { \\prime } \\sim \\pi ( s ^ { \\prime } ) } } & { } & { ( 5 ) } \\\\ { \\bar { s } _ { 0 } \\sim d _ { 0 } , \\bar { u } _ { 0 } \\sim d _ { 0 } ) , ( \\bar { s } _ { 0 } , \\bar { u } , s ^ { \\prime } ) \\sim d ^ { \\mathcal { D } } , a ^ { \\prime } \\sim \\pi ( \\bar { s } ^ { \\prime } ) } & { } & { } \\\\ { k \\big ( \\phi ( s , a ) , \\phi ( \\bar { s } , \\bar { a } ) \\big ) + ( 1 - \\gamma ) ^ { 2 } k \\big ( \\phi ( s _ { 0 } , a _ { 0 } ) , \\phi ( \\bar { s } _ { 0 } , \\bar { a } _ { 0 } ) \\big ) + \\gamma ^ { 2 } k \\big ( \\phi ( s ^ { \\prime } , a ^ { \\prime } ) , \\phi ( \\bar { s } ^ { \\prime } , \\bar { a } ^ { \\prime } ) \\big ) } & { } & { } \\\\ { - 2 ( 1 - \\gamma ) k \\big ( \\phi ( s _ { 0 } , a _ { 0 } ) , \\phi ( \\bar { s } , \\bar { a } ) \\big ) - 2 \\gamma k \\big ( \\phi ( s ^ { \\prime } , a ^ { \\prime } ) , \\phi ( \\bar { s } , \\bar { a } ) \\big ) + 2 \\gamma ( 1 - \\gamma ) k \\big ( \\phi ( s ^ { \\prime } , a ^ { \\prime } ) , \\phi ( \\bar { s } _ { 0 } , \\bar { a } _ { 0 } ) \\big ) \\bigg ] . } & { } & \\end{array}\n$$",
+        "text_format": "latex",
+        "bbox": [
+            181,
+            184,
+            828,
+            279
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "text",
+        "text": "This completes our derivation for the tractable upper bound of policy evaluation error (Eq. (3)), whose direct dependence on $d ^ { \\pi }$ is eliminated by Eq. (5). Finally, we can train a model by minimizing the upper bound that encourages us to learn balanced representation while improving data fitness, where the trained model can readily provide a model-based OPE. ",
+        "bbox": [
+            173,
+            281,
+            825,
+            338
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "text",
+        "text": "The $\\mathrm { I P M } _ { \\mathcal { G } } ( d _ { \\phi } ^ { \\pi } , d _ { \\phi } ^ { D } ) ^ { 2 }$ in Eq. (5) can be estimated via finite random samples, and we denote its sampled-based estimator as $\\widehat { \\mathrm { I P M } } ( d _ { \\phi } ^ { \\pi } , d _ { \\phi } ^ { D } ) ^ { 2 }$ . We show in the following that, a valid upper bound can be established based on the sample-based estimators instead of the exact terms in the RHS of Eq. (3) under certain conditions. ",
+        "bbox": [
+            173,
+            343,
+            825,
+            407
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "text",
+        "text": "Theorem 4.3. Given an MDP $M$ , its estimate $\\widehat { M }$ with a bijective representation function $\\phi$ , i.e. $( \\widehat { T } , \\widehat { R } ) = \\langle \\widehat { T } _ { z } \\circ \\phi , \\widehat { R } _ { z } \\circ \\phi \\rangle$ , and an RKHS $\\begin{array} { r } { \\mathcal { H } _ { k } \\quad \\subset \\quad ( \\mathcal { Z } \\quad \\to \\quad \\mathbb { R } ) } \\end{array}$ induced by a universal kernel $k$ such that $\\mathrm { s u p } _ { z \\in \\mathcal { Z } } k ( z , z ) ~ = ~ \\bar { k }$ , assume that $f _ { \\phi , \\widehat { R } _ { z } , \\widehat { T } _ { z } } \\left( \\widehat { z } \\right) \\ =$ $\\begin{array} { r } { \\mathbb { E } _ { T , \\pi } \\Big [ \\sum _ { t = 0 } ^ { \\infty } \\gamma ^ { t } \\mathcal { E } _ { \\phi , \\widehat { R } _ { z } , \\widehat { T } _ { z } } \\big ( s _ { t } , a _ { t } \\big ) \\Big | \\big ( s _ { 0 } , a _ { 0 } \\big ) = \\phi ^ { - 1 } \\big ( z ) \\Big ] \\in \\mathcal { H } _ { k } } \\end{array}$ with $B _ { \\phi } = \\| f _ { \\phi , \\widehat { R } _ { z } , \\widehat { T } _ { z } } \\| _ { \\mathcal { H } _ { k } }$ and the loss is bounded by $\\begin{array} { r } { \\bar { \\mathcal { E } } = \\operatorname* { s u p } _ { s \\in { \\mathcal { S } } , a \\in \\mathcal { A } } \\mathcal { E } _ { \\phi , \\widehat { R } _ { z } , \\widehat { T } _ { z } } ( s , a ) } \\end{array}$ . Let n be the number of data in $\\mathcal { D }$ . With probability $1 - 2 \\delta$ , ",
+        "bbox": [
+            173,
+            410,
+            825,
+            515
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "equation",
+        "img_path": "images/1cfc254f1512a52e679a50996bd4958636bc46ff2883210d61add703a1dd3887.jpg",
+        "text": "$$\nR ^ { \\pi } - { \\widehat R } ^ { \\pi } \\Big | \\leq \\frac { 1 } { n } \\sum _ { ( s , a ) \\in { \\mathcal D } } \\mathcal { E } _ { \\phi , { \\widehat R } _ { z } , { \\widehat T } _ { z } } ( s , a ) + B _ { \\phi } \\widehat { \\Gamma \\mathrm { M } } \\big ( d _ { \\phi } ^ { \\pi } , d _ { \\phi } ^ { { D } } \\big ) + \\sqrt { \\frac { \\bar { \\mathcal E } ^ { 2 } } { 2 n } \\log \\frac { 1 } { \\delta } } + B _ { \\phi } \\sqrt { \\frac { \\bar { k } } { n } } \\left( 4 + \\sqrt { 8 \\log \\frac { 3 } { \\delta } } \\right) .\n$$",
+        "text_format": "latex",
+        "bbox": [
+            181,
+            516,
+            826,
+            561
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "text",
+        "text": "This result can be proved by adapting the convergence results of the empirical estimate of the MMD (Gretton et al., 2012) and Hoeffding’s inequality (Hoeffding, 1963). With the choice of an RKHS $\\mathcal { H } _ { k }$ , we can now interpret $B _ { \\phi }$ as the RKHS norm $\\| f _ { \\phi , \\widehat { R } _ { z } , \\widehat { T } _ { z } } \\| _ { \\mathcal { H } _ { k } }$ , which captures the magbnitude and the smoothness of the expected cumulative model loss $f _ { \\phi , \\widehat { R } _ { z } , \\widehat { T } _ { z } }$ . In general, assuming smooth underlying dynamics, we can expect $B _ { \\phi }$ to be small when the model error is small. Furthermore, although $\\bar { k }$ depends on the kernel function we use, we can always let √ $\\bar { k } = 1$ and subsume it into $B _ { \\phi }$ as long as it is bounded, i.e. using $\\tilde { B } _ { \\phi } \\triangleq B _ { \\phi } \\sqrt { \\bar { k } }$ . In the next section, we develop algorithms based on practical approximations of Eq. (3). ",
+        "bbox": [
+            173,
+            570,
+            826,
+            693
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "text",
+        "text": "Detailed comparison to RepBM As previously stated, RepBM (Liu et al., 2018b) is a modelbased finite-horizon OPE algorithm that trains the model to have balanced representation $\\phi$ , which is encouraged to be invariant under the target and behavior policies. Specifically, given the deterministic target policy $\\pi$ and the behavior policy $\\mu$ , at each timestep $t$ , it defines the factual distribution on $\\mathcal { Z }$ given that the actions until timestep $t$ have been executed according to the policy $\\pi$ : $p _ { \\phi , t } ^ { F } ( z ) = \\bar { \\mathrm { P r } } ( z _ { t } | M , \\mu , a _ { 0 : t } = \\pi ( s _ { 0 : t } ) )$ and the counterfactual distribution on $\\mathcal { Z }$ given the same condition except the action at timestep $t$ : $p _ { \\phi , t } ^ { C F } ( z ) = \\mathrm { P r } ( z _ { t } | M , \\mu , a _ { 0 : t - 1 } = \\pi ( s _ { 0 : t - 1 } ) , a _ { t } \\neq \\pi ( s _ { t } ) )$ . Then, RepBM bounds the OPE error as,1 ",
+        "bbox": [
+            173,
+            705,
+            825,
+            825
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "equation",
+        "img_path": "images/89841c17654cdd55d12f821f9c6ff9daaeed1eb8fcc65fba27a07c65eb3a0a3a.jpg",
+        "text": "$$\nR ^ { \\pi } - \\widehat { R } ^ { \\pi } \\Big | \\leq ( 1 - \\gamma ) \\sum _ { t = 0 } ^ { \\infty } \\gamma ^ { t } \\left( \\mathbb { E } _ { p ( s _ { t } | M , \\mu , a _ { 0 : t } = \\pi ( s _ { 0 : t } ) ) } [ \\mathcal { E } _ { \\phi , \\widehat { R } _ { z } , \\widehat { T } _ { z } } ( s _ { t } , \\pi ( s _ { t } ) ) ] + B _ { \\phi , t } [ \\mathrm { P M } _ { \\mathcal { G } _ { t } } ( p _ { \\phi , t } ^ { F } , p _ { \\phi , t } ^ { C F } ) \\right) .\n$$",
+        "text_format": "latex",
+        "bbox": [
+            181,
+            827,
+            828,
+            868
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "text",
+        "text": "Although RepBM achieves performance improvement over other OPE algorithms, we found a number of practical challenges: from the definition of the $\\mathrm { I P M } _ { \\mathcal { G } _ { t } } ( p _ { \\phi , t } ^ { F } , p _ { \\phi , t } ^ { C F } )$ , it requires a discrete-action environment and a deterministic policy $\\pi$ , which cannot be met by many practical RL settings. In addition, since the sample-based estimation of $\\mathrm { I P M } _ { \\mathcal { G } _ { t } } ( p _ { \\phi , t } ^ { F } , p _ { \\phi , t } ^ { C F } )$ requires samples consistent with the policy $\\pi$ , i.e. $a _ { 0 : t - 1 } = \\pi ( s _ { 0 : t - 1 } )$ , the algorithm would reject exponentially many samples with respect to $t$ , which is the curse of horizon (Liu et al., 2018a). When there is a large difference between the behavior and the target policies in long-term tasks, their implementation becomes close to using the maximum likelihood objective, which can also be observed empirically in our experiments. In contrast, our work is free from the abovementioned practical limitations by performing balancing between the discounted stationary distribution $d ^ { \\pi }$ and the data distribution $d ^ { \\bar { D } }$ , leveraging the recent advances in stationary distribution correction estimation (i.e. the DualDICE trick) to overcome the difficulties pertinent to the expectation concerning $d ^ { \\pi }$ required to evaluate the IPM in the objective. ",
+        "bbox": [
+            174,
+            871,
+            823,
+            901
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "text",
+        "text": "",
+        "bbox": [
+            173,
+            103,
+            825,
+            246
+        ],
+        "page_idx": 5
+    },
+    {
+        "type": "text",
+        "text": "4.2 REPRESENTATION BALANCING WITH STATIONARY DISTRIBUTION ESTIMATION ",
+        "bbox": [
+            171,
+            261,
+            763,
+            276
+        ],
+        "page_idx": 5
+    },
+    {
+        "type": "text",
+        "text": "In the following, we describe algorithms for OPE and offline RL based on the practical approximations to Eq. (3), which we call the RepB-SDE framework. ",
+        "bbox": [
+            173,
+            286,
+            823,
+            315
+        ],
+        "page_idx": 5
+    },
+    {
+        "type": "text",
+        "text": "Objective for off-policy evaluation As we mentioned earlier, we aim to minimize the upper bound of OPE error $| R ^ { \\pi } - { \\widehat { R } } ^ { \\pi } |$ specified in Theorem 4.2. To make the RHS of Eq. (3) tractable for optimization, we replace the intractable total variation distance with KL-divergence, which can be easily minimized by maximizing the data log-likelihood. We also replace the IPM with its samplebased estimator to obtain the learning objective: ",
+        "bbox": [
+            174,
+            330,
+            823,
+            404
+        ],
+        "page_idx": 5
+    },
+    {
+        "type": "equation",
+        "img_path": "images/0237e51f5b62910fe3675c51bd2d0830f1b1eb5740055f62c049167cc00a62aa.jpg",
+        "text": "$$\n\\operatorname* { m i n } _ { \\phi , \\widehat { R } _ { z } , \\widehat { T } _ { z } } \\frac { 1 } { n } \\sum _ { \\substack { ( s , a , s ^ { \\prime } , r ) \\in \\mathcal { D } } } \\Big [ \\underbrace { - \\log \\widehat { R } _ { z } \\big ( r | \\phi ( s , a ) \\big ) } _ { D _ { \\mathrm { K L } } ( R ( r | s , a ) | | \\widehat { R } _ { z } ( r | \\phi ( s , a ) ) ) D _ { \\mathrm { K L } } ( T ( s ^ { \\prime } | s , a ) | | \\widehat { T } _ { z } ( s ^ { \\prime } | \\phi ( s , a ) ) ) } \\Big ] + \\alpha _ { M } \\widehat { \\mathrm { I P M } } ( d _ { \\phi } ^ { \\pi } , d _ { \\phi } ^ { \\mathcal { D } } )\n$$",
+        "text_format": "latex",
+        "bbox": [
+            181,
+            410,
+            795,
+            458
+        ],
+        "page_idx": 5
+    },
+    {
+        "type": "text",
+        "text": "The constant $B _ { \\phi }$ in Theorem 4.2 depends on the function classes and cannot be estimated, and thus, we replace it with a tunable hyperparameter $\\alpha _ { M }$ that balances between data fitness and representation invariance. Remark that $\\alpha _ { M } = 0$ recovers the simple maximum-likelihood objective. ",
+        "bbox": [
+            176,
+            463,
+            825,
+            506
+        ],
+        "page_idx": 5
+    },
+    {
+        "type": "text",
+        "text": "By simulating the target policy $\\pi$ under the environment model $( \\widehat { T } , \\widehat { R } )$ obtained by minimizing Eq. (6), it is possible to perform a model-based OPE that approximately minimizes the upper bound of OPE error. ",
+        "bbox": [
+            174,
+            512,
+            825,
+            556
+        ],
+        "page_idx": 5
+    },
+    {
+        "type": "text",
+        "text": "Objectives for offline model-based RL By rearranging Eq. (3), we have, ",
+        "bbox": [
+            174,
+            570,
+            669,
+            587
+        ],
+        "page_idx": 5
+    },
+    {
+        "type": "equation",
+        "img_path": "images/d8af21f1c1d33ba304d1be665588f047c405ec5681877518b5352d4c61c8146e.jpg",
+        "text": "$$\n\\begin{array} { r } { { \\cal R } ^ { \\pi } \\geq \\widehat { \\cal R } ^ { \\pi } - \\mathbb { E } _ { ( s , a ) \\sim \\mathcal { D } } \\left[ \\mathcal { E } _ { \\phi , \\widehat { \\cal R } _ { z } , \\widehat { T } _ { z } } ( s , a ) \\right] - { \\cal B } _ { \\phi } \\mathrm { I P M } _ { \\mathcal { G } } ( d _ { \\phi } ^ { \\pi } , d _ { \\phi } ^ { \\mathcal { D } } ) . } \\end{array}\n$$",
+        "text_format": "latex",
+        "bbox": [
+            297,
+            592,
+            699,
+            619
+        ],
+        "page_idx": 5
+    },
+    {
+        "type": "text",
+        "text": "Then, we can maximize the RHS of Eq. (7) to get the model and the policy that maximizes the lower bound of true return $R ^ { \\pi }$ . Similar to the derivation of Eq. (6), we replace the total variation distance by KL-divergence to obtain the following learning objectives: ",
+        "bbox": [
+            174,
+            625,
+            826,
+            666
+        ],
+        "page_idx": 5
+    },
+    {
+        "type": "equation",
+        "img_path": "images/55972eb1ba29537874f0985674ed88ed72de9579cd85bfcb0b113cc436a53192.jpg",
+        "text": "$$\n\\begin{array} { r l } & { \\mathcal { L } _ { \\widehat { M } } ( \\widehat { M } , \\pi , \\alpha _ { M } ) = \\mathbb { E } _ { d ^ { p } } \\Big [ - \\log \\widehat { R } _ { z } \\big ( r | \\phi ( s , a ) \\big ) - \\log \\widehat { T } _ { z } \\big ( s ^ { \\prime } | \\phi ( s , a ) \\big ) \\Big ] + \\alpha _ { M } \\mathrm { I P M } _ { \\mathcal { G } } \\big ( d _ { \\phi } ^ { \\pi } , d _ { \\phi } ^ { { D } } \\big ) , } \\\\ & { \\mathcal { I } _ { \\pi } ( \\pi , \\widehat { M } , \\alpha _ { \\pi } ) = \\mathbb { E } _ { \\widehat { M } , \\pi } \\left[ \\displaystyle \\sum _ { t = 0 } ^ { \\infty } \\gamma ^ { t } r _ { t } \\right] - \\alpha _ { \\pi } \\mathrm { I P M } _ { \\mathcal { G } } \\big ( d _ { \\phi } ^ { \\pi } , d _ { \\phi } ^ { { D } } \\big ) . } \\end{array}\n$$",
+        "text_format": "latex",
+        "bbox": [
+            189,
+            671,
+            787,
+            744
+        ],
+        "page_idx": 5
+    },
+    {
+        "type": "text",
+        "text": "where the expectation in Eq. (9) can be optimized using various model-based RL algorithms, e.g. with planning (Chua et al., 2018) or using a model-free learner (Janner et al., 2019). By iterating between the minimization of ${ \\mathcal { L } } _ { \\widehat { M } }$ with respect to $\\widehat { M }$ and the maximization of $\\mathcal { I } _ { \\pi }$ with respect to $\\pi$ cby stochastic gradient method, it is possible to perform offline model-based RL that approximately maximizes the lower bound of the true return $R ^ { \\pi }$ . ",
+        "bbox": [
+            173,
+            747,
+            825,
+            820
+        ],
+        "page_idx": 5
+    },
+    {
+        "type": "text",
+        "text": "Implementation details Following the recent practice (Chua et al., 2018; Janner et al., 2019; Yu et al., 2020), we model the dynamics $( \\widehat { T } , \\widehat { R } )$ using a bootstrap ensemble of neural networks. To optimize a policy based on the objective, we perform full rollouts (until reaching the terminal states or maximum timesteps) using learned dynamics $( \\widehat { T } , \\widehat { R } )$ . During obtaining the rollouts, we pessimistically augment the estimated reward function using the penalty proportional to the bootstrapped uncertainty, which helped the algorithm to perform robustly. We suspect the difficulty in calculating accurate IPM estimation is what makes the additional pessimism beneficial. We store the generated experiences to a separate dataset $\\widehat { \\mathcal { D } }$ and update the policy $\\pi$ with IPM-regularized soft actor-critic (SAC) (Haarnoja et al., 2018) using samples from both datasets $\\mathcal { D } \\cup \\widehat { \\mathcal { D } }$ similar to MBPO (Janner et al., 2019). ",
+        "bbox": [
+            173,
+            834,
+            825,
+            924
+        ],
+        "page_idx": 5
+    },
+    {
+        "type": "image",
+        "img_path": "images/f7e55e839858979ed3073f247288e2993a18881c7140dc8bb252ba65b110e1f2.jpg",
+        "image_caption": [
+            "Figure 1: The OPE results of different model learning algorithms with varying off-policyness on the $\\mathbf { X }$ -axis. The y-axis plots the normalized individual MSE on test trajectories where the performance of the baseline model is set to 1. The tasks used are infinite-horizon discounted environments $\\gamma = 0 . 9 8$ (HIV), $\\gamma = 0 . 9 9$ (others)), where we truncated at $t = 1 0 0 0$ . The experiments are repeated 200 times, and the error bars indicate $9 5 \\%$ confidence intervals. "
+        ],
+        "image_footnote": [],
+        "bbox": [
+            171,
+            101,
+            826,
+            251
+        ],
+        "page_idx": 6
+    },
+    {
+        "type": "text",
+        "text": "",
+        "bbox": [
+            174,
+            362,
+            825,
+            424
+        ],
+        "page_idx": 6
+    },
+    {
+        "type": "text",
+        "text": "Since the presented model objective requires a policy $\\pi$ to perform a balancing, we initially trained the model and the policy using $\\alpha _ { M } = \\alpha _ { \\pi } = 0$ : by ${ \\widehat { \\widetilde M } } _ { 0 } = \\arg \\operatorname* { m i n } _ { { \\widehat M } } { \\mathcal { L } } _ { { \\widehat M } } ^ { - } ( { \\widehat M } , \\cdot , 0 )$ and $\\pi _ { 0 } = \\arg \\operatorname* { m i n } _ { \\pi } \\mathcal { L } _ { \\pi } ( \\pi , \\widehat { M } _ { 0 } , 0 )$ . Then, we retrained the model and the policy using $\\pi _ { 0 }$ : $\\widehat { M } _ { 1 } \\ =$ $\\mathrm { a r g m i n } _ { \\widehat { M } } \\mathcal { L } _ { \\widehat { M } } ( \\widehat { M } , \\pi _ { 0 } , \\alpha _ { M } )$ and $\\begin{array} { r } { \\pi _ { 1 } = \\arg \\operatorname* { m i n } _ { \\pi } \\mathcal { L } _ { \\pi } ( \\pi , \\widehat { M } _ { 1 } , \\alpha _ { \\pi } ) } \\end{array}$ for some non-negative $\\alpha _ { M }$ and $\\alpha _ { \\pi }$ . c cWhile it is desirable to repeat the optimization of the model and the policy until convergence, we did not observe significant improvement after the first iteration and reported the performance of the policy after the first iteration, $\\pi _ { 1 }$ . ",
+        "bbox": [
+            173,
+            430,
+            825,
+            537
+        ],
+        "page_idx": 6
+    },
+    {
+        "type": "text",
+        "text": "5 EXPERIMENTS ",
+        "text_level": 1,
+        "bbox": [
+            176,
+            563,
+            326,
+            579
+        ],
+        "page_idx": 6
+    },
+    {
+        "type": "text",
+        "text": "We demonstrate the effectiveness of the RepB-SDE framework, by comparing the OPE performance of Eq. (6) to that of RepBM and evaluating the presented model-based offline RL algorithm on the benchmarks. The code used to produce the results is available online.2 A detailed description of the experiments can be found in Appendix B. ",
+        "bbox": [
+            174,
+            597,
+            825,
+            654
+        ],
+        "page_idx": 6
+    },
+    {
+        "type": "text",
+        "text": "5.1 MODEL-BASED OFF-POLICY EVALUATION ",
+        "text_level": 1,
+        "bbox": [
+            174,
+            676,
+            501,
+            689
+        ],
+        "page_idx": 6
+    },
+    {
+        "type": "text",
+        "text": "For the sake of comparison with RepBM, we test our OPE algorithm on three continuous-state discrete-action tasks where the goal is to evaluate a deterministic target policy. We trained a suboptimal deterministic target policy $\\pi$ and used an $\\epsilon$ -greedy policy with various values of $\\epsilon$ as the data collection policy. In each experiment, we trained environment models for a fixed number of epochs concerning three different objectives: simple maximum-likelihood baseline, step-wise representation balancing objective used in RepBM (Liu et al., 2018b), and the presented OPE objective of RepB-SDE (Eq. (6)). We measured the individual mean squared error (Liu et al., 2018b). ",
+        "bbox": [
+            173,
+            703,
+            825,
+            801
+        ],
+        "page_idx": 6
+    },
+    {
+        "type": "text",
+        "text": "The normalized error of each algorithm, relative to the error of the baseline valued at 1, is presented in Figure 1. We can observe that the presented objective of RepB-SDE can reduce the OPE error from the baseline significantly, outperforming RepBM in most of the cases. As the off-policyness between the policies (\u000f) increases, representation balancing algorithms should more benefit compared to the maximum-likelihood baseline in principle. However, the result shows that the performance of RepBM merely increases due to the increased sample rejection rate under large $\\epsilon$ . ",
+        "bbox": [
+            174,
+            808,
+            825,
+            891
+        ],
+        "page_idx": 6
+    },
+    {
+        "type": "table",
+        "img_path": "images/49f566ca115dde3869f3a7c649db4ca701dd408c978bc5e3af8a0e88cc803964.jpg",
+        "table_caption": [
+            "Table 1: Normalized scores on D4RL MuJoCo benchmark datasets (Fu et al., 2020) where the score of 0 corresponds to a random policy and 100 corresponds to a converged SAC policy. All results (except MF, which is taken from Fu et al. (2020) and Kumar et al. (2020)) are averaged over 5 runs, where $\\pm$ denotes the standard error. The highest scores are highlighted with boldface. "
+        ],
+        "table_footnote": [],
+        "table_body": "<table><tr><td>Dataset type</td><td>Environment</td><td>RepB-SDE (ours)</td><td>RP</td><td>Base</td><td>MOPO</td><td>MF</td><td>BC</td></tr><tr><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td></tr><tr><td>Random</td><td>Walker2d</td><td>21.1 ± 1.0</td><td>18.4</td><td>16.4</td><td>1.3</td><td>7.3 BEAR</td><td>0.0</td></tr><tr><td>Medium</td><td>Walker2d</td><td>72.1 ± 1.9</td><td>56.3</td><td>5.5</td><td>-0.1</td><td>81.1 BRAC</td><td>7.7</td></tr><tr><td>Med-Replay</td><td>Walker2d</td><td>49.8 ± 11.4</td><td>41.1</td><td>6.2</td><td>47.8</td><td>26.7 CQL</td><td>8.0</td></tr><tr><td>Med-Expert</td><td>Walker2d</td><td>88.8 ± 6.9</td><td>72.6</td><td>51.7</td><td>32.4</td><td>98.7 CQL</td><td>3.3</td></tr><tr><td>Random</td><td>Hopper</td><td>8.6 ±1.0</td><td>8.3</td><td>8.3</td><td>9.1</td><td>12.2 BRAC</td><td>9.0</td></tr><tr><td>Medium</td><td>Hopper</td><td>34.0 ± 2.8</td><td>27.5</td><td>19.8</td><td>19.2</td><td>58.0 CQL</td><td>34.5</td></tr><tr><td>Med-Replay</td><td>Hopper</td><td>62.2 ± 6.7</td><td>49.8</td><td>32.9</td><td>80.8</td><td>48.6 CQL</td><td>13.2</td></tr><tr><td>Med-Expert</td><td>Hopper</td><td>82.6 ± 7.0</td><td>74.0</td><td>19.1</td><td>23.2</td><td>111.0 CQL</td><td>41.9</td></tr><tr><td>Random</td><td>HalfCheetah</td><td>32.9 ± 1.1</td><td>31.3</td><td>26.1</td><td>29.9</td><td>35.4 CQL</td><td>-2.4</td></tr><tr><td>Medium</td><td>HalfCheetah</td><td>49.1 ± 0.3</td><td>47.3</td><td>22.1</td><td>45.0</td><td>46.3 BRAC</td><td>31.0</td></tr><tr><td>Med-Replay</td><td>HalfCheetah</td><td>57.5 ± 0.8</td><td>53.6</td><td>54.0</td><td>53.8</td><td>47.7 BRAC</td><td>14.7</td></tr><tr><td>Med-Expert</td><td>HalfCheetah</td><td>55.4 ± 8.3</td><td>36.7</td><td>31.9</td><td>91.0</td><td>64.7 BCQ</td><td>29.5</td></tr></table>",
+        "bbox": [
+            176,
+            172,
+            821,
+            382
+        ],
+        "page_idx": 7
+    },
+    {
+        "type": "text",
+        "text": "5.2 OFFLINE REINFORCEMENT LEARNING WITH REPRESENTATION BALANCED MODEL ",
+        "bbox": [
+            169,
+            422,
+            777,
+            435
+        ],
+        "page_idx": 7
+    },
+    {
+        "type": "text",
+        "text": "We evaluate the offline model-based RL algorithm presented in Section 4.2 on a subset of datasets in the D4RL benchmark (Fu et al., 2020): using four types of datasets (Random, Medium, MediumReplay, and Medium-Expert) from three different MuJoCo environments (HalfCheetah-v2, Hopperv2, and Walker2d-v2) (Todorov et al., 2012). Random dataset contains $1 0 ^ { 6 }$ experience tuples from a random policy. Medium dataset contains $1 0 ^ { 6 }$ experience tuples from a policy trained to approximately $1 / 3$ the performance of the expert, which is an agent trained to completion with SAC. Med-Replay dataset contains $1 0 ^ { 5 }$ $2 \\times \\mathrm { i 0 ^ { 5 } }$ for Walker2d-v2) experience tuples, which are from the replay buffer of a policy trained up to the performance of the medium agent. Med-Expert dataset is a Medium dataset combined with $1 0 ^ { 6 }$ samples from the expert. This experimental setting exactly follows that of (Yu et al., 2020; Argenson & Dulac-Arnold, 2020). ",
+        "bbox": [
+            173,
+            450,
+            825,
+            590
+        ],
+        "page_idx": 7
+    },
+    {
+        "type": "text",
+        "text": "The normalized score of each algorithm is presented in Table 1. MF denotes the best score from offline model-free algorithms (taken from Fu et al. (2020) and Kumar et al. (2020)), including SAC (Haarnoja et al., 2018), BCQ (Fujimoto et al., 2019), BEAR (Kumar et al., 2019), BRAC (Wu et al., 2019), AWR (Peng et al., 2019), cREM (Agarwal et al., 2020), AlgaeDICE (Nachum et al., 2019b), and CQL (Kumar et al., 2020). The actual algorithm that achieves the reported score is presented next to the numbers. Base shows the performance of the most naive baseline, which attempts to maximize the estimated policy return under the maximum-likelihood model. RP denotes the performance of Base equipped with the appropriate reward penalty using the bootstrapped uncertainty of the model, which is equivalent to $\\pi _ { 0 }$ described in Section 4.2. RepB-SDE denotes the performance after a single iteration of our algorithm, corresponding to $\\pi _ { 1 }$ . We also provide BC, the performance of direct behavior cloning from the data, and MOPO (Yu et al., 2020), an offline model-based RL algorithm that optimizes policy based on truncated rollouts with the heuristic reward penalty. ",
+        "bbox": [
+            173,
+            597,
+            825,
+            777
+        ],
+        "page_idx": 7
+    },
+    {
+        "type": "text",
+        "text": "The significant gap between RepB-SDE and RP in the results shows the advantage brought by our framework that encourages balanced representation. While our approach was less successful on some of the datasets (mostly on the Hopper-v2 environment), we hypothesize that the conservative training techniques: the behavior regularization approaches exploited in the model-free algorithms, the rollout truncation technique in MOPO, and the pessimistic training based on the bootstrapped uncertainty estimates adopted in our algorithm exhibit their strengths in different datasets. For example, it may be the case that the ensemble models are overconfident especially in Hopper-v2, and should be regularized with more explicit methods. Nevertheless, we emphasize that the presented framework can be used jointly with any other conservative training technique to improve their performance. ",
+        "bbox": [
+            174,
+            785,
+            825,
+            924
+        ],
+        "page_idx": 7
+    },
+    {
+        "type": "text",
+        "text": "6 CONCLUSION AND FUTURE WORK ",
+        "text_level": 1,
+        "bbox": [
+            174,
+            102,
+            495,
+            117
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "In this paper, we presented RepB-SDE, a framework for balancing the model representation with stationary distribution estimation, aiming at obtaining a model robust to the distribution shift that arises in off-policy and offline RL. We started from the theoretical observation that the model-based policy evaluation error can be upper-bounded by the data fitness and the distance between two distributions in the representation space. Motivated by the bound, we presented a model learning objective for off-policy evaluation and model-based offline policy optimization. RepB-SDE can be seen as an extension of RepBM, which addresses the curse of horizon by leveraging the recent advances in stationary distribution correction estimation (i.e. the DualDICE trick). Using stationary distribution also frees us from other limitations of RepBM to be applied to more practical settings. To the best of our knowledge, it is the first attempt to introduce an augmented objective for the learning of model robust to a specific distribution shift in offline RL. ",
+        "bbox": [
+            174,
+            135,
+            825,
+            286
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "In the experiments, we empirically demonstrated that we can significantly reduce the OPE error from the baseline, outperforming RepBM in most cases. We also showed that the robust model also helps in the offline model-based policy optimization, yielding the state-of-the-art performance in a representative set of D4RL benchmarks. We emphasize that our approach can be directly adopted in many other model-based offline RL algorithms. ",
+        "bbox": [
+            174,
+            294,
+            823,
+            363
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "There are a number of promising directions for future work. Most importantly, we have not leveraged the learned representation in the policy when optimizing the policy, yet it is very natural to do so. We can easily incorporate the representation into the policy by assuming energy-based models, but this would make the computation of entropy intractable in entropy-regularized policy optimization algorithms. It would be also interesting to see if the proposed framework for learning balanced representation can benefit off-policy (and offline) model-free methods. ",
+        "bbox": [
+            174,
+            371,
+            825,
+            454
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "ACKNOWLEDGMENTS ",
+        "text_level": 1,
+        "bbox": [
+            176,
+            473,
+            326,
+            484
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "This work was supported by the National Research Foundation (NRF) of Korea (NRF2019M3F2A1072238 and NRF-2019R1A2C1087634), and the Ministry of Science and Information communication Technology (MSIT) of Korea (IITP No. 2019-0-00075, IITP No. 2020-0-00940 and IITP No. 2017-0-01779 XAI). ",
+        "bbox": [
+            174,
+            497,
+            825,
+            551
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "REFERENCES ",
+        "text_level": 1,
+        "bbox": [
+            176,
+            575,
+            285,
+            590
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "Rishabh Agarwal, Dale Schuurmans, and Mohammad Norouzi. An optimistic perspective on offline reinforcement learning. In International Conference on Machine Learning, 2020. ",
+        "bbox": [
+            174,
+            599,
+            823,
+            627
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "Arthur Argenson and Gabriel Dulac-Arnold. Model-based offline planning. arXiv preprint arXiv:2008.05556, 2020. ",
+        "bbox": [
+            169,
+            638,
+            825,
+            667
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "Peter L Bartlett and Shahar Mendelson. Rademacher and Gaussian complexities: Risk bounds and structural results. Journal of Machine Learning Research, 3(Nov):463–482, 2002. ",
+        "bbox": [
+            173,
+            679,
+            823,
+            708
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "Shai Ben-David, John Blitzer, Koby Crammer, and Fernando Pereira. Analysis of representations for domain adaptation. In Advances in Neural Information Processing Systems, 2007. ",
+        "bbox": [
+            174,
+            719,
+            823,
+            750
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "Greg Brockman, Vicki Cheung, Ludwig Pettersson, Jonas Schneider, John Schulman, Jie Tang, and Wojciech Zaremba. OpenAI Gym. arXiv preprint arXiv:1606.01540, 2016. ",
+        "bbox": [
+            174,
+            760,
+            823,
+            789
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "Kurtland Chua, Roberto Calandra, Rowan McAllister, and Sergey Levine. Deep reinforcement learning in a handful of trials using probabilistic dynamics models. In Advances in Neural Information Processing Systems, 2018. ",
+        "bbox": [
+            174,
+            800,
+            821,
+            843
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "Justin Fu, Aviral Kumar, Ofir Nachum, George Tucker, and Sergey Levine. D4RL: Datasets for deep data-driven reinforcement learning. arXiv preprint arXiv:2004.07219, 2020. ",
+        "bbox": [
+            169,
+            854,
+            823,
+            883
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "Scott Fujimoto, David Meger, and Doina Precup. Off-policy deep reinforcement learning without exploration. In International Conference on Machine Learning, 2019. ",
+        "bbox": [
+            174,
+            895,
+            820,
+            924
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "Carles Gelada and Marc G Bellemare. Off-policy deep reinforcement learning by bootstrapping the covariate shift. In AAAI Conference on Artificial Intelligence, 2019. ",
+        "bbox": [
+            171,
+            103,
+            825,
+            132
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "Arthur Gretton, Karsten M Borgwardt, Malte J Rasch, Bernhard Scholkopf, and Alexander Smola.¨ A kernel two-sample test. Journal of Machine Learning Research, 13(Mar):723–773, 2012. ",
+        "bbox": [
+            173,
+            140,
+            823,
+            170
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "Tuomas Haarnoja, Aurick Zhou, Pieter Abbeel, and Sergey Levine. Soft actor-critic: Off-policy maximum entropy deep reinforcement learning with a stochastic actor. In International Conference on Machine Learning, 2018. ",
+        "bbox": [
+            176,
+            178,
+            821,
+            220
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "Assaf Hallak and Shie Mannor. Consistent on-line off-policy evaluation. In International Conference on Machine Learning, 2017. ",
+        "bbox": [
+            171,
+            228,
+            823,
+            258
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "Wassily Hoeffding. Probability inequalities for sums of bounded random variables. Journal of the American Statistical Association, 58(301):13–30, 1963. ",
+        "bbox": [
+            174,
+            266,
+            823,
+            295
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "Michael Janner, Justin Fu, Marvin Zhang, and Sergey Levine. When to trust your model: Modelbased policy optimization. In Advances in Neural Information Processing Systems, 2019. ",
+        "bbox": [
+            174,
+            303,
+            821,
+            333
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "Nan Jiang and Lihong Li. Doubly robust off-policy value evaluation for reinforcement learning. In International Conference on Machine Learning, 2016. ",
+        "bbox": [
+            174,
+            340,
+            823,
+            371
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "Fredrik D Johansson, Nathan Kallus, Uri Shalit, and David Sontag. Learning weighted representations for generalization across designs. arXiv preprint arXiv:1802.08598, 2018. ",
+        "bbox": [
+            171,
+            377,
+            821,
+            407
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "Michael Kearns and Satinder Singh. Near-optimal reinforcement learning in polynomial time. Machine learning, 49(2-3):209–232, 2002. ",
+        "bbox": [
+            171,
+            415,
+            823,
+            445
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "Rahul Kidambi, Aravind Rajeswaran, Praneeth Netrapalli, and Thorsten Joachims. MOReL: Modelbased offline reinforcement learning. In Advances in Neural Information Processing Systems, 2020. ",
+        "bbox": [
+            174,
+            452,
+            823,
+            496
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "Diederik P Kingma and Jimmy Ba. Adam: A method for stochastic optimization. arXiv preprint arXiv:1412.6980, 2014. ",
+        "bbox": [
+            173,
+            503,
+            823,
+            532
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "Aviral Kumar, Justin Fu, Matthew Soh, George Tucker, and Sergey Levine. Stabilizing off-policy q-learning via bootstrapping error reduction. In Advances in Neural Information Processing Systems, 2019. ",
+        "bbox": [
+            174,
+            541,
+            823,
+            584
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "Aviral Kumar, Aurick Zhou, George Tucker, and Sergey Levine. Conservative Q-learning for offline reinforcement learning. In Advances in Neural Information Processing Systems, 2020. ",
+        "bbox": [
+            171,
+            592,
+            823,
+            622
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "Byung-Jun Lee, Jongmin Lee, Peter Vrancx, Dongho Kim, and Kee-Eung Kim. Batch reinforcement learning with hyperparameter gradients. In International Conference on Machine Learning, 2020. ",
+        "bbox": [
+            174,
+            628,
+            823,
+            660
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "Sergey Levine, Aviral Kumar, George Tucker, and Justin Fu. Offline reinforcement learning: Tutorial, review, and perspectives on open problems. arXiv preprint arXiv:2005.01643, 2020. ",
+        "bbox": [
+            173,
+            666,
+            821,
+            696
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "Timothy P Lillicrap, Jonathan J Hunt, Alexander Pritzel, Nicolas Heess, Tom Erez, Yuval Tassa, David Silver, and Daan Wierstra. Continuous control with deep reinforcement learning. In International Conference on Learning Representations, 2016. ",
+        "bbox": [
+            176,
+            704,
+            821,
+            747
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "Qiang Liu, Lihong Li, Ziyang Tang, and Dengyong Zhou. Breaking the curse of horizon: Infinitehorizon off-policy estimation. In Advances in Neural Information Processing Systems, 2018a. ",
+        "bbox": [
+            171,
+            755,
+            823,
+            785
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "Yao Liu, Omer Gottesman, Aniruddh Raghu, Matthieu Komorowski, Aldo A Faisal, Finale DoshiVelez, and Emma Brunskill. Representation balancing MDPs for off-policy policy evaluation. In Advances in Neural Information Processing Systems, 2018b. ",
+        "bbox": [
+            176,
+            792,
+            821,
+            835
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "Yao Liu, Adith Swaminathan, Alekh Agarwal, and Emma Brunskill. Off-policy policy gradient with state distribution correction. arXiv preprint arXiv:1904.08473, 2019. ",
+        "bbox": [
+            173,
+            843,
+            821,
+            873
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "Tatsuya Matsushima, Hiroki Furuta, Yutaka Matsuo, Ofir Nachum, and Shixiang Gu. Deploymentefficient reinforcement learning via model-based offline optimization. arXiv preprint arXiv:2006.03647, 2020. ",
+        "bbox": [
+            176,
+            882,
+            823,
+            924
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "Colin McDiarmid. On the method of bounded differences. Surveys in combinatorics, 141(1):148– 188, 1989. ",
+        "bbox": [
+            173,
+            103,
+            823,
+            132
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Volodymyr Mnih, Koray Kavukcuoglu, David Silver, Andrei A Rusu, Joel Veness, Marc G Bellemare, Alex Graves, Martin Riedmiller, Andreas K Fidjeland, Georg Ostrovski, et al. Human-level control through deep reinforcement learning. Nature, 518(7540):529–533, 2015. ",
+        "bbox": [
+            176,
+            140,
+            821,
+            184
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Ali Mousavi, Lihong Li, Qiang Liu, and Denny Zhou. Black-box off-policy estimation for infinitehorizon reinforcement learning. In International Conference on Learning Representations, 2020. ",
+        "bbox": [
+            171,
+            191,
+            823,
+            220
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Alfred Muller. Integral probability metrics and their generating classes of functions. ¨ Advances in Applied Probability, 29(2):429–443, 1997. ",
+        "bbox": [
+            173,
+            228,
+            823,
+            258
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Ofir Nachum, Yinlam Chow, Bo Dai, and Lihong Li. DualDICE: Behavior-agnostic estimation of discounted stationary distribution corrections. In Advances in Neural Information Processing Systems, 2019a. ",
+        "bbox": [
+            173,
+            266,
+            823,
+            309
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Ofir Nachum, Bo Dai, Ilya Kostrikov, Yinlam Chow, Lihong Li, and Dale Schuurmans. AlgaeDICE: Policy gradient from arbitrary experience. arXiv preprint arXiv:1912.02074, 2019b. ",
+        "bbox": [
+            171,
+            316,
+            823,
+            347
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Xue Bin Peng, Aviral Kumar, Grace Zhang, and Sergey Levine. Advantage-weighted regression: Simple and scalable off-policy reinforcement learning. arXiv preprint arXiv:1910.00177, 2019. ",
+        "bbox": [
+            171,
+            354,
+            821,
+            385
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Doina Precup. Eligibility traces for off-policy policy evaluation. Computer Science Department Faculty Publication Series, pp. 80, 2000. ",
+        "bbox": [
+            173,
+            392,
+            823,
+            421
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Prajit Ramachandran, Barret Zoph, and Quoc V Le. Swish: a self-gated activation function. arXiv preprint arXiv:1710.05941, 7, 2017. ",
+        "bbox": [
+            173,
+            429,
+            823,
+            458
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "John Schulman, Filip Wolski, Prafulla Dhariwal, Alec Radford, and Oleg Klimov. Proximal policy optimization algorithms. arXiv preprint arXiv:1707.06347, 2017. ",
+        "bbox": [
+            171,
+            467,
+            823,
+            496
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Uri Shalit, Fredrik D Johansson, and David Sontag. Estimating individual treatment effect: generalization bounds and algorithms. In International Conference on Machine Learning, 2017. ",
+        "bbox": [
+            171,
+            503,
+            821,
+            534
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Bharath K Sriperumbudur, Kenji Fukumizu, Arthur Gretton, Bernhard Scholkopf, and Gert RG ¨ Lanckriet. On integral probability metrics, $\\phi$ -divergences and binary classification. arXiv preprint arXiv:0901.2698, 2009. ",
+        "bbox": [
+            173,
+            541,
+            823,
+            584
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Richard S Sutton, Andrew G Barto, et al. Introduction to reinforcement learning, volume 135. MIT press Cambridge, 1998. ",
+        "bbox": [
+            171,
+            592,
+            823,
+            622
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Richard S Sutton, Hamid R Maei, and Csaba Szepesvari. A convergent ´ $o ( n )$ temporal-difference algorithm for off-policy learning with linear function approximation. In Advances in neural information processing systems, 2009. ",
+        "bbox": [
+            174,
+            628,
+            823,
+            672
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Phillip Swazinna, Steffen Udluft, and Thomas Runkler. Overcoming model bias for robust offline deep reinforcement learning. arXiv preprint arXiv:2008.05533, 2020. ",
+        "bbox": [
+            171,
+            680,
+            823,
+            710
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Ziyang Tang, Yihao Feng, Lihong Li, Dengyong Zhou, and Qiang Liu. Doubly robust bias reduction in infinite horizon off-policy estimation. In International Conference on Learning Representations, 2020. ",
+        "bbox": [
+            174,
+            718,
+            825,
+            761
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Philip Thomas and Emma Brunskill. Data-efficient off-policy policy evaluation for reinforcement learning. In International Conference on Machine Learning, 2016. ",
+        "bbox": [
+            173,
+            768,
+            821,
+            799
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Emanuel Todorov, Tom Erez, and Yuval Tassa. Mujoco: A physics engine for model-based control. In International Conference on Intelligent Robots and Systems, 2012. ",
+        "bbox": [
+            173,
+            806,
+            820,
+            837
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Yifan Wu, George Tucker, and Ofir Nachum. Behavior regularized offline reinforcement learning. arXiv preprint arXiv:1911.11361, 2019. ",
+        "bbox": [
+            174,
+            843,
+            821,
+            873
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Tianhe Yu, Garrett Thomas, Lantao Yu, Stefano Ermon, James Zou, Sergey Levine, Chelsea Finn, and Tengyu Ma. MOPO: Model-based offline policy optimization. In Advances in Neural Information Processing Systems, 2020. ",
+        "bbox": [
+            174,
+            881,
+            823,
+            924
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Rich Zemel, Yu Wu, Kevin Swersky, Toni Pitassi, and Cynthia Dwork. Learning fair representations. In International Conference on Machine Learning, 2013. ",
+        "bbox": [
+            171,
+            103,
+            823,
+            132
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "Ruiyi Zhang, Bo Dai, Lihong Li, and Dale Schuurmans. GenDICE: Generalized offline estimation of stationary values. In International Conference on Learning Representations, 2020. ",
+        "bbox": [
+            173,
+            140,
+            823,
+            170
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "A PROOFS ",
+        "text_level": 1,
+        "bbox": [
+            174,
+            102,
+            277,
+            118
+        ],
+        "page_idx": 12
+    },
+    {
+        "type": "text",
+        "text": "Lemma 4.1. Given an MDP $M$ and its estimate $\\widehat { M }$ with a bijective representation function $\\phi ,$ , i.e. $( \\widehat { T } , \\widehat { R } ) = \\langle \\widehat { T } _ { z } \\circ \\phi , \\widehat { R } _ { z } \\circ \\phi \\rangle$ , the policy evaluation error of a policy $\\pi$ can be bounded by: ",
+        "bbox": [
+            173,
+            131,
+            823,
+            166
+        ],
+        "page_idx": 12
+    },
+    {
+        "type": "equation",
+        "img_path": "images/33d91a00582c4e53b97278296f355c5673f244ded589d301c9aa46c446a8f0ee.jpg",
+        "text": "$$\n\\begin{array} { r } { \\left| R ^ { \\pi } - \\widehat { R } ^ { \\pi } \\right| \\leq \\mathbb { E } _ { ( s , a ) \\sim d ^ { \\pi } } \\left[ \\mathcal { E } _ { \\phi , \\widehat { R } _ { z } , \\widehat { T } _ { z } } ( s , a ) \\right] } \\end{array}\n$$",
+        "text_format": "latex",
+        "bbox": [
+            362,
+            172,
+            633,
+            200
+        ],
+        "page_idx": 12
+    },
+    {
+        "type": "text",
+        "text": "Proof. We define a value function $V ^ { \\pi } ( s ) = \\mathbb { E } _ { M , \\pi } [ \\sum _ { t = 0 } ^ { \\infty } \\gamma ^ { t } r _ { t } | s _ { 0 } = s ]$ , the expected discounted return starting from the state $s$ , for the following proof. Due to its definition, we can write $R ^ { \\pi } =$ $( 1 - \\gamma ) \\mathbb { E } _ { s \\sim d _ { 0 } } \\overline { { [ V ^ { \\pi } ( s ) ] } }$ . The following recursive equation also holds: ",
+        "bbox": [
+            173,
+            215,
+            826,
+            261
+        ],
+        "page_idx": 12
+    },
+    {
+        "type": "equation",
+        "img_path": "images/c1ea46570341074db9f0ff962d493395189cbc0ddf5ce0bd5a734eb9b1fa2818.jpg",
+        "text": "$$\nV ^ { \\pi } ( s ) = \\mathbb { E } _ { a \\sim \\pi ( s ) } \\left[ r ( s , a ) + \\gamma \\mathbb { E } _ { s ^ { \\prime } \\sim T ( s , a ) } [ V ( s ^ { \\prime } ) ] \\right] ,\n$$",
+        "text_format": "latex",
+        "bbox": [
+            331,
+            267,
+            665,
+            287
+        ],
+        "page_idx": 12
+    },
+    {
+        "type": "text",
+        "text": "and similarly with an approximate value function $\\widehat { V } ^ { \\pi } ( s ) = \\mathbb { E } _ { \\widehat { M } , \\pi } [ \\sum _ { t = 0 } ^ { \\infty } \\gamma ^ { t } r _ { t } | s _ { 0 } = s ]$ given an MDP estimate $\\widehat { M }$ . Then, ",
+        "bbox": [
+            173,
+            296,
+            823,
+            333
+        ],
+        "page_idx": 12
+    },
+    {
+        "type": "equation",
+        "img_path": "images/b2eca14fd51e0827850edb012e3393c12214cd4f5f440e18f3f30a62554ad408.jpg",
+        "text": "$$\n\\begin{array} { r l } & { \\displaystyle \\frac { 1 } { 1 - \\gamma } \\Big | R ^ { \\pi } - \\widehat { R } ^ { \\pi } \\Big | = \\left| \\mathbb { E } _ { d _ { 0 } } \\left[ V ^ { \\pi } ( s _ { 0 } ) - \\widehat { V } ^ { \\pi } ( s _ { 0 } ) \\right] \\right| } \\\\ & { = \\left| \\mathbb { E } _ { d _ { 0 } , \\pi } \\left[ r _ { 0 } - \\widehat { r } _ { 0 } \\right] + \\gamma \\mathbb { E } _ { d _ { 0 } } \\left[ \\int \\left( V ^ { \\pi } ( s _ { 1 } ) T ( s _ { 1 } | s _ { 0 } , a _ { 0 } ) - \\widehat { V } ^ { \\pi } ( s _ { 1 } ) \\widehat { T } ( s _ { 1 } | s _ { 0 } , a _ { 0 } ) \\right) \\pi ( a _ { 0 } | s _ { 0 } ) d s _ { 1 } d a _ { 0 } \\right] \\right| } \\\\ & { = \\left| \\mathbb { E } _ { d _ { 0 } , \\pi } \\left[ r _ { 0 } - \\widehat { r } _ { 0 } \\right] + \\gamma \\mathbb { E } _ { d _ { 0 } } \\left[ \\int \\left( V ^ { \\pi } ( s _ { 1 } ) T ( s _ { 1 } | s _ { 0 } , a _ { 0 } ) - \\widehat { V } ^ { \\pi } ( s _ { 1 } ) T ( s _ { 1 } | s _ { 0 } , a _ { 0 } ) \\right. \\right. } \\\\ & { \\qquad \\left. \\left. + \\widehat { V } ^ { \\pi } ( s _ { 1 } ) T ( s _ { 1 } | s _ { 0 } , a _ { 0 } ) - \\widehat { V } ^ { \\pi } ( s _ { 1 } ) \\widehat { T } ( s _ { 1 } | s _ { 0 } , a _ { 0 } ) \\right) \\pi ( a _ { 0 } | s _ { 0 } ) d s _ { 1 } d a _ { 0 } \\right] \\right| } \\\\ & { = \\left| \\mathbb { E } _ { d _ { 0 } , \\pi } \\left[ r _ { 0 } - \\widehat { r } _ { 0 } \\right] + \\gamma \\mathbb { E } _ { d _ { 0 } } \\left[ \\int \\left( V ^ { \\pi } ( s _ { 1 } ) - \\widehat { V } ^ { \\pi } ( s _ { 1 } ) \\right) T ( s _ { 1 } | s _ { 0 } , a _ { 0 } ) \\pi ( a _ { 0 } | s _ { 0 } ) d s _ { 1 } d a _ { 0 } \\right] \\right| } \\end{array}\n$$",
+        "text_format": "latex",
+        "bbox": [
+            181,
+            339,
+            820,
+            517
+        ],
+        "page_idx": 12
+    },
+    {
+        "type": "equation",
+        "img_path": "images/4f76507e4d0bde5af4efb2747350527ebfb713650f52d5c558cf11544b51ad73.jpg",
+        "text": "$$\n\\begin{array} { r l } & { \\qquad + \\gamma \\mathbb { E } _ { d _ { 0 } } \\bigg [ \\int \\hat { V } ^ { x } ( s _ { 1 } ) \\bigg ( T ( s _ { 1 } | s _ { 0 } , a _ { 0 } ) - \\hat { T } ( s _ { 1 } | s _ { 0 } , a _ { 0 } ) \\bigg ) \\pi ( a _ { 0 } | s _ { 0 } ) d s _ { 1 } d a _ { 0 } \\bigg ] \\bigg | } \\\\ & { = \\left| \\mathbb { E } _ { ( s , a ) \\times d ^ { c } } \\left[ \\int \\bigg ( R ( r | s , a ) - \\hat { R } ( r | s , a ) \\bigg ) d r + \\gamma \\int \\hat { V } ^ { x } ( s ^ { \\prime } ) \\bigg ( T ( s ^ { \\prime } | s , a ) - \\hat { T } ( s ^ { \\prime } | s , a ) \\bigg ) d s ^ { \\prime } \\right] \\right| } \\\\ & { \\leq \\mathbb { E } _ { ( s , a ) \\sim d ^ { c } } \\left[ \\int \\bigg | R ( r | s , a ) - \\hat { R } ( r | s , a ) \\bigg | d r + \\gamma \\int \\hat { V } ^ { \\tau } ( s ^ { \\prime } ) \\bigg | T ( s ^ { \\prime } | s , a ) - \\hat { T } ( s ^ { \\prime } | s , a ) \\bigg | d s ^ { \\prime } \\right] } \\\\ & { \\leq \\mathbb { E } _ { ( s , a ) \\sim d ^ { c } } \\left[ \\int \\bigg | R ( r | s , a ) - \\hat { R } ( r | s , a ) \\bigg | d r + \\gamma \\int \\frac { r _ { \\operatorname* { m a x } } } { 1 - \\gamma } \\bigg | T ( s ^ { \\prime } | s , a ) - \\hat { T } ( s ^ { \\prime } | s , a ) \\bigg | d s ^ { \\prime } \\right] } \\\\ & { = 2 \\mathbb { E } _ { ( s , a ) \\sim d ^ { c } } \\left[ D _ { T V } \\big ( R ( r | s , a ) \\big | \\hat { R } _ { s } ( r | \\phi ( s , a ) ) \\big ) \\right] } \\\\ & { \\qquad + \\frac { 2 \\gamma r _ { \\operatorname* { m a x } } } { 1 - \\gamma } \\mathbb { E } _ { ( s , a ) \\sim d ^ { c } } \\left[ D _ { T V } \\big ( T ( s ^ { \\prime } | s , a ) \\big | \\hat { T } _ { 2 } ( s ^ { \\prime } | \\phi ( s , a ) ) \\big ) \\right] } \\end{array}\n$$",
+        "text_format": "latex",
+        "bbox": [
+            181,
+            534,
+            784,
+            736
+        ],
+        "page_idx": 12
+    },
+    {
+        "type": "text",
+        "text": "Theorem 4.2. Given an MDP $M$ and its estimate $\\widehat { M }$ with a bijective representation function $\\phi$ , i.e. $( \\widehat { T } , \\widehat { R } ) = \\langle \\widehat { T } _ { z } \\circ \\phi , \\widehat { R } _ { z } \\circ \\phi \\rangle$ , assume that there exists a constant $B _ { \\phi } > 0$ and a function class ${ \\mathcal { G } } \\triangleq \\{ g : { \\mathcal { Z } } \\to \\mathbb { R } \\}$ b such that $\\begin{array} { r } { \\frac { 1 } { B _ { \\phi } } \\mathcal { E } _ { \\phi , \\widehat { R } _ { z } , \\widehat { T } _ { z } } \\left( \\phi ^ { - 1 } ( \\cdot ) \\right) \\in \\mathcal { G } . } \\end{array}$ . Then, for any policy $\\pi$ , ",
+        "bbox": [
+            173,
+            762,
+            825,
+            815
+        ],
+        "page_idx": 12
+    },
+    {
+        "type": "equation",
+        "img_path": "images/20f445a1a1724a9bf59376c67dcf590916f33dfe41f2db895ef78f051bceb746.jpg",
+        "text": "$$\n\\left| R ^ { \\pi } - \\widehat { R } ^ { \\pi } \\right| \\leq \\mathbb { E } _ { ( s , a ) \\sim d ^ { \\mathcal { D } } } \\Big [ \\mathcal { E } _ { \\phi , \\widehat { R } _ { z } , \\widehat { T } _ { z } } ( s , a ) \\Big ] + B _ { \\phi } \\mathrm { I P M } _ { \\mathcal { G } } ( d _ { \\phi } ^ { \\pi } , d _ { \\phi } ^ { \\mathcal { D } } )\n$$",
+        "text_format": "latex",
+        "bbox": [
+            290,
+            824,
+            705,
+            852
+        ],
+        "page_idx": 12
+    },
+    {
+        "type": "text",
+        "text": "Proof. From Lemma 4.1, we directly have: ",
+        "bbox": [
+            174,
+            867,
+            459,
+            883
+        ],
+        "page_idx": 12
+    },
+    {
+        "type": "equation",
+        "img_path": "images/c3454a9ebef3285947433d845c8fbed141601c968a8c8c8c532f53ffce967a75.jpg",
+        "text": "$$\n\\Big | R ^ { \\pi } - \\widehat { R } ^ { \\pi } \\Big | \\leq \\mathbb { E } _ { ( s , a ) \\sim d ^ { D } } \\Big [ \\mathcal { E } _ { \\phi , \\widehat { R } _ { z } , \\widehat { T } _ { z } } ( s , a ) \\Big ] + \\int \\mathcal { E } _ { \\phi , \\widehat { R } _ { z } , \\widehat { T } _ { z } } ( s , a ) \\left( d ^ { \\pi } ( s , a ) - d ^ { D } ( s , a ) \\right) d s d a .\n$$",
+        "text_format": "latex",
+        "bbox": [
+            202,
+            890,
+            794,
+            922
+        ],
+        "page_idx": 12
+    },
+    {
+        "type": "text",
+        "text": "Then, ",
+        "bbox": [
+            173,
+            103,
+            214,
+            117
+        ],
+        "page_idx": 13
+    },
+    {
+        "type": "equation",
+        "img_path": "images/a720027c9c93c42352414970b7e78611a3535be8979d5601c76031d1aa1ace27.jpg",
+        "text": "$$\n\\begin{array} { r l } & { \\displaystyle \\int \\mathcal { E } _ { \\phi , \\widehat { R } _ { + } , \\widehat { T } _ { z } } ( s , a ) \\left( d ^ { \\pi } ( s , a ) - d ^ { D } ( s , a ) \\right) d s d a } \\\\ & { \\quad = \\displaystyle \\int \\mathcal { E } _ { \\phi , \\widehat { R } _ { \\epsilon } , \\widehat { T } _ { z } } ( \\phi ^ { - 1 } ( z ) ) \\left( d ^ { \\pi } ( z ) - d ^ { D } \\left( \\phi ^ { - 1 } ( \\phi ^ { - 1 } ( z ) ) \\right) \\right) \\left. \\frac { d ( s , a ) } { d z } \\right. d z } \\\\ & { \\quad = \\displaystyle \\int \\mathcal { E } _ { \\phi , \\widehat { R } _ { \\epsilon } , \\widehat { T } _ { z } } ( \\phi ^ { - 1 } ( z ) ) \\left( d _ { \\phi } ^ { \\pi } ( z ) - d _ { \\phi } ^ { D } ( z ) \\right) d z } \\\\ & { \\quad \\leq { B } _ { \\phi } \\displaystyle \\left. \\int \\frac { 1 } { B _ { \\phi } } \\mathcal { E } _ { \\phi , \\widehat { R } _ { \\epsilon } , \\widehat { T } _ { z } } ( \\phi ^ { - 1 } ( z ) ) \\left( d _ { \\phi } ^ { \\pi } ( z ) - d _ { \\phi } ^ { D } ( z ) \\right) d z \\right. \\qquad \\left( { B } _ { \\phi } > 0 \\right) } \\\\ & { \\quad \\leq { B } _ { \\phi \\in \\nabla } \\displaystyle \\left. \\int \\mathcal { J } ( { \\boldsymbol \\jmath } ^ { \\pi } ) \\left( d _ { \\phi } ^ { \\pi } ( z ) - d _ { \\phi } ^ { D } ( z ) \\right) d z \\right. \\qquad \\left( \\frac { 1 } { B _ { \\phi } } \\mathcal { E } _ { \\phi , \\widehat { R } _ { \\epsilon } , \\widehat { T } _ { z } } ( \\phi ^ { - 1 } ( z ) ) \\in \\mathcal { G } \\right) } \\\\ & { \\quad = { B } _ { \\phi } \\mathrm { F N M } _ { \\phi } \\left( a _ { \\phi } ^ { \\pi } , d _ { \\phi } ^ { \\theta } \\right) } \\end{array}\n$$",
+        "text_format": "latex",
+        "bbox": [
+            232,
+            119,
+            764,
+            316
+        ],
+        "page_idx": 13
+    },
+    {
+        "type": "text",
+        "text": "We state some previous results, which are required for further proof. ",
+        "bbox": [
+            174,
+            343,
+            625,
+            359
+        ],
+        "page_idx": 13
+    },
+    {
+        "type": "text",
+        "text": "Theorem A.1 (McDiarmid’s inequality (McDiarmid, 1989)). Let $\\{ X _ { i } \\} _ { i = 1 } ^ { n }$ be independent random variables taking values in set $\\mathcal { X }$ , and assume that $f : \\mathcal { X } ^ { n }  \\mathbb { R }$ satisfies ",
+        "bbox": [
+            176,
+            361,
+            825,
+            390
+        ],
+        "page_idx": 13
+    },
+    {
+        "type": "equation",
+        "img_path": "images/5322ea8d75e48088fee9c308cdb79057d3b98b8e0ba25e9b81ea214f2d4e441e.jpg",
+        "text": "$$\n\\operatorname* { s u p } _ { \\{ x _ { i } \\} _ { i = 1 } ^ { n } \\in \\mathcal { X } ^ { n } , \\tilde { x } \\in \\mathcal { X } } | f ( \\{ x _ { i } \\} _ { i = 1 } ^ { n } ) - f ( x _ { 1 } , . . . , x _ { i - 1 } , \\tilde { x } , x _ { i + 1 } , . . . , x _ { n } ) | \\leq c _ { i } .\n$$",
+        "text_format": "latex",
+        "bbox": [
+            272,
+            392,
+            725,
+            420
+        ],
+        "page_idx": 13
+    },
+    {
+        "type": "text",
+        "text": "Then for every $\\epsilon > 0$ , ",
+        "bbox": [
+            174,
+            422,
+            315,
+            438
+        ],
+        "page_idx": 13
+    },
+    {
+        "type": "equation",
+        "img_path": "images/a019a80950c3511cf145cf95eb3e05d6994a05270feb86d4ea47853f31ee3a30.jpg",
+        "text": "$$\n{ \\operatorname* { P r } } \\{ f ( \\{ X _ { i } \\} _ { i = 1 } ^ { n } ) - \\mathbb { E } _ { \\{ X _ { i } \\} _ { i = 1 } ^ { n } } [ f ( \\{ X _ { i } \\} _ { i = 1 } ^ { n } ) ] \\geq \\epsilon \\} \\leq \\exp \\left( - \\frac { 2 \\epsilon ^ { 2 } } { \\sum _ { i = 1 } ^ { n } c _ { i } ^ { 2 } } \\right) .\n$$",
+        "text_format": "latex",
+        "bbox": [
+            266,
+            440,
+            735,
+            477
+        ],
+        "page_idx": 13
+    },
+    {
+        "type": "text",
+        "text": "Lemma A.2 (Rademacher complexity of RKHS (Bartlett & Mendelson, 2002)). Let $\\mathcal { F }$ be a unit ball in a universal RKHS on the compact domain $\\mathcal { X }$ , with kernel bounded according to $0 \\leq k ( x , x ^ { \\prime } ) \\leq \\bar { k }$ . Let $\\{ x _ { i } \\} _ { i = 1 } ^ { n }$ be an i.i.d. sample of size n drawn according to a probability measure $p$ on $\\mathcal { X }$ , and let $\\sigma _ { i }$ be i.i.d. and take values in $\\{ - 1 , 1 \\}$ with equal probability. The Rademacher complexity, which is defined as below, is upper bounded as: ",
+        "bbox": [
+            173,
+            484,
+            826,
+            556
+        ],
+        "page_idx": 13
+    },
+    {
+        "type": "equation",
+        "img_path": "images/49bf738bde67b3d72bbf1989e2bb9b1edbd9d5aef690648e56bf852f6d38f497.jpg",
+        "text": "$$\nR _ { n } ( \\mathcal { F } ) \\triangleq \\mathbb { E } _ { \\{ x _ { i } \\} _ { i = 1 } ^ { n } , \\sigma } \\operatorname* { s u p } _ { f \\in \\mathcal { F } } \\left| \\frac { 1 } { n } \\sum _ { i = 1 } ^ { n } \\sigma _ { i } f ( x _ { i } ) \\right| \\leq \\sqrt { \\frac { \\bar { k } } { n } } .\n$$",
+        "text_format": "latex",
+        "bbox": [
+            331,
+            558,
+            666,
+            601
+        ],
+        "page_idx": 13
+    },
+    {
+        "type": "text",
+        "text": "The upper bound is followed by Lemma 22 of (Bartlett & Mendelson, 2002). ",
+        "bbox": [
+            176,
+            602,
+            671,
+            617
+        ],
+        "page_idx": 13
+    },
+    {
+        "type": "text",
+        "text": "Now we prove the following using the results above. ",
+        "bbox": [
+            173,
+            626,
+            517,
+            642
+        ],
+        "page_idx": 13
+    },
+    {
+        "type": "text",
+        "text": "Lemma A.3. Let $\\mathcal { H } _ { k }$ be a RKHS associated with universal kernel $k ( \\cdot , \\cdot )$ . Let $\\langle \\cdot , \\cdot \\rangle _ { \\mathcal { H } _ { k } }$ be the inner product of $\\mathcal { H } _ { k }$ , which satisfies the reproducing property $\\nu ( z ) = \\langle \\nu , \\dot { k } ( \\cdot , z ) \\rangle _ { \\mathcal { H } _ { k } }$ . When $\\mathcal { G }$ is chosen such that ",
+        "bbox": [
+            173,
+            643,
+            825,
+            685
+        ],
+        "page_idx": 13
+    },
+    {
+        "type": "equation",
+        "img_path": "images/418aa03f21e5d2b92df1f9e04e68d4c5d9ee40bc6d1f112a6a369ea021ca8ca0.jpg",
+        "text": "$$\n\\begin{array} { r } { \\boldsymbol { \\operatorname { \\mathrm { ? } } } = \\biggl \\{ g \\in ( \\mathcal { Z } \\to \\mathbb { R } ) : g ( z ) = \\nu ( z ) - \\gamma \\mathbb { E } _ { s ^ { \\prime } \\sim T ( \\phi ^ { - 1 } ( z ) ) } [ \\nu ( \\phi ( s ^ { \\prime } , a ^ { \\prime } ) ) ] , \\nu \\in ( \\mathcal { Z } \\to \\mathbb { R } ) , \\langle \\nu , \\nu \\rangle _ { \\mathcal { H } _ { k } } \\leq 1 \\biggr \\} , } \\end{array}\n$$",
+        "text_format": "latex",
+        "bbox": [
+            181,
+            684,
+            823,
+            726
+        ],
+        "page_idx": 13
+    },
+    {
+        "type": "text",
+        "text": "the $\\mathrm { I P M } _ { \\mathcal { G } } ( d _ { \\phi } ^ { \\pi } , d _ { \\phi } ^ { D } )$ has the following closed form definition: ",
+        "bbox": [
+            173,
+            728,
+            566,
+            744
+        ],
+        "page_idx": 13
+    },
+    {
+        "type": "equation",
+        "img_path": "images/3444fb4a6f7a51d186fdee2a116daacd8024fae5a586d50294f891655a6837a3.jpg",
+        "text": "$$\n\\begin{array} { r l } & { \\mathrm { [ P M } _ { \\mathcal { G } } ( d _ { \\phi } ^ { \\pi } , d _ { \\phi } ^ { \\mathcal { D } } ) ^ { 2 } = \\mathbb { E } _ { s _ { 0 } \\sim d _ { 0 } , a _ { 0 } \\sim \\pi ( s _ { 0 } ) , ( s , a , s ^ { \\prime } ) \\sim d ^ { \\mathcal { D } } , a ^ { \\prime } \\sim \\pi ( s ^ { \\prime } ) } } \\\\ & { \\quad \\quad k ( \\phi ( s , a ) , \\phi ( \\bar { s } , \\bar { a } ) ) + ( 1 - \\gamma ) ^ { 2 } k ( \\phi ( s _ { 0 } , a _ { 0 } ) , \\phi ( \\bar { s } _ { 0 } , \\bar { a } _ { 0 } ) ) + \\gamma ^ { 2 } k ( \\phi ( s ^ { \\prime } , a ^ { \\prime } ) , \\phi ( \\bar { s } ^ { \\prime } , \\bar { a } ^ { \\prime } ) ) } \\\\ & { \\quad - 2 ( 1 - \\gamma ) k ( \\phi ( s _ { 0 } , a _ { 0 } ) , \\phi ( \\bar { s } , \\bar { a } ) ) - 2 \\gamma k ( \\phi ( s ^ { \\prime } , a ^ { \\prime } ) , \\phi ( \\bar { s } , \\bar { a } ) ) + 2 \\gamma ( 1 - \\gamma ) k ( \\phi ( s ^ { \\prime } , a ^ { \\prime } ) , \\phi ( \\bar { s } _ { 0 } , \\bar { a } _ { 0 } ) ) \\bigg ] . } \\end{array}\n$$",
+        "text_format": "latex",
+        "bbox": [
+            181,
+            748,
+            825,
+            842
+        ],
+        "page_idx": 13
+    },
+    {
+        "type": "text",
+        "text": "Furthermore, suppose that $\\bar { k } \\triangleq \\operatorname* { s u p } _ { z \\in \\mathcal { Z } } k ( z , z )$ . The estimator $\\widehat { \\mathrm { I P M } } ( d _ { \\phi } ^ { \\pi } , d _ { \\phi } ^ { D } ) ^ { 2 }$ , which is the samplebased estimation of $\\mathrm { I P M } _ { \\mathcal { G } } ( d _ { \\phi } ^ { \\pi } , d _ { \\phi } ^ { \\mathcal { D } } ) ^ { 2 }$ from $n$ samples, satisfies with probability at least $1 - \\delta$ , ",
+        "bbox": [
+            173,
+            847,
+            823,
+            883
+        ],
+        "page_idx": 13
+    },
+    {
+        "type": "equation",
+        "img_path": "images/851944251370897dcaa5c34d1a2a719012871dff65c5fb85cc5dc91831383dcd.jpg",
+        "text": "$$\n\\left| \\mathrm { I P M } _ { \\mathcal { G } } ( d _ { \\phi } ^ { \\pi } , d _ { \\phi } ^ { D } ) - \\widehat { \\mathrm { I P M } } ( d _ { \\phi } ^ { \\pi } , d _ { \\phi } ^ { D } ) \\right| \\leq \\sqrt { \\frac { \\bar { k } } { n } } \\left( 4 + \\sqrt { 8 \\log \\frac { 3 } { \\delta } } \\right) .\n$$",
+        "text_format": "latex",
+        "bbox": [
+            299,
+            887,
+            697,
+            928
+        ],
+        "page_idx": 13
+    },
+    {
+        "type": "text",
+        "text": "Proof. In the below we write in shorthand, $\\operatorname { I P M } _ { \\mathcal { G } }$ to denote $\\mathrm { I P M } _ { \\mathcal { G } } ( d _ { \\phi } ^ { \\pi } , d _ { \\phi } ^ { D } )$ . As in Eq. (4), we can rewrite the IPM term as: ",
+        "bbox": [
+            173,
+            102,
+            825,
+            132
+        ],
+        "page_idx": 14
+    },
+    {
+        "type": "equation",
+        "img_path": "images/1e9af79db67c46d6c1abe815776e0ef1c6841c3eeeba68a8b65a28072a4f3e9e.jpg",
+        "text": "$$\n\\mathrm { I P M } _ { \\mathcal { G } } = \\underset { \\nu \\in \\mathcal { F } } { \\operatorname* { s u p } } \\left| ( 1 - \\gamma ) \\mathbb { E } _ { \\underset { a \\sim \\pi ( s ) } { s \\sim d _ { 0 } } } \\left[ \\nu ( \\phi ( s , a ) ) \\right] - \\mathbb { E } _ { ( s , a , s ^ { \\prime } ) \\sim d ^ { \\mathcal { D } } } \\left[ \\nu ( \\phi ( s , a ) ) - \\gamma \\nu ( \\phi ( s ^ { \\prime } , a ^ { \\prime } ) ) \\right] \\right| ,\n$$",
+        "text_format": "latex",
+        "bbox": [
+            209,
+            132,
+            787,
+            175
+        ],
+        "page_idx": 14
+    },
+    {
+        "type": "text",
+        "text": "and $\\mathcal { F }$ here becomes $\\mathcal { F } = \\{ \\nu \\in ( \\mathcal { Z }  \\mathbb { R } ) : \\langle \\nu , \\nu \\rangle _ { \\mathcal { H } _ { k } } \\leq 1 \\}$ , a unit ball in RKHS $\\mathcal { H } _ { k }$ . Using the reproducing property of $\\mathcal { H } _ { k }$ : ",
+        "bbox": [
+            173,
+            176,
+            823,
+            207
+        ],
+        "page_idx": 14
+    },
+    {
+        "type": "equation",
+        "img_path": "images/bba865bd9db1881f88626ac635d3f86efbab27a10ed37efebb2b5f94e92bc50a.jpg",
+        "text": "$$\n\\begin{array} { r l } & { \\mathrm { I P M } _ { \\mathcal { G } } ^ { 2 } = \\Bigg \\{ \\underset { s \\in \\mathcal { F } } { \\operatorname* { s u p } } \\Bigg \\{ ( 1 - \\gamma ) \\mathbb { E } _ { \\underset { \\sigma \\sim \\infty } { s \\operatorname* { s d } } } \\left[ \\nu ( \\phi s , a ) \\right] \\big ] - \\mathbb { E } _ { ( s , a , s ^ { \\prime } ) \\sim d } \\mathbb { P } \\big [ \\nu ( \\phi ( s , a ) ) - \\gamma \\nu ( \\phi ( s ^ { \\prime } , a ^ { \\prime } ) ) \\big ] \\Bigg \\} ^ { 2 } } \\\\ & { \\qquad = \\underset { s \\in \\mathcal { F } } { \\operatorname* { s u p } } \\Bigg \\{ ( 1 - \\gamma ) \\mathbb { E } _ { \\underset { \\sigma \\sim \\infty } { s \\operatorname* { s u p } } } \\big [ \\nu ( \\phi ( s , a ) ) \\big ] - \\mathbb { E } _ { ( s , a , s ^ { \\prime } ) \\sim d } \\mathbb { P } \\big [ \\nu ( \\phi ( s , a ) ) - \\gamma \\nu ( \\phi ( s ^ { \\prime } , a ^ { \\prime } ) ) \\big ] \\Bigg \\} ^ { 2 } } \\\\ & { \\qquad = \\underset { s \\in \\mathcal { F } } { \\operatorname* { s u p } } \\Bigg \\{ ( 1 - \\gamma ) \\mathbb { E } _ { \\underset { \\sigma \\sim \\infty } { s \\operatorname* { s u p } } } \\big [ ( \\nu , k ( \\cdot , \\phi ( s , a ) ) ) u _ { k } \\big ] } \\\\ & { \\qquad \\quad - \\mathbb { E } _ { ( s , a , s ^ { \\prime } ) \\sim d } \\mathbb { P } \\big [ \\nu , k ( \\cdot , \\cdot , \\phi ( s , a ) ) \\big ] u _ { k } - \\gamma \\big ( \\nu , k ( \\cdot , \\phi ( s ^ { \\prime } , a ^ { \\prime } ) ) \\big ) _ { k k } \\bigg ] \\Bigg \\} ^ { 2 } } \\\\ & { \\qquad - \\mathbb { E } _ { ( s , a , s ^ { \\prime } ) \\sim d } \\rho ^ { D } \\left[ \\big \\langle \\nu , k ( \\cdot , \\phi ( s , a ) ) \\big \\rangle _ { \\mathcal { H } _ { k } } - \\gamma \\big \\langle \\nu , k ( \\cdot , \\phi ( s ^ { \\prime } , a ^ { \\prime } ) ) \\big \\rangle _ { \\mathcal { H } _ { k } } \\right] \\Bigg \\} ^ { 2 } } \\\\ & { \\qquad = \\underset { s \\in \\mathcal { F } } { \\operatorname* { s u p } } \\big \\{ \\nu , \\nu ^ { * } \\} _ { \\mathcal { H } _ { k } } ^ { 2 } , } \\end{array}\n$$",
+        "text_format": "latex",
+        "bbox": [
+            192,
+            209,
+            802,
+            409
+        ],
+        "page_idx": 14
+    },
+    {
+        "type": "text",
+        "text": "where ",
+        "bbox": [
+            174,
+            409,
+            217,
+            421
+        ],
+        "page_idx": 14
+    },
+    {
+        "type": "equation",
+        "img_path": "images/acf18b5e3942dbcc477617e3bfbb93d7359240124fce41eeb9f3e3bc0fad7800.jpg",
+        "text": "$$\n\\begin{array} { r } { \\nu ^ { * } ( \\cdot ) = ( 1 - \\gamma ) \\mathbb { E } _ { \\stackrel { s \\sim d _ { 0 } } { a \\sim \\pi ( s ) } } \\left[ k ( \\cdot , \\phi ( s , a ) ) \\right] - \\mathbb { E } _ { ( s , a , s ^ { \\prime } ) \\sim d ^ { \\mathcal { D } } } \\left[ k ( \\cdot , \\phi ( s , a ) ) - \\gamma k ( \\cdot , \\phi ( s ^ { \\prime } , a ^ { \\prime } ) ) \\right] . } \\end{array}\n$$",
+        "text_format": "latex",
+        "bbox": [
+            214,
+            424,
+            781,
+            454
+        ],
+        "page_idx": 14
+    },
+    {
+        "type": "text",
+        "text": "Due to the Cauchy-Schwarz inequality and $\\langle \\nu , \\nu \\rangle _ { \\mathcal { H } _ { k } } \\leq 1 , \\forall \\nu \\in \\mathcal { F }$ , ",
+        "bbox": [
+            173,
+            457,
+            619,
+            472
+        ],
+        "page_idx": 14
+    },
+    {
+        "type": "equation",
+        "img_path": "images/4bbe514d3b1e199bd94f527f270ca24f68cb7324d0641a8347f32aab732c9bc9.jpg",
+        "text": "$$\n\\langle \\nu , \\nu ^ { * } \\rangle _ { \\mathcal { H } _ { k } } ^ { 2 } \\le \\langle \\nu , \\nu \\rangle _ { \\mathcal { H } _ { k } } \\langle \\nu ^ { * } , \\nu ^ { * } \\rangle _ { \\mathcal { H } _ { k } } \\le \\langle \\nu ^ { * } , \\nu ^ { * } \\rangle _ { \\mathcal { H } _ { k } }\n$$",
+        "text_format": "latex",
+        "bbox": [
+            343,
+            473,
+            651,
+            493
+        ],
+        "page_idx": 14
+    },
+    {
+        "type": "text",
+        "text": "holds and $\\begin{array} { r } { \\operatorname* { s u p } _ { \\nu \\in \\mathcal { F } } \\langle \\nu , \\nu ^ { * } \\rangle _ { \\mathcal { H } _ { k } } ^ { 2 } = \\langle \\nu ^ { * } , \\nu ^ { * } \\rangle _ { \\mathcal { H } _ { k } } } \\end{array}$ . Using the property that $\\langle k ( \\cdot , z ) , k ( \\cdot , \\bar { z } ) \\rangle _ { \\mathcal { H } _ { k } } = k ( z , \\bar { z } )$ we can derive the closed form expression in the lemma from $\\langle \\nu ^ { * } , \\nu ^ { * } \\rangle _ { \\mathcal { H } _ { k } }$ . ",
+        "bbox": [
+            176,
+            496,
+            818,
+            526
+        ],
+        "page_idx": 14
+    },
+    {
+        "type": "text",
+        "text": "Now we prove the error bound of the estimator. First, we divide $\\mathrm { I P M } _ { \\mathcal { G } } ( d _ { \\phi } ^ { \\pi } , d _ { \\phi } ^ { D } )$ into three parts: ",
+        "bbox": [
+            173,
+            531,
+            799,
+            549
+        ],
+        "page_idx": 14
+    },
+    {
+        "type": "equation",
+        "img_path": "images/3e4f9a1c0fd26489b8afe531f101ab6a89e2c21c7642455f36285f71d18e8f81.jpg",
+        "text": "$$\n\\begin{array} { r l } & { \\mathrm { I P M } _ { \\mathcal { G } } ( d _ { \\phi } ^ { \\pi } , d _ { \\phi } ^ { D } ) = \\underset { \\nu \\in \\mathcal { F } } { \\operatorname* { s u p } } | f _ { 1 } ( \\nu ) + f _ { 2 } ( \\nu ) + f _ { 3 } ( \\nu ) | } \\\\ & { f _ { 1 } ( \\nu ) = ( 1 - \\gamma ) \\mathbb { E } _ { s \\sim d _ { 0 } , a \\sim \\pi ( s ) } [ \\nu ( \\phi ( s , a ) ) ] , \\qquad f _ { 2 } ( \\nu ) = - \\mathbb { E } _ { ( s , a ) \\sim d ^ { D } } \\left[ \\nu ( \\phi ( s , a ) ) \\right] , } \\\\ & { f _ { 3 } ( \\nu ) = \\mathbb { E } _ { ( s , a , s ^ { \\prime } ) \\sim d ^ { D } , a \\cdot \\sim \\pi ( s ^ { \\prime } ) } \\left[ \\gamma \\nu ( \\phi ( s ^ { \\prime } , a ^ { \\prime } ) ) \\right] . } \\end{array}\n$$",
+        "text_format": "latex",
+        "bbox": [
+            254,
+            551,
+            799,
+            619
+        ],
+        "page_idx": 14
+    },
+    {
+        "type": "text",
+        "text": "Given $n$ samples $\\{ s _ { 0 } ^ { ( i ) } , a _ { 0 } ^ { ( i ) } , s ^ { ( i ) } , a ^ { ( i ) } , s ^ { \\prime ( i ) } , a ^ { \\prime ( i ) } \\} _ { i = 1 } ^ { n }$ from the generative process $s _ { 0 } ^ { ( i ) } \\sim d _ { 0 } , a _ { 0 } ^ { ( i ) } \\sim$ $\\pi ( s _ { 0 } ^ { ( i ) } ) , ( s ^ { ( i ) } , a ^ { ( i ) } , s ^ { \\prime ( i ) } ) ~ \\sim ~ d ^ { D } , a ^ { \\prime ( i ) } ~ \\sim ~ \\pi ( s ^ { \\prime ( i ) } ) ~ \\forall i ,$ , we define the sample-based estimator $\\widehat { \\mathrm { I P M } } ( d _ { \\phi } ^ { \\pi } , d _ { \\phi } ^ { D } )$ : ",
+        "bbox": [
+            173,
+            621,
+            825,
+            676
+        ],
+        "page_idx": 14
+    },
+    {
+        "type": "equation",
+        "img_path": "images/c8bc39946b4a96a2690e87ba300f82a151470d102975b6f07871841d792ad7d0.jpg",
+        "text": "$$\n\\begin{array} { l } { { \\displaystyle \\widehat { \\mathrm { I P M } } ( d _ { \\phi } ^ { \\pi } , d _ { \\phi } ^ { \\mathcal { D } } ) ^ { 2 } = \\frac { 1 } { n ^ { 2 } } \\sum _ { i , j } \\bigg [ k ( \\phi ( s ^ { ( i ) } , a ^ { ( i ) } ) , \\phi ( s ^ { ( j ) } , a ^ { ( j ) } ) ) + ( 1 - \\gamma ) ^ { 2 } k ( \\phi ( s _ { 0 } ^ { ( i ) } , a _ { 0 } ^ { ( i ) } ) , \\phi ( s _ { 0 } ^ { ( j ) } , a _ { 0 } ^ { ( j ) } ) ) } } \\\\ { { \\displaystyle \\qquad + \\gamma ^ { 2 } k ( \\phi ( s ^ { \\prime ( i ) } , a ^ { \\prime ( i ) } ) , \\phi ( s ^ { \\prime ( j ) } , a ^ { \\prime ( j ) } ) ) - 2 ( 1 - \\gamma ) k ( \\phi ( s _ { 0 } ^ { ( i ) } , a _ { 0 } ^ { ( i ) } ) , \\phi ( s ^ { ( j ) } , a ^ { ( j ) } ) ) } } \\\\ { { \\displaystyle \\qquad - 2 \\gamma k ( \\phi ( s ^ { \\prime ( i ) } , a ^ { \\prime ( i ) } ) , \\phi ( s ^ { ( j ) } , a ^ { ( j ) } ) ) + 2 \\gamma ( 1 - \\gamma ) k ( \\phi ( s ^ { \\prime ( i ) } , a ^ { \\prime ( i ) } ) , \\phi ( s _ { 0 } ^ { ( j ) } , a _ { 0 } ^ { ( j ) } ) ) \\bigg ] . } } \\end{array}\n$$",
+        "text_format": "latex",
+        "bbox": [
+            183,
+            678,
+            812,
+            777
+        ],
+        "page_idx": 14
+    },
+    {
+        "type": "text",
+        "text": "By deriving in reverse order, we can recover its another definition in supremum, which can be divided into three parts: ",
+        "bbox": [
+            173,
+            776,
+            825,
+            805
+        ],
+        "page_idx": 14
+    },
+    {
+        "type": "equation",
+        "img_path": "images/48a4c81c6e8b78cd0a30ed57cc221fcd7dfa0fef593289ae20126a390d8e9431.jpg",
+        "text": "$$\n\\begin{array} { r l } & { \\widehat { \\mathrm { I P M } } ( d _ { \\phi } ^ { \\pi } , d _ { \\phi } ^ { 2 D } ) = \\displaystyle \\operatorname* { s u p } _ { \\nu \\in \\mathcal { F } } \\Big | \\widehat { f } _ { 1 } ( \\nu ) + \\widehat { f } _ { 2 } ( \\nu ) + \\widehat { f } _ { 3 } ( \\nu ) \\Big | } \\\\ & { \\mathrm { w h e r e } \\widehat { f } _ { 1 } ( \\nu ) = \\displaystyle \\frac { 1 - \\gamma } { n } \\sum _ { i = 1 } ^ { n } \\nu \\left( \\phi \\left( s _ { 0 } ^ { ( i ) } , a _ { 0 } ^ { ( i ) } \\right) \\right) , \\widehat { f } _ { 2 } ( \\nu ) = - \\displaystyle \\frac { 1 } { n } \\sum _ { i = 1 } ^ { n } \\nu \\left( \\phi \\left( s ^ { ( i ) } , a ^ { ( i ) } \\right) \\right) , } \\\\ & { \\widehat { f } _ { 3 } ( \\nu ) = \\displaystyle \\frac { \\gamma } { n } \\sum _ { i = 1 } ^ { n } \\nu \\left( \\phi \\left( s ^ { \\prime ( i ) } , a ^ { \\prime ( i ) } \\right) \\right) . } \\end{array}\n$$",
+        "text_format": "latex",
+        "bbox": [
+            215,
+            805,
+            779,
+            924
+        ],
+        "page_idx": 14
+    },
+    {
+        "type": "text",
+        "text": "We can bound the error of sample-based estimator with individual errors as: ",
+        "bbox": [
+            171,
+            103,
+            671,
+            118
+        ],
+        "page_idx": 15
+    },
+    {
+        "type": "equation",
+        "img_path": "images/b0c68ff1a53c4effe6dff949d0bead0c77d624f5418e389ef1debb02c711ae45.jpg",
+        "text": "$$\n\\begin{array} { r l } { \\mathrm { I P M } _ { \\mathcal { G } } ( d _ { \\phi } ^ { \\pi } , d _ { \\phi } ^ { 2 } ) - \\widehat { \\mathrm { I P M } } ( d _ { \\phi } ^ { \\pi } , d _ { \\phi } ^ { \\mathcal { D } } ) \\Bigg | = \\Bigg | \\underset { \\nu \\in \\mathcal { F } } { \\operatorname* { s u p } } | f _ { 1 } ( \\nu ) + f _ { 2 } ( \\nu ) + f _ { 3 } ( \\nu ) | - \\underset { \\nu \\in \\mathcal { F } } { \\operatorname* { s u p } } | \\widehat { f } _ { 1 } ( \\nu ) + \\widehat { f } _ { 2 } ( \\nu ) + \\widehat { f } _ { 3 } ( \\nu ) | \\Bigg | } & { } \\\\ { \\leq \\underset { \\nu \\in \\mathcal { F } } { \\operatorname* { s u p } } | | f _ { 1 } ( \\nu ) + f _ { 2 } ( \\nu ) + f _ { 3 } ( \\nu ) | - | \\widehat { f } _ { 1 } ( \\nu ) + \\widehat { f } _ { 2 } ( \\nu ) + \\widehat { f } _ { 3 } ( \\nu ) | | } & { } \\\\ { \\leq \\underset { \\nu \\in \\mathcal { F } } { \\operatorname* { s u p } } | f _ { 1 } ( \\nu ) + f _ { 2 } ( \\nu ) + f _ { 3 } ( \\nu ) - \\widehat { f } _ { 1 } ( \\nu ) - \\widehat { f } _ { 2 } ( \\nu ) - \\widehat { f } _ { 3 } ( \\nu ) | } & { } \\\\ { \\leq \\underset { \\nu \\in \\mathcal { F } } { \\operatorname* { s u p } } [ | f _ { 1 } ( \\nu ) - \\widehat { f } _ { 1 } ( \\nu ) | + | f _ { 2 } ( \\nu ) - \\widehat { f } _ { 2 } ( \\nu ) | + | f _ { 3 } ( \\nu ) - \\widehat { f } _ { 3 } ( \\nu ) | ] } & { } \\\\ { \\leq \\underset { \\nu \\in \\mathcal { F } } { \\operatorname* { s u p } } [ | f _ { 1 } ( \\nu ) - \\widehat { f } _ { 1 } ( \\nu ) | + | \\widehat { f } _ { 2 } ( \\nu ) - \\widehat { f } _ { 2 } ( \\nu ) | + | f _ { 3 } ( \\nu ) - \\widehat { f } _ { 3 } ( \\nu ) | ] } & { } \\\\  \\leq \\underset { \\nu \\in \\mathcal { F } } { \\operatorname* { s u p } } | f _ { 1 } ( \\nu ) - \\widehat { f } _ { 1 } ( \\nu ) | + \\underset { \\nu \\in \\mathcal { F } } { \\operatorname* { s u p } } | f _   \\end{array}\n$$",
+        "text_format": "latex",
+        "bbox": [
+            181,
+            122,
+            838,
+            287
+        ],
+        "page_idx": 15
+    },
+    {
+        "type": "text",
+        "text": "We then observe that ",
+        "bbox": [
+            173,
+            290,
+            313,
+            304
+        ],
+        "page_idx": 15
+    },
+    {
+        "type": "equation",
+        "img_path": "images/7baccde9ecc5c794c743a3d16fa77db89bf9fece95437c64ff8f77484699b903.jpg",
+        "text": "$$\n\\begin{array} { l } { \\displaystyle \\underset { \\nu \\in \\mathcal { F } } { \\operatorname* { s u p } } \\left| f _ { 1 } ( \\nu ) - \\widehat { f _ { 1 } } ( \\nu ) \\right| = ( 1 - \\gamma ) \\Biggl \\{ \\mathbb { E } _ { s \\sim d _ { 0 } , a \\sim \\pi ( s ) } [ k ( \\phi ( s , a ) , \\phi ( \\bar { s } , \\bar { a } ) ) ] } \\\\ { \\displaystyle - \\frac { 2 } { n } \\sum _ { i = 1 } ^ { n } \\mathbb { E } _ { \\bar { s } \\sim d _ { 0 } , \\bar { a } \\sim \\pi ( s ) } \\left[ k \\left( \\phi \\left( s ^ { ( i ) } , a ^ { ( i ) } \\right) , \\phi ( \\bar { s } , \\bar { a } ) \\right) \\right] } \\\\ { \\displaystyle + \\frac { 1 } { n ^ { 2 } } \\sum _ { i = 1 } ^ { n } \\sum _ { j = 1 } ^ { n } k \\left( \\phi \\left( s ^ { ( i ) } , a ^ { ( i ) } \\right) , \\phi \\left( s ^ { ( j ) } , a ^ { ( j ) } \\right) \\right) \\Biggl \\} ^ { 1 / 2 } , } \\end{array}\n$$",
+        "text_format": "latex",
+        "bbox": [
+            205,
+            308,
+            790,
+            431
+        ],
+        "page_idx": 15
+    },
+    {
+        "type": "text",
+        "text": "which shows that changing $s ^ { ( i ) } , a ^ { ( i ) }$ results in changes of ${ \\mathrm { s u p } } _ { \\nu \\in { \\mathcal { F } } } \\left| f _ { 1 } ( \\nu ) - { \\widehat { f } } _ { 1 } ( \\nu ) \\right|$ in magnitude of at most $2 ( 1 - \\gamma ) \\bar { k } ^ { 1 / 2 } / n$ where ${ \\bar { k } } = \\operatorname* { s u p } _ { z \\in { \\mathcal { Z } } } k ( z , z )$ . Therefore, by McDiarmid’s inequality (Theorem A.1), ",
+        "bbox": [
+            174,
+            435,
+            826,
+            489
+        ],
+        "page_idx": 15
+    },
+    {
+        "type": "equation",
+        "img_path": "images/ff86b5f506316d6e41e4f1efcef98f23ccb6537b98b1135dfc3e9d6e477d1eb0.jpg",
+        "text": "$$\n\\operatorname* { P r } \\left\\{ \\operatorname* { s u p } _ { \\nu \\in \\mathcal { F } } \\left| f _ { 1 } ( \\nu ) - \\widehat f _ { 1 } ( \\nu ) \\right| - \\mathbb { E } _ { s ^ { ( i ) } , a ^ { ( i ) } } \\left[ \\operatorname* { s u p } _ { \\nu \\in \\mathcal { F } } \\left| f _ { 1 } ( \\nu ) - \\widehat f _ { 1 } ( \\nu ) \\right| \\right] \\geq \\epsilon \\right\\} \\leq \\exp \\left( - \\frac { n \\epsilon ^ { 2 } } { 2 ( 1 - \\gamma ) ^ { 2 } \\bar { k } } \\right) .\n$$",
+        "text_format": "latex",
+        "bbox": [
+            191,
+            493,
+            805,
+            529
+        ],
+        "page_idx": 15
+    },
+    {
+        "type": "text",
+        "text": "Also, ",
+        "bbox": [
+            174,
+            534,
+            210,
+            547
+        ],
+        "page_idx": 15
+    },
+    {
+        "type": "equation",
+        "img_path": "images/2bd7abe170a7ba5c16bd7e0b5e3d05a75f74804940fc49c3b0a3aa28c8176a71.jpg",
+        "text": "$$\n\\begin{array} { r l } & { \\mathbb { E } _ { s ^ { ( i ) } , \\sigma ^ { ( k ) } } [ \\underset { s ^ { ( r , F ) } } { \\operatorname* { s u p } } [ f _ { 1 } ( \\nu ) - \\widehat { f } _ { 1 } ( \\nu ) ] ] } \\\\ & { = \\frac { 1 - \\gamma } { n } \\underset { s ^ { ( i ) } , a ^ { ( i ) } , b ^ { ( i ) } } { \\operatorname* { s u p } } [ \\underset { s ^ { ( r , F ) } } { \\operatorname* { s u p } } [ \\underset { s ^ { ( i ) } , a ^ { ( i ) } , b ^ { ( i ) } } { \\operatorname* { s u p } } [ \\underset { i = 1 } { \\overset { n } { \\sum } } \\nu ( \\phi ( \\tilde { s } _ { 0 } ^ { ( i ) } , \\bar { a } _ { 0 } ^ { ( i ) } ) ) ] - \\underset { i = 1 } { \\overset { n } { \\sum } } \\nu ( \\phi ( s _ { 0 } ^ { ( i ) } , a _ { 0 } ^ { ( i ) } ) ) ) ] } \\\\ & { \\leq \\frac { 1 - \\gamma } { n } \\underset { s ^ { ( i ) } , a ^ { ( i ) } , b ^ { ( i ) } , a ^ { ( i ) } , b ^ { ( i ) } } { \\operatorname* { s u p } } [ \\underset { s ^ { ( r , F ) } } { \\operatorname* { s u p } } [ \\underset { i = 1 } { \\overset { n } { \\sum } } \\nu ( \\phi ( \\frac { \\sigma ^ { ( i ) } } { s } , \\bar { a } _ { 0 } ^ { ( i ) } ) ) - \\underset { i = 1 } { \\overset { n } { \\sum } } \\nu ( \\phi ( s _ { 0 } ^ { ( i ) } , a _ { 0 } ^ { ( i ) } ) ) ) ] } \\\\ & { = \\frac { 1 - \\gamma } { n } \\underset { s ^ { ( i ) } , a ^ { ( i ) } , b ^ { ( i ) } , a ^ { ( i ) } , b ^ { ( i ) } } { \\operatorname* { s u p } } [ \\underset { s ^ { ( r , F ) } } { \\operatorname* { s u p } } [ \\underset { i = 1 } { \\overset { n } { \\sum } } \\sigma ^ { ( i ) 1 } \\{ \\nu ( \\phi ( \\tilde { s } _ { 0 } ^ { ( i ) } , \\bar { a } _ { 0 } ^ { ( i ) } ) ) - \\nu ( \\phi ( s _ { 0 } ^ { ( i ) } , a _ { 0 } ^ { ( i ) } ) ) \\} ] } \\\\ &  \\leq 2 ( 1 - \\gamma )  \\end{array}\n$$",
+        "text_format": "latex",
+        "bbox": [
+            202,
+            551,
+            795,
+            753
+        ],
+        "page_idx": 15
+    },
+    {
+        "type": "text",
+        "text": "where the last inequality is from Lemma A.2. Combining the results, we get ",
+        "bbox": [
+            173,
+            755,
+            674,
+            771
+        ],
+        "page_idx": 15
+    },
+    {
+        "type": "equation",
+        "img_path": "images/7f260310c1dac7b688d86d84bfd912dda7131f603013ed051c33a3a5dc9d6360.jpg",
+        "text": "$$\n\\operatorname* { P r } \\left\\{ \\operatorname* { s u p } _ { \\nu \\in \\mathcal { F } } \\left| f _ { 1 } ( \\nu ) - \\widehat { f } _ { 1 } ( \\nu ) \\right| - 2 ( 1 - \\gamma ) \\sqrt { \\frac { \\bar { k } } { n } } \\geq \\epsilon \\right\\} \\leq \\exp \\left( - \\frac { n \\epsilon ^ { 2 } } { 2 ( 1 - \\gamma ) ^ { 2 } \\bar { k } } \\right) .\n$$",
+        "text_format": "latex",
+        "bbox": [
+            251,
+            773,
+            745,
+            818
+        ],
+        "page_idx": 15
+    },
+    {
+        "type": "text",
+        "text": "Similarly, we derive bounds for $f _ { 2 }$ and $f _ { 3 }$ respectively: ",
+        "bbox": [
+            173,
+            820,
+            534,
+            837
+        ],
+        "page_idx": 15
+    },
+    {
+        "type": "equation",
+        "img_path": "images/09fb04b83ee29d40bceeaf191a8532c8bdaa6bd877c408d43fa97c37a0c896ab.jpg",
+        "text": "$$\n\\begin{array} { l l } { { \\displaystyle \\operatorname* { P r } \\left\\{ \\operatorname* { s u p } _ { \\nu \\in \\mathcal { F } } \\Big | f _ { 2 } ( \\nu ) - \\widehat { f } _ { 2 } ( \\nu ) \\Big | - 2 \\sqrt \\frac { \\bar { k } } { n } \\geq \\epsilon \\right\\} \\leq \\exp \\left( - \\frac { n \\epsilon ^ { 2 } } { 2 \\bar { k } } \\right) , } } & { { } } \\\\ { { \\displaystyle \\operatorname* { P r } \\left\\{ \\operatorname* { s u p } _ { \\nu \\in \\mathcal { F } } \\Big | f _ { 3 } ( \\nu ) - \\widehat { f } _ { 3 } ( \\nu ) \\Big | - 2 \\gamma \\sqrt \\frac { \\bar { k } } { n } \\geq \\epsilon \\right\\} \\leq \\exp \\left( - \\frac { n \\epsilon ^ { 2 } } { 2 \\gamma ^ { 2 } \\bar { k } } \\right) . } } \\end{array}\n$$",
+        "text_format": "latex",
+        "bbox": [
+            292,
+            840,
+            705,
+            929
+        ],
+        "page_idx": 15
+    },
+    {
+        "type": "text",
+        "text": "By letting RHS of above bounds to be $\\delta / 3$ and using union bound, we get, with probability $1 { - } \\delta$ , we get ",
+        "bbox": [
+            169,
+            102,
+            825,
+            131
+        ],
+        "page_idx": 16
+    },
+    {
+        "type": "equation",
+        "img_path": "images/7b40792537770ffb37a39911addd230c15068105f605b926770d665fa2554f28.jpg",
+        "text": "$$\n\\left| \\mathrm { I P M } _ { \\mathcal { G } } ( d _ { \\phi } ^ { \\pi } , d _ { \\phi } ^ { D } ) - \\widehat { \\mathrm { I P M } } ( d _ { \\phi } ^ { \\pi } , d _ { \\phi } ^ { D } ) \\right| \\leq \\sqrt { \\frac { \\bar { k } } { n } } \\left( 4 + \\sqrt { 8 \\log \\frac { 3 } { \\delta } } \\right) .\n$$",
+        "text_format": "latex",
+        "bbox": [
+            297,
+            128,
+            699,
+            171
+        ],
+        "page_idx": 16
+    },
+    {
+        "type": "text",
+        "text": "The relationship between $\\mathcal { F }$ and $\\mathcal { G }$ ",
+        "bbox": [
+            174,
+            204,
+            401,
+            219
+        ],
+        "page_idx": 16
+    },
+    {
+        "type": "equation",
+        "img_path": "images/46d48b0cf72d25c32debc83d9775166ef3e7e3825d5a91191375f7d2e616b369.jpg",
+        "text": "$$\n\\mathcal { J } = \\left\\{ g \\in ( \\mathcal { Z } \\to \\mathbb { R } ) : g ( z ) = \\nu ( z ) - \\gamma \\mathbb { E } _ { s ^ { \\prime } \\sim T ( \\phi ^ { - 1 } ( z ) ) } [ \\nu ( \\phi ( s ^ { \\prime } , a ^ { \\prime } ) ) ] , \\nu \\in ( \\mathcal { Z } \\to \\mathbb { R } ) , \\langle \\nu , \\nu \\rangle _ { \\mathcal { H } _ { k } } \\leq 1 \\right\\}\n$$",
+        "text_format": "latex",
+        "bbox": [
+            181,
+            223,
+            825,
+            265
+        ],
+        "page_idx": 16
+    },
+    {
+        "type": "text",
+        "text": "show that when the conditional expectation $\\mathbb { E } _ { s ^ { \\prime } \\sim T ( \\phi ^ { - 1 } ( \\cdot ) ) , a ^ { \\prime } \\sim \\pi ( s ^ { \\prime } ) } [ \\nu ( \\phi ( s ^ { \\prime } , a ^ { \\prime } ) ) ] : \\mathcal { Z } \\to \\mathbb { R }$ is a function in RKHS $\\mathcal { H } _ { k }$ , $\\mathcal { G }$ also becomes a subset of $\\mathcal { H } _ { k }$ . ",
+        "bbox": [
+            174,
+            271,
+            821,
+            300
+        ],
+        "page_idx": 16
+    },
+    {
+        "type": "text",
+        "text": "Then we can prove the following Theorem. ",
+        "bbox": [
+            174,
+            306,
+            457,
+            321
+        ],
+        "page_idx": 16
+    },
+    {
+        "type": "text",
+        "text": "Theorem 4.3. Given an MDP $M$ , its estimate $\\widehat { M }$ with a bijective representation function $\\phi$ , i.e. $( \\widehat { T } , \\widehat { R } ) = \\langle \\widehat { T } _ { z } \\circ \\phi , \\widehat { R } _ { z } \\circ \\phi \\rangle$ , and an RKHS $\\begin{array} { r } { \\mathcal { H } _ { k } \\quad \\subset \\quad ( \\mathcal { Z } \\quad \\to \\quad \\mathbb { R } ) } \\end{array}$ induced by a universal kernel $k$ such that $\\mathrm { s u p } _ { z \\in \\mathcal { Z } } k ( z , z ) ~ = ~ \\bar { k }$ , assume that $f _ { \\phi , \\widehat { R } _ { z } , \\widehat { T } _ { z } } \\left( \\widehat { z } \\right) \\ =$ $\\begin{array} { r } { \\mathbb { E } _ { T , \\pi } \\Big [ \\sum _ { t = 0 } ^ { \\infty } \\gamma ^ { t } \\mathcal { E } _ { \\phi , \\widehat { R } _ { z } , \\widehat { T } _ { z } } \\big ( s _ { t } , a _ { t } \\big ) \\Big | \\big ( s _ { 0 } , a _ { 0 } \\big ) = \\phi ^ { - 1 } \\big ( z \\big ) \\Big ] \\in \\mathcal { H } _ { k } } \\end{array}$ with $B _ { \\phi } = \\| f _ { \\phi , \\widehat { R } _ { z } , \\widehat { T } _ { z } } \\| _ { \\mathcal { H } _ { k } }$ and the loss is bounded by $\\begin{array} { r } { \\bar { \\mathcal { E } } = \\operatorname* { s u p } _ { s \\in { \\mathcal { S } } , a \\in \\mathcal { A } } \\mathcal { E } _ { \\phi , \\widehat { R } _ { z } , \\widehat { T } _ { z } } ( s , a ) } \\end{array}$ . Let n be the number of data in $\\mathcal { D }$ . With probability $1 - 2 \\delta$ , ",
+        "bbox": [
+            171,
+            325,
+            826,
+            431
+        ],
+        "page_idx": 16
+    },
+    {
+        "type": "equation",
+        "img_path": "images/38315fc6b9bb6240f5da5e207809b2ab70cd1e301d17089118a70400757cb45d.jpg",
+        "text": "$$\nR ^ { \\pi } - { \\widehat R } ^ { \\pi } \\Big | \\leq \\frac { 1 } { n } \\sum _ { ( s , a ) \\in { \\mathcal D } } { \\mathcal E } _ { \\phi , { \\widehat R } _ { z } , { \\widehat T } _ { z } } ( s , a ) + B _ { \\phi } \\widehat \\Pi \\widehat M \\big ( d _ { \\phi } ^ { \\pi } , d _ { \\phi } ^ { { D } } \\big ) + \\sqrt { \\frac { \\bar { \\mathcal E } ^ { 2 } } { 2 n } \\log { \\frac { 1 } { \\delta } } } + B _ { \\phi } \\sqrt { \\frac { \\bar { k } } { n } } \\left( 4 + \\sqrt { 8 \\log { \\frac { 3 } { \\delta } } } \\right) .\n$$",
+        "text_format": "latex",
+        "bbox": [
+            181,
+            435,
+            826,
+            481
+        ],
+        "page_idx": 16
+    },
+    {
+        "type": "text",
+        "text": "Proof. Applying the Hoeffding inequality (Hoeffding, 1963), with probability $1 - \\delta$ , we get: ",
+        "bbox": [
+            171,
+            493,
+            782,
+            511
+        ],
+        "page_idx": 16
+    },
+    {
+        "type": "equation",
+        "img_path": "images/a143e9f568260a599d52a8c6d831e1d3e60ae1e101d09e4e264932b139e1cebe.jpg",
+        "text": "$$\n\\mathbb { E } _ { ( s , a ) \\sim d ^ { \\mathcal { D } } } \\left[ \\mathcal { E } _ { \\phi , \\widehat { R } _ { z } , \\widehat { T } _ { z } } ( s , a ) \\right] \\leq \\frac { 1 } { n } \\sum _ { ( s , a ) \\sim \\mathcal { D } } \\mathcal { E } _ { \\phi , \\widehat { R } _ { z } , \\widehat { T } _ { z } } ( s , a ) + \\sqrt { \\frac { \\bar { \\mathcal { E } } ^ { 2 } } { 2 n } \\log \\frac { 1 } { \\delta } } .\n$$",
+        "text_format": "latex",
+        "bbox": [
+            266,
+            516,
+            730,
+            560
+        ],
+        "page_idx": 16
+    },
+    {
+        "type": "text",
+        "text": "By using an union bound with Eq. (13) and substituting terms in Eq. (3), we recover the result. ",
+        "bbox": [
+            169,
+            566,
+            794,
+            583
+        ],
+        "page_idx": 16
+    },
+    {
+        "type": "text",
+        "text": "B EXPERIMENT DETAILS ",
+        "text_level": 1,
+        "bbox": [
+            176,
+            102,
+            397,
+            117
+        ],
+        "page_idx": 17
+    },
+    {
+        "type": "text",
+        "text": "B.1 COMPUTING INFRASTRUCTURE ",
+        "text_level": 1,
+        "bbox": [
+            176,
+            133,
+            434,
+            148
+        ],
+        "page_idx": 17
+    },
+    {
+        "type": "text",
+        "text": "All experiments were conducted on the Google Cloud Platform. Specifically, we used computeoptimized machines (c2-standard-4) that provide 4 vCPUs and 16 GB memory for the evaluation experiment of Section 5.1, and we used high-memory machines (n1-highmem-4), which provide 4 vCPUs and 26GB memory, equipped with an Nvidia Tesla K80 GPU for the RL experiment of Section 5.2. ",
+        "bbox": [
+            174,
+            160,
+            825,
+            231
+        ],
+        "page_idx": 17
+    },
+    {
+        "type": "text",
+        "text": "B.2 DETAILS OF THE OPE EXPERIMENT ",
+        "text_level": 1,
+        "bbox": [
+            178,
+            248,
+            464,
+            262
+        ],
+        "page_idx": 17
+    },
+    {
+        "type": "text",
+        "text": "Task details We did not modify CartPole-v0 environment and Acrobot-v1 environment from the original implementation of OpenAI Gym (Brockman et al., 2016) except for the maximum trajectory length. We ran PPO (Schulman et al., 2017) to optimize policies to reach a certain performance level and set them as the target policies for CartPole-v0 and Acrobot-v1. For the HIV simulator, we used the code adapted by Liu et al. (2018b), which is originally from the implementation of $\\mathsf { R L P y } ^ { 3 }$ . We modified the environment to have more randomness in the initial state (up to $1 0 \\%$ perturbation from the baseline initial state) and to use the reward function that gives the logarithm of original reward values, as the original reward function scales up to $1 0 ^ { 1 0 }$ . We used a tree-based fitted ${ \\bf q }$ -iteration algorithm implemented by Liu et al. (2018b) to optimize the target policy for the HIV simulator. All the other details are shown in Table 2. We assume that the termination conditions of tasks are known in prior. ",
+        "bbox": [
+            173,
+            275,
+            825,
+            428
+        ],
+        "page_idx": 17
+    },
+    {
+        "type": "table",
+        "img_path": "images/ff1293c514c4b257cc52982e06a45ba9a96fe99a6b4e136baaa72b7cf5b218a4.jpg",
+        "table_caption": [
+            "Table 2: Task settings of OPE experiments. "
+        ],
+        "table_footnote": [],
+        "table_body": "<table><tr><td></td><td>CartPole-vO</td><td>Acrobot-v1</td><td>HIV simulator</td></tr><tr><td></td><td></td><td></td><td></td></tr><tr><td>state space dimension</td><td>4</td><td>6</td><td>6</td></tr><tr><td># of actions</td><td>2</td><td>3</td><td>4</td></tr><tr><td>discount rate y</td><td>0.99</td><td>0.99</td><td>0.98</td></tr><tr><td># of trajectories</td><td>200</td><td>200</td><td>50</td></tr><tr><td>max length of training traj.</td><td>200</td><td>200</td><td>200</td></tr><tr><td>max length of rollouts for evaluation</td><td>1000</td><td>1000</td><td>1000</td></tr><tr><td>discounted return of target policy</td><td>89.4</td><td>-56.88</td><td>803.2</td></tr></table>",
+        "bbox": [
+            217,
+            472,
+            782,
+            603
+        ],
+        "page_idx": 17
+    },
+    {
+        "type": "text",
+        "text": "Model and algorithm details The model we learn is composed of a representation module and a dynamics module. To be consistent with the experiment settings in Liu et al. (2018b), we use a representation module of a single hidden layer feed-forward network that takes the state as input and outputs representation. We squashed the representation between $( - 1 , 1 )$ using the tanh activation function. The dynamics module is also a single hidden layer feed-forward network that takes representation as input and outputs state difference and reward prediction for each action. We use the swish activation function (Ramachandran et al., 2017) for the hidden layers of two modules. As a whole, the model can also be seen as a feed-forward network with three hidden layers of varying activation functions, where the output of the second hidden layer is the representation we regularize. ",
+        "bbox": [
+            173,
+            640,
+            825,
+            766
+        ],
+        "page_idx": 17
+    },
+    {
+        "type": "text",
+        "text": "For the purpose of comparison, we minimize the $L 2$ distance between the model prediction and the desired outcome from data, which corresponds to using a model of Gaussian predictive distribution with fixed variance. We standardized the inputs and outputs of the neural network and used Adam (Kingma & Ba, 2014) with a learning rate of $3 \\times 1 0 ^ { - 4 }$ for the optimization. When compared to the similar experiments conducted in Liu et al. (2018b), we used a larger and more expressive model with more optimization steps with a smaller learning rate for a more accurate comparison. While the derivation of the RepB-SDE objective was based on the state-action representation function, we use state representation in this experiment for direct comparison with RepBM, which uses state representation (it can be also understood as using action invariant kernel). We follow the choice of Liu et al. (2018b) and use dot product kernel $k ( \\phi ( s ) , \\phi ( \\bar { s } ) ) = \\phi ( s ) ^ { \\top } \\phi ( \\bar { s } )$ for the OPE experiment, which is not universal but allows us to avoid search of kernel hyperparameters, such as length-scales. ",
+        "bbox": [
+            174,
+            772,
+            825,
+            897
+        ],
+        "page_idx": 17
+    },
+    {
+        "type": "text",
+        "text": "",
+        "bbox": [
+            171,
+            102,
+            823,
+            132
+        ],
+        "page_idx": 18
+    },
+    {
+        "type": "text",
+        "text": "After the training, we generate another 200 trajectories (50 in case of HIV simulator), and rollout in both true and simulated (based on learned model) environments to evaluate models. We measure the individual MSE, which is ",
+        "bbox": [
+            176,
+            138,
+            823,
+            181
+        ],
+        "page_idx": 18
+    },
+    {
+        "type": "equation",
+        "img_path": "images/8f8df59f7994214c467ec82d6d2b271632da645a86884b5b445941a6e4e829c3.jpg",
+        "text": "$$\n\\mathrm { I n d i v i d u a l ~ M S E } = \\mathbb { E } _ { s _ { 0 } \\sim d _ { 0 } } \\bigg [ \\bigg ( \\mathbb { E } _ { \\widehat { M } , \\pi } \\Big [ \\sum _ { t = 0 } ^ { \\infty } \\gamma ^ { t } r _ { t } | s _ { 0 } \\Big ] - \\mathbb { E } _ { M , \\pi } \\Big [ \\sum _ { t = 0 } ^ { \\infty } \\gamma ^ { t } r _ { t } | s _ { 0 } \\Big ] \\bigg ) ^ { 2 } \\bigg ]\n$$",
+        "text_format": "latex",
+        "bbox": [
+            245,
+            188,
+            750,
+            228
+        ],
+        "page_idx": 18
+    },
+    {
+        "type": "text",
+        "text": "for measuring the performance of each model. Whole experiment, from sampling data to learning and evaluating the model, is repeated 200 times with different random seeds. ",
+        "bbox": [
+            174,
+            234,
+            823,
+            262
+        ],
+        "page_idx": 18
+    },
+    {
+        "type": "text",
+        "text": "Choice and effect of hyperparameter $\\alpha$ For choosing hyperparameter $\\alpha$ for each algorithm, we searched over $\\alpha \\in \\{ 0 . 0 0 1 , 0 . 0 1 , 0 . 1 , 1 , 1 0 \\}$ for each off-policyness $\\epsilon$ and for each environment. Chosen αs were mainly $\\alpha \\in$ $\\lbrace 0 . 0 0 1 , 0 . 0 1 \\rbrace$ for CartPole-v0, $\\alpha \\in \\{ 1 , 1 0 \\}$ for Acrobotv1, and $\\alpha \\in \\{ 0 . 0 1 , 0 . 1 \\}$ for HIV simulator for both algorithms. In general, large $\\alpha$ was beneficial when high off-policyness (\u000f) is present and/or the task is hard to generalize. On the right we show the example of effect of varying $\\alpha$ in CartPole-v0. ",
+        "bbox": [
+            174,
+            268,
+            545,
+            407
+        ],
+        "page_idx": 18
+    },
+    {
+        "type": "image",
+        "img_path": "images/1c9ccb5c2c41ddb695510e4975c65cec88c2fe83b3762b2ff076037bc3277679.jpg",
+        "image_caption": [],
+        "image_footnote": [],
+        "bbox": [
+            553,
+            270,
+            818,
+            402
+        ],
+        "page_idx": 18
+    },
+    {
+        "type": "text",
+        "text": "Comparison to other baselines In Figure 2, the OPE results with other model-free baselines are presented. FQE: Fitted Q-evaluation, IS: step-wise importance sampling, DR: doubly robust estimator based on step-wise importance sampling using the value function learned with fitted Qevaluation (Jiang & Li, 2016), DualDICE: stationary distribution correction algorithm (Nachum et al., 2019a). We used the implementation provided by the authors in case of DualDICE. All results are normalized to set the MSE of model-based baseline to be 1. Here, we used average MSEs instead of individual MSEs since importance sampling estimators are not suitable in computing individual MSEs. The results show that model-based methods are more robust to increasing off-policyness when compared to FQE. The results of model-based methods on Acrobot is relatively worse due to the difficult dynamics of Acrobot environment. ",
+        "bbox": [
+            173,
+            421,
+            825,
+            561
+        ],
+        "page_idx": 18
+    },
+    {
+        "type": "image",
+        "img_path": "images/1eb2b85496bae8da7d8f1105ae49dddbe789c2255747b8997f70703f372d78dd.jpg",
+        "image_caption": [
+            "Figure 2: The OPE results compared to other model-free baselines. All experiments are repeated 200 times and the error bars denote $9 5 \\%$ confidence interval. "
+        ],
+        "image_footnote": [],
+        "bbox": [
+            173,
+            573,
+            826,
+            722
+        ],
+        "page_idx": 18
+    },
+    {
+        "type": "text",
+        "text": "B.3 DETAILS OF THE OFFLINE RL EXPERIMENTS ",
+        "text_level": 1,
+        "bbox": [
+            174,
+            786,
+            524,
+            800
+        ],
+        "page_idx": 18
+    },
+    {
+        "type": "text",
+        "text": "Task details We used 12 datasets in D4RL $\\mathrm { F u }$ et al., 2020) over four dataset types and three environments as specified in the main text. In Table 1, normalized scores suggested by (Fu et al., 2020) is used to report the result, where score of 0 corresponds to a random policy and 100 corresponds to a converged SAC policy. In HalfCheetah-v2, score of 0 means the undiscounted return of $- 2 8 0$ , and score of 1 means the undiscounted return of 12135. In Hopper-v2, score of 0 means the undiscounted return of $- 2 0$ , and score of 1 means the undiscounted return of 3234. In Walker2d-v2 score of 0 means the undiscounted return of 2, and score of 1 means the undiscounted return of 4592. We assume that the termination conditions of tasks are known in prior. ",
+        "bbox": [
+            173,
+            811,
+            825,
+            924
+        ],
+        "page_idx": 18
+    },
+    {
+        "type": "text",
+        "text": "Representation balancing maximizing undiscounted return As we report the undiscounted sum of rewards in the experiments, the maximization of lower bound of $R ^ { \\pi }$ may result in an underutilization of experiences of later timesteps. One way to mitigate the mismatch is to optimize the policy by maximizing the returns starting from the states in the dataset. It corresponds to maximizing $\\begin{array} { r } { \\tilde { R } ^ { \\pi } = ( 1 - \\gamma ) \\mathbb { E } , M , \\pi \\bigl [ \\sum _ { t = 0 } ^ { \\infty } \\gamma ^ { t } r _ { t } \\bigr ] } \\end{array}$ instead of $R ^ { \\pi }$ , where the expectation respect to the initial state distribution $d _ { 0 }$ is altered with the data distribution $d ^ { \\mathcal { D } }$ . ",
+        "bbox": [
+            173,
+            103,
+            825,
+            191
+        ],
+        "page_idx": 19
+    },
+    {
+        "type": "text",
+        "text": "Consequently, to bound the error of ${ \\tilde { R } } ^ { \\pi }$ , the representation should be balanced with another dis  \ncounted stationary distribution $\\begin{array} { r } { \\tilde { d } ^ { \\pi } ( s , a ) \\triangleq ( 1 - \\gamma ) \\sum _ { t = 0 } ^ { \\infty } \\gamma ^ { t } \\operatorname* { P r } ( s _ { t } = s , a _ { t } = a | s _ { 0 } \\sim d ^ { \\mathcal { D } } , T , \\pi ) } \\end{array}$ , the $\\pi$   \nderivations can be easily adapted by noting that: ",
+        "bbox": [
+            173,
+            196,
+            825,
+            257
+        ],
+        "page_idx": 19
+    },
+    {
+        "type": "equation",
+        "img_path": "images/14ef59d4631f76c8b8c1477cb06be9a612301e6f5f17325e6aa561ff27047134.jpg",
+        "text": "$$\n\\mathrm { P M } _ { \\mathcal { G } } ( \\tilde { d } _ { \\phi } ^ { \\pi } , d _ { \\phi } ^ { \\mathcal { D } } ) = \\operatorname* { s u p } _ { \\nu \\in \\mathcal { F } } \\left| ( 1 - \\gamma ) \\mathbb { E } _ { \\alpha \\sim \\pi ( s ) } \\left[ \\nu \\big ( \\phi ( s , a ) \\big ) \\right] - \\mathbb { E } _ { ( s , a , s ^ { \\prime } ) \\sim d ^ { \\mathcal { D } } } \\left[ \\nu \\big ( \\phi ( s , a ) \\big ) - \\gamma \\nu \\big ( \\phi ( s ^ { \\prime } , a ^ { \\prime } ) \\big ) \\right] \\right| ,\n$$",
+        "text_format": "latex",
+        "bbox": [
+            181,
+            262,
+            826,
+            306
+        ],
+        "page_idx": 19
+    },
+    {
+        "type": "text",
+        "text": "and changing the initial state sampling distributions to $d ^ { \\mathcal { D } }$ during the estimation of IPM. ",
+        "bbox": [
+            171,
+            314,
+            750,
+            330
+        ],
+        "page_idx": 19
+    },
+    {
+        "type": "text",
+        "text": "Model and algorithm details Similar to the model used in the OPE experiment, the model we learn is composed of a representation module and a dynamics module. A representation module is a feed-forward network with two hidden layers that takes the state-action pair as input and outputs representation through the tanh activation function. The dynamics module is a single hidden layer network that takes representation as input and outputs parameters of diagonal Gaussian distribution predicting state difference and reward. We use 200 hidden units for all intermediate layers including the representation layer. Across all domains, we train an ensemble of 7 models and pick the best 5 models on their validation error on hold-out set of 1000 transitions in the dataset. The inputs and outputs of the neural network is normalized. We present the pseudo-code of the presented Representation Balancing Offline Model-based RL algorithm below. ",
+        "bbox": [
+            173,
+            344,
+            825,
+            484
+        ],
+        "page_idx": 19
+    },
+    {
+        "type": "text",
+        "text": "Input: Offline dataset $\\mathcal { D }$ , previous policy $\\pi$ Output: Optimized policy $\\pi$   \n1: Sample $K$ independent datasets with replacement from $\\mathcal { D }$ .   \n2: Train bootstrapped ensemble of $K$ models $\\{ \\widehat { T } _ { i } , \\widehat { R } _ { i } \\} _ { i = 0 } ^ { K }$ minimizing Eq. (8) (adapted with $\\tilde { d } _ { \\phi } ^ { \\pi } ,$ ).   \n3: for rep ${ \\mathcal { a } } t = 0 , 1 , \\ldots$ do   \n4: for rollout $\\mathbf { \\Sigma } = 0 , 1 , \\dots , B$ do   \n5: Sample initial rollout state $s _ { 0 }$ from $\\mathcal { D }$ .   \n6: for $t = 0 , 1 , \\ldots$ do   \n7: Sample an action $a _ { t } \\sim \\pi ( s _ { t } )$ .   \n8: Randomly pick $( \\widehat { T } _ { i } , \\widehat { R } _ { i } )$ and sample $( s _ { t + 1 } , r _ { t } ) \\sim ( \\widehat { T } _ { i } , \\widehat { R } _ { i } ) ( s _ { t } , a _ { t } ) .$   \n9: Compute $ \\tilde { r } _ { t } = r _ { t } - \\gamma \\lambda \\sqrt { \\mathbb { V } _ { K } [ \\mu ( s _ { t } , a _ { t } ) ] } | | _ { 2 }$ and store $\\left( { { s _ { t } } , { a _ { t } } , \\tilde { r } _ { t } , { s _ { t + 1 } } } \\right)$ in $\\widehat { \\mathcal { D } }$ .   \n10: end for   \n11: end for   \n12: Draw samples from $\\mathcal { D }$ to compute $\\widehat { \\mathrm { I P M } } ( \\tilde { d } _ { \\phi } ^ { \\pi } , d _ { \\phi } ^ { D } )$ .   \n13: Draw samples from $\\mathcal { D }$ and $\\widehat { \\mathcal { D } }$ to update critic $Q$ .   \n14: Maximize $\\mathbb { E } _ { s \\sim \\mathcal { D } \\cup \\widehat { \\mathcal { D } } , a \\sim \\pi ( s ) } [ Q ( s , a ) - \\tau \\log \\pi ( a | s ) ] - \\alpha _ { \\pi } \\widehat { \\mathrm { I P M } } ( \\tilde { d } _ { \\phi } ^ { \\pi } , d _ { \\phi } ^ { \\mathcal { D } } )$ to update $\\pi$ .   \n15: end for ",
+        "bbox": [
+            178,
+            517,
+            818,
+            761
+        ],
+        "page_idx": 19
+    },
+    {
+        "type": "text",
+        "text": "For the result of MOPO (Yu et al., 2020), we ran the code kindly provided by the authors4 without any modification on the algorithm or the hyperparameters. All algorithms we experimented (Base, RP, RepB-SDE) share all the hyperparameters in common except the ones associated with changing objectives. We run SAC on the full rollouts from the trained ensemble models as shown in Algorithm 1. The common hyperparameters shared among algorithms are shown in Table 3. We simply tried the listed hyperparameters and not tuned them further. For RP and RepB-SDE, we penalized the reward from simulated environments with the standard deviation of prediction means of neural network ensembles. We used standardized output of all 7 neural networks to compute the reward penalty. We first search over the reward penalty coefficient $\\lambda \\in \\{ 0 , 2 , 5 , 7 , 1 0 , 1 5 \\}$ that grants RP the best performance. We shared same $\\lambda$ for RepB-SDE and searched over $\\alpha _ { M } \\in \\{ 0 . 0 1 , 0 . 1 , 1 \\}$ , $\\alpha _ { \\pi } ~ \\in ~ \\{ 0 , 0 . 0 1 , 0 . 1 , 1 , 1 0 \\}$ . We ran the algorithms for $1 . 5 \\times 1 0 ^ { 6 }$ gradient updates, except for HalfCheetah-Medium-Expert where we ran for $5 . 0 \\times 1 0 ^ { 6 }$ . In Table 4, we present the standard errors of results, which was omitted in the main text. ",
+        "bbox": [
+            173,
+            786,
+            825,
+            898
+        ],
+        "page_idx": 19
+    },
+    {
+        "type": "table",
+        "img_path": "images/d1bf4bbb33fafb716bba9830bdaa0eb212a3b37e4999d735955451123d3aec44.jpg",
+        "table_caption": [
+            "Table 3: Common hyperparameters used in offline RL experiments. "
+        ],
+        "table_footnote": [],
+        "table_body": "<table><tr><td>Parameter Value</td></tr><tr><td>optimizer Adam (Kingma &amp; Ba, 2014)</td></tr><tr><td>learning rate 3×10-4</td></tr><tr><td>discount factor y 0.99</td></tr><tr><td>number of samples per minibatch 256</td></tr><tr><td>target smoothing coefficient T 5×10-3</td></tr><tr><td>[actor/critic] number of hidden layers 2</td></tr><tr><td>[actor/critic] number of hidden units per layer 256</td></tr><tr><td>[actor/critic] non-linearity ReLU</td></tr><tr><td># of rollouts 104</td></tr><tr><td>max length of rollouts 103</td></tr><tr><td>rollout buffer size 5×107</td></tr></table>",
+        "bbox": [
+            238,
+            132,
+            761,
+            319
+        ],
+        "page_idx": 20
+    },
+    {
+        "type": "table",
+        "img_path": "images/3284e1e9c9ccda068f8a8ccbf08fd85271914a97e4b455c2e3669098579f8a52.jpg",
+        "table_caption": [
+            "Table 4: Normalized scores on D4RL MuJoCo benchmark datasets (Fu et al., 2020) with standard errors fully specified. "
+        ],
+        "table_footnote": [],
+        "table_body": "<table><tr><td>Dataset type</td><td>Environment</td><td>RepB-SDE (ours)</td><td>RP</td><td>Base</td><td>MOPO</td></tr><tr><td></td><td></td><td></td><td></td><td></td><td></td></tr><tr><td>Random</td><td>Walker2d-v2</td><td>21.1 ± 1.0</td><td>18.4 ±1.8</td><td>16.4 ± 2.2</td><td>1.3 ± 0.5</td></tr><tr><td>Medium</td><td>Walker2d-v2</td><td>72.1 ± 1.9</td><td>56.3 ± 4.3</td><td>5.5 ± 0.4</td><td>-0.1 ± 0.0</td></tr><tr><td>Med-Replay</td><td>Walker2d-v2</td><td>49.8 ± 11.4</td><td>41.1 ±10.3</td><td>6.2 ± 0.7</td><td>47.8 ± 4.8</td></tr><tr><td>Med-Expert</td><td>Walker2d-v2</td><td>88.8 ± 6.9</td><td>72.6 ± 5.1</td><td>51.7 ± 7.1</td><td>32.4 ± 8.8</td></tr><tr><td>Random</td><td>Hopper-v2</td><td>8.6 ±1.0</td><td>8.3 ±0.2</td><td>8.3 ±0.2</td><td>9.1 ± 0.3</td></tr><tr><td>Medium</td><td>Hopper-v2</td><td>34.0 ± 2.8</td><td>27.5 ± 3.1</td><td>19.8 ± 0.8</td><td>19.2 ± 4.5</td></tr><tr><td>Med-Replay</td><td>Hopper-v2</td><td>62.2 ± 6.7</td><td>49.8 ± 8.7</td><td>32.9 ± 5.5</td><td>80.8 ± 2.9</td></tr><tr><td>Med-Expert</td><td>Hopper-v2</td><td>82.6 ± 7.0</td><td>74.0 ± 7.2</td><td>19.1 ± 1.1</td><td>23.2 ± 1.3</td></tr><tr><td>Random</td><td>HalfCheetah-v2</td><td>32.9 ± 1.1</td><td>31.3 ±1.5</td><td>26.1 ± 6.2</td><td>29.9 ± 1.2</td></tr><tr><td>Medium</td><td>HalfCheetah-v2</td><td>49.1 ± 0.3</td><td>47.3 ± 0.3</td><td>22.1 ± 4.4</td><td>45.0 ± 0.4</td></tr><tr><td>Med-Replay</td><td>HalfCheetah-v2</td><td>57.5 ± 0.8</td><td>53.6 ± 0.8</td><td>54.0 ± 2.4</td><td>53.8 ± 1.1</td></tr><tr><td>Med-Expert</td><td>HalfCheetah-v2</td><td>55.4 ± 8.3</td><td>36.7 ± 12.0</td><td>31.9 ± 10.5</td><td>91.0 ± 2.5</td></tr></table>",
+        "bbox": [
+            174,
+            382,
+            821,
+            580
+        ],
+        "page_idx": 20
+    },
+    {
+        "type": "text",
+        "text": "",
+        "bbox": [
+            174,
+            613,
+            825,
+            683
+        ],
+        "page_idx": 20
+    },
+    {
+        "type": "text",
+        "text": "In addition to the performance after the final iteration presented in Table 1, we also present the learning curves of the experiment in Figure 3, and the effect of the choice of $\\alpha _ { M }$ and $\\alpha _ { \\pi }$ in Figure 4. The Figure 4 shows that the presented regularization is robust to the choice of $\\alpha _ { M } , \\alpha _ { \\pi }$ except for too large $\\alpha _ { M }$ , consistently improving from RP, which corresponds to $\\alpha _ { M } = 0 , \\alpha _ { \\pi } = 0$ . ",
+        "bbox": [
+            174,
+            690,
+            825,
+            746
+        ],
+        "page_idx": 20
+    },
+    {
+        "type": "image",
+        "img_path": "images/6da69fd1b89d4ec41b0f3dfa3d8233be66aedb17d085f5919e4c11e946c54cdc.jpg",
+        "image_caption": [
+            "Figure 3: The learning curves of the D4RL experiment. "
+        ],
+        "image_footnote": [],
+        "bbox": [
+            173,
+            126,
+            821,
+            536
+        ],
+        "page_idx": 21
+    },
+    {
+        "type": "image",
+        "img_path": "images/bcf6e9d1e6e698ee7c20d46468f932a177986219b3341a07904605f54aa078ee.jpg",
+        "image_caption": [
+            "Figure 4: The effect of varying hyperparameter in the D4RL experiment. "
+        ],
+        "image_footnote": [],
+        "bbox": [
+            171,
+            622,
+            826,
+            868
+        ],
+        "page_idx": 21
+    }
+]
\ No newline at end of file
diff --git a/parse/train/QpNz8r_Ri2Y/QpNz8r_Ri2Y_middle.json b/parse/train/QpNz8r_Ri2Y/QpNz8r_Ri2Y_middle.json
new file mode 100644
index 0000000000000000000000000000000000000000..d30510f25807d412e2c0e2e0c39cca6eeeca9d57
--- /dev/null
+++ b/parse/train/QpNz8r_Ri2Y/QpNz8r_Ri2Y_middle.json
@@ -0,0 +1,58366 @@
+{
+    "pdf_info": [
+        {
+            "preproc_blocks": [
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        79,
+                        504,
+                        116
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                78,
+                                335,
+                                97
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        78,
+                                        335,
+                                        97
+                                    ],
+                                    "score": 1.0,
+                                    "content": "REPRESENTATION BALANCING",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                107,
+                                99,
+                                505,
+                                117
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        99,
+                                        505,
+                                        117
+                                    ],
+                                    "score": 1.0,
+                                    "content": "OFFLINE MODEL-BASED REINFORCEMENT LEARNING",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        }
+                    ],
+                    "index": 0.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        114,
+                        133,
+                        344,
+                        146
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                113,
+                                133,
+                                344,
+                                149
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        113,
+                                        133,
+                                        312,
+                                        149
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Byung-Jun Lee1,3, Jongmin Lee1 & Kee-Eung",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        313,
+                                        134,
+                                        344,
+                                        146
+                                    ],
+                                    "score": 0.49,
+                                    "content": "\\mathbf { K i m } ^ { 1 , 2 }",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 2
+                        }
+                    ],
+                    "index": 2
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        112,
+                        147,
+                        386,
+                        192
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                112,
+                                145,
+                                355,
+                                160
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        112,
+                                        145,
+                                        355,
+                                        160
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1School of Computing, KAIST, Daejeon, Republic of Korea",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                110,
+                                156,
+                                360,
+                                172
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        110,
+                                        156,
+                                        360,
+                                        172
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2Graduate School of AI, KAIST, Daejeon, Republic of Korea",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                111,
+                                168,
+                                290,
+                                183
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        111,
+                                        168,
+                                        290,
+                                        183
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3Gauss Labs Inc., Seoul, Republic of Korea",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                113,
+                                181,
+                                388,
+                                195
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        113,
+                                        181,
+                                        388,
+                                        195
+                                    ],
+                                    "score": 1.0,
+                                    "content": "{bjlee,jmlee}@ai.kaist.ac.kr, kekim@kaist.ac.k",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        }
+                    ],
+                    "index": 4.5
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        278,
+                        221,
+                        333,
+                        233
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                276,
+                                220,
+                                336,
+                                235
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        276,
+                                        220,
+                                        336,
+                                        235
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ABSTRACT",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        }
+                    ],
+                    "index": 7
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        143,
+                        247,
+                        468,
+                        411
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                142,
+                                246,
+                                469,
+                                260
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        142,
+                                        246,
+                                        469,
+                                        260
+                                    ],
+                                    "score": 1.0,
+                                    "content": "One of the main challenges in offline and off-policy reinforcement learning is to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                258,
+                                470,
+                                271
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        258,
+                                        470,
+                                        271
+                                    ],
+                                    "score": 1.0,
+                                    "content": "cope with the distribution shift that arises from the mismatch between the target",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                269,
+                                470,
+                                282
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        269,
+                                        470,
+                                        282
+                                    ],
+                                    "score": 1.0,
+                                    "content": "policy and the data collection policy. In this paper, we focus on a model-based",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                281,
+                                470,
+                                292
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        281,
+                                        470,
+                                        292
+                                    ],
+                                    "score": 1.0,
+                                    "content": "approach, particularly on learning the representation for a robust model of the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                291,
+                                470,
+                                304
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        291,
+                                        470,
+                                        304
+                                    ],
+                                    "score": 1.0,
+                                    "content": "environment under the distribution shift, which has been first studied by Repre-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                302,
+                                469,
+                                314
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        302,
+                                        469,
+                                        314
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sentation Balancing MDP (RepBM). Although this prior work has shown promis-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                313,
+                                469,
+                                326
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        313,
+                                        469,
+                                        326
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ing results, there are a number of shortcomings that still hinder its applicability",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                324,
+                                469,
+                                336
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        324,
+                                        469,
+                                        336
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to practical tasks. In particular, we address the curse of horizon exhibited by",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                334,
+                                470,
+                                348
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        334,
+                                        470,
+                                        348
+                                    ],
+                                    "score": 1.0,
+                                    "content": "RepBM, rejecting most of the pre-collected data in long-term tasks. We present a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                346,
+                                470,
+                                359
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        346,
+                                        470,
+                                        359
+                                    ],
+                                    "score": 1.0,
+                                    "content": "new objective for model learning motivated by recent advances in the estimation",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                142,
+                                357,
+                                469,
+                                369
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        142,
+                                        357,
+                                        469,
+                                        369
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of stationary distribution corrections. This effectively overcomes the aforemen-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                142,
+                                368,
+                                469,
+                                380
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        142,
+                                        368,
+                                        469,
+                                        380
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tioned limitation of RepBM, as well as naturally extending to continuous action",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                378,
+                                470,
+                                393
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        378,
+                                        470,
+                                        393
+                                    ],
+                                    "score": 1.0,
+                                    "content": "spaces and stochastic policies. We also present an offline model-based policy op-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                390,
+                                470,
+                                402
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        390,
+                                        470,
+                                        402
+                                    ],
+                                    "score": 1.0,
+                                    "content": "timization using this new objective, yielding the state-of-the-art performance in a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                401,
+                                339,
+                                412
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        401,
+                                        339,
+                                        412
+                                    ],
+                                    "score": 1.0,
+                                    "content": "representative set of benchmark offline RL tasks.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        }
+                    ],
+                    "index": 15
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        435,
+                        206,
+                        448
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                434,
+                                208,
+                                451
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        434,
+                                        208,
+                                        451
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1 INTRODUCTION",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        }
+                    ],
+                    "index": 23
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        461,
+                        505,
+                        539
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                461,
+                                505,
+                                474
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        461,
+                                        505,
+                                        474
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Reinforcement learning (RL) has accomplished remarkable results in a wide range of domains, but",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                473,
+                                505,
+                                484
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        473,
+                                        505,
+                                        484
+                                    ],
+                                    "score": 1.0,
+                                    "content": "its successes were mostly based on a large number of online interactions with the environment.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                483,
+                                505,
+                                496
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        483,
+                                        505,
+                                        496
+                                    ],
+                                    "score": 1.0,
+                                    "content": "However, in many real-world tasks, exploratory online interactions are either very expensive or",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                495,
+                                505,
+                                507
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        495,
+                                        505,
+                                        507
+                                    ],
+                                    "score": 1.0,
+                                    "content": "dangerous (e.g. robotics, autonomous driving, and healthcare), and applying a standard online RL",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                505,
+                                505,
+                                518
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        505,
+                                        505,
+                                        518
+                                    ],
+                                    "score": 1.0,
+                                    "content": "would be impractical. Consequently, the ability to optimize RL agents reliably without online inter-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                516,
+                                505,
+                                529
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        516,
+                                        505,
+                                        529
+                                    ],
+                                    "score": 1.0,
+                                    "content": "actions has been considered as a key to practical deployment, which is the main goal of batch RL,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                528,
+                                382,
+                                539
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        528,
+                                        382,
+                                        539
+                                    ],
+                                    "score": 1.0,
+                                    "content": "also known as offline RL (Fujimoto et al., 2019; Levine et al., 2020).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        }
+                    ],
+                    "index": 27
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        545,
+                        505,
+                        621
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                544,
+                                505,
+                                557
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        544,
+                                        505,
+                                        557
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In an offline RL algorithm, accurate policy evaluation and reliable policy improvement are both",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                554,
+                                505,
+                                568
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        554,
+                                        505,
+                                        568
+                                    ],
+                                    "score": 1.0,
+                                    "content": "crucial for the successful training of the agent. Evaluating policies in offline RL is essentially an",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                566,
+                                505,
+                                578
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        566,
+                                        505,
+                                        578
+                                    ],
+                                    "score": 1.0,
+                                    "content": "off-policy evaluation (OPE) task, which aims to evaluate the target policy given the dataset collected",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                577,
+                                506,
+                                590
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        577,
+                                        506,
+                                        590
+                                    ],
+                                    "score": 1.0,
+                                    "content": "from the behavior policy. The difference between the target and the behavior policies causes a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                587,
+                                505,
+                                601
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        587,
+                                        505,
+                                        601
+                                    ],
+                                    "score": 1.0,
+                                    "content": "distribution shift in the estimation, which needs to be adequately addressed for accurate policy",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                599,
+                                506,
+                                612
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        599,
+                                        506,
+                                        612
+                                    ],
+                                    "score": 1.0,
+                                    "content": "evaluation. OPE itself is one of the long-standing hard problems in RL (Sutton et al., 1998; 2009;",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                610,
+                                322,
+                                622
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        610,
+                                        322,
+                                        622
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Thomas & Brunskill, 2016; Hallak & Mannor, 2017).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        }
+                    ],
+                    "index": 34
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        627,
+                        504,
+                        704
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                627,
+                                505,
+                                639
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        627,
+                                        505,
+                                        639
+                                    ],
+                                    "score": 1.0,
+                                    "content": "However, recent offline RL studies mainly focus on how to improve the policy conservatively while",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                639,
+                                505,
+                                650
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        639,
+                                        505,
+                                        650
+                                    ],
+                                    "score": 1.0,
+                                    "content": "using a common policy evaluation technique without much considerations for the distribution shift,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                649,
+                                506,
+                                662
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        649,
+                                        506,
+                                        662
+                                    ],
+                                    "score": 1.0,
+                                    "content": "e.g. mean squared temporal difference error minimization or maximum-likelihood training of en-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                660,
+                                506,
+                                672
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        660,
+                                        506,
+                                        672
+                                    ],
+                                    "score": 1.0,
+                                    "content": "vironment model (Fujimoto et al., 2019; Kumar et al., 2019; Yu et al., 2020). While conservative",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                671,
+                                506,
+                                683
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        671,
+                                        506,
+                                        683
+                                    ],
+                                    "score": 1.0,
+                                    "content": "policy improvement helps the policy evaluation by reducing the off-policyness, we hypothesize that",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                682,
+                                506,
+                                694
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        682,
+                                        506,
+                                        694
+                                    ],
+                                    "score": 1.0,
+                                    "content": "addressing the distribution shift explicitly during the policy evaluation can further improve the over-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                693,
+                                431,
+                                705
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        693,
+                                        431,
+                                        705
+                                    ],
+                                    "score": 1.0,
+                                    "content": "all performance, since it can provide a better foundation for policy improvement.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        }
+                    ],
+                    "index": 41
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        709,
+                        503,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                709,
+                                505,
+                                722
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        709,
+                                        505,
+                                        722
+                                    ],
+                                    "score": 1.0,
+                                    "content": "To this end, we aim to explicitly address the distribution shift of the OPE estimator used in the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                720,
+                                505,
+                                733
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        720,
+                                        505,
+                                        733
+                                    ],
+                                    "score": 1.0,
+                                    "content": "offline RL algorithm. In particular, we focus on the model-based approach, where we train an",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        }
+                    ],
+                    "index": 45.5
+                }
+            ],
+            "page_idx": 0,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        107,
+                        27,
+                        292,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                25,
+                                293,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        25,
+                                        293,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Published as a conference paper at ICLR 2021",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        303,
+                        751,
+                        308,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                302,
+                                751,
+                                308,
+                                761
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        302,
+                                        751,
+                                        308,
+                                        761
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        79,
+                        504,
+                        116
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                78,
+                                335,
+                                97
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        78,
+                                        335,
+                                        97
+                                    ],
+                                    "score": 1.0,
+                                    "content": "REPRESENTATION BALANCING",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                107,
+                                99,
+                                505,
+                                117
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        99,
+                                        505,
+                                        117
+                                    ],
+                                    "score": 1.0,
+                                    "content": "OFFLINE MODEL-BASED REINFORCEMENT LEARNING",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        }
+                    ],
+                    "index": 0.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        114,
+                        133,
+                        344,
+                        146
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                113,
+                                133,
+                                344,
+                                149
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        113,
+                                        133,
+                                        312,
+                                        149
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Byung-Jun Lee1,3, Jongmin Lee1 & Kee-Eung",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        313,
+                                        134,
+                                        344,
+                                        146
+                                    ],
+                                    "score": 0.49,
+                                    "content": "\\mathbf { K i m } ^ { 1 , 2 }",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 2
+                        }
+                    ],
+                    "index": 2,
+                    "bbox_fs": [
+                        113,
+                        133,
+                        344,
+                        149
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        112,
+                        147,
+                        386,
+                        192
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                112,
+                                145,
+                                355,
+                                160
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        112,
+                                        145,
+                                        355,
+                                        160
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1School of Computing, KAIST, Daejeon, Republic of Korea",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                110,
+                                156,
+                                360,
+                                172
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        110,
+                                        156,
+                                        360,
+                                        172
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2Graduate School of AI, KAIST, Daejeon, Republic of Korea",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                111,
+                                168,
+                                290,
+                                183
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        111,
+                                        168,
+                                        290,
+                                        183
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3Gauss Labs Inc., Seoul, Republic of Korea",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                113,
+                                181,
+                                388,
+                                195
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        113,
+                                        181,
+                                        388,
+                                        195
+                                    ],
+                                    "score": 1.0,
+                                    "content": "{bjlee,jmlee}@ai.kaist.ac.kr, kekim@kaist.ac.k",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        }
+                    ],
+                    "index": 4.5,
+                    "bbox_fs": [
+                        110,
+                        145,
+                        388,
+                        195
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        278,
+                        221,
+                        333,
+                        233
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                276,
+                                220,
+                                336,
+                                235
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        276,
+                                        220,
+                                        336,
+                                        235
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ABSTRACT",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        }
+                    ],
+                    "index": 7
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        143,
+                        247,
+                        468,
+                        411
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                142,
+                                246,
+                                469,
+                                260
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        142,
+                                        246,
+                                        469,
+                                        260
+                                    ],
+                                    "score": 1.0,
+                                    "content": "One of the main challenges in offline and off-policy reinforcement learning is to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                258,
+                                470,
+                                271
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        258,
+                                        470,
+                                        271
+                                    ],
+                                    "score": 1.0,
+                                    "content": "cope with the distribution shift that arises from the mismatch between the target",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                269,
+                                470,
+                                282
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        269,
+                                        470,
+                                        282
+                                    ],
+                                    "score": 1.0,
+                                    "content": "policy and the data collection policy. In this paper, we focus on a model-based",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                281,
+                                470,
+                                292
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        281,
+                                        470,
+                                        292
+                                    ],
+                                    "score": 1.0,
+                                    "content": "approach, particularly on learning the representation for a robust model of the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                291,
+                                470,
+                                304
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        291,
+                                        470,
+                                        304
+                                    ],
+                                    "score": 1.0,
+                                    "content": "environment under the distribution shift, which has been first studied by Repre-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                302,
+                                469,
+                                314
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        302,
+                                        469,
+                                        314
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sentation Balancing MDP (RepBM). Although this prior work has shown promis-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                313,
+                                469,
+                                326
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        313,
+                                        469,
+                                        326
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ing results, there are a number of shortcomings that still hinder its applicability",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                324,
+                                469,
+                                336
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        324,
+                                        469,
+                                        336
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to practical tasks. In particular, we address the curse of horizon exhibited by",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                334,
+                                470,
+                                348
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        334,
+                                        470,
+                                        348
+                                    ],
+                                    "score": 1.0,
+                                    "content": "RepBM, rejecting most of the pre-collected data in long-term tasks. We present a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                346,
+                                470,
+                                359
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        346,
+                                        470,
+                                        359
+                                    ],
+                                    "score": 1.0,
+                                    "content": "new objective for model learning motivated by recent advances in the estimation",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                142,
+                                357,
+                                469,
+                                369
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        142,
+                                        357,
+                                        469,
+                                        369
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of stationary distribution corrections. This effectively overcomes the aforemen-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                142,
+                                368,
+                                469,
+                                380
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        142,
+                                        368,
+                                        469,
+                                        380
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tioned limitation of RepBM, as well as naturally extending to continuous action",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                378,
+                                470,
+                                393
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        378,
+                                        470,
+                                        393
+                                    ],
+                                    "score": 1.0,
+                                    "content": "spaces and stochastic policies. We also present an offline model-based policy op-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                390,
+                                470,
+                                402
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        390,
+                                        470,
+                                        402
+                                    ],
+                                    "score": 1.0,
+                                    "content": "timization using this new objective, yielding the state-of-the-art performance in a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                401,
+                                339,
+                                412
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        401,
+                                        339,
+                                        412
+                                    ],
+                                    "score": 1.0,
+                                    "content": "representative set of benchmark offline RL tasks.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        }
+                    ],
+                    "index": 15,
+                    "bbox_fs": [
+                        141,
+                        246,
+                        470,
+                        412
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        435,
+                        206,
+                        448
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                434,
+                                208,
+                                451
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        434,
+                                        208,
+                                        451
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1 INTRODUCTION",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        }
+                    ],
+                    "index": 23
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        461,
+                        505,
+                        539
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                461,
+                                505,
+                                474
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        461,
+                                        505,
+                                        474
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Reinforcement learning (RL) has accomplished remarkable results in a wide range of domains, but",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                473,
+                                505,
+                                484
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        473,
+                                        505,
+                                        484
+                                    ],
+                                    "score": 1.0,
+                                    "content": "its successes were mostly based on a large number of online interactions with the environment.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                483,
+                                505,
+                                496
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        483,
+                                        505,
+                                        496
+                                    ],
+                                    "score": 1.0,
+                                    "content": "However, in many real-world tasks, exploratory online interactions are either very expensive or",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                495,
+                                505,
+                                507
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        495,
+                                        505,
+                                        507
+                                    ],
+                                    "score": 1.0,
+                                    "content": "dangerous (e.g. robotics, autonomous driving, and healthcare), and applying a standard online RL",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                505,
+                                505,
+                                518
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        505,
+                                        505,
+                                        518
+                                    ],
+                                    "score": 1.0,
+                                    "content": "would be impractical. Consequently, the ability to optimize RL agents reliably without online inter-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                516,
+                                505,
+                                529
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        516,
+                                        505,
+                                        529
+                                    ],
+                                    "score": 1.0,
+                                    "content": "actions has been considered as a key to practical deployment, which is the main goal of batch RL,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                528,
+                                382,
+                                539
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        528,
+                                        382,
+                                        539
+                                    ],
+                                    "score": 1.0,
+                                    "content": "also known as offline RL (Fujimoto et al., 2019; Levine et al., 2020).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        }
+                    ],
+                    "index": 27,
+                    "bbox_fs": [
+                        105,
+                        461,
+                        505,
+                        539
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        545,
+                        505,
+                        621
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                544,
+                                505,
+                                557
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        544,
+                                        505,
+                                        557
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In an offline RL algorithm, accurate policy evaluation and reliable policy improvement are both",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                554,
+                                505,
+                                568
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        554,
+                                        505,
+                                        568
+                                    ],
+                                    "score": 1.0,
+                                    "content": "crucial for the successful training of the agent. Evaluating policies in offline RL is essentially an",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                566,
+                                505,
+                                578
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        566,
+                                        505,
+                                        578
+                                    ],
+                                    "score": 1.0,
+                                    "content": "off-policy evaluation (OPE) task, which aims to evaluate the target policy given the dataset collected",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                577,
+                                506,
+                                590
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        577,
+                                        506,
+                                        590
+                                    ],
+                                    "score": 1.0,
+                                    "content": "from the behavior policy. The difference between the target and the behavior policies causes a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                587,
+                                505,
+                                601
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        587,
+                                        505,
+                                        601
+                                    ],
+                                    "score": 1.0,
+                                    "content": "distribution shift in the estimation, which needs to be adequately addressed for accurate policy",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                599,
+                                506,
+                                612
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        599,
+                                        506,
+                                        612
+                                    ],
+                                    "score": 1.0,
+                                    "content": "evaluation. OPE itself is one of the long-standing hard problems in RL (Sutton et al., 1998; 2009;",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                610,
+                                322,
+                                622
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        610,
+                                        322,
+                                        622
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Thomas & Brunskill, 2016; Hallak & Mannor, 2017).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        }
+                    ],
+                    "index": 34,
+                    "bbox_fs": [
+                        105,
+                        544,
+                        506,
+                        622
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        627,
+                        504,
+                        704
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                627,
+                                505,
+                                639
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        627,
+                                        505,
+                                        639
+                                    ],
+                                    "score": 1.0,
+                                    "content": "However, recent offline RL studies mainly focus on how to improve the policy conservatively while",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                639,
+                                505,
+                                650
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        639,
+                                        505,
+                                        650
+                                    ],
+                                    "score": 1.0,
+                                    "content": "using a common policy evaluation technique without much considerations for the distribution shift,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                649,
+                                506,
+                                662
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        649,
+                                        506,
+                                        662
+                                    ],
+                                    "score": 1.0,
+                                    "content": "e.g. mean squared temporal difference error minimization or maximum-likelihood training of en-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                660,
+                                506,
+                                672
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        660,
+                                        506,
+                                        672
+                                    ],
+                                    "score": 1.0,
+                                    "content": "vironment model (Fujimoto et al., 2019; Kumar et al., 2019; Yu et al., 2020). While conservative",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                671,
+                                506,
+                                683
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        671,
+                                        506,
+                                        683
+                                    ],
+                                    "score": 1.0,
+                                    "content": "policy improvement helps the policy evaluation by reducing the off-policyness, we hypothesize that",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                682,
+                                506,
+                                694
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        682,
+                                        506,
+                                        694
+                                    ],
+                                    "score": 1.0,
+                                    "content": "addressing the distribution shift explicitly during the policy evaluation can further improve the over-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                693,
+                                431,
+                                705
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        693,
+                                        431,
+                                        705
+                                    ],
+                                    "score": 1.0,
+                                    "content": "all performance, since it can provide a better foundation for policy improvement.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        }
+                    ],
+                    "index": 41,
+                    "bbox_fs": [
+                        104,
+                        627,
+                        506,
+                        705
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        709,
+                        503,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                709,
+                                505,
+                                722
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        709,
+                                        505,
+                                        722
+                                    ],
+                                    "score": 1.0,
+                                    "content": "To this end, we aim to explicitly address the distribution shift of the OPE estimator used in the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                720,
+                                505,
+                                733
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        720,
+                                        505,
+                                        733
+                                    ],
+                                    "score": 1.0,
+                                    "content": "offline RL algorithm. In particular, we focus on the model-based approach, where we train an",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                82,
+                                505,
+                                95
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        82,
+                                        505,
+                                        95
+                                    ],
+                                    "score": 1.0,
+                                    "content": "environment model robust to the distribution shift. One of the notable prior works is Representation",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                94,
+                                505,
+                                106
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        94,
+                                        505,
+                                        106
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Balancing MDP (RepBM) (Liu et al., 2018b), which regularizes the representation learning of the",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                104,
+                                505,
+                                117
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        104,
+                                        505,
+                                        117
+                                    ],
+                                    "score": 1.0,
+                                    "content": "model to be invariant between the distributions. However, despite the promising results by RepBM,",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                116,
+                                505,
+                                127
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        116,
+                                        505,
+                                        127
+                                    ],
+                                    "score": 1.0,
+                                    "content": "its step-wise estimation of the distance between the distributions has a few drawbacks that limit the",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                126,
+                                505,
+                                139
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        126,
+                                        505,
+                                        139
+                                    ],
+                                    "score": 1.0,
+                                    "content": "algorithm from being practical: not only it assumes a discrete-action task where the target policy is",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                137,
+                                505,
+                                150
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        137,
+                                        505,
+                                        150
+                                    ],
+                                    "score": 1.0,
+                                    "content": "deterministic, but it also performs poorly in long-term tasks due to the curse of horizon of step-wise",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                149,
+                                330,
+                                161
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        149,
+                                        330,
+                                        161
+                                    ],
+                                    "score": 1.0,
+                                    "content": "importance sampling (IS) estimators (Liu et al., 2018a).",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 6
+                        }
+                    ],
+                    "index": 45.5,
+                    "bbox_fs": [
+                        105,
+                        709,
+                        505,
+                        733
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        82,
+                        505,
+                        160
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                82,
+                                505,
+                                95
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        82,
+                                        505,
+                                        95
+                                    ],
+                                    "score": 1.0,
+                                    "content": "environment model robust to the distribution shift. One of the notable prior works is Representation",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                94,
+                                505,
+                                106
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        94,
+                                        505,
+                                        106
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Balancing MDP (RepBM) (Liu et al., 2018b), which regularizes the representation learning of the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                104,
+                                505,
+                                117
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        104,
+                                        505,
+                                        117
+                                    ],
+                                    "score": 1.0,
+                                    "content": "model to be invariant between the distributions. However, despite the promising results by RepBM,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                116,
+                                505,
+                                127
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        116,
+                                        505,
+                                        127
+                                    ],
+                                    "score": 1.0,
+                                    "content": "its step-wise estimation of the distance between the distributions has a few drawbacks that limit the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                126,
+                                505,
+                                139
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        126,
+                                        505,
+                                        139
+                                    ],
+                                    "score": 1.0,
+                                    "content": "algorithm from being practical: not only it assumes a discrete-action task where the target policy is",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                137,
+                                505,
+                                150
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        137,
+                                        505,
+                                        150
+                                    ],
+                                    "score": 1.0,
+                                    "content": "deterministic, but it also performs poorly in long-term tasks due to the curse of horizon of step-wise",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                149,
+                                330,
+                                161
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        149,
+                                        330,
+                                        161
+                                    ],
+                                    "score": 1.0,
+                                    "content": "importance sampling (IS) estimators (Liu et al., 2018a).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        }
+                    ],
+                    "index": 3
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        165,
+                        505,
+                        286
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                164,
+                                505,
+                                178
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        164,
+                                        505,
+                                        178
+                                    ],
+                                    "score": 1.0,
+                                    "content": "To address these limitations, we present the Representation Balancing with Stationary Distribution",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                176,
+                                505,
+                                189
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        176,
+                                        505,
+                                        189
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Estimation (RepB-SDE) framework, where we aim to learn a balanced representation by regular-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                187,
+                                506,
+                                199
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        187,
+                                        506,
+                                        199
+                                    ],
+                                    "score": 1.0,
+                                    "content": "izing, in the representation space, the distance between the data distribution and the discounted",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                198,
+                                505,
+                                211
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        198,
+                                        505,
+                                        211
+                                    ],
+                                    "score": 1.0,
+                                    "content": "stationary distribution induced by the target policy. Motivated by the recent advances in estimating",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                209,
+                                506,
+                                222
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        209,
+                                        506,
+                                        222
+                                    ],
+                                    "score": 1.0,
+                                    "content": "stationary distribution corrections, we present a new representation balancing objective to train a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                220,
+                                505,
+                                233
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        220,
+                                        505,
+                                        233
+                                    ],
+                                    "score": 1.0,
+                                    "content": "model of the environment that no longer suffers from the curse of horizon. We empirically show",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                231,
+                                505,
+                                243
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        231,
+                                        505,
+                                        243
+                                    ],
+                                    "score": 1.0,
+                                    "content": "that the model trained by the RepB-SDE objective is robust to the distribution shift for the OPE task,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                242,
+                                506,
+                                255
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        242,
+                                        506,
+                                        255
+                                    ],
+                                    "score": 1.0,
+                                    "content": "particularly when the difference between the target and the behavior is large. We also introduce a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                252,
+                                505,
+                                265
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        252,
+                                        505,
+                                        265
+                                    ],
+                                    "score": 1.0,
+                                    "content": "model-based offline RL algorithm based on the RepB-SDE framework and report its performance on",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                263,
+                                505,
+                                277
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        263,
+                                        505,
+                                        277
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the D4RL benchmark (Fu et al., 2020), showing the state-of-the-art performance in a representative",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                276,
+                                155,
+                                286
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        276,
+                                        155,
+                                        286
+                                    ],
+                                    "score": 1.0,
+                                    "content": "set of tasks.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        }
+                    ],
+                    "index": 12
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        305,
+                        209,
+                        318
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                304,
+                                210,
+                                320
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        304,
+                                        210,
+                                        320
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2 RELATED WORK",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        }
+                    ],
+                    "index": 18
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        332,
+                        505,
+                        420
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                331,
+                                505,
+                                345
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        331,
+                                        505,
+                                        345
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Learning balanced representation Learning a representation invariant to specific aspects of data",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                343,
+                                504,
+                                356
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        343,
+                                        504,
+                                        356
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is an established method for overcoming distribution shift that arises in unsupervised domain adap-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                354,
+                                505,
+                                366
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        354,
+                                        505,
+                                        366
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tation (Ben-David et al., 2007; Zemel et al., 2013) and in causal inference from observational",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                365,
+                                505,
+                                377
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        365,
+                                        505,
+                                        377
+                                    ],
+                                    "score": 1.0,
+                                    "content": "data (Shalit et al., 2017; Johansson et al., 2018). They have shown that imposing a bound on the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                376,
+                                505,
+                                389
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        376,
+                                        505,
+                                        389
+                                    ],
+                                    "score": 1.0,
+                                    "content": "generalization error under the distribution shift leads to the objective that learns a balanced repre-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                387,
+                                505,
+                                399
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        387,
+                                        505,
+                                        399
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sentation such that the training and the test distributions look similar. RepBM (Liu et al., 2018b)",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                399,
+                                505,
+                                411
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        399,
+                                        505,
+                                        411
+                                    ],
+                                    "score": 1.0,
+                                    "content": "can be seen as a direct extension to the sequential case, which encourages the representation to be",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                408,
+                                366,
+                                422
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        408,
+                                        366,
+                                        422
+                                    ],
+                                    "score": 1.0,
+                                    "content": "invariant under the target and behavior policies in each timestep.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        }
+                    ],
+                    "index": 22.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        435,
+                        505,
+                        557
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                434,
+                                505,
+                                448
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        434,
+                                        505,
+                                        448
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Stationary distribution correction estimation (DICE) Step-wise importance sampling (IS) es-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                446,
+                                506,
+                                460
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        446,
+                                        506,
+                                        460
+                                    ],
+                                    "score": 1.0,
+                                    "content": "timators (Precup, 2000) compute importance weights by taking the product of per-step distribution",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                457,
+                                505,
+                                471
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        457,
+                                        505,
+                                        471
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ratios. Consequently, these methods suffer from exponentially high variance in the lengths of trajec-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                469,
+                                506,
+                                482
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        469,
+                                        506,
+                                        482
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tories, which is a phenomenon called the curse of horizon (Liu et al., 2018a). Recently, techniques of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                479,
+                                506,
+                                493
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        479,
+                                        506,
+                                        493
+                                    ],
+                                    "score": 1.0,
+                                    "content": "computing a stationary DIstribution Correction Estimation (DICE) have made remarkable progress",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                491,
+                                505,
+                                504
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        491,
+                                        428,
+                                        504
+                                    ],
+                                    "score": 1.0,
+                                    "content": "that effectively addresses the curse of horizon (Liu et al., 2018a; Nachum et al.,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        428,
+                                        491,
+                                        454,
+                                        501
+                                    ],
+                                    "score": 0.71,
+                                    "content": "2 0 1 9 \\mathrm { a }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        455,
+                                        491,
+                                        505,
+                                        504
+                                    ],
+                                    "score": 1.0,
+                                    "content": "; Tang et al.,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                501,
+                                505,
+                                514
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        501,
+                                        505,
+                                        514
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2020; Zhang et al., 2020; Mousavi et al., 2020). DICE has been also used to explicitly address the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                512,
+                                505,
+                                525
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        512,
+                                        505,
+                                        525
+                                    ],
+                                    "score": 1.0,
+                                    "content": "distribution shift in online model-free RL, by directly applying IS on the policy and action-value",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                522,
+                                505,
+                                537
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        522,
+                                        505,
+                                        537
+                                    ],
+                                    "score": 1.0,
+                                    "content": "objectives (Liu et al., 2019; Gelada & Bellemare, 2019). We adopt one of the estimation techniques,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                534,
+                                505,
+                                547
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        534,
+                                        505,
+                                        547
+                                    ],
+                                    "score": 1.0,
+                                    "content": "DualDICE (Nachum et al., 2019a), to measure the distance between the stationary distribution and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                545,
+                                299,
+                                559
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        545,
+                                        299,
+                                        559
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the data distribution in the representation space.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        }
+                    ],
+                    "index": 32
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        572,
+                        505,
+                        671
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                572,
+                                504,
+                                585
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        572,
+                                        504,
+                                        585
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Offline reinforcement learning There are extensive studies on improving standard online model-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                581,
+                                505,
+                                597
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        581,
+                                        505,
+                                        597
+                                    ],
+                                    "score": 1.0,
+                                    "content": "free RL algorithms (Mnih et al., 2015; Lillicrap et al., 2016; Haarnoja et al., 2018) for stable learning",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                593,
+                                505,
+                                607
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        593,
+                                        505,
+                                        607
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in the offline setting. The main idea behind them is to conservatively improve policy by (1) quanti-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                604,
+                                505,
+                                618
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        604,
+                                        505,
+                                        618
+                                    ],
+                                    "score": 1.0,
+                                    "content": "fying the uncertainty of value function estimate, e.g. using bootstrapped ensembles (Kumar et al.,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                616,
+                                506,
+                                629
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        616,
+                                        506,
+                                        629
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2019; Agarwal et al., 2020), or/and (2) constraining the optimized target policy to be close to the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                625,
+                                506,
+                                640
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        625,
+                                        506,
+                                        640
+                                    ],
+                                    "score": 1.0,
+                                    "content": "behavior policy (i.e. behavior regularization approaches) (Fujimoto et al., 2019; Kumar et al., 2019;",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                636,
+                                506,
+                                651
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        636,
+                                        506,
+                                        651
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Wu et al., 2019; Lee et al., 2020). A notable exception is AlgaeDICE (Nachum et al., 2019b), which",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                647,
+                                505,
+                                663
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        647,
+                                        505,
+                                        663
+                                    ],
+                                    "score": 1.0,
+                                    "content": "implicitly uses DICE to regularize the discounted stationary distribution induced by the target policy",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                660,
+                                336,
+                                673
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        660,
+                                        336,
+                                        673
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to be kept inside of the data support, similar to this work.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        }
+                    ],
+                    "index": 42
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        677,
+                        504,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                676,
+                                505,
+                                689
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        676,
+                                        505,
+                                        689
+                                    ],
+                                    "score": 1.0,
+                                    "content": "On the other hand, Yu et al. (2020) argued that the model-based approach can be advantageous",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 47
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                688,
+                                505,
+                                700
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        688,
+                                        505,
+                                        700
+                                    ],
+                                    "score": 1.0,
+                                    "content": "due to its ability to generalize predictions on the states outside of the data support. They introduce",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 48
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                698,
+                                505,
+                                712
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        698,
+                                        505,
+                                        712
+                                    ],
+                                    "score": 1.0,
+                                    "content": "MOPO (Yu et al., 2020), which uses truncated rollouts and penalized rewards for conservative policy",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 49
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                709,
+                                505,
+                                723
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        709,
+                                        505,
+                                        723
+                                    ],
+                                    "score": 1.0,
+                                    "content": "improvement. MOReL (Kidambi et al., 2020) trains a state-action novelty detector and use it to pe-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 50
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                720,
+                                505,
+                                733
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        720,
+                                        505,
+                                        733
+                                    ],
+                                    "score": 1.0,
+                                    "content": "nalize rewards in the data-sparse region. Matsushima et al. (2020), MOOSE (Swazinna et al., 2020)",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 51
+                        }
+                    ],
+                    "index": 49
+                }
+            ],
+            "page_idx": 1,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        108,
+                        27,
+                        292,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                26,
+                                293,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        26,
+                                        293,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Published as a conference paper at ICLR 2021",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        302,
+                        751,
+                        309,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                301,
+                                750,
+                                310,
+                                763
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        301,
+                                        750,
+                                        310,
+                                        763
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        82,
+                        505,
+                        160
+                    ],
+                    "lines": [],
+                    "index": 3,
+                    "bbox_fs": [
+                        105,
+                        82,
+                        505,
+                        161
+                    ],
+                    "lines_deleted": true
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        165,
+                        505,
+                        286
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                164,
+                                505,
+                                178
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        164,
+                                        505,
+                                        178
+                                    ],
+                                    "score": 1.0,
+                                    "content": "To address these limitations, we present the Representation Balancing with Stationary Distribution",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                176,
+                                505,
+                                189
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        176,
+                                        505,
+                                        189
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Estimation (RepB-SDE) framework, where we aim to learn a balanced representation by regular-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                187,
+                                506,
+                                199
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        187,
+                                        506,
+                                        199
+                                    ],
+                                    "score": 1.0,
+                                    "content": "izing, in the representation space, the distance between the data distribution and the discounted",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                198,
+                                505,
+                                211
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        198,
+                                        505,
+                                        211
+                                    ],
+                                    "score": 1.0,
+                                    "content": "stationary distribution induced by the target policy. Motivated by the recent advances in estimating",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                209,
+                                506,
+                                222
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        209,
+                                        506,
+                                        222
+                                    ],
+                                    "score": 1.0,
+                                    "content": "stationary distribution corrections, we present a new representation balancing objective to train a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                220,
+                                505,
+                                233
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        220,
+                                        505,
+                                        233
+                                    ],
+                                    "score": 1.0,
+                                    "content": "model of the environment that no longer suffers from the curse of horizon. We empirically show",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                231,
+                                505,
+                                243
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        231,
+                                        505,
+                                        243
+                                    ],
+                                    "score": 1.0,
+                                    "content": "that the model trained by the RepB-SDE objective is robust to the distribution shift for the OPE task,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                242,
+                                506,
+                                255
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        242,
+                                        506,
+                                        255
+                                    ],
+                                    "score": 1.0,
+                                    "content": "particularly when the difference between the target and the behavior is large. We also introduce a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                252,
+                                505,
+                                265
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        252,
+                                        505,
+                                        265
+                                    ],
+                                    "score": 1.0,
+                                    "content": "model-based offline RL algorithm based on the RepB-SDE framework and report its performance on",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                263,
+                                505,
+                                277
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        263,
+                                        505,
+                                        277
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the D4RL benchmark (Fu et al., 2020), showing the state-of-the-art performance in a representative",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                276,
+                                155,
+                                286
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        276,
+                                        155,
+                                        286
+                                    ],
+                                    "score": 1.0,
+                                    "content": "set of tasks.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        }
+                    ],
+                    "index": 12,
+                    "bbox_fs": [
+                        105,
+                        164,
+                        506,
+                        286
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        305,
+                        209,
+                        318
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                304,
+                                210,
+                                320
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        304,
+                                        210,
+                                        320
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2 RELATED WORK",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        }
+                    ],
+                    "index": 18
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        332,
+                        505,
+                        420
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                331,
+                                505,
+                                345
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        331,
+                                        505,
+                                        345
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Learning balanced representation Learning a representation invariant to specific aspects of data",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                343,
+                                504,
+                                356
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        343,
+                                        504,
+                                        356
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is an established method for overcoming distribution shift that arises in unsupervised domain adap-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                354,
+                                505,
+                                366
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        354,
+                                        505,
+                                        366
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tation (Ben-David et al., 2007; Zemel et al., 2013) and in causal inference from observational",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                365,
+                                505,
+                                377
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        365,
+                                        505,
+                                        377
+                                    ],
+                                    "score": 1.0,
+                                    "content": "data (Shalit et al., 2017; Johansson et al., 2018). They have shown that imposing a bound on the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                376,
+                                505,
+                                389
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        376,
+                                        505,
+                                        389
+                                    ],
+                                    "score": 1.0,
+                                    "content": "generalization error under the distribution shift leads to the objective that learns a balanced repre-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                387,
+                                505,
+                                399
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        387,
+                                        505,
+                                        399
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sentation such that the training and the test distributions look similar. RepBM (Liu et al., 2018b)",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                399,
+                                505,
+                                411
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        399,
+                                        505,
+                                        411
+                                    ],
+                                    "score": 1.0,
+                                    "content": "can be seen as a direct extension to the sequential case, which encourages the representation to be",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                408,
+                                366,
+                                422
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        408,
+                                        366,
+                                        422
+                                    ],
+                                    "score": 1.0,
+                                    "content": "invariant under the target and behavior policies in each timestep.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        }
+                    ],
+                    "index": 22.5,
+                    "bbox_fs": [
+                        105,
+                        331,
+                        505,
+                        422
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        435,
+                        505,
+                        557
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                434,
+                                505,
+                                448
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        434,
+                                        505,
+                                        448
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Stationary distribution correction estimation (DICE) Step-wise importance sampling (IS) es-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                446,
+                                506,
+                                460
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        446,
+                                        506,
+                                        460
+                                    ],
+                                    "score": 1.0,
+                                    "content": "timators (Precup, 2000) compute importance weights by taking the product of per-step distribution",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                457,
+                                505,
+                                471
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        457,
+                                        505,
+                                        471
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ratios. Consequently, these methods suffer from exponentially high variance in the lengths of trajec-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                469,
+                                506,
+                                482
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        469,
+                                        506,
+                                        482
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tories, which is a phenomenon called the curse of horizon (Liu et al., 2018a). Recently, techniques of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                479,
+                                506,
+                                493
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        479,
+                                        506,
+                                        493
+                                    ],
+                                    "score": 1.0,
+                                    "content": "computing a stationary DIstribution Correction Estimation (DICE) have made remarkable progress",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                491,
+                                505,
+                                504
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        491,
+                                        428,
+                                        504
+                                    ],
+                                    "score": 1.0,
+                                    "content": "that effectively addresses the curse of horizon (Liu et al., 2018a; Nachum et al.,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        428,
+                                        491,
+                                        454,
+                                        501
+                                    ],
+                                    "score": 0.71,
+                                    "content": "2 0 1 9 \\mathrm { a }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        455,
+                                        491,
+                                        505,
+                                        504
+                                    ],
+                                    "score": 1.0,
+                                    "content": "; Tang et al.,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                501,
+                                505,
+                                514
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        501,
+                                        505,
+                                        514
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2020; Zhang et al., 2020; Mousavi et al., 2020). DICE has been also used to explicitly address the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                512,
+                                505,
+                                525
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        512,
+                                        505,
+                                        525
+                                    ],
+                                    "score": 1.0,
+                                    "content": "distribution shift in online model-free RL, by directly applying IS on the policy and action-value",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                522,
+                                505,
+                                537
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        522,
+                                        505,
+                                        537
+                                    ],
+                                    "score": 1.0,
+                                    "content": "objectives (Liu et al., 2019; Gelada & Bellemare, 2019). We adopt one of the estimation techniques,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                534,
+                                505,
+                                547
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        534,
+                                        505,
+                                        547
+                                    ],
+                                    "score": 1.0,
+                                    "content": "DualDICE (Nachum et al., 2019a), to measure the distance between the stationary distribution and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                545,
+                                299,
+                                559
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        545,
+                                        299,
+                                        559
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the data distribution in the representation space.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        }
+                    ],
+                    "index": 32,
+                    "bbox_fs": [
+                        105,
+                        434,
+                        506,
+                        559
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        572,
+                        505,
+                        671
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                572,
+                                504,
+                                585
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        572,
+                                        504,
+                                        585
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Offline reinforcement learning There are extensive studies on improving standard online model-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                581,
+                                505,
+                                597
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        581,
+                                        505,
+                                        597
+                                    ],
+                                    "score": 1.0,
+                                    "content": "free RL algorithms (Mnih et al., 2015; Lillicrap et al., 2016; Haarnoja et al., 2018) for stable learning",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                593,
+                                505,
+                                607
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        593,
+                                        505,
+                                        607
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in the offline setting. The main idea behind them is to conservatively improve policy by (1) quanti-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                604,
+                                505,
+                                618
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        604,
+                                        505,
+                                        618
+                                    ],
+                                    "score": 1.0,
+                                    "content": "fying the uncertainty of value function estimate, e.g. using bootstrapped ensembles (Kumar et al.,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                616,
+                                506,
+                                629
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        616,
+                                        506,
+                                        629
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2019; Agarwal et al., 2020), or/and (2) constraining the optimized target policy to be close to the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                625,
+                                506,
+                                640
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        625,
+                                        506,
+                                        640
+                                    ],
+                                    "score": 1.0,
+                                    "content": "behavior policy (i.e. behavior regularization approaches) (Fujimoto et al., 2019; Kumar et al., 2019;",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                636,
+                                506,
+                                651
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        636,
+                                        506,
+                                        651
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Wu et al., 2019; Lee et al., 2020). A notable exception is AlgaeDICE (Nachum et al., 2019b), which",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                647,
+                                505,
+                                663
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        647,
+                                        505,
+                                        663
+                                    ],
+                                    "score": 1.0,
+                                    "content": "implicitly uses DICE to regularize the discounted stationary distribution induced by the target policy",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                660,
+                                336,
+                                673
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        660,
+                                        336,
+                                        673
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to be kept inside of the data support, similar to this work.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        }
+                    ],
+                    "index": 42,
+                    "bbox_fs": [
+                        104,
+                        572,
+                        506,
+                        673
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        677,
+                        504,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                676,
+                                505,
+                                689
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        676,
+                                        505,
+                                        689
+                                    ],
+                                    "score": 1.0,
+                                    "content": "On the other hand, Yu et al. (2020) argued that the model-based approach can be advantageous",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 47
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                688,
+                                505,
+                                700
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        688,
+                                        505,
+                                        700
+                                    ],
+                                    "score": 1.0,
+                                    "content": "due to its ability to generalize predictions on the states outside of the data support. They introduce",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 48
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                698,
+                                505,
+                                712
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        698,
+                                        505,
+                                        712
+                                    ],
+                                    "score": 1.0,
+                                    "content": "MOPO (Yu et al., 2020), which uses truncated rollouts and penalized rewards for conservative policy",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 49
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                709,
+                                505,
+                                723
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        709,
+                                        505,
+                                        723
+                                    ],
+                                    "score": 1.0,
+                                    "content": "improvement. MOReL (Kidambi et al., 2020) trains a state-action novelty detector and use it to pe-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 50
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                720,
+                                505,
+                                733
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        720,
+                                        505,
+                                        733
+                                    ],
+                                    "score": 1.0,
+                                    "content": "nalize rewards in the data-sparse region. Matsushima et al. (2020), MOOSE (Swazinna et al., 2020)",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 51
+                        }
+                    ],
+                    "index": 49,
+                    "bbox_fs": [
+                        105,
+                        676,
+                        505,
+                        733
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        82,
+                        504,
+                        105
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                81,
+                                505,
+                                95
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        81,
+                                        505,
+                                        95
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and MBOP (Argenson & Dulac-Arnold, 2020) guide their policy optimization using the behavior",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                93,
+                                335,
+                                107
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        93,
+                                        335,
+                                        107
+                                    ],
+                                    "score": 1.0,
+                                    "content": "policy, similar to the behavior regularization approaches.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        }
+                    ],
+                    "index": 0.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        110,
+                        505,
+                        166
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                110,
+                                504,
+                                123
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        110,
+                                        504,
+                                        123
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Note that these aforementioned offline RL methods build on the standard approximate dynamic pro-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                121,
+                                505,
+                                134
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        121,
+                                        505,
+                                        134
+                                    ],
+                                    "score": 1.0,
+                                    "content": "gramming algorithm for action-value estimation (model-free) or on a maximum-likelihood environ-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                132,
+                                505,
+                                145
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        132,
+                                        505,
+                                        145
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ment model (model-based), without explicitly addressing the distribution shift in the estimator. In",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                144,
+                                505,
+                                155
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        144,
+                                        505,
+                                        155
+                                    ],
+                                    "score": 1.0,
+                                    "content": "contrast, we augment the objective for model learning to obtain a robust model under the distribution",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                153,
+                                406,
+                                168
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        153,
+                                        406,
+                                        168
+                                    ],
+                                    "score": 1.0,
+                                    "content": "shift, which is the first attempt for offline RL to the best of our knowledge.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        }
+                    ],
+                    "index": 4
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        181,
+                        208,
+                        194
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                181,
+                                209,
+                                196
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        181,
+                                        209,
+                                        196
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3 PRELIMINARIES",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        }
+                    ],
+                    "index": 7
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        205,
+                        506,
+                        309
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                204,
+                                506,
+                                219
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        204,
+                                        352,
+                                        219
+                                    ],
+                                    "score": 1.0,
+                                    "content": "A Markov Decision Process (MDP) is specified by a tuple",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        353,
+                                        206,
+                                        456,
+                                        218
+                                    ],
+                                    "score": 0.91,
+                                    "content": "M = \\langle S , A , T , R , d _ { 0 } , \\gamma \\rangle",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        457,
+                                        204,
+                                        506,
+                                        219
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", consisting",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                217,
+                                506,
+                                229
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        217,
+                                        168,
+                                        229
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of state space",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        168,
+                                        217,
+                                        176,
+                                        227
+                                    ],
+                                    "score": 0.72,
+                                    "content": "s",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        177,
+                                        217,
+                                        237,
+                                        229
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", action space",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        238,
+                                        218,
+                                        247,
+                                        227
+                                    ],
+                                    "score": 0.75,
+                                    "content": "\\mathcal { A }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        247,
+                                        217,
+                                        332,
+                                        229
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", transition function",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        333,
+                                        218,
+                                        431,
+                                        229
+                                    ],
+                                    "score": 0.88,
+                                    "content": "T : \\mathcal { S } \\times \\mathcal { A }  \\Delta ( \\mathcal { S } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        431,
+                                        217,
+                                        506,
+                                        229
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", reward function",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                227,
+                                506,
+                                241
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        228,
+                                        226,
+                                        240
+                                    ],
+                                    "score": 0.89,
+                                    "content": "R : \\mathcal { S } \\times \\overline { { \\mathcal { A } } } \\to \\Delta ( [ 0 , r _ { \\operatorname* { m a x } } ] )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        226,
+                                        227,
+                                        329,
+                                        241
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", initial state distribution",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        329,
+                                        228,
+                                        340,
+                                        239
+                                    ],
+                                    "score": 0.88,
+                                    "content": "d _ { 0 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        340,
+                                        227,
+                                        418,
+                                        241
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", and discount rate",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        419,
+                                        230,
+                                        425,
+                                        240
+                                    ],
+                                    "score": 0.81,
+                                    "content": "\\gamma",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        426,
+                                        227,
+                                        506,
+                                        241
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". In this paper, we",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                238,
+                                505,
+                                251
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        239,
+                                        267,
+                                        251
+                                    ],
+                                    "score": 1.0,
+                                    "content": "mainly focus on continuous state space",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        267,
+                                        238,
+                                        305,
+                                        250
+                                    ],
+                                    "score": 0.93,
+                                    "content": "S \\subseteq \\mathbb { R } ^ { d _ { s } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        306,
+                                        239,
+                                        505,
+                                        251
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and conduct experiments on both discrete action",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                248,
+                                506,
+                                263
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        248,
+                                        136,
+                                        263
+                                    ],
+                                    "score": 1.0,
+                                    "content": "spaces",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        136,
+                                        250,
+                                        209,
+                                        262
+                                    ],
+                                    "score": 0.93,
+                                    "content": "{ \\cal A } = \\{ a _ { 0 } , . . . a _ { n _ { a } } \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        209,
+                                        248,
+                                        333,
+                                        263
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and continuous action spaces",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        333,
+                                        249,
+                                        374,
+                                        261
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\mathcal { A } \\subseteq \\mathbb { R } ^ { d _ { a } ^ { \\star } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        375,
+                                        248,
+                                        434,
+                                        263
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Given MDP",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        434,
+                                        250,
+                                        446,
+                                        260
+                                    ],
+                                    "score": 0.73,
+                                    "content": "M",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        447,
+                                        248,
+                                        493,
+                                        263
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and policy",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        494,
+                                        252,
+                                        501,
+                                        260
+                                    ],
+                                    "score": 0.71,
+                                    "content": "\\pi",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        502,
+                                        248,
+                                        506,
+                                        263
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                260,
+                                506,
+                                273
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        260,
+                                        506,
+                                        273
+                                    ],
+                                    "score": 1.0,
+                                    "content": "which is a (stochastic) mapping from state to action, the trajectory can be generated in the form of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                107,
+                                272,
+                                505,
+                                285
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        274,
+                                        199,
+                                        283
+                                    ],
+                                    "score": 0.87,
+                                    "content": "s _ { 0 } , a _ { 0 } , r _ { 0 } , s _ { 1 } , a _ { 1 } , r _ { 1 } , \\ldots",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        199,
+                                        272,
+                                        230,
+                                        285
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", where",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        231,
+                                        272,
+                                        266,
+                                        283
+                                    ],
+                                    "score": 0.92,
+                                    "content": "s _ { 0 } \\sim d _ { 0 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        266,
+                                        272,
+                                        357,
+                                        285
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and for each timestep",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        358,
+                                        272,
+                                        383,
+                                        283
+                                    ],
+                                    "score": 0.86,
+                                    "content": "t \\geq 0",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        384,
+                                        272,
+                                        388,
+                                        285
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        388,
+                                        272,
+                                        435,
+                                        284
+                                    ],
+                                    "score": 0.83,
+                                    "content": "a _ { t } \\sim \\pi ( s _ { t } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        435,
+                                        272,
+                                        439,
+                                        285
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        440,
+                                        272,
+                                        501,
+                                        284
+                                    ],
+                                    "score": 0.89,
+                                    "content": "\\boldsymbol { r } _ { t } \\sim R ( s _ { t } , a _ { t } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        501,
+                                        272,
+                                        505,
+                                        285
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                282,
+                                506,
+                                296
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        282,
+                                        124,
+                                        296
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        124,
+                                        284,
+                                        194,
+                                        295
+                                    ],
+                                    "score": 0.93,
+                                    "content": "s _ { t + 1 } \\sim T ( s _ { t } , a _ { t } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        194,
+                                        282,
+                                        506,
+                                        296
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". The goal of RL is to optimize or evaluate a policy, based on the normalized",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                292,
+                                356,
+                                311
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        292,
+                                        219,
+                                        311
+                                    ],
+                                    "score": 1.0,
+                                    "content": "expected discounted return:",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        220,
+                                        294,
+                                        350,
+                                        308
+                                    ],
+                                    "score": 0.9,
+                                    "content": "\\begin{array} { r } { R ^ { \\pi } \\triangleq ( 1 - \\gamma ) \\mathbb { E } _ { M , \\pi } \\left[ \\sum _ { t = 0 } ^ { \\infty } \\gamma ^ { t } r _ { t } \\right] } \\end{array}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        351,
+                                        292,
+                                        356,
+                                        311
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        }
+                    ],
+                    "index": 12
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        312,
+                        504,
+                        335
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                311,
+                                505,
+                                325
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        311,
+                                        505,
+                                        325
+                                    ],
+                                    "score": 1.0,
+                                    "content": "A useful and important concept throughout the paper is the discounted stationary distribution, which",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                324,
+                                289,
+                                336
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        324,
+                                        289,
+                                        336
+                                    ],
+                                    "score": 1.0,
+                                    "content": "represents the long-term occupancy of states:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        }
+                    ],
+                    "index": 17.5
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        200,
+                        338,
+                        410,
+                        371
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                200,
+                                338,
+                                410,
+                                371
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        200,
+                                        338,
+                                        410,
+                                        371
+                                    ],
+                                    "score": 0.93,
+                                    "content": "d ^ { \\pi } ( s , a ) \\triangleq ( 1 - \\gamma ) \\sum _ { t = 0 } ^ { \\infty } \\gamma ^ { t } \\operatorname* { P r } ( s _ { t } = s , a _ { t } = a | M , \\pi ) .",
+                                    "type": "interline_equation",
+                                    "image_path": "4e97c5ca6686900ac9b0f9bd4d01b84db14920e7a7632d71f5cbe853c1d7c3e0.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 19.5,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                200,
+                                338,
+                                410,
+                                354.5
+                            ],
+                            "spans": [],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                200,
+                                354.5,
+                                410,
+                                371.0
+                            ],
+                            "spans": [],
+                            "index": 20
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        104,
+                        374,
+                        474,
+                        387
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                371,
+                                472,
+                                389
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        371,
+                                        280,
+                                        389
+                                    ],
+                                    "score": 1.0,
+                                    "content": "From the definition, it can be observed that",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        281,
+                                        375,
+                                        294,
+                                        384
+                                    ],
+                                    "score": 0.88,
+                                    "content": "R ^ { \\pi }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        295,
+                                        371,
+                                        373,
+                                        389
+                                    ],
+                                    "score": 1.0,
+                                    "content": "can be obtained by",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        374,
+                                        374,
+                                        472,
+                                        387
+                                    ],
+                                    "score": 0.92,
+                                    "content": "R ^ { \\pi } = \\mathbb { E } _ { ( s , a ) \\sim d ^ { \\pi } } \\left[ r ( s , a ) \\right]",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 21
+                        }
+                    ],
+                    "index": 21
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        397,
+                        505,
+                        442
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                397,
+                                505,
+                                411
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        397,
+                                        426,
+                                        411
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Offline RL and off-policy evaluation In this paper, we focus on the offline",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        426,
+                                        398,
+                                        440,
+                                        408
+                                    ],
+                                    "score": 0.61,
+                                    "content": "R L",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        441,
+                                        397,
+                                        505,
+                                        411
+                                    ],
+                                    "score": 1.0,
+                                    "content": "problem where",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                406,
+                                507,
+                                423
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        406,
+                                        271,
+                                        423
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the agent can only access a static dataset",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        272,
+                                        408,
+                                        371,
+                                        421
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\mathcal { D } = \\{ ( \\bar { s } _ { i } , a _ { i } , r _ { i } , s _ { i } ^ { \\prime } ) \\} _ { i = 1 } ^ { N }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        371,
+                                        406,
+                                        470,
+                                        423
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for the maximization of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        471,
+                                        409,
+                                        484,
+                                        419
+                                    ],
+                                    "score": 0.86,
+                                    "content": "R ^ { \\pi }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        484,
+                                        406,
+                                        507,
+                                        423
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". We",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                420,
+                                505,
+                                432
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        420,
+                                        505,
+                                        432
+                                    ],
+                                    "score": 1.0,
+                                    "content": "consider a behavior-agnostic setting where we do not have any knowledge of the data collection",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                430,
+                                374,
+                                443
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        431,
+                                        357,
+                                        443
+                                    ],
+                                    "score": 1.0,
+                                    "content": "process. We denote the empirical distribution of the dataset by",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        358,
+                                        430,
+                                        371,
+                                        441
+                                    ],
+                                    "score": 0.89,
+                                    "content": "\\dot { d } ^ { \\mathcal { D } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        371,
+                                        431,
+                                        374,
+                                        443
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        }
+                    ],
+                    "index": 23.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        447,
+                        506,
+                        615
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                447,
+                                505,
+                                460
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        447,
+                                        324,
+                                        460
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Before improving policy, we first aim to better evaluate",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        325,
+                                        448,
+                                        338,
+                                        458
+                                    ],
+                                    "score": 0.88,
+                                    "content": "R ^ { \\pi }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        339,
+                                        447,
+                                        421,
+                                        460
+                                    ],
+                                    "score": 1.0,
+                                    "content": "given a target policy",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        421,
+                                        450,
+                                        429,
+                                        457
+                                    ],
+                                    "score": 0.76,
+                                    "content": "\\pi",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        429,
+                                        447,
+                                        505,
+                                        460
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and a static dataset",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                459,
+                                505,
+                                471
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        459,
+                                        115,
+                                        469
+                                    ],
+                                    "score": 0.75,
+                                    "content": "\\mathcal { D }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        116,
+                                        459,
+                                        505,
+                                        471
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", which corresponds to an off-policy evaluation (OPE) problem. We mainly focus on a model-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                469,
+                                505,
+                                484
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        471,
+                                        412,
+                                        484
+                                    ],
+                                    "score": 1.0,
+                                    "content": "based approach where the algorithm first estimates the unknown dynamics",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        412,
+                                        469,
+                                        433,
+                                        483
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\widehat { T } , \\widehat { R }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        434,
+                                        471,
+                                        505,
+                                        484
+                                    ],
+                                    "score": 1.0,
+                                    "content": "using the dataset",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                482,
+                                506,
+                                497
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        484,
+                                        115,
+                                        495
+                                    ],
+                                    "score": 0.76,
+                                    "content": "\\mathcal { D }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        116,
+                                        483,
+                                        268,
+                                        497
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". This defines an approximate MDP",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        268,
+                                        482,
+                                        371,
+                                        497
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\widehat { M } = \\langle S , A , \\widehat { T } , \\widehat { R } , d _ { 0 } , \\gamma \\rangle",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        372,
+                                        483,
+                                        506,
+                                        497
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", with the approximate expected",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                102,
+                                492,
+                                509,
+                                516
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        102,
+                                        492,
+                                        178,
+                                        516
+                                    ],
+                                    "score": 1.0,
+                                    "content": "discounted return",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        178,
+                                        496,
+                                        307,
+                                        512
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\begin{array} { r } { \\widehat { R } ^ { \\pi } \\triangleq ( 1 - \\gamma ) \\mathbb { E } _ { \\widehat { M } , \\pi } \\left[ \\sum _ { t = 0 } ^ { \\infty } \\gamma ^ { t } r _ { t } \\right] } \\end{array}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        307,
+                                        492,
+                                        366,
+                                        516
+                                    ],
+                                    "score": 1.0,
+                                    "content": "obtained from",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        366,
+                                        496,
+                                        378,
+                                        508
+                                    ],
+                                    "score": 0.87,
+                                    "content": "\\widehat { M }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        378,
+                                        492,
+                                        509,
+                                        516
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". In this paper, we are interested",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                511,
+                                505,
+                                527
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        511,
+                                        190,
+                                        527
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in the MDP estimate",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        190,
+                                        511,
+                                        202,
+                                        524
+                                    ],
+                                    "score": 0.85,
+                                    "content": "\\widehat { M }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        202,
+                                        511,
+                                        448,
+                                        527
+                                    ],
+                                    "score": 1.0,
+                                    "content": "that can effectively reduce the error in the evaluation of policy",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        448,
+                                        516,
+                                        455,
+                                        524
+                                    ],
+                                    "score": 0.57,
+                                    "content": "\\pi",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        456,
+                                        511,
+                                        459,
+                                        527
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        459,
+                                        511,
+                                        501,
+                                        526
+                                    ],
+                                    "score": 0.91,
+                                    "content": "| R ^ { \\pi } - { \\widehat R } ^ { \\pi } |",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        501,
+                                        511,
+                                        505,
+                                        527
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                530,
+                                505,
+                                543
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        530,
+                                        505,
+                                        543
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In order to do so, we need to learn a good representation of a model that results in a small OPE error.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                541,
+                                505,
+                                554
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        541,
+                                        291,
+                                        554
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We assume a bijective representation function",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        291,
+                                        542,
+                                        355,
+                                        553
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\phi : \\mathcal { S } \\times \\mathcal { A }  \\mathcal { Z }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        356,
+                                        541,
+                                        383,
+                                        554
+                                    ],
+                                    "score": 1.0,
+                                    "content": "where",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        383,
+                                        541,
+                                        421,
+                                        553
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\mathcal { Z } \\subseteq \\mathbb { R } ^ { d _ { z } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        421,
+                                        541,
+                                        505,
+                                        554
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is the representation",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                553,
+                                505,
+                                565
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        553,
+                                        479,
+                                        565
+                                    ],
+                                    "score": 1.0,
+                                    "content": "space. We define the transition and the reward models in terms of the representation function",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        480,
+                                        553,
+                                        487,
+                                        564
+                                    ],
+                                    "score": 0.84,
+                                    "content": "\\phi",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        487,
+                                        553,
+                                        505,
+                                        565
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", i.e.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                562,
+                                506,
+                                579
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        563,
+                                        157,
+                                        577
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\bar { \\widehat { T } } = \\widehat { T } _ { z } \\circ \\phi",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        158,
+                                        562,
+                                        176,
+                                        579
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        176,
+                                        563,
+                                        228,
+                                        577
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\widehat { R } = \\widehat { R } _ { z } \\circ \\phi",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        229,
+                                        562,
+                                        427,
+                                        579
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". In practice, where we use a neural network for",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        427,
+                                        564,
+                                        436,
+                                        576
+                                    ],
+                                    "score": 0.86,
+                                    "content": "\\widehat { T }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        436,
+                                        562,
+                                        455,
+                                        579
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        455,
+                                        564,
+                                        464,
+                                        576
+                                    ],
+                                    "score": 0.72,
+                                    "content": "\\widehat { R }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        464,
+                                        562,
+                                        468,
+                                        579
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        468,
+                                        568,
+                                        475,
+                                        576
+                                    ],
+                                    "score": 0.6,
+                                    "content": "z",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        475,
+                                        562,
+                                        506,
+                                        579
+                                    ],
+                                    "score": 1.0,
+                                    "content": "can be",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                576,
+                                505,
+                                589
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        576,
+                                        366,
+                                        589
+                                    ],
+                                    "score": 1.0,
+                                    "content": "chosen to be the output of an intermediate hidden layer, making",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        366,
+                                        577,
+                                        374,
+                                        588
+                                    ],
+                                    "score": 0.84,
+                                    "content": "\\phi",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        374,
+                                        576,
+                                        505,
+                                        589
+                                    ],
+                                    "score": 1.0,
+                                    "content": "represented by lower layers and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                586,
+                                504,
+                                604
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        587,
+                                        135,
+                                        601
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\widehat { T } _ { z } , \\widehat { R } _ { z }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        135,
+                                        586,
+                                        304,
+                                        604
+                                    ],
+                                    "score": 1.0,
+                                    "content": "by the remaining upper layers. We define",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        305,
+                                        589,
+                                        329,
+                                        603
+                                    ],
+                                    "score": 0.93,
+                                    "content": "d _ { \\phi } ^ { \\pi } ( z )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        329,
+                                        586,
+                                        495,
+                                        604
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the discounted stationary distribution on",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        495,
+                                        590,
+                                        504,
+                                        599
+                                    ],
+                                    "score": 0.76,
+                                    "content": "\\mathcal { Z }",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                601,
+                                473,
+                                617
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        601,
+                                        153,
+                                        617
+                                    ],
+                                    "score": 1.0,
+                                    "content": "induced by",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        153,
+                                        603,
+                                        187,
+                                        615
+                                    ],
+                                    "score": 0.93,
+                                    "content": "d ^ { \\pi } ( s , a )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        187,
+                                        601,
+                                        322,
+                                        617
+                                    ],
+                                    "score": 1.0,
+                                    "content": "under the representation function",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        323,
+                                        603,
+                                        370,
+                                        615
+                                    ],
+                                    "score": 0.93,
+                                    "content": "z = \\phi ( s , a )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        370,
+                                        601,
+                                        443,
+                                        617
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", and similarly for",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        443,
+                                        602,
+                                        469,
+                                        617
+                                    ],
+                                    "score": 0.93,
+                                    "content": "d _ { \\phi } ^ { { \\cal D } } ( z )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        469,
+                                        601,
+                                        473,
+                                        617
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        }
+                    ],
+                    "index": 32
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        630,
+                        425,
+                        644
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                630,
+                                426,
+                                645
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        630,
+                                        426,
+                                        645
+                                    ],
+                                    "score": 1.0,
+                                    "content": "4 REPRESENTATION BALANCING OFFLINE MODEL-BASED RL",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        }
+                    ],
+                    "index": 39
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        104,
+                        655,
+                        470,
+                        666
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                655,
+                                470,
+                                666
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        655,
+                                        470,
+                                        666
+                                    ],
+                                    "score": 1.0,
+                                    "content": "4.1 GENERALIZATION ERROR BOUND FOR MODEL-BASED OFF-POLICY EVALUATION",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        }
+                    ],
+                    "index": 40
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        672,
+                        505,
+                        711
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                673,
+                                506,
+                                689
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        673,
+                                        228,
+                                        689
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We aim to construct a model",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        228,
+                                        673,
+                                        240,
+                                        685
+                                    ],
+                                    "score": 0.87,
+                                    "content": "\\widehat { M }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        240,
+                                        673,
+                                        311,
+                                        689
+                                    ],
+                                    "score": 1.0,
+                                    "content": "from the dataset",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        311,
+                                        676,
+                                        320,
+                                        685
+                                    ],
+                                    "score": 0.8,
+                                    "content": "\\mathcal { D }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        321,
+                                        673,
+                                        494,
+                                        689
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", which can accurately evaluate the policy",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        494,
+                                        678,
+                                        501,
+                                        685
+                                    ],
+                                    "score": 0.7,
+                                    "content": "\\pi",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        501,
+                                        673,
+                                        506,
+                                        689
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                685,
+                                507,
+                                704
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        685,
+                                        356,
+                                        704
+                                    ],
+                                    "score": 1.0,
+                                    "content": "by minimizing a good upper bound of policy evaluation error",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        356,
+                                        686,
+                                        401,
+                                        700
+                                    ],
+                                    "score": 0.93,
+                                    "content": "| R ^ { \\pi } - { \\widehat { R } } ^ { \\pi } |",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        401,
+                                        685,
+                                        507,
+                                        704
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". We define the following",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                699,
+                                307,
+                                712
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        699,
+                                        307,
+                                        712
+                                    ],
+                                    "score": 1.0,
+                                    "content": "point-wise model loss for notational convenience:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        }
+                    ],
+                    "index": 42
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        115,
+                        714,
+                        493,
+                        736
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                115,
+                                714,
+                                493,
+                                736
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        714,
+                                        493,
+                                        736
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\begin{array} { r } { \\mathcal { E } _ { \\phi , \\widehat { R } _ { z } , \\widehat { T } _ { z } } ( s , a ) = c _ { R } D _ { T V } \\Big ( R ( r | s , a ) \\Big | \\Big | \\widehat { R } _ { z } \\big ( r | \\phi ( s , a ) \\big ) \\Big ) + c _ { T } D _ { T V } \\Big ( T ( s ^ { \\prime } | s , a ) \\Big | \\Big | \\widehat { T } _ { z } \\big ( s ^ { \\prime } | \\phi ( s , a ) \\big ) \\Big ) , } \\end{array}",
+                                    "type": "interline_equation",
+                                    "image_path": "ad510db3114669b45b366d79bc47298250fb467c2c679abf0cf02629ee163245.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 44,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                115,
+                                714,
+                                493,
+                                736
+                            ],
+                            "spans": [],
+                            "index": 44
+                        }
+                    ]
+                }
+            ],
+            "page_idx": 2,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        108,
+                        27,
+                        292,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                26,
+                                293,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        26,
+                                        293,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Published as a conference paper at ICLR 2021",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        302,
+                        751,
+                        309,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                301,
+                                750,
+                                309,
+                                762
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        301,
+                                        750,
+                                        309,
+                                        762
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        82,
+                        504,
+                        105
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                81,
+                                505,
+                                95
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        81,
+                                        505,
+                                        95
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and MBOP (Argenson & Dulac-Arnold, 2020) guide their policy optimization using the behavior",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                93,
+                                335,
+                                107
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        93,
+                                        335,
+                                        107
+                                    ],
+                                    "score": 1.0,
+                                    "content": "policy, similar to the behavior regularization approaches.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        }
+                    ],
+                    "index": 0.5,
+                    "bbox_fs": [
+                        105,
+                        81,
+                        505,
+                        107
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        110,
+                        505,
+                        166
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                110,
+                                504,
+                                123
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        110,
+                                        504,
+                                        123
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Note that these aforementioned offline RL methods build on the standard approximate dynamic pro-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                121,
+                                505,
+                                134
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        121,
+                                        505,
+                                        134
+                                    ],
+                                    "score": 1.0,
+                                    "content": "gramming algorithm for action-value estimation (model-free) or on a maximum-likelihood environ-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                132,
+                                505,
+                                145
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        132,
+                                        505,
+                                        145
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ment model (model-based), without explicitly addressing the distribution shift in the estimator. In",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                144,
+                                505,
+                                155
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        144,
+                                        505,
+                                        155
+                                    ],
+                                    "score": 1.0,
+                                    "content": "contrast, we augment the objective for model learning to obtain a robust model under the distribution",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                153,
+                                406,
+                                168
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        153,
+                                        406,
+                                        168
+                                    ],
+                                    "score": 1.0,
+                                    "content": "shift, which is the first attempt for offline RL to the best of our knowledge.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        }
+                    ],
+                    "index": 4,
+                    "bbox_fs": [
+                        105,
+                        110,
+                        505,
+                        168
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        181,
+                        208,
+                        194
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                181,
+                                209,
+                                196
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        181,
+                                        209,
+                                        196
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3 PRELIMINARIES",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        }
+                    ],
+                    "index": 7
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        205,
+                        506,
+                        309
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                204,
+                                506,
+                                219
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        204,
+                                        352,
+                                        219
+                                    ],
+                                    "score": 1.0,
+                                    "content": "A Markov Decision Process (MDP) is specified by a tuple",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        353,
+                                        206,
+                                        456,
+                                        218
+                                    ],
+                                    "score": 0.91,
+                                    "content": "M = \\langle S , A , T , R , d _ { 0 } , \\gamma \\rangle",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        457,
+                                        204,
+                                        506,
+                                        219
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", consisting",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                217,
+                                506,
+                                229
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        217,
+                                        168,
+                                        229
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of state space",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        168,
+                                        217,
+                                        176,
+                                        227
+                                    ],
+                                    "score": 0.72,
+                                    "content": "s",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        177,
+                                        217,
+                                        237,
+                                        229
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", action space",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        238,
+                                        218,
+                                        247,
+                                        227
+                                    ],
+                                    "score": 0.75,
+                                    "content": "\\mathcal { A }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        247,
+                                        217,
+                                        332,
+                                        229
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", transition function",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        333,
+                                        218,
+                                        431,
+                                        229
+                                    ],
+                                    "score": 0.88,
+                                    "content": "T : \\mathcal { S } \\times \\mathcal { A }  \\Delta ( \\mathcal { S } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        431,
+                                        217,
+                                        506,
+                                        229
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", reward function",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                227,
+                                506,
+                                241
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        228,
+                                        226,
+                                        240
+                                    ],
+                                    "score": 0.89,
+                                    "content": "R : \\mathcal { S } \\times \\overline { { \\mathcal { A } } } \\to \\Delta ( [ 0 , r _ { \\operatorname* { m a x } } ] )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        226,
+                                        227,
+                                        329,
+                                        241
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", initial state distribution",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        329,
+                                        228,
+                                        340,
+                                        239
+                                    ],
+                                    "score": 0.88,
+                                    "content": "d _ { 0 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        340,
+                                        227,
+                                        418,
+                                        241
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", and discount rate",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        419,
+                                        230,
+                                        425,
+                                        240
+                                    ],
+                                    "score": 0.81,
+                                    "content": "\\gamma",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        426,
+                                        227,
+                                        506,
+                                        241
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". In this paper, we",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                238,
+                                505,
+                                251
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        239,
+                                        267,
+                                        251
+                                    ],
+                                    "score": 1.0,
+                                    "content": "mainly focus on continuous state space",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        267,
+                                        238,
+                                        305,
+                                        250
+                                    ],
+                                    "score": 0.93,
+                                    "content": "S \\subseteq \\mathbb { R } ^ { d _ { s } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        306,
+                                        239,
+                                        505,
+                                        251
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and conduct experiments on both discrete action",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                248,
+                                506,
+                                263
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        248,
+                                        136,
+                                        263
+                                    ],
+                                    "score": 1.0,
+                                    "content": "spaces",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        136,
+                                        250,
+                                        209,
+                                        262
+                                    ],
+                                    "score": 0.93,
+                                    "content": "{ \\cal A } = \\{ a _ { 0 } , . . . a _ { n _ { a } } \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        209,
+                                        248,
+                                        333,
+                                        263
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and continuous action spaces",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        333,
+                                        249,
+                                        374,
+                                        261
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\mathcal { A } \\subseteq \\mathbb { R } ^ { d _ { a } ^ { \\star } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        375,
+                                        248,
+                                        434,
+                                        263
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Given MDP",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        434,
+                                        250,
+                                        446,
+                                        260
+                                    ],
+                                    "score": 0.73,
+                                    "content": "M",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        447,
+                                        248,
+                                        493,
+                                        263
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and policy",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        494,
+                                        252,
+                                        501,
+                                        260
+                                    ],
+                                    "score": 0.71,
+                                    "content": "\\pi",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        502,
+                                        248,
+                                        506,
+                                        263
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                260,
+                                506,
+                                273
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        260,
+                                        506,
+                                        273
+                                    ],
+                                    "score": 1.0,
+                                    "content": "which is a (stochastic) mapping from state to action, the trajectory can be generated in the form of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                107,
+                                272,
+                                505,
+                                285
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        274,
+                                        199,
+                                        283
+                                    ],
+                                    "score": 0.87,
+                                    "content": "s _ { 0 } , a _ { 0 } , r _ { 0 } , s _ { 1 } , a _ { 1 } , r _ { 1 } , \\ldots",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        199,
+                                        272,
+                                        230,
+                                        285
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", where",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        231,
+                                        272,
+                                        266,
+                                        283
+                                    ],
+                                    "score": 0.92,
+                                    "content": "s _ { 0 } \\sim d _ { 0 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        266,
+                                        272,
+                                        357,
+                                        285
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and for each timestep",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        358,
+                                        272,
+                                        383,
+                                        283
+                                    ],
+                                    "score": 0.86,
+                                    "content": "t \\geq 0",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        384,
+                                        272,
+                                        388,
+                                        285
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        388,
+                                        272,
+                                        435,
+                                        284
+                                    ],
+                                    "score": 0.83,
+                                    "content": "a _ { t } \\sim \\pi ( s _ { t } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        435,
+                                        272,
+                                        439,
+                                        285
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        440,
+                                        272,
+                                        501,
+                                        284
+                                    ],
+                                    "score": 0.89,
+                                    "content": "\\boldsymbol { r } _ { t } \\sim R ( s _ { t } , a _ { t } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        501,
+                                        272,
+                                        505,
+                                        285
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                282,
+                                506,
+                                296
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        282,
+                                        124,
+                                        296
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        124,
+                                        284,
+                                        194,
+                                        295
+                                    ],
+                                    "score": 0.93,
+                                    "content": "s _ { t + 1 } \\sim T ( s _ { t } , a _ { t } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        194,
+                                        282,
+                                        506,
+                                        296
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". The goal of RL is to optimize or evaluate a policy, based on the normalized",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                292,
+                                356,
+                                311
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        292,
+                                        219,
+                                        311
+                                    ],
+                                    "score": 1.0,
+                                    "content": "expected discounted return:",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        220,
+                                        294,
+                                        350,
+                                        308
+                                    ],
+                                    "score": 0.9,
+                                    "content": "\\begin{array} { r } { R ^ { \\pi } \\triangleq ( 1 - \\gamma ) \\mathbb { E } _ { M , \\pi } \\left[ \\sum _ { t = 0 } ^ { \\infty } \\gamma ^ { t } r _ { t } \\right] } \\end{array}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        351,
+                                        292,
+                                        356,
+                                        311
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        }
+                    ],
+                    "index": 12,
+                    "bbox_fs": [
+                        104,
+                        204,
+                        506,
+                        311
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        312,
+                        504,
+                        335
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                311,
+                                505,
+                                325
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        311,
+                                        505,
+                                        325
+                                    ],
+                                    "score": 1.0,
+                                    "content": "A useful and important concept throughout the paper is the discounted stationary distribution, which",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                324,
+                                289,
+                                336
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        324,
+                                        289,
+                                        336
+                                    ],
+                                    "score": 1.0,
+                                    "content": "represents the long-term occupancy of states:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        }
+                    ],
+                    "index": 17.5,
+                    "bbox_fs": [
+                        105,
+                        311,
+                        505,
+                        336
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        200,
+                        338,
+                        410,
+                        371
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                200,
+                                338,
+                                410,
+                                371
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        200,
+                                        338,
+                                        410,
+                                        371
+                                    ],
+                                    "score": 0.93,
+                                    "content": "d ^ { \\pi } ( s , a ) \\triangleq ( 1 - \\gamma ) \\sum _ { t = 0 } ^ { \\infty } \\gamma ^ { t } \\operatorname* { P r } ( s _ { t } = s , a _ { t } = a | M , \\pi ) .",
+                                    "type": "interline_equation",
+                                    "image_path": "4e97c5ca6686900ac9b0f9bd4d01b84db14920e7a7632d71f5cbe853c1d7c3e0.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 19.5,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                200,
+                                338,
+                                410,
+                                354.5
+                            ],
+                            "spans": [],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                200,
+                                354.5,
+                                410,
+                                371.0
+                            ],
+                            "spans": [],
+                            "index": 20
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        104,
+                        374,
+                        474,
+                        387
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                371,
+                                472,
+                                389
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        371,
+                                        280,
+                                        389
+                                    ],
+                                    "score": 1.0,
+                                    "content": "From the definition, it can be observed that",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        281,
+                                        375,
+                                        294,
+                                        384
+                                    ],
+                                    "score": 0.88,
+                                    "content": "R ^ { \\pi }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        295,
+                                        371,
+                                        373,
+                                        389
+                                    ],
+                                    "score": 1.0,
+                                    "content": "can be obtained by",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        374,
+                                        374,
+                                        472,
+                                        387
+                                    ],
+                                    "score": 0.92,
+                                    "content": "R ^ { \\pi } = \\mathbb { E } _ { ( s , a ) \\sim d ^ { \\pi } } \\left[ r ( s , a ) \\right]",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 21
+                        }
+                    ],
+                    "index": 21,
+                    "bbox_fs": [
+                        104,
+                        371,
+                        472,
+                        389
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        397,
+                        505,
+                        442
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                397,
+                                505,
+                                411
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        397,
+                                        426,
+                                        411
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Offline RL and off-policy evaluation In this paper, we focus on the offline",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        426,
+                                        398,
+                                        440,
+                                        408
+                                    ],
+                                    "score": 0.61,
+                                    "content": "R L",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        441,
+                                        397,
+                                        505,
+                                        411
+                                    ],
+                                    "score": 1.0,
+                                    "content": "problem where",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                406,
+                                507,
+                                423
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        406,
+                                        271,
+                                        423
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the agent can only access a static dataset",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        272,
+                                        408,
+                                        371,
+                                        421
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\mathcal { D } = \\{ ( \\bar { s } _ { i } , a _ { i } , r _ { i } , s _ { i } ^ { \\prime } ) \\} _ { i = 1 } ^ { N }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        371,
+                                        406,
+                                        470,
+                                        423
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for the maximization of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        471,
+                                        409,
+                                        484,
+                                        419
+                                    ],
+                                    "score": 0.86,
+                                    "content": "R ^ { \\pi }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        484,
+                                        406,
+                                        507,
+                                        423
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". We",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                420,
+                                505,
+                                432
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        420,
+                                        505,
+                                        432
+                                    ],
+                                    "score": 1.0,
+                                    "content": "consider a behavior-agnostic setting where we do not have any knowledge of the data collection",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                430,
+                                374,
+                                443
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        431,
+                                        357,
+                                        443
+                                    ],
+                                    "score": 1.0,
+                                    "content": "process. We denote the empirical distribution of the dataset by",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        358,
+                                        430,
+                                        371,
+                                        441
+                                    ],
+                                    "score": 0.89,
+                                    "content": "\\dot { d } ^ { \\mathcal { D } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        371,
+                                        431,
+                                        374,
+                                        443
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        }
+                    ],
+                    "index": 23.5,
+                    "bbox_fs": [
+                        104,
+                        397,
+                        507,
+                        443
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        447,
+                        506,
+                        615
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                447,
+                                505,
+                                460
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        447,
+                                        324,
+                                        460
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Before improving policy, we first aim to better evaluate",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        325,
+                                        448,
+                                        338,
+                                        458
+                                    ],
+                                    "score": 0.88,
+                                    "content": "R ^ { \\pi }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        339,
+                                        447,
+                                        421,
+                                        460
+                                    ],
+                                    "score": 1.0,
+                                    "content": "given a target policy",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        421,
+                                        450,
+                                        429,
+                                        457
+                                    ],
+                                    "score": 0.76,
+                                    "content": "\\pi",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        429,
+                                        447,
+                                        505,
+                                        460
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and a static dataset",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                459,
+                                505,
+                                471
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        459,
+                                        115,
+                                        469
+                                    ],
+                                    "score": 0.75,
+                                    "content": "\\mathcal { D }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        116,
+                                        459,
+                                        505,
+                                        471
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", which corresponds to an off-policy evaluation (OPE) problem. We mainly focus on a model-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                469,
+                                505,
+                                484
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        471,
+                                        412,
+                                        484
+                                    ],
+                                    "score": 1.0,
+                                    "content": "based approach where the algorithm first estimates the unknown dynamics",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        412,
+                                        469,
+                                        433,
+                                        483
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\widehat { T } , \\widehat { R }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        434,
+                                        471,
+                                        505,
+                                        484
+                                    ],
+                                    "score": 1.0,
+                                    "content": "using the dataset",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                482,
+                                506,
+                                497
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        484,
+                                        115,
+                                        495
+                                    ],
+                                    "score": 0.76,
+                                    "content": "\\mathcal { D }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        116,
+                                        483,
+                                        268,
+                                        497
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". This defines an approximate MDP",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        268,
+                                        482,
+                                        371,
+                                        497
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\widehat { M } = \\langle S , A , \\widehat { T } , \\widehat { R } , d _ { 0 } , \\gamma \\rangle",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        372,
+                                        483,
+                                        506,
+                                        497
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", with the approximate expected",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                102,
+                                492,
+                                509,
+                                516
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        102,
+                                        492,
+                                        178,
+                                        516
+                                    ],
+                                    "score": 1.0,
+                                    "content": "discounted return",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        178,
+                                        496,
+                                        307,
+                                        512
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\begin{array} { r } { \\widehat { R } ^ { \\pi } \\triangleq ( 1 - \\gamma ) \\mathbb { E } _ { \\widehat { M } , \\pi } \\left[ \\sum _ { t = 0 } ^ { \\infty } \\gamma ^ { t } r _ { t } \\right] } \\end{array}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        307,
+                                        492,
+                                        366,
+                                        516
+                                    ],
+                                    "score": 1.0,
+                                    "content": "obtained from",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        366,
+                                        496,
+                                        378,
+                                        508
+                                    ],
+                                    "score": 0.87,
+                                    "content": "\\widehat { M }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        378,
+                                        492,
+                                        509,
+                                        516
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". In this paper, we are interested",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                511,
+                                505,
+                                527
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        511,
+                                        190,
+                                        527
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in the MDP estimate",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        190,
+                                        511,
+                                        202,
+                                        524
+                                    ],
+                                    "score": 0.85,
+                                    "content": "\\widehat { M }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        202,
+                                        511,
+                                        448,
+                                        527
+                                    ],
+                                    "score": 1.0,
+                                    "content": "that can effectively reduce the error in the evaluation of policy",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        448,
+                                        516,
+                                        455,
+                                        524
+                                    ],
+                                    "score": 0.57,
+                                    "content": "\\pi",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        456,
+                                        511,
+                                        459,
+                                        527
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        459,
+                                        511,
+                                        501,
+                                        526
+                                    ],
+                                    "score": 0.91,
+                                    "content": "| R ^ { \\pi } - { \\widehat R } ^ { \\pi } |",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        501,
+                                        511,
+                                        505,
+                                        527
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                530,
+                                505,
+                                543
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        530,
+                                        505,
+                                        543
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In order to do so, we need to learn a good representation of a model that results in a small OPE error.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                541,
+                                505,
+                                554
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        541,
+                                        291,
+                                        554
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We assume a bijective representation function",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        291,
+                                        542,
+                                        355,
+                                        553
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\phi : \\mathcal { S } \\times \\mathcal { A }  \\mathcal { Z }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        356,
+                                        541,
+                                        383,
+                                        554
+                                    ],
+                                    "score": 1.0,
+                                    "content": "where",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        383,
+                                        541,
+                                        421,
+                                        553
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\mathcal { Z } \\subseteq \\mathbb { R } ^ { d _ { z } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        421,
+                                        541,
+                                        505,
+                                        554
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is the representation",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                553,
+                                505,
+                                565
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        553,
+                                        479,
+                                        565
+                                    ],
+                                    "score": 1.0,
+                                    "content": "space. We define the transition and the reward models in terms of the representation function",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        480,
+                                        553,
+                                        487,
+                                        564
+                                    ],
+                                    "score": 0.84,
+                                    "content": "\\phi",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        487,
+                                        553,
+                                        505,
+                                        565
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", i.e.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                562,
+                                506,
+                                579
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        563,
+                                        157,
+                                        577
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\bar { \\widehat { T } } = \\widehat { T } _ { z } \\circ \\phi",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        158,
+                                        562,
+                                        176,
+                                        579
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        176,
+                                        563,
+                                        228,
+                                        577
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\widehat { R } = \\widehat { R } _ { z } \\circ \\phi",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        229,
+                                        562,
+                                        427,
+                                        579
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". In practice, where we use a neural network for",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        427,
+                                        564,
+                                        436,
+                                        576
+                                    ],
+                                    "score": 0.86,
+                                    "content": "\\widehat { T }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        436,
+                                        562,
+                                        455,
+                                        579
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        455,
+                                        564,
+                                        464,
+                                        576
+                                    ],
+                                    "score": 0.72,
+                                    "content": "\\widehat { R }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        464,
+                                        562,
+                                        468,
+                                        579
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        468,
+                                        568,
+                                        475,
+                                        576
+                                    ],
+                                    "score": 0.6,
+                                    "content": "z",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        475,
+                                        562,
+                                        506,
+                                        579
+                                    ],
+                                    "score": 1.0,
+                                    "content": "can be",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                576,
+                                505,
+                                589
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        576,
+                                        366,
+                                        589
+                                    ],
+                                    "score": 1.0,
+                                    "content": "chosen to be the output of an intermediate hidden layer, making",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        366,
+                                        577,
+                                        374,
+                                        588
+                                    ],
+                                    "score": 0.84,
+                                    "content": "\\phi",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        374,
+                                        576,
+                                        505,
+                                        589
+                                    ],
+                                    "score": 1.0,
+                                    "content": "represented by lower layers and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                586,
+                                504,
+                                604
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        587,
+                                        135,
+                                        601
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\widehat { T } _ { z } , \\widehat { R } _ { z }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        135,
+                                        586,
+                                        304,
+                                        604
+                                    ],
+                                    "score": 1.0,
+                                    "content": "by the remaining upper layers. We define",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        305,
+                                        589,
+                                        329,
+                                        603
+                                    ],
+                                    "score": 0.93,
+                                    "content": "d _ { \\phi } ^ { \\pi } ( z )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        329,
+                                        586,
+                                        495,
+                                        604
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the discounted stationary distribution on",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        495,
+                                        590,
+                                        504,
+                                        599
+                                    ],
+                                    "score": 0.76,
+                                    "content": "\\mathcal { Z }",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                601,
+                                473,
+                                617
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        601,
+                                        153,
+                                        617
+                                    ],
+                                    "score": 1.0,
+                                    "content": "induced by",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        153,
+                                        603,
+                                        187,
+                                        615
+                                    ],
+                                    "score": 0.93,
+                                    "content": "d ^ { \\pi } ( s , a )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        187,
+                                        601,
+                                        322,
+                                        617
+                                    ],
+                                    "score": 1.0,
+                                    "content": "under the representation function",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        323,
+                                        603,
+                                        370,
+                                        615
+                                    ],
+                                    "score": 0.93,
+                                    "content": "z = \\phi ( s , a )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        370,
+                                        601,
+                                        443,
+                                        617
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", and similarly for",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        443,
+                                        602,
+                                        469,
+                                        617
+                                    ],
+                                    "score": 0.93,
+                                    "content": "d _ { \\phi } ^ { { \\cal D } } ( z )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        469,
+                                        601,
+                                        473,
+                                        617
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        }
+                    ],
+                    "index": 32,
+                    "bbox_fs": [
+                        102,
+                        447,
+                        509,
+                        617
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        630,
+                        425,
+                        644
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                630,
+                                426,
+                                645
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        630,
+                                        426,
+                                        645
+                                    ],
+                                    "score": 1.0,
+                                    "content": "4 REPRESENTATION BALANCING OFFLINE MODEL-BASED RL",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        }
+                    ],
+                    "index": 39
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        104,
+                        655,
+                        470,
+                        666
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                655,
+                                470,
+                                666
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        655,
+                                        470,
+                                        666
+                                    ],
+                                    "score": 1.0,
+                                    "content": "4.1 GENERALIZATION ERROR BOUND FOR MODEL-BASED OFF-POLICY EVALUATION",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        }
+                    ],
+                    "index": 40
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        672,
+                        505,
+                        711
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                673,
+                                506,
+                                689
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        673,
+                                        228,
+                                        689
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We aim to construct a model",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        228,
+                                        673,
+                                        240,
+                                        685
+                                    ],
+                                    "score": 0.87,
+                                    "content": "\\widehat { M }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        240,
+                                        673,
+                                        311,
+                                        689
+                                    ],
+                                    "score": 1.0,
+                                    "content": "from the dataset",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        311,
+                                        676,
+                                        320,
+                                        685
+                                    ],
+                                    "score": 0.8,
+                                    "content": "\\mathcal { D }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        321,
+                                        673,
+                                        494,
+                                        689
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", which can accurately evaluate the policy",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        494,
+                                        678,
+                                        501,
+                                        685
+                                    ],
+                                    "score": 0.7,
+                                    "content": "\\pi",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        501,
+                                        673,
+                                        506,
+                                        689
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                685,
+                                507,
+                                704
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        685,
+                                        356,
+                                        704
+                                    ],
+                                    "score": 1.0,
+                                    "content": "by minimizing a good upper bound of policy evaluation error",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        356,
+                                        686,
+                                        401,
+                                        700
+                                    ],
+                                    "score": 0.93,
+                                    "content": "| R ^ { \\pi } - { \\widehat { R } } ^ { \\pi } |",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        401,
+                                        685,
+                                        507,
+                                        704
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". We define the following",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                699,
+                                307,
+                                712
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        699,
+                                        307,
+                                        712
+                                    ],
+                                    "score": 1.0,
+                                    "content": "point-wise model loss for notational convenience:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        }
+                    ],
+                    "index": 42,
+                    "bbox_fs": [
+                        104,
+                        673,
+                        507,
+                        712
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        115,
+                        714,
+                        493,
+                        736
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                115,
+                                714,
+                                493,
+                                736
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        714,
+                                        493,
+                                        736
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\begin{array} { r } { \\mathcal { E } _ { \\phi , \\widehat { R } _ { z } , \\widehat { T } _ { z } } ( s , a ) = c _ { R } D _ { T V } \\Big ( R ( r | s , a ) \\Big | \\Big | \\widehat { R } _ { z } \\big ( r | \\phi ( s , a ) \\big ) \\Big ) + c _ { T } D _ { T V } \\Big ( T ( s ^ { \\prime } | s , a ) \\Big | \\Big | \\widehat { T } _ { z } \\big ( s ^ { \\prime } | \\phi ( s , a ) \\big ) \\Big ) , } \\end{array}",
+                                    "type": "interline_equation",
+                                    "image_path": "ad510db3114669b45b366d79bc47298250fb467c2c679abf0cf02629ee163245.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 44,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                115,
+                                714,
+                                493,
+                                736
+                            ],
+                            "spans": [],
+                            "index": 44
+                        }
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        82,
+                        505,
+                        116
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                82,
+                                505,
+                                96
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        82,
+                                        133,
+                                        96
+                                    ],
+                                    "score": 1.0,
+                                    "content": "where",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        133,
+                                        82,
+                                        195,
+                                        95
+                                    ],
+                                    "score": 0.92,
+                                    "content": "c _ { R } = 2 ( 1 - \\gamma )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        195,
+                                        82,
+                                        213,
+                                        96
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        213,
+                                        83,
+                                        269,
+                                        94
+                                    ],
+                                    "score": 0.92,
+                                    "content": "c _ { T } = 2 \\gamma r _ { \\operatorname* { m a x } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        269,
+                                        82,
+                                        505,
+                                        96
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Then, we start by restating the simulation lemma (Kearns",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                93,
+                                505,
+                                106
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        93,
+                                        505,
+                                        106
+                                    ],
+                                    "score": 1.0,
+                                    "content": "& Singh, 2002) to bound the policy evaluation error in terms of the point-wise model loss. The proof",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                104,
+                                218,
+                                117
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        104,
+                                        218,
+                                        117
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is available in Appendix A.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        }
+                    ],
+                    "index": 1
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        119,
+                        504,
+                        146
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                118,
+                                505,
+                                134
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        118,
+                                        226,
+                                        134
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Lemma 4.1. Given an MDP",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        226,
+                                        121,
+                                        238,
+                                        131
+                                    ],
+                                    "score": 0.52,
+                                    "content": "M",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        239,
+                                        118,
+                                        306,
+                                        134
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and its estimate",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        306,
+                                        118,
+                                        318,
+                                        131
+                                    ],
+                                    "score": 0.86,
+                                    "content": "\\widehat { M }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        318,
+                                        118,
+                                        479,
+                                        134
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with a bijective representation function",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        479,
+                                        122,
+                                        487,
+                                        133
+                                    ],
+                                    "score": 0.79,
+                                    "content": "\\phi _ { : }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        487,
+                                        118,
+                                        505,
+                                        134
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", i.e.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                107,
+                                131,
+                                460,
+                                148
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        131,
+                                        213,
+                                        146
+                                    ],
+                                    "score": 0.92,
+                                    "content": "( \\widehat { T } , \\widehat { R } ) = \\langle \\widehat { T } _ { z } \\circ \\phi , \\widehat { R } _ { z } \\circ \\phi \\rangle",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        213,
+                                        131,
+                                        371,
+                                        148
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", the policy evaluation error of a policy",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        371,
+                                        136,
+                                        378,
+                                        144
+                                    ],
+                                    "score": 0.63,
+                                    "content": "\\pi",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        379,
+                                        131,
+                                        460,
+                                        148
+                                    ],
+                                    "score": 1.0,
+                                    "content": "can be bounded by:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        }
+                    ],
+                    "index": 3.5
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        222,
+                        149,
+                        388,
+                        171
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                222,
+                                149,
+                                388,
+                                171
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        222,
+                                        149,
+                                        388,
+                                        171
+                                    ],
+                                    "score": 0.94,
+                                    "content": "\\begin{array} { r } { \\left| R ^ { \\pi } - \\widehat { R } ^ { \\pi } \\right| \\leq \\mathbb { E } _ { ( s , a ) \\sim d ^ { \\pi } } \\left[ \\mathcal { E } _ { \\phi , \\widehat { R } _ { z } , \\widehat { T } _ { z } } ( s , a ) \\right] } \\end{array}",
+                                    "type": "interline_equation",
+                                    "image_path": "be4430fb6e26dd6d13dc5a5eadf0c4ec02c96ae654aff21a36528c073cefe5c6.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 5,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                222,
+                                149,
+                                388,
+                                171
+                            ],
+                            "spans": [],
+                            "index": 5
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        178,
+                        505,
+                        246
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                178,
+                                505,
+                                192
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        178,
+                                        505,
+                                        192
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The Lemma 4.1 has a natural interpretation: if the model error is small in the states frequently",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                190,
+                                506,
+                                204
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        190,
+                                        231,
+                                        204
+                                    ],
+                                    "score": 1.0,
+                                    "content": "visited by following the policy",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        232,
+                                        193,
+                                        239,
+                                        201
+                                    ],
+                                    "score": 0.76,
+                                    "content": "\\pi",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        239,
+                                        190,
+                                        506,
+                                        204
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", the resultant policy evaluation error will also be small. However,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                202,
+                                505,
+                                213
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        202,
+                                        505,
+                                        213
+                                    ],
+                                    "score": 1.0,
+                                    "content": "minimizing the RHS of Eq. (1) in the off-policy evaluation (OPE) task is generally intractable since",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                213,
+                                505,
+                                225
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        213,
+                                        169,
+                                        225
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the distribution",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        170,
+                                        213,
+                                        181,
+                                        222
+                                    ],
+                                    "score": 0.87,
+                                    "content": "d ^ { \\pi }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        182,
+                                        213,
+                                        505,
+                                        225
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is not directly accessible. Therefore, the common practice has been to construct",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                223,
+                                506,
+                                236
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        223,
+                                        248,
+                                        236
+                                    ],
+                                    "score": 1.0,
+                                    "content": "a maximum-likelihood MDP using",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        248,
+                                        224,
+                                        257,
+                                        233
+                                    ],
+                                    "score": 0.82,
+                                    "content": "\\mathcal { D }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        258,
+                                        223,
+                                        506,
+                                        236
+                                    ],
+                                    "score": 1.0,
+                                    "content": "while ignoring the distribution shift, but its OPE performance",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                235,
+                                180,
+                                247
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        235,
+                                        180,
+                                        247
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is not guaranteed.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        }
+                    ],
+                    "index": 8.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        250,
+                        505,
+                        297
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                250,
+                                505,
+                                264
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        250,
+                                        505,
+                                        264
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Instead, we will derive a tractable upper bound on the policy evaluation error by eliminating the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                262,
+                                504,
+                                274
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        262,
+                                        195,
+                                        274
+                                    ],
+                                    "score": 1.0,
+                                    "content": "direct dependence on",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        195,
+                                        263,
+                                        207,
+                                        272
+                                    ],
+                                    "score": 0.86,
+                                    "content": "d ^ { \\pi }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        207,
+                                        262,
+                                        504,
+                                        274
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in Eq. (1). To this end, we adopt the distance metric between two distri-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                272,
+                                505,
+                                286
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        272,
+                                        221,
+                                        286
+                                    ],
+                                    "score": 1.0,
+                                    "content": "butions over representations",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        221,
+                                        273,
+                                        232,
+                                        286
+                                    ],
+                                    "score": 0.88,
+                                    "content": "d _ { \\phi } ^ { \\pi }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        233,
+                                        272,
+                                        250,
+                                        286
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        251,
+                                        272,
+                                        264,
+                                        286
+                                    ],
+                                    "score": 0.9,
+                                    "content": "d _ { \\phi } ^ { \\mathcal { D } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        264,
+                                        272,
+                                        505,
+                                        286
+                                    ],
+                                    "score": 1.0,
+                                    "content": "that can bound their difference in expectations, which is the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                285,
+                                307,
+                                297
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        285,
+                                        307,
+                                        297
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Integral Probability Metric (IPM) (Muller, 1997): ¨",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        }
+                    ],
+                    "index": 13.5
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        210,
+                        299,
+                        401,
+                        321
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                210,
+                                299,
+                                401,
+                                321
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        210,
+                                        299,
+                                        401,
+                                        321
+                                    ],
+                                    "score": 0.95,
+                                    "content": "\\mathrm { I P M } _ { \\mathcal { G } } ( p , q ) = \\operatorname* { s u p } _ { g \\in \\mathcal { G } } \\big | \\mathbb { E } _ { z \\sim p } [ g ( z ) ] - \\mathbb { E } _ { z \\sim q } [ g ( z ) ] \\big | .",
+                                    "type": "interline_equation",
+                                    "image_path": "b86cdd7b9a7b014a45536a084ffd03b37556682b8e9c250e180c7355c8694f81.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 16,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                210,
+                                299,
+                                401,
+                                321
+                            ],
+                            "spans": [],
+                            "index": 16
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        325,
+                        505,
+                        348
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                325,
+                                505,
+                                336
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        325,
+                                        218,
+                                        336
+                                    ],
+                                    "score": 1.0,
+                                    "content": "where particular choices of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        218,
+                                        326,
+                                        226,
+                                        336
+                                    ],
+                                    "score": 0.84,
+                                    "content": "\\mathcal { G }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        226,
+                                        325,
+                                        505,
+                                        336
+                                    ],
+                                    "score": 1.0,
+                                    "content": "make the IPM equivalent to different well-known distances of distri-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                336,
+                                466,
+                                349
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        336,
+                                        466,
+                                        349
+                                    ],
+                                    "score": 1.0,
+                                    "content": "butions, e.g. total variation distance or Wasserstein distance (Sriperumbudur et al., 2009).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        }
+                    ],
+                    "index": 17.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        351,
+                        505,
+                        393
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                349,
+                                505,
+                                366
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        349,
+                                        235,
+                                        366
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Theorem 4.2. Given an MDP",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        235,
+                                        353,
+                                        247,
+                                        362
+                                    ],
+                                    "score": 0.59,
+                                    "content": "M",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        247,
+                                        349,
+                                        317,
+                                        366
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and its estimate",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        317,
+                                        350,
+                                        329,
+                                        362
+                                    ],
+                                    "score": 0.86,
+                                    "content": "\\widehat { M }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        329,
+                                        349,
+                                        494,
+                                        366
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with a bijective representation function",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        495,
+                                        353,
+                                        501,
+                                        364
+                                    ],
+                                    "score": 0.76,
+                                    "content": "\\phi",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        501,
+                                        349,
+                                        505,
+                                        366
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                362,
+                                505,
+                                398
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        362,
+                                        124,
+                                        379
+                                    ],
+                                    "score": 1.0,
+                                    "content": "i.e.",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        107,
+                                        378,
+                                        182,
+                                        390
+                                    ],
+                                    "score": 0.91,
+                                    "content": "{ \\mathcal { G } } \\triangleq \\{ g : { \\mathcal { Z } } \\to \\mathbb { R } \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        124,
+                                        363,
+                                        235,
+                                        377
+                                    ],
+                                    "score": 0.91,
+                                    "content": "( \\widehat { T } , \\widehat { R } ) = \\langle \\widehat { T } _ { z } \\circ \\phi , \\widehat { R } _ { z } \\circ \\phi \\rangle",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        182,
+                                        370,
+                                        222,
+                                        398
+                                    ],
+                                    "score": 1.0,
+                                    "content": "b such that",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        222,
+                                        377,
+                                        325,
+                                        393
+                                    ],
+                                    "score": 0.94,
+                                    "content": "\\begin{array} { r } { \\frac { 1 } { B _ { \\phi } } \\mathcal { E } _ { \\phi , \\widehat { R } _ { z } , \\widehat { T } _ { z } } \\left( \\phi ^ { - 1 } ( \\cdot ) \\right) \\in \\mathcal { G } } \\end{array}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        235,
+                                        362,
+                                        384,
+                                        379
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", assume that there exists a constant",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        325,
+                                        370,
+                                        412,
+                                        398
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Then, for any policy",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        384,
+                                        365,
+                                        419,
+                                        378
+                                    ],
+                                    "score": 0.93,
+                                    "content": "B _ { \\phi } > 0",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        412,
+                                        381,
+                                        419,
+                                        388
+                                    ],
+                                    "score": 0.73,
+                                    "content": "\\pi",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        419,
+                                        362,
+                                        438,
+                                        379
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        419,
+                                        370,
+                                        426,
+                                        398
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        439,
+                                        367,
+                                        445,
+                                        375
+                                    ],
+                                    "score": 0.26,
+                                    "content": "a",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        445,
+                                        362,
+                                        505,
+                                        379
+                                    ],
+                                    "score": 1.0,
+                                    "content": "function class",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        }
+                    ],
+                    "index": 19.5
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        179,
+                        397,
+                        431,
+                        418
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                179,
+                                397,
+                                431,
+                                418
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        179,
+                                        397,
+                                        431,
+                                        418
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\left| R ^ { \\pi } - \\widehat { R } ^ { \\pi } \\right| \\leq \\mathbb { E } _ { ( s , a ) \\sim d ^ { \\mathcal { D } } } \\Big [ \\mathcal { E } _ { \\phi , \\widehat { R } _ { z } , \\widehat { T } _ { z } } ( s , a ) \\Big ] + B _ { \\phi } \\mathrm { I P M } _ { \\mathcal { G } } ( d _ { \\phi } ^ { \\pi } , d _ { \\phi } ^ { \\mathcal { D } } )",
+                                    "type": "interline_equation",
+                                    "image_path": "50332b075b998d81ff520371d10d176b4273f51b2bb0d2bb4ffa9954584b2c6d.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 21,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                179,
+                                397,
+                                431,
+                                418
+                            ],
+                            "spans": [],
+                            "index": 21
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        426,
+                        505,
+                        518
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                426,
+                                506,
+                                439
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        426,
+                                        506,
+                                        439
+                                    ],
+                                    "score": 1.0,
+                                    "content": "This theorem is an adaptation of Lemma 1 of Shalit et al. (2017) to an infinite horizon model-based",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                435,
+                                508,
+                                455
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        435,
+                                        335,
+                                        455
+                                    ],
+                                    "score": 1.0,
+                                    "content": "policy evaluation and can be derived by the definition of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        335,
+                                        438,
+                                        396,
+                                        452
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\mathrm { I P M } _ { \\mathcal { G } } ( d _ { \\phi } ^ { \\pi } , d _ { \\phi } ^ { D } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        396,
+                                        435,
+                                        508,
+                                        455
+                                    ],
+                                    "score": 1.0,
+                                    "content": "since it serves as an upper",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                450,
+                                505,
+                                462
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        450,
+                                        348,
+                                        462
+                                    ],
+                                    "score": 1.0,
+                                    "content": "bound of the difference in the expectations of any function in",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        348,
+                                        452,
+                                        356,
+                                        461
+                                    ],
+                                    "score": 0.82,
+                                    "content": "\\mathcal { G }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        356,
+                                        450,
+                                        505,
+                                        462
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". The first term in Eq. (3) corresponds",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                460,
+                                505,
+                                473
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        461,
+                                        249,
+                                        473
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to the fitness to the data following",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        249,
+                                        460,
+                                        262,
+                                        471
+                                    ],
+                                    "score": 0.87,
+                                    "content": "d ^ { \\mathcal { D } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        262,
+                                        461,
+                                        505,
+                                        473
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", while the second term serves as a regularizer. To see this,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                471,
+                                505,
+                                483
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        471,
+                                        505,
+                                        483
+                                    ],
+                                    "score": 1.0,
+                                    "content": "minimizing the second term would yield a near-constant representation function, which would be",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                483,
+                                505,
+                                495
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        483,
+                                        505,
+                                        495
+                                    ],
+                                    "score": 1.0,
+                                    "content": "bad for the first term since it cannot distinguish states and actions well enough. It shows a natural",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                494,
+                                506,
+                                506
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        494,
+                                        506,
+                                        506
+                                    ],
+                                    "score": 1.0,
+                                    "content": "trade-off between optimizing the model that fits data better and learning the representation that is",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                502,
+                                253,
+                                521
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        502,
+                                        205,
+                                        521
+                                    ],
+                                    "score": 1.0,
+                                    "content": "invariant with respect to",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        205,
+                                        505,
+                                        217,
+                                        519
+                                    ],
+                                    "score": 0.89,
+                                    "content": "d _ { \\phi } ^ { \\pi }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        217,
+                                        502,
+                                        235,
+                                        521
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        235,
+                                        504,
+                                        248,
+                                        519
+                                    ],
+                                    "score": 0.91,
+                                    "content": "d _ { \\phi } ^ { \\mathcal { D } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        248,
+                                        502,
+                                        253,
+                                        521
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        }
+                    ],
+                    "index": 25.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        522,
+                        505,
+                        567
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                523,
+                                505,
+                                534
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        523,
+                                        461,
+                                        534
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Nevertheless, RHS of Eq. (3) still cannot be evaluated naively due to its dependence on",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        462,
+                                        523,
+                                        473,
+                                        533
+                                    ],
+                                    "score": 0.84,
+                                    "content": "d ^ { \\pi }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        473,
+                                        523,
+                                        505,
+                                        534
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in esti-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                533,
+                                505,
+                                545
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        533,
+                                        505,
+                                        545
+                                    ],
+                                    "score": 1.0,
+                                    "content": "mating the IPM. We address this challenge via a change of variable, which is known as a DualDICE",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                544,
+                                505,
+                                557
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        544,
+                                        257,
+                                        557
+                                    ],
+                                    "score": 1.0,
+                                    "content": "trick (Nachum et al., 2019a). Define",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        257,
+                                        545,
+                                        307,
+                                        555
+                                    ],
+                                    "score": 0.9,
+                                    "content": "\\nu : \\mathcal { Z } \\to \\mathbb { R }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        307,
+                                        544,
+                                        505,
+                                        557
+                                    ],
+                                    "score": 1.0,
+                                    "content": "as an arbitrary function of state-action pairs that",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                555,
+                                144,
+                                567
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        555,
+                                        144,
+                                        567
+                                    ],
+                                    "score": 1.0,
+                                    "content": "satisfies:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        }
+                    ],
+                    "index": 31.5
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        156,
+                        568,
+                        455,
+                        596
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                156,
+                                568,
+                                455,
+                                596
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        156,
+                                        568,
+                                        455,
+                                        596
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\nu \\big ( \\phi ( s , a ) \\big ) \\triangleq g \\big ( \\phi ( s , a ) \\big ) + \\gamma \\mathbb { E } _ { s ^ { \\prime } \\sim T ( s , a ) } \\Big [ \\nu \\big ( \\phi ( s ^ { \\prime } , a ^ { \\prime } ) \\big ) \\Big ] , \\forall ( s , a ) \\in \\mathcal { S } \\times \\mathcal { A } .",
+                                    "type": "interline_equation",
+                                    "image_path": "18aa3fba76be38e90e93d6e50a0189bfe6bb43416fb2cfa8334b5cd0ee563228.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 35,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                156,
+                                568,
+                                455,
+                                577.3333333333334
+                            ],
+                            "spans": [],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                156,
+                                577.3333333333334,
+                                455,
+                                586.6666666666667
+                            ],
+                            "spans": [],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                156,
+                                586.6666666666667,
+                                455,
+                                596.0000000000001
+                            ],
+                            "spans": [],
+                            "index": 36
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        599,
+                        240,
+                        610
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                599,
+                                241,
+                                612
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        599,
+                                        241,
+                                        612
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Then we can rewrite the IPM as:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        }
+                    ],
+                    "index": 37
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        116,
+                        612,
+                        481,
+                        708
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                116,
+                                612,
+                                481,
+                                708
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        612,
+                                        481,
+                                        708
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\begin{array} { r l } & { \\displaystyle \\mathrm { I P M } _ { \\mathcal { G } } ( d _ { \\phi } ^ { \\pi } , d _ { \\phi } ^ { \\mathcal { D } } ) = \\underset { g \\in \\mathcal { G } } { \\operatorname* { s u p } } \\left| \\mathbb { E } _ { ( s , a ) \\sim d ^ { \\pi } } \\Big [ g \\big ( \\phi ( s , a ) \\big ) \\Big ] - \\mathbb { E } _ { ( s , a ) \\sim d ^ { \\mathcal { D } } } \\Big [ g \\big ( \\phi ( s , a ) \\big ) \\Big ] \\right| } \\\\ & { \\displaystyle \\quad = \\operatorname* { s u p } _ { \\nu \\in \\mathcal { F } } \\left| ( 1 - \\gamma ) \\mathbb { E } \\qquad \\Big [ \\nu \\big ( \\phi ( s , a ) \\big ) \\Big ] - \\mathbb { E } _ { ( s , a , s ^ { \\prime } ) \\sim d ^ { \\mathcal { D } } } \\Big [ \\nu \\big ( \\phi ( s , a ) \\big ) - \\gamma \\nu \\big ( \\phi ( s ^ { \\prime } , a ^ { \\prime } ) \\big ) \\Big ] \\right| , } \\\\ & { \\displaystyle \\quad \\quad \\quad \\quad \\quad \\quad \\quad \\quad \\quad \\quad \\quad \\quad \\quad \\quad \\quad \\quad \\quad \\quad \\quad \\quad \\quad \\quad \\quad \\quad \\quad \\quad \\quad \\quad \\quad \\quad \\quad \\quad a ^ { \\prime } \\sim \\pi ^ { \\prime } } \\end{array}",
+                                    "type": "interline_equation",
+                                    "image_path": "70a7132ac2d1c0745a05676ac0d5bdbaa8b1b9e9f9212c8bae7665f886215dc3.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 39,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                116,
+                                612,
+                                481,
+                                644.0
+                            ],
+                            "spans": [],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                644.0,
+                                481,
+                                676.0
+                            ],
+                            "spans": [],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                676.0,
+                                481,
+                                708.0
+                            ],
+                            "spans": [],
+                            "index": 40
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        709,
+                        504,
+                        733
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                708,
+                                506,
+                                722
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        708,
+                                        419,
+                                        722
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In other words, we are now taking a supremum over the new function class",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        420,
+                                        710,
+                                        429,
+                                        720
+                                    ],
+                                    "score": 0.84,
+                                    "content": "\\mathcal { F }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        429,
+                                        708,
+                                        506,
+                                        722
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", which captures a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                720,
+                                505,
+                                733
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        720,
+                                        317,
+                                        733
+                                    ],
+                                    "score": 1.0,
+                                    "content": "function that returns the expected discounted sum of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        318,
+                                        721,
+                                        366,
+                                        732
+                                    ],
+                                    "score": 0.92,
+                                    "content": "g ( \\phi ( s _ { t } , a _ { t } ) )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        366,
+                                        720,
+                                        450,
+                                        733
+                                    ],
+                                    "score": 1.0,
+                                    "content": "following the policy",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        451,
+                                        722,
+                                        458,
+                                        730
+                                    ],
+                                    "score": 0.77,
+                                    "content": "\\pi",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        458,
+                                        720,
+                                        505,
+                                        733
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in an MDP",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        }
+                    ],
+                    "index": 41.5
+                }
+            ],
+            "page_idx": 3,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        107,
+                        26,
+                        292,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                25,
+                                293,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        25,
+                                        293,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Published as a conference paper at ICLR 2021",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        302,
+                        751,
+                        309,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                301,
+                                750,
+                                310,
+                                762
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        301,
+                                        750,
+                                        310,
+                                        762
+                                    ],
+                                    "score": 1.0,
+                                    "content": "4",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        82,
+                        505,
+                        116
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                82,
+                                505,
+                                96
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        82,
+                                        133,
+                                        96
+                                    ],
+                                    "score": 1.0,
+                                    "content": "where",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        133,
+                                        82,
+                                        195,
+                                        95
+                                    ],
+                                    "score": 0.92,
+                                    "content": "c _ { R } = 2 ( 1 - \\gamma )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        195,
+                                        82,
+                                        213,
+                                        96
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        213,
+                                        83,
+                                        269,
+                                        94
+                                    ],
+                                    "score": 0.92,
+                                    "content": "c _ { T } = 2 \\gamma r _ { \\operatorname* { m a x } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        269,
+                                        82,
+                                        505,
+                                        96
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Then, we start by restating the simulation lemma (Kearns",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                93,
+                                505,
+                                106
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        93,
+                                        505,
+                                        106
+                                    ],
+                                    "score": 1.0,
+                                    "content": "& Singh, 2002) to bound the policy evaluation error in terms of the point-wise model loss. The proof",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                104,
+                                218,
+                                117
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        104,
+                                        218,
+                                        117
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is available in Appendix A.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        }
+                    ],
+                    "index": 1,
+                    "bbox_fs": [
+                        105,
+                        82,
+                        505,
+                        117
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        119,
+                        504,
+                        146
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                118,
+                                505,
+                                134
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        118,
+                                        226,
+                                        134
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Lemma 4.1. Given an MDP",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        226,
+                                        121,
+                                        238,
+                                        131
+                                    ],
+                                    "score": 0.52,
+                                    "content": "M",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        239,
+                                        118,
+                                        306,
+                                        134
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and its estimate",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        306,
+                                        118,
+                                        318,
+                                        131
+                                    ],
+                                    "score": 0.86,
+                                    "content": "\\widehat { M }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        318,
+                                        118,
+                                        479,
+                                        134
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with a bijective representation function",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        479,
+                                        122,
+                                        487,
+                                        133
+                                    ],
+                                    "score": 0.79,
+                                    "content": "\\phi _ { : }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        487,
+                                        118,
+                                        505,
+                                        134
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", i.e.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                107,
+                                131,
+                                460,
+                                148
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        131,
+                                        213,
+                                        146
+                                    ],
+                                    "score": 0.92,
+                                    "content": "( \\widehat { T } , \\widehat { R } ) = \\langle \\widehat { T } _ { z } \\circ \\phi , \\widehat { R } _ { z } \\circ \\phi \\rangle",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        213,
+                                        131,
+                                        371,
+                                        148
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", the policy evaluation error of a policy",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        371,
+                                        136,
+                                        378,
+                                        144
+                                    ],
+                                    "score": 0.63,
+                                    "content": "\\pi",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        379,
+                                        131,
+                                        460,
+                                        148
+                                    ],
+                                    "score": 1.0,
+                                    "content": "can be bounded by:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        }
+                    ],
+                    "index": 3.5,
+                    "bbox_fs": [
+                        105,
+                        118,
+                        505,
+                        148
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        222,
+                        149,
+                        388,
+                        171
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                222,
+                                149,
+                                388,
+                                171
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        222,
+                                        149,
+                                        388,
+                                        171
+                                    ],
+                                    "score": 0.94,
+                                    "content": "\\begin{array} { r } { \\left| R ^ { \\pi } - \\widehat { R } ^ { \\pi } \\right| \\leq \\mathbb { E } _ { ( s , a ) \\sim d ^ { \\pi } } \\left[ \\mathcal { E } _ { \\phi , \\widehat { R } _ { z } , \\widehat { T } _ { z } } ( s , a ) \\right] } \\end{array}",
+                                    "type": "interline_equation",
+                                    "image_path": "be4430fb6e26dd6d13dc5a5eadf0c4ec02c96ae654aff21a36528c073cefe5c6.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 5,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                222,
+                                149,
+                                388,
+                                171
+                            ],
+                            "spans": [],
+                            "index": 5
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        178,
+                        505,
+                        246
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                178,
+                                505,
+                                192
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        178,
+                                        505,
+                                        192
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The Lemma 4.1 has a natural interpretation: if the model error is small in the states frequently",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                190,
+                                506,
+                                204
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        190,
+                                        231,
+                                        204
+                                    ],
+                                    "score": 1.0,
+                                    "content": "visited by following the policy",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        232,
+                                        193,
+                                        239,
+                                        201
+                                    ],
+                                    "score": 0.76,
+                                    "content": "\\pi",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        239,
+                                        190,
+                                        506,
+                                        204
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", the resultant policy evaluation error will also be small. However,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                202,
+                                505,
+                                213
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        202,
+                                        505,
+                                        213
+                                    ],
+                                    "score": 1.0,
+                                    "content": "minimizing the RHS of Eq. (1) in the off-policy evaluation (OPE) task is generally intractable since",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                213,
+                                505,
+                                225
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        213,
+                                        169,
+                                        225
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the distribution",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        170,
+                                        213,
+                                        181,
+                                        222
+                                    ],
+                                    "score": 0.87,
+                                    "content": "d ^ { \\pi }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        182,
+                                        213,
+                                        505,
+                                        225
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is not directly accessible. Therefore, the common practice has been to construct",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                223,
+                                506,
+                                236
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        223,
+                                        248,
+                                        236
+                                    ],
+                                    "score": 1.0,
+                                    "content": "a maximum-likelihood MDP using",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        248,
+                                        224,
+                                        257,
+                                        233
+                                    ],
+                                    "score": 0.82,
+                                    "content": "\\mathcal { D }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        258,
+                                        223,
+                                        506,
+                                        236
+                                    ],
+                                    "score": 1.0,
+                                    "content": "while ignoring the distribution shift, but its OPE performance",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                235,
+                                180,
+                                247
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        235,
+                                        180,
+                                        247
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is not guaranteed.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        }
+                    ],
+                    "index": 8.5,
+                    "bbox_fs": [
+                        105,
+                        178,
+                        506,
+                        247
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        250,
+                        505,
+                        297
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                250,
+                                505,
+                                264
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        250,
+                                        505,
+                                        264
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Instead, we will derive a tractable upper bound on the policy evaluation error by eliminating the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                262,
+                                504,
+                                274
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        262,
+                                        195,
+                                        274
+                                    ],
+                                    "score": 1.0,
+                                    "content": "direct dependence on",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        195,
+                                        263,
+                                        207,
+                                        272
+                                    ],
+                                    "score": 0.86,
+                                    "content": "d ^ { \\pi }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        207,
+                                        262,
+                                        504,
+                                        274
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in Eq. (1). To this end, we adopt the distance metric between two distri-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                272,
+                                505,
+                                286
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        272,
+                                        221,
+                                        286
+                                    ],
+                                    "score": 1.0,
+                                    "content": "butions over representations",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        221,
+                                        273,
+                                        232,
+                                        286
+                                    ],
+                                    "score": 0.88,
+                                    "content": "d _ { \\phi } ^ { \\pi }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        233,
+                                        272,
+                                        250,
+                                        286
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        251,
+                                        272,
+                                        264,
+                                        286
+                                    ],
+                                    "score": 0.9,
+                                    "content": "d _ { \\phi } ^ { \\mathcal { D } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        264,
+                                        272,
+                                        505,
+                                        286
+                                    ],
+                                    "score": 1.0,
+                                    "content": "that can bound their difference in expectations, which is the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                285,
+                                307,
+                                297
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        285,
+                                        307,
+                                        297
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Integral Probability Metric (IPM) (Muller, 1997): ¨",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        }
+                    ],
+                    "index": 13.5,
+                    "bbox_fs": [
+                        104,
+                        250,
+                        505,
+                        297
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        210,
+                        299,
+                        401,
+                        321
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                210,
+                                299,
+                                401,
+                                321
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        210,
+                                        299,
+                                        401,
+                                        321
+                                    ],
+                                    "score": 0.95,
+                                    "content": "\\mathrm { I P M } _ { \\mathcal { G } } ( p , q ) = \\operatorname* { s u p } _ { g \\in \\mathcal { G } } \\big | \\mathbb { E } _ { z \\sim p } [ g ( z ) ] - \\mathbb { E } _ { z \\sim q } [ g ( z ) ] \\big | .",
+                                    "type": "interline_equation",
+                                    "image_path": "b86cdd7b9a7b014a45536a084ffd03b37556682b8e9c250e180c7355c8694f81.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 16,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                210,
+                                299,
+                                401,
+                                321
+                            ],
+                            "spans": [],
+                            "index": 16
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        325,
+                        505,
+                        348
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                325,
+                                505,
+                                336
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        325,
+                                        218,
+                                        336
+                                    ],
+                                    "score": 1.0,
+                                    "content": "where particular choices of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        218,
+                                        326,
+                                        226,
+                                        336
+                                    ],
+                                    "score": 0.84,
+                                    "content": "\\mathcal { G }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        226,
+                                        325,
+                                        505,
+                                        336
+                                    ],
+                                    "score": 1.0,
+                                    "content": "make the IPM equivalent to different well-known distances of distri-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                336,
+                                466,
+                                349
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        336,
+                                        466,
+                                        349
+                                    ],
+                                    "score": 1.0,
+                                    "content": "butions, e.g. total variation distance or Wasserstein distance (Sriperumbudur et al., 2009).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        }
+                    ],
+                    "index": 17.5,
+                    "bbox_fs": [
+                        105,
+                        325,
+                        505,
+                        349
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        351,
+                        505,
+                        393
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                349,
+                                505,
+                                366
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        349,
+                                        235,
+                                        366
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Theorem 4.2. Given an MDP",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        235,
+                                        353,
+                                        247,
+                                        362
+                                    ],
+                                    "score": 0.59,
+                                    "content": "M",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        247,
+                                        349,
+                                        317,
+                                        366
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and its estimate",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        317,
+                                        350,
+                                        329,
+                                        362
+                                    ],
+                                    "score": 0.86,
+                                    "content": "\\widehat { M }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        329,
+                                        349,
+                                        494,
+                                        366
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with a bijective representation function",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        495,
+                                        353,
+                                        501,
+                                        364
+                                    ],
+                                    "score": 0.76,
+                                    "content": "\\phi",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        501,
+                                        349,
+                                        505,
+                                        366
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                362,
+                                505,
+                                398
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        362,
+                                        124,
+                                        379
+                                    ],
+                                    "score": 1.0,
+                                    "content": "i.e.",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        107,
+                                        378,
+                                        182,
+                                        390
+                                    ],
+                                    "score": 0.91,
+                                    "content": "{ \\mathcal { G } } \\triangleq \\{ g : { \\mathcal { Z } } \\to \\mathbb { R } \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        124,
+                                        363,
+                                        235,
+                                        377
+                                    ],
+                                    "score": 0.91,
+                                    "content": "( \\widehat { T } , \\widehat { R } ) = \\langle \\widehat { T } _ { z } \\circ \\phi , \\widehat { R } _ { z } \\circ \\phi \\rangle",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        182,
+                                        370,
+                                        222,
+                                        398
+                                    ],
+                                    "score": 1.0,
+                                    "content": "b such that",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        222,
+                                        377,
+                                        325,
+                                        393
+                                    ],
+                                    "score": 0.94,
+                                    "content": "\\begin{array} { r } { \\frac { 1 } { B _ { \\phi } } \\mathcal { E } _ { \\phi , \\widehat { R } _ { z } , \\widehat { T } _ { z } } \\left( \\phi ^ { - 1 } ( \\cdot ) \\right) \\in \\mathcal { G } } \\end{array}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        235,
+                                        362,
+                                        384,
+                                        379
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", assume that there exists a constant",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        325,
+                                        370,
+                                        412,
+                                        398
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Then, for any policy",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        384,
+                                        365,
+                                        419,
+                                        378
+                                    ],
+                                    "score": 0.93,
+                                    "content": "B _ { \\phi } > 0",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        412,
+                                        381,
+                                        419,
+                                        388
+                                    ],
+                                    "score": 0.73,
+                                    "content": "\\pi",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        419,
+                                        362,
+                                        438,
+                                        379
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        419,
+                                        370,
+                                        426,
+                                        398
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        439,
+                                        367,
+                                        445,
+                                        375
+                                    ],
+                                    "score": 0.26,
+                                    "content": "a",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        445,
+                                        362,
+                                        505,
+                                        379
+                                    ],
+                                    "score": 1.0,
+                                    "content": "function class",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        }
+                    ],
+                    "index": 19.5,
+                    "bbox_fs": [
+                        105,
+                        349,
+                        505,
+                        398
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        179,
+                        397,
+                        431,
+                        418
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                179,
+                                397,
+                                431,
+                                418
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        179,
+                                        397,
+                                        431,
+                                        418
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\left| R ^ { \\pi } - \\widehat { R } ^ { \\pi } \\right| \\leq \\mathbb { E } _ { ( s , a ) \\sim d ^ { \\mathcal { D } } } \\Big [ \\mathcal { E } _ { \\phi , \\widehat { R } _ { z } , \\widehat { T } _ { z } } ( s , a ) \\Big ] + B _ { \\phi } \\mathrm { I P M } _ { \\mathcal { G } } ( d _ { \\phi } ^ { \\pi } , d _ { \\phi } ^ { \\mathcal { D } } )",
+                                    "type": "interline_equation",
+                                    "image_path": "50332b075b998d81ff520371d10d176b4273f51b2bb0d2bb4ffa9954584b2c6d.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 21,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                179,
+                                397,
+                                431,
+                                418
+                            ],
+                            "spans": [],
+                            "index": 21
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        426,
+                        505,
+                        518
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                426,
+                                506,
+                                439
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        426,
+                                        506,
+                                        439
+                                    ],
+                                    "score": 1.0,
+                                    "content": "This theorem is an adaptation of Lemma 1 of Shalit et al. (2017) to an infinite horizon model-based",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                435,
+                                508,
+                                455
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        435,
+                                        335,
+                                        455
+                                    ],
+                                    "score": 1.0,
+                                    "content": "policy evaluation and can be derived by the definition of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        335,
+                                        438,
+                                        396,
+                                        452
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\mathrm { I P M } _ { \\mathcal { G } } ( d _ { \\phi } ^ { \\pi } , d _ { \\phi } ^ { D } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        396,
+                                        435,
+                                        508,
+                                        455
+                                    ],
+                                    "score": 1.0,
+                                    "content": "since it serves as an upper",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                450,
+                                505,
+                                462
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        450,
+                                        348,
+                                        462
+                                    ],
+                                    "score": 1.0,
+                                    "content": "bound of the difference in the expectations of any function in",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        348,
+                                        452,
+                                        356,
+                                        461
+                                    ],
+                                    "score": 0.82,
+                                    "content": "\\mathcal { G }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        356,
+                                        450,
+                                        505,
+                                        462
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". The first term in Eq. (3) corresponds",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                460,
+                                505,
+                                473
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        461,
+                                        249,
+                                        473
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to the fitness to the data following",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        249,
+                                        460,
+                                        262,
+                                        471
+                                    ],
+                                    "score": 0.87,
+                                    "content": "d ^ { \\mathcal { D } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        262,
+                                        461,
+                                        505,
+                                        473
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", while the second term serves as a regularizer. To see this,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                471,
+                                505,
+                                483
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        471,
+                                        505,
+                                        483
+                                    ],
+                                    "score": 1.0,
+                                    "content": "minimizing the second term would yield a near-constant representation function, which would be",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                483,
+                                505,
+                                495
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        483,
+                                        505,
+                                        495
+                                    ],
+                                    "score": 1.0,
+                                    "content": "bad for the first term since it cannot distinguish states and actions well enough. It shows a natural",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                494,
+                                506,
+                                506
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        494,
+                                        506,
+                                        506
+                                    ],
+                                    "score": 1.0,
+                                    "content": "trade-off between optimizing the model that fits data better and learning the representation that is",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                502,
+                                253,
+                                521
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        502,
+                                        205,
+                                        521
+                                    ],
+                                    "score": 1.0,
+                                    "content": "invariant with respect to",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        205,
+                                        505,
+                                        217,
+                                        519
+                                    ],
+                                    "score": 0.89,
+                                    "content": "d _ { \\phi } ^ { \\pi }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        217,
+                                        502,
+                                        235,
+                                        521
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        235,
+                                        504,
+                                        248,
+                                        519
+                                    ],
+                                    "score": 0.91,
+                                    "content": "d _ { \\phi } ^ { \\mathcal { D } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        248,
+                                        502,
+                                        253,
+                                        521
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        }
+                    ],
+                    "index": 25.5,
+                    "bbox_fs": [
+                        104,
+                        426,
+                        508,
+                        521
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        522,
+                        505,
+                        567
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                523,
+                                505,
+                                534
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        523,
+                                        461,
+                                        534
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Nevertheless, RHS of Eq. (3) still cannot be evaluated naively due to its dependence on",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        462,
+                                        523,
+                                        473,
+                                        533
+                                    ],
+                                    "score": 0.84,
+                                    "content": "d ^ { \\pi }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        473,
+                                        523,
+                                        505,
+                                        534
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in esti-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                533,
+                                505,
+                                545
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        533,
+                                        505,
+                                        545
+                                    ],
+                                    "score": 1.0,
+                                    "content": "mating the IPM. We address this challenge via a change of variable, which is known as a DualDICE",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                544,
+                                505,
+                                557
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        544,
+                                        257,
+                                        557
+                                    ],
+                                    "score": 1.0,
+                                    "content": "trick (Nachum et al., 2019a). Define",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        257,
+                                        545,
+                                        307,
+                                        555
+                                    ],
+                                    "score": 0.9,
+                                    "content": "\\nu : \\mathcal { Z } \\to \\mathbb { R }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        307,
+                                        544,
+                                        505,
+                                        557
+                                    ],
+                                    "score": 1.0,
+                                    "content": "as an arbitrary function of state-action pairs that",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                555,
+                                144,
+                                567
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        555,
+                                        144,
+                                        567
+                                    ],
+                                    "score": 1.0,
+                                    "content": "satisfies:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        }
+                    ],
+                    "index": 31.5,
+                    "bbox_fs": [
+                        105,
+                        523,
+                        505,
+                        567
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        156,
+                        568,
+                        455,
+                        596
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                156,
+                                568,
+                                455,
+                                596
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        156,
+                                        568,
+                                        455,
+                                        596
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\nu \\big ( \\phi ( s , a ) \\big ) \\triangleq g \\big ( \\phi ( s , a ) \\big ) + \\gamma \\mathbb { E } _ { s ^ { \\prime } \\sim T ( s , a ) } \\Big [ \\nu \\big ( \\phi ( s ^ { \\prime } , a ^ { \\prime } ) \\big ) \\Big ] , \\forall ( s , a ) \\in \\mathcal { S } \\times \\mathcal { A } .",
+                                    "type": "interline_equation",
+                                    "image_path": "18aa3fba76be38e90e93d6e50a0189bfe6bb43416fb2cfa8334b5cd0ee563228.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 35,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                156,
+                                568,
+                                455,
+                                577.3333333333334
+                            ],
+                            "spans": [],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                156,
+                                577.3333333333334,
+                                455,
+                                586.6666666666667
+                            ],
+                            "spans": [],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                156,
+                                586.6666666666667,
+                                455,
+                                596.0000000000001
+                            ],
+                            "spans": [],
+                            "index": 36
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        599,
+                        240,
+                        610
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                599,
+                                241,
+                                612
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        599,
+                                        241,
+                                        612
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Then we can rewrite the IPM as:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        }
+                    ],
+                    "index": 37,
+                    "bbox_fs": [
+                        106,
+                        599,
+                        241,
+                        612
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        116,
+                        612,
+                        481,
+                        708
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                116,
+                                612,
+                                481,
+                                708
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        612,
+                                        481,
+                                        708
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\begin{array} { r l } & { \\displaystyle \\mathrm { I P M } _ { \\mathcal { G } } ( d _ { \\phi } ^ { \\pi } , d _ { \\phi } ^ { \\mathcal { D } } ) = \\underset { g \\in \\mathcal { G } } { \\operatorname* { s u p } } \\left| \\mathbb { E } _ { ( s , a ) \\sim d ^ { \\pi } } \\Big [ g \\big ( \\phi ( s , a ) \\big ) \\Big ] - \\mathbb { E } _ { ( s , a ) \\sim d ^ { \\mathcal { D } } } \\Big [ g \\big ( \\phi ( s , a ) \\big ) \\Big ] \\right| } \\\\ & { \\displaystyle \\quad = \\operatorname* { s u p } _ { \\nu \\in \\mathcal { F } } \\left| ( 1 - \\gamma ) \\mathbb { E } \\qquad \\Big [ \\nu \\big ( \\phi ( s , a ) \\big ) \\Big ] - \\mathbb { E } _ { ( s , a , s ^ { \\prime } ) \\sim d ^ { \\mathcal { D } } } \\Big [ \\nu \\big ( \\phi ( s , a ) \\big ) - \\gamma \\nu \\big ( \\phi ( s ^ { \\prime } , a ^ { \\prime } ) \\big ) \\Big ] \\right| , } \\\\ & { \\displaystyle \\quad \\quad \\quad \\quad \\quad \\quad \\quad \\quad \\quad \\quad \\quad \\quad \\quad \\quad \\quad \\quad \\quad \\quad \\quad \\quad \\quad \\quad \\quad \\quad \\quad \\quad \\quad \\quad \\quad \\quad \\quad \\quad a ^ { \\prime } \\sim \\pi ^ { \\prime } } \\end{array}",
+                                    "type": "interline_equation",
+                                    "image_path": "70a7132ac2d1c0745a05676ac0d5bdbaa8b1b9e9f9212c8bae7665f886215dc3.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 39,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                116,
+                                612,
+                                481,
+                                644.0
+                            ],
+                            "spans": [],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                644.0,
+                                481,
+                                676.0
+                            ],
+                            "spans": [],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                676.0,
+                                481,
+                                708.0
+                            ],
+                            "spans": [],
+                            "index": 40
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        709,
+                        504,
+                        733
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                708,
+                                506,
+                                722
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        708,
+                                        419,
+                                        722
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In other words, we are now taking a supremum over the new function class",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        420,
+                                        710,
+                                        429,
+                                        720
+                                    ],
+                                    "score": 0.84,
+                                    "content": "\\mathcal { F }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        429,
+                                        708,
+                                        506,
+                                        722
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", which captures a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                720,
+                                505,
+                                733
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        720,
+                                        317,
+                                        733
+                                    ],
+                                    "score": 1.0,
+                                    "content": "function that returns the expected discounted sum of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        318,
+                                        721,
+                                        366,
+                                        732
+                                    ],
+                                    "score": 0.92,
+                                    "content": "g ( \\phi ( s _ { t } , a _ { t } ) )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        366,
+                                        720,
+                                        450,
+                                        733
+                                    ],
+                                    "score": 1.0,
+                                    "content": "following the policy",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        451,
+                                        722,
+                                        458,
+                                        730
+                                    ],
+                                    "score": 0.77,
+                                    "content": "\\pi",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        458,
+                                        720,
+                                        505,
+                                        733
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in an MDP",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        }
+                    ],
+                    "index": 41.5,
+                    "bbox_fs": [
+                        105,
+                        708,
+                        506,
+                        733
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        82,
+                        504,
+                        105
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                81,
+                                505,
+                                96
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        83,
+                                        119,
+                                        92
+                                    ],
+                                    "score": 0.72,
+                                    "content": "M",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        119,
+                                        81,
+                                        242,
+                                        96
+                                    ],
+                                    "score": 1.0,
+                                    "content": "given an initial representation",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        243,
+                                        85,
+                                        249,
+                                        92
+                                    ],
+                                    "score": 0.73,
+                                    "content": "z",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        249,
+                                        81,
+                                        394,
+                                        96
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". While it is now difficult to choose",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        395,
+                                        83,
+                                        405,
+                                        92
+                                    ],
+                                    "score": 0.82,
+                                    "content": "\\mathcal { F }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        405,
+                                        81,
+                                        428,
+                                        96
+                                    ],
+                                    "score": 1.0,
+                                    "content": "from",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        428,
+                                        83,
+                                        435,
+                                        93
+                                    ],
+                                    "score": 0.74,
+                                    "content": "\\mathcal { G }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        436,
+                                        81,
+                                        505,
+                                        96
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", Eq. (3) still can",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                93,
+                                353,
+                                105
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        93,
+                                        340,
+                                        105
+                                    ],
+                                    "score": 1.0,
+                                    "content": "be kept valid by using a sufficiently rich function class for",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        340,
+                                        94,
+                                        350,
+                                        104
+                                    ],
+                                    "score": 0.83,
+                                    "content": "\\mathcal { F }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        350,
+                                        93,
+                                        353,
+                                        105
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        }
+                    ],
+                    "index": 0.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        110,
+                        505,
+                        144
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                110,
+                                505,
+                                122
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        110,
+                                        209,
+                                        122
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In this work, we choose",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        209,
+                                        111,
+                                        219,
+                                        121
+                                    ],
+                                    "score": 0.84,
+                                    "content": "\\mathcal { F }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        219,
+                                        110,
+                                        505,
+                                        122
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to be the family of functions in the unit ball in a reproducing kernel",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                121,
+                                505,
+                                134
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        121,
+                                        198,
+                                        134
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Hilbert space (RKHS)",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        199,
+                                        122,
+                                        213,
+                                        132
+                                    ],
+                                    "score": 0.89,
+                                    "content": "\\mathcal { H } _ { k }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        213,
+                                        121,
+                                        278,
+                                        134
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with the kernel",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        278,
+                                        122,
+                                        285,
+                                        132
+                                    ],
+                                    "score": 0.81,
+                                    "content": "k",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        285,
+                                        121,
+                                        505,
+                                        134
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", which allows the following closed-form formula (see",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                132,
+                                258,
+                                145
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        132,
+                                        258,
+                                        145
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Lemma A.3 in Appendix for details):",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        }
+                    ],
+                    "index": 3
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        111,
+                        146,
+                        507,
+                        221
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                111,
+                                146,
+                                507,
+                                221
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        111,
+                                        146,
+                                        507,
+                                        221
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\begin{array} { r l r } { \\mathrm { P M } _ { \\mathcal { G } } ( d _ { \\phi } ^ { \\pi } , d _ { \\phi } ^ { \\mathcal { D } } ) ^ { 2 } = \\mathbb { E } _ { s _ { 0 } \\sim d _ { 0 } , a _ { 0 } \\sim \\pi ( s _ { 0 } ) , ( s , a , s ^ { \\prime } ) \\sim d ^ { \\mathcal { D } } , a ^ { \\prime } \\sim \\pi ( s ^ { \\prime } ) } } & { } & { ( 5 ) } \\\\ { \\bar { s } _ { 0 } \\sim d _ { 0 } , \\bar { u } _ { 0 } \\sim d _ { 0 } ) , ( \\bar { s } _ { 0 } , \\bar { u } , s ^ { \\prime } ) \\sim d ^ { \\mathcal { D } } , a ^ { \\prime } \\sim \\pi ( \\bar { s } ^ { \\prime } ) } & { } & { } \\\\ { k \\big ( \\phi ( s , a ) , \\phi ( \\bar { s } , \\bar { a } ) \\big ) + ( 1 - \\gamma ) ^ { 2 } k \\big ( \\phi ( s _ { 0 } , a _ { 0 } ) , \\phi ( \\bar { s } _ { 0 } , \\bar { a } _ { 0 } ) \\big ) + \\gamma ^ { 2 } k \\big ( \\phi ( s ^ { \\prime } , a ^ { \\prime } ) , \\phi ( \\bar { s } ^ { \\prime } , \\bar { a } ^ { \\prime } ) \\big ) } & { } & { } \\\\ { - 2 ( 1 - \\gamma ) k \\big ( \\phi ( s _ { 0 } , a _ { 0 } ) , \\phi ( \\bar { s } , \\bar { a } ) \\big ) - 2 \\gamma k \\big ( \\phi ( s ^ { \\prime } , a ^ { \\prime } ) , \\phi ( \\bar { s } , \\bar { a } ) \\big ) + 2 \\gamma ( 1 - \\gamma ) k \\big ( \\phi ( s ^ { \\prime } , a ^ { \\prime } ) , \\phi ( \\bar { s } _ { 0 } , \\bar { a } _ { 0 } ) \\big ) \\bigg ] . } & { } & \\end{array}",
+                                    "type": "interline_equation",
+                                    "image_path": "1953b31a1cd60337242d69b916f242145a0e39f20e0603f1049a6db643290d90.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 6,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                111,
+                                146,
+                                507,
+                                171.0
+                            ],
+                            "spans": [],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                111,
+                                171.0,
+                                507,
+                                196.0
+                            ],
+                            "spans": [],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                111,
+                                196.0,
+                                507,
+                                221.0
+                            ],
+                            "spans": [],
+                            "index": 7
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        223,
+                        505,
+                        268
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                223,
+                                505,
+                                236
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        223,
+                                        505,
+                                        236
+                                    ],
+                                    "score": 1.0,
+                                    "content": "This completes our derivation for the tractable upper bound of policy evaluation error (Eq. (3)),",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                233,
+                                505,
+                                248
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        233,
+                                        219,
+                                        248
+                                    ],
+                                    "score": 1.0,
+                                    "content": "whose direct dependence on",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        220,
+                                        235,
+                                        231,
+                                        245
+                                    ],
+                                    "score": 0.86,
+                                    "content": "d ^ { \\pi }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        231,
+                                        233,
+                                        505,
+                                        248
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is eliminated by Eq. (5). Finally, we can train a model by minimizing",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                244,
+                                506,
+                                260
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        244,
+                                        506,
+                                        260
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the upper bound that encourages us to learn balanced representation while improving data fitness,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                257,
+                                369,
+                                268
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        257,
+                                        369,
+                                        268
+                                    ],
+                                    "score": 1.0,
+                                    "content": "where the trained model can readily provide a model-based OPE.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        }
+                    ],
+                    "index": 9.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        272,
+                        505,
+                        323
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                272,
+                                506,
+                                287
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        272,
+                                        126,
+                                        287
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        126,
+                                        272,
+                                        192,
+                                        287
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\mathrm { I P M } _ { \\mathcal { G } } ( d _ { \\phi } ^ { \\pi } , d _ { \\phi } ^ { D } ) ^ { 2 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        192,
+                                        272,
+                                        506,
+                                        287
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in Eq. (5) can be estimated via finite random samples, and we denote its",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                286,
+                                506,
+                                302
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        288,
+                                        221,
+                                        302
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sampled-based estimator as",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        222,
+                                        286,
+                                        281,
+                                        302
+                                    ],
+                                    "score": 0.94,
+                                    "content": "\\widehat { \\mathrm { I P M } } ( d _ { \\phi } ^ { \\pi } , d _ { \\phi } ^ { D } ) ^ { 2 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        282,
+                                        288,
+                                        506,
+                                        302
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". We show in the following that, a valid upper bound",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                300,
+                                506,
+                                313
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        300,
+                                        506,
+                                        313
+                                    ],
+                                    "score": 1.0,
+                                    "content": "can be established based on the sample-based estimators instead of the exact terms in the RHS of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                312,
+                                238,
+                                324
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        312,
+                                        238,
+                                        324
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Eq. (3) under certain conditions.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        }
+                    ],
+                    "index": 13.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        325,
+                        505,
+                        408
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                324,
+                                506,
+                                340
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        324,
+                                        249,
+                                        340
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Theorem 4.3. Given an MDP",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        249,
+                                        328,
+                                        261,
+                                        338
+                                    ],
+                                    "score": 0.62,
+                                    "content": "M",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        261,
+                                        324,
+                                        325,
+                                        340
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", its estimate",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        325,
+                                        325,
+                                        338,
+                                        338
+                                    ],
+                                    "score": 0.83,
+                                    "content": "\\widehat { M }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        338,
+                                        324,
+                                        506,
+                                        340
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with a bijective representation func-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                337,
+                                506,
+                                353
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        337,
+                                        128,
+                                        353
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tion",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        128,
+                                        340,
+                                        136,
+                                        352
+                                    ],
+                                    "score": 0.76,
+                                    "content": "\\phi",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        136,
+                                        337,
+                                        173,
+                                        353
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", i.e.",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        173,
+                                        338,
+                                        308,
+                                        352
+                                    ],
+                                    "score": 0.91,
+                                    "content": "( \\widehat { T } , \\widehat { R } ) = \\langle \\widehat { T } _ { z } \\circ \\phi , \\widehat { R } _ { z } \\circ \\phi \\rangle",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        309,
+                                        337,
+                                        387,
+                                        353
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", and an RKHS",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        388,
+                                        340,
+                                        487,
+                                        352
+                                    ],
+                                    "score": 0.63,
+                                    "content": "\\begin{array} { r } { \\mathcal { H } _ { k } \\quad \\subset \\quad ( \\mathcal { Z } \\quad \\to \\quad \\mathbb { R } ) } \\end{array}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        487,
+                                        337,
+                                        506,
+                                        353
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                349,
+                                504,
+                                366
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        349,
+                                        232,
+                                        366
+                                    ],
+                                    "score": 1.0,
+                                    "content": "duced by a universal kernel",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        232,
+                                        352,
+                                        240,
+                                        361
+                                    ],
+                                    "score": 0.74,
+                                    "content": "k",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        240,
+                                        349,
+                                        287,
+                                        366
+                                    ],
+                                    "score": 1.0,
+                                    "content": "such that",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        287,
+                                        351,
+                                        377,
+                                        363
+                                    ],
+                                    "score": 0.89,
+                                    "content": "\\mathrm { s u p } _ { z \\in \\mathcal { Z } } k ( z , z ) ~ = ~ \\bar { k }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        378,
+                                        349,
+                                        440,
+                                        366
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", assume that",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        441,
+                                        351,
+                                        504,
+                                        366
+                                    ],
+                                    "score": 0.9,
+                                    "content": "f _ { \\phi , \\widehat { R } _ { z } , \\widehat { T } _ { z } } \\left( \\widehat { z } \\right) \\ =",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                362,
+                                508,
+                                386
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        364,
+                                        344,
+                                        384
+                                    ],
+                                    "score": 0.9,
+                                    "content": "\\begin{array} { r } { \\mathbb { E } _ { T , \\pi } \\Big [ \\sum _ { t = 0 } ^ { \\infty } \\gamma ^ { t } \\mathcal { E } _ { \\phi , \\widehat { R } _ { z } , \\widehat { T } _ { z } } \\big ( s _ { t } , a _ { t } \\big ) \\Big | \\big ( s _ { 0 } , a _ { 0 } \\big ) = \\phi ^ { - 1 } \\big ( z ) \\Big ] \\in \\mathcal { H } _ { k } } \\end{array}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        344,
+                                        362,
+                                        366,
+                                        386
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        367,
+                                        368,
+                                        451,
+                                        383
+                                    ],
+                                    "score": 0.93,
+                                    "content": "B _ { \\phi } = \\| f _ { \\phi , \\widehat { R } _ { z } , \\widehat { T } _ { z } } \\| _ { \\mathcal { H } _ { k } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        451,
+                                        362,
+                                        508,
+                                        386
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and the loss",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                380,
+                                506,
+                                400
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        380,
+                                        165,
+                                        400
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is bounded by",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        165,
+                                        385,
+                                        293,
+                                        398
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\begin{array} { r } { \\bar { \\mathcal { E } } = \\operatorname* { s u p } _ { s \\in { \\mathcal { S } } , a \\in \\mathcal { A } } \\mathcal { E } _ { \\phi , \\widehat { R } _ { z } , \\widehat { T } _ { z } } ( s , a ) } \\end{array}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        293,
+                                        380,
+                                        423,
+                                        400
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Let n be the number of data in",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        424,
+                                        385,
+                                        432,
+                                        394
+                                    ],
+                                    "score": 0.76,
+                                    "content": "\\mathcal { D }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        433,
+                                        380,
+                                        506,
+                                        400
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". With probability",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                394,
+                                140,
+                                410
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        397,
+                                        135,
+                                        408
+                                    ],
+                                    "score": 0.64,
+                                    "content": "1 - 2 \\delta",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        135,
+                                        394,
+                                        140,
+                                        410
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        }
+                    ],
+                    "index": 18.5
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        111,
+                        409,
+                        506,
+                        445
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                111,
+                                409,
+                                506,
+                                445
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        111,
+                                        409,
+                                        506,
+                                        445
+                                    ],
+                                    "score": 0.89,
+                                    "content": "R ^ { \\pi } - { \\widehat R } ^ { \\pi } \\Big | \\leq \\frac { 1 } { n } \\sum _ { ( s , a ) \\in { \\mathcal D } } \\mathcal { E } _ { \\phi , { \\widehat R } _ { z } , { \\widehat T } _ { z } } ( s , a ) + B _ { \\phi } \\widehat { \\Gamma \\mathrm { M } } \\big ( d _ { \\phi } ^ { \\pi } , d _ { \\phi } ^ { { D } } \\big ) + \\sqrt { \\frac { \\bar { \\mathcal E } ^ { 2 } } { 2 n } \\log \\frac { 1 } { \\delta } } + B _ { \\phi } \\sqrt { \\frac { \\bar { k } } { n } } \\left( 4 + \\sqrt { 8 \\log \\frac { 3 } { \\delta } } \\right) .",
+                                    "type": "interline_equation",
+                                    "image_path": "1cfc254f1512a52e679a50996bd4958636bc46ff2883210d61add703a1dd3887.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 23,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                111,
+                                409,
+                                506,
+                                421.0
+                            ],
+                            "spans": [],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                111,
+                                421.0,
+                                506,
+                                433.0
+                            ],
+                            "spans": [],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                111,
+                                433.0,
+                                506,
+                                445.0
+                            ],
+                            "spans": [],
+                            "index": 24
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        452,
+                        506,
+                        549
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                452,
+                                506,
+                                466
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        452,
+                                        506,
+                                        466
+                                    ],
+                                    "score": 1.0,
+                                    "content": "This result can be proved by adapting the convergence results of the empirical estimate of the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                464,
+                                506,
+                                477
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        464,
+                                        506,
+                                        477
+                                    ],
+                                    "score": 1.0,
+                                    "content": "MMD (Gretton et al., 2012) and Hoeffding’s inequality (Hoeffding, 1963). With the choice of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                474,
+                                506,
+                                491
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        474,
+                                        147,
+                                        491
+                                    ],
+                                    "score": 1.0,
+                                    "content": "an RKHS",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        147,
+                                        475,
+                                        162,
+                                        487
+                                    ],
+                                    "score": 0.88,
+                                    "content": "\\mathcal { H } _ { k }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        162,
+                                        474,
+                                        251,
+                                        491
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", we can now interpret",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        252,
+                                        476,
+                                        266,
+                                        487
+                                    ],
+                                    "score": 0.89,
+                                    "content": "B _ { \\phi }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        266,
+                                        474,
+                                        345,
+                                        491
+                                    ],
+                                    "score": 1.0,
+                                    "content": "as the RKHS norm",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        345,
+                                        475,
+                                        401,
+                                        489
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\| f _ { \\phi , \\widehat { R } _ { z } , \\widehat { T } _ { z } } \\| _ { \\mathcal { H } _ { k } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        401,
+                                        474,
+                                        506,
+                                        491
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", which captures the mag-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                486,
+                                506,
+                                504
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        486,
+                                        376,
+                                        504
+                                    ],
+                                    "score": 1.0,
+                                    "content": "bnitude and the smoothness of the expected cumulative model loss",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        377,
+                                        489,
+                                        411,
+                                        502
+                                    ],
+                                    "score": 0.91,
+                                    "content": "f _ { \\phi , \\widehat { R } _ { z } , \\widehat { T } _ { z } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        411,
+                                        486,
+                                        506,
+                                        504
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". In general, assuming",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                500,
+                                505,
+                                513
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        500,
+                                        285,
+                                        513
+                                    ],
+                                    "score": 1.0,
+                                    "content": "smooth underlying dynamics, we can expect",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        285,
+                                        501,
+                                        299,
+                                        513
+                                    ],
+                                    "score": 0.9,
+                                    "content": "B _ { \\phi }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        299,
+                                        500,
+                                        505,
+                                        513
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to be small when the model error is small. Further-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                511,
+                                506,
+                                525
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        513,
+                                        170,
+                                        525
+                                    ],
+                                    "score": 1.0,
+                                    "content": "more, although",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        170,
+                                        511,
+                                        177,
+                                        523
+                                    ],
+                                    "score": 0.81,
+                                    "content": "\\bar { k }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        178,
+                                        513,
+                                        414,
+                                        525
+                                    ],
+                                    "score": 1.0,
+                                    "content": "depends on the kernel function we use, we can always let √",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        414,
+                                        511,
+                                        441,
+                                        523
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\bar { k } = 1",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        441,
+                                        513,
+                                        506,
+                                        525
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and subsume it",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                523,
+                                506,
+                                539
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        524,
+                                        124,
+                                        539
+                                    ],
+                                    "score": 1.0,
+                                    "content": "into",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        124,
+                                        526,
+                                        138,
+                                        538
+                                    ],
+                                    "score": 0.9,
+                                    "content": "B _ { \\phi }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        139,
+                                        524,
+                                        276,
+                                        539
+                                    ],
+                                    "score": 1.0,
+                                    "content": "as long as it is bounded, i.e. using",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        277,
+                                        523,
+                                        331,
+                                        538
+                                    ],
+                                    "score": 0.94,
+                                    "content": "\\tilde { B } _ { \\phi } \\triangleq B _ { \\phi } \\sqrt { \\bar { k } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        331,
+                                        524,
+                                        506,
+                                        539
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". In the next section, we develop algorithms",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                537,
+                                289,
+                                550
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        537,
+                                        289,
+                                        550
+                                    ],
+                                    "score": 1.0,
+                                    "content": "based on practical approximations of Eq. (3).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        }
+                    ],
+                    "index": 28.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        559,
+                        505,
+                        654
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                559,
+                                505,
+                                572
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        559,
+                                        505,
+                                        572
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Detailed comparison to RepBM As previously stated, RepBM (Liu et al., 2018b) is a model-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                571,
+                                505,
+                                582
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        571,
+                                        466,
+                                        582
+                                    ],
+                                    "score": 1.0,
+                                    "content": "based finite-horizon OPE algorithm that trains the model to have balanced representation",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        467,
+                                        571,
+                                        474,
+                                        582
+                                    ],
+                                    "score": 0.83,
+                                    "content": "\\phi",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        474,
+                                        571,
+                                        505,
+                                        582
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", which",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                582,
+                                505,
+                                594
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        582,
+                                        505,
+                                        594
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is encouraged to be invariant under the target and behavior policies. Specifically, given the deter-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                592,
+                                506,
+                                605
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        592,
+                                        195,
+                                        605
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ministic target policy",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        195,
+                                        595,
+                                        203,
+                                        603
+                                    ],
+                                    "score": 0.74,
+                                    "content": "\\pi",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        203,
+                                        592,
+                                        303,
+                                        605
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and the behavior policy",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        303,
+                                        595,
+                                        310,
+                                        604
+                                    ],
+                                    "score": 0.77,
+                                    "content": "\\mu",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        311,
+                                        592,
+                                        383,
+                                        605
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", at each timestep",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        383,
+                                        594,
+                                        388,
+                                        603
+                                    ],
+                                    "score": 0.67,
+                                    "content": "t",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        388,
+                                        592,
+                                        506,
+                                        605
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", it defines the factual distri-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                602,
+                                506,
+                                618
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        602,
+                                        148,
+                                        618
+                                    ],
+                                    "score": 1.0,
+                                    "content": "bution on",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        149,
+                                        604,
+                                        158,
+                                        613
+                                    ],
+                                    "score": 0.79,
+                                    "content": "\\mathcal { Z }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        158,
+                                        602,
+                                        307,
+                                        618
+                                    ],
+                                    "score": 1.0,
+                                    "content": "given that the actions until timestep",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        308,
+                                        605,
+                                        313,
+                                        614
+                                    ],
+                                    "score": 0.72,
+                                    "content": "t",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        313,
+                                        602,
+                                        493,
+                                        618
+                                    ],
+                                    "score": 1.0,
+                                    "content": "have been executed according to the policy",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        494,
+                                        605,
+                                        501,
+                                        614
+                                    ],
+                                    "score": 0.68,
+                                    "content": "\\pi",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        501,
+                                        602,
+                                        506,
+                                        618
+                                    ],
+                                    "score": 1.0,
+                                    "content": ":",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                613,
+                                507,
+                                629
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        614,
+                                        267,
+                                        628
+                                    ],
+                                    "score": 0.9,
+                                    "content": "p _ { \\phi , t } ^ { F } ( z ) = \\bar { \\mathrm { P r } } ( z _ { t } | M , \\mu , a _ { 0 : t } = \\pi ( s _ { 0 : t } ) )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        268,
+                                        613,
+                                        429,
+                                        629
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and the counterfactual distribution on",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        429,
+                                        615,
+                                        439,
+                                        624
+                                    ],
+                                    "score": 0.83,
+                                    "content": "\\mathcal { Z }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        439,
+                                        613,
+                                        507,
+                                        629
+                                    ],
+                                    "score": 1.0,
+                                    "content": "given the same",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                625,
+                                506,
+                                644
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        625,
+                                        263,
+                                        644
+                                    ],
+                                    "score": 1.0,
+                                    "content": "condition except the action at timestep",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        264,
+                                        629,
+                                        269,
+                                        638
+                                    ],
+                                    "score": 0.48,
+                                    "content": "t",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        269,
+                                        625,
+                                        273,
+                                        644
+                                    ],
+                                    "score": 1.0,
+                                    "content": ":",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        273,
+                                        627,
+                                        501,
+                                        642
+                                    ],
+                                    "score": 0.87,
+                                    "content": "p _ { \\phi , t } ^ { C F } ( z ) = \\mathrm { P r } ( z _ { t } | M , \\mu , a _ { 0 : t - 1 } = \\pi ( s _ { 0 : t - 1 } ) , a _ { t } \\neq \\pi ( s _ { t } ) )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        501,
+                                        625,
+                                        506,
+                                        644
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                641,
+                                271,
+                                654
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        641,
+                                        271,
+                                        654
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Then, RepBM bounds the OPE error as,1",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        }
+                    ],
+                    "index": 36.5
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        111,
+                        655,
+                        507,
+                        688
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                111,
+                                655,
+                                507,
+                                688
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        111,
+                                        655,
+                                        507,
+                                        688
+                                    ],
+                                    "score": 0.89,
+                                    "content": "R ^ { \\pi } - \\widehat { R } ^ { \\pi } \\Big | \\leq ( 1 - \\gamma ) \\sum _ { t = 0 } ^ { \\infty } \\gamma ^ { t } \\left( \\mathbb { E } _ { p ( s _ { t } | M , \\mu , a _ { 0 : t } = \\pi ( s _ { 0 : t } ) ) } [ \\mathcal { E } _ { \\phi , \\widehat { R } _ { z } , \\widehat { T } _ { z } } ( s _ { t } , \\pi ( s _ { t } ) ) ] + B _ { \\phi , t } [ \\mathrm { P M } _ { \\mathcal { G } _ { t } } ( p _ { \\phi , t } ^ { F } , p _ { \\phi , t } ^ { C F } ) \\right) .",
+                                    "type": "interline_equation",
+                                    "image_path": "89841c17654cdd55d12f821f9c6ff9daaeed1eb8fcc65fba27a07c65eb3a0a3a.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 42,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                111,
+                                655,
+                                507,
+                                666.0
+                            ],
+                            "spans": [],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                111,
+                                666.0,
+                                507,
+                                677.0
+                            ],
+                            "spans": [],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                111,
+                                677.0,
+                                507,
+                                688.0
+                            ],
+                            "spans": [],
+                            "index": 43
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        690,
+                        504,
+                        714
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                689,
+                                505,
+                                703
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        689,
+                                        505,
+                                        703
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Although RepBM achieves performance improvement over other OPE algorithms, we found a num-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                103,
+                                696,
+                                508,
+                                718
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        103,
+                                        696,
+                                        317,
+                                        718
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ber of practical challenges: from the definition of the",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        317,
+                                        700,
+                                        392,
+                                        714
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\mathrm { I P M } _ { \\mathcal { G } _ { t } } ( p _ { \\phi , t } ^ { F } , p _ { \\phi , t } ^ { C F } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        392,
+                                        696,
+                                        508,
+                                        718
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", it requires a discrete-action",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        }
+                    ],
+                    "index": 44.5
+                }
+            ],
+            "page_idx": 4,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        107,
+                        27,
+                        292,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                26,
+                                293,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        26,
+                                        293,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Published as a conference paper at ICLR 2021",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        118,
+                        721,
+                        401,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                119,
+                                719,
+                                402,
+                                734
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        119,
+                                        719,
+                                        402,
+                                        734
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1We adapted their formulation to the infinite horizon discounted MDP setting.",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        302,
+                        751,
+                        309,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                301,
+                                750,
+                                310,
+                                762
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        301,
+                                        750,
+                                        310,
+                                        762
+                                    ],
+                                    "score": 1.0,
+                                    "content": "5",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        82,
+                        504,
+                        105
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                81,
+                                505,
+                                96
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        83,
+                                        119,
+                                        92
+                                    ],
+                                    "score": 0.72,
+                                    "content": "M",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        119,
+                                        81,
+                                        242,
+                                        96
+                                    ],
+                                    "score": 1.0,
+                                    "content": "given an initial representation",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        243,
+                                        85,
+                                        249,
+                                        92
+                                    ],
+                                    "score": 0.73,
+                                    "content": "z",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        249,
+                                        81,
+                                        394,
+                                        96
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". While it is now difficult to choose",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        395,
+                                        83,
+                                        405,
+                                        92
+                                    ],
+                                    "score": 0.82,
+                                    "content": "\\mathcal { F }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        405,
+                                        81,
+                                        428,
+                                        96
+                                    ],
+                                    "score": 1.0,
+                                    "content": "from",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        428,
+                                        83,
+                                        435,
+                                        93
+                                    ],
+                                    "score": 0.74,
+                                    "content": "\\mathcal { G }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        436,
+                                        81,
+                                        505,
+                                        96
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", Eq. (3) still can",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                93,
+                                353,
+                                105
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        93,
+                                        340,
+                                        105
+                                    ],
+                                    "score": 1.0,
+                                    "content": "be kept valid by using a sufficiently rich function class for",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        340,
+                                        94,
+                                        350,
+                                        104
+                                    ],
+                                    "score": 0.83,
+                                    "content": "\\mathcal { F }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        350,
+                                        93,
+                                        353,
+                                        105
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        }
+                    ],
+                    "index": 0.5,
+                    "bbox_fs": [
+                        106,
+                        81,
+                        505,
+                        105
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        110,
+                        505,
+                        144
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                110,
+                                505,
+                                122
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        110,
+                                        209,
+                                        122
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In this work, we choose",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        209,
+                                        111,
+                                        219,
+                                        121
+                                    ],
+                                    "score": 0.84,
+                                    "content": "\\mathcal { F }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        219,
+                                        110,
+                                        505,
+                                        122
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to be the family of functions in the unit ball in a reproducing kernel",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                121,
+                                505,
+                                134
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        121,
+                                        198,
+                                        134
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Hilbert space (RKHS)",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        199,
+                                        122,
+                                        213,
+                                        132
+                                    ],
+                                    "score": 0.89,
+                                    "content": "\\mathcal { H } _ { k }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        213,
+                                        121,
+                                        278,
+                                        134
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with the kernel",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        278,
+                                        122,
+                                        285,
+                                        132
+                                    ],
+                                    "score": 0.81,
+                                    "content": "k",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        285,
+                                        121,
+                                        505,
+                                        134
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", which allows the following closed-form formula (see",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                132,
+                                258,
+                                145
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        132,
+                                        258,
+                                        145
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Lemma A.3 in Appendix for details):",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        }
+                    ],
+                    "index": 3,
+                    "bbox_fs": [
+                        105,
+                        110,
+                        505,
+                        145
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        111,
+                        146,
+                        507,
+                        221
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                111,
+                                146,
+                                507,
+                                221
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        111,
+                                        146,
+                                        507,
+                                        221
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\begin{array} { r l r } { \\mathrm { P M } _ { \\mathcal { G } } ( d _ { \\phi } ^ { \\pi } , d _ { \\phi } ^ { \\mathcal { D } } ) ^ { 2 } = \\mathbb { E } _ { s _ { 0 } \\sim d _ { 0 } , a _ { 0 } \\sim \\pi ( s _ { 0 } ) , ( s , a , s ^ { \\prime } ) \\sim d ^ { \\mathcal { D } } , a ^ { \\prime } \\sim \\pi ( s ^ { \\prime } ) } } & { } & { ( 5 ) } \\\\ { \\bar { s } _ { 0 } \\sim d _ { 0 } , \\bar { u } _ { 0 } \\sim d _ { 0 } ) , ( \\bar { s } _ { 0 } , \\bar { u } , s ^ { \\prime } ) \\sim d ^ { \\mathcal { D } } , a ^ { \\prime } \\sim \\pi ( \\bar { s } ^ { \\prime } ) } & { } & { } \\\\ { k \\big ( \\phi ( s , a ) , \\phi ( \\bar { s } , \\bar { a } ) \\big ) + ( 1 - \\gamma ) ^ { 2 } k \\big ( \\phi ( s _ { 0 } , a _ { 0 } ) , \\phi ( \\bar { s } _ { 0 } , \\bar { a } _ { 0 } ) \\big ) + \\gamma ^ { 2 } k \\big ( \\phi ( s ^ { \\prime } , a ^ { \\prime } ) , \\phi ( \\bar { s } ^ { \\prime } , \\bar { a } ^ { \\prime } ) \\big ) } & { } & { } \\\\ { - 2 ( 1 - \\gamma ) k \\big ( \\phi ( s _ { 0 } , a _ { 0 } ) , \\phi ( \\bar { s } , \\bar { a } ) \\big ) - 2 \\gamma k \\big ( \\phi ( s ^ { \\prime } , a ^ { \\prime } ) , \\phi ( \\bar { s } , \\bar { a } ) \\big ) + 2 \\gamma ( 1 - \\gamma ) k \\big ( \\phi ( s ^ { \\prime } , a ^ { \\prime } ) , \\phi ( \\bar { s } _ { 0 } , \\bar { a } _ { 0 } ) \\big ) \\bigg ] . } & { } & \\end{array}",
+                                    "type": "interline_equation",
+                                    "image_path": "1953b31a1cd60337242d69b916f242145a0e39f20e0603f1049a6db643290d90.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 6,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                111,
+                                146,
+                                507,
+                                171.0
+                            ],
+                            "spans": [],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                111,
+                                171.0,
+                                507,
+                                196.0
+                            ],
+                            "spans": [],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                111,
+                                196.0,
+                                507,
+                                221.0
+                            ],
+                            "spans": [],
+                            "index": 7
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        223,
+                        505,
+                        268
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                223,
+                                505,
+                                236
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        223,
+                                        505,
+                                        236
+                                    ],
+                                    "score": 1.0,
+                                    "content": "This completes our derivation for the tractable upper bound of policy evaluation error (Eq. (3)),",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                233,
+                                505,
+                                248
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        233,
+                                        219,
+                                        248
+                                    ],
+                                    "score": 1.0,
+                                    "content": "whose direct dependence on",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        220,
+                                        235,
+                                        231,
+                                        245
+                                    ],
+                                    "score": 0.86,
+                                    "content": "d ^ { \\pi }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        231,
+                                        233,
+                                        505,
+                                        248
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is eliminated by Eq. (5). Finally, we can train a model by minimizing",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                244,
+                                506,
+                                260
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        244,
+                                        506,
+                                        260
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the upper bound that encourages us to learn balanced representation while improving data fitness,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                257,
+                                369,
+                                268
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        257,
+                                        369,
+                                        268
+                                    ],
+                                    "score": 1.0,
+                                    "content": "where the trained model can readily provide a model-based OPE.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        }
+                    ],
+                    "index": 9.5,
+                    "bbox_fs": [
+                        105,
+                        223,
+                        506,
+                        268
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        272,
+                        505,
+                        323
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                272,
+                                506,
+                                287
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        272,
+                                        126,
+                                        287
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        126,
+                                        272,
+                                        192,
+                                        287
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\mathrm { I P M } _ { \\mathcal { G } } ( d _ { \\phi } ^ { \\pi } , d _ { \\phi } ^ { D } ) ^ { 2 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        192,
+                                        272,
+                                        506,
+                                        287
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in Eq. (5) can be estimated via finite random samples, and we denote its",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                286,
+                                506,
+                                302
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        288,
+                                        221,
+                                        302
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sampled-based estimator as",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        222,
+                                        286,
+                                        281,
+                                        302
+                                    ],
+                                    "score": 0.94,
+                                    "content": "\\widehat { \\mathrm { I P M } } ( d _ { \\phi } ^ { \\pi } , d _ { \\phi } ^ { D } ) ^ { 2 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        282,
+                                        288,
+                                        506,
+                                        302
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". We show in the following that, a valid upper bound",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                300,
+                                506,
+                                313
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        300,
+                                        506,
+                                        313
+                                    ],
+                                    "score": 1.0,
+                                    "content": "can be established based on the sample-based estimators instead of the exact terms in the RHS of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                312,
+                                238,
+                                324
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        312,
+                                        238,
+                                        324
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Eq. (3) under certain conditions.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        }
+                    ],
+                    "index": 13.5,
+                    "bbox_fs": [
+                        105,
+                        272,
+                        506,
+                        324
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        325,
+                        505,
+                        408
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                324,
+                                506,
+                                340
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        324,
+                                        249,
+                                        340
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Theorem 4.3. Given an MDP",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        249,
+                                        328,
+                                        261,
+                                        338
+                                    ],
+                                    "score": 0.62,
+                                    "content": "M",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        261,
+                                        324,
+                                        325,
+                                        340
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", its estimate",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        325,
+                                        325,
+                                        338,
+                                        338
+                                    ],
+                                    "score": 0.83,
+                                    "content": "\\widehat { M }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        338,
+                                        324,
+                                        506,
+                                        340
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with a bijective representation func-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                337,
+                                506,
+                                353
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        337,
+                                        128,
+                                        353
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tion",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        128,
+                                        340,
+                                        136,
+                                        352
+                                    ],
+                                    "score": 0.76,
+                                    "content": "\\phi",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        136,
+                                        337,
+                                        173,
+                                        353
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", i.e.",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        173,
+                                        338,
+                                        308,
+                                        352
+                                    ],
+                                    "score": 0.91,
+                                    "content": "( \\widehat { T } , \\widehat { R } ) = \\langle \\widehat { T } _ { z } \\circ \\phi , \\widehat { R } _ { z } \\circ \\phi \\rangle",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        309,
+                                        337,
+                                        387,
+                                        353
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", and an RKHS",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        388,
+                                        340,
+                                        487,
+                                        352
+                                    ],
+                                    "score": 0.63,
+                                    "content": "\\begin{array} { r } { \\mathcal { H } _ { k } \\quad \\subset \\quad ( \\mathcal { Z } \\quad \\to \\quad \\mathbb { R } ) } \\end{array}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        487,
+                                        337,
+                                        506,
+                                        353
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                349,
+                                504,
+                                366
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        349,
+                                        232,
+                                        366
+                                    ],
+                                    "score": 1.0,
+                                    "content": "duced by a universal kernel",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        232,
+                                        352,
+                                        240,
+                                        361
+                                    ],
+                                    "score": 0.74,
+                                    "content": "k",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        240,
+                                        349,
+                                        287,
+                                        366
+                                    ],
+                                    "score": 1.0,
+                                    "content": "such that",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        287,
+                                        351,
+                                        377,
+                                        363
+                                    ],
+                                    "score": 0.89,
+                                    "content": "\\mathrm { s u p } _ { z \\in \\mathcal { Z } } k ( z , z ) ~ = ~ \\bar { k }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        378,
+                                        349,
+                                        440,
+                                        366
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", assume that",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        441,
+                                        351,
+                                        504,
+                                        366
+                                    ],
+                                    "score": 0.9,
+                                    "content": "f _ { \\phi , \\widehat { R } _ { z } , \\widehat { T } _ { z } } \\left( \\widehat { z } \\right) \\ =",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                362,
+                                508,
+                                386
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        364,
+                                        344,
+                                        384
+                                    ],
+                                    "score": 0.9,
+                                    "content": "\\begin{array} { r } { \\mathbb { E } _ { T , \\pi } \\Big [ \\sum _ { t = 0 } ^ { \\infty } \\gamma ^ { t } \\mathcal { E } _ { \\phi , \\widehat { R } _ { z } , \\widehat { T } _ { z } } \\big ( s _ { t } , a _ { t } \\big ) \\Big | \\big ( s _ { 0 } , a _ { 0 } \\big ) = \\phi ^ { - 1 } \\big ( z ) \\Big ] \\in \\mathcal { H } _ { k } } \\end{array}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        344,
+                                        362,
+                                        366,
+                                        386
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        367,
+                                        368,
+                                        451,
+                                        383
+                                    ],
+                                    "score": 0.93,
+                                    "content": "B _ { \\phi } = \\| f _ { \\phi , \\widehat { R } _ { z } , \\widehat { T } _ { z } } \\| _ { \\mathcal { H } _ { k } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        451,
+                                        362,
+                                        508,
+                                        386
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and the loss",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                380,
+                                506,
+                                400
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        380,
+                                        165,
+                                        400
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is bounded by",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        165,
+                                        385,
+                                        293,
+                                        398
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\begin{array} { r } { \\bar { \\mathcal { E } } = \\operatorname* { s u p } _ { s \\in { \\mathcal { S } } , a \\in \\mathcal { A } } \\mathcal { E } _ { \\phi , \\widehat { R } _ { z } , \\widehat { T } _ { z } } ( s , a ) } \\end{array}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        293,
+                                        380,
+                                        423,
+                                        400
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Let n be the number of data in",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        424,
+                                        385,
+                                        432,
+                                        394
+                                    ],
+                                    "score": 0.76,
+                                    "content": "\\mathcal { D }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        433,
+                                        380,
+                                        506,
+                                        400
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". With probability",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                394,
+                                140,
+                                410
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        397,
+                                        135,
+                                        408
+                                    ],
+                                    "score": 0.64,
+                                    "content": "1 - 2 \\delta",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        135,
+                                        394,
+                                        140,
+                                        410
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        }
+                    ],
+                    "index": 18.5,
+                    "bbox_fs": [
+                        104,
+                        324,
+                        508,
+                        410
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        111,
+                        409,
+                        506,
+                        445
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                111,
+                                409,
+                                506,
+                                445
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        111,
+                                        409,
+                                        506,
+                                        445
+                                    ],
+                                    "score": 0.89,
+                                    "content": "R ^ { \\pi } - { \\widehat R } ^ { \\pi } \\Big | \\leq \\frac { 1 } { n } \\sum _ { ( s , a ) \\in { \\mathcal D } } \\mathcal { E } _ { \\phi , { \\widehat R } _ { z } , { \\widehat T } _ { z } } ( s , a ) + B _ { \\phi } \\widehat { \\Gamma \\mathrm { M } } \\big ( d _ { \\phi } ^ { \\pi } , d _ { \\phi } ^ { { D } } \\big ) + \\sqrt { \\frac { \\bar { \\mathcal E } ^ { 2 } } { 2 n } \\log \\frac { 1 } { \\delta } } + B _ { \\phi } \\sqrt { \\frac { \\bar { k } } { n } } \\left( 4 + \\sqrt { 8 \\log \\frac { 3 } { \\delta } } \\right) .",
+                                    "type": "interline_equation",
+                                    "image_path": "1cfc254f1512a52e679a50996bd4958636bc46ff2883210d61add703a1dd3887.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 23,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                111,
+                                409,
+                                506,
+                                421.0
+                            ],
+                            "spans": [],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                111,
+                                421.0,
+                                506,
+                                433.0
+                            ],
+                            "spans": [],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                111,
+                                433.0,
+                                506,
+                                445.0
+                            ],
+                            "spans": [],
+                            "index": 24
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        452,
+                        506,
+                        549
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                452,
+                                506,
+                                466
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        452,
+                                        506,
+                                        466
+                                    ],
+                                    "score": 1.0,
+                                    "content": "This result can be proved by adapting the convergence results of the empirical estimate of the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                464,
+                                506,
+                                477
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        464,
+                                        506,
+                                        477
+                                    ],
+                                    "score": 1.0,
+                                    "content": "MMD (Gretton et al., 2012) and Hoeffding’s inequality (Hoeffding, 1963). With the choice of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                474,
+                                506,
+                                491
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        474,
+                                        147,
+                                        491
+                                    ],
+                                    "score": 1.0,
+                                    "content": "an RKHS",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        147,
+                                        475,
+                                        162,
+                                        487
+                                    ],
+                                    "score": 0.88,
+                                    "content": "\\mathcal { H } _ { k }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        162,
+                                        474,
+                                        251,
+                                        491
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", we can now interpret",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        252,
+                                        476,
+                                        266,
+                                        487
+                                    ],
+                                    "score": 0.89,
+                                    "content": "B _ { \\phi }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        266,
+                                        474,
+                                        345,
+                                        491
+                                    ],
+                                    "score": 1.0,
+                                    "content": "as the RKHS norm",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        345,
+                                        475,
+                                        401,
+                                        489
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\| f _ { \\phi , \\widehat { R } _ { z } , \\widehat { T } _ { z } } \\| _ { \\mathcal { H } _ { k } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        401,
+                                        474,
+                                        506,
+                                        491
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", which captures the mag-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                486,
+                                506,
+                                504
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        486,
+                                        376,
+                                        504
+                                    ],
+                                    "score": 1.0,
+                                    "content": "bnitude and the smoothness of the expected cumulative model loss",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        377,
+                                        489,
+                                        411,
+                                        502
+                                    ],
+                                    "score": 0.91,
+                                    "content": "f _ { \\phi , \\widehat { R } _ { z } , \\widehat { T } _ { z } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        411,
+                                        486,
+                                        506,
+                                        504
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". In general, assuming",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                500,
+                                505,
+                                513
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        500,
+                                        285,
+                                        513
+                                    ],
+                                    "score": 1.0,
+                                    "content": "smooth underlying dynamics, we can expect",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        285,
+                                        501,
+                                        299,
+                                        513
+                                    ],
+                                    "score": 0.9,
+                                    "content": "B _ { \\phi }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        299,
+                                        500,
+                                        505,
+                                        513
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to be small when the model error is small. Further-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                511,
+                                506,
+                                525
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        513,
+                                        170,
+                                        525
+                                    ],
+                                    "score": 1.0,
+                                    "content": "more, although",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        170,
+                                        511,
+                                        177,
+                                        523
+                                    ],
+                                    "score": 0.81,
+                                    "content": "\\bar { k }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        178,
+                                        513,
+                                        414,
+                                        525
+                                    ],
+                                    "score": 1.0,
+                                    "content": "depends on the kernel function we use, we can always let √",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        414,
+                                        511,
+                                        441,
+                                        523
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\bar { k } = 1",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        441,
+                                        513,
+                                        506,
+                                        525
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and subsume it",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                523,
+                                506,
+                                539
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        524,
+                                        124,
+                                        539
+                                    ],
+                                    "score": 1.0,
+                                    "content": "into",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        124,
+                                        526,
+                                        138,
+                                        538
+                                    ],
+                                    "score": 0.9,
+                                    "content": "B _ { \\phi }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        139,
+                                        524,
+                                        276,
+                                        539
+                                    ],
+                                    "score": 1.0,
+                                    "content": "as long as it is bounded, i.e. using",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        277,
+                                        523,
+                                        331,
+                                        538
+                                    ],
+                                    "score": 0.94,
+                                    "content": "\\tilde { B } _ { \\phi } \\triangleq B _ { \\phi } \\sqrt { \\bar { k } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        331,
+                                        524,
+                                        506,
+                                        539
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". In the next section, we develop algorithms",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                537,
+                                289,
+                                550
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        537,
+                                        289,
+                                        550
+                                    ],
+                                    "score": 1.0,
+                                    "content": "based on practical approximations of Eq. (3).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        }
+                    ],
+                    "index": 28.5,
+                    "bbox_fs": [
+                        104,
+                        452,
+                        506,
+                        550
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        559,
+                        505,
+                        654
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                559,
+                                505,
+                                572
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        559,
+                                        505,
+                                        572
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Detailed comparison to RepBM As previously stated, RepBM (Liu et al., 2018b) is a model-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                571,
+                                505,
+                                582
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        571,
+                                        466,
+                                        582
+                                    ],
+                                    "score": 1.0,
+                                    "content": "based finite-horizon OPE algorithm that trains the model to have balanced representation",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        467,
+                                        571,
+                                        474,
+                                        582
+                                    ],
+                                    "score": 0.83,
+                                    "content": "\\phi",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        474,
+                                        571,
+                                        505,
+                                        582
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", which",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                582,
+                                505,
+                                594
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        582,
+                                        505,
+                                        594
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is encouraged to be invariant under the target and behavior policies. Specifically, given the deter-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                592,
+                                506,
+                                605
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        592,
+                                        195,
+                                        605
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ministic target policy",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        195,
+                                        595,
+                                        203,
+                                        603
+                                    ],
+                                    "score": 0.74,
+                                    "content": "\\pi",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        203,
+                                        592,
+                                        303,
+                                        605
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and the behavior policy",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        303,
+                                        595,
+                                        310,
+                                        604
+                                    ],
+                                    "score": 0.77,
+                                    "content": "\\mu",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        311,
+                                        592,
+                                        383,
+                                        605
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", at each timestep",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        383,
+                                        594,
+                                        388,
+                                        603
+                                    ],
+                                    "score": 0.67,
+                                    "content": "t",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        388,
+                                        592,
+                                        506,
+                                        605
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", it defines the factual distri-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                602,
+                                506,
+                                618
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        602,
+                                        148,
+                                        618
+                                    ],
+                                    "score": 1.0,
+                                    "content": "bution on",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        149,
+                                        604,
+                                        158,
+                                        613
+                                    ],
+                                    "score": 0.79,
+                                    "content": "\\mathcal { Z }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        158,
+                                        602,
+                                        307,
+                                        618
+                                    ],
+                                    "score": 1.0,
+                                    "content": "given that the actions until timestep",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        308,
+                                        605,
+                                        313,
+                                        614
+                                    ],
+                                    "score": 0.72,
+                                    "content": "t",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        313,
+                                        602,
+                                        493,
+                                        618
+                                    ],
+                                    "score": 1.0,
+                                    "content": "have been executed according to the policy",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        494,
+                                        605,
+                                        501,
+                                        614
+                                    ],
+                                    "score": 0.68,
+                                    "content": "\\pi",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        501,
+                                        602,
+                                        506,
+                                        618
+                                    ],
+                                    "score": 1.0,
+                                    "content": ":",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                613,
+                                507,
+                                629
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        614,
+                                        267,
+                                        628
+                                    ],
+                                    "score": 0.9,
+                                    "content": "p _ { \\phi , t } ^ { F } ( z ) = \\bar { \\mathrm { P r } } ( z _ { t } | M , \\mu , a _ { 0 : t } = \\pi ( s _ { 0 : t } ) )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        268,
+                                        613,
+                                        429,
+                                        629
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and the counterfactual distribution on",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        429,
+                                        615,
+                                        439,
+                                        624
+                                    ],
+                                    "score": 0.83,
+                                    "content": "\\mathcal { Z }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        439,
+                                        613,
+                                        507,
+                                        629
+                                    ],
+                                    "score": 1.0,
+                                    "content": "given the same",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                625,
+                                506,
+                                644
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        625,
+                                        263,
+                                        644
+                                    ],
+                                    "score": 1.0,
+                                    "content": "condition except the action at timestep",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        264,
+                                        629,
+                                        269,
+                                        638
+                                    ],
+                                    "score": 0.48,
+                                    "content": "t",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        269,
+                                        625,
+                                        273,
+                                        644
+                                    ],
+                                    "score": 1.0,
+                                    "content": ":",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        273,
+                                        627,
+                                        501,
+                                        642
+                                    ],
+                                    "score": 0.87,
+                                    "content": "p _ { \\phi , t } ^ { C F } ( z ) = \\mathrm { P r } ( z _ { t } | M , \\mu , a _ { 0 : t - 1 } = \\pi ( s _ { 0 : t - 1 } ) , a _ { t } \\neq \\pi ( s _ { t } ) )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        501,
+                                        625,
+                                        506,
+                                        644
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                641,
+                                271,
+                                654
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        641,
+                                        271,
+                                        654
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Then, RepBM bounds the OPE error as,1",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        }
+                    ],
+                    "index": 36.5,
+                    "bbox_fs": [
+                        104,
+                        559,
+                        507,
+                        654
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        111,
+                        655,
+                        507,
+                        688
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                111,
+                                655,
+                                507,
+                                688
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        111,
+                                        655,
+                                        507,
+                                        688
+                                    ],
+                                    "score": 0.89,
+                                    "content": "R ^ { \\pi } - \\widehat { R } ^ { \\pi } \\Big | \\leq ( 1 - \\gamma ) \\sum _ { t = 0 } ^ { \\infty } \\gamma ^ { t } \\left( \\mathbb { E } _ { p ( s _ { t } | M , \\mu , a _ { 0 : t } = \\pi ( s _ { 0 : t } ) ) } [ \\mathcal { E } _ { \\phi , \\widehat { R } _ { z } , \\widehat { T } _ { z } } ( s _ { t } , \\pi ( s _ { t } ) ) ] + B _ { \\phi , t } [ \\mathrm { P M } _ { \\mathcal { G } _ { t } } ( p _ { \\phi , t } ^ { F } , p _ { \\phi , t } ^ { C F } ) \\right) .",
+                                    "type": "interline_equation",
+                                    "image_path": "89841c17654cdd55d12f821f9c6ff9daaeed1eb8fcc65fba27a07c65eb3a0a3a.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 42,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                111,
+                                655,
+                                507,
+                                666.0
+                            ],
+                            "spans": [],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                111,
+                                666.0,
+                                507,
+                                677.0
+                            ],
+                            "spans": [],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                111,
+                                677.0,
+                                507,
+                                688.0
+                            ],
+                            "spans": [],
+                            "index": 43
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        690,
+                        504,
+                        714
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                689,
+                                505,
+                                703
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        689,
+                                        505,
+                                        703
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Although RepBM achieves performance improvement over other OPE algorithms, we found a num-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                103,
+                                696,
+                                508,
+                                718
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        103,
+                                        696,
+                                        317,
+                                        718
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ber of practical challenges: from the definition of the",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        317,
+                                        700,
+                                        392,
+                                        714
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\mathrm { I P M } _ { \\mathcal { G } _ { t } } ( p _ { \\phi , t } ^ { F } , p _ { \\phi , t } ^ { C F } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        392,
+                                        696,
+                                        508,
+                                        718
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", it requires a discrete-action",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                82,
+                                505,
+                                95
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        82,
+                                        267,
+                                        95
+                                    ],
+                                    "score": 1.0,
+                                    "content": "environment and a deterministic policy",
+                                    "type": "text",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        267,
+                                        85,
+                                        274,
+                                        92
+                                    ],
+                                    "score": 0.73,
+                                    "content": "\\pi",
+                                    "type": "inline_equation",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        275,
+                                        82,
+                                        505,
+                                        95
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", which cannot be met by many practical RL settings. In",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                103,
+                                89,
+                                508,
+                                111
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        103,
+                                        89,
+                                        295,
+                                        111
+                                    ],
+                                    "score": 1.0,
+                                    "content": "addition, since the sample-based estimation of",
+                                    "type": "text",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        296,
+                                        93,
+                                        370,
+                                        107
+                                    ],
+                                    "score": 0.94,
+                                    "content": "\\mathrm { I P M } _ { \\mathcal { G } _ { t } } ( p _ { \\phi , t } ^ { F } , p _ { \\phi , t } ^ { C F } )",
+                                    "type": "inline_equation",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        371,
+                                        89,
+                                        508,
+                                        111
+                                    ],
+                                    "score": 1.0,
+                                    "content": "requires samples consistent with",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                105,
+                                505,
+                                120
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        105,
+                                        149,
+                                        120
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the policy",
+                                    "type": "text",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        149,
+                                        108,
+                                        156,
+                                        117
+                                    ],
+                                    "score": 0.7,
+                                    "content": "\\pi",
+                                    "type": "inline_equation",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        156,
+                                        105,
+                                        176,
+                                        120
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", i.e.",
+                                    "type": "text",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        176,
+                                        106,
+                                        255,
+                                        118
+                                    ],
+                                    "score": 0.92,
+                                    "content": "a _ { 0 : t - 1 } = \\pi ( s _ { 0 : t - 1 } )",
+                                    "type": "inline_equation",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        256,
+                                        105,
+                                        505,
+                                        120
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", the algorithm would reject exponentially many samples with",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                118,
+                                504,
+                                129
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        118,
+                                        148,
+                                        129
+                                    ],
+                                    "score": 1.0,
+                                    "content": "respect to",
+                                    "type": "text",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        148,
+                                        118,
+                                        153,
+                                        127
+                                    ],
+                                    "score": 0.71,
+                                    "content": "t",
+                                    "type": "inline_equation",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        154,
+                                        118,
+                                        504,
+                                        129
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", which is the curse of horizon (Liu et al., 2018a). When there is a large difference be-",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                128,
+                                504,
+                                140
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        128,
+                                        504,
+                                        140
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tween the behavior and the target policies in long-term tasks, their implementation becomes close to",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                138,
+                                505,
+                                152
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        138,
+                                        505,
+                                        152
+                                    ],
+                                    "score": 1.0,
+                                    "content": "using the maximum likelihood objective, which can also be observed empirically in our experiments.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                147,
+                                506,
+                                164
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        147,
+                                        506,
+                                        164
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In contrast, our work is free from the abovementioned practical limitations by performing balancing",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                160,
+                                505,
+                                173
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        160,
+                                        291,
+                                        173
+                                    ],
+                                    "score": 1.0,
+                                    "content": "between the discounted stationary distribution",
+                                    "type": "text",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        292,
+                                        161,
+                                        303,
+                                        171
+                                    ],
+                                    "score": 0.86,
+                                    "content": "d ^ { \\pi }",
+                                    "type": "inline_equation",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        303,
+                                        160,
+                                        403,
+                                        173
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and the data distribution",
+                                    "type": "text",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        403,
+                                        160,
+                                        416,
+                                        171
+                                    ],
+                                    "score": 0.87,
+                                    "content": "d ^ { \\bar { D } }",
+                                    "type": "inline_equation",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        416,
+                                        160,
+                                        505,
+                                        173
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", leveraging the recent",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                172,
+                                505,
+                                184
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        172,
+                                        505,
+                                        184
+                                    ],
+                                    "score": 1.0,
+                                    "content": "advances in stationary distribution correction estimation (i.e. the DualDICE trick) to overcome the",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                183,
+                                504,
+                                196
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        183,
+                                        309,
+                                        196
+                                    ],
+                                    "score": 1.0,
+                                    "content": "difficulties pertinent to the expectation concerning",
+                                    "type": "text",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        309,
+                                        183,
+                                        321,
+                                        193
+                                    ],
+                                    "score": 0.85,
+                                    "content": "d ^ { \\pi }",
+                                    "type": "inline_equation",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        321,
+                                        183,
+                                        504,
+                                        196
+                                    ],
+                                    "score": 1.0,
+                                    "content": "required to evaluate the IPM in the objective.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 9
+                        }
+                    ],
+                    "index": 44.5,
+                    "bbox_fs": [
+                        103,
+                        689,
+                        508,
+                        718
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        82,
+                        505,
+                        195
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                82,
+                                505,
+                                95
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        82,
+                                        267,
+                                        95
+                                    ],
+                                    "score": 1.0,
+                                    "content": "environment and a deterministic policy",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        267,
+                                        85,
+                                        274,
+                                        92
+                                    ],
+                                    "score": 0.73,
+                                    "content": "\\pi",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        275,
+                                        82,
+                                        505,
+                                        95
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", which cannot be met by many practical RL settings. In",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                103,
+                                89,
+                                508,
+                                111
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        103,
+                                        89,
+                                        295,
+                                        111
+                                    ],
+                                    "score": 1.0,
+                                    "content": "addition, since the sample-based estimation of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        296,
+                                        93,
+                                        370,
+                                        107
+                                    ],
+                                    "score": 0.94,
+                                    "content": "\\mathrm { I P M } _ { \\mathcal { G } _ { t } } ( p _ { \\phi , t } ^ { F } , p _ { \\phi , t } ^ { C F } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        371,
+                                        89,
+                                        508,
+                                        111
+                                    ],
+                                    "score": 1.0,
+                                    "content": "requires samples consistent with",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                105,
+                                505,
+                                120
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        105,
+                                        149,
+                                        120
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the policy",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        149,
+                                        108,
+                                        156,
+                                        117
+                                    ],
+                                    "score": 0.7,
+                                    "content": "\\pi",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        156,
+                                        105,
+                                        176,
+                                        120
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", i.e.",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        176,
+                                        106,
+                                        255,
+                                        118
+                                    ],
+                                    "score": 0.92,
+                                    "content": "a _ { 0 : t - 1 } = \\pi ( s _ { 0 : t - 1 } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        256,
+                                        105,
+                                        505,
+                                        120
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", the algorithm would reject exponentially many samples with",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                118,
+                                504,
+                                129
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        118,
+                                        148,
+                                        129
+                                    ],
+                                    "score": 1.0,
+                                    "content": "respect to",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        148,
+                                        118,
+                                        153,
+                                        127
+                                    ],
+                                    "score": 0.71,
+                                    "content": "t",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        154,
+                                        118,
+                                        504,
+                                        129
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", which is the curse of horizon (Liu et al., 2018a). When there is a large difference be-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                128,
+                                504,
+                                140
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        128,
+                                        504,
+                                        140
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tween the behavior and the target policies in long-term tasks, their implementation becomes close to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                138,
+                                505,
+                                152
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        138,
+                                        505,
+                                        152
+                                    ],
+                                    "score": 1.0,
+                                    "content": "using the maximum likelihood objective, which can also be observed empirically in our experiments.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                147,
+                                506,
+                                164
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        147,
+                                        506,
+                                        164
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In contrast, our work is free from the abovementioned practical limitations by performing balancing",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                160,
+                                505,
+                                173
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        160,
+                                        291,
+                                        173
+                                    ],
+                                    "score": 1.0,
+                                    "content": "between the discounted stationary distribution",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        292,
+                                        161,
+                                        303,
+                                        171
+                                    ],
+                                    "score": 0.86,
+                                    "content": "d ^ { \\pi }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        303,
+                                        160,
+                                        403,
+                                        173
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and the data distribution",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        403,
+                                        160,
+                                        416,
+                                        171
+                                    ],
+                                    "score": 0.87,
+                                    "content": "d ^ { \\bar { D } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        416,
+                                        160,
+                                        505,
+                                        173
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", leveraging the recent",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                172,
+                                505,
+                                184
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        172,
+                                        505,
+                                        184
+                                    ],
+                                    "score": 1.0,
+                                    "content": "advances in stationary distribution correction estimation (i.e. the DualDICE trick) to overcome the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                183,
+                                504,
+                                196
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        183,
+                                        309,
+                                        196
+                                    ],
+                                    "score": 1.0,
+                                    "content": "difficulties pertinent to the expectation concerning",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        309,
+                                        183,
+                                        321,
+                                        193
+                                    ],
+                                    "score": 0.85,
+                                    "content": "d ^ { \\pi }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        321,
+                                        183,
+                                        504,
+                                        196
+                                    ],
+                                    "score": 1.0,
+                                    "content": "required to evaluate the IPM in the objective.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        }
+                    ],
+                    "index": 4.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        207,
+                        467,
+                        219
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                207,
+                                469,
+                                220
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        207,
+                                        469,
+                                        220
+                                    ],
+                                    "score": 1.0,
+                                    "content": "4.2 REPRESENTATION BALANCING WITH STATIONARY DISTRIBUTION ESTIMATION",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        }
+                    ],
+                    "index": 10
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        227,
+                        504,
+                        250
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                227,
+                                504,
+                                241
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        227,
+                                        504,
+                                        241
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In the following, we describe algorithms for OPE and offline RL based on the practical approxima-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                239,
+                                338,
+                                251
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        239,
+                                        338,
+                                        251
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tions to Eq. (3), which we call the RepB-SDE framework.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        }
+                    ],
+                    "index": 11.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        262,
+                        504,
+                        320
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                262,
+                                505,
+                                275
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        262,
+                                        505,
+                                        275
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Objective for off-policy evaluation As we mentioned earlier, we aim to minimize the upper",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                274,
+                                505,
+                                288
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        275,
+                                        188,
+                                        288
+                                    ],
+                                    "score": 1.0,
+                                    "content": "bound of OPE error",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        189,
+                                        274,
+                                        233,
+                                        288
+                                    ],
+                                    "score": 0.93,
+                                    "content": "| R ^ { \\pi } - { \\widehat { R } } ^ { \\pi } |",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        234,
+                                        275,
+                                        505,
+                                        288
+                                    ],
+                                    "score": 1.0,
+                                    "content": "specified in Theorem 4.2. To make the RHS of Eq. (3) tractable for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                286,
+                                506,
+                                299
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        286,
+                                        506,
+                                        299
+                                    ],
+                                    "score": 1.0,
+                                    "content": "optimization, we replace the intractable total variation distance with KL-divergence, which can be",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                298,
+                                504,
+                                310
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        298,
+                                        504,
+                                        310
+                                    ],
+                                    "score": 1.0,
+                                    "content": "easily minimized by maximizing the data log-likelihood. We also replace the IPM with its sample-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                309,
+                                300,
+                                321
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        309,
+                                        300,
+                                        321
+                                    ],
+                                    "score": 1.0,
+                                    "content": "based estimator to obtain the learning objective:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        }
+                    ],
+                    "index": 15
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        111,
+                        325,
+                        487,
+                        363
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                111,
+                                325,
+                                487,
+                                363
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        111,
+                                        325,
+                                        487,
+                                        363
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\operatorname* { m i n } _ { \\phi , \\widehat { R } _ { z } , \\widehat { T } _ { z } } \\frac { 1 } { n } \\sum _ { \\substack { ( s , a , s ^ { \\prime } , r ) \\in \\mathcal { D } } } \\Big [ \\underbrace { - \\log \\widehat { R } _ { z } \\big ( r | \\phi ( s , a ) \\big ) } _ { D _ { \\mathrm { K L } } ( R ( r | s , a ) | | \\widehat { R } _ { z } ( r | \\phi ( s , a ) ) ) D _ { \\mathrm { K L } } ( T ( s ^ { \\prime } | s , a ) | | \\widehat { T } _ { z } ( s ^ { \\prime } | \\phi ( s , a ) ) ) } \\Big ] + \\alpha _ { M } \\widehat { \\mathrm { I P M } } ( d _ { \\phi } ^ { \\pi } , d _ { \\phi } ^ { \\mathcal { D } } )",
+                                    "type": "interline_equation",
+                                    "image_path": "0237e51f5b62910fe3675c51bd2d0830f1b1eb5740055f62c049167cc00a62aa.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 19,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                111,
+                                325,
+                                487,
+                                337.6666666666667
+                            ],
+                            "spans": [],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                111,
+                                337.6666666666667,
+                                487,
+                                350.33333333333337
+                            ],
+                            "spans": [],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                111,
+                                350.33333333333337,
+                                487,
+                                363.00000000000006
+                            ],
+                            "spans": [],
+                            "index": 20
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        367,
+                        505,
+                        401
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                366,
+                                505,
+                                380
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        366,
+                                        160,
+                                        380
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The constant",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        160,
+                                        367,
+                                        174,
+                                        379
+                                    ],
+                                    "score": 0.9,
+                                    "content": "B _ { \\phi }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        174,
+                                        366,
+                                        505,
+                                        380
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in Theorem 4.2 depends on the function classes and cannot be estimated, and thus,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                378,
+                                504,
+                                390
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        378,
+                                        285,
+                                        390
+                                    ],
+                                    "score": 1.0,
+                                    "content": "we replace it with a tunable hyperparameter",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        285,
+                                        380,
+                                        302,
+                                        389
+                                    ],
+                                    "score": 0.87,
+                                    "content": "\\alpha _ { M }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        302,
+                                        378,
+                                        504,
+                                        390
+                                    ],
+                                    "score": 1.0,
+                                    "content": "that balances between data fitness and representa-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                389,
+                                465,
+                                402
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        389,
+                                        222,
+                                        402
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tion invariance. Remark that",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        223,
+                                        390,
+                                        257,
+                                        400
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\alpha _ { M } = 0",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        258,
+                                        389,
+                                        465,
+                                        402
+                                    ],
+                                    "score": 1.0,
+                                    "content": "recovers the simple maximum-likelihood objective.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        }
+                    ],
+                    "index": 22
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        406,
+                        505,
+                        441
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                405,
+                                505,
+                                422
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        406,
+                                        238,
+                                        422
+                                    ],
+                                    "score": 1.0,
+                                    "content": "By simulating the target policy",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        239,
+                                        410,
+                                        246,
+                                        418
+                                    ],
+                                    "score": 0.77,
+                                    "content": "\\pi",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        247,
+                                        406,
+                                        374,
+                                        422
+                                    ],
+                                    "score": 1.0,
+                                    "content": "under the environment model",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        374,
+                                        405,
+                                        402,
+                                        420
+                                    ],
+                                    "score": 0.92,
+                                    "content": "( \\widehat { T } , \\widehat { R } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        402,
+                                        406,
+                                        505,
+                                        422
+                                    ],
+                                    "score": 1.0,
+                                    "content": "obtained by minimizing",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                419,
+                                505,
+                                431
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        419,
+                                        505,
+                                        431
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Eq. (6), it is possible to perform a model-based OPE that approximately minimizes the upper bound",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                428,
+                                163,
+                                442
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        428,
+                                        163,
+                                        442
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of OPE error.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        }
+                    ],
+                    "index": 25
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        452,
+                        410,
+                        465
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                451,
+                                412,
+                                468
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        451,
+                                        412,
+                                        468
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Objectives for offline model-based RL By rearranging Eq. (3), we have,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        }
+                    ],
+                    "index": 27
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        182,
+                        469,
+                        428,
+                        491
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                182,
+                                469,
+                                428,
+                                491
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        182,
+                                        469,
+                                        428,
+                                        491
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\begin{array} { r } { { \\cal R } ^ { \\pi } \\geq \\widehat { \\cal R } ^ { \\pi } - \\mathbb { E } _ { ( s , a ) \\sim \\mathcal { D } } \\left[ \\mathcal { E } _ { \\phi , \\widehat { \\cal R } _ { z } , \\widehat { T } _ { z } } ( s , a ) \\right] - { \\cal B } _ { \\phi } \\mathrm { I P M } _ { \\mathcal { G } } ( d _ { \\phi } ^ { \\pi } , d _ { \\phi } ^ { \\mathcal { D } } ) . } \\end{array}",
+                                    "type": "interline_equation",
+                                    "image_path": "d8af21f1c1d33ba304d1be665588f047c405ec5681877518b5352d4c61c8146e.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 28,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                182,
+                                469,
+                                428,
+                                491
+                            ],
+                            "spans": [],
+                            "index": 28
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        495,
+                        506,
+                        528
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                493,
+                                505,
+                                508
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        493,
+                                        505,
+                                        508
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Then, we can maximize the RHS of Eq. (7) to get the model and the policy that maximizes the lower",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                505,
+                                505,
+                                518
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        505,
+                                        189,
+                                        518
+                                    ],
+                                    "score": 1.0,
+                                    "content": "bound of true return",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        190,
+                                        506,
+                                        203,
+                                        516
+                                    ],
+                                    "score": 0.86,
+                                    "content": "R ^ { \\pi }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        204,
+                                        505,
+                                        505,
+                                        518
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Similar to the derivation of Eq. (6), we replace the total variation distance",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                517,
+                                355,
+                                529
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        517,
+                                        355,
+                                        529
+                                    ],
+                                    "score": 1.0,
+                                    "content": "by KL-divergence to obtain the following learning objectives:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        }
+                    ],
+                    "index": 30
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        116,
+                        532,
+                        482,
+                        590
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                116,
+                                532,
+                                482,
+                                590
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        532,
+                                        482,
+                                        590
+                                    ],
+                                    "score": 0.94,
+                                    "content": "\\begin{array} { r l } & { \\mathcal { L } _ { \\widehat { M } } ( \\widehat { M } , \\pi , \\alpha _ { M } ) = \\mathbb { E } _ { d ^ { p } } \\Big [ - \\log \\widehat { R } _ { z } \\big ( r | \\phi ( s , a ) \\big ) - \\log \\widehat { T } _ { z } \\big ( s ^ { \\prime } | \\phi ( s , a ) \\big ) \\Big ] + \\alpha _ { M } \\mathrm { I P M } _ { \\mathcal { G } } \\big ( d _ { \\phi } ^ { \\pi } , d _ { \\phi } ^ { { D } } \\big ) , } \\\\ & { \\mathcal { I } _ { \\pi } ( \\pi , \\widehat { M } , \\alpha _ { \\pi } ) = \\mathbb { E } _ { \\widehat { M } , \\pi } \\left[ \\displaystyle \\sum _ { t = 0 } ^ { \\infty } \\gamma ^ { t } r _ { t } \\right] - \\alpha _ { \\pi } \\mathrm { I P M } _ { \\mathcal { G } } \\big ( d _ { \\phi } ^ { \\pi } , d _ { \\phi } ^ { { D } } \\big ) . } \\end{array}",
+                                    "type": "interline_equation",
+                                    "image_path": "55972eb1ba29537874f0985674ed88ed72de9579cd85bfcb0b113cc436a53192.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 33,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                116,
+                                532,
+                                482,
+                                551.3333333333334
+                            ],
+                            "spans": [],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                551.3333333333334,
+                                482,
+                                570.6666666666667
+                            ],
+                            "spans": [],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                570.6666666666667,
+                                482,
+                                590.0000000000001
+                            ],
+                            "spans": [],
+                            "index": 34
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        592,
+                        505,
+                        650
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                590,
+                                506,
+                                606
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        590,
+                                        506,
+                                        606
+                                    ],
+                                    "score": 1.0,
+                                    "content": "where the expectation in Eq. (9) can be optimized using various model-based RL algorithms, e.g.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                602,
+                                505,
+                                617
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        602,
+                                        505,
+                                        617
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with planning (Chua et al., 2018) or using a model-free learner (Janner et al., 2019). By iterating",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                614,
+                                504,
+                                630
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        616,
+                                        224,
+                                        630
+                                    ],
+                                    "score": 1.0,
+                                    "content": "between the minimization of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        225,
+                                        616,
+                                        242,
+                                        630
+                                    ],
+                                    "score": 0.9,
+                                    "content": "{ \\mathcal { L } } _ { \\widehat { M } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        242,
+                                        616,
+                                        305,
+                                        630
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with respect to",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        305,
+                                        614,
+                                        317,
+                                        626
+                                    ],
+                                    "score": 0.87,
+                                    "content": "\\widehat { M }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        317,
+                                        616,
+                                        419,
+                                        630
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and the maximization of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        420,
+                                        617,
+                                        433,
+                                        627
+                                    ],
+                                    "score": 0.89,
+                                    "content": "\\mathcal { I } _ { \\pi }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        433,
+                                        616,
+                                        496,
+                                        630
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with respect to",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        497,
+                                        618,
+                                        504,
+                                        626
+                                    ],
+                                    "score": 0.71,
+                                    "content": "\\pi",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                626,
+                                505,
+                                641
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        626,
+                                        505,
+                                        641
+                                    ],
+                                    "score": 1.0,
+                                    "content": "cby stochastic gradient method, it is possible to perform offline model-based RL that approximately",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                638,
+                                307,
+                                650
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        638,
+                                        289,
+                                        650
+                                    ],
+                                    "score": 1.0,
+                                    "content": "maximizes the lower bound of the true return",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        289,
+                                        639,
+                                        303,
+                                        648
+                                    ],
+                                    "score": 0.87,
+                                    "content": "R ^ { \\pi }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        303,
+                                        638,
+                                        307,
+                                        650
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        }
+                    ],
+                    "index": 37
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        661,
+                        505,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                661,
+                                505,
+                                675
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        661,
+                                        505,
+                                        675
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Implementation details Following the recent practice (Chua et al., 2018; Janner et al., 2019; Yu",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                672,
+                                506,
+                                689
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        672,
+                                        256,
+                                        689
+                                    ],
+                                    "score": 1.0,
+                                    "content": "et al., 2020), we model the dynamics",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        257,
+                                        673,
+                                        284,
+                                        687
+                                    ],
+                                    "score": 0.93,
+                                    "content": "( \\widehat { T } , \\widehat { R } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        284,
+                                        672,
+                                        506,
+                                        689
+                                    ],
+                                    "score": 1.0,
+                                    "content": "using a bootstrap ensemble of neural networks. To op-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                685,
+                                505,
+                                699
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        685,
+                                        505,
+                                        699
+                                    ],
+                                    "score": 1.0,
+                                    "content": "timize a policy based on the objective, we perform full rollouts (until reaching the terminal states or",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                696,
+                                506,
+                                712
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        696,
+                                        291,
+                                        712
+                                    ],
+                                    "score": 1.0,
+                                    "content": "maximum timesteps) using learned dynamics",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        291,
+                                        698,
+                                        318,
+                                        711
+                                    ],
+                                    "score": 0.93,
+                                    "content": "( \\widehat { T } , \\widehat { R } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        318,
+                                        696,
+                                        506,
+                                        712
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". During obtaining the rollouts, we pessimisti-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                709,
+                                505,
+                                722
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        709,
+                                        505,
+                                        722
+                                    ],
+                                    "score": 1.0,
+                                    "content": "cally augment the estimated reward function using the penalty proportional to the bootstrapped",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                720,
+                                505,
+                                734
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        720,
+                                        505,
+                                        734
+                                    ],
+                                    "score": 1.0,
+                                    "content": "uncertainty, which helped the algorithm to perform robustly. We suspect the difficulty in calculating",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        }
+                    ],
+                    "index": 42.5
+                }
+            ],
+            "page_idx": 5,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        108,
+                        27,
+                        292,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                25,
+                                293,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        25,
+                                        293,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Published as a conference paper at ICLR 2021",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        302,
+                        752,
+                        309,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                302,
+                                751,
+                                310,
+                                762
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        302,
+                                        751,
+                                        310,
+                                        762
+                                    ],
+                                    "score": 1.0,
+                                    "content": "6",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        82,
+                        505,
+                        195
+                    ],
+                    "lines": [],
+                    "index": 4.5,
+                    "bbox_fs": [
+                        103,
+                        82,
+                        508,
+                        196
+                    ],
+                    "lines_deleted": true
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        207,
+                        467,
+                        219
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                207,
+                                469,
+                                220
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        207,
+                                        469,
+                                        220
+                                    ],
+                                    "score": 1.0,
+                                    "content": "4.2 REPRESENTATION BALANCING WITH STATIONARY DISTRIBUTION ESTIMATION",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        }
+                    ],
+                    "index": 10,
+                    "bbox_fs": [
+                        105,
+                        207,
+                        469,
+                        220
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        227,
+                        504,
+                        250
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                227,
+                                504,
+                                241
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        227,
+                                        504,
+                                        241
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In the following, we describe algorithms for OPE and offline RL based on the practical approxima-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                239,
+                                338,
+                                251
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        239,
+                                        338,
+                                        251
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tions to Eq. (3), which we call the RepB-SDE framework.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        }
+                    ],
+                    "index": 11.5,
+                    "bbox_fs": [
+                        105,
+                        227,
+                        504,
+                        251
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        262,
+                        504,
+                        320
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                262,
+                                505,
+                                275
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        262,
+                                        505,
+                                        275
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Objective for off-policy evaluation As we mentioned earlier, we aim to minimize the upper",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                274,
+                                505,
+                                288
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        275,
+                                        188,
+                                        288
+                                    ],
+                                    "score": 1.0,
+                                    "content": "bound of OPE error",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        189,
+                                        274,
+                                        233,
+                                        288
+                                    ],
+                                    "score": 0.93,
+                                    "content": "| R ^ { \\pi } - { \\widehat { R } } ^ { \\pi } |",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        234,
+                                        275,
+                                        505,
+                                        288
+                                    ],
+                                    "score": 1.0,
+                                    "content": "specified in Theorem 4.2. To make the RHS of Eq. (3) tractable for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                286,
+                                506,
+                                299
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        286,
+                                        506,
+                                        299
+                                    ],
+                                    "score": 1.0,
+                                    "content": "optimization, we replace the intractable total variation distance with KL-divergence, which can be",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                298,
+                                504,
+                                310
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        298,
+                                        504,
+                                        310
+                                    ],
+                                    "score": 1.0,
+                                    "content": "easily minimized by maximizing the data log-likelihood. We also replace the IPM with its sample-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                309,
+                                300,
+                                321
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        309,
+                                        300,
+                                        321
+                                    ],
+                                    "score": 1.0,
+                                    "content": "based estimator to obtain the learning objective:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        }
+                    ],
+                    "index": 15,
+                    "bbox_fs": [
+                        105,
+                        262,
+                        506,
+                        321
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        111,
+                        325,
+                        487,
+                        363
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                111,
+                                325,
+                                487,
+                                363
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        111,
+                                        325,
+                                        487,
+                                        363
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\operatorname* { m i n } _ { \\phi , \\widehat { R } _ { z } , \\widehat { T } _ { z } } \\frac { 1 } { n } \\sum _ { \\substack { ( s , a , s ^ { \\prime } , r ) \\in \\mathcal { D } } } \\Big [ \\underbrace { - \\log \\widehat { R } _ { z } \\big ( r | \\phi ( s , a ) \\big ) } _ { D _ { \\mathrm { K L } } ( R ( r | s , a ) | | \\widehat { R } _ { z } ( r | \\phi ( s , a ) ) ) D _ { \\mathrm { K L } } ( T ( s ^ { \\prime } | s , a ) | | \\widehat { T } _ { z } ( s ^ { \\prime } | \\phi ( s , a ) ) ) } \\Big ] + \\alpha _ { M } \\widehat { \\mathrm { I P M } } ( d _ { \\phi } ^ { \\pi } , d _ { \\phi } ^ { \\mathcal { D } } )",
+                                    "type": "interline_equation",
+                                    "image_path": "0237e51f5b62910fe3675c51bd2d0830f1b1eb5740055f62c049167cc00a62aa.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 19,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                111,
+                                325,
+                                487,
+                                337.6666666666667
+                            ],
+                            "spans": [],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                111,
+                                337.6666666666667,
+                                487,
+                                350.33333333333337
+                            ],
+                            "spans": [],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                111,
+                                350.33333333333337,
+                                487,
+                                363.00000000000006
+                            ],
+                            "spans": [],
+                            "index": 20
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        367,
+                        505,
+                        401
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                366,
+                                505,
+                                380
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        366,
+                                        160,
+                                        380
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The constant",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        160,
+                                        367,
+                                        174,
+                                        379
+                                    ],
+                                    "score": 0.9,
+                                    "content": "B _ { \\phi }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        174,
+                                        366,
+                                        505,
+                                        380
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in Theorem 4.2 depends on the function classes and cannot be estimated, and thus,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                378,
+                                504,
+                                390
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        378,
+                                        285,
+                                        390
+                                    ],
+                                    "score": 1.0,
+                                    "content": "we replace it with a tunable hyperparameter",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        285,
+                                        380,
+                                        302,
+                                        389
+                                    ],
+                                    "score": 0.87,
+                                    "content": "\\alpha _ { M }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        302,
+                                        378,
+                                        504,
+                                        390
+                                    ],
+                                    "score": 1.0,
+                                    "content": "that balances between data fitness and representa-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                389,
+                                465,
+                                402
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        389,
+                                        222,
+                                        402
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tion invariance. Remark that",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        223,
+                                        390,
+                                        257,
+                                        400
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\alpha _ { M } = 0",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        258,
+                                        389,
+                                        465,
+                                        402
+                                    ],
+                                    "score": 1.0,
+                                    "content": "recovers the simple maximum-likelihood objective.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        }
+                    ],
+                    "index": 22,
+                    "bbox_fs": [
+                        105,
+                        366,
+                        505,
+                        402
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        406,
+                        505,
+                        441
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                405,
+                                505,
+                                422
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        406,
+                                        238,
+                                        422
+                                    ],
+                                    "score": 1.0,
+                                    "content": "By simulating the target policy",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        239,
+                                        410,
+                                        246,
+                                        418
+                                    ],
+                                    "score": 0.77,
+                                    "content": "\\pi",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        247,
+                                        406,
+                                        374,
+                                        422
+                                    ],
+                                    "score": 1.0,
+                                    "content": "under the environment model",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        374,
+                                        405,
+                                        402,
+                                        420
+                                    ],
+                                    "score": 0.92,
+                                    "content": "( \\widehat { T } , \\widehat { R } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        402,
+                                        406,
+                                        505,
+                                        422
+                                    ],
+                                    "score": 1.0,
+                                    "content": "obtained by minimizing",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                419,
+                                505,
+                                431
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        419,
+                                        505,
+                                        431
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Eq. (6), it is possible to perform a model-based OPE that approximately minimizes the upper bound",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                428,
+                                163,
+                                442
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        428,
+                                        163,
+                                        442
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of OPE error.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        }
+                    ],
+                    "index": 25,
+                    "bbox_fs": [
+                        105,
+                        405,
+                        505,
+                        442
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        452,
+                        410,
+                        465
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                451,
+                                412,
+                                468
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        451,
+                                        412,
+                                        468
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Objectives for offline model-based RL By rearranging Eq. (3), we have,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        }
+                    ],
+                    "index": 27,
+                    "bbox_fs": [
+                        105,
+                        451,
+                        412,
+                        468
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        182,
+                        469,
+                        428,
+                        491
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                182,
+                                469,
+                                428,
+                                491
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        182,
+                                        469,
+                                        428,
+                                        491
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\begin{array} { r } { { \\cal R } ^ { \\pi } \\geq \\widehat { \\cal R } ^ { \\pi } - \\mathbb { E } _ { ( s , a ) \\sim \\mathcal { D } } \\left[ \\mathcal { E } _ { \\phi , \\widehat { \\cal R } _ { z } , \\widehat { T } _ { z } } ( s , a ) \\right] - { \\cal B } _ { \\phi } \\mathrm { I P M } _ { \\mathcal { G } } ( d _ { \\phi } ^ { \\pi } , d _ { \\phi } ^ { \\mathcal { D } } ) . } \\end{array}",
+                                    "type": "interline_equation",
+                                    "image_path": "d8af21f1c1d33ba304d1be665588f047c405ec5681877518b5352d4c61c8146e.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 28,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                182,
+                                469,
+                                428,
+                                491
+                            ],
+                            "spans": [],
+                            "index": 28
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        495,
+                        506,
+                        528
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                493,
+                                505,
+                                508
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        493,
+                                        505,
+                                        508
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Then, we can maximize the RHS of Eq. (7) to get the model and the policy that maximizes the lower",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                505,
+                                505,
+                                518
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        505,
+                                        189,
+                                        518
+                                    ],
+                                    "score": 1.0,
+                                    "content": "bound of true return",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        190,
+                                        506,
+                                        203,
+                                        516
+                                    ],
+                                    "score": 0.86,
+                                    "content": "R ^ { \\pi }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        204,
+                                        505,
+                                        505,
+                                        518
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Similar to the derivation of Eq. (6), we replace the total variation distance",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                517,
+                                355,
+                                529
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        517,
+                                        355,
+                                        529
+                                    ],
+                                    "score": 1.0,
+                                    "content": "by KL-divergence to obtain the following learning objectives:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        }
+                    ],
+                    "index": 30,
+                    "bbox_fs": [
+                        105,
+                        493,
+                        505,
+                        529
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        116,
+                        532,
+                        482,
+                        590
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                116,
+                                532,
+                                482,
+                                590
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        532,
+                                        482,
+                                        590
+                                    ],
+                                    "score": 0.94,
+                                    "content": "\\begin{array} { r l } & { \\mathcal { L } _ { \\widehat { M } } ( \\widehat { M } , \\pi , \\alpha _ { M } ) = \\mathbb { E } _ { d ^ { p } } \\Big [ - \\log \\widehat { R } _ { z } \\big ( r | \\phi ( s , a ) \\big ) - \\log \\widehat { T } _ { z } \\big ( s ^ { \\prime } | \\phi ( s , a ) \\big ) \\Big ] + \\alpha _ { M } \\mathrm { I P M } _ { \\mathcal { G } } \\big ( d _ { \\phi } ^ { \\pi } , d _ { \\phi } ^ { { D } } \\big ) , } \\\\ & { \\mathcal { I } _ { \\pi } ( \\pi , \\widehat { M } , \\alpha _ { \\pi } ) = \\mathbb { E } _ { \\widehat { M } , \\pi } \\left[ \\displaystyle \\sum _ { t = 0 } ^ { \\infty } \\gamma ^ { t } r _ { t } \\right] - \\alpha _ { \\pi } \\mathrm { I P M } _ { \\mathcal { G } } \\big ( d _ { \\phi } ^ { \\pi } , d _ { \\phi } ^ { { D } } \\big ) . } \\end{array}",
+                                    "type": "interline_equation",
+                                    "image_path": "55972eb1ba29537874f0985674ed88ed72de9579cd85bfcb0b113cc436a53192.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 33,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                116,
+                                532,
+                                482,
+                                551.3333333333334
+                            ],
+                            "spans": [],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                551.3333333333334,
+                                482,
+                                570.6666666666667
+                            ],
+                            "spans": [],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                570.6666666666667,
+                                482,
+                                590.0000000000001
+                            ],
+                            "spans": [],
+                            "index": 34
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        592,
+                        505,
+                        650
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                590,
+                                506,
+                                606
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        590,
+                                        506,
+                                        606
+                                    ],
+                                    "score": 1.0,
+                                    "content": "where the expectation in Eq. (9) can be optimized using various model-based RL algorithms, e.g.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                602,
+                                505,
+                                617
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        602,
+                                        505,
+                                        617
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with planning (Chua et al., 2018) or using a model-free learner (Janner et al., 2019). By iterating",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                614,
+                                504,
+                                630
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        616,
+                                        224,
+                                        630
+                                    ],
+                                    "score": 1.0,
+                                    "content": "between the minimization of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        225,
+                                        616,
+                                        242,
+                                        630
+                                    ],
+                                    "score": 0.9,
+                                    "content": "{ \\mathcal { L } } _ { \\widehat { M } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        242,
+                                        616,
+                                        305,
+                                        630
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with respect to",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        305,
+                                        614,
+                                        317,
+                                        626
+                                    ],
+                                    "score": 0.87,
+                                    "content": "\\widehat { M }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        317,
+                                        616,
+                                        419,
+                                        630
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and the maximization of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        420,
+                                        617,
+                                        433,
+                                        627
+                                    ],
+                                    "score": 0.89,
+                                    "content": "\\mathcal { I } _ { \\pi }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        433,
+                                        616,
+                                        496,
+                                        630
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with respect to",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        497,
+                                        618,
+                                        504,
+                                        626
+                                    ],
+                                    "score": 0.71,
+                                    "content": "\\pi",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                626,
+                                505,
+                                641
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        626,
+                                        505,
+                                        641
+                                    ],
+                                    "score": 1.0,
+                                    "content": "cby stochastic gradient method, it is possible to perform offline model-based RL that approximately",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                638,
+                                307,
+                                650
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        638,
+                                        289,
+                                        650
+                                    ],
+                                    "score": 1.0,
+                                    "content": "maximizes the lower bound of the true return",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        289,
+                                        639,
+                                        303,
+                                        648
+                                    ],
+                                    "score": 0.87,
+                                    "content": "R ^ { \\pi }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        303,
+                                        638,
+                                        307,
+                                        650
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        }
+                    ],
+                    "index": 37,
+                    "bbox_fs": [
+                        105,
+                        590,
+                        506,
+                        650
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        661,
+                        505,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                661,
+                                505,
+                                675
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        661,
+                                        505,
+                                        675
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Implementation details Following the recent practice (Chua et al., 2018; Janner et al., 2019; Yu",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                672,
+                                506,
+                                689
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        672,
+                                        256,
+                                        689
+                                    ],
+                                    "score": 1.0,
+                                    "content": "et al., 2020), we model the dynamics",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        257,
+                                        673,
+                                        284,
+                                        687
+                                    ],
+                                    "score": 0.93,
+                                    "content": "( \\widehat { T } , \\widehat { R } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        284,
+                                        672,
+                                        506,
+                                        689
+                                    ],
+                                    "score": 1.0,
+                                    "content": "using a bootstrap ensemble of neural networks. To op-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                685,
+                                505,
+                                699
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        685,
+                                        505,
+                                        699
+                                    ],
+                                    "score": 1.0,
+                                    "content": "timize a policy based on the objective, we perform full rollouts (until reaching the terminal states or",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                696,
+                                506,
+                                712
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        696,
+                                        291,
+                                        712
+                                    ],
+                                    "score": 1.0,
+                                    "content": "maximum timesteps) using learned dynamics",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        291,
+                                        698,
+                                        318,
+                                        711
+                                    ],
+                                    "score": 0.93,
+                                    "content": "( \\widehat { T } , \\widehat { R } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        318,
+                                        696,
+                                        506,
+                                        712
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". During obtaining the rollouts, we pessimisti-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                709,
+                                505,
+                                722
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        709,
+                                        505,
+                                        722
+                                    ],
+                                    "score": 1.0,
+                                    "content": "cally augment the estimated reward function using the penalty proportional to the bootstrapped",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                720,
+                                505,
+                                734
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        720,
+                                        505,
+                                        734
+                                    ],
+                                    "score": 1.0,
+                                    "content": "uncertainty, which helped the algorithm to perform robustly. We suspect the difficulty in calculating",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                287,
+                                505,
+                                299
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        287,
+                                        505,
+                                        299
+                                    ],
+                                    "score": 1.0,
+                                    "content": "accurate IPM estimation is what makes the additional pessimism beneficial. We store the generated",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                298,
+                                505,
+                                312
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        300,
+                                        241,
+                                        312
+                                    ],
+                                    "score": 1.0,
+                                    "content": "experiences to a separate dataset",
+                                    "type": "text",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        241,
+                                        298,
+                                        251,
+                                        310
+                                    ],
+                                    "score": 0.85,
+                                    "content": "\\widehat { \\mathcal { D } }",
+                                    "type": "inline_equation",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        251,
+                                        300,
+                                        342,
+                                        312
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and update the policy",
+                                    "type": "text",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        342,
+                                        302,
+                                        350,
+                                        310
+                                    ],
+                                    "score": 0.74,
+                                    "content": "\\pi",
+                                    "type": "inline_equation",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        351,
+                                        300,
+                                        505,
+                                        312
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with IPM-regularized soft actor-critic",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                311,
+                                505,
+                                325
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        313,
+                                        368,
+                                        325
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(SAC) (Haarnoja et al., 2018) using samples from both datasets",
+                                    "type": "text",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        369,
+                                        311,
+                                        398,
+                                        324
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\mathcal { D } \\cup \\widehat { \\mathcal { D } }",
+                                    "type": "inline_equation",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        399,
+                                        313,
+                                        505,
+                                        325
+                                    ],
+                                    "score": 1.0,
+                                    "content": "similar to MBPO (Janner",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                324,
+                                159,
+                                337
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        324,
+                                        159,
+                                        337
+                                    ],
+                                    "score": 1.0,
+                                    "content": "et al., 2019).",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 11
+                        }
+                    ],
+                    "index": 42.5,
+                    "bbox_fs": [
+                        104,
+                        661,
+                        506,
+                        734
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "image",
+                    "bbox": [
+                        105,
+                        80,
+                        506,
+                        199
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                105,
+                                80,
+                                506,
+                                199
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        80,
+                                        506,
+                                        199
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                80,
+                                                506,
+                                                199
+                                            ],
+                                            "score": 0.971,
+                                            "type": "image",
+                                            "image_path": "f7e55e839858979ed3073f247288e2993a18881c7140dc8bb252ba65b110e1f2.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 1,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        80,
+                                        506,
+                                        119.66666666666666
+                                    ],
+                                    "spans": [],
+                                    "index": 0
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        119.66666666666666,
+                                        506,
+                                        159.33333333333331
+                                    ],
+                                    "spans": [],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        159.33333333333331,
+                                        506,
+                                        198.99999999999997
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                106,
+                                207,
+                                505,
+                                262
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        207,
+                                        505,
+                                        219
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                207,
+                                                505,
+                                                219
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 1: The OPE results of different model learning algorithms with varying off-policyness on the",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 3
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        218,
+                                        506,
+                                        231
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                220,
+                                                113,
+                                                228
+                                            ],
+                                            "score": 0.45,
+                                            "content": "\\mathbf { X }",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                113,
+                                                218,
+                                                506,
+                                                231
+                                            ],
+                                            "score": 1.0,
+                                            "content": "-axis. The y-axis plots the normalized individual MSE on test trajectories where the performance of",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 4
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        229,
+                                        504,
+                                        241
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                229,
+                                                465,
+                                                241
+                                            ],
+                                            "score": 1.0,
+                                            "content": "the baseline model is set to 1. The tasks used are infinite-horizon discounted environments",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                465,
+                                                229,
+                                                504,
+                                                240
+                                            ],
+                                            "score": 0.87,
+                                            "content": "\\gamma = 0 . 9 8",
+                                            "type": "inline_equation"
+                                        }
+                                    ],
+                                    "index": 5
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        239,
+                                        506,
+                                        254
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                239,
+                                                135,
+                                                254
+                                            ],
+                                            "score": 1.0,
+                                            "content": "(HIV),",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                135,
+                                                240,
+                                                173,
+                                                252
+                                            ],
+                                            "score": 0.89,
+                                            "content": "\\gamma = 0 . 9 9",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                174,
+                                                239,
+                                                301,
+                                                254
+                                            ],
+                                            "score": 1.0,
+                                            "content": "(others)), where we truncated at",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                301,
+                                                240,
+                                                339,
+                                                250
+                                            ],
+                                            "score": 0.88,
+                                            "content": "t = 1 0 0 0",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                340,
+                                                239,
+                                                506,
+                                                254
+                                            ],
+                                            "score": 1.0,
+                                            "content": ". The experiments are repeated 200 times,",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 6
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        251,
+                                        318,
+                                        263
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                251,
+                                                213,
+                                                263
+                                            ],
+                                            "score": 1.0,
+                                            "content": "and the error bars indicate",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                213,
+                                                251,
+                                                233,
+                                                262
+                                            ],
+                                            "score": 0.89,
+                                            "content": "9 5 \\%",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                233,
+                                                251,
+                                                318,
+                                                263
+                                            ],
+                                            "score": 1.0,
+                                            "content": "confidence intervals.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 7
+                                }
+                            ],
+                            "index": 5
+                        }
+                    ],
+                    "index": 3.0
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        287,
+                        505,
+                        336
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                287,
+                                505,
+                                299
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        287,
+                                        505,
+                                        299
+                                    ],
+                                    "score": 1.0,
+                                    "content": "accurate IPM estimation is what makes the additional pessimism beneficial. We store the generated",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                298,
+                                505,
+                                312
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        300,
+                                        241,
+                                        312
+                                    ],
+                                    "score": 1.0,
+                                    "content": "experiences to a separate dataset",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        241,
+                                        298,
+                                        251,
+                                        310
+                                    ],
+                                    "score": 0.85,
+                                    "content": "\\widehat { \\mathcal { D } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        251,
+                                        300,
+                                        342,
+                                        312
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and update the policy",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        342,
+                                        302,
+                                        350,
+                                        310
+                                    ],
+                                    "score": 0.74,
+                                    "content": "\\pi",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        351,
+                                        300,
+                                        505,
+                                        312
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with IPM-regularized soft actor-critic",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                311,
+                                505,
+                                325
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        313,
+                                        368,
+                                        325
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(SAC) (Haarnoja et al., 2018) using samples from both datasets",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        369,
+                                        311,
+                                        398,
+                                        324
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\mathcal { D } \\cup \\widehat { \\mathcal { D } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        399,
+                                        313,
+                                        505,
+                                        325
+                                    ],
+                                    "score": 1.0,
+                                    "content": "similar to MBPO (Janner",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                324,
+                                159,
+                                337
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        324,
+                                        159,
+                                        337
+                                    ],
+                                    "score": 1.0,
+                                    "content": "et al., 2019).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        }
+                    ],
+                    "index": 9.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        341,
+                        505,
+                        426
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                340,
+                                505,
+                                354
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        340,
+                                        339,
+                                        354
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Since the presented model objective requires a policy",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        339,
+                                        344,
+                                        347,
+                                        351
+                                    ],
+                                    "score": 0.76,
+                                    "content": "\\pi",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        347,
+                                        340,
+                                        505,
+                                        354
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to perform a balancing, we initially",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                352,
+                                506,
+                                369
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        352,
+                                        270,
+                                        369
+                                    ],
+                                    "score": 1.0,
+                                    "content": "trained the model and the policy using",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        270,
+                                        354,
+                                        338,
+                                        366
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\alpha _ { M } = \\alpha _ { \\pi } = 0",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        339,
+                                        352,
+                                        359,
+                                        369
+                                    ],
+                                    "score": 1.0,
+                                    "content": ": by",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        360,
+                                        352,
+                                        486,
+                                        368
+                                    ],
+                                    "score": 0.92,
+                                    "content": "{ \\widehat { \\widetilde M } } _ { 0 } = \\arg \\operatorname* { m i n } _ { { \\widehat M } } { \\mathcal { L } } _ { { \\widehat M } } ^ { - } ( { \\widehat M } , \\cdot , 0 )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        486,
+                                        352,
+                                        506,
+                                        369
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                366,
+                                505,
+                                382
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        366,
+                                        230,
+                                        381
+                                    ],
+                                    "score": 0.9,
+                                    "content": "\\pi _ { 0 } = \\arg \\operatorname* { m i n } _ { \\pi } \\mathcal { L } _ { \\pi } ( \\pi , \\widehat { M } _ { 0 } , 0 )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        231,
+                                        367,
+                                        455,
+                                        382
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Then, we retrained the model and the policy using",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        456,
+                                        369,
+                                        467,
+                                        380
+                                    ],
+                                    "score": 0.76,
+                                    "content": "\\pi _ { 0 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        468,
+                                        367,
+                                        474,
+                                        382
+                                    ],
+                                    "score": 1.0,
+                                    "content": ":",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        475,
+                                        366,
+                                        505,
+                                        380
+                                    ],
+                                    "score": 0.88,
+                                    "content": "\\widehat { M } _ { 1 } \\ =",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                380,
+                                505,
+                                396
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        381,
+                                        220,
+                                        396
+                                    ],
+                                    "score": 0.89,
+                                    "content": "\\mathrm { a r g m i n } _ { \\widehat { M } } \\mathcal { L } _ { \\widehat { M } } ( \\widehat { M } , \\pi _ { 0 } , \\alpha _ { M } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        221,
+                                        381,
+                                        236,
+                                        396
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        237,
+                                        380,
+                                        361,
+                                        394
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\begin{array} { r } { \\pi _ { 1 } = \\arg \\operatorname* { m i n } _ { \\pi } \\mathcal { L } _ { \\pi } ( \\pi , \\widehat { M } _ { 1 } , \\alpha _ { \\pi } ) } \\end{array}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        362,
+                                        381,
+                                        453,
+                                        396
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for some non-negative",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        454,
+                                        384,
+                                        470,
+                                        393
+                                    ],
+                                    "score": 0.86,
+                                    "content": "\\alpha _ { M }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        471,
+                                        381,
+                                        488,
+                                        396
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        488,
+                                        384,
+                                        501,
+                                        393
+                                    ],
+                                    "score": 0.84,
+                                    "content": "\\alpha _ { \\pi }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        501,
+                                        381,
+                                        505,
+                                        396
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                392,
+                                505,
+                                406
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        392,
+                                        505,
+                                        406
+                                    ],
+                                    "score": 1.0,
+                                    "content": "c cWhile it is desirable to repeat the optimization of the model and the policy until convergence, we",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                403,
+                                505,
+                                417
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        403,
+                                        505,
+                                        417
+                                    ],
+                                    "score": 1.0,
+                                    "content": "did not observe significant improvement after the first iteration and reported the performance of the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                415,
+                                241,
+                                429
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        415,
+                                        225,
+                                        429
+                                    ],
+                                    "score": 1.0,
+                                    "content": "policy after the first iteration,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        225,
+                                        417,
+                                        236,
+                                        426
+                                    ],
+                                    "score": 0.84,
+                                    "content": "\\pi _ { 1 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        237,
+                                        415,
+                                        241,
+                                        429
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        }
+                    ],
+                    "index": 15
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        446,
+                        200,
+                        459
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                445,
+                                201,
+                                461
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        445,
+                                        201,
+                                        461
+                                    ],
+                                    "score": 1.0,
+                                    "content": "5 EXPERIMENTS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        }
+                    ],
+                    "index": 19
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        473,
+                        505,
+                        518
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                473,
+                                505,
+                                486
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        473,
+                                        505,
+                                        486
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We demonstrate the effectiveness of the RepB-SDE framework, by comparing the OPE performance",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                484,
+                                505,
+                                497
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        484,
+                                        505,
+                                        497
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of Eq. (6) to that of RepBM and evaluating the presented model-based offline RL algorithm on the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                495,
+                                505,
+                                508
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        495,
+                                        505,
+                                        508
+                                    ],
+                                    "score": 1.0,
+                                    "content": "benchmarks. The code used to produce the results is available online.2 A detailed description of the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                507,
+                                275,
+                                518
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        507,
+                                        275,
+                                        518
+                                    ],
+                                    "score": 1.0,
+                                    "content": "experiments can be found in Appendix B.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        }
+                    ],
+                    "index": 21.5
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        536,
+                        307,
+                        546
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                535,
+                                308,
+                                548
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        535,
+                                        308,
+                                        548
+                                    ],
+                                    "score": 1.0,
+                                    "content": "5.1 MODEL-BASED OFF-POLICY EVALUATION",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        }
+                    ],
+                    "index": 24
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        557,
+                        505,
+                        635
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                556,
+                                505,
+                                570
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        556,
+                                        505,
+                                        570
+                                    ],
+                                    "score": 1.0,
+                                    "content": "For the sake of comparison with RepBM, we test our OPE algorithm on three continuous-state",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                569,
+                                505,
+                                581
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        569,
+                                        505,
+                                        581
+                                    ],
+                                    "score": 1.0,
+                                    "content": "discrete-action tasks where the goal is to evaluate a deterministic target policy. We trained a sub-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                579,
+                                505,
+                                592
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        579,
+                                        249,
+                                        592
+                                    ],
+                                    "score": 1.0,
+                                    "content": "optimal deterministic target policy",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        250,
+                                        581,
+                                        257,
+                                        589
+                                    ],
+                                    "score": 0.7,
+                                    "content": "\\pi",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        258,
+                                        579,
+                                        311,
+                                        592
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and used an",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        312,
+                                        582,
+                                        317,
+                                        589
+                                    ],
+                                    "score": 0.53,
+                                    "content": "\\epsilon",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        317,
+                                        579,
+                                        471,
+                                        592
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-greedy policy with various values of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        471,
+                                        581,
+                                        477,
+                                        589
+                                    ],
+                                    "score": 0.61,
+                                    "content": "\\epsilon",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        478,
+                                        579,
+                                        505,
+                                        592
+                                    ],
+                                    "score": 1.0,
+                                    "content": "as the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                590,
+                                506,
+                                603
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        590,
+                                        506,
+                                        603
+                                    ],
+                                    "score": 1.0,
+                                    "content": "data collection policy. In each experiment, we trained environment models for a fixed number of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                600,
+                                505,
+                                616
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        600,
+                                        505,
+                                        616
+                                    ],
+                                    "score": 1.0,
+                                    "content": "epochs concerning three different objectives: simple maximum-likelihood baseline, step-wise rep-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                612,
+                                505,
+                                624
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        612,
+                                        505,
+                                        624
+                                    ],
+                                    "score": 1.0,
+                                    "content": "resentation balancing objective used in RepBM (Liu et al., 2018b), and the presented OPE objective",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                623,
+                                475,
+                                636
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        623,
+                                        475,
+                                        636
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of RepB-SDE (Eq. (6)). We measured the individual mean squared error (Liu et al., 2018b).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        }
+                    ],
+                    "index": 28
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        640,
+                        505,
+                        706
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                640,
+                                505,
+                                653
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        640,
+                                        505,
+                                        653
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The normalized error of each algorithm, relative to the error of the baseline valued at 1, is presented",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                651,
+                                506,
+                                663
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        651,
+                                        506,
+                                        663
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in Figure 1. We can observe that the presented objective of RepB-SDE can reduce the OPE error",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                663,
+                                504,
+                                674
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        663,
+                                        504,
+                                        674
+                                    ],
+                                    "score": 1.0,
+                                    "content": "from the baseline significantly, outperforming RepBM in most of the cases. As the off-policyness be-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                673,
+                                505,
+                                686
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        673,
+                                        505,
+                                        686
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tween the policies (\u000f) increases, representation balancing algorithms should more benefit compared",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                684,
+                                506,
+                                697
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        684,
+                                        506,
+                                        697
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to the maximum-likelihood baseline in principle. However, the result shows that the performance of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                695,
+                                434,
+                                708
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        695,
+                                        424,
+                                        708
+                                    ],
+                                    "score": 1.0,
+                                    "content": "RepBM merely increases due to the increased sample rejection rate under large",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        425,
+                                        697,
+                                        430,
+                                        705
+                                    ],
+                                    "score": 0.47,
+                                    "content": "\\epsilon",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        430,
+                                        695,
+                                        434,
+                                        708
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        }
+                    ],
+                    "index": 34.5
+                }
+            ],
+            "page_idx": 6,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        119,
+                        722,
+                        318,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                118,
+                                720,
+                                320,
+                                734
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        118,
+                                        720,
+                                        320,
+                                        734
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2https://github.com/dlqudwns/repb-sde",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        107,
+                        27,
+                        292,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                26,
+                                293,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        26,
+                                        293,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Published as a conference paper at ICLR 2021",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        302,
+                        751,
+                        309,
+                        759
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                302,
+                                750,
+                                309,
+                                762
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        302,
+                                        750,
+                                        309,
+                                        762
+                                    ],
+                                    "score": 1.0,
+                                    "content": "7",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "image",
+                    "bbox": [
+                        105,
+                        80,
+                        506,
+                        199
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                105,
+                                80,
+                                506,
+                                199
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        80,
+                                        506,
+                                        199
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                80,
+                                                506,
+                                                199
+                                            ],
+                                            "score": 0.971,
+                                            "type": "image",
+                                            "image_path": "f7e55e839858979ed3073f247288e2993a18881c7140dc8bb252ba65b110e1f2.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 1,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        80,
+                                        506,
+                                        119.66666666666666
+                                    ],
+                                    "spans": [],
+                                    "index": 0
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        119.66666666666666,
+                                        506,
+                                        159.33333333333331
+                                    ],
+                                    "spans": [],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        159.33333333333331,
+                                        506,
+                                        198.99999999999997
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                106,
+                                207,
+                                505,
+                                262
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        207,
+                                        505,
+                                        219
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                207,
+                                                505,
+                                                219
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 1: The OPE results of different model learning algorithms with varying off-policyness on the",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 3
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        218,
+                                        506,
+                                        231
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                220,
+                                                113,
+                                                228
+                                            ],
+                                            "score": 0.45,
+                                            "content": "\\mathbf { X }",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                113,
+                                                218,
+                                                506,
+                                                231
+                                            ],
+                                            "score": 1.0,
+                                            "content": "-axis. The y-axis plots the normalized individual MSE on test trajectories where the performance of",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 4
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        229,
+                                        504,
+                                        241
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                229,
+                                                465,
+                                                241
+                                            ],
+                                            "score": 1.0,
+                                            "content": "the baseline model is set to 1. The tasks used are infinite-horizon discounted environments",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                465,
+                                                229,
+                                                504,
+                                                240
+                                            ],
+                                            "score": 0.87,
+                                            "content": "\\gamma = 0 . 9 8",
+                                            "type": "inline_equation"
+                                        }
+                                    ],
+                                    "index": 5
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        239,
+                                        506,
+                                        254
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                239,
+                                                135,
+                                                254
+                                            ],
+                                            "score": 1.0,
+                                            "content": "(HIV),",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                135,
+                                                240,
+                                                173,
+                                                252
+                                            ],
+                                            "score": 0.89,
+                                            "content": "\\gamma = 0 . 9 9",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                174,
+                                                239,
+                                                301,
+                                                254
+                                            ],
+                                            "score": 1.0,
+                                            "content": "(others)), where we truncated at",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                301,
+                                                240,
+                                                339,
+                                                250
+                                            ],
+                                            "score": 0.88,
+                                            "content": "t = 1 0 0 0",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                340,
+                                                239,
+                                                506,
+                                                254
+                                            ],
+                                            "score": 1.0,
+                                            "content": ". The experiments are repeated 200 times,",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 6
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        251,
+                                        318,
+                                        263
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                251,
+                                                213,
+                                                263
+                                            ],
+                                            "score": 1.0,
+                                            "content": "and the error bars indicate",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                213,
+                                                251,
+                                                233,
+                                                262
+                                            ],
+                                            "score": 0.89,
+                                            "content": "9 5 \\%",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                233,
+                                                251,
+                                                318,
+                                                263
+                                            ],
+                                            "score": 1.0,
+                                            "content": "confidence intervals.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 7
+                                }
+                            ],
+                            "index": 5
+                        }
+                    ],
+                    "index": 3.0
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        287,
+                        505,
+                        336
+                    ],
+                    "lines": [],
+                    "index": 9.5,
+                    "bbox_fs": [
+                        105,
+                        287,
+                        505,
+                        337
+                    ],
+                    "lines_deleted": true
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        341,
+                        505,
+                        426
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                340,
+                                505,
+                                354
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        340,
+                                        339,
+                                        354
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Since the presented model objective requires a policy",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        339,
+                                        344,
+                                        347,
+                                        351
+                                    ],
+                                    "score": 0.76,
+                                    "content": "\\pi",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        347,
+                                        340,
+                                        505,
+                                        354
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to perform a balancing, we initially",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                352,
+                                506,
+                                369
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        352,
+                                        270,
+                                        369
+                                    ],
+                                    "score": 1.0,
+                                    "content": "trained the model and the policy using",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        270,
+                                        354,
+                                        338,
+                                        366
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\alpha _ { M } = \\alpha _ { \\pi } = 0",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        339,
+                                        352,
+                                        359,
+                                        369
+                                    ],
+                                    "score": 1.0,
+                                    "content": ": by",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        360,
+                                        352,
+                                        486,
+                                        368
+                                    ],
+                                    "score": 0.92,
+                                    "content": "{ \\widehat { \\widetilde M } } _ { 0 } = \\arg \\operatorname* { m i n } _ { { \\widehat M } } { \\mathcal { L } } _ { { \\widehat M } } ^ { - } ( { \\widehat M } , \\cdot , 0 )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        486,
+                                        352,
+                                        506,
+                                        369
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                366,
+                                505,
+                                382
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        366,
+                                        230,
+                                        381
+                                    ],
+                                    "score": 0.9,
+                                    "content": "\\pi _ { 0 } = \\arg \\operatorname* { m i n } _ { \\pi } \\mathcal { L } _ { \\pi } ( \\pi , \\widehat { M } _ { 0 } , 0 )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        231,
+                                        367,
+                                        455,
+                                        382
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Then, we retrained the model and the policy using",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        456,
+                                        369,
+                                        467,
+                                        380
+                                    ],
+                                    "score": 0.76,
+                                    "content": "\\pi _ { 0 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        468,
+                                        367,
+                                        474,
+                                        382
+                                    ],
+                                    "score": 1.0,
+                                    "content": ":",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        475,
+                                        366,
+                                        505,
+                                        380
+                                    ],
+                                    "score": 0.88,
+                                    "content": "\\widehat { M } _ { 1 } \\ =",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                380,
+                                505,
+                                396
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        381,
+                                        220,
+                                        396
+                                    ],
+                                    "score": 0.89,
+                                    "content": "\\mathrm { a r g m i n } _ { \\widehat { M } } \\mathcal { L } _ { \\widehat { M } } ( \\widehat { M } , \\pi _ { 0 } , \\alpha _ { M } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        221,
+                                        381,
+                                        236,
+                                        396
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        237,
+                                        380,
+                                        361,
+                                        394
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\begin{array} { r } { \\pi _ { 1 } = \\arg \\operatorname* { m i n } _ { \\pi } \\mathcal { L } _ { \\pi } ( \\pi , \\widehat { M } _ { 1 } , \\alpha _ { \\pi } ) } \\end{array}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        362,
+                                        381,
+                                        453,
+                                        396
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for some non-negative",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        454,
+                                        384,
+                                        470,
+                                        393
+                                    ],
+                                    "score": 0.86,
+                                    "content": "\\alpha _ { M }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        471,
+                                        381,
+                                        488,
+                                        396
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        488,
+                                        384,
+                                        501,
+                                        393
+                                    ],
+                                    "score": 0.84,
+                                    "content": "\\alpha _ { \\pi }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        501,
+                                        381,
+                                        505,
+                                        396
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                392,
+                                505,
+                                406
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        392,
+                                        505,
+                                        406
+                                    ],
+                                    "score": 1.0,
+                                    "content": "c cWhile it is desirable to repeat the optimization of the model and the policy until convergence, we",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                403,
+                                505,
+                                417
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        403,
+                                        505,
+                                        417
+                                    ],
+                                    "score": 1.0,
+                                    "content": "did not observe significant improvement after the first iteration and reported the performance of the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                415,
+                                241,
+                                429
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        415,
+                                        225,
+                                        429
+                                    ],
+                                    "score": 1.0,
+                                    "content": "policy after the first iteration,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        225,
+                                        417,
+                                        236,
+                                        426
+                                    ],
+                                    "score": 0.84,
+                                    "content": "\\pi _ { 1 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        237,
+                                        415,
+                                        241,
+                                        429
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        }
+                    ],
+                    "index": 15,
+                    "bbox_fs": [
+                        105,
+                        340,
+                        506,
+                        429
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        446,
+                        200,
+                        459
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                445,
+                                201,
+                                461
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        445,
+                                        201,
+                                        461
+                                    ],
+                                    "score": 1.0,
+                                    "content": "5 EXPERIMENTS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        }
+                    ],
+                    "index": 19
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        473,
+                        505,
+                        518
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                473,
+                                505,
+                                486
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        473,
+                                        505,
+                                        486
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We demonstrate the effectiveness of the RepB-SDE framework, by comparing the OPE performance",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                484,
+                                505,
+                                497
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        484,
+                                        505,
+                                        497
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of Eq. (6) to that of RepBM and evaluating the presented model-based offline RL algorithm on the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                495,
+                                505,
+                                508
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        495,
+                                        505,
+                                        508
+                                    ],
+                                    "score": 1.0,
+                                    "content": "benchmarks. The code used to produce the results is available online.2 A detailed description of the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                507,
+                                275,
+                                518
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        507,
+                                        275,
+                                        518
+                                    ],
+                                    "score": 1.0,
+                                    "content": "experiments can be found in Appendix B.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        }
+                    ],
+                    "index": 21.5,
+                    "bbox_fs": [
+                        105,
+                        473,
+                        505,
+                        518
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        536,
+                        307,
+                        546
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                535,
+                                308,
+                                548
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        535,
+                                        308,
+                                        548
+                                    ],
+                                    "score": 1.0,
+                                    "content": "5.1 MODEL-BASED OFF-POLICY EVALUATION",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        }
+                    ],
+                    "index": 24
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        557,
+                        505,
+                        635
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                556,
+                                505,
+                                570
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        556,
+                                        505,
+                                        570
+                                    ],
+                                    "score": 1.0,
+                                    "content": "For the sake of comparison with RepBM, we test our OPE algorithm on three continuous-state",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                569,
+                                505,
+                                581
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        569,
+                                        505,
+                                        581
+                                    ],
+                                    "score": 1.0,
+                                    "content": "discrete-action tasks where the goal is to evaluate a deterministic target policy. We trained a sub-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                579,
+                                505,
+                                592
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        579,
+                                        249,
+                                        592
+                                    ],
+                                    "score": 1.0,
+                                    "content": "optimal deterministic target policy",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        250,
+                                        581,
+                                        257,
+                                        589
+                                    ],
+                                    "score": 0.7,
+                                    "content": "\\pi",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        258,
+                                        579,
+                                        311,
+                                        592
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and used an",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        312,
+                                        582,
+                                        317,
+                                        589
+                                    ],
+                                    "score": 0.53,
+                                    "content": "\\epsilon",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        317,
+                                        579,
+                                        471,
+                                        592
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-greedy policy with various values of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        471,
+                                        581,
+                                        477,
+                                        589
+                                    ],
+                                    "score": 0.61,
+                                    "content": "\\epsilon",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        478,
+                                        579,
+                                        505,
+                                        592
+                                    ],
+                                    "score": 1.0,
+                                    "content": "as the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                590,
+                                506,
+                                603
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        590,
+                                        506,
+                                        603
+                                    ],
+                                    "score": 1.0,
+                                    "content": "data collection policy. In each experiment, we trained environment models for a fixed number of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                600,
+                                505,
+                                616
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        600,
+                                        505,
+                                        616
+                                    ],
+                                    "score": 1.0,
+                                    "content": "epochs concerning three different objectives: simple maximum-likelihood baseline, step-wise rep-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                612,
+                                505,
+                                624
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        612,
+                                        505,
+                                        624
+                                    ],
+                                    "score": 1.0,
+                                    "content": "resentation balancing objective used in RepBM (Liu et al., 2018b), and the presented OPE objective",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                623,
+                                475,
+                                636
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        623,
+                                        475,
+                                        636
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of RepB-SDE (Eq. (6)). We measured the individual mean squared error (Liu et al., 2018b).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        }
+                    ],
+                    "index": 28,
+                    "bbox_fs": [
+                        105,
+                        556,
+                        506,
+                        636
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        640,
+                        505,
+                        706
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                640,
+                                505,
+                                653
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        640,
+                                        505,
+                                        653
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The normalized error of each algorithm, relative to the error of the baseline valued at 1, is presented",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                651,
+                                506,
+                                663
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        651,
+                                        506,
+                                        663
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in Figure 1. We can observe that the presented objective of RepB-SDE can reduce the OPE error",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                663,
+                                504,
+                                674
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        663,
+                                        504,
+                                        674
+                                    ],
+                                    "score": 1.0,
+                                    "content": "from the baseline significantly, outperforming RepBM in most of the cases. As the off-policyness be-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                673,
+                                505,
+                                686
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        673,
+                                        505,
+                                        686
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tween the policies (\u000f) increases, representation balancing algorithms should more benefit compared",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                684,
+                                506,
+                                697
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        684,
+                                        506,
+                                        697
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to the maximum-likelihood baseline in principle. However, the result shows that the performance of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                695,
+                                434,
+                                708
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        695,
+                                        424,
+                                        708
+                                    ],
+                                    "score": 1.0,
+                                    "content": "RepBM merely increases due to the increased sample rejection rate under large",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        425,
+                                        697,
+                                        430,
+                                        705
+                                    ],
+                                    "score": 0.47,
+                                    "content": "\\epsilon",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        430,
+                                        695,
+                                        434,
+                                        708
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        }
+                    ],
+                    "index": 34.5,
+                    "bbox_fs": [
+                        105,
+                        640,
+                        506,
+                        708
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "table",
+                    "bbox": [
+                        108,
+                        137,
+                        503,
+                        303
+                    ],
+                    "blocks": [
+                        {
+                            "type": "table_caption",
+                            "bbox": [
+                                106,
+                                80,
+                                505,
+                                125
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        79,
+                                        505,
+                                        93
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                79,
+                                                505,
+                                                93
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Table 1: Normalized scores on D4RL MuJoCo benchmark datasets (Fu et al., 2020) where the score",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 0
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        91,
+                                        505,
+                                        104
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                91,
+                                                505,
+                                                104
+                                            ],
+                                            "score": 1.0,
+                                            "content": "of 0 corresponds to a random policy and 100 corresponds to a converged SAC policy. All results",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        101,
+                                        506,
+                                        116
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                101,
+                                                506,
+                                                116
+                                            ],
+                                            "score": 1.0,
+                                            "content": "(except MF, which is taken from Fu et al. (2020) and Kumar et al. (2020)) are averaged over 5 runs,",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 2
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        112,
+                                        451,
+                                        126
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                112,
+                                                133,
+                                                126
+                                            ],
+                                            "score": 1.0,
+                                            "content": "where",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                133,
+                                                114,
+                                                143,
+                                                124
+                                            ],
+                                            "score": 0.67,
+                                            "content": "\\pm",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                144,
+                                                112,
+                                                451,
+                                                126
+                                            ],
+                                            "score": 1.0,
+                                            "content": "denotes the standard error. The highest scores are highlighted with boldface.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 3
+                                }
+                            ],
+                            "index": 1.5
+                        },
+                        {
+                            "type": "table_body",
+                            "bbox": [
+                                108,
+                                137,
+                                503,
+                                303
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        108,
+                                        137,
+                                        503,
+                                        303
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                108,
+                                                137,
+                                                503,
+                                                303
+                                            ],
+                                            "score": 0.981,
+                                            "html": "<table><tr><td>Dataset type</td><td>Environment</td><td>RepB-SDE (ours)</td><td>RP</td><td>Base</td><td>MOPO</td><td>MF</td><td>BC</td></tr><tr><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td></tr><tr><td>Random</td><td>Walker2d</td><td>21.1 ± 1.0</td><td>18.4</td><td>16.4</td><td>1.3</td><td>7.3 BEAR</td><td>0.0</td></tr><tr><td>Medium</td><td>Walker2d</td><td>72.1 ± 1.9</td><td>56.3</td><td>5.5</td><td>-0.1</td><td>81.1 BRAC</td><td>7.7</td></tr><tr><td>Med-Replay</td><td>Walker2d</td><td>49.8 ± 11.4</td><td>41.1</td><td>6.2</td><td>47.8</td><td>26.7 CQL</td><td>8.0</td></tr><tr><td>Med-Expert</td><td>Walker2d</td><td>88.8 ± 6.9</td><td>72.6</td><td>51.7</td><td>32.4</td><td>98.7 CQL</td><td>3.3</td></tr><tr><td>Random</td><td>Hopper</td><td>8.6 ±1.0</td><td>8.3</td><td>8.3</td><td>9.1</td><td>12.2 BRAC</td><td>9.0</td></tr><tr><td>Medium</td><td>Hopper</td><td>34.0 ± 2.8</td><td>27.5</td><td>19.8</td><td>19.2</td><td>58.0 CQL</td><td>34.5</td></tr><tr><td>Med-Replay</td><td>Hopper</td><td>62.2 ± 6.7</td><td>49.8</td><td>32.9</td><td>80.8</td><td>48.6 CQL</td><td>13.2</td></tr><tr><td>Med-Expert</td><td>Hopper</td><td>82.6 ± 7.0</td><td>74.0</td><td>19.1</td><td>23.2</td><td>111.0 CQL</td><td>41.9</td></tr><tr><td>Random</td><td>HalfCheetah</td><td>32.9 ± 1.1</td><td>31.3</td><td>26.1</td><td>29.9</td><td>35.4 CQL</td><td>-2.4</td></tr><tr><td>Medium</td><td>HalfCheetah</td><td>49.1 ± 0.3</td><td>47.3</td><td>22.1</td><td>45.0</td><td>46.3 BRAC</td><td>31.0</td></tr><tr><td>Med-Replay</td><td>HalfCheetah</td><td>57.5 ± 0.8</td><td>53.6</td><td>54.0</td><td>53.8</td><td>47.7 BRAC</td><td>14.7</td></tr><tr><td>Med-Expert</td><td>HalfCheetah</td><td>55.4 ± 8.3</td><td>36.7</td><td>31.9</td><td>91.0</td><td>64.7 BCQ</td><td>29.5</td></tr></table>",
+                                            "type": "table",
+                                            "image_path": "49f566ca115dde3869f3a7c649db4ca701dd408c978bc5e3af8a0e88cc803964.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 5,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        108,
+                                        137,
+                                        503,
+                                        192.33333333333334
+                                    ],
+                                    "spans": [],
+                                    "index": 4
+                                },
+                                {
+                                    "bbox": [
+                                        108,
+                                        192.33333333333334,
+                                        503,
+                                        247.66666666666669
+                                    ],
+                                    "spans": [],
+                                    "index": 5
+                                },
+                                {
+                                    "bbox": [
+                                        108,
+                                        247.66666666666669,
+                                        503,
+                                        303.0
+                                    ],
+                                    "spans": [],
+                                    "index": 6
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 3.25
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        104,
+                        335,
+                        476,
+                        345
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                334,
+                                478,
+                                347
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        334,
+                                        478,
+                                        347
+                                    ],
+                                    "score": 1.0,
+                                    "content": "5.2 OFFLINE REINFORCEMENT LEARNING WITH REPRESENTATION BALANCED MODEL",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        }
+                    ],
+                    "index": 7
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        357,
+                        505,
+                        468
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                358,
+                                505,
+                                370
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        358,
+                                        505,
+                                        370
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We evaluate the offline model-based RL algorithm presented in Section 4.2 on a subset of datasets",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                368,
+                                505,
+                                381
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        368,
+                                        505,
+                                        381
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in the D4RL benchmark (Fu et al., 2020): using four types of datasets (Random, Medium, Medium-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                380,
+                                505,
+                                393
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        380,
+                                        505,
+                                        393
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Replay, and Medium-Expert) from three different MuJoCo environments (HalfCheetah-v2, Hopper-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                390,
+                                505,
+                                403
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        390,
+                                        394,
+                                        403
+                                    ],
+                                    "score": 1.0,
+                                    "content": "v2, and Walker2d-v2) (Todorov et al., 2012). Random dataset contains",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        394,
+                                        390,
+                                        410,
+                                        401
+                                    ],
+                                    "score": 0.92,
+                                    "content": "1 0 ^ { 6 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        410,
+                                        390,
+                                        505,
+                                        403
+                                    ],
+                                    "score": 1.0,
+                                    "content": "experience tuples from",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                401,
+                                505,
+                                415
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        402,
+                                        283,
+                                        415
+                                    ],
+                                    "score": 1.0,
+                                    "content": "a random policy. Medium dataset contains",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        284,
+                                        401,
+                                        299,
+                                        412
+                                    ],
+                                    "score": 0.87,
+                                    "content": "1 0 ^ { 6 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        299,
+                                        402,
+                                        505,
+                                        415
+                                    ],
+                                    "score": 1.0,
+                                    "content": "experience tuples from a policy trained to approx-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                412,
+                                505,
+                                425
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        412,
+                                        140,
+                                        425
+                                    ],
+                                    "score": 1.0,
+                                    "content": "imately",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        140,
+                                        413,
+                                        157,
+                                        424
+                                    ],
+                                    "score": 0.35,
+                                    "content": "1 / 3",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        157,
+                                        412,
+                                        505,
+                                        425
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the performance of the expert, which is an agent trained to completion with SAC.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                422,
+                                505,
+                                437
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        422,
+                                        227,
+                                        437
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Med-Replay dataset contains",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        227,
+                                        423,
+                                        243,
+                                        434
+                                    ],
+                                    "score": 0.86,
+                                    "content": "1 0 ^ { 5 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        243,
+                                        422,
+                                        247,
+                                        437
+                                    ],
+                                    "score": 0.0,
+                                    "content": "",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        248,
+                                        423,
+                                        280,
+                                        434
+                                    ],
+                                    "score": 0.89,
+                                    "content": "2 \\times \\mathrm { i 0 ^ { 5 } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        280,
+                                        422,
+                                        505,
+                                        437
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for Walker2d-v2) experience tuples, which are from the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                435,
+                                506,
+                                447
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        435,
+                                        506,
+                                        447
+                                    ],
+                                    "score": 1.0,
+                                    "content": "replay buffer of a policy trained up to the performance of the medium agent. Med-Expert dataset",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                444,
+                                505,
+                                459
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        444,
+                                        252,
+                                        459
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is a Medium dataset combined with",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        253,
+                                        445,
+                                        268,
+                                        456
+                                    ],
+                                    "score": 0.89,
+                                    "content": "1 0 ^ { 6 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        269,
+                                        444,
+                                        505,
+                                        459
+                                    ],
+                                    "score": 1.0,
+                                    "content": "samples from the expert. This experimental setting exactly",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                456,
+                                375,
+                                469
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        456,
+                                        375,
+                                        469
+                                    ],
+                                    "score": 1.0,
+                                    "content": "follows that of (Yu et al., 2020; Argenson & Dulac-Arnold, 2020).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        }
+                    ],
+                    "index": 12.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        473,
+                        505,
+                        616
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                473,
+                                505,
+                                486
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        473,
+                                        505,
+                                        486
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The normalized score of each algorithm is presented in Table 1. MF denotes the best score from",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                483,
+                                506,
+                                498
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        483,
+                                        506,
+                                        498
+                                    ],
+                                    "score": 1.0,
+                                    "content": "offline model-free algorithms (taken from Fu et al. (2020) and Kumar et al. (2020)), including",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                496,
+                                505,
+                                507
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        496,
+                                        505,
+                                        507
+                                    ],
+                                    "score": 1.0,
+                                    "content": "SAC (Haarnoja et al., 2018), BCQ (Fujimoto et al., 2019), BEAR (Kumar et al., 2019), BRAC (Wu",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                506,
+                                506,
+                                519
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        506,
+                                        506,
+                                        519
+                                    ],
+                                    "score": 1.0,
+                                    "content": "et al., 2019), AWR (Peng et al., 2019), cREM (Agarwal et al., 2020), AlgaeDICE (Nachum et al.,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                516,
+                                506,
+                                530
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        516,
+                                        506,
+                                        530
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2019b), and CQL (Kumar et al., 2020). The actual algorithm that achieves the reported score is",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                528,
+                                506,
+                                541
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        528,
+                                        506,
+                                        541
+                                    ],
+                                    "score": 1.0,
+                                    "content": "presented next to the numbers. Base shows the performance of the most naive baseline, which at-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                538,
+                                505,
+                                551
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        538,
+                                        505,
+                                        551
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tempts to maximize the estimated policy return under the maximum-likelihood model. RP denotes",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                550,
+                                505,
+                                563
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        550,
+                                        505,
+                                        563
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the performance of Base equipped with the appropriate reward penalty using the bootstrapped un-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                561,
+                                505,
+                                574
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        561,
+                                        291,
+                                        574
+                                    ],
+                                    "score": 1.0,
+                                    "content": "certainty of the model, which is equivalent to",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        291,
+                                        563,
+                                        303,
+                                        572
+                                    ],
+                                    "score": 0.86,
+                                    "content": "\\pi _ { 0 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        303,
+                                        561,
+                                        505,
+                                        574
+                                    ],
+                                    "score": 1.0,
+                                    "content": "described in Section 4.2. RepB-SDE denotes the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                572,
+                                505,
+                                585
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        572,
+                                        398,
+                                        585
+                                    ],
+                                    "score": 1.0,
+                                    "content": "performance after a single iteration of our algorithm, corresponding to",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        398,
+                                        574,
+                                        409,
+                                        583
+                                    ],
+                                    "score": 0.84,
+                                    "content": "\\pi _ { 1 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        410,
+                                        572,
+                                        505,
+                                        585
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". We also provide BC,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                583,
+                                505,
+                                596
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        583,
+                                        505,
+                                        596
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the performance of direct behavior cloning from the data, and MOPO (Yu et al., 2020), an offline",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                594,
+                                506,
+                                608
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        594,
+                                        506,
+                                        608
+                                    ],
+                                    "score": 1.0,
+                                    "content": "model-based RL algorithm that optimizes policy based on truncated rollouts with the heuristic re-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                603,
+                                162,
+                                619
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        603,
+                                        162,
+                                        619
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ward penalty.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        }
+                    ],
+                    "index": 24
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        622,
+                        505,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                621,
+                                505,
+                                635
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        621,
+                                        505,
+                                        635
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The significant gap between RepB-SDE and RP in the results shows the advantage brought by our",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                633,
+                                505,
+                                646
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        633,
+                                        505,
+                                        646
+                                    ],
+                                    "score": 1.0,
+                                    "content": "framework that encourages balanced representation. While our approach was less successful on",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                644,
+                                505,
+                                656
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        644,
+                                        505,
+                                        656
+                                    ],
+                                    "score": 1.0,
+                                    "content": "some of the datasets (mostly on the Hopper-v2 environment), we hypothesize that the conservative",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                656,
+                                505,
+                                667
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        656,
+                                        505,
+                                        667
+                                    ],
+                                    "score": 1.0,
+                                    "content": "training techniques: the behavior regularization approaches exploited in the model-free algorithms,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                666,
+                                505,
+                                678
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        666,
+                                        505,
+                                        678
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the rollout truncation technique in MOPO, and the pessimistic training based on the bootstrapped",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                677,
+                                504,
+                                689
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        677,
+                                        504,
+                                        689
+                                    ],
+                                    "score": 1.0,
+                                    "content": "uncertainty estimates adopted in our algorithm exhibit their strengths in different datasets. For ex-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                687,
+                                505,
+                                700
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        687,
+                                        505,
+                                        700
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ample, it may be the case that the ensemble models are overconfident especially in Hopper-v2, and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                699,
+                                505,
+                                711
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        699,
+                                        505,
+                                        711
+                                    ],
+                                    "score": 1.0,
+                                    "content": "should be regularized with more explicit methods. Nevertheless, we emphasize that the presented",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                709,
+                                505,
+                                723
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        709,
+                                        505,
+                                        723
+                                    ],
+                                    "score": 1.0,
+                                    "content": "framework can be used jointly with any other conservative training technique to improve their per-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                720,
+                                150,
+                                734
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        720,
+                                        150,
+                                        734
+                                    ],
+                                    "score": 1.0,
+                                    "content": "formance.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        }
+                    ],
+                    "index": 35.5
+                }
+            ],
+            "page_idx": 7,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        108,
+                        27,
+                        292,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                25,
+                                293,
+                                39
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        25,
+                                        293,
+                                        39
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Published as a conference paper at ICLR 2021",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        302,
+                        751,
+                        308,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                300,
+                                750,
+                                309,
+                                761
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        300,
+                                        750,
+                                        309,
+                                        761
+                                    ],
+                                    "score": 1.0,
+                                    "content": "8",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "table",
+                    "bbox": [
+                        108,
+                        137,
+                        503,
+                        303
+                    ],
+                    "blocks": [
+                        {
+                            "type": "table_caption",
+                            "bbox": [
+                                106,
+                                80,
+                                505,
+                                125
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        79,
+                                        505,
+                                        93
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                79,
+                                                505,
+                                                93
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Table 1: Normalized scores on D4RL MuJoCo benchmark datasets (Fu et al., 2020) where the score",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 0
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        91,
+                                        505,
+                                        104
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                91,
+                                                505,
+                                                104
+                                            ],
+                                            "score": 1.0,
+                                            "content": "of 0 corresponds to a random policy and 100 corresponds to a converged SAC policy. All results",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        101,
+                                        506,
+                                        116
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                101,
+                                                506,
+                                                116
+                                            ],
+                                            "score": 1.0,
+                                            "content": "(except MF, which is taken from Fu et al. (2020) and Kumar et al. (2020)) are averaged over 5 runs,",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 2
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        112,
+                                        451,
+                                        126
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                112,
+                                                133,
+                                                126
+                                            ],
+                                            "score": 1.0,
+                                            "content": "where",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                133,
+                                                114,
+                                                143,
+                                                124
+                                            ],
+                                            "score": 0.67,
+                                            "content": "\\pm",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                144,
+                                                112,
+                                                451,
+                                                126
+                                            ],
+                                            "score": 1.0,
+                                            "content": "denotes the standard error. The highest scores are highlighted with boldface.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 3
+                                }
+                            ],
+                            "index": 1.5
+                        },
+                        {
+                            "type": "table_body",
+                            "bbox": [
+                                108,
+                                137,
+                                503,
+                                303
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        108,
+                                        137,
+                                        503,
+                                        303
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                108,
+                                                137,
+                                                503,
+                                                303
+                                            ],
+                                            "score": 0.981,
+                                            "html": "<table><tr><td>Dataset type</td><td>Environment</td><td>RepB-SDE (ours)</td><td>RP</td><td>Base</td><td>MOPO</td><td>MF</td><td>BC</td></tr><tr><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td></tr><tr><td>Random</td><td>Walker2d</td><td>21.1 ± 1.0</td><td>18.4</td><td>16.4</td><td>1.3</td><td>7.3 BEAR</td><td>0.0</td></tr><tr><td>Medium</td><td>Walker2d</td><td>72.1 ± 1.9</td><td>56.3</td><td>5.5</td><td>-0.1</td><td>81.1 BRAC</td><td>7.7</td></tr><tr><td>Med-Replay</td><td>Walker2d</td><td>49.8 ± 11.4</td><td>41.1</td><td>6.2</td><td>47.8</td><td>26.7 CQL</td><td>8.0</td></tr><tr><td>Med-Expert</td><td>Walker2d</td><td>88.8 ± 6.9</td><td>72.6</td><td>51.7</td><td>32.4</td><td>98.7 CQL</td><td>3.3</td></tr><tr><td>Random</td><td>Hopper</td><td>8.6 ±1.0</td><td>8.3</td><td>8.3</td><td>9.1</td><td>12.2 BRAC</td><td>9.0</td></tr><tr><td>Medium</td><td>Hopper</td><td>34.0 ± 2.8</td><td>27.5</td><td>19.8</td><td>19.2</td><td>58.0 CQL</td><td>34.5</td></tr><tr><td>Med-Replay</td><td>Hopper</td><td>62.2 ± 6.7</td><td>49.8</td><td>32.9</td><td>80.8</td><td>48.6 CQL</td><td>13.2</td></tr><tr><td>Med-Expert</td><td>Hopper</td><td>82.6 ± 7.0</td><td>74.0</td><td>19.1</td><td>23.2</td><td>111.0 CQL</td><td>41.9</td></tr><tr><td>Random</td><td>HalfCheetah</td><td>32.9 ± 1.1</td><td>31.3</td><td>26.1</td><td>29.9</td><td>35.4 CQL</td><td>-2.4</td></tr><tr><td>Medium</td><td>HalfCheetah</td><td>49.1 ± 0.3</td><td>47.3</td><td>22.1</td><td>45.0</td><td>46.3 BRAC</td><td>31.0</td></tr><tr><td>Med-Replay</td><td>HalfCheetah</td><td>57.5 ± 0.8</td><td>53.6</td><td>54.0</td><td>53.8</td><td>47.7 BRAC</td><td>14.7</td></tr><tr><td>Med-Expert</td><td>HalfCheetah</td><td>55.4 ± 8.3</td><td>36.7</td><td>31.9</td><td>91.0</td><td>64.7 BCQ</td><td>29.5</td></tr></table>",
+                                            "type": "table",
+                                            "image_path": "49f566ca115dde3869f3a7c649db4ca701dd408c978bc5e3af8a0e88cc803964.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 5,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        108,
+                                        137,
+                                        503,
+                                        192.33333333333334
+                                    ],
+                                    "spans": [],
+                                    "index": 4
+                                },
+                                {
+                                    "bbox": [
+                                        108,
+                                        192.33333333333334,
+                                        503,
+                                        247.66666666666669
+                                    ],
+                                    "spans": [],
+                                    "index": 5
+                                },
+                                {
+                                    "bbox": [
+                                        108,
+                                        247.66666666666669,
+                                        503,
+                                        303.0
+                                    ],
+                                    "spans": [],
+                                    "index": 6
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 3.25
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        104,
+                        335,
+                        476,
+                        345
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                334,
+                                478,
+                                347
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        334,
+                                        478,
+                                        347
+                                    ],
+                                    "score": 1.0,
+                                    "content": "5.2 OFFLINE REINFORCEMENT LEARNING WITH REPRESENTATION BALANCED MODEL",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        }
+                    ],
+                    "index": 7,
+                    "bbox_fs": [
+                        105,
+                        334,
+                        478,
+                        347
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        357,
+                        505,
+                        468
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                358,
+                                505,
+                                370
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        358,
+                                        505,
+                                        370
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We evaluate the offline model-based RL algorithm presented in Section 4.2 on a subset of datasets",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                368,
+                                505,
+                                381
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        368,
+                                        505,
+                                        381
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in the D4RL benchmark (Fu et al., 2020): using four types of datasets (Random, Medium, Medium-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                380,
+                                505,
+                                393
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        380,
+                                        505,
+                                        393
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Replay, and Medium-Expert) from three different MuJoCo environments (HalfCheetah-v2, Hopper-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                390,
+                                505,
+                                403
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        390,
+                                        394,
+                                        403
+                                    ],
+                                    "score": 1.0,
+                                    "content": "v2, and Walker2d-v2) (Todorov et al., 2012). Random dataset contains",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        394,
+                                        390,
+                                        410,
+                                        401
+                                    ],
+                                    "score": 0.92,
+                                    "content": "1 0 ^ { 6 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        410,
+                                        390,
+                                        505,
+                                        403
+                                    ],
+                                    "score": 1.0,
+                                    "content": "experience tuples from",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                401,
+                                505,
+                                415
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        402,
+                                        283,
+                                        415
+                                    ],
+                                    "score": 1.0,
+                                    "content": "a random policy. Medium dataset contains",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        284,
+                                        401,
+                                        299,
+                                        412
+                                    ],
+                                    "score": 0.87,
+                                    "content": "1 0 ^ { 6 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        299,
+                                        402,
+                                        505,
+                                        415
+                                    ],
+                                    "score": 1.0,
+                                    "content": "experience tuples from a policy trained to approx-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                412,
+                                505,
+                                425
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        412,
+                                        140,
+                                        425
+                                    ],
+                                    "score": 1.0,
+                                    "content": "imately",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        140,
+                                        413,
+                                        157,
+                                        424
+                                    ],
+                                    "score": 0.35,
+                                    "content": "1 / 3",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        157,
+                                        412,
+                                        505,
+                                        425
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the performance of the expert, which is an agent trained to completion with SAC.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                422,
+                                505,
+                                437
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        422,
+                                        227,
+                                        437
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Med-Replay dataset contains",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        227,
+                                        423,
+                                        243,
+                                        434
+                                    ],
+                                    "score": 0.86,
+                                    "content": "1 0 ^ { 5 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        243,
+                                        422,
+                                        247,
+                                        437
+                                    ],
+                                    "score": 0.0,
+                                    "content": "",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        248,
+                                        423,
+                                        280,
+                                        434
+                                    ],
+                                    "score": 0.89,
+                                    "content": "2 \\times \\mathrm { i 0 ^ { 5 } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        280,
+                                        422,
+                                        505,
+                                        437
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for Walker2d-v2) experience tuples, which are from the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                435,
+                                506,
+                                447
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        435,
+                                        506,
+                                        447
+                                    ],
+                                    "score": 1.0,
+                                    "content": "replay buffer of a policy trained up to the performance of the medium agent. Med-Expert dataset",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                444,
+                                505,
+                                459
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        444,
+                                        252,
+                                        459
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is a Medium dataset combined with",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        253,
+                                        445,
+                                        268,
+                                        456
+                                    ],
+                                    "score": 0.89,
+                                    "content": "1 0 ^ { 6 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        269,
+                                        444,
+                                        505,
+                                        459
+                                    ],
+                                    "score": 1.0,
+                                    "content": "samples from the expert. This experimental setting exactly",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                456,
+                                375,
+                                469
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        456,
+                                        375,
+                                        469
+                                    ],
+                                    "score": 1.0,
+                                    "content": "follows that of (Yu et al., 2020; Argenson & Dulac-Arnold, 2020).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        }
+                    ],
+                    "index": 12.5,
+                    "bbox_fs": [
+                        104,
+                        358,
+                        506,
+                        469
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        473,
+                        505,
+                        616
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                473,
+                                505,
+                                486
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        473,
+                                        505,
+                                        486
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The normalized score of each algorithm is presented in Table 1. MF denotes the best score from",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                483,
+                                506,
+                                498
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        483,
+                                        506,
+                                        498
+                                    ],
+                                    "score": 1.0,
+                                    "content": "offline model-free algorithms (taken from Fu et al. (2020) and Kumar et al. (2020)), including",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                496,
+                                505,
+                                507
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        496,
+                                        505,
+                                        507
+                                    ],
+                                    "score": 1.0,
+                                    "content": "SAC (Haarnoja et al., 2018), BCQ (Fujimoto et al., 2019), BEAR (Kumar et al., 2019), BRAC (Wu",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                506,
+                                506,
+                                519
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        506,
+                                        506,
+                                        519
+                                    ],
+                                    "score": 1.0,
+                                    "content": "et al., 2019), AWR (Peng et al., 2019), cREM (Agarwal et al., 2020), AlgaeDICE (Nachum et al.,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                516,
+                                506,
+                                530
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        516,
+                                        506,
+                                        530
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2019b), and CQL (Kumar et al., 2020). The actual algorithm that achieves the reported score is",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                528,
+                                506,
+                                541
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        528,
+                                        506,
+                                        541
+                                    ],
+                                    "score": 1.0,
+                                    "content": "presented next to the numbers. Base shows the performance of the most naive baseline, which at-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                538,
+                                505,
+                                551
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        538,
+                                        505,
+                                        551
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tempts to maximize the estimated policy return under the maximum-likelihood model. RP denotes",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                550,
+                                505,
+                                563
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        550,
+                                        505,
+                                        563
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the performance of Base equipped with the appropriate reward penalty using the bootstrapped un-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                561,
+                                505,
+                                574
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        561,
+                                        291,
+                                        574
+                                    ],
+                                    "score": 1.0,
+                                    "content": "certainty of the model, which is equivalent to",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        291,
+                                        563,
+                                        303,
+                                        572
+                                    ],
+                                    "score": 0.86,
+                                    "content": "\\pi _ { 0 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        303,
+                                        561,
+                                        505,
+                                        574
+                                    ],
+                                    "score": 1.0,
+                                    "content": "described in Section 4.2. RepB-SDE denotes the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                572,
+                                505,
+                                585
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        572,
+                                        398,
+                                        585
+                                    ],
+                                    "score": 1.0,
+                                    "content": "performance after a single iteration of our algorithm, corresponding to",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        398,
+                                        574,
+                                        409,
+                                        583
+                                    ],
+                                    "score": 0.84,
+                                    "content": "\\pi _ { 1 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        410,
+                                        572,
+                                        505,
+                                        585
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". We also provide BC,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                583,
+                                505,
+                                596
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        583,
+                                        505,
+                                        596
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the performance of direct behavior cloning from the data, and MOPO (Yu et al., 2020), an offline",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                594,
+                                506,
+                                608
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        594,
+                                        506,
+                                        608
+                                    ],
+                                    "score": 1.0,
+                                    "content": "model-based RL algorithm that optimizes policy based on truncated rollouts with the heuristic re-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                603,
+                                162,
+                                619
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        603,
+                                        162,
+                                        619
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ward penalty.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        }
+                    ],
+                    "index": 24,
+                    "bbox_fs": [
+                        105,
+                        473,
+                        506,
+                        619
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        622,
+                        505,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                621,
+                                505,
+                                635
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        621,
+                                        505,
+                                        635
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The significant gap between RepB-SDE and RP in the results shows the advantage brought by our",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                633,
+                                505,
+                                646
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        633,
+                                        505,
+                                        646
+                                    ],
+                                    "score": 1.0,
+                                    "content": "framework that encourages balanced representation. While our approach was less successful on",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                644,
+                                505,
+                                656
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        644,
+                                        505,
+                                        656
+                                    ],
+                                    "score": 1.0,
+                                    "content": "some of the datasets (mostly on the Hopper-v2 environment), we hypothesize that the conservative",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                656,
+                                505,
+                                667
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        656,
+                                        505,
+                                        667
+                                    ],
+                                    "score": 1.0,
+                                    "content": "training techniques: the behavior regularization approaches exploited in the model-free algorithms,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                666,
+                                505,
+                                678
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        666,
+                                        505,
+                                        678
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the rollout truncation technique in MOPO, and the pessimistic training based on the bootstrapped",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                677,
+                                504,
+                                689
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        677,
+                                        504,
+                                        689
+                                    ],
+                                    "score": 1.0,
+                                    "content": "uncertainty estimates adopted in our algorithm exhibit their strengths in different datasets. For ex-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                687,
+                                505,
+                                700
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        687,
+                                        505,
+                                        700
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ample, it may be the case that the ensemble models are overconfident especially in Hopper-v2, and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                699,
+                                505,
+                                711
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        699,
+                                        505,
+                                        711
+                                    ],
+                                    "score": 1.0,
+                                    "content": "should be regularized with more explicit methods. Nevertheless, we emphasize that the presented",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                709,
+                                505,
+                                723
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        709,
+                                        505,
+                                        723
+                                    ],
+                                    "score": 1.0,
+                                    "content": "framework can be used jointly with any other conservative training technique to improve their per-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                720,
+                                150,
+                                734
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        720,
+                                        150,
+                                        734
+                                    ],
+                                    "score": 1.0,
+                                    "content": "formance.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        }
+                    ],
+                    "index": 35.5,
+                    "bbox_fs": [
+                        105,
+                        621,
+                        505,
+                        734
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        81,
+                        303,
+                        93
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                79,
+                                304,
+                                96
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        79,
+                                        304,
+                                        96
+                                    ],
+                                    "score": 1.0,
+                                    "content": "6 CONCLUSION AND FUTURE WORK",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        }
+                    ],
+                    "index": 0
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        107,
+                        505,
+                        227
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                106,
+                                505,
+                                120
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        106,
+                                        505,
+                                        120
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In this paper, we presented RepB-SDE, a framework for balancing the model representation with sta-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                118,
+                                505,
+                                130
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        118,
+                                        505,
+                                        130
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tionary distribution estimation, aiming at obtaining a model robust to the distribution shift that arises",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                127,
+                                505,
+                                143
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        127,
+                                        505,
+                                        143
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in off-policy and offline RL. We started from the theoretical observation that the model-based policy",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                140,
+                                505,
+                                151
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        140,
+                                        505,
+                                        151
+                                    ],
+                                    "score": 1.0,
+                                    "content": "evaluation error can be upper-bounded by the data fitness and the distance between two distributions",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                151,
+                                505,
+                                163
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        151,
+                                        505,
+                                        163
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in the representation space. Motivated by the bound, we presented a model learning objective for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                162,
+                                506,
+                                174
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        162,
+                                        506,
+                                        174
+                                    ],
+                                    "score": 1.0,
+                                    "content": "off-policy evaluation and model-based offline policy optimization. RepB-SDE can be seen as an",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                172,
+                                506,
+                                186
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        172,
+                                        506,
+                                        186
+                                    ],
+                                    "score": 1.0,
+                                    "content": "extension of RepBM, which addresses the curse of horizon by leveraging the recent advances in",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                183,
+                                505,
+                                196
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        183,
+                                        505,
+                                        196
+                                    ],
+                                    "score": 1.0,
+                                    "content": "stationary distribution correction estimation (i.e. the DualDICE trick). Using stationary distribution",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                194,
+                                505,
+                                207
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        194,
+                                        505,
+                                        207
+                                    ],
+                                    "score": 1.0,
+                                    "content": "also frees us from other limitations of RepBM to be applied to more practical settings. To the best of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                205,
+                                505,
+                                218
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        205,
+                                        505,
+                                        218
+                                    ],
+                                    "score": 1.0,
+                                    "content": "our knowledge, it is the first attempt to introduce an augmented objective for the learning of model",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                217,
+                                308,
+                                228
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        217,
+                                        308,
+                                        228
+                                    ],
+                                    "score": 1.0,
+                                    "content": "robust to a specific distribution shift in offline RL.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        }
+                    ],
+                    "index": 6
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        233,
+                        504,
+                        288
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                234,
+                                505,
+                                245
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        234,
+                                        505,
+                                        245
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In the experiments, we empirically demonstrated that we can significantly reduce the OPE error",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                245,
+                                505,
+                                257
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        245,
+                                        505,
+                                        257
+                                    ],
+                                    "score": 1.0,
+                                    "content": "from the baseline, outperforming RepBM in most cases. We also showed that the robust model also",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                255,
+                                506,
+                                268
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        255,
+                                        506,
+                                        268
+                                    ],
+                                    "score": 1.0,
+                                    "content": "helps in the offline model-based policy optimization, yielding the state-of-the-art performance in a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                266,
+                                506,
+                                280
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        266,
+                                        506,
+                                        280
+                                    ],
+                                    "score": 1.0,
+                                    "content": "representative set of D4RL benchmarks. We emphasize that our approach can be directly adopted in",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                278,
+                                298,
+                                290
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        278,
+                                        298,
+                                        290
+                                    ],
+                                    "score": 1.0,
+                                    "content": "many other model-based offline RL algorithms.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        }
+                    ],
+                    "index": 14
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        294,
+                        505,
+                        360
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                293,
+                                505,
+                                307
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        293,
+                                        505,
+                                        307
+                                    ],
+                                    "score": 1.0,
+                                    "content": "There are a number of promising directions for future work. Most importantly, we have not leveraged",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                304,
+                                505,
+                                318
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        304,
+                                        505,
+                                        318
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the learned representation in the policy when optimizing the policy, yet it is very natural to do so.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                315,
+                                506,
+                                330
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        315,
+                                        506,
+                                        330
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We can easily incorporate the representation into the policy by assuming energy-based models, but",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                326,
+                                505,
+                                340
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        326,
+                                        505,
+                                        340
+                                    ],
+                                    "score": 1.0,
+                                    "content": "this would make the computation of entropy intractable in entropy-regularized policy optimization",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                337,
+                                505,
+                                350
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        337,
+                                        505,
+                                        350
+                                    ],
+                                    "score": 1.0,
+                                    "content": "algorithms. It would be also interesting to see if the proposed framework for learning balanced",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                349,
+                                390,
+                                361
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        349,
+                                        390,
+                                        361
+                                    ],
+                                    "score": 1.0,
+                                    "content": "representation can benefit off-policy (and offline) model-free methods.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        }
+                    ],
+                    "index": 19.5
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        375,
+                        200,
+                        384
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                107,
+                                375,
+                                200,
+                                385
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        375,
+                                        200,
+                                        385
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ACKNOWLEDGMENTS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        }
+                    ],
+                    "index": 23
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        394,
+                        505,
+                        437
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                393,
+                                505,
+                                407
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        393,
+                                        505,
+                                        407
+                                    ],
+                                    "score": 1.0,
+                                    "content": "This work was supported by the National Research Foundation (NRF) of Korea (NRF-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                404,
+                                505,
+                                417
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        404,
+                                        505,
+                                        417
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2019M3F2A1072238 and NRF-2019R1A2C1087634), and the Ministry of Science and Information",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                416,
+                                505,
+                                428
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        416,
+                                        505,
+                                        428
+                                    ],
+                                    "score": 1.0,
+                                    "content": "communication Technology (MSIT) of Korea (IITP No. 2019-0-00075, IITP No. 2020-0-00940 and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                426,
+                                229,
+                                439
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        426,
+                                        229,
+                                        439
+                                    ],
+                                    "score": 1.0,
+                                    "content": "IITP No. 2017-0-01779 XAI).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        }
+                    ],
+                    "index": 25.5
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        456,
+                        175,
+                        468
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                456,
+                                176,
+                                468
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        456,
+                                        176,
+                                        468
+                                    ],
+                                    "score": 1.0,
+                                    "content": "REFERENCES",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        }
+                    ],
+                    "index": 28
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        475,
+                        504,
+                        497
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                474,
+                                505,
+                                487
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        474,
+                                        505,
+                                        487
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Rishabh Agarwal, Dale Schuurmans, and Mohammad Norouzi. An optimistic perspective on offline",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                485,
+                                444,
+                                498
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        485,
+                                        444,
+                                        498
+                                    ],
+                                    "score": 1.0,
+                                    "content": "reinforcement learning. In International Conference on Machine Learning, 2020.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        }
+                    ],
+                    "index": 29.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        104,
+                        506,
+                        505,
+                        529
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                505,
+                                505,
+                                520
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        505,
+                                        505,
+                                        520
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Arthur Argenson and Gabriel Dulac-Arnold. Model-based offline planning. arXiv preprint",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                518,
+                                218,
+                                529
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        518,
+                                        218,
+                                        529
+                                    ],
+                                    "score": 1.0,
+                                    "content": "arXiv:2008.05556, 2020.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        }
+                    ],
+                    "index": 31.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        538,
+                        504,
+                        561
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                538,
+                                505,
+                                550
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        538,
+                                        505,
+                                        550
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Peter L Bartlett and Shahar Mendelson. Rademacher and Gaussian complexities: Risk bounds and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                549,
+                                446,
+                                562
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        549,
+                                        446,
+                                        562
+                                    ],
+                                    "score": 1.0,
+                                    "content": "structural results. Journal of Machine Learning Research, 3(Nov):463–482, 2002.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        }
+                    ],
+                    "index": 33.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        570,
+                        504,
+                        594
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                569,
+                                506,
+                                583
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        569,
+                                        506,
+                                        583
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Shai Ben-David, John Blitzer, Koby Crammer, and Fernando Pereira. Analysis of representations",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                581,
+                                460,
+                                594
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        581,
+                                        460,
+                                        594
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for domain adaptation. In Advances in Neural Information Processing Systems, 2007.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        }
+                    ],
+                    "index": 35.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        602,
+                        504,
+                        625
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                603,
+                                505,
+                                615
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        603,
+                                        505,
+                                        615
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Greg Brockman, Vicki Cheung, Ludwig Pettersson, Jonas Schneider, John Schulman, Jie Tang, and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                614,
+                                420,
+                                626
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        614,
+                                        420,
+                                        626
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Wojciech Zaremba. OpenAI Gym. arXiv preprint arXiv:1606.01540, 2016.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        }
+                    ],
+                    "index": 37.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        634,
+                        503,
+                        668
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                634,
+                                505,
+                                648
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        634,
+                                        505,
+                                        648
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Kurtland Chua, Roberto Calandra, Rowan McAllister, and Sergey Levine. Deep reinforcement learn-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                646,
+                                505,
+                                658
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        646,
+                                        505,
+                                        658
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ing in a handful of trials using probabilistic dynamics models. In Advances in Neural Information",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                657,
+                                224,
+                                669
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        657,
+                                        224,
+                                        669
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Processing Systems, 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        }
+                    ],
+                    "index": 40
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        104,
+                        677,
+                        504,
+                        700
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                676,
+                                505,
+                                690
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        676,
+                                        505,
+                                        690
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Justin Fu, Aviral Kumar, Ofir Nachum, George Tucker, and Sergey Levine. D4RL: Datasets for deep",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                688,
+                                423,
+                                701
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        688,
+                                        423,
+                                        701
+                                    ],
+                                    "score": 1.0,
+                                    "content": "data-driven reinforcement learning. arXiv preprint arXiv:2004.07219, 2020.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        }
+                    ],
+                    "index": 42.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        709,
+                        502,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                709,
+                                504,
+                                722
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        709,
+                                        504,
+                                        722
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Scott Fujimoto, David Meger, and Doina Precup. Off-policy deep reinforcement learning without",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                720,
+                                398,
+                                733
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        720,
+                                        398,
+                                        733
+                                    ],
+                                    "score": 1.0,
+                                    "content": "exploration. In International Conference on Machine Learning, 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        }
+                    ],
+                    "index": 44.5
+                }
+            ],
+            "page_idx": 8,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        108,
+                        27,
+                        292,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                25,
+                                293,
+                                39
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        25,
+                                        293,
+                                        39
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Published as a conference paper at ICLR 2021",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        302,
+                        751,
+                        308,
+                        759
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                302,
+                                751,
+                                309,
+                                762
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        302,
+                                        751,
+                                        309,
+                                        762
+                                    ],
+                                    "score": 1.0,
+                                    "content": "9",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        81,
+                        303,
+                        93
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                79,
+                                304,
+                                96
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        79,
+                                        304,
+                                        96
+                                    ],
+                                    "score": 1.0,
+                                    "content": "6 CONCLUSION AND FUTURE WORK",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        }
+                    ],
+                    "index": 0
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        107,
+                        505,
+                        227
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                106,
+                                505,
+                                120
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        106,
+                                        505,
+                                        120
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In this paper, we presented RepB-SDE, a framework for balancing the model representation with sta-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                118,
+                                505,
+                                130
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        118,
+                                        505,
+                                        130
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tionary distribution estimation, aiming at obtaining a model robust to the distribution shift that arises",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                127,
+                                505,
+                                143
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        127,
+                                        505,
+                                        143
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in off-policy and offline RL. We started from the theoretical observation that the model-based policy",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                140,
+                                505,
+                                151
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        140,
+                                        505,
+                                        151
+                                    ],
+                                    "score": 1.0,
+                                    "content": "evaluation error can be upper-bounded by the data fitness and the distance between two distributions",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                151,
+                                505,
+                                163
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        151,
+                                        505,
+                                        163
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in the representation space. Motivated by the bound, we presented a model learning objective for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                162,
+                                506,
+                                174
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        162,
+                                        506,
+                                        174
+                                    ],
+                                    "score": 1.0,
+                                    "content": "off-policy evaluation and model-based offline policy optimization. RepB-SDE can be seen as an",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                172,
+                                506,
+                                186
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        172,
+                                        506,
+                                        186
+                                    ],
+                                    "score": 1.0,
+                                    "content": "extension of RepBM, which addresses the curse of horizon by leveraging the recent advances in",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                183,
+                                505,
+                                196
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        183,
+                                        505,
+                                        196
+                                    ],
+                                    "score": 1.0,
+                                    "content": "stationary distribution correction estimation (i.e. the DualDICE trick). Using stationary distribution",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                194,
+                                505,
+                                207
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        194,
+                                        505,
+                                        207
+                                    ],
+                                    "score": 1.0,
+                                    "content": "also frees us from other limitations of RepBM to be applied to more practical settings. To the best of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                205,
+                                505,
+                                218
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        205,
+                                        505,
+                                        218
+                                    ],
+                                    "score": 1.0,
+                                    "content": "our knowledge, it is the first attempt to introduce an augmented objective for the learning of model",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                217,
+                                308,
+                                228
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        217,
+                                        308,
+                                        228
+                                    ],
+                                    "score": 1.0,
+                                    "content": "robust to a specific distribution shift in offline RL.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        }
+                    ],
+                    "index": 6,
+                    "bbox_fs": [
+                        105,
+                        106,
+                        506,
+                        228
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        233,
+                        504,
+                        288
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                234,
+                                505,
+                                245
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        234,
+                                        505,
+                                        245
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In the experiments, we empirically demonstrated that we can significantly reduce the OPE error",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                245,
+                                505,
+                                257
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        245,
+                                        505,
+                                        257
+                                    ],
+                                    "score": 1.0,
+                                    "content": "from the baseline, outperforming RepBM in most cases. We also showed that the robust model also",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                255,
+                                506,
+                                268
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        255,
+                                        506,
+                                        268
+                                    ],
+                                    "score": 1.0,
+                                    "content": "helps in the offline model-based policy optimization, yielding the state-of-the-art performance in a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                266,
+                                506,
+                                280
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        266,
+                                        506,
+                                        280
+                                    ],
+                                    "score": 1.0,
+                                    "content": "representative set of D4RL benchmarks. We emphasize that our approach can be directly adopted in",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                278,
+                                298,
+                                290
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        278,
+                                        298,
+                                        290
+                                    ],
+                                    "score": 1.0,
+                                    "content": "many other model-based offline RL algorithms.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        }
+                    ],
+                    "index": 14,
+                    "bbox_fs": [
+                        105,
+                        234,
+                        506,
+                        290
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        294,
+                        505,
+                        360
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                293,
+                                505,
+                                307
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        293,
+                                        505,
+                                        307
+                                    ],
+                                    "score": 1.0,
+                                    "content": "There are a number of promising directions for future work. Most importantly, we have not leveraged",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                304,
+                                505,
+                                318
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        304,
+                                        505,
+                                        318
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the learned representation in the policy when optimizing the policy, yet it is very natural to do so.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                315,
+                                506,
+                                330
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        315,
+                                        506,
+                                        330
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We can easily incorporate the representation into the policy by assuming energy-based models, but",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                326,
+                                505,
+                                340
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        326,
+                                        505,
+                                        340
+                                    ],
+                                    "score": 1.0,
+                                    "content": "this would make the computation of entropy intractable in entropy-regularized policy optimization",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                337,
+                                505,
+                                350
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        337,
+                                        505,
+                                        350
+                                    ],
+                                    "score": 1.0,
+                                    "content": "algorithms. It would be also interesting to see if the proposed framework for learning balanced",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                349,
+                                390,
+                                361
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        349,
+                                        390,
+                                        361
+                                    ],
+                                    "score": 1.0,
+                                    "content": "representation can benefit off-policy (and offline) model-free methods.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        }
+                    ],
+                    "index": 19.5,
+                    "bbox_fs": [
+                        105,
+                        293,
+                        506,
+                        361
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        375,
+                        200,
+                        384
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                107,
+                                375,
+                                200,
+                                385
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        375,
+                                        200,
+                                        385
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ACKNOWLEDGMENTS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        }
+                    ],
+                    "index": 23
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        394,
+                        505,
+                        437
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                393,
+                                505,
+                                407
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        393,
+                                        505,
+                                        407
+                                    ],
+                                    "score": 1.0,
+                                    "content": "This work was supported by the National Research Foundation (NRF) of Korea (NRF-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                404,
+                                505,
+                                417
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        404,
+                                        505,
+                                        417
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2019M3F2A1072238 and NRF-2019R1A2C1087634), and the Ministry of Science and Information",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                416,
+                                505,
+                                428
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        416,
+                                        505,
+                                        428
+                                    ],
+                                    "score": 1.0,
+                                    "content": "communication Technology (MSIT) of Korea (IITP No. 2019-0-00075, IITP No. 2020-0-00940 and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                426,
+                                229,
+                                439
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        426,
+                                        229,
+                                        439
+                                    ],
+                                    "score": 1.0,
+                                    "content": "IITP No. 2017-0-01779 XAI).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        }
+                    ],
+                    "index": 25.5,
+                    "bbox_fs": [
+                        105,
+                        393,
+                        505,
+                        439
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        456,
+                        175,
+                        468
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                456,
+                                176,
+                                468
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        456,
+                                        176,
+                                        468
+                                    ],
+                                    "score": 1.0,
+                                    "content": "REFERENCES",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        }
+                    ],
+                    "index": 28
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        475,
+                        504,
+                        497
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                474,
+                                505,
+                                487
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        474,
+                                        505,
+                                        487
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Rishabh Agarwal, Dale Schuurmans, and Mohammad Norouzi. An optimistic perspective on offline",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                485,
+                                444,
+                                498
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        485,
+                                        444,
+                                        498
+                                    ],
+                                    "score": 1.0,
+                                    "content": "reinforcement learning. In International Conference on Machine Learning, 2020.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        }
+                    ],
+                    "index": 29.5,
+                    "bbox_fs": [
+                        105,
+                        474,
+                        505,
+                        498
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        104,
+                        506,
+                        505,
+                        529
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                505,
+                                505,
+                                520
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        505,
+                                        505,
+                                        520
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Arthur Argenson and Gabriel Dulac-Arnold. Model-based offline planning. arXiv preprint",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                518,
+                                218,
+                                529
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        518,
+                                        218,
+                                        529
+                                    ],
+                                    "score": 1.0,
+                                    "content": "arXiv:2008.05556, 2020.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        }
+                    ],
+                    "index": 31.5,
+                    "bbox_fs": [
+                        105,
+                        505,
+                        505,
+                        529
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        538,
+                        504,
+                        561
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                538,
+                                505,
+                                550
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        538,
+                                        505,
+                                        550
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Peter L Bartlett and Shahar Mendelson. Rademacher and Gaussian complexities: Risk bounds and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                549,
+                                446,
+                                562
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        549,
+                                        446,
+                                        562
+                                    ],
+                                    "score": 1.0,
+                                    "content": "structural results. Journal of Machine Learning Research, 3(Nov):463–482, 2002.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        }
+                    ],
+                    "index": 33.5,
+                    "bbox_fs": [
+                        106,
+                        538,
+                        505,
+                        562
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        570,
+                        504,
+                        594
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                569,
+                                506,
+                                583
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        569,
+                                        506,
+                                        583
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Shai Ben-David, John Blitzer, Koby Crammer, and Fernando Pereira. Analysis of representations",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                581,
+                                460,
+                                594
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        581,
+                                        460,
+                                        594
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for domain adaptation. In Advances in Neural Information Processing Systems, 2007.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        }
+                    ],
+                    "index": 35.5,
+                    "bbox_fs": [
+                        105,
+                        569,
+                        506,
+                        594
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        602,
+                        504,
+                        625
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                603,
+                                505,
+                                615
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        603,
+                                        505,
+                                        615
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Greg Brockman, Vicki Cheung, Ludwig Pettersson, Jonas Schneider, John Schulman, Jie Tang, and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                614,
+                                420,
+                                626
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        614,
+                                        420,
+                                        626
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Wojciech Zaremba. OpenAI Gym. arXiv preprint arXiv:1606.01540, 2016.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        }
+                    ],
+                    "index": 37.5,
+                    "bbox_fs": [
+                        106,
+                        603,
+                        505,
+                        626
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        634,
+                        503,
+                        668
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                634,
+                                505,
+                                648
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        634,
+                                        505,
+                                        648
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Kurtland Chua, Roberto Calandra, Rowan McAllister, and Sergey Levine. Deep reinforcement learn-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                646,
+                                505,
+                                658
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        646,
+                                        505,
+                                        658
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ing in a handful of trials using probabilistic dynamics models. In Advances in Neural Information",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                657,
+                                224,
+                                669
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        657,
+                                        224,
+                                        669
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Processing Systems, 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        }
+                    ],
+                    "index": 40,
+                    "bbox_fs": [
+                        105,
+                        634,
+                        505,
+                        669
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        104,
+                        677,
+                        504,
+                        700
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                676,
+                                505,
+                                690
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        676,
+                                        505,
+                                        690
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Justin Fu, Aviral Kumar, Ofir Nachum, George Tucker, and Sergey Levine. D4RL: Datasets for deep",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                688,
+                                423,
+                                701
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        688,
+                                        423,
+                                        701
+                                    ],
+                                    "score": 1.0,
+                                    "content": "data-driven reinforcement learning. arXiv preprint arXiv:2004.07219, 2020.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        }
+                    ],
+                    "index": 42.5,
+                    "bbox_fs": [
+                        105,
+                        676,
+                        505,
+                        701
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        709,
+                        502,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                709,
+                                504,
+                                722
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        709,
+                                        504,
+                                        722
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Scott Fujimoto, David Meger, and Doina Precup. Off-policy deep reinforcement learning without",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                720,
+                                398,
+                                733
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        720,
+                                        398,
+                                        733
+                                    ],
+                                    "score": 1.0,
+                                    "content": "exploration. In International Conference on Machine Learning, 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        }
+                    ],
+                    "index": 44.5,
+                    "bbox_fs": [
+                        106,
+                        709,
+                        504,
+                        733
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        82,
+                        505,
+                        105
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                81,
+                                505,
+                                95
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        81,
+                                        505,
+                                        95
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Carles Gelada and Marc G Bellemare. Off-policy deep reinforcement learning by bootstrapping the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                93,
+                                389,
+                                106
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        93,
+                                        389,
+                                        106
+                                    ],
+                                    "score": 1.0,
+                                    "content": "covariate shift. In AAAI Conference on Artificial Intelligence, 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        }
+                    ],
+                    "index": 0.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        111,
+                        504,
+                        135
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                111,
+                                504,
+                                124
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        111,
+                                        504,
+                                        124
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Arthur Gretton, Karsten M Borgwardt, Malte J Rasch, Bernhard Scholkopf, and Alexander Smola.¨",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                123,
+                                484,
+                                136
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        123,
+                                        484,
+                                        136
+                                    ],
+                                    "score": 1.0,
+                                    "content": "A kernel two-sample test. Journal of Machine Learning Research, 13(Mar):723–773, 2012.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        }
+                    ],
+                    "index": 2.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        141,
+                        503,
+                        175
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                140,
+                                505,
+                                155
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        140,
+                                        505,
+                                        155
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Tuomas Haarnoja, Aurick Zhou, Pieter Abbeel, and Sergey Levine. Soft actor-critic: Off-policy",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                153,
+                                505,
+                                165
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        153,
+                                        505,
+                                        165
+                                    ],
+                                    "score": 1.0,
+                                    "content": "maximum entropy deep reinforcement learning with a stochastic actor. In International Confer-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                164,
+                                253,
+                                176
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        164,
+                                        253,
+                                        176
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ence on Machine Learning, 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        }
+                    ],
+                    "index": 5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        181,
+                        504,
+                        205
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                181,
+                                504,
+                                195
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        181,
+                                        504,
+                                        195
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Assaf Hallak and Shie Mannor. Consistent on-line off-policy evaluation. In International Confer-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                193,
+                                253,
+                                205
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        193,
+                                        253,
+                                        205
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ence on Machine Learning, 2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        }
+                    ],
+                    "index": 7.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        211,
+                        504,
+                        234
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                211,
+                                505,
+                                224
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        211,
+                                        505,
+                                        224
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Wassily Hoeffding. Probability inequalities for sums of bounded random variables. Journal of the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                222,
+                                340,
+                                234
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        222,
+                                        340,
+                                        234
+                                    ],
+                                    "score": 1.0,
+                                    "content": "American Statistical Association, 58(301):13–30, 1963.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        }
+                    ],
+                    "index": 9.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        240,
+                        503,
+                        264
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                240,
+                                505,
+                                254
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        240,
+                                        505,
+                                        254
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Michael Janner, Justin Fu, Marvin Zhang, and Sergey Levine. When to trust your model: Model-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                252,
+                                475,
+                                264
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        252,
+                                        475,
+                                        264
+                                    ],
+                                    "score": 1.0,
+                                    "content": "based policy optimization. In Advances in Neural Information Processing Systems, 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        }
+                    ],
+                    "index": 11.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        270,
+                        504,
+                        294
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                270,
+                                505,
+                                283
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        270,
+                                        505,
+                                        283
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Nan Jiang and Lihong Li. Doubly robust off-policy value evaluation for reinforcement learning. In",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                281,
+                                336,
+                                294
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        281,
+                                        336,
+                                        294
+                                    ],
+                                    "score": 1.0,
+                                    "content": "International Conference on Machine Learning, 2016.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        }
+                    ],
+                    "index": 13.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        299,
+                        503,
+                        323
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                299,
+                                504,
+                                313
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        299,
+                                        504,
+                                        313
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Fredrik D Johansson, Nathan Kallus, Uri Shalit, and David Sontag. Learning weighted representa-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                311,
+                                437,
+                                324
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        311,
+                                        437,
+                                        324
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tions for generalization across designs. arXiv preprint arXiv:1802.08598, 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        }
+                    ],
+                    "index": 15.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        329,
+                        504,
+                        353
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                329,
+                                505,
+                                342
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        329,
+                                        505,
+                                        342
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Michael Kearns and Satinder Singh. Near-optimal reinforcement learning in polynomial time. Ma-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                340,
+                                277,
+                                353
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        340,
+                                        277,
+                                        353
+                                    ],
+                                    "score": 1.0,
+                                    "content": "chine learning, 49(2-3):209–232, 2002.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        }
+                    ],
+                    "index": 17.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        358,
+                        504,
+                        393
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                359,
+                                505,
+                                371
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        359,
+                                        505,
+                                        371
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Rahul Kidambi, Aravind Rajeswaran, Praneeth Netrapalli, and Thorsten Joachims. MOReL: Model-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                370,
+                                505,
+                                383
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        370,
+                                        505,
+                                        383
+                                    ],
+                                    "score": 1.0,
+                                    "content": "based offline reinforcement learning. In Advances in Neural Information Processing Systems,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                380,
+                                142,
+                                393
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        380,
+                                        142,
+                                        393
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2020.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        }
+                    ],
+                    "index": 20
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        399,
+                        504,
+                        422
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                398,
+                                506,
+                                414
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        398,
+                                        506,
+                                        414
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Diederik P Kingma and Jimmy Ba. Adam: A method for stochastic optimization. arXiv preprint",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                411,
+                                214,
+                                422
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        411,
+                                        214,
+                                        422
+                                    ],
+                                    "score": 1.0,
+                                    "content": "arXiv:1412.6980, 2014.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        }
+                    ],
+                    "index": 22.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        429,
+                        504,
+                        463
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                428,
+                                505,
+                                443
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        428,
+                                        505,
+                                        443
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Aviral Kumar, Justin Fu, Matthew Soh, George Tucker, and Sergey Levine. Stabilizing off-policy",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                441,
+                                505,
+                                453
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        441,
+                                        505,
+                                        453
+                                    ],
+                                    "score": 1.0,
+                                    "content": "q-learning via bootstrapping error reduction. In Advances in Neural Information Processing Sys-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                451,
+                                165,
+                                463
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        451,
+                                        165,
+                                        463
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tems, 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        }
+                    ],
+                    "index": 25
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        469,
+                        504,
+                        493
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                469,
+                                505,
+                                482
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        469,
+                                        505,
+                                        482
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Aviral Kumar, Aurick Zhou, George Tucker, and Sergey Levine. Conservative Q-learning for offline",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                481,
+                                463,
+                                493
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        481,
+                                        463,
+                                        493
+                                    ],
+                                    "score": 1.0,
+                                    "content": "reinforcement learning. In Advances in Neural Information Processing Systems, 2020.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        }
+                    ],
+                    "index": 27.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        498,
+                        504,
+                        523
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                499,
+                                505,
+                                512
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        499,
+                                        505,
+                                        512
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Byung-Jun Lee, Jongmin Lee, Peter Vrancx, Dongho Kim, and Kee-Eung Kim. Batch reinforcement",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                510,
+                                505,
+                                523
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        510,
+                                        505,
+                                        523
+                                    ],
+                                    "score": 1.0,
+                                    "content": "learning with hyperparameter gradients. In International Conference on Machine Learning, 2020.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        }
+                    ],
+                    "index": 29.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        528,
+                        503,
+                        552
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                528,
+                                505,
+                                541
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        528,
+                                        505,
+                                        541
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Sergey Levine, Aviral Kumar, George Tucker, and Justin Fu. Offline reinforcement learning: Tuto-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                540,
+                                474,
+                                553
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        540,
+                                        474,
+                                        553
+                                    ],
+                                    "score": 1.0,
+                                    "content": "rial, review, and perspectives on open problems. arXiv preprint arXiv:2005.01643, 2020.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        }
+                    ],
+                    "index": 31.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        558,
+                        503,
+                        592
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                557,
+                                505,
+                                572
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        557,
+                                        505,
+                                        572
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Timothy P Lillicrap, Jonathan J Hunt, Alexander Pritzel, Nicolas Heess, Tom Erez, Yuval Tassa,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                569,
+                                505,
+                                582
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        569,
+                                        505,
+                                        582
+                                    ],
+                                    "score": 1.0,
+                                    "content": "David Silver, and Daan Wierstra. Continuous control with deep reinforcement learning. In Inter-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                581,
+                                345,
+                                592
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        581,
+                                        345,
+                                        592
+                                    ],
+                                    "score": 1.0,
+                                    "content": "national Conference on Learning Representations, 2016.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        }
+                    ],
+                    "index": 34
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        598,
+                        504,
+                        622
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                599,
+                                505,
+                                612
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        599,
+                                        505,
+                                        612
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Qiang Liu, Lihong Li, Ziyang Tang, and Dengyong Zhou. Breaking the curse of horizon: Infinite-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                610,
+                                492,
+                                623
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        610,
+                                        492,
+                                        623
+                                    ],
+                                    "score": 1.0,
+                                    "content": "horizon off-policy estimation. In Advances in Neural Information Processing Systems, 2018a.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        }
+                    ],
+                    "index": 36.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        628,
+                        503,
+                        662
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                628,
+                                505,
+                                641
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        628,
+                                        505,
+                                        641
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Yao Liu, Omer Gottesman, Aniruddh Raghu, Matthieu Komorowski, Aldo A Faisal, Finale Doshi-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                117,
+                                639,
+                                505,
+                                652
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        117,
+                                        639,
+                                        505,
+                                        652
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Velez, and Emma Brunskill. Representation balancing MDPs for off-policy policy evaluation. In",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                650,
+                                360,
+                                663
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        650,
+                                        360,
+                                        663
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Advances in Neural Information Processing Systems, 2018b.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        }
+                    ],
+                    "index": 39
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        668,
+                        503,
+                        692
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                668,
+                                505,
+                                681
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        668,
+                                        505,
+                                        681
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Yao Liu, Adith Swaminathan, Alekh Agarwal, and Emma Brunskill. Off-policy policy gradient with",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                680,
+                                394,
+                                692
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        680,
+                                        394,
+                                        692
+                                    ],
+                                    "score": 1.0,
+                                    "content": "state distribution correction. arXiv preprint arXiv:1904.08473, 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        }
+                    ],
+                    "index": 41.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        699,
+                        504,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                698,
+                                505,
+                                711
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        698,
+                                        505,
+                                        711
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Tatsuya Matsushima, Hiroki Furuta, Yutaka Matsuo, Ofir Nachum, and Shixiang Gu. Deployment-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                709,
+                                505,
+                                722
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        709,
+                                        429,
+                                        722
+                                    ],
+                                    "score": 1.0,
+                                    "content": "efficient reinforcement learning via model-based offline optimization.",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        439,
+                                        711,
+                                        505,
+                                        722
+                                    ],
+                                    "score": 1.0,
+                                    "content": "arXiv preprint",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                720,
+                                219,
+                                732
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        720,
+                                        219,
+                                        732
+                                    ],
+                                    "score": 1.0,
+                                    "content": "arXiv:2006.03647, 2020.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        }
+                    ],
+                    "index": 44
+                }
+            ],
+            "page_idx": 9,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        108,
+                        27,
+                        292,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                26,
+                                293,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        26,
+                                        293,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Published as a conference paper at ICLR 2021",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        300,
+                        751,
+                        311,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                299,
+                                750,
+                                313,
+                                764
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        299,
+                                        750,
+                                        313,
+                                        764
+                                    ],
+                                    "score": 1.0,
+                                    "content": "10",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        82,
+                        505,
+                        105
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                81,
+                                505,
+                                95
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        81,
+                                        505,
+                                        95
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Carles Gelada and Marc G Bellemare. Off-policy deep reinforcement learning by bootstrapping the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                93,
+                                389,
+                                106
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        93,
+                                        389,
+                                        106
+                                    ],
+                                    "score": 1.0,
+                                    "content": "covariate shift. In AAAI Conference on Artificial Intelligence, 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        }
+                    ],
+                    "index": 0.5,
+                    "bbox_fs": [
+                        106,
+                        81,
+                        505,
+                        106
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        111,
+                        504,
+                        135
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                111,
+                                504,
+                                124
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        111,
+                                        504,
+                                        124
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Arthur Gretton, Karsten M Borgwardt, Malte J Rasch, Bernhard Scholkopf, and Alexander Smola.¨",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                123,
+                                484,
+                                136
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        123,
+                                        484,
+                                        136
+                                    ],
+                                    "score": 1.0,
+                                    "content": "A kernel two-sample test. Journal of Machine Learning Research, 13(Mar):723–773, 2012.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        }
+                    ],
+                    "index": 2.5,
+                    "bbox_fs": [
+                        106,
+                        111,
+                        504,
+                        136
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        141,
+                        503,
+                        175
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                140,
+                                505,
+                                155
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        140,
+                                        505,
+                                        155
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Tuomas Haarnoja, Aurick Zhou, Pieter Abbeel, and Sergey Levine. Soft actor-critic: Off-policy",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                153,
+                                505,
+                                165
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        153,
+                                        505,
+                                        165
+                                    ],
+                                    "score": 1.0,
+                                    "content": "maximum entropy deep reinforcement learning with a stochastic actor. In International Confer-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                164,
+                                253,
+                                176
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        164,
+                                        253,
+                                        176
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ence on Machine Learning, 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        }
+                    ],
+                    "index": 5,
+                    "bbox_fs": [
+                        106,
+                        140,
+                        505,
+                        176
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        181,
+                        504,
+                        205
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                181,
+                                504,
+                                195
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        181,
+                                        504,
+                                        195
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Assaf Hallak and Shie Mannor. Consistent on-line off-policy evaluation. In International Confer-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                193,
+                                253,
+                                205
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        193,
+                                        253,
+                                        205
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ence on Machine Learning, 2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        }
+                    ],
+                    "index": 7.5,
+                    "bbox_fs": [
+                        106,
+                        181,
+                        504,
+                        205
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        211,
+                        504,
+                        234
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                211,
+                                505,
+                                224
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        211,
+                                        505,
+                                        224
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Wassily Hoeffding. Probability inequalities for sums of bounded random variables. Journal of the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                222,
+                                340,
+                                234
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        222,
+                                        340,
+                                        234
+                                    ],
+                                    "score": 1.0,
+                                    "content": "American Statistical Association, 58(301):13–30, 1963.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        }
+                    ],
+                    "index": 9.5,
+                    "bbox_fs": [
+                        106,
+                        211,
+                        505,
+                        234
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        240,
+                        503,
+                        264
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                240,
+                                505,
+                                254
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        240,
+                                        505,
+                                        254
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Michael Janner, Justin Fu, Marvin Zhang, and Sergey Levine. When to trust your model: Model-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                252,
+                                475,
+                                264
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        252,
+                                        475,
+                                        264
+                                    ],
+                                    "score": 1.0,
+                                    "content": "based policy optimization. In Advances in Neural Information Processing Systems, 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        }
+                    ],
+                    "index": 11.5,
+                    "bbox_fs": [
+                        105,
+                        240,
+                        505,
+                        264
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        270,
+                        504,
+                        294
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                270,
+                                505,
+                                283
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        270,
+                                        505,
+                                        283
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Nan Jiang and Lihong Li. Doubly robust off-policy value evaluation for reinforcement learning. In",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                281,
+                                336,
+                                294
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        281,
+                                        336,
+                                        294
+                                    ],
+                                    "score": 1.0,
+                                    "content": "International Conference on Machine Learning, 2016.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        }
+                    ],
+                    "index": 13.5,
+                    "bbox_fs": [
+                        105,
+                        270,
+                        505,
+                        294
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        299,
+                        503,
+                        323
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                299,
+                                504,
+                                313
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        299,
+                                        504,
+                                        313
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Fredrik D Johansson, Nathan Kallus, Uri Shalit, and David Sontag. Learning weighted representa-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                311,
+                                437,
+                                324
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        311,
+                                        437,
+                                        324
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tions for generalization across designs. arXiv preprint arXiv:1802.08598, 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        }
+                    ],
+                    "index": 15.5,
+                    "bbox_fs": [
+                        105,
+                        299,
+                        504,
+                        324
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        329,
+                        504,
+                        353
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                329,
+                                505,
+                                342
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        329,
+                                        505,
+                                        342
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Michael Kearns and Satinder Singh. Near-optimal reinforcement learning in polynomial time. Ma-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                340,
+                                277,
+                                353
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        340,
+                                        277,
+                                        353
+                                    ],
+                                    "score": 1.0,
+                                    "content": "chine learning, 49(2-3):209–232, 2002.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        }
+                    ],
+                    "index": 17.5,
+                    "bbox_fs": [
+                        106,
+                        329,
+                        505,
+                        353
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        358,
+                        504,
+                        393
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                359,
+                                505,
+                                371
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        359,
+                                        505,
+                                        371
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Rahul Kidambi, Aravind Rajeswaran, Praneeth Netrapalli, and Thorsten Joachims. MOReL: Model-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                370,
+                                505,
+                                383
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        370,
+                                        505,
+                                        383
+                                    ],
+                                    "score": 1.0,
+                                    "content": "based offline reinforcement learning. In Advances in Neural Information Processing Systems,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                380,
+                                142,
+                                393
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        380,
+                                        142,
+                                        393
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2020.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        }
+                    ],
+                    "index": 20,
+                    "bbox_fs": [
+                        105,
+                        359,
+                        505,
+                        393
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        399,
+                        504,
+                        422
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                398,
+                                506,
+                                414
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        398,
+                                        506,
+                                        414
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Diederik P Kingma and Jimmy Ba. Adam: A method for stochastic optimization. arXiv preprint",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                411,
+                                214,
+                                422
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        411,
+                                        214,
+                                        422
+                                    ],
+                                    "score": 1.0,
+                                    "content": "arXiv:1412.6980, 2014.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        }
+                    ],
+                    "index": 22.5,
+                    "bbox_fs": [
+                        105,
+                        398,
+                        506,
+                        422
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        429,
+                        504,
+                        463
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                428,
+                                505,
+                                443
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        428,
+                                        505,
+                                        443
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Aviral Kumar, Justin Fu, Matthew Soh, George Tucker, and Sergey Levine. Stabilizing off-policy",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                441,
+                                505,
+                                453
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        441,
+                                        505,
+                                        453
+                                    ],
+                                    "score": 1.0,
+                                    "content": "q-learning via bootstrapping error reduction. In Advances in Neural Information Processing Sys-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                451,
+                                165,
+                                463
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        451,
+                                        165,
+                                        463
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tems, 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        }
+                    ],
+                    "index": 25,
+                    "bbox_fs": [
+                        105,
+                        428,
+                        505,
+                        463
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        469,
+                        504,
+                        493
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                469,
+                                505,
+                                482
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        469,
+                                        505,
+                                        482
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Aviral Kumar, Aurick Zhou, George Tucker, and Sergey Levine. Conservative Q-learning for offline",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                481,
+                                463,
+                                493
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        481,
+                                        463,
+                                        493
+                                    ],
+                                    "score": 1.0,
+                                    "content": "reinforcement learning. In Advances in Neural Information Processing Systems, 2020.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        }
+                    ],
+                    "index": 27.5,
+                    "bbox_fs": [
+                        106,
+                        469,
+                        505,
+                        493
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        498,
+                        504,
+                        523
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                499,
+                                505,
+                                512
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        499,
+                                        505,
+                                        512
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Byung-Jun Lee, Jongmin Lee, Peter Vrancx, Dongho Kim, and Kee-Eung Kim. Batch reinforcement",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                510,
+                                505,
+                                523
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        510,
+                                        505,
+                                        523
+                                    ],
+                                    "score": 1.0,
+                                    "content": "learning with hyperparameter gradients. In International Conference on Machine Learning, 2020.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        }
+                    ],
+                    "index": 29.5,
+                    "bbox_fs": [
+                        106,
+                        499,
+                        505,
+                        523
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        528,
+                        503,
+                        552
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                528,
+                                505,
+                                541
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        528,
+                                        505,
+                                        541
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Sergey Levine, Aviral Kumar, George Tucker, and Justin Fu. Offline reinforcement learning: Tuto-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                540,
+                                474,
+                                553
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        540,
+                                        474,
+                                        553
+                                    ],
+                                    "score": 1.0,
+                                    "content": "rial, review, and perspectives on open problems. arXiv preprint arXiv:2005.01643, 2020.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        }
+                    ],
+                    "index": 31.5,
+                    "bbox_fs": [
+                        106,
+                        528,
+                        505,
+                        553
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        558,
+                        503,
+                        592
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                557,
+                                505,
+                                572
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        557,
+                                        505,
+                                        572
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Timothy P Lillicrap, Jonathan J Hunt, Alexander Pritzel, Nicolas Heess, Tom Erez, Yuval Tassa,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                569,
+                                505,
+                                582
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        569,
+                                        505,
+                                        582
+                                    ],
+                                    "score": 1.0,
+                                    "content": "David Silver, and Daan Wierstra. Continuous control with deep reinforcement learning. In Inter-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                581,
+                                345,
+                                592
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        581,
+                                        345,
+                                        592
+                                    ],
+                                    "score": 1.0,
+                                    "content": "national Conference on Learning Representations, 2016.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        }
+                    ],
+                    "index": 34,
+                    "bbox_fs": [
+                        106,
+                        557,
+                        505,
+                        592
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        598,
+                        504,
+                        622
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                599,
+                                505,
+                                612
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        599,
+                                        505,
+                                        612
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Qiang Liu, Lihong Li, Ziyang Tang, and Dengyong Zhou. Breaking the curse of horizon: Infinite-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                610,
+                                492,
+                                623
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        610,
+                                        492,
+                                        623
+                                    ],
+                                    "score": 1.0,
+                                    "content": "horizon off-policy estimation. In Advances in Neural Information Processing Systems, 2018a.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        }
+                    ],
+                    "index": 36.5,
+                    "bbox_fs": [
+                        106,
+                        599,
+                        505,
+                        623
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        628,
+                        503,
+                        662
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                628,
+                                505,
+                                641
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        628,
+                                        505,
+                                        641
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Yao Liu, Omer Gottesman, Aniruddh Raghu, Matthieu Komorowski, Aldo A Faisal, Finale Doshi-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                117,
+                                639,
+                                505,
+                                652
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        117,
+                                        639,
+                                        505,
+                                        652
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Velez, and Emma Brunskill. Representation balancing MDPs for off-policy policy evaluation. In",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                650,
+                                360,
+                                663
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        650,
+                                        360,
+                                        663
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Advances in Neural Information Processing Systems, 2018b.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        }
+                    ],
+                    "index": 39,
+                    "bbox_fs": [
+                        106,
+                        628,
+                        505,
+                        663
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        668,
+                        503,
+                        692
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                668,
+                                505,
+                                681
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        668,
+                                        505,
+                                        681
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Yao Liu, Adith Swaminathan, Alekh Agarwal, and Emma Brunskill. Off-policy policy gradient with",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                680,
+                                394,
+                                692
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        680,
+                                        394,
+                                        692
+                                    ],
+                                    "score": 1.0,
+                                    "content": "state distribution correction. arXiv preprint arXiv:1904.08473, 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        }
+                    ],
+                    "index": 41.5,
+                    "bbox_fs": [
+                        106,
+                        668,
+                        505,
+                        692
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        699,
+                        504,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                698,
+                                505,
+                                711
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        698,
+                                        505,
+                                        711
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Tatsuya Matsushima, Hiroki Furuta, Yutaka Matsuo, Ofir Nachum, and Shixiang Gu. Deployment-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                709,
+                                505,
+                                722
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        709,
+                                        429,
+                                        722
+                                    ],
+                                    "score": 1.0,
+                                    "content": "efficient reinforcement learning via model-based offline optimization.",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        439,
+                                        711,
+                                        505,
+                                        722
+                                    ],
+                                    "score": 1.0,
+                                    "content": "arXiv preprint",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                720,
+                                219,
+                                732
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        720,
+                                        219,
+                                        732
+                                    ],
+                                    "score": 1.0,
+                                    "content": "arXiv:2006.03647, 2020.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        }
+                    ],
+                    "index": 44,
+                    "bbox_fs": [
+                        105,
+                        698,
+                        505,
+                        732
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        82,
+                        504,
+                        105
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                82,
+                                505,
+                                95
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        82,
+                                        505,
+                                        95
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Colin McDiarmid. On the method of bounded differences. Surveys in combinatorics, 141(1):148–",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                92,
+                                162,
+                                106
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        92,
+                                        162,
+                                        106
+                                    ],
+                                    "score": 1.0,
+                                    "content": "188, 1989.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        }
+                    ],
+                    "index": 0.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        111,
+                        503,
+                        146
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                111,
+                                505,
+                                125
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        111,
+                                        505,
+                                        125
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Volodymyr Mnih, Koray Kavukcuoglu, David Silver, Andrei A Rusu, Joel Veness, Marc G Belle-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                123,
+                                505,
+                                136
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        123,
+                                        505,
+                                        136
+                                    ],
+                                    "score": 1.0,
+                                    "content": "mare, Alex Graves, Martin Riedmiller, Andreas K Fidjeland, Georg Ostrovski, et al. Human-level",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                134,
+                                441,
+                                147
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        134,
+                                        441,
+                                        147
+                                    ],
+                                    "score": 1.0,
+                                    "content": "control through deep reinforcement learning. Nature, 518(7540):529–533, 2015.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        }
+                    ],
+                    "index": 3
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        152,
+                        504,
+                        175
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                153,
+                                504,
+                                164
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        153,
+                                        504,
+                                        164
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Ali Mousavi, Lihong Li, Qiang Liu, and Denny Zhou. Black-box off-policy estimation for infinite-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                163,
+                                505,
+                                176
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        163,
+                                        505,
+                                        176
+                                    ],
+                                    "score": 1.0,
+                                    "content": "horizon reinforcement learning. In International Conference on Learning Representations, 2020.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        }
+                    ],
+                    "index": 5.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        181,
+                        504,
+                        205
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                182,
+                                506,
+                                195
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        182,
+                                        506,
+                                        195
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Alfred Muller. Integral probability metrics and their generating classes of functions. ¨ Advances in",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                192,
+                                290,
+                                205
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        192,
+                                        290,
+                                        205
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Applied Probability, 29(2):429–443, 1997.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        }
+                    ],
+                    "index": 7.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        211,
+                        504,
+                        245
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                210,
+                                505,
+                                225
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        210,
+                                        505,
+                                        225
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Ofir Nachum, Yinlam Chow, Bo Dai, and Lihong Li. DualDICE: Behavior-agnostic estimation",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                221,
+                                506,
+                                236
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        221,
+                                        506,
+                                        236
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of discounted stationary distribution corrections. In Advances in Neural Information Processing",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                233,
+                                182,
+                                246
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        233,
+                                        182,
+                                        246
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Systems, 2019a.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        }
+                    ],
+                    "index": 10
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        251,
+                        504,
+                        275
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                251,
+                                505,
+                                264
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        251,
+                                        505,
+                                        264
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Ofir Nachum, Bo Dai, Ilya Kostrikov, Yinlam Chow, Lihong Li, and Dale Schuurmans. AlgaeDICE:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                263,
+                                454,
+                                276
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        263,
+                                        454,
+                                        276
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Policy gradient from arbitrary experience. arXiv preprint arXiv:1912.02074, 2019b.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        }
+                    ],
+                    "index": 12.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        281,
+                        503,
+                        305
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                281,
+                                505,
+                                294
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        281,
+                                        505,
+                                        294
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Xue Bin Peng, Aviral Kumar, Grace Zhang, and Sergey Levine. Advantage-weighted regression:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                293,
+                                500,
+                                306
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        293,
+                                        500,
+                                        306
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Simple and scalable off-policy reinforcement learning. arXiv preprint arXiv:1910.00177, 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        }
+                    ],
+                    "index": 14.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        311,
+                        504,
+                        334
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                310,
+                                506,
+                                325
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        310,
+                                        506,
+                                        325
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Doina Precup. Eligibility traces for off-policy policy evaluation. Computer Science Department",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                322,
+                                284,
+                                334
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        322,
+                                        284,
+                                        334
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Faculty Publication Series, pp. 80, 2000.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        }
+                    ],
+                    "index": 16.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        340,
+                        504,
+                        363
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                340,
+                                506,
+                                353
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        340,
+                                        506,
+                                        353
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Prajit Ramachandran, Barret Zoph, and Quoc V Le. Swish: a self-gated activation function. arXiv",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                352,
+                                264,
+                                364
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        352,
+                                        264,
+                                        364
+                                    ],
+                                    "score": 1.0,
+                                    "content": "preprint arXiv:1710.05941, 7, 2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        }
+                    ],
+                    "index": 18.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        370,
+                        504,
+                        393
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                369,
+                                505,
+                                384
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        369,
+                                        505,
+                                        384
+                                    ],
+                                    "score": 1.0,
+                                    "content": "John Schulman, Filip Wolski, Prafulla Dhariwal, Alec Radford, and Oleg Klimov. Proximal policy",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                381,
+                                380,
+                                393
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        381,
+                                        380,
+                                        393
+                                    ],
+                                    "score": 1.0,
+                                    "content": "optimization algorithms. arXiv preprint arXiv:1707.06347, 2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        }
+                    ],
+                    "index": 20.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        399,
+                        503,
+                        423
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                399,
+                                505,
+                                413
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        399,
+                                        505,
+                                        413
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Uri Shalit, Fredrik D Johansson, and David Sontag. Estimating individual treatment effect: gener-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                411,
+                                481,
+                                424
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        411,
+                                        481,
+                                        424
+                                    ],
+                                    "score": 1.0,
+                                    "content": "alization bounds and algorithms. In International Conference on Machine Learning, 2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        }
+                    ],
+                    "index": 22.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        429,
+                        504,
+                        463
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                429,
+                                505,
+                                441
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        429,
+                                        505,
+                                        441
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Bharath K Sriperumbudur, Kenji Fukumizu, Arthur Gretton, Bernhard Scholkopf, and Gert RG ¨",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                440,
+                                505,
+                                452
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        440,
+                                        286,
+                                        452
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Lanckriet. On integral probability metrics,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        286,
+                                        441,
+                                        293,
+                                        452
+                                    ],
+                                    "score": 0.85,
+                                    "content": "\\phi",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        294,
+                                        440,
+                                        505,
+                                        452
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-divergences and binary classification. arXiv preprint",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                451,
+                                214,
+                                463
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        451,
+                                        214,
+                                        463
+                                    ],
+                                    "score": 1.0,
+                                    "content": "arXiv:0901.2698, 2009.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        }
+                    ],
+                    "index": 25
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        469,
+                        504,
+                        493
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                469,
+                                505,
+                                482
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        469,
+                                        505,
+                                        482
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Richard S Sutton, Andrew G Barto, et al. Introduction to reinforcement learning, volume 135. MIT",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                481,
+                                214,
+                                493
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        481,
+                                        214,
+                                        493
+                                    ],
+                                    "score": 1.0,
+                                    "content": "press Cambridge, 1998.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        }
+                    ],
+                    "index": 27.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        498,
+                        504,
+                        533
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                498,
+                                505,
+                                513
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        498,
+                                        402,
+                                        513
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Richard S Sutton, Hamid R Maei, and Csaba Szepesvari. A convergent ´",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        402,
+                                        499,
+                                        421,
+                                        511
+                                    ],
+                                    "score": 0.91,
+                                    "content": "o ( n )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        422,
+                                        498,
+                                        505,
+                                        513
+                                    ],
+                                    "score": 1.0,
+                                    "content": "temporal-difference",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                510,
+                                505,
+                                523
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        510,
+                                        505,
+                                        523
+                                    ],
+                                    "score": 1.0,
+                                    "content": "algorithm for off-policy learning with linear function approximation. In Advances in neural in-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                114,
+                                521,
+                                263,
+                                533
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        114,
+                                        521,
+                                        263,
+                                        533
+                                    ],
+                                    "score": 1.0,
+                                    "content": "formation processing systems, 2009.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        }
+                    ],
+                    "index": 30
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        539,
+                        504,
+                        563
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                539,
+                                505,
+                                552
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        539,
+                                        505,
+                                        552
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Phillip Swazinna, Steffen Udluft, and Thomas Runkler. Overcoming model bias for robust offline",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                550,
+                                397,
+                                563
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        550,
+                                        397,
+                                        563
+                                    ],
+                                    "score": 1.0,
+                                    "content": "deep reinforcement learning. arXiv preprint arXiv:2008.05533, 2020.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        }
+                    ],
+                    "index": 32.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        569,
+                        505,
+                        603
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                569,
+                                505,
+                                582
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        569,
+                                        505,
+                                        582
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Ziyang Tang, Yihao Feng, Lihong Li, Dengyong Zhou, and Qiang Liu. Doubly robust bias reduction",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                581,
+                                505,
+                                593
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        581,
+                                        505,
+                                        593
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in infinite horizon off-policy estimation. In International Conference on Learning Representa-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                591,
+                                165,
+                                603
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        591,
+                                        165,
+                                        603
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tions, 2020.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        }
+                    ],
+                    "index": 35
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        609,
+                        503,
+                        633
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                609,
+                                505,
+                                622
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        609,
+                                        505,
+                                        622
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Philip Thomas and Emma Brunskill. Data-efficient off-policy policy evaluation for reinforcement",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                114,
+                                620,
+                                386,
+                                634
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        114,
+                                        620,
+                                        386,
+                                        634
+                                    ],
+                                    "score": 1.0,
+                                    "content": "learning. In International Conference on Machine Learning, 2016.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        }
+                    ],
+                    "index": 37.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        639,
+                        502,
+                        663
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                639,
+                                504,
+                                652
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        639,
+                                        504,
+                                        652
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Emanuel Todorov, Tom Erez, and Yuval Tassa. Mujoco: A physics engine for model-based control.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                650,
+                                395,
+                                662
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        650,
+                                        395,
+                                        662
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In International Conference on Intelligent Robots and Systems, 2012.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        }
+                    ],
+                    "index": 39.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        668,
+                        503,
+                        692
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                666,
+                                506,
+                                684
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        666,
+                                        506,
+                                        684
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Yifan Wu, George Tucker, and Ofir Nachum. Behavior regularized offline reinforcement learning.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                680,
+                                279,
+                                692
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        680,
+                                        279,
+                                        692
+                                    ],
+                                    "score": 1.0,
+                                    "content": "arXiv preprint arXiv:1911.11361, 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        }
+                    ],
+                    "index": 41.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        698,
+                        504,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                697,
+                                506,
+                                712
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        697,
+                                        506,
+                                        712
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Tianhe Yu, Garrett Thomas, Lantao Yu, Stefano Ermon, James Zou, Sergey Levine, Chelsea Finn,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                709,
+                                505,
+                                722
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        709,
+                                        505,
+                                        722
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and Tengyu Ma. MOPO: Model-based offline policy optimization. In Advances in Neural Infor-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                721,
+                                254,
+                                733
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        721,
+                                        254,
+                                        733
+                                    ],
+                                    "score": 1.0,
+                                    "content": "mation Processing Systems, 2020.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        }
+                    ],
+                    "index": 44
+                }
+            ],
+            "page_idx": 10,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        108,
+                        27,
+                        292,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                26,
+                                293,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        26,
+                                        293,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Published as a conference paper at ICLR 2021",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        300,
+                        751,
+                        310,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                299,
+                                750,
+                                312,
+                                765
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        299,
+                                        750,
+                                        312,
+                                        765
+                                    ],
+                                    "score": 1.0,
+                                    "content": "",
+                                    "type": "text",
+                                    "height": 15,
+                                    "width": 13
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        82,
+                        504,
+                        105
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                82,
+                                505,
+                                95
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        82,
+                                        505,
+                                        95
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Colin McDiarmid. On the method of bounded differences. Surveys in combinatorics, 141(1):148–",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                92,
+                                162,
+                                106
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        92,
+                                        162,
+                                        106
+                                    ],
+                                    "score": 1.0,
+                                    "content": "188, 1989.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        }
+                    ],
+                    "index": 0.5,
+                    "bbox_fs": [
+                        106,
+                        82,
+                        505,
+                        106
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        111,
+                        503,
+                        146
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                111,
+                                505,
+                                125
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        111,
+                                        505,
+                                        125
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Volodymyr Mnih, Koray Kavukcuoglu, David Silver, Andrei A Rusu, Joel Veness, Marc G Belle-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                123,
+                                505,
+                                136
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        123,
+                                        505,
+                                        136
+                                    ],
+                                    "score": 1.0,
+                                    "content": "mare, Alex Graves, Martin Riedmiller, Andreas K Fidjeland, Georg Ostrovski, et al. Human-level",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                134,
+                                441,
+                                147
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        134,
+                                        441,
+                                        147
+                                    ],
+                                    "score": 1.0,
+                                    "content": "control through deep reinforcement learning. Nature, 518(7540):529–533, 2015.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        }
+                    ],
+                    "index": 3,
+                    "bbox_fs": [
+                        106,
+                        111,
+                        505,
+                        147
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        152,
+                        504,
+                        175
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                153,
+                                504,
+                                164
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        153,
+                                        504,
+                                        164
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Ali Mousavi, Lihong Li, Qiang Liu, and Denny Zhou. Black-box off-policy estimation for infinite-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                163,
+                                505,
+                                176
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        163,
+                                        505,
+                                        176
+                                    ],
+                                    "score": 1.0,
+                                    "content": "horizon reinforcement learning. In International Conference on Learning Representations, 2020.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        }
+                    ],
+                    "index": 5.5,
+                    "bbox_fs": [
+                        106,
+                        153,
+                        505,
+                        176
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        181,
+                        504,
+                        205
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                182,
+                                506,
+                                195
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        182,
+                                        506,
+                                        195
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Alfred Muller. Integral probability metrics and their generating classes of functions. ¨ Advances in",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                192,
+                                290,
+                                205
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        192,
+                                        290,
+                                        205
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Applied Probability, 29(2):429–443, 1997.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        }
+                    ],
+                    "index": 7.5,
+                    "bbox_fs": [
+                        106,
+                        182,
+                        506,
+                        205
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        211,
+                        504,
+                        245
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                210,
+                                505,
+                                225
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        210,
+                                        505,
+                                        225
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Ofir Nachum, Yinlam Chow, Bo Dai, and Lihong Li. DualDICE: Behavior-agnostic estimation",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                221,
+                                506,
+                                236
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        221,
+                                        506,
+                                        236
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of discounted stationary distribution corrections. In Advances in Neural Information Processing",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                233,
+                                182,
+                                246
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        233,
+                                        182,
+                                        246
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Systems, 2019a.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        }
+                    ],
+                    "index": 10,
+                    "bbox_fs": [
+                        105,
+                        210,
+                        506,
+                        246
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        251,
+                        504,
+                        275
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                251,
+                                505,
+                                264
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        251,
+                                        505,
+                                        264
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Ofir Nachum, Bo Dai, Ilya Kostrikov, Yinlam Chow, Lihong Li, and Dale Schuurmans. AlgaeDICE:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                263,
+                                454,
+                                276
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        263,
+                                        454,
+                                        276
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Policy gradient from arbitrary experience. arXiv preprint arXiv:1912.02074, 2019b.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        }
+                    ],
+                    "index": 12.5,
+                    "bbox_fs": [
+                        106,
+                        251,
+                        505,
+                        276
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        281,
+                        503,
+                        305
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                281,
+                                505,
+                                294
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        281,
+                                        505,
+                                        294
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Xue Bin Peng, Aviral Kumar, Grace Zhang, and Sergey Levine. Advantage-weighted regression:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                293,
+                                500,
+                                306
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        293,
+                                        500,
+                                        306
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Simple and scalable off-policy reinforcement learning. arXiv preprint arXiv:1910.00177, 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        }
+                    ],
+                    "index": 14.5,
+                    "bbox_fs": [
+                        106,
+                        281,
+                        505,
+                        306
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        311,
+                        504,
+                        334
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                310,
+                                506,
+                                325
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        310,
+                                        506,
+                                        325
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Doina Precup. Eligibility traces for off-policy policy evaluation. Computer Science Department",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                322,
+                                284,
+                                334
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        322,
+                                        284,
+                                        334
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Faculty Publication Series, pp. 80, 2000.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        }
+                    ],
+                    "index": 16.5,
+                    "bbox_fs": [
+                        106,
+                        310,
+                        506,
+                        334
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        340,
+                        504,
+                        363
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                340,
+                                506,
+                                353
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        340,
+                                        506,
+                                        353
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Prajit Ramachandran, Barret Zoph, and Quoc V Le. Swish: a self-gated activation function. arXiv",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                352,
+                                264,
+                                364
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        352,
+                                        264,
+                                        364
+                                    ],
+                                    "score": 1.0,
+                                    "content": "preprint arXiv:1710.05941, 7, 2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        }
+                    ],
+                    "index": 18.5,
+                    "bbox_fs": [
+                        106,
+                        340,
+                        506,
+                        364
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        370,
+                        504,
+                        393
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                369,
+                                505,
+                                384
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        369,
+                                        505,
+                                        384
+                                    ],
+                                    "score": 1.0,
+                                    "content": "John Schulman, Filip Wolski, Prafulla Dhariwal, Alec Radford, and Oleg Klimov. Proximal policy",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                381,
+                                380,
+                                393
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        381,
+                                        380,
+                                        393
+                                    ],
+                                    "score": 1.0,
+                                    "content": "optimization algorithms. arXiv preprint arXiv:1707.06347, 2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        }
+                    ],
+                    "index": 20.5,
+                    "bbox_fs": [
+                        105,
+                        369,
+                        505,
+                        393
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        399,
+                        503,
+                        423
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                399,
+                                505,
+                                413
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        399,
+                                        505,
+                                        413
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Uri Shalit, Fredrik D Johansson, and David Sontag. Estimating individual treatment effect: gener-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                411,
+                                481,
+                                424
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        411,
+                                        481,
+                                        424
+                                    ],
+                                    "score": 1.0,
+                                    "content": "alization bounds and algorithms. In International Conference on Machine Learning, 2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        }
+                    ],
+                    "index": 22.5,
+                    "bbox_fs": [
+                        106,
+                        399,
+                        505,
+                        424
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        429,
+                        504,
+                        463
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                429,
+                                505,
+                                441
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        429,
+                                        505,
+                                        441
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Bharath K Sriperumbudur, Kenji Fukumizu, Arthur Gretton, Bernhard Scholkopf, and Gert RG ¨",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                440,
+                                505,
+                                452
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        440,
+                                        286,
+                                        452
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Lanckriet. On integral probability metrics,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        286,
+                                        441,
+                                        293,
+                                        452
+                                    ],
+                                    "score": 0.85,
+                                    "content": "\\phi",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        294,
+                                        440,
+                                        505,
+                                        452
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-divergences and binary classification. arXiv preprint",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                451,
+                                214,
+                                463
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        451,
+                                        214,
+                                        463
+                                    ],
+                                    "score": 1.0,
+                                    "content": "arXiv:0901.2698, 2009.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        }
+                    ],
+                    "index": 25,
+                    "bbox_fs": [
+                        106,
+                        429,
+                        505,
+                        463
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        469,
+                        504,
+                        493
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                469,
+                                505,
+                                482
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        469,
+                                        505,
+                                        482
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Richard S Sutton, Andrew G Barto, et al. Introduction to reinforcement learning, volume 135. MIT",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                481,
+                                214,
+                                493
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        481,
+                                        214,
+                                        493
+                                    ],
+                                    "score": 1.0,
+                                    "content": "press Cambridge, 1998.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        }
+                    ],
+                    "index": 27.5,
+                    "bbox_fs": [
+                        106,
+                        469,
+                        505,
+                        493
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        498,
+                        504,
+                        533
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                498,
+                                505,
+                                513
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        498,
+                                        402,
+                                        513
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Richard S Sutton, Hamid R Maei, and Csaba Szepesvari. A convergent ´",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        402,
+                                        499,
+                                        421,
+                                        511
+                                    ],
+                                    "score": 0.91,
+                                    "content": "o ( n )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        422,
+                                        498,
+                                        505,
+                                        513
+                                    ],
+                                    "score": 1.0,
+                                    "content": "temporal-difference",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                510,
+                                505,
+                                523
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        510,
+                                        505,
+                                        523
+                                    ],
+                                    "score": 1.0,
+                                    "content": "algorithm for off-policy learning with linear function approximation. In Advances in neural in-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                114,
+                                521,
+                                263,
+                                533
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        114,
+                                        521,
+                                        263,
+                                        533
+                                    ],
+                                    "score": 1.0,
+                                    "content": "formation processing systems, 2009.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        }
+                    ],
+                    "index": 30,
+                    "bbox_fs": [
+                        105,
+                        498,
+                        505,
+                        533
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        539,
+                        504,
+                        563
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                539,
+                                505,
+                                552
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        539,
+                                        505,
+                                        552
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Phillip Swazinna, Steffen Udluft, and Thomas Runkler. Overcoming model bias for robust offline",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                550,
+                                397,
+                                563
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        550,
+                                        397,
+                                        563
+                                    ],
+                                    "score": 1.0,
+                                    "content": "deep reinforcement learning. arXiv preprint arXiv:2008.05533, 2020.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        }
+                    ],
+                    "index": 32.5,
+                    "bbox_fs": [
+                        105,
+                        539,
+                        505,
+                        563
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        569,
+                        505,
+                        603
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                569,
+                                505,
+                                582
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        569,
+                                        505,
+                                        582
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Ziyang Tang, Yihao Feng, Lihong Li, Dengyong Zhou, and Qiang Liu. Doubly robust bias reduction",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                581,
+                                505,
+                                593
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        581,
+                                        505,
+                                        593
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in infinite horizon off-policy estimation. In International Conference on Learning Representa-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                591,
+                                165,
+                                603
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        591,
+                                        165,
+                                        603
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tions, 2020.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        }
+                    ],
+                    "index": 35,
+                    "bbox_fs": [
+                        105,
+                        569,
+                        505,
+                        603
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        609,
+                        503,
+                        633
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                609,
+                                505,
+                                622
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        609,
+                                        505,
+                                        622
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Philip Thomas and Emma Brunskill. Data-efficient off-policy policy evaluation for reinforcement",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                114,
+                                620,
+                                386,
+                                634
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        114,
+                                        620,
+                                        386,
+                                        634
+                                    ],
+                                    "score": 1.0,
+                                    "content": "learning. In International Conference on Machine Learning, 2016.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        }
+                    ],
+                    "index": 37.5,
+                    "bbox_fs": [
+                        106,
+                        609,
+                        505,
+                        634
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        639,
+                        502,
+                        663
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                639,
+                                504,
+                                652
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        639,
+                                        504,
+                                        652
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Emanuel Todorov, Tom Erez, and Yuval Tassa. Mujoco: A physics engine for model-based control.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                650,
+                                395,
+                                662
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        650,
+                                        395,
+                                        662
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In International Conference on Intelligent Robots and Systems, 2012.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        }
+                    ],
+                    "index": 39.5,
+                    "bbox_fs": [
+                        106,
+                        639,
+                        504,
+                        662
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        668,
+                        503,
+                        692
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                666,
+                                506,
+                                684
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        666,
+                                        506,
+                                        684
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Yifan Wu, George Tucker, and Ofir Nachum. Behavior regularized offline reinforcement learning.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                680,
+                                279,
+                                692
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        680,
+                                        279,
+                                        692
+                                    ],
+                                    "score": 1.0,
+                                    "content": "arXiv preprint arXiv:1911.11361, 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        }
+                    ],
+                    "index": 41.5,
+                    "bbox_fs": [
+                        105,
+                        666,
+                        506,
+                        692
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        698,
+                        504,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                697,
+                                506,
+                                712
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        697,
+                                        506,
+                                        712
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Tianhe Yu, Garrett Thomas, Lantao Yu, Stefano Ermon, James Zou, Sergey Levine, Chelsea Finn,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                709,
+                                505,
+                                722
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        709,
+                                        505,
+                                        722
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and Tengyu Ma. MOPO: Model-based offline policy optimization. In Advances in Neural Infor-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                721,
+                                254,
+                                733
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        721,
+                                        254,
+                                        733
+                                    ],
+                                    "score": 1.0,
+                                    "content": "mation Processing Systems, 2020.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        }
+                    ],
+                    "index": 44,
+                    "bbox_fs": [
+                        105,
+                        697,
+                        506,
+                        733
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        82,
+                        504,
+                        105
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                81,
+                                505,
+                                96
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        81,
+                                        505,
+                                        96
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Rich Zemel, Yu Wu, Kevin Swersky, Toni Pitassi, and Cynthia Dwork. Learning fair representations.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                93,
+                                346,
+                                106
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        93,
+                                        346,
+                                        106
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In International Conference on Machine Learning, 2013.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        }
+                    ],
+                    "index": 0.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        111,
+                        504,
+                        135
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                112,
+                                505,
+                                125
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        112,
+                                        505,
+                                        125
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Ruiyi Zhang, Bo Dai, Lihong Li, and Dale Schuurmans. GenDICE: Generalized offline estimation",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                123,
+                                459,
+                                136
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        123,
+                                        459,
+                                        136
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of stationary values. In International Conference on Learning Representations, 2020.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        }
+                    ],
+                    "index": 2.5
+                }
+            ],
+            "page_idx": 11,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        107,
+                        27,
+                        293,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                26,
+                                293,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        26,
+                                        293,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Published as a conference paper at ICLR 2021",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        300,
+                        751,
+                        311,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                299,
+                                750,
+                                313,
+                                764
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        299,
+                                        750,
+                                        313,
+                                        764
+                                    ],
+                                    "score": 1.0,
+                                    "content": "12",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        82,
+                        504,
+                        105
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                81,
+                                505,
+                                96
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        81,
+                                        505,
+                                        96
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Rich Zemel, Yu Wu, Kevin Swersky, Toni Pitassi, and Cynthia Dwork. Learning fair representations.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                93,
+                                346,
+                                106
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        93,
+                                        346,
+                                        106
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In International Conference on Machine Learning, 2013.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        }
+                    ],
+                    "index": 0.5,
+                    "bbox_fs": [
+                        105,
+                        81,
+                        505,
+                        106
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        111,
+                        504,
+                        135
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                112,
+                                505,
+                                125
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        112,
+                                        505,
+                                        125
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Ruiyi Zhang, Bo Dai, Lihong Li, and Dale Schuurmans. GenDICE: Generalized offline estimation",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                123,
+                                459,
+                                136
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        123,
+                                        459,
+                                        136
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of stationary values. In International Conference on Learning Representations, 2020.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        }
+                    ],
+                    "index": 2.5,
+                    "bbox_fs": [
+                        106,
+                        112,
+                        505,
+                        136
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        81,
+                        170,
+                        94
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                79,
+                                171,
+                                97
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        79,
+                                        171,
+                                        97
+                                    ],
+                                    "score": 1.0,
+                                    "content": "A PROOFS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        }
+                    ],
+                    "index": 0
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        104,
+                        504,
+                        132
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                104,
+                                505,
+                                120
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        104,
+                                        226,
+                                        120
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Lemma 4.1. Given an MDP",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        226,
+                                        107,
+                                        238,
+                                        117
+                                    ],
+                                    "score": 0.38,
+                                    "content": "M",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        238,
+                                        104,
+                                        305,
+                                        120
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and its estimate",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        306,
+                                        104,
+                                        318,
+                                        117
+                                    ],
+                                    "score": 0.84,
+                                    "content": "\\widehat { M }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        318,
+                                        104,
+                                        479,
+                                        120
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with a bijective representation function",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        479,
+                                        107,
+                                        487,
+                                        118
+                                    ],
+                                    "score": 0.81,
+                                    "content": "\\phi ,",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        487,
+                                        104,
+                                        505,
+                                        120
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", i.e.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                117,
+                                460,
+                                134
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        117,
+                                        213,
+                                        132
+                                    ],
+                                    "score": 0.93,
+                                    "content": "( \\widehat { T } , \\widehat { R } ) = \\langle \\widehat { T } _ { z } \\circ \\phi , \\widehat { R } _ { z } \\circ \\phi \\rangle",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        213,
+                                        117,
+                                        371,
+                                        134
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", the policy evaluation error of a policy",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        371,
+                                        122,
+                                        378,
+                                        129
+                                    ],
+                                    "score": 0.64,
+                                    "content": "\\pi",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        379,
+                                        117,
+                                        460,
+                                        134
+                                    ],
+                                    "score": 1.0,
+                                    "content": "can be bounded by:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        }
+                    ],
+                    "index": 1.5
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        222,
+                        137,
+                        388,
+                        159
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                222,
+                                137,
+                                388,
+                                159
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        222,
+                                        137,
+                                        388,
+                                        159
+                                    ],
+                                    "score": 0.94,
+                                    "content": "\\begin{array} { r } { \\left| R ^ { \\pi } - \\widehat { R } ^ { \\pi } \\right| \\leq \\mathbb { E } _ { ( s , a ) \\sim d ^ { \\pi } } \\left[ \\mathcal { E } _ { \\phi , \\widehat { R } _ { z } , \\widehat { T } _ { z } } ( s , a ) \\right] } \\end{array}",
+                                    "type": "interline_equation",
+                                    "image_path": "33d91a00582c4e53b97278296f355c5673f244ded589d301c9aa46c446a8f0ee.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 3,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                222,
+                                137,
+                                388,
+                                159
+                            ],
+                            "spans": [],
+                            "index": 3
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        171,
+                        506,
+                        207
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                102,
+                                168,
+                                508,
+                                190
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        102,
+                                        168,
+                                        250,
+                                        190
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Proof. We define a value function",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        251,
+                                        172,
+                                        400,
+                                        186
+                                    ],
+                                    "score": 0.92,
+                                    "content": "V ^ { \\pi } ( s ) = \\mathbb { E } _ { M , \\pi } [ \\sum _ { t = 0 } ^ { \\infty } \\gamma ^ { t } r _ { t } | s _ { 0 } = s ]",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        400,
+                                        168,
+                                        508,
+                                        190
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", the expected discounted",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                184,
+                                505,
+                                196
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        184,
+                                        225,
+                                        196
+                                    ],
+                                    "score": 1.0,
+                                    "content": "return starting from the state",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        225,
+                                        186,
+                                        231,
+                                        194
+                                    ],
+                                    "score": 0.74,
+                                    "content": "s",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        231,
+                                        184,
+                                        478,
+                                        196
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", for the following proof. Due to its definition, we can write",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        478,
+                                        184,
+                                        505,
+                                        195
+                                    ],
+                                    "score": 0.86,
+                                    "content": "R ^ { \\pi } =",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                107,
+                                195,
+                                379,
+                                207
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        195,
+                                        194,
+                                        207
+                                    ],
+                                    "score": 0.94,
+                                    "content": "( 1 - \\gamma ) \\mathbb { E } _ { s \\sim d _ { 0 } } \\overline { { [ V ^ { \\pi } ( s ) ] } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        195,
+                                        195,
+                                        379,
+                                        207
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". The following recursive equation also holds:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        }
+                    ],
+                    "index": 5
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        203,
+                        212,
+                        407,
+                        228
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                203,
+                                212,
+                                407,
+                                228
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        203,
+                                        212,
+                                        407,
+                                        228
+                                    ],
+                                    "score": 0.88,
+                                    "content": "V ^ { \\pi } ( s ) = \\mathbb { E } _ { a \\sim \\pi ( s ) } \\left[ r ( s , a ) + \\gamma \\mathbb { E } _ { s ^ { \\prime } \\sim T ( s , a ) } [ V ( s ^ { \\prime } ) ] \\right] ,",
+                                    "type": "interline_equation",
+                                    "image_path": "c1ea46570341074db9f0ff962d493395189cbc0ddf5ce0bd5a734eb9b1fa2818.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 7,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                203,
+                                212,
+                                407,
+                                228
+                            ],
+                            "spans": [],
+                            "index": 7
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        235,
+                        504,
+                        264
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                103,
+                                232,
+                                507,
+                                252
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        103,
+                                        232,
+                                        303,
+                                        252
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and similarly with an approximate value function",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        303,
+                                        234,
+                                        445,
+                                        251
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\widehat { V } ^ { \\pi } ( s ) = \\mathbb { E } _ { \\widehat { M } , \\pi } [ \\sum _ { t = 0 } ^ { \\infty } \\gamma ^ { t } r _ { t } | s _ { 0 } = s ]",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        445,
+                                        232,
+                                        507,
+                                        252
+                                    ],
+                                    "score": 1.0,
+                                    "content": "given an MDP",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                248,
+                                186,
+                                266
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        248,
+                                        142,
+                                        266
+                                    ],
+                                    "score": 1.0,
+                                    "content": "estimate",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        142,
+                                        250,
+                                        154,
+                                        262
+                                    ],
+                                    "score": 0.86,
+                                    "content": "\\widehat { M }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        155,
+                                        248,
+                                        186,
+                                        266
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Then,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        }
+                    ],
+                    "index": 8.5
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        111,
+                        269,
+                        502,
+                        410
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                111,
+                                269,
+                                502,
+                                410
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        111,
+                                        269,
+                                        502,
+                                        410
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\begin{array} { r l } & { \\displaystyle \\frac { 1 } { 1 - \\gamma } \\Big | R ^ { \\pi } - \\widehat { R } ^ { \\pi } \\Big | = \\left| \\mathbb { E } _ { d _ { 0 } } \\left[ V ^ { \\pi } ( s _ { 0 } ) - \\widehat { V } ^ { \\pi } ( s _ { 0 } ) \\right] \\right| } \\\\ & { = \\left| \\mathbb { E } _ { d _ { 0 } , \\pi } \\left[ r _ { 0 } - \\widehat { r } _ { 0 } \\right] + \\gamma \\mathbb { E } _ { d _ { 0 } } \\left[ \\int \\left( V ^ { \\pi } ( s _ { 1 } ) T ( s _ { 1 } | s _ { 0 } , a _ { 0 } ) - \\widehat { V } ^ { \\pi } ( s _ { 1 } ) \\widehat { T } ( s _ { 1 } | s _ { 0 } , a _ { 0 } ) \\right) \\pi ( a _ { 0 } | s _ { 0 } ) d s _ { 1 } d a _ { 0 } \\right] \\right| } \\\\ & { = \\left| \\mathbb { E } _ { d _ { 0 } , \\pi } \\left[ r _ { 0 } - \\widehat { r } _ { 0 } \\right] + \\gamma \\mathbb { E } _ { d _ { 0 } } \\left[ \\int \\left( V ^ { \\pi } ( s _ { 1 } ) T ( s _ { 1 } | s _ { 0 } , a _ { 0 } ) - \\widehat { V } ^ { \\pi } ( s _ { 1 } ) T ( s _ { 1 } | s _ { 0 } , a _ { 0 } ) \\right. \\right. } \\\\ & { \\qquad \\left. \\left. + \\widehat { V } ^ { \\pi } ( s _ { 1 } ) T ( s _ { 1 } | s _ { 0 } , a _ { 0 } ) - \\widehat { V } ^ { \\pi } ( s _ { 1 } ) \\widehat { T } ( s _ { 1 } | s _ { 0 } , a _ { 0 } ) \\right) \\pi ( a _ { 0 } | s _ { 0 } ) d s _ { 1 } d a _ { 0 } \\right] \\right| } \\\\ & { = \\left| \\mathbb { E } _ { d _ { 0 } , \\pi } \\left[ r _ { 0 } - \\widehat { r } _ { 0 } \\right] + \\gamma \\mathbb { E } _ { d _ { 0 } } \\left[ \\int \\left( V ^ { \\pi } ( s _ { 1 } ) - \\widehat { V } ^ { \\pi } ( s _ { 1 } ) \\right) T ( s _ { 1 } | s _ { 0 } , a _ { 0 } ) \\pi ( a _ { 0 } | s _ { 0 } ) d s _ { 1 } d a _ { 0 } \\right] \\right| } \\end{array}",
+                                    "type": "interline_equation",
+                                    "image_path": "b2eca14fd51e0827850edb012e3393c12214cd4f5f440e18f3f30a62554ad408.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 11,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                111,
+                                269,
+                                502,
+                                316.0
+                            ],
+                            "spans": [],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                111,
+                                316.0,
+                                502,
+                                363.0
+                            ],
+                            "spans": [],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                111,
+                                363.0,
+                                502,
+                                410.0
+                            ],
+                            "spans": [],
+                            "index": 12
+                        }
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        111,
+                        423,
+                        480,
+                        583
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                111,
+                                423,
+                                480,
+                                583
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        111,
+                                        423,
+                                        480,
+                                        583
+                                    ],
+                                    "score": 0.95,
+                                    "content": "\\begin{array} { r l } & { \\qquad + \\gamma \\mathbb { E } _ { d _ { 0 } } \\bigg [ \\int \\hat { V } ^ { x } ( s _ { 1 } ) \\bigg ( T ( s _ { 1 } | s _ { 0 } , a _ { 0 } ) - \\hat { T } ( s _ { 1 } | s _ { 0 } , a _ { 0 } ) \\bigg ) \\pi ( a _ { 0 } | s _ { 0 } ) d s _ { 1 } d a _ { 0 } \\bigg ] \\bigg | } \\\\ & { = \\left| \\mathbb { E } _ { ( s , a ) \\times d ^ { c } } \\left[ \\int \\bigg ( R ( r | s , a ) - \\hat { R } ( r | s , a ) \\bigg ) d r + \\gamma \\int \\hat { V } ^ { x } ( s ^ { \\prime } ) \\bigg ( T ( s ^ { \\prime } | s , a ) - \\hat { T } ( s ^ { \\prime } | s , a ) \\bigg ) d s ^ { \\prime } \\right] \\right| } \\\\ & { \\leq \\mathbb { E } _ { ( s , a ) \\sim d ^ { c } } \\left[ \\int \\bigg | R ( r | s , a ) - \\hat { R } ( r | s , a ) \\bigg | d r + \\gamma \\int \\hat { V } ^ { \\tau } ( s ^ { \\prime } ) \\bigg | T ( s ^ { \\prime } | s , a ) - \\hat { T } ( s ^ { \\prime } | s , a ) \\bigg | d s ^ { \\prime } \\right] } \\\\ & { \\leq \\mathbb { E } _ { ( s , a ) \\sim d ^ { c } } \\left[ \\int \\bigg | R ( r | s , a ) - \\hat { R } ( r | s , a ) \\bigg | d r + \\gamma \\int \\frac { r _ { \\operatorname* { m a x } } } { 1 - \\gamma } \\bigg | T ( s ^ { \\prime } | s , a ) - \\hat { T } ( s ^ { \\prime } | s , a ) \\bigg | d s ^ { \\prime } \\right] } \\\\ & { = 2 \\mathbb { E } _ { ( s , a ) \\sim d ^ { c } } \\left[ D _ { T V } \\big ( R ( r | s , a ) \\big | \\hat { R } _ { s } ( r | \\phi ( s , a ) ) \\big ) \\right] } \\\\ & { \\qquad + \\frac { 2 \\gamma r _ { \\operatorname* { m a x } } } { 1 - \\gamma } \\mathbb { E } _ { ( s , a ) \\sim d ^ { c } } \\left[ D _ { T V } \\big ( T ( s ^ { \\prime } | s , a ) \\big | \\hat { T } _ { 2 } ( s ^ { \\prime } | \\phi ( s , a ) ) \\big ) \\right] } \\end{array}",
+                                    "type": "interline_equation",
+                                    "image_path": "4f76507e4d0bde5af4efb2747350527ebfb713650f52d5c558cf11544b51ad73.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 14,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                111,
+                                423,
+                                480,
+                                476.3333333333333
+                            ],
+                            "spans": [],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                111,
+                                476.3333333333333,
+                                480,
+                                529.6666666666666
+                            ],
+                            "spans": [],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                111,
+                                529.6666666666666,
+                                480,
+                                583.0
+                            ],
+                            "spans": [],
+                            "index": 15
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        604,
+                        505,
+                        646
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                603,
+                                505,
+                                620
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        603,
+                                        235,
+                                        620
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Theorem 4.2. Given an MDP",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        235,
+                                        606,
+                                        247,
+                                        616
+                                    ],
+                                    "score": 0.5,
+                                    "content": "M",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        247,
+                                        603,
+                                        317,
+                                        620
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and its estimate",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        317,
+                                        604,
+                                        329,
+                                        617
+                                    ],
+                                    "score": 0.85,
+                                    "content": "\\widehat { M }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        330,
+                                        603,
+                                        494,
+                                        620
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with a bijective representation function",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        495,
+                                        609,
+                                        501,
+                                        618
+                                    ],
+                                    "score": 0.82,
+                                    "content": "\\phi",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        501,
+                                        603,
+                                        505,
+                                        620
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                616,
+                                506,
+                                633
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        616,
+                                        124,
+                                        633
+                                    ],
+                                    "score": 1.0,
+                                    "content": "i.e.",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        124,
+                                        617,
+                                        235,
+                                        631
+                                    ],
+                                    "score": 0.92,
+                                    "content": "( \\widehat { T } , \\widehat { R } ) = \\langle \\widehat { T } _ { z } \\circ \\phi , \\widehat { R } _ { z } \\circ \\phi \\rangle",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        235,
+                                        616,
+                                        384,
+                                        633
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", assume that there exists a constant",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        384,
+                                        619,
+                                        419,
+                                        632
+                                    ],
+                                    "score": 0.92,
+                                    "content": "B _ { \\phi } > 0",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        419,
+                                        616,
+                                        506,
+                                        633
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and a function class",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                107,
+                                625,
+                                427,
+                                651
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        632,
+                                        182,
+                                        644
+                                    ],
+                                    "score": 0.91,
+                                    "content": "{ \\mathcal { G } } \\triangleq \\{ g : { \\mathcal { Z } } \\to \\mathbb { R } \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        182,
+                                        625,
+                                        222,
+                                        651
+                                    ],
+                                    "score": 1.0,
+                                    "content": "b such that",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        222,
+                                        631,
+                                        325,
+                                        647
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\begin{array} { r } { \\frac { 1 } { B _ { \\phi } } \\mathcal { E } _ { \\phi , \\widehat { R } _ { z } , \\widehat { T } _ { z } } \\left( \\phi ^ { - 1 } ( \\cdot ) \\right) \\in \\mathcal { G } . } \\end{array}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        326,
+                                        625,
+                                        412,
+                                        651
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Then, for any policy",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        412,
+                                        635,
+                                        419,
+                                        642
+                                    ],
+                                    "score": 0.7,
+                                    "content": "\\pi",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        419,
+                                        625,
+                                        427,
+                                        651
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        }
+                    ],
+                    "index": 17
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        178,
+                        653,
+                        432,
+                        675
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                178,
+                                653,
+                                432,
+                                675
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        178,
+                                        653,
+                                        432,
+                                        675
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\left| R ^ { \\pi } - \\widehat { R } ^ { \\pi } \\right| \\leq \\mathbb { E } _ { ( s , a ) \\sim d ^ { \\mathcal { D } } } \\Big [ \\mathcal { E } _ { \\phi , \\widehat { R } _ { z } , \\widehat { T } _ { z } } ( s , a ) \\Big ] + B _ { \\phi } \\mathrm { I P M } _ { \\mathcal { G } } ( d _ { \\phi } ^ { \\pi } , d _ { \\phi } ^ { \\mathcal { D } } )",
+                                    "type": "interline_equation",
+                                    "image_path": "20f445a1a1724a9bf59376c67dcf590916f33dfe41f2db895ef78f051bceb746.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 19,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                178,
+                                653,
+                                432,
+                                675
+                            ],
+                            "spans": [],
+                            "index": 19
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        687,
+                        281,
+                        700
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                686,
+                                281,
+                                701
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        686,
+                                        281,
+                                        701
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Proof. From Lemma 4.1, we directly have:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        }
+                    ],
+                    "index": 20
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        124,
+                        705,
+                        486,
+                        731
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                124,
+                                705,
+                                486,
+                                731
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        124,
+                                        705,
+                                        486,
+                                        731
+                                    ],
+                                    "score": 0.95,
+                                    "content": "\\Big | R ^ { \\pi } - \\widehat { R } ^ { \\pi } \\Big | \\leq \\mathbb { E } _ { ( s , a ) \\sim d ^ { D } } \\Big [ \\mathcal { E } _ { \\phi , \\widehat { R } _ { z } , \\widehat { T } _ { z } } ( s , a ) \\Big ] + \\int \\mathcal { E } _ { \\phi , \\widehat { R } _ { z } , \\widehat { T } _ { z } } ( s , a ) \\left( d ^ { \\pi } ( s , a ) - d ^ { D } ( s , a ) \\right) d s d a .",
+                                    "type": "interline_equation",
+                                    "image_path": "c3454a9ebef3285947433d845c8fbed141601c968a8c8c8c532f53ffce967a75.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 21,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                124,
+                                705,
+                                486,
+                                731
+                            ],
+                            "spans": [],
+                            "index": 21
+                        }
+                    ]
+                }
+            ],
+            "page_idx": 12,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        107,
+                        26,
+                        293,
+                        38
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                25,
+                                293,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        25,
+                                        293,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Published as a conference paper at ICLR 2021",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        300,
+                        751,
+                        311,
+                        761
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                299,
+                                750,
+                                313,
+                                764
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        299,
+                                        750,
+                                        313,
+                                        764
+                                    ],
+                                    "score": 1.0,
+                                    "content": "13",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        494,
+                        586,
+                        505,
+                        597
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                496,
+                                587,
+                                505,
+                                597
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        496,
+                                        587,
+                                        505,
+                                        597
+                                    ],
+                                    "score": 0.999,
+                                    "content": "□",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        81,
+                        170,
+                        94
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                79,
+                                171,
+                                97
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        79,
+                                        171,
+                                        97
+                                    ],
+                                    "score": 1.0,
+                                    "content": "A PROOFS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        }
+                    ],
+                    "index": 0
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        104,
+                        504,
+                        132
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                104,
+                                505,
+                                120
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        104,
+                                        226,
+                                        120
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Lemma 4.1. Given an MDP",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        226,
+                                        107,
+                                        238,
+                                        117
+                                    ],
+                                    "score": 0.38,
+                                    "content": "M",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        238,
+                                        104,
+                                        305,
+                                        120
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and its estimate",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        306,
+                                        104,
+                                        318,
+                                        117
+                                    ],
+                                    "score": 0.84,
+                                    "content": "\\widehat { M }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        318,
+                                        104,
+                                        479,
+                                        120
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with a bijective representation function",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        479,
+                                        107,
+                                        487,
+                                        118
+                                    ],
+                                    "score": 0.81,
+                                    "content": "\\phi ,",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        487,
+                                        104,
+                                        505,
+                                        120
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", i.e.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                117,
+                                460,
+                                134
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        117,
+                                        213,
+                                        132
+                                    ],
+                                    "score": 0.93,
+                                    "content": "( \\widehat { T } , \\widehat { R } ) = \\langle \\widehat { T } _ { z } \\circ \\phi , \\widehat { R } _ { z } \\circ \\phi \\rangle",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        213,
+                                        117,
+                                        371,
+                                        134
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", the policy evaluation error of a policy",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        371,
+                                        122,
+                                        378,
+                                        129
+                                    ],
+                                    "score": 0.64,
+                                    "content": "\\pi",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        379,
+                                        117,
+                                        460,
+                                        134
+                                    ],
+                                    "score": 1.0,
+                                    "content": "can be bounded by:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        }
+                    ],
+                    "index": 1.5,
+                    "bbox_fs": [
+                        105,
+                        104,
+                        505,
+                        134
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        222,
+                        137,
+                        388,
+                        159
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                222,
+                                137,
+                                388,
+                                159
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        222,
+                                        137,
+                                        388,
+                                        159
+                                    ],
+                                    "score": 0.94,
+                                    "content": "\\begin{array} { r } { \\left| R ^ { \\pi } - \\widehat { R } ^ { \\pi } \\right| \\leq \\mathbb { E } _ { ( s , a ) \\sim d ^ { \\pi } } \\left[ \\mathcal { E } _ { \\phi , \\widehat { R } _ { z } , \\widehat { T } _ { z } } ( s , a ) \\right] } \\end{array}",
+                                    "type": "interline_equation",
+                                    "image_path": "33d91a00582c4e53b97278296f355c5673f244ded589d301c9aa46c446a8f0ee.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 3,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                222,
+                                137,
+                                388,
+                                159
+                            ],
+                            "spans": [],
+                            "index": 3
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        171,
+                        506,
+                        207
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                102,
+                                168,
+                                508,
+                                190
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        102,
+                                        168,
+                                        250,
+                                        190
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Proof. We define a value function",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        251,
+                                        172,
+                                        400,
+                                        186
+                                    ],
+                                    "score": 0.92,
+                                    "content": "V ^ { \\pi } ( s ) = \\mathbb { E } _ { M , \\pi } [ \\sum _ { t = 0 } ^ { \\infty } \\gamma ^ { t } r _ { t } | s _ { 0 } = s ]",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        400,
+                                        168,
+                                        508,
+                                        190
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", the expected discounted",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                184,
+                                505,
+                                196
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        184,
+                                        225,
+                                        196
+                                    ],
+                                    "score": 1.0,
+                                    "content": "return starting from the state",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        225,
+                                        186,
+                                        231,
+                                        194
+                                    ],
+                                    "score": 0.74,
+                                    "content": "s",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        231,
+                                        184,
+                                        478,
+                                        196
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", for the following proof. Due to its definition, we can write",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        478,
+                                        184,
+                                        505,
+                                        195
+                                    ],
+                                    "score": 0.86,
+                                    "content": "R ^ { \\pi } =",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                107,
+                                195,
+                                379,
+                                207
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        195,
+                                        194,
+                                        207
+                                    ],
+                                    "score": 0.94,
+                                    "content": "( 1 - \\gamma ) \\mathbb { E } _ { s \\sim d _ { 0 } } \\overline { { [ V ^ { \\pi } ( s ) ] } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        195,
+                                        195,
+                                        379,
+                                        207
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". The following recursive equation also holds:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        }
+                    ],
+                    "index": 5,
+                    "bbox_fs": [
+                        102,
+                        168,
+                        508,
+                        207
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        203,
+                        212,
+                        407,
+                        228
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                203,
+                                212,
+                                407,
+                                228
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        203,
+                                        212,
+                                        407,
+                                        228
+                                    ],
+                                    "score": 0.88,
+                                    "content": "V ^ { \\pi } ( s ) = \\mathbb { E } _ { a \\sim \\pi ( s ) } \\left[ r ( s , a ) + \\gamma \\mathbb { E } _ { s ^ { \\prime } \\sim T ( s , a ) } [ V ( s ^ { \\prime } ) ] \\right] ,",
+                                    "type": "interline_equation",
+                                    "image_path": "c1ea46570341074db9f0ff962d493395189cbc0ddf5ce0bd5a734eb9b1fa2818.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 7,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                203,
+                                212,
+                                407,
+                                228
+                            ],
+                            "spans": [],
+                            "index": 7
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        235,
+                        504,
+                        264
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                103,
+                                232,
+                                507,
+                                252
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        103,
+                                        232,
+                                        303,
+                                        252
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and similarly with an approximate value function",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        303,
+                                        234,
+                                        445,
+                                        251
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\widehat { V } ^ { \\pi } ( s ) = \\mathbb { E } _ { \\widehat { M } , \\pi } [ \\sum _ { t = 0 } ^ { \\infty } \\gamma ^ { t } r _ { t } | s _ { 0 } = s ]",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        445,
+                                        232,
+                                        507,
+                                        252
+                                    ],
+                                    "score": 1.0,
+                                    "content": "given an MDP",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                248,
+                                186,
+                                266
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        248,
+                                        142,
+                                        266
+                                    ],
+                                    "score": 1.0,
+                                    "content": "estimate",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        142,
+                                        250,
+                                        154,
+                                        262
+                                    ],
+                                    "score": 0.86,
+                                    "content": "\\widehat { M }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        155,
+                                        248,
+                                        186,
+                                        266
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Then,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        }
+                    ],
+                    "index": 8.5,
+                    "bbox_fs": [
+                        103,
+                        232,
+                        507,
+                        266
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        111,
+                        269,
+                        502,
+                        410
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                111,
+                                269,
+                                502,
+                                410
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        111,
+                                        269,
+                                        502,
+                                        410
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\begin{array} { r l } & { \\displaystyle \\frac { 1 } { 1 - \\gamma } \\Big | R ^ { \\pi } - \\widehat { R } ^ { \\pi } \\Big | = \\left| \\mathbb { E } _ { d _ { 0 } } \\left[ V ^ { \\pi } ( s _ { 0 } ) - \\widehat { V } ^ { \\pi } ( s _ { 0 } ) \\right] \\right| } \\\\ & { = \\left| \\mathbb { E } _ { d _ { 0 } , \\pi } \\left[ r _ { 0 } - \\widehat { r } _ { 0 } \\right] + \\gamma \\mathbb { E } _ { d _ { 0 } } \\left[ \\int \\left( V ^ { \\pi } ( s _ { 1 } ) T ( s _ { 1 } | s _ { 0 } , a _ { 0 } ) - \\widehat { V } ^ { \\pi } ( s _ { 1 } ) \\widehat { T } ( s _ { 1 } | s _ { 0 } , a _ { 0 } ) \\right) \\pi ( a _ { 0 } | s _ { 0 } ) d s _ { 1 } d a _ { 0 } \\right] \\right| } \\\\ & { = \\left| \\mathbb { E } _ { d _ { 0 } , \\pi } \\left[ r _ { 0 } - \\widehat { r } _ { 0 } \\right] + \\gamma \\mathbb { E } _ { d _ { 0 } } \\left[ \\int \\left( V ^ { \\pi } ( s _ { 1 } ) T ( s _ { 1 } | s _ { 0 } , a _ { 0 } ) - \\widehat { V } ^ { \\pi } ( s _ { 1 } ) T ( s _ { 1 } | s _ { 0 } , a _ { 0 } ) \\right. \\right. } \\\\ & { \\qquad \\left. \\left. + \\widehat { V } ^ { \\pi } ( s _ { 1 } ) T ( s _ { 1 } | s _ { 0 } , a _ { 0 } ) - \\widehat { V } ^ { \\pi } ( s _ { 1 } ) \\widehat { T } ( s _ { 1 } | s _ { 0 } , a _ { 0 } ) \\right) \\pi ( a _ { 0 } | s _ { 0 } ) d s _ { 1 } d a _ { 0 } \\right] \\right| } \\\\ & { = \\left| \\mathbb { E } _ { d _ { 0 } , \\pi } \\left[ r _ { 0 } - \\widehat { r } _ { 0 } \\right] + \\gamma \\mathbb { E } _ { d _ { 0 } } \\left[ \\int \\left( V ^ { \\pi } ( s _ { 1 } ) - \\widehat { V } ^ { \\pi } ( s _ { 1 } ) \\right) T ( s _ { 1 } | s _ { 0 } , a _ { 0 } ) \\pi ( a _ { 0 } | s _ { 0 } ) d s _ { 1 } d a _ { 0 } \\right] \\right| } \\end{array}",
+                                    "type": "interline_equation",
+                                    "image_path": "b2eca14fd51e0827850edb012e3393c12214cd4f5f440e18f3f30a62554ad408.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 11,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                111,
+                                269,
+                                502,
+                                316.0
+                            ],
+                            "spans": [],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                111,
+                                316.0,
+                                502,
+                                363.0
+                            ],
+                            "spans": [],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                111,
+                                363.0,
+                                502,
+                                410.0
+                            ],
+                            "spans": [],
+                            "index": 12
+                        }
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        111,
+                        423,
+                        480,
+                        583
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                111,
+                                423,
+                                480,
+                                583
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        111,
+                                        423,
+                                        480,
+                                        583
+                                    ],
+                                    "score": 0.95,
+                                    "content": "\\begin{array} { r l } & { \\qquad + \\gamma \\mathbb { E } _ { d _ { 0 } } \\bigg [ \\int \\hat { V } ^ { x } ( s _ { 1 } ) \\bigg ( T ( s _ { 1 } | s _ { 0 } , a _ { 0 } ) - \\hat { T } ( s _ { 1 } | s _ { 0 } , a _ { 0 } ) \\bigg ) \\pi ( a _ { 0 } | s _ { 0 } ) d s _ { 1 } d a _ { 0 } \\bigg ] \\bigg | } \\\\ & { = \\left| \\mathbb { E } _ { ( s , a ) \\times d ^ { c } } \\left[ \\int \\bigg ( R ( r | s , a ) - \\hat { R } ( r | s , a ) \\bigg ) d r + \\gamma \\int \\hat { V } ^ { x } ( s ^ { \\prime } ) \\bigg ( T ( s ^ { \\prime } | s , a ) - \\hat { T } ( s ^ { \\prime } | s , a ) \\bigg ) d s ^ { \\prime } \\right] \\right| } \\\\ & { \\leq \\mathbb { E } _ { ( s , a ) \\sim d ^ { c } } \\left[ \\int \\bigg | R ( r | s , a ) - \\hat { R } ( r | s , a ) \\bigg | d r + \\gamma \\int \\hat { V } ^ { \\tau } ( s ^ { \\prime } ) \\bigg | T ( s ^ { \\prime } | s , a ) - \\hat { T } ( s ^ { \\prime } | s , a ) \\bigg | d s ^ { \\prime } \\right] } \\\\ & { \\leq \\mathbb { E } _ { ( s , a ) \\sim d ^ { c } } \\left[ \\int \\bigg | R ( r | s , a ) - \\hat { R } ( r | s , a ) \\bigg | d r + \\gamma \\int \\frac { r _ { \\operatorname* { m a x } } } { 1 - \\gamma } \\bigg | T ( s ^ { \\prime } | s , a ) - \\hat { T } ( s ^ { \\prime } | s , a ) \\bigg | d s ^ { \\prime } \\right] } \\\\ & { = 2 \\mathbb { E } _ { ( s , a ) \\sim d ^ { c } } \\left[ D _ { T V } \\big ( R ( r | s , a ) \\big | \\hat { R } _ { s } ( r | \\phi ( s , a ) ) \\big ) \\right] } \\\\ & { \\qquad + \\frac { 2 \\gamma r _ { \\operatorname* { m a x } } } { 1 - \\gamma } \\mathbb { E } _ { ( s , a ) \\sim d ^ { c } } \\left[ D _ { T V } \\big ( T ( s ^ { \\prime } | s , a ) \\big | \\hat { T } _ { 2 } ( s ^ { \\prime } | \\phi ( s , a ) ) \\big ) \\right] } \\end{array}",
+                                    "type": "interline_equation",
+                                    "image_path": "4f76507e4d0bde5af4efb2747350527ebfb713650f52d5c558cf11544b51ad73.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 14,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                111,
+                                423,
+                                480,
+                                476.3333333333333
+                            ],
+                            "spans": [],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                111,
+                                476.3333333333333,
+                                480,
+                                529.6666666666666
+                            ],
+                            "spans": [],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                111,
+                                529.6666666666666,
+                                480,
+                                583.0
+                            ],
+                            "spans": [],
+                            "index": 15
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        604,
+                        505,
+                        646
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                603,
+                                505,
+                                620
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        603,
+                                        235,
+                                        620
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Theorem 4.2. Given an MDP",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        235,
+                                        606,
+                                        247,
+                                        616
+                                    ],
+                                    "score": 0.5,
+                                    "content": "M",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        247,
+                                        603,
+                                        317,
+                                        620
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and its estimate",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        317,
+                                        604,
+                                        329,
+                                        617
+                                    ],
+                                    "score": 0.85,
+                                    "content": "\\widehat { M }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        330,
+                                        603,
+                                        494,
+                                        620
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with a bijective representation function",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        495,
+                                        609,
+                                        501,
+                                        618
+                                    ],
+                                    "score": 0.82,
+                                    "content": "\\phi",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        501,
+                                        603,
+                                        505,
+                                        620
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                616,
+                                506,
+                                633
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        616,
+                                        124,
+                                        633
+                                    ],
+                                    "score": 1.0,
+                                    "content": "i.e.",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        124,
+                                        617,
+                                        235,
+                                        631
+                                    ],
+                                    "score": 0.92,
+                                    "content": "( \\widehat { T } , \\widehat { R } ) = \\langle \\widehat { T } _ { z } \\circ \\phi , \\widehat { R } _ { z } \\circ \\phi \\rangle",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        235,
+                                        616,
+                                        384,
+                                        633
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", assume that there exists a constant",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        384,
+                                        619,
+                                        419,
+                                        632
+                                    ],
+                                    "score": 0.92,
+                                    "content": "B _ { \\phi } > 0",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        419,
+                                        616,
+                                        506,
+                                        633
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and a function class",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                107,
+                                625,
+                                427,
+                                651
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        632,
+                                        182,
+                                        644
+                                    ],
+                                    "score": 0.91,
+                                    "content": "{ \\mathcal { G } } \\triangleq \\{ g : { \\mathcal { Z } } \\to \\mathbb { R } \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        182,
+                                        625,
+                                        222,
+                                        651
+                                    ],
+                                    "score": 1.0,
+                                    "content": "b such that",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        222,
+                                        631,
+                                        325,
+                                        647
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\begin{array} { r } { \\frac { 1 } { B _ { \\phi } } \\mathcal { E } _ { \\phi , \\widehat { R } _ { z } , \\widehat { T } _ { z } } \\left( \\phi ^ { - 1 } ( \\cdot ) \\right) \\in \\mathcal { G } . } \\end{array}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        326,
+                                        625,
+                                        412,
+                                        651
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Then, for any policy",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        412,
+                                        635,
+                                        419,
+                                        642
+                                    ],
+                                    "score": 0.7,
+                                    "content": "\\pi",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        419,
+                                        625,
+                                        427,
+                                        651
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        }
+                    ],
+                    "index": 17,
+                    "bbox_fs": [
+                        105,
+                        603,
+                        506,
+                        651
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        178,
+                        653,
+                        432,
+                        675
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                178,
+                                653,
+                                432,
+                                675
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        178,
+                                        653,
+                                        432,
+                                        675
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\left| R ^ { \\pi } - \\widehat { R } ^ { \\pi } \\right| \\leq \\mathbb { E } _ { ( s , a ) \\sim d ^ { \\mathcal { D } } } \\Big [ \\mathcal { E } _ { \\phi , \\widehat { R } _ { z } , \\widehat { T } _ { z } } ( s , a ) \\Big ] + B _ { \\phi } \\mathrm { I P M } _ { \\mathcal { G } } ( d _ { \\phi } ^ { \\pi } , d _ { \\phi } ^ { \\mathcal { D } } )",
+                                    "type": "interline_equation",
+                                    "image_path": "20f445a1a1724a9bf59376c67dcf590916f33dfe41f2db895ef78f051bceb746.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 19,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                178,
+                                653,
+                                432,
+                                675
+                            ],
+                            "spans": [],
+                            "index": 19
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        687,
+                        281,
+                        700
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                686,
+                                281,
+                                701
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        686,
+                                        281,
+                                        701
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Proof. From Lemma 4.1, we directly have:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        }
+                    ],
+                    "index": 20,
+                    "bbox_fs": [
+                        106,
+                        686,
+                        281,
+                        701
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        124,
+                        705,
+                        486,
+                        731
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                124,
+                                705,
+                                486,
+                                731
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        124,
+                                        705,
+                                        486,
+                                        731
+                                    ],
+                                    "score": 0.95,
+                                    "content": "\\Big | R ^ { \\pi } - \\widehat { R } ^ { \\pi } \\Big | \\leq \\mathbb { E } _ { ( s , a ) \\sim d ^ { D } } \\Big [ \\mathcal { E } _ { \\phi , \\widehat { R } _ { z } , \\widehat { T } _ { z } } ( s , a ) \\Big ] + \\int \\mathcal { E } _ { \\phi , \\widehat { R } _ { z } , \\widehat { T } _ { z } } ( s , a ) \\left( d ^ { \\pi } ( s , a ) - d ^ { D } ( s , a ) \\right) d s d a .",
+                                    "type": "interline_equation",
+                                    "image_path": "c3454a9ebef3285947433d845c8fbed141601c968a8c8c8c532f53ffce967a75.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 21,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                124,
+                                705,
+                                486,
+                                731
+                            ],
+                            "spans": [],
+                            "index": 21
+                        }
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        82,
+                        131,
+                        93
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                81,
+                                134,
+                                96
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        81,
+                                        134,
+                                        96
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Then,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        }
+                    ],
+                    "index": 0
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        142,
+                        95,
+                        468,
+                        251
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                142,
+                                95,
+                                468,
+                                251
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        142,
+                                        95,
+                                        468,
+                                        251
+                                    ],
+                                    "score": 0.95,
+                                    "content": "\\begin{array} { r l } & { \\displaystyle \\int \\mathcal { E } _ { \\phi , \\widehat { R } _ { + } , \\widehat { T } _ { z } } ( s , a ) \\left( d ^ { \\pi } ( s , a ) - d ^ { D } ( s , a ) \\right) d s d a } \\\\ & { \\quad = \\displaystyle \\int \\mathcal { E } _ { \\phi , \\widehat { R } _ { \\epsilon } , \\widehat { T } _ { z } } ( \\phi ^ { - 1 } ( z ) ) \\left( d ^ { \\pi } ( z ) - d ^ { D } \\left( \\phi ^ { - 1 } ( \\phi ^ { - 1 } ( z ) ) \\right) \\right) \\left. \\frac { d ( s , a ) } { d z } \\right. d z } \\\\ & { \\quad = \\displaystyle \\int \\mathcal { E } _ { \\phi , \\widehat { R } _ { \\epsilon } , \\widehat { T } _ { z } } ( \\phi ^ { - 1 } ( z ) ) \\left( d _ { \\phi } ^ { \\pi } ( z ) - d _ { \\phi } ^ { D } ( z ) \\right) d z } \\\\ & { \\quad \\leq { B } _ { \\phi } \\displaystyle \\left. \\int \\frac { 1 } { B _ { \\phi } } \\mathcal { E } _ { \\phi , \\widehat { R } _ { \\epsilon } , \\widehat { T } _ { z } } ( \\phi ^ { - 1 } ( z ) ) \\left( d _ { \\phi } ^ { \\pi } ( z ) - d _ { \\phi } ^ { D } ( z ) \\right) d z \\right. \\qquad \\left( { B } _ { \\phi } > 0 \\right) } \\\\ & { \\quad \\leq { B } _ { \\phi \\in \\nabla } \\displaystyle \\left. \\int \\mathcal { J } ( { \\boldsymbol \\jmath } ^ { \\pi } ) \\left( d _ { \\phi } ^ { \\pi } ( z ) - d _ { \\phi } ^ { D } ( z ) \\right) d z \\right. \\qquad \\left( \\frac { 1 } { B _ { \\phi } } \\mathcal { E } _ { \\phi , \\widehat { R } _ { \\epsilon } , \\widehat { T } _ { z } } ( \\phi ^ { - 1 } ( z ) ) \\in \\mathcal { G } \\right) } \\\\ & { \\quad = { B } _ { \\phi } \\mathrm { F N M } _ { \\phi } \\left( a _ { \\phi } ^ { \\pi } , d _ { \\phi } ^ { \\theta } \\right) } \\end{array}",
+                                    "type": "interline_equation",
+                                    "image_path": "a720027c9c93c42352414970b7e78611a3535be8979d5601c76031d1aa1ace27.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 2,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                142,
+                                95,
+                                468,
+                                147.0
+                            ],
+                            "spans": [],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                142,
+                                147.0,
+                                468,
+                                199.0
+                            ],
+                            "spans": [],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                142,
+                                199.0,
+                                468,
+                                251.0
+                            ],
+                            "spans": [],
+                            "index": 3
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        272,
+                        383,
+                        285
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                272,
+                                381,
+                                286
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        272,
+                                        381,
+                                        286
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We state some previous results, which are required for further proof.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        }
+                    ],
+                    "index": 4
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        286,
+                        505,
+                        309
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                285,
+                                506,
+                                302
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        285,
+                                        371,
+                                        302
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Theorem A.1 (McDiarmid’s inequality (McDiarmid, 1989)). Let",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        371,
+                                        286,
+                                        407,
+                                        299
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\{ X _ { i } \\} _ { i = 1 } ^ { n }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        407,
+                                        285,
+                                        506,
+                                        302
+                                    ],
+                                    "score": 1.0,
+                                    "content": "be independent random",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                296,
+                                394,
+                                311
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        296,
+                                        225,
+                                        311
+                                    ],
+                                    "score": 1.0,
+                                    "content": "variables taking values in set",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        226,
+                                        298,
+                                        235,
+                                        307
+                                    ],
+                                    "score": 0.79,
+                                    "content": "\\mathcal { X }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        235,
+                                        296,
+                                        306,
+                                        311
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", and assume that",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        307,
+                                        298,
+                                        358,
+                                        309
+                                    ],
+                                    "score": 0.92,
+                                    "content": "f : \\mathcal { X } ^ { n }  \\mathbb { R }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        358,
+                                        296,
+                                        394,
+                                        311
+                                    ],
+                                    "score": 1.0,
+                                    "content": "satisfies",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        }
+                    ],
+                    "index": 5.5
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        167,
+                        311,
+                        444,
+                        333
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                167,
+                                311,
+                                444,
+                                333
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        167,
+                                        311,
+                                        444,
+                                        333
+                                    ],
+                                    "score": 0.87,
+                                    "content": "\\operatorname* { s u p } _ { \\{ x _ { i } \\} _ { i = 1 } ^ { n } \\in \\mathcal { X } ^ { n } , \\tilde { x } \\in \\mathcal { X } } | f ( \\{ x _ { i } \\} _ { i = 1 } ^ { n } ) - f ( x _ { 1 } , . . . , x _ { i - 1 } , \\tilde { x } , x _ { i + 1 } , . . . , x _ { n } ) | \\leq c _ { i } .",
+                                    "type": "interline_equation",
+                                    "image_path": "5322ea8d75e48088fee9c308cdb79057d3b98b8e0ba25e9b81ea214f2d4e441e.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 7,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                167,
+                                311,
+                                444,
+                                333
+                            ],
+                            "spans": [],
+                            "index": 7
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        335,
+                        193,
+                        347
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                334,
+                                194,
+                                349
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        334,
+                                        167,
+                                        349
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Then for every",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        167,
+                                        336,
+                                        190,
+                                        346
+                                    ],
+                                    "score": 0.89,
+                                    "content": "\\epsilon > 0",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        191,
+                                        334,
+                                        194,
+                                        349
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        }
+                    ],
+                    "index": 8
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        163,
+                        349,
+                        450,
+                        378
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                163,
+                                349,
+                                450,
+                                378
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        163,
+                                        349,
+                                        450,
+                                        378
+                                    ],
+                                    "score": 0.91,
+                                    "content": "{ \\operatorname* { P r } } \\{ f ( \\{ X _ { i } \\} _ { i = 1 } ^ { n } ) - \\mathbb { E } _ { \\{ X _ { i } \\} _ { i = 1 } ^ { n } } [ f ( \\{ X _ { i } \\} _ { i = 1 } ^ { n } ) ] \\geq \\epsilon \\} \\leq \\exp \\left( - \\frac { 2 \\epsilon ^ { 2 } } { \\sum _ { i = 1 } ^ { n } c _ { i } ^ { 2 } } \\right) .",
+                                    "type": "interline_equation",
+                                    "image_path": "a019a80950c3511cf145cf95eb3e05d6994a05270feb86d4ea47853f31ee3a30.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 9,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                163,
+                                349,
+                                450,
+                                378
+                            ],
+                            "spans": [],
+                            "index": 9
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        384,
+                        506,
+                        441
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                385,
+                                505,
+                                397
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        385,
+                                        441,
+                                        397
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Lemma A.2 (Rademacher complexity of RKHS (Bartlett & Mendelson, 2002)). Let",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        441,
+                                        386,
+                                        450,
+                                        395
+                                    ],
+                                    "score": 0.77,
+                                    "content": "\\mathcal { F }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        451,
+                                        385,
+                                        505,
+                                        397
+                                    ],
+                                    "score": 1.0,
+                                    "content": "be a unit ball",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                396,
+                                505,
+                                409
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        396,
+                                        284,
+                                        409
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in a universal RKHS on the compact domain",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        284,
+                                        397,
+                                        293,
+                                        406
+                                    ],
+                                    "score": 0.8,
+                                    "content": "\\mathcal { X }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        293,
+                                        396,
+                                        432,
+                                        409
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", with kernel bounded according to",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        432,
+                                        396,
+                                        501,
+                                        408
+                                    ],
+                                    "score": 0.92,
+                                    "content": "0 \\leq k ( x , x ^ { \\prime } ) \\leq \\bar { k }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        501,
+                                        396,
+                                        505,
+                                        409
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                406,
+                                506,
+                                421
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        406,
+                                        122,
+                                        421
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Let",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        122,
+                                        407,
+                                        155,
+                                        419
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\{ x _ { i } \\} _ { i = 1 } ^ { n }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        155,
+                                        406,
+                                        441,
+                                        421
+                                    ],
+                                    "score": 1.0,
+                                    "content": "be an i.i.d. sample of size n drawn according to a probability measure",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        441,
+                                        410,
+                                        448,
+                                        419
+                                    ],
+                                    "score": 0.28,
+                                    "content": "p",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        448,
+                                        406,
+                                        461,
+                                        421
+                                    ],
+                                    "score": 1.0,
+                                    "content": "on",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        462,
+                                        409,
+                                        470,
+                                        417
+                                    ],
+                                    "score": 0.82,
+                                    "content": "\\mathcal { X }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        471,
+                                        406,
+                                        506,
+                                        421
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", and let",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                107,
+                                417,
+                                506,
+                                431
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        420,
+                                        117,
+                                        429
+                                    ],
+                                    "score": 0.82,
+                                    "content": "\\sigma _ { i }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        117,
+                                        417,
+                                        225,
+                                        431
+                                    ],
+                                    "score": 1.0,
+                                    "content": "be i.i.d. and take values in",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        226,
+                                        418,
+                                        258,
+                                        430
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\{ - 1 , 1 \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        258,
+                                        417,
+                                        506,
+                                        431
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with equal probability. The Rademacher complexity, which is",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                429,
+                                263,
+                                442
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        429,
+                                        263,
+                                        442
+                                    ],
+                                    "score": 1.0,
+                                    "content": "defined as below, is upper bounded as:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        }
+                    ],
+                    "index": 12
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        203,
+                        442,
+                        408,
+                        476
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                203,
+                                442,
+                                408,
+                                476
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        203,
+                                        442,
+                                        408,
+                                        476
+                                    ],
+                                    "score": 0.94,
+                                    "content": "R _ { n } ( \\mathcal { F } ) \\triangleq \\mathbb { E } _ { \\{ x _ { i } \\} _ { i = 1 } ^ { n } , \\sigma } \\operatorname* { s u p } _ { f \\in \\mathcal { F } } \\left| \\frac { 1 } { n } \\sum _ { i = 1 } ^ { n } \\sigma _ { i } f ( x _ { i } ) \\right| \\leq \\sqrt { \\frac { \\bar { k } } { n } } .",
+                                    "type": "interline_equation",
+                                    "image_path": "49bf738bde67b3d72bbf1989e2bb9b1edbd9d5aef690648e56bf852f6d38f497.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 15.5,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                203,
+                                442,
+                                408,
+                                459.0
+                            ],
+                            "spans": [],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                203,
+                                459.0,
+                                408,
+                                476.0
+                            ],
+                            "spans": [],
+                            "index": 16
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        477,
+                        411,
+                        489
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                476,
+                                413,
+                                491
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        476,
+                                        413,
+                                        491
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The upper bound is followed by Lemma 22 of (Bartlett & Mendelson, 2002).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        }
+                    ],
+                    "index": 17
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        496,
+                        317,
+                        509
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                496,
+                                318,
+                                510
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        496,
+                                        318,
+                                        510
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Now we prove the following using the results above.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        }
+                    ],
+                    "index": 18
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        510,
+                        505,
+                        543
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                509,
+                                505,
+                                523
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        509,
+                                        180,
+                                        523
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Lemma A.3. Let",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        180,
+                                        511,
+                                        194,
+                                        522
+                                    ],
+                                    "score": 0.89,
+                                    "content": "\\mathcal { H } _ { k }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        195,
+                                        509,
+                                        376,
+                                        523
+                                    ],
+                                    "score": 1.0,
+                                    "content": "be a RKHS associated with universal kernel",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        376,
+                                        510,
+                                        400,
+                                        522
+                                    ],
+                                    "score": 0.88,
+                                    "content": "k ( \\cdot , \\cdot )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        401,
+                                        509,
+                                        422,
+                                        523
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Let",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        423,
+                                        510,
+                                        452,
+                                        523
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\langle \\cdot , \\cdot \\rangle _ { \\mathcal { H } _ { k } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        452,
+                                        509,
+                                        505,
+                                        523
+                                    ],
+                                    "score": 1.0,
+                                    "content": "be the inner",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                521,
+                                505,
+                                534
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        521,
+                                        151,
+                                        534
+                                    ],
+                                    "score": 1.0,
+                                    "content": "product of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        151,
+                                        522,
+                                        165,
+                                        533
+                                    ],
+                                    "score": 0.89,
+                                    "content": "\\mathcal { H } _ { k }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        165,
+                                        521,
+                                        335,
+                                        534
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", which satisfies the reproducing property",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        335,
+                                        522,
+                                        423,
+                                        534
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\nu ( z ) = \\langle \\nu , \\dot { k } ( \\cdot , z ) \\rangle _ { \\mathcal { H } _ { k } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        424,
+                                        521,
+                                        456,
+                                        534
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". When",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        456,
+                                        522,
+                                        464,
+                                        532
+                                    ],
+                                    "score": 0.81,
+                                    "content": "\\mathcal { G }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        464,
+                                        521,
+                                        505,
+                                        534
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is chosen",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                532,
+                                147,
+                                544
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        532,
+                                        147,
+                                        544
+                                    ],
+                                    "score": 1.0,
+                                    "content": "such that",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        }
+                    ],
+                    "index": 20
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        111,
+                        542,
+                        504,
+                        575
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                111,
+                                542,
+                                504,
+                                575
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        111,
+                                        542,
+                                        504,
+                                        575
+                                    ],
+                                    "score": 0.89,
+                                    "content": "\\begin{array} { r } { \\boldsymbol { \\operatorname { \\mathrm { ? } } } = \\biggl \\{ g \\in ( \\mathcal { Z } \\to \\mathbb { R } ) : g ( z ) = \\nu ( z ) - \\gamma \\mathbb { E } _ { s ^ { \\prime } \\sim T ( \\phi ^ { - 1 } ( z ) ) } [ \\nu ( \\phi ( s ^ { \\prime } , a ^ { \\prime } ) ) ] , \\nu \\in ( \\mathcal { Z } \\to \\mathbb { R } ) , \\langle \\nu , \\nu \\rangle _ { \\mathcal { H } _ { k } } \\leq 1 \\biggr \\} , } \\end{array}",
+                                    "type": "interline_equation",
+                                    "image_path": "418aa03f21e5d2b92df1f9e04e68d4c5d9ee40bc6d1f112a6a369ea021ca8ca0.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 23,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                111,
+                                542,
+                                504,
+                                553.0
+                            ],
+                            "spans": [],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                111,
+                                553.0,
+                                504,
+                                564.0
+                            ],
+                            "spans": [],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                111,
+                                564.0,
+                                504,
+                                575.0
+                            ],
+                            "spans": [],
+                            "index": 24
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        577,
+                        347,
+                        590
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                576,
+                                348,
+                                592
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        576,
+                                        121,
+                                        592
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        121,
+                                        577,
+                                        182,
+                                        592
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\mathrm { I P M } _ { \\mathcal { G } } ( d _ { \\phi } ^ { \\pi } , d _ { \\phi } ^ { D } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        182,
+                                        576,
+                                        348,
+                                        592
+                                    ],
+                                    "score": 1.0,
+                                    "content": "has the following closed form definition:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        }
+                    ],
+                    "index": 25
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        111,
+                        593,
+                        505,
+                        667
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                111,
+                                593,
+                                505,
+                                667
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        111,
+                                        593,
+                                        505,
+                                        667
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\begin{array} { r l } & { \\mathrm { [ P M } _ { \\mathcal { G } } ( d _ { \\phi } ^ { \\pi } , d _ { \\phi } ^ { \\mathcal { D } } ) ^ { 2 } = \\mathbb { E } _ { s _ { 0 } \\sim d _ { 0 } , a _ { 0 } \\sim \\pi ( s _ { 0 } ) , ( s , a , s ^ { \\prime } ) \\sim d ^ { \\mathcal { D } } , a ^ { \\prime } \\sim \\pi ( s ^ { \\prime } ) } } \\\\ & { \\quad \\quad k ( \\phi ( s , a ) , \\phi ( \\bar { s } , \\bar { a } ) ) + ( 1 - \\gamma ) ^ { 2 } k ( \\phi ( s _ { 0 } , a _ { 0 } ) , \\phi ( \\bar { s } _ { 0 } , \\bar { a } _ { 0 } ) ) + \\gamma ^ { 2 } k ( \\phi ( s ^ { \\prime } , a ^ { \\prime } ) , \\phi ( \\bar { s } ^ { \\prime } , \\bar { a } ^ { \\prime } ) ) } \\\\ & { \\quad - 2 ( 1 - \\gamma ) k ( \\phi ( s _ { 0 } , a _ { 0 } ) , \\phi ( \\bar { s } , \\bar { a } ) ) - 2 \\gamma k ( \\phi ( s ^ { \\prime } , a ^ { \\prime } ) , \\phi ( \\bar { s } , \\bar { a } ) ) + 2 \\gamma ( 1 - \\gamma ) k ( \\phi ( s ^ { \\prime } , a ^ { \\prime } ) , \\phi ( \\bar { s } _ { 0 } , \\bar { a } _ { 0 } ) ) \\bigg ] . } \\end{array}",
+                                    "type": "interline_equation",
+                                    "image_path": "3444fb4a6f7a51d186fdee2a116daacd8024fae5a586d50294f891655a6837a3.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 27,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                111,
+                                593,
+                                505,
+                                617.6666666666666
+                            ],
+                            "spans": [],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                111,
+                                617.6666666666666,
+                                505,
+                                642.3333333333333
+                            ],
+                            "spans": [],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                111,
+                                642.3333333333333,
+                                505,
+                                666.9999999999999
+                            ],
+                            "spans": [],
+                            "index": 28
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        671,
+                        504,
+                        700
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                671,
+                                505,
+                                688
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        671,
+                                        215,
+                                        688
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Furthermore, suppose that",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        216,
+                                        673,
+                                        295,
+                                        686
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\bar { k } \\triangleq \\operatorname* { s u p } _ { z \\in \\mathcal { Z } } k ( z , z )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        295,
+                                        671,
+                                        358,
+                                        688
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". The estimator",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        358,
+                                        672,
+                                        417,
+                                        687
+                                    ],
+                                    "score": 0.94,
+                                    "content": "\\widehat { \\mathrm { I P M } } ( d _ { \\phi } ^ { \\pi } , d _ { \\phi } ^ { D } ) ^ { 2 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        417,
+                                        671,
+                                        505,
+                                        688
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", which is the sample-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                684,
+                                479,
+                                702
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        684,
+                                        187,
+                                        702
+                                    ],
+                                    "score": 1.0,
+                                    "content": "based estimation of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        187,
+                                        687,
+                                        252,
+                                        701
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\mathrm { I P M } _ { \\mathcal { G } } ( d _ { \\phi } ^ { \\pi } , d _ { \\phi } ^ { \\mathcal { D } } ) ^ { 2 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        252,
+                                        684,
+                                        275,
+                                        702
+                                    ],
+                                    "score": 1.0,
+                                    "content": "from",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        275,
+                                        691,
+                                        281,
+                                        697
+                                    ],
+                                    "score": 0.62,
+                                    "content": "n",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        281,
+                                        684,
+                                        452,
+                                        702
+                                    ],
+                                    "score": 1.0,
+                                    "content": "samples, satisfies with probability at least",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        452,
+                                        687,
+                                        474,
+                                        698
+                                    ],
+                                    "score": 0.86,
+                                    "content": "1 - \\delta",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        475,
+                                        684,
+                                        479,
+                                        702
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        }
+                    ],
+                    "index": 29.5
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        183,
+                        703,
+                        427,
+                        735
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                183,
+                                703,
+                                427,
+                                735
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        183,
+                                        703,
+                                        427,
+                                        735
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\left| \\mathrm { I P M } _ { \\mathcal { G } } ( d _ { \\phi } ^ { \\pi } , d _ { \\phi } ^ { D } ) - \\widehat { \\mathrm { I P M } } ( d _ { \\phi } ^ { \\pi } , d _ { \\phi } ^ { D } ) \\right| \\leq \\sqrt { \\frac { \\bar { k } } { n } } \\left( 4 + \\sqrt { 8 \\log \\frac { 3 } { \\delta } } \\right) .",
+                                    "type": "interline_equation",
+                                    "image_path": "851944251370897dcaa5c34d1a2a719012871dff65c5fb85cc5dc91831383dcd.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 31.5,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                183,
+                                703,
+                                427,
+                                719.0
+                            ],
+                            "spans": [],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                183,
+                                719.0,
+                                427,
+                                735.0
+                            ],
+                            "spans": [],
+                            "index": 32
+                        }
+                    ]
+                }
+            ],
+            "page_idx": 13,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        107,
+                        26,
+                        293,
+                        38
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                25,
+                                293,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        25,
+                                        293,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Published as a conference paper at ICLR 2021",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        300,
+                        751,
+                        311,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                299,
+                                750,
+                                313,
+                                764
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        299,
+                                        750,
+                                        313,
+                                        764
+                                    ],
+                                    "score": 1.0,
+                                    "content": "14",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        494,
+                        249,
+                        505,
+                        260
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                496,
+                                251,
+                                505,
+                                261
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        496,
+                                        251,
+                                        505,
+                                        261
+                                    ],
+                                    "score": 0.999,
+                                    "content": "□",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        82,
+                        131,
+                        93
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                81,
+                                134,
+                                96
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        81,
+                                        134,
+                                        96
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Then,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        }
+                    ],
+                    "index": 0,
+                    "bbox_fs": [
+                        106,
+                        81,
+                        134,
+                        96
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        142,
+                        95,
+                        468,
+                        251
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                142,
+                                95,
+                                468,
+                                251
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        142,
+                                        95,
+                                        468,
+                                        251
+                                    ],
+                                    "score": 0.95,
+                                    "content": "\\begin{array} { r l } & { \\displaystyle \\int \\mathcal { E } _ { \\phi , \\widehat { R } _ { + } , \\widehat { T } _ { z } } ( s , a ) \\left( d ^ { \\pi } ( s , a ) - d ^ { D } ( s , a ) \\right) d s d a } \\\\ & { \\quad = \\displaystyle \\int \\mathcal { E } _ { \\phi , \\widehat { R } _ { \\epsilon } , \\widehat { T } _ { z } } ( \\phi ^ { - 1 } ( z ) ) \\left( d ^ { \\pi } ( z ) - d ^ { D } \\left( \\phi ^ { - 1 } ( \\phi ^ { - 1 } ( z ) ) \\right) \\right) \\left. \\frac { d ( s , a ) } { d z } \\right. d z } \\\\ & { \\quad = \\displaystyle \\int \\mathcal { E } _ { \\phi , \\widehat { R } _ { \\epsilon } , \\widehat { T } _ { z } } ( \\phi ^ { - 1 } ( z ) ) \\left( d _ { \\phi } ^ { \\pi } ( z ) - d _ { \\phi } ^ { D } ( z ) \\right) d z } \\\\ & { \\quad \\leq { B } _ { \\phi } \\displaystyle \\left. \\int \\frac { 1 } { B _ { \\phi } } \\mathcal { E } _ { \\phi , \\widehat { R } _ { \\epsilon } , \\widehat { T } _ { z } } ( \\phi ^ { - 1 } ( z ) ) \\left( d _ { \\phi } ^ { \\pi } ( z ) - d _ { \\phi } ^ { D } ( z ) \\right) d z \\right. \\qquad \\left( { B } _ { \\phi } > 0 \\right) } \\\\ & { \\quad \\leq { B } _ { \\phi \\in \\nabla } \\displaystyle \\left. \\int \\mathcal { J } ( { \\boldsymbol \\jmath } ^ { \\pi } ) \\left( d _ { \\phi } ^ { \\pi } ( z ) - d _ { \\phi } ^ { D } ( z ) \\right) d z \\right. \\qquad \\left( \\frac { 1 } { B _ { \\phi } } \\mathcal { E } _ { \\phi , \\widehat { R } _ { \\epsilon } , \\widehat { T } _ { z } } ( \\phi ^ { - 1 } ( z ) ) \\in \\mathcal { G } \\right) } \\\\ & { \\quad = { B } _ { \\phi } \\mathrm { F N M } _ { \\phi } \\left( a _ { \\phi } ^ { \\pi } , d _ { \\phi } ^ { \\theta } \\right) } \\end{array}",
+                                    "type": "interline_equation",
+                                    "image_path": "a720027c9c93c42352414970b7e78611a3535be8979d5601c76031d1aa1ace27.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 2,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                142,
+                                95,
+                                468,
+                                147.0
+                            ],
+                            "spans": [],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                142,
+                                147.0,
+                                468,
+                                199.0
+                            ],
+                            "spans": [],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                142,
+                                199.0,
+                                468,
+                                251.0
+                            ],
+                            "spans": [],
+                            "index": 3
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        272,
+                        383,
+                        285
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                272,
+                                381,
+                                286
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        272,
+                                        381,
+                                        286
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We state some previous results, which are required for further proof.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        }
+                    ],
+                    "index": 4,
+                    "bbox_fs": [
+                        106,
+                        272,
+                        381,
+                        286
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        286,
+                        505,
+                        309
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                285,
+                                506,
+                                302
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        285,
+                                        371,
+                                        302
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Theorem A.1 (McDiarmid’s inequality (McDiarmid, 1989)). Let",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        371,
+                                        286,
+                                        407,
+                                        299
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\{ X _ { i } \\} _ { i = 1 } ^ { n }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        407,
+                                        285,
+                                        506,
+                                        302
+                                    ],
+                                    "score": 1.0,
+                                    "content": "be independent random",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                296,
+                                394,
+                                311
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        296,
+                                        225,
+                                        311
+                                    ],
+                                    "score": 1.0,
+                                    "content": "variables taking values in set",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        226,
+                                        298,
+                                        235,
+                                        307
+                                    ],
+                                    "score": 0.79,
+                                    "content": "\\mathcal { X }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        235,
+                                        296,
+                                        306,
+                                        311
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", and assume that",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        307,
+                                        298,
+                                        358,
+                                        309
+                                    ],
+                                    "score": 0.92,
+                                    "content": "f : \\mathcal { X } ^ { n }  \\mathbb { R }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        358,
+                                        296,
+                                        394,
+                                        311
+                                    ],
+                                    "score": 1.0,
+                                    "content": "satisfies",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        }
+                    ],
+                    "index": 5.5,
+                    "bbox_fs": [
+                        105,
+                        285,
+                        506,
+                        311
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        167,
+                        311,
+                        444,
+                        333
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                167,
+                                311,
+                                444,
+                                333
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        167,
+                                        311,
+                                        444,
+                                        333
+                                    ],
+                                    "score": 0.87,
+                                    "content": "\\operatorname* { s u p } _ { \\{ x _ { i } \\} _ { i = 1 } ^ { n } \\in \\mathcal { X } ^ { n } , \\tilde { x } \\in \\mathcal { X } } | f ( \\{ x _ { i } \\} _ { i = 1 } ^ { n } ) - f ( x _ { 1 } , . . . , x _ { i - 1 } , \\tilde { x } , x _ { i + 1 } , . . . , x _ { n } ) | \\leq c _ { i } .",
+                                    "type": "interline_equation",
+                                    "image_path": "5322ea8d75e48088fee9c308cdb79057d3b98b8e0ba25e9b81ea214f2d4e441e.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 7,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                167,
+                                311,
+                                444,
+                                333
+                            ],
+                            "spans": [],
+                            "index": 7
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        335,
+                        193,
+                        347
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                334,
+                                194,
+                                349
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        334,
+                                        167,
+                                        349
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Then for every",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        167,
+                                        336,
+                                        190,
+                                        346
+                                    ],
+                                    "score": 0.89,
+                                    "content": "\\epsilon > 0",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        191,
+                                        334,
+                                        194,
+                                        349
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        }
+                    ],
+                    "index": 8,
+                    "bbox_fs": [
+                        106,
+                        334,
+                        194,
+                        349
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        163,
+                        349,
+                        450,
+                        378
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                163,
+                                349,
+                                450,
+                                378
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        163,
+                                        349,
+                                        450,
+                                        378
+                                    ],
+                                    "score": 0.91,
+                                    "content": "{ \\operatorname* { P r } } \\{ f ( \\{ X _ { i } \\} _ { i = 1 } ^ { n } ) - \\mathbb { E } _ { \\{ X _ { i } \\} _ { i = 1 } ^ { n } } [ f ( \\{ X _ { i } \\} _ { i = 1 } ^ { n } ) ] \\geq \\epsilon \\} \\leq \\exp \\left( - \\frac { 2 \\epsilon ^ { 2 } } { \\sum _ { i = 1 } ^ { n } c _ { i } ^ { 2 } } \\right) .",
+                                    "type": "interline_equation",
+                                    "image_path": "a019a80950c3511cf145cf95eb3e05d6994a05270feb86d4ea47853f31ee3a30.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 9,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                163,
+                                349,
+                                450,
+                                378
+                            ],
+                            "spans": [],
+                            "index": 9
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        384,
+                        506,
+                        441
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                385,
+                                505,
+                                397
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        385,
+                                        441,
+                                        397
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Lemma A.2 (Rademacher complexity of RKHS (Bartlett & Mendelson, 2002)). Let",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        441,
+                                        386,
+                                        450,
+                                        395
+                                    ],
+                                    "score": 0.77,
+                                    "content": "\\mathcal { F }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        451,
+                                        385,
+                                        505,
+                                        397
+                                    ],
+                                    "score": 1.0,
+                                    "content": "be a unit ball",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                396,
+                                505,
+                                409
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        396,
+                                        284,
+                                        409
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in a universal RKHS on the compact domain",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        284,
+                                        397,
+                                        293,
+                                        406
+                                    ],
+                                    "score": 0.8,
+                                    "content": "\\mathcal { X }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        293,
+                                        396,
+                                        432,
+                                        409
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", with kernel bounded according to",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        432,
+                                        396,
+                                        501,
+                                        408
+                                    ],
+                                    "score": 0.92,
+                                    "content": "0 \\leq k ( x , x ^ { \\prime } ) \\leq \\bar { k }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        501,
+                                        396,
+                                        505,
+                                        409
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                406,
+                                506,
+                                421
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        406,
+                                        122,
+                                        421
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Let",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        122,
+                                        407,
+                                        155,
+                                        419
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\{ x _ { i } \\} _ { i = 1 } ^ { n }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        155,
+                                        406,
+                                        441,
+                                        421
+                                    ],
+                                    "score": 1.0,
+                                    "content": "be an i.i.d. sample of size n drawn according to a probability measure",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        441,
+                                        410,
+                                        448,
+                                        419
+                                    ],
+                                    "score": 0.28,
+                                    "content": "p",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        448,
+                                        406,
+                                        461,
+                                        421
+                                    ],
+                                    "score": 1.0,
+                                    "content": "on",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        462,
+                                        409,
+                                        470,
+                                        417
+                                    ],
+                                    "score": 0.82,
+                                    "content": "\\mathcal { X }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        471,
+                                        406,
+                                        506,
+                                        421
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", and let",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                107,
+                                417,
+                                506,
+                                431
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        420,
+                                        117,
+                                        429
+                                    ],
+                                    "score": 0.82,
+                                    "content": "\\sigma _ { i }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        117,
+                                        417,
+                                        225,
+                                        431
+                                    ],
+                                    "score": 1.0,
+                                    "content": "be i.i.d. and take values in",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        226,
+                                        418,
+                                        258,
+                                        430
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\{ - 1 , 1 \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        258,
+                                        417,
+                                        506,
+                                        431
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with equal probability. The Rademacher complexity, which is",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                429,
+                                263,
+                                442
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        429,
+                                        263,
+                                        442
+                                    ],
+                                    "score": 1.0,
+                                    "content": "defined as below, is upper bounded as:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        }
+                    ],
+                    "index": 12,
+                    "bbox_fs": [
+                        105,
+                        385,
+                        506,
+                        442
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        203,
+                        442,
+                        408,
+                        476
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                203,
+                                442,
+                                408,
+                                476
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        203,
+                                        442,
+                                        408,
+                                        476
+                                    ],
+                                    "score": 0.94,
+                                    "content": "R _ { n } ( \\mathcal { F } ) \\triangleq \\mathbb { E } _ { \\{ x _ { i } \\} _ { i = 1 } ^ { n } , \\sigma } \\operatorname* { s u p } _ { f \\in \\mathcal { F } } \\left| \\frac { 1 } { n } \\sum _ { i = 1 } ^ { n } \\sigma _ { i } f ( x _ { i } ) \\right| \\leq \\sqrt { \\frac { \\bar { k } } { n } } .",
+                                    "type": "interline_equation",
+                                    "image_path": "49bf738bde67b3d72bbf1989e2bb9b1edbd9d5aef690648e56bf852f6d38f497.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 15.5,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                203,
+                                442,
+                                408,
+                                459.0
+                            ],
+                            "spans": [],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                203,
+                                459.0,
+                                408,
+                                476.0
+                            ],
+                            "spans": [],
+                            "index": 16
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        477,
+                        411,
+                        489
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                476,
+                                413,
+                                491
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        476,
+                                        413,
+                                        491
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The upper bound is followed by Lemma 22 of (Bartlett & Mendelson, 2002).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        }
+                    ],
+                    "index": 17,
+                    "bbox_fs": [
+                        105,
+                        476,
+                        413,
+                        491
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        496,
+                        317,
+                        509
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                496,
+                                318,
+                                510
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        496,
+                                        318,
+                                        510
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Now we prove the following using the results above.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        }
+                    ],
+                    "index": 18,
+                    "bbox_fs": [
+                        106,
+                        496,
+                        318,
+                        510
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        510,
+                        505,
+                        543
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                509,
+                                505,
+                                523
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        509,
+                                        180,
+                                        523
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Lemma A.3. Let",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        180,
+                                        511,
+                                        194,
+                                        522
+                                    ],
+                                    "score": 0.89,
+                                    "content": "\\mathcal { H } _ { k }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        195,
+                                        509,
+                                        376,
+                                        523
+                                    ],
+                                    "score": 1.0,
+                                    "content": "be a RKHS associated with universal kernel",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        376,
+                                        510,
+                                        400,
+                                        522
+                                    ],
+                                    "score": 0.88,
+                                    "content": "k ( \\cdot , \\cdot )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        401,
+                                        509,
+                                        422,
+                                        523
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Let",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        423,
+                                        510,
+                                        452,
+                                        523
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\langle \\cdot , \\cdot \\rangle _ { \\mathcal { H } _ { k } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        452,
+                                        509,
+                                        505,
+                                        523
+                                    ],
+                                    "score": 1.0,
+                                    "content": "be the inner",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                521,
+                                505,
+                                534
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        521,
+                                        151,
+                                        534
+                                    ],
+                                    "score": 1.0,
+                                    "content": "product of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        151,
+                                        522,
+                                        165,
+                                        533
+                                    ],
+                                    "score": 0.89,
+                                    "content": "\\mathcal { H } _ { k }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        165,
+                                        521,
+                                        335,
+                                        534
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", which satisfies the reproducing property",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        335,
+                                        522,
+                                        423,
+                                        534
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\nu ( z ) = \\langle \\nu , \\dot { k } ( \\cdot , z ) \\rangle _ { \\mathcal { H } _ { k } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        424,
+                                        521,
+                                        456,
+                                        534
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". When",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        456,
+                                        522,
+                                        464,
+                                        532
+                                    ],
+                                    "score": 0.81,
+                                    "content": "\\mathcal { G }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        464,
+                                        521,
+                                        505,
+                                        534
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is chosen",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                532,
+                                147,
+                                544
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        532,
+                                        147,
+                                        544
+                                    ],
+                                    "score": 1.0,
+                                    "content": "such that",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        }
+                    ],
+                    "index": 20,
+                    "bbox_fs": [
+                        105,
+                        509,
+                        505,
+                        544
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        111,
+                        542,
+                        504,
+                        575
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                111,
+                                542,
+                                504,
+                                575
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        111,
+                                        542,
+                                        504,
+                                        575
+                                    ],
+                                    "score": 0.89,
+                                    "content": "\\begin{array} { r } { \\boldsymbol { \\operatorname { \\mathrm { ? } } } = \\biggl \\{ g \\in ( \\mathcal { Z } \\to \\mathbb { R } ) : g ( z ) = \\nu ( z ) - \\gamma \\mathbb { E } _ { s ^ { \\prime } \\sim T ( \\phi ^ { - 1 } ( z ) ) } [ \\nu ( \\phi ( s ^ { \\prime } , a ^ { \\prime } ) ) ] , \\nu \\in ( \\mathcal { Z } \\to \\mathbb { R } ) , \\langle \\nu , \\nu \\rangle _ { \\mathcal { H } _ { k } } \\leq 1 \\biggr \\} , } \\end{array}",
+                                    "type": "interline_equation",
+                                    "image_path": "418aa03f21e5d2b92df1f9e04e68d4c5d9ee40bc6d1f112a6a369ea021ca8ca0.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 23,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                111,
+                                542,
+                                504,
+                                553.0
+                            ],
+                            "spans": [],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                111,
+                                553.0,
+                                504,
+                                564.0
+                            ],
+                            "spans": [],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                111,
+                                564.0,
+                                504,
+                                575.0
+                            ],
+                            "spans": [],
+                            "index": 24
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        577,
+                        347,
+                        590
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                576,
+                                348,
+                                592
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        576,
+                                        121,
+                                        592
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        121,
+                                        577,
+                                        182,
+                                        592
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\mathrm { I P M } _ { \\mathcal { G } } ( d _ { \\phi } ^ { \\pi } , d _ { \\phi } ^ { D } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        182,
+                                        576,
+                                        348,
+                                        592
+                                    ],
+                                    "score": 1.0,
+                                    "content": "has the following closed form definition:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        }
+                    ],
+                    "index": 25,
+                    "bbox_fs": [
+                        106,
+                        576,
+                        348,
+                        592
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        111,
+                        593,
+                        505,
+                        667
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                111,
+                                593,
+                                505,
+                                667
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        111,
+                                        593,
+                                        505,
+                                        667
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\begin{array} { r l } & { \\mathrm { [ P M } _ { \\mathcal { G } } ( d _ { \\phi } ^ { \\pi } , d _ { \\phi } ^ { \\mathcal { D } } ) ^ { 2 } = \\mathbb { E } _ { s _ { 0 } \\sim d _ { 0 } , a _ { 0 } \\sim \\pi ( s _ { 0 } ) , ( s , a , s ^ { \\prime } ) \\sim d ^ { \\mathcal { D } } , a ^ { \\prime } \\sim \\pi ( s ^ { \\prime } ) } } \\\\ & { \\quad \\quad k ( \\phi ( s , a ) , \\phi ( \\bar { s } , \\bar { a } ) ) + ( 1 - \\gamma ) ^ { 2 } k ( \\phi ( s _ { 0 } , a _ { 0 } ) , \\phi ( \\bar { s } _ { 0 } , \\bar { a } _ { 0 } ) ) + \\gamma ^ { 2 } k ( \\phi ( s ^ { \\prime } , a ^ { \\prime } ) , \\phi ( \\bar { s } ^ { \\prime } , \\bar { a } ^ { \\prime } ) ) } \\\\ & { \\quad - 2 ( 1 - \\gamma ) k ( \\phi ( s _ { 0 } , a _ { 0 } ) , \\phi ( \\bar { s } , \\bar { a } ) ) - 2 \\gamma k ( \\phi ( s ^ { \\prime } , a ^ { \\prime } ) , \\phi ( \\bar { s } , \\bar { a } ) ) + 2 \\gamma ( 1 - \\gamma ) k ( \\phi ( s ^ { \\prime } , a ^ { \\prime } ) , \\phi ( \\bar { s } _ { 0 } , \\bar { a } _ { 0 } ) ) \\bigg ] . } \\end{array}",
+                                    "type": "interline_equation",
+                                    "image_path": "3444fb4a6f7a51d186fdee2a116daacd8024fae5a586d50294f891655a6837a3.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 27,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                111,
+                                593,
+                                505,
+                                617.6666666666666
+                            ],
+                            "spans": [],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                111,
+                                617.6666666666666,
+                                505,
+                                642.3333333333333
+                            ],
+                            "spans": [],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                111,
+                                642.3333333333333,
+                                505,
+                                666.9999999999999
+                            ],
+                            "spans": [],
+                            "index": 28
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        671,
+                        504,
+                        700
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                671,
+                                505,
+                                688
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        671,
+                                        215,
+                                        688
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Furthermore, suppose that",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        216,
+                                        673,
+                                        295,
+                                        686
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\bar { k } \\triangleq \\operatorname* { s u p } _ { z \\in \\mathcal { Z } } k ( z , z )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        295,
+                                        671,
+                                        358,
+                                        688
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". The estimator",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        358,
+                                        672,
+                                        417,
+                                        687
+                                    ],
+                                    "score": 0.94,
+                                    "content": "\\widehat { \\mathrm { I P M } } ( d _ { \\phi } ^ { \\pi } , d _ { \\phi } ^ { D } ) ^ { 2 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        417,
+                                        671,
+                                        505,
+                                        688
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", which is the sample-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                684,
+                                479,
+                                702
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        684,
+                                        187,
+                                        702
+                                    ],
+                                    "score": 1.0,
+                                    "content": "based estimation of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        187,
+                                        687,
+                                        252,
+                                        701
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\mathrm { I P M } _ { \\mathcal { G } } ( d _ { \\phi } ^ { \\pi } , d _ { \\phi } ^ { \\mathcal { D } } ) ^ { 2 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        252,
+                                        684,
+                                        275,
+                                        702
+                                    ],
+                                    "score": 1.0,
+                                    "content": "from",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        275,
+                                        691,
+                                        281,
+                                        697
+                                    ],
+                                    "score": 0.62,
+                                    "content": "n",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        281,
+                                        684,
+                                        452,
+                                        702
+                                    ],
+                                    "score": 1.0,
+                                    "content": "samples, satisfies with probability at least",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        452,
+                                        687,
+                                        474,
+                                        698
+                                    ],
+                                    "score": 0.86,
+                                    "content": "1 - \\delta",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        475,
+                                        684,
+                                        479,
+                                        702
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        }
+                    ],
+                    "index": 29.5,
+                    "bbox_fs": [
+                        105,
+                        671,
+                        505,
+                        702
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        183,
+                        703,
+                        427,
+                        735
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                183,
+                                703,
+                                427,
+                                735
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        183,
+                                        703,
+                                        427,
+                                        735
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\left| \\mathrm { I P M } _ { \\mathcal { G } } ( d _ { \\phi } ^ { \\pi } , d _ { \\phi } ^ { D } ) - \\widehat { \\mathrm { I P M } } ( d _ { \\phi } ^ { \\pi } , d _ { \\phi } ^ { D } ) \\right| \\leq \\sqrt { \\frac { \\bar { k } } { n } } \\left( 4 + \\sqrt { 8 \\log \\frac { 3 } { \\delta } } \\right) .",
+                                    "type": "interline_equation",
+                                    "image_path": "851944251370897dcaa5c34d1a2a719012871dff65c5fb85cc5dc91831383dcd.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 31.5,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                183,
+                                703,
+                                427,
+                                719.0
+                            ],
+                            "spans": [],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                183,
+                                719.0,
+                                427,
+                                735.0
+                            ],
+                            "spans": [],
+                            "index": 32
+                        }
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        81,
+                        505,
+                        105
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                80,
+                                505,
+                                96
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        80,
+                                        281,
+                                        96
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Proof. In the below we write in shorthand,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        282,
+                                        82,
+                                        308,
+                                        94
+                                    ],
+                                    "score": 0.9,
+                                    "content": "\\operatorname { I P M } _ { \\mathcal { G } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        309,
+                                        80,
+                                        350,
+                                        96
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to denote",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        350,
+                                        82,
+                                        411,
+                                        96
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\mathrm { I P M } _ { \\mathcal { G } } ( d _ { \\phi } ^ { \\pi } , d _ { \\phi } ^ { D } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        411,
+                                        80,
+                                        505,
+                                        96
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". As in Eq. (4), we can",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                95,
+                                206,
+                                106
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        95,
+                                        206,
+                                        106
+                                    ],
+                                    "score": 1.0,
+                                    "content": "rewrite the IPM term as:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        }
+                    ],
+                    "index": 0.5
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        128,
+                        105,
+                        482,
+                        139
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                128,
+                                105,
+                                482,
+                                139
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        128,
+                                        105,
+                                        482,
+                                        139
+                                    ],
+                                    "score": 0.9,
+                                    "content": "\\mathrm { I P M } _ { \\mathcal { G } } = \\underset { \\nu \\in \\mathcal { F } } { \\operatorname* { s u p } } \\left| ( 1 - \\gamma ) \\mathbb { E } _ { \\underset { a \\sim \\pi ( s ) } { s \\sim d _ { 0 } } } \\left[ \\nu ( \\phi ( s , a ) ) \\right] - \\mathbb { E } _ { ( s , a , s ^ { \\prime } ) \\sim d ^ { \\mathcal { D } } } \\left[ \\nu ( \\phi ( s , a ) ) - \\gamma \\nu ( \\phi ( s ^ { \\prime } , a ^ { \\prime } ) ) \\right] \\right| ,",
+                                    "type": "interline_equation",
+                                    "image_path": "1e9af79db67c46d6c1abe815776e0ef1c6841c3eeeba68a8b65a28072a4f3e9e.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 3,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                128,
+                                105,
+                                482,
+                                116.33333333333333
+                            ],
+                            "spans": [],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                128,
+                                116.33333333333333,
+                                482,
+                                127.66666666666666
+                            ],
+                            "spans": [],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                128,
+                                127.66666666666666,
+                                482,
+                                139.0
+                            ],
+                            "spans": [],
+                            "index": 4
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        140,
+                        504,
+                        164
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                139,
+                                506,
+                                155
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        139,
+                                        124,
+                                        155
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        124,
+                                        141,
+                                        134,
+                                        151
+                                    ],
+                                    "score": 0.84,
+                                    "content": "\\mathcal { F }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        134,
+                                        139,
+                                        193,
+                                        155
+                                    ],
+                                    "score": 1.0,
+                                    "content": "here becomes",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        194,
+                                        141,
+                                        353,
+                                        153
+                                    ],
+                                    "score": 0.9,
+                                    "content": "\\mathcal { F } = \\{ \\nu \\in ( \\mathcal { Z }  \\mathbb { R } ) : \\langle \\nu , \\nu \\rangle _ { \\mathcal { H } _ { k } } \\leq 1 \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        353,
+                                        139,
+                                        442,
+                                        155
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", a unit ball in RKHS",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        443,
+                                        141,
+                                        457,
+                                        152
+                                    ],
+                                    "score": 0.88,
+                                    "content": "\\mathcal { H } _ { k }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        457,
+                                        139,
+                                        506,
+                                        155
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Using the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                152,
+                                223,
+                                165
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        152,
+                                        204,
+                                        165
+                                    ],
+                                    "score": 1.0,
+                                    "content": "reproducing property of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        204,
+                                        153,
+                                        218,
+                                        163
+                                    ],
+                                    "score": 0.89,
+                                    "content": "\\mathcal { H } _ { k }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        219,
+                                        152,
+                                        223,
+                                        165
+                                    ],
+                                    "score": 1.0,
+                                    "content": ":",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        }
+                    ],
+                    "index": 5.5
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        118,
+                        166,
+                        491,
+                        324
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                118,
+                                166,
+                                491,
+                                324
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        118,
+                                        166,
+                                        491,
+                                        324
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\begin{array} { r l } & { \\mathrm { I P M } _ { \\mathcal { G } } ^ { 2 } = \\Bigg \\{ \\underset { s \\in \\mathcal { F } } { \\operatorname* { s u p } } \\Bigg \\{ ( 1 - \\gamma ) \\mathbb { E } _ { \\underset { \\sigma \\sim \\infty } { s \\operatorname* { s d } } } \\left[ \\nu ( \\phi s , a ) \\right] \\big ] - \\mathbb { E } _ { ( s , a , s ^ { \\prime } ) \\sim d } \\mathbb { P } \\big [ \\nu ( \\phi ( s , a ) ) - \\gamma \\nu ( \\phi ( s ^ { \\prime } , a ^ { \\prime } ) ) \\big ] \\Bigg \\} ^ { 2 } } \\\\ & { \\qquad = \\underset { s \\in \\mathcal { F } } { \\operatorname* { s u p } } \\Bigg \\{ ( 1 - \\gamma ) \\mathbb { E } _ { \\underset { \\sigma \\sim \\infty } { s \\operatorname* { s u p } } } \\big [ \\nu ( \\phi ( s , a ) ) \\big ] - \\mathbb { E } _ { ( s , a , s ^ { \\prime } ) \\sim d } \\mathbb { P } \\big [ \\nu ( \\phi ( s , a ) ) - \\gamma \\nu ( \\phi ( s ^ { \\prime } , a ^ { \\prime } ) ) \\big ] \\Bigg \\} ^ { 2 } } \\\\ & { \\qquad = \\underset { s \\in \\mathcal { F } } { \\operatorname* { s u p } } \\Bigg \\{ ( 1 - \\gamma ) \\mathbb { E } _ { \\underset { \\sigma \\sim \\infty } { s \\operatorname* { s u p } } } \\big [ ( \\nu , k ( \\cdot , \\phi ( s , a ) ) ) u _ { k } \\big ] } \\\\ & { \\qquad \\quad - \\mathbb { E } _ { ( s , a , s ^ { \\prime } ) \\sim d } \\mathbb { P } \\big [ \\nu , k ( \\cdot , \\cdot , \\phi ( s , a ) ) \\big ] u _ { k } - \\gamma \\big ( \\nu , k ( \\cdot , \\phi ( s ^ { \\prime } , a ^ { \\prime } ) ) \\big ) _ { k k } \\bigg ] \\Bigg \\} ^ { 2 } } \\\\ & { \\qquad - \\mathbb { E } _ { ( s , a , s ^ { \\prime } ) \\sim d } \\rho ^ { D } \\left[ \\big \\langle \\nu , k ( \\cdot , \\phi ( s , a ) ) \\big \\rangle _ { \\mathcal { H } _ { k } } - \\gamma \\big \\langle \\nu , k ( \\cdot , \\phi ( s ^ { \\prime } , a ^ { \\prime } ) ) \\big \\rangle _ { \\mathcal { H } _ { k } } \\right] \\Bigg \\} ^ { 2 } } \\\\ & { \\qquad = \\underset { s \\in \\mathcal { F } } { \\operatorname* { s u p } } \\big \\{ \\nu , \\nu ^ { * } \\} _ { \\mathcal { H } _ { k } } ^ { 2 } , } \\end{array}",
+                                    "type": "interline_equation",
+                                    "image_path": "bba865bd9db1881f88626ac635d3f86efbab27a10ed37efebb2b5f94e92bc50a.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 8,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                118,
+                                166,
+                                491,
+                                218.66666666666666
+                            ],
+                            "spans": [],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                118,
+                                218.66666666666666,
+                                491,
+                                271.3333333333333
+                            ],
+                            "spans": [],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                118,
+                                271.3333333333333,
+                                491,
+                                324.0
+                            ],
+                            "spans": [],
+                            "index": 9
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        324,
+                        133,
+                        334
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                322,
+                                135,
+                                336
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        322,
+                                        135,
+                                        336
+                                    ],
+                                    "score": 1.0,
+                                    "content": "where",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        }
+                    ],
+                    "index": 10
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        131,
+                        336,
+                        478,
+                        360
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                131,
+                                336,
+                                478,
+                                360
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        131,
+                                        336,
+                                        478,
+                                        360
+                                    ],
+                                    "score": 0.87,
+                                    "content": "\\begin{array} { r } { \\nu ^ { * } ( \\cdot ) = ( 1 - \\gamma ) \\mathbb { E } _ { \\stackrel { s \\sim d _ { 0 } } { a \\sim \\pi ( s ) } } \\left[ k ( \\cdot , \\phi ( s , a ) ) \\right] - \\mathbb { E } _ { ( s , a , s ^ { \\prime } ) \\sim d ^ { \\mathcal { D } } } \\left[ k ( \\cdot , \\phi ( s , a ) ) - \\gamma k ( \\cdot , \\phi ( s ^ { \\prime } , a ^ { \\prime } ) ) \\right] . } \\end{array}",
+                                    "type": "interline_equation",
+                                    "image_path": "acf18b5e3942dbcc477617e3bfbb93d7359240124fce41eeb9f3e3bc0fad7800.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 11,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                131,
+                                336,
+                                478,
+                                360
+                            ],
+                            "spans": [],
+                            "index": 11
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        362,
+                        379,
+                        374
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                361,
+                                379,
+                                376
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        361,
+                                        280,
+                                        376
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Due to the Cauchy-Schwarz inequality and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        281,
+                                        362,
+                                        375,
+                                        374
+                                    ],
+                                    "score": 0.84,
+                                    "content": "\\langle \\nu , \\nu \\rangle _ { \\mathcal { H } _ { k } } \\leq 1 , \\forall \\nu \\in \\mathcal { F }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        375,
+                                        361,
+                                        379,
+                                        376
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        }
+                    ],
+                    "index": 12
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        210,
+                        375,
+                        399,
+                        391
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                210,
+                                375,
+                                399,
+                                391
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        210,
+                                        375,
+                                        399,
+                                        391
+                                    ],
+                                    "score": 0.87,
+                                    "content": "\\langle \\nu , \\nu ^ { * } \\rangle _ { \\mathcal { H } _ { k } } ^ { 2 } \\le \\langle \\nu , \\nu \\rangle _ { \\mathcal { H } _ { k } } \\langle \\nu ^ { * } , \\nu ^ { * } \\rangle _ { \\mathcal { H } _ { k } } \\le \\langle \\nu ^ { * } , \\nu ^ { * } \\rangle _ { \\mathcal { H } _ { k } }",
+                                    "type": "interline_equation",
+                                    "image_path": "4bbe514d3b1e199bd94f527f270ca24f68cb7324d0641a8347f32aab732c9bc9.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 13,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                210,
+                                375,
+                                399,
+                                391
+                            ],
+                            "spans": [],
+                            "index": 13
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        393,
+                        501,
+                        417
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                391,
+                                501,
+                                408
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        391,
+                                        149,
+                                        408
+                                    ],
+                                    "score": 1.0,
+                                    "content": "holds and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        149,
+                                        393,
+                                        280,
+                                        406
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\begin{array} { r } { \\operatorname* { s u p } _ { \\nu \\in \\mathcal { F } } \\langle \\nu , \\nu ^ { * } \\rangle _ { \\mathcal { H } _ { k } } ^ { 2 } = \\langle \\nu ^ { * } , \\nu ^ { * } \\rangle _ { \\mathcal { H } _ { k } } } \\end{array}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        280,
+                                        391,
+                                        382,
+                                        408
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Using the property that",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        383,
+                                        393,
+                                        501,
+                                        405
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\langle k ( \\cdot , z ) , k ( \\cdot , \\bar { z } ) \\rangle _ { \\mathcal { H } _ { k } } = k ( z , \\bar { z } )",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                402,
+                                399,
+                                421
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        402,
+                                        350,
+                                        421
+                                    ],
+                                    "score": 1.0,
+                                    "content": "we can derive the closed form expression in the lemma from",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        350,
+                                        405,
+                                        394,
+                                        418
+                                    ],
+                                    "score": 0.94,
+                                    "content": "\\langle \\nu ^ { * } , \\nu ^ { * } \\rangle _ { \\mathcal { H } _ { k } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        395,
+                                        402,
+                                        399,
+                                        421
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        }
+                    ],
+                    "index": 14.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        421,
+                        489,
+                        435
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                421,
+                                491,
+                                436
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        421,
+                                        363,
+                                        436
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Now we prove the error bound of the estimator. First, we divide",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        363,
+                                        421,
+                                        424,
+                                        436
+                                    ],
+                                    "score": 0.94,
+                                    "content": "\\mathrm { I P M } _ { \\mathcal { G } } ( d _ { \\phi } ^ { \\pi } , d _ { \\phi } ^ { D } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        424,
+                                        421,
+                                        491,
+                                        436
+                                    ],
+                                    "score": 1.0,
+                                    "content": "into three parts:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        }
+                    ],
+                    "index": 16
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        156,
+                        437,
+                        489,
+                        491
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                156,
+                                437,
+                                489,
+                                491
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        156,
+                                        437,
+                                        489,
+                                        491
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\begin{array} { r l } & { \\mathrm { I P M } _ { \\mathcal { G } } ( d _ { \\phi } ^ { \\pi } , d _ { \\phi } ^ { D } ) = \\underset { \\nu \\in \\mathcal { F } } { \\operatorname* { s u p } } | f _ { 1 } ( \\nu ) + f _ { 2 } ( \\nu ) + f _ { 3 } ( \\nu ) | } \\\\ & { f _ { 1 } ( \\nu ) = ( 1 - \\gamma ) \\mathbb { E } _ { s \\sim d _ { 0 } , a \\sim \\pi ( s ) } [ \\nu ( \\phi ( s , a ) ) ] , \\qquad f _ { 2 } ( \\nu ) = - \\mathbb { E } _ { ( s , a ) \\sim d ^ { D } } \\left[ \\nu ( \\phi ( s , a ) ) \\right] , } \\\\ & { f _ { 3 } ( \\nu ) = \\mathbb { E } _ { ( s , a , s ^ { \\prime } ) \\sim d ^ { D } , a \\cdot \\sim \\pi ( s ^ { \\prime } ) } \\left[ \\gamma \\nu ( \\phi ( s ^ { \\prime } , a ^ { \\prime } ) ) \\right] . } \\end{array}",
+                                    "type": "interline_equation",
+                                    "image_path": "3e4f9a1c0fd26489b8afe531f101ab6a89e2c21c7642455f36285f71d18e8f81.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 18,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                156,
+                                437,
+                                489,
+                                455.0
+                            ],
+                            "spans": [],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                156,
+                                455.0,
+                                489,
+                                473.0
+                            ],
+                            "spans": [],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                156,
+                                473.0,
+                                489,
+                                491.0
+                            ],
+                            "spans": [],
+                            "index": 19
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        492,
+                        505,
+                        536
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                101,
+                                485,
+                                505,
+                                515
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        101,
+                                        485,
+                                        133,
+                                        515
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Given",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        133,
+                                        497,
+                                        141,
+                                        505
+                                    ],
+                                    "score": 0.75,
+                                    "content": "n",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        141,
+                                        485,
+                                        178,
+                                        515
+                                    ],
+                                    "score": 1.0,
+                                    "content": "samples",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        178,
+                                        492,
+                                        316,
+                                        507
+                                    ],
+                                    "score": 0.9,
+                                    "content": "\\{ s _ { 0 } ^ { ( i ) } , a _ { 0 } ^ { ( i ) } , s ^ { ( i ) } , a ^ { ( i ) } , s ^ { \\prime ( i ) } , a ^ { \\prime ( i ) } \\} _ { i = 1 } ^ { n }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        316,
+                                        485,
+                                        433,
+                                        515
+                                    ],
+                                    "score": 1.0,
+                                    "content": "from the generative process",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        433,
+                                        492,
+                                        505,
+                                        507
+                                    ],
+                                    "score": 0.92,
+                                    "content": "s _ { 0 } ^ { ( i ) } \\sim d _ { 0 } , a _ { 0 } ^ { ( i ) } \\sim",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                504,
+                                508,
+                                524
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        507,
+                                        335,
+                                        521
+                                    ],
+                                    "score": 0.87,
+                                    "content": "\\pi ( s _ { 0 } ^ { ( i ) } ) , ( s ^ { ( i ) } , a ^ { ( i ) } , s ^ { \\prime ( i ) } ) ~ \\sim ~ d ^ { D } , a ^ { \\prime ( i ) } ~ \\sim ~ \\pi ( s ^ { \\prime ( i ) } ) ~ \\forall i ,",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        335,
+                                        504,
+                                        508,
+                                        524
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", we define the sample-based estimator",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                519,
+                                168,
+                                538
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        521,
+                                        162,
+                                        536
+                                    ],
+                                    "score": 0.89,
+                                    "content": "\\widehat { \\mathrm { I P M } } ( d _ { \\phi } ^ { \\pi } , d _ { \\phi } ^ { D } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        162,
+                                        519,
+                                        168,
+                                        538
+                                    ],
+                                    "score": 1.0,
+                                    "content": ":",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        }
+                    ],
+                    "index": 21
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        112,
+                        537,
+                        497,
+                        616
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                112,
+                                537,
+                                497,
+                                616
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        112,
+                                        537,
+                                        497,
+                                        616
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\begin{array} { l } { { \\displaystyle \\widehat { \\mathrm { I P M } } ( d _ { \\phi } ^ { \\pi } , d _ { \\phi } ^ { \\mathcal { D } } ) ^ { 2 } = \\frac { 1 } { n ^ { 2 } } \\sum _ { i , j } \\bigg [ k ( \\phi ( s ^ { ( i ) } , a ^ { ( i ) } ) , \\phi ( s ^ { ( j ) } , a ^ { ( j ) } ) ) + ( 1 - \\gamma ) ^ { 2 } k ( \\phi ( s _ { 0 } ^ { ( i ) } , a _ { 0 } ^ { ( i ) } ) , \\phi ( s _ { 0 } ^ { ( j ) } , a _ { 0 } ^ { ( j ) } ) ) } } \\\\ { { \\displaystyle \\qquad + \\gamma ^ { 2 } k ( \\phi ( s ^ { \\prime ( i ) } , a ^ { \\prime ( i ) } ) , \\phi ( s ^ { \\prime ( j ) } , a ^ { \\prime ( j ) } ) ) - 2 ( 1 - \\gamma ) k ( \\phi ( s _ { 0 } ^ { ( i ) } , a _ { 0 } ^ { ( i ) } ) , \\phi ( s ^ { ( j ) } , a ^ { ( j ) } ) ) } } \\\\ { { \\displaystyle \\qquad - 2 \\gamma k ( \\phi ( s ^ { \\prime ( i ) } , a ^ { \\prime ( i ) } ) , \\phi ( s ^ { ( j ) } , a ^ { ( j ) } ) ) + 2 \\gamma ( 1 - \\gamma ) k ( \\phi ( s ^ { \\prime ( i ) } , a ^ { \\prime ( i ) } ) , \\phi ( s _ { 0 } ^ { ( j ) } , a _ { 0 } ^ { ( j ) } ) ) \\bigg ] . } } \\end{array}",
+                                    "type": "interline_equation",
+                                    "image_path": "c8bc39946b4a96a2690e87ba300f82a151470d102975b6f07871841d792ad7d0.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 24,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                112,
+                                537,
+                                497,
+                                563.3333333333334
+                            ],
+                            "spans": [],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                112,
+                                563.3333333333334,
+                                497,
+                                589.6666666666667
+                            ],
+                            "spans": [],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                112,
+                                589.6666666666667,
+                                497,
+                                616.0000000000001
+                            ],
+                            "spans": [],
+                            "index": 25
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        615,
+                        505,
+                        638
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                615,
+                                505,
+                                628
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        615,
+                                        505,
+                                        628
+                                    ],
+                                    "score": 1.0,
+                                    "content": "By deriving in reverse order, we can recover its another definition in supremum, which can be",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                626,
+                                204,
+                                639
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        626,
+                                        204,
+                                        639
+                                    ],
+                                    "score": 1.0,
+                                    "content": "divided into three parts:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        }
+                    ],
+                    "index": 26.5
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        132,
+                        638,
+                        477,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                132,
+                                638,
+                                477,
+                                732
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        132,
+                                        638,
+                                        477,
+                                        732
+                                    ],
+                                    "score": 0.94,
+                                    "content": "\\begin{array} { r l } & { \\widehat { \\mathrm { I P M } } ( d _ { \\phi } ^ { \\pi } , d _ { \\phi } ^ { 2 D } ) = \\displaystyle \\operatorname* { s u p } _ { \\nu \\in \\mathcal { F } } \\Big | \\widehat { f } _ { 1 } ( \\nu ) + \\widehat { f } _ { 2 } ( \\nu ) + \\widehat { f } _ { 3 } ( \\nu ) \\Big | } \\\\ & { \\mathrm { w h e r e } \\widehat { f } _ { 1 } ( \\nu ) = \\displaystyle \\frac { 1 - \\gamma } { n } \\sum _ { i = 1 } ^ { n } \\nu \\left( \\phi \\left( s _ { 0 } ^ { ( i ) } , a _ { 0 } ^ { ( i ) } \\right) \\right) , \\widehat { f } _ { 2 } ( \\nu ) = - \\displaystyle \\frac { 1 } { n } \\sum _ { i = 1 } ^ { n } \\nu \\left( \\phi \\left( s ^ { ( i ) } , a ^ { ( i ) } \\right) \\right) , } \\\\ & { \\widehat { f } _ { 3 } ( \\nu ) = \\displaystyle \\frac { \\gamma } { n } \\sum _ { i = 1 } ^ { n } \\nu \\left( \\phi \\left( s ^ { \\prime ( i ) } , a ^ { \\prime ( i ) } \\right) \\right) . } \\end{array}",
+                                    "type": "interline_equation",
+                                    "image_path": "48a4c81c6e8b78cd0a30ed57cc221fcd7dfa0fef593289ae20126a390d8e9431.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 29,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                132,
+                                638,
+                                477,
+                                669.3333333333334
+                            ],
+                            "spans": [],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                132,
+                                669.3333333333334,
+                                477,
+                                700.6666666666667
+                            ],
+                            "spans": [],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                132,
+                                700.6666666666667,
+                                477,
+                                732.0000000000001
+                            ],
+                            "spans": [],
+                            "index": 30
+                        }
+                    ]
+                }
+            ],
+            "page_idx": 14,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        107,
+                        26,
+                        293,
+                        38
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                25,
+                                293,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        25,
+                                        293,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Published as a conference paper at ICLR 2021",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        300,
+                        751,
+                        311,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                299,
+                                750,
+                                313,
+                                764
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        299,
+                                        750,
+                                        313,
+                                        764
+                                    ],
+                                    "score": 1.0,
+                                    "content": "15",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        81,
+                        505,
+                        105
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                80,
+                                505,
+                                96
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        80,
+                                        281,
+                                        96
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Proof. In the below we write in shorthand,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        282,
+                                        82,
+                                        308,
+                                        94
+                                    ],
+                                    "score": 0.9,
+                                    "content": "\\operatorname { I P M } _ { \\mathcal { G } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        309,
+                                        80,
+                                        350,
+                                        96
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to denote",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        350,
+                                        82,
+                                        411,
+                                        96
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\mathrm { I P M } _ { \\mathcal { G } } ( d _ { \\phi } ^ { \\pi } , d _ { \\phi } ^ { D } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        411,
+                                        80,
+                                        505,
+                                        96
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". As in Eq. (4), we can",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                95,
+                                206,
+                                106
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        95,
+                                        206,
+                                        106
+                                    ],
+                                    "score": 1.0,
+                                    "content": "rewrite the IPM term as:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        }
+                    ],
+                    "index": 0.5,
+                    "bbox_fs": [
+                        105,
+                        80,
+                        505,
+                        106
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        128,
+                        105,
+                        482,
+                        139
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                128,
+                                105,
+                                482,
+                                139
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        128,
+                                        105,
+                                        482,
+                                        139
+                                    ],
+                                    "score": 0.9,
+                                    "content": "\\mathrm { I P M } _ { \\mathcal { G } } = \\underset { \\nu \\in \\mathcal { F } } { \\operatorname* { s u p } } \\left| ( 1 - \\gamma ) \\mathbb { E } _ { \\underset { a \\sim \\pi ( s ) } { s \\sim d _ { 0 } } } \\left[ \\nu ( \\phi ( s , a ) ) \\right] - \\mathbb { E } _ { ( s , a , s ^ { \\prime } ) \\sim d ^ { \\mathcal { D } } } \\left[ \\nu ( \\phi ( s , a ) ) - \\gamma \\nu ( \\phi ( s ^ { \\prime } , a ^ { \\prime } ) ) \\right] \\right| ,",
+                                    "type": "interline_equation",
+                                    "image_path": "1e9af79db67c46d6c1abe815776e0ef1c6841c3eeeba68a8b65a28072a4f3e9e.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 3,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                128,
+                                105,
+                                482,
+                                116.33333333333333
+                            ],
+                            "spans": [],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                128,
+                                116.33333333333333,
+                                482,
+                                127.66666666666666
+                            ],
+                            "spans": [],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                128,
+                                127.66666666666666,
+                                482,
+                                139.0
+                            ],
+                            "spans": [],
+                            "index": 4
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        140,
+                        504,
+                        164
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                139,
+                                506,
+                                155
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        139,
+                                        124,
+                                        155
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        124,
+                                        141,
+                                        134,
+                                        151
+                                    ],
+                                    "score": 0.84,
+                                    "content": "\\mathcal { F }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        134,
+                                        139,
+                                        193,
+                                        155
+                                    ],
+                                    "score": 1.0,
+                                    "content": "here becomes",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        194,
+                                        141,
+                                        353,
+                                        153
+                                    ],
+                                    "score": 0.9,
+                                    "content": "\\mathcal { F } = \\{ \\nu \\in ( \\mathcal { Z }  \\mathbb { R } ) : \\langle \\nu , \\nu \\rangle _ { \\mathcal { H } _ { k } } \\leq 1 \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        353,
+                                        139,
+                                        442,
+                                        155
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", a unit ball in RKHS",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        443,
+                                        141,
+                                        457,
+                                        152
+                                    ],
+                                    "score": 0.88,
+                                    "content": "\\mathcal { H } _ { k }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        457,
+                                        139,
+                                        506,
+                                        155
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Using the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                152,
+                                223,
+                                165
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        152,
+                                        204,
+                                        165
+                                    ],
+                                    "score": 1.0,
+                                    "content": "reproducing property of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        204,
+                                        153,
+                                        218,
+                                        163
+                                    ],
+                                    "score": 0.89,
+                                    "content": "\\mathcal { H } _ { k }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        219,
+                                        152,
+                                        223,
+                                        165
+                                    ],
+                                    "score": 1.0,
+                                    "content": ":",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        }
+                    ],
+                    "index": 5.5,
+                    "bbox_fs": [
+                        105,
+                        139,
+                        506,
+                        165
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        118,
+                        166,
+                        491,
+                        324
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                118,
+                                166,
+                                491,
+                                324
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        118,
+                                        166,
+                                        491,
+                                        324
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\begin{array} { r l } & { \\mathrm { I P M } _ { \\mathcal { G } } ^ { 2 } = \\Bigg \\{ \\underset { s \\in \\mathcal { F } } { \\operatorname* { s u p } } \\Bigg \\{ ( 1 - \\gamma ) \\mathbb { E } _ { \\underset { \\sigma \\sim \\infty } { s \\operatorname* { s d } } } \\left[ \\nu ( \\phi s , a ) \\right] \\big ] - \\mathbb { E } _ { ( s , a , s ^ { \\prime } ) \\sim d } \\mathbb { P } \\big [ \\nu ( \\phi ( s , a ) ) - \\gamma \\nu ( \\phi ( s ^ { \\prime } , a ^ { \\prime } ) ) \\big ] \\Bigg \\} ^ { 2 } } \\\\ & { \\qquad = \\underset { s \\in \\mathcal { F } } { \\operatorname* { s u p } } \\Bigg \\{ ( 1 - \\gamma ) \\mathbb { E } _ { \\underset { \\sigma \\sim \\infty } { s \\operatorname* { s u p } } } \\big [ \\nu ( \\phi ( s , a ) ) \\big ] - \\mathbb { E } _ { ( s , a , s ^ { \\prime } ) \\sim d } \\mathbb { P } \\big [ \\nu ( \\phi ( s , a ) ) - \\gamma \\nu ( \\phi ( s ^ { \\prime } , a ^ { \\prime } ) ) \\big ] \\Bigg \\} ^ { 2 } } \\\\ & { \\qquad = \\underset { s \\in \\mathcal { F } } { \\operatorname* { s u p } } \\Bigg \\{ ( 1 - \\gamma ) \\mathbb { E } _ { \\underset { \\sigma \\sim \\infty } { s \\operatorname* { s u p } } } \\big [ ( \\nu , k ( \\cdot , \\phi ( s , a ) ) ) u _ { k } \\big ] } \\\\ & { \\qquad \\quad - \\mathbb { E } _ { ( s , a , s ^ { \\prime } ) \\sim d } \\mathbb { P } \\big [ \\nu , k ( \\cdot , \\cdot , \\phi ( s , a ) ) \\big ] u _ { k } - \\gamma \\big ( \\nu , k ( \\cdot , \\phi ( s ^ { \\prime } , a ^ { \\prime } ) ) \\big ) _ { k k } \\bigg ] \\Bigg \\} ^ { 2 } } \\\\ & { \\qquad - \\mathbb { E } _ { ( s , a , s ^ { \\prime } ) \\sim d } \\rho ^ { D } \\left[ \\big \\langle \\nu , k ( \\cdot , \\phi ( s , a ) ) \\big \\rangle _ { \\mathcal { H } _ { k } } - \\gamma \\big \\langle \\nu , k ( \\cdot , \\phi ( s ^ { \\prime } , a ^ { \\prime } ) ) \\big \\rangle _ { \\mathcal { H } _ { k } } \\right] \\Bigg \\} ^ { 2 } } \\\\ & { \\qquad = \\underset { s \\in \\mathcal { F } } { \\operatorname* { s u p } } \\big \\{ \\nu , \\nu ^ { * } \\} _ { \\mathcal { H } _ { k } } ^ { 2 } , } \\end{array}",
+                                    "type": "interline_equation",
+                                    "image_path": "bba865bd9db1881f88626ac635d3f86efbab27a10ed37efebb2b5f94e92bc50a.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 8,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                118,
+                                166,
+                                491,
+                                218.66666666666666
+                            ],
+                            "spans": [],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                118,
+                                218.66666666666666,
+                                491,
+                                271.3333333333333
+                            ],
+                            "spans": [],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                118,
+                                271.3333333333333,
+                                491,
+                                324.0
+                            ],
+                            "spans": [],
+                            "index": 9
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        324,
+                        133,
+                        334
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                322,
+                                135,
+                                336
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        322,
+                                        135,
+                                        336
+                                    ],
+                                    "score": 1.0,
+                                    "content": "where",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        }
+                    ],
+                    "index": 10,
+                    "bbox_fs": [
+                        105,
+                        322,
+                        135,
+                        336
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        131,
+                        336,
+                        478,
+                        360
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                131,
+                                336,
+                                478,
+                                360
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        131,
+                                        336,
+                                        478,
+                                        360
+                                    ],
+                                    "score": 0.87,
+                                    "content": "\\begin{array} { r } { \\nu ^ { * } ( \\cdot ) = ( 1 - \\gamma ) \\mathbb { E } _ { \\stackrel { s \\sim d _ { 0 } } { a \\sim \\pi ( s ) } } \\left[ k ( \\cdot , \\phi ( s , a ) ) \\right] - \\mathbb { E } _ { ( s , a , s ^ { \\prime } ) \\sim d ^ { \\mathcal { D } } } \\left[ k ( \\cdot , \\phi ( s , a ) ) - \\gamma k ( \\cdot , \\phi ( s ^ { \\prime } , a ^ { \\prime } ) ) \\right] . } \\end{array}",
+                                    "type": "interline_equation",
+                                    "image_path": "acf18b5e3942dbcc477617e3bfbb93d7359240124fce41eeb9f3e3bc0fad7800.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 11,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                131,
+                                336,
+                                478,
+                                360
+                            ],
+                            "spans": [],
+                            "index": 11
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        362,
+                        379,
+                        374
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                361,
+                                379,
+                                376
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        361,
+                                        280,
+                                        376
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Due to the Cauchy-Schwarz inequality and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        281,
+                                        362,
+                                        375,
+                                        374
+                                    ],
+                                    "score": 0.84,
+                                    "content": "\\langle \\nu , \\nu \\rangle _ { \\mathcal { H } _ { k } } \\leq 1 , \\forall \\nu \\in \\mathcal { F }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        375,
+                                        361,
+                                        379,
+                                        376
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        }
+                    ],
+                    "index": 12,
+                    "bbox_fs": [
+                        105,
+                        361,
+                        379,
+                        376
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        210,
+                        375,
+                        399,
+                        391
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                210,
+                                375,
+                                399,
+                                391
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        210,
+                                        375,
+                                        399,
+                                        391
+                                    ],
+                                    "score": 0.87,
+                                    "content": "\\langle \\nu , \\nu ^ { * } \\rangle _ { \\mathcal { H } _ { k } } ^ { 2 } \\le \\langle \\nu , \\nu \\rangle _ { \\mathcal { H } _ { k } } \\langle \\nu ^ { * } , \\nu ^ { * } \\rangle _ { \\mathcal { H } _ { k } } \\le \\langle \\nu ^ { * } , \\nu ^ { * } \\rangle _ { \\mathcal { H } _ { k } }",
+                                    "type": "interline_equation",
+                                    "image_path": "4bbe514d3b1e199bd94f527f270ca24f68cb7324d0641a8347f32aab732c9bc9.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 13,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                210,
+                                375,
+                                399,
+                                391
+                            ],
+                            "spans": [],
+                            "index": 13
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        393,
+                        501,
+                        417
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                391,
+                                501,
+                                408
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        391,
+                                        149,
+                                        408
+                                    ],
+                                    "score": 1.0,
+                                    "content": "holds and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        149,
+                                        393,
+                                        280,
+                                        406
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\begin{array} { r } { \\operatorname* { s u p } _ { \\nu \\in \\mathcal { F } } \\langle \\nu , \\nu ^ { * } \\rangle _ { \\mathcal { H } _ { k } } ^ { 2 } = \\langle \\nu ^ { * } , \\nu ^ { * } \\rangle _ { \\mathcal { H } _ { k } } } \\end{array}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        280,
+                                        391,
+                                        382,
+                                        408
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Using the property that",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        383,
+                                        393,
+                                        501,
+                                        405
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\langle k ( \\cdot , z ) , k ( \\cdot , \\bar { z } ) \\rangle _ { \\mathcal { H } _ { k } } = k ( z , \\bar { z } )",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                402,
+                                399,
+                                421
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        402,
+                                        350,
+                                        421
+                                    ],
+                                    "score": 1.0,
+                                    "content": "we can derive the closed form expression in the lemma from",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        350,
+                                        405,
+                                        394,
+                                        418
+                                    ],
+                                    "score": 0.94,
+                                    "content": "\\langle \\nu ^ { * } , \\nu ^ { * } \\rangle _ { \\mathcal { H } _ { k } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        395,
+                                        402,
+                                        399,
+                                        421
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        }
+                    ],
+                    "index": 14.5,
+                    "bbox_fs": [
+                        105,
+                        391,
+                        501,
+                        421
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        421,
+                        489,
+                        435
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                421,
+                                491,
+                                436
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        421,
+                                        363,
+                                        436
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Now we prove the error bound of the estimator. First, we divide",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        363,
+                                        421,
+                                        424,
+                                        436
+                                    ],
+                                    "score": 0.94,
+                                    "content": "\\mathrm { I P M } _ { \\mathcal { G } } ( d _ { \\phi } ^ { \\pi } , d _ { \\phi } ^ { D } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        424,
+                                        421,
+                                        491,
+                                        436
+                                    ],
+                                    "score": 1.0,
+                                    "content": "into three parts:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        }
+                    ],
+                    "index": 16,
+                    "bbox_fs": [
+                        105,
+                        421,
+                        491,
+                        436
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        156,
+                        437,
+                        489,
+                        491
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                156,
+                                437,
+                                489,
+                                491
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        156,
+                                        437,
+                                        489,
+                                        491
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\begin{array} { r l } & { \\mathrm { I P M } _ { \\mathcal { G } } ( d _ { \\phi } ^ { \\pi } , d _ { \\phi } ^ { D } ) = \\underset { \\nu \\in \\mathcal { F } } { \\operatorname* { s u p } } | f _ { 1 } ( \\nu ) + f _ { 2 } ( \\nu ) + f _ { 3 } ( \\nu ) | } \\\\ & { f _ { 1 } ( \\nu ) = ( 1 - \\gamma ) \\mathbb { E } _ { s \\sim d _ { 0 } , a \\sim \\pi ( s ) } [ \\nu ( \\phi ( s , a ) ) ] , \\qquad f _ { 2 } ( \\nu ) = - \\mathbb { E } _ { ( s , a ) \\sim d ^ { D } } \\left[ \\nu ( \\phi ( s , a ) ) \\right] , } \\\\ & { f _ { 3 } ( \\nu ) = \\mathbb { E } _ { ( s , a , s ^ { \\prime } ) \\sim d ^ { D } , a \\cdot \\sim \\pi ( s ^ { \\prime } ) } \\left[ \\gamma \\nu ( \\phi ( s ^ { \\prime } , a ^ { \\prime } ) ) \\right] . } \\end{array}",
+                                    "type": "interline_equation",
+                                    "image_path": "3e4f9a1c0fd26489b8afe531f101ab6a89e2c21c7642455f36285f71d18e8f81.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 18,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                156,
+                                437,
+                                489,
+                                455.0
+                            ],
+                            "spans": [],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                156,
+                                455.0,
+                                489,
+                                473.0
+                            ],
+                            "spans": [],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                156,
+                                473.0,
+                                489,
+                                491.0
+                            ],
+                            "spans": [],
+                            "index": 19
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        492,
+                        505,
+                        536
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                101,
+                                485,
+                                505,
+                                515
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        101,
+                                        485,
+                                        133,
+                                        515
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Given",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        133,
+                                        497,
+                                        141,
+                                        505
+                                    ],
+                                    "score": 0.75,
+                                    "content": "n",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        141,
+                                        485,
+                                        178,
+                                        515
+                                    ],
+                                    "score": 1.0,
+                                    "content": "samples",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        178,
+                                        492,
+                                        316,
+                                        507
+                                    ],
+                                    "score": 0.9,
+                                    "content": "\\{ s _ { 0 } ^ { ( i ) } , a _ { 0 } ^ { ( i ) } , s ^ { ( i ) } , a ^ { ( i ) } , s ^ { \\prime ( i ) } , a ^ { \\prime ( i ) } \\} _ { i = 1 } ^ { n }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        316,
+                                        485,
+                                        433,
+                                        515
+                                    ],
+                                    "score": 1.0,
+                                    "content": "from the generative process",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        433,
+                                        492,
+                                        505,
+                                        507
+                                    ],
+                                    "score": 0.92,
+                                    "content": "s _ { 0 } ^ { ( i ) } \\sim d _ { 0 } , a _ { 0 } ^ { ( i ) } \\sim",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                504,
+                                508,
+                                524
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        507,
+                                        335,
+                                        521
+                                    ],
+                                    "score": 0.87,
+                                    "content": "\\pi ( s _ { 0 } ^ { ( i ) } ) , ( s ^ { ( i ) } , a ^ { ( i ) } , s ^ { \\prime ( i ) } ) ~ \\sim ~ d ^ { D } , a ^ { \\prime ( i ) } ~ \\sim ~ \\pi ( s ^ { \\prime ( i ) } ) ~ \\forall i ,",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        335,
+                                        504,
+                                        508,
+                                        524
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", we define the sample-based estimator",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                519,
+                                168,
+                                538
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        521,
+                                        162,
+                                        536
+                                    ],
+                                    "score": 0.89,
+                                    "content": "\\widehat { \\mathrm { I P M } } ( d _ { \\phi } ^ { \\pi } , d _ { \\phi } ^ { D } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        162,
+                                        519,
+                                        168,
+                                        538
+                                    ],
+                                    "score": 1.0,
+                                    "content": ":",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        }
+                    ],
+                    "index": 21,
+                    "bbox_fs": [
+                        101,
+                        485,
+                        508,
+                        538
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        112,
+                        537,
+                        497,
+                        616
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                112,
+                                537,
+                                497,
+                                616
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        112,
+                                        537,
+                                        497,
+                                        616
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\begin{array} { l } { { \\displaystyle \\widehat { \\mathrm { I P M } } ( d _ { \\phi } ^ { \\pi } , d _ { \\phi } ^ { \\mathcal { D } } ) ^ { 2 } = \\frac { 1 } { n ^ { 2 } } \\sum _ { i , j } \\bigg [ k ( \\phi ( s ^ { ( i ) } , a ^ { ( i ) } ) , \\phi ( s ^ { ( j ) } , a ^ { ( j ) } ) ) + ( 1 - \\gamma ) ^ { 2 } k ( \\phi ( s _ { 0 } ^ { ( i ) } , a _ { 0 } ^ { ( i ) } ) , \\phi ( s _ { 0 } ^ { ( j ) } , a _ { 0 } ^ { ( j ) } ) ) } } \\\\ { { \\displaystyle \\qquad + \\gamma ^ { 2 } k ( \\phi ( s ^ { \\prime ( i ) } , a ^ { \\prime ( i ) } ) , \\phi ( s ^ { \\prime ( j ) } , a ^ { \\prime ( j ) } ) ) - 2 ( 1 - \\gamma ) k ( \\phi ( s _ { 0 } ^ { ( i ) } , a _ { 0 } ^ { ( i ) } ) , \\phi ( s ^ { ( j ) } , a ^ { ( j ) } ) ) } } \\\\ { { \\displaystyle \\qquad - 2 \\gamma k ( \\phi ( s ^ { \\prime ( i ) } , a ^ { \\prime ( i ) } ) , \\phi ( s ^ { ( j ) } , a ^ { ( j ) } ) ) + 2 \\gamma ( 1 - \\gamma ) k ( \\phi ( s ^ { \\prime ( i ) } , a ^ { \\prime ( i ) } ) , \\phi ( s _ { 0 } ^ { ( j ) } , a _ { 0 } ^ { ( j ) } ) ) \\bigg ] . } } \\end{array}",
+                                    "type": "interline_equation",
+                                    "image_path": "c8bc39946b4a96a2690e87ba300f82a151470d102975b6f07871841d792ad7d0.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 24,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                112,
+                                537,
+                                497,
+                                563.3333333333334
+                            ],
+                            "spans": [],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                112,
+                                563.3333333333334,
+                                497,
+                                589.6666666666667
+                            ],
+                            "spans": [],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                112,
+                                589.6666666666667,
+                                497,
+                                616.0000000000001
+                            ],
+                            "spans": [],
+                            "index": 25
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        615,
+                        505,
+                        638
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                615,
+                                505,
+                                628
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        615,
+                                        505,
+                                        628
+                                    ],
+                                    "score": 1.0,
+                                    "content": "By deriving in reverse order, we can recover its another definition in supremum, which can be",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                626,
+                                204,
+                                639
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        626,
+                                        204,
+                                        639
+                                    ],
+                                    "score": 1.0,
+                                    "content": "divided into three parts:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        }
+                    ],
+                    "index": 26.5,
+                    "bbox_fs": [
+                        105,
+                        615,
+                        505,
+                        639
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        132,
+                        638,
+                        477,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                132,
+                                638,
+                                477,
+                                732
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        132,
+                                        638,
+                                        477,
+                                        732
+                                    ],
+                                    "score": 0.94,
+                                    "content": "\\begin{array} { r l } & { \\widehat { \\mathrm { I P M } } ( d _ { \\phi } ^ { \\pi } , d _ { \\phi } ^ { 2 D } ) = \\displaystyle \\operatorname* { s u p } _ { \\nu \\in \\mathcal { F } } \\Big | \\widehat { f } _ { 1 } ( \\nu ) + \\widehat { f } _ { 2 } ( \\nu ) + \\widehat { f } _ { 3 } ( \\nu ) \\Big | } \\\\ & { \\mathrm { w h e r e } \\widehat { f } _ { 1 } ( \\nu ) = \\displaystyle \\frac { 1 - \\gamma } { n } \\sum _ { i = 1 } ^ { n } \\nu \\left( \\phi \\left( s _ { 0 } ^ { ( i ) } , a _ { 0 } ^ { ( i ) } \\right) \\right) , \\widehat { f } _ { 2 } ( \\nu ) = - \\displaystyle \\frac { 1 } { n } \\sum _ { i = 1 } ^ { n } \\nu \\left( \\phi \\left( s ^ { ( i ) } , a ^ { ( i ) } \\right) \\right) , } \\\\ & { \\widehat { f } _ { 3 } ( \\nu ) = \\displaystyle \\frac { \\gamma } { n } \\sum _ { i = 1 } ^ { n } \\nu \\left( \\phi \\left( s ^ { \\prime ( i ) } , a ^ { \\prime ( i ) } \\right) \\right) . } \\end{array}",
+                                    "type": "interline_equation",
+                                    "image_path": "48a4c81c6e8b78cd0a30ed57cc221fcd7dfa0fef593289ae20126a390d8e9431.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 29,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                132,
+                                638,
+                                477,
+                                669.3333333333334
+                            ],
+                            "spans": [],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                132,
+                                669.3333333333334,
+                                477,
+                                700.6666666666667
+                            ],
+                            "spans": [],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                132,
+                                700.6666666666667,
+                                477,
+                                732.0000000000001
+                            ],
+                            "spans": [],
+                            "index": 30
+                        }
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        82,
+                        411,
+                        94
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                82,
+                                412,
+                                95
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        82,
+                                        412,
+                                        95
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We can bound the error of sample-based estimator with individual errors as:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        }
+                    ],
+                    "index": 0
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        111,
+                        97,
+                        513,
+                        228
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                111,
+                                97,
+                                513,
+                                228
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        111,
+                                        97,
+                                        513,
+                                        228
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\begin{array} { r l } { \\mathrm { I P M } _ { \\mathcal { G } } ( d _ { \\phi } ^ { \\pi } , d _ { \\phi } ^ { 2 } ) - \\widehat { \\mathrm { I P M } } ( d _ { \\phi } ^ { \\pi } , d _ { \\phi } ^ { \\mathcal { D } } ) \\Bigg | = \\Bigg | \\underset { \\nu \\in \\mathcal { F } } { \\operatorname* { s u p } } | f _ { 1 } ( \\nu ) + f _ { 2 } ( \\nu ) + f _ { 3 } ( \\nu ) | - \\underset { \\nu \\in \\mathcal { F } } { \\operatorname* { s u p } } | \\widehat { f } _ { 1 } ( \\nu ) + \\widehat { f } _ { 2 } ( \\nu ) + \\widehat { f } _ { 3 } ( \\nu ) | \\Bigg | } & { } \\\\ { \\leq \\underset { \\nu \\in \\mathcal { F } } { \\operatorname* { s u p } } | | f _ { 1 } ( \\nu ) + f _ { 2 } ( \\nu ) + f _ { 3 } ( \\nu ) | - | \\widehat { f } _ { 1 } ( \\nu ) + \\widehat { f } _ { 2 } ( \\nu ) + \\widehat { f } _ { 3 } ( \\nu ) | | } & { } \\\\ { \\leq \\underset { \\nu \\in \\mathcal { F } } { \\operatorname* { s u p } } | f _ { 1 } ( \\nu ) + f _ { 2 } ( \\nu ) + f _ { 3 } ( \\nu ) - \\widehat { f } _ { 1 } ( \\nu ) - \\widehat { f } _ { 2 } ( \\nu ) - \\widehat { f } _ { 3 } ( \\nu ) | } & { } \\\\ { \\leq \\underset { \\nu \\in \\mathcal { F } } { \\operatorname* { s u p } } [ | f _ { 1 } ( \\nu ) - \\widehat { f } _ { 1 } ( \\nu ) | + | f _ { 2 } ( \\nu ) - \\widehat { f } _ { 2 } ( \\nu ) | + | f _ { 3 } ( \\nu ) - \\widehat { f } _ { 3 } ( \\nu ) | ] } & { } \\\\ { \\leq \\underset { \\nu \\in \\mathcal { F } } { \\operatorname* { s u p } } [ | f _ { 1 } ( \\nu ) - \\widehat { f } _ { 1 } ( \\nu ) | + | \\widehat { f } _ { 2 } ( \\nu ) - \\widehat { f } _ { 2 } ( \\nu ) | + | f _ { 3 } ( \\nu ) - \\widehat { f } _ { 3 } ( \\nu ) | ] } & { } \\\\  \\leq \\underset { \\nu \\in \\mathcal { F } } { \\operatorname* { s u p } } | f _ { 1 } ( \\nu ) - \\widehat { f } _ { 1 } ( \\nu ) | + \\underset { \\nu \\in \\mathcal { F } } { \\operatorname* { s u p } } | f _   \\end{array}",
+                                    "type": "interline_equation",
+                                    "image_path": "b0c68ff1a53c4effe6dff949d0bead0c77d624f5418e389ef1debb02c711ae45.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 2,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                111,
+                                97,
+                                513,
+                                140.66666666666666
+                            ],
+                            "spans": [],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                111,
+                                140.66666666666666,
+                                513,
+                                184.33333333333331
+                            ],
+                            "spans": [],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                111,
+                                184.33333333333331,
+                                513,
+                                227.99999999999997
+                            ],
+                            "spans": [],
+                            "index": 3
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        230,
+                        192,
+                        241
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                228,
+                                193,
+                                243
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        228,
+                                        193,
+                                        243
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We then observe that",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        }
+                    ],
+                    "index": 4
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        126,
+                        244,
+                        484,
+                        342
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                126,
+                                244,
+                                484,
+                                342
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        244,
+                                        484,
+                                        342
+                                    ],
+                                    "score": 0.95,
+                                    "content": "\\begin{array} { l } { \\displaystyle \\underset { \\nu \\in \\mathcal { F } } { \\operatorname* { s u p } } \\left| f _ { 1 } ( \\nu ) - \\widehat { f _ { 1 } } ( \\nu ) \\right| = ( 1 - \\gamma ) \\Biggl \\{ \\mathbb { E } _ { s \\sim d _ { 0 } , a \\sim \\pi ( s ) } [ k ( \\phi ( s , a ) , \\phi ( \\bar { s } , \\bar { a } ) ) ] } \\\\ { \\displaystyle - \\frac { 2 } { n } \\sum _ { i = 1 } ^ { n } \\mathbb { E } _ { \\bar { s } \\sim d _ { 0 } , \\bar { a } \\sim \\pi ( s ) } \\left[ k \\left( \\phi \\left( s ^ { ( i ) } , a ^ { ( i ) } \\right) , \\phi ( \\bar { s } , \\bar { a } ) \\right) \\right] } \\\\ { \\displaystyle + \\frac { 1 } { n ^ { 2 } } \\sum _ { i = 1 } ^ { n } \\sum _ { j = 1 } ^ { n } k \\left( \\phi \\left( s ^ { ( i ) } , a ^ { ( i ) } \\right) , \\phi \\left( s ^ { ( j ) } , a ^ { ( j ) } \\right) \\right) \\Biggl \\} ^ { 1 / 2 } , } \\end{array}",
+                                    "type": "interline_equation",
+                                    "image_path": "7baccde9ecc5c794c743a3d16fa77db89bf9fece95437c64ff8f77484699b903.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 6,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                126,
+                                244,
+                                484,
+                                276.6666666666667
+                            ],
+                            "spans": [],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                276.6666666666667,
+                                484,
+                                309.33333333333337
+                            ],
+                            "spans": [],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                309.33333333333337,
+                                484,
+                                342.00000000000006
+                            ],
+                            "spans": [],
+                            "index": 7
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        345,
+                        506,
+                        388
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                345,
+                                506,
+                                366
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        345,
+                                        222,
+                                        364
+                                    ],
+                                    "score": 1.0,
+                                    "content": "which shows that changing",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        222,
+                                        348,
+                                        256,
+                                        361
+                                    ],
+                                    "score": 0.94,
+                                    "content": "s ^ { ( i ) } , a ^ { ( i ) }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        256,
+                                        345,
+                                        348,
+                                        364
+                                    ],
+                                    "score": 1.0,
+                                    "content": "results in changes of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        348,
+                                        345,
+                                        447,
+                                        366
+                                    ],
+                                    "score": 0.94,
+                                    "content": "{ \\mathrm { s u p } } _ { \\nu \\in { \\mathcal { F } } } \\left| f _ { 1 } ( \\nu ) - { \\widehat { f } } _ { 1 } ( \\nu ) \\right|",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        447,
+                                        345,
+                                        506,
+                                        364
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in magnitude",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                362,
+                                505,
+                                380
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        362,
+                                        153,
+                                        380
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of at most",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        154,
+                                        364,
+                                        221,
+                                        378
+                                    ],
+                                    "score": 0.92,
+                                    "content": "2 ( 1 - \\gamma ) \\bar { k } ^ { 1 / 2 } / n",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        221,
+                                        362,
+                                        252,
+                                        380
+                                    ],
+                                    "score": 1.0,
+                                    "content": "where",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        252,
+                                        364,
+                                        337,
+                                        378
+                                    ],
+                                    "score": 0.92,
+                                    "content": "{ \\bar { k } } = \\operatorname* { s u p } _ { z \\in { \\mathcal { Z } } } k ( z , z )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        338,
+                                        362,
+                                        505,
+                                        380
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Therefore, by McDiarmid’s inequality",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                374,
+                                172,
+                                389
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        374,
+                                        172,
+                                        389
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(Theorem A.1),",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        }
+                    ],
+                    "index": 9
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        117,
+                        391,
+                        493,
+                        419
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                117,
+                                391,
+                                493,
+                                419
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        117,
+                                        391,
+                                        493,
+                                        419
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\operatorname* { P r } \\left\\{ \\operatorname* { s u p } _ { \\nu \\in \\mathcal { F } } \\left| f _ { 1 } ( \\nu ) - \\widehat f _ { 1 } ( \\nu ) \\right| - \\mathbb { E } _ { s ^ { ( i ) } , a ^ { ( i ) } } \\left[ \\operatorname* { s u p } _ { \\nu \\in \\mathcal { F } } \\left| f _ { 1 } ( \\nu ) - \\widehat f _ { 1 } ( \\nu ) \\right| \\right] \\geq \\epsilon \\right\\} \\leq \\exp \\left( - \\frac { n \\epsilon ^ { 2 } } { 2 ( 1 - \\gamma ) ^ { 2 } \\bar { k } } \\right) .",
+                                    "type": "interline_equation",
+                                    "image_path": "ff86b5f506316d6e41e4f1efcef98f23ccb6537b98b1135dfc3e9d6e477d1eb0.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 11,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                117,
+                                391,
+                                493,
+                                419
+                            ],
+                            "spans": [],
+                            "index": 11
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        423,
+                        129,
+                        434
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                421,
+                                132,
+                                437
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        421,
+                                        132,
+                                        437
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Also,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        }
+                    ],
+                    "index": 12
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        124,
+                        437,
+                        487,
+                        597
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                124,
+                                437,
+                                487,
+                                597
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        124,
+                                        437,
+                                        487,
+                                        597
+                                    ],
+                                    "score": 0.94,
+                                    "content": "\\begin{array} { r l } & { \\mathbb { E } _ { s ^ { ( i ) } , \\sigma ^ { ( k ) } } [ \\underset { s ^ { ( r , F ) } } { \\operatorname* { s u p } } [ f _ { 1 } ( \\nu ) - \\widehat { f } _ { 1 } ( \\nu ) ] ] } \\\\ & { = \\frac { 1 - \\gamma } { n } \\underset { s ^ { ( i ) } , a ^ { ( i ) } , b ^ { ( i ) } } { \\operatorname* { s u p } } [ \\underset { s ^ { ( r , F ) } } { \\operatorname* { s u p } } [ \\underset { s ^ { ( i ) } , a ^ { ( i ) } , b ^ { ( i ) } } { \\operatorname* { s u p } } [ \\underset { i = 1 } { \\overset { n } { \\sum } } \\nu ( \\phi ( \\tilde { s } _ { 0 } ^ { ( i ) } , \\bar { a } _ { 0 } ^ { ( i ) } ) ) ] - \\underset { i = 1 } { \\overset { n } { \\sum } } \\nu ( \\phi ( s _ { 0 } ^ { ( i ) } , a _ { 0 } ^ { ( i ) } ) ) ) ] } \\\\ & { \\leq \\frac { 1 - \\gamma } { n } \\underset { s ^ { ( i ) } , a ^ { ( i ) } , b ^ { ( i ) } , a ^ { ( i ) } , b ^ { ( i ) } } { \\operatorname* { s u p } } [ \\underset { s ^ { ( r , F ) } } { \\operatorname* { s u p } } [ \\underset { i = 1 } { \\overset { n } { \\sum } } \\nu ( \\phi ( \\frac { \\sigma ^ { ( i ) } } { s } , \\bar { a } _ { 0 } ^ { ( i ) } ) ) - \\underset { i = 1 } { \\overset { n } { \\sum } } \\nu ( \\phi ( s _ { 0 } ^ { ( i ) } , a _ { 0 } ^ { ( i ) } ) ) ) ] } \\\\ & { = \\frac { 1 - \\gamma } { n } \\underset { s ^ { ( i ) } , a ^ { ( i ) } , b ^ { ( i ) } , a ^ { ( i ) } , b ^ { ( i ) } } { \\operatorname* { s u p } } [ \\underset { s ^ { ( r , F ) } } { \\operatorname* { s u p } } [ \\underset { i = 1 } { \\overset { n } { \\sum } } \\sigma ^ { ( i ) 1 } \\{ \\nu ( \\phi ( \\tilde { s } _ { 0 } ^ { ( i ) } , \\bar { a } _ { 0 } ^ { ( i ) } ) ) - \\nu ( \\phi ( s _ { 0 } ^ { ( i ) } , a _ { 0 } ^ { ( i ) } ) ) \\} ] } \\\\ &  \\leq 2 ( 1 - \\gamma )  \\end{array}",
+                                    "type": "interline_equation",
+                                    "image_path": "2bd7abe170a7ba5c16bd7e0b5e3d05a75f74804940fc49c3b0a3aa28c8176a71.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 14,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                124,
+                                437,
+                                487,
+                                490.3333333333333
+                            ],
+                            "spans": [],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                124,
+                                490.3333333333333,
+                                487,
+                                543.6666666666666
+                            ],
+                            "spans": [],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                124,
+                                543.6666666666666,
+                                487,
+                                597.0
+                            ],
+                            "spans": [],
+                            "index": 15
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        598,
+                        413,
+                        611
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                596,
+                                413,
+                                614
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        596,
+                                        413,
+                                        614
+                                    ],
+                                    "score": 1.0,
+                                    "content": "where the last inequality is from Lemma A.2. Combining the results, we get",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        }
+                    ],
+                    "index": 16
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        154,
+                        613,
+                        456,
+                        648
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                154,
+                                613,
+                                456,
+                                648
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        154,
+                                        613,
+                                        456,
+                                        648
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\operatorname* { P r } \\left\\{ \\operatorname* { s u p } _ { \\nu \\in \\mathcal { F } } \\left| f _ { 1 } ( \\nu ) - \\widehat { f } _ { 1 } ( \\nu ) \\right| - 2 ( 1 - \\gamma ) \\sqrt { \\frac { \\bar { k } } { n } } \\geq \\epsilon \\right\\} \\leq \\exp \\left( - \\frac { n \\epsilon ^ { 2 } } { 2 ( 1 - \\gamma ) ^ { 2 } \\bar { k } } \\right) .",
+                                    "type": "interline_equation",
+                                    "image_path": "7f260310c1dac7b688d86d84bfd912dda7131f603013ed051c33a3a5dc9d6360.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 18,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                154,
+                                613,
+                                456,
+                                624.6666666666666
+                            ],
+                            "spans": [],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                154,
+                                624.6666666666666,
+                                456,
+                                636.3333333333333
+                            ],
+                            "spans": [],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                154,
+                                636.3333333333333,
+                                456,
+                                647.9999999999999
+                            ],
+                            "spans": [],
+                            "index": 19
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        650,
+                        327,
+                        663
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                648,
+                                328,
+                                665
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        648,
+                                        235,
+                                        665
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Similarly, we derive bounds for",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        235,
+                                        651,
+                                        245,
+                                        662
+                                    ],
+                                    "score": 0.89,
+                                    "content": "f _ { 2 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        245,
+                                        648,
+                                        263,
+                                        665
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        263,
+                                        651,
+                                        273,
+                                        662
+                                    ],
+                                    "score": 0.88,
+                                    "content": "f _ { 3 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        274,
+                                        648,
+                                        328,
+                                        665
+                                    ],
+                                    "score": 1.0,
+                                    "content": "respectively:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        }
+                    ],
+                    "index": 20
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        179,
+                        666,
+                        432,
+                        736
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                179,
+                                666,
+                                432,
+                                736
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        179,
+                                        666,
+                                        432,
+                                        736
+                                    ],
+                                    "score": 0.78,
+                                    "content": "\\begin{array} { l l } { { \\displaystyle \\operatorname* { P r } \\left\\{ \\operatorname* { s u p } _ { \\nu \\in \\mathcal { F } } \\Big | f _ { 2 } ( \\nu ) - \\widehat { f } _ { 2 } ( \\nu ) \\Big | - 2 \\sqrt \\frac { \\bar { k } } { n } \\geq \\epsilon \\right\\} \\leq \\exp \\left( - \\frac { n \\epsilon ^ { 2 } } { 2 \\bar { k } } \\right) , } } & { { } } \\\\ { { \\displaystyle \\operatorname* { P r } \\left\\{ \\operatorname* { s u p } _ { \\nu \\in \\mathcal { F } } \\Big | f _ { 3 } ( \\nu ) - \\widehat { f } _ { 3 } ( \\nu ) \\Big | - 2 \\gamma \\sqrt \\frac { \\bar { k } } { n } \\geq \\epsilon \\right\\} \\leq \\exp \\left( - \\frac { n \\epsilon ^ { 2 } } { 2 \\gamma ^ { 2 } \\bar { k } } \\right) . } } \\end{array}",
+                                    "type": "interline_equation",
+                                    "image_path": "09fb04b83ee29d40bceeaf191a8532c8bdaa6bd877c408d43fa97c37a0c896ab.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 22,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                179,
+                                666,
+                                432,
+                                689.3333333333334
+                            ],
+                            "spans": [],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                179,
+                                689.3333333333334,
+                                432,
+                                712.6666666666667
+                            ],
+                            "spans": [],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                179,
+                                712.6666666666667,
+                                432,
+                                736.0000000000001
+                            ],
+                            "spans": [],
+                            "index": 23
+                        }
+                    ]
+                }
+            ],
+            "page_idx": 15,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        107,
+                        26,
+                        293,
+                        38
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                25,
+                                293,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        25,
+                                        293,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Published as a conference paper at ICLR 2021",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        300,
+                        751,
+                        311,
+                        761
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                299,
+                                750,
+                                313,
+                                764
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        299,
+                                        750,
+                                        313,
+                                        764
+                                    ],
+                                    "score": 1.0,
+                                    "content": "16",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        82,
+                        411,
+                        94
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                82,
+                                412,
+                                95
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        82,
+                                        412,
+                                        95
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We can bound the error of sample-based estimator with individual errors as:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        }
+                    ],
+                    "index": 0,
+                    "bbox_fs": [
+                        106,
+                        82,
+                        412,
+                        95
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        111,
+                        97,
+                        513,
+                        228
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                111,
+                                97,
+                                513,
+                                228
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        111,
+                                        97,
+                                        513,
+                                        228
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\begin{array} { r l } { \\mathrm { I P M } _ { \\mathcal { G } } ( d _ { \\phi } ^ { \\pi } , d _ { \\phi } ^ { 2 } ) - \\widehat { \\mathrm { I P M } } ( d _ { \\phi } ^ { \\pi } , d _ { \\phi } ^ { \\mathcal { D } } ) \\Bigg | = \\Bigg | \\underset { \\nu \\in \\mathcal { F } } { \\operatorname* { s u p } } | f _ { 1 } ( \\nu ) + f _ { 2 } ( \\nu ) + f _ { 3 } ( \\nu ) | - \\underset { \\nu \\in \\mathcal { F } } { \\operatorname* { s u p } } | \\widehat { f } _ { 1 } ( \\nu ) + \\widehat { f } _ { 2 } ( \\nu ) + \\widehat { f } _ { 3 } ( \\nu ) | \\Bigg | } & { } \\\\ { \\leq \\underset { \\nu \\in \\mathcal { F } } { \\operatorname* { s u p } } | | f _ { 1 } ( \\nu ) + f _ { 2 } ( \\nu ) + f _ { 3 } ( \\nu ) | - | \\widehat { f } _ { 1 } ( \\nu ) + \\widehat { f } _ { 2 } ( \\nu ) + \\widehat { f } _ { 3 } ( \\nu ) | | } & { } \\\\ { \\leq \\underset { \\nu \\in \\mathcal { F } } { \\operatorname* { s u p } } | f _ { 1 } ( \\nu ) + f _ { 2 } ( \\nu ) + f _ { 3 } ( \\nu ) - \\widehat { f } _ { 1 } ( \\nu ) - \\widehat { f } _ { 2 } ( \\nu ) - \\widehat { f } _ { 3 } ( \\nu ) | } & { } \\\\ { \\leq \\underset { \\nu \\in \\mathcal { F } } { \\operatorname* { s u p } } [ | f _ { 1 } ( \\nu ) - \\widehat { f } _ { 1 } ( \\nu ) | + | f _ { 2 } ( \\nu ) - \\widehat { f } _ { 2 } ( \\nu ) | + | f _ { 3 } ( \\nu ) - \\widehat { f } _ { 3 } ( \\nu ) | ] } & { } \\\\ { \\leq \\underset { \\nu \\in \\mathcal { F } } { \\operatorname* { s u p } } [ | f _ { 1 } ( \\nu ) - \\widehat { f } _ { 1 } ( \\nu ) | + | \\widehat { f } _ { 2 } ( \\nu ) - \\widehat { f } _ { 2 } ( \\nu ) | + | f _ { 3 } ( \\nu ) - \\widehat { f } _ { 3 } ( \\nu ) | ] } & { } \\\\  \\leq \\underset { \\nu \\in \\mathcal { F } } { \\operatorname* { s u p } } | f _ { 1 } ( \\nu ) - \\widehat { f } _ { 1 } ( \\nu ) | + \\underset { \\nu \\in \\mathcal { F } } { \\operatorname* { s u p } } | f _   \\end{array}",
+                                    "type": "interline_equation",
+                                    "image_path": "b0c68ff1a53c4effe6dff949d0bead0c77d624f5418e389ef1debb02c711ae45.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 2,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                111,
+                                97,
+                                513,
+                                140.66666666666666
+                            ],
+                            "spans": [],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                111,
+                                140.66666666666666,
+                                513,
+                                184.33333333333331
+                            ],
+                            "spans": [],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                111,
+                                184.33333333333331,
+                                513,
+                                227.99999999999997
+                            ],
+                            "spans": [],
+                            "index": 3
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        230,
+                        192,
+                        241
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                228,
+                                193,
+                                243
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        228,
+                                        193,
+                                        243
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We then observe that",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        }
+                    ],
+                    "index": 4,
+                    "bbox_fs": [
+                        106,
+                        228,
+                        193,
+                        243
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        126,
+                        244,
+                        484,
+                        342
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                126,
+                                244,
+                                484,
+                                342
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        244,
+                                        484,
+                                        342
+                                    ],
+                                    "score": 0.95,
+                                    "content": "\\begin{array} { l } { \\displaystyle \\underset { \\nu \\in \\mathcal { F } } { \\operatorname* { s u p } } \\left| f _ { 1 } ( \\nu ) - \\widehat { f _ { 1 } } ( \\nu ) \\right| = ( 1 - \\gamma ) \\Biggl \\{ \\mathbb { E } _ { s \\sim d _ { 0 } , a \\sim \\pi ( s ) } [ k ( \\phi ( s , a ) , \\phi ( \\bar { s } , \\bar { a } ) ) ] } \\\\ { \\displaystyle - \\frac { 2 } { n } \\sum _ { i = 1 } ^ { n } \\mathbb { E } _ { \\bar { s } \\sim d _ { 0 } , \\bar { a } \\sim \\pi ( s ) } \\left[ k \\left( \\phi \\left( s ^ { ( i ) } , a ^ { ( i ) } \\right) , \\phi ( \\bar { s } , \\bar { a } ) \\right) \\right] } \\\\ { \\displaystyle + \\frac { 1 } { n ^ { 2 } } \\sum _ { i = 1 } ^ { n } \\sum _ { j = 1 } ^ { n } k \\left( \\phi \\left( s ^ { ( i ) } , a ^ { ( i ) } \\right) , \\phi \\left( s ^ { ( j ) } , a ^ { ( j ) } \\right) \\right) \\Biggl \\} ^ { 1 / 2 } , } \\end{array}",
+                                    "type": "interline_equation",
+                                    "image_path": "7baccde9ecc5c794c743a3d16fa77db89bf9fece95437c64ff8f77484699b903.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 6,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                126,
+                                244,
+                                484,
+                                276.6666666666667
+                            ],
+                            "spans": [],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                276.6666666666667,
+                                484,
+                                309.33333333333337
+                            ],
+                            "spans": [],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                309.33333333333337,
+                                484,
+                                342.00000000000006
+                            ],
+                            "spans": [],
+                            "index": 7
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        345,
+                        506,
+                        388
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                345,
+                                506,
+                                366
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        345,
+                                        222,
+                                        364
+                                    ],
+                                    "score": 1.0,
+                                    "content": "which shows that changing",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        222,
+                                        348,
+                                        256,
+                                        361
+                                    ],
+                                    "score": 0.94,
+                                    "content": "s ^ { ( i ) } , a ^ { ( i ) }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        256,
+                                        345,
+                                        348,
+                                        364
+                                    ],
+                                    "score": 1.0,
+                                    "content": "results in changes of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        348,
+                                        345,
+                                        447,
+                                        366
+                                    ],
+                                    "score": 0.94,
+                                    "content": "{ \\mathrm { s u p } } _ { \\nu \\in { \\mathcal { F } } } \\left| f _ { 1 } ( \\nu ) - { \\widehat { f } } _ { 1 } ( \\nu ) \\right|",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        447,
+                                        345,
+                                        506,
+                                        364
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in magnitude",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                362,
+                                505,
+                                380
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        362,
+                                        153,
+                                        380
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of at most",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        154,
+                                        364,
+                                        221,
+                                        378
+                                    ],
+                                    "score": 0.92,
+                                    "content": "2 ( 1 - \\gamma ) \\bar { k } ^ { 1 / 2 } / n",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        221,
+                                        362,
+                                        252,
+                                        380
+                                    ],
+                                    "score": 1.0,
+                                    "content": "where",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        252,
+                                        364,
+                                        337,
+                                        378
+                                    ],
+                                    "score": 0.92,
+                                    "content": "{ \\bar { k } } = \\operatorname* { s u p } _ { z \\in { \\mathcal { Z } } } k ( z , z )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        338,
+                                        362,
+                                        505,
+                                        380
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Therefore, by McDiarmid’s inequality",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                374,
+                                172,
+                                389
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        374,
+                                        172,
+                                        389
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(Theorem A.1),",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        }
+                    ],
+                    "index": 9,
+                    "bbox_fs": [
+                        104,
+                        345,
+                        506,
+                        389
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        117,
+                        391,
+                        493,
+                        419
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                117,
+                                391,
+                                493,
+                                419
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        117,
+                                        391,
+                                        493,
+                                        419
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\operatorname* { P r } \\left\\{ \\operatorname* { s u p } _ { \\nu \\in \\mathcal { F } } \\left| f _ { 1 } ( \\nu ) - \\widehat f _ { 1 } ( \\nu ) \\right| - \\mathbb { E } _ { s ^ { ( i ) } , a ^ { ( i ) } } \\left[ \\operatorname* { s u p } _ { \\nu \\in \\mathcal { F } } \\left| f _ { 1 } ( \\nu ) - \\widehat f _ { 1 } ( \\nu ) \\right| \\right] \\geq \\epsilon \\right\\} \\leq \\exp \\left( - \\frac { n \\epsilon ^ { 2 } } { 2 ( 1 - \\gamma ) ^ { 2 } \\bar { k } } \\right) .",
+                                    "type": "interline_equation",
+                                    "image_path": "ff86b5f506316d6e41e4f1efcef98f23ccb6537b98b1135dfc3e9d6e477d1eb0.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 11,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                117,
+                                391,
+                                493,
+                                419
+                            ],
+                            "spans": [],
+                            "index": 11
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        423,
+                        129,
+                        434
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                421,
+                                132,
+                                437
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        421,
+                                        132,
+                                        437
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Also,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        }
+                    ],
+                    "index": 12,
+                    "bbox_fs": [
+                        105,
+                        421,
+                        132,
+                        437
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        124,
+                        437,
+                        487,
+                        597
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                124,
+                                437,
+                                487,
+                                597
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        124,
+                                        437,
+                                        487,
+                                        597
+                                    ],
+                                    "score": 0.94,
+                                    "content": "\\begin{array} { r l } & { \\mathbb { E } _ { s ^ { ( i ) } , \\sigma ^ { ( k ) } } [ \\underset { s ^ { ( r , F ) } } { \\operatorname* { s u p } } [ f _ { 1 } ( \\nu ) - \\widehat { f } _ { 1 } ( \\nu ) ] ] } \\\\ & { = \\frac { 1 - \\gamma } { n } \\underset { s ^ { ( i ) } , a ^ { ( i ) } , b ^ { ( i ) } } { \\operatorname* { s u p } } [ \\underset { s ^ { ( r , F ) } } { \\operatorname* { s u p } } [ \\underset { s ^ { ( i ) } , a ^ { ( i ) } , b ^ { ( i ) } } { \\operatorname* { s u p } } [ \\underset { i = 1 } { \\overset { n } { \\sum } } \\nu ( \\phi ( \\tilde { s } _ { 0 } ^ { ( i ) } , \\bar { a } _ { 0 } ^ { ( i ) } ) ) ] - \\underset { i = 1 } { \\overset { n } { \\sum } } \\nu ( \\phi ( s _ { 0 } ^ { ( i ) } , a _ { 0 } ^ { ( i ) } ) ) ) ] } \\\\ & { \\leq \\frac { 1 - \\gamma } { n } \\underset { s ^ { ( i ) } , a ^ { ( i ) } , b ^ { ( i ) } , a ^ { ( i ) } , b ^ { ( i ) } } { \\operatorname* { s u p } } [ \\underset { s ^ { ( r , F ) } } { \\operatorname* { s u p } } [ \\underset { i = 1 } { \\overset { n } { \\sum } } \\nu ( \\phi ( \\frac { \\sigma ^ { ( i ) } } { s } , \\bar { a } _ { 0 } ^ { ( i ) } ) ) - \\underset { i = 1 } { \\overset { n } { \\sum } } \\nu ( \\phi ( s _ { 0 } ^ { ( i ) } , a _ { 0 } ^ { ( i ) } ) ) ) ] } \\\\ & { = \\frac { 1 - \\gamma } { n } \\underset { s ^ { ( i ) } , a ^ { ( i ) } , b ^ { ( i ) } , a ^ { ( i ) } , b ^ { ( i ) } } { \\operatorname* { s u p } } [ \\underset { s ^ { ( r , F ) } } { \\operatorname* { s u p } } [ \\underset { i = 1 } { \\overset { n } { \\sum } } \\sigma ^ { ( i ) 1 } \\{ \\nu ( \\phi ( \\tilde { s } _ { 0 } ^ { ( i ) } , \\bar { a } _ { 0 } ^ { ( i ) } ) ) - \\nu ( \\phi ( s _ { 0 } ^ { ( i ) } , a _ { 0 } ^ { ( i ) } ) ) \\} ] } \\\\ &  \\leq 2 ( 1 - \\gamma )  \\end{array}",
+                                    "type": "interline_equation",
+                                    "image_path": "2bd7abe170a7ba5c16bd7e0b5e3d05a75f74804940fc49c3b0a3aa28c8176a71.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 14,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                124,
+                                437,
+                                487,
+                                490.3333333333333
+                            ],
+                            "spans": [],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                124,
+                                490.3333333333333,
+                                487,
+                                543.6666666666666
+                            ],
+                            "spans": [],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                124,
+                                543.6666666666666,
+                                487,
+                                597.0
+                            ],
+                            "spans": [],
+                            "index": 15
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        598,
+                        413,
+                        611
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                596,
+                                413,
+                                614
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        596,
+                                        413,
+                                        614
+                                    ],
+                                    "score": 1.0,
+                                    "content": "where the last inequality is from Lemma A.2. Combining the results, we get",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        }
+                    ],
+                    "index": 16,
+                    "bbox_fs": [
+                        105,
+                        596,
+                        413,
+                        614
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        154,
+                        613,
+                        456,
+                        648
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                154,
+                                613,
+                                456,
+                                648
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        154,
+                                        613,
+                                        456,
+                                        648
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\operatorname* { P r } \\left\\{ \\operatorname* { s u p } _ { \\nu \\in \\mathcal { F } } \\left| f _ { 1 } ( \\nu ) - \\widehat { f } _ { 1 } ( \\nu ) \\right| - 2 ( 1 - \\gamma ) \\sqrt { \\frac { \\bar { k } } { n } } \\geq \\epsilon \\right\\} \\leq \\exp \\left( - \\frac { n \\epsilon ^ { 2 } } { 2 ( 1 - \\gamma ) ^ { 2 } \\bar { k } } \\right) .",
+                                    "type": "interline_equation",
+                                    "image_path": "7f260310c1dac7b688d86d84bfd912dda7131f603013ed051c33a3a5dc9d6360.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 18,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                154,
+                                613,
+                                456,
+                                624.6666666666666
+                            ],
+                            "spans": [],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                154,
+                                624.6666666666666,
+                                456,
+                                636.3333333333333
+                            ],
+                            "spans": [],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                154,
+                                636.3333333333333,
+                                456,
+                                647.9999999999999
+                            ],
+                            "spans": [],
+                            "index": 19
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        650,
+                        327,
+                        663
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                648,
+                                328,
+                                665
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        648,
+                                        235,
+                                        665
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Similarly, we derive bounds for",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        235,
+                                        651,
+                                        245,
+                                        662
+                                    ],
+                                    "score": 0.89,
+                                    "content": "f _ { 2 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        245,
+                                        648,
+                                        263,
+                                        665
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        263,
+                                        651,
+                                        273,
+                                        662
+                                    ],
+                                    "score": 0.88,
+                                    "content": "f _ { 3 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        274,
+                                        648,
+                                        328,
+                                        665
+                                    ],
+                                    "score": 1.0,
+                                    "content": "respectively:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        }
+                    ],
+                    "index": 20,
+                    "bbox_fs": [
+                        105,
+                        648,
+                        328,
+                        665
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        179,
+                        666,
+                        432,
+                        736
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                179,
+                                666,
+                                432,
+                                736
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        179,
+                                        666,
+                                        432,
+                                        736
+                                    ],
+                                    "score": 0.78,
+                                    "content": "\\begin{array} { l l } { { \\displaystyle \\operatorname* { P r } \\left\\{ \\operatorname* { s u p } _ { \\nu \\in \\mathcal { F } } \\Big | f _ { 2 } ( \\nu ) - \\widehat { f } _ { 2 } ( \\nu ) \\Big | - 2 \\sqrt \\frac { \\bar { k } } { n } \\geq \\epsilon \\right\\} \\leq \\exp \\left( - \\frac { n \\epsilon ^ { 2 } } { 2 \\bar { k } } \\right) , } } & { { } } \\\\ { { \\displaystyle \\operatorname* { P r } \\left\\{ \\operatorname* { s u p } _ { \\nu \\in \\mathcal { F } } \\Big | f _ { 3 } ( \\nu ) - \\widehat { f } _ { 3 } ( \\nu ) \\Big | - 2 \\gamma \\sqrt \\frac { \\bar { k } } { n } \\geq \\epsilon \\right\\} \\leq \\exp \\left( - \\frac { n \\epsilon ^ { 2 } } { 2 \\gamma ^ { 2 } \\bar { k } } \\right) . } } \\end{array}",
+                                    "type": "interline_equation",
+                                    "image_path": "09fb04b83ee29d40bceeaf191a8532c8bdaa6bd877c408d43fa97c37a0c896ab.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 22,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                179,
+                                666,
+                                432,
+                                689.3333333333334
+                            ],
+                            "spans": [],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                179,
+                                689.3333333333334,
+                                432,
+                                712.6666666666667
+                            ],
+                            "spans": [],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                179,
+                                712.6666666666667,
+                                432,
+                                736.0000000000001
+                            ],
+                            "spans": [],
+                            "index": 23
+                        }
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        104,
+                        81,
+                        505,
+                        104
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                80,
+                                505,
+                                97
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        80,
+                                        261,
+                                        97
+                                    ],
+                                    "score": 1.0,
+                                    "content": "By letting RHS of above bounds to be",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        261,
+                                        82,
+                                        277,
+                                        95
+                                    ],
+                                    "score": 0.87,
+                                    "content": "\\delta / 3",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        277,
+                                        80,
+                                        473,
+                                        97
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and using union bound, we get, with probability",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        473,
+                                        83,
+                                        487,
+                                        93
+                                    ],
+                                    "score": 0.36,
+                                    "content": "1 { - } \\delta",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        487,
+                                        80,
+                                        505,
+                                        97
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", we",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                93,
+                                123,
+                                107
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        93,
+                                        123,
+                                        107
+                                    ],
+                                    "score": 1.0,
+                                    "content": "get",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        }
+                    ],
+                    "index": 0.5
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        182,
+                        102,
+                        428,
+                        136
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                182,
+                                102,
+                                428,
+                                136
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        182,
+                                        102,
+                                        428,
+                                        136
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\left| \\mathrm { I P M } _ { \\mathcal { G } } ( d _ { \\phi } ^ { \\pi } , d _ { \\phi } ^ { D } ) - \\widehat { \\mathrm { I P M } } ( d _ { \\phi } ^ { \\pi } , d _ { \\phi } ^ { D } ) \\right| \\leq \\sqrt { \\frac { \\bar { k } } { n } } \\left( 4 + \\sqrt { 8 \\log \\frac { 3 } { \\delta } } \\right) .",
+                                    "type": "interline_equation",
+                                    "image_path": "7b40792537770ffb37a39911addd230c15068105f605b926770d665fa2554f28.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 3,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                182,
+                                102,
+                                428,
+                                113.33333333333333
+                            ],
+                            "spans": [],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                182,
+                                113.33333333333333,
+                                428,
+                                124.66666666666666
+                            ],
+                            "spans": [],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                182,
+                                124.66666666666666,
+                                428,
+                                136.0
+                            ],
+                            "spans": [],
+                            "index": 4
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        162,
+                        246,
+                        174
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                161,
+                                245,
+                                174
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        161,
+                                        210,
+                                        174
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The relationship between",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        210,
+                                        163,
+                                        219,
+                                        172
+                                    ],
+                                    "score": 0.85,
+                                    "content": "\\mathcal { F }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        220,
+                                        161,
+                                        237,
+                                        174
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        237,
+                                        163,
+                                        245,
+                                        173
+                                    ],
+                                    "score": 0.82,
+                                    "content": "\\mathcal { G }",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 5
+                        }
+                    ],
+                    "index": 5
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        111,
+                        177,
+                        505,
+                        210
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                111,
+                                177,
+                                505,
+                                210
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        111,
+                                        177,
+                                        505,
+                                        210
+                                    ],
+                                    "score": 0.89,
+                                    "content": "\\mathcal { J } = \\left\\{ g \\in ( \\mathcal { Z } \\to \\mathbb { R } ) : g ( z ) = \\nu ( z ) - \\gamma \\mathbb { E } _ { s ^ { \\prime } \\sim T ( \\phi ^ { - 1 } ( z ) ) } [ \\nu ( \\phi ( s ^ { \\prime } , a ^ { \\prime } ) ) ] , \\nu \\in ( \\mathcal { Z } \\to \\mathbb { R } ) , \\langle \\nu , \\nu \\rangle _ { \\mathcal { H } _ { k } } \\leq 1 \\right\\}",
+                                    "type": "interline_equation",
+                                    "image_path": "46d48b0cf72d25c32debc83d9775166ef3e7e3825d5a91191375f7d2e616b369.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 7,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                111,
+                                177,
+                                505,
+                                188.0
+                            ],
+                            "spans": [],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                111,
+                                188.0,
+                                505,
+                                199.0
+                            ],
+                            "spans": [],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                111,
+                                199.0,
+                                505,
+                                210.0
+                            ],
+                            "spans": [],
+                            "index": 8
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        215,
+                        503,
+                        238
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                213,
+                                506,
+                                231
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        213,
+                                        282,
+                                        231
+                                    ],
+                                    "score": 1.0,
+                                    "content": "show that when the conditional expectation",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        283,
+                                        215,
+                                        464,
+                                        228
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\mathbb { E } _ { s ^ { \\prime } \\sim T ( \\phi ^ { - 1 } ( \\cdot ) ) , a ^ { \\prime } \\sim \\pi ( s ^ { \\prime } ) } [ \\nu ( \\phi ( s ^ { \\prime } , a ^ { \\prime } ) ) ] : \\mathcal { Z } \\to \\mathbb { R }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        464,
+                                        213,
+                                        506,
+                                        231
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is a func-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                225,
+                                311,
+                                239
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        225,
+                                        163,
+                                        239
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tion in RKHS",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        164,
+                                        226,
+                                        178,
+                                        237
+                                    ],
+                                    "score": 0.79,
+                                    "content": "\\mathcal { H } _ { k }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        178,
+                                        225,
+                                        182,
+                                        239
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        182,
+                                        227,
+                                        190,
+                                        237
+                                    ],
+                                    "score": 0.72,
+                                    "content": "\\mathcal { G }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        190,
+                                        225,
+                                        292,
+                                        239
+                                    ],
+                                    "score": 1.0,
+                                    "content": "also becomes a subset of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        292,
+                                        227,
+                                        306,
+                                        237
+                                    ],
+                                    "score": 0.88,
+                                    "content": "\\mathcal { H } _ { k }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        307,
+                                        225,
+                                        311,
+                                        239
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        }
+                    ],
+                    "index": 9.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        243,
+                        280,
+                        255
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                243,
+                                281,
+                                256
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        243,
+                                        281,
+                                        256
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Then we can prove the following Theorem.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        }
+                    ],
+                    "index": 11
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        258,
+                        506,
+                        342
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                257,
+                                505,
+                                273
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        257,
+                                        249,
+                                        273
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Theorem 4.3. Given an MDP",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        249,
+                                        261,
+                                        261,
+                                        270
+                                    ],
+                                    "score": 0.66,
+                                    "content": "M",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        261,
+                                        257,
+                                        325,
+                                        273
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", its estimate",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        325,
+                                        258,
+                                        338,
+                                        271
+                                    ],
+                                    "score": 0.85,
+                                    "content": "\\widehat { M }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        339,
+                                        257,
+                                        505,
+                                        273
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with a bijective representation func-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                270,
+                                506,
+                                287
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        270,
+                                        128,
+                                        287
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tion",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        128,
+                                        273,
+                                        136,
+                                        285
+                                    ],
+                                    "score": 0.76,
+                                    "content": "\\phi",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        136,
+                                        270,
+                                        173,
+                                        287
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", i.e.",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        173,
+                                        271,
+                                        308,
+                                        285
+                                    ],
+                                    "score": 0.91,
+                                    "content": "( \\widehat { T } , \\widehat { R } ) = \\langle \\widehat { T } _ { z } \\circ \\phi , \\widehat { R } _ { z } \\circ \\phi \\rangle",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        309,
+                                        270,
+                                        387,
+                                        287
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", and an RKHS",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        387,
+                                        273,
+                                        487,
+                                        285
+                                    ],
+                                    "score": 0.59,
+                                    "content": "\\begin{array} { r } { \\mathcal { H } _ { k } \\quad \\subset \\quad ( \\mathcal { Z } \\quad \\to \\quad \\mathbb { R } ) } \\end{array}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        488,
+                                        270,
+                                        506,
+                                        287
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                282,
+                                504,
+                                299
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        282,
+                                        232,
+                                        298
+                                    ],
+                                    "score": 1.0,
+                                    "content": "duced by a universal kernel",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        232,
+                                        285,
+                                        240,
+                                        294
+                                    ],
+                                    "score": 0.73,
+                                    "content": "k",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        240,
+                                        282,
+                                        287,
+                                        298
+                                    ],
+                                    "score": 1.0,
+                                    "content": "such that",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        287,
+                                        284,
+                                        378,
+                                        297
+                                    ],
+                                    "score": 0.9,
+                                    "content": "\\mathrm { s u p } _ { z \\in \\mathcal { Z } } k ( z , z ) ~ = ~ \\bar { k }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        378,
+                                        282,
+                                        440,
+                                        298
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", assume that",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        441,
+                                        284,
+                                        504,
+                                        299
+                                    ],
+                                    "score": 0.89,
+                                    "content": "f _ { \\phi , \\widehat { R } _ { z } , \\widehat { T } _ { z } } \\left( \\widehat { z } \\right) \\ =",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                296,
+                                508,
+                                318
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        297,
+                                        344,
+                                        317
+                                    ],
+                                    "score": 0.9,
+                                    "content": "\\begin{array} { r } { \\mathbb { E } _ { T , \\pi } \\Big [ \\sum _ { t = 0 } ^ { \\infty } \\gamma ^ { t } \\mathcal { E } _ { \\phi , \\widehat { R } _ { z } , \\widehat { T } _ { z } } \\big ( s _ { t } , a _ { t } \\big ) \\Big | \\big ( s _ { 0 } , a _ { 0 } \\big ) = \\phi ^ { - 1 } \\big ( z \\big ) \\Big ] \\in \\mathcal { H } _ { k } } \\end{array}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        344,
+                                        296,
+                                        366,
+                                        318
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        367,
+                                        301,
+                                        451,
+                                        316
+                                    ],
+                                    "score": 0.93,
+                                    "content": "B _ { \\phi } = \\| f _ { \\phi , \\widehat { R } _ { z } , \\widehat { T } _ { z } } \\| _ { \\mathcal { H } _ { k } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        451,
+                                        296,
+                                        508,
+                                        318
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and the loss",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                103,
+                                312,
+                                506,
+                                334
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        103,
+                                        312,
+                                        165,
+                                        334
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is bounded by",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        165,
+                                        317,
+                                        293,
+                                        331
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\begin{array} { r } { \\bar { \\mathcal { E } } = \\operatorname* { s u p } _ { s \\in { \\mathcal { S } } , a \\in \\mathcal { A } } \\mathcal { E } _ { \\phi , \\widehat { R } _ { z } , \\widehat { T } _ { z } } ( s , a ) } \\end{array}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        293,
+                                        312,
+                                        423,
+                                        334
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Let n be the number of data in",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        424,
+                                        317,
+                                        432,
+                                        327
+                                    ],
+                                    "score": 0.78,
+                                    "content": "\\mathcal { D }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        433,
+                                        312,
+                                        506,
+                                        334
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". With probability",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                327,
+                                140,
+                                343
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        329,
+                                        135,
+                                        340
+                                    ],
+                                    "score": 0.64,
+                                    "content": "1 - 2 \\delta",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        135,
+                                        327,
+                                        140,
+                                        343
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        }
+                    ],
+                    "index": 14.5
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        111,
+                        345,
+                        506,
+                        381
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                111,
+                                345,
+                                506,
+                                381
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        111,
+                                        345,
+                                        506,
+                                        381
+                                    ],
+                                    "score": 0.89,
+                                    "content": "R ^ { \\pi } - { \\widehat R } ^ { \\pi } \\Big | \\leq \\frac { 1 } { n } \\sum _ { ( s , a ) \\in { \\mathcal D } } { \\mathcal E } _ { \\phi , { \\widehat R } _ { z } , { \\widehat T } _ { z } } ( s , a ) + B _ { \\phi } \\widehat \\Pi \\widehat M \\big ( d _ { \\phi } ^ { \\pi } , d _ { \\phi } ^ { { D } } \\big ) + \\sqrt { \\frac { \\bar { \\mathcal E } ^ { 2 } } { 2 n } \\log { \\frac { 1 } { \\delta } } } + B _ { \\phi } \\sqrt { \\frac { \\bar { k } } { n } } \\left( 4 + \\sqrt { 8 \\log { \\frac { 3 } { \\delta } } } \\right) .",
+                                    "type": "interline_equation",
+                                    "image_path": "38315fc6b9bb6240f5da5e207809b2ab70cd1e301d17089118a70400757cb45d.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 19,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                111,
+                                345,
+                                506,
+                                357.0
+                            ],
+                            "spans": [],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                111,
+                                357.0,
+                                506,
+                                369.0
+                            ],
+                            "spans": [],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                111,
+                                369.0,
+                                506,
+                                381.0
+                            ],
+                            "spans": [],
+                            "index": 20
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        391,
+                        479,
+                        405
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                391,
+                                479,
+                                406
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        391,
+                                        420,
+                                        406
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Proof. Applying the Hoeffding inequality (Hoeffding, 1963), with probability",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        421,
+                                        393,
+                                        443,
+                                        403
+                                    ],
+                                    "score": 0.84,
+                                    "content": "1 - \\delta",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        444,
+                                        391,
+                                        479,
+                                        406
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", we get:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        }
+                    ],
+                    "index": 21
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        163,
+                        409,
+                        447,
+                        444
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                163,
+                                409,
+                                447,
+                                444
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        163,
+                                        409,
+                                        447,
+                                        444
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\mathbb { E } _ { ( s , a ) \\sim d ^ { \\mathcal { D } } } \\left[ \\mathcal { E } _ { \\phi , \\widehat { R } _ { z } , \\widehat { T } _ { z } } ( s , a ) \\right] \\leq \\frac { 1 } { n } \\sum _ { ( s , a ) \\sim \\mathcal { D } } \\mathcal { E } _ { \\phi , \\widehat { R } _ { z } , \\widehat { T } _ { z } } ( s , a ) + \\sqrt { \\frac { \\bar { \\mathcal { E } } ^ { 2 } } { 2 n } \\log \\frac { 1 } { \\delta } } .",
+                                    "type": "interline_equation",
+                                    "image_path": "a143e9f568260a599d52a8c6d831e1d3e60ae1e101d09e4e264932b139e1cebe.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 23,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                163,
+                                409,
+                                447,
+                                420.6666666666667
+                            ],
+                            "spans": [],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                163,
+                                420.6666666666667,
+                                447,
+                                432.33333333333337
+                            ],
+                            "spans": [],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                163,
+                                432.33333333333337,
+                                447,
+                                444.00000000000006
+                            ],
+                            "spans": [],
+                            "index": 24
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        104,
+                        449,
+                        486,
+                        462
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                448,
+                                486,
+                                463
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        448,
+                                        486,
+                                        463
+                                    ],
+                                    "score": 1.0,
+                                    "content": "By using an union bound with Eq. (13) and substituting terms in Eq. (3), we recover the result.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        }
+                    ],
+                    "index": 25
+                }
+            ],
+            "page_idx": 16,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        107,
+                        26,
+                        293,
+                        38
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                25,
+                                293,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        25,
+                                        293,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Published as a conference paper at ICLR 2021",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        300,
+                        751,
+                        311,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                299,
+                                750,
+                                313,
+                                764
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        299,
+                                        750,
+                                        313,
+                                        764
+                                    ],
+                                    "score": 1.0,
+                                    "content": "17",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        494,
+                        138,
+                        505,
+                        149
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                495,
+                                140,
+                                505,
+                                150
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        495,
+                                        140,
+                                        505,
+                                        150
+                                    ],
+                                    "score": 0.999,
+                                    "content": "□",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        495,
+                        450,
+                        505,
+                        460
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                496,
+                                451,
+                                504,
+                                461
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        496,
+                                        451,
+                                        504,
+                                        461
+                                    ],
+                                    "score": 0.996,
+                                    "content": "□",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        104,
+                        81,
+                        505,
+                        104
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                80,
+                                505,
+                                97
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        80,
+                                        261,
+                                        97
+                                    ],
+                                    "score": 1.0,
+                                    "content": "By letting RHS of above bounds to be",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        261,
+                                        82,
+                                        277,
+                                        95
+                                    ],
+                                    "score": 0.87,
+                                    "content": "\\delta / 3",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        277,
+                                        80,
+                                        473,
+                                        97
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and using union bound, we get, with probability",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        473,
+                                        83,
+                                        487,
+                                        93
+                                    ],
+                                    "score": 0.36,
+                                    "content": "1 { - } \\delta",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        487,
+                                        80,
+                                        505,
+                                        97
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", we",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                93,
+                                123,
+                                107
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        93,
+                                        123,
+                                        107
+                                    ],
+                                    "score": 1.0,
+                                    "content": "get",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        }
+                    ],
+                    "index": 0.5,
+                    "bbox_fs": [
+                        105,
+                        80,
+                        505,
+                        107
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        182,
+                        102,
+                        428,
+                        136
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                182,
+                                102,
+                                428,
+                                136
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        182,
+                                        102,
+                                        428,
+                                        136
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\left| \\mathrm { I P M } _ { \\mathcal { G } } ( d _ { \\phi } ^ { \\pi } , d _ { \\phi } ^ { D } ) - \\widehat { \\mathrm { I P M } } ( d _ { \\phi } ^ { \\pi } , d _ { \\phi } ^ { D } ) \\right| \\leq \\sqrt { \\frac { \\bar { k } } { n } } \\left( 4 + \\sqrt { 8 \\log \\frac { 3 } { \\delta } } \\right) .",
+                                    "type": "interline_equation",
+                                    "image_path": "7b40792537770ffb37a39911addd230c15068105f605b926770d665fa2554f28.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 3,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                182,
+                                102,
+                                428,
+                                113.33333333333333
+                            ],
+                            "spans": [],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                182,
+                                113.33333333333333,
+                                428,
+                                124.66666666666666
+                            ],
+                            "spans": [],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                182,
+                                124.66666666666666,
+                                428,
+                                136.0
+                            ],
+                            "spans": [],
+                            "index": 4
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        162,
+                        246,
+                        174
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                161,
+                                245,
+                                174
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        161,
+                                        210,
+                                        174
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The relationship between",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        210,
+                                        163,
+                                        219,
+                                        172
+                                    ],
+                                    "score": 0.85,
+                                    "content": "\\mathcal { F }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        220,
+                                        161,
+                                        237,
+                                        174
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        237,
+                                        163,
+                                        245,
+                                        173
+                                    ],
+                                    "score": 0.82,
+                                    "content": "\\mathcal { G }",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 5
+                        }
+                    ],
+                    "index": 5,
+                    "bbox_fs": [
+                        106,
+                        161,
+                        245,
+                        174
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        111,
+                        177,
+                        505,
+                        210
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                111,
+                                177,
+                                505,
+                                210
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        111,
+                                        177,
+                                        505,
+                                        210
+                                    ],
+                                    "score": 0.89,
+                                    "content": "\\mathcal { J } = \\left\\{ g \\in ( \\mathcal { Z } \\to \\mathbb { R } ) : g ( z ) = \\nu ( z ) - \\gamma \\mathbb { E } _ { s ^ { \\prime } \\sim T ( \\phi ^ { - 1 } ( z ) ) } [ \\nu ( \\phi ( s ^ { \\prime } , a ^ { \\prime } ) ) ] , \\nu \\in ( \\mathcal { Z } \\to \\mathbb { R } ) , \\langle \\nu , \\nu \\rangle _ { \\mathcal { H } _ { k } } \\leq 1 \\right\\}",
+                                    "type": "interline_equation",
+                                    "image_path": "46d48b0cf72d25c32debc83d9775166ef3e7e3825d5a91191375f7d2e616b369.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 7,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                111,
+                                177,
+                                505,
+                                188.0
+                            ],
+                            "spans": [],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                111,
+                                188.0,
+                                505,
+                                199.0
+                            ],
+                            "spans": [],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                111,
+                                199.0,
+                                505,
+                                210.0
+                            ],
+                            "spans": [],
+                            "index": 8
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        215,
+                        503,
+                        238
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                213,
+                                506,
+                                231
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        213,
+                                        282,
+                                        231
+                                    ],
+                                    "score": 1.0,
+                                    "content": "show that when the conditional expectation",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        283,
+                                        215,
+                                        464,
+                                        228
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\mathbb { E } _ { s ^ { \\prime } \\sim T ( \\phi ^ { - 1 } ( \\cdot ) ) , a ^ { \\prime } \\sim \\pi ( s ^ { \\prime } ) } [ \\nu ( \\phi ( s ^ { \\prime } , a ^ { \\prime } ) ) ] : \\mathcal { Z } \\to \\mathbb { R }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        464,
+                                        213,
+                                        506,
+                                        231
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is a func-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                225,
+                                311,
+                                239
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        225,
+                                        163,
+                                        239
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tion in RKHS",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        164,
+                                        226,
+                                        178,
+                                        237
+                                    ],
+                                    "score": 0.79,
+                                    "content": "\\mathcal { H } _ { k }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        178,
+                                        225,
+                                        182,
+                                        239
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        182,
+                                        227,
+                                        190,
+                                        237
+                                    ],
+                                    "score": 0.72,
+                                    "content": "\\mathcal { G }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        190,
+                                        225,
+                                        292,
+                                        239
+                                    ],
+                                    "score": 1.0,
+                                    "content": "also becomes a subset of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        292,
+                                        227,
+                                        306,
+                                        237
+                                    ],
+                                    "score": 0.88,
+                                    "content": "\\mathcal { H } _ { k }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        307,
+                                        225,
+                                        311,
+                                        239
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        }
+                    ],
+                    "index": 9.5,
+                    "bbox_fs": [
+                        105,
+                        213,
+                        506,
+                        239
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        243,
+                        280,
+                        255
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                243,
+                                281,
+                                256
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        243,
+                                        281,
+                                        256
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Then we can prove the following Theorem.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        }
+                    ],
+                    "index": 11,
+                    "bbox_fs": [
+                        106,
+                        243,
+                        281,
+                        256
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        258,
+                        506,
+                        342
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                257,
+                                505,
+                                273
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        257,
+                                        249,
+                                        273
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Theorem 4.3. Given an MDP",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        249,
+                                        261,
+                                        261,
+                                        270
+                                    ],
+                                    "score": 0.66,
+                                    "content": "M",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        261,
+                                        257,
+                                        325,
+                                        273
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", its estimate",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        325,
+                                        258,
+                                        338,
+                                        271
+                                    ],
+                                    "score": 0.85,
+                                    "content": "\\widehat { M }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        339,
+                                        257,
+                                        505,
+                                        273
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with a bijective representation func-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                270,
+                                506,
+                                287
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        270,
+                                        128,
+                                        287
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tion",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        128,
+                                        273,
+                                        136,
+                                        285
+                                    ],
+                                    "score": 0.76,
+                                    "content": "\\phi",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        136,
+                                        270,
+                                        173,
+                                        287
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", i.e.",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        173,
+                                        271,
+                                        308,
+                                        285
+                                    ],
+                                    "score": 0.91,
+                                    "content": "( \\widehat { T } , \\widehat { R } ) = \\langle \\widehat { T } _ { z } \\circ \\phi , \\widehat { R } _ { z } \\circ \\phi \\rangle",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        309,
+                                        270,
+                                        387,
+                                        287
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", and an RKHS",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        387,
+                                        273,
+                                        487,
+                                        285
+                                    ],
+                                    "score": 0.59,
+                                    "content": "\\begin{array} { r } { \\mathcal { H } _ { k } \\quad \\subset \\quad ( \\mathcal { Z } \\quad \\to \\quad \\mathbb { R } ) } \\end{array}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        488,
+                                        270,
+                                        506,
+                                        287
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                282,
+                                504,
+                                299
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        282,
+                                        232,
+                                        298
+                                    ],
+                                    "score": 1.0,
+                                    "content": "duced by a universal kernel",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        232,
+                                        285,
+                                        240,
+                                        294
+                                    ],
+                                    "score": 0.73,
+                                    "content": "k",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        240,
+                                        282,
+                                        287,
+                                        298
+                                    ],
+                                    "score": 1.0,
+                                    "content": "such that",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        287,
+                                        284,
+                                        378,
+                                        297
+                                    ],
+                                    "score": 0.9,
+                                    "content": "\\mathrm { s u p } _ { z \\in \\mathcal { Z } } k ( z , z ) ~ = ~ \\bar { k }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        378,
+                                        282,
+                                        440,
+                                        298
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", assume that",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        441,
+                                        284,
+                                        504,
+                                        299
+                                    ],
+                                    "score": 0.89,
+                                    "content": "f _ { \\phi , \\widehat { R } _ { z } , \\widehat { T } _ { z } } \\left( \\widehat { z } \\right) \\ =",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                296,
+                                508,
+                                318
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        297,
+                                        344,
+                                        317
+                                    ],
+                                    "score": 0.9,
+                                    "content": "\\begin{array} { r } { \\mathbb { E } _ { T , \\pi } \\Big [ \\sum _ { t = 0 } ^ { \\infty } \\gamma ^ { t } \\mathcal { E } _ { \\phi , \\widehat { R } _ { z } , \\widehat { T } _ { z } } \\big ( s _ { t } , a _ { t } \\big ) \\Big | \\big ( s _ { 0 } , a _ { 0 } \\big ) = \\phi ^ { - 1 } \\big ( z \\big ) \\Big ] \\in \\mathcal { H } _ { k } } \\end{array}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        344,
+                                        296,
+                                        366,
+                                        318
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        367,
+                                        301,
+                                        451,
+                                        316
+                                    ],
+                                    "score": 0.93,
+                                    "content": "B _ { \\phi } = \\| f _ { \\phi , \\widehat { R } _ { z } , \\widehat { T } _ { z } } \\| _ { \\mathcal { H } _ { k } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        451,
+                                        296,
+                                        508,
+                                        318
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and the loss",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                103,
+                                312,
+                                506,
+                                334
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        103,
+                                        312,
+                                        165,
+                                        334
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is bounded by",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        165,
+                                        317,
+                                        293,
+                                        331
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\begin{array} { r } { \\bar { \\mathcal { E } } = \\operatorname* { s u p } _ { s \\in { \\mathcal { S } } , a \\in \\mathcal { A } } \\mathcal { E } _ { \\phi , \\widehat { R } _ { z } , \\widehat { T } _ { z } } ( s , a ) } \\end{array}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        293,
+                                        312,
+                                        423,
+                                        334
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Let n be the number of data in",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        424,
+                                        317,
+                                        432,
+                                        327
+                                    ],
+                                    "score": 0.78,
+                                    "content": "\\mathcal { D }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        433,
+                                        312,
+                                        506,
+                                        334
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". With probability",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                327,
+                                140,
+                                343
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        329,
+                                        135,
+                                        340
+                                    ],
+                                    "score": 0.64,
+                                    "content": "1 - 2 \\delta",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        135,
+                                        327,
+                                        140,
+                                        343
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        }
+                    ],
+                    "index": 14.5,
+                    "bbox_fs": [
+                        103,
+                        257,
+                        508,
+                        343
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        111,
+                        345,
+                        506,
+                        381
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                111,
+                                345,
+                                506,
+                                381
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        111,
+                                        345,
+                                        506,
+                                        381
+                                    ],
+                                    "score": 0.89,
+                                    "content": "R ^ { \\pi } - { \\widehat R } ^ { \\pi } \\Big | \\leq \\frac { 1 } { n } \\sum _ { ( s , a ) \\in { \\mathcal D } } { \\mathcal E } _ { \\phi , { \\widehat R } _ { z } , { \\widehat T } _ { z } } ( s , a ) + B _ { \\phi } \\widehat \\Pi \\widehat M \\big ( d _ { \\phi } ^ { \\pi } , d _ { \\phi } ^ { { D } } \\big ) + \\sqrt { \\frac { \\bar { \\mathcal E } ^ { 2 } } { 2 n } \\log { \\frac { 1 } { \\delta } } } + B _ { \\phi } \\sqrt { \\frac { \\bar { k } } { n } } \\left( 4 + \\sqrt { 8 \\log { \\frac { 3 } { \\delta } } } \\right) .",
+                                    "type": "interline_equation",
+                                    "image_path": "38315fc6b9bb6240f5da5e207809b2ab70cd1e301d17089118a70400757cb45d.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 19,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                111,
+                                345,
+                                506,
+                                357.0
+                            ],
+                            "spans": [],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                111,
+                                357.0,
+                                506,
+                                369.0
+                            ],
+                            "spans": [],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                111,
+                                369.0,
+                                506,
+                                381.0
+                            ],
+                            "spans": [],
+                            "index": 20
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        391,
+                        479,
+                        405
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                391,
+                                479,
+                                406
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        391,
+                                        420,
+                                        406
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Proof. Applying the Hoeffding inequality (Hoeffding, 1963), with probability",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        421,
+                                        393,
+                                        443,
+                                        403
+                                    ],
+                                    "score": 0.84,
+                                    "content": "1 - \\delta",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        444,
+                                        391,
+                                        479,
+                                        406
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", we get:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        }
+                    ],
+                    "index": 21,
+                    "bbox_fs": [
+                        105,
+                        391,
+                        479,
+                        406
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        163,
+                        409,
+                        447,
+                        444
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                163,
+                                409,
+                                447,
+                                444
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        163,
+                                        409,
+                                        447,
+                                        444
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\mathbb { E } _ { ( s , a ) \\sim d ^ { \\mathcal { D } } } \\left[ \\mathcal { E } _ { \\phi , \\widehat { R } _ { z } , \\widehat { T } _ { z } } ( s , a ) \\right] \\leq \\frac { 1 } { n } \\sum _ { ( s , a ) \\sim \\mathcal { D } } \\mathcal { E } _ { \\phi , \\widehat { R } _ { z } , \\widehat { T } _ { z } } ( s , a ) + \\sqrt { \\frac { \\bar { \\mathcal { E } } ^ { 2 } } { 2 n } \\log \\frac { 1 } { \\delta } } .",
+                                    "type": "interline_equation",
+                                    "image_path": "a143e9f568260a599d52a8c6d831e1d3e60ae1e101d09e4e264932b139e1cebe.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 23,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                163,
+                                409,
+                                447,
+                                420.6666666666667
+                            ],
+                            "spans": [],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                163,
+                                420.6666666666667,
+                                447,
+                                432.33333333333337
+                            ],
+                            "spans": [],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                163,
+                                432.33333333333337,
+                                447,
+                                444.00000000000006
+                            ],
+                            "spans": [],
+                            "index": 24
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        104,
+                        449,
+                        486,
+                        462
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                448,
+                                486,
+                                463
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        448,
+                                        486,
+                                        463
+                                    ],
+                                    "score": 1.0,
+                                    "content": "By using an union bound with Eq. (13) and substituting terms in Eq. (3), we recover the result.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        }
+                    ],
+                    "index": 25,
+                    "bbox_fs": [
+                        105,
+                        448,
+                        486,
+                        463
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        81,
+                        243,
+                        93
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                81,
+                                244,
+                                95
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        81,
+                                        244,
+                                        95
+                                    ],
+                                    "score": 1.0,
+                                    "content": "B EXPERIMENT DETAILS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        }
+                    ],
+                    "index": 0
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        106,
+                        266,
+                        118
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                106,
+                                267,
+                                119
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        106,
+                                        267,
+                                        119
+                                    ],
+                                    "score": 1.0,
+                                    "content": "B.1 COMPUTING INFRASTRUCTURE",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        }
+                    ],
+                    "index": 1
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        127,
+                        505,
+                        183
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                127,
+                                505,
+                                141
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        127,
+                                        505,
+                                        141
+                                    ],
+                                    "score": 1.0,
+                                    "content": "All experiments were conducted on the Google Cloud Platform. Specifically, we used compute-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                138,
+                                505,
+                                151
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        138,
+                                        505,
+                                        151
+                                    ],
+                                    "score": 1.0,
+                                    "content": "optimized machines (c2-standard-4) that provide 4 vCPUs and 16 GB memory for the evaluation",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                149,
+                                505,
+                                163
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        149,
+                                        505,
+                                        163
+                                    ],
+                                    "score": 1.0,
+                                    "content": "experiment of Section 5.1, and we used high-memory machines (n1-highmem-4), which provide",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                160,
+                                506,
+                                173
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        160,
+                                        506,
+                                        173
+                                    ],
+                                    "score": 1.0,
+                                    "content": "4 vCPUs and 26GB memory, equipped with an Nvidia Tesla K80 GPU for the RL experiment of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                172,
+                                156,
+                                183
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        172,
+                                        156,
+                                        183
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Section 5.2.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        }
+                    ],
+                    "index": 4
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        109,
+                        197,
+                        284,
+                        208
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                197,
+                                286,
+                                210
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        197,
+                                        286,
+                                        210
+                                    ],
+                                    "score": 1.0,
+                                    "content": "B.2 DETAILS OF THE OPE EXPERIMENT",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        }
+                    ],
+                    "index": 7
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        218,
+                        505,
+                        339
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                218,
+                                506,
+                                230
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        218,
+                                        506,
+                                        230
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Task details We did not modify CartPole-v0 environment and Acrobot-v1 environment from the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                228,
+                                505,
+                                243
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        228,
+                                        505,
+                                        243
+                                    ],
+                                    "score": 1.0,
+                                    "content": "original implementation of OpenAI Gym (Brockman et al., 2016) except for the maximum trajectory",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                240,
+                                505,
+                                253
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        240,
+                                        505,
+                                        253
+                                    ],
+                                    "score": 1.0,
+                                    "content": "length. We ran PPO (Schulman et al., 2017) to optimize policies to reach a certain performance level",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                251,
+                                505,
+                                263
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        251,
+                                        505,
+                                        263
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and set them as the target policies for CartPole-v0 and Acrobot-v1. For the HIV simulator, we used",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                261,
+                                506,
+                                275
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        261,
+                                        456,
+                                        275
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the code adapted by Liu et al. (2018b), which is originally from the implementation of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        456,
+                                        262,
+                                        484,
+                                        273
+                                    ],
+                                    "score": 0.68,
+                                    "content": "\\mathsf { R L P y } ^ { 3 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        484,
+                                        261,
+                                        506,
+                                        275
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". We",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                273,
+                                505,
+                                285
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        274,
+                                        411,
+                                        285
+                                    ],
+                                    "score": 1.0,
+                                    "content": "modified the environment to have more randomness in the initial state (up to",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        412,
+                                        273,
+                                        431,
+                                        284
+                                    ],
+                                    "score": 0.89,
+                                    "content": "1 0 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        432,
+                                        274,
+                                        505,
+                                        285
+                                    ],
+                                    "score": 1.0,
+                                    "content": "perturbation from",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                284,
+                                505,
+                                297
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        284,
+                                        505,
+                                        297
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the baseline initial state) and to use the reward function that gives the logarithm of original reward",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                294,
+                                505,
+                                308
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        294,
+                                        319,
+                                        308
+                                    ],
+                                    "score": 1.0,
+                                    "content": "values, as the original reward function scales up to",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        320,
+                                        294,
+                                        339,
+                                        306
+                                    ],
+                                    "score": 0.84,
+                                    "content": "1 0 ^ { 1 0 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        339,
+                                        294,
+                                        460,
+                                        308
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". We used a tree-based fitted",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        461,
+                                        297,
+                                        468,
+                                        307
+                                    ],
+                                    "score": 0.66,
+                                    "content": "{ \\bf q }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        468,
+                                        294,
+                                        505,
+                                        308
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-iteration",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                306,
+                                505,
+                                318
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        306,
+                                        505,
+                                        318
+                                    ],
+                                    "score": 1.0,
+                                    "content": "algorithm implemented by Liu et al. (2018b) to optimize the target policy for the HIV simulator. All",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                316,
+                                505,
+                                329
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        316,
+                                        505,
+                                        329
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the other details are shown in Table 2. We assume that the termination conditions of tasks are known",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                327,
+                                141,
+                                342
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        327,
+                                        141,
+                                        342
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in prior.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        }
+                    ],
+                    "index": 13
+                },
+                {
+                    "type": "table",
+                    "bbox": [
+                        133,
+                        374,
+                        479,
+                        478
+                    ],
+                    "blocks": [
+                        {
+                            "type": "table_caption",
+                            "bbox": [
+                                219,
+                                350,
+                                392,
+                                362
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        217,
+                                        349,
+                                        393,
+                                        364
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                217,
+                                                349,
+                                                393,
+                                                364
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Table 2: Task settings of OPE experiments.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 19
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "type": "table_body",
+                            "bbox": [
+                                133,
+                                374,
+                                479,
+                                478
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        133,
+                                        374,
+                                        479,
+                                        478
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                133,
+                                                374,
+                                                479,
+                                                478
+                                            ],
+                                            "score": 0.977,
+                                            "html": "<table><tr><td></td><td>CartPole-vO</td><td>Acrobot-v1</td><td>HIV simulator</td></tr><tr><td></td><td></td><td></td><td></td></tr><tr><td>state space dimension</td><td>4</td><td>6</td><td>6</td></tr><tr><td># of actions</td><td>2</td><td>3</td><td>4</td></tr><tr><td>discount rate y</td><td>0.99</td><td>0.99</td><td>0.98</td></tr><tr><td># of trajectories</td><td>200</td><td>200</td><td>50</td></tr><tr><td>max length of training traj.</td><td>200</td><td>200</td><td>200</td></tr><tr><td>max length of rollouts for evaluation</td><td>1000</td><td>1000</td><td>1000</td></tr><tr><td>discounted return of target policy</td><td>89.4</td><td>-56.88</td><td>803.2</td></tr></table>",
+                                            "type": "table",
+                                            "image_path": "ff1293c514c4b257cc52982e06a45ba9a96fe99a6b4e136baaa72b7cf5b218a4.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 21,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        133,
+                                        374,
+                                        479,
+                                        408.6666666666667
+                                    ],
+                                    "spans": [],
+                                    "index": 20
+                                },
+                                {
+                                    "bbox": [
+                                        133,
+                                        408.6666666666667,
+                                        479,
+                                        443.33333333333337
+                                    ],
+                                    "spans": [],
+                                    "index": 21
+                                },
+                                {
+                                    "bbox": [
+                                        133,
+                                        443.33333333333337,
+                                        479,
+                                        478.00000000000006
+                                    ],
+                                    "spans": [],
+                                    "index": 22
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 20.0
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        507,
+                        505,
+                        607
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                507,
+                                504,
+                                519
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        507,
+                                        504,
+                                        519
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Model and algorithm details The model we learn is composed of a representation module and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                518,
+                                506,
+                                531
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        518,
+                                        506,
+                                        531
+                                    ],
+                                    "score": 1.0,
+                                    "content": "a dynamics module. To be consistent with the experiment settings in Liu et al. (2018b), we use a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                530,
+                                506,
+                                542
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        530,
+                                        506,
+                                        542
+                                    ],
+                                    "score": 1.0,
+                                    "content": "representation module of a single hidden layer feed-forward network that takes the state as input",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                541,
+                                505,
+                                553
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        541,
+                                        384,
+                                        553
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and outputs representation. We squashed the representation between",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        384,
+                                        541,
+                                        414,
+                                        552
+                                    ],
+                                    "score": 0.92,
+                                    "content": "( - 1 , 1 )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        415,
+                                        541,
+                                        505,
+                                        553
+                                    ],
+                                    "score": 1.0,
+                                    "content": "using the tanh activa-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                551,
+                                506,
+                                564
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        551,
+                                        506,
+                                        564
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tion function. The dynamics module is also a single hidden layer feed-forward network that takes",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                562,
+                                506,
+                                575
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        562,
+                                        506,
+                                        575
+                                    ],
+                                    "score": 1.0,
+                                    "content": "representation as input and outputs state difference and reward prediction for each action. We use",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                574,
+                                505,
+                                585
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        574,
+                                        505,
+                                        585
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the swish activation function (Ramachandran et al., 2017) for the hidden layers of two modules. As",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                582,
+                                506,
+                                599
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        582,
+                                        506,
+                                        599
+                                    ],
+                                    "score": 1.0,
+                                    "content": "a whole, the model can also be seen as a feed-forward network with three hidden layers of varying",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                595,
+                                504,
+                                609
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        595,
+                                        504,
+                                        609
+                                    ],
+                                    "score": 1.0,
+                                    "content": "activation functions, where the output of the second hidden layer is the representation we regularize.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        }
+                    ],
+                    "index": 27
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        612,
+                        505,
+                        711
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                611,
+                                505,
+                                625
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        611,
+                                        303,
+                                        625
+                                    ],
+                                    "score": 1.0,
+                                    "content": "For the purpose of comparison, we minimize the",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        303,
+                                        613,
+                                        316,
+                                        623
+                                    ],
+                                    "score": 0.67,
+                                    "content": "L 2",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        317,
+                                        611,
+                                        505,
+                                        625
+                                    ],
+                                    "score": 1.0,
+                                    "content": "distance between the model prediction and the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                623,
+                                504,
+                                635
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        623,
+                                        504,
+                                        635
+                                    ],
+                                    "score": 1.0,
+                                    "content": "desired outcome from data, which corresponds to using a model of Gaussian predictive distribu-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                634,
+                                505,
+                                646
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        634,
+                                        505,
+                                        646
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tion with fixed variance. We standardized the inputs and outputs of the neural network and used",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                644,
+                                506,
+                                658
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        644,
+                                        313,
+                                        658
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Adam (Kingma & Ba, 2014) with a learning rate of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        313,
+                                        645,
+                                        351,
+                                        655
+                                    ],
+                                    "score": 0.92,
+                                    "content": "3 \\times 1 0 ^ { - 4 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        352,
+                                        644,
+                                        506,
+                                        658
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for the optimization. When compared",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                655,
+                                505,
+                                669
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        655,
+                                        505,
+                                        669
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to the similar experiments conducted in Liu et al. (2018b), we used a larger and more expressive",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                667,
+                                505,
+                                680
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        667,
+                                        505,
+                                        680
+                                    ],
+                                    "score": 1.0,
+                                    "content": "model with more optimization steps with a smaller learning rate for a more accurate comparison.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                678,
+                                504,
+                                690
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        678,
+                                        504,
+                                        690
+                                    ],
+                                    "score": 1.0,
+                                    "content": "While the derivation of the RepB-SDE objective was based on the state-action representation func-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                689,
+                                506,
+                                702
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        689,
+                                        506,
+                                        702
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tion, we use state representation in this experiment for direct comparison with RepBM, which uses",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                700,
+                                505,
+                                713
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        700,
+                                        505,
+                                        713
+                                    ],
+                                    "score": 1.0,
+                                    "content": "state representation (it can be also understood as using action invariant kernel). We follow the choice",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        }
+                    ],
+                    "index": 36
+                }
+            ],
+            "page_idx": 17,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        119,
+                        721,
+                        274,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                118,
+                                719,
+                                275,
+                                734
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        118,
+                                        719,
+                                        275,
+                                        734
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3https://github.com/rlpy/rlpy",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        107,
+                        27,
+                        292,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                26,
+                                293,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        26,
+                                        293,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Published as a conference paper at ICLR 2021",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        300,
+                        751,
+                        311,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                299,
+                                750,
+                                313,
+                                763
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        299,
+                                        750,
+                                        313,
+                                        763
+                                    ],
+                                    "score": 1.0,
+                                    "content": "18",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        81,
+                        243,
+                        93
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                81,
+                                244,
+                                95
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        81,
+                                        244,
+                                        95
+                                    ],
+                                    "score": 1.0,
+                                    "content": "B EXPERIMENT DETAILS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        }
+                    ],
+                    "index": 0
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        106,
+                        266,
+                        118
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                106,
+                                267,
+                                119
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        106,
+                                        267,
+                                        119
+                                    ],
+                                    "score": 1.0,
+                                    "content": "B.1 COMPUTING INFRASTRUCTURE",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        }
+                    ],
+                    "index": 1
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        127,
+                        505,
+                        183
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                127,
+                                505,
+                                141
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        127,
+                                        505,
+                                        141
+                                    ],
+                                    "score": 1.0,
+                                    "content": "All experiments were conducted on the Google Cloud Platform. Specifically, we used compute-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                138,
+                                505,
+                                151
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        138,
+                                        505,
+                                        151
+                                    ],
+                                    "score": 1.0,
+                                    "content": "optimized machines (c2-standard-4) that provide 4 vCPUs and 16 GB memory for the evaluation",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                149,
+                                505,
+                                163
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        149,
+                                        505,
+                                        163
+                                    ],
+                                    "score": 1.0,
+                                    "content": "experiment of Section 5.1, and we used high-memory machines (n1-highmem-4), which provide",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                160,
+                                506,
+                                173
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        160,
+                                        506,
+                                        173
+                                    ],
+                                    "score": 1.0,
+                                    "content": "4 vCPUs and 26GB memory, equipped with an Nvidia Tesla K80 GPU for the RL experiment of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                172,
+                                156,
+                                183
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        172,
+                                        156,
+                                        183
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Section 5.2.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        }
+                    ],
+                    "index": 4,
+                    "bbox_fs": [
+                        105,
+                        127,
+                        506,
+                        183
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        109,
+                        197,
+                        284,
+                        208
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                197,
+                                286,
+                                210
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        197,
+                                        286,
+                                        210
+                                    ],
+                                    "score": 1.0,
+                                    "content": "B.2 DETAILS OF THE OPE EXPERIMENT",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        }
+                    ],
+                    "index": 7
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        218,
+                        505,
+                        339
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                218,
+                                506,
+                                230
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        218,
+                                        506,
+                                        230
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Task details We did not modify CartPole-v0 environment and Acrobot-v1 environment from the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                228,
+                                505,
+                                243
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        228,
+                                        505,
+                                        243
+                                    ],
+                                    "score": 1.0,
+                                    "content": "original implementation of OpenAI Gym (Brockman et al., 2016) except for the maximum trajectory",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                240,
+                                505,
+                                253
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        240,
+                                        505,
+                                        253
+                                    ],
+                                    "score": 1.0,
+                                    "content": "length. We ran PPO (Schulman et al., 2017) to optimize policies to reach a certain performance level",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                251,
+                                505,
+                                263
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        251,
+                                        505,
+                                        263
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and set them as the target policies for CartPole-v0 and Acrobot-v1. For the HIV simulator, we used",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                261,
+                                506,
+                                275
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        261,
+                                        456,
+                                        275
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the code adapted by Liu et al. (2018b), which is originally from the implementation of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        456,
+                                        262,
+                                        484,
+                                        273
+                                    ],
+                                    "score": 0.68,
+                                    "content": "\\mathsf { R L P y } ^ { 3 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        484,
+                                        261,
+                                        506,
+                                        275
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". We",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                273,
+                                505,
+                                285
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        274,
+                                        411,
+                                        285
+                                    ],
+                                    "score": 1.0,
+                                    "content": "modified the environment to have more randomness in the initial state (up to",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        412,
+                                        273,
+                                        431,
+                                        284
+                                    ],
+                                    "score": 0.89,
+                                    "content": "1 0 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        432,
+                                        274,
+                                        505,
+                                        285
+                                    ],
+                                    "score": 1.0,
+                                    "content": "perturbation from",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                284,
+                                505,
+                                297
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        284,
+                                        505,
+                                        297
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the baseline initial state) and to use the reward function that gives the logarithm of original reward",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                294,
+                                505,
+                                308
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        294,
+                                        319,
+                                        308
+                                    ],
+                                    "score": 1.0,
+                                    "content": "values, as the original reward function scales up to",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        320,
+                                        294,
+                                        339,
+                                        306
+                                    ],
+                                    "score": 0.84,
+                                    "content": "1 0 ^ { 1 0 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        339,
+                                        294,
+                                        460,
+                                        308
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". We used a tree-based fitted",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        461,
+                                        297,
+                                        468,
+                                        307
+                                    ],
+                                    "score": 0.66,
+                                    "content": "{ \\bf q }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        468,
+                                        294,
+                                        505,
+                                        308
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-iteration",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                306,
+                                505,
+                                318
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        306,
+                                        505,
+                                        318
+                                    ],
+                                    "score": 1.0,
+                                    "content": "algorithm implemented by Liu et al. (2018b) to optimize the target policy for the HIV simulator. All",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                316,
+                                505,
+                                329
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        316,
+                                        505,
+                                        329
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the other details are shown in Table 2. We assume that the termination conditions of tasks are known",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                327,
+                                141,
+                                342
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        327,
+                                        141,
+                                        342
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in prior.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        }
+                    ],
+                    "index": 13,
+                    "bbox_fs": [
+                        104,
+                        218,
+                        506,
+                        342
+                    ]
+                },
+                {
+                    "type": "table",
+                    "bbox": [
+                        133,
+                        374,
+                        479,
+                        478
+                    ],
+                    "blocks": [
+                        {
+                            "type": "table_caption",
+                            "bbox": [
+                                219,
+                                350,
+                                392,
+                                362
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        217,
+                                        349,
+                                        393,
+                                        364
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                217,
+                                                349,
+                                                393,
+                                                364
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Table 2: Task settings of OPE experiments.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 19
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "type": "table_body",
+                            "bbox": [
+                                133,
+                                374,
+                                479,
+                                478
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        133,
+                                        374,
+                                        479,
+                                        478
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                133,
+                                                374,
+                                                479,
+                                                478
+                                            ],
+                                            "score": 0.977,
+                                            "html": "<table><tr><td></td><td>CartPole-vO</td><td>Acrobot-v1</td><td>HIV simulator</td></tr><tr><td></td><td></td><td></td><td></td></tr><tr><td>state space dimension</td><td>4</td><td>6</td><td>6</td></tr><tr><td># of actions</td><td>2</td><td>3</td><td>4</td></tr><tr><td>discount rate y</td><td>0.99</td><td>0.99</td><td>0.98</td></tr><tr><td># of trajectories</td><td>200</td><td>200</td><td>50</td></tr><tr><td>max length of training traj.</td><td>200</td><td>200</td><td>200</td></tr><tr><td>max length of rollouts for evaluation</td><td>1000</td><td>1000</td><td>1000</td></tr><tr><td>discounted return of target policy</td><td>89.4</td><td>-56.88</td><td>803.2</td></tr></table>",
+                                            "type": "table",
+                                            "image_path": "ff1293c514c4b257cc52982e06a45ba9a96fe99a6b4e136baaa72b7cf5b218a4.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 21,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        133,
+                                        374,
+                                        479,
+                                        408.6666666666667
+                                    ],
+                                    "spans": [],
+                                    "index": 20
+                                },
+                                {
+                                    "bbox": [
+                                        133,
+                                        408.6666666666667,
+                                        479,
+                                        443.33333333333337
+                                    ],
+                                    "spans": [],
+                                    "index": 21
+                                },
+                                {
+                                    "bbox": [
+                                        133,
+                                        443.33333333333337,
+                                        479,
+                                        478.00000000000006
+                                    ],
+                                    "spans": [],
+                                    "index": 22
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 20.0
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        507,
+                        505,
+                        607
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                507,
+                                504,
+                                519
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        507,
+                                        504,
+                                        519
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Model and algorithm details The model we learn is composed of a representation module and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                518,
+                                506,
+                                531
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        518,
+                                        506,
+                                        531
+                                    ],
+                                    "score": 1.0,
+                                    "content": "a dynamics module. To be consistent with the experiment settings in Liu et al. (2018b), we use a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                530,
+                                506,
+                                542
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        530,
+                                        506,
+                                        542
+                                    ],
+                                    "score": 1.0,
+                                    "content": "representation module of a single hidden layer feed-forward network that takes the state as input",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                541,
+                                505,
+                                553
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        541,
+                                        384,
+                                        553
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and outputs representation. We squashed the representation between",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        384,
+                                        541,
+                                        414,
+                                        552
+                                    ],
+                                    "score": 0.92,
+                                    "content": "( - 1 , 1 )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        415,
+                                        541,
+                                        505,
+                                        553
+                                    ],
+                                    "score": 1.0,
+                                    "content": "using the tanh activa-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                551,
+                                506,
+                                564
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        551,
+                                        506,
+                                        564
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tion function. The dynamics module is also a single hidden layer feed-forward network that takes",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                562,
+                                506,
+                                575
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        562,
+                                        506,
+                                        575
+                                    ],
+                                    "score": 1.0,
+                                    "content": "representation as input and outputs state difference and reward prediction for each action. We use",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                574,
+                                505,
+                                585
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        574,
+                                        505,
+                                        585
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the swish activation function (Ramachandran et al., 2017) for the hidden layers of two modules. As",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                582,
+                                506,
+                                599
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        582,
+                                        506,
+                                        599
+                                    ],
+                                    "score": 1.0,
+                                    "content": "a whole, the model can also be seen as a feed-forward network with three hidden layers of varying",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                595,
+                                504,
+                                609
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        595,
+                                        504,
+                                        609
+                                    ],
+                                    "score": 1.0,
+                                    "content": "activation functions, where the output of the second hidden layer is the representation we regularize.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        }
+                    ],
+                    "index": 27,
+                    "bbox_fs": [
+                        104,
+                        507,
+                        506,
+                        609
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        612,
+                        505,
+                        711
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                611,
+                                505,
+                                625
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        611,
+                                        303,
+                                        625
+                                    ],
+                                    "score": 1.0,
+                                    "content": "For the purpose of comparison, we minimize the",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        303,
+                                        613,
+                                        316,
+                                        623
+                                    ],
+                                    "score": 0.67,
+                                    "content": "L 2",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        317,
+                                        611,
+                                        505,
+                                        625
+                                    ],
+                                    "score": 1.0,
+                                    "content": "distance between the model prediction and the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                623,
+                                504,
+                                635
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        623,
+                                        504,
+                                        635
+                                    ],
+                                    "score": 1.0,
+                                    "content": "desired outcome from data, which corresponds to using a model of Gaussian predictive distribu-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                634,
+                                505,
+                                646
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        634,
+                                        505,
+                                        646
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tion with fixed variance. We standardized the inputs and outputs of the neural network and used",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                644,
+                                506,
+                                658
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        644,
+                                        313,
+                                        658
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Adam (Kingma & Ba, 2014) with a learning rate of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        313,
+                                        645,
+                                        351,
+                                        655
+                                    ],
+                                    "score": 0.92,
+                                    "content": "3 \\times 1 0 ^ { - 4 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        352,
+                                        644,
+                                        506,
+                                        658
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for the optimization. When compared",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                655,
+                                505,
+                                669
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        655,
+                                        505,
+                                        669
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to the similar experiments conducted in Liu et al. (2018b), we used a larger and more expressive",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                667,
+                                505,
+                                680
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        667,
+                                        505,
+                                        680
+                                    ],
+                                    "score": 1.0,
+                                    "content": "model with more optimization steps with a smaller learning rate for a more accurate comparison.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                678,
+                                504,
+                                690
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        678,
+                                        504,
+                                        690
+                                    ],
+                                    "score": 1.0,
+                                    "content": "While the derivation of the RepB-SDE objective was based on the state-action representation func-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                689,
+                                506,
+                                702
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        689,
+                                        506,
+                                        702
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tion, we use state representation in this experiment for direct comparison with RepBM, which uses",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                700,
+                                505,
+                                713
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        700,
+                                        505,
+                                        713
+                                    ],
+                                    "score": 1.0,
+                                    "content": "state representation (it can be also understood as using action invariant kernel). We follow the choice",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                81,
+                                505,
+                                96
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        81,
+                                        293,
+                                        96
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of Liu et al. (2018b) and use dot product kernel",
+                                    "type": "text",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        294,
+                                        82,
+                                        406,
+                                        95
+                                    ],
+                                    "score": 0.92,
+                                    "content": "k ( \\phi ( s ) , \\phi ( \\bar { s } ) ) = \\phi ( s ) ^ { \\top } \\phi ( \\bar { s } )",
+                                    "type": "inline_equation",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        406,
+                                        81,
+                                        505,
+                                        96
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for the OPE experiment,",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                93,
+                                505,
+                                106
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        93,
+                                        505,
+                                        106
+                                    ],
+                                    "score": 1.0,
+                                    "content": "which is not universal but allows us to avoid search of kernel hyperparameters, such as length-scales.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 1
+                        }
+                    ],
+                    "index": 36,
+                    "bbox_fs": [
+                        105,
+                        611,
+                        506,
+                        713
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        81,
+                        504,
+                        105
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                81,
+                                505,
+                                96
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        81,
+                                        293,
+                                        96
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of Liu et al. (2018b) and use dot product kernel",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        294,
+                                        82,
+                                        406,
+                                        95
+                                    ],
+                                    "score": 0.92,
+                                    "content": "k ( \\phi ( s ) , \\phi ( \\bar { s } ) ) = \\phi ( s ) ^ { \\top } \\phi ( \\bar { s } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        406,
+                                        81,
+                                        505,
+                                        96
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for the OPE experiment,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                93,
+                                505,
+                                106
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        93,
+                                        505,
+                                        106
+                                    ],
+                                    "score": 1.0,
+                                    "content": "which is not universal but allows us to avoid search of kernel hyperparameters, such as length-scales.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        }
+                    ],
+                    "index": 0.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        110,
+                        504,
+                        144
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                110,
+                                506,
+                                122
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        110,
+                                        506,
+                                        122
+                                    ],
+                                    "score": 1.0,
+                                    "content": "After the training, we generate another 200 trajectories (50 in case of HIV simulator), and rollout",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                122,
+                                505,
+                                133
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        122,
+                                        505,
+                                        133
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in both true and simulated (based on learned model) environments to evaluate models. We measure",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                132,
+                                226,
+                                143
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        132,
+                                        226,
+                                        143
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the individual MSE, which is",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        }
+                    ],
+                    "index": 3
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        150,
+                        149,
+                        459,
+                        181
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                150,
+                                149,
+                                459,
+                                181
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        150,
+                                        149,
+                                        459,
+                                        181
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\mathrm { I n d i v i d u a l ~ M S E } = \\mathbb { E } _ { s _ { 0 } \\sim d _ { 0 } } \\bigg [ \\bigg ( \\mathbb { E } _ { \\widehat { M } , \\pi } \\Big [ \\sum _ { t = 0 } ^ { \\infty } \\gamma ^ { t } r _ { t } | s _ { 0 } \\Big ] - \\mathbb { E } _ { M , \\pi } \\Big [ \\sum _ { t = 0 } ^ { \\infty } \\gamma ^ { t } r _ { t } | s _ { 0 } \\Big ] \\bigg ) ^ { 2 } \\bigg ]",
+                                    "type": "interline_equation",
+                                    "image_path": "8f8df59f7994214c467ec82d6d2b271632da645a86884b5b445941a6e4e829c3.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 6,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                150,
+                                149,
+                                459,
+                                159.66666666666666
+                            ],
+                            "spans": [],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                150,
+                                159.66666666666666,
+                                459,
+                                170.33333333333331
+                            ],
+                            "spans": [],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                150,
+                                170.33333333333331,
+                                459,
+                                180.99999999999997
+                            ],
+                            "spans": [],
+                            "index": 7
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        186,
+                        504,
+                        208
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                185,
+                                505,
+                                200
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        185,
+                                        505,
+                                        200
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for measuring the performance of each model. Whole experiment, from sampling data to learning",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                197,
+                                415,
+                                209
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        197,
+                                        415,
+                                        209
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and evaluating the model, is repeated 200 times with different random seeds.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        }
+                    ],
+                    "index": 8.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        213,
+                        334,
+                        323
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                212,
+                                335,
+                                226
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        212,
+                                        264,
+                                        226
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Choice and effect of hyperparameter",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        264,
+                                        215,
+                                        272,
+                                        223
+                                    ],
+                                    "score": 0.75,
+                                    "content": "\\alpha",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        272,
+                                        212,
+                                        335,
+                                        226
+                                    ],
+                                    "score": 1.0,
+                                    "content": "For choosing",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                224,
+                                335,
+                                236
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        224,
+                                        172,
+                                        236
+                                    ],
+                                    "score": 1.0,
+                                    "content": "hyperparameter",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        172,
+                                        226,
+                                        180,
+                                        234
+                                    ],
+                                    "score": 0.76,
+                                    "content": "\\alpha",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        180,
+                                        224,
+                                        335,
+                                        236
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for each algorithm, we searched over",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                235,
+                                334,
+                                247
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        235,
+                                        227,
+                                        247
+                                    ],
+                                    "score": 0.9,
+                                    "content": "\\alpha \\in \\{ 0 . 0 0 1 , 0 . 0 1 , 0 . 1 , 1 , 1 0 \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        228,
+                                        235,
+                                        327,
+                                        247
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for each off-policyness",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        328,
+                                        237,
+                                        334,
+                                        245
+                                    ],
+                                    "score": 0.55,
+                                    "content": "\\epsilon",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                245,
+                                334,
+                                258
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        245,
+                                        315,
+                                        258
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and for each environment. Chosen αs were mainly",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        316,
+                                        247,
+                                        334,
+                                        257
+                                    ],
+                                    "score": 0.82,
+                                    "content": "\\alpha \\in",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                107,
+                                257,
+                                335,
+                                269
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        257,
+                                        163,
+                                        268
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\lbrace 0 . 0 0 1 , 0 . 0 1 \\rbrace",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        163,
+                                        257,
+                                        231,
+                                        269
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for CartPole-v0,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        232,
+                                        257,
+                                        281,
+                                        269
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\alpha \\in \\{ 1 , 1 0 \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        281,
+                                        257,
+                                        335,
+                                        269
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for Acrobot-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                267,
+                                335,
+                                280
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        267,
+                                        140,
+                                        279
+                                    ],
+                                    "score": 1.0,
+                                    "content": "v1, and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        140,
+                                        268,
+                                        207,
+                                        280
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\alpha \\in \\{ 0 . 0 1 , 0 . 1 \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        208,
+                                        267,
+                                        335,
+                                        279
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for HIV simulator for both al-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                279,
+                                335,
+                                290
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        279,
+                                        219,
+                                        290
+                                    ],
+                                    "score": 1.0,
+                                    "content": "gorithms. In general, large",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        219,
+                                        281,
+                                        227,
+                                        289
+                                    ],
+                                    "score": 0.76,
+                                    "content": "\\alpha",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        228,
+                                        279,
+                                        335,
+                                        290
+                                    ],
+                                    "score": 1.0,
+                                    "content": "was beneficial when high",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                290,
+                                335,
+                                302
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        290,
+                                        335,
+                                        302
+                                    ],
+                                    "score": 1.0,
+                                    "content": "off-policyness (\u000f) is present and/or the task is hard to gen-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                300,
+                                335,
+                                313
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        300,
+                                        335,
+                                        313
+                                    ],
+                                    "score": 1.0,
+                                    "content": "eralize. On the right we show the example of effect of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                312,
+                                212,
+                                323
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        312,
+                                        139,
+                                        323
+                                    ],
+                                    "score": 1.0,
+                                    "content": "varying",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        139,
+                                        314,
+                                        147,
+                                        322
+                                    ],
+                                    "score": 0.76,
+                                    "content": "\\alpha",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        147,
+                                        312,
+                                        212,
+                                        323
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in CartPole-v0.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        }
+                    ],
+                    "index": 14.5
+                },
+                {
+                    "type": "image",
+                    "bbox": [
+                        339,
+                        214,
+                        501,
+                        319
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                339,
+                                214,
+                                501,
+                                319
+                            ],
+                            "group_id": 1,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        339,
+                                        214,
+                                        501,
+                                        319
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                339,
+                                                214,
+                                                501,
+                                                319
+                                            ],
+                                            "score": 0.95,
+                                            "type": "image",
+                                            "image_path": "1c9ccb5c2c41ddb695510e4975c65cec88c2fe83b3762b2ff076037bc3277679.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 23.5,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        339,
+                                        214,
+                                        501,
+                                        227.125
+                                    ],
+                                    "spans": [],
+                                    "index": 20
+                                },
+                                {
+                                    "bbox": [
+                                        339,
+                                        227.125,
+                                        501,
+                                        240.25
+                                    ],
+                                    "spans": [],
+                                    "index": 21
+                                },
+                                {
+                                    "bbox": [
+                                        339,
+                                        240.25,
+                                        501,
+                                        253.375
+                                    ],
+                                    "spans": [],
+                                    "index": 22
+                                },
+                                {
+                                    "bbox": [
+                                        339,
+                                        253.375,
+                                        501,
+                                        266.5
+                                    ],
+                                    "spans": [],
+                                    "index": 23
+                                },
+                                {
+                                    "bbox": [
+                                        339,
+                                        266.5,
+                                        501,
+                                        279.625
+                                    ],
+                                    "spans": [],
+                                    "index": 24
+                                },
+                                {
+                                    "bbox": [
+                                        339,
+                                        279.625,
+                                        501,
+                                        292.75
+                                    ],
+                                    "spans": [],
+                                    "index": 25
+                                },
+                                {
+                                    "bbox": [
+                                        339,
+                                        292.75,
+                                        501,
+                                        305.875
+                                    ],
+                                    "spans": [],
+                                    "index": 26
+                                },
+                                {
+                                    "bbox": [
+                                        339,
+                                        305.875,
+                                        501,
+                                        319.0
+                                    ],
+                                    "spans": [],
+                                    "index": 27
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 23.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        334,
+                        505,
+                        445
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                335,
+                                505,
+                                347
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        335,
+                                        505,
+                                        347
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Comparison to other baselines In Figure 2, the OPE results with other model-free baselines",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                345,
+                                506,
+                                358
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        345,
+                                        506,
+                                        358
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are presented. FQE: Fitted Q-evaluation, IS: step-wise importance sampling, DR: doubly robust",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                356,
+                                506,
+                                370
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        356,
+                                        506,
+                                        370
+                                    ],
+                                    "score": 1.0,
+                                    "content": "estimator based on step-wise importance sampling using the value function learned with fitted Q-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                368,
+                                505,
+                                380
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        368,
+                                        505,
+                                        380
+                                    ],
+                                    "score": 1.0,
+                                    "content": "evaluation (Jiang & Li, 2016), DualDICE: stationary distribution correction algorithm (Nachum",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                378,
+                                505,
+                                391
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        378,
+                                        505,
+                                        391
+                                    ],
+                                    "score": 1.0,
+                                    "content": "et al., 2019a). We used the implementation provided by the authors in case of DualDICE. All results",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                390,
+                                505,
+                                401
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        390,
+                                        505,
+                                        401
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are normalized to set the MSE of model-based baseline to be 1. Here, we used average MSEs instead",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                399,
+                                505,
+                                413
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        399,
+                                        505,
+                                        413
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of individual MSEs since importance sampling estimators are not suitable in computing individual",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                410,
+                                505,
+                                424
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        410,
+                                        505,
+                                        424
+                                    ],
+                                    "score": 1.0,
+                                    "content": "MSEs. The results show that model-based methods are more robust to increasing off-policyness",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                423,
+                                505,
+                                434
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        423,
+                                        505,
+                                        434
+                                    ],
+                                    "score": 1.0,
+                                    "content": "when compared to FQE. The results of model-based methods on Acrobot is relatively worse due to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                434,
+                                296,
+                                446
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        434,
+                                        296,
+                                        446
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the difficult dynamics of Acrobot environment.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        }
+                    ],
+                    "index": 32.5
+                },
+                {
+                    "type": "image",
+                    "bbox": [
+                        106,
+                        454,
+                        506,
+                        572
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                106,
+                                454,
+                                506,
+                                572
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        454,
+                                        506,
+                                        572
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                454,
+                                                506,
+                                                572
+                                            ],
+                                            "score": 0.969,
+                                            "type": "image",
+                                            "image_path": "1eb2b85496bae8da7d8f1105ae49dddbe789c2255747b8997f70703f372d78dd.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 39,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        454,
+                                        506,
+                                        493.3333333333333
+                                    ],
+                                    "spans": [],
+                                    "index": 38
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        493.3333333333333,
+                                        506,
+                                        532.6666666666666
+                                    ],
+                                    "spans": [],
+                                    "index": 39
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        532.6666666666666,
+                                        506,
+                                        572.0
+                                    ],
+                                    "spans": [],
+                                    "index": 40
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                105,
+                                579,
+                                504,
+                                602
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        579,
+                                        505,
+                                        592
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                579,
+                                                505,
+                                                592
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 2: The OPE results compared to other model-free baselines. All experiments are repeated",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 41
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        590,
+                                        351,
+                                        602
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                590,
+                                                249,
+                                                602
+                                            ],
+                                            "score": 1.0,
+                                            "content": "200 times and the error bars denote",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                249,
+                                                591,
+                                                269,
+                                                601
+                                            ],
+                                            "score": 0.87,
+                                            "content": "9 5 \\%",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                270,
+                                                590,
+                                                351,
+                                                602
+                                            ],
+                                            "score": 1.0,
+                                            "content": "confidence interval.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 42
+                                }
+                            ],
+                            "index": 41.5
+                        }
+                    ],
+                    "index": 40.25
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        623,
+                        321,
+                        634
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                623,
+                                322,
+                                635
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        623,
+                                        322,
+                                        635
+                                    ],
+                                    "score": 1.0,
+                                    "content": "B.3 DETAILS OF THE OFFLINE RL EXPERIMENTS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        }
+                    ],
+                    "index": 43
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        643,
+                        505,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                644,
+                                505,
+                                656
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        644,
+                                        289,
+                                        656
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Task details We used 12 datasets in D4RL",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        289,
+                                        644,
+                                        302,
+                                        654
+                                    ],
+                                    "score": 0.26,
+                                    "content": "\\mathrm { F u }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        302,
+                                        644,
+                                        505,
+                                        656
+                                    ],
+                                    "score": 1.0,
+                                    "content": "et al., 2020) over four dataset types and three envi-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                654,
+                                505,
+                                667
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        654,
+                                        505,
+                                        667
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ronments as specified in the main text. In Table 1, normalized scores suggested by (Fu et al., 2020)",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                666,
+                                505,
+                                678
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        666,
+                                        505,
+                                        678
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is used to report the result, where score of 0 corresponds to a random policy and 100 corresponds",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                677,
+                                505,
+                                689
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        677,
+                                        477,
+                                        689
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to a converged SAC policy. In HalfCheetah-v2, score of 0 means the undiscounted return of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        478,
+                                        677,
+                                        501,
+                                        687
+                                    ],
+                                    "score": 0.52,
+                                    "content": "- 2 8 0",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        502,
+                                        677,
+                                        505,
+                                        689
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 47
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                688,
+                                504,
+                                699
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        688,
+                                        504,
+                                        699
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and score of 1 means the undiscounted return of 12135. In Hopper-v2, score of 0 means the undis-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 48
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                699,
+                                506,
+                                711
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        699,
+                                        177,
+                                        711
+                                    ],
+                                    "score": 1.0,
+                                    "content": "counted return of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        177,
+                                        699,
+                                        196,
+                                        709
+                                    ],
+                                    "score": 0.77,
+                                    "content": "- 2 0",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        196,
+                                        699,
+                                        506,
+                                        711
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", and score of 1 means the undiscounted return of 3234. In Walker2d-v2 score",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 49
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                710,
+                                505,
+                                721
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        710,
+                                        505,
+                                        721
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of 0 means the undiscounted return of 2, and score of 1 means the undiscounted return of 4592. We",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 50
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                720,
+                                374,
+                                733
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        720,
+                                        374,
+                                        733
+                                    ],
+                                    "score": 1.0,
+                                    "content": "assume that the termination conditions of tasks are known in prior.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 51
+                        }
+                    ],
+                    "index": 47.5
+                }
+            ],
+            "page_idx": 18,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        108,
+                        27,
+                        292,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                26,
+                                293,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        26,
+                                        293,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Published as a conference paper at ICLR 2021",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        301,
+                        751,
+                        311,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                299,
+                                750,
+                                312,
+                                764
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        299,
+                                        750,
+                                        312,
+                                        764
+                                    ],
+                                    "score": 1.0,
+                                    "content": "",
+                                    "type": "text",
+                                    "height": 14,
+                                    "width": 13
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        81,
+                        504,
+                        105
+                    ],
+                    "lines": [],
+                    "index": 0.5,
+                    "bbox_fs": [
+                        105,
+                        81,
+                        505,
+                        106
+                    ],
+                    "lines_deleted": true
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        110,
+                        504,
+                        144
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                110,
+                                506,
+                                122
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        110,
+                                        506,
+                                        122
+                                    ],
+                                    "score": 1.0,
+                                    "content": "After the training, we generate another 200 trajectories (50 in case of HIV simulator), and rollout",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                122,
+                                505,
+                                133
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        122,
+                                        505,
+                                        133
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in both true and simulated (based on learned model) environments to evaluate models. We measure",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                132,
+                                226,
+                                143
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        132,
+                                        226,
+                                        143
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the individual MSE, which is",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        }
+                    ],
+                    "index": 3,
+                    "bbox_fs": [
+                        106,
+                        110,
+                        506,
+                        143
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        150,
+                        149,
+                        459,
+                        181
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                150,
+                                149,
+                                459,
+                                181
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        150,
+                                        149,
+                                        459,
+                                        181
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\mathrm { I n d i v i d u a l ~ M S E } = \\mathbb { E } _ { s _ { 0 } \\sim d _ { 0 } } \\bigg [ \\bigg ( \\mathbb { E } _ { \\widehat { M } , \\pi } \\Big [ \\sum _ { t = 0 } ^ { \\infty } \\gamma ^ { t } r _ { t } | s _ { 0 } \\Big ] - \\mathbb { E } _ { M , \\pi } \\Big [ \\sum _ { t = 0 } ^ { \\infty } \\gamma ^ { t } r _ { t } | s _ { 0 } \\Big ] \\bigg ) ^ { 2 } \\bigg ]",
+                                    "type": "interline_equation",
+                                    "image_path": "8f8df59f7994214c467ec82d6d2b271632da645a86884b5b445941a6e4e829c3.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 6,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                150,
+                                149,
+                                459,
+                                159.66666666666666
+                            ],
+                            "spans": [],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                150,
+                                159.66666666666666,
+                                459,
+                                170.33333333333331
+                            ],
+                            "spans": [],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                150,
+                                170.33333333333331,
+                                459,
+                                180.99999999999997
+                            ],
+                            "spans": [],
+                            "index": 7
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        186,
+                        504,
+                        208
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                185,
+                                505,
+                                200
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        185,
+                                        505,
+                                        200
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for measuring the performance of each model. Whole experiment, from sampling data to learning",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                197,
+                                415,
+                                209
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        197,
+                                        415,
+                                        209
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and evaluating the model, is repeated 200 times with different random seeds.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        }
+                    ],
+                    "index": 8.5,
+                    "bbox_fs": [
+                        105,
+                        185,
+                        505,
+                        209
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        213,
+                        334,
+                        323
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                212,
+                                335,
+                                226
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        212,
+                                        264,
+                                        226
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Choice and effect of hyperparameter",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        264,
+                                        215,
+                                        272,
+                                        223
+                                    ],
+                                    "score": 0.75,
+                                    "content": "\\alpha",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        272,
+                                        212,
+                                        335,
+                                        226
+                                    ],
+                                    "score": 1.0,
+                                    "content": "For choosing",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                224,
+                                335,
+                                236
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        224,
+                                        172,
+                                        236
+                                    ],
+                                    "score": 1.0,
+                                    "content": "hyperparameter",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        172,
+                                        226,
+                                        180,
+                                        234
+                                    ],
+                                    "score": 0.76,
+                                    "content": "\\alpha",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        180,
+                                        224,
+                                        335,
+                                        236
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for each algorithm, we searched over",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                235,
+                                334,
+                                247
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        235,
+                                        227,
+                                        247
+                                    ],
+                                    "score": 0.9,
+                                    "content": "\\alpha \\in \\{ 0 . 0 0 1 , 0 . 0 1 , 0 . 1 , 1 , 1 0 \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        228,
+                                        235,
+                                        327,
+                                        247
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for each off-policyness",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        328,
+                                        237,
+                                        334,
+                                        245
+                                    ],
+                                    "score": 0.55,
+                                    "content": "\\epsilon",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                245,
+                                334,
+                                258
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        245,
+                                        315,
+                                        258
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and for each environment. Chosen αs were mainly",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        316,
+                                        247,
+                                        334,
+                                        257
+                                    ],
+                                    "score": 0.82,
+                                    "content": "\\alpha \\in",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                107,
+                                257,
+                                335,
+                                269
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        257,
+                                        163,
+                                        268
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\lbrace 0 . 0 0 1 , 0 . 0 1 \\rbrace",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        163,
+                                        257,
+                                        231,
+                                        269
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for CartPole-v0,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        232,
+                                        257,
+                                        281,
+                                        269
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\alpha \\in \\{ 1 , 1 0 \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        281,
+                                        257,
+                                        335,
+                                        269
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for Acrobot-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                267,
+                                335,
+                                280
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        267,
+                                        140,
+                                        279
+                                    ],
+                                    "score": 1.0,
+                                    "content": "v1, and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        140,
+                                        268,
+                                        207,
+                                        280
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\alpha \\in \\{ 0 . 0 1 , 0 . 1 \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        208,
+                                        267,
+                                        335,
+                                        279
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for HIV simulator for both al-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                279,
+                                335,
+                                290
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        279,
+                                        219,
+                                        290
+                                    ],
+                                    "score": 1.0,
+                                    "content": "gorithms. In general, large",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        219,
+                                        281,
+                                        227,
+                                        289
+                                    ],
+                                    "score": 0.76,
+                                    "content": "\\alpha",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        228,
+                                        279,
+                                        335,
+                                        290
+                                    ],
+                                    "score": 1.0,
+                                    "content": "was beneficial when high",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                290,
+                                335,
+                                302
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        290,
+                                        335,
+                                        302
+                                    ],
+                                    "score": 1.0,
+                                    "content": "off-policyness (\u000f) is present and/or the task is hard to gen-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                300,
+                                335,
+                                313
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        300,
+                                        335,
+                                        313
+                                    ],
+                                    "score": 1.0,
+                                    "content": "eralize. On the right we show the example of effect of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                312,
+                                212,
+                                323
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        312,
+                                        139,
+                                        323
+                                    ],
+                                    "score": 1.0,
+                                    "content": "varying",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        139,
+                                        314,
+                                        147,
+                                        322
+                                    ],
+                                    "score": 0.76,
+                                    "content": "\\alpha",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        147,
+                                        312,
+                                        212,
+                                        323
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in CartPole-v0.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        }
+                    ],
+                    "index": 14.5,
+                    "bbox_fs": [
+                        105,
+                        212,
+                        335,
+                        323
+                    ]
+                },
+                {
+                    "type": "image",
+                    "bbox": [
+                        339,
+                        214,
+                        501,
+                        319
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                339,
+                                214,
+                                501,
+                                319
+                            ],
+                            "group_id": 1,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        339,
+                                        214,
+                                        501,
+                                        319
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                339,
+                                                214,
+                                                501,
+                                                319
+                                            ],
+                                            "score": 0.95,
+                                            "type": "image",
+                                            "image_path": "1c9ccb5c2c41ddb695510e4975c65cec88c2fe83b3762b2ff076037bc3277679.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 23.5,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        339,
+                                        214,
+                                        501,
+                                        227.125
+                                    ],
+                                    "spans": [],
+                                    "index": 20
+                                },
+                                {
+                                    "bbox": [
+                                        339,
+                                        227.125,
+                                        501,
+                                        240.25
+                                    ],
+                                    "spans": [],
+                                    "index": 21
+                                },
+                                {
+                                    "bbox": [
+                                        339,
+                                        240.25,
+                                        501,
+                                        253.375
+                                    ],
+                                    "spans": [],
+                                    "index": 22
+                                },
+                                {
+                                    "bbox": [
+                                        339,
+                                        253.375,
+                                        501,
+                                        266.5
+                                    ],
+                                    "spans": [],
+                                    "index": 23
+                                },
+                                {
+                                    "bbox": [
+                                        339,
+                                        266.5,
+                                        501,
+                                        279.625
+                                    ],
+                                    "spans": [],
+                                    "index": 24
+                                },
+                                {
+                                    "bbox": [
+                                        339,
+                                        279.625,
+                                        501,
+                                        292.75
+                                    ],
+                                    "spans": [],
+                                    "index": 25
+                                },
+                                {
+                                    "bbox": [
+                                        339,
+                                        292.75,
+                                        501,
+                                        305.875
+                                    ],
+                                    "spans": [],
+                                    "index": 26
+                                },
+                                {
+                                    "bbox": [
+                                        339,
+                                        305.875,
+                                        501,
+                                        319.0
+                                    ],
+                                    "spans": [],
+                                    "index": 27
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 23.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        334,
+                        505,
+                        445
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                335,
+                                505,
+                                347
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        335,
+                                        505,
+                                        347
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Comparison to other baselines In Figure 2, the OPE results with other model-free baselines",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                345,
+                                506,
+                                358
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        345,
+                                        506,
+                                        358
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are presented. FQE: Fitted Q-evaluation, IS: step-wise importance sampling, DR: doubly robust",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                356,
+                                506,
+                                370
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        356,
+                                        506,
+                                        370
+                                    ],
+                                    "score": 1.0,
+                                    "content": "estimator based on step-wise importance sampling using the value function learned with fitted Q-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                368,
+                                505,
+                                380
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        368,
+                                        505,
+                                        380
+                                    ],
+                                    "score": 1.0,
+                                    "content": "evaluation (Jiang & Li, 2016), DualDICE: stationary distribution correction algorithm (Nachum",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                378,
+                                505,
+                                391
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        378,
+                                        505,
+                                        391
+                                    ],
+                                    "score": 1.0,
+                                    "content": "et al., 2019a). We used the implementation provided by the authors in case of DualDICE. All results",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                390,
+                                505,
+                                401
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        390,
+                                        505,
+                                        401
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are normalized to set the MSE of model-based baseline to be 1. Here, we used average MSEs instead",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                399,
+                                505,
+                                413
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        399,
+                                        505,
+                                        413
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of individual MSEs since importance sampling estimators are not suitable in computing individual",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                410,
+                                505,
+                                424
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        410,
+                                        505,
+                                        424
+                                    ],
+                                    "score": 1.0,
+                                    "content": "MSEs. The results show that model-based methods are more robust to increasing off-policyness",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                423,
+                                505,
+                                434
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        423,
+                                        505,
+                                        434
+                                    ],
+                                    "score": 1.0,
+                                    "content": "when compared to FQE. The results of model-based methods on Acrobot is relatively worse due to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                434,
+                                296,
+                                446
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        434,
+                                        296,
+                                        446
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the difficult dynamics of Acrobot environment.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        }
+                    ],
+                    "index": 32.5,
+                    "bbox_fs": [
+                        105,
+                        335,
+                        506,
+                        446
+                    ]
+                },
+                {
+                    "type": "image",
+                    "bbox": [
+                        106,
+                        454,
+                        506,
+                        572
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                106,
+                                454,
+                                506,
+                                572
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        454,
+                                        506,
+                                        572
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                454,
+                                                506,
+                                                572
+                                            ],
+                                            "score": 0.969,
+                                            "type": "image",
+                                            "image_path": "1eb2b85496bae8da7d8f1105ae49dddbe789c2255747b8997f70703f372d78dd.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 39,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        454,
+                                        506,
+                                        493.3333333333333
+                                    ],
+                                    "spans": [],
+                                    "index": 38
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        493.3333333333333,
+                                        506,
+                                        532.6666666666666
+                                    ],
+                                    "spans": [],
+                                    "index": 39
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        532.6666666666666,
+                                        506,
+                                        572.0
+                                    ],
+                                    "spans": [],
+                                    "index": 40
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                105,
+                                579,
+                                504,
+                                602
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        579,
+                                        505,
+                                        592
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                579,
+                                                505,
+                                                592
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 2: The OPE results compared to other model-free baselines. All experiments are repeated",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 41
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        590,
+                                        351,
+                                        602
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                590,
+                                                249,
+                                                602
+                                            ],
+                                            "score": 1.0,
+                                            "content": "200 times and the error bars denote",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                249,
+                                                591,
+                                                269,
+                                                601
+                                            ],
+                                            "score": 0.87,
+                                            "content": "9 5 \\%",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                270,
+                                                590,
+                                                351,
+                                                602
+                                            ],
+                                            "score": 1.0,
+                                            "content": "confidence interval.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 42
+                                }
+                            ],
+                            "index": 41.5
+                        }
+                    ],
+                    "index": 40.25
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        623,
+                        321,
+                        634
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                623,
+                                322,
+                                635
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        623,
+                                        322,
+                                        635
+                                    ],
+                                    "score": 1.0,
+                                    "content": "B.3 DETAILS OF THE OFFLINE RL EXPERIMENTS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        }
+                    ],
+                    "index": 43
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        643,
+                        505,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                644,
+                                505,
+                                656
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        644,
+                                        289,
+                                        656
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Task details We used 12 datasets in D4RL",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        289,
+                                        644,
+                                        302,
+                                        654
+                                    ],
+                                    "score": 0.26,
+                                    "content": "\\mathrm { F u }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        302,
+                                        644,
+                                        505,
+                                        656
+                                    ],
+                                    "score": 1.0,
+                                    "content": "et al., 2020) over four dataset types and three envi-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                654,
+                                505,
+                                667
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        654,
+                                        505,
+                                        667
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ronments as specified in the main text. In Table 1, normalized scores suggested by (Fu et al., 2020)",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                666,
+                                505,
+                                678
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        666,
+                                        505,
+                                        678
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is used to report the result, where score of 0 corresponds to a random policy and 100 corresponds",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                677,
+                                505,
+                                689
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        677,
+                                        477,
+                                        689
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to a converged SAC policy. In HalfCheetah-v2, score of 0 means the undiscounted return of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        478,
+                                        677,
+                                        501,
+                                        687
+                                    ],
+                                    "score": 0.52,
+                                    "content": "- 2 8 0",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        502,
+                                        677,
+                                        505,
+                                        689
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 47
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                688,
+                                504,
+                                699
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        688,
+                                        504,
+                                        699
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and score of 1 means the undiscounted return of 12135. In Hopper-v2, score of 0 means the undis-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 48
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                699,
+                                506,
+                                711
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        699,
+                                        177,
+                                        711
+                                    ],
+                                    "score": 1.0,
+                                    "content": "counted return of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        177,
+                                        699,
+                                        196,
+                                        709
+                                    ],
+                                    "score": 0.77,
+                                    "content": "- 2 0",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        196,
+                                        699,
+                                        506,
+                                        711
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", and score of 1 means the undiscounted return of 3234. In Walker2d-v2 score",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 49
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                710,
+                                505,
+                                721
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        710,
+                                        505,
+                                        721
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of 0 means the undiscounted return of 2, and score of 1 means the undiscounted return of 4592. We",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 50
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                720,
+                                374,
+                                733
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        720,
+                                        374,
+                                        733
+                                    ],
+                                    "score": 1.0,
+                                    "content": "assume that the termination conditions of tasks are known in prior.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 51
+                        }
+                    ],
+                    "index": 47.5,
+                    "bbox_fs": [
+                        104,
+                        644,
+                        506,
+                        733
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        82,
+                        505,
+                        152
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                82,
+                                505,
+                                95
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        82,
+                                        505,
+                                        95
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Representation balancing maximizing undiscounted return As we report the undiscounted sum",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                93,
+                                505,
+                                106
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        93,
+                                        388,
+                                        106
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of rewards in the experiments, the maximization of lower bound of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        389,
+                                        94,
+                                        403,
+                                        104
+                                    ],
+                                    "score": 0.87,
+                                    "content": "R ^ { \\pi }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        403,
+                                        93,
+                                        505,
+                                        106
+                                    ],
+                                    "score": 1.0,
+                                    "content": "may result in an under-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                104,
+                                505,
+                                117
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        104,
+                                        505,
+                                        117
+                                    ],
+                                    "score": 1.0,
+                                    "content": "utilization of experiences of later timesteps. One way to mitigate the mismatch is to optimize the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                114,
+                                506,
+                                129
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        114,
+                                        506,
+                                        129
+                                    ],
+                                    "score": 1.0,
+                                    "content": "policy by maximizing the returns starting from the states in the dataset. It corresponds to maximizing",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                107,
+                                121,
+                                509,
+                                146
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        127,
+                                        266,
+                                        141
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\begin{array} { r } { \\tilde { R } ^ { \\pi } = ( 1 - \\gamma ) \\mathbb { E } , M , \\pi \\bigl [ \\sum _ { t = 0 } ^ { \\infty } \\gamma ^ { t } r _ { t } \\bigr ] } \\end{array}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        266,
+                                        121,
+                                        311,
+                                        146
+                                    ],
+                                    "score": 1.0,
+                                    "content": "instead of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        311,
+                                        128,
+                                        325,
+                                        138
+                                    ],
+                                    "score": 0.87,
+                                    "content": "R ^ { \\pi }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        325,
+                                        121,
+                                        509,
+                                        146
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", where the expectation respect to the initial",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                140,
+                                345,
+                                152
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        140,
+                                        175,
+                                        152
+                                    ],
+                                    "score": 1.0,
+                                    "content": "state distribution",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        176,
+                                        141,
+                                        186,
+                                        151
+                                    ],
+                                    "score": 0.88,
+                                    "content": "d _ { 0 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        187,
+                                        140,
+                                        328,
+                                        152
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is altered with the data distribution",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        329,
+                                        140,
+                                        342,
+                                        150
+                                    ],
+                                    "score": 0.88,
+                                    "content": "d ^ { \\mathcal { D } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        342,
+                                        140,
+                                        345,
+                                        152
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        }
+                    ],
+                    "index": 2.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        156,
+                        505,
+                        204
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                156,
+                                505,
+                                169
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        156,
+                                        257,
+                                        169
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Consequently, to bound the error of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        257,
+                                        156,
+                                        271,
+                                        167
+                                    ],
+                                    "score": 0.88,
+                                    "content": "{ \\tilde { R } } ^ { \\pi }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        271,
+                                        156,
+                                        505,
+                                        169
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", the representation should be balanced with another dis-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                101,
+                                163,
+                                509,
+                                192
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        101,
+                                        163,
+                                        231,
+                                        192
+                                    ],
+                                    "score": 1.0,
+                                    "content": "counted stationary distribution",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        231,
+                                        169,
+                                        486,
+                                        182
+                                    ],
+                                    "score": 0.87,
+                                    "content": "\\begin{array} { r } { \\tilde { d } ^ { \\pi } ( s , a ) \\triangleq ( 1 - \\gamma ) \\sum _ { t = 0 } ^ { \\infty } \\gamma ^ { t } \\operatorname* { P r } ( s _ { t } = s , a _ { t } = a | s _ { 0 } \\sim d ^ { \\mathcal { D } } , T , \\pi ) } \\end{array}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        487,
+                                        163,
+                                        509,
+                                        192
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                242,
+                                183,
+                                249,
+                                191
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        242,
+                                        183,
+                                        249,
+                                        191
+                                    ],
+                                    "score": 0.76,
+                                    "content": "\\pi",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                192,
+                                301,
+                                204
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        192,
+                                        301,
+                                        204
+                                    ],
+                                    "score": 1.0,
+                                    "content": "derivations can be easily adapted by noting that:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        }
+                    ],
+                    "index": 7.5
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        111,
+                        208,
+                        506,
+                        243
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                111,
+                                208,
+                                506,
+                                243
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        111,
+                                        208,
+                                        506,
+                                        243
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\mathrm { P M } _ { \\mathcal { G } } ( \\tilde { d } _ { \\phi } ^ { \\pi } , d _ { \\phi } ^ { \\mathcal { D } } ) = \\operatorname* { s u p } _ { \\nu \\in \\mathcal { F } } \\left| ( 1 - \\gamma ) \\mathbb { E } _ { \\alpha \\sim \\pi ( s ) } \\left[ \\nu \\big ( \\phi ( s , a ) \\big ) \\right] - \\mathbb { E } _ { ( s , a , s ^ { \\prime } ) \\sim d ^ { \\mathcal { D } } } \\left[ \\nu \\big ( \\phi ( s , a ) \\big ) - \\gamma \\nu \\big ( \\phi ( s ^ { \\prime } , a ^ { \\prime } ) \\big ) \\right] \\right| ,",
+                                    "type": "interline_equation",
+                                    "image_path": "14ef59d4631f76c8b8c1477cb06be9a612301e6f5f17325e6aa561ff27047134.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 11,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                111,
+                                208,
+                                506,
+                                219.66666666666666
+                            ],
+                            "spans": [],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                111,
+                                219.66666666666666,
+                                506,
+                                231.33333333333331
+                            ],
+                            "spans": [],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                111,
+                                231.33333333333331,
+                                506,
+                                242.99999999999997
+                            ],
+                            "spans": [],
+                            "index": 12
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        249,
+                        459,
+                        262
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                248,
+                                461,
+                                263
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        248,
+                                        325,
+                                        263
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and changing the initial state sampling distributions to",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        325,
+                                        249,
+                                        338,
+                                        260
+                                    ],
+                                    "score": 0.88,
+                                    "content": "d ^ { \\mathcal { D } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        338,
+                                        248,
+                                        461,
+                                        263
+                                    ],
+                                    "score": 1.0,
+                                    "content": "during the estimation of IPM.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        }
+                    ],
+                    "index": 13
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        273,
+                        505,
+                        384
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                273,
+                                505,
+                                286
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        273,
+                                        505,
+                                        286
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Model and algorithm details Similar to the model used in the OPE experiment, the model we",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                286,
+                                505,
+                                297
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        286,
+                                        505,
+                                        297
+                                    ],
+                                    "score": 1.0,
+                                    "content": "learn is composed of a representation module and a dynamics module. A representation module is",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                295,
+                                505,
+                                309
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        295,
+                                        505,
+                                        309
+                                    ],
+                                    "score": 1.0,
+                                    "content": "a feed-forward network with two hidden layers that takes the state-action pair as input and outputs",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                307,
+                                506,
+                                320
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        307,
+                                        506,
+                                        320
+                                    ],
+                                    "score": 1.0,
+                                    "content": "representation through the tanh activation function. The dynamics module is a single hidden layer",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                318,
+                                506,
+                                331
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        318,
+                                        506,
+                                        331
+                                    ],
+                                    "score": 1.0,
+                                    "content": "network that takes representation as input and outputs parameters of diagonal Gaussian distribution",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                329,
+                                505,
+                                342
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        329,
+                                        505,
+                                        342
+                                    ],
+                                    "score": 1.0,
+                                    "content": "predicting state difference and reward. We use 200 hidden units for all intermediate layers including",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                340,
+                                506,
+                                352
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        340,
+                                        506,
+                                        352
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the representation layer. Across all domains, we train an ensemble of 7 models and pick the best",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                350,
+                                505,
+                                363
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        350,
+                                        505,
+                                        363
+                                    ],
+                                    "score": 1.0,
+                                    "content": "5 models on their validation error on hold-out set of 1000 transitions in the dataset. The inputs",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                362,
+                                506,
+                                374
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        362,
+                                        506,
+                                        374
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and outputs of the neural network is normalized. We present the pseudo-code of the presented",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                372,
+                                381,
+                                385
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        372,
+                                        381,
+                                        385
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Representation Balancing Offline Model-based RL algorithm below.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        }
+                    ],
+                    "index": 18.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        109,
+                        410,
+                        501,
+                        603
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                124,
+                                410,
+                                433,
+                                423
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        124,
+                                        410,
+                                        214,
+                                        423
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Input: Offline dataset",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        214,
+                                        411,
+                                        223,
+                                        420
+                                    ],
+                                    "score": 0.78,
+                                    "content": "\\mathcal { D }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        223,
+                                        410,
+                                        291,
+                                        423
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", previous policy",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        291,
+                                        413,
+                                        299,
+                                        420
+                                    ],
+                                    "score": 0.68,
+                                    "content": "\\pi",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        316,
+                                        410,
+                                        425,
+                                        423
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Output: Optimized policy",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        426,
+                                        413,
+                                        433,
+                                        421
+                                    ],
+                                    "score": 0.48,
+                                    "content": "\\pi",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                111,
+                                423,
+                                359,
+                                435
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        111,
+                                        423,
+                                        155,
+                                        435
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1: Sample",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        156,
+                                        424,
+                                        167,
+                                        434
+                                    ],
+                                    "score": 0.8,
+                                    "content": "K",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        167,
+                                        423,
+                                        346,
+                                        435
+                                    ],
+                                    "score": 1.0,
+                                    "content": "independent datasets with replacement from",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        346,
+                                        424,
+                                        355,
+                                        433
+                                    ],
+                                    "score": 0.73,
+                                    "content": "\\mathcal { D }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        355,
+                                        423,
+                                        359,
+                                        435
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                108,
+                                432,
+                                500,
+                                451
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        108,
+                                        432,
+                                        252,
+                                        451
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2: Train bootstrapped ensemble of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        252,
+                                        437,
+                                        263,
+                                        446
+                                    ],
+                                    "score": 0.82,
+                                    "content": "K",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        263,
+                                        432,
+                                        295,
+                                        451
+                                    ],
+                                    "score": 1.0,
+                                    "content": "models",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        295,
+                                        434,
+                                        344,
+                                        449
+                                    ],
+                                    "score": 0.94,
+                                    "content": "\\{ \\widehat { T } _ { i } , \\widehat { R } _ { i } \\} _ { i = 0 } ^ { K }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        344,
+                                        432,
+                                        480,
+                                        451
+                                    ],
+                                    "score": 1.0,
+                                    "content": "minimizing Eq. (8) (adapted with",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        481,
+                                        434,
+                                        493,
+                                        451
+                                    ],
+                                    "score": 0.86,
+                                    "content": "\\tilde { d } _ { \\phi } ^ { \\pi } ,",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        494,
+                                        432,
+                                        500,
+                                        451
+                                    ],
+                                    "score": 1.0,
+                                    "content": ").",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                111,
+                                448,
+                                226,
+                                460
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        111,
+                                        448,
+                                        156,
+                                        460
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3: for rep",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        156,
+                                        448,
+                                        211,
+                                        460
+                                    ],
+                                    "score": 0.7,
+                                    "content": "{ \\mathcal { a } } t = 0 , 1 , \\ldots",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        211,
+                                        448,
+                                        226,
+                                        460
+                                    ],
+                                    "score": 1.0,
+                                    "content": "do",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                110,
+                                458,
+                                257,
+                                471
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        110,
+                                        460,
+                                        121,
+                                        471
+                                    ],
+                                    "score": 1.0,
+                                    "content": "4:",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        137,
+                                        458,
+                                        185,
+                                        471
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for rollout",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        185,
+                                        460,
+                                        243,
+                                        471
+                                    ],
+                                    "score": 0.49,
+                                    "content": "\\mathbf { \\Sigma } = 0 , 1 , \\dots , B",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        243,
+                                        458,
+                                        257,
+                                        471
+                                    ],
+                                    "score": 1.0,
+                                    "content": "do",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                110,
+                                469,
+                                307,
+                                482
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        110,
+                                        470,
+                                        121,
+                                        482
+                                    ],
+                                    "score": 1.0,
+                                    "content": "5:",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        153,
+                                        469,
+                                        261,
+                                        482
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Sample initial rollout state",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        261,
+                                        472,
+                                        271,
+                                        481
+                                    ],
+                                    "score": 0.82,
+                                    "content": "s _ { 0 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        272,
+                                        469,
+                                        294,
+                                        482
+                                    ],
+                                    "score": 1.0,
+                                    "content": "from",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        295,
+                                        471,
+                                        304,
+                                        480
+                                    ],
+                                    "score": 0.82,
+                                    "content": "\\mathcal { D }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        304,
+                                        469,
+                                        307,
+                                        482
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                110,
+                                479,
+                                232,
+                                494
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        110,
+                                        481,
+                                        121,
+                                        493
+                                    ],
+                                    "score": 1.0,
+                                    "content": "6:",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        152,
+                                        479,
+                                        168,
+                                        494
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        168,
+                                        481,
+                                        217,
+                                        492
+                                    ],
+                                    "score": 0.65,
+                                    "content": "t = 0 , 1 , \\ldots",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        218,
+                                        479,
+                                        232,
+                                        494
+                                    ],
+                                    "score": 1.0,
+                                    "content": "do",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                110,
+                                492,
+                                288,
+                                505
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        110,
+                                        492,
+                                        121,
+                                        505
+                                    ],
+                                    "score": 1.0,
+                                    "content": "7:",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        167,
+                                        492,
+                                        239,
+                                        505
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Sample an action",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        240,
+                                        492,
+                                        284,
+                                        504
+                                    ],
+                                    "score": 0.91,
+                                    "content": "a _ { t } \\sim \\pi ( s _ { t } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        285,
+                                        492,
+                                        288,
+                                        505
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                110,
+                                503,
+                                429,
+                                518
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        110,
+                                        506,
+                                        121,
+                                        518
+                                    ],
+                                    "score": 1.0,
+                                    "content": "8:",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        167,
+                                        503,
+                                        232,
+                                        518
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Randomly pick",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        232,
+                                        504,
+                                        265,
+                                        518
+                                    ],
+                                    "score": 0.91,
+                                    "content": "( \\widehat { T } _ { i } , \\widehat { R } _ { i } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        266,
+                                        503,
+                                        315,
+                                        518
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and sample",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        315,
+                                        503,
+                                        429,
+                                        518
+                                    ],
+                                    "score": 0.88,
+                                    "content": "( s _ { t + 1 } , r _ { t } ) \\sim ( \\widehat { T } _ { i } , \\widehat { R } _ { i } ) ( s _ { t } , a _ { t } ) .",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                110,
+                                517,
+                                474,
+                                537
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        110,
+                                        520,
+                                        121,
+                                        533
+                                    ],
+                                    "score": 1.0,
+                                    "content": "9:",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        167,
+                                        518,
+                                        207,
+                                        534
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Compute",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        207,
+                                        517,
+                                        343,
+                                        537
+                                    ],
+                                    "score": 0.9,
+                                    "content": " \\tilde { r } _ { t } = r _ { t } - \\gamma \\lambda \\sqrt { \\mathbb { V } _ { K } [ \\mu ( s _ { t } , a _ { t } ) ] } | | _ { 2 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        344,
+                                        518,
+                                        385,
+                                        535
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and store",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        385,
+                                        519,
+                                        450,
+                                        532
+                                    ],
+                                    "score": 0.86,
+                                    "content": "\\left( { { s _ { t } } , { a _ { t } } , \\tilde { r } _ { t } , { s _ { t + 1 } } } \\right)",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        450,
+                                        518,
+                                        461,
+                                        535
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        461,
+                                        518,
+                                        470,
+                                        530
+                                    ],
+                                    "score": 0.82,
+                                    "content": "\\widehat { \\mathcal { D } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        471,
+                                        518,
+                                        474,
+                                        535
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                534,
+                                187,
+                                547
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        534,
+                                        121,
+                                        547
+                                    ],
+                                    "score": 1.0,
+                                    "content": "10:",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        152,
+                                        534,
+                                        187,
+                                        547
+                                    ],
+                                    "score": 1.0,
+                                    "content": "end for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                544,
+                                172,
+                                558
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        546,
+                                        121,
+                                        558
+                                    ],
+                                    "score": 1.0,
+                                    "content": "11:",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        137,
+                                        544,
+                                        172,
+                                        558
+                                    ],
+                                    "score": 1.0,
+                                    "content": "end for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                555,
+                                335,
+                                571
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        557,
+                                        121,
+                                        569
+                                    ],
+                                    "score": 1.0,
+                                    "content": "12:",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        137,
+                                        555,
+                                        219,
+                                        570
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Draw samples from",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        219,
+                                        557,
+                                        228,
+                                        567
+                                    ],
+                                    "score": 0.8,
+                                    "content": "\\mathcal { D }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        229,
+                                        555,
+                                        277,
+                                        570
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to compute",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        277,
+                                        555,
+                                        332,
+                                        571
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\widehat { \\mathrm { I P M } } ( \\tilde { d } _ { \\phi } ^ { \\pi } , d _ { \\phi } ^ { D } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        333,
+                                        555,
+                                        335,
+                                        570
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                570,
+                                332,
+                                585
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        571,
+                                        121,
+                                        584
+                                    ],
+                                    "score": 1.0,
+                                    "content": "13:",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        137,
+                                        570,
+                                        219,
+                                        585
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Draw samples from",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        219,
+                                        572,
+                                        228,
+                                        582
+                                    ],
+                                    "score": 0.8,
+                                    "content": "\\mathcal { D }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        229,
+                                        570,
+                                        246,
+                                        585
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        246,
+                                        570,
+                                        256,
+                                        582
+                                    ],
+                                    "score": 0.85,
+                                    "content": "\\widehat { \\mathcal { D } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        257,
+                                        570,
+                                        319,
+                                        585
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to update critic",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        320,
+                                        572,
+                                        328,
+                                        583
+                                    ],
+                                    "score": 0.83,
+                                    "content": "Q",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        329,
+                                        570,
+                                        332,
+                                        585
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                582,
+                                473,
+                                602
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        585,
+                                        121,
+                                        597
+                                    ],
+                                    "score": 1.0,
+                                    "content": "14:",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        135,
+                                        582,
+                                        181,
+                                        602
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Maximize",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        181,
+                                        583,
+                                        420,
+                                        600
+                                    ],
+                                    "score": 0.88,
+                                    "content": "\\mathbb { E } _ { s \\sim \\mathcal { D } \\cup \\widehat { \\mathcal { D } } , a \\sim \\pi ( s ) } [ Q ( s , a ) - \\tau \\log \\pi ( a | s ) ] - \\alpha _ { \\pi } \\widehat { \\mathrm { I P M } } ( \\tilde { d } _ { \\phi } ^ { \\pi } , d _ { \\phi } ^ { \\mathcal { D } } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        420,
+                                        582,
+                                        460,
+                                        602
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to update",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        461,
+                                        587,
+                                        468,
+                                        595
+                                    ],
+                                    "score": 0.64,
+                                    "content": "\\pi",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        468,
+                                        582,
+                                        473,
+                                        602
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                596,
+                                156,
+                                606
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        597,
+                                        120,
+                                        606
+                                    ],
+                                    "score": 0.928,
+                                    "content": "15:",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        125,
+                                        596,
+                                        156,
+                                        606
+                                    ],
+                                    "score": 0.957,
+                                    "content": "end for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        }
+                    ],
+                    "index": 31.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        623,
+                        505,
+                        712
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                623,
+                                505,
+                                637
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        623,
+                                        505,
+                                        637
+                                    ],
+                                    "score": 1.0,
+                                    "content": "For the result of MOPO (Yu et al., 2020), we ran the code kindly provided by the authors4 without",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                634,
+                                505,
+                                648
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        634,
+                                        505,
+                                        648
+                                    ],
+                                    "score": 1.0,
+                                    "content": "any modification on the algorithm or the hyperparameters. All algorithms we experimented (Base,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                645,
+                                506,
+                                659
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        645,
+                                        506,
+                                        659
+                                    ],
+                                    "score": 1.0,
+                                    "content": "RP, RepB-SDE) share all the hyperparameters in common except the ones associated with changing",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                657,
+                                504,
+                                669
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        657,
+                                        504,
+                                        669
+                                    ],
+                                    "score": 1.0,
+                                    "content": "objectives. We run SAC on the full rollouts from the trained ensemble models as shown in Algo-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                667,
+                                505,
+                                680
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        667,
+                                        505,
+                                        680
+                                    ],
+                                    "score": 1.0,
+                                    "content": "rithm 1. The common hyperparameters shared among algorithms are shown in Table 3. We simply",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                679,
+                                505,
+                                691
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        679,
+                                        505,
+                                        691
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tried the listed hyperparameters and not tuned them further. For RP and RepB-SDE, we penalized",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                690,
+                                505,
+                                701
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        690,
+                                        505,
+                                        701
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the reward from simulated environments with the standard deviation of prediction means of neural",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                701,
+                                505,
+                                712
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        701,
+                                        505,
+                                        712
+                                    ],
+                                    "score": 1.0,
+                                    "content": "network ensembles. We used standardized output of all 7 neural networks to compute the reward",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 47
+                        }
+                    ],
+                    "index": 43.5
+                }
+            ],
+            "page_idx": 19,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        118,
+                        722,
+                        312,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                118,
+                                719,
+                                313,
+                                734
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        118,
+                                        719,
+                                        313,
+                                        734
+                                    ],
+                                    "score": 1.0,
+                                    "content": "4https://github.com/tianheyu927/mopo",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        108,
+                        27,
+                        292,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                26,
+                                293,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        26,
+                                        293,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Published as a conference paper at ICLR 2021",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        300,
+                        751,
+                        311,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                298,
+                                750,
+                                313,
+                                763
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        298,
+                                        750,
+                                        313,
+                                        763
+                                    ],
+                                    "score": 1.0,
+                                    "content": "20",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        82,
+                        505,
+                        152
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                82,
+                                505,
+                                95
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        82,
+                                        505,
+                                        95
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Representation balancing maximizing undiscounted return As we report the undiscounted sum",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                93,
+                                505,
+                                106
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        93,
+                                        388,
+                                        106
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of rewards in the experiments, the maximization of lower bound of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        389,
+                                        94,
+                                        403,
+                                        104
+                                    ],
+                                    "score": 0.87,
+                                    "content": "R ^ { \\pi }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        403,
+                                        93,
+                                        505,
+                                        106
+                                    ],
+                                    "score": 1.0,
+                                    "content": "may result in an under-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                104,
+                                505,
+                                117
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        104,
+                                        505,
+                                        117
+                                    ],
+                                    "score": 1.0,
+                                    "content": "utilization of experiences of later timesteps. One way to mitigate the mismatch is to optimize the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                114,
+                                506,
+                                129
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        114,
+                                        506,
+                                        129
+                                    ],
+                                    "score": 1.0,
+                                    "content": "policy by maximizing the returns starting from the states in the dataset. It corresponds to maximizing",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                107,
+                                121,
+                                509,
+                                146
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        127,
+                                        266,
+                                        141
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\begin{array} { r } { \\tilde { R } ^ { \\pi } = ( 1 - \\gamma ) \\mathbb { E } , M , \\pi \\bigl [ \\sum _ { t = 0 } ^ { \\infty } \\gamma ^ { t } r _ { t } \\bigr ] } \\end{array}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        266,
+                                        121,
+                                        311,
+                                        146
+                                    ],
+                                    "score": 1.0,
+                                    "content": "instead of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        311,
+                                        128,
+                                        325,
+                                        138
+                                    ],
+                                    "score": 0.87,
+                                    "content": "R ^ { \\pi }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        325,
+                                        121,
+                                        509,
+                                        146
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", where the expectation respect to the initial",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                140,
+                                345,
+                                152
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        140,
+                                        175,
+                                        152
+                                    ],
+                                    "score": 1.0,
+                                    "content": "state distribution",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        176,
+                                        141,
+                                        186,
+                                        151
+                                    ],
+                                    "score": 0.88,
+                                    "content": "d _ { 0 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        187,
+                                        140,
+                                        328,
+                                        152
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is altered with the data distribution",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        329,
+                                        140,
+                                        342,
+                                        150
+                                    ],
+                                    "score": 0.88,
+                                    "content": "d ^ { \\mathcal { D } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        342,
+                                        140,
+                                        345,
+                                        152
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        }
+                    ],
+                    "index": 2.5,
+                    "bbox_fs": [
+                        105,
+                        82,
+                        509,
+                        152
+                    ]
+                },
+                {
+                    "type": "list",
+                    "bbox": [
+                        106,
+                        156,
+                        505,
+                        204
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                156,
+                                505,
+                                169
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        156,
+                                        257,
+                                        169
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Consequently, to bound the error of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        257,
+                                        156,
+                                        271,
+                                        167
+                                    ],
+                                    "score": 0.88,
+                                    "content": "{ \\tilde { R } } ^ { \\pi }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        271,
+                                        156,
+                                        505,
+                                        169
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", the representation should be balanced with another dis-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                101,
+                                163,
+                                509,
+                                192
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        101,
+                                        163,
+                                        231,
+                                        192
+                                    ],
+                                    "score": 1.0,
+                                    "content": "counted stationary distribution",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        231,
+                                        169,
+                                        486,
+                                        182
+                                    ],
+                                    "score": 0.87,
+                                    "content": "\\begin{array} { r } { \\tilde { d } ^ { \\pi } ( s , a ) \\triangleq ( 1 - \\gamma ) \\sum _ { t = 0 } ^ { \\infty } \\gamma ^ { t } \\operatorname* { P r } ( s _ { t } = s , a _ { t } = a | s _ { 0 } \\sim d ^ { \\mathcal { D } } , T , \\pi ) } \\end{array}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        487,
+                                        163,
+                                        509,
+                                        192
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                242,
+                                183,
+                                249,
+                                191
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        242,
+                                        183,
+                                        249,
+                                        191
+                                    ],
+                                    "score": 0.76,
+                                    "content": "\\pi",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 8,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                192,
+                                301,
+                                204
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        192,
+                                        301,
+                                        204
+                                    ],
+                                    "score": 1.0,
+                                    "content": "derivations can be easily adapted by noting that:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9,
+                            "is_list_start_line": true,
+                            "is_list_end_line": true
+                        }
+                    ],
+                    "index": 7.5,
+                    "bbox_fs": [
+                        101,
+                        156,
+                        509,
+                        204
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        111,
+                        208,
+                        506,
+                        243
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                111,
+                                208,
+                                506,
+                                243
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        111,
+                                        208,
+                                        506,
+                                        243
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\mathrm { P M } _ { \\mathcal { G } } ( \\tilde { d } _ { \\phi } ^ { \\pi } , d _ { \\phi } ^ { \\mathcal { D } } ) = \\operatorname* { s u p } _ { \\nu \\in \\mathcal { F } } \\left| ( 1 - \\gamma ) \\mathbb { E } _ { \\alpha \\sim \\pi ( s ) } \\left[ \\nu \\big ( \\phi ( s , a ) \\big ) \\right] - \\mathbb { E } _ { ( s , a , s ^ { \\prime } ) \\sim d ^ { \\mathcal { D } } } \\left[ \\nu \\big ( \\phi ( s , a ) \\big ) - \\gamma \\nu \\big ( \\phi ( s ^ { \\prime } , a ^ { \\prime } ) \\big ) \\right] \\right| ,",
+                                    "type": "interline_equation",
+                                    "image_path": "14ef59d4631f76c8b8c1477cb06be9a612301e6f5f17325e6aa561ff27047134.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 11,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                111,
+                                208,
+                                506,
+                                219.66666666666666
+                            ],
+                            "spans": [],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                111,
+                                219.66666666666666,
+                                506,
+                                231.33333333333331
+                            ],
+                            "spans": [],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                111,
+                                231.33333333333331,
+                                506,
+                                242.99999999999997
+                            ],
+                            "spans": [],
+                            "index": 12
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        249,
+                        459,
+                        262
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                248,
+                                461,
+                                263
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        248,
+                                        325,
+                                        263
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and changing the initial state sampling distributions to",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        325,
+                                        249,
+                                        338,
+                                        260
+                                    ],
+                                    "score": 0.88,
+                                    "content": "d ^ { \\mathcal { D } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        338,
+                                        248,
+                                        461,
+                                        263
+                                    ],
+                                    "score": 1.0,
+                                    "content": "during the estimation of IPM.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        }
+                    ],
+                    "index": 13,
+                    "bbox_fs": [
+                        105,
+                        248,
+                        461,
+                        263
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        273,
+                        505,
+                        384
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                273,
+                                505,
+                                286
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        273,
+                                        505,
+                                        286
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Model and algorithm details Similar to the model used in the OPE experiment, the model we",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                286,
+                                505,
+                                297
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        286,
+                                        505,
+                                        297
+                                    ],
+                                    "score": 1.0,
+                                    "content": "learn is composed of a representation module and a dynamics module. A representation module is",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                295,
+                                505,
+                                309
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        295,
+                                        505,
+                                        309
+                                    ],
+                                    "score": 1.0,
+                                    "content": "a feed-forward network with two hidden layers that takes the state-action pair as input and outputs",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                307,
+                                506,
+                                320
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        307,
+                                        506,
+                                        320
+                                    ],
+                                    "score": 1.0,
+                                    "content": "representation through the tanh activation function. The dynamics module is a single hidden layer",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                318,
+                                506,
+                                331
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        318,
+                                        506,
+                                        331
+                                    ],
+                                    "score": 1.0,
+                                    "content": "network that takes representation as input and outputs parameters of diagonal Gaussian distribution",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                329,
+                                505,
+                                342
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        329,
+                                        505,
+                                        342
+                                    ],
+                                    "score": 1.0,
+                                    "content": "predicting state difference and reward. We use 200 hidden units for all intermediate layers including",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                340,
+                                506,
+                                352
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        340,
+                                        506,
+                                        352
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the representation layer. Across all domains, we train an ensemble of 7 models and pick the best",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                350,
+                                505,
+                                363
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        350,
+                                        505,
+                                        363
+                                    ],
+                                    "score": 1.0,
+                                    "content": "5 models on their validation error on hold-out set of 1000 transitions in the dataset. The inputs",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                362,
+                                506,
+                                374
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        362,
+                                        506,
+                                        374
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and outputs of the neural network is normalized. We present the pseudo-code of the presented",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                372,
+                                381,
+                                385
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        372,
+                                        381,
+                                        385
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Representation Balancing Offline Model-based RL algorithm below.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        }
+                    ],
+                    "index": 18.5,
+                    "bbox_fs": [
+                        104,
+                        273,
+                        506,
+                        385
+                    ]
+                },
+                {
+                    "type": "index",
+                    "bbox": [
+                        109,
+                        410,
+                        501,
+                        603
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                124,
+                                410,
+                                433,
+                                423
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        124,
+                                        410,
+                                        214,
+                                        423
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Input: Offline dataset",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        214,
+                                        411,
+                                        223,
+                                        420
+                                    ],
+                                    "score": 0.78,
+                                    "content": "\\mathcal { D }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        223,
+                                        410,
+                                        291,
+                                        423
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", previous policy",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        291,
+                                        413,
+                                        299,
+                                        420
+                                    ],
+                                    "score": 0.68,
+                                    "content": "\\pi",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        316,
+                                        410,
+                                        425,
+                                        423
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Output: Optimized policy",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        426,
+                                        413,
+                                        433,
+                                        421
+                                    ],
+                                    "score": 0.48,
+                                    "content": "\\pi",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 24,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                111,
+                                423,
+                                359,
+                                435
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        111,
+                                        423,
+                                        155,
+                                        435
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1: Sample",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        156,
+                                        424,
+                                        167,
+                                        434
+                                    ],
+                                    "score": 0.8,
+                                    "content": "K",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        167,
+                                        423,
+                                        346,
+                                        435
+                                    ],
+                                    "score": 1.0,
+                                    "content": "independent datasets with replacement from",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        346,
+                                        424,
+                                        355,
+                                        433
+                                    ],
+                                    "score": 0.73,
+                                    "content": "\\mathcal { D }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        355,
+                                        423,
+                                        359,
+                                        435
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                108,
+                                432,
+                                500,
+                                451
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        108,
+                                        432,
+                                        252,
+                                        451
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2: Train bootstrapped ensemble of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        252,
+                                        437,
+                                        263,
+                                        446
+                                    ],
+                                    "score": 0.82,
+                                    "content": "K",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        263,
+                                        432,
+                                        295,
+                                        451
+                                    ],
+                                    "score": 1.0,
+                                    "content": "models",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        295,
+                                        434,
+                                        344,
+                                        449
+                                    ],
+                                    "score": 0.94,
+                                    "content": "\\{ \\widehat { T } _ { i } , \\widehat { R } _ { i } \\} _ { i = 0 } ^ { K }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        344,
+                                        432,
+                                        480,
+                                        451
+                                    ],
+                                    "score": 1.0,
+                                    "content": "minimizing Eq. (8) (adapted with",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        481,
+                                        434,
+                                        493,
+                                        451
+                                    ],
+                                    "score": 0.86,
+                                    "content": "\\tilde { d } _ { \\phi } ^ { \\pi } ,",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        494,
+                                        432,
+                                        500,
+                                        451
+                                    ],
+                                    "score": 1.0,
+                                    "content": ").",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                111,
+                                448,
+                                226,
+                                460
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        111,
+                                        448,
+                                        156,
+                                        460
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3: for rep",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        156,
+                                        448,
+                                        211,
+                                        460
+                                    ],
+                                    "score": 0.7,
+                                    "content": "{ \\mathcal { a } } t = 0 , 1 , \\ldots",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        211,
+                                        448,
+                                        226,
+                                        460
+                                    ],
+                                    "score": 1.0,
+                                    "content": "do",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                110,
+                                458,
+                                257,
+                                471
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        110,
+                                        460,
+                                        121,
+                                        471
+                                    ],
+                                    "score": 1.0,
+                                    "content": "4:",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        137,
+                                        458,
+                                        185,
+                                        471
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for rollout",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        185,
+                                        460,
+                                        243,
+                                        471
+                                    ],
+                                    "score": 0.49,
+                                    "content": "\\mathbf { \\Sigma } = 0 , 1 , \\dots , B",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        243,
+                                        458,
+                                        257,
+                                        471
+                                    ],
+                                    "score": 1.0,
+                                    "content": "do",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                110,
+                                469,
+                                307,
+                                482
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        110,
+                                        470,
+                                        121,
+                                        482
+                                    ],
+                                    "score": 1.0,
+                                    "content": "5:",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        153,
+                                        469,
+                                        261,
+                                        482
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Sample initial rollout state",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        261,
+                                        472,
+                                        271,
+                                        481
+                                    ],
+                                    "score": 0.82,
+                                    "content": "s _ { 0 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        272,
+                                        469,
+                                        294,
+                                        482
+                                    ],
+                                    "score": 1.0,
+                                    "content": "from",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        295,
+                                        471,
+                                        304,
+                                        480
+                                    ],
+                                    "score": 0.82,
+                                    "content": "\\mathcal { D }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        304,
+                                        469,
+                                        307,
+                                        482
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                110,
+                                479,
+                                232,
+                                494
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        110,
+                                        481,
+                                        121,
+                                        493
+                                    ],
+                                    "score": 1.0,
+                                    "content": "6:",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        152,
+                                        479,
+                                        168,
+                                        494
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        168,
+                                        481,
+                                        217,
+                                        492
+                                    ],
+                                    "score": 0.65,
+                                    "content": "t = 0 , 1 , \\ldots",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        218,
+                                        479,
+                                        232,
+                                        494
+                                    ],
+                                    "score": 1.0,
+                                    "content": "do",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                110,
+                                492,
+                                288,
+                                505
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        110,
+                                        492,
+                                        121,
+                                        505
+                                    ],
+                                    "score": 1.0,
+                                    "content": "7:",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        167,
+                                        492,
+                                        239,
+                                        505
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Sample an action",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        240,
+                                        492,
+                                        284,
+                                        504
+                                    ],
+                                    "score": 0.91,
+                                    "content": "a _ { t } \\sim \\pi ( s _ { t } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        285,
+                                        492,
+                                        288,
+                                        505
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                110,
+                                503,
+                                429,
+                                518
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        110,
+                                        506,
+                                        121,
+                                        518
+                                    ],
+                                    "score": 1.0,
+                                    "content": "8:",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        167,
+                                        503,
+                                        232,
+                                        518
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Randomly pick",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        232,
+                                        504,
+                                        265,
+                                        518
+                                    ],
+                                    "score": 0.91,
+                                    "content": "( \\widehat { T } _ { i } , \\widehat { R } _ { i } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        266,
+                                        503,
+                                        315,
+                                        518
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and sample",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        315,
+                                        503,
+                                        429,
+                                        518
+                                    ],
+                                    "score": 0.88,
+                                    "content": "( s _ { t + 1 } , r _ { t } ) \\sim ( \\widehat { T } _ { i } , \\widehat { R } _ { i } ) ( s _ { t } , a _ { t } ) .",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 32,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                110,
+                                517,
+                                474,
+                                537
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        110,
+                                        520,
+                                        121,
+                                        533
+                                    ],
+                                    "score": 1.0,
+                                    "content": "9:",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        167,
+                                        518,
+                                        207,
+                                        534
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Compute",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        207,
+                                        517,
+                                        343,
+                                        537
+                                    ],
+                                    "score": 0.9,
+                                    "content": " \\tilde { r } _ { t } = r _ { t } - \\gamma \\lambda \\sqrt { \\mathbb { V } _ { K } [ \\mu ( s _ { t } , a _ { t } ) ] } | | _ { 2 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        344,
+                                        518,
+                                        385,
+                                        535
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and store",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        385,
+                                        519,
+                                        450,
+                                        532
+                                    ],
+                                    "score": 0.86,
+                                    "content": "\\left( { { s _ { t } } , { a _ { t } } , \\tilde { r } _ { t } , { s _ { t + 1 } } } \\right)",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        450,
+                                        518,
+                                        461,
+                                        535
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        461,
+                                        518,
+                                        470,
+                                        530
+                                    ],
+                                    "score": 0.82,
+                                    "content": "\\widehat { \\mathcal { D } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        471,
+                                        518,
+                                        474,
+                                        535
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                534,
+                                187,
+                                547
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        534,
+                                        121,
+                                        547
+                                    ],
+                                    "score": 1.0,
+                                    "content": "10:",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        152,
+                                        534,
+                                        187,
+                                        547
+                                    ],
+                                    "score": 1.0,
+                                    "content": "end for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                544,
+                                172,
+                                558
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        546,
+                                        121,
+                                        558
+                                    ],
+                                    "score": 1.0,
+                                    "content": "11:",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        137,
+                                        544,
+                                        172,
+                                        558
+                                    ],
+                                    "score": 1.0,
+                                    "content": "end for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                555,
+                                335,
+                                571
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        557,
+                                        121,
+                                        569
+                                    ],
+                                    "score": 1.0,
+                                    "content": "12:",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        137,
+                                        555,
+                                        219,
+                                        570
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Draw samples from",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        219,
+                                        557,
+                                        228,
+                                        567
+                                    ],
+                                    "score": 0.8,
+                                    "content": "\\mathcal { D }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        229,
+                                        555,
+                                        277,
+                                        570
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to compute",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        277,
+                                        555,
+                                        332,
+                                        571
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\widehat { \\mathrm { I P M } } ( \\tilde { d } _ { \\phi } ^ { \\pi } , d _ { \\phi } ^ { D } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        333,
+                                        555,
+                                        335,
+                                        570
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                570,
+                                332,
+                                585
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        571,
+                                        121,
+                                        584
+                                    ],
+                                    "score": 1.0,
+                                    "content": "13:",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        137,
+                                        570,
+                                        219,
+                                        585
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Draw samples from",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        219,
+                                        572,
+                                        228,
+                                        582
+                                    ],
+                                    "score": 0.8,
+                                    "content": "\\mathcal { D }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        229,
+                                        570,
+                                        246,
+                                        585
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        246,
+                                        570,
+                                        256,
+                                        582
+                                    ],
+                                    "score": 0.85,
+                                    "content": "\\widehat { \\mathcal { D } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        257,
+                                        570,
+                                        319,
+                                        585
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to update critic",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        320,
+                                        572,
+                                        328,
+                                        583
+                                    ],
+                                    "score": 0.83,
+                                    "content": "Q",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        329,
+                                        570,
+                                        332,
+                                        585
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                582,
+                                473,
+                                602
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        585,
+                                        121,
+                                        597
+                                    ],
+                                    "score": 1.0,
+                                    "content": "14:",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        135,
+                                        582,
+                                        181,
+                                        602
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Maximize",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        181,
+                                        583,
+                                        420,
+                                        600
+                                    ],
+                                    "score": 0.88,
+                                    "content": "\\mathbb { E } _ { s \\sim \\mathcal { D } \\cup \\widehat { \\mathcal { D } } , a \\sim \\pi ( s ) } [ Q ( s , a ) - \\tau \\log \\pi ( a | s ) ] - \\alpha _ { \\pi } \\widehat { \\mathrm { I P M } } ( \\tilde { d } _ { \\phi } ^ { \\pi } , d _ { \\phi } ^ { \\mathcal { D } } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        420,
+                                        582,
+                                        460,
+                                        602
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to update",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        461,
+                                        587,
+                                        468,
+                                        595
+                                    ],
+                                    "score": 0.64,
+                                    "content": "\\pi",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        468,
+                                        582,
+                                        473,
+                                        602
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                596,
+                                156,
+                                606
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        597,
+                                        120,
+                                        606
+                                    ],
+                                    "score": 0.928,
+                                    "content": "15:",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        125,
+                                        596,
+                                        156,
+                                        606
+                                    ],
+                                    "score": 0.957,
+                                    "content": "end for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39,
+                            "is_list_start_line": true
+                        }
+                    ],
+                    "index": 31.5,
+                    "bbox_fs": [
+                        106,
+                        410,
+                        500,
+                        606
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        623,
+                        505,
+                        712
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                623,
+                                505,
+                                637
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        623,
+                                        505,
+                                        637
+                                    ],
+                                    "score": 1.0,
+                                    "content": "For the result of MOPO (Yu et al., 2020), we ran the code kindly provided by the authors4 without",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                634,
+                                505,
+                                648
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        634,
+                                        505,
+                                        648
+                                    ],
+                                    "score": 1.0,
+                                    "content": "any modification on the algorithm or the hyperparameters. All algorithms we experimented (Base,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                645,
+                                506,
+                                659
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        645,
+                                        506,
+                                        659
+                                    ],
+                                    "score": 1.0,
+                                    "content": "RP, RepB-SDE) share all the hyperparameters in common except the ones associated with changing",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                657,
+                                504,
+                                669
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        657,
+                                        504,
+                                        669
+                                    ],
+                                    "score": 1.0,
+                                    "content": "objectives. We run SAC on the full rollouts from the trained ensemble models as shown in Algo-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                667,
+                                505,
+                                680
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        667,
+                                        505,
+                                        680
+                                    ],
+                                    "score": 1.0,
+                                    "content": "rithm 1. The common hyperparameters shared among algorithms are shown in Table 3. We simply",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                679,
+                                505,
+                                691
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        679,
+                                        505,
+                                        691
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tried the listed hyperparameters and not tuned them further. For RP and RepB-SDE, we penalized",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                690,
+                                505,
+                                701
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        690,
+                                        505,
+                                        701
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the reward from simulated environments with the standard deviation of prediction means of neural",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                701,
+                                505,
+                                712
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        701,
+                                        505,
+                                        712
+                                    ],
+                                    "score": 1.0,
+                                    "content": "network ensembles. We used standardized output of all 7 neural networks to compute the reward",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 47
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                485,
+                                505,
+                                498
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        485,
+                                        348,
+                                        498
+                                    ],
+                                    "score": 1.0,
+                                    "content": "penalty. We first search over the reward penalty coefficient",
+                                    "type": "text",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        349,
+                                        486,
+                                        442,
+                                        498
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\lambda \\in \\{ 0 , 2 , 5 , 7 , 1 0 , 1 5 \\}",
+                                    "type": "inline_equation",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        442,
+                                        485,
+                                        505,
+                                        498
+                                    ],
+                                    "score": 1.0,
+                                    "content": "that grants RP",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                496,
+                                505,
+                                510
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        496,
+                                        267,
+                                        510
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the best performance. We shared same",
+                                    "type": "text",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        267,
+                                        497,
+                                        275,
+                                        507
+                                    ],
+                                    "score": 0.68,
+                                    "content": "\\lambda",
+                                    "type": "inline_equation",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        275,
+                                        496,
+                                        416,
+                                        510
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for RepB-SDE and searched over",
+                                    "type": "text",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        416,
+                                        497,
+                                        501,
+                                        509
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\alpha _ { M } \\in \\{ 0 . 0 1 , 0 . 1 , 1 \\}",
+                                    "type": "inline_equation",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        501,
+                                        496,
+                                        505,
+                                        510
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                507,
+                                506,
+                                521
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        508,
+                                        216,
+                                        520
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\alpha _ { \\pi } ~ \\in ~ \\{ 0 , 0 . 0 1 , 0 . 1 , 1 , 1 0 \\}",
+                                    "type": "inline_equation",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        217,
+                                        507,
+                                        340,
+                                        521
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". We ran the algorithms for",
+                                    "type": "text",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        340,
+                                        507,
+                                        383,
+                                        519
+                                    ],
+                                    "score": 0.91,
+                                    "content": "1 . 5 \\times 1 0 ^ { 6 }",
+                                    "type": "inline_equation",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        384,
+                                        507,
+                                        506,
+                                        521
+                                    ],
+                                    "score": 1.0,
+                                    "content": "gradient updates, except for",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                516,
+                                506,
+                                532
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        516,
+                                        303,
+                                        532
+                                    ],
+                                    "score": 1.0,
+                                    "content": "HalfCheetah-Medium-Expert where we ran for",
+                                    "type": "text",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        303,
+                                        519,
+                                        346,
+                                        529
+                                    ],
+                                    "score": 0.91,
+                                    "content": "5 . 0 \\times 1 0 ^ { 6 }",
+                                    "type": "inline_equation",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        346,
+                                        516,
+                                        506,
+                                        532
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". In Table 4, we present the standard",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                530,
+                                318,
+                                542
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        530,
+                                        318,
+                                        542
+                                    ],
+                                    "score": 1.0,
+                                    "content": "errors of results, which was omitted in the main text.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 13
+                        }
+                    ],
+                    "index": 43.5,
+                    "bbox_fs": [
+                        104,
+                        623,
+                        506,
+                        712
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "table",
+                    "bbox": [
+                        146,
+                        105,
+                        466,
+                        253
+                    ],
+                    "blocks": [
+                        {
+                            "type": "table_caption",
+                            "bbox": [
+                                169,
+                                80,
+                                441,
+                                92
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        169,
+                                        78,
+                                        442,
+                                        95
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                169,
+                                                78,
+                                                442,
+                                                95
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Table 3: Common hyperparameters used in offline RL experiments.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 0
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "type": "table_body",
+                            "bbox": [
+                                146,
+                                105,
+                                466,
+                                253
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        146,
+                                        105,
+                                        466,
+                                        253
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                146,
+                                                105,
+                                                466,
+                                                253
+                                            ],
+                                            "score": 0.968,
+                                            "html": "<table><tr><td>Parameter Value</td></tr><tr><td>optimizer Adam (Kingma &amp; Ba, 2014)</td></tr><tr><td>learning rate 3×10-4</td></tr><tr><td>discount factor y 0.99</td></tr><tr><td>number of samples per minibatch 256</td></tr><tr><td>target smoothing coefficient T 5×10-3</td></tr><tr><td>[actor/critic] number of hidden layers 2</td></tr><tr><td>[actor/critic] number of hidden units per layer 256</td></tr><tr><td>[actor/critic] non-linearity ReLU</td></tr><tr><td># of rollouts 104</td></tr><tr><td>max length of rollouts 103</td></tr><tr><td>rollout buffer size 5×107</td></tr></table>",
+                                            "type": "table",
+                                            "image_path": "d1bf4bbb33fafb716bba9830bdaa0eb212a3b37e4999d735955451123d3aec44.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 2,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        146,
+                                        105,
+                                        466,
+                                        154.33333333333334
+                                    ],
+                                    "spans": [],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        146,
+                                        154.33333333333334,
+                                        466,
+                                        203.66666666666669
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                },
+                                {
+                                    "bbox": [
+                                        146,
+                                        203.66666666666669,
+                                        466,
+                                        253.00000000000003
+                                    ],
+                                    "spans": [],
+                                    "index": 3
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 1.0
+                },
+                {
+                    "type": "table",
+                    "bbox": [
+                        107,
+                        303,
+                        503,
+                        460
+                    ],
+                    "blocks": [
+                        {
+                            "type": "table_caption",
+                            "bbox": [
+                                107,
+                                268,
+                                504,
+                                291
+                            ],
+                            "group_id": 1,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        268,
+                                        504,
+                                        279
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                268,
+                                                504,
+                                                279
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Table 4: Normalized scores on D4RL MuJoCo benchmark datasets (Fu et al., 2020) with standard",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 4
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        279,
+                                        194,
+                                        291
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                279,
+                                                194,
+                                                291
+                                            ],
+                                            "score": 1.0,
+                                            "content": "errors fully specified.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 5
+                                }
+                            ],
+                            "index": 4.5
+                        },
+                        {
+                            "type": "table_body",
+                            "bbox": [
+                                107,
+                                303,
+                                503,
+                                460
+                            ],
+                            "group_id": 1,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        303,
+                                        503,
+                                        460
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                107,
+                                                303,
+                                                503,
+                                                460
+                                            ],
+                                            "score": 0.976,
+                                            "html": "<table><tr><td>Dataset type</td><td>Environment</td><td>RepB-SDE (ours)</td><td>RP</td><td>Base</td><td>MOPO</td></tr><tr><td></td><td></td><td></td><td></td><td></td><td></td></tr><tr><td>Random</td><td>Walker2d-v2</td><td>21.1 ± 1.0</td><td>18.4 ±1.8</td><td>16.4 ± 2.2</td><td>1.3 ± 0.5</td></tr><tr><td>Medium</td><td>Walker2d-v2</td><td>72.1 ± 1.9</td><td>56.3 ± 4.3</td><td>5.5 ± 0.4</td><td>-0.1 ± 0.0</td></tr><tr><td>Med-Replay</td><td>Walker2d-v2</td><td>49.8 ± 11.4</td><td>41.1 ±10.3</td><td>6.2 ± 0.7</td><td>47.8 ± 4.8</td></tr><tr><td>Med-Expert</td><td>Walker2d-v2</td><td>88.8 ± 6.9</td><td>72.6 ± 5.1</td><td>51.7 ± 7.1</td><td>32.4 ± 8.8</td></tr><tr><td>Random</td><td>Hopper-v2</td><td>8.6 ±1.0</td><td>8.3 ±0.2</td><td>8.3 ±0.2</td><td>9.1 ± 0.3</td></tr><tr><td>Medium</td><td>Hopper-v2</td><td>34.0 ± 2.8</td><td>27.5 ± 3.1</td><td>19.8 ± 0.8</td><td>19.2 ± 4.5</td></tr><tr><td>Med-Replay</td><td>Hopper-v2</td><td>62.2 ± 6.7</td><td>49.8 ± 8.7</td><td>32.9 ± 5.5</td><td>80.8 ± 2.9</td></tr><tr><td>Med-Expert</td><td>Hopper-v2</td><td>82.6 ± 7.0</td><td>74.0 ± 7.2</td><td>19.1 ± 1.1</td><td>23.2 ± 1.3</td></tr><tr><td>Random</td><td>HalfCheetah-v2</td><td>32.9 ± 1.1</td><td>31.3 ±1.5</td><td>26.1 ± 6.2</td><td>29.9 ± 1.2</td></tr><tr><td>Medium</td><td>HalfCheetah-v2</td><td>49.1 ± 0.3</td><td>47.3 ± 0.3</td><td>22.1 ± 4.4</td><td>45.0 ± 0.4</td></tr><tr><td>Med-Replay</td><td>HalfCheetah-v2</td><td>57.5 ± 0.8</td><td>53.6 ± 0.8</td><td>54.0 ± 2.4</td><td>53.8 ± 1.1</td></tr><tr><td>Med-Expert</td><td>HalfCheetah-v2</td><td>55.4 ± 8.3</td><td>36.7 ± 12.0</td><td>31.9 ± 10.5</td><td>91.0 ± 2.5</td></tr></table>",
+                                            "type": "table",
+                                            "image_path": "3284e1e9c9ccda068f8a8ccbf08fd85271914a97e4b455c2e3669098579f8a52.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 7,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        303,
+                                        503,
+                                        355.3333333333333
+                                    ],
+                                    "spans": [],
+                                    "index": 6
+                                },
+                                {
+                                    "bbox": [
+                                        107,
+                                        355.3333333333333,
+                                        503,
+                                        407.66666666666663
+                                    ],
+                                    "spans": [],
+                                    "index": 7
+                                },
+                                {
+                                    "bbox": [
+                                        107,
+                                        407.66666666666663,
+                                        503,
+                                        459.99999999999994
+                                    ],
+                                    "spans": [],
+                                    "index": 8
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 5.75
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        486,
+                        505,
+                        541
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                485,
+                                505,
+                                498
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        485,
+                                        348,
+                                        498
+                                    ],
+                                    "score": 1.0,
+                                    "content": "penalty. We first search over the reward penalty coefficient",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        349,
+                                        486,
+                                        442,
+                                        498
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\lambda \\in \\{ 0 , 2 , 5 , 7 , 1 0 , 1 5 \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        442,
+                                        485,
+                                        505,
+                                        498
+                                    ],
+                                    "score": 1.0,
+                                    "content": "that grants RP",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                496,
+                                505,
+                                510
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        496,
+                                        267,
+                                        510
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the best performance. We shared same",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        267,
+                                        497,
+                                        275,
+                                        507
+                                    ],
+                                    "score": 0.68,
+                                    "content": "\\lambda",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        275,
+                                        496,
+                                        416,
+                                        510
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for RepB-SDE and searched over",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        416,
+                                        497,
+                                        501,
+                                        509
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\alpha _ { M } \\in \\{ 0 . 0 1 , 0 . 1 , 1 \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        501,
+                                        496,
+                                        505,
+                                        510
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                507,
+                                506,
+                                521
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        508,
+                                        216,
+                                        520
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\alpha _ { \\pi } ~ \\in ~ \\{ 0 , 0 . 0 1 , 0 . 1 , 1 , 1 0 \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        217,
+                                        507,
+                                        340,
+                                        521
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". We ran the algorithms for",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        340,
+                                        507,
+                                        383,
+                                        519
+                                    ],
+                                    "score": 0.91,
+                                    "content": "1 . 5 \\times 1 0 ^ { 6 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        384,
+                                        507,
+                                        506,
+                                        521
+                                    ],
+                                    "score": 1.0,
+                                    "content": "gradient updates, except for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                516,
+                                506,
+                                532
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        516,
+                                        303,
+                                        532
+                                    ],
+                                    "score": 1.0,
+                                    "content": "HalfCheetah-Medium-Expert where we ran for",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        303,
+                                        519,
+                                        346,
+                                        529
+                                    ],
+                                    "score": 0.91,
+                                    "content": "5 . 0 \\times 1 0 ^ { 6 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        346,
+                                        516,
+                                        506,
+                                        532
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". In Table 4, we present the standard",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                530,
+                                318,
+                                542
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        530,
+                                        318,
+                                        542
+                                    ],
+                                    "score": 1.0,
+                                    "content": "errors of results, which was omitted in the main text.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        }
+                    ],
+                    "index": 11
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        547,
+                        505,
+                        591
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                547,
+                                504,
+                                559
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        547,
+                                        504,
+                                        559
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In addition to the performance after the final iteration presented in Table 1, we also present the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                558,
+                                505,
+                                569
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        558,
+                                        408,
+                                        569
+                                    ],
+                                    "score": 1.0,
+                                    "content": "learning curves of the experiment in Figure 3, and the effect of the choice of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        409,
+                                        559,
+                                        425,
+                                        569
+                                    ],
+                                    "score": 0.88,
+                                    "content": "\\alpha _ { M }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        425,
+                                        558,
+                                        443,
+                                        569
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        443,
+                                        559,
+                                        456,
+                                        569
+                                    ],
+                                    "score": 0.86,
+                                    "content": "\\alpha _ { \\pi }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        456,
+                                        558,
+                                        505,
+                                        569
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in Figure 4.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                568,
+                                505,
+                                582
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        568,
+                                        427,
+                                        582
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The Figure 4 shows that the presented regularization is robust to the choice of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        427,
+                                        570,
+                                        460,
+                                        580
+                                    ],
+                                    "score": 0.89,
+                                    "content": "\\alpha _ { M } , \\alpha _ { \\pi }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        460,
+                                        568,
+                                        505,
+                                        582
+                                    ],
+                                    "score": 1.0,
+                                    "content": "except for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                579,
+                                460,
+                                592
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        579,
+                                        144,
+                                        592
+                                    ],
+                                    "score": 1.0,
+                                    "content": "too large",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        144,
+                                        582,
+                                        160,
+                                        591
+                                    ],
+                                    "score": 0.87,
+                                    "content": "\\alpha _ { M }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        160,
+                                        579,
+                                        386,
+                                        592
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", consistently improving from RP, which corresponds to",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        387,
+                                        580,
+                                        455,
+                                        591
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\alpha _ { M } = 0 , \\alpha _ { \\pi } = 0",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        455,
+                                        579,
+                                        460,
+                                        592
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        }
+                    ],
+                    "index": 15.5
+                }
+            ],
+            "page_idx": 20,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        107,
+                        27,
+                        292,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                26,
+                                293,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        26,
+                                        293,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Published as a conference paper at ICLR 2021",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        300,
+                        751,
+                        310,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                298,
+                                750,
+                                312,
+                                764
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        298,
+                                        750,
+                                        312,
+                                        764
+                                    ],
+                                    "score": 1.0,
+                                    "content": "",
+                                    "type": "text",
+                                    "height": 14,
+                                    "width": 14
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "table",
+                    "bbox": [
+                        146,
+                        105,
+                        466,
+                        253
+                    ],
+                    "blocks": [
+                        {
+                            "type": "table_caption",
+                            "bbox": [
+                                169,
+                                80,
+                                441,
+                                92
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        169,
+                                        78,
+                                        442,
+                                        95
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                169,
+                                                78,
+                                                442,
+                                                95
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Table 3: Common hyperparameters used in offline RL experiments.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 0
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "type": "table_body",
+                            "bbox": [
+                                146,
+                                105,
+                                466,
+                                253
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        146,
+                                        105,
+                                        466,
+                                        253
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                146,
+                                                105,
+                                                466,
+                                                253
+                                            ],
+                                            "score": 0.968,
+                                            "html": "<table><tr><td>Parameter Value</td></tr><tr><td>optimizer Adam (Kingma &amp; Ba, 2014)</td></tr><tr><td>learning rate 3×10-4</td></tr><tr><td>discount factor y 0.99</td></tr><tr><td>number of samples per minibatch 256</td></tr><tr><td>target smoothing coefficient T 5×10-3</td></tr><tr><td>[actor/critic] number of hidden layers 2</td></tr><tr><td>[actor/critic] number of hidden units per layer 256</td></tr><tr><td>[actor/critic] non-linearity ReLU</td></tr><tr><td># of rollouts 104</td></tr><tr><td>max length of rollouts 103</td></tr><tr><td>rollout buffer size 5×107</td></tr></table>",
+                                            "type": "table",
+                                            "image_path": "d1bf4bbb33fafb716bba9830bdaa0eb212a3b37e4999d735955451123d3aec44.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 2,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        146,
+                                        105,
+                                        466,
+                                        154.33333333333334
+                                    ],
+                                    "spans": [],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        146,
+                                        154.33333333333334,
+                                        466,
+                                        203.66666666666669
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                },
+                                {
+                                    "bbox": [
+                                        146,
+                                        203.66666666666669,
+                                        466,
+                                        253.00000000000003
+                                    ],
+                                    "spans": [],
+                                    "index": 3
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 1.0
+                },
+                {
+                    "type": "table",
+                    "bbox": [
+                        107,
+                        303,
+                        503,
+                        460
+                    ],
+                    "blocks": [
+                        {
+                            "type": "table_caption",
+                            "bbox": [
+                                107,
+                                268,
+                                504,
+                                291
+                            ],
+                            "group_id": 1,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        268,
+                                        504,
+                                        279
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                268,
+                                                504,
+                                                279
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Table 4: Normalized scores on D4RL MuJoCo benchmark datasets (Fu et al., 2020) with standard",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 4
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        279,
+                                        194,
+                                        291
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                279,
+                                                194,
+                                                291
+                                            ],
+                                            "score": 1.0,
+                                            "content": "errors fully specified.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 5
+                                }
+                            ],
+                            "index": 4.5
+                        },
+                        {
+                            "type": "table_body",
+                            "bbox": [
+                                107,
+                                303,
+                                503,
+                                460
+                            ],
+                            "group_id": 1,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        303,
+                                        503,
+                                        460
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                107,
+                                                303,
+                                                503,
+                                                460
+                                            ],
+                                            "score": 0.976,
+                                            "html": "<table><tr><td>Dataset type</td><td>Environment</td><td>RepB-SDE (ours)</td><td>RP</td><td>Base</td><td>MOPO</td></tr><tr><td></td><td></td><td></td><td></td><td></td><td></td></tr><tr><td>Random</td><td>Walker2d-v2</td><td>21.1 ± 1.0</td><td>18.4 ±1.8</td><td>16.4 ± 2.2</td><td>1.3 ± 0.5</td></tr><tr><td>Medium</td><td>Walker2d-v2</td><td>72.1 ± 1.9</td><td>56.3 ± 4.3</td><td>5.5 ± 0.4</td><td>-0.1 ± 0.0</td></tr><tr><td>Med-Replay</td><td>Walker2d-v2</td><td>49.8 ± 11.4</td><td>41.1 ±10.3</td><td>6.2 ± 0.7</td><td>47.8 ± 4.8</td></tr><tr><td>Med-Expert</td><td>Walker2d-v2</td><td>88.8 ± 6.9</td><td>72.6 ± 5.1</td><td>51.7 ± 7.1</td><td>32.4 ± 8.8</td></tr><tr><td>Random</td><td>Hopper-v2</td><td>8.6 ±1.0</td><td>8.3 ±0.2</td><td>8.3 ±0.2</td><td>9.1 ± 0.3</td></tr><tr><td>Medium</td><td>Hopper-v2</td><td>34.0 ± 2.8</td><td>27.5 ± 3.1</td><td>19.8 ± 0.8</td><td>19.2 ± 4.5</td></tr><tr><td>Med-Replay</td><td>Hopper-v2</td><td>62.2 ± 6.7</td><td>49.8 ± 8.7</td><td>32.9 ± 5.5</td><td>80.8 ± 2.9</td></tr><tr><td>Med-Expert</td><td>Hopper-v2</td><td>82.6 ± 7.0</td><td>74.0 ± 7.2</td><td>19.1 ± 1.1</td><td>23.2 ± 1.3</td></tr><tr><td>Random</td><td>HalfCheetah-v2</td><td>32.9 ± 1.1</td><td>31.3 ±1.5</td><td>26.1 ± 6.2</td><td>29.9 ± 1.2</td></tr><tr><td>Medium</td><td>HalfCheetah-v2</td><td>49.1 ± 0.3</td><td>47.3 ± 0.3</td><td>22.1 ± 4.4</td><td>45.0 ± 0.4</td></tr><tr><td>Med-Replay</td><td>HalfCheetah-v2</td><td>57.5 ± 0.8</td><td>53.6 ± 0.8</td><td>54.0 ± 2.4</td><td>53.8 ± 1.1</td></tr><tr><td>Med-Expert</td><td>HalfCheetah-v2</td><td>55.4 ± 8.3</td><td>36.7 ± 12.0</td><td>31.9 ± 10.5</td><td>91.0 ± 2.5</td></tr></table>",
+                                            "type": "table",
+                                            "image_path": "3284e1e9c9ccda068f8a8ccbf08fd85271914a97e4b455c2e3669098579f8a52.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 7,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        303,
+                                        503,
+                                        355.3333333333333
+                                    ],
+                                    "spans": [],
+                                    "index": 6
+                                },
+                                {
+                                    "bbox": [
+                                        107,
+                                        355.3333333333333,
+                                        503,
+                                        407.66666666666663
+                                    ],
+                                    "spans": [],
+                                    "index": 7
+                                },
+                                {
+                                    "bbox": [
+                                        107,
+                                        407.66666666666663,
+                                        503,
+                                        459.99999999999994
+                                    ],
+                                    "spans": [],
+                                    "index": 8
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 5.75
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        486,
+                        505,
+                        541
+                    ],
+                    "lines": [],
+                    "index": 11,
+                    "bbox_fs": [
+                        105,
+                        485,
+                        506,
+                        542
+                    ],
+                    "lines_deleted": true
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        547,
+                        505,
+                        591
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                547,
+                                504,
+                                559
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        547,
+                                        504,
+                                        559
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In addition to the performance after the final iteration presented in Table 1, we also present the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                558,
+                                505,
+                                569
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        558,
+                                        408,
+                                        569
+                                    ],
+                                    "score": 1.0,
+                                    "content": "learning curves of the experiment in Figure 3, and the effect of the choice of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        409,
+                                        559,
+                                        425,
+                                        569
+                                    ],
+                                    "score": 0.88,
+                                    "content": "\\alpha _ { M }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        425,
+                                        558,
+                                        443,
+                                        569
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        443,
+                                        559,
+                                        456,
+                                        569
+                                    ],
+                                    "score": 0.86,
+                                    "content": "\\alpha _ { \\pi }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        456,
+                                        558,
+                                        505,
+                                        569
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in Figure 4.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                568,
+                                505,
+                                582
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        568,
+                                        427,
+                                        582
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The Figure 4 shows that the presented regularization is robust to the choice of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        427,
+                                        570,
+                                        460,
+                                        580
+                                    ],
+                                    "score": 0.89,
+                                    "content": "\\alpha _ { M } , \\alpha _ { \\pi }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        460,
+                                        568,
+                                        505,
+                                        582
+                                    ],
+                                    "score": 1.0,
+                                    "content": "except for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                579,
+                                460,
+                                592
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        579,
+                                        144,
+                                        592
+                                    ],
+                                    "score": 1.0,
+                                    "content": "too large",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        144,
+                                        582,
+                                        160,
+                                        591
+                                    ],
+                                    "score": 0.87,
+                                    "content": "\\alpha _ { M }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        160,
+                                        579,
+                                        386,
+                                        592
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", consistently improving from RP, which corresponds to",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        387,
+                                        580,
+                                        455,
+                                        591
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\alpha _ { M } = 0 , \\alpha _ { \\pi } = 0",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        455,
+                                        579,
+                                        460,
+                                        592
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        }
+                    ],
+                    "index": 15.5,
+                    "bbox_fs": [
+                        105,
+                        547,
+                        505,
+                        592
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "image",
+                    "bbox": [
+                        106,
+                        100,
+                        503,
+                        425
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                106,
+                                100,
+                                503,
+                                425
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        100,
+                                        503,
+                                        425
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                100,
+                                                503,
+                                                425
+                                            ],
+                                            "score": 0.973,
+                                            "type": "image",
+                                            "image_path": "6da69fd1b89d4ec41b0f3dfa3d8233be66aedb17d085f5919e4c11e946c54cdc.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 1,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        100,
+                                        503,
+                                        208.33333333333331
+                                    ],
+                                    "spans": [],
+                                    "index": 0
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        208.33333333333331,
+                                        503,
+                                        316.66666666666663
+                                    ],
+                                    "spans": [],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        316.66666666666663,
+                                        503,
+                                        424.99999999999994
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                193,
+                                432,
+                                416,
+                                444
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        193,
+                                        431,
+                                        417,
+                                        446
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                193,
+                                                431,
+                                                417,
+                                                446
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 3: The learning curves of the D4RL experiment.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 3
+                                }
+                            ],
+                            "index": 3
+                        }
+                    ],
+                    "index": 2.0
+                },
+                {
+                    "type": "image",
+                    "bbox": [
+                        105,
+                        493,
+                        506,
+                        688
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                105,
+                                493,
+                                506,
+                                688
+                            ],
+                            "group_id": 1,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        493,
+                                        506,
+                                        688
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                493,
+                                                506,
+                                                688
+                                            ],
+                                            "score": 0.961,
+                                            "type": "image",
+                                            "image_path": "bcf6e9d1e6e698ee7c20d46468f932a177986219b3341a07904605f54aa078ee.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 5,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        493,
+                                        506,
+                                        558.0
+                                    ],
+                                    "spans": [],
+                                    "index": 4
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        558.0,
+                                        506,
+                                        623.0
+                                    ],
+                                    "spans": [],
+                                    "index": 5
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        623.0,
+                                        506,
+                                        688.0
+                                    ],
+                                    "spans": [],
+                                    "index": 6
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                159,
+                                695,
+                                451,
+                                708
+                            ],
+                            "group_id": 1,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        159,
+                                        695,
+                                        452,
+                                        709
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                159,
+                                                695,
+                                                452,
+                                                709
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 4: The effect of varying hyperparameter in the D4RL experiment.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 7
+                                }
+                            ],
+                            "index": 7
+                        }
+                    ],
+                    "index": 6.0
+                }
+            ],
+            "page_idx": 21,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        107,
+                        27,
+                        292,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                26,
+                                293,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        26,
+                                        293,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Published as a conference paper at ICLR 2021",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        300,
+                        751,
+                        311,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                298,
+                                750,
+                                313,
+                                765
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        298,
+                                        750,
+                                        313,
+                                        765
+                                    ],
+                                    "score": 1.0,
+                                    "content": "",
+                                    "type": "text",
+                                    "height": 15,
+                                    "width": 15
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "image",
+                    "bbox": [
+                        106,
+                        100,
+                        503,
+                        425
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                106,
+                                100,
+                                503,
+                                425
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        100,
+                                        503,
+                                        425
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                100,
+                                                503,
+                                                425
+                                            ],
+                                            "score": 0.973,
+                                            "type": "image",
+                                            "image_path": "6da69fd1b89d4ec41b0f3dfa3d8233be66aedb17d085f5919e4c11e946c54cdc.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 1,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        100,
+                                        503,
+                                        208.33333333333331
+                                    ],
+                                    "spans": [],
+                                    "index": 0
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        208.33333333333331,
+                                        503,
+                                        316.66666666666663
+                                    ],
+                                    "spans": [],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        316.66666666666663,
+                                        503,
+                                        424.99999999999994
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                193,
+                                432,
+                                416,
+                                444
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        193,
+                                        431,
+                                        417,
+                                        446
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                193,
+                                                431,
+                                                417,
+                                                446
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 3: The learning curves of the D4RL experiment.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 3
+                                }
+                            ],
+                            "index": 3
+                        }
+                    ],
+                    "index": 2.0
+                },
+                {
+                    "type": "image",
+                    "bbox": [
+                        105,
+                        493,
+                        506,
+                        688
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                105,
+                                493,
+                                506,
+                                688
+                            ],
+                            "group_id": 1,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        493,
+                                        506,
+                                        688
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                493,
+                                                506,
+                                                688
+                                            ],
+                                            "score": 0.961,
+                                            "type": "image",
+                                            "image_path": "bcf6e9d1e6e698ee7c20d46468f932a177986219b3341a07904605f54aa078ee.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 5,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        493,
+                                        506,
+                                        558.0
+                                    ],
+                                    "spans": [],
+                                    "index": 4
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        558.0,
+                                        506,
+                                        623.0
+                                    ],
+                                    "spans": [],
+                                    "index": 5
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        623.0,
+                                        506,
+                                        688.0
+                                    ],
+                                    "spans": [],
+                                    "index": 6
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                159,
+                                695,
+                                451,
+                                708
+                            ],
+                            "group_id": 1,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        159,
+                                        695,
+                                        452,
+                                        709
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                159,
+                                                695,
+                                                452,
+                                                709
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 4: The effect of varying hyperparameter in the D4RL experiment.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 7
+                                }
+                            ],
+                            "index": 7
+                        }
+                    ],
+                    "index": 6.0
+                }
+            ]
+        }
+    ],
+    "_backend": "pipeline",
+    "_version_name": "2.2.2"
+}
\ No newline at end of file
diff --git a/parse/train/QpNz8r_Ri2Y/QpNz8r_Ri2Y_model.json b/parse/train/QpNz8r_Ri2Y/QpNz8r_Ri2Y_model.json
new file mode 100644
index 0000000000000000000000000000000000000000..d45a84c56c5c27fbd965e3ef35ca11aea5b19f10
--- /dev/null
+++ b/parse/train/QpNz8r_Ri2Y/QpNz8r_Ri2Y_model.json
@@ -0,0 +1,32393 @@
+[
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    398,
+                    688,
+                    1302,
+                    688,
+                    1302,
+                    1144,
+                    398,
+                    1144
+                ],
+                "score": 0.983
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1514,
+                    1403,
+                    1514,
+                    1403,
+                    1726,
+                    298,
+                    1726
+                ],
+                "score": 0.98
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1283,
+                    1403,
+                    1283,
+                    1403,
+                    1498,
+                    298,
+                    1498
+                ],
+                "score": 0.98
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1742,
+                    1402,
+                    1742,
+                    1402,
+                    1958,
+                    298,
+                    1958
+                ],
+                "score": 0.979
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    298,
+                    220,
+                    1400,
+                    220,
+                    1400,
+                    324,
+                    298,
+                    324
+                ],
+                "score": 0.957
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1972,
+                    1398,
+                    1972,
+                    1398,
+                    2034,
+                    297,
+                    2034
+                ],
+                "score": 0.937
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    302,
+                    1211,
+                    573,
+                    1211,
+                    573,
+                    1246,
+                    302,
+                    1246
+                ],
+                "score": 0.89
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    773,
+                    616,
+                    927,
+                    616,
+                    927,
+                    649,
+                    773,
+                    649
+                ],
+                "score": 0.881
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    299,
+                    75,
+                    813,
+                    75,
+                    813,
+                    104,
+                    299,
+                    104
+                ],
+                "score": 0.877
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    312,
+                    411,
+                    1073,
+                    411,
+                    1073,
+                    536,
+                    312,
+                    536
+                ],
+                "score": 0.782
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    318,
+                    372,
+                    957,
+                    372,
+                    957,
+                    408,
+                    318,
+                    408
+                ],
+                "score": 0.754
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    842,
+                    2088,
+                    857,
+                    2088,
+                    857,
+                    2112,
+                    842,
+                    2112
+                ],
+                "score": 0.621
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    870,
+                    374,
+                    957,
+                    374,
+                    957,
+                    406,
+                    870,
+                    406
+                ],
+                "score": 0.49,
+                "latex": "\\mathbf { K i m } ^ { 1 , 2 }"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    218.0,
+                    932.0,
+                    218.0,
+                    932.0,
+                    270.0,
+                    296.0,
+                    270.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    298.0,
+                    277.0,
+                    1404.0,
+                    277.0,
+                    1404.0,
+                    326.0,
+                    298.0,
+                    326.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1208.0,
+                    579.0,
+                    1208.0,
+                    579.0,
+                    1255.0,
+                    294.0,
+                    1255.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    768.0,
+                    612.0,
+                    934.0,
+                    612.0,
+                    934.0,
+                    655.0,
+                    768.0,
+                    655.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    72.0,
+                    815.0,
+                    72.0,
+                    815.0,
+                    108.0,
+                    295.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    841.0,
+                    2088.0,
+                    858.0,
+                    2088.0,
+                    858.0,
+                    2116.0,
+                    841.0,
+                    2116.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    395.0,
+                    686.0,
+                    1304.0,
+                    686.0,
+                    1304.0,
+                    723.0,
+                    395.0,
+                    723.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    393.0,
+                    719.0,
+                    1307.0,
+                    719.0,
+                    1307.0,
+                    753.0,
+                    393.0,
+                    753.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    392.0,
+                    748.0,
+                    1308.0,
+                    748.0,
+                    1308.0,
+                    784.0,
+                    392.0,
+                    784.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    393.0,
+                    781.0,
+                    1306.0,
+                    781.0,
+                    1306.0,
+                    813.0,
+                    393.0,
+                    813.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    392.0,
+                    809.0,
+                    1306.0,
+                    809.0,
+                    1306.0,
+                    845.0,
+                    392.0,
+                    845.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    393.0,
+                    840.0,
+                    1305.0,
+                    840.0,
+                    1305.0,
+                    874.0,
+                    393.0,
+                    874.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    393.0,
+                    871.0,
+                    1304.0,
+                    871.0,
+                    1304.0,
+                    907.0,
+                    393.0,
+                    907.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    392.0,
+                    901.0,
+                    1304.0,
+                    901.0,
+                    1304.0,
+                    936.0,
+                    392.0,
+                    936.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    392.0,
+                    930.0,
+                    1308.0,
+                    930.0,
+                    1308.0,
+                    967.0,
+                    392.0,
+                    967.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    394.0,
+                    963.0,
+                    1306.0,
+                    963.0,
+                    1306.0,
+                    998.0,
+                    394.0,
+                    998.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    395.0,
+                    994.0,
+                    1305.0,
+                    994.0,
+                    1305.0,
+                    1025.0,
+                    395.0,
+                    1025.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    395.0,
+                    1024.0,
+                    1305.0,
+                    1024.0,
+                    1305.0,
+                    1056.0,
+                    395.0,
+                    1056.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    392.0,
+                    1052.0,
+                    1306.0,
+                    1052.0,
+                    1306.0,
+                    1092.0,
+                    392.0,
+                    1092.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    394.0,
+                    1085.0,
+                    1307.0,
+                    1085.0,
+                    1307.0,
+                    1118.0,
+                    394.0,
+                    1118.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    393.0,
+                    1115.0,
+                    944.0,
+                    1115.0,
+                    944.0,
+                    1147.0,
+                    393.0,
+                    1147.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1512.0,
+                    1405.0,
+                    1512.0,
+                    1405.0,
+                    1548.0,
+                    293.0,
+                    1548.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1541.0,
+                    1405.0,
+                    1541.0,
+                    1405.0,
+                    1579.0,
+                    292.0,
+                    1579.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1574.0,
+                    1405.0,
+                    1574.0,
+                    1405.0,
+                    1608.0,
+                    294.0,
+                    1608.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1604.0,
+                    1408.0,
+                    1604.0,
+                    1408.0,
+                    1640.0,
+                    293.0,
+                    1640.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1631.0,
+                    1404.0,
+                    1631.0,
+                    1404.0,
+                    1671.0,
+                    292.0,
+                    1671.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1664.0,
+                    1407.0,
+                    1664.0,
+                    1407.0,
+                    1700.0,
+                    292.0,
+                    1700.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1696.0,
+                    897.0,
+                    1696.0,
+                    897.0,
+                    1730.0,
+                    296.0,
+                    1730.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1283.0,
+                    1405.0,
+                    1283.0,
+                    1405.0,
+                    1318.0,
+                    294.0,
+                    1318.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1315.0,
+                    1403.0,
+                    1315.0,
+                    1403.0,
+                    1346.0,
+                    296.0,
+                    1346.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1344.0,
+                    1405.0,
+                    1344.0,
+                    1405.0,
+                    1379.0,
+                    293.0,
+                    1379.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1375.0,
+                    1404.0,
+                    1375.0,
+                    1404.0,
+                    1409.0,
+                    293.0,
+                    1409.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1405.0,
+                    1405.0,
+                    1405.0,
+                    1405.0,
+                    1440.0,
+                    294.0,
+                    1440.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1436.0,
+                    1405.0,
+                    1436.0,
+                    1405.0,
+                    1470.0,
+                    294.0,
+                    1470.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1468.0,
+                    1063.0,
+                    1468.0,
+                    1063.0,
+                    1499.0,
+                    296.0,
+                    1499.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1743.0,
+                    1404.0,
+                    1743.0,
+                    1404.0,
+                    1777.0,
+                    294.0,
+                    1777.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1775.0,
+                    1404.0,
+                    1775.0,
+                    1404.0,
+                    1807.0,
+                    293.0,
+                    1807.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1803.0,
+                    1406.0,
+                    1803.0,
+                    1406.0,
+                    1840.0,
+                    291.0,
+                    1840.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1834.0,
+                    1406.0,
+                    1834.0,
+                    1406.0,
+                    1869.0,
+                    294.0,
+                    1869.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1864.0,
+                    1407.0,
+                    1864.0,
+                    1407.0,
+                    1899.0,
+                    292.0,
+                    1899.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1896.0,
+                    1406.0,
+                    1896.0,
+                    1406.0,
+                    1930.0,
+                    294.0,
+                    1930.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1925.0,
+                    1199.0,
+                    1925.0,
+                    1199.0,
+                    1961.0,
+                    293.0,
+                    1961.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1970.0,
+                    1403.0,
+                    1970.0,
+                    1403.0,
+                    2007.0,
+                    293.0,
+                    2007.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    2002.0,
+                    1405.0,
+                    2002.0,
+                    1405.0,
+                    2037.0,
+                    295.0,
+                    2037.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    312.0,
+                    403.0,
+                    987.0,
+                    403.0,
+                    987.0,
+                    445.0,
+                    312.0,
+                    445.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    308.0,
+                    435.0,
+                    1000.0,
+                    435.0,
+                    1000.0,
+                    478.0,
+                    308.0,
+                    478.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    310.0,
+                    467.0,
+                    806.0,
+                    467.0,
+                    806.0,
+                    511.0,
+                    310.0,
+                    511.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    314.0,
+                    504.0,
+                    1080.0,
+                    504.0,
+                    1080.0,
+                    542.0,
+                    314.0,
+                    542.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    316.0,
+                    370.0,
+                    869.0,
+                    370.0,
+                    869.0,
+                    414.0,
+                    316.0,
+                    414.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    958.0,
+                    370.0,
+                    962.0,
+                    370.0,
+                    962.0,
+                    414.0,
+                    958.0,
+                    414.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 0,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    460,
+                    1404,
+                    460,
+                    1404,
+                    796,
+                    298,
+                    796
+                ],
+                "score": 0.983
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1589,
+                    1404,
+                    1589,
+                    1404,
+                    1866,
+                    298,
+                    1866
+                ],
+                "score": 0.983
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1210,
+                    1404,
+                    1210,
+                    1404,
+                    1548,
+                    298,
+                    1548
+                ],
+                "score": 0.981
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    924,
+                    1404,
+                    924,
+                    1404,
+                    1169,
+                    298,
+                    1169
+                ],
+                "score": 0.981
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    229,
+                    1403,
+                    229,
+                    1403,
+                    445,
+                    298,
+                    445
+                ],
+                "score": 0.979
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    1881,
+                    1401,
+                    1881,
+                    1401,
+                    2034,
+                    299,
+                    2034
+                ],
+                "score": 0.976
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    301,
+                    849,
+                    581,
+                    849,
+                    581,
+                    884,
+                    301,
+                    884
+                ],
+                "score": 0.903
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    300,
+                    76,
+                    812,
+                    76,
+                    812,
+                    104,
+                    300,
+                    104
+                ],
+                "score": 0.895
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    841,
+                    2088,
+                    858,
+                    2088,
+                    858,
+                    2112,
+                    841,
+                    2112
+                ],
+                "score": 0.695
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    841,
+                    2088,
+                    859,
+                    2088,
+                    859,
+                    2112,
+                    841,
+                    2112
+                ],
+                "score": 0.188
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1190,
+                    1365,
+                    1263,
+                    1365,
+                    1263,
+                    1394,
+                    1190,
+                    1394
+                ],
+                "score": 0.71,
+                "latex": "2 0 1 9 \\mathrm { a }"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    846.0,
+                    586.0,
+                    846.0,
+                    586.0,
+                    890.0,
+                    292.0,
+                    890.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    73.0,
+                    815.0,
+                    73.0,
+                    815.0,
+                    108.0,
+                    295.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    839.0,
+                    2085.0,
+                    862.0,
+                    2085.0,
+                    862.0,
+                    2120.0,
+                    839.0,
+                    2120.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    838.0,
+                    2085.0,
+                    862.0,
+                    2085.0,
+                    862.0,
+                    2120.0,
+                    838.0,
+                    2120.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    458.0,
+                    1404.0,
+                    458.0,
+                    1404.0,
+                    495.0,
+                    295.0,
+                    495.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    489.0,
+                    1404.0,
+                    489.0,
+                    1404.0,
+                    525.0,
+                    293.0,
+                    525.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    521.0,
+                    1406.0,
+                    521.0,
+                    1406.0,
+                    555.0,
+                    293.0,
+                    555.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    550.0,
+                    1405.0,
+                    550.0,
+                    1405.0,
+                    588.0,
+                    293.0,
+                    588.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    583.0,
+                    1406.0,
+                    583.0,
+                    1406.0,
+                    618.0,
+                    294.0,
+                    618.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    613.0,
+                    1405.0,
+                    613.0,
+                    1405.0,
+                    648.0,
+                    293.0,
+                    648.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    644.0,
+                    1404.0,
+                    644.0,
+                    1404.0,
+                    675.0,
+                    296.0,
+                    675.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    673.0,
+                    1407.0,
+                    673.0,
+                    1407.0,
+                    710.0,
+                    292.0,
+                    710.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    702.0,
+                    1404.0,
+                    702.0,
+                    1404.0,
+                    738.0,
+                    293.0,
+                    738.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    733.0,
+                    1405.0,
+                    733.0,
+                    1405.0,
+                    770.0,
+                    293.0,
+                    770.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    767.0,
+                    433.0,
+                    767.0,
+                    433.0,
+                    795.0,
+                    296.0,
+                    795.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1589.0,
+                    1402.0,
+                    1589.0,
+                    1402.0,
+                    1625.0,
+                    294.0,
+                    1625.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1615.0,
+                    1405.0,
+                    1615.0,
+                    1405.0,
+                    1659.0,
+                    292.0,
+                    1659.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1649.0,
+                    1404.0,
+                    1649.0,
+                    1404.0,
+                    1688.0,
+                    292.0,
+                    1688.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1679.0,
+                    1405.0,
+                    1679.0,
+                    1405.0,
+                    1718.0,
+                    293.0,
+                    1718.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1712.0,
+                    1406.0,
+                    1712.0,
+                    1406.0,
+                    1748.0,
+                    294.0,
+                    1748.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1738.0,
+                    1406.0,
+                    1738.0,
+                    1406.0,
+                    1780.0,
+                    291.0,
+                    1780.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1768.0,
+                    1406.0,
+                    1768.0,
+                    1406.0,
+                    1811.0,
+                    292.0,
+                    1811.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1799.0,
+                    1404.0,
+                    1799.0,
+                    1404.0,
+                    1843.0,
+                    292.0,
+                    1843.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1834.0,
+                    935.0,
+                    1834.0,
+                    935.0,
+                    1870.0,
+                    294.0,
+                    1870.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1208.0,
+                    1404.0,
+                    1208.0,
+                    1404.0,
+                    1247.0,
+                    293.0,
+                    1247.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1240.0,
+                    1406.0,
+                    1240.0,
+                    1406.0,
+                    1279.0,
+                    292.0,
+                    1279.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1271.0,
+                    1405.0,
+                    1271.0,
+                    1405.0,
+                    1311.0,
+                    292.0,
+                    1311.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1304.0,
+                    1406.0,
+                    1304.0,
+                    1406.0,
+                    1339.0,
+                    294.0,
+                    1339.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1332.0,
+                    1406.0,
+                    1332.0,
+                    1406.0,
+                    1370.0,
+                    293.0,
+                    1370.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1365.0,
+                    1189.0,
+                    1365.0,
+                    1189.0,
+                    1400.0,
+                    294.0,
+                    1400.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1264.0,
+                    1365.0,
+                    1405.0,
+                    1365.0,
+                    1405.0,
+                    1400.0,
+                    1264.0,
+                    1400.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1393.0,
+                    1405.0,
+                    1393.0,
+                    1405.0,
+                    1430.0,
+                    293.0,
+                    1430.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1423.0,
+                    1405.0,
+                    1423.0,
+                    1405.0,
+                    1460.0,
+                    293.0,
+                    1460.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1451.0,
+                    1405.0,
+                    1451.0,
+                    1405.0,
+                    1494.0,
+                    292.0,
+                    1494.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1486.0,
+                    1405.0,
+                    1486.0,
+                    1405.0,
+                    1521.0,
+                    293.0,
+                    1521.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1516.0,
+                    833.0,
+                    1516.0,
+                    833.0,
+                    1553.0,
+                    293.0,
+                    1553.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    922.0,
+                    1405.0,
+                    922.0,
+                    1405.0,
+                    960.0,
+                    292.0,
+                    960.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    955.0,
+                    1402.0,
+                    955.0,
+                    1402.0,
+                    989.0,
+                    293.0,
+                    989.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    985.0,
+                    1404.0,
+                    985.0,
+                    1404.0,
+                    1019.0,
+                    297.0,
+                    1019.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1015.0,
+                    1405.0,
+                    1015.0,
+                    1405.0,
+                    1049.0,
+                    294.0,
+                    1049.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1045.0,
+                    1405.0,
+                    1045.0,
+                    1405.0,
+                    1082.0,
+                    292.0,
+                    1082.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1076.0,
+                    1404.0,
+                    1076.0,
+                    1404.0,
+                    1110.0,
+                    294.0,
+                    1110.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1109.0,
+                    1405.0,
+                    1109.0,
+                    1405.0,
+                    1142.0,
+                    294.0,
+                    1142.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1136.0,
+                    1017.0,
+                    1136.0,
+                    1017.0,
+                    1174.0,
+                    293.0,
+                    1174.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    230.0,
+                    1405.0,
+                    230.0,
+                    1405.0,
+                    264.0,
+                    294.0,
+                    264.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    262.0,
+                    1405.0,
+                    262.0,
+                    1405.0,
+                    296.0,
+                    294.0,
+                    296.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    291.0,
+                    1404.0,
+                    291.0,
+                    1404.0,
+                    326.0,
+                    294.0,
+                    326.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    323.0,
+                    1404.0,
+                    323.0,
+                    1404.0,
+                    354.0,
+                    296.0,
+                    354.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    351.0,
+                    1405.0,
+                    351.0,
+                    1405.0,
+                    388.0,
+                    293.0,
+                    388.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    383.0,
+                    1405.0,
+                    383.0,
+                    1405.0,
+                    417.0,
+                    294.0,
+                    417.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    414.0,
+                    919.0,
+                    414.0,
+                    919.0,
+                    448.0,
+                    294.0,
+                    448.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1880.0,
+                    1405.0,
+                    1880.0,
+                    1405.0,
+                    1915.0,
+                    295.0,
+                    1915.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    1913.0,
+                    1403.0,
+                    1913.0,
+                    1403.0,
+                    1946.0,
+                    297.0,
+                    1946.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1939.0,
+                    1403.0,
+                    1939.0,
+                    1403.0,
+                    1980.0,
+                    293.0,
+                    1980.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1972.0,
+                    1405.0,
+                    1972.0,
+                    1405.0,
+                    2009.0,
+                    294.0,
+                    2009.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    2002.0,
+                    1405.0,
+                    2002.0,
+                    1405.0,
+                    2038.0,
+                    293.0,
+                    2038.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 1,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    570,
+                    1406,
+                    570,
+                    1406,
+                    859,
+                    297,
+                    859
+                ],
+                "score": 0.982
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    306,
+                    1403,
+                    306,
+                    1403,
+                    462,
+                    298,
+                    462
+                ],
+                "score": 0.978
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1105,
+                    1404,
+                    1105,
+                    1404,
+                    1230,
+                    297,
+                    1230
+                ],
+                "score": 0.97
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1869,
+                    1405,
+                    1869,
+                    1405,
+                    1975,
+                    298,
+                    1975
+                ],
+                "score": 0.967
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    556,
+                    944,
+                    1142,
+                    944,
+                    1142,
+                    1028,
+                    556,
+                    1028
+                ],
+                "score": 0.956
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    295,
+                    228,
+                    1402,
+                    228,
+                    1402,
+                    294,
+                    295,
+                    294
+                ],
+                "score": 0.954
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    296,
+                    868,
+                    1401,
+                    868,
+                    1401,
+                    931,
+                    296,
+                    931
+                ],
+                "score": 0.953
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    291,
+                    1040,
+                    1318,
+                    1040,
+                    1318,
+                    1075,
+                    291,
+                    1075
+                ],
+                "score": 0.925
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    300,
+                    75,
+                    813,
+                    75,
+                    813,
+                    105,
+                    300,
+                    105
+                ],
+                "score": 0.907
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    300,
+                    505,
+                    578,
+                    505,
+                    578,
+                    541,
+                    300,
+                    541
+                ],
+                "score": 0.907
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    299,
+                    1752,
+                    1181,
+                    1752,
+                    1181,
+                    1789,
+                    299,
+                    1789
+                ],
+                "score": 0.898
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    308,
+                    1985,
+                    1375,
+                    1985,
+                    1375,
+                    2046,
+                    308,
+                    2046
+                ],
+                "score": 0.857
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    291,
+                    1820,
+                    1306,
+                    1820,
+                    1306,
+                    1852,
+                    291,
+                    1852
+                ],
+                "score": 0.838
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    292,
+                    1243,
+                    1406,
+                    1243,
+                    1406,
+                    1711,
+                    292,
+                    1711
+                ],
+                "score": 0.696
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    841,
+                    2087,
+                    860,
+                    2087,
+                    860,
+                    2112,
+                    841,
+                    2112
+                ],
+                "score": 0.642
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    841,
+                    2087,
+                    859,
+                    2087,
+                    859,
+                    2112,
+                    841,
+                    2112
+                ],
+                "score": 0.613
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    380,
+                    695,
+                    582,
+                    695,
+                    582,
+                    729,
+                    380,
+                    729
+                ],
+                "score": 0.93,
+                "latex": "{ \\cal A } = \\{ a _ { 0 } , . . . a _ { n _ { a } } \\}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    898,
+                    1676,
+                    1029,
+                    1676,
+                    1029,
+                    1710,
+                    898,
+                    1710
+                ],
+                "score": 0.93,
+                "latex": "z = \\phi ( s , a )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1233,
+                    1674,
+                    1304,
+                    1674,
+                    1304,
+                    1715,
+                    1233,
+                    1715
+                ],
+                "score": 0.93,
+                "latex": "d _ { \\phi } ^ { { \\cal D } } ( z )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    491,
+                    1566,
+                    636,
+                    1566,
+                    636,
+                    1604,
+                    491,
+                    1604
+                ],
+                "score": 0.93,
+                "latex": "\\widehat { R } = \\widehat { R } _ { z } \\circ \\phi"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    346,
+                    789,
+                    539,
+                    789,
+                    539,
+                    821,
+                    346,
+                    821
+                ],
+                "score": 0.93,
+                "latex": "s _ { t + 1 } \\sim T ( s _ { t } , a _ { t } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    427,
+                    1676,
+                    520,
+                    1676,
+                    520,
+                    1710,
+                    427,
+                    1710
+                ],
+                "score": 0.93,
+                "latex": "d ^ { \\pi } ( s , a )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    990,
+                    1908,
+                    1115,
+                    1908,
+                    1115,
+                    1947,
+                    990,
+                    1947
+                ],
+                "score": 0.93,
+                "latex": "| R ^ { \\pi } - { \\widehat { R } } ^ { \\pi } |"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    848,
+                    1638,
+                    915,
+                    1638,
+                    915,
+                    1677,
+                    848,
+                    1677
+                ],
+                "score": 0.93,
+                "latex": "d _ { \\phi } ^ { \\pi } ( z )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    744,
+                    663,
+                    849,
+                    663,
+                    849,
+                    696,
+                    744,
+                    696
+                ],
+                "score": 0.93,
+                "latex": "S \\subseteq \\mathbb { R } ^ { d _ { s } }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    297,
+                    1566,
+                    438,
+                    1566,
+                    438,
+                    1605,
+                    297,
+                    1605
+                ],
+                "score": 0.93,
+                "latex": "\\bar { \\widehat { T } } = \\widehat { T } _ { z } \\circ \\phi"
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    556,
+                    940,
+                    1141,
+                    940,
+                    1141,
+                    1031,
+                    556,
+                    1031
+                ],
+                "score": 0.93,
+                "latex": "d ^ { \\pi } ( s , a ) \\triangleq ( 1 - \\gamma ) \\sum _ { t = 0 } ^ { \\infty } \\gamma ^ { t } \\operatorname* { P r } ( s _ { t } = s , a _ { t } = a | M , \\pi ) ."
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1066,
+                    1504,
+                    1170,
+                    1504,
+                    1170,
+                    1537,
+                    1066,
+                    1537
+                ],
+                "score": 0.93,
+                "latex": "\\mathcal { Z } \\subseteq \\mathbb { R } ^ { d _ { z } }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    497,
+                    1379,
+                    853,
+                    1379,
+                    853,
+                    1424,
+                    497,
+                    1424
+                ],
+                "score": 0.92,
+                "latex": "\\begin{array} { r } { \\widehat { R } ^ { \\pi } \\triangleq ( 1 - \\gamma ) \\mathbb { E } _ { \\widehat { M } , \\pi } \\left[ \\sum _ { t = 0 } ^ { \\infty } \\gamma ^ { t } r _ { t } \\right] } \\end{array}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    926,
+                    693,
+                    1041,
+                    693,
+                    1041,
+                    726,
+                    926,
+                    726
+                ],
+                "score": 0.92,
+                "latex": "\\mathcal { A } \\subseteq \\mathbb { R } ^ { d _ { a } ^ { \\star } }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1147,
+                    1304,
+                    1205,
+                    1304,
+                    1205,
+                    1342,
+                    1147,
+                    1342
+                ],
+                "score": 0.92,
+                "latex": "\\widehat { T } , \\widehat { R }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    296,
+                    1632,
+                    375,
+                    1632,
+                    375,
+                    1670,
+                    296,
+                    1670
+                ],
+                "score": 0.92,
+                "latex": "\\widehat { T } _ { z } , \\widehat { R } _ { z }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    746,
+                    1340,
+                    1033,
+                    1340,
+                    1033,
+                    1381,
+                    746,
+                    1381
+                ],
+                "score": 0.92,
+                "latex": "\\widehat { M } = \\langle S , A , \\widehat { T } , \\widehat { R } , d _ { 0 } , \\gamma \\rangle"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    642,
+                    758,
+                    740,
+                    758,
+                    740,
+                    788,
+                    642,
+                    788
+                ],
+                "score": 0.92,
+                "latex": "s _ { 0 } \\sim d _ { 0 }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1039,
+                    1041,
+                    1313,
+                    1041,
+                    1313,
+                    1077,
+                    1039,
+                    1077
+                ],
+                "score": 0.92,
+                "latex": "R ^ { \\pi } = \\mathbb { E } _ { ( s , a ) \\sim d ^ { \\pi } } \\left[ r ( s , a ) \\right]"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    756,
+                    1135,
+                    1032,
+                    1135,
+                    1032,
+                    1170,
+                    756,
+                    1170
+                ],
+                "score": 0.91,
+                "latex": "\\mathcal { D } = \\{ ( \\bar { s } _ { i } , a _ { i } , r _ { i } , s _ { i } ^ { \\prime } ) \\} _ { i = 1 } ^ { N }"
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    322,
+                    1984,
+                    1371,
+                    1984,
+                    1371,
+                    2046,
+                    322,
+                    2046
+                ],
+                "score": 0.91,
+                "latex": "\\begin{array} { r } { \\mathcal { E } _ { \\phi , \\widehat { R } _ { z } , \\widehat { T } _ { z } } ( s , a ) = c _ { R } D _ { T V } \\Big ( R ( r | s , a ) \\Big | \\Big | \\widehat { R } _ { z } \\big ( r | \\phi ( s , a ) \\big ) \\Big ) + c _ { T } D _ { T V } \\Big ( T ( s ^ { \\prime } | s , a ) \\Big | \\Big | \\widehat { T } _ { z } \\big ( s ^ { \\prime } | \\phi ( s , a ) \\big ) \\Big ) , } \\end{array}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    981,
+                    573,
+                    1269,
+                    573,
+                    1269,
+                    606,
+                    981,
+                    606
+                ],
+                "score": 0.91,
+                "latex": "M = \\langle S , A , T , R , d _ { 0 } , \\gamma \\rangle"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    811,
+                    1507,
+                    988,
+                    1507,
+                    988,
+                    1537,
+                    811,
+                    1537
+                ],
+                "score": 0.91,
+                "latex": "\\phi : \\mathcal { S } \\times \\mathcal { A }  \\mathcal { Z }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1276,
+                    1422,
+                    1393,
+                    1422,
+                    1393,
+                    1462,
+                    1276,
+                    1462
+                ],
+                "score": 0.91,
+                "latex": "| R ^ { \\pi } - { \\widehat R } ^ { \\pi } |"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    612,
+                    819,
+                    974,
+                    819,
+                    974,
+                    858,
+                    612,
+                    858
+                ],
+                "score": 0.9,
+                "latex": "\\begin{array} { r } { R ^ { \\pi } \\triangleq ( 1 - \\gamma ) \\mathbb { E } _ { M , \\pi } \\left[ \\sum _ { t = 0 } ^ { \\infty } \\gamma ^ { t } r _ { t } \\right] } \\end{array}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    297,
+                    634,
+                    628,
+                    634,
+                    628,
+                    668,
+                    297,
+                    668
+                ],
+                "score": 0.89,
+                "latex": "R : \\mathcal { S } \\times \\overline { { \\mathcal { A } } } \\to \\Delta ( [ 0 , r _ { \\operatorname* { m a x } } ] )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1223,
+                    756,
+                    1392,
+                    756,
+                    1392,
+                    790,
+                    1223,
+                    790
+                ],
+                "score": 0.89,
+                "latex": "\\boldsymbol { r } _ { t } \\sim R ( s _ { t } , a _ { t } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    995,
+                    1196,
+                    1031,
+                    1196,
+                    1031,
+                    1225,
+                    995,
+                    1225
+                ],
+                "score": 0.89,
+                "latex": "\\dot { d } ^ { \\mathcal { D } }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    781,
+                    1042,
+                    819,
+                    1042,
+                    819,
+                    1069,
+                    781,
+                    1069
+                ],
+                "score": 0.88,
+                "latex": "R ^ { \\pi }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    925,
+                    606,
+                    1198,
+                    606,
+                    1198,
+                    637,
+                    925,
+                    637
+                ],
+                "score": 0.88,
+                "latex": "T : \\mathcal { S } \\times \\mathcal { A }  \\Delta ( \\mathcal { S } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    915,
+                    636,
+                    945,
+                    636,
+                    945,
+                    666,
+                    915,
+                    666
+                ],
+                "score": 0.88,
+                "latex": "d _ { 0 }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    903,
+                    1246,
+                    941,
+                    1246,
+                    941,
+                    1273,
+                    903,
+                    1273
+                ],
+                "score": 0.88,
+                "latex": "R ^ { \\pi }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    635,
+                    1871,
+                    668,
+                    1871,
+                    668,
+                    1905,
+                    635,
+                    1905
+                ],
+                "score": 0.87,
+                "latex": "\\widehat { M }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1018,
+                    1379,
+                    1051,
+                    1379,
+                    1051,
+                    1412,
+                    1018,
+                    1412
+                ],
+                "score": 0.87,
+                "latex": "\\widehat { M }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    298,
+                    763,
+                    554,
+                    763,
+                    554,
+                    788,
+                    298,
+                    788
+                ],
+                "score": 0.87,
+                "latex": "s _ { 0 } , a _ { 0 } , r _ { 0 } , s _ { 1 } , a _ { 1 } , r _ { 1 } , \\ldots"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1309,
+                    1137,
+                    1346,
+                    1137,
+                    1346,
+                    1165,
+                    1309,
+                    1165
+                ],
+                "score": 0.86,
+                "latex": "R ^ { \\pi }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1188,
+                    1567,
+                    1212,
+                    1567,
+                    1212,
+                    1600,
+                    1188,
+                    1600
+                ],
+                "score": 0.86,
+                "latex": "\\widehat { T }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    995,
+                    758,
+                    1066,
+                    758,
+                    1066,
+                    788,
+                    995,
+                    788
+                ],
+                "score": 0.86,
+                "latex": "t \\geq 0"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    529,
+                    1422,
+                    562,
+                    1422,
+                    562,
+                    1456,
+                    529,
+                    1456
+                ],
+                "score": 0.85,
+                "latex": "\\widehat { M }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1019,
+                    1605,
+                    1039,
+                    1605,
+                    1039,
+                    1636,
+                    1019,
+                    1636
+                ],
+                "score": 0.84,
+                "latex": "\\phi"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1334,
+                    1538,
+                    1353,
+                    1538,
+                    1353,
+                    1569,
+                    1334,
+                    1569
+                ],
+                "score": 0.84,
+                "latex": "\\phi"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1079,
+                    757,
+                    1210,
+                    757,
+                    1210,
+                    789,
+                    1079,
+                    789
+                ],
+                "score": 0.83,
+                "latex": "a _ { t } \\sim \\pi ( s _ { t } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1164,
+                    641,
+                    1183,
+                    641,
+                    1183,
+                    667,
+                    1164,
+                    667
+                ],
+                "score": 0.81,
+                "latex": "\\gamma"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    866,
+                    1879,
+                    891,
+                    1879,
+                    891,
+                    1905,
+                    866,
+                    1905
+                ],
+                "score": 0.8,
+                "latex": "\\mathcal { D }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1171,
+                    1250,
+                    1192,
+                    1250,
+                    1192,
+                    1272,
+                    1171,
+                    1272
+                ],
+                "score": 0.76,
+                "latex": "\\pi"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1376,
+                    1640,
+                    1401,
+                    1640,
+                    1401,
+                    1666,
+                    1376,
+                    1666
+                ],
+                "score": 0.76,
+                "latex": "\\mathcal { Z }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    297,
+                    1347,
+                    322,
+                    1347,
+                    322,
+                    1375,
+                    297,
+                    1375
+                ],
+                "score": 0.76,
+                "latex": "\\mathcal { D }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    297,
+                    1276,
+                    322,
+                    1276,
+                    322,
+                    1303,
+                    297,
+                    1303
+                ],
+                "score": 0.75,
+                "latex": "\\mathcal { D }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    662,
+                    606,
+                    687,
+                    606,
+                    687,
+                    632,
+                    662,
+                    632
+                ],
+                "score": 0.75,
+                "latex": "\\mathcal { A }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1207,
+                    696,
+                    1241,
+                    696,
+                    1241,
+                    724,
+                    1207,
+                    724
+                ],
+                "score": 0.73,
+                "latex": "M"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1265,
+                    1567,
+                    1290,
+                    1567,
+                    1290,
+                    1601,
+                    1265,
+                    1601
+                ],
+                "score": 0.72,
+                "latex": "\\widehat { R }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    468,
+                    605,
+                    491,
+                    605,
+                    491,
+                    632,
+                    468,
+                    632
+                ],
+                "score": 0.72,
+                "latex": "s"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1373,
+                    701,
+                    1394,
+                    701,
+                    1394,
+                    724,
+                    1373,
+                    724
+                ],
+                "score": 0.71,
+                "latex": "\\pi"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1374,
+                    1885,
+                    1393,
+                    1885,
+                    1393,
+                    1905,
+                    1374,
+                    1905
+                ],
+                "score": 0.7,
+                "latex": "\\pi"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1186,
+                    1108,
+                    1224,
+                    1108,
+                    1224,
+                    1135,
+                    1186,
+                    1135
+                ],
+                "score": 0.61,
+                "latex": "R L"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1302,
+                    1580,
+                    1320,
+                    1580,
+                    1320,
+                    1600,
+                    1302,
+                    1600
+                ],
+                "score": 0.6,
+                "latex": "z"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1247,
+                    1435,
+                    1266,
+                    1435,
+                    1266,
+                    1456,
+                    1247,
+                    1456
+                ],
+                "score": 0.57,
+                "latex": "\\pi"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    73.0,
+                    815.0,
+                    73.0,
+                    815.0,
+                    108.0,
+                    295.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    503.0,
+                    582.0,
+                    503.0,
+                    582.0,
+                    546.0,
+                    293.0,
+                    546.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1751.0,
+                    1186.0,
+                    1751.0,
+                    1186.0,
+                    1792.0,
+                    293.0,
+                    1792.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1821.0,
+                    1306.0,
+                    1821.0,
+                    1306.0,
+                    1852.0,
+                    296.0,
+                    1852.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    838.0,
+                    2086.0,
+                    860.0,
+                    2086.0,
+                    860.0,
+                    2117.0,
+                    838.0,
+                    2117.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    839.0,
+                    2086.0,
+                    860.0,
+                    2086.0,
+                    860.0,
+                    2117.0,
+                    839.0,
+                    2117.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    569.0,
+                    980.0,
+                    569.0,
+                    980.0,
+                    609.0,
+                    292.0,
+                    609.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1270.0,
+                    569.0,
+                    1406.0,
+                    569.0,
+                    1406.0,
+                    609.0,
+                    1270.0,
+                    609.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    604.0,
+                    467.0,
+                    604.0,
+                    467.0,
+                    638.0,
+                    295.0,
+                    638.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    492.0,
+                    604.0,
+                    661.0,
+                    604.0,
+                    661.0,
+                    638.0,
+                    492.0,
+                    638.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    688.0,
+                    604.0,
+                    924.0,
+                    604.0,
+                    924.0,
+                    638.0,
+                    688.0,
+                    638.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1199.0,
+                    604.0,
+                    1406.0,
+                    604.0,
+                    1406.0,
+                    638.0,
+                    1199.0,
+                    638.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    629.0,
+                    631.0,
+                    914.0,
+                    631.0,
+                    914.0,
+                    672.0,
+                    629.0,
+                    672.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    946.0,
+                    631.0,
+                    1163.0,
+                    631.0,
+                    1163.0,
+                    672.0,
+                    946.0,
+                    672.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1184.0,
+                    631.0,
+                    1406.0,
+                    631.0,
+                    1406.0,
+                    672.0,
+                    1184.0,
+                    672.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    665.0,
+                    743.0,
+                    665.0,
+                    743.0,
+                    699.0,
+                    294.0,
+                    699.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    850.0,
+                    665.0,
+                    1404.0,
+                    665.0,
+                    1404.0,
+                    699.0,
+                    850.0,
+                    699.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    691.0,
+                    379.0,
+                    691.0,
+                    379.0,
+                    733.0,
+                    291.0,
+                    733.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    583.0,
+                    691.0,
+                    925.0,
+                    691.0,
+                    925.0,
+                    733.0,
+                    583.0,
+                    733.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1042.0,
+                    691.0,
+                    1206.0,
+                    691.0,
+                    1206.0,
+                    733.0,
+                    1042.0,
+                    733.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1242.0,
+                    691.0,
+                    1372.0,
+                    691.0,
+                    1372.0,
+                    733.0,
+                    1242.0,
+                    733.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1395.0,
+                    691.0,
+                    1408.0,
+                    691.0,
+                    1408.0,
+                    733.0,
+                    1395.0,
+                    733.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    724.0,
+                    1407.0,
+                    724.0,
+                    1407.0,
+                    761.0,
+                    292.0,
+                    761.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    757.0,
+                    297.0,
+                    757.0,
+                    297.0,
+                    792.0,
+                    292.0,
+                    792.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    555.0,
+                    757.0,
+                    641.0,
+                    757.0,
+                    641.0,
+                    792.0,
+                    555.0,
+                    792.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    741.0,
+                    757.0,
+                    994.0,
+                    757.0,
+                    994.0,
+                    792.0,
+                    741.0,
+                    792.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1067.0,
+                    757.0,
+                    1078.0,
+                    757.0,
+                    1078.0,
+                    792.0,
+                    1067.0,
+                    792.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1211.0,
+                    757.0,
+                    1222.0,
+                    757.0,
+                    1222.0,
+                    792.0,
+                    1211.0,
+                    792.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1393.0,
+                    757.0,
+                    1403.0,
+                    757.0,
+                    1403.0,
+                    792.0,
+                    1393.0,
+                    792.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    785.0,
+                    345.0,
+                    785.0,
+                    345.0,
+                    823.0,
+                    292.0,
+                    823.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    540.0,
+                    785.0,
+                    1407.0,
+                    785.0,
+                    1407.0,
+                    823.0,
+                    540.0,
+                    823.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    290.0,
+                    813.0,
+                    611.0,
+                    813.0,
+                    611.0,
+                    866.0,
+                    290.0,
+                    866.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    975.0,
+                    813.0,
+                    991.0,
+                    813.0,
+                    991.0,
+                    866.0,
+                    975.0,
+                    866.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    307.0,
+                    1402.0,
+                    307.0,
+                    1402.0,
+                    343.0,
+                    294.0,
+                    343.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    337.0,
+                    1405.0,
+                    337.0,
+                    1405.0,
+                    373.0,
+                    293.0,
+                    373.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    369.0,
+                    1404.0,
+                    369.0,
+                    1404.0,
+                    403.0,
+                    294.0,
+                    403.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    400.0,
+                    1404.0,
+                    400.0,
+                    1404.0,
+                    432.0,
+                    294.0,
+                    432.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    427.0,
+                    1129.0,
+                    427.0,
+                    1129.0,
+                    467.0,
+                    293.0,
+                    467.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1103.0,
+                    1185.0,
+                    1103.0,
+                    1185.0,
+                    1142.0,
+                    294.0,
+                    1142.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1225.0,
+                    1103.0,
+                    1405.0,
+                    1103.0,
+                    1405.0,
+                    1142.0,
+                    1225.0,
+                    1142.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    290.0,
+                    1129.0,
+                    755.0,
+                    1129.0,
+                    755.0,
+                    1177.0,
+                    290.0,
+                    1177.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1033.0,
+                    1129.0,
+                    1308.0,
+                    1129.0,
+                    1308.0,
+                    1177.0,
+                    1033.0,
+                    1177.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1347.0,
+                    1129.0,
+                    1409.0,
+                    1129.0,
+                    1409.0,
+                    1177.0,
+                    1347.0,
+                    1177.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1169.0,
+                    1404.0,
+                    1169.0,
+                    1404.0,
+                    1201.0,
+                    295.0,
+                    1201.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1198.0,
+                    994.0,
+                    1198.0,
+                    994.0,
+                    1231.0,
+                    292.0,
+                    1231.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1032.0,
+                    1198.0,
+                    1041.0,
+                    1198.0,
+                    1041.0,
+                    1231.0,
+                    1032.0,
+                    1231.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1870.0,
+                    634.0,
+                    1870.0,
+                    634.0,
+                    1916.0,
+                    293.0,
+                    1916.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    669.0,
+                    1870.0,
+                    865.0,
+                    1870.0,
+                    865.0,
+                    1916.0,
+                    669.0,
+                    1916.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    892.0,
+                    1870.0,
+                    1373.0,
+                    1870.0,
+                    1373.0,
+                    1916.0,
+                    892.0,
+                    1916.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1394.0,
+                    1870.0,
+                    1406.0,
+                    1870.0,
+                    1406.0,
+                    1916.0,
+                    1394.0,
+                    1916.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    290.0,
+                    1904.0,
+                    989.0,
+                    1904.0,
+                    989.0,
+                    1956.0,
+                    290.0,
+                    1956.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1116.0,
+                    1904.0,
+                    1409.0,
+                    1904.0,
+                    1409.0,
+                    1956.0,
+                    1116.0,
+                    1956.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1943.0,
+                    855.0,
+                    1943.0,
+                    855.0,
+                    1979.0,
+                    292.0,
+                    1979.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    227.0,
+                    1404.0,
+                    227.0,
+                    1404.0,
+                    266.0,
+                    294.0,
+                    266.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    260.0,
+                    933.0,
+                    260.0,
+                    933.0,
+                    299.0,
+                    293.0,
+                    299.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    866.0,
+                    1405.0,
+                    866.0,
+                    1405.0,
+                    905.0,
+                    294.0,
+                    905.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    900.0,
+                    805.0,
+                    900.0,
+                    805.0,
+                    935.0,
+                    292.0,
+                    935.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    290.0,
+                    1033.0,
+                    780.0,
+                    1033.0,
+                    780.0,
+                    1082.0,
+                    290.0,
+                    1082.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    820.0,
+                    1033.0,
+                    1038.0,
+                    1033.0,
+                    1038.0,
+                    1082.0,
+                    820.0,
+                    1082.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1314.0,
+                    1033.0,
+                    1323.0,
+                    1033.0,
+                    1323.0,
+                    1082.0,
+                    1314.0,
+                    1082.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1244.0,
+                    902.0,
+                    1244.0,
+                    902.0,
+                    1279.0,
+                    294.0,
+                    1279.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    942.0,
+                    1244.0,
+                    1170.0,
+                    1244.0,
+                    1170.0,
+                    1279.0,
+                    942.0,
+                    1279.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1193.0,
+                    1244.0,
+                    1405.0,
+                    1244.0,
+                    1405.0,
+                    1279.0,
+                    1193.0,
+                    1279.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    1275.0,
+                    1404.0,
+                    1275.0,
+                    1404.0,
+                    1311.0,
+                    323.0,
+                    1311.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1310.0,
+                    1146.0,
+                    1310.0,
+                    1146.0,
+                    1345.0,
+                    294.0,
+                    1345.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1206.0,
+                    1310.0,
+                    1405.0,
+                    1310.0,
+                    1405.0,
+                    1345.0,
+                    1206.0,
+                    1345.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1344.0,
+                    296.0,
+                    1344.0,
+                    296.0,
+                    1383.0,
+                    293.0,
+                    1383.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    1344.0,
+                    745.0,
+                    1344.0,
+                    745.0,
+                    1383.0,
+                    323.0,
+                    1383.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1034.0,
+                    1344.0,
+                    1407.0,
+                    1344.0,
+                    1407.0,
+                    1383.0,
+                    1034.0,
+                    1383.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    285.0,
+                    1368.0,
+                    496.0,
+                    1368.0,
+                    496.0,
+                    1435.0,
+                    285.0,
+                    1435.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    854.0,
+                    1368.0,
+                    1017.0,
+                    1368.0,
+                    1017.0,
+                    1435.0,
+                    854.0,
+                    1435.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1052.0,
+                    1368.0,
+                    1416.0,
+                    1368.0,
+                    1416.0,
+                    1435.0,
+                    1052.0,
+                    1435.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1422.0,
+                    528.0,
+                    1422.0,
+                    528.0,
+                    1464.0,
+                    291.0,
+                    1464.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    563.0,
+                    1422.0,
+                    1246.0,
+                    1422.0,
+                    1246.0,
+                    1464.0,
+                    563.0,
+                    1464.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1267.0,
+                    1422.0,
+                    1275.0,
+                    1422.0,
+                    1275.0,
+                    1464.0,
+                    1267.0,
+                    1464.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1394.0,
+                    1422.0,
+                    1404.0,
+                    1422.0,
+                    1404.0,
+                    1464.0,
+                    1394.0,
+                    1464.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1474.0,
+                    1404.0,
+                    1474.0,
+                    1404.0,
+                    1510.0,
+                    295.0,
+                    1510.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1505.0,
+                    810.0,
+                    1505.0,
+                    810.0,
+                    1540.0,
+                    295.0,
+                    1540.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    989.0,
+                    1505.0,
+                    1065.0,
+                    1505.0,
+                    1065.0,
+                    1540.0,
+                    989.0,
+                    1540.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1171.0,
+                    1505.0,
+                    1404.0,
+                    1505.0,
+                    1404.0,
+                    1540.0,
+                    1171.0,
+                    1540.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1537.0,
+                    1333.0,
+                    1537.0,
+                    1333.0,
+                    1571.0,
+                    293.0,
+                    1571.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1354.0,
+                    1537.0,
+                    1404.0,
+                    1537.0,
+                    1404.0,
+                    1571.0,
+                    1354.0,
+                    1571.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    290.0,
+                    1563.0,
+                    296.0,
+                    1563.0,
+                    296.0,
+                    1611.0,
+                    290.0,
+                    1611.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    439.0,
+                    1563.0,
+                    490.0,
+                    1563.0,
+                    490.0,
+                    1611.0,
+                    439.0,
+                    1611.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    637.0,
+                    1563.0,
+                    1187.0,
+                    1563.0,
+                    1187.0,
+                    1611.0,
+                    637.0,
+                    1611.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1213.0,
+                    1563.0,
+                    1264.0,
+                    1563.0,
+                    1264.0,
+                    1611.0,
+                    1213.0,
+                    1611.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1291.0,
+                    1563.0,
+                    1301.0,
+                    1563.0,
+                    1301.0,
+                    1611.0,
+                    1291.0,
+                    1611.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1321.0,
+                    1563.0,
+                    1407.0,
+                    1563.0,
+                    1407.0,
+                    1611.0,
+                    1321.0,
+                    1611.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1601.0,
+                    1018.0,
+                    1601.0,
+                    1018.0,
+                    1638.0,
+                    293.0,
+                    1638.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1040.0,
+                    1601.0,
+                    1405.0,
+                    1601.0,
+                    1405.0,
+                    1638.0,
+                    1040.0,
+                    1638.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1630.0,
+                    295.0,
+                    1630.0,
+                    295.0,
+                    1679.0,
+                    291.0,
+                    1679.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    376.0,
+                    1630.0,
+                    847.0,
+                    1630.0,
+                    847.0,
+                    1679.0,
+                    376.0,
+                    1679.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    916.0,
+                    1630.0,
+                    1375.0,
+                    1630.0,
+                    1375.0,
+                    1679.0,
+                    916.0,
+                    1679.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1402.0,
+                    1630.0,
+                    1408.0,
+                    1630.0,
+                    1408.0,
+                    1679.0,
+                    1402.0,
+                    1679.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1670.0,
+                    426.0,
+                    1670.0,
+                    426.0,
+                    1715.0,
+                    291.0,
+                    1715.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    521.0,
+                    1670.0,
+                    897.0,
+                    1670.0,
+                    897.0,
+                    1715.0,
+                    521.0,
+                    1715.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1030.0,
+                    1670.0,
+                    1232.0,
+                    1670.0,
+                    1232.0,
+                    1715.0,
+                    1030.0,
+                    1715.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1305.0,
+                    1670.0,
+                    1316.0,
+                    1670.0,
+                    1316.0,
+                    1715.0,
+                    1305.0,
+                    1715.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 2,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    296,
+                    1184,
+                    1405,
+                    1184,
+                    1405,
+                    1440,
+                    296,
+                    1440
+                ],
+                "score": 0.981
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    296,
+                    497,
+                    1404,
+                    497,
+                    1404,
+                    684,
+                    296,
+                    684
+                ],
+                "score": 0.98
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1452,
+                    1405,
+                    1452,
+                    1405,
+                    1575,
+                    298,
+                    1575
+                ],
+                "score": 0.977
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    296,
+                    696,
+                    1404,
+                    696,
+                    1404,
+                    825,
+                    296,
+                    825
+                ],
+                "score": 0.977
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    296,
+                    975,
+                    1404,
+                    975,
+                    1404,
+                    1092,
+                    296,
+                    1092
+                ],
+                "score": 0.974
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    326,
+                    1700,
+                    1342,
+                    1700,
+                    1342,
+                    1965,
+                    326,
+                    1965
+                ],
+                "score": 0.971
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    228,
+                    1403,
+                    228,
+                    1403,
+                    323,
+                    298,
+                    323
+                ],
+                "score": 0.958
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    433,
+                    1581,
+                    1260,
+                    1581,
+                    1260,
+                    1656,
+                    433,
+                    1656
+                ],
+                "score": 0.947
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    501,
+                    1102,
+                    1199,
+                    1102,
+                    1199,
+                    1163,
+                    501,
+                    1163
+                ],
+                "score": 0.945
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    903,
+                    1403,
+                    903,
+                    1403,
+                    967,
+                    298,
+                    967
+                ],
+                "score": 0.943
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    583,
+                    833,
+                    1115,
+                    833,
+                    1115,
+                    893,
+                    583,
+                    893
+                ],
+                "score": 0.938
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    293,
+                    331,
+                    1400,
+                    331,
+                    1400,
+                    406,
+                    293,
+                    406
+                ],
+                "score": 0.937
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1664,
+                    668,
+                    1664,
+                    668,
+                    1696,
+                    298,
+                    1696
+                ],
+                "score": 0.933
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    295,
+                    1972,
+                    1402,
+                    1972,
+                    1402,
+                    2037,
+                    295,
+                    2037
+                ],
+                "score": 0.932
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    299,
+                    74,
+                    813,
+                    74,
+                    813,
+                    105,
+                    299,
+                    105
+                ],
+                "score": 0.918
+            },
+            {
+                "category_id": 9,
+                "poly": [
+                    1366,
+                    839,
+                    1400,
+                    839,
+                    1400,
+                    869,
+                    1366,
+                    869
+                ],
+                "score": 0.885
+            },
+            {
+                "category_id": 9,
+                "poly": [
+                    1365,
+                    429,
+                    1400,
+                    429,
+                    1400,
+                    460,
+                    1365,
+                    460
+                ],
+                "score": 0.88
+            },
+            {
+                "category_id": 9,
+                "poly": [
+                    1365,
+                    1117,
+                    1400,
+                    1117,
+                    1400,
+                    1148,
+                    1365,
+                    1148
+                ],
+                "score": 0.874
+            },
+            {
+                "category_id": 9,
+                "poly": [
+                    1366,
+                    1806,
+                    1401,
+                    1806,
+                    1401,
+                    1836,
+                    1366,
+                    1836
+                ],
+                "score": 0.806
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    840,
+                    2088,
+                    859,
+                    2088,
+                    859,
+                    2112,
+                    840,
+                    2112
+                ],
+                "score": 0.781
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    620,
+                    415,
+                    1078,
+                    415,
+                    1078,
+                    478,
+                    620,
+                    478
+                ],
+                "score": 0.569
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    622,
+                    416,
+                    1077,
+                    416,
+                    1077,
+                    478,
+                    622,
+                    478
+                ],
+                "score": 0.343
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    584,
+                    832,
+                    1115,
+                    832,
+                    1115,
+                    894,
+                    584,
+                    894
+                ],
+                "score": 0.95,
+                "latex": "\\mathrm { I P M } _ { \\mathcal { G } } ( p , q ) = \\operatorname* { s u p } _ { g \\in \\mathcal { G } } \\big | \\mathbb { E } _ { z \\sim p } [ g ( z ) ] - \\mathbb { E } _ { z \\sim q } [ g ( z ) ] \\big | ."
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    619,
+                    414,
+                    1079,
+                    414,
+                    1079,
+                    475,
+                    619,
+                    475
+                ],
+                "score": 0.94,
+                "latex": "\\begin{array} { r } { \\left| R ^ { \\pi } - \\widehat { R } ^ { \\pi } \\right| \\leq \\mathbb { E } _ { ( s , a ) \\sim d ^ { \\pi } } \\left[ \\mathcal { E } _ { \\phi , \\widehat { R } _ { z } , \\widehat { T } _ { z } } ( s , a ) \\right] } \\end{array}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    619,
+                    1049,
+                    904,
+                    1049,
+                    904,
+                    1093,
+                    619,
+                    1093
+                ],
+                "score": 0.94,
+                "latex": "\\begin{array} { r } { \\frac { 1 } { B _ { \\phi } } \\mathcal { E } _ { \\phi , \\widehat { R } _ { z } , \\widehat { T } _ { z } } \\left( \\phi ^ { - 1 } ( \\cdot ) \\right) \\in \\mathcal { G } } \\end{array}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1068,
+                    1016,
+                    1165,
+                    1016,
+                    1165,
+                    1051,
+                    1068,
+                    1051
+                ],
+                "score": 0.93,
+                "latex": "B _ { \\phi } > 0"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    933,
+                    1217,
+                    1101,
+                    1217,
+                    1101,
+                    1256,
+                    933,
+                    1256
+                ],
+                "score": 0.93,
+                "latex": "\\mathrm { I P M } _ { \\mathcal { G } } ( d _ { \\phi } ^ { \\pi } , d _ { \\phi } ^ { D } )"
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    434,
+                    1579,
+                    1264,
+                    1579,
+                    1264,
+                    1658,
+                    434,
+                    1658
+                ],
+                "score": 0.93,
+                "latex": "\\nu \\big ( \\phi ( s , a ) \\big ) \\triangleq g \\big ( \\phi ( s , a ) \\big ) + \\gamma \\mathbb { E } _ { s ^ { \\prime } \\sim T ( s , a ) } \\Big [ \\nu \\big ( \\phi ( s ^ { \\prime } , a ^ { \\prime } ) \\big ) \\Big ] , \\forall ( s , a ) \\in \\mathcal { S } \\times \\mathcal { A } ."
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    372,
+                    230,
+                    542,
+                    230,
+                    542,
+                    264,
+                    372,
+                    264
+                ],
+                "score": 0.92,
+                "latex": "c _ { R } = 2 ( 1 - \\gamma )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    884,
+                    2003,
+                    1017,
+                    2003,
+                    1017,
+                    2036,
+                    884,
+                    2036
+                ],
+                "score": 0.92,
+                "latex": "g ( \\phi ( s _ { t } , a _ { t } ) )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    298,
+                    366,
+                    592,
+                    366,
+                    592,
+                    406,
+                    298,
+                    406
+                ],
+                "score": 0.92,
+                "latex": "( \\widehat { T } , \\widehat { R } ) = \\langle \\widehat { T } _ { z } \\circ \\phi , \\widehat { R } _ { z } \\circ \\phi \\rangle"
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    498,
+                    1103,
+                    1199,
+                    1103,
+                    1199,
+                    1162,
+                    498,
+                    1162
+                ],
+                "score": 0.92,
+                "latex": "\\left| R ^ { \\pi } - \\widehat { R } ^ { \\pi } \\right| \\leq \\mathbb { E } _ { ( s , a ) \\sim d ^ { \\mathcal { D } } } \\Big [ \\mathcal { E } _ { \\phi , \\widehat { R } _ { z } , \\widehat { T } _ { z } } ( s , a ) \\Big ] + B _ { \\phi } \\mathrm { I P M } _ { \\mathcal { G } } ( d _ { \\phi } ^ { \\pi } , d _ { \\phi } ^ { \\mathcal { D } } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    594,
+                    232,
+                    748,
+                    232,
+                    748,
+                    263,
+                    594,
+                    263
+                ],
+                "score": 0.92,
+                "latex": "c _ { T } = 2 \\gamma r _ { \\operatorname* { m a x } }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    298,
+                    1050,
+                    506,
+                    1050,
+                    506,
+                    1085,
+                    298,
+                    1085
+                ],
+                "score": 0.91,
+                "latex": "{ \\mathcal { G } } \\triangleq \\{ g : { \\mathcal { Z } } \\to \\mathbb { R } \\}"
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    324,
+                    1701,
+                    1337,
+                    1701,
+                    1337,
+                    1967,
+                    324,
+                    1967
+                ],
+                "score": 0.91,
+                "latex": "\\begin{array} { r l } & { \\displaystyle \\mathrm { I P M } _ { \\mathcal { G } } ( d _ { \\phi } ^ { \\pi } , d _ { \\phi } ^ { \\mathcal { D } } ) = \\underset { g \\in \\mathcal { G } } { \\operatorname* { s u p } } \\left| \\mathbb { E } _ { ( s , a ) \\sim d ^ { \\pi } } \\Big [ g \\big ( \\phi ( s , a ) \\big ) \\Big ] - \\mathbb { E } _ { ( s , a ) \\sim d ^ { \\mathcal { D } } } \\Big [ g \\big ( \\phi ( s , a ) \\big ) \\Big ] \\right| } \\\\ & { \\displaystyle \\quad = \\operatorname* { s u p } _ { \\nu \\in \\mathcal { F } } \\left| ( 1 - \\gamma ) \\mathbb { E } \\qquad \\Big [ \\nu \\big ( \\phi ( s , a ) \\big ) \\Big ] - \\mathbb { E } _ { ( s , a , s ^ { \\prime } ) \\sim d ^ { \\mathcal { D } } } \\Big [ \\nu \\big ( \\phi ( s , a ) \\big ) - \\gamma \\nu \\big ( \\phi ( s ^ { \\prime } , a ^ { \\prime } ) \\big ) \\Big ] \\right| , } \\\\ & { \\displaystyle \\quad \\quad \\quad \\quad \\quad \\quad \\quad \\quad \\quad \\quad \\quad \\quad \\quad \\quad \\quad \\quad \\quad \\quad \\quad \\quad \\quad \\quad \\quad \\quad \\quad \\quad \\quad \\quad \\quad \\quad \\quad \\quad a ^ { \\prime } \\sim \\pi ^ { \\prime } } \\end{array}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    347,
+                    1009,
+                    653,
+                    1009,
+                    653,
+                    1049,
+                    347,
+                    1049
+                ],
+                "score": 0.91,
+                "latex": "( \\widehat { T } , \\widehat { R } ) = \\langle \\widehat { T } _ { z } \\circ \\phi , \\widehat { R } _ { z } \\circ \\phi \\rangle"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    654,
+                    1402,
+                    690,
+                    1402,
+                    690,
+                    1442,
+                    654,
+                    1442
+                ],
+                "score": 0.91,
+                "latex": "d _ { \\phi } ^ { \\mathcal { D } }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    716,
+                    1515,
+                    853,
+                    1515,
+                    853,
+                    1543,
+                    716,
+                    1543
+                ],
+                "score": 0.9,
+                "latex": "\\nu : \\mathcal { Z } \\to \\mathbb { R }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    698,
+                    758,
+                    734,
+                    758,
+                    734,
+                    797,
+                    698,
+                    797
+                ],
+                "score": 0.9,
+                "latex": "d _ { \\phi } ^ { \\mathcal { D } }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    571,
+                    1404,
+                    603,
+                    1404,
+                    603,
+                    1442,
+                    571,
+                    1442
+                ],
+                "score": 0.89,
+                "latex": "d _ { \\phi } ^ { \\pi }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    615,
+                    761,
+                    647,
+                    761,
+                    647,
+                    797,
+                    615,
+                    797
+                ],
+                "score": 0.88,
+                "latex": "d _ { \\phi } ^ { \\pi }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    473,
+                    592,
+                    505,
+                    592,
+                    505,
+                    619,
+                    473,
+                    619
+                ],
+                "score": 0.87,
+                "latex": "d ^ { \\pi }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    693,
+                    1280,
+                    729,
+                    1280,
+                    729,
+                    1310,
+                    693,
+                    1310
+                ],
+                "score": 0.87,
+                "latex": "d ^ { \\mathcal { D } }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    544,
+                    731,
+                    575,
+                    731,
+                    575,
+                    757,
+                    544,
+                    757
+                ],
+                "score": 0.86,
+                "latex": "d ^ { \\pi }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    851,
+                    330,
+                    884,
+                    330,
+                    884,
+                    365,
+                    851,
+                    365
+                ],
+                "score": 0.86,
+                "latex": "\\widehat { M }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    882,
+                    973,
+                    915,
+                    973,
+                    915,
+                    1008,
+                    882,
+                    1008
+                ],
+                "score": 0.86,
+                "latex": "\\widehat { M }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1284,
+                    1455,
+                    1315,
+                    1455,
+                    1315,
+                    1481,
+                    1284,
+                    1481
+                ],
+                "score": 0.84,
+                "latex": "d ^ { \\pi }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    608,
+                    906,
+                    629,
+                    906,
+                    629,
+                    934,
+                    608,
+                    934
+                ],
+                "score": 0.84,
+                "latex": "\\mathcal { G }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1167,
+                    1974,
+                    1193,
+                    1974,
+                    1193,
+                    2001,
+                    1167,
+                    2001
+                ],
+                "score": 0.84,
+                "latex": "\\mathcal { F }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    690,
+                    623,
+                    716,
+                    623,
+                    716,
+                    649,
+                    690,
+                    649
+                ],
+                "score": 0.82,
+                "latex": "\\mathcal { D }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    969,
+                    1257,
+                    989,
+                    1257,
+                    989,
+                    1282,
+                    969,
+                    1282
+                ],
+                "score": 0.82,
+                "latex": "\\mathcal { G }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1333,
+                    339,
+                    1353,
+                    339,
+                    1353,
+                    370,
+                    1333,
+                    370
+                ],
+                "score": 0.79,
+                "latex": "\\phi _ { : }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1253,
+                    2008,
+                    1273,
+                    2008,
+                    1273,
+                    2030,
+                    1253,
+                    2030
+                ],
+                "score": 0.77,
+                "latex": "\\pi"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    645,
+                    537,
+                    665,
+                    537,
+                    665,
+                    559,
+                    645,
+                    559
+                ],
+                "score": 0.76,
+                "latex": "\\pi"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1375,
+                    983,
+                    1393,
+                    983,
+                    1393,
+                    1013,
+                    1375,
+                    1013
+                ],
+                "score": 0.76,
+                "latex": "\\phi"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1147,
+                    1059,
+                    1165,
+                    1059,
+                    1165,
+                    1080,
+                    1147,
+                    1080
+                ],
+                "score": 0.73,
+                "latex": "\\pi"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1032,
+                    379,
+                    1052,
+                    379,
+                    1052,
+                    400,
+                    1032,
+                    400
+                ],
+                "score": 0.63,
+                "latex": "\\pi"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    654,
+                    981,
+                    687,
+                    981,
+                    687,
+                    1008,
+                    654,
+                    1008
+                ],
+                "score": 0.59,
+                "latex": "M"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    629,
+                    338,
+                    663,
+                    338,
+                    663,
+                    365,
+                    629,
+                    365
+                ],
+                "score": 0.52,
+                "latex": "M"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1220,
+                    1021,
+                    1237,
+                    1021,
+                    1237,
+                    1043,
+                    1220,
+                    1043
+                ],
+                "score": 0.26,
+                "latex": "a"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    72.0,
+                    815.0,
+                    72.0,
+                    815.0,
+                    108.0,
+                    295.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    838.0,
+                    2085.0,
+                    863.0,
+                    2085.0,
+                    863.0,
+                    2118.0,
+                    838.0,
+                    2118.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1186.0,
+                    1406.0,
+                    1186.0,
+                    1406.0,
+                    1221.0,
+                    296.0,
+                    1221.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    289.0,
+                    1209.0,
+                    932.0,
+                    1209.0,
+                    932.0,
+                    1264.0,
+                    289.0,
+                    1264.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1102.0,
+                    1209.0,
+                    1413.0,
+                    1209.0,
+                    1413.0,
+                    1264.0,
+                    1102.0,
+                    1264.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1251.0,
+                    968.0,
+                    1251.0,
+                    968.0,
+                    1286.0,
+                    294.0,
+                    1286.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    990.0,
+                    1251.0,
+                    1405.0,
+                    1251.0,
+                    1405.0,
+                    1286.0,
+                    990.0,
+                    1286.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1281.0,
+                    692.0,
+                    1281.0,
+                    692.0,
+                    1316.0,
+                    294.0,
+                    1316.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    730.0,
+                    1281.0,
+                    1405.0,
+                    1281.0,
+                    1405.0,
+                    1316.0,
+                    730.0,
+                    1316.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1311.0,
+                    1405.0,
+                    1311.0,
+                    1405.0,
+                    1344.0,
+                    293.0,
+                    1344.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1344.0,
+                    1403.0,
+                    1344.0,
+                    1403.0,
+                    1375.0,
+                    295.0,
+                    1375.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1373.0,
+                    1406.0,
+                    1373.0,
+                    1406.0,
+                    1407.0,
+                    295.0,
+                    1407.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    289.0,
+                    1395.0,
+                    570.0,
+                    1395.0,
+                    570.0,
+                    1449.0,
+                    289.0,
+                    1449.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    604.0,
+                    1395.0,
+                    653.0,
+                    1395.0,
+                    653.0,
+                    1449.0,
+                    604.0,
+                    1449.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    691.0,
+                    1395.0,
+                    704.0,
+                    1395.0,
+                    704.0,
+                    1449.0,
+                    691.0,
+                    1449.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    496.0,
+                    1404.0,
+                    496.0,
+                    1404.0,
+                    535.0,
+                    294.0,
+                    535.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    528.0,
+                    644.0,
+                    528.0,
+                    644.0,
+                    569.0,
+                    293.0,
+                    569.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    666.0,
+                    528.0,
+                    1407.0,
+                    528.0,
+                    1407.0,
+                    569.0,
+                    666.0,
+                    569.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    562.0,
+                    1405.0,
+                    562.0,
+                    1405.0,
+                    594.0,
+                    295.0,
+                    594.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    593.0,
+                    472.0,
+                    593.0,
+                    472.0,
+                    625.0,
+                    295.0,
+                    625.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    506.0,
+                    593.0,
+                    1405.0,
+                    593.0,
+                    1405.0,
+                    625.0,
+                    506.0,
+                    625.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    621.0,
+                    689.0,
+                    621.0,
+                    689.0,
+                    657.0,
+                    293.0,
+                    657.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    717.0,
+                    621.0,
+                    1406.0,
+                    621.0,
+                    1406.0,
+                    657.0,
+                    717.0,
+                    657.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    653.0,
+                    500.0,
+                    653.0,
+                    500.0,
+                    687.0,
+                    294.0,
+                    687.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1453.0,
+                    1283.0,
+                    1453.0,
+                    1283.0,
+                    1486.0,
+                    295.0,
+                    1486.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1316.0,
+                    1453.0,
+                    1403.0,
+                    1453.0,
+                    1403.0,
+                    1486.0,
+                    1316.0,
+                    1486.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1483.0,
+                    1405.0,
+                    1483.0,
+                    1405.0,
+                    1516.0,
+                    293.0,
+                    1516.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1513.0,
+                    715.0,
+                    1513.0,
+                    715.0,
+                    1548.0,
+                    293.0,
+                    1548.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    854.0,
+                    1513.0,
+                    1405.0,
+                    1513.0,
+                    1405.0,
+                    1548.0,
+                    854.0,
+                    1548.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1543.0,
+                    400.0,
+                    1543.0,
+                    400.0,
+                    1577.0,
+                    293.0,
+                    1577.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    695.0,
+                    1405.0,
+                    695.0,
+                    1405.0,
+                    734.0,
+                    293.0,
+                    734.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    729.0,
+                    543.0,
+                    729.0,
+                    543.0,
+                    763.0,
+                    296.0,
+                    763.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    576.0,
+                    729.0,
+                    1402.0,
+                    729.0,
+                    1402.0,
+                    763.0,
+                    576.0,
+                    763.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    757.0,
+                    614.0,
+                    757.0,
+                    614.0,
+                    796.0,
+                    291.0,
+                    796.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    648.0,
+                    757.0,
+                    697.0,
+                    757.0,
+                    697.0,
+                    796.0,
+                    648.0,
+                    796.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    735.0,
+                    757.0,
+                    1405.0,
+                    757.0,
+                    1405.0,
+                    796.0,
+                    735.0,
+                    796.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    794.0,
+                    853.0,
+                    794.0,
+                    853.0,
+                    827.0,
+                    295.0,
+                    827.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    972.0,
+                    653.0,
+                    972.0,
+                    653.0,
+                    1017.0,
+                    294.0,
+                    1017.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    688.0,
+                    972.0,
+                    881.0,
+                    972.0,
+                    881.0,
+                    1017.0,
+                    688.0,
+                    1017.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    916.0,
+                    972.0,
+                    1374.0,
+                    972.0,
+                    1374.0,
+                    1017.0,
+                    916.0,
+                    1017.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1394.0,
+                    972.0,
+                    1404.0,
+                    972.0,
+                    1404.0,
+                    1017.0,
+                    1394.0,
+                    1017.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1008.0,
+                    346.0,
+                    1008.0,
+                    346.0,
+                    1054.0,
+                    295.0,
+                    1054.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    654.0,
+                    1008.0,
+                    1067.0,
+                    1008.0,
+                    1067.0,
+                    1054.0,
+                    654.0,
+                    1054.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1166.0,
+                    1008.0,
+                    1219.0,
+                    1008.0,
+                    1219.0,
+                    1054.0,
+                    1166.0,
+                    1054.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1238.0,
+                    1008.0,
+                    1405.0,
+                    1008.0,
+                    1405.0,
+                    1054.0,
+                    1238.0,
+                    1054.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    287.0,
+                    1029.0,
+                    297.0,
+                    1029.0,
+                    297.0,
+                    1108.0,
+                    287.0,
+                    1108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    507.0,
+                    1029.0,
+                    618.0,
+                    1029.0,
+                    618.0,
+                    1108.0,
+                    507.0,
+                    1108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    905.0,
+                    1029.0,
+                    1146.0,
+                    1029.0,
+                    1146.0,
+                    1108.0,
+                    905.0,
+                    1108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1166.0,
+                    1029.0,
+                    1186.0,
+                    1029.0,
+                    1186.0,
+                    1108.0,
+                    1166.0,
+                    1108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    229.0,
+                    371.0,
+                    229.0,
+                    371.0,
+                    267.0,
+                    294.0,
+                    267.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    543.0,
+                    229.0,
+                    593.0,
+                    229.0,
+                    593.0,
+                    267.0,
+                    543.0,
+                    267.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    749.0,
+                    229.0,
+                    1404.0,
+                    229.0,
+                    1404.0,
+                    267.0,
+                    749.0,
+                    267.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    260.0,
+                    1405.0,
+                    260.0,
+                    1405.0,
+                    296.0,
+                    293.0,
+                    296.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    291.0,
+                    606.0,
+                    291.0,
+                    606.0,
+                    325.0,
+                    293.0,
+                    325.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    903.0,
+                    607.0,
+                    903.0,
+                    607.0,
+                    935.0,
+                    297.0,
+                    935.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    630.0,
+                    903.0,
+                    1403.0,
+                    903.0,
+                    1403.0,
+                    935.0,
+                    630.0,
+                    935.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    934.0,
+                    1295.0,
+                    934.0,
+                    1295.0,
+                    970.0,
+                    294.0,
+                    970.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    330.0,
+                    628.0,
+                    330.0,
+                    628.0,
+                    374.0,
+                    292.0,
+                    374.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    664.0,
+                    330.0,
+                    850.0,
+                    330.0,
+                    850.0,
+                    374.0,
+                    664.0,
+                    374.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    885.0,
+                    330.0,
+                    1332.0,
+                    330.0,
+                    1332.0,
+                    374.0,
+                    885.0,
+                    374.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1354.0,
+                    330.0,
+                    1405.0,
+                    330.0,
+                    1405.0,
+                    374.0,
+                    1354.0,
+                    374.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    593.0,
+                    364.0,
+                    1031.0,
+                    364.0,
+                    1031.0,
+                    412.0,
+                    593.0,
+                    412.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1053.0,
+                    364.0,
+                    1279.0,
+                    364.0,
+                    1279.0,
+                    412.0,
+                    1053.0,
+                    412.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1664.0,
+                    670.0,
+                    1664.0,
+                    670.0,
+                    1700.0,
+                    296.0,
+                    1700.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1968.0,
+                    1166.0,
+                    1968.0,
+                    1166.0,
+                    2008.0,
+                    294.0,
+                    2008.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1194.0,
+                    1968.0,
+                    1407.0,
+                    1968.0,
+                    1407.0,
+                    2008.0,
+                    1194.0,
+                    2008.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    2001.0,
+                    883.0,
+                    2001.0,
+                    883.0,
+                    2037.0,
+                    297.0,
+                    2037.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1018.0,
+                    2001.0,
+                    1252.0,
+                    2001.0,
+                    1252.0,
+                    2037.0,
+                    1018.0,
+                    2037.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1274.0,
+                    2001.0,
+                    1404.0,
+                    2001.0,
+                    1404.0,
+                    2037.0,
+                    1274.0,
+                    2037.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 3,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    295,
+                    1258,
+                    1406,
+                    1258,
+                    1406,
+                    1526,
+                    295,
+                    1526
+                ],
+                "score": 0.98
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    295,
+                    904,
+                    1405,
+                    904,
+                    1405,
+                    1135,
+                    295,
+                    1135
+                ],
+                "score": 0.975
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    757,
+                    1405,
+                    757,
+                    1405,
+                    899,
+                    297,
+                    899
+                ],
+                "score": 0.97
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    622,
+                    1404,
+                    622,
+                    1404,
+                    746,
+                    297,
+                    746
+                ],
+                "score": 0.969
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    307,
+                    1403,
+                    307,
+                    1403,
+                    400,
+                    297,
+                    400
+                ],
+                "score": 0.961
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    292,
+                    228,
+                    1401,
+                    228,
+                    1401,
+                    294,
+                    292,
+                    294
+                ],
+                "score": 0.953
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    296,
+                    408,
+                    1403,
+                    408,
+                    1403,
+                    615,
+                    296,
+                    615
+                ],
+                "score": 0.943
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    295,
+                    1553,
+                    1405,
+                    1553,
+                    1405,
+                    1817,
+                    295,
+                    1817
+                ],
+                "score": 0.938
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    298,
+                    1144,
+                    1403,
+                    1144,
+                    1403,
+                    1236,
+                    298,
+                    1236
+                ],
+                "score": 0.922
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    299,
+                    75,
+                    813,
+                    75,
+                    813,
+                    105,
+                    299,
+                    105
+                ],
+                "score": 0.913
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    1917,
+                    1402,
+                    1917,
+                    1402,
+                    1984,
+                    299,
+                    1984
+                ],
+                "score": 0.91
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    330,
+                    2003,
+                    1114,
+                    2003,
+                    1114,
+                    2035,
+                    330,
+                    2035
+                ],
+                "score": 0.9
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    302,
+                    1826,
+                    1398,
+                    1826,
+                    1398,
+                    1909,
+                    302,
+                    1909
+                ],
+                "score": 0.872
+            },
+            {
+                "category_id": 9,
+                "poly": [
+                    1365,
+                    430,
+                    1401,
+                    430,
+                    1401,
+                    461,
+                    1365,
+                    461
+                ],
+                "score": 0.856
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    841,
+                    2088,
+                    859,
+                    2088,
+                    859,
+                    2112,
+                    841,
+                    2112
+                ],
+                "score": 0.767
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    770,
+                    1455,
+                    920,
+                    1455,
+                    920,
+                    1496,
+                    770,
+                    1496
+                ],
+                "score": 0.94,
+                "latex": "\\tilde { B } _ { \\phi } \\triangleq B _ { \\phi } \\sqrt { \\bar { k } }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    617,
+                    796,
+                    783,
+                    796,
+                    783,
+                    841,
+                    617,
+                    841
+                ],
+                "score": 0.94,
+                "latex": "\\widehat { \\mathrm { I P M } } ( d _ { \\phi } ^ { \\pi } , d _ { \\phi } ^ { D } ) ^ { 2 }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1020,
+                    1023,
+                    1254,
+                    1023,
+                    1254,
+                    1065,
+                    1020,
+                    1065
+                ],
+                "score": 0.93,
+                "latex": "B _ { \\phi } = \\| f _ { \\phi , \\widehat { R } _ { z } , \\widehat { T } _ { z } } \\| _ { \\mathcal { H } _ { k } }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    883,
+                    1947,
+                    1090,
+                    1947,
+                    1090,
+                    1986,
+                    883,
+                    1986
+                ],
+                "score": 0.93,
+                "latex": "\\mathrm { I P M } _ { \\mathcal { G } _ { t } } ( p _ { \\phi , t } ^ { F } , p _ { \\phi , t } ^ { C F } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    352,
+                    758,
+                    534,
+                    758,
+                    534,
+                    799,
+                    352,
+                    799
+                ],
+                "score": 0.93,
+                "latex": "\\mathrm { I P M } _ { \\mathcal { G } } ( d _ { \\phi } ^ { \\pi } , d _ { \\phi } ^ { D } ) ^ { 2 }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    961,
+                    1322,
+                    1115,
+                    1322,
+                    1115,
+                    1360,
+                    961,
+                    1360
+                ],
+                "score": 0.92,
+                "latex": "\\| f _ { \\phi , \\widehat { R } _ { z } , \\widehat { T } _ { z } } \\| _ { \\mathcal { H } _ { k } }"
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    310,
+                    406,
+                    1409,
+                    406,
+                    1409,
+                    614,
+                    310,
+                    614
+                ],
+                "score": 0.92,
+                "latex": "\\begin{array} { r l r } { \\mathrm { P M } _ { \\mathcal { G } } ( d _ { \\phi } ^ { \\pi } , d _ { \\phi } ^ { \\mathcal { D } } ) ^ { 2 } = \\mathbb { E } _ { s _ { 0 } \\sim d _ { 0 } , a _ { 0 } \\sim \\pi ( s _ { 0 } ) , ( s , a , s ^ { \\prime } ) \\sim d ^ { \\mathcal { D } } , a ^ { \\prime } \\sim \\pi ( s ^ { \\prime } ) } } & { } & { ( 5 ) } \\\\ { \\bar { s } _ { 0 } \\sim d _ { 0 } , \\bar { u } _ { 0 } \\sim d _ { 0 } ) , ( \\bar { s } _ { 0 } , \\bar { u } , s ^ { \\prime } ) \\sim d ^ { \\mathcal { D } } , a ^ { \\prime } \\sim \\pi ( \\bar { s } ^ { \\prime } ) } & { } & { } \\\\ { k \\big ( \\phi ( s , a ) , \\phi ( \\bar { s } , \\bar { a } ) \\big ) + ( 1 - \\gamma ) ^ { 2 } k \\big ( \\phi ( s _ { 0 } , a _ { 0 } ) , \\phi ( \\bar { s } _ { 0 } , \\bar { a } _ { 0 } ) \\big ) + \\gamma ^ { 2 } k \\big ( \\phi ( s ^ { \\prime } , a ^ { \\prime } ) , \\phi ( \\bar { s } ^ { \\prime } , \\bar { a } ^ { \\prime } ) \\big ) } & { } & { } \\\\ { - 2 ( 1 - \\gamma ) k \\big ( \\phi ( s _ { 0 } , a _ { 0 } ) , \\phi ( \\bar { s } , \\bar { a } ) \\big ) - 2 \\gamma k \\big ( \\phi ( s ^ { \\prime } , a ^ { \\prime } ) , \\phi ( \\bar { s } , \\bar { a } ) \\big ) + 2 \\gamma ( 1 - \\gamma ) k \\big ( \\phi ( s ^ { \\prime } , a ^ { \\prime } ) , \\phi ( \\bar { s } _ { 0 } , \\bar { a } _ { 0 } ) \\big ) \\bigg ] . } & { } & \\end{array}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1048,
+                    1359,
+                    1143,
+                    1359,
+                    1143,
+                    1395,
+                    1048,
+                    1395
+                ],
+                "score": 0.91,
+                "latex": "f _ { \\phi , \\widehat { R } _ { z } , \\widehat { T } _ { z } }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1151,
+                    1422,
+                    1225,
+                    1422,
+                    1225,
+                    1454,
+                    1151,
+                    1454
+                ],
+                "score": 0.91,
+                "latex": "\\bar { k } = 1"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    460,
+                    1070,
+                    814,
+                    1070,
+                    814,
+                    1108,
+                    460,
+                    1108
+                ],
+                "score": 0.91,
+                "latex": "\\begin{array} { r } { \\bar { \\mathcal { E } } = \\operatorname* { s u p } _ { s \\in { \\mathcal { S } } , a \\in \\mathcal { A } } \\mathcal { E } _ { \\phi , \\widehat { R } _ { z } , \\widehat { T } _ { z } } ( s , a ) } \\end{array}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    483,
+                    941,
+                    858,
+                    941,
+                    858,
+                    979,
+                    483,
+                    979
+                ],
+                "score": 0.91,
+                "latex": "( \\widehat { T } , \\widehat { R } ) = \\langle \\widehat { T } _ { z } \\circ \\phi , \\widehat { R } _ { z } \\circ \\phi \\rangle"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    295,
+                    1013,
+                    956,
+                    1013,
+                    956,
+                    1069,
+                    295,
+                    1069
+                ],
+                "score": 0.9,
+                "latex": "\\begin{array} { r } { \\mathbb { E } _ { T , \\pi } \\Big [ \\sum _ { t = 0 } ^ { \\infty } \\gamma ^ { t } \\mathcal { E } _ { \\phi , \\widehat { R } _ { z } , \\widehat { T } _ { z } } \\big ( s _ { t } , a _ { t } \\big ) \\Big | \\big ( s _ { 0 } , a _ { 0 } \\big ) = \\phi ^ { - 1 } \\big ( z ) \\Big ] \\in \\mathcal { H } _ { k } } \\end{array}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    297,
+                    1708,
+                    744,
+                    1708,
+                    744,
+                    1747,
+                    297,
+                    1747
+                ],
+                "score": 0.9,
+                "latex": "p _ { \\phi , t } ^ { F } ( z ) = \\bar { \\mathrm { P r } } ( z _ { t } | M , \\mu , a _ { 0 : t } = \\pi ( s _ { 0 : t } ) )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    794,
+                    1392,
+                    832,
+                    1392,
+                    832,
+                    1425,
+                    794,
+                    1425
+                ],
+                "score": 0.9,
+                "latex": "B _ { \\phi }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    347,
+                    1462,
+                    386,
+                    1462,
+                    386,
+                    1495,
+                    347,
+                    1495
+                ],
+                "score": 0.9,
+                "latex": "B _ { \\phi }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1225,
+                    977,
+                    1402,
+                    977,
+                    1402,
+                    1018,
+                    1225,
+                    1018
+                ],
+                "score": 0.9,
+                "latex": "f _ { \\phi , \\widehat { R } _ { z } , \\widehat { T } _ { z } } \\left( \\widehat { z } \\right) \\ ="
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    799,
+                    976,
+                    1049,
+                    976,
+                    1049,
+                    1011,
+                    799,
+                    1011
+                ],
+                "score": 0.89,
+                "latex": "\\mathrm { s u p } _ { z \\in \\mathcal { Z } } k ( z , z ) ~ = ~ \\bar { k }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    700,
+                    1323,
+                    739,
+                    1323,
+                    739,
+                    1355,
+                    700,
+                    1355
+                ],
+                "score": 0.89,
+                "latex": "B _ { \\phi }"
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    310,
+                    1138,
+                    1406,
+                    1138,
+                    1406,
+                    1238,
+                    310,
+                    1238
+                ],
+                "score": 0.89,
+                "latex": "R ^ { \\pi } - { \\widehat R } ^ { \\pi } \\Big | \\leq \\frac { 1 } { n } \\sum _ { ( s , a ) \\in { \\mathcal D } } \\mathcal { E } _ { \\phi , { \\widehat R } _ { z } , { \\widehat T } _ { z } } ( s , a ) + B _ { \\phi } \\widehat { \\Gamma \\mathrm { M } } \\big ( d _ { \\phi } ^ { \\pi } , d _ { \\phi } ^ { { D } } \\big ) + \\sqrt { \\frac { \\bar { \\mathcal E } ^ { 2 } } { 2 n } \\log \\frac { 1 } { \\delta } } + B _ { \\phi } \\sqrt { \\frac { \\bar { k } } { n } } \\left( 4 + \\sqrt { 8 \\log \\frac { 3 } { \\delta } } \\right) ."
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    553,
+                    340,
+                    593,
+                    340,
+                    593,
+                    369,
+                    553,
+                    369
+                ],
+                "score": 0.89,
+                "latex": "\\mathcal { H } _ { k }"
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    310,
+                    1822,
+                    1410,
+                    1822,
+                    1410,
+                    1912,
+                    310,
+                    1912
+                ],
+                "score": 0.89,
+                "latex": "R ^ { \\pi } - \\widehat { R } ^ { \\pi } \\Big | \\leq ( 1 - \\gamma ) \\sum _ { t = 0 } ^ { \\infty } \\gamma ^ { t } \\left( \\mathbb { E } _ { p ( s _ { t } | M , \\mu , a _ { 0 : t } = \\pi ( s _ { 0 : t } ) ) } [ \\mathcal { E } _ { \\phi , \\widehat { R } _ { z } , \\widehat { T } _ { z } } ( s _ { t } , \\pi ( s _ { t } ) ) ] + B _ { \\phi , t } [ \\mathrm { P M } _ { \\mathcal { G } _ { t } } ( p _ { \\phi , t } ^ { F } , p _ { \\phi , t } ^ { C F } ) \\right) ."
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    410,
+                    1322,
+                    450,
+                    1322,
+                    450,
+                    1353,
+                    410,
+                    1353
+                ],
+                "score": 0.88,
+                "latex": "\\mathcal { H } _ { k }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    760,
+                    1743,
+                    1393,
+                    1743,
+                    1393,
+                    1784,
+                    760,
+                    1784
+                ],
+                "score": 0.87,
+                "latex": "p _ { \\phi , t } ^ { C F } ( z ) = \\mathrm { P r } ( z _ { t } | M , \\mu , a _ { 0 : t - 1 } = \\pi ( s _ { 0 : t - 1 } ) , a _ { t } \\neq \\pi ( s _ { t } ) )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    612,
+                    654,
+                    643,
+                    654,
+                    643,
+                    681,
+                    612,
+                    681
+                ],
+                "score": 0.86,
+                "latex": "d ^ { \\pi }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    583,
+                    310,
+                    610,
+                    310,
+                    610,
+                    337,
+                    583,
+                    337
+                ],
+                "score": 0.84,
+                "latex": "\\mathcal { F }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1298,
+                    1588,
+                    1318,
+                    1588,
+                    1318,
+                    1619,
+                    1298,
+                    1619
+                ],
+                "score": 0.83,
+                "latex": "\\phi"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    947,
+                    262,
+                    973,
+                    262,
+                    973,
+                    289,
+                    947,
+                    289
+                ],
+                "score": 0.83,
+                "latex": "\\mathcal { F }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    905,
+                    904,
+                    940,
+                    904,
+                    940,
+                    940,
+                    905,
+                    940
+                ],
+                "score": 0.83,
+                "latex": "\\widehat { M }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1194,
+                    1711,
+                    1220,
+                    1711,
+                    1220,
+                    1736,
+                    1194,
+                    1736
+                ],
+                "score": 0.83,
+                "latex": "\\mathcal { Z }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1098,
+                    232,
+                    1125,
+                    232,
+                    1125,
+                    258,
+                    1098,
+                    258
+                ],
+                "score": 0.82,
+                "latex": "\\mathcal { F }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    774,
+                    340,
+                    792,
+                    340,
+                    792,
+                    367,
+                    774,
+                    367
+                ],
+                "score": 0.81,
+                "latex": "k"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    474,
+                    1422,
+                    494,
+                    1422,
+                    494,
+                    1454,
+                    474,
+                    1454
+                ],
+                "score": 0.81,
+                "latex": "\\bar { k }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    414,
+                    1680,
+                    439,
+                    1680,
+                    439,
+                    1705,
+                    414,
+                    1705
+                ],
+                "score": 0.79,
+                "latex": "\\mathcal { Z }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    843,
+                    1654,
+                    863,
+                    1654,
+                    863,
+                    1679,
+                    843,
+                    1679
+                ],
+                "score": 0.77,
+                "latex": "\\mu"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    358,
+                    947,
+                    379,
+                    947,
+                    379,
+                    979,
+                    358,
+                    979
+                ],
+                "score": 0.76,
+                "latex": "\\phi"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1178,
+                    1070,
+                    1202,
+                    1070,
+                    1202,
+                    1096,
+                    1178,
+                    1096
+                ],
+                "score": 0.76,
+                "latex": "\\mathcal { D }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1190,
+                    232,
+                    1211,
+                    232,
+                    1211,
+                    260,
+                    1190,
+                    260
+                ],
+                "score": 0.74,
+                "latex": "\\mathcal { G }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    543,
+                    1653,
+                    565,
+                    1653,
+                    565,
+                    1676,
+                    543,
+                    1676
+                ],
+                "score": 0.74,
+                "latex": "\\pi"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    646,
+                    979,
+                    667,
+                    979,
+                    667,
+                    1004,
+                    646,
+                    1004
+                ],
+                "score": 0.74,
+                "latex": "k"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    675,
+                    237,
+                    692,
+                    237,
+                    692,
+                    258,
+                    675,
+                    258
+                ],
+                "score": 0.73,
+                "latex": "z"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    297,
+                    231,
+                    331,
+                    231,
+                    331,
+                    258,
+                    297,
+                    258
+                ],
+                "score": 0.72,
+                "latex": "M"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    856,
+                    1682,
+                    871,
+                    1682,
+                    871,
+                    1706,
+                    856,
+                    1706
+                ],
+                "score": 0.72,
+                "latex": "t"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1373,
+                    1683,
+                    1393,
+                    1683,
+                    1393,
+                    1707,
+                    1373,
+                    1707
+                ],
+                "score": 0.68,
+                "latex": "\\pi"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1065,
+                    1651,
+                    1079,
+                    1651,
+                    1079,
+                    1676,
+                    1065,
+                    1676
+                ],
+                "score": 0.67,
+                "latex": "t"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    297,
+                    1103,
+                    375,
+                    1103,
+                    375,
+                    1134,
+                    297,
+                    1134
+                ],
+                "score": 0.64,
+                "latex": "1 - 2 \\delta"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1078,
+                    945,
+                    1353,
+                    945,
+                    1353,
+                    978,
+                    1078,
+                    978
+                ],
+                "score": 0.63,
+                "latex": "\\begin{array} { r } { \\mathcal { H } _ { k } \\quad \\subset \\quad ( \\mathcal { Z } \\quad \\to \\quad \\mathbb { R } ) } \\end{array}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    693,
+                    912,
+                    726,
+                    912,
+                    726,
+                    939,
+                    693,
+                    939
+                ],
+                "score": 0.62,
+                "latex": "M"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1077,
+                    946,
+                    1119,
+                    946,
+                    1119,
+                    977,
+                    1077,
+                    977
+                ],
+                "score": 0.53,
+                "latex": "\\mathcal { H } _ { k }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    734,
+                    1749,
+                    748,
+                    1749,
+                    748,
+                    1774,
+                    734,
+                    1774
+                ],
+                "score": 0.48,
+                "latex": "t"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    73.0,
+                    815.0,
+                    73.0,
+                    815.0,
+                    108.0,
+                    295.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    331.0,
+                    1998.0,
+                    1117.0,
+                    1998.0,
+                    1117.0,
+                    2041.0,
+                    331.0,
+                    2041.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    838.0,
+                    2085.0,
+                    862.0,
+                    2085.0,
+                    862.0,
+                    2119.0,
+                    838.0,
+                    2119.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1258.0,
+                    1407.0,
+                    1258.0,
+                    1407.0,
+                    1297.0,
+                    294.0,
+                    1297.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1291.0,
+                    1406.0,
+                    1291.0,
+                    1406.0,
+                    1327.0,
+                    294.0,
+                    1327.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    290.0,
+                    1318.0,
+                    409.0,
+                    1318.0,
+                    409.0,
+                    1365.0,
+                    290.0,
+                    1365.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    451.0,
+                    1318.0,
+                    699.0,
+                    1318.0,
+                    699.0,
+                    1365.0,
+                    451.0,
+                    1365.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    740.0,
+                    1318.0,
+                    960.0,
+                    1318.0,
+                    960.0,
+                    1365.0,
+                    740.0,
+                    1365.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1116.0,
+                    1318.0,
+                    1407.0,
+                    1318.0,
+                    1407.0,
+                    1365.0,
+                    1116.0,
+                    1365.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1350.0,
+                    1047.0,
+                    1350.0,
+                    1047.0,
+                    1401.0,
+                    291.0,
+                    1401.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1144.0,
+                    1350.0,
+                    1408.0,
+                    1350.0,
+                    1408.0,
+                    1401.0,
+                    1144.0,
+                    1401.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1391.0,
+                    793.0,
+                    1391.0,
+                    793.0,
+                    1426.0,
+                    295.0,
+                    1426.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    833.0,
+                    1391.0,
+                    1404.0,
+                    1391.0,
+                    1404.0,
+                    1426.0,
+                    833.0,
+                    1426.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1425.0,
+                    473.0,
+                    1425.0,
+                    473.0,
+                    1460.0,
+                    294.0,
+                    1460.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    495.0,
+                    1425.0,
+                    1150.0,
+                    1425.0,
+                    1150.0,
+                    1460.0,
+                    495.0,
+                    1460.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1226.0,
+                    1425.0,
+                    1406.0,
+                    1425.0,
+                    1406.0,
+                    1460.0,
+                    1226.0,
+                    1460.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1456.0,
+                    346.0,
+                    1456.0,
+                    346.0,
+                    1499.0,
+                    292.0,
+                    1499.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    387.0,
+                    1456.0,
+                    769.0,
+                    1456.0,
+                    769.0,
+                    1499.0,
+                    387.0,
+                    1499.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    921.0,
+                    1456.0,
+                    1406.0,
+                    1456.0,
+                    1406.0,
+                    1499.0,
+                    921.0,
+                    1499.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1493.0,
+                    804.0,
+                    1493.0,
+                    804.0,
+                    1529.0,
+                    295.0,
+                    1529.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    902.0,
+                    692.0,
+                    902.0,
+                    692.0,
+                    947.0,
+                    293.0,
+                    947.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    727.0,
+                    902.0,
+                    904.0,
+                    902.0,
+                    904.0,
+                    947.0,
+                    727.0,
+                    947.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    941.0,
+                    902.0,
+                    1406.0,
+                    902.0,
+                    1406.0,
+                    947.0,
+                    941.0,
+                    947.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    290.0,
+                    937.0,
+                    357.0,
+                    937.0,
+                    357.0,
+                    983.0,
+                    290.0,
+                    983.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    380.0,
+                    937.0,
+                    482.0,
+                    937.0,
+                    482.0,
+                    983.0,
+                    380.0,
+                    983.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    859.0,
+                    937.0,
+                    1076.0,
+                    937.0,
+                    1076.0,
+                    983.0,
+                    859.0,
+                    983.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1354.0,
+                    937.0,
+                    1406.0,
+                    937.0,
+                    1406.0,
+                    983.0,
+                    1354.0,
+                    983.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    970.0,
+                    645.0,
+                    970.0,
+                    645.0,
+                    1017.0,
+                    292.0,
+                    1017.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    668.0,
+                    970.0,
+                    798.0,
+                    970.0,
+                    798.0,
+                    1017.0,
+                    668.0,
+                    1017.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1050.0,
+                    970.0,
+                    1224.0,
+                    970.0,
+                    1224.0,
+                    1017.0,
+                    1050.0,
+                    1017.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1403.0,
+                    970.0,
+                    1406.0,
+                    970.0,
+                    1406.0,
+                    1017.0,
+                    1403.0,
+                    1017.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    285.0,
+                    1008.0,
+                    294.0,
+                    1008.0,
+                    294.0,
+                    1073.0,
+                    285.0,
+                    1073.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    957.0,
+                    1008.0,
+                    1019.0,
+                    1008.0,
+                    1019.0,
+                    1073.0,
+                    957.0,
+                    1073.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1255.0,
+                    1008.0,
+                    1412.0,
+                    1008.0,
+                    1412.0,
+                    1073.0,
+                    1255.0,
+                    1073.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    289.0,
+                    1057.0,
+                    459.0,
+                    1057.0,
+                    459.0,
+                    1113.0,
+                    289.0,
+                    1113.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    815.0,
+                    1057.0,
+                    1177.0,
+                    1057.0,
+                    1177.0,
+                    1113.0,
+                    815.0,
+                    1113.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1203.0,
+                    1057.0,
+                    1408.0,
+                    1057.0,
+                    1408.0,
+                    1113.0,
+                    1203.0,
+                    1113.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    290.0,
+                    1097.0,
+                    296.0,
+                    1097.0,
+                    296.0,
+                    1140.0,
+                    290.0,
+                    1140.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    376.0,
+                    1097.0,
+                    390.0,
+                    1097.0,
+                    390.0,
+                    1140.0,
+                    376.0,
+                    1140.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    756.0,
+                    351.0,
+                    756.0,
+                    351.0,
+                    799.0,
+                    295.0,
+                    799.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    535.0,
+                    756.0,
+                    1406.0,
+                    756.0,
+                    1406.0,
+                    799.0,
+                    535.0,
+                    799.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    800.0,
+                    616.0,
+                    800.0,
+                    616.0,
+                    839.0,
+                    294.0,
+                    839.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    784.0,
+                    800.0,
+                    1406.0,
+                    800.0,
+                    1406.0,
+                    839.0,
+                    784.0,
+                    839.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    836.0,
+                    1406.0,
+                    836.0,
+                    1406.0,
+                    871.0,
+                    295.0,
+                    871.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    868.0,
+                    662.0,
+                    868.0,
+                    662.0,
+                    900.0,
+                    296.0,
+                    900.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    620.0,
+                    1404.0,
+                    620.0,
+                    1404.0,
+                    658.0,
+                    292.0,
+                    658.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    649.0,
+                    611.0,
+                    649.0,
+                    611.0,
+                    691.0,
+                    294.0,
+                    691.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    644.0,
+                    649.0,
+                    1405.0,
+                    649.0,
+                    1405.0,
+                    691.0,
+                    644.0,
+                    691.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    680.0,
+                    1408.0,
+                    680.0,
+                    1408.0,
+                    723.0,
+                    292.0,
+                    723.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    714.0,
+                    1025.0,
+                    714.0,
+                    1025.0,
+                    747.0,
+                    296.0,
+                    747.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    308.0,
+                    582.0,
+                    308.0,
+                    582.0,
+                    341.0,
+                    293.0,
+                    341.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    611.0,
+                    308.0,
+                    1404.0,
+                    308.0,
+                    1404.0,
+                    341.0,
+                    611.0,
+                    341.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    337.0,
+                    552.0,
+                    337.0,
+                    552.0,
+                    373.0,
+                    294.0,
+                    373.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    594.0,
+                    337.0,
+                    773.0,
+                    337.0,
+                    773.0,
+                    373.0,
+                    594.0,
+                    373.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    793.0,
+                    337.0,
+                    1405.0,
+                    337.0,
+                    1405.0,
+                    373.0,
+                    793.0,
+                    373.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    369.0,
+                    717.0,
+                    369.0,
+                    717.0,
+                    403.0,
+                    295.0,
+                    403.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    332.0,
+                    226.0,
+                    674.0,
+                    226.0,
+                    674.0,
+                    268.0,
+                    332.0,
+                    268.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    693.0,
+                    226.0,
+                    1097.0,
+                    226.0,
+                    1097.0,
+                    268.0,
+                    693.0,
+                    268.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1126.0,
+                    226.0,
+                    1189.0,
+                    226.0,
+                    1189.0,
+                    268.0,
+                    1126.0,
+                    268.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1212.0,
+                    226.0,
+                    1404.0,
+                    226.0,
+                    1404.0,
+                    268.0,
+                    1212.0,
+                    268.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    261.0,
+                    946.0,
+                    261.0,
+                    946.0,
+                    293.0,
+                    295.0,
+                    293.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    974.0,
+                    261.0,
+                    982.0,
+                    261.0,
+                    982.0,
+                    293.0,
+                    974.0,
+                    293.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1555.0,
+                    1403.0,
+                    1555.0,
+                    1403.0,
+                    1590.0,
+                    294.0,
+                    1590.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1588.0,
+                    1297.0,
+                    1588.0,
+                    1297.0,
+                    1619.0,
+                    295.0,
+                    1619.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1319.0,
+                    1588.0,
+                    1403.0,
+                    1588.0,
+                    1403.0,
+                    1619.0,
+                    1319.0,
+                    1619.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1617.0,
+                    1405.0,
+                    1617.0,
+                    1405.0,
+                    1652.0,
+                    293.0,
+                    1652.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1646.0,
+                    542.0,
+                    1646.0,
+                    542.0,
+                    1682.0,
+                    292.0,
+                    1682.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    566.0,
+                    1646.0,
+                    842.0,
+                    1646.0,
+                    842.0,
+                    1682.0,
+                    566.0,
+                    1682.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    864.0,
+                    1646.0,
+                    1064.0,
+                    1646.0,
+                    1064.0,
+                    1682.0,
+                    864.0,
+                    1682.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1080.0,
+                    1646.0,
+                    1406.0,
+                    1646.0,
+                    1406.0,
+                    1682.0,
+                    1080.0,
+                    1682.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    290.0,
+                    1674.0,
+                    413.0,
+                    1674.0,
+                    413.0,
+                    1717.0,
+                    290.0,
+                    1717.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    440.0,
+                    1674.0,
+                    855.0,
+                    1674.0,
+                    855.0,
+                    1717.0,
+                    440.0,
+                    1717.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    872.0,
+                    1674.0,
+                    1372.0,
+                    1674.0,
+                    1372.0,
+                    1717.0,
+                    872.0,
+                    1717.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1394.0,
+                    1674.0,
+                    1406.0,
+                    1674.0,
+                    1406.0,
+                    1717.0,
+                    1394.0,
+                    1717.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1703.0,
+                    296.0,
+                    1703.0,
+                    296.0,
+                    1749.0,
+                    292.0,
+                    1749.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    745.0,
+                    1703.0,
+                    1193.0,
+                    1703.0,
+                    1193.0,
+                    1749.0,
+                    745.0,
+                    1749.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1221.0,
+                    1703.0,
+                    1410.0,
+                    1703.0,
+                    1410.0,
+                    1749.0,
+                    1221.0,
+                    1749.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    289.0,
+                    1738.0,
+                    733.0,
+                    1738.0,
+                    733.0,
+                    1790.0,
+                    289.0,
+                    1790.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    749.0,
+                    1738.0,
+                    759.0,
+                    1738.0,
+                    759.0,
+                    1790.0,
+                    749.0,
+                    1790.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1394.0,
+                    1738.0,
+                    1407.0,
+                    1738.0,
+                    1407.0,
+                    1790.0,
+                    1394.0,
+                    1790.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1782.0,
+                    755.0,
+                    1782.0,
+                    755.0,
+                    1818.0,
+                    295.0,
+                    1818.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1915.0,
+                    1404.0,
+                    1915.0,
+                    1404.0,
+                    1953.0,
+                    294.0,
+                    1953.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    288.0,
+                    1934.0,
+                    882.0,
+                    1934.0,
+                    882.0,
+                    1995.0,
+                    288.0,
+                    1995.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1091.0,
+                    1934.0,
+                    1412.0,
+                    1934.0,
+                    1412.0,
+                    1995.0,
+                    1091.0,
+                    1995.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 4,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    228,
+                    1405,
+                    228,
+                    1405,
+                    542,
+                    297,
+                    542
+                ],
+                "score": 0.983
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1838,
+                    1403,
+                    1838,
+                    1403,
+                    2035,
+                    297,
+                    2035
+                ],
+                "score": 0.98
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    729,
+                    1402,
+                    729,
+                    1402,
+                    891,
+                    298,
+                    891
+                ],
+                "score": 0.98
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1645,
+                    1403,
+                    1645,
+                    1403,
+                    1807,
+                    297,
+                    1807
+                ],
+                "score": 0.978
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    1375,
+                    1406,
+                    1375,
+                    1406,
+                    1468,
+                    299,
+                    1468
+                ],
+                "score": 0.975
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1129,
+                    1404,
+                    1129,
+                    1404,
+                    1227,
+                    298,
+                    1227
+                ],
+                "score": 0.97
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    300,
+                    1021,
+                    1405,
+                    1021,
+                    1405,
+                    1115,
+                    300,
+                    1115
+                ],
+                "score": 0.962
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    311,
+                    905,
+                    1340,
+                    905,
+                    1340,
+                    1010,
+                    311,
+                    1010
+                ],
+                "score": 0.955
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    296,
+                    633,
+                    1401,
+                    633,
+                    1401,
+                    697,
+                    296,
+                    697
+                ],
+                "score": 0.951
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    343,
+                    1544,
+                    981,
+                    1544,
+                    981,
+                    1635,
+                    343,
+                    1635
+                ],
+                "score": 0.94
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    507,
+                    1305,
+                    1189,
+                    1305,
+                    1189,
+                    1365,
+                    507,
+                    1365
+                ],
+                "score": 0.938
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    300,
+                    75,
+                    813,
+                    75,
+                    813,
+                    104,
+                    300,
+                    104
+                ],
+                "score": 0.902
+            },
+            {
+                "category_id": 9,
+                "poly": [
+                    1366,
+                    1319,
+                    1400,
+                    1319,
+                    1400,
+                    1349,
+                    1366,
+                    1349
+                ],
+                "score": 0.888
+            },
+            {
+                "category_id": 9,
+                "poly": [
+                    1366,
+                    1574,
+                    1400,
+                    1574,
+                    1400,
+                    1603,
+                    1366,
+                    1603
+                ],
+                "score": 0.867
+            },
+            {
+                "category_id": 9,
+                "poly": [
+                    1366,
+                    1497,
+                    1400,
+                    1497,
+                    1400,
+                    1527,
+                    1366,
+                    1527
+                ],
+                "score": 0.851
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    840,
+                    2089,
+                    859,
+                    2089,
+                    859,
+                    2112,
+                    840,
+                    2112
+                ],
+                "score": 0.792
+            },
+            {
+                "category_id": 9,
+                "poly": [
+                    1366,
+                    922,
+                    1400,
+                    922,
+                    1400,
+                    953,
+                    1366,
+                    953
+                ],
+                "score": 0.706
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    294,
+                    576,
+                    1298,
+                    576,
+                    1298,
+                    610,
+                    294,
+                    610
+                ],
+                "score": 0.676
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    325,
+                    1482,
+                    1337,
+                    1482,
+                    1337,
+                    1540,
+                    325,
+                    1540
+                ],
+                "score": 0.656
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1257,
+                    1140,
+                    1257,
+                    1140,
+                    1294,
+                    298,
+                    1294
+                ],
+                "score": 0.618
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    323,
+                    1482,
+                    1336,
+                    1482,
+                    1336,
+                    1540,
+                    323,
+                    1540
+                ],
+                "score": 0.54
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    294,
+                    576,
+                    1298,
+                    576,
+                    1298,
+                    610,
+                    294,
+                    610
+                ],
+                "score": 0.307
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    298,
+                    1257,
+                    1140,
+                    1257,
+                    1140,
+                    1294,
+                    298,
+                    1294
+                ],
+                "score": 0.269
+            },
+            {
+                "category_id": 9,
+                "poly": [
+                    1366,
+                    923,
+                    1400,
+                    923,
+                    1400,
+                    953,
+                    1366,
+                    953
+                ],
+                "score": 0.096
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    823,
+                    259,
+                    1030,
+                    259,
+                    1030,
+                    299,
+                    823,
+                    299
+                ],
+                "score": 0.94,
+                "latex": "\\mathrm { I P M } _ { \\mathcal { G } _ { t } } ( p _ { \\phi , t } ^ { F } , p _ { \\phi , t } ^ { C F } )"
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    324,
+                    1479,
+                    1339,
+                    1479,
+                    1339,
+                    1640,
+                    324,
+                    1640
+                ],
+                "score": 0.94,
+                "latex": "\\begin{array} { r l } & { \\mathcal { L } _ { \\widehat { M } } ( \\widehat { M } , \\pi , \\alpha _ { M } ) = \\mathbb { E } _ { d ^ { p } } \\Big [ - \\log \\widehat { R } _ { z } \\big ( r | \\phi ( s , a ) \\big ) - \\log \\widehat { T } _ { z } \\big ( s ^ { \\prime } | \\phi ( s , a ) \\big ) \\Big ] + \\alpha _ { M } \\mathrm { I P M } _ { \\mathcal { G } } \\big ( d _ { \\phi } ^ { \\pi } , d _ { \\phi } ^ { { D } } \\big ) , } \\\\ & { \\mathcal { I } _ { \\pi } ( \\pi , \\widehat { M } , \\alpha _ { \\pi } ) = \\mathbb { E } _ { \\widehat { M } , \\pi } \\left[ \\displaystyle \\sum _ { t = 0 } ^ { \\infty } \\gamma ^ { t } r _ { t } \\right] - \\alpha _ { \\pi } \\mathrm { I P M } _ { \\mathcal { G } } \\big ( d _ { \\phi } ^ { \\pi } , d _ { \\phi } ^ { { D } } \\big ) . } \\end{array}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    525,
+                    762,
+                    649,
+                    762,
+                    649,
+                    800,
+                    525,
+                    800
+                ],
+                "score": 0.93,
+                "latex": "| R ^ { \\pi } - { \\widehat { R } } ^ { \\pi } |"
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    508,
+                    1305,
+                    1191,
+                    1305,
+                    1191,
+                    1365,
+                    508,
+                    1365
+                ],
+                "score": 0.93,
+                "latex": "\\begin{array} { r } { { \\cal R } ^ { \\pi } \\geq \\widehat { \\cal R } ^ { \\pi } - \\mathbb { E } _ { ( s , a ) \\sim \\mathcal { D } } \\left[ \\mathcal { E } _ { \\phi , \\widehat { \\cal R } _ { z } , \\widehat { T } _ { z } } ( s , a ) \\right] - { \\cal B } _ { \\phi } \\mathrm { I P M } _ { \\mathcal { G } } ( d _ { \\phi } ^ { \\pi } , d _ { \\phi } ^ { \\mathcal { D } } ) . } \\end{array}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    714,
+                    1872,
+                    790,
+                    1872,
+                    790,
+                    1909,
+                    714,
+                    1909
+                ],
+                "score": 0.93,
+                "latex": "( \\widehat { T } , \\widehat { R } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    810,
+                    1941,
+                    885,
+                    1941,
+                    885,
+                    1976,
+                    810,
+                    1976
+                ],
+                "score": 0.93,
+                "latex": "( \\widehat { T } , \\widehat { R } )"
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    311,
+                    903,
+                    1353,
+                    903,
+                    1353,
+                    1010,
+                    311,
+                    1010
+                ],
+                "score": 0.93,
+                "latex": "\\operatorname* { m i n } _ { \\phi , \\widehat { R } _ { z } , \\widehat { T } _ { z } } \\frac { 1 } { n } \\sum _ { \\substack { ( s , a , s ^ { \\prime } , r ) \\in \\mathcal { D } } } \\Big [ \\underbrace { - \\log \\widehat { R } _ { z } \\big ( r | \\phi ( s , a ) \\big ) } _ { D _ { \\mathrm { K L } } ( R ( r | s , a ) | | \\widehat { R } _ { z } ( r | \\phi ( s , a ) ) ) D _ { \\mathrm { K L } } ( T ( s ^ { \\prime } | s , a ) | | \\widehat { T } _ { z } ( s ^ { \\prime } | \\phi ( s , a ) ) ) } \\Big ] + \\alpha _ { M } \\widehat { \\mathrm { I P M } } ( d _ { \\phi } ^ { \\pi } , d _ { \\phi } ^ { \\mathcal { D } } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    491,
+                    295,
+                    711,
+                    295,
+                    711,
+                    330,
+                    491,
+                    330
+                ],
+                "score": 0.92,
+                "latex": "a _ { 0 : t - 1 } = \\pi ( s _ { 0 : t - 1 } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1041,
+                    1127,
+                    1118,
+                    1127,
+                    1118,
+                    1168,
+                    1041,
+                    1168
+                ],
+                "score": 0.92,
+                "latex": "( \\widehat { T } , \\widehat { R } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    620,
+                    1084,
+                    716,
+                    1084,
+                    716,
+                    1113,
+                    620,
+                    1113
+                ],
+                "score": 0.91,
+                "latex": "\\alpha _ { M } = 0"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    625,
+                    1713,
+                    673,
+                    1713,
+                    673,
+                    1750,
+                    625,
+                    1750
+                ],
+                "score": 0.9,
+                "latex": "{ \\mathcal { L } } _ { \\widehat { M } }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    446,
+                    1022,
+                    484,
+                    1022,
+                    484,
+                    1055,
+                    446,
+                    1055
+                ],
+                "score": 0.9,
+                "latex": "B _ { \\phi }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1167,
+                    1714,
+                    1203,
+                    1714,
+                    1203,
+                    1744,
+                    1167,
+                    1744
+                ],
+                "score": 0.89,
+                "latex": "\\mathcal { I } _ { \\pi }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    849,
+                    1707,
+                    882,
+                    1707,
+                    882,
+                    1741,
+                    849,
+                    1741
+                ],
+                "score": 0.87,
+                "latex": "\\widehat { M }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    794,
+                    1056,
+                    839,
+                    1056,
+                    839,
+                    1082,
+                    794,
+                    1082
+                ],
+                "score": 0.87,
+                "latex": "\\alpha _ { M }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    805,
+                    1776,
+                    842,
+                    1776,
+                    842,
+                    1802,
+                    805,
+                    1802
+                ],
+                "score": 0.87,
+                "latex": "R ^ { \\pi }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1121,
+                    446,
+                    1156,
+                    446,
+                    1156,
+                    476,
+                    1121,
+                    476
+                ],
+                "score": 0.87,
+                "latex": "d ^ { \\bar { D } }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    528,
+                    1408,
+                    566,
+                    1408,
+                    566,
+                    1435,
+                    528,
+                    1435
+                ],
+                "score": 0.86,
+                "latex": "R ^ { \\pi }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    812,
+                    449,
+                    843,
+                    449,
+                    843,
+                    477,
+                    812,
+                    477
+                ],
+                "score": 0.86,
+                "latex": "d ^ { \\pi }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    861,
+                    511,
+                    892,
+                    511,
+                    892,
+                    537,
+                    861,
+                    537
+                ],
+                "score": 0.85,
+                "latex": "d ^ { \\pi }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    664,
+                    1141,
+                    686,
+                    1141,
+                    686,
+                    1162,
+                    664,
+                    1162
+                ],
+                "score": 0.77,
+                "latex": "\\pi"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    743,
+                    237,
+                    763,
+                    237,
+                    763,
+                    258,
+                    743,
+                    258
+                ],
+                "score": 0.73,
+                "latex": "\\pi"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    413,
+                    330,
+                    427,
+                    330,
+                    427,
+                    354,
+                    413,
+                    354
+                ],
+                "score": 0.71,
+                "latex": "t"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1381,
+                    1719,
+                    1401,
+                    1719,
+                    1401,
+                    1741,
+                    1381,
+                    1741
+                ],
+                "score": 0.71,
+                "latex": "\\pi"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    415,
+                    302,
+                    435,
+                    302,
+                    435,
+                    325,
+                    415,
+                    325
+                ],
+                "score": 0.7,
+                "latex": "\\pi"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    72.0,
+                    815.0,
+                    72.0,
+                    815.0,
+                    108.0,
+                    295.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    840.0,
+                    2087.0,
+                    862.0,
+                    2087.0,
+                    862.0,
+                    2118.0,
+                    840.0,
+                    2118.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    575.0,
+                    1303.0,
+                    575.0,
+                    1303.0,
+                    613.0,
+                    292.0,
+                    613.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1253.0,
+                    1145.0,
+                    1253.0,
+                    1145.0,
+                    1302.0,
+                    293.0,
+                    1302.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    230.0,
+                    742.0,
+                    230.0,
+                    742.0,
+                    266.0,
+                    294.0,
+                    266.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    764.0,
+                    230.0,
+                    1405.0,
+                    230.0,
+                    1405.0,
+                    266.0,
+                    764.0,
+                    266.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    287.0,
+                    249.0,
+                    822.0,
+                    249.0,
+                    822.0,
+                    310.0,
+                    287.0,
+                    310.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1031.0,
+                    249.0,
+                    1412.0,
+                    249.0,
+                    1412.0,
+                    310.0,
+                    1031.0,
+                    310.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    294.0,
+                    414.0,
+                    294.0,
+                    414.0,
+                    335.0,
+                    294.0,
+                    335.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    436.0,
+                    294.0,
+                    490.0,
+                    294.0,
+                    490.0,
+                    335.0,
+                    436.0,
+                    335.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    712.0,
+                    294.0,
+                    1405.0,
+                    294.0,
+                    1405.0,
+                    335.0,
+                    712.0,
+                    335.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    328.0,
+                    412.0,
+                    328.0,
+                    412.0,
+                    360.0,
+                    295.0,
+                    360.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    428.0,
+                    328.0,
+                    1402.0,
+                    328.0,
+                    1402.0,
+                    360.0,
+                    428.0,
+                    360.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    358.0,
+                    1402.0,
+                    358.0,
+                    1402.0,
+                    390.0,
+                    295.0,
+                    390.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    386.0,
+                    1403.0,
+                    386.0,
+                    1403.0,
+                    423.0,
+                    295.0,
+                    423.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    411.0,
+                    1406.0,
+                    411.0,
+                    1406.0,
+                    458.0,
+                    291.0,
+                    458.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    446.0,
+                    811.0,
+                    446.0,
+                    811.0,
+                    483.0,
+                    294.0,
+                    483.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    844.0,
+                    446.0,
+                    1120.0,
+                    446.0,
+                    1120.0,
+                    483.0,
+                    844.0,
+                    483.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1157.0,
+                    446.0,
+                    1405.0,
+                    446.0,
+                    1405.0,
+                    483.0,
+                    1157.0,
+                    483.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    479.0,
+                    1403.0,
+                    479.0,
+                    1403.0,
+                    512.0,
+                    296.0,
+                    512.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    509.0,
+                    860.0,
+                    509.0,
+                    860.0,
+                    545.0,
+                    295.0,
+                    545.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    893.0,
+                    509.0,
+                    1402.0,
+                    509.0,
+                    1402.0,
+                    545.0,
+                    893.0,
+                    545.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1838.0,
+                    1405.0,
+                    1838.0,
+                    1405.0,
+                    1876.0,
+                    295.0,
+                    1876.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1868.0,
+                    713.0,
+                    1868.0,
+                    713.0,
+                    1915.0,
+                    291.0,
+                    1915.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    791.0,
+                    1868.0,
+                    1407.0,
+                    1868.0,
+                    1407.0,
+                    1915.0,
+                    791.0,
+                    1915.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1905.0,
+                    1404.0,
+                    1905.0,
+                    1404.0,
+                    1942.0,
+                    295.0,
+                    1942.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1936.0,
+                    809.0,
+                    1936.0,
+                    809.0,
+                    1980.0,
+                    292.0,
+                    1980.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    886.0,
+                    1936.0,
+                    1407.0,
+                    1936.0,
+                    1407.0,
+                    1980.0,
+                    886.0,
+                    1980.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1971.0,
+                    1405.0,
+                    1971.0,
+                    1405.0,
+                    2008.0,
+                    295.0,
+                    2008.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    2000.0,
+                    1405.0,
+                    2000.0,
+                    1405.0,
+                    2040.0,
+                    294.0,
+                    2040.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    729.0,
+                    1404.0,
+                    729.0,
+                    1404.0,
+                    766.0,
+                    296.0,
+                    766.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    764.0,
+                    524.0,
+                    764.0,
+                    524.0,
+                    800.0,
+                    293.0,
+                    800.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    650.0,
+                    764.0,
+                    1404.0,
+                    764.0,
+                    1404.0,
+                    800.0,
+                    650.0,
+                    800.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    796.0,
+                    1406.0,
+                    796.0,
+                    1406.0,
+                    833.0,
+                    293.0,
+                    833.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    828.0,
+                    1402.0,
+                    828.0,
+                    1402.0,
+                    862.0,
+                    294.0,
+                    862.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    859.0,
+                    834.0,
+                    859.0,
+                    834.0,
+                    893.0,
+                    294.0,
+                    893.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1641.0,
+                    1407.0,
+                    1641.0,
+                    1407.0,
+                    1686.0,
+                    294.0,
+                    1686.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1674.0,
+                    1405.0,
+                    1674.0,
+                    1405.0,
+                    1715.0,
+                    294.0,
+                    1715.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1712.0,
+                    624.0,
+                    1712.0,
+                    624.0,
+                    1751.0,
+                    293.0,
+                    1751.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    674.0,
+                    1712.0,
+                    848.0,
+                    1712.0,
+                    848.0,
+                    1751.0,
+                    674.0,
+                    1751.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    883.0,
+                    1712.0,
+                    1166.0,
+                    1712.0,
+                    1166.0,
+                    1751.0,
+                    883.0,
+                    1751.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1204.0,
+                    1712.0,
+                    1380.0,
+                    1712.0,
+                    1380.0,
+                    1751.0,
+                    1204.0,
+                    1751.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1402.0,
+                    1712.0,
+                    1405.0,
+                    1712.0,
+                    1405.0,
+                    1751.0,
+                    1402.0,
+                    1751.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1741.0,
+                    1403.0,
+                    1741.0,
+                    1403.0,
+                    1782.0,
+                    294.0,
+                    1782.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1774.0,
+                    804.0,
+                    1774.0,
+                    804.0,
+                    1808.0,
+                    292.0,
+                    1808.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    843.0,
+                    1774.0,
+                    854.0,
+                    1774.0,
+                    854.0,
+                    1808.0,
+                    843.0,
+                    1808.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1372.0,
+                    1404.0,
+                    1372.0,
+                    1404.0,
+                    1412.0,
+                    294.0,
+                    1412.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1405.0,
+                    527.0,
+                    1405.0,
+                    527.0,
+                    1441.0,
+                    294.0,
+                    1441.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    567.0,
+                    1405.0,
+                    1404.0,
+                    1405.0,
+                    1404.0,
+                    1441.0,
+                    567.0,
+                    1441.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1437.0,
+                    987.0,
+                    1437.0,
+                    987.0,
+                    1471.0,
+                    296.0,
+                    1471.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1128.0,
+                    663.0,
+                    1128.0,
+                    663.0,
+                    1173.0,
+                    292.0,
+                    1173.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    687.0,
+                    1128.0,
+                    1040.0,
+                    1128.0,
+                    1040.0,
+                    1173.0,
+                    687.0,
+                    1173.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1119.0,
+                    1128.0,
+                    1405.0,
+                    1128.0,
+                    1405.0,
+                    1173.0,
+                    1119.0,
+                    1173.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1164.0,
+                    1404.0,
+                    1164.0,
+                    1404.0,
+                    1199.0,
+                    294.0,
+                    1199.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1191.0,
+                    454.0,
+                    1191.0,
+                    454.0,
+                    1230.0,
+                    292.0,
+                    1230.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1019.0,
+                    445.0,
+                    1019.0,
+                    445.0,
+                    1056.0,
+                    295.0,
+                    1056.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    485.0,
+                    1019.0,
+                    1405.0,
+                    1019.0,
+                    1405.0,
+                    1056.0,
+                    485.0,
+                    1056.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1051.0,
+                    793.0,
+                    1051.0,
+                    793.0,
+                    1085.0,
+                    296.0,
+                    1085.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    840.0,
+                    1051.0,
+                    1402.0,
+                    1051.0,
+                    1402.0,
+                    1085.0,
+                    840.0,
+                    1085.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1081.0,
+                    619.0,
+                    1081.0,
+                    619.0,
+                    1118.0,
+                    294.0,
+                    1118.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    717.0,
+                    1081.0,
+                    1293.0,
+                    1081.0,
+                    1293.0,
+                    1118.0,
+                    717.0,
+                    1118.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    633.0,
+                    1401.0,
+                    633.0,
+                    1401.0,
+                    670.0,
+                    292.0,
+                    670.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    666.0,
+                    941.0,
+                    666.0,
+                    941.0,
+                    698.0,
+                    297.0,
+                    698.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    575.0,
+                    1303.0,
+                    575.0,
+                    1303.0,
+                    613.0,
+                    292.0,
+                    613.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1253.0,
+                    1145.0,
+                    1253.0,
+                    1145.0,
+                    1302.0,
+                    293.0,
+                    1302.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 5,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    948,
+                    1405,
+                    948,
+                    1405,
+                    1186,
+                    297,
+                    1186
+                ],
+                "score": 0.979
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1548,
+                    1403,
+                    1548,
+                    1403,
+                    1764,
+                    297,
+                    1764
+                ],
+                "score": 0.978
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1779,
+                    1404,
+                    1779,
+                    1404,
+                    1963,
+                    298,
+                    1963
+                ],
+                "score": 0.975
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1316,
+                    1404,
+                    1316,
+                    1404,
+                    1440,
+                    298,
+                    1440
+                ],
+                "score": 0.975
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    799,
+                    1403,
+                    799,
+                    1403,
+                    934,
+                    298,
+                    934
+                ],
+                "score": 0.972
+            },
+            {
+                "category_id": 3,
+                "poly": [
+                    294,
+                    223,
+                    1406,
+                    223,
+                    1406,
+                    554,
+                    294,
+                    554
+                ],
+                "score": 0.971
+            },
+            {
+                "category_id": 4,
+                "poly": [
+                    296,
+                    575,
+                    1405,
+                    575,
+                    1405,
+                    729,
+                    296,
+                    729
+                ],
+                "score": 0.969
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    331,
+                    2006,
+                    886,
+                    2006,
+                    886,
+                    2034,
+                    331,
+                    2034
+                ],
+                "score": 0.916
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    300,
+                    1241,
+                    557,
+                    1241,
+                    557,
+                    1276,
+                    300,
+                    1276
+                ],
+                "score": 0.904
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    299,
+                    76,
+                    812,
+                    76,
+                    812,
+                    104,
+                    299,
+                    104
+                ],
+                "score": 0.889
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    298,
+                    1489,
+                    854,
+                    1489,
+                    854,
+                    1518,
+                    298,
+                    1518
+                ],
+                "score": 0.859
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    842,
+                    2088,
+                    858,
+                    2088,
+                    858,
+                    2111,
+                    842,
+                    2111
+                ],
+                "score": 0.7
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    841,
+                    2088,
+                    859,
+                    2088,
+                    859,
+                    2111,
+                    841,
+                    2111
+                ],
+                "score": 0.127
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    659,
+                    1057,
+                    1005,
+                    1057,
+                    1005,
+                    1097,
+                    659,
+                    1097
+                ],
+                "score": 0.92,
+                "latex": "\\begin{array} { r } { \\pi _ { 1 } = \\arg \\operatorname* { m i n } _ { \\pi } \\mathcal { L } _ { \\pi } ( \\pi , \\widehat { M } _ { 1 } , \\alpha _ { \\pi } ) } \\end{array}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1000,
+                    978,
+                    1351,
+                    978,
+                    1351,
+                    1023,
+                    1000,
+                    1023
+                ],
+                "score": 0.92,
+                "latex": "{ \\widehat { \\widetilde M } } _ { 0 } = \\arg \\operatorname* { m i n } _ { { \\widehat M } } { \\mathcal { L } } _ { { \\widehat M } } ^ { - } ( { \\widehat M } , \\cdot , 0 )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1025,
+                    866,
+                    1108,
+                    866,
+                    1108,
+                    900,
+                    1025,
+                    900
+                ],
+                "score": 0.91,
+                "latex": "\\mathcal { D } \\cup \\widehat { \\mathcal { D } }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    751,
+                    986,
+                    941,
+                    986,
+                    941,
+                    1017,
+                    751,
+                    1017
+                ],
+                "score": 0.91,
+                "latex": "\\alpha _ { M } = \\alpha _ { \\pi } = 0"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    297,
+                    1018,
+                    641,
+                    1018,
+                    641,
+                    1059,
+                    297,
+                    1059
+                ],
+                "score": 0.9,
+                "latex": "\\pi _ { 0 } = \\arg \\operatorname* { m i n } _ { \\pi } \\mathcal { L } _ { \\pi } ( \\pi , \\widehat { M } _ { 0 } , 0 )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    377,
+                    669,
+                    483,
+                    669,
+                    483,
+                    701,
+                    377,
+                    701
+                ],
+                "score": 0.89,
+                "latex": "\\gamma = 0 . 9 9"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    297,
+                    1059,
+                    613,
+                    1059,
+                    613,
+                    1101,
+                    297,
+                    1101
+                ],
+                "score": 0.89,
+                "latex": "\\mathrm { a r g m i n } _ { \\widehat { M } } \\mathcal { L } _ { \\widehat { M } } ( \\widehat { M } , \\pi _ { 0 } , \\alpha _ { M } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    593,
+                    698,
+                    648,
+                    698,
+                    648,
+                    728,
+                    593,
+                    728
+                ],
+                "score": 0.89,
+                "latex": "9 5 \\%"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    838,
+                    669,
+                    944,
+                    669,
+                    944,
+                    697,
+                    838,
+                    697
+                ],
+                "score": 0.88,
+                "latex": "t = 1 0 0 0"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1320,
+                    1019,
+                    1404,
+                    1019,
+                    1404,
+                    1058,
+                    1320,
+                    1058
+                ],
+                "score": 0.88,
+                "latex": "\\widehat { M } _ { 1 } \\ ="
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1293,
+                    638,
+                    1402,
+                    638,
+                    1402,
+                    669,
+                    1293,
+                    669
+                ],
+                "score": 0.87,
+                "latex": "\\gamma = 0 . 9 8"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1262,
+                    1069,
+                    1308,
+                    1069,
+                    1308,
+                    1094,
+                    1262,
+                    1094
+                ],
+                "score": 0.86,
+                "latex": "\\alpha _ { M }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    671,
+                    829,
+                    698,
+                    829,
+                    698,
+                    863,
+                    671,
+                    863
+                ],
+                "score": 0.85,
+                "latex": "\\widehat { \\mathcal { D } }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    627,
+                    1160,
+                    658,
+                    1160,
+                    658,
+                    1185,
+                    627,
+                    1185
+                ],
+                "score": 0.84,
+                "latex": "\\pi _ { 1 }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1358,
+                    1069,
+                    1393,
+                    1069,
+                    1393,
+                    1094,
+                    1358,
+                    1094
+                ],
+                "score": 0.84,
+                "latex": "\\alpha _ { \\pi }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    943,
+                    957,
+                    965,
+                    957,
+                    965,
+                    977,
+                    943,
+                    977
+                ],
+                "score": 0.76,
+                "latex": "\\pi"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1267,
+                    1027,
+                    1299,
+                    1027,
+                    1299,
+                    1058,
+                    1267,
+                    1058
+                ],
+                "score": 0.76,
+                "latex": "\\pi _ { 0 }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    952,
+                    841,
+                    974,
+                    841,
+                    974,
+                    863,
+                    952,
+                    863
+                ],
+                "score": 0.74,
+                "latex": "\\pi"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    695,
+                    1616,
+                    716,
+                    1616,
+                    716,
+                    1638,
+                    695,
+                    1638
+                ],
+                "score": 0.7,
+                "latex": "\\pi"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1310,
+                    1616,
+                    1327,
+                    1616,
+                    1327,
+                    1638,
+                    1310,
+                    1638
+                ],
+                "score": 0.61,
+                "latex": "\\epsilon"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    867,
+                    1617,
+                    881,
+                    1617,
+                    881,
+                    1638,
+                    867,
+                    1638
+                ],
+                "score": 0.53,
+                "latex": "\\epsilon"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1181,
+                    1938,
+                    1195,
+                    1938,
+                    1195,
+                    1959,
+                    1181,
+                    1959
+                ],
+                "score": 0.47,
+                "latex": "\\epsilon"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    297,
+                    613,
+                    315,
+                    613,
+                    315,
+                    635,
+                    297,
+                    635
+                ],
+                "score": 0.45,
+                "latex": "\\mathbf { X }"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    218.0,
+                    326.0,
+                    218.0,
+                    326.0,
+                    497.0,
+                    291.0,
+                    497.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1163.0,
+                    225.0,
+                    1335.0,
+                    225.0,
+                    1335.0,
+                    260.0,
+                    1163.0,
+                    260.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    451.0,
+                    229.0,
+                    594.0,
+                    229.0,
+                    594.0,
+                    257.0,
+                    451.0,
+                    257.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    816.0,
+                    229.0,
+                    954.0,
+                    229.0,
+                    954.0,
+                    257.0,
+                    816.0,
+                    257.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    321.0,
+                    330.0,
+                    366.0,
+                    330.0,
+                    366.0,
+                    362.0,
+                    321.0,
+                    362.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    375.0,
+                    266.0,
+                    403.0,
+                    266.0,
+                    403.0,
+                    379.0,
+                    375.0,
+                    379.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    408.0,
+                    313.0,
+                    437.0,
+                    313.0,
+                    437.0,
+                    377.0,
+                    408.0,
+                    377.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    442.0,
+                    320.0,
+                    467.0,
+                    320.0,
+                    467.0,
+                    375.0,
+                    442.0,
+                    375.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    475.0,
+                    331.0,
+                    503.0,
+                    331.0,
+                    503.0,
+                    374.0,
+                    475.0,
+                    374.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    508.0,
+                    335.0,
+                    534.0,
+                    335.0,
+                    534.0,
+                    380.0,
+                    508.0,
+                    380.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    541.0,
+                    332.0,
+                    568.0,
+                    332.0,
+                    568.0,
+                    371.0,
+                    541.0,
+                    371.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    574.0,
+                    315.0,
+                    603.0,
+                    315.0,
+                    603.0,
+                    376.0,
+                    574.0,
+                    376.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    604.0,
+                    304.0,
+                    637.0,
+                    304.0,
+                    637.0,
+                    347.0,
+                    604.0,
+                    347.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    639.0,
+                    302.0,
+                    667.0,
+                    302.0,
+                    667.0,
+                    343.0,
+                    639.0,
+                    343.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    682.0,
+                    278.0,
+                    734.0,
+                    278.0,
+                    734.0,
+                    382.0,
+                    682.0,
+                    382.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    740.0,
+                    267.0,
+                    768.0,
+                    267.0,
+                    768.0,
+                    329.0,
+                    740.0,
+                    329.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    774.0,
+                    292.0,
+                    800.0,
+                    292.0,
+                    800.0,
+                    343.0,
+                    774.0,
+                    343.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    807.0,
+                    315.0,
+                    831.0,
+                    315.0,
+                    831.0,
+                    348.0,
+                    807.0,
+                    348.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    841.0,
+                    287.0,
+                    864.0,
+                    287.0,
+                    864.0,
+                    325.0,
+                    841.0,
+                    325.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    875.0,
+                    288.0,
+                    895.0,
+                    288.0,
+                    895.0,
+                    335.0,
+                    875.0,
+                    335.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    906.0,
+                    278.0,
+                    969.0,
+                    278.0,
+                    969.0,
+                    355.0,
+                    906.0,
+                    355.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    970.0,
+                    288.0,
+                    998.0,
+                    288.0,
+                    998.0,
+                    398.0,
+                    970.0,
+                    398.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1003.0,
+                    305.0,
+                    1031.0,
+                    305.0,
+                    1031.0,
+                    340.0,
+                    1003.0,
+                    340.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1048.0,
+                    294.0,
+                    1095.0,
+                    294.0,
+                    1095.0,
+                    326.0,
+                    1048.0,
+                    326.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1104.0,
+                    329.0,
+                    1129.0,
+                    329.0,
+                    1129.0,
+                    400.0,
+                    1104.0,
+                    400.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1138.0,
+                    293.0,
+                    1164.0,
+                    293.0,
+                    1164.0,
+                    337.0,
+                    1138.0,
+                    337.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1172.0,
+                    291.0,
+                    1195.0,
+                    291.0,
+                    1195.0,
+                    329.0,
+                    1172.0,
+                    329.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1204.0,
+                    305.0,
+                    1230.0,
+                    305.0,
+                    1230.0,
+                    354.0,
+                    1204.0,
+                    354.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1239.0,
+                    300.0,
+                    1259.0,
+                    300.0,
+                    1259.0,
+                    348.0,
+                    1239.0,
+                    348.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1272.0,
+                    322.0,
+                    1292.0,
+                    322.0,
+                    1292.0,
+                    342.0,
+                    1272.0,
+                    342.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1306.0,
+                    291.0,
+                    1323.0,
+                    291.0,
+                    1323.0,
+                    375.0,
+                    1306.0,
+                    375.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1339.0,
+                    319.0,
+                    1358.0,
+                    319.0,
+                    1358.0,
+                    344.0,
+                    1339.0,
+                    344.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1369.0,
+                    309.0,
+                    1393.0,
+                    309.0,
+                    1393.0,
+                    374.0,
+                    1369.0,
+                    374.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    940.0,
+                    376.0,
+                    962.0,
+                    376.0,
+                    962.0,
+                    414.0,
+                    940.0,
+                    414.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1275.0,
+                    386.0,
+                    1292.0,
+                    386.0,
+                    1292.0,
+                    402.0,
+                    1275.0,
+                    402.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1339.0,
+                    382.0,
+                    1356.0,
+                    382.0,
+                    1356.0,
+                    417.0,
+                    1339.0,
+                    417.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    321.0,
+                    423.0,
+                    368.0,
+                    423.0,
+                    368.0,
+                    456.0,
+                    321.0,
+                    456.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    608.0,
+                    395.0,
+                    633.0,
+                    395.0,
+                    633.0,
+                    422.0,
+                    608.0,
+                    422.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    638.0,
+                    404.0,
+                    668.0,
+                    404.0,
+                    668.0,
+                    455.0,
+                    638.0,
+                    455.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1003.0,
+                    408.0,
+                    1031.0,
+                    408.0,
+                    1031.0,
+                    453.0,
+                    1003.0,
+                    453.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1047.0,
+                    415.0,
+                    1099.0,
+                    415.0,
+                    1099.0,
+                    451.0,
+                    1047.0,
+                    451.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    368.0,
+                    460.0,
+                    414.0,
+                    460.0,
+                    414.0,
+                    490.0,
+                    368.0,
+                    490.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    431.0,
+                    460.0,
+                    479.0,
+                    460.0,
+                    479.0,
+                    490.0,
+                    431.0,
+                    490.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    498.0,
+                    458.0,
+                    544.0,
+                    458.0,
+                    544.0,
+                    490.0,
+                    498.0,
+                    490.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    563.0,
+                    460.0,
+                    610.0,
+                    460.0,
+                    610.0,
+                    490.0,
+                    563.0,
+                    490.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    629.0,
+                    460.0,
+                    676.0,
+                    460.0,
+                    676.0,
+                    490.0,
+                    629.0,
+                    490.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    733.0,
+                    460.0,
+                    779.0,
+                    460.0,
+                    779.0,
+                    490.0,
+                    733.0,
+                    490.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    794.0,
+                    460.0,
+                    844.0,
+                    460.0,
+                    844.0,
+                    490.0,
+                    794.0,
+                    490.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    861.0,
+                    460.0,
+                    908.0,
+                    460.0,
+                    908.0,
+                    490.0,
+                    861.0,
+                    490.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    927.0,
+                    460.0,
+                    975.0,
+                    460.0,
+                    975.0,
+                    490.0,
+                    927.0,
+                    490.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    993.0,
+                    460.0,
+                    1039.0,
+                    460.0,
+                    1039.0,
+                    490.0,
+                    993.0,
+                    490.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1096.0,
+                    460.0,
+                    1143.0,
+                    460.0,
+                    1143.0,
+                    490.0,
+                    1096.0,
+                    490.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1159.0,
+                    460.0,
+                    1207.0,
+                    460.0,
+                    1207.0,
+                    490.0,
+                    1159.0,
+                    490.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1225.0,
+                    460.0,
+                    1273.0,
+                    460.0,
+                    1273.0,
+                    490.0,
+                    1225.0,
+                    490.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1292.0,
+                    461.0,
+                    1337.0,
+                    461.0,
+                    1337.0,
+                    488.0,
+                    1292.0,
+                    488.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1356.0,
+                    460.0,
+                    1403.0,
+                    460.0,
+                    1403.0,
+                    490.0,
+                    1356.0,
+                    490.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    381.0,
+                    486.0,
+                    659.0,
+                    486.0,
+                    659.0,
+                    520.0,
+                    381.0,
+                    520.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    745.0,
+                    485.0,
+                    1023.0,
+                    485.0,
+                    1023.0,
+                    520.0,
+                    745.0,
+                    520.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1109.0,
+                    485.0,
+                    1387.0,
+                    485.0,
+                    1387.0,
+                    520.0,
+                    1109.0,
+                    520.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    566.0,
+                    524.0,
+                    714.0,
+                    524.0,
+                    714.0,
+                    553.0,
+                    566.0,
+                    553.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    755.0,
+                    524.0,
+                    779.0,
+                    524.0,
+                    779.0,
+                    554.0,
+                    755.0,
+                    554.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    796.0,
+                    522.0,
+                    880.0,
+                    522.0,
+                    880.0,
+                    555.0,
+                    796.0,
+                    555.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    919.0,
+                    524.0,
+                    942.0,
+                    524.0,
+                    942.0,
+                    554.0,
+                    919.0,
+                    554.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    960.0,
+                    523.0,
+                    1135.0,
+                    523.0,
+                    1135.0,
+                    555.0,
+                    960.0,
+                    555.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    577.0,
+                    1403.0,
+                    577.0,
+                    1403.0,
+                    610.0,
+                    295.0,
+                    610.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    606.0,
+                    296.0,
+                    606.0,
+                    296.0,
+                    643.0,
+                    293.0,
+                    643.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    316.0,
+                    606.0,
+                    1406.0,
+                    606.0,
+                    1406.0,
+                    643.0,
+                    316.0,
+                    643.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    637.0,
+                    1292.0,
+                    637.0,
+                    1292.0,
+                    671.0,
+                    295.0,
+                    671.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    664.0,
+                    376.0,
+                    664.0,
+                    376.0,
+                    708.0,
+                    293.0,
+                    708.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    484.0,
+                    664.0,
+                    837.0,
+                    664.0,
+                    837.0,
+                    708.0,
+                    484.0,
+                    708.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    945.0,
+                    664.0,
+                    1406.0,
+                    664.0,
+                    1406.0,
+                    708.0,
+                    945.0,
+                    708.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    698.0,
+                    592.0,
+                    698.0,
+                    592.0,
+                    731.0,
+                    295.0,
+                    731.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    649.0,
+                    698.0,
+                    886.0,
+                    698.0,
+                    886.0,
+                    731.0,
+                    649.0,
+                    731.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    328.0,
+                    2000.0,
+                    889.0,
+                    2000.0,
+                    889.0,
+                    2040.0,
+                    328.0,
+                    2040.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1238.0,
+                    561.0,
+                    1238.0,
+                    561.0,
+                    1282.0,
+                    292.0,
+                    1282.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    73.0,
+                    815.0,
+                    73.0,
+                    815.0,
+                    108.0,
+                    295.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1487.0,
+                    858.0,
+                    1487.0,
+                    858.0,
+                    1523.0,
+                    295.0,
+                    1523.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    840.0,
+                    2086.0,
+                    860.0,
+                    2086.0,
+                    860.0,
+                    2119.0,
+                    840.0,
+                    2119.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    840.0,
+                    2086.0,
+                    860.0,
+                    2086.0,
+                    860.0,
+                    2119.0,
+                    840.0,
+                    2119.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    946.0,
+                    942.0,
+                    946.0,
+                    942.0,
+                    985.0,
+                    294.0,
+                    985.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    966.0,
+                    946.0,
+                    1405.0,
+                    946.0,
+                    1405.0,
+                    985.0,
+                    966.0,
+                    985.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    980.0,
+                    750.0,
+                    980.0,
+                    750.0,
+                    1025.0,
+                    294.0,
+                    1025.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    942.0,
+                    980.0,
+                    999.0,
+                    980.0,
+                    999.0,
+                    1025.0,
+                    942.0,
+                    1025.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1352.0,
+                    980.0,
+                    1406.0,
+                    980.0,
+                    1406.0,
+                    1025.0,
+                    1352.0,
+                    1025.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1021.0,
+                    296.0,
+                    1021.0,
+                    296.0,
+                    1062.0,
+                    291.0,
+                    1062.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    642.0,
+                    1021.0,
+                    1266.0,
+                    1021.0,
+                    1266.0,
+                    1062.0,
+                    642.0,
+                    1062.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1300.0,
+                    1021.0,
+                    1319.0,
+                    1021.0,
+                    1319.0,
+                    1062.0,
+                    1300.0,
+                    1062.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    614.0,
+                    1060.0,
+                    658.0,
+                    1060.0,
+                    658.0,
+                    1101.0,
+                    614.0,
+                    1101.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1006.0,
+                    1060.0,
+                    1261.0,
+                    1060.0,
+                    1261.0,
+                    1101.0,
+                    1006.0,
+                    1101.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1309.0,
+                    1060.0,
+                    1357.0,
+                    1060.0,
+                    1357.0,
+                    1101.0,
+                    1309.0,
+                    1101.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1394.0,
+                    1060.0,
+                    1405.0,
+                    1060.0,
+                    1405.0,
+                    1101.0,
+                    1394.0,
+                    1101.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1089.0,
+                    1405.0,
+                    1089.0,
+                    1405.0,
+                    1130.0,
+                    294.0,
+                    1130.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1122.0,
+                    1405.0,
+                    1122.0,
+                    1405.0,
+                    1159.0,
+                    294.0,
+                    1159.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1153.0,
+                    626.0,
+                    1153.0,
+                    626.0,
+                    1192.0,
+                    292.0,
+                    1192.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    659.0,
+                    1153.0,
+                    671.0,
+                    1153.0,
+                    671.0,
+                    1192.0,
+                    659.0,
+                    1192.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1547.0,
+                    1405.0,
+                    1547.0,
+                    1405.0,
+                    1584.0,
+                    294.0,
+                    1584.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1581.0,
+                    1403.0,
+                    1581.0,
+                    1403.0,
+                    1615.0,
+                    295.0,
+                    1615.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1609.0,
+                    694.0,
+                    1609.0,
+                    694.0,
+                    1646.0,
+                    293.0,
+                    1646.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    717.0,
+                    1609.0,
+                    866.0,
+                    1609.0,
+                    866.0,
+                    1646.0,
+                    717.0,
+                    1646.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    882.0,
+                    1609.0,
+                    1309.0,
+                    1609.0,
+                    1309.0,
+                    1646.0,
+                    882.0,
+                    1646.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1328.0,
+                    1609.0,
+                    1405.0,
+                    1609.0,
+                    1405.0,
+                    1646.0,
+                    1328.0,
+                    1646.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1641.0,
+                    1406.0,
+                    1641.0,
+                    1406.0,
+                    1676.0,
+                    293.0,
+                    1676.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1668.0,
+                    1404.0,
+                    1668.0,
+                    1404.0,
+                    1712.0,
+                    292.0,
+                    1712.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1702.0,
+                    1405.0,
+                    1702.0,
+                    1405.0,
+                    1736.0,
+                    293.0,
+                    1736.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1731.0,
+                    1320.0,
+                    1731.0,
+                    1320.0,
+                    1768.0,
+                    294.0,
+                    1768.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1778.0,
+                    1405.0,
+                    1778.0,
+                    1405.0,
+                    1814.0,
+                    294.0,
+                    1814.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1809.0,
+                    1406.0,
+                    1809.0,
+                    1406.0,
+                    1844.0,
+                    294.0,
+                    1844.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1842.0,
+                    1402.0,
+                    1842.0,
+                    1402.0,
+                    1874.0,
+                    296.0,
+                    1874.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1871.0,
+                    1404.0,
+                    1871.0,
+                    1404.0,
+                    1907.0,
+                    294.0,
+                    1907.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1901.0,
+                    1406.0,
+                    1901.0,
+                    1406.0,
+                    1937.0,
+                    293.0,
+                    1937.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1932.0,
+                    1180.0,
+                    1932.0,
+                    1180.0,
+                    1967.0,
+                    294.0,
+                    1967.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1196.0,
+                    1932.0,
+                    1207.0,
+                    1932.0,
+                    1207.0,
+                    1967.0,
+                    1196.0,
+                    1967.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1314.0,
+                    1405.0,
+                    1314.0,
+                    1405.0,
+                    1352.0,
+                    293.0,
+                    1352.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1345.0,
+                    1405.0,
+                    1345.0,
+                    1405.0,
+                    1383.0,
+                    292.0,
+                    1383.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1377.0,
+                    1405.0,
+                    1377.0,
+                    1405.0,
+                    1413.0,
+                    293.0,
+                    1413.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1411.0,
+                    764.0,
+                    1411.0,
+                    764.0,
+                    1441.0,
+                    293.0,
+                    1441.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    799.0,
+                    1404.0,
+                    799.0,
+                    1404.0,
+                    833.0,
+                    293.0,
+                    833.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    835.0,
+                    670.0,
+                    835.0,
+                    670.0,
+                    869.0,
+                    293.0,
+                    869.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    699.0,
+                    835.0,
+                    951.0,
+                    835.0,
+                    951.0,
+                    869.0,
+                    699.0,
+                    869.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    975.0,
+                    835.0,
+                    1403.0,
+                    835.0,
+                    1403.0,
+                    869.0,
+                    975.0,
+                    869.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    870.0,
+                    1024.0,
+                    870.0,
+                    1024.0,
+                    904.0,
+                    293.0,
+                    904.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1109.0,
+                    870.0,
+                    1405.0,
+                    870.0,
+                    1405.0,
+                    904.0,
+                    1109.0,
+                    904.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    902.0,
+                    444.0,
+                    902.0,
+                    444.0,
+                    937.0,
+                    293.0,
+                    937.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 6,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1316,
+                    1404,
+                    1316,
+                    1404,
+                    1712,
+                    297,
+                    1712
+                ],
+                "score": 0.982
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1728,
+                    1404,
+                    1728,
+                    1404,
+                    2034,
+                    298,
+                    2034
+                ],
+                "score": 0.982
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    994,
+                    1404,
+                    994,
+                    1404,
+                    1301,
+                    297,
+                    1301
+                ],
+                "score": 0.981
+            },
+            {
+                "category_id": 5,
+                "poly": [
+                    302,
+                    382,
+                    1398,
+                    382,
+                    1398,
+                    843,
+                    302,
+                    843
+                ],
+                "score": 0.981,
+                "html": "<table><tr><td>Dataset type</td><td>Environment</td><td>RepB-SDE (ours)</td><td>RP</td><td>Base</td><td>MOPO</td><td>MF</td><td>BC</td></tr><tr><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td></td></tr><tr><td>Random</td><td>Walker2d</td><td>21.1 ± 1.0</td><td>18.4</td><td>16.4</td><td>1.3</td><td>7.3 BEAR</td><td>0.0</td></tr><tr><td>Medium</td><td>Walker2d</td><td>72.1 ± 1.9</td><td>56.3</td><td>5.5</td><td>-0.1</td><td>81.1 BRAC</td><td>7.7</td></tr><tr><td>Med-Replay</td><td>Walker2d</td><td>49.8 ± 11.4</td><td>41.1</td><td>6.2</td><td>47.8</td><td>26.7 CQL</td><td>8.0</td></tr><tr><td>Med-Expert</td><td>Walker2d</td><td>88.8 ± 6.9</td><td>72.6</td><td>51.7</td><td>32.4</td><td>98.7 CQL</td><td>3.3</td></tr><tr><td>Random</td><td>Hopper</td><td>8.6 ±1.0</td><td>8.3</td><td>8.3</td><td>9.1</td><td>12.2 BRAC</td><td>9.0</td></tr><tr><td>Medium</td><td>Hopper</td><td>34.0 ± 2.8</td><td>27.5</td><td>19.8</td><td>19.2</td><td>58.0 CQL</td><td>34.5</td></tr><tr><td>Med-Replay</td><td>Hopper</td><td>62.2 ± 6.7</td><td>49.8</td><td>32.9</td><td>80.8</td><td>48.6 CQL</td><td>13.2</td></tr><tr><td>Med-Expert</td><td>Hopper</td><td>82.6 ± 7.0</td><td>74.0</td><td>19.1</td><td>23.2</td><td>111.0 CQL</td><td>41.9</td></tr><tr><td>Random</td><td>HalfCheetah</td><td>32.9 ± 1.1</td><td>31.3</td><td>26.1</td><td>29.9</td><td>35.4 CQL</td><td>-2.4</td></tr><tr><td>Medium</td><td>HalfCheetah</td><td>49.1 ± 0.3</td><td>47.3</td><td>22.1</td><td>45.0</td><td>46.3 BRAC</td><td>31.0</td></tr><tr><td>Med-Replay</td><td>HalfCheetah</td><td>57.5 ± 0.8</td><td>53.6</td><td>54.0</td><td>53.8</td><td>47.7 BRAC</td><td>14.7</td></tr><tr><td>Med-Expert</td><td>HalfCheetah</td><td>55.4 ± 8.3</td><td>36.7</td><td>31.9</td><td>91.0</td><td>64.7 BCQ</td><td>29.5</td></tr></table>"
+            },
+            {
+                "category_id": 6,
+                "poly": [
+                    296,
+                    223,
+                    1404,
+                    223,
+                    1404,
+                    348,
+                    296,
+                    348
+                ],
+                "score": 0.945
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    300,
+                    76,
+                    813,
+                    76,
+                    813,
+                    104,
+                    300,
+                    104
+                ],
+                "score": 0.894
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    840,
+                    2088,
+                    858,
+                    2088,
+                    858,
+                    2112,
+                    840,
+                    2112
+                ],
+                "score": 0.791
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    290,
+                    932,
+                    1323,
+                    932,
+                    1323,
+                    961,
+                    290,
+                    961
+                ],
+                "score": 0.554
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    290,
+                    932,
+                    1323,
+                    932,
+                    1323,
+                    961,
+                    290,
+                    961
+                ],
+                "score": 0.217
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1097,
+                    1086,
+                    1140,
+                    1086,
+                    1140,
+                    1115,
+                    1097,
+                    1115
+                ],
+                "score": 0.92,
+                "latex": "1 0 ^ { 6 }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    689,
+                    1177,
+                    779,
+                    1177,
+                    779,
+                    1208,
+                    689,
+                    1208
+                ],
+                "score": 0.89,
+                "latex": "2 \\times \\mathrm { i 0 ^ { 5 } }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    703,
+                    1238,
+                    747,
+                    1238,
+                    747,
+                    1269,
+                    703,
+                    1269
+                ],
+                "score": 0.89,
+                "latex": "1 0 ^ { 6 }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    789,
+                    1116,
+                    832,
+                    1116,
+                    832,
+                    1146,
+                    789,
+                    1146
+                ],
+                "score": 0.87,
+                "latex": "1 0 ^ { 6 }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    811,
+                    1564,
+                    843,
+                    1564,
+                    843,
+                    1591,
+                    811,
+                    1591
+                ],
+                "score": 0.86,
+                "latex": "\\pi _ { 0 }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    633,
+                    1177,
+                    676,
+                    1177,
+                    676,
+                    1208,
+                    633,
+                    1208
+                ],
+                "score": 0.86,
+                "latex": "1 0 ^ { 5 }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1107,
+                    1596,
+                    1138,
+                    1596,
+                    1138,
+                    1621,
+                    1107,
+                    1621
+                ],
+                "score": 0.84,
+                "latex": "\\pi _ { 1 }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    371,
+                    318,
+                    399,
+                    318,
+                    399,
+                    345,
+                    371,
+                    345
+                ],
+                "score": 0.67,
+                "latex": "\\pm"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    391,
+                    1148,
+                    437,
+                    1148,
+                    437,
+                    1180,
+                    391,
+                    1180
+                ],
+                "score": 0.35,
+                "latex": "1 / 3"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    221.0,
+                    1405.0,
+                    221.0,
+                    1405.0,
+                    260.0,
+                    294.0,
+                    260.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    254.0,
+                    1405.0,
+                    254.0,
+                    1405.0,
+                    290.0,
+                    294.0,
+                    290.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    283.0,
+                    1407.0,
+                    283.0,
+                    1407.0,
+                    323.0,
+                    294.0,
+                    323.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    312.0,
+                    370.0,
+                    312.0,
+                    370.0,
+                    351.0,
+                    294.0,
+                    351.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    400.0,
+                    312.0,
+                    1253.0,
+                    312.0,
+                    1253.0,
+                    351.0,
+                    400.0,
+                    351.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    72.0,
+                    816.0,
+                    72.0,
+                    816.0,
+                    110.0,
+                    294.0,
+                    110.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    836.0,
+                    2085.0,
+                    859.0,
+                    2085.0,
+                    859.0,
+                    2116.0,
+                    836.0,
+                    2116.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    928.0,
+                    1329.0,
+                    928.0,
+                    1329.0,
+                    965.0,
+                    294.0,
+                    965.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1315.0,
+                    1405.0,
+                    1315.0,
+                    1405.0,
+                    1351.0,
+                    295.0,
+                    1351.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1342.0,
+                    1406.0,
+                    1342.0,
+                    1406.0,
+                    1386.0,
+                    292.0,
+                    1386.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1378.0,
+                    1405.0,
+                    1378.0,
+                    1405.0,
+                    1411.0,
+                    295.0,
+                    1411.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1407.0,
+                    1406.0,
+                    1407.0,
+                    1406.0,
+                    1443.0,
+                    294.0,
+                    1443.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1436.0,
+                    1406.0,
+                    1436.0,
+                    1406.0,
+                    1474.0,
+                    292.0,
+                    1474.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1468.0,
+                    1406.0,
+                    1468.0,
+                    1406.0,
+                    1504.0,
+                    292.0,
+                    1504.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1497.0,
+                    1404.0,
+                    1497.0,
+                    1404.0,
+                    1533.0,
+                    294.0,
+                    1533.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1530.0,
+                    1404.0,
+                    1530.0,
+                    1404.0,
+                    1566.0,
+                    295.0,
+                    1566.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1560.0,
+                    810.0,
+                    1560.0,
+                    810.0,
+                    1597.0,
+                    295.0,
+                    1597.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    844.0,
+                    1560.0,
+                    1405.0,
+                    1560.0,
+                    1405.0,
+                    1597.0,
+                    844.0,
+                    1597.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1590.0,
+                    1106.0,
+                    1590.0,
+                    1106.0,
+                    1626.0,
+                    295.0,
+                    1626.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1139.0,
+                    1590.0,
+                    1404.0,
+                    1590.0,
+                    1404.0,
+                    1626.0,
+                    1139.0,
+                    1626.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1620.0,
+                    1404.0,
+                    1620.0,
+                    1404.0,
+                    1656.0,
+                    295.0,
+                    1656.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1650.0,
+                    1406.0,
+                    1650.0,
+                    1406.0,
+                    1690.0,
+                    294.0,
+                    1690.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1677.0,
+                    452.0,
+                    1677.0,
+                    452.0,
+                    1721.0,
+                    294.0,
+                    1721.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1726.0,
+                    1405.0,
+                    1726.0,
+                    1405.0,
+                    1764.0,
+                    295.0,
+                    1764.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1759.0,
+                    1405.0,
+                    1759.0,
+                    1405.0,
+                    1795.0,
+                    294.0,
+                    1795.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1791.0,
+                    1404.0,
+                    1791.0,
+                    1404.0,
+                    1823.0,
+                    294.0,
+                    1823.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1823.0,
+                    1405.0,
+                    1823.0,
+                    1405.0,
+                    1854.0,
+                    296.0,
+                    1854.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1851.0,
+                    1405.0,
+                    1851.0,
+                    1405.0,
+                    1886.0,
+                    294.0,
+                    1886.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1882.0,
+                    1402.0,
+                    1882.0,
+                    1402.0,
+                    1914.0,
+                    296.0,
+                    1914.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1911.0,
+                    1405.0,
+                    1911.0,
+                    1405.0,
+                    1947.0,
+                    294.0,
+                    1947.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1942.0,
+                    1404.0,
+                    1942.0,
+                    1404.0,
+                    1976.0,
+                    295.0,
+                    1976.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1970.0,
+                    1404.0,
+                    1970.0,
+                    1404.0,
+                    2009.0,
+                    292.0,
+                    2009.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    2001.0,
+                    417.0,
+                    2001.0,
+                    417.0,
+                    2039.0,
+                    293.0,
+                    2039.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    996.0,
+                    1404.0,
+                    996.0,
+                    1404.0,
+                    1028.0,
+                    296.0,
+                    1028.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1024.0,
+                    1404.0,
+                    1024.0,
+                    1404.0,
+                    1061.0,
+                    291.0,
+                    1061.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1056.0,
+                    1404.0,
+                    1056.0,
+                    1404.0,
+                    1092.0,
+                    294.0,
+                    1092.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1085.0,
+                    1096.0,
+                    1085.0,
+                    1096.0,
+                    1121.0,
+                    295.0,
+                    1121.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1141.0,
+                    1085.0,
+                    1404.0,
+                    1085.0,
+                    1404.0,
+                    1121.0,
+                    1141.0,
+                    1121.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1117.0,
+                    788.0,
+                    1117.0,
+                    788.0,
+                    1153.0,
+                    292.0,
+                    1153.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    833.0,
+                    1117.0,
+                    1404.0,
+                    1117.0,
+                    1404.0,
+                    1153.0,
+                    833.0,
+                    1153.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1146.0,
+                    390.0,
+                    1146.0,
+                    390.0,
+                    1182.0,
+                    292.0,
+                    1182.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    438.0,
+                    1146.0,
+                    1404.0,
+                    1146.0,
+                    1404.0,
+                    1182.0,
+                    438.0,
+                    1182.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1174.0,
+                    632.0,
+                    1174.0,
+                    632.0,
+                    1215.0,
+                    292.0,
+                    1215.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    677.0,
+                    1174.0,
+                    688.0,
+                    1174.0,
+                    688.0,
+                    1215.0,
+                    677.0,
+                    1215.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    780.0,
+                    1174.0,
+                    1405.0,
+                    1174.0,
+                    1405.0,
+                    1215.0,
+                    780.0,
+                    1215.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1209.0,
+                    1406.0,
+                    1209.0,
+                    1406.0,
+                    1243.0,
+                    294.0,
+                    1243.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    290.0,
+                    1235.0,
+                    702.0,
+                    1235.0,
+                    702.0,
+                    1277.0,
+                    290.0,
+                    1277.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    748.0,
+                    1235.0,
+                    1405.0,
+                    1235.0,
+                    1405.0,
+                    1277.0,
+                    748.0,
+                    1277.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1267.0,
+                    1043.0,
+                    1267.0,
+                    1043.0,
+                    1303.0,
+                    294.0,
+                    1303.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    928.0,
+                    1329.0,
+                    928.0,
+                    1329.0,
+                    965.0,
+                    294.0,
+                    965.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 7,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    298,
+                    1403,
+                    298,
+                    1403,
+                    633,
+                    298,
+                    633
+                ],
+                "score": 0.983
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    817,
+                    1404,
+                    817,
+                    1404,
+                    1001,
+                    299,
+                    1001
+                ],
+                "score": 0.978
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    649,
+                    1402,
+                    649,
+                    1402,
+                    802,
+                    299,
+                    802
+                ],
+                "score": 0.976
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    1095,
+                    1403,
+                    1095,
+                    1403,
+                    1216,
+                    299,
+                    1216
+                ],
+                "score": 0.962
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    300,
+                    76,
+                    813,
+                    76,
+                    813,
+                    104,
+                    300,
+                    104
+                ],
+                "score": 0.876
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    301,
+                    1042,
+                    557,
+                    1042,
+                    557,
+                    1069,
+                    301,
+                    1069
+                ],
+                "score": 0.839
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1320,
+                    1401,
+                    1320,
+                    1401,
+                    1382,
+                    298,
+                    1382
+                ],
+                "score": 0.809
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    300,
+                    1267,
+                    487,
+                    1267,
+                    487,
+                    1300,
+                    300,
+                    1300
+                ],
+                "score": 0.794
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    298,
+                    226,
+                    842,
+                    226,
+                    842,
+                    261,
+                    298,
+                    261
+                ],
+                "score": 0.785
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    840,
+                    2088,
+                    858,
+                    2088,
+                    858,
+                    2111,
+                    840,
+                    2111
+                ],
+                "score": 0.767
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    291,
+                    1407,
+                    1403,
+                    1407,
+                    1403,
+                    1471,
+                    291,
+                    1471
+                ],
+                "score": 0.665
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1673,
+                    1401,
+                    1673,
+                    1401,
+                    1738,
+                    298,
+                    1738
+                ],
+                "score": 0.653
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    1763,
+                    1399,
+                    1763,
+                    1399,
+                    1856,
+                    299,
+                    1856
+                ],
+                "score": 0.641
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1971,
+                    1397,
+                    1971,
+                    1397,
+                    2034,
+                    298,
+                    2034
+                ],
+                "score": 0.634
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    291,
+                    1882,
+                    1401,
+                    1882,
+                    1401,
+                    1946,
+                    291,
+                    1946
+                ],
+                "score": 0.623
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    1584,
+                    1402,
+                    1584,
+                    1402,
+                    1650,
+                    299,
+                    1650
+                ],
+                "score": 0.563
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    296,
+                    1495,
+                    1400,
+                    1495,
+                    1400,
+                    1561,
+                    296,
+                    1561
+                ],
+                "score": 0.525
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    72.0,
+                    816.0,
+                    72.0,
+                    816.0,
+                    110.0,
+                    294.0,
+                    110.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    299.0,
+                    1043.0,
+                    558.0,
+                    1043.0,
+                    558.0,
+                    1071.0,
+                    299.0,
+                    1071.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    1269.0,
+                    489.0,
+                    1269.0,
+                    489.0,
+                    1302.0,
+                    297.0,
+                    1302.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    221.0,
+                    847.0,
+                    221.0,
+                    847.0,
+                    268.0,
+                    292.0,
+                    268.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    839.0,
+                    2087.0,
+                    860.0,
+                    2087.0,
+                    860.0,
+                    2117.0,
+                    839.0,
+                    2117.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    295.0,
+                    1404.0,
+                    295.0,
+                    1404.0,
+                    335.0,
+                    292.0,
+                    335.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    328.0,
+                    1405.0,
+                    328.0,
+                    1405.0,
+                    363.0,
+                    293.0,
+                    363.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    355.0,
+                    1404.0,
+                    355.0,
+                    1404.0,
+                    398.0,
+                    292.0,
+                    398.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    391.0,
+                    1404.0,
+                    391.0,
+                    1404.0,
+                    422.0,
+                    296.0,
+                    422.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    422.0,
+                    1404.0,
+                    422.0,
+                    1404.0,
+                    453.0,
+                    296.0,
+                    453.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    451.0,
+                    1407.0,
+                    451.0,
+                    1407.0,
+                    485.0,
+                    294.0,
+                    485.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    479.0,
+                    1407.0,
+                    479.0,
+                    1407.0,
+                    517.0,
+                    293.0,
+                    517.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    511.0,
+                    1405.0,
+                    511.0,
+                    1405.0,
+                    546.0,
+                    294.0,
+                    546.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    540.0,
+                    1405.0,
+                    540.0,
+                    1405.0,
+                    577.0,
+                    293.0,
+                    577.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    572.0,
+                    1404.0,
+                    572.0,
+                    1404.0,
+                    606.0,
+                    294.0,
+                    606.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    604.0,
+                    858.0,
+                    604.0,
+                    858.0,
+                    635.0,
+                    296.0,
+                    635.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    815.0,
+                    1405.0,
+                    815.0,
+                    1405.0,
+                    853.0,
+                    294.0,
+                    853.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    847.0,
+                    1405.0,
+                    847.0,
+                    1405.0,
+                    885.0,
+                    294.0,
+                    885.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    877.0,
+                    1408.0,
+                    877.0,
+                    1408.0,
+                    917.0,
+                    294.0,
+                    917.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    908.0,
+                    1405.0,
+                    908.0,
+                    1405.0,
+                    946.0,
+                    294.0,
+                    946.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    938.0,
+                    1404.0,
+                    938.0,
+                    1404.0,
+                    974.0,
+                    295.0,
+                    974.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    971.0,
+                    1085.0,
+                    971.0,
+                    1085.0,
+                    1004.0,
+                    293.0,
+                    1004.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    650.0,
+                    1404.0,
+                    650.0,
+                    1404.0,
+                    683.0,
+                    295.0,
+                    683.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    681.0,
+                    1404.0,
+                    681.0,
+                    1404.0,
+                    714.0,
+                    295.0,
+                    714.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    710.0,
+                    1407.0,
+                    710.0,
+                    1407.0,
+                    747.0,
+                    294.0,
+                    747.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    741.0,
+                    1406.0,
+                    741.0,
+                    1406.0,
+                    778.0,
+                    295.0,
+                    778.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    773.0,
+                    828.0,
+                    773.0,
+                    828.0,
+                    806.0,
+                    297.0,
+                    806.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1092.0,
+                    1405.0,
+                    1092.0,
+                    1405.0,
+                    1131.0,
+                    293.0,
+                    1131.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1123.0,
+                    1404.0,
+                    1123.0,
+                    1404.0,
+                    1160.0,
+                    293.0,
+                    1160.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1157.0,
+                    1404.0,
+                    1157.0,
+                    1404.0,
+                    1190.0,
+                    293.0,
+                    1190.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1184.0,
+                    638.0,
+                    1184.0,
+                    638.0,
+                    1220.0,
+                    294.0,
+                    1220.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1318.0,
+                    1403.0,
+                    1318.0,
+                    1403.0,
+                    1355.0,
+                    293.0,
+                    1355.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    321.0,
+                    1349.0,
+                    1234.0,
+                    1349.0,
+                    1234.0,
+                    1385.0,
+                    321.0,
+                    1385.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1403.0,
+                    1405.0,
+                    1403.0,
+                    1405.0,
+                    1447.0,
+                    293.0,
+                    1447.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    1439.0,
+                    608.0,
+                    1439.0,
+                    608.0,
+                    1470.0,
+                    322.0,
+                    1470.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1675.0,
+                    1405.0,
+                    1675.0,
+                    1405.0,
+                    1711.0,
+                    296.0,
+                    1711.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    1706.0,
+                    1169.0,
+                    1706.0,
+                    1169.0,
+                    1739.0,
+                    323.0,
+                    1739.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1762.0,
+                    1403.0,
+                    1762.0,
+                    1403.0,
+                    1800.0,
+                    294.0,
+                    1800.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    1795.0,
+                    1404.0,
+                    1795.0,
+                    1404.0,
+                    1829.0,
+                    323.0,
+                    1829.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    324.0,
+                    1825.0,
+                    624.0,
+                    1825.0,
+                    624.0,
+                    1860.0,
+                    324.0,
+                    1860.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1972.0,
+                    1402.0,
+                    1972.0,
+                    1402.0,
+                    2008.0,
+                    295.0,
+                    2008.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    2001.0,
+                    1106.0,
+                    2001.0,
+                    1106.0,
+                    2037.0,
+                    323.0,
+                    2037.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1879.0,
+                    1404.0,
+                    1879.0,
+                    1404.0,
+                    1919.0,
+                    294.0,
+                    1919.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    1913.0,
+                    1176.0,
+                    1913.0,
+                    1176.0,
+                    1949.0,
+                    322.0,
+                    1949.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1583.0,
+                    1406.0,
+                    1583.0,
+                    1406.0,
+                    1622.0,
+                    294.0,
+                    1622.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    1616.0,
+                    1279.0,
+                    1616.0,
+                    1279.0,
+                    1651.0,
+                    322.0,
+                    1651.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1497.0,
+                    1404.0,
+                    1497.0,
+                    1404.0,
+                    1529.0,
+                    296.0,
+                    1529.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    321.0,
+                    1527.0,
+                    1239.0,
+                    1527.0,
+                    1239.0,
+                    1562.0,
+                    321.0,
+                    1562.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 8,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 2,
+                "poly": [
+                    300,
+                    75,
+                    813,
+                    75,
+                    813,
+                    105,
+                    300,
+                    105
+                ],
+                "score": 0.89
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    836,
+                    2088,
+                    866,
+                    2088,
+                    866,
+                    2113,
+                    836,
+                    2113
+                ],
+                "score": 0.83
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    292,
+                    228,
+                    1403,
+                    228,
+                    1403,
+                    294,
+                    292,
+                    294
+                ],
+                "score": 0.664
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    296,
+                    1110,
+                    1402,
+                    1110,
+                    1402,
+                    1174,
+                    296,
+                    1174
+                ],
+                "score": 0.655
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1192,
+                    1400,
+                    1192,
+                    1400,
+                    1287,
+                    298,
+                    1287
+                ],
+                "score": 0.648
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    997,
+                    1401,
+                    997,
+                    1401,
+                    1093,
+                    299,
+                    1093
+                ],
+                "score": 0.644
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    296,
+                    1468,
+                    1399,
+                    1468,
+                    1399,
+                    1535,
+                    296,
+                    1535
+                ],
+                "score": 0.634
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    293,
+                    1304,
+                    1402,
+                    1304,
+                    1402,
+                    1371,
+                    293,
+                    1371
+                ],
+                "score": 0.62
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    588,
+                    1401,
+                    588,
+                    1401,
+                    652,
+                    298,
+                    652
+                ],
+                "score": 0.619
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1386,
+                    1401,
+                    1386,
+                    1401,
+                    1453,
+                    298,
+                    1453
+                ],
+                "score": 0.615
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    294,
+                    915,
+                    1401,
+                    915,
+                    1401,
+                    981,
+                    294,
+                    981
+                ],
+                "score": 0.608
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    301,
+                    1550,
+                    1399,
+                    1550,
+                    1399,
+                    1647,
+                    301,
+                    1647
+                ],
+                "score": 0.598
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    294,
+                    833,
+                    1398,
+                    833,
+                    1398,
+                    899,
+                    294,
+                    899
+                ],
+                "score": 0.597
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    294,
+                    505,
+                    1400,
+                    505,
+                    1400,
+                    570,
+                    294,
+                    570
+                ],
+                "score": 0.579
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    669,
+                    1399,
+                    669,
+                    1399,
+                    735,
+                    298,
+                    735
+                ],
+                "score": 0.577
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    752,
+                    1400,
+                    752,
+                    1400,
+                    817,
+                    298,
+                    817
+                ],
+                "score": 0.574
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    310,
+                    1400,
+                    310,
+                    1400,
+                    377,
+                    297,
+                    377
+                ],
+                "score": 0.568
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    294,
+                    1663,
+                    1401,
+                    1663,
+                    1401,
+                    1729,
+                    294,
+                    1729
+                ],
+                "score": 0.564
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1857,
+                    1399,
+                    1857,
+                    1399,
+                    1923,
+                    297,
+                    1923
+                ],
+                "score": 0.548
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    300,
+                    1745,
+                    1399,
+                    1745,
+                    1399,
+                    1841,
+                    300,
+                    1841
+                ],
+                "score": 0.546
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    301,
+                    392,
+                    1399,
+                    392,
+                    1399,
+                    488,
+                    301,
+                    488
+                ],
+                "score": 0.53
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    300,
+                    1942,
+                    1402,
+                    1942,
+                    1402,
+                    2034,
+                    300,
+                    2034
+                ],
+                "score": 0.439
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    73.0,
+                    815.0,
+                    73.0,
+                    815.0,
+                    108.0,
+                    295.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    832.0,
+                    2085.0,
+                    870.0,
+                    2085.0,
+                    870.0,
+                    2123.0,
+                    832.0,
+                    2123.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    227.0,
+                    1403.0,
+                    227.0,
+                    1403.0,
+                    266.0,
+                    295.0,
+                    266.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    260.0,
+                    1082.0,
+                    260.0,
+                    1082.0,
+                    296.0,
+                    323.0,
+                    296.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1106.0,
+                    1406.0,
+                    1106.0,
+                    1406.0,
+                    1150.0,
+                    293.0,
+                    1150.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    1142.0,
+                    596.0,
+                    1142.0,
+                    596.0,
+                    1173.0,
+                    323.0,
+                    1173.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1191.0,
+                    1404.0,
+                    1191.0,
+                    1404.0,
+                    1231.0,
+                    293.0,
+                    1231.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    1225.0,
+                    1404.0,
+                    1225.0,
+                    1404.0,
+                    1259.0,
+                    323.0,
+                    1259.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    320.0,
+                    1254.0,
+                    459.0,
+                    1254.0,
+                    459.0,
+                    1287.0,
+                    320.0,
+                    1287.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    998.0,
+                    1405.0,
+                    998.0,
+                    1405.0,
+                    1033.0,
+                    294.0,
+                    1033.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    1028.0,
+                    1405.0,
+                    1028.0,
+                    1405.0,
+                    1066.0,
+                    322.0,
+                    1066.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    1058.0,
+                    395.0,
+                    1058.0,
+                    395.0,
+                    1093.0,
+                    322.0,
+                    1093.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1469.0,
+                    1403.0,
+                    1469.0,
+                    1403.0,
+                    1505.0,
+                    296.0,
+                    1505.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    1500.0,
+                    1317.0,
+                    1500.0,
+                    1317.0,
+                    1537.0,
+                    322.0,
+                    1537.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1305.0,
+                    1405.0,
+                    1305.0,
+                    1405.0,
+                    1341.0,
+                    296.0,
+                    1341.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    321.0,
+                    1338.0,
+                    1288.0,
+                    1338.0,
+                    1288.0,
+                    1371.0,
+                    321.0,
+                    1371.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    588.0,
+                    1403.0,
+                    588.0,
+                    1403.0,
+                    624.0,
+                    297.0,
+                    624.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    619.0,
+                    946.0,
+                    619.0,
+                    946.0,
+                    651.0,
+                    323.0,
+                    651.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1387.0,
+                    1405.0,
+                    1387.0,
+                    1405.0,
+                    1423.0,
+                    296.0,
+                    1423.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    1418.0,
+                    1403.0,
+                    1418.0,
+                    1403.0,
+                    1455.0,
+                    322.0,
+                    1455.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    916.0,
+                    1403.0,
+                    916.0,
+                    1403.0,
+                    952.0,
+                    296.0,
+                    952.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    324.0,
+                    945.0,
+                    771.0,
+                    945.0,
+                    771.0,
+                    982.0,
+                    324.0,
+                    982.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1548.0,
+                    1404.0,
+                    1548.0,
+                    1404.0,
+                    1590.0,
+                    295.0,
+                    1590.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    321.0,
+                    1582.0,
+                    1404.0,
+                    1582.0,
+                    1404.0,
+                    1618.0,
+                    321.0,
+                    1618.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    321.0,
+                    1615.0,
+                    960.0,
+                    1615.0,
+                    960.0,
+                    1646.0,
+                    321.0,
+                    1646.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    831.0,
+                    1402.0,
+                    831.0,
+                    1402.0,
+                    871.0,
+                    294.0,
+                    871.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    324.0,
+                    865.0,
+                    1216.0,
+                    865.0,
+                    1216.0,
+                    901.0,
+                    324.0,
+                    901.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    504.0,
+                    1402.0,
+                    504.0,
+                    1402.0,
+                    543.0,
+                    296.0,
+                    543.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    538.0,
+                    705.0,
+                    538.0,
+                    705.0,
+                    570.0,
+                    322.0,
+                    570.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    668.0,
+                    1404.0,
+                    668.0,
+                    1404.0,
+                    707.0,
+                    293.0,
+                    707.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    701.0,
+                    1320.0,
+                    701.0,
+                    1320.0,
+                    736.0,
+                    322.0,
+                    736.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    752.0,
+                    1404.0,
+                    752.0,
+                    1404.0,
+                    788.0,
+                    294.0,
+                    788.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    783.0,
+                    934.0,
+                    783.0,
+                    934.0,
+                    817.0,
+                    322.0,
+                    817.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    310.0,
+                    1402.0,
+                    310.0,
+                    1402.0,
+                    346.0,
+                    296.0,
+                    346.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    342.0,
+                    1346.0,
+                    342.0,
+                    1346.0,
+                    379.0,
+                    323.0,
+                    379.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    1665.0,
+                    1403.0,
+                    1665.0,
+                    1403.0,
+                    1701.0,
+                    297.0,
+                    1701.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    1695.0,
+                    1369.0,
+                    1695.0,
+                    1369.0,
+                    1731.0,
+                    322.0,
+                    1731.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1856.0,
+                    1403.0,
+                    1856.0,
+                    1403.0,
+                    1894.0,
+                    296.0,
+                    1894.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    1889.0,
+                    1097.0,
+                    1889.0,
+                    1097.0,
+                    1924.0,
+                    322.0,
+                    1924.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    1746.0,
+                    1403.0,
+                    1746.0,
+                    1403.0,
+                    1781.0,
+                    297.0,
+                    1781.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    325.0,
+                    1776.0,
+                    1404.0,
+                    1776.0,
+                    1404.0,
+                    1813.0,
+                    325.0,
+                    1813.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    321.0,
+                    1808.0,
+                    1002.0,
+                    1808.0,
+                    1002.0,
+                    1842.0,
+                    321.0,
+                    1842.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    390.0,
+                    1404.0,
+                    390.0,
+                    1404.0,
+                    431.0,
+                    295.0,
+                    431.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    425.0,
+                    1403.0,
+                    425.0,
+                    1403.0,
+                    460.0,
+                    322.0,
+                    460.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    456.0,
+                    705.0,
+                    456.0,
+                    705.0,
+                    489.0,
+                    322.0,
+                    489.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1939.0,
+                    1403.0,
+                    1939.0,
+                    1403.0,
+                    1977.0,
+                    294.0,
+                    1977.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    321.0,
+                    1971.0,
+                    1194.0,
+                    1971.0,
+                    1194.0,
+                    2006.0,
+                    321.0,
+                    2006.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1220.0,
+                    1975.0,
+                    1404.0,
+                    1975.0,
+                    1404.0,
+                    2006.0,
+                    1220.0,
+                    2006.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    321.0,
+                    2001.0,
+                    609.0,
+                    2001.0,
+                    609.0,
+                    2034.0,
+                    321.0,
+                    2034.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 9,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 2,
+                "poly": [
+                    300,
+                    75,
+                    813,
+                    75,
+                    813,
+                    105,
+                    300,
+                    105
+                ],
+                "score": 0.886
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    836,
+                    2088,
+                    863,
+                    2088,
+                    863,
+                    2113,
+                    836,
+                    2113
+                ],
+                "score": 0.818
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    295,
+                    228,
+                    1402,
+                    228,
+                    1402,
+                    293,
+                    295,
+                    293
+                ],
+                "score": 0.706
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    945,
+                    1402,
+                    945,
+                    1402,
+                    1011,
+                    297,
+                    1011
+                ],
+                "score": 0.702
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    292,
+                    1304,
+                    1402,
+                    1304,
+                    1402,
+                    1370,
+                    292,
+                    1370
+                ],
+                "score": 0.698
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    294,
+                    1028,
+                    1401,
+                    1028,
+                    1401,
+                    1093,
+                    294,
+                    1093
+                ],
+                "score": 0.683
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    864,
+                    1402,
+                    864,
+                    1402,
+                    929,
+                    297,
+                    929
+                ],
+                "score": 0.683
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    587,
+                    1401,
+                    587,
+                    1401,
+                    682,
+                    297,
+                    682
+                ],
+                "score": 0.661
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    294,
+                    699,
+                    1402,
+                    699,
+                    1402,
+                    765,
+                    294,
+                    765
+                ],
+                "score": 0.66
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1192,
+                    1402,
+                    1192,
+                    1402,
+                    1287,
+                    297,
+                    1287
+                ],
+                "score": 0.648
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    294,
+                    782,
+                    1399,
+                    782,
+                    1399,
+                    848,
+                    294,
+                    848
+                ],
+                "score": 0.641
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1857,
+                    1398,
+                    1857,
+                    1398,
+                    1923,
+                    298,
+                    1923
+                ],
+                "score": 0.641
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    293,
+                    1499,
+                    1402,
+                    1499,
+                    1402,
+                    1564,
+                    293,
+                    1564
+                ],
+                "score": 0.639
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    295,
+                    1693,
+                    1399,
+                    1693,
+                    1399,
+                    1760,
+                    295,
+                    1760
+                ],
+                "score": 0.637
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1775,
+                    1397,
+                    1775,
+                    1397,
+                    1842,
+                    297,
+                    1842
+                ],
+                "score": 0.629
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    1386,
+                    1400,
+                    1386,
+                    1400,
+                    1482,
+                    299,
+                    1482
+                ],
+                "score": 0.624
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1581,
+                    1403,
+                    1581,
+                    1403,
+                    1676,
+                    298,
+                    1676
+                ],
+                "score": 0.619
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    293,
+                    1110,
+                    1399,
+                    1110,
+                    1399,
+                    1176,
+                    293,
+                    1176
+                ],
+                "score": 0.618
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    296,
+                    505,
+                    1402,
+                    505,
+                    1402,
+                    570,
+                    296,
+                    570
+                ],
+                "score": 0.595
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    301,
+                    310,
+                    1398,
+                    310,
+                    1398,
+                    406,
+                    301,
+                    406
+                ],
+                "score": 0.586
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    293,
+                    423,
+                    1401,
+                    423,
+                    1401,
+                    488,
+                    293,
+                    488
+                ],
+                "score": 0.558
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    1941,
+                    1402,
+                    1941,
+                    1402,
+                    2035,
+                    299,
+                    2035
+                ],
+                "score": 0.541
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1118,
+                    1388,
+                    1172,
+                    1388,
+                    1172,
+                    1422,
+                    1118,
+                    1422
+                ],
+                "score": 0.91,
+                "latex": "o ( n )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    797,
+                    1227,
+                    816,
+                    1227,
+                    816,
+                    1257,
+                    797,
+                    1257
+                ],
+                "score": 0.85,
+                "latex": "\\phi"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    73.0,
+                    815.0,
+                    73.0,
+                    815.0,
+                    108.0,
+                    295.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    832.0,
+                    2085.0,
+                    868.0,
+                    2085.0,
+                    868.0,
+                    2125.0,
+                    832.0,
+                    2125.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    228.0,
+                    1404.0,
+                    228.0,
+                    1404.0,
+                    264.0,
+                    295.0,
+                    264.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    258.0,
+                    451.0,
+                    258.0,
+                    451.0,
+                    295.0,
+                    323.0,
+                    295.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    946.0,
+                    1406.0,
+                    946.0,
+                    1406.0,
+                    982.0,
+                    295.0,
+                    982.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    320.0,
+                    978.0,
+                    736.0,
+                    978.0,
+                    736.0,
+                    1012.0,
+                    320.0,
+                    1012.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1305.0,
+                    1404.0,
+                    1305.0,
+                    1404.0,
+                    1341.0,
+                    295.0,
+                    1341.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    321.0,
+                    1337.0,
+                    595.0,
+                    1337.0,
+                    595.0,
+                    1371.0,
+                    321.0,
+                    1371.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1026.0,
+                    1405.0,
+                    1026.0,
+                    1405.0,
+                    1068.0,
+                    292.0,
+                    1068.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    1061.0,
+                    1057.0,
+                    1061.0,
+                    1057.0,
+                    1094.0,
+                    323.0,
+                    1094.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    863.0,
+                    1406.0,
+                    863.0,
+                    1406.0,
+                    903.0,
+                    295.0,
+                    903.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    895.0,
+                    789.0,
+                    895.0,
+                    789.0,
+                    930.0,
+                    322.0,
+                    930.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    586.0,
+                    1405.0,
+                    586.0,
+                    1405.0,
+                    625.0,
+                    293.0,
+                    625.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    320.0,
+                    615.0,
+                    1406.0,
+                    615.0,
+                    1406.0,
+                    658.0,
+                    320.0,
+                    658.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    324.0,
+                    649.0,
+                    508.0,
+                    649.0,
+                    508.0,
+                    684.0,
+                    324.0,
+                    684.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    698.0,
+                    1403.0,
+                    698.0,
+                    1403.0,
+                    735.0,
+                    296.0,
+                    735.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    731.0,
+                    1263.0,
+                    731.0,
+                    1263.0,
+                    767.0,
+                    323.0,
+                    767.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1193.0,
+                    1404.0,
+                    1193.0,
+                    1404.0,
+                    1227.0,
+                    296.0,
+                    1227.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    1224.0,
+                    796.0,
+                    1224.0,
+                    796.0,
+                    1258.0,
+                    322.0,
+                    1258.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    817.0,
+                    1224.0,
+                    1403.0,
+                    1224.0,
+                    1403.0,
+                    1258.0,
+                    817.0,
+                    1258.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    321.0,
+                    1254.0,
+                    597.0,
+                    1254.0,
+                    597.0,
+                    1287.0,
+                    321.0,
+                    1287.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    783.0,
+                    1403.0,
+                    783.0,
+                    1403.0,
+                    819.0,
+                    295.0,
+                    819.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    324.0,
+                    814.0,
+                    1391.0,
+                    814.0,
+                    1391.0,
+                    850.0,
+                    324.0,
+                    850.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1850.0,
+                    1406.0,
+                    1850.0,
+                    1406.0,
+                    1900.0,
+                    293.0,
+                    1900.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    1890.0,
+                    775.0,
+                    1890.0,
+                    775.0,
+                    1924.0,
+                    323.0,
+                    1924.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1498.0,
+                    1405.0,
+                    1498.0,
+                    1405.0,
+                    1535.0,
+                    293.0,
+                    1535.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    1530.0,
+                    1103.0,
+                    1530.0,
+                    1103.0,
+                    1565.0,
+                    322.0,
+                    1565.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1694.0,
+                    1404.0,
+                    1694.0,
+                    1404.0,
+                    1730.0,
+                    295.0,
+                    1730.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    319.0,
+                    1723.0,
+                    1073.0,
+                    1723.0,
+                    1073.0,
+                    1762.0,
+                    319.0,
+                    1762.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1776.0,
+                    1402.0,
+                    1776.0,
+                    1402.0,
+                    1812.0,
+                    296.0,
+                    1812.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    1807.0,
+                    1099.0,
+                    1807.0,
+                    1099.0,
+                    1841.0,
+                    322.0,
+                    1841.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1384.0,
+                    1117.0,
+                    1384.0,
+                    1117.0,
+                    1425.0,
+                    293.0,
+                    1425.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1173.0,
+                    1384.0,
+                    1404.0,
+                    1384.0,
+                    1404.0,
+                    1425.0,
+                    1173.0,
+                    1425.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    1419.0,
+                    1404.0,
+                    1419.0,
+                    1404.0,
+                    1454.0,
+                    323.0,
+                    1454.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    319.0,
+                    1449.0,
+                    733.0,
+                    1449.0,
+                    733.0,
+                    1483.0,
+                    319.0,
+                    1483.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1581.0,
+                    1404.0,
+                    1581.0,
+                    1404.0,
+                    1619.0,
+                    294.0,
+                    1619.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    1614.0,
+                    1403.0,
+                    1614.0,
+                    1403.0,
+                    1648.0,
+                    322.0,
+                    1648.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    1644.0,
+                    461.0,
+                    1644.0,
+                    461.0,
+                    1676.0,
+                    323.0,
+                    1676.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1109.0,
+                    1404.0,
+                    1109.0,
+                    1404.0,
+                    1148.0,
+                    295.0,
+                    1148.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    1142.0,
+                    1337.0,
+                    1142.0,
+                    1337.0,
+                    1178.0,
+                    323.0,
+                    1178.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    507.0,
+                    1406.0,
+                    507.0,
+                    1406.0,
+                    543.0,
+                    296.0,
+                    543.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    321.0,
+                    536.0,
+                    806.0,
+                    536.0,
+                    806.0,
+                    571.0,
+                    321.0,
+                    571.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    310.0,
+                    1403.0,
+                    310.0,
+                    1403.0,
+                    348.0,
+                    296.0,
+                    348.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    342.0,
+                    1404.0,
+                    342.0,
+                    1404.0,
+                    379.0,
+                    322.0,
+                    379.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    321.0,
+                    373.0,
+                    1226.0,
+                    373.0,
+                    1226.0,
+                    409.0,
+                    321.0,
+                    409.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    426.0,
+                    1402.0,
+                    426.0,
+                    1402.0,
+                    458.0,
+                    297.0,
+                    458.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    320.0,
+                    453.0,
+                    1403.0,
+                    453.0,
+                    1403.0,
+                    490.0,
+                    320.0,
+                    490.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1938.0,
+                    1406.0,
+                    1938.0,
+                    1406.0,
+                    1978.0,
+                    293.0,
+                    1978.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    324.0,
+                    1972.0,
+                    1404.0,
+                    1972.0,
+                    1404.0,
+                    2006.0,
+                    324.0,
+                    2006.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    324.0,
+                    2003.0,
+                    706.0,
+                    2003.0,
+                    706.0,
+                    2037.0,
+                    324.0,
+                    2037.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 10,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 2,
+                "poly": [
+                    299,
+                    75,
+                    814,
+                    75,
+                    814,
+                    105,
+                    299,
+                    105
+                ],
+                "score": 0.87
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    836,
+                    2088,
+                    865,
+                    2088,
+                    865,
+                    2113,
+                    836,
+                    2113
+                ],
+                "score": 0.834
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    294,
+                    228,
+                    1401,
+                    228,
+                    1401,
+                    294,
+                    294,
+                    294
+                ],
+                "score": 0.579
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    295,
+                    311,
+                    1402,
+                    311,
+                    1402,
+                    377,
+                    295,
+                    377
+                ],
+                "score": 0.53
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    73.0,
+                    815.0,
+                    73.0,
+                    815.0,
+                    108.0,
+                    296.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    832.0,
+                    2085.0,
+                    870.0,
+                    2085.0,
+                    870.0,
+                    2124.0,
+                    832.0,
+                    2124.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    225.0,
+                    1405.0,
+                    225.0,
+                    1405.0,
+                    269.0,
+                    293.0,
+                    269.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    260.0,
+                    963.0,
+                    260.0,
+                    963.0,
+                    296.0,
+                    323.0,
+                    296.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    312.0,
+                    1404.0,
+                    312.0,
+                    1404.0,
+                    348.0,
+                    295.0,
+                    348.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    343.0,
+                    1277.0,
+                    343.0,
+                    1277.0,
+                    378.0,
+                    322.0,
+                    378.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 11,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    296,
+                    477,
+                    1406,
+                    477,
+                    1406,
+                    576,
+                    296,
+                    576
+                ],
+                "score": 0.976
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    296,
+                    1679,
+                    1404,
+                    1679,
+                    1404,
+                    1797,
+                    296,
+                    1797
+                ],
+                "score": 0.975
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    305,
+                    751,
+                    1385,
+                    751,
+                    1385,
+                    1617,
+                    305,
+                    1617
+                ],
+                "score": 0.963
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    653,
+                    1400,
+                    653,
+                    1400,
+                    735,
+                    297,
+                    735
+                ],
+                "score": 0.951
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    620,
+                    381,
+                    1078,
+                    381,
+                    1078,
+                    445,
+                    620,
+                    445
+                ],
+                "score": 0.948
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    500,
+                    1814,
+                    1201,
+                    1814,
+                    1201,
+                    1877,
+                    500,
+                    1877
+                ],
+                "score": 0.944
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    295,
+                    291,
+                    1401,
+                    291,
+                    1401,
+                    367,
+                    295,
+                    367
+                ],
+                "score": 0.942
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    564,
+                    592,
+                    1133,
+                    592,
+                    1133,
+                    638,
+                    564,
+                    638
+                ],
+                "score": 0.929
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    1909,
+                    782,
+                    1909,
+                    782,
+                    1945,
+                    299,
+                    1945
+                ],
+                "score": 0.929
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    298,
+                    73,
+                    814,
+                    73,
+                    814,
+                    106,
+                    298,
+                    106
+                ],
+                "score": 0.922
+            },
+            {
+                "category_id": 9,
+                "poly": [
+                    1365,
+                    397,
+                    1401,
+                    397,
+                    1401,
+                    429,
+                    1365,
+                    429
+                ],
+                "score": 0.886
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    342,
+                    1961,
+                    1352,
+                    1961,
+                    1352,
+                    2031,
+                    342,
+                    2031
+                ],
+                "score": 0.87
+            },
+            {
+                "category_id": 9,
+                "poly": [
+                    1364,
+                    1829,
+                    1401,
+                    1829,
+                    1401,
+                    1860,
+                    1364,
+                    1860
+                ],
+                "score": 0.868
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    836,
+                    2087,
+                    865,
+                    2087,
+                    865,
+                    2114,
+                    836,
+                    2114
+                ],
+                "score": 0.867
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    1374,
+                    1628,
+                    1404,
+                    1628,
+                    1404,
+                    1659,
+                    1374,
+                    1659
+                ],
+                "score": 0.838
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    299,
+                    225,
+                    474,
+                    225,
+                    474,
+                    262,
+                    299,
+                    262
+                ],
+                "score": 0.758
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    225,
+                    474,
+                    225,
+                    474,
+                    262,
+                    299,
+                    262
+                ],
+                "score": 0.174
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    346,
+                    1961,
+                    1350,
+                    1961,
+                    1350,
+                    2031,
+                    346,
+                    2031
+                ],
+                "score": 0.174
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    310,
+                    1176,
+                    1336,
+                    1176,
+                    1336,
+                    1620,
+                    310,
+                    1620
+                ],
+                "score": 0.95,
+                "latex": "\\begin{array} { r l } & { \\qquad + \\gamma \\mathbb { E } _ { d _ { 0 } } \\bigg [ \\int \\hat { V } ^ { x } ( s _ { 1 } ) \\bigg ( T ( s _ { 1 } | s _ { 0 } , a _ { 0 } ) - \\hat { T } ( s _ { 1 } | s _ { 0 } , a _ { 0 } ) \\bigg ) \\pi ( a _ { 0 } | s _ { 0 } ) d s _ { 1 } d a _ { 0 } \\bigg ] \\bigg | } \\\\ & { = \\left| \\mathbb { E } _ { ( s , a ) \\times d ^ { c } } \\left[ \\int \\bigg ( R ( r | s , a ) - \\hat { R } ( r | s , a ) \\bigg ) d r + \\gamma \\int \\hat { V } ^ { x } ( s ^ { \\prime } ) \\bigg ( T ( s ^ { \\prime } | s , a ) - \\hat { T } ( s ^ { \\prime } | s , a ) \\bigg ) d s ^ { \\prime } \\right] \\right| } \\\\ & { \\leq \\mathbb { E } _ { ( s , a ) \\sim d ^ { c } } \\left[ \\int \\bigg | R ( r | s , a ) - \\hat { R } ( r | s , a ) \\bigg | d r + \\gamma \\int \\hat { V } ^ { \\tau } ( s ^ { \\prime } ) \\bigg | T ( s ^ { \\prime } | s , a ) - \\hat { T } ( s ^ { \\prime } | s , a ) \\bigg | d s ^ { \\prime } \\right] } \\\\ & { \\leq \\mathbb { E } _ { ( s , a ) \\sim d ^ { c } } \\left[ \\int \\bigg | R ( r | s , a ) - \\hat { R } ( r | s , a ) \\bigg | d r + \\gamma \\int \\frac { r _ { \\operatorname* { m a x } } } { 1 - \\gamma } \\bigg | T ( s ^ { \\prime } | s , a ) - \\hat { T } ( s ^ { \\prime } | s , a ) \\bigg | d s ^ { \\prime } \\right] } \\\\ & { = 2 \\mathbb { E } _ { ( s , a ) \\sim d ^ { c } } \\left[ D _ { T V } \\big ( R ( r | s , a ) \\big | \\hat { R } _ { s } ( r | \\phi ( s , a ) ) \\big ) \\right] } \\\\ & { \\qquad + \\frac { 2 \\gamma r _ { \\operatorname* { m a x } } } { 1 - \\gamma } \\mathbb { E } _ { ( s , a ) \\sim d ^ { c } } \\left[ D _ { T V } \\big ( T ( s ^ { \\prime } | s , a ) \\big | \\hat { T } _ { 2 } ( s ^ { \\prime } | \\phi ( s , a ) ) \\big ) \\right] } \\end{array}"
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    345,
+                    1959,
+                    1350,
+                    1959,
+                    1350,
+                    2031,
+                    345,
+                    2031
+                ],
+                "score": 0.95,
+                "latex": "\\Big | R ^ { \\pi } - \\widehat { R } ^ { \\pi } \\Big | \\leq \\mathbb { E } _ { ( s , a ) \\sim d ^ { D } } \\Big [ \\mathcal { E } _ { \\phi , \\widehat { R } _ { z } , \\widehat { T } _ { z } } ( s , a ) \\Big ] + \\int \\mathcal { E } _ { \\phi , \\widehat { R } _ { z } , \\widehat { T } _ { z } } ( s , a ) \\left( d ^ { \\pi } ( s , a ) - d ^ { D } ( s , a ) \\right) d s d a ."
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    298,
+                    542,
+                    541,
+                    542,
+                    541,
+                    576,
+                    298,
+                    576
+                ],
+                "score": 0.94,
+                "latex": "( 1 - \\gamma ) \\mathbb { E } _ { s \\sim d _ { 0 } } \\overline { { [ V ^ { \\pi } ( s ) ] } }"
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    619,
+                    382,
+                    1079,
+                    382,
+                    1079,
+                    442,
+                    619,
+                    442
+                ],
+                "score": 0.94,
+                "latex": "\\begin{array} { r } { \\left| R ^ { \\pi } - \\widehat { R } ^ { \\pi } \\right| \\leq \\mathbb { E } _ { ( s , a ) \\sim d ^ { \\pi } } \\left[ \\mathcal { E } _ { \\phi , \\widehat { R } _ { z } , \\widehat { T } _ { z } } ( s , a ) \\right] } \\end{array}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    297,
+                    327,
+                    593,
+                    327,
+                    593,
+                    367,
+                    297,
+                    367
+                ],
+                "score": 0.93,
+                "latex": "( \\widehat { T } , \\widehat { R } ) = \\langle \\widehat { T } _ { z } \\circ \\phi , \\widehat { R } _ { z } \\circ \\phi \\rangle"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    619,
+                    1754,
+                    905,
+                    1754,
+                    905,
+                    1798,
+                    619,
+                    1798
+                ],
+                "score": 0.93,
+                "latex": "\\begin{array} { r } { \\frac { 1 } { B _ { \\phi } } \\mathcal { E } _ { \\phi , \\widehat { R } _ { z } , \\widehat { T } _ { z } } \\left( \\phi ^ { - 1 } ( \\cdot ) \\right) \\in \\mathcal { G } . } \\end{array}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    844,
+                    652,
+                    1237,
+                    652,
+                    1237,
+                    699,
+                    844,
+                    699
+                ],
+                "score": 0.93,
+                "latex": "\\widehat { V } ^ { \\pi } ( s ) = \\mathbb { E } _ { \\widehat { M } , \\pi } [ \\sum _ { t = 0 } ^ { \\infty } \\gamma ^ { t } r _ { t } | s _ { 0 } = s ]"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1068,
+                    1722,
+                    1165,
+                    1722,
+                    1165,
+                    1756,
+                    1068,
+                    1756
+                ],
+                "score": 0.92,
+                "latex": "B _ { \\phi } > 0"
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    310,
+                    749,
+                    1396,
+                    749,
+                    1396,
+                    1140,
+                    310,
+                    1140
+                ],
+                "score": 0.92,
+                "latex": "\\begin{array} { r l } & { \\displaystyle \\frac { 1 } { 1 - \\gamma } \\Big | R ^ { \\pi } - \\widehat { R } ^ { \\pi } \\Big | = \\left| \\mathbb { E } _ { d _ { 0 } } \\left[ V ^ { \\pi } ( s _ { 0 } ) - \\widehat { V } ^ { \\pi } ( s _ { 0 } ) \\right] \\right| } \\\\ & { = \\left| \\mathbb { E } _ { d _ { 0 } , \\pi } \\left[ r _ { 0 } - \\widehat { r } _ { 0 } \\right] + \\gamma \\mathbb { E } _ { d _ { 0 } } \\left[ \\int \\left( V ^ { \\pi } ( s _ { 1 } ) T ( s _ { 1 } | s _ { 0 } , a _ { 0 } ) - \\widehat { V } ^ { \\pi } ( s _ { 1 } ) \\widehat { T } ( s _ { 1 } | s _ { 0 } , a _ { 0 } ) \\right) \\pi ( a _ { 0 } | s _ { 0 } ) d s _ { 1 } d a _ { 0 } \\right] \\right| } \\\\ & { = \\left| \\mathbb { E } _ { d _ { 0 } , \\pi } \\left[ r _ { 0 } - \\widehat { r } _ { 0 } \\right] + \\gamma \\mathbb { E } _ { d _ { 0 } } \\left[ \\int \\left( V ^ { \\pi } ( s _ { 1 } ) T ( s _ { 1 } | s _ { 0 } , a _ { 0 } ) - \\widehat { V } ^ { \\pi } ( s _ { 1 } ) T ( s _ { 1 } | s _ { 0 } , a _ { 0 } ) \\right. \\right. } \\\\ & { \\qquad \\left. \\left. + \\widehat { V } ^ { \\pi } ( s _ { 1 } ) T ( s _ { 1 } | s _ { 0 } , a _ { 0 } ) - \\widehat { V } ^ { \\pi } ( s _ { 1 } ) \\widehat { T } ( s _ { 1 } | s _ { 0 } , a _ { 0 } ) \\right) \\pi ( a _ { 0 } | s _ { 0 } ) d s _ { 1 } d a _ { 0 } \\right] \\right| } \\\\ & { = \\left| \\mathbb { E } _ { d _ { 0 } , \\pi } \\left[ r _ { 0 } - \\widehat { r } _ { 0 } \\right] + \\gamma \\mathbb { E } _ { d _ { 0 } } \\left[ \\int \\left( V ^ { \\pi } ( s _ { 1 } ) - \\widehat { V } ^ { \\pi } ( s _ { 1 } ) \\right) T ( s _ { 1 } | s _ { 0 } , a _ { 0 } ) \\pi ( a _ { 0 } | s _ { 0 } ) d s _ { 1 } d a _ { 0 } \\right] \\right| } \\end{array}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    698,
+                    480,
+                    1112,
+                    480,
+                    1112,
+                    517,
+                    698,
+                    517
+                ],
+                "score": 0.92,
+                "latex": "V ^ { \\pi } ( s ) = \\mathbb { E } _ { M , \\pi } [ \\sum _ { t = 0 } ^ { \\infty } \\gamma ^ { t } r _ { t } | s _ { 0 } = s ]"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    346,
+                    1715,
+                    654,
+                    1715,
+                    654,
+                    1755,
+                    346,
+                    1755
+                ],
+                "score": 0.92,
+                "latex": "( \\widehat { T } , \\widehat { R } ) = \\langle \\widehat { T } _ { z } \\circ \\phi , \\widehat { R } _ { z } \\circ \\phi \\rangle"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    298,
+                    1756,
+                    506,
+                    1756,
+                    506,
+                    1791,
+                    298,
+                    1791
+                ],
+                "score": 0.91,
+                "latex": "{ \\mathcal { G } } \\triangleq \\{ g : { \\mathcal { Z } } \\to \\mathbb { R } \\}"
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    497,
+                    1814,
+                    1200,
+                    1814,
+                    1200,
+                    1875,
+                    497,
+                    1875
+                ],
+                "score": 0.91,
+                "latex": "\\left| R ^ { \\pi } - \\widehat { R } ^ { \\pi } \\right| \\leq \\mathbb { E } _ { ( s , a ) \\sim d ^ { \\mathcal { D } } } \\Big [ \\mathcal { E } _ { \\phi , \\widehat { R } _ { z } , \\widehat { T } _ { z } } ( s , a ) \\Big ] + B _ { \\phi } \\mathrm { I P M } _ { \\mathcal { G } } ( d _ { \\phi } ^ { \\pi } , d _ { \\phi } ^ { \\mathcal { D } } )"
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    565,
+                    591,
+                    1133,
+                    591,
+                    1133,
+                    636,
+                    565,
+                    636
+                ],
+                "score": 0.88,
+                "latex": "V ^ { \\pi } ( s ) = \\mathbb { E } _ { a \\sim \\pi ( s ) } \\left[ r ( s , a ) + \\gamma \\mathbb { E } _ { s ^ { \\prime } \\sim T ( s , a ) } [ V ( s ^ { \\prime } ) ] \\right] ,"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    397,
+                    696,
+                    430,
+                    696,
+                    430,
+                    730,
+                    397,
+                    730
+                ],
+                "score": 0.86,
+                "latex": "\\widehat { M }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1329,
+                    513,
+                    1403,
+                    513,
+                    1403,
+                    543,
+                    1329,
+                    543
+                ],
+                "score": 0.86,
+                "latex": "R ^ { \\pi } ="
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    882,
+                    1678,
+                    916,
+                    1678,
+                    916,
+                    1714,
+                    882,
+                    1714
+                ],
+                "score": 0.85,
+                "latex": "\\widehat { M }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    850,
+                    290,
+                    884,
+                    290,
+                    884,
+                    325,
+                    850,
+                    325
+                ],
+                "score": 0.84,
+                "latex": "\\widehat { M }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1375,
+                    1692,
+                    1392,
+                    1692,
+                    1392,
+                    1718,
+                    1375,
+                    1718
+                ],
+                "score": 0.82,
+                "latex": "\\phi"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1333,
+                    299,
+                    1354,
+                    299,
+                    1354,
+                    330,
+                    1333,
+                    330
+                ],
+                "score": 0.81,
+                "latex": "\\phi ,"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    627,
+                    519,
+                    643,
+                    519,
+                    643,
+                    540,
+                    627,
+                    540
+                ],
+                "score": 0.74,
+                "latex": "s"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1146,
+                    1765,
+                    1165,
+                    1765,
+                    1165,
+                    1785,
+                    1146,
+                    1785
+                ],
+                "score": 0.7,
+                "latex": "\\pi"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1032,
+                    340,
+                    1052,
+                    340,
+                    1052,
+                    361,
+                    1032,
+                    361
+                ],
+                "score": 0.64,
+                "latex": "\\pi"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    654,
+                    1686,
+                    687,
+                    1686,
+                    687,
+                    1713,
+                    654,
+                    1713
+                ],
+                "score": 0.5,
+                "latex": "M"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    629,
+                    298,
+                    662,
+                    298,
+                    662,
+                    326,
+                    629,
+                    326
+                ],
+                "score": 0.38,
+                "latex": "M"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    72.0,
+                    815.0,
+                    72.0,
+                    815.0,
+                    108.0,
+                    296.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    832.0,
+                    2085.0,
+                    870.0,
+                    2085.0,
+                    870.0,
+                    2123.0,
+                    832.0,
+                    2123.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1378.0,
+                    1633.0,
+                    1403.0,
+                    1633.0,
+                    1403.0,
+                    1661.0,
+                    1378.0,
+                    1661.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    220.0,
+                    477.0,
+                    220.0,
+                    477.0,
+                    270.0,
+                    292.0,
+                    270.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    286.0,
+                    468.0,
+                    697.0,
+                    468.0,
+                    697.0,
+                    530.0,
+                    286.0,
+                    530.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1113.0,
+                    468.0,
+                    1413.0,
+                    468.0,
+                    1413.0,
+                    530.0,
+                    1113.0,
+                    530.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    512.0,
+                    626.0,
+                    512.0,
+                    626.0,
+                    545.0,
+                    293.0,
+                    545.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    644.0,
+                    512.0,
+                    1328.0,
+                    512.0,
+                    1328.0,
+                    545.0,
+                    644.0,
+                    545.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    542.0,
+                    543.0,
+                    1055.0,
+                    543.0,
+                    1055.0,
+                    577.0,
+                    542.0,
+                    577.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1677.0,
+                    653.0,
+                    1677.0,
+                    653.0,
+                    1723.0,
+                    294.0,
+                    1723.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    688.0,
+                    1677.0,
+                    881.0,
+                    1677.0,
+                    881.0,
+                    1723.0,
+                    688.0,
+                    1723.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    917.0,
+                    1677.0,
+                    1374.0,
+                    1677.0,
+                    1374.0,
+                    1723.0,
+                    917.0,
+                    1723.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1393.0,
+                    1677.0,
+                    1404.0,
+                    1677.0,
+                    1404.0,
+                    1723.0,
+                    1393.0,
+                    1723.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1713.0,
+                    345.0,
+                    1713.0,
+                    345.0,
+                    1760.0,
+                    294.0,
+                    1760.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    655.0,
+                    1713.0,
+                    1067.0,
+                    1713.0,
+                    1067.0,
+                    1760.0,
+                    655.0,
+                    1760.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1166.0,
+                    1713.0,
+                    1406.0,
+                    1713.0,
+                    1406.0,
+                    1760.0,
+                    1166.0,
+                    1760.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    286.0,
+                    1738.0,
+                    297.0,
+                    1738.0,
+                    297.0,
+                    1810.0,
+                    286.0,
+                    1810.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    507.0,
+                    1738.0,
+                    618.0,
+                    1738.0,
+                    618.0,
+                    1810.0,
+                    507.0,
+                    1810.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    906.0,
+                    1738.0,
+                    1145.0,
+                    1738.0,
+                    1145.0,
+                    1810.0,
+                    906.0,
+                    1810.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1166.0,
+                    1738.0,
+                    1187.0,
+                    1738.0,
+                    1187.0,
+                    1810.0,
+                    1166.0,
+                    1810.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    288.0,
+                    645.0,
+                    843.0,
+                    645.0,
+                    843.0,
+                    702.0,
+                    288.0,
+                    702.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1238.0,
+                    645.0,
+                    1411.0,
+                    645.0,
+                    1411.0,
+                    702.0,
+                    1238.0,
+                    702.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    691.0,
+                    396.0,
+                    691.0,
+                    396.0,
+                    740.0,
+                    291.0,
+                    740.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    431.0,
+                    691.0,
+                    518.0,
+                    691.0,
+                    518.0,
+                    740.0,
+                    431.0,
+                    740.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    289.0,
+                    628.0,
+                    289.0,
+                    628.0,
+                    335.0,
+                    294.0,
+                    335.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    663.0,
+                    289.0,
+                    849.0,
+                    289.0,
+                    849.0,
+                    335.0,
+                    663.0,
+                    335.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    885.0,
+                    289.0,
+                    1332.0,
+                    289.0,
+                    1332.0,
+                    335.0,
+                    885.0,
+                    335.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1355.0,
+                    289.0,
+                    1403.0,
+                    289.0,
+                    1403.0,
+                    335.0,
+                    1355.0,
+                    335.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    594.0,
+                    325.0,
+                    1031.0,
+                    325.0,
+                    1031.0,
+                    373.0,
+                    594.0,
+                    373.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1053.0,
+                    325.0,
+                    1279.0,
+                    325.0,
+                    1279.0,
+                    373.0,
+                    1053.0,
+                    373.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1906.0,
+                    783.0,
+                    1906.0,
+                    783.0,
+                    1948.0,
+                    295.0,
+                    1948.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    220.0,
+                    477.0,
+                    220.0,
+                    477.0,
+                    270.0,
+                    292.0,
+                    270.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 12,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    295,
+                    1068,
+                    1406,
+                    1068,
+                    1406,
+                    1225,
+                    295,
+                    1225
+                ],
+                "score": 0.978
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    399,
+                    267,
+                    1298,
+                    267,
+                    1298,
+                    694,
+                    399,
+                    694
+                ],
+                "score": 0.969
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    299,
+                    1650,
+                    1400,
+                    1650,
+                    1400,
+                    1854,
+                    299,
+                    1854
+                ],
+                "score": 0.962
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    295,
+                    1419,
+                    1404,
+                    1419,
+                    1404,
+                    1510,
+                    295,
+                    1510
+                ],
+                "score": 0.957
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    566,
+                    1232,
+                    1136,
+                    1232,
+                    1136,
+                    1321,
+                    566,
+                    1321
+                ],
+                "score": 0.949
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    508,
+                    1954,
+                    1187,
+                    1954,
+                    1187,
+                    2043,
+                    508,
+                    2043
+                ],
+                "score": 0.945
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    452,
+                    974,
+                    1248,
+                    974,
+                    1248,
+                    1049,
+                    452,
+                    1049
+                ],
+                "score": 0.938
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    295,
+                    1865,
+                    1400,
+                    1865,
+                    1400,
+                    1945,
+                    295,
+                    1945
+                ],
+                "score": 0.935
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    301,
+                    796,
+                    1403,
+                    796,
+                    1403,
+                    859,
+                    301,
+                    859
+                ],
+                "score": 0.931
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    302,
+                    1326,
+                    1143,
+                    1326,
+                    1143,
+                    1361,
+                    302,
+                    1361
+                ],
+                "score": 0.924
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    758,
+                    1065,
+                    758,
+                    1065,
+                    792,
+                    298,
+                    792
+                ],
+                "score": 0.921
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    933,
+                    538,
+                    933,
+                    538,
+                    966,
+                    298,
+                    966
+                ],
+                "score": 0.92
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    460,
+                    865,
+                    1236,
+                    865,
+                    1236,
+                    925,
+                    460,
+                    925
+                ],
+                "score": 0.917
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    298,
+                    73,
+                    814,
+                    73,
+                    814,
+                    106,
+                    298,
+                    106
+                ],
+                "score": 0.916
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1380,
+                    883,
+                    1380,
+                    883,
+                    1414,
+                    297,
+                    1414
+                ],
+                "score": 0.911
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    230,
+                    366,
+                    230,
+                    366,
+                    261,
+                    297,
+                    261
+                ],
+                "score": 0.908
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    300,
+                    1518,
+                    1399,
+                    1518,
+                    1399,
+                    1597,
+                    300,
+                    1597
+                ],
+                "score": 0.904
+            },
+            {
+                "category_id": 9,
+                "poly": [
+                    1351,
+                    1260,
+                    1401,
+                    1260,
+                    1401,
+                    1292,
+                    1351,
+                    1292
+                ],
+                "score": 0.899
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    296,
+                    1604,
+                    965,
+                    1604,
+                    965,
+                    1641,
+                    296,
+                    1641
+                ],
+                "score": 0.886
+            },
+            {
+                "category_id": 9,
+                "poly": [
+                    1351,
+                    869,
+                    1400,
+                    869,
+                    1400,
+                    900,
+                    1351,
+                    900
+                ],
+                "score": 0.883
+            },
+            {
+                "category_id": 9,
+                "poly": [
+                    1351,
+                    993,
+                    1400,
+                    993,
+                    1400,
+                    1025,
+                    1351,
+                    1025
+                ],
+                "score": 0.876
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    835,
+                    2087,
+                    865,
+                    2087,
+                    865,
+                    2113,
+                    835,
+                    2113
+                ],
+                "score": 0.868
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    1374,
+                    693,
+                    1404,
+                    693,
+                    1404,
+                    724,
+                    1374,
+                    724
+                ],
+                "score": 0.84
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    395,
+                    265,
+                    1300,
+                    265,
+                    1300,
+                    699,
+                    395,
+                    699
+                ],
+                "score": 0.95,
+                "latex": "\\begin{array} { r l } & { \\displaystyle \\int \\mathcal { E } _ { \\phi , \\widehat { R } _ { + } , \\widehat { T } _ { z } } ( s , a ) \\left( d ^ { \\pi } ( s , a ) - d ^ { D } ( s , a ) \\right) d s d a } \\\\ & { \\quad = \\displaystyle \\int \\mathcal { E } _ { \\phi , \\widehat { R } _ { \\epsilon } , \\widehat { T } _ { z } } ( \\phi ^ { - 1 } ( z ) ) \\left( d ^ { \\pi } ( z ) - d ^ { D } \\left( \\phi ^ { - 1 } ( \\phi ^ { - 1 } ( z ) ) \\right) \\right) \\left. \\frac { d ( s , a ) } { d z } \\right. d z } \\\\ & { \\quad = \\displaystyle \\int \\mathcal { E } _ { \\phi , \\widehat { R } _ { \\epsilon } , \\widehat { T } _ { z } } ( \\phi ^ { - 1 } ( z ) ) \\left( d _ { \\phi } ^ { \\pi } ( z ) - d _ { \\phi } ^ { D } ( z ) \\right) d z } \\\\ & { \\quad \\leq { B } _ { \\phi } \\displaystyle \\left. \\int \\frac { 1 } { B _ { \\phi } } \\mathcal { E } _ { \\phi , \\widehat { R } _ { \\epsilon } , \\widehat { T } _ { z } } ( \\phi ^ { - 1 } ( z ) ) \\left( d _ { \\phi } ^ { \\pi } ( z ) - d _ { \\phi } ^ { D } ( z ) \\right) d z \\right. \\qquad \\left( { B } _ { \\phi } > 0 \\right) } \\\\ & { \\quad \\leq { B } _ { \\phi \\in \\nabla } \\displaystyle \\left. \\int \\mathcal { J } ( { \\boldsymbol \\jmath } ^ { \\pi } ) \\left( d _ { \\phi } ^ { \\pi } ( z ) - d _ { \\phi } ^ { D } ( z ) \\right) d z \\right. \\qquad \\left( \\frac { 1 } { B _ { \\phi } } \\mathcal { E } _ { \\phi , \\widehat { R } _ { \\epsilon } , \\widehat { T } _ { z } } ( \\phi ^ { - 1 } ( z ) ) \\in \\mathcal { G } \\right) } \\\\ & { \\quad = { B } _ { \\phi } \\mathrm { F N M } _ { \\phi } \\left( a _ { \\phi } ^ { \\pi } , d _ { \\phi } ^ { \\theta } \\right) } \\end{array}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    996,
+                    1867,
+                    1159,
+                    1867,
+                    1159,
+                    1911,
+                    996,
+                    1911
+                ],
+                "score": 0.94,
+                "latex": "\\widehat { \\mathrm { I P M } } ( d _ { \\phi } ^ { \\pi } , d _ { \\phi } ^ { D } ) ^ { 2 }"
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    565,
+                    1229,
+                    1135,
+                    1229,
+                    1135,
+                    1323,
+                    565,
+                    1323
+                ],
+                "score": 0.94,
+                "latex": "R _ { n } ( \\mathcal { F } ) \\triangleq \\mathbb { E } _ { \\{ x _ { i } \\} _ { i = 1 } ^ { n } , \\sigma } \\operatorname* { s u p } _ { f \\in \\mathcal { F } } \\left| \\frac { 1 } { n } \\sum _ { i = 1 } ^ { n } \\sigma _ { i } f ( x _ { i } ) \\right| \\leq \\sqrt { \\frac { \\bar { k } } { n } } ."
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    338,
+                    1603,
+                    507,
+                    1603,
+                    507,
+                    1645,
+                    338,
+                    1645
+                ],
+                "score": 0.93,
+                "latex": "\\mathrm { I P M } _ { \\mathcal { G } } ( d _ { \\phi } ^ { \\pi } , d _ { \\phi } ^ { D } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    628,
+                    1163,
+                    718,
+                    1163,
+                    718,
+                    1196,
+                    628,
+                    1196
+                ],
+                "score": 0.93,
+                "latex": "\\{ - 1 , 1 \\}"
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    509,
+                    1955,
+                    1187,
+                    1955,
+                    1187,
+                    2044,
+                    509,
+                    2044
+                ],
+                "score": 0.93,
+                "latex": "\\left| \\mathrm { I P M } _ { \\mathcal { G } } ( d _ { \\phi } ^ { \\pi } , d _ { \\phi } ^ { D } ) - \\widehat { \\mathrm { I P M } } ( d _ { \\phi } ^ { \\pi } , d _ { \\phi } ^ { D } ) \\right| \\leq \\sqrt { \\frac { \\bar { k } } { n } } \\left( 4 + \\sqrt { 8 \\log \\frac { 3 } { \\delta } } \\right) ."
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    521,
+                    1909,
+                    701,
+                    1909,
+                    701,
+                    1948,
+                    521,
+                    1948
+                ],
+                "score": 0.92,
+                "latex": "\\mathrm { I P M } _ { \\mathcal { G } } ( d _ { \\phi } ^ { \\pi } , d _ { \\phi } ^ { \\mathcal { D } } ) ^ { 2 }"
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    310,
+                    1648,
+                    1403,
+                    1648,
+                    1403,
+                    1855,
+                    310,
+                    1855
+                ],
+                "score": 0.92,
+                "latex": "\\begin{array} { r l } & { \\mathrm { [ P M } _ { \\mathcal { G } } ( d _ { \\phi } ^ { \\pi } , d _ { \\phi } ^ { \\mathcal { D } } ) ^ { 2 } = \\mathbb { E } _ { s _ { 0 } \\sim d _ { 0 } , a _ { 0 } \\sim \\pi ( s _ { 0 } ) , ( s , a , s ^ { \\prime } ) \\sim d ^ { \\mathcal { D } } , a ^ { \\prime } \\sim \\pi ( s ^ { \\prime } ) } } \\\\ & { \\quad \\quad k ( \\phi ( s , a ) , \\phi ( \\bar { s } , \\bar { a } ) ) + ( 1 - \\gamma ) ^ { 2 } k ( \\phi ( s _ { 0 } , a _ { 0 } ) , \\phi ( \\bar { s } _ { 0 } , \\bar { a } _ { 0 } ) ) + \\gamma ^ { 2 } k ( \\phi ( s ^ { \\prime } , a ^ { \\prime } ) , \\phi ( \\bar { s } ^ { \\prime } , \\bar { a } ^ { \\prime } ) ) } \\\\ & { \\quad - 2 ( 1 - \\gamma ) k ( \\phi ( s _ { 0 } , a _ { 0 } ) , \\phi ( \\bar { s } , \\bar { a } ) ) - 2 \\gamma k ( \\phi ( s ^ { \\prime } , a ^ { \\prime } ) , \\phi ( \\bar { s } , \\bar { a } ) ) + 2 \\gamma ( 1 - \\gamma ) k ( \\phi ( s ^ { \\prime } , a ^ { \\prime } ) , \\phi ( \\bar { s } _ { 0 } , \\bar { a } _ { 0 } ) ) \\bigg ] . } \\end{array}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    932,
+                    1450,
+                    1177,
+                    1450,
+                    1177,
+                    1484,
+                    932,
+                    1484
+                ],
+                "score": 0.92,
+                "latex": "\\nu ( z ) = \\langle \\nu , \\dot { k } ( \\cdot , z ) \\rangle _ { \\mathcal { H } _ { k } }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1201,
+                    1100,
+                    1393,
+                    1100,
+                    1393,
+                    1135,
+                    1201,
+                    1135
+                ],
+                "score": 0.92,
+                "latex": "0 \\leq k ( x , x ^ { \\prime } ) \\leq \\bar { k }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    853,
+                    829,
+                    996,
+                    829,
+                    996,
+                    860,
+                    853,
+                    860
+                ],
+                "score": 0.92,
+                "latex": "f : \\mathcal { X } ^ { n }  \\mathbb { R }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    341,
+                    1132,
+                    432,
+                    1132,
+                    432,
+                    1166,
+                    341,
+                    1166
+                ],
+                "score": 0.92,
+                "latex": "\\{ x _ { i } \\} _ { i = 1 } ^ { n }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1033,
+                    797,
+                    1131,
+                    797,
+                    1131,
+                    831,
+                    1033,
+                    831
+                ],
+                "score": 0.91,
+                "latex": "\\{ X _ { i } \\} _ { i = 1 } ^ { n }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    600,
+                    1871,
+                    821,
+                    1871,
+                    821,
+                    1907,
+                    600,
+                    1907
+                ],
+                "score": 0.91,
+                "latex": "\\bar { k } \\triangleq \\operatorname* { s u p } _ { z \\in \\mathcal { Z } } k ( z , z )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1175,
+                    1419,
+                    1257,
+                    1419,
+                    1257,
+                    1453,
+                    1175,
+                    1453
+                ],
+                "score": 0.91,
+                "latex": "\\langle \\cdot , \\cdot \\rangle _ { \\mathcal { H } _ { k } }"
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    453,
+                    970,
+                    1250,
+                    970,
+                    1250,
+                    1050,
+                    453,
+                    1050
+                ],
+                "score": 0.91,
+                "latex": "{ \\operatorname* { P r } } \\{ f ( \\{ X _ { i } \\} _ { i = 1 } ^ { n } ) - \\mathbb { E } _ { \\{ X _ { i } \\} _ { i = 1 } ^ { n } } [ f ( \\{ X _ { i } \\} _ { i = 1 } ^ { n } ) ] \\geq \\epsilon \\} \\leq \\exp \\left( - \\frac { 2 \\epsilon ^ { 2 } } { \\sum _ { i = 1 } ^ { n } c _ { i } ^ { 2 } } \\right) ."
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    466,
+                    936,
+                    530,
+                    936,
+                    530,
+                    963,
+                    466,
+                    963
+                ],
+                "score": 0.89,
+                "latex": "\\epsilon > 0"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    421,
+                    1451,
+                    459,
+                    1451,
+                    459,
+                    1481,
+                    421,
+                    1481
+                ],
+                "score": 0.89,
+                "latex": "\\mathcal { H } _ { k }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    501,
+                    1420,
+                    541,
+                    1420,
+                    541,
+                    1450,
+                    501,
+                    1450
+                ],
+                "score": 0.89,
+                "latex": "\\mathcal { H } _ { k }"
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    310,
+                    1507,
+                    1402,
+                    1507,
+                    1402,
+                    1599,
+                    310,
+                    1599
+                ],
+                "score": 0.89,
+                "latex": "\\begin{array} { r } { \\boldsymbol { \\operatorname { \\mathrm { ? } } } = \\biggl \\{ g \\in ( \\mathcal { Z } \\to \\mathbb { R } ) : g ( z ) = \\nu ( z ) - \\gamma \\mathbb { E } _ { s ^ { \\prime } \\sim T ( \\phi ^ { - 1 } ( z ) ) } [ \\nu ( \\phi ( s ^ { \\prime } , a ^ { \\prime } ) ) ] , \\nu \\in ( \\mathcal { Z } \\to \\mathbb { R } ) , \\langle \\nu , \\nu \\rangle _ { \\mathcal { H } _ { k } } \\leq 1 \\biggr \\} , } \\end{array}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1047,
+                    1419,
+                    1113,
+                    1419,
+                    1113,
+                    1451,
+                    1047,
+                    1451
+                ],
+                "score": 0.88,
+                "latex": "k ( \\cdot , \\cdot )"
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    465,
+                    865,
+                    1236,
+                    865,
+                    1236,
+                    927,
+                    465,
+                    927
+                ],
+                "score": 0.87,
+                "latex": "\\operatorname* { s u p } _ { \\{ x _ { i } \\} _ { i = 1 } ^ { n } \\in \\mathcal { X } ^ { n } , \\tilde { x } \\in \\mathcal { X } } | f ( \\{ x _ { i } \\} _ { i = 1 } ^ { n } ) - f ( x _ { 1 } , . . . , x _ { i - 1 } , \\tilde { x } , x _ { i + 1 } , . . . , x _ { n } ) | \\leq c _ { i } ."
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1257,
+                    1911,
+                    1319,
+                    1911,
+                    1319,
+                    1939,
+                    1257,
+                    1939
+                ],
+                "score": 0.86,
+                "latex": "1 - \\delta"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1284,
+                    1137,
+                    1308,
+                    1137,
+                    1308,
+                    1160,
+                    1284,
+                    1160
+                ],
+                "score": 0.82,
+                "latex": "\\mathcal { X }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    298,
+                    1167,
+                    325,
+                    1167,
+                    325,
+                    1194,
+                    298,
+                    1194
+                ],
+                "score": 0.82,
+                "latex": "\\sigma _ { i }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1268,
+                    1451,
+                    1289,
+                    1451,
+                    1289,
+                    1480,
+                    1268,
+                    1480
+                ],
+                "score": 0.81,
+                "latex": "\\mathcal { G }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    790,
+                    1104,
+                    814,
+                    1104,
+                    814,
+                    1128,
+                    790,
+                    1128
+                ],
+                "score": 0.8,
+                "latex": "\\mathcal { X }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    628,
+                    829,
+                    653,
+                    829,
+                    653,
+                    855,
+                    628,
+                    855
+                ],
+                "score": 0.79,
+                "latex": "\\mathcal { X }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1227,
+                    1074,
+                    1252,
+                    1074,
+                    1252,
+                    1099,
+                    1227,
+                    1099
+                ],
+                "score": 0.77,
+                "latex": "\\mathcal { F }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    765,
+                    1922,
+                    782,
+                    1922,
+                    782,
+                    1937,
+                    765,
+                    1937
+                ],
+                "score": 0.62,
+                "latex": "n"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1227,
+                    1140,
+                    1245,
+                    1140,
+                    1245,
+                    1165,
+                    1227,
+                    1165
+                ],
+                "score": 0.28,
+                "latex": "p"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    72.0,
+                    815.0,
+                    72.0,
+                    815.0,
+                    108.0,
+                    296.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    832.0,
+                    2085.0,
+                    870.0,
+                    2085.0,
+                    870.0,
+                    2124.0,
+                    832.0,
+                    2124.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1378.0,
+                    698.0,
+                    1403.0,
+                    698.0,
+                    1403.0,
+                    725.0,
+                    1378.0,
+                    725.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1072.0,
+                    1226.0,
+                    1072.0,
+                    1226.0,
+                    1105.0,
+                    295.0,
+                    1105.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1253.0,
+                    1072.0,
+                    1404.0,
+                    1072.0,
+                    1404.0,
+                    1105.0,
+                    1253.0,
+                    1105.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1100.0,
+                    789.0,
+                    1100.0,
+                    789.0,
+                    1137.0,
+                    293.0,
+                    1137.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    815.0,
+                    1100.0,
+                    1200.0,
+                    1100.0,
+                    1200.0,
+                    1137.0,
+                    815.0,
+                    1137.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1394.0,
+                    1100.0,
+                    1404.0,
+                    1100.0,
+                    1404.0,
+                    1137.0,
+                    1394.0,
+                    1137.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1129.0,
+                    340.0,
+                    1129.0,
+                    340.0,
+                    1171.0,
+                    293.0,
+                    1171.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    433.0,
+                    1129.0,
+                    1226.0,
+                    1129.0,
+                    1226.0,
+                    1171.0,
+                    433.0,
+                    1171.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1246.0,
+                    1129.0,
+                    1283.0,
+                    1129.0,
+                    1283.0,
+                    1171.0,
+                    1246.0,
+                    1171.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1309.0,
+                    1129.0,
+                    1407.0,
+                    1129.0,
+                    1407.0,
+                    1171.0,
+                    1309.0,
+                    1171.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1161.0,
+                    297.0,
+                    1161.0,
+                    297.0,
+                    1199.0,
+                    294.0,
+                    1199.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    326.0,
+                    1161.0,
+                    627.0,
+                    1161.0,
+                    627.0,
+                    1199.0,
+                    326.0,
+                    1199.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    719.0,
+                    1161.0,
+                    1406.0,
+                    1161.0,
+                    1406.0,
+                    1199.0,
+                    719.0,
+                    1199.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1192.0,
+                    733.0,
+                    1192.0,
+                    733.0,
+                    1228.0,
+                    294.0,
+                    1228.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1414.0,
+                    500.0,
+                    1414.0,
+                    500.0,
+                    1455.0,
+                    292.0,
+                    1455.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    542.0,
+                    1414.0,
+                    1046.0,
+                    1414.0,
+                    1046.0,
+                    1455.0,
+                    542.0,
+                    1455.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1114.0,
+                    1414.0,
+                    1174.0,
+                    1414.0,
+                    1174.0,
+                    1455.0,
+                    1114.0,
+                    1455.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1258.0,
+                    1414.0,
+                    1405.0,
+                    1414.0,
+                    1405.0,
+                    1455.0,
+                    1258.0,
+                    1455.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1449.0,
+                    420.0,
+                    1449.0,
+                    420.0,
+                    1486.0,
+                    292.0,
+                    1486.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    460.0,
+                    1449.0,
+                    931.0,
+                    1449.0,
+                    931.0,
+                    1486.0,
+                    460.0,
+                    1486.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1178.0,
+                    1449.0,
+                    1267.0,
+                    1449.0,
+                    1267.0,
+                    1486.0,
+                    1178.0,
+                    1486.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1290.0,
+                    1449.0,
+                    1405.0,
+                    1449.0,
+                    1405.0,
+                    1486.0,
+                    1290.0,
+                    1486.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1479.0,
+                    409.0,
+                    1479.0,
+                    409.0,
+                    1513.0,
+                    293.0,
+                    1513.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1864.0,
+                    599.0,
+                    1864.0,
+                    599.0,
+                    1912.0,
+                    294.0,
+                    1912.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    822.0,
+                    1864.0,
+                    995.0,
+                    1864.0,
+                    995.0,
+                    1912.0,
+                    822.0,
+                    1912.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1160.0,
+                    1864.0,
+                    1404.0,
+                    1864.0,
+                    1404.0,
+                    1912.0,
+                    1160.0,
+                    1912.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1902.0,
+                    520.0,
+                    1902.0,
+                    520.0,
+                    1950.0,
+                    293.0,
+                    1950.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    702.0,
+                    1902.0,
+                    764.0,
+                    1902.0,
+                    764.0,
+                    1950.0,
+                    702.0,
+                    1950.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    783.0,
+                    1902.0,
+                    1256.0,
+                    1902.0,
+                    1256.0,
+                    1950.0,
+                    783.0,
+                    1950.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1320.0,
+                    1902.0,
+                    1332.0,
+                    1902.0,
+                    1332.0,
+                    1950.0,
+                    1320.0,
+                    1950.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    792.0,
+                    1032.0,
+                    792.0,
+                    1032.0,
+                    839.0,
+                    295.0,
+                    839.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1132.0,
+                    792.0,
+                    1407.0,
+                    792.0,
+                    1407.0,
+                    839.0,
+                    1132.0,
+                    839.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    824.0,
+                    627.0,
+                    824.0,
+                    627.0,
+                    864.0,
+                    294.0,
+                    864.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    654.0,
+                    824.0,
+                    852.0,
+                    824.0,
+                    852.0,
+                    864.0,
+                    654.0,
+                    864.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    997.0,
+                    824.0,
+                    1096.0,
+                    824.0,
+                    1096.0,
+                    864.0,
+                    997.0,
+                    864.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1323.0,
+                    1148.0,
+                    1323.0,
+                    1148.0,
+                    1365.0,
+                    294.0,
+                    1365.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    756.0,
+                    1060.0,
+                    756.0,
+                    1060.0,
+                    796.0,
+                    296.0,
+                    796.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    929.0,
+                    465.0,
+                    929.0,
+                    465.0,
+                    971.0,
+                    295.0,
+                    971.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    531.0,
+                    929.0,
+                    541.0,
+                    929.0,
+                    541.0,
+                    971.0,
+                    531.0,
+                    971.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1380.0,
+                    885.0,
+                    1380.0,
+                    885.0,
+                    1417.0,
+                    295.0,
+                    1417.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    226.0,
+                    373.0,
+                    226.0,
+                    373.0,
+                    269.0,
+                    295.0,
+                    269.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1602.0,
+                    337.0,
+                    1602.0,
+                    337.0,
+                    1646.0,
+                    295.0,
+                    1646.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    508.0,
+                    1602.0,
+                    967.0,
+                    1602.0,
+                    967.0,
+                    1646.0,
+                    508.0,
+                    1646.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 13,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 8,
+                "poly": [
+                    332,
+                    463,
+                    1362,
+                    463,
+                    1362,
+                    894,
+                    332,
+                    894
+                ],
+                "score": 0.969
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    312,
+                    1495,
+                    1389,
+                    1495,
+                    1389,
+                    1703,
+                    312,
+                    1703
+                ],
+                "score": 0.965
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    295,
+                    1369,
+                    1405,
+                    1369,
+                    1405,
+                    1490,
+                    295,
+                    1490
+                ],
+                "score": 0.965
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    296,
+                    226,
+                    1404,
+                    226,
+                    1404,
+                    294,
+                    296,
+                    294
+                ],
+                "score": 0.94
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    301,
+                    1092,
+                    1394,
+                    1092,
+                    1394,
+                    1161,
+                    301,
+                    1161
+                ],
+                "score": 0.935
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    295,
+                    1711,
+                    1403,
+                    1711,
+                    1403,
+                    1774,
+                    295,
+                    1774
+                ],
+                "score": 0.935
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    583,
+                    1046,
+                    1111,
+                    1046,
+                    1111,
+                    1087,
+                    583,
+                    1087
+                ],
+                "score": 0.926
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    298,
+                    73,
+                    814,
+                    73,
+                    814,
+                    106,
+                    298,
+                    106
+                ],
+                "score": 0.921
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    376,
+                    1779,
+                    1327,
+                    1779,
+                    1327,
+                    2030,
+                    376,
+                    2030
+                ],
+                "score": 0.916
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    295,
+                    1007,
+                    1054,
+                    1007,
+                    1054,
+                    1041,
+                    295,
+                    1041
+                ],
+                "score": 0.908
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    900,
+                    371,
+                    900,
+                    371,
+                    929,
+                    298,
+                    929
+                ],
+                "score": 0.902
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    295,
+                    1172,
+                    1359,
+                    1172,
+                    1359,
+                    1210,
+                    295,
+                    1210
+                ],
+                "score": 0.892
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    295,
+                    390,
+                    1402,
+                    390,
+                    1402,
+                    456,
+                    295,
+                    456
+                ],
+                "score": 0.891
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    367,
+                    933,
+                    1330,
+                    933,
+                    1330,
+                    1000,
+                    367,
+                    1000
+                ],
+                "score": 0.886
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    358,
+                    302,
+                    1347,
+                    302,
+                    1347,
+                    385,
+                    358,
+                    385
+                ],
+                "score": 0.875
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    836,
+                    2087,
+                    865,
+                    2087,
+                    865,
+                    2113,
+                    836,
+                    2113
+                ],
+                "score": 0.859
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    432,
+                    1217,
+                    965,
+                    1217,
+                    965,
+                    1273,
+                    432,
+                    1273
+                ],
+                "score": 0.554
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    334,
+                    1217,
+                    1364,
+                    1217,
+                    1364,
+                    1364,
+                    334,
+                    1364
+                ],
+                "score": 0.32
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    433,
+                    1217,
+                    964,
+                    1217,
+                    964,
+                    1272,
+                    433,
+                    1272
+                ],
+                "score": 0.195
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    442,
+                    1324,
+                    953,
+                    1324,
+                    953,
+                    1363,
+                    442,
+                    1363
+                ],
+                "score": 0.189
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    367,
+                    1774,
+                    1326,
+                    1774,
+                    1326,
+                    2034,
+                    367,
+                    2034
+                ],
+                "score": 0.94,
+                "latex": "\\begin{array} { r l } & { \\widehat { \\mathrm { I P M } } ( d _ { \\phi } ^ { \\pi } , d _ { \\phi } ^ { 2 D } ) = \\displaystyle \\operatorname* { s u p } _ { \\nu \\in \\mathcal { F } } \\Big | \\widehat { f } _ { 1 } ( \\nu ) + \\widehat { f } _ { 2 } ( \\nu ) + \\widehat { f } _ { 3 } ( \\nu ) \\Big | } \\\\ & { \\mathrm { w h e r e } \\widehat { f } _ { 1 } ( \\nu ) = \\displaystyle \\frac { 1 - \\gamma } { n } \\sum _ { i = 1 } ^ { n } \\nu \\left( \\phi \\left( s _ { 0 } ^ { ( i ) } , a _ { 0 } ^ { ( i ) } \\right) \\right) , \\widehat { f } _ { 2 } ( \\nu ) = - \\displaystyle \\frac { 1 } { n } \\sum _ { i = 1 } ^ { n } \\nu \\left( \\phi \\left( s ^ { ( i ) } , a ^ { ( i ) } \\right) \\right) , } \\\\ & { \\widehat { f } _ { 3 } ( \\nu ) = \\displaystyle \\frac { \\gamma } { n } \\sum _ { i = 1 } ^ { n } \\nu \\left( \\phi \\left( s ^ { \\prime ( i ) } , a ^ { \\prime ( i ) } \\right) \\right) . } \\end{array}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1011,
+                    1172,
+                    1179,
+                    1172,
+                    1179,
+                    1213,
+                    1011,
+                    1213
+                ],
+                "score": 0.94,
+                "latex": "\\mathrm { I P M } _ { \\mathcal { G } } ( d _ { \\phi } ^ { \\pi } , d _ { \\phi } ^ { D } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    974,
+                    1127,
+                    1097,
+                    1127,
+                    1097,
+                    1162,
+                    974,
+                    1162
+                ],
+                "score": 0.94,
+                "latex": "\\langle \\nu ^ { * } , \\nu ^ { * } \\rangle _ { \\mathcal { H } _ { k } }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    415,
+                    1092,
+                    778,
+                    1092,
+                    778,
+                    1130,
+                    415,
+                    1130
+                ],
+                "score": 0.93,
+                "latex": "\\begin{array} { r } { \\operatorname* { s u p } _ { \\nu \\in \\mathcal { F } } \\langle \\nu , \\nu ^ { * } \\rangle _ { \\mathcal { H } _ { k } } ^ { 2 } = \\langle \\nu ^ { * } , \\nu ^ { * } \\rangle _ { \\mathcal { H } _ { k } } } \\end{array}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    974,
+                    228,
+                    1143,
+                    228,
+                    1143,
+                    269,
+                    974,
+                    269
+                ],
+                "score": 0.93,
+                "latex": "\\mathrm { I P M } _ { \\mathcal { G } } ( d _ { \\phi } ^ { \\pi } , d _ { \\phi } ^ { D } )"
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    330,
+                    463,
+                    1366,
+                    463,
+                    1366,
+                    901,
+                    330,
+                    901
+                ],
+                "score": 0.92,
+                "latex": "\\begin{array} { r l } & { \\mathrm { I P M } _ { \\mathcal { G } } ^ { 2 } = \\Bigg \\{ \\underset { s \\in \\mathcal { F } } { \\operatorname* { s u p } } \\Bigg \\{ ( 1 - \\gamma ) \\mathbb { E } _ { \\underset { \\sigma \\sim \\infty } { s \\operatorname* { s d } } } \\left[ \\nu ( \\phi s , a ) \\right] \\big ] - \\mathbb { E } _ { ( s , a , s ^ { \\prime } ) \\sim d } \\mathbb { P } \\big [ \\nu ( \\phi ( s , a ) ) - \\gamma \\nu ( \\phi ( s ^ { \\prime } , a ^ { \\prime } ) ) \\big ] \\Bigg \\} ^ { 2 } } \\\\ & { \\qquad = \\underset { s \\in \\mathcal { F } } { \\operatorname* { s u p } } \\Bigg \\{ ( 1 - \\gamma ) \\mathbb { E } _ { \\underset { \\sigma \\sim \\infty } { s \\operatorname* { s u p } } } \\big [ \\nu ( \\phi ( s , a ) ) \\big ] - \\mathbb { E } _ { ( s , a , s ^ { \\prime } ) \\sim d } \\mathbb { P } \\big [ \\nu ( \\phi ( s , a ) ) - \\gamma \\nu ( \\phi ( s ^ { \\prime } , a ^ { \\prime } ) ) \\big ] \\Bigg \\} ^ { 2 } } \\\\ & { \\qquad = \\underset { s \\in \\mathcal { F } } { \\operatorname* { s u p } } \\Bigg \\{ ( 1 - \\gamma ) \\mathbb { E } _ { \\underset { \\sigma \\sim \\infty } { s \\operatorname* { s u p } } } \\big [ ( \\nu , k ( \\cdot , \\phi ( s , a ) ) ) u _ { k } \\big ] } \\\\ & { \\qquad \\quad - \\mathbb { E } _ { ( s , a , s ^ { \\prime } ) \\sim d } \\mathbb { P } \\big [ \\nu , k ( \\cdot , \\cdot , \\phi ( s , a ) ) \\big ] u _ { k } - \\gamma \\big ( \\nu , k ( \\cdot , \\phi ( s ^ { \\prime } , a ^ { \\prime } ) ) \\big ) _ { k k } \\bigg ] \\Bigg \\} ^ { 2 } } \\\\ & { \\qquad - \\mathbb { E } _ { ( s , a , s ^ { \\prime } ) \\sim d } \\rho ^ { D } \\left[ \\big \\langle \\nu , k ( \\cdot , \\phi ( s , a ) ) \\big \\rangle _ { \\mathcal { H } _ { k } } - \\gamma \\big \\langle \\nu , k ( \\cdot , \\phi ( s ^ { \\prime } , a ^ { \\prime } ) ) \\big \\rangle _ { \\mathcal { H } _ { k } } \\right] \\Bigg \\} ^ { 2 } } \\\\ & { \\qquad = \\underset { s \\in \\mathcal { F } } { \\operatorname* { s u p } } \\big \\{ \\nu , \\nu ^ { * } \\} _ { \\mathcal { H } _ { k } } ^ { 2 } , } \\end{array}"
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    313,
+                    1492,
+                    1383,
+                    1492,
+                    1383,
+                    1712,
+                    313,
+                    1712
+                ],
+                "score": 0.92,
+                "latex": "\\begin{array} { l } { { \\displaystyle \\widehat { \\mathrm { I P M } } ( d _ { \\phi } ^ { \\pi } , d _ { \\phi } ^ { \\mathcal { D } } ) ^ { 2 } = \\frac { 1 } { n ^ { 2 } } \\sum _ { i , j } \\bigg [ k ( \\phi ( s ^ { ( i ) } , a ^ { ( i ) } ) , \\phi ( s ^ { ( j ) } , a ^ { ( j ) } ) ) + ( 1 - \\gamma ) ^ { 2 } k ( \\phi ( s _ { 0 } ^ { ( i ) } , a _ { 0 } ^ { ( i ) } ) , \\phi ( s _ { 0 } ^ { ( j ) } , a _ { 0 } ^ { ( j ) } ) ) } } \\\\ { { \\displaystyle \\qquad + \\gamma ^ { 2 } k ( \\phi ( s ^ { \\prime ( i ) } , a ^ { \\prime ( i ) } ) , \\phi ( s ^ { \\prime ( j ) } , a ^ { \\prime ( j ) } ) ) - 2 ( 1 - \\gamma ) k ( \\phi ( s _ { 0 } ^ { ( i ) } , a _ { 0 } ^ { ( i ) } ) , \\phi ( s ^ { ( j ) } , a ^ { ( j ) } ) ) } } \\\\ { { \\displaystyle \\qquad - 2 \\gamma k ( \\phi ( s ^ { \\prime ( i ) } , a ^ { \\prime ( i ) } ) , \\phi ( s ^ { ( j ) } , a ^ { ( j ) } ) ) + 2 \\gamma ( 1 - \\gamma ) k ( \\phi ( s ^ { \\prime ( i ) } , a ^ { \\prime ( i ) } ) , \\phi ( s _ { 0 } ^ { ( j ) } , a _ { 0 } ^ { ( j ) } ) ) \\bigg ] . } } \\end{array}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1064,
+                    1092,
+                    1394,
+                    1092,
+                    1394,
+                    1127,
+                    1064,
+                    1127
+                ],
+                "score": 0.92,
+                "latex": "\\langle k ( \\cdot , z ) , k ( \\cdot , \\bar { z } ) \\rangle _ { \\mathcal { H } _ { k } } = k ( z , \\bar { z } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1204,
+                    1367,
+                    1405,
+                    1367,
+                    1405,
+                    1410,
+                    1204,
+                    1410
+                ],
+                "score": 0.92,
+                "latex": "s _ { 0 } ^ { ( i ) } \\sim d _ { 0 } , a _ { 0 } ^ { ( i ) } \\sim"
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    434,
+                    1214,
+                    1359,
+                    1214,
+                    1359,
+                    1365,
+                    434,
+                    1365
+                ],
+                "score": 0.91,
+                "latex": "\\begin{array} { r l } & { \\mathrm { I P M } _ { \\mathcal { G } } ( d _ { \\phi } ^ { \\pi } , d _ { \\phi } ^ { D } ) = \\underset { \\nu \\in \\mathcal { F } } { \\operatorname* { s u p } } | f _ { 1 } ( \\nu ) + f _ { 2 } ( \\nu ) + f _ { 3 } ( \\nu ) | } \\\\ & { f _ { 1 } ( \\nu ) = ( 1 - \\gamma ) \\mathbb { E } _ { s \\sim d _ { 0 } , a \\sim \\pi ( s ) } [ \\nu ( \\phi ( s , a ) ) ] , \\qquad f _ { 2 } ( \\nu ) = - \\mathbb { E } _ { ( s , a ) \\sim d ^ { D } } \\left[ \\nu ( \\phi ( s , a ) ) \\right] , } \\\\ & { f _ { 3 } ( \\nu ) = \\mathbb { E } _ { ( s , a , s ^ { \\prime } ) \\sim d ^ { D } , a \\cdot \\sim \\pi ( s ^ { \\prime } ) } \\left[ \\gamma \\nu ( \\phi ( s ^ { \\prime } , a ^ { \\prime } ) ) \\right] . } \\end{array}"
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    356,
+                    292,
+                    1339,
+                    292,
+                    1339,
+                    387,
+                    356,
+                    387
+                ],
+                "score": 0.9,
+                "latex": "\\mathrm { I P M } _ { \\mathcal { G } } = \\underset { \\nu \\in \\mathcal { F } } { \\operatorname* { s u p } } \\left| ( 1 - \\gamma ) \\mathbb { E } _ { \\underset { a \\sim \\pi ( s ) } { s \\sim d _ { 0 } } } \\left[ \\nu ( \\phi ( s , a ) ) \\right] - \\mathbb { E } _ { ( s , a , s ^ { \\prime } ) \\sim d ^ { \\mathcal { D } } } \\left[ \\nu ( \\phi ( s , a ) ) - \\gamma \\nu ( \\phi ( s ^ { \\prime } , a ^ { \\prime } ) ) \\right] \\right| ,"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    784,
+                    230,
+                    858,
+                    230,
+                    858,
+                    263,
+                    784,
+                    263
+                ],
+                "score": 0.9,
+                "latex": "\\operatorname { I P M } _ { \\mathcal { G } }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    496,
+                    1369,
+                    879,
+                    1369,
+                    879,
+                    1410,
+                    496,
+                    1410
+                ],
+                "score": 0.9,
+                "latex": "\\{ s _ { 0 } ^ { ( i ) } , a _ { 0 } ^ { ( i ) } , s ^ { ( i ) } , a ^ { ( i ) } , s ^ { \\prime ( i ) } , a ^ { \\prime ( i ) } \\} _ { i = 1 } ^ { n }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    539,
+                    392,
+                    982,
+                    392,
+                    982,
+                    426,
+                    539,
+                    426
+                ],
+                "score": 0.9,
+                "latex": "\\mathcal { F } = \\{ \\nu \\in ( \\mathcal { Z }  \\mathbb { R } ) : \\langle \\nu , \\nu \\rangle _ { \\mathcal { H } _ { k } } \\leq 1 \\}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    569,
+                    425,
+                    608,
+                    425,
+                    608,
+                    454,
+                    569,
+                    454
+                ],
+                "score": 0.89,
+                "latex": "\\mathcal { H } _ { k }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    297,
+                    1449,
+                    451,
+                    1449,
+                    451,
+                    1491,
+                    297,
+                    1491
+                ],
+                "score": 0.89,
+                "latex": "\\widehat { \\mathrm { I P M } } ( d _ { \\phi } ^ { \\pi } , d _ { \\phi } ^ { D } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1231,
+                    393,
+                    1271,
+                    393,
+                    1271,
+                    423,
+                    1231,
+                    423
+                ],
+                "score": 0.88,
+                "latex": "\\mathcal { H } _ { k }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    293,
+                    1411,
+                    932,
+                    1411,
+                    932,
+                    1449,
+                    293,
+                    1449
+                ],
+                "score": 0.87,
+                "latex": "\\pi ( s _ { 0 } ^ { ( i ) } ) , ( s ^ { ( i ) } , a ^ { ( i ) } , s ^ { \\prime ( i ) } ) ~ \\sim ~ d ^ { D } , a ^ { \\prime ( i ) } ~ \\sim ~ \\pi ( s ^ { \\prime ( i ) } ) ~ \\forall i ,"
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    366,
+                    934,
+                    1330,
+                    934,
+                    1330,
+                    1002,
+                    366,
+                    1002
+                ],
+                "score": 0.87,
+                "latex": "\\begin{array} { r } { \\nu ^ { * } ( \\cdot ) = ( 1 - \\gamma ) \\mathbb { E } _ { \\stackrel { s \\sim d _ { 0 } } { a \\sim \\pi ( s ) } } \\left[ k ( \\cdot , \\phi ( s , a ) ) \\right] - \\mathbb { E } _ { ( s , a , s ^ { \\prime } ) \\sim d ^ { \\mathcal { D } } } \\left[ k ( \\cdot , \\phi ( s , a ) ) - \\gamma k ( \\cdot , \\phi ( s ^ { \\prime } , a ^ { \\prime } ) ) \\right] . } \\end{array}"
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    584,
+                    1044,
+                    1111,
+                    1044,
+                    1111,
+                    1087,
+                    584,
+                    1087
+                ],
+                "score": 0.87,
+                "latex": "\\langle \\nu , \\nu ^ { * } \\rangle _ { \\mathcal { H } _ { k } } ^ { 2 } \\le \\langle \\nu , \\nu \\rangle _ { \\mathcal { H } _ { k } } \\langle \\nu ^ { * } , \\nu ^ { * } \\rangle _ { \\mathcal { H } _ { k } } \\le \\langle \\nu ^ { * } , \\nu ^ { * } \\rangle _ { \\mathcal { H } _ { k } }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    346,
+                    394,
+                    373,
+                    394,
+                    373,
+                    421,
+                    346,
+                    421
+                ],
+                "score": 0.84,
+                "latex": "\\mathcal { F }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    781,
+                    1007,
+                    1043,
+                    1007,
+                    1043,
+                    1041,
+                    781,
+                    1041
+                ],
+                "score": 0.84,
+                "latex": "\\langle \\nu , \\nu \\rangle _ { \\mathcal { H } _ { k } } \\leq 1 , \\forall \\nu \\in \\mathcal { F }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    372,
+                    1383,
+                    393,
+                    1383,
+                    393,
+                    1403,
+                    372,
+                    1403
+                ],
+                "score": 0.75,
+                "latex": "n"
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    434,
+                    1215,
+                    969,
+                    1215,
+                    969,
+                    1275,
+                    434,
+                    1275
+                ],
+                "score": 0.71,
+                "latex": "\\mathrm { I P M } _ { \\mathcal { G } } ( d _ { \\phi } ^ { \\pi } , d _ { \\phi } ^ { D } ) = \\operatorname* { s u p } _ { \\nu \\in \\mathcal { F } } | f _ { 1 } ( \\nu ) + f _ { 2 } ( \\nu ) + f _ { 3 } ( \\nu ) |"
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    441,
+                    1322,
+                    960,
+                    1322,
+                    960,
+                    1362,
+                    441,
+                    1362
+                ],
+                "score": 0.53,
+                "latex": "f _ { 3 } ( \\nu ) = \\mathbb { E } _ { ( s , a , s ^ { \\prime } ) \\sim d ^ { \\mathcal { D } } , a ^ { \\prime } \\sim \\pi ( s ^ { \\prime } ) } \\left[ \\gamma \\nu ( \\phi ( s ^ { \\prime } , a ^ { \\prime } ) ) \\right] ."
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    899,
+                    1414,
+                    929,
+                    1414,
+                    929,
+                    1444,
+                    899,
+                    1444
+                ],
+                "score": 0.29,
+                "latex": "\\forall i"
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    449,
+                    1281,
+                    1007,
+                    1281,
+                    1007,
+                    1317,
+                    449,
+                    1317
+                ],
+                "score": 0.29,
+                "latex": "f _ { 1 } ( \\nu ) = ( 1 - \\gamma ) \\mathbb { E } _ { s \\sim d _ { 0 } , a \\sim \\pi ( s ) } [ \\nu ( \\phi ( s , a ) ) ] , \\qquad f _ { 2 } ("
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    72.0,
+                    815.0,
+                    72.0,
+                    815.0,
+                    108.0,
+                    296.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    832.0,
+                    2085.0,
+                    870.0,
+                    2085.0,
+                    870.0,
+                    2124.0,
+                    832.0,
+                    2124.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    283.0,
+                    1348.0,
+                    371.0,
+                    1348.0,
+                    371.0,
+                    1431.0,
+                    283.0,
+                    1431.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    394.0,
+                    1348.0,
+                    495.0,
+                    1348.0,
+                    495.0,
+                    1431.0,
+                    394.0,
+                    1431.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    880.0,
+                    1348.0,
+                    1203.0,
+                    1348.0,
+                    1203.0,
+                    1431.0,
+                    880.0,
+                    1431.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    933.0,
+                    1400.0,
+                    1412.0,
+                    1400.0,
+                    1412.0,
+                    1457.0,
+                    933.0,
+                    1457.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1442.0,
+                    296.0,
+                    1442.0,
+                    296.0,
+                    1496.0,
+                    293.0,
+                    1496.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    452.0,
+                    1442.0,
+                    468.0,
+                    1442.0,
+                    468.0,
+                    1496.0,
+                    452.0,
+                    1496.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1241.0,
+                    1365.0,
+                    1364.0,
+                    1365.0,
+                    1364.0,
+                    1416.5,
+                    1241.0,
+                    1416.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    223.0,
+                    783.0,
+                    223.0,
+                    783.0,
+                    268.0,
+                    293.0,
+                    268.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    859.0,
+                    223.0,
+                    973.0,
+                    223.0,
+                    973.0,
+                    268.0,
+                    859.0,
+                    268.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1144.0,
+                    223.0,
+                    1404.0,
+                    223.0,
+                    1404.0,
+                    268.0,
+                    1144.0,
+                    268.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    264.0,
+                    574.0,
+                    264.0,
+                    574.0,
+                    297.0,
+                    295.0,
+                    297.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1087.0,
+                    414.0,
+                    1087.0,
+                    414.0,
+                    1136.0,
+                    292.0,
+                    1136.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    779.0,
+                    1087.0,
+                    1063.0,
+                    1087.0,
+                    1063.0,
+                    1136.0,
+                    779.0,
+                    1136.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1395.0,
+                    1087.0,
+                    1399.0,
+                    1087.0,
+                    1399.0,
+                    1136.0,
+                    1395.0,
+                    1136.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1118.0,
+                    973.0,
+                    1118.0,
+                    973.0,
+                    1170.0,
+                    293.0,
+                    1170.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1098.0,
+                    1118.0,
+                    1111.0,
+                    1118.0,
+                    1111.0,
+                    1170.0,
+                    1098.0,
+                    1170.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1711.0,
+                    1403.0,
+                    1711.0,
+                    1403.0,
+                    1746.0,
+                    294.0,
+                    1746.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1740.0,
+                    567.0,
+                    1740.0,
+                    567.0,
+                    1776.0,
+                    294.0,
+                    1776.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1003.0,
+                    780.0,
+                    1003.0,
+                    780.0,
+                    1045.0,
+                    294.0,
+                    1045.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1044.0,
+                    1003.0,
+                    1053.0,
+                    1003.0,
+                    1053.0,
+                    1045.0,
+                    1044.0,
+                    1045.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    896.0,
+                    376.0,
+                    896.0,
+                    376.0,
+                    934.0,
+                    293.0,
+                    934.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1172.0,
+                    1010.0,
+                    1172.0,
+                    1010.0,
+                    1213.0,
+                    294.0,
+                    1213.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1180.0,
+                    1172.0,
+                    1365.0,
+                    1172.0,
+                    1365.0,
+                    1213.0,
+                    1180.0,
+                    1213.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    387.0,
+                    345.0,
+                    387.0,
+                    345.0,
+                    431.0,
+                    292.0,
+                    431.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    374.0,
+                    387.0,
+                    538.0,
+                    387.0,
+                    538.0,
+                    431.0,
+                    374.0,
+                    431.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    983.0,
+                    387.0,
+                    1230.0,
+                    387.0,
+                    1230.0,
+                    431.0,
+                    983.0,
+                    431.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1272.0,
+                    387.0,
+                    1407.0,
+                    387.0,
+                    1407.0,
+                    431.0,
+                    1272.0,
+                    431.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    423.0,
+                    568.0,
+                    423.0,
+                    568.0,
+                    459.0,
+                    294.0,
+                    459.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    609.0,
+                    423.0,
+                    621.0,
+                    423.0,
+                    621.0,
+                    459.0,
+                    609.0,
+                    459.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 14,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    961,
+                    1406,
+                    961,
+                    1406,
+                    1079,
+                    298,
+                    1079
+                ],
+                "score": 0.969
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    340,
+                    1216,
+                    1351,
+                    1216,
+                    1351,
+                    1655,
+                    340,
+                    1655
+                ],
+                "score": 0.968
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    500,
+                    1851,
+                    1196,
+                    1851,
+                    1196,
+                    2045,
+                    500,
+                    2045
+                ],
+                "score": 0.965
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    351,
+                    677,
+                    1348,
+                    677,
+                    1348,
+                    947,
+                    351,
+                    947
+                ],
+                "score": 0.963
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    295,
+                    273,
+                    1471,
+                    273,
+                    1471,
+                    632,
+                    295,
+                    632
+                ],
+                "score": 0.963
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    428,
+                    1706,
+                    1265,
+                    1706,
+                    1265,
+                    1797,
+                    428,
+                    1797
+                ],
+                "score": 0.948
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1807,
+                    911,
+                    1807,
+                    911,
+                    1842,
+                    297,
+                    1842
+                ],
+                "score": 0.926
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    640,
+                    536,
+                    640,
+                    536,
+                    672,
+                    297,
+                    672
+                ],
+                "score": 0.926
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    294,
+                    228,
+                    1144,
+                    228,
+                    1144,
+                    263,
+                    294,
+                    263
+                ],
+                "score": 0.923
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1663,
+                    1148,
+                    1663,
+                    1148,
+                    1698,
+                    297,
+                    1698
+                ],
+                "score": 0.922
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    298,
+                    73,
+                    814,
+                    73,
+                    814,
+                    106,
+                    298,
+                    106
+                ],
+                "score": 0.92
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    325,
+                    1088,
+                    1368,
+                    1088,
+                    1368,
+                    1167,
+                    325,
+                    1167
+                ],
+                "score": 0.897
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1176,
+                    361,
+                    1176,
+                    361,
+                    1206,
+                    298,
+                    1206
+                ],
+                "score": 0.846
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    835,
+                    2087,
+                    866,
+                    2087,
+                    866,
+                    2114,
+                    835,
+                    2114
+                ],
+                "score": 0.784
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    835,
+                    2087,
+                    866,
+                    2087,
+                    866,
+                    2114,
+                    835,
+                    2114
+                ],
+                "score": 0.16
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    350,
+                    678,
+                    1346,
+                    678,
+                    1346,
+                    952,
+                    350,
+                    952
+                ],
+                "score": 0.95,
+                "latex": "\\begin{array} { l } { \\displaystyle \\underset { \\nu \\in \\mathcal { F } } { \\operatorname* { s u p } } \\left| f _ { 1 } ( \\nu ) - \\widehat { f _ { 1 } } ( \\nu ) \\right| = ( 1 - \\gamma ) \\Biggl \\{ \\mathbb { E } _ { s \\sim d _ { 0 } , a \\sim \\pi ( s ) } [ k ( \\phi ( s , a ) , \\phi ( \\bar { s } , \\bar { a } ) ) ] } \\\\ { \\displaystyle - \\frac { 2 } { n } \\sum _ { i = 1 } ^ { n } \\mathbb { E } _ { \\bar { s } \\sim d _ { 0 } , \\bar { a } \\sim \\pi ( s ) } \\left[ k \\left( \\phi \\left( s ^ { ( i ) } , a ^ { ( i ) } \\right) , \\phi ( \\bar { s } , \\bar { a } ) \\right) \\right] } \\\\ { \\displaystyle + \\frac { 1 } { n ^ { 2 } } \\sum _ { i = 1 } ^ { n } \\sum _ { j = 1 } ^ { n } k \\left( \\phi \\left( s ^ { ( i ) } , a ^ { ( i ) } \\right) , \\phi \\left( s ^ { ( j ) } , a ^ { ( j ) } \\right) \\right) \\Biggl \\} ^ { 1 / 2 } , } \\end{array}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    968,
+                    960,
+                    1242,
+                    960,
+                    1242,
+                    1017,
+                    968,
+                    1017
+                ],
+                "score": 0.94,
+                "latex": "{ \\mathrm { s u p } } _ { \\nu \\in { \\mathcal { F } } } \\left| f _ { 1 } ( \\nu ) - { \\widehat { f } } _ { 1 } ( \\nu ) \\right|"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    618,
+                    967,
+                    712,
+                    967,
+                    712,
+                    1003,
+                    618,
+                    1003
+                ],
+                "score": 0.94,
+                "latex": "s ^ { ( i ) } , a ^ { ( i ) }"
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    346,
+                    1215,
+                    1354,
+                    1215,
+                    1354,
+                    1660,
+                    346,
+                    1660
+                ],
+                "score": 0.94,
+                "latex": "\\begin{array} { r l } & { \\mathbb { E } _ { s ^ { ( i ) } , \\sigma ^ { ( k ) } } [ \\underset { s ^ { ( r , F ) } } { \\operatorname* { s u p } } [ f _ { 1 } ( \\nu ) - \\widehat { f } _ { 1 } ( \\nu ) ] ] } \\\\ & { = \\frac { 1 - \\gamma } { n } \\underset { s ^ { ( i ) } , a ^ { ( i ) } , b ^ { ( i ) } } { \\operatorname* { s u p } } [ \\underset { s ^ { ( r , F ) } } { \\operatorname* { s u p } } [ \\underset { s ^ { ( i ) } , a ^ { ( i ) } , b ^ { ( i ) } } { \\operatorname* { s u p } } [ \\underset { i = 1 } { \\overset { n } { \\sum } } \\nu ( \\phi ( \\tilde { s } _ { 0 } ^ { ( i ) } , \\bar { a } _ { 0 } ^ { ( i ) } ) ) ] - \\underset { i = 1 } { \\overset { n } { \\sum } } \\nu ( \\phi ( s _ { 0 } ^ { ( i ) } , a _ { 0 } ^ { ( i ) } ) ) ) ] } \\\\ & { \\leq \\frac { 1 - \\gamma } { n } \\underset { s ^ { ( i ) } , a ^ { ( i ) } , b ^ { ( i ) } , a ^ { ( i ) } , b ^ { ( i ) } } { \\operatorname* { s u p } } [ \\underset { s ^ { ( r , F ) } } { \\operatorname* { s u p } } [ \\underset { i = 1 } { \\overset { n } { \\sum } } \\nu ( \\phi ( \\frac { \\sigma ^ { ( i ) } } { s } , \\bar { a } _ { 0 } ^ { ( i ) } ) ) - \\underset { i = 1 } { \\overset { n } { \\sum } } \\nu ( \\phi ( s _ { 0 } ^ { ( i ) } , a _ { 0 } ^ { ( i ) } ) ) ) ] } \\\\ & { = \\frac { 1 - \\gamma } { n } \\underset { s ^ { ( i ) } , a ^ { ( i ) } , b ^ { ( i ) } , a ^ { ( i ) } , b ^ { ( i ) } } { \\operatorname* { s u p } } [ \\underset { s ^ { ( r , F ) } } { \\operatorname* { s u p } } [ \\underset { i = 1 } { \\overset { n } { \\sum } } \\sigma ^ { ( i ) 1 } \\{ \\nu ( \\phi ( \\tilde { s } _ { 0 } ^ { ( i ) } , \\bar { a } _ { 0 } ^ { ( i ) } ) ) - \\nu ( \\phi ( s _ { 0 } ^ { ( i ) } , a _ { 0 } ^ { ( i ) } ) ) \\} ] } \\\\ &  \\leq 2 ( 1 - \\gamma )  \\end{array}"
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    430,
+                    1705,
+                    1268,
+                    1705,
+                    1268,
+                    1800,
+                    430,
+                    1800
+                ],
+                "score": 0.92,
+                "latex": "\\operatorname* { P r } \\left\\{ \\operatorname* { s u p } _ { \\nu \\in \\mathcal { F } } \\left| f _ { 1 } ( \\nu ) - \\widehat { f } _ { 1 } ( \\nu ) \\right| - 2 ( 1 - \\gamma ) \\sqrt { \\frac { \\bar { k } } { n } } \\geq \\epsilon \\right\\} \\leq \\exp \\left( - \\frac { n \\epsilon ^ { 2 } } { 2 ( 1 - \\gamma ) ^ { 2 } \\bar { k } } \\right) ."
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    428,
+                    1012,
+                    615,
+                    1012,
+                    615,
+                    1050,
+                    428,
+                    1050
+                ],
+                "score": 0.92,
+                "latex": "2 ( 1 - \\gamma ) \\bar { k } ^ { 1 / 2 } / n"
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    310,
+                    271,
+                    1427,
+                    271,
+                    1427,
+                    636,
+                    310,
+                    636
+                ],
+                "score": 0.92,
+                "latex": "\\begin{array} { r l } { \\mathrm { I P M } _ { \\mathcal { G } } ( d _ { \\phi } ^ { \\pi } , d _ { \\phi } ^ { 2 } ) - \\widehat { \\mathrm { I P M } } ( d _ { \\phi } ^ { \\pi } , d _ { \\phi } ^ { \\mathcal { D } } ) \\Bigg | = \\Bigg | \\underset { \\nu \\in \\mathcal { F } } { \\operatorname* { s u p } } | f _ { 1 } ( \\nu ) + f _ { 2 } ( \\nu ) + f _ { 3 } ( \\nu ) | - \\underset { \\nu \\in \\mathcal { F } } { \\operatorname* { s u p } } | \\widehat { f } _ { 1 } ( \\nu ) + \\widehat { f } _ { 2 } ( \\nu ) + \\widehat { f } _ { 3 } ( \\nu ) | \\Bigg | } & { } \\\\ { \\leq \\underset { \\nu \\in \\mathcal { F } } { \\operatorname* { s u p } } | | f _ { 1 } ( \\nu ) + f _ { 2 } ( \\nu ) + f _ { 3 } ( \\nu ) | - | \\widehat { f } _ { 1 } ( \\nu ) + \\widehat { f } _ { 2 } ( \\nu ) + \\widehat { f } _ { 3 } ( \\nu ) | | } & { } \\\\ { \\leq \\underset { \\nu \\in \\mathcal { F } } { \\operatorname* { s u p } } | f _ { 1 } ( \\nu ) + f _ { 2 } ( \\nu ) + f _ { 3 } ( \\nu ) - \\widehat { f } _ { 1 } ( \\nu ) - \\widehat { f } _ { 2 } ( \\nu ) - \\widehat { f } _ { 3 } ( \\nu ) | } & { } \\\\ { \\leq \\underset { \\nu \\in \\mathcal { F } } { \\operatorname* { s u p } } [ | f _ { 1 } ( \\nu ) - \\widehat { f } _ { 1 } ( \\nu ) | + | f _ { 2 } ( \\nu ) - \\widehat { f } _ { 2 } ( \\nu ) | + | f _ { 3 } ( \\nu ) - \\widehat { f } _ { 3 } ( \\nu ) | ] } & { } \\\\ { \\leq \\underset { \\nu \\in \\mathcal { F } } { \\operatorname* { s u p } } [ | f _ { 1 } ( \\nu ) - \\widehat { f } _ { 1 } ( \\nu ) | + | \\widehat { f } _ { 2 } ( \\nu ) - \\widehat { f } _ { 2 } ( \\nu ) | + | f _ { 3 } ( \\nu ) - \\widehat { f } _ { 3 } ( \\nu ) | ] } & { } \\\\  \\leq \\underset { \\nu \\in \\mathcal { F } } { \\operatorname* { s u p } } | f _ { 1 } ( \\nu ) - \\widehat { f } _ { 1 } ( \\nu ) | + \\underset { \\nu \\in \\mathcal { F } } { \\operatorname* { s u p } } | f _   \\end{array}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    701,
+                    1013,
+                    938,
+                    1013,
+                    938,
+                    1051,
+                    701,
+                    1051
+                ],
+                "score": 0.92,
+                "latex": "{ \\bar { k } } = \\operatorname* { s u p } _ { z \\in { \\mathcal { Z } } } k ( z , z )"
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    327,
+                    1087,
+                    1371,
+                    1087,
+                    1371,
+                    1166,
+                    327,
+                    1166
+                ],
+                "score": 0.91,
+                "latex": "\\operatorname* { P r } \\left\\{ \\operatorname* { s u p } _ { \\nu \\in \\mathcal { F } } \\left| f _ { 1 } ( \\nu ) - \\widehat f _ { 1 } ( \\nu ) \\right| - \\mathbb { E } _ { s ^ { ( i ) } , a ^ { ( i ) } } \\left[ \\operatorname* { s u p } _ { \\nu \\in \\mathcal { F } } \\left| f _ { 1 } ( \\nu ) - \\widehat f _ { 1 } ( \\nu ) \\right| \\right] \\geq \\epsilon \\right\\} \\leq \\exp \\left( - \\frac { n \\epsilon ^ { 2 } } { 2 ( 1 - \\gamma ) ^ { 2 } \\bar { k } } \\right) ."
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    654,
+                    1810,
+                    681,
+                    1810,
+                    681,
+                    1841,
+                    654,
+                    1841
+                ],
+                "score": 0.89,
+                "latex": "f _ { 2 }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    733,
+                    1810,
+                    761,
+                    1810,
+                    761,
+                    1841,
+                    733,
+                    1841
+                ],
+                "score": 0.88,
+                "latex": "f _ { 3 }"
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    498,
+                    1850,
+                    1202,
+                    1850,
+                    1202,
+                    2046,
+                    498,
+                    2046
+                ],
+                "score": 0.78,
+                "latex": "\\begin{array} { l l } { { \\displaystyle \\operatorname* { P r } \\left\\{ \\operatorname* { s u p } _ { \\nu \\in \\mathcal { F } } \\Big | f _ { 2 } ( \\nu ) - \\widehat { f } _ { 2 } ( \\nu ) \\Big | - 2 \\sqrt \\frac { \\bar { k } } { n } \\geq \\epsilon \\right\\} \\leq \\exp \\left( - \\frac { n \\epsilon ^ { 2 } } { 2 \\bar { k } } \\right) , } } & { { } } \\\\ { { \\displaystyle \\operatorname* { P r } \\left\\{ \\operatorname* { s u p } _ { \\nu \\in \\mathcal { F } } \\Big | f _ { 3 } ( \\nu ) - \\widehat { f } _ { 3 } ( \\nu ) \\Big | - 2 \\gamma \\sqrt \\frac { \\bar { k } } { n } \\geq \\epsilon \\right\\} \\leq \\exp \\left( - \\frac { n \\epsilon ^ { 2 } } { 2 \\gamma ^ { 2 } \\bar { k } } \\right) . } } \\end{array}"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    72.0,
+                    815.0,
+                    72.0,
+                    815.0,
+                    108.0,
+                    296.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    832.0,
+                    2085.0,
+                    871.0,
+                    2085.0,
+                    871.0,
+                    2123.0,
+                    832.0,
+                    2123.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    832.0,
+                    2085.0,
+                    871.0,
+                    2085.0,
+                    871.0,
+                    2123.0,
+                    832.0,
+                    2123.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    960.0,
+                    617.0,
+                    960.0,
+                    617.0,
+                    1012.0,
+                    292.0,
+                    1012.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    713.0,
+                    960.0,
+                    967.0,
+                    960.0,
+                    967.0,
+                    1012.0,
+                    713.0,
+                    1012.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1243.0,
+                    960.0,
+                    1406.0,
+                    960.0,
+                    1406.0,
+                    1012.0,
+                    1243.0,
+                    1012.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1006.0,
+                    427.0,
+                    1006.0,
+                    427.0,
+                    1057.0,
+                    291.0,
+                    1057.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    616.0,
+                    1006.0,
+                    700.0,
+                    1006.0,
+                    700.0,
+                    1057.0,
+                    616.0,
+                    1057.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    939.0,
+                    1006.0,
+                    1403.0,
+                    1006.0,
+                    1403.0,
+                    1057.0,
+                    939.0,
+                    1057.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1040.0,
+                    478.0,
+                    1040.0,
+                    478.0,
+                    1083.0,
+                    294.0,
+                    1083.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1801.0,
+                    653.0,
+                    1801.0,
+                    653.0,
+                    1849.0,
+                    294.0,
+                    1849.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    682.0,
+                    1801.0,
+                    732.0,
+                    1801.0,
+                    732.0,
+                    1849.0,
+                    682.0,
+                    1849.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    762.0,
+                    1801.0,
+                    912.0,
+                    1801.0,
+                    912.0,
+                    1849.0,
+                    762.0,
+                    1849.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    636.0,
+                    537.0,
+                    636.0,
+                    537.0,
+                    675.0,
+                    295.0,
+                    675.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    228.0,
+                    1145.0,
+                    228.0,
+                    1145.0,
+                    266.0,
+                    296.0,
+                    266.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1657.0,
+                    1149.0,
+                    1657.0,
+                    1149.0,
+                    1706.0,
+                    294.0,
+                    1706.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1171.0,
+                    368.0,
+                    1171.0,
+                    368.0,
+                    1215.0,
+                    293.0,
+                    1215.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 15,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    294,
+                    717,
+                    1406,
+                    717,
+                    1406,
+                    951,
+                    294,
+                    951
+                ],
+                "score": 0.975
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    453,
+                    1143,
+                    1242,
+                    1143,
+                    1242,
+                    1234,
+                    453,
+                    1234
+                ],
+                "score": 0.948
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    297,
+                    963,
+                    1403,
+                    963,
+                    1403,
+                    1059,
+                    297,
+                    1059
+                ],
+                "score": 0.944
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    509,
+                    293,
+                    1187,
+                    293,
+                    1187,
+                    378,
+                    509,
+                    378
+                ],
+                "score": 0.94
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    291,
+                    227,
+                    1403,
+                    227,
+                    1403,
+                    290,
+                    291,
+                    290
+                ],
+                "score": 0.939
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    294,
+                    1087,
+                    1331,
+                    1087,
+                    1331,
+                    1125,
+                    294,
+                    1125
+                ],
+                "score": 0.926
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    298,
+                    73,
+                    814,
+                    73,
+                    814,
+                    106,
+                    298,
+                    106
+                ],
+                "score": 0.922
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    291,
+                    1248,
+                    1350,
+                    1248,
+                    1350,
+                    1285,
+                    291,
+                    1285
+                ],
+                "score": 0.922
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    450,
+                    685,
+                    450,
+                    685,
+                    484,
+                    298,
+                    484
+                ],
+                "score": 0.92
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    676,
+                    780,
+                    676,
+                    780,
+                    710,
+                    299,
+                    710
+                ],
+                "score": 0.913
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    598,
+                    1399,
+                    598,
+                    1399,
+                    662,
+                    299,
+                    662
+                ],
+                "score": 0.88
+            },
+            {
+                "category_id": 9,
+                "poly": [
+                    1351,
+                    318,
+                    1401,
+                    318,
+                    1401,
+                    350,
+                    1351,
+                    350
+                ],
+                "score": 0.873
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    836,
+                    2087,
+                    865,
+                    2087,
+                    865,
+                    2112,
+                    836,
+                    2112
+                ],
+                "score": 0.848
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    1373,
+                    385,
+                    1404,
+                    385,
+                    1404,
+                    415,
+                    1373,
+                    415
+                ],
+                "score": 0.818
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    1375,
+                    1251,
+                    1403,
+                    1251,
+                    1403,
+                    1280,
+                    1375,
+                    1280
+                ],
+                "score": 0.613
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    300,
+                    496,
+                    1397,
+                    496,
+                    1397,
+                    583,
+                    300,
+                    583
+                ],
+                "score": 0.396
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    507,
+                    286,
+                    1189,
+                    286,
+                    1189,
+                    380,
+                    507,
+                    380
+                ],
+                "score": 0.93,
+                "latex": "\\left| \\mathrm { I P M } _ { \\mathcal { G } } ( d _ { \\phi } ^ { \\pi } , d _ { \\phi } ^ { D } ) - \\widehat { \\mathrm { I P M } } ( d _ { \\phi } ^ { \\pi } , d _ { \\phi } ^ { D } ) \\right| \\leq \\sqrt { \\frac { \\bar { k } } { n } } \\left( 4 + \\sqrt { 8 \\log \\frac { 3 } { \\delta } } \\right) ."
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1020,
+                    837,
+                    1254,
+                    837,
+                    1254,
+                    878,
+                    1020,
+                    878
+                ],
+                "score": 0.93,
+                "latex": "B _ { \\phi } = \\| f _ { \\phi , \\widehat { R } _ { z } , \\widehat { T } _ { z } } \\| _ { \\mathcal { H } _ { k } }"
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    455,
+                    1137,
+                    1243,
+                    1137,
+                    1243,
+                    1235,
+                    455,
+                    1235
+                ],
+                "score": 0.93,
+                "latex": "\\mathbb { E } _ { ( s , a ) \\sim d ^ { \\mathcal { D } } } \\left[ \\mathcal { E } _ { \\phi , \\widehat { R } _ { z } , \\widehat { T } _ { z } } ( s , a ) \\right] \\leq \\frac { 1 } { n } \\sum _ { ( s , a ) \\sim \\mathcal { D } } \\mathcal { E } _ { \\phi , \\widehat { R } _ { z } , \\widehat { T } _ { z } } ( s , a ) + \\sqrt { \\frac { \\bar { \\mathcal { E } } ^ { 2 } } { 2 n } \\log \\frac { 1 } { \\delta } } ."
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    787,
+                    599,
+                    1289,
+                    599,
+                    1289,
+                    636,
+                    787,
+                    636
+                ],
+                "score": 0.93,
+                "latex": "\\mathbb { E } _ { s ^ { \\prime } \\sim T ( \\phi ^ { - 1 } ( \\cdot ) ) , a ^ { \\prime } \\sim \\pi ( s ^ { \\prime } ) } [ \\nu ( \\phi ( s ^ { \\prime } , a ^ { \\prime } ) ) ] : \\mathcal { Z } \\to \\mathbb { R }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    483,
+                    754,
+                    858,
+                    754,
+                    858,
+                    793,
+                    483,
+                    793
+                ],
+                "score": 0.91,
+                "latex": "( \\widehat { T } , \\widehat { R } ) = \\langle \\widehat { T } _ { z } \\circ \\phi , \\widehat { R } _ { z } \\circ \\phi \\rangle"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    460,
+                    883,
+                    814,
+                    883,
+                    814,
+                    922,
+                    460,
+                    922
+                ],
+                "score": 0.91,
+                "latex": "\\begin{array} { r } { \\bar { \\mathcal { E } } = \\operatorname* { s u p } _ { s \\in { \\mathcal { S } } , a \\in \\mathcal { A } } \\mathcal { E } _ { \\phi , \\widehat { R } _ { z } , \\widehat { T } _ { z } } ( s , a ) } \\end{array}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    294,
+                    827,
+                    956,
+                    827,
+                    956,
+                    882,
+                    294,
+                    882
+                ],
+                "score": 0.9,
+                "latex": "\\begin{array} { r } { \\mathbb { E } _ { T , \\pi } \\Big [ \\sum _ { t = 0 } ^ { \\infty } \\gamma ^ { t } \\mathcal { E } _ { \\phi , \\widehat { R } _ { z } , \\widehat { T } _ { z } } \\big ( s _ { t } , a _ { t } \\big ) \\Big | \\big ( s _ { 0 } , a _ { 0 } \\big ) = \\phi ^ { - 1 } \\big ( z \\big ) \\Big ] \\in \\mathcal { H } _ { k } } \\end{array}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    799,
+                    790,
+                    1050,
+                    790,
+                    1050,
+                    825,
+                    799,
+                    825
+                ],
+                "score": 0.9,
+                "latex": "\\mathrm { s u p } _ { z \\in \\mathcal { Z } } k ( z , z ) ~ = ~ \\bar { k }"
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    310,
+                    960,
+                    1406,
+                    960,
+                    1406,
+                    1059,
+                    310,
+                    1059
+                ],
+                "score": 0.89,
+                "latex": "R ^ { \\pi } - { \\widehat R } ^ { \\pi } \\Big | \\leq \\frac { 1 } { n } \\sum _ { ( s , a ) \\in { \\mathcal D } } { \\mathcal E } _ { \\phi , { \\widehat R } _ { z } , { \\widehat T } _ { z } } ( s , a ) + B _ { \\phi } \\widehat \\Pi \\widehat M \\big ( d _ { \\phi } ^ { \\pi } , d _ { \\phi } ^ { { D } } \\big ) + \\sqrt { \\frac { \\bar { \\mathcal E } ^ { 2 } } { 2 n } \\log { \\frac { 1 } { \\delta } } } + B _ { \\phi } \\sqrt { \\frac { \\bar { k } } { n } } \\left( 4 + \\sqrt { 8 \\log { \\frac { 3 } { \\delta } } } \\right) ."
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1225,
+                    791,
+                    1402,
+                    791,
+                    1402,
+                    831,
+                    1225,
+                    831
+                ],
+                "score": 0.89,
+                "latex": "f _ { \\phi , \\widehat { R } _ { z } , \\widehat { T } _ { z } } \\left( \\widehat { z } \\right) \\ ="
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    310,
+                    492,
+                    1403,
+                    492,
+                    1403,
+                    585,
+                    310,
+                    585
+                ],
+                "score": 0.89,
+                "latex": "\\mathcal { J } = \\left\\{ g \\in ( \\mathcal { Z } \\to \\mathbb { R } ) : g ( z ) = \\nu ( z ) - \\gamma \\mathbb { E } _ { s ^ { \\prime } \\sim T ( \\phi ^ { - 1 } ( z ) ) } [ \\nu ( \\phi ( s ^ { \\prime } , a ^ { \\prime } ) ) ] , \\nu \\in ( \\mathcal { Z } \\to \\mathbb { R } ) , \\langle \\nu , \\nu \\rangle _ { \\mathcal { H } _ { k } } \\leq 1 \\right\\}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    813,
+                    633,
+                    852,
+                    633,
+                    852,
+                    661,
+                    813,
+                    661
+                ],
+                "score": 0.88,
+                "latex": "\\mathcal { H } _ { k }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    727,
+                    230,
+                    771,
+                    230,
+                    771,
+                    264,
+                    727,
+                    264
+                ],
+                "score": 0.87,
+                "latex": "\\delta / 3"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    585,
+                    453,
+                    611,
+                    453,
+                    611,
+                    479,
+                    585,
+                    479
+                ],
+                "score": 0.85,
+                "latex": "\\mathcal { F }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    905,
+                    717,
+                    941,
+                    717,
+                    941,
+                    753,
+                    905,
+                    753
+                ],
+                "score": 0.85,
+                "latex": "\\widehat { M }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1170,
+                    1092,
+                    1233,
+                    1092,
+                    1233,
+                    1120,
+                    1170,
+                    1120
+                ],
+                "score": 0.84,
+                "latex": "1 - \\delta"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    661,
+                    453,
+                    682,
+                    453,
+                    682,
+                    481,
+                    661,
+                    481
+                ],
+                "score": 0.82,
+                "latex": "\\mathcal { G }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    456,
+                    630,
+                    496,
+                    630,
+                    496,
+                    661,
+                    456,
+                    661
+                ],
+                "score": 0.79,
+                "latex": "\\mathcal { H } _ { k }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1178,
+                    883,
+                    1202,
+                    883,
+                    1202,
+                    909,
+                    1178,
+                    909
+                ],
+                "score": 0.78,
+                "latex": "\\mathcal { D }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    358,
+                    761,
+                    379,
+                    761,
+                    379,
+                    793,
+                    358,
+                    793
+                ],
+                "score": 0.76,
+                "latex": "\\phi"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    646,
+                    793,
+                    667,
+                    793,
+                    667,
+                    819,
+                    646,
+                    819
+                ],
+                "score": 0.73,
+                "latex": "k"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    507,
+                    631,
+                    529,
+                    631,
+                    529,
+                    660,
+                    507,
+                    660
+                ],
+                "score": 0.72,
+                "latex": "\\mathcal { G }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    693,
+                    725,
+                    725,
+                    725,
+                    725,
+                    752,
+                    693,
+                    752
+                ],
+                "score": 0.66,
+                "latex": "M"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    297,
+                    916,
+                    375,
+                    916,
+                    375,
+                    947,
+                    297,
+                    947
+                ],
+                "score": 0.64,
+                "latex": "1 - 2 \\delta"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1077,
+                    759,
+                    1355,
+                    759,
+                    1355,
+                    792,
+                    1077,
+                    792
+                ],
+                "score": 0.59,
+                "latex": "\\begin{array} { r } { \\mathcal { H } _ { k } \\quad \\subset \\quad ( \\mathcal { Z } \\quad \\to \\quad \\mathbb { R } ) } \\end{array}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1078,
+                    760,
+                    1120,
+                    760,
+                    1120,
+                    791,
+                    1078,
+                    791
+                ],
+                "score": 0.52,
+                "latex": "\\mathcal { H } _ { k }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1316,
+                    231,
+                    1354,
+                    231,
+                    1354,
+                    259,
+                    1316,
+                    259
+                ],
+                "score": 0.36,
+                "latex": "1 { - } \\delta"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    72.0,
+                    815.0,
+                    72.0,
+                    815.0,
+                    108.0,
+                    296.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    832.0,
+                    2085.0,
+                    870.0,
+                    2085.0,
+                    870.0,
+                    2124.0,
+                    832.0,
+                    2124.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1377.0,
+                    389.0,
+                    1403.0,
+                    389.0,
+                    1403.0,
+                    417.0,
+                    1377.0,
+                    417.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1379.0,
+                    1255.0,
+                    1402.0,
+                    1255.0,
+                    1402.0,
+                    1282.0,
+                    1379.0,
+                    1282.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    715.0,
+                    692.0,
+                    715.0,
+                    692.0,
+                    760.0,
+                    293.0,
+                    760.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    726.0,
+                    715.0,
+                    904.0,
+                    715.0,
+                    904.0,
+                    760.0,
+                    726.0,
+                    760.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    942.0,
+                    715.0,
+                    1404.0,
+                    715.0,
+                    1404.0,
+                    760.0,
+                    942.0,
+                    760.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    750.0,
+                    357.0,
+                    750.0,
+                    357.0,
+                    798.0,
+                    291.0,
+                    798.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    380.0,
+                    750.0,
+                    482.0,
+                    750.0,
+                    482.0,
+                    798.0,
+                    380.0,
+                    798.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    859.0,
+                    750.0,
+                    1076.0,
+                    750.0,
+                    1076.0,
+                    798.0,
+                    859.0,
+                    798.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1356.0,
+                    750.0,
+                    1407.0,
+                    750.0,
+                    1407.0,
+                    798.0,
+                    1356.0,
+                    798.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    786.0,
+                    645.0,
+                    786.0,
+                    645.0,
+                    829.0,
+                    292.0,
+                    829.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    668.0,
+                    786.0,
+                    798.0,
+                    786.0,
+                    798.0,
+                    829.0,
+                    668.0,
+                    829.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1051.0,
+                    786.0,
+                    1224.0,
+                    786.0,
+                    1224.0,
+                    829.0,
+                    1051.0,
+                    829.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1403.0,
+                    786.0,
+                    1406.0,
+                    786.0,
+                    1406.0,
+                    829.0,
+                    1403.0,
+                    829.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    287.0,
+                    824.0,
+                    293.0,
+                    824.0,
+                    293.0,
+                    885.0,
+                    287.0,
+                    885.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    957.0,
+                    824.0,
+                    1019.0,
+                    824.0,
+                    1019.0,
+                    885.0,
+                    957.0,
+                    885.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1255.0,
+                    824.0,
+                    1412.0,
+                    824.0,
+                    1412.0,
+                    885.0,
+                    1255.0,
+                    885.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    288.0,
+                    869.0,
+                    459.0,
+                    869.0,
+                    459.0,
+                    928.0,
+                    288.0,
+                    928.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    815.0,
+                    869.0,
+                    1177.0,
+                    869.0,
+                    1177.0,
+                    928.0,
+                    815.0,
+                    928.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1203.0,
+                    869.0,
+                    1408.0,
+                    869.0,
+                    1408.0,
+                    928.0,
+                    1203.0,
+                    928.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    910.0,
+                    296.0,
+                    910.0,
+                    296.0,
+                    953.0,
+                    291.0,
+                    953.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    376.0,
+                    910.0,
+                    390.0,
+                    910.0,
+                    390.0,
+                    953.0,
+                    376.0,
+                    953.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    224.0,
+                    726.0,
+                    224.0,
+                    726.0,
+                    270.0,
+                    292.0,
+                    270.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    772.0,
+                    224.0,
+                    1315.0,
+                    224.0,
+                    1315.0,
+                    270.0,
+                    772.0,
+                    270.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1355.0,
+                    224.0,
+                    1405.0,
+                    224.0,
+                    1405.0,
+                    270.0,
+                    1355.0,
+                    270.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    261.0,
+                    343.0,
+                    261.0,
+                    343.0,
+                    299.0,
+                    292.0,
+                    299.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1087.0,
+                    1169.0,
+                    1087.0,
+                    1169.0,
+                    1129.0,
+                    293.0,
+                    1129.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1234.0,
+                    1087.0,
+                    1331.0,
+                    1087.0,
+                    1331.0,
+                    1129.0,
+                    1234.0,
+                    1129.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1247.0,
+                    1350.0,
+                    1247.0,
+                    1350.0,
+                    1287.0,
+                    292.0,
+                    1287.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    449.0,
+                    584.0,
+                    449.0,
+                    584.0,
+                    486.0,
+                    297.0,
+                    486.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    612.0,
+                    449.0,
+                    660.0,
+                    449.0,
+                    660.0,
+                    486.0,
+                    612.0,
+                    486.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    675.0,
+                    782.0,
+                    675.0,
+                    782.0,
+                    713.0,
+                    295.0,
+                    713.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    594.0,
+                    786.0,
+                    594.0,
+                    786.0,
+                    642.0,
+                    293.0,
+                    642.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1290.0,
+                    594.0,
+                    1408.0,
+                    594.0,
+                    1408.0,
+                    642.0,
+                    1290.0,
+                    642.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    627.0,
+                    455.0,
+                    627.0,
+                    455.0,
+                    665.0,
+                    294.0,
+                    665.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    497.0,
+                    627.0,
+                    506.0,
+                    627.0,
+                    506.0,
+                    665.0,
+                    497.0,
+                    665.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    530.0,
+                    627.0,
+                    812.0,
+                    627.0,
+                    812.0,
+                    665.0,
+                    530.0,
+                    665.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    853.0,
+                    627.0,
+                    864.0,
+                    627.0,
+                    864.0,
+                    665.0,
+                    853.0,
+                    665.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 16,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1410,
+                    1404,
+                    1410,
+                    1404,
+                    1687,
+                    297,
+                    1687
+                ],
+                "score": 0.981
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1701,
+                    1404,
+                    1701,
+                    1404,
+                    1977,
+                    298,
+                    1977
+                ],
+                "score": 0.98
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    606,
+                    1405,
+                    606,
+                    1405,
+                    942,
+                    297,
+                    942
+                ],
+                "score": 0.98
+            },
+            {
+                "category_id": 5,
+                "poly": [
+                    371,
+                    1041,
+                    1333,
+                    1041,
+                    1333,
+                    1329,
+                    371,
+                    1329
+                ],
+                "score": 0.977,
+                "html": "<table><tr><td></td><td>CartPole-vO</td><td>Acrobot-v1</td><td>HIV simulator</td></tr><tr><td></td><td></td><td></td><td></td></tr><tr><td>state space dimension</td><td>4</td><td>6</td><td>6</td></tr><tr><td># of actions</td><td>2</td><td>3</td><td>4</td></tr><tr><td>discount rate y</td><td>0.99</td><td>0.99</td><td>0.98</td></tr><tr><td># of trajectories</td><td>200</td><td>200</td><td>50</td></tr><tr><td>max length of training traj.</td><td>200</td><td>200</td><td>200</td></tr><tr><td>max length of rollouts for evaluation</td><td>1000</td><td>1000</td><td>1000</td></tr><tr><td>discounted return of target policy</td><td>89.4</td><td>-56.88</td><td>803.2</td></tr></table>"
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    354,
+                    1404,
+                    354,
+                    1404,
+                    509,
+                    298,
+                    509
+                ],
+                "score": 0.975
+            },
+            {
+                "category_id": 6,
+                "poly": [
+                    610,
+                    974,
+                    1090,
+                    974,
+                    1090,
+                    1007,
+                    610,
+                    1007
+                ],
+                "score": 0.91
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    332,
+                    2005,
+                    763,
+                    2005,
+                    763,
+                    2034,
+                    332,
+                    2034
+                ],
+                "score": 0.91
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    303,
+                    548,
+                    790,
+                    548,
+                    790,
+                    580,
+                    303,
+                    580
+                ],
+                "score": 0.881
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    300,
+                    226,
+                    675,
+                    226,
+                    675,
+                    261,
+                    300,
+                    261
+                ],
+                "score": 0.881
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    301,
+                    297,
+                    740,
+                    297,
+                    740,
+                    328,
+                    301,
+                    328
+                ],
+                "score": 0.879
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    299,
+                    75,
+                    813,
+                    75,
+                    813,
+                    105,
+                    299,
+                    105
+                ],
+                "score": 0.877
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    836,
+                    2088,
+                    865,
+                    2088,
+                    865,
+                    2112,
+                    836,
+                    2112
+                ],
+                "score": 0.849
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    871,
+                    1792,
+                    977,
+                    1792,
+                    977,
+                    1822,
+                    871,
+                    1822
+                ],
+                "score": 0.92,
+                "latex": "3 \\times 1 0 ^ { - 4 }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1068,
+                    1503,
+                    1152,
+                    1503,
+                    1152,
+                    1536,
+                    1068,
+                    1536
+                ],
+                "score": 0.92,
+                "latex": "( - 1 , 1 )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1145,
+                    759,
+                    1199,
+                    759,
+                    1199,
+                    790,
+                    1145,
+                    790
+                ],
+                "score": 0.89,
+                "latex": "1 0 \\%"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    889,
+                    819,
+                    942,
+                    819,
+                    942,
+                    850,
+                    889,
+                    850
+                ],
+                "score": 0.84,
+                "latex": "1 0 ^ { 1 0 }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1268,
+                    728,
+                    1346,
+                    728,
+                    1346,
+                    761,
+                    1268,
+                    761
+                ],
+                "score": 0.68,
+                "latex": "\\mathsf { R L P y } ^ { 3 }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    843,
+                    1703,
+                    880,
+                    1703,
+                    880,
+                    1731,
+                    843,
+                    1731
+                ],
+                "score": 0.67,
+                "latex": "L 2"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1281,
+                    825,
+                    1300,
+                    825,
+                    1300,
+                    853,
+                    1281,
+                    853
+                ],
+                "score": 0.66,
+                "latex": "{ \\bf q }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    532,
+                    1175,
+                    554,
+                    1175,
+                    554,
+                    1202,
+                    532,
+                    1202
+                ],
+                "score": 0.55,
+                "latex": "\\gamma"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    604.0,
+                    971.0,
+                    1093.0,
+                    971.0,
+                    1093.0,
+                    1012.0,
+                    604.0,
+                    1012.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    330.0,
+                    1998.0,
+                    766.0,
+                    1998.0,
+                    766.0,
+                    2041.0,
+                    330.0,
+                    2041.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    548.0,
+                    795.0,
+                    548.0,
+                    795.0,
+                    584.0,
+                    295.0,
+                    584.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    225.0,
+                    678.0,
+                    225.0,
+                    678.0,
+                    266.0,
+                    294.0,
+                    266.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    296.0,
+                    744.0,
+                    296.0,
+                    744.0,
+                    331.0,
+                    295.0,
+                    331.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    73.0,
+                    815.0,
+                    73.0,
+                    815.0,
+                    108.0,
+                    295.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    832.0,
+                    2085.0,
+                    870.0,
+                    2085.0,
+                    870.0,
+                    2122.0,
+                    832.0,
+                    2122.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1410.0,
+                    1402.0,
+                    1410.0,
+                    1402.0,
+                    1442.0,
+                    296.0,
+                    1442.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1440.0,
+                    1406.0,
+                    1440.0,
+                    1406.0,
+                    1477.0,
+                    294.0,
+                    1477.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1473.0,
+                    1406.0,
+                    1473.0,
+                    1406.0,
+                    1508.0,
+                    292.0,
+                    1508.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1504.0,
+                    1067.0,
+                    1504.0,
+                    1067.0,
+                    1537.0,
+                    296.0,
+                    1537.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1153.0,
+                    1504.0,
+                    1404.0,
+                    1504.0,
+                    1404.0,
+                    1537.0,
+                    1153.0,
+                    1537.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1533.0,
+                    1406.0,
+                    1533.0,
+                    1406.0,
+                    1569.0,
+                    294.0,
+                    1569.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1563.0,
+                    1406.0,
+                    1563.0,
+                    1406.0,
+                    1598.0,
+                    292.0,
+                    1598.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    1597.0,
+                    1404.0,
+                    1597.0,
+                    1404.0,
+                    1626.0,
+                    297.0,
+                    1626.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    290.0,
+                    1617.0,
+                    1406.0,
+                    1617.0,
+                    1406.0,
+                    1666.0,
+                    290.0,
+                    1666.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1653.0,
+                    1402.0,
+                    1653.0,
+                    1402.0,
+                    1693.0,
+                    294.0,
+                    1693.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1699.0,
+                    842.0,
+                    1699.0,
+                    842.0,
+                    1737.0,
+                    293.0,
+                    1737.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    881.0,
+                    1699.0,
+                    1405.0,
+                    1699.0,
+                    1405.0,
+                    1737.0,
+                    881.0,
+                    1737.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1733.0,
+                    1402.0,
+                    1733.0,
+                    1402.0,
+                    1766.0,
+                    296.0,
+                    1766.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1763.0,
+                    1404.0,
+                    1763.0,
+                    1404.0,
+                    1796.0,
+                    294.0,
+                    1796.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1789.0,
+                    870.0,
+                    1789.0,
+                    870.0,
+                    1830.0,
+                    293.0,
+                    1830.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    978.0,
+                    1789.0,
+                    1406.0,
+                    1789.0,
+                    1406.0,
+                    1830.0,
+                    978.0,
+                    1830.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1822.0,
+                    1404.0,
+                    1822.0,
+                    1404.0,
+                    1859.0,
+                    293.0,
+                    1859.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1853.0,
+                    1404.0,
+                    1853.0,
+                    1404.0,
+                    1891.0,
+                    293.0,
+                    1891.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    1885.0,
+                    1401.0,
+                    1885.0,
+                    1401.0,
+                    1917.0,
+                    297.0,
+                    1917.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1915.0,
+                    1406.0,
+                    1915.0,
+                    1406.0,
+                    1950.0,
+                    293.0,
+                    1950.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1945.0,
+                    1405.0,
+                    1945.0,
+                    1405.0,
+                    1981.0,
+                    293.0,
+                    1981.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    607.0,
+                    1406.0,
+                    607.0,
+                    1406.0,
+                    641.0,
+                    295.0,
+                    641.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    635.0,
+                    1405.0,
+                    635.0,
+                    1405.0,
+                    676.0,
+                    292.0,
+                    676.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    669.0,
+                    1405.0,
+                    669.0,
+                    1405.0,
+                    703.0,
+                    295.0,
+                    703.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    698.0,
+                    1405.0,
+                    698.0,
+                    1405.0,
+                    733.0,
+                    294.0,
+                    733.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    727.0,
+                    1267.0,
+                    727.0,
+                    1267.0,
+                    765.0,
+                    294.0,
+                    765.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1347.0,
+                    727.0,
+                    1406.0,
+                    727.0,
+                    1406.0,
+                    765.0,
+                    1347.0,
+                    765.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    762.0,
+                    1144.0,
+                    762.0,
+                    1144.0,
+                    793.0,
+                    295.0,
+                    793.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1200.0,
+                    762.0,
+                    1405.0,
+                    762.0,
+                    1405.0,
+                    793.0,
+                    1200.0,
+                    793.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    790.0,
+                    1405.0,
+                    790.0,
+                    1405.0,
+                    825.0,
+                    294.0,
+                    825.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    817.0,
+                    888.0,
+                    817.0,
+                    888.0,
+                    856.0,
+                    294.0,
+                    856.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    943.0,
+                    817.0,
+                    1280.0,
+                    817.0,
+                    1280.0,
+                    856.0,
+                    943.0,
+                    856.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1301.0,
+                    817.0,
+                    1405.0,
+                    817.0,
+                    1405.0,
+                    856.0,
+                    1301.0,
+                    856.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    851.0,
+                    1405.0,
+                    851.0,
+                    1405.0,
+                    885.0,
+                    295.0,
+                    885.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    879.0,
+                    1405.0,
+                    879.0,
+                    1405.0,
+                    916.0,
+                    292.0,
+                    916.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    290.0,
+                    909.0,
+                    394.0,
+                    909.0,
+                    394.0,
+                    950.0,
+                    290.0,
+                    950.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    354.0,
+                    1404.0,
+                    354.0,
+                    1404.0,
+                    392.0,
+                    294.0,
+                    392.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    385.0,
+                    1405.0,
+                    385.0,
+                    1405.0,
+                    422.0,
+                    294.0,
+                    422.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    416.0,
+                    1405.0,
+                    416.0,
+                    1405.0,
+                    453.0,
+                    292.0,
+                    453.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    446.0,
+                    1406.0,
+                    446.0,
+                    1406.0,
+                    483.0,
+                    292.0,
+                    483.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    478.0,
+                    435.0,
+                    478.0,
+                    435.0,
+                    509.0,
+                    294.0,
+                    509.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 17,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1787,
+                    1404,
+                    1787,
+                    1404,
+                    2035,
+                    297,
+                    2035
+                ],
+                "score": 0.98
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    593,
+                    930,
+                    593,
+                    930,
+                    899,
+                    298,
+                    899
+                ],
+                "score": 0.98
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    930,
+                    1405,
+                    930,
+                    1405,
+                    1237,
+                    297,
+                    1237
+                ],
+                "score": 0.978
+            },
+            {
+                "category_id": 3,
+                "poly": [
+                    295,
+                    1262,
+                    1406,
+                    1262,
+                    1406,
+                    1591,
+                    295,
+                    1591
+                ],
+                "score": 0.969
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    300,
+                    307,
+                    1402,
+                    307,
+                    1402,
+                    400,
+                    300,
+                    400
+                ],
+                "score": 0.956
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    518,
+                    1401,
+                    518,
+                    1401,
+                    580,
+                    299,
+                    580
+                ],
+                "score": 0.952
+            },
+            {
+                "category_id": 3,
+                "poly": [
+                    943,
+                    597,
+                    1394,
+                    597,
+                    1394,
+                    887,
+                    943,
+                    887
+                ],
+                "score": 0.95
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    421,
+                    415,
+                    1277,
+                    415,
+                    1277,
+                    503,
+                    421,
+                    503
+                ],
+                "score": 0.949
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    294,
+                    226,
+                    1401,
+                    226,
+                    1401,
+                    294,
+                    294,
+                    294
+                ],
+                "score": 0.948
+            },
+            {
+                "category_id": 4,
+                "poly": [
+                    293,
+                    1611,
+                    1401,
+                    1611,
+                    1401,
+                    1673,
+                    293,
+                    1673
+                ],
+                "score": 0.92
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    300,
+                    75,
+                    813,
+                    75,
+                    813,
+                    105,
+                    300,
+                    105
+                ],
+                "score": 0.912
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    837,
+                    2088,
+                    865,
+                    2088,
+                    865,
+                    2112,
+                    837,
+                    2112
+                ],
+                "score": 0.857
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    299,
+                    1731,
+                    893,
+                    1731,
+                    893,
+                    1763,
+                    299,
+                    1763
+                ],
+                "score": 0.773
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    1731,
+                    893,
+                    1731,
+                    893,
+                    1763,
+                    299,
+                    1763
+                ],
+                "score": 0.116
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    645,
+                    714,
+                    781,
+                    714,
+                    781,
+                    748,
+                    645,
+                    748
+                ],
+                "score": 0.93,
+                "latex": "\\alpha \\in \\{ 1 , 1 0 \\}"
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    418,
+                    414,
+                    1276,
+                    414,
+                    1276,
+                    504,
+                    418,
+                    504
+                ],
+                "score": 0.93,
+                "latex": "\\mathrm { I n d i v i d u a l ~ M S E } = \\mathbb { E } _ { s _ { 0 } \\sim d _ { 0 } } \\bigg [ \\bigg ( \\mathbb { E } _ { \\widehat { M } , \\pi } \\Big [ \\sum _ { t = 0 } ^ { \\infty } \\gamma ^ { t } r _ { t } | s _ { 0 } \\Big ] - \\mathbb { E } _ { M , \\pi } \\Big [ \\sum _ { t = 0 } ^ { \\infty } \\gamma ^ { t } r _ { t } | s _ { 0 } \\Big ] \\bigg ) ^ { 2 } \\bigg ]"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    817,
+                    228,
+                    1128,
+                    228,
+                    1128,
+                    264,
+                    817,
+                    264
+                ],
+                "score": 0.92,
+                "latex": "k ( \\phi ( s ) , \\phi ( \\bar { s } ) ) = \\phi ( s ) ^ { \\top } \\phi ( \\bar { s } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    391,
+                    746,
+                    577,
+                    746,
+                    577,
+                    779,
+                    391,
+                    779
+                ],
+                "score": 0.92,
+                "latex": "\\alpha \\in \\{ 0 . 0 1 , 0 . 1 \\}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    299,
+                    714,
+                    453,
+                    714,
+                    453,
+                    747,
+                    299,
+                    747
+                ],
+                "score": 0.91,
+                "latex": "\\lbrace 0 . 0 0 1 , 0 . 0 1 \\rbrace"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    297,
+                    654,
+                    633,
+                    654,
+                    633,
+                    688,
+                    297,
+                    688
+                ],
+                "score": 0.9,
+                "latex": "\\alpha \\in \\{ 0 . 0 0 1 , 0 . 0 1 , 0 . 1 , 1 , 1 0 \\}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    694,
+                    1642,
+                    749,
+                    1642,
+                    749,
+                    1671,
+                    694,
+                    1671
+                ],
+                "score": 0.87,
+                "latex": "9 5 \\%"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    878,
+                    687,
+                    929,
+                    687,
+                    929,
+                    715,
+                    878,
+                    715
+                ],
+                "score": 0.82,
+                "latex": "\\alpha \\in"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    494,
+                    1943,
+                    545,
+                    1943,
+                    545,
+                    1972,
+                    494,
+                    1972
+                ],
+                "score": 0.77,
+                "latex": "- 2 0"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    388,
+                    873,
+                    410,
+                    873,
+                    410,
+                    895,
+                    388,
+                    895
+                ],
+                "score": 0.76,
+                "latex": "\\alpha"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    479,
+                    630,
+                    501,
+                    630,
+                    501,
+                    651,
+                    479,
+                    651
+                ],
+                "score": 0.76,
+                "latex": "\\alpha"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    611,
+                    781,
+                    633,
+                    781,
+                    633,
+                    803,
+                    611,
+                    803
+                ],
+                "score": 0.76,
+                "latex": "\\alpha"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    735,
+                    599,
+                    757,
+                    599,
+                    757,
+                    621,
+                    735,
+                    621
+                ],
+                "score": 0.75,
+                "latex": "\\alpha"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    912,
+                    661,
+                    929,
+                    661,
+                    929,
+                    682,
+                    912,
+                    682
+                ],
+                "score": 0.55,
+                "latex": "\\epsilon"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1328,
+                    1882,
+                    1394,
+                    1882,
+                    1394,
+                    1911,
+                    1328,
+                    1911
+                ],
+                "score": 0.52,
+                "latex": "- 2 8 0"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    805,
+                    1791,
+                    839,
+                    1791,
+                    839,
+                    1819,
+                    805,
+                    1819
+                ],
+                "score": 0.26,
+                "latex": "\\mathrm { F u }"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1260.0,
+                    328.0,
+                    1260.0,
+                    328.0,
+                    1526.0,
+                    292.0,
+                    1526.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    448.0,
+                    1264.0,
+                    593.0,
+                    1264.0,
+                    593.0,
+                    1296.0,
+                    448.0,
+                    1296.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    818.0,
+                    1264.0,
+                    956.0,
+                    1264.0,
+                    956.0,
+                    1296.0,
+                    818.0,
+                    1296.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1170.0,
+                    1264.0,
+                    1341.0,
+                    1264.0,
+                    1341.0,
+                    1295.0,
+                    1170.0,
+                    1295.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    319.0,
+                    1297.0,
+                    370.0,
+                    1297.0,
+                    370.0,
+                    1323.0,
+                    319.0,
+                    1323.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    572.0,
+                    1303.0,
+                    594.0,
+                    1303.0,
+                    594.0,
+                    1341.0,
+                    572.0,
+                    1341.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    688.0,
+                    1297.0,
+                    741.0,
+                    1297.0,
+                    741.0,
+                    1334.0,
+                    688.0,
+                    1334.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    777.0,
+                    1302.0,
+                    808.0,
+                    1302.0,
+                    808.0,
+                    1359.0,
+                    777.0,
+                    1359.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    844.0,
+                    1310.0,
+                    868.0,
+                    1310.0,
+                    868.0,
+                    1344.0,
+                    844.0,
+                    1344.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1053.0,
+                    1288.0,
+                    1108.0,
+                    1288.0,
+                    1108.0,
+                    1321.0,
+                    1053.0,
+                    1321.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1119.0,
+                    1311.0,
+                    1141.0,
+                    1311.0,
+                    1141.0,
+                    1334.0,
+                    1119.0,
+                    1334.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1213.0,
+                    1303.0,
+                    1235.0,
+                    1303.0,
+                    1235.0,
+                    1333.0,
+                    1213.0,
+                    1333.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    321.0,
+                    1358.0,
+                    372.0,
+                    1358.0,
+                    372.0,
+                    1457.0,
+                    321.0,
+                    1457.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    380.0,
+                    1366.0,
+                    409.0,
+                    1366.0,
+                    409.0,
+                    1453.0,
+                    380.0,
+                    1453.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    412.0,
+                    1348.0,
+                    438.0,
+                    1348.0,
+                    438.0,
+                    1446.0,
+                    412.0,
+                    1446.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    444.0,
+                    1361.0,
+                    471.0,
+                    1361.0,
+                    471.0,
+                    1437.0,
+                    444.0,
+                    1437.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    476.0,
+                    1355.0,
+                    501.0,
+                    1355.0,
+                    501.0,
+                    1434.0,
+                    476.0,
+                    1434.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    506.0,
+                    1342.0,
+                    533.0,
+                    1342.0,
+                    533.0,
+                    1433.0,
+                    506.0,
+                    1433.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    538.0,
+                    1323.0,
+                    565.0,
+                    1323.0,
+                    565.0,
+                    1436.0,
+                    538.0,
+                    1436.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    572.0,
+                    1369.0,
+                    593.0,
+                    1369.0,
+                    593.0,
+                    1425.0,
+                    572.0,
+                    1425.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    602.0,
+                    1347.0,
+                    627.0,
+                    1347.0,
+                    627.0,
+                    1406.0,
+                    602.0,
+                    1406.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    632.0,
+                    1344.0,
+                    660.0,
+                    1344.0,
+                    660.0,
+                    1413.0,
+                    632.0,
+                    1413.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    685.0,
+                    1358.0,
+                    739.0,
+                    1358.0,
+                    739.0,
+                    1392.0,
+                    685.0,
+                    1392.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    747.0,
+                    1375.0,
+                    839.0,
+                    1375.0,
+                    839.0,
+                    1435.0,
+                    747.0,
+                    1435.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    844.0,
+                    1359.0,
+                    868.0,
+                    1359.0,
+                    868.0,
+                    1423.0,
+                    844.0,
+                    1423.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    871.0,
+                    1370.0,
+                    901.0,
+                    1370.0,
+                    901.0,
+                    1439.0,
+                    871.0,
+                    1439.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    906.0,
+                    1374.0,
+                    934.0,
+                    1374.0,
+                    934.0,
+                    1441.0,
+                    906.0,
+                    1441.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    938.0,
+                    1378.0,
+                    964.0,
+                    1378.0,
+                    964.0,
+                    1453.0,
+                    938.0,
+                    1453.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    969.0,
+                    1358.0,
+                    996.0,
+                    1358.0,
+                    996.0,
+                    1431.0,
+                    969.0,
+                    1431.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    999.0,
+                    1377.0,
+                    1028.0,
+                    1377.0,
+                    1028.0,
+                    1442.0,
+                    999.0,
+                    1442.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1052.0,
+                    1358.0,
+                    1107.0,
+                    1358.0,
+                    1107.0,
+                    1388.0,
+                    1052.0,
+                    1388.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1115.0,
+                    1422.0,
+                    1143.0,
+                    1422.0,
+                    1143.0,
+                    1458.0,
+                    1115.0,
+                    1458.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1148.0,
+                    1336.0,
+                    1173.0,
+                    1336.0,
+                    1173.0,
+                    1452.0,
+                    1148.0,
+                    1452.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1178.0,
+                    1323.0,
+                    1205.0,
+                    1323.0,
+                    1205.0,
+                    1451.0,
+                    1178.0,
+                    1451.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1213.0,
+                    1389.0,
+                    1239.0,
+                    1389.0,
+                    1239.0,
+                    1466.0,
+                    1213.0,
+                    1466.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1240.0,
+                    1348.0,
+                    1271.0,
+                    1348.0,
+                    1271.0,
+                    1465.0,
+                    1240.0,
+                    1465.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1276.0,
+                    1325.0,
+                    1301.0,
+                    1325.0,
+                    1301.0,
+                    1457.0,
+                    1276.0,
+                    1457.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1306.0,
+                    1340.0,
+                    1331.0,
+                    1340.0,
+                    1331.0,
+                    1456.0,
+                    1306.0,
+                    1456.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1337.0,
+                    1333.0,
+                    1364.0,
+                    1333.0,
+                    1364.0,
+                    1455.0,
+                    1337.0,
+                    1455.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1369.0,
+                    1336.0,
+                    1394.0,
+                    1336.0,
+                    1394.0,
+                    1452.0,
+                    1369.0,
+                    1452.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    380.0,
+                    1456.0,
+                    409.0,
+                    1456.0,
+                    409.0,
+                    1480.0,
+                    380.0,
+                    1480.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    441.0,
+                    1459.0,
+                    472.0,
+                    1459.0,
+                    472.0,
+                    1484.0,
+                    441.0,
+                    1484.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    477.0,
+                    1459.0,
+                    500.0,
+                    1459.0,
+                    500.0,
+                    1480.0,
+                    477.0,
+                    1480.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    538.0,
+                    1457.0,
+                    564.0,
+                    1457.0,
+                    564.0,
+                    1481.0,
+                    538.0,
+                    1481.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    572.0,
+                    1462.0,
+                    596.0,
+                    1462.0,
+                    596.0,
+                    1491.0,
+                    572.0,
+                    1491.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    602.0,
+                    1467.0,
+                    627.0,
+                    1467.0,
+                    627.0,
+                    1494.0,
+                    602.0,
+                    1494.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    630.0,
+                    1466.0,
+                    658.0,
+                    1466.0,
+                    658.0,
+                    1488.0,
+                    630.0,
+                    1488.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    748.0,
+                    1465.0,
+                    775.0,
+                    1465.0,
+                    775.0,
+                    1492.0,
+                    748.0,
+                    1492.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    779.0,
+                    1458.0,
+                    805.0,
+                    1458.0,
+                    805.0,
+                    1485.0,
+                    779.0,
+                    1485.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    813.0,
+                    1455.0,
+                    837.0,
+                    1455.0,
+                    837.0,
+                    1478.0,
+                    813.0,
+                    1478.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    845.0,
+                    1452.0,
+                    866.0,
+                    1452.0,
+                    866.0,
+                    1472.0,
+                    845.0,
+                    1472.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    874.0,
+                    1459.0,
+                    901.0,
+                    1459.0,
+                    901.0,
+                    1486.0,
+                    874.0,
+                    1486.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    906.0,
+                    1465.0,
+                    931.0,
+                    1465.0,
+                    931.0,
+                    1492.0,
+                    906.0,
+                    1492.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1117.0,
+                    1467.0,
+                    1142.0,
+                    1467.0,
+                    1142.0,
+                    1492.0,
+                    1117.0,
+                    1492.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1151.0,
+                    1455.0,
+                    1171.0,
+                    1455.0,
+                    1171.0,
+                    1474.0,
+                    1151.0,
+                    1474.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1184.0,
+                    1446.0,
+                    1202.0,
+                    1446.0,
+                    1202.0,
+                    1470.0,
+                    1184.0,
+                    1470.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    371.0,
+                    1495.0,
+                    418.0,
+                    1495.0,
+                    418.0,
+                    1526.0,
+                    371.0,
+                    1526.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    433.0,
+                    1495.0,
+                    481.0,
+                    1495.0,
+                    481.0,
+                    1526.0,
+                    433.0,
+                    1526.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    496.0,
+                    1495.0,
+                    543.0,
+                    1495.0,
+                    543.0,
+                    1526.0,
+                    496.0,
+                    1526.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    560.0,
+                    1495.0,
+                    606.0,
+                    1495.0,
+                    606.0,
+                    1526.0,
+                    560.0,
+                    1526.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    622.0,
+                    1495.0,
+                    668.0,
+                    1495.0,
+                    668.0,
+                    1526.0,
+                    622.0,
+                    1526.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    738.0,
+                    1496.0,
+                    785.0,
+                    1496.0,
+                    785.0,
+                    1526.0,
+                    738.0,
+                    1526.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    801.0,
+                    1495.0,
+                    848.0,
+                    1495.0,
+                    848.0,
+                    1526.0,
+                    801.0,
+                    1526.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    864.0,
+                    1494.0,
+                    911.0,
+                    1494.0,
+                    911.0,
+                    1526.0,
+                    864.0,
+                    1526.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    927.0,
+                    1495.0,
+                    974.0,
+                    1495.0,
+                    974.0,
+                    1526.0,
+                    927.0,
+                    1526.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    991.0,
+                    1495.0,
+                    1036.0,
+                    1495.0,
+                    1036.0,
+                    1526.0,
+                    991.0,
+                    1526.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1108.0,
+                    1497.0,
+                    1152.0,
+                    1497.0,
+                    1152.0,
+                    1524.0,
+                    1108.0,
+                    1524.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1170.0,
+                    1497.0,
+                    1219.0,
+                    1497.0,
+                    1219.0,
+                    1525.0,
+                    1170.0,
+                    1525.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1230.0,
+                    1497.0,
+                    1341.0,
+                    1497.0,
+                    1341.0,
+                    1525.0,
+                    1230.0,
+                    1525.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1356.0,
+                    1497.0,
+                    1400.0,
+                    1497.0,
+                    1400.0,
+                    1524.0,
+                    1356.0,
+                    1524.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    379.0,
+                    1523.0,
+                    659.0,
+                    1523.0,
+                    659.0,
+                    1557.0,
+                    379.0,
+                    1557.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    745.0,
+                    1519.0,
+                    1025.0,
+                    1519.0,
+                    1025.0,
+                    1557.0,
+                    745.0,
+                    1557.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1115.0,
+                    1523.0,
+                    1394.0,
+                    1523.0,
+                    1394.0,
+                    1557.0,
+                    1115.0,
+                    1557.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    422.0,
+                    1559.0,
+                    574.0,
+                    1559.0,
+                    574.0,
+                    1591.0,
+                    422.0,
+                    1591.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    615.0,
+                    1559.0,
+                    640.0,
+                    1559.0,
+                    640.0,
+                    1590.0,
+                    615.0,
+                    1590.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    656.0,
+                    1556.0,
+                    717.0,
+                    1556.0,
+                    717.0,
+                    1593.0,
+                    656.0,
+                    1593.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    757.0,
+                    1559.0,
+                    781.0,
+                    1559.0,
+                    781.0,
+                    1591.0,
+                    757.0,
+                    1591.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    796.0,
+                    1557.0,
+                    853.0,
+                    1557.0,
+                    853.0,
+                    1592.0,
+                    796.0,
+                    1592.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    890.0,
+                    1557.0,
+                    917.0,
+                    1557.0,
+                    917.0,
+                    1591.0,
+                    890.0,
+                    1591.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    932.0,
+                    1559.0,
+                    965.0,
+                    1559.0,
+                    965.0,
+                    1590.0,
+                    932.0,
+                    1590.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1003.0,
+                    1557.0,
+                    1031.0,
+                    1557.0,
+                    1031.0,
+                    1592.0,
+                    1003.0,
+                    1592.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1046.0,
+                    1558.0,
+                    1090.0,
+                    1558.0,
+                    1090.0,
+                    1591.0,
+                    1046.0,
+                    1591.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1129.0,
+                    1559.0,
+                    1153.0,
+                    1559.0,
+                    1153.0,
+                    1591.0,
+                    1129.0,
+                    1591.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1171.0,
+                    1558.0,
+                    1277.0,
+                    1558.0,
+                    1277.0,
+                    1590.0,
+                    1171.0,
+                    1590.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    672.25,
+                    1414.5,
+                    742.25,
+                    1414.5,
+                    742.25,
+                    1449.5,
+                    672.25,
+                    1449.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1050.25,
+                    1422.5,
+                    1111.25,
+                    1422.5,
+                    1111.25,
+                    1458.0,
+                    1050.25,
+                    1458.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    500.0,
+                    1459.0,
+                    543.0,
+                    1459.0,
+                    543.0,
+                    1479.0,
+                    500.0,
+                    1479.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1073.0,
+                    600.0,
+                    1265.0,
+                    600.0,
+                    1265.0,
+                    625.0,
+                    1073.0,
+                    625.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    942.0,
+                    630.0,
+                    971.0,
+                    630.0,
+                    971.0,
+                    837.0,
+                    942.0,
+                    837.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    964.0,
+                    672.0,
+                    1017.0,
+                    672.0,
+                    1017.0,
+                    699.0,
+                    964.0,
+                    699.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1253.0,
+                    712.0,
+                    1269.0,
+                    712.0,
+                    1269.0,
+                    733.0,
+                    1253.0,
+                    733.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1362.0,
+                    741.0,
+                    1380.0,
+                    741.0,
+                    1380.0,
+                    767.0,
+                    1362.0,
+                    767.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    967.0,
+                    758.0,
+                    1018.0,
+                    758.0,
+                    1018.0,
+                    788.0,
+                    967.0,
+                    788.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1030.0,
+                    767.0,
+                    1049.0,
+                    767.0,
+                    1049.0,
+                    800.0,
+                    1030.0,
+                    800.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1139.0,
+                    766.0,
+                    1159.0,
+                    766.0,
+                    1159.0,
+                    807.0,
+                    1139.0,
+                    807.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1251.0,
+                    776.0,
+                    1271.0,
+                    776.0,
+                    1271.0,
+                    804.0,
+                    1251.0,
+                    804.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1002.0,
+                    814.0,
+                    1072.0,
+                    814.0,
+                    1072.0,
+                    845.0,
+                    1002.0,
+                    845.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1121.0,
+                    817.0,
+                    1176.0,
+                    817.0,
+                    1176.0,
+                    842.0,
+                    1121.0,
+                    842.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1234.0,
+                    814.0,
+                    1284.0,
+                    814.0,
+                    1284.0,
+                    844.0,
+                    1234.0,
+                    844.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1351.0,
+                    814.0,
+                    1390.0,
+                    814.0,
+                    1390.0,
+                    845.0,
+                    1351.0,
+                    845.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1194.0,
+                    843.0,
+                    1214.0,
+                    843.0,
+                    1214.0,
+                    864.0,
+                    1194.0,
+                    864.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    956.0,
+                    857.0,
+                    1080.0,
+                    857.0,
+                    1080.0,
+                    885.0,
+                    956.0,
+                    885.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1113.0,
+                    860.0,
+                    1134.0,
+                    860.0,
+                    1134.0,
+                    883.0,
+                    1113.0,
+                    883.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1145.0,
+                    859.0,
+                    1216.0,
+                    859.0,
+                    1216.0,
+                    885.0,
+                    1145.0,
+                    885.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1250.0,
+                    861.0,
+                    1267.0,
+                    861.0,
+                    1267.0,
+                    880.0,
+                    1250.0,
+                    880.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1283.0,
+                    859.0,
+                    1384.0,
+                    859.0,
+                    1384.0,
+                    885.0,
+                    1283.0,
+                    885.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1610.0,
+                    1404.0,
+                    1610.0,
+                    1404.0,
+                    1646.0,
+                    295.0,
+                    1646.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1639.0,
+                    693.0,
+                    1639.0,
+                    693.0,
+                    1674.0,
+                    295.0,
+                    1674.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    750.0,
+                    1639.0,
+                    975.0,
+                    1639.0,
+                    975.0,
+                    1674.0,
+                    750.0,
+                    1674.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    73.0,
+                    815.0,
+                    73.0,
+                    815.0,
+                    108.0,
+                    295.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    832.0,
+                    2085.0,
+                    869.0,
+                    2085.0,
+                    869.0,
+                    2124.0,
+                    832.0,
+                    2124.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1733.0,
+                    895.0,
+                    1733.0,
+                    895.0,
+                    1765.0,
+                    295.0,
+                    1765.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1789.0,
+                    804.0,
+                    1789.0,
+                    804.0,
+                    1823.0,
+                    296.0,
+                    1823.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    840.0,
+                    1789.0,
+                    1404.0,
+                    1789.0,
+                    1404.0,
+                    1823.0,
+                    840.0,
+                    1823.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1819.0,
+                    1405.0,
+                    1819.0,
+                    1405.0,
+                    1855.0,
+                    291.0,
+                    1855.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1852.0,
+                    1404.0,
+                    1852.0,
+                    1404.0,
+                    1886.0,
+                    294.0,
+                    1886.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1882.0,
+                    1327.0,
+                    1882.0,
+                    1327.0,
+                    1915.0,
+                    292.0,
+                    1915.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1395.0,
+                    1882.0,
+                    1404.0,
+                    1882.0,
+                    1404.0,
+                    1915.0,
+                    1395.0,
+                    1915.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1913.0,
+                    1402.0,
+                    1913.0,
+                    1402.0,
+                    1944.0,
+                    296.0,
+                    1944.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1942.0,
+                    493.0,
+                    1942.0,
+                    493.0,
+                    1976.0,
+                    294.0,
+                    1976.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    546.0,
+                    1942.0,
+                    1406.0,
+                    1942.0,
+                    1406.0,
+                    1976.0,
+                    546.0,
+                    1976.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1974.0,
+                    1404.0,
+                    1974.0,
+                    1404.0,
+                    2004.0,
+                    296.0,
+                    2004.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    2002.0,
+                    1039.0,
+                    2002.0,
+                    1039.0,
+                    2037.0,
+                    292.0,
+                    2037.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    591.0,
+                    734.0,
+                    591.0,
+                    734.0,
+                    628.0,
+                    296.0,
+                    628.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    758.0,
+                    591.0,
+                    932.0,
+                    591.0,
+                    932.0,
+                    628.0,
+                    758.0,
+                    628.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    623.0,
+                    478.0,
+                    623.0,
+                    478.0,
+                    656.0,
+                    294.0,
+                    656.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    502.0,
+                    623.0,
+                    933.0,
+                    623.0,
+                    933.0,
+                    656.0,
+                    502.0,
+                    656.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    653.0,
+                    296.0,
+                    653.0,
+                    296.0,
+                    688.0,
+                    293.0,
+                    688.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    634.0,
+                    653.0,
+                    911.0,
+                    653.0,
+                    911.0,
+                    688.0,
+                    634.0,
+                    688.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    683.0,
+                    877.0,
+                    683.0,
+                    877.0,
+                    718.0,
+                    294.0,
+                    718.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    454.0,
+                    714.0,
+                    644.0,
+                    714.0,
+                    644.0,
+                    748.0,
+                    454.0,
+                    748.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    782.0,
+                    714.0,
+                    932.0,
+                    714.0,
+                    932.0,
+                    748.0,
+                    782.0,
+                    748.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    744.0,
+                    390.0,
+                    744.0,
+                    390.0,
+                    777.0,
+                    294.0,
+                    777.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    578.0,
+                    744.0,
+                    931.0,
+                    744.0,
+                    931.0,
+                    777.0,
+                    578.0,
+                    777.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    776.0,
+                    610.0,
+                    776.0,
+                    610.0,
+                    808.0,
+                    295.0,
+                    808.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    634.0,
+                    776.0,
+                    931.0,
+                    776.0,
+                    931.0,
+                    808.0,
+                    634.0,
+                    808.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    807.0,
+                    932.0,
+                    807.0,
+                    932.0,
+                    840.0,
+                    296.0,
+                    840.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    836.0,
+                    933.0,
+                    836.0,
+                    933.0,
+                    870.0,
+                    296.0,
+                    870.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    868.0,
+                    387.0,
+                    868.0,
+                    387.0,
+                    899.0,
+                    295.0,
+                    899.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    411.0,
+                    868.0,
+                    589.0,
+                    868.0,
+                    589.0,
+                    899.0,
+                    411.0,
+                    899.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    932.0,
+                    1403.0,
+                    932.0,
+                    1403.0,
+                    964.0,
+                    297.0,
+                    964.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    961.0,
+                    1406.0,
+                    961.0,
+                    1406.0,
+                    997.0,
+                    295.0,
+                    997.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    991.0,
+                    1406.0,
+                    991.0,
+                    1406.0,
+                    1030.0,
+                    294.0,
+                    1030.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1024.0,
+                    1403.0,
+                    1024.0,
+                    1403.0,
+                    1056.0,
+                    296.0,
+                    1056.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1052.0,
+                    1405.0,
+                    1052.0,
+                    1405.0,
+                    1087.0,
+                    294.0,
+                    1087.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1084.0,
+                    1403.0,
+                    1084.0,
+                    1403.0,
+                    1115.0,
+                    295.0,
+                    1115.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1111.0,
+                    1405.0,
+                    1111.0,
+                    1405.0,
+                    1148.0,
+                    292.0,
+                    1148.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1141.0,
+                    1405.0,
+                    1141.0,
+                    1405.0,
+                    1180.0,
+                    292.0,
+                    1180.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1176.0,
+                    1403.0,
+                    1176.0,
+                    1403.0,
+                    1208.0,
+                    296.0,
+                    1208.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1207.0,
+                    823.0,
+                    1207.0,
+                    823.0,
+                    1239.0,
+                    296.0,
+                    1239.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    308.0,
+                    1406.0,
+                    308.0,
+                    1406.0,
+                    341.0,
+                    296.0,
+                    341.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    340.0,
+                    1404.0,
+                    340.0,
+                    1404.0,
+                    370.0,
+                    295.0,
+                    370.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    369.0,
+                    628.0,
+                    369.0,
+                    628.0,
+                    399.0,
+                    296.0,
+                    399.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    514.0,
+                    1405.0,
+                    514.0,
+                    1405.0,
+                    557.0,
+                    293.0,
+                    557.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    548.0,
+                    1153.0,
+                    548.0,
+                    1153.0,
+                    582.0,
+                    294.0,
+                    582.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    225.0,
+                    816.0,
+                    225.0,
+                    816.0,
+                    268.0,
+                    292.0,
+                    268.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1129.0,
+                    225.0,
+                    1405.0,
+                    225.0,
+                    1405.0,
+                    268.0,
+                    1129.0,
+                    268.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    259.0,
+                    1404.0,
+                    259.0,
+                    1404.0,
+                    297.0,
+                    294.0,
+                    297.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1733.0,
+                    895.0,
+                    1733.0,
+                    895.0,
+                    1765.0,
+                    295.0,
+                    1765.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 18,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    296,
+                    760,
+                    1405,
+                    760,
+                    1405,
+                    1068,
+                    296,
+                    1068
+                ],
+                "score": 0.982
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    229,
+                    1403,
+                    229,
+                    1403,
+                    423,
+                    297,
+                    423
+                ],
+                "score": 0.977
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    435,
+                    1404,
+                    435,
+                    1404,
+                    567,
+                    297,
+                    567
+                ],
+                "score": 0.977
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    296,
+                    1733,
+                    1404,
+                    1733,
+                    1404,
+                    1980,
+                    296,
+                    1980
+                ],
+                "score": 0.938
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    329,
+                    2006,
+                    868,
+                    2006,
+                    868,
+                    2034,
+                    329,
+                    2034
+                ],
+                "score": 0.922
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    296,
+                    584,
+                    1401,
+                    584,
+                    1401,
+                    676,
+                    296,
+                    676
+                ],
+                "score": 0.922
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    294,
+                    693,
+                    1277,
+                    693,
+                    1277,
+                    728,
+                    294,
+                    728
+                ],
+                "score": 0.918
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    300,
+                    75,
+                    813,
+                    75,
+                    813,
+                    105,
+                    300,
+                    105
+                ],
+                "score": 0.895
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    834,
+                    2088,
+                    865,
+                    2088,
+                    865,
+                    2113,
+                    834,
+                    2113
+                ],
+                "score": 0.857
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    303,
+                    1140,
+                    1392,
+                    1140,
+                    1392,
+                    1676,
+                    303,
+                    1676
+                ],
+                "score": 0.794
+            },
+            {
+                "category_id": 6,
+                "poly": [
+                    297,
+                    1103,
+                    1015,
+                    1103,
+                    1015,
+                    1137,
+                    297,
+                    1137
+                ],
+                "score": 0.53
+            },
+            {
+                "category_id": 5,
+                "poly": [
+                    303,
+                    1140,
+                    1392,
+                    1140,
+                    1392,
+                    1676,
+                    303,
+                    1676
+                ],
+                "score": 0.226,
+                "html": "<table><tr><td colspan=\"2\">Input: Offline dataset D, previous policy π</td></tr><tr><td colspan=\"2\">1: Sample K independent datasets with replacement from D.</td></tr><tr><td>2:</td><td>Train bootstrapped ensemble of K models {Ti, Ri}Ko minimizing Eq. (8) (adapted with d).</td></tr><tr><td>3:</td><td>for repeat = 0,1,...do</td></tr><tr><td>4:</td><td>for rollout= 0,1,...,B do</td></tr><tr><td>5:</td><td>Sample initial rollout state so from D.</td></tr><tr><td>6:</td><td>for t= 0,1,...do</td></tr><tr><td>7:</td><td>Sample an action at ~ π(st).</td></tr><tr><td>8:</td><td>Randomly pick (Ti,Ri) and sample (St+1,rt) ~ (Ti,Ri)(st, at). and store (St, at,rt, St+1) in D.</td></tr><tr><td>9:</td><td>Compute rt = rt-γ&gt;|√Vκ[μ(st,at)]||2</td></tr><tr><td>10: 11:</td><td>end for</td></tr><tr><td>12:</td><td>end for Draw samples from D to compute IPM(dπ, d).</td></tr><tr><td>13:</td><td>Draw samples from D and D to update critic Q.</td></tr><tr><td>14:</td><td>Maximize Es~DuD,a~π(s)[Q(s,a) - T log π(a|s)] - απIPM(d,dD) to update π.</td></tr><tr><td></td><td></td></tr></table>"
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    297,
+                    1103,
+                    1015,
+                    1103,
+                    1015,
+                    1137,
+                    297,
+                    1137
+                ],
+                "score": 0.222
+            },
+            {
+                "category_id": 7,
+                "poly": [
+                    300,
+                    1664,
+                    433,
+                    1664,
+                    433,
+                    1693,
+                    300,
+                    1693
+                ],
+                "score": 0.12
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    822,
+                    1207,
+                    957,
+                    1207,
+                    957,
+                    1248,
+                    822,
+                    1248
+                ],
+                "score": 0.94,
+                "latex": "\\{ \\widehat { T } _ { i } , \\widehat { R } _ { i } \\} _ { i = 0 } ^ { K }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    771,
+                    1542,
+                    924,
+                    1542,
+                    924,
+                    1587,
+                    771,
+                    1587
+                ],
+                "score": 0.93,
+                "latex": "\\widehat { \\mathrm { I P M } } ( \\tilde { d } _ { \\phi } ^ { \\pi } , d _ { \\phi } ^ { D } )"
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    310,
+                    579,
+                    1406,
+                    579,
+                    1406,
+                    677,
+                    310,
+                    677
+                ],
+                "score": 0.92,
+                "latex": "\\mathrm { P M } _ { \\mathcal { G } } ( \\tilde { d } _ { \\phi } ^ { \\pi } , d _ { \\phi } ^ { \\mathcal { D } } ) = \\operatorname* { s u p } _ { \\nu \\in \\mathcal { F } } \\left| ( 1 - \\gamma ) \\mathbb { E } _ { \\alpha \\sim \\pi ( s ) } \\left[ \\nu \\big ( \\phi ( s , a ) \\big ) \\right] - \\mathbb { E } _ { ( s , a , s ^ { \\prime } ) \\sim d ^ { \\mathcal { D } } } \\left[ \\nu \\big ( \\phi ( s , a ) \\big ) - \\gamma \\nu \\big ( \\phi ( s ^ { \\prime } , a ^ { \\prime } ) \\big ) \\right] \\right| ,"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    298,
+                    353,
+                    740,
+                    353,
+                    740,
+                    392,
+                    298,
+                    392
+                ],
+                "score": 0.91,
+                "latex": "\\begin{array} { r } { \\tilde { R } ^ { \\pi } = ( 1 - \\gamma ) \\mathbb { E } , M , \\pi \\bigl [ \\sum _ { t = 0 } ^ { \\infty } \\gamma ^ { t } r _ { t } \\bigr ] } \\end{array}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    647,
+                    1400,
+                    738,
+                    1400,
+                    738,
+                    1439,
+                    647,
+                    1439
+                ],
+                "score": 0.91,
+                "latex": "( \\widehat { T } _ { i } , \\widehat { R } _ { i } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    667,
+                    1368,
+                    791,
+                    1368,
+                    791,
+                    1402,
+                    667,
+                    1402
+                ],
+                "score": 0.91,
+                "latex": "a _ { t } \\sim \\pi ( s _ { t } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    577,
+                    1438,
+                    955,
+                    1438,
+                    955,
+                    1492,
+                    577,
+                    1492
+                ],
+                "score": 0.9,
+                "latex": " \\tilde { r } _ { t } = r _ { t } - \\gamma \\lambda \\sqrt { \\mathbb { V } _ { K } [ \\mu ( s _ { t } , a _ { t } ) ] } | | _ { 2 }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    914,
+                    389,
+                    950,
+                    389,
+                    950,
+                    419,
+                    914,
+                    419
+                ],
+                "score": 0.88,
+                "latex": "d ^ { \\mathcal { D } }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    504,
+                    1621,
+                    1168,
+                    1621,
+                    1168,
+                    1667,
+                    504,
+                    1667
+                ],
+                "score": 0.88,
+                "latex": "\\mathbb { E } _ { s \\sim \\mathcal { D } \\cup \\widehat { \\mathcal { D } } , a \\sim \\pi ( s ) } [ Q ( s , a ) - \\tau \\log \\pi ( a | s ) ] - \\alpha _ { \\pi } \\widehat { \\mathrm { I P M } } ( \\tilde { d } _ { \\phi } ^ { \\pi } , d _ { \\phi } ^ { \\mathcal { D } } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    715,
+                    434,
+                    753,
+                    434,
+                    753,
+                    466,
+                    715,
+                    466
+                ],
+                "score": 0.88,
+                "latex": "{ \\tilde { R } } ^ { \\pi }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    904,
+                    693,
+                    940,
+                    693,
+                    940,
+                    723,
+                    904,
+                    723
+                ],
+                "score": 0.88,
+                "latex": "d ^ { \\mathcal { D } }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    489,
+                    393,
+                    519,
+                    393,
+                    519,
+                    422,
+                    489,
+                    422
+                ],
+                "score": 0.88,
+                "latex": "d _ { 0 }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    876,
+                    1399,
+                    1194,
+                    1399,
+                    1194,
+                    1439,
+                    876,
+                    1439
+                ],
+                "score": 0.88,
+                "latex": "( s _ { t + 1 } , r _ { t } ) \\sim ( \\widehat { T } _ { i } , \\widehat { R } _ { i } ) ( s _ { t } , a _ { t } ) ."
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    865,
+                    357,
+                    903,
+                    357,
+                    903,
+                    385,
+                    865,
+                    385
+                ],
+                "score": 0.87,
+                "latex": "R ^ { \\pi }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    644,
+                    470,
+                    1352,
+                    470,
+                    1352,
+                    508,
+                    644,
+                    508
+                ],
+                "score": 0.87,
+                "latex": "\\begin{array} { r } { \\tilde { d } ^ { \\pi } ( s , a ) \\triangleq ( 1 - \\gamma ) \\sum _ { t = 0 } ^ { \\infty } \\gamma ^ { t } \\operatorname* { P r } ( s _ { t } = s , a _ { t } = a | s _ { 0 } \\sim d ^ { \\mathcal { D } } , T , \\pi ) } \\end{array}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1081,
+                    262,
+                    1120,
+                    262,
+                    1120,
+                    290,
+                    1081,
+                    290
+                ],
+                "score": 0.87,
+                "latex": "R ^ { \\pi }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1071,
+                    1444,
+                    1250,
+                    1444,
+                    1250,
+                    1480,
+                    1071,
+                    1480
+                ],
+                "score": 0.86,
+                "latex": "\\left( { { s _ { t } } , { a _ { t } } , \\tilde { r } _ { t } , { s _ { t + 1 } } } \\right)"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1337,
+                    1207,
+                    1372,
+                    1207,
+                    1372,
+                    1253,
+                    1337,
+                    1253
+                ],
+                "score": 0.86,
+                "latex": "\\tilde { d } _ { \\phi } ^ { \\pi } ,"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    686,
+                    1584,
+                    713,
+                    1584,
+                    713,
+                    1617,
+                    686,
+                    1617
+                ],
+                "score": 0.85,
+                "latex": "\\widehat { \\mathcal { D } }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    889,
+                    1591,
+                    913,
+                    1591,
+                    913,
+                    1621,
+                    889,
+                    1621
+                ],
+                "score": 0.83,
+                "latex": "Q"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    820,
+                    1309,
+                    845,
+                    1309,
+                    845,
+                    1335,
+                    820,
+                    1335
+                ],
+                "score": 0.82,
+                "latex": "\\mathcal { D }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1283,
+                    1440,
+                    1308,
+                    1440,
+                    1308,
+                    1474,
+                    1283,
+                    1474
+                ],
+                "score": 0.82,
+                "latex": "\\widehat { \\mathcal { D } }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    727,
+                    1313,
+                    755,
+                    1313,
+                    755,
+                    1338,
+                    727,
+                    1338
+                ],
+                "score": 0.82,
+                "latex": "s _ { 0 }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    702,
+                    1214,
+                    731,
+                    1214,
+                    731,
+                    1241,
+                    702,
+                    1241
+                ],
+                "score": 0.82,
+                "latex": "K"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    434,
+                    1178,
+                    464,
+                    1178,
+                    464,
+                    1206,
+                    434,
+                    1206
+                ],
+                "score": 0.8,
+                "latex": "K"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    610,
+                    1549,
+                    636,
+                    1549,
+                    636,
+                    1576,
+                    610,
+                    1576
+                ],
+                "score": 0.8,
+                "latex": "\\mathcal { D }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    610,
+                    1590,
+                    636,
+                    1590,
+                    636,
+                    1617,
+                    610,
+                    1617
+                ],
+                "score": 0.8,
+                "latex": "\\mathcal { D }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    597,
+                    1143,
+                    621,
+                    1143,
+                    621,
+                    1169,
+                    597,
+                    1169
+                ],
+                "score": 0.78,
+                "latex": "\\mathcal { D }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    673,
+                    510,
+                    693,
+                    510,
+                    693,
+                    531,
+                    673,
+                    531
+                ],
+                "score": 0.76,
+                "latex": "\\pi"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    963,
+                    1178,
+                    987,
+                    1178,
+                    987,
+                    1205,
+                    963,
+                    1205
+                ],
+                "score": 0.73,
+                "latex": "\\mathcal { D }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    435,
+                    1247,
+                    587,
+                    1247,
+                    587,
+                    1278,
+                    435,
+                    1278
+                ],
+                "score": 0.7,
+                "latex": "{ \\mathcal { a } } t = 0 , 1 , \\ldots"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    810,
+                    1148,
+                    831,
+                    1148,
+                    831,
+                    1169,
+                    810,
+                    1169
+                ],
+                "score": 0.68,
+                "latex": "\\pi"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    468,
+                    1338,
+                    605,
+                    1338,
+                    605,
+                    1369,
+                    468,
+                    1369
+                ],
+                "score": 0.65,
+                "latex": "t = 0 , 1 , \\ldots"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1281,
+                    1633,
+                    1300,
+                    1633,
+                    1300,
+                    1654,
+                    1281,
+                    1654
+                ],
+                "score": 0.64,
+                "latex": "\\pi"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    515,
+                    1279,
+                    675,
+                    1279,
+                    675,
+                    1309,
+                    515,
+                    1309
+                ],
+                "score": 0.49,
+                "latex": "\\mathbf { \\Sigma } = 0 , 1 , \\dots , B"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1184,
+                    1148,
+                    1205,
+                    1148,
+                    1205,
+                    1170,
+                    1184,
+                    1170
+                ],
+                "score": 0.48,
+                "latex": "\\pi"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    330.0,
+                    1999.0,
+                    872.0,
+                    1999.0,
+                    872.0,
+                    2041.0,
+                    330.0,
+                    2041.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    73.0,
+                    815.0,
+                    73.0,
+                    815.0,
+                    108.0,
+                    295.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    830.0,
+                    2084.0,
+                    871.0,
+                    2084.0,
+                    871.0,
+                    2121.0,
+                    830.0,
+                    2121.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    1101.0,
+                    1017.0,
+                    1101.0,
+                    1017.0,
+                    1141.0,
+                    297.0,
+                    1141.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    1101.0,
+                    1017.0,
+                    1101.0,
+                    1017.0,
+                    1141.0,
+                    297.0,
+                    1141.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    298.0,
+                    1659.0,
+                    438.0,
+                    1659.0,
+                    438.0,
+                    1696.0,
+                    298.0,
+                    1696.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    760.0,
+                    1405.0,
+                    760.0,
+                    1405.0,
+                    797.0,
+                    294.0,
+                    797.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    795.0,
+                    1403.0,
+                    795.0,
+                    1403.0,
+                    827.0,
+                    295.0,
+                    827.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    822.0,
+                    1405.0,
+                    822.0,
+                    1405.0,
+                    859.0,
+                    291.0,
+                    859.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    853.0,
+                    1406.0,
+                    853.0,
+                    1406.0,
+                    891.0,
+                    291.0,
+                    891.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    885.0,
+                    1406.0,
+                    885.0,
+                    1406.0,
+                    920.0,
+                    294.0,
+                    920.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    915.0,
+                    1405.0,
+                    915.0,
+                    1405.0,
+                    951.0,
+                    294.0,
+                    951.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    946.0,
+                    1406.0,
+                    946.0,
+                    1406.0,
+                    978.0,
+                    295.0,
+                    978.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    973.0,
+                    1405.0,
+                    973.0,
+                    1405.0,
+                    1011.0,
+                    293.0,
+                    1011.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1006.0,
+                    1406.0,
+                    1006.0,
+                    1406.0,
+                    1040.0,
+                    294.0,
+                    1040.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1036.0,
+                    1061.0,
+                    1036.0,
+                    1061.0,
+                    1071.0,
+                    294.0,
+                    1071.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    229.0,
+                    1405.0,
+                    229.0,
+                    1405.0,
+                    266.0,
+                    295.0,
+                    266.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    259.0,
+                    1080.0,
+                    259.0,
+                    1080.0,
+                    296.0,
+                    295.0,
+                    296.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1121.0,
+                    259.0,
+                    1405.0,
+                    259.0,
+                    1405.0,
+                    296.0,
+                    1121.0,
+                    296.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    291.0,
+                    1405.0,
+                    291.0,
+                    1405.0,
+                    327.0,
+                    295.0,
+                    327.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    318.0,
+                    1407.0,
+                    318.0,
+                    1407.0,
+                    360.0,
+                    292.0,
+                    360.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    286.0,
+                    338.0,
+                    297.0,
+                    338.0,
+                    297.0,
+                    407.0,
+                    286.0,
+                    407.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    741.0,
+                    338.0,
+                    864.0,
+                    338.0,
+                    864.0,
+                    407.0,
+                    741.0,
+                    407.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    904.0,
+                    338.0,
+                    1415.0,
+                    338.0,
+                    1415.0,
+                    407.0,
+                    904.0,
+                    407.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    389.0,
+                    488.0,
+                    389.0,
+                    488.0,
+                    424.0,
+                    292.0,
+                    424.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    520.0,
+                    389.0,
+                    913.0,
+                    389.0,
+                    913.0,
+                    424.0,
+                    520.0,
+                    424.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    951.0,
+                    389.0,
+                    960.0,
+                    389.0,
+                    960.0,
+                    424.0,
+                    951.0,
+                    424.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    436.0,
+                    714.0,
+                    436.0,
+                    714.0,
+                    472.0,
+                    295.0,
+                    472.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    754.0,
+                    436.0,
+                    1405.0,
+                    436.0,
+                    1405.0,
+                    472.0,
+                    754.0,
+                    472.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    283.0,
+                    454.0,
+                    643.0,
+                    454.0,
+                    643.0,
+                    536.0,
+                    283.0,
+                    536.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1353.0,
+                    454.0,
+                    1415.0,
+                    454.0,
+                    1415.0,
+                    536.0,
+                    1353.0,
+                    536.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    534.0,
+                    837.0,
+                    534.0,
+                    837.0,
+                    567.0,
+                    295.0,
+                    567.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1732.0,
+                    1405.0,
+                    1732.0,
+                    1405.0,
+                    1770.0,
+                    293.0,
+                    1770.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1763.0,
+                    1405.0,
+                    1763.0,
+                    1405.0,
+                    1801.0,
+                    293.0,
+                    1801.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1793.0,
+                    1406.0,
+                    1793.0,
+                    1406.0,
+                    1832.0,
+                    293.0,
+                    1832.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1825.0,
+                    1402.0,
+                    1825.0,
+                    1402.0,
+                    1860.0,
+                    294.0,
+                    1860.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1853.0,
+                    1405.0,
+                    1853.0,
+                    1405.0,
+                    1891.0,
+                    291.0,
+                    1891.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1887.0,
+                    1405.0,
+                    1887.0,
+                    1405.0,
+                    1920.0,
+                    295.0,
+                    1920.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1918.0,
+                    1404.0,
+                    1918.0,
+                    1404.0,
+                    1948.0,
+                    296.0,
+                    1948.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1949.0,
+                    1404.0,
+                    1949.0,
+                    1404.0,
+                    1979.0,
+                    295.0,
+                    1979.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    690.0,
+                    903.0,
+                    690.0,
+                    903.0,
+                    733.0,
+                    292.0,
+                    733.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    941.0,
+                    690.0,
+                    1281.0,
+                    690.0,
+                    1281.0,
+                    733.0,
+                    941.0,
+                    733.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    345.0,
+                    1141.0,
+                    596.0,
+                    1141.0,
+                    596.0,
+                    1176.0,
+                    345.0,
+                    1176.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    622.0,
+                    1141.0,
+                    809.0,
+                    1141.0,
+                    809.0,
+                    1176.0,
+                    622.0,
+                    1176.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    832.0,
+                    1141.0,
+                    835.0,
+                    1141.0,
+                    835.0,
+                    1176.0,
+                    832.0,
+                    1176.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    880.0,
+                    1141.0,
+                    1183.0,
+                    1141.0,
+                    1183.0,
+                    1177.0,
+                    880.0,
+                    1177.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    309.0,
+                    1176.0,
+                    433.0,
+                    1176.0,
+                    433.0,
+                    1210.0,
+                    309.0,
+                    1210.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    465.0,
+                    1176.0,
+                    962.0,
+                    1176.0,
+                    962.0,
+                    1210.0,
+                    465.0,
+                    1210.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    988.0,
+                    1176.0,
+                    999.0,
+                    1176.0,
+                    999.0,
+                    1210.0,
+                    988.0,
+                    1210.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    301.0,
+                    1200.0,
+                    701.0,
+                    1200.0,
+                    701.0,
+                    1254.0,
+                    301.0,
+                    1254.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    732.0,
+                    1200.0,
+                    821.0,
+                    1200.0,
+                    821.0,
+                    1254.0,
+                    732.0,
+                    1254.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    958.0,
+                    1200.0,
+                    1336.0,
+                    1200.0,
+                    1336.0,
+                    1254.0,
+                    958.0,
+                    1254.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1373.0,
+                    1200.0,
+                    1390.0,
+                    1200.0,
+                    1390.0,
+                    1254.0,
+                    1373.0,
+                    1254.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    310.0,
+                    1247.0,
+                    434.0,
+                    1247.0,
+                    434.0,
+                    1278.0,
+                    310.0,
+                    1278.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    588.0,
+                    1247.0,
+                    630.0,
+                    1247.0,
+                    630.0,
+                    1278.0,
+                    588.0,
+                    1278.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    307.0,
+                    1278.0,
+                    338.0,
+                    1278.0,
+                    338.0,
+                    1310.0,
+                    307.0,
+                    1310.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    383.0,
+                    1274.0,
+                    514.0,
+                    1274.0,
+                    514.0,
+                    1309.0,
+                    383.0,
+                    1309.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    676.0,
+                    1274.0,
+                    715.0,
+                    1274.0,
+                    715.0,
+                    1309.0,
+                    676.0,
+                    1309.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    307.0,
+                    1307.0,
+                    338.0,
+                    1307.0,
+                    338.0,
+                    1340.0,
+                    307.0,
+                    1340.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    425.0,
+                    1305.0,
+                    726.0,
+                    1305.0,
+                    726.0,
+                    1340.0,
+                    425.0,
+                    1340.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    756.0,
+                    1305.0,
+                    819.0,
+                    1305.0,
+                    819.0,
+                    1340.0,
+                    756.0,
+                    1340.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    846.0,
+                    1305.0,
+                    855.0,
+                    1305.0,
+                    855.0,
+                    1340.0,
+                    846.0,
+                    1340.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    307.0,
+                    1337.0,
+                    338.0,
+                    1337.0,
+                    338.0,
+                    1371.0,
+                    307.0,
+                    1371.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    423.0,
+                    1333.0,
+                    467.0,
+                    1333.0,
+                    467.0,
+                    1373.0,
+                    423.0,
+                    1373.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    606.0,
+                    1333.0,
+                    645.0,
+                    1333.0,
+                    645.0,
+                    1373.0,
+                    606.0,
+                    1373.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    307.0,
+                    1368.0,
+                    338.0,
+                    1368.0,
+                    338.0,
+                    1403.0,
+                    307.0,
+                    1403.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    466.0,
+                    1367.0,
+                    666.0,
+                    1367.0,
+                    666.0,
+                    1403.0,
+                    466.0,
+                    1403.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    792.0,
+                    1367.0,
+                    800.0,
+                    1367.0,
+                    800.0,
+                    1403.0,
+                    792.0,
+                    1403.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    307.0,
+                    1407.0,
+                    338.0,
+                    1407.0,
+                    338.0,
+                    1440.0,
+                    307.0,
+                    1440.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    466.0,
+                    1398.0,
+                    646.0,
+                    1398.0,
+                    646.0,
+                    1440.0,
+                    466.0,
+                    1440.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    739.0,
+                    1398.0,
+                    875.0,
+                    1398.0,
+                    875.0,
+                    1440.0,
+                    739.0,
+                    1440.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1195.0,
+                    1398.0,
+                    1202.0,
+                    1398.0,
+                    1202.0,
+                    1440.0,
+                    1195.0,
+                    1440.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    307.0,
+                    1447.0,
+                    337.0,
+                    1447.0,
+                    337.0,
+                    1483.0,
+                    307.0,
+                    1483.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    465.0,
+                    1439.0,
+                    576.0,
+                    1439.0,
+                    576.0,
+                    1485.0,
+                    465.0,
+                    1485.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    956.0,
+                    1441.0,
+                    1070.0,
+                    1441.0,
+                    1070.0,
+                    1488.0,
+                    956.0,
+                    1488.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1251.0,
+                    1441.0,
+                    1282.0,
+                    1441.0,
+                    1282.0,
+                    1488.0,
+                    1251.0,
+                    1488.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1309.0,
+                    1441.0,
+                    1319.0,
+                    1441.0,
+                    1319.0,
+                    1488.0,
+                    1309.0,
+                    1488.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1486.0,
+                    338.0,
+                    1486.0,
+                    338.0,
+                    1520.0,
+                    296.0,
+                    1520.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    423.0,
+                    1484.0,
+                    522.0,
+                    1484.0,
+                    522.0,
+                    1521.0,
+                    423.0,
+                    1521.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1517.0,
+                    338.0,
+                    1517.0,
+                    338.0,
+                    1552.0,
+                    296.0,
+                    1552.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    381.0,
+                    1512.0,
+                    480.0,
+                    1512.0,
+                    480.0,
+                    1552.0,
+                    381.0,
+                    1552.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1548.0,
+                    338.0,
+                    1548.0,
+                    338.0,
+                    1583.0,
+                    296.0,
+                    1583.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    381.0,
+                    1542.0,
+                    609.0,
+                    1542.0,
+                    609.0,
+                    1586.0,
+                    381.0,
+                    1586.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    637.0,
+                    1542.0,
+                    770.0,
+                    1542.0,
+                    770.0,
+                    1586.0,
+                    637.0,
+                    1586.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    925.0,
+                    1542.0,
+                    933.0,
+                    1542.0,
+                    933.0,
+                    1586.0,
+                    925.0,
+                    1586.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1588.0,
+                    338.0,
+                    1588.0,
+                    338.0,
+                    1623.0,
+                    296.0,
+                    1623.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    382.0,
+                    1584.0,
+                    609.0,
+                    1584.0,
+                    609.0,
+                    1625.0,
+                    382.0,
+                    1625.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    637.0,
+                    1584.0,
+                    685.0,
+                    1584.0,
+                    685.0,
+                    1625.0,
+                    637.0,
+                    1625.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    714.0,
+                    1584.0,
+                    888.0,
+                    1584.0,
+                    888.0,
+                    1625.0,
+                    714.0,
+                    1625.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    914.0,
+                    1584.0,
+                    924.0,
+                    1584.0,
+                    924.0,
+                    1625.0,
+                    914.0,
+                    1625.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1627.0,
+                    338.0,
+                    1627.0,
+                    338.0,
+                    1660.0,
+                    295.0,
+                    1660.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    377.0,
+                    1618.0,
+                    503.0,
+                    1618.0,
+                    503.0,
+                    1673.0,
+                    377.0,
+                    1673.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1169.0,
+                    1618.0,
+                    1280.0,
+                    1618.0,
+                    1280.0,
+                    1673.0,
+                    1169.0,
+                    1673.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1301.0,
+                    1618.0,
+                    1316.0,
+                    1618.0,
+                    1316.0,
+                    1673.0,
+                    1301.0,
+                    1673.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1660.0,
+                    336.0,
+                    1660.0,
+                    336.0,
+                    1686.0,
+                    296.0,
+                    1686.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    348.0,
+                    1658.0,
+                    435.0,
+                    1658.0,
+                    435.0,
+                    1686.0,
+                    348.0,
+                    1686.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 19,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 5,
+                "poly": [
+                    298,
+                    844,
+                    1398,
+                    844,
+                    1398,
+                    1279,
+                    298,
+                    1279
+                ],
+                "score": 0.976,
+                "html": "<table><tr><td>Dataset type</td><td>Environment</td><td>RepB-SDE (ours)</td><td>RP</td><td>Base</td><td>MOPO</td></tr><tr><td></td><td></td><td></td><td></td><td></td><td></td></tr><tr><td>Random</td><td>Walker2d-v2</td><td>21.1 ± 1.0</td><td>18.4 ±1.8</td><td>16.4 ± 2.2</td><td>1.3 ± 0.5</td></tr><tr><td>Medium</td><td>Walker2d-v2</td><td>72.1 ± 1.9</td><td>56.3 ± 4.3</td><td>5.5 ± 0.4</td><td>-0.1 ± 0.0</td></tr><tr><td>Med-Replay</td><td>Walker2d-v2</td><td>49.8 ± 11.4</td><td>41.1 ±10.3</td><td>6.2 ± 0.7</td><td>47.8 ± 4.8</td></tr><tr><td>Med-Expert</td><td>Walker2d-v2</td><td>88.8 ± 6.9</td><td>72.6 ± 5.1</td><td>51.7 ± 7.1</td><td>32.4 ± 8.8</td></tr><tr><td>Random</td><td>Hopper-v2</td><td>8.6 ±1.0</td><td>8.3 ±0.2</td><td>8.3 ±0.2</td><td>9.1 ± 0.3</td></tr><tr><td>Medium</td><td>Hopper-v2</td><td>34.0 ± 2.8</td><td>27.5 ± 3.1</td><td>19.8 ± 0.8</td><td>19.2 ± 4.5</td></tr><tr><td>Med-Replay</td><td>Hopper-v2</td><td>62.2 ± 6.7</td><td>49.8 ± 8.7</td><td>32.9 ± 5.5</td><td>80.8 ± 2.9</td></tr><tr><td>Med-Expert</td><td>Hopper-v2</td><td>82.6 ± 7.0</td><td>74.0 ± 7.2</td><td>19.1 ± 1.1</td><td>23.2 ± 1.3</td></tr><tr><td>Random</td><td>HalfCheetah-v2</td><td>32.9 ± 1.1</td><td>31.3 ±1.5</td><td>26.1 ± 6.2</td><td>29.9 ± 1.2</td></tr><tr><td>Medium</td><td>HalfCheetah-v2</td><td>49.1 ± 0.3</td><td>47.3 ± 0.3</td><td>22.1 ± 4.4</td><td>45.0 ± 0.4</td></tr><tr><td>Med-Replay</td><td>HalfCheetah-v2</td><td>57.5 ± 0.8</td><td>53.6 ± 0.8</td><td>54.0 ± 2.4</td><td>53.8 ± 1.1</td></tr><tr><td>Med-Expert</td><td>HalfCheetah-v2</td><td>55.4 ± 8.3</td><td>36.7 ± 12.0</td><td>31.9 ± 10.5</td><td>91.0 ± 2.5</td></tr></table>"
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1350,
+                    1404,
+                    1350,
+                    1404,
+                    1505,
+                    298,
+                    1505
+                ],
+                "score": 0.974
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1520,
+                    1404,
+                    1520,
+                    1404,
+                    1644,
+                    298,
+                    1644
+                ],
+                "score": 0.973
+            },
+            {
+                "category_id": 5,
+                "poly": [
+                    407,
+                    292,
+                    1295,
+                    292,
+                    1295,
+                    703,
+                    407,
+                    703
+                ],
+                "score": 0.968,
+                "html": "<table><tr><td>Parameter Value</td></tr><tr><td>optimizer Adam (Kingma &amp; Ba, 2014)</td></tr><tr><td>learning rate 3×10-4</td></tr><tr><td>discount factor y 0.99</td></tr><tr><td>number of samples per minibatch 256</td></tr><tr><td>target smoothing coefficient T 5×10-3</td></tr><tr><td>[actor/critic] number of hidden layers 2</td></tr><tr><td>[actor/critic] number of hidden units per layer 256</td></tr><tr><td>[actor/critic] non-linearity ReLU</td></tr><tr><td># of rollouts 104</td></tr><tr><td>max length of rollouts 103</td></tr><tr><td>rollout buffer size 5×107</td></tr></table>"
+            },
+            {
+                "category_id": 6,
+                "poly": [
+                    298,
+                    745,
+                    1402,
+                    745,
+                    1402,
+                    809,
+                    298,
+                    809
+                ],
+                "score": 0.934
+            },
+            {
+                "category_id": 6,
+                "poly": [
+                    471,
+                    224,
+                    1225,
+                    224,
+                    1225,
+                    258,
+                    471,
+                    258
+                ],
+                "score": 0.897
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    299,
+                    76,
+                    812,
+                    76,
+                    812,
+                    104,
+                    299,
+                    104
+                ],
+                "score": 0.889
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    834,
+                    2088,
+                    863,
+                    2088,
+                    863,
+                    2113,
+                    834,
+                    2113
+                ],
+                "score": 0.808
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    297,
+                    1412,
+                    602,
+                    1412,
+                    602,
+                    1445,
+                    297,
+                    1445
+                ],
+                "score": 0.92,
+                "latex": "\\alpha _ { \\pi } ~ \\in ~ \\{ 0 , 0 . 0 1 , 0 . 1 , 1 , 1 0 \\}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    970,
+                    1351,
+                    1229,
+                    1351,
+                    1229,
+                    1384,
+                    970,
+                    1384
+                ],
+                "score": 0.92,
+                "latex": "\\lambda \\in \\{ 0 , 2 , 5 , 7 , 1 0 , 1 5 \\}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1075,
+                    1613,
+                    1265,
+                    1613,
+                    1265,
+                    1643,
+                    1075,
+                    1643
+                ],
+                "score": 0.91,
+                "latex": "\\alpha _ { M } = 0 , \\alpha _ { \\pi } = 0"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    844,
+                    1442,
+                    962,
+                    1442,
+                    962,
+                    1472,
+                    844,
+                    1472
+                ],
+                "score": 0.91,
+                "latex": "5 . 0 \\times 1 0 ^ { 6 }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    946,
+                    1411,
+                    1066,
+                    1411,
+                    1066,
+                    1442,
+                    946,
+                    1442
+                ],
+                "score": 0.91,
+                "latex": "1 . 5 \\times 1 0 ^ { 6 }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1158,
+                    1382,
+                    1393,
+                    1382,
+                    1393,
+                    1415,
+                    1158,
+                    1415
+                ],
+                "score": 0.91,
+                "latex": "\\alpha _ { M } \\in \\{ 0 . 0 1 , 0 . 1 , 1 \\}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1188,
+                    1586,
+                    1279,
+                    1586,
+                    1279,
+                    1613,
+                    1188,
+                    1613
+                ],
+                "score": 0.89,
+                "latex": "\\alpha _ { M } , \\alpha _ { \\pi }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1137,
+                    1554,
+                    1182,
+                    1554,
+                    1182,
+                    1581,
+                    1137,
+                    1581
+                ],
+                "score": 0.88,
+                "latex": "\\alpha _ { M }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    401,
+                    1617,
+                    446,
+                    1617,
+                    446,
+                    1643,
+                    401,
+                    1643
+                ],
+                "score": 0.87,
+                "latex": "\\alpha _ { M }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1232,
+                    1554,
+                    1267,
+                    1554,
+                    1267,
+                    1581,
+                    1232,
+                    1581
+                ],
+                "score": 0.86,
+                "latex": "\\alpha _ { \\pi }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    590,
+                    427,
+                    611,
+                    427,
+                    611,
+                    453,
+                    590,
+                    453
+                ],
+                "score": 0.72,
+                "latex": "\\gamma"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    744,
+                    1383,
+                    764,
+                    1383,
+                    764,
+                    1410,
+                    744,
+                    1410
+                ],
+                "score": 0.68,
+                "latex": "\\lambda"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    734,
+                    488,
+                    754,
+                    488,
+                    754,
+                    510,
+                    734,
+                    510
+                ],
+                "score": 0.55,
+                "latex": "\\tau"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1235,
+                    603,
+                    1280,
+                    603,
+                    1280,
+                    633,
+                    1235,
+                    633
+                ],
+                "score": 0.46,
+                "latex": "1 0 ^ { 4 }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1235,
+                    635,
+                    1280,
+                    635,
+                    1280,
+                    664,
+                    1235,
+                    664
+                ],
+                "score": 0.39,
+                "latex": "1 0 ^ { 3 }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1186,
+                    666,
+                    1280,
+                    666,
+                    1280,
+                    699,
+                    1186,
+                    699
+                ],
+                "score": 0.26,
+                "latex": "5 \\times 1 0 ^ { 7 }"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    745.0,
+                    1402.0,
+                    745.0,
+                    1402.0,
+                    777.0,
+                    296.0,
+                    777.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    776.0,
+                    541.0,
+                    776.0,
+                    541.0,
+                    810.0,
+                    293.0,
+                    810.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    470.0,
+                    219.0,
+                    1228.0,
+                    219.0,
+                    1228.0,
+                    264.0,
+                    470.0,
+                    264.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    73.0,
+                    815.0,
+                    73.0,
+                    815.0,
+                    108.0,
+                    295.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    830.0,
+                    2085.0,
+                    868.0,
+                    2085.0,
+                    868.0,
+                    2124.0,
+                    830.0,
+                    2124.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1349.0,
+                    969.0,
+                    1349.0,
+                    969.0,
+                    1386.0,
+                    293.0,
+                    1386.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1230.0,
+                    1349.0,
+                    1405.0,
+                    1349.0,
+                    1405.0,
+                    1386.0,
+                    1230.0,
+                    1386.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1378.0,
+                    743.0,
+                    1378.0,
+                    743.0,
+                    1418.0,
+                    293.0,
+                    1418.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    765.0,
+                    1378.0,
+                    1157.0,
+                    1378.0,
+                    1157.0,
+                    1418.0,
+                    765.0,
+                    1418.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1394.0,
+                    1378.0,
+                    1405.0,
+                    1378.0,
+                    1405.0,
+                    1418.0,
+                    1394.0,
+                    1418.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    289.0,
+                    1410.0,
+                    296.0,
+                    1410.0,
+                    296.0,
+                    1449.0,
+                    289.0,
+                    1449.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    603.0,
+                    1410.0,
+                    945.0,
+                    1410.0,
+                    945.0,
+                    1449.0,
+                    603.0,
+                    1449.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1067.0,
+                    1410.0,
+                    1407.0,
+                    1410.0,
+                    1407.0,
+                    1449.0,
+                    1067.0,
+                    1449.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1435.0,
+                    843.0,
+                    1435.0,
+                    843.0,
+                    1479.0,
+                    292.0,
+                    1479.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    963.0,
+                    1435.0,
+                    1406.0,
+                    1435.0,
+                    1406.0,
+                    1479.0,
+                    963.0,
+                    1479.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1474.0,
+                    886.0,
+                    1474.0,
+                    886.0,
+                    1507.0,
+                    296.0,
+                    1507.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1520.0,
+                    1401.0,
+                    1520.0,
+                    1401.0,
+                    1553.0,
+                    294.0,
+                    1553.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1550.0,
+                    1136.0,
+                    1550.0,
+                    1136.0,
+                    1583.0,
+                    294.0,
+                    1583.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1183.0,
+                    1550.0,
+                    1231.0,
+                    1550.0,
+                    1231.0,
+                    1583.0,
+                    1183.0,
+                    1583.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1268.0,
+                    1550.0,
+                    1404.0,
+                    1550.0,
+                    1404.0,
+                    1583.0,
+                    1268.0,
+                    1583.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1578.0,
+                    1187.0,
+                    1578.0,
+                    1187.0,
+                    1617.0,
+                    292.0,
+                    1617.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1280.0,
+                    1578.0,
+                    1405.0,
+                    1578.0,
+                    1405.0,
+                    1617.0,
+                    1280.0,
+                    1617.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1611.0,
+                    400.0,
+                    1611.0,
+                    400.0,
+                    1646.0,
+                    292.0,
+                    1646.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    447.0,
+                    1611.0,
+                    1074.0,
+                    1611.0,
+                    1074.0,
+                    1646.0,
+                    447.0,
+                    1646.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1266.0,
+                    1611.0,
+                    1278.0,
+                    1611.0,
+                    1278.0,
+                    1646.0,
+                    1266.0,
+                    1646.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 20,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 3,
+                "poly": [
+                    295,
+                    278,
+                    1398,
+                    278,
+                    1398,
+                    1182,
+                    295,
+                    1182
+                ],
+                "score": 0.973
+            },
+            {
+                "category_id": 3,
+                "poly": [
+                    294,
+                    1370,
+                    1408,
+                    1370,
+                    1408,
+                    1912,
+                    294,
+                    1912
+                ],
+                "score": 0.961
+            },
+            {
+                "category_id": 4,
+                "poly": [
+                    537,
+                    1201,
+                    1157,
+                    1201,
+                    1157,
+                    1235,
+                    537,
+                    1235
+                ],
+                "score": 0.925
+            },
+            {
+                "category_id": 4,
+                "poly": [
+                    443,
+                    1932,
+                    1254,
+                    1932,
+                    1254,
+                    1968,
+                    443,
+                    1968
+                ],
+                "score": 0.924
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    299,
+                    76,
+                    812,
+                    76,
+                    812,
+                    104,
+                    299,
+                    104
+                ],
+                "score": 0.893
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    834,
+                    2088,
+                    865,
+                    2088,
+                    865,
+                    2113,
+                    834,
+                    2113
+                ],
+                "score": 0.861
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    839,
+                    1377,
+                    871,
+                    1377,
+                    871,
+                    1403,
+                    839,
+                    1403
+                ],
+                "score": 0.85,
+                "latex": "\\alpha _ { \\pi }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    749,
+                    1378,
+                    784,
+                    1378,
+                    784,
+                    1403,
+                    749,
+                    1403
+                ],
+                "score": 0.84,
+                "latex": "\\alpha _ { M }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1294,
+                    1579,
+                    1378,
+                    1579,
+                    1378,
+                    1603,
+                    1294,
+                    1603
+                ],
+                "score": 0.67,
+                "latex": "\\alpha _ { \\pi } = 0 . 0 )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1187,
+                    1578,
+                    1286,
+                    1578,
+                    1286,
+                    1604,
+                    1187,
+                    1604
+                ],
+                "score": 0.66,
+                "latex": "( \\alpha _ { M } = 0 . 0 1"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1313,
+                    1550,
+                    1395,
+                    1550,
+                    1395,
+                    1573,
+                    1313,
+                    1573
+                ],
+                "score": 0.62,
+                "latex": "\\alpha _ { \\pi } = 0 . 0"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1220,
+                    1549,
+                    1306,
+                    1549,
+                    1306,
+                    1574,
+                    1220,
+                    1574
+                ],
+                "score": 0.6,
+                "latex": "\\alpha _ { M } = 0 . 0"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    490,
+                    1123,
+                    530,
+                    1123,
+                    530,
+                    1146,
+                    490,
+                    1146
+                ],
+                "score": 0.26,
+                "latex": "( 1 0 ^ { 6 } )"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    384.0,
+                    283.0,
+                    566.0,
+                    283.0,
+                    566.0,
+                    307.0,
+                    384.0,
+                    307.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    651.0,
+                    282.0,
+                    831.0,
+                    282.0,
+                    831.0,
+                    306.0,
+                    651.0,
+                    306.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    884.0,
+                    279.0,
+                    1131.0,
+                    279.0,
+                    1131.0,
+                    308.0,
+                    884.0,
+                    308.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1151.0,
+                    279.0,
+                    1394.0,
+                    279.0,
+                    1394.0,
+                    308.0,
+                    1151.0,
+                    308.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    638.0,
+                    311.0,
+                    843.0,
+                    311.0,
+                    843.0,
+                    326.0,
+                    638.0,
+                    326.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    847.0,
+                    298.0,
+                    912.0,
+                    298.0,
+                    912.0,
+                    327.0,
+                    847.0,
+                    327.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    315.0,
+                    318.0,
+                    372.0,
+                    318.0,
+                    372.0,
+                    348.0,
+                    315.0,
+                    348.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1112.0,
+                    323.0,
+                    1170.0,
+                    323.0,
+                    1170.0,
+                    354.0,
+                    1112.0,
+                    354.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    580.0,
+                    343.0,
+                    638.0,
+                    343.0,
+                    638.0,
+                    374.0,
+                    580.0,
+                    374.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    348.0,
+                    320.0,
+                    348.0,
+                    320.0,
+                    489.0,
+                    291.0,
+                    489.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    547.0,
+                    356.0,
+                    575.0,
+                    356.0,
+                    575.0,
+                    370.0,
+                    547.0,
+                    370.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    326.0,
+                    365.0,
+                    370.0,
+                    365.0,
+                    370.0,
+                    392.0,
+                    326.0,
+                    392.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    848.0,
+                    366.0,
+                    905.0,
+                    366.0,
+                    905.0,
+                    396.0,
+                    848.0,
+                    396.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1008.0,
+                    389.0,
+                    1019.0,
+                    389.0,
+                    1019.0,
+                    397.0,
+                    1008.0,
+                    397.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1115.0,
+                    370.0,
+                    1169.0,
+                    370.0,
+                    1169.0,
+                    396.0,
+                    1115.0,
+                    396.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    325.0,
+                    410.0,
+                    369.0,
+                    410.0,
+                    369.0,
+                    436.0,
+                    325.0,
+                    436.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    581.0,
+                    396.0,
+                    638.0,
+                    396.0,
+                    638.0,
+                    426.0,
+                    581.0,
+                    426.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1115.0,
+                    415.0,
+                    1169.0,
+                    415.0,
+                    1169.0,
+                    441.0,
+                    1115.0,
+                    441.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    848.0,
+                    434.0,
+                    904.0,
+                    434.0,
+                    904.0,
+                    464.0,
+                    848.0,
+                    464.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    581.0,
+                    449.0,
+                    637.0,
+                    449.0,
+                    637.0,
+                    479.0,
+                    581.0,
+                    479.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    974.0,
+                    450.0,
+                    984.0,
+                    450.0,
+                    984.0,
+                    460.0,
+                    974.0,
+                    460.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    325.0,
+                    455.0,
+                    369.0,
+                    455.0,
+                    369.0,
+                    482.0,
+                    325.0,
+                    482.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    387.0,
+                    459.0,
+                    395.0,
+                    459.0,
+                    395.0,
+                    469.0,
+                    387.0,
+                    469.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1113.0,
+                    458.0,
+                    1170.0,
+                    458.0,
+                    1170.0,
+                    488.0,
+                    1113.0,
+                    488.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    402.0,
+                    476.0,
+                    413.0,
+                    476.0,
+                    413.0,
+                    485.0,
+                    402.0,
+                    485.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1225.0,
+                    476.0,
+                    1237.0,
+                    476.0,
+                    1237.0,
+                    487.0,
+                    1225.0,
+                    487.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    938.0,
+                    487.0,
+                    964.0,
+                    487.0,
+                    964.0,
+                    499.0,
+                    938.0,
+                    499.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    562.0,
+                    494.0,
+                    573.0,
+                    494.0,
+                    573.0,
+                    505.0,
+                    562.0,
+                    505.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    804.0,
+                    495.0,
+                    836.0,
+                    495.0,
+                    836.0,
+                    504.0,
+                    804.0,
+                    504.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    348.0,
+                    502.0,
+                    368.0,
+                    502.0,
+                    368.0,
+                    524.0,
+                    348.0,
+                    524.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    613.0,
+                    507.0,
+                    635.0,
+                    507.0,
+                    635.0,
+                    529.0,
+                    613.0,
+                    529.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    646.0,
+                    504.0,
+                    662.0,
+                    504.0,
+                    662.0,
+                    519.0,
+                    646.0,
+                    519.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    880.0,
+                    507.0,
+                    900.0,
+                    507.0,
+                    900.0,
+                    529.0,
+                    880.0,
+                    529.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1147.0,
+                    505.0,
+                    1167.0,
+                    505.0,
+                    1167.0,
+                    529.0,
+                    1147.0,
+                    529.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    354.0,
+                    530.0,
+                    595.0,
+                    530.0,
+                    595.0,
+                    558.0,
+                    354.0,
+                    558.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    622.0,
+                    530.0,
+                    823.0,
+                    530.0,
+                    823.0,
+                    558.0,
+                    622.0,
+                    558.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    824.0,
+                    533.0,
+                    860.0,
+                    533.0,
+                    860.0,
+                    557.0,
+                    824.0,
+                    557.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    887.0,
+                    530.0,
+                    1128.0,
+                    530.0,
+                    1128.0,
+                    558.0,
+                    887.0,
+                    558.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1152.0,
+                    530.0,
+                    1393.0,
+                    530.0,
+                    1393.0,
+                    558.0,
+                    1152.0,
+                    558.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    392.0,
+                    564.0,
+                    560.0,
+                    564.0,
+                    560.0,
+                    592.0,
+                    392.0,
+                    592.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    659.0,
+                    564.0,
+                    825.0,
+                    564.0,
+                    825.0,
+                    592.0,
+                    659.0,
+                    592.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    895.0,
+                    564.0,
+                    1121.0,
+                    564.0,
+                    1121.0,
+                    592.0,
+                    895.0,
+                    592.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1162.0,
+                    564.0,
+                    1385.0,
+                    564.0,
+                    1385.0,
+                    592.0,
+                    1162.0,
+                    592.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    325.0,
+                    588.0,
+                    370.0,
+                    588.0,
+                    370.0,
+                    615.0,
+                    325.0,
+                    615.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    382.0,
+                    596.0,
+                    433.0,
+                    596.0,
+                    433.0,
+                    608.0,
+                    382.0,
+                    608.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    452.0,
+                    596.0,
+                    575.0,
+                    596.0,
+                    575.0,
+                    608.0,
+                    452.0,
+                    608.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    645.0,
+                    594.0,
+                    717.0,
+                    594.0,
+                    717.0,
+                    607.0,
+                    645.0,
+                    607.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    847.0,
+                    583.0,
+                    913.0,
+                    583.0,
+                    913.0,
+                    612.0,
+                    847.0,
+                    612.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1179.0,
+                    596.0,
+                    1245.0,
+                    596.0,
+                    1245.0,
+                    608.0,
+                    1179.0,
+                    608.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1335.0,
+                    598.0,
+                    1369.0,
+                    598.0,
+                    1369.0,
+                    607.0,
+                    1335.0,
+                    607.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1078.0,
+                    608.0,
+                    1098.0,
+                    608.0,
+                    1098.0,
+                    621.0,
+                    1078.0,
+                    621.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    300.0,
+                    630.0,
+                    369.0,
+                    630.0,
+                    369.0,
+                    772.0,
+                    300.0,
+                    772.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    531.0,
+                    648.0,
+                    577.0,
+                    648.0,
+                    577.0,
+                    666.0,
+                    531.0,
+                    666.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    581.0,
+                    618.0,
+                    638.0,
+                    618.0,
+                    638.0,
+                    705.0,
+                    581.0,
+                    705.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    667.0,
+                    669.0,
+                    731.0,
+                    669.0,
+                    731.0,
+                    685.0,
+                    667.0,
+                    685.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    847.0,
+                    650.0,
+                    904.0,
+                    650.0,
+                    904.0,
+                    680.0,
+                    847.0,
+                    680.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1113.0,
+                    623.0,
+                    1171.0,
+                    623.0,
+                    1171.0,
+                    707.0,
+                    1113.0,
+                    707.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    659.0,
+                    717.0,
+                    673.0,
+                    717.0,
+                    673.0,
+                    767.0,
+                    659.0,
+                    767.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    749.0,
+                    714.0,
+                    768.0,
+                    714.0,
+                    768.0,
+                    722.0,
+                    749.0,
+                    722.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    848.0,
+                    720.0,
+                    903.0,
+                    720.0,
+                    903.0,
+                    746.0,
+                    848.0,
+                    746.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    919.0,
+                    722.0,
+                    935.0,
+                    722.0,
+                    935.0,
+                    797.0,
+                    919.0,
+                    797.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1013.0,
+                    715.0,
+                    1106.0,
+                    715.0,
+                    1106.0,
+                    734.0,
+                    1013.0,
+                    734.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1345.0,
+                    722.0,
+                    1367.0,
+                    722.0,
+                    1367.0,
+                    731.0,
+                    1345.0,
+                    731.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    325.0,
+                    743.0,
+                    370.0,
+                    743.0,
+                    370.0,
+                    769.0,
+                    325.0,
+                    769.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    590.0,
+                    730.0,
+                    637.0,
+                    730.0,
+                    637.0,
+                    761.0,
+                    590.0,
+                    761.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    681.0,
+                    733.0,
+                    695.0,
+                    733.0,
+                    695.0,
+                    749.0,
+                    681.0,
+                    749.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1115.0,
+                    734.0,
+                    1169.0,
+                    734.0,
+                    1169.0,
+                    760.0,
+                    1115.0,
+                    760.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1070.0,
+                    766.0,
+                    1098.0,
+                    766.0,
+                    1098.0,
+                    775.0,
+                    1070.0,
+                    775.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    348.0,
+                    795.0,
+                    369.0,
+                    795.0,
+                    369.0,
+                    818.0,
+                    348.0,
+                    818.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    613.0,
+                    790.0,
+                    635.0,
+                    790.0,
+                    635.0,
+                    814.0,
+                    613.0,
+                    814.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    882.0,
+                    790.0,
+                    900.0,
+                    790.0,
+                    900.0,
+                    813.0,
+                    882.0,
+                    813.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1147.0,
+                    789.0,
+                    1169.0,
+                    789.0,
+                    1169.0,
+                    813.0,
+                    1147.0,
+                    813.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    354.0,
+                    814.0,
+                    596.0,
+                    814.0,
+                    596.0,
+                    842.0,
+                    354.0,
+                    842.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    620.0,
+                    814.0,
+                    863.0,
+                    814.0,
+                    863.0,
+                    842.0,
+                    620.0,
+                    842.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    887.0,
+                    814.0,
+                    1128.0,
+                    814.0,
+                    1128.0,
+                    842.0,
+                    887.0,
+                    842.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1152.0,
+                    814.0,
+                    1395.0,
+                    814.0,
+                    1395.0,
+                    842.0,
+                    1152.0,
+                    842.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    372.0,
+                    848.0,
+                    578.0,
+                    848.0,
+                    578.0,
+                    872.0,
+                    372.0,
+                    872.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    640.0,
+                    848.0,
+                    844.0,
+                    848.0,
+                    844.0,
+                    872.0,
+                    640.0,
+                    872.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    875.0,
+                    848.0,
+                    1404.0,
+                    848.0,
+                    1404.0,
+                    872.0,
+                    875.0,
+                    872.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    314.0,
+                    881.0,
+                    372.0,
+                    881.0,
+                    372.0,
+                    911.0,
+                    314.0,
+                    911.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    413.0,
+                    892.0,
+                    580.0,
+                    892.0,
+                    580.0,
+                    916.0,
+                    413.0,
+                    916.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    582.0,
+                    877.0,
+                    637.0,
+                    877.0,
+                    637.0,
+                    903.0,
+                    582.0,
+                    903.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    692.0,
+                    892.0,
+                    704.0,
+                    892.0,
+                    704.0,
+                    901.0,
+                    692.0,
+                    901.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    714.0,
+                    887.0,
+                    845.0,
+                    887.0,
+                    845.0,
+                    904.0,
+                    714.0,
+                    904.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1324.0,
+                    889.0,
+                    1352.0,
+                    889.0,
+                    1352.0,
+                    902.0,
+                    1324.0,
+                    902.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    290.0,
+                    912.0,
+                    370.0,
+                    912.0,
+                    370.0,
+                    1056.0,
+                    290.0,
+                    1056.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    381.0,
+                    950.0,
+                    393.0,
+                    950.0,
+                    393.0,
+                    1035.0,
+                    381.0,
+                    1035.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    410.0,
+                    911.0,
+                    433.0,
+                    911.0,
+                    433.0,
+                    933.0,
+                    410.0,
+                    933.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    581.0,
+                    936.0,
+                    638.0,
+                    936.0,
+                    638.0,
+                    966.0,
+                    581.0,
+                    966.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    647.0,
+                    950.0,
+                    659.0,
+                    950.0,
+                    659.0,
+                    961.0,
+                    647.0,
+                    961.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    667.0,
+                    949.0,
+                    710.0,
+                    949.0,
+                    710.0,
+                    963.0,
+                    667.0,
+                    963.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    849.0,
+                    906.0,
+                    902.0,
+                    906.0,
+                    902.0,
+                    932.0,
+                    849.0,
+                    932.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1105.0,
+                    898.0,
+                    1170.0,
+                    898.0,
+                    1170.0,
+                    967.0,
+                    1105.0,
+                    967.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    849.0,
+                    957.0,
+                    903.0,
+                    957.0,
+                    903.0,
+                    984.0,
+                    849.0,
+                    984.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    315.0,
+                    970.0,
+                    370.0,
+                    970.0,
+                    370.0,
+                    995.0,
+                    315.0,
+                    995.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1113.0,
+                    980.0,
+                    1170.0,
+                    980.0,
+                    1170.0,
+                    1010.0,
+                    1113.0,
+                    1010.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    582.0,
+                    999.0,
+                    636.0,
+                    999.0,
+                    636.0,
+                    1025.0,
+                    582.0,
+                    1025.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1293.0,
+                    1002.0,
+                    1325.0,
+                    1002.0,
+                    1325.0,
+                    1011.0,
+                    1293.0,
+                    1011.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    316.0,
+                    1012.0,
+                    369.0,
+                    1012.0,
+                    369.0,
+                    1080.0,
+                    316.0,
+                    1080.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    613.0,
+                    1060.0,
+                    635.0,
+                    1060.0,
+                    635.0,
+                    1084.0,
+                    613.0,
+                    1084.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    645.0,
+                    1007.0,
+                    668.0,
+                    1007.0,
+                    668.0,
+                    1092.0,
+                    645.0,
+                    1092.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    847.0,
+                    1007.0,
+                    903.0,
+                    1007.0,
+                    903.0,
+                    1086.0,
+                    847.0,
+                    1086.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1113.0,
+                    1021.0,
+                    1170.0,
+                    1021.0,
+                    1170.0,
+                    1089.0,
+                    1113.0,
+                    1089.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    353.0,
+                    1095.0,
+                    598.0,
+                    1095.0,
+                    598.0,
+                    1127.0,
+                    353.0,
+                    1127.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    620.0,
+                    1096.0,
+                    863.0,
+                    1096.0,
+                    863.0,
+                    1124.0,
+                    620.0,
+                    1124.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    887.0,
+                    1096.0,
+                    1128.0,
+                    1096.0,
+                    1128.0,
+                    1124.0,
+                    887.0,
+                    1124.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1154.0,
+                    1096.0,
+                    1395.0,
+                    1096.0,
+                    1395.0,
+                    1124.0,
+                    1154.0,
+                    1124.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    411.0,
+                    1121.0,
+                    489.0,
+                    1121.0,
+                    489.0,
+                    1146.0,
+                    411.0,
+                    1146.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    531.0,
+                    1121.0,
+                    537.0,
+                    1121.0,
+                    537.0,
+                    1146.0,
+                    531.0,
+                    1146.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    678.0,
+                    1121.0,
+                    802.0,
+                    1121.0,
+                    802.0,
+                    1146.0,
+                    678.0,
+                    1146.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    944.0,
+                    1121.0,
+                    1070.0,
+                    1121.0,
+                    1070.0,
+                    1146.0,
+                    944.0,
+                    1146.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1211.0,
+                    1121.0,
+                    1336.0,
+                    1121.0,
+                    1336.0,
+                    1146.0,
+                    1211.0,
+                    1146.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    526.0,
+                    1154.0,
+                    627.0,
+                    1154.0,
+                    627.0,
+                    1183.0,
+                    526.0,
+                    1183.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    650.0,
+                    1162.0,
+                    684.0,
+                    1162.0,
+                    684.0,
+                    1174.0,
+                    650.0,
+                    1174.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    690.0,
+                    1155.0,
+                    742.0,
+                    1155.0,
+                    742.0,
+                    1182.0,
+                    690.0,
+                    1182.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    806.0,
+                    1157.0,
+                    840.0,
+                    1157.0,
+                    840.0,
+                    1181.0,
+                    806.0,
+                    1181.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    902.0,
+                    1154.0,
+                    996.0,
+                    1154.0,
+                    996.0,
+                    1179.0,
+                    902.0,
+                    1179.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1016.0,
+                    1154.0,
+                    1097.0,
+                    1154.0,
+                    1097.0,
+                    1183.0,
+                    1016.0,
+                    1183.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1117.0,
+                    1160.0,
+                    1156.0,
+                    1160.0,
+                    1156.0,
+                    1177.0,
+                    1117.0,
+                    1177.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1160.0,
+                    1157.0,
+                    1195.0,
+                    1157.0,
+                    1195.0,
+                    1181.0,
+                    1160.0,
+                    1181.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    439.0,
+                    351.0,
+                    501.0,
+                    351.0,
+                    501.0,
+                    385.0,
+                    439.0,
+                    385.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    498.0,
+                    355.5,
+                    545.0,
+                    355.5,
+                    545.0,
+                    368.0,
+                    498.0,
+                    368.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    462.25,
+                    471.0,
+                    560.25,
+                    471.0,
+                    560.25,
+                    501.0,
+                    462.25,
+                    501.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    668.0,
+                    714.0,
+                    687.0,
+                    714.0,
+                    687.0,
+                    729.5,
+                    668.0,
+                    729.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    765.0,
+                    714.0,
+                    845.0,
+                    714.0,
+                    845.0,
+                    737.5,
+                    765.0,
+                    737.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    378.0,
+                    692.5,
+                    438.0,
+                    692.5,
+                    438.0,
+                    709.5,
+                    378.0,
+                    709.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    679.25,
+                    714.5,
+                    724.25,
+                    714.5,
+                    724.25,
+                    733.0,
+                    679.25,
+                    733.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1240.75,
+                    752.0,
+                    1315.75,
+                    752.0,
+                    1315.75,
+                    769.5,
+                    1240.75,
+                    769.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    629.0,
+                    759.0,
+                    694.0,
+                    759.0,
+                    694.0,
+                    774.0,
+                    629.0,
+                    774.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1019.0,
+                    891.0,
+                    1110.0,
+                    891.0,
+                    1110.0,
+                    914.5,
+                    1019.0,
+                    914.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1240.0,
+                    884.0,
+                    1329.0,
+                    884.0,
+                    1329.0,
+                    904.0,
+                    1240.0,
+                    904.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    690.0,
+                    905.5,
+                    724.0,
+                    905.5,
+                    724.0,
+                    917.5,
+                    690.0,
+                    917.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    721.0,
+                    899.0,
+                    792.0,
+                    899.0,
+                    792.0,
+                    911.5,
+                    721.0,
+                    911.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    429.0,
+                    912.0,
+                    478.0,
+                    912.0,
+                    478.0,
+                    928.5,
+                    429.0,
+                    928.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    950.0,
+                    902.5,
+                    1032.0,
+                    902.5,
+                    1032.0,
+                    931.5,
+                    950.0,
+                    931.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    702.75,
+                    967.0,
+                    715.75,
+                    967.0,
+                    715.75,
+                    976.0,
+                    702.75,
+                    976.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    912.0,
+                    940.5,
+                    972.0,
+                    940.5,
+                    972.0,
+                    955.0,
+                    912.0,
+                    955.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    917.0,
+                    1002.5,
+                    940.0,
+                    1002.5,
+                    940.0,
+                    1005.0,
+                    917.0,
+                    1005.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1193.25,
+                    1043.0,
+                    1258.25,
+                    1043.0,
+                    1258.25,
+                    1067.5,
+                    1193.25,
+                    1067.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    369.0,
+                    1020.0,
+                    420.0,
+                    1020.0,
+                    420.0,
+                    1026.0,
+                    369.0,
+                    1026.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    545.0,
+                    1370.0,
+                    748.0,
+                    1370.0,
+                    748.0,
+                    1405.0,
+                    545.0,
+                    1405.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    785.0,
+                    1370.0,
+                    838.0,
+                    1370.0,
+                    838.0,
+                    1405.0,
+                    785.0,
+                    1405.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    872.0,
+                    1370.0,
+                    1152.0,
+                    1370.0,
+                    1152.0,
+                    1405.0,
+                    872.0,
+                    1405.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    320.0,
+                    1449.0,
+                    388.0,
+                    1449.0,
+                    388.0,
+                    1484.0,
+                    320.0,
+                    1484.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    711.0,
+                    1449.0,
+                    779.0,
+                    1449.0,
+                    779.0,
+                    1484.0,
+                    711.0,
+                    1484.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    320.0,
+                    1499.0,
+                    392.0,
+                    1499.0,
+                    392.0,
+                    1534.0,
+                    320.0,
+                    1534.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    709.0,
+                    1499.0,
+                    780.0,
+                    1499.0,
+                    780.0,
+                    1534.0,
+                    709.0,
+                    1534.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1176.0,
+                    1486.0,
+                    1220.0,
+                    1486.0,
+                    1220.0,
+                    1547.0,
+                    1176.0,
+                    1547.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1551.0,
+                    389.0,
+                    1551.0,
+                    389.0,
+                    1719.0,
+                    293.0,
+                    1719.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    712.0,
+                    1552.0,
+                    779.0,
+                    1552.0,
+                    779.0,
+                    1586.0,
+                    712.0,
+                    1586.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1177.0,
+                    1544.0,
+                    1219.0,
+                    1544.0,
+                    1219.0,
+                    1577.0,
+                    1177.0,
+                    1577.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1307.0,
+                    1544.0,
+                    1312.0,
+                    1544.0,
+                    1312.0,
+                    1577.0,
+                    1307.0,
+                    1577.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1396.0,
+                    1544.0,
+                    1406.0,
+                    1544.0,
+                    1406.0,
+                    1577.0,
+                    1396.0,
+                    1577.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1179.0,
+                    1575.0,
+                    1186.0,
+                    1575.0,
+                    1186.0,
+                    1607.0,
+                    1179.0,
+                    1607.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1287.0,
+                    1575.0,
+                    1293.0,
+                    1575.0,
+                    1293.0,
+                    1607.0,
+                    1287.0,
+                    1607.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1379.0,
+                    1575.0,
+                    1388.0,
+                    1575.0,
+                    1388.0,
+                    1607.0,
+                    1379.0,
+                    1607.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    319.0,
+                    1603.0,
+                    388.0,
+                    1603.0,
+                    388.0,
+                    1637.0,
+                    319.0,
+                    1637.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    711.0,
+                    1602.0,
+                    780.0,
+                    1602.0,
+                    780.0,
+                    1637.0,
+                    711.0,
+                    1637.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1179.0,
+                    1605.0,
+                    1376.0,
+                    1605.0,
+                    1376.0,
+                    1637.0,
+                    1179.0,
+                    1637.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1176.0,
+                    1632.0,
+                    1378.0,
+                    1632.0,
+                    1378.0,
+                    1666.0,
+                    1176.0,
+                    1666.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    320.0,
+                    1654.0,
+                    387.0,
+                    1654.0,
+                    387.0,
+                    1687.0,
+                    320.0,
+                    1687.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    709.0,
+                    1654.0,
+                    779.0,
+                    1654.0,
+                    779.0,
+                    1686.0,
+                    709.0,
+                    1686.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1179.0,
+                    1664.0,
+                    1387.0,
+                    1664.0,
+                    1387.0,
+                    1696.0,
+                    1179.0,
+                    1696.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    320.0,
+                    1706.0,
+                    387.0,
+                    1706.0,
+                    387.0,
+                    1740.0,
+                    320.0,
+                    1740.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    409.0,
+                    1719.0,
+                    426.0,
+                    1719.0,
+                    426.0,
+                    1777.0,
+                    409.0,
+                    1777.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    711.0,
+                    1706.0,
+                    780.0,
+                    1706.0,
+                    780.0,
+                    1740.0,
+                    711.0,
+                    1740.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1176.0,
+                    1693.0,
+                    1388.0,
+                    1693.0,
+                    1388.0,
+                    1756.0,
+                    1176.0,
+                    1756.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    331.0,
+                    1755.0,
+                    389.0,
+                    1755.0,
+                    389.0,
+                    1791.0,
+                    331.0,
+                    1791.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    722.0,
+                    1755.0,
+                    781.0,
+                    1755.0,
+                    781.0,
+                    1791.0,
+                    722.0,
+                    1791.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1179.0,
+                    1753.0,
+                    1376.0,
+                    1753.0,
+                    1376.0,
+                    1785.0,
+                    1179.0,
+                    1785.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1177.0,
+                    1784.0,
+                    1387.0,
+                    1784.0,
+                    1387.0,
+                    1814.0,
+                    1177.0,
+                    1814.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    360.0,
+                    1810.0,
+                    386.0,
+                    1810.0,
+                    386.0,
+                    1838.0,
+                    360.0,
+                    1838.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    749.0,
+                    1808.0,
+                    778.0,
+                    1808.0,
+                    778.0,
+                    1840.0,
+                    749.0,
+                    1840.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    369.0,
+                    1850.0,
+                    416.0,
+                    1850.0,
+                    416.0,
+                    1882.0,
+                    369.0,
+                    1882.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    430.0,
+                    1850.0,
+                    606.0,
+                    1850.0,
+                    606.0,
+                    1880.0,
+                    430.0,
+                    1880.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    623.0,
+                    1852.0,
+                    670.0,
+                    1852.0,
+                    670.0,
+                    1879.0,
+                    623.0,
+                    1879.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    685.0,
+                    1852.0,
+                    732.0,
+                    1852.0,
+                    732.0,
+                    1879.0,
+                    685.0,
+                    1879.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    760.0,
+                    1852.0,
+                    805.0,
+                    1852.0,
+                    805.0,
+                    1879.0,
+                    760.0,
+                    1879.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    822.0,
+                    1852.0,
+                    871.0,
+                    1852.0,
+                    871.0,
+                    1880.0,
+                    822.0,
+                    1880.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    884.0,
+                    1849.0,
+                    1120.0,
+                    1849.0,
+                    1120.0,
+                    1882.0,
+                    884.0,
+                    1882.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    473.0,
+                    1877.0,
+                    627.0,
+                    1877.0,
+                    627.0,
+                    1913.0,
+                    473.0,
+                    1913.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    862.0,
+                    1877.0,
+                    1018.0,
+                    1877.0,
+                    1018.0,
+                    1913.0,
+                    862.0,
+                    1913.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    538.0,
+                    1199.0,
+                    1160.0,
+                    1199.0,
+                    1160.0,
+                    1239.0,
+                    538.0,
+                    1239.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    442.0,
+                    1931.0,
+                    1257.0,
+                    1931.0,
+                    1257.0,
+                    1971.0,
+                    442.0,
+                    1971.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    73.0,
+                    815.0,
+                    73.0,
+                    815.0,
+                    108.0,
+                    295.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    829.0,
+                    2084.0,
+                    871.0,
+                    2084.0,
+                    871.0,
+                    2125.0,
+                    829.0,
+                    2125.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 21,
+            "width": 1700,
+            "height": 2200
+        }
+    }
+]
\ No newline at end of file
diff --git a/parse/train/S1en0sRqKm/S1en0sRqKm.md b/parse/train/S1en0sRqKm/S1en0sRqKm.md
new file mode 100644
index 0000000000000000000000000000000000000000..1f92ccdb0cbd1e6f106d88480b02141820409763
--- /dev/null
+++ b/parse/train/S1en0sRqKm/S1en0sRqKm.md
@@ -0,0 +1,200 @@
+# ON THE COMPUTATIONAL INEFFICIENCY OF LARGE BATCH SIZES FOR STOCHASTIC GRADIENT DESCENT
+
+Anonymous authors Paper under double-blind review
+
+# ABSTRACT
+
+Increasing the mini-batch size for stochastic gradient descent offers significant opportunities to reduce wall-clock training time, but there are a variety of theoretical and systems challenges that impede the widespread success of this technique (Das et al., 2016; Keskar et al., 2016). We investigate these issues, with an emphasis on time to convergence and total computational cost, through an extensive empirical analysis of network training across several architectures and problem domains, including image classification, image segmentation, and language modeling. Although it is common practice to increase the batch size in order to fully exploit available computational resources, we find a substantially more nuanced picture. Our main finding is that across a wide range of network architectures and problem domains, increasing the batch size beyond a certain point yields no decrease in wall-clock time to convergence for either train or test loss. This batch size is usually substantially below the capacity of current systems. We show that popular training strategies for large batch size optimization begin to fail before we can populate all available compute resources, and we show that the point at which these methods break down depends more on attributes like model architecture and data complexity than it does directly on the size of the dataset.
+
+# 1 INTRODUCTION
+
+Mini-batch stochastic gradient descent (SGD) is the dominant optimization method for training deep neural networks (DNNs) (Bengio & LeCun, 2007; Bottou, 2010). In the face of unprecedented growth in dataset size, a large body of work has attempted to scale SGD to train DNN models on increasingly large datasets, while keeping wall-clock time manageable (Iandola et al., 2015; Goyal et al., 2017; Smith & Le, 2018; Devarakonda et al., 2017). The most common approach to train large models at scale is distributed synchronous mini-batch SGD, which exploits additional computational resources through data parallelism. This technique reduces wall-clock training time by increasing the mini-batch size, i.e., the number of examples used to compute a stochastic estimate of the gradient of the loss function at each training iteration, while holding the number of epochs constant. Proponents of large batch size training often argue that the merits stem from its ability to decrease wall-clock training time while maintaining final model performance. Indeed, an enormous amount of work has gone into designing systems that seem to operate under an assumption that equates large batch size training with machine learning at scale (Goyal et al., 2017; Jia et al., 2018; Puri et al., 2018).
+
+Increasing the batch size improves the scaling performance of SGD per epoch, and there are significant challenges in building efficient distributed systems that are able to exploit additional computational resources to use large batch sizes (Jia et al., 2018). However, even if we were able to address these systems challenges, there are still more fundamental limitations to this approach. Large batch sizes often negatively impact important performance metrics of interest, including total computational cost (which usually determines monetary cost) and prediction quality.
+
+In this paper, we will measure the total computational cost as the number of training iterations times the work done per iteration—in order to simplify measurements, we use the number of training iterations as a proxy for the wall-clock time. We do this because the implementation of parallel algorithms depends on software and hardware choices, and our goal is to draw more general conclusions about the performance of SGD-based methods.
+
+Based on this model for total computational cost and wall-clock time, the following should be clear: unless increasing the batch size leads to a commensurate decrease in the total number of training iterations needed to find a good model, large batch training will result in greater total computational cost with little-to-no decrease in wall-clock training time.
+
+Based on our empirical results across a range of datasets and architectures, we find that as the batch size becomes larger, there are three main phases of scaling behavior for convergence speed: (1) there is a small regime of batch sizes in which increasing the batch size results in linear gains in convergence speed; (2) there is a larger regime of batch sizes that results in sublinear gains in convergence speed—in this regime, increasing the batch size can improve wall-clock training time at the expense of greater total computational cost; (3) eventually, we reach a third regime where a higher batch size results in marginal or non-existent reductions in convergence speed. In our experiments, we find that this third regime begins at a batch size that is too small to let us fully utilize available compute. Training past this batch size increases the total computational cost without reducing wall-clock training time or prediction quality. While there has been considerable excitement around heuristics that have been shown to make large batch training practical for certain problems (Goyal et al., 2017; Smith & Le, 2018), we demonstrate that these techniques still suffer from the same convergence trends we observe, and they often decrease stability of the training process.
+
+Recent work has observed that the final test performance of models trained with large batch sizes degrades after training for a fixed number of epochs (Yao et al., 2018; Keskar et al., 2016). This phenomenon is known as the generalization gap. Previous work addressing this problem has focused on training for more iterations in the large batch case (Hoffer et al., 2017) or adopting various heuristics to select a learning rate for larger batch sizes (Goyal et al., 2017; Smith & Le, 2018). Based on our empirical results, we find that existing techniques to mitigate the generalization gap do not work on some problems, and for other problems they only work for batch sizes that do not allow us to fully utilize our available compute. Perhaps more importantly, they do little to affect the diminishing returns in rates of convergence for training loss as batch size increases.
+
+Our objective is to understand the behavior of SGD and existing large batch techniques for many network architectures and problem domains, e.g., image classification/segmentation and natural language processing (NLP). We observe markedly worse performance for these techniques in domains other than image classification, where large batch optimization has received the most attention (Jia et al., 2018; You et al., 2017b). Because we eschew the challenges of an efficient distributed implementation by measuring number of iterations instead of wall-clock time, our results assume the most optimistic circumstances for large batch training. Our key observations are:
+
+• Increasing the batch size beyond a certain point yields no improvement in wall-clock time to convergence, even for a system with perfect parallelism. We observe that larger batch sizes result in a limited reduction in the number of training iterations needed to achieve low training or test error, and that eventually these gains become near-zero. Increasing the batch size leads to a significant increase in generalization error, which cannot be mitigated by existing techniques. We observe that these techniques often result in divergent training behavior or that they only mitigate degradation in test performance for small batch sizes relative to available compute. Dataset size is not the only factor determining the computational efficiency of large batch training. We observe that both the diminishing returns in convergence speed and the failure of existing methods correlate more with factors like model architecture and data complexity than dataset size alone. As a result, training time may significantly increase with dataset size in spite of increasingly available compute resources.
+
+In Section 2, we review the formulation of SGD as well as existing strategies to train with large batch sizes. In Section 3, we review recent theoretical results regarding the convergence rates of SGD in highly over-parameterized settings and discuss the potential impact of these results on the computational efficiency of SGD for deep learning. Section 4 presents our empirical results that demonstrate the inefficiencies of training SGD with large batch sizes, and we show that these persist when using existing large batch optimization techniques.
+
+# 2 BACKGROUND AND RELATED WORK
+
+Stochastic Gradient Descent. SGD is the most widely used algorithm to train DNN models. The model is parameterized by weights $\mathbf { w } \in \mathbb { R } ^ { d }$ , and the objective is to minimize the empirical loss over $n$ data points $\mathbf { x } _ { i }$ :
+
+$$
+L ( \mathbf { w } ) = \frac { 1 } { n } \sum _ { i = 1 } ^ { n } \ell ( \mathbf { w } , \mathbf { x } _ { i } ) ,
+$$
+
+where $\ell ( \cdot , \cdot )$ is a loss, e.g., cross-entropy or squared error. This loss gives a corresponding gradient
+
+$$
+\mathbf { g } ( \mathbf { w } ) : = \nabla L ( \mathbf { w } ) = \frac { 1 } { n } \sum _ { i = 1 } ^ { n } \nabla \ell ( \mathbf { w } , \mathbf { x } _ { i } ) .
+$$
+
+A mini-batch $B _ { m }$ of size $m < n$ is a collection of $m$ indices randomly drawn from the set $\{ 1 , \ldots , n \}$ , and we can use it to form an unbiased estimate of the gradient at iteration $k$ , as well as the corresponding SGD update:
+
+$$
+\mathbf { g } _ { m } ( \mathbf { w } _ { k } ) = \frac { 1 } { m } \sum _ { i \in \mathcal { B } _ { m } } \nabla \ell ( \mathbf { w } _ { k } , \mathbf { x } _ { i } ) \quad \mathrm { ~ a n d ~ } \quad \mathbf { w } _ { k + 1 } = \mathbf { w } _ { k } - \eta _ { k } \mathbf { g } _ { m } ( \mathbf { w } _ { k } ) ,
+$$
+
+where $\eta _ { k } > 0$ is the learning rate for iteration $k$ . One iteration of training for SGD corresponds to a single gradient computation / weight update. One epoch corresponds to $n / m$ iterations of training. This constitutes a single pass over the dataset, assuming the dataset is sampled without replacement.
+
+Efficient distributed systems reduce wall-clock training time by parallelizing gradient calculations across many machines. When the batch size is large enough to populate all available compute resources, this allows us to amortize the cost of coordination for each weight update.
+
+Existing large batch techniques. With the hope of keeping training times manageable as dataset sizes escalate, recent work has focused on the development of techniques that allow practitioners to increase the batch size to make use of growing computational resources (Jin et al., 2016; Jia et al., 2018; You et al., 2017a). However, there is a growing body of theoretical and empirical results suggesting that large batch sizes adversely affect the generalization performance of the final model (Yao et al., 2018; Keskar et al., 2016; Devarakonda et al., 2017).
+
+In response to this, recent work has proposed changing two parameters in relation to batch size: the number of training iterations and the learning rate. However, they also make assumptions that limit the effectiveness of their proposals as useful heuristics for practitioners.
+
+• Training longer: Hoffer et al. (2017) suggest increasing the number of training iterations. Even if this does reduce the generalization gap, it significantly increases both wall-clock training time and computational cost. Moreover, in some problems it does not lead to minima with better generalization performance (as we found when running our experiments). • Square root LR scaling: Scaling the learning rate as $\eta _ { 0 } \propto \sqrt { m }$ attempts to keep the weight increment length statistics constant, but the distance between SGD iterates is governed more by properties of the objective function than the ratio of learning rate to batch size (Chaudhari & Soatto, 2017; Zhu et al., 2018). This rule has also been found to be empirically sub-optimal in various problem domains (Krizhevsky, 2014). Linear LR scaling: The performance of large batch training can also be improved by using the linear scaling rule, which suggests choosing a learning rate proportional to the batch size $( \eta _ { 0 } \propto m )$ (Goyal et al., 2017). There are two motivations for this rule: the first assumes that one large-batch gradient step should resemble a series of small-batch gradient steps in order for convergence rates to improve linearly (Goyal et al., 2017); the other regards the SGD update equation as the Euler-Maruyama discretization of a stochastic differential equation (Sauer, 2012; Xing et al., 2018), and attempts to maintain a constant level of minibatch noise to help SGD explore the loss landscape (Chaudhari & Soatto, 2017; Zhu et al., 2018; Smith & Le, 2018).
+
+Both justifications for the linear scaling rule implicitly impose strong conditions on the loss function by requiring that it behave linearly near SGD iterates; therefore, if the loss function is highly nonlinear along the SGD trajectory or the step size is not small enough, then we should not expect these rules to provide useful guidance for many problems. Whereas several groups have successfully used this rule to train on the ImageNet dataset in under an hour, e.g. (Goyal et al., 2017; You et al., 2017b), applying this heuristic to other datasets has not led to similarly impressive results so far (Puri et al., 2018).
+
+The focus of this paper, however, is on more fundamental limitations of large batch training, and we empirically show that the above approaches fail to prevent diminishing returns in the rate of convergence for large batch sizes. We believe that these diminishing returns are of more immediate concern than the generalization gap and warrant more careful examination: if we cannot even minimize training error quickly, there is no real opportunity to minimize test error quickly, regardless of the difference in final test error across batch sizes by the time the model has converged.
+
+# 3 CRITICAL BATCH SIZES AND DIMINISHING RETURNS
+
+The convergence rate of SGD, denoted by $k _ { \epsilon } ( m )$ , is the number of iterations needed to achieve training error less than a fixed constant $\epsilon > 0$ by using SGD with batch size $m$ (we will drop the subscript $\epsilon$ when it is unambiguous). In order to guarantee that large batch sizes speed up training, $k ( m )$ should continue to decrease near-linearly with $m$ . Otherwise, a larger batch size increases computational cost with only limited reductions in wall-clock training time. For near-constant $k ( m )$ , the benefit of large batch sizes becomes near-zero.
+
+Ma et al. (2017) showed theoretically that in convex, over-parameterized settings, the reduction in convergence time obtained by increasing the batch size decays dramatically to a near-constant level after a critical batch size that is independent of the dataset size. This speedup is measured with respect to the number of SGD iterations required to reach some fixed loss error for some baseline batch size $m _ { 0 }$ , and for this purpose we define the speedup ratio $s ( m ; m _ { 0 } ) = k ( m _ { 0 } ) / k ( m )$ . The speedup ratio represents the amount of time we save by increasing the batch size to $m$ . Beyond the critical batch size mentioned above, even with no communication overhead and unlimited resources (where each batch size requires the same amount of wall-clock time to process) we would prefer to use the critical batch size because it requires less overall computation.
+
+This result is surprising because researchers have asserted that it should be possible to achieve linear gains in convergence speed so long as the batch size is small relative to dataset size (Smith & Le, 2018). This will present significant difficulties for future optimization work (large mini-batch training) because it prevents us from using large batch sizes as a catch-all approach to quickly train models on large datasets.
+
+# 4 EMPIRICAL EVALUATION
+
+Recent work studying large batch training has looked primarily at image classification (Jastrzebski et al., 2018; Yao et al., 2018), especially on the ImageNet dataset (Deng et al., 2009). We perform large batch size experiments across both traditional image classification (IC) tasks (such as on CIFAR-10/100 (Krizhevsky & Hinton, 2009)), as well as previously unexplored tasks like image segmentation (IS) using the Cityscapes dataset (Cordts et al., 2016), and natural language processing (NLP) using the WikiText-2 dataset (Merity et al., 2016). We also test how these results vary across other modern DNN architectures, namely ResNets (He et al., 2016), LSTMs (Hochreiter & Schmidhuber, 1997; Gers et al., 2000), AlexNet (Krizhevsky et al., 2012), VGG (Simonyan & Zisserman, 2014), Dilated Residual Networks (Yu et al., 2017), and MobileNetV2 (Sandler et al., 2014). We tested all of the large batch training techniques described in Section 2. We tried training longer based on the work of Hoffer et al. (2017), but we found that this necessarily cannot improve the convergence speed and often does not improve final test performance. The two other techniques include the square root scaling rule strategy (SRSR) and the linear scaling rule strategy (LSR). For the latter, we used a warm-up period at the start of training as suggested by Goyal et al. (2017). Table 1 reports our datasets, models and different training strategies. For each model, we evaluated against a base learning rate strategy (BLR) that used the same learning rate across all batch sizes. We selected this learning rate based on its performance on a small baseline batch size.
+
+![](images/b45c386a5ec3b1dee617dd3ed5dfe50c26a309dd84ec1efd0f4dcdac283a893c.jpg)  
+Figure 1: Contour plots of training losses for various problem domains on a log scale. Lighter colors indicate lower loss values. Since we train each batch size for a fixed number of epochs, the total number of training iterations scales down linearly. For each loss value, we can observe how many iterations it takes to converge to that value given a particular batch size, by tracing the level curve for the associated color. For all problems, there is a batch size after which the number of training iterations necessary to converge does not decrease.
+
+# 4.1 DIMINISHING RETURNS IN RATES OF CONVERGENCE
+
+We demonstrate the rapidly diminishing returns in rates of convergence across various problem domains and network configurations. Researchers increase the batch size in an attempt to achieve nearly linear speedups in convergence compared to a small mini-batch size. In particular, if the speedup is near-linear, i.e. $s ( m ; \bar { m } _ { 0 } ) = k ( \bar { m } _ { 0 } ) / k ( m ) \approx m / m _ { 0 }$ , then the computational cost remains nearly constant for large and small mini-batch SGD. However, if $s ( m ) \ \bar { \ll } \ m / m _ { 0 }$ , then the benefit of using large batch size training is negligible.
+
+In Figure 1, we show contour plots of training loss as a function of both the batch size and the number of training iterations of ResNet34 on CIFAR-10, an LSTM on WikiText-2, and DRN-D-22 on Cityscapes. Consider, for example, the contour plot for ResNet34 trained on CIFAR-10. We can see that as the batch size increases from 16 to 2048, the number of SGD iterations needed to achieve a particular loss value decreases linearly. Exceeding this regime, however, the speedup ratio becomes increasingly sublinear and soon we have $s ( m ; \bar { m _ { 0 } } ) \ll \bar { m } / m _ { 0 }$ . For batch size 8196, the training procedure does not achieve the lowest training loss, and from this perspective, even if we did not care about computational cost or training time, we would not be able to find an accurate model. We observe even worse scaling behavior for test performance (please see Figure 5 for details).
+
+Table 1: A description of the problem configurations and training strategies used in this paper. $\eta _ { 0 }$ is the initial learning rate, $W$ is the number of epochs used for warm-up in the linear scaling rule, $E$ is the total number of epochs trained   
+
+<table><tr><td>Dataset</td><td>Task</td><td>Architecture</td><td>Training Strategy</td><td>BS range</td></tr><tr><td> MNIST</td><td>IC</td><td>ResNet34</td><td>BLR,LSR (no = 0.1, W = 10, E = 200)</td><td>26 -214</td></tr><tr><td>CIFAR-10</td><td>IC</td><td>AlexNet, MobileNetV2 ResNet34, VGG16</td><td>BLR,LSR, SRSR (mo = 0.1,W=10,E= 200)</td><td>26-214</td></tr><tr><td>CIFAR-100</td><td>IC</td><td>ResNet34</td><td>BLR,LSR (no = 0.1,W = 10, E = 200)</td><td>2 -214</td></tr><tr><td>SVHN</td><td>IC</td><td>ResNet34</td><td>BLR,LSR (no =0.1,W =10,E= 200)</td><td>26-214</td></tr><tr><td> WikiText-2</td><td> NLP</td><td>LSTM</td><td>BLR,LSR (no = 20,W = 3,E = 40)</td><td>2³ -210</td></tr><tr><td>Cityscapes</td><td>IS</td><td>DRN-D-22</td><td>BLR,LSR (no = 0.01,W = 10,E= 100)</td><td>23 -211</td></tr></table>
+
+![](images/2f39da181bbafada9ee2bf61d16e3221282ee2f410345f6758d773bf18d978e5.jpg)  
+Figure 2: On the left: speedup curves when applying several popular techniques to avoid the generalization gap. Base LR uses a single learning rate for all batch sizes. On the right: the effect of the linear approximation error on final test accuracy when using the linear LR scaling rule.
+
+For NLP and IS, note that the gain from large batch training diminishes even faster. Neither the LSTM on WikiText-2 nor DRN-D-22 on Cityscapes can reach their respective baseline performances after reasonably small batch sizes of about 32 and 64, respectively. Although Puri et al. (2018) showed that training on the Amazon Reviews dataset (McAuley et al., 2015) can be done within 4 hours, they tune hyper-parameters heavily. This poses an issue for many practical deployments because these problems are often already slow to train.
+
+# 4.2 EXISTING STRATEGIES BREAK DOWN FOR LARGE BATCH SIZES
+
+We further explore how training with the linear and square root scaling rules compares to training with a fixed baseline learning rate (BLR) that does not change with batch size. In the left subfigure of Figure 2, we show the speedup curves of BLR, LSR, and SRSR strategies for ResNet34 on CIFAR10. Note that LSR and SRSR outperform BLR from batch size 256 to 2048 which implies that LSR and SRSR can help the model train for small-to-medium batch sizes. However, the speedup of LSR and SRSR is still worse than the ideal linear case, and the curves plateau quickly after a batch size of 2048, at which point BLR becomes better than LSR and SRSR. This means that for certain problems, scaling up the learning rate to compensate for an increased batch size hurts performance.
+
+In the right subfigure of Figure 2, we plot the test performance and the approximation error for LSR of ResNet34 on CIFAR-10. We measure the approximation error at the end of training, with final weights $\mathbf { w } ^ { * }$ . We take this error to be the absolute difference between the true loss value $L ( \mathbf { w } )$ and the linear approximation at $\mathbf { w } ^ { * }$ , given by $\hat { L } ( \mathbf { w } ) = L ( \mathbf { w } ^ { * } ) + \langle \mathbf { g } _ { m } ( \mathbf { w } ^ { * } ) , \mathbf { w } - \mathbf { w } ^ { * } \rangle$ . The approximation is calculated for $\mathbf { w } = \mathbf { w } ^ { * } - \eta \frac { m } { m _ { 0 } } \mathbf { g } _ { m } \big ( \mathbf { w } ^ { * } \big )$ to understand the behavior of the approximation along the trajectory for a single SGD iterate using the LSR. It appears that there exists a strong relationship between linear approximation error and test accuracy: as the linear approximation error increases, the test accuracy drops. Note the transition that happens at the critical batch size of 2048. After this point, the test accuracy drops significantly and the linear approximation error exceeds 1, showing that we quickly exit the regime in which the linear approximation is valid.
+
+![](images/8e5a9c2868609d6811e018f9aa938f74636a8e7c872b18be0773d9a10cb99683.jpg)  
+Figure 3: Speedup curves across different problem configurations. Left: different architectures result in different rates of convergence on CIFAR-10. Right: ResNet34 exhibits different rates of convergence on CIFAR-10, CIFAR-100, and SVHN. Loss thresholds are obtained by computing the lower quartile of loss values achieved by the largest batch size.
+
+# 4.3 CONVERGENCE SPEED HAS A WEAK DEPENDENCE ON DATASET SIZE
+
+Previous works have conjectured that the maximum batch size that can result in a good model is proportional to the size of the whole dataset (Smith et al., 2017; Smith & Le, 2018). However, for convex, over-parameterized problems, Ma et al. (2017) show that there is a model-dependent critical batch size after which we observe rapidly diminishing returns in convergence speed. In this section, to observe if a similar critical batch size exists in the non-convex case, we compare how changing model architecture or data complexity affects the shapes of speedup curves compared to changing the dataset size alone.
+
+First, in order to show that these diminishing returns depend on data complexity and DNN architecture, we plot speedup curves in Figure 3 to compare the scaling behaviors across different models and dataset configurations. For the error threshold , we chose the lowest quartile loss value reached by the largest batch size to make a fair comparison across configurations. This setup actually favors the large batch case, because there are lower loss thresholds that are attainable only in the small batch case. On the left, for the CIFAR-10 dataset, we compared four model architectures. For each architecture, we plotted the speedup curve obtained by training this model on the dataset for various batch sizes. The variety of speedup curve shapes indicates that model architecture is an important factor in determining the convergence speed of training for large batch sizes. For MobileNetV2/AlexNet, the diminishing returns become visible when batch size is 1024. However, for VGG16/ResNet34, the speedup does not flatten out until batch size 8196. Hence, in practice, the choice of model strongly affects our ability to use large batch sizes in SGD.
+
+On the right, in order to investigate the effect of problem complexity, we compared the performance of ResNet34 on four datasets of the same size: CIFAR-10, CIFAR-100, MNIST, and the SVHN dataset (we cut off MNIST and SVHN to $5 0 k$ training examples each). Although all problems display diminishing returns in rates of convergence, the point at which the curves plateau varies according to problem complexity. It is not hard to see that, for simpler problems such as SVHN, the curves flatten out later than for harder problems (e.g. CIFAR-10/100).
+
+In all of the above cases, the diminishing rates of return in convergence speed become visible after only moderate increases in the batch size. Previous works have only studied convergence behavior for a fairly limited range of batch sizes (e.g., up to 4096 for CIFAR-10) (Hoffer et al., 2017; Keskar et al., 2016). By increasing the batch size past this point, it becomes immediately apparent that the primary issue with large batch size optimization is training speed, not the generalization gap.
+
+![](images/96bab39fb3c982480548f52bd526a04980c2093b687c837893866a29eef23e63.jpg)  
+Figure 4: Speedup curves as dataset size varies for different datasets. Even as dataset size increases back up to the baseline of $1 0 0 \%$ , there is no noticeable improvement in convergence speed.
+
+In order to test whether the sublinear behavior of $s ( m ; m _ { 0 } )$ depends primarily on dataset size, we compare the speedup curves obtained when training a single model on different fractions of the original training data. We trained ResNet34 models on the CIFAR-10 and SVHN datasets (for SVHN in this experiment, we train on all $6 0 0 k$ available training images). For each dataset, we trained on $1 0 0 \%$ , $5 0 \%$ , and then $2 5 \%$ of the available training data.
+
+In Figure 4, we plot the resulting speedup curves for the various partitions. In order to maintain a fair comparison (as baseline loss values change for different dataset sizes), we again choose the loss threshold to be the lower quartile of loss values obtained by the largest batch size.1 Notably, the batch size at which the curves begin to plateau remains constant as dataset size changes. For ResNet34 on CIFAR-10, the linear speedup behavior breaks around batch size 128 for all three curves. By a batch size of 1024, all curves have flattened. We can see similar behavior for ResNet34 on SVHN. Overall, looking back to Figure 3, the choice of model and the complexity of the dataset appear to be more related to the shape of speedup curve than dataset size alone.
+
+# 5 CONCLUSION
+
+By experimenting across a wide range of network architectures and problem domains, we find that, after a certain point, increasing the batch size fails to decrease wall-clock time to convergence and results in low computational efficiency, even assuming perfect parallelism. The critical batch size after which these returns diminish tends to be small relative to existing system capabilities. These trends present impediments to progress in developing effective machine learning systems that are capable of handling growing data demands.
+
+Recent works also suggest heuristics to decrease the generalization gap, but we find that these heuristics cannot be used to solve the underlying issue of training convergence speed. Moreover, we find that they usually only help decrease the generalization error in a small-to-medium batch size regime. There does not seem to be a simple training heuristic to improve large batch performance in general.
+
+These results suggest that we should not assume that increasing the batch size for larger datasets will keep training times manageable for all problems. Even though it is a natural form of data parallelism for large-scale optimization, alternative forms of parallelism should be explored to utilize all of our data more efficiently.
+
+# REFERENCES
+
+Yoshua Bengio and Yann LeCun. Scaling Learning Algorithms Towards AI. In Large Scale Kernel Machines. MIT Press, 2007.
+
+Leon Bottou. Large-scale machine learning with stochastic gradient descent. In ´ Proceedings of COMPSTAT’2010, pp. 177–186. Springer, 2010.
+
+Pratik Chaudhari and Stefano Soatto. Stochastic gradient descent performs variational inference, converges to limit cycles for deep networks. arXiv:1710.11029, 10 2017.
+
+Marius Cordts, Mohamed Omran, Sebastian Ramos, Timo Rehfeld, Markus Enzweiler, Rodrigo Benenson, Uwe Franke, Stefan Roth, and Bernt Schiele. The Cityscapes Dataset for Semantic Urban Scene Understanding. In Proc. of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2016.
+
+Dipankar Das, Sasikanth Avancha, Dheevatsa Mudigere, Karthikeyan Vaidyanathan, Srinivas Sridharan, Dhiraj D. Kalamkar, Bharat Kaul, and Pradeep Dubey. Distributed Deep Learning Using Synchronous Stochastic Gradient Descent. arXiv:1602.06709, 2016.
+
+Jia Deng, Wei Dong, Richard Socher, Li-Jia Li, Kai Li, and Li Fei-Fei. ImageNet: A large-scale hierarchical image database. In Computer Vision and Pattern Recognition, 2009. CVPR 2009. IEEE Conference on, pp. 248–255. IEEE, 2009.
+
+Aditya Devarakonda, Maxim Naumov, and Michael Garland. AdaBatch: Adaptive Batch Sizes for Training Deep Neural Networks. arXiv:1712.02029, 2017.
+
+Felix A. Gers, Jurgen A. Schmidhuber, and Fred A. Cummins. Learning to Forget: Continual ¨ Prediction with LSTM. Neural Computing, 12(10):2451–2471, 10 2000.
+
+Priya Goyal, Piotr Dollar, Ross Girshick, Pieter Noordhuis, Lukasz Wesolowski, Aapo Kyrola, An-´ drew Tulloch, Yangqing Jia, and Kaiming He. Accurate, large minibatch SGD: training ImageNet in 1 hour. arXiv:1706.02677, 2017.
+
+Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image recognition. In Proceedings of the IEEE conference on computer vision and pattern recognition, pp. 770–778, 2016.
+
+Sepp Hochreiter and Jurgen Schmidhuber. Long Short-Term Memory. ¨ Neural Computing, 9(8): 1735–1780, 11 1997.
+
+Elad Hoffer, Itay Hubara, and Daniel Soudry. Train longer, generalize better: closing the generalization gap in large batch training of neural networks. In Advances in Neural Information Processing Systems, pp. 1731–1741, 2017.
+
+Forrest N. Iandola, Khalid Ashraf, Matthew W. Moskewicz, and Kurt Keutzer. FireCaffe: near-linear acceleration of deep neural network training on compute clusters. arXiv:1511.00175, 11 2015.
+
+Stanislaw Jastrzebski, Zachary Kenton, Devansh Arpit, Nicolas Ballas, Asja Fischer, Yoshua Bengio, and Amos J. Storkey. Three Factors Influencing Minima in SGD. arXiv:1711.04623, 11 2018.
+
+Xianyan Jia, Shutao Song, Wei He, Yangzihao Wang, Haidong Rong, Feihu Zhou, Liqiang Xie, Zhenyu Guo, Yuanzhou Yang, Liwei Yu, Tiegang Chen, Guangxiao Hu, Shaohuai Shi, and Xiaowen Chu. Highly Scalable Deep Learning Training System with Mixed-Precision: Training ImageNet in Four Minutes. arXiv:1807.11205, 07 2018.
+
+Peter H. Jin, Qiaochu Yuan, Forrest N. Iandola, and Kurt Keutzer. How to scale distributed deep learning? arXiv:1611.04581, 11 2016.
+
+Nitish Shirish Keskar, Dheevatsa Mudigere, Jorge Nocedal, Mikhail Smelyanskiy, and Ping Tak Peter Tang. On large-batch training for deep learning: Generalization gap and sharp minima. arXiv:1609.04836, 2016.
+
+Alex Krizhevsky. One weird trick for parallelizing convolutional neural networks. arXiv:1404.5997, 04 2014.
+
+Alex Krizhevsky and Geoffrey Hinton. Learning multiple layers of features from tiny images. Technical report, Citeseer, 2009.
+
+Alex Krizhevsky, Ilya Sutskever, and Geoffrey E. Hinton. ImageNet Classification with Deep Convolutional Neural Networks. In Proceedings of the 25th International Conference on Neural Information Processing Systems - Volume 1, NIPS’12, pp. 1097–1105. Curran Associates Inc., 2012.
+
+Siyuan Ma, Raef Bassily, and Mikhail Belkin. The Power of Interpolation: Understanding the Effectiveness of SGD in Modern Over-parametrized Learning. arXiv:1712.06559, 2017.
+
+Julian McAuley, Christopher Targett, Qinfeng Shi, and Anton van den Hengel. Image-Based Recommendations on Styles and Substitutes. In Proceedings of the 38th International ACM SIGIR Conference on Research and Development in Information Retrieval, SIGIR ’15, pp. 43–52. ACM, 2015.
+
+Stephen Merity, Caiming Xiong, James Bradbury, and Richard Socher. Pointer Sentinel Mixture Models. arXiv:1609.07843, 09 2016.
+
+Raul Puri, Robert Kirby, Nikolai Yakovenko, and Bryan Catanzaro. Large Scale Language Modeling: Converging on 40GB of Text in Four Hours. arXiv:1808.01371, 08 2018.
+
+Mark Sandler, Andrew G. Howard, Menglong Zhu, Andrey Zhmoginov, and Liang-Chieh Chen. Very Deep Convolutional Networks for Large-Scale Image Recognition. arXiv:1801.04381, 09 2014.
+
+Timothy Sauer. Numerical solution of stochastic differential equations in finance. In Handbook of computational finance, pp. 529–550. Springer, 2012.
+
+Karen Simonyan and Andrew Zisserman. Very Deep Convolutional Networks for Large-Scale Image Recognition. arXiv:1409.1556, 09 2014.
+
+Samuel L. Smith and Quoc V. Le. A Bayesian Perspective on Generalization and Stochastic Gradient Descent. In International Conference on Learning Representations, 2018.
+
+Samuel L Smith, Pieter-Jan Kindermans, and Quoc V Le. Don’t Decay the Learning Rate, Increase the Batch Size. arXiv:1711.00489, 2017.
+
+Chen Xing, Devansh Arpit, Christos Tsirigotis, and Yoshua Bengio. A Walk with SGD. arXiv:1802.08770, 2018.
+
+Zhewei Yao, Amir Gholami, Qi Lei, Kurt Keutzer, and Michael W Mahoney. Hessian-based Analysis of Large Batch Training and Robustness to Adversaries. arXiv:1802.08241, 2018.
+
+Yang You, Igor Gitman, and Boris Ginsburg. Large Batch Training of Convolutional Networks. arXiv:1708.03888, 2017a.
+
+Yang You, Zhao Zhang, Cho-Jui Hsieh, and James Demmel. 100-epoch ImageNet Training with AlexNet in 24 Minutes. arXiv:1709.05011, 2017b.
+
+Fisher Yu, Vladlen Koltun, and Thomas Funkhouser. Dilated Residual Networks. In 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pp. 636–644, 07 2017.
+
+Zhanxing Zhu, Jingfeng Wu, Bing Yu, Lei Wu, and Jinwen Ma. The Anisotropic Noise in Stochastic Gradient Descent: Its Behavior of Escaping from Minima and Regularization Effects. arXiv:1803.00195, 2018.
+
+![](images/e2d22d36762c86eb7114a858df3deb4cc64fe830391fe3ba4a35dd5954e98afa.jpg)  
+Figure 5: Contour plots of test losses for various problem domains on a log scale. The test losses for BLR are on the left, while the losses for the LSR strategy are on the right. Lighter colors indicate lower loss values. Since we train each batch size for a fixed number of epochs, the total number of training iterations scales down linearly. For each loss value, we can observe how many iterations it takes to converge to that value given a particular batch size, by tracing the level curve for the associated color. For all problems, there is a batch size after which the number of training iterations necessary to converge does not decrease.
\ No newline at end of file
diff --git a/parse/train/S1en0sRqKm/S1en0sRqKm_content_list.json b/parse/train/S1en0sRqKm/S1en0sRqKm_content_list.json
new file mode 100644
index 0000000000000000000000000000000000000000..e53c9a52e1032d45fc5fddf87900dd394bce579f
--- /dev/null
+++ b/parse/train/S1en0sRqKm/S1en0sRqKm_content_list.json
@@ -0,0 +1,1122 @@
+[
+    {
+        "type": "text",
+        "text": "ON THE COMPUTATIONAL INEFFICIENCY OF LARGE BATCH SIZES FOR STOCHASTIC GRADIENT DESCENT ",
+        "text_level": 1,
+        "bbox": [
+            176,
+            99,
+            823,
+            147
+        ],
+        "page_idx": 0
+    },
+    {
+        "type": "text",
+        "text": "Anonymous authors Paper under double-blind review ",
+        "bbox": [
+            184,
+            170,
+            400,
+            198
+        ],
+        "page_idx": 0
+    },
+    {
+        "type": "text",
+        "text": "ABSTRACT ",
+        "text_level": 1,
+        "bbox": [
+            454,
+            234,
+            544,
+            251
+        ],
+        "page_idx": 0
+    },
+    {
+        "type": "text",
+        "text": "Increasing the mini-batch size for stochastic gradient descent offers significant opportunities to reduce wall-clock training time, but there are a variety of theoretical and systems challenges that impede the widespread success of this technique (Das et al., 2016; Keskar et al., 2016). We investigate these issues, with an emphasis on time to convergence and total computational cost, through an extensive empirical analysis of network training across several architectures and problem domains, including image classification, image segmentation, and language modeling. Although it is common practice to increase the batch size in order to fully exploit available computational resources, we find a substantially more nuanced picture. Our main finding is that across a wide range of network architectures and problem domains, increasing the batch size beyond a certain point yields no decrease in wall-clock time to convergence for either train or test loss. This batch size is usually substantially below the capacity of current systems. We show that popular training strategies for large batch size optimization begin to fail before we can populate all available compute resources, and we show that the point at which these methods break down depends more on attributes like model architecture and data complexity than it does directly on the size of the dataset. ",
+        "bbox": [
+            233,
+            266,
+            764,
+            502
+        ],
+        "page_idx": 0
+    },
+    {
+        "type": "text",
+        "text": "1 INTRODUCTION ",
+        "text_level": 1,
+        "bbox": [
+            176,
+            530,
+            334,
+            546
+        ],
+        "page_idx": 0
+    },
+    {
+        "type": "text",
+        "text": "Mini-batch stochastic gradient descent (SGD) is the dominant optimization method for training deep neural networks (DNNs) (Bengio & LeCun, 2007; Bottou, 2010). In the face of unprecedented growth in dataset size, a large body of work has attempted to scale SGD to train DNN models on increasingly large datasets, while keeping wall-clock time manageable (Iandola et al., 2015; Goyal et al., 2017; Smith & Le, 2018; Devarakonda et al., 2017). The most common approach to train large models at scale is distributed synchronous mini-batch SGD, which exploits additional computational resources through data parallelism. This technique reduces wall-clock training time by increasing the mini-batch size, i.e., the number of examples used to compute a stochastic estimate of the gradient of the loss function at each training iteration, while holding the number of epochs constant. Proponents of large batch size training often argue that the merits stem from its ability to decrease wall-clock training time while maintaining final model performance. Indeed, an enormous amount of work has gone into designing systems that seem to operate under an assumption that equates large batch size training with machine learning at scale (Goyal et al., 2017; Jia et al., 2018; Puri et al., 2018). ",
+        "bbox": [
+            174,
+            563,
+            825,
+            756
+        ],
+        "page_idx": 0
+    },
+    {
+        "type": "text",
+        "text": "Increasing the batch size improves the scaling performance of SGD per epoch, and there are significant challenges in building efficient distributed systems that are able to exploit additional computational resources to use large batch sizes (Jia et al., 2018). However, even if we were able to address these systems challenges, there are still more fundamental limitations to this approach. Large batch sizes often negatively impact important performance metrics of interest, including total computational cost (which usually determines monetary cost) and prediction quality. ",
+        "bbox": [
+            174,
+            763,
+            823,
+            847
+        ],
+        "page_idx": 0
+    },
+    {
+        "type": "text",
+        "text": "In this paper, we will measure the total computational cost as the number of training iterations times the work done per iteration—in order to simplify measurements, we use the number of training iterations as a proxy for the wall-clock time. We do this because the implementation of parallel algorithms depends on software and hardware choices, and our goal is to draw more general conclusions about the performance of SGD-based methods. ",
+        "bbox": [
+            174,
+            854,
+            823,
+            922
+        ],
+        "page_idx": 0
+    },
+    {
+        "type": "text",
+        "text": "Based on this model for total computational cost and wall-clock time, the following should be clear: unless increasing the batch size leads to a commensurate decrease in the total number of training iterations needed to find a good model, large batch training will result in greater total computational cost with little-to-no decrease in wall-clock training time. ",
+        "bbox": [
+            174,
+            103,
+            823,
+            159
+        ],
+        "page_idx": 1
+    },
+    {
+        "type": "text",
+        "text": "Based on our empirical results across a range of datasets and architectures, we find that as the batch size becomes larger, there are three main phases of scaling behavior for convergence speed: (1) there is a small regime of batch sizes in which increasing the batch size results in linear gains in convergence speed; (2) there is a larger regime of batch sizes that results in sublinear gains in convergence speed—in this regime, increasing the batch size can improve wall-clock training time at the expense of greater total computational cost; (3) eventually, we reach a third regime where a higher batch size results in marginal or non-existent reductions in convergence speed. In our experiments, we find that this third regime begins at a batch size that is too small to let us fully utilize available compute. Training past this batch size increases the total computational cost without reducing wall-clock training time or prediction quality. While there has been considerable excitement around heuristics that have been shown to make large batch training practical for certain problems (Goyal et al., 2017; Smith & Le, 2018), we demonstrate that these techniques still suffer from the same convergence trends we observe, and they often decrease stability of the training process. ",
+        "bbox": [
+            174,
+            166,
+            825,
+            361
+        ],
+        "page_idx": 1
+    },
+    {
+        "type": "text",
+        "text": "Recent work has observed that the final test performance of models trained with large batch sizes degrades after training for a fixed number of epochs (Yao et al., 2018; Keskar et al., 2016). This phenomenon is known as the generalization gap. Previous work addressing this problem has focused on training for more iterations in the large batch case (Hoffer et al., 2017) or adopting various heuristics to select a learning rate for larger batch sizes (Goyal et al., 2017; Smith & Le, 2018). Based on our empirical results, we find that existing techniques to mitigate the generalization gap do not work on some problems, and for other problems they only work for batch sizes that do not allow us to fully utilize our available compute. Perhaps more importantly, they do little to affect the diminishing returns in rates of convergence for training loss as batch size increases. ",
+        "bbox": [
+            174,
+            367,
+            825,
+            493
+        ],
+        "page_idx": 1
+    },
+    {
+        "type": "text",
+        "text": "Our objective is to understand the behavior of SGD and existing large batch techniques for many network architectures and problem domains, e.g., image classification/segmentation and natural language processing (NLP). We observe markedly worse performance for these techniques in domains other than image classification, where large batch optimization has received the most attention (Jia et al., 2018; You et al., 2017b). Because we eschew the challenges of an efficient distributed implementation by measuring number of iterations instead of wall-clock time, our results assume the most optimistic circumstances for large batch training. Our key observations are: ",
+        "bbox": [
+            174,
+            500,
+            825,
+            597
+        ],
+        "page_idx": 1
+    },
+    {
+        "type": "text",
+        "text": "• Increasing the batch size beyond a certain point yields no improvement in wall-clock time to convergence, even for a system with perfect parallelism. We observe that larger batch sizes result in a limited reduction in the number of training iterations needed to achieve low training or test error, and that eventually these gains become near-zero. Increasing the batch size leads to a significant increase in generalization error, which cannot be mitigated by existing techniques. We observe that these techniques often result in divergent training behavior or that they only mitigate degradation in test performance for small batch sizes relative to available compute. Dataset size is not the only factor determining the computational efficiency of large batch training. We observe that both the diminishing returns in convergence speed and the failure of existing methods correlate more with factors like model architecture and data complexity than dataset size alone. As a result, training time may significantly increase with dataset size in spite of increasingly available compute resources. ",
+        "bbox": [
+            215,
+            613,
+            826,
+            818
+        ],
+        "page_idx": 1
+    },
+    {
+        "type": "text",
+        "text": "In Section 2, we review the formulation of SGD as well as existing strategies to train with large batch sizes. In Section 3, we review recent theoretical results regarding the convergence rates of SGD in highly over-parameterized settings and discuss the potential impact of these results on the computational efficiency of SGD for deep learning. Section 4 presents our empirical results that demonstrate the inefficiencies of training SGD with large batch sizes, and we show that these persist when using existing large batch optimization techniques. ",
+        "bbox": [
+            174,
+            840,
+            825,
+            924
+        ],
+        "page_idx": 1
+    },
+    {
+        "type": "text",
+        "text": "2 BACKGROUND AND RELATED WORK ",
+        "text_level": 1,
+        "bbox": [
+            174,
+            102,
+            513,
+            118
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "text",
+        "text": "Stochastic Gradient Descent. SGD is the most widely used algorithm to train DNN models. The model is parameterized by weights $\\mathbf { w } \\in \\mathbb { R } ^ { d }$ , and the objective is to minimize the empirical loss over $n$ data points $\\mathbf { x } _ { i }$ : ",
+        "bbox": [
+            173,
+            133,
+            825,
+            175
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "equation",
+        "img_path": "images/887595ea6a4113fb0007840a5a7f971572079b5726c91bec59fd88f90a821b85.jpg",
+        "text": "$$\nL ( \\mathbf { w } ) = \\frac { 1 } { n } \\sum _ { i = 1 } ^ { n } \\ell ( \\mathbf { w } , \\mathbf { x } _ { i } ) ,\n$$",
+        "text_format": "latex",
+        "bbox": [
+            415,
+            172,
+            581,
+            215
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "text",
+        "text": "where $\\ell ( \\cdot , \\cdot )$ is a loss, e.g., cross-entropy or squared error. This loss gives a corresponding gradient ",
+        "bbox": [
+            176,
+            217,
+            820,
+            233
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "equation",
+        "img_path": "images/5ce375a0c0b4eba72df983bc418c05ba2ee824ff817e740a7b4a68960b2471aa.jpg",
+        "text": "$$\n\\mathbf { g } ( \\mathbf { w } ) : = \\nabla L ( \\mathbf { w } ) = \\frac { 1 } { n } \\sum _ { i = 1 } ^ { n } \\nabla \\ell ( \\mathbf { w } , \\mathbf { x } _ { i } ) .\n$$",
+        "text_format": "latex",
+        "bbox": [
+            370,
+            238,
+            627,
+            280
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "text",
+        "text": "A mini-batch $B _ { m }$ of size $m < n$ is a collection of $m$ indices randomly drawn from the set $\\{ 1 , \\ldots , n \\}$ , and we can use it to form an unbiased estimate of the gradient at iteration $k$ , as well as the corresponding SGD update: ",
+        "bbox": [
+            174,
+            286,
+            825,
+            329
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "equation",
+        "img_path": "images/83e3f1f25ec3b8d3387d926f3bff0f8c4141099ab08b61b4c0c318b9c7452052.jpg",
+        "text": "$$\n\\mathbf { g } _ { m } ( \\mathbf { w } _ { k } ) = \\frac { 1 } { m } \\sum _ { i \\in \\mathcal { B } _ { m } } \\nabla \\ell ( \\mathbf { w } _ { k } , \\mathbf { x } _ { i } ) \\quad \\mathrm { ~ a n d ~ } \\quad \\mathbf { w } _ { k + 1 } = \\mathbf { w } _ { k } - \\eta _ { k } \\mathbf { g } _ { m } ( \\mathbf { w } _ { k } ) ,\n$$",
+        "text_format": "latex",
+        "bbox": [
+            266,
+            334,
+            730,
+            373
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "text",
+        "text": "where $\\eta _ { k } > 0$ is the learning rate for iteration $k$ . One iteration of training for SGD corresponds to a single gradient computation / weight update. One epoch corresponds to $n / m$ iterations of training. This constitutes a single pass over the dataset, assuming the dataset is sampled without replacement. ",
+        "bbox": [
+            176,
+            378,
+            825,
+            421
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "text",
+        "text": "Efficient distributed systems reduce wall-clock training time by parallelizing gradient calculations across many machines. When the batch size is large enough to populate all available compute resources, this allows us to amortize the cost of coordination for each weight update. ",
+        "bbox": [
+            176,
+            428,
+            821,
+            470
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "text",
+        "text": "Existing large batch techniques. With the hope of keeping training times manageable as dataset sizes escalate, recent work has focused on the development of techniques that allow practitioners to increase the batch size to make use of growing computational resources (Jin et al., 2016; Jia et al., 2018; You et al., 2017a). However, there is a growing body of theoretical and empirical results suggesting that large batch sizes adversely affect the generalization performance of the final model (Yao et al., 2018; Keskar et al., 2016; Devarakonda et al., 2017). ",
+        "bbox": [
+            173,
+            477,
+            825,
+            561
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "text",
+        "text": "In response to this, recent work has proposed changing two parameters in relation to batch size: the number of training iterations and the learning rate. However, they also make assumptions that limit the effectiveness of their proposals as useful heuristics for practitioners. ",
+        "bbox": [
+            176,
+            568,
+            823,
+            609
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "text",
+        "text": "• Training longer: Hoffer et al. (2017) suggest increasing the number of training iterations. Even if this does reduce the generalization gap, it significantly increases both wall-clock training time and computational cost. Moreover, in some problems it does not lead to minima with better generalization performance (as we found when running our experiments). • Square root LR scaling: Scaling the learning rate as $\\eta _ { 0 } \\propto \\sqrt { m }$ attempts to keep the weight increment length statistics constant, but the distance between SGD iterates is governed more by properties of the objective function than the ratio of learning rate to batch size (Chaudhari & Soatto, 2017; Zhu et al., 2018). This rule has also been found to be empirically sub-optimal in various problem domains (Krizhevsky, 2014). Linear LR scaling: The performance of large batch training can also be improved by using the linear scaling rule, which suggests choosing a learning rate proportional to the batch size $( \\eta _ { 0 } \\propto m )$ (Goyal et al., 2017). There are two motivations for this rule: the first assumes that one large-batch gradient step should resemble a series of small-batch gradient steps in order for convergence rates to improve linearly (Goyal et al., 2017); the other regards the SGD update equation as the Euler-Maruyama discretization of a stochastic differential equation (Sauer, 2012; Xing et al., 2018), and attempts to maintain a constant level of minibatch noise to help SGD explore the loss landscape (Chaudhari & Soatto, 2017; Zhu et al., 2018; Smith & Le, 2018). ",
+        "bbox": [
+            215,
+            622,
+            825,
+            882
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "text",
+        "text": "Both justifications for the linear scaling rule implicitly impose strong conditions on the loss function by requiring that it behave linearly near SGD iterates; therefore, if the loss function is highly nonlinear along the SGD trajectory or the step size is not small enough, then we should not expect these rules to provide useful guidance for many problems. Whereas several groups have successfully used this rule to train on the ImageNet dataset in under an hour, e.g. (Goyal et al., 2017; You et al., 2017b), applying this heuristic to other datasets has not led to similarly impressive results so far (Puri et al., 2018). ",
+        "bbox": [
+            176,
+            895,
+            820,
+            924
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "text",
+        "text": "",
+        "bbox": [
+            174,
+            103,
+            825,
+            172
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "The focus of this paper, however, is on more fundamental limitations of large batch training, and we empirically show that the above approaches fail to prevent diminishing returns in the rate of convergence for large batch sizes. We believe that these diminishing returns are of more immediate concern than the generalization gap and warrant more careful examination: if we cannot even minimize training error quickly, there is no real opportunity to minimize test error quickly, regardless of the difference in final test error across batch sizes by the time the model has converged. ",
+        "bbox": [
+            174,
+            180,
+            825,
+            263
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "3 CRITICAL BATCH SIZES AND DIMINISHING RETURNS ",
+        "text_level": 1,
+        "bbox": [
+            174,
+            292,
+            650,
+            309
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "The convergence rate of SGD, denoted by $k _ { \\epsilon } ( m )$ , is the number of iterations needed to achieve training error less than a fixed constant $\\epsilon > 0$ by using SGD with batch size $m$ (we will drop the subscript $\\epsilon$ when it is unambiguous). In order to guarantee that large batch sizes speed up training, $k ( m )$ should continue to decrease near-linearly with $m$ . Otherwise, a larger batch size increases computational cost with only limited reductions in wall-clock training time. For near-constant $k ( m )$ , the benefit of large batch sizes becomes near-zero. ",
+        "bbox": [
+            174,
+            329,
+            825,
+            412
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "Ma et al. (2017) showed theoretically that in convex, over-parameterized settings, the reduction in convergence time obtained by increasing the batch size decays dramatically to a near-constant level after a critical batch size that is independent of the dataset size. This speedup is measured with respect to the number of SGD iterations required to reach some fixed loss error for some baseline batch size $m _ { 0 }$ , and for this purpose we define the speedup ratio $s ( m ; m _ { 0 } ) = k ( m _ { 0 } ) / k ( m )$ . The speedup ratio represents the amount of time we save by increasing the batch size to $m$ . Beyond the critical batch size mentioned above, even with no communication overhead and unlimited resources (where each batch size requires the same amount of wall-clock time to process) we would prefer to use the critical batch size because it requires less overall computation. ",
+        "bbox": [
+            174,
+            420,
+            825,
+            545
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "This result is surprising because researchers have asserted that it should be possible to achieve linear gains in convergence speed so long as the batch size is small relative to dataset size (Smith & Le, 2018). This will present significant difficulties for future optimization work (large mini-batch training) because it prevents us from using large batch sizes as a catch-all approach to quickly train models on large datasets. ",
+        "bbox": [
+            174,
+            553,
+            823,
+            622
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "4 EMPIRICAL EVALUATION ",
+        "text_level": 1,
+        "bbox": [
+            176,
+            651,
+            415,
+            666
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "Recent work studying large batch training has looked primarily at image classification (Jastrzebski et al., 2018; Yao et al., 2018), especially on the ImageNet dataset (Deng et al., 2009). We perform large batch size experiments across both traditional image classification (IC) tasks (such as on CIFAR-10/100 (Krizhevsky & Hinton, 2009)), as well as previously unexplored tasks like image segmentation (IS) using the Cityscapes dataset (Cordts et al., 2016), and natural language processing (NLP) using the WikiText-2 dataset (Merity et al., 2016). We also test how these results vary across other modern DNN architectures, namely ResNets (He et al., 2016), LSTMs (Hochreiter & Schmidhuber, 1997; Gers et al., 2000), AlexNet (Krizhevsky et al., 2012), VGG (Simonyan & Zisserman, 2014), Dilated Residual Networks (Yu et al., 2017), and MobileNetV2 (Sandler et al., 2014). We tested all of the large batch training techniques described in Section 2. We tried training longer based on the work of Hoffer et al. (2017), but we found that this necessarily cannot improve the convergence speed and often does not improve final test performance. The two other techniques include the square root scaling rule strategy (SRSR) and the linear scaling rule strategy (LSR). For the latter, we used a warm-up period at the start of training as suggested by Goyal et al. (2017). Table 1 reports our datasets, models and different training strategies. For each model, we evaluated against a base learning rate strategy (BLR) that used the same learning rate across all batch sizes. We selected this learning rate based on its performance on a small baseline batch size. ",
+        "bbox": [
+            174,
+            688,
+            825,
+            924
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "image",
+        "img_path": "images/b45c386a5ec3b1dee617dd3ed5dfe50c26a309dd84ec1efd0f4dcdac283a893c.jpg",
+        "image_caption": [
+            "Figure 1: Contour plots of training losses for various problem domains on a log scale. Lighter colors indicate lower loss values. Since we train each batch size for a fixed number of epochs, the total number of training iterations scales down linearly. For each loss value, we can observe how many iterations it takes to converge to that value given a particular batch size, by tracing the level curve for the associated color. For all problems, there is a batch size after which the number of training iterations necessary to converge does not decrease. "
+        ],
+        "image_footnote": [],
+        "bbox": [
+            187,
+            107,
+            810,
+            667
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "text",
+        "text": "4.1 DIMINISHING RETURNS IN RATES OF CONVERGENCE ",
+        "text_level": 1,
+        "bbox": [
+            174,
+            776,
+            584,
+            791
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "text",
+        "text": "We demonstrate the rapidly diminishing returns in rates of convergence across various problem domains and network configurations. Researchers increase the batch size in an attempt to achieve nearly linear speedups in convergence compared to a small mini-batch size. In particular, if the speedup is near-linear, i.e. $s ( m ; \\bar { m } _ { 0 } ) = k ( \\bar { m } _ { 0 } ) / k ( m ) \\approx m / m _ { 0 }$ , then the computational cost remains nearly constant for large and small mini-batch SGD. However, if $s ( m ) \\ \\bar { \\ll } \\ m / m _ { 0 }$ , then the benefit of using large batch size training is negligible. ",
+        "bbox": [
+            174,
+            804,
+            825,
+            888
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "text",
+        "text": "In Figure 1, we show contour plots of training loss as a function of both the batch size and the number of training iterations of ResNet34 on CIFAR-10, an LSTM on WikiText-2, and DRN-D-22 on Cityscapes. Consider, for example, the contour plot for ResNet34 trained on CIFAR-10. We can see that as the batch size increases from 16 to 2048, the number of SGD iterations needed to achieve a particular loss value decreases linearly. Exceeding this regime, however, the speedup ratio becomes increasingly sublinear and soon we have $s ( m ; \\bar { m _ { 0 } } ) \\ll \\bar { m } / m _ { 0 }$ . For batch size 8196, the training procedure does not achieve the lowest training loss, and from this perspective, even if we did not care about computational cost or training time, we would not be able to find an accurate model. We observe even worse scaling behavior for test performance (please see Figure 5 for details). ",
+        "bbox": [
+            174,
+            895,
+            823,
+            924
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "table",
+        "img_path": "images/6850766aebd7ac2141d3cf5d090844fbc7e2c84a240d31b09a191a647538a886.jpg",
+        "table_caption": [
+            "Table 1: A description of the problem configurations and training strategies used in this paper. $\\eta _ { 0 }$ is the initial learning rate, $W$ is the number of epochs used for warm-up in the linear scaling rule, $E$ is the total number of epochs trained "
+        ],
+        "table_footnote": [],
+        "table_body": "<table><tr><td>Dataset</td><td>Task</td><td>Architecture</td><td>Training Strategy</td><td>BS range</td></tr><tr><td> MNIST</td><td>IC</td><td>ResNet34</td><td>BLR,LSR (no = 0.1, W = 10, E = 200)</td><td>26 -214</td></tr><tr><td>CIFAR-10</td><td>IC</td><td>AlexNet, MobileNetV2 ResNet34, VGG16</td><td>BLR,LSR, SRSR (mo = 0.1,W=10,E= 200)</td><td>26-214</td></tr><tr><td>CIFAR-100</td><td>IC</td><td>ResNet34</td><td>BLR,LSR (no = 0.1,W = 10, E = 200)</td><td>2 -214</td></tr><tr><td>SVHN</td><td>IC</td><td>ResNet34</td><td>BLR,LSR (no =0.1,W =10,E= 200)</td><td>26-214</td></tr><tr><td> WikiText-2</td><td> NLP</td><td>LSTM</td><td>BLR,LSR (no = 20,W = 3,E = 40)</td><td>2³ -210</td></tr><tr><td>Cityscapes</td><td>IS</td><td>DRN-D-22</td><td>BLR,LSR (no = 0.01,W = 10,E= 100)</td><td>23 -211</td></tr></table>",
+        "bbox": [
+            174,
+            151,
+            823,
+            273
+        ],
+        "page_idx": 5
+    },
+    {
+        "type": "image",
+        "img_path": "images/2f39da181bbafada9ee2bf61d16e3221282ee2f410345f6758d773bf18d978e5.jpg",
+        "image_caption": [
+            "Figure 2: On the left: speedup curves when applying several popular techniques to avoid the generalization gap. Base LR uses a single learning rate for all batch sizes. On the right: the effect of the linear approximation error on final test accuracy when using the linear LR scaling rule. "
+        ],
+        "image_footnote": [],
+        "bbox": [
+            194,
+            297,
+            803,
+            462
+        ],
+        "page_idx": 5
+    },
+    {
+        "type": "text",
+        "text": "",
+        "bbox": [
+            174,
+            542,
+            825,
+            640
+        ],
+        "page_idx": 5
+    },
+    {
+        "type": "text",
+        "text": "For NLP and IS, note that the gain from large batch training diminishes even faster. Neither the LSTM on WikiText-2 nor DRN-D-22 on Cityscapes can reach their respective baseline performances after reasonably small batch sizes of about 32 and 64, respectively. Although Puri et al. (2018) showed that training on the Amazon Reviews dataset (McAuley et al., 2015) can be done within 4 hours, they tune hyper-parameters heavily. This poses an issue for many practical deployments because these problems are often already slow to train. ",
+        "bbox": [
+            174,
+            647,
+            825,
+            731
+        ],
+        "page_idx": 5
+    },
+    {
+        "type": "text",
+        "text": "4.2 EXISTING STRATEGIES BREAK DOWN FOR LARGE BATCH SIZES ",
+        "text_level": 1,
+        "bbox": [
+            176,
+            751,
+            660,
+            765
+        ],
+        "page_idx": 5
+    },
+    {
+        "type": "text",
+        "text": "We further explore how training with the linear and square root scaling rules compares to training with a fixed baseline learning rate (BLR) that does not change with batch size. In the left subfigure of Figure 2, we show the speedup curves of BLR, LSR, and SRSR strategies for ResNet34 on CIFAR10. Note that LSR and SRSR outperform BLR from batch size 256 to 2048 which implies that LSR and SRSR can help the model train for small-to-medium batch sizes. However, the speedup of LSR and SRSR is still worse than the ideal linear case, and the curves plateau quickly after a batch size of 2048, at which point BLR becomes better than LSR and SRSR. This means that for certain problems, scaling up the learning rate to compensate for an increased batch size hurts performance. ",
+        "bbox": [
+            174,
+            776,
+            825,
+            888
+        ],
+        "page_idx": 5
+    },
+    {
+        "type": "text",
+        "text": "In the right subfigure of Figure 2, we plot the test performance and the approximation error for LSR of ResNet34 on CIFAR-10. We measure the approximation error at the end of training, with final weights $\\mathbf { w } ^ { * }$ . We take this error to be the absolute difference between the true loss value $L ( \\mathbf { w } )$ and the linear approximation at $\\mathbf { w } ^ { * }$ , given by $\\hat { L } ( \\mathbf { w } ) = L ( \\mathbf { w } ^ { * } ) + \\langle \\mathbf { g } _ { m } ( \\mathbf { w } ^ { * } ) , \\mathbf { w } - \\mathbf { w } ^ { * } \\rangle$ . The approximation is calculated for $\\mathbf { w } = \\mathbf { w } ^ { * } - \\eta \\frac { m } { m _ { 0 } } \\mathbf { g } _ { m } \\big ( \\mathbf { w } ^ { * } \\big )$ to understand the behavior of the approximation along the trajectory for a single SGD iterate using the LSR. It appears that there exists a strong relationship between linear approximation error and test accuracy: as the linear approximation error increases, the test accuracy drops. Note the transition that happens at the critical batch size of 2048. After this point, the test accuracy drops significantly and the linear approximation error exceeds 1, showing that we quickly exit the regime in which the linear approximation is valid. ",
+        "bbox": [
+            173,
+            895,
+            821,
+            924
+        ],
+        "page_idx": 5
+    },
+    {
+        "type": "image",
+        "img_path": "images/8e5a9c2868609d6811e018f9aa938f74636a8e7c872b18be0773d9a10cb99683.jpg",
+        "image_caption": [
+            "Figure 3: Speedup curves across different problem configurations. Left: different architectures result in different rates of convergence on CIFAR-10. Right: ResNet34 exhibits different rates of convergence on CIFAR-10, CIFAR-100, and SVHN. Loss thresholds are obtained by computing the lower quartile of loss values achieved by the largest batch size. "
+        ],
+        "image_footnote": [],
+        "bbox": [
+            236,
+            113,
+            759,
+            258
+        ],
+        "page_idx": 6
+    },
+    {
+        "type": "text",
+        "text": "",
+        "bbox": [
+            173,
+            352,
+            825,
+            468
+        ],
+        "page_idx": 6
+    },
+    {
+        "type": "text",
+        "text": "4.3 CONVERGENCE SPEED HAS A WEAK DEPENDENCE ON DATASET SIZE ",
+        "text_level": 1,
+        "bbox": [
+            176,
+            487,
+            686,
+            501
+        ],
+        "page_idx": 6
+    },
+    {
+        "type": "text",
+        "text": "Previous works have conjectured that the maximum batch size that can result in a good model is proportional to the size of the whole dataset (Smith et al., 2017; Smith & Le, 2018). However, for convex, over-parameterized problems, Ma et al. (2017) show that there is a model-dependent critical batch size after which we observe rapidly diminishing returns in convergence speed. In this section, to observe if a similar critical batch size exists in the non-convex case, we compare how changing model architecture or data complexity affects the shapes of speedup curves compared to changing the dataset size alone. ",
+        "bbox": [
+            174,
+            513,
+            825,
+            609
+        ],
+        "page_idx": 6
+    },
+    {
+        "type": "text",
+        "text": "First, in order to show that these diminishing returns depend on data complexity and DNN architecture, we plot speedup curves in Figure 3 to compare the scaling behaviors across different models and dataset configurations. For the error threshold \u000f, we chose the lowest quartile loss value reached by the largest batch size to make a fair comparison across configurations. This setup actually favors the large batch case, because there are lower loss thresholds that are attainable only in the small batch case. On the left, for the CIFAR-10 dataset, we compared four model architectures. For each architecture, we plotted the speedup curve obtained by training this model on the dataset for various batch sizes. The variety of speedup curve shapes indicates that model architecture is an important factor in determining the convergence speed of training for large batch sizes. For MobileNetV2/AlexNet, the diminishing returns become visible when batch size is 1024. However, for VGG16/ResNet34, the speedup does not flatten out until batch size 8196. Hence, in practice, the choice of model strongly affects our ability to use large batch sizes in SGD. ",
+        "bbox": [
+            173,
+            617,
+            825,
+            784
+        ],
+        "page_idx": 6
+    },
+    {
+        "type": "text",
+        "text": "On the right, in order to investigate the effect of problem complexity, we compared the performance of ResNet34 on four datasets of the same size: CIFAR-10, CIFAR-100, MNIST, and the SVHN dataset (we cut off MNIST and SVHN to $5 0 k$ training examples each). Although all problems display diminishing returns in rates of convergence, the point at which the curves plateau varies according to problem complexity. It is not hard to see that, for simpler problems such as SVHN, the curves flatten out later than for harder problems (e.g. CIFAR-10/100). ",
+        "bbox": [
+            174,
+            791,
+            825,
+            875
+        ],
+        "page_idx": 6
+    },
+    {
+        "type": "text",
+        "text": "In all of the above cases, the diminishing rates of return in convergence speed become visible after only moderate increases in the batch size. Previous works have only studied convergence behavior for a fairly limited range of batch sizes (e.g., up to 4096 for CIFAR-10) (Hoffer et al., 2017; Keskar et al., 2016). By increasing the batch size past this point, it becomes immediately apparent that the primary issue with large batch size optimization is training speed, not the generalization gap. ",
+        "bbox": [
+            176,
+            882,
+            823,
+            924
+        ],
+        "page_idx": 6
+    },
+    {
+        "type": "image",
+        "img_path": "images/96bab39fb3c982480548f52bd526a04980c2093b687c837893866a29eef23e63.jpg",
+        "image_caption": [
+            "Figure 4: Speedup curves as dataset size varies for different datasets. Even as dataset size increases back up to the baseline of $1 0 0 \\%$ , there is no noticeable improvement in convergence speed. "
+        ],
+        "image_footnote": [],
+        "bbox": [
+            251,
+            113,
+            745,
+            261
+        ],
+        "page_idx": 7
+    },
+    {
+        "type": "text",
+        "text": "",
+        "bbox": [
+            171,
+            330,
+            823,
+            358
+        ],
+        "page_idx": 7
+    },
+    {
+        "type": "text",
+        "text": "In order to test whether the sublinear behavior of $s ( m ; m _ { 0 } )$ depends primarily on dataset size, we compare the speedup curves obtained when training a single model on different fractions of the original training data. We trained ResNet34 models on the CIFAR-10 and SVHN datasets (for SVHN in this experiment, we train on all $6 0 0 k$ available training images). For each dataset, we trained on $1 0 0 \\%$ , $5 0 \\%$ , and then $2 5 \\%$ of the available training data. ",
+        "bbox": [
+            174,
+            364,
+            825,
+            435
+        ],
+        "page_idx": 7
+    },
+    {
+        "type": "text",
+        "text": "In Figure 4, we plot the resulting speedup curves for the various partitions. In order to maintain a fair comparison (as baseline loss values change for different dataset sizes), we again choose the loss threshold to be the lower quartile of loss values obtained by the largest batch size.1 Notably, the batch size at which the curves begin to plateau remains constant as dataset size changes. For ResNet34 on CIFAR-10, the linear speedup behavior breaks around batch size 128 for all three curves. By a batch size of 1024, all curves have flattened. We can see similar behavior for ResNet34 on SVHN. Overall, looking back to Figure 3, the choice of model and the complexity of the dataset appear to be more related to the shape of speedup curve than dataset size alone. ",
+        "bbox": [
+            174,
+            443,
+            825,
+            554
+        ],
+        "page_idx": 7
+    },
+    {
+        "type": "text",
+        "text": "5 CONCLUSION ",
+        "text_level": 1,
+        "bbox": [
+            176,
+            574,
+            318,
+            589
+        ],
+        "page_idx": 7
+    },
+    {
+        "type": "text",
+        "text": "By experimenting across a wide range of network architectures and problem domains, we find that, after a certain point, increasing the batch size fails to decrease wall-clock time to convergence and results in low computational efficiency, even assuming perfect parallelism. The critical batch size after which these returns diminish tends to be small relative to existing system capabilities. These trends present impediments to progress in developing effective machine learning systems that are capable of handling growing data demands. ",
+        "bbox": [
+            174,
+            604,
+            825,
+            689
+        ],
+        "page_idx": 7
+    },
+    {
+        "type": "text",
+        "text": "Recent works also suggest heuristics to decrease the generalization gap, but we find that these heuristics cannot be used to solve the underlying issue of training convergence speed. Moreover, we find that they usually only help decrease the generalization error in a small-to-medium batch size regime. There does not seem to be a simple training heuristic to improve large batch performance in general. ",
+        "bbox": [
+            174,
+            695,
+            825,
+            752
+        ],
+        "page_idx": 7
+    },
+    {
+        "type": "text",
+        "text": "These results suggest that we should not assume that increasing the batch size for larger datasets will keep training times manageable for all problems. Even though it is a natural form of data parallelism for large-scale optimization, alternative forms of parallelism should be explored to utilize all of our data more efficiently. ",
+        "bbox": [
+            174,
+            758,
+            823,
+            815
+        ],
+        "page_idx": 7
+    },
+    {
+        "type": "text",
+        "text": "REFERENCES ",
+        "text_level": 1,
+        "bbox": [
+            176,
+            102,
+            287,
+            118
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "Yoshua Bengio and Yann LeCun. Scaling Learning Algorithms Towards AI. In Large Scale Kernel Machines. MIT Press, 2007. ",
+        "bbox": [
+            173,
+            126,
+            825,
+            155
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "Leon Bottou. Large-scale machine learning with stochastic gradient descent. In ´ Proceedings of COMPSTAT’2010, pp. 177–186. Springer, 2010. ",
+        "bbox": [
+            171,
+            165,
+            825,
+            194
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "Pratik Chaudhari and Stefano Soatto. Stochastic gradient descent performs variational inference, converges to limit cycles for deep networks. arXiv:1710.11029, 10 2017. ",
+        "bbox": [
+            171,
+            203,
+            823,
+            232
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "Marius Cordts, Mohamed Omran, Sebastian Ramos, Timo Rehfeld, Markus Enzweiler, Rodrigo Benenson, Uwe Franke, Stefan Roth, and Bernt Schiele. The Cityscapes Dataset for Semantic Urban Scene Understanding. In Proc. of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2016. ",
+        "bbox": [
+            173,
+            241,
+            826,
+            297
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "Dipankar Das, Sasikanth Avancha, Dheevatsa Mudigere, Karthikeyan Vaidyanathan, Srinivas Sridharan, Dhiraj D. Kalamkar, Bharat Kaul, and Pradeep Dubey. Distributed Deep Learning Using Synchronous Stochastic Gradient Descent. arXiv:1602.06709, 2016. ",
+        "bbox": [
+            174,
+            308,
+            825,
+            351
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "Jia Deng, Wei Dong, Richard Socher, Li-Jia Li, Kai Li, and Li Fei-Fei. ImageNet: A large-scale hierarchical image database. In Computer Vision and Pattern Recognition, 2009. CVPR 2009. IEEE Conference on, pp. 248–255. IEEE, 2009. ",
+        "bbox": [
+            173,
+            359,
+            825,
+            404
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "Aditya Devarakonda, Maxim Naumov, and Michael Garland. AdaBatch: Adaptive Batch Sizes for Training Deep Neural Networks. arXiv:1712.02029, 2017. ",
+        "bbox": [
+            173,
+            412,
+            825,
+            441
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "Felix A. Gers, Jurgen A. Schmidhuber, and Fred A. Cummins. Learning to Forget: Continual ¨ Prediction with LSTM. Neural Computing, 12(10):2451–2471, 10 2000. ",
+        "bbox": [
+            173,
+            450,
+            825,
+            481
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "Priya Goyal, Piotr Dollar, Ross Girshick, Pieter Noordhuis, Lukasz Wesolowski, Aapo Kyrola, An-´ drew Tulloch, Yangqing Jia, and Kaiming He. Accurate, large minibatch SGD: training ImageNet in 1 hour. arXiv:1706.02677, 2017. ",
+        "bbox": [
+            174,
+            489,
+            823,
+            532
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image recognition. In Proceedings of the IEEE conference on computer vision and pattern recognition, pp. 770–778, 2016. ",
+        "bbox": [
+            173,
+            541,
+            823,
+            584
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "Sepp Hochreiter and Jurgen Schmidhuber. Long Short-Term Memory. ¨ Neural Computing, 9(8): 1735–1780, 11 1997. ",
+        "bbox": [
+            173,
+            594,
+            821,
+            623
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "Elad Hoffer, Itay Hubara, and Daniel Soudry. Train longer, generalize better: closing the generalization gap in large batch training of neural networks. In Advances in Neural Information Processing Systems, pp. 1731–1741, 2017. ",
+        "bbox": [
+            173,
+            632,
+            825,
+            676
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "Forrest N. Iandola, Khalid Ashraf, Matthew W. Moskewicz, and Kurt Keutzer. FireCaffe: near-linear acceleration of deep neural network training on compute clusters. arXiv:1511.00175, 11 2015. ",
+        "bbox": [
+            173,
+            685,
+            821,
+            714
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "Stanislaw Jastrzebski, Zachary Kenton, Devansh Arpit, Nicolas Ballas, Asja Fischer, Yoshua Bengio, and Amos J. Storkey. Three Factors Influencing Minima in SGD. arXiv:1711.04623, 11 2018. ",
+        "bbox": [
+            173,
+            723,
+            825,
+            766
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "Xianyan Jia, Shutao Song, Wei He, Yangzihao Wang, Haidong Rong, Feihu Zhou, Liqiang Xie, Zhenyu Guo, Yuanzhou Yang, Liwei Yu, Tiegang Chen, Guangxiao Hu, Shaohuai Shi, and Xiaowen Chu. Highly Scalable Deep Learning Training System with Mixed-Precision: Training ImageNet in Four Minutes. arXiv:1807.11205, 07 2018. ",
+        "bbox": [
+            174,
+            776,
+            825,
+            833
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "Peter H. Jin, Qiaochu Yuan, Forrest N. Iandola, and Kurt Keutzer. How to scale distributed deep learning? arXiv:1611.04581, 11 2016. ",
+        "bbox": [
+            166,
+            843,
+            823,
+            872
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "Nitish Shirish Keskar, Dheevatsa Mudigere, Jorge Nocedal, Mikhail Smelyanskiy, and Ping Tak Peter Tang. On large-batch training for deep learning: Generalization gap and sharp minima. arXiv:1609.04836, 2016. ",
+        "bbox": [
+            174,
+            881,
+            823,
+            924
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "Alex Krizhevsky. One weird trick for parallelizing convolutional neural networks. arXiv:1404.5997, 04 2014. ",
+        "bbox": [
+            171,
+            103,
+            823,
+            132
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "Alex Krizhevsky and Geoffrey Hinton. Learning multiple layers of features from tiny images. Technical report, Citeseer, 2009. ",
+        "bbox": [
+            173,
+            141,
+            823,
+            170
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "Alex Krizhevsky, Ilya Sutskever, and Geoffrey E. Hinton. ImageNet Classification with Deep Convolutional Neural Networks. In Proceedings of the 25th International Conference on Neural Information Processing Systems - Volume 1, NIPS’12, pp. 1097–1105. Curran Associates Inc., 2012. ",
+        "bbox": [
+            174,
+            178,
+            825,
+            234
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "Siyuan Ma, Raef Bassily, and Mikhail Belkin. The Power of Interpolation: Understanding the Effectiveness of SGD in Modern Over-parametrized Learning. arXiv:1712.06559, 2017. ",
+        "bbox": [
+            176,
+            243,
+            821,
+            273
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "Julian McAuley, Christopher Targett, Qinfeng Shi, and Anton van den Hengel. Image-Based Recommendations on Styles and Substitutes. In Proceedings of the 38th International ACM SIGIR Conference on Research and Development in Information Retrieval, SIGIR ’15, pp. 43–52. ACM, 2015. ",
+        "bbox": [
+            173,
+            281,
+            825,
+            338
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "Stephen Merity, Caiming Xiong, James Bradbury, and Richard Socher. Pointer Sentinel Mixture Models. arXiv:1609.07843, 09 2016. ",
+        "bbox": [
+            174,
+            347,
+            823,
+            376
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "Raul Puri, Robert Kirby, Nikolai Yakovenko, and Bryan Catanzaro. Large Scale Language Modeling: Converging on 40GB of Text in Four Hours. arXiv:1808.01371, 08 2018. ",
+        "bbox": [
+            173,
+            385,
+            821,
+            414
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "Mark Sandler, Andrew G. Howard, Menglong Zhu, Andrey Zhmoginov, and Liang-Chieh Chen. Very Deep Convolutional Networks for Large-Scale Image Recognition. arXiv:1801.04381, 09 2014. ",
+        "bbox": [
+            173,
+            422,
+            826,
+            464
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "Timothy Sauer. Numerical solution of stochastic differential equations in finance. In Handbook of computational finance, pp. 529–550. Springer, 2012. ",
+        "bbox": [
+            174,
+            474,
+            823,
+            503
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "Karen Simonyan and Andrew Zisserman. Very Deep Convolutional Networks for Large-Scale Image Recognition. arXiv:1409.1556, 09 2014. ",
+        "bbox": [
+            174,
+            511,
+            823,
+            541
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "Samuel L. Smith and Quoc V. Le. A Bayesian Perspective on Generalization and Stochastic Gradient Descent. In International Conference on Learning Representations, 2018. ",
+        "bbox": [
+            174,
+            549,
+            823,
+            579
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "Samuel L Smith, Pieter-Jan Kindermans, and Quoc V Le. Don’t Decay the Learning Rate, Increase the Batch Size. arXiv:1711.00489, 2017. ",
+        "bbox": [
+            173,
+            587,
+            823,
+            616
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "Chen Xing, Devansh Arpit, Christos Tsirigotis, and Yoshua Bengio. A Walk with SGD. arXiv:1802.08770, 2018. ",
+        "bbox": [
+            174,
+            625,
+            823,
+            654
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "Zhewei Yao, Amir Gholami, Qi Lei, Kurt Keutzer, and Michael W Mahoney. Hessian-based Analysis of Large Batch Training and Robustness to Adversaries. arXiv:1802.08241, 2018. ",
+        "bbox": [
+            173,
+            662,
+            825,
+            693
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "Yang You, Igor Gitman, and Boris Ginsburg. Large Batch Training of Convolutional Networks. arXiv:1708.03888, 2017a. ",
+        "bbox": [
+            173,
+            700,
+            823,
+            729
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "Yang You, Zhao Zhang, Cho-Jui Hsieh, and James Demmel. 100-epoch ImageNet Training with AlexNet in 24 Minutes. arXiv:1709.05011, 2017b. ",
+        "bbox": [
+            171,
+            738,
+            825,
+            767
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "Fisher Yu, Vladlen Koltun, and Thomas Funkhouser. Dilated Residual Networks. In 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pp. 636–644, 07 2017. ",
+        "bbox": [
+            169,
+            776,
+            823,
+            806
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "Zhanxing Zhu, Jingfeng Wu, Bing Yu, Lei Wu, and Jinwen Ma. The Anisotropic Noise in Stochastic Gradient Descent: Its Behavior of Escaping from Minima and Regularization Effects. arXiv:1803.00195, 2018. ",
+        "bbox": [
+            174,
+            814,
+            823,
+            857
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "image",
+        "img_path": "images/e2d22d36762c86eb7114a858df3deb4cc64fe830391fe3ba4a35dd5954e98afa.jpg",
+        "image_caption": [
+            "Figure 5: Contour plots of test losses for various problem domains on a log scale. The test losses for BLR are on the left, while the losses for the LSR strategy are on the right. Lighter colors indicate lower loss values. Since we train each batch size for a fixed number of epochs, the total number of training iterations scales down linearly. For each loss value, we can observe how many iterations it takes to converge to that value given a particular batch size, by tracing the level curve for the associated color. For all problems, there is a batch size after which the number of training iterations necessary to converge does not decrease. "
+        ],
+        "image_footnote": [],
+        "bbox": [
+            187,
+            146,
+            810,
+            705
+        ],
+        "page_idx": 10
+    }
+]
\ No newline at end of file
diff --git a/parse/train/S1en0sRqKm/S1en0sRqKm_middle.json b/parse/train/S1en0sRqKm/S1en0sRqKm_middle.json
new file mode 100644
index 0000000000000000000000000000000000000000..ccf9fbf87c95599b5d4e94cbf1b15f51a620a6ab
--- /dev/null
+++ b/parse/train/S1en0sRqKm/S1en0sRqKm_middle.json
@@ -0,0 +1,24631 @@
+{
+    "pdf_info": [
+        {
+            "preproc_blocks": [
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        79,
+                        504,
+                        117
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                77,
+                                505,
+                                98
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        77,
+                                        505,
+                                        98
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ON THE COMPUTATIONAL INEFFICIENCY OF LARGE",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                99,
+                                500,
+                                117
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        99,
+                                        500,
+                                        117
+                                    ],
+                                    "score": 1.0,
+                                    "content": "BATCH SIZES FOR STOCHASTIC GRADIENT DESCENT",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        }
+                    ],
+                    "index": 0.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        113,
+                        135,
+                        245,
+                        157
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                113,
+                                136,
+                                201,
+                                147
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        113,
+                                        136,
+                                        201,
+                                        147
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Anonymous authors",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                111,
+                                146,
+                                245,
+                                158
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        111,
+                                        146,
+                                        245,
+                                        158
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Paper under double-blind review",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        }
+                    ],
+                    "index": 2.5
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        278,
+                        186,
+                        333,
+                        199
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                276,
+                                186,
+                                335,
+                                200
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        276,
+                                        186,
+                                        335,
+                                        200
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ABSTRACT",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        }
+                    ],
+                    "index": 4
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        143,
+                        211,
+                        468,
+                        398
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                141,
+                                212,
+                                469,
+                                225
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        212,
+                                        469,
+                                        225
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Increasing the mini-batch size for stochastic gradient descent offers significant op-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                223,
+                                470,
+                                235
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        223,
+                                        470,
+                                        235
+                                    ],
+                                    "score": 1.0,
+                                    "content": "portunities to reduce wall-clock training time, but there are a variety of theoretical",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                142,
+                                234,
+                                469,
+                                246
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        142,
+                                        234,
+                                        469,
+                                        246
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and systems challenges that impede the widespread success of this technique (Das",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                245,
+                                469,
+                                257
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        245,
+                                        469,
+                                        257
+                                    ],
+                                    "score": 1.0,
+                                    "content": "et al., 2016; Keskar et al., 2016). We investigate these issues, with an emphasis on",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                142,
+                                257,
+                                469,
+                                267
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        142,
+                                        257,
+                                        469,
+                                        267
+                                    ],
+                                    "score": 1.0,
+                                    "content": "time to convergence and total computational cost, through an extensive empirical",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                142,
+                                267,
+                                469,
+                                280
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        142,
+                                        267,
+                                        469,
+                                        280
+                                    ],
+                                    "score": 1.0,
+                                    "content": "analysis of network training across several architectures and problem domains,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                277,
+                                470,
+                                291
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        277,
+                                        470,
+                                        291
+                                    ],
+                                    "score": 1.0,
+                                    "content": "including image classification, image segmentation, and language modeling. Al-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                288,
+                                470,
+                                302
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        288,
+                                        470,
+                                        302
+                                    ],
+                                    "score": 1.0,
+                                    "content": "though it is common practice to increase the batch size in order to fully exploit",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                142,
+                                300,
+                                469,
+                                312
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        142,
+                                        300,
+                                        469,
+                                        312
+                                    ],
+                                    "score": 1.0,
+                                    "content": "available computational resources, we find a substantially more nuanced picture.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                142,
+                                311,
+                                469,
+                                323
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        142,
+                                        311,
+                                        469,
+                                        323
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Our main finding is that across a wide range of network architectures and prob-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                321,
+                                470,
+                                334
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        321,
+                                        470,
+                                        334
+                                    ],
+                                    "score": 1.0,
+                                    "content": "lem domains, increasing the batch size beyond a certain point yields no decrease",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                332,
+                                470,
+                                345
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        332,
+                                        470,
+                                        345
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in wall-clock time to convergence for either train or test loss. This batch size is",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                343,
+                                470,
+                                357
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        343,
+                                        470,
+                                        357
+                                    ],
+                                    "score": 1.0,
+                                    "content": "usually substantially below the capacity of current systems. We show that pop-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                354,
+                                470,
+                                367
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        354,
+                                        470,
+                                        367
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ular training strategies for large batch size optimization begin to fail before we",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                366,
+                                469,
+                                378
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        366,
+                                        469,
+                                        378
+                                    ],
+                                    "score": 1.0,
+                                    "content": "can populate all available compute resources, and we show that the point at which",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                376,
+                                469,
+                                388
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        376,
+                                        469,
+                                        388
+                                    ],
+                                    "score": 1.0,
+                                    "content": "these methods break down depends more on attributes like model architecture and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                387,
+                                393,
+                                399
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        387,
+                                        393,
+                                        399
+                                    ],
+                                    "score": 1.0,
+                                    "content": "data complexity than it does directly on the size of the dataset.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        }
+                    ],
+                    "index": 13
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        420,
+                        205,
+                        433
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                419,
+                                208,
+                                436
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        419,
+                                        208,
+                                        436
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1 INTRODUCTION",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        }
+                    ],
+                    "index": 22
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        446,
+                        505,
+                        599
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                445,
+                                505,
+                                459
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        445,
+                                        505,
+                                        459
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Mini-batch stochastic gradient descent (SGD) is the dominant optimization method for training deep",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                456,
+                                505,
+                                469
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        456,
+                                        505,
+                                        469
+                                    ],
+                                    "score": 1.0,
+                                    "content": "neural networks (DNNs) (Bengio & LeCun, 2007; Bottou, 2010). In the face of unprecedented",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                467,
+                                506,
+                                480
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        467,
+                                        506,
+                                        480
+                                    ],
+                                    "score": 1.0,
+                                    "content": "growth in dataset size, a large body of work has attempted to scale SGD to train DNN models on",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                479,
+                                505,
+                                491
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        479,
+                                        505,
+                                        491
+                                    ],
+                                    "score": 1.0,
+                                    "content": "increasingly large datasets, while keeping wall-clock time manageable (Iandola et al., 2015; Goyal",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                488,
+                                506,
+                                502
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        488,
+                                        506,
+                                        502
+                                    ],
+                                    "score": 1.0,
+                                    "content": "et al., 2017; Smith & Le, 2018; Devarakonda et al., 2017). The most common approach to train",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                500,
+                                505,
+                                513
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        500,
+                                        505,
+                                        513
+                                    ],
+                                    "score": 1.0,
+                                    "content": "large models at scale is distributed synchronous mini-batch SGD, which exploits additional com-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                511,
+                                505,
+                                524
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        511,
+                                        505,
+                                        524
+                                    ],
+                                    "score": 1.0,
+                                    "content": "putational resources through data parallelism. This technique reduces wall-clock training time by",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                523,
+                                505,
+                                534
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        523,
+                                        505,
+                                        534
+                                    ],
+                                    "score": 1.0,
+                                    "content": "increasing the mini-batch size, i.e., the number of examples used to compute a stochastic estimate",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                533,
+                                505,
+                                546
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        533,
+                                        505,
+                                        546
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of the gradient of the loss function at each training iteration, while holding the number of epochs",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                544,
+                                504,
+                                557
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        544,
+                                        504,
+                                        557
+                                    ],
+                                    "score": 1.0,
+                                    "content": "constant. Proponents of large batch size training often argue that the merits stem from its ability to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                556,
+                                505,
+                                567
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        556,
+                                        505,
+                                        567
+                                    ],
+                                    "score": 1.0,
+                                    "content": "decrease wall-clock training time while maintaining final model performance. Indeed, an enormous",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                565,
+                                506,
+                                579
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        565,
+                                        506,
+                                        579
+                                    ],
+                                    "score": 1.0,
+                                    "content": "amount of work has gone into designing systems that seem to operate under an assumption that",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                577,
+                                505,
+                                590
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        577,
+                                        505,
+                                        590
+                                    ],
+                                    "score": 1.0,
+                                    "content": "equates large batch size training with machine learning at scale (Goyal et al., 2017; Jia et al., 2018;",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                588,
+                                178,
+                                600
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        588,
+                                        178,
+                                        600
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Puri et al., 2018).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        }
+                    ],
+                    "index": 29.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        605,
+                        504,
+                        671
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                604,
+                                505,
+                                617
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        604,
+                                        505,
+                                        617
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Increasing the batch size improves the scaling performance of SGD per epoch, and there are signifi-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                615,
+                                505,
+                                628
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        615,
+                                        505,
+                                        628
+                                    ],
+                                    "score": 1.0,
+                                    "content": "cant challenges in building efficient distributed systems that are able to exploit additional computa-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                627,
+                                505,
+                                639
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        627,
+                                        505,
+                                        639
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tional resources to use large batch sizes (Jia et al., 2018). However, even if we were able to address",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                638,
+                                505,
+                                651
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        638,
+                                        505,
+                                        651
+                                    ],
+                                    "score": 1.0,
+                                    "content": "these systems challenges, there are still more fundamental limitations to this approach. Large batch",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                648,
+                                505,
+                                662
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        648,
+                                        505,
+                                        662
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sizes often negatively impact important performance metrics of interest, including total computa-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                659,
+                                411,
+                                673
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        659,
+                                        411,
+                                        673
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tional cost (which usually determines monetary cost) and prediction quality.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        }
+                    ],
+                    "index": 39.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        677,
+                        504,
+                        731
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                676,
+                                505,
+                                689
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        676,
+                                        505,
+                                        689
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In this paper, we will measure the total computational cost as the number of training iterations times",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                686,
+                                505,
+                                701
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        686,
+                                        505,
+                                        701
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the work done per iteration—in order to simplify measurements, we use the number of training",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                698,
+                                504,
+                                711
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        698,
+                                        504,
+                                        711
+                                    ],
+                                    "score": 1.0,
+                                    "content": "iterations as a proxy for the wall-clock time. We do this because the implementation of parallel",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                709,
+                                505,
+                                722
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        709,
+                                        505,
+                                        722
+                                    ],
+                                    "score": 1.0,
+                                    "content": "algorithms depends on software and hardware choices, and our goal is to draw more general conclu-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                721,
+                                319,
+                                732
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        721,
+                                        319,
+                                        732
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sions about the performance of SGD-based methods.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 47
+                        }
+                    ],
+                    "index": 45
+                }
+            ],
+            "page_idx": 0,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        107,
+                        27,
+                        308,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                107,
+                                26,
+                                308,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        26,
+                                        308,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Under review as a conference paper at ICLR 2019",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        303,
+                        751,
+                        308,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                302,
+                                751,
+                                308,
+                                761
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        302,
+                                        751,
+                                        308,
+                                        761
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        79,
+                        504,
+                        117
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                77,
+                                505,
+                                98
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        77,
+                                        505,
+                                        98
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ON THE COMPUTATIONAL INEFFICIENCY OF LARGE",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                99,
+                                500,
+                                117
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        99,
+                                        500,
+                                        117
+                                    ],
+                                    "score": 1.0,
+                                    "content": "BATCH SIZES FOR STOCHASTIC GRADIENT DESCENT",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        }
+                    ],
+                    "index": 0.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        113,
+                        135,
+                        245,
+                        157
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                113,
+                                136,
+                                201,
+                                147
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        113,
+                                        136,
+                                        201,
+                                        147
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Anonymous authors",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                111,
+                                146,
+                                245,
+                                158
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        111,
+                                        146,
+                                        245,
+                                        158
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Paper under double-blind review",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        }
+                    ],
+                    "index": 2.5,
+                    "bbox_fs": [
+                        111,
+                        136,
+                        245,
+                        158
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        278,
+                        186,
+                        333,
+                        199
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                276,
+                                186,
+                                335,
+                                200
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        276,
+                                        186,
+                                        335,
+                                        200
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ABSTRACT",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        }
+                    ],
+                    "index": 4
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        143,
+                        211,
+                        468,
+                        398
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                141,
+                                212,
+                                469,
+                                225
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        212,
+                                        469,
+                                        225
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Increasing the mini-batch size for stochastic gradient descent offers significant op-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                223,
+                                470,
+                                235
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        223,
+                                        470,
+                                        235
+                                    ],
+                                    "score": 1.0,
+                                    "content": "portunities to reduce wall-clock training time, but there are a variety of theoretical",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                142,
+                                234,
+                                469,
+                                246
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        142,
+                                        234,
+                                        469,
+                                        246
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and systems challenges that impede the widespread success of this technique (Das",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                245,
+                                469,
+                                257
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        245,
+                                        469,
+                                        257
+                                    ],
+                                    "score": 1.0,
+                                    "content": "et al., 2016; Keskar et al., 2016). We investigate these issues, with an emphasis on",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                142,
+                                257,
+                                469,
+                                267
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        142,
+                                        257,
+                                        469,
+                                        267
+                                    ],
+                                    "score": 1.0,
+                                    "content": "time to convergence and total computational cost, through an extensive empirical",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                142,
+                                267,
+                                469,
+                                280
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        142,
+                                        267,
+                                        469,
+                                        280
+                                    ],
+                                    "score": 1.0,
+                                    "content": "analysis of network training across several architectures and problem domains,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                277,
+                                470,
+                                291
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        277,
+                                        470,
+                                        291
+                                    ],
+                                    "score": 1.0,
+                                    "content": "including image classification, image segmentation, and language modeling. Al-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                288,
+                                470,
+                                302
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        288,
+                                        470,
+                                        302
+                                    ],
+                                    "score": 1.0,
+                                    "content": "though it is common practice to increase the batch size in order to fully exploit",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                142,
+                                300,
+                                469,
+                                312
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        142,
+                                        300,
+                                        469,
+                                        312
+                                    ],
+                                    "score": 1.0,
+                                    "content": "available computational resources, we find a substantially more nuanced picture.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                142,
+                                311,
+                                469,
+                                323
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        142,
+                                        311,
+                                        469,
+                                        323
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Our main finding is that across a wide range of network architectures and prob-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                321,
+                                470,
+                                334
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        321,
+                                        470,
+                                        334
+                                    ],
+                                    "score": 1.0,
+                                    "content": "lem domains, increasing the batch size beyond a certain point yields no decrease",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                332,
+                                470,
+                                345
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        332,
+                                        470,
+                                        345
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in wall-clock time to convergence for either train or test loss. This batch size is",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                343,
+                                470,
+                                357
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        343,
+                                        470,
+                                        357
+                                    ],
+                                    "score": 1.0,
+                                    "content": "usually substantially below the capacity of current systems. We show that pop-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                354,
+                                470,
+                                367
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        354,
+                                        470,
+                                        367
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ular training strategies for large batch size optimization begin to fail before we",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                366,
+                                469,
+                                378
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        366,
+                                        469,
+                                        378
+                                    ],
+                                    "score": 1.0,
+                                    "content": "can populate all available compute resources, and we show that the point at which",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                376,
+                                469,
+                                388
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        376,
+                                        469,
+                                        388
+                                    ],
+                                    "score": 1.0,
+                                    "content": "these methods break down depends more on attributes like model architecture and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                387,
+                                393,
+                                399
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        387,
+                                        393,
+                                        399
+                                    ],
+                                    "score": 1.0,
+                                    "content": "data complexity than it does directly on the size of the dataset.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        }
+                    ],
+                    "index": 13,
+                    "bbox_fs": [
+                        141,
+                        212,
+                        470,
+                        399
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        420,
+                        205,
+                        433
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                419,
+                                208,
+                                436
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        419,
+                                        208,
+                                        436
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1 INTRODUCTION",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        }
+                    ],
+                    "index": 22
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        446,
+                        505,
+                        599
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                445,
+                                505,
+                                459
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        445,
+                                        505,
+                                        459
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Mini-batch stochastic gradient descent (SGD) is the dominant optimization method for training deep",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                456,
+                                505,
+                                469
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        456,
+                                        505,
+                                        469
+                                    ],
+                                    "score": 1.0,
+                                    "content": "neural networks (DNNs) (Bengio & LeCun, 2007; Bottou, 2010). In the face of unprecedented",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                467,
+                                506,
+                                480
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        467,
+                                        506,
+                                        480
+                                    ],
+                                    "score": 1.0,
+                                    "content": "growth in dataset size, a large body of work has attempted to scale SGD to train DNN models on",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                479,
+                                505,
+                                491
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        479,
+                                        505,
+                                        491
+                                    ],
+                                    "score": 1.0,
+                                    "content": "increasingly large datasets, while keeping wall-clock time manageable (Iandola et al., 2015; Goyal",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                488,
+                                506,
+                                502
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        488,
+                                        506,
+                                        502
+                                    ],
+                                    "score": 1.0,
+                                    "content": "et al., 2017; Smith & Le, 2018; Devarakonda et al., 2017). The most common approach to train",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                500,
+                                505,
+                                513
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        500,
+                                        505,
+                                        513
+                                    ],
+                                    "score": 1.0,
+                                    "content": "large models at scale is distributed synchronous mini-batch SGD, which exploits additional com-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                511,
+                                505,
+                                524
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        511,
+                                        505,
+                                        524
+                                    ],
+                                    "score": 1.0,
+                                    "content": "putational resources through data parallelism. This technique reduces wall-clock training time by",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                523,
+                                505,
+                                534
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        523,
+                                        505,
+                                        534
+                                    ],
+                                    "score": 1.0,
+                                    "content": "increasing the mini-batch size, i.e., the number of examples used to compute a stochastic estimate",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                533,
+                                505,
+                                546
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        533,
+                                        505,
+                                        546
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of the gradient of the loss function at each training iteration, while holding the number of epochs",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                544,
+                                504,
+                                557
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        544,
+                                        504,
+                                        557
+                                    ],
+                                    "score": 1.0,
+                                    "content": "constant. Proponents of large batch size training often argue that the merits stem from its ability to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                556,
+                                505,
+                                567
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        556,
+                                        505,
+                                        567
+                                    ],
+                                    "score": 1.0,
+                                    "content": "decrease wall-clock training time while maintaining final model performance. Indeed, an enormous",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                565,
+                                506,
+                                579
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        565,
+                                        506,
+                                        579
+                                    ],
+                                    "score": 1.0,
+                                    "content": "amount of work has gone into designing systems that seem to operate under an assumption that",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                577,
+                                505,
+                                590
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        577,
+                                        505,
+                                        590
+                                    ],
+                                    "score": 1.0,
+                                    "content": "equates large batch size training with machine learning at scale (Goyal et al., 2017; Jia et al., 2018;",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                588,
+                                178,
+                                600
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        588,
+                                        178,
+                                        600
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Puri et al., 2018).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        }
+                    ],
+                    "index": 29.5,
+                    "bbox_fs": [
+                        104,
+                        445,
+                        506,
+                        600
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        605,
+                        504,
+                        671
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                604,
+                                505,
+                                617
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        604,
+                                        505,
+                                        617
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Increasing the batch size improves the scaling performance of SGD per epoch, and there are signifi-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                615,
+                                505,
+                                628
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        615,
+                                        505,
+                                        628
+                                    ],
+                                    "score": 1.0,
+                                    "content": "cant challenges in building efficient distributed systems that are able to exploit additional computa-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                627,
+                                505,
+                                639
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        627,
+                                        505,
+                                        639
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tional resources to use large batch sizes (Jia et al., 2018). However, even if we were able to address",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                638,
+                                505,
+                                651
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        638,
+                                        505,
+                                        651
+                                    ],
+                                    "score": 1.0,
+                                    "content": "these systems challenges, there are still more fundamental limitations to this approach. Large batch",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                648,
+                                505,
+                                662
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        648,
+                                        505,
+                                        662
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sizes often negatively impact important performance metrics of interest, including total computa-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                659,
+                                411,
+                                673
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        659,
+                                        411,
+                                        673
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tional cost (which usually determines monetary cost) and prediction quality.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        }
+                    ],
+                    "index": 39.5,
+                    "bbox_fs": [
+                        105,
+                        604,
+                        505,
+                        673
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        677,
+                        504,
+                        731
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                676,
+                                505,
+                                689
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        676,
+                                        505,
+                                        689
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In this paper, we will measure the total computational cost as the number of training iterations times",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                686,
+                                505,
+                                701
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        686,
+                                        505,
+                                        701
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the work done per iteration—in order to simplify measurements, we use the number of training",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                698,
+                                504,
+                                711
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        698,
+                                        504,
+                                        711
+                                    ],
+                                    "score": 1.0,
+                                    "content": "iterations as a proxy for the wall-clock time. We do this because the implementation of parallel",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                709,
+                                505,
+                                722
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        709,
+                                        505,
+                                        722
+                                    ],
+                                    "score": 1.0,
+                                    "content": "algorithms depends on software and hardware choices, and our goal is to draw more general conclu-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                721,
+                                319,
+                                732
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        721,
+                                        319,
+                                        732
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sions about the performance of SGD-based methods.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 47
+                        }
+                    ],
+                    "index": 45,
+                    "bbox_fs": [
+                        105,
+                        676,
+                        505,
+                        732
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        82,
+                        504,
+                        126
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                83,
+                                505,
+                                95
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        83,
+                                        505,
+                                        95
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Based on this model for total computational cost and wall-clock time, the following should be clear:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                92,
+                                506,
+                                107
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        92,
+                                        506,
+                                        107
+                                    ],
+                                    "score": 1.0,
+                                    "content": "unless increasing the batch size leads to a commensurate decrease in the total number of training",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                105,
+                                505,
+                                117
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        105,
+                                        505,
+                                        117
+                                    ],
+                                    "score": 1.0,
+                                    "content": "iterations needed to find a good model, large batch training will result in greater total computational",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                116,
+                                338,
+                                128
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        116,
+                                        338,
+                                        128
+                                    ],
+                                    "score": 1.0,
+                                    "content": "cost with little-to-no decrease in wall-clock training time.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        }
+                    ],
+                    "index": 1.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        132,
+                        505,
+                        286
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                132,
+                                505,
+                                144
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        132,
+                                        505,
+                                        144
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Based on our empirical results across a range of datasets and architectures, we find that as the batch",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                142,
+                                505,
+                                157
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        142,
+                                        505,
+                                        157
+                                    ],
+                                    "score": 1.0,
+                                    "content": "size becomes larger, there are three main phases of scaling behavior for convergence speed: (1)",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                154,
+                                506,
+                                167
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        154,
+                                        506,
+                                        167
+                                    ],
+                                    "score": 1.0,
+                                    "content": "there is a small regime of batch sizes in which increasing the batch size results in linear gains",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                165,
+                                505,
+                                178
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        165,
+                                        505,
+                                        178
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in convergence speed; (2) there is a larger regime of batch sizes that results in sublinear gains",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                175,
+                                506,
+                                190
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        175,
+                                        506,
+                                        190
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in convergence speed—in this regime, increasing the batch size can improve wall-clock training",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                186,
+                                506,
+                                200
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        186,
+                                        506,
+                                        200
+                                    ],
+                                    "score": 1.0,
+                                    "content": "time at the expense of greater total computational cost; (3) eventually, we reach a third regime",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                198,
+                                506,
+                                211
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        198,
+                                        506,
+                                        211
+                                    ],
+                                    "score": 1.0,
+                                    "content": "where a higher batch size results in marginal or non-existent reductions in convergence speed. In",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                209,
+                                506,
+                                222
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        209,
+                                        506,
+                                        222
+                                    ],
+                                    "score": 1.0,
+                                    "content": "our experiments, we find that this third regime begins at a batch size that is too small to let us",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                219,
+                                506,
+                                233
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        219,
+                                        506,
+                                        233
+                                    ],
+                                    "score": 1.0,
+                                    "content": "fully utilize available compute. Training past this batch size increases the total computational cost",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                231,
+                                505,
+                                244
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        231,
+                                        505,
+                                        244
+                                    ],
+                                    "score": 1.0,
+                                    "content": "without reducing wall-clock training time or prediction quality. While there has been considerable",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                242,
+                                505,
+                                254
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        242,
+                                        505,
+                                        254
+                                    ],
+                                    "score": 1.0,
+                                    "content": "excitement around heuristics that have been shown to make large batch training practical for certain",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                252,
+                                506,
+                                266
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        252,
+                                        506,
+                                        266
+                                    ],
+                                    "score": 1.0,
+                                    "content": "problems (Goyal et al., 2017; Smith & Le, 2018), we demonstrate that these techniques still suffer",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                263,
+                                505,
+                                278
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        263,
+                                        505,
+                                        278
+                                    ],
+                                    "score": 1.0,
+                                    "content": "from the same convergence trends we observe, and they often decrease stability of the training",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                276,
+                                141,
+                                287
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        276,
+                                        141,
+                                        287
+                                    ],
+                                    "score": 1.0,
+                                    "content": "process.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        }
+                    ],
+                    "index": 10.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        291,
+                        505,
+                        391
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                291,
+                                505,
+                                304
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        291,
+                                        505,
+                                        304
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Recent work has observed that the final test performance of models trained with large batch sizes",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                303,
+                                505,
+                                315
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        303,
+                                        505,
+                                        315
+                                    ],
+                                    "score": 1.0,
+                                    "content": "degrades after training for a fixed number of epochs (Yao et al., 2018; Keskar et al., 2016). This",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                313,
+                                505,
+                                326
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        313,
+                                        505,
+                                        326
+                                    ],
+                                    "score": 1.0,
+                                    "content": "phenomenon is known as the generalization gap. Previous work addressing this problem has focused",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                324,
+                                505,
+                                338
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        324,
+                                        505,
+                                        338
+                                    ],
+                                    "score": 1.0,
+                                    "content": "on training for more iterations in the large batch case (Hoffer et al., 2017) or adopting various",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                335,
+                                505,
+                                349
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        335,
+                                        505,
+                                        349
+                                    ],
+                                    "score": 1.0,
+                                    "content": "heuristics to select a learning rate for larger batch sizes (Goyal et al., 2017; Smith & Le, 2018).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                344,
+                                506,
+                                361
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        344,
+                                        506,
+                                        361
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Based on our empirical results, we find that existing techniques to mitigate the generalization gap",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                358,
+                                505,
+                                370
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        358,
+                                        505,
+                                        370
+                                    ],
+                                    "score": 1.0,
+                                    "content": "do not work on some problems, and for other problems they only work for batch sizes that do not",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                367,
+                                506,
+                                381
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        367,
+                                        506,
+                                        381
+                                    ],
+                                    "score": 1.0,
+                                    "content": "allow us to fully utilize our available compute. Perhaps more importantly, they do little to affect the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                380,
+                                441,
+                                393
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        380,
+                                        441,
+                                        393
+                                    ],
+                                    "score": 1.0,
+                                    "content": "diminishing returns in rates of convergence for training loss as batch size increases.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        }
+                    ],
+                    "index": 22
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        396,
+                        505,
+                        473
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                395,
+                                505,
+                                410
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        395,
+                                        505,
+                                        410
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Our objective is to understand the behavior of SGD and existing large batch techniques for many",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                407,
+                                505,
+                                420
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        407,
+                                        505,
+                                        420
+                                    ],
+                                    "score": 1.0,
+                                    "content": "network architectures and problem domains, e.g., image classification/segmentation and natural lan-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                418,
+                                506,
+                                431
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        418,
+                                        506,
+                                        431
+                                    ],
+                                    "score": 1.0,
+                                    "content": "guage processing (NLP). We observe markedly worse performance for these techniques in domains",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                430,
+                                505,
+                                441
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        430,
+                                        505,
+                                        441
+                                    ],
+                                    "score": 1.0,
+                                    "content": "other than image classification, where large batch optimization has received the most attention (Jia",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                440,
+                                505,
+                                453
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        440,
+                                        505,
+                                        453
+                                    ],
+                                    "score": 1.0,
+                                    "content": "et al., 2018; You et al., 2017b). Because we eschew the challenges of an efficient distributed im-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                451,
+                                505,
+                                464
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        451,
+                                        505,
+                                        464
+                                    ],
+                                    "score": 1.0,
+                                    "content": "plementation by measuring number of iterations instead of wall-clock time, our results assume the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                462,
+                                431,
+                                475
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        462,
+                                        431,
+                                        475
+                                    ],
+                                    "score": 1.0,
+                                    "content": "most optimistic circumstances for large batch training. Our key observations are:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        }
+                    ],
+                    "index": 30
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        132,
+                        486,
+                        506,
+                        648
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                136,
+                                485,
+                                506,
+                                499
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        136,
+                                        485,
+                                        506,
+                                        499
+                                    ],
+                                    "score": 1.0,
+                                    "content": "• Increasing the batch size beyond a certain point yields no improvement in wall-clock",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                497,
+                                506,
+                                510
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        497,
+                                        506,
+                                        510
+                                    ],
+                                    "score": 1.0,
+                                    "content": "time to convergence, even for a system with perfect parallelism. We observe that larger",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                507,
+                                505,
+                                520
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        507,
+                                        505,
+                                        520
+                                    ],
+                                    "score": 1.0,
+                                    "content": "batch sizes result in a limited reduction in the number of training iterations needed to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                142,
+                                519,
+                                477,
+                                532
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        142,
+                                        519,
+                                        477,
+                                        532
+                                    ],
+                                    "score": 1.0,
+                                    "content": "achieve low training or test error, and that eventually these gains become near-zero.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                137,
+                                539,
+                                505,
+                                552
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        137,
+                                        539,
+                                        505,
+                                        552
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Increasing the batch size leads to a significant increase in generalization error, which",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                550,
+                                506,
+                                563
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        550,
+                                        506,
+                                        563
+                                    ],
+                                    "score": 1.0,
+                                    "content": "cannot be mitigated by existing techniques. We observe that these techniques often result",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                142,
+                                561,
+                                505,
+                                574
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        142,
+                                        561,
+                                        505,
+                                        574
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in divergent training behavior or that they only mitigate degradation in test performance for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                571,
+                                332,
+                                586
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        571,
+                                        332,
+                                        586
+                                    ],
+                                    "score": 1.0,
+                                    "content": "small batch sizes relative to available compute.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                138,
+                                590,
+                                505,
+                                607
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        138,
+                                        590,
+                                        505,
+                                        607
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Dataset size is not the only factor determining the computational efficiency of large",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                603,
+                                505,
+                                617
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        603,
+                                        505,
+                                        617
+                                    ],
+                                    "score": 1.0,
+                                    "content": "batch training. We observe that both the diminishing returns in convergence speed and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                614,
+                                505,
+                                627
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        614,
+                                        505,
+                                        627
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the failure of existing methods correlate more with factors like model architecture and data",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                142,
+                                626,
+                                505,
+                                638
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        142,
+                                        626,
+                                        505,
+                                        638
+                                    ],
+                                    "score": 1.0,
+                                    "content": "complexity than dataset size alone. As a result, training time may significantly increase",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                636,
+                                422,
+                                650
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        636,
+                                        422,
+                                        650
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with dataset size in spite of increasingly available compute resources.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        }
+                    ],
+                    "index": 40
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        666,
+                        505,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                664,
+                                505,
+                                678
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        664,
+                                        505,
+                                        678
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In Section 2, we review the formulation of SGD as well as existing strategies to train with large",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 47
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                676,
+                                505,
+                                690
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        676,
+                                        505,
+                                        690
+                                    ],
+                                    "score": 1.0,
+                                    "content": "batch sizes. In Section 3, we review recent theoretical results regarding the convergence rates of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 48
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                687,
+                                506,
+                                700
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        687,
+                                        506,
+                                        700
+                                    ],
+                                    "score": 1.0,
+                                    "content": "SGD in highly over-parameterized settings and discuss the potential impact of these results on the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 49
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                699,
+                                506,
+                                712
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        699,
+                                        506,
+                                        712
+                                    ],
+                                    "score": 1.0,
+                                    "content": "computational efficiency of SGD for deep learning. Section 4 presents our empirical results that",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 50
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                710,
+                                505,
+                                721
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        710,
+                                        505,
+                                        721
+                                    ],
+                                    "score": 1.0,
+                                    "content": "demonstrate the inefficiencies of training SGD with large batch sizes, and we show that these persist",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 51
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                720,
+                                336,
+                                733
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        720,
+                                        336,
+                                        733
+                                    ],
+                                    "score": 1.0,
+                                    "content": "when using existing large batch optimization techniques.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 52
+                        }
+                    ],
+                    "index": 49.5
+                }
+            ],
+            "page_idx": 1,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        107,
+                        27,
+                        308,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                107,
+                                26,
+                                308,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        26,
+                                        308,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Under review as a conference paper at ICLR 2019",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        302,
+                        751,
+                        309,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                301,
+                                750,
+                                310,
+                                763
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        301,
+                                        750,
+                                        310,
+                                        763
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        82,
+                        504,
+                        126
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                83,
+                                505,
+                                95
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        83,
+                                        505,
+                                        95
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Based on this model for total computational cost and wall-clock time, the following should be clear:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                92,
+                                506,
+                                107
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        92,
+                                        506,
+                                        107
+                                    ],
+                                    "score": 1.0,
+                                    "content": "unless increasing the batch size leads to a commensurate decrease in the total number of training",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                105,
+                                505,
+                                117
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        105,
+                                        505,
+                                        117
+                                    ],
+                                    "score": 1.0,
+                                    "content": "iterations needed to find a good model, large batch training will result in greater total computational",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                116,
+                                338,
+                                128
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        116,
+                                        338,
+                                        128
+                                    ],
+                                    "score": 1.0,
+                                    "content": "cost with little-to-no decrease in wall-clock training time.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        }
+                    ],
+                    "index": 1.5,
+                    "bbox_fs": [
+                        105,
+                        83,
+                        506,
+                        128
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        132,
+                        505,
+                        286
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                132,
+                                505,
+                                144
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        132,
+                                        505,
+                                        144
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Based on our empirical results across a range of datasets and architectures, we find that as the batch",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                142,
+                                505,
+                                157
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        142,
+                                        505,
+                                        157
+                                    ],
+                                    "score": 1.0,
+                                    "content": "size becomes larger, there are three main phases of scaling behavior for convergence speed: (1)",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                154,
+                                506,
+                                167
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        154,
+                                        506,
+                                        167
+                                    ],
+                                    "score": 1.0,
+                                    "content": "there is a small regime of batch sizes in which increasing the batch size results in linear gains",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                165,
+                                505,
+                                178
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        165,
+                                        505,
+                                        178
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in convergence speed; (2) there is a larger regime of batch sizes that results in sublinear gains",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                175,
+                                506,
+                                190
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        175,
+                                        506,
+                                        190
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in convergence speed—in this regime, increasing the batch size can improve wall-clock training",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                186,
+                                506,
+                                200
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        186,
+                                        506,
+                                        200
+                                    ],
+                                    "score": 1.0,
+                                    "content": "time at the expense of greater total computational cost; (3) eventually, we reach a third regime",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                198,
+                                506,
+                                211
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        198,
+                                        506,
+                                        211
+                                    ],
+                                    "score": 1.0,
+                                    "content": "where a higher batch size results in marginal or non-existent reductions in convergence speed. In",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                209,
+                                506,
+                                222
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        209,
+                                        506,
+                                        222
+                                    ],
+                                    "score": 1.0,
+                                    "content": "our experiments, we find that this third regime begins at a batch size that is too small to let us",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                219,
+                                506,
+                                233
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        219,
+                                        506,
+                                        233
+                                    ],
+                                    "score": 1.0,
+                                    "content": "fully utilize available compute. Training past this batch size increases the total computational cost",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                231,
+                                505,
+                                244
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        231,
+                                        505,
+                                        244
+                                    ],
+                                    "score": 1.0,
+                                    "content": "without reducing wall-clock training time or prediction quality. While there has been considerable",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                242,
+                                505,
+                                254
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        242,
+                                        505,
+                                        254
+                                    ],
+                                    "score": 1.0,
+                                    "content": "excitement around heuristics that have been shown to make large batch training practical for certain",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                252,
+                                506,
+                                266
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        252,
+                                        506,
+                                        266
+                                    ],
+                                    "score": 1.0,
+                                    "content": "problems (Goyal et al., 2017; Smith & Le, 2018), we demonstrate that these techniques still suffer",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                263,
+                                505,
+                                278
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        263,
+                                        505,
+                                        278
+                                    ],
+                                    "score": 1.0,
+                                    "content": "from the same convergence trends we observe, and they often decrease stability of the training",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                276,
+                                141,
+                                287
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        276,
+                                        141,
+                                        287
+                                    ],
+                                    "score": 1.0,
+                                    "content": "process.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        }
+                    ],
+                    "index": 10.5,
+                    "bbox_fs": [
+                        104,
+                        132,
+                        506,
+                        287
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        291,
+                        505,
+                        391
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                291,
+                                505,
+                                304
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        291,
+                                        505,
+                                        304
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Recent work has observed that the final test performance of models trained with large batch sizes",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                303,
+                                505,
+                                315
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        303,
+                                        505,
+                                        315
+                                    ],
+                                    "score": 1.0,
+                                    "content": "degrades after training for a fixed number of epochs (Yao et al., 2018; Keskar et al., 2016). This",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                313,
+                                505,
+                                326
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        313,
+                                        505,
+                                        326
+                                    ],
+                                    "score": 1.0,
+                                    "content": "phenomenon is known as the generalization gap. Previous work addressing this problem has focused",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                324,
+                                505,
+                                338
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        324,
+                                        505,
+                                        338
+                                    ],
+                                    "score": 1.0,
+                                    "content": "on training for more iterations in the large batch case (Hoffer et al., 2017) or adopting various",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                335,
+                                505,
+                                349
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        335,
+                                        505,
+                                        349
+                                    ],
+                                    "score": 1.0,
+                                    "content": "heuristics to select a learning rate for larger batch sizes (Goyal et al., 2017; Smith & Le, 2018).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                344,
+                                506,
+                                361
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        344,
+                                        506,
+                                        361
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Based on our empirical results, we find that existing techniques to mitigate the generalization gap",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                358,
+                                505,
+                                370
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        358,
+                                        505,
+                                        370
+                                    ],
+                                    "score": 1.0,
+                                    "content": "do not work on some problems, and for other problems they only work for batch sizes that do not",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                367,
+                                506,
+                                381
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        367,
+                                        506,
+                                        381
+                                    ],
+                                    "score": 1.0,
+                                    "content": "allow us to fully utilize our available compute. Perhaps more importantly, they do little to affect the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                380,
+                                441,
+                                393
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        380,
+                                        441,
+                                        393
+                                    ],
+                                    "score": 1.0,
+                                    "content": "diminishing returns in rates of convergence for training loss as batch size increases.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        }
+                    ],
+                    "index": 22,
+                    "bbox_fs": [
+                        104,
+                        291,
+                        506,
+                        393
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        396,
+                        505,
+                        473
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                395,
+                                505,
+                                410
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        395,
+                                        505,
+                                        410
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Our objective is to understand the behavior of SGD and existing large batch techniques for many",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                407,
+                                505,
+                                420
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        407,
+                                        505,
+                                        420
+                                    ],
+                                    "score": 1.0,
+                                    "content": "network architectures and problem domains, e.g., image classification/segmentation and natural lan-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                418,
+                                506,
+                                431
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        418,
+                                        506,
+                                        431
+                                    ],
+                                    "score": 1.0,
+                                    "content": "guage processing (NLP). We observe markedly worse performance for these techniques in domains",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                430,
+                                505,
+                                441
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        430,
+                                        505,
+                                        441
+                                    ],
+                                    "score": 1.0,
+                                    "content": "other than image classification, where large batch optimization has received the most attention (Jia",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                440,
+                                505,
+                                453
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        440,
+                                        505,
+                                        453
+                                    ],
+                                    "score": 1.0,
+                                    "content": "et al., 2018; You et al., 2017b). Because we eschew the challenges of an efficient distributed im-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                451,
+                                505,
+                                464
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        451,
+                                        505,
+                                        464
+                                    ],
+                                    "score": 1.0,
+                                    "content": "plementation by measuring number of iterations instead of wall-clock time, our results assume the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                462,
+                                431,
+                                475
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        462,
+                                        431,
+                                        475
+                                    ],
+                                    "score": 1.0,
+                                    "content": "most optimistic circumstances for large batch training. Our key observations are:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        }
+                    ],
+                    "index": 30,
+                    "bbox_fs": [
+                        105,
+                        395,
+                        506,
+                        475
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        132,
+                        486,
+                        506,
+                        648
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                136,
+                                485,
+                                506,
+                                499
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        136,
+                                        485,
+                                        506,
+                                        499
+                                    ],
+                                    "score": 1.0,
+                                    "content": "• Increasing the batch size beyond a certain point yields no improvement in wall-clock",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                497,
+                                506,
+                                510
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        497,
+                                        506,
+                                        510
+                                    ],
+                                    "score": 1.0,
+                                    "content": "time to convergence, even for a system with perfect parallelism. We observe that larger",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                507,
+                                505,
+                                520
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        507,
+                                        505,
+                                        520
+                                    ],
+                                    "score": 1.0,
+                                    "content": "batch sizes result in a limited reduction in the number of training iterations needed to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                142,
+                                519,
+                                477,
+                                532
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        142,
+                                        519,
+                                        477,
+                                        532
+                                    ],
+                                    "score": 1.0,
+                                    "content": "achieve low training or test error, and that eventually these gains become near-zero.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                137,
+                                539,
+                                505,
+                                552
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        137,
+                                        539,
+                                        505,
+                                        552
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Increasing the batch size leads to a significant increase in generalization error, which",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                550,
+                                506,
+                                563
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        550,
+                                        506,
+                                        563
+                                    ],
+                                    "score": 1.0,
+                                    "content": "cannot be mitigated by existing techniques. We observe that these techniques often result",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                142,
+                                561,
+                                505,
+                                574
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        142,
+                                        561,
+                                        505,
+                                        574
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in divergent training behavior or that they only mitigate degradation in test performance for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                571,
+                                332,
+                                586
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        571,
+                                        332,
+                                        586
+                                    ],
+                                    "score": 1.0,
+                                    "content": "small batch sizes relative to available compute.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                138,
+                                590,
+                                505,
+                                607
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        138,
+                                        590,
+                                        505,
+                                        607
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Dataset size is not the only factor determining the computational efficiency of large",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                603,
+                                505,
+                                617
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        603,
+                                        505,
+                                        617
+                                    ],
+                                    "score": 1.0,
+                                    "content": "batch training. We observe that both the diminishing returns in convergence speed and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                614,
+                                505,
+                                627
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        614,
+                                        505,
+                                        627
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the failure of existing methods correlate more with factors like model architecture and data",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                142,
+                                626,
+                                505,
+                                638
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        142,
+                                        626,
+                                        505,
+                                        638
+                                    ],
+                                    "score": 1.0,
+                                    "content": "complexity than dataset size alone. As a result, training time may significantly increase",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                636,
+                                422,
+                                650
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        636,
+                                        422,
+                                        650
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with dataset size in spite of increasingly available compute resources.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        }
+                    ],
+                    "index": 40,
+                    "bbox_fs": [
+                        136,
+                        485,
+                        506,
+                        650
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        666,
+                        505,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                664,
+                                505,
+                                678
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        664,
+                                        505,
+                                        678
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In Section 2, we review the formulation of SGD as well as existing strategies to train with large",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 47
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                676,
+                                505,
+                                690
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        676,
+                                        505,
+                                        690
+                                    ],
+                                    "score": 1.0,
+                                    "content": "batch sizes. In Section 3, we review recent theoretical results regarding the convergence rates of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 48
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                687,
+                                506,
+                                700
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        687,
+                                        506,
+                                        700
+                                    ],
+                                    "score": 1.0,
+                                    "content": "SGD in highly over-parameterized settings and discuss the potential impact of these results on the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 49
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                699,
+                                506,
+                                712
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        699,
+                                        506,
+                                        712
+                                    ],
+                                    "score": 1.0,
+                                    "content": "computational efficiency of SGD for deep learning. Section 4 presents our empirical results that",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 50
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                710,
+                                505,
+                                721
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        710,
+                                        505,
+                                        721
+                                    ],
+                                    "score": 1.0,
+                                    "content": "demonstrate the inefficiencies of training SGD with large batch sizes, and we show that these persist",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 51
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                720,
+                                336,
+                                733
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        720,
+                                        336,
+                                        733
+                                    ],
+                                    "score": 1.0,
+                                    "content": "when using existing large batch optimization techniques.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 52
+                        }
+                    ],
+                    "index": 49.5,
+                    "bbox_fs": [
+                        105,
+                        664,
+                        506,
+                        733
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        81,
+                        314,
+                        94
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                79,
+                                315,
+                                96
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        79,
+                                        315,
+                                        96
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2 BACKGROUND AND RELATED WORK",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        }
+                    ],
+                    "index": 0
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        106,
+                        505,
+                        139
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                105,
+                                505,
+                                118
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        105,
+                                        505,
+                                        118
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Stochastic Gradient Descent. SGD is the most widely used algorithm to train DNN models. The",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                117,
+                                505,
+                                129
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        117,
+                                        247,
+                                        129
+                                    ],
+                                    "score": 1.0,
+                                    "content": "model is parameterized by weights",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        247,
+                                        117,
+                                        280,
+                                        127
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\mathbf { w } \\in \\mathbb { R } ^ { d }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        281,
+                                        117,
+                                        505,
+                                        129
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", and the objective is to minimize the empirical loss over",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                127,
+                                176,
+                                142
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        131,
+                                        114,
+                                        138
+                                    ],
+                                    "score": 0.7,
+                                    "content": "n",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        114,
+                                        127,
+                                        160,
+                                        142
+                                    ],
+                                    "score": 1.0,
+                                    "content": "data points",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        161,
+                                        130,
+                                        171,
+                                        139
+                                    ],
+                                    "score": 0.85,
+                                    "content": "\\mathbf { x } _ { i }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        172,
+                                        127,
+                                        176,
+                                        142
+                                    ],
+                                    "score": 1.0,
+                                    "content": ":",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        }
+                    ],
+                    "index": 2
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        254,
+                        137,
+                        356,
+                        171
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                254,
+                                137,
+                                356,
+                                171
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        254,
+                                        137,
+                                        356,
+                                        171
+                                    ],
+                                    "score": 0.95,
+                                    "content": "L ( \\mathbf { w } ) = \\frac { 1 } { n } \\sum _ { i = 1 } ^ { n } \\ell ( \\mathbf { w } , \\mathbf { x } _ { i } ) ,",
+                                    "type": "interline_equation",
+                                    "image_path": "887595ea6a4113fb0007840a5a7f971572079b5726c91bec59fd88f90a821b85.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 4.5,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                254,
+                                137,
+                                356,
+                                154.0
+                            ],
+                            "spans": [],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                254,
+                                154.0,
+                                356,
+                                171.0
+                            ],
+                            "spans": [],
+                            "index": 5
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        172,
+                        502,
+                        185
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                170,
+                                502,
+                                186
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        170,
+                                        133,
+                                        186
+                                    ],
+                                    "score": 1.0,
+                                    "content": "where",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        133,
+                                        172,
+                                        156,
+                                        185
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\ell ( \\cdot , \\cdot )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        156,
+                                        170,
+                                        502,
+                                        186
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is a loss, e.g., cross-entropy or squared error. This loss gives a corresponding gradient",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        }
+                    ],
+                    "index": 6
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        227,
+                        189,
+                        384,
+                        222
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                227,
+                                189,
+                                384,
+                                222
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        227,
+                                        189,
+                                        384,
+                                        222
+                                    ],
+                                    "score": 0.95,
+                                    "content": "\\mathbf { g } ( \\mathbf { w } ) : = \\nabla L ( \\mathbf { w } ) = \\frac { 1 } { n } \\sum _ { i = 1 } ^ { n } \\nabla \\ell ( \\mathbf { w } , \\mathbf { x } _ { i } ) .",
+                                    "type": "interline_equation",
+                                    "image_path": "5ce375a0c0b4eba72df983bc418c05ba2ee824ff817e740a7b4a68960b2471aa.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 7.5,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                227,
+                                189,
+                                384,
+                                205.5
+                            ],
+                            "spans": [],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                227,
+                                205.5,
+                                384,
+                                222.0
+                            ],
+                            "spans": [],
+                            "index": 8
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        227,
+                        505,
+                        261
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                225,
+                                505,
+                                240
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        225,
+                                        160,
+                                        240
+                                    ],
+                                    "score": 1.0,
+                                    "content": "A mini-batch",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        161,
+                                        228,
+                                        176,
+                                        238
+                                    ],
+                                    "score": 0.89,
+                                    "content": "B _ { m }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        176,
+                                        225,
+                                        204,
+                                        240
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of size",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        204,
+                                        228,
+                                        234,
+                                        237
+                                    ],
+                                    "score": 0.89,
+                                    "content": "m < n",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        234,
+                                        225,
+                                        301,
+                                        240
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is a collection of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        301,
+                                        229,
+                                        311,
+                                        237
+                                    ],
+                                    "score": 0.79,
+                                    "content": "m",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        311,
+                                        225,
+                                        457,
+                                        240
+                                    ],
+                                    "score": 1.0,
+                                    "content": "indices randomly drawn from the set",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        457,
+                                        227,
+                                        501,
+                                        239
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\{ 1 , \\ldots , n \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        501,
+                                        225,
+                                        505,
+                                        240
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                237,
+                                505,
+                                250
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        237,
+                                        409,
+                                        250
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and we can use it to form an unbiased estimate of the gradient at iteration",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        410,
+                                        239,
+                                        416,
+                                        248
+                                    ],
+                                    "score": 0.81,
+                                    "content": "k",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        416,
+                                        237,
+                                        505,
+                                        250
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", as well as the corre-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                249,
+                                200,
+                                262
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        249,
+                                        200,
+                                        262
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sponding SGD update:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        }
+                    ],
+                    "index": 10
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        163,
+                        265,
+                        447,
+                        296
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                163,
+                                265,
+                                447,
+                                296
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        163,
+                                        265,
+                                        447,
+                                        296
+                                    ],
+                                    "score": 0.94,
+                                    "content": "\\mathbf { g } _ { m } ( \\mathbf { w } _ { k } ) = \\frac { 1 } { m } \\sum _ { i \\in \\mathcal { B } _ { m } } \\nabla \\ell ( \\mathbf { w } _ { k } , \\mathbf { x } _ { i } ) \\quad \\mathrm { ~ a n d ~ } \\quad \\mathbf { w } _ { k + 1 } = \\mathbf { w } _ { k } - \\eta _ { k } \\mathbf { g } _ { m } ( \\mathbf { w } _ { k } ) ,",
+                                    "type": "interline_equation",
+                                    "image_path": "83e3f1f25ec3b8d3387d926f3bff0f8c4141099ab08b61b4c0c318b9c7452052.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 13,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                163,
+                                265,
+                                447,
+                                275.3333333333333
+                            ],
+                            "spans": [],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                163,
+                                275.3333333333333,
+                                447,
+                                285.66666666666663
+                            ],
+                            "spans": [],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                163,
+                                285.66666666666663,
+                                447,
+                                295.99999999999994
+                            ],
+                            "spans": [],
+                            "index": 14
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        300,
+                        505,
+                        334
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                300,
+                                506,
+                                313
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        300,
+                                        133,
+                                        313
+                                    ],
+                                    "score": 1.0,
+                                    "content": "where",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        133,
+                                        301,
+                                        163,
+                                        312
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\eta _ { k } > 0",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        163,
+                                        300,
+                                        290,
+                                        313
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is the learning rate for iteration",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        290,
+                                        301,
+                                        297,
+                                        311
+                                    ],
+                                    "score": 0.78,
+                                    "content": "k",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        297,
+                                        300,
+                                        506,
+                                        313
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". One iteration of training for SGD corresponds to a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                311,
+                                505,
+                                325
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        311,
+                                        396,
+                                        325
+                                    ],
+                                    "score": 1.0,
+                                    "content": "single gradient computation / weight update. One epoch corresponds to",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        396,
+                                        312,
+                                        417,
+                                        324
+                                    ],
+                                    "score": 0.92,
+                                    "content": "n / m",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        417,
+                                        311,
+                                        505,
+                                        325
+                                    ],
+                                    "score": 1.0,
+                                    "content": "iterations of training.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                321,
+                                505,
+                                335
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        321,
+                                        505,
+                                        335
+                                    ],
+                                    "score": 1.0,
+                                    "content": "This constitutes a single pass over the dataset, assuming the dataset is sampled without replacement.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        }
+                    ],
+                    "index": 16
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        339,
+                        503,
+                        373
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                339,
+                                505,
+                                353
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        339,
+                                        505,
+                                        353
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Efficient distributed systems reduce wall-clock training time by parallelizing gradient calculations",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                350,
+                                504,
+                                362
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        350,
+                                        504,
+                                        362
+                                    ],
+                                    "score": 1.0,
+                                    "content": "across many machines. When the batch size is large enough to populate all available compute",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                362,
+                                446,
+                                374
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        362,
+                                        446,
+                                        374
+                                    ],
+                                    "score": 1.0,
+                                    "content": "resources, this allows us to amortize the cost of coordination for each weight update.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        }
+                    ],
+                    "index": 19
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        378,
+                        505,
+                        445
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                378,
+                                506,
+                                391
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        378,
+                                        506,
+                                        391
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Existing large batch techniques. With the hope of keeping training times manageable as dataset",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                390,
+                                505,
+                                401
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        390,
+                                        505,
+                                        401
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sizes escalate, recent work has focused on the development of techniques that allow practitioners",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                400,
+                                506,
+                                413
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        400,
+                                        506,
+                                        413
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to increase the batch size to make use of growing computational resources (Jin et al., 2016; Jia",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                411,
+                                505,
+                                424
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        411,
+                                        505,
+                                        424
+                                    ],
+                                    "score": 1.0,
+                                    "content": "et al., 2018; You et al., 2017a). However, there is a growing body of theoretical and empirical",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                422,
+                                506,
+                                435
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        422,
+                                        506,
+                                        435
+                                    ],
+                                    "score": 1.0,
+                                    "content": "results suggesting that large batch sizes adversely affect the generalization performance of the final",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                433,
+                                392,
+                                445
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        433,
+                                        392,
+                                        445
+                                    ],
+                                    "score": 1.0,
+                                    "content": "model (Yao et al., 2018; Keskar et al., 2016; Devarakonda et al., 2017).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        }
+                    ],
+                    "index": 23.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        450,
+                        504,
+                        483
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                450,
+                                505,
+                                463
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        450,
+                                        505,
+                                        463
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In response to this, recent work has proposed changing two parameters in relation to batch size: the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                461,
+                                505,
+                                473
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        461,
+                                        505,
+                                        473
+                                    ],
+                                    "score": 1.0,
+                                    "content": "number of training iterations and the learning rate. However, they also make assumptions that limit",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                472,
+                                395,
+                                484
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        472,
+                                        395,
+                                        484
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the effectiveness of their proposals as useful heuristics for practitioners.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        }
+                    ],
+                    "index": 28
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        132,
+                        493,
+                        505,
+                        699
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                133,
+                                493,
+                                505,
+                                506
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        133,
+                                        493,
+                                        505,
+                                        506
+                                    ],
+                                    "score": 1.0,
+                                    "content": "• Training longer: Hoffer et al. (2017) suggest increasing the number of training iterations.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                504,
+                                506,
+                                518
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        504,
+                                        506,
+                                        518
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Even if this does reduce the generalization gap, it significantly increases both wall-clock",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                516,
+                                505,
+                                529
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        516,
+                                        505,
+                                        529
+                                    ],
+                                    "score": 1.0,
+                                    "content": "training time and computational cost. Moreover, in some problems it does not lead to min-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                142,
+                                528,
+                                502,
+                                539
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        142,
+                                        528,
+                                        502,
+                                        539
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ima with better generalization performance (as we found when running our experiments).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                132,
+                                541,
+                                505,
+                                555
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        132,
+                                        542,
+                                        367,
+                                        555
+                                    ],
+                                    "score": 1.0,
+                                    "content": "• Square root LR scaling: Scaling the learning rate as",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        368,
+                                        541,
+                                        415,
+                                        553
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\eta _ { 0 } \\propto \\sqrt { m }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        416,
+                                        542,
+                                        505,
+                                        555
+                                    ],
+                                    "score": 1.0,
+                                    "content": "attempts to keep the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                142,
+                                553,
+                                505,
+                                564
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        142,
+                                        553,
+                                        505,
+                                        564
+                                    ],
+                                    "score": 1.0,
+                                    "content": "weight increment length statistics constant, but the distance between SGD iterates is gov-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                564,
+                                505,
+                                575
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        564,
+                                        505,
+                                        575
+                                    ],
+                                    "score": 1.0,
+                                    "content": "erned more by properties of the objective function than the ratio of learning rate to batch",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                142,
+                                574,
+                                505,
+                                586
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        142,
+                                        574,
+                                        505,
+                                        586
+                                    ],
+                                    "score": 1.0,
+                                    "content": "size (Chaudhari & Soatto, 2017; Zhu et al., 2018). This rule has also been found to be",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                586,
+                                435,
+                                597
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        586,
+                                        435,
+                                        597
+                                    ],
+                                    "score": 1.0,
+                                    "content": "empirically sub-optimal in various problem domains (Krizhevsky, 2014).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                137,
+                                600,
+                                506,
+                                613
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        137,
+                                        600,
+                                        506,
+                                        613
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Linear LR scaling: The performance of large batch training can also be improved by using",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                611,
+                                506,
+                                624
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        611,
+                                        506,
+                                        624
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the linear scaling rule, which suggests choosing a learning rate proportional to the batch size",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                144,
+                                622,
+                                506,
+                                635
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        144,
+                                        623,
+                                        183,
+                                        634
+                                    ],
+                                    "score": 0.84,
+                                    "content": "( \\eta _ { 0 } \\propto m )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        184,
+                                        622,
+                                        506,
+                                        635
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(Goyal et al., 2017). There are two motivations for this rule: the first assumes",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                632,
+                                505,
+                                646
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        632,
+                                        505,
+                                        646
+                                    ],
+                                    "score": 1.0,
+                                    "content": "that one large-batch gradient step should resemble a series of small-batch gradient steps",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                644,
+                                506,
+                                657
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        644,
+                                        506,
+                                        657
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in order for convergence rates to improve linearly (Goyal et al., 2017); the other regards",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                655,
+                                505,
+                                666
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        655,
+                                        505,
+                                        666
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the SGD update equation as the Euler-Maruyama discretization of a stochastic differential",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                666,
+                                505,
+                                679
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        666,
+                                        505,
+                                        679
+                                    ],
+                                    "score": 1.0,
+                                    "content": "equation (Sauer, 2012; Xing et al., 2018), and attempts to maintain a constant level of mini-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                677,
+                                505,
+                                690
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        677,
+                                        505,
+                                        690
+                                    ],
+                                    "score": 1.0,
+                                    "content": "batch noise to help SGD explore the loss landscape (Chaudhari & Soatto, 2017; Zhu et al.,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        },
+                        {
+                            "bbox": [
+                                142,
+                                688,
+                                248,
+                                700
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        142,
+                                        688,
+                                        248,
+                                        700
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2018; Smith & Le, 2018).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 47
+                        }
+                    ],
+                    "index": 38.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        709,
+                        502,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                709,
+                                504,
+                                722
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        709,
+                                        504,
+                                        722
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Both justifications for the linear scaling rule implicitly impose strong conditions on the loss func-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 48
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                719,
+                                504,
+                                734
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        719,
+                                        504,
+                                        734
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tion by requiring that it behave linearly near SGD iterates; therefore, if the loss function is highly",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 49
+                        }
+                    ],
+                    "index": 48.5
+                }
+            ],
+            "page_idx": 2,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        106,
+                        26,
+                        308,
+                        38
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                25,
+                                308,
+                                39
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        25,
+                                        308,
+                                        39
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Under review as a conference paper at ICLR 2019",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        302,
+                        751,
+                        309,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                301,
+                                750,
+                                310,
+                                762
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        301,
+                                        750,
+                                        310,
+                                        762
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        81,
+                        314,
+                        94
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                79,
+                                315,
+                                96
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        79,
+                                        315,
+                                        96
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2 BACKGROUND AND RELATED WORK",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        }
+                    ],
+                    "index": 0
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        106,
+                        505,
+                        139
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                105,
+                                505,
+                                118
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        105,
+                                        505,
+                                        118
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Stochastic Gradient Descent. SGD is the most widely used algorithm to train DNN models. The",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                117,
+                                505,
+                                129
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        117,
+                                        247,
+                                        129
+                                    ],
+                                    "score": 1.0,
+                                    "content": "model is parameterized by weights",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        247,
+                                        117,
+                                        280,
+                                        127
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\mathbf { w } \\in \\mathbb { R } ^ { d }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        281,
+                                        117,
+                                        505,
+                                        129
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", and the objective is to minimize the empirical loss over",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                127,
+                                176,
+                                142
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        131,
+                                        114,
+                                        138
+                                    ],
+                                    "score": 0.7,
+                                    "content": "n",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        114,
+                                        127,
+                                        160,
+                                        142
+                                    ],
+                                    "score": 1.0,
+                                    "content": "data points",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        161,
+                                        130,
+                                        171,
+                                        139
+                                    ],
+                                    "score": 0.85,
+                                    "content": "\\mathbf { x } _ { i }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        172,
+                                        127,
+                                        176,
+                                        142
+                                    ],
+                                    "score": 1.0,
+                                    "content": ":",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        }
+                    ],
+                    "index": 2,
+                    "bbox_fs": [
+                        105,
+                        105,
+                        505,
+                        142
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        254,
+                        137,
+                        356,
+                        171
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                254,
+                                137,
+                                356,
+                                171
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        254,
+                                        137,
+                                        356,
+                                        171
+                                    ],
+                                    "score": 0.95,
+                                    "content": "L ( \\mathbf { w } ) = \\frac { 1 } { n } \\sum _ { i = 1 } ^ { n } \\ell ( \\mathbf { w } , \\mathbf { x } _ { i } ) ,",
+                                    "type": "interline_equation",
+                                    "image_path": "887595ea6a4113fb0007840a5a7f971572079b5726c91bec59fd88f90a821b85.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 4.5,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                254,
+                                137,
+                                356,
+                                154.0
+                            ],
+                            "spans": [],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                254,
+                                154.0,
+                                356,
+                                171.0
+                            ],
+                            "spans": [],
+                            "index": 5
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        172,
+                        502,
+                        185
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                170,
+                                502,
+                                186
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        170,
+                                        133,
+                                        186
+                                    ],
+                                    "score": 1.0,
+                                    "content": "where",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        133,
+                                        172,
+                                        156,
+                                        185
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\ell ( \\cdot , \\cdot )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        156,
+                                        170,
+                                        502,
+                                        186
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is a loss, e.g., cross-entropy or squared error. This loss gives a corresponding gradient",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        }
+                    ],
+                    "index": 6,
+                    "bbox_fs": [
+                        105,
+                        170,
+                        502,
+                        186
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        227,
+                        189,
+                        384,
+                        222
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                227,
+                                189,
+                                384,
+                                222
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        227,
+                                        189,
+                                        384,
+                                        222
+                                    ],
+                                    "score": 0.95,
+                                    "content": "\\mathbf { g } ( \\mathbf { w } ) : = \\nabla L ( \\mathbf { w } ) = \\frac { 1 } { n } \\sum _ { i = 1 } ^ { n } \\nabla \\ell ( \\mathbf { w } , \\mathbf { x } _ { i } ) .",
+                                    "type": "interline_equation",
+                                    "image_path": "5ce375a0c0b4eba72df983bc418c05ba2ee824ff817e740a7b4a68960b2471aa.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 7.5,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                227,
+                                189,
+                                384,
+                                205.5
+                            ],
+                            "spans": [],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                227,
+                                205.5,
+                                384,
+                                222.0
+                            ],
+                            "spans": [],
+                            "index": 8
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        227,
+                        505,
+                        261
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                225,
+                                505,
+                                240
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        225,
+                                        160,
+                                        240
+                                    ],
+                                    "score": 1.0,
+                                    "content": "A mini-batch",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        161,
+                                        228,
+                                        176,
+                                        238
+                                    ],
+                                    "score": 0.89,
+                                    "content": "B _ { m }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        176,
+                                        225,
+                                        204,
+                                        240
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of size",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        204,
+                                        228,
+                                        234,
+                                        237
+                                    ],
+                                    "score": 0.89,
+                                    "content": "m < n",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        234,
+                                        225,
+                                        301,
+                                        240
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is a collection of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        301,
+                                        229,
+                                        311,
+                                        237
+                                    ],
+                                    "score": 0.79,
+                                    "content": "m",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        311,
+                                        225,
+                                        457,
+                                        240
+                                    ],
+                                    "score": 1.0,
+                                    "content": "indices randomly drawn from the set",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        457,
+                                        227,
+                                        501,
+                                        239
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\{ 1 , \\ldots , n \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        501,
+                                        225,
+                                        505,
+                                        240
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                237,
+                                505,
+                                250
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        237,
+                                        409,
+                                        250
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and we can use it to form an unbiased estimate of the gradient at iteration",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        410,
+                                        239,
+                                        416,
+                                        248
+                                    ],
+                                    "score": 0.81,
+                                    "content": "k",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        416,
+                                        237,
+                                        505,
+                                        250
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", as well as the corre-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                249,
+                                200,
+                                262
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        249,
+                                        200,
+                                        262
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sponding SGD update:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        }
+                    ],
+                    "index": 10,
+                    "bbox_fs": [
+                        105,
+                        225,
+                        505,
+                        262
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        163,
+                        265,
+                        447,
+                        296
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                163,
+                                265,
+                                447,
+                                296
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        163,
+                                        265,
+                                        447,
+                                        296
+                                    ],
+                                    "score": 0.94,
+                                    "content": "\\mathbf { g } _ { m } ( \\mathbf { w } _ { k } ) = \\frac { 1 } { m } \\sum _ { i \\in \\mathcal { B } _ { m } } \\nabla \\ell ( \\mathbf { w } _ { k } , \\mathbf { x } _ { i } ) \\quad \\mathrm { ~ a n d ~ } \\quad \\mathbf { w } _ { k + 1 } = \\mathbf { w } _ { k } - \\eta _ { k } \\mathbf { g } _ { m } ( \\mathbf { w } _ { k } ) ,",
+                                    "type": "interline_equation",
+                                    "image_path": "83e3f1f25ec3b8d3387d926f3bff0f8c4141099ab08b61b4c0c318b9c7452052.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 13,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                163,
+                                265,
+                                447,
+                                275.3333333333333
+                            ],
+                            "spans": [],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                163,
+                                275.3333333333333,
+                                447,
+                                285.66666666666663
+                            ],
+                            "spans": [],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                163,
+                                285.66666666666663,
+                                447,
+                                295.99999999999994
+                            ],
+                            "spans": [],
+                            "index": 14
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        300,
+                        505,
+                        334
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                300,
+                                506,
+                                313
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        300,
+                                        133,
+                                        313
+                                    ],
+                                    "score": 1.0,
+                                    "content": "where",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        133,
+                                        301,
+                                        163,
+                                        312
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\eta _ { k } > 0",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        163,
+                                        300,
+                                        290,
+                                        313
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is the learning rate for iteration",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        290,
+                                        301,
+                                        297,
+                                        311
+                                    ],
+                                    "score": 0.78,
+                                    "content": "k",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        297,
+                                        300,
+                                        506,
+                                        313
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". One iteration of training for SGD corresponds to a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                311,
+                                505,
+                                325
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        311,
+                                        396,
+                                        325
+                                    ],
+                                    "score": 1.0,
+                                    "content": "single gradient computation / weight update. One epoch corresponds to",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        396,
+                                        312,
+                                        417,
+                                        324
+                                    ],
+                                    "score": 0.92,
+                                    "content": "n / m",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        417,
+                                        311,
+                                        505,
+                                        325
+                                    ],
+                                    "score": 1.0,
+                                    "content": "iterations of training.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                321,
+                                505,
+                                335
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        321,
+                                        505,
+                                        335
+                                    ],
+                                    "score": 1.0,
+                                    "content": "This constitutes a single pass over the dataset, assuming the dataset is sampled without replacement.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        }
+                    ],
+                    "index": 16,
+                    "bbox_fs": [
+                        105,
+                        300,
+                        506,
+                        335
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        339,
+                        503,
+                        373
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                339,
+                                505,
+                                353
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        339,
+                                        505,
+                                        353
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Efficient distributed systems reduce wall-clock training time by parallelizing gradient calculations",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                350,
+                                504,
+                                362
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        350,
+                                        504,
+                                        362
+                                    ],
+                                    "score": 1.0,
+                                    "content": "across many machines. When the batch size is large enough to populate all available compute",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                362,
+                                446,
+                                374
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        362,
+                                        446,
+                                        374
+                                    ],
+                                    "score": 1.0,
+                                    "content": "resources, this allows us to amortize the cost of coordination for each weight update.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        }
+                    ],
+                    "index": 19,
+                    "bbox_fs": [
+                        106,
+                        339,
+                        505,
+                        374
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        378,
+                        505,
+                        445
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                378,
+                                506,
+                                391
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        378,
+                                        506,
+                                        391
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Existing large batch techniques. With the hope of keeping training times manageable as dataset",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                390,
+                                505,
+                                401
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        390,
+                                        505,
+                                        401
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sizes escalate, recent work has focused on the development of techniques that allow practitioners",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                400,
+                                506,
+                                413
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        400,
+                                        506,
+                                        413
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to increase the batch size to make use of growing computational resources (Jin et al., 2016; Jia",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                411,
+                                505,
+                                424
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        411,
+                                        505,
+                                        424
+                                    ],
+                                    "score": 1.0,
+                                    "content": "et al., 2018; You et al., 2017a). However, there is a growing body of theoretical and empirical",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                422,
+                                506,
+                                435
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        422,
+                                        506,
+                                        435
+                                    ],
+                                    "score": 1.0,
+                                    "content": "results suggesting that large batch sizes adversely affect the generalization performance of the final",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                433,
+                                392,
+                                445
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        433,
+                                        392,
+                                        445
+                                    ],
+                                    "score": 1.0,
+                                    "content": "model (Yao et al., 2018; Keskar et al., 2016; Devarakonda et al., 2017).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        }
+                    ],
+                    "index": 23.5,
+                    "bbox_fs": [
+                        105,
+                        378,
+                        506,
+                        445
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        450,
+                        504,
+                        483
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                450,
+                                505,
+                                463
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        450,
+                                        505,
+                                        463
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In response to this, recent work has proposed changing two parameters in relation to batch size: the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                461,
+                                505,
+                                473
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        461,
+                                        505,
+                                        473
+                                    ],
+                                    "score": 1.0,
+                                    "content": "number of training iterations and the learning rate. However, they also make assumptions that limit",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                472,
+                                395,
+                                484
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        472,
+                                        395,
+                                        484
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the effectiveness of their proposals as useful heuristics for practitioners.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        }
+                    ],
+                    "index": 28,
+                    "bbox_fs": [
+                        105,
+                        450,
+                        505,
+                        484
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        132,
+                        493,
+                        505,
+                        699
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                133,
+                                493,
+                                505,
+                                506
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        133,
+                                        493,
+                                        505,
+                                        506
+                                    ],
+                                    "score": 1.0,
+                                    "content": "• Training longer: Hoffer et al. (2017) suggest increasing the number of training iterations.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                504,
+                                506,
+                                518
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        504,
+                                        506,
+                                        518
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Even if this does reduce the generalization gap, it significantly increases both wall-clock",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                516,
+                                505,
+                                529
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        516,
+                                        505,
+                                        529
+                                    ],
+                                    "score": 1.0,
+                                    "content": "training time and computational cost. Moreover, in some problems it does not lead to min-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                142,
+                                528,
+                                502,
+                                539
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        142,
+                                        528,
+                                        502,
+                                        539
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ima with better generalization performance (as we found when running our experiments).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                132,
+                                541,
+                                505,
+                                555
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        132,
+                                        542,
+                                        367,
+                                        555
+                                    ],
+                                    "score": 1.0,
+                                    "content": "• Square root LR scaling: Scaling the learning rate as",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        368,
+                                        541,
+                                        415,
+                                        553
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\eta _ { 0 } \\propto \\sqrt { m }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        416,
+                                        542,
+                                        505,
+                                        555
+                                    ],
+                                    "score": 1.0,
+                                    "content": "attempts to keep the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                142,
+                                553,
+                                505,
+                                564
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        142,
+                                        553,
+                                        505,
+                                        564
+                                    ],
+                                    "score": 1.0,
+                                    "content": "weight increment length statistics constant, but the distance between SGD iterates is gov-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                564,
+                                505,
+                                575
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        564,
+                                        505,
+                                        575
+                                    ],
+                                    "score": 1.0,
+                                    "content": "erned more by properties of the objective function than the ratio of learning rate to batch",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                142,
+                                574,
+                                505,
+                                586
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        142,
+                                        574,
+                                        505,
+                                        586
+                                    ],
+                                    "score": 1.0,
+                                    "content": "size (Chaudhari & Soatto, 2017; Zhu et al., 2018). This rule has also been found to be",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                586,
+                                435,
+                                597
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        586,
+                                        435,
+                                        597
+                                    ],
+                                    "score": 1.0,
+                                    "content": "empirically sub-optimal in various problem domains (Krizhevsky, 2014).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                137,
+                                600,
+                                506,
+                                613
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        137,
+                                        600,
+                                        506,
+                                        613
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Linear LR scaling: The performance of large batch training can also be improved by using",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                611,
+                                506,
+                                624
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        611,
+                                        506,
+                                        624
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the linear scaling rule, which suggests choosing a learning rate proportional to the batch size",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                144,
+                                622,
+                                506,
+                                635
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        144,
+                                        623,
+                                        183,
+                                        634
+                                    ],
+                                    "score": 0.84,
+                                    "content": "( \\eta _ { 0 } \\propto m )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        184,
+                                        622,
+                                        506,
+                                        635
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(Goyal et al., 2017). There are two motivations for this rule: the first assumes",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                632,
+                                505,
+                                646
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        632,
+                                        505,
+                                        646
+                                    ],
+                                    "score": 1.0,
+                                    "content": "that one large-batch gradient step should resemble a series of small-batch gradient steps",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                644,
+                                506,
+                                657
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        644,
+                                        506,
+                                        657
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in order for convergence rates to improve linearly (Goyal et al., 2017); the other regards",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                655,
+                                505,
+                                666
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        655,
+                                        505,
+                                        666
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the SGD update equation as the Euler-Maruyama discretization of a stochastic differential",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                666,
+                                505,
+                                679
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        666,
+                                        505,
+                                        679
+                                    ],
+                                    "score": 1.0,
+                                    "content": "equation (Sauer, 2012; Xing et al., 2018), and attempts to maintain a constant level of mini-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                677,
+                                505,
+                                690
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        677,
+                                        505,
+                                        690
+                                    ],
+                                    "score": 1.0,
+                                    "content": "batch noise to help SGD explore the loss landscape (Chaudhari & Soatto, 2017; Zhu et al.,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        },
+                        {
+                            "bbox": [
+                                142,
+                                688,
+                                248,
+                                700
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        142,
+                                        688,
+                                        248,
+                                        700
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2018; Smith & Le, 2018).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 47
+                        }
+                    ],
+                    "index": 38.5,
+                    "bbox_fs": [
+                        132,
+                        493,
+                        506,
+                        700
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        709,
+                        502,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                709,
+                                504,
+                                722
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        709,
+                                        504,
+                                        722
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Both justifications for the linear scaling rule implicitly impose strong conditions on the loss func-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 48
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                719,
+                                504,
+                                734
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        719,
+                                        504,
+                                        734
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tion by requiring that it behave linearly near SGD iterates; therefore, if the loss function is highly",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 49
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                82,
+                                506,
+                                95
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        82,
+                                        506,
+                                        95
+                                    ],
+                                    "score": 1.0,
+                                    "content": "nonlinear along the SGD trajectory or the step size is not small enough, then we should not expect",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                92,
+                                505,
+                                107
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        92,
+                                        505,
+                                        107
+                                    ],
+                                    "score": 1.0,
+                                    "content": "these rules to provide useful guidance for many problems. Whereas several groups have success-",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                104,
+                                505,
+                                117
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        104,
+                                        505,
+                                        117
+                                    ],
+                                    "score": 1.0,
+                                    "content": "fully used this rule to train on the ImageNet dataset in under an hour, e.g. (Goyal et al., 2017; You",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                115,
+                                505,
+                                128
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        115,
+                                        505,
+                                        128
+                                    ],
+                                    "score": 1.0,
+                                    "content": "et al., 2017b), applying this heuristic to other datasets has not led to similarly impressive results so",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                126,
+                                195,
+                                139
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        126,
+                                        195,
+                                        139
+                                    ],
+                                    "score": 1.0,
+                                    "content": "far (Puri et al., 2018).",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 4
+                        }
+                    ],
+                    "index": 48.5,
+                    "bbox_fs": [
+                        105,
+                        709,
+                        504,
+                        734
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        82,
+                        505,
+                        137
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                82,
+                                506,
+                                95
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        82,
+                                        506,
+                                        95
+                                    ],
+                                    "score": 1.0,
+                                    "content": "nonlinear along the SGD trajectory or the step size is not small enough, then we should not expect",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                92,
+                                505,
+                                107
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        92,
+                                        505,
+                                        107
+                                    ],
+                                    "score": 1.0,
+                                    "content": "these rules to provide useful guidance for many problems. Whereas several groups have success-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                104,
+                                505,
+                                117
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        104,
+                                        505,
+                                        117
+                                    ],
+                                    "score": 1.0,
+                                    "content": "fully used this rule to train on the ImageNet dataset in under an hour, e.g. (Goyal et al., 2017; You",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                115,
+                                505,
+                                128
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        115,
+                                        505,
+                                        128
+                                    ],
+                                    "score": 1.0,
+                                    "content": "et al., 2017b), applying this heuristic to other datasets has not led to similarly impressive results so",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                126,
+                                195,
+                                139
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        126,
+                                        195,
+                                        139
+                                    ],
+                                    "score": 1.0,
+                                    "content": "far (Puri et al., 2018).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        }
+                    ],
+                    "index": 2
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        143,
+                        505,
+                        209
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                142,
+                                505,
+                                155
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        142,
+                                        505,
+                                        155
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The focus of this paper, however, is on more fundamental limitations of large batch training, and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                154,
+                                506,
+                                166
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        154,
+                                        506,
+                                        166
+                                    ],
+                                    "score": 1.0,
+                                    "content": "we empirically show that the above approaches fail to prevent diminishing returns in the rate of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                165,
+                                506,
+                                177
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        165,
+                                        506,
+                                        177
+                                    ],
+                                    "score": 1.0,
+                                    "content": "convergence for large batch sizes. We believe that these diminishing returns are of more immediate",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                176,
+                                505,
+                                188
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        176,
+                                        505,
+                                        188
+                                    ],
+                                    "score": 1.0,
+                                    "content": "concern than the generalization gap and warrant more careful examination: if we cannot even mini-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                187,
+                                506,
+                                200
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        187,
+                                        506,
+                                        200
+                                    ],
+                                    "score": 1.0,
+                                    "content": "mize training error quickly, there is no real opportunity to minimize test error quickly, regardless of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                197,
+                                457,
+                                211
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        197,
+                                        457,
+                                        211
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the difference in final test error across batch sizes by the time the model has converged.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        }
+                    ],
+                    "index": 7.5
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        232,
+                        398,
+                        245
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                231,
+                                399,
+                                247
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        231,
+                                        399,
+                                        247
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3 CRITICAL BATCH SIZES AND DIMINISHING RETURNS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        }
+                    ],
+                    "index": 11
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        261,
+                        505,
+                        327
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                261,
+                                505,
+                                274
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        261,
+                                        284,
+                                        274
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The convergence rate of SGD, denoted by",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        284,
+                                        261,
+                                        311,
+                                        273
+                                    ],
+                                    "score": 0.92,
+                                    "content": "k _ { \\epsilon } ( m )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        311,
+                                        261,
+                                        505,
+                                        274
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", is the number of iterations needed to achieve",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                272,
+                                505,
+                                285
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        272,
+                                        268,
+                                        285
+                                    ],
+                                    "score": 1.0,
+                                    "content": "training error less than a fixed constant",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        268,
+                                        273,
+                                        294,
+                                        283
+                                    ],
+                                    "score": 0.89,
+                                    "content": "\\epsilon > 0",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        294,
+                                        272,
+                                        420,
+                                        285
+                                    ],
+                                    "score": 1.0,
+                                    "content": "by using SGD with batch size",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        420,
+                                        274,
+                                        431,
+                                        283
+                                    ],
+                                    "score": 0.7,
+                                    "content": "m",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        431,
+                                        272,
+                                        505,
+                                        285
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(we will drop the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                281,
+                                506,
+                                298
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        281,
+                                        145,
+                                        298
+                                    ],
+                                    "score": 1.0,
+                                    "content": "subscript",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        145,
+                                        285,
+                                        151,
+                                        293
+                                    ],
+                                    "score": 0.63,
+                                    "content": "\\epsilon",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        151,
+                                        281,
+                                        506,
+                                        298
+                                    ],
+                                    "score": 1.0,
+                                    "content": "when it is unambiguous). In order to guarantee that large batch sizes speed up training,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                107,
+                                294,
+                                505,
+                                307
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        294,
+                                        129,
+                                        307
+                                    ],
+                                    "score": 0.91,
+                                    "content": "k ( m )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        130,
+                                        294,
+                                        324,
+                                        307
+                                    ],
+                                    "score": 1.0,
+                                    "content": "should continue to decrease near-linearly with",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        324,
+                                        296,
+                                        333,
+                                        304
+                                    ],
+                                    "score": 0.7,
+                                    "content": "m",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        334,
+                                        294,
+                                        505,
+                                        307
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Otherwise, a larger batch size increases",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                305,
+                                505,
+                                318
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        306,
+                                        478,
+                                        318
+                                    ],
+                                    "score": 1.0,
+                                    "content": "computational cost with only limited reductions in wall-clock training time. For near-constant",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        478,
+                                        305,
+                                        501,
+                                        317
+                                    ],
+                                    "score": 0.94,
+                                    "content": "k ( m )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        501,
+                                        306,
+                                        505,
+                                        318
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                317,
+                                309,
+                                329
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        317,
+                                        309,
+                                        329
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the benefit of large batch sizes becomes near-zero.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        }
+                    ],
+                    "index": 14.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        333,
+                        505,
+                        432
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                334,
+                                505,
+                                345
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        334,
+                                        505,
+                                        345
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Ma et al. (2017) showed theoretically that in convex, over-parameterized settings, the reduction in",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                345,
+                                505,
+                                356
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        345,
+                                        505,
+                                        356
+                                    ],
+                                    "score": 1.0,
+                                    "content": "convergence time obtained by increasing the batch size decays dramatically to a near-constant level",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                356,
+                                505,
+                                367
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        356,
+                                        505,
+                                        367
+                                    ],
+                                    "score": 1.0,
+                                    "content": "after a critical batch size that is independent of the dataset size. This speedup is measured with",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                366,
+                                505,
+                                379
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        366,
+                                        505,
+                                        379
+                                    ],
+                                    "score": 1.0,
+                                    "content": "respect to the number of SGD iterations required to reach some fixed loss error for some baseline",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                377,
+                                505,
+                                390
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        377,
+                                        150,
+                                        390
+                                    ],
+                                    "score": 1.0,
+                                    "content": "batch size",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        150,
+                                        379,
+                                        164,
+                                        388
+                                    ],
+                                    "score": 0.86,
+                                    "content": "m _ { 0 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        165,
+                                        377,
+                                        371,
+                                        390
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", and for this purpose we define the speedup ratio",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        371,
+                                        377,
+                                        480,
+                                        389
+                                    ],
+                                    "score": 0.91,
+                                    "content": "s ( m ; m _ { 0 } ) = k ( m _ { 0 } ) / k ( m )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        480,
+                                        377,
+                                        505,
+                                        390
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". The",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                388,
+                                505,
+                                401
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        388,
+                                        442,
+                                        401
+                                    ],
+                                    "score": 1.0,
+                                    "content": "speedup ratio represents the amount of time we save by increasing the batch size to",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        442,
+                                        390,
+                                        452,
+                                        398
+                                    ],
+                                    "score": 0.57,
+                                    "content": "m",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        452,
+                                        388,
+                                        505,
+                                        401
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Beyond the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                399,
+                                505,
+                                411
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        399,
+                                        505,
+                                        411
+                                    ],
+                                    "score": 1.0,
+                                    "content": "critical batch size mentioned above, even with no communication overhead and unlimited resources",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                410,
+                                505,
+                                423
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        410,
+                                        505,
+                                        423
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(where each batch size requires the same amount of wall-clock time to process) we would prefer to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                421,
+                                387,
+                                433
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        421,
+                                        387,
+                                        433
+                                    ],
+                                    "score": 1.0,
+                                    "content": "use the critical batch size because it requires less overall computation.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        }
+                    ],
+                    "index": 22
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        438,
+                        504,
+                        493
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                438,
+                                505,
+                                450
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        438,
+                                        505,
+                                        450
+                                    ],
+                                    "score": 1.0,
+                                    "content": "This result is surprising because researchers have asserted that it should be possible to achieve linear",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                449,
+                                506,
+                                461
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        449,
+                                        506,
+                                        461
+                                    ],
+                                    "score": 1.0,
+                                    "content": "gains in convergence speed so long as the batch size is small relative to dataset size (Smith &",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                459,
+                                505,
+                                473
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        459,
+                                        505,
+                                        473
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Le, 2018). This will present significant difficulties for future optimization work (large mini-batch",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                471,
+                                506,
+                                484
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        471,
+                                        506,
+                                        484
+                                    ],
+                                    "score": 1.0,
+                                    "content": "training) because it prevents us from using large batch sizes as a catch-all approach to quickly train",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                481,
+                                209,
+                                495
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        481,
+                                        209,
+                                        495
+                                    ],
+                                    "score": 1.0,
+                                    "content": "models on large datasets.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        }
+                    ],
+                    "index": 29
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        516,
+                        254,
+                        528
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                514,
+                                256,
+                                531
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        514,
+                                        256,
+                                        531
+                                    ],
+                                    "score": 1.0,
+                                    "content": "4 EMPIRICAL EVALUATION",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        }
+                    ],
+                    "index": 32
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        545,
+                        505,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                544,
+                                505,
+                                557
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        544,
+                                        505,
+                                        557
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Recent work studying large batch training has looked primarily at image classification (Jastrzebski",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                555,
+                                505,
+                                569
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        555,
+                                        505,
+                                        569
+                                    ],
+                                    "score": 1.0,
+                                    "content": "et al., 2018; Yao et al., 2018), especially on the ImageNet dataset (Deng et al., 2009). We per-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                568,
+                                505,
+                                579
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        568,
+                                        505,
+                                        579
+                                    ],
+                                    "score": 1.0,
+                                    "content": "form large batch size experiments across both traditional image classification (IC) tasks (such as",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                577,
+                                506,
+                                592
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        577,
+                                        506,
+                                        592
+                                    ],
+                                    "score": 1.0,
+                                    "content": "on CIFAR-10/100 (Krizhevsky & Hinton, 2009)), as well as previously unexplored tasks like image",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                587,
+                                506,
+                                604
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        587,
+                                        506,
+                                        604
+                                    ],
+                                    "score": 1.0,
+                                    "content": "segmentation (IS) using the Cityscapes dataset (Cordts et al., 2016), and natural language processing",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                600,
+                                506,
+                                613
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        600,
+                                        506,
+                                        613
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(NLP) using the WikiText-2 dataset (Merity et al., 2016). We also test how these results vary across",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                611,
+                                504,
+                                623
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        611,
+                                        504,
+                                        623
+                                    ],
+                                    "score": 1.0,
+                                    "content": "other modern DNN architectures, namely ResNets (He et al., 2016), LSTMs (Hochreiter & Schmid-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                621,
+                                505,
+                                635
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        621,
+                                        505,
+                                        635
+                                    ],
+                                    "score": 1.0,
+                                    "content": "huber, 1997; Gers et al., 2000), AlexNet (Krizhevsky et al., 2012), VGG (Simonyan & Zisserman,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                632,
+                                505,
+                                645
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        632,
+                                        505,
+                                        645
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2014), Dilated Residual Networks (Yu et al., 2017), and MobileNetV2 (Sandler et al., 2014). We",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                642,
+                                506,
+                                658
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        642,
+                                        506,
+                                        658
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tested all of the large batch training techniques described in Section 2. We tried training longer",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                654,
+                                505,
+                                668
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        654,
+                                        505,
+                                        668
+                                    ],
+                                    "score": 1.0,
+                                    "content": "based on the work of Hoffer et al. (2017), but we found that this necessarily cannot improve the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                665,
+                                506,
+                                679
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        665,
+                                        506,
+                                        679
+                                    ],
+                                    "score": 1.0,
+                                    "content": "convergence speed and often does not improve final test performance. The two other techniques",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                677,
+                                505,
+                                690
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        677,
+                                        505,
+                                        690
+                                    ],
+                                    "score": 1.0,
+                                    "content": "include the square root scaling rule strategy (SRSR) and the linear scaling rule strategy (LSR). For",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                687,
+                                505,
+                                701
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        687,
+                                        505,
+                                        701
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the latter, we used a warm-up period at the start of training as suggested by Goyal et al. (2017).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                698,
+                                506,
+                                712
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        698,
+                                        506,
+                                        712
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Table 1 reports our datasets, models and different training strategies. For each model, we evaluated",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 47
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                709,
+                                506,
+                                723
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        709,
+                                        506,
+                                        723
+                                    ],
+                                    "score": 1.0,
+                                    "content": "against a base learning rate strategy (BLR) that used the same learning rate across all batch sizes.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 48
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                721,
+                                452,
+                                732
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        721,
+                                        452,
+                                        732
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We selected this learning rate based on its performance on a small baseline batch size.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 49
+                        }
+                    ],
+                    "index": 41
+                }
+            ],
+            "page_idx": 3,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        107,
+                        27,
+                        308,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                107,
+                                26,
+                                308,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        26,
+                                        308,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Under review as a conference paper at ICLR 2019",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        302,
+                        752,
+                        308,
+                        759
+                    ],
+                    "lines": []
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        82,
+                        505,
+                        137
+                    ],
+                    "lines": [],
+                    "index": 2,
+                    "bbox_fs": [
+                        105,
+                        82,
+                        506,
+                        139
+                    ],
+                    "lines_deleted": true
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        143,
+                        505,
+                        209
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                142,
+                                505,
+                                155
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        142,
+                                        505,
+                                        155
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The focus of this paper, however, is on more fundamental limitations of large batch training, and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                154,
+                                506,
+                                166
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        154,
+                                        506,
+                                        166
+                                    ],
+                                    "score": 1.0,
+                                    "content": "we empirically show that the above approaches fail to prevent diminishing returns in the rate of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                165,
+                                506,
+                                177
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        165,
+                                        506,
+                                        177
+                                    ],
+                                    "score": 1.0,
+                                    "content": "convergence for large batch sizes. We believe that these diminishing returns are of more immediate",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                176,
+                                505,
+                                188
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        176,
+                                        505,
+                                        188
+                                    ],
+                                    "score": 1.0,
+                                    "content": "concern than the generalization gap and warrant more careful examination: if we cannot even mini-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                187,
+                                506,
+                                200
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        187,
+                                        506,
+                                        200
+                                    ],
+                                    "score": 1.0,
+                                    "content": "mize training error quickly, there is no real opportunity to minimize test error quickly, regardless of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                197,
+                                457,
+                                211
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        197,
+                                        457,
+                                        211
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the difference in final test error across batch sizes by the time the model has converged.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        }
+                    ],
+                    "index": 7.5,
+                    "bbox_fs": [
+                        105,
+                        142,
+                        506,
+                        211
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        232,
+                        398,
+                        245
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                231,
+                                399,
+                                247
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        231,
+                                        399,
+                                        247
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3 CRITICAL BATCH SIZES AND DIMINISHING RETURNS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        }
+                    ],
+                    "index": 11
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        261,
+                        505,
+                        327
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                261,
+                                505,
+                                274
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        261,
+                                        284,
+                                        274
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The convergence rate of SGD, denoted by",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        284,
+                                        261,
+                                        311,
+                                        273
+                                    ],
+                                    "score": 0.92,
+                                    "content": "k _ { \\epsilon } ( m )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        311,
+                                        261,
+                                        505,
+                                        274
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", is the number of iterations needed to achieve",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                272,
+                                505,
+                                285
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        272,
+                                        268,
+                                        285
+                                    ],
+                                    "score": 1.0,
+                                    "content": "training error less than a fixed constant",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        268,
+                                        273,
+                                        294,
+                                        283
+                                    ],
+                                    "score": 0.89,
+                                    "content": "\\epsilon > 0",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        294,
+                                        272,
+                                        420,
+                                        285
+                                    ],
+                                    "score": 1.0,
+                                    "content": "by using SGD with batch size",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        420,
+                                        274,
+                                        431,
+                                        283
+                                    ],
+                                    "score": 0.7,
+                                    "content": "m",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        431,
+                                        272,
+                                        505,
+                                        285
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(we will drop the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                281,
+                                506,
+                                298
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        281,
+                                        145,
+                                        298
+                                    ],
+                                    "score": 1.0,
+                                    "content": "subscript",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        145,
+                                        285,
+                                        151,
+                                        293
+                                    ],
+                                    "score": 0.63,
+                                    "content": "\\epsilon",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        151,
+                                        281,
+                                        506,
+                                        298
+                                    ],
+                                    "score": 1.0,
+                                    "content": "when it is unambiguous). In order to guarantee that large batch sizes speed up training,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                107,
+                                294,
+                                505,
+                                307
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        294,
+                                        129,
+                                        307
+                                    ],
+                                    "score": 0.91,
+                                    "content": "k ( m )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        130,
+                                        294,
+                                        324,
+                                        307
+                                    ],
+                                    "score": 1.0,
+                                    "content": "should continue to decrease near-linearly with",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        324,
+                                        296,
+                                        333,
+                                        304
+                                    ],
+                                    "score": 0.7,
+                                    "content": "m",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        334,
+                                        294,
+                                        505,
+                                        307
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Otherwise, a larger batch size increases",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                305,
+                                505,
+                                318
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        306,
+                                        478,
+                                        318
+                                    ],
+                                    "score": 1.0,
+                                    "content": "computational cost with only limited reductions in wall-clock training time. For near-constant",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        478,
+                                        305,
+                                        501,
+                                        317
+                                    ],
+                                    "score": 0.94,
+                                    "content": "k ( m )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        501,
+                                        306,
+                                        505,
+                                        318
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                317,
+                                309,
+                                329
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        317,
+                                        309,
+                                        329
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the benefit of large batch sizes becomes near-zero.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        }
+                    ],
+                    "index": 14.5,
+                    "bbox_fs": [
+                        105,
+                        261,
+                        506,
+                        329
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        333,
+                        505,
+                        432
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                334,
+                                505,
+                                345
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        334,
+                                        505,
+                                        345
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Ma et al. (2017) showed theoretically that in convex, over-parameterized settings, the reduction in",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                345,
+                                505,
+                                356
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        345,
+                                        505,
+                                        356
+                                    ],
+                                    "score": 1.0,
+                                    "content": "convergence time obtained by increasing the batch size decays dramatically to a near-constant level",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                356,
+                                505,
+                                367
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        356,
+                                        505,
+                                        367
+                                    ],
+                                    "score": 1.0,
+                                    "content": "after a critical batch size that is independent of the dataset size. This speedup is measured with",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                366,
+                                505,
+                                379
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        366,
+                                        505,
+                                        379
+                                    ],
+                                    "score": 1.0,
+                                    "content": "respect to the number of SGD iterations required to reach some fixed loss error for some baseline",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                377,
+                                505,
+                                390
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        377,
+                                        150,
+                                        390
+                                    ],
+                                    "score": 1.0,
+                                    "content": "batch size",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        150,
+                                        379,
+                                        164,
+                                        388
+                                    ],
+                                    "score": 0.86,
+                                    "content": "m _ { 0 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        165,
+                                        377,
+                                        371,
+                                        390
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", and for this purpose we define the speedup ratio",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        371,
+                                        377,
+                                        480,
+                                        389
+                                    ],
+                                    "score": 0.91,
+                                    "content": "s ( m ; m _ { 0 } ) = k ( m _ { 0 } ) / k ( m )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        480,
+                                        377,
+                                        505,
+                                        390
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". The",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                388,
+                                505,
+                                401
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        388,
+                                        442,
+                                        401
+                                    ],
+                                    "score": 1.0,
+                                    "content": "speedup ratio represents the amount of time we save by increasing the batch size to",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        442,
+                                        390,
+                                        452,
+                                        398
+                                    ],
+                                    "score": 0.57,
+                                    "content": "m",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        452,
+                                        388,
+                                        505,
+                                        401
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Beyond the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                399,
+                                505,
+                                411
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        399,
+                                        505,
+                                        411
+                                    ],
+                                    "score": 1.0,
+                                    "content": "critical batch size mentioned above, even with no communication overhead and unlimited resources",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                410,
+                                505,
+                                423
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        410,
+                                        505,
+                                        423
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(where each batch size requires the same amount of wall-clock time to process) we would prefer to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                421,
+                                387,
+                                433
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        421,
+                                        387,
+                                        433
+                                    ],
+                                    "score": 1.0,
+                                    "content": "use the critical batch size because it requires less overall computation.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        }
+                    ],
+                    "index": 22,
+                    "bbox_fs": [
+                        105,
+                        334,
+                        505,
+                        433
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        438,
+                        504,
+                        493
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                438,
+                                505,
+                                450
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        438,
+                                        505,
+                                        450
+                                    ],
+                                    "score": 1.0,
+                                    "content": "This result is surprising because researchers have asserted that it should be possible to achieve linear",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                449,
+                                506,
+                                461
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        449,
+                                        506,
+                                        461
+                                    ],
+                                    "score": 1.0,
+                                    "content": "gains in convergence speed so long as the batch size is small relative to dataset size (Smith &",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                459,
+                                505,
+                                473
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        459,
+                                        505,
+                                        473
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Le, 2018). This will present significant difficulties for future optimization work (large mini-batch",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                471,
+                                506,
+                                484
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        471,
+                                        506,
+                                        484
+                                    ],
+                                    "score": 1.0,
+                                    "content": "training) because it prevents us from using large batch sizes as a catch-all approach to quickly train",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                481,
+                                209,
+                                495
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        481,
+                                        209,
+                                        495
+                                    ],
+                                    "score": 1.0,
+                                    "content": "models on large datasets.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        }
+                    ],
+                    "index": 29,
+                    "bbox_fs": [
+                        105,
+                        438,
+                        506,
+                        495
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        516,
+                        254,
+                        528
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                514,
+                                256,
+                                531
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        514,
+                                        256,
+                                        531
+                                    ],
+                                    "score": 1.0,
+                                    "content": "4 EMPIRICAL EVALUATION",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        }
+                    ],
+                    "index": 32
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        545,
+                        505,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                544,
+                                505,
+                                557
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        544,
+                                        505,
+                                        557
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Recent work studying large batch training has looked primarily at image classification (Jastrzebski",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                555,
+                                505,
+                                569
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        555,
+                                        505,
+                                        569
+                                    ],
+                                    "score": 1.0,
+                                    "content": "et al., 2018; Yao et al., 2018), especially on the ImageNet dataset (Deng et al., 2009). We per-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                568,
+                                505,
+                                579
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        568,
+                                        505,
+                                        579
+                                    ],
+                                    "score": 1.0,
+                                    "content": "form large batch size experiments across both traditional image classification (IC) tasks (such as",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                577,
+                                506,
+                                592
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        577,
+                                        506,
+                                        592
+                                    ],
+                                    "score": 1.0,
+                                    "content": "on CIFAR-10/100 (Krizhevsky & Hinton, 2009)), as well as previously unexplored tasks like image",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                587,
+                                506,
+                                604
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        587,
+                                        506,
+                                        604
+                                    ],
+                                    "score": 1.0,
+                                    "content": "segmentation (IS) using the Cityscapes dataset (Cordts et al., 2016), and natural language processing",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                600,
+                                506,
+                                613
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        600,
+                                        506,
+                                        613
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(NLP) using the WikiText-2 dataset (Merity et al., 2016). We also test how these results vary across",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                611,
+                                504,
+                                623
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        611,
+                                        504,
+                                        623
+                                    ],
+                                    "score": 1.0,
+                                    "content": "other modern DNN architectures, namely ResNets (He et al., 2016), LSTMs (Hochreiter & Schmid-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                621,
+                                505,
+                                635
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        621,
+                                        505,
+                                        635
+                                    ],
+                                    "score": 1.0,
+                                    "content": "huber, 1997; Gers et al., 2000), AlexNet (Krizhevsky et al., 2012), VGG (Simonyan & Zisserman,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                632,
+                                505,
+                                645
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        632,
+                                        505,
+                                        645
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2014), Dilated Residual Networks (Yu et al., 2017), and MobileNetV2 (Sandler et al., 2014). We",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                642,
+                                506,
+                                658
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        642,
+                                        506,
+                                        658
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tested all of the large batch training techniques described in Section 2. We tried training longer",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                654,
+                                505,
+                                668
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        654,
+                                        505,
+                                        668
+                                    ],
+                                    "score": 1.0,
+                                    "content": "based on the work of Hoffer et al. (2017), but we found that this necessarily cannot improve the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                665,
+                                506,
+                                679
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        665,
+                                        506,
+                                        679
+                                    ],
+                                    "score": 1.0,
+                                    "content": "convergence speed and often does not improve final test performance. The two other techniques",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                677,
+                                505,
+                                690
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        677,
+                                        505,
+                                        690
+                                    ],
+                                    "score": 1.0,
+                                    "content": "include the square root scaling rule strategy (SRSR) and the linear scaling rule strategy (LSR). For",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                687,
+                                505,
+                                701
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        687,
+                                        505,
+                                        701
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the latter, we used a warm-up period at the start of training as suggested by Goyal et al. (2017).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                698,
+                                506,
+                                712
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        698,
+                                        506,
+                                        712
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Table 1 reports our datasets, models and different training strategies. For each model, we evaluated",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 47
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                709,
+                                506,
+                                723
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        709,
+                                        506,
+                                        723
+                                    ],
+                                    "score": 1.0,
+                                    "content": "against a base learning rate strategy (BLR) that used the same learning rate across all batch sizes.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 48
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                721,
+                                452,
+                                732
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        721,
+                                        452,
+                                        732
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We selected this learning rate based on its performance on a small baseline batch size.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 49
+                        }
+                    ],
+                    "index": 41,
+                    "bbox_fs": [
+                        104,
+                        544,
+                        506,
+                        732
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "image",
+                    "bbox": [
+                        115,
+                        85,
+                        496,
+                        529
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                115,
+                                85,
+                                496,
+                                529
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        85,
+                                        496,
+                                        529
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                115,
+                                                85,
+                                                496,
+                                                529
+                                            ],
+                                            "score": 0.975,
+                                            "type": "image",
+                                            "image_path": "b45c386a5ec3b1dee617dd3ed5dfe50c26a309dd84ec1efd0f4dcdac283a893c.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 1,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        85,
+                                        496,
+                                        233.0
+                                    ],
+                                    "spans": [],
+                                    "index": 0
+                                },
+                                {
+                                    "bbox": [
+                                        115,
+                                        233.0,
+                                        496,
+                                        381.0
+                                    ],
+                                    "spans": [],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        115,
+                                        381.0,
+                                        496,
+                                        529.0
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                106,
+                                539,
+                                505,
+                                590
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        538,
+                                        505,
+                                        551
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                538,
+                                                505,
+                                                551
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 1: Contour plots of training losses for various problem domains on a log scale. Lighter colors indicate",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 3
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        548,
+                                        505,
+                                        561
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                548,
+                                                505,
+                                                561
+                                            ],
+                                            "score": 1.0,
+                                            "content": "lower loss values. Since we train each batch size for a fixed number of epochs, the total number of training",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 4
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        558,
+                                        505,
+                                        572
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                558,
+                                                505,
+                                                572
+                                            ],
+                                            "score": 1.0,
+                                            "content": "iterations scales down linearly. For each loss value, we can observe how many iterations it takes to converge",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 5
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        568,
+                                        505,
+                                        582
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                568,
+                                                505,
+                                                582
+                                            ],
+                                            "score": 1.0,
+                                            "content": "to that value given a particular batch size, by tracing the level curve for the associated color. For all problems,",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 6
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        579,
+                                        488,
+                                        591
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                579,
+                                                488,
+                                                591
+                                            ],
+                                            "score": 1.0,
+                                            "content": "there is a batch size after which the number of training iterations necessary to converge does not decrease.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 7
+                                }
+                            ],
+                            "index": 5
+                        }
+                    ],
+                    "index": 3.0
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        615,
+                        358,
+                        627
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                615,
+                                359,
+                                628
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        615,
+                                        359,
+                                        628
+                                    ],
+                                    "score": 1.0,
+                                    "content": "4.1 DIMINISHING RETURNS IN RATES OF CONVERGENCE",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        }
+                    ],
+                    "index": 8
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        637,
+                        505,
+                        704
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                637,
+                                505,
+                                651
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        637,
+                                        505,
+                                        651
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We demonstrate the rapidly diminishing returns in rates of convergence across various problem do-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                649,
+                                505,
+                                661
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        649,
+                                        505,
+                                        661
+                                    ],
+                                    "score": 1.0,
+                                    "content": "mains and network configurations. Researchers increase the batch size in an attempt to achieve",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                660,
+                                505,
+                                673
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        660,
+                                        505,
+                                        673
+                                    ],
+                                    "score": 1.0,
+                                    "content": "nearly linear speedups in convergence compared to a small mini-batch size. In particular, if the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                671,
+                                505,
+                                683
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        671,
+                                        218,
+                                        683
+                                    ],
+                                    "score": 1.0,
+                                    "content": "speedup is near-linear, i.e.",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        219,
+                                        671,
+                                        371,
+                                        683
+                                    ],
+                                    "score": 0.91,
+                                    "content": "s ( m ; \\bar { m } _ { 0 } ) = k ( \\bar { m } _ { 0 } ) / k ( m ) \\approx m / m _ { 0 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        371,
+                                        671,
+                                        505,
+                                        683
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", then the computational cost re-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                681,
+                                505,
+                                694
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        682,
+                                        399,
+                                        694
+                                    ],
+                                    "score": 1.0,
+                                    "content": "mains nearly constant for large and small mini-batch SGD. However, if",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        399,
+                                        681,
+                                        466,
+                                        694
+                                    ],
+                                    "score": 0.93,
+                                    "content": "s ( m ) \\ \\bar { \\ll } \\ m / m _ { 0 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        466,
+                                        682,
+                                        505,
+                                        694
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", then the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                692,
+                                322,
+                                705
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        692,
+                                        322,
+                                        705
+                                    ],
+                                    "score": 1.0,
+                                    "content": "benefit of using large batch size training is negligible.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        }
+                    ],
+                    "index": 11.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        709,
+                        504,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                709,
+                                505,
+                                722
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        709,
+                                        505,
+                                        722
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In Figure 1, we show contour plots of training loss as a function of both the batch size and the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                721,
+                                505,
+                                731
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        721,
+                                        505,
+                                        731
+                                    ],
+                                    "score": 1.0,
+                                    "content": "number of training iterations of ResNet34 on CIFAR-10, an LSTM on WikiText-2, and DRN-D-22",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        }
+                    ],
+                    "index": 15.5
+                }
+            ],
+            "page_idx": 4,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        107,
+                        27,
+                        308,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                107,
+                                26,
+                                308,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        26,
+                                        308,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Under review as a conference paper at ICLR 2019",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        302,
+                        751,
+                        308,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                302,
+                                750,
+                                309,
+                                763
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        302,
+                                        750,
+                                        309,
+                                        763
+                                    ],
+                                    "score": 1.0,
+                                    "content": "5",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "image",
+                    "bbox": [
+                        115,
+                        85,
+                        496,
+                        529
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                115,
+                                85,
+                                496,
+                                529
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        85,
+                                        496,
+                                        529
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                115,
+                                                85,
+                                                496,
+                                                529
+                                            ],
+                                            "score": 0.975,
+                                            "type": "image",
+                                            "image_path": "b45c386a5ec3b1dee617dd3ed5dfe50c26a309dd84ec1efd0f4dcdac283a893c.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 1,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        85,
+                                        496,
+                                        233.0
+                                    ],
+                                    "spans": [],
+                                    "index": 0
+                                },
+                                {
+                                    "bbox": [
+                                        115,
+                                        233.0,
+                                        496,
+                                        381.0
+                                    ],
+                                    "spans": [],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        115,
+                                        381.0,
+                                        496,
+                                        529.0
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                106,
+                                539,
+                                505,
+                                590
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        538,
+                                        505,
+                                        551
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                538,
+                                                505,
+                                                551
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 1: Contour plots of training losses for various problem domains on a log scale. Lighter colors indicate",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 3
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        548,
+                                        505,
+                                        561
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                548,
+                                                505,
+                                                561
+                                            ],
+                                            "score": 1.0,
+                                            "content": "lower loss values. Since we train each batch size for a fixed number of epochs, the total number of training",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 4
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        558,
+                                        505,
+                                        572
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                558,
+                                                505,
+                                                572
+                                            ],
+                                            "score": 1.0,
+                                            "content": "iterations scales down linearly. For each loss value, we can observe how many iterations it takes to converge",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 5
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        568,
+                                        505,
+                                        582
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                568,
+                                                505,
+                                                582
+                                            ],
+                                            "score": 1.0,
+                                            "content": "to that value given a particular batch size, by tracing the level curve for the associated color. For all problems,",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 6
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        579,
+                                        488,
+                                        591
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                579,
+                                                488,
+                                                591
+                                            ],
+                                            "score": 1.0,
+                                            "content": "there is a batch size after which the number of training iterations necessary to converge does not decrease.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 7
+                                }
+                            ],
+                            "index": 5
+                        }
+                    ],
+                    "index": 3.0
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        615,
+                        358,
+                        627
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                615,
+                                359,
+                                628
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        615,
+                                        359,
+                                        628
+                                    ],
+                                    "score": 1.0,
+                                    "content": "4.1 DIMINISHING RETURNS IN RATES OF CONVERGENCE",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        }
+                    ],
+                    "index": 8
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        637,
+                        505,
+                        704
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                637,
+                                505,
+                                651
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        637,
+                                        505,
+                                        651
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We demonstrate the rapidly diminishing returns in rates of convergence across various problem do-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                649,
+                                505,
+                                661
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        649,
+                                        505,
+                                        661
+                                    ],
+                                    "score": 1.0,
+                                    "content": "mains and network configurations. Researchers increase the batch size in an attempt to achieve",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                660,
+                                505,
+                                673
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        660,
+                                        505,
+                                        673
+                                    ],
+                                    "score": 1.0,
+                                    "content": "nearly linear speedups in convergence compared to a small mini-batch size. In particular, if the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                671,
+                                505,
+                                683
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        671,
+                                        218,
+                                        683
+                                    ],
+                                    "score": 1.0,
+                                    "content": "speedup is near-linear, i.e.",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        219,
+                                        671,
+                                        371,
+                                        683
+                                    ],
+                                    "score": 0.91,
+                                    "content": "s ( m ; \\bar { m } _ { 0 } ) = k ( \\bar { m } _ { 0 } ) / k ( m ) \\approx m / m _ { 0 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        371,
+                                        671,
+                                        505,
+                                        683
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", then the computational cost re-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                681,
+                                505,
+                                694
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        682,
+                                        399,
+                                        694
+                                    ],
+                                    "score": 1.0,
+                                    "content": "mains nearly constant for large and small mini-batch SGD. However, if",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        399,
+                                        681,
+                                        466,
+                                        694
+                                    ],
+                                    "score": 0.93,
+                                    "content": "s ( m ) \\ \\bar { \\ll } \\ m / m _ { 0 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        466,
+                                        682,
+                                        505,
+                                        694
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", then the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                692,
+                                322,
+                                705
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        692,
+                                        322,
+                                        705
+                                    ],
+                                    "score": 1.0,
+                                    "content": "benefit of using large batch size training is negligible.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        }
+                    ],
+                    "index": 11.5,
+                    "bbox_fs": [
+                        105,
+                        637,
+                        505,
+                        705
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        709,
+                        504,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                709,
+                                505,
+                                722
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        709,
+                                        505,
+                                        722
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In Figure 1, we show contour plots of training loss as a function of both the batch size and the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                721,
+                                505,
+                                731
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        721,
+                                        505,
+                                        731
+                                    ],
+                                    "score": 1.0,
+                                    "content": "number of training iterations of ResNet34 on CIFAR-10, an LSTM on WikiText-2, and DRN-D-22",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                430,
+                                505,
+                                442
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        430,
+                                        505,
+                                        442
+                                    ],
+                                    "score": 1.0,
+                                    "content": "on Cityscapes. Consider, for example, the contour plot for ResNet34 trained on CIFAR-10. We",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                442,
+                                505,
+                                453
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        442,
+                                        505,
+                                        453
+                                    ],
+                                    "score": 1.0,
+                                    "content": "can see that as the batch size increases from 16 to 2048, the number of SGD iterations needed to",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                453,
+                                505,
+                                465
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        453,
+                                        505,
+                                        465
+                                    ],
+                                    "score": 1.0,
+                                    "content": "achieve a particular loss value decreases linearly. Exceeding this regime, however, the speedup ratio",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                463,
+                                505,
+                                476
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        464,
+                                        312,
+                                        476
+                                    ],
+                                    "score": 1.0,
+                                    "content": "becomes increasingly sublinear and soon we have",
+                                    "type": "text",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        313,
+                                        463,
+                                        397,
+                                        475
+                                    ],
+                                    "score": 0.92,
+                                    "content": "s ( m ; \\bar { m _ { 0 } } ) \\ll \\bar { m } / m _ { 0 }",
+                                    "type": "inline_equation",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        398,
+                                        464,
+                                        505,
+                                        476
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". For batch size 8196, the",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                474,
+                                505,
+                                487
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        474,
+                                        505,
+                                        487
+                                    ],
+                                    "score": 1.0,
+                                    "content": "training procedure does not achieve the lowest training loss, and from this perspective, even if we did",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                485,
+                                505,
+                                497
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        485,
+                                        505,
+                                        497
+                                    ],
+                                    "score": 1.0,
+                                    "content": "not care about computational cost or training time, we would not be able to find an accurate model.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                496,
+                                484,
+                                509
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        496,
+                                        484,
+                                        509
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We observe even worse scaling behavior for test performance (please see Figure 5 for details).",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 18
+                        }
+                    ],
+                    "index": 15.5,
+                    "bbox_fs": [
+                        105,
+                        709,
+                        505,
+                        731
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "table",
+                    "bbox": [
+                        107,
+                        120,
+                        504,
+                        217
+                    ],
+                    "blocks": [
+                        {
+                            "type": "table_caption",
+                            "bbox": [
+                                106,
+                                81,
+                                505,
+                                111
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        80,
+                                        505,
+                                        92
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                80,
+                                                449,
+                                                92
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Table 1: A description of the problem configurations and training strategies used in this paper.",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                449,
+                                                83,
+                                                459,
+                                                92
+                                            ],
+                                            "score": 0.81,
+                                            "content": "\\eta _ { 0 }",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                460,
+                                                80,
+                                                505,
+                                                92
+                                            ],
+                                            "score": 1.0,
+                                            "content": "is the initial",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 0
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        91,
+                                        506,
+                                        102
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                91,
+                                                156,
+                                                102
+                                            ],
+                                            "score": 1.0,
+                                            "content": "learning rate,",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                156,
+                                                91,
+                                                168,
+                                                100
+                                            ],
+                                            "score": 0.33,
+                                            "content": "W",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                168,
+                                                91,
+                                                415,
+                                                102
+                                            ],
+                                            "score": 1.0,
+                                            "content": "is the number of epochs used for warm-up in the linear scaling rule,",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                416,
+                                                91,
+                                                424,
+                                                100
+                                            ],
+                                            "score": 0.82,
+                                            "content": "E",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                424,
+                                                91,
+                                                506,
+                                                102
+                                            ],
+                                            "score": 1.0,
+                                            "content": "is the total number of",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        101,
+                                        162,
+                                        111
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                101,
+                                                162,
+                                                111
+                                            ],
+                                            "score": 1.0,
+                                            "content": "epochs trained",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 2
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "type": "table_body",
+                            "bbox": [
+                                107,
+                                120,
+                                504,
+                                217
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        120,
+                                        504,
+                                        217
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                107,
+                                                120,
+                                                504,
+                                                217
+                                            ],
+                                            "score": 0.977,
+                                            "html": "<table><tr><td>Dataset</td><td>Task</td><td>Architecture</td><td>Training Strategy</td><td>BS range</td></tr><tr><td> MNIST</td><td>IC</td><td>ResNet34</td><td>BLR,LSR (no = 0.1, W = 10, E = 200)</td><td>26 -214</td></tr><tr><td>CIFAR-10</td><td>IC</td><td>AlexNet, MobileNetV2 ResNet34, VGG16</td><td>BLR,LSR, SRSR (mo = 0.1,W=10,E= 200)</td><td>26-214</td></tr><tr><td>CIFAR-100</td><td>IC</td><td>ResNet34</td><td>BLR,LSR (no = 0.1,W = 10, E = 200)</td><td>2 -214</td></tr><tr><td>SVHN</td><td>IC</td><td>ResNet34</td><td>BLR,LSR (no =0.1,W =10,E= 200)</td><td>26-214</td></tr><tr><td> WikiText-2</td><td> NLP</td><td>LSTM</td><td>BLR,LSR (no = 20,W = 3,E = 40)</td><td>2³ -210</td></tr><tr><td>Cityscapes</td><td>IS</td><td>DRN-D-22</td><td>BLR,LSR (no = 0.01,W = 10,E= 100)</td><td>23 -211</td></tr></table>",
+                                            "type": "table",
+                                            "image_path": "6850766aebd7ac2141d3cf5d090844fbc7e2c84a240d31b09a191a647538a886.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 4,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        120,
+                                        504,
+                                        152.33333333333334
+                                    ],
+                                    "spans": [],
+                                    "index": 3
+                                },
+                                {
+                                    "bbox": [
+                                        107,
+                                        152.33333333333334,
+                                        504,
+                                        184.66666666666669
+                                    ],
+                                    "spans": [],
+                                    "index": 4
+                                },
+                                {
+                                    "bbox": [
+                                        107,
+                                        184.66666666666669,
+                                        504,
+                                        217.00000000000003
+                                    ],
+                                    "spans": [],
+                                    "index": 5
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 2.5
+                },
+                {
+                    "type": "image",
+                    "bbox": [
+                        119,
+                        236,
+                        492,
+                        366
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                119,
+                                236,
+                                492,
+                                366
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        119,
+                                        236,
+                                        492,
+                                        366
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                119,
+                                                236,
+                                                492,
+                                                366
+                                            ],
+                                            "score": 0.963,
+                                            "type": "image",
+                                            "image_path": "2f39da181bbafada9ee2bf61d16e3221282ee2f410345f6758d773bf18d978e5.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 7,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        119,
+                                        236,
+                                        492,
+                                        279.3333333333333
+                                    ],
+                                    "spans": [],
+                                    "index": 6
+                                },
+                                {
+                                    "bbox": [
+                                        119,
+                                        279.3333333333333,
+                                        492,
+                                        322.66666666666663
+                                    ],
+                                    "spans": [],
+                                    "index": 7
+                                },
+                                {
+                                    "bbox": [
+                                        119,
+                                        322.66666666666663,
+                                        492,
+                                        365.99999999999994
+                                    ],
+                                    "spans": [],
+                                    "index": 8
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                107,
+                                377,
+                                505,
+                                407
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        376,
+                                        505,
+                                        389
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                376,
+                                                505,
+                                                389
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 2: On the left: speedup curves when applying several popular techniques to avoid the generalization",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 9
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        386,
+                                        506,
+                                        399
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                386,
+                                                506,
+                                                399
+                                            ],
+                                            "score": 1.0,
+                                            "content": "gap. Base LR uses a single learning rate for all batch sizes. On the right: the effect of the linear approximation",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 10
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        397,
+                                        343,
+                                        408
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                397,
+                                                343,
+                                                408
+                                            ],
+                                            "score": 1.0,
+                                            "content": "error on final test accuracy when using the linear LR scaling rule.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 11
+                                }
+                            ],
+                            "index": 10
+                        }
+                    ],
+                    "index": 8.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        430,
+                        505,
+                        507
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                430,
+                                505,
+                                442
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        430,
+                                        505,
+                                        442
+                                    ],
+                                    "score": 1.0,
+                                    "content": "on Cityscapes. Consider, for example, the contour plot for ResNet34 trained on CIFAR-10. We",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                442,
+                                505,
+                                453
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        442,
+                                        505,
+                                        453
+                                    ],
+                                    "score": 1.0,
+                                    "content": "can see that as the batch size increases from 16 to 2048, the number of SGD iterations needed to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                453,
+                                505,
+                                465
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        453,
+                                        505,
+                                        465
+                                    ],
+                                    "score": 1.0,
+                                    "content": "achieve a particular loss value decreases linearly. Exceeding this regime, however, the speedup ratio",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                463,
+                                505,
+                                476
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        464,
+                                        312,
+                                        476
+                                    ],
+                                    "score": 1.0,
+                                    "content": "becomes increasingly sublinear and soon we have",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        313,
+                                        463,
+                                        397,
+                                        475
+                                    ],
+                                    "score": 0.92,
+                                    "content": "s ( m ; \\bar { m _ { 0 } } ) \\ll \\bar { m } / m _ { 0 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        398,
+                                        464,
+                                        505,
+                                        476
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". For batch size 8196, the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                474,
+                                505,
+                                487
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        474,
+                                        505,
+                                        487
+                                    ],
+                                    "score": 1.0,
+                                    "content": "training procedure does not achieve the lowest training loss, and from this perspective, even if we did",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                485,
+                                505,
+                                497
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        485,
+                                        505,
+                                        497
+                                    ],
+                                    "score": 1.0,
+                                    "content": "not care about computational cost or training time, we would not be able to find an accurate model.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                496,
+                                484,
+                                509
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        496,
+                                        484,
+                                        509
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We observe even worse scaling behavior for test performance (please see Figure 5 for details).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        }
+                    ],
+                    "index": 15
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        513,
+                        505,
+                        579
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                513,
+                                505,
+                                525
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        513,
+                                        505,
+                                        525
+                                    ],
+                                    "score": 1.0,
+                                    "content": "For NLP and IS, note that the gain from large batch training diminishes even faster. Neither the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                524,
+                                505,
+                                537
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        524,
+                                        505,
+                                        537
+                                    ],
+                                    "score": 1.0,
+                                    "content": "LSTM on WikiText-2 nor DRN-D-22 on Cityscapes can reach their respective baseline perfor-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                536,
+                                505,
+                                547
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        536,
+                                        505,
+                                        547
+                                    ],
+                                    "score": 1.0,
+                                    "content": "mances after reasonably small batch sizes of about 32 and 64, respectively. Although Puri et al.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                546,
+                                505,
+                                559
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        546,
+                                        505,
+                                        559
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(2018) showed that training on the Amazon Reviews dataset (McAuley et al., 2015) can be done",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                556,
+                                505,
+                                570
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        556,
+                                        505,
+                                        570
+                                    ],
+                                    "score": 1.0,
+                                    "content": "within 4 hours, they tune hyper-parameters heavily. This poses an issue for many practical deploy-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                569,
+                                353,
+                                580
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        569,
+                                        353,
+                                        580
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ments because these problems are often already slow to train.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        }
+                    ],
+                    "index": 21.5
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        595,
+                        404,
+                        606
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                594,
+                                405,
+                                607
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        594,
+                                        405,
+                                        607
+                                    ],
+                                    "score": 1.0,
+                                    "content": "4.2 EXISTING STRATEGIES BREAK DOWN FOR LARGE BATCH SIZES",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        }
+                    ],
+                    "index": 25
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        615,
+                        505,
+                        704
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                615,
+                                505,
+                                630
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        615,
+                                        505,
+                                        630
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We further explore how training with the linear and square root scaling rules compares to training",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                627,
+                                506,
+                                639
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        627,
+                                        506,
+                                        639
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with a fixed baseline learning rate (BLR) that does not change with batch size. In the left subfigure of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                638,
+                                505,
+                                650
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        638,
+                                        505,
+                                        650
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Figure 2, we show the speedup curves of BLR, LSR, and SRSR strategies for ResNet34 on CIFAR-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                649,
+                                505,
+                                660
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        649,
+                                        505,
+                                        660
+                                    ],
+                                    "score": 1.0,
+                                    "content": "10. Note that LSR and SRSR outperform BLR from batch size 256 to 2048 which implies that",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                659,
+                                506,
+                                672
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        659,
+                                        506,
+                                        672
+                                    ],
+                                    "score": 1.0,
+                                    "content": "LSR and SRSR can help the model train for small-to-medium batch sizes. However, the speedup of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                671,
+                                506,
+                                683
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        671,
+                                        506,
+                                        683
+                                    ],
+                                    "score": 1.0,
+                                    "content": "LSR and SRSR is still worse than the ideal linear case, and the curves plateau quickly after a batch",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                682,
+                                505,
+                                693
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        682,
+                                        505,
+                                        693
+                                    ],
+                                    "score": 1.0,
+                                    "content": "size of 2048, at which point BLR becomes better than LSR and SRSR. This means that for certain",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                693,
+                                505,
+                                705
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        693,
+                                        505,
+                                        705
+                                    ],
+                                    "score": 1.0,
+                                    "content": "problems, scaling up the learning rate to compensate for an increased batch size hurts performance.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        }
+                    ],
+                    "index": 29.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        709,
+                        503,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                709,
+                                505,
+                                722
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        709,
+                                        505,
+                                        722
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In the right subfigure of Figure 2, we plot the test performance and the approximation error for LSR",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                720,
+                                505,
+                                733
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        720,
+                                        505,
+                                        733
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of ResNet34 on CIFAR-10. We measure the approximation error at the end of training, with final",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        }
+                    ],
+                    "index": 34.5
+                }
+            ],
+            "page_idx": 5,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        107,
+                        27,
+                        308,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                107,
+                                26,
+                                308,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        26,
+                                        308,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Under review as a conference paper at ICLR 2019",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        302,
+                        752,
+                        308,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                302,
+                                751,
+                                309,
+                                762
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        302,
+                                        751,
+                                        309,
+                                        762
+                                    ],
+                                    "score": 1.0,
+                                    "content": "6",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "table",
+                    "bbox": [
+                        107,
+                        120,
+                        504,
+                        217
+                    ],
+                    "blocks": [
+                        {
+                            "type": "table_caption",
+                            "bbox": [
+                                106,
+                                81,
+                                505,
+                                111
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        80,
+                                        505,
+                                        92
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                80,
+                                                449,
+                                                92
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Table 1: A description of the problem configurations and training strategies used in this paper.",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                449,
+                                                83,
+                                                459,
+                                                92
+                                            ],
+                                            "score": 0.81,
+                                            "content": "\\eta _ { 0 }",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                460,
+                                                80,
+                                                505,
+                                                92
+                                            ],
+                                            "score": 1.0,
+                                            "content": "is the initial",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 0
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        91,
+                                        506,
+                                        102
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                91,
+                                                156,
+                                                102
+                                            ],
+                                            "score": 1.0,
+                                            "content": "learning rate,",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                156,
+                                                91,
+                                                168,
+                                                100
+                                            ],
+                                            "score": 0.33,
+                                            "content": "W",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                168,
+                                                91,
+                                                415,
+                                                102
+                                            ],
+                                            "score": 1.0,
+                                            "content": "is the number of epochs used for warm-up in the linear scaling rule,",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                416,
+                                                91,
+                                                424,
+                                                100
+                                            ],
+                                            "score": 0.82,
+                                            "content": "E",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                424,
+                                                91,
+                                                506,
+                                                102
+                                            ],
+                                            "score": 1.0,
+                                            "content": "is the total number of",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        101,
+                                        162,
+                                        111
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                101,
+                                                162,
+                                                111
+                                            ],
+                                            "score": 1.0,
+                                            "content": "epochs trained",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 2
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "type": "table_body",
+                            "bbox": [
+                                107,
+                                120,
+                                504,
+                                217
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        120,
+                                        504,
+                                        217
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                107,
+                                                120,
+                                                504,
+                                                217
+                                            ],
+                                            "score": 0.977,
+                                            "html": "<table><tr><td>Dataset</td><td>Task</td><td>Architecture</td><td>Training Strategy</td><td>BS range</td></tr><tr><td> MNIST</td><td>IC</td><td>ResNet34</td><td>BLR,LSR (no = 0.1, W = 10, E = 200)</td><td>26 -214</td></tr><tr><td>CIFAR-10</td><td>IC</td><td>AlexNet, MobileNetV2 ResNet34, VGG16</td><td>BLR,LSR, SRSR (mo = 0.1,W=10,E= 200)</td><td>26-214</td></tr><tr><td>CIFAR-100</td><td>IC</td><td>ResNet34</td><td>BLR,LSR (no = 0.1,W = 10, E = 200)</td><td>2 -214</td></tr><tr><td>SVHN</td><td>IC</td><td>ResNet34</td><td>BLR,LSR (no =0.1,W =10,E= 200)</td><td>26-214</td></tr><tr><td> WikiText-2</td><td> NLP</td><td>LSTM</td><td>BLR,LSR (no = 20,W = 3,E = 40)</td><td>2³ -210</td></tr><tr><td>Cityscapes</td><td>IS</td><td>DRN-D-22</td><td>BLR,LSR (no = 0.01,W = 10,E= 100)</td><td>23 -211</td></tr></table>",
+                                            "type": "table",
+                                            "image_path": "6850766aebd7ac2141d3cf5d090844fbc7e2c84a240d31b09a191a647538a886.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 4,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        120,
+                                        504,
+                                        152.33333333333334
+                                    ],
+                                    "spans": [],
+                                    "index": 3
+                                },
+                                {
+                                    "bbox": [
+                                        107,
+                                        152.33333333333334,
+                                        504,
+                                        184.66666666666669
+                                    ],
+                                    "spans": [],
+                                    "index": 4
+                                },
+                                {
+                                    "bbox": [
+                                        107,
+                                        184.66666666666669,
+                                        504,
+                                        217.00000000000003
+                                    ],
+                                    "spans": [],
+                                    "index": 5
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 2.5
+                },
+                {
+                    "type": "image",
+                    "bbox": [
+                        119,
+                        236,
+                        492,
+                        366
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                119,
+                                236,
+                                492,
+                                366
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        119,
+                                        236,
+                                        492,
+                                        366
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                119,
+                                                236,
+                                                492,
+                                                366
+                                            ],
+                                            "score": 0.963,
+                                            "type": "image",
+                                            "image_path": "2f39da181bbafada9ee2bf61d16e3221282ee2f410345f6758d773bf18d978e5.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 7,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        119,
+                                        236,
+                                        492,
+                                        279.3333333333333
+                                    ],
+                                    "spans": [],
+                                    "index": 6
+                                },
+                                {
+                                    "bbox": [
+                                        119,
+                                        279.3333333333333,
+                                        492,
+                                        322.66666666666663
+                                    ],
+                                    "spans": [],
+                                    "index": 7
+                                },
+                                {
+                                    "bbox": [
+                                        119,
+                                        322.66666666666663,
+                                        492,
+                                        365.99999999999994
+                                    ],
+                                    "spans": [],
+                                    "index": 8
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                107,
+                                377,
+                                505,
+                                407
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        376,
+                                        505,
+                                        389
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                376,
+                                                505,
+                                                389
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 2: On the left: speedup curves when applying several popular techniques to avoid the generalization",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 9
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        386,
+                                        506,
+                                        399
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                386,
+                                                506,
+                                                399
+                                            ],
+                                            "score": 1.0,
+                                            "content": "gap. Base LR uses a single learning rate for all batch sizes. On the right: the effect of the linear approximation",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 10
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        397,
+                                        343,
+                                        408
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                397,
+                                                343,
+                                                408
+                                            ],
+                                            "score": 1.0,
+                                            "content": "error on final test accuracy when using the linear LR scaling rule.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 11
+                                }
+                            ],
+                            "index": 10
+                        }
+                    ],
+                    "index": 8.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        430,
+                        505,
+                        507
+                    ],
+                    "lines": [],
+                    "index": 15,
+                    "bbox_fs": [
+                        105,
+                        430,
+                        505,
+                        509
+                    ],
+                    "lines_deleted": true
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        513,
+                        505,
+                        579
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                513,
+                                505,
+                                525
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        513,
+                                        505,
+                                        525
+                                    ],
+                                    "score": 1.0,
+                                    "content": "For NLP and IS, note that the gain from large batch training diminishes even faster. Neither the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                524,
+                                505,
+                                537
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        524,
+                                        505,
+                                        537
+                                    ],
+                                    "score": 1.0,
+                                    "content": "LSTM on WikiText-2 nor DRN-D-22 on Cityscapes can reach their respective baseline perfor-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                536,
+                                505,
+                                547
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        536,
+                                        505,
+                                        547
+                                    ],
+                                    "score": 1.0,
+                                    "content": "mances after reasonably small batch sizes of about 32 and 64, respectively. Although Puri et al.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                546,
+                                505,
+                                559
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        546,
+                                        505,
+                                        559
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(2018) showed that training on the Amazon Reviews dataset (McAuley et al., 2015) can be done",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                556,
+                                505,
+                                570
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        556,
+                                        505,
+                                        570
+                                    ],
+                                    "score": 1.0,
+                                    "content": "within 4 hours, they tune hyper-parameters heavily. This poses an issue for many practical deploy-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                569,
+                                353,
+                                580
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        569,
+                                        353,
+                                        580
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ments because these problems are often already slow to train.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        }
+                    ],
+                    "index": 21.5,
+                    "bbox_fs": [
+                        105,
+                        513,
+                        505,
+                        580
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        595,
+                        404,
+                        606
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                594,
+                                405,
+                                607
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        594,
+                                        405,
+                                        607
+                                    ],
+                                    "score": 1.0,
+                                    "content": "4.2 EXISTING STRATEGIES BREAK DOWN FOR LARGE BATCH SIZES",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        }
+                    ],
+                    "index": 25
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        615,
+                        505,
+                        704
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                615,
+                                505,
+                                630
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        615,
+                                        505,
+                                        630
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We further explore how training with the linear and square root scaling rules compares to training",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                627,
+                                506,
+                                639
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        627,
+                                        506,
+                                        639
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with a fixed baseline learning rate (BLR) that does not change with batch size. In the left subfigure of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                638,
+                                505,
+                                650
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        638,
+                                        505,
+                                        650
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Figure 2, we show the speedup curves of BLR, LSR, and SRSR strategies for ResNet34 on CIFAR-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                649,
+                                505,
+                                660
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        649,
+                                        505,
+                                        660
+                                    ],
+                                    "score": 1.0,
+                                    "content": "10. Note that LSR and SRSR outperform BLR from batch size 256 to 2048 which implies that",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                659,
+                                506,
+                                672
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        659,
+                                        506,
+                                        672
+                                    ],
+                                    "score": 1.0,
+                                    "content": "LSR and SRSR can help the model train for small-to-medium batch sizes. However, the speedup of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                671,
+                                506,
+                                683
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        671,
+                                        506,
+                                        683
+                                    ],
+                                    "score": 1.0,
+                                    "content": "LSR and SRSR is still worse than the ideal linear case, and the curves plateau quickly after a batch",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                682,
+                                505,
+                                693
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        682,
+                                        505,
+                                        693
+                                    ],
+                                    "score": 1.0,
+                                    "content": "size of 2048, at which point BLR becomes better than LSR and SRSR. This means that for certain",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                693,
+                                505,
+                                705
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        693,
+                                        505,
+                                        705
+                                    ],
+                                    "score": 1.0,
+                                    "content": "problems, scaling up the learning rate to compensate for an increased batch size hurts performance.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        }
+                    ],
+                    "index": 29.5,
+                    "bbox_fs": [
+                        105,
+                        615,
+                        506,
+                        705
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        709,
+                        503,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                709,
+                                505,
+                                722
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        709,
+                                        505,
+                                        722
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In the right subfigure of Figure 2, we plot the test performance and the approximation error for LSR",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                720,
+                                505,
+                                733
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        720,
+                                        505,
+                                        733
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of ResNet34 on CIFAR-10. We measure the approximation error at the end of training, with final",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                279,
+                                505,
+                                292
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        279,
+                                        140,
+                                        292
+                                    ],
+                                    "score": 1.0,
+                                    "content": "weights",
+                                    "type": "text",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        140,
+                                        280,
+                                        154,
+                                        290
+                                    ],
+                                    "score": 0.81,
+                                    "content": "\\mathbf { w } ^ { * }",
+                                    "type": "inline_equation",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        154,
+                                        279,
+                                        462,
+                                        292
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". We take this error to be the absolute difference between the true loss value",
+                                    "type": "text",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        462,
+                                        279,
+                                        486,
+                                        291
+                                    ],
+                                    "score": 0.9,
+                                    "content": "L ( \\mathbf { w } )",
+                                    "type": "inline_equation",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        487,
+                                        279,
+                                        505,
+                                        292
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                291,
+                                505,
+                                305
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        291,
+                                        215,
+                                        305
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the linear approximation at",
+                                    "type": "text",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        216,
+                                        293,
+                                        229,
+                                        303
+                                    ],
+                                    "score": 0.82,
+                                    "content": "\\mathbf { w } ^ { * }",
+                                    "type": "inline_equation",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        230,
+                                        291,
+                                        269,
+                                        305
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", given by",
+                                    "type": "text",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        270,
+                                        291,
+                                        423,
+                                        304
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\hat { L } ( \\mathbf { w } ) = L ( \\mathbf { w } ^ { * } ) + \\langle \\mathbf { g } _ { m } ( \\mathbf { w } ^ { * } ) , \\mathbf { w } - \\mathbf { w } ^ { * } \\rangle",
+                                    "type": "inline_equation",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        423,
+                                        291,
+                                        505,
+                                        305
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". The approximation",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                101,
+                                298,
+                                509,
+                                323
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        101,
+                                        298,
+                                        172,
+                                        323
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is calculated for",
+                                    "type": "text",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        172,
+                                        303,
+                                        271,
+                                        317
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\mathbf { w } = \\mathbf { w } ^ { * } - \\eta \\frac { m } { m _ { 0 } } \\mathbf { g } _ { m } \\big ( \\mathbf { w } ^ { * } \\big )",
+                                    "type": "inline_equation",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        272,
+                                        298,
+                                        509,
+                                        323
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to understand the behavior of the approximation along the",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                315,
+                                505,
+                                329
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        315,
+                                        505,
+                                        329
+                                    ],
+                                    "score": 1.0,
+                                    "content": "trajectory for a single SGD iterate using the LSR. It appears that there exists a strong relationship",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                326,
+                                506,
+                                339
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        326,
+                                        506,
+                                        339
+                                    ],
+                                    "score": 1.0,
+                                    "content": "between linear approximation error and test accuracy: as the linear approximation error increases,",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                338,
+                                506,
+                                350
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        338,
+                                        506,
+                                        350
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the test accuracy drops. Note the transition that happens at the critical batch size of 2048. After this",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                348,
+                                506,
+                                362
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        348,
+                                        506,
+                                        362
+                                    ],
+                                    "score": 1.0,
+                                    "content": "point, the test accuracy drops significantly and the linear approximation error exceeds 1, showing",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                359,
+                                403,
+                                372
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        359,
+                                        403,
+                                        372
+                                    ],
+                                    "score": 1.0,
+                                    "content": "that we quickly exit the regime in which the linear approximation is valid.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 14
+                        }
+                    ],
+                    "index": 34.5,
+                    "bbox_fs": [
+                        106,
+                        709,
+                        505,
+                        733
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "image",
+                    "bbox": [
+                        145,
+                        90,
+                        465,
+                        205
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                145,
+                                90,
+                                465,
+                                205
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        145,
+                                        90,
+                                        465,
+                                        205
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                145,
+                                                90,
+                                                465,
+                                                205
+                                            ],
+                                            "score": 0.958,
+                                            "type": "image",
+                                            "image_path": "8e5a9c2868609d6811e018f9aa938f74636a8e7c872b18be0773d9a10cb99683.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 1,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        145,
+                                        90,
+                                        465,
+                                        128.33333333333334
+                                    ],
+                                    "spans": [],
+                                    "index": 0
+                                },
+                                {
+                                    "bbox": [
+                                        145,
+                                        128.33333333333334,
+                                        465,
+                                        166.66666666666669
+                                    ],
+                                    "spans": [],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        145,
+                                        166.66666666666669,
+                                        465,
+                                        205.00000000000003
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                106,
+                                216,
+                                505,
+                                256
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        215,
+                                        505,
+                                        228
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                215,
+                                                505,
+                                                228
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 3: Speedup curves across different problem configurations. Left: different architectures result in differ-",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 3
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        226,
+                                        505,
+                                        238
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                226,
+                                                505,
+                                                238
+                                            ],
+                                            "score": 1.0,
+                                            "content": "ent rates of convergence on CIFAR-10. Right: ResNet34 exhibits different rates of convergence on CIFAR-10,",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 4
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        235,
+                                        505,
+                                        248
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                235,
+                                                505,
+                                                248
+                                            ],
+                                            "score": 1.0,
+                                            "content": "CIFAR-100, and SVHN. Loss thresholds are obtained by computing the lower quartile of loss values achieved",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 5
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        246,
+                                        198,
+                                        258
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                246,
+                                                198,
+                                                258
+                                            ],
+                                            "score": 1.0,
+                                            "content": "by the largest batch size.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 6
+                                }
+                            ],
+                            "index": 4.5
+                        }
+                    ],
+                    "index": 2.75
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        279,
+                        505,
+                        371
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                279,
+                                505,
+                                292
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        279,
+                                        140,
+                                        292
+                                    ],
+                                    "score": 1.0,
+                                    "content": "weights",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        140,
+                                        280,
+                                        154,
+                                        290
+                                    ],
+                                    "score": 0.81,
+                                    "content": "\\mathbf { w } ^ { * }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        154,
+                                        279,
+                                        462,
+                                        292
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". We take this error to be the absolute difference between the true loss value",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        462,
+                                        279,
+                                        486,
+                                        291
+                                    ],
+                                    "score": 0.9,
+                                    "content": "L ( \\mathbf { w } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        487,
+                                        279,
+                                        505,
+                                        292
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                291,
+                                505,
+                                305
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        291,
+                                        215,
+                                        305
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the linear approximation at",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        216,
+                                        293,
+                                        229,
+                                        303
+                                    ],
+                                    "score": 0.82,
+                                    "content": "\\mathbf { w } ^ { * }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        230,
+                                        291,
+                                        269,
+                                        305
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", given by",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        270,
+                                        291,
+                                        423,
+                                        304
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\hat { L } ( \\mathbf { w } ) = L ( \\mathbf { w } ^ { * } ) + \\langle \\mathbf { g } _ { m } ( \\mathbf { w } ^ { * } ) , \\mathbf { w } - \\mathbf { w } ^ { * } \\rangle",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        423,
+                                        291,
+                                        505,
+                                        305
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". The approximation",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                101,
+                                298,
+                                509,
+                                323
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        101,
+                                        298,
+                                        172,
+                                        323
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is calculated for",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        172,
+                                        303,
+                                        271,
+                                        317
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\mathbf { w } = \\mathbf { w } ^ { * } - \\eta \\frac { m } { m _ { 0 } } \\mathbf { g } _ { m } \\big ( \\mathbf { w } ^ { * } \\big )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        272,
+                                        298,
+                                        509,
+                                        323
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to understand the behavior of the approximation along the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                315,
+                                505,
+                                329
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        315,
+                                        505,
+                                        329
+                                    ],
+                                    "score": 1.0,
+                                    "content": "trajectory for a single SGD iterate using the LSR. It appears that there exists a strong relationship",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                326,
+                                506,
+                                339
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        326,
+                                        506,
+                                        339
+                                    ],
+                                    "score": 1.0,
+                                    "content": "between linear approximation error and test accuracy: as the linear approximation error increases,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                338,
+                                506,
+                                350
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        338,
+                                        506,
+                                        350
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the test accuracy drops. Note the transition that happens at the critical batch size of 2048. After this",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                348,
+                                506,
+                                362
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        348,
+                                        506,
+                                        362
+                                    ],
+                                    "score": 1.0,
+                                    "content": "point, the test accuracy drops significantly and the linear approximation error exceeds 1, showing",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                359,
+                                403,
+                                372
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        359,
+                                        403,
+                                        372
+                                    ],
+                                    "score": 1.0,
+                                    "content": "that we quickly exit the regime in which the linear approximation is valid.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        }
+                    ],
+                    "index": 10.5
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        386,
+                        420,
+                        397
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                384,
+                                423,
+                                398
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        384,
+                                        423,
+                                        398
+                                    ],
+                                    "score": 1.0,
+                                    "content": "4.3 CONVERGENCE SPEED HAS A WEAK DEPENDENCE ON DATASET SIZE",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        }
+                    ],
+                    "index": 15
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        407,
+                        505,
+                        483
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                408,
+                                505,
+                                419
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        408,
+                                        505,
+                                        419
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Previous works have conjectured that the maximum batch size that can result in a good model is",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                417,
+                                506,
+                                430
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        417,
+                                        506,
+                                        430
+                                    ],
+                                    "score": 1.0,
+                                    "content": "proportional to the size of the whole dataset (Smith et al., 2017; Smith & Le, 2018). However, for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                429,
+                                505,
+                                441
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        429,
+                                        505,
+                                        441
+                                    ],
+                                    "score": 1.0,
+                                    "content": "convex, over-parameterized problems, Ma et al. (2017) show that there is a model-dependent critical",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                439,
+                                505,
+                                452
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        439,
+                                        505,
+                                        452
+                                    ],
+                                    "score": 1.0,
+                                    "content": "batch size after which we observe rapidly diminishing returns in convergence speed. In this section,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                449,
+                                505,
+                                464
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        449,
+                                        505,
+                                        464
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to observe if a similar critical batch size exists in the non-convex case, we compare how changing",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                461,
+                                505,
+                                475
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        461,
+                                        505,
+                                        475
+                                    ],
+                                    "score": 1.0,
+                                    "content": "model architecture or data complexity affects the shapes of speedup curves compared to changing",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                472,
+                                196,
+                                484
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        472,
+                                        196,
+                                        484
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the dataset size alone.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        }
+                    ],
+                    "index": 19
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        489,
+                        505,
+                        621
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                489,
+                                505,
+                                502
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        489,
+                                        505,
+                                        502
+                                    ],
+                                    "score": 1.0,
+                                    "content": "First, in order to show that these diminishing returns depend on data complexity and DNN architec-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                501,
+                                505,
+                                512
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        501,
+                                        505,
+                                        512
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ture, we plot speedup curves in Figure 3 to compare the scaling behaviors across different models",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                511,
+                                505,
+                                524
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        511,
+                                        505,
+                                        524
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and dataset configurations. For the error threshold \u000f, we chose the lowest quartile loss value reached",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                523,
+                                505,
+                                535
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        523,
+                                        505,
+                                        535
+                                    ],
+                                    "score": 1.0,
+                                    "content": "by the largest batch size to make a fair comparison across configurations. This setup actually favors",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                533,
+                                505,
+                                545
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        533,
+                                        505,
+                                        545
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the large batch case, because there are lower loss thresholds that are attainable only in the small batch",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                544,
+                                505,
+                                556
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        544,
+                                        505,
+                                        556
+                                    ],
+                                    "score": 1.0,
+                                    "content": "case. On the left, for the CIFAR-10 dataset, we compared four model architectures. For each archi-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                555,
+                                505,
+                                568
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        555,
+                                        505,
+                                        568
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tecture, we plotted the speedup curve obtained by training this model on the dataset for various batch",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                567,
+                                505,
+                                579
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        567,
+                                        505,
+                                        579
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sizes. The variety of speedup curve shapes indicates that model architecture is an important factor in",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                577,
+                                506,
+                                590
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        577,
+                                        506,
+                                        590
+                                    ],
+                                    "score": 1.0,
+                                    "content": "determining the convergence speed of training for large batch sizes. For MobileNetV2/AlexNet, the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                588,
+                                505,
+                                600
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        588,
+                                        505,
+                                        600
+                                    ],
+                                    "score": 1.0,
+                                    "content": "diminishing returns become visible when batch size is 1024. However, for VGG16/ResNet34, the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                599,
+                                505,
+                                612
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        599,
+                                        505,
+                                        612
+                                    ],
+                                    "score": 1.0,
+                                    "content": "speedup does not flatten out until batch size 8196. Hence, in practice, the choice of model strongly",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                610,
+                                308,
+                                622
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        610,
+                                        308,
+                                        622
+                                    ],
+                                    "score": 1.0,
+                                    "content": "affects our ability to use large batch sizes in SGD.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        }
+                    ],
+                    "index": 28.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        627,
+                        505,
+                        693
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                626,
+                                505,
+                                640
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        626,
+                                        505,
+                                        640
+                                    ],
+                                    "score": 1.0,
+                                    "content": "On the right, in order to investigate the effect of problem complexity, we compared the performance",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                638,
+                                505,
+                                650
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        638,
+                                        505,
+                                        650
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of ResNet34 on four datasets of the same size: CIFAR-10, CIFAR-100, MNIST, and the SVHN",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                648,
+                                505,
+                                662
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        648,
+                                        283,
+                                        662
+                                    ],
+                                    "score": 1.0,
+                                    "content": "dataset (we cut off MNIST and SVHN to",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        284,
+                                        649,
+                                        300,
+                                        659
+                                    ],
+                                    "score": 0.85,
+                                    "content": "5 0 k",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        301,
+                                        648,
+                                        505,
+                                        662
+                                    ],
+                                    "score": 1.0,
+                                    "content": "training examples each). Although all problems",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                659,
+                                505,
+                                673
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        659,
+                                        505,
+                                        673
+                                    ],
+                                    "score": 1.0,
+                                    "content": "display diminishing returns in rates of convergence, the point at which the curves plateau varies",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                671,
+                                505,
+                                683
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        671,
+                                        505,
+                                        683
+                                    ],
+                                    "score": 1.0,
+                                    "content": "according to problem complexity. It is not hard to see that, for simpler problems such as SVHN, the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                682,
+                                388,
+                                694
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        682,
+                                        388,
+                                        694
+                                    ],
+                                    "score": 1.0,
+                                    "content": "curves flatten out later than for harder problems (e.g. CIFAR-10/100).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        }
+                    ],
+                    "index": 37.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        699,
+                        504,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                698,
+                                505,
+                                711
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        698,
+                                        505,
+                                        711
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In all of the above cases, the diminishing rates of return in convergence speed become visible after",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                710,
+                                505,
+                                722
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        710,
+                                        505,
+                                        722
+                                    ],
+                                    "score": 1.0,
+                                    "content": "only moderate increases in the batch size. Previous works have only studied convergence behavior",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                720,
+                                506,
+                                734
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        720,
+                                        506,
+                                        734
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for a fairly limited range of batch sizes (e.g., up to 4096 for CIFAR-10) (Hoffer et al., 2017; Keskar",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        }
+                    ],
+                    "index": 42
+                }
+            ],
+            "page_idx": 6,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        107,
+                        27,
+                        308,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                107,
+                                26,
+                                308,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        26,
+                                        308,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Under review as a conference paper at ICLR 2019",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        303,
+                        751,
+                        309,
+                        759
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                302,
+                                750,
+                                309,
+                                762
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        302,
+                                        750,
+                                        309,
+                                        762
+                                    ],
+                                    "score": 1.0,
+                                    "content": "7",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "image",
+                    "bbox": [
+                        145,
+                        90,
+                        465,
+                        205
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                145,
+                                90,
+                                465,
+                                205
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        145,
+                                        90,
+                                        465,
+                                        205
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                145,
+                                                90,
+                                                465,
+                                                205
+                                            ],
+                                            "score": 0.958,
+                                            "type": "image",
+                                            "image_path": "8e5a9c2868609d6811e018f9aa938f74636a8e7c872b18be0773d9a10cb99683.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 1,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        145,
+                                        90,
+                                        465,
+                                        128.33333333333334
+                                    ],
+                                    "spans": [],
+                                    "index": 0
+                                },
+                                {
+                                    "bbox": [
+                                        145,
+                                        128.33333333333334,
+                                        465,
+                                        166.66666666666669
+                                    ],
+                                    "spans": [],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        145,
+                                        166.66666666666669,
+                                        465,
+                                        205.00000000000003
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                106,
+                                216,
+                                505,
+                                256
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        215,
+                                        505,
+                                        228
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                215,
+                                                505,
+                                                228
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 3: Speedup curves across different problem configurations. Left: different architectures result in differ-",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 3
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        226,
+                                        505,
+                                        238
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                226,
+                                                505,
+                                                238
+                                            ],
+                                            "score": 1.0,
+                                            "content": "ent rates of convergence on CIFAR-10. Right: ResNet34 exhibits different rates of convergence on CIFAR-10,",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 4
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        235,
+                                        505,
+                                        248
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                235,
+                                                505,
+                                                248
+                                            ],
+                                            "score": 1.0,
+                                            "content": "CIFAR-100, and SVHN. Loss thresholds are obtained by computing the lower quartile of loss values achieved",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 5
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        246,
+                                        198,
+                                        258
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                246,
+                                                198,
+                                                258
+                                            ],
+                                            "score": 1.0,
+                                            "content": "by the largest batch size.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 6
+                                }
+                            ],
+                            "index": 4.5
+                        }
+                    ],
+                    "index": 2.75
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        279,
+                        505,
+                        371
+                    ],
+                    "lines": [],
+                    "index": 10.5,
+                    "bbox_fs": [
+                        101,
+                        279,
+                        509,
+                        372
+                    ],
+                    "lines_deleted": true
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        386,
+                        420,
+                        397
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                384,
+                                423,
+                                398
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        384,
+                                        423,
+                                        398
+                                    ],
+                                    "score": 1.0,
+                                    "content": "4.3 CONVERGENCE SPEED HAS A WEAK DEPENDENCE ON DATASET SIZE",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        }
+                    ],
+                    "index": 15
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        407,
+                        505,
+                        483
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                408,
+                                505,
+                                419
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        408,
+                                        505,
+                                        419
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Previous works have conjectured that the maximum batch size that can result in a good model is",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                417,
+                                506,
+                                430
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        417,
+                                        506,
+                                        430
+                                    ],
+                                    "score": 1.0,
+                                    "content": "proportional to the size of the whole dataset (Smith et al., 2017; Smith & Le, 2018). However, for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                429,
+                                505,
+                                441
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        429,
+                                        505,
+                                        441
+                                    ],
+                                    "score": 1.0,
+                                    "content": "convex, over-parameterized problems, Ma et al. (2017) show that there is a model-dependent critical",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                439,
+                                505,
+                                452
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        439,
+                                        505,
+                                        452
+                                    ],
+                                    "score": 1.0,
+                                    "content": "batch size after which we observe rapidly diminishing returns in convergence speed. In this section,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                449,
+                                505,
+                                464
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        449,
+                                        505,
+                                        464
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to observe if a similar critical batch size exists in the non-convex case, we compare how changing",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                461,
+                                505,
+                                475
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        461,
+                                        505,
+                                        475
+                                    ],
+                                    "score": 1.0,
+                                    "content": "model architecture or data complexity affects the shapes of speedup curves compared to changing",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                472,
+                                196,
+                                484
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        472,
+                                        196,
+                                        484
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the dataset size alone.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        }
+                    ],
+                    "index": 19,
+                    "bbox_fs": [
+                        104,
+                        408,
+                        506,
+                        484
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        489,
+                        505,
+                        621
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                489,
+                                505,
+                                502
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        489,
+                                        505,
+                                        502
+                                    ],
+                                    "score": 1.0,
+                                    "content": "First, in order to show that these diminishing returns depend on data complexity and DNN architec-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                501,
+                                505,
+                                512
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        501,
+                                        505,
+                                        512
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ture, we plot speedup curves in Figure 3 to compare the scaling behaviors across different models",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                511,
+                                505,
+                                524
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        511,
+                                        505,
+                                        524
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and dataset configurations. For the error threshold \u000f, we chose the lowest quartile loss value reached",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                523,
+                                505,
+                                535
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        523,
+                                        505,
+                                        535
+                                    ],
+                                    "score": 1.0,
+                                    "content": "by the largest batch size to make a fair comparison across configurations. This setup actually favors",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                533,
+                                505,
+                                545
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        533,
+                                        505,
+                                        545
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the large batch case, because there are lower loss thresholds that are attainable only in the small batch",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                544,
+                                505,
+                                556
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        544,
+                                        505,
+                                        556
+                                    ],
+                                    "score": 1.0,
+                                    "content": "case. On the left, for the CIFAR-10 dataset, we compared four model architectures. For each archi-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                555,
+                                505,
+                                568
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        555,
+                                        505,
+                                        568
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tecture, we plotted the speedup curve obtained by training this model on the dataset for various batch",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                567,
+                                505,
+                                579
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        567,
+                                        505,
+                                        579
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sizes. The variety of speedup curve shapes indicates that model architecture is an important factor in",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                577,
+                                506,
+                                590
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        577,
+                                        506,
+                                        590
+                                    ],
+                                    "score": 1.0,
+                                    "content": "determining the convergence speed of training for large batch sizes. For MobileNetV2/AlexNet, the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                588,
+                                505,
+                                600
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        588,
+                                        505,
+                                        600
+                                    ],
+                                    "score": 1.0,
+                                    "content": "diminishing returns become visible when batch size is 1024. However, for VGG16/ResNet34, the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                599,
+                                505,
+                                612
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        599,
+                                        505,
+                                        612
+                                    ],
+                                    "score": 1.0,
+                                    "content": "speedup does not flatten out until batch size 8196. Hence, in practice, the choice of model strongly",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                610,
+                                308,
+                                622
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        610,
+                                        308,
+                                        622
+                                    ],
+                                    "score": 1.0,
+                                    "content": "affects our ability to use large batch sizes in SGD.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        }
+                    ],
+                    "index": 28.5,
+                    "bbox_fs": [
+                        105,
+                        489,
+                        506,
+                        622
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        627,
+                        505,
+                        693
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                626,
+                                505,
+                                640
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        626,
+                                        505,
+                                        640
+                                    ],
+                                    "score": 1.0,
+                                    "content": "On the right, in order to investigate the effect of problem complexity, we compared the performance",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                638,
+                                505,
+                                650
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        638,
+                                        505,
+                                        650
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of ResNet34 on four datasets of the same size: CIFAR-10, CIFAR-100, MNIST, and the SVHN",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                648,
+                                505,
+                                662
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        648,
+                                        283,
+                                        662
+                                    ],
+                                    "score": 1.0,
+                                    "content": "dataset (we cut off MNIST and SVHN to",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        284,
+                                        649,
+                                        300,
+                                        659
+                                    ],
+                                    "score": 0.85,
+                                    "content": "5 0 k",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        301,
+                                        648,
+                                        505,
+                                        662
+                                    ],
+                                    "score": 1.0,
+                                    "content": "training examples each). Although all problems",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                659,
+                                505,
+                                673
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        659,
+                                        505,
+                                        673
+                                    ],
+                                    "score": 1.0,
+                                    "content": "display diminishing returns in rates of convergence, the point at which the curves plateau varies",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                671,
+                                505,
+                                683
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        671,
+                                        505,
+                                        683
+                                    ],
+                                    "score": 1.0,
+                                    "content": "according to problem complexity. It is not hard to see that, for simpler problems such as SVHN, the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                682,
+                                388,
+                                694
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        682,
+                                        388,
+                                        694
+                                    ],
+                                    "score": 1.0,
+                                    "content": "curves flatten out later than for harder problems (e.g. CIFAR-10/100).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        }
+                    ],
+                    "index": 37.5,
+                    "bbox_fs": [
+                        105,
+                        626,
+                        505,
+                        694
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        699,
+                        504,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                698,
+                                505,
+                                711
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        698,
+                                        505,
+                                        711
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In all of the above cases, the diminishing rates of return in convergence speed become visible after",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                710,
+                                505,
+                                722
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        710,
+                                        505,
+                                        722
+                                    ],
+                                    "score": 1.0,
+                                    "content": "only moderate increases in the batch size. Previous works have only studied convergence behavior",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                720,
+                                506,
+                                734
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        720,
+                                        506,
+                                        734
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for a fairly limited range of batch sizes (e.g., up to 4096 for CIFAR-10) (Hoffer et al., 2017; Keskar",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                261,
+                                504,
+                                274
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        261,
+                                        504,
+                                        274
+                                    ],
+                                    "score": 1.0,
+                                    "content": "et al., 2016). By increasing the batch size past this point, it becomes immediately apparent that the",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                272,
+                                479,
+                                288
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        272,
+                                        479,
+                                        288
+                                    ],
+                                    "score": 1.0,
+                                    "content": "primary issue with large batch size optimization is training speed, not the generalization gap.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 6
+                        }
+                    ],
+                    "index": 42,
+                    "bbox_fs": [
+                        105,
+                        698,
+                        506,
+                        734
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "image",
+                    "bbox": [
+                        154,
+                        90,
+                        456,
+                        207
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                154,
+                                90,
+                                456,
+                                207
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        154,
+                                        90,
+                                        456,
+                                        207
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                154,
+                                                90,
+                                                456,
+                                                207
+                                            ],
+                                            "score": 0.908,
+                                            "type": "image",
+                                            "image_path": "96bab39fb3c982480548f52bd526a04980c2093b687c837893866a29eef23e63.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 1,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        154,
+                                        90,
+                                        456,
+                                        129.0
+                                    ],
+                                    "spans": [],
+                                    "index": 0
+                                },
+                                {
+                                    "bbox": [
+                                        154,
+                                        129.0,
+                                        456,
+                                        168.0
+                                    ],
+                                    "spans": [],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        154,
+                                        168.0,
+                                        456,
+                                        207.0
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                106,
+                                218,
+                                505,
+                                240
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        217,
+                                        505,
+                                        230
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                217,
+                                                505,
+                                                230
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 4: Speedup curves as dataset size varies for different datasets. Even as dataset size increases",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 3
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        228,
+                                        472,
+                                        241
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                228,
+                                                211,
+                                                241
+                                            ],
+                                            "score": 1.0,
+                                            "content": "back up to the baseline of",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                212,
+                                                228,
+                                                236,
+                                                239
+                                            ],
+                                            "score": 0.89,
+                                            "content": "1 0 0 \\%",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                236,
+                                                228,
+                                                472,
+                                                241
+                                            ],
+                                            "score": 1.0,
+                                            "content": ", there is no noticeable improvement in convergence speed.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 4
+                                }
+                            ],
+                            "index": 3.5
+                        }
+                    ],
+                    "index": 2.25
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        262,
+                        504,
+                        284
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                261,
+                                504,
+                                274
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        261,
+                                        504,
+                                        274
+                                    ],
+                                    "score": 1.0,
+                                    "content": "et al., 2016). By increasing the batch size past this point, it becomes immediately apparent that the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                272,
+                                479,
+                                288
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        272,
+                                        479,
+                                        288
+                                    ],
+                                    "score": 1.0,
+                                    "content": "primary issue with large batch size optimization is training speed, not the generalization gap.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        }
+                    ],
+                    "index": 5.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        289,
+                        505,
+                        345
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                289,
+                                505,
+                                302
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        289,
+                                        308,
+                                        302
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In order to test whether the sublinear behavior of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        309,
+                                        290,
+                                        348,
+                                        302
+                                    ],
+                                    "score": 0.93,
+                                    "content": "s ( m ; m _ { 0 } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        348,
+                                        289,
+                                        505,
+                                        302
+                                    ],
+                                    "score": 1.0,
+                                    "content": "depends primarily on dataset size, we",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                301,
+                                505,
+                                312
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        301,
+                                        505,
+                                        312
+                                    ],
+                                    "score": 1.0,
+                                    "content": "compare the speedup curves obtained when training a single model on different fractions of the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                311,
+                                505,
+                                324
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        311,
+                                        505,
+                                        324
+                                    ],
+                                    "score": 1.0,
+                                    "content": "original training data. We trained ResNet34 models on the CIFAR-10 and SVHN datasets (for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                321,
+                                506,
+                                336
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        321,
+                                        280,
+                                        336
+                                    ],
+                                    "score": 1.0,
+                                    "content": "SVHN in this experiment, we train on all",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        281,
+                                        323,
+                                        303,
+                                        333
+                                    ],
+                                    "score": 0.85,
+                                    "content": "6 0 0 k",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        303,
+                                        321,
+                                        506,
+                                        336
+                                    ],
+                                    "score": 1.0,
+                                    "content": "available training images). For each dataset, we",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                333,
+                                376,
+                                347
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        333,
+                                        149,
+                                        347
+                                    ],
+                                    "score": 1.0,
+                                    "content": "trained on",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        149,
+                                        334,
+                                        173,
+                                        345
+                                    ],
+                                    "score": 0.86,
+                                    "content": "1 0 0 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        174,
+                                        333,
+                                        177,
+                                        347
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        177,
+                                        334,
+                                        197,
+                                        345
+                                    ],
+                                    "score": 0.86,
+                                    "content": "5 0 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        197,
+                                        333,
+                                        237,
+                                        347
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", and then",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        237,
+                                        334,
+                                        257,
+                                        344
+                                    ],
+                                    "score": 0.88,
+                                    "content": "2 5 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        257,
+                                        333,
+                                        376,
+                                        347
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of the available training data.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        }
+                    ],
+                    "index": 9
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        351,
+                        505,
+                        439
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                350,
+                                506,
+                                363
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        350,
+                                        506,
+                                        363
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In Figure 4, we plot the resulting speedup curves for the various partitions. In order to maintain",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                362,
+                                505,
+                                374
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        362,
+                                        505,
+                                        374
+                                    ],
+                                    "score": 1.0,
+                                    "content": "a fair comparison (as baseline loss values change for different dataset sizes), we again choose the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                371,
+                                505,
+                                385
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        371,
+                                        505,
+                                        385
+                                    ],
+                                    "score": 1.0,
+                                    "content": "loss threshold to be the lower quartile of loss values obtained by the largest batch size.1 Notably,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                384,
+                                505,
+                                396
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        384,
+                                        505,
+                                        396
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the batch size at which the curves begin to plateau remains constant as dataset size changes. For",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                394,
+                                505,
+                                407
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        394,
+                                        505,
+                                        407
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ResNet34 on CIFAR-10, the linear speedup behavior breaks around batch size 128 for all three",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                405,
+                                505,
+                                417
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        405,
+                                        505,
+                                        417
+                                    ],
+                                    "score": 1.0,
+                                    "content": "curves. By a batch size of 1024, all curves have flattened. We can see similar behavior for ResNet34",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                417,
+                                506,
+                                429
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        417,
+                                        506,
+                                        429
+                                    ],
+                                    "score": 1.0,
+                                    "content": "on SVHN. Overall, looking back to Figure 3, the choice of model and the complexity of the dataset",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                428,
+                                425,
+                                439
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        428,
+                                        425,
+                                        439
+                                    ],
+                                    "score": 1.0,
+                                    "content": "appear to be more related to the shape of speedup curve than dataset size alone.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        }
+                    ],
+                    "index": 15.5
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        455,
+                        195,
+                        467
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                453,
+                                197,
+                                470
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        453,
+                                        197,
+                                        470
+                                    ],
+                                    "score": 1.0,
+                                    "content": "5 CONCLUSION",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        }
+                    ],
+                    "index": 20
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        479,
+                        505,
+                        546
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                480,
+                                505,
+                                492
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        480,
+                                        505,
+                                        492
+                                    ],
+                                    "score": 1.0,
+                                    "content": "By experimenting across a wide range of network architectures and problem domains, we find that,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                491,
+                                505,
+                                502
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        491,
+                                        505,
+                                        502
+                                    ],
+                                    "score": 1.0,
+                                    "content": "after a certain point, increasing the batch size fails to decrease wall-clock time to convergence and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                502,
+                                505,
+                                515
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        502,
+                                        505,
+                                        515
+                                    ],
+                                    "score": 1.0,
+                                    "content": "results in low computational efficiency, even assuming perfect parallelism. The critical batch size",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                512,
+                                505,
+                                525
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        512,
+                                        505,
+                                        525
+                                    ],
+                                    "score": 1.0,
+                                    "content": "after which these returns diminish tends to be small relative to existing system capabilities. These",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                523,
+                                506,
+                                537
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        523,
+                                        506,
+                                        537
+                                    ],
+                                    "score": 1.0,
+                                    "content": "trends present impediments to progress in developing effective machine learning systems that are",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                535,
+                                282,
+                                547
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        535,
+                                        282,
+                                        547
+                                    ],
+                                    "score": 1.0,
+                                    "content": "capable of handling growing data demands.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        }
+                    ],
+                    "index": 23.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        551,
+                        505,
+                        596
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                551,
+                                505,
+                                564
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        551,
+                                        505,
+                                        564
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Recent works also suggest heuristics to decrease the generalization gap, but we find that these heuris-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                563,
+                                505,
+                                576
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        563,
+                                        505,
+                                        576
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tics cannot be used to solve the underlying issue of training convergence speed. Moreover, we find",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                573,
+                                505,
+                                586
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        573,
+                                        505,
+                                        586
+                                    ],
+                                    "score": 1.0,
+                                    "content": "that they usually only help decrease the generalization error in a small-to-medium batch size regime.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                583,
+                                505,
+                                597
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        583,
+                                        505,
+                                        597
+                                    ],
+                                    "score": 1.0,
+                                    "content": "There does not seem to be a simple training heuristic to improve large batch performance in general.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        }
+                    ],
+                    "index": 28.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        601,
+                        504,
+                        646
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                600,
+                                506,
+                                614
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        600,
+                                        506,
+                                        614
+                                    ],
+                                    "score": 1.0,
+                                    "content": "These results suggest that we should not assume that increasing the batch size for larger datasets will",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                613,
+                                505,
+                                624
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        613,
+                                        505,
+                                        624
+                                    ],
+                                    "score": 1.0,
+                                    "content": "keep training times manageable for all problems. Even though it is a natural form of data parallelism",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                623,
+                                506,
+                                636
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        623,
+                                        506,
+                                        636
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for large-scale optimization, alternative forms of parallelism should be explored to utilize all of our",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                634,
+                                192,
+                                647
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        634,
+                                        192,
+                                        647
+                                    ],
+                                    "score": 1.0,
+                                    "content": "data more efficiently.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        }
+                    ],
+                    "index": 32.5
+                }
+            ],
+            "page_idx": 7,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        107,
+                        701,
+                        505,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                118,
+                                699,
+                                505,
+                                713
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        118,
+                                        699,
+                                        505,
+                                        713
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1We observed that the loss threshold for a smaller partition is higher than that of the full dataset. This may",
+                                    "type": "text"
+                                }
+                            ]
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                711,
+                                506,
+                                723
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        711,
+                                        506,
+                                        723
+                                    ],
+                                    "score": 1.0,
+                                    "content": "be because, as we decrease dataset size, the large batch behavior that determines our threshold approaches that",
+                                    "type": "text"
+                                }
+                            ]
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                721,
+                                486,
+                                733
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        721,
+                                        486,
+                                        733
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of vanilla gradient descent, which typically displays poor training convergence speed for DNN problems.",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        302,
+                        752,
+                        308,
+                        759
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                300,
+                                750,
+                                309,
+                                761
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        300,
+                                        750,
+                                        309,
+                                        761
+                                    ],
+                                    "score": 1.0,
+                                    "content": "",
+                                    "type": "text",
+                                    "height": 11,
+                                    "width": 9
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        107,
+                        27,
+                        308,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                107,
+                                26,
+                                308,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        26,
+                                        308,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Under review as a conference paper at ICLR 2019",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "image",
+                    "bbox": [
+                        154,
+                        90,
+                        456,
+                        207
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                154,
+                                90,
+                                456,
+                                207
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        154,
+                                        90,
+                                        456,
+                                        207
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                154,
+                                                90,
+                                                456,
+                                                207
+                                            ],
+                                            "score": 0.908,
+                                            "type": "image",
+                                            "image_path": "96bab39fb3c982480548f52bd526a04980c2093b687c837893866a29eef23e63.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 1,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        154,
+                                        90,
+                                        456,
+                                        129.0
+                                    ],
+                                    "spans": [],
+                                    "index": 0
+                                },
+                                {
+                                    "bbox": [
+                                        154,
+                                        129.0,
+                                        456,
+                                        168.0
+                                    ],
+                                    "spans": [],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        154,
+                                        168.0,
+                                        456,
+                                        207.0
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                106,
+                                218,
+                                505,
+                                240
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        217,
+                                        505,
+                                        230
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                217,
+                                                505,
+                                                230
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 4: Speedup curves as dataset size varies for different datasets. Even as dataset size increases",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 3
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        228,
+                                        472,
+                                        241
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                228,
+                                                211,
+                                                241
+                                            ],
+                                            "score": 1.0,
+                                            "content": "back up to the baseline of",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                212,
+                                                228,
+                                                236,
+                                                239
+                                            ],
+                                            "score": 0.89,
+                                            "content": "1 0 0 \\%",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                236,
+                                                228,
+                                                472,
+                                                241
+                                            ],
+                                            "score": 1.0,
+                                            "content": ", there is no noticeable improvement in convergence speed.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 4
+                                }
+                            ],
+                            "index": 3.5
+                        }
+                    ],
+                    "index": 2.25
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        262,
+                        504,
+                        284
+                    ],
+                    "lines": [],
+                    "index": 5.5,
+                    "bbox_fs": [
+                        104,
+                        261,
+                        504,
+                        288
+                    ],
+                    "lines_deleted": true
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        289,
+                        505,
+                        345
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                289,
+                                505,
+                                302
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        289,
+                                        308,
+                                        302
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In order to test whether the sublinear behavior of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        309,
+                                        290,
+                                        348,
+                                        302
+                                    ],
+                                    "score": 0.93,
+                                    "content": "s ( m ; m _ { 0 } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        348,
+                                        289,
+                                        505,
+                                        302
+                                    ],
+                                    "score": 1.0,
+                                    "content": "depends primarily on dataset size, we",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                301,
+                                505,
+                                312
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        301,
+                                        505,
+                                        312
+                                    ],
+                                    "score": 1.0,
+                                    "content": "compare the speedup curves obtained when training a single model on different fractions of the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                311,
+                                505,
+                                324
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        311,
+                                        505,
+                                        324
+                                    ],
+                                    "score": 1.0,
+                                    "content": "original training data. We trained ResNet34 models on the CIFAR-10 and SVHN datasets (for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                321,
+                                506,
+                                336
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        321,
+                                        280,
+                                        336
+                                    ],
+                                    "score": 1.0,
+                                    "content": "SVHN in this experiment, we train on all",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        281,
+                                        323,
+                                        303,
+                                        333
+                                    ],
+                                    "score": 0.85,
+                                    "content": "6 0 0 k",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        303,
+                                        321,
+                                        506,
+                                        336
+                                    ],
+                                    "score": 1.0,
+                                    "content": "available training images). For each dataset, we",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                333,
+                                376,
+                                347
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        333,
+                                        149,
+                                        347
+                                    ],
+                                    "score": 1.0,
+                                    "content": "trained on",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        149,
+                                        334,
+                                        173,
+                                        345
+                                    ],
+                                    "score": 0.86,
+                                    "content": "1 0 0 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        174,
+                                        333,
+                                        177,
+                                        347
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        177,
+                                        334,
+                                        197,
+                                        345
+                                    ],
+                                    "score": 0.86,
+                                    "content": "5 0 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        197,
+                                        333,
+                                        237,
+                                        347
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", and then",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        237,
+                                        334,
+                                        257,
+                                        344
+                                    ],
+                                    "score": 0.88,
+                                    "content": "2 5 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        257,
+                                        333,
+                                        376,
+                                        347
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of the available training data.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        }
+                    ],
+                    "index": 9,
+                    "bbox_fs": [
+                        105,
+                        289,
+                        506,
+                        347
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        351,
+                        505,
+                        439
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                350,
+                                506,
+                                363
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        350,
+                                        506,
+                                        363
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In Figure 4, we plot the resulting speedup curves for the various partitions. In order to maintain",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                362,
+                                505,
+                                374
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        362,
+                                        505,
+                                        374
+                                    ],
+                                    "score": 1.0,
+                                    "content": "a fair comparison (as baseline loss values change for different dataset sizes), we again choose the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                371,
+                                505,
+                                385
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        371,
+                                        505,
+                                        385
+                                    ],
+                                    "score": 1.0,
+                                    "content": "loss threshold to be the lower quartile of loss values obtained by the largest batch size.1 Notably,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                384,
+                                505,
+                                396
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        384,
+                                        505,
+                                        396
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the batch size at which the curves begin to plateau remains constant as dataset size changes. For",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                394,
+                                505,
+                                407
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        394,
+                                        505,
+                                        407
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ResNet34 on CIFAR-10, the linear speedup behavior breaks around batch size 128 for all three",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                405,
+                                505,
+                                417
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        405,
+                                        505,
+                                        417
+                                    ],
+                                    "score": 1.0,
+                                    "content": "curves. By a batch size of 1024, all curves have flattened. We can see similar behavior for ResNet34",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                417,
+                                506,
+                                429
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        417,
+                                        506,
+                                        429
+                                    ],
+                                    "score": 1.0,
+                                    "content": "on SVHN. Overall, looking back to Figure 3, the choice of model and the complexity of the dataset",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                428,
+                                425,
+                                439
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        428,
+                                        425,
+                                        439
+                                    ],
+                                    "score": 1.0,
+                                    "content": "appear to be more related to the shape of speedup curve than dataset size alone.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        }
+                    ],
+                    "index": 15.5,
+                    "bbox_fs": [
+                        104,
+                        350,
+                        506,
+                        439
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        455,
+                        195,
+                        467
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                453,
+                                197,
+                                470
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        453,
+                                        197,
+                                        470
+                                    ],
+                                    "score": 1.0,
+                                    "content": "5 CONCLUSION",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        }
+                    ],
+                    "index": 20
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        479,
+                        505,
+                        546
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                480,
+                                505,
+                                492
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        480,
+                                        505,
+                                        492
+                                    ],
+                                    "score": 1.0,
+                                    "content": "By experimenting across a wide range of network architectures and problem domains, we find that,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                491,
+                                505,
+                                502
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        491,
+                                        505,
+                                        502
+                                    ],
+                                    "score": 1.0,
+                                    "content": "after a certain point, increasing the batch size fails to decrease wall-clock time to convergence and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                502,
+                                505,
+                                515
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        502,
+                                        505,
+                                        515
+                                    ],
+                                    "score": 1.0,
+                                    "content": "results in low computational efficiency, even assuming perfect parallelism. The critical batch size",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                512,
+                                505,
+                                525
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        512,
+                                        505,
+                                        525
+                                    ],
+                                    "score": 1.0,
+                                    "content": "after which these returns diminish tends to be small relative to existing system capabilities. These",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                523,
+                                506,
+                                537
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        523,
+                                        506,
+                                        537
+                                    ],
+                                    "score": 1.0,
+                                    "content": "trends present impediments to progress in developing effective machine learning systems that are",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                535,
+                                282,
+                                547
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        535,
+                                        282,
+                                        547
+                                    ],
+                                    "score": 1.0,
+                                    "content": "capable of handling growing data demands.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        }
+                    ],
+                    "index": 23.5,
+                    "bbox_fs": [
+                        105,
+                        480,
+                        506,
+                        547
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        551,
+                        505,
+                        596
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                551,
+                                505,
+                                564
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        551,
+                                        505,
+                                        564
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Recent works also suggest heuristics to decrease the generalization gap, but we find that these heuris-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                563,
+                                505,
+                                576
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        563,
+                                        505,
+                                        576
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tics cannot be used to solve the underlying issue of training convergence speed. Moreover, we find",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                573,
+                                505,
+                                586
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        573,
+                                        505,
+                                        586
+                                    ],
+                                    "score": 1.0,
+                                    "content": "that they usually only help decrease the generalization error in a small-to-medium batch size regime.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                583,
+                                505,
+                                597
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        583,
+                                        505,
+                                        597
+                                    ],
+                                    "score": 1.0,
+                                    "content": "There does not seem to be a simple training heuristic to improve large batch performance in general.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        }
+                    ],
+                    "index": 28.5,
+                    "bbox_fs": [
+                        105,
+                        551,
+                        505,
+                        597
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        601,
+                        504,
+                        646
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                600,
+                                506,
+                                614
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        600,
+                                        506,
+                                        614
+                                    ],
+                                    "score": 1.0,
+                                    "content": "These results suggest that we should not assume that increasing the batch size for larger datasets will",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                613,
+                                505,
+                                624
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        613,
+                                        505,
+                                        624
+                                    ],
+                                    "score": 1.0,
+                                    "content": "keep training times manageable for all problems. Even though it is a natural form of data parallelism",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                623,
+                                506,
+                                636
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        623,
+                                        506,
+                                        636
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for large-scale optimization, alternative forms of parallelism should be explored to utilize all of our",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                634,
+                                192,
+                                647
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        634,
+                                        192,
+                                        647
+                                    ],
+                                    "score": 1.0,
+                                    "content": "data more efficiently.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        }
+                    ],
+                    "index": 32.5,
+                    "bbox_fs": [
+                        105,
+                        600,
+                        506,
+                        647
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        81,
+                        176,
+                        94
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                82,
+                                176,
+                                94
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        82,
+                                        176,
+                                        94
+                                    ],
+                                    "score": 1.0,
+                                    "content": "REFERENCES",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        }
+                    ],
+                    "index": 0
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        100,
+                        505,
+                        123
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                99,
+                                505,
+                                112
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        99,
+                                        505,
+                                        112
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Yoshua Bengio and Yann LeCun. Scaling Learning Algorithms Towards AI. In Large Scale Kernel",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                111,
+                                232,
+                                122
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        111,
+                                        232,
+                                        122
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Machines. MIT Press, 2007.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        }
+                    ],
+                    "index": 1.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        131,
+                        505,
+                        154
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                129,
+                                506,
+                                144
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        129,
+                                        506,
+                                        144
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Leon Bottou. Large-scale machine learning with stochastic gradient descent. In ´ Proceedings of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                117,
+                                142,
+                                313,
+                                155
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        117,
+                                        142,
+                                        313,
+                                        155
+                                    ],
+                                    "score": 1.0,
+                                    "content": "COMPSTAT’2010, pp. 177–186. Springer, 2010.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        }
+                    ],
+                    "index": 3.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        161,
+                        504,
+                        184
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                160,
+                                505,
+                                174
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        160,
+                                        505,
+                                        174
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Pratik Chaudhari and Stefano Soatto. Stochastic gradient descent performs variational inference,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                173,
+                                411,
+                                185
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        173,
+                                        411,
+                                        185
+                                    ],
+                                    "score": 1.0,
+                                    "content": "converges to limit cycles for deep networks. arXiv:1710.11029, 10 2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        }
+                    ],
+                    "index": 5.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        191,
+                        506,
+                        236
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                191,
+                                506,
+                                205
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        191,
+                                        506,
+                                        205
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Marius Cordts, Mohamed Omran, Sebastian Ramos, Timo Rehfeld, Markus Enzweiler, Rodrigo",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                204,
+                                505,
+                                214
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        204,
+                                        505,
+                                        214
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Benenson, Uwe Franke, Stefan Roth, and Bernt Schiele. The Cityscapes Dataset for Semantic",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                213,
+                                506,
+                                227
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        213,
+                                        506,
+                                        227
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Urban Scene Understanding. In Proc. of the IEEE Conference on Computer Vision and Pattern",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                225,
+                                228,
+                                237
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        225,
+                                        228,
+                                        237
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Recognition (CVPR), 2016.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        }
+                    ],
+                    "index": 8.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        244,
+                        505,
+                        278
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                244,
+                                504,
+                                257
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        244,
+                                        504,
+                                        257
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Dipankar Das, Sasikanth Avancha, Dheevatsa Mudigere, Karthikeyan Vaidyanathan, Srinivas Srid-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                114,
+                                253,
+                                506,
+                                270
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        114,
+                                        253,
+                                        506,
+                                        270
+                                    ],
+                                    "score": 1.0,
+                                    "content": "haran, Dhiraj D. Kalamkar, Bharat Kaul, and Pradeep Dubey. Distributed Deep Learning Using",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                266,
+                                392,
+                                279
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        266,
+                                        392,
+                                        279
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Synchronous Stochastic Gradient Descent. arXiv:1602.06709, 2016.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        }
+                    ],
+                    "index": 12
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        285,
+                        505,
+                        320
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                285,
+                                505,
+                                299
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        285,
+                                        505,
+                                        299
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Jia Deng, Wei Dong, Richard Socher, Li-Jia Li, Kai Li, and Li Fei-Fei. ImageNet: A large-scale",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                297,
+                                504,
+                                309
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        297,
+                                        504,
+                                        309
+                                    ],
+                                    "score": 1.0,
+                                    "content": "hierarchical image database. In Computer Vision and Pattern Recognition, 2009. CVPR 2009.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                308,
+                                310,
+                                320
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        308,
+                                        310,
+                                        320
+                                    ],
+                                    "score": 1.0,
+                                    "content": "IEEE Conference on, pp. 248–255. IEEE, 2009.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        }
+                    ],
+                    "index": 15
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        327,
+                        505,
+                        350
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                326,
+                                505,
+                                341
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        326,
+                                        505,
+                                        341
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Aditya Devarakonda, Maxim Naumov, and Michael Garland. AdaBatch: Adaptive Batch Sizes for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                339,
+                                354,
+                                351
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        339,
+                                        354,
+                                        351
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Training Deep Neural Networks. arXiv:1712.02029, 2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        }
+                    ],
+                    "index": 17.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        357,
+                        505,
+                        381
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                357,
+                                505,
+                                371
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        357,
+                                        505,
+                                        371
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Felix A. Gers, Jurgen A. Schmidhuber, and Fred A. Cummins. Learning to Forget: Continual ¨",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                369,
+                                408,
+                                382
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        369,
+                                        408,
+                                        382
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Prediction with LSTM. Neural Computing, 12(10):2451–2471, 10 2000.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        }
+                    ],
+                    "index": 19.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        388,
+                        504,
+                        422
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                387,
+                                504,
+                                401
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        387,
+                                        504,
+                                        401
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Priya Goyal, Piotr Dollar, Ross Girshick, Pieter Noordhuis, Lukasz Wesolowski, Aapo Kyrola, An-´",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                399,
+                                505,
+                                412
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        399,
+                                        505,
+                                        412
+                                    ],
+                                    "score": 1.0,
+                                    "content": "drew Tulloch, Yangqing Jia, and Kaiming He. Accurate, large minibatch SGD: training ImageNet",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                410,
+                                262,
+                                422
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        410,
+                                        262,
+                                        422
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in 1 hour. arXiv:1706.02677, 2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        }
+                    ],
+                    "index": 22
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        429,
+                        504,
+                        463
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                428,
+                                504,
+                                444
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        428,
+                                        504,
+                                        444
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image recog-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                439,
+                                505,
+                                455
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        439,
+                                        505,
+                                        455
+                                    ],
+                                    "score": 1.0,
+                                    "content": "nition. In Proceedings of the IEEE conference on computer vision and pattern recognition, pp.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                117,
+                                452,
+                                181,
+                                463
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        117,
+                                        452,
+                                        181,
+                                        463
+                                    ],
+                                    "score": 1.0,
+                                    "content": "770–778, 2016.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        }
+                    ],
+                    "index": 25
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        471,
+                        503,
+                        494
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                470,
+                                505,
+                                485
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        470,
+                                        505,
+                                        485
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Sepp Hochreiter and Jurgen Schmidhuber. Long Short-Term Memory. ¨ Neural Computing, 9(8):",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                482,
+                                204,
+                                493
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        482,
+                                        204,
+                                        493
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1735–1780, 11 1997.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        }
+                    ],
+                    "index": 27.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        501,
+                        505,
+                        536
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                501,
+                                505,
+                                515
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        501,
+                                        505,
+                                        515
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Elad Hoffer, Itay Hubara, and Daniel Soudry. Train longer, generalize better: closing the generaliza-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                512,
+                                505,
+                                527
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        512,
+                                        505,
+                                        527
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tion gap in large batch training of neural networks. In Advances in Neural Information Processing",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                524,
+                                243,
+                                536
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        524,
+                                        243,
+                                        536
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Systems, pp. 1731–1741, 2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        }
+                    ],
+                    "index": 30
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        543,
+                        503,
+                        566
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                107,
+                                544,
+                                504,
+                                555
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        544,
+                                        504,
+                                        555
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Forrest N. Iandola, Khalid Ashraf, Matthew W. Moskewicz, and Kurt Keutzer. FireCaffe: near-linear",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                554,
+                                497,
+                                567
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        554,
+                                        497,
+                                        567
+                                    ],
+                                    "score": 1.0,
+                                    "content": "acceleration of deep neural network training on compute clusters. arXiv:1511.00175, 11 2015.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        }
+                    ],
+                    "index": 32.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        573,
+                        505,
+                        607
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                574,
+                                504,
+                                586
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        574,
+                                        504,
+                                        586
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Stanislaw Jastrzebski, Zachary Kenton, Devansh Arpit, Nicolas Ballas, Asja Fischer, Yoshua Ben-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                585,
+                                504,
+                                597
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        585,
+                                        504,
+                                        597
+                                    ],
+                                    "score": 1.0,
+                                    "content": "gio, and Amos J. Storkey. Three Factors Influencing Minima in SGD. arXiv:1711.04623, 11",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                595,
+                                143,
+                                608
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        595,
+                                        143,
+                                        608
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        }
+                    ],
+                    "index": 35
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        615,
+                        505,
+                        660
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                615,
+                                505,
+                                628
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        615,
+                                        505,
+                                        628
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Xianyan Jia, Shutao Song, Wei He, Yangzihao Wang, Haidong Rong, Feihu Zhou, Liqiang Xie,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                626,
+                                505,
+                                639
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        626,
+                                        505,
+                                        639
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Zhenyu Guo, Yuanzhou Yang, Liwei Yu, Tiegang Chen, Guangxiao Hu, Shaohuai Shi, and Xi-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                637,
+                                505,
+                                651
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        637,
+                                        505,
+                                        651
+                                    ],
+                                    "score": 1.0,
+                                    "content": "aowen Chu. Highly Scalable Deep Learning Training System with Mixed-Precision: Training",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                648,
+                                343,
+                                660
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        648,
+                                        343,
+                                        660
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ImageNet in Four Minutes. arXiv:1807.11205, 07 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        }
+                    ],
+                    "index": 38.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        102,
+                        668,
+                        504,
+                        691
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                666,
+                                505,
+                                681
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        666,
+                                        505,
+                                        681
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Peter H. Jin, Qiaochu Yuan, Forrest N. Iandola, and Kurt Keutzer. How to scale distributed deep",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                678,
+                                273,
+                                691
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        678,
+                                        273,
+                                        691
+                                    ],
+                                    "score": 1.0,
+                                    "content": "learning? arXiv:1611.04581, 11 2016.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        }
+                    ],
+                    "index": 41.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        698,
+                        504,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                698,
+                                504,
+                                711
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        698,
+                                        504,
+                                        711
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Nitish Shirish Keskar, Dheevatsa Mudigere, Jorge Nocedal, Mikhail Smelyanskiy, and Ping Tak Pe-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                709,
+                                505,
+                                722
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        709,
+                                        505,
+                                        722
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ter Tang. On large-batch training for deep learning: Generalization gap and sharp minima.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                720,
+                                219,
+                                732
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        720,
+                                        219,
+                                        732
+                                    ],
+                                    "score": 1.0,
+                                    "content": "arXiv:1609.04836, 2016.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        }
+                    ],
+                    "index": 44
+                }
+            ],
+            "page_idx": 8,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        107,
+                        27,
+                        308,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                107,
+                                26,
+                                308,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        26,
+                                        308,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Under review as a conference paper at ICLR 2019",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        302,
+                        751,
+                        309,
+                        759
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                302,
+                                751,
+                                309,
+                                762
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        302,
+                                        751,
+                                        309,
+                                        762
+                                    ],
+                                    "score": 1.0,
+                                    "content": "9",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        81,
+                        176,
+                        94
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                82,
+                                176,
+                                94
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        82,
+                                        176,
+                                        94
+                                    ],
+                                    "score": 1.0,
+                                    "content": "REFERENCES",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        }
+                    ],
+                    "index": 0
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        100,
+                        505,
+                        123
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                99,
+                                505,
+                                112
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        99,
+                                        505,
+                                        112
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Yoshua Bengio and Yann LeCun. Scaling Learning Algorithms Towards AI. In Large Scale Kernel",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                111,
+                                232,
+                                122
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        111,
+                                        232,
+                                        122
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Machines. MIT Press, 2007.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        }
+                    ],
+                    "index": 1.5,
+                    "bbox_fs": [
+                        106,
+                        99,
+                        505,
+                        122
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        131,
+                        505,
+                        154
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                129,
+                                506,
+                                144
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        129,
+                                        506,
+                                        144
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Leon Bottou. Large-scale machine learning with stochastic gradient descent. In ´ Proceedings of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                117,
+                                142,
+                                313,
+                                155
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        117,
+                                        142,
+                                        313,
+                                        155
+                                    ],
+                                    "score": 1.0,
+                                    "content": "COMPSTAT’2010, pp. 177–186. Springer, 2010.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        }
+                    ],
+                    "index": 3.5,
+                    "bbox_fs": [
+                        105,
+                        129,
+                        506,
+                        155
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        161,
+                        504,
+                        184
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                160,
+                                505,
+                                174
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        160,
+                                        505,
+                                        174
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Pratik Chaudhari and Stefano Soatto. Stochastic gradient descent performs variational inference,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                173,
+                                411,
+                                185
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        173,
+                                        411,
+                                        185
+                                    ],
+                                    "score": 1.0,
+                                    "content": "converges to limit cycles for deep networks. arXiv:1710.11029, 10 2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        }
+                    ],
+                    "index": 5.5,
+                    "bbox_fs": [
+                        105,
+                        160,
+                        505,
+                        185
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        191,
+                        506,
+                        236
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                191,
+                                506,
+                                205
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        191,
+                                        506,
+                                        205
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Marius Cordts, Mohamed Omran, Sebastian Ramos, Timo Rehfeld, Markus Enzweiler, Rodrigo",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                204,
+                                505,
+                                214
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        204,
+                                        505,
+                                        214
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Benenson, Uwe Franke, Stefan Roth, and Bernt Schiele. The Cityscapes Dataset for Semantic",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                213,
+                                506,
+                                227
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        213,
+                                        506,
+                                        227
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Urban Scene Understanding. In Proc. of the IEEE Conference on Computer Vision and Pattern",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                225,
+                                228,
+                                237
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        225,
+                                        228,
+                                        237
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Recognition (CVPR), 2016.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        }
+                    ],
+                    "index": 8.5,
+                    "bbox_fs": [
+                        105,
+                        191,
+                        506,
+                        237
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        244,
+                        505,
+                        278
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                244,
+                                504,
+                                257
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        244,
+                                        504,
+                                        257
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Dipankar Das, Sasikanth Avancha, Dheevatsa Mudigere, Karthikeyan Vaidyanathan, Srinivas Srid-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                114,
+                                253,
+                                506,
+                                270
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        114,
+                                        253,
+                                        506,
+                                        270
+                                    ],
+                                    "score": 1.0,
+                                    "content": "haran, Dhiraj D. Kalamkar, Bharat Kaul, and Pradeep Dubey. Distributed Deep Learning Using",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                266,
+                                392,
+                                279
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        266,
+                                        392,
+                                        279
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Synchronous Stochastic Gradient Descent. arXiv:1602.06709, 2016.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        }
+                    ],
+                    "index": 12,
+                    "bbox_fs": [
+                        106,
+                        244,
+                        506,
+                        279
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        285,
+                        505,
+                        320
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                285,
+                                505,
+                                299
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        285,
+                                        505,
+                                        299
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Jia Deng, Wei Dong, Richard Socher, Li-Jia Li, Kai Li, and Li Fei-Fei. ImageNet: A large-scale",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                297,
+                                504,
+                                309
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        297,
+                                        504,
+                                        309
+                                    ],
+                                    "score": 1.0,
+                                    "content": "hierarchical image database. In Computer Vision and Pattern Recognition, 2009. CVPR 2009.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                308,
+                                310,
+                                320
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        308,
+                                        310,
+                                        320
+                                    ],
+                                    "score": 1.0,
+                                    "content": "IEEE Conference on, pp. 248–255. IEEE, 2009.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        }
+                    ],
+                    "index": 15,
+                    "bbox_fs": [
+                        106,
+                        285,
+                        505,
+                        320
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        327,
+                        505,
+                        350
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                326,
+                                505,
+                                341
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        326,
+                                        505,
+                                        341
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Aditya Devarakonda, Maxim Naumov, and Michael Garland. AdaBatch: Adaptive Batch Sizes for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                339,
+                                354,
+                                351
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        339,
+                                        354,
+                                        351
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Training Deep Neural Networks. arXiv:1712.02029, 2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        }
+                    ],
+                    "index": 17.5,
+                    "bbox_fs": [
+                        105,
+                        326,
+                        505,
+                        351
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        357,
+                        505,
+                        381
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                357,
+                                505,
+                                371
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        357,
+                                        505,
+                                        371
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Felix A. Gers, Jurgen A. Schmidhuber, and Fred A. Cummins. Learning to Forget: Continual ¨",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                369,
+                                408,
+                                382
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        369,
+                                        408,
+                                        382
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Prediction with LSTM. Neural Computing, 12(10):2451–2471, 10 2000.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        }
+                    ],
+                    "index": 19.5,
+                    "bbox_fs": [
+                        105,
+                        357,
+                        505,
+                        382
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        388,
+                        504,
+                        422
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                387,
+                                504,
+                                401
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        387,
+                                        504,
+                                        401
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Priya Goyal, Piotr Dollar, Ross Girshick, Pieter Noordhuis, Lukasz Wesolowski, Aapo Kyrola, An-´",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                399,
+                                505,
+                                412
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        399,
+                                        505,
+                                        412
+                                    ],
+                                    "score": 1.0,
+                                    "content": "drew Tulloch, Yangqing Jia, and Kaiming He. Accurate, large minibatch SGD: training ImageNet",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                410,
+                                262,
+                                422
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        410,
+                                        262,
+                                        422
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in 1 hour. arXiv:1706.02677, 2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        }
+                    ],
+                    "index": 22,
+                    "bbox_fs": [
+                        105,
+                        387,
+                        505,
+                        422
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        429,
+                        504,
+                        463
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                428,
+                                504,
+                                444
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        428,
+                                        504,
+                                        444
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image recog-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                439,
+                                505,
+                                455
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        439,
+                                        505,
+                                        455
+                                    ],
+                                    "score": 1.0,
+                                    "content": "nition. In Proceedings of the IEEE conference on computer vision and pattern recognition, pp.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                117,
+                                452,
+                                181,
+                                463
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        117,
+                                        452,
+                                        181,
+                                        463
+                                    ],
+                                    "score": 1.0,
+                                    "content": "770–778, 2016.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        }
+                    ],
+                    "index": 25,
+                    "bbox_fs": [
+                        105,
+                        428,
+                        505,
+                        463
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        471,
+                        503,
+                        494
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                470,
+                                505,
+                                485
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        470,
+                                        505,
+                                        485
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Sepp Hochreiter and Jurgen Schmidhuber. Long Short-Term Memory. ¨ Neural Computing, 9(8):",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                482,
+                                204,
+                                493
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        482,
+                                        204,
+                                        493
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1735–1780, 11 1997.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        }
+                    ],
+                    "index": 27.5,
+                    "bbox_fs": [
+                        105,
+                        470,
+                        505,
+                        493
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        501,
+                        505,
+                        536
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                501,
+                                505,
+                                515
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        501,
+                                        505,
+                                        515
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Elad Hoffer, Itay Hubara, and Daniel Soudry. Train longer, generalize better: closing the generaliza-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                512,
+                                505,
+                                527
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        512,
+                                        505,
+                                        527
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tion gap in large batch training of neural networks. In Advances in Neural Information Processing",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                524,
+                                243,
+                                536
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        524,
+                                        243,
+                                        536
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Systems, pp. 1731–1741, 2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        }
+                    ],
+                    "index": 30,
+                    "bbox_fs": [
+                        105,
+                        501,
+                        505,
+                        536
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        543,
+                        503,
+                        566
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                107,
+                                544,
+                                504,
+                                555
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        544,
+                                        504,
+                                        555
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Forrest N. Iandola, Khalid Ashraf, Matthew W. Moskewicz, and Kurt Keutzer. FireCaffe: near-linear",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                554,
+                                497,
+                                567
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        554,
+                                        497,
+                                        567
+                                    ],
+                                    "score": 1.0,
+                                    "content": "acceleration of deep neural network training on compute clusters. arXiv:1511.00175, 11 2015.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        }
+                    ],
+                    "index": 32.5,
+                    "bbox_fs": [
+                        107,
+                        544,
+                        504,
+                        567
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        573,
+                        505,
+                        607
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                574,
+                                504,
+                                586
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        574,
+                                        504,
+                                        586
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Stanislaw Jastrzebski, Zachary Kenton, Devansh Arpit, Nicolas Ballas, Asja Fischer, Yoshua Ben-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                585,
+                                504,
+                                597
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        585,
+                                        504,
+                                        597
+                                    ],
+                                    "score": 1.0,
+                                    "content": "gio, and Amos J. Storkey. Three Factors Influencing Minima in SGD. arXiv:1711.04623, 11",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                595,
+                                143,
+                                608
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        595,
+                                        143,
+                                        608
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        }
+                    ],
+                    "index": 35,
+                    "bbox_fs": [
+                        106,
+                        574,
+                        504,
+                        608
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        615,
+                        505,
+                        660
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                615,
+                                505,
+                                628
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        615,
+                                        505,
+                                        628
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Xianyan Jia, Shutao Song, Wei He, Yangzihao Wang, Haidong Rong, Feihu Zhou, Liqiang Xie,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                626,
+                                505,
+                                639
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        626,
+                                        505,
+                                        639
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Zhenyu Guo, Yuanzhou Yang, Liwei Yu, Tiegang Chen, Guangxiao Hu, Shaohuai Shi, and Xi-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                637,
+                                505,
+                                651
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        637,
+                                        505,
+                                        651
+                                    ],
+                                    "score": 1.0,
+                                    "content": "aowen Chu. Highly Scalable Deep Learning Training System with Mixed-Precision: Training",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                648,
+                                343,
+                                660
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        648,
+                                        343,
+                                        660
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ImageNet in Four Minutes. arXiv:1807.11205, 07 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        }
+                    ],
+                    "index": 38.5,
+                    "bbox_fs": [
+                        105,
+                        615,
+                        505,
+                        660
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        102,
+                        668,
+                        504,
+                        691
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                666,
+                                505,
+                                681
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        666,
+                                        505,
+                                        681
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Peter H. Jin, Qiaochu Yuan, Forrest N. Iandola, and Kurt Keutzer. How to scale distributed deep",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                678,
+                                273,
+                                691
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        678,
+                                        273,
+                                        691
+                                    ],
+                                    "score": 1.0,
+                                    "content": "learning? arXiv:1611.04581, 11 2016.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        }
+                    ],
+                    "index": 41.5,
+                    "bbox_fs": [
+                        105,
+                        666,
+                        505,
+                        691
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        698,
+                        504,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                698,
+                                504,
+                                711
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        698,
+                                        504,
+                                        711
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Nitish Shirish Keskar, Dheevatsa Mudigere, Jorge Nocedal, Mikhail Smelyanskiy, and Ping Tak Pe-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                709,
+                                505,
+                                722
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        709,
+                                        505,
+                                        722
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ter Tang. On large-batch training for deep learning: Generalization gap and sharp minima.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                720,
+                                219,
+                                732
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        720,
+                                        219,
+                                        732
+                                    ],
+                                    "score": 1.0,
+                                    "content": "arXiv:1609.04836, 2016.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        }
+                    ],
+                    "index": 44,
+                    "bbox_fs": [
+                        105,
+                        698,
+                        505,
+                        732
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        82,
+                        504,
+                        105
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                81,
+                                506,
+                                96
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        81,
+                                        506,
+                                        96
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Alex Krizhevsky. One weird trick for parallelizing convolutional neural networks. arXiv:1404.5997,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                92,
+                                155,
+                                106
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        92,
+                                        155,
+                                        106
+                                    ],
+                                    "score": 1.0,
+                                    "content": "04 2014.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        }
+                    ],
+                    "index": 0.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        112,
+                        504,
+                        135
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                111,
+                                505,
+                                126
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        111,
+                                        505,
+                                        126
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Alex Krizhevsky and Geoffrey Hinton. Learning multiple layers of features from tiny images. Tech-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                123,
+                                230,
+                                135
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        123,
+                                        230,
+                                        135
+                                    ],
+                                    "score": 1.0,
+                                    "content": "nical report, Citeseer, 2009.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        }
+                    ],
+                    "index": 2.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        141,
+                        505,
+                        186
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                142,
+                                505,
+                                155
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        142,
+                                        505,
+                                        155
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Alex Krizhevsky, Ilya Sutskever, and Geoffrey E. Hinton. ImageNet Classification with Deep Con-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                117,
+                                154,
+                                505,
+                                165
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        117,
+                                        154,
+                                        505,
+                                        165
+                                    ],
+                                    "score": 1.0,
+                                    "content": "volutional Neural Networks. In Proceedings of the 25th International Conference on Neural",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                164,
+                                506,
+                                177
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        164,
+                                        506,
+                                        177
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Information Processing Systems - Volume 1, NIPS’12, pp. 1097–1105. Curran Associates Inc.,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                174,
+                                142,
+                                186
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        174,
+                                        142,
+                                        186
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2012.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        }
+                    ],
+                    "index": 5.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        193,
+                        503,
+                        217
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                194,
+                                505,
+                                207
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        194,
+                                        505,
+                                        207
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Siyuan Ma, Raef Bassily, and Mikhail Belkin. The Power of Interpolation: Understanding the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                204,
+                                472,
+                                217
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        204,
+                                        472,
+                                        217
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Effectiveness of SGD in Modern Over-parametrized Learning. arXiv:1712.06559, 2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        }
+                    ],
+                    "index": 8.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        223,
+                        505,
+                        268
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                224,
+                                505,
+                                237
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        224,
+                                        505,
+                                        237
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Julian McAuley, Christopher Targett, Qinfeng Shi, and Anton van den Hengel. Image-Based Rec-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                235,
+                                506,
+                                247
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        235,
+                                        506,
+                                        247
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ommendations on Styles and Substitutes. In Proceedings of the 38th International ACM SIGIR",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                245,
+                                506,
+                                259
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        245,
+                                        506,
+                                        259
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Conference on Research and Development in Information Retrieval, SIGIR ’15, pp. 43–52. ACM,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                257,
+                                142,
+                                268
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        257,
+                                        142,
+                                        268
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2015.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        }
+                    ],
+                    "index": 11.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        275,
+                        504,
+                        298
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                275,
+                                504,
+                                288
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        275,
+                                        504,
+                                        288
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Stephen Merity, Caiming Xiong, James Bradbury, and Richard Socher. Pointer Sentinel Mixture",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                287,
+                                267,
+                                298
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        287,
+                                        267,
+                                        298
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Models. arXiv:1609.07843, 09 2016.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        }
+                    ],
+                    "index": 14.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        305,
+                        503,
+                        328
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                306,
+                                505,
+                                318
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        306,
+                                        505,
+                                        318
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Raul Puri, Robert Kirby, Nikolai Yakovenko, and Bryan Catanzaro. Large Scale Language Model-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                316,
+                                430,
+                                329
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        316,
+                                        430,
+                                        329
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ing: Converging on 40GB of Text in Four Hours. arXiv:1808.01371, 08 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        }
+                    ],
+                    "index": 16.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        335,
+                        506,
+                        368
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                335,
+                                505,
+                                348
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        335,
+                                        505,
+                                        348
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Mark Sandler, Andrew G. Howard, Menglong Zhu, Andrey Zhmoginov, and Liang-Chieh Chen.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                345,
+                                506,
+                                359
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        345,
+                                        506,
+                                        359
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Very Deep Convolutional Networks for Large-Scale Image Recognition. arXiv:1801.04381, 09",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                356,
+                                143,
+                                369
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        356,
+                                        143,
+                                        369
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2014.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        }
+                    ],
+                    "index": 19
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        376,
+                        504,
+                        399
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                107,
+                                375,
+                                505,
+                                389
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        375,
+                                        505,
+                                        389
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Timothy Sauer. Numerical solution of stochastic differential equations in finance. In Handbook of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                388,
+                                329,
+                                400
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        388,
+                                        329,
+                                        400
+                                    ],
+                                    "score": 1.0,
+                                    "content": "computational finance, pp. 529–550. Springer, 2012.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        }
+                    ],
+                    "index": 21.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        405,
+                        504,
+                        429
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                404,
+                                506,
+                                420
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        404,
+                                        506,
+                                        420
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Karen Simonyan and Andrew Zisserman. Very Deep Convolutional Networks for Large-Scale Image",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                417,
+                                281,
+                                429
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        417,
+                                        281,
+                                        429
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Recognition. arXiv:1409.1556, 09 2014.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        }
+                    ],
+                    "index": 23.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        435,
+                        504,
+                        459
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                435,
+                                505,
+                                448
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        435,
+                                        505,
+                                        448
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Samuel L. Smith and Quoc V. Le. A Bayesian Perspective on Generalization and Stochastic Gradient",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                447,
+                                414,
+                                459
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        447,
+                                        414,
+                                        459
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Descent. In International Conference on Learning Representations, 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        }
+                    ],
+                    "index": 25.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        465,
+                        504,
+                        488
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                464,
+                                505,
+                                480
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        464,
+                                        505,
+                                        480
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Samuel L Smith, Pieter-Jan Kindermans, and Quoc V Le. Don’t Decay the Learning Rate, Increase",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                477,
+                                283,
+                                488
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        477,
+                                        283,
+                                        488
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the Batch Size. arXiv:1711.00489, 2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        }
+                    ],
+                    "index": 27.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        495,
+                        504,
+                        518
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                495,
+                                505,
+                                509
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        495,
+                                        505,
+                                        509
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Chen Xing, Devansh Arpit, Christos Tsirigotis, and Yoshua Bengio. A Walk with SGD.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                507,
+                                219,
+                                518
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        507,
+                                        219,
+                                        518
+                                    ],
+                                    "score": 1.0,
+                                    "content": "arXiv:1802.08770, 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        }
+                    ],
+                    "index": 29.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        525,
+                        505,
+                        549
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                526,
+                                505,
+                                538
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        526,
+                                        505,
+                                        538
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Zhewei Yao, Amir Gholami, Qi Lei, Kurt Keutzer, and Michael W Mahoney. Hessian-based Anal-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                536,
+                                465,
+                                549
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        536,
+                                        465,
+                                        549
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ysis of Large Batch Training and Robustness to Adversaries. arXiv:1802.08241, 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        }
+                    ],
+                    "index": 31.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        555,
+                        504,
+                        578
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                555,
+                                505,
+                                568
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        555,
+                                        505,
+                                        568
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Yang You, Igor Gitman, and Boris Ginsburg. Large Batch Training of Convolutional Networks.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                567,
+                                223,
+                                578
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        567,
+                                        223,
+                                        578
+                                    ],
+                                    "score": 1.0,
+                                    "content": "arXiv:1708.03888, 2017a.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        }
+                    ],
+                    "index": 33.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        585,
+                        505,
+                        608
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                585,
+                                504,
+                                598
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        585,
+                                        504,
+                                        598
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Yang You, Zhao Zhang, Cho-Jui Hsieh, and James Demmel. 100-epoch ImageNet Training with",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                117,
+                                596,
+                                321,
+                                608
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        117,
+                                        596,
+                                        321,
+                                        608
+                                    ],
+                                    "score": 1.0,
+                                    "content": "AlexNet in 24 Minutes. arXiv:1709.05011, 2017b.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        }
+                    ],
+                    "index": 35.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        104,
+                        615,
+                        504,
+                        639
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                614,
+                                506,
+                                628
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        614,
+                                        506,
+                                        628
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Fisher Yu, Vladlen Koltun, and Thomas Funkhouser. Dilated Residual Networks. In 2017 IEEE",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                626,
+                                475,
+                                639
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        626,
+                                        475,
+                                        639
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Conference on Computer Vision and Pattern Recognition (CVPR), pp. 636–644, 07 2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        }
+                    ],
+                    "index": 37.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        645,
+                        504,
+                        679
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                644,
+                                506,
+                                658
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        644,
+                                        506,
+                                        658
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Zhanxing Zhu, Jingfeng Wu, Bing Yu, Lei Wu, and Jinwen Ma. The Anisotropic Noise in",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                655,
+                                506,
+                                669
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        655,
+                                        506,
+                                        669
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Stochastic Gradient Descent: Its Behavior of Escaping from Minima and Regularization Effects.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                668,
+                                219,
+                                678
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        668,
+                                        219,
+                                        678
+                                    ],
+                                    "score": 1.0,
+                                    "content": "arXiv:1803.00195, 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        }
+                    ],
+                    "index": 40
+                }
+            ],
+            "page_idx": 9,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        107,
+                        27,
+                        308,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                107,
+                                26,
+                                308,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        26,
+                                        308,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Under review as a conference paper at ICLR 2019",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        300,
+                        751,
+                        311,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                299,
+                                750,
+                                313,
+                                764
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        299,
+                                        750,
+                                        313,
+                                        764
+                                    ],
+                                    "score": 1.0,
+                                    "content": "10",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        82,
+                        504,
+                        105
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                81,
+                                506,
+                                96
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        81,
+                                        506,
+                                        96
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Alex Krizhevsky. One weird trick for parallelizing convolutional neural networks. arXiv:1404.5997,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                92,
+                                155,
+                                106
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        92,
+                                        155,
+                                        106
+                                    ],
+                                    "score": 1.0,
+                                    "content": "04 2014.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        }
+                    ],
+                    "index": 0.5,
+                    "bbox_fs": [
+                        105,
+                        81,
+                        506,
+                        106
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        112,
+                        504,
+                        135
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                111,
+                                505,
+                                126
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        111,
+                                        505,
+                                        126
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Alex Krizhevsky and Geoffrey Hinton. Learning multiple layers of features from tiny images. Tech-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                123,
+                                230,
+                                135
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        123,
+                                        230,
+                                        135
+                                    ],
+                                    "score": 1.0,
+                                    "content": "nical report, Citeseer, 2009.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        }
+                    ],
+                    "index": 2.5,
+                    "bbox_fs": [
+                        106,
+                        111,
+                        505,
+                        135
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        141,
+                        505,
+                        186
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                142,
+                                505,
+                                155
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        142,
+                                        505,
+                                        155
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Alex Krizhevsky, Ilya Sutskever, and Geoffrey E. Hinton. ImageNet Classification with Deep Con-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                117,
+                                154,
+                                505,
+                                165
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        117,
+                                        154,
+                                        505,
+                                        165
+                                    ],
+                                    "score": 1.0,
+                                    "content": "volutional Neural Networks. In Proceedings of the 25th International Conference on Neural",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                164,
+                                506,
+                                177
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        164,
+                                        506,
+                                        177
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Information Processing Systems - Volume 1, NIPS’12, pp. 1097–1105. Curran Associates Inc.,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                174,
+                                142,
+                                186
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        174,
+                                        142,
+                                        186
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2012.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        }
+                    ],
+                    "index": 5.5,
+                    "bbox_fs": [
+                        105,
+                        142,
+                        506,
+                        186
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        193,
+                        503,
+                        217
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                194,
+                                505,
+                                207
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        194,
+                                        505,
+                                        207
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Siyuan Ma, Raef Bassily, and Mikhail Belkin. The Power of Interpolation: Understanding the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                204,
+                                472,
+                                217
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        204,
+                                        472,
+                                        217
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Effectiveness of SGD in Modern Over-parametrized Learning. arXiv:1712.06559, 2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        }
+                    ],
+                    "index": 8.5,
+                    "bbox_fs": [
+                        106,
+                        194,
+                        505,
+                        217
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        223,
+                        505,
+                        268
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                224,
+                                505,
+                                237
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        224,
+                                        505,
+                                        237
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Julian McAuley, Christopher Targett, Qinfeng Shi, and Anton van den Hengel. Image-Based Rec-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                235,
+                                506,
+                                247
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        235,
+                                        506,
+                                        247
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ommendations on Styles and Substitutes. In Proceedings of the 38th International ACM SIGIR",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                245,
+                                506,
+                                259
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        245,
+                                        506,
+                                        259
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Conference on Research and Development in Information Retrieval, SIGIR ’15, pp. 43–52. ACM,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                257,
+                                142,
+                                268
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        257,
+                                        142,
+                                        268
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2015.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        }
+                    ],
+                    "index": 11.5,
+                    "bbox_fs": [
+                        105,
+                        224,
+                        506,
+                        268
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        275,
+                        504,
+                        298
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                275,
+                                504,
+                                288
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        275,
+                                        504,
+                                        288
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Stephen Merity, Caiming Xiong, James Bradbury, and Richard Socher. Pointer Sentinel Mixture",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                287,
+                                267,
+                                298
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        287,
+                                        267,
+                                        298
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Models. arXiv:1609.07843, 09 2016.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        }
+                    ],
+                    "index": 14.5,
+                    "bbox_fs": [
+                        106,
+                        275,
+                        504,
+                        298
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        305,
+                        503,
+                        328
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                306,
+                                505,
+                                318
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        306,
+                                        505,
+                                        318
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Raul Puri, Robert Kirby, Nikolai Yakovenko, and Bryan Catanzaro. Large Scale Language Model-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                316,
+                                430,
+                                329
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        316,
+                                        430,
+                                        329
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ing: Converging on 40GB of Text in Four Hours. arXiv:1808.01371, 08 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        }
+                    ],
+                    "index": 16.5,
+                    "bbox_fs": [
+                        106,
+                        306,
+                        505,
+                        329
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        335,
+                        506,
+                        368
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                335,
+                                505,
+                                348
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        335,
+                                        505,
+                                        348
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Mark Sandler, Andrew G. Howard, Menglong Zhu, Andrey Zhmoginov, and Liang-Chieh Chen.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                345,
+                                506,
+                                359
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        345,
+                                        506,
+                                        359
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Very Deep Convolutional Networks for Large-Scale Image Recognition. arXiv:1801.04381, 09",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                356,
+                                143,
+                                369
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        356,
+                                        143,
+                                        369
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2014.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        }
+                    ],
+                    "index": 19,
+                    "bbox_fs": [
+                        105,
+                        335,
+                        506,
+                        369
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        376,
+                        504,
+                        399
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                107,
+                                375,
+                                505,
+                                389
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        375,
+                                        505,
+                                        389
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Timothy Sauer. Numerical solution of stochastic differential equations in finance. In Handbook of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                388,
+                                329,
+                                400
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        388,
+                                        329,
+                                        400
+                                    ],
+                                    "score": 1.0,
+                                    "content": "computational finance, pp. 529–550. Springer, 2012.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        }
+                    ],
+                    "index": 21.5,
+                    "bbox_fs": [
+                        107,
+                        375,
+                        505,
+                        400
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        405,
+                        504,
+                        429
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                404,
+                                506,
+                                420
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        404,
+                                        506,
+                                        420
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Karen Simonyan and Andrew Zisserman. Very Deep Convolutional Networks for Large-Scale Image",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                417,
+                                281,
+                                429
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        417,
+                                        281,
+                                        429
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Recognition. arXiv:1409.1556, 09 2014.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        }
+                    ],
+                    "index": 23.5,
+                    "bbox_fs": [
+                        105,
+                        404,
+                        506,
+                        429
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        435,
+                        504,
+                        459
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                435,
+                                505,
+                                448
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        435,
+                                        505,
+                                        448
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Samuel L. Smith and Quoc V. Le. A Bayesian Perspective on Generalization and Stochastic Gradient",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                447,
+                                414,
+                                459
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        447,
+                                        414,
+                                        459
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Descent. In International Conference on Learning Representations, 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        }
+                    ],
+                    "index": 25.5,
+                    "bbox_fs": [
+                        106,
+                        435,
+                        505,
+                        459
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        465,
+                        504,
+                        488
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                464,
+                                505,
+                                480
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        464,
+                                        505,
+                                        480
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Samuel L Smith, Pieter-Jan Kindermans, and Quoc V Le. Don’t Decay the Learning Rate, Increase",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                477,
+                                283,
+                                488
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        477,
+                                        283,
+                                        488
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the Batch Size. arXiv:1711.00489, 2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        }
+                    ],
+                    "index": 27.5,
+                    "bbox_fs": [
+                        106,
+                        464,
+                        505,
+                        488
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        495,
+                        504,
+                        518
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                495,
+                                505,
+                                509
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        495,
+                                        505,
+                                        509
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Chen Xing, Devansh Arpit, Christos Tsirigotis, and Yoshua Bengio. A Walk with SGD.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                507,
+                                219,
+                                518
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        507,
+                                        219,
+                                        518
+                                    ],
+                                    "score": 1.0,
+                                    "content": "arXiv:1802.08770, 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        }
+                    ],
+                    "index": 29.5,
+                    "bbox_fs": [
+                        105,
+                        495,
+                        505,
+                        518
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        525,
+                        505,
+                        549
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                526,
+                                505,
+                                538
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        526,
+                                        505,
+                                        538
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Zhewei Yao, Amir Gholami, Qi Lei, Kurt Keutzer, and Michael W Mahoney. Hessian-based Anal-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                536,
+                                465,
+                                549
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        536,
+                                        465,
+                                        549
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ysis of Large Batch Training and Robustness to Adversaries. arXiv:1802.08241, 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        }
+                    ],
+                    "index": 31.5,
+                    "bbox_fs": [
+                        106,
+                        526,
+                        505,
+                        549
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        555,
+                        504,
+                        578
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                555,
+                                505,
+                                568
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        555,
+                                        505,
+                                        568
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Yang You, Igor Gitman, and Boris Ginsburg. Large Batch Training of Convolutional Networks.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                567,
+                                223,
+                                578
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        567,
+                                        223,
+                                        578
+                                    ],
+                                    "score": 1.0,
+                                    "content": "arXiv:1708.03888, 2017a.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        }
+                    ],
+                    "index": 33.5,
+                    "bbox_fs": [
+                        106,
+                        555,
+                        505,
+                        578
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        585,
+                        505,
+                        608
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                585,
+                                504,
+                                598
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        585,
+                                        504,
+                                        598
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Yang You, Zhao Zhang, Cho-Jui Hsieh, and James Demmel. 100-epoch ImageNet Training with",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                117,
+                                596,
+                                321,
+                                608
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        117,
+                                        596,
+                                        321,
+                                        608
+                                    ],
+                                    "score": 1.0,
+                                    "content": "AlexNet in 24 Minutes. arXiv:1709.05011, 2017b.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        }
+                    ],
+                    "index": 35.5,
+                    "bbox_fs": [
+                        106,
+                        585,
+                        504,
+                        608
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        104,
+                        615,
+                        504,
+                        639
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                614,
+                                506,
+                                628
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        614,
+                                        506,
+                                        628
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Fisher Yu, Vladlen Koltun, and Thomas Funkhouser. Dilated Residual Networks. In 2017 IEEE",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                626,
+                                475,
+                                639
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        626,
+                                        475,
+                                        639
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Conference on Computer Vision and Pattern Recognition (CVPR), pp. 636–644, 07 2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        }
+                    ],
+                    "index": 37.5,
+                    "bbox_fs": [
+                        105,
+                        614,
+                        506,
+                        639
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        645,
+                        504,
+                        679
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                644,
+                                506,
+                                658
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        644,
+                                        506,
+                                        658
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Zhanxing Zhu, Jingfeng Wu, Bing Yu, Lei Wu, and Jinwen Ma. The Anisotropic Noise in",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                655,
+                                506,
+                                669
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        655,
+                                        506,
+                                        669
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Stochastic Gradient Descent: Its Behavior of Escaping from Minima and Regularization Effects.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                668,
+                                219,
+                                678
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        668,
+                                        219,
+                                        678
+                                    ],
+                                    "score": 1.0,
+                                    "content": "arXiv:1803.00195, 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        }
+                    ],
+                    "index": 40,
+                    "bbox_fs": [
+                        105,
+                        644,
+                        506,
+                        678
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "image",
+                    "bbox": [
+                        115,
+                        116,
+                        496,
+                        559
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                115,
+                                116,
+                                496,
+                                559
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        116,
+                                        496,
+                                        559
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                115,
+                                                116,
+                                                496,
+                                                559
+                                            ],
+                                            "score": 0.976,
+                                            "type": "image",
+                                            "image_path": "e2d22d36762c86eb7114a858df3deb4cc64fe830391fe3ba4a35dd5954e98afa.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 1,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        116,
+                                        496,
+                                        263.66666666666663
+                                    ],
+                                    "spans": [],
+                                    "index": 0
+                                },
+                                {
+                                    "bbox": [
+                                        115,
+                                        263.66666666666663,
+                                        496,
+                                        411.33333333333326
+                                    ],
+                                    "spans": [],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        115,
+                                        411.33333333333326,
+                                        496,
+                                        558.9999999999999
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                106,
+                                567,
+                                506,
+                                628
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        566,
+                                        506,
+                                        579
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                566,
+                                                506,
+                                                579
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 5: Contour plots of test losses for various problem domains on a log scale. The test losses for BLR",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 3
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        577,
+                                        505,
+                                        590
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                577,
+                                                505,
+                                                590
+                                            ],
+                                            "score": 1.0,
+                                            "content": "are on the left, while the losses for the LSR strategy are on the right. Lighter colors indicate lower loss values.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 4
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        587,
+                                        505,
+                                        599
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                587,
+                                                505,
+                                                599
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Since we train each batch size for a fixed number of epochs, the total number of training iterations scales down",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 5
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        597,
+                                        506,
+                                        609
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                597,
+                                                506,
+                                                609
+                                            ],
+                                            "score": 1.0,
+                                            "content": "linearly. For each loss value, we can observe how many iterations it takes to converge to that value given a",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 6
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        608,
+                                        505,
+                                        619
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                608,
+                                                505,
+                                                619
+                                            ],
+                                            "score": 1.0,
+                                            "content": "particular batch size, by tracing the level curve for the associated color. For all problems, there is a batch size",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 7
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        617,
+                                        416,
+                                        629
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                617,
+                                                416,
+                                                629
+                                            ],
+                                            "score": 1.0,
+                                            "content": "after which the number of training iterations necessary to converge does not decrease.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 8
+                                }
+                            ],
+                            "index": 5.5
+                        }
+                    ],
+                    "index": 3.25
+                }
+            ],
+            "page_idx": 10,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        106,
+                        27,
+                        308,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                26,
+                                308,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        26,
+                                        308,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Under review as a conference paper at ICLR 2019",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        300,
+                        751,
+                        310,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                299,
+                                750,
+                                312,
+                                765
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        299,
+                                        750,
+                                        312,
+                                        765
+                                    ],
+                                    "score": 1.0,
+                                    "content": "",
+                                    "type": "text",
+                                    "height": 15,
+                                    "width": 13
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        108,
+                        83,
+                        244,
+                        93
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                81,
+                                244,
+                                95
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        81,
+                                        244,
+                                        95
+                                    ],
+                                    "score": 1.0,
+                                    "content": "A ADDITIONAL RESULTS",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "image",
+                    "bbox": [
+                        115,
+                        116,
+                        496,
+                        559
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                115,
+                                116,
+                                496,
+                                559
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        116,
+                                        496,
+                                        559
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                115,
+                                                116,
+                                                496,
+                                                559
+                                            ],
+                                            "score": 0.976,
+                                            "type": "image",
+                                            "image_path": "e2d22d36762c86eb7114a858df3deb4cc64fe830391fe3ba4a35dd5954e98afa.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 1,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        116,
+                                        496,
+                                        263.66666666666663
+                                    ],
+                                    "spans": [],
+                                    "index": 0
+                                },
+                                {
+                                    "bbox": [
+                                        115,
+                                        263.66666666666663,
+                                        496,
+                                        411.33333333333326
+                                    ],
+                                    "spans": [],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        115,
+                                        411.33333333333326,
+                                        496,
+                                        558.9999999999999
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                106,
+                                567,
+                                506,
+                                628
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        566,
+                                        506,
+                                        579
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                566,
+                                                506,
+                                                579
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 5: Contour plots of test losses for various problem domains on a log scale. The test losses for BLR",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 3
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        577,
+                                        505,
+                                        590
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                577,
+                                                505,
+                                                590
+                                            ],
+                                            "score": 1.0,
+                                            "content": "are on the left, while the losses for the LSR strategy are on the right. Lighter colors indicate lower loss values.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 4
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        587,
+                                        505,
+                                        599
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                587,
+                                                505,
+                                                599
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Since we train each batch size for a fixed number of epochs, the total number of training iterations scales down",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 5
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        597,
+                                        506,
+                                        609
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                597,
+                                                506,
+                                                609
+                                            ],
+                                            "score": 1.0,
+                                            "content": "linearly. For each loss value, we can observe how many iterations it takes to converge to that value given a",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 6
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        608,
+                                        505,
+                                        619
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                608,
+                                                505,
+                                                619
+                                            ],
+                                            "score": 1.0,
+                                            "content": "particular batch size, by tracing the level curve for the associated color. For all problems, there is a batch size",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 7
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        617,
+                                        416,
+                                        629
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                617,
+                                                416,
+                                                629
+                                            ],
+                                            "score": 1.0,
+                                            "content": "after which the number of training iterations necessary to converge does not decrease.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 8
+                                }
+                            ],
+                            "index": 5.5
+                        }
+                    ],
+                    "index": 3.25
+                }
+            ]
+        }
+    ],
+    "_backend": "pipeline",
+    "_version_name": "2.2.2"
+}
\ No newline at end of file
diff --git a/parse/train/S1en0sRqKm/S1en0sRqKm_model.json b/parse/train/S1en0sRqKm/S1en0sRqKm_model.json
new file mode 100644
index 0000000000000000000000000000000000000000..b16ee0b22c113b178f81ade5946b3214e994f099
--- /dev/null
+++ b/parse/train/S1en0sRqKm/S1en0sRqKm_model.json
@@ -0,0 +1,14662 @@
+[
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1239,
+                    1404,
+                    1239,
+                    1404,
+                    1665,
+                    298,
+                    1665
+                ],
+                "score": 0.984
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    398,
+                    587,
+                    1302,
+                    587,
+                    1302,
+                    1106,
+                    398,
+                    1106
+                ],
+                "score": 0.982
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1681,
+                    1401,
+                    1681,
+                    1401,
+                    1865,
+                    298,
+                    1865
+                ],
+                "score": 0.98
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    1881,
+                    1400,
+                    1881,
+                    1400,
+                    2033,
+                    299,
+                    2033
+                ],
+                "score": 0.976
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    301,
+                    220,
+                    1400,
+                    220,
+                    1400,
+                    325,
+                    301,
+                    325
+                ],
+                "score": 0.964
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    298,
+                    75,
+                    857,
+                    75,
+                    857,
+                    105,
+                    298,
+                    105
+                ],
+                "score": 0.901
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    302,
+                    1168,
+                    572,
+                    1168,
+                    572,
+                    1203,
+                    302,
+                    1203
+                ],
+                "score": 0.891
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    773,
+                    519,
+                    926,
+                    519,
+                    926,
+                    553,
+                    773,
+                    553
+                ],
+                "score": 0.863
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    314,
+                    377,
+                    681,
+                    377,
+                    681,
+                    438,
+                    314,
+                    438
+                ],
+                "score": 0.822
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    842,
+                    2088,
+                    857,
+                    2088,
+                    857,
+                    2112,
+                    842,
+                    2112
+                ],
+                "score": 0.601
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    215.0,
+                    1405.0,
+                    215.0,
+                    1405.0,
+                    274.0,
+                    292.0,
+                    274.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    276.0,
+                    1389.0,
+                    276.0,
+                    1389.0,
+                    325.0,
+                    296.0,
+                    325.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    298.0,
+                    73.0,
+                    858.0,
+                    73.0,
+                    858.0,
+                    108.0,
+                    298.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1165.0,
+                    579.0,
+                    1165.0,
+                    579.0,
+                    1212.0,
+                    293.0,
+                    1212.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    769.0,
+                    518.0,
+                    932.0,
+                    518.0,
+                    932.0,
+                    556.0,
+                    769.0,
+                    556.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    841.0,
+                    2088.0,
+                    858.0,
+                    2088.0,
+                    858.0,
+                    2116.0,
+                    841.0,
+                    2116.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1237.0,
+                    1405.0,
+                    1237.0,
+                    1405.0,
+                    1275.0,
+                    293.0,
+                    1275.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1267.0,
+                    1405.0,
+                    1267.0,
+                    1405.0,
+                    1303.0,
+                    293.0,
+                    1303.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1299.0,
+                    1407.0,
+                    1299.0,
+                    1407.0,
+                    1336.0,
+                    292.0,
+                    1336.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1331.0,
+                    1405.0,
+                    1331.0,
+                    1405.0,
+                    1366.0,
+                    293.0,
+                    1366.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1358.0,
+                    1406.0,
+                    1358.0,
+                    1406.0,
+                    1396.0,
+                    292.0,
+                    1396.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1391.0,
+                    1404.0,
+                    1391.0,
+                    1404.0,
+                    1427.0,
+                    293.0,
+                    1427.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1420.0,
+                    1405.0,
+                    1420.0,
+                    1405.0,
+                    1458.0,
+                    291.0,
+                    1458.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1455.0,
+                    1405.0,
+                    1455.0,
+                    1405.0,
+                    1486.0,
+                    294.0,
+                    1486.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1482.0,
+                    1404.0,
+                    1482.0,
+                    1404.0,
+                    1518.0,
+                    294.0,
+                    1518.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1513.0,
+                    1402.0,
+                    1513.0,
+                    1402.0,
+                    1548.0,
+                    293.0,
+                    1548.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1546.0,
+                    1405.0,
+                    1546.0,
+                    1405.0,
+                    1577.0,
+                    296.0,
+                    1577.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1572.0,
+                    1408.0,
+                    1572.0,
+                    1408.0,
+                    1611.0,
+                    292.0,
+                    1611.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1605.0,
+                    1405.0,
+                    1605.0,
+                    1405.0,
+                    1640.0,
+                    294.0,
+                    1640.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1635.0,
+                    495.0,
+                    1635.0,
+                    495.0,
+                    1667.0,
+                    294.0,
+                    1667.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    394.0,
+                    589.0,
+                    1303.0,
+                    589.0,
+                    1303.0,
+                    625.0,
+                    394.0,
+                    625.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    392.0,
+                    620.0,
+                    1306.0,
+                    620.0,
+                    1306.0,
+                    655.0,
+                    392.0,
+                    655.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    395.0,
+                    652.0,
+                    1305.0,
+                    652.0,
+                    1305.0,
+                    685.0,
+                    395.0,
+                    685.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    393.0,
+                    681.0,
+                    1305.0,
+                    681.0,
+                    1305.0,
+                    716.0,
+                    393.0,
+                    716.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    395.0,
+                    714.0,
+                    1305.0,
+                    714.0,
+                    1305.0,
+                    744.0,
+                    395.0,
+                    744.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    395.0,
+                    744.0,
+                    1304.0,
+                    744.0,
+                    1304.0,
+                    778.0,
+                    395.0,
+                    778.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    394.0,
+                    772.0,
+                    1306.0,
+                    772.0,
+                    1306.0,
+                    809.0,
+                    394.0,
+                    809.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    394.0,
+                    802.0,
+                    1306.0,
+                    802.0,
+                    1306.0,
+                    839.0,
+                    394.0,
+                    839.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    395.0,
+                    835.0,
+                    1304.0,
+                    835.0,
+                    1304.0,
+                    868.0,
+                    395.0,
+                    868.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    395.0,
+                    865.0,
+                    1304.0,
+                    865.0,
+                    1304.0,
+                    898.0,
+                    395.0,
+                    898.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    393.0,
+                    894.0,
+                    1306.0,
+                    894.0,
+                    1306.0,
+                    929.0,
+                    393.0,
+                    929.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    393.0,
+                    924.0,
+                    1307.0,
+                    924.0,
+                    1307.0,
+                    960.0,
+                    393.0,
+                    960.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    393.0,
+                    954.0,
+                    1306.0,
+                    954.0,
+                    1306.0,
+                    993.0,
+                    393.0,
+                    993.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    394.0,
+                    986.0,
+                    1306.0,
+                    986.0,
+                    1306.0,
+                    1020.0,
+                    394.0,
+                    1020.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    393.0,
+                    1017.0,
+                    1305.0,
+                    1017.0,
+                    1305.0,
+                    1050.0,
+                    393.0,
+                    1050.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    393.0,
+                    1046.0,
+                    1305.0,
+                    1046.0,
+                    1305.0,
+                    1079.0,
+                    393.0,
+                    1079.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    394.0,
+                    1076.0,
+                    1093.0,
+                    1076.0,
+                    1093.0,
+                    1110.0,
+                    394.0,
+                    1110.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1680.0,
+                    1403.0,
+                    1680.0,
+                    1403.0,
+                    1716.0,
+                    294.0,
+                    1716.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1711.0,
+                    1403.0,
+                    1711.0,
+                    1403.0,
+                    1746.0,
+                    294.0,
+                    1746.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1744.0,
+                    1403.0,
+                    1744.0,
+                    1403.0,
+                    1776.0,
+                    296.0,
+                    1776.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1773.0,
+                    1405.0,
+                    1773.0,
+                    1405.0,
+                    1809.0,
+                    294.0,
+                    1809.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1802.0,
+                    1405.0,
+                    1802.0,
+                    1405.0,
+                    1839.0,
+                    293.0,
+                    1839.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1833.0,
+                    1144.0,
+                    1833.0,
+                    1144.0,
+                    1871.0,
+                    293.0,
+                    1871.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1878.0,
+                    1405.0,
+                    1878.0,
+                    1405.0,
+                    1916.0,
+                    293.0,
+                    1916.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1908.0,
+                    1405.0,
+                    1908.0,
+                    1405.0,
+                    1948.0,
+                    294.0,
+                    1948.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1941.0,
+                    1402.0,
+                    1941.0,
+                    1402.0,
+                    1977.0,
+                    293.0,
+                    1977.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1971.0,
+                    1404.0,
+                    1971.0,
+                    1404.0,
+                    2008.0,
+                    295.0,
+                    2008.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    2003.0,
+                    888.0,
+                    2003.0,
+                    888.0,
+                    2036.0,
+                    295.0,
+                    2036.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    315.0,
+                    378.0,
+                    560.0,
+                    378.0,
+                    560.0,
+                    411.0,
+                    315.0,
+                    411.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    311.0,
+                    407.0,
+                    682.0,
+                    407.0,
+                    682.0,
+                    440.0,
+                    311.0,
+                    440.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 0,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    368,
+                    1404,
+                    368,
+                    1404,
+                    795,
+                    298,
+                    795
+                ],
+                "score": 0.981
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    811,
+                    1404,
+                    811,
+                    1404,
+                    1087,
+                    298,
+                    1087
+                ],
+                "score": 0.98
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1101,
+                    1403,
+                    1101,
+                    1403,
+                    1316,
+                    298,
+                    1316
+                ],
+                "score": 0.974
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1850,
+                    1404,
+                    1850,
+                    1404,
+                    2034,
+                    298,
+                    2034
+                ],
+                "score": 0.974
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    229,
+                    1402,
+                    229,
+                    1402,
+                    352,
+                    299,
+                    352
+                ],
+                "score": 0.973
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    299,
+                    76,
+                    857,
+                    76,
+                    857,
+                    104,
+                    299,
+                    104
+                ],
+                "score": 0.867
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    368,
+                    1350,
+                    1406,
+                    1350,
+                    1406,
+                    1800,
+                    368,
+                    1800
+                ],
+                "score": 0.758
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    841,
+                    2089,
+                    858,
+                    2089,
+                    858,
+                    2112,
+                    841,
+                    2112
+                ],
+                "score": 0.7
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    841,
+                    2088,
+                    859,
+                    2088,
+                    859,
+                    2112,
+                    841,
+                    2112
+                ],
+                "score": 0.175
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    300,
+                    76,
+                    855,
+                    76,
+                    855,
+                    104,
+                    300,
+                    104
+                ],
+                "score": 0.101
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    298.0,
+                    73.0,
+                    858.0,
+                    73.0,
+                    858.0,
+                    108.0,
+                    298.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    839.0,
+                    2086.0,
+                    862.0,
+                    2086.0,
+                    862.0,
+                    2121.0,
+                    839.0,
+                    2121.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    838.0,
+                    2085.0,
+                    862.0,
+                    2085.0,
+                    862.0,
+                    2120.0,
+                    838.0,
+                    2120.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    73.0,
+                    859.0,
+                    73.0,
+                    859.0,
+                    108.0,
+                    297.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    367.0,
+                    1404.0,
+                    367.0,
+                    1404.0,
+                    402.0,
+                    294.0,
+                    402.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    397.0,
+                    1404.0,
+                    397.0,
+                    1404.0,
+                    437.0,
+                    293.0,
+                    437.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    428.0,
+                    1406.0,
+                    428.0,
+                    1406.0,
+                    465.0,
+                    293.0,
+                    465.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    461.0,
+                    1405.0,
+                    461.0,
+                    1405.0,
+                    497.0,
+                    293.0,
+                    497.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    488.0,
+                    1408.0,
+                    488.0,
+                    1408.0,
+                    530.0,
+                    291.0,
+                    530.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    519.0,
+                    1406.0,
+                    519.0,
+                    1406.0,
+                    557.0,
+                    293.0,
+                    557.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    550.0,
+                    1406.0,
+                    550.0,
+                    1406.0,
+                    587.0,
+                    295.0,
+                    587.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    583.0,
+                    1406.0,
+                    583.0,
+                    1406.0,
+                    617.0,
+                    292.0,
+                    617.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    611.0,
+                    1406.0,
+                    611.0,
+                    1406.0,
+                    649.0,
+                    293.0,
+                    649.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    642.0,
+                    1405.0,
+                    642.0,
+                    1405.0,
+                    678.0,
+                    294.0,
+                    678.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    673.0,
+                    1405.0,
+                    673.0,
+                    1405.0,
+                    708.0,
+                    294.0,
+                    708.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    702.0,
+                    1406.0,
+                    702.0,
+                    1406.0,
+                    739.0,
+                    292.0,
+                    739.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    731.0,
+                    1405.0,
+                    731.0,
+                    1405.0,
+                    773.0,
+                    291.0,
+                    773.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    768.0,
+                    394.0,
+                    768.0,
+                    394.0,
+                    798.0,
+                    292.0,
+                    798.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    811.0,
+                    1405.0,
+                    811.0,
+                    1405.0,
+                    846.0,
+                    295.0,
+                    846.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    843.0,
+                    1405.0,
+                    843.0,
+                    1405.0,
+                    876.0,
+                    296.0,
+                    876.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    872.0,
+                    1405.0,
+                    872.0,
+                    1405.0,
+                    907.0,
+                    293.0,
+                    907.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    902.0,
+                    1405.0,
+                    902.0,
+                    1405.0,
+                    940.0,
+                    291.0,
+                    940.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    933.0,
+                    1404.0,
+                    933.0,
+                    1404.0,
+                    970.0,
+                    293.0,
+                    970.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    958.0,
+                    1406.0,
+                    958.0,
+                    1406.0,
+                    1004.0,
+                    292.0,
+                    1004.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    996.0,
+                    1405.0,
+                    996.0,
+                    1405.0,
+                    1029.0,
+                    296.0,
+                    1029.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1022.0,
+                    1406.0,
+                    1022.0,
+                    1406.0,
+                    1060.0,
+                    293.0,
+                    1060.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1056.0,
+                    1227.0,
+                    1056.0,
+                    1227.0,
+                    1093.0,
+                    294.0,
+                    1093.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1098.0,
+                    1404.0,
+                    1098.0,
+                    1404.0,
+                    1141.0,
+                    293.0,
+                    1141.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1133.0,
+                    1404.0,
+                    1133.0,
+                    1404.0,
+                    1168.0,
+                    293.0,
+                    1168.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1163.0,
+                    1406.0,
+                    1163.0,
+                    1406.0,
+                    1199.0,
+                    292.0,
+                    1199.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1196.0,
+                    1403.0,
+                    1196.0,
+                    1403.0,
+                    1227.0,
+                    296.0,
+                    1227.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1224.0,
+                    1405.0,
+                    1224.0,
+                    1405.0,
+                    1259.0,
+                    294.0,
+                    1259.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1255.0,
+                    1405.0,
+                    1255.0,
+                    1405.0,
+                    1290.0,
+                    294.0,
+                    1290.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1286.0,
+                    1199.0,
+                    1286.0,
+                    1199.0,
+                    1320.0,
+                    294.0,
+                    1320.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1847.0,
+                    1405.0,
+                    1847.0,
+                    1405.0,
+                    1886.0,
+                    292.0,
+                    1886.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1879.0,
+                    1405.0,
+                    1879.0,
+                    1405.0,
+                    1917.0,
+                    295.0,
+                    1917.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1909.0,
+                    1406.0,
+                    1909.0,
+                    1406.0,
+                    1947.0,
+                    293.0,
+                    1947.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1942.0,
+                    1406.0,
+                    1942.0,
+                    1406.0,
+                    1978.0,
+                    294.0,
+                    1978.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    1974.0,
+                    1405.0,
+                    1974.0,
+                    1405.0,
+                    2005.0,
+                    297.0,
+                    2005.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    2002.0,
+                    934.0,
+                    2002.0,
+                    934.0,
+                    2038.0,
+                    295.0,
+                    2038.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    231.0,
+                    1403.0,
+                    231.0,
+                    1403.0,
+                    264.0,
+                    295.0,
+                    264.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    258.0,
+                    1406.0,
+                    258.0,
+                    1406.0,
+                    299.0,
+                    294.0,
+                    299.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    293.0,
+                    1404.0,
+                    293.0,
+                    1404.0,
+                    325.0,
+                    295.0,
+                    325.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    323.0,
+                    941.0,
+                    323.0,
+                    941.0,
+                    356.0,
+                    295.0,
+                    356.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    380.0,
+                    1349.0,
+                    1406.0,
+                    1349.0,
+                    1406.0,
+                    1387.0,
+                    380.0,
+                    1387.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    394.0,
+                    1381.0,
+                    1406.0,
+                    1381.0,
+                    1406.0,
+                    1419.0,
+                    394.0,
+                    1419.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    394.0,
+                    1410.0,
+                    1405.0,
+                    1410.0,
+                    1405.0,
+                    1447.0,
+                    394.0,
+                    1447.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    395.0,
+                    1444.0,
+                    1326.0,
+                    1444.0,
+                    1326.0,
+                    1478.0,
+                    395.0,
+                    1478.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    381.0,
+                    1498.0,
+                    1404.0,
+                    1498.0,
+                    1404.0,
+                    1534.0,
+                    381.0,
+                    1534.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    393.0,
+                    1530.0,
+                    1406.0,
+                    1530.0,
+                    1406.0,
+                    1565.0,
+                    393.0,
+                    1565.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    395.0,
+                    1561.0,
+                    1405.0,
+                    1561.0,
+                    1405.0,
+                    1596.0,
+                    395.0,
+                    1596.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    393.0,
+                    1587.0,
+                    924.0,
+                    1587.0,
+                    924.0,
+                    1628.0,
+                    393.0,
+                    1628.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    384.0,
+                    1641.0,
+                    1405.0,
+                    1641.0,
+                    1405.0,
+                    1687.0,
+                    384.0,
+                    1687.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    394.0,
+                    1677.0,
+                    1405.0,
+                    1677.0,
+                    1405.0,
+                    1715.0,
+                    394.0,
+                    1715.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    394.0,
+                    1706.0,
+                    1405.0,
+                    1706.0,
+                    1405.0,
+                    1744.0,
+                    394.0,
+                    1744.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    395.0,
+                    1740.0,
+                    1404.0,
+                    1740.0,
+                    1404.0,
+                    1774.0,
+                    395.0,
+                    1774.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    394.0,
+                    1768.0,
+                    1173.0,
+                    1768.0,
+                    1173.0,
+                    1806.0,
+                    394.0,
+                    1806.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 1,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    631,
+                    1403,
+                    631,
+                    1403,
+                    725,
+                    299,
+                    725
+                ],
+                "score": 0.975
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    295,
+                    1404,
+                    295,
+                    1404,
+                    388,
+                    297,
+                    388
+                ],
+                "score": 0.974
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1051,
+                    1404,
+                    1051,
+                    1404,
+                    1237,
+                    297,
+                    1237
+                ],
+                "score": 0.969
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    300,
+                    1251,
+                    1400,
+                    1251,
+                    1400,
+                    1344,
+                    300,
+                    1344
+                ],
+                "score": 0.963
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    300,
+                    834,
+                    1404,
+                    834,
+                    1404,
+                    930,
+                    300,
+                    930
+                ],
+                "score": 0.962
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    709,
+                    389,
+                    989,
+                    389,
+                    989,
+                    472,
+                    709,
+                    472
+                ],
+                "score": 0.959
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    301,
+                    943,
+                    1399,
+                    943,
+                    1399,
+                    1038,
+                    301,
+                    1038
+                ],
+                "score": 0.958
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    630,
+                    532,
+                    1066,
+                    532,
+                    1066,
+                    615,
+                    630,
+                    615
+                ],
+                "score": 0.955
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    368,
+                    1371,
+                    1404,
+                    1371,
+                    1404,
+                    1942,
+                    368,
+                    1942
+                ],
+                "score": 0.949
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    453,
+                    741,
+                    1246,
+                    741,
+                    1246,
+                    821,
+                    453,
+                    821
+                ],
+                "score": 0.946
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    300,
+                    1972,
+                    1397,
+                    1972,
+                    1397,
+                    2034,
+                    300,
+                    2034
+                ],
+                "score": 0.941
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    298,
+                    225,
+                    873,
+                    225,
+                    873,
+                    263,
+                    298,
+                    263
+                ],
+                "score": 0.925
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    300,
+                    479,
+                    1395,
+                    479,
+                    1395,
+                    514,
+                    300,
+                    514
+                ],
+                "score": 0.921
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    297,
+                    74,
+                    858,
+                    74,
+                    858,
+                    106,
+                    297,
+                    106
+                ],
+                "score": 0.92
+            },
+            {
+                "category_id": 9,
+                "poly": [
+                    1366,
+                    412,
+                    1400,
+                    412,
+                    1400,
+                    443,
+                    1366,
+                    443
+                ],
+                "score": 0.887
+            },
+            {
+                "category_id": 9,
+                "poly": [
+                    1366,
+                    758,
+                    1400,
+                    758,
+                    1400,
+                    788,
+                    1366,
+                    788
+                ],
+                "score": 0.877
+            },
+            {
+                "category_id": 9,
+                "poly": [
+                    1366,
+                    555,
+                    1400,
+                    555,
+                    1400,
+                    586,
+                    1366,
+                    586
+                ],
+                "score": 0.875
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    841,
+                    2088,
+                    858,
+                    2088,
+                    858,
+                    2112,
+                    841,
+                    2112
+                ],
+                "score": 0.634
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    841,
+                    2088,
+                    859,
+                    2088,
+                    859,
+                    2112,
+                    841,
+                    2112
+                ],
+                "score": 0.402
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    707,
+                    382,
+                    989,
+                    382,
+                    989,
+                    475,
+                    707,
+                    475
+                ],
+                "score": 0.95,
+                "latex": "L ( \\mathbf { w } ) = \\frac { 1 } { n } \\sum _ { i = 1 } ^ { n } \\ell ( \\mathbf { w } , \\mathbf { x } _ { i } ) ,"
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    631,
+                    527,
+                    1068,
+                    527,
+                    1068,
+                    619,
+                    631,
+                    619
+                ],
+                "score": 0.95,
+                "latex": "\\mathbf { g } ( \\mathbf { w } ) : = \\nabla L ( \\mathbf { w } ) = \\frac { 1 } { n } \\sum _ { i = 1 } ^ { n } \\nabla \\ell ( \\mathbf { w } , \\mathbf { x } _ { i } ) ."
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    454,
+                    737,
+                    1243,
+                    737,
+                    1243,
+                    823,
+                    454,
+                    823
+                ],
+                "score": 0.94,
+                "latex": "\\mathbf { g } _ { m } ( \\mathbf { w } _ { k } ) = \\frac { 1 } { m } \\sum _ { i \\in \\mathcal { B } _ { m } } \\nabla \\ell ( \\mathbf { w } _ { k } , \\mathbf { x } _ { i } ) \\quad \\mathrm { ~ a n d ~ } \\quad \\mathbf { w } _ { k + 1 } = \\mathbf { w } _ { k } - \\eta _ { k } \\mathbf { g } _ { m } ( \\mathbf { w } _ { k } ) ,"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1272,
+                    632,
+                    1393,
+                    632,
+                    1393,
+                    666,
+                    1272,
+                    666
+                ],
+                "score": 0.93,
+                "latex": "\\{ 1 , \\ldots , n \\}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    372,
+                    480,
+                    435,
+                    480,
+                    435,
+                    514,
+                    372,
+                    514
+                ],
+                "score": 0.92,
+                "latex": "\\ell ( \\cdot , \\cdot )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1101,
+                    867,
+                    1159,
+                    867,
+                    1159,
+                    900,
+                    1101,
+                    900
+                ],
+                "score": 0.92,
+                "latex": "n / m"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1023,
+                    1503,
+                    1155,
+                    1503,
+                    1155,
+                    1538,
+                    1023,
+                    1538
+                ],
+                "score": 0.91,
+                "latex": "\\eta _ { 0 } \\propto \\sqrt { m }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    371,
+                    838,
+                    454,
+                    838,
+                    454,
+                    868,
+                    371,
+                    868
+                ],
+                "score": 0.91,
+                "latex": "\\eta _ { k } > 0"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    688,
+                    325,
+                    780,
+                    325,
+                    780,
+                    355,
+                    688,
+                    355
+                ],
+                "score": 0.91,
+                "latex": "\\mathbf { w } \\in \\mathbb { R } ^ { d }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    448,
+                    634,
+                    489,
+                    634,
+                    489,
+                    663,
+                    448,
+                    663
+                ],
+                "score": 0.89,
+                "latex": "B _ { m }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    569,
+                    636,
+                    651,
+                    636,
+                    651,
+                    661,
+                    569,
+                    661
+                ],
+                "score": 0.89,
+                "latex": "m < n"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    448,
+                    362,
+                    477,
+                    362,
+                    477,
+                    388,
+                    448,
+                    388
+                ],
+                "score": 0.85,
+                "latex": "\\mathbf { x } _ { i }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    400,
+                    1731,
+                    511,
+                    1731,
+                    511,
+                    1762,
+                    400,
+                    1762
+                ],
+                "score": 0.84,
+                "latex": "( \\eta _ { 0 } \\propto m )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1139,
+                    664,
+                    1157,
+                    664,
+                    1157,
+                    690,
+                    1139,
+                    690
+                ],
+                "score": 0.81,
+                "latex": "k"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    838,
+                    638,
+                    865,
+                    638,
+                    865,
+                    660,
+                    838,
+                    660
+                ],
+                "score": 0.79,
+                "latex": "m"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    808,
+                    838,
+                    826,
+                    838,
+                    826,
+                    864,
+                    808,
+                    864
+                ],
+                "score": 0.78,
+                "latex": "k"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    297,
+                    364,
+                    317,
+                    364,
+                    317,
+                    385,
+                    297,
+                    385
+                ],
+                "score": 0.7,
+                "latex": "n"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    222.0,
+                    877.0,
+                    222.0,
+                    877.0,
+                    268.0,
+                    291.0,
+                    268.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    72.0,
+                    858.0,
+                    72.0,
+                    858.0,
+                    109.0,
+                    297.0,
+                    109.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    839.0,
+                    2085.0,
+                    860.0,
+                    2085.0,
+                    860.0,
+                    2116.0,
+                    839.0,
+                    2116.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    838.0,
+                    2085.0,
+                    862.0,
+                    2085.0,
+                    862.0,
+                    2118.0,
+                    838.0,
+                    2118.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    627.0,
+                    447.0,
+                    627.0,
+                    447.0,
+                    668.0,
+                    293.0,
+                    668.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    490.0,
+                    627.0,
+                    568.0,
+                    627.0,
+                    568.0,
+                    668.0,
+                    490.0,
+                    668.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    652.0,
+                    627.0,
+                    837.0,
+                    627.0,
+                    837.0,
+                    668.0,
+                    652.0,
+                    668.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    866.0,
+                    627.0,
+                    1271.0,
+                    627.0,
+                    1271.0,
+                    668.0,
+                    866.0,
+                    668.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1394.0,
+                    627.0,
+                    1405.0,
+                    627.0,
+                    1405.0,
+                    668.0,
+                    1394.0,
+                    668.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    661.0,
+                    1138.0,
+                    661.0,
+                    1138.0,
+                    696.0,
+                    294.0,
+                    696.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1158.0,
+                    661.0,
+                    1405.0,
+                    661.0,
+                    1405.0,
+                    696.0,
+                    1158.0,
+                    696.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    693.0,
+                    557.0,
+                    693.0,
+                    557.0,
+                    729.0,
+                    293.0,
+                    729.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    294.0,
+                    1404.0,
+                    294.0,
+                    1404.0,
+                    329.0,
+                    295.0,
+                    329.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    325.0,
+                    687.0,
+                    325.0,
+                    687.0,
+                    361.0,
+                    294.0,
+                    361.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    781.0,
+                    325.0,
+                    1405.0,
+                    325.0,
+                    1405.0,
+                    361.0,
+                    781.0,
+                    361.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    355.0,
+                    296.0,
+                    355.0,
+                    296.0,
+                    395.0,
+                    293.0,
+                    395.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    318.0,
+                    355.0,
+                    447.0,
+                    355.0,
+                    447.0,
+                    395.0,
+                    318.0,
+                    395.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    478.0,
+                    355.0,
+                    491.0,
+                    355.0,
+                    491.0,
+                    395.0,
+                    478.0,
+                    395.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1050.0,
+                    1406.0,
+                    1050.0,
+                    1406.0,
+                    1087.0,
+                    294.0,
+                    1087.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1084.0,
+                    1405.0,
+                    1084.0,
+                    1405.0,
+                    1116.0,
+                    295.0,
+                    1116.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1112.0,
+                    1406.0,
+                    1112.0,
+                    1406.0,
+                    1148.0,
+                    292.0,
+                    1148.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1143.0,
+                    1405.0,
+                    1143.0,
+                    1405.0,
+                    1178.0,
+                    294.0,
+                    1178.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1173.0,
+                    1406.0,
+                    1173.0,
+                    1406.0,
+                    1209.0,
+                    294.0,
+                    1209.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1205.0,
+                    1091.0,
+                    1205.0,
+                    1091.0,
+                    1237.0,
+                    295.0,
+                    1237.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1250.0,
+                    1405.0,
+                    1250.0,
+                    1405.0,
+                    1288.0,
+                    294.0,
+                    1288.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1282.0,
+                    1404.0,
+                    1282.0,
+                    1404.0,
+                    1316.0,
+                    295.0,
+                    1316.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1313.0,
+                    1098.0,
+                    1313.0,
+                    1098.0,
+                    1347.0,
+                    296.0,
+                    1347.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    834.0,
+                    370.0,
+                    834.0,
+                    370.0,
+                    872.0,
+                    295.0,
+                    872.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    455.0,
+                    834.0,
+                    807.0,
+                    834.0,
+                    807.0,
+                    872.0,
+                    455.0,
+                    872.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    827.0,
+                    834.0,
+                    1408.0,
+                    834.0,
+                    1408.0,
+                    872.0,
+                    827.0,
+                    872.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    865.0,
+                    1100.0,
+                    865.0,
+                    1100.0,
+                    904.0,
+                    293.0,
+                    904.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1160.0,
+                    865.0,
+                    1404.0,
+                    865.0,
+                    1404.0,
+                    904.0,
+                    1160.0,
+                    904.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    894.0,
+                    1404.0,
+                    894.0,
+                    1404.0,
+                    933.0,
+                    294.0,
+                    933.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    943.0,
+                    1404.0,
+                    943.0,
+                    1404.0,
+                    981.0,
+                    295.0,
+                    981.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    973.0,
+                    1402.0,
+                    973.0,
+                    1402.0,
+                    1008.0,
+                    295.0,
+                    1008.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1006.0,
+                    1239.0,
+                    1006.0,
+                    1239.0,
+                    1040.0,
+                    296.0,
+                    1040.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    372.0,
+                    1372.0,
+                    1405.0,
+                    1372.0,
+                    1405.0,
+                    1408.0,
+                    372.0,
+                    1408.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    393.0,
+                    1401.0,
+                    1407.0,
+                    1401.0,
+                    1407.0,
+                    1439.0,
+                    393.0,
+                    1439.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    394.0,
+                    1434.0,
+                    1405.0,
+                    1434.0,
+                    1405.0,
+                    1470.0,
+                    394.0,
+                    1470.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    395.0,
+                    1467.0,
+                    1395.0,
+                    1467.0,
+                    1395.0,
+                    1499.0,
+                    395.0,
+                    1499.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    368.0,
+                    1506.0,
+                    1022.0,
+                    1506.0,
+                    1022.0,
+                    1542.0,
+                    368.0,
+                    1542.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1156.0,
+                    1506.0,
+                    1405.0,
+                    1506.0,
+                    1405.0,
+                    1542.0,
+                    1156.0,
+                    1542.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    395.0,
+                    1537.0,
+                    1404.0,
+                    1537.0,
+                    1404.0,
+                    1569.0,
+                    395.0,
+                    1569.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    394.0,
+                    1567.0,
+                    1404.0,
+                    1567.0,
+                    1404.0,
+                    1599.0,
+                    394.0,
+                    1599.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    395.0,
+                    1597.0,
+                    1404.0,
+                    1597.0,
+                    1404.0,
+                    1630.0,
+                    395.0,
+                    1630.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    394.0,
+                    1629.0,
+                    1210.0,
+                    1629.0,
+                    1210.0,
+                    1661.0,
+                    394.0,
+                    1661.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    381.0,
+                    1667.0,
+                    1406.0,
+                    1667.0,
+                    1406.0,
+                    1705.0,
+                    381.0,
+                    1705.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    394.0,
+                    1698.0,
+                    1407.0,
+                    1698.0,
+                    1407.0,
+                    1735.0,
+                    394.0,
+                    1735.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    393.0,
+                    1730.0,
+                    399.0,
+                    1730.0,
+                    399.0,
+                    1764.0,
+                    393.0,
+                    1764.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    512.0,
+                    1730.0,
+                    1407.0,
+                    1730.0,
+                    1407.0,
+                    1764.0,
+                    512.0,
+                    1764.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    393.0,
+                    1758.0,
+                    1405.0,
+                    1758.0,
+                    1405.0,
+                    1796.0,
+                    393.0,
+                    1796.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    393.0,
+                    1790.0,
+                    1407.0,
+                    1790.0,
+                    1407.0,
+                    1827.0,
+                    393.0,
+                    1827.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    394.0,
+                    1821.0,
+                    1405.0,
+                    1821.0,
+                    1405.0,
+                    1852.0,
+                    394.0,
+                    1852.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    394.0,
+                    1851.0,
+                    1405.0,
+                    1851.0,
+                    1405.0,
+                    1887.0,
+                    394.0,
+                    1887.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    394.0,
+                    1881.0,
+                    1405.0,
+                    1881.0,
+                    1405.0,
+                    1917.0,
+                    394.0,
+                    1917.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    395.0,
+                    1912.0,
+                    691.0,
+                    1912.0,
+                    691.0,
+                    1945.0,
+                    395.0,
+                    1945.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1970.0,
+                    1402.0,
+                    1970.0,
+                    1402.0,
+                    2007.0,
+                    294.0,
+                    2007.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1999.0,
+                    1402.0,
+                    1999.0,
+                    1402.0,
+                    2040.0,
+                    294.0,
+                    2040.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    474.0,
+                    371.0,
+                    474.0,
+                    371.0,
+                    519.0,
+                    294.0,
+                    519.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    436.0,
+                    474.0,
+                    1395.0,
+                    474.0,
+                    1395.0,
+                    519.0,
+                    436.0,
+                    519.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 2,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1514,
+                    1404,
+                    1514,
+                    1404,
+                    2035,
+                    298,
+                    2035
+                ],
+                "score": 0.985
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    927,
+                    1403,
+                    927,
+                    1403,
+                    1201,
+                    298,
+                    1201
+                ],
+                "score": 0.983
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    399,
+                    1404,
+                    399,
+                    1404,
+                    583,
+                    299,
+                    583
+                ],
+                "score": 0.978
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    725,
+                    1403,
+                    725,
+                    1403,
+                    911,
+                    299,
+                    911
+                ],
+                "score": 0.977
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1217,
+                    1402,
+                    1217,
+                    1402,
+                    1371,
+                    298,
+                    1371
+                ],
+                "score": 0.976
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    229,
+                    1403,
+                    229,
+                    1403,
+                    383,
+                    298,
+                    383
+                ],
+                "score": 0.975
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    299,
+                    645,
+                    1106,
+                    645,
+                    1106,
+                    682,
+                    299,
+                    682
+                ],
+                "score": 0.899
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    302,
+                    1434,
+                    707,
+                    1434,
+                    707,
+                    1469,
+                    302,
+                    1469
+                ],
+                "score": 0.863
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    299,
+                    76,
+                    857,
+                    76,
+                    857,
+                    104,
+                    299,
+                    104
+                ],
+                "score": 0.849
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    841,
+                    2089,
+                    858,
+                    2089,
+                    858,
+                    2111,
+                    841,
+                    2111
+                ],
+                "score": 0.756
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    300,
+                    76,
+                    855,
+                    76,
+                    855,
+                    104,
+                    300,
+                    104
+                ],
+                "score": 0.186
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1330,
+                    849,
+                    1393,
+                    849,
+                    1393,
+                    883,
+                    1330,
+                    883
+                ],
+                "score": 0.94,
+                "latex": "k ( m )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    791,
+                    727,
+                    865,
+                    727,
+                    865,
+                    760,
+                    791,
+                    760
+                ],
+                "score": 0.92,
+                "latex": "k _ { \\epsilon } ( m )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1033,
+                    1048,
+                    1334,
+                    1048,
+                    1334,
+                    1082,
+                    1033,
+                    1082
+                ],
+                "score": 0.91,
+                "latex": "s ( m ; m _ { 0 } ) = k ( m _ { 0 } ) / k ( m )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    298,
+                    819,
+                    361,
+                    819,
+                    361,
+                    853,
+                    298,
+                    853
+                ],
+                "score": 0.91,
+                "latex": "k ( m )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    746,
+                    761,
+                    818,
+                    761,
+                    818,
+                    788,
+                    746,
+                    788
+                ],
+                "score": 0.89,
+                "latex": "\\epsilon > 0"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    418,
+                    1053,
+                    458,
+                    1053,
+                    458,
+                    1080,
+                    418,
+                    1080
+                ],
+                "score": 0.86,
+                "latex": "m _ { 0 }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    901,
+                    824,
+                    927,
+                    824,
+                    927,
+                    847,
+                    901,
+                    847
+                ],
+                "score": 0.7,
+                "latex": "m"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1169,
+                    763,
+                    1198,
+                    763,
+                    1198,
+                    787,
+                    1169,
+                    787
+                ],
+                "score": 0.7,
+                "latex": "m"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    404,
+                    794,
+                    420,
+                    794,
+                    420,
+                    816,
+                    404,
+                    816
+                ],
+                "score": 0.63,
+                "latex": "\\epsilon"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1230,
+                    1086,
+                    1257,
+                    1086,
+                    1257,
+                    1108,
+                    1230,
+                    1108
+                ],
+                "score": 0.57,
+                "latex": "m"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    642.0,
+                    1111.0,
+                    642.0,
+                    1111.0,
+                    688.0,
+                    291.0,
+                    688.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1430.0,
+                    713.0,
+                    1430.0,
+                    713.0,
+                    1476.0,
+                    293.0,
+                    1476.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    298.0,
+                    73.0,
+                    858.0,
+                    73.0,
+                    858.0,
+                    108.0,
+                    298.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    838.0,
+                    2086.0,
+                    862.0,
+                    2086.0,
+                    862.0,
+                    2118.0,
+                    838.0,
+                    2118.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    73.0,
+                    859.0,
+                    73.0,
+                    859.0,
+                    108.0,
+                    297.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1513.0,
+                    1404.0,
+                    1513.0,
+                    1404.0,
+                    1549.0,
+                    293.0,
+                    1549.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1544.0,
+                    1404.0,
+                    1544.0,
+                    1404.0,
+                    1583.0,
+                    292.0,
+                    1583.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1578.0,
+                    1405.0,
+                    1578.0,
+                    1405.0,
+                    1610.0,
+                    296.0,
+                    1610.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1604.0,
+                    1406.0,
+                    1604.0,
+                    1406.0,
+                    1646.0,
+                    291.0,
+                    1646.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1633.0,
+                    1408.0,
+                    1633.0,
+                    1408.0,
+                    1678.0,
+                    291.0,
+                    1678.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1667.0,
+                    1406.0,
+                    1667.0,
+                    1406.0,
+                    1705.0,
+                    293.0,
+                    1705.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1699.0,
+                    1402.0,
+                    1699.0,
+                    1402.0,
+                    1732.0,
+                    294.0,
+                    1732.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1725.0,
+                    1405.0,
+                    1725.0,
+                    1405.0,
+                    1764.0,
+                    292.0,
+                    1764.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1756.0,
+                    1405.0,
+                    1756.0,
+                    1405.0,
+                    1794.0,
+                    292.0,
+                    1794.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1785.0,
+                    1408.0,
+                    1785.0,
+                    1408.0,
+                    1828.0,
+                    291.0,
+                    1828.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1818.0,
+                    1405.0,
+                    1818.0,
+                    1405.0,
+                    1856.0,
+                    293.0,
+                    1856.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1848.0,
+                    1408.0,
+                    1848.0,
+                    1408.0,
+                    1888.0,
+                    292.0,
+                    1888.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1882.0,
+                    1405.0,
+                    1882.0,
+                    1405.0,
+                    1918.0,
+                    292.0,
+                    1918.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1909.0,
+                    1405.0,
+                    1909.0,
+                    1405.0,
+                    1948.0,
+                    291.0,
+                    1948.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1940.0,
+                    1406.0,
+                    1940.0,
+                    1406.0,
+                    1978.0,
+                    293.0,
+                    1978.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1971.0,
+                    1406.0,
+                    1971.0,
+                    1406.0,
+                    2009.0,
+                    292.0,
+                    2009.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    2003.0,
+                    1256.0,
+                    2003.0,
+                    1256.0,
+                    2036.0,
+                    297.0,
+                    2036.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    929.0,
+                    1404.0,
+                    929.0,
+                    1404.0,
+                    961.0,
+                    296.0,
+                    961.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    959.0,
+                    1404.0,
+                    959.0,
+                    1404.0,
+                    991.0,
+                    294.0,
+                    991.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    990.0,
+                    1404.0,
+                    990.0,
+                    1404.0,
+                    1022.0,
+                    294.0,
+                    1022.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1017.0,
+                    1405.0,
+                    1017.0,
+                    1405.0,
+                    1054.0,
+                    292.0,
+                    1054.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1048.0,
+                    417.0,
+                    1048.0,
+                    417.0,
+                    1085.0,
+                    294.0,
+                    1085.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    459.0,
+                    1048.0,
+                    1032.0,
+                    1048.0,
+                    1032.0,
+                    1085.0,
+                    459.0,
+                    1085.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1335.0,
+                    1048.0,
+                    1405.0,
+                    1048.0,
+                    1405.0,
+                    1085.0,
+                    1335.0,
+                    1085.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1080.0,
+                    1229.0,
+                    1080.0,
+                    1229.0,
+                    1115.0,
+                    293.0,
+                    1115.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1258.0,
+                    1080.0,
+                    1405.0,
+                    1080.0,
+                    1405.0,
+                    1115.0,
+                    1258.0,
+                    1115.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1111.0,
+                    1405.0,
+                    1111.0,
+                    1405.0,
+                    1143.0,
+                    294.0,
+                    1143.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1141.0,
+                    1405.0,
+                    1141.0,
+                    1405.0,
+                    1177.0,
+                    294.0,
+                    1177.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1172.0,
+                    1076.0,
+                    1172.0,
+                    1076.0,
+                    1204.0,
+                    296.0,
+                    1204.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    397.0,
+                    1404.0,
+                    397.0,
+                    1404.0,
+                    433.0,
+                    297.0,
+                    433.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    430.0,
+                    1406.0,
+                    430.0,
+                    1406.0,
+                    463.0,
+                    293.0,
+                    463.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    459.0,
+                    1406.0,
+                    459.0,
+                    1406.0,
+                    494.0,
+                    293.0,
+                    494.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    490.0,
+                    1404.0,
+                    490.0,
+                    1404.0,
+                    524.0,
+                    293.0,
+                    524.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    521.0,
+                    1406.0,
+                    521.0,
+                    1406.0,
+                    556.0,
+                    293.0,
+                    556.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    549.0,
+                    1270.0,
+                    549.0,
+                    1270.0,
+                    587.0,
+                    294.0,
+                    587.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    726.0,
+                    790.0,
+                    726.0,
+                    790.0,
+                    762.0,
+                    297.0,
+                    762.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    866.0,
+                    726.0,
+                    1405.0,
+                    726.0,
+                    1405.0,
+                    762.0,
+                    866.0,
+                    762.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    758.0,
+                    745.0,
+                    758.0,
+                    745.0,
+                    794.0,
+                    295.0,
+                    794.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    819.0,
+                    758.0,
+                    1168.0,
+                    758.0,
+                    1168.0,
+                    794.0,
+                    819.0,
+                    794.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1199.0,
+                    758.0,
+                    1404.0,
+                    758.0,
+                    1404.0,
+                    794.0,
+                    1199.0,
+                    794.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    783.0,
+                    403.0,
+                    783.0,
+                    403.0,
+                    828.0,
+                    292.0,
+                    828.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    421.0,
+                    783.0,
+                    1407.0,
+                    783.0,
+                    1407.0,
+                    828.0,
+                    421.0,
+                    828.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    362.0,
+                    819.0,
+                    900.0,
+                    819.0,
+                    900.0,
+                    855.0,
+                    362.0,
+                    855.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    928.0,
+                    819.0,
+                    1405.0,
+                    819.0,
+                    1405.0,
+                    855.0,
+                    928.0,
+                    855.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    850.0,
+                    1329.0,
+                    850.0,
+                    1329.0,
+                    884.0,
+                    293.0,
+                    884.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1394.0,
+                    850.0,
+                    1405.0,
+                    850.0,
+                    1405.0,
+                    884.0,
+                    1394.0,
+                    884.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    881.0,
+                    860.0,
+                    881.0,
+                    860.0,
+                    915.0,
+                    295.0,
+                    915.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    1217.0,
+                    1404.0,
+                    1217.0,
+                    1404.0,
+                    1250.0,
+                    297.0,
+                    1250.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1248.0,
+                    1406.0,
+                    1248.0,
+                    1406.0,
+                    1282.0,
+                    293.0,
+                    1282.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1277.0,
+                    1404.0,
+                    1277.0,
+                    1404.0,
+                    1314.0,
+                    292.0,
+                    1314.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1309.0,
+                    1406.0,
+                    1309.0,
+                    1406.0,
+                    1346.0,
+                    294.0,
+                    1346.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1338.0,
+                    582.0,
+                    1338.0,
+                    582.0,
+                    1375.0,
+                    295.0,
+                    1375.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    228.0,
+                    1407.0,
+                    228.0,
+                    1407.0,
+                    266.0,
+                    292.0,
+                    266.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    258.0,
+                    1405.0,
+                    258.0,
+                    1405.0,
+                    298.0,
+                    293.0,
+                    298.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    290.0,
+                    1405.0,
+                    290.0,
+                    1405.0,
+                    327.0,
+                    293.0,
+                    327.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    321.0,
+                    1405.0,
+                    321.0,
+                    1405.0,
+                    358.0,
+                    293.0,
+                    358.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    351.0,
+                    543.0,
+                    351.0,
+                    543.0,
+                    388.0,
+                    293.0,
+                    388.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 3,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1772,
+                    1403,
+                    1772,
+                    1403,
+                    1958,
+                    298,
+                    1958
+                ],
+                "score": 0.977
+            },
+            {
+                "category_id": 3,
+                "poly": [
+                    322,
+                    237,
+                    1379,
+                    237,
+                    1379,
+                    1470,
+                    322,
+                    1470
+                ],
+                "score": 0.975
+            },
+            {
+                "category_id": 4,
+                "poly": [
+                    296,
+                    1498,
+                    1405,
+                    1498,
+                    1405,
+                    1640,
+                    296,
+                    1640
+                ],
+                "score": 0.958
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1972,
+                    1401,
+                    1972,
+                    1401,
+                    2034,
+                    298,
+                    2034
+                ],
+                "score": 0.938
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    298,
+                    1710,
+                    995,
+                    1710,
+                    995,
+                    1742,
+                    298,
+                    1742
+                ],
+                "score": 0.893
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    298,
+                    76,
+                    857,
+                    76,
+                    857,
+                    104,
+                    298,
+                    104
+                ],
+                "score": 0.816
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    841,
+                    2088,
+                    858,
+                    2088,
+                    858,
+                    2112,
+                    841,
+                    2112
+                ],
+                "score": 0.721
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    300,
+                    76,
+                    855,
+                    76,
+                    855,
+                    104,
+                    300,
+                    104
+                ],
+                "score": 0.248
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1111,
+                    1894,
+                    1295,
+                    1894,
+                    1295,
+                    1928,
+                    1111,
+                    1928
+                ],
+                "score": 0.93,
+                "latex": "s ( m ) \\ \\bar { \\ll } \\ m / m _ { 0 }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    609,
+                    1864,
+                    1032,
+                    1864,
+                    1032,
+                    1898,
+                    609,
+                    1898
+                ],
+                "score": 0.91,
+                "latex": "s ( m ; \\bar { m } _ { 0 } ) = k ( \\bar { m } _ { 0 } ) / k ( m ) \\approx m / m _ { 0 }"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    421.0,
+                    242.0,
+                    650.0,
+                    242.0,
+                    650.0,
+                    286.0,
+                    421.0,
+                    286.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    982.0,
+                    242.0,
+                    1219.0,
+                    242.0,
+                    1219.0,
+                    286.0,
+                    982.0,
+                    286.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    422.0,
+                    302.0,
+                    677.0,
+                    302.0,
+                    677.0,
+                    341.0,
+                    422.0,
+                    341.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    768.0,
+                    330.0,
+                    819.0,
+                    330.0,
+                    819.0,
+                    359.0,
+                    768.0,
+                    359.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    937.0,
+                    334.0,
+                    948.0,
+                    334.0,
+                    948.0,
+                    345.0,
+                    937.0,
+                    345.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    980.0,
+                    302.0,
+                    1234.0,
+                    302.0,
+                    1234.0,
+                    341.0,
+                    980.0,
+                    341.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1329.0,
+                    320.0,
+                    1375.0,
+                    320.0,
+                    1375.0,
+                    349.0,
+                    1329.0,
+                    349.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    343.0,
+                    374.0,
+                    380.0,
+                    374.0,
+                    380.0,
+                    408.0,
+                    343.0,
+                    408.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    899.0,
+                    374.0,
+                    938.0,
+                    374.0,
+                    938.0,
+                    408.0,
+                    899.0,
+                    408.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    317.0,
+                    416.0,
+                    380.0,
+                    416.0,
+                    380.0,
+                    537.0,
+                    317.0,
+                    537.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    874.0,
+                    416.0,
+                    938.0,
+                    416.0,
+                    938.0,
+                    540.0,
+                    874.0,
+                    540.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    343.0,
+                    490.0,
+                    380.0,
+                    490.0,
+                    380.0,
+                    524.0,
+                    343.0,
+                    524.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    899.0,
+                    487.0,
+                    939.0,
+                    487.0,
+                    939.0,
+                    526.0,
+                    899.0,
+                    526.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    350.0,
+                    531.0,
+                    379.0,
+                    531.0,
+                    379.0,
+                    562.0,
+                    350.0,
+                    562.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    908.0,
+                    531.0,
+                    937.0,
+                    531.0,
+                    937.0,
+                    562.0,
+                    908.0,
+                    562.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    351.0,
+                    570.0,
+                    379.0,
+                    570.0,
+                    379.0,
+                    599.0,
+                    351.0,
+                    599.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    908.0,
+                    569.0,
+                    937.0,
+                    569.0,
+                    937.0,
+                    601.0,
+                    908.0,
+                    601.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    351.0,
+                    608.0,
+                    394.0,
+                    608.0,
+                    394.0,
+                    648.0,
+                    351.0,
+                    648.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    425.0,
+                    620.0,
+                    457.0,
+                    620.0,
+                    457.0,
+                    649.0,
+                    425.0,
+                    649.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    488.0,
+                    620.0,
+                    520.0,
+                    620.0,
+                    520.0,
+                    649.0,
+                    488.0,
+                    649.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    543.0,
+                    619.0,
+                    586.0,
+                    619.0,
+                    586.0,
+                    651.0,
+                    543.0,
+                    651.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    609.0,
+                    619.0,
+                    650.0,
+                    619.0,
+                    650.0,
+                    651.0,
+                    609.0,
+                    651.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    670.0,
+                    619.0,
+                    711.0,
+                    619.0,
+                    711.0,
+                    651.0,
+                    670.0,
+                    651.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    766.0,
+                    594.0,
+                    819.0,
+                    594.0,
+                    819.0,
+                    632.0,
+                    766.0,
+                    632.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    908.0,
+                    608.0,
+                    952.0,
+                    608.0,
+                    952.0,
+                    649.0,
+                    908.0,
+                    649.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    983.0,
+                    622.0,
+                    1013.0,
+                    622.0,
+                    1013.0,
+                    649.0,
+                    983.0,
+                    649.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1042.0,
+                    619.0,
+                    1079.0,
+                    619.0,
+                    1079.0,
+                    651.0,
+                    1042.0,
+                    651.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1104.0,
+                    619.0,
+                    1144.0,
+                    619.0,
+                    1144.0,
+                    651.0,
+                    1104.0,
+                    651.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1227.0,
+                    619.0,
+                    1268.0,
+                    619.0,
+                    1268.0,
+                    651.0,
+                    1227.0,
+                    651.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1326.0,
+                    609.0,
+                    1373.0,
+                    609.0,
+                    1373.0,
+                    638.0,
+                    1326.0,
+                    638.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    488.0,
+                    638.0,
+                    614.0,
+                    638.0,
+                    614.0,
+                    679.0,
+                    488.0,
+                    679.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1044.0,
+                    638.0,
+                    1172.0,
+                    638.0,
+                    1172.0,
+                    679.0,
+                    1044.0,
+                    679.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    348.0,
+                    715.0,
+                    389.0,
+                    715.0,
+                    389.0,
+                    748.0,
+                    348.0,
+                    748.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    420.0,
+                    699.0,
+                    695.0,
+                    699.0,
+                    695.0,
+                    743.0,
+                    420.0,
+                    743.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    776.0,
+                    726.0,
+                    826.0,
+                    726.0,
+                    826.0,
+                    761.0,
+                    776.0,
+                    761.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    893.0,
+                    714.0,
+                    939.0,
+                    714.0,
+                    939.0,
+                    747.0,
+                    893.0,
+                    747.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    968.0,
+                    699.0,
+                    1246.0,
+                    699.0,
+                    1246.0,
+                    743.0,
+                    968.0,
+                    743.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1326.0,
+                    729.0,
+                    1376.0,
+                    729.0,
+                    1376.0,
+                    758.0,
+                    1326.0,
+                    758.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    346.0,
+                    764.0,
+                    385.0,
+                    764.0,
+                    385.0,
+                    792.0,
+                    346.0,
+                    792.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    899.0,
+                    762.0,
+                    938.0,
+                    762.0,
+                    938.0,
+                    795.0,
+                    899.0,
+                    795.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    325.0,
+                    817.0,
+                    357.0,
+                    817.0,
+                    357.0,
+                    931.0,
+                    325.0,
+                    931.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    360.0,
+                    910.0,
+                    383.0,
+                    910.0,
+                    383.0,
+                    937.0,
+                    360.0,
+                    937.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    874.0,
+                    813.0,
+                    937.0,
+                    813.0,
+                    937.0,
+                    940.0,
+                    874.0,
+                    940.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    357.0,
+                    955.0,
+                    386.0,
+                    955.0,
+                    386.0,
+                    987.0,
+                    357.0,
+                    987.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    908.0,
+                    956.0,
+                    937.0,
+                    956.0,
+                    937.0,
+                    988.0,
+                    908.0,
+                    988.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    355.0,
+                    1003.0,
+                    389.0,
+                    1003.0,
+                    389.0,
+                    1038.0,
+                    355.0,
+                    1038.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    414.0,
+                    1019.0,
+                    443.0,
+                    1019.0,
+                    443.0,
+                    1048.0,
+                    414.0,
+                    1048.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    499.0,
+                    1019.0,
+                    529.0,
+                    1019.0,
+                    529.0,
+                    1047.0,
+                    499.0,
+                    1047.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    585.0,
+                    1019.0,
+                    614.0,
+                    1019.0,
+                    614.0,
+                    1048.0,
+                    585.0,
+                    1048.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    666.0,
+                    1017.0,
+                    706.0,
+                    1017.0,
+                    706.0,
+                    1049.0,
+                    666.0,
+                    1049.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    906.0,
+                    1003.0,
+                    939.0,
+                    1003.0,
+                    939.0,
+                    1038.0,
+                    906.0,
+                    1038.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    963.0,
+                    1019.0,
+                    994.0,
+                    1019.0,
+                    994.0,
+                    1048.0,
+                    963.0,
+                    1048.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1049.0,
+                    1019.0,
+                    1079.0,
+                    1019.0,
+                    1079.0,
+                    1047.0,
+                    1049.0,
+                    1047.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1136.0,
+                    1019.0,
+                    1165.0,
+                    1019.0,
+                    1165.0,
+                    1048.0,
+                    1136.0,
+                    1048.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1216.0,
+                    1017.0,
+                    1255.0,
+                    1017.0,
+                    1255.0,
+                    1049.0,
+                    1216.0,
+                    1049.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1327.0,
+                    997.0,
+                    1374.0,
+                    997.0,
+                    1374.0,
+                    1031.0,
+                    1327.0,
+                    1031.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    495.0,
+                    1035.0,
+                    620.0,
+                    1035.0,
+                    620.0,
+                    1076.0,
+                    495.0,
+                    1076.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1046.0,
+                    1035.0,
+                    1171.0,
+                    1035.0,
+                    1171.0,
+                    1075.0,
+                    1046.0,
+                    1075.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    443.0,
+                    1101.0,
+                    671.0,
+                    1101.0,
+                    671.0,
+                    1135.0,
+                    443.0,
+                    1135.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    774.0,
+                    1124.0,
+                    825.0,
+                    1124.0,
+                    825.0,
+                    1153.0,
+                    774.0,
+                    1153.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    895.0,
+                    1114.0,
+                    937.0,
+                    1114.0,
+                    937.0,
+                    1141.0,
+                    895.0,
+                    1141.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    994.0,
+                    1101.0,
+                    1222.0,
+                    1101.0,
+                    1222.0,
+                    1135.0,
+                    994.0,
+                    1135.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1326.0,
+                    1130.0,
+                    1376.0,
+                    1130.0,
+                    1376.0,
+                    1160.0,
+                    1326.0,
+                    1160.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    350.0,
+                    1177.0,
+                    387.0,
+                    1177.0,
+                    387.0,
+                    1210.0,
+                    350.0,
+                    1210.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    897.0,
+                    1176.0,
+                    938.0,
+                    1176.0,
+                    938.0,
+                    1208.0,
+                    897.0,
+                    1208.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    325.0,
+                    1215.0,
+                    387.0,
+                    1215.0,
+                    387.0,
+                    1328.0,
+                    325.0,
+                    1328.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    871.0,
+                    1213.0,
+                    938.0,
+                    1213.0,
+                    938.0,
+                    1335.0,
+                    871.0,
+                    1335.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    899.0,
+                    1303.0,
+                    938.0,
+                    1303.0,
+                    938.0,
+                    1338.0,
+                    899.0,
+                    1338.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    350.0,
+                    1369.0,
+                    389.0,
+                    1369.0,
+                    389.0,
+                    1402.0,
+                    350.0,
+                    1402.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    774.0,
+                    1381.0,
+                    825.0,
+                    1381.0,
+                    825.0,
+                    1416.0,
+                    774.0,
+                    1416.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    899.0,
+                    1369.0,
+                    938.0,
+                    1369.0,
+                    938.0,
+                    1403.0,
+                    899.0,
+                    1403.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1326.0,
+                    1394.0,
+                    1373.0,
+                    1394.0,
+                    1373.0,
+                    1423.0,
+                    1326.0,
+                    1423.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    422.0,
+                    1419.0,
+                    449.0,
+                    1419.0,
+                    449.0,
+                    1445.0,
+                    422.0,
+                    1445.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    518.0,
+                    1417.0,
+                    547.0,
+                    1417.0,
+                    547.0,
+                    1445.0,
+                    518.0,
+                    1445.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    618.0,
+                    1419.0,
+                    645.0,
+                    1419.0,
+                    645.0,
+                    1442.0,
+                    618.0,
+                    1442.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    706.0,
+                    1412.0,
+                    803.0,
+                    1412.0,
+                    803.0,
+                    1448.0,
+                    706.0,
+                    1448.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    973.0,
+                    1419.0,
+                    999.0,
+                    1419.0,
+                    999.0,
+                    1445.0,
+                    973.0,
+                    1445.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1070.0,
+                    1419.0,
+                    1097.0,
+                    1419.0,
+                    1097.0,
+                    1442.0,
+                    1070.0,
+                    1442.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1166.0,
+                    1417.0,
+                    1195.0,
+                    1417.0,
+                    1195.0,
+                    1445.0,
+                    1166.0,
+                    1445.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1257.0,
+                    1412.0,
+                    1353.0,
+                    1412.0,
+                    1353.0,
+                    1448.0,
+                    1257.0,
+                    1448.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    495.0,
+                    1432.0,
+                    620.0,
+                    1432.0,
+                    620.0,
+                    1473.0,
+                    495.0,
+                    1473.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1046.0,
+                    1432.0,
+                    1171.0,
+                    1432.0,
+                    1171.0,
+                    1473.0,
+                    1046.0,
+                    1473.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    895.75,
+                    335.0,
+                    939.75,
+                    335.0,
+                    939.75,
+                    367.0,
+                    895.75,
+                    367.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    337.0,
+                    337.5,
+                    384.0,
+                    337.5,
+                    384.0,
+                    364.5,
+                    337.0,
+                    364.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    332.0,
+                    415.0,
+                    387.0,
+                    415.0,
+                    387.0,
+                    442.0,
+                    332.0,
+                    442.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    888.0,
+                    415.0,
+                    945.0,
+                    415.0,
+                    945.0,
+                    442.5,
+                    888.0,
+                    442.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1161.0,
+                    618.5,
+                    1209.0,
+                    618.5,
+                    1209.0,
+                    650.5,
+                    1161.0,
+                    650.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    343.0,
+                    813.5,
+                    392.0,
+                    813.5,
+                    392.0,
+                    839.5,
+                    343.0,
+                    839.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    892.0,
+                    810.5,
+                    944.0,
+                    810.5,
+                    944.0,
+                    842.0,
+                    892.0,
+                    842.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    344.0,
+                    858.5,
+                    391.0,
+                    858.5,
+                    391.0,
+                    885.5,
+                    344.0,
+                    885.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    890.25,
+                    856.5,
+                    945.25,
+                    856.5,
+                    945.25,
+                    887.5,
+                    890.25,
+                    887.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    776.0,
+                    1000.0,
+                    822.0,
+                    1000.0,
+                    822.0,
+                    1029.0,
+                    776.0,
+                    1029.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    346.0,
+                    1112.5,
+                    390.0,
+                    1112.5,
+                    390.0,
+                    1143.0,
+                    346.0,
+                    1143.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    345.25,
+                    1304.0,
+                    389.25,
+                    1304.0,
+                    389.25,
+                    1336.0,
+                    345.25,
+                    1336.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1496.0,
+                    1403.0,
+                    1496.0,
+                    1403.0,
+                    1532.0,
+                    295.0,
+                    1532.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1524.0,
+                    1405.0,
+                    1524.0,
+                    1405.0,
+                    1561.0,
+                    293.0,
+                    1561.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1550.0,
+                    1405.0,
+                    1550.0,
+                    1405.0,
+                    1590.0,
+                    293.0,
+                    1590.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1580.0,
+                    1405.0,
+                    1580.0,
+                    1405.0,
+                    1617.0,
+                    293.0,
+                    1617.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1609.0,
+                    1357.0,
+                    1609.0,
+                    1357.0,
+                    1643.0,
+                    294.0,
+                    1643.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1710.0,
+                    998.0,
+                    1710.0,
+                    998.0,
+                    1745.0,
+                    295.0,
+                    1745.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    298.0,
+                    73.0,
+                    858.0,
+                    73.0,
+                    858.0,
+                    108.0,
+                    298.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    839.0,
+                    2085.0,
+                    861.0,
+                    2085.0,
+                    861.0,
+                    2120.0,
+                    839.0,
+                    2120.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    73.0,
+                    859.0,
+                    73.0,
+                    859.0,
+                    108.0,
+                    297.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1771.0,
+                    1405.0,
+                    1771.0,
+                    1405.0,
+                    1809.0,
+                    294.0,
+                    1809.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1805.0,
+                    1404.0,
+                    1805.0,
+                    1404.0,
+                    1837.0,
+                    296.0,
+                    1837.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1834.0,
+                    1405.0,
+                    1834.0,
+                    1405.0,
+                    1870.0,
+                    294.0,
+                    1870.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1865.0,
+                    608.0,
+                    1865.0,
+                    608.0,
+                    1899.0,
+                    294.0,
+                    1899.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1033.0,
+                    1865.0,
+                    1405.0,
+                    1865.0,
+                    1405.0,
+                    1899.0,
+                    1033.0,
+                    1899.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1896.0,
+                    1110.0,
+                    1896.0,
+                    1110.0,
+                    1928.0,
+                    296.0,
+                    1928.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1296.0,
+                    1896.0,
+                    1404.0,
+                    1896.0,
+                    1404.0,
+                    1928.0,
+                    1296.0,
+                    1928.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1924.0,
+                    896.0,
+                    1924.0,
+                    896.0,
+                    1959.0,
+                    294.0,
+                    1959.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1971.0,
+                    1405.0,
+                    1971.0,
+                    1405.0,
+                    2007.0,
+                    294.0,
+                    2007.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    2003.0,
+                    1403.0,
+                    2003.0,
+                    1403.0,
+                    2033.0,
+                    296.0,
+                    2033.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 4,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1711,
+                    1404,
+                    1711,
+                    1404,
+                    1957,
+                    298,
+                    1957
+                ],
+                "score": 0.981
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1426,
+                    1403,
+                    1426,
+                    1403,
+                    1610,
+                    298,
+                    1610
+                ],
+                "score": 0.98
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1197,
+                    1403,
+                    1197,
+                    1403,
+                    1411,
+                    298,
+                    1411
+                ],
+                "score": 0.978
+            },
+            {
+                "category_id": 5,
+                "poly": [
+                    299,
+                    336,
+                    1400,
+                    336,
+                    1400,
+                    603,
+                    299,
+                    603
+                ],
+                "score": 0.977,
+                "html": "<table><tr><td>Dataset</td><td>Task</td><td>Architecture</td><td>Training Strategy</td><td>BS range</td></tr><tr><td> MNIST</td><td>IC</td><td>ResNet34</td><td>BLR,LSR (no = 0.1, W = 10, E = 200)</td><td>26 -214</td></tr><tr><td>CIFAR-10</td><td>IC</td><td>AlexNet, MobileNetV2 ResNet34, VGG16</td><td>BLR,LSR, SRSR (mo = 0.1,W=10,E= 200)</td><td>26-214</td></tr><tr><td>CIFAR-100</td><td>IC</td><td>ResNet34</td><td>BLR,LSR (no = 0.1,W = 10, E = 200)</td><td>2 -214</td></tr><tr><td>SVHN</td><td>IC</td><td>ResNet34</td><td>BLR,LSR (no =0.1,W =10,E= 200)</td><td>26-214</td></tr><tr><td> WikiText-2</td><td> NLP</td><td>LSTM</td><td>BLR,LSR (no = 20,W = 3,E = 40)</td><td>2³ -210</td></tr><tr><td>Cityscapes</td><td>IS</td><td>DRN-D-22</td><td>BLR,LSR (no = 0.01,W = 10,E= 100)</td><td>23 -211</td></tr></table>"
+            },
+            {
+                "category_id": 3,
+                "poly": [
+                    333,
+                    658,
+                    1368,
+                    658,
+                    1368,
+                    1017,
+                    333,
+                    1017
+                ],
+                "score": 0.963
+            },
+            {
+                "category_id": 4,
+                "poly": [
+                    298,
+                    1048,
+                    1403,
+                    1048,
+                    1403,
+                    1133,
+                    298,
+                    1133
+                ],
+                "score": 0.961
+            },
+            {
+                "category_id": 6,
+                "poly": [
+                    297,
+                    226,
+                    1403,
+                    226,
+                    1403,
+                    309,
+                    297,
+                    309
+                ],
+                "score": 0.939
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1972,
+                    1399,
+                    1972,
+                    1399,
+                    2034,
+                    297,
+                    2034
+                ],
+                "score": 0.936
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    298,
+                    76,
+                    857,
+                    76,
+                    857,
+                    104,
+                    298,
+                    104
+                ],
+                "score": 0.893
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    301,
+                    1653,
+                    1124,
+                    1653,
+                    1124,
+                    1684,
+                    301,
+                    1684
+                ],
+                "score": 0.87
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    840,
+                    2089,
+                    858,
+                    2089,
+                    858,
+                    2112,
+                    840,
+                    2112
+                ],
+                "score": 0.78
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    870,
+                    1288,
+                    1105,
+                    1288,
+                    1105,
+                    1322,
+                    870,
+                    1322
+                ],
+                "score": 0.92,
+                "latex": "s ( m ; \\bar { m _ { 0 } } ) \\ll \\bar { m } / m _ { 0 }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1156,
+                    255,
+                    1179,
+                    255,
+                    1179,
+                    279,
+                    1156,
+                    279
+                ],
+                "score": 0.82,
+                "latex": "E"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1249,
+                    232,
+                    1277,
+                    232,
+                    1277,
+                    256,
+                    1249,
+                    256
+                ],
+                "score": 0.81,
+                "latex": "\\eta _ { 0 }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    940,
+                    477,
+                    1244,
+                    477,
+                    1244,
+                    505,
+                    940,
+                    505
+                ],
+                "score": 0.79,
+                "latex": "( \\eta _ { 0 } = 0 . 1 , W = 1 0 , E = 2 0 0 )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    948,
+                    510,
+                    1241,
+                    510,
+                    1241,
+                    533,
+                    948,
+                    533
+                ],
+                "score": 0.77,
+                "latex": "\\eta _ { 0 } = 0 . 1 , W = 1 0 , E = 2 0 0"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    932,
+                    569,
+                    1252,
+                    569,
+                    1252,
+                    594,
+                    932,
+                    594
+                ],
+                "score": 0.74,
+                "latex": "( \\eta _ { 0 } = 0 . 0 1 , W = 1 0 , E = 1 0 0 )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    950,
+                    540,
+                    1241,
+                    540,
+                    1241,
+                    564,
+                    950,
+                    564
+                ],
+                "score": 0.67,
+                "latex": "( \\eta _ { 0 } = 2 0 , W = 3 , E = 4 0 )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1291,
+                    537,
+                    1375,
+                    537,
+                    1375,
+                    560,
+                    1291,
+                    560
+                ],
+                "score": 0.59,
+                "latex": "2 ^ { 3 } - 2 ^ { 1 0 }"
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    880,
+                    445,
+                    1194,
+                    445,
+                    1194,
+                    474,
+                    880,
+                    474
+                ],
+                "score": 0.58,
+                "latex": "( \\eta _ { 0 } = 0 . 1 , W = 1 0 , E = 2 0 0 )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    882,
+                    445,
+                    1194,
+                    445,
+                    1194,
+                    474,
+                    882,
+                    474
+                ],
+                "score": 0.45,
+                "latex": "( \\eta _ { 0 } = 0 . 1 , W = 1 0 , E = 2 0 0 )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1291,
+                    477,
+                    1375,
+                    477,
+                    1375,
+                    501,
+                    1291,
+                    501
+                ],
+                "score": 0.45,
+                "latex": "2 ^ { 6 } - 2 ^ { 1 4 }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1290,
+                    507,
+                    1375,
+                    507,
+                    1375,
+                    530,
+                    1290,
+                    530
+                ],
+                "score": 0.35,
+                "latex": "2 ^ { 6 } - 2 ^ { 1 4 }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    436,
+                    254,
+                    467,
+                    254,
+                    467,
+                    280,
+                    436,
+                    280
+                ],
+                "score": 0.33,
+                "latex": "W"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1291,
+                    566,
+                    1374,
+                    566,
+                    1374,
+                    591,
+                    1291,
+                    591
+                ],
+                "score": 0.27,
+                "latex": "2 ^ { 3 } - 2 ^ { 1 1 }"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    432.0,
+                    666.0,
+                    500.0,
+                    666.0,
+                    500.0,
+                    689.0,
+                    432.0,
+                    689.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    355.0,
+                    675.0,
+                    380.0,
+                    675.0,
+                    380.0,
+                    700.0,
+                    355.0,
+                    700.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    431.0,
+                    683.0,
+                    548.0,
+                    683.0,
+                    548.0,
+                    721.0,
+                    431.0,
+                    721.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    429.0,
+                    708.0,
+                    588.0,
+                    708.0,
+                    588.0,
+                    743.0,
+                    429.0,
+                    743.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    940.0,
+                    721.0,
+                    978.0,
+                    721.0,
+                    978.0,
+                    974.0,
+                    940.0,
+                    974.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    431.0,
+                    738.0,
+                    479.0,
+                    738.0,
+                    479.0,
+                    763.0,
+                    431.0,
+                    763.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    357.0,
+                    741.0,
+                    378.0,
+                    741.0,
+                    378.0,
+                    762.0,
+                    357.0,
+                    762.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    330.0,
+                    742.0,
+                    368.0,
+                    742.0,
+                    368.0,
+                    892.0,
+                    330.0,
+                    892.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    976.0,
+                    743.0,
+                    1004.0,
+                    743.0,
+                    1004.0,
+                    767.0,
+                    976.0,
+                    767.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1323.0,
+                    755.0,
+                    1345.0,
+                    755.0,
+                    1345.0,
+                    774.0,
+                    1323.0,
+                    774.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1337.0,
+                    762.0,
+                    1367.0,
+                    762.0,
+                    1367.0,
+                    937.0,
+                    1337.0,
+                    937.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    969.0,
+                    800.0,
+                    1003.0,
+                    800.0,
+                    1003.0,
+                    823.0,
+                    969.0,
+                    823.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1323.0,
+                    800.0,
+                    1346.0,
+                    800.0,
+                    1346.0,
+                    823.0,
+                    1323.0,
+                    823.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    356.0,
+                    805.0,
+                    378.0,
+                    805.0,
+                    378.0,
+                    826.0,
+                    356.0,
+                    826.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1323.0,
+                    848.0,
+                    1344.0,
+                    848.0,
+                    1344.0,
+                    867.0,
+                    1323.0,
+                    867.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    968.0,
+                    857.0,
+                    995.0,
+                    857.0,
+                    995.0,
+                    879.0,
+                    968.0,
+                    879.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    990.0,
+                    859.0,
+                    999.0,
+                    859.0,
+                    999.0,
+                    869.0,
+                    990.0,
+                    869.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    356.0,
+                    869.0,
+                    378.0,
+                    869.0,
+                    378.0,
+                    889.0,
+                    356.0,
+                    889.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    642.0,
+                    885.0,
+                    652.0,
+                    885.0,
+                    652.0,
+                    894.0,
+                    642.0,
+                    894.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1325.0,
+                    895.0,
+                    1345.0,
+                    895.0,
+                    1345.0,
+                    915.0,
+                    1325.0,
+                    915.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    969.0,
+                    914.0,
+                    994.0,
+                    914.0,
+                    994.0,
+                    933.0,
+                    969.0,
+                    933.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    357.0,
+                    932.0,
+                    380.0,
+                    932.0,
+                    380.0,
+                    955.0,
+                    357.0,
+                    955.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1323.0,
+                    943.0,
+                    1345.0,
+                    943.0,
+                    1345.0,
+                    963.0,
+                    1323.0,
+                    963.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    391.0,
+                    976.0,
+                    411.0,
+                    976.0,
+                    411.0,
+                    995.0,
+                    391.0,
+                    995.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    439.0,
+                    975.0,
+                    460.0,
+                    975.0,
+                    460.0,
+                    995.0,
+                    439.0,
+                    995.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    488.0,
+                    975.0,
+                    510.0,
+                    975.0,
+                    510.0,
+                    995.0,
+                    488.0,
+                    995.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    537.0,
+                    975.0,
+                    560.0,
+                    975.0,
+                    560.0,
+                    995.0,
+                    537.0,
+                    995.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    579.0,
+                    974.0,
+                    611.0,
+                    974.0,
+                    611.0,
+                    994.0,
+                    579.0,
+                    994.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    631.0,
+                    974.0,
+                    660.0,
+                    974.0,
+                    660.0,
+                    996.0,
+                    631.0,
+                    996.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    964.0,
+                    968.0,
+                    1005.0,
+                    968.0,
+                    1005.0,
+                    989.0,
+                    964.0,
+                    989.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1050.0,
+                    977.0,
+                    1073.0,
+                    977.0,
+                    1073.0,
+                    1000.0,
+                    1050.0,
+                    1000.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1133.0,
+                    976.0,
+                    1153.0,
+                    976.0,
+                    1153.0,
+                    998.0,
+                    1133.0,
+                    998.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    547.0,
+                    990.0,
+                    646.0,
+                    990.0,
+                    646.0,
+                    1021.0,
+                    547.0,
+                    1021.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1114.0,
+                    988.0,
+                    1215.0,
+                    988.0,
+                    1215.0,
+                    1022.0,
+                    1114.0,
+                    1022.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    507.0,
+                    802.0,
+                    800.0,
+                    802.0,
+                    800.0,
+                    818.5,
+                    507.0,
+                    818.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    675.0,
+                    972.0,
+                    714.0,
+                    972.0,
+                    714.0,
+                    996.0,
+                    675.0,
+                    996.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    725.0,
+                    972.5,
+                    762.0,
+                    972.5,
+                    762.0,
+                    995.5,
+                    725.0,
+                    995.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    774.0,
+                    973.0,
+                    811.0,
+                    973.0,
+                    811.0,
+                    994.5,
+                    774.0,
+                    994.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1206.0,
+                    974.5,
+                    1244.0,
+                    974.5,
+                    1244.0,
+                    997.5,
+                    1206.0,
+                    997.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1289.0,
+                    975.5,
+                    1322.0,
+                    975.5,
+                    1322.0,
+                    998.0,
+                    1289.0,
+                    998.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1047.0,
+                    1404.0,
+                    1047.0,
+                    1404.0,
+                    1082.0,
+                    294.0,
+                    1082.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1074.0,
+                    1406.0,
+                    1074.0,
+                    1406.0,
+                    1109.0,
+                    292.0,
+                    1109.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1104.0,
+                    955.0,
+                    1104.0,
+                    955.0,
+                    1136.0,
+                    293.0,
+                    1136.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    223.0,
+                    1248.0,
+                    223.0,
+                    1248.0,
+                    258.0,
+                    294.0,
+                    258.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1278.0,
+                    223.0,
+                    1405.0,
+                    223.0,
+                    1405.0,
+                    258.0,
+                    1278.0,
+                    258.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    253.0,
+                    435.0,
+                    253.0,
+                    435.0,
+                    285.0,
+                    293.0,
+                    285.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    468.0,
+                    253.0,
+                    1155.0,
+                    253.0,
+                    1155.0,
+                    285.0,
+                    468.0,
+                    285.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1180.0,
+                    253.0,
+                    1407.0,
+                    253.0,
+                    1407.0,
+                    285.0,
+                    1180.0,
+                    285.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    282.0,
+                    451.0,
+                    282.0,
+                    451.0,
+                    311.0,
+                    292.0,
+                    311.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    298.0,
+                    73.0,
+                    858.0,
+                    73.0,
+                    858.0,
+                    108.0,
+                    298.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1651.0,
+                    1127.0,
+                    1651.0,
+                    1127.0,
+                    1688.0,
+                    294.0,
+                    1688.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    840.0,
+                    2087.0,
+                    861.0,
+                    2087.0,
+                    861.0,
+                    2118.0,
+                    840.0,
+                    2118.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1709.0,
+                    1405.0,
+                    1709.0,
+                    1405.0,
+                    1750.0,
+                    293.0,
+                    1750.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1743.0,
+                    1406.0,
+                    1743.0,
+                    1406.0,
+                    1776.0,
+                    294.0,
+                    1776.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1773.0,
+                    1404.0,
+                    1773.0,
+                    1404.0,
+                    1806.0,
+                    293.0,
+                    1806.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    1805.0,
+                    1405.0,
+                    1805.0,
+                    1405.0,
+                    1835.0,
+                    297.0,
+                    1835.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1833.0,
+                    1406.0,
+                    1833.0,
+                    1406.0,
+                    1869.0,
+                    292.0,
+                    1869.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1864.0,
+                    1406.0,
+                    1864.0,
+                    1406.0,
+                    1898.0,
+                    293.0,
+                    1898.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1896.0,
+                    1404.0,
+                    1896.0,
+                    1404.0,
+                    1927.0,
+                    296.0,
+                    1927.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1927.0,
+                    1404.0,
+                    1927.0,
+                    1404.0,
+                    1960.0,
+                    296.0,
+                    1960.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1425.0,
+                    1405.0,
+                    1425.0,
+                    1405.0,
+                    1461.0,
+                    293.0,
+                    1461.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1457.0,
+                    1404.0,
+                    1457.0,
+                    1404.0,
+                    1493.0,
+                    294.0,
+                    1493.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1489.0,
+                    1404.0,
+                    1489.0,
+                    1404.0,
+                    1521.0,
+                    294.0,
+                    1521.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1518.0,
+                    1405.0,
+                    1518.0,
+                    1405.0,
+                    1554.0,
+                    296.0,
+                    1554.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1545.0,
+                    1405.0,
+                    1545.0,
+                    1405.0,
+                    1586.0,
+                    293.0,
+                    1586.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1581.0,
+                    982.0,
+                    1581.0,
+                    982.0,
+                    1613.0,
+                    296.0,
+                    1613.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1197.0,
+                    1405.0,
+                    1197.0,
+                    1405.0,
+                    1230.0,
+                    293.0,
+                    1230.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1228.0,
+                    1405.0,
+                    1228.0,
+                    1405.0,
+                    1261.0,
+                    292.0,
+                    1261.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1259.0,
+                    1405.0,
+                    1259.0,
+                    1405.0,
+                    1293.0,
+                    294.0,
+                    1293.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1289.0,
+                    869.0,
+                    1289.0,
+                    869.0,
+                    1324.0,
+                    294.0,
+                    1324.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1106.0,
+                    1289.0,
+                    1405.0,
+                    1289.0,
+                    1405.0,
+                    1324.0,
+                    1106.0,
+                    1324.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1319.0,
+                    1405.0,
+                    1319.0,
+                    1405.0,
+                    1353.0,
+                    294.0,
+                    1353.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1349.0,
+                    1405.0,
+                    1349.0,
+                    1405.0,
+                    1382.0,
+                    292.0,
+                    1382.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1380.0,
+                    1345.0,
+                    1380.0,
+                    1345.0,
+                    1414.0,
+                    296.0,
+                    1414.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1971.0,
+                    1404.0,
+                    1971.0,
+                    1404.0,
+                    2007.0,
+                    295.0,
+                    2007.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    2002.0,
+                    1404.0,
+                    2002.0,
+                    1404.0,
+                    2037.0,
+                    295.0,
+                    2037.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 5,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1360,
+                    1404,
+                    1360,
+                    1404,
+                    1726,
+                    297,
+                    1726
+                ],
+                "score": 0.983
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    775,
+                    1404,
+                    775,
+                    1404,
+                    1031,
+                    297,
+                    1031
+                ],
+                "score": 0.981
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1131,
+                    1404,
+                    1131,
+                    1404,
+                    1344,
+                    298,
+                    1344
+                ],
+                "score": 0.981
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1742,
+                    1403,
+                    1742,
+                    1403,
+                    1926,
+                    298,
+                    1926
+                ],
+                "score": 0.979
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    300,
+                    1943,
+                    1401,
+                    1943,
+                    1401,
+                    2034,
+                    300,
+                    2034
+                ],
+                "score": 0.971
+            },
+            {
+                "category_id": 4,
+                "poly": [
+                    296,
+                    601,
+                    1405,
+                    601,
+                    1405,
+                    713,
+                    296,
+                    713
+                ],
+                "score": 0.961
+            },
+            {
+                "category_id": 3,
+                "poly": [
+                    403,
+                    252,
+                    1294,
+                    252,
+                    1294,
+                    571,
+                    403,
+                    571
+                ],
+                "score": 0.958
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    300,
+                    1074,
+                    1168,
+                    1074,
+                    1168,
+                    1103,
+                    300,
+                    1103
+                ],
+                "score": 0.843
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    298,
+                    76,
+                    857,
+                    76,
+                    857,
+                    104,
+                    298,
+                    104
+                ],
+                "score": 0.818
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    842,
+                    2088,
+                    858,
+                    2088,
+                    858,
+                    2111,
+                    842,
+                    2111
+                ],
+                "score": 0.646
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    842,
+                    2088,
+                    859,
+                    2088,
+                    859,
+                    2111,
+                    842,
+                    2111
+                ],
+                "score": 0.229
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    300,
+                    76,
+                    855,
+                    76,
+                    855,
+                    104,
+                    300,
+                    104
+                ],
+                "score": 0.201
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    750,
+                    809,
+                    1176,
+                    809,
+                    1176,
+                    847,
+                    750,
+                    847
+                ],
+                "score": 0.93,
+                "latex": "\\hat { L } ( \\mathbf { w } ) = L ( \\mathbf { w } ^ { * } ) + \\langle \\mathbf { g } _ { m } ( \\mathbf { w } ^ { * } ) , \\mathbf { w } - \\mathbf { w } ^ { * } \\rangle"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    480,
+                    844,
+                    755,
+                    844,
+                    755,
+                    882,
+                    480,
+                    882
+                ],
+                "score": 0.92,
+                "latex": "\\mathbf { w } = \\mathbf { w } ^ { * } - \\eta \\frac { m } { m _ { 0 } } \\mathbf { g } _ { m } \\big ( \\mathbf { w } ^ { * } \\big )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1286,
+                    777,
+                    1352,
+                    777,
+                    1352,
+                    811,
+                    1286,
+                    811
+                ],
+                "score": 0.9,
+                "latex": "L ( \\mathbf { w } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    789,
+                    1804,
+                    836,
+                    1804,
+                    836,
+                    1833,
+                    789,
+                    1833
+                ],
+                "score": 0.85,
+                "latex": "5 0 k"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    600,
+                    815,
+                    638,
+                    815,
+                    638,
+                    842,
+                    600,
+                    842
+                ],
+                "score": 0.82,
+                "latex": "\\mathbf { w } ^ { * }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    391,
+                    779,
+                    429,
+                    779,
+                    429,
+                    806,
+                    391,
+                    806
+                ],
+                "score": 0.81,
+                "latex": "\\mathbf { w } ^ { * }"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    599.0,
+                    1405.0,
+                    599.0,
+                    1405.0,
+                    636.0,
+                    295.0,
+                    636.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    628.0,
+                    1403.0,
+                    628.0,
+                    1403.0,
+                    662.0,
+                    293.0,
+                    662.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    653.0,
+                    1405.0,
+                    653.0,
+                    1405.0,
+                    691.0,
+                    294.0,
+                    691.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    685.0,
+                    550.0,
+                    685.0,
+                    550.0,
+                    718.0,
+                    295.0,
+                    718.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    441.0,
+                    253.0,
+                    773.0,
+                    253.0,
+                    773.0,
+                    276.0,
+                    441.0,
+                    276.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    940.0,
+                    252.0,
+                    1279.0,
+                    252.0,
+                    1279.0,
+                    276.0,
+                    940.0,
+                    276.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    425.0,
+                    278.0,
+                    452.0,
+                    278.0,
+                    452.0,
+                    302.0,
+                    425.0,
+                    302.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    928.0,
+                    278.0,
+                    954.0,
+                    278.0,
+                    954.0,
+                    303.0,
+                    928.0,
+                    303.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    496.0,
+                    293.0,
+                    682.0,
+                    293.0,
+                    682.0,
+                    322.0,
+                    496.0,
+                    322.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    999.0,
+                    297.0,
+                    1082.0,
+                    297.0,
+                    1082.0,
+                    320.0,
+                    999.0,
+                    320.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    498.0,
+                    319.0,
+                    681.0,
+                    319.0,
+                    681.0,
+                    344.0,
+                    498.0,
+                    344.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    999.0,
+                    320.0,
+                    1075.0,
+                    320.0,
+                    1075.0,
+                    343.0,
+                    999.0,
+                    343.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    403.0,
+                    332.0,
+                    448.0,
+                    332.0,
+                    448.0,
+                    482.0,
+                    403.0,
+                    482.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    496.0,
+                    341.0,
+                    688.0,
+                    341.0,
+                    688.0,
+                    366.0,
+                    496.0,
+                    366.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    904.0,
+                    332.0,
+                    950.0,
+                    332.0,
+                    950.0,
+                    483.0,
+                    904.0,
+                    483.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    997.0,
+                    342.0,
+                    1054.0,
+                    342.0,
+                    1054.0,
+                    365.0,
+                    997.0,
+                    365.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    496.0,
+                    363.0,
+                    720.0,
+                    363.0,
+                    720.0,
+                    387.0,
+                    496.0,
+                    387.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    999.0,
+                    365.0,
+                    1057.0,
+                    365.0,
+                    1057.0,
+                    385.0,
+                    999.0,
+                    385.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    427.0,
+                    384.0,
+                    449.0,
+                    384.0,
+                    449.0,
+                    407.0,
+                    427.0,
+                    407.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    458.0,
+                    385.0,
+                    541.0,
+                    385.0,
+                    541.0,
+                    409.0,
+                    458.0,
+                    409.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    930.0,
+                    384.0,
+                    951.0,
+                    384.0,
+                    951.0,
+                    406.0,
+                    930.0,
+                    406.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    963.0,
+                    390.0,
+                    997.0,
+                    390.0,
+                    997.0,
+                    403.0,
+                    963.0,
+                    403.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    998.0,
+                    385.0,
+                    1044.0,
+                    385.0,
+                    1044.0,
+                    409.0,
+                    998.0,
+                    409.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    428.0,
+                    440.0,
+                    447.0,
+                    440.0,
+                    447.0,
+                    457.0,
+                    428.0,
+                    457.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    930.0,
+                    438.0,
+                    951.0,
+                    438.0,
+                    951.0,
+                    459.0,
+                    930.0,
+                    459.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    427.0,
+                    491.0,
+                    448.0,
+                    491.0,
+                    448.0,
+                    513.0,
+                    427.0,
+                    513.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    929.0,
+                    491.0,
+                    952.0,
+                    491.0,
+                    952.0,
+                    513.0,
+                    929.0,
+                    513.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    437.0,
+                    528.0,
+                    459.0,
+                    528.0,
+                    459.0,
+                    550.0,
+                    437.0,
+                    550.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    518.0,
+                    527.0,
+                    543.0,
+                    527.0,
+                    543.0,
+                    553.0,
+                    518.0,
+                    553.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    597.0,
+                    526.0,
+                    630.0,
+                    526.0,
+                    630.0,
+                    554.0,
+                    597.0,
+                    554.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1021.0,
+                    527.0,
+                    1046.0,
+                    527.0,
+                    1046.0,
+                    553.0,
+                    1021.0,
+                    553.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1099.0,
+                    526.0,
+                    1131.0,
+                    526.0,
+                    1131.0,
+                    554.0,
+                    1099.0,
+                    554.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1265.0,
+                    525.0,
+                    1297.0,
+                    525.0,
+                    1297.0,
+                    554.0,
+                    1265.0,
+                    554.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    562.0,
+                    542.0,
+                    664.0,
+                    542.0,
+                    664.0,
+                    574.0,
+                    562.0,
+                    574.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1066.0,
+                    542.0,
+                    1167.0,
+                    542.0,
+                    1167.0,
+                    574.0,
+                    1066.0,
+                    574.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    718.0,
+                    327.5,
+                    785.0,
+                    327.5,
+                    785.0,
+                    339.5,
+                    718.0,
+                    339.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1172.75,
+                    341.0,
+                    1289.75,
+                    341.0,
+                    1289.75,
+                    356.5,
+                    1172.75,
+                    356.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1055.0,
+                    432.5,
+                    1120.0,
+                    432.5,
+                    1120.0,
+                    446.5,
+                    1055.0,
+                    446.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    675.75,
+                    526.0,
+                    714.75,
+                    526.0,
+                    714.75,
+                    552.0,
+                    675.75,
+                    552.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    759.25,
+                    525.0,
+                    797.25,
+                    525.0,
+                    797.25,
+                    552.0,
+                    759.25,
+                    552.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1176.0,
+                    525.5,
+                    1219.0,
+                    525.5,
+                    1219.0,
+                    550.0,
+                    1176.0,
+                    550.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    931.0,
+                    528.0,
+                    973.0,
+                    528.0,
+                    973.0,
+                    541.5,
+                    931.0,
+                    541.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1069.0,
+                    1175.0,
+                    1069.0,
+                    1175.0,
+                    1107.0,
+                    293.0,
+                    1107.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    298.0,
+                    73.0,
+                    858.0,
+                    73.0,
+                    858.0,
+                    108.0,
+                    298.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    840.0,
+                    2086.0,
+                    860.0,
+                    2086.0,
+                    860.0,
+                    2119.0,
+                    840.0,
+                    2119.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    840.0,
+                    2087.0,
+                    861.0,
+                    2087.0,
+                    861.0,
+                    2119.0,
+                    840.0,
+                    2119.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    73.0,
+                    859.0,
+                    73.0,
+                    859.0,
+                    108.0,
+                    297.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1359.0,
+                    1404.0,
+                    1359.0,
+                    1404.0,
+                    1395.0,
+                    292.0,
+                    1395.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1393.0,
+                    1405.0,
+                    1393.0,
+                    1405.0,
+                    1424.0,
+                    292.0,
+                    1424.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1422.0,
+                    1405.0,
+                    1422.0,
+                    1405.0,
+                    1456.0,
+                    295.0,
+                    1456.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1453.0,
+                    1404.0,
+                    1453.0,
+                    1404.0,
+                    1487.0,
+                    295.0,
+                    1487.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1482.0,
+                    1404.0,
+                    1482.0,
+                    1404.0,
+                    1516.0,
+                    294.0,
+                    1516.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1513.0,
+                    1404.0,
+                    1513.0,
+                    1404.0,
+                    1547.0,
+                    292.0,
+                    1547.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1544.0,
+                    1405.0,
+                    1544.0,
+                    1405.0,
+                    1578.0,
+                    294.0,
+                    1578.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1575.0,
+                    1405.0,
+                    1575.0,
+                    1405.0,
+                    1609.0,
+                    294.0,
+                    1609.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1604.0,
+                    1406.0,
+                    1604.0,
+                    1406.0,
+                    1639.0,
+                    292.0,
+                    1639.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1634.0,
+                    1404.0,
+                    1634.0,
+                    1404.0,
+                    1668.0,
+                    294.0,
+                    1668.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1664.0,
+                    1405.0,
+                    1664.0,
+                    1405.0,
+                    1702.0,
+                    294.0,
+                    1702.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1696.0,
+                    858.0,
+                    1696.0,
+                    858.0,
+                    1730.0,
+                    295.0,
+                    1730.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    777.0,
+                    390.0,
+                    777.0,
+                    390.0,
+                    812.0,
+                    295.0,
+                    812.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    430.0,
+                    777.0,
+                    1285.0,
+                    777.0,
+                    1285.0,
+                    812.0,
+                    430.0,
+                    812.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1353.0,
+                    777.0,
+                    1405.0,
+                    777.0,
+                    1405.0,
+                    812.0,
+                    1353.0,
+                    812.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    810.0,
+                    599.0,
+                    810.0,
+                    599.0,
+                    849.0,
+                    294.0,
+                    849.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    639.0,
+                    810.0,
+                    749.0,
+                    810.0,
+                    749.0,
+                    849.0,
+                    639.0,
+                    849.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1177.0,
+                    810.0,
+                    1405.0,
+                    810.0,
+                    1405.0,
+                    849.0,
+                    1177.0,
+                    849.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    282.0,
+                    828.0,
+                    479.0,
+                    828.0,
+                    479.0,
+                    898.0,
+                    282.0,
+                    898.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    756.0,
+                    828.0,
+                    1415.0,
+                    828.0,
+                    1415.0,
+                    898.0,
+                    756.0,
+                    898.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    876.0,
+                    1405.0,
+                    876.0,
+                    1405.0,
+                    914.0,
+                    294.0,
+                    914.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    908.0,
+                    1406.0,
+                    908.0,
+                    1406.0,
+                    944.0,
+                    292.0,
+                    944.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    939.0,
+                    1406.0,
+                    939.0,
+                    1406.0,
+                    973.0,
+                    294.0,
+                    973.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    968.0,
+                    1406.0,
+                    968.0,
+                    1406.0,
+                    1008.0,
+                    291.0,
+                    1008.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    999.0,
+                    1122.0,
+                    999.0,
+                    1122.0,
+                    1034.0,
+                    293.0,
+                    1034.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1134.0,
+                    1404.0,
+                    1134.0,
+                    1404.0,
+                    1164.0,
+                    294.0,
+                    1164.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1160.0,
+                    1406.0,
+                    1160.0,
+                    1406.0,
+                    1197.0,
+                    292.0,
+                    1197.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1192.0,
+                    1405.0,
+                    1192.0,
+                    1405.0,
+                    1226.0,
+                    292.0,
+                    1226.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1222.0,
+                    1405.0,
+                    1222.0,
+                    1405.0,
+                    1258.0,
+                    292.0,
+                    1258.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1248.0,
+                    1405.0,
+                    1248.0,
+                    1405.0,
+                    1291.0,
+                    292.0,
+                    1291.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1281.0,
+                    1405.0,
+                    1281.0,
+                    1405.0,
+                    1322.0,
+                    291.0,
+                    1322.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1312.0,
+                    546.0,
+                    1312.0,
+                    546.0,
+                    1346.0,
+                    295.0,
+                    1346.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1740.0,
+                    1405.0,
+                    1740.0,
+                    1405.0,
+                    1778.0,
+                    293.0,
+                    1778.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1774.0,
+                    1403.0,
+                    1774.0,
+                    1403.0,
+                    1806.0,
+                    294.0,
+                    1806.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1801.0,
+                    788.0,
+                    1801.0,
+                    788.0,
+                    1839.0,
+                    293.0,
+                    1839.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    837.0,
+                    1801.0,
+                    1405.0,
+                    1801.0,
+                    1405.0,
+                    1839.0,
+                    837.0,
+                    1839.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1833.0,
+                    1405.0,
+                    1833.0,
+                    1405.0,
+                    1871.0,
+                    293.0,
+                    1871.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1864.0,
+                    1405.0,
+                    1864.0,
+                    1405.0,
+                    1899.0,
+                    293.0,
+                    1899.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1896.0,
+                    1078.0,
+                    1896.0,
+                    1078.0,
+                    1928.0,
+                    296.0,
+                    1928.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1941.0,
+                    1405.0,
+                    1941.0,
+                    1405.0,
+                    1977.0,
+                    294.0,
+                    1977.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1973.0,
+                    1405.0,
+                    1973.0,
+                    1405.0,
+                    2006.0,
+                    296.0,
+                    2006.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    2002.0,
+                    1406.0,
+                    2002.0,
+                    1406.0,
+                    2039.0,
+                    294.0,
+                    2039.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 6,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    975,
+                    1404,
+                    975,
+                    1404,
+                    1220,
+                    298,
+                    1220
+                ],
+                "score": 0.979
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1333,
+                    1403,
+                    1333,
+                    1403,
+                    1518,
+                    298,
+                    1518
+                ],
+                "score": 0.978
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    805,
+                    1403,
+                    805,
+                    1403,
+                    959,
+                    298,
+                    959
+                ],
+                "score": 0.976
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    1671,
+                    1402,
+                    1671,
+                    1402,
+                    1795,
+                    299,
+                    1795
+                ],
+                "score": 0.973
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    1533,
+                    1403,
+                    1533,
+                    1403,
+                    1656,
+                    299,
+                    1656
+                ],
+                "score": 0.971
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    298,
+                    1949,
+                    1403,
+                    1949,
+                    1403,
+                    2034,
+                    298,
+                    2034
+                ],
+                "score": 0.953
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    293,
+                    728,
+                    1400,
+                    728,
+                    1400,
+                    791,
+                    293,
+                    791
+                ],
+                "score": 0.952
+            },
+            {
+                "category_id": 4,
+                "poly": [
+                    296,
+                    606,
+                    1404,
+                    606,
+                    1404,
+                    668,
+                    296,
+                    668
+                ],
+                "score": 0.947
+            },
+            {
+                "category_id": 3,
+                "poly": [
+                    430,
+                    251,
+                    1269,
+                    251,
+                    1269,
+                    577,
+                    430,
+                    577
+                ],
+                "score": 0.908
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    300,
+                    1264,
+                    543,
+                    1264,
+                    543,
+                    1299,
+                    300,
+                    1299
+                ],
+                "score": 0.894
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    840,
+                    2089,
+                    858,
+                    2089,
+                    858,
+                    2111,
+                    840,
+                    2111
+                ],
+                "score": 0.804
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    298,
+                    76,
+                    857,
+                    76,
+                    857,
+                    104,
+                    298,
+                    104
+                ],
+                "score": 0.745
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    299,
+                    76,
+                    854,
+                    76,
+                    854,
+                    103,
+                    299,
+                    103
+                ],
+                "score": 0.34
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    859,
+                    806,
+                    968,
+                    806,
+                    968,
+                    840,
+                    859,
+                    840
+                ],
+                "score": 0.93,
+                "latex": "s ( m ; m _ { 0 } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    589,
+                    636,
+                    656,
+                    636,
+                    656,
+                    666,
+                    589,
+                    666
+                ],
+                "score": 0.89,
+                "latex": "1 0 0 \\%"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    660,
+                    928,
+                    715,
+                    928,
+                    715,
+                    958,
+                    660,
+                    958
+                ],
+                "score": 0.88,
+                "latex": "2 5 \\%"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    494,
+                    928,
+                    548,
+                    928,
+                    548,
+                    959,
+                    494,
+                    959
+                ],
+                "score": 0.86,
+                "latex": "5 0 \\%"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    416,
+                    928,
+                    483,
+                    928,
+                    483,
+                    959,
+                    416,
+                    959
+                ],
+                "score": 0.86,
+                "latex": "1 0 0 \\%"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    781,
+                    898,
+                    842,
+                    898,
+                    842,
+                    927,
+                    781,
+                    927
+                ],
+                "score": 0.85,
+                "latex": "6 0 0 k"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    330.0,
+                    1943.0,
+                    1404.0,
+                    1943.0,
+                    1404.0,
+                    1983.0,
+                    330.0,
+                    1983.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1975.0,
+                    1407.0,
+                    1975.0,
+                    1407.0,
+                    2009.0,
+                    293.0,
+                    2009.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    2003.0,
+                    1350.0,
+                    2003.0,
+                    1350.0,
+                    2038.0,
+                    293.0,
+                    2038.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    605.0,
+                    1405.0,
+                    605.0,
+                    1405.0,
+                    641.0,
+                    295.0,
+                    641.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    634.0,
+                    588.0,
+                    634.0,
+                    588.0,
+                    672.0,
+                    294.0,
+                    672.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    657.0,
+                    634.0,
+                    1312.0,
+                    634.0,
+                    1312.0,
+                    672.0,
+                    657.0,
+                    672.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    462.0,
+                    251.0,
+                    816.0,
+                    251.0,
+                    816.0,
+                    278.0,
+                    462.0,
+                    278.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    921.0,
+                    251.0,
+                    1250.0,
+                    251.0,
+                    1250.0,
+                    278.0,
+                    921.0,
+                    278.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    453.0,
+                    286.0,
+                    482.0,
+                    286.0,
+                    482.0,
+                    316.0,
+                    453.0,
+                    316.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    526.0,
+                    297.0,
+                    658.0,
+                    297.0,
+                    658.0,
+                    321.0,
+                    526.0,
+                    321.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    901.0,
+                    287.0,
+                    927.0,
+                    287.0,
+                    927.0,
+                    314.0,
+                    901.0,
+                    314.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    972.0,
+                    297.0,
+                    1085.0,
+                    297.0,
+                    1085.0,
+                    322.0,
+                    972.0,
+                    322.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    527.0,
+                    321.0,
+                    651.0,
+                    321.0,
+                    651.0,
+                    343.0,
+                    527.0,
+                    343.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    972.0,
+                    319.0,
+                    1077.0,
+                    319.0,
+                    1077.0,
+                    345.0,
+                    972.0,
+                    345.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    430.0,
+                    338.0,
+                    479.0,
+                    338.0,
+                    479.0,
+                    489.0,
+                    430.0,
+                    489.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    526.0,
+                    343.0,
+                    650.0,
+                    343.0,
+                    650.0,
+                    367.0,
+                    526.0,
+                    367.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    878.0,
+                    339.0,
+                    910.0,
+                    339.0,
+                    910.0,
+                    488.0,
+                    878.0,
+                    488.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    973.0,
+                    343.0,
+                    1076.0,
+                    343.0,
+                    1076.0,
+                    367.0,
+                    973.0,
+                    367.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    456.0,
+                    388.0,
+                    478.0,
+                    388.0,
+                    478.0,
+                    411.0,
+                    456.0,
+                    411.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    905.0,
+                    389.0,
+                    922.0,
+                    389.0,
+                    922.0,
+                    409.0,
+                    905.0,
+                    409.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    456.0,
+                    438.0,
+                    477.0,
+                    438.0,
+                    477.0,
+                    460.0,
+                    456.0,
+                    460.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    903.0,
+                    439.0,
+                    923.0,
+                    439.0,
+                    923.0,
+                    460.0,
+                    903.0,
+                    460.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    454.0,
+                    486.0,
+                    479.0,
+                    486.0,
+                    479.0,
+                    511.0,
+                    454.0,
+                    511.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    902.0,
+                    487.0,
+                    923.0,
+                    487.0,
+                    923.0,
+                    509.0,
+                    902.0,
+                    509.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    465.0,
+                    535.0,
+                    487.0,
+                    535.0,
+                    487.0,
+                    557.0,
+                    465.0,
+                    557.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    539.0,
+                    535.0,
+                    561.0,
+                    535.0,
+                    561.0,
+                    557.0,
+                    539.0,
+                    557.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    614.0,
+                    533.0,
+                    638.0,
+                    533.0,
+                    638.0,
+                    558.0,
+                    614.0,
+                    558.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    685.0,
+                    532.0,
+                    717.0,
+                    532.0,
+                    717.0,
+                    559.0,
+                    685.0,
+                    559.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    760.0,
+                    534.0,
+                    791.0,
+                    534.0,
+                    791.0,
+                    558.0,
+                    760.0,
+                    558.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    913.0,
+                    535.0,
+                    933.0,
+                    535.0,
+                    933.0,
+                    556.0,
+                    913.0,
+                    556.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    987.0,
+                    535.0,
+                    1008.0,
+                    535.0,
+                    1008.0,
+                    557.0,
+                    987.0,
+                    557.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1060.0,
+                    533.0,
+                    1086.0,
+                    533.0,
+                    1086.0,
+                    558.0,
+                    1060.0,
+                    558.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1128.0,
+                    534.0,
+                    1162.0,
+                    534.0,
+                    1162.0,
+                    556.0,
+                    1128.0,
+                    556.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1204.0,
+                    533.0,
+                    1239.0,
+                    533.0,
+                    1239.0,
+                    557.0,
+                    1204.0,
+                    557.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    595.0,
+                    551.0,
+                    695.0,
+                    551.0,
+                    695.0,
+                    580.0,
+                    595.0,
+                    580.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1043.0,
+                    551.0,
+                    1142.0,
+                    551.0,
+                    1142.0,
+                    580.0,
+                    1043.0,
+                    580.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1190.0,
+                    316.5,
+                    1266.0,
+                    316.5,
+                    1266.0,
+                    330.0,
+                    1190.0,
+                    330.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    896.0,
+                    342.5,
+                    929.0,
+                    342.5,
+                    929.0,
+                    357.5,
+                    896.0,
+                    357.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    666.0,
+                    329.0,
+                    782.0,
+                    329.0,
+                    782.0,
+                    343.0,
+                    666.0,
+                    343.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1260.0,
+                    549.0,
+                    1260.0,
+                    549.0,
+                    1308.0,
+                    292.0,
+                    1308.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    836.0,
+                    2085.0,
+                    859.0,
+                    2085.0,
+                    859.0,
+                    2116.0,
+                    836.0,
+                    2116.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    298.0,
+                    73.0,
+                    858.0,
+                    73.0,
+                    858.0,
+                    108.0,
+                    298.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    73.0,
+                    859.0,
+                    73.0,
+                    859.0,
+                    107.0,
+                    297.0,
+                    107.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    974.0,
+                    1406.0,
+                    974.0,
+                    1406.0,
+                    1011.0,
+                    293.0,
+                    1011.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1006.0,
+                    1404.0,
+                    1006.0,
+                    1404.0,
+                    1040.0,
+                    293.0,
+                    1040.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1033.0,
+                    1405.0,
+                    1033.0,
+                    1405.0,
+                    1072.0,
+                    291.0,
+                    1072.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1068.0,
+                    1405.0,
+                    1068.0,
+                    1405.0,
+                    1101.0,
+                    294.0,
+                    1101.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1097.0,
+                    1405.0,
+                    1097.0,
+                    1405.0,
+                    1131.0,
+                    294.0,
+                    1131.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1127.0,
+                    1405.0,
+                    1127.0,
+                    1405.0,
+                    1161.0,
+                    293.0,
+                    1161.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1159.0,
+                    1406.0,
+                    1159.0,
+                    1406.0,
+                    1192.0,
+                    294.0,
+                    1192.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1190.0,
+                    1183.0,
+                    1190.0,
+                    1183.0,
+                    1222.0,
+                    293.0,
+                    1222.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1334.0,
+                    1405.0,
+                    1334.0,
+                    1405.0,
+                    1368.0,
+                    293.0,
+                    1368.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1364.0,
+                    1403.0,
+                    1364.0,
+                    1403.0,
+                    1396.0,
+                    296.0,
+                    1396.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1395.0,
+                    1405.0,
+                    1395.0,
+                    1405.0,
+                    1431.0,
+                    293.0,
+                    1431.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1423.0,
+                    1405.0,
+                    1423.0,
+                    1405.0,
+                    1461.0,
+                    293.0,
+                    1461.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1455.0,
+                    1407.0,
+                    1455.0,
+                    1407.0,
+                    1492.0,
+                    293.0,
+                    1492.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1487.0,
+                    786.0,
+                    1487.0,
+                    786.0,
+                    1522.0,
+                    296.0,
+                    1522.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    805.0,
+                    858.0,
+                    805.0,
+                    858.0,
+                    841.0,
+                    294.0,
+                    841.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    969.0,
+                    805.0,
+                    1405.0,
+                    805.0,
+                    1405.0,
+                    841.0,
+                    969.0,
+                    841.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    837.0,
+                    1404.0,
+                    837.0,
+                    1404.0,
+                    869.0,
+                    294.0,
+                    869.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    866.0,
+                    1405.0,
+                    866.0,
+                    1405.0,
+                    902.0,
+                    293.0,
+                    902.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    894.0,
+                    780.0,
+                    894.0,
+                    780.0,
+                    934.0,
+                    293.0,
+                    934.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    843.0,
+                    894.0,
+                    1407.0,
+                    894.0,
+                    1407.0,
+                    934.0,
+                    843.0,
+                    934.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    926.0,
+                    415.0,
+                    926.0,
+                    415.0,
+                    964.0,
+                    293.0,
+                    964.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    484.0,
+                    926.0,
+                    493.0,
+                    926.0,
+                    493.0,
+                    964.0,
+                    484.0,
+                    964.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    549.0,
+                    926.0,
+                    659.0,
+                    926.0,
+                    659.0,
+                    964.0,
+                    549.0,
+                    964.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    716.0,
+                    926.0,
+                    1045.0,
+                    926.0,
+                    1045.0,
+                    964.0,
+                    716.0,
+                    964.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1669.0,
+                    1406.0,
+                    1669.0,
+                    1406.0,
+                    1706.0,
+                    294.0,
+                    1706.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1703.0,
+                    1404.0,
+                    1703.0,
+                    1404.0,
+                    1735.0,
+                    294.0,
+                    1735.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1731.0,
+                    1406.0,
+                    1731.0,
+                    1406.0,
+                    1768.0,
+                    292.0,
+                    1768.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1762.0,
+                    536.0,
+                    1762.0,
+                    536.0,
+                    1798.0,
+                    293.0,
+                    1798.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1533.0,
+                    1404.0,
+                    1533.0,
+                    1404.0,
+                    1569.0,
+                    294.0,
+                    1569.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1564.0,
+                    1405.0,
+                    1564.0,
+                    1405.0,
+                    1600.0,
+                    294.0,
+                    1600.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1594.0,
+                    1403.0,
+                    1594.0,
+                    1403.0,
+                    1629.0,
+                    294.0,
+                    1629.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1622.0,
+                    1403.0,
+                    1622.0,
+                    1403.0,
+                    1660.0,
+                    294.0,
+                    1660.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    726.0,
+                    1402.0,
+                    726.0,
+                    1402.0,
+                    762.0,
+                    295.0,
+                    762.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    756.0,
+                    1333.0,
+                    756.0,
+                    1333.0,
+                    800.0,
+                    291.0,
+                    800.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 7,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 2,
+                "poly": [
+                    298,
+                    75,
+                    857,
+                    75,
+                    857,
+                    105,
+                    298,
+                    105
+                ],
+                "score": 0.897
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    295,
+                    279,
+                    1404,
+                    279,
+                    1404,
+                    342,
+                    295,
+                    342
+                ],
+                "score": 0.822
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    295,
+                    1509,
+                    1398,
+                    1509,
+                    1398,
+                    1574,
+                    295,
+                    1574
+                ],
+                "score": 0.813
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1193,
+                    1400,
+                    1193,
+                    1400,
+                    1288,
+                    297,
+                    1288
+                ],
+                "score": 0.801
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    296,
+                    1309,
+                    1398,
+                    1309,
+                    1398,
+                    1373,
+                    296,
+                    1373
+                ],
+                "score": 0.794
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    840,
+                    2088,
+                    859,
+                    2088,
+                    859,
+                    2111,
+                    840,
+                    2111
+                ],
+                "score": 0.769
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    300,
+                    227,
+                    489,
+                    227,
+                    489,
+                    262,
+                    300,
+                    262
+                ],
+                "score": 0.759
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    1078,
+                    1400,
+                    1078,
+                    1400,
+                    1173,
+                    299,
+                    1173
+                ],
+                "score": 0.756
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    292,
+                    364,
+                    1403,
+                    364,
+                    1403,
+                    428,
+                    292,
+                    428
+                ],
+                "score": 0.754
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1710,
+                    1404,
+                    1710,
+                    1404,
+                    1834,
+                    298,
+                    1834
+                ],
+                "score": 0.754
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1394,
+                    1403,
+                    1394,
+                    1403,
+                    1489,
+                    297,
+                    1489
+                ],
+                "score": 0.746
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    296,
+                    909,
+                    1404,
+                    909,
+                    1404,
+                    974,
+                    296,
+                    974
+                ],
+                "score": 0.744
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1594,
+                    1404,
+                    1594,
+                    1404,
+                    1688,
+                    297,
+                    1688
+                ],
+                "score": 0.739
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    292,
+                    448,
+                    1401,
+                    448,
+                    1401,
+                    513,
+                    292,
+                    513
+                ],
+                "score": 0.736
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    1941,
+                    1402,
+                    1941,
+                    1402,
+                    2034,
+                    299,
+                    2034
+                ],
+                "score": 0.72
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    296,
+                    993,
+                    1405,
+                    993,
+                    1405,
+                    1060,
+                    296,
+                    1060
+                ],
+                "score": 0.696
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    533,
+                    1406,
+                    533,
+                    1406,
+                    658,
+                    297,
+                    658
+                ],
+                "score": 0.685
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    679,
+                    1404,
+                    679,
+                    1404,
+                    774,
+                    298,
+                    774
+                ],
+                "score": 0.667
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    286,
+                    1856,
+                    1402,
+                    1856,
+                    1402,
+                    1920,
+                    286,
+                    1920
+                ],
+                "score": 0.664
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    793,
+                    1405,
+                    793,
+                    1405,
+                    889,
+                    297,
+                    889
+                ],
+                "score": 0.597
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    300,
+                    227,
+                    489,
+                    227,
+                    489,
+                    262,
+                    300,
+                    262
+                ],
+                "score": 0.105
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    298.0,
+                    73.0,
+                    858.0,
+                    73.0,
+                    858.0,
+                    108.0,
+                    298.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    839.0,
+                    2087.0,
+                    861.0,
+                    2087.0,
+                    861.0,
+                    2117.0,
+                    839.0,
+                    2117.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    230.0,
+                    490.0,
+                    230.0,
+                    490.0,
+                    263.0,
+                    297.0,
+                    263.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    277.0,
+                    1405.0,
+                    277.0,
+                    1405.0,
+                    313.0,
+                    297.0,
+                    313.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    309.0,
+                    645.0,
+                    309.0,
+                    645.0,
+                    341.0,
+                    322.0,
+                    341.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    298.0,
+                    1512.0,
+                    1402.0,
+                    1512.0,
+                    1402.0,
+                    1544.0,
+                    298.0,
+                    1544.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    320.0,
+                    1540.0,
+                    1381.0,
+                    1540.0,
+                    1381.0,
+                    1576.0,
+                    320.0,
+                    1576.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1189.0,
+                    1402.0,
+                    1189.0,
+                    1402.0,
+                    1236.0,
+                    292.0,
+                    1236.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    321.0,
+                    1221.0,
+                    1404.0,
+                    1221.0,
+                    1404.0,
+                    1266.0,
+                    321.0,
+                    1266.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    325.0,
+                    1256.0,
+                    504.0,
+                    1256.0,
+                    504.0,
+                    1288.0,
+                    325.0,
+                    1288.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1306.0,
+                    1403.0,
+                    1306.0,
+                    1403.0,
+                    1349.0,
+                    294.0,
+                    1349.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    1340.0,
+                    568.0,
+                    1340.0,
+                    568.0,
+                    1372.0,
+                    322.0,
+                    1372.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1077.0,
+                    1402.0,
+                    1077.0,
+                    1402.0,
+                    1116.0,
+                    293.0,
+                    1116.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    1110.0,
+                    1404.0,
+                    1110.0,
+                    1404.0,
+                    1146.0,
+                    323.0,
+                    1146.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    1139.0,
+                    728.0,
+                    1139.0,
+                    728.0,
+                    1174.0,
+                    322.0,
+                    1174.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    360.0,
+                    1408.0,
+                    360.0,
+                    1408.0,
+                    402.0,
+                    293.0,
+                    402.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    325.0,
+                    395.0,
+                    870.0,
+                    395.0,
+                    870.0,
+                    431.0,
+                    325.0,
+                    431.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1710.0,
+                    1404.0,
+                    1710.0,
+                    1404.0,
+                    1746.0,
+                    294.0,
+                    1746.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    321.0,
+                    1740.0,
+                    1405.0,
+                    1740.0,
+                    1405.0,
+                    1777.0,
+                    321.0,
+                    1777.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    320.0,
+                    1771.0,
+                    1405.0,
+                    1771.0,
+                    1405.0,
+                    1809.0,
+                    320.0,
+                    1809.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    1802.0,
+                    955.0,
+                    1802.0,
+                    955.0,
+                    1835.0,
+                    323.0,
+                    1835.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1393.0,
+                    1403.0,
+                    1393.0,
+                    1403.0,
+                    1432.0,
+                    294.0,
+                    1432.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    321.0,
+                    1423.0,
+                    1405.0,
+                    1423.0,
+                    1405.0,
+                    1465.0,
+                    321.0,
+                    1465.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    1456.0,
+                    677.0,
+                    1456.0,
+                    677.0,
+                    1490.0,
+                    322.0,
+                    1490.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    907.0,
+                    1405.0,
+                    907.0,
+                    1405.0,
+                    948.0,
+                    293.0,
+                    948.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    942.0,
+                    985.0,
+                    942.0,
+                    985.0,
+                    975.0,
+                    323.0,
+                    975.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1595.0,
+                    1402.0,
+                    1595.0,
+                    1402.0,
+                    1629.0,
+                    296.0,
+                    1629.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    321.0,
+                    1625.0,
+                    1402.0,
+                    1625.0,
+                    1402.0,
+                    1660.0,
+                    321.0,
+                    1660.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    320.0,
+                    1653.0,
+                    398.0,
+                    1653.0,
+                    398.0,
+                    1691.0,
+                    320.0,
+                    1691.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    446.0,
+                    1404.0,
+                    446.0,
+                    1404.0,
+                    486.0,
+                    294.0,
+                    486.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    481.0,
+                    1142.0,
+                    481.0,
+                    1142.0,
+                    514.0,
+                    323.0,
+                    514.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1939.0,
+                    1402.0,
+                    1939.0,
+                    1402.0,
+                    1977.0,
+                    294.0,
+                    1977.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    1972.0,
+                    1403.0,
+                    1972.0,
+                    1403.0,
+                    2008.0,
+                    322.0,
+                    2008.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    2002.0,
+                    611.0,
+                    2002.0,
+                    611.0,
+                    2036.0,
+                    323.0,
+                    2036.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    993.0,
+                    1405.0,
+                    993.0,
+                    1405.0,
+                    1033.0,
+                    294.0,
+                    1033.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    1025.0,
+                    1136.0,
+                    1025.0,
+                    1136.0,
+                    1062.0,
+                    323.0,
+                    1062.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    531.0,
+                    1406.0,
+                    531.0,
+                    1406.0,
+                    570.0,
+                    292.0,
+                    570.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    324.0,
+                    567.0,
+                    1403.0,
+                    567.0,
+                    1403.0,
+                    597.0,
+                    324.0,
+                    597.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    321.0,
+                    593.0,
+                    1406.0,
+                    593.0,
+                    1406.0,
+                    631.0,
+                    321.0,
+                    631.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    627.0,
+                    636.0,
+                    627.0,
+                    636.0,
+                    660.0,
+                    323.0,
+                    660.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    680.0,
+                    1401.0,
+                    680.0,
+                    1401.0,
+                    714.0,
+                    296.0,
+                    714.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    318.0,
+                    704.0,
+                    1406.0,
+                    704.0,
+                    1406.0,
+                    751.0,
+                    318.0,
+                    751.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    741.0,
+                    1091.0,
+                    741.0,
+                    1091.0,
+                    775.0,
+                    323.0,
+                    775.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1851.0,
+                    1405.0,
+                    1851.0,
+                    1405.0,
+                    1894.0,
+                    292.0,
+                    1894.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    1885.0,
+                    760.0,
+                    1885.0,
+                    760.0,
+                    1921.0,
+                    322.0,
+                    1921.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    793.0,
+                    1403.0,
+                    793.0,
+                    1403.0,
+                    831.0,
+                    295.0,
+                    831.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    324.0,
+                    826.0,
+                    1402.0,
+                    826.0,
+                    1402.0,
+                    861.0,
+                    324.0,
+                    861.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    856.0,
+                    862.0,
+                    856.0,
+                    862.0,
+                    890.0,
+                    322.0,
+                    890.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    230.0,
+                    490.0,
+                    230.0,
+                    490.0,
+                    263.0,
+                    297.0,
+                    263.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 8,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    394,
+                    1404,
+                    394,
+                    1404,
+                    518,
+                    298,
+                    518
+                ],
+                "score": 0.943
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    621,
+                    1405,
+                    621,
+                    1405,
+                    746,
+                    297,
+                    746
+                ],
+                "score": 0.931
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    294,
+                    229,
+                    1401,
+                    229,
+                    1401,
+                    294,
+                    294,
+                    294
+                ],
+                "score": 0.922
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    295,
+                    312,
+                    1400,
+                    312,
+                    1400,
+                    376,
+                    295,
+                    376
+                ],
+                "score": 0.92
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    298,
+                    75,
+                    857,
+                    75,
+                    857,
+                    105,
+                    298,
+                    105
+                ],
+                "score": 0.902
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    301,
+                    538,
+                    1399,
+                    538,
+                    1399,
+                    603,
+                    301,
+                    603
+                ],
+                "score": 0.899
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    766,
+                    1400,
+                    766,
+                    1400,
+                    829,
+                    298,
+                    829
+                ],
+                "score": 0.892
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    296,
+                    848,
+                    1399,
+                    848,
+                    1399,
+                    913,
+                    296,
+                    913
+                ],
+                "score": 0.89
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1127,
+                    1402,
+                    1127,
+                    1402,
+                    1193,
+                    298,
+                    1193
+                ],
+                "score": 0.848
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    1045,
+                    1402,
+                    1045,
+                    1402,
+                    1111,
+                    299,
+                    1111
+                ],
+                "score": 0.847
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    296,
+                    1460,
+                    1403,
+                    1460,
+                    1403,
+                    1525,
+                    296,
+                    1525
+                ],
+                "score": 0.846
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    836,
+                    2088,
+                    866,
+                    2088,
+                    866,
+                    2113,
+                    836,
+                    2113
+                ],
+                "score": 0.846
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    295,
+                    1544,
+                    1402,
+                    1544,
+                    1402,
+                    1607,
+                    295,
+                    1607
+                ],
+                "score": 0.835
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1377,
+                    1400,
+                    1377,
+                    1400,
+                    1441,
+                    298,
+                    1441
+                ],
+                "score": 0.799
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    294,
+                    1627,
+                    1403,
+                    1627,
+                    1403,
+                    1691,
+                    294,
+                    1691
+                ],
+                "score": 0.798
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    932,
+                    1406,
+                    932,
+                    1406,
+                    1024,
+                    297,
+                    1024
+                ],
+                "score": 0.798
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1210,
+                    1400,
+                    1210,
+                    1400,
+                    1276,
+                    298,
+                    1276
+                ],
+                "score": 0.781
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1294,
+                    1402,
+                    1294,
+                    1402,
+                    1358,
+                    297,
+                    1358
+                ],
+                "score": 0.773
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    291,
+                    1709,
+                    1401,
+                    1709,
+                    1401,
+                    1775,
+                    291,
+                    1775
+                ],
+                "score": 0.733
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1793,
+                    1402,
+                    1793,
+                    1402,
+                    1887,
+                    298,
+                    1887
+                ],
+                "score": 0.653
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    298.0,
+                    73.0,
+                    858.0,
+                    73.0,
+                    858.0,
+                    108.0,
+                    298.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    832.0,
+                    2085.0,
+                    870.0,
+                    2085.0,
+                    870.0,
+                    2123.0,
+                    832.0,
+                    2123.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    395.0,
+                    1405.0,
+                    395.0,
+                    1405.0,
+                    432.0,
+                    294.0,
+                    432.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    325.0,
+                    428.0,
+                    1405.0,
+                    428.0,
+                    1405.0,
+                    461.0,
+                    325.0,
+                    461.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    321.0,
+                    457.0,
+                    1408.0,
+                    457.0,
+                    1408.0,
+                    493.0,
+                    321.0,
+                    493.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    321.0,
+                    486.0,
+                    396.0,
+                    486.0,
+                    396.0,
+                    519.0,
+                    321.0,
+                    519.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    623.0,
+                    1405.0,
+                    623.0,
+                    1405.0,
+                    659.0,
+                    292.0,
+                    659.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    654.0,
+                    1406.0,
+                    654.0,
+                    1406.0,
+                    687.0,
+                    323.0,
+                    687.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    321.0,
+                    682.0,
+                    1406.0,
+                    682.0,
+                    1406.0,
+                    720.0,
+                    321.0,
+                    720.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    321.0,
+                    714.0,
+                    395.0,
+                    714.0,
+                    395.0,
+                    747.0,
+                    321.0,
+                    747.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    227.0,
+                    1406.0,
+                    227.0,
+                    1406.0,
+                    267.0,
+                    294.0,
+                    267.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    321.0,
+                    258.0,
+                    431.0,
+                    258.0,
+                    431.0,
+                    295.0,
+                    321.0,
+                    295.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    311.0,
+                    1404.0,
+                    311.0,
+                    1404.0,
+                    350.0,
+                    295.0,
+                    350.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    321.0,
+                    344.0,
+                    639.0,
+                    344.0,
+                    639.0,
+                    377.0,
+                    321.0,
+                    377.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    540.0,
+                    1404.0,
+                    540.0,
+                    1404.0,
+                    576.0,
+                    296.0,
+                    576.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    569.0,
+                    1312.0,
+                    569.0,
+                    1312.0,
+                    605.0,
+                    322.0,
+                    605.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    764.0,
+                    1402.0,
+                    764.0,
+                    1402.0,
+                    800.0,
+                    296.0,
+                    800.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    324.0,
+                    798.0,
+                    744.0,
+                    798.0,
+                    744.0,
+                    829.0,
+                    324.0,
+                    829.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    850.0,
+                    1403.0,
+                    850.0,
+                    1403.0,
+                    886.0,
+                    295.0,
+                    886.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    879.0,
+                    1197.0,
+                    879.0,
+                    1197.0,
+                    914.0,
+                    322.0,
+                    914.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1124.0,
+                    1406.0,
+                    1124.0,
+                    1406.0,
+                    1168.0,
+                    293.0,
+                    1168.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    1160.0,
+                    782.0,
+                    1160.0,
+                    782.0,
+                    1192.0,
+                    323.0,
+                    1192.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    298.0,
+                    1044.0,
+                    1404.0,
+                    1044.0,
+                    1404.0,
+                    1081.0,
+                    298.0,
+                    1081.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    1078.0,
+                    916.0,
+                    1078.0,
+                    916.0,
+                    1112.0,
+                    323.0,
+                    1112.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1463.0,
+                    1403.0,
+                    1463.0,
+                    1403.0,
+                    1495.0,
+                    296.0,
+                    1495.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    320.0,
+                    1491.0,
+                    1292.0,
+                    1491.0,
+                    1292.0,
+                    1526.0,
+                    320.0,
+                    1526.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    1544.0,
+                    1403.0,
+                    1544.0,
+                    1403.0,
+                    1579.0,
+                    297.0,
+                    1579.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    1576.0,
+                    621.0,
+                    1576.0,
+                    621.0,
+                    1606.0,
+                    323.0,
+                    1606.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1376.0,
+                    1404.0,
+                    1376.0,
+                    1404.0,
+                    1415.0,
+                    294.0,
+                    1415.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    1409.0,
+                    611.0,
+                    1409.0,
+                    611.0,
+                    1440.0,
+                    323.0,
+                    1440.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    1626.0,
+                    1401.0,
+                    1626.0,
+                    1401.0,
+                    1662.0,
+                    297.0,
+                    1662.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    325.0,
+                    1658.0,
+                    894.0,
+                    1658.0,
+                    894.0,
+                    1690.0,
+                    325.0,
+                    1690.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    931.0,
+                    1403.0,
+                    931.0,
+                    1403.0,
+                    968.0,
+                    294.0,
+                    968.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    961.0,
+                    1406.0,
+                    961.0,
+                    1406.0,
+                    998.0,
+                    323.0,
+                    998.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    320.0,
+                    990.0,
+                    398.0,
+                    990.0,
+                    398.0,
+                    1027.0,
+                    320.0,
+                    1027.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1211.0,
+                    1405.0,
+                    1211.0,
+                    1405.0,
+                    1247.0,
+                    296.0,
+                    1247.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    1242.0,
+                    1152.0,
+                    1242.0,
+                    1152.0,
+                    1277.0,
+                    323.0,
+                    1277.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1291.0,
+                    1404.0,
+                    1291.0,
+                    1404.0,
+                    1334.0,
+                    295.0,
+                    1334.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    324.0,
+                    1326.0,
+                    787.0,
+                    1326.0,
+                    787.0,
+                    1358.0,
+                    324.0,
+                    1358.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1708.0,
+                    1406.0,
+                    1708.0,
+                    1406.0,
+                    1746.0,
+                    294.0,
+                    1746.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    324.0,
+                    1741.0,
+                    1322.0,
+                    1741.0,
+                    1322.0,
+                    1777.0,
+                    324.0,
+                    1777.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1790.0,
+                    1406.0,
+                    1790.0,
+                    1406.0,
+                    1830.0,
+                    292.0,
+                    1830.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    321.0,
+                    1821.0,
+                    1406.0,
+                    1821.0,
+                    1406.0,
+                    1861.0,
+                    321.0,
+                    1861.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    1856.0,
+                    610.0,
+                    1856.0,
+                    610.0,
+                    1885.0,
+                    323.0,
+                    1885.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 9,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 3,
+                "poly": [
+                    322,
+                    324,
+                    1379,
+                    324,
+                    1379,
+                    1555,
+                    322,
+                    1555
+                ],
+                "score": 0.976
+            },
+            {
+                "category_id": 4,
+                "poly": [
+                    296,
+                    1576,
+                    1406,
+                    1576,
+                    1406,
+                    1747,
+                    296,
+                    1747
+                ],
+                "score": 0.961
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    297,
+                    75,
+                    857,
+                    75,
+                    857,
+                    105,
+                    297,
+                    105
+                ],
+                "score": 0.902
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    836,
+                    2088,
+                    863,
+                    2088,
+                    863,
+                    2113,
+                    836,
+                    2113
+                ],
+                "score": 0.847
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    301,
+                    231,
+                    678,
+                    231,
+                    678,
+                    260,
+                    301,
+                    260
+                ],
+                "score": 0.753
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    435.0,
+                    327.0,
+                    660.0,
+                    327.0,
+                    660.0,
+                    366.0,
+                    435.0,
+                    366.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    988.0,
+                    329.0,
+                    1209.0,
+                    329.0,
+                    1209.0,
+                    364.0,
+                    988.0,
+                    364.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    426.0,
+                    382.0,
+                    677.0,
+                    382.0,
+                    677.0,
+                    417.0,
+                    426.0,
+                    417.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    981.0,
+                    384.0,
+                    1234.0,
+                    384.0,
+                    1234.0,
+                    415.0,
+                    981.0,
+                    415.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    767.0,
+                    406.0,
+                    819.0,
+                    406.0,
+                    819.0,
+                    440.0,
+                    767.0,
+                    440.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1323.0,
+                    406.0,
+                    1376.0,
+                    406.0,
+                    1376.0,
+                    433.0,
+                    1323.0,
+                    433.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    343.0,
+                    461.0,
+                    380.0,
+                    461.0,
+                    380.0,
+                    493.0,
+                    343.0,
+                    493.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    317.0,
+                    494.0,
+                    382.0,
+                    494.0,
+                    382.0,
+                    617.0,
+                    317.0,
+                    617.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    387.0,
+                    492.0,
+                    410.0,
+                    492.0,
+                    410.0,
+                    508.0,
+                    387.0,
+                    508.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    874.0,
+                    494.0,
+                    912.0,
+                    494.0,
+                    912.0,
+                    617.0,
+                    874.0,
+                    617.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    342.0,
+                    549.0,
+                    382.0,
+                    549.0,
+                    382.0,
+                    589.0,
+                    342.0,
+                    589.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    899.0,
+                    550.0,
+                    939.0,
+                    550.0,
+                    939.0,
+                    589.0,
+                    899.0,
+                    589.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    901.0,
+                    596.0,
+                    938.0,
+                    596.0,
+                    938.0,
+                    630.0,
+                    901.0,
+                    630.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1103.0,
+                    593.0,
+                    1144.0,
+                    593.0,
+                    1144.0,
+                    609.0,
+                    1103.0,
+                    609.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    348.0,
+                    641.0,
+                    382.0,
+                    641.0,
+                    382.0,
+                    676.0,
+                    348.0,
+                    676.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    905.0,
+                    641.0,
+                    939.0,
+                    641.0,
+                    939.0,
+                    676.0,
+                    905.0,
+                    676.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    764.0,
+                    673.0,
+                    816.0,
+                    673.0,
+                    816.0,
+                    704.0,
+                    764.0,
+                    704.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1325.0,
+                    676.0,
+                    1373.0,
+                    676.0,
+                    1373.0,
+                    705.0,
+                    1325.0,
+                    705.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    428.0,
+                    700.0,
+                    457.0,
+                    700.0,
+                    457.0,
+                    729.0,
+                    428.0,
+                    729.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    490.0,
+                    698.0,
+                    520.0,
+                    698.0,
+                    520.0,
+                    726.0,
+                    490.0,
+                    726.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    545.0,
+                    697.0,
+                    588.0,
+                    697.0,
+                    588.0,
+                    727.0,
+                    545.0,
+                    727.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    609.0,
+                    697.0,
+                    650.0,
+                    697.0,
+                    650.0,
+                    729.0,
+                    609.0,
+                    729.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    924.0,
+                    698.0,
+                    952.0,
+                    698.0,
+                    952.0,
+                    729.0,
+                    924.0,
+                    729.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    985.0,
+                    700.0,
+                    1015.0,
+                    700.0,
+                    1015.0,
+                    729.0,
+                    985.0,
+                    729.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1048.0,
+                    698.0,
+                    1076.0,
+                    698.0,
+                    1076.0,
+                    727.0,
+                    1048.0,
+                    727.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1100.0,
+                    698.0,
+                    1143.0,
+                    698.0,
+                    1143.0,
+                    724.0,
+                    1100.0,
+                    724.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1166.0,
+                    697.0,
+                    1206.0,
+                    697.0,
+                    1206.0,
+                    730.0,
+                    1166.0,
+                    730.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    489.0,
+                    716.0,
+                    614.0,
+                    716.0,
+                    614.0,
+                    756.0,
+                    489.0,
+                    756.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1047.0,
+                    716.0,
+                    1172.0,
+                    716.0,
+                    1172.0,
+                    756.0,
+                    1047.0,
+                    756.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    348.0,
+                    793.0,
+                    389.0,
+                    793.0,
+                    389.0,
+                    826.0,
+                    348.0,
+                    826.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    420.0,
+                    777.0,
+                    695.0,
+                    777.0,
+                    695.0,
+                    819.0,
+                    420.0,
+                    819.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    777.0,
+                    797.0,
+                    825.0,
+                    797.0,
+                    825.0,
+                    826.0,
+                    777.0,
+                    826.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    899.0,
+                    793.0,
+                    939.0,
+                    793.0,
+                    939.0,
+                    826.0,
+                    899.0,
+                    826.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    971.0,
+                    777.0,
+                    1244.0,
+                    777.0,
+                    1244.0,
+                    819.0,
+                    971.0,
+                    819.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1326.0,
+                    809.0,
+                    1375.0,
+                    809.0,
+                    1375.0,
+                    834.0,
+                    1326.0,
+                    834.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    385.0,
+                    830.0,
+                    396.0,
+                    830.0,
+                    396.0,
+                    841.0,
+                    385.0,
+                    841.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    425.0,
+                    860.0,
+                    436.0,
+                    860.0,
+                    436.0,
+                    870.0,
+                    425.0,
+                    870.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    898.0,
+                    838.0,
+                    941.0,
+                    838.0,
+                    941.0,
+                    876.0,
+                    898.0,
+                    876.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    325.0,
+                    893.0,
+                    361.0,
+                    893.0,
+                    361.0,
+                    1013.0,
+                    325.0,
+                    1013.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    874.0,
+                    892.0,
+                    939.0,
+                    892.0,
+                    939.0,
+                    1015.0,
+                    874.0,
+                    1015.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    358.0,
+                    992.0,
+                    385.0,
+                    992.0,
+                    385.0,
+                    1023.0,
+                    358.0,
+                    1023.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    908.0,
+                    987.0,
+                    935.0,
+                    987.0,
+                    935.0,
+                    1019.0,
+                    908.0,
+                    1019.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    357.0,
+                    1041.0,
+                    387.0,
+                    1041.0,
+                    387.0,
+                    1071.0,
+                    357.0,
+                    1071.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    908.0,
+                    1037.0,
+                    935.0,
+                    1037.0,
+                    935.0,
+                    1066.0,
+                    908.0,
+                    1066.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    777.0,
+                    1064.0,
+                    824.0,
+                    1064.0,
+                    824.0,
+                    1093.0,
+                    777.0,
+                    1093.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    414.0,
+                    1096.0,
+                    443.0,
+                    1096.0,
+                    443.0,
+                    1125.0,
+                    414.0,
+                    1125.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    501.0,
+                    1099.0,
+                    528.0,
+                    1099.0,
+                    528.0,
+                    1123.0,
+                    501.0,
+                    1123.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    585.0,
+                    1096.0,
+                    615.0,
+                    1096.0,
+                    615.0,
+                    1125.0,
+                    585.0,
+                    1125.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    666.0,
+                    1095.0,
+                    706.0,
+                    1095.0,
+                    706.0,
+                    1127.0,
+                    666.0,
+                    1127.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    908.0,
+                    1084.0,
+                    937.0,
+                    1084.0,
+                    937.0,
+                    1113.0,
+                    908.0,
+                    1113.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    965.0,
+                    1098.0,
+                    994.0,
+                    1098.0,
+                    994.0,
+                    1125.0,
+                    965.0,
+                    1125.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1049.0,
+                    1098.0,
+                    1079.0,
+                    1098.0,
+                    1079.0,
+                    1125.0,
+                    1049.0,
+                    1125.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1134.0,
+                    1096.0,
+                    1165.0,
+                    1096.0,
+                    1165.0,
+                    1125.0,
+                    1134.0,
+                    1125.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1216.0,
+                    1095.0,
+                    1257.0,
+                    1095.0,
+                    1257.0,
+                    1127.0,
+                    1216.0,
+                    1127.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1327.0,
+                    1080.0,
+                    1372.0,
+                    1080.0,
+                    1372.0,
+                    1104.0,
+                    1327.0,
+                    1104.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    495.0,
+                    1114.0,
+                    621.0,
+                    1114.0,
+                    621.0,
+                    1154.0,
+                    495.0,
+                    1154.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1045.0,
+                    1114.0,
+                    1172.0,
+                    1114.0,
+                    1172.0,
+                    1154.0,
+                    1045.0,
+                    1154.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    443.0,
+                    1178.0,
+                    670.0,
+                    1178.0,
+                    670.0,
+                    1213.0,
+                    443.0,
+                    1213.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    899.0,
+                    1190.0,
+                    938.0,
+                    1190.0,
+                    938.0,
+                    1225.0,
+                    899.0,
+                    1225.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    993.0,
+                    1175.0,
+                    1221.0,
+                    1175.0,
+                    1221.0,
+                    1214.0,
+                    993.0,
+                    1214.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    767.0,
+                    1208.0,
+                    826.0,
+                    1208.0,
+                    826.0,
+                    1257.0,
+                    767.0,
+                    1257.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1316.0,
+                    1214.0,
+                    1330.0,
+                    1214.0,
+                    1330.0,
+                    1267.0,
+                    1316.0,
+                    1267.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    350.0,
+                    1256.0,
+                    387.0,
+                    1256.0,
+                    387.0,
+                    1289.0,
+                    350.0,
+                    1289.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    896.0,
+                    1257.0,
+                    936.0,
+                    1257.0,
+                    936.0,
+                    1284.0,
+                    896.0,
+                    1284.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1326.0,
+                    1256.0,
+                    1376.0,
+                    1256.0,
+                    1376.0,
+                    1285.0,
+                    1326.0,
+                    1285.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    1293.0,
+                    360.0,
+                    1293.0,
+                    360.0,
+                    1412.0,
+                    323.0,
+                    1412.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    871.0,
+                    1292.0,
+                    912.0,
+                    1292.0,
+                    912.0,
+                    1415.0,
+                    871.0,
+                    1415.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    959.0,
+                    1304.0,
+                    974.0,
+                    1304.0,
+                    974.0,
+                    1315.0,
+                    959.0,
+                    1315.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    350.0,
+                    1319.0,
+                    387.0,
+                    1319.0,
+                    387.0,
+                    1353.0,
+                    350.0,
+                    1353.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    901.0,
+                    1319.0,
+                    939.0,
+                    1319.0,
+                    939.0,
+                    1353.0,
+                    901.0,
+                    1353.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    350.0,
+                    1383.0,
+                    387.0,
+                    1383.0,
+                    387.0,
+                    1416.0,
+                    350.0,
+                    1416.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    777.0,
+                    1428.0,
+                    825.0,
+                    1428.0,
+                    825.0,
+                    1463.0,
+                    777.0,
+                    1463.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    348.0,
+                    1447.0,
+                    389.0,
+                    1447.0,
+                    389.0,
+                    1480.0,
+                    348.0,
+                    1480.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    899.0,
+                    1447.0,
+                    939.0,
+                    1447.0,
+                    939.0,
+                    1480.0,
+                    899.0,
+                    1480.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1327.0,
+                    1455.0,
+                    1373.0,
+                    1455.0,
+                    1373.0,
+                    1484.0,
+                    1327.0,
+                    1484.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    421.0,
+                    1494.0,
+                    449.0,
+                    1494.0,
+                    449.0,
+                    1523.0,
+                    421.0,
+                    1523.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    518.0,
+                    1494.0,
+                    547.0,
+                    1494.0,
+                    547.0,
+                    1522.0,
+                    518.0,
+                    1522.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    615.0,
+                    1495.0,
+                    645.0,
+                    1495.0,
+                    645.0,
+                    1523.0,
+                    615.0,
+                    1523.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    706.0,
+                    1490.0,
+                    802.0,
+                    1490.0,
+                    802.0,
+                    1526.0,
+                    706.0,
+                    1526.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    970.0,
+                    1494.0,
+                    999.0,
+                    1494.0,
+                    999.0,
+                    1523.0,
+                    970.0,
+                    1523.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1067.0,
+                    1494.0,
+                    1098.0,
+                    1494.0,
+                    1098.0,
+                    1523.0,
+                    1067.0,
+                    1523.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1168.0,
+                    1494.0,
+                    1195.0,
+                    1494.0,
+                    1195.0,
+                    1523.0,
+                    1168.0,
+                    1523.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1257.0,
+                    1490.0,
+                    1353.0,
+                    1490.0,
+                    1353.0,
+                    1526.0,
+                    1257.0,
+                    1526.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    497.0,
+                    1513.0,
+                    618.0,
+                    1513.0,
+                    618.0,
+                    1550.0,
+                    497.0,
+                    1550.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1046.0,
+                    1510.0,
+                    1169.0,
+                    1510.0,
+                    1169.0,
+                    1550.0,
+                    1046.0,
+                    1550.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    337.0,
+                    416.5,
+                    385.0,
+                    416.5,
+                    385.0,
+                    448.5,
+                    337.0,
+                    448.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    893.75,
+                    416.5,
+                    940.75,
+                    416.5,
+                    940.75,
+                    449.5,
+                    893.75,
+                    449.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    895.0,
+                    460.5,
+                    941.0,
+                    460.5,
+                    941.0,
+                    492.5,
+                    895.0,
+                    492.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    892.0,
+                    502.5,
+                    945.0,
+                    502.5,
+                    945.0,
+                    541.5,
+                    892.0,
+                    541.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    336.0,
+                    596.0,
+                    388.0,
+                    596.0,
+                    388.0,
+                    627.5,
+                    336.0,
+                    627.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    362.0,
+                    699.0,
+                    399.0,
+                    699.0,
+                    399.0,
+                    725.5,
+                    362.0,
+                    725.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    665.0,
+                    695.5,
+                    716.0,
+                    695.5,
+                    716.0,
+                    728.5,
+                    665.0,
+                    728.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1224.0,
+                    696.5,
+                    1271.0,
+                    696.5,
+                    1271.0,
+                    729.5,
+                    1224.0,
+                    729.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    341.0,
+                    840.5,
+                    395.0,
+                    840.5,
+                    395.0,
+                    872.5,
+                    341.0,
+                    872.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    343.25,
+                    893.0,
+                    393.25,
+                    893.0,
+                    393.25,
+                    922.0,
+                    343.25,
+                    922.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    891.75,
+                    888.0,
+                    943.75,
+                    888.0,
+                    943.75,
+                    920.0,
+                    891.75,
+                    920.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    344.75,
+                    941.0,
+                    390.75,
+                    941.0,
+                    390.75,
+                    967.0,
+                    344.75,
+                    967.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    346.0,
+                    1191.5,
+                    390.0,
+                    1191.5,
+                    390.0,
+                    1221.5,
+                    346.0,
+                    1221.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    891.0,
+                    1381.5,
+                    945.0,
+                    1381.5,
+                    945.0,
+                    1412.5,
+                    891.0,
+                    1412.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1574.0,
+                    1406.0,
+                    1574.0,
+                    1406.0,
+                    1609.0,
+                    294.0,
+                    1609.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1603.0,
+                    1403.0,
+                    1603.0,
+                    1403.0,
+                    1639.0,
+                    293.0,
+                    1639.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1632.0,
+                    1404.0,
+                    1632.0,
+                    1404.0,
+                    1664.0,
+                    294.0,
+                    1664.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1659.0,
+                    1407.0,
+                    1659.0,
+                    1407.0,
+                    1693.0,
+                    294.0,
+                    1693.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1691.0,
+                    1404.0,
+                    1691.0,
+                    1404.0,
+                    1720.0,
+                    294.0,
+                    1720.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1716.0,
+                    1157.0,
+                    1716.0,
+                    1157.0,
+                    1749.0,
+                    294.0,
+                    1749.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    73.0,
+                    858.0,
+                    73.0,
+                    858.0,
+                    108.0,
+                    297.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    832.0,
+                    2085.0,
+                    868.0,
+                    2085.0,
+                    868.0,
+                    2125.0,
+                    832.0,
+                    2125.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    227.0,
+                    679.0,
+                    227.0,
+                    679.0,
+                    266.0,
+                    296.0,
+                    266.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 10,
+            "width": 1700,
+            "height": 2200
+        }
+    }
+]
\ No newline at end of file
diff --git a/parse/train/SJl3h2EYvS/SJl3h2EYvS.md b/parse/train/SJl3h2EYvS/SJl3h2EYvS.md
new file mode 100644
index 0000000000000000000000000000000000000000..7c83a6abf1782c75118bf347ba98c06367232a12
--- /dev/null
+++ b/parse/train/SJl3h2EYvS/SJl3h2EYvS.md
@@ -0,0 +1,224 @@
+# CLAREL: CLASSIFICATION VIA RETRIEVAL LOSS FOR ZERO-SHOT LEARNING
+
+Anonymous authors Paper under double-blind review
+
+# ABSTRACT
+
+We address the problem of learning fine-grained cross-modal representations. We propose an instance-based deep metric learning approach in joint visual and textual space. The key novelty of this paper is that it shows that using per-image semantic supervision leads to substantial improvement in zero-shot performance over using class-only supervision. On top of that, we provide a probabilistic justification for a metric rescaling approach that solves a very common problem in the generalized zero-shot learning setting, i.e., classifying test images from unseen classes as one of the classes seen during training. We evaluate our approach on two finegrained zero-shot learning datasets: CUB and FLOWERS. We find that on the generalized zero-shot classification task CLAREL consistently outperforms the existing approaches on both datasets.
+
+# 1 INTRODUCTION
+
+Deep learning-based approaches have demonstrated superior flexibility and generalization capabilities in information processing on a wide variety of tasks, such as vision, speech and language (LeCun et al., 2015). However, it has been widely realized that the transfer of deep representations to real-world applications is challenging due to the typical reliance on massive hand-labeled datasets. Learning in the low-labeled data regime, especially in the zero-shot (Wang et al., 2019) and the few-shot (Wang & Yao, 2019) setups, have recently received significant attention in the literature. In the problem of zero-shot learning (ZSL), the objective is to recognize categories that have not been seen during the training (Larochelle et al., 2008). This is typically done by relying on anchor embeddings learned in one modality as prototypes and by associating a query embedding from the other modality with the closest prototype. In the generalized ZSL (GZSL) case (Xian et al., 2018c), the objective is more challenging as recognition is performed in the joint space of seen and unseen categories. ZSL, as well as its generalized counterpart, provide a viable framework to learn cross-modal representations that are flexible and adaptive. For example, in this paradigm, the adaptation to a new classification task based on text/image representation space alignment could be as easy as defining/appending/modifying a set of text sentences to define classes of new classifiers. This is an especially relevant problem as machine learning is challenged with the long tail of classes, and the idea of learning from pairs of images and sentences, abundant on the web, looks like a natural solution. Therefore, in this paper we specifically target the fine-grained scenario of paired images and their respective text descriptions. The uniqueness of this scenario is in the fact that the co-occurance of image and text provides a rich source of information. The ways of leveraging this source have not been sufficiently explored in the context of GZSL. Although we focus exclusively on the GZSL recognition setup in this paper, we believe that the research in this direction has potential to enable zero-shot flexibility in a wider array of high-level tasks such as segmentation or conditional image generation (Zhang et al., 2018). The contributions of this work can be characterized under the following two themes.
+
+Instance-based training loss. Most prominent zero-shot learning approaches rely heavily on classlevel modality alignment (Xian et al., 2018c). We propose a new composite loss function that balances instance-based pairwise image/text retrieval loss and the usual classifier loss. The retrieval loss term does not use class labels. We demonstrate that the class-level information is important, but in the fine-grained text/image pairing scenarios, most of the GZSL accuracy can be extracted from the instance-based retrieval loss. To the best of our knowledge, this type of training has not been used in the GZSL literature. Its impressive performance opens up new promising research directions.
+
+![](images/c5306e54b7867e3eb7aee45c6499213721005da29ec547a7ed5a3a76968503c1.jpg)  
+Figure 1: The architecture and training diagram describing the proposed method. Each batch consists of randomly sampled instances, i.e. pairs of images and their corresponding texts. Images are embedded via ResNet and texts are embedded via a CNN/LSTM stack. Image and text features are projected via a fully connected layer into the same dimensional space. In this space, distances between text and image features from different instances are computed. The negative distances are fed into softmax to train on both the image and the text retrieval tasks. The image retrieval task consists of retrieving the image corresponding to the given text of the same instance and the text retrieval task is vice versa. In addition to that, image and text embeddings are trained on auxiliary image and text classification tasks on the class labels corresponding to instances.
+
+Metric space rescaling. Metric-based ZSL approaches rely on distances between prototypes and query embeddings during inference. They are known to suffer from imbalanced performance on seen and unseen classes (Liu et al., 2018). Previous work proposed to use a heuristic trick, calibrated stacking (Chao et al., 2016) or calibration (Das & Lee, 2019), to solve the problem. We refer to this technique as metric rescaling in our work, and provide a sound probabilistic justification for it.
+
+# 2 PROPOSED METHOD
+
+In this paper, we specifically target the fine-grained visual description scenario, as defined by Reed et al. (2016). In this setting, the dataset consists of a number of images from a given set of classes and each image is accompanied by a number of textual descriptions. The task is to learn a joint representation space for images and texts that can be used for zero-shot recognition. An instance of the zero-shot multimodal representation learning problem can then be defined as follows. Given a training set $S = \{ ( v _ { n } , t _ { n } , y _ { n } ) \mid v _ { n } \in \mathcal { V }$ , $t _ { n } \in \mathcal { T } , y _ { n } \in \mathcal { V } , n = 1 \ldots N \}$ of image, text and label tuples, we are interested in finding representations $f _ { \phi } : \mathcal { V } \to \mathcal { Z }$ of image, parameterized by $\phi$ , and $f _ { \theta } : \mathcal { T }  \mathcal { Z }$ of text, parameterized by $\theta$ , in a common embedding space $\mathcal { Z }$ . Furthermore, GZSL problem is defined using the sets of seen $\mathcal { V } ^ { t r }$ and unseen $\mathcal { V } ^ { t s }$ classes, such that $\mathcal { V } = \mathcal { V } ^ { t r } \cup \mathcal { V } ^ { t s }$ and $\bar { \mathcal { V } } ^ { t r } \cap \mathcal { V } ^ { t s } = \emptyset$ . The training set will then only contain the seen classes, i.e. $S ^ { t r } = \{ ( v _ { n } , t _ { n } , y _ { n } ) \mid v _ { n } \in$ $\nu$ , $t _ { n } \in \mathcal { T } , y _ { n } \in \mathcal { y } ^ { t r } \}$ and the task is to build a classifier function $g : \mathcal { Z } \times \mathcal { Z }  \mathcal { V }$ . This is different from the ZSL scenario focusing on $g : \mathcal { Z } \times \mathcal { Z }  \mathcal { Y } ^ { t s }$ .
+
+To build $g$ , most approaches to joint representation learning rely on class labeling to train a representation. For example, all the methods reviewed by Xian et al. (2018c) require the access to class labels at train time. We hypothesise that in the fine-grained learning scenario, such as the one described by Reed et al. (2016), a lot of information can be extracted simply from pairwise image/text cooccurrences. The class labels really only become critically necessary when we define class prototypes, i.e. at zero-shot test time. Following this intuition, we define a composite loss function that relies both on the pairwise relationships and on the class labels. The high-level description of the proposed framework is depicted in Figure 1. The framework enables us, among other things, to experiment with the effects of train-time availability of class labels on the quality of zero-shot representations. The framework is based on projecting texts and images into a common space and then learning a representation based on a mixture of four loss functions: a pairwise text retrieval loss, a pairwise image retrieval loss, a text classifier loss and an image classifier loss (see Algorithm 1).
+
+Algorithm 1 Loss calculation for a single optimization iteration of the proposed method. $N$ is the number of instances in the training set ${ \mathcal { S } } ^ { t r }$ , $B$ is the number of instances per batch, $C$ is the number of classes in the train set. $\mathrm { R A N D O M S A M P L E } ( S , B )$ denotes a set of $B$ elements chosen uniformly at random from a set $s$ , without replacement.
+
+<table><tr><td colspan="3">Input: Training set Str = {(U1,t1,yi),...,(UN,tN,yN)},λ ∈ [0,1], κ ∈ [0,1]. Output: The loss J(,θ) for a randomly sampled training batch. I← RANDOMSAMPLE({1,...,N},B) &gt;Select B instance indices for batch</td></tr><tr><td>Jrc(Θ),J1c(Φ)←0,0 foriinZdo</td><td colspan="3">Initialize classification losses Embed images and texts Image classifier probabilities</td></tr><tr><td colspan="3">Zui,Zt←f(Ui）,fθ(ti)</td></tr><tr><td colspan="2">p ← softmax(Wizui +b1) PT ← sOftmax(WTZt +br)</td></tr><tr><td colspan="2">Textclassifierprobabilities JTc(0) ← JTc(0)+ crossentropy(pr,yi) Text classification loss</td></tr><tr><td colspan="2">JIc(𝜙) ← Jic(Φ)+ crossentropy(p1, yi) Image classification loss end for</td></tr><tr><td colspan="2">JTR(Φ,0),J1R(Φ,0)←0,0 Initialize retrieval losses</td></tr><tr><td colspan="2">foriinZdo 1 Text retrieval loss</td></tr><tr><td colspan="2">JTR(Φ,0)←JTR(Φ,0)+ d(zu,Zt）+log∑ exp(-d(Zu,Ztj)) B j∈I</td></tr><tr><td colspan="2">1-B JIR(Φ,0)←JIR(Φ,0)+ d(Zui,Zti) + log∑exp(-d(Zt , Zuj))|&gt; Image retrieval loss</td></tr><tr><td colspan="2">j∈I</td></tr><tr><td colspan="2">J(,θ)←λJTR(𝜙,0)+(1-λ)JIR(Φ,0) Add retrieval loss to the total loss J(Φ,θ)←(1-κ)J(,0)+(Jrc(0)+JIc()) Add classification loss to the total loss</td></tr></table>
+
+# 2.1 RETRIEVAL LOSS FUNCTION
+
+Pairwise cross-modal loss function is based solely on the pairwise relationships between texts and images. We choose to use the metric learning approach to capture the relationship between images and texts. Now, suppose $d$ is a metric $d : \mathcal { Z } \times \mathcal { Z }  \mathbb { R } ^ { + }$ , $v _ { i }$ is an image and $\tau = \{ t _ { j ^ { \prime } } \}$ is a collection of arbitrary texts sampled uniformly at random, of which text $t _ { j }$ belongs to $v _ { i }$ . We propose the following model for the probability of image $v _ { i }$ and text $t _ { j }$ to belong to the same object instance:
+
+$$
+p _ { \phi , \theta } ( i = j | v _ { i } , t _ { j } , \tau ) = \frac { \exp ( - d ( f _ { \phi } ( v _ { i } ) , f _ { \theta } ( t _ { j } ) ) ) } { \sum _ { t _ { j ^ { \prime } } \in \tau } \exp ( - d ( f _ { \phi } ( v _ { i } ) , f _ { \theta } ( t _ { j ^ { \prime } } ) ) ) } \ .
+$$
+
+The learning is then based on the following cross-entropy loss defined on the batch of size $B$ :
+
+$$
+J _ { T R } ( \phi , \theta ) = - \frac { 1 } { B } \sum _ { i , j = 1 } ^ { B } \ell _ { i , j } \log p _ { \phi , \theta } ( i = j | v _ { i } , t _ { j } , \{ t _ { j ^ { \prime } } \} _ { j ^ { \prime } = 1 } ^ { B } ) ,
+$$
+
+where $\ell _ { i , j }$ is a binary indicator of the true match $\mathcal { \ell } _ { i , j } = 1$ , if $i = j$ and 0 otherwise). Note that the expression above has the interpretation of the text retrieval loss. It attains its smallest value when for each image in the batch we manage to assign probability 1 to its respective text and 0 to all other texts. This can be further expanded as:
+
+$$
+J _ { T R } ( \phi , \theta ) = \frac { 1 } { B } \sum _ { i = 1 } ^ { B } \left( d ( f _ { \phi } ( v _ { i } ) , f _ { \theta } ( t _ { i } ) ) + \log \Big [ \sum _ { t _ { j ^ { \prime } } \in \tau } \exp ( - d ( f _ { \phi } ( v _ { i } ) , f _ { \theta } ( t _ { j ^ { \prime } } ) ) ) \Big ] \right) .
+$$
+
+Exchanging the order of image and text in the probability model (1) leads to the image retrieval loss, $J _ { I R } ( \phi , \bar { \theta } )$ . The two losses are mixed using parameter $\lambda \in [ 0 , 1 ]$ as shown in Algorithm 1.
+
+The pairwise retrieval loss functions are responsible for the modality alignment. In addition to those, we propose to include, as mentioned above, the usual image and text classifier losses. These losses are responsible for reducing the intraclass variability of representations. The classifier losses are added to the retrieval losses using a mixing parameter $\kappa \in [ 0 , 1 ]$ as shown in Algorithm 1.
+
+# 2.2 BALANCING ACCURACY ON THE SEEN AND UNSEEN CLASSES
+
+Let us define class prototypes $\mathbf p ( y )$ , each based on the set of texts $\mathcal { T } _ { y }$ belonging to class $y$ , $\mathbf { \{ p ( }  y \mathbf { ) = }$ $\begin{array} { r } { \frac { 1 } { | T _ { y } | } \sum _ { t _ { i } \in \mathcal { T } _ { y } } f _ { \theta } ( t _ { i } ) \ | \ y \in \mathcal { V } \} } \end{array}$ . In the context of GZSL, the nearest neighbor decision rule for a given image $v$ and its features ${ \bf z } _ { v } = f _ { \phi } ( v )$ has the following form:
+
+$$
+\boldsymbol { \widehat { y } } = \arg \operatorname* { m i n } _ { \boldsymbol { y } \in \mathcal { V } } d ( \mathbf { z } _ { v } , \mathbf { p } ( \boldsymbol { y } ) ) .
+$$
+
+The most acute problem faced in this setup is the accuracy imbalance between seen and unseen classes. A very representative case clearly outlining the imbalance problem is presented in Table 6 of (Xian et al., 2018c), where accuracy on the seen classes is always significantly greater than the accuracy on unseen ones. In order to measure and control the imbalance, three metrics are commonly used to assess the classification performance in the GZSL scenario: the Top-1 accuracy on the seen categories (s), the Top-1 accuracy on the unseen categories $\mathbf { \Pi } ^ { ( \mathbf { u } ) }$ and their harmonic mean, $\mathbf { H } = \mathbf { u } \cdot \mathbf { s } / ( \mathbf { u } + \mathbf { s } )$ . The main metric to assess GZSL performance is then $\mathbf { H }$ , which quantifies both $\mathbf { u }$ and s.
+
+To formalize the problem, we first introduce $y _ { v }$ , the true class label of image $v$ . Mathematically, the main GZSL pain point is that $\mathbb { P } \{ \widehat { y } \in \mathcal { y } ^ { t r } | y _ { v } \in \mathcal { y } ^ { t s } \}$ is significantly greater than $\mathbb { P } \{ \widehat { y } \in \mathcal { y } ^ { t s } | y _ { v } \in$ $\mathcal { V } ^ { t r } \}$ b b. In other words, the problem is that a given image is more likely to be confused with one of the seen classes if it belongs to an unseen class than vice versa. Our approach to solving the problem is based on the following probabilistic representation of the event space for the decision rule in Equation (4):
+
+$$
+\mathbb { P } \{ \boldsymbol { { \widehat { y } } } \in { \mathcal { Y } } ^ { t r } | y _ { v } \in { \mathcal { Y } } ^ { t s } \} = \mathbb { P } \left\{ \operatorname* { m i n } _ { y \in { \mathcal { Y } } ^ { t r } } d ( \mathbf { z } _ { v } , \mathbf { p } ( y ) ) < \operatorname* { m i n } _ { y \in { \mathcal { Y } } ^ { t s } } d ( \mathbf { z } _ { v } , \mathbf { p } ( y ) ) \mid y _ { v } \in { \mathcal { Y } } ^ { t s } \right\} .
+$$
+
+Rephrasing, the most acute GZSL error happens when the prototype of one of the seen classes is closer to an image embedding from an unseen class than any of the prototypes of the unseen classes.
+
+To rectify the situation we propose the following very direct solution to balance $\mathbb { P } \{ \widehat { y } \in \mathcal { y } ^ { t r } | y _ { v } \in \mathcal { y } ^ { t s } \}$ and $\mathbb { P } \{ \hat { \mathcal { Y } } \in \mathcal { Y } ^ { t s } | y _ { v } \in \mathcal { Y } ^ { t \bar { r } } \}$ . We introduce a positive scalar $\alpha \in \mathbb { R } ^ { + }$ band scale all the distances bcorresponding to the seen prototypes by $1 + \alpha$ . This gives rise to the following scaled distance $d _ { \alpha }$ :
+
+$$
+d _ { \alpha } ( \mathbf { z } _ { v } , \mathbf { p } ( y ) ) = \left\{ \begin{array} { l l } { ( 1 + \alpha ) d ( \mathbf { z } _ { v } , \mathbf { p } ( y ) ) , } & { \mathrm { i f ~ } y \in \mathcal { Y } ^ { t r } } \\ { d ( \mathbf { z } _ { v } , \mathbf { p } ( y ) ) , } & { \mathrm { o t h e r w i s e } } \end{array} \right. .
+$$
+
+The misclassification between unseen as seen classes for the classifier $\widehat { y } _ { \alpha }$ , based on (6) is then:
+
+$$
+\mathbb { P } \{ \widehat { y } _ { \alpha } \in \mathcal { Y } ^ { t r } | y _ { v } \in \mathcal { Y } ^ { t s } \} = \mathbb { P } \left\{ \left( 1 + \alpha \right) \operatorname* { m i n } _ { y \in \mathcal { Y } ^ { t r } } d ( \mathbf { z } _ { v } , \mathbf { p } ( y ) ) < \operatorname* { m i n } _ { y \in \mathcal { Y } ^ { t s } } d ( \mathbf { z } _ { v } , \mathbf { p } ( y ) ) | y _ { v } \in \mathcal { Y } ^ { t s } \right\} \ ,
+$$
+
+and it has the following property: for any $0 \leq \alpha _ { 1 } \leq \alpha _ { 2 }$ , $\mathbb { P } \{ \widehat { y } _ { \alpha _ { 1 } } ~ \in ~ \mathcal { Y } ^ { t r } | y _ { v } ~ \in ~ \mathcal { Y } ^ { t s } \} ~ \geq ~ \mathbb { P } \{ \widehat { y } _ { \alpha _ { 2 } } ~ \in ~$ $\mathcal { V } ^ { t r } | y _ { v } \in \mathcal { V } ^ { t s } \}$ , i.e. $\mathbb { P } \{ \widehat { y } _ { \alpha } \in \mathcal { V } ^ { t \bar { r } } | y _ { v } \in \mathcal { V } ^ { t s } \}$ b   is a monotone non-increasing function of $\alpha$ b and we bcan reduce it by increasing $\alpha$ (please refer to Appendix A for a proof). Consider now $\mathbb { P } \{ \widehat { y } _ { \alpha } \in$ $\mathcal { V } ^ { t r } | y _ { v } \in \mathcal { V } ^ { t r } \big \}$ , which is a probability that we classify an image $v$ bfrom one of the seen classes as still one of the seen classes. Using exactly the same chain of arguments as in Appendix A, it is straightforward to show that the probability is a non-increasing function of $\alpha$ . Hence the probability $\mathbb { P } \{ \hat { y } _ { \alpha } ^ { \sim } \in \mathcal { V } ^ { t s } | y _ { v } \in \mathcal { V } ^ { t r } \} = 1 - \mathbb { P } \{ \hat { y } _ { \alpha } \in \mathcal { V } ^ { \hat { t r } } | y _ { v } \in \mathcal { V } ^ { t r } \}$ is a non-decreasing function of $\alpha$ . Therefore, as $\alpha$ increases, we expect more classification errors in classifying images from seen classes, because some of them will be classified as one of the unseen classes.
+
+To sum up, given the arguments presented above we expect that by varying $\alpha > 0$ we can balance the error rate $\mathbb { P } \mathbf { \bar { \{ y _ { \alpha } \in \mathcal { V } ^ { t r } \vert y _ { v } \in \mathcal { V } ^ { t s } \} } }$ of leaking the unseen class images into seen class classification bdecision and the error rate $\mathbb { P } \{ \widehat { y } _ { \alpha } ^ { } \in \mathcal { V } ^ { t s } | y _ { v } ^ { - } \in \mathcal { V } ^ { t r } \}$ of leaking the seen class images into unseen bclass classification decision. This is possible as we just showed above that $\mathbb { P } \{ \widehat { y } _ { \alpha } \in \mathcal { V } ^ { t r } | y _ { v } \in \mathcal { V } ^ { t s } \}$ is a non-increasing function of $\alpha$ , while $\mathbb { P } \{ \widehat { y } _ { \alpha } \in \mathcal { V } ^ { t s } | y _ { v } \in \mathcal { V } ^ { t r } \}$ b is a non-decreasing one. It is also important to emphasize that $\alpha$ bis applied only to distances between the query embedding and the prototypes of seen classes and it is constant over seen classes. Therefore, the application of $\alpha$ does not at all affect the classification error rates either within $\mathcal { V } ^ { t r }$ or within $\mathcal { V } ^ { t s }$ . Varying $\alpha$ balances exclusively the classification errors arising from transitions between seen and unseen class labels. We study the empirical aspects of balancing $\alpha$ in Section 4.4.
+
+Table 1: Generalized zero-shot Top-1 classification accuracy.   
+
+<table><tr><td></td><td colspan="3">CUB</td><td colspan="3">FLOWERS</td></tr><tr><td></td><td>u</td><td>S</td><td>H</td><td>u</td><td>s</td><td>H</td></tr><tr><td>CADA-VAE (Schönfeld et al., 2019)</td><td>n/a</td><td>n/a</td><td>53.4</td><td>n/a</td><td>n/a</td><td>n/a</td></tr><tr><td>f-CLSWGAN (Xian et al., 2018d)</td><td>50.3</td><td>58.3</td><td>54.0</td><td>59.0</td><td>73.8</td><td>65.6</td></tr><tr><td>f-VAEGAN-D2 (Xian et al.,2019)</td><td>48.4</td><td>60.1</td><td>53.6</td><td>56.8</td><td>74.9</td><td>64.6</td></tr><tr><td>cycle-(U)WGAN (Felix et al., 2018)</td><td>47.9</td><td>59.3</td><td>53.0</td><td>61.6</td><td>69.2</td><td>65.2</td></tr><tr><td>COSMO+f-CLSWGAN (Atzmon &amp; Chechik, 2019)</td><td>n/a</td><td>n/a</td><td>n/a</td><td>59.6</td><td>81.4</td><td>68.8</td></tr><tr><td>CLAREL (Ours)</td><td>59.3</td><td>52.6</td><td>55.8</td><td>73.0</td><td>73.6</td><td>73.3</td></tr></table>
+
+Table 2: Zero-shot Top-1 classification accuracy.   
+
+<table><tr><td></td><td>CUB</td><td>FLOWERS</td></tr><tr><td>CADA-VAE (Schonfeld et al., 2019)</td><td>n/a</td><td>n/a</td></tr><tr><td>f-CLSWGAN (Xian et al., 2018d)</td><td>57.3</td><td>67.2</td></tr><tr><td>f-VAEGAN-D2 (Xian et al., 2019)</td><td>61.0</td><td>67.7</td></tr><tr><td>cycle-(U)WGAN (Felix et al., 2018)</td><td>58.6</td><td>70.3</td></tr><tr><td>CLAREL (Ours)</td><td>66.7</td><td>76.8</td></tr></table>
+
+# 3 RELATED WORK
+
+ZSL approaches aim at recognizing objects belonging to classes unseen during training (Larochelle et al., 2008; Palatucci et al., 2009). This has been extended to the GZSL framework in which the decision space consists of both seen and unseen classes (Socher et al., 2013; Xian et al., 2018c). The classical zero-shot approaches build a joint visual-semantic space, relying on a linear cross-modal compatibility function (e.g. dot-product between query embedding and semantic prototypes or a variation of a hinge loss) (Frome et al., 2013; Akata et al., 2015; 2016; Reed et al., 2016). Non-linear variants of the compatibility has also been explored (Xian et al., 2016; Socher et al., 2013). Extending previously proposed cross-modal transfer approaches based on auto-encoders (Hubert Tsai et al., 2017) and cross-domain learning (Gretton et al., 2007), more recent line of work (Schönfeld et al., 2019; Xian et al., 2018d; 2019; Felix et al., 2018; Verma et al., 2018) relies on combining these approaches and their variations with dataset augmentation tools such as GAN (Goodfellow et al., 2014) and VAE (Kingma & Welling, 2014). It is argued that the use of those tools helps to resolve one of the prominent problems in GZSL scenario: classifying images from unseen classes as one of the seen classes. There exist approaches that try to tackle this same problem via temperature calibration (Liu et al., 2018) originally proposed by Hinton et al. (2015). Chao et al. (2016); Das & Lee (2019) proposed an approach to seen/unseen accuracy balancing that is very similar to ours, based on heuristic arguments. We extend this line of work here by providing a probabilistic justification for the balancing effect observed when applying metric rescaling. Atzmon & Chechik (2019) propose a more sophisticated way to deal with seen/unseen imbalance via adaptive confidence smoothing and gating, yet as authors note it is much simpler to train than the existing GAN-based zero-shot approaches. In this work, we introduce arguably the simplest zero-shot representation training approach of all, and we demonstrate that when the image level text information is available, it achieves the state-of-the-art results on GZSL task on two well-known datasets.
+
+# 4 EXPERIMENTAL RESULTS
+
+# 4.1 DATASETS
+
+We focus on learning embeddings for fine-grained visual descriptions and test them in ZSL/GZSL scenario. To test the quality of trained embeddings we focus on datasets that provide paired images and text descriptions, such as Caltech-UCSD-Birds (CUB) (Welinder et al., 2010) and Oxford Flowers (FLOWERS) (Nilsback & Zisserman, 2008), that were augmented with textual descriptions by Reed et al. (2016). We use the GZSL splits proposed by Xian et al. (2018c). The attribute-based datasets, such as SUN (Patterson et al., 2014) and AWA (Lampert et al., 2014) do not contain this information and do not have a notion of entity of a class in them. They are out of the scope of the current paper.
+
+![](images/6c68eb4c7e42a907d13836d18bd23fbf8d9cba1c7819eebd0cbae1133409eec9.jpg)  
+Figure 2: Harmonic mean Top-1 accuracy on seen and unseen, $\mathbf { H }$ , against the value of $\alpha$ on the validation set. The curves represent the mean and $9 5 \%$ confidence intervals over 10 optimization runs. Results are stable over different runs. $\mathbf { H }$ exhibits a distinct inverted U-shape w.r.t. $\alpha$ .
+
+# 4.2 ARCHITECTURE AND TRAINING DETAILS
+
+Below, we provide more detailed description of parameters used to build and train the architecture depicted in Figure 1. We use exactly the same hyperparameter settings for CUB and FLOWERS. The text feature extractor is built by cascading two ResNet blocks, followed by a BiLSTM. Each ResNet block has 3 convolutional/batch norm layers. The number of filters in the ResNet blocks is 128 and 256, BiLSTM has 512 filters for forward and backward branches (1024 total). All variables in the convolutional stack (including the batch normalization parameters $\gamma$ and $\beta$ ) are L2-penalized with weight 0.001. The image feature extractor is a ResNet-101 with fixed weights pretrained on the split of ImageNet proposed by Xian et al. (2018c). In this work we use precomputed image features, available in (Xian et al., 2018a) for CUB and in (Xian et al., 2018b) for FLOWERS. Image and text features are projected in the common embedding space of size 1024 with FC layers with no non-linearity. They are preceded with a dropout of 0.25. The trainable components of the model are trained for $1 5 0 \mathrm { k }$ batches of size 32 using SGD with initial learning rate of 0.1 that is annealed by a factor of 10 every $5 0 \mathrm { k }$ batches. For each batch, we sample 32 instances, each instance includes a vector of precomputed ResNet-101 features and 10 text descriptions corresponding to it, according to the original dataset definition Reed et al. (2016). All 10 text descriptions are processed via the CNN/LSTM stack and the resulting embeddings are average pooled to create a vector representation of length 1024.
+
+# 4.3 KEY RESULTS
+
+Our key empirical results are compared in Table 1 and in Table 2 against the latest state of the art. Our results are based on the settings of $\lambda = 0 . 5$ , $\kappa = 0 . 5$ and $\alpha$ selected on the validation sets of CUB and FLOWERS datasets. Please refer to Section 4.5 for the analysis of stability with respect to the choices of $\lambda$ and $\kappa$ and Sections 2.2 and 4.4 for more details on the selection of $\alpha$ . The combination of the proposed training method and the rebalancing of the metric space results in the state-of-the-art performance. Most of the current methods rely on the dataset augmentation techniques based on GANs, VAEs or combinations thereof. Those are clearly complementary w.r.t. our method and their addition to the training procedure is likely to further boost the performance of our proposed approach. However, this is outside of the scope of the current work. Moreover, the proposed method is state-of-the-art on FLOWERS even when compared against (Atzmon & Chechik, 2019) that uses both more sophisticated GAN based embedding learning approach and a more sophisticated seen/unseen error rate balancing based on COSMO. It is important to note that Atzmon & Chechik (2019) did not report the sentence level results on CUB. Yet, when applied on attributes together with f-CLSWGAN (Xian et al., 2018d) COSMO resulted in $0 . 8 \%$ performance drop and when applied with LAGO (Atzmon & Chechik, 2018) it achieved $0 . 5 \%$ improvement over the attribute based state of the art.
+
+![](images/3f384a4e76f0bc5c58077471f4b2954c58239b2f85fada1da5ff845d8c8bbbae.jpg)  
+Figure 3: Harmonic mean Top-1 accuracy on seen and unseen, H, against $\lambda$ , the relative weight of image and text retrieval loss terms. $\lambda = 0$ corresponds to the case of image retrieval loss having weight 1 and text retrieval loss having weight 0. Mean over 10 optimization runs.
+
+![](images/a157ed151bbb217f07fbd4f1af909679c63192783007d447465bf462d40594e9.jpg)  
+Figure 4: The plot of the harmonic mean Top-1 accuracy on seen and unseen, $\mathbf { H }$ , against $\kappa$ , the relative weight of the retrieval and the classification loss terms. $\kappa = 0$ corresponds to the case of classification loss having weight 0. The curves represent the mean over 10 optimization runs.
+
+# 4.4 ON THE SEEN/UNSEEN ACCURACY BALANCING
+
+Figure 2 demonstrates the plot of harmonic mean Top-1 accuracy, $\mathbf { H }$ , against the value of $\alpha$ on the validation sets of CUB and FLOWERS datasets. The validation set is constructed by further splitting the train set on both datasets. For example, CUB has a train set of 5875 images from 100 seen classes and a validation set of 2946 images from 50 unseen classes. We further divide the train set into 4700 train images from 100 seen classes, 1175 seen validation images $( 4 7 0 0 + 1 1 7 5 = 5 8 7 5 )$ and we use all the 2946 images from 50 classes as the unseen validation set. Once the value of $\alpha$ is determined we train the representation on the full train $^ +$ val subset and report results on the test split (the usual practice in GZSL). We confirm on the validation set that H exhibits an inverted U-shape behavior as a function of $\alpha$ , which was theoretically predicted in Section 2.2. Therefore, $\alpha$ can be selected on the validation set and then applied to re-scale the metric space to balance the accuracy on seen and unseen classes during test time as described in Section 2.2.
+
+# 4.5 ABLATION STUDY
+
+Figure 3 presents the results of the ablation study on the importance of image and text retrieval losses. We see that all of the Top-1 accuracies $( \mathbf { H } , \mathbf { s } , \mathbf { u } )$ are stable in the range of $\lambda \in [ 0 . 2 , 0 . 9 ]$ , when both losses have tangible weight. Removing either text or image retrieval losses (setting $\lambda$ to 0 or 1 respectively) leads to performance drop in both cases. Removing the text retrieval loss (case $\lambda = 0$ ) results in the most significant drop. This is due to the fact that the text retrieval task is more tightly related to the GZSL task. At the batch level, retrieving the right text given an image is equivalent to identifying the correct class encoded by a text prototype during ZSL inference step. The image retrieval task is not directly related to solving the ZSL problem and yet it does yield a positive regularizing effect on both CUB and FLOWERS.
+
+Figure 4 shows the results of the ablation study of the interplay between the retrieval loss and the classification loss. We observe, just as in the case with $\lambda$ , that there exists a reasonably flat and stable range of $\kappa \in [ 0 . 2 , 0 . 6 ]$ . The range for $\kappa$ is a bit smaller. $\kappa = 1$ results in the catastrophic performance drop: the classification losses by themselves do not enforce any modality alignment (please refer to Fig. 1 and Algorithm 1 clearly demonstrating this).
+
+Table 3: Generalized zero-shot Top-1 classification accuracy, ablation study.   
+
+<table><tr><td></td><td></td><td></td><td></td><td>CUB</td><td></td><td colspan="3">FLOWERS</td></tr><tr><td>α</td><td>入</td><td>K</td><td>u</td><td>S</td><td>H</td><td>u</td><td>s</td><td>H</td></tr><tr><td>0.0</td><td>0.5</td><td>0.5</td><td>38.3</td><td>65.3</td><td>48.3</td><td>55.1</td><td>84.6</td><td>66.7</td></tr><tr><td>0.0</td><td>0.5</td><td>0.0</td><td>39.3</td><td>57.5</td><td>46.7</td><td>54.0</td><td>78.1</td><td>63.8</td></tr><tr><td>√</td><td>0.5</td><td>0.0</td><td>53.8</td><td>49.6</td><td>51.6</td><td>71.7</td><td>67.2</td><td>69.4</td></tr><tr><td>√</td><td>0.0</td><td>0.5</td><td>47.4</td><td>36.6</td><td>41.3</td><td>51.5</td><td>60.5</td><td>55.6</td></tr><tr><td>√</td><td>1.0</td><td>0.5</td><td>53.9</td><td>53.8</td><td>53.8</td><td>69.5</td><td>73.9</td><td>71.6</td></tr><tr><td>厂</td><td>0.5</td><td>0.5</td><td>59.3</td><td>52.6</td><td>55.8</td><td>73.0</td><td>73.6</td><td>73.3</td></tr></table>
+
+Table 3 studies the effects of different loss terms on the harmonic mean Top-1 accuracy H. The best result is achieved when all loss terms are active and when the metric space rescaling is on (the case of $\lambda = 0 . 5$ , $\kappa = 0 . 5$ and $\alpha$ is checked, the last line in the table). Comparing this with the case when there is no metric space rescaling (first line with $\alpha = 0$ ), we see that the rescaling helps to decrease the gap between seen and unseen classification accuracy. For CUB, the discrepancy reduction is from around $30 \%$ to around $6 \%$ , for FLOWERS it is from around $30 \%$ to around $1 \%$ . We would like to stress that we only use images and texts from the training set to achieve that. Going to the second line in the table (the image/text classification loss is inactive, $\kappa = 0$ ) and comparing it to the first one, we assess the effect of the image/text classification loss. It barely affects the performance on unseen set, but it significantly boosts the classification accuracy on the seen set (around $8 \%$ on both datasets). This is logical: adding a classifier loss results in a better classifier of the test images from the seen classes. This alone does not make it a better GZSL classifier, however. Only when applied together with metric space rescaling, this results in the performance boost (please refer to lines 1 and 6 in Table 3). Our interpretation is that the addition of the image/text classifier loss helps to reduce the intraclass variability in embeddings and provides for tighter clustering. However, this also leads to overfit on the classification task. This is accounted for by metric rescaling that enables the learnings from the image/text classification task be transferred effectively into the GZSL task.
+
+The comparison of the last four rows of Table 3 leads us to believe that all the proposed loss terms outlined in Fig. 1 and Algorithm 1 are important for achieving the state-of-the-art performance. Excluding any one of them (corresponding to the extreme values $\lambda = 0$ , $\lambda = 1$ , $\kappa = 0$ ) leads to performance deterioration. Finally, an interesting observation can be made by comparing line 3 of Table 3 with performance of algorithms in Table 1. In this case our algorithm does not use any class labels and relies on training using exclusively the retrieval losses that can be calculated only based on the pairwise relationships between texts and images. We can see that using this type of supervision alone already results in a very high-quality representation. The representation is competitive against the latest GAN/VAE based approaches on CUB and is state-of-the-art on FLOWERS. This opens up new exploration avenues showing that in the case when very fine-grained modality outputs are available (image and text description pairs being a very prominent example), the high-quality representations may be learned without relying on manually supplied class labels.
+
+# 5 CONCLUSIONS
+
+We propose and empirically validate two improvements to the process of learning fine-grained crossmodal representations. First, we confirm the hypothesis that in the context of paired images and texts, a deep metric learning approach can be driven by an instance-based retrieval loss resulting in competitive generalized zero shot classification results. Combined with an additional class label based image/text crossentropy term this results in state-of-the-art performance on two well known datasets, CUB and FLOWERS. This is an interesting result demonstrating that high-quality deep representations can be trained relying largely on pairwise relationships between modalities. On top of that, we propose a solution to one of the prominent problems in GZSL: classifying instances of unseen classes as seen ones. We mathematically analyze and empirically validate the method of adjusting a single scalar that transcends in its effectiveness advanced dataset augmentation and training approaches based on GANs and VAEs.
+
+# REFERENCES
+
+Zeynep Akata, Scott Reed, Daniel Walter, Honglak Lee, and Bernt Schiele. Evaluation of output embeddings for fine-grained image classification. In CVPR, 2015.
+
+Zeynep Akata, Florent Perronnin, Zaid Harchaoui, and Cordelia Schmid. Label-embedding for image classification. TPAMI, 2016.
+
+Yuval Atzmon and Gal Chechik. Probabilistic AND-OR attribute grouping for zero-shot learning. In UAI, 2018.
+
+Yuval Atzmon and Gal Chechik. Adaptive confidence smoothing for generalized zero-shot learning. In CVPR, 2019.
+
+Wei-Lun Chao, Soravit Changpinyo, Boqing Gong, and Fei Sha. An empirical study and analysis of generalized zero-shot learning for object recognition in the wild. In ECCV (2), pp. 52–68, 2016.
+
+Debasmit Das and C Lee. Zero-shot image recognition using relational matching, adaptation and calibration. In International Joint Conference on Neural Networks, 2019.
+
+Rafael Felix, Vijay Kumar B G, Ian Reid, and Gustavo Carneiro. Multi-modal cycle-consistent generalized zero-shot learning. In ECCV, 2018.
+
+Andrea Frome, Greg S Corrado, Jon Shlens, Samy Bengio, Jeff Dean, Marc Aurelio Ranzato, and Tomas Mikolov. DeViSE: A deep visual-semantic embedding model. In NIPS, pp. 2121–2129, 2013.
+
+Ian Goodfellow, Jean Pouget-Abadie, Mehdi Mirza, Bing Xu, David Warde-Farley, Sherjil Ozair, Aaron Courville, and Yoshua Bengio. Generative adversarial nets. In NIPS, pp. 2672–2680, 2014.
+
+Arthur Gretton, Karsten M Borgwardt, Malte Rasch, Bernhard Schölkopf, and Alex J Smola. A kernel method for the two-sample-problem. In NIPS, 2007.
+
+Geoffrey Hinton, Oriol Vinyals, and Jeffrey Dean. Distilling the knowledge in a neural network. In NIPS Workshop, 2015.
+
+Yao-Hung Hubert Tsai, Liang-Kang Huang, and Ruslan Salakhutdinov. Learning robust visualsemantic embeddings. In CVPR, 2017.
+
+Diederik P. Kingma and Max Welling. Auto-encoding variational bayes. In ICLR, 2014.
+
+C. H. Lampert, H. Nickisch, and S. Harmeling. Attribute-based classification for zero-shot visual object categorization. TPAMI, 2014.
+
+Hugo Larochelle, Dumitru Erhan, and Yoshua Bengio. Zero-data learning of new tasks. In AAAI, 2008.
+
+Yann LeCun, Yoshua Bengio, and Geoffrey Hinton. Deep learning. Nature, 2015.
+
+Shichen Liu, Mingsheng Long, Jianmin Wang, and Michael I Jordan. Generalized zero-shot learning with deep calibration network. In NIPS, 2018.
+
+M. Nilsback and A. Zisserman. Automated flower classification over a large number of classes. In 2008 Sixth Indian Conference on Computer Vision, Graphics Image Processing, 2008.
+
+Mark Palatucci, Dean Pomerleau, Geoffrey E Hinton, and Tom M Mitchell. Zero-shot learning with semantic output codes. In NIPS, 2009.
+
+Genevieve Patterson, Chen Xu, Hang Su, and James Hays. The sun attribute database: Beyond categories for deeper scene understanding. IJCV, 2014.
+
+Scott E. Reed, Zeynep Akata, Honglak Lee, and Bernt Schiele. Learning deep representations of fine-grained visual descriptions. In CVPR, 2016.
+
+Edgar Schönfeld, Sayna Ebrahimi, Samarth Sinha, Trevor Darrell, and Zeynep Akata. Generalized zero-and few-shot learning via aligned variational autoencoders. CVPR, 2019.
+
+Richard Socher, Milind Ganjoo, Christopher D Manning, and Andrew Ng. Zero-shot learning through cross-modal transfer. In NIPS, 2013.   
+Vinay Kumar Verma, Gundeep Arora, Ashish Mishra, and Piyush Rai. Generalized zero-shot learning via synthesized examples. In CVPR, 2018.   
+Wei Wang, Vincent W. Zheng, Han Yu, and Chunyan Miao. A survey of zero-shot learning: Settings, methods, and applications. ACM Trans. Intell. Syst. Technol., 2019.   
+Yaqing Wang and Quanming Yao. Few-shot learning: A survey. In arXiv, 2019.   
+P. Welinder, S. Branson, T. Mita, C. Wah, F. Schroff, S. Belongie, and P. Perona. Caltech-UCSD Birds 200. Technical report, California Institute of Technology, 2010.   
+Yongqin Xian, Zeynep Akata, Gaurav Sharma, Quynh N. Nguyen, Matthias Hein, and Bernt Schiele. Latent embeddings for zero-shot classification. In CVPR, 2016.   
+Yongqin Xian, Christoph H. Lampert, Bernt Schiele, and Zeynep Akata. Pretrained CUB features, 2018a. URL http://datasets.d2.mpi-inf.mpg.de/xian/xlsa17.zip.   
+Yongqin Xian, Christoph H. Lampert, Bernt Schiele, and Zeynep Akata. Pretrained FLOWERS features, 2018b. URL http://datasets.d2.mpi-inf.mpg.de/xian/cvpr18xian. zip.   
+Yongqin Xian, H. Christoph Lampert, Bernt Schiele, and Zeynep Akata. Zero-shot learning: A comprehensive evaluation of the good, the bad and the ugly. TPAMI, 2018c.   
+Yongqin Xian, Tobias Lorenz, Bernt Schiele, and Zeynep Akata. Feature generating networks for zero-shot learning. In CVPR, 2018d.   
+Yongqin Xian, Saurabh Sharma, Bernt Schiele, and Zeynep Akata. f-vaegan-d2: A feature generating framework for any-shot learning. CVPR, 2019.   
+H. Zhang, T. Xu, H. Li, S. Zhang, X. Wang, X. Huang, and D. N. Metaxas. StackGAN $^ { + + }$ : Realistic image synthesis with stacked generative adversarial networks. PAMI, 2018.
+
+# A THE ANALYSIS OF ERROR RATES
+
+We show that $\mathbb { P } \{ \widehat { y } ~ \in ~ \mathcal { Y } ^ { t r } | y _ { v } ~ \in ~ \mathcal { Y } ^ { t s } \} ~ \geq ~ \mathbb { P } \{ \widehat { y } _ { \alpha } ~ \in ~ \mathcal { Y } ^ { t r } | y _ { v } ~ \in ~ \mathcal { Y } ^ { t s } \}$ . Let us define $\delta _ { t r } \equiv$ $\begin{array} { r } { \operatorname* { m i n } _ { y \in \mathcal { y } ^ { t r } } d ( \mathbf { z } _ { v } , \mathbf { p } ( y ) ) } \end{array}$ and $\begin{array} { r } { \delta _ { t s } \equiv \operatorname* { m i n } _ { y \in \mathcal { V } ^ { t s } } d ( \mathbf { z } _ { v } , \mathbf { p } ( y ) ) } \end{array}$ , then Equation (7) can be rewritten as:
+
+$$
+\mathbb { P } \{ \widehat { y } _ { \alpha } \in \mathcal { V } ^ { t r } | y _ { v } \in \mathcal { V } ^ { t s } \} = \mathbb { P } \left\{ ( 1 + \alpha ) \delta _ { t r } < \delta _ { t s } | y _ { v } \in \mathcal { V } ^ { t s } \right\} \mathrm { ~ . ~ }
+$$
+
+Let us consider the probability of event $\delta _ { t r } < \delta _ { t s }$ and decompose it as follows:
+
+$$
+\begin{array} { r l } & { \mathbb { P } \left\{ \delta _ { t r } < \delta _ { t s } \middle | y _ { v } \in \mathcal { Y } ^ { t s } \right\} = \mathbb { P } \left\{ ( 1 + \alpha ) \delta _ { t r } < ( 1 + \alpha ) \delta _ { t s } | y _ { v } \in \mathcal { Y } ^ { t s } \right\} } \\ & { = \mathbb { P } \left\{ ( 1 + \alpha ) \delta _ { t r } < \delta _ { t s } \cup \delta _ { t s } \leq ( 1 + \alpha ) \delta _ { t r } < ( 1 + \alpha ) \delta _ { t s } | y _ { v } \in \mathcal { Y } ^ { t s } \right\} } \\ & { = \mathbb { P } \left\{ ( 1 + \alpha ) \delta _ { t r } < \delta _ { t s } | y _ { v } \in \mathcal { Y } ^ { t s } \right\} + \mathbb { P } \left\{ \delta _ { t s } \leq ( 1 + \alpha ) \delta _ { t r } < ( 1 + \alpha ) \delta _ { t s } | y _ { v } \in \mathcal { Y } ^ { t s } \right\} } \\ & { - \mathbb { P } \left\{ ( 1 + \alpha ) \delta _ { t r } < \delta _ { t s } \cap \delta _ { t s } \leq ( 1 + \alpha ) \delta _ { t r } < ( 1 + \alpha ) \delta _ { t s } | y _ { v } \in \mathcal { Y } ^ { t s } \right\} } \\ & { = \mathbb { P } \left\{ ( 1 + \alpha ) \delta _ { t r } < \delta _ { t s } | y _ { v } \in \mathcal { Y } ^ { t s } \right\} + \mathbb { P } \left\{ \delta _ { t s } \leq ( 1 + \alpha ) \delta _ { t r } < ( 1 + \alpha ) \delta _ { t s } | y _ { v } \in \mathcal { Y } ^ { t s } \right\} } \end{array}
+$$
+
+The transitions are based on the relationship between probabilities of arbitrary events $A$ and $B$ , $\mathbb { P } \{ A \cup B \} = \mathbb { P } \{ A \} + \mathbb { P } \{ B \} - \mathbb { P } \{ A \cap B \}$ , and in our case ${ \mathbb { P } } \{ A \cap B \} = 0$ . This implies that:
+
+$$
+\begin{array} { r l r } {  { \mathbb { P } \{ \widehat { y } _ { \alpha } \in \mathcal { Y } ^ { t r } | y _ { v } \in \mathcal { Y } ^ { t s } \} = \mathbb { P } \{ \widehat { y } \in \mathcal { Y } ^ { t r } | y _ { v } \in \mathcal { Y } ^ { t s } \} - \mathbb { P } \{ \frac { \delta _ { t s } } { ( 1 + \alpha ) } \leq \delta _ { t r } < \delta _ { t s } | y _ { v } \in \mathcal { Y } ^ { t s } \} } } \\ & { } & { \leq \mathbb { P } \{ \widehat { y } \in \mathcal { Y } ^ { t r } | y _ { v } \in \mathcal { Y } ^ { t s } \} . } \end{array}
+$$
+
+We have just shown that for a non-negative $\alpha$ the probability of misclassifying an image from an unseen class as one of the seen classes is smaller for the decision rule $\widehat { y } _ { \alpha }$ than for the original decision rule $\widehat { y }$ . In fact, we can make a stronger claim. Since $\delta _ { t s }$ and $\delta _ { t r }$ are non-negative, it is bclear that the length of interval $[ \delta _ { t s } / ( 1 + \alpha ) , \delta _ { t s } \overline { { ) } }$ increases as $\alpha$ increases, and hence probability that $\delta _ { t r }$ falls in this interval is non-decreasing with increasing $\alpha$ . Thus we have for any $0 \leq \alpha _ { 1 } \leq \alpha _ { 2 }$ , $\mathbb { P } \{ \widehat { y } _ { \alpha _ { 1 } } \in \mathcal { Y } ^ { t r } | y _ { v } \in \mathcal { Y } ^ { t s } \} \geq \mathbb { P } \{ \widehat { y } _ { \alpha _ { 2 } } \in \mathcal { Y } ^ { \bar { t } r } | y _ { v } \in \mathcal { Y } ^ { t s } \}$ , i.e. $\mathbb { P } \{ \widehat { y } _ { \alpha } \in \mathcal { V } ^ { t r } | y _ { v } \in \mathcal { V } ^ { t s } \}$ is a monotone bnon-increasing function of $\alpha$ b band we can reduce it by increasing $\alpha$ .
\ No newline at end of file
diff --git a/parse/train/SJl3h2EYvS/SJl3h2EYvS_content_list.json b/parse/train/SJl3h2EYvS/SJl3h2EYvS_content_list.json
new file mode 100644
index 0000000000000000000000000000000000000000..f4338fb4d6f2fbb0540cad4502283c122aa7fd35
--- /dev/null
+++ b/parse/train/SJl3h2EYvS/SJl3h2EYvS_content_list.json
@@ -0,0 +1,1106 @@
+[
+    {
+        "type": "text",
+        "text": "CLAREL: CLASSIFICATION VIA RETRIEVAL LOSS FOR ZERO-SHOT LEARNING ",
+        "text_level": 1,
+        "bbox": [
+            174,
+            99,
+            823,
+            146
+        ],
+        "page_idx": 0
+    },
+    {
+        "type": "text",
+        "text": "Anonymous authors Paper under double-blind review ",
+        "bbox": [
+            183,
+            170,
+            398,
+            198
+        ],
+        "page_idx": 0
+    },
+    {
+        "type": "text",
+        "text": "ABSTRACT ",
+        "text_level": 1,
+        "bbox": [
+            454,
+            234,
+            544,
+            251
+        ],
+        "page_idx": 0
+    },
+    {
+        "type": "text",
+        "text": "We address the problem of learning fine-grained cross-modal representations. We propose an instance-based deep metric learning approach in joint visual and textual space. The key novelty of this paper is that it shows that using per-image semantic supervision leads to substantial improvement in zero-shot performance over using class-only supervision. On top of that, we provide a probabilistic justification for a metric rescaling approach that solves a very common problem in the generalized zero-shot learning setting, i.e., classifying test images from unseen classes as one of the classes seen during training. We evaluate our approach on two finegrained zero-shot learning datasets: CUB and FLOWERS. We find that on the generalized zero-shot classification task CLAREL consistently outperforms the existing approaches on both datasets. ",
+        "bbox": [
+            233,
+            268,
+            766,
+            421
+        ],
+        "page_idx": 0
+    },
+    {
+        "type": "text",
+        "text": "1 INTRODUCTION ",
+        "text_level": 1,
+        "bbox": [
+            176,
+            453,
+            336,
+            469
+        ],
+        "page_idx": 0
+    },
+    {
+        "type": "text",
+        "text": "Deep learning-based approaches have demonstrated superior flexibility and generalization capabilities in information processing on a wide variety of tasks, such as vision, speech and language (LeCun et al., 2015). However, it has been widely realized that the transfer of deep representations to real-world applications is challenging due to the typical reliance on massive hand-labeled datasets. Learning in the low-labeled data regime, especially in the zero-shot (Wang et al., 2019) and the few-shot (Wang & Yao, 2019) setups, have recently received significant attention in the literature. In the problem of zero-shot learning (ZSL), the objective is to recognize categories that have not been seen during the training (Larochelle et al., 2008). This is typically done by relying on anchor embeddings learned in one modality as prototypes and by associating a query embedding from the other modality with the closest prototype. In the generalized ZSL (GZSL) case (Xian et al., 2018c), the objective is more challenging as recognition is performed in the joint space of seen and unseen categories. ZSL, as well as its generalized counterpart, provide a viable framework to learn cross-modal representations that are flexible and adaptive. For example, in this paradigm, the adaptation to a new classification task based on text/image representation space alignment could be as easy as defining/appending/modifying a set of text sentences to define classes of new classifiers. This is an especially relevant problem as machine learning is challenged with the long tail of classes, and the idea of learning from pairs of images and sentences, abundant on the web, looks like a natural solution. Therefore, in this paper we specifically target the fine-grained scenario of paired images and their respective text descriptions. The uniqueness of this scenario is in the fact that the co-occurance of image and text provides a rich source of information. The ways of leveraging this source have not been sufficiently explored in the context of GZSL. Although we focus exclusively on the GZSL recognition setup in this paper, we believe that the research in this direction has potential to enable zero-shot flexibility in a wider array of high-level tasks such as segmentation or conditional image generation (Zhang et al., 2018). The contributions of this work can be characterized under the following two themes. ",
+        "bbox": [
+            174,
+            487,
+            825,
+            819
+        ],
+        "page_idx": 0
+    },
+    {
+        "type": "text",
+        "text": "Instance-based training loss. Most prominent zero-shot learning approaches rely heavily on classlevel modality alignment (Xian et al., 2018c). We propose a new composite loss function that balances instance-based pairwise image/text retrieval loss and the usual classifier loss. The retrieval loss term does not use class labels. We demonstrate that the class-level information is important, but in the fine-grained text/image pairing scenarios, most of the GZSL accuracy can be extracted from the instance-based retrieval loss. To the best of our knowledge, this type of training has not been used in the GZSL literature. Its impressive performance opens up new promising research directions. ",
+        "bbox": [
+            174,
+            827,
+            825,
+            924
+        ],
+        "page_idx": 0
+    },
+    {
+        "type": "image",
+        "img_path": "images/c5306e54b7867e3eb7aee45c6499213721005da29ec547a7ed5a3a76968503c1.jpg",
+        "image_caption": [
+            "Figure 1: The architecture and training diagram describing the proposed method. Each batch consists of randomly sampled instances, i.e. pairs of images and their corresponding texts. Images are embedded via ResNet and texts are embedded via a CNN/LSTM stack. Image and text features are projected via a fully connected layer into the same dimensional space. In this space, distances between text and image features from different instances are computed. The negative distances are fed into softmax to train on both the image and the text retrieval tasks. The image retrieval task consists of retrieving the image corresponding to the given text of the same instance and the text retrieval task is vice versa. In addition to that, image and text embeddings are trained on auxiliary image and text classification tasks on the class labels corresponding to instances. "
+        ],
+        "image_footnote": [],
+        "bbox": [
+            250,
+            98,
+            753,
+            291
+        ],
+        "page_idx": 1
+    },
+    {
+        "type": "text",
+        "text": "Metric space rescaling. Metric-based ZSL approaches rely on distances between prototypes and query embeddings during inference. They are known to suffer from imbalanced performance on seen and unseen classes (Liu et al., 2018). Previous work proposed to use a heuristic trick, calibrated stacking (Chao et al., 2016) or calibration (Das & Lee, 2019), to solve the problem. We refer to this technique as metric rescaling in our work, and provide a sound probabilistic justification for it. ",
+        "bbox": [
+            174,
+            457,
+            825,
+            527
+        ],
+        "page_idx": 1
+    },
+    {
+        "type": "text",
+        "text": "2 PROPOSED METHOD ",
+        "text_level": 1,
+        "bbox": [
+            176,
+            549,
+            375,
+            565
+        ],
+        "page_idx": 1
+    },
+    {
+        "type": "text",
+        "text": "In this paper, we specifically target the fine-grained visual description scenario, as defined by Reed et al. (2016). In this setting, the dataset consists of a number of images from a given set of classes and each image is accompanied by a number of textual descriptions. The task is to learn a joint representation space for images and texts that can be used for zero-shot recognition. An instance of the zero-shot multimodal representation learning problem can then be defined as follows. Given a training set $S = \\{ ( v _ { n } , t _ { n } , y _ { n } ) \\mid v _ { n } \\in \\mathcal { V }$ , $t _ { n } \\in \\mathcal { T } , y _ { n } \\in \\mathcal { V } , n = 1 \\ldots N \\}$ of image, text and label tuples, we are interested in finding representations $f _ { \\phi } : \\mathcal { V } \\to \\mathcal { Z }$ of image, parameterized by $\\phi$ , and $f _ { \\theta } : \\mathcal { T }  \\mathcal { Z }$ of text, parameterized by $\\theta$ , in a common embedding space $\\mathcal { Z }$ . Furthermore, GZSL problem is defined using the sets of seen $\\mathcal { V } ^ { t r }$ and unseen $\\mathcal { V } ^ { t s }$ classes, such that $\\mathcal { V } = \\mathcal { V } ^ { t r } \\cup \\mathcal { V } ^ { t s }$ and $\\bar { \\mathcal { V } } ^ { t r } \\cap \\mathcal { V } ^ { t s } = \\emptyset$ . The training set will then only contain the seen classes, i.e. $S ^ { t r } = \\{ ( v _ { n } , t _ { n } , y _ { n } ) \\mid v _ { n } \\in$ $\\nu$ , $t _ { n } \\in \\mathcal { T } , y _ { n } \\in \\mathcal { y } ^ { t r } \\}$ and the task is to build a classifier function $g : \\mathcal { Z } \\times \\mathcal { Z }  \\mathcal { V }$ . This is different from the ZSL scenario focusing on $g : \\mathcal { Z } \\times \\mathcal { Z }  \\mathcal { Y } ^ { t s }$ . ",
+        "bbox": [
+            173,
+            583,
+            825,
+            751
+        ],
+        "page_idx": 1
+    },
+    {
+        "type": "text",
+        "text": "To build $g$ , most approaches to joint representation learning rely on class labeling to train a representation. For example, all the methods reviewed by Xian et al. (2018c) require the access to class labels at train time. We hypothesise that in the fine-grained learning scenario, such as the one described by Reed et al. (2016), a lot of information can be extracted simply from pairwise image/text cooccurrences. The class labels really only become critically necessary when we define class prototypes, i.e. at zero-shot test time. Following this intuition, we define a composite loss function that relies both on the pairwise relationships and on the class labels. The high-level description of the proposed framework is depicted in Figure 1. The framework enables us, among other things, to experiment with the effects of train-time availability of class labels on the quality of zero-shot representations. The framework is based on projecting texts and images into a common space and then learning a representation based on a mixture of four loss functions: a pairwise text retrieval loss, a pairwise image retrieval loss, a text classifier loss and an image classifier loss (see Algorithm 1). ",
+        "bbox": [
+            173,
+            757,
+            826,
+            924
+        ],
+        "page_idx": 1
+    },
+    {
+        "type": "text",
+        "text": "Algorithm 1 Loss calculation for a single optimization iteration of the proposed method. $N$ is the number of instances in the training set ${ \\mathcal { S } } ^ { t r }$ , $B$ is the number of instances per batch, $C$ is the number of classes in the train set. $\\mathrm { R A N D O M S A M P L E } ( S , B )$ denotes a set of $B$ elements chosen uniformly at random from a set $s$ , without replacement. ",
+        "bbox": [
+            173,
+            102,
+            825,
+            160
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "table",
+        "img_path": "images/18df0181b501f401a7b2d506b2c75380a51ed2ff3f2b3eb40ec700d88788894d.jpg",
+        "table_caption": [],
+        "table_footnote": [],
+        "table_body": "<table><tr><td colspan=\"3\">Input: Training set Str = {(U1,t1,yi),...,(UN,tN,yN)},λ ∈ [0,1], κ ∈ [0,1]. Output: The loss J(,θ) for a randomly sampled training batch. I← RANDOMSAMPLE({1,...,N},B) &gt;Select B instance indices for batch</td></tr><tr><td>Jrc(Θ),J1c(Φ)←0,0 foriinZdo</td><td colspan=\"3\">Initialize classification losses Embed images and texts Image classifier probabilities</td></tr><tr><td colspan=\"3\">Zui,Zt←f(Ui）,fθ(ti)</td></tr><tr><td colspan=\"2\">p ← softmax(Wizui +b1) PT ← sOftmax(WTZt +br)</td></tr><tr><td colspan=\"2\">Textclassifierprobabilities JTc(0) ← JTc(0)+ crossentropy(pr,yi) Text classification loss</td></tr><tr><td colspan=\"2\">JIc(𝜙) ← Jic(Φ)+ crossentropy(p1, yi) Image classification loss end for</td></tr><tr><td colspan=\"2\">JTR(Φ,0),J1R(Φ,0)←0,0 Initialize retrieval losses</td></tr><tr><td colspan=\"2\">foriinZdo 1 Text retrieval loss</td></tr><tr><td colspan=\"2\">JTR(Φ,0)←JTR(Φ,0)+ d(zu,Zt）+log∑ exp(-d(Zu,Ztj)) B j∈I</td></tr><tr><td colspan=\"2\">1-B JIR(Φ,0)←JIR(Φ,0)+ d(Zui,Zti) + log∑exp(-d(Zt , Zuj))|&gt; Image retrieval loss</td></tr><tr><td colspan=\"2\">j∈I</td></tr><tr><td colspan=\"2\">J(,θ)←λJTR(𝜙,0)+(1-λ)JIR(Φ,0) Add retrieval loss to the total loss J(Φ,θ)←(1-κ)J(,0)+(Jrc(0)+JIc()) Add classification loss to the total loss</td></tr></table>",
+        "bbox": [
+            174,
+            162,
+            836,
+            487
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "text",
+        "text": "2.1 RETRIEVAL LOSS FUNCTION ",
+        "text_level": 1,
+        "bbox": [
+            176,
+            511,
+            411,
+            525
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "text",
+        "text": "Pairwise cross-modal loss function is based solely on the pairwise relationships between texts and images. We choose to use the metric learning approach to capture the relationship between images and texts. Now, suppose $d$ is a metric $d : \\mathcal { Z } \\times \\mathcal { Z }  \\mathbb { R } ^ { + }$ , $v _ { i }$ is an image and $\\tau = \\{ t _ { j ^ { \\prime } } \\}$ is a collection of arbitrary texts sampled uniformly at random, of which text $t _ { j }$ belongs to $v _ { i }$ . We propose the following model for the probability of image $v _ { i }$ and text $t _ { j }$ to belong to the same object instance: ",
+        "bbox": [
+            173,
+            536,
+            825,
+            608
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "equation",
+        "img_path": "images/66c9fefbdd5debf1a919106a675316eaef51240da8ba8bff3384cde5d307f350.jpg",
+        "text": "$$\np _ { \\phi , \\theta } ( i = j | v _ { i } , t _ { j } , \\tau ) = \\frac { \\exp ( - d ( f _ { \\phi } ( v _ { i } ) , f _ { \\theta } ( t _ { j } ) ) ) } { \\sum _ { t _ { j ^ { \\prime } } \\in \\tau } \\exp ( - d ( f _ { \\phi } ( v _ { i } ) , f _ { \\theta } ( t _ { j ^ { \\prime } } ) ) ) } \\ .\n$$",
+        "text_format": "latex",
+        "bbox": [
+            305,
+            612,
+            692,
+            650
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "text",
+        "text": "The learning is then based on the following cross-entropy loss defined on the batch of size $B$ : ",
+        "bbox": [
+            169,
+            652,
+            784,
+            667
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "equation",
+        "img_path": "images/2b97ee593256906efa0bca123ca74f839ff6c75fd4f2cca4d9bdd6cf46a14163.jpg",
+        "text": "$$\nJ _ { T R } ( \\phi , \\theta ) = - \\frac { 1 } { B } \\sum _ { i , j = 1 } ^ { B } \\ell _ { i , j } \\log p _ { \\phi , \\theta } ( i = j | v _ { i } , t _ { j } , \\{ t _ { j ^ { \\prime } } \\} _ { j ^ { \\prime } = 1 } ^ { B } ) ,\n$$",
+        "text_format": "latex",
+        "bbox": [
+            302,
+            671,
+            694,
+            715
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "text",
+        "text": "where $\\ell _ { i , j }$ is a binary indicator of the true match $\\mathcal { \\ell } _ { i , j } = 1$ , if $i = j$ and 0 otherwise). Note that the expression above has the interpretation of the text retrieval loss. It attains its smallest value when for each image in the batch we manage to assign probability 1 to its respective text and 0 to all other texts. This can be further expanded as: ",
+        "bbox": [
+            174,
+            719,
+            825,
+            776
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "equation",
+        "img_path": "images/e94db0ef9c2a0a6f62952bdf1653f0a4f3ffe6247f5fd866035cda9e491a975b.jpg",
+        "text": "$$\nJ _ { T R } ( \\phi , \\theta ) = \\frac { 1 } { B } \\sum _ { i = 1 } ^ { B } \\left( d ( f _ { \\phi } ( v _ { i } ) , f _ { \\theta } ( t _ { i } ) ) + \\log \\Big [ \\sum _ { t _ { j ^ { \\prime } } \\in \\tau } \\exp ( - d ( f _ { \\phi } ( v _ { i } ) , f _ { \\theta } ( t _ { j ^ { \\prime } } ) ) ) \\Big ] \\right) .\n$$",
+        "text_format": "latex",
+        "bbox": [
+            223,
+            779,
+            776,
+            829
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "text",
+        "text": "Exchanging the order of image and text in the probability model (1) leads to the image retrieval loss, $J _ { I R } ( \\phi , \\bar { \\theta } )$ . The two losses are mixed using parameter $\\lambda \\in [ 0 , 1 ]$ as shown in Algorithm 1. ",
+        "bbox": [
+            174,
+            832,
+            823,
+            862
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "text",
+        "text": "The pairwise retrieval loss functions are responsible for the modality alignment. In addition to those, we propose to include, as mentioned above, the usual image and text classifier losses. These losses are responsible for reducing the intraclass variability of representations. The classifier losses are added to the retrieval losses using a mixing parameter $\\kappa \\in [ 0 , 1 ]$ as shown in Algorithm 1. ",
+        "bbox": [
+            174,
+            867,
+            825,
+            924
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "text",
+        "text": "2.2 BALANCING ACCURACY ON THE SEEN AND UNSEEN CLASSES ",
+        "text_level": 1,
+        "bbox": [
+            174,
+            103,
+            648,
+            118
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "Let us define class prototypes $\\mathbf p ( y )$ , each based on the set of texts $\\mathcal { T } _ { y }$ belonging to class $y$ , $\\mathbf { \\{ p ( }  y \\mathbf { ) = }$ $\\begin{array} { r } { \\frac { 1 } { | T _ { y } | } \\sum _ { t _ { i } \\in \\mathcal { T } _ { y } } f _ { \\theta } ( t _ { i } ) \\ | \\ y \\in \\mathcal { V } \\} } \\end{array}$ . In the context of GZSL, the nearest neighbor decision rule for a given image $v$ and its features ${ \\bf z } _ { v } = f _ { \\phi } ( v )$ has the following form: ",
+        "bbox": [
+            173,
+            128,
+            825,
+            180
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "equation",
+        "img_path": "images/3d3a73b561686471c4dfe86df22f349395ea59a86c528b9db4f01ce94ccf5acf.jpg",
+        "text": "$$\n\\boldsymbol { \\widehat { y } } = \\arg \\operatorname* { m i n } _ { \\boldsymbol { y } \\in \\mathcal { V } } d ( \\mathbf { z } _ { v } , \\mathbf { p } ( \\boldsymbol { y } ) ) .\n$$",
+        "text_format": "latex",
+        "bbox": [
+            410,
+            188,
+            586,
+            212
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "The most acute problem faced in this setup is the accuracy imbalance between seen and unseen classes. A very representative case clearly outlining the imbalance problem is presented in Table 6 of (Xian et al., 2018c), where accuracy on the seen classes is always significantly greater than the accuracy on unseen ones. In order to measure and control the imbalance, three metrics are commonly used to assess the classification performance in the GZSL scenario: the Top-1 accuracy on the seen categories (s), the Top-1 accuracy on the unseen categories $\\mathbf { \\Pi } ^ { ( \\mathbf { u } ) }$ and their harmonic mean, $\\mathbf { H } = \\mathbf { u } \\cdot \\mathbf { s } / ( \\mathbf { u } + \\mathbf { s } )$ . The main metric to assess GZSL performance is then $\\mathbf { H }$ , which quantifies both $\\mathbf { u }$ and s. ",
+        "bbox": [
+            173,
+            219,
+            825,
+            319
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "To formalize the problem, we first introduce $y _ { v }$ , the true class label of image $v$ . Mathematically, the main GZSL pain point is that $\\mathbb { P } \\{ \\widehat { y } \\in \\mathcal { y } ^ { t r } | y _ { v } \\in \\mathcal { y } ^ { t s } \\}$ is significantly greater than $\\mathbb { P } \\{ \\widehat { y } \\in \\mathcal { y } ^ { t s } | y _ { v } \\in$ $\\mathcal { V } ^ { t r } \\}$ b b. In other words, the problem is that a given image is more likely to be confused with one of the seen classes if it belongs to an unseen class than vice versa. Our approach to solving the problem is based on the following probabilistic representation of the event space for the decision rule in Equation (4): ",
+        "bbox": [
+            173,
+            324,
+            825,
+            410
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "equation",
+        "img_path": "images/c4992af813698d95795db7215aceffe43193c03d6626652d1543fc22759ab44a.jpg",
+        "text": "$$\n\\mathbb { P } \\{ \\boldsymbol { { \\widehat { y } } } \\in { \\mathcal { Y } } ^ { t r } | y _ { v } \\in { \\mathcal { Y } } ^ { t s } \\} = \\mathbb { P } \\left\\{ \\operatorname* { m i n } _ { y \\in { \\mathcal { Y } } ^ { t r } } d ( \\mathbf { z } _ { v } , \\mathbf { p } ( y ) ) < \\operatorname* { m i n } _ { y \\in { \\mathcal { Y } } ^ { t s } } d ( \\mathbf { z } _ { v } , \\mathbf { p } ( y ) ) \\mid y _ { v } \\in { \\mathcal { Y } } ^ { t s } \\right\\} .\n$$",
+        "text_format": "latex",
+        "bbox": [
+            227,
+            416,
+            769,
+            449
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "Rephrasing, the most acute GZSL error happens when the prototype of one of the seen classes is closer to an image embedding from an unseen class than any of the prototypes of the unseen classes. ",
+        "bbox": [
+            173,
+            457,
+            825,
+            486
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "To rectify the situation we propose the following very direct solution to balance $\\mathbb { P } \\{ \\widehat { y } \\in \\mathcal { y } ^ { t r } | y _ { v } \\in \\mathcal { y } ^ { t s } \\}$ and $\\mathbb { P } \\{ \\hat { \\mathcal { Y } } \\in \\mathcal { Y } ^ { t s } | y _ { v } \\in \\mathcal { Y } ^ { t \\bar { r } } \\}$ . We introduce a positive scalar $\\alpha \\in \\mathbb { R } ^ { + }$ band scale all the distances bcorresponding to the seen prototypes by $1 + \\alpha$ . This gives rise to the following scaled distance $d _ { \\alpha }$ : ",
+        "bbox": [
+            174,
+            492,
+            825,
+            535
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "equation",
+        "img_path": "images/a62834e649aa08ed171d61fabc0cba31952b1b0e4dc94b2e737d501dd6da9e62.jpg",
+        "text": "$$\nd _ { \\alpha } ( \\mathbf { z } _ { v } , \\mathbf { p } ( y ) ) = \\left\\{ \\begin{array} { l l } { ( 1 + \\alpha ) d ( \\mathbf { z } _ { v } , \\mathbf { p } ( y ) ) , } & { \\mathrm { i f ~ } y \\in \\mathcal { Y } ^ { t r } } \\\\ { d ( \\mathbf { z } _ { v } , \\mathbf { p } ( y ) ) , } & { \\mathrm { o t h e r w i s e } } \\end{array} \\right. .\n$$",
+        "text_format": "latex",
+        "bbox": [
+            323,
+            542,
+            676,
+            582
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "The misclassification between unseen as seen classes for the classifier $\\widehat { y } _ { \\alpha }$ , based on (6) is then: ",
+        "bbox": [
+            174,
+            588,
+            790,
+            603
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "equation",
+        "img_path": "images/6c11b414afece46ca498ac50b8688e023ff6de855077e8453e68919cc2b737ba.jpg",
+        "text": "$$\n\\mathbb { P } \\{ \\widehat { y } _ { \\alpha } \\in \\mathcal { Y } ^ { t r } | y _ { v } \\in \\mathcal { Y } ^ { t s } \\} = \\mathbb { P } \\left\\{ \\left( 1 + \\alpha \\right) \\operatorname* { m i n } _ { y \\in \\mathcal { Y } ^ { t r } } d ( \\mathbf { z } _ { v } , \\mathbf { p } ( y ) ) < \\operatorname* { m i n } _ { y \\in \\mathcal { Y } ^ { t s } } d ( \\mathbf { z } _ { v } , \\mathbf { p } ( y ) ) | y _ { v } \\in \\mathcal { Y } ^ { t s } \\right\\} \\ ,\n$$",
+        "text_format": "latex",
+        "bbox": [
+            192,
+            609,
+            784,
+            643
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "and it has the following property: for any $0 \\leq \\alpha _ { 1 } \\leq \\alpha _ { 2 }$ , $\\mathbb { P } \\{ \\widehat { y } _ { \\alpha _ { 1 } } ~ \\in ~ \\mathcal { Y } ^ { t r } | y _ { v } ~ \\in ~ \\mathcal { Y } ^ { t s } \\} ~ \\geq ~ \\mathbb { P } \\{ \\widehat { y } _ { \\alpha _ { 2 } } ~ \\in ~$ $\\mathcal { V } ^ { t r } | y _ { v } \\in \\mathcal { V } ^ { t s } \\}$ , i.e. $\\mathbb { P } \\{ \\widehat { y } _ { \\alpha } \\in \\mathcal { V } ^ { t \\bar { r } } | y _ { v } \\in \\mathcal { V } ^ { t s } \\}$ b   is a monotone non-increasing function of $\\alpha$ b and we bcan reduce it by increasing $\\alpha$ (please refer to Appendix A for a proof). Consider now $\\mathbb { P } \\{ \\widehat { y } _ { \\alpha } \\in$ $\\mathcal { V } ^ { t r } | y _ { v } \\in \\mathcal { V } ^ { t r } \\big \\}$ , which is a probability that we classify an image $v$ bfrom one of the seen classes as still one of the seen classes. Using exactly the same chain of arguments as in Appendix A, it is straightforward to show that the probability is a non-increasing function of $\\alpha$ . Hence the probability $\\mathbb { P } \\{ \\hat { y } _ { \\alpha } ^ { \\sim } \\in \\mathcal { V } ^ { t s } | y _ { v } \\in \\mathcal { V } ^ { t r } \\} = 1 - \\mathbb { P } \\{ \\hat { y } _ { \\alpha } \\in \\mathcal { V } ^ { \\hat { t r } } | y _ { v } \\in \\mathcal { V } ^ { t r } \\}$ is a non-decreasing function of $\\alpha$ . Therefore, as $\\alpha$ increases, we expect more classification errors in classifying images from seen classes, because some of them will be classified as one of the unseen classes. ",
+        "bbox": [
+            173,
+            651,
+            825,
+            779
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "To sum up, given the arguments presented above we expect that by varying $\\alpha > 0$ we can balance the error rate $\\mathbb { P } \\mathbf { \\bar { \\{ y _ { \\alpha } \\in \\mathcal { V } ^ { t r } \\vert y _ { v } \\in \\mathcal { V } ^ { t s } \\} } }$ of leaking the unseen class images into seen class classification bdecision and the error rate $\\mathbb { P } \\{ \\widehat { y } _ { \\alpha } ^ { } \\in \\mathcal { V } ^ { t s } | y _ { v } ^ { - } \\in \\mathcal { V } ^ { t r } \\}$ of leaking the seen class images into unseen bclass classification decision. This is possible as we just showed above that $\\mathbb { P } \\{ \\widehat { y } _ { \\alpha } \\in \\mathcal { V } ^ { t r } | y _ { v } \\in \\mathcal { V } ^ { t s } \\}$ is a non-increasing function of $\\alpha$ , while $\\mathbb { P } \\{ \\widehat { y } _ { \\alpha } \\in \\mathcal { V } ^ { t s } | y _ { v } \\in \\mathcal { V } ^ { t r } \\}$ b is a non-decreasing one. It is also important to emphasize that $\\alpha$ bis applied only to distances between the query embedding and the prototypes of seen classes and it is constant over seen classes. Therefore, the application of $\\alpha$ does not at all affect the classification error rates either within $\\mathcal { V } ^ { t r }$ or within $\\mathcal { V } ^ { t s }$ . Varying $\\alpha$ balances exclusively the classification errors arising from transitions between seen and unseen class labels. We study the empirical aspects of balancing $\\alpha$ in Section 4.4. ",
+        "bbox": [
+            173,
+            784,
+            825,
+            924
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "table",
+        "img_path": "images/12bbcc390447bbfe9373f37f6f92b9541094bd9ffb0478a62b265d975ef854ee.jpg",
+        "table_caption": [
+            "Table 1: Generalized zero-shot Top-1 classification accuracy. "
+        ],
+        "table_footnote": [],
+        "table_body": "<table><tr><td></td><td colspan=\"3\">CUB</td><td colspan=\"3\">FLOWERS</td></tr><tr><td></td><td>u</td><td>S</td><td>H</td><td>u</td><td>s</td><td>H</td></tr><tr><td>CADA-VAE (Schönfeld et al., 2019)</td><td>n/a</td><td>n/a</td><td>53.4</td><td>n/a</td><td>n/a</td><td>n/a</td></tr><tr><td>f-CLSWGAN (Xian et al., 2018d)</td><td>50.3</td><td>58.3</td><td>54.0</td><td>59.0</td><td>73.8</td><td>65.6</td></tr><tr><td>f-VAEGAN-D2 (Xian et al.,2019)</td><td>48.4</td><td>60.1</td><td>53.6</td><td>56.8</td><td>74.9</td><td>64.6</td></tr><tr><td>cycle-(U)WGAN (Felix et al., 2018)</td><td>47.9</td><td>59.3</td><td>53.0</td><td>61.6</td><td>69.2</td><td>65.2</td></tr><tr><td>COSMO+f-CLSWGAN (Atzmon &amp; Chechik, 2019)</td><td>n/a</td><td>n/a</td><td>n/a</td><td>59.6</td><td>81.4</td><td>68.8</td></tr><tr><td>CLAREL (Ours)</td><td>59.3</td><td>52.6</td><td>55.8</td><td>73.0</td><td>73.6</td><td>73.3</td></tr></table>",
+        "bbox": [
+            173,
+            127,
+            846,
+            250
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "table",
+        "img_path": "images/892a986457f15b9754b348d5c3429df71973148b0d2ec102398a7aeb1a84ccab.jpg",
+        "table_caption": [
+            "Table 2: Zero-shot Top-1 classification accuracy. "
+        ],
+        "table_footnote": [],
+        "table_body": "<table><tr><td></td><td>CUB</td><td>FLOWERS</td></tr><tr><td>CADA-VAE (Schonfeld et al., 2019)</td><td>n/a</td><td>n/a</td></tr><tr><td>f-CLSWGAN (Xian et al., 2018d)</td><td>57.3</td><td>67.2</td></tr><tr><td>f-VAEGAN-D2 (Xian et al., 2019)</td><td>61.0</td><td>67.7</td></tr><tr><td>cycle-(U)WGAN (Felix et al., 2018)</td><td>58.6</td><td>70.3</td></tr><tr><td>CLAREL (Ours)</td><td>66.7</td><td>76.8</td></tr></table>",
+        "bbox": [
+            282,
+            291,
+            715,
+            387
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "text",
+        "text": "3 RELATED WORK ",
+        "text_level": 1,
+        "bbox": [
+            176,
+            414,
+            344,
+            430
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "text",
+        "text": "ZSL approaches aim at recognizing objects belonging to classes unseen during training (Larochelle et al., 2008; Palatucci et al., 2009). This has been extended to the GZSL framework in which the decision space consists of both seen and unseen classes (Socher et al., 2013; Xian et al., 2018c). The classical zero-shot approaches build a joint visual-semantic space, relying on a linear cross-modal compatibility function (e.g. dot-product between query embedding and semantic prototypes or a variation of a hinge loss) (Frome et al., 2013; Akata et al., 2015; 2016; Reed et al., 2016). Non-linear variants of the compatibility has also been explored (Xian et al., 2016; Socher et al., 2013). Extending previously proposed cross-modal transfer approaches based on auto-encoders (Hubert Tsai et al., 2017) and cross-domain learning (Gretton et al., 2007), more recent line of work (Schönfeld et al., 2019; Xian et al., 2018d; 2019; Felix et al., 2018; Verma et al., 2018) relies on combining these approaches and their variations with dataset augmentation tools such as GAN (Goodfellow et al., 2014) and VAE (Kingma & Welling, 2014). It is argued that the use of those tools helps to resolve one of the prominent problems in GZSL scenario: classifying images from unseen classes as one of the seen classes. There exist approaches that try to tackle this same problem via temperature calibration (Liu et al., 2018) originally proposed by Hinton et al. (2015). Chao et al. (2016); Das & Lee (2019) proposed an approach to seen/unseen accuracy balancing that is very similar to ours, based on heuristic arguments. We extend this line of work here by providing a probabilistic justification for the balancing effect observed when applying metric rescaling. Atzmon & Chechik (2019) propose a more sophisticated way to deal with seen/unseen imbalance via adaptive confidence smoothing and gating, yet as authors note it is much simpler to train than the existing GAN-based zero-shot approaches. In this work, we introduce arguably the simplest zero-shot representation training approach of all, and we demonstrate that when the image level text information is available, it achieves the state-of-the-art results on GZSL task on two well-known datasets. ",
+        "bbox": [
+            173,
+            446,
+            825,
+            766
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "text",
+        "text": "4 EXPERIMENTAL RESULTS ",
+        "text_level": 1,
+        "bbox": [
+            176,
+            792,
+            418,
+            808
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "text",
+        "text": "4.1 DATASETS ",
+        "text_level": 1,
+        "bbox": [
+            174,
+            825,
+            287,
+            840
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "text",
+        "text": "We focus on learning embeddings for fine-grained visual descriptions and test them in ZSL/GZSL scenario. To test the quality of trained embeddings we focus on datasets that provide paired images and text descriptions, such as Caltech-UCSD-Birds (CUB) (Welinder et al., 2010) and Oxford Flowers (FLOWERS) (Nilsback & Zisserman, 2008), that were augmented with textual descriptions by Reed et al. (2016). We use the GZSL splits proposed by Xian et al. (2018c). The attribute-based datasets, such as SUN (Patterson et al., 2014) and AWA (Lampert et al., 2014) do not contain this information and do not have a notion of entity of a class in them. They are out of the scope of the current paper. ",
+        "bbox": [
+            174,
+            854,
+            825,
+            924
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "image",
+        "img_path": "images/6c68eb4c7e42a907d13836d18bd23fbf8d9cba1c7819eebd0cbae1133409eec9.jpg",
+        "image_caption": [
+            "Figure 2: Harmonic mean Top-1 accuracy on seen and unseen, $\\mathbf { H }$ , against the value of $\\alpha$ on the validation set. The curves represent the mean and $9 5 \\%$ confidence intervals over 10 optimization runs. Results are stable over different runs. $\\mathbf { H }$ exhibits a distinct inverted U-shape w.r.t. $\\alpha$ . "
+        ],
+        "image_footnote": [],
+        "bbox": [
+            220,
+            102,
+            776,
+            229
+        ],
+        "page_idx": 5
+    },
+    {
+        "type": "text",
+        "text": "",
+        "bbox": [
+            176,
+            320,
+            823,
+            349
+        ],
+        "page_idx": 5
+    },
+    {
+        "type": "text",
+        "text": "4.2 ARCHITECTURE AND TRAINING DETAILS ",
+        "text_level": 1,
+        "bbox": [
+            178,
+            377,
+            500,
+            391
+        ],
+        "page_idx": 5
+    },
+    {
+        "type": "text",
+        "text": "Below, we provide more detailed description of parameters used to build and train the architecture depicted in Figure 1. We use exactly the same hyperparameter settings for CUB and FLOWERS. The text feature extractor is built by cascading two ResNet blocks, followed by a BiLSTM. Each ResNet block has 3 convolutional/batch norm layers. The number of filters in the ResNet blocks is 128 and 256, BiLSTM has 512 filters for forward and backward branches (1024 total). All variables in the convolutional stack (including the batch normalization parameters $\\gamma$ and $\\beta$ ) are L2-penalized with weight 0.001. The image feature extractor is a ResNet-101 with fixed weights pretrained on the split of ImageNet proposed by Xian et al. (2018c). In this work we use precomputed image features, available in (Xian et al., 2018a) for CUB and in (Xian et al., 2018b) for FLOWERS. Image and text features are projected in the common embedding space of size 1024 with FC layers with no non-linearity. They are preceded with a dropout of 0.25. The trainable components of the model are trained for $1 5 0 \\mathrm { k }$ batches of size 32 using SGD with initial learning rate of 0.1 that is annealed by a factor of 10 every $5 0 \\mathrm { k }$ batches. For each batch, we sample 32 instances, each instance includes a vector of precomputed ResNet-101 features and 10 text descriptions corresponding to it, according to the original dataset definition Reed et al. (2016). All 10 text descriptions are processed via the CNN/LSTM stack and the resulting embeddings are average pooled to create a vector representation of length 1024. ",
+        "bbox": [
+            174,
+            407,
+            825,
+            642
+        ],
+        "page_idx": 5
+    },
+    {
+        "type": "text",
+        "text": "4.3 KEY RESULTS ",
+        "text_level": 1,
+        "bbox": [
+            174,
+            671,
+            313,
+            685
+        ],
+        "page_idx": 5
+    },
+    {
+        "type": "text",
+        "text": "Our key empirical results are compared in Table 1 and in Table 2 against the latest state of the art. Our results are based on the settings of $\\lambda = 0 . 5$ , $\\kappa = 0 . 5$ and $\\alpha$ selected on the validation sets of CUB and FLOWERS datasets. Please refer to Section 4.5 for the analysis of stability with respect to the choices of $\\lambda$ and $\\kappa$ and Sections 2.2 and 4.4 for more details on the selection of $\\alpha$ . The combination of the proposed training method and the rebalancing of the metric space results in the state-of-the-art performance. Most of the current methods rely on the dataset augmentation techniques based on GANs, VAEs or combinations thereof. Those are clearly complementary w.r.t. our method and their addition to the training procedure is likely to further boost the performance of our proposed approach. However, this is outside of the scope of the current work. Moreover, the proposed method is state-of-the-art on FLOWERS even when compared against (Atzmon & Chechik, 2019) that uses both more sophisticated GAN based embedding learning approach and a more sophisticated seen/unseen error rate balancing based on COSMO. It is important to note that Atzmon & Chechik (2019) did not report the sentence level results on CUB. Yet, when applied on attributes together with f-CLSWGAN (Xian et al., 2018d) COSMO resulted in $0 . 8 \\%$ performance drop and when applied with LAGO (Atzmon & Chechik, 2018) it achieved $0 . 5 \\%$ improvement over the attribute based state of the art. ",
+        "bbox": [
+            174,
+            702,
+            825,
+            922
+        ],
+        "page_idx": 5
+    },
+    {
+        "type": "image",
+        "img_path": "images/3f384a4e76f0bc5c58077471f4b2954c58239b2f85fada1da5ff845d8c8bbbae.jpg",
+        "image_caption": [
+            "Figure 3: Harmonic mean Top-1 accuracy on seen and unseen, H, against $\\lambda$ , the relative weight of image and text retrieval loss terms. $\\lambda = 0$ corresponds to the case of image retrieval loss having weight 1 and text retrieval loss having weight 0. Mean over 10 optimization runs. "
+        ],
+        "image_footnote": [],
+        "bbox": [
+            220,
+            102,
+            777,
+            229
+        ],
+        "page_idx": 6
+    },
+    {
+        "type": "image",
+        "img_path": "images/a157ed151bbb217f07fbd4f1af909679c63192783007d447465bf462d40594e9.jpg",
+        "image_caption": [
+            "Figure 4: The plot of the harmonic mean Top-1 accuracy on seen and unseen, $\\mathbf { H }$ , against $\\kappa$ , the relative weight of the retrieval and the classification loss terms. $\\kappa = 0$ corresponds to the case of classification loss having weight 0. The curves represent the mean over 10 optimization runs. "
+        ],
+        "image_footnote": [],
+        "bbox": [
+            220,
+            303,
+            777,
+            431
+        ],
+        "page_idx": 6
+    },
+    {
+        "type": "text",
+        "text": "4.4 ON THE SEEN/UNSEEN ACCURACY BALANCING ",
+        "text_level": 1,
+        "bbox": [
+            179,
+            518,
+            537,
+            531
+        ],
+        "page_idx": 6
+    },
+    {
+        "type": "text",
+        "text": "Figure 2 demonstrates the plot of harmonic mean Top-1 accuracy, $\\mathbf { H }$ , against the value of $\\alpha$ on the validation sets of CUB and FLOWERS datasets. The validation set is constructed by further splitting the train set on both datasets. For example, CUB has a train set of 5875 images from 100 seen classes and a validation set of 2946 images from 50 unseen classes. We further divide the train set into 4700 train images from 100 seen classes, 1175 seen validation images $( 4 7 0 0 + 1 1 7 5 = 5 8 7 5 )$ and we use all the 2946 images from 50 classes as the unseen validation set. Once the value of $\\alpha$ is determined we train the representation on the full train $^ +$ val subset and report results on the test split (the usual practice in GZSL). We confirm on the validation set that H exhibits an inverted U-shape behavior as a function of $\\alpha$ , which was theoretically predicted in Section 2.2. Therefore, $\\alpha$ can be selected on the validation set and then applied to re-scale the metric space to balance the accuracy on seen and unseen classes during test time as described in Section 2.2. ",
+        "bbox": [
+            173,
+            545,
+            825,
+            698
+        ],
+        "page_idx": 6
+    },
+    {
+        "type": "text",
+        "text": "4.5 ABLATION STUDY ",
+        "text_level": 1,
+        "bbox": [
+            176,
+            722,
+            339,
+            734
+        ],
+        "page_idx": 6
+    },
+    {
+        "type": "text",
+        "text": "Figure 3 presents the results of the ablation study on the importance of image and text retrieval losses. We see that all of the Top-1 accuracies $( \\mathbf { H } , \\mathbf { s } , \\mathbf { u } )$ are stable in the range of $\\lambda \\in [ 0 . 2 , 0 . 9 ]$ , when both losses have tangible weight. Removing either text or image retrieval losses (setting $\\lambda$ to 0 or 1 respectively) leads to performance drop in both cases. Removing the text retrieval loss (case $\\lambda = 0$ ) results in the most significant drop. This is due to the fact that the text retrieval task is more tightly related to the GZSL task. At the batch level, retrieving the right text given an image is equivalent to identifying the correct class encoded by a text prototype during ZSL inference step. The image retrieval task is not directly related to solving the ZSL problem and yet it does yield a positive regularizing effect on both CUB and FLOWERS. ",
+        "bbox": [
+            173,
+            750,
+            825,
+            875
+        ],
+        "page_idx": 6
+    },
+    {
+        "type": "text",
+        "text": "Figure 4 shows the results of the ablation study of the interplay between the retrieval loss and the classification loss. We observe, just as in the case with $\\lambda$ , that there exists a reasonably flat and stable range of $\\kappa \\in [ 0 . 2 , 0 . 6 ]$ . The range for $\\kappa$ is a bit smaller. $\\kappa = 1$ results in the catastrophic performance drop: the classification losses by themselves do not enforce any modality alignment (please refer to Fig. 1 and Algorithm 1 clearly demonstrating this). ",
+        "bbox": [
+            174,
+            882,
+            823,
+            924
+        ],
+        "page_idx": 6
+    },
+    {
+        "type": "table",
+        "img_path": "images/a9ed39c401c5a3ac797b5f3447e0276b6810d59aee48a6e4a3f12e525762ff7f.jpg",
+        "table_caption": [
+            "Table 3: Generalized zero-shot Top-1 classification accuracy, ablation study. "
+        ],
+        "table_footnote": [],
+        "table_body": "<table><tr><td></td><td></td><td></td><td></td><td>CUB</td><td></td><td colspan=\"3\">FLOWERS</td></tr><tr><td>α</td><td>入</td><td>K</td><td>u</td><td>S</td><td>H</td><td>u</td><td>s</td><td>H</td></tr><tr><td>0.0</td><td>0.5</td><td>0.5</td><td>38.3</td><td>65.3</td><td>48.3</td><td>55.1</td><td>84.6</td><td>66.7</td></tr><tr><td>0.0</td><td>0.5</td><td>0.0</td><td>39.3</td><td>57.5</td><td>46.7</td><td>54.0</td><td>78.1</td><td>63.8</td></tr><tr><td>√</td><td>0.5</td><td>0.0</td><td>53.8</td><td>49.6</td><td>51.6</td><td>71.7</td><td>67.2</td><td>69.4</td></tr><tr><td>√</td><td>0.0</td><td>0.5</td><td>47.4</td><td>36.6</td><td>41.3</td><td>51.5</td><td>60.5</td><td>55.6</td></tr><tr><td>√</td><td>1.0</td><td>0.5</td><td>53.9</td><td>53.8</td><td>53.8</td><td>69.5</td><td>73.9</td><td>71.6</td></tr><tr><td>厂</td><td>0.5</td><td>0.5</td><td>59.3</td><td>52.6</td><td>55.8</td><td>73.0</td><td>73.6</td><td>73.3</td></tr></table>",
+        "bbox": [
+            292,
+            126,
+            705,
+            248
+        ],
+        "page_idx": 7
+    },
+    {
+        "type": "text",
+        "text": "",
+        "bbox": [
+            174,
+            273,
+            821,
+            303
+        ],
+        "page_idx": 7
+    },
+    {
+        "type": "text",
+        "text": "Table 3 studies the effects of different loss terms on the harmonic mean Top-1 accuracy H. The best result is achieved when all loss terms are active and when the metric space rescaling is on (the case of $\\lambda = 0 . 5$ , $\\kappa = 0 . 5$ and $\\alpha$ is checked, the last line in the table). Comparing this with the case when there is no metric space rescaling (first line with $\\alpha = 0$ ), we see that the rescaling helps to decrease the gap between seen and unseen classification accuracy. For CUB, the discrepancy reduction is from around $30 \\%$ to around $6 \\%$ , for FLOWERS it is from around $30 \\%$ to around $1 \\%$ . We would like to stress that we only use images and texts from the training set to achieve that. Going to the second line in the table (the image/text classification loss is inactive, $\\kappa = 0$ ) and comparing it to the first one, we assess the effect of the image/text classification loss. It barely affects the performance on unseen set, but it significantly boosts the classification accuracy on the seen set (around $8 \\%$ on both datasets). This is logical: adding a classifier loss results in a better classifier of the test images from the seen classes. This alone does not make it a better GZSL classifier, however. Only when applied together with metric space rescaling, this results in the performance boost (please refer to lines 1 and 6 in Table 3). Our interpretation is that the addition of the image/text classifier loss helps to reduce the intraclass variability in embeddings and provides for tighter clustering. However, this also leads to overfit on the classification task. This is accounted for by metric rescaling that enables the learnings from the image/text classification task be transferred effectively into the GZSL task. ",
+        "bbox": [
+            173,
+            309,
+            825,
+            545
+        ],
+        "page_idx": 7
+    },
+    {
+        "type": "text",
+        "text": "The comparison of the last four rows of Table 3 leads us to believe that all the proposed loss terms outlined in Fig. 1 and Algorithm 1 are important for achieving the state-of-the-art performance. Excluding any one of them (corresponding to the extreme values $\\lambda = 0$ , $\\lambda = 1$ , $\\kappa = 0$ ) leads to performance deterioration. Finally, an interesting observation can be made by comparing line 3 of Table 3 with performance of algorithms in Table 1. In this case our algorithm does not use any class labels and relies on training using exclusively the retrieval losses that can be calculated only based on the pairwise relationships between texts and images. We can see that using this type of supervision alone already results in a very high-quality representation. The representation is competitive against the latest GAN/VAE based approaches on CUB and is state-of-the-art on FLOWERS. This opens up new exploration avenues showing that in the case when very fine-grained modality outputs are available (image and text description pairs being a very prominent example), the high-quality representations may be learned without relying on manually supplied class labels. ",
+        "bbox": [
+            173,
+            553,
+            825,
+            719
+        ],
+        "page_idx": 7
+    },
+    {
+        "type": "text",
+        "text": "5 CONCLUSIONS ",
+        "text_level": 1,
+        "bbox": [
+            176,
+            739,
+            328,
+            755
+        ],
+        "page_idx": 7
+    },
+    {
+        "type": "text",
+        "text": "We propose and empirically validate two improvements to the process of learning fine-grained crossmodal representations. First, we confirm the hypothesis that in the context of paired images and texts, a deep metric learning approach can be driven by an instance-based retrieval loss resulting in competitive generalized zero shot classification results. Combined with an additional class label based image/text crossentropy term this results in state-of-the-art performance on two well known datasets, CUB and FLOWERS. This is an interesting result demonstrating that high-quality deep representations can be trained relying largely on pairwise relationships between modalities. On top of that, we propose a solution to one of the prominent problems in GZSL: classifying instances of unseen classes as seen ones. We mathematically analyze and empirically validate the method of adjusting a single scalar that transcends in its effectiveness advanced dataset augmentation and training approaches based on GANs and VAEs. ",
+        "bbox": [
+            174,
+            770,
+            825,
+            922
+        ],
+        "page_idx": 7
+    },
+    {
+        "type": "text",
+        "text": "REFERENCES ",
+        "text_level": 1,
+        "bbox": [
+            174,
+            102,
+            287,
+            117
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "Zeynep Akata, Scott Reed, Daniel Walter, Honglak Lee, and Bernt Schiele. Evaluation of output embeddings for fine-grained image classification. In CVPR, 2015. ",
+        "bbox": [
+            174,
+            126,
+            823,
+            155
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "Zeynep Akata, Florent Perronnin, Zaid Harchaoui, and Cordelia Schmid. Label-embedding for image classification. TPAMI, 2016. ",
+        "bbox": [
+            173,
+            162,
+            821,
+            191
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "Yuval Atzmon and Gal Chechik. Probabilistic AND-OR attribute grouping for zero-shot learning. In UAI, 2018. ",
+        "bbox": [
+            173,
+            199,
+            821,
+            229
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "Yuval Atzmon and Gal Chechik. Adaptive confidence smoothing for generalized zero-shot learning. In CVPR, 2019. ",
+        "bbox": [
+            173,
+            237,
+            823,
+            266
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "Wei-Lun Chao, Soravit Changpinyo, Boqing Gong, and Fei Sha. An empirical study and analysis of generalized zero-shot learning for object recognition in the wild. In ECCV (2), pp. 52–68, 2016. ",
+        "bbox": [
+            173,
+            273,
+            825,
+            304
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "Debasmit Das and C Lee. Zero-shot image recognition using relational matching, adaptation and calibration. In International Joint Conference on Neural Networks, 2019. ",
+        "bbox": [
+            174,
+            311,
+            821,
+            342
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "Rafael Felix, Vijay Kumar B G, Ian Reid, and Gustavo Carneiro. Multi-modal cycle-consistent generalized zero-shot learning. In ECCV, 2018. ",
+        "bbox": [
+            173,
+            348,
+            823,
+            378
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "Andrea Frome, Greg S Corrado, Jon Shlens, Samy Bengio, Jeff Dean, Marc Aurelio Ranzato, and Tomas Mikolov. DeViSE: A deep visual-semantic embedding model. In NIPS, pp. 2121–2129, 2013. ",
+        "bbox": [
+            174,
+            386,
+            825,
+            429
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "Ian Goodfellow, Jean Pouget-Abadie, Mehdi Mirza, Bing Xu, David Warde-Farley, Sherjil Ozair, Aaron Courville, and Yoshua Bengio. Generative adversarial nets. In NIPS, pp. 2672–2680, 2014. ",
+        "bbox": [
+            173,
+            438,
+            825,
+            467
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "Arthur Gretton, Karsten M Borgwardt, Malte Rasch, Bernhard Schölkopf, and Alex J Smola. A kernel method for the two-sample-problem. In NIPS, 2007. ",
+        "bbox": [
+            173,
+            474,
+            825,
+            503
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "Geoffrey Hinton, Oriol Vinyals, and Jeffrey Dean. Distilling the knowledge in a neural network. In NIPS Workshop, 2015. ",
+        "bbox": [
+            174,
+            512,
+            823,
+            541
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "Yao-Hung Hubert Tsai, Liang-Kang Huang, and Ruslan Salakhutdinov. Learning robust visualsemantic embeddings. In CVPR, 2017. ",
+        "bbox": [
+            173,
+            549,
+            825,
+            579
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "Diederik P. Kingma and Max Welling. Auto-encoding variational bayes. In ICLR, 2014. ",
+        "bbox": [
+            171,
+            587,
+            753,
+            602
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "C. H. Lampert, H. Nickisch, and S. Harmeling. Attribute-based classification for zero-shot visual object categorization. TPAMI, 2014. ",
+        "bbox": [
+            176,
+            611,
+            825,
+            640
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "Hugo Larochelle, Dumitru Erhan, and Yoshua Bengio. Zero-data learning of new tasks. In AAAI, 2008. ",
+        "bbox": [
+            176,
+            647,
+            825,
+            676
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "Yann LeCun, Yoshua Bengio, and Geoffrey Hinton. Deep learning. Nature, 2015. ",
+        "bbox": [
+            176,
+            684,
+            712,
+            700
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "Shichen Liu, Mingsheng Long, Jianmin Wang, and Michael I Jordan. Generalized zero-shot learning with deep calibration network. In NIPS, 2018. ",
+        "bbox": [
+            173,
+            708,
+            818,
+            738
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "M. Nilsback and A. Zisserman. Automated flower classification over a large number of classes. In 2008 Sixth Indian Conference on Computer Vision, Graphics Image Processing, 2008. ",
+        "bbox": [
+            173,
+            746,
+            823,
+            775
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "Mark Palatucci, Dean Pomerleau, Geoffrey E Hinton, and Tom M Mitchell. Zero-shot learning with semantic output codes. In NIPS, 2009. ",
+        "bbox": [
+            174,
+            782,
+            823,
+            813
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "Genevieve Patterson, Chen Xu, Hang Su, and James Hays. The sun attribute database: Beyond categories for deeper scene understanding. IJCV, 2014. ",
+        "bbox": [
+            173,
+            820,
+            823,
+            849
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "Scott E. Reed, Zeynep Akata, Honglak Lee, and Bernt Schiele. Learning deep representations of fine-grained visual descriptions. In CVPR, 2016. ",
+        "bbox": [
+            173,
+            858,
+            821,
+            887
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "Edgar Schönfeld, Sayna Ebrahimi, Samarth Sinha, Trevor Darrell, and Zeynep Akata. Generalized zero-and few-shot learning via aligned variational autoencoders. CVPR, 2019. ",
+        "bbox": [
+            174,
+            895,
+            820,
+            924
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "Richard Socher, Milind Ganjoo, Christopher D Manning, and Andrew Ng. Zero-shot learning through cross-modal transfer. In NIPS, 2013.   \nVinay Kumar Verma, Gundeep Arora, Ashish Mishra, and Piyush Rai. Generalized zero-shot learning via synthesized examples. In CVPR, 2018.   \nWei Wang, Vincent W. Zheng, Han Yu, and Chunyan Miao. A survey of zero-shot learning: Settings, methods, and applications. ACM Trans. Intell. Syst. Technol., 2019.   \nYaqing Wang and Quanming Yao. Few-shot learning: A survey. In arXiv, 2019.   \nP. Welinder, S. Branson, T. Mita, C. Wah, F. Schroff, S. Belongie, and P. Perona. Caltech-UCSD Birds 200. Technical report, California Institute of Technology, 2010.   \nYongqin Xian, Zeynep Akata, Gaurav Sharma, Quynh N. Nguyen, Matthias Hein, and Bernt Schiele. Latent embeddings for zero-shot classification. In CVPR, 2016.   \nYongqin Xian, Christoph H. Lampert, Bernt Schiele, and Zeynep Akata. Pretrained CUB features, 2018a. URL http://datasets.d2.mpi-inf.mpg.de/xian/xlsa17.zip.   \nYongqin Xian, Christoph H. Lampert, Bernt Schiele, and Zeynep Akata. Pretrained FLOWERS features, 2018b. URL http://datasets.d2.mpi-inf.mpg.de/xian/cvpr18xian. zip.   \nYongqin Xian, H. Christoph Lampert, Bernt Schiele, and Zeynep Akata. Zero-shot learning: A comprehensive evaluation of the good, the bad and the ugly. TPAMI, 2018c.   \nYongqin Xian, Tobias Lorenz, Bernt Schiele, and Zeynep Akata. Feature generating networks for zero-shot learning. In CVPR, 2018d.   \nYongqin Xian, Saurabh Sharma, Bernt Schiele, and Zeynep Akata. f-vaegan-d2: A feature generating framework for any-shot learning. CVPR, 2019.   \nH. Zhang, T. Xu, H. Li, S. Zhang, X. Wang, X. Huang, and D. N. Metaxas. StackGAN $^ { + + }$ : Realistic image synthesis with stacked generative adversarial networks. PAMI, 2018. ",
+        "bbox": [
+            171,
+            102,
+            828,
+            522
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "A THE ANALYSIS OF ERROR RATES ",
+        "text_level": 1,
+        "bbox": [
+            174,
+            541,
+            488,
+            558
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "We show that $\\mathbb { P } \\{ \\widehat { y } ~ \\in ~ \\mathcal { Y } ^ { t r } | y _ { v } ~ \\in ~ \\mathcal { Y } ^ { t s } \\} ~ \\geq ~ \\mathbb { P } \\{ \\widehat { y } _ { \\alpha } ~ \\in ~ \\mathcal { Y } ^ { t r } | y _ { v } ~ \\in ~ \\mathcal { Y } ^ { t s } \\}$ . Let us define $\\delta _ { t r } \\equiv$ $\\begin{array} { r } { \\operatorname* { m i n } _ { y \\in \\mathcal { y } ^ { t r } } d ( \\mathbf { z } _ { v } , \\mathbf { p } ( y ) ) } \\end{array}$ and $\\begin{array} { r } { \\delta _ { t s } \\equiv \\operatorname* { m i n } _ { y \\in \\mathcal { V } ^ { t s } } d ( \\mathbf { z } _ { v } , \\mathbf { p } ( y ) ) } \\end{array}$ , then Equation (7) can be rewritten as: ",
+        "bbox": [
+            171,
+            570,
+            825,
+            602
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "equation",
+        "img_path": "images/d406a9252d80daf106d9a4cfc77b0b1de57e941418018b7e6ea684fcf96c68c2.jpg",
+        "text": "$$\n\\mathbb { P } \\{ \\widehat { y } _ { \\alpha } \\in \\mathcal { V } ^ { t r } | y _ { v } \\in \\mathcal { V } ^ { t s } \\} = \\mathbb { P } \\left\\{ ( 1 + \\alpha ) \\delta _ { t r } < \\delta _ { t s } | y _ { v } \\in \\mathcal { V } ^ { t s } \\right\\} \\mathrm { ~ . ~ }\n$$",
+        "text_format": "latex",
+        "bbox": [
+            299,
+            604,
+            694,
+            621
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "Let us consider the probability of event $\\delta _ { t r } < \\delta _ { t s }$ and decompose it as follows: ",
+        "bbox": [
+            174,
+            622,
+            691,
+            636
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "equation",
+        "img_path": "images/1125cd16d5d95a1f8c8e948f94949c974d4895b575a12f7fa163ec98d67420e9.jpg",
+        "text": "$$\n\\begin{array} { r l } & { \\mathbb { P } \\left\\{ \\delta _ { t r } < \\delta _ { t s } \\middle | y _ { v } \\in \\mathcal { Y } ^ { t s } \\right\\} = \\mathbb { P } \\left\\{ ( 1 + \\alpha ) \\delta _ { t r } < ( 1 + \\alpha ) \\delta _ { t s } | y _ { v } \\in \\mathcal { Y } ^ { t s } \\right\\} } \\\\ & { = \\mathbb { P } \\left\\{ ( 1 + \\alpha ) \\delta _ { t r } < \\delta _ { t s } \\cup \\delta _ { t s } \\leq ( 1 + \\alpha ) \\delta _ { t r } < ( 1 + \\alpha ) \\delta _ { t s } | y _ { v } \\in \\mathcal { Y } ^ { t s } \\right\\} } \\\\ & { = \\mathbb { P } \\left\\{ ( 1 + \\alpha ) \\delta _ { t r } < \\delta _ { t s } | y _ { v } \\in \\mathcal { Y } ^ { t s } \\right\\} + \\mathbb { P } \\left\\{ \\delta _ { t s } \\leq ( 1 + \\alpha ) \\delta _ { t r } < ( 1 + \\alpha ) \\delta _ { t s } | y _ { v } \\in \\mathcal { Y } ^ { t s } \\right\\} } \\\\ & { - \\mathbb { P } \\left\\{ ( 1 + \\alpha ) \\delta _ { t r } < \\delta _ { t s } \\cap \\delta _ { t s } \\leq ( 1 + \\alpha ) \\delta _ { t r } < ( 1 + \\alpha ) \\delta _ { t s } | y _ { v } \\in \\mathcal { Y } ^ { t s } \\right\\} } \\\\ & { = \\mathbb { P } \\left\\{ ( 1 + \\alpha ) \\delta _ { t r } < \\delta _ { t s } | y _ { v } \\in \\mathcal { Y } ^ { t s } \\right\\} + \\mathbb { P } \\left\\{ \\delta _ { t s } \\leq ( 1 + \\alpha ) \\delta _ { t r } < ( 1 + \\alpha ) \\delta _ { t s } | y _ { v } \\in \\mathcal { Y } ^ { t s } \\right\\} } \\end{array}\n$$",
+        "text_format": "latex",
+        "bbox": [
+            214,
+            635,
+            772,
+            734
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "The transitions are based on the relationship between probabilities of arbitrary events $A$ and $B$ , $\\mathbb { P } \\{ A \\cup B \\} = \\mathbb { P } \\{ A \\} + \\mathbb { P } \\{ B \\} - \\mathbb { P } \\{ A \\cap B \\}$ , and in our case ${ \\mathbb { P } } \\{ A \\cap B \\} = 0$ . This implies that: ",
+        "bbox": [
+            171,
+            734,
+            828,
+            762
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "equation",
+        "img_path": "images/c82fe815b758df297244b22978a323eab8837f91f51cbd01bee778cf0f808da0.jpg",
+        "text": "$$\n\\begin{array} { r l r } {  { \\mathbb { P } \\{ \\widehat { y } _ { \\alpha } \\in \\mathcal { Y } ^ { t r } | y _ { v } \\in \\mathcal { Y } ^ { t s } \\} = \\mathbb { P } \\{ \\widehat { y } \\in \\mathcal { Y } ^ { t r } | y _ { v } \\in \\mathcal { Y } ^ { t s } \\} - \\mathbb { P } \\{ \\frac { \\delta _ { t s } } { ( 1 + \\alpha ) } \\leq \\delta _ { t r } < \\delta _ { t s } | y _ { v } \\in \\mathcal { Y } ^ { t s } \\} } } \\\\ & { } & { \\leq \\mathbb { P } \\{ \\widehat { y } \\in \\mathcal { Y } ^ { t r } | y _ { v } \\in \\mathcal { Y } ^ { t s } \\} . } \\end{array}\n$$",
+        "text_format": "latex",
+        "bbox": [
+            209,
+            763,
+            787,
+            818
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "We have just shown that for a non-negative $\\alpha$ the probability of misclassifying an image from an unseen class as one of the seen classes is smaller for the decision rule $\\widehat { y } _ { \\alpha }$ than for the original decision rule $\\widehat { y }$ . In fact, we can make a stronger claim. Since $\\delta _ { t s }$ and $\\delta _ { t r }$ are non-negative, it is bclear that the length of interval $[ \\delta _ { t s } / ( 1 + \\alpha ) , \\delta _ { t s } \\overline { { ) } }$ increases as $\\alpha$ increases, and hence probability that $\\delta _ { t r }$ falls in this interval is non-decreasing with increasing $\\alpha$ . Thus we have for any $0 \\leq \\alpha _ { 1 } \\leq \\alpha _ { 2 }$ , $\\mathbb { P } \\{ \\widehat { y } _ { \\alpha _ { 1 } } \\in \\mathcal { Y } ^ { t r } | y _ { v } \\in \\mathcal { Y } ^ { t s } \\} \\geq \\mathbb { P } \\{ \\widehat { y } _ { \\alpha _ { 2 } } \\in \\mathcal { Y } ^ { \\bar { t } r } | y _ { v } \\in \\mathcal { Y } ^ { t s } \\}$ , i.e. $\\mathbb { P } \\{ \\widehat { y } _ { \\alpha } \\in \\mathcal { V } ^ { t r } | y _ { v } \\in \\mathcal { V } ^ { t s } \\}$ is a monotone bnon-increasing function of $\\alpha$ b band we can reduce it by increasing $\\alpha$ . ",
+        "bbox": [
+            173,
+            823,
+            826,
+            921
+        ],
+        "page_idx": 9
+    }
+]
\ No newline at end of file
diff --git a/parse/train/SJl3h2EYvS/SJl3h2EYvS_middle.json b/parse/train/SJl3h2EYvS/SJl3h2EYvS_middle.json
new file mode 100644
index 0000000000000000000000000000000000000000..3dff72f2c0fda8c6cccc51a0b94f13f32f6f5157
--- /dev/null
+++ b/parse/train/SJl3h2EYvS/SJl3h2EYvS_middle.json
@@ -0,0 +1,28738 @@
+{
+    "pdf_info": [
+        {
+            "preproc_blocks": [
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        79,
+                        504,
+                        116
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                79,
+                                505,
+                                99
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        79,
+                                        505,
+                                        99
+                                    ],
+                                    "score": 1.0,
+                                    "content": "CLAREL: CLASSIFICATION VIA RETRIEVAL LOSS FOR",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                101,
+                                274,
+                                118
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        101,
+                                        274,
+                                        118
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ZERO-SHOT LEARNING",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        }
+                    ],
+                    "index": 0.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        112,
+                        135,
+                        244,
+                        157
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                113,
+                                135,
+                                201,
+                                147
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        113,
+                                        135,
+                                        201,
+                                        147
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Anonymous authors",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                111,
+                                146,
+                                245,
+                                158
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        111,
+                                        146,
+                                        245,
+                                        158
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Paper under double-blind review",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        }
+                    ],
+                    "index": 2.5
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        278,
+                        186,
+                        333,
+                        199
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                277,
+                                186,
+                                335,
+                                200
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        277,
+                                        186,
+                                        335,
+                                        200
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ABSTRACT",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        }
+                    ],
+                    "index": 4
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        143,
+                        213,
+                        469,
+                        334
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                142,
+                                213,
+                                469,
+                                225
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        142,
+                                        213,
+                                        469,
+                                        225
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We address the problem of learning fine-grained cross-modal representations. We",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                224,
+                                470,
+                                237
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        224,
+                                        470,
+                                        237
+                                    ],
+                                    "score": 1.0,
+                                    "content": "propose an instance-based deep metric learning approach in joint visual and textual",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                235,
+                                469,
+                                248
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        235,
+                                        469,
+                                        248
+                                    ],
+                                    "score": 1.0,
+                                    "content": "space. The key novelty of this paper is that it shows that using per-image semantic",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                245,
+                                469,
+                                259
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        245,
+                                        469,
+                                        259
+                                    ],
+                                    "score": 1.0,
+                                    "content": "supervision leads to substantial improvement in zero-shot performance over using",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                142,
+                                258,
+                                470,
+                                270
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        142,
+                                        258,
+                                        470,
+                                        270
+                                    ],
+                                    "score": 1.0,
+                                    "content": "class-only supervision. On top of that, we provide a probabilistic justification for a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                268,
+                                470,
+                                281
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        268,
+                                        470,
+                                        281
+                                    ],
+                                    "score": 1.0,
+                                    "content": "metric rescaling approach that solves a very common problem in the generalized",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                280,
+                                470,
+                                291
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        280,
+                                        470,
+                                        291
+                                    ],
+                                    "score": 1.0,
+                                    "content": "zero-shot learning setting, i.e., classifying test images from unseen classes as",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                290,
+                                470,
+                                302
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        290,
+                                        470,
+                                        302
+                                    ],
+                                    "score": 1.0,
+                                    "content": "one of the classes seen during training. We evaluate our approach on two fine-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                301,
+                                469,
+                                312
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        301,
+                                        469,
+                                        312
+                                    ],
+                                    "score": 1.0,
+                                    "content": "grained zero-shot learning datasets: CUB and FLOWERS. We find that on the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                311,
+                                469,
+                                325
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        311,
+                                        469,
+                                        325
+                                    ],
+                                    "score": 1.0,
+                                    "content": "generalized zero-shot classification task CLAREL consistently outperforms the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                323,
+                                293,
+                                335
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        323,
+                                        293,
+                                        335
+                                    ],
+                                    "score": 1.0,
+                                    "content": "existing approaches on both datasets.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        }
+                    ],
+                    "index": 10
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        359,
+                        206,
+                        372
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                358,
+                                208,
+                                375
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        358,
+                                        208,
+                                        375
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1 INTRODUCTION",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        }
+                    ],
+                    "index": 16
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        386,
+                        505,
+                        649
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                386,
+                                505,
+                                398
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        386,
+                                        505,
+                                        398
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Deep learning-based approaches have demonstrated superior flexibility and generalization capabilities",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                397,
+                                506,
+                                410
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        397,
+                                        506,
+                                        410
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in information processing on a wide variety of tasks, such as vision, speech and language (LeCun et al.,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                407,
+                                505,
+                                420
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        407,
+                                        505,
+                                        420
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2015). However, it has been widely realized that the transfer of deep representations to real-world",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                419,
+                                506,
+                                432
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        419,
+                                        506,
+                                        432
+                                    ],
+                                    "score": 1.0,
+                                    "content": "applications is challenging due to the typical reliance on massive hand-labeled datasets. Learning in",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                428,
+                                506,
+                                443
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        428,
+                                        506,
+                                        443
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the low-labeled data regime, especially in the zero-shot (Wang et al., 2019) and the few-shot (Wang",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                440,
+                                505,
+                                453
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        440,
+                                        505,
+                                        453
+                                    ],
+                                    "score": 1.0,
+                                    "content": "& Yao, 2019) setups, have recently received significant attention in the literature. In the problem of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                452,
+                                505,
+                                464
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        452,
+                                        505,
+                                        464
+                                    ],
+                                    "score": 1.0,
+                                    "content": "zero-shot learning (ZSL), the objective is to recognize categories that have not been seen during the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                462,
+                                505,
+                                476
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        462,
+                                        505,
+                                        476
+                                    ],
+                                    "score": 1.0,
+                                    "content": "training (Larochelle et al., 2008). This is typically done by relying on anchor embeddings learned in",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                473,
+                                505,
+                                487
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        473,
+                                        505,
+                                        487
+                                    ],
+                                    "score": 1.0,
+                                    "content": "one modality as prototypes and by associating a query embedding from the other modality with the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                484,
+                                506,
+                                498
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        484,
+                                        506,
+                                        498
+                                    ],
+                                    "score": 1.0,
+                                    "content": "closest prototype. In the generalized ZSL (GZSL) case (Xian et al., 2018c), the objective is more",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                495,
+                                505,
+                                508
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        495,
+                                        505,
+                                        508
+                                    ],
+                                    "score": 1.0,
+                                    "content": "challenging as recognition is performed in the joint space of seen and unseen categories. ZSL, as well",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                506,
+                                506,
+                                519
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        506,
+                                        506,
+                                        519
+                                    ],
+                                    "score": 1.0,
+                                    "content": "as its generalized counterpart, provide a viable framework to learn cross-modal representations that",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                518,
+                                506,
+                                530
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        518,
+                                        506,
+                                        530
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are flexible and adaptive. For example, in this paradigm, the adaptation to a new classification task",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                527,
+                                506,
+                                542
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        527,
+                                        506,
+                                        542
+                                    ],
+                                    "score": 1.0,
+                                    "content": "based on text/image representation space alignment could be as easy as defining/appending/modifying",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                540,
+                                505,
+                                552
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        540,
+                                        505,
+                                        552
+                                    ],
+                                    "score": 1.0,
+                                    "content": "a set of text sentences to define classes of new classifiers. This is an especially relevant problem as",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                549,
+                                506,
+                                563
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        549,
+                                        506,
+                                        563
+                                    ],
+                                    "score": 1.0,
+                                    "content": "machine learning is challenged with the long tail of classes, and the idea of learning from pairs of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                560,
+                                506,
+                                575
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        560,
+                                        506,
+                                        575
+                                    ],
+                                    "score": 1.0,
+                                    "content": "images and sentences, abundant on the web, looks like a natural solution. Therefore, in this paper we",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                572,
+                                507,
+                                585
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        572,
+                                        507,
+                                        585
+                                    ],
+                                    "score": 1.0,
+                                    "content": "specifically target the fine-grained scenario of paired images and their respective text descriptions.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                583,
+                                505,
+                                595
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        583,
+                                        505,
+                                        595
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The uniqueness of this scenario is in the fact that the co-occurance of image and text provides a rich",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                594,
+                                505,
+                                606
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        594,
+                                        505,
+                                        606
+                                    ],
+                                    "score": 1.0,
+                                    "content": "source of information. The ways of leveraging this source have not been sufficiently explored in the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                604,
+                                506,
+                                619
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        604,
+                                        506,
+                                        619
+                                    ],
+                                    "score": 1.0,
+                                    "content": "context of GZSL. Although we focus exclusively on the GZSL recognition setup in this paper, we",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                615,
+                                505,
+                                629
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        615,
+                                        505,
+                                        629
+                                    ],
+                                    "score": 1.0,
+                                    "content": "believe that the research in this direction has potential to enable zero-shot flexibility in a wider array",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                627,
+                                505,
+                                640
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        627,
+                                        505,
+                                        640
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of high-level tasks such as segmentation or conditional image generation (Zhang et al., 2018). The",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                638,
+                                426,
+                                651
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        638,
+                                        426,
+                                        651
+                                    ],
+                                    "score": 1.0,
+                                    "content": "contributions of this work can be characterized under the following two themes.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        }
+                    ],
+                    "index": 28.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        655,
+                        505,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                654,
+                                507,
+                                668
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        654,
+                                        507,
+                                        668
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Instance-based training loss. Most prominent zero-shot learning approaches rely heavily on class-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                666,
+                                505,
+                                678
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        666,
+                                        505,
+                                        678
+                                    ],
+                                    "score": 1.0,
+                                    "content": "level modality alignment (Xian et al., 2018c). We propose a new composite loss function that balances",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                676,
+                                505,
+                                689
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        676,
+                                        505,
+                                        689
+                                    ],
+                                    "score": 1.0,
+                                    "content": "instance-based pairwise image/text retrieval loss and the usual classifier loss. The retrieval loss term",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                687,
+                                505,
+                                700
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        687,
+                                        505,
+                                        700
+                                    ],
+                                    "score": 1.0,
+                                    "content": "does not use class labels. We demonstrate that the class-level information is important, but in the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                699,
+                                505,
+                                711
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        699,
+                                        505,
+                                        711
+                                    ],
+                                    "score": 1.0,
+                                    "content": "fine-grained text/image pairing scenarios, most of the GZSL accuracy can be extracted from the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                709,
+                                505,
+                                722
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        709,
+                                        505,
+                                        722
+                                    ],
+                                    "score": 1.0,
+                                    "content": "instance-based retrieval loss. To the best of our knowledge, this type of training has not been used in",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                720,
+                                480,
+                                734
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        720,
+                                        480,
+                                        734
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the GZSL literature. Its impressive performance opens up new promising research directions.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 47
+                        }
+                    ],
+                    "index": 44
+                }
+            ],
+            "page_idx": 0,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        107,
+                        27,
+                        308,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                107,
+                                26,
+                                308,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        26,
+                                        308,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Under review as a conference paper at ICLR 2020",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        302,
+                        751,
+                        308,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                302,
+                                751,
+                                308,
+                                761
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        302,
+                                        751,
+                                        308,
+                                        761
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        79,
+                        504,
+                        116
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                79,
+                                505,
+                                99
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        79,
+                                        505,
+                                        99
+                                    ],
+                                    "score": 1.0,
+                                    "content": "CLAREL: CLASSIFICATION VIA RETRIEVAL LOSS FOR",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                101,
+                                274,
+                                118
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        101,
+                                        274,
+                                        118
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ZERO-SHOT LEARNING",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        }
+                    ],
+                    "index": 0.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        112,
+                        135,
+                        244,
+                        157
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                113,
+                                135,
+                                201,
+                                147
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        113,
+                                        135,
+                                        201,
+                                        147
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Anonymous authors",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                111,
+                                146,
+                                245,
+                                158
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        111,
+                                        146,
+                                        245,
+                                        158
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Paper under double-blind review",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        }
+                    ],
+                    "index": 2.5,
+                    "bbox_fs": [
+                        111,
+                        135,
+                        245,
+                        158
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        278,
+                        186,
+                        333,
+                        199
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                277,
+                                186,
+                                335,
+                                200
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        277,
+                                        186,
+                                        335,
+                                        200
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ABSTRACT",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        }
+                    ],
+                    "index": 4
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        143,
+                        213,
+                        469,
+                        334
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                142,
+                                213,
+                                469,
+                                225
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        142,
+                                        213,
+                                        469,
+                                        225
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We address the problem of learning fine-grained cross-modal representations. We",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                224,
+                                470,
+                                237
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        224,
+                                        470,
+                                        237
+                                    ],
+                                    "score": 1.0,
+                                    "content": "propose an instance-based deep metric learning approach in joint visual and textual",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                235,
+                                469,
+                                248
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        235,
+                                        469,
+                                        248
+                                    ],
+                                    "score": 1.0,
+                                    "content": "space. The key novelty of this paper is that it shows that using per-image semantic",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                245,
+                                469,
+                                259
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        245,
+                                        469,
+                                        259
+                                    ],
+                                    "score": 1.0,
+                                    "content": "supervision leads to substantial improvement in zero-shot performance over using",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                142,
+                                258,
+                                470,
+                                270
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        142,
+                                        258,
+                                        470,
+                                        270
+                                    ],
+                                    "score": 1.0,
+                                    "content": "class-only supervision. On top of that, we provide a probabilistic justification for a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                268,
+                                470,
+                                281
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        268,
+                                        470,
+                                        281
+                                    ],
+                                    "score": 1.0,
+                                    "content": "metric rescaling approach that solves a very common problem in the generalized",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                280,
+                                470,
+                                291
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        280,
+                                        470,
+                                        291
+                                    ],
+                                    "score": 1.0,
+                                    "content": "zero-shot learning setting, i.e., classifying test images from unseen classes as",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                290,
+                                470,
+                                302
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        290,
+                                        470,
+                                        302
+                                    ],
+                                    "score": 1.0,
+                                    "content": "one of the classes seen during training. We evaluate our approach on two fine-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                301,
+                                469,
+                                312
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        301,
+                                        469,
+                                        312
+                                    ],
+                                    "score": 1.0,
+                                    "content": "grained zero-shot learning datasets: CUB and FLOWERS. We find that on the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                311,
+                                469,
+                                325
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        311,
+                                        469,
+                                        325
+                                    ],
+                                    "score": 1.0,
+                                    "content": "generalized zero-shot classification task CLAREL consistently outperforms the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                323,
+                                293,
+                                335
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        323,
+                                        293,
+                                        335
+                                    ],
+                                    "score": 1.0,
+                                    "content": "existing approaches on both datasets.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        }
+                    ],
+                    "index": 10,
+                    "bbox_fs": [
+                        141,
+                        213,
+                        470,
+                        335
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        359,
+                        206,
+                        372
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                358,
+                                208,
+                                375
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        358,
+                                        208,
+                                        375
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1 INTRODUCTION",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        }
+                    ],
+                    "index": 16
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        386,
+                        505,
+                        649
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                386,
+                                505,
+                                398
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        386,
+                                        505,
+                                        398
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Deep learning-based approaches have demonstrated superior flexibility and generalization capabilities",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                397,
+                                506,
+                                410
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        397,
+                                        506,
+                                        410
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in information processing on a wide variety of tasks, such as vision, speech and language (LeCun et al.,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                407,
+                                505,
+                                420
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        407,
+                                        505,
+                                        420
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2015). However, it has been widely realized that the transfer of deep representations to real-world",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                419,
+                                506,
+                                432
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        419,
+                                        506,
+                                        432
+                                    ],
+                                    "score": 1.0,
+                                    "content": "applications is challenging due to the typical reliance on massive hand-labeled datasets. Learning in",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                428,
+                                506,
+                                443
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        428,
+                                        506,
+                                        443
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the low-labeled data regime, especially in the zero-shot (Wang et al., 2019) and the few-shot (Wang",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                440,
+                                505,
+                                453
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        440,
+                                        505,
+                                        453
+                                    ],
+                                    "score": 1.0,
+                                    "content": "& Yao, 2019) setups, have recently received significant attention in the literature. In the problem of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                452,
+                                505,
+                                464
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        452,
+                                        505,
+                                        464
+                                    ],
+                                    "score": 1.0,
+                                    "content": "zero-shot learning (ZSL), the objective is to recognize categories that have not been seen during the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                462,
+                                505,
+                                476
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        462,
+                                        505,
+                                        476
+                                    ],
+                                    "score": 1.0,
+                                    "content": "training (Larochelle et al., 2008). This is typically done by relying on anchor embeddings learned in",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                473,
+                                505,
+                                487
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        473,
+                                        505,
+                                        487
+                                    ],
+                                    "score": 1.0,
+                                    "content": "one modality as prototypes and by associating a query embedding from the other modality with the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                484,
+                                506,
+                                498
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        484,
+                                        506,
+                                        498
+                                    ],
+                                    "score": 1.0,
+                                    "content": "closest prototype. In the generalized ZSL (GZSL) case (Xian et al., 2018c), the objective is more",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                495,
+                                505,
+                                508
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        495,
+                                        505,
+                                        508
+                                    ],
+                                    "score": 1.0,
+                                    "content": "challenging as recognition is performed in the joint space of seen and unseen categories. ZSL, as well",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                506,
+                                506,
+                                519
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        506,
+                                        506,
+                                        519
+                                    ],
+                                    "score": 1.0,
+                                    "content": "as its generalized counterpart, provide a viable framework to learn cross-modal representations that",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                518,
+                                506,
+                                530
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        518,
+                                        506,
+                                        530
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are flexible and adaptive. For example, in this paradigm, the adaptation to a new classification task",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                527,
+                                506,
+                                542
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        527,
+                                        506,
+                                        542
+                                    ],
+                                    "score": 1.0,
+                                    "content": "based on text/image representation space alignment could be as easy as defining/appending/modifying",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                540,
+                                505,
+                                552
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        540,
+                                        505,
+                                        552
+                                    ],
+                                    "score": 1.0,
+                                    "content": "a set of text sentences to define classes of new classifiers. This is an especially relevant problem as",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                549,
+                                506,
+                                563
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        549,
+                                        506,
+                                        563
+                                    ],
+                                    "score": 1.0,
+                                    "content": "machine learning is challenged with the long tail of classes, and the idea of learning from pairs of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                560,
+                                506,
+                                575
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        560,
+                                        506,
+                                        575
+                                    ],
+                                    "score": 1.0,
+                                    "content": "images and sentences, abundant on the web, looks like a natural solution. Therefore, in this paper we",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                572,
+                                507,
+                                585
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        572,
+                                        507,
+                                        585
+                                    ],
+                                    "score": 1.0,
+                                    "content": "specifically target the fine-grained scenario of paired images and their respective text descriptions.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                583,
+                                505,
+                                595
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        583,
+                                        505,
+                                        595
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The uniqueness of this scenario is in the fact that the co-occurance of image and text provides a rich",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                594,
+                                505,
+                                606
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        594,
+                                        505,
+                                        606
+                                    ],
+                                    "score": 1.0,
+                                    "content": "source of information. The ways of leveraging this source have not been sufficiently explored in the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                604,
+                                506,
+                                619
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        604,
+                                        506,
+                                        619
+                                    ],
+                                    "score": 1.0,
+                                    "content": "context of GZSL. Although we focus exclusively on the GZSL recognition setup in this paper, we",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                615,
+                                505,
+                                629
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        615,
+                                        505,
+                                        629
+                                    ],
+                                    "score": 1.0,
+                                    "content": "believe that the research in this direction has potential to enable zero-shot flexibility in a wider array",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                627,
+                                505,
+                                640
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        627,
+                                        505,
+                                        640
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of high-level tasks such as segmentation or conditional image generation (Zhang et al., 2018). The",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                638,
+                                426,
+                                651
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        638,
+                                        426,
+                                        651
+                                    ],
+                                    "score": 1.0,
+                                    "content": "contributions of this work can be characterized under the following two themes.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        }
+                    ],
+                    "index": 28.5,
+                    "bbox_fs": [
+                        104,
+                        386,
+                        507,
+                        651
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        655,
+                        505,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                654,
+                                507,
+                                668
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        654,
+                                        507,
+                                        668
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Instance-based training loss. Most prominent zero-shot learning approaches rely heavily on class-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                666,
+                                505,
+                                678
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        666,
+                                        505,
+                                        678
+                                    ],
+                                    "score": 1.0,
+                                    "content": "level modality alignment (Xian et al., 2018c). We propose a new composite loss function that balances",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                676,
+                                505,
+                                689
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        676,
+                                        505,
+                                        689
+                                    ],
+                                    "score": 1.0,
+                                    "content": "instance-based pairwise image/text retrieval loss and the usual classifier loss. The retrieval loss term",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                687,
+                                505,
+                                700
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        687,
+                                        505,
+                                        700
+                                    ],
+                                    "score": 1.0,
+                                    "content": "does not use class labels. We demonstrate that the class-level information is important, but in the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                699,
+                                505,
+                                711
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        699,
+                                        505,
+                                        711
+                                    ],
+                                    "score": 1.0,
+                                    "content": "fine-grained text/image pairing scenarios, most of the GZSL accuracy can be extracted from the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                709,
+                                505,
+                                722
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        709,
+                                        505,
+                                        722
+                                    ],
+                                    "score": 1.0,
+                                    "content": "instance-based retrieval loss. To the best of our knowledge, this type of training has not been used in",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                720,
+                                480,
+                                734
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        720,
+                                        480,
+                                        734
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the GZSL literature. Its impressive performance opens up new promising research directions.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 47
+                        }
+                    ],
+                    "index": 44,
+                    "bbox_fs": [
+                        105,
+                        654,
+                        507,
+                        734
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "image",
+                    "bbox": [
+                        153,
+                        78,
+                        461,
+                        231
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                153,
+                                78,
+                                461,
+                                231
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        153,
+                                        78,
+                                        461,
+                                        231
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                153,
+                                                78,
+                                                461,
+                                                231
+                                            ],
+                                            "score": 0.974,
+                                            "type": "image",
+                                            "image_path": "c5306e54b7867e3eb7aee45c6499213721005da29ec547a7ed5a3a76968503c1.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 1,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        153,
+                                        78,
+                                        461,
+                                        129.0
+                                    ],
+                                    "spans": [],
+                                    "index": 0
+                                },
+                                {
+                                    "bbox": [
+                                        153,
+                                        129.0,
+                                        461,
+                                        180.0
+                                    ],
+                                    "spans": [],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        153,
+                                        180.0,
+                                        461,
+                                        231.0
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                106,
+                                239,
+                                505,
+                                338
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        238,
+                                        506,
+                                        252
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                238,
+                                                506,
+                                                252
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 1: The architecture and training diagram describing the proposed method. Each batch consists",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 3
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        250,
+                                        505,
+                                        264
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                250,
+                                                505,
+                                                264
+                                            ],
+                                            "score": 1.0,
+                                            "content": "of randomly sampled instances, i.e. pairs of images and their corresponding texts. Images are",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 4
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        261,
+                                        505,
+                                        275
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                261,
+                                                505,
+                                                275
+                                            ],
+                                            "score": 1.0,
+                                            "content": "embedded via ResNet and texts are embedded via a CNN/LSTM stack. Image and text features",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 5
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        271,
+                                        506,
+                                        286
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                271,
+                                                506,
+                                                286
+                                            ],
+                                            "score": 1.0,
+                                            "content": "are projected via a fully connected layer into the same dimensional space. In this space, distances",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 6
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        283,
+                                        505,
+                                        296
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                283,
+                                                505,
+                                                296
+                                            ],
+                                            "score": 1.0,
+                                            "content": "between text and image features from different instances are computed. The negative distances are",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 7
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        294,
+                                        505,
+                                        306
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                294,
+                                                505,
+                                                306
+                                            ],
+                                            "score": 1.0,
+                                            "content": "fed into softmax to train on both the image and the text retrieval tasks. The image retrieval task",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 8
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        305,
+                                        506,
+                                        318
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                305,
+                                                506,
+                                                318
+                                            ],
+                                            "score": 1.0,
+                                            "content": "consists of retrieving the image corresponding to the given text of the same instance and the text",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 9
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        315,
+                                        505,
+                                        330
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                315,
+                                                505,
+                                                330
+                                            ],
+                                            "score": 1.0,
+                                            "content": "retrieval task is vice versa. In addition to that, image and text embeddings are trained on auxiliary",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 10
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        327,
+                                        430,
+                                        339
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                327,
+                                                430,
+                                                339
+                                            ],
+                                            "score": 1.0,
+                                            "content": "image and text classification tasks on the class labels corresponding to instances.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 11
+                                }
+                            ],
+                            "index": 7
+                        }
+                    ],
+                    "index": 4.0
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        362,
+                        505,
+                        418
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                362,
+                                505,
+                                375
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        362,
+                                        505,
+                                        375
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Metric space rescaling. Metric-based ZSL approaches rely on distances between prototypes and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                372,
+                                506,
+                                387
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        372,
+                                        506,
+                                        387
+                                    ],
+                                    "score": 1.0,
+                                    "content": "query embeddings during inference. They are known to suffer from imbalanced performance on seen",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                384,
+                                505,
+                                397
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        384,
+                                        505,
+                                        397
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and unseen classes (Liu et al., 2018). Previous work proposed to use a heuristic trick, calibrated",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                394,
+                                505,
+                                408
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        394,
+                                        505,
+                                        408
+                                    ],
+                                    "score": 1.0,
+                                    "content": "stacking (Chao et al., 2016) or calibration (Das & Lee, 2019), to solve the problem. We refer to this",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                406,
+                                485,
+                                418
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        406,
+                                        485,
+                                        418
+                                    ],
+                                    "score": 1.0,
+                                    "content": "technique as metric rescaling in our work, and provide a sound probabilistic justification for it.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        }
+                    ],
+                    "index": 14
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        435,
+                        230,
+                        448
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                435,
+                                232,
+                                451
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        435,
+                                        232,
+                                        451
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2 PROPOSED METHOD",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        }
+                    ],
+                    "index": 17
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        462,
+                        505,
+                        595
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                462,
+                                506,
+                                475
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        462,
+                                        506,
+                                        475
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In this paper, we specifically target the fine-grained visual description scenario, as defined by Reed",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                474,
+                                506,
+                                486
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        474,
+                                        506,
+                                        486
+                                    ],
+                                    "score": 1.0,
+                                    "content": "et al. (2016). In this setting, the dataset consists of a number of images from a given set of classes",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                484,
+                                506,
+                                497
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        484,
+                                        506,
+                                        497
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and each image is accompanied by a number of textual descriptions. The task is to learn a joint",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                496,
+                                506,
+                                507
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        496,
+                                        506,
+                                        507
+                                    ],
+                                    "score": 1.0,
+                                    "content": "representation space for images and texts that can be used for zero-shot recognition. An instance of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                506,
+                                506,
+                                519
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        506,
+                                        506,
+                                        519
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the zero-shot multimodal representation learning problem can then be defined as follows. Given a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                516,
+                                506,
+                                530
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        516,
+                                        155,
+                                        530
+                                    ],
+                                    "score": 1.0,
+                                    "content": "training set",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        155,
+                                        517,
+                                        271,
+                                        529
+                                    ],
+                                    "score": 0.28,
+                                    "content": "S = \\{ ( v _ { n } , t _ { n } , y _ { n } ) \\mid v _ { n } \\in \\mathcal { V }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        271,
+                                        516,
+                                        276,
+                                        530
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        276,
+                                        517,
+                                        404,
+                                        529
+                                    ],
+                                    "score": 0.5,
+                                    "content": "t _ { n } \\in \\mathcal { T } , y _ { n } \\in \\mathcal { V } , n = 1 \\ldots N \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        405,
+                                        516,
+                                        506,
+                                        530
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of image, text and label",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                528,
+                                505,
+                                541
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        528,
+                                        312,
+                                        541
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tuples, we are interested in finding representations",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        312,
+                                        529,
+                                        362,
+                                        540
+                                    ],
+                                    "score": 0.92,
+                                    "content": "f _ { \\phi } : \\mathcal { V } \\to \\mathcal { Z }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        363,
+                                        528,
+                                        477,
+                                        541
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of image, parameterized by",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        477,
+                                        529,
+                                        484,
+                                        540
+                                    ],
+                                    "score": 0.83,
+                                    "content": "\\phi",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        484,
+                                        528,
+                                        505,
+                                        541
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                107,
+                                539,
+                                506,
+                                552
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        540,
+                                        160,
+                                        551
+                                    ],
+                                    "score": 0.91,
+                                    "content": "f _ { \\theta } : \\mathcal { T }  \\mathcal { Z }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        160,
+                                        539,
+                                        266,
+                                        552
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of text, parameterized by",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        267,
+                                        540,
+                                        273,
+                                        549
+                                    ],
+                                    "score": 0.77,
+                                    "content": "\\theta",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        273,
+                                        539,
+                                        406,
+                                        552
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", in a common embedding space",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        406,
+                                        540,
+                                        416,
+                                        549
+                                    ],
+                                    "score": 0.8,
+                                    "content": "\\mathcal { Z }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        416,
+                                        539,
+                                        506,
+                                        552
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Furthermore, GZSL",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                549,
+                                506,
+                                563
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        549,
+                                        271,
+                                        563
+                                    ],
+                                    "score": 1.0,
+                                    "content": "problem is defined using the sets of seen",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        271,
+                                        550,
+                                        286,
+                                        561
+                                    ],
+                                    "score": 0.89,
+                                    "content": "\\mathcal { V } ^ { t r }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        287,
+                                        549,
+                                        335,
+                                        563
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and unseen",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        335,
+                                        550,
+                                        351,
+                                        561
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\mathcal { V } ^ { t s }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        351,
+                                        549,
+                                        424,
+                                        563
+                                    ],
+                                    "score": 1.0,
+                                    "content": "classes, such that",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        424,
+                                        550,
+                                        487,
+                                        561
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\mathcal { V } = \\mathcal { V } ^ { t r } \\cup \\mathcal { V } ^ { t s }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        487,
+                                        549,
+                                        506,
+                                        563
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                107,
+                                559,
+                                504,
+                                574
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        561,
+                                        164,
+                                        572
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\bar { \\mathcal { V } } ^ { t r } \\cap \\mathcal { V } ^ { t s } = \\emptyset",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        164,
+                                        559,
+                                        396,
+                                        574
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". The training set will then only contain the seen classes, i.e.",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        396,
+                                        561,
+                                        504,
+                                        573
+                                    ],
+                                    "score": 0.88,
+                                    "content": "S ^ { t r } = \\{ ( v _ { n } , t _ { n } , y _ { n } ) \\mid v _ { n } \\in",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                571,
+                                506,
+                                585
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        573,
+                                        115,
+                                        583
+                                    ],
+                                    "score": 0.4,
+                                    "content": "\\nu",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        115,
+                                        571,
+                                        118,
+                                        585
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        118,
+                                        572,
+                                        198,
+                                        584
+                                    ],
+                                    "score": 0.86,
+                                    "content": "t _ { n } \\in \\mathcal { T } , y _ { n } \\in \\mathcal { y } ^ { t r } \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        199,
+                                        571,
+                                        369,
+                                        585
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and the task is to build a classifier function",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        370,
+                                        573,
+                                        436,
+                                        583
+                                    ],
+                                    "score": 0.9,
+                                    "content": "g : \\mathcal { Z } \\times \\mathcal { Z }  \\mathcal { V }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        436,
+                                        571,
+                                        506,
+                                        585
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". This is different",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                581,
+                                325,
+                                595
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        581,
+                                        248,
+                                        595
+                                    ],
+                                    "score": 1.0,
+                                    "content": "from the ZSL scenario focusing on",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        249,
+                                        583,
+                                        322,
+                                        594
+                                    ],
+                                    "score": 0.91,
+                                    "content": "g : \\mathcal { Z } \\times \\mathcal { Z }  \\mathcal { Y } ^ { t s }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        322,
+                                        581,
+                                        325,
+                                        595
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        }
+                    ],
+                    "index": 23.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        600,
+                        506,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                600,
+                                507,
+                                614
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        600,
+                                        141,
+                                        614
+                                    ],
+                                    "score": 1.0,
+                                    "content": "To build",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        141,
+                                        602,
+                                        148,
+                                        612
+                                    ],
+                                    "score": 0.78,
+                                    "content": "g",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        148,
+                                        600,
+                                        507,
+                                        614
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", most approaches to joint representation learning rely on class labeling to train a represen-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                611,
+                                506,
+                                623
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        611,
+                                        506,
+                                        623
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tation. For example, all the methods reviewed by Xian et al. (2018c) require the access to class labels",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                622,
+                                506,
+                                634
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        622,
+                                        506,
+                                        634
+                                    ],
+                                    "score": 1.0,
+                                    "content": "at train time. We hypothesise that in the fine-grained learning scenario, such as the one described",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                633,
+                                506,
+                                645
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        633,
+                                        506,
+                                        645
+                                    ],
+                                    "score": 1.0,
+                                    "content": "by Reed et al. (2016), a lot of information can be extracted simply from pairwise image/text co-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                643,
+                                507,
+                                657
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        643,
+                                        507,
+                                        657
+                                    ],
+                                    "score": 1.0,
+                                    "content": "occurrences. The class labels really only become critically necessary when we define class prototypes,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                655,
+                                505,
+                                666
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        655,
+                                        505,
+                                        666
+                                    ],
+                                    "score": 1.0,
+                                    "content": "i.e. at zero-shot test time. Following this intuition, we define a composite loss function that relies",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                666,
+                                506,
+                                678
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        666,
+                                        506,
+                                        678
+                                    ],
+                                    "score": 1.0,
+                                    "content": "both on the pairwise relationships and on the class labels. The high-level description of the proposed",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                676,
+                                506,
+                                690
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        676,
+                                        506,
+                                        690
+                                    ],
+                                    "score": 1.0,
+                                    "content": "framework is depicted in Figure 1. The framework enables us, among other things, to experiment",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                687,
+                                506,
+                                700
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        687,
+                                        506,
+                                        700
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with the effects of train-time availability of class labels on the quality of zero-shot representations.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                698,
+                                506,
+                                712
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        698,
+                                        506,
+                                        712
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The framework is based on projecting texts and images into a common space and then learning a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                710,
+                                506,
+                                722
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        710,
+                                        506,
+                                        722
+                                    ],
+                                    "score": 1.0,
+                                    "content": "representation based on a mixture of four loss functions: a pairwise text retrieval loss, a pairwise",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                721,
+                                456,
+                                733
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        721,
+                                        456,
+                                        733
+                                    ],
+                                    "score": 1.0,
+                                    "content": "image retrieval loss, a text classifier loss and an image classifier loss (see Algorithm 1).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        }
+                    ],
+                    "index": 35.5
+                }
+            ],
+            "page_idx": 1,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        107,
+                        27,
+                        308,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                25,
+                                309,
+                                39
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        25,
+                                        309,
+                                        39
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Under review as a conference paper at ICLR 2020",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        302,
+                        751,
+                        309,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                301,
+                                750,
+                                310,
+                                763
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        301,
+                                        750,
+                                        310,
+                                        763
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "image",
+                    "bbox": [
+                        153,
+                        78,
+                        461,
+                        231
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                153,
+                                78,
+                                461,
+                                231
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        153,
+                                        78,
+                                        461,
+                                        231
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                153,
+                                                78,
+                                                461,
+                                                231
+                                            ],
+                                            "score": 0.974,
+                                            "type": "image",
+                                            "image_path": "c5306e54b7867e3eb7aee45c6499213721005da29ec547a7ed5a3a76968503c1.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 1,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        153,
+                                        78,
+                                        461,
+                                        129.0
+                                    ],
+                                    "spans": [],
+                                    "index": 0
+                                },
+                                {
+                                    "bbox": [
+                                        153,
+                                        129.0,
+                                        461,
+                                        180.0
+                                    ],
+                                    "spans": [],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        153,
+                                        180.0,
+                                        461,
+                                        231.0
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                106,
+                                239,
+                                505,
+                                338
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        238,
+                                        506,
+                                        252
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                238,
+                                                506,
+                                                252
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 1: The architecture and training diagram describing the proposed method. Each batch consists",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 3
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        250,
+                                        505,
+                                        264
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                250,
+                                                505,
+                                                264
+                                            ],
+                                            "score": 1.0,
+                                            "content": "of randomly sampled instances, i.e. pairs of images and their corresponding texts. Images are",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 4
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        261,
+                                        505,
+                                        275
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                261,
+                                                505,
+                                                275
+                                            ],
+                                            "score": 1.0,
+                                            "content": "embedded via ResNet and texts are embedded via a CNN/LSTM stack. Image and text features",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 5
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        271,
+                                        506,
+                                        286
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                271,
+                                                506,
+                                                286
+                                            ],
+                                            "score": 1.0,
+                                            "content": "are projected via a fully connected layer into the same dimensional space. In this space, distances",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 6
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        283,
+                                        505,
+                                        296
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                283,
+                                                505,
+                                                296
+                                            ],
+                                            "score": 1.0,
+                                            "content": "between text and image features from different instances are computed. The negative distances are",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 7
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        294,
+                                        505,
+                                        306
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                294,
+                                                505,
+                                                306
+                                            ],
+                                            "score": 1.0,
+                                            "content": "fed into softmax to train on both the image and the text retrieval tasks. The image retrieval task",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 8
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        305,
+                                        506,
+                                        318
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                305,
+                                                506,
+                                                318
+                                            ],
+                                            "score": 1.0,
+                                            "content": "consists of retrieving the image corresponding to the given text of the same instance and the text",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 9
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        315,
+                                        505,
+                                        330
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                315,
+                                                505,
+                                                330
+                                            ],
+                                            "score": 1.0,
+                                            "content": "retrieval task is vice versa. In addition to that, image and text embeddings are trained on auxiliary",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 10
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        327,
+                                        430,
+                                        339
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                327,
+                                                430,
+                                                339
+                                            ],
+                                            "score": 1.0,
+                                            "content": "image and text classification tasks on the class labels corresponding to instances.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 11
+                                }
+                            ],
+                            "index": 7
+                        }
+                    ],
+                    "index": 4.0
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        362,
+                        505,
+                        418
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                362,
+                                505,
+                                375
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        362,
+                                        505,
+                                        375
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Metric space rescaling. Metric-based ZSL approaches rely on distances between prototypes and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                372,
+                                506,
+                                387
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        372,
+                                        506,
+                                        387
+                                    ],
+                                    "score": 1.0,
+                                    "content": "query embeddings during inference. They are known to suffer from imbalanced performance on seen",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                384,
+                                505,
+                                397
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        384,
+                                        505,
+                                        397
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and unseen classes (Liu et al., 2018). Previous work proposed to use a heuristic trick, calibrated",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                394,
+                                505,
+                                408
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        394,
+                                        505,
+                                        408
+                                    ],
+                                    "score": 1.0,
+                                    "content": "stacking (Chao et al., 2016) or calibration (Das & Lee, 2019), to solve the problem. We refer to this",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                406,
+                                485,
+                                418
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        406,
+                                        485,
+                                        418
+                                    ],
+                                    "score": 1.0,
+                                    "content": "technique as metric rescaling in our work, and provide a sound probabilistic justification for it.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        }
+                    ],
+                    "index": 14,
+                    "bbox_fs": [
+                        105,
+                        362,
+                        506,
+                        418
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        435,
+                        230,
+                        448
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                435,
+                                232,
+                                451
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        435,
+                                        232,
+                                        451
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2 PROPOSED METHOD",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        }
+                    ],
+                    "index": 17
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        462,
+                        505,
+                        595
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                462,
+                                506,
+                                475
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        462,
+                                        506,
+                                        475
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In this paper, we specifically target the fine-grained visual description scenario, as defined by Reed",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                474,
+                                506,
+                                486
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        474,
+                                        506,
+                                        486
+                                    ],
+                                    "score": 1.0,
+                                    "content": "et al. (2016). In this setting, the dataset consists of a number of images from a given set of classes",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                484,
+                                506,
+                                497
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        484,
+                                        506,
+                                        497
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and each image is accompanied by a number of textual descriptions. The task is to learn a joint",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                496,
+                                506,
+                                507
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        496,
+                                        506,
+                                        507
+                                    ],
+                                    "score": 1.0,
+                                    "content": "representation space for images and texts that can be used for zero-shot recognition. An instance of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                506,
+                                506,
+                                519
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        506,
+                                        506,
+                                        519
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the zero-shot multimodal representation learning problem can then be defined as follows. Given a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                516,
+                                506,
+                                530
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        516,
+                                        155,
+                                        530
+                                    ],
+                                    "score": 1.0,
+                                    "content": "training set",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        155,
+                                        517,
+                                        271,
+                                        529
+                                    ],
+                                    "score": 0.28,
+                                    "content": "S = \\{ ( v _ { n } , t _ { n } , y _ { n } ) \\mid v _ { n } \\in \\mathcal { V }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        271,
+                                        516,
+                                        276,
+                                        530
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        276,
+                                        517,
+                                        404,
+                                        529
+                                    ],
+                                    "score": 0.5,
+                                    "content": "t _ { n } \\in \\mathcal { T } , y _ { n } \\in \\mathcal { V } , n = 1 \\ldots N \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        405,
+                                        516,
+                                        506,
+                                        530
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of image, text and label",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                528,
+                                505,
+                                541
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        528,
+                                        312,
+                                        541
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tuples, we are interested in finding representations",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        312,
+                                        529,
+                                        362,
+                                        540
+                                    ],
+                                    "score": 0.92,
+                                    "content": "f _ { \\phi } : \\mathcal { V } \\to \\mathcal { Z }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        363,
+                                        528,
+                                        477,
+                                        541
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of image, parameterized by",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        477,
+                                        529,
+                                        484,
+                                        540
+                                    ],
+                                    "score": 0.83,
+                                    "content": "\\phi",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        484,
+                                        528,
+                                        505,
+                                        541
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                107,
+                                539,
+                                506,
+                                552
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        540,
+                                        160,
+                                        551
+                                    ],
+                                    "score": 0.91,
+                                    "content": "f _ { \\theta } : \\mathcal { T }  \\mathcal { Z }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        160,
+                                        539,
+                                        266,
+                                        552
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of text, parameterized by",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        267,
+                                        540,
+                                        273,
+                                        549
+                                    ],
+                                    "score": 0.77,
+                                    "content": "\\theta",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        273,
+                                        539,
+                                        406,
+                                        552
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", in a common embedding space",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        406,
+                                        540,
+                                        416,
+                                        549
+                                    ],
+                                    "score": 0.8,
+                                    "content": "\\mathcal { Z }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        416,
+                                        539,
+                                        506,
+                                        552
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Furthermore, GZSL",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                549,
+                                506,
+                                563
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        549,
+                                        271,
+                                        563
+                                    ],
+                                    "score": 1.0,
+                                    "content": "problem is defined using the sets of seen",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        271,
+                                        550,
+                                        286,
+                                        561
+                                    ],
+                                    "score": 0.89,
+                                    "content": "\\mathcal { V } ^ { t r }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        287,
+                                        549,
+                                        335,
+                                        563
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and unseen",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        335,
+                                        550,
+                                        351,
+                                        561
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\mathcal { V } ^ { t s }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        351,
+                                        549,
+                                        424,
+                                        563
+                                    ],
+                                    "score": 1.0,
+                                    "content": "classes, such that",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        424,
+                                        550,
+                                        487,
+                                        561
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\mathcal { V } = \\mathcal { V } ^ { t r } \\cup \\mathcal { V } ^ { t s }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        487,
+                                        549,
+                                        506,
+                                        563
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                107,
+                                559,
+                                504,
+                                574
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        561,
+                                        164,
+                                        572
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\bar { \\mathcal { V } } ^ { t r } \\cap \\mathcal { V } ^ { t s } = \\emptyset",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        164,
+                                        559,
+                                        396,
+                                        574
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". The training set will then only contain the seen classes, i.e.",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        396,
+                                        561,
+                                        504,
+                                        573
+                                    ],
+                                    "score": 0.88,
+                                    "content": "S ^ { t r } = \\{ ( v _ { n } , t _ { n } , y _ { n } ) \\mid v _ { n } \\in",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                571,
+                                506,
+                                585
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        573,
+                                        115,
+                                        583
+                                    ],
+                                    "score": 0.4,
+                                    "content": "\\nu",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        115,
+                                        571,
+                                        118,
+                                        585
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        118,
+                                        572,
+                                        198,
+                                        584
+                                    ],
+                                    "score": 0.86,
+                                    "content": "t _ { n } \\in \\mathcal { T } , y _ { n } \\in \\mathcal { y } ^ { t r } \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        199,
+                                        571,
+                                        369,
+                                        585
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and the task is to build a classifier function",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        370,
+                                        573,
+                                        436,
+                                        583
+                                    ],
+                                    "score": 0.9,
+                                    "content": "g : \\mathcal { Z } \\times \\mathcal { Z }  \\mathcal { V }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        436,
+                                        571,
+                                        506,
+                                        585
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". This is different",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                581,
+                                325,
+                                595
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        581,
+                                        248,
+                                        595
+                                    ],
+                                    "score": 1.0,
+                                    "content": "from the ZSL scenario focusing on",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        249,
+                                        583,
+                                        322,
+                                        594
+                                    ],
+                                    "score": 0.91,
+                                    "content": "g : \\mathcal { Z } \\times \\mathcal { Z }  \\mathcal { Y } ^ { t s }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        322,
+                                        581,
+                                        325,
+                                        595
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        }
+                    ],
+                    "index": 23.5,
+                    "bbox_fs": [
+                        104,
+                        462,
+                        506,
+                        595
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        600,
+                        506,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                600,
+                                507,
+                                614
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        600,
+                                        141,
+                                        614
+                                    ],
+                                    "score": 1.0,
+                                    "content": "To build",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        141,
+                                        602,
+                                        148,
+                                        612
+                                    ],
+                                    "score": 0.78,
+                                    "content": "g",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        148,
+                                        600,
+                                        507,
+                                        614
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", most approaches to joint representation learning rely on class labeling to train a represen-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                611,
+                                506,
+                                623
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        611,
+                                        506,
+                                        623
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tation. For example, all the methods reviewed by Xian et al. (2018c) require the access to class labels",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                622,
+                                506,
+                                634
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        622,
+                                        506,
+                                        634
+                                    ],
+                                    "score": 1.0,
+                                    "content": "at train time. We hypothesise that in the fine-grained learning scenario, such as the one described",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                633,
+                                506,
+                                645
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        633,
+                                        506,
+                                        645
+                                    ],
+                                    "score": 1.0,
+                                    "content": "by Reed et al. (2016), a lot of information can be extracted simply from pairwise image/text co-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                643,
+                                507,
+                                657
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        643,
+                                        507,
+                                        657
+                                    ],
+                                    "score": 1.0,
+                                    "content": "occurrences. The class labels really only become critically necessary when we define class prototypes,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                655,
+                                505,
+                                666
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        655,
+                                        505,
+                                        666
+                                    ],
+                                    "score": 1.0,
+                                    "content": "i.e. at zero-shot test time. Following this intuition, we define a composite loss function that relies",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                666,
+                                506,
+                                678
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        666,
+                                        506,
+                                        678
+                                    ],
+                                    "score": 1.0,
+                                    "content": "both on the pairwise relationships and on the class labels. The high-level description of the proposed",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                676,
+                                506,
+                                690
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        676,
+                                        506,
+                                        690
+                                    ],
+                                    "score": 1.0,
+                                    "content": "framework is depicted in Figure 1. The framework enables us, among other things, to experiment",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                687,
+                                506,
+                                700
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        687,
+                                        506,
+                                        700
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with the effects of train-time availability of class labels on the quality of zero-shot representations.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                698,
+                                506,
+                                712
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        698,
+                                        506,
+                                        712
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The framework is based on projecting texts and images into a common space and then learning a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                710,
+                                506,
+                                722
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        710,
+                                        506,
+                                        722
+                                    ],
+                                    "score": 1.0,
+                                    "content": "representation based on a mixture of four loss functions: a pairwise text retrieval loss, a pairwise",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                721,
+                                456,
+                                733
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        721,
+                                        456,
+                                        733
+                                    ],
+                                    "score": 1.0,
+                                    "content": "image retrieval loss, a text classifier loss and an image classifier loss (see Algorithm 1).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        }
+                    ],
+                    "index": 35.5,
+                    "bbox_fs": [
+                        104,
+                        600,
+                        507,
+                        733
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        81,
+                        505,
+                        127
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                81,
+                                505,
+                                94
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        81,
+                                        469,
+                                        94
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Algorithm 1 Loss calculation for a single optimization iteration of the proposed method.",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        469,
+                                        83,
+                                        479,
+                                        92
+                                    ],
+                                    "score": 0.77,
+                                    "content": "N",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        480,
+                                        81,
+                                        505,
+                                        94
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                92,
+                                507,
+                                106
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        92,
+                                        261,
+                                        106
+                                    ],
+                                    "score": 1.0,
+                                    "content": "number of instances in the training set",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        261,
+                                        93,
+                                        276,
+                                        104
+                                    ],
+                                    "score": 0.85,
+                                    "content": "{ \\mathcal { S } } ^ { t r }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        276,
+                                        92,
+                                        280,
+                                        106
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        280,
+                                        93,
+                                        289,
+                                        104
+                                    ],
+                                    "score": 0.81,
+                                    "content": "B",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        290,
+                                        92,
+                                        438,
+                                        106
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is the number of instances per batch,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        438,
+                                        93,
+                                        447,
+                                        104
+                                    ],
+                                    "score": 0.82,
+                                    "content": "C",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        448,
+                                        92,
+                                        507,
+                                        106
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is the number",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                104,
+                                506,
+                                116
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        104,
+                                        208,
+                                        116
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of classes in the train set.",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        208,
+                                        105,
+                                        312,
+                                        116
+                                    ],
+                                    "score": 0.34,
+                                    "content": "\\mathrm { R A N D O M S A M P L E } ( S , B )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        312,
+                                        104,
+                                        376,
+                                        116
+                                    ],
+                                    "score": 1.0,
+                                    "content": "denotes a set of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        376,
+                                        105,
+                                        386,
+                                        114
+                                    ],
+                                    "score": 0.8,
+                                    "content": "B",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        386,
+                                        104,
+                                        506,
+                                        116
+                                    ],
+                                    "score": 1.0,
+                                    "content": "elements chosen uniformly at",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                115,
+                                280,
+                                127
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        115,
+                                        181,
+                                        127
+                                    ],
+                                    "score": 1.0,
+                                    "content": "random from a set",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        182,
+                                        116,
+                                        190,
+                                        125
+                                    ],
+                                    "score": 0.78,
+                                    "content": "s",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        190,
+                                        115,
+                                        280,
+                                        127
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", without replacement.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        }
+                    ],
+                    "index": 1.5
+                },
+                {
+                    "type": "table",
+                    "bbox": [
+                        107,
+                        129,
+                        512,
+                        386
+                    ],
+                    "blocks": [
+                        {
+                            "type": "table_body",
+                            "bbox": [
+                                107,
+                                129,
+                                512,
+                                386
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        129,
+                                        505,
+                                        386
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                107,
+                                                129,
+                                                505,
+                                                386
+                                            ],
+                                            "score": 0.764,
+                                            "html": "<table><tr><td colspan=\"3\">Input: Training set Str = {(U1,t1,yi),...,(UN,tN,yN)},λ ∈ [0,1], κ ∈ [0,1]. Output: The loss J(,θ) for a randomly sampled training batch. I← RANDOMSAMPLE({1,...,N},B) &gt;Select B instance indices for batch</td></tr><tr><td>Jrc(Θ),J1c(Φ)←0,0 foriinZdo</td><td colspan=\"3\">Initialize classification losses Embed images and texts Image classifier probabilities</td></tr><tr><td colspan=\"3\">Zui,Zt←f(Ui）,fθ(ti)</td></tr><tr><td colspan=\"2\">p ← softmax(Wizui +b1) PT ← sOftmax(WTZt +br)</td></tr><tr><td colspan=\"2\">Textclassifierprobabilities JTc(0) ← JTc(0)+ crossentropy(pr,yi) Text classification loss</td></tr><tr><td colspan=\"2\">JIc(𝜙) ← Jic(Φ)+ crossentropy(p1, yi) Image classification loss end for</td></tr><tr><td colspan=\"2\">JTR(Φ,0),J1R(Φ,0)←0,0 Initialize retrieval losses</td></tr><tr><td colspan=\"2\">foriinZdo 1 Text retrieval loss</td></tr><tr><td colspan=\"2\">JTR(Φ,0)←JTR(Φ,0)+ d(zu,Zt）+log∑ exp(-d(Zu,Ztj)) B j∈I</td></tr><tr><td colspan=\"2\">1-B JIR(Φ,0)←JIR(Φ,0)+ d(Zui,Zti) + log∑exp(-d(Zt , Zuj))|&gt; Image retrieval loss</td></tr><tr><td colspan=\"2\">j∈I</td></tr><tr><td colspan=\"2\">J(,θ)←λJTR(𝜙,0)+(1-λ)JIR(Φ,0) Add retrieval loss to the total loss J(Φ,θ)←(1-κ)J(,0)+(Jrc(0)+JIc()) Add classification loss to the total loss</td></tr></table>",
+                                            "type": "table",
+                                            "image_path": "18df0181b501f401a7b2d506b2c75380a51ed2ff3f2b3eb40ec700d88788894d.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 5,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        129,
+                                        512,
+                                        214.66666666666669
+                                    ],
+                                    "spans": [],
+                                    "index": 4
+                                },
+                                {
+                                    "bbox": [
+                                        107,
+                                        214.66666666666669,
+                                        512,
+                                        300.33333333333337
+                                    ],
+                                    "spans": [],
+                                    "index": 5
+                                },
+                                {
+                                    "bbox": [
+                                        107,
+                                        300.33333333333337,
+                                        512,
+                                        386.00000000000006
+                                    ],
+                                    "spans": [],
+                                    "index": 6
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 5
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        405,
+                        252,
+                        416
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                405,
+                                253,
+                                418
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        405,
+                                        253,
+                                        418
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2.1 RETRIEVAL LOSS FUNCTION",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        }
+                    ],
+                    "index": 7
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        425,
+                        505,
+                        482
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                425,
+                                505,
+                                438
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        425,
+                                        505,
+                                        438
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Pairwise cross-modal loss function is based solely on the pairwise relationships between texts and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                436,
+                                505,
+                                450
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        436,
+                                        505,
+                                        450
+                                    ],
+                                    "score": 1.0,
+                                    "content": "images. We choose to use the metric learning approach to capture the relationship between images",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                446,
+                                505,
+                                461
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        446,
+                                        204,
+                                        461
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and texts. Now, suppose",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        204,
+                                        448,
+                                        210,
+                                        458
+                                    ],
+                                    "score": 0.8,
+                                    "content": "d",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        211,
+                                        446,
+                                        255,
+                                        461
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is a metric",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        255,
+                                        448,
+                                        327,
+                                        459
+                                    ],
+                                    "score": 0.9,
+                                    "content": "d : \\mathcal { Z } \\times \\mathcal { Z }  \\mathbb { R } ^ { + }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        328,
+                                        446,
+                                        331,
+                                        461
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        332,
+                                        449,
+                                        342,
+                                        459
+                                    ],
+                                    "score": 0.81,
+                                    "content": "v _ { i }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        342,
+                                        446,
+                                        406,
+                                        461
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is an image and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        406,
+                                        448,
+                                        447,
+                                        460
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\tau = \\{ t _ { j ^ { \\prime } } \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        447,
+                                        446,
+                                        505,
+                                        461
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is a collection",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                458,
+                                505,
+                                471
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        458,
+                                        365,
+                                        471
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of arbitrary texts sampled uniformly at random, of which text",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        365,
+                                        460,
+                                        374,
+                                        471
+                                    ],
+                                    "score": 0.86,
+                                    "content": "t _ { j }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        374,
+                                        458,
+                                        422,
+                                        471
+                                    ],
+                                    "score": 1.0,
+                                    "content": "belongs to",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        422,
+                                        461,
+                                        431,
+                                        470
+                                    ],
+                                    "score": 0.86,
+                                    "content": "v _ { i }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        432,
+                                        458,
+                                        505,
+                                        471
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". We propose the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                470,
+                                495,
+                                483
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        470,
+                                        287,
+                                        483
+                                    ],
+                                    "score": 1.0,
+                                    "content": "following model for the probability of image",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        288,
+                                        471,
+                                        297,
+                                        481
+                                    ],
+                                    "score": 0.86,
+                                    "content": "v _ { i }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        297,
+                                        470,
+                                        332,
+                                        483
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and text",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        332,
+                                        471,
+                                        341,
+                                        482
+                                    ],
+                                    "score": 0.88,
+                                    "content": "t _ { j }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        342,
+                                        470,
+                                        495,
+                                        483
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to belong to the same object instance:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        }
+                    ],
+                    "index": 10
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        187,
+                        485,
+                        424,
+                        515
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                187,
+                                485,
+                                424,
+                                515
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        187,
+                                        485,
+                                        424,
+                                        515
+                                    ],
+                                    "score": 0.93,
+                                    "content": "p _ { \\phi , \\theta } ( i = j | v _ { i } , t _ { j } , \\tau ) = \\frac { \\exp ( - d ( f _ { \\phi } ( v _ { i } ) , f _ { \\theta } ( t _ { j } ) ) ) } { \\sum _ { t _ { j ^ { \\prime } } \\in \\tau } \\exp ( - d ( f _ { \\phi } ( v _ { i } ) , f _ { \\theta } ( t _ { j ^ { \\prime } } ) ) ) } \\ .",
+                                    "type": "interline_equation",
+                                    "image_path": "66c9fefbdd5debf1a919106a675316eaef51240da8ba8bff3384cde5d307f350.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 13.5,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                187,
+                                485,
+                                424,
+                                500.0
+                            ],
+                            "spans": [],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                187,
+                                500.0,
+                                424,
+                                515.0
+                            ],
+                            "spans": [],
+                            "index": 14
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        104,
+                        517,
+                        480,
+                        529
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                516,
+                                482,
+                                530
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        516,
+                                        468,
+                                        530
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The learning is then based on the following cross-entropy loss defined on the batch of size",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        468,
+                                        518,
+                                        477,
+                                        527
+                                    ],
+                                    "score": 0.84,
+                                    "content": "B",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        477,
+                                        516,
+                                        482,
+                                        530
+                                    ],
+                                    "score": 1.0,
+                                    "content": ":",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        }
+                    ],
+                    "index": 15
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        185,
+                        532,
+                        425,
+                        567
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                185,
+                                532,
+                                425,
+                                567
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        185,
+                                        532,
+                                        425,
+                                        567
+                                    ],
+                                    "score": 0.94,
+                                    "content": "J _ { T R } ( \\phi , \\theta ) = - \\frac { 1 } { B } \\sum _ { i , j = 1 } ^ { B } \\ell _ { i , j } \\log p _ { \\phi , \\theta } ( i = j | v _ { i } , t _ { j } , \\{ t _ { j ^ { \\prime } } \\} _ { j ^ { \\prime } = 1 } ^ { B } ) ,",
+                                    "type": "interline_equation",
+                                    "image_path": "2b97ee593256906efa0bca123ca74f839ff6c75fd4f2cca4d9bdd6cf46a14163.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 16.5,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                185,
+                                532,
+                                425,
+                                549.5
+                            ],
+                            "spans": [],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                185,
+                                549.5,
+                                425,
+                                567.0
+                            ],
+                            "spans": [],
+                            "index": 17
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        570,
+                        505,
+                        615
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                570,
+                                505,
+                                583
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        570,
+                                        133,
+                                        583
+                                    ],
+                                    "score": 1.0,
+                                    "content": "where",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        134,
+                                        571,
+                                        148,
+                                        583
+                                    ],
+                                    "score": 0.9,
+                                    "content": "\\ell _ { i , j }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        148,
+                                        570,
+                                        307,
+                                        583
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is a binary indicator of the true match",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        308,
+                                        570,
+                                        341,
+                                        583
+                                    ],
+                                    "score": 0.89,
+                                    "content": "\\mathcal { \\ell } _ { i , j } = 1",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        342,
+                                        570,
+                                        354,
+                                        583
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", if",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        354,
+                                        571,
+                                        376,
+                                        582
+                                    ],
+                                    "score": 0.91,
+                                    "content": "i = j",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        377,
+                                        570,
+                                        505,
+                                        583
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and 0 otherwise). Note that the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                581,
+                                506,
+                                594
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        581,
+                                        506,
+                                        594
+                                    ],
+                                    "score": 1.0,
+                                    "content": "expression above has the interpretation of the text retrieval loss. It attains its smallest value when for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                593,
+                                505,
+                                604
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        593,
+                                        505,
+                                        604
+                                    ],
+                                    "score": 1.0,
+                                    "content": "each image in the batch we manage to assign probability 1 to its respective text and 0 to all other",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                604,
+                                263,
+                                615
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        604,
+                                        263,
+                                        615
+                                    ],
+                                    "score": 1.0,
+                                    "content": "texts. This can be further expanded as:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        }
+                    ],
+                    "index": 19.5
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        137,
+                        617,
+                        475,
+                        657
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                137,
+                                617,
+                                475,
+                                657
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        137,
+                                        617,
+                                        475,
+                                        657
+                                    ],
+                                    "score": 0.93,
+                                    "content": "J _ { T R } ( \\phi , \\theta ) = \\frac { 1 } { B } \\sum _ { i = 1 } ^ { B } \\left( d ( f _ { \\phi } ( v _ { i } ) , f _ { \\theta } ( t _ { i } ) ) + \\log \\Big [ \\sum _ { t _ { j ^ { \\prime } } \\in \\tau } \\exp ( - d ( f _ { \\phi } ( v _ { i } ) , f _ { \\theta } ( t _ { j ^ { \\prime } } ) ) ) \\Big ] \\right) .",
+                                    "type": "interline_equation",
+                                    "image_path": "e94db0ef9c2a0a6f62952bdf1653f0a4f3ffe6247f5fd866035cda9e491a975b.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 23,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                137,
+                                617,
+                                475,
+                                630.3333333333334
+                            ],
+                            "spans": [],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                137,
+                                630.3333333333334,
+                                475,
+                                643.6666666666667
+                            ],
+                            "spans": [],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                137,
+                                643.6666666666667,
+                                475,
+                                657.0000000000001
+                            ],
+                            "spans": [],
+                            "index": 24
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        659,
+                        504,
+                        683
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                659,
+                                506,
+                                673
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        659,
+                                        506,
+                                        673
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Exchanging the order of image and text in the probability model (1) leads to the image retrieval loss,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                107,
+                                670,
+                                465,
+                                684
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        671,
+                                        147,
+                                        683
+                                    ],
+                                    "score": 0.92,
+                                    "content": "J _ { I R } ( \\phi , \\bar { \\theta } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        147,
+                                        670,
+                                        321,
+                                        684
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". The two losses are mixed using parameter",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        322,
+                                        671,
+                                        360,
+                                        683
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\lambda \\in [ 0 , 1 ]",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        360,
+                                        670,
+                                        465,
+                                        684
+                                    ],
+                                    "score": 1.0,
+                                    "content": "as shown in Algorithm 1.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        }
+                    ],
+                    "index": 25.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        687,
+                        505,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                687,
+                                507,
+                                700
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        687,
+                                        507,
+                                        700
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The pairwise retrieval loss functions are responsible for the modality alignment. In addition to those,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                699,
+                                505,
+                                711
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        699,
+                                        505,
+                                        711
+                                    ],
+                                    "score": 1.0,
+                                    "content": "we propose to include, as mentioned above, the usual image and text classifier losses. These losses",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                710,
+                                505,
+                                722
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        710,
+                                        505,
+                                        722
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are responsible for reducing the intraclass variability of representations. The classifier losses are",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                720,
+                                467,
+                                734
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        720,
+                                        324,
+                                        734
+                                    ],
+                                    "score": 1.0,
+                                    "content": "added to the retrieval losses using a mixing parameter",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        324,
+                                        722,
+                                        362,
+                                        732
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\kappa \\in [ 0 , 1 ]",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        362,
+                                        720,
+                                        467,
+                                        734
+                                    ],
+                                    "score": 1.0,
+                                    "content": "as shown in Algorithm 1.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        }
+                    ],
+                    "index": 28.5
+                }
+            ],
+            "page_idx": 2,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        107,
+                        26,
+                        308,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                25,
+                                308,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        25,
+                                        308,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Under review as a conference paper at ICLR 2020",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        302,
+                        751,
+                        309,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                301,
+                                750,
+                                310,
+                                762
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        301,
+                                        750,
+                                        310,
+                                        762
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        81,
+                        505,
+                        127
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                81,
+                                505,
+                                94
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        81,
+                                        469,
+                                        94
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Algorithm 1 Loss calculation for a single optimization iteration of the proposed method.",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        469,
+                                        83,
+                                        479,
+                                        92
+                                    ],
+                                    "score": 0.77,
+                                    "content": "N",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        480,
+                                        81,
+                                        505,
+                                        94
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                92,
+                                507,
+                                106
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        92,
+                                        261,
+                                        106
+                                    ],
+                                    "score": 1.0,
+                                    "content": "number of instances in the training set",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        261,
+                                        93,
+                                        276,
+                                        104
+                                    ],
+                                    "score": 0.85,
+                                    "content": "{ \\mathcal { S } } ^ { t r }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        276,
+                                        92,
+                                        280,
+                                        106
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        280,
+                                        93,
+                                        289,
+                                        104
+                                    ],
+                                    "score": 0.81,
+                                    "content": "B",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        290,
+                                        92,
+                                        438,
+                                        106
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is the number of instances per batch,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        438,
+                                        93,
+                                        447,
+                                        104
+                                    ],
+                                    "score": 0.82,
+                                    "content": "C",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        448,
+                                        92,
+                                        507,
+                                        106
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is the number",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                104,
+                                506,
+                                116
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        104,
+                                        208,
+                                        116
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of classes in the train set.",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        208,
+                                        105,
+                                        312,
+                                        116
+                                    ],
+                                    "score": 0.34,
+                                    "content": "\\mathrm { R A N D O M S A M P L E } ( S , B )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        312,
+                                        104,
+                                        376,
+                                        116
+                                    ],
+                                    "score": 1.0,
+                                    "content": "denotes a set of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        376,
+                                        105,
+                                        386,
+                                        114
+                                    ],
+                                    "score": 0.8,
+                                    "content": "B",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        386,
+                                        104,
+                                        506,
+                                        116
+                                    ],
+                                    "score": 1.0,
+                                    "content": "elements chosen uniformly at",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                115,
+                                280,
+                                127
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        115,
+                                        181,
+                                        127
+                                    ],
+                                    "score": 1.0,
+                                    "content": "random from a set",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        182,
+                                        116,
+                                        190,
+                                        125
+                                    ],
+                                    "score": 0.78,
+                                    "content": "s",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        190,
+                                        115,
+                                        280,
+                                        127
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", without replacement.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        }
+                    ],
+                    "index": 1.5,
+                    "bbox_fs": [
+                        105,
+                        81,
+                        507,
+                        127
+                    ]
+                },
+                {
+                    "type": "table",
+                    "bbox": [
+                        107,
+                        129,
+                        512,
+                        386
+                    ],
+                    "blocks": [
+                        {
+                            "type": "table_body",
+                            "bbox": [
+                                107,
+                                129,
+                                512,
+                                386
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        129,
+                                        505,
+                                        386
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                107,
+                                                129,
+                                                505,
+                                                386
+                                            ],
+                                            "score": 0.764,
+                                            "html": "<table><tr><td colspan=\"3\">Input: Training set Str = {(U1,t1,yi),...,(UN,tN,yN)},λ ∈ [0,1], κ ∈ [0,1]. Output: The loss J(,θ) for a randomly sampled training batch. I← RANDOMSAMPLE({1,...,N},B) &gt;Select B instance indices for batch</td></tr><tr><td>Jrc(Θ),J1c(Φ)←0,0 foriinZdo</td><td colspan=\"3\">Initialize classification losses Embed images and texts Image classifier probabilities</td></tr><tr><td colspan=\"3\">Zui,Zt←f(Ui）,fθ(ti)</td></tr><tr><td colspan=\"2\">p ← softmax(Wizui +b1) PT ← sOftmax(WTZt +br)</td></tr><tr><td colspan=\"2\">Textclassifierprobabilities JTc(0) ← JTc(0)+ crossentropy(pr,yi) Text classification loss</td></tr><tr><td colspan=\"2\">JIc(𝜙) ← Jic(Φ)+ crossentropy(p1, yi) Image classification loss end for</td></tr><tr><td colspan=\"2\">JTR(Φ,0),J1R(Φ,0)←0,0 Initialize retrieval losses</td></tr><tr><td colspan=\"2\">foriinZdo 1 Text retrieval loss</td></tr><tr><td colspan=\"2\">JTR(Φ,0)←JTR(Φ,0)+ d(zu,Zt）+log∑ exp(-d(Zu,Ztj)) B j∈I</td></tr><tr><td colspan=\"2\">1-B JIR(Φ,0)←JIR(Φ,0)+ d(Zui,Zti) + log∑exp(-d(Zt , Zuj))|&gt; Image retrieval loss</td></tr><tr><td colspan=\"2\">j∈I</td></tr><tr><td colspan=\"2\">J(,θ)←λJTR(𝜙,0)+(1-λ)JIR(Φ,0) Add retrieval loss to the total loss J(Φ,θ)←(1-κ)J(,0)+(Jrc(0)+JIc()) Add classification loss to the total loss</td></tr></table>",
+                                            "type": "table",
+                                            "image_path": "18df0181b501f401a7b2d506b2c75380a51ed2ff3f2b3eb40ec700d88788894d.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 5,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        129,
+                                        512,
+                                        214.66666666666669
+                                    ],
+                                    "spans": [],
+                                    "index": 4
+                                },
+                                {
+                                    "bbox": [
+                                        107,
+                                        214.66666666666669,
+                                        512,
+                                        300.33333333333337
+                                    ],
+                                    "spans": [],
+                                    "index": 5
+                                },
+                                {
+                                    "bbox": [
+                                        107,
+                                        300.33333333333337,
+                                        512,
+                                        386.00000000000006
+                                    ],
+                                    "spans": [],
+                                    "index": 6
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 5
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        405,
+                        252,
+                        416
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                405,
+                                253,
+                                418
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        405,
+                                        253,
+                                        418
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2.1 RETRIEVAL LOSS FUNCTION",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        }
+                    ],
+                    "index": 7
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        425,
+                        505,
+                        482
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                425,
+                                505,
+                                438
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        425,
+                                        505,
+                                        438
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Pairwise cross-modal loss function is based solely on the pairwise relationships between texts and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                436,
+                                505,
+                                450
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        436,
+                                        505,
+                                        450
+                                    ],
+                                    "score": 1.0,
+                                    "content": "images. We choose to use the metric learning approach to capture the relationship between images",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                446,
+                                505,
+                                461
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        446,
+                                        204,
+                                        461
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and texts. Now, suppose",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        204,
+                                        448,
+                                        210,
+                                        458
+                                    ],
+                                    "score": 0.8,
+                                    "content": "d",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        211,
+                                        446,
+                                        255,
+                                        461
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is a metric",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        255,
+                                        448,
+                                        327,
+                                        459
+                                    ],
+                                    "score": 0.9,
+                                    "content": "d : \\mathcal { Z } \\times \\mathcal { Z }  \\mathbb { R } ^ { + }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        328,
+                                        446,
+                                        331,
+                                        461
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        332,
+                                        449,
+                                        342,
+                                        459
+                                    ],
+                                    "score": 0.81,
+                                    "content": "v _ { i }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        342,
+                                        446,
+                                        406,
+                                        461
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is an image and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        406,
+                                        448,
+                                        447,
+                                        460
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\tau = \\{ t _ { j ^ { \\prime } } \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        447,
+                                        446,
+                                        505,
+                                        461
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is a collection",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                458,
+                                505,
+                                471
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        458,
+                                        365,
+                                        471
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of arbitrary texts sampled uniformly at random, of which text",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        365,
+                                        460,
+                                        374,
+                                        471
+                                    ],
+                                    "score": 0.86,
+                                    "content": "t _ { j }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        374,
+                                        458,
+                                        422,
+                                        471
+                                    ],
+                                    "score": 1.0,
+                                    "content": "belongs to",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        422,
+                                        461,
+                                        431,
+                                        470
+                                    ],
+                                    "score": 0.86,
+                                    "content": "v _ { i }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        432,
+                                        458,
+                                        505,
+                                        471
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". We propose the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                470,
+                                495,
+                                483
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        470,
+                                        287,
+                                        483
+                                    ],
+                                    "score": 1.0,
+                                    "content": "following model for the probability of image",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        288,
+                                        471,
+                                        297,
+                                        481
+                                    ],
+                                    "score": 0.86,
+                                    "content": "v _ { i }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        297,
+                                        470,
+                                        332,
+                                        483
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and text",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        332,
+                                        471,
+                                        341,
+                                        482
+                                    ],
+                                    "score": 0.88,
+                                    "content": "t _ { j }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        342,
+                                        470,
+                                        495,
+                                        483
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to belong to the same object instance:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        }
+                    ],
+                    "index": 10,
+                    "bbox_fs": [
+                        105,
+                        425,
+                        505,
+                        483
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        187,
+                        485,
+                        424,
+                        515
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                187,
+                                485,
+                                424,
+                                515
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        187,
+                                        485,
+                                        424,
+                                        515
+                                    ],
+                                    "score": 0.93,
+                                    "content": "p _ { \\phi , \\theta } ( i = j | v _ { i } , t _ { j } , \\tau ) = \\frac { \\exp ( - d ( f _ { \\phi } ( v _ { i } ) , f _ { \\theta } ( t _ { j } ) ) ) } { \\sum _ { t _ { j ^ { \\prime } } \\in \\tau } \\exp ( - d ( f _ { \\phi } ( v _ { i } ) , f _ { \\theta } ( t _ { j ^ { \\prime } } ) ) ) } \\ .",
+                                    "type": "interline_equation",
+                                    "image_path": "66c9fefbdd5debf1a919106a675316eaef51240da8ba8bff3384cde5d307f350.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 13.5,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                187,
+                                485,
+                                424,
+                                500.0
+                            ],
+                            "spans": [],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                187,
+                                500.0,
+                                424,
+                                515.0
+                            ],
+                            "spans": [],
+                            "index": 14
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        104,
+                        517,
+                        480,
+                        529
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                516,
+                                482,
+                                530
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        516,
+                                        468,
+                                        530
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The learning is then based on the following cross-entropy loss defined on the batch of size",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        468,
+                                        518,
+                                        477,
+                                        527
+                                    ],
+                                    "score": 0.84,
+                                    "content": "B",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        477,
+                                        516,
+                                        482,
+                                        530
+                                    ],
+                                    "score": 1.0,
+                                    "content": ":",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        }
+                    ],
+                    "index": 15,
+                    "bbox_fs": [
+                        105,
+                        516,
+                        482,
+                        530
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        185,
+                        532,
+                        425,
+                        567
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                185,
+                                532,
+                                425,
+                                567
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        185,
+                                        532,
+                                        425,
+                                        567
+                                    ],
+                                    "score": 0.94,
+                                    "content": "J _ { T R } ( \\phi , \\theta ) = - \\frac { 1 } { B } \\sum _ { i , j = 1 } ^ { B } \\ell _ { i , j } \\log p _ { \\phi , \\theta } ( i = j | v _ { i } , t _ { j } , \\{ t _ { j ^ { \\prime } } \\} _ { j ^ { \\prime } = 1 } ^ { B } ) ,",
+                                    "type": "interline_equation",
+                                    "image_path": "2b97ee593256906efa0bca123ca74f839ff6c75fd4f2cca4d9bdd6cf46a14163.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 16.5,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                185,
+                                532,
+                                425,
+                                549.5
+                            ],
+                            "spans": [],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                185,
+                                549.5,
+                                425,
+                                567.0
+                            ],
+                            "spans": [],
+                            "index": 17
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        570,
+                        505,
+                        615
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                570,
+                                505,
+                                583
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        570,
+                                        133,
+                                        583
+                                    ],
+                                    "score": 1.0,
+                                    "content": "where",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        134,
+                                        571,
+                                        148,
+                                        583
+                                    ],
+                                    "score": 0.9,
+                                    "content": "\\ell _ { i , j }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        148,
+                                        570,
+                                        307,
+                                        583
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is a binary indicator of the true match",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        308,
+                                        570,
+                                        341,
+                                        583
+                                    ],
+                                    "score": 0.89,
+                                    "content": "\\mathcal { \\ell } _ { i , j } = 1",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        342,
+                                        570,
+                                        354,
+                                        583
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", if",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        354,
+                                        571,
+                                        376,
+                                        582
+                                    ],
+                                    "score": 0.91,
+                                    "content": "i = j",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        377,
+                                        570,
+                                        505,
+                                        583
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and 0 otherwise). Note that the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                581,
+                                506,
+                                594
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        581,
+                                        506,
+                                        594
+                                    ],
+                                    "score": 1.0,
+                                    "content": "expression above has the interpretation of the text retrieval loss. It attains its smallest value when for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                593,
+                                505,
+                                604
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        593,
+                                        505,
+                                        604
+                                    ],
+                                    "score": 1.0,
+                                    "content": "each image in the batch we manage to assign probability 1 to its respective text and 0 to all other",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                604,
+                                263,
+                                615
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        604,
+                                        263,
+                                        615
+                                    ],
+                                    "score": 1.0,
+                                    "content": "texts. This can be further expanded as:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        }
+                    ],
+                    "index": 19.5,
+                    "bbox_fs": [
+                        105,
+                        570,
+                        506,
+                        615
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        137,
+                        617,
+                        475,
+                        657
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                137,
+                                617,
+                                475,
+                                657
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        137,
+                                        617,
+                                        475,
+                                        657
+                                    ],
+                                    "score": 0.93,
+                                    "content": "J _ { T R } ( \\phi , \\theta ) = \\frac { 1 } { B } \\sum _ { i = 1 } ^ { B } \\left( d ( f _ { \\phi } ( v _ { i } ) , f _ { \\theta } ( t _ { i } ) ) + \\log \\Big [ \\sum _ { t _ { j ^ { \\prime } } \\in \\tau } \\exp ( - d ( f _ { \\phi } ( v _ { i } ) , f _ { \\theta } ( t _ { j ^ { \\prime } } ) ) ) \\Big ] \\right) .",
+                                    "type": "interline_equation",
+                                    "image_path": "e94db0ef9c2a0a6f62952bdf1653f0a4f3ffe6247f5fd866035cda9e491a975b.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 23,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                137,
+                                617,
+                                475,
+                                630.3333333333334
+                            ],
+                            "spans": [],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                137,
+                                630.3333333333334,
+                                475,
+                                643.6666666666667
+                            ],
+                            "spans": [],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                137,
+                                643.6666666666667,
+                                475,
+                                657.0000000000001
+                            ],
+                            "spans": [],
+                            "index": 24
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        659,
+                        504,
+                        683
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                659,
+                                506,
+                                673
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        659,
+                                        506,
+                                        673
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Exchanging the order of image and text in the probability model (1) leads to the image retrieval loss,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                107,
+                                670,
+                                465,
+                                684
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        671,
+                                        147,
+                                        683
+                                    ],
+                                    "score": 0.92,
+                                    "content": "J _ { I R } ( \\phi , \\bar { \\theta } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        147,
+                                        670,
+                                        321,
+                                        684
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". The two losses are mixed using parameter",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        322,
+                                        671,
+                                        360,
+                                        683
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\lambda \\in [ 0 , 1 ]",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        360,
+                                        670,
+                                        465,
+                                        684
+                                    ],
+                                    "score": 1.0,
+                                    "content": "as shown in Algorithm 1.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        }
+                    ],
+                    "index": 25.5,
+                    "bbox_fs": [
+                        105,
+                        659,
+                        506,
+                        684
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        687,
+                        505,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                687,
+                                507,
+                                700
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        687,
+                                        507,
+                                        700
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The pairwise retrieval loss functions are responsible for the modality alignment. In addition to those,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                699,
+                                505,
+                                711
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        699,
+                                        505,
+                                        711
+                                    ],
+                                    "score": 1.0,
+                                    "content": "we propose to include, as mentioned above, the usual image and text classifier losses. These losses",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                710,
+                                505,
+                                722
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        710,
+                                        505,
+                                        722
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are responsible for reducing the intraclass variability of representations. The classifier losses are",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                720,
+                                467,
+                                734
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        720,
+                                        324,
+                                        734
+                                    ],
+                                    "score": 1.0,
+                                    "content": "added to the retrieval losses using a mixing parameter",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        324,
+                                        722,
+                                        362,
+                                        732
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\kappa \\in [ 0 , 1 ]",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        362,
+                                        720,
+                                        467,
+                                        734
+                                    ],
+                                    "score": 1.0,
+                                    "content": "as shown in Algorithm 1.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        }
+                    ],
+                    "index": 28.5,
+                    "bbox_fs": [
+                        105,
+                        687,
+                        507,
+                        734
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        82,
+                        397,
+                        94
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                82,
+                                398,
+                                96
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        82,
+                                        398,
+                                        96
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2.2 BALANCING ACCURACY ON THE SEEN AND UNSEEN CLASSES",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        }
+                    ],
+                    "index": 0
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        102,
+                        505,
+                        143
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                101,
+                                506,
+                                117
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        101,
+                                        227,
+                                        117
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Let us define class prototypes",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        227,
+                                        103,
+                                        247,
+                                        115
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\mathbf p ( y )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        248,
+                                        101,
+                                        370,
+                                        117
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", each based on the set of texts",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        370,
+                                        104,
+                                        382,
+                                        117
+                                    ],
+                                    "score": 0.89,
+                                    "content": "\\mathcal { T } _ { y }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        382,
+                                        101,
+                                        457,
+                                        117
+                                    ],
+                                    "score": 1.0,
+                                    "content": "belonging to class",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        457,
+                                        105,
+                                        464,
+                                        115
+                                    ],
+                                    "score": 0.63,
+                                    "content": "y",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        464,
+                                        101,
+                                        468,
+                                        117
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        468,
+                                        103,
+                                        506,
+                                        116
+                                    ],
+                                    "score": 0.88,
+                                    "content": "\\mathbf { \\{ p ( }  y \\mathbf { ) = }",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                107,
+                                115,
+                                506,
+                                133
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        115,
+                                        221,
+                                        133
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\begin{array} { r } { \\frac { 1 } { | T _ { y } | } \\sum _ { t _ { i } \\in \\mathcal { T } _ { y } } f _ { \\theta } ( t _ { i } ) \\ | \\ y \\in \\mathcal { V } \\} } \\end{array}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        222,
+                                        115,
+                                        506,
+                                        131
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". In the context of GZSL, the nearest neighbor decision rule for a given",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                131,
+                                348,
+                                144
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        131,
+                                        133,
+                                        144
+                                    ],
+                                    "score": 1.0,
+                                    "content": "image",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        133,
+                                        134,
+                                        140,
+                                        141
+                                    ],
+                                    "score": 0.77,
+                                    "content": "v",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        140,
+                                        131,
+                                        204,
+                                        144
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and its features",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        204,
+                                        131,
+                                        251,
+                                        144
+                                    ],
+                                    "score": 0.93,
+                                    "content": "{ \\bf z } _ { v } = f _ { \\phi } ( v )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        252,
+                                        131,
+                                        348,
+                                        144
+                                    ],
+                                    "score": 1.0,
+                                    "content": "has the following form:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        }
+                    ],
+                    "index": 2
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        251,
+                        149,
+                        359,
+                        168
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                251,
+                                149,
+                                359,
+                                168
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        251,
+                                        149,
+                                        359,
+                                        168
+                                    ],
+                                    "score": 0.94,
+                                    "content": "\\boldsymbol { \\widehat { y } } = \\arg \\operatorname* { m i n } _ { \\boldsymbol { y } \\in \\mathcal { V } } d ( \\mathbf { z } _ { v } , \\mathbf { p } ( \\boldsymbol { y } ) ) .",
+                                    "type": "interline_equation",
+                                    "image_path": "3d3a73b561686471c4dfe86df22f349395ea59a86c528b9db4f01ce94ccf5acf.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 4,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                251,
+                                149,
+                                359,
+                                168
+                            ],
+                            "spans": [],
+                            "index": 4
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        174,
+                        505,
+                        253
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                175,
+                                506,
+                                187
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        175,
+                                        506,
+                                        187
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The most acute problem faced in this setup is the accuracy imbalance between seen and unseen classes.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                186,
+                                505,
+                                198
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        186,
+                                        505,
+                                        198
+                                    ],
+                                    "score": 1.0,
+                                    "content": "A very representative case clearly outlining the imbalance problem is presented in Table 6 of (Xian",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                196,
+                                505,
+                                210
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        196,
+                                        505,
+                                        210
+                                    ],
+                                    "score": 1.0,
+                                    "content": "et al., 2018c), where accuracy on the seen classes is always significantly greater than the accuracy on",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                209,
+                                505,
+                                220
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        209,
+                                        505,
+                                        220
+                                    ],
+                                    "score": 1.0,
+                                    "content": "unseen ones. In order to measure and control the imbalance, three metrics are commonly used to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                219,
+                                505,
+                                231
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        219,
+                                        505,
+                                        231
+                                    ],
+                                    "score": 1.0,
+                                    "content": "assess the classification performance in the GZSL scenario: the Top-1 accuracy on the seen categories",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                230,
+                                505,
+                                243
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        230,
+                                        296,
+                                        243
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(s), the Top-1 accuracy on the unseen categories",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        296,
+                                        231,
+                                        309,
+                                        241
+                                    ],
+                                    "score": 0.31,
+                                    "content": "\\mathbf { \\Pi } ^ { ( \\mathbf { u } ) }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        309,
+                                        230,
+                                        412,
+                                        243
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and their harmonic mean,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        412,
+                                        230,
+                                        483,
+                                        242
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\mathbf { H } = \\mathbf { u } \\cdot \\mathbf { s } / ( \\mathbf { u } + \\mathbf { s } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        483,
+                                        230,
+                                        505,
+                                        243
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". The",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                241,
+                                438,
+                                253
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        241,
+                                        303,
+                                        253
+                                    ],
+                                    "score": 1.0,
+                                    "content": "main metric to assess GZSL performance is then",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        303,
+                                        241,
+                                        312,
+                                        251
+                                    ],
+                                    "score": 0.58,
+                                    "content": "\\mathbf { H }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        312,
+                                        241,
+                                        403,
+                                        253
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", which quantifies both",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        404,
+                                        243,
+                                        411,
+                                        251
+                                    ],
+                                    "score": 0.44,
+                                    "content": "\\mathbf { u }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        411,
+                                        241,
+                                        438,
+                                        253
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and s.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        }
+                    ],
+                    "index": 8
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        257,
+                        505,
+                        325
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                257,
+                                505,
+                                270
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        257,
+                                        284,
+                                        270
+                                    ],
+                                    "score": 1.0,
+                                    "content": "To formalize the problem, we first introduce",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        284,
+                                        260,
+                                        295,
+                                        269
+                                    ],
+                                    "score": 0.83,
+                                    "content": "y _ { v }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        295,
+                                        257,
+                                        412,
+                                        270
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", the true class label of image",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        413,
+                                        260,
+                                        419,
+                                        268
+                                    ],
+                                    "score": 0.68,
+                                    "content": "v",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        419,
+                                        257,
+                                        505,
+                                        270
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Mathematically, the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                266,
+                                505,
+                                283
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        266,
+                                        229,
+                                        283
+                                    ],
+                                    "score": 1.0,
+                                    "content": "main GZSL pain point is that",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        230,
+                                        269,
+                                        320,
+                                        281
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\mathbb { P } \\{ \\widehat { y } \\in \\mathcal { y } ^ { t r } | y _ { v } \\in \\mathcal { y } ^ { t s } \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        321,
+                                        266,
+                                        436,
+                                        283
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is significantly greater than",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        436,
+                                        269,
+                                        505,
+                                        281
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\mathbb { P } \\{ \\widehat { y } \\in \\mathcal { y } ^ { t s } | y _ { v } \\in",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                278,
+                                506,
+                                293
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        280,
+                                        127,
+                                        292
+                                    ],
+                                    "score": 0.9,
+                                    "content": "\\mathcal { V } ^ { t r } \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        128,
+                                        278,
+                                        506,
+                                        293
+                                    ],
+                                    "score": 1.0,
+                                    "content": "b b. In other words, the problem is that a given image is more likely to be confused with one of the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                291,
+                                505,
+                                303
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        291,
+                                        505,
+                                        303
+                                    ],
+                                    "score": 1.0,
+                                    "content": "seen classes if it belongs to an unseen class than vice versa. Our approach to solving the problem",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                300,
+                                506,
+                                315
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        300,
+                                        506,
+                                        315
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is based on the following probabilistic representation of the event space for the decision rule in",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                311,
+                                163,
+                                326
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        311,
+                                        163,
+                                        326
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Equation (4):",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        }
+                    ],
+                    "index": 14.5
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        139,
+                        330,
+                        471,
+                        356
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                139,
+                                330,
+                                471,
+                                356
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        139,
+                                        330,
+                                        471,
+                                        356
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\mathbb { P } \\{ \\boldsymbol { { \\widehat { y } } } \\in { \\mathcal { Y } } ^ { t r } | y _ { v } \\in { \\mathcal { Y } } ^ { t s } \\} = \\mathbb { P } \\left\\{ \\operatorname* { m i n } _ { y \\in { \\mathcal { Y } } ^ { t r } } d ( \\mathbf { z } _ { v } , \\mathbf { p } ( y ) ) < \\operatorname* { m i n } _ { y \\in { \\mathcal { Y } } ^ { t s } } d ( \\mathbf { z } _ { v } , \\mathbf { p } ( y ) ) \\mid y _ { v } \\in { \\mathcal { Y } } ^ { t s } \\right\\} .",
+                                    "type": "interline_equation",
+                                    "image_path": "c4992af813698d95795db7215aceffe43193c03d6626652d1543fc22759ab44a.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 18,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                139,
+                                330,
+                                471,
+                                356
+                            ],
+                            "spans": [],
+                            "index": 18
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        362,
+                        505,
+                        385
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                362,
+                                505,
+                                375
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        362,
+                                        505,
+                                        375
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Rephrasing, the most acute GZSL error happens when the prototype of one of the seen classes is",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                374,
+                                506,
+                                387
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        374,
+                                        506,
+                                        387
+                                    ],
+                                    "score": 1.0,
+                                    "content": "closer to an image embedding from an unseen class than any of the prototypes of the unseen classes.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        }
+                    ],
+                    "index": 19.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        390,
+                        505,
+                        424
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                389,
+                                504,
+                                403
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        389,
+                                        415,
+                                        403
+                                    ],
+                                    "score": 1.0,
+                                    "content": "To rectify the situation we propose the following very direct solution to balance",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        415,
+                                        390,
+                                        504,
+                                        403
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\mathbb { P } \\{ \\widehat { y } \\in \\mathcal { y } ^ { t r } | y _ { v } \\in \\mathcal { y } ^ { t s } \\}",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                401,
+                                506,
+                                414
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        401,
+                                        124,
+                                        414
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        124,
+                                        401,
+                                        219,
+                                        414
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\mathbb { P } \\{ \\hat { \\mathcal { Y } } \\in \\mathcal { Y } ^ { t s } | y _ { v } \\in \\mathcal { Y } ^ { t \\bar { r } } \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        220,
+                                        401,
+                                        355,
+                                        414
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". We introduce a positive scalar",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        356,
+                                        401,
+                                        392,
+                                        412
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\alpha \\in \\mathbb { R } ^ { + }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        392,
+                                        401,
+                                        506,
+                                        414
+                                    ],
+                                    "score": 1.0,
+                                    "content": "band scale all the distances",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                412,
+                                502,
+                                426
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        412,
+                                        268,
+                                        426
+                                    ],
+                                    "score": 1.0,
+                                    "content": "bcorresponding to the seen prototypes by",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        268,
+                                        413,
+                                        293,
+                                        423
+                                    ],
+                                    "score": 0.89,
+                                    "content": "1 + \\alpha",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        293,
+                                        412,
+                                        486,
+                                        426
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". This gives rise to the following scaled distance",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        486,
+                                        413,
+                                        498,
+                                        424
+                                    ],
+                                    "score": 0.88,
+                                    "content": "d _ { \\alpha }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        498,
+                                        412,
+                                        502,
+                                        426
+                                    ],
+                                    "score": 1.0,
+                                    "content": ":",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        }
+                    ],
+                    "index": 22
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        198,
+                        430,
+                        414,
+                        461
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                198,
+                                430,
+                                414,
+                                461
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        198,
+                                        430,
+                                        414,
+                                        461
+                                    ],
+                                    "score": 0.92,
+                                    "content": "d _ { \\alpha } ( \\mathbf { z } _ { v } , \\mathbf { p } ( y ) ) = \\left\\{ \\begin{array} { l l } { ( 1 + \\alpha ) d ( \\mathbf { z } _ { v } , \\mathbf { p } ( y ) ) , } & { \\mathrm { i f ~ } y \\in \\mathcal { Y } ^ { t r } } \\\\ { d ( \\mathbf { z } _ { v } , \\mathbf { p } ( y ) ) , } & { \\mathrm { o t h e r w i s e } } \\end{array} \\right. .",
+                                    "type": "interline_equation",
+                                    "image_path": "a62834e649aa08ed171d61fabc0cba31952b1b0e4dc94b2e737d501dd6da9e62.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 24.5,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                198,
+                                430,
+                                414,
+                                445.5
+                            ],
+                            "spans": [],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                198,
+                                445.5,
+                                414,
+                                461.0
+                            ],
+                            "spans": [],
+                            "index": 25
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        466,
+                        484,
+                        478
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                465,
+                                486,
+                                479
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        465,
+                                        387,
+                                        479
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The misclassification between unseen as seen classes for the classifier",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        387,
+                                        466,
+                                        398,
+                                        478
+                                    ],
+                                    "score": 0.89,
+                                    "content": "\\widehat { y } _ { \\alpha }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        398,
+                                        465,
+                                        486,
+                                        479
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", based on (6) is then:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        }
+                    ],
+                    "index": 26
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        118,
+                        483,
+                        480,
+                        510
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                118,
+                                483,
+                                480,
+                                510
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        118,
+                                        483,
+                                        480,
+                                        510
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\mathbb { P } \\{ \\widehat { y } _ { \\alpha } \\in \\mathcal { Y } ^ { t r } | y _ { v } \\in \\mathcal { Y } ^ { t s } \\} = \\mathbb { P } \\left\\{ \\left( 1 + \\alpha \\right) \\operatorname* { m i n } _ { y \\in \\mathcal { Y } ^ { t r } } d ( \\mathbf { z } _ { v } , \\mathbf { p } ( y ) ) < \\operatorname* { m i n } _ { y \\in \\mathcal { Y } ^ { t s } } d ( \\mathbf { z } _ { v } , \\mathbf { p } ( y ) ) | y _ { v } \\in \\mathcal { Y } ^ { t s } \\right\\} \\ ,",
+                                    "type": "interline_equation",
+                                    "image_path": "6c11b414afece46ca498ac50b8688e023ff6de855077e8453e68919cc2b737ba.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 27,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                118,
+                                483,
+                                480,
+                                510
+                            ],
+                            "spans": [],
+                            "index": 27
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        516,
+                        505,
+                        617
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                515,
+                                504,
+                                531
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        515,
+                                        283,
+                                        531
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and it has the following property: for any",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        283,
+                                        518,
+                                        343,
+                                        529
+                                    ],
+                                    "score": 0.81,
+                                    "content": "0 \\leq \\alpha _ { 1 } \\leq \\alpha _ { 2 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        343,
+                                        515,
+                                        347,
+                                        531
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        347,
+                                        516,
+                                        504,
+                                        530
+                                    ],
+                                    "score": 0.86,
+                                    "content": "\\mathbb { P } \\{ \\widehat { y } _ { \\alpha _ { 1 } } ~ \\in ~ \\mathcal { Y } ^ { t r } | y _ { v } ~ \\in ~ \\mathcal { Y } ^ { t s } \\} ~ \\geq ~ \\mathbb { P } \\{ \\widehat { y } _ { \\alpha _ { 2 } } ~ \\in ~",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                107,
+                                525,
+                                507,
+                                543
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        528,
+                                        168,
+                                        540
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\mathcal { V } ^ { t r } | y _ { v } \\in \\mathcal { V } ^ { t s } \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        169,
+                                        525,
+                                        190,
+                                        543
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", i.e.",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        191,
+                                        528,
+                                        289,
+                                        540
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\mathbb { P } \\{ \\widehat { y } _ { \\alpha } \\in \\mathcal { V } ^ { t \\bar { r } } | y _ { v } \\in \\mathcal { V } ^ { t s } \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        289,
+                                        525,
+                                        463,
+                                        543
+                                    ],
+                                    "score": 1.0,
+                                    "content": "b   is a monotone non-increasing function of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        464,
+                                        531,
+                                        471,
+                                        538
+                                    ],
+                                    "score": 0.79,
+                                    "content": "\\alpha",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        472,
+                                        525,
+                                        507,
+                                        543
+                                    ],
+                                    "score": 1.0,
+                                    "content": "b and we",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                538,
+                                505,
+                                552
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        538,
+                                        222,
+                                        552
+                                    ],
+                                    "score": 1.0,
+                                    "content": "bcan reduce it by increasing",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        222,
+                                        542,
+                                        230,
+                                        550
+                                    ],
+                                    "score": 0.77,
+                                    "content": "\\alpha",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        231,
+                                        538,
+                                        469,
+                                        552
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(please refer to Appendix A for a proof). Consider now",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        470,
+                                        539,
+                                        505,
+                                        551
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\mathbb { P } \\{ \\widehat { y } _ { \\alpha } \\in",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                107,
+                                549,
+                                506,
+                                563
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        550,
+                                        167,
+                                        562
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\mathcal { V } ^ { t r } | y _ { v } \\in \\mathcal { V } ^ { t r } \\big \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        168,
+                                        549,
+                                        368,
+                                        563
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", which is a probability that we classify an image",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        369,
+                                        552,
+                                        375,
+                                        560
+                                    ],
+                                    "score": 0.69,
+                                    "content": "v",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        376,
+                                        549,
+                                        506,
+                                        563
+                                    ],
+                                    "score": 1.0,
+                                    "content": "bfrom one of the seen classes as",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                560,
+                                506,
+                                575
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        560,
+                                        506,
+                                        575
+                                    ],
+                                    "score": 1.0,
+                                    "content": "still one of the seen classes. Using exactly the same chain of arguments as in Appendix A, it is",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                572,
+                                505,
+                                585
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        572,
+                                        405,
+                                        585
+                                    ],
+                                    "score": 1.0,
+                                    "content": "straightforward to show that the probability is a non-increasing function of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        405,
+                                        574,
+                                        412,
+                                        582
+                                    ],
+                                    "score": 0.75,
+                                    "content": "\\alpha",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        413,
+                                        572,
+                                        505,
+                                        585
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Hence the probability",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                581,
+                                507,
+                                597
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        582,
+                                        324,
+                                        595
+                                    ],
+                                    "score": 0.86,
+                                    "content": "\\mathbb { P } \\{ \\hat { y } _ { \\alpha } ^ { \\sim } \\in \\mathcal { V } ^ { t s } | y _ { v } \\in \\mathcal { V } ^ { t r } \\} = 1 - \\mathbb { P } \\{ \\hat { y } _ { \\alpha } \\in \\mathcal { V } ^ { \\hat { t r } } | y _ { v } \\in \\mathcal { V } ^ { t r } \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        325,
+                                        581,
+                                        450,
+                                        597
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is a non-decreasing function of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        451,
+                                        585,
+                                        458,
+                                        593
+                                    ],
+                                    "score": 0.73,
+                                    "content": "\\alpha",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        459,
+                                        581,
+                                        507,
+                                        597
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Therefore,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                595,
+                                505,
+                                606
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        595,
+                                        117,
+                                        606
+                                    ],
+                                    "score": 1.0,
+                                    "content": "as",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        117,
+                                        596,
+                                        125,
+                                        604
+                                    ],
+                                    "score": 0.79,
+                                    "content": "\\alpha",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        126,
+                                        595,
+                                        505,
+                                        606
+                                    ],
+                                    "score": 1.0,
+                                    "content": "increases, we expect more classification errors in classifying images from seen classes, because",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                604,
+                                349,
+                                618
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        604,
+                                        349,
+                                        618
+                                    ],
+                                    "score": 1.0,
+                                    "content": "some of them will be classified as one of the unseen classes.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        }
+                    ],
+                    "index": 32
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        621,
+                        505,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                622,
+                                505,
+                                635
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        622,
+                                        401,
+                                        635
+                                    ],
+                                    "score": 1.0,
+                                    "content": "To sum up, given the arguments presented above we expect that by varying",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        402,
+                                        623,
+                                        428,
+                                        632
+                                    ],
+                                    "score": 0.89,
+                                    "content": "\\alpha > 0",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        428,
+                                        622,
+                                        505,
+                                        635
+                                    ],
+                                    "score": 1.0,
+                                    "content": "we can balance the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                632,
+                                506,
+                                646
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        632,
+                                        147,
+                                        646
+                                    ],
+                                    "score": 1.0,
+                                    "content": "error rate",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        147,
+                                        632,
+                                        243,
+                                        645
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\mathbb { P } \\mathbf { \\bar { \\{ y _ { \\alpha } \\in \\mathcal { V } ^ { t r } \\vert y _ { v } \\in \\mathcal { V } ^ { t s } \\} } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        243,
+                                        632,
+                                        506,
+                                        646
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of leaking the unseen class images into seen class classification",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                643,
+                                506,
+                                657
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        643,
+                                        218,
+                                        657
+                                    ],
+                                    "score": 1.0,
+                                    "content": "bdecision and the error rate",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        218,
+                                        644,
+                                        317,
+                                        656
+                                    ],
+                                    "score": 0.9,
+                                    "content": "\\mathbb { P } \\{ \\widehat { y } _ { \\alpha } ^ { } \\in \\mathcal { V } ^ { t s } | y _ { v } ^ { - } \\in \\mathcal { V } ^ { t r } \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        318,
+                                        643,
+                                        506,
+                                        657
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of leaking the seen class images into unseen",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                654,
+                                504,
+                                667
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        654,
+                                        409,
+                                        667
+                                    ],
+                                    "score": 1.0,
+                                    "content": "bclass classification decision. This is possible as we just showed above that",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        410,
+                                        654,
+                                        504,
+                                        667
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\mathbb { P } \\{ \\widehat { y } _ { \\alpha } \\in \\mathcal { V } ^ { t r } | y _ { v } \\in \\mathcal { V } ^ { t s } \\}",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                665,
+                                506,
+                                679
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        665,
+                                        233,
+                                        679
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is a non-increasing function of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        234,
+                                        668,
+                                        241,
+                                        676
+                                    ],
+                                    "score": 0.78,
+                                    "content": "\\alpha",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        241,
+                                        665,
+                                        270,
+                                        679
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", while",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        270,
+                                        666,
+                                        365,
+                                        678
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\mathbb { P } \\{ \\widehat { y } _ { \\alpha } \\in \\mathcal { V } ^ { t s } | y _ { v } \\in \\mathcal { V } ^ { t r } \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        365,
+                                        665,
+                                        506,
+                                        679
+                                    ],
+                                    "score": 1.0,
+                                    "content": "b is a non-decreasing one. It is also",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                677,
+                                505,
+                                689
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        677,
+                                        223,
+                                        689
+                                    ],
+                                    "score": 1.0,
+                                    "content": "important to emphasize that",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        224,
+                                        679,
+                                        231,
+                                        687
+                                    ],
+                                    "score": 0.77,
+                                    "content": "\\alpha",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        232,
+                                        677,
+                                        505,
+                                        689
+                                    ],
+                                    "score": 1.0,
+                                    "content": "bis applied only to distances between the query embedding and the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                687,
+                                505,
+                                700
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        687,
+                                        475,
+                                        700
+                                    ],
+                                    "score": 1.0,
+                                    "content": "prototypes of seen classes and it is constant over seen classes. Therefore, the application of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        475,
+                                        690,
+                                        483,
+                                        698
+                                    ],
+                                    "score": 0.77,
+                                    "content": "\\alpha",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        483,
+                                        687,
+                                        505,
+                                        700
+                                    ],
+                                    "score": 1.0,
+                                    "content": "does",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                697,
+                                506,
+                                712
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        697,
+                                        343,
+                                        712
+                                    ],
+                                    "score": 1.0,
+                                    "content": "not at all affect the classification error rates either within",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        343,
+                                        698,
+                                        359,
+                                        709
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\mathcal { V } ^ { t r }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        359,
+                                        697,
+                                        401,
+                                        712
+                                    ],
+                                    "score": 1.0,
+                                    "content": "or within",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        401,
+                                        698,
+                                        416,
+                                        710
+                                    ],
+                                    "score": 0.89,
+                                    "content": "\\mathcal { V } ^ { t s }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        417,
+                                        697,
+                                        457,
+                                        712
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Varying",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        458,
+                                        700,
+                                        466,
+                                        709
+                                    ],
+                                    "score": 0.76,
+                                    "content": "\\alpha",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        466,
+                                        697,
+                                        506,
+                                        712
+                                    ],
+                                    "score": 1.0,
+                                    "content": "balances",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                709,
+                                505,
+                                722
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        709,
+                                        505,
+                                        722
+                                    ],
+                                    "score": 1.0,
+                                    "content": "exclusively the classification errors arising from transitions between seen and unseen class labels. We",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                720,
+                                338,
+                                733
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        720,
+                                        269,
+                                        733
+                                    ],
+                                    "score": 1.0,
+                                    "content": "study the empirical aspects of balancing",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        269,
+                                        724,
+                                        276,
+                                        730
+                                    ],
+                                    "score": 0.77,
+                                    "content": "\\alpha",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        277,
+                                        720,
+                                        338,
+                                        733
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in Section 4.4.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        }
+                    ],
+                    "index": 41.5
+                }
+            ],
+            "page_idx": 3,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        106,
+                        26,
+                        308,
+                        38
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                25,
+                                308,
+                                39
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        25,
+                                        308,
+                                        39
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Under review as a conference paper at ICLR 2020",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        302,
+                        751,
+                        308,
+                        759
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                301,
+                                750,
+                                309,
+                                762
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        301,
+                                        750,
+                                        309,
+                                        762
+                                    ],
+                                    "score": 1.0,
+                                    "content": "",
+                                    "type": "text",
+                                    "height": 12,
+                                    "width": 8
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        82,
+                        397,
+                        94
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                82,
+                                398,
+                                96
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        82,
+                                        398,
+                                        96
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2.2 BALANCING ACCURACY ON THE SEEN AND UNSEEN CLASSES",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        }
+                    ],
+                    "index": 0
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        102,
+                        505,
+                        143
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                101,
+                                506,
+                                117
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        101,
+                                        227,
+                                        117
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Let us define class prototypes",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        227,
+                                        103,
+                                        247,
+                                        115
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\mathbf p ( y )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        248,
+                                        101,
+                                        370,
+                                        117
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", each based on the set of texts",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        370,
+                                        104,
+                                        382,
+                                        117
+                                    ],
+                                    "score": 0.89,
+                                    "content": "\\mathcal { T } _ { y }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        382,
+                                        101,
+                                        457,
+                                        117
+                                    ],
+                                    "score": 1.0,
+                                    "content": "belonging to class",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        457,
+                                        105,
+                                        464,
+                                        115
+                                    ],
+                                    "score": 0.63,
+                                    "content": "y",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        464,
+                                        101,
+                                        468,
+                                        117
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        468,
+                                        103,
+                                        506,
+                                        116
+                                    ],
+                                    "score": 0.88,
+                                    "content": "\\mathbf { \\{ p ( }  y \\mathbf { ) = }",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                107,
+                                115,
+                                506,
+                                133
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        115,
+                                        221,
+                                        133
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\begin{array} { r } { \\frac { 1 } { | T _ { y } | } \\sum _ { t _ { i } \\in \\mathcal { T } _ { y } } f _ { \\theta } ( t _ { i } ) \\ | \\ y \\in \\mathcal { V } \\} } \\end{array}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        222,
+                                        115,
+                                        506,
+                                        131
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". In the context of GZSL, the nearest neighbor decision rule for a given",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                131,
+                                348,
+                                144
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        131,
+                                        133,
+                                        144
+                                    ],
+                                    "score": 1.0,
+                                    "content": "image",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        133,
+                                        134,
+                                        140,
+                                        141
+                                    ],
+                                    "score": 0.77,
+                                    "content": "v",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        140,
+                                        131,
+                                        204,
+                                        144
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and its features",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        204,
+                                        131,
+                                        251,
+                                        144
+                                    ],
+                                    "score": 0.93,
+                                    "content": "{ \\bf z } _ { v } = f _ { \\phi } ( v )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        252,
+                                        131,
+                                        348,
+                                        144
+                                    ],
+                                    "score": 1.0,
+                                    "content": "has the following form:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        }
+                    ],
+                    "index": 2,
+                    "bbox_fs": [
+                        104,
+                        101,
+                        506,
+                        144
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        251,
+                        149,
+                        359,
+                        168
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                251,
+                                149,
+                                359,
+                                168
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        251,
+                                        149,
+                                        359,
+                                        168
+                                    ],
+                                    "score": 0.94,
+                                    "content": "\\boldsymbol { \\widehat { y } } = \\arg \\operatorname* { m i n } _ { \\boldsymbol { y } \\in \\mathcal { V } } d ( \\mathbf { z } _ { v } , \\mathbf { p } ( \\boldsymbol { y } ) ) .",
+                                    "type": "interline_equation",
+                                    "image_path": "3d3a73b561686471c4dfe86df22f349395ea59a86c528b9db4f01ce94ccf5acf.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 4,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                251,
+                                149,
+                                359,
+                                168
+                            ],
+                            "spans": [],
+                            "index": 4
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        174,
+                        505,
+                        253
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                175,
+                                506,
+                                187
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        175,
+                                        506,
+                                        187
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The most acute problem faced in this setup is the accuracy imbalance between seen and unseen classes.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                186,
+                                505,
+                                198
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        186,
+                                        505,
+                                        198
+                                    ],
+                                    "score": 1.0,
+                                    "content": "A very representative case clearly outlining the imbalance problem is presented in Table 6 of (Xian",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                196,
+                                505,
+                                210
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        196,
+                                        505,
+                                        210
+                                    ],
+                                    "score": 1.0,
+                                    "content": "et al., 2018c), where accuracy on the seen classes is always significantly greater than the accuracy on",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                209,
+                                505,
+                                220
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        209,
+                                        505,
+                                        220
+                                    ],
+                                    "score": 1.0,
+                                    "content": "unseen ones. In order to measure and control the imbalance, three metrics are commonly used to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                219,
+                                505,
+                                231
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        219,
+                                        505,
+                                        231
+                                    ],
+                                    "score": 1.0,
+                                    "content": "assess the classification performance in the GZSL scenario: the Top-1 accuracy on the seen categories",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                230,
+                                505,
+                                243
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        230,
+                                        296,
+                                        243
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(s), the Top-1 accuracy on the unseen categories",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        296,
+                                        231,
+                                        309,
+                                        241
+                                    ],
+                                    "score": 0.31,
+                                    "content": "\\mathbf { \\Pi } ^ { ( \\mathbf { u } ) }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        309,
+                                        230,
+                                        412,
+                                        243
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and their harmonic mean,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        412,
+                                        230,
+                                        483,
+                                        242
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\mathbf { H } = \\mathbf { u } \\cdot \\mathbf { s } / ( \\mathbf { u } + \\mathbf { s } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        483,
+                                        230,
+                                        505,
+                                        243
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". The",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                241,
+                                438,
+                                253
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        241,
+                                        303,
+                                        253
+                                    ],
+                                    "score": 1.0,
+                                    "content": "main metric to assess GZSL performance is then",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        303,
+                                        241,
+                                        312,
+                                        251
+                                    ],
+                                    "score": 0.58,
+                                    "content": "\\mathbf { H }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        312,
+                                        241,
+                                        403,
+                                        253
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", which quantifies both",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        404,
+                                        243,
+                                        411,
+                                        251
+                                    ],
+                                    "score": 0.44,
+                                    "content": "\\mathbf { u }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        411,
+                                        241,
+                                        438,
+                                        253
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and s.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        }
+                    ],
+                    "index": 8,
+                    "bbox_fs": [
+                        105,
+                        175,
+                        506,
+                        253
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        257,
+                        505,
+                        325
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                257,
+                                505,
+                                270
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        257,
+                                        284,
+                                        270
+                                    ],
+                                    "score": 1.0,
+                                    "content": "To formalize the problem, we first introduce",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        284,
+                                        260,
+                                        295,
+                                        269
+                                    ],
+                                    "score": 0.83,
+                                    "content": "y _ { v }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        295,
+                                        257,
+                                        412,
+                                        270
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", the true class label of image",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        413,
+                                        260,
+                                        419,
+                                        268
+                                    ],
+                                    "score": 0.68,
+                                    "content": "v",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        419,
+                                        257,
+                                        505,
+                                        270
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Mathematically, the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                266,
+                                505,
+                                283
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        266,
+                                        229,
+                                        283
+                                    ],
+                                    "score": 1.0,
+                                    "content": "main GZSL pain point is that",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        230,
+                                        269,
+                                        320,
+                                        281
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\mathbb { P } \\{ \\widehat { y } \\in \\mathcal { y } ^ { t r } | y _ { v } \\in \\mathcal { y } ^ { t s } \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        321,
+                                        266,
+                                        436,
+                                        283
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is significantly greater than",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        436,
+                                        269,
+                                        505,
+                                        281
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\mathbb { P } \\{ \\widehat { y } \\in \\mathcal { y } ^ { t s } | y _ { v } \\in",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                278,
+                                506,
+                                293
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        280,
+                                        127,
+                                        292
+                                    ],
+                                    "score": 0.9,
+                                    "content": "\\mathcal { V } ^ { t r } \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        128,
+                                        278,
+                                        506,
+                                        293
+                                    ],
+                                    "score": 1.0,
+                                    "content": "b b. In other words, the problem is that a given image is more likely to be confused with one of the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                291,
+                                505,
+                                303
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        291,
+                                        505,
+                                        303
+                                    ],
+                                    "score": 1.0,
+                                    "content": "seen classes if it belongs to an unseen class than vice versa. Our approach to solving the problem",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                300,
+                                506,
+                                315
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        300,
+                                        506,
+                                        315
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is based on the following probabilistic representation of the event space for the decision rule in",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                311,
+                                163,
+                                326
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        311,
+                                        163,
+                                        326
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Equation (4):",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        }
+                    ],
+                    "index": 14.5,
+                    "bbox_fs": [
+                        104,
+                        257,
+                        506,
+                        326
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        139,
+                        330,
+                        471,
+                        356
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                139,
+                                330,
+                                471,
+                                356
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        139,
+                                        330,
+                                        471,
+                                        356
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\mathbb { P } \\{ \\boldsymbol { { \\widehat { y } } } \\in { \\mathcal { Y } } ^ { t r } | y _ { v } \\in { \\mathcal { Y } } ^ { t s } \\} = \\mathbb { P } \\left\\{ \\operatorname* { m i n } _ { y \\in { \\mathcal { Y } } ^ { t r } } d ( \\mathbf { z } _ { v } , \\mathbf { p } ( y ) ) < \\operatorname* { m i n } _ { y \\in { \\mathcal { Y } } ^ { t s } } d ( \\mathbf { z } _ { v } , \\mathbf { p } ( y ) ) \\mid y _ { v } \\in { \\mathcal { Y } } ^ { t s } \\right\\} .",
+                                    "type": "interline_equation",
+                                    "image_path": "c4992af813698d95795db7215aceffe43193c03d6626652d1543fc22759ab44a.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 18,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                139,
+                                330,
+                                471,
+                                356
+                            ],
+                            "spans": [],
+                            "index": 18
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        362,
+                        505,
+                        385
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                362,
+                                505,
+                                375
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        362,
+                                        505,
+                                        375
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Rephrasing, the most acute GZSL error happens when the prototype of one of the seen classes is",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                374,
+                                506,
+                                387
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        374,
+                                        506,
+                                        387
+                                    ],
+                                    "score": 1.0,
+                                    "content": "closer to an image embedding from an unseen class than any of the prototypes of the unseen classes.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        }
+                    ],
+                    "index": 19.5,
+                    "bbox_fs": [
+                        106,
+                        362,
+                        506,
+                        387
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        390,
+                        505,
+                        424
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                389,
+                                504,
+                                403
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        389,
+                                        415,
+                                        403
+                                    ],
+                                    "score": 1.0,
+                                    "content": "To rectify the situation we propose the following very direct solution to balance",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        415,
+                                        390,
+                                        504,
+                                        403
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\mathbb { P } \\{ \\widehat { y } \\in \\mathcal { y } ^ { t r } | y _ { v } \\in \\mathcal { y } ^ { t s } \\}",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                401,
+                                506,
+                                414
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        401,
+                                        124,
+                                        414
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        124,
+                                        401,
+                                        219,
+                                        414
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\mathbb { P } \\{ \\hat { \\mathcal { Y } } \\in \\mathcal { Y } ^ { t s } | y _ { v } \\in \\mathcal { Y } ^ { t \\bar { r } } \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        220,
+                                        401,
+                                        355,
+                                        414
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". We introduce a positive scalar",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        356,
+                                        401,
+                                        392,
+                                        412
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\alpha \\in \\mathbb { R } ^ { + }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        392,
+                                        401,
+                                        506,
+                                        414
+                                    ],
+                                    "score": 1.0,
+                                    "content": "band scale all the distances",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                412,
+                                502,
+                                426
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        412,
+                                        268,
+                                        426
+                                    ],
+                                    "score": 1.0,
+                                    "content": "bcorresponding to the seen prototypes by",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        268,
+                                        413,
+                                        293,
+                                        423
+                                    ],
+                                    "score": 0.89,
+                                    "content": "1 + \\alpha",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        293,
+                                        412,
+                                        486,
+                                        426
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". This gives rise to the following scaled distance",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        486,
+                                        413,
+                                        498,
+                                        424
+                                    ],
+                                    "score": 0.88,
+                                    "content": "d _ { \\alpha }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        498,
+                                        412,
+                                        502,
+                                        426
+                                    ],
+                                    "score": 1.0,
+                                    "content": ":",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        }
+                    ],
+                    "index": 22,
+                    "bbox_fs": [
+                        105,
+                        389,
+                        506,
+                        426
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        198,
+                        430,
+                        414,
+                        461
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                198,
+                                430,
+                                414,
+                                461
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        198,
+                                        430,
+                                        414,
+                                        461
+                                    ],
+                                    "score": 0.92,
+                                    "content": "d _ { \\alpha } ( \\mathbf { z } _ { v } , \\mathbf { p } ( y ) ) = \\left\\{ \\begin{array} { l l } { ( 1 + \\alpha ) d ( \\mathbf { z } _ { v } , \\mathbf { p } ( y ) ) , } & { \\mathrm { i f ~ } y \\in \\mathcal { Y } ^ { t r } } \\\\ { d ( \\mathbf { z } _ { v } , \\mathbf { p } ( y ) ) , } & { \\mathrm { o t h e r w i s e } } \\end{array} \\right. .",
+                                    "type": "interline_equation",
+                                    "image_path": "a62834e649aa08ed171d61fabc0cba31952b1b0e4dc94b2e737d501dd6da9e62.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 24.5,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                198,
+                                430,
+                                414,
+                                445.5
+                            ],
+                            "spans": [],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                198,
+                                445.5,
+                                414,
+                                461.0
+                            ],
+                            "spans": [],
+                            "index": 25
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        466,
+                        484,
+                        478
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                465,
+                                486,
+                                479
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        465,
+                                        387,
+                                        479
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The misclassification between unseen as seen classes for the classifier",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        387,
+                                        466,
+                                        398,
+                                        478
+                                    ],
+                                    "score": 0.89,
+                                    "content": "\\widehat { y } _ { \\alpha }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        398,
+                                        465,
+                                        486,
+                                        479
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", based on (6) is then:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        }
+                    ],
+                    "index": 26,
+                    "bbox_fs": [
+                        105,
+                        465,
+                        486,
+                        479
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        118,
+                        483,
+                        480,
+                        510
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                118,
+                                483,
+                                480,
+                                510
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        118,
+                                        483,
+                                        480,
+                                        510
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\mathbb { P } \\{ \\widehat { y } _ { \\alpha } \\in \\mathcal { Y } ^ { t r } | y _ { v } \\in \\mathcal { Y } ^ { t s } \\} = \\mathbb { P } \\left\\{ \\left( 1 + \\alpha \\right) \\operatorname* { m i n } _ { y \\in \\mathcal { Y } ^ { t r } } d ( \\mathbf { z } _ { v } , \\mathbf { p } ( y ) ) < \\operatorname* { m i n } _ { y \\in \\mathcal { Y } ^ { t s } } d ( \\mathbf { z } _ { v } , \\mathbf { p } ( y ) ) | y _ { v } \\in \\mathcal { Y } ^ { t s } \\right\\} \\ ,",
+                                    "type": "interline_equation",
+                                    "image_path": "6c11b414afece46ca498ac50b8688e023ff6de855077e8453e68919cc2b737ba.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 27,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                118,
+                                483,
+                                480,
+                                510
+                            ],
+                            "spans": [],
+                            "index": 27
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        516,
+                        505,
+                        617
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                515,
+                                504,
+                                531
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        515,
+                                        283,
+                                        531
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and it has the following property: for any",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        283,
+                                        518,
+                                        343,
+                                        529
+                                    ],
+                                    "score": 0.81,
+                                    "content": "0 \\leq \\alpha _ { 1 } \\leq \\alpha _ { 2 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        343,
+                                        515,
+                                        347,
+                                        531
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        347,
+                                        516,
+                                        504,
+                                        530
+                                    ],
+                                    "score": 0.86,
+                                    "content": "\\mathbb { P } \\{ \\widehat { y } _ { \\alpha _ { 1 } } ~ \\in ~ \\mathcal { Y } ^ { t r } | y _ { v } ~ \\in ~ \\mathcal { Y } ^ { t s } \\} ~ \\geq ~ \\mathbb { P } \\{ \\widehat { y } _ { \\alpha _ { 2 } } ~ \\in ~",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                107,
+                                525,
+                                507,
+                                543
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        528,
+                                        168,
+                                        540
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\mathcal { V } ^ { t r } | y _ { v } \\in \\mathcal { V } ^ { t s } \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        169,
+                                        525,
+                                        190,
+                                        543
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", i.e.",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        191,
+                                        528,
+                                        289,
+                                        540
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\mathbb { P } \\{ \\widehat { y } _ { \\alpha } \\in \\mathcal { V } ^ { t \\bar { r } } | y _ { v } \\in \\mathcal { V } ^ { t s } \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        289,
+                                        525,
+                                        463,
+                                        543
+                                    ],
+                                    "score": 1.0,
+                                    "content": "b   is a monotone non-increasing function of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        464,
+                                        531,
+                                        471,
+                                        538
+                                    ],
+                                    "score": 0.79,
+                                    "content": "\\alpha",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        472,
+                                        525,
+                                        507,
+                                        543
+                                    ],
+                                    "score": 1.0,
+                                    "content": "b and we",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                538,
+                                505,
+                                552
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        538,
+                                        222,
+                                        552
+                                    ],
+                                    "score": 1.0,
+                                    "content": "bcan reduce it by increasing",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        222,
+                                        542,
+                                        230,
+                                        550
+                                    ],
+                                    "score": 0.77,
+                                    "content": "\\alpha",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        231,
+                                        538,
+                                        469,
+                                        552
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(please refer to Appendix A for a proof). Consider now",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        470,
+                                        539,
+                                        505,
+                                        551
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\mathbb { P } \\{ \\widehat { y } _ { \\alpha } \\in",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                107,
+                                549,
+                                506,
+                                563
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        550,
+                                        167,
+                                        562
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\mathcal { V } ^ { t r } | y _ { v } \\in \\mathcal { V } ^ { t r } \\big \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        168,
+                                        549,
+                                        368,
+                                        563
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", which is a probability that we classify an image",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        369,
+                                        552,
+                                        375,
+                                        560
+                                    ],
+                                    "score": 0.69,
+                                    "content": "v",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        376,
+                                        549,
+                                        506,
+                                        563
+                                    ],
+                                    "score": 1.0,
+                                    "content": "bfrom one of the seen classes as",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                560,
+                                506,
+                                575
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        560,
+                                        506,
+                                        575
+                                    ],
+                                    "score": 1.0,
+                                    "content": "still one of the seen classes. Using exactly the same chain of arguments as in Appendix A, it is",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                572,
+                                505,
+                                585
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        572,
+                                        405,
+                                        585
+                                    ],
+                                    "score": 1.0,
+                                    "content": "straightforward to show that the probability is a non-increasing function of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        405,
+                                        574,
+                                        412,
+                                        582
+                                    ],
+                                    "score": 0.75,
+                                    "content": "\\alpha",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        413,
+                                        572,
+                                        505,
+                                        585
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Hence the probability",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                581,
+                                507,
+                                597
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        582,
+                                        324,
+                                        595
+                                    ],
+                                    "score": 0.86,
+                                    "content": "\\mathbb { P } \\{ \\hat { y } _ { \\alpha } ^ { \\sim } \\in \\mathcal { V } ^ { t s } | y _ { v } \\in \\mathcal { V } ^ { t r } \\} = 1 - \\mathbb { P } \\{ \\hat { y } _ { \\alpha } \\in \\mathcal { V } ^ { \\hat { t r } } | y _ { v } \\in \\mathcal { V } ^ { t r } \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        325,
+                                        581,
+                                        450,
+                                        597
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is a non-decreasing function of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        451,
+                                        585,
+                                        458,
+                                        593
+                                    ],
+                                    "score": 0.73,
+                                    "content": "\\alpha",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        459,
+                                        581,
+                                        507,
+                                        597
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Therefore,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                595,
+                                505,
+                                606
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        595,
+                                        117,
+                                        606
+                                    ],
+                                    "score": 1.0,
+                                    "content": "as",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        117,
+                                        596,
+                                        125,
+                                        604
+                                    ],
+                                    "score": 0.79,
+                                    "content": "\\alpha",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        126,
+                                        595,
+                                        505,
+                                        606
+                                    ],
+                                    "score": 1.0,
+                                    "content": "increases, we expect more classification errors in classifying images from seen classes, because",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                604,
+                                349,
+                                618
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        604,
+                                        349,
+                                        618
+                                    ],
+                                    "score": 1.0,
+                                    "content": "some of them will be classified as one of the unseen classes.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        }
+                    ],
+                    "index": 32,
+                    "bbox_fs": [
+                        104,
+                        515,
+                        507,
+                        618
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        621,
+                        505,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                622,
+                                505,
+                                635
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        622,
+                                        401,
+                                        635
+                                    ],
+                                    "score": 1.0,
+                                    "content": "To sum up, given the arguments presented above we expect that by varying",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        402,
+                                        623,
+                                        428,
+                                        632
+                                    ],
+                                    "score": 0.89,
+                                    "content": "\\alpha > 0",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        428,
+                                        622,
+                                        505,
+                                        635
+                                    ],
+                                    "score": 1.0,
+                                    "content": "we can balance the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                632,
+                                506,
+                                646
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        632,
+                                        147,
+                                        646
+                                    ],
+                                    "score": 1.0,
+                                    "content": "error rate",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        147,
+                                        632,
+                                        243,
+                                        645
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\mathbb { P } \\mathbf { \\bar { \\{ y _ { \\alpha } \\in \\mathcal { V } ^ { t r } \\vert y _ { v } \\in \\mathcal { V } ^ { t s } \\} } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        243,
+                                        632,
+                                        506,
+                                        646
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of leaking the unseen class images into seen class classification",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                643,
+                                506,
+                                657
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        643,
+                                        218,
+                                        657
+                                    ],
+                                    "score": 1.0,
+                                    "content": "bdecision and the error rate",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        218,
+                                        644,
+                                        317,
+                                        656
+                                    ],
+                                    "score": 0.9,
+                                    "content": "\\mathbb { P } \\{ \\widehat { y } _ { \\alpha } ^ { } \\in \\mathcal { V } ^ { t s } | y _ { v } ^ { - } \\in \\mathcal { V } ^ { t r } \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        318,
+                                        643,
+                                        506,
+                                        657
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of leaking the seen class images into unseen",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                654,
+                                504,
+                                667
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        654,
+                                        409,
+                                        667
+                                    ],
+                                    "score": 1.0,
+                                    "content": "bclass classification decision. This is possible as we just showed above that",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        410,
+                                        654,
+                                        504,
+                                        667
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\mathbb { P } \\{ \\widehat { y } _ { \\alpha } \\in \\mathcal { V } ^ { t r } | y _ { v } \\in \\mathcal { V } ^ { t s } \\}",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                665,
+                                506,
+                                679
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        665,
+                                        233,
+                                        679
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is a non-increasing function of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        234,
+                                        668,
+                                        241,
+                                        676
+                                    ],
+                                    "score": 0.78,
+                                    "content": "\\alpha",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        241,
+                                        665,
+                                        270,
+                                        679
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", while",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        270,
+                                        666,
+                                        365,
+                                        678
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\mathbb { P } \\{ \\widehat { y } _ { \\alpha } \\in \\mathcal { V } ^ { t s } | y _ { v } \\in \\mathcal { V } ^ { t r } \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        365,
+                                        665,
+                                        506,
+                                        679
+                                    ],
+                                    "score": 1.0,
+                                    "content": "b is a non-decreasing one. It is also",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                677,
+                                505,
+                                689
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        677,
+                                        223,
+                                        689
+                                    ],
+                                    "score": 1.0,
+                                    "content": "important to emphasize that",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        224,
+                                        679,
+                                        231,
+                                        687
+                                    ],
+                                    "score": 0.77,
+                                    "content": "\\alpha",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        232,
+                                        677,
+                                        505,
+                                        689
+                                    ],
+                                    "score": 1.0,
+                                    "content": "bis applied only to distances between the query embedding and the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                687,
+                                505,
+                                700
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        687,
+                                        475,
+                                        700
+                                    ],
+                                    "score": 1.0,
+                                    "content": "prototypes of seen classes and it is constant over seen classes. Therefore, the application of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        475,
+                                        690,
+                                        483,
+                                        698
+                                    ],
+                                    "score": 0.77,
+                                    "content": "\\alpha",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        483,
+                                        687,
+                                        505,
+                                        700
+                                    ],
+                                    "score": 1.0,
+                                    "content": "does",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                697,
+                                506,
+                                712
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        697,
+                                        343,
+                                        712
+                                    ],
+                                    "score": 1.0,
+                                    "content": "not at all affect the classification error rates either within",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        343,
+                                        698,
+                                        359,
+                                        709
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\mathcal { V } ^ { t r }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        359,
+                                        697,
+                                        401,
+                                        712
+                                    ],
+                                    "score": 1.0,
+                                    "content": "or within",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        401,
+                                        698,
+                                        416,
+                                        710
+                                    ],
+                                    "score": 0.89,
+                                    "content": "\\mathcal { V } ^ { t s }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        417,
+                                        697,
+                                        457,
+                                        712
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Varying",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        458,
+                                        700,
+                                        466,
+                                        709
+                                    ],
+                                    "score": 0.76,
+                                    "content": "\\alpha",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        466,
+                                        697,
+                                        506,
+                                        712
+                                    ],
+                                    "score": 1.0,
+                                    "content": "balances",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                709,
+                                505,
+                                722
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        709,
+                                        505,
+                                        722
+                                    ],
+                                    "score": 1.0,
+                                    "content": "exclusively the classification errors arising from transitions between seen and unseen class labels. We",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                720,
+                                338,
+                                733
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        720,
+                                        269,
+                                        733
+                                    ],
+                                    "score": 1.0,
+                                    "content": "study the empirical aspects of balancing",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        269,
+                                        724,
+                                        276,
+                                        730
+                                    ],
+                                    "score": 0.77,
+                                    "content": "\\alpha",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        277,
+                                        720,
+                                        338,
+                                        733
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in Section 4.4.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        }
+                    ],
+                    "index": 41.5,
+                    "bbox_fs": [
+                        104,
+                        622,
+                        506,
+                        733
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "table",
+                    "bbox": [
+                        106,
+                        101,
+                        518,
+                        198
+                    ],
+                    "blocks": [
+                        {
+                            "type": "table_caption",
+                            "bbox": [
+                                182,
+                                80,
+                                427,
+                                92
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        181,
+                                        78,
+                                        428,
+                                        94
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                181,
+                                                78,
+                                                428,
+                                                94
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Table 1: Generalized zero-shot Top-1 classification accuracy.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 0
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "type": "table_body",
+                            "bbox": [
+                                106,
+                                101,
+                                518,
+                                198
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        101,
+                                        518,
+                                        198
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                101,
+                                                518,
+                                                198
+                                            ],
+                                            "score": 0.982,
+                                            "html": "<table><tr><td></td><td colspan=\"3\">CUB</td><td colspan=\"3\">FLOWERS</td></tr><tr><td></td><td>u</td><td>S</td><td>H</td><td>u</td><td>s</td><td>H</td></tr><tr><td>CADA-VAE (Schönfeld et al., 2019)</td><td>n/a</td><td>n/a</td><td>53.4</td><td>n/a</td><td>n/a</td><td>n/a</td></tr><tr><td>f-CLSWGAN (Xian et al., 2018d)</td><td>50.3</td><td>58.3</td><td>54.0</td><td>59.0</td><td>73.8</td><td>65.6</td></tr><tr><td>f-VAEGAN-D2 (Xian et al.,2019)</td><td>48.4</td><td>60.1</td><td>53.6</td><td>56.8</td><td>74.9</td><td>64.6</td></tr><tr><td>cycle-(U)WGAN (Felix et al., 2018)</td><td>47.9</td><td>59.3</td><td>53.0</td><td>61.6</td><td>69.2</td><td>65.2</td></tr><tr><td>COSMO+f-CLSWGAN (Atzmon &amp; Chechik, 2019)</td><td>n/a</td><td>n/a</td><td>n/a</td><td>59.6</td><td>81.4</td><td>68.8</td></tr><tr><td>CLAREL (Ours)</td><td>59.3</td><td>52.6</td><td>55.8</td><td>73.0</td><td>73.6</td><td>73.3</td></tr></table>",
+                                            "type": "table",
+                                            "image_path": "12bbcc390447bbfe9373f37f6f92b9541094bd9ffb0478a62b265d975ef854ee.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 2,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        101,
+                                        518,
+                                        133.33333333333334
+                                    ],
+                                    "spans": [],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        133.33333333333334,
+                                        518,
+                                        165.66666666666669
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        165.66666666666669,
+                                        518,
+                                        198.00000000000003
+                                    ],
+                                    "spans": [],
+                                    "index": 3
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 1.0
+                },
+                {
+                    "type": "table",
+                    "bbox": [
+                        173,
+                        231,
+                        438,
+                        307
+                    ],
+                    "blocks": [
+                        {
+                            "type": "table_caption",
+                            "bbox": [
+                                207,
+                                210,
+                                402,
+                                223
+                            ],
+                            "group_id": 1,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        206,
+                                        209,
+                                        404,
+                                        225
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                206,
+                                                209,
+                                                404,
+                                                225
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Table 2: Zero-shot Top-1 classification accuracy.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 4
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "type": "table_body",
+                            "bbox": [
+                                173,
+                                231,
+                                438,
+                                307
+                            ],
+                            "group_id": 1,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        173,
+                                        231,
+                                        438,
+                                        307
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                173,
+                                                231,
+                                                438,
+                                                307
+                                            ],
+                                            "score": 0.981,
+                                            "html": "<table><tr><td></td><td>CUB</td><td>FLOWERS</td></tr><tr><td>CADA-VAE (Schonfeld et al., 2019)</td><td>n/a</td><td>n/a</td></tr><tr><td>f-CLSWGAN (Xian et al., 2018d)</td><td>57.3</td><td>67.2</td></tr><tr><td>f-VAEGAN-D2 (Xian et al., 2019)</td><td>61.0</td><td>67.7</td></tr><tr><td>cycle-(U)WGAN (Felix et al., 2018)</td><td>58.6</td><td>70.3</td></tr><tr><td>CLAREL (Ours)</td><td>66.7</td><td>76.8</td></tr></table>",
+                                            "type": "table",
+                                            "image_path": "892a986457f15b9754b348d5c3429df71973148b0d2ec102398a7aeb1a84ccab.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 6,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        173,
+                                        231,
+                                        438,
+                                        256.3333333333333
+                                    ],
+                                    "spans": [],
+                                    "index": 5
+                                },
+                                {
+                                    "bbox": [
+                                        173,
+                                        256.3333333333333,
+                                        438,
+                                        281.66666666666663
+                                    ],
+                                    "spans": [],
+                                    "index": 6
+                                },
+                                {
+                                    "bbox": [
+                                        173,
+                                        281.66666666666663,
+                                        438,
+                                        306.99999999999994
+                                    ],
+                                    "spans": [],
+                                    "index": 7
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 5.0
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        328,
+                        211,
+                        341
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                327,
+                                213,
+                                343
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        327,
+                                        213,
+                                        343
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3 RELATED WORK",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        }
+                    ],
+                    "index": 8
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        354,
+                        505,
+                        607
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                354,
+                                506,
+                                369
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        354,
+                                        506,
+                                        369
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ZSL approaches aim at recognizing objects belonging to classes unseen during training (Larochelle",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                366,
+                                506,
+                                379
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        366,
+                                        506,
+                                        379
+                                    ],
+                                    "score": 1.0,
+                                    "content": "et al., 2008; Palatucci et al., 2009). This has been extended to the GZSL framework in which the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                377,
+                                505,
+                                390
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        377,
+                                        505,
+                                        390
+                                    ],
+                                    "score": 1.0,
+                                    "content": "decision space consists of both seen and unseen classes (Socher et al., 2013; Xian et al., 2018c). The",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                388,
+                                505,
+                                401
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        388,
+                                        505,
+                                        401
+                                    ],
+                                    "score": 1.0,
+                                    "content": "classical zero-shot approaches build a joint visual-semantic space, relying on a linear cross-modal",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                399,
+                                506,
+                                413
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        399,
+                                        506,
+                                        413
+                                    ],
+                                    "score": 1.0,
+                                    "content": "compatibility function (e.g. dot-product between query embedding and semantic prototypes or a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                411,
+                                506,
+                                423
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        411,
+                                        506,
+                                        423
+                                    ],
+                                    "score": 1.0,
+                                    "content": "variation of a hinge loss) (Frome et al., 2013; Akata et al., 2015; 2016; Reed et al., 2016). Non-linear",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                420,
+                                506,
+                                435
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        420,
+                                        506,
+                                        435
+                                    ],
+                                    "score": 1.0,
+                                    "content": "variants of the compatibility has also been explored (Xian et al., 2016; Socher et al., 2013). Extending",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                432,
+                                506,
+                                445
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        432,
+                                        506,
+                                        445
+                                    ],
+                                    "score": 1.0,
+                                    "content": "previously proposed cross-modal transfer approaches based on auto-encoders (Hubert Tsai et al.,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                442,
+                                507,
+                                456
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        442,
+                                        507,
+                                        456
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2017) and cross-domain learning (Gretton et al., 2007), more recent line of work (Schönfeld et al.,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                453,
+                                506,
+                                467
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        453,
+                                        506,
+                                        467
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2019; Xian et al., 2018d; 2019; Felix et al., 2018; Verma et al., 2018) relies on combining these",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                465,
+                                507,
+                                478
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        465,
+                                        507,
+                                        478
+                                    ],
+                                    "score": 1.0,
+                                    "content": "approaches and their variations with dataset augmentation tools such as GAN (Goodfellow et al.,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                476,
+                                505,
+                                488
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        476,
+                                        505,
+                                        488
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2014) and VAE (Kingma & Welling, 2014). It is argued that the use of those tools helps to resolve",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                487,
+                                505,
+                                500
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        487,
+                                        505,
+                                        500
+                                    ],
+                                    "score": 1.0,
+                                    "content": "one of the prominent problems in GZSL scenario: classifying images from unseen classes as one",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                497,
+                                506,
+                                511
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        497,
+                                        506,
+                                        511
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of the seen classes. There exist approaches that try to tackle this same problem via temperature",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                509,
+                                506,
+                                522
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        509,
+                                        506,
+                                        522
+                                    ],
+                                    "score": 1.0,
+                                    "content": "calibration (Liu et al., 2018) originally proposed by Hinton et al. (2015). Chao et al. (2016); Das &",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                520,
+                                506,
+                                533
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        520,
+                                        506,
+                                        533
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Lee (2019) proposed an approach to seen/unseen accuracy balancing that is very similar to ours, based",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                531,
+                                506,
+                                543
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        531,
+                                        506,
+                                        543
+                                    ],
+                                    "score": 1.0,
+                                    "content": "on heuristic arguments. We extend this line of work here by providing a probabilistic justification for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                541,
+                                506,
+                                555
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        541,
+                                        506,
+                                        555
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the balancing effect observed when applying metric rescaling. Atzmon & Chechik (2019) propose",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                552,
+                                505,
+                                566
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        552,
+                                        505,
+                                        566
+                                    ],
+                                    "score": 1.0,
+                                    "content": "a more sophisticated way to deal with seen/unseen imbalance via adaptive confidence smoothing",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                564,
+                                506,
+                                577
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        564,
+                                        506,
+                                        577
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and gating, yet as authors note it is much simpler to train than the existing GAN-based zero-shot",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                574,
+                                506,
+                                589
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        574,
+                                        506,
+                                        589
+                                    ],
+                                    "score": 1.0,
+                                    "content": "approaches. In this work, we introduce arguably the simplest zero-shot representation training",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                586,
+                                506,
+                                599
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        586,
+                                        506,
+                                        599
+                                    ],
+                                    "score": 1.0,
+                                    "content": "approach of all, and we demonstrate that when the image level text information is available, it",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                597,
+                                422,
+                                609
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        597,
+                                        422,
+                                        609
+                                    ],
+                                    "score": 1.0,
+                                    "content": "achieves the state-of-the-art results on GZSL task on two well-known datasets.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        }
+                    ],
+                    "index": 20
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        628,
+                        256,
+                        640
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                626,
+                                258,
+                                642
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        626,
+                                        258,
+                                        642
+                                    ],
+                                    "score": 1.0,
+                                    "content": "4 EXPERIMENTAL RESULTS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        }
+                    ],
+                    "index": 32
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        654,
+                        176,
+                        666
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                654,
+                                177,
+                                667
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        654,
+                                        177,
+                                        667
+                                    ],
+                                    "score": 1.0,
+                                    "content": "4.1 DATASETS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        }
+                    ],
+                    "index": 33
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        677,
+                        505,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                676,
+                                506,
+                                689
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        676,
+                                        506,
+                                        689
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We focus on learning embeddings for fine-grained visual descriptions and test them in ZSL/GZSL",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                687,
+                                506,
+                                702
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        687,
+                                        506,
+                                        702
+                                    ],
+                                    "score": 1.0,
+                                    "content": "scenario. To test the quality of trained embeddings we focus on datasets that provide paired images",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                699,
+                                505,
+                                711
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        699,
+                                        505,
+                                        711
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and text descriptions, such as Caltech-UCSD-Birds (CUB) (Welinder et al., 2010) and Oxford Flowers",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                709,
+                                505,
+                                722
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        709,
+                                        505,
+                                        722
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(FLOWERS) (Nilsback & Zisserman, 2008), that were augmented with textual descriptions by Reed",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                720,
+                                506,
+                                734
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        720,
+                                        506,
+                                        734
+                                    ],
+                                    "score": 1.0,
+                                    "content": "et al. (2016). We use the GZSL splits proposed by Xian et al. (2018c). The attribute-based datasets,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        }
+                    ],
+                    "index": 36
+                }
+            ],
+            "page_idx": 4,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        107,
+                        27,
+                        308,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                107,
+                                26,
+                                308,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        26,
+                                        308,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Under review as a conference paper at ICLR 2020",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        302,
+                        751,
+                        308,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                302,
+                                750,
+                                309,
+                                763
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        302,
+                                        750,
+                                        309,
+                                        763
+                                    ],
+                                    "score": 1.0,
+                                    "content": "5",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "table",
+                    "bbox": [
+                        106,
+                        101,
+                        518,
+                        198
+                    ],
+                    "blocks": [
+                        {
+                            "type": "table_caption",
+                            "bbox": [
+                                182,
+                                80,
+                                427,
+                                92
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        181,
+                                        78,
+                                        428,
+                                        94
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                181,
+                                                78,
+                                                428,
+                                                94
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Table 1: Generalized zero-shot Top-1 classification accuracy.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 0
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "type": "table_body",
+                            "bbox": [
+                                106,
+                                101,
+                                518,
+                                198
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        101,
+                                        518,
+                                        198
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                101,
+                                                518,
+                                                198
+                                            ],
+                                            "score": 0.982,
+                                            "html": "<table><tr><td></td><td colspan=\"3\">CUB</td><td colspan=\"3\">FLOWERS</td></tr><tr><td></td><td>u</td><td>S</td><td>H</td><td>u</td><td>s</td><td>H</td></tr><tr><td>CADA-VAE (Schönfeld et al., 2019)</td><td>n/a</td><td>n/a</td><td>53.4</td><td>n/a</td><td>n/a</td><td>n/a</td></tr><tr><td>f-CLSWGAN (Xian et al., 2018d)</td><td>50.3</td><td>58.3</td><td>54.0</td><td>59.0</td><td>73.8</td><td>65.6</td></tr><tr><td>f-VAEGAN-D2 (Xian et al.,2019)</td><td>48.4</td><td>60.1</td><td>53.6</td><td>56.8</td><td>74.9</td><td>64.6</td></tr><tr><td>cycle-(U)WGAN (Felix et al., 2018)</td><td>47.9</td><td>59.3</td><td>53.0</td><td>61.6</td><td>69.2</td><td>65.2</td></tr><tr><td>COSMO+f-CLSWGAN (Atzmon &amp; Chechik, 2019)</td><td>n/a</td><td>n/a</td><td>n/a</td><td>59.6</td><td>81.4</td><td>68.8</td></tr><tr><td>CLAREL (Ours)</td><td>59.3</td><td>52.6</td><td>55.8</td><td>73.0</td><td>73.6</td><td>73.3</td></tr></table>",
+                                            "type": "table",
+                                            "image_path": "12bbcc390447bbfe9373f37f6f92b9541094bd9ffb0478a62b265d975ef854ee.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 2,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        101,
+                                        518,
+                                        133.33333333333334
+                                    ],
+                                    "spans": [],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        133.33333333333334,
+                                        518,
+                                        165.66666666666669
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        165.66666666666669,
+                                        518,
+                                        198.00000000000003
+                                    ],
+                                    "spans": [],
+                                    "index": 3
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 1.0
+                },
+                {
+                    "type": "table",
+                    "bbox": [
+                        173,
+                        231,
+                        438,
+                        307
+                    ],
+                    "blocks": [
+                        {
+                            "type": "table_caption",
+                            "bbox": [
+                                207,
+                                210,
+                                402,
+                                223
+                            ],
+                            "group_id": 1,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        206,
+                                        209,
+                                        404,
+                                        225
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                206,
+                                                209,
+                                                404,
+                                                225
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Table 2: Zero-shot Top-1 classification accuracy.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 4
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "type": "table_body",
+                            "bbox": [
+                                173,
+                                231,
+                                438,
+                                307
+                            ],
+                            "group_id": 1,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        173,
+                                        231,
+                                        438,
+                                        307
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                173,
+                                                231,
+                                                438,
+                                                307
+                                            ],
+                                            "score": 0.981,
+                                            "html": "<table><tr><td></td><td>CUB</td><td>FLOWERS</td></tr><tr><td>CADA-VAE (Schonfeld et al., 2019)</td><td>n/a</td><td>n/a</td></tr><tr><td>f-CLSWGAN (Xian et al., 2018d)</td><td>57.3</td><td>67.2</td></tr><tr><td>f-VAEGAN-D2 (Xian et al., 2019)</td><td>61.0</td><td>67.7</td></tr><tr><td>cycle-(U)WGAN (Felix et al., 2018)</td><td>58.6</td><td>70.3</td></tr><tr><td>CLAREL (Ours)</td><td>66.7</td><td>76.8</td></tr></table>",
+                                            "type": "table",
+                                            "image_path": "892a986457f15b9754b348d5c3429df71973148b0d2ec102398a7aeb1a84ccab.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 6,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        173,
+                                        231,
+                                        438,
+                                        256.3333333333333
+                                    ],
+                                    "spans": [],
+                                    "index": 5
+                                },
+                                {
+                                    "bbox": [
+                                        173,
+                                        256.3333333333333,
+                                        438,
+                                        281.66666666666663
+                                    ],
+                                    "spans": [],
+                                    "index": 6
+                                },
+                                {
+                                    "bbox": [
+                                        173,
+                                        281.66666666666663,
+                                        438,
+                                        306.99999999999994
+                                    ],
+                                    "spans": [],
+                                    "index": 7
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 5.0
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        328,
+                        211,
+                        341
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                327,
+                                213,
+                                343
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        327,
+                                        213,
+                                        343
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3 RELATED WORK",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        }
+                    ],
+                    "index": 8
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        354,
+                        505,
+                        607
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                354,
+                                506,
+                                369
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        354,
+                                        506,
+                                        369
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ZSL approaches aim at recognizing objects belonging to classes unseen during training (Larochelle",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                366,
+                                506,
+                                379
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        366,
+                                        506,
+                                        379
+                                    ],
+                                    "score": 1.0,
+                                    "content": "et al., 2008; Palatucci et al., 2009). This has been extended to the GZSL framework in which the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                377,
+                                505,
+                                390
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        377,
+                                        505,
+                                        390
+                                    ],
+                                    "score": 1.0,
+                                    "content": "decision space consists of both seen and unseen classes (Socher et al., 2013; Xian et al., 2018c). The",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                388,
+                                505,
+                                401
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        388,
+                                        505,
+                                        401
+                                    ],
+                                    "score": 1.0,
+                                    "content": "classical zero-shot approaches build a joint visual-semantic space, relying on a linear cross-modal",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                399,
+                                506,
+                                413
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        399,
+                                        506,
+                                        413
+                                    ],
+                                    "score": 1.0,
+                                    "content": "compatibility function (e.g. dot-product between query embedding and semantic prototypes or a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                411,
+                                506,
+                                423
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        411,
+                                        506,
+                                        423
+                                    ],
+                                    "score": 1.0,
+                                    "content": "variation of a hinge loss) (Frome et al., 2013; Akata et al., 2015; 2016; Reed et al., 2016). Non-linear",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                420,
+                                506,
+                                435
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        420,
+                                        506,
+                                        435
+                                    ],
+                                    "score": 1.0,
+                                    "content": "variants of the compatibility has also been explored (Xian et al., 2016; Socher et al., 2013). Extending",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                432,
+                                506,
+                                445
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        432,
+                                        506,
+                                        445
+                                    ],
+                                    "score": 1.0,
+                                    "content": "previously proposed cross-modal transfer approaches based on auto-encoders (Hubert Tsai et al.,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                442,
+                                507,
+                                456
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        442,
+                                        507,
+                                        456
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2017) and cross-domain learning (Gretton et al., 2007), more recent line of work (Schönfeld et al.,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                453,
+                                506,
+                                467
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        453,
+                                        506,
+                                        467
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2019; Xian et al., 2018d; 2019; Felix et al., 2018; Verma et al., 2018) relies on combining these",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                465,
+                                507,
+                                478
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        465,
+                                        507,
+                                        478
+                                    ],
+                                    "score": 1.0,
+                                    "content": "approaches and their variations with dataset augmentation tools such as GAN (Goodfellow et al.,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                476,
+                                505,
+                                488
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        476,
+                                        505,
+                                        488
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2014) and VAE (Kingma & Welling, 2014). It is argued that the use of those tools helps to resolve",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                487,
+                                505,
+                                500
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        487,
+                                        505,
+                                        500
+                                    ],
+                                    "score": 1.0,
+                                    "content": "one of the prominent problems in GZSL scenario: classifying images from unseen classes as one",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                497,
+                                506,
+                                511
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        497,
+                                        506,
+                                        511
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of the seen classes. There exist approaches that try to tackle this same problem via temperature",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                509,
+                                506,
+                                522
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        509,
+                                        506,
+                                        522
+                                    ],
+                                    "score": 1.0,
+                                    "content": "calibration (Liu et al., 2018) originally proposed by Hinton et al. (2015). Chao et al. (2016); Das &",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                520,
+                                506,
+                                533
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        520,
+                                        506,
+                                        533
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Lee (2019) proposed an approach to seen/unseen accuracy balancing that is very similar to ours, based",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                531,
+                                506,
+                                543
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        531,
+                                        506,
+                                        543
+                                    ],
+                                    "score": 1.0,
+                                    "content": "on heuristic arguments. We extend this line of work here by providing a probabilistic justification for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                541,
+                                506,
+                                555
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        541,
+                                        506,
+                                        555
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the balancing effect observed when applying metric rescaling. Atzmon & Chechik (2019) propose",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                552,
+                                505,
+                                566
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        552,
+                                        505,
+                                        566
+                                    ],
+                                    "score": 1.0,
+                                    "content": "a more sophisticated way to deal with seen/unseen imbalance via adaptive confidence smoothing",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                564,
+                                506,
+                                577
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        564,
+                                        506,
+                                        577
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and gating, yet as authors note it is much simpler to train than the existing GAN-based zero-shot",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                574,
+                                506,
+                                589
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        574,
+                                        506,
+                                        589
+                                    ],
+                                    "score": 1.0,
+                                    "content": "approaches. In this work, we introduce arguably the simplest zero-shot representation training",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                586,
+                                506,
+                                599
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        586,
+                                        506,
+                                        599
+                                    ],
+                                    "score": 1.0,
+                                    "content": "approach of all, and we demonstrate that when the image level text information is available, it",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                597,
+                                422,
+                                609
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        597,
+                                        422,
+                                        609
+                                    ],
+                                    "score": 1.0,
+                                    "content": "achieves the state-of-the-art results on GZSL task on two well-known datasets.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        }
+                    ],
+                    "index": 20,
+                    "bbox_fs": [
+                        104,
+                        354,
+                        507,
+                        609
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        628,
+                        256,
+                        640
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                626,
+                                258,
+                                642
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        626,
+                                        258,
+                                        642
+                                    ],
+                                    "score": 1.0,
+                                    "content": "4 EXPERIMENTAL RESULTS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        }
+                    ],
+                    "index": 32
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        654,
+                        176,
+                        666
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                654,
+                                177,
+                                667
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        654,
+                                        177,
+                                        667
+                                    ],
+                                    "score": 1.0,
+                                    "content": "4.1 DATASETS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        }
+                    ],
+                    "index": 33
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        677,
+                        505,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                676,
+                                506,
+                                689
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        676,
+                                        506,
+                                        689
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We focus on learning embeddings for fine-grained visual descriptions and test them in ZSL/GZSL",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                687,
+                                506,
+                                702
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        687,
+                                        506,
+                                        702
+                                    ],
+                                    "score": 1.0,
+                                    "content": "scenario. To test the quality of trained embeddings we focus on datasets that provide paired images",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                699,
+                                505,
+                                711
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        699,
+                                        505,
+                                        711
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and text descriptions, such as Caltech-UCSD-Birds (CUB) (Welinder et al., 2010) and Oxford Flowers",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                709,
+                                505,
+                                722
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        709,
+                                        505,
+                                        722
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(FLOWERS) (Nilsback & Zisserman, 2008), that were augmented with textual descriptions by Reed",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                720,
+                                506,
+                                734
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        720,
+                                        506,
+                                        734
+                                    ],
+                                    "score": 1.0,
+                                    "content": "et al. (2016). We use the GZSL splits proposed by Xian et al. (2018c). The attribute-based datasets,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                255,
+                                505,
+                                266
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        255,
+                                        505,
+                                        266
+                                    ],
+                                    "score": 1.0,
+                                    "content": "such as SUN (Patterson et al., 2014) and AWA (Lampert et al., 2014) do not contain this information",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                264,
+                                505,
+                                279
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        264,
+                                        505,
+                                        279
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and do not have a notion of entity of a class in them. They are out of the scope of the current paper.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 7
+                        }
+                    ],
+                    "index": 36,
+                    "bbox_fs": [
+                        105,
+                        676,
+                        506,
+                        734
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "image",
+                    "bbox": [
+                        135,
+                        81,
+                        475,
+                        182
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                135,
+                                81,
+                                475,
+                                182
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        135,
+                                        81,
+                                        475,
+                                        182
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                135,
+                                                81,
+                                                475,
+                                                182
+                                            ],
+                                            "score": 0.968,
+                                            "type": "image",
+                                            "image_path": "6c68eb4c7e42a907d13836d18bd23fbf8d9cba1c7819eebd0cbae1133409eec9.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 1,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        135,
+                                        81,
+                                        475,
+                                        114.66666666666666
+                                    ],
+                                    "spans": [],
+                                    "index": 0
+                                },
+                                {
+                                    "bbox": [
+                                        135,
+                                        114.66666666666666,
+                                        475,
+                                        148.33333333333331
+                                    ],
+                                    "spans": [],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        135,
+                                        148.33333333333331,
+                                        475,
+                                        181.99999999999997
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                107,
+                                191,
+                                504,
+                                225
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        191,
+                                        505,
+                                        205
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                191,
+                                                369,
+                                                205
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 2: Harmonic mean Top-1 accuracy on seen and unseen,",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                370,
+                                                192,
+                                                379,
+                                                202
+                                            ],
+                                            "score": 0.26,
+                                            "content": "\\mathbf { H }",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                380,
+                                                191,
+                                                467,
+                                                205
+                                            ],
+                                            "score": 1.0,
+                                            "content": ", against the value of",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                468,
+                                                194,
+                                                475,
+                                                201
+                                            ],
+                                            "score": 0.77,
+                                            "content": "\\alpha",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                475,
+                                                191,
+                                                505,
+                                                205
+                                            ],
+                                            "score": 1.0,
+                                            "content": "on the",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 3
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        203,
+                                        506,
+                                        215
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                203,
+                                                312,
+                                                215
+                                            ],
+                                            "score": 1.0,
+                                            "content": "validation set. The curves represent the mean and",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                312,
+                                                203,
+                                                332,
+                                                213
+                                            ],
+                                            "score": 0.87,
+                                            "content": "9 5 \\%",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                333,
+                                                203,
+                                                506,
+                                                215
+                                            ],
+                                            "score": 1.0,
+                                            "content": "confidence intervals over 10 optimization",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 4
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        213,
+                                        468,
+                                        226
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                213,
+                                                280,
+                                                226
+                                            ],
+                                            "score": 1.0,
+                                            "content": "runs. Results are stable over different runs.",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                281,
+                                                214,
+                                                290,
+                                                224
+                                            ],
+                                            "score": 0.27,
+                                            "content": "\\mathbf { H }",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                291,
+                                                213,
+                                                456,
+                                                226
+                                            ],
+                                            "score": 1.0,
+                                            "content": "exhibits a distinct inverted U-shape w.r.t.",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                456,
+                                                216,
+                                                464,
+                                                223
+                                            ],
+                                            "score": 0.64,
+                                            "content": "\\alpha",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                464,
+                                                213,
+                                                468,
+                                                226
+                                            ],
+                                            "score": 1.0,
+                                            "content": ".",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 5
+                                }
+                            ],
+                            "index": 4
+                        }
+                    ],
+                    "index": 2.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        254,
+                        504,
+                        277
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                255,
+                                505,
+                                266
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        255,
+                                        505,
+                                        266
+                                    ],
+                                    "score": 1.0,
+                                    "content": "such as SUN (Patterson et al., 2014) and AWA (Lampert et al., 2014) do not contain this information",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                264,
+                                505,
+                                279
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        264,
+                                        505,
+                                        279
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and do not have a notion of entity of a class in them. They are out of the scope of the current paper.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        }
+                    ],
+                    "index": 6.5
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        109,
+                        299,
+                        306,
+                        310
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                298,
+                                308,
+                                312
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        298,
+                                        308,
+                                        312
+                                    ],
+                                    "score": 1.0,
+                                    "content": "4.2 ARCHITECTURE AND TRAINING DETAILS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        }
+                    ],
+                    "index": 8
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        323,
+                        505,
+                        509
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                322,
+                                506,
+                                336
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        322,
+                                        506,
+                                        336
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Below, we provide more detailed description of parameters used to build and train the architecture",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                334,
+                                506,
+                                347
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        334,
+                                        506,
+                                        347
+                                    ],
+                                    "score": 1.0,
+                                    "content": "depicted in Figure 1. We use exactly the same hyperparameter settings for CUB and FLOWERS.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                345,
+                                504,
+                                357
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        345,
+                                        504,
+                                        357
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The text feature extractor is built by cascading two ResNet blocks, followed by a BiLSTM. Each",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                356,
+                                506,
+                                369
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        356,
+                                        506,
+                                        369
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ResNet block has 3 convolutional/batch norm layers. The number of filters in the ResNet blocks is",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                367,
+                                505,
+                                379
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        367,
+                                        505,
+                                        379
+                                    ],
+                                    "score": 1.0,
+                                    "content": "128 and 256, BiLSTM has 512 filters for forward and backward branches (1024 total). All variables",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                378,
+                                505,
+                                391
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        378,
+                                        397,
+                                        391
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in the convolutional stack (including the batch normalization parameters",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        398,
+                                        380,
+                                        406,
+                                        389
+                                    ],
+                                    "score": 0.82,
+                                    "content": "\\gamma",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        406,
+                                        378,
+                                        423,
+                                        391
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        423,
+                                        379,
+                                        431,
+                                        389
+                                    ],
+                                    "score": 0.8,
+                                    "content": "\\beta",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        431,
+                                        378,
+                                        505,
+                                        391
+                                    ],
+                                    "score": 1.0,
+                                    "content": ") are L2-penalized",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                388,
+                                505,
+                                402
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        388,
+                                        505,
+                                        402
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with weight 0.001. The image feature extractor is a ResNet-101 with fixed weights pretrained on",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                399,
+                                505,
+                                413
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        399,
+                                        505,
+                                        413
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the split of ImageNet proposed by Xian et al. (2018c). In this work we use precomputed image",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                409,
+                                505,
+                                425
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        409,
+                                        505,
+                                        425
+                                    ],
+                                    "score": 1.0,
+                                    "content": "features, available in (Xian et al., 2018a) for CUB and in (Xian et al., 2018b) for FLOWERS. Image",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                421,
+                                506,
+                                434
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        421,
+                                        506,
+                                        434
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and text features are projected in the common embedding space of size 1024 with FC layers with no",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                433,
+                                505,
+                                445
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        433,
+                                        505,
+                                        445
+                                    ],
+                                    "score": 1.0,
+                                    "content": "non-linearity. They are preceded with a dropout of 0.25. The trainable components of the model are",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                443,
+                                505,
+                                456
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        443,
+                                        152,
+                                        456
+                                    ],
+                                    "score": 1.0,
+                                    "content": "trained for",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        153,
+                                        444,
+                                        174,
+                                        454
+                                    ],
+                                    "score": 0.41,
+                                    "content": "1 5 0 \\mathrm { k }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        174,
+                                        443,
+                                        505,
+                                        456
+                                    ],
+                                    "score": 1.0,
+                                    "content": "batches of size 32 using SGD with initial learning rate of 0.1 that is annealed by",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                455,
+                                506,
+                                468
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        455,
+                                        187,
+                                        468
+                                    ],
+                                    "score": 1.0,
+                                    "content": "a factor of 10 every",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        187,
+                                        455,
+                                        204,
+                                        465
+                                    ],
+                                    "score": 0.45,
+                                    "content": "5 0 \\mathrm { k }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        205,
+                                        455,
+                                        506,
+                                        468
+                                    ],
+                                    "score": 1.0,
+                                    "content": "batches. For each batch, we sample 32 instances, each instance includes a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                465,
+                                505,
+                                480
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        465,
+                                        505,
+                                        480
+                                    ],
+                                    "score": 1.0,
+                                    "content": "vector of precomputed ResNet-101 features and 10 text descriptions corresponding to it, according",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                477,
+                                505,
+                                490
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        477,
+                                        505,
+                                        490
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to the original dataset definition Reed et al. (2016). All 10 text descriptions are processed via the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                486,
+                                506,
+                                502
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        486,
+                                        506,
+                                        502
+                                    ],
+                                    "score": 1.0,
+                                    "content": "CNN/LSTM stack and the resulting embeddings are average pooled to create a vector representation",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                498,
+                                170,
+                                511
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        498,
+                                        170,
+                                        511
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of length 1024.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        }
+                    ],
+                    "index": 17
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        532,
+                        192,
+                        543
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                530,
+                                194,
+                                545
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        530,
+                                        194,
+                                        545
+                                    ],
+                                    "score": 1.0,
+                                    "content": "4.3 KEY RESULTS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        }
+                    ],
+                    "index": 26
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        556,
+                        505,
+                        731
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                556,
+                                505,
+                                568
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        556,
+                                        505,
+                                        568
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Our key empirical results are compared in Table 1 and in Table 2 against the latest state of the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                567,
+                                505,
+                                579
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        567,
+                                        291,
+                                        579
+                                    ],
+                                    "score": 1.0,
+                                    "content": "art. Our results are based on the settings of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        291,
+                                        567,
+                                        327,
+                                        578
+                                    ],
+                                    "score": 0.89,
+                                    "content": "\\lambda = 0 . 5",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        327,
+                                        567,
+                                        331,
+                                        579
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        331,
+                                        567,
+                                        367,
+                                        578
+                                    ],
+                                    "score": 0.87,
+                                    "content": "\\kappa = 0 . 5",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        368,
+                                        567,
+                                        387,
+                                        579
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        387,
+                                        569,
+                                        396,
+                                        577
+                                    ],
+                                    "score": 0.77,
+                                    "content": "\\alpha",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        396,
+                                        567,
+                                        505,
+                                        579
+                                    ],
+                                    "score": 1.0,
+                                    "content": "selected on the validation",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                578,
+                                505,
+                                590
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        578,
+                                        505,
+                                        590
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sets of CUB and FLOWERS datasets. Please refer to Section 4.5 for the analysis of stability with",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                589,
+                                506,
+                                601
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        589,
+                                        209,
+                                        601
+                                    ],
+                                    "score": 1.0,
+                                    "content": "respect to the choices of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        209,
+                                        590,
+                                        217,
+                                        599
+                                    ],
+                                    "score": 0.77,
+                                    "content": "\\lambda",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        217,
+                                        589,
+                                        235,
+                                        601
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        236,
+                                        591,
+                                        243,
+                                        599
+                                    ],
+                                    "score": 0.76,
+                                    "content": "\\kappa",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        243,
+                                        589,
+                                        495,
+                                        601
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and Sections 2.2 and 4.4 for more details on the selection of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        496,
+                                        591,
+                                        503,
+                                        599
+                                    ],
+                                    "score": 0.73,
+                                    "content": "\\alpha",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        503,
+                                        589,
+                                        506,
+                                        601
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                599,
+                                505,
+                                613
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        599,
+                                        505,
+                                        613
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The combination of the proposed training method and the rebalancing of the metric space results",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                611,
+                                505,
+                                623
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        611,
+                                        505,
+                                        623
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in the state-of-the-art performance. Most of the current methods rely on the dataset augmentation",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                621,
+                                507,
+                                635
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        621,
+                                        507,
+                                        635
+                                    ],
+                                    "score": 1.0,
+                                    "content": "techniques based on GANs, VAEs or combinations thereof. Those are clearly complementary w.r.t.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                632,
+                                506,
+                                646
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        632,
+                                        506,
+                                        646
+                                    ],
+                                    "score": 1.0,
+                                    "content": "our method and their addition to the training procedure is likely to further boost the performance",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                643,
+                                506,
+                                657
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        643,
+                                        506,
+                                        657
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of our proposed approach. However, this is outside of the scope of the current work. Moreover,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                654,
+                                506,
+                                668
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        654,
+                                        506,
+                                        668
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the proposed method is state-of-the-art on FLOWERS even when compared against (Atzmon &",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                665,
+                                506,
+                                679
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        665,
+                                        506,
+                                        679
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Chechik, 2019) that uses both more sophisticated GAN based embedding learning approach and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                678,
+                                505,
+                                689
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        678,
+                                        505,
+                                        689
+                                    ],
+                                    "score": 1.0,
+                                    "content": "a more sophisticated seen/unseen error rate balancing based on COSMO. It is important to note",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                686,
+                                506,
+                                701
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        686,
+                                        506,
+                                        701
+                                    ],
+                                    "score": 1.0,
+                                    "content": "that Atzmon & Chechik (2019) did not report the sentence level results on CUB. Yet, when applied",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                699,
+                                505,
+                                711
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        699,
+                                        429,
+                                        711
+                                    ],
+                                    "score": 1.0,
+                                    "content": "on attributes together with f-CLSWGAN (Xian et al., 2018d) COSMO resulted in",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        430,
+                                        699,
+                                        452,
+                                        709
+                                    ],
+                                    "score": 0.86,
+                                    "content": "0 . 8 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        452,
+                                        699,
+                                        505,
+                                        711
+                                    ],
+                                    "score": 1.0,
+                                    "content": "performance",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                709,
+                                506,
+                                722
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        709,
+                                        405,
+                                        722
+                                    ],
+                                    "score": 1.0,
+                                    "content": "drop and when applied with LAGO (Atzmon & Chechik, 2018) it achieved",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        405,
+                                        710,
+                                        429,
+                                        720
+                                    ],
+                                    "score": 0.84,
+                                    "content": "0 . 5 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        430,
+                                        709,
+                                        506,
+                                        722
+                                    ],
+                                    "score": 1.0,
+                                    "content": "improvement over",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                720,
+                                243,
+                                732
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        720,
+                                        243,
+                                        732
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the attribute based state of the art.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        }
+                    ],
+                    "index": 34.5
+                }
+            ],
+            "page_idx": 5,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        106,
+                        27,
+                        308,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                107,
+                                25,
+                                308,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        25,
+                                        308,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Under review as a conference paper at ICLR 2020",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        302,
+                        752,
+                        308,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                302,
+                                751,
+                                309,
+                                762
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        302,
+                                        751,
+                                        309,
+                                        762
+                                    ],
+                                    "score": 1.0,
+                                    "content": "6",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "image",
+                    "bbox": [
+                        135,
+                        81,
+                        475,
+                        182
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                135,
+                                81,
+                                475,
+                                182
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        135,
+                                        81,
+                                        475,
+                                        182
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                135,
+                                                81,
+                                                475,
+                                                182
+                                            ],
+                                            "score": 0.968,
+                                            "type": "image",
+                                            "image_path": "6c68eb4c7e42a907d13836d18bd23fbf8d9cba1c7819eebd0cbae1133409eec9.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 1,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        135,
+                                        81,
+                                        475,
+                                        114.66666666666666
+                                    ],
+                                    "spans": [],
+                                    "index": 0
+                                },
+                                {
+                                    "bbox": [
+                                        135,
+                                        114.66666666666666,
+                                        475,
+                                        148.33333333333331
+                                    ],
+                                    "spans": [],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        135,
+                                        148.33333333333331,
+                                        475,
+                                        181.99999999999997
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                107,
+                                191,
+                                504,
+                                225
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        191,
+                                        505,
+                                        205
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                191,
+                                                369,
+                                                205
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 2: Harmonic mean Top-1 accuracy on seen and unseen,",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                370,
+                                                192,
+                                                379,
+                                                202
+                                            ],
+                                            "score": 0.26,
+                                            "content": "\\mathbf { H }",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                380,
+                                                191,
+                                                467,
+                                                205
+                                            ],
+                                            "score": 1.0,
+                                            "content": ", against the value of",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                468,
+                                                194,
+                                                475,
+                                                201
+                                            ],
+                                            "score": 0.77,
+                                            "content": "\\alpha",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                475,
+                                                191,
+                                                505,
+                                                205
+                                            ],
+                                            "score": 1.0,
+                                            "content": "on the",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 3
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        203,
+                                        506,
+                                        215
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                203,
+                                                312,
+                                                215
+                                            ],
+                                            "score": 1.0,
+                                            "content": "validation set. The curves represent the mean and",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                312,
+                                                203,
+                                                332,
+                                                213
+                                            ],
+                                            "score": 0.87,
+                                            "content": "9 5 \\%",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                333,
+                                                203,
+                                                506,
+                                                215
+                                            ],
+                                            "score": 1.0,
+                                            "content": "confidence intervals over 10 optimization",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 4
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        213,
+                                        468,
+                                        226
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                213,
+                                                280,
+                                                226
+                                            ],
+                                            "score": 1.0,
+                                            "content": "runs. Results are stable over different runs.",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                281,
+                                                214,
+                                                290,
+                                                224
+                                            ],
+                                            "score": 0.27,
+                                            "content": "\\mathbf { H }",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                291,
+                                                213,
+                                                456,
+                                                226
+                                            ],
+                                            "score": 1.0,
+                                            "content": "exhibits a distinct inverted U-shape w.r.t.",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                456,
+                                                216,
+                                                464,
+                                                223
+                                            ],
+                                            "score": 0.64,
+                                            "content": "\\alpha",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                464,
+                                                213,
+                                                468,
+                                                226
+                                            ],
+                                            "score": 1.0,
+                                            "content": ".",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 5
+                                }
+                            ],
+                            "index": 4
+                        }
+                    ],
+                    "index": 2.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        254,
+                        504,
+                        277
+                    ],
+                    "lines": [],
+                    "index": 6.5,
+                    "bbox_fs": [
+                        106,
+                        255,
+                        505,
+                        279
+                    ],
+                    "lines_deleted": true
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        109,
+                        299,
+                        306,
+                        310
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                298,
+                                308,
+                                312
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        298,
+                                        308,
+                                        312
+                                    ],
+                                    "score": 1.0,
+                                    "content": "4.2 ARCHITECTURE AND TRAINING DETAILS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        }
+                    ],
+                    "index": 8
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        323,
+                        505,
+                        509
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                322,
+                                506,
+                                336
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        322,
+                                        506,
+                                        336
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Below, we provide more detailed description of parameters used to build and train the architecture",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                334,
+                                506,
+                                347
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        334,
+                                        506,
+                                        347
+                                    ],
+                                    "score": 1.0,
+                                    "content": "depicted in Figure 1. We use exactly the same hyperparameter settings for CUB and FLOWERS.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                345,
+                                504,
+                                357
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        345,
+                                        504,
+                                        357
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The text feature extractor is built by cascading two ResNet blocks, followed by a BiLSTM. Each",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                356,
+                                506,
+                                369
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        356,
+                                        506,
+                                        369
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ResNet block has 3 convolutional/batch norm layers. The number of filters in the ResNet blocks is",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                367,
+                                505,
+                                379
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        367,
+                                        505,
+                                        379
+                                    ],
+                                    "score": 1.0,
+                                    "content": "128 and 256, BiLSTM has 512 filters for forward and backward branches (1024 total). All variables",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                378,
+                                505,
+                                391
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        378,
+                                        397,
+                                        391
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in the convolutional stack (including the batch normalization parameters",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        398,
+                                        380,
+                                        406,
+                                        389
+                                    ],
+                                    "score": 0.82,
+                                    "content": "\\gamma",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        406,
+                                        378,
+                                        423,
+                                        391
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        423,
+                                        379,
+                                        431,
+                                        389
+                                    ],
+                                    "score": 0.8,
+                                    "content": "\\beta",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        431,
+                                        378,
+                                        505,
+                                        391
+                                    ],
+                                    "score": 1.0,
+                                    "content": ") are L2-penalized",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                388,
+                                505,
+                                402
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        388,
+                                        505,
+                                        402
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with weight 0.001. The image feature extractor is a ResNet-101 with fixed weights pretrained on",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                399,
+                                505,
+                                413
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        399,
+                                        505,
+                                        413
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the split of ImageNet proposed by Xian et al. (2018c). In this work we use precomputed image",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                409,
+                                505,
+                                425
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        409,
+                                        505,
+                                        425
+                                    ],
+                                    "score": 1.0,
+                                    "content": "features, available in (Xian et al., 2018a) for CUB and in (Xian et al., 2018b) for FLOWERS. Image",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                421,
+                                506,
+                                434
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        421,
+                                        506,
+                                        434
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and text features are projected in the common embedding space of size 1024 with FC layers with no",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                433,
+                                505,
+                                445
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        433,
+                                        505,
+                                        445
+                                    ],
+                                    "score": 1.0,
+                                    "content": "non-linearity. They are preceded with a dropout of 0.25. The trainable components of the model are",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                443,
+                                505,
+                                456
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        443,
+                                        152,
+                                        456
+                                    ],
+                                    "score": 1.0,
+                                    "content": "trained for",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        153,
+                                        444,
+                                        174,
+                                        454
+                                    ],
+                                    "score": 0.41,
+                                    "content": "1 5 0 \\mathrm { k }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        174,
+                                        443,
+                                        505,
+                                        456
+                                    ],
+                                    "score": 1.0,
+                                    "content": "batches of size 32 using SGD with initial learning rate of 0.1 that is annealed by",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                455,
+                                506,
+                                468
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        455,
+                                        187,
+                                        468
+                                    ],
+                                    "score": 1.0,
+                                    "content": "a factor of 10 every",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        187,
+                                        455,
+                                        204,
+                                        465
+                                    ],
+                                    "score": 0.45,
+                                    "content": "5 0 \\mathrm { k }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        205,
+                                        455,
+                                        506,
+                                        468
+                                    ],
+                                    "score": 1.0,
+                                    "content": "batches. For each batch, we sample 32 instances, each instance includes a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                465,
+                                505,
+                                480
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        465,
+                                        505,
+                                        480
+                                    ],
+                                    "score": 1.0,
+                                    "content": "vector of precomputed ResNet-101 features and 10 text descriptions corresponding to it, according",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                477,
+                                505,
+                                490
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        477,
+                                        505,
+                                        490
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to the original dataset definition Reed et al. (2016). All 10 text descriptions are processed via the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                486,
+                                506,
+                                502
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        486,
+                                        506,
+                                        502
+                                    ],
+                                    "score": 1.0,
+                                    "content": "CNN/LSTM stack and the resulting embeddings are average pooled to create a vector representation",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                498,
+                                170,
+                                511
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        498,
+                                        170,
+                                        511
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of length 1024.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        }
+                    ],
+                    "index": 17,
+                    "bbox_fs": [
+                        105,
+                        322,
+                        506,
+                        511
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        532,
+                        192,
+                        543
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                530,
+                                194,
+                                545
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        530,
+                                        194,
+                                        545
+                                    ],
+                                    "score": 1.0,
+                                    "content": "4.3 KEY RESULTS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        }
+                    ],
+                    "index": 26
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        556,
+                        505,
+                        731
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                556,
+                                505,
+                                568
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        556,
+                                        505,
+                                        568
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Our key empirical results are compared in Table 1 and in Table 2 against the latest state of the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                567,
+                                505,
+                                579
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        567,
+                                        291,
+                                        579
+                                    ],
+                                    "score": 1.0,
+                                    "content": "art. Our results are based on the settings of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        291,
+                                        567,
+                                        327,
+                                        578
+                                    ],
+                                    "score": 0.89,
+                                    "content": "\\lambda = 0 . 5",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        327,
+                                        567,
+                                        331,
+                                        579
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        331,
+                                        567,
+                                        367,
+                                        578
+                                    ],
+                                    "score": 0.87,
+                                    "content": "\\kappa = 0 . 5",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        368,
+                                        567,
+                                        387,
+                                        579
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        387,
+                                        569,
+                                        396,
+                                        577
+                                    ],
+                                    "score": 0.77,
+                                    "content": "\\alpha",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        396,
+                                        567,
+                                        505,
+                                        579
+                                    ],
+                                    "score": 1.0,
+                                    "content": "selected on the validation",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                578,
+                                505,
+                                590
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        578,
+                                        505,
+                                        590
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sets of CUB and FLOWERS datasets. Please refer to Section 4.5 for the analysis of stability with",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                589,
+                                506,
+                                601
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        589,
+                                        209,
+                                        601
+                                    ],
+                                    "score": 1.0,
+                                    "content": "respect to the choices of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        209,
+                                        590,
+                                        217,
+                                        599
+                                    ],
+                                    "score": 0.77,
+                                    "content": "\\lambda",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        217,
+                                        589,
+                                        235,
+                                        601
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        236,
+                                        591,
+                                        243,
+                                        599
+                                    ],
+                                    "score": 0.76,
+                                    "content": "\\kappa",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        243,
+                                        589,
+                                        495,
+                                        601
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and Sections 2.2 and 4.4 for more details on the selection of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        496,
+                                        591,
+                                        503,
+                                        599
+                                    ],
+                                    "score": 0.73,
+                                    "content": "\\alpha",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        503,
+                                        589,
+                                        506,
+                                        601
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                599,
+                                505,
+                                613
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        599,
+                                        505,
+                                        613
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The combination of the proposed training method and the rebalancing of the metric space results",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                611,
+                                505,
+                                623
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        611,
+                                        505,
+                                        623
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in the state-of-the-art performance. Most of the current methods rely on the dataset augmentation",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                621,
+                                507,
+                                635
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        621,
+                                        507,
+                                        635
+                                    ],
+                                    "score": 1.0,
+                                    "content": "techniques based on GANs, VAEs or combinations thereof. Those are clearly complementary w.r.t.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                632,
+                                506,
+                                646
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        632,
+                                        506,
+                                        646
+                                    ],
+                                    "score": 1.0,
+                                    "content": "our method and their addition to the training procedure is likely to further boost the performance",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                643,
+                                506,
+                                657
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        643,
+                                        506,
+                                        657
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of our proposed approach. However, this is outside of the scope of the current work. Moreover,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                654,
+                                506,
+                                668
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        654,
+                                        506,
+                                        668
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the proposed method is state-of-the-art on FLOWERS even when compared against (Atzmon &",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                665,
+                                506,
+                                679
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        665,
+                                        506,
+                                        679
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Chechik, 2019) that uses both more sophisticated GAN based embedding learning approach and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                678,
+                                505,
+                                689
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        678,
+                                        505,
+                                        689
+                                    ],
+                                    "score": 1.0,
+                                    "content": "a more sophisticated seen/unseen error rate balancing based on COSMO. It is important to note",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                686,
+                                506,
+                                701
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        686,
+                                        506,
+                                        701
+                                    ],
+                                    "score": 1.0,
+                                    "content": "that Atzmon & Chechik (2019) did not report the sentence level results on CUB. Yet, when applied",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                699,
+                                505,
+                                711
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        699,
+                                        429,
+                                        711
+                                    ],
+                                    "score": 1.0,
+                                    "content": "on attributes together with f-CLSWGAN (Xian et al., 2018d) COSMO resulted in",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        430,
+                                        699,
+                                        452,
+                                        709
+                                    ],
+                                    "score": 0.86,
+                                    "content": "0 . 8 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        452,
+                                        699,
+                                        505,
+                                        711
+                                    ],
+                                    "score": 1.0,
+                                    "content": "performance",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                709,
+                                506,
+                                722
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        709,
+                                        405,
+                                        722
+                                    ],
+                                    "score": 1.0,
+                                    "content": "drop and when applied with LAGO (Atzmon & Chechik, 2018) it achieved",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        405,
+                                        710,
+                                        429,
+                                        720
+                                    ],
+                                    "score": 0.84,
+                                    "content": "0 . 5 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        430,
+                                        709,
+                                        506,
+                                        722
+                                    ],
+                                    "score": 1.0,
+                                    "content": "improvement over",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                720,
+                                243,
+                                732
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        720,
+                                        243,
+                                        732
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the attribute based state of the art.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        }
+                    ],
+                    "index": 34.5,
+                    "bbox_fs": [
+                        105,
+                        556,
+                        507,
+                        732
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "image",
+                    "bbox": [
+                        135,
+                        81,
+                        476,
+                        182
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                135,
+                                81,
+                                476,
+                                182
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        135,
+                                        81,
+                                        476,
+                                        182
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                135,
+                                                81,
+                                                476,
+                                                182
+                                            ],
+                                            "score": 0.967,
+                                            "type": "image",
+                                            "image_path": "3f384a4e76f0bc5c58077471f4b2954c58239b2f85fada1da5ff845d8c8bbbae.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 1,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        135,
+                                        81,
+                                        476,
+                                        114.66666666666666
+                                    ],
+                                    "spans": [],
+                                    "index": 0
+                                },
+                                {
+                                    "bbox": [
+                                        135,
+                                        114.66666666666666,
+                                        476,
+                                        148.33333333333331
+                                    ],
+                                    "spans": [],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        135,
+                                        148.33333333333331,
+                                        476,
+                                        181.99999999999997
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                106,
+                                190,
+                                505,
+                                225
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        191,
+                                        506,
+                                        203
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                191,
+                                                406,
+                                                203
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 3: Harmonic mean Top-1 accuracy on seen and unseen, H, against",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                406,
+                                                191,
+                                                413,
+                                                201
+                                            ],
+                                            "score": 0.73,
+                                            "content": "\\lambda",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                413,
+                                                191,
+                                                506,
+                                                203
+                                            ],
+                                            "score": 1.0,
+                                            "content": ", the relative weight of",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 3
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        201,
+                                        506,
+                                        216
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                201,
+                                                254,
+                                                216
+                                            ],
+                                            "score": 1.0,
+                                            "content": "image and text retrieval loss terms.",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                254,
+                                                202,
+                                                281,
+                                                212
+                                            ],
+                                            "score": 0.9,
+                                            "content": "\\lambda = 0",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                281,
+                                                201,
+                                                506,
+                                                216
+                                            ],
+                                            "score": 1.0,
+                                            "content": "corresponds to the case of image retrieval loss having",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 4
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        212,
+                                        433,
+                                        226
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                212,
+                                                433,
+                                                226
+                                            ],
+                                            "score": 1.0,
+                                            "content": "weight 1 and text retrieval loss having weight 0. Mean over 10 optimization runs.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 5
+                                }
+                            ],
+                            "index": 4
+                        }
+                    ],
+                    "index": 2.5
+                },
+                {
+                    "type": "image",
+                    "bbox": [
+                        135,
+                        240,
+                        476,
+                        342
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                135,
+                                240,
+                                476,
+                                342
+                            ],
+                            "group_id": 1,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        135,
+                                        240,
+                                        476,
+                                        342
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                135,
+                                                240,
+                                                476,
+                                                342
+                                            ],
+                                            "score": 0.97,
+                                            "type": "image",
+                                            "image_path": "a157ed151bbb217f07fbd4f1af909679c63192783007d447465bf462d40594e9.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 7,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        135,
+                                        240,
+                                        476,
+                                        274.0
+                                    ],
+                                    "spans": [],
+                                    "index": 6
+                                },
+                                {
+                                    "bbox": [
+                                        135,
+                                        274.0,
+                                        476,
+                                        308.0
+                                    ],
+                                    "spans": [],
+                                    "index": 7
+                                },
+                                {
+                                    "bbox": [
+                                        135,
+                                        308.0,
+                                        476,
+                                        342.0
+                                    ],
+                                    "spans": [],
+                                    "index": 8
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                106,
+                                351,
+                                506,
+                                384
+                            ],
+                            "group_id": 1,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        351,
+                                        505,
+                                        363
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                351,
+                                                433,
+                                                363
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 4: The plot of the harmonic mean Top-1 accuracy on seen and unseen,",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                433,
+                                                351,
+                                                442,
+                                                361
+                                            ],
+                                            "score": 0.4,
+                                            "content": "\\mathbf { H }",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                443,
+                                                351,
+                                                478,
+                                                363
+                                            ],
+                                            "score": 1.0,
+                                            "content": ", against",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                478,
+                                                353,
+                                                486,
+                                                361
+                                            ],
+                                            "score": 0.72,
+                                            "content": "\\kappa",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                486,
+                                                351,
+                                                505,
+                                                363
+                                            ],
+                                            "score": 1.0,
+                                            "content": ", the",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 9
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        362,
+                                        506,
+                                        374
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                362,
+                                                368,
+                                                374
+                                            ],
+                                            "score": 1.0,
+                                            "content": "relative weight of the retrieval and the classification loss terms.",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                368,
+                                                362,
+                                                395,
+                                                372
+                                            ],
+                                            "score": 0.9,
+                                            "content": "\\kappa = 0",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                395,
+                                                362,
+                                                506,
+                                                374
+                                            ],
+                                            "score": 1.0,
+                                            "content": "corresponds to the case of",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 10
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        372,
+                                        480,
+                                        385
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                372,
+                                                480,
+                                                385
+                                            ],
+                                            "score": 1.0,
+                                            "content": "classification loss having weight 0. The curves represent the mean over 10 optimization runs.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 11
+                                }
+                            ],
+                            "index": 10
+                        }
+                    ],
+                    "index": 8.5
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        110,
+                        411,
+                        329,
+                        421
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                107,
+                                410,
+                                332,
+                                422
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        410,
+                                        332,
+                                        422
+                                    ],
+                                    "score": 1.0,
+                                    "content": "4.4 ON THE SEEN/UNSEEN ACCURACY BALANCING",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        }
+                    ],
+                    "index": 12
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        432,
+                        505,
+                        553
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                432,
+                                505,
+                                445
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        432,
+                                        374,
+                                        445
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Figure 2 demonstrates the plot of harmonic mean Top-1 accuracy,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        375,
+                                        433,
+                                        384,
+                                        443
+                                    ],
+                                    "score": 0.25,
+                                    "content": "\\mathbf { H }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        384,
+                                        432,
+                                        469,
+                                        445
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", against the value of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        469,
+                                        435,
+                                        477,
+                                        442
+                                    ],
+                                    "score": 0.77,
+                                    "content": "\\alpha",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        477,
+                                        432,
+                                        505,
+                                        445
+                                    ],
+                                    "score": 1.0,
+                                    "content": "on the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                441,
+                                505,
+                                457
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        441,
+                                        505,
+                                        457
+                                    ],
+                                    "score": 1.0,
+                                    "content": "validation sets of CUB and FLOWERS datasets. The validation set is constructed by further splitting",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                454,
+                                505,
+                                466
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        454,
+                                        505,
+                                        466
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the train set on both datasets. For example, CUB has a train set of 5875 images from 100 seen classes",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                465,
+                                505,
+                                477
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        465,
+                                        505,
+                                        477
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and a validation set of 2946 images from 50 unseen classes. We further divide the train set into 4700",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                475,
+                                505,
+                                488
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        475,
+                                        369,
+                                        488
+                                    ],
+                                    "score": 1.0,
+                                    "content": "train images from 100 seen classes, 1175 seen validation images",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        369,
+                                        476,
+                                        455,
+                                        487
+                                    ],
+                                    "score": 0.91,
+                                    "content": "( 4 7 0 0 + 1 1 7 5 = 5 8 7 5 )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        456,
+                                        475,
+                                        505,
+                                        488
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and we use",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                487,
+                                506,
+                                499
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        487,
+                                        439,
+                                        499
+                                    ],
+                                    "score": 1.0,
+                                    "content": "all the 2946 images from 50 classes as the unseen validation set. Once the value of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        439,
+                                        489,
+                                        447,
+                                        497
+                                    ],
+                                    "score": 0.79,
+                                    "content": "\\alpha",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        448,
+                                        487,
+                                        506,
+                                        499
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is determined",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                498,
+                                505,
+                                510
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        498,
+                                        279,
+                                        510
+                                    ],
+                                    "score": 1.0,
+                                    "content": "we train the representation on the full train",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        279,
+                                        500,
+                                        285,
+                                        507
+                                    ],
+                                    "score": 0.31,
+                                    "content": "^ +",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        286,
+                                        498,
+                                        505,
+                                        510
+                                    ],
+                                    "score": 1.0,
+                                    "content": "val subset and report results on the test split (the usual",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                509,
+                                506,
+                                522
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        509,
+                                        506,
+                                        522
+                                    ],
+                                    "score": 1.0,
+                                    "content": "practice in GZSL). We confirm on the validation set that H exhibits an inverted U-shape behavior as",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                520,
+                                505,
+                                532
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        520,
+                                        161,
+                                        532
+                                    ],
+                                    "score": 1.0,
+                                    "content": "a function of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        161,
+                                        522,
+                                        169,
+                                        530
+                                    ],
+                                    "score": 0.74,
+                                    "content": "\\alpha",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        169,
+                                        520,
+                                        419,
+                                        532
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", which was theoretically predicted in Section 2.2. Therefore,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        419,
+                                        522,
+                                        427,
+                                        530
+                                    ],
+                                    "score": 0.74,
+                                    "content": "\\alpha",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        428,
+                                        520,
+                                        505,
+                                        532
+                                    ],
+                                    "score": 1.0,
+                                    "content": "can be selected on",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                531,
+                                505,
+                                543
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        531,
+                                        505,
+                                        543
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the validation set and then applied to re-scale the metric space to balance the accuracy on seen and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                542,
+                                344,
+                                554
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        542,
+                                        344,
+                                        554
+                                    ],
+                                    "score": 1.0,
+                                    "content": "unseen classes during test time as described in Section 2.2.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        }
+                    ],
+                    "index": 18
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        572,
+                        208,
+                        582
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                570,
+                                209,
+                                585
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        570,
+                                        209,
+                                        585
+                                    ],
+                                    "score": 1.0,
+                                    "content": "4.5 ABLATION STUDY",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        }
+                    ],
+                    "index": 24
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        594,
+                        505,
+                        693
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                594,
+                                505,
+                                606
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        594,
+                                        505,
+                                        606
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Figure 3 presents the results of the ablation study on the importance of image and text retrieval",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                605,
+                                506,
+                                618
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        605,
+                                        302,
+                                        618
+                                    ],
+                                    "score": 1.0,
+                                    "content": "losses. We see that all of the Top-1 accuracies",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        302,
+                                        605,
+                                        338,
+                                        616
+                                    ],
+                                    "score": 0.33,
+                                    "content": "( \\mathbf { H } , \\mathbf { s } , \\mathbf { u } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        338,
+                                        605,
+                                        446,
+                                        618
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are stable in the range of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        446,
+                                        605,
+                                        502,
+                                        617
+                                    ],
+                                    "score": 0.84,
+                                    "content": "\\lambda \\in [ 0 . 2 , 0 . 9 ]",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        502,
+                                        605,
+                                        506,
+                                        618
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                615,
+                                504,
+                                629
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        615,
+                                        496,
+                                        629
+                                    ],
+                                    "score": 1.0,
+                                    "content": "when both losses have tangible weight. Removing either text or image retrieval losses (setting",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        497,
+                                        617,
+                                        504,
+                                        626
+                                    ],
+                                    "score": 0.72,
+                                    "content": "\\lambda",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                627,
+                                505,
+                                639
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        627,
+                                        505,
+                                        639
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to 0 or 1 respectively) leads to performance drop in both cases. Removing the text retrieval loss",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                638,
+                                505,
+                                650
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        639,
+                                        129,
+                                        650
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(case",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        129,
+                                        638,
+                                        155,
+                                        648
+                                    ],
+                                    "score": 0.88,
+                                    "content": "\\lambda = 0",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        155,
+                                        639,
+                                        505,
+                                        650
+                                    ],
+                                    "score": 1.0,
+                                    "content": ") results in the most significant drop. This is due to the fact that the text retrieval task is",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                648,
+                                506,
+                                663
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        648,
+                                        506,
+                                        663
+                                    ],
+                                    "score": 1.0,
+                                    "content": "more tightly related to the GZSL task. At the batch level, retrieving the right text given an image",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                659,
+                                507,
+                                674
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        659,
+                                        507,
+                                        674
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is equivalent to identifying the correct class encoded by a text prototype during ZSL inference step.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                671,
+                                506,
+                                684
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        671,
+                                        506,
+                                        684
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The image retrieval task is not directly related to solving the ZSL problem and yet it does yield a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                681,
+                                339,
+                                694
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        681,
+                                        339,
+                                        694
+                                    ],
+                                    "score": 1.0,
+                                    "content": "positive regularizing effect on both CUB and FLOWERS.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        }
+                    ],
+                    "index": 29
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        699,
+                        504,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                698,
+                                506,
+                                712
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        698,
+                                        506,
+                                        712
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Figure 4 shows the results of the ablation study of the interplay between the retrieval loss and the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                710,
+                                505,
+                                721
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        710,
+                                        322,
+                                        721
+                                    ],
+                                    "score": 1.0,
+                                    "content": "classification loss. We observe, just as in the case with",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        322,
+                                        710,
+                                        329,
+                                        720
+                                    ],
+                                    "score": 0.78,
+                                    "content": "\\lambda",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        329,
+                                        710,
+                                        505,
+                                        721
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", that there exists a reasonably flat and stable",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                720,
+                                506,
+                                734
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        720,
+                                        141,
+                                        734
+                                    ],
+                                    "score": 1.0,
+                                    "content": "range of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        141,
+                                        720,
+                                        195,
+                                        732
+                                    ],
+                                    "score": 0.9,
+                                    "content": "\\kappa \\in [ 0 . 2 , 0 . 6 ]",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        196,
+                                        720,
+                                        255,
+                                        734
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". The range for",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        255,
+                                        722,
+                                        262,
+                                        730
+                                    ],
+                                    "score": 0.76,
+                                    "content": "\\kappa",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        263,
+                                        720,
+                                        325,
+                                        734
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is a bit smaller.",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        325,
+                                        721,
+                                        351,
+                                        730
+                                    ],
+                                    "score": 0.89,
+                                    "content": "\\kappa = 1",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        351,
+                                        720,
+                                        506,
+                                        734
+                                    ],
+                                    "score": 1.0,
+                                    "content": "results in the catastrophic performance",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        }
+                    ],
+                    "index": 35
+                }
+            ],
+            "page_idx": 6,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        302,
+                        751,
+                        308,
+                        759
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                302,
+                                750,
+                                309,
+                                762
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        302,
+                                        750,
+                                        309,
+                                        762
+                                    ],
+                                    "score": 1.0,
+                                    "content": "7",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        106,
+                        27,
+                        308,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                26,
+                                308,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        26,
+                                        308,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Under review as a conference paper at ICLR 2020",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "image",
+                    "bbox": [
+                        135,
+                        81,
+                        476,
+                        182
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                135,
+                                81,
+                                476,
+                                182
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        135,
+                                        81,
+                                        476,
+                                        182
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                135,
+                                                81,
+                                                476,
+                                                182
+                                            ],
+                                            "score": 0.967,
+                                            "type": "image",
+                                            "image_path": "3f384a4e76f0bc5c58077471f4b2954c58239b2f85fada1da5ff845d8c8bbbae.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 1,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        135,
+                                        81,
+                                        476,
+                                        114.66666666666666
+                                    ],
+                                    "spans": [],
+                                    "index": 0
+                                },
+                                {
+                                    "bbox": [
+                                        135,
+                                        114.66666666666666,
+                                        476,
+                                        148.33333333333331
+                                    ],
+                                    "spans": [],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        135,
+                                        148.33333333333331,
+                                        476,
+                                        181.99999999999997
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                106,
+                                190,
+                                505,
+                                225
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        191,
+                                        506,
+                                        203
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                191,
+                                                406,
+                                                203
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 3: Harmonic mean Top-1 accuracy on seen and unseen, H, against",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                406,
+                                                191,
+                                                413,
+                                                201
+                                            ],
+                                            "score": 0.73,
+                                            "content": "\\lambda",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                413,
+                                                191,
+                                                506,
+                                                203
+                                            ],
+                                            "score": 1.0,
+                                            "content": ", the relative weight of",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 3
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        201,
+                                        506,
+                                        216
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                201,
+                                                254,
+                                                216
+                                            ],
+                                            "score": 1.0,
+                                            "content": "image and text retrieval loss terms.",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                254,
+                                                202,
+                                                281,
+                                                212
+                                            ],
+                                            "score": 0.9,
+                                            "content": "\\lambda = 0",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                281,
+                                                201,
+                                                506,
+                                                216
+                                            ],
+                                            "score": 1.0,
+                                            "content": "corresponds to the case of image retrieval loss having",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 4
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        212,
+                                        433,
+                                        226
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                212,
+                                                433,
+                                                226
+                                            ],
+                                            "score": 1.0,
+                                            "content": "weight 1 and text retrieval loss having weight 0. Mean over 10 optimization runs.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 5
+                                }
+                            ],
+                            "index": 4
+                        }
+                    ],
+                    "index": 2.5
+                },
+                {
+                    "type": "image",
+                    "bbox": [
+                        135,
+                        240,
+                        476,
+                        342
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                135,
+                                240,
+                                476,
+                                342
+                            ],
+                            "group_id": 1,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        135,
+                                        240,
+                                        476,
+                                        342
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                135,
+                                                240,
+                                                476,
+                                                342
+                                            ],
+                                            "score": 0.97,
+                                            "type": "image",
+                                            "image_path": "a157ed151bbb217f07fbd4f1af909679c63192783007d447465bf462d40594e9.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 7,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        135,
+                                        240,
+                                        476,
+                                        274.0
+                                    ],
+                                    "spans": [],
+                                    "index": 6
+                                },
+                                {
+                                    "bbox": [
+                                        135,
+                                        274.0,
+                                        476,
+                                        308.0
+                                    ],
+                                    "spans": [],
+                                    "index": 7
+                                },
+                                {
+                                    "bbox": [
+                                        135,
+                                        308.0,
+                                        476,
+                                        342.0
+                                    ],
+                                    "spans": [],
+                                    "index": 8
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                106,
+                                351,
+                                506,
+                                384
+                            ],
+                            "group_id": 1,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        351,
+                                        505,
+                                        363
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                351,
+                                                433,
+                                                363
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 4: The plot of the harmonic mean Top-1 accuracy on seen and unseen,",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                433,
+                                                351,
+                                                442,
+                                                361
+                                            ],
+                                            "score": 0.4,
+                                            "content": "\\mathbf { H }",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                443,
+                                                351,
+                                                478,
+                                                363
+                                            ],
+                                            "score": 1.0,
+                                            "content": ", against",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                478,
+                                                353,
+                                                486,
+                                                361
+                                            ],
+                                            "score": 0.72,
+                                            "content": "\\kappa",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                486,
+                                                351,
+                                                505,
+                                                363
+                                            ],
+                                            "score": 1.0,
+                                            "content": ", the",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 9
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        362,
+                                        506,
+                                        374
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                362,
+                                                368,
+                                                374
+                                            ],
+                                            "score": 1.0,
+                                            "content": "relative weight of the retrieval and the classification loss terms.",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                368,
+                                                362,
+                                                395,
+                                                372
+                                            ],
+                                            "score": 0.9,
+                                            "content": "\\kappa = 0",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                395,
+                                                362,
+                                                506,
+                                                374
+                                            ],
+                                            "score": 1.0,
+                                            "content": "corresponds to the case of",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 10
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        372,
+                                        480,
+                                        385
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                372,
+                                                480,
+                                                385
+                                            ],
+                                            "score": 1.0,
+                                            "content": "classification loss having weight 0. The curves represent the mean over 10 optimization runs.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 11
+                                }
+                            ],
+                            "index": 10
+                        }
+                    ],
+                    "index": 8.5
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        110,
+                        411,
+                        329,
+                        421
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                107,
+                                410,
+                                332,
+                                422
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        410,
+                                        332,
+                                        422
+                                    ],
+                                    "score": 1.0,
+                                    "content": "4.4 ON THE SEEN/UNSEEN ACCURACY BALANCING",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        }
+                    ],
+                    "index": 12
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        432,
+                        505,
+                        553
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                432,
+                                505,
+                                445
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        432,
+                                        374,
+                                        445
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Figure 2 demonstrates the plot of harmonic mean Top-1 accuracy,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        375,
+                                        433,
+                                        384,
+                                        443
+                                    ],
+                                    "score": 0.25,
+                                    "content": "\\mathbf { H }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        384,
+                                        432,
+                                        469,
+                                        445
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", against the value of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        469,
+                                        435,
+                                        477,
+                                        442
+                                    ],
+                                    "score": 0.77,
+                                    "content": "\\alpha",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        477,
+                                        432,
+                                        505,
+                                        445
+                                    ],
+                                    "score": 1.0,
+                                    "content": "on the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                441,
+                                505,
+                                457
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        441,
+                                        505,
+                                        457
+                                    ],
+                                    "score": 1.0,
+                                    "content": "validation sets of CUB and FLOWERS datasets. The validation set is constructed by further splitting",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                454,
+                                505,
+                                466
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        454,
+                                        505,
+                                        466
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the train set on both datasets. For example, CUB has a train set of 5875 images from 100 seen classes",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                465,
+                                505,
+                                477
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        465,
+                                        505,
+                                        477
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and a validation set of 2946 images from 50 unseen classes. We further divide the train set into 4700",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                475,
+                                505,
+                                488
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        475,
+                                        369,
+                                        488
+                                    ],
+                                    "score": 1.0,
+                                    "content": "train images from 100 seen classes, 1175 seen validation images",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        369,
+                                        476,
+                                        455,
+                                        487
+                                    ],
+                                    "score": 0.91,
+                                    "content": "( 4 7 0 0 + 1 1 7 5 = 5 8 7 5 )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        456,
+                                        475,
+                                        505,
+                                        488
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and we use",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                487,
+                                506,
+                                499
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        487,
+                                        439,
+                                        499
+                                    ],
+                                    "score": 1.0,
+                                    "content": "all the 2946 images from 50 classes as the unseen validation set. Once the value of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        439,
+                                        489,
+                                        447,
+                                        497
+                                    ],
+                                    "score": 0.79,
+                                    "content": "\\alpha",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        448,
+                                        487,
+                                        506,
+                                        499
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is determined",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                498,
+                                505,
+                                510
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        498,
+                                        279,
+                                        510
+                                    ],
+                                    "score": 1.0,
+                                    "content": "we train the representation on the full train",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        279,
+                                        500,
+                                        285,
+                                        507
+                                    ],
+                                    "score": 0.31,
+                                    "content": "^ +",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        286,
+                                        498,
+                                        505,
+                                        510
+                                    ],
+                                    "score": 1.0,
+                                    "content": "val subset and report results on the test split (the usual",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                509,
+                                506,
+                                522
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        509,
+                                        506,
+                                        522
+                                    ],
+                                    "score": 1.0,
+                                    "content": "practice in GZSL). We confirm on the validation set that H exhibits an inverted U-shape behavior as",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                520,
+                                505,
+                                532
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        520,
+                                        161,
+                                        532
+                                    ],
+                                    "score": 1.0,
+                                    "content": "a function of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        161,
+                                        522,
+                                        169,
+                                        530
+                                    ],
+                                    "score": 0.74,
+                                    "content": "\\alpha",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        169,
+                                        520,
+                                        419,
+                                        532
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", which was theoretically predicted in Section 2.2. Therefore,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        419,
+                                        522,
+                                        427,
+                                        530
+                                    ],
+                                    "score": 0.74,
+                                    "content": "\\alpha",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        428,
+                                        520,
+                                        505,
+                                        532
+                                    ],
+                                    "score": 1.0,
+                                    "content": "can be selected on",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                531,
+                                505,
+                                543
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        531,
+                                        505,
+                                        543
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the validation set and then applied to re-scale the metric space to balance the accuracy on seen and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                542,
+                                344,
+                                554
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        542,
+                                        344,
+                                        554
+                                    ],
+                                    "score": 1.0,
+                                    "content": "unseen classes during test time as described in Section 2.2.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        }
+                    ],
+                    "index": 18,
+                    "bbox_fs": [
+                        104,
+                        432,
+                        506,
+                        554
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        572,
+                        208,
+                        582
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                570,
+                                209,
+                                585
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        570,
+                                        209,
+                                        585
+                                    ],
+                                    "score": 1.0,
+                                    "content": "4.5 ABLATION STUDY",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        }
+                    ],
+                    "index": 24
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        594,
+                        505,
+                        693
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                594,
+                                505,
+                                606
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        594,
+                                        505,
+                                        606
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Figure 3 presents the results of the ablation study on the importance of image and text retrieval",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                605,
+                                506,
+                                618
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        605,
+                                        302,
+                                        618
+                                    ],
+                                    "score": 1.0,
+                                    "content": "losses. We see that all of the Top-1 accuracies",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        302,
+                                        605,
+                                        338,
+                                        616
+                                    ],
+                                    "score": 0.33,
+                                    "content": "( \\mathbf { H } , \\mathbf { s } , \\mathbf { u } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        338,
+                                        605,
+                                        446,
+                                        618
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are stable in the range of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        446,
+                                        605,
+                                        502,
+                                        617
+                                    ],
+                                    "score": 0.84,
+                                    "content": "\\lambda \\in [ 0 . 2 , 0 . 9 ]",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        502,
+                                        605,
+                                        506,
+                                        618
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                615,
+                                504,
+                                629
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        615,
+                                        496,
+                                        629
+                                    ],
+                                    "score": 1.0,
+                                    "content": "when both losses have tangible weight. Removing either text or image retrieval losses (setting",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        497,
+                                        617,
+                                        504,
+                                        626
+                                    ],
+                                    "score": 0.72,
+                                    "content": "\\lambda",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                627,
+                                505,
+                                639
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        627,
+                                        505,
+                                        639
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to 0 or 1 respectively) leads to performance drop in both cases. Removing the text retrieval loss",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                638,
+                                505,
+                                650
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        639,
+                                        129,
+                                        650
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(case",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        129,
+                                        638,
+                                        155,
+                                        648
+                                    ],
+                                    "score": 0.88,
+                                    "content": "\\lambda = 0",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        155,
+                                        639,
+                                        505,
+                                        650
+                                    ],
+                                    "score": 1.0,
+                                    "content": ") results in the most significant drop. This is due to the fact that the text retrieval task is",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                648,
+                                506,
+                                663
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        648,
+                                        506,
+                                        663
+                                    ],
+                                    "score": 1.0,
+                                    "content": "more tightly related to the GZSL task. At the batch level, retrieving the right text given an image",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                659,
+                                507,
+                                674
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        659,
+                                        507,
+                                        674
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is equivalent to identifying the correct class encoded by a text prototype during ZSL inference step.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                671,
+                                506,
+                                684
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        671,
+                                        506,
+                                        684
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The image retrieval task is not directly related to solving the ZSL problem and yet it does yield a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                681,
+                                339,
+                                694
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        681,
+                                        339,
+                                        694
+                                    ],
+                                    "score": 1.0,
+                                    "content": "positive regularizing effect on both CUB and FLOWERS.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        }
+                    ],
+                    "index": 29,
+                    "bbox_fs": [
+                        104,
+                        594,
+                        507,
+                        694
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        699,
+                        504,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                698,
+                                506,
+                                712
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        698,
+                                        506,
+                                        712
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Figure 4 shows the results of the ablation study of the interplay between the retrieval loss and the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                710,
+                                505,
+                                721
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        710,
+                                        322,
+                                        721
+                                    ],
+                                    "score": 1.0,
+                                    "content": "classification loss. We observe, just as in the case with",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        322,
+                                        710,
+                                        329,
+                                        720
+                                    ],
+                                    "score": 0.78,
+                                    "content": "\\lambda",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        329,
+                                        710,
+                                        505,
+                                        721
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", that there exists a reasonably flat and stable",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                720,
+                                506,
+                                734
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        720,
+                                        141,
+                                        734
+                                    ],
+                                    "score": 1.0,
+                                    "content": "range of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        141,
+                                        720,
+                                        195,
+                                        732
+                                    ],
+                                    "score": 0.9,
+                                    "content": "\\kappa \\in [ 0 . 2 , 0 . 6 ]",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        196,
+                                        720,
+                                        255,
+                                        734
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". The range for",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        255,
+                                        722,
+                                        262,
+                                        730
+                                    ],
+                                    "score": 0.76,
+                                    "content": "\\kappa",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        263,
+                                        720,
+                                        325,
+                                        734
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is a bit smaller.",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        325,
+                                        721,
+                                        351,
+                                        730
+                                    ],
+                                    "score": 0.89,
+                                    "content": "\\kappa = 1",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        351,
+                                        720,
+                                        506,
+                                        734
+                                    ],
+                                    "score": 1.0,
+                                    "content": "results in the catastrophic performance",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                217,
+                                505,
+                                231
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        217,
+                                        505,
+                                        231
+                                    ],
+                                    "score": 1.0,
+                                    "content": "drop: the classification losses by themselves do not enforce any modality alignment (please refer to",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                229,
+                                312,
+                                241
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        229,
+                                        312,
+                                        241
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Fig. 1 and Algorithm 1 clearly demonstrating this).",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 5
+                        }
+                    ],
+                    "index": 35,
+                    "bbox_fs": [
+                        105,
+                        698,
+                        506,
+                        734
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "table",
+                    "bbox": [
+                        179,
+                        100,
+                        432,
+                        197
+                    ],
+                    "blocks": [
+                        {
+                            "type": "table_caption",
+                            "bbox": [
+                                152,
+                                80,
+                                457,
+                                92
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        151,
+                                        78,
+                                        458,
+                                        95
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                151,
+                                                78,
+                                                458,
+                                                95
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Table 3: Generalized zero-shot Top-1 classification accuracy, ablation study.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 0
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "type": "table_body",
+                            "bbox": [
+                                179,
+                                100,
+                                432,
+                                197
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        179,
+                                        100,
+                                        432,
+                                        197
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                179,
+                                                100,
+                                                432,
+                                                197
+                                            ],
+                                            "score": 0.981,
+                                            "html": "<table><tr><td></td><td></td><td></td><td></td><td>CUB</td><td></td><td colspan=\"3\">FLOWERS</td></tr><tr><td>α</td><td>入</td><td>K</td><td>u</td><td>S</td><td>H</td><td>u</td><td>s</td><td>H</td></tr><tr><td>0.0</td><td>0.5</td><td>0.5</td><td>38.3</td><td>65.3</td><td>48.3</td><td>55.1</td><td>84.6</td><td>66.7</td></tr><tr><td>0.0</td><td>0.5</td><td>0.0</td><td>39.3</td><td>57.5</td><td>46.7</td><td>54.0</td><td>78.1</td><td>63.8</td></tr><tr><td>√</td><td>0.5</td><td>0.0</td><td>53.8</td><td>49.6</td><td>51.6</td><td>71.7</td><td>67.2</td><td>69.4</td></tr><tr><td>√</td><td>0.0</td><td>0.5</td><td>47.4</td><td>36.6</td><td>41.3</td><td>51.5</td><td>60.5</td><td>55.6</td></tr><tr><td>√</td><td>1.0</td><td>0.5</td><td>53.9</td><td>53.8</td><td>53.8</td><td>69.5</td><td>73.9</td><td>71.6</td></tr><tr><td>厂</td><td>0.5</td><td>0.5</td><td>59.3</td><td>52.6</td><td>55.8</td><td>73.0</td><td>73.6</td><td>73.3</td></tr></table>",
+                                            "type": "table",
+                                            "image_path": "a9ed39c401c5a3ac797b5f3447e0276b6810d59aee48a6e4a3f12e525762ff7f.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 2,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        179,
+                                        100,
+                                        432,
+                                        132.33333333333334
+                                    ],
+                                    "spans": [],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        179,
+                                        132.33333333333334,
+                                        432,
+                                        164.66666666666669
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                },
+                                {
+                                    "bbox": [
+                                        179,
+                                        164.66666666666669,
+                                        432,
+                                        197.00000000000003
+                                    ],
+                                    "spans": [],
+                                    "index": 3
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 1.0
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        217,
+                        503,
+                        240
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                217,
+                                505,
+                                231
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        217,
+                                        505,
+                                        231
+                                    ],
+                                    "score": 1.0,
+                                    "content": "drop: the classification losses by themselves do not enforce any modality alignment (please refer to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                229,
+                                312,
+                                241
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        229,
+                                        312,
+                                        241
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Fig. 1 and Algorithm 1 clearly demonstrating this).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        }
+                    ],
+                    "index": 4.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        245,
+                        505,
+                        432
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                245,
+                                506,
+                                258
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        245,
+                                        506,
+                                        258
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Table 3 studies the effects of different loss terms on the harmonic mean Top-1 accuracy H. The best",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                256,
+                                505,
+                                269
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        256,
+                                        505,
+                                        269
+                                    ],
+                                    "score": 1.0,
+                                    "content": "result is achieved when all loss terms are active and when the metric space rescaling is on (the case of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                267,
+                                505,
+                                280
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        267,
+                                        140,
+                                        278
+                                    ],
+                                    "score": 0.87,
+                                    "content": "\\lambda = 0 . 5",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        141,
+                                        267,
+                                        144,
+                                        280
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        144,
+                                        268,
+                                        178,
+                                        278
+                                    ],
+                                    "score": 0.87,
+                                    "content": "\\kappa = 0 . 5",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        179,
+                                        267,
+                                        196,
+                                        280
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        197,
+                                        269,
+                                        205,
+                                        278
+                                    ],
+                                    "score": 0.78,
+                                    "content": "\\alpha",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        205,
+                                        267,
+                                        505,
+                                        280
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is checked, the last line in the table). Comparing this with the case when",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                277,
+                                505,
+                                291
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        277,
+                                        301,
+                                        291
+                                    ],
+                                    "score": 1.0,
+                                    "content": "there is no metric space rescaling (first line with",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        301,
+                                        279,
+                                        327,
+                                        289
+                                    ],
+                                    "score": 0.89,
+                                    "content": "\\alpha = 0",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        327,
+                                        277,
+                                        505,
+                                        291
+                                    ],
+                                    "score": 1.0,
+                                    "content": "), we see that the rescaling helps to decrease",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                289,
+                                505,
+                                302
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        289,
+                                        505,
+                                        302
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the gap between seen and unseen classification accuracy. For CUB, the discrepancy reduction is from",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                299,
+                                506,
+                                313
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        299,
+                                        137,
+                                        313
+                                    ],
+                                    "score": 1.0,
+                                    "content": "around",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        137,
+                                        300,
+                                        157,
+                                        311
+                                    ],
+                                    "score": 0.88,
+                                    "content": "30 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        158,
+                                        299,
+                                        199,
+                                        313
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to around",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        199,
+                                        300,
+                                        214,
+                                        311
+                                    ],
+                                    "score": 0.87,
+                                    "content": "6 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        214,
+                                        299,
+                                        352,
+                                        313
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", for FLOWERS it is from around",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        352,
+                                        300,
+                                        372,
+                                        311
+                                    ],
+                                    "score": 0.88,
+                                    "content": "30 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        372,
+                                        299,
+                                        414,
+                                        313
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to around",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        415,
+                                        300,
+                                        429,
+                                        311
+                                    ],
+                                    "score": 0.86,
+                                    "content": "1 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        429,
+                                        299,
+                                        506,
+                                        313
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". We would like to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                312,
+                                505,
+                                324
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        312,
+                                        505,
+                                        324
+                                    ],
+                                    "score": 1.0,
+                                    "content": "stress that we only use images and texts from the training set to achieve that. Going to the second",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                321,
+                                507,
+                                336
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        321,
+                                        346,
+                                        336
+                                    ],
+                                    "score": 1.0,
+                                    "content": "line in the table (the image/text classification loss is inactive,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        347,
+                                        323,
+                                        372,
+                                        333
+                                    ],
+                                    "score": 0.89,
+                                    "content": "\\kappa = 0",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        372,
+                                        321,
+                                        507,
+                                        336
+                                    ],
+                                    "score": 1.0,
+                                    "content": ") and comparing it to the first one,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                333,
+                                506,
+                                346
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        333,
+                                        506,
+                                        346
+                                    ],
+                                    "score": 1.0,
+                                    "content": "we assess the effect of the image/text classification loss. It barely affects the performance on unseen",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                344,
+                                506,
+                                357
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        344,
+                                        420,
+                                        357
+                                    ],
+                                    "score": 1.0,
+                                    "content": "set, but it significantly boosts the classification accuracy on the seen set (around",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        420,
+                                        344,
+                                        434,
+                                        355
+                                    ],
+                                    "score": 0.88,
+                                    "content": "8 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        435,
+                                        344,
+                                        506,
+                                        357
+                                    ],
+                                    "score": 1.0,
+                                    "content": "on both datasets).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                354,
+                                506,
+                                368
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        354,
+                                        506,
+                                        368
+                                    ],
+                                    "score": 1.0,
+                                    "content": "This is logical: adding a classifier loss results in a better classifier of the test images from the seen",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                366,
+                                506,
+                                379
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        366,
+                                        506,
+                                        379
+                                    ],
+                                    "score": 1.0,
+                                    "content": "classes. This alone does not make it a better GZSL classifier, however. Only when applied together",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                377,
+                                505,
+                                390
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        377,
+                                        505,
+                                        390
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with metric space rescaling, this results in the performance boost (please refer to lines 1 and 6 in",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                388,
+                                505,
+                                400
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        388,
+                                        505,
+                                        400
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Table 3). Our interpretation is that the addition of the image/text classifier loss helps to reduce the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                399,
+                                505,
+                                411
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        399,
+                                        505,
+                                        411
+                                    ],
+                                    "score": 1.0,
+                                    "content": "intraclass variability in embeddings and provides for tighter clustering. However, this also leads to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                408,
+                                506,
+                                424
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        408,
+                                        506,
+                                        424
+                                    ],
+                                    "score": 1.0,
+                                    "content": "overfit on the classification task. This is accounted for by metric rescaling that enables the learnings",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                421,
+                                444,
+                                434
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        421,
+                                        444,
+                                        434
+                                    ],
+                                    "score": 1.0,
+                                    "content": "from the image/text classification task be transferred effectively into the GZSL task.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        }
+                    ],
+                    "index": 14
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        438,
+                        505,
+                        570
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                437,
+                                505,
+                                450
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        437,
+                                        505,
+                                        450
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The comparison of the last four rows of Table 3 leads us to believe that all the proposed loss terms",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                449,
+                                507,
+                                461
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        449,
+                                        507,
+                                        461
+                                    ],
+                                    "score": 1.0,
+                                    "content": "outlined in Fig. 1 and Algorithm 1 are important for achieving the state-of-the-art performance.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                460,
+                                505,
+                                472
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        460,
+                                        380,
+                                        472
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Excluding any one of them (corresponding to the extreme values",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        380,
+                                        460,
+                                        409,
+                                        470
+                                    ],
+                                    "score": 0.84,
+                                    "content": "\\lambda = 0",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        409,
+                                        460,
+                                        414,
+                                        472
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        414,
+                                        460,
+                                        443,
+                                        470
+                                    ],
+                                    "score": 0.78,
+                                    "content": "\\lambda = 1",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        443,
+                                        460,
+                                        447,
+                                        472
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        447,
+                                        460,
+                                        477,
+                                        470
+                                    ],
+                                    "score": 0.81,
+                                    "content": "\\kappa = 0",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        477,
+                                        460,
+                                        505,
+                                        472
+                                    ],
+                                    "score": 1.0,
+                                    "content": ") leads",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                470,
+                                506,
+                                484
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        470,
+                                        506,
+                                        484
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to performance deterioration. Finally, an interesting observation can be made by comparing line",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                482,
+                                506,
+                                494
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        482,
+                                        506,
+                                        494
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3 of Table 3 with performance of algorithms in Table 1. In this case our algorithm does not use",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                493,
+                                505,
+                                505
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        493,
+                                        505,
+                                        505
+                                    ],
+                                    "score": 1.0,
+                                    "content": "any class labels and relies on training using exclusively the retrieval losses that can be calculated",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                504,
+                                505,
+                                516
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        504,
+                                        505,
+                                        516
+                                    ],
+                                    "score": 1.0,
+                                    "content": "only based on the pairwise relationships between texts and images. We can see that using this",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                514,
+                                505,
+                                527
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        514,
+                                        505,
+                                        527
+                                    ],
+                                    "score": 1.0,
+                                    "content": "type of supervision alone already results in a very high-quality representation. The representation",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                525,
+                                505,
+                                538
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        525,
+                                        505,
+                                        538
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is competitive against the latest GAN/VAE based approaches on CUB and is state-of-the-art on",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                535,
+                                506,
+                                550
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        535,
+                                        506,
+                                        550
+                                    ],
+                                    "score": 1.0,
+                                    "content": "FLOWERS. This opens up new exploration avenues showing that in the case when very fine-grained",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                547,
+                                506,
+                                561
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        547,
+                                        506,
+                                        561
+                                    ],
+                                    "score": 1.0,
+                                    "content": "modality outputs are available (image and text description pairs being a very prominent example), the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                559,
+                                486,
+                                571
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        559,
+                                        486,
+                                        571
+                                    ],
+                                    "score": 1.0,
+                                    "content": "high-quality representations may be learned without relying on manually supplied class labels.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        }
+                    ],
+                    "index": 28.5
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        586,
+                        201,
+                        598
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                583,
+                                203,
+                                601
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        583,
+                                        203,
+                                        601
+                                    ],
+                                    "score": 1.0,
+                                    "content": "5 CONCLUSIONS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        }
+                    ],
+                    "index": 35
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        610,
+                        505,
+                        731
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                611,
+                                506,
+                                624
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        611,
+                                        506,
+                                        624
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We propose and empirically validate two improvements to the process of learning fine-grained cross-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                622,
+                                505,
+                                634
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        622,
+                                        505,
+                                        634
+                                    ],
+                                    "score": 1.0,
+                                    "content": "modal representations. First, we confirm the hypothesis that in the context of paired images and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                632,
+                                505,
+                                646
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        632,
+                                        505,
+                                        646
+                                    ],
+                                    "score": 1.0,
+                                    "content": "texts, a deep metric learning approach can be driven by an instance-based retrieval loss resulting",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                644,
+                                505,
+                                655
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        644,
+                                        505,
+                                        655
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in competitive generalized zero shot classification results. Combined with an additional class label",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                654,
+                                505,
+                                667
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        654,
+                                        505,
+                                        667
+                                    ],
+                                    "score": 1.0,
+                                    "content": "based image/text crossentropy term this results in state-of-the-art performance on two well known",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                665,
+                                505,
+                                678
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        665,
+                                        505,
+                                        678
+                                    ],
+                                    "score": 1.0,
+                                    "content": "datasets, CUB and FLOWERS. This is an interesting result demonstrating that high-quality deep",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                676,
+                                506,
+                                691
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        676,
+                                        506,
+                                        691
+                                    ],
+                                    "score": 1.0,
+                                    "content": "representations can be trained relying largely on pairwise relationships between modalities. On top",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                688,
+                                505,
+                                700
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        688,
+                                        505,
+                                        700
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of that, we propose a solution to one of the prominent problems in GZSL: classifying instances",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                699,
+                                504,
+                                711
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        699,
+                                        504,
+                                        711
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of unseen classes as seen ones. We mathematically analyze and empirically validate the method",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                710,
+                                506,
+                                722
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        710,
+                                        506,
+                                        722
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of adjusting a single scalar that transcends in its effectiveness advanced dataset augmentation and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                721,
+                                298,
+                                732
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        721,
+                                        298,
+                                        732
+                                    ],
+                                    "score": 1.0,
+                                    "content": "training approaches based on GANs and VAEs.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        }
+                    ],
+                    "index": 41
+                }
+            ],
+            "page_idx": 7,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        106,
+                        27,
+                        308,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                26,
+                                308,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        26,
+                                        308,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Under review as a conference paper at ICLR 2020",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        302,
+                        751,
+                        308,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                300,
+                                750,
+                                309,
+                                761
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        300,
+                                        750,
+                                        309,
+                                        761
+                                    ],
+                                    "score": 1.0,
+                                    "content": "8",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "table",
+                    "bbox": [
+                        179,
+                        100,
+                        432,
+                        197
+                    ],
+                    "blocks": [
+                        {
+                            "type": "table_caption",
+                            "bbox": [
+                                152,
+                                80,
+                                457,
+                                92
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        151,
+                                        78,
+                                        458,
+                                        95
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                151,
+                                                78,
+                                                458,
+                                                95
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Table 3: Generalized zero-shot Top-1 classification accuracy, ablation study.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 0
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "type": "table_body",
+                            "bbox": [
+                                179,
+                                100,
+                                432,
+                                197
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        179,
+                                        100,
+                                        432,
+                                        197
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                179,
+                                                100,
+                                                432,
+                                                197
+                                            ],
+                                            "score": 0.981,
+                                            "html": "<table><tr><td></td><td></td><td></td><td></td><td>CUB</td><td></td><td colspan=\"3\">FLOWERS</td></tr><tr><td>α</td><td>入</td><td>K</td><td>u</td><td>S</td><td>H</td><td>u</td><td>s</td><td>H</td></tr><tr><td>0.0</td><td>0.5</td><td>0.5</td><td>38.3</td><td>65.3</td><td>48.3</td><td>55.1</td><td>84.6</td><td>66.7</td></tr><tr><td>0.0</td><td>0.5</td><td>0.0</td><td>39.3</td><td>57.5</td><td>46.7</td><td>54.0</td><td>78.1</td><td>63.8</td></tr><tr><td>√</td><td>0.5</td><td>0.0</td><td>53.8</td><td>49.6</td><td>51.6</td><td>71.7</td><td>67.2</td><td>69.4</td></tr><tr><td>√</td><td>0.0</td><td>0.5</td><td>47.4</td><td>36.6</td><td>41.3</td><td>51.5</td><td>60.5</td><td>55.6</td></tr><tr><td>√</td><td>1.0</td><td>0.5</td><td>53.9</td><td>53.8</td><td>53.8</td><td>69.5</td><td>73.9</td><td>71.6</td></tr><tr><td>厂</td><td>0.5</td><td>0.5</td><td>59.3</td><td>52.6</td><td>55.8</td><td>73.0</td><td>73.6</td><td>73.3</td></tr></table>",
+                                            "type": "table",
+                                            "image_path": "a9ed39c401c5a3ac797b5f3447e0276b6810d59aee48a6e4a3f12e525762ff7f.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 2,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        179,
+                                        100,
+                                        432,
+                                        132.33333333333334
+                                    ],
+                                    "spans": [],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        179,
+                                        132.33333333333334,
+                                        432,
+                                        164.66666666666669
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                },
+                                {
+                                    "bbox": [
+                                        179,
+                                        164.66666666666669,
+                                        432,
+                                        197.00000000000003
+                                    ],
+                                    "spans": [],
+                                    "index": 3
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 1.0
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        217,
+                        503,
+                        240
+                    ],
+                    "lines": [],
+                    "index": 4.5,
+                    "bbox_fs": [
+                        105,
+                        217,
+                        505,
+                        241
+                    ],
+                    "lines_deleted": true
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        245,
+                        505,
+                        432
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                245,
+                                506,
+                                258
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        245,
+                                        506,
+                                        258
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Table 3 studies the effects of different loss terms on the harmonic mean Top-1 accuracy H. The best",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                256,
+                                505,
+                                269
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        256,
+                                        505,
+                                        269
+                                    ],
+                                    "score": 1.0,
+                                    "content": "result is achieved when all loss terms are active and when the metric space rescaling is on (the case of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                267,
+                                505,
+                                280
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        267,
+                                        140,
+                                        278
+                                    ],
+                                    "score": 0.87,
+                                    "content": "\\lambda = 0 . 5",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        141,
+                                        267,
+                                        144,
+                                        280
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        144,
+                                        268,
+                                        178,
+                                        278
+                                    ],
+                                    "score": 0.87,
+                                    "content": "\\kappa = 0 . 5",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        179,
+                                        267,
+                                        196,
+                                        280
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        197,
+                                        269,
+                                        205,
+                                        278
+                                    ],
+                                    "score": 0.78,
+                                    "content": "\\alpha",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        205,
+                                        267,
+                                        505,
+                                        280
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is checked, the last line in the table). Comparing this with the case when",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                277,
+                                505,
+                                291
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        277,
+                                        301,
+                                        291
+                                    ],
+                                    "score": 1.0,
+                                    "content": "there is no metric space rescaling (first line with",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        301,
+                                        279,
+                                        327,
+                                        289
+                                    ],
+                                    "score": 0.89,
+                                    "content": "\\alpha = 0",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        327,
+                                        277,
+                                        505,
+                                        291
+                                    ],
+                                    "score": 1.0,
+                                    "content": "), we see that the rescaling helps to decrease",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                289,
+                                505,
+                                302
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        289,
+                                        505,
+                                        302
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the gap between seen and unseen classification accuracy. For CUB, the discrepancy reduction is from",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                299,
+                                506,
+                                313
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        299,
+                                        137,
+                                        313
+                                    ],
+                                    "score": 1.0,
+                                    "content": "around",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        137,
+                                        300,
+                                        157,
+                                        311
+                                    ],
+                                    "score": 0.88,
+                                    "content": "30 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        158,
+                                        299,
+                                        199,
+                                        313
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to around",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        199,
+                                        300,
+                                        214,
+                                        311
+                                    ],
+                                    "score": 0.87,
+                                    "content": "6 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        214,
+                                        299,
+                                        352,
+                                        313
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", for FLOWERS it is from around",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        352,
+                                        300,
+                                        372,
+                                        311
+                                    ],
+                                    "score": 0.88,
+                                    "content": "30 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        372,
+                                        299,
+                                        414,
+                                        313
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to around",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        415,
+                                        300,
+                                        429,
+                                        311
+                                    ],
+                                    "score": 0.86,
+                                    "content": "1 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        429,
+                                        299,
+                                        506,
+                                        313
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". We would like to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                312,
+                                505,
+                                324
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        312,
+                                        505,
+                                        324
+                                    ],
+                                    "score": 1.0,
+                                    "content": "stress that we only use images and texts from the training set to achieve that. Going to the second",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                321,
+                                507,
+                                336
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        321,
+                                        346,
+                                        336
+                                    ],
+                                    "score": 1.0,
+                                    "content": "line in the table (the image/text classification loss is inactive,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        347,
+                                        323,
+                                        372,
+                                        333
+                                    ],
+                                    "score": 0.89,
+                                    "content": "\\kappa = 0",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        372,
+                                        321,
+                                        507,
+                                        336
+                                    ],
+                                    "score": 1.0,
+                                    "content": ") and comparing it to the first one,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                333,
+                                506,
+                                346
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        333,
+                                        506,
+                                        346
+                                    ],
+                                    "score": 1.0,
+                                    "content": "we assess the effect of the image/text classification loss. It barely affects the performance on unseen",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                344,
+                                506,
+                                357
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        344,
+                                        420,
+                                        357
+                                    ],
+                                    "score": 1.0,
+                                    "content": "set, but it significantly boosts the classification accuracy on the seen set (around",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        420,
+                                        344,
+                                        434,
+                                        355
+                                    ],
+                                    "score": 0.88,
+                                    "content": "8 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        435,
+                                        344,
+                                        506,
+                                        357
+                                    ],
+                                    "score": 1.0,
+                                    "content": "on both datasets).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                354,
+                                506,
+                                368
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        354,
+                                        506,
+                                        368
+                                    ],
+                                    "score": 1.0,
+                                    "content": "This is logical: adding a classifier loss results in a better classifier of the test images from the seen",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                366,
+                                506,
+                                379
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        366,
+                                        506,
+                                        379
+                                    ],
+                                    "score": 1.0,
+                                    "content": "classes. This alone does not make it a better GZSL classifier, however. Only when applied together",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                377,
+                                505,
+                                390
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        377,
+                                        505,
+                                        390
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with metric space rescaling, this results in the performance boost (please refer to lines 1 and 6 in",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                388,
+                                505,
+                                400
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        388,
+                                        505,
+                                        400
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Table 3). Our interpretation is that the addition of the image/text classifier loss helps to reduce the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                399,
+                                505,
+                                411
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        399,
+                                        505,
+                                        411
+                                    ],
+                                    "score": 1.0,
+                                    "content": "intraclass variability in embeddings and provides for tighter clustering. However, this also leads to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                408,
+                                506,
+                                424
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        408,
+                                        506,
+                                        424
+                                    ],
+                                    "score": 1.0,
+                                    "content": "overfit on the classification task. This is accounted for by metric rescaling that enables the learnings",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                421,
+                                444,
+                                434
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        421,
+                                        444,
+                                        434
+                                    ],
+                                    "score": 1.0,
+                                    "content": "from the image/text classification task be transferred effectively into the GZSL task.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        }
+                    ],
+                    "index": 14,
+                    "bbox_fs": [
+                        104,
+                        245,
+                        507,
+                        434
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        438,
+                        505,
+                        570
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                437,
+                                505,
+                                450
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        437,
+                                        505,
+                                        450
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The comparison of the last four rows of Table 3 leads us to believe that all the proposed loss terms",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                449,
+                                507,
+                                461
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        449,
+                                        507,
+                                        461
+                                    ],
+                                    "score": 1.0,
+                                    "content": "outlined in Fig. 1 and Algorithm 1 are important for achieving the state-of-the-art performance.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                460,
+                                505,
+                                472
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        460,
+                                        380,
+                                        472
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Excluding any one of them (corresponding to the extreme values",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        380,
+                                        460,
+                                        409,
+                                        470
+                                    ],
+                                    "score": 0.84,
+                                    "content": "\\lambda = 0",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        409,
+                                        460,
+                                        414,
+                                        472
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        414,
+                                        460,
+                                        443,
+                                        470
+                                    ],
+                                    "score": 0.78,
+                                    "content": "\\lambda = 1",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        443,
+                                        460,
+                                        447,
+                                        472
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        447,
+                                        460,
+                                        477,
+                                        470
+                                    ],
+                                    "score": 0.81,
+                                    "content": "\\kappa = 0",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        477,
+                                        460,
+                                        505,
+                                        472
+                                    ],
+                                    "score": 1.0,
+                                    "content": ") leads",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                470,
+                                506,
+                                484
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        470,
+                                        506,
+                                        484
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to performance deterioration. Finally, an interesting observation can be made by comparing line",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                482,
+                                506,
+                                494
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        482,
+                                        506,
+                                        494
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3 of Table 3 with performance of algorithms in Table 1. In this case our algorithm does not use",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                493,
+                                505,
+                                505
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        493,
+                                        505,
+                                        505
+                                    ],
+                                    "score": 1.0,
+                                    "content": "any class labels and relies on training using exclusively the retrieval losses that can be calculated",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                504,
+                                505,
+                                516
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        504,
+                                        505,
+                                        516
+                                    ],
+                                    "score": 1.0,
+                                    "content": "only based on the pairwise relationships between texts and images. We can see that using this",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                514,
+                                505,
+                                527
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        514,
+                                        505,
+                                        527
+                                    ],
+                                    "score": 1.0,
+                                    "content": "type of supervision alone already results in a very high-quality representation. The representation",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                525,
+                                505,
+                                538
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        525,
+                                        505,
+                                        538
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is competitive against the latest GAN/VAE based approaches on CUB and is state-of-the-art on",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                535,
+                                506,
+                                550
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        535,
+                                        506,
+                                        550
+                                    ],
+                                    "score": 1.0,
+                                    "content": "FLOWERS. This opens up new exploration avenues showing that in the case when very fine-grained",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                547,
+                                506,
+                                561
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        547,
+                                        506,
+                                        561
+                                    ],
+                                    "score": 1.0,
+                                    "content": "modality outputs are available (image and text description pairs being a very prominent example), the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                559,
+                                486,
+                                571
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        559,
+                                        486,
+                                        571
+                                    ],
+                                    "score": 1.0,
+                                    "content": "high-quality representations may be learned without relying on manually supplied class labels.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        }
+                    ],
+                    "index": 28.5,
+                    "bbox_fs": [
+                        104,
+                        437,
+                        507,
+                        571
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        586,
+                        201,
+                        598
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                583,
+                                203,
+                                601
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        583,
+                                        203,
+                                        601
+                                    ],
+                                    "score": 1.0,
+                                    "content": "5 CONCLUSIONS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        }
+                    ],
+                    "index": 35
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        610,
+                        505,
+                        731
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                611,
+                                506,
+                                624
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        611,
+                                        506,
+                                        624
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We propose and empirically validate two improvements to the process of learning fine-grained cross-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                622,
+                                505,
+                                634
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        622,
+                                        505,
+                                        634
+                                    ],
+                                    "score": 1.0,
+                                    "content": "modal representations. First, we confirm the hypothesis that in the context of paired images and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                632,
+                                505,
+                                646
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        632,
+                                        505,
+                                        646
+                                    ],
+                                    "score": 1.0,
+                                    "content": "texts, a deep metric learning approach can be driven by an instance-based retrieval loss resulting",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                644,
+                                505,
+                                655
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        644,
+                                        505,
+                                        655
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in competitive generalized zero shot classification results. Combined with an additional class label",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                654,
+                                505,
+                                667
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        654,
+                                        505,
+                                        667
+                                    ],
+                                    "score": 1.0,
+                                    "content": "based image/text crossentropy term this results in state-of-the-art performance on two well known",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                665,
+                                505,
+                                678
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        665,
+                                        505,
+                                        678
+                                    ],
+                                    "score": 1.0,
+                                    "content": "datasets, CUB and FLOWERS. This is an interesting result demonstrating that high-quality deep",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                676,
+                                506,
+                                691
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        676,
+                                        506,
+                                        691
+                                    ],
+                                    "score": 1.0,
+                                    "content": "representations can be trained relying largely on pairwise relationships between modalities. On top",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                688,
+                                505,
+                                700
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        688,
+                                        505,
+                                        700
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of that, we propose a solution to one of the prominent problems in GZSL: classifying instances",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                699,
+                                504,
+                                711
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        699,
+                                        504,
+                                        711
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of unseen classes as seen ones. We mathematically analyze and empirically validate the method",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                710,
+                                506,
+                                722
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        710,
+                                        506,
+                                        722
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of adjusting a single scalar that transcends in its effectiveness advanced dataset augmentation and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                721,
+                                298,
+                                732
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        721,
+                                        298,
+                                        732
+                                    ],
+                                    "score": 1.0,
+                                    "content": "training approaches based on GANs and VAEs.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        }
+                    ],
+                    "index": 41,
+                    "bbox_fs": [
+                        105,
+                        611,
+                        506,
+                        732
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        81,
+                        176,
+                        93
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                82,
+                                176,
+                                95
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        82,
+                                        176,
+                                        95
+                                    ],
+                                    "score": 1.0,
+                                    "content": "REFERENCES",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        }
+                    ],
+                    "index": 0
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        100,
+                        504,
+                        123
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                99,
+                                505,
+                                113
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        99,
+                                        505,
+                                        113
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Zeynep Akata, Scott Reed, Daniel Walter, Honglak Lee, and Bernt Schiele. Evaluation of output",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                110,
+                                382,
+                                123
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        110,
+                                        382,
+                                        123
+                                    ],
+                                    "score": 1.0,
+                                    "content": "embeddings for fine-grained image classification. In CVPR, 2015.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        }
+                    ],
+                    "index": 1.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        129,
+                        503,
+                        152
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                127,
+                                505,
+                                144
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        127,
+                                        505,
+                                        144
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Zeynep Akata, Florent Perronnin, Zaid Harchaoui, and Cordelia Schmid. Label-embedding for image",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                140,
+                                233,
+                                152
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        140,
+                                        233,
+                                        152
+                                    ],
+                                    "score": 1.0,
+                                    "content": "classification. TPAMI, 2016.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        }
+                    ],
+                    "index": 3.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        158,
+                        503,
+                        182
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                158,
+                                505,
+                                172
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        158,
+                                        505,
+                                        172
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Yuval Atzmon and Gal Chechik. Probabilistic AND-OR attribute grouping for zero-shot learning. In",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                169,
+                                163,
+                                182
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        169,
+                                        163,
+                                        182
+                                    ],
+                                    "score": 1.0,
+                                    "content": "UAI, 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        }
+                    ],
+                    "index": 5.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        188,
+                        504,
+                        211
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                186,
+                                505,
+                                203
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        186,
+                                        505,
+                                        203
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Yuval Atzmon and Gal Chechik. Adaptive confidence smoothing for generalized zero-shot learning.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                199,
+                                182,
+                                211
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        199,
+                                        182,
+                                        211
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In CVPR, 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        }
+                    ],
+                    "index": 7.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        217,
+                        505,
+                        241
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                218,
+                                506,
+                                231
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        218,
+                                        506,
+                                        231
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Wei-Lun Chao, Soravit Changpinyo, Boqing Gong, and Fei Sha. An empirical study and analysis of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                229,
+                                503,
+                                242
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        229,
+                                        503,
+                                        242
+                                    ],
+                                    "score": 1.0,
+                                    "content": "generalized zero-shot learning for object recognition in the wild. In ECCV (2), pp. 52–68, 2016.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        }
+                    ],
+                    "index": 9.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        247,
+                        503,
+                        271
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                246,
+                                505,
+                                261
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        246,
+                                        505,
+                                        261
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Debasmit Das and C Lee. Zero-shot image recognition using relational matching, adaptation and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                258,
+                                411,
+                                271
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        258,
+                                        411,
+                                        271
+                                    ],
+                                    "score": 1.0,
+                                    "content": "calibration. In International Joint Conference on Neural Networks, 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        }
+                    ],
+                    "index": 11.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        276,
+                        504,
+                        300
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                275,
+                                505,
+                                291
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        275,
+                                        505,
+                                        291
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Rafael Felix, Vijay Kumar B G, Ian Reid, and Gustavo Carneiro. Multi-modal cycle-consistent",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                288,
+                                309,
+                                300
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        288,
+                                        309,
+                                        300
+                                    ],
+                                    "score": 1.0,
+                                    "content": "generalized zero-shot learning. In ECCV, 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        }
+                    ],
+                    "index": 13.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        306,
+                        505,
+                        340
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                306,
+                                506,
+                                320
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        306,
+                                        506,
+                                        320
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Andrea Frome, Greg S Corrado, Jon Shlens, Samy Bengio, Jeff Dean, Marc Aurelio Ranzato, and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                317,
+                                506,
+                                330
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        317,
+                                        506,
+                                        330
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Tomas Mikolov. DeViSE: A deep visual-semantic embedding model. In NIPS, pp. 2121–2129,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                327,
+                                144,
+                                340
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        327,
+                                        144,
+                                        340
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2013.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        }
+                    ],
+                    "index": 16
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        347,
+                        505,
+                        370
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                346,
+                                506,
+                                360
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        346,
+                                        506,
+                                        360
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Ian Goodfellow, Jean Pouget-Abadie, Mehdi Mirza, Bing Xu, David Warde-Farley, Sherjil Ozair,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                358,
+                                507,
+                                371
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        358,
+                                        507,
+                                        371
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Aaron Courville, and Yoshua Bengio. Generative adversarial nets. In NIPS, pp. 2672–2680, 2014.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        }
+                    ],
+                    "index": 18.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        376,
+                        505,
+                        399
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                376,
+                                505,
+                                389
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        376,
+                                        505,
+                                        389
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Arthur Gretton, Karsten M Borgwardt, Malte Rasch, Bernhard Schölkopf, and Alex J Smola. A",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                387,
+                                356,
+                                400
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        387,
+                                        356,
+                                        400
+                                    ],
+                                    "score": 1.0,
+                                    "content": "kernel method for the two-sample-problem. In NIPS, 2007.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        }
+                    ],
+                    "index": 20.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        406,
+                        504,
+                        429
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                405,
+                                505,
+                                420
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        405,
+                                        505,
+                                        420
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Geoffrey Hinton, Oriol Vinyals, and Jeffrey Dean. Distilling the knowledge in a neural network. In",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                114,
+                                416,
+                                209,
+                                429
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        114,
+                                        416,
+                                        209,
+                                        429
+                                    ],
+                                    "score": 1.0,
+                                    "content": "NIPS Workshop, 2015.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        }
+                    ],
+                    "index": 22.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        435,
+                        505,
+                        459
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                107,
+                                435,
+                                506,
+                                448
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        435,
+                                        506,
+                                        448
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Yao-Hung Hubert Tsai, Liang-Kang Huang, and Ruslan Salakhutdinov. Learning robust visual-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                447,
+                                274,
+                                459
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        447,
+                                        274,
+                                        459
+                                    ],
+                                    "score": 1.0,
+                                    "content": "semantic embeddings. In CVPR, 2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        }
+                    ],
+                    "index": 24.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        465,
+                        461,
+                        477
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                464,
+                                462,
+                                479
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        464,
+                                        462,
+                                        479
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Diederik P. Kingma and Max Welling. Auto-encoding variational bayes. In ICLR, 2014.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        }
+                    ],
+                    "index": 26
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        484,
+                        505,
+                        507
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                484,
+                                505,
+                                497
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        484,
+                                        505,
+                                        497
+                                    ],
+                                    "score": 1.0,
+                                    "content": "C. H. Lampert, H. Nickisch, and S. Harmeling. Attribute-based classification for zero-shot visual",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                495,
+                                264,
+                                507
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        495,
+                                        264,
+                                        507
+                                    ],
+                                    "score": 1.0,
+                                    "content": "object categorization. TPAMI, 2014.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        }
+                    ],
+                    "index": 27.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        513,
+                        505,
+                        536
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                513,
+                                506,
+                                526
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        513,
+                                        506,
+                                        526
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Hugo Larochelle, Dumitru Erhan, and Yoshua Bengio. Zero-data learning of new tasks. In AAAI,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                114,
+                                523,
+                                143,
+                                537
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        114,
+                                        523,
+                                        143,
+                                        537
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2008.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        }
+                    ],
+                    "index": 29.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        542,
+                        436,
+                        555
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                542,
+                                435,
+                                556
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        542,
+                                        435,
+                                        556
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Yann LeCun, Yoshua Bengio, and Geoffrey Hinton. Deep learning. Nature, 2015.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        }
+                    ],
+                    "index": 31
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        561,
+                        501,
+                        585
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                560,
+                                504,
+                                575
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        560,
+                                        504,
+                                        575
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Shichen Liu, Mingsheng Long, Jianmin Wang, and Michael I Jordan. Generalized zero-shot learning",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                572,
+                                303,
+                                585
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        572,
+                                        303,
+                                        585
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with deep calibration network. In NIPS, 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        }
+                    ],
+                    "index": 32.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        591,
+                        504,
+                        614
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                591,
+                                505,
+                                604
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        591,
+                                        505,
+                                        604
+                                    ],
+                                    "score": 1.0,
+                                    "content": "M. Nilsback and A. Zisserman. Automated flower classification over a large number of classes. In",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                602,
+                                462,
+                                615
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        602,
+                                        462,
+                                        615
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2008 Sixth Indian Conference on Computer Vision, Graphics Image Processing, 2008.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        }
+                    ],
+                    "index": 34.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        620,
+                        504,
+                        644
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                620,
+                                505,
+                                633
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        620,
+                                        505,
+                                        633
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Mark Palatucci, Dean Pomerleau, Geoffrey E Hinton, and Tom M Mitchell. Zero-shot learning with",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                632,
+                                273,
+                                644
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        632,
+                                        273,
+                                        644
+                                    ],
+                                    "score": 1.0,
+                                    "content": "semantic output codes. In NIPS, 2009.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        }
+                    ],
+                    "index": 36.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        650,
+                        504,
+                        673
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                650,
+                                505,
+                                663
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        650,
+                                        505,
+                                        663
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Genevieve Patterson, Chen Xu, Hang Su, and James Hays. The sun attribute database: Beyond",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                661,
+                                340,
+                                674
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        661,
+                                        340,
+                                        674
+                                    ],
+                                    "score": 1.0,
+                                    "content": "categories for deeper scene understanding. IJCV, 2014.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        }
+                    ],
+                    "index": 38.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        680,
+                        503,
+                        703
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                680,
+                                505,
+                                693
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        680,
+                                        505,
+                                        693
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Scott E. Reed, Zeynep Akata, Honglak Lee, and Bernt Schiele. Learning deep representations of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                690,
+                                313,
+                                703
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        690,
+                                        313,
+                                        703
+                                    ],
+                                    "score": 1.0,
+                                    "content": "fine-grained visual descriptions. In CVPR, 2016.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        }
+                    ],
+                    "index": 40.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        709,
+                        502,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                709,
+                                505,
+                                722
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        709,
+                                        505,
+                                        722
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Edgar Schönfeld, Sayna Ebrahimi, Samarth Sinha, Trevor Darrell, and Zeynep Akata. Generalized",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                720,
+                                432,
+                                732
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        720,
+                                        432,
+                                        732
+                                    ],
+                                    "score": 1.0,
+                                    "content": "zero-and few-shot learning via aligned variational autoencoders. CVPR, 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        }
+                    ],
+                    "index": 42.5
+                }
+            ],
+            "page_idx": 8,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        107,
+                        27,
+                        308,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                107,
+                                26,
+                                308,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        26,
+                                        308,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Under review as a conference paper at ICLR 2020",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        302,
+                        751,
+                        309,
+                        759
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                302,
+                                751,
+                                309,
+                                762
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        302,
+                                        751,
+                                        309,
+                                        762
+                                    ],
+                                    "score": 1.0,
+                                    "content": "9",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        81,
+                        176,
+                        93
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                82,
+                                176,
+                                95
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        82,
+                                        176,
+                                        95
+                                    ],
+                                    "score": 1.0,
+                                    "content": "REFERENCES",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        }
+                    ],
+                    "index": 0
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        100,
+                        504,
+                        123
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                99,
+                                505,
+                                113
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        99,
+                                        505,
+                                        113
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Zeynep Akata, Scott Reed, Daniel Walter, Honglak Lee, and Bernt Schiele. Evaluation of output",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                110,
+                                382,
+                                123
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        110,
+                                        382,
+                                        123
+                                    ],
+                                    "score": 1.0,
+                                    "content": "embeddings for fine-grained image classification. In CVPR, 2015.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        }
+                    ],
+                    "index": 1.5,
+                    "bbox_fs": [
+                        105,
+                        99,
+                        505,
+                        123
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        129,
+                        503,
+                        152
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                127,
+                                505,
+                                144
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        127,
+                                        505,
+                                        144
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Zeynep Akata, Florent Perronnin, Zaid Harchaoui, and Cordelia Schmid. Label-embedding for image",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                140,
+                                233,
+                                152
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        140,
+                                        233,
+                                        152
+                                    ],
+                                    "score": 1.0,
+                                    "content": "classification. TPAMI, 2016.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        }
+                    ],
+                    "index": 3.5,
+                    "bbox_fs": [
+                        104,
+                        127,
+                        505,
+                        152
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        158,
+                        503,
+                        182
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                158,
+                                505,
+                                172
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        158,
+                                        505,
+                                        172
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Yuval Atzmon and Gal Chechik. Probabilistic AND-OR attribute grouping for zero-shot learning. In",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                169,
+                                163,
+                                182
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        169,
+                                        163,
+                                        182
+                                    ],
+                                    "score": 1.0,
+                                    "content": "UAI, 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        }
+                    ],
+                    "index": 5.5,
+                    "bbox_fs": [
+                        106,
+                        158,
+                        505,
+                        182
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        188,
+                        504,
+                        211
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                186,
+                                505,
+                                203
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        186,
+                                        505,
+                                        203
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Yuval Atzmon and Gal Chechik. Adaptive confidence smoothing for generalized zero-shot learning.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                199,
+                                182,
+                                211
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        199,
+                                        182,
+                                        211
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In CVPR, 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        }
+                    ],
+                    "index": 7.5,
+                    "bbox_fs": [
+                        106,
+                        186,
+                        505,
+                        211
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        217,
+                        505,
+                        241
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                218,
+                                506,
+                                231
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        218,
+                                        506,
+                                        231
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Wei-Lun Chao, Soravit Changpinyo, Boqing Gong, and Fei Sha. An empirical study and analysis of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                229,
+                                503,
+                                242
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        229,
+                                        503,
+                                        242
+                                    ],
+                                    "score": 1.0,
+                                    "content": "generalized zero-shot learning for object recognition in the wild. In ECCV (2), pp. 52–68, 2016.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        }
+                    ],
+                    "index": 9.5,
+                    "bbox_fs": [
+                        106,
+                        218,
+                        506,
+                        242
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        247,
+                        503,
+                        271
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                246,
+                                505,
+                                261
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        246,
+                                        505,
+                                        261
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Debasmit Das and C Lee. Zero-shot image recognition using relational matching, adaptation and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                258,
+                                411,
+                                271
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        258,
+                                        411,
+                                        271
+                                    ],
+                                    "score": 1.0,
+                                    "content": "calibration. In International Joint Conference on Neural Networks, 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        }
+                    ],
+                    "index": 11.5,
+                    "bbox_fs": [
+                        105,
+                        246,
+                        505,
+                        271
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        276,
+                        504,
+                        300
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                275,
+                                505,
+                                291
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        275,
+                                        505,
+                                        291
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Rafael Felix, Vijay Kumar B G, Ian Reid, and Gustavo Carneiro. Multi-modal cycle-consistent",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                288,
+                                309,
+                                300
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        288,
+                                        309,
+                                        300
+                                    ],
+                                    "score": 1.0,
+                                    "content": "generalized zero-shot learning. In ECCV, 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        }
+                    ],
+                    "index": 13.5,
+                    "bbox_fs": [
+                        105,
+                        275,
+                        505,
+                        300
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        306,
+                        505,
+                        340
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                306,
+                                506,
+                                320
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        306,
+                                        506,
+                                        320
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Andrea Frome, Greg S Corrado, Jon Shlens, Samy Bengio, Jeff Dean, Marc Aurelio Ranzato, and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                317,
+                                506,
+                                330
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        317,
+                                        506,
+                                        330
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Tomas Mikolov. DeViSE: A deep visual-semantic embedding model. In NIPS, pp. 2121–2129,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                327,
+                                144,
+                                340
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        327,
+                                        144,
+                                        340
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2013.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        }
+                    ],
+                    "index": 16,
+                    "bbox_fs": [
+                        106,
+                        306,
+                        506,
+                        340
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        347,
+                        505,
+                        370
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                346,
+                                506,
+                                360
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        346,
+                                        506,
+                                        360
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Ian Goodfellow, Jean Pouget-Abadie, Mehdi Mirza, Bing Xu, David Warde-Farley, Sherjil Ozair,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                358,
+                                507,
+                                371
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        358,
+                                        507,
+                                        371
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Aaron Courville, and Yoshua Bengio. Generative adversarial nets. In NIPS, pp. 2672–2680, 2014.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        }
+                    ],
+                    "index": 18.5,
+                    "bbox_fs": [
+                        105,
+                        346,
+                        507,
+                        371
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        376,
+                        505,
+                        399
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                376,
+                                505,
+                                389
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        376,
+                                        505,
+                                        389
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Arthur Gretton, Karsten M Borgwardt, Malte Rasch, Bernhard Schölkopf, and Alex J Smola. A",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                387,
+                                356,
+                                400
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        387,
+                                        356,
+                                        400
+                                    ],
+                                    "score": 1.0,
+                                    "content": "kernel method for the two-sample-problem. In NIPS, 2007.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        }
+                    ],
+                    "index": 20.5,
+                    "bbox_fs": [
+                        106,
+                        376,
+                        505,
+                        400
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        406,
+                        504,
+                        429
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                405,
+                                505,
+                                420
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        405,
+                                        505,
+                                        420
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Geoffrey Hinton, Oriol Vinyals, and Jeffrey Dean. Distilling the knowledge in a neural network. In",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                114,
+                                416,
+                                209,
+                                429
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        114,
+                                        416,
+                                        209,
+                                        429
+                                    ],
+                                    "score": 1.0,
+                                    "content": "NIPS Workshop, 2015.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        }
+                    ],
+                    "index": 22.5,
+                    "bbox_fs": [
+                        105,
+                        405,
+                        505,
+                        429
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        435,
+                        505,
+                        459
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                107,
+                                435,
+                                506,
+                                448
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        435,
+                                        506,
+                                        448
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Yao-Hung Hubert Tsai, Liang-Kang Huang, and Ruslan Salakhutdinov. Learning robust visual-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                447,
+                                274,
+                                459
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        447,
+                                        274,
+                                        459
+                                    ],
+                                    "score": 1.0,
+                                    "content": "semantic embeddings. In CVPR, 2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        }
+                    ],
+                    "index": 24.5,
+                    "bbox_fs": [
+                        107,
+                        435,
+                        506,
+                        459
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        465,
+                        461,
+                        477
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                464,
+                                462,
+                                479
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        464,
+                                        462,
+                                        479
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Diederik P. Kingma and Max Welling. Auto-encoding variational bayes. In ICLR, 2014.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        }
+                    ],
+                    "index": 26,
+                    "bbox_fs": [
+                        105,
+                        464,
+                        462,
+                        479
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        484,
+                        505,
+                        507
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                484,
+                                505,
+                                497
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        484,
+                                        505,
+                                        497
+                                    ],
+                                    "score": 1.0,
+                                    "content": "C. H. Lampert, H. Nickisch, and S. Harmeling. Attribute-based classification for zero-shot visual",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                495,
+                                264,
+                                507
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        495,
+                                        264,
+                                        507
+                                    ],
+                                    "score": 1.0,
+                                    "content": "object categorization. TPAMI, 2014.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        }
+                    ],
+                    "index": 27.5,
+                    "bbox_fs": [
+                        106,
+                        484,
+                        505,
+                        507
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        513,
+                        505,
+                        536
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                513,
+                                506,
+                                526
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        513,
+                                        506,
+                                        526
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Hugo Larochelle, Dumitru Erhan, and Yoshua Bengio. Zero-data learning of new tasks. In AAAI,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                114,
+                                523,
+                                143,
+                                537
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        114,
+                                        523,
+                                        143,
+                                        537
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2008.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        }
+                    ],
+                    "index": 29.5,
+                    "bbox_fs": [
+                        106,
+                        513,
+                        506,
+                        537
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        542,
+                        436,
+                        555
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                542,
+                                435,
+                                556
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        542,
+                                        435,
+                                        556
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Yann LeCun, Yoshua Bengio, and Geoffrey Hinton. Deep learning. Nature, 2015.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        }
+                    ],
+                    "index": 31,
+                    "bbox_fs": [
+                        106,
+                        542,
+                        435,
+                        556
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        561,
+                        501,
+                        585
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                560,
+                                504,
+                                575
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        560,
+                                        504,
+                                        575
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Shichen Liu, Mingsheng Long, Jianmin Wang, and Michael I Jordan. Generalized zero-shot learning",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                572,
+                                303,
+                                585
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        572,
+                                        303,
+                                        585
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with deep calibration network. In NIPS, 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        }
+                    ],
+                    "index": 32.5,
+                    "bbox_fs": [
+                        105,
+                        560,
+                        504,
+                        585
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        591,
+                        504,
+                        614
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                591,
+                                505,
+                                604
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        591,
+                                        505,
+                                        604
+                                    ],
+                                    "score": 1.0,
+                                    "content": "M. Nilsback and A. Zisserman. Automated flower classification over a large number of classes. In",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                602,
+                                462,
+                                615
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        602,
+                                        462,
+                                        615
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2008 Sixth Indian Conference on Computer Vision, Graphics Image Processing, 2008.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        }
+                    ],
+                    "index": 34.5,
+                    "bbox_fs": [
+                        105,
+                        591,
+                        505,
+                        615
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        620,
+                        504,
+                        644
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                620,
+                                505,
+                                633
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        620,
+                                        505,
+                                        633
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Mark Palatucci, Dean Pomerleau, Geoffrey E Hinton, and Tom M Mitchell. Zero-shot learning with",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                632,
+                                273,
+                                644
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        632,
+                                        273,
+                                        644
+                                    ],
+                                    "score": 1.0,
+                                    "content": "semantic output codes. In NIPS, 2009.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        }
+                    ],
+                    "index": 36.5,
+                    "bbox_fs": [
+                        105,
+                        620,
+                        505,
+                        644
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        650,
+                        504,
+                        673
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                650,
+                                505,
+                                663
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        650,
+                                        505,
+                                        663
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Genevieve Patterson, Chen Xu, Hang Su, and James Hays. The sun attribute database: Beyond",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                661,
+                                340,
+                                674
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        661,
+                                        340,
+                                        674
+                                    ],
+                                    "score": 1.0,
+                                    "content": "categories for deeper scene understanding. IJCV, 2014.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        }
+                    ],
+                    "index": 38.5,
+                    "bbox_fs": [
+                        106,
+                        650,
+                        505,
+                        674
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        680,
+                        503,
+                        703
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                680,
+                                505,
+                                693
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        680,
+                                        505,
+                                        693
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Scott E. Reed, Zeynep Akata, Honglak Lee, and Bernt Schiele. Learning deep representations of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                690,
+                                313,
+                                703
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        690,
+                                        313,
+                                        703
+                                    ],
+                                    "score": 1.0,
+                                    "content": "fine-grained visual descriptions. In CVPR, 2016.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        }
+                    ],
+                    "index": 40.5,
+                    "bbox_fs": [
+                        106,
+                        680,
+                        505,
+                        703
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        709,
+                        502,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                709,
+                                505,
+                                722
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        709,
+                                        505,
+                                        722
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Edgar Schönfeld, Sayna Ebrahimi, Samarth Sinha, Trevor Darrell, and Zeynep Akata. Generalized",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                720,
+                                432,
+                                732
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        720,
+                                        432,
+                                        732
+                                    ],
+                                    "score": 1.0,
+                                    "content": "zero-and few-shot learning via aligned variational autoencoders. CVPR, 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        }
+                    ],
+                    "index": 42.5,
+                    "bbox_fs": [
+                        106,
+                        709,
+                        505,
+                        732
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        81,
+                        507,
+                        414
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                82,
+                                505,
+                                95
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        82,
+                                        505,
+                                        95
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Richard Socher, Milind Ganjoo, Christopher D Manning, and Andrew Ng. Zero-shot learning through",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                94,
+                                265,
+                                105
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        94,
+                                        265,
+                                        105
+                                    ],
+                                    "score": 1.0,
+                                    "content": "cross-modal transfer. In NIPS, 2013.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                110,
+                                505,
+                                124
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        110,
+                                        505,
+                                        124
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Vinay Kumar Verma, Gundeep Arora, Ashish Mishra, and Piyush Rai. Generalized zero-shot learning",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                122,
+                                289,
+                                133
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        122,
+                                        289,
+                                        133
+                                    ],
+                                    "score": 1.0,
+                                    "content": "via synthesized examples. In CVPR, 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                138,
+                                506,
+                                152
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        138,
+                                        506,
+                                        152
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Wei Wang, Vincent W. Zheng, Han Yu, and Chunyan Miao. A survey of zero-shot learning: Settings,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                117,
+                                150,
+                                387,
+                                162
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        117,
+                                        150,
+                                        387,
+                                        162
+                                    ],
+                                    "score": 1.0,
+                                    "content": "methods, and applications. ACM Trans. Intell. Syst. Technol., 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                166,
+                                429,
+                                180
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        166,
+                                        429,
+                                        180
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Yaqing Wang and Quanming Yao. Few-shot learning: A survey. In arXiv, 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                182,
+                                505,
+                                196
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        182,
+                                        505,
+                                        196
+                                    ],
+                                    "score": 1.0,
+                                    "content": "P. Welinder, S. Branson, T. Mita, C. Wah, F. Schroff, S. Belongie, and P. Perona. Caltech-UCSD",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                194,
+                                395,
+                                207
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        194,
+                                        395,
+                                        207
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Birds 200. Technical report, California Institute of Technology, 2010.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                211,
+                                506,
+                                224
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        211,
+                                        506,
+                                        224
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Yongqin Xian, Zeynep Akata, Gaurav Sharma, Quynh N. Nguyen, Matthias Hein, and Bernt Schiele.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                222,
+                                371,
+                                234
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        222,
+                                        371,
+                                        234
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Latent embeddings for zero-shot classification. In CVPR, 2016.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                239,
+                                507,
+                                252
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        239,
+                                        507,
+                                        252
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Yongqin Xian, Christoph H. Lampert, Bernt Schiele, and Zeynep Akata. Pretrained CUB features,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                249,
+                                468,
+                                263
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        249,
+                                        468,
+                                        263
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2018a. URL http://datasets.d2.mpi-inf.mpg.de/xian/xlsa17.zip.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                266,
+                                505,
+                                280
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        266,
+                                        505,
+                                        280
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Yongqin Xian, Christoph H. Lampert, Bernt Schiele, and Zeynep Akata. Pretrained FLOWERS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                278,
+                                508,
+                                291
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        278,
+                                        508,
+                                        291
+                                    ],
+                                    "score": 1.0,
+                                    "content": "features, 2018b. URL http://datasets.d2.mpi-inf.mpg.de/xian/cvpr18xian.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                117,
+                                289,
+                                140,
+                                302
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        117,
+                                        289,
+                                        140,
+                                        302
+                                    ],
+                                    "score": 1.0,
+                                    "content": "zip.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                306,
+                                506,
+                                320
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        306,
+                                        506,
+                                        320
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Yongqin Xian, H. Christoph Lampert, Bernt Schiele, and Zeynep Akata. Zero-shot learning: A",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                317,
+                                422,
+                                330
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        317,
+                                        422,
+                                        330
+                                    ],
+                                    "score": 1.0,
+                                    "content": "comprehensive evaluation of the good, the bad and the ugly. TPAMI, 2018c.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                333,
+                                505,
+                                347
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        333,
+                                        505,
+                                        347
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Yongqin Xian, Tobias Lorenz, Bernt Schiele, and Zeynep Akata. Feature generating networks for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                345,
+                                264,
+                                357
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        345,
+                                        264,
+                                        357
+                                    ],
+                                    "score": 1.0,
+                                    "content": "zero-shot learning. In CVPR, 2018d.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                361,
+                                505,
+                                375
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        361,
+                                        505,
+                                        375
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Yongqin Xian, Saurabh Sharma, Bernt Schiele, and Zeynep Akata. f-vaegan-d2: A feature generating",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                373,
+                                307,
+                                385
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        373,
+                                        307,
+                                        385
+                                    ],
+                                    "score": 1.0,
+                                    "content": "framework for any-shot learning. CVPR, 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                389,
+                                505,
+                                403
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        389,
+                                        451,
+                                        403
+                                    ],
+                                    "score": 1.0,
+                                    "content": "H. Zhang, T. Xu, H. Li, S. Zhang, X. Wang, X. Huang, and D. N. Metaxas. StackGAN",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        451,
+                                        390,
+                                        463,
+                                        400
+                                    ],
+                                    "score": 0.4,
+                                    "content": "^ { + + }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        464,
+                                        389,
+                                        505,
+                                        403
+                                    ],
+                                    "score": 1.0,
+                                    "content": ": Realistic",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                401,
+                                420,
+                                414
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        401,
+                                        420,
+                                        414
+                                    ],
+                                    "score": 1.0,
+                                    "content": "image synthesis with stacked generative adversarial networks. PAMI, 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        }
+                    ],
+                    "index": 11.5
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        429,
+                        299,
+                        442
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                428,
+                                299,
+                                444
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        428,
+                                        299,
+                                        444
+                                    ],
+                                    "score": 1.0,
+                                    "content": "A THE ANALYSIS OF ERROR RATES",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        }
+                    ],
+                    "index": 24
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        452,
+                        505,
+                        477
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                451,
+                                505,
+                                467
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        451,
+                                        169,
+                                        467
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We show that",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        169,
+                                        453,
+                                        404,
+                                        465
+                                    ],
+                                    "score": 0.86,
+                                    "content": "\\mathbb { P } \\{ \\widehat { y } ~ \\in ~ \\mathcal { Y } ^ { t r } | y _ { v } ~ \\in ~ \\mathcal { Y } ^ { t s } \\} ~ \\geq ~ \\mathbb { P } \\{ \\widehat { y } _ { \\alpha } ~ \\in ~ \\mathcal { Y } ^ { t r } | y _ { v } ~ \\in ~ \\mathcal { Y } ^ { t s } \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        404,
+                                        451,
+                                        476,
+                                        467
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Let us define",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        476,
+                                        453,
+                                        505,
+                                        466
+                                    ],
+                                    "score": 0.85,
+                                    "content": "\\delta _ { t r } \\equiv",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                107,
+                                464,
+                                483,
+                                479
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        465,
+                                        195,
+                                        477
+                                    ],
+                                    "score": 0.88,
+                                    "content": "\\begin{array} { r } { \\operatorname* { m i n } _ { y \\in \\mathcal { y } ^ { t r } } d ( \\mathbf { z } _ { v } , \\mathbf { p } ( y ) ) } \\end{array}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        195,
+                                        464,
+                                        213,
+                                        479
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        214,
+                                        466,
+                                        326,
+                                        477
+                                    ],
+                                    "score": 0.87,
+                                    "content": "\\begin{array} { r } { \\delta _ { t s } \\equiv \\operatorname* { m i n } _ { y \\in \\mathcal { V } ^ { t s } } d ( \\mathbf { z } _ { v } , \\mathbf { p } ( y ) ) } \\end{array}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        326,
+                                        464,
+                                        483,
+                                        479
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", then Equation (7) can be rewritten as:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        }
+                    ],
+                    "index": 25.5
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        183,
+                        479,
+                        425,
+                        492
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                183,
+                                479,
+                                425,
+                                492
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        183,
+                                        479,
+                                        425,
+                                        492
+                                    ],
+                                    "score": 0.81,
+                                    "content": "\\mathbb { P } \\{ \\widehat { y } _ { \\alpha } \\in \\mathcal { V } ^ { t r } | y _ { v } \\in \\mathcal { V } ^ { t s } \\} = \\mathbb { P } \\left\\{ ( 1 + \\alpha ) \\delta _ { t r } < \\delta _ { t s } | y _ { v } \\in \\mathcal { V } ^ { t s } \\right\\} \\mathrm { ~ . ~ }",
+                                    "type": "interline_equation",
+                                    "image_path": "d406a9252d80daf106d9a4cfc77b0b1de57e941418018b7e6ea684fcf96c68c2.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 27,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                183,
+                                479,
+                                425,
+                                492
+                            ],
+                            "spans": [],
+                            "index": 27
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        493,
+                        423,
+                        504
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                492,
+                                422,
+                                506
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        492,
+                                        265,
+                                        506
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Let us consider the probability of event",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        266,
+                                        493,
+                                        303,
+                                        504
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\delta _ { t r } < \\delta _ { t s }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        303,
+                                        492,
+                                        422,
+                                        506
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and decompose it as follows:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        }
+                    ],
+                    "index": 28
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        131,
+                        503,
+                        473,
+                        582
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                131,
+                                503,
+                                473,
+                                582
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        131,
+                                        503,
+                                        473,
+                                        582
+                                    ],
+                                    "score": 0.67,
+                                    "content": "\\begin{array} { r l } & { \\mathbb { P } \\left\\{ \\delta _ { t r } < \\delta _ { t s } \\middle | y _ { v } \\in \\mathcal { Y } ^ { t s } \\right\\} = \\mathbb { P } \\left\\{ ( 1 + \\alpha ) \\delta _ { t r } < ( 1 + \\alpha ) \\delta _ { t s } | y _ { v } \\in \\mathcal { Y } ^ { t s } \\right\\} } \\\\ & { = \\mathbb { P } \\left\\{ ( 1 + \\alpha ) \\delta _ { t r } < \\delta _ { t s } \\cup \\delta _ { t s } \\leq ( 1 + \\alpha ) \\delta _ { t r } < ( 1 + \\alpha ) \\delta _ { t s } | y _ { v } \\in \\mathcal { Y } ^ { t s } \\right\\} } \\\\ & { = \\mathbb { P } \\left\\{ ( 1 + \\alpha ) \\delta _ { t r } < \\delta _ { t s } | y _ { v } \\in \\mathcal { Y } ^ { t s } \\right\\} + \\mathbb { P } \\left\\{ \\delta _ { t s } \\leq ( 1 + \\alpha ) \\delta _ { t r } < ( 1 + \\alpha ) \\delta _ { t s } | y _ { v } \\in \\mathcal { Y } ^ { t s } \\right\\} } \\\\ & { - \\mathbb { P } \\left\\{ ( 1 + \\alpha ) \\delta _ { t r } < \\delta _ { t s } \\cap \\delta _ { t s } \\leq ( 1 + \\alpha ) \\delta _ { t r } < ( 1 + \\alpha ) \\delta _ { t s } | y _ { v } \\in \\mathcal { Y } ^ { t s } \\right\\} } \\\\ & { = \\mathbb { P } \\left\\{ ( 1 + \\alpha ) \\delta _ { t r } < \\delta _ { t s } | y _ { v } \\in \\mathcal { Y } ^ { t s } \\right\\} + \\mathbb { P } \\left\\{ \\delta _ { t s } \\leq ( 1 + \\alpha ) \\delta _ { t r } < ( 1 + \\alpha ) \\delta _ { t s } | y _ { v } \\in \\mathcal { Y } ^ { t s } \\right\\} } \\end{array}",
+                                    "type": "interline_equation",
+                                    "image_path": "1125cd16d5d95a1f8c8e948f94949c974d4895b575a12f7fa163ec98d67420e9.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 30,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                131,
+                                503,
+                                473,
+                                529.3333333333334
+                            ],
+                            "spans": [],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                131,
+                                529.3333333333334,
+                                473,
+                                555.6666666666667
+                            ],
+                            "spans": [],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                131,
+                                555.6666666666667,
+                                473,
+                                582.0000000000001
+                            ],
+                            "spans": [],
+                            "index": 31
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        582,
+                        507,
+                        604
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                580,
+                                507,
+                                595
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        580,
+                                        464,
+                                        595
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The transitions are based on the relationship between probabilities of arbitrary events",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        465,
+                                        582,
+                                        473,
+                                        591
+                                    ],
+                                    "score": 0.82,
+                                    "content": "A",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        474,
+                                        580,
+                                        493,
+                                        595
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        493,
+                                        582,
+                                        502,
+                                        591
+                                    ],
+                                    "score": 0.83,
+                                    "content": "B",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        503,
+                                        580,
+                                        507,
+                                        595
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                592,
+                                483,
+                                606
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        592,
+                                        278,
+                                        605
+                                    ],
+                                    "score": 0.88,
+                                    "content": "\\mathbb { P } \\{ A \\cup B \\} = \\mathbb { P } \\{ A \\} + \\mathbb { P } \\{ B \\} - \\mathbb { P } \\{ A \\cap B \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        279,
+                                        592,
+                                        344,
+                                        606
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", and in our case",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        345,
+                                        592,
+                                        407,
+                                        605
+                                    ],
+                                    "score": 0.92,
+                                    "content": "{ \\mathbb { P } } \\{ A \\cap B \\} = 0",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        407,
+                                        592,
+                                        483,
+                                        606
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". This implies that:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        }
+                    ],
+                    "index": 32.5
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        128,
+                        605,
+                        482,
+                        648
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                128,
+                                605,
+                                482,
+                                648
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        128,
+                                        605,
+                                        482,
+                                        648
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\begin{array} { r l r } {  { \\mathbb { P } \\{ \\widehat { y } _ { \\alpha } \\in \\mathcal { Y } ^ { t r } | y _ { v } \\in \\mathcal { Y } ^ { t s } \\} = \\mathbb { P } \\{ \\widehat { y } \\in \\mathcal { Y } ^ { t r } | y _ { v } \\in \\mathcal { Y } ^ { t s } \\} - \\mathbb { P } \\{ \\frac { \\delta _ { t s } } { ( 1 + \\alpha ) } \\leq \\delta _ { t r } < \\delta _ { t s } | y _ { v } \\in \\mathcal { Y } ^ { t s } \\} } } \\\\ & { } & { \\leq \\mathbb { P } \\{ \\widehat { y } \\in \\mathcal { Y } ^ { t r } | y _ { v } \\in \\mathcal { Y } ^ { t s } \\} . } \\end{array}",
+                                    "type": "interline_equation",
+                                    "image_path": "c82fe815b758df297244b22978a323eab8837f91f51cbd01bee778cf0f808da0.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 35,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                128,
+                                605,
+                                482,
+                                619.3333333333334
+                            ],
+                            "spans": [],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                128,
+                                619.3333333333334,
+                                482,
+                                633.6666666666667
+                            ],
+                            "spans": [],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                128,
+                                633.6666666666667,
+                                482,
+                                648.0000000000001
+                            ],
+                            "spans": [],
+                            "index": 36
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        652,
+                        506,
+                        730
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                653,
+                                505,
+                                665
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        653,
+                                        285,
+                                        665
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We have just shown that for a non-negative",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        286,
+                                        655,
+                                        294,
+                                        663
+                                    ],
+                                    "score": 0.8,
+                                    "content": "\\alpha",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        294,
+                                        653,
+                                        505,
+                                        665
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the probability of misclassifying an image from an",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                664,
+                                505,
+                                676
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        664,
+                                        404,
+                                        676
+                                    ],
+                                    "score": 1.0,
+                                    "content": "unseen class as one of the seen classes is smaller for the decision rule",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        405,
+                                        664,
+                                        416,
+                                        675
+                                    ],
+                                    "score": 0.88,
+                                    "content": "\\widehat { y } _ { \\alpha }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        417,
+                                        664,
+                                        505,
+                                        676
+                                    ],
+                                    "score": 1.0,
+                                    "content": "than for the original",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                673,
+                                506,
+                                688
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        673,
+                                        162,
+                                        688
+                                    ],
+                                    "score": 1.0,
+                                    "content": "decision rule",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        162,
+                                        675,
+                                        169,
+                                        686
+                                    ],
+                                    "score": 0.81,
+                                    "content": "\\widehat { y }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        170,
+                                        673,
+                                        365,
+                                        688
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". In fact, we can make a stronger claim. Since",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        366,
+                                        675,
+                                        378,
+                                        686
+                                    ],
+                                    "score": 0.89,
+                                    "content": "\\delta _ { t s }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        379,
+                                        673,
+                                        398,
+                                        688
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        398,
+                                        675,
+                                        411,
+                                        686
+                                    ],
+                                    "score": 0.9,
+                                    "content": "\\delta _ { t r }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        411,
+                                        673,
+                                        506,
+                                        688
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are non-negative, it is",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                685,
+                                506,
+                                699
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        685,
+                                        229,
+                                        699
+                                    ],
+                                    "score": 1.0,
+                                    "content": "bclear that the length of interval",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        230,
+                                        685,
+                                        301,
+                                        698
+                                    ],
+                                    "score": 0.93,
+                                    "content": "[ \\delta _ { t s } / ( 1 + \\alpha ) , \\delta _ { t s } \\overline { { ) } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        301,
+                                        685,
+                                        351,
+                                        699
+                                    ],
+                                    "score": 1.0,
+                                    "content": "increases as",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        351,
+                                        687,
+                                        359,
+                                        695
+                                    ],
+                                    "score": 0.78,
+                                    "content": "\\alpha",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        359,
+                                        685,
+                                        506,
+                                        699
+                                    ],
+                                    "score": 1.0,
+                                    "content": "increases, and hence probability that",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                107,
+                                695,
+                                507,
+                                711
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        697,
+                                        119,
+                                        708
+                                    ],
+                                    "score": 0.87,
+                                    "content": "\\delta _ { t r }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        119,
+                                        695,
+                                        342,
+                                        711
+                                    ],
+                                    "score": 1.0,
+                                    "content": "falls in this interval is non-decreasing with increasing",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        343,
+                                        699,
+                                        350,
+                                        707
+                                    ],
+                                    "score": 0.73,
+                                    "content": "\\alpha",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        350,
+                                        695,
+                                        446,
+                                        711
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Thus we have for any",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        447,
+                                        697,
+                                        502,
+                                        708
+                                    ],
+                                    "score": 0.91,
+                                    "content": "0 \\leq \\alpha _ { 1 } \\leq \\alpha _ { 2 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        503,
+                                        695,
+                                        507,
+                                        711
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                706,
+                                506,
+                                721
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        707,
+                                        324,
+                                        720
+                                    ],
+                                    "score": 0.87,
+                                    "content": "\\mathbb { P } \\{ \\widehat { y } _ { \\alpha _ { 1 } } \\in \\mathcal { Y } ^ { t r } | y _ { v } \\in \\mathcal { Y } ^ { t s } \\} \\geq \\mathbb { P } \\{ \\widehat { y } _ { \\alpha _ { 2 } } \\in \\mathcal { Y } ^ { \\bar { t } r } | y _ { v } \\in \\mathcal { Y } ^ { t s } \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        324,
+                                        706,
+                                        345,
+                                        721
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", i.e.",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        345,
+                                        707,
+                                        443,
+                                        720
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\mathbb { P } \\{ \\widehat { y } _ { \\alpha } \\in \\mathcal { V } ^ { t r } | y _ { v } \\in \\mathcal { V } ^ { t s } \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        443,
+                                        706,
+                                        506,
+                                        721
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is a monotone",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                718,
+                                376,
+                                732
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        718,
+                                        214,
+                                        732
+                                    ],
+                                    "score": 1.0,
+                                    "content": "bnon-increasing function of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        215,
+                                        721,
+                                        222,
+                                        728
+                                    ],
+                                    "score": 0.82,
+                                    "content": "\\alpha",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        223,
+                                        718,
+                                        364,
+                                        732
+                                    ],
+                                    "score": 1.0,
+                                    "content": "b band we can reduce it by increasing",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        364,
+                                        721,
+                                        371,
+                                        728
+                                    ],
+                                    "score": 0.78,
+                                    "content": "\\alpha",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        372,
+                                        718,
+                                        376,
+                                        732
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        }
+                    ],
+                    "index": 40
+                }
+            ],
+            "page_idx": 9,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        107,
+                        27,
+                        308,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                107,
+                                26,
+                                308,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        26,
+                                        308,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Under review as a conference paper at ICLR 2020",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        300,
+                        751,
+                        311,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                298,
+                                750,
+                                312,
+                                764
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        298,
+                                        750,
+                                        312,
+                                        764
+                                    ],
+                                    "score": 1.0,
+                                    "content": "10",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "list",
+                    "bbox": [
+                        105,
+                        81,
+                        507,
+                        414
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                82,
+                                505,
+                                95
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        82,
+                                        505,
+                                        95
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Richard Socher, Milind Ganjoo, Christopher D Manning, and Andrew Ng. Zero-shot learning through",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                94,
+                                265,
+                                105
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        94,
+                                        265,
+                                        105
+                                    ],
+                                    "score": 1.0,
+                                    "content": "cross-modal transfer. In NIPS, 2013.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                110,
+                                505,
+                                124
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        110,
+                                        505,
+                                        124
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Vinay Kumar Verma, Gundeep Arora, Ashish Mishra, and Piyush Rai. Generalized zero-shot learning",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                122,
+                                289,
+                                133
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        122,
+                                        289,
+                                        133
+                                    ],
+                                    "score": 1.0,
+                                    "content": "via synthesized examples. In CVPR, 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                138,
+                                506,
+                                152
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        138,
+                                        506,
+                                        152
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Wei Wang, Vincent W. Zheng, Han Yu, and Chunyan Miao. A survey of zero-shot learning: Settings,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                117,
+                                150,
+                                387,
+                                162
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        117,
+                                        150,
+                                        387,
+                                        162
+                                    ],
+                                    "score": 1.0,
+                                    "content": "methods, and applications. ACM Trans. Intell. Syst. Technol., 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                166,
+                                429,
+                                180
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        166,
+                                        429,
+                                        180
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Yaqing Wang and Quanming Yao. Few-shot learning: A survey. In arXiv, 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6,
+                            "is_list_start_line": true,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                182,
+                                505,
+                                196
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        182,
+                                        505,
+                                        196
+                                    ],
+                                    "score": 1.0,
+                                    "content": "P. Welinder, S. Branson, T. Mita, C. Wah, F. Schroff, S. Belongie, and P. Perona. Caltech-UCSD",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                194,
+                                395,
+                                207
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        194,
+                                        395,
+                                        207
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Birds 200. Technical report, California Institute of Technology, 2010.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                211,
+                                506,
+                                224
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        211,
+                                        506,
+                                        224
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Yongqin Xian, Zeynep Akata, Gaurav Sharma, Quynh N. Nguyen, Matthias Hein, and Bernt Schiele.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                222,
+                                371,
+                                234
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        222,
+                                        371,
+                                        234
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Latent embeddings for zero-shot classification. In CVPR, 2016.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                239,
+                                507,
+                                252
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        239,
+                                        507,
+                                        252
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Yongqin Xian, Christoph H. Lampert, Bernt Schiele, and Zeynep Akata. Pretrained CUB features,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                249,
+                                468,
+                                263
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        249,
+                                        468,
+                                        263
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2018a. URL http://datasets.d2.mpi-inf.mpg.de/xian/xlsa17.zip.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                266,
+                                505,
+                                280
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        266,
+                                        505,
+                                        280
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Yongqin Xian, Christoph H. Lampert, Bernt Schiele, and Zeynep Akata. Pretrained FLOWERS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                278,
+                                508,
+                                291
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        278,
+                                        508,
+                                        291
+                                    ],
+                                    "score": 1.0,
+                                    "content": "features, 2018b. URL http://datasets.d2.mpi-inf.mpg.de/xian/cvpr18xian.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                117,
+                                289,
+                                140,
+                                302
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        117,
+                                        289,
+                                        140,
+                                        302
+                                    ],
+                                    "score": 1.0,
+                                    "content": "zip.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                306,
+                                506,
+                                320
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        306,
+                                        506,
+                                        320
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Yongqin Xian, H. Christoph Lampert, Bernt Schiele, and Zeynep Akata. Zero-shot learning: A",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                317,
+                                422,
+                                330
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        317,
+                                        422,
+                                        330
+                                    ],
+                                    "score": 1.0,
+                                    "content": "comprehensive evaluation of the good, the bad and the ugly. TPAMI, 2018c.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                333,
+                                505,
+                                347
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        333,
+                                        505,
+                                        347
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Yongqin Xian, Tobias Lorenz, Bernt Schiele, and Zeynep Akata. Feature generating networks for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                345,
+                                264,
+                                357
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        345,
+                                        264,
+                                        357
+                                    ],
+                                    "score": 1.0,
+                                    "content": "zero-shot learning. In CVPR, 2018d.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                361,
+                                505,
+                                375
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        361,
+                                        505,
+                                        375
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Yongqin Xian, Saurabh Sharma, Bernt Schiele, and Zeynep Akata. f-vaegan-d2: A feature generating",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                373,
+                                307,
+                                385
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        373,
+                                        307,
+                                        385
+                                    ],
+                                    "score": 1.0,
+                                    "content": "framework for any-shot learning. CVPR, 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                389,
+                                505,
+                                403
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        389,
+                                        451,
+                                        403
+                                    ],
+                                    "score": 1.0,
+                                    "content": "H. Zhang, T. Xu, H. Li, S. Zhang, X. Wang, X. Huang, and D. N. Metaxas. StackGAN",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        451,
+                                        390,
+                                        463,
+                                        400
+                                    ],
+                                    "score": 0.4,
+                                    "content": "^ { + + }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        464,
+                                        389,
+                                        505,
+                                        403
+                                    ],
+                                    "score": 1.0,
+                                    "content": ": Realistic",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                401,
+                                420,
+                                414
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        401,
+                                        420,
+                                        414
+                                    ],
+                                    "score": 1.0,
+                                    "content": "image synthesis with stacked generative adversarial networks. PAMI, 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23,
+                            "is_list_end_line": true
+                        }
+                    ],
+                    "index": 11.5,
+                    "bbox_fs": [
+                        104,
+                        82,
+                        508,
+                        414
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        429,
+                        299,
+                        442
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                428,
+                                299,
+                                444
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        428,
+                                        299,
+                                        444
+                                    ],
+                                    "score": 1.0,
+                                    "content": "A THE ANALYSIS OF ERROR RATES",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        }
+                    ],
+                    "index": 24
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        452,
+                        505,
+                        477
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                451,
+                                505,
+                                467
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        451,
+                                        169,
+                                        467
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We show that",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        169,
+                                        453,
+                                        404,
+                                        465
+                                    ],
+                                    "score": 0.86,
+                                    "content": "\\mathbb { P } \\{ \\widehat { y } ~ \\in ~ \\mathcal { Y } ^ { t r } | y _ { v } ~ \\in ~ \\mathcal { Y } ^ { t s } \\} ~ \\geq ~ \\mathbb { P } \\{ \\widehat { y } _ { \\alpha } ~ \\in ~ \\mathcal { Y } ^ { t r } | y _ { v } ~ \\in ~ \\mathcal { Y } ^ { t s } \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        404,
+                                        451,
+                                        476,
+                                        467
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Let us define",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        476,
+                                        453,
+                                        505,
+                                        466
+                                    ],
+                                    "score": 0.85,
+                                    "content": "\\delta _ { t r } \\equiv",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                107,
+                                464,
+                                483,
+                                479
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        465,
+                                        195,
+                                        477
+                                    ],
+                                    "score": 0.88,
+                                    "content": "\\begin{array} { r } { \\operatorname* { m i n } _ { y \\in \\mathcal { y } ^ { t r } } d ( \\mathbf { z } _ { v } , \\mathbf { p } ( y ) ) } \\end{array}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        195,
+                                        464,
+                                        213,
+                                        479
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        214,
+                                        466,
+                                        326,
+                                        477
+                                    ],
+                                    "score": 0.87,
+                                    "content": "\\begin{array} { r } { \\delta _ { t s } \\equiv \\operatorname* { m i n } _ { y \\in \\mathcal { V } ^ { t s } } d ( \\mathbf { z } _ { v } , \\mathbf { p } ( y ) ) } \\end{array}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        326,
+                                        464,
+                                        483,
+                                        479
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", then Equation (7) can be rewritten as:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        }
+                    ],
+                    "index": 25.5,
+                    "bbox_fs": [
+                        105,
+                        451,
+                        505,
+                        479
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        183,
+                        479,
+                        425,
+                        492
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                183,
+                                479,
+                                425,
+                                492
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        183,
+                                        479,
+                                        425,
+                                        492
+                                    ],
+                                    "score": 0.81,
+                                    "content": "\\mathbb { P } \\{ \\widehat { y } _ { \\alpha } \\in \\mathcal { V } ^ { t r } | y _ { v } \\in \\mathcal { V } ^ { t s } \\} = \\mathbb { P } \\left\\{ ( 1 + \\alpha ) \\delta _ { t r } < \\delta _ { t s } | y _ { v } \\in \\mathcal { V } ^ { t s } \\right\\} \\mathrm { ~ . ~ }",
+                                    "type": "interline_equation",
+                                    "image_path": "d406a9252d80daf106d9a4cfc77b0b1de57e941418018b7e6ea684fcf96c68c2.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 27,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                183,
+                                479,
+                                425,
+                                492
+                            ],
+                            "spans": [],
+                            "index": 27
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        493,
+                        423,
+                        504
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                492,
+                                422,
+                                506
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        492,
+                                        265,
+                                        506
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Let us consider the probability of event",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        266,
+                                        493,
+                                        303,
+                                        504
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\delta _ { t r } < \\delta _ { t s }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        303,
+                                        492,
+                                        422,
+                                        506
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and decompose it as follows:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        }
+                    ],
+                    "index": 28,
+                    "bbox_fs": [
+                        106,
+                        492,
+                        422,
+                        506
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        131,
+                        503,
+                        473,
+                        582
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                131,
+                                503,
+                                473,
+                                582
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        131,
+                                        503,
+                                        473,
+                                        582
+                                    ],
+                                    "score": 0.67,
+                                    "content": "\\begin{array} { r l } & { \\mathbb { P } \\left\\{ \\delta _ { t r } < \\delta _ { t s } \\middle | y _ { v } \\in \\mathcal { Y } ^ { t s } \\right\\} = \\mathbb { P } \\left\\{ ( 1 + \\alpha ) \\delta _ { t r } < ( 1 + \\alpha ) \\delta _ { t s } | y _ { v } \\in \\mathcal { Y } ^ { t s } \\right\\} } \\\\ & { = \\mathbb { P } \\left\\{ ( 1 + \\alpha ) \\delta _ { t r } < \\delta _ { t s } \\cup \\delta _ { t s } \\leq ( 1 + \\alpha ) \\delta _ { t r } < ( 1 + \\alpha ) \\delta _ { t s } | y _ { v } \\in \\mathcal { Y } ^ { t s } \\right\\} } \\\\ & { = \\mathbb { P } \\left\\{ ( 1 + \\alpha ) \\delta _ { t r } < \\delta _ { t s } | y _ { v } \\in \\mathcal { Y } ^ { t s } \\right\\} + \\mathbb { P } \\left\\{ \\delta _ { t s } \\leq ( 1 + \\alpha ) \\delta _ { t r } < ( 1 + \\alpha ) \\delta _ { t s } | y _ { v } \\in \\mathcal { Y } ^ { t s } \\right\\} } \\\\ & { - \\mathbb { P } \\left\\{ ( 1 + \\alpha ) \\delta _ { t r } < \\delta _ { t s } \\cap \\delta _ { t s } \\leq ( 1 + \\alpha ) \\delta _ { t r } < ( 1 + \\alpha ) \\delta _ { t s } | y _ { v } \\in \\mathcal { Y } ^ { t s } \\right\\} } \\\\ & { = \\mathbb { P } \\left\\{ ( 1 + \\alpha ) \\delta _ { t r } < \\delta _ { t s } | y _ { v } \\in \\mathcal { Y } ^ { t s } \\right\\} + \\mathbb { P } \\left\\{ \\delta _ { t s } \\leq ( 1 + \\alpha ) \\delta _ { t r } < ( 1 + \\alpha ) \\delta _ { t s } | y _ { v } \\in \\mathcal { Y } ^ { t s } \\right\\} } \\end{array}",
+                                    "type": "interline_equation",
+                                    "image_path": "1125cd16d5d95a1f8c8e948f94949c974d4895b575a12f7fa163ec98d67420e9.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 30,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                131,
+                                503,
+                                473,
+                                529.3333333333334
+                            ],
+                            "spans": [],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                131,
+                                529.3333333333334,
+                                473,
+                                555.6666666666667
+                            ],
+                            "spans": [],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                131,
+                                555.6666666666667,
+                                473,
+                                582.0000000000001
+                            ],
+                            "spans": [],
+                            "index": 31
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        582,
+                        507,
+                        604
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                580,
+                                507,
+                                595
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        580,
+                                        464,
+                                        595
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The transitions are based on the relationship between probabilities of arbitrary events",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        465,
+                                        582,
+                                        473,
+                                        591
+                                    ],
+                                    "score": 0.82,
+                                    "content": "A",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        474,
+                                        580,
+                                        493,
+                                        595
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        493,
+                                        582,
+                                        502,
+                                        591
+                                    ],
+                                    "score": 0.83,
+                                    "content": "B",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        503,
+                                        580,
+                                        507,
+                                        595
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                592,
+                                483,
+                                606
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        592,
+                                        278,
+                                        605
+                                    ],
+                                    "score": 0.88,
+                                    "content": "\\mathbb { P } \\{ A \\cup B \\} = \\mathbb { P } \\{ A \\} + \\mathbb { P } \\{ B \\} - \\mathbb { P } \\{ A \\cap B \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        279,
+                                        592,
+                                        344,
+                                        606
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", and in our case",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        345,
+                                        592,
+                                        407,
+                                        605
+                                    ],
+                                    "score": 0.92,
+                                    "content": "{ \\mathbb { P } } \\{ A \\cap B \\} = 0",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        407,
+                                        592,
+                                        483,
+                                        606
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". This implies that:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        }
+                    ],
+                    "index": 32.5,
+                    "bbox_fs": [
+                        105,
+                        580,
+                        507,
+                        606
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        128,
+                        605,
+                        482,
+                        648
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                128,
+                                605,
+                                482,
+                                648
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        128,
+                                        605,
+                                        482,
+                                        648
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\begin{array} { r l r } {  { \\mathbb { P } \\{ \\widehat { y } _ { \\alpha } \\in \\mathcal { Y } ^ { t r } | y _ { v } \\in \\mathcal { Y } ^ { t s } \\} = \\mathbb { P } \\{ \\widehat { y } \\in \\mathcal { Y } ^ { t r } | y _ { v } \\in \\mathcal { Y } ^ { t s } \\} - \\mathbb { P } \\{ \\frac { \\delta _ { t s } } { ( 1 + \\alpha ) } \\leq \\delta _ { t r } < \\delta _ { t s } | y _ { v } \\in \\mathcal { Y } ^ { t s } \\} } } \\\\ & { } & { \\leq \\mathbb { P } \\{ \\widehat { y } \\in \\mathcal { Y } ^ { t r } | y _ { v } \\in \\mathcal { Y } ^ { t s } \\} . } \\end{array}",
+                                    "type": "interline_equation",
+                                    "image_path": "c82fe815b758df297244b22978a323eab8837f91f51cbd01bee778cf0f808da0.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 35,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                128,
+                                605,
+                                482,
+                                619.3333333333334
+                            ],
+                            "spans": [],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                128,
+                                619.3333333333334,
+                                482,
+                                633.6666666666667
+                            ],
+                            "spans": [],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                128,
+                                633.6666666666667,
+                                482,
+                                648.0000000000001
+                            ],
+                            "spans": [],
+                            "index": 36
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        652,
+                        506,
+                        730
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                653,
+                                505,
+                                665
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        653,
+                                        285,
+                                        665
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We have just shown that for a non-negative",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        286,
+                                        655,
+                                        294,
+                                        663
+                                    ],
+                                    "score": 0.8,
+                                    "content": "\\alpha",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        294,
+                                        653,
+                                        505,
+                                        665
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the probability of misclassifying an image from an",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                664,
+                                505,
+                                676
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        664,
+                                        404,
+                                        676
+                                    ],
+                                    "score": 1.0,
+                                    "content": "unseen class as one of the seen classes is smaller for the decision rule",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        405,
+                                        664,
+                                        416,
+                                        675
+                                    ],
+                                    "score": 0.88,
+                                    "content": "\\widehat { y } _ { \\alpha }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        417,
+                                        664,
+                                        505,
+                                        676
+                                    ],
+                                    "score": 1.0,
+                                    "content": "than for the original",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                673,
+                                506,
+                                688
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        673,
+                                        162,
+                                        688
+                                    ],
+                                    "score": 1.0,
+                                    "content": "decision rule",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        162,
+                                        675,
+                                        169,
+                                        686
+                                    ],
+                                    "score": 0.81,
+                                    "content": "\\widehat { y }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        170,
+                                        673,
+                                        365,
+                                        688
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". In fact, we can make a stronger claim. Since",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        366,
+                                        675,
+                                        378,
+                                        686
+                                    ],
+                                    "score": 0.89,
+                                    "content": "\\delta _ { t s }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        379,
+                                        673,
+                                        398,
+                                        688
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        398,
+                                        675,
+                                        411,
+                                        686
+                                    ],
+                                    "score": 0.9,
+                                    "content": "\\delta _ { t r }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        411,
+                                        673,
+                                        506,
+                                        688
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are non-negative, it is",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                685,
+                                506,
+                                699
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        685,
+                                        229,
+                                        699
+                                    ],
+                                    "score": 1.0,
+                                    "content": "bclear that the length of interval",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        230,
+                                        685,
+                                        301,
+                                        698
+                                    ],
+                                    "score": 0.93,
+                                    "content": "[ \\delta _ { t s } / ( 1 + \\alpha ) , \\delta _ { t s } \\overline { { ) } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        301,
+                                        685,
+                                        351,
+                                        699
+                                    ],
+                                    "score": 1.0,
+                                    "content": "increases as",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        351,
+                                        687,
+                                        359,
+                                        695
+                                    ],
+                                    "score": 0.78,
+                                    "content": "\\alpha",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        359,
+                                        685,
+                                        506,
+                                        699
+                                    ],
+                                    "score": 1.0,
+                                    "content": "increases, and hence probability that",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                107,
+                                695,
+                                507,
+                                711
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        697,
+                                        119,
+                                        708
+                                    ],
+                                    "score": 0.87,
+                                    "content": "\\delta _ { t r }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        119,
+                                        695,
+                                        342,
+                                        711
+                                    ],
+                                    "score": 1.0,
+                                    "content": "falls in this interval is non-decreasing with increasing",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        343,
+                                        699,
+                                        350,
+                                        707
+                                    ],
+                                    "score": 0.73,
+                                    "content": "\\alpha",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        350,
+                                        695,
+                                        446,
+                                        711
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Thus we have for any",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        447,
+                                        697,
+                                        502,
+                                        708
+                                    ],
+                                    "score": 0.91,
+                                    "content": "0 \\leq \\alpha _ { 1 } \\leq \\alpha _ { 2 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        503,
+                                        695,
+                                        507,
+                                        711
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                706,
+                                506,
+                                721
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        707,
+                                        324,
+                                        720
+                                    ],
+                                    "score": 0.87,
+                                    "content": "\\mathbb { P } \\{ \\widehat { y } _ { \\alpha _ { 1 } } \\in \\mathcal { Y } ^ { t r } | y _ { v } \\in \\mathcal { Y } ^ { t s } \\} \\geq \\mathbb { P } \\{ \\widehat { y } _ { \\alpha _ { 2 } } \\in \\mathcal { Y } ^ { \\bar { t } r } | y _ { v } \\in \\mathcal { Y } ^ { t s } \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        324,
+                                        706,
+                                        345,
+                                        721
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", i.e.",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        345,
+                                        707,
+                                        443,
+                                        720
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\mathbb { P } \\{ \\widehat { y } _ { \\alpha } \\in \\mathcal { V } ^ { t r } | y _ { v } \\in \\mathcal { V } ^ { t s } \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        443,
+                                        706,
+                                        506,
+                                        721
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is a monotone",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                718,
+                                376,
+                                732
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        718,
+                                        214,
+                                        732
+                                    ],
+                                    "score": 1.0,
+                                    "content": "bnon-increasing function of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        215,
+                                        721,
+                                        222,
+                                        728
+                                    ],
+                                    "score": 0.82,
+                                    "content": "\\alpha",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        223,
+                                        718,
+                                        364,
+                                        732
+                                    ],
+                                    "score": 1.0,
+                                    "content": "b band we can reduce it by increasing",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        364,
+                                        721,
+                                        371,
+                                        728
+                                    ],
+                                    "score": 0.78,
+                                    "content": "\\alpha",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        372,
+                                        718,
+                                        376,
+                                        732
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        }
+                    ],
+                    "index": 40,
+                    "bbox_fs": [
+                        105,
+                        653,
+                        507,
+                        732
+                    ]
+                }
+            ]
+        }
+    ],
+    "_backend": "pipeline",
+    "_version_name": "2.2.2"
+}
\ No newline at end of file
diff --git a/parse/train/SJl3h2EYvS/SJl3h2EYvS_model.json b/parse/train/SJl3h2EYvS/SJl3h2EYvS_model.json
new file mode 100644
index 0000000000000000000000000000000000000000..4b6f98e163a9a97aef31960287a7f85b64c16a4a
--- /dev/null
+++ b/parse/train/SJl3h2EYvS/SJl3h2EYvS_model.json
@@ -0,0 +1,15003 @@
+[
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1073,
+                    1404,
+                    1073,
+                    1404,
+                    1803,
+                    298,
+                    1803
+                ],
+                "score": 0.984
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    398,
+                    592,
+                    1303,
+                    592,
+                    1303,
+                    928,
+                    398,
+                    928
+                ],
+                "score": 0.982
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1820,
+                    1403,
+                    1820,
+                    1403,
+                    2034,
+                    298,
+                    2034
+                ],
+                "score": 0.979
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    299,
+                    221,
+                    1400,
+                    221,
+                    1400,
+                    323,
+                    299,
+                    323
+                ],
+                "score": 0.96
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    313,
+                    377,
+                    680,
+                    377,
+                    680,
+                    438,
+                    313,
+                    438
+                ],
+                "score": 0.931
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    302,
+                    999,
+                    573,
+                    999,
+                    573,
+                    1034,
+                    302,
+                    1034
+                ],
+                "score": 0.9
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    298,
+                    75,
+                    858,
+                    75,
+                    858,
+                    105,
+                    298,
+                    105
+                ],
+                "score": 0.898
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    773,
+                    519,
+                    927,
+                    519,
+                    927,
+                    553,
+                    773,
+                    553
+                ],
+                "score": 0.872
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    841,
+                    2088,
+                    856,
+                    2088,
+                    856,
+                    2112,
+                    841,
+                    2112
+                ],
+                "score": 0.723
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    221.0,
+                    1403.0,
+                    221.0,
+                    1403.0,
+                    275.0,
+                    295.0,
+                    275.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    283.0,
+                    763.0,
+                    283.0,
+                    763.0,
+                    328.0,
+                    297.0,
+                    328.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    996.0,
+                    579.0,
+                    996.0,
+                    579.0,
+                    1043.0,
+                    294.0,
+                    1043.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    298.0,
+                    73.0,
+                    858.0,
+                    73.0,
+                    858.0,
+                    108.0,
+                    298.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    770.0,
+                    518.0,
+                    932.0,
+                    518.0,
+                    932.0,
+                    556.0,
+                    770.0,
+                    556.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    840.0,
+                    2088.0,
+                    857.0,
+                    2088.0,
+                    857.0,
+                    2116.0,
+                    840.0,
+                    2116.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1074.0,
+                    1404.0,
+                    1074.0,
+                    1404.0,
+                    1108.0,
+                    294.0,
+                    1108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1105.0,
+                    1406.0,
+                    1105.0,
+                    1406.0,
+                    1139.0,
+                    293.0,
+                    1139.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1133.0,
+                    1405.0,
+                    1133.0,
+                    1405.0,
+                    1168.0,
+                    294.0,
+                    1168.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1164.0,
+                    1406.0,
+                    1164.0,
+                    1406.0,
+                    1201.0,
+                    292.0,
+                    1201.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1190.0,
+                    1406.0,
+                    1190.0,
+                    1406.0,
+                    1233.0,
+                    292.0,
+                    1233.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1224.0,
+                    1405.0,
+                    1224.0,
+                    1405.0,
+                    1259.0,
+                    293.0,
+                    1259.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1256.0,
+                    1404.0,
+                    1256.0,
+                    1404.0,
+                    1291.0,
+                    294.0,
+                    1291.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1285.0,
+                    1405.0,
+                    1285.0,
+                    1405.0,
+                    1323.0,
+                    293.0,
+                    1323.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1316.0,
+                    1405.0,
+                    1316.0,
+                    1405.0,
+                    1354.0,
+                    292.0,
+                    1354.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1345.0,
+                    1408.0,
+                    1345.0,
+                    1408.0,
+                    1385.0,
+                    292.0,
+                    1385.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1377.0,
+                    1405.0,
+                    1377.0,
+                    1405.0,
+                    1412.0,
+                    294.0,
+                    1412.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1408.0,
+                    1406.0,
+                    1408.0,
+                    1406.0,
+                    1442.0,
+                    293.0,
+                    1442.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1439.0,
+                    1406.0,
+                    1439.0,
+                    1406.0,
+                    1474.0,
+                    294.0,
+                    1474.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1465.0,
+                    1406.0,
+                    1465.0,
+                    1406.0,
+                    1507.0,
+                    292.0,
+                    1507.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1501.0,
+                    1405.0,
+                    1501.0,
+                    1405.0,
+                    1536.0,
+                    293.0,
+                    1536.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1527.0,
+                    1408.0,
+                    1527.0,
+                    1408.0,
+                    1566.0,
+                    292.0,
+                    1566.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1558.0,
+                    1406.0,
+                    1558.0,
+                    1406.0,
+                    1599.0,
+                    292.0,
+                    1599.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1590.0,
+                    1409.0,
+                    1590.0,
+                    1409.0,
+                    1627.0,
+                    292.0,
+                    1627.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1621.0,
+                    1405.0,
+                    1621.0,
+                    1405.0,
+                    1655.0,
+                    294.0,
+                    1655.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1652.0,
+                    1404.0,
+                    1652.0,
+                    1404.0,
+                    1686.0,
+                    293.0,
+                    1686.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1678.0,
+                    1406.0,
+                    1678.0,
+                    1406.0,
+                    1720.0,
+                    291.0,
+                    1720.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1710.0,
+                    1404.0,
+                    1710.0,
+                    1404.0,
+                    1749.0,
+                    293.0,
+                    1749.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1742.0,
+                    1405.0,
+                    1742.0,
+                    1405.0,
+                    1780.0,
+                    293.0,
+                    1780.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1773.0,
+                    1186.0,
+                    1773.0,
+                    1186.0,
+                    1809.0,
+                    295.0,
+                    1809.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    397.0,
+                    594.0,
+                    1303.0,
+                    594.0,
+                    1303.0,
+                    627.0,
+                    397.0,
+                    627.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    392.0,
+                    623.0,
+                    1306.0,
+                    623.0,
+                    1306.0,
+                    660.0,
+                    392.0,
+                    660.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    393.0,
+                    655.0,
+                    1305.0,
+                    655.0,
+                    1305.0,
+                    690.0,
+                    393.0,
+                    690.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    392.0,
+                    683.0,
+                    1305.0,
+                    683.0,
+                    1305.0,
+                    722.0,
+                    392.0,
+                    722.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    395.0,
+                    717.0,
+                    1307.0,
+                    717.0,
+                    1307.0,
+                    750.0,
+                    395.0,
+                    750.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    392.0,
+                    745.0,
+                    1308.0,
+                    745.0,
+                    1308.0,
+                    781.0,
+                    392.0,
+                    781.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    394.0,
+                    778.0,
+                    1306.0,
+                    778.0,
+                    1306.0,
+                    810.0,
+                    394.0,
+                    810.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    394.0,
+                    808.0,
+                    1307.0,
+                    808.0,
+                    1307.0,
+                    841.0,
+                    394.0,
+                    841.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    394.0,
+                    838.0,
+                    1304.0,
+                    838.0,
+                    1304.0,
+                    869.0,
+                    394.0,
+                    869.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    392.0,
+                    866.0,
+                    1305.0,
+                    866.0,
+                    1305.0,
+                    905.0,
+                    392.0,
+                    905.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    394.0,
+                    899.0,
+                    814.0,
+                    899.0,
+                    814.0,
+                    931.0,
+                    394.0,
+                    931.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1819.0,
+                    1409.0,
+                    1819.0,
+                    1409.0,
+                    1857.0,
+                    293.0,
+                    1857.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1851.0,
+                    1405.0,
+                    1851.0,
+                    1405.0,
+                    1885.0,
+                    294.0,
+                    1885.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1880.0,
+                    1405.0,
+                    1880.0,
+                    1405.0,
+                    1916.0,
+                    293.0,
+                    1916.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1911.0,
+                    1404.0,
+                    1911.0,
+                    1404.0,
+                    1947.0,
+                    293.0,
+                    1947.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1943.0,
+                    1405.0,
+                    1943.0,
+                    1405.0,
+                    1977.0,
+                    294.0,
+                    1977.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1971.0,
+                    1405.0,
+                    1971.0,
+                    1405.0,
+                    2007.0,
+                    293.0,
+                    2007.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    2002.0,
+                    1336.0,
+                    2002.0,
+                    1336.0,
+                    2039.0,
+                    293.0,
+                    2039.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    315.0,
+                    377.0,
+                    560.0,
+                    377.0,
+                    560.0,
+                    410.0,
+                    315.0,
+                    410.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    311.0,
+                    407.0,
+                    682.0,
+                    407.0,
+                    682.0,
+                    440.0,
+                    311.0,
+                    440.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 0,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1668,
+                    1406,
+                    1668,
+                    1406,
+                    2034,
+                    297,
+                    2034
+                ],
+                "score": 0.984
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1285,
+                    1405,
+                    1285,
+                    1405,
+                    1653,
+                    297,
+                    1653
+                ],
+                "score": 0.984
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1006,
+                    1403,
+                    1006,
+                    1403,
+                    1162,
+                    298,
+                    1162
+                ],
+                "score": 0.979
+            },
+            {
+                "category_id": 3,
+                "poly": [
+                    426,
+                    219,
+                    1283,
+                    219,
+                    1283,
+                    643,
+                    426,
+                    643
+                ],
+                "score": 0.974
+            },
+            {
+                "category_id": 4,
+                "poly": [
+                    296,
+                    665,
+                    1405,
+                    665,
+                    1405,
+                    941,
+                    296,
+                    941
+                ],
+                "score": 0.961
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    301,
+                    1211,
+                    641,
+                    1211,
+                    641,
+                    1247,
+                    301,
+                    1247
+                ],
+                "score": 0.902
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    298,
+                    76,
+                    858,
+                    76,
+                    858,
+                    104,
+                    298,
+                    104
+                ],
+                "score": 0.878
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    841,
+                    2088,
+                    858,
+                    2088,
+                    858,
+                    2112,
+                    841,
+                    2112
+                ],
+                "score": 0.717
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    841,
+                    2088,
+                    859,
+                    2088,
+                    859,
+                    2112,
+                    841,
+                    2112
+                ],
+                "score": 0.129
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1179,
+                    1529,
+                    1353,
+                    1529,
+                    1353,
+                    1559,
+                    1179,
+                    1559
+                ],
+                "score": 0.93,
+                "latex": "\\mathcal { V } = \\mathcal { V } ^ { t r } \\cup \\mathcal { V } ^ { t s }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    933,
+                    1529,
+                    976,
+                    1529,
+                    976,
+                    1560,
+                    933,
+                    1560
+                ],
+                "score": 0.92,
+                "latex": "\\mathcal { V } ^ { t s }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    868,
+                    1471,
+                    1008,
+                    1471,
+                    1008,
+                    1502,
+                    868,
+                    1502
+                ],
+                "score": 0.92,
+                "latex": "f _ { \\phi } : \\mathcal { V } \\to \\mathcal { Z }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    298,
+                    1500,
+                    445,
+                    1500,
+                    445,
+                    1531,
+                    298,
+                    1531
+                ],
+                "score": 0.91,
+                "latex": "f _ { \\theta } : \\mathcal { T }  \\mathcal { Z }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    692,
+                    1620,
+                    895,
+                    1620,
+                    895,
+                    1652,
+                    692,
+                    1652
+                ],
+                "score": 0.91,
+                "latex": "g : \\mathcal { Z } \\times \\mathcal { Z }  \\mathcal { Y } ^ { t s }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    298,
+                    1559,
+                    456,
+                    1559,
+                    456,
+                    1590,
+                    298,
+                    1590
+                ],
+                "score": 0.91,
+                "latex": "\\bar { \\mathcal { V } } ^ { t r } \\cap \\mathcal { V } ^ { t s } = \\emptyset"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1028,
+                    1593,
+                    1212,
+                    1593,
+                    1212,
+                    1622,
+                    1028,
+                    1622
+                ],
+                "score": 0.9,
+                "latex": "g : \\mathcal { Z } \\times \\mathcal { Z }  \\mathcal { V }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    754,
+                    1529,
+                    797,
+                    1529,
+                    797,
+                    1559,
+                    754,
+                    1559
+                ],
+                "score": 0.89,
+                "latex": "\\mathcal { V } ^ { t r }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1101,
+                    1560,
+                    1401,
+                    1560,
+                    1401,
+                    1593,
+                    1101,
+                    1593
+                ],
+                "score": 0.88,
+                "latex": "S ^ { t r } = \\{ ( v _ { n } , t _ { n } , y _ { n } ) \\mid v _ { n } \\in"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    330,
+                    1591,
+                    552,
+                    1591,
+                    552,
+                    1623,
+                    330,
+                    1623
+                ],
+                "score": 0.86,
+                "latex": "t _ { n } \\in \\mathcal { T } , y _ { n } \\in \\mathcal { y } ^ { t r } \\}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1327,
+                    1470,
+                    1346,
+                    1470,
+                    1346,
+                    1500,
+                    1327,
+                    1500
+                ],
+                "score": 0.83,
+                "latex": "\\phi"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1130,
+                    1501,
+                    1157,
+                    1501,
+                    1157,
+                    1527,
+                    1130,
+                    1527
+                ],
+                "score": 0.8,
+                "latex": "\\mathcal { Z }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    394,
+                    1674,
+                    412,
+                    1674,
+                    412,
+                    1700,
+                    394,
+                    1700
+                ],
+                "score": 0.78,
+                "latex": "g"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    742,
+                    1501,
+                    759,
+                    1501,
+                    759,
+                    1527,
+                    742,
+                    1527
+                ],
+                "score": 0.77,
+                "latex": "\\theta"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    769,
+                    1438,
+                    1124,
+                    1438,
+                    1124,
+                    1471,
+                    769,
+                    1471
+                ],
+                "score": 0.5,
+                "latex": "t _ { n } \\in \\mathcal { T } , y _ { n } \\in \\mathcal { V } , n = 1 \\ldots N \\}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    297,
+                    1592,
+                    320,
+                    1592,
+                    320,
+                    1620,
+                    297,
+                    1620
+                ],
+                "score": 0.4,
+                "latex": "\\nu"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    433,
+                    1438,
+                    753,
+                    1438,
+                    753,
+                    1472,
+                    433,
+                    1472
+                ],
+                "score": 0.28,
+                "latex": "S = \\{ ( v _ { n } , t _ { n } , y _ { n } ) \\mid v _ { n } \\in \\mathcal { V }"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    830.0,
+                    235.0,
+                    886.0,
+                    235.0,
+                    886.0,
+                    262.0,
+                    830.0,
+                    262.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    915.0,
+                    226.0,
+                    985.0,
+                    226.0,
+                    985.0,
+                    256.0,
+                    915.0,
+                    256.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1013.0,
+                    228.0,
+                    1080.0,
+                    228.0,
+                    1080.0,
+                    255.0,
+                    1013.0,
+                    255.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1104.0,
+                    240.0,
+                    1141.0,
+                    240.0,
+                    1141.0,
+                    261.0,
+                    1104.0,
+                    261.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    832.0,
+                    256.0,
+                    905.0,
+                    256.0,
+                    905.0,
+                    280.0,
+                    832.0,
+                    280.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    947.0,
+                    262.0,
+                    956.0,
+                    262.0,
+                    956.0,
+                    270.0,
+                    947.0,
+                    270.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1104.0,
+                    258.0,
+                    1176.0,
+                    258.0,
+                    1176.0,
+                    282.0,
+                    1104.0,
+                    282.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    935.0,
+                    291.0,
+                    966.0,
+                    291.0,
+                    966.0,
+                    313.0,
+                    935.0,
+                    313.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1031.0,
+                    291.0,
+                    1063.0,
+                    291.0,
+                    1063.0,
+                    312.0,
+                    1031.0,
+                    312.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1102.0,
+                    337.0,
+                    1199.0,
+                    337.0,
+                    1199.0,
+                    359.0,
+                    1102.0,
+                    359.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    681.0,
+                    354.0,
+                    743.0,
+                    354.0,
+                    743.0,
+                    379.0,
+                    681.0,
+                    379.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    848.0,
+                    358.0,
+                    861.0,
+                    358.0,
+                    861.0,
+                    370.0,
+                    848.0,
+                    370.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1118.0,
+                    364.0,
+                    1183.0,
+                    364.0,
+                    1183.0,
+                    385.0,
+                    1118.0,
+                    385.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    445.0,
+                    456.0,
+                    579.0,
+                    456.0,
+                    579.0,
+                    483.0,
+                    445.0,
+                    483.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1143.0,
+                    456.0,
+                    1160.0,
+                    456.0,
+                    1160.0,
+                    479.0,
+                    1143.0,
+                    479.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1228.0,
+                    409.0,
+                    1280.0,
+                    409.0,
+                    1280.0,
+                    522.0,
+                    1228.0,
+                    522.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    448.0,
+                    475.0,
+                    581.0,
+                    475.0,
+                    581.0,
+                    501.0,
+                    448.0,
+                    501.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    665.0,
+                    476.0,
+                    762.0,
+                    476.0,
+                    762.0,
+                    500.0,
+                    665.0,
+                    500.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    844.0,
+                    480.0,
+                    867.0,
+                    480.0,
+                    867.0,
+                    499.0,
+                    844.0,
+                    499.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    447.0,
+                    493.0,
+                    575.0,
+                    493.0,
+                    575.0,
+                    523.0,
+                    447.0,
+                    523.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    663.0,
+                    1406.0,
+                    663.0,
+                    1406.0,
+                    701.0,
+                    293.0,
+                    701.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    695.0,
+                    1405.0,
+                    695.0,
+                    1405.0,
+                    734.0,
+                    293.0,
+                    734.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    726.0,
+                    1405.0,
+                    726.0,
+                    1405.0,
+                    764.0,
+                    293.0,
+                    764.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    755.0,
+                    1406.0,
+                    755.0,
+                    1406.0,
+                    795.0,
+                    293.0,
+                    795.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    787.0,
+                    1405.0,
+                    787.0,
+                    1405.0,
+                    824.0,
+                    294.0,
+                    824.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    819.0,
+                    1403.0,
+                    819.0,
+                    1403.0,
+                    851.0,
+                    295.0,
+                    851.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    849.0,
+                    1407.0,
+                    849.0,
+                    1407.0,
+                    885.0,
+                    294.0,
+                    885.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    877.0,
+                    1403.0,
+                    877.0,
+                    1403.0,
+                    917.0,
+                    293.0,
+                    917.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    911.0,
+                    1196.0,
+                    911.0,
+                    1196.0,
+                    944.0,
+                    296.0,
+                    944.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1210.0,
+                    645.0,
+                    1210.0,
+                    645.0,
+                    1253.0,
+                    292.0,
+                    1253.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    70.0,
+                    859.0,
+                    70.0,
+                    859.0,
+                    109.0,
+                    296.0,
+                    109.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    839.0,
+                    2085.0,
+                    862.0,
+                    2085.0,
+                    862.0,
+                    2120.0,
+                    839.0,
+                    2120.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    838.0,
+                    2085.0,
+                    862.0,
+                    2085.0,
+                    862.0,
+                    2120.0,
+                    838.0,
+                    2120.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1667.0,
+                    393.0,
+                    1667.0,
+                    393.0,
+                    1707.0,
+                    291.0,
+                    1707.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    413.0,
+                    1667.0,
+                    1409.0,
+                    1667.0,
+                    1409.0,
+                    1707.0,
+                    413.0,
+                    1707.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1699.0,
+                    1406.0,
+                    1699.0,
+                    1406.0,
+                    1733.0,
+                    294.0,
+                    1733.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1728.0,
+                    1406.0,
+                    1728.0,
+                    1406.0,
+                    1762.0,
+                    292.0,
+                    1762.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1760.0,
+                    1408.0,
+                    1760.0,
+                    1408.0,
+                    1794.0,
+                    295.0,
+                    1794.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1787.0,
+                    1409.0,
+                    1787.0,
+                    1409.0,
+                    1827.0,
+                    291.0,
+                    1827.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1822.0,
+                    1403.0,
+                    1822.0,
+                    1403.0,
+                    1852.0,
+                    294.0,
+                    1852.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1850.0,
+                    1406.0,
+                    1850.0,
+                    1406.0,
+                    1886.0,
+                    292.0,
+                    1886.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1880.0,
+                    1406.0,
+                    1880.0,
+                    1406.0,
+                    1917.0,
+                    292.0,
+                    1917.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1909.0,
+                    1408.0,
+                    1909.0,
+                    1408.0,
+                    1947.0,
+                    294.0,
+                    1947.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1941.0,
+                    1406.0,
+                    1941.0,
+                    1406.0,
+                    1978.0,
+                    292.0,
+                    1978.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1974.0,
+                    1406.0,
+                    1974.0,
+                    1406.0,
+                    2008.0,
+                    294.0,
+                    2008.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    2003.0,
+                    1267.0,
+                    2003.0,
+                    1267.0,
+                    2037.0,
+                    295.0,
+                    2037.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1286.0,
+                    1406.0,
+                    1286.0,
+                    1406.0,
+                    1320.0,
+                    294.0,
+                    1320.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1317.0,
+                    1406.0,
+                    1317.0,
+                    1406.0,
+                    1351.0,
+                    294.0,
+                    1351.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1347.0,
+                    1406.0,
+                    1347.0,
+                    1406.0,
+                    1381.0,
+                    295.0,
+                    1381.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1378.0,
+                    1406.0,
+                    1378.0,
+                    1406.0,
+                    1410.0,
+                    292.0,
+                    1410.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1406.0,
+                    1408.0,
+                    1406.0,
+                    1408.0,
+                    1443.0,
+                    292.0,
+                    1443.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1436.0,
+                    432.0,
+                    1436.0,
+                    432.0,
+                    1474.0,
+                    291.0,
+                    1474.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    754.0,
+                    1436.0,
+                    768.0,
+                    1436.0,
+                    768.0,
+                    1474.0,
+                    754.0,
+                    1474.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1125.0,
+                    1436.0,
+                    1407.0,
+                    1436.0,
+                    1407.0,
+                    1474.0,
+                    1125.0,
+                    1474.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1469.0,
+                    867.0,
+                    1469.0,
+                    867.0,
+                    1503.0,
+                    295.0,
+                    1503.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1009.0,
+                    1469.0,
+                    1326.0,
+                    1469.0,
+                    1326.0,
+                    1503.0,
+                    1009.0,
+                    1503.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1347.0,
+                    1469.0,
+                    1405.0,
+                    1469.0,
+                    1405.0,
+                    1503.0,
+                    1347.0,
+                    1503.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1498.0,
+                    297.0,
+                    1498.0,
+                    297.0,
+                    1534.0,
+                    292.0,
+                    1534.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    446.0,
+                    1498.0,
+                    741.0,
+                    1498.0,
+                    741.0,
+                    1534.0,
+                    446.0,
+                    1534.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    760.0,
+                    1498.0,
+                    1129.0,
+                    1498.0,
+                    1129.0,
+                    1534.0,
+                    760.0,
+                    1534.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1158.0,
+                    1498.0,
+                    1408.0,
+                    1498.0,
+                    1408.0,
+                    1534.0,
+                    1158.0,
+                    1534.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    290.0,
+                    1525.0,
+                    753.0,
+                    1525.0,
+                    753.0,
+                    1564.0,
+                    290.0,
+                    1564.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    798.0,
+                    1525.0,
+                    932.0,
+                    1525.0,
+                    932.0,
+                    1564.0,
+                    798.0,
+                    1564.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    977.0,
+                    1525.0,
+                    1178.0,
+                    1525.0,
+                    1178.0,
+                    1564.0,
+                    977.0,
+                    1564.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1354.0,
+                    1525.0,
+                    1408.0,
+                    1525.0,
+                    1408.0,
+                    1564.0,
+                    1354.0,
+                    1564.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1554.0,
+                    297.0,
+                    1554.0,
+                    297.0,
+                    1597.0,
+                    291.0,
+                    1597.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    457.0,
+                    1554.0,
+                    1100.0,
+                    1554.0,
+                    1100.0,
+                    1597.0,
+                    457.0,
+                    1597.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1402.0,
+                    1554.0,
+                    1406.0,
+                    1554.0,
+                    1406.0,
+                    1597.0,
+                    1402.0,
+                    1597.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    321.0,
+                    1587.0,
+                    329.0,
+                    1587.0,
+                    329.0,
+                    1625.0,
+                    321.0,
+                    1625.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    553.0,
+                    1587.0,
+                    1027.0,
+                    1587.0,
+                    1027.0,
+                    1625.0,
+                    553.0,
+                    1625.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1213.0,
+                    1587.0,
+                    1407.0,
+                    1587.0,
+                    1407.0,
+                    1625.0,
+                    1213.0,
+                    1625.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1616.0,
+                    691.0,
+                    1616.0,
+                    691.0,
+                    1654.0,
+                    291.0,
+                    1654.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    896.0,
+                    1616.0,
+                    905.0,
+                    1616.0,
+                    905.0,
+                    1654.0,
+                    896.0,
+                    1654.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1007.0,
+                    1404.0,
+                    1007.0,
+                    1404.0,
+                    1044.0,
+                    294.0,
+                    1044.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1036.0,
+                    1406.0,
+                    1036.0,
+                    1406.0,
+                    1076.0,
+                    292.0,
+                    1076.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1069.0,
+                    1404.0,
+                    1069.0,
+                    1404.0,
+                    1103.0,
+                    294.0,
+                    1103.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1097.0,
+                    1405.0,
+                    1097.0,
+                    1405.0,
+                    1135.0,
+                    294.0,
+                    1135.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1129.0,
+                    1348.0,
+                    1129.0,
+                    1348.0,
+                    1163.0,
+                    296.0,
+                    1163.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 1,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1182,
+                    1405,
+                    1182,
+                    1405,
+                    1339,
+                    297,
+                    1339
+                ],
+                "score": 0.98
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1910,
+                    1405,
+                    1910,
+                    1405,
+                    2036,
+                    298,
+                    2036
+                ],
+                "score": 0.976
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    1584,
+                    1404,
+                    1584,
+                    1404,
+                    1709,
+                    299,
+                    1709
+                ],
+                "score": 0.974
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    517,
+                    1350,
+                    1177,
+                    1350,
+                    1177,
+                    1429,
+                    517,
+                    1429
+                ],
+                "score": 0.956
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1833,
+                    1401,
+                    1833,
+                    1401,
+                    1899,
+                    298,
+                    1899
+                ],
+                "score": 0.951
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    517,
+                    1483,
+                    1184,
+                    1483,
+                    1184,
+                    1574,
+                    517,
+                    1574
+                ],
+                "score": 0.95
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    290,
+                    1437,
+                    1334,
+                    1437,
+                    1334,
+                    1471,
+                    290,
+                    1471
+                ],
+                "score": 0.929
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    382,
+                    1720,
+                    1315,
+                    1720,
+                    1315,
+                    1824,
+                    382,
+                    1824
+                ],
+                "score": 0.919
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    298,
+                    74,
+                    858,
+                    74,
+                    858,
+                    105,
+                    298,
+                    105
+                ],
+                "score": 0.917
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    300,
+                    1127,
+                    701,
+                    1127,
+                    701,
+                    1158,
+                    300,
+                    1158
+                ],
+                "score": 0.911
+            },
+            {
+                "category_id": 9,
+                "poly": [
+                    1366,
+                    1371,
+                    1400,
+                    1371,
+                    1400,
+                    1401,
+                    1366,
+                    1401
+                ],
+                "score": 0.893
+            },
+            {
+                "category_id": 9,
+                "poly": [
+                    1366,
+                    1512,
+                    1400,
+                    1512,
+                    1400,
+                    1543,
+                    1366,
+                    1543
+                ],
+                "score": 0.89
+            },
+            {
+                "category_id": 9,
+                "poly": [
+                    1366,
+                    1758,
+                    1400,
+                    1758,
+                    1400,
+                    1788,
+                    1366,
+                    1788
+                ],
+                "score": 0.856
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    296,
+                    226,
+                    1404,
+                    226,
+                    1404,
+                    354,
+                    296,
+                    354
+                ],
+                "score": 0.803
+            },
+            {
+                "category_id": 5,
+                "poly": [
+                    299,
+                    359,
+                    1405,
+                    359,
+                    1405,
+                    1074,
+                    299,
+                    1074
+                ],
+                "score": 0.764,
+                "html": "<table><tr><td colspan=\"3\">Input: Training set Str = {(U1,t1,yi),...,(UN,tN,yN)},λ ∈ [0,1], κ ∈ [0,1]. Output: The loss J(,θ) for a randomly sampled training batch. I← RANDOMSAMPLE({1,...,N},B) &gt;Select B instance indices for batch</td></tr><tr><td>Jrc(Θ),J1c(Φ)←0,0 foriinZdo</td><td colspan=\"3\">Initialize classification losses Embed images and texts Image classifier probabilities</td></tr><tr><td colspan=\"3\">Zui,Zt←f(Ui）,fθ(ti)</td></tr><tr><td colspan=\"2\">p ← softmax(Wizui +b1) PT ← sOftmax(WTZt +br)</td></tr><tr><td colspan=\"2\">Textclassifierprobabilities JTc(0) ← JTc(0)+ crossentropy(pr,yi) Text classification loss</td></tr><tr><td colspan=\"2\">JIc(𝜙) ← Jic(Φ)+ crossentropy(p1, yi) Image classification loss end for</td></tr><tr><td colspan=\"2\">JTR(Φ,0),J1R(Φ,0)←0,0 Initialize retrieval losses</td></tr><tr><td colspan=\"2\">foriinZdo 1 Text retrieval loss</td></tr><tr><td colspan=\"2\">JTR(Φ,0)←JTR(Φ,0)+ d(zu,Zt）+log∑ exp(-d(Zu,Ztj)) B j∈I</td></tr><tr><td colspan=\"2\">1-B JIR(Φ,0)←JIR(Φ,0)+ d(Zui,Zti) + log∑exp(-d(Zt , Zuj))|&gt; Image retrieval loss</td></tr><tr><td colspan=\"2\">j∈I</td></tr><tr><td colspan=\"2\">J(,θ)←λJTR(𝜙,0)+(1-λ)JIR(Φ,0) Add retrieval loss to the total loss J(Φ,θ)←(1-κ)J(,0)+(Jrc(0)+JIc()) Add classification loss to the total loss</td></tr></table>"
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    841,
+                    2088,
+                    859,
+                    2088,
+                    859,
+                    2112,
+                    841,
+                    2112
+                ],
+                "score": 0.65
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    841,
+                    2088,
+                    859,
+                    2088,
+                    859,
+                    2112,
+                    841,
+                    2112
+                ],
+                "score": 0.55
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    324,
+                    968,
+                    415,
+                    968,
+                    415,
+                    995,
+                    324,
+                    995
+                ],
+                "score": 0.289
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    516,
+                    1479,
+                    1182,
+                    1479,
+                    1182,
+                    1577,
+                    516,
+                    1577
+                ],
+                "score": 0.94,
+                "latex": "J _ { T R } ( \\phi , \\theta ) = - \\frac { 1 } { B } \\sum _ { i , j = 1 } ^ { B } \\ell _ { i , j } \\log p _ { \\phi , \\theta } ( i = j | v _ { i } , t _ { j } , \\{ t _ { j ^ { \\prime } } \\} _ { j ^ { \\prime } = 1 } ^ { B } ) ,"
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    520,
+                    1349,
+                    1178,
+                    1349,
+                    1178,
+                    1431,
+                    520,
+                    1431
+                ],
+                "score": 0.93,
+                "latex": "p _ { \\phi , \\theta } ( i = j | v _ { i } , t _ { j } , \\tau ) = \\frac { \\exp ( - d ( f _ { \\phi } ( v _ { i } ) , f _ { \\theta } ( t _ { j } ) ) ) } { \\sum _ { t _ { j ^ { \\prime } } \\in \\tau } \\exp ( - d ( f _ { \\phi } ( v _ { i } ) , f _ { \\theta } ( t _ { j ^ { \\prime } } ) ) ) } \\ ."
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    381,
+                    1716,
+                    1321,
+                    1716,
+                    1321,
+                    1827,
+                    381,
+                    1827
+                ],
+                "score": 0.93,
+                "latex": "J _ { T R } ( \\phi , \\theta ) = \\frac { 1 } { B } \\sum _ { i = 1 } ^ { B } \\left( d ( f _ { \\phi } ( v _ { i } ) , f _ { \\theta } ( t _ { i } ) ) + \\log \\Big [ \\sum _ { t _ { j ^ { \\prime } } \\in \\tau } \\exp ( - d ( f _ { \\phi } ( v _ { i } ) , f _ { \\theta } ( t _ { j ^ { \\prime } } ) ) ) \\Big ] \\right) ."
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1130,
+                    1245,
+                    1242,
+                    1245,
+                    1242,
+                    1279,
+                    1130,
+                    1279
+                ],
+                "score": 0.93,
+                "latex": "\\tau = \\{ t _ { j ^ { \\prime } } \\}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    901,
+                    2006,
+                    1007,
+                    2006,
+                    1007,
+                    2036,
+                    901,
+                    2036
+                ],
+                "score": 0.92,
+                "latex": "\\kappa \\in [ 0 , 1 ]"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    298,
+                    1864,
+                    409,
+                    1864,
+                    409,
+                    1898,
+                    298,
+                    1898
+                ],
+                "score": 0.92,
+                "latex": "J _ { I R } ( \\phi , \\bar { \\theta } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    895,
+                    1866,
+                    1001,
+                    1866,
+                    1001,
+                    1898,
+                    895,
+                    1898
+                ],
+                "score": 0.91,
+                "latex": "\\lambda \\in [ 0 , 1 ]"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    985,
+                    1587,
+                    1047,
+                    1587,
+                    1047,
+                    1618,
+                    985,
+                    1618
+                ],
+                "score": 0.91,
+                "latex": "i = j"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    507,
+                    398,
+                    591,
+                    398,
+                    591,
+                    430,
+                    507,
+                    430
+                ],
+                "score": 0.9,
+                "latex": "J ( \\phi , \\theta )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    373,
+                    1587,
+                    412,
+                    1587,
+                    412,
+                    1621,
+                    373,
+                    1621
+                ],
+                "score": 0.9,
+                "latex": "\\ell _ { i , j }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    711,
+                    1245,
+                    911,
+                    1245,
+                    911,
+                    1275,
+                    711,
+                    1275
+                ],
+                "score": 0.9,
+                "latex": "d : \\mathcal { Z } \\times \\mathcal { Z }  \\mathbb { R } ^ { + }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    856,
+                    1586,
+                    949,
+                    1586,
+                    949,
+                    1621,
+                    856,
+                    1621
+                ],
+                "score": 0.89,
+                "latex": "\\mathcal { \\ell } _ { i , j } = 1"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    924,
+                    1310,
+                    949,
+                    1310,
+                    949,
+                    1341,
+                    924,
+                    1341
+                ],
+                "score": 0.88,
+                "latex": "t _ { j }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    525,
+                    365,
+                    962,
+                    365,
+                    962,
+                    400,
+                    525,
+                    400
+                ],
+                "score": 0.87,
+                "latex": "\\mathcal { S } ^ { t r } = \\{ ( v _ { 1 } , t _ { 1 } , y _ { 1 } ) , \\dots , ( v _ { N } , t _ { N } , y _ { N } ) \\} _ { } ,"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1015,
+                    1278,
+                    1040,
+                    1278,
+                    1040,
+                    1310,
+                    1015,
+                    1310
+                ],
+                "score": 0.86,
+                "latex": "t _ { j }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    800,
+                    1311,
+                    826,
+                    1311,
+                    826,
+                    1337,
+                    800,
+                    1337
+                ],
+                "score": 0.86,
+                "latex": "v _ { i }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1174,
+                    1282,
+                    1199,
+                    1282,
+                    1199,
+                    1306,
+                    1174,
+                    1306
+                ],
+                "score": 0.86,
+                "latex": "v _ { i }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    726,
+                    260,
+                    767,
+                    260,
+                    767,
+                    289,
+                    726,
+                    289
+                ],
+                "score": 0.85,
+                "latex": "{ \\mathcal { S } } ^ { t r }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    326,
+                    705,
+                    646,
+                    705,
+                    646,
+                    740,
+                    326,
+                    740
+                ],
+                "score": 0.85,
+                "latex": "J _ { T R } ( \\phi , \\theta ) , J _ { I R } ( \\phi , \\theta )  0 , 0"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1302,
+                    1440,
+                    1326,
+                    1440,
+                    1326,
+                    1466,
+                    1302,
+                    1466
+                ],
+                "score": 0.84,
+                "latex": "B"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1219,
+                    261,
+                    1244,
+                    261,
+                    1244,
+                    289,
+                    1219,
+                    289
+                ],
+                "score": 0.82,
+                "latex": "C"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    779,
+                    261,
+                    805,
+                    261,
+                    805,
+                    289,
+                    779,
+                    289
+                ],
+                "score": 0.81,
+                "latex": "B"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    923,
+                    1248,
+                    950,
+                    1248,
+                    950,
+                    1276,
+                    923,
+                    1276
+                ],
+                "score": 0.81,
+                "latex": "v _ { i }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1047,
+                    292,
+                    1073,
+                    292,
+                    1073,
+                    318,
+                    1047,
+                    318
+                ],
+                "score": 0.8,
+                "latex": "B"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    568,
+                    1247,
+                    586,
+                    1247,
+                    586,
+                    1273,
+                    568,
+                    1273
+                ],
+                "score": 0.8,
+                "latex": "d"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    506,
+                    323,
+                    528,
+                    323,
+                    528,
+                    349,
+                    506,
+                    349
+                ],
+                "score": 0.78,
+                "latex": "s"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1090,
+                    366,
+                    1198,
+                    366,
+                    1198,
+                    400,
+                    1090,
+                    400
+                ],
+                "score": 0.77,
+                "latex": "\\kappa \\in [ 0 , 1 ]"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1305,
+                    232,
+                    1333,
+                    232,
+                    1333,
+                    257,
+                    1305,
+                    257
+                ],
+                "score": 0.77,
+                "latex": "N"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1079,
+                    429,
+                    1105,
+                    429,
+                    1105,
+                    455,
+                    1079,
+                    455
+                ],
+                "score": 0.76,
+                "latex": "B"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    366,
+                    741,
+                    379,
+                    741,
+                    379,
+                    765,
+                    366,
+                    765
+                ],
+                "score": 0.75,
+                "latex": "i"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    325,
+                    458,
+                    594,
+                    458,
+                    594,
+                    491,
+                    325,
+                    491
+                ],
+                "score": 0.75,
+                "latex": "J _ { T C } ( \\theta ) , J _ { I C } ( \\phi ) \\gets 0 , 0"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    367,
+                    493,
+                    380,
+                    493,
+                    380,
+                    515,
+                    367,
+                    515
+                ],
+                "score": 0.74,
+                "latex": "i"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    792,
+                    291,
+                    867,
+                    291,
+                    867,
+                    323,
+                    792,
+                    323
+                ],
+                "score": 0.72,
+                "latex": "( S , B )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    411,
+                    492,
+                    432,
+                    492,
+                    432,
+                    515,
+                    411,
+                    515
+                ],
+                "score": 0.71,
+                "latex": "\\mathcal { T }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    411,
+                    740,
+                    432,
+                    740,
+                    432,
+                    764,
+                    411,
+                    764
+                ],
+                "score": 0.68,
+                "latex": "\\mathcal { T }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    971,
+                    365,
+                    1079,
+                    365,
+                    1079,
+                    400,
+                    971,
+                    400
+                ],
+                "score": 0.65,
+                "latex": "\\lambda \\in [ 0 , 1 ]"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    324,
+                    428,
+                    383,
+                    428,
+                    383,
+                    456,
+                    324,
+                    456
+                ],
+                "score": 0.44,
+                "latex": "\\tau \\gets"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1044,
+                    462,
+                    1066,
+                    462,
+                    1066,
+                    487,
+                    1044,
+                    487
+                ],
+                "score": 0.44,
+                "latex": "\\triangleright"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    352,
+                    554,
+                    694,
+                    554,
+                    694,
+                    582,
+                    352,
+                    582
+                ],
+                "score": 0.42,
+                "latex": "p _ { I }  \\mathrm { s o f t m a x } ( \\mathbf { W } _ { I } \\mathbf { z } _ { v _ { i } } + \\mathbf { b } _ { I } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    981,
+                    432,
+                    1003,
+                    432,
+                    1003,
+                    455,
+                    981,
+                    455
+                ],
+                "score": 0.39,
+                "latex": "\\triangleright"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1103,
+                    650,
+                    1125,
+                    650,
+                    1125,
+                    674,
+                    1103,
+                    674
+                ],
+                "score": 0.34,
+                "latex": "\\triangleright"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    580,
+                    292,
+                    867,
+                    292,
+                    867,
+                    324,
+                    580,
+                    324
+                ],
+                "score": 0.34,
+                "latex": "\\mathrm { R A N D O M S A M P L E } ( S , B )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    324,
+                    428,
+                    782,
+                    428,
+                    782,
+                    460,
+                    324,
+                    460
+                ],
+                "score": 0.32,
+                "latex": "\\mathcal { T } \\gets \\mathrm { R A N D O M S A M P L E } ( \\{ 1 , \\dots , N \\} , B )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1098,
+                    711,
+                    1120,
+                    711,
+                    1120,
+                    734,
+                    1098,
+                    734
+                ],
+                "score": 0.32,
+                "latex": "\\triangleright"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    352,
+                    517,
+                    637,
+                    517,
+                    637,
+                    552,
+                    352,
+                    552
+                ],
+                "score": 0.32,
+                "latex": "\\mathbf { z } _ { v _ { i } } , \\mathbf { z } _ { t _ { i } } \\gets f _ { \\phi } ( v _ { i } ) , f _ { \\theta } ( t _ { i } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    353,
+                    611,
+                    625,
+                    611,
+                    625,
+                    644,
+                    353,
+                    644
+                ],
+                "score": 0.3,
+                "latex": "\\begin{array} { r } { J _ { T C } ( \\theta )  J _ { T C } ( \\theta ) + \\frac { 1 } { B } } \\end{array}"
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    335,
+                    428,
+                    1424,
+                    428,
+                    1424,
+                    970,
+                    335,
+                    970
+                ],
+                "score": 0.28,
+                "latex": "\\begin{array} { r l r l } & { \\mathbf { \\Phi } _ { \\mathrm { P } } ^ { \\dagger } \\{ \\mathbf { \\Phi } \\} } & { \\approx \\mathrm { S e l s . N o n } \\alpha \\mathrm { N o n } \\alpha \\mathrm { N o l A } \\mathrm { N o r E } \\{ [ \\mathrm { S i , ~ \\dots , ~ } ] , B \\} } & { \\quad \\mathrm { ~ s ~ o ~ s i d e l . ~ f l i m i z a t e s ~ f i n g ~ h o t ~ a t ~ e n t ~ h e t ~ } } \\\\ & { \\mathrm { f r o f ~ } \\{ \\alpha ( \\phi ) , \\mathcal { F } _ { L } ( \\phi ) - \\mathrm { 0 , ~ } \\mathbf { 0 } , \\mathbf { 0 } \\} } & { \\quad \\mathrm { ~ p ~ t i m i l a l s e ~ \\Lambda ~ f l i c i u l e s ~ a s i ~ e x ~ } } \\\\ & { \\mathrm { a r ~ t i m i z ~ \\phi } _ { \\mathrm { P } } \\{ \\alpha _ { \\mathrm { P } } , \\mathbf { 0 } , \\mathbf { \\Phi } _ { \\mathrm { P } } ^ { \\dagger } \\{ \\mathbf { \\Phi } \\} } & { \\quad \\mathrm { f o r ~ } \\{ \\alpha _ { \\mathrm { P } } , \\mathbf { 0 } \\} } & { \\quad \\mathrm { ~ b ~ E n b e l d ~ i n d ~ a g e s ~ c a l ~ t e x ~ } } \\\\ & { \\mathrm { p . ~ t e . ~ s o l u m i x ~ } \\{ \\mathrm { W i g . ~ z } _ { R } ( \\phi ) + \\mathrm { ~ b y ~ z } \\} } & { \\quad \\mathrm { ~ p ~ r a l s ~ e s l e a s i ~ e x ~ f r o r ~ a b l i t i s ~ } } \\\\ & { \\mathrm { f r e c t . ~ c o l s t ~ t o n s i n g ~ a r c e l t y ~ } \\mathrm { f o r ~ } \\{ [ \\mathrm { F i g . ~ \\alpha } ] , \\mathbf { 0 } \\} } & { \\quad \\mathrm { ~ b ~ r a t e s l e a s i ~ e x i s ~ i f r o ~ a b l i t i s ~ } } \\\\ & { \\mathrm { f r e c t . ~ c o l s t ~ t e n t e r a l ~ r e t o r ~ a l l ~ p . } } \\\\ &  \\mathcal { H } _ { \\mathrm { P } } ( \\phi ) + \\mathrm { ~ \\mathcal { F } _ { L } ( \\phi ) + \\frac { 1 } { 2 } \\mathrm { ~ p ~ c o u s i n t e x ~ p o r ~ } ( \\phi \\mathrm { r } , \\phi ) } & { \\quad \\mathrm { ~ b ~ m i l i s e ~ c o l s ~ s i f r o ~ a t ~ i n d ~ i n s e ~ } } \\\\ & { \\mathrm { f r o p ~ t i m i g h . ~ } } \\\\ & { \\mathrm { a t ~ a d y ~ c e t a i c ~ i s ~ e x p t ~ e n t ~ e n t ~ } } \\\\ & { \\mathrm { f r ~ d e t ~ i n f ~ a ~ s i g m a ~ } } \\\\ &  \\mathrm  f r ~ d e t ~ i n f ~ a ~ s i \\end{array}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    600,
+                    428,
+                    781,
+                    428,
+                    781,
+                    460,
+                    600,
+                    460
+                ],
+                "score": 0.28,
+                "latex": "( \\{ 1 , \\ldots , N \\} , B )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1047,
+                    554,
+                    1068,
+                    554,
+                    1068,
+                    577,
+                    1047,
+                    577
+                ],
+                "score": 0.26,
+                "latex": "\\triangleright"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    72.0,
+                    857.0,
+                    72.0,
+                    857.0,
+                    108.0,
+                    297.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1126.0,
+                    703.0,
+                    1126.0,
+                    703.0,
+                    1162.0,
+                    295.0,
+                    1162.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    838.0,
+                    2085.0,
+                    862.0,
+                    2085.0,
+                    862.0,
+                    2118.0,
+                    838.0,
+                    2118.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    838.0,
+                    2085.0,
+                    862.0,
+                    2085.0,
+                    862.0,
+                    2118.0,
+                    838.0,
+                    2118.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    964.0,
+                    419.0,
+                    964.0,
+                    419.0,
+                    1000.0,
+                    322.0,
+                    1000.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1183.0,
+                    1403.0,
+                    1183.0,
+                    1403.0,
+                    1218.0,
+                    294.0,
+                    1218.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1212.0,
+                    1403.0,
+                    1212.0,
+                    1403.0,
+                    1250.0,
+                    295.0,
+                    1250.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1240.0,
+                    567.0,
+                    1240.0,
+                    567.0,
+                    1283.0,
+                    294.0,
+                    1283.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    587.0,
+                    1240.0,
+                    710.0,
+                    1240.0,
+                    710.0,
+                    1283.0,
+                    587.0,
+                    1283.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    912.0,
+                    1240.0,
+                    922.0,
+                    1240.0,
+                    922.0,
+                    1283.0,
+                    912.0,
+                    1283.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    951.0,
+                    1240.0,
+                    1129.0,
+                    1240.0,
+                    1129.0,
+                    1283.0,
+                    951.0,
+                    1283.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1243.0,
+                    1240.0,
+                    1405.0,
+                    1240.0,
+                    1405.0,
+                    1283.0,
+                    1243.0,
+                    1283.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1273.0,
+                    1014.0,
+                    1273.0,
+                    1014.0,
+                    1311.0,
+                    294.0,
+                    1311.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1041.0,
+                    1273.0,
+                    1173.0,
+                    1273.0,
+                    1173.0,
+                    1311.0,
+                    1041.0,
+                    1311.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1200.0,
+                    1273.0,
+                    1405.0,
+                    1273.0,
+                    1405.0,
+                    1311.0,
+                    1200.0,
+                    1311.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1306.0,
+                    799.0,
+                    1306.0,
+                    799.0,
+                    1343.0,
+                    295.0,
+                    1343.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    827.0,
+                    1306.0,
+                    923.0,
+                    1306.0,
+                    923.0,
+                    1343.0,
+                    827.0,
+                    1343.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    950.0,
+                    1306.0,
+                    1377.0,
+                    1306.0,
+                    1377.0,
+                    1343.0,
+                    950.0,
+                    1343.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1911.0,
+                    1410.0,
+                    1911.0,
+                    1410.0,
+                    1947.0,
+                    293.0,
+                    1947.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1942.0,
+                    1405.0,
+                    1942.0,
+                    1405.0,
+                    1977.0,
+                    293.0,
+                    1977.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1974.0,
+                    1405.0,
+                    1974.0,
+                    1405.0,
+                    2007.0,
+                    295.0,
+                    2007.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    2002.0,
+                    900.0,
+                    2002.0,
+                    900.0,
+                    2039.0,
+                    293.0,
+                    2039.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1008.0,
+                    2002.0,
+                    1299.0,
+                    2002.0,
+                    1299.0,
+                    2039.0,
+                    1008.0,
+                    2039.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1584.0,
+                    372.0,
+                    1584.0,
+                    372.0,
+                    1621.0,
+                    294.0,
+                    1621.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    413.0,
+                    1584.0,
+                    855.0,
+                    1584.0,
+                    855.0,
+                    1621.0,
+                    413.0,
+                    1621.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    950.0,
+                    1584.0,
+                    984.0,
+                    1584.0,
+                    984.0,
+                    1621.0,
+                    950.0,
+                    1621.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1048.0,
+                    1584.0,
+                    1405.0,
+                    1584.0,
+                    1405.0,
+                    1621.0,
+                    1048.0,
+                    1621.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1616.0,
+                    1406.0,
+                    1616.0,
+                    1406.0,
+                    1650.0,
+                    294.0,
+                    1650.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1648.0,
+                    1405.0,
+                    1648.0,
+                    1405.0,
+                    1680.0,
+                    294.0,
+                    1680.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1678.0,
+                    731.0,
+                    1678.0,
+                    731.0,
+                    1711.0,
+                    295.0,
+                    1711.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1831.0,
+                    1407.0,
+                    1831.0,
+                    1407.0,
+                    1870.0,
+                    294.0,
+                    1870.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    410.0,
+                    1863.0,
+                    894.0,
+                    1863.0,
+                    894.0,
+                    1900.0,
+                    410.0,
+                    1900.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1002.0,
+                    1863.0,
+                    1294.0,
+                    1863.0,
+                    1294.0,
+                    1900.0,
+                    1002.0,
+                    1900.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1435.0,
+                    1301.0,
+                    1435.0,
+                    1301.0,
+                    1473.0,
+                    292.0,
+                    1473.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1327.0,
+                    1435.0,
+                    1339.0,
+                    1435.0,
+                    1339.0,
+                    1473.0,
+                    1327.0,
+                    1473.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    227.0,
+                    1304.0,
+                    227.0,
+                    1304.0,
+                    263.0,
+                    294.0,
+                    263.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1334.0,
+                    227.0,
+                    1405.0,
+                    227.0,
+                    1405.0,
+                    263.0,
+                    1334.0,
+                    263.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    256.0,
+                    725.0,
+                    256.0,
+                    725.0,
+                    296.0,
+                    293.0,
+                    296.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    768.0,
+                    256.0,
+                    778.0,
+                    256.0,
+                    778.0,
+                    296.0,
+                    768.0,
+                    296.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    806.0,
+                    256.0,
+                    1218.0,
+                    256.0,
+                    1218.0,
+                    296.0,
+                    806.0,
+                    296.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1245.0,
+                    256.0,
+                    1409.0,
+                    256.0,
+                    1409.0,
+                    296.0,
+                    1245.0,
+                    296.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    290.0,
+                    579.0,
+                    290.0,
+                    579.0,
+                    323.0,
+                    295.0,
+                    323.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    868.0,
+                    290.0,
+                    1046.0,
+                    290.0,
+                    1046.0,
+                    323.0,
+                    868.0,
+                    323.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1074.0,
+                    290.0,
+                    1406.0,
+                    290.0,
+                    1406.0,
+                    323.0,
+                    1074.0,
+                    323.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    322.0,
+                    505.0,
+                    322.0,
+                    505.0,
+                    355.0,
+                    294.0,
+                    355.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    529.0,
+                    322.0,
+                    778.0,
+                    322.0,
+                    778.0,
+                    355.0,
+                    529.0,
+                    355.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 2,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1727,
+                    1405,
+                    1727,
+                    1405,
+                    2035,
+                    297,
+                    2035
+                ],
+                "score": 0.984
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    296,
+                    1435,
+                    1405,
+                    1435,
+                    1405,
+                    1715,
+                    296,
+                    1715
+                ],
+                "score": 0.984
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    296,
+                    485,
+                    1405,
+                    485,
+                    1405,
+                    704,
+                    296,
+                    704
+                ],
+                "score": 0.982
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    296,
+                    716,
+                    1404,
+                    716,
+                    1404,
+                    903,
+                    296,
+                    903
+                ],
+                "score": 0.981
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    296,
+                    286,
+                    1404,
+                    286,
+                    1404,
+                    399,
+                    296,
+                    399
+                ],
+                "score": 0.977
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1085,
+                    1404,
+                    1085,
+                    1404,
+                    1180,
+                    298,
+                    1180
+                ],
+                "score": 0.976
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    548,
+                    1196,
+                    1141,
+                    1196,
+                    1141,
+                    1278,
+                    548,
+                    1278
+                ],
+                "score": 0.955
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1007,
+                    1403,
+                    1007,
+                    1403,
+                    1072,
+                    297,
+                    1072
+                ],
+                "score": 0.951
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    700,
+                    415,
+                    994,
+                    415,
+                    994,
+                    468,
+                    700,
+                    468
+                ],
+                "score": 0.947
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    386,
+                    920,
+                    1304,
+                    920,
+                    1304,
+                    990,
+                    386,
+                    990
+                ],
+                "score": 0.941
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1295,
+                    1346,
+                    1295,
+                    1346,
+                    1329,
+                    298,
+                    1329
+                ],
+                "score": 0.934
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    327,
+                    1345,
+                    1335,
+                    1345,
+                    1335,
+                    1416,
+                    327,
+                    1416
+                ],
+                "score": 0.926
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    297,
+                    74,
+                    858,
+                    74,
+                    858,
+                    106,
+                    297,
+                    106
+                ],
+                "score": 0.922
+            },
+            {
+                "category_id": 9,
+                "poly": [
+                    1366,
+                    420,
+                    1400,
+                    420,
+                    1400,
+                    449,
+                    1366,
+                    449
+                ],
+                "score": 0.883
+            },
+            {
+                "category_id": 9,
+                "poly": [
+                    1366,
+                    1223,
+                    1400,
+                    1223,
+                    1400,
+                    1252,
+                    1366,
+                    1252
+                ],
+                "score": 0.88
+            },
+            {
+                "category_id": 9,
+                "poly": [
+                    1366,
+                    939,
+                    1400,
+                    939,
+                    1400,
+                    969,
+                    1366,
+                    969
+                ],
+                "score": 0.853
+            },
+            {
+                "category_id": 9,
+                "poly": [
+                    1367,
+                    1366,
+                    1400,
+                    1366,
+                    1400,
+                    1398,
+                    1367,
+                    1398
+                ],
+                "score": 0.829
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    840,
+                    2088,
+                    857,
+                    2088,
+                    857,
+                    2111,
+                    840,
+                    2111
+                ],
+                "score": 0.79
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    298,
+                    228,
+                    1104,
+                    228,
+                    1104,
+                    263,
+                    298,
+                    263
+                ],
+                "score": 0.664
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    228,
+                    1104,
+                    228,
+                    1104,
+                    263,
+                    298,
+                    263
+                ],
+                "score": 0.275
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    699,
+                    414,
+                    999,
+                    414,
+                    999,
+                    469,
+                    699,
+                    469
+                ],
+                "score": 0.94,
+                "latex": "\\boldsymbol { \\widehat { y } } = \\arg \\operatorname* { m i n } _ { \\boldsymbol { y } \\in \\mathcal { V } } d ( \\mathbf { z } _ { v } , \\mathbf { p } ( \\boldsymbol { y } ) ) ."
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    568,
+                    366,
+                    699,
+                    366,
+                    699,
+                    401,
+                    568,
+                    401
+                ],
+                "score": 0.93,
+                "latex": "{ \\bf z } _ { v } = f _ { \\phi } ( v )"
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    388,
+                    918,
+                    1309,
+                    918,
+                    1309,
+                    991,
+                    388,
+                    991
+                ],
+                "score": 0.93,
+                "latex": "\\mathbb { P } \\{ \\boldsymbol { { \\widehat { y } } } \\in { \\mathcal { Y } } ^ { t r } | y _ { v } \\in { \\mathcal { Y } } ^ { t s } \\} = \\mathbb { P } \\left\\{ \\operatorname* { m i n } _ { y \\in { \\mathcal { Y } } ^ { t r } } d ( \\mathbf { z } _ { v } , \\mathbf { p } ( y ) ) < \\operatorname* { m i n } _ { y \\in { \\mathcal { Y } } ^ { t s } } d ( \\mathbf { z } _ { v } , \\mathbf { p } ( y ) ) \\mid y _ { v } \\in { \\mathcal { Y } } ^ { t s } \\right\\} ."
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    531,
+                    1468,
+                    804,
+                    1468,
+                    804,
+                    1502,
+                    531,
+                    1502
+                ],
+                "score": 0.93,
+                "latex": "\\mathbb { P } \\{ \\widehat { y } _ { \\alpha } \\in \\mathcal { V } ^ { t \\bar { r } } | y _ { v } \\in \\mathcal { V } ^ { t s } \\}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    347,
+                    1116,
+                    611,
+                    1116,
+                    611,
+                    1150,
+                    347,
+                    1150
+                ],
+                "score": 0.93,
+                "latex": "\\mathbb { P } \\{ \\hat { \\mathcal { Y } } \\in \\mathcal { Y } ^ { t s } | y _ { v } \\in \\mathcal { Y } ^ { t \\bar { r } } \\}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1146,
+                    640,
+                    1343,
+                    640,
+                    1343,
+                    674,
+                    1146,
+                    674
+                ],
+                "score": 0.92,
+                "latex": "\\mathbf { H } = \\mathbf { u } \\cdot \\mathbf { s } / ( \\mathbf { u } + \\mathbf { s } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    298,
+                    1528,
+                    466,
+                    1528,
+                    466,
+                    1563,
+                    298,
+                    1563
+                ],
+                "score": 0.92,
+                "latex": "\\mathcal { V } ^ { t r } | y _ { v } \\in \\mathcal { V } ^ { t r } \\big \\}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1139,
+                    1819,
+                    1401,
+                    1819,
+                    1401,
+                    1854,
+                    1139,
+                    1854
+                ],
+                "score": 0.92,
+                "latex": "\\mathbb { P } \\{ \\widehat { y } _ { \\alpha } \\in \\mathcal { V } ^ { t r } | y _ { v } \\in \\mathcal { V } ^ { t s } \\}"
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    550,
+                    1195,
+                    1150,
+                    1195,
+                    1150,
+                    1281,
+                    550,
+                    1281
+                ],
+                "score": 0.92,
+                "latex": "d _ { \\alpha } ( \\mathbf { z } _ { v } , \\mathbf { p } ( y ) ) = \\left\\{ \\begin{array} { l l } { ( 1 + \\alpha ) d ( \\mathbf { z } _ { v } , \\mathbf { p } ( y ) ) , } & { \\mathrm { i f ~ } y \\in \\mathcal { Y } ^ { t r } } \\\\ { d ( \\mathbf { z } _ { v } , \\mathbf { p } ( y ) ) , } & { \\mathrm { o t h e r w i s e } } \\end{array} \\right. ."
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    298,
+                    1467,
+                    469,
+                    1467,
+                    469,
+                    1502,
+                    298,
+                    1502
+                ],
+                "score": 0.92,
+                "latex": "\\mathcal { V } ^ { t r } | y _ { v } \\in \\mathcal { V } ^ { t s } \\}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    410,
+                    1758,
+                    676,
+                    1758,
+                    676,
+                    1792,
+                    410,
+                    1792
+                ],
+                "score": 0.92,
+                "latex": "\\mathbb { P } \\mathbf { \\bar { \\{ y _ { \\alpha } \\in \\mathcal { V } ^ { t r } \\vert y _ { v } \\in \\mathcal { V } ^ { t s } \\} } }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    639,
+                    748,
+                    891,
+                    748,
+                    891,
+                    782,
+                    639,
+                    782
+                ],
+                "score": 0.92,
+                "latex": "\\mathbb { P } \\{ \\widehat { y } \\in \\mathcal { y } ^ { t r } | y _ { v } \\in \\mathcal { y } ^ { t s } \\}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    989,
+                    1116,
+                    1090,
+                    1116,
+                    1090,
+                    1145,
+                    989,
+                    1145
+                ],
+                "score": 0.92,
+                "latex": "\\alpha \\in \\mathbb { R } ^ { + }"
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    328,
+                    1343,
+                    1335,
+                    1343,
+                    1335,
+                    1417,
+                    328,
+                    1417
+                ],
+                "score": 0.92,
+                "latex": "\\mathbb { P } \\{ \\widehat { y } _ { \\alpha } \\in \\mathcal { Y } ^ { t r } | y _ { v } \\in \\mathcal { Y } ^ { t s } \\} = \\mathbb { P } \\left\\{ \\left( 1 + \\alpha \\right) \\operatorname* { m i n } _ { y \\in \\mathcal { Y } ^ { t r } } d ( \\mathbf { z } _ { v } , \\mathbf { p } ( y ) ) < \\operatorname* { m i n } _ { y \\in \\mathcal { Y } ^ { t s } } d ( \\mathbf { z } _ { v } , \\mathbf { p } ( y ) ) | y _ { v } \\in \\mathcal { Y } ^ { t s } \\right\\} \\ ,"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    299,
+                    322,
+                    616,
+                    322,
+                    616,
+                    370,
+                    299,
+                    370
+                ],
+                "score": 0.92,
+                "latex": "\\begin{array} { r } { \\frac { 1 } { | T _ { y } | } \\sum _ { t _ { i } \\in \\mathcal { T } _ { y } } f _ { \\theta } ( t _ { i } ) \\ | \\ y \\in \\mathcal { V } \\} } \\end{array}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1154,
+                    1084,
+                    1401,
+                    1084,
+                    1401,
+                    1120,
+                    1154,
+                    1120
+                ],
+                "score": 0.91,
+                "latex": "\\mathbb { P } \\{ \\widehat { y } \\in \\mathcal { y } ^ { t r } | y _ { v } \\in \\mathcal { y } ^ { t s } \\}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    632,
+                    288,
+                    688,
+                    288,
+                    688,
+                    322,
+                    632,
+                    322
+                ],
+                "score": 0.91,
+                "latex": "\\mathbf p ( y )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    955,
+                    1941,
+                    998,
+                    1941,
+                    998,
+                    1972,
+                    955,
+                    1972
+                ],
+                "score": 0.91,
+                "latex": "\\mathcal { V } ^ { t r }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1306,
+                    1499,
+                    1403,
+                    1499,
+                    1403,
+                    1533,
+                    1306,
+                    1533
+                ],
+                "score": 0.91,
+                "latex": "\\mathbb { P } \\{ \\widehat { y } _ { \\alpha } \\in"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1213,
+                    748,
+                    1403,
+                    748,
+                    1403,
+                    782,
+                    1213,
+                    782
+                ],
+                "score": 0.91,
+                "latex": "\\mathbb { P } \\{ \\widehat { y } \\in \\mathcal { y } ^ { t s } | y _ { v } \\in"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    752,
+                    1850,
+                    1015,
+                    1850,
+                    1015,
+                    1884,
+                    752,
+                    1884
+                ],
+                "score": 0.91,
+                "latex": "\\mathbb { P } \\{ \\widehat { y } _ { \\alpha } \\in \\mathcal { V } ^ { t s } | y _ { v } \\in \\mathcal { V } ^ { t r } \\}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    608,
+                    1789,
+                    883,
+                    1789,
+                    883,
+                    1823,
+                    608,
+                    1823
+                ],
+                "score": 0.9,
+                "latex": "\\mathbb { P } \\{ \\widehat { y } _ { \\alpha } ^ { } \\in \\mathcal { V } ^ { t s } | y _ { v } ^ { - } \\in \\mathcal { V } ^ { t r } \\}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    297,
+                    778,
+                    355,
+                    778,
+                    355,
+                    812,
+                    297,
+                    812
+                ],
+                "score": 0.9,
+                "latex": "\\mathcal { V } ^ { t r } \\}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    746,
+                    1149,
+                    815,
+                    1149,
+                    815,
+                    1177,
+                    746,
+                    1177
+                ],
+                "score": 0.89,
+                "latex": "1 + \\alpha"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1076,
+                    1297,
+                    1107,
+                    1297,
+                    1107,
+                    1329,
+                    1076,
+                    1329
+                ],
+                "score": 0.89,
+                "latex": "\\widehat { y } _ { \\alpha }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1115,
+                    1941,
+                    1158,
+                    1941,
+                    1158,
+                    1973,
+                    1115,
+                    1973
+                ],
+                "score": 0.89,
+                "latex": "\\mathcal { V } ^ { t s }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1117,
+                    1731,
+                    1189,
+                    1731,
+                    1189,
+                    1758,
+                    1117,
+                    1758
+                ],
+                "score": 0.89,
+                "latex": "\\alpha > 0"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1030,
+                    289,
+                    1062,
+                    289,
+                    1062,
+                    326,
+                    1030,
+                    326
+                ],
+                "score": 0.89,
+                "latex": "\\mathcal { T } _ { y }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1351,
+                    1148,
+                    1384,
+                    1148,
+                    1384,
+                    1178,
+                    1351,
+                    1178
+                ],
+                "score": 0.88,
+                "latex": "d _ { \\alpha }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1301,
+                    287,
+                    1406,
+                    287,
+                    1406,
+                    323,
+                    1301,
+                    323
+                ],
+                "score": 0.88,
+                "latex": "\\mathbf { \\{ p ( }  y \\mathbf { ) = }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    966,
+                    1436,
+                    1401,
+                    1436,
+                    1401,
+                    1473,
+                    966,
+                    1473
+                ],
+                "score": 0.86,
+                "latex": "\\mathbb { P } \\{ \\widehat { y } _ { \\alpha _ { 1 } } ~ \\in ~ \\mathcal { Y } ^ { t r } | y _ { v } ~ \\in ~ \\mathcal { Y } ^ { t s } \\} ~ \\geq ~ \\mathbb { P } \\{ \\widehat { y } _ { \\alpha _ { 2 } } ~ \\in ~"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    296,
+                    1619,
+                    902,
+                    1619,
+                    902,
+                    1654,
+                    296,
+                    1654
+                ],
+                "score": 0.86,
+                "latex": "\\mathbb { P } \\{ \\hat { y } _ { \\alpha } ^ { \\sim } \\in \\mathcal { V } ^ { t s } | y _ { v } \\in \\mathcal { V } ^ { t r } \\} = 1 - \\mathbb { P } \\{ \\hat { y } _ { \\alpha } \\in \\mathcal { V } ^ { \\hat { t r } } | y _ { v } \\in \\mathcal { V } ^ { t r } \\}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    791,
+                    723,
+                    820,
+                    723,
+                    820,
+                    749,
+                    791,
+                    749
+                ],
+                "score": 0.83,
+                "latex": "y _ { v }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    788,
+                    1439,
+                    954,
+                    1439,
+                    954,
+                    1471,
+                    788,
+                    1471
+                ],
+                "score": 0.81,
+                "latex": "0 \\leq \\alpha _ { 1 } \\leq \\alpha _ { 2 }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    327,
+                    1658,
+                    349,
+                    1658,
+                    349,
+                    1679,
+                    327,
+                    1679
+                ],
+                "score": 0.79,
+                "latex": "\\alpha"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1289,
+                    1475,
+                    1311,
+                    1475,
+                    1311,
+                    1496,
+                    1289,
+                    1496
+                ],
+                "score": 0.79,
+                "latex": "\\alpha"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    650,
+                    1856,
+                    671,
+                    1856,
+                    671,
+                    1878,
+                    650,
+                    1878
+                ],
+                "score": 0.78,
+                "latex": "\\alpha"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    623,
+                    1887,
+                    644,
+                    1887,
+                    644,
+                    1909,
+                    623,
+                    1909
+                ],
+                "score": 0.77,
+                "latex": "\\alpha"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1322,
+                    1917,
+                    1343,
+                    1917,
+                    1343,
+                    1939,
+                    1322,
+                    1939
+                ],
+                "score": 0.77,
+                "latex": "\\alpha"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    372,
+                    374,
+                    390,
+                    374,
+                    390,
+                    394,
+                    372,
+                    394
+                ],
+                "score": 0.77,
+                "latex": "v"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    749,
+                    2013,
+                    769,
+                    2013,
+                    769,
+                    2030,
+                    749,
+                    2030
+                ],
+                "score": 0.77,
+                "latex": "\\alpha"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    619,
+                    1506,
+                    641,
+                    1506,
+                    641,
+                    1528,
+                    619,
+                    1528
+                ],
+                "score": 0.77,
+                "latex": "\\alpha"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1273,
+                    1947,
+                    1296,
+                    1947,
+                    1296,
+                    1970,
+                    1273,
+                    1970
+                ],
+                "score": 0.76,
+                "latex": "\\alpha"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1127,
+                    1596,
+                    1147,
+                    1596,
+                    1147,
+                    1618,
+                    1127,
+                    1618
+                ],
+                "score": 0.75,
+                "latex": "\\alpha"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1253,
+                    1625,
+                    1274,
+                    1625,
+                    1274,
+                    1648,
+                    1253,
+                    1648
+                ],
+                "score": 0.73,
+                "latex": "\\alpha"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1025,
+                    1535,
+                    1044,
+                    1535,
+                    1044,
+                    1557,
+                    1025,
+                    1557
+                ],
+                "score": 0.69,
+                "latex": "v"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1148,
+                    723,
+                    1165,
+                    723,
+                    1165,
+                    745,
+                    1148,
+                    745
+                ],
+                "score": 0.68,
+                "latex": "v"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1272,
+                    293,
+                    1290,
+                    293,
+                    1290,
+                    321,
+                    1272,
+                    321
+                ],
+                "score": 0.63,
+                "latex": "y"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    843,
+                    672,
+                    868,
+                    672,
+                    868,
+                    698,
+                    843,
+                    698
+                ],
+                "score": 0.58,
+                "latex": "\\mathbf { H }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1123,
+                    676,
+                    1143,
+                    676,
+                    1143,
+                    698,
+                    1123,
+                    698
+                ],
+                "score": 0.44,
+                "latex": "\\mathbf { u }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    824,
+                    642,
+                    859,
+                    642,
+                    859,
+                    671,
+                    824,
+                    671
+                ],
+                "score": 0.31,
+                "latex": "\\mathbf { \\Pi } ^ { ( \\mathbf { u } ) }"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    72.0,
+                    858.0,
+                    72.0,
+                    858.0,
+                    109.0,
+                    297.0,
+                    109.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    838.0,
+                    2086.0,
+                    861.0,
+                    2086.0,
+                    861.0,
+                    2118.0,
+                    838.0,
+                    2118.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    228.0,
+                    1106.0,
+                    228.0,
+                    1106.0,
+                    267.0,
+                    294.0,
+                    267.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1728.0,
+                    1116.0,
+                    1728.0,
+                    1116.0,
+                    1764.0,
+                    295.0,
+                    1764.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1190.0,
+                    1728.0,
+                    1405.0,
+                    1728.0,
+                    1405.0,
+                    1764.0,
+                    1190.0,
+                    1764.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1756.0,
+                    409.0,
+                    1756.0,
+                    409.0,
+                    1796.0,
+                    292.0,
+                    1796.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    677.0,
+                    1756.0,
+                    1406.0,
+                    1756.0,
+                    1406.0,
+                    1796.0,
+                    677.0,
+                    1796.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1787.0,
+                    607.0,
+                    1787.0,
+                    607.0,
+                    1826.0,
+                    294.0,
+                    1826.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    884.0,
+                    1787.0,
+                    1407.0,
+                    1787.0,
+                    1407.0,
+                    1826.0,
+                    884.0,
+                    1826.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1817.0,
+                    1138.0,
+                    1817.0,
+                    1138.0,
+                    1855.0,
+                    292.0,
+                    1855.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1402.0,
+                    1817.0,
+                    1405.0,
+                    1817.0,
+                    1405.0,
+                    1855.0,
+                    1402.0,
+                    1855.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1848.0,
+                    649.0,
+                    1848.0,
+                    649.0,
+                    1887.0,
+                    291.0,
+                    1887.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    672.0,
+                    1848.0,
+                    751.0,
+                    1848.0,
+                    751.0,
+                    1887.0,
+                    672.0,
+                    1887.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1016.0,
+                    1848.0,
+                    1406.0,
+                    1848.0,
+                    1406.0,
+                    1887.0,
+                    1016.0,
+                    1887.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1881.0,
+                    622.0,
+                    1881.0,
+                    622.0,
+                    1915.0,
+                    294.0,
+                    1915.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    645.0,
+                    1881.0,
+                    1405.0,
+                    1881.0,
+                    1405.0,
+                    1915.0,
+                    645.0,
+                    1915.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1910.0,
+                    1321.0,
+                    1910.0,
+                    1321.0,
+                    1946.0,
+                    292.0,
+                    1946.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1344.0,
+                    1910.0,
+                    1405.0,
+                    1910.0,
+                    1405.0,
+                    1946.0,
+                    1344.0,
+                    1946.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1938.0,
+                    954.0,
+                    1938.0,
+                    954.0,
+                    1978.0,
+                    292.0,
+                    1978.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    999.0,
+                    1938.0,
+                    1114.0,
+                    1938.0,
+                    1114.0,
+                    1978.0,
+                    999.0,
+                    1978.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1159.0,
+                    1938.0,
+                    1272.0,
+                    1938.0,
+                    1272.0,
+                    1978.0,
+                    1159.0,
+                    1978.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1297.0,
+                    1938.0,
+                    1406.0,
+                    1938.0,
+                    1406.0,
+                    1978.0,
+                    1297.0,
+                    1978.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1972.0,
+                    1405.0,
+                    1972.0,
+                    1405.0,
+                    2006.0,
+                    294.0,
+                    2006.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    2002.0,
+                    748.0,
+                    2002.0,
+                    748.0,
+                    2038.0,
+                    295.0,
+                    2038.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    770.0,
+                    2002.0,
+                    940.0,
+                    2002.0,
+                    940.0,
+                    2038.0,
+                    770.0,
+                    2038.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1433.0,
+                    787.0,
+                    1433.0,
+                    787.0,
+                    1477.0,
+                    291.0,
+                    1477.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    955.0,
+                    1433.0,
+                    965.0,
+                    1433.0,
+                    965.0,
+                    1477.0,
+                    955.0,
+                    1477.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1402.0,
+                    1433.0,
+                    1407.0,
+                    1433.0,
+                    1407.0,
+                    1477.0,
+                    1402.0,
+                    1477.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1460.0,
+                    297.0,
+                    1460.0,
+                    297.0,
+                    1509.0,
+                    291.0,
+                    1509.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    470.0,
+                    1460.0,
+                    530.0,
+                    1460.0,
+                    530.0,
+                    1509.0,
+                    470.0,
+                    1509.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    805.0,
+                    1460.0,
+                    1288.0,
+                    1460.0,
+                    1288.0,
+                    1509.0,
+                    805.0,
+                    1509.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1312.0,
+                    1460.0,
+                    1410.0,
+                    1460.0,
+                    1410.0,
+                    1509.0,
+                    1312.0,
+                    1509.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1497.0,
+                    618.0,
+                    1497.0,
+                    618.0,
+                    1534.0,
+                    294.0,
+                    1534.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    642.0,
+                    1497.0,
+                    1305.0,
+                    1497.0,
+                    1305.0,
+                    1534.0,
+                    642.0,
+                    1534.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1525.0,
+                    297.0,
+                    1525.0,
+                    297.0,
+                    1566.0,
+                    293.0,
+                    1566.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    467.0,
+                    1525.0,
+                    1024.0,
+                    1525.0,
+                    1024.0,
+                    1566.0,
+                    467.0,
+                    1566.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1045.0,
+                    1525.0,
+                    1408.0,
+                    1525.0,
+                    1408.0,
+                    1566.0,
+                    1045.0,
+                    1566.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1558.0,
+                    1408.0,
+                    1558.0,
+                    1408.0,
+                    1599.0,
+                    293.0,
+                    1599.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1590.0,
+                    1126.0,
+                    1590.0,
+                    1126.0,
+                    1626.0,
+                    294.0,
+                    1626.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1148.0,
+                    1590.0,
+                    1405.0,
+                    1590.0,
+                    1405.0,
+                    1626.0,
+                    1148.0,
+                    1626.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1615.0,
+                    295.0,
+                    1615.0,
+                    295.0,
+                    1659.0,
+                    291.0,
+                    1659.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    903.0,
+                    1615.0,
+                    1252.0,
+                    1615.0,
+                    1252.0,
+                    1659.0,
+                    903.0,
+                    1659.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1275.0,
+                    1615.0,
+                    1411.0,
+                    1615.0,
+                    1411.0,
+                    1659.0,
+                    1275.0,
+                    1659.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1653.0,
+                    326.0,
+                    1653.0,
+                    326.0,
+                    1686.0,
+                    294.0,
+                    1686.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    350.0,
+                    1653.0,
+                    1405.0,
+                    1653.0,
+                    1405.0,
+                    1686.0,
+                    350.0,
+                    1686.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1679.0,
+                    970.0,
+                    1679.0,
+                    970.0,
+                    1718.0,
+                    293.0,
+                    1718.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    487.0,
+                    1408.0,
+                    487.0,
+                    1408.0,
+                    522.0,
+                    294.0,
+                    522.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    519.0,
+                    1405.0,
+                    519.0,
+                    1405.0,
+                    552.0,
+                    293.0,
+                    552.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    547.0,
+                    1405.0,
+                    547.0,
+                    1405.0,
+                    585.0,
+                    293.0,
+                    585.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    582.0,
+                    1403.0,
+                    582.0,
+                    1403.0,
+                    613.0,
+                    295.0,
+                    613.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    609.0,
+                    1405.0,
+                    609.0,
+                    1405.0,
+                    644.0,
+                    294.0,
+                    644.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    641.0,
+                    823.0,
+                    641.0,
+                    823.0,
+                    675.0,
+                    295.0,
+                    675.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    860.0,
+                    641.0,
+                    1145.0,
+                    641.0,
+                    1145.0,
+                    675.0,
+                    860.0,
+                    675.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1344.0,
+                    641.0,
+                    1403.0,
+                    641.0,
+                    1403.0,
+                    675.0,
+                    1344.0,
+                    675.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    671.0,
+                    842.0,
+                    671.0,
+                    842.0,
+                    705.0,
+                    294.0,
+                    705.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    869.0,
+                    671.0,
+                    1122.0,
+                    671.0,
+                    1122.0,
+                    705.0,
+                    869.0,
+                    705.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1144.0,
+                    671.0,
+                    1218.0,
+                    671.0,
+                    1218.0,
+                    705.0,
+                    1144.0,
+                    705.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    716.0,
+                    790.0,
+                    716.0,
+                    790.0,
+                    752.0,
+                    294.0,
+                    752.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    821.0,
+                    716.0,
+                    1147.0,
+                    716.0,
+                    1147.0,
+                    752.0,
+                    821.0,
+                    752.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1166.0,
+                    716.0,
+                    1405.0,
+                    716.0,
+                    1405.0,
+                    752.0,
+                    1166.0,
+                    752.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    290.0,
+                    740.0,
+                    638.0,
+                    740.0,
+                    638.0,
+                    788.0,
+                    290.0,
+                    788.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    892.0,
+                    740.0,
+                    1212.0,
+                    740.0,
+                    1212.0,
+                    788.0,
+                    892.0,
+                    788.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1404.0,
+                    740.0,
+                    1409.0,
+                    740.0,
+                    1409.0,
+                    788.0,
+                    1404.0,
+                    788.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    774.0,
+                    296.0,
+                    774.0,
+                    296.0,
+                    816.0,
+                    293.0,
+                    816.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    356.0,
+                    774.0,
+                    1406.0,
+                    774.0,
+                    1406.0,
+                    816.0,
+                    356.0,
+                    816.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    810.0,
+                    1404.0,
+                    810.0,
+                    1404.0,
+                    842.0,
+                    295.0,
+                    842.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    835.0,
+                    1406.0,
+                    835.0,
+                    1406.0,
+                    876.0,
+                    291.0,
+                    876.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    864.0,
+                    453.0,
+                    864.0,
+                    453.0,
+                    908.0,
+                    293.0,
+                    908.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    283.0,
+                    631.0,
+                    283.0,
+                    631.0,
+                    327.0,
+                    291.0,
+                    327.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    689.0,
+                    283.0,
+                    1029.0,
+                    283.0,
+                    1029.0,
+                    327.0,
+                    689.0,
+                    327.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1063.0,
+                    283.0,
+                    1271.0,
+                    283.0,
+                    1271.0,
+                    327.0,
+                    1063.0,
+                    327.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1291.0,
+                    283.0,
+                    1300.0,
+                    283.0,
+                    1300.0,
+                    327.0,
+                    1291.0,
+                    327.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    321.0,
+                    298.0,
+                    321.0,
+                    298.0,
+                    365.0,
+                    295.0,
+                    365.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    617.0,
+                    321.0,
+                    1407.0,
+                    321.0,
+                    1407.0,
+                    365.0,
+                    617.0,
+                    365.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    365.0,
+                    371.0,
+                    365.0,
+                    371.0,
+                    402.0,
+                    295.0,
+                    402.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    391.0,
+                    365.0,
+                    567.0,
+                    365.0,
+                    567.0,
+                    402.0,
+                    391.0,
+                    402.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    700.0,
+                    365.0,
+                    968.0,
+                    365.0,
+                    968.0,
+                    402.0,
+                    700.0,
+                    402.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1083.0,
+                    1153.0,
+                    1083.0,
+                    1153.0,
+                    1120.0,
+                    292.0,
+                    1120.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1114.0,
+                    346.0,
+                    1114.0,
+                    346.0,
+                    1152.0,
+                    294.0,
+                    1152.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    612.0,
+                    1114.0,
+                    988.0,
+                    1114.0,
+                    988.0,
+                    1152.0,
+                    612.0,
+                    1152.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1091.0,
+                    1114.0,
+                    1406.0,
+                    1114.0,
+                    1406.0,
+                    1152.0,
+                    1091.0,
+                    1152.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1147.0,
+                    745.0,
+                    1147.0,
+                    745.0,
+                    1184.0,
+                    293.0,
+                    1184.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    816.0,
+                    1147.0,
+                    1350.0,
+                    1147.0,
+                    1350.0,
+                    1184.0,
+                    816.0,
+                    1184.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1385.0,
+                    1147.0,
+                    1397.0,
+                    1147.0,
+                    1397.0,
+                    1184.0,
+                    1385.0,
+                    1184.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1007.0,
+                    1405.0,
+                    1007.0,
+                    1405.0,
+                    1043.0,
+                    295.0,
+                    1043.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1039.0,
+                    1408.0,
+                    1039.0,
+                    1408.0,
+                    1076.0,
+                    295.0,
+                    1076.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1293.0,
+                    1075.0,
+                    1293.0,
+                    1075.0,
+                    1331.0,
+                    293.0,
+                    1331.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1108.0,
+                    1293.0,
+                    1351.0,
+                    1293.0,
+                    1351.0,
+                    1331.0,
+                    1108.0,
+                    1331.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    228.0,
+                    1106.0,
+                    228.0,
+                    1106.0,
+                    267.0,
+                    294.0,
+                    267.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 3,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    986,
+                    1405,
+                    986,
+                    1405,
+                    1688,
+                    297,
+                    1688
+                ],
+                "score": 0.985
+            },
+            {
+                "category_id": 5,
+                "poly": [
+                    297,
+                    281,
+                    1439,
+                    281,
+                    1439,
+                    550,
+                    297,
+                    550
+                ],
+                "score": 0.982,
+                "html": "<table><tr><td></td><td colspan=\"3\">CUB</td><td colspan=\"3\">FLOWERS</td></tr><tr><td></td><td>u</td><td>S</td><td>H</td><td>u</td><td>s</td><td>H</td></tr><tr><td>CADA-VAE (Schönfeld et al., 2019)</td><td>n/a</td><td>n/a</td><td>53.4</td><td>n/a</td><td>n/a</td><td>n/a</td></tr><tr><td>f-CLSWGAN (Xian et al., 2018d)</td><td>50.3</td><td>58.3</td><td>54.0</td><td>59.0</td><td>73.8</td><td>65.6</td></tr><tr><td>f-VAEGAN-D2 (Xian et al.,2019)</td><td>48.4</td><td>60.1</td><td>53.6</td><td>56.8</td><td>74.9</td><td>64.6</td></tr><tr><td>cycle-(U)WGAN (Felix et al., 2018)</td><td>47.9</td><td>59.3</td><td>53.0</td><td>61.6</td><td>69.2</td><td>65.2</td></tr><tr><td>COSMO+f-CLSWGAN (Atzmon &amp; Chechik, 2019)</td><td>n/a</td><td>n/a</td><td>n/a</td><td>59.6</td><td>81.4</td><td>68.8</td></tr><tr><td>CLAREL (Ours)</td><td>59.3</td><td>52.6</td><td>55.8</td><td>73.0</td><td>73.6</td><td>73.3</td></tr></table>"
+            },
+            {
+                "category_id": 5,
+                "poly": [
+                    483,
+                    644,
+                    1217,
+                    644,
+                    1217,
+                    853,
+                    483,
+                    853
+                ],
+                "score": 0.981,
+                "html": "<table><tr><td></td><td>CUB</td><td>FLOWERS</td></tr><tr><td>CADA-VAE (Schonfeld et al., 2019)</td><td>n/a</td><td>n/a</td></tr><tr><td>f-CLSWGAN (Xian et al., 2018d)</td><td>57.3</td><td>67.2</td></tr><tr><td>f-VAEGAN-D2 (Xian et al., 2019)</td><td>61.0</td><td>67.7</td></tr><tr><td>cycle-(U)WGAN (Felix et al., 2018)</td><td>58.6</td><td>70.3</td></tr><tr><td>CLAREL (Ours)</td><td>66.7</td><td>76.8</td></tr></table>"
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1881,
+                    1403,
+                    1881,
+                    1403,
+                    2034,
+                    298,
+                    2034
+                ],
+                "score": 0.975
+            },
+            {
+                "category_id": 6,
+                "poly": [
+                    576,
+                    586,
+                    1118,
+                    586,
+                    1118,
+                    620,
+                    576,
+                    620
+                ],
+                "score": 0.915
+            },
+            {
+                "category_id": 6,
+                "poly": [
+                    508,
+                    224,
+                    1188,
+                    224,
+                    1188,
+                    258,
+                    508,
+                    258
+                ],
+                "score": 0.914
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    300,
+                    912,
+                    587,
+                    912,
+                    587,
+                    948,
+                    300,
+                    948
+                ],
+                "score": 0.913
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    298,
+                    75,
+                    858,
+                    75,
+                    858,
+                    105,
+                    298,
+                    105
+                ],
+                "score": 0.913
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    299,
+                    1819,
+                    490,
+                    1819,
+                    490,
+                    1851,
+                    299,
+                    1851
+                ],
+                "score": 0.886
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    302,
+                    1745,
+                    713,
+                    1745,
+                    713,
+                    1779,
+                    302,
+                    1779
+                ],
+                "score": 0.882
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    841,
+                    2088,
+                    858,
+                    2088,
+                    858,
+                    2112,
+                    841,
+                    2112
+                ],
+                "score": 0.705
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    574.0,
+                    581.0,
+                    1123.0,
+                    581.0,
+                    1123.0,
+                    625.0,
+                    574.0,
+                    625.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    505.0,
+                    219.0,
+                    1191.0,
+                    219.0,
+                    1191.0,
+                    263.0,
+                    505.0,
+                    263.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    910.0,
+                    593.0,
+                    910.0,
+                    593.0,
+                    953.0,
+                    291.0,
+                    953.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    298.0,
+                    73.0,
+                    858.0,
+                    73.0,
+                    858.0,
+                    108.0,
+                    298.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1819.0,
+                    494.0,
+                    1819.0,
+                    494.0,
+                    1855.0,
+                    294.0,
+                    1855.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1741.0,
+                    717.0,
+                    1741.0,
+                    717.0,
+                    1784.0,
+                    293.0,
+                    1784.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    839.0,
+                    2085.0,
+                    861.0,
+                    2085.0,
+                    861.0,
+                    2120.0,
+                    839.0,
+                    2120.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    985.0,
+                    1406.0,
+                    985.0,
+                    1406.0,
+                    1025.0,
+                    292.0,
+                    1025.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1019.0,
+                    1406.0,
+                    1019.0,
+                    1406.0,
+                    1053.0,
+                    292.0,
+                    1053.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1049.0,
+                    1405.0,
+                    1049.0,
+                    1405.0,
+                    1084.0,
+                    294.0,
+                    1084.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1080.0,
+                    1405.0,
+                    1080.0,
+                    1405.0,
+                    1114.0,
+                    294.0,
+                    1114.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1109.0,
+                    1408.0,
+                    1109.0,
+                    1408.0,
+                    1148.0,
+                    294.0,
+                    1148.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1142.0,
+                    1407.0,
+                    1142.0,
+                    1407.0,
+                    1175.0,
+                    292.0,
+                    1175.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1169.0,
+                    1407.0,
+                    1169.0,
+                    1407.0,
+                    1209.0,
+                    291.0,
+                    1209.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1200.0,
+                    1408.0,
+                    1200.0,
+                    1408.0,
+                    1238.0,
+                    292.0,
+                    1238.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1228.0,
+                    1410.0,
+                    1228.0,
+                    1410.0,
+                    1269.0,
+                    292.0,
+                    1269.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1261.0,
+                    1407.0,
+                    1261.0,
+                    1407.0,
+                    1298.0,
+                    294.0,
+                    1298.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1292.0,
+                    1410.0,
+                    1292.0,
+                    1410.0,
+                    1329.0,
+                    292.0,
+                    1329.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1323.0,
+                    1405.0,
+                    1323.0,
+                    1405.0,
+                    1358.0,
+                    295.0,
+                    1358.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1353.0,
+                    1405.0,
+                    1353.0,
+                    1405.0,
+                    1390.0,
+                    292.0,
+                    1390.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1383.0,
+                    1407.0,
+                    1383.0,
+                    1407.0,
+                    1420.0,
+                    292.0,
+                    1420.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1416.0,
+                    1407.0,
+                    1416.0,
+                    1407.0,
+                    1451.0,
+                    295.0,
+                    1451.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1446.0,
+                    1406.0,
+                    1446.0,
+                    1406.0,
+                    1481.0,
+                    294.0,
+                    1481.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1476.0,
+                    1406.0,
+                    1476.0,
+                    1406.0,
+                    1511.0,
+                    294.0,
+                    1511.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1505.0,
+                    1406.0,
+                    1505.0,
+                    1406.0,
+                    1543.0,
+                    294.0,
+                    1543.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1536.0,
+                    1405.0,
+                    1536.0,
+                    1405.0,
+                    1573.0,
+                    291.0,
+                    1573.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1568.0,
+                    1406.0,
+                    1568.0,
+                    1406.0,
+                    1603.0,
+                    295.0,
+                    1603.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1595.0,
+                    1406.0,
+                    1595.0,
+                    1406.0,
+                    1637.0,
+                    292.0,
+                    1637.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1628.0,
+                    1408.0,
+                    1628.0,
+                    1408.0,
+                    1664.0,
+                    292.0,
+                    1664.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1659.0,
+                    1174.0,
+                    1659.0,
+                    1174.0,
+                    1692.0,
+                    295.0,
+                    1692.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1880.0,
+                    1407.0,
+                    1880.0,
+                    1407.0,
+                    1915.0,
+                    294.0,
+                    1915.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1909.0,
+                    1407.0,
+                    1909.0,
+                    1407.0,
+                    1951.0,
+                    292.0,
+                    1951.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1943.0,
+                    1405.0,
+                    1943.0,
+                    1405.0,
+                    1976.0,
+                    294.0,
+                    1976.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1971.0,
+                    1405.0,
+                    1971.0,
+                    1405.0,
+                    2008.0,
+                    294.0,
+                    2008.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    2000.0,
+                    1408.0,
+                    2000.0,
+                    1408.0,
+                    2039.0,
+                    292.0,
+                    2039.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 4,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    899,
+                    1404,
+                    899,
+                    1404,
+                    1416,
+                    298,
+                    1416
+                ],
+                "score": 0.984
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1546,
+                    1405,
+                    1546,
+                    1405,
+                    2033,
+                    298,
+                    2033
+                ],
+                "score": 0.983
+            },
+            {
+                "category_id": 3,
+                "poly": [
+                    376,
+                    225,
+                    1322,
+                    225,
+                    1322,
+                    508,
+                    376,
+                    508
+                ],
+                "score": 0.968
+            },
+            {
+                "category_id": 4,
+                "poly": [
+                    298,
+                    532,
+                    1402,
+                    532,
+                    1402,
+                    627,
+                    298,
+                    627
+                ],
+                "score": 0.955
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    300,
+                    707,
+                    1400,
+                    707,
+                    1400,
+                    770,
+                    300,
+                    770
+                ],
+                "score": 0.941
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    297,
+                    76,
+                    858,
+                    76,
+                    858,
+                    105,
+                    297,
+                    105
+                ],
+                "score": 0.9
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    299,
+                    1478,
+                    534,
+                    1478,
+                    534,
+                    1509,
+                    299,
+                    1509
+                ],
+                "score": 0.889
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    840,
+                    2089,
+                    858,
+                    2089,
+                    858,
+                    2112,
+                    840,
+                    2112
+                ],
+                "score": 0.773
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    303,
+                    832,
+                    851,
+                    832,
+                    851,
+                    863,
+                    303,
+                    863
+                ],
+                "score": 0.767
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    810,
+                    1577,
+                    909,
+                    1577,
+                    909,
+                    1606,
+                    810,
+                    1606
+                ],
+                "score": 0.89,
+                "latex": "\\lambda = 0 . 5"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    868,
+                    565,
+                    924,
+                    565,
+                    924,
+                    594,
+                    868,
+                    594
+                ],
+                "score": 0.87,
+                "latex": "9 5 \\%"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    922,
+                    1577,
+                    1022,
+                    1577,
+                    1022,
+                    1606,
+                    922,
+                    1606
+                ],
+                "score": 0.87,
+                "latex": "\\kappa = 0 . 5"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1195,
+                    1943,
+                    1256,
+                    1943,
+                    1256,
+                    1972,
+                    1195,
+                    1972
+                ],
+                "score": 0.86,
+                "latex": "0 . 8 \\%"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1126,
+                    1973,
+                    1194,
+                    1973,
+                    1194,
+                    2002,
+                    1126,
+                    2002
+                ],
+                "score": 0.84,
+                "latex": "0 . 5 \\%"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1106,
+                    1056,
+                    1128,
+                    1056,
+                    1128,
+                    1083,
+                    1106,
+                    1083
+                ],
+                "score": 0.82,
+                "latex": "\\gamma"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1176,
+                    1053,
+                    1198,
+                    1053,
+                    1198,
+                    1083,
+                    1176,
+                    1083
+                ],
+                "score": 0.8,
+                "latex": "\\beta"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1077,
+                    1581,
+                    1100,
+                    1581,
+                    1100,
+                    1605,
+                    1077,
+                    1605
+                ],
+                "score": 0.77,
+                "latex": "\\alpha"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1300,
+                    539,
+                    1321,
+                    539,
+                    1321,
+                    561,
+                    1300,
+                    561
+                ],
+                "score": 0.77,
+                "latex": "\\alpha"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    583,
+                    1639,
+                    603,
+                    1639,
+                    603,
+                    1665,
+                    583,
+                    1665
+                ],
+                "score": 0.77,
+                "latex": "\\lambda"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    656,
+                    1642,
+                    676,
+                    1642,
+                    676,
+                    1665,
+                    656,
+                    1665
+                ],
+                "score": 0.76,
+                "latex": "\\kappa"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1378,
+                    1643,
+                    1398,
+                    1643,
+                    1398,
+                    1665,
+                    1378,
+                    1665
+                ],
+                "score": 0.73,
+                "latex": "\\alpha"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1269,
+                    601,
+                    1290,
+                    601,
+                    1290,
+                    622,
+                    1269,
+                    622
+                ],
+                "score": 0.64,
+                "latex": "\\alpha"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    521,
+                    1265,
+                    569,
+                    1265,
+                    569,
+                    1294,
+                    521,
+                    1294
+                ],
+                "score": 0.45,
+                "latex": "5 0 \\mathrm { k }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    425,
+                    1234,
+                    485,
+                    1234,
+                    485,
+                    1263,
+                    425,
+                    1263
+                ],
+                "score": 0.41,
+                "latex": "1 5 0 \\mathrm { k }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    781,
+                    595,
+                    808,
+                    595,
+                    808,
+                    623,
+                    781,
+                    623
+                ],
+                "score": 0.27,
+                "latex": "\\mathbf { H }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1028,
+                    534,
+                    1055,
+                    534,
+                    1055,
+                    562,
+                    1028,
+                    562
+                ],
+                "score": 0.26,
+                "latex": "\\mathbf { H }"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    390.0,
+                    230.0,
+                    425.0,
+                    230.0,
+                    425.0,
+                    262.0,
+                    390.0,
+                    262.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    389.0,
+                    269.0,
+                    427.0,
+                    269.0,
+                    427.0,
+                    303.0,
+                    389.0,
+                    303.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    900.0,
+                    260.0,
+                    934.0,
+                    260.0,
+                    934.0,
+                    289.0,
+                    900.0,
+                    289.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    389.0,
+                    345.0,
+                    427.0,
+                    345.0,
+                    427.0,
+                    381.0,
+                    389.0,
+                    381.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    899.0,
+                    351.0,
+                    934.0,
+                    351.0,
+                    934.0,
+                    383.0,
+                    899.0,
+                    383.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    389.0,
+                    383.0,
+                    427.0,
+                    383.0,
+                    427.0,
+                    419.0,
+                    389.0,
+                    419.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    899.0,
+                    396.0,
+                    935.0,
+                    396.0,
+                    935.0,
+                    426.0,
+                    899.0,
+                    426.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    396.0,
+                    415.0,
+                    449.0,
+                    415.0,
+                    449.0,
+                    450.0,
+                    396.0,
+                    450.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    483.0,
+                    418.0,
+                    533.0,
+                    418.0,
+                    533.0,
+                    449.0,
+                    483.0,
+                    449.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    567.0,
+                    415.0,
+                    623.0,
+                    415.0,
+                    623.0,
+                    450.0,
+                    567.0,
+                    450.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    654.0,
+                    418.0,
+                    705.0,
+                    418.0,
+                    705.0,
+                    449.0,
+                    654.0,
+                    449.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    740.0,
+                    418.0,
+                    791.0,
+                    418.0,
+                    791.0,
+                    449.0,
+                    740.0,
+                    449.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    906.0,
+                    417.0,
+                    957.0,
+                    417.0,
+                    957.0,
+                    448.0,
+                    906.0,
+                    448.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    991.0,
+                    418.0,
+                    1042.0,
+                    418.0,
+                    1042.0,
+                    449.0,
+                    991.0,
+                    449.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1077.0,
+                    418.0,
+                    1129.0,
+                    418.0,
+                    1129.0,
+                    448.0,
+                    1077.0,
+                    448.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1163.0,
+                    418.0,
+                    1214.0,
+                    418.0,
+                    1214.0,
+                    449.0,
+                    1163.0,
+                    449.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1249.0,
+                    418.0,
+                    1300.0,
+                    418.0,
+                    1300.0,
+                    449.0,
+                    1249.0,
+                    449.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    602.0,
+                    439.0,
+                    629.0,
+                    439.0,
+                    629.0,
+                    467.0,
+                    602.0,
+                    467.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1112.0,
+                    440.0,
+                    1136.0,
+                    440.0,
+                    1136.0,
+                    463.0,
+                    1112.0,
+                    463.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    543.0,
+                    476.0,
+                    644.0,
+                    476.0,
+                    644.0,
+                    509.0,
+                    543.0,
+                    509.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1021.0,
+                    475.0,
+                    1185.0,
+                    475.0,
+                    1185.0,
+                    509.0,
+                    1021.0,
+                    509.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    662.25,
+                    290.0,
+                    720.25,
+                    290.0,
+                    720.25,
+                    305.0,
+                    662.25,
+                    305.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    367.0,
+                    307.0,
+                    430.0,
+                    307.0,
+                    430.0,
+                    343.0,
+                    367.0,
+                    343.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    730.25,
+                    353.0,
+                    794.25,
+                    353.0,
+                    794.25,
+                    367.0,
+                    730.25,
+                    367.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    872.0,
+                    304.5,
+                    941.0,
+                    304.5,
+                    941.0,
+                    340.5,
+                    872.0,
+                    340.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    953.0,
+                    328.5,
+                    1010.0,
+                    328.5,
+                    1010.0,
+                    343.5,
+                    953.0,
+                    343.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    532.0,
+                    1027.0,
+                    532.0,
+                    1027.0,
+                    570.0,
+                    294.0,
+                    570.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1056.0,
+                    532.0,
+                    1299.0,
+                    532.0,
+                    1299.0,
+                    570.0,
+                    1056.0,
+                    570.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1322.0,
+                    532.0,
+                    1404.0,
+                    532.0,
+                    1404.0,
+                    570.0,
+                    1322.0,
+                    570.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    565.0,
+                    867.0,
+                    565.0,
+                    867.0,
+                    599.0,
+                    294.0,
+                    599.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    925.0,
+                    565.0,
+                    1406.0,
+                    565.0,
+                    1406.0,
+                    599.0,
+                    925.0,
+                    599.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    593.0,
+                    780.0,
+                    593.0,
+                    780.0,
+                    629.0,
+                    293.0,
+                    629.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    809.0,
+                    593.0,
+                    1268.0,
+                    593.0,
+                    1268.0,
+                    629.0,
+                    809.0,
+                    629.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1291.0,
+                    593.0,
+                    1300.0,
+                    593.0,
+                    1300.0,
+                    629.0,
+                    1291.0,
+                    629.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    298.0,
+                    72.0,
+                    858.0,
+                    72.0,
+                    858.0,
+                    108.0,
+                    298.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1474.0,
+                    539.0,
+                    1474.0,
+                    539.0,
+                    1515.0,
+                    292.0,
+                    1515.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    840.0,
+                    2087.0,
+                    861.0,
+                    2087.0,
+                    861.0,
+                    2118.0,
+                    840.0,
+                    2118.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    830.0,
+                    857.0,
+                    830.0,
+                    857.0,
+                    867.0,
+                    296.0,
+                    867.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    897.0,
+                    1406.0,
+                    897.0,
+                    1406.0,
+                    934.0,
+                    293.0,
+                    934.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    929.0,
+                    1408.0,
+                    929.0,
+                    1408.0,
+                    965.0,
+                    294.0,
+                    965.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    960.0,
+                    1402.0,
+                    960.0,
+                    1402.0,
+                    992.0,
+                    294.0,
+                    992.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    989.0,
+                    1406.0,
+                    989.0,
+                    1406.0,
+                    1025.0,
+                    293.0,
+                    1025.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1021.0,
+                    1405.0,
+                    1021.0,
+                    1405.0,
+                    1053.0,
+                    294.0,
+                    1053.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1051.0,
+                    1105.0,
+                    1051.0,
+                    1105.0,
+                    1087.0,
+                    293.0,
+                    1087.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1129.0,
+                    1051.0,
+                    1175.0,
+                    1051.0,
+                    1175.0,
+                    1087.0,
+                    1129.0,
+                    1087.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1199.0,
+                    1051.0,
+                    1405.0,
+                    1051.0,
+                    1405.0,
+                    1087.0,
+                    1199.0,
+                    1087.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1080.0,
+                    1405.0,
+                    1080.0,
+                    1405.0,
+                    1118.0,
+                    293.0,
+                    1118.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1109.0,
+                    1405.0,
+                    1109.0,
+                    1405.0,
+                    1149.0,
+                    293.0,
+                    1149.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1138.0,
+                    1405.0,
+                    1138.0,
+                    1405.0,
+                    1181.0,
+                    292.0,
+                    1181.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1172.0,
+                    1406.0,
+                    1172.0,
+                    1406.0,
+                    1208.0,
+                    294.0,
+                    1208.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1205.0,
+                    1405.0,
+                    1205.0,
+                    1405.0,
+                    1237.0,
+                    294.0,
+                    1237.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1231.0,
+                    424.0,
+                    1231.0,
+                    424.0,
+                    1269.0,
+                    293.0,
+                    1269.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    486.0,
+                    1231.0,
+                    1404.0,
+                    1231.0,
+                    1404.0,
+                    1269.0,
+                    486.0,
+                    1269.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1264.0,
+                    520.0,
+                    1264.0,
+                    520.0,
+                    1300.0,
+                    293.0,
+                    1300.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    570.0,
+                    1264.0,
+                    1408.0,
+                    1264.0,
+                    1408.0,
+                    1300.0,
+                    570.0,
+                    1300.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1292.0,
+                    1405.0,
+                    1292.0,
+                    1405.0,
+                    1334.0,
+                    292.0,
+                    1334.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1326.0,
+                    1404.0,
+                    1326.0,
+                    1404.0,
+                    1362.0,
+                    293.0,
+                    1362.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1352.0,
+                    1406.0,
+                    1352.0,
+                    1406.0,
+                    1395.0,
+                    292.0,
+                    1395.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1384.0,
+                    473.0,
+                    1384.0,
+                    473.0,
+                    1421.0,
+                    293.0,
+                    1421.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1546.0,
+                    1405.0,
+                    1546.0,
+                    1405.0,
+                    1580.0,
+                    296.0,
+                    1580.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1577.0,
+                    809.0,
+                    1577.0,
+                    809.0,
+                    1610.0,
+                    292.0,
+                    1610.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    910.0,
+                    1577.0,
+                    921.0,
+                    1577.0,
+                    921.0,
+                    1610.0,
+                    910.0,
+                    1610.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1023.0,
+                    1577.0,
+                    1076.0,
+                    1577.0,
+                    1076.0,
+                    1610.0,
+                    1023.0,
+                    1610.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1101.0,
+                    1577.0,
+                    1405.0,
+                    1577.0,
+                    1405.0,
+                    1610.0,
+                    1101.0,
+                    1610.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1607.0,
+                    1405.0,
+                    1607.0,
+                    1405.0,
+                    1640.0,
+                    293.0,
+                    1640.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1638.0,
+                    582.0,
+                    1638.0,
+                    582.0,
+                    1672.0,
+                    293.0,
+                    1672.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    604.0,
+                    1638.0,
+                    655.0,
+                    1638.0,
+                    655.0,
+                    1672.0,
+                    604.0,
+                    1672.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    677.0,
+                    1638.0,
+                    1377.0,
+                    1638.0,
+                    1377.0,
+                    1672.0,
+                    677.0,
+                    1672.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1399.0,
+                    1638.0,
+                    1407.0,
+                    1638.0,
+                    1407.0,
+                    1672.0,
+                    1399.0,
+                    1672.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1665.0,
+                    1405.0,
+                    1665.0,
+                    1405.0,
+                    1704.0,
+                    293.0,
+                    1704.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1699.0,
+                    1405.0,
+                    1699.0,
+                    1405.0,
+                    1733.0,
+                    295.0,
+                    1733.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1727.0,
+                    1409.0,
+                    1727.0,
+                    1409.0,
+                    1765.0,
+                    293.0,
+                    1765.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1758.0,
+                    1407.0,
+                    1758.0,
+                    1407.0,
+                    1797.0,
+                    292.0,
+                    1797.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1788.0,
+                    1407.0,
+                    1788.0,
+                    1407.0,
+                    1825.0,
+                    293.0,
+                    1825.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1819.0,
+                    1407.0,
+                    1819.0,
+                    1407.0,
+                    1856.0,
+                    293.0,
+                    1856.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1849.0,
+                    1407.0,
+                    1849.0,
+                    1407.0,
+                    1887.0,
+                    293.0,
+                    1887.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1884.0,
+                    1403.0,
+                    1884.0,
+                    1403.0,
+                    1914.0,
+                    293.0,
+                    1914.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1908.0,
+                    1407.0,
+                    1908.0,
+                    1407.0,
+                    1949.0,
+                    292.0,
+                    1949.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1942.0,
+                    1194.0,
+                    1942.0,
+                    1194.0,
+                    1977.0,
+                    295.0,
+                    1977.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1257.0,
+                    1942.0,
+                    1405.0,
+                    1942.0,
+                    1405.0,
+                    1977.0,
+                    1257.0,
+                    1977.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1972.0,
+                    1125.0,
+                    1972.0,
+                    1125.0,
+                    2008.0,
+                    293.0,
+                    2008.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1195.0,
+                    1972.0,
+                    1407.0,
+                    1972.0,
+                    1407.0,
+                    2008.0,
+                    1195.0,
+                    2008.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    2002.0,
+                    677.0,
+                    2002.0,
+                    677.0,
+                    2035.0,
+                    295.0,
+                    2035.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    709.0,
+                    1403.0,
+                    709.0,
+                    1403.0,
+                    741.0,
+                    297.0,
+                    741.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    735.0,
+                    1404.0,
+                    735.0,
+                    1404.0,
+                    777.0,
+                    295.0,
+                    777.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 5,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1201,
+                    1405,
+                    1201,
+                    1405,
+                    1538,
+                    297,
+                    1538
+                ],
+                "score": 0.979
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1650,
+                    1405,
+                    1650,
+                    1405,
+                    1926,
+                    297,
+                    1926
+                ],
+                "score": 0.977
+            },
+            {
+                "category_id": 3,
+                "poly": [
+                    375,
+                    669,
+                    1324,
+                    669,
+                    1324,
+                    951,
+                    375,
+                    951
+                ],
+                "score": 0.97
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    1942,
+                    1402,
+                    1942,
+                    1402,
+                    2035,
+                    299,
+                    2035
+                ],
+                "score": 0.968
+            },
+            {
+                "category_id": 3,
+                "poly": [
+                    375,
+                    225,
+                    1324,
+                    225,
+                    1324,
+                    506,
+                    375,
+                    506
+                ],
+                "score": 0.967
+            },
+            {
+                "category_id": 4,
+                "poly": [
+                    297,
+                    530,
+                    1405,
+                    530,
+                    1405,
+                    626,
+                    297,
+                    626
+                ],
+                "score": 0.957
+            },
+            {
+                "category_id": 4,
+                "poly": [
+                    297,
+                    975,
+                    1406,
+                    975,
+                    1406,
+                    1069,
+                    297,
+                    1069
+                ],
+                "score": 0.953
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    300,
+                    1591,
+                    578,
+                    1591,
+                    578,
+                    1619,
+                    300,
+                    1619
+                ],
+                "score": 0.811
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    841,
+                    2087,
+                    858,
+                    2087,
+                    858,
+                    2111,
+                    841,
+                    2111
+                ],
+                "score": 0.748
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    297,
+                    76,
+                    858,
+                    76,
+                    858,
+                    104,
+                    297,
+                    104
+                ],
+                "score": 0.687
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    307,
+                    1142,
+                    916,
+                    1142,
+                    916,
+                    1171,
+                    307,
+                    1171
+                ],
+                "score": 0.671
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    307,
+                    1142,
+                    916,
+                    1142,
+                    916,
+                    1171,
+                    307,
+                    1171
+                ],
+                "score": 0.188
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    297,
+                    76,
+                    857,
+                    76,
+                    857,
+                    105,
+                    297,
+                    105
+                ],
+                "score": 0.169
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1027,
+                    1324,
+                    1266,
+                    1324,
+                    1266,
+                    1354,
+                    1027,
+                    1354
+                ],
+                "score": 0.91,
+                "latex": "( 4 7 0 0 + 1 1 7 5 = 5 8 7 5 )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    707,
+                    563,
+                    782,
+                    563,
+                    782,
+                    591,
+                    707,
+                    591
+                ],
+                "score": 0.9,
+                "latex": "\\lambda = 0"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    394,
+                    2002,
+                    544,
+                    2002,
+                    544,
+                    2036,
+                    394,
+                    2036
+                ],
+                "score": 0.9,
+                "latex": "\\kappa \\in [ 0 . 2 , 0 . 6 ]"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1024,
+                    1007,
+                    1098,
+                    1007,
+                    1098,
+                    1034,
+                    1024,
+                    1034
+                ],
+                "score": 0.9,
+                "latex": "\\kappa = 0"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    905,
+                    2004,
+                    975,
+                    2004,
+                    975,
+                    2030,
+                    905,
+                    2030
+                ],
+                "score": 0.89,
+                "latex": "\\kappa = 1"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    360,
+                    1774,
+                    431,
+                    1774,
+                    431,
+                    1802,
+                    360,
+                    1802
+                ],
+                "score": 0.88,
+                "latex": "\\lambda = 0"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1240,
+                    1681,
+                    1396,
+                    1681,
+                    1396,
+                    1714,
+                    1240,
+                    1714
+                ],
+                "score": 0.84,
+                "latex": "\\lambda \\in [ 0 . 2 , 0 . 9 ]"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1222,
+                    1359,
+                    1244,
+                    1359,
+                    1244,
+                    1382,
+                    1222,
+                    1382
+                ],
+                "score": 0.79,
+                "latex": "\\alpha"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    897,
+                    1974,
+                    915,
+                    1974,
+                    915,
+                    2000,
+                    897,
+                    2000
+                ],
+                "score": 0.78,
+                "latex": "\\lambda"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1304,
+                    1209,
+                    1325,
+                    1209,
+                    1325,
+                    1230,
+                    1304,
+                    1230
+                ],
+                "score": 0.77,
+                "latex": "\\alpha"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    711,
+                    2008,
+                    730,
+                    2008,
+                    730,
+                    2030,
+                    711,
+                    2030
+                ],
+                "score": 0.76,
+                "latex": "\\kappa"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1166,
+                    1451,
+                    1188,
+                    1451,
+                    1188,
+                    1474,
+                    1166,
+                    1474
+                ],
+                "score": 0.74,
+                "latex": "\\alpha"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    449,
+                    1451,
+                    470,
+                    1451,
+                    470,
+                    1474,
+                    449,
+                    1474
+                ],
+                "score": 0.74,
+                "latex": "\\alpha"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1129,
+                    533,
+                    1148,
+                    533,
+                    1148,
+                    560,
+                    1129,
+                    560
+                ],
+                "score": 0.73,
+                "latex": "\\lambda"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1381,
+                    1714,
+                    1402,
+                    1714,
+                    1402,
+                    1740,
+                    1381,
+                    1740
+                ],
+                "score": 0.72,
+                "latex": "\\lambda"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1330,
+                    982,
+                    1350,
+                    982,
+                    1350,
+                    1004,
+                    1330,
+                    1004
+                ],
+                "score": 0.72,
+                "latex": "\\kappa"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1204,
+                    976,
+                    1230,
+                    976,
+                    1230,
+                    1004,
+                    1204,
+                    1004
+                ],
+                "score": 0.4,
+                "latex": "\\mathbf { H }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    841,
+                    1682,
+                    940,
+                    1682,
+                    940,
+                    1713,
+                    841,
+                    1713
+                ],
+                "score": 0.33,
+                "latex": "( \\mathbf { H } , \\mathbf { s } , \\mathbf { u } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    777,
+                    1390,
+                    794,
+                    1390,
+                    794,
+                    1411,
+                    777,
+                    1411
+                ],
+                "score": 0.31,
+                "latex": "^ +"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1042,
+                    1203,
+                    1068,
+                    1203,
+                    1068,
+                    1231,
+                    1042,
+                    1231
+                ],
+                "score": 0.25,
+                "latex": "\\mathbf { H }"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    901.0,
+                    669.0,
+                    934.0,
+                    669.0,
+                    934.0,
+                    695.0,
+                    901.0,
+                    695.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    393.0,
+                    685.0,
+                    422.0,
+                    685.0,
+                    422.0,
+                    711.0,
+                    393.0,
+                    711.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    901.0,
+                    698.0,
+                    931.0,
+                    698.0,
+                    931.0,
+                    731.0,
+                    901.0,
+                    731.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    373.0,
+                    707.0,
+                    401.0,
+                    707.0,
+                    401.0,
+                    835.0,
+                    373.0,
+                    835.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    881.0,
+                    707.0,
+                    910.0,
+                    707.0,
+                    910.0,
+                    839.0,
+                    881.0,
+                    839.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    394.0,
+                    724.0,
+                    424.0,
+                    724.0,
+                    424.0,
+                    753.0,
+                    394.0,
+                    753.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    902.0,
+                    729.0,
+                    930.0,
+                    729.0,
+                    930.0,
+                    756.0,
+                    902.0,
+                    756.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    393.0,
+                    769.0,
+                    422.0,
+                    769.0,
+                    422.0,
+                    794.0,
+                    393.0,
+                    794.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    902.0,
+                    760.0,
+                    930.0,
+                    760.0,
+                    930.0,
+                    786.0,
+                    902.0,
+                    786.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    901.0,
+                    792.0,
+                    933.0,
+                    792.0,
+                    933.0,
+                    817.0,
+                    901.0,
+                    817.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    980.0,
+                    792.0,
+                    998.0,
+                    792.0,
+                    998.0,
+                    811.0,
+                    980.0,
+                    811.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    393.0,
+                    811.0,
+                    421.0,
+                    811.0,
+                    421.0,
+                    836.0,
+                    393.0,
+                    836.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    468.0,
+                    812.0,
+                    492.0,
+                    812.0,
+                    492.0,
+                    834.0,
+                    468.0,
+                    834.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    978.0,
+                    811.0,
+                    1000.0,
+                    811.0,
+                    1000.0,
+                    834.0,
+                    978.0,
+                    834.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    467.0,
+                    830.0,
+                    494.0,
+                    830.0,
+                    494.0,
+                    856.0,
+                    467.0,
+                    856.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    901.0,
+                    822.0,
+                    931.0,
+                    822.0,
+                    931.0,
+                    847.0,
+                    901.0,
+                    847.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    974.0,
+                    830.0,
+                    1004.0,
+                    830.0,
+                    1004.0,
+                    857.0,
+                    974.0,
+                    857.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    478.0,
+                    862.0,
+                    515.0,
+                    862.0,
+                    515.0,
+                    891.0,
+                    478.0,
+                    891.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    556.0,
+                    863.0,
+                    592.0,
+                    863.0,
+                    592.0,
+                    888.0,
+                    556.0,
+                    888.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    628.0,
+                    862.0,
+                    667.0,
+                    862.0,
+                    667.0,
+                    890.0,
+                    628.0,
+                    890.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    704.0,
+                    862.0,
+                    742.0,
+                    862.0,
+                    742.0,
+                    890.0,
+                    704.0,
+                    890.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    781.0,
+                    862.0,
+                    817.0,
+                    862.0,
+                    817.0,
+                    890.0,
+                    781.0,
+                    890.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    986.0,
+                    862.0,
+                    1025.0,
+                    862.0,
+                    1025.0,
+                    890.0,
+                    986.0,
+                    890.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1064.0,
+                    863.0,
+                    1100.0,
+                    863.0,
+                    1100.0,
+                    888.0,
+                    1064.0,
+                    888.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1137.0,
+                    862.0,
+                    1176.0,
+                    862.0,
+                    1176.0,
+                    889.0,
+                    1137.0,
+                    889.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1213.0,
+                    862.0,
+                    1252.0,
+                    862.0,
+                    1252.0,
+                    890.0,
+                    1213.0,
+                    890.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1289.0,
+                    862.0,
+                    1327.0,
+                    862.0,
+                    1327.0,
+                    890.0,
+                    1289.0,
+                    890.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    600.0,
+                    884.0,
+                    621.0,
+                    884.0,
+                    621.0,
+                    904.0,
+                    600.0,
+                    904.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1110.0,
+                    884.0,
+                    1128.0,
+                    884.0,
+                    1128.0,
+                    903.0,
+                    1110.0,
+                    903.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    544.0,
+                    919.0,
+                    643.0,
+                    919.0,
+                    643.0,
+                    951.0,
+                    544.0,
+                    951.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1022.0,
+                    919.0,
+                    1184.0,
+                    919.0,
+                    1184.0,
+                    950.0,
+                    1022.0,
+                    950.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    462.0,
+                    789.5,
+                    496.0,
+                    789.5,
+                    496.0,
+                    811.5,
+                    462.0,
+                    811.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    399.0,
+                    852.5,
+                    429.0,
+                    852.5,
+                    429.0,
+                    888.5,
+                    399.0,
+                    888.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    906.25,
+                    854.0,
+                    939.25,
+                    854.0,
+                    939.25,
+                    888.5,
+                    906.25,
+                    888.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    898.0,
+                    220.0,
+                    937.0,
+                    220.0,
+                    937.0,
+                    263.0,
+                    898.0,
+                    263.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    393.0,
+                    241.0,
+                    422.0,
+                    241.0,
+                    422.0,
+                    268.0,
+                    393.0,
+                    268.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    896.0,
+                    242.0,
+                    941.0,
+                    242.0,
+                    941.0,
+                    353.0,
+                    896.0,
+                    353.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    373.0,
+                    264.0,
+                    401.0,
+                    264.0,
+                    401.0,
+                    391.0,
+                    373.0,
+                    391.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    882.0,
+                    264.0,
+                    911.0,
+                    264.0,
+                    911.0,
+                    397.0,
+                    882.0,
+                    397.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    391.0,
+                    280.0,
+                    423.0,
+                    280.0,
+                    423.0,
+                    312.0,
+                    391.0,
+                    312.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    392.0,
+                    323.0,
+                    424.0,
+                    323.0,
+                    424.0,
+                    355.0,
+                    392.0,
+                    355.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    770.0,
+                    348.0,
+                    787.0,
+                    348.0,
+                    787.0,
+                    367.0,
+                    770.0,
+                    367.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    899.0,
+                    345.0,
+                    933.0,
+                    345.0,
+                    933.0,
+                    377.0,
+                    899.0,
+                    377.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1277.0,
+                    347.0,
+                    1295.0,
+                    347.0,
+                    1295.0,
+                    368.0,
+                    1277.0,
+                    368.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    393.0,
+                    367.0,
+                    422.0,
+                    367.0,
+                    422.0,
+                    393.0,
+                    393.0,
+                    393.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    769.0,
+                    370.0,
+                    790.0,
+                    370.0,
+                    790.0,
+                    413.0,
+                    769.0,
+                    413.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    900.0,
+                    376.0,
+                    931.0,
+                    376.0,
+                    931.0,
+                    408.0,
+                    900.0,
+                    408.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1275.0,
+                    367.0,
+                    1301.0,
+                    367.0,
+                    1301.0,
+                    415.0,
+                    1275.0,
+                    415.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    477.0,
+                    417.0,
+                    516.0,
+                    417.0,
+                    516.0,
+                    447.0,
+                    477.0,
+                    447.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    553.0,
+                    417.0,
+                    593.0,
+                    417.0,
+                    593.0,
+                    446.0,
+                    553.0,
+                    446.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    628.0,
+                    419.0,
+                    667.0,
+                    419.0,
+                    667.0,
+                    446.0,
+                    628.0,
+                    446.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    704.0,
+                    417.0,
+                    742.0,
+                    417.0,
+                    742.0,
+                    447.0,
+                    704.0,
+                    447.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    781.0,
+                    419.0,
+                    818.0,
+                    419.0,
+                    818.0,
+                    447.0,
+                    781.0,
+                    447.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    986.0,
+                    417.0,
+                    1025.0,
+                    417.0,
+                    1025.0,
+                    447.0,
+                    986.0,
+                    447.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1061.0,
+                    417.0,
+                    1101.0,
+                    417.0,
+                    1101.0,
+                    446.0,
+                    1061.0,
+                    446.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1137.0,
+                    419.0,
+                    1176.0,
+                    419.0,
+                    1176.0,
+                    446.0,
+                    1137.0,
+                    446.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1213.0,
+                    417.0,
+                    1252.0,
+                    417.0,
+                    1252.0,
+                    447.0,
+                    1213.0,
+                    447.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1289.0,
+                    419.0,
+                    1327.0,
+                    419.0,
+                    1327.0,
+                    447.0,
+                    1289.0,
+                    447.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    598.0,
+                    437.0,
+                    620.0,
+                    437.0,
+                    620.0,
+                    462.0,
+                    598.0,
+                    462.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1107.0,
+                    437.0,
+                    1129.0,
+                    437.0,
+                    1129.0,
+                    462.0,
+                    1107.0,
+                    462.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    544.0,
+                    474.0,
+                    643.0,
+                    474.0,
+                    643.0,
+                    507.0,
+                    544.0,
+                    507.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1022.0,
+                    473.0,
+                    1184.0,
+                    473.0,
+                    1184.0,
+                    507.0,
+                    1022.0,
+                    507.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    396.25,
+                    408.5,
+                    432.25,
+                    408.5,
+                    432.25,
+                    445.0,
+                    396.25,
+                    445.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    904.25,
+                    409.5,
+                    942.25,
+                    409.5,
+                    942.25,
+                    445.0,
+                    904.25,
+                    445.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    531.0,
+                    1128.0,
+                    531.0,
+                    1128.0,
+                    566.0,
+                    296.0,
+                    566.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1149.0,
+                    531.0,
+                    1406.0,
+                    531.0,
+                    1406.0,
+                    566.0,
+                    1149.0,
+                    566.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    560.0,
+                    706.0,
+                    560.0,
+                    706.0,
+                    600.0,
+                    292.0,
+                    600.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    783.0,
+                    560.0,
+                    1406.0,
+                    560.0,
+                    1406.0,
+                    600.0,
+                    783.0,
+                    600.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    591.0,
+                    1203.0,
+                    591.0,
+                    1203.0,
+                    628.0,
+                    294.0,
+                    628.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    976.0,
+                    1203.0,
+                    976.0,
+                    1203.0,
+                    1010.0,
+                    295.0,
+                    1010.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1231.0,
+                    976.0,
+                    1329.0,
+                    976.0,
+                    1329.0,
+                    1010.0,
+                    1231.0,
+                    1010.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1351.0,
+                    976.0,
+                    1403.0,
+                    976.0,
+                    1403.0,
+                    1010.0,
+                    1351.0,
+                    1010.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1006.0,
+                    1023.0,
+                    1006.0,
+                    1023.0,
+                    1040.0,
+                    294.0,
+                    1040.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1099.0,
+                    1006.0,
+                    1406.0,
+                    1006.0,
+                    1406.0,
+                    1040.0,
+                    1099.0,
+                    1040.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1035.0,
+                    1334.0,
+                    1035.0,
+                    1334.0,
+                    1072.0,
+                    294.0,
+                    1072.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1585.0,
+                    583.0,
+                    1585.0,
+                    583.0,
+                    1625.0,
+                    293.0,
+                    1625.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    839.0,
+                    2086.0,
+                    860.0,
+                    2086.0,
+                    860.0,
+                    2118.0,
+                    839.0,
+                    2118.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    73.0,
+                    858.0,
+                    73.0,
+                    858.0,
+                    108.0,
+                    297.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    299.0,
+                    1139.0,
+                    923.0,
+                    1139.0,
+                    923.0,
+                    1174.0,
+                    299.0,
+                    1174.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    298.0,
+                    72.0,
+                    858.0,
+                    72.0,
+                    858.0,
+                    108.0,
+                    298.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1202.0,
+                    1041.0,
+                    1202.0,
+                    1041.0,
+                    1237.0,
+                    295.0,
+                    1237.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1069.0,
+                    1202.0,
+                    1303.0,
+                    1202.0,
+                    1303.0,
+                    1237.0,
+                    1069.0,
+                    1237.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1326.0,
+                    1202.0,
+                    1405.0,
+                    1202.0,
+                    1405.0,
+                    1237.0,
+                    1326.0,
+                    1237.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1227.0,
+                    1405.0,
+                    1227.0,
+                    1405.0,
+                    1271.0,
+                    291.0,
+                    1271.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1263.0,
+                    1403.0,
+                    1263.0,
+                    1403.0,
+                    1297.0,
+                    295.0,
+                    1297.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1293.0,
+                    1403.0,
+                    1293.0,
+                    1403.0,
+                    1326.0,
+                    295.0,
+                    1326.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1322.0,
+                    1026.0,
+                    1322.0,
+                    1026.0,
+                    1357.0,
+                    295.0,
+                    1357.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1267.0,
+                    1322.0,
+                    1405.0,
+                    1322.0,
+                    1405.0,
+                    1357.0,
+                    1267.0,
+                    1357.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1353.0,
+                    1221.0,
+                    1353.0,
+                    1221.0,
+                    1388.0,
+                    294.0,
+                    1388.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1245.0,
+                    1353.0,
+                    1406.0,
+                    1353.0,
+                    1406.0,
+                    1388.0,
+                    1245.0,
+                    1388.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1384.0,
+                    776.0,
+                    1384.0,
+                    776.0,
+                    1418.0,
+                    292.0,
+                    1418.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    795.0,
+                    1384.0,
+                    1405.0,
+                    1384.0,
+                    1405.0,
+                    1418.0,
+                    795.0,
+                    1418.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1414.0,
+                    1406.0,
+                    1414.0,
+                    1406.0,
+                    1451.0,
+                    291.0,
+                    1451.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1446.0,
+                    448.0,
+                    1446.0,
+                    448.0,
+                    1478.0,
+                    295.0,
+                    1478.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    471.0,
+                    1446.0,
+                    1165.0,
+                    1446.0,
+                    1165.0,
+                    1478.0,
+                    471.0,
+                    1478.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1189.0,
+                    1446.0,
+                    1403.0,
+                    1446.0,
+                    1403.0,
+                    1478.0,
+                    1189.0,
+                    1478.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1476.0,
+                    1405.0,
+                    1476.0,
+                    1405.0,
+                    1510.0,
+                    295.0,
+                    1510.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1507.0,
+                    957.0,
+                    1507.0,
+                    957.0,
+                    1540.0,
+                    294.0,
+                    1540.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1652.0,
+                    1405.0,
+                    1652.0,
+                    1405.0,
+                    1685.0,
+                    296.0,
+                    1685.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1681.0,
+                    840.0,
+                    1681.0,
+                    840.0,
+                    1718.0,
+                    294.0,
+                    1718.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    941.0,
+                    1681.0,
+                    1239.0,
+                    1681.0,
+                    1239.0,
+                    1718.0,
+                    941.0,
+                    1718.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1397.0,
+                    1681.0,
+                    1407.0,
+                    1681.0,
+                    1407.0,
+                    1718.0,
+                    1397.0,
+                    1718.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1711.0,
+                    1380.0,
+                    1711.0,
+                    1380.0,
+                    1748.0,
+                    294.0,
+                    1748.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1403.0,
+                    1711.0,
+                    1406.0,
+                    1711.0,
+                    1406.0,
+                    1748.0,
+                    1403.0,
+                    1748.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1744.0,
+                    1403.0,
+                    1744.0,
+                    1403.0,
+                    1776.0,
+                    296.0,
+                    1776.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1775.0,
+                    359.0,
+                    1775.0,
+                    359.0,
+                    1808.0,
+                    296.0,
+                    1808.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    432.0,
+                    1775.0,
+                    1405.0,
+                    1775.0,
+                    1405.0,
+                    1808.0,
+                    432.0,
+                    1808.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1800.0,
+                    1406.0,
+                    1800.0,
+                    1406.0,
+                    1843.0,
+                    291.0,
+                    1843.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1831.0,
+                    1410.0,
+                    1831.0,
+                    1410.0,
+                    1873.0,
+                    291.0,
+                    1873.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1864.0,
+                    1407.0,
+                    1864.0,
+                    1407.0,
+                    1900.0,
+                    294.0,
+                    1900.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1894.0,
+                    944.0,
+                    1894.0,
+                    944.0,
+                    1929.0,
+                    294.0,
+                    1929.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1939.0,
+                    1406.0,
+                    1939.0,
+                    1406.0,
+                    1978.0,
+                    294.0,
+                    1978.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    1974.0,
+                    896.0,
+                    1974.0,
+                    896.0,
+                    2004.0,
+                    297.0,
+                    2004.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    916.0,
+                    1974.0,
+                    1403.0,
+                    1974.0,
+                    1403.0,
+                    2004.0,
+                    916.0,
+                    2004.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    2002.0,
+                    393.0,
+                    2002.0,
+                    393.0,
+                    2039.0,
+                    294.0,
+                    2039.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    545.0,
+                    2002.0,
+                    710.0,
+                    2002.0,
+                    710.0,
+                    2039.0,
+                    545.0,
+                    2039.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    731.0,
+                    2002.0,
+                    904.0,
+                    2002.0,
+                    904.0,
+                    2039.0,
+                    731.0,
+                    2039.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    976.0,
+                    2002.0,
+                    1406.0,
+                    2002.0,
+                    1406.0,
+                    2039.0,
+                    976.0,
+                    2039.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    299.0,
+                    1139.0,
+                    923.0,
+                    1139.0,
+                    923.0,
+                    1174.0,
+                    299.0,
+                    1174.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 6,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    683,
+                    1405,
+                    683,
+                    1405,
+                    1201,
+                    297,
+                    1201
+                ],
+                "score": 0.984
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1217,
+                    1404,
+                    1217,
+                    1404,
+                    1584,
+                    297,
+                    1584
+                ],
+                "score": 0.983
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1697,
+                    1405,
+                    1697,
+                    1405,
+                    2033,
+                    298,
+                    2033
+                ],
+                "score": 0.982
+            },
+            {
+                "category_id": 5,
+                "poly": [
+                    498,
+                    280,
+                    1202,
+                    280,
+                    1202,
+                    548,
+                    498,
+                    548
+                ],
+                "score": 0.981,
+                "html": "<table><tr><td></td><td></td><td></td><td></td><td>CUB</td><td></td><td colspan=\"3\">FLOWERS</td></tr><tr><td>α</td><td>入</td><td>K</td><td>u</td><td>S</td><td>H</td><td>u</td><td>s</td><td>H</td></tr><tr><td>0.0</td><td>0.5</td><td>0.5</td><td>38.3</td><td>65.3</td><td>48.3</td><td>55.1</td><td>84.6</td><td>66.7</td></tr><tr><td>0.0</td><td>0.5</td><td>0.0</td><td>39.3</td><td>57.5</td><td>46.7</td><td>54.0</td><td>78.1</td><td>63.8</td></tr><tr><td>√</td><td>0.5</td><td>0.0</td><td>53.8</td><td>49.6</td><td>51.6</td><td>71.7</td><td>67.2</td><td>69.4</td></tr><tr><td>√</td><td>0.0</td><td>0.5</td><td>47.4</td><td>36.6</td><td>41.3</td><td>51.5</td><td>60.5</td><td>55.6</td></tr><tr><td>√</td><td>1.0</td><td>0.5</td><td>53.9</td><td>53.8</td><td>53.8</td><td>69.5</td><td>73.9</td><td>71.6</td></tr><tr><td>厂</td><td>0.5</td><td>0.5</td><td>59.3</td><td>52.6</td><td>55.8</td><td>73.0</td><td>73.6</td><td>73.3</td></tr></table>"
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    605,
+                    1399,
+                    605,
+                    1399,
+                    668,
+                    299,
+                    668
+                ],
+                "score": 0.958
+            },
+            {
+                "category_id": 6,
+                "poly": [
+                    423,
+                    224,
+                    1272,
+                    224,
+                    1272,
+                    258,
+                    423,
+                    258
+                ],
+                "score": 0.915
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    297,
+                    75,
+                    858,
+                    75,
+                    858,
+                    105,
+                    297,
+                    105
+                ],
+                "score": 0.912
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    301,
+                    1628,
+                    560,
+                    1628,
+                    560,
+                    1663,
+                    301,
+                    1663
+                ],
+                "score": 0.889
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    840,
+                    2088,
+                    858,
+                    2088,
+                    858,
+                    2112,
+                    840,
+                    2112
+                ],
+                "score": 0.796
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    964,
+                    898,
+                    1035,
+                    898,
+                    1035,
+                    925,
+                    964,
+                    925
+                ],
+                "score": 0.89,
+                "latex": "\\kappa = 0"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    838,
+                    776,
+                    910,
+                    776,
+                    910,
+                    804,
+                    838,
+                    804
+                ],
+                "score": 0.89,
+                "latex": "\\alpha = 0"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    979,
+                    836,
+                    1035,
+                    836,
+                    1035,
+                    865,
+                    979,
+                    865
+                ],
+                "score": 0.88,
+                "latex": "30 \\%"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1168,
+                    958,
+                    1208,
+                    958,
+                    1208,
+                    987,
+                    1168,
+                    987
+                ],
+                "score": 0.88,
+                "latex": "8 \\%"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    382,
+                    836,
+                    438,
+                    836,
+                    438,
+                    865,
+                    382,
+                    865
+                ],
+                "score": 0.88,
+                "latex": "30 \\%"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    554,
+                    836,
+                    596,
+                    836,
+                    596,
+                    865,
+                    554,
+                    865
+                ],
+                "score": 0.87,
+                "latex": "6 \\%"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    297,
+                    744,
+                    391,
+                    744,
+                    391,
+                    774,
+                    297,
+                    774
+                ],
+                "score": 0.87,
+                "latex": "\\lambda = 0 . 5"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    402,
+                    745,
+                    497,
+                    745,
+                    497,
+                    774,
+                    402,
+                    774
+                ],
+                "score": 0.87,
+                "latex": "\\kappa = 0 . 5"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1153,
+                    836,
+                    1193,
+                    836,
+                    1193,
+                    864,
+                    1153,
+                    864
+                ],
+                "score": 0.86,
+                "latex": "1 \\%"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1057,
+                    1279,
+                    1137,
+                    1279,
+                    1137,
+                    1307,
+                    1057,
+                    1307
+                ],
+                "score": 0.84,
+                "latex": "\\lambda = 0"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1244,
+                    1280,
+                    1325,
+                    1280,
+                    1325,
+                    1307,
+                    1244,
+                    1307
+                ],
+                "score": 0.81,
+                "latex": "\\kappa = 0"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1151,
+                    1279,
+                    1231,
+                    1279,
+                    1231,
+                    1307,
+                    1151,
+                    1307
+                ],
+                "score": 0.78,
+                "latex": "\\lambda = 1"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    548,
+                    749,
+                    570,
+                    749,
+                    570,
+                    773,
+                    548,
+                    773
+                ],
+                "score": 0.78,
+                "latex": "\\alpha"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    421.0,
+                    217.0,
+                    1274.0,
+                    217.0,
+                    1274.0,
+                    264.0,
+                    421.0,
+                    264.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    73.0,
+                    858.0,
+                    73.0,
+                    858.0,
+                    108.0,
+                    297.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1621.0,
+                    566.0,
+                    1621.0,
+                    566.0,
+                    1672.0,
+                    291.0,
+                    1672.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    836.0,
+                    2085.0,
+                    859.0,
+                    2085.0,
+                    859.0,
+                    2116.0,
+                    836.0,
+                    2116.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    681.0,
+                    1407.0,
+                    681.0,
+                    1407.0,
+                    718.0,
+                    294.0,
+                    718.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    713.0,
+                    1405.0,
+                    713.0,
+                    1405.0,
+                    748.0,
+                    294.0,
+                    748.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    392.0,
+                    744.0,
+                    401.0,
+                    744.0,
+                    401.0,
+                    779.0,
+                    392.0,
+                    779.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    498.0,
+                    744.0,
+                    547.0,
+                    744.0,
+                    547.0,
+                    779.0,
+                    498.0,
+                    779.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    571.0,
+                    744.0,
+                    1403.0,
+                    744.0,
+                    1403.0,
+                    779.0,
+                    571.0,
+                    779.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    772.0,
+                    837.0,
+                    772.0,
+                    837.0,
+                    811.0,
+                    292.0,
+                    811.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    911.0,
+                    772.0,
+                    1405.0,
+                    772.0,
+                    1405.0,
+                    811.0,
+                    911.0,
+                    811.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    804.0,
+                    1403.0,
+                    804.0,
+                    1403.0,
+                    840.0,
+                    295.0,
+                    840.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    833.0,
+                    381.0,
+                    833.0,
+                    381.0,
+                    872.0,
+                    292.0,
+                    872.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    439.0,
+                    833.0,
+                    553.0,
+                    833.0,
+                    553.0,
+                    872.0,
+                    439.0,
+                    872.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    597.0,
+                    833.0,
+                    978.0,
+                    833.0,
+                    978.0,
+                    872.0,
+                    597.0,
+                    872.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1036.0,
+                    833.0,
+                    1152.0,
+                    833.0,
+                    1152.0,
+                    872.0,
+                    1036.0,
+                    872.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1194.0,
+                    833.0,
+                    1407.0,
+                    833.0,
+                    1407.0,
+                    872.0,
+                    1194.0,
+                    872.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    867.0,
+                    1405.0,
+                    867.0,
+                    1405.0,
+                    900.0,
+                    295.0,
+                    900.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    892.0,
+                    963.0,
+                    892.0,
+                    963.0,
+                    936.0,
+                    291.0,
+                    936.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1036.0,
+                    892.0,
+                    1410.0,
+                    892.0,
+                    1410.0,
+                    936.0,
+                    1036.0,
+                    936.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    927.0,
+                    1407.0,
+                    927.0,
+                    1407.0,
+                    963.0,
+                    294.0,
+                    963.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    957.0,
+                    1167.0,
+                    957.0,
+                    1167.0,
+                    993.0,
+                    294.0,
+                    993.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1209.0,
+                    957.0,
+                    1407.0,
+                    957.0,
+                    1407.0,
+                    993.0,
+                    1209.0,
+                    993.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    986.0,
+                    1406.0,
+                    986.0,
+                    1406.0,
+                    1024.0,
+                    294.0,
+                    1024.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1018.0,
+                    1406.0,
+                    1018.0,
+                    1406.0,
+                    1054.0,
+                    294.0,
+                    1054.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1048.0,
+                    1405.0,
+                    1048.0,
+                    1405.0,
+                    1084.0,
+                    294.0,
+                    1084.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1080.0,
+                    1403.0,
+                    1080.0,
+                    1403.0,
+                    1112.0,
+                    295.0,
+                    1112.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1111.0,
+                    1403.0,
+                    1111.0,
+                    1403.0,
+                    1143.0,
+                    296.0,
+                    1143.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1135.0,
+                    1406.0,
+                    1135.0,
+                    1406.0,
+                    1178.0,
+                    292.0,
+                    1178.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1170.0,
+                    1234.0,
+                    1170.0,
+                    1234.0,
+                    1206.0,
+                    295.0,
+                    1206.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1216.0,
+                    1404.0,
+                    1216.0,
+                    1404.0,
+                    1252.0,
+                    294.0,
+                    1252.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1248.0,
+                    1409.0,
+                    1248.0,
+                    1409.0,
+                    1283.0,
+                    295.0,
+                    1283.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1279.0,
+                    1056.0,
+                    1279.0,
+                    1056.0,
+                    1313.0,
+                    295.0,
+                    1313.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1138.0,
+                    1279.0,
+                    1150.0,
+                    1279.0,
+                    1150.0,
+                    1313.0,
+                    1138.0,
+                    1313.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1232.0,
+                    1279.0,
+                    1243.0,
+                    1279.0,
+                    1243.0,
+                    1313.0,
+                    1232.0,
+                    1313.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1326.0,
+                    1279.0,
+                    1405.0,
+                    1279.0,
+                    1405.0,
+                    1313.0,
+                    1326.0,
+                    1313.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1307.0,
+                    1406.0,
+                    1307.0,
+                    1406.0,
+                    1346.0,
+                    291.0,
+                    1346.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1340.0,
+                    1406.0,
+                    1340.0,
+                    1406.0,
+                    1374.0,
+                    294.0,
+                    1374.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1370.0,
+                    1405.0,
+                    1370.0,
+                    1405.0,
+                    1404.0,
+                    294.0,
+                    1404.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1402.0,
+                    1404.0,
+                    1402.0,
+                    1404.0,
+                    1436.0,
+                    296.0,
+                    1436.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1430.0,
+                    1405.0,
+                    1430.0,
+                    1405.0,
+                    1466.0,
+                    292.0,
+                    1466.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1461.0,
+                    1404.0,
+                    1461.0,
+                    1404.0,
+                    1495.0,
+                    294.0,
+                    1495.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1488.0,
+                    1406.0,
+                    1488.0,
+                    1406.0,
+                    1528.0,
+                    291.0,
+                    1528.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1521.0,
+                    1406.0,
+                    1521.0,
+                    1406.0,
+                    1559.0,
+                    292.0,
+                    1559.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1554.0,
+                    1351.0,
+                    1554.0,
+                    1351.0,
+                    1588.0,
+                    295.0,
+                    1588.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1699.0,
+                    1407.0,
+                    1699.0,
+                    1407.0,
+                    1734.0,
+                    296.0,
+                    1734.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1728.0,
+                    1403.0,
+                    1728.0,
+                    1403.0,
+                    1762.0,
+                    295.0,
+                    1762.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1757.0,
+                    1405.0,
+                    1757.0,
+                    1405.0,
+                    1797.0,
+                    292.0,
+                    1797.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1791.0,
+                    1403.0,
+                    1791.0,
+                    1403.0,
+                    1822.0,
+                    295.0,
+                    1822.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1819.0,
+                    1403.0,
+                    1819.0,
+                    1403.0,
+                    1854.0,
+                    295.0,
+                    1854.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1849.0,
+                    1405.0,
+                    1849.0,
+                    1405.0,
+                    1886.0,
+                    293.0,
+                    1886.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1880.0,
+                    1406.0,
+                    1880.0,
+                    1406.0,
+                    1920.0,
+                    292.0,
+                    1920.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1912.0,
+                    1405.0,
+                    1912.0,
+                    1405.0,
+                    1947.0,
+                    295.0,
+                    1947.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1944.0,
+                    1402.0,
+                    1944.0,
+                    1402.0,
+                    1975.0,
+                    296.0,
+                    1975.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1973.0,
+                    1406.0,
+                    1973.0,
+                    1406.0,
+                    2007.0,
+                    295.0,
+                    2007.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    2004.0,
+                    830.0,
+                    2004.0,
+                    830.0,
+                    2036.0,
+                    293.0,
+                    2036.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    603.0,
+                    1405.0,
+                    603.0,
+                    1405.0,
+                    643.0,
+                    295.0,
+                    643.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    637.0,
+                    868.0,
+                    637.0,
+                    868.0,
+                    672.0,
+                    294.0,
+                    672.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 7,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 2,
+                "poly": [
+                    298,
+                    75,
+                    858,
+                    75,
+                    858,
+                    105,
+                    298,
+                    105
+                ],
+                "score": 0.894
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    840,
+                    2088,
+                    859,
+                    2088,
+                    859,
+                    2111,
+                    840,
+                    2111
+                ],
+                "score": 0.773
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    299,
+                    227,
+                    489,
+                    227,
+                    489,
+                    261,
+                    299,
+                    261
+                ],
+                "score": 0.751
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    278,
+                    1400,
+                    278,
+                    1400,
+                    342,
+                    298,
+                    342
+                ],
+                "score": 0.554
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    851,
+                    1405,
+                    851,
+                    1405,
+                    946,
+                    298,
+                    946
+                ],
+                "score": 0.548
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    769,
+                    1401,
+                    769,
+                    1401,
+                    835,
+                    297,
+                    835
+                ],
+                "score": 0.531
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    300,
+                    1426,
+                    1404,
+                    1426,
+                    1404,
+                    1491,
+                    300,
+                    1491
+                ],
+                "score": 0.529
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    294,
+                    1292,
+                    1283,
+                    1292,
+                    1283,
+                    1327,
+                    294,
+                    1327
+                ],
+                "score": 0.524
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    605,
+                    1403,
+                    605,
+                    1403,
+                    672,
+                    297,
+                    672
+                ],
+                "score": 0.519
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    301,
+                    1508,
+                    1212,
+                    1508,
+                    1212,
+                    1544,
+                    301,
+                    1544
+                ],
+                "score": 0.515
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    687,
+                    1399,
+                    687,
+                    1399,
+                    753,
+                    299,
+                    753
+                ],
+                "score": 0.507
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    296,
+                    1560,
+                    1394,
+                    1560,
+                    1394,
+                    1625,
+                    296,
+                    1625
+                ],
+                "score": 0.504
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    296,
+                    359,
+                    1398,
+                    359,
+                    1398,
+                    424,
+                    296,
+                    424
+                ],
+                "score": 0.503
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    523,
+                    1400,
+                    523,
+                    1400,
+                    587,
+                    297,
+                    587
+                ],
+                "score": 0.503
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    441,
+                    1399,
+                    441,
+                    1399,
+                    506,
+                    297,
+                    506
+                ],
+                "score": 0.493
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1046,
+                    1403,
+                    1046,
+                    1403,
+                    1111,
+                    297,
+                    1111
+                ],
+                "score": 0.485
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    301,
+                    1345,
+                    1404,
+                    1345,
+                    1404,
+                    1409,
+                    301,
+                    1409
+                ],
+                "score": 0.483
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    964,
+                    1405,
+                    964,
+                    1405,
+                    1030,
+                    297,
+                    1030
+                ],
+                "score": 0.481
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    296,
+                    1210,
+                    1403,
+                    1210,
+                    1403,
+                    1276,
+                    296,
+                    1276
+                ],
+                "score": 0.479
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    295,
+                    1643,
+                    1402,
+                    1643,
+                    1402,
+                    1707,
+                    295,
+                    1707
+                ],
+                "score": 0.46
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1128,
+                    1400,
+                    1128,
+                    1400,
+                    1193,
+                    298,
+                    1193
+                ],
+                "score": 0.458
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    1971,
+                    1397,
+                    1971,
+                    1397,
+                    2035,
+                    299,
+                    2035
+                ],
+                "score": 0.444
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1724,
+                    1400,
+                    1724,
+                    1400,
+                    1789,
+                    298,
+                    1789
+                ],
+                "score": 0.439
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    296,
+                    1889,
+                    1399,
+                    1889,
+                    1399,
+                    1953,
+                    296,
+                    1953
+                ],
+                "score": 0.422
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1807,
+                    1400,
+                    1807,
+                    1400,
+                    1872,
+                    297,
+                    1872
+                ],
+                "score": 0.407
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    227,
+                    489,
+                    227,
+                    489,
+                    261,
+                    299,
+                    261
+                ],
+                "score": 0.105
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    298.0,
+                    73.0,
+                    858.0,
+                    73.0,
+                    858.0,
+                    108.0,
+                    298.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    839.0,
+                    2087.0,
+                    861.0,
+                    2087.0,
+                    861.0,
+                    2117.0,
+                    839.0,
+                    2117.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    228.0,
+                    491.0,
+                    228.0,
+                    491.0,
+                    264.0,
+                    297.0,
+                    264.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    275.0,
+                    1405.0,
+                    275.0,
+                    1405.0,
+                    316.0,
+                    293.0,
+                    316.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    324.0,
+                    307.0,
+                    1063.0,
+                    307.0,
+                    1063.0,
+                    343.0,
+                    324.0,
+                    343.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    851.0,
+                    1406.0,
+                    851.0,
+                    1406.0,
+                    889.0,
+                    295.0,
+                    889.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    321.0,
+                    883.0,
+                    1407.0,
+                    883.0,
+                    1407.0,
+                    918.0,
+                    321.0,
+                    918.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    320.0,
+                    911.0,
+                    401.0,
+                    911.0,
+                    401.0,
+                    947.0,
+                    320.0,
+                    947.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    766.0,
+                    1405.0,
+                    766.0,
+                    1405.0,
+                    809.0,
+                    292.0,
+                    809.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    802.0,
+                    861.0,
+                    802.0,
+                    861.0,
+                    836.0,
+                    322.0,
+                    836.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1426.0,
+                    1408.0,
+                    1426.0,
+                    1408.0,
+                    1462.0,
+                    295.0,
+                    1462.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    319.0,
+                    1453.0,
+                    398.0,
+                    1453.0,
+                    398.0,
+                    1492.0,
+                    319.0,
+                    1492.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1291.0,
+                    1285.0,
+                    1291.0,
+                    1285.0,
+                    1333.0,
+                    292.0,
+                    1333.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    606.0,
+                    1407.0,
+                    606.0,
+                    1407.0,
+                    642.0,
+                    296.0,
+                    642.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    321.0,
+                    637.0,
+                    1398.0,
+                    637.0,
+                    1398.0,
+                    674.0,
+                    321.0,
+                    674.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1506.0,
+                    1210.0,
+                    1506.0,
+                    1210.0,
+                    1547.0,
+                    295.0,
+                    1547.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    686.0,
+                    1403.0,
+                    686.0,
+                    1403.0,
+                    725.0,
+                    294.0,
+                    725.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    719.0,
+                    1143.0,
+                    719.0,
+                    1143.0,
+                    754.0,
+                    323.0,
+                    754.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1558.0,
+                    1400.0,
+                    1558.0,
+                    1400.0,
+                    1599.0,
+                    293.0,
+                    1599.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    1591.0,
+                    844.0,
+                    1591.0,
+                    844.0,
+                    1627.0,
+                    323.0,
+                    1627.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    353.0,
+                    1404.0,
+                    353.0,
+                    1404.0,
+                    402.0,
+                    291.0,
+                    402.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    324.0,
+                    391.0,
+                    649.0,
+                    391.0,
+                    649.0,
+                    424.0,
+                    324.0,
+                    424.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    519.0,
+                    1404.0,
+                    519.0,
+                    1404.0,
+                    564.0,
+                    295.0,
+                    564.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    555.0,
+                    508.0,
+                    555.0,
+                    508.0,
+                    587.0,
+                    323.0,
+                    587.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    439.0,
+                    1403.0,
+                    439.0,
+                    1403.0,
+                    479.0,
+                    295.0,
+                    479.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    472.0,
+                    454.0,
+                    472.0,
+                    454.0,
+                    506.0,
+                    323.0,
+                    506.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1045.0,
+                    1405.0,
+                    1045.0,
+                    1405.0,
+                    1082.0,
+                    295.0,
+                    1082.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    1077.0,
+                    989.0,
+                    1077.0,
+                    989.0,
+                    1113.0,
+                    322.0,
+                    1113.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1345.0,
+                    1404.0,
+                    1345.0,
+                    1404.0,
+                    1381.0,
+                    296.0,
+                    1381.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    321.0,
+                    1376.0,
+                    736.0,
+                    1376.0,
+                    736.0,
+                    1410.0,
+                    321.0,
+                    1410.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    963.0,
+                    1408.0,
+                    963.0,
+                    1408.0,
+                    1002.0,
+                    294.0,
+                    1002.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    996.0,
+                    1410.0,
+                    996.0,
+                    1410.0,
+                    1032.0,
+                    323.0,
+                    1032.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    298.0,
+                    1211.0,
+                    1408.0,
+                    1211.0,
+                    1408.0,
+                    1247.0,
+                    298.0,
+                    1247.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    321.0,
+                    1243.0,
+                    763.0,
+                    1243.0,
+                    763.0,
+                    1277.0,
+                    321.0,
+                    1277.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1642.0,
+                    1404.0,
+                    1642.0,
+                    1404.0,
+                    1679.0,
+                    294.0,
+                    1679.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    1674.0,
+                    1285.0,
+                    1674.0,
+                    1285.0,
+                    1710.0,
+                    323.0,
+                    1710.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1127.0,
+                    1405.0,
+                    1127.0,
+                    1405.0,
+                    1167.0,
+                    294.0,
+                    1167.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    319.0,
+                    1158.0,
+                    582.0,
+                    1158.0,
+                    582.0,
+                    1194.0,
+                    319.0,
+                    1194.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1971.0,
+                    1403.0,
+                    1971.0,
+                    1403.0,
+                    2007.0,
+                    295.0,
+                    2007.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    2002.0,
+                    1200.0,
+                    2002.0,
+                    1200.0,
+                    2035.0,
+                    323.0,
+                    2035.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1724.0,
+                    1404.0,
+                    1724.0,
+                    1404.0,
+                    1760.0,
+                    294.0,
+                    1760.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    1758.0,
+                    760.0,
+                    1758.0,
+                    760.0,
+                    1789.0,
+                    323.0,
+                    1789.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1889.0,
+                    1404.0,
+                    1889.0,
+                    1404.0,
+                    1925.0,
+                    296.0,
+                    1925.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    324.0,
+                    1918.0,
+                    870.0,
+                    1918.0,
+                    870.0,
+                    1954.0,
+                    324.0,
+                    1954.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    1807.0,
+                    1404.0,
+                    1807.0,
+                    1404.0,
+                    1843.0,
+                    297.0,
+                    1843.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    1838.0,
+                    947.0,
+                    1838.0,
+                    947.0,
+                    1873.0,
+                    322.0,
+                    1873.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    228.0,
+                    491.0,
+                    228.0,
+                    491.0,
+                    264.0,
+                    297.0,
+                    264.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 8,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    295,
+                    1813,
+                    1406,
+                    1813,
+                    1406,
+                    2030,
+                    295,
+                    2030
+                ],
+                "score": 0.979
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    371,
+                    1403,
+                    1321,
+                    1403,
+                    1321,
+                    1616,
+                    371,
+                    1616
+                ],
+                "score": 0.964
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    357,
+                    1684,
+                    1334,
+                    1684,
+                    1334,
+                    1797,
+                    357,
+                    1797
+                ],
+                "score": 0.959
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    294,
+                    1257,
+                    1405,
+                    1257,
+                    1405,
+                    1326,
+                    294,
+                    1326
+                ],
+                "score": 0.939
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    293,
+                    1618,
+                    1411,
+                    1618,
+                    1411,
+                    1680,
+                    293,
+                    1680
+                ],
+                "score": 0.938
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    298,
+                    1194,
+                    832,
+                    1194,
+                    832,
+                    1230,
+                    298,
+                    1230
+                ],
+                "score": 0.92
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    293,
+                    225,
+                    1411,
+                    225,
+                    1411,
+                    1151,
+                    293,
+                    1151
+                ],
+                "score": 0.913
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    298,
+                    75,
+                    858,
+                    75,
+                    858,
+                    105,
+                    298,
+                    105
+                ],
+                "score": 0.899
+            },
+            {
+                "category_id": 9,
+                "poly": [
+                    1366,
+                    1763,
+                    1400,
+                    1763,
+                    1400,
+                    1792,
+                    1366,
+                    1792
+                ],
+                "score": 0.876
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    836,
+                    2088,
+                    865,
+                    2088,
+                    865,
+                    2113,
+                    836,
+                    2113
+                ],
+                "score": 0.867
+            },
+            {
+                "category_id": 9,
+                "poly": [
+                    1365,
+                    1335,
+                    1401,
+                    1335,
+                    1401,
+                    1364,
+                    1365,
+                    1364
+                ],
+                "score": 0.862
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1370,
+                    1176,
+                    1370,
+                    1176,
+                    1401,
+                    298,
+                    1401
+                ],
+                "score": 0.851
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    498,
+                    1330,
+                    1182,
+                    1330,
+                    1182,
+                    1368,
+                    498,
+                    1368
+                ],
+                "score": 0.757
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    498,
+                    1330,
+                    1182,
+                    1330,
+                    1182,
+                    1368,
+                    498,
+                    1368
+                ],
+                "score": 0.168
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    961,
+                    1966,
+                    1232,
+                    1966,
+                    1232,
+                    2000,
+                    961,
+                    2000
+                ],
+                "score": 0.93,
+                "latex": "\\mathbb { P } \\{ \\widehat { y } _ { \\alpha } \\in \\mathcal { V } ^ { t r } | y _ { v } \\in \\mathcal { V } ^ { t s } \\}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    639,
+                    1905,
+                    837,
+                    1905,
+                    837,
+                    1939,
+                    639,
+                    1939
+                ],
+                "score": 0.93,
+                "latex": "[ \\delta _ { t s } / ( 1 + \\alpha ) , \\delta _ { t s } \\overline { { ) } }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    959,
+                    1647,
+                    1131,
+                    1647,
+                    1131,
+                    1681,
+                    959,
+                    1681
+                ],
+                "score": 0.92,
+                "latex": "{ \\mathbb { P } } \\{ A \\cap B \\} = 0"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    739,
+                    1371,
+                    843,
+                    1371,
+                    843,
+                    1401,
+                    739,
+                    1401
+                ],
+                "score": 0.91,
+                "latex": "\\delta _ { t r } < \\delta _ { t s }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1242,
+                    1938,
+                    1397,
+                    1938,
+                    1397,
+                    1968,
+                    1242,
+                    1968
+                ],
+                "score": 0.91,
+                "latex": "0 \\leq \\alpha _ { 1 } \\leq \\alpha _ { 2 }"
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    356,
+                    1682,
+                    1339,
+                    1682,
+                    1339,
+                    1801,
+                    356,
+                    1801
+                ],
+                "score": 0.91,
+                "latex": "\\begin{array} { r l r } {  { \\mathbb { P } \\{ \\widehat { y } _ { \\alpha } \\in \\mathcal { Y } ^ { t r } | y _ { v } \\in \\mathcal { Y } ^ { t s } \\} = \\mathbb { P } \\{ \\widehat { y } \\in \\mathcal { Y } ^ { t r } | y _ { v } \\in \\mathcal { Y } ^ { t s } \\} - \\mathbb { P } \\{ \\frac { \\delta _ { t s } } { ( 1 + \\alpha ) } \\leq \\delta _ { t r } < \\delta _ { t s } | y _ { v } \\in \\mathcal { Y } ^ { t s } \\} } } \\\\ & { } & { \\leq \\mathbb { P } \\{ \\widehat { y } \\in \\mathcal { Y } ^ { t r } | y _ { v } \\in \\mathcal { Y } ^ { t s } \\} . } \\end{array}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1107,
+                    1877,
+                    1143,
+                    1877,
+                    1143,
+                    1906,
+                    1107,
+                    1906
+                ],
+                "score": 0.9,
+                "latex": "\\delta _ { t r }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1017,
+                    1876,
+                    1052,
+                    1876,
+                    1052,
+                    1906,
+                    1017,
+                    1906
+                ],
+                "score": 0.89,
+                "latex": "\\delta _ { t s }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    298,
+                    1292,
+                    543,
+                    1292,
+                    543,
+                    1326,
+                    298,
+                    1326
+                ],
+                "score": 0.88,
+                "latex": "\\begin{array} { r } { \\operatorname* { m i n } _ { y \\in \\mathcal { y } ^ { t r } } d ( \\mathbf { z } _ { v } , \\mathbf { p } ( y ) ) } \\end{array}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    296,
+                    1646,
+                    774,
+                    1646,
+                    774,
+                    1682,
+                    296,
+                    1682
+                ],
+                "score": 0.88,
+                "latex": "\\mathbb { P } \\{ A \\cup B \\} = \\mathbb { P } \\{ A \\} + \\mathbb { P } \\{ B \\} - \\mathbb { P } \\{ A \\cap B \\}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1125,
+                    1845,
+                    1158,
+                    1845,
+                    1158,
+                    1876,
+                    1125,
+                    1876
+                ],
+                "score": 0.88,
+                "latex": "\\widehat { y } _ { \\alpha }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    298,
+                    1937,
+                    332,
+                    1937,
+                    332,
+                    1967,
+                    298,
+                    1967
+                ],
+                "score": 0.87,
+                "latex": "\\delta _ { t r }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    595,
+                    1295,
+                    906,
+                    1295,
+                    906,
+                    1327,
+                    595,
+                    1327
+                ],
+                "score": 0.87,
+                "latex": "\\begin{array} { r } { \\delta _ { t s } \\equiv \\operatorname* { m i n } _ { y \\in \\mathcal { V } ^ { t s } } d ( \\mathbf { z } _ { v } , \\mathbf { p } ( y ) ) } \\end{array}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    296,
+                    1966,
+                    900,
+                    1966,
+                    900,
+                    2000,
+                    296,
+                    2000
+                ],
+                "score": 0.87,
+                "latex": "\\mathbb { P } \\{ \\widehat { y } _ { \\alpha _ { 1 } } \\in \\mathcal { Y } ^ { t r } | y _ { v } \\in \\mathcal { Y } ^ { t s } \\} \\geq \\mathbb { P } \\{ \\widehat { y } _ { \\alpha _ { 2 } } \\in \\mathcal { Y } ^ { \\bar { t } r } | y _ { v } \\in \\mathcal { Y } ^ { t s } \\}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    471,
+                    1260,
+                    1123,
+                    1260,
+                    1123,
+                    1294,
+                    471,
+                    1294
+                ],
+                "score": 0.86,
+                "latex": "\\mathbb { P } \\{ \\widehat { y } ~ \\in ~ \\mathcal { Y } ^ { t r } | y _ { v } ~ \\in ~ \\mathcal { Y } ^ { t s } \\} ~ \\geq ~ \\mathbb { P } \\{ \\widehat { y } _ { \\alpha } ~ \\in ~ \\mathcal { Y } ^ { t r } | y _ { v } ~ \\in ~ \\mathcal { Y } ^ { t s } \\}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1324,
+                    1261,
+                    1403,
+                    1261,
+                    1403,
+                    1295,
+                    1324,
+                    1295
+                ],
+                "score": 0.85,
+                "latex": "\\delta _ { t r } \\equiv"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1372,
+                    1618,
+                    1397,
+                    1618,
+                    1397,
+                    1644,
+                    1372,
+                    1644
+                ],
+                "score": 0.83,
+                "latex": "B"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    598,
+                    2004,
+                    619,
+                    2004,
+                    619,
+                    2024,
+                    598,
+                    2024
+                ],
+                "score": 0.82,
+                "latex": "\\alpha"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1292,
+                    1619,
+                    1316,
+                    1619,
+                    1316,
+                    1644,
+                    1292,
+                    1644
+                ],
+                "score": 0.82,
+                "latex": "A"
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    510,
+                    1331,
+                    1182,
+                    1331,
+                    1182,
+                    1368,
+                    510,
+                    1368
+                ],
+                "score": 0.81,
+                "latex": "\\mathbb { P } \\{ \\widehat { y } _ { \\alpha } \\in \\mathcal { V } ^ { t r } | y _ { v } \\in \\mathcal { V } ^ { t s } \\} = \\mathbb { P } \\left\\{ ( 1 + \\alpha ) \\delta _ { t r } < \\delta _ { t s } | y _ { v } \\in \\mathcal { V } ^ { t s } \\right\\} \\mathrm { ~ . ~ }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    452,
+                    1876,
+                    472,
+                    1876,
+                    472,
+                    1908,
+                    452,
+                    1908
+                ],
+                "score": 0.81,
+                "latex": "\\widehat { y }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    795,
+                    1821,
+                    817,
+                    1821,
+                    817,
+                    1842,
+                    795,
+                    1842
+                ],
+                "score": 0.8,
+                "latex": "\\alpha"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    977,
+                    1911,
+                    998,
+                    1911,
+                    998,
+                    1933,
+                    977,
+                    1933
+                ],
+                "score": 0.78,
+                "latex": "\\alpha"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1013,
+                    2005,
+                    1033,
+                    2005,
+                    1033,
+                    2024,
+                    1013,
+                    2024
+                ],
+                "score": 0.78,
+                "latex": "\\alpha"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    953,
+                    1943,
+                    973,
+                    1943,
+                    973,
+                    1964,
+                    953,
+                    1964
+                ],
+                "score": 0.73,
+                "latex": "\\alpha"
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    365,
+                    1398,
+                    1314,
+                    1398,
+                    1314,
+                    1619,
+                    365,
+                    1619
+                ],
+                "score": 0.67,
+                "latex": "\\begin{array} { r l } & { \\mathbb { P } \\left\\{ \\delta _ { t r } < \\delta _ { t s } \\middle | y _ { v } \\in \\mathcal { Y } ^ { t s } \\right\\} = \\mathbb { P } \\left\\{ ( 1 + \\alpha ) \\delta _ { t r } < ( 1 + \\alpha ) \\delta _ { t s } | y _ { v } \\in \\mathcal { Y } ^ { t s } \\right\\} } \\\\ & { = \\mathbb { P } \\left\\{ ( 1 + \\alpha ) \\delta _ { t r } < \\delta _ { t s } \\cup \\delta _ { t s } \\leq ( 1 + \\alpha ) \\delta _ { t r } < ( 1 + \\alpha ) \\delta _ { t s } | y _ { v } \\in \\mathcal { Y } ^ { t s } \\right\\} } \\\\ & { = \\mathbb { P } \\left\\{ ( 1 + \\alpha ) \\delta _ { t r } < \\delta _ { t s } | y _ { v } \\in \\mathcal { Y } ^ { t s } \\right\\} + \\mathbb { P } \\left\\{ \\delta _ { t s } \\leq ( 1 + \\alpha ) \\delta _ { t r } < ( 1 + \\alpha ) \\delta _ { t s } | y _ { v } \\in \\mathcal { Y } ^ { t s } \\right\\} } \\\\ & { - \\mathbb { P } \\left\\{ ( 1 + \\alpha ) \\delta _ { t r } < \\delta _ { t s } \\cap \\delta _ { t s } \\leq ( 1 + \\alpha ) \\delta _ { t r } < ( 1 + \\alpha ) \\delta _ { t s } | y _ { v } \\in \\mathcal { Y } ^ { t s } \\right\\} } \\\\ & { = \\mathbb { P } \\left\\{ ( 1 + \\alpha ) \\delta _ { t r } < \\delta _ { t s } | y _ { v } \\in \\mathcal { Y } ^ { t s } \\right\\} + \\mathbb { P } \\left\\{ \\delta _ { t s } \\leq ( 1 + \\alpha ) \\delta _ { t r } < ( 1 + \\alpha ) \\delta _ { t s } | y _ { v } \\in \\mathcal { Y } ^ { t s } \\right\\} } \\end{array}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1255,
+                    1086,
+                    1288,
+                    1086,
+                    1288,
+                    1112,
+                    1255,
+                    1112
+                ],
+                "score": 0.4,
+                "latex": "^ { + + }"
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    386,
+                    1404,
+                    1133,
+                    1404,
+                    1133,
+                    1441,
+                    386,
+                    1441
+                ],
+                "score": 0.38,
+                "latex": "\\mathbb { \\rho } \\left\\{ \\delta _ { t r } < \\delta _ { t s } | y _ { v } \\in \\mathcal { V } ^ { t s } \\right\\} = \\mathbb { P } \\left\\{ ( 1 + \\alpha ) \\delta _ { t r } < ( 1 + \\alpha ) \\delta _ { t s } | y _ { v } \\in \\mathcal { V } ^ { t s } \\right\\}"
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    370,
+                    1476,
+                    1311,
+                    1476,
+                    1311,
+                    1568,
+                    370,
+                    1568
+                ],
+                "score": 0.27,
+                "latex": "\\begin{array} { r l } & { = \\mathbb { P } \\left\\{ ( 1 + \\alpha ) \\delta _ { t r } < \\delta _ { t s } | y _ { v } \\in \\mathcal { D } ^ { t s } \\right\\} + \\mathbb { P } \\left\\{ \\delta _ { t s } \\leq ( 1 + \\alpha ) \\delta _ { t r } < ( 1 + \\alpha ) \\delta _ { t s } | y _ { v } \\in \\mathcal { D } ^ { t s } \\right\\} } \\\\ & { - \\mathbb { P } \\left\\{ ( 1 + \\alpha ) \\delta _ { t r } < \\delta _ { t s } \\cap \\delta _ { t s } \\leq ( 1 + \\alpha ) \\delta _ { t r } < ( 1 + \\alpha ) \\delta _ { t s } | y _ { v } \\in \\mathcal { D } ^ { t s } \\right\\} } \\end{array}"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1189.0,
+                    833.0,
+                    1189.0,
+                    833.0,
+                    1235.0,
+                    294.0,
+                    1235.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    298.0,
+                    73.0,
+                    858.0,
+                    73.0,
+                    858.0,
+                    108.0,
+                    298.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    830.0,
+                    2085.0,
+                    869.0,
+                    2085.0,
+                    869.0,
+                    2123.0,
+                    830.0,
+                    2123.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1814.0,
+                    794.0,
+                    1814.0,
+                    794.0,
+                    1848.0,
+                    295.0,
+                    1848.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    818.0,
+                    1814.0,
+                    1404.0,
+                    1814.0,
+                    1404.0,
+                    1848.0,
+                    818.0,
+                    1848.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1845.0,
+                    1124.0,
+                    1845.0,
+                    1124.0,
+                    1879.0,
+                    294.0,
+                    1879.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1159.0,
+                    1845.0,
+                    1404.0,
+                    1845.0,
+                    1404.0,
+                    1879.0,
+                    1159.0,
+                    1879.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1872.0,
+                    451.0,
+                    1872.0,
+                    451.0,
+                    1913.0,
+                    292.0,
+                    1913.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    473.0,
+                    1872.0,
+                    1016.0,
+                    1872.0,
+                    1016.0,
+                    1913.0,
+                    473.0,
+                    1913.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1053.0,
+                    1872.0,
+                    1106.0,
+                    1872.0,
+                    1106.0,
+                    1913.0,
+                    1053.0,
+                    1913.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1144.0,
+                    1872.0,
+                    1407.0,
+                    1872.0,
+                    1407.0,
+                    1913.0,
+                    1144.0,
+                    1913.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1904.0,
+                    638.0,
+                    1904.0,
+                    638.0,
+                    1943.0,
+                    293.0,
+                    1943.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    838.0,
+                    1904.0,
+                    976.0,
+                    1904.0,
+                    976.0,
+                    1943.0,
+                    838.0,
+                    1943.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    999.0,
+                    1904.0,
+                    1406.0,
+                    1904.0,
+                    1406.0,
+                    1943.0,
+                    999.0,
+                    1943.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1931.0,
+                    297.0,
+                    1931.0,
+                    297.0,
+                    1975.0,
+                    293.0,
+                    1975.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    333.0,
+                    1931.0,
+                    952.0,
+                    1931.0,
+                    952.0,
+                    1975.0,
+                    333.0,
+                    1975.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    974.0,
+                    1931.0,
+                    1241.0,
+                    1931.0,
+                    1241.0,
+                    1975.0,
+                    974.0,
+                    1975.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1398.0,
+                    1931.0,
+                    1409.0,
+                    1931.0,
+                    1409.0,
+                    1975.0,
+                    1398.0,
+                    1975.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    290.0,
+                    1962.0,
+                    295.0,
+                    1962.0,
+                    295.0,
+                    2004.0,
+                    290.0,
+                    2004.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    901.0,
+                    1962.0,
+                    960.0,
+                    1962.0,
+                    960.0,
+                    2004.0,
+                    901.0,
+                    2004.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1233.0,
+                    1962.0,
+                    1408.0,
+                    1962.0,
+                    1408.0,
+                    2004.0,
+                    1233.0,
+                    2004.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1995.0,
+                    597.0,
+                    1995.0,
+                    597.0,
+                    2034.0,
+                    292.0,
+                    2034.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    620.0,
+                    1995.0,
+                    1012.0,
+                    1995.0,
+                    1012.0,
+                    2034.0,
+                    620.0,
+                    2034.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1034.0,
+                    1995.0,
+                    1047.0,
+                    1995.0,
+                    1047.0,
+                    2034.0,
+                    1034.0,
+                    2034.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1253.0,
+                    470.0,
+                    1253.0,
+                    470.0,
+                    1298.0,
+                    293.0,
+                    1298.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1124.0,
+                    1253.0,
+                    1323.0,
+                    1253.0,
+                    1323.0,
+                    1298.0,
+                    1124.0,
+                    1298.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1290.0,
+                    297.0,
+                    1290.0,
+                    297.0,
+                    1331.0,
+                    294.0,
+                    1331.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    544.0,
+                    1290.0,
+                    594.0,
+                    1290.0,
+                    594.0,
+                    1331.0,
+                    544.0,
+                    1331.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    907.0,
+                    1290.0,
+                    1344.0,
+                    1290.0,
+                    1344.0,
+                    1331.0,
+                    907.0,
+                    1331.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1612.0,
+                    1291.0,
+                    1612.0,
+                    1291.0,
+                    1653.0,
+                    293.0,
+                    1653.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1317.0,
+                    1612.0,
+                    1371.0,
+                    1612.0,
+                    1371.0,
+                    1653.0,
+                    1317.0,
+                    1653.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1398.0,
+                    1612.0,
+                    1409.0,
+                    1612.0,
+                    1409.0,
+                    1653.0,
+                    1398.0,
+                    1653.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    775.0,
+                    1645.0,
+                    958.0,
+                    1645.0,
+                    958.0,
+                    1684.0,
+                    775.0,
+                    1684.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1132.0,
+                    1645.0,
+                    1344.0,
+                    1645.0,
+                    1344.0,
+                    1684.0,
+                    1132.0,
+                    1684.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    230.0,
+                    1403.0,
+                    230.0,
+                    1403.0,
+                    266.0,
+                    295.0,
+                    266.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    262.0,
+                    738.0,
+                    262.0,
+                    738.0,
+                    294.0,
+                    323.0,
+                    294.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    306.0,
+                    1404.0,
+                    306.0,
+                    1404.0,
+                    346.0,
+                    295.0,
+                    346.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    324.0,
+                    340.0,
+                    804.0,
+                    340.0,
+                    804.0,
+                    372.0,
+                    324.0,
+                    372.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    384.0,
+                    1408.0,
+                    384.0,
+                    1408.0,
+                    423.0,
+                    294.0,
+                    423.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    325.0,
+                    419.0,
+                    1077.0,
+                    419.0,
+                    1077.0,
+                    451.0,
+                    325.0,
+                    451.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    462.0,
+                    1192.0,
+                    462.0,
+                    1192.0,
+                    502.0,
+                    295.0,
+                    502.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    508.0,
+                    1404.0,
+                    508.0,
+                    1404.0,
+                    547.0,
+                    291.0,
+                    547.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    540.0,
+                    1099.0,
+                    540.0,
+                    1099.0,
+                    575.0,
+                    322.0,
+                    575.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    588.0,
+                    1408.0,
+                    588.0,
+                    1408.0,
+                    624.0,
+                    296.0,
+                    624.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    619.0,
+                    1033.0,
+                    619.0,
+                    1033.0,
+                    651.0,
+                    323.0,
+                    651.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    665.0,
+                    1409.0,
+                    665.0,
+                    1409.0,
+                    702.0,
+                    295.0,
+                    702.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    694.0,
+                    1300.0,
+                    694.0,
+                    1300.0,
+                    733.0,
+                    322.0,
+                    733.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    741.0,
+                    1405.0,
+                    741.0,
+                    1405.0,
+                    779.0,
+                    295.0,
+                    779.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    320.0,
+                    773.0,
+                    1412.0,
+                    773.0,
+                    1412.0,
+                    810.0,
+                    320.0,
+                    810.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    325.0,
+                    805.0,
+                    389.0,
+                    805.0,
+                    389.0,
+                    841.0,
+                    325.0,
+                    841.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    850.0,
+                    1406.0,
+                    850.0,
+                    1406.0,
+                    889.0,
+                    295.0,
+                    889.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    882.0,
+                    1173.0,
+                    882.0,
+                    1173.0,
+                    918.0,
+                    322.0,
+                    918.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    927.0,
+                    1405.0,
+                    927.0,
+                    1405.0,
+                    965.0,
+                    295.0,
+                    965.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    960.0,
+                    736.0,
+                    960.0,
+                    736.0,
+                    992.0,
+                    322.0,
+                    992.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1003.0,
+                    1404.0,
+                    1003.0,
+                    1404.0,
+                    1043.0,
+                    295.0,
+                    1043.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    1037.0,
+                    854.0,
+                    1037.0,
+                    854.0,
+                    1071.0,
+                    322.0,
+                    1071.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1082.0,
+                    1254.0,
+                    1082.0,
+                    1254.0,
+                    1120.0,
+                    294.0,
+                    1120.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1289.0,
+                    1082.0,
+                    1404.0,
+                    1082.0,
+                    1404.0,
+                    1120.0,
+                    1289.0,
+                    1120.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    1114.0,
+                    1167.0,
+                    1114.0,
+                    1167.0,
+                    1150.0,
+                    322.0,
+                    1150.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1367.0,
+                    738.0,
+                    1367.0,
+                    738.0,
+                    1406.0,
+                    295.0,
+                    1406.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    844.0,
+                    1367.0,
+                    1174.0,
+                    1367.0,
+                    1174.0,
+                    1406.0,
+                    844.0,
+                    1406.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1183.0,
+                    1324.0,
+                    1190.0,
+                    1324.0,
+                    1190.0,
+                    1375.0,
+                    1183.0,
+                    1375.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 9,
+            "width": 1700,
+            "height": 2200
+        }
+    }
+]
\ No newline at end of file
diff --git a/parse/train/SkC_7v5gx/SkC_7v5gx.md b/parse/train/SkC_7v5gx/SkC_7v5gx.md
new file mode 100644
index 0000000000000000000000000000000000000000..be42a68a19fd9b8bb78d2af0847bb006ef0c729f
--- /dev/null
+++ b/parse/train/SkC_7v5gx/SkC_7v5gx.md
@@ -0,0 +1,305 @@
+# THE POWER OF SPARSITY IN CONVOLUTIONAL NEURAL NETWORKS
+
+Mark Sandler and Andrey Zhmoginov
+
+Soravit Changpinyo ∗ Department of Computer Science University of Southern California Los Angeles, CA 90020, USA schangpi@usc.edu
+
+Google Inc. 1600 Amphitheatre Parkway Mountain View, CA 94043, USA {sandler,azhmogin}@google.com
+
+# ABSTRACT
+
+Deep convolutional networks are well-known for their high computational and memory demands. Given limited resources, how does one design a network that balances its size, training time, and prediction accuracy? A surprisingly effective approach to trade accuracy for size and speed is to simply reduce the number of channels in each convolutional layer by a fixed fraction and retrain the network. In many cases this leads to significantly smaller networks with only minimal changes to accuracy. In this paper, we take a step further by empirically examining a strategy for deactivating connections between filters in convolutional layers in a way that allows us to harvest savings both in run-time and memory for many network architectures. More specifically, we generalize 2D convolution to use a channel-wise sparse connection structure and show that this leads to significantly better results than the baseline approach for large networks including VGG and Inception V3.
+
+# 1 INTRODUCTION
+
+Deep neural networks combined with large-scale labeled data have become a standard recipe for achieving state-of-the-art performance on supervised learning tasks in recent years. Despite of their success, the capability of deep neural networks to model highly nonlinear functions comes with high computational and memory demands both during the model training and inference. In particular, the number of parameters of neural network models is often designed to be huge to account for the scale, diversity, and complexity of data that they learn from. While advances in hardware have somewhat alleviated the issue, network size, speed, and power consumption are all limiting factors when it comes to production deployment on mobile and embedded devices. On the other hand, it is wellknown that there is significant redundancy among the weights of neural networks. For example, Denil et al. (2013) show that it is possible to learn less than $5 \%$ of the network parameters and predict the rest without losing predictive accuracy. This evidence suggests that neural networks are often over-parameterized.
+
+These motivate the research on neural network compression. However, several immediate questions arise: Are these parameters easy to identify? Could we just make the network $5 \%$ of its size and retrain? Or are more advanced methods required? There is an extensive literature in the last few years that explores the question of network compression using advanced techniques, including network prunning, loss-based compression, quantization, and matrix decomposition. We overview many of these directions in the next section. However, there is surprisingly little research on whether this over-parameterization can simply be re-captured by more efficient architectures that could be obtained from original architectures via simple transformations.
+
+Our approach is inspired by a very simple yet successful method called depth multiplier (Howard, 2017). In this method the depth (the number of channels) of each convolutional layer in a given network is simply reduced by a fixed fraction and the network is retrained. We generalize this approach by removing the constraint that every input filter (or channel) must be fully connected to every output filter. Instead, we use a sparse connection matrix, where each output convolution channel is connected only to a small random fraction of the input channels. Note that, for convolutional networks, this still allows for efficient computation since the one channel spatial convolution across the entire plane remains unchanged.
+
+We empirically demonstrate the effectiveness of our approach on four networks (MNIST, CIFAR Net, Inception-V3 and VGG-16) of different sizes. Our results suggest that our approach outperforms dense convolutions with depth multiplier at high compression rates.
+
+For Inception V3 (Szegedy et al., 2016), we show that we can train a network with only about 300K of convolutional parameters1 and about 100M multiply-adds that achieves above $52 \%$ accuracy after it is fully trained. The corresponding depth-multiplier network has only about $41 \%$ accuracy. Another network that we consider is VGG-16n, a slightly modified version of VGG-16 (Simonyan & Zisserman, 2015), with $7 \mathbf { x }$ fewer parameters and similar accuracy.2 We found VGG-16n to start training much faster than the original VGG-16 which was trained incrementally in the original literature. We explore the impact of sparsification and the number of parameters on the quality of the network by building the networks up to $3 0 \mathrm { x }$ smaller than VGG-16n $2 0 0 \mathrm { x }$ smaller than the original VGG-16).
+
+In terms of model flexibility, sparse connections allow for an incremental training approach, where connection structure between layers can be densified as training progresses. More importantly, the incremental training approach can potentially speed up the training significantly due to savings in the early stages of training.
+
+The rest of the paper is organized as follows. Section 2 summarizes relevant work. We describe our approach in Section 3 and then present some intuition in Section 4. Finally, we show our experimental results in Section 5.
+
+# 2 RELATED WORK
+
+# 2.1 COMPRESSION TECHNIQUES FOR NEURAL NETWORKS
+
+Our work is closely related to a compression technique based on network pruning. However, the important difference is that we do not try to select the connections which are redundant. Instead, we just fix a random connectivity pattern and let the network train around it. We also give a brief overview of other two popular techniques: quantization and decomposition, though these directions are not the main focus and could be complementary to our work.
+
+Network pruning Much initial work on neural network compression focuses on removing unimportant connections using weight decay. Hanson & Pratt (1989) introduce hyperbolic and exponential biases to the objective. Optimal Brain Damage (LeCun et al., 1989) and Optimal Brain Surgeon (Hassibi & Stork, 1993) prune the networks based on second-order derivatives of the objectives. Recent work by Han et al. (2015; 2016a) alternates between pruning near-zero weights, which are encouraged by $\ell { 1 }$ or $\ell 2$ regularization, and retraining the pruned networks.
+
+More complex regularizers have also been considered. Wen et al. (2016) and Li et al. (2016) put structured sparsity regularizers on the weights, while Murray & Chiang (2015) put them on the hidden units. Feng & Darrell (2015) explore a nonparametric prior based on the Indian buffet processes (Griffiths & Ghahramani, 2011) on layers. Hu et al. (2016) prune neurons based on the analysis of their outputs on a large dataset. Anwar et al. (2015b) consider special sparsity patterns: channel-wise (removing a feature map/channel from a layer), kernel-wise (removing all connections between two feature maps in consecutive layers), and intra-kernel-strided (removing connections between two features with particular stride and offset). They also propose to use particle filter to decide the importance of connections and paths during training.
+
+Another line of work explores fixed network architectures with some subsets of connections removed. For example, LeCun et al. (1998) remove connections between the first two convolutional feature maps in a completely uniform manner. This is similar to our approach but they only consider a pre-defined pattern in which the same number of input feature map are assigned to each output feature map (Random Connection Table in Torch’s SpatialConvolutionMap function). Further, they do not explore how sparse connections affect performance compared to dense networks. Along a similar vein, Cires¸an et al. (2011) remove random connections in their MNIST experiments. However, they do not try to preserve the spatial convolutional density and it might be a challenge to harvest the savings on existing hardware. Ioannou et al. (2016a) explore three types of hierarchical arrangements of filter groups for CNNs, which depend on different assumptions about co-dependency of filters within each layer. These arrangements include columnar topologies inspired by AlexNet (Krizhevsky et al., 2012), tree-like topologies previously used by Ioannou et al. (2016b), and root-like topologies. Finally, Howard (2017) proposes the depth multiplier method to scale down the number of filters in each convolutional layer by a factor. In this case, depth multiplier can be thought of channel-wise pruning mentioned in (Anwar et al., 2015b). However, depth multiplier modifies the network architectures before training and removes each layer’s feature maps in a uniform manner.
+
+With the exception of (Anwar et al., 2015b; Li et al., 2016; Ioannou et al., 2016a) and depth multiplier (Howard, 2017), the above previous work performs connection pruning that leads to irregular network architectures. Thus, those techniques require additional efforts to represent network connections and might or might not allow for direct computational savings.
+
+Quantization Reducing the degree of redundancy of model parameters can be done in the form of quantization of network parameters. Hwang & Sung (2014); Arora et al. (2014) and Courbariaux et al. (2015; 2016); Rastegari et al. (2016) propose to train CNNs with ternary weights and binary weights, respectively. Gong et al. (2014) use vector quantization for parameters in fully connected layers. Anwar et al. (2015a) quantize a network with the squared error minimization. Chen et al. (2015) randomly group network parameters using a hash function. We note that this technique could be complementary to network pruning. For example, Han et al. (2016a) combine connection pruning in (Han et al., 2015) with quantization and Huffman coding.
+
+Decomposition Another approach is based on low-rank decomposition of the parameters. Decomposition methods include truncated SVD (Denton et al., 2014), decomposition to rank-1 bases (Jaderberg et al., 2014), CP decomposition (PARAFAC or CANDECOMP) (Lebedev et al., 2015), Tensor-Train decomposition of Oseledets (2011) (Novikov et al., 2015), sparse dictionary learning of Mairal et al. (2009) and PCA (Liu et al., 2015), asymmetric (3D) decomposition using reconstruction loss of non-linear responses combined with a rank selection method based on PCA accumulated energy (Zhang et al., 2015b;a), and Tucker decomposition using the kernel tensor reconstruction loss combined with a rank selection method based on global analytic variational Bayesian matrix factorization (Kim et al., 2016).
+
+# 2.2 REGULARIZATION OF NEURAL NETWORKS
+
+Hinton et al. (2012); Srivastava et al. (2014) propose Dropout for regularizing fully connected layers within neural networks layers by randomly setting a subset of activations to zero during training. Wan et al. (2013) later propose DropConnect, a generalization of Dropout that instead randomly sets a subset of weights or connections to zero. Our approach could be thought as related to DropConnect, but (1) we remove connections before training; (2) we focus on connections between convolutional layers; and (3) we kill connections in a more regular manner by restricting connection patterns to be the same along spatial dimensions.
+
+Recently, Han et al. (2016b) and Jin et al. (2016) propose a form of regularization where dropped connections are unfrozen and the network is retrained. This idea is similar to our incremental training approach. However, (1) we do not start with a full network; (2) we do not unfreeze connections all at once; and (3) we preserve regularity of the convolution operation.
+
+# 2.3 NEURAL NETWORK ARCHITECTURES
+
+Network compression and architectures are closely related. The goal of compression is to remove redundancy in network parameters; therefore, the knowledge about traits that determine architecture’s success would be desirable. Other than the discovery that depth is an important factor (Ba & Caruana, 2014), little is known about such traits.
+
+Some previous work performs architecture search but without the main goal of doing compression (Murray & Chiang, 2015; De Brabandere et al., 2016). Recent work proposes shortcut/skip connections to convolutional networks. See, among others, highway networks (Srivastava et al., 2015), residual networks (He et al., 2016a;b), networks with stochastic depth (Huang et al., 2016b), and densely connected convolutional networks (Huang et al., 2016a).
+
+# 3 APPROACH
+
+A CNN architecture consist of (1) convolutional layers, (2) pooling layers, (3) fully connected layers, and (4) a topology that governs how these layers are organized. Given an architecture, our general goal is to transform it into another architecture with a smaller number of parameters. In this paper, we limit ourselves to transformation functions that keep the general topology of the input architecture intact. Moreover, the main focus will be on the convolutional layers and convolution operations, as they impose highest computational and memory burden for most if not all large networks.
+
+# 3.1 DEPTH MULTIPLIER
+
+We first give a description of the depth multiplier method used in Howard (2017). Given a hyperparameter $\alpha \in ( 0 , 1 ]$ , the depth multiplier approach scales down the number of filters in each convolutional layers by $\alpha$ . Note that depth here refers to the third dimension of the activation volume of a single layer, not the number of layers in the whole network.
+
+Let $n _ { l - 1 }$ and $n _ { l }$ be the number of input and output filters at layer $l$ , respectively. After the operation $n _ { l - 1 }$ and $n _ { l }$ become $\lceil \alpha n _ { l - 1 } \rceil$ and $\lceil \alpha n _ { l } \rceil$ and the number of parameters (and the number of multiplications) becomes $\approx \alpha ^ { 2 }$ of the original number.
+
+The result of this operation is a network that is both $1 / \alpha ^ { 2 }$ smaller and faster. Many large networks can be significantly reduced in size using this method with only a small loss of precision (Howard, 2017). It is our belief that this method establishes a strong baseline to which any other advanced techniques should compare themselves. To the best of our knowledge, we are not aware of such comparisons in the literature.
+
+# 3.2 SPARSE RANDOM
+
+Instead of looking at depth multiplier as deactivating channels in the convolutional layers, we can look at it from the perspective of deactivating connections. From this point of view, depth multiplier kills the connections between two convolutional layers such that (a) the connection patterns are still the same across spatial dimensions and (b) all “alive” input channels are fully connected to all “alive” output channels.
+
+We generalize this approach by relaxing (b) while maintaining (a). That is, for every output channel, we connect it to a small subset of input channels. In other words, dense connections between a small number of channels become sparse connections between larger number of channels. This can be summarized in Fig. 1. The advantage of this is that the actual convolution can still be computed efficiently because sparsity is introduced only at the outer loop of the convolution operation and we can still take the advantage of the continuous memory layout. For more details regarding implementations of the two approaches, please refer to the Appendix.
+
+More concretely, let $n _ { l - 1 }$ and $n _ { l }$ be the number of channels of layer $l - 1$ and layer $l$ , respectively. For a sparsity coefficient $\alpha$ , each output filter $j$ only connects to an $\alpha$ fraction of filters of the previous layer. Thus, instead of having a connectivity matrix $W _ { s i j }$ of dimension $k ^ { 2 } \times n _ { l - 1 } \times n _ { l }$ , we have a sparse matrix with non-zero entries at $W _ { s a _ { i j } j }$ , where $a _ { i j }$ is an index matrix of dimension $k ^ { 2 } \times \alpha n _ { l - 1 } \times n _ { l }$ and $k$ is the kernel size.
+
+# 3.2.1 INCREMENTAL TRAINING
+
+In contrast to depth multiplier, a sparse convolutional network defines a connection pattern on a much bigger network. Therefore, an interesting extension is to consider incremental training: we start with a network that only contains a small fraction of connections (in our experiments we use $1 \%$ and $0 . 1 \%$ ) and add connections over time. This is motivated by an intuition that the network can use learned channels in new contexts by introducing additional connections. The potential practical advantage of this approach is that since we start training with very small networks and grow them over time, this approach has a potential to speed up the whole training process significantly. We note that depth multiplier will not benefit from this approach as any newly activated connections would require learning new filters from scratch.
+
+![](images/66fc8a8512a12ff04f8f7dc6aab17648934eff7a1161c2536e3fa54b1fa83cde.jpg)  
+Figure 1: Connection tensors of depth multiplier (left) and sparse random (right) approaches for $n _ { l - 1 } = 5$ and $n _ { l } = 1 0$ . Yellow denotes active connections. For both approaches, the connection pattern is the same across spatial dimension and fixed before training. However, in the sparse random approach, each output channel is connected to a (possibly) different subset of input channels, and vice versa.
+
+# 4 ANALYSIS
+
+In this section, we approach a question of why sparse convolutions are frequently more efficient than the dense convolutions with the same number of parameters. Our main intuition is that the sparse convolutional networks promote diversity. It is much harder to learn equivalent set of channels as, at high sparsity, channels have distinct connection structure or even overlapping connections. This can be formalized with a simple observation that any dense network is in fact a part of an exponentially large equivalence class, which is guaranteed to produce the same output for every input.
+
+Lemma 1 Any dense convolutional neural network with no cross-channel nonlinearities, distinct weights and biases, and with l hidden layers of sizes $n _ { 1 }$ , $n _ { 2 }$ , . . . , $n _ { l }$ , has at least $\textstyle \prod _ { i = 1 } ^ { l } n _ { i } !$ distinct equivalent networks which produce the same output.
+
+Proof Let $I$ denote the input to the network, $C _ { i }$ be the convolutional operator, $\sigma _ { i }$ denote the nonlinearity operator applied to the $i$ -th convolution layer and $S$ be a final transformation (e.g. softmax classifier). We assume that $\sigma _ { i }$ is a function that operates on each of the channels independently. We note that this is the case for almost any modern network. The output of the network can then be written as:
+
+$$
+\mathcal { N } ( I ) \equiv S \circ \sigma _ { l } \circ C _ { l } \circ \sigma _ { l - 1 } \circ \cdots \circ \sigma _ { 1 } \circ C _ { 1 } ( I )
+$$
+
+where we use $\circ$ to denote function composition to avoid numerous parentheses. The convolution operator $C _ { i }$ operates on input with $n _ { i - 1 }$ channels and produces an output with $n _ { i }$ channels. Now, fix arbitrary set of permutation functions $\pi _ { i }$ , where $\pi _ { i }$ can permute depth of size $n _ { i }$ . Since $\pi _ { i }$ is a linear function, it follows that $C _ { i } ^ { \prime } = \pi _ { i } ^ { - 1 } C _ { i } \pi _ { i - 1 }$ is a valid convolutional operator, which can be obtained from $C _ { i }$ by permuting its bias according to $\pi _ { i }$ and its weight matrix along input and output dimensions according to $\pi _ { i - 1 }$ and $\pi _ { i }$ respectively. For a new network defined as:
+
+$$
+\mathcal { N } ^ { \prime } ( I ) = S ^ { \prime } \circ \sigma _ { l } \circ C _ { l } ^ { \prime } \circ \sigma _ { l - 1 } \circ \cdots \circ \sigma _ { 1 } \circ C _ { 1 } ^ { \prime } ( I ) ,
+$$
+
+where $\pi _ { 0 }$ is an identity operator and $S ^ { \prime } \equiv S \circ \pi _ { l }$ , we claim that $\mathcal { N } ^ { \prime } ( I ) \equiv \mathcal { N } ( I )$ . Indeed, since nonlinearities do not apply cross-depth we have $\pi _ { n } \sigma _ { n } \pi _ { n } ^ { - 1 } \equiv \sigma _ { n }$ and thus:
+
+$$
+\begin{array} { r l } & { \mathscr { N } ^ { \prime } ( I ) = S ^ { \prime } \circ \sigma _ { l } \circ C _ { l } ^ { \prime } \circ \sigma _ { l - 1 } \circ \cdots \circ \sigma _ { 1 } \circ C _ { 1 } ^ { \prime } ( I ) = } \\ & { \qquad = S \circ \pi _ { l } \circ \sigma _ { l } \circ \pi _ { l } ^ { - 1 } \circ C _ { l } \circ \pi _ { l - 1 } \circ \cdots \circ \pi _ { 1 } \circ \sigma _ { 1 } \circ \pi _ { 1 } ^ { - 1 } \circ C _ { 1 } ( I ) = \mathscr { N } ( I ) . } \end{array}
+$$
+
+Thus, any set of permutations on hidden units defines an equivalent network.
+
+It is obvious that sparse networks are much more immune to parameter permutation – indeed every channel at layer $l$ is likely to have a unique tree describing its connection matrix all the way down. Exploring this direction is an interesting open question.
+
+# 5 EXPERIMENTS
+
+In this section, we demonstrate the effectiveness of the sparse random approach by comparing it to the depth multiplier approach at different compression rates. Moreover, we examine several settings in the incremental training where connections gradually become active during the training process.
+
+# 5.1 SETUP
+
+Networks and Datasets Our experiments are conducted on 4 networks for 3 different datasets.   
+All our experiments use open-source TensorFlow networks Abadi et al. (2015).
+
+MNIST AND CIFAR-10 We use standard networks provided by TensorFlow. For MNIST, it has 3-layer convolutional layers and achieves $9 9 . 5 \%$ accuracy when fully trained. For CIFAR-10, it has 2 convolutional layers and achieves $87 \%$ accuracy.
+
+IMAGENET We use open source Inception-V3 (Szegedy et al., 2016) network and a slightly modified version of VGG-16 (Simonyan & Zisserman, 2015) called VGG-16n on ImageNet ILSVRC 2012 (Deng et al., 2009; Russakovsky et al., 2015).
+
+Random connections Connections are activated according to their likelihood from the uniform distribution. In addition, they are activated in such a way that there are no connections going in or coming out of dead filters (i.e., any connection must have a path to input image and a path to the final prediction.). All connections in fully connected layers are retained.
+
+Implementation details All code is implemented in TensorFlow (Abadi et al., 2015). Deactivating connections is done by applying masks to parameter tensors. The Inception-v3 and VGG-16n networks are trained on 8 Tesla K80 GPUs, each with batch size 256 (32 per gpu) and batch normalization was used for all networks.
+
+5.2 COMPARISON BETWEEN SPARSE RANDOM AND DEPTH MULTIPLIER
+
+# 5.2.1 MNIST AND CIFAR-10
+
+We first compare depth multiplier and sparse random for the two small networks on MNIST and CIFAR-10. We compare the accuracy of the two approaches when the numbers of connections are roughly the same, based on a hyperparameter $\alpha$ . For dense convolutions, we pick a multiplier $\alpha$ and each filter depth is scaled down by $\sqrt { \alpha }$ and then rounded up. In sparse convolutions, a fraction $\alpha$ of connections are randomly deactivated if those parameters connect at least two filters on each layer; otherwise, a fraction of $\sqrt { \alpha }$ is used instead if the parameters connect layers with only one filter left. The accuracy numbers are averaged over 5 rounds for MNIST and 2 rounds on CIFAR-10.
+
+We show in Fig. 2 and Fig. 3 that the sparse networks have comparable or higher accuracy for the same number of parameters, with comparable accuracy at higher density. We note however that these networks are so small that at high compression rates most of operations are concentrated at the first layer, which is negligible for large networks. Moreover, in MNIST example, the size of network changes most dramatically from 2000 to 2 million parameters, while affecting accuracy only by $1 \%$ . This observation suggests that there might be benefits of maintaining the number of filters to be high and/or breaking the symmetry of connections. We explore this in the next section.
+
+# 5.2.2 INCEPTION-V3 ON IMAGENET
+
+We consider different values of sparsity ranging from 0.003 to 1, and depth multiplier from 0.05 to 1. Our experiments show (see Table 1 and Fig. 4) significant advantage of sparse networks over equivalently sized dense networks. We note that due to time constraints the reported quantitative numbers are preliminary, as the networks have not finished converging. We expect the final numbers to match the reported number for Inception V3 Szegedy et al. (2016), and the smaller networks to have comparable improvement.
+
+![](images/65926ca6ffb9b876a5add68d622ccc4f1c7a0fd94934ef67cf453485cb422141.jpg)  
+Figure 2: Comparison of accuracy (averaged over 5 rounds) vs. Number of parameters/Number of multiply-adds between dense and sparse convolutions on MNIST dataset. Note that though sparse convolution result in better parameter trade-off curve, the multiply-add curve shows the opposite pattern.
+
+Table 1: Inception V3: Preliminary quantitative results after 100 Epochs. Note the smallest sparse network is actually a hybrid network - we used both depth multiplier (0.5) and sparsity (0.01). The number of parameters is the number of parameters excluding the softmax layer.
+
+Accuracy for sparse convolutions   
+
+<table><tr><td>Sparsity</td><td>MAdds Params</td><td>P@1</td></tr><tr><td>0.50/0.01 0.003</td><td>43.0M 90k 158k</td><td>40.3 46.1</td></tr><tr><td>0.01</td><td>82.0M 104M</td><td>52.3</td></tr><tr><td>0.03</td><td>287k 208M 724k</td><td></td></tr><tr><td>0.10</td><td></td><td>59.5</td></tr><tr><td>0.30</td><td>628M 2.3 M</td><td>67.2</td></tr><tr><td></td><td>1.80 B 6.6M</td><td>73</td></tr><tr><td>0.60</td><td>3.50B 13M 22 M</td><td>75</td></tr><tr><td>1.00</td><td>5.70 B</td><td>77</td></tr></table>
+
+Accuracy for Depth Multiplier   
+
+<table><tr><td>Multiplier</td><td>MAdds</td><td>Params P@1</td></tr><tr><td>0.05</td><td>55.0M 56k</td><td>24.6</td></tr><tr><td>0.10</td><td>75.0M</td><td>170k 38.6</td></tr><tr><td>0.20</td><td>183M 718k</td><td>54.2</td></tr><tr><td>0.30</td><td>439M 1.8M</td><td>64.0</td></tr><tr><td>0.50</td><td>1.40B 5.4M</td><td>72.3</td></tr><tr><td>0.80</td><td>3.40B 13M</td><td>75.6</td></tr></table>
+
+Original network: 5.70 B 22 M 77 (78.8)
+
+# 5.2.3 VGG-16 ON IMAGENET
+
+In our experiments with the VGG-16 network (Simonyan & Zisserman, 2015), we modify the model architecture (calling it VGG-16n) by removing the two fully-connected layers with depth 4096 and replacing them with a $2 \times 2$ maxpool layer followed by a $3 \times 3$ convolutional layer with the depth of 1024. This alone sped up our training significantly. The comparison between depth multiplier and sparse connection approaches is shown in Fig. 5. The modified VGG-16n network has about 7 times fewer parameters, but appears to have comparable precision.
+
+![](images/f4cf6c184c4020dc5874321b477ae62c97ddf25c9d68322157c1a605a04d32e7.jpg)  
+Figure 3: Comparison of accuracy (averaged over 2 rounds) vs. Number of parameters/Number of multiply-adds between dense and sparse convolutions on CIFAR-10 dataset.
+
+![](images/1cf4cc4464141afef87dfda693d69954f6e151da17bb74348a13d18643a16d98.jpg)  
+Figure 4: Inception V3: Comparison of Precision $@ 1$ vs. Number of parameters/Number of multiply-adds between dense and sparse convolutions on ImageNet/Inception-V3. The full network corresponds to the right-most point of the curve.
+
+![](images/a57e0b1ff2bb1e2c18547e1723279c8b784b21823938c56fcb8383c2d7879efa.jpg)  
+Figure 5: VGG 16: Preliminary Quantitative Results. Comparison of Precision $@ 1$ vs. Number of parameters/Number of multiply-adds between dense and sparse convolutions on ImageNet/VGG16n. The full network corresponds to the right-most point of the curve. Original VGG-16 as described in Simonyan & Zisserman (2015) (blue star) and the same model trained by us from scratch (red cross) are also shown.
+
+# 5.3 INCREMENTAL TRAINING
+
+![](images/f704c8a1c6906e9b1cd74e0b702c578d2acd95067c474ee8b78c09edf9df95a7.jpg)  
+Figure 6: Incremental Training Of Inception V3: We show Precision $@ 1$ during the training process, where the networks densify over time. The saturation points show where the networks actually reach their full density.
+
+Finally, we show that incremental training is a promising direction. We start with a very sparse model and increase its density over time, using the approach described in Sect. 3.2.1. We note that a naive approach where we simply add filters results in training process basically equivalent to as if it started from scratch in every step. On the other hand, when the network densifies over time, all channels already possess some discriminative power and that information is utilized.
+
+In our experiments, we initially start training Inception-V3 with only $1 \%$ or $0 . 1 \%$ of connections enabled. Then, we double the number of connections every $T$ steps. We use $T = 1 0 , 0 0 0$ , $T =$ 25, 000 and $T = 5 0 , 0 0 0$ . The results are presented in Fig. 6. We show that the networks trained with the incremental approach regardless of the doubling period can catch up with the full Inception-V3 network (in some cases with small gains). Moreover, they recover very quickly from adding more (untrained) connections. In fact, the recovery is so fast that it is shorter than our saving interval for all the networks except for the network with 10K doubling period (resulting in the sharp drop). We believe that incremental training is a promising direction to speeding up the training of large convolutional neural networks since early stages of the training require much less computation.
+
+# 6 CONCLUSION
+
+We have proposed a new compression technique that uses a sparse random connection structure between input-output filters in convolutional layers of CNNs. We fix this structure before training and use the same structure across spatial dimensions to harvest savings from modern hardware. We show that this approach is especially useful at very high compression rates for large networks. For example, this simple method when applied to Inception V3 (Fig. 4), achieves AlexNet-level accuracy (Krizhevsky et al., 2012) with fewer than 400K parameters and VGG-level one (Fig. 5) with roughly $3 . 5 \mathrm { M }$ parameters. The simplicity of our approach is instructive in that it establishes a strong baseline to compare against when developing more advanced techniques. On the other hand, the uncanny match in performance of dense and equivalently-sized sparse networks with sparsity $> 0 . 1$ suggests that there might be some fundamental property of network architectures that is controlled by the number of parameters, regardless of how they are organized. Exploring this further might yield additional insights on understanding neural networks.
+
+In addition, we show that our method leads to an interesting novel incremental training technique, where we take advantage of sparse (and smaller) models to build a dense network. One interesting open direction is to enable incremental training not to simply densify the network over time, but also increase the number of channels. This would allow us to grow the network without having to fix its original shape in place.
+
+# REFERENCES
+
+Mart´ın Abadi, Ashish Agarwal, Paul Barham, Eugene Brevdo, Zhifeng Chen, Craig Citro, Greg S. Corrado, Andy Davis, Jeffrey Dean, Matthieu Devin, Sanjay Ghemawat, Ian Goodfellow, Andrew Harp, Geoffrey Irving, Michael Isard, Yangqing Jia, Rafal Jozefowicz, Lukasz Kaiser, Manjunath Kudlur, Josh Levenberg, Dan Mane, Rajat Monga, Sherry Moore, Derek Murray, Chris Olah, Mike Schuster, Jonathon Shlens, Benoit ´ Steiner, Ilya Sutskever, Kunal Talwar, Paul Tucker, Vincent Vanhoucke, Vijay Vasudevan, Fernanda Viegas, ´ Oriol Vinyals, Pete Warden, Martin Wattenberg, Martin Wicke, Yuan Yu, and Xiaoqiang Zheng. TensorFlow: Large-scale machine learning on heterogeneous systems, 2015. URL http://tensorflow.org/. Software available from tensorflow.org.
+
+Sajid Anwar, Kyuyeon Hwang, and Wonyong Sung. Fixed point optimization of deep convolutional neural networks for object recognition. In ICASSP, 2015a.
+
+Sajid Anwar, Kyuyeon Hwang, and Wonyong Sung. Structured pruning of deep convolutional neural networks. arXiv preprint arXiv:1512.08571, 2015b.
+
+Sanjeev Arora, Aditya Bhaskara, Rong Ge, and Tengyu Ma. Provable bounds for learning some deep representations. In ICML, 2014.
+
+Jimmy Ba and Rich Caruana. Do deep nets really need to be deep? In NIPS, 2014.
+
+Wenlin Chen, James T. Wilson, Stephen Tyree, Kilian Q. Weinberger, and Yixin Chen. Compressing neural networks with the hashing trick. In ICML, 2015.
+
+Dan C. Cires¸an, Ueli Meier, Jonathan Masci, Luca M. Gambardella, and Jurgen Schmidhuber. High- ¨ performance neural networks for visual object classification. arXiv preprint arXiv:1102.0183, 2011.   
+Matthieu Courbariaux, Yoshua Bengio, and Jean-Pierre David. Binaryconnect: Training deep neural networks with binary weights during propagations. In NIPS, 2015.   
+Matthieu Courbariaux, Itay Hubara, Daniel Soudry, Ran El-Yaniv, and Yoshua Bengio. Binarized neural networks: Training deep neural networks with weights and activations constrained to $+ 1$ or -1. arXiv preprint arXiv:1602.02830, 2016.   
+Bert De Brabandere, Xu Jia, Tinne Tuytelaars, and Luc Van Gool. Dynamic filter networks. In NIPS, 2016.   
+Jia Deng, Wei Dong, Richard Socher, Li-Jia Li, Kai Li, and Li Fei-Fei. Imagenet: A large-scale hierarchical image database. In CVPR, 2009.   
+Misha Denil, Babak Shakibi, Laurent Dinh, Marc Aurelio Ranzato, and Nando de Freitas. Predicting parameters in deep learning. In NIPS, 2013.   
+Emily L. Denton, Wojciech Zaremba, Joan Bruna, Yann LeCun, and Rob Fergus. Exploiting linear structure within convolutional networks for efficient evaluation. In NIPS, 2014.   
+Jiashi Feng and Trevor Darrell. Learning the structure of deep convolutional networks. In ICCV, 2015.   
+Yunchao Gong, Liu Liu, Ming Yang, and Lubomir Bourdev. Compressing deep convolutional networks using vector quantization. arXiv preprint arXiv:1412.6115, 2014.   
+Thomas L. Griffiths and Zoubin Ghahramani. The indian buffet process: An introduction and review. Journal of Machine Learning Research, 12(Apr):1185–1224, 2011.   
+Song Han, Jeff Pool, John Tran, and William Dally. Learning both weights and connections for efficient neural network. In NIPS, 2015.   
+Song Han, Huizi Mao, and William J. Dally. Deep compression: Compressing deep neural network with pruning, trained quantization and huffman coding. In ICLR, 2016a.   
+Song Han, Jeff Pool, Sharan Narang, Huizi Mao, Shijian Tang, Erich Elsen, Bryan Catanzaro, John Tran, and William J. Dally. Dsd: Regularizing deep neural networks with dense-sparse-dense training flow. arXiv preprint arXiv:1607.04381, 2016b.   
+Stephen Jose Hanson and Lorien Y. Pratt. Comparing biases for minimal network construction with back- ´ propagation. In NIPS, 1989.   
+Babak Hassibi and David G. Stork. Second order derivatives for network pruning: Optimal brain surgeon. In NIPS, 1993.   
+Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image recognition. In CVPR, 2016a.   
+Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Identity mappings in deep residual networks. In ECCV, 2016b.   
+Geoffrey E. Hinton, Nitish Srivastava, Alex Krizhevsky, Ilya Sutskever, and Ruslan Salakhutdinov. Improving neural networks by preventing co-adaptation of feature detectors. arXiv preprint arXiv:1207.0580, 2012.   
+Andrew G. Howard. Mobilenets: Efficient convolutional neural networks for mobile vision applications. In forthcoming, 2017.   
+Hengyuan Hu, Rui Peng, Yu-Wing Tai, and Chi-Keung Tang. Network trimming: A data-driven neuron pruning approach towards efficient deep architectures. arXiv preprint arXiv:1607.03250, 2016.   
+Gao Huang, Zhuang Liu, and Kilian Q. Weinberger. Densely connected convolutional networks. arXiv preprint arXiv:1608.06993, 2016a.   
+Gao Huang, Yu Sun, Zhuang Liu, Daniel Sedra, and Kilian Weinberger. Deep networks with stochastic depth. arXiv preprint arXiv:1603.09382, 2016b.   
+Kyuyeon Hwang and Wonyong Sung. Fixed-point feedforward deep neural network design using weights $+ 1$ , 0, and -1. In IEEE Workshop on Signal Processing Systems (SiPS), 2014.
+
+Yani Ioannou, Duncan Robertson, Roberto Cipolla, and Antonio Criminisi. Deep roots: Improving cnn efficiency with hierarchical filter groups. arXiv preprint arXiv:1605.06489, 2016a.
+
+Yani Ioannou, Duncan Robertson, Jamie Shotton, Roberto Cipolla, and Antonio Criminisi. Training cnns with low-rank filters for efficient image classification. In ICLR, 2016b.
+
+Max Jaderberg, Andrea Vedaldi, and Andrew Zisserman. Speeding up convolutional neural networks with low rank expansions. In BMVC, 2014.
+
+Xiaojie Jin, Xiaotong Yuan, Jiashi Feng, and Shuicheng Yan. Training skinny deep neural networks with iterative hard thresholding methods. arXiv preprint arXiv:1607.05423, 2016.
+
+Yong-Deok Kim, Eunhyeok Park, Sungjoo Yoo, Taelim Choi, Lu Yang, and Dongjun Shin. Compression of deep convolutional neural networks for fast and low power mobile applications. In ICLR, 2016.
+
+Alex Krizhevsky, Ilya Sutskever, and Geoffrey E. Hinton. Imagenet classification with deep convolutional neural networks. In NIPS, 2012.
+
+Vadim Lebedev, Yaroslav Ganin, Maksim Rakhuba, Ivan Oseledets, and Victor Lempitsky. Speeding-up convolutional neural networks using fine-tuned cp-decomposition. In ICLR, 2015.
+
+Yann LeCun, John S. Denker, Sara A. Solla, Richard E. Howard, and Lawrence D. Jackel. Optimal brain damage. In NIPS, 1989.
+
+Yann LeCun, Leon Bottou, Yoshua Bengio, and Patrick Haffner. Gradient-based learning applied to document ´ recognition. Proceedings of the IEEE, 86(11):2278–2324, 1998.
+
+Hao Li, Asim Kadav, Igor Durdanovic, Hanan Samet, and Hans Peter Graf. Pruning filters for efficient convnets. arXiv preprint arXiv:1608.08710, 2016.
+
+Baoyuan Liu, Min Wang, Hassan Foroosh, Marshall Tappen, and Marianna Pensky. Sparse convolutional neural networks. In CVPR, 2015.
+
+Julien Mairal, Francis Bach, Jean Ponce, and Guillermo Sapiro. Online dictionary learning for sparse coding. In ICML, 2009.
+
+Kenton Murray and David Chiang. Auto-sizing neural networks: With applications to n-gram language models. In EMNLP, 2015.
+
+Alexander Novikov, Dmitrii Podoprikhin, Anton Osokin, and Dmitry P. Vetrov. Tensorizing neural networks. In NIPS, 2015.
+
+Ivan V. Oseledets. Tensor-train decomposition. SIAM Journal on Scientific Computing, 33(5):2295–2317, 2011.
+
+Mohammad Rastegari, Vicente Ordonez, Joseph Redmon, and Ali Farhadi. Xnor-net: Imagenet classification using binary convolutional neural networks. In ECCV, 2016.
+
+Olga Russakovsky, Jia Deng, Hao Su, Jonathan Krause, Sanjeev Satheesh, Sean Ma, Zhiheng Huang, Andrej Karpathy, Aditya Khosla, Michael Bernstein, et al. Imagenet large scale visual recognition challenge. International Journal of Computer Vision, 115(3):211–252, 2015.
+
+Karen Simonyan and Andrew Zisserman. Very deep convolutional networks for large-scale image recognition. In ICLR, 2015.
+
+Nitish Srivastava, Geoffrey E. Hinton, Alex Krizhevsky, Ilya Sutskever, and Ruslan Salakhutdinov. Dropout: a simple way to prevent neural networks from overfitting. Journal of Machine Learning Research, 15(1): 1929–1958, 2014.
+
+Rupesh K. Srivastava, Klaus Greff, and Jurgen Schmidhuber. Training very deep networks. In ¨ NIPS, 2015.
+
+Christian Szegedy, Vincent Vanhoucke, Sergey Ioffe, Jonathon Shlens, and Zbigniew Wojna. Rethinking the inception architecture for computer vision. In CVPR, 2016.
+
+Li Wan, Matthew Zeiler, Sixin Zhang, Yann LeCun, and Rob Fergus. Regularization of neural networks using dropconnect. In ICML, 2013.
+
+Wei Wen, Chunpeng Wu, Yandan Wang, Yiran Chen, and Hai Li. Learning structured sparsity in deep neural networks. In NIPS, 2016.
+
+Xiangyu Zhang, Jianhua Zou, Kaiming He, and Jian Sun. Accelerating very deep convolutional networks for classification and detection. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2015a.
+
+Xiangyu Zhang, Jianhua Zou, Xiang Ming, Kaiming He, and Jian Sun. Efficient and accurate approximations of nonlinear convolutional networks. In CVPR, 2015b.
+
+# A ADDITIONAL DETAILS ON DENSE VS. SPARSE CONVOLUTIONS
+
+We contrast naive implementations of dense and sparse convolutions (cf. Sect. 3) in Algorithm. 1 and Algorithm 2. We emphasize that we do not use sparse matrices and only introduce sparsity from channel to channel. Thus, walltime will be mostly in terms of Multiply-Adds; the basic operation (convolving the entire image plane in Line 8 of both algorithms) is unchanged.
+
+# Algorithm 1 Naive implementation of dense convolution
+
+1: Inputs:   
+2: - input: Data tensor   
+3: - W : Parameter tensor   
+4: - input channels: Array of input channel IDs   
+5: - output channels: Array of output channel IDs   
+6: for $i$ in input channels do   
+7: for $o$ in output channels do   
+8: $o u t p u t [ o ] \gets o u t p u t [ o ] + \mathrm { c o n v o l v e } ( i n p u t [ i ] , W [ i , o , . . . ] )$   
+9: end for   
+10: end for   
+11: return output
+
+Algorithm 2 Naive implementation of sparse convolution
+
+1: Inputs:   
+2: - input: Data tensor   
+3: - W : Parameter tensor   
+4: - input channels: Array of input channel IDs   
+5: - output channels connected to i: Array of array of output channel IDs specifying connec  
+tions to each input channel   
+6: for $i$ in input channels do   
+7: for index, $o$ in enumerate(output channels connected to i[i]) do   
+8: output[o] $\gets$ output[o] $^ +$ convolve(input[i], W [i, index, . . .])   
+9: end for   
+10: end for   
+11: return output
\ No newline at end of file
diff --git a/parse/train/SkC_7v5gx/SkC_7v5gx_content_list.json b/parse/train/SkC_7v5gx/SkC_7v5gx_content_list.json
new file mode 100644
index 0000000000000000000000000000000000000000..49266790d0bd1bdfa23912b6fc0220a544c9d8b8
--- /dev/null
+++ b/parse/train/SkC_7v5gx/SkC_7v5gx_content_list.json
@@ -0,0 +1,1450 @@
+[
+    {
+        "type": "text",
+        "text": "THE POWER OF SPARSITY IN CONVOLUTIONAL NEURAL NETWORKS ",
+        "text_level": 1,
+        "bbox": [
+            174,
+            99,
+            647,
+            146
+        ],
+        "page_idx": 0
+    },
+    {
+        "type": "text",
+        "text": "Mark Sandler and Andrey Zhmoginov ",
+        "bbox": [
+            478,
+            171,
+            743,
+            184
+        ],
+        "page_idx": 0
+    },
+    {
+        "type": "text",
+        "text": "Soravit Changpinyo ∗ Department of Computer Science University of Southern California Los Angeles, CA 90020, USA schangpi@usc.edu ",
+        "bbox": [
+            183,
+            170,
+            405,
+            239
+        ],
+        "page_idx": 0
+    },
+    {
+        "type": "text",
+        "text": "Google Inc. 1600 Amphitheatre Parkway Mountain View, CA 94043, USA {sandler,azhmogin}@google.com ",
+        "bbox": [
+            478,
+            186,
+            759,
+            239
+        ],
+        "page_idx": 0
+    },
+    {
+        "type": "text",
+        "text": "ABSTRACT ",
+        "text_level": 1,
+        "bbox": [
+            454,
+            276,
+            544,
+            291
+        ],
+        "page_idx": 0
+    },
+    {
+        "type": "text",
+        "text": "Deep convolutional networks are well-known for their high computational and memory demands. Given limited resources, how does one design a network that balances its size, training time, and prediction accuracy? A surprisingly effective approach to trade accuracy for size and speed is to simply reduce the number of channels in each convolutional layer by a fixed fraction and retrain the network. In many cases this leads to significantly smaller networks with only minimal changes to accuracy. In this paper, we take a step further by empirically examining a strategy for deactivating connections between filters in convolutional layers in a way that allows us to harvest savings both in run-time and memory for many network architectures. More specifically, we generalize 2D convolution to use a channel-wise sparse connection structure and show that this leads to significantly better results than the baseline approach for large networks including VGG and Inception V3. ",
+        "bbox": [
+            233,
+            308,
+            764,
+            488
+        ],
+        "page_idx": 0
+    },
+    {
+        "type": "text",
+        "text": "1 INTRODUCTION ",
+        "text_level": 1,
+        "bbox": [
+            176,
+            510,
+            336,
+            525
+        ],
+        "page_idx": 0
+    },
+    {
+        "type": "text",
+        "text": "Deep neural networks combined with large-scale labeled data have become a standard recipe for achieving state-of-the-art performance on supervised learning tasks in recent years. Despite of their success, the capability of deep neural networks to model highly nonlinear functions comes with high computational and memory demands both during the model training and inference. In particular, the number of parameters of neural network models is often designed to be huge to account for the scale, diversity, and complexity of data that they learn from. While advances in hardware have somewhat alleviated the issue, network size, speed, and power consumption are all limiting factors when it comes to production deployment on mobile and embedded devices. On the other hand, it is wellknown that there is significant redundancy among the weights of neural networks. For example, Denil et al. (2013) show that it is possible to learn less than $5 \\%$ of the network parameters and predict the rest without losing predictive accuracy. This evidence suggests that neural networks are often over-parameterized. ",
+        "bbox": [
+            174,
+            540,
+            825,
+            705
+        ],
+        "page_idx": 0
+    },
+    {
+        "type": "text",
+        "text": "These motivate the research on neural network compression. However, several immediate questions arise: Are these parameters easy to identify? Could we just make the network $5 \\%$ of its size and retrain? Or are more advanced methods required? There is an extensive literature in the last few years that explores the question of network compression using advanced techniques, including network prunning, loss-based compression, quantization, and matrix decomposition. We overview many of these directions in the next section. However, there is surprisingly little research on whether this over-parameterization can simply be re-captured by more efficient architectures that could be obtained from original architectures via simple transformations. ",
+        "bbox": [
+            174,
+            713,
+            825,
+            824
+        ],
+        "page_idx": 0
+    },
+    {
+        "type": "text",
+        "text": "Our approach is inspired by a very simple yet successful method called depth multiplier (Howard, 2017). In this method the depth (the number of channels) of each convolutional layer in a given network is simply reduced by a fixed fraction and the network is retrained. We generalize this approach by removing the constraint that every input filter (or channel) must be fully connected to every output filter. Instead, we use a sparse connection matrix, where each output convolution channel is connected only to a small random fraction of the input channels. Note that, for convolutional networks, this still allows for efficient computation since the one channel spatial convolution across the entire plane remains unchanged. ",
+        "bbox": [
+            174,
+            832,
+            823,
+            901
+        ],
+        "page_idx": 0
+    },
+    {
+        "type": "text",
+        "text": "",
+        "bbox": [
+            176,
+            103,
+            823,
+            146
+        ],
+        "page_idx": 1
+    },
+    {
+        "type": "text",
+        "text": "We empirically demonstrate the effectiveness of our approach on four networks (MNIST, CIFAR Net, Inception-V3 and VGG-16) of different sizes. Our results suggest that our approach outperforms dense convolutions with depth multiplier at high compression rates. ",
+        "bbox": [
+            176,
+            152,
+            821,
+            195
+        ],
+        "page_idx": 1
+    },
+    {
+        "type": "text",
+        "text": "For Inception V3 (Szegedy et al., 2016), we show that we can train a network with only about 300K of convolutional parameters1 and about 100M multiply-adds that achieves above $52 \\%$ accuracy after it is fully trained. The corresponding depth-multiplier network has only about $41 \\%$ accuracy. Another network that we consider is VGG-16n, a slightly modified version of VGG-16 (Simonyan & Zisserman, 2015), with $7 \\mathbf { x }$ fewer parameters and similar accuracy.2 We found VGG-16n to start training much faster than the original VGG-16 which was trained incrementally in the original literature. We explore the impact of sparsification and the number of parameters on the quality of the network by building the networks up to $3 0 \\mathrm { x }$ smaller than VGG-16n $2 0 0 \\mathrm { x }$ smaller than the original VGG-16). ",
+        "bbox": [
+            174,
+            202,
+            825,
+            325
+        ],
+        "page_idx": 1
+    },
+    {
+        "type": "text",
+        "text": "In terms of model flexibility, sparse connections allow for an incremental training approach, where connection structure between layers can be densified as training progresses. More importantly, the incremental training approach can potentially speed up the training significantly due to savings in the early stages of training. ",
+        "bbox": [
+            174,
+            334,
+            825,
+            390
+        ],
+        "page_idx": 1
+    },
+    {
+        "type": "text",
+        "text": "The rest of the paper is organized as follows. Section 2 summarizes relevant work. We describe our approach in Section 3 and then present some intuition in Section 4. Finally, we show our experimental results in Section 5. ",
+        "bbox": [
+            176,
+            397,
+            825,
+            438
+        ],
+        "page_idx": 1
+    },
+    {
+        "type": "text",
+        "text": "2 RELATED WORK ",
+        "text_level": 1,
+        "bbox": [
+            176,
+            454,
+            344,
+            469
+        ],
+        "page_idx": 1
+    },
+    {
+        "type": "text",
+        "text": "2.1 COMPRESSION TECHNIQUES FOR NEURAL NETWORKS ",
+        "text_level": 1,
+        "bbox": [
+            174,
+            486,
+            588,
+            500
+        ],
+        "page_idx": 1
+    },
+    {
+        "type": "text",
+        "text": "Our work is closely related to a compression technique based on network pruning. However, the important difference is that we do not try to select the connections which are redundant. Instead, we just fix a random connectivity pattern and let the network train around it. We also give a brief overview of other two popular techniques: quantization and decomposition, though these directions are not the main focus and could be complementary to our work. ",
+        "bbox": [
+            174,
+            512,
+            825,
+            582
+        ],
+        "page_idx": 1
+    },
+    {
+        "type": "text",
+        "text": "Network pruning Much initial work on neural network compression focuses on removing unimportant connections using weight decay. Hanson & Pratt (1989) introduce hyperbolic and exponential biases to the objective. Optimal Brain Damage (LeCun et al., 1989) and Optimal Brain Surgeon (Hassibi & Stork, 1993) prune the networks based on second-order derivatives of the objectives. Recent work by Han et al. (2015; 2016a) alternates between pruning near-zero weights, which are encouraged by $\\ell { 1 }$ or $\\ell 2$ regularization, and retraining the pruned networks. ",
+        "bbox": [
+            174,
+            594,
+            825,
+            678
+        ],
+        "page_idx": 1
+    },
+    {
+        "type": "text",
+        "text": "More complex regularizers have also been considered. Wen et al. (2016) and Li et al. (2016) put structured sparsity regularizers on the weights, while Murray & Chiang (2015) put them on the hidden units. Feng & Darrell (2015) explore a nonparametric prior based on the Indian buffet processes (Griffiths & Ghahramani, 2011) on layers. Hu et al. (2016) prune neurons based on the analysis of their outputs on a large dataset. Anwar et al. (2015b) consider special sparsity patterns: channel-wise (removing a feature map/channel from a layer), kernel-wise (removing all connections between two feature maps in consecutive layers), and intra-kernel-strided (removing connections between two features with particular stride and offset). They also propose to use particle filter to decide the importance of connections and paths during training. ",
+        "bbox": [
+            174,
+            684,
+            825,
+            810
+        ],
+        "page_idx": 1
+    },
+    {
+        "type": "text",
+        "text": "Another line of work explores fixed network architectures with some subsets of connections removed. For example, LeCun et al. (1998) remove connections between the first two convolutional feature maps in a completely uniform manner. This is similar to our approach but they only consider a pre-defined pattern in which the same number of input feature map are assigned to each output feature map (Random Connection Table in Torch’s SpatialConvolutionMap function). Further, they do not explore how sparse connections affect performance compared to dense networks. Along a similar vein, Cires¸an et al. (2011) remove random connections in their MNIST experiments. However, they do not try to preserve the spatial convolutional density and it might be a challenge to harvest the savings on existing hardware. Ioannou et al. (2016a) explore three types of hierarchical arrangements of filter groups for CNNs, which depend on different assumptions about co-dependency of filters within each layer. These arrangements include columnar topologies inspired by AlexNet (Krizhevsky et al., 2012), tree-like topologies previously used by Ioannou et al. (2016b), and root-like topologies. Finally, Howard (2017) proposes the depth multiplier method to scale down the number of filters in each convolutional layer by a factor. In this case, depth multiplier can be thought of channel-wise pruning mentioned in (Anwar et al., 2015b). However, depth multiplier modifies the network architectures before training and removes each layer’s feature maps in a uniform manner. ",
+        "bbox": [
+            176,
+            818,
+            821,
+            859
+        ],
+        "page_idx": 1
+    },
+    {
+        "type": "text",
+        "text": "",
+        "bbox": [
+            174,
+            103,
+            825,
+            297
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "text",
+        "text": "With the exception of (Anwar et al., 2015b; Li et al., 2016; Ioannou et al., 2016a) and depth multiplier (Howard, 2017), the above previous work performs connection pruning that leads to irregular network architectures. Thus, those techniques require additional efforts to represent network connections and might or might not allow for direct computational savings. ",
+        "bbox": [
+            174,
+            304,
+            823,
+            361
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "text",
+        "text": "Quantization Reducing the degree of redundancy of model parameters can be done in the form of quantization of network parameters. Hwang & Sung (2014); Arora et al. (2014) and Courbariaux et al. (2015; 2016); Rastegari et al. (2016) propose to train CNNs with ternary weights and binary weights, respectively. Gong et al. (2014) use vector quantization for parameters in fully connected layers. Anwar et al. (2015a) quantize a network with the squared error minimization. Chen et al. (2015) randomly group network parameters using a hash function. We note that this technique could be complementary to network pruning. For example, Han et al. (2016a) combine connection pruning in (Han et al., 2015) with quantization and Huffman coding. ",
+        "bbox": [
+            174,
+            375,
+            825,
+            486
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "text",
+        "text": "Decomposition Another approach is based on low-rank decomposition of the parameters. Decomposition methods include truncated SVD (Denton et al., 2014), decomposition to rank-1 bases (Jaderberg et al., 2014), CP decomposition (PARAFAC or CANDECOMP) (Lebedev et al., 2015), Tensor-Train decomposition of Oseledets (2011) (Novikov et al., 2015), sparse dictionary learning of Mairal et al. (2009) and PCA (Liu et al., 2015), asymmetric (3D) decomposition using reconstruction loss of non-linear responses combined with a rank selection method based on PCA accumulated energy (Zhang et al., 2015b;a), and Tucker decomposition using the kernel tensor reconstruction loss combined with a rank selection method based on global analytic variational Bayesian matrix factorization (Kim et al., 2016). ",
+        "bbox": [
+            173,
+            500,
+            825,
+            625
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "text",
+        "text": "2.2 REGULARIZATION OF NEURAL NETWORKS ",
+        "text_level": 1,
+        "bbox": [
+            174,
+            638,
+            513,
+            654
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "text",
+        "text": "Hinton et al. (2012); Srivastava et al. (2014) propose Dropout for regularizing fully connected layers within neural networks layers by randomly setting a subset of activations to zero during training. Wan et al. (2013) later propose DropConnect, a generalization of Dropout that instead randomly sets a subset of weights or connections to zero. Our approach could be thought as related to DropConnect, but (1) we remove connections before training; (2) we focus on connections between convolutional layers; and (3) we kill connections in a more regular manner by restricting connection patterns to be the same along spatial dimensions. ",
+        "bbox": [
+            174,
+            666,
+            823,
+            763
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "text",
+        "text": "Recently, Han et al. (2016b) and Jin et al. (2016) propose a form of regularization where dropped connections are unfrozen and the network is retrained. This idea is similar to our incremental training approach. However, (1) we do not start with a full network; (2) we do not unfreeze connections all at once; and (3) we preserve regularity of the convolution operation. ",
+        "bbox": [
+            174,
+            770,
+            823,
+            827
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "text",
+        "text": "2.3 NEURAL NETWORK ARCHITECTURES ",
+        "text_level": 1,
+        "bbox": [
+            176,
+            840,
+            468,
+            854
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "text",
+        "text": "Network compression and architectures are closely related. The goal of compression is to remove redundancy in network parameters; therefore, the knowledge about traits that determine architecture’s success would be desirable. Other than the discovery that depth is an important factor (Ba & Caruana, 2014), little is known about such traits. ",
+        "bbox": [
+            176,
+            867,
+            823,
+            922
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "text",
+        "text": "Some previous work performs architecture search but without the main goal of doing compression (Murray & Chiang, 2015; De Brabandere et al., 2016). Recent work proposes shortcut/skip connections to convolutional networks. See, among others, highway networks (Srivastava et al., 2015), residual networks (He et al., 2016a;b), networks with stochastic depth (Huang et al., 2016b), and densely connected convolutional networks (Huang et al., 2016a). ",
+        "bbox": [
+            174,
+            103,
+            825,
+            174
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "3 APPROACH",
+        "text_level": 1,
+        "bbox": [
+            174,
+            188,
+            299,
+            204
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "A CNN architecture consist of (1) convolutional layers, (2) pooling layers, (3) fully connected layers, and (4) a topology that governs how these layers are organized. Given an architecture, our general goal is to transform it into another architecture with a smaller number of parameters. In this paper, we limit ourselves to transformation functions that keep the general topology of the input architecture intact. Moreover, the main focus will be on the convolutional layers and convolution operations, as they impose highest computational and memory burden for most if not all large networks. ",
+        "bbox": [
+            174,
+            218,
+            825,
+            301
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "3.1 DEPTH MULTIPLIER ",
+        "text_level": 1,
+        "bbox": [
+            176,
+            314,
+            351,
+            327
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "We first give a description of the depth multiplier method used in Howard (2017). Given a hyperparameter $\\alpha \\in ( 0 , 1 ]$ , the depth multiplier approach scales down the number of filters in each convolutional layers by $\\alpha$ . Note that depth here refers to the third dimension of the activation volume of a single layer, not the number of layers in the whole network. ",
+        "bbox": [
+            174,
+            340,
+            825,
+            396
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "Let $n _ { l - 1 }$ and $n _ { l }$ be the number of input and output filters at layer $l$ , respectively. After the operation $n _ { l - 1 }$ and $n _ { l }$ become $\\lceil \\alpha n _ { l - 1 } \\rceil$ and $\\lceil \\alpha n _ { l } \\rceil$ and the number of parameters (and the number of multiplications) becomes $\\approx \\alpha ^ { 2 }$ of the original number. ",
+        "bbox": [
+            176,
+            404,
+            821,
+            445
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "The result of this operation is a network that is both $1 / \\alpha ^ { 2 }$ smaller and faster. Many large networks can be significantly reduced in size using this method with only a small loss of precision (Howard, 2017). It is our belief that this method establishes a strong baseline to which any other advanced techniques should compare themselves. To the best of our knowledge, we are not aware of such comparisons in the literature. ",
+        "bbox": [
+            174,
+            452,
+            825,
+            522
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "3.2 SPARSE RANDOM ",
+        "text_level": 1,
+        "bbox": [
+            176,
+            535,
+            338,
+            549
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "Instead of looking at depth multiplier as deactivating channels in the convolutional layers, we can look at it from the perspective of deactivating connections. From this point of view, depth multiplier kills the connections between two convolutional layers such that (a) the connection patterns are still the same across spatial dimensions and (b) all “alive” input channels are fully connected to all “alive” output channels. ",
+        "bbox": [
+            174,
+            561,
+            825,
+            631
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "We generalize this approach by relaxing (b) while maintaining (a). That is, for every output channel, we connect it to a small subset of input channels. In other words, dense connections between a small number of channels become sparse connections between larger number of channels. This can be summarized in Fig. 1. The advantage of this is that the actual convolution can still be computed efficiently because sparsity is introduced only at the outer loop of the convolution operation and we can still take the advantage of the continuous memory layout. For more details regarding implementations of the two approaches, please refer to the Appendix. ",
+        "bbox": [
+            174,
+            638,
+            825,
+            736
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "More concretely, let $n _ { l - 1 }$ and $n _ { l }$ be the number of channels of layer $l - 1$ and layer $l$ , respectively. For a sparsity coefficient $\\alpha$ , each output filter $j$ only connects to an $\\alpha$ fraction of filters of the previous layer. Thus, instead of having a connectivity matrix $W _ { s i j }$ of dimension $k ^ { 2 } \\times n _ { l - 1 } \\times n _ { l }$ , we have a sparse matrix with non-zero entries at $W _ { s a _ { i j } j }$ , where $a _ { i j }$ is an index matrix of dimension $k ^ { 2 } \\times \\alpha n _ { l - 1 } \\times n _ { l }$ and $k$ is the kernel size. ",
+        "bbox": [
+            174,
+            742,
+            825,
+            814
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "3.2.1 INCREMENTAL TRAINING ",
+        "text_level": 1,
+        "bbox": [
+            176,
+            827,
+            403,
+            842
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "In contrast to depth multiplier, a sparse convolutional network defines a connection pattern on a much bigger network. Therefore, an interesting extension is to consider incremental training: we start with a network that only contains a small fraction of connections (in our experiments we use $1 \\%$ and $0 . 1 \\%$ ) and add connections over time. This is motivated by an intuition that the network can use learned channels in new contexts by introducing additional connections. The potential practical advantage of this approach is that since we start training with very small networks and grow them over time, this approach has a potential to speed up the whole training process significantly. We note that depth multiplier will not benefit from this approach as any newly activated connections would require learning new filters from scratch. ",
+        "bbox": [
+            174,
+            853,
+            823,
+            924
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "image",
+        "img_path": "images/66fc8a8512a12ff04f8f7dc6aab17648934eff7a1161c2536e3fa54b1fa83cde.jpg",
+        "image_caption": [
+            "Figure 1: Connection tensors of depth multiplier (left) and sparse random (right) approaches for $n _ { l - 1 } = 5$ and $n _ { l } = 1 0$ . Yellow denotes active connections. For both approaches, the connection pattern is the same across spatial dimension and fixed before training. However, in the sparse random approach, each output channel is connected to a (possibly) different subset of input channels, and vice versa. "
+        ],
+        "image_footnote": [],
+        "bbox": [
+            204,
+            106,
+            782,
+            185
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "text",
+        "text": "",
+        "bbox": [
+            174,
+            281,
+            825,
+            338
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "text",
+        "text": "4 ANALYSIS ",
+        "text_level": 1,
+        "bbox": [
+            174,
+            356,
+            290,
+            371
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "text",
+        "text": "In this section, we approach a question of why sparse convolutions are frequently more efficient than the dense convolutions with the same number of parameters. Our main intuition is that the sparse convolutional networks promote diversity. It is much harder to learn equivalent set of channels as, at high sparsity, channels have distinct connection structure or even overlapping connections. This can be formalized with a simple observation that any dense network is in fact a part of an exponentially large equivalence class, which is guaranteed to produce the same output for every input. ",
+        "bbox": [
+            173,
+            386,
+            825,
+            470
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "text",
+        "text": "Lemma 1 Any dense convolutional neural network with no cross-channel nonlinearities, distinct weights and biases, and with l hidden layers of sizes $n _ { 1 }$ , $n _ { 2 }$ , . . . , $n _ { l }$ , has at least $\\textstyle \\prod _ { i = 1 } ^ { l } n _ { i } !$ distinct equivalent networks which produce the same output. ",
+        "bbox": [
+            173,
+            484,
+            825,
+            530
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "text",
+        "text": "Proof Let $I$ denote the input to the network, $C _ { i }$ be the convolutional operator, $\\sigma _ { i }$ denote the nonlinearity operator applied to the $i$ -th convolution layer and $S$ be a final transformation (e.g. softmax classifier). We assume that $\\sigma _ { i }$ is a function that operates on each of the channels independently. We note that this is the case for almost any modern network. The output of the network can then be written as: ",
+        "bbox": [
+            173,
+            542,
+            825,
+            613
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "equation",
+        "img_path": "images/beea7d489507a37291c18e46788f828e6040412e7e84a7604ead6cd1d103d3a8.jpg",
+        "text": "$$\n\\mathcal { N } ( I ) \\equiv S \\circ \\sigma _ { l } \\circ C _ { l } \\circ \\sigma _ { l - 1 } \\circ \\cdots \\circ \\sigma _ { 1 } \\circ C _ { 1 } ( I )\n$$",
+        "text_format": "latex",
+        "bbox": [
+            346,
+            617,
+            650,
+            635
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "text",
+        "text": "where we use $\\circ$ to denote function composition to avoid numerous parentheses. The convolution operator $C _ { i }$ operates on input with $n _ { i - 1 }$ channels and produces an output with $n _ { i }$ channels. Now, fix arbitrary set of permutation functions $\\pi _ { i }$ , where $\\pi _ { i }$ can permute depth of size $n _ { i }$ . Since $\\pi _ { i }$ is a linear function, it follows that $C _ { i } ^ { \\prime } = \\pi _ { i } ^ { - 1 } C _ { i } \\pi _ { i - 1 }$ is a valid convolutional operator, which can be obtained from $C _ { i }$ by permuting its bias according to $\\pi _ { i }$ and its weight matrix along input and output dimensions according to $\\pi _ { i - 1 }$ and $\\pi _ { i }$ respectively. For a new network defined as: ",
+        "bbox": [
+            173,
+            643,
+            825,
+            728
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "equation",
+        "img_path": "images/f9d507d6a130c4c9052e073184e7bcd8a6aa91888fa306ef72272247738d2624.jpg",
+        "text": "$$\n\\mathcal { N } ^ { \\prime } ( I ) = S ^ { \\prime } \\circ \\sigma _ { l } \\circ C _ { l } ^ { \\prime } \\circ \\sigma _ { l - 1 } \\circ \\cdots \\circ \\sigma _ { 1 } \\circ C _ { 1 } ^ { \\prime } ( I ) ,\n$$",
+        "text_format": "latex",
+        "bbox": [
+            339,
+            739,
+            656,
+            757
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "text",
+        "text": "where $\\pi _ { 0 }$ is an identity operator and $S ^ { \\prime } \\equiv S \\circ \\pi _ { l }$ , we claim that $\\mathcal { N } ^ { \\prime } ( I ) \\equiv \\mathcal { N } ( I )$ . Indeed, since nonlinearities do not apply cross-depth we have $\\pi _ { n } \\sigma _ { n } \\pi _ { n } ^ { - 1 } \\equiv \\sigma _ { n }$ and thus: ",
+        "bbox": [
+            173,
+            767,
+            823,
+            797
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "equation",
+        "img_path": "images/f024c18d6595d3e2ba9aa755d9d14deecbf1af5c05c6f9f759b65544ecec492d.jpg",
+        "text": "$$\n\\begin{array} { r l } & { \\mathscr { N } ^ { \\prime } ( I ) = S ^ { \\prime } \\circ \\sigma _ { l } \\circ C _ { l } ^ { \\prime } \\circ \\sigma _ { l - 1 } \\circ \\cdots \\circ \\sigma _ { 1 } \\circ C _ { 1 } ^ { \\prime } ( I ) = } \\\\ & { \\qquad = S \\circ \\pi _ { l } \\circ \\sigma _ { l } \\circ \\pi _ { l } ^ { - 1 } \\circ C _ { l } \\circ \\pi _ { l - 1 } \\circ \\cdots \\circ \\pi _ { 1 } \\circ \\sigma _ { 1 } \\circ \\pi _ { 1 } ^ { - 1 } \\circ C _ { 1 } ( I ) = \\mathscr { N } ( I ) . } \\end{array}\n$$",
+        "text_format": "latex",
+        "bbox": [
+            187,
+            809,
+            808,
+            852
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "text",
+        "text": "Thus, any set of permutations on hidden units defines an equivalent network. ",
+        "bbox": [
+            176,
+            859,
+            676,
+            875
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "text",
+        "text": "It is obvious that sparse networks are much more immune to parameter permutation – indeed every channel at layer $l$ is likely to have a unique tree describing its connection matrix all the way down. Exploring this direction is an interesting open question. ",
+        "bbox": [
+            174,
+            881,
+            823,
+            924
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "text",
+        "text": "5 EXPERIMENTS ",
+        "text_level": 1,
+        "bbox": [
+            174,
+            102,
+            326,
+            118
+        ],
+        "page_idx": 5
+    },
+    {
+        "type": "text",
+        "text": "In this section, we demonstrate the effectiveness of the sparse random approach by comparing it to the depth multiplier approach at different compression rates. Moreover, we examine several settings in the incremental training where connections gradually become active during the training process. ",
+        "bbox": [
+            174,
+            133,
+            825,
+            176
+        ],
+        "page_idx": 5
+    },
+    {
+        "type": "text",
+        "text": "5.1 SETUP ",
+        "text_level": 1,
+        "bbox": [
+            174,
+            193,
+            261,
+            207
+        ],
+        "page_idx": 5
+    },
+    {
+        "type": "text",
+        "text": "Networks and Datasets Our experiments are conducted on 4 networks for 3 different datasets.   \nAll our experiments use open-source TensorFlow networks Abadi et al. (2015). ",
+        "bbox": [
+            176,
+            222,
+            823,
+            250
+        ],
+        "page_idx": 5
+    },
+    {
+        "type": "text",
+        "text": "MNIST AND CIFAR-10 We use standard networks provided by TensorFlow. For MNIST, it has 3-layer convolutional layers and achieves $9 9 . 5 \\%$ accuracy when fully trained. For CIFAR-10, it has 2 convolutional layers and achieves $87 \\%$ accuracy. ",
+        "bbox": [
+            176,
+            267,
+            823,
+            309
+        ],
+        "page_idx": 5
+    },
+    {
+        "type": "text",
+        "text": "IMAGENET We use open source Inception-V3 (Szegedy et al., 2016) network and a slightly modified version of VGG-16 (Simonyan & Zisserman, 2015) called VGG-16n on ImageNet ILSVRC 2012 (Deng et al., 2009; Russakovsky et al., 2015). ",
+        "bbox": [
+            176,
+            327,
+            821,
+            369
+        ],
+        "page_idx": 5
+    },
+    {
+        "type": "text",
+        "text": "Random connections Connections are activated according to their likelihood from the uniform distribution. In addition, they are activated in such a way that there are no connections going in or coming out of dead filters (i.e., any connection must have a path to input image and a path to the final prediction.). All connections in fully connected layers are retained. ",
+        "bbox": [
+            174,
+            387,
+            825,
+            443
+        ],
+        "page_idx": 5
+    },
+    {
+        "type": "text",
+        "text": "Implementation details All code is implemented in TensorFlow (Abadi et al., 2015). Deactivating connections is done by applying masks to parameter tensors. The Inception-v3 and VGG-16n networks are trained on 8 Tesla K80 GPUs, each with batch size 256 (32 per gpu) and batch normalization was used for all networks. ",
+        "bbox": [
+            176,
+            460,
+            823,
+            516
+        ],
+        "page_idx": 5
+    },
+    {
+        "type": "text",
+        "text": "5.2 COMPARISON BETWEEN SPARSE RANDOM AND DEPTH MULTIPLIER ",
+        "bbox": [
+            176,
+            536,
+            676,
+            549
+        ],
+        "page_idx": 5
+    },
+    {
+        "type": "text",
+        "text": "5.2.1 MNIST AND CIFAR-10 ",
+        "text_level": 1,
+        "bbox": [
+            176,
+            563,
+            398,
+            578
+        ],
+        "page_idx": 5
+    },
+    {
+        "type": "text",
+        "text": "We first compare depth multiplier and sparse random for the two small networks on MNIST and CIFAR-10. We compare the accuracy of the two approaches when the numbers of connections are roughly the same, based on a hyperparameter $\\alpha$ . For dense convolutions, we pick a multiplier $\\alpha$ and each filter depth is scaled down by $\\sqrt { \\alpha }$ and then rounded up. In sparse convolutions, a fraction $\\alpha$ of connections are randomly deactivated if those parameters connect at least two filters on each layer; otherwise, a fraction of $\\sqrt { \\alpha }$ is used instead if the parameters connect layers with only one filter left. The accuracy numbers are averaged over 5 rounds for MNIST and 2 rounds on CIFAR-10. ",
+        "bbox": [
+            174,
+            592,
+            825,
+            689
+        ],
+        "page_idx": 5
+    },
+    {
+        "type": "text",
+        "text": "We show in Fig. 2 and Fig. 3 that the sparse networks have comparable or higher accuracy for the same number of parameters, with comparable accuracy at higher density. We note however that these networks are so small that at high compression rates most of operations are concentrated at the first layer, which is negligible for large networks. Moreover, in MNIST example, the size of network changes most dramatically from 2000 to 2 million parameters, while affecting accuracy only by $1 \\%$ . This observation suggests that there might be benefits of maintaining the number of filters to be high and/or breaking the symmetry of connections. We explore this in the next section. ",
+        "bbox": [
+            174,
+            696,
+            825,
+            794
+        ],
+        "page_idx": 5
+    },
+    {
+        "type": "text",
+        "text": "5.2.2 INCEPTION-V3 ON IMAGENET",
+        "text_level": 1,
+        "bbox": [
+            176,
+            810,
+            436,
+            825
+        ],
+        "page_idx": 5
+    },
+    {
+        "type": "text",
+        "text": "We consider different values of sparsity ranging from 0.003 to 1, and depth multiplier from 0.05 to 1. Our experiments show (see Table 1 and Fig. 4) significant advantage of sparse networks over equivalently sized dense networks. We note that due to time constraints the reported quantitative numbers are preliminary, as the networks have not finished converging. We expect the final numbers to match the reported number for Inception V3 Szegedy et al. (2016), and the smaller networks to have comparable improvement. ",
+        "bbox": [
+            174,
+            840,
+            825,
+            922
+        ],
+        "page_idx": 5
+    },
+    {
+        "type": "image",
+        "img_path": "images/65926ca6ffb9b876a5add68d622ccc4f1c7a0fd94934ef67cf453485cb422141.jpg",
+        "image_caption": [
+            "Figure 2: Comparison of accuracy (averaged over 5 rounds) vs. Number of parameters/Number of multiply-adds between dense and sparse convolutions on MNIST dataset. Note that though sparse convolution result in better parameter trade-off curve, the multiply-add curve shows the opposite pattern. "
+        ],
+        "image_footnote": [],
+        "bbox": [
+            204,
+            101,
+            784,
+            271
+        ],
+        "page_idx": 6
+    },
+    {
+        "type": "text",
+        "text": "Table 1: Inception V3: Preliminary quantitative results after 100 Epochs. Note the smallest sparse network is actually a hybrid network - we used both depth multiplier (0.5) and sparsity (0.01). The number of parameters is the number of parameters excluding the softmax layer. ",
+        "bbox": [
+            176,
+            340,
+            825,
+            383
+        ],
+        "page_idx": 6
+    },
+    {
+        "type": "table",
+        "img_path": "images/b34ddde6d213af5b60d226747df5ee3604eee0c1cb8d2d5689b2231902ee2059.jpg",
+        "table_caption": [
+            "Accuracy for sparse convolutions "
+        ],
+        "table_footnote": [],
+        "table_body": "<table><tr><td>Sparsity</td><td>MAdds Params</td><td>P@1</td></tr><tr><td>0.50/0.01 0.003</td><td>43.0M 90k 158k</td><td>40.3 46.1</td></tr><tr><td>0.01</td><td>82.0M 104M</td><td>52.3</td></tr><tr><td>0.03</td><td>287k 208M 724k</td><td></td></tr><tr><td>0.10</td><td></td><td>59.5</td></tr><tr><td>0.30</td><td>628M 2.3 M</td><td>67.2</td></tr><tr><td></td><td>1.80 B 6.6M</td><td>73</td></tr><tr><td>0.60</td><td>3.50B 13M 22 M</td><td>75</td></tr><tr><td>1.00</td><td>5.70 B</td><td>77</td></tr></table>",
+        "bbox": [
+            215,
+            409,
+            486,
+            537
+        ],
+        "page_idx": 6
+    },
+    {
+        "type": "table",
+        "img_path": "images/f13c73e145cd97e43929b7f6bff27130ece79ed9e82f3f37297cc3a19a01a5c9.jpg",
+        "table_caption": [
+            "Accuracy for Depth Multiplier "
+        ],
+        "table_footnote": [],
+        "table_body": "<table><tr><td>Multiplier</td><td>MAdds</td><td>Params P@1</td></tr><tr><td>0.05</td><td>55.0M 56k</td><td>24.6</td></tr><tr><td>0.10</td><td>75.0M</td><td>170k 38.6</td></tr><tr><td>0.20</td><td>183M 718k</td><td>54.2</td></tr><tr><td>0.30</td><td>439M 1.8M</td><td>64.0</td></tr><tr><td>0.50</td><td>1.40B 5.4M</td><td>72.3</td></tr><tr><td>0.80</td><td>3.40B 13M</td><td>75.6</td></tr></table>",
+        "bbox": [
+            534,
+            422,
+            815,
+            523
+        ],
+        "page_idx": 6
+    },
+    {
+        "type": "text",
+        "text": "Original network: 5.70 B 22 M 77 (78.8) ",
+        "bbox": [
+            194,
+            549,
+            509,
+            564
+        ],
+        "page_idx": 6
+    },
+    {
+        "type": "text",
+        "text": "5.2.3 VGG-16 ON IMAGENET ",
+        "text_level": 1,
+        "bbox": [
+            174,
+            594,
+            397,
+            609
+        ],
+        "page_idx": 6
+    },
+    {
+        "type": "text",
+        "text": "In our experiments with the VGG-16 network (Simonyan & Zisserman, 2015), we modify the model architecture (calling it VGG-16n) by removing the two fully-connected layers with depth 4096 and replacing them with a $2 \\times 2$ maxpool layer followed by a $3 \\times 3$ convolutional layer with the depth of 1024. This alone sped up our training significantly. The comparison between depth multiplier and sparse connection approaches is shown in Fig. 5. The modified VGG-16n network has about 7 times fewer parameters, but appears to have comparable precision. ",
+        "bbox": [
+            173,
+            622,
+            825,
+            707
+        ],
+        "page_idx": 6
+    },
+    {
+        "type": "image",
+        "img_path": "images/f4cf6c184c4020dc5874321b477ae62c97ddf25c9d68322157c1a605a04d32e7.jpg",
+        "image_caption": [
+            "Figure 3: Comparison of accuracy (averaged over 2 rounds) vs. Number of parameters/Number of multiply-adds between dense and sparse convolutions on CIFAR-10 dataset. "
+        ],
+        "image_footnote": [],
+        "bbox": [
+            222,
+            734,
+            767,
+            896
+        ],
+        "page_idx": 6
+    },
+    {
+        "type": "image",
+        "img_path": "images/1cf4cc4464141afef87dfda693d69954f6e151da17bb74348a13d18643a16d98.jpg",
+        "image_caption": [
+            "Figure 4: Inception V3: Comparison of Precision $@ 1$ vs. Number of parameters/Number of multiply-adds between dense and sparse convolutions on ImageNet/Inception-V3. The full network corresponds to the right-most point of the curve. "
+        ],
+        "image_footnote": [],
+        "bbox": [
+            204,
+            102,
+            787,
+            273
+        ],
+        "page_idx": 7
+    },
+    {
+        "type": "image",
+        "img_path": "images/a57e0b1ff2bb1e2c18547e1723279c8b784b21823938c56fcb8383c2d7879efa.jpg",
+        "image_caption": [
+            "Figure 5: VGG 16: Preliminary Quantitative Results. Comparison of Precision $@ 1$ vs. Number of parameters/Number of multiply-adds between dense and sparse convolutions on ImageNet/VGG16n. The full network corresponds to the right-most point of the curve. Original VGG-16 as described in Simonyan & Zisserman (2015) (blue star) and the same model trained by us from scratch (red cross) are also shown. "
+        ],
+        "image_footnote": [],
+        "bbox": [
+            205,
+            335,
+            785,
+            507
+        ],
+        "page_idx": 7
+    },
+    {
+        "type": "text",
+        "text": "5.3 INCREMENTAL TRAINING ",
+        "text_level": 1,
+        "bbox": [
+            174,
+            609,
+            392,
+            622
+        ],
+        "page_idx": 7
+    },
+    {
+        "type": "image",
+        "img_path": "images/f704c8a1c6906e9b1cd74e0b702c578d2acd95067c474ee8b78c09edf9df95a7.jpg",
+        "image_caption": [
+            "Figure 6: Incremental Training Of Inception V3: We show Precision $@ 1$ during the training process, where the networks densify over time. The saturation points show where the networks actually reach their full density. "
+        ],
+        "image_footnote": [],
+        "bbox": [
+            310,
+            657,
+            676,
+            881
+        ],
+        "page_idx": 7
+    },
+    {
+        "type": "text",
+        "text": "Finally, we show that incremental training is a promising direction. We start with a very sparse model and increase its density over time, using the approach described in Sect. 3.2.1. We note that a naive approach where we simply add filters results in training process basically equivalent to as if it started from scratch in every step. On the other hand, when the network densifies over time, all channels already possess some discriminative power and that information is utilized. ",
+        "bbox": [
+            174,
+            103,
+            825,
+            172
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "In our experiments, we initially start training Inception-V3 with only $1 \\%$ or $0 . 1 \\%$ of connections enabled. Then, we double the number of connections every $T$ steps. We use $T = 1 0 , 0 0 0$ , $T =$ 25, 000 and $T = 5 0 , 0 0 0$ . The results are presented in Fig. 6. We show that the networks trained with the incremental approach regardless of the doubling period can catch up with the full Inception-V3 network (in some cases with small gains). Moreover, they recover very quickly from adding more (untrained) connections. In fact, the recovery is so fast that it is shorter than our saving interval for all the networks except for the network with 10K doubling period (resulting in the sharp drop). We believe that incremental training is a promising direction to speeding up the training of large convolutional neural networks since early stages of the training require much less computation. ",
+        "bbox": [
+            174,
+            180,
+            825,
+            305
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "6 CONCLUSION ",
+        "text_level": 1,
+        "bbox": [
+            176,
+            320,
+            318,
+            337
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "We have proposed a new compression technique that uses a sparse random connection structure between input-output filters in convolutional layers of CNNs. We fix this structure before training and use the same structure across spatial dimensions to harvest savings from modern hardware. We show that this approach is especially useful at very high compression rates for large networks. For example, this simple method when applied to Inception V3 (Fig. 4), achieves AlexNet-level accuracy (Krizhevsky et al., 2012) with fewer than 400K parameters and VGG-level one (Fig. 5) with roughly $3 . 5 \\mathrm { M }$ parameters. The simplicity of our approach is instructive in that it establishes a strong baseline to compare against when developing more advanced techniques. On the other hand, the uncanny match in performance of dense and equivalently-sized sparse networks with sparsity $> 0 . 1$ suggests that there might be some fundamental property of network architectures that is controlled by the number of parameters, regardless of how they are organized. Exploring this further might yield additional insights on understanding neural networks. ",
+        "bbox": [
+            173,
+            352,
+            825,
+            517
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "In addition, we show that our method leads to an interesting novel incremental training technique, where we take advantage of sparse (and smaller) models to build a dense network. One interesting open direction is to enable incremental training not to simply densify the network over time, but also increase the number of channels. This would allow us to grow the network without having to fix its original shape in place. ",
+        "bbox": [
+            174,
+            525,
+            825,
+            594
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "REFERENCES ",
+        "text_level": 1,
+        "bbox": [
+            174,
+            611,
+            285,
+            626
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "Mart´ın Abadi, Ashish Agarwal, Paul Barham, Eugene Brevdo, Zhifeng Chen, Craig Citro, Greg S. Corrado, Andy Davis, Jeffrey Dean, Matthieu Devin, Sanjay Ghemawat, Ian Goodfellow, Andrew Harp, Geoffrey Irving, Michael Isard, Yangqing Jia, Rafal Jozefowicz, Lukasz Kaiser, Manjunath Kudlur, Josh Levenberg, Dan Mane, Rajat Monga, Sherry Moore, Derek Murray, Chris Olah, Mike Schuster, Jonathon Shlens, Benoit ´ Steiner, Ilya Sutskever, Kunal Talwar, Paul Tucker, Vincent Vanhoucke, Vijay Vasudevan, Fernanda Viegas, ´ Oriol Vinyals, Pete Warden, Martin Wattenberg, Martin Wicke, Yuan Yu, and Xiaoqiang Zheng. TensorFlow: Large-scale machine learning on heterogeneous systems, 2015. URL http://tensorflow.org/. Software available from tensorflow.org. ",
+        "bbox": [
+            176,
+            640,
+            825,
+            742
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "Sajid Anwar, Kyuyeon Hwang, and Wonyong Sung. Fixed point optimization of deep convolutional neural networks for object recognition. In ICASSP, 2015a. ",
+        "bbox": [
+            169,
+            755,
+            823,
+            781
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "Sajid Anwar, Kyuyeon Hwang, and Wonyong Sung. Structured pruning of deep convolutional neural networks. arXiv preprint arXiv:1512.08571, 2015b. ",
+        "bbox": [
+            171,
+            792,
+            821,
+            819
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "Sanjeev Arora, Aditya Bhaskara, Rong Ge, and Tengyu Ma. Provable bounds for learning some deep representations. In ICML, 2014. ",
+        "bbox": [
+            171,
+            832,
+            821,
+            858
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "Jimmy Ba and Rich Caruana. Do deep nets really need to be deep? In NIPS, 2014. ",
+        "bbox": [
+            173,
+            871,
+            663,
+            885
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "Wenlin Chen, James T. Wilson, Stephen Tyree, Kilian Q. Weinberger, and Yixin Chen. Compressing neural networks with the hashing trick. In ICML, 2015. ",
+        "bbox": [
+            173,
+            897,
+            825,
+            922
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "Dan C. Cires¸an, Ueli Meier, Jonathan Masci, Luca M. Gambardella, and Jurgen Schmidhuber. High- ¨ performance neural networks for visual object classification. arXiv preprint arXiv:1102.0183, 2011.   \nMatthieu Courbariaux, Yoshua Bengio, and Jean-Pierre David. Binaryconnect: Training deep neural networks with binary weights during propagations. In NIPS, 2015.   \nMatthieu Courbariaux, Itay Hubara, Daniel Soudry, Ran El-Yaniv, and Yoshua Bengio. Binarized neural networks: Training deep neural networks with weights and activations constrained to $+ 1$ or -1. arXiv preprint arXiv:1602.02830, 2016.   \nBert De Brabandere, Xu Jia, Tinne Tuytelaars, and Luc Van Gool. Dynamic filter networks. In NIPS, 2016.   \nJia Deng, Wei Dong, Richard Socher, Li-Jia Li, Kai Li, and Li Fei-Fei. Imagenet: A large-scale hierarchical image database. In CVPR, 2009.   \nMisha Denil, Babak Shakibi, Laurent Dinh, Marc Aurelio Ranzato, and Nando de Freitas. Predicting parameters in deep learning. In NIPS, 2013.   \nEmily L. Denton, Wojciech Zaremba, Joan Bruna, Yann LeCun, and Rob Fergus. Exploiting linear structure within convolutional networks for efficient evaluation. In NIPS, 2014.   \nJiashi Feng and Trevor Darrell. Learning the structure of deep convolutional networks. In ICCV, 2015.   \nYunchao Gong, Liu Liu, Ming Yang, and Lubomir Bourdev. Compressing deep convolutional networks using vector quantization. arXiv preprint arXiv:1412.6115, 2014.   \nThomas L. Griffiths and Zoubin Ghahramani. The indian buffet process: An introduction and review. Journal of Machine Learning Research, 12(Apr):1185–1224, 2011.   \nSong Han, Jeff Pool, John Tran, and William Dally. Learning both weights and connections for efficient neural network. In NIPS, 2015.   \nSong Han, Huizi Mao, and William J. Dally. Deep compression: Compressing deep neural network with pruning, trained quantization and huffman coding. In ICLR, 2016a.   \nSong Han, Jeff Pool, Sharan Narang, Huizi Mao, Shijian Tang, Erich Elsen, Bryan Catanzaro, John Tran, and William J. Dally. Dsd: Regularizing deep neural networks with dense-sparse-dense training flow. arXiv preprint arXiv:1607.04381, 2016b.   \nStephen Jose Hanson and Lorien Y. Pratt. Comparing biases for minimal network construction with back- ´ propagation. In NIPS, 1989.   \nBabak Hassibi and David G. Stork. Second order derivatives for network pruning: Optimal brain surgeon. In NIPS, 1993.   \nKaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image recognition. In CVPR, 2016a.   \nKaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Identity mappings in deep residual networks. In ECCV, 2016b.   \nGeoffrey E. Hinton, Nitish Srivastava, Alex Krizhevsky, Ilya Sutskever, and Ruslan Salakhutdinov. Improving neural networks by preventing co-adaptation of feature detectors. arXiv preprint arXiv:1207.0580, 2012.   \nAndrew G. Howard. Mobilenets: Efficient convolutional neural networks for mobile vision applications. In forthcoming, 2017.   \nHengyuan Hu, Rui Peng, Yu-Wing Tai, and Chi-Keung Tang. Network trimming: A data-driven neuron pruning approach towards efficient deep architectures. arXiv preprint arXiv:1607.03250, 2016.   \nGao Huang, Zhuang Liu, and Kilian Q. Weinberger. Densely connected convolutional networks. arXiv preprint arXiv:1608.06993, 2016a.   \nGao Huang, Yu Sun, Zhuang Liu, Daniel Sedra, and Kilian Weinberger. Deep networks with stochastic depth. arXiv preprint arXiv:1603.09382, 2016b.   \nKyuyeon Hwang and Wonyong Sung. Fixed-point feedforward deep neural network design using weights $+ 1$ , 0, and -1. In IEEE Workshop on Signal Processing Systems (SiPS), 2014. ",
+        "bbox": [
+            171,
+            94,
+            828,
+            929
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "Yani Ioannou, Duncan Robertson, Roberto Cipolla, and Antonio Criminisi. Deep roots: Improving cnn efficiency with hierarchical filter groups. arXiv preprint arXiv:1605.06489, 2016a. ",
+        "bbox": [
+            171,
+            103,
+            821,
+            131
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Yani Ioannou, Duncan Robertson, Jamie Shotton, Roberto Cipolla, and Antonio Criminisi. Training cnns with low-rank filters for efficient image classification. In ICLR, 2016b. ",
+        "bbox": [
+            174,
+            140,
+            821,
+            166
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Max Jaderberg, Andrea Vedaldi, and Andrew Zisserman. Speeding up convolutional neural networks with low rank expansions. In BMVC, 2014. ",
+        "bbox": [
+            176,
+            174,
+            821,
+            202
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Xiaojie Jin, Xiaotong Yuan, Jiashi Feng, and Shuicheng Yan. Training skinny deep neural networks with iterative hard thresholding methods. arXiv preprint arXiv:1607.05423, 2016. ",
+        "bbox": [
+            174,
+            210,
+            823,
+            238
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Yong-Deok Kim, Eunhyeok Park, Sungjoo Yoo, Taelim Choi, Lu Yang, and Dongjun Shin. Compression of deep convolutional neural networks for fast and low power mobile applications. In ICLR, 2016. ",
+        "bbox": [
+            173,
+            246,
+            823,
+            273
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Alex Krizhevsky, Ilya Sutskever, and Geoffrey E. Hinton. Imagenet classification with deep convolutional neural networks. In NIPS, 2012. ",
+        "bbox": [
+            174,
+            281,
+            823,
+            308
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Vadim Lebedev, Yaroslav Ganin, Maksim Rakhuba, Ivan Oseledets, and Victor Lempitsky. Speeding-up convolutional neural networks using fine-tuned cp-decomposition. In ICLR, 2015. ",
+        "bbox": [
+            173,
+            316,
+            821,
+            344
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Yann LeCun, John S. Denker, Sara A. Solla, Richard E. Howard, and Lawrence D. Jackel. Optimal brain damage. In NIPS, 1989. ",
+        "bbox": [
+            174,
+            352,
+            823,
+            378
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Yann LeCun, Leon Bottou, Yoshua Bengio, and Patrick Haffner. Gradient-based learning applied to document ´ recognition. Proceedings of the IEEE, 86(11):2278–2324, 1998. ",
+        "bbox": [
+            174,
+            387,
+            823,
+            415
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Hao Li, Asim Kadav, Igor Durdanovic, Hanan Samet, and Hans Peter Graf. Pruning filters for efficient convnets. arXiv preprint arXiv:1608.08710, 2016. ",
+        "bbox": [
+            173,
+            422,
+            823,
+            450
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Baoyuan Liu, Min Wang, Hassan Foroosh, Marshall Tappen, and Marianna Pensky. Sparse convolutional neural networks. In CVPR, 2015. ",
+        "bbox": [
+            173,
+            458,
+            821,
+            486
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Julien Mairal, Francis Bach, Jean Ponce, and Guillermo Sapiro. Online dictionary learning for sparse coding. In ICML, 2009. ",
+        "bbox": [
+            171,
+            494,
+            821,
+            521
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Kenton Murray and David Chiang. Auto-sizing neural networks: With applications to n-gram language models. In EMNLP, 2015. ",
+        "bbox": [
+            173,
+            530,
+            823,
+            558
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Alexander Novikov, Dmitrii Podoprikhin, Anton Osokin, and Dmitry P. Vetrov. Tensorizing neural networks. In NIPS, 2015. ",
+        "bbox": [
+            173,
+            565,
+            821,
+            592
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Ivan V. Oseledets. Tensor-train decomposition. SIAM Journal on Scientific Computing, 33(5):2295–2317, 2011. ",
+        "bbox": [
+            173,
+            601,
+            821,
+            627
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Mohammad Rastegari, Vicente Ordonez, Joseph Redmon, and Ali Farhadi. Xnor-net: Imagenet classification using binary convolutional neural networks. In ECCV, 2016. ",
+        "bbox": [
+            173,
+            636,
+            823,
+            662
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Olga Russakovsky, Jia Deng, Hao Su, Jonathan Krause, Sanjeev Satheesh, Sean Ma, Zhiheng Huang, Andrej Karpathy, Aditya Khosla, Michael Bernstein, et al. Imagenet large scale visual recognition challenge. International Journal of Computer Vision, 115(3):211–252, 2015. ",
+        "bbox": [
+            174,
+            671,
+            821,
+            712
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Karen Simonyan and Andrew Zisserman. Very deep convolutional networks for large-scale image recognition. In ICLR, 2015. ",
+        "bbox": [
+            173,
+            719,
+            823,
+            746
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Nitish Srivastava, Geoffrey E. Hinton, Alex Krizhevsky, Ilya Sutskever, and Ruslan Salakhutdinov. Dropout: a simple way to prevent neural networks from overfitting. Journal of Machine Learning Research, 15(1): 1929–1958, 2014. ",
+        "bbox": [
+            174,
+            755,
+            821,
+            795
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Rupesh K. Srivastava, Klaus Greff, and Jurgen Schmidhuber. Training very deep networks. In ¨ NIPS, 2015. ",
+        "bbox": [
+            171,
+            803,
+            803,
+            818
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Christian Szegedy, Vincent Vanhoucke, Sergey Ioffe, Jonathon Shlens, and Zbigniew Wojna. Rethinking the inception architecture for computer vision. In CVPR, 2016. ",
+        "bbox": [
+            171,
+            827,
+            823,
+            853
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Li Wan, Matthew Zeiler, Sixin Zhang, Yann LeCun, and Rob Fergus. Regularization of neural networks using dropconnect. In ICML, 2013. ",
+        "bbox": [
+            173,
+            862,
+            823,
+            888
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Wei Wen, Chunpeng Wu, Yandan Wang, Yiran Chen, and Hai Li. Learning structured sparsity in deep neural networks. In NIPS, 2016. ",
+        "bbox": [
+            171,
+            897,
+            825,
+            924
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Xiangyu Zhang, Jianhua Zou, Kaiming He, and Jian Sun. Accelerating very deep convolutional networks for classification and detection. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2015a. ",
+        "bbox": [
+            173,
+            104,
+            823,
+            131
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "Xiangyu Zhang, Jianhua Zou, Xiang Ming, Kaiming He, and Jian Sun. Efficient and accurate approximations of nonlinear convolutional networks. In CVPR, 2015b. ",
+        "bbox": [
+            174,
+            140,
+            823,
+            165
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "A ADDITIONAL DETAILS ON DENSE VS. SPARSE CONVOLUTIONS ",
+        "text_level": 1,
+        "bbox": [
+            176,
+            103,
+            722,
+            117
+        ],
+        "page_idx": 12
+    },
+    {
+        "type": "text",
+        "text": "We contrast naive implementations of dense and sparse convolutions (cf. Sect. 3) in Algorithm. 1 and Algorithm 2. We emphasize that we do not use sparse matrices and only introduce sparsity from channel to channel. Thus, walltime will be mostly in terms of Multiply-Adds; the basic operation (convolving the entire image plane in Line 8 of both algorithms) is unchanged. ",
+        "bbox": [
+            174,
+            131,
+            825,
+            188
+        ],
+        "page_idx": 12
+    },
+    {
+        "type": "text",
+        "text": "Algorithm 1 Naive implementation of dense convolution ",
+        "text_level": 1,
+        "bbox": [
+            173,
+            203,
+            549,
+            218
+        ],
+        "page_idx": 12
+    },
+    {
+        "type": "text",
+        "text": "1: Inputs:   \n2: - input: Data tensor   \n3: - W : Parameter tensor   \n4: - input channels: Array of input channel IDs   \n5: - output channels: Array of output channel IDs   \n6: for $i$ in input channels do   \n7: for $o$ in output channels do   \n8: $o u t p u t [ o ] \\gets o u t p u t [ o ] + \\mathrm { c o n v o l v e } ( i n p u t [ i ] , W [ i , o , . . . ] )$   \n9: end for   \n10: end for   \n11: return output ",
+        "bbox": [
+            179,
+            223,
+            629,
+            381
+        ],
+        "page_idx": 12
+    },
+    {
+        "type": "text",
+        "text": "Algorithm 2 Naive implementation of sparse convolution ",
+        "bbox": [
+            174,
+            409,
+            552,
+            422
+        ],
+        "page_idx": 12
+    },
+    {
+        "type": "text",
+        "text": "1: Inputs:   \n2: - input: Data tensor   \n3: - W : Parameter tensor   \n4: - input channels: Array of input channel IDs   \n5: - output channels connected to i: Array of array of output channel IDs specifying connec  \ntions to each input channel   \n6: for $i$ in input channels do   \n7: for index, $o$ in enumerate(output channels connected to i[i]) do   \n8: output[o] $\\gets$ output[o] $^ +$ convolve(input[i], W [i, index, . . .])   \n9: end for   \n10: end for   \n11: return output ",
+        "bbox": [
+            179,
+            429,
+            825,
+            603
+        ],
+        "page_idx": 12
+    }
+]
\ No newline at end of file
diff --git a/parse/train/SkC_7v5gx/SkC_7v5gx_middle.json b/parse/train/SkC_7v5gx/SkC_7v5gx_middle.json
new file mode 100644
index 0000000000000000000000000000000000000000..e81e57a65c534b417525fd51fde9efda52767f5a
--- /dev/null
+++ b/parse/train/SkC_7v5gx/SkC_7v5gx_middle.json
@@ -0,0 +1,32303 @@
+{
+    "pdf_info": [
+        {
+            "preproc_blocks": [
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        79,
+                        396,
+                        116
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                78,
+                                320,
+                                97
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        78,
+                                        320,
+                                        97
+                                    ],
+                                    "score": 1.0,
+                                    "content": "THE POWER OF SPARSITY IN",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                97,
+                                397,
+                                118
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        97,
+                                        397,
+                                        118
+                                    ],
+                                    "score": 1.0,
+                                    "content": "CONVOLUTIONAL NEURAL NETWORKS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        }
+                    ],
+                    "index": 0.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        293,
+                        136,
+                        455,
+                        146
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                291,
+                                133,
+                                456,
+                                150
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        291,
+                                        133,
+                                        456,
+                                        150
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Mark Sandler and Andrey Zhmoginov",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        }
+                    ],
+                    "index": 3
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        112,
+                        135,
+                        248,
+                        190
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                112,
+                                135,
+                                204,
+                                148
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        112,
+                                        135,
+                                        201,
+                                        148
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Soravit Changpinyo",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        201,
+                                        136,
+                                        204,
+                                        141
+                                    ],
+                                    "score": 1.0,
+                                    "content": "∗",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                111,
+                                146,
+                                249,
+                                159
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        111,
+                                        146,
+                                        249,
+                                        159
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Department of Computer Science",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                112,
+                                157,
+                                249,
+                                168
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        112,
+                                        157,
+                                        249,
+                                        168
+                                    ],
+                                    "score": 1.0,
+                                    "content": "University of Southern California",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                112,
+                                168,
+                                236,
+                                180
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        112,
+                                        168,
+                                        236,
+                                        180
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Los Angeles, CA 90020, USA",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                113,
+                                180,
+                                210,
+                                191
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        113,
+                                        180,
+                                        210,
+                                        191
+                                    ],
+                                    "score": 1.0,
+                                    "content": "schangpi@usc.edu",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        }
+                    ],
+                    "index": 6
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        293,
+                        148,
+                        465,
+                        190
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                291,
+                                146,
+                                341,
+                                158
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        291,
+                                        146,
+                                        341,
+                                        158
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Google Inc.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                291,
+                                155,
+                                409,
+                                170
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        291,
+                                        155,
+                                        409,
+                                        170
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1600 Amphitheatre Parkway",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                291,
+                                168,
+                                426,
+                                179
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        291,
+                                        168,
+                                        426,
+                                        179
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Mountain View, CA 94043, USA",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                292,
+                                179,
+                                466,
+                                191
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        292,
+                                        179,
+                                        466,
+                                        191
+                                    ],
+                                    "score": 1.0,
+                                    "content": "{sandler,azhmogin}@google.com",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        }
+                    ],
+                    "index": 8.0
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        278,
+                        219,
+                        333,
+                        231
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                276,
+                                218,
+                                336,
+                                234
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        276,
+                                        218,
+                                        336,
+                                        234
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ABSTRACT",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        }
+                    ],
+                    "index": 12
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        143,
+                        244,
+                        468,
+                        387
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                141,
+                                245,
+                                469,
+                                257
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        245,
+                                        469,
+                                        257
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Deep convolutional networks are well-known for their high computational and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                256,
+                                469,
+                                268
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        256,
+                                        469,
+                                        268
+                                    ],
+                                    "score": 1.0,
+                                    "content": "memory demands. Given limited resources, how does one design a network that",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                266,
+                                469,
+                                279
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        266,
+                                        469,
+                                        279
+                                    ],
+                                    "score": 1.0,
+                                    "content": "balances its size, training time, and prediction accuracy? A surprisingly effective",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                278,
+                                470,
+                                289
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        278,
+                                        470,
+                                        289
+                                    ],
+                                    "score": 1.0,
+                                    "content": "approach to trade accuracy for size and speed is to simply reduce the number of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                288,
+                                470,
+                                301
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        288,
+                                        470,
+                                        301
+                                    ],
+                                    "score": 1.0,
+                                    "content": "channels in each convolutional layer by a fixed fraction and retrain the network. In",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                299,
+                                470,
+                                312
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        299,
+                                        470,
+                                        312
+                                    ],
+                                    "score": 1.0,
+                                    "content": "many cases this leads to significantly smaller networks with only minimal changes",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                310,
+                                470,
+                                324
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        310,
+                                        470,
+                                        324
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to accuracy. In this paper, we take a step further by empirically examining a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                142,
+                                322,
+                                470,
+                                334
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        142,
+                                        322,
+                                        470,
+                                        334
+                                    ],
+                                    "score": 1.0,
+                                    "content": "strategy for deactivating connections between filters in convolutional layers in a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                332,
+                                470,
+                                346
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        332,
+                                        470,
+                                        346
+                                    ],
+                                    "score": 1.0,
+                                    "content": "way that allows us to harvest savings both in run-time and memory for many",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                343,
+                                470,
+                                356
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        343,
+                                        470,
+                                        356
+                                    ],
+                                    "score": 1.0,
+                                    "content": "network architectures. More specifically, we generalize 2D convolution to use a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                354,
+                                469,
+                                367
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        354,
+                                        469,
+                                        367
+                                    ],
+                                    "score": 1.0,
+                                    "content": "channel-wise sparse connection structure and show that this leads to significantly",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                365,
+                                470,
+                                378
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        365,
+                                        470,
+                                        378
+                                    ],
+                                    "score": 1.0,
+                                    "content": "better results than the baseline approach for large networks including VGG and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                375,
+                                200,
+                                389
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        375,
+                                        200,
+                                        389
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Inception V3.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        }
+                    ],
+                    "index": 19
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        404,
+                        206,
+                        416
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                403,
+                                208,
+                                420
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        403,
+                                        208,
+                                        420
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1 INTRODUCTION",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        }
+                    ],
+                    "index": 26
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        428,
+                        505,
+                        559
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                428,
+                                505,
+                                440
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        428,
+                                        505,
+                                        440
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Deep neural networks combined with large-scale labeled data have become a standard recipe for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                439,
+                                505,
+                                451
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        439,
+                                        505,
+                                        451
+                                    ],
+                                    "score": 1.0,
+                                    "content": "achieving state-of-the-art performance on supervised learning tasks in recent years. Despite of their",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                449,
+                                505,
+                                462
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        449,
+                                        505,
+                                        462
+                                    ],
+                                    "score": 1.0,
+                                    "content": "success, the capability of deep neural networks to model highly nonlinear functions comes with high",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                461,
+                                505,
+                                473
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        461,
+                                        505,
+                                        473
+                                    ],
+                                    "score": 1.0,
+                                    "content": "computational and memory demands both during the model training and inference. In particular, the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                472,
+                                505,
+                                484
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        472,
+                                        505,
+                                        484
+                                    ],
+                                    "score": 1.0,
+                                    "content": "number of parameters of neural network models is often designed to be huge to account for the scale,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                482,
+                                506,
+                                495
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        482,
+                                        506,
+                                        495
+                                    ],
+                                    "score": 1.0,
+                                    "content": "diversity, and complexity of data that they learn from. While advances in hardware have somewhat",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                494,
+                                506,
+                                505
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        494,
+                                        506,
+                                        505
+                                    ],
+                                    "score": 1.0,
+                                    "content": "alleviated the issue, network size, speed, and power consumption are all limiting factors when it",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                505,
+                                505,
+                                516
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        505,
+                                        505,
+                                        516
+                                    ],
+                                    "score": 1.0,
+                                    "content": "comes to production deployment on mobile and embedded devices. On the other hand, it is well-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                516,
+                                505,
+                                528
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        516,
+                                        505,
+                                        528
+                                    ],
+                                    "score": 1.0,
+                                    "content": "known that there is significant redundancy among the weights of neural networks. For example,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                526,
+                                505,
+                                538
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        526,
+                                        359,
+                                        538
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Denil et al. (2013) show that it is possible to learn less than",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        360,
+                                        527,
+                                        375,
+                                        537
+                                    ],
+                                    "score": 0.86,
+                                    "content": "5 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        375,
+                                        526,
+                                        505,
+                                        538
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of the network parameters and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                538,
+                                505,
+                                549
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        538,
+                                        505,
+                                        549
+                                    ],
+                                    "score": 1.0,
+                                    "content": "predict the rest without losing predictive accuracy. This evidence suggests that neural networks are",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                549,
+                                211,
+                                561
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        549,
+                                        211,
+                                        561
+                                    ],
+                                    "score": 1.0,
+                                    "content": "often over-parameterized.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        }
+                    ],
+                    "index": 32.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        565,
+                        505,
+                        653
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                565,
+                                505,
+                                578
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        565,
+                                        505,
+                                        578
+                                    ],
+                                    "score": 1.0,
+                                    "content": "These motivate the research on neural network compression. However, several immediate ques-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                576,
+                                505,
+                                588
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        576,
+                                        447,
+                                        588
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tions arise: Are these parameters easy to identify? Could we just make the network",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        447,
+                                        576,
+                                        462,
+                                        587
+                                    ],
+                                    "score": 0.85,
+                                    "content": "5 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        462,
+                                        576,
+                                        505,
+                                        588
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of its size",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                588,
+                                505,
+                                599
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        588,
+                                        505,
+                                        599
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and retrain? Or are more advanced methods required? There is an extensive literature in the last",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                597,
+                                506,
+                                612
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        597,
+                                        506,
+                                        612
+                                    ],
+                                    "score": 1.0,
+                                    "content": "few years that explores the question of network compression using advanced techniques, including",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                609,
+                                506,
+                                623
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        609,
+                                        506,
+                                        623
+                                    ],
+                                    "score": 1.0,
+                                    "content": "network prunning, loss-based compression, quantization, and matrix decomposition. We overview",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                621,
+                                505,
+                                632
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        621,
+                                        505,
+                                        632
+                                    ],
+                                    "score": 1.0,
+                                    "content": "many of these directions in the next section. However, there is surprisingly little research on whether",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                631,
+                                506,
+                                644
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        631,
+                                        506,
+                                        644
+                                    ],
+                                    "score": 1.0,
+                                    "content": "this over-parameterization can simply be re-captured by more efficient architectures that could be",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                642,
+                                363,
+                                654
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        642,
+                                        363,
+                                        654
+                                    ],
+                                    "score": 1.0,
+                                    "content": "obtained from original architectures via simple transformations.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        }
+                    ],
+                    "index": 42.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        659,
+                        504,
+                        714
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                659,
+                                505,
+                                671
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        659,
+                                        505,
+                                        671
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Our approach is inspired by a very simple yet successful method called depth multiplier (Howard,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 47
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                669,
+                                505,
+                                683
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        669,
+                                        505,
+                                        683
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2017). In this method the depth (the number of channels) of each convolutional layer in a given",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 48
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                679,
+                                506,
+                                695
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        679,
+                                        506,
+                                        695
+                                    ],
+                                    "score": 1.0,
+                                    "content": "network is simply reduced by a fixed fraction and the network is retrained. We generalize this ap-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 49
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                691,
+                                506,
+                                705
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        691,
+                                        506,
+                                        705
+                                    ],
+                                    "score": 1.0,
+                                    "content": "proach by removing the constraint that every input filter (or channel) must be fully connected to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 50
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                703,
+                                505,
+                                716
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        703,
+                                        505,
+                                        716
+                                    ],
+                                    "score": 1.0,
+                                    "content": "every output filter. Instead, we use a sparse connection matrix, where each output convolution chan-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 51
+                        }
+                    ],
+                    "index": 49
+                }
+            ],
+            "page_idx": 0,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        117,
+                        721,
+                        417,
+                        731
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                118,
+                                719,
+                                418,
+                                734
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        118,
+                                        719,
+                                        418,
+                                        734
+                                    ],
+                                    "score": 1.0,
+                                    "content": "∗The work was done while the author was doing an internship at Google Research.",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        107,
+                        27,
+                        308,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                25,
+                                308,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        25,
+                                        308,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Under review as a conference paper at ICLR 2017",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        303,
+                        752,
+                        308,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                302,
+                                751,
+                                309,
+                                762
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        302,
+                                        751,
+                                        309,
+                                        762
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        79,
+                        396,
+                        116
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                78,
+                                320,
+                                97
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        78,
+                                        320,
+                                        97
+                                    ],
+                                    "score": 1.0,
+                                    "content": "THE POWER OF SPARSITY IN",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                97,
+                                397,
+                                118
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        97,
+                                        397,
+                                        118
+                                    ],
+                                    "score": 1.0,
+                                    "content": "CONVOLUTIONAL NEURAL NETWORKS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        }
+                    ],
+                    "index": 0.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        293,
+                        136,
+                        455,
+                        146
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                291,
+                                133,
+                                456,
+                                150
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        291,
+                                        133,
+                                        456,
+                                        150
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Mark Sandler and Andrey Zhmoginov",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        }
+                    ],
+                    "index": 3,
+                    "bbox_fs": [
+                        291,
+                        133,
+                        456,
+                        150
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        112,
+                        135,
+                        248,
+                        190
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                112,
+                                135,
+                                204,
+                                148
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        112,
+                                        135,
+                                        201,
+                                        148
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Soravit Changpinyo",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        201,
+                                        136,
+                                        204,
+                                        141
+                                    ],
+                                    "score": 1.0,
+                                    "content": "∗",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                111,
+                                146,
+                                249,
+                                159
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        111,
+                                        146,
+                                        249,
+                                        159
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Department of Computer Science",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                112,
+                                157,
+                                249,
+                                168
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        112,
+                                        157,
+                                        249,
+                                        168
+                                    ],
+                                    "score": 1.0,
+                                    "content": "University of Southern California",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                112,
+                                168,
+                                236,
+                                180
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        112,
+                                        168,
+                                        236,
+                                        180
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Los Angeles, CA 90020, USA",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                113,
+                                180,
+                                210,
+                                191
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        113,
+                                        180,
+                                        210,
+                                        191
+                                    ],
+                                    "score": 1.0,
+                                    "content": "schangpi@usc.edu",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        }
+                    ],
+                    "index": 6,
+                    "bbox_fs": [
+                        111,
+                        135,
+                        249,
+                        191
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        293,
+                        148,
+                        465,
+                        190
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                291,
+                                146,
+                                341,
+                                158
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        291,
+                                        146,
+                                        341,
+                                        158
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Google Inc.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                291,
+                                155,
+                                409,
+                                170
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        291,
+                                        155,
+                                        409,
+                                        170
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1600 Amphitheatre Parkway",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                291,
+                                168,
+                                426,
+                                179
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        291,
+                                        168,
+                                        426,
+                                        179
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Mountain View, CA 94043, USA",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                292,
+                                179,
+                                466,
+                                191
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        292,
+                                        179,
+                                        466,
+                                        191
+                                    ],
+                                    "score": 1.0,
+                                    "content": "{sandler,azhmogin}@google.com",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        }
+                    ],
+                    "index": 8.0,
+                    "bbox_fs": [
+                        291,
+                        146,
+                        466,
+                        191
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        278,
+                        219,
+                        333,
+                        231
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                276,
+                                218,
+                                336,
+                                234
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        276,
+                                        218,
+                                        336,
+                                        234
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ABSTRACT",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        }
+                    ],
+                    "index": 12
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        143,
+                        244,
+                        468,
+                        387
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                141,
+                                245,
+                                469,
+                                257
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        245,
+                                        469,
+                                        257
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Deep convolutional networks are well-known for their high computational and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                256,
+                                469,
+                                268
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        256,
+                                        469,
+                                        268
+                                    ],
+                                    "score": 1.0,
+                                    "content": "memory demands. Given limited resources, how does one design a network that",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                266,
+                                469,
+                                279
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        266,
+                                        469,
+                                        279
+                                    ],
+                                    "score": 1.0,
+                                    "content": "balances its size, training time, and prediction accuracy? A surprisingly effective",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                278,
+                                470,
+                                289
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        278,
+                                        470,
+                                        289
+                                    ],
+                                    "score": 1.0,
+                                    "content": "approach to trade accuracy for size and speed is to simply reduce the number of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                288,
+                                470,
+                                301
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        288,
+                                        470,
+                                        301
+                                    ],
+                                    "score": 1.0,
+                                    "content": "channels in each convolutional layer by a fixed fraction and retrain the network. In",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                299,
+                                470,
+                                312
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        299,
+                                        470,
+                                        312
+                                    ],
+                                    "score": 1.0,
+                                    "content": "many cases this leads to significantly smaller networks with only minimal changes",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                310,
+                                470,
+                                324
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        310,
+                                        470,
+                                        324
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to accuracy. In this paper, we take a step further by empirically examining a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                142,
+                                322,
+                                470,
+                                334
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        142,
+                                        322,
+                                        470,
+                                        334
+                                    ],
+                                    "score": 1.0,
+                                    "content": "strategy for deactivating connections between filters in convolutional layers in a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                332,
+                                470,
+                                346
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        332,
+                                        470,
+                                        346
+                                    ],
+                                    "score": 1.0,
+                                    "content": "way that allows us to harvest savings both in run-time and memory for many",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                343,
+                                470,
+                                356
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        343,
+                                        470,
+                                        356
+                                    ],
+                                    "score": 1.0,
+                                    "content": "network architectures. More specifically, we generalize 2D convolution to use a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                354,
+                                469,
+                                367
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        354,
+                                        469,
+                                        367
+                                    ],
+                                    "score": 1.0,
+                                    "content": "channel-wise sparse connection structure and show that this leads to significantly",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                365,
+                                470,
+                                378
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        365,
+                                        470,
+                                        378
+                                    ],
+                                    "score": 1.0,
+                                    "content": "better results than the baseline approach for large networks including VGG and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                375,
+                                200,
+                                389
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        375,
+                                        200,
+                                        389
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Inception V3.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        }
+                    ],
+                    "index": 19,
+                    "bbox_fs": [
+                        141,
+                        245,
+                        470,
+                        389
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        404,
+                        206,
+                        416
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                403,
+                                208,
+                                420
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        403,
+                                        208,
+                                        420
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1 INTRODUCTION",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        }
+                    ],
+                    "index": 26
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        428,
+                        505,
+                        559
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                428,
+                                505,
+                                440
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        428,
+                                        505,
+                                        440
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Deep neural networks combined with large-scale labeled data have become a standard recipe for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                439,
+                                505,
+                                451
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        439,
+                                        505,
+                                        451
+                                    ],
+                                    "score": 1.0,
+                                    "content": "achieving state-of-the-art performance on supervised learning tasks in recent years. Despite of their",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                449,
+                                505,
+                                462
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        449,
+                                        505,
+                                        462
+                                    ],
+                                    "score": 1.0,
+                                    "content": "success, the capability of deep neural networks to model highly nonlinear functions comes with high",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                461,
+                                505,
+                                473
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        461,
+                                        505,
+                                        473
+                                    ],
+                                    "score": 1.0,
+                                    "content": "computational and memory demands both during the model training and inference. In particular, the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                472,
+                                505,
+                                484
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        472,
+                                        505,
+                                        484
+                                    ],
+                                    "score": 1.0,
+                                    "content": "number of parameters of neural network models is often designed to be huge to account for the scale,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                482,
+                                506,
+                                495
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        482,
+                                        506,
+                                        495
+                                    ],
+                                    "score": 1.0,
+                                    "content": "diversity, and complexity of data that they learn from. While advances in hardware have somewhat",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                494,
+                                506,
+                                505
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        494,
+                                        506,
+                                        505
+                                    ],
+                                    "score": 1.0,
+                                    "content": "alleviated the issue, network size, speed, and power consumption are all limiting factors when it",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                505,
+                                505,
+                                516
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        505,
+                                        505,
+                                        516
+                                    ],
+                                    "score": 1.0,
+                                    "content": "comes to production deployment on mobile and embedded devices. On the other hand, it is well-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                516,
+                                505,
+                                528
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        516,
+                                        505,
+                                        528
+                                    ],
+                                    "score": 1.0,
+                                    "content": "known that there is significant redundancy among the weights of neural networks. For example,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                526,
+                                505,
+                                538
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        526,
+                                        359,
+                                        538
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Denil et al. (2013) show that it is possible to learn less than",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        360,
+                                        527,
+                                        375,
+                                        537
+                                    ],
+                                    "score": 0.86,
+                                    "content": "5 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        375,
+                                        526,
+                                        505,
+                                        538
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of the network parameters and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                538,
+                                505,
+                                549
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        538,
+                                        505,
+                                        549
+                                    ],
+                                    "score": 1.0,
+                                    "content": "predict the rest without losing predictive accuracy. This evidence suggests that neural networks are",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                549,
+                                211,
+                                561
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        549,
+                                        211,
+                                        561
+                                    ],
+                                    "score": 1.0,
+                                    "content": "often over-parameterized.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        }
+                    ],
+                    "index": 32.5,
+                    "bbox_fs": [
+                        105,
+                        428,
+                        506,
+                        561
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        565,
+                        505,
+                        653
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                565,
+                                505,
+                                578
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        565,
+                                        505,
+                                        578
+                                    ],
+                                    "score": 1.0,
+                                    "content": "These motivate the research on neural network compression. However, several immediate ques-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                576,
+                                505,
+                                588
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        576,
+                                        447,
+                                        588
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tions arise: Are these parameters easy to identify? Could we just make the network",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        447,
+                                        576,
+                                        462,
+                                        587
+                                    ],
+                                    "score": 0.85,
+                                    "content": "5 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        462,
+                                        576,
+                                        505,
+                                        588
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of its size",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                588,
+                                505,
+                                599
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        588,
+                                        505,
+                                        599
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and retrain? Or are more advanced methods required? There is an extensive literature in the last",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                597,
+                                506,
+                                612
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        597,
+                                        506,
+                                        612
+                                    ],
+                                    "score": 1.0,
+                                    "content": "few years that explores the question of network compression using advanced techniques, including",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                609,
+                                506,
+                                623
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        609,
+                                        506,
+                                        623
+                                    ],
+                                    "score": 1.0,
+                                    "content": "network prunning, loss-based compression, quantization, and matrix decomposition. We overview",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                621,
+                                505,
+                                632
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        621,
+                                        505,
+                                        632
+                                    ],
+                                    "score": 1.0,
+                                    "content": "many of these directions in the next section. However, there is surprisingly little research on whether",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                631,
+                                506,
+                                644
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        631,
+                                        506,
+                                        644
+                                    ],
+                                    "score": 1.0,
+                                    "content": "this over-parameterization can simply be re-captured by more efficient architectures that could be",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                642,
+                                363,
+                                654
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        642,
+                                        363,
+                                        654
+                                    ],
+                                    "score": 1.0,
+                                    "content": "obtained from original architectures via simple transformations.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        }
+                    ],
+                    "index": 42.5,
+                    "bbox_fs": [
+                        105,
+                        565,
+                        506,
+                        654
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        659,
+                        504,
+                        714
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                659,
+                                505,
+                                671
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        659,
+                                        505,
+                                        671
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Our approach is inspired by a very simple yet successful method called depth multiplier (Howard,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 47
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                669,
+                                505,
+                                683
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        669,
+                                        505,
+                                        683
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2017). In this method the depth (the number of channels) of each convolutional layer in a given",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 48
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                679,
+                                506,
+                                695
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        679,
+                                        506,
+                                        695
+                                    ],
+                                    "score": 1.0,
+                                    "content": "network is simply reduced by a fixed fraction and the network is retrained. We generalize this ap-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 49
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                691,
+                                506,
+                                705
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        691,
+                                        506,
+                                        705
+                                    ],
+                                    "score": 1.0,
+                                    "content": "proach by removing the constraint that every input filter (or channel) must be fully connected to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 50
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                703,
+                                505,
+                                716
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        703,
+                                        505,
+                                        716
+                                    ],
+                                    "score": 1.0,
+                                    "content": "every output filter. Instead, we use a sparse connection matrix, where each output convolution chan-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 51
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                83,
+                                505,
+                                93
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        83,
+                                        505,
+                                        93
+                                    ],
+                                    "score": 1.0,
+                                    "content": "nel is connected only to a small random fraction of the input channels. Note that, for convolutional",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                93,
+                                505,
+                                106
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        93,
+                                        505,
+                                        106
+                                    ],
+                                    "score": 1.0,
+                                    "content": "networks, this still allows for efficient computation since the one channel spatial convolution across",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                105,
+                                252,
+                                117
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        105,
+                                        252,
+                                        117
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the entire plane remains unchanged.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 2
+                        }
+                    ],
+                    "index": 49,
+                    "bbox_fs": [
+                        104,
+                        659,
+                        506,
+                        716
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        82,
+                        504,
+                        116
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                83,
+                                505,
+                                93
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        83,
+                                        505,
+                                        93
+                                    ],
+                                    "score": 1.0,
+                                    "content": "nel is connected only to a small random fraction of the input channels. Note that, for convolutional",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                93,
+                                505,
+                                106
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        93,
+                                        505,
+                                        106
+                                    ],
+                                    "score": 1.0,
+                                    "content": "networks, this still allows for efficient computation since the one channel spatial convolution across",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                105,
+                                252,
+                                117
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        105,
+                                        252,
+                                        117
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the entire plane remains unchanged.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        }
+                    ],
+                    "index": 1
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        121,
+                        503,
+                        155
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                121,
+                                505,
+                                135
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        121,
+                                        505,
+                                        135
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We empirically demonstrate the effectiveness of our approach on four networks (MNIST, CIFAR",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                132,
+                                505,
+                                145
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        132,
+                                        505,
+                                        145
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Net, Inception-V3 and VGG-16) of different sizes. Our results suggest that our approach outper-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                143,
+                                404,
+                                156
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        143,
+                                        404,
+                                        156
+                                    ],
+                                    "score": 1.0,
+                                    "content": "forms dense convolutions with depth multiplier at high compression rates.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        }
+                    ],
+                    "index": 4
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        160,
+                        505,
+                        258
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                160,
+                                505,
+                                173
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        160,
+                                        505,
+                                        173
+                                    ],
+                                    "score": 1.0,
+                                    "content": "For Inception V3 (Szegedy et al., 2016), we show that we can train a network with only about 300K",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                171,
+                                504,
+                                183
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        171,
+                                        432,
+                                        183
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of convolutional parameters1 and about 100M multiply-adds that achieves above",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        432,
+                                        171,
+                                        452,
+                                        182
+                                    ],
+                                    "score": 0.87,
+                                    "content": "52 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        452,
+                                        171,
+                                        504,
+                                        183
+                                    ],
+                                    "score": 1.0,
+                                    "content": "accuracy af-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                181,
+                                505,
+                                195
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        181,
+                                        443,
+                                        195
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ter it is fully trained. The corresponding depth-multiplier network has only about",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        443,
+                                        182,
+                                        464,
+                                        193
+                                    ],
+                                    "score": 0.85,
+                                    "content": "41 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        464,
+                                        181,
+                                        505,
+                                        195
+                                    ],
+                                    "score": 1.0,
+                                    "content": "accuracy.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                193,
+                                505,
+                                206
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        193,
+                                        505,
+                                        206
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Another network that we consider is VGG-16n, a slightly modified version of VGG-16 (Simonyan",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                204,
+                                506,
+                                217
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        204,
+                                        212,
+                                        217
+                                    ],
+                                    "score": 1.0,
+                                    "content": "& Zisserman, 2015), with",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        213,
+                                        204,
+                                        225,
+                                        214
+                                    ],
+                                    "score": 0.68,
+                                    "content": "7 \\mathbf { x }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        225,
+                                        204,
+                                        506,
+                                        217
+                                    ],
+                                    "score": 1.0,
+                                    "content": "fewer parameters and similar accuracy.2 We found VGG-16n to start",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                216,
+                                505,
+                                227
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        216,
+                                        505,
+                                        227
+                                    ],
+                                    "score": 1.0,
+                                    "content": "training much faster than the original VGG-16 which was trained incrementally in the original lit-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                227,
+                                505,
+                                239
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        227,
+                                        505,
+                                        239
+                                    ],
+                                    "score": 1.0,
+                                    "content": "erature. We explore the impact of sparsification and the number of parameters on the quality of the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                236,
+                                506,
+                                251
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        236,
+                                        267,
+                                        251
+                                    ],
+                                    "score": 1.0,
+                                    "content": "network by building the networks up to",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        267,
+                                        237,
+                                        284,
+                                        248
+                                    ],
+                                    "score": 0.5,
+                                    "content": "3 0 \\mathrm { x }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        284,
+                                        236,
+                                        382,
+                                        251
+                                    ],
+                                    "score": 1.0,
+                                    "content": "smaller than VGG-16n",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        383,
+                                        237,
+                                        404,
+                                        248
+                                    ],
+                                    "score": 0.48,
+                                    "content": "2 0 0 \\mathrm { x }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        404,
+                                        236,
+                                        506,
+                                        251
+                                    ],
+                                    "score": 1.0,
+                                    "content": "smaller than the original",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                246,
+                                150,
+                                261
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        246,
+                                        150,
+                                        261
+                                    ],
+                                    "score": 1.0,
+                                    "content": "VGG-16).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        }
+                    ],
+                    "index": 10
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        265,
+                        505,
+                        309
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                264,
+                                505,
+                                278
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        264,
+                                        505,
+                                        278
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In terms of model flexibility, sparse connections allow for an incremental training approach, where",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                276,
+                                505,
+                                288
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        276,
+                                        505,
+                                        288
+                                    ],
+                                    "score": 1.0,
+                                    "content": "connection structure between layers can be densified as training progresses. More importantly, the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                286,
+                                506,
+                                299
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        286,
+                                        506,
+                                        299
+                                    ],
+                                    "score": 1.0,
+                                    "content": "incremental training approach can potentially speed up the training significantly due to savings in",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                297,
+                                217,
+                                311
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        297,
+                                        217,
+                                        311
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the early stages of training.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        }
+                    ],
+                    "index": 16.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        315,
+                        505,
+                        347
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                315,
+                                505,
+                                327
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        315,
+                                        505,
+                                        327
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The rest of the paper is organized as follows. Section 2 summarizes relevant work. We describe",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                326,
+                                505,
+                                338
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        326,
+                                        505,
+                                        338
+                                    ],
+                                    "score": 1.0,
+                                    "content": "our approach in Section 3 and then present some intuition in Section 4. Finally, we show our",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                337,
+                                243,
+                                348
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        337,
+                                        243,
+                                        348
+                                    ],
+                                    "score": 1.0,
+                                    "content": "experimental results in Section 5.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        }
+                    ],
+                    "index": 20
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        360,
+                        211,
+                        372
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                358,
+                                213,
+                                375
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        358,
+                                        213,
+                                        375
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2 RELATED WORK",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        }
+                    ],
+                    "index": 22
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        385,
+                        360,
+                        396
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                384,
+                                361,
+                                397
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        384,
+                                        361,
+                                        397
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2.1 COMPRESSION TECHNIQUES FOR NEURAL NETWORKS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        }
+                    ],
+                    "index": 23
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        406,
+                        505,
+                        461
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                405,
+                                505,
+                                419
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        405,
+                                        505,
+                                        419
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Our work is closely related to a compression technique based on network pruning. However, the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                417,
+                                505,
+                                429
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        417,
+                                        505,
+                                        429
+                                    ],
+                                    "score": 1.0,
+                                    "content": "important difference is that we do not try to select the connections which are redundant. Instead,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                428,
+                                505,
+                                440
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        428,
+                                        505,
+                                        440
+                                    ],
+                                    "score": 1.0,
+                                    "content": "we just fix a random connectivity pattern and let the network train around it. We also give a brief",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                439,
+                                505,
+                                451
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        439,
+                                        505,
+                                        451
+                                    ],
+                                    "score": 1.0,
+                                    "content": "overview of other two popular techniques: quantization and decomposition, though these directions",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                450,
+                                366,
+                                462
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        450,
+                                        366,
+                                        462
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are not the main focus and could be complementary to our work.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        }
+                    ],
+                    "index": 26
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        471,
+                        505,
+                        537
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                471,
+                                505,
+                                484
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        471,
+                                        505,
+                                        484
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Network pruning Much initial work on neural network compression focuses on removing unim-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                483,
+                                505,
+                                495
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        483,
+                                        505,
+                                        495
+                                    ],
+                                    "score": 1.0,
+                                    "content": "portant connections using weight decay. Hanson & Pratt (1989) introduce hyperbolic and exponen-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                493,
+                                505,
+                                506
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        493,
+                                        505,
+                                        506
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tial biases to the objective. Optimal Brain Damage (LeCun et al., 1989) and Optimal Brain Surgeon",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                504,
+                                505,
+                                517
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        504,
+                                        505,
+                                        517
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(Hassibi & Stork, 1993) prune the networks based on second-order derivatives of the objectives.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                515,
+                                506,
+                                529
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        515,
+                                        506,
+                                        529
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Recent work by Han et al. (2015; 2016a) alternates between pruning near-zero weights, which are",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                525,
+                                406,
+                                539
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        525,
+                                        167,
+                                        539
+                                    ],
+                                    "score": 1.0,
+                                    "content": "encouraged by",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        168,
+                                        527,
+                                        178,
+                                        536
+                                    ],
+                                    "score": 0.26,
+                                    "content": "\\ell { 1 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        178,
+                                        525,
+                                        189,
+                                        539
+                                    ],
+                                    "score": 1.0,
+                                    "content": "or",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        190,
+                                        526,
+                                        200,
+                                        536
+                                    ],
+                                    "score": 0.62,
+                                    "content": "\\ell 2",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        201,
+                                        525,
+                                        406,
+                                        539
+                                    ],
+                                    "score": 1.0,
+                                    "content": "regularization, and retraining the pruned networks.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        }
+                    ],
+                    "index": 31.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        542,
+                        505,
+                        642
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                543,
+                                505,
+                                555
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        543,
+                                        505,
+                                        555
+                                    ],
+                                    "score": 1.0,
+                                    "content": "More complex regularizers have also been considered. Wen et al. (2016) and Li et al. (2016) put",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                553,
+                                505,
+                                567
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        553,
+                                        505,
+                                        567
+                                    ],
+                                    "score": 1.0,
+                                    "content": "structured sparsity regularizers on the weights, while Murray & Chiang (2015) put them on the hid-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                565,
+                                505,
+                                577
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        565,
+                                        505,
+                                        577
+                                    ],
+                                    "score": 1.0,
+                                    "content": "den units. Feng & Darrell (2015) explore a nonparametric prior based on the Indian buffet processes",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                576,
+                                506,
+                                588
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        576,
+                                        506,
+                                        588
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(Griffiths & Ghahramani, 2011) on layers. Hu et al. (2016) prune neurons based on the analysis of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                586,
+                                506,
+                                601
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        586,
+                                        506,
+                                        601
+                                    ],
+                                    "score": 1.0,
+                                    "content": "their outputs on a large dataset. Anwar et al. (2015b) consider special sparsity patterns: channel-wise",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                597,
+                                506,
+                                612
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        597,
+                                        506,
+                                        612
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(removing a feature map/channel from a layer), kernel-wise (removing all connections between two",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                609,
+                                505,
+                                621
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        609,
+                                        505,
+                                        621
+                                    ],
+                                    "score": 1.0,
+                                    "content": "feature maps in consecutive layers), and intra-kernel-strided (removing connections between two",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                619,
+                                506,
+                                633
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        619,
+                                        506,
+                                        633
+                                    ],
+                                    "score": 1.0,
+                                    "content": "features with particular stride and offset). They also propose to use particle filter to decide the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                630,
+                                320,
+                                644
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        630,
+                                        320,
+                                        644
+                                    ],
+                                    "score": 1.0,
+                                    "content": "importance of connections and paths during training.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        }
+                    ],
+                    "index": 39
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        648,
+                        503,
+                        681
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                647,
+                                505,
+                                659
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        647,
+                                        505,
+                                        659
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Another line of work explores fixed network architectures with some subsets of connections re-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                659,
+                                505,
+                                671
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        659,
+                                        505,
+                                        671
+                                    ],
+                                    "score": 1.0,
+                                    "content": "moved. For example, LeCun et al. (1998) remove connections between the first two convolutional",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                669,
+                                505,
+                                683
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        669,
+                                        505,
+                                        683
+                                    ],
+                                    "score": 1.0,
+                                    "content": "feature maps in a completely uniform manner. This is similar to our approach but they only con-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        }
+                    ],
+                    "index": 45
+                }
+            ],
+            "page_idx": 1,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        106,
+                        690,
+                        505,
+                        731
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                119,
+                                689,
+                                505,
+                                703
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        119,
+                                        689,
+                                        505,
+                                        703
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1Here and elsewhere, we ignore the parameters for the softmax classifier since they simply describe a linear",
+                                    "type": "text"
+                                }
+                            ]
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                700,
+                                285,
+                                712
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        700,
+                                        285,
+                                        712
+                                    ],
+                                    "score": 1.0,
+                                    "content": "transformation and depend on number of classes.",
+                                    "type": "text"
+                                }
+                            ]
+                        },
+                        {
+                            "bbox": [
+                                117,
+                                708,
+                                506,
+                                725
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        117,
+                                        708,
+                                        506,
+                                        725
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2As of this writing, VGG networks have not finished training, but the training trajectory suggest similar",
+                                    "type": "text"
+                                }
+                            ]
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                721,
+                                156,
+                                732
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        721,
+                                        156,
+                                        732
+                                    ],
+                                    "score": 1.0,
+                                    "content": "performance.",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        107,
+                        27,
+                        308,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                107,
+                                26,
+                                308,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        26,
+                                        308,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Under review as a conference paper at ICLR 2017",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        302,
+                        751,
+                        309,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                301,
+                                750,
+                                310,
+                                763
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        301,
+                                        750,
+                                        310,
+                                        763
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        82,
+                        504,
+                        116
+                    ],
+                    "lines": [],
+                    "index": 1,
+                    "bbox_fs": [
+                        105,
+                        83,
+                        505,
+                        117
+                    ],
+                    "lines_deleted": true
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        121,
+                        503,
+                        155
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                121,
+                                505,
+                                135
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        121,
+                                        505,
+                                        135
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We empirically demonstrate the effectiveness of our approach on four networks (MNIST, CIFAR",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                132,
+                                505,
+                                145
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        132,
+                                        505,
+                                        145
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Net, Inception-V3 and VGG-16) of different sizes. Our results suggest that our approach outper-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                143,
+                                404,
+                                156
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        143,
+                                        404,
+                                        156
+                                    ],
+                                    "score": 1.0,
+                                    "content": "forms dense convolutions with depth multiplier at high compression rates.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        }
+                    ],
+                    "index": 4,
+                    "bbox_fs": [
+                        105,
+                        121,
+                        505,
+                        156
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        160,
+                        505,
+                        258
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                160,
+                                505,
+                                173
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        160,
+                                        505,
+                                        173
+                                    ],
+                                    "score": 1.0,
+                                    "content": "For Inception V3 (Szegedy et al., 2016), we show that we can train a network with only about 300K",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                171,
+                                504,
+                                183
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        171,
+                                        432,
+                                        183
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of convolutional parameters1 and about 100M multiply-adds that achieves above",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        432,
+                                        171,
+                                        452,
+                                        182
+                                    ],
+                                    "score": 0.87,
+                                    "content": "52 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        452,
+                                        171,
+                                        504,
+                                        183
+                                    ],
+                                    "score": 1.0,
+                                    "content": "accuracy af-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                181,
+                                505,
+                                195
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        181,
+                                        443,
+                                        195
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ter it is fully trained. The corresponding depth-multiplier network has only about",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        443,
+                                        182,
+                                        464,
+                                        193
+                                    ],
+                                    "score": 0.85,
+                                    "content": "41 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        464,
+                                        181,
+                                        505,
+                                        195
+                                    ],
+                                    "score": 1.0,
+                                    "content": "accuracy.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                193,
+                                505,
+                                206
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        193,
+                                        505,
+                                        206
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Another network that we consider is VGG-16n, a slightly modified version of VGG-16 (Simonyan",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                204,
+                                506,
+                                217
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        204,
+                                        212,
+                                        217
+                                    ],
+                                    "score": 1.0,
+                                    "content": "& Zisserman, 2015), with",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        213,
+                                        204,
+                                        225,
+                                        214
+                                    ],
+                                    "score": 0.68,
+                                    "content": "7 \\mathbf { x }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        225,
+                                        204,
+                                        506,
+                                        217
+                                    ],
+                                    "score": 1.0,
+                                    "content": "fewer parameters and similar accuracy.2 We found VGG-16n to start",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                216,
+                                505,
+                                227
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        216,
+                                        505,
+                                        227
+                                    ],
+                                    "score": 1.0,
+                                    "content": "training much faster than the original VGG-16 which was trained incrementally in the original lit-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                227,
+                                505,
+                                239
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        227,
+                                        505,
+                                        239
+                                    ],
+                                    "score": 1.0,
+                                    "content": "erature. We explore the impact of sparsification and the number of parameters on the quality of the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                236,
+                                506,
+                                251
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        236,
+                                        267,
+                                        251
+                                    ],
+                                    "score": 1.0,
+                                    "content": "network by building the networks up to",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        267,
+                                        237,
+                                        284,
+                                        248
+                                    ],
+                                    "score": 0.5,
+                                    "content": "3 0 \\mathrm { x }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        284,
+                                        236,
+                                        382,
+                                        251
+                                    ],
+                                    "score": 1.0,
+                                    "content": "smaller than VGG-16n",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        383,
+                                        237,
+                                        404,
+                                        248
+                                    ],
+                                    "score": 0.48,
+                                    "content": "2 0 0 \\mathrm { x }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        404,
+                                        236,
+                                        506,
+                                        251
+                                    ],
+                                    "score": 1.0,
+                                    "content": "smaller than the original",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                246,
+                                150,
+                                261
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        246,
+                                        150,
+                                        261
+                                    ],
+                                    "score": 1.0,
+                                    "content": "VGG-16).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        }
+                    ],
+                    "index": 10,
+                    "bbox_fs": [
+                        104,
+                        160,
+                        506,
+                        261
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        265,
+                        505,
+                        309
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                264,
+                                505,
+                                278
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        264,
+                                        505,
+                                        278
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In terms of model flexibility, sparse connections allow for an incremental training approach, where",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                276,
+                                505,
+                                288
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        276,
+                                        505,
+                                        288
+                                    ],
+                                    "score": 1.0,
+                                    "content": "connection structure between layers can be densified as training progresses. More importantly, the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                286,
+                                506,
+                                299
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        286,
+                                        506,
+                                        299
+                                    ],
+                                    "score": 1.0,
+                                    "content": "incremental training approach can potentially speed up the training significantly due to savings in",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                297,
+                                217,
+                                311
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        297,
+                                        217,
+                                        311
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the early stages of training.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        }
+                    ],
+                    "index": 16.5,
+                    "bbox_fs": [
+                        105,
+                        264,
+                        506,
+                        311
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        315,
+                        505,
+                        347
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                315,
+                                505,
+                                327
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        315,
+                                        505,
+                                        327
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The rest of the paper is organized as follows. Section 2 summarizes relevant work. We describe",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                326,
+                                505,
+                                338
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        326,
+                                        505,
+                                        338
+                                    ],
+                                    "score": 1.0,
+                                    "content": "our approach in Section 3 and then present some intuition in Section 4. Finally, we show our",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                337,
+                                243,
+                                348
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        337,
+                                        243,
+                                        348
+                                    ],
+                                    "score": 1.0,
+                                    "content": "experimental results in Section 5.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        }
+                    ],
+                    "index": 20,
+                    "bbox_fs": [
+                        106,
+                        315,
+                        505,
+                        348
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        360,
+                        211,
+                        372
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                358,
+                                213,
+                                375
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        358,
+                                        213,
+                                        375
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2 RELATED WORK",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        }
+                    ],
+                    "index": 22
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        385,
+                        360,
+                        396
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                384,
+                                361,
+                                397
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        384,
+                                        361,
+                                        397
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2.1 COMPRESSION TECHNIQUES FOR NEURAL NETWORKS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        }
+                    ],
+                    "index": 23
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        406,
+                        505,
+                        461
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                405,
+                                505,
+                                419
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        405,
+                                        505,
+                                        419
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Our work is closely related to a compression technique based on network pruning. However, the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                417,
+                                505,
+                                429
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        417,
+                                        505,
+                                        429
+                                    ],
+                                    "score": 1.0,
+                                    "content": "important difference is that we do not try to select the connections which are redundant. Instead,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                428,
+                                505,
+                                440
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        428,
+                                        505,
+                                        440
+                                    ],
+                                    "score": 1.0,
+                                    "content": "we just fix a random connectivity pattern and let the network train around it. We also give a brief",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                439,
+                                505,
+                                451
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        439,
+                                        505,
+                                        451
+                                    ],
+                                    "score": 1.0,
+                                    "content": "overview of other two popular techniques: quantization and decomposition, though these directions",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                450,
+                                366,
+                                462
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        450,
+                                        366,
+                                        462
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are not the main focus and could be complementary to our work.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        }
+                    ],
+                    "index": 26,
+                    "bbox_fs": [
+                        106,
+                        405,
+                        505,
+                        462
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        471,
+                        505,
+                        537
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                471,
+                                505,
+                                484
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        471,
+                                        505,
+                                        484
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Network pruning Much initial work on neural network compression focuses on removing unim-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                483,
+                                505,
+                                495
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        483,
+                                        505,
+                                        495
+                                    ],
+                                    "score": 1.0,
+                                    "content": "portant connections using weight decay. Hanson & Pratt (1989) introduce hyperbolic and exponen-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                493,
+                                505,
+                                506
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        493,
+                                        505,
+                                        506
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tial biases to the objective. Optimal Brain Damage (LeCun et al., 1989) and Optimal Brain Surgeon",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                504,
+                                505,
+                                517
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        504,
+                                        505,
+                                        517
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(Hassibi & Stork, 1993) prune the networks based on second-order derivatives of the objectives.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                515,
+                                506,
+                                529
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        515,
+                                        506,
+                                        529
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Recent work by Han et al. (2015; 2016a) alternates between pruning near-zero weights, which are",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                525,
+                                406,
+                                539
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        525,
+                                        167,
+                                        539
+                                    ],
+                                    "score": 1.0,
+                                    "content": "encouraged by",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        168,
+                                        527,
+                                        178,
+                                        536
+                                    ],
+                                    "score": 0.26,
+                                    "content": "\\ell { 1 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        178,
+                                        525,
+                                        189,
+                                        539
+                                    ],
+                                    "score": 1.0,
+                                    "content": "or",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        190,
+                                        526,
+                                        200,
+                                        536
+                                    ],
+                                    "score": 0.62,
+                                    "content": "\\ell 2",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        201,
+                                        525,
+                                        406,
+                                        539
+                                    ],
+                                    "score": 1.0,
+                                    "content": "regularization, and retraining the pruned networks.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        }
+                    ],
+                    "index": 31.5,
+                    "bbox_fs": [
+                        105,
+                        471,
+                        506,
+                        539
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        542,
+                        505,
+                        642
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                543,
+                                505,
+                                555
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        543,
+                                        505,
+                                        555
+                                    ],
+                                    "score": 1.0,
+                                    "content": "More complex regularizers have also been considered. Wen et al. (2016) and Li et al. (2016) put",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                553,
+                                505,
+                                567
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        553,
+                                        505,
+                                        567
+                                    ],
+                                    "score": 1.0,
+                                    "content": "structured sparsity regularizers on the weights, while Murray & Chiang (2015) put them on the hid-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                565,
+                                505,
+                                577
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        565,
+                                        505,
+                                        577
+                                    ],
+                                    "score": 1.0,
+                                    "content": "den units. Feng & Darrell (2015) explore a nonparametric prior based on the Indian buffet processes",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                576,
+                                506,
+                                588
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        576,
+                                        506,
+                                        588
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(Griffiths & Ghahramani, 2011) on layers. Hu et al. (2016) prune neurons based on the analysis of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                586,
+                                506,
+                                601
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        586,
+                                        506,
+                                        601
+                                    ],
+                                    "score": 1.0,
+                                    "content": "their outputs on a large dataset. Anwar et al. (2015b) consider special sparsity patterns: channel-wise",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                597,
+                                506,
+                                612
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        597,
+                                        506,
+                                        612
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(removing a feature map/channel from a layer), kernel-wise (removing all connections between two",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                609,
+                                505,
+                                621
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        609,
+                                        505,
+                                        621
+                                    ],
+                                    "score": 1.0,
+                                    "content": "feature maps in consecutive layers), and intra-kernel-strided (removing connections between two",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                619,
+                                506,
+                                633
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        619,
+                                        506,
+                                        633
+                                    ],
+                                    "score": 1.0,
+                                    "content": "features with particular stride and offset). They also propose to use particle filter to decide the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                630,
+                                320,
+                                644
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        630,
+                                        320,
+                                        644
+                                    ],
+                                    "score": 1.0,
+                                    "content": "importance of connections and paths during training.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        }
+                    ],
+                    "index": 39,
+                    "bbox_fs": [
+                        105,
+                        543,
+                        506,
+                        644
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        648,
+                        503,
+                        681
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                647,
+                                505,
+                                659
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        647,
+                                        505,
+                                        659
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Another line of work explores fixed network architectures with some subsets of connections re-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                659,
+                                505,
+                                671
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        659,
+                                        505,
+                                        671
+                                    ],
+                                    "score": 1.0,
+                                    "content": "moved. For example, LeCun et al. (1998) remove connections between the first two convolutional",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                669,
+                                505,
+                                683
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        669,
+                                        505,
+                                        683
+                                    ],
+                                    "score": 1.0,
+                                    "content": "feature maps in a completely uniform manner. This is similar to our approach but they only con-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                83,
+                                505,
+                                95
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        83,
+                                        505,
+                                        95
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sider a pre-defined pattern in which the same number of input feature map are assigned to each",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                93,
+                                505,
+                                106
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        93,
+                                        505,
+                                        106
+                                    ],
+                                    "score": 1.0,
+                                    "content": "output feature map (Random Connection Table in Torch’s SpatialConvolutionMap function). Fur-",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                105,
+                                505,
+                                117
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        105,
+                                        505,
+                                        117
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ther, they do not explore how sparse connections affect performance compared to dense networks.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                115,
+                                505,
+                                128
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        115,
+                                        505,
+                                        128
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Along a similar vein, Cires¸an et al. (2011) remove random connections in their MNIST experi-",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                127,
+                                506,
+                                139
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        127,
+                                        506,
+                                        139
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ments. However, they do not try to preserve the spatial convolutional density and it might be a",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                137,
+                                506,
+                                150
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        137,
+                                        506,
+                                        150
+                                    ],
+                                    "score": 1.0,
+                                    "content": "challenge to harvest the savings on existing hardware. Ioannou et al. (2016a) explore three types of",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                149,
+                                506,
+                                161
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        149,
+                                        506,
+                                        161
+                                    ],
+                                    "score": 1.0,
+                                    "content": "hierarchical arrangements of filter groups for CNNs, which depend on different assumptions about",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                159,
+                                505,
+                                172
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        159,
+                                        505,
+                                        172
+                                    ],
+                                    "score": 1.0,
+                                    "content": "co-dependency of filters within each layer. These arrangements include columnar topologies in-",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                170,
+                                506,
+                                183
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        170,
+                                        506,
+                                        183
+                                    ],
+                                    "score": 1.0,
+                                    "content": "spired by AlexNet (Krizhevsky et al., 2012), tree-like topologies previously used by Ioannou et al.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                181,
+                                506,
+                                194
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        181,
+                                        506,
+                                        194
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(2016b), and root-like topologies. Finally, Howard (2017) proposes the depth multiplier method to",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                192,
+                                505,
+                                205
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        192,
+                                        505,
+                                        205
+                                    ],
+                                    "score": 1.0,
+                                    "content": "scale down the number of filters in each convolutional layer by a factor. In this case, depth multi-",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                203,
+                                505,
+                                216
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        203,
+                                        505,
+                                        216
+                                    ],
+                                    "score": 1.0,
+                                    "content": "plier can be thought of channel-wise pruning mentioned in (Anwar et al., 2015b). However, depth",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                213,
+                                505,
+                                227
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        213,
+                                        505,
+                                        227
+                                    ],
+                                    "score": 1.0,
+                                    "content": "multiplier modifies the network architectures before training and removes each layer’s feature maps",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                225,
+                                193,
+                                237
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        225,
+                                        193,
+                                        237
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in a uniform manner.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 13
+                        }
+                    ],
+                    "index": 45,
+                    "bbox_fs": [
+                        106,
+                        647,
+                        505,
+                        683
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        82,
+                        505,
+                        236
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                83,
+                                505,
+                                95
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        83,
+                                        505,
+                                        95
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sider a pre-defined pattern in which the same number of input feature map are assigned to each",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                93,
+                                505,
+                                106
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        93,
+                                        505,
+                                        106
+                                    ],
+                                    "score": 1.0,
+                                    "content": "output feature map (Random Connection Table in Torch’s SpatialConvolutionMap function). Fur-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                105,
+                                505,
+                                117
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        105,
+                                        505,
+                                        117
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ther, they do not explore how sparse connections affect performance compared to dense networks.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                115,
+                                505,
+                                128
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        115,
+                                        505,
+                                        128
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Along a similar vein, Cires¸an et al. (2011) remove random connections in their MNIST experi-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                127,
+                                506,
+                                139
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        127,
+                                        506,
+                                        139
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ments. However, they do not try to preserve the spatial convolutional density and it might be a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                137,
+                                506,
+                                150
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        137,
+                                        506,
+                                        150
+                                    ],
+                                    "score": 1.0,
+                                    "content": "challenge to harvest the savings on existing hardware. Ioannou et al. (2016a) explore three types of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                149,
+                                506,
+                                161
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        149,
+                                        506,
+                                        161
+                                    ],
+                                    "score": 1.0,
+                                    "content": "hierarchical arrangements of filter groups for CNNs, which depend on different assumptions about",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                159,
+                                505,
+                                172
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        159,
+                                        505,
+                                        172
+                                    ],
+                                    "score": 1.0,
+                                    "content": "co-dependency of filters within each layer. These arrangements include columnar topologies in-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                170,
+                                506,
+                                183
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        170,
+                                        506,
+                                        183
+                                    ],
+                                    "score": 1.0,
+                                    "content": "spired by AlexNet (Krizhevsky et al., 2012), tree-like topologies previously used by Ioannou et al.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                181,
+                                506,
+                                194
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        181,
+                                        506,
+                                        194
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(2016b), and root-like topologies. Finally, Howard (2017) proposes the depth multiplier method to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                192,
+                                505,
+                                205
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        192,
+                                        505,
+                                        205
+                                    ],
+                                    "score": 1.0,
+                                    "content": "scale down the number of filters in each convolutional layer by a factor. In this case, depth multi-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                203,
+                                505,
+                                216
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        203,
+                                        505,
+                                        216
+                                    ],
+                                    "score": 1.0,
+                                    "content": "plier can be thought of channel-wise pruning mentioned in (Anwar et al., 2015b). However, depth",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                213,
+                                505,
+                                227
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        213,
+                                        505,
+                                        227
+                                    ],
+                                    "score": 1.0,
+                                    "content": "multiplier modifies the network architectures before training and removes each layer’s feature maps",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                225,
+                                193,
+                                237
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        225,
+                                        193,
+                                        237
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in a uniform manner.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        }
+                    ],
+                    "index": 6.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        241,
+                        504,
+                        286
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                242,
+                                504,
+                                254
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        242,
+                                        504,
+                                        254
+                                    ],
+                                    "score": 1.0,
+                                    "content": "With the exception of (Anwar et al., 2015b; Li et al., 2016; Ioannou et al., 2016a) and depth multi-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                252,
+                                506,
+                                266
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        252,
+                                        506,
+                                        266
+                                    ],
+                                    "score": 1.0,
+                                    "content": "plier (Howard, 2017), the above previous work performs connection pruning that leads to irregular",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                264,
+                                505,
+                                276
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        264,
+                                        505,
+                                        276
+                                    ],
+                                    "score": 1.0,
+                                    "content": "network architectures. Thus, those techniques require additional efforts to represent network con-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                275,
+                                394,
+                                288
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        275,
+                                        394,
+                                        288
+                                    ],
+                                    "score": 1.0,
+                                    "content": "nections and might or might not allow for direct computational savings.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        }
+                    ],
+                    "index": 15.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        297,
+                        505,
+                        385
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                297,
+                                506,
+                                310
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        297,
+                                        506,
+                                        310
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Quantization Reducing the degree of redundancy of model parameters can be done in the form",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                308,
+                                506,
+                                321
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        308,
+                                        506,
+                                        321
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of quantization of network parameters. Hwang & Sung (2014); Arora et al. (2014) and Courbariaux",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                319,
+                                505,
+                                332
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        319,
+                                        505,
+                                        332
+                                    ],
+                                    "score": 1.0,
+                                    "content": "et al. (2015; 2016); Rastegari et al. (2016) propose to train CNNs with ternary weights and binary",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                330,
+                                505,
+                                343
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        330,
+                                        505,
+                                        343
+                                    ],
+                                    "score": 1.0,
+                                    "content": "weights, respectively. Gong et al. (2014) use vector quantization for parameters in fully connected",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                341,
+                                505,
+                                353
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        341,
+                                        505,
+                                        353
+                                    ],
+                                    "score": 1.0,
+                                    "content": "layers. Anwar et al. (2015a) quantize a network with the squared error minimization. Chen et al.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                351,
+                                506,
+                                366
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        351,
+                                        506,
+                                        366
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(2015) randomly group network parameters using a hash function. We note that this technique could",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                362,
+                                505,
+                                377
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        362,
+                                        505,
+                                        377
+                                    ],
+                                    "score": 1.0,
+                                    "content": "be complementary to network pruning. For example, Han et al. (2016a) combine connection pruning",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                374,
+                                349,
+                                388
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        374,
+                                        349,
+                                        388
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in (Han et al., 2015) with quantization and Huffman coding.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        }
+                    ],
+                    "index": 21.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        396,
+                        505,
+                        495
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                397,
+                                505,
+                                410
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        397,
+                                        505,
+                                        410
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Decomposition Another approach is based on low-rank decomposition of the parameters. De-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                408,
+                                505,
+                                420
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        408,
+                                        505,
+                                        420
+                                    ],
+                                    "score": 1.0,
+                                    "content": "composition methods include truncated SVD (Denton et al., 2014), decomposition to rank-1 bases",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                417,
+                                506,
+                                432
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        417,
+                                        506,
+                                        432
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(Jaderberg et al., 2014), CP decomposition (PARAFAC or CANDECOMP) (Lebedev et al., 2015),",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                427,
+                                505,
+                                443
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        427,
+                                        505,
+                                        443
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Tensor-Train decomposition of Oseledets (2011) (Novikov et al., 2015), sparse dictionary learning",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                439,
+                                506,
+                                455
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        439,
+                                        506,
+                                        455
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of Mairal et al. (2009) and PCA (Liu et al., 2015), asymmetric (3D) decomposition using reconstruc-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                451,
+                                505,
+                                464
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        451,
+                                        505,
+                                        464
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tion loss of non-linear responses combined with a rank selection method based on PCA accumulated",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                461,
+                                506,
+                                475
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        461,
+                                        506,
+                                        475
+                                    ],
+                                    "score": 1.0,
+                                    "content": "energy (Zhang et al., 2015b;a), and Tucker decomposition using the kernel tensor reconstruction loss",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                473,
+                                506,
+                                486
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        473,
+                                        506,
+                                        486
+                                    ],
+                                    "score": 1.0,
+                                    "content": "combined with a rank selection method based on global analytic variational Bayesian matrix factor-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                484,
+                                212,
+                                497
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        484,
+                                        212,
+                                        497
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ization (Kim et al., 2016).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        }
+                    ],
+                    "index": 30
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        506,
+                        314,
+                        518
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                505,
+                                315,
+                                519
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        505,
+                                        315,
+                                        519
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2.2 REGULARIZATION OF NEURAL NETWORKS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        }
+                    ],
+                    "index": 35
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        528,
+                        504,
+                        605
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                528,
+                                506,
+                                542
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        528,
+                                        506,
+                                        542
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Hinton et al. (2012); Srivastava et al. (2014) propose Dropout for regularizing fully connected layers",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                539,
+                                505,
+                                552
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        539,
+                                        505,
+                                        552
+                                    ],
+                                    "score": 1.0,
+                                    "content": "within neural networks layers by randomly setting a subset of activations to zero during training.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                550,
+                                505,
+                                563
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        550,
+                                        505,
+                                        563
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Wan et al. (2013) later propose DropConnect, a generalization of Dropout that instead randomly sets",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                561,
+                                505,
+                                573
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        561,
+                                        505,
+                                        573
+                                    ],
+                                    "score": 1.0,
+                                    "content": "a subset of weights or connections to zero. Our approach could be thought as related to DropCon-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                572,
+                                505,
+                                584
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        572,
+                                        505,
+                                        584
+                                    ],
+                                    "score": 1.0,
+                                    "content": "nect, but (1) we remove connections before training; (2) we focus on connections between convolu-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                583,
+                                505,
+                                596
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        583,
+                                        505,
+                                        596
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tional layers; and (3) we kill connections in a more regular manner by restricting connection patterns",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                594,
+                                270,
+                                606
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        594,
+                                        270,
+                                        606
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to be the same along spatial dimensions.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        }
+                    ],
+                    "index": 39
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        610,
+                        504,
+                        655
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                610,
+                                506,
+                                623
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        610,
+                                        506,
+                                        623
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Recently, Han et al. (2016b) and Jin et al. (2016) propose a form of regularization where dropped",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                621,
+                                506,
+                                635
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        621,
+                                        506,
+                                        635
+                                    ],
+                                    "score": 1.0,
+                                    "content": "connections are unfrozen and the network is retrained. This idea is similar to our incremental training",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                633,
+                                506,
+                                645
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        633,
+                                        506,
+                                        645
+                                    ],
+                                    "score": 1.0,
+                                    "content": "approach. However, (1) we do not start with a full network; (2) we do not unfreeze connections all",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                644,
+                                380,
+                                656
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        644,
+                                        380,
+                                        656
+                                    ],
+                                    "score": 1.0,
+                                    "content": "at once; and (3) we preserve regularity of the convolution operation.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        }
+                    ],
+                    "index": 44.5
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        666,
+                        287,
+                        677
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                666,
+                                290,
+                                678
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        666,
+                                        290,
+                                        678
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2.3 NEURAL NETWORK ARCHITECTURES",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 47
+                        }
+                    ],
+                    "index": 47
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        687,
+                        504,
+                        731
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                687,
+                                505,
+                                700
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        687,
+                                        505,
+                                        700
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Network compression and architectures are closely related. The goal of compression is to remove",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 48
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                699,
+                                505,
+                                711
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        699,
+                                        505,
+                                        711
+                                    ],
+                                    "score": 1.0,
+                                    "content": "redundancy in network parameters; therefore, the knowledge about traits that determine architec-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 49
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                709,
+                                505,
+                                722
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        709,
+                                        505,
+                                        722
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ture’s success would be desirable. Other than the discovery that depth is an important factor (Ba &",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 50
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                720,
+                                302,
+                                732
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        720,
+                                        302,
+                                        732
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Caruana, 2014), little is known about such traits.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 51
+                        }
+                    ],
+                    "index": 49.5
+                }
+            ],
+            "page_idx": 2,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        107,
+                        27,
+                        308,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                107,
+                                26,
+                                308,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        26,
+                                        308,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Under review as a conference paper at ICLR 2017",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        303,
+                        751,
+                        309,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                301,
+                                750,
+                                310,
+                                762
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        301,
+                                        750,
+                                        310,
+                                        762
+                                    ],
+                                    "score": 1.0,
+                                    "content": "",
+                                    "type": "text",
+                                    "height": 12,
+                                    "width": 9
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        82,
+                        505,
+                        236
+                    ],
+                    "lines": [],
+                    "index": 6.5,
+                    "bbox_fs": [
+                        105,
+                        83,
+                        506,
+                        237
+                    ],
+                    "lines_deleted": true
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        241,
+                        504,
+                        286
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                242,
+                                504,
+                                254
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        242,
+                                        504,
+                                        254
+                                    ],
+                                    "score": 1.0,
+                                    "content": "With the exception of (Anwar et al., 2015b; Li et al., 2016; Ioannou et al., 2016a) and depth multi-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                252,
+                                506,
+                                266
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        252,
+                                        506,
+                                        266
+                                    ],
+                                    "score": 1.0,
+                                    "content": "plier (Howard, 2017), the above previous work performs connection pruning that leads to irregular",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                264,
+                                505,
+                                276
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        264,
+                                        505,
+                                        276
+                                    ],
+                                    "score": 1.0,
+                                    "content": "network architectures. Thus, those techniques require additional efforts to represent network con-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                275,
+                                394,
+                                288
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        275,
+                                        394,
+                                        288
+                                    ],
+                                    "score": 1.0,
+                                    "content": "nections and might or might not allow for direct computational savings.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        }
+                    ],
+                    "index": 15.5,
+                    "bbox_fs": [
+                        105,
+                        242,
+                        506,
+                        288
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        297,
+                        505,
+                        385
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                297,
+                                506,
+                                310
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        297,
+                                        506,
+                                        310
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Quantization Reducing the degree of redundancy of model parameters can be done in the form",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                308,
+                                506,
+                                321
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        308,
+                                        506,
+                                        321
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of quantization of network parameters. Hwang & Sung (2014); Arora et al. (2014) and Courbariaux",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                319,
+                                505,
+                                332
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        319,
+                                        505,
+                                        332
+                                    ],
+                                    "score": 1.0,
+                                    "content": "et al. (2015; 2016); Rastegari et al. (2016) propose to train CNNs with ternary weights and binary",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                330,
+                                505,
+                                343
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        330,
+                                        505,
+                                        343
+                                    ],
+                                    "score": 1.0,
+                                    "content": "weights, respectively. Gong et al. (2014) use vector quantization for parameters in fully connected",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                341,
+                                505,
+                                353
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        341,
+                                        505,
+                                        353
+                                    ],
+                                    "score": 1.0,
+                                    "content": "layers. Anwar et al. (2015a) quantize a network with the squared error minimization. Chen et al.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                351,
+                                506,
+                                366
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        351,
+                                        506,
+                                        366
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(2015) randomly group network parameters using a hash function. We note that this technique could",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                362,
+                                505,
+                                377
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        362,
+                                        505,
+                                        377
+                                    ],
+                                    "score": 1.0,
+                                    "content": "be complementary to network pruning. For example, Han et al. (2016a) combine connection pruning",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                374,
+                                349,
+                                388
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        374,
+                                        349,
+                                        388
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in (Han et al., 2015) with quantization and Huffman coding.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        }
+                    ],
+                    "index": 21.5,
+                    "bbox_fs": [
+                        104,
+                        297,
+                        506,
+                        388
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        396,
+                        505,
+                        495
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                397,
+                                505,
+                                410
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        397,
+                                        505,
+                                        410
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Decomposition Another approach is based on low-rank decomposition of the parameters. De-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                408,
+                                505,
+                                420
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        408,
+                                        505,
+                                        420
+                                    ],
+                                    "score": 1.0,
+                                    "content": "composition methods include truncated SVD (Denton et al., 2014), decomposition to rank-1 bases",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                417,
+                                506,
+                                432
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        417,
+                                        506,
+                                        432
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(Jaderberg et al., 2014), CP decomposition (PARAFAC or CANDECOMP) (Lebedev et al., 2015),",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                427,
+                                505,
+                                443
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        427,
+                                        505,
+                                        443
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Tensor-Train decomposition of Oseledets (2011) (Novikov et al., 2015), sparse dictionary learning",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                439,
+                                506,
+                                455
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        439,
+                                        506,
+                                        455
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of Mairal et al. (2009) and PCA (Liu et al., 2015), asymmetric (3D) decomposition using reconstruc-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                451,
+                                505,
+                                464
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        451,
+                                        505,
+                                        464
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tion loss of non-linear responses combined with a rank selection method based on PCA accumulated",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                461,
+                                506,
+                                475
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        461,
+                                        506,
+                                        475
+                                    ],
+                                    "score": 1.0,
+                                    "content": "energy (Zhang et al., 2015b;a), and Tucker decomposition using the kernel tensor reconstruction loss",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                473,
+                                506,
+                                486
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        473,
+                                        506,
+                                        486
+                                    ],
+                                    "score": 1.0,
+                                    "content": "combined with a rank selection method based on global analytic variational Bayesian matrix factor-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                484,
+                                212,
+                                497
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        484,
+                                        212,
+                                        497
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ization (Kim et al., 2016).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        }
+                    ],
+                    "index": 30,
+                    "bbox_fs": [
+                        104,
+                        397,
+                        506,
+                        497
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        506,
+                        314,
+                        518
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                505,
+                                315,
+                                519
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        505,
+                                        315,
+                                        519
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2.2 REGULARIZATION OF NEURAL NETWORKS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        }
+                    ],
+                    "index": 35
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        528,
+                        504,
+                        605
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                528,
+                                506,
+                                542
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        528,
+                                        506,
+                                        542
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Hinton et al. (2012); Srivastava et al. (2014) propose Dropout for regularizing fully connected layers",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                539,
+                                505,
+                                552
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        539,
+                                        505,
+                                        552
+                                    ],
+                                    "score": 1.0,
+                                    "content": "within neural networks layers by randomly setting a subset of activations to zero during training.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                550,
+                                505,
+                                563
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        550,
+                                        505,
+                                        563
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Wan et al. (2013) later propose DropConnect, a generalization of Dropout that instead randomly sets",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                561,
+                                505,
+                                573
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        561,
+                                        505,
+                                        573
+                                    ],
+                                    "score": 1.0,
+                                    "content": "a subset of weights or connections to zero. Our approach could be thought as related to DropCon-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                572,
+                                505,
+                                584
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        572,
+                                        505,
+                                        584
+                                    ],
+                                    "score": 1.0,
+                                    "content": "nect, but (1) we remove connections before training; (2) we focus on connections between convolu-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                583,
+                                505,
+                                596
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        583,
+                                        505,
+                                        596
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tional layers; and (3) we kill connections in a more regular manner by restricting connection patterns",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                594,
+                                270,
+                                606
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        594,
+                                        270,
+                                        606
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to be the same along spatial dimensions.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        }
+                    ],
+                    "index": 39,
+                    "bbox_fs": [
+                        105,
+                        528,
+                        506,
+                        606
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        610,
+                        504,
+                        655
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                610,
+                                506,
+                                623
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        610,
+                                        506,
+                                        623
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Recently, Han et al. (2016b) and Jin et al. (2016) propose a form of regularization where dropped",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                621,
+                                506,
+                                635
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        621,
+                                        506,
+                                        635
+                                    ],
+                                    "score": 1.0,
+                                    "content": "connections are unfrozen and the network is retrained. This idea is similar to our incremental training",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                633,
+                                506,
+                                645
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        633,
+                                        506,
+                                        645
+                                    ],
+                                    "score": 1.0,
+                                    "content": "approach. However, (1) we do not start with a full network; (2) we do not unfreeze connections all",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                644,
+                                380,
+                                656
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        644,
+                                        380,
+                                        656
+                                    ],
+                                    "score": 1.0,
+                                    "content": "at once; and (3) we preserve regularity of the convolution operation.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        }
+                    ],
+                    "index": 44.5,
+                    "bbox_fs": [
+                        105,
+                        610,
+                        506,
+                        656
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        666,
+                        287,
+                        677
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                666,
+                                290,
+                                678
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        666,
+                                        290,
+                                        678
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2.3 NEURAL NETWORK ARCHITECTURES",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 47
+                        }
+                    ],
+                    "index": 47
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        687,
+                        504,
+                        731
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                687,
+                                505,
+                                700
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        687,
+                                        505,
+                                        700
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Network compression and architectures are closely related. The goal of compression is to remove",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 48
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                699,
+                                505,
+                                711
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        699,
+                                        505,
+                                        711
+                                    ],
+                                    "score": 1.0,
+                                    "content": "redundancy in network parameters; therefore, the knowledge about traits that determine architec-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 49
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                709,
+                                505,
+                                722
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        709,
+                                        505,
+                                        722
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ture’s success would be desirable. Other than the discovery that depth is an important factor (Ba &",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 50
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                720,
+                                302,
+                                732
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        720,
+                                        302,
+                                        732
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Caruana, 2014), little is known about such traits.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 51
+                        }
+                    ],
+                    "index": 49.5,
+                    "bbox_fs": [
+                        105,
+                        687,
+                        505,
+                        732
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        82,
+                        505,
+                        138
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                82,
+                                505,
+                                96
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        82,
+                                        505,
+                                        96
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Some previous work performs architecture search but without the main goal of doing compression",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                93,
+                                505,
+                                106
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        93,
+                                        505,
+                                        106
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(Murray & Chiang, 2015; De Brabandere et al., 2016). Recent work proposes shortcut/skip con-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                104,
+                                505,
+                                118
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        104,
+                                        505,
+                                        118
+                                    ],
+                                    "score": 1.0,
+                                    "content": "nections to convolutional networks. See, among others, highway networks (Srivastava et al., 2015),",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                115,
+                                505,
+                                128
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        115,
+                                        505,
+                                        128
+                                    ],
+                                    "score": 1.0,
+                                    "content": "residual networks (He et al., 2016a;b), networks with stochastic depth (Huang et al., 2016b), and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                126,
+                                367,
+                                140
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        126,
+                                        367,
+                                        140
+                                    ],
+                                    "score": 1.0,
+                                    "content": "densely connected convolutional networks (Huang et al., 2016a).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        }
+                    ],
+                    "index": 2
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        149,
+                        183,
+                        162
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                147,
+                                185,
+                                164
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        147,
+                                        185,
+                                        164
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3 APPROACH",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        }
+                    ],
+                    "index": 5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        173,
+                        505,
+                        239
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                172,
+                                505,
+                                186
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        172,
+                                        505,
+                                        186
+                                    ],
+                                    "score": 1.0,
+                                    "content": "A CNN architecture consist of (1) convolutional layers, (2) pooling layers, (3) fully connected layers,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                185,
+                                505,
+                                196
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        185,
+                                        505,
+                                        196
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and (4) a topology that governs how these layers are organized. Given an architecture, our general",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                194,
+                                505,
+                                209
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        194,
+                                        505,
+                                        209
+                                    ],
+                                    "score": 1.0,
+                                    "content": "goal is to transform it into another architecture with a smaller number of parameters. In this paper,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                206,
+                                505,
+                                218
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        206,
+                                        505,
+                                        218
+                                    ],
+                                    "score": 1.0,
+                                    "content": "we limit ourselves to transformation functions that keep the general topology of the input architec-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                217,
+                                505,
+                                230
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        217,
+                                        505,
+                                        230
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ture intact. Moreover, the main focus will be on the convolutional layers and convolution operations,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                228,
+                                477,
+                                240
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        228,
+                                        477,
+                                        240
+                                    ],
+                                    "score": 1.0,
+                                    "content": "as they impose highest computational and memory burden for most if not all large networks.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        }
+                    ],
+                    "index": 8.5
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        249,
+                        215,
+                        259
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                248,
+                                216,
+                                261
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        248,
+                                        216,
+                                        261
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3.1 DEPTH MULTIPLIER",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        }
+                    ],
+                    "index": 12
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        270,
+                        505,
+                        314
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                269,
+                                505,
+                                282
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        269,
+                                        505,
+                                        282
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We first give a description of the depth multiplier method used in Howard (2017). Given a hy-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                281,
+                                506,
+                                293
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        281,
+                                        162,
+                                        293
+                                    ],
+                                    "score": 1.0,
+                                    "content": "perparameter",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        163,
+                                        281,
+                                        206,
+                                        293
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\alpha \\in ( 0 , 1 ]",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        206,
+                                        281,
+                                        506,
+                                        293
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", the depth multiplier approach scales down the number of filters in each",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                293,
+                                505,
+                                304
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        293,
+                                        203,
+                                        304
+                                    ],
+                                    "score": 1.0,
+                                    "content": "convolutional layers by",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        203,
+                                        294,
+                                        211,
+                                        302
+                                    ],
+                                    "score": 0.72,
+                                    "content": "\\alpha",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        211,
+                                        293,
+                                        505,
+                                        304
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Note that depth here refers to the third dimension of the activation vol-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                304,
+                                384,
+                                315
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        304,
+                                        384,
+                                        315
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ume of a single layer, not the number of layers in the whole network.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        }
+                    ],
+                    "index": 14.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        320,
+                        503,
+                        353
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                319,
+                                505,
+                                333
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        319,
+                                        123,
+                                        333
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Let",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        123,
+                                        322,
+                                        144,
+                                        332
+                                    ],
+                                    "score": 0.87,
+                                    "content": "n _ { l - 1 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        144,
+                                        319,
+                                        163,
+                                        333
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        163,
+                                        322,
+                                        173,
+                                        331
+                                    ],
+                                    "score": 0.82,
+                                    "content": "n _ { l }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        174,
+                                        319,
+                                        378,
+                                        333
+                                    ],
+                                    "score": 1.0,
+                                    "content": "be the number of input and output filters at layer",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        378,
+                                        321,
+                                        383,
+                                        330
+                                    ],
+                                    "score": 0.57,
+                                    "content": "l",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        383,
+                                        319,
+                                        505,
+                                        333
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", respectively. After the oper-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                331,
+                                505,
+                                343
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        331,
+                                        129,
+                                        343
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ation",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        129,
+                                        333,
+                                        150,
+                                        343
+                                    ],
+                                    "score": 0.87,
+                                    "content": "n _ { l - 1 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        150,
+                                        331,
+                                        169,
+                                        343
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        169,
+                                        333,
+                                        180,
+                                        342
+                                    ],
+                                    "score": 0.83,
+                                    "content": "n _ { l }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        180,
+                                        331,
+                                        217,
+                                        343
+                                    ],
+                                    "score": 1.0,
+                                    "content": "become",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        217,
+                                        331,
+                                        252,
+                                        343
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\lceil \\alpha n _ { l - 1 } \\rceil",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        252,
+                                        331,
+                                        272,
+                                        343
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        272,
+                                        331,
+                                        297,
+                                        343
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\lceil \\alpha n _ { l } \\rceil",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        297,
+                                        331,
+                                        505,
+                                        343
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and the number of parameters (and the number of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                342,
+                                327,
+                                355
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        342,
+                                        209,
+                                        355
+                                    ],
+                                    "score": 1.0,
+                                    "content": "multiplications) becomes",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        209,
+                                        342,
+                                        232,
+                                        352
+                                    ],
+                                    "score": 0.87,
+                                    "content": "\\approx \\alpha ^ { 2 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        232,
+                                        342,
+                                        327,
+                                        355
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of the original number.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        }
+                    ],
+                    "index": 18
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        358,
+                        505,
+                        414
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                358,
+                                505,
+                                371
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        359,
+                                        317,
+                                        371
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The result of this operation is a network that is both",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        317,
+                                        358,
+                                        339,
+                                        371
+                                    ],
+                                    "score": 0.91,
+                                    "content": "1 / \\alpha ^ { 2 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        340,
+                                        359,
+                                        505,
+                                        371
+                                    ],
+                                    "score": 1.0,
+                                    "content": "smaller and faster. Many large networks",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                370,
+                                505,
+                                382
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        370,
+                                        505,
+                                        382
+                                    ],
+                                    "score": 1.0,
+                                    "content": "can be significantly reduced in size using this method with only a small loss of precision (Howard,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                380,
+                                505,
+                                394
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        380,
+                                        505,
+                                        394
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2017). It is our belief that this method establishes a strong baseline to which any other advanced",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                391,
+                                505,
+                                405
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        391,
+                                        505,
+                                        405
+                                    ],
+                                    "score": 1.0,
+                                    "content": "techniques should compare themselves. To the best of our knowledge, we are not aware of such",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                403,
+                                224,
+                                415
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        403,
+                                        224,
+                                        415
+                                    ],
+                                    "score": 1.0,
+                                    "content": "comparisons in the literature.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        }
+                    ],
+                    "index": 22
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        424,
+                        207,
+                        435
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                423,
+                                208,
+                                437
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        423,
+                                        208,
+                                        437
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3.2 SPARSE RANDOM",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        }
+                    ],
+                    "index": 25
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        445,
+                        505,
+                        500
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                446,
+                                505,
+                                457
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        446,
+                                        505,
+                                        457
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Instead of looking at depth multiplier as deactivating channels in the convolutional layers, we can",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                456,
+                                505,
+                                468
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        456,
+                                        505,
+                                        468
+                                    ],
+                                    "score": 1.0,
+                                    "content": "look at it from the perspective of deactivating connections. From this point of view, depth multiplier",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                466,
+                                505,
+                                480
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        466,
+                                        505,
+                                        480
+                                    ],
+                                    "score": 1.0,
+                                    "content": "kills the connections between two convolutional layers such that (a) the connection patterns are",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                477,
+                                505,
+                                491
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        477,
+                                        505,
+                                        491
+                                    ],
+                                    "score": 1.0,
+                                    "content": "still the same across spatial dimensions and (b) all “alive” input channels are fully connected to all",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                488,
+                                205,
+                                502
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        488,
+                                        205,
+                                        502
+                                    ],
+                                    "score": 1.0,
+                                    "content": "“alive” output channels.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        }
+                    ],
+                    "index": 28
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        506,
+                        505,
+                        583
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                506,
+                                505,
+                                519
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        506,
+                                        505,
+                                        519
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We generalize this approach by relaxing (b) while maintaining (a). That is, for every output chan-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                518,
+                                505,
+                                529
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        518,
+                                        505,
+                                        529
+                                    ],
+                                    "score": 1.0,
+                                    "content": "nel, we connect it to a small subset of input channels. In other words, dense connections between",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                528,
+                                505,
+                                540
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        528,
+                                        505,
+                                        540
+                                    ],
+                                    "score": 1.0,
+                                    "content": "a small number of channels become sparse connections between larger number of channels. This",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                539,
+                                505,
+                                551
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        539,
+                                        505,
+                                        551
+                                    ],
+                                    "score": 1.0,
+                                    "content": "can be summarized in Fig. 1. The advantage of this is that the actual convolution can still be com-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                550,
+                                505,
+                                562
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        550,
+                                        505,
+                                        562
+                                    ],
+                                    "score": 1.0,
+                                    "content": "puted efficiently because sparsity is introduced only at the outer loop of the convolution operation",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                561,
+                                505,
+                                574
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        561,
+                                        505,
+                                        574
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and we can still take the advantage of the continuous memory layout. For more details regarding",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                572,
+                                387,
+                                585
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        572,
+                                        387,
+                                        585
+                                    ],
+                                    "score": 1.0,
+                                    "content": "implementations of the two approaches, please refer to the Appendix.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        }
+                    ],
+                    "index": 34
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        588,
+                        505,
+                        645
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                589,
+                                504,
+                                601
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        589,
+                                        189,
+                                        601
+                                    ],
+                                    "score": 1.0,
+                                    "content": "More concretely, let",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        189,
+                                        590,
+                                        210,
+                                        600
+                                    ],
+                                    "score": 0.9,
+                                    "content": "n _ { l - 1 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        210,
+                                        589,
+                                        228,
+                                        601
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        228,
+                                        590,
+                                        239,
+                                        600
+                                    ],
+                                    "score": 0.84,
+                                    "content": "n _ { l }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        239,
+                                        589,
+                                        381,
+                                        601
+                                    ],
+                                    "score": 1.0,
+                                    "content": "be the number of channels of layer",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        382,
+                                        589,
+                                        403,
+                                        599
+                                    ],
+                                    "score": 0.73,
+                                    "content": "l - 1",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        404,
+                                        589,
+                                        444,
+                                        601
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and layer",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        445,
+                                        590,
+                                        449,
+                                        599
+                                    ],
+                                    "score": 0.49,
+                                    "content": "l",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        450,
+                                        589,
+                                        504,
+                                        601
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", respectively.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                600,
+                                505,
+                                612
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        600,
+                                        213,
+                                        612
+                                    ],
+                                    "score": 1.0,
+                                    "content": "For a sparsity coefficient",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        213,
+                                        602,
+                                        221,
+                                        610
+                                    ],
+                                    "score": 0.74,
+                                    "content": "\\alpha",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        221,
+                                        600,
+                                        300,
+                                        612
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", each output filter",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        300,
+                                        601,
+                                        307,
+                                        612
+                                    ],
+                                    "score": 0.82,
+                                    "content": "j",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        307,
+                                        600,
+                                        394,
+                                        612
+                                    ],
+                                    "score": 1.0,
+                                    "content": "only connects to an",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        394,
+                                        602,
+                                        402,
+                                        610
+                                    ],
+                                    "score": 0.77,
+                                    "content": "\\alpha",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        403,
+                                        600,
+                                        505,
+                                        612
+                                    ],
+                                    "score": 1.0,
+                                    "content": "fraction of filters of the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                609,
+                                506,
+                                624
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        609,
+                                        356,
+                                        624
+                                    ],
+                                    "score": 1.0,
+                                    "content": "previous layer. Thus, instead of having a connectivity matrix",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        357,
+                                        611,
+                                        378,
+                                        623
+                                    ],
+                                    "score": 0.9,
+                                    "content": "W _ { s i j }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        379,
+                                        609,
+                                        436,
+                                        624
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of dimension",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        436,
+                                        610,
+                                        501,
+                                        622
+                                    ],
+                                    "score": 0.91,
+                                    "content": "k ^ { 2 } \\times n _ { l - 1 } \\times n _ { l }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        501,
+                                        609,
+                                        506,
+                                        624
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                620,
+                                506,
+                                637
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        620,
+                                        301,
+                                        637
+                                    ],
+                                    "score": 1.0,
+                                    "content": "we have a sparse matrix with non-zero entries at",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        302,
+                                        622,
+                                        330,
+                                        634
+                                    ],
+                                    "score": 0.92,
+                                    "content": "W _ { s a _ { i j } j }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        331,
+                                        620,
+                                        361,
+                                        637
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", where",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        361,
+                                        624,
+                                        375,
+                                        634
+                                    ],
+                                    "score": 0.88,
+                                    "content": "a _ { i j }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        375,
+                                        620,
+                                        506,
+                                        637
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is an index matrix of dimension",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                632,
+                                275,
+                                648
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        633,
+                                        176,
+                                        646
+                                    ],
+                                    "score": 0.93,
+                                    "content": "k ^ { 2 } \\times \\alpha n _ { l - 1 } \\times n _ { l }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        177,
+                                        632,
+                                        195,
+                                        648
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        195,
+                                        634,
+                                        202,
+                                        644
+                                    ],
+                                    "score": 0.8,
+                                    "content": "k",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        202,
+                                        632,
+                                        275,
+                                        648
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is the kernel size.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        }
+                    ],
+                    "index": 40
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        655,
+                        247,
+                        667
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                655,
+                                249,
+                                668
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        655,
+                                        249,
+                                        668
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3.2.1 INCREMENTAL TRAINING",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        }
+                    ],
+                    "index": 43
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        676,
+                        504,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                677,
+                                506,
+                                689
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        677,
+                                        506,
+                                        689
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In contrast to depth multiplier, a sparse convolutional network defines a connection pattern on a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                687,
+                                506,
+                                700
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        687,
+                                        506,
+                                        700
+                                    ],
+                                    "score": 1.0,
+                                    "content": "much bigger network. Therefore, an interesting extension is to consider incremental training: we",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                698,
+                                505,
+                                712
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        698,
+                                        505,
+                                        712
+                                    ],
+                                    "score": 1.0,
+                                    "content": "start with a network that only contains a small fraction of connections (in our experiments we use",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                708,
+                                506,
+                                723
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        710,
+                                        121,
+                                        720
+                                    ],
+                                    "score": 0.86,
+                                    "content": "1 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        122,
+                                        708,
+                                        138,
+                                        723
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        139,
+                                        710,
+                                        162,
+                                        721
+                                    ],
+                                    "score": 0.85,
+                                    "content": "0 . 1 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        162,
+                                        708,
+                                        506,
+                                        723
+                                    ],
+                                    "score": 1.0,
+                                    "content": ") and add connections over time. This is motivated by an intuition that the network can",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 47
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                721,
+                                505,
+                                732
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        721,
+                                        505,
+                                        732
+                                    ],
+                                    "score": 1.0,
+                                    "content": "use learned channels in new contexts by introducing additional connections. The potential practical",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 48
+                        }
+                    ],
+                    "index": 46
+                }
+            ],
+            "page_idx": 3,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        107,
+                        27,
+                        308,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                107,
+                                25,
+                                308,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        25,
+                                        308,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Under review as a conference paper at ICLR 2017",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        302,
+                        752,
+                        308,
+                        759
+                    ],
+                    "lines": []
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        82,
+                        505,
+                        138
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                82,
+                                505,
+                                96
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        82,
+                                        505,
+                                        96
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Some previous work performs architecture search but without the main goal of doing compression",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                93,
+                                505,
+                                106
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        93,
+                                        505,
+                                        106
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(Murray & Chiang, 2015; De Brabandere et al., 2016). Recent work proposes shortcut/skip con-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                104,
+                                505,
+                                118
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        104,
+                                        505,
+                                        118
+                                    ],
+                                    "score": 1.0,
+                                    "content": "nections to convolutional networks. See, among others, highway networks (Srivastava et al., 2015),",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                115,
+                                505,
+                                128
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        115,
+                                        505,
+                                        128
+                                    ],
+                                    "score": 1.0,
+                                    "content": "residual networks (He et al., 2016a;b), networks with stochastic depth (Huang et al., 2016b), and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                126,
+                                367,
+                                140
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        126,
+                                        367,
+                                        140
+                                    ],
+                                    "score": 1.0,
+                                    "content": "densely connected convolutional networks (Huang et al., 2016a).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        }
+                    ],
+                    "index": 2,
+                    "bbox_fs": [
+                        105,
+                        82,
+                        505,
+                        140
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        149,
+                        183,
+                        162
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                147,
+                                185,
+                                164
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        147,
+                                        185,
+                                        164
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3 APPROACH",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        }
+                    ],
+                    "index": 5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        173,
+                        505,
+                        239
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                172,
+                                505,
+                                186
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        172,
+                                        505,
+                                        186
+                                    ],
+                                    "score": 1.0,
+                                    "content": "A CNN architecture consist of (1) convolutional layers, (2) pooling layers, (3) fully connected layers,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                185,
+                                505,
+                                196
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        185,
+                                        505,
+                                        196
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and (4) a topology that governs how these layers are organized. Given an architecture, our general",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                194,
+                                505,
+                                209
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        194,
+                                        505,
+                                        209
+                                    ],
+                                    "score": 1.0,
+                                    "content": "goal is to transform it into another architecture with a smaller number of parameters. In this paper,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                206,
+                                505,
+                                218
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        206,
+                                        505,
+                                        218
+                                    ],
+                                    "score": 1.0,
+                                    "content": "we limit ourselves to transformation functions that keep the general topology of the input architec-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                217,
+                                505,
+                                230
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        217,
+                                        505,
+                                        230
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ture intact. Moreover, the main focus will be on the convolutional layers and convolution operations,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                228,
+                                477,
+                                240
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        228,
+                                        477,
+                                        240
+                                    ],
+                                    "score": 1.0,
+                                    "content": "as they impose highest computational and memory burden for most if not all large networks.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        }
+                    ],
+                    "index": 8.5,
+                    "bbox_fs": [
+                        105,
+                        172,
+                        505,
+                        240
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        249,
+                        215,
+                        259
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                248,
+                                216,
+                                261
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        248,
+                                        216,
+                                        261
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3.1 DEPTH MULTIPLIER",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        }
+                    ],
+                    "index": 12
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        270,
+                        505,
+                        314
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                269,
+                                505,
+                                282
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        269,
+                                        505,
+                                        282
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We first give a description of the depth multiplier method used in Howard (2017). Given a hy-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                281,
+                                506,
+                                293
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        281,
+                                        162,
+                                        293
+                                    ],
+                                    "score": 1.0,
+                                    "content": "perparameter",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        163,
+                                        281,
+                                        206,
+                                        293
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\alpha \\in ( 0 , 1 ]",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        206,
+                                        281,
+                                        506,
+                                        293
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", the depth multiplier approach scales down the number of filters in each",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                293,
+                                505,
+                                304
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        293,
+                                        203,
+                                        304
+                                    ],
+                                    "score": 1.0,
+                                    "content": "convolutional layers by",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        203,
+                                        294,
+                                        211,
+                                        302
+                                    ],
+                                    "score": 0.72,
+                                    "content": "\\alpha",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        211,
+                                        293,
+                                        505,
+                                        304
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Note that depth here refers to the third dimension of the activation vol-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                304,
+                                384,
+                                315
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        304,
+                                        384,
+                                        315
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ume of a single layer, not the number of layers in the whole network.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        }
+                    ],
+                    "index": 14.5,
+                    "bbox_fs": [
+                        105,
+                        269,
+                        506,
+                        315
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        320,
+                        503,
+                        353
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                319,
+                                505,
+                                333
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        319,
+                                        123,
+                                        333
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Let",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        123,
+                                        322,
+                                        144,
+                                        332
+                                    ],
+                                    "score": 0.87,
+                                    "content": "n _ { l - 1 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        144,
+                                        319,
+                                        163,
+                                        333
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        163,
+                                        322,
+                                        173,
+                                        331
+                                    ],
+                                    "score": 0.82,
+                                    "content": "n _ { l }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        174,
+                                        319,
+                                        378,
+                                        333
+                                    ],
+                                    "score": 1.0,
+                                    "content": "be the number of input and output filters at layer",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        378,
+                                        321,
+                                        383,
+                                        330
+                                    ],
+                                    "score": 0.57,
+                                    "content": "l",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        383,
+                                        319,
+                                        505,
+                                        333
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", respectively. After the oper-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                331,
+                                505,
+                                343
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        331,
+                                        129,
+                                        343
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ation",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        129,
+                                        333,
+                                        150,
+                                        343
+                                    ],
+                                    "score": 0.87,
+                                    "content": "n _ { l - 1 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        150,
+                                        331,
+                                        169,
+                                        343
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        169,
+                                        333,
+                                        180,
+                                        342
+                                    ],
+                                    "score": 0.83,
+                                    "content": "n _ { l }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        180,
+                                        331,
+                                        217,
+                                        343
+                                    ],
+                                    "score": 1.0,
+                                    "content": "become",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        217,
+                                        331,
+                                        252,
+                                        343
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\lceil \\alpha n _ { l - 1 } \\rceil",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        252,
+                                        331,
+                                        272,
+                                        343
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        272,
+                                        331,
+                                        297,
+                                        343
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\lceil \\alpha n _ { l } \\rceil",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        297,
+                                        331,
+                                        505,
+                                        343
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and the number of parameters (and the number of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                342,
+                                327,
+                                355
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        342,
+                                        209,
+                                        355
+                                    ],
+                                    "score": 1.0,
+                                    "content": "multiplications) becomes",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        209,
+                                        342,
+                                        232,
+                                        352
+                                    ],
+                                    "score": 0.87,
+                                    "content": "\\approx \\alpha ^ { 2 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        232,
+                                        342,
+                                        327,
+                                        355
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of the original number.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        }
+                    ],
+                    "index": 18,
+                    "bbox_fs": [
+                        105,
+                        319,
+                        505,
+                        355
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        358,
+                        505,
+                        414
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                358,
+                                505,
+                                371
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        359,
+                                        317,
+                                        371
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The result of this operation is a network that is both",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        317,
+                                        358,
+                                        339,
+                                        371
+                                    ],
+                                    "score": 0.91,
+                                    "content": "1 / \\alpha ^ { 2 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        340,
+                                        359,
+                                        505,
+                                        371
+                                    ],
+                                    "score": 1.0,
+                                    "content": "smaller and faster. Many large networks",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                370,
+                                505,
+                                382
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        370,
+                                        505,
+                                        382
+                                    ],
+                                    "score": 1.0,
+                                    "content": "can be significantly reduced in size using this method with only a small loss of precision (Howard,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                380,
+                                505,
+                                394
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        380,
+                                        505,
+                                        394
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2017). It is our belief that this method establishes a strong baseline to which any other advanced",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                391,
+                                505,
+                                405
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        391,
+                                        505,
+                                        405
+                                    ],
+                                    "score": 1.0,
+                                    "content": "techniques should compare themselves. To the best of our knowledge, we are not aware of such",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                403,
+                                224,
+                                415
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        403,
+                                        224,
+                                        415
+                                    ],
+                                    "score": 1.0,
+                                    "content": "comparisons in the literature.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        }
+                    ],
+                    "index": 22,
+                    "bbox_fs": [
+                        105,
+                        358,
+                        505,
+                        415
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        424,
+                        207,
+                        435
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                423,
+                                208,
+                                437
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        423,
+                                        208,
+                                        437
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3.2 SPARSE RANDOM",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        }
+                    ],
+                    "index": 25
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        445,
+                        505,
+                        500
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                446,
+                                505,
+                                457
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        446,
+                                        505,
+                                        457
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Instead of looking at depth multiplier as deactivating channels in the convolutional layers, we can",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                456,
+                                505,
+                                468
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        456,
+                                        505,
+                                        468
+                                    ],
+                                    "score": 1.0,
+                                    "content": "look at it from the perspective of deactivating connections. From this point of view, depth multiplier",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                466,
+                                505,
+                                480
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        466,
+                                        505,
+                                        480
+                                    ],
+                                    "score": 1.0,
+                                    "content": "kills the connections between two convolutional layers such that (a) the connection patterns are",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                477,
+                                505,
+                                491
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        477,
+                                        505,
+                                        491
+                                    ],
+                                    "score": 1.0,
+                                    "content": "still the same across spatial dimensions and (b) all “alive” input channels are fully connected to all",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                488,
+                                205,
+                                502
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        488,
+                                        205,
+                                        502
+                                    ],
+                                    "score": 1.0,
+                                    "content": "“alive” output channels.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        }
+                    ],
+                    "index": 28,
+                    "bbox_fs": [
+                        105,
+                        446,
+                        505,
+                        502
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        506,
+                        505,
+                        583
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                506,
+                                505,
+                                519
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        506,
+                                        505,
+                                        519
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We generalize this approach by relaxing (b) while maintaining (a). That is, for every output chan-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                518,
+                                505,
+                                529
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        518,
+                                        505,
+                                        529
+                                    ],
+                                    "score": 1.0,
+                                    "content": "nel, we connect it to a small subset of input channels. In other words, dense connections between",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                528,
+                                505,
+                                540
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        528,
+                                        505,
+                                        540
+                                    ],
+                                    "score": 1.0,
+                                    "content": "a small number of channels become sparse connections between larger number of channels. This",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                539,
+                                505,
+                                551
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        539,
+                                        505,
+                                        551
+                                    ],
+                                    "score": 1.0,
+                                    "content": "can be summarized in Fig. 1. The advantage of this is that the actual convolution can still be com-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                550,
+                                505,
+                                562
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        550,
+                                        505,
+                                        562
+                                    ],
+                                    "score": 1.0,
+                                    "content": "puted efficiently because sparsity is introduced only at the outer loop of the convolution operation",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                561,
+                                505,
+                                574
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        561,
+                                        505,
+                                        574
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and we can still take the advantage of the continuous memory layout. For more details regarding",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                572,
+                                387,
+                                585
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        572,
+                                        387,
+                                        585
+                                    ],
+                                    "score": 1.0,
+                                    "content": "implementations of the two approaches, please refer to the Appendix.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        }
+                    ],
+                    "index": 34,
+                    "bbox_fs": [
+                        105,
+                        506,
+                        505,
+                        585
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        588,
+                        505,
+                        645
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                589,
+                                504,
+                                601
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        589,
+                                        189,
+                                        601
+                                    ],
+                                    "score": 1.0,
+                                    "content": "More concretely, let",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        189,
+                                        590,
+                                        210,
+                                        600
+                                    ],
+                                    "score": 0.9,
+                                    "content": "n _ { l - 1 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        210,
+                                        589,
+                                        228,
+                                        601
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        228,
+                                        590,
+                                        239,
+                                        600
+                                    ],
+                                    "score": 0.84,
+                                    "content": "n _ { l }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        239,
+                                        589,
+                                        381,
+                                        601
+                                    ],
+                                    "score": 1.0,
+                                    "content": "be the number of channels of layer",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        382,
+                                        589,
+                                        403,
+                                        599
+                                    ],
+                                    "score": 0.73,
+                                    "content": "l - 1",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        404,
+                                        589,
+                                        444,
+                                        601
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and layer",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        445,
+                                        590,
+                                        449,
+                                        599
+                                    ],
+                                    "score": 0.49,
+                                    "content": "l",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        450,
+                                        589,
+                                        504,
+                                        601
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", respectively.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                600,
+                                505,
+                                612
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        600,
+                                        213,
+                                        612
+                                    ],
+                                    "score": 1.0,
+                                    "content": "For a sparsity coefficient",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        213,
+                                        602,
+                                        221,
+                                        610
+                                    ],
+                                    "score": 0.74,
+                                    "content": "\\alpha",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        221,
+                                        600,
+                                        300,
+                                        612
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", each output filter",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        300,
+                                        601,
+                                        307,
+                                        612
+                                    ],
+                                    "score": 0.82,
+                                    "content": "j",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        307,
+                                        600,
+                                        394,
+                                        612
+                                    ],
+                                    "score": 1.0,
+                                    "content": "only connects to an",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        394,
+                                        602,
+                                        402,
+                                        610
+                                    ],
+                                    "score": 0.77,
+                                    "content": "\\alpha",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        403,
+                                        600,
+                                        505,
+                                        612
+                                    ],
+                                    "score": 1.0,
+                                    "content": "fraction of filters of the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                609,
+                                506,
+                                624
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        609,
+                                        356,
+                                        624
+                                    ],
+                                    "score": 1.0,
+                                    "content": "previous layer. Thus, instead of having a connectivity matrix",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        357,
+                                        611,
+                                        378,
+                                        623
+                                    ],
+                                    "score": 0.9,
+                                    "content": "W _ { s i j }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        379,
+                                        609,
+                                        436,
+                                        624
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of dimension",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        436,
+                                        610,
+                                        501,
+                                        622
+                                    ],
+                                    "score": 0.91,
+                                    "content": "k ^ { 2 } \\times n _ { l - 1 } \\times n _ { l }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        501,
+                                        609,
+                                        506,
+                                        624
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                620,
+                                506,
+                                637
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        620,
+                                        301,
+                                        637
+                                    ],
+                                    "score": 1.0,
+                                    "content": "we have a sparse matrix with non-zero entries at",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        302,
+                                        622,
+                                        330,
+                                        634
+                                    ],
+                                    "score": 0.92,
+                                    "content": "W _ { s a _ { i j } j }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        331,
+                                        620,
+                                        361,
+                                        637
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", where",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        361,
+                                        624,
+                                        375,
+                                        634
+                                    ],
+                                    "score": 0.88,
+                                    "content": "a _ { i j }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        375,
+                                        620,
+                                        506,
+                                        637
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is an index matrix of dimension",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                632,
+                                275,
+                                648
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        633,
+                                        176,
+                                        646
+                                    ],
+                                    "score": 0.93,
+                                    "content": "k ^ { 2 } \\times \\alpha n _ { l - 1 } \\times n _ { l }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        177,
+                                        632,
+                                        195,
+                                        648
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        195,
+                                        634,
+                                        202,
+                                        644
+                                    ],
+                                    "score": 0.8,
+                                    "content": "k",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        202,
+                                        632,
+                                        275,
+                                        648
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is the kernel size.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        }
+                    ],
+                    "index": 40,
+                    "bbox_fs": [
+                        105,
+                        589,
+                        506,
+                        648
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        655,
+                        247,
+                        667
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                655,
+                                249,
+                                668
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        655,
+                                        249,
+                                        668
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3.2.1 INCREMENTAL TRAINING",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        }
+                    ],
+                    "index": 43
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        676,
+                        504,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                677,
+                                506,
+                                689
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        677,
+                                        506,
+                                        689
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In contrast to depth multiplier, a sparse convolutional network defines a connection pattern on a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                687,
+                                506,
+                                700
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        687,
+                                        506,
+                                        700
+                                    ],
+                                    "score": 1.0,
+                                    "content": "much bigger network. Therefore, an interesting extension is to consider incremental training: we",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                698,
+                                505,
+                                712
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        698,
+                                        505,
+                                        712
+                                    ],
+                                    "score": 1.0,
+                                    "content": "start with a network that only contains a small fraction of connections (in our experiments we use",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                708,
+                                506,
+                                723
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        710,
+                                        121,
+                                        720
+                                    ],
+                                    "score": 0.86,
+                                    "content": "1 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        122,
+                                        708,
+                                        138,
+                                        723
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        139,
+                                        710,
+                                        162,
+                                        721
+                                    ],
+                                    "score": 0.85,
+                                    "content": "0 . 1 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        162,
+                                        708,
+                                        506,
+                                        723
+                                    ],
+                                    "score": 1.0,
+                                    "content": ") and add connections over time. This is motivated by an intuition that the network can",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 47
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                721,
+                                505,
+                                732
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        721,
+                                        505,
+                                        732
+                                    ],
+                                    "score": 1.0,
+                                    "content": "use learned channels in new contexts by introducing additional connections. The potential practical",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 48
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                224,
+                                504,
+                                236
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        224,
+                                        504,
+                                        236
+                                    ],
+                                    "score": 1.0,
+                                    "content": "advantage of this approach is that since we start training with very small networks and grow them",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                236,
+                                505,
+                                248
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        236,
+                                        505,
+                                        248
+                                    ],
+                                    "score": 1.0,
+                                    "content": "over time, this approach has a potential to speed up the whole training process significantly. We note",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                246,
+                                505,
+                                258
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        246,
+                                        505,
+                                        258
+                                    ],
+                                    "score": 1.0,
+                                    "content": "that depth multiplier will not benefit from this approach as any newly activated connections would",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                258,
+                                271,
+                                269
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        258,
+                                        271,
+                                        269
+                                    ],
+                                    "score": 1.0,
+                                    "content": "require learning new filters from scratch.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 11
+                        }
+                    ],
+                    "index": 46,
+                    "bbox_fs": [
+                        105,
+                        677,
+                        506,
+                        732
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "image",
+                    "bbox": [
+                        125,
+                        84,
+                        479,
+                        147
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                125,
+                                84,
+                                479,
+                                147
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        125,
+                                        84,
+                                        479,
+                                        147
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                125,
+                                                84,
+                                                479,
+                                                147
+                                            ],
+                                            "score": 0.956,
+                                            "type": "image",
+                                            "image_path": "66fc8a8512a12ff04f8f7dc6aab17648934eff7a1161c2536e3fa54b1fa83cde.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 1,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        125,
+                                        84,
+                                        479,
+                                        105.0
+                                    ],
+                                    "spans": [],
+                                    "index": 0
+                                },
+                                {
+                                    "bbox": [
+                                        125,
+                                        105.0,
+                                        479,
+                                        126.0
+                                    ],
+                                    "spans": [],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        125,
+                                        126.0,
+                                        479,
+                                        147.0
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                106,
+                                149,
+                                506,
+                                204
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        149,
+                                        505,
+                                        160
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                149,
+                                                505,
+                                                160
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 1: Connection tensors of depth multiplier (left) and sparse random (right) approaches for",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 3
+                                },
+                                {
+                                    "bbox": [
+                                        107,
+                                        159,
+                                        505,
+                                        172
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                107,
+                                                160,
+                                                148,
+                                                171
+                                            ],
+                                            "score": 0.91,
+                                            "content": "n _ { l - 1 } = 5",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                148,
+                                                159,
+                                                167,
+                                                172
+                                            ],
+                                            "score": 1.0,
+                                            "content": "and",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                167,
+                                                160,
+                                                203,
+                                                171
+                                            ],
+                                            "score": 0.91,
+                                            "content": "n _ { l } = 1 0",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                203,
+                                                159,
+                                                505,
+                                                172
+                                            ],
+                                            "score": 1.0,
+                                            "content": ". Yellow denotes active connections. For both approaches, the connection",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 4
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        170,
+                                        505,
+                                        183
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                170,
+                                                505,
+                                                183
+                                            ],
+                                            "score": 1.0,
+                                            "content": "pattern is the same across spatial dimension and fixed before training. However, in the sparse random",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 5
+                                },
+                                {
+                                    "bbox": [
+                                        104,
+                                        181,
+                                        506,
+                                        195
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                104,
+                                                181,
+                                                506,
+                                                195
+                                            ],
+                                            "score": 1.0,
+                                            "content": "approach, each output channel is connected to a (possibly) different subset of input channels, and",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 6
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        194,
+                                        151,
+                                        204
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                194,
+                                                151,
+                                                204
+                                            ],
+                                            "score": 1.0,
+                                            "content": "vice versa.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 7
+                                }
+                            ],
+                            "index": 5
+                        }
+                    ],
+                    "index": 3.0
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        223,
+                        505,
+                        268
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                224,
+                                504,
+                                236
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        224,
+                                        504,
+                                        236
+                                    ],
+                                    "score": 1.0,
+                                    "content": "advantage of this approach is that since we start training with very small networks and grow them",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                236,
+                                505,
+                                248
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        236,
+                                        505,
+                                        248
+                                    ],
+                                    "score": 1.0,
+                                    "content": "over time, this approach has a potential to speed up the whole training process significantly. We note",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                246,
+                                505,
+                                258
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        246,
+                                        505,
+                                        258
+                                    ],
+                                    "score": 1.0,
+                                    "content": "that depth multiplier will not benefit from this approach as any newly activated connections would",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                258,
+                                271,
+                                269
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        258,
+                                        271,
+                                        269
+                                    ],
+                                    "score": 1.0,
+                                    "content": "require learning new filters from scratch.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        }
+                    ],
+                    "index": 9.5
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        282,
+                        178,
+                        294
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                280,
+                                180,
+                                297
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        280,
+                                        180,
+                                        297
+                                    ],
+                                    "score": 1.0,
+                                    "content": "4 ANALYSIS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        }
+                    ],
+                    "index": 12
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        306,
+                        505,
+                        373
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                307,
+                                505,
+                                318
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        307,
+                                        505,
+                                        318
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In this section, we approach a question of why sparse convolutions are frequently more efficient than",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                317,
+                                505,
+                                330
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        317,
+                                        505,
+                                        330
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the dense convolutions with the same number of parameters. Our main intuition is that the sparse",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                329,
+                                505,
+                                340
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        329,
+                                        505,
+                                        340
+                                    ],
+                                    "score": 1.0,
+                                    "content": "convolutional networks promote diversity. It is much harder to learn equivalent set of channels as, at",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                339,
+                                506,
+                                352
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        339,
+                                        506,
+                                        352
+                                    ],
+                                    "score": 1.0,
+                                    "content": "high sparsity, channels have distinct connection structure or even overlapping connections. This can",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                349,
+                                505,
+                                363
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        349,
+                                        505,
+                                        363
+                                    ],
+                                    "score": 1.0,
+                                    "content": "be formalized with a simple observation that any dense network is in fact a part of an exponentially",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                362,
+                                459,
+                                373
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        362,
+                                        459,
+                                        373
+                                    ],
+                                    "score": 1.0,
+                                    "content": "large equivalence class, which is guaranteed to produce the same output for every input.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        }
+                    ],
+                    "index": 15.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        384,
+                        505,
+                        420
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                384,
+                                505,
+                                397
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        384,
+                                        505,
+                                        397
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Lemma 1 Any dense convolutional neural network with no cross-channel nonlinearities, distinct",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                395,
+                                505,
+                                411
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        395,
+                                        321,
+                                        411
+                                    ],
+                                    "score": 1.0,
+                                    "content": "weights and biases, and with l hidden layers of sizes",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        321,
+                                        399,
+                                        332,
+                                        408
+                                    ],
+                                    "score": 0.75,
+                                    "content": "n _ { 1 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        333,
+                                        395,
+                                        336,
+                                        411
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        337,
+                                        399,
+                                        348,
+                                        408
+                                    ],
+                                    "score": 0.62,
+                                    "content": "n _ { 2 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        349,
+                                        395,
+                                        370,
+                                        411
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", . . . ,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        370,
+                                        399,
+                                        381,
+                                        408
+                                    ],
+                                    "score": 0.68,
+                                    "content": "n _ { l }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        381,
+                                        395,
+                                        433,
+                                        411
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", has at least",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        434,
+                                        395,
+                                        472,
+                                        410
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\textstyle \\prod _ { i = 1 } ^ { l } n _ { i } !",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        472,
+                                        395,
+                                        505,
+                                        411
+                                    ],
+                                    "score": 1.0,
+                                    "content": "distinct",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                408,
+                                317,
+                                421
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        408,
+                                        317,
+                                        421
+                                    ],
+                                    "score": 1.0,
+                                    "content": "equivalent networks which produce the same output.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        }
+                    ],
+                    "index": 20
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        430,
+                        505,
+                        486
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                429,
+                                505,
+                                443
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        429,
+                                        149,
+                                        443
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Proof Let",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        150,
+                                        431,
+                                        156,
+                                        441
+                                    ],
+                                    "score": 0.74,
+                                    "content": "I",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        157,
+                                        429,
+                                        290,
+                                        443
+                                    ],
+                                    "score": 1.0,
+                                    "content": "denote the input to the network,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        291,
+                                        431,
+                                        302,
+                                        442
+                                    ],
+                                    "score": 0.87,
+                                    "content": "C _ { i }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        302,
+                                        429,
+                                        427,
+                                        443
+                                    ],
+                                    "score": 1.0,
+                                    "content": "be the convolutional operator,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        427,
+                                        433,
+                                        438,
+                                        442
+                                    ],
+                                    "score": 0.85,
+                                    "content": "\\sigma _ { i }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        438,
+                                        429,
+                                        505,
+                                        443
+                                    ],
+                                    "score": 1.0,
+                                    "content": "denote the non-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                441,
+                                505,
+                                454
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        441,
+                                        234,
+                                        454
+                                    ],
+                                    "score": 1.0,
+                                    "content": "linearity operator applied to the",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        235,
+                                        442,
+                                        239,
+                                        451
+                                    ],
+                                    "score": 0.72,
+                                    "content": "i",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        240,
+                                        441,
+                                        340,
+                                        454
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-th convolution layer and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        340,
+                                        442,
+                                        348,
+                                        452
+                                    ],
+                                    "score": 0.81,
+                                    "content": "S",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        349,
+                                        441,
+                                        505,
+                                        454
+                                    ],
+                                    "score": 1.0,
+                                    "content": "be a final transformation (e.g. softmax",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                452,
+                                505,
+                                465
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        452,
+                                        215,
+                                        465
+                                    ],
+                                    "score": 1.0,
+                                    "content": "classifier). We assume that",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        216,
+                                        454,
+                                        226,
+                                        464
+                                    ],
+                                    "score": 0.86,
+                                    "content": "\\sigma _ { i }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        226,
+                                        452,
+                                        505,
+                                        465
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is a function that operates on each of the channels independently. We",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                463,
+                                505,
+                                476
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        463,
+                                        505,
+                                        476
+                                    ],
+                                    "score": 1.0,
+                                    "content": "note that this is the case for almost any modern network. The output of the network can then be",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                474,
+                                152,
+                                487
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        474,
+                                        152,
+                                        487
+                                    ],
+                                    "score": 1.0,
+                                    "content": "written as:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        }
+                    ],
+                    "index": 24
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        212,
+                        489,
+                        398,
+                        503
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                212,
+                                489,
+                                398,
+                                503
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        212,
+                                        489,
+                                        398,
+                                        503
+                                    ],
+                                    "score": 0.89,
+                                    "content": "\\mathcal { N } ( I ) \\equiv S \\circ \\sigma _ { l } \\circ C _ { l } \\circ \\sigma _ { l - 1 } \\circ \\cdots \\circ \\sigma _ { 1 } \\circ C _ { 1 } ( I )",
+                                    "type": "interline_equation",
+                                    "image_path": "beea7d489507a37291c18e46788f828e6040412e7e84a7604ead6cd1d103d3a8.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 27,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                212,
+                                489,
+                                398,
+                                503
+                            ],
+                            "spans": [],
+                            "index": 27
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        510,
+                        505,
+                        577
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                511,
+                                505,
+                                523
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        511,
+                                        165,
+                                        523
+                                    ],
+                                    "score": 1.0,
+                                    "content": "where we use",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        165,
+                                        513,
+                                        172,
+                                        521
+                                    ],
+                                    "score": 0.68,
+                                    "content": "\\circ",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        173,
+                                        511,
+                                        505,
+                                        523
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to denote function composition to avoid numerous parentheses. The convolution",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                522,
+                                505,
+                                534
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        522,
+                                        142,
+                                        534
+                                    ],
+                                    "score": 1.0,
+                                    "content": "operator",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        143,
+                                        523,
+                                        154,
+                                        533
+                                    ],
+                                    "score": 0.87,
+                                    "content": "C _ { i }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        155,
+                                        522,
+                                        249,
+                                        534
+                                    ],
+                                    "score": 1.0,
+                                    "content": "operates on input with",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        249,
+                                        523,
+                                        270,
+                                        534
+                                    ],
+                                    "score": 0.88,
+                                    "content": "n _ { i - 1 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        270,
+                                        522,
+                                        427,
+                                        534
+                                    ],
+                                    "score": 1.0,
+                                    "content": "channels and produces an output with",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        427,
+                                        523,
+                                        438,
+                                        533
+                                    ],
+                                    "score": 0.85,
+                                    "content": "n _ { i }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        438,
+                                        522,
+                                        505,
+                                        534
+                                    ],
+                                    "score": 1.0,
+                                    "content": "channels. Now,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                533,
+                                506,
+                                546
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        533,
+                                        276,
+                                        546
+                                    ],
+                                    "score": 1.0,
+                                    "content": "fix arbitrary set of permutation functions",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        276,
+                                        535,
+                                        286,
+                                        543
+                                    ],
+                                    "score": 0.79,
+                                    "content": "\\pi _ { i }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        286,
+                                        533,
+                                        318,
+                                        546
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", where",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        319,
+                                        535,
+                                        329,
+                                        544
+                                    ],
+                                    "score": 0.85,
+                                    "content": "\\pi _ { i }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        330,
+                                        533,
+                                        440,
+                                        546
+                                    ],
+                                    "score": 1.0,
+                                    "content": "can permute depth of size",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        441,
+                                        535,
+                                        451,
+                                        544
+                                    ],
+                                    "score": 0.85,
+                                    "content": "n _ { i }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        451,
+                                        533,
+                                        483,
+                                        546
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Since",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        483,
+                                        534,
+                                        494,
+                                        544
+                                    ],
+                                    "score": 0.85,
+                                    "content": "\\pi _ { i }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        494,
+                                        533,
+                                        506,
+                                        546
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                544,
+                                505,
+                                557
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        544,
+                                        238,
+                                        557
+                                    ],
+                                    "score": 1.0,
+                                    "content": "a linear function, it follows that",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        238,
+                                        544,
+                                        311,
+                                        556
+                                    ],
+                                    "score": 0.92,
+                                    "content": "C _ { i } ^ { \\prime } = \\pi _ { i } ^ { - 1 } C _ { i } \\pi _ { i - 1 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        311,
+                                        544,
+                                        505,
+                                        557
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is a valid convolutional operator, which can be",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                554,
+                                506,
+                                567
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        554,
+                                        165,
+                                        567
+                                    ],
+                                    "score": 1.0,
+                                    "content": "obtained from",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        165,
+                                        556,
+                                        176,
+                                        566
+                                    ],
+                                    "score": 0.87,
+                                    "content": "C _ { i }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        177,
+                                        554,
+                                        315,
+                                        567
+                                    ],
+                                    "score": 1.0,
+                                    "content": "by permuting its bias according to",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        315,
+                                        556,
+                                        325,
+                                        566
+                                    ],
+                                    "score": 0.86,
+                                    "content": "\\pi _ { i }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        326,
+                                        554,
+                                        506,
+                                        567
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and its weight matrix along input and output",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                565,
+                                432,
+                                578
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        565,
+                                        206,
+                                        578
+                                    ],
+                                    "score": 1.0,
+                                    "content": "dimensions according to",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        207,
+                                        567,
+                                        227,
+                                        578
+                                    ],
+                                    "score": 0.89,
+                                    "content": "\\pi _ { i - 1 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        227,
+                                        565,
+                                        245,
+                                        578
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        245,
+                                        567,
+                                        255,
+                                        577
+                                    ],
+                                    "score": 0.85,
+                                    "content": "\\pi _ { i }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        255,
+                                        565,
+                                        432,
+                                        578
+                                    ],
+                                    "score": 1.0,
+                                    "content": "respectively. For a new network defined as:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        }
+                    ],
+                    "index": 30.5
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        208,
+                        586,
+                        402,
+                        600
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                208,
+                                586,
+                                402,
+                                600
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        208,
+                                        586,
+                                        402,
+                                        600
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\mathcal { N } ^ { \\prime } ( I ) = S ^ { \\prime } \\circ \\sigma _ { l } \\circ C _ { l } ^ { \\prime } \\circ \\sigma _ { l - 1 } \\circ \\cdots \\circ \\sigma _ { 1 } \\circ C _ { 1 } ^ { \\prime } ( I ) ,",
+                                    "type": "interline_equation",
+                                    "image_path": "f9d507d6a130c4c9052e073184e7bcd8a6aa91888fa306ef72272247738d2624.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 34,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                208,
+                                586,
+                                402,
+                                600
+                            ],
+                            "spans": [],
+                            "index": 34
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        608,
+                        504,
+                        632
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                608,
+                                506,
+                                621
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        608,
+                                        134,
+                                        621
+                                    ],
+                                    "score": 1.0,
+                                    "content": "where",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        134,
+                                        611,
+                                        146,
+                                        620
+                                    ],
+                                    "score": 0.85,
+                                    "content": "\\pi _ { 0 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        146,
+                                        608,
+                                        259,
+                                        621
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is an identity operator and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        259,
+                                        609,
+                                        312,
+                                        620
+                                    ],
+                                    "score": 0.91,
+                                    "content": "S ^ { \\prime } \\equiv S \\circ \\pi _ { l }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        312,
+                                        608,
+                                        376,
+                                        621
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", we claim that",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        376,
+                                        608,
+                                        442,
+                                        621
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\mathcal { N } ^ { \\prime } ( I ) \\equiv \\mathcal { N } ( I )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        442,
+                                        608,
+                                        506,
+                                        621
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Indeed, since",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                618,
+                                406,
+                                634
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        618,
+                                        299,
+                                        634
+                                    ],
+                                    "score": 1.0,
+                                    "content": "nonlinearities do not apply cross-depth we have",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        299,
+                                        619,
+                                        363,
+                                        632
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\pi _ { n } \\sigma _ { n } \\pi _ { n } ^ { - 1 } \\equiv \\sigma _ { n }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        363,
+                                        618,
+                                        406,
+                                        634
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and thus:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        }
+                    ],
+                    "index": 35.5
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        115,
+                        641,
+                        495,
+                        675
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                115,
+                                641,
+                                495,
+                                675
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        641,
+                                        495,
+                                        675
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\begin{array} { r l } & { \\mathscr { N } ^ { \\prime } ( I ) = S ^ { \\prime } \\circ \\sigma _ { l } \\circ C _ { l } ^ { \\prime } \\circ \\sigma _ { l - 1 } \\circ \\cdots \\circ \\sigma _ { 1 } \\circ C _ { 1 } ^ { \\prime } ( I ) = } \\\\ & { \\qquad = S \\circ \\pi _ { l } \\circ \\sigma _ { l } \\circ \\pi _ { l } ^ { - 1 } \\circ C _ { l } \\circ \\pi _ { l - 1 } \\circ \\cdots \\circ \\pi _ { 1 } \\circ \\sigma _ { 1 } \\circ \\pi _ { 1 } ^ { - 1 } \\circ C _ { 1 } ( I ) = \\mathscr { N } ( I ) . } \\end{array}",
+                                    "type": "interline_equation",
+                                    "image_path": "f024c18d6595d3e2ba9aa755d9d14deecbf1af5c05c6f9f759b65544ecec492d.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 38,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                115,
+                                641,
+                                495,
+                                652.3333333333334
+                            ],
+                            "spans": [],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                652.3333333333334,
+                                495,
+                                663.6666666666667
+                            ],
+                            "spans": [],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                663.6666666666667,
+                                495,
+                                675.0000000000001
+                            ],
+                            "spans": [],
+                            "index": 39
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        681,
+                        414,
+                        693
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                680,
+                                415,
+                                695
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        680,
+                                        415,
+                                        695
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Thus, any set of permutations on hidden units defines an equivalent network.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        }
+                    ],
+                    "index": 40
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        698,
+                        504,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                698,
+                                505,
+                                712
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        698,
+                                        505,
+                                        712
+                                    ],
+                                    "score": 1.0,
+                                    "content": "It is obvious that sparse networks are much more immune to parameter permutation – indeed every",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                709,
+                                506,
+                                723
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        709,
+                                        173,
+                                        723
+                                    ],
+                                    "score": 1.0,
+                                    "content": "channel at layer",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        173,
+                                        711,
+                                        177,
+                                        720
+                                    ],
+                                    "score": 0.66,
+                                    "content": "l",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        178,
+                                        709,
+                                        506,
+                                        723
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is likely to have a unique tree describing its connection matrix all the way down.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                720,
+                                329,
+                                733
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        720,
+                                        329,
+                                        733
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Exploring this direction is an interesting open question.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        }
+                    ],
+                    "index": 42
+                }
+            ],
+            "page_idx": 4,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        106,
+                        27,
+                        308,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                26,
+                                308,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        26,
+                                        308,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Under review as a conference paper at ICLR 2017",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        495,
+                        682,
+                        504,
+                        691
+                    ],
+                    "lines": []
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        302,
+                        751,
+                        308,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                302,
+                                750,
+                                309,
+                                763
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        302,
+                                        750,
+                                        309,
+                                        763
+                                    ],
+                                    "score": 1.0,
+                                    "content": "5",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "image",
+                    "bbox": [
+                        125,
+                        84,
+                        479,
+                        147
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                125,
+                                84,
+                                479,
+                                147
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        125,
+                                        84,
+                                        479,
+                                        147
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                125,
+                                                84,
+                                                479,
+                                                147
+                                            ],
+                                            "score": 0.956,
+                                            "type": "image",
+                                            "image_path": "66fc8a8512a12ff04f8f7dc6aab17648934eff7a1161c2536e3fa54b1fa83cde.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 1,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        125,
+                                        84,
+                                        479,
+                                        105.0
+                                    ],
+                                    "spans": [],
+                                    "index": 0
+                                },
+                                {
+                                    "bbox": [
+                                        125,
+                                        105.0,
+                                        479,
+                                        126.0
+                                    ],
+                                    "spans": [],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        125,
+                                        126.0,
+                                        479,
+                                        147.0
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                106,
+                                149,
+                                506,
+                                204
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        149,
+                                        505,
+                                        160
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                149,
+                                                505,
+                                                160
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 1: Connection tensors of depth multiplier (left) and sparse random (right) approaches for",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 3
+                                },
+                                {
+                                    "bbox": [
+                                        107,
+                                        159,
+                                        505,
+                                        172
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                107,
+                                                160,
+                                                148,
+                                                171
+                                            ],
+                                            "score": 0.91,
+                                            "content": "n _ { l - 1 } = 5",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                148,
+                                                159,
+                                                167,
+                                                172
+                                            ],
+                                            "score": 1.0,
+                                            "content": "and",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                167,
+                                                160,
+                                                203,
+                                                171
+                                            ],
+                                            "score": 0.91,
+                                            "content": "n _ { l } = 1 0",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                203,
+                                                159,
+                                                505,
+                                                172
+                                            ],
+                                            "score": 1.0,
+                                            "content": ". Yellow denotes active connections. For both approaches, the connection",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 4
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        170,
+                                        505,
+                                        183
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                170,
+                                                505,
+                                                183
+                                            ],
+                                            "score": 1.0,
+                                            "content": "pattern is the same across spatial dimension and fixed before training. However, in the sparse random",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 5
+                                },
+                                {
+                                    "bbox": [
+                                        104,
+                                        181,
+                                        506,
+                                        195
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                104,
+                                                181,
+                                                506,
+                                                195
+                                            ],
+                                            "score": 1.0,
+                                            "content": "approach, each output channel is connected to a (possibly) different subset of input channels, and",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 6
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        194,
+                                        151,
+                                        204
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                194,
+                                                151,
+                                                204
+                                            ],
+                                            "score": 1.0,
+                                            "content": "vice versa.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 7
+                                }
+                            ],
+                            "index": 5
+                        }
+                    ],
+                    "index": 3.0
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        223,
+                        505,
+                        268
+                    ],
+                    "lines": [],
+                    "index": 9.5,
+                    "bbox_fs": [
+                        105,
+                        224,
+                        505,
+                        269
+                    ],
+                    "lines_deleted": true
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        282,
+                        178,
+                        294
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                280,
+                                180,
+                                297
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        280,
+                                        180,
+                                        297
+                                    ],
+                                    "score": 1.0,
+                                    "content": "4 ANALYSIS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        }
+                    ],
+                    "index": 12
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        306,
+                        505,
+                        373
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                307,
+                                505,
+                                318
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        307,
+                                        505,
+                                        318
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In this section, we approach a question of why sparse convolutions are frequently more efficient than",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                317,
+                                505,
+                                330
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        317,
+                                        505,
+                                        330
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the dense convolutions with the same number of parameters. Our main intuition is that the sparse",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                329,
+                                505,
+                                340
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        329,
+                                        505,
+                                        340
+                                    ],
+                                    "score": 1.0,
+                                    "content": "convolutional networks promote diversity. It is much harder to learn equivalent set of channels as, at",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                339,
+                                506,
+                                352
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        339,
+                                        506,
+                                        352
+                                    ],
+                                    "score": 1.0,
+                                    "content": "high sparsity, channels have distinct connection structure or even overlapping connections. This can",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                349,
+                                505,
+                                363
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        349,
+                                        505,
+                                        363
+                                    ],
+                                    "score": 1.0,
+                                    "content": "be formalized with a simple observation that any dense network is in fact a part of an exponentially",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                362,
+                                459,
+                                373
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        362,
+                                        459,
+                                        373
+                                    ],
+                                    "score": 1.0,
+                                    "content": "large equivalence class, which is guaranteed to produce the same output for every input.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        }
+                    ],
+                    "index": 15.5,
+                    "bbox_fs": [
+                        105,
+                        307,
+                        506,
+                        373
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        384,
+                        505,
+                        420
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                384,
+                                505,
+                                397
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        384,
+                                        505,
+                                        397
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Lemma 1 Any dense convolutional neural network with no cross-channel nonlinearities, distinct",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                395,
+                                505,
+                                411
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        395,
+                                        321,
+                                        411
+                                    ],
+                                    "score": 1.0,
+                                    "content": "weights and biases, and with l hidden layers of sizes",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        321,
+                                        399,
+                                        332,
+                                        408
+                                    ],
+                                    "score": 0.75,
+                                    "content": "n _ { 1 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        333,
+                                        395,
+                                        336,
+                                        411
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        337,
+                                        399,
+                                        348,
+                                        408
+                                    ],
+                                    "score": 0.62,
+                                    "content": "n _ { 2 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        349,
+                                        395,
+                                        370,
+                                        411
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", . . . ,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        370,
+                                        399,
+                                        381,
+                                        408
+                                    ],
+                                    "score": 0.68,
+                                    "content": "n _ { l }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        381,
+                                        395,
+                                        433,
+                                        411
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", has at least",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        434,
+                                        395,
+                                        472,
+                                        410
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\textstyle \\prod _ { i = 1 } ^ { l } n _ { i } !",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        472,
+                                        395,
+                                        505,
+                                        411
+                                    ],
+                                    "score": 1.0,
+                                    "content": "distinct",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                408,
+                                317,
+                                421
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        408,
+                                        317,
+                                        421
+                                    ],
+                                    "score": 1.0,
+                                    "content": "equivalent networks which produce the same output.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        }
+                    ],
+                    "index": 20,
+                    "bbox_fs": [
+                        105,
+                        384,
+                        505,
+                        421
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        430,
+                        505,
+                        486
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                429,
+                                505,
+                                443
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        429,
+                                        149,
+                                        443
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Proof Let",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        150,
+                                        431,
+                                        156,
+                                        441
+                                    ],
+                                    "score": 0.74,
+                                    "content": "I",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        157,
+                                        429,
+                                        290,
+                                        443
+                                    ],
+                                    "score": 1.0,
+                                    "content": "denote the input to the network,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        291,
+                                        431,
+                                        302,
+                                        442
+                                    ],
+                                    "score": 0.87,
+                                    "content": "C _ { i }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        302,
+                                        429,
+                                        427,
+                                        443
+                                    ],
+                                    "score": 1.0,
+                                    "content": "be the convolutional operator,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        427,
+                                        433,
+                                        438,
+                                        442
+                                    ],
+                                    "score": 0.85,
+                                    "content": "\\sigma _ { i }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        438,
+                                        429,
+                                        505,
+                                        443
+                                    ],
+                                    "score": 1.0,
+                                    "content": "denote the non-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                441,
+                                505,
+                                454
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        441,
+                                        234,
+                                        454
+                                    ],
+                                    "score": 1.0,
+                                    "content": "linearity operator applied to the",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        235,
+                                        442,
+                                        239,
+                                        451
+                                    ],
+                                    "score": 0.72,
+                                    "content": "i",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        240,
+                                        441,
+                                        340,
+                                        454
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-th convolution layer and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        340,
+                                        442,
+                                        348,
+                                        452
+                                    ],
+                                    "score": 0.81,
+                                    "content": "S",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        349,
+                                        441,
+                                        505,
+                                        454
+                                    ],
+                                    "score": 1.0,
+                                    "content": "be a final transformation (e.g. softmax",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                452,
+                                505,
+                                465
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        452,
+                                        215,
+                                        465
+                                    ],
+                                    "score": 1.0,
+                                    "content": "classifier). We assume that",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        216,
+                                        454,
+                                        226,
+                                        464
+                                    ],
+                                    "score": 0.86,
+                                    "content": "\\sigma _ { i }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        226,
+                                        452,
+                                        505,
+                                        465
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is a function that operates on each of the channels independently. We",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                463,
+                                505,
+                                476
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        463,
+                                        505,
+                                        476
+                                    ],
+                                    "score": 1.0,
+                                    "content": "note that this is the case for almost any modern network. The output of the network can then be",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                474,
+                                152,
+                                487
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        474,
+                                        152,
+                                        487
+                                    ],
+                                    "score": 1.0,
+                                    "content": "written as:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        }
+                    ],
+                    "index": 24,
+                    "bbox_fs": [
+                        105,
+                        429,
+                        505,
+                        487
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        212,
+                        489,
+                        398,
+                        503
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                212,
+                                489,
+                                398,
+                                503
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        212,
+                                        489,
+                                        398,
+                                        503
+                                    ],
+                                    "score": 0.89,
+                                    "content": "\\mathcal { N } ( I ) \\equiv S \\circ \\sigma _ { l } \\circ C _ { l } \\circ \\sigma _ { l - 1 } \\circ \\cdots \\circ \\sigma _ { 1 } \\circ C _ { 1 } ( I )",
+                                    "type": "interline_equation",
+                                    "image_path": "beea7d489507a37291c18e46788f828e6040412e7e84a7604ead6cd1d103d3a8.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 27,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                212,
+                                489,
+                                398,
+                                503
+                            ],
+                            "spans": [],
+                            "index": 27
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        510,
+                        505,
+                        577
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                511,
+                                505,
+                                523
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        511,
+                                        165,
+                                        523
+                                    ],
+                                    "score": 1.0,
+                                    "content": "where we use",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        165,
+                                        513,
+                                        172,
+                                        521
+                                    ],
+                                    "score": 0.68,
+                                    "content": "\\circ",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        173,
+                                        511,
+                                        505,
+                                        523
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to denote function composition to avoid numerous parentheses. The convolution",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                522,
+                                505,
+                                534
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        522,
+                                        142,
+                                        534
+                                    ],
+                                    "score": 1.0,
+                                    "content": "operator",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        143,
+                                        523,
+                                        154,
+                                        533
+                                    ],
+                                    "score": 0.87,
+                                    "content": "C _ { i }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        155,
+                                        522,
+                                        249,
+                                        534
+                                    ],
+                                    "score": 1.0,
+                                    "content": "operates on input with",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        249,
+                                        523,
+                                        270,
+                                        534
+                                    ],
+                                    "score": 0.88,
+                                    "content": "n _ { i - 1 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        270,
+                                        522,
+                                        427,
+                                        534
+                                    ],
+                                    "score": 1.0,
+                                    "content": "channels and produces an output with",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        427,
+                                        523,
+                                        438,
+                                        533
+                                    ],
+                                    "score": 0.85,
+                                    "content": "n _ { i }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        438,
+                                        522,
+                                        505,
+                                        534
+                                    ],
+                                    "score": 1.0,
+                                    "content": "channels. Now,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                533,
+                                506,
+                                546
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        533,
+                                        276,
+                                        546
+                                    ],
+                                    "score": 1.0,
+                                    "content": "fix arbitrary set of permutation functions",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        276,
+                                        535,
+                                        286,
+                                        543
+                                    ],
+                                    "score": 0.79,
+                                    "content": "\\pi _ { i }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        286,
+                                        533,
+                                        318,
+                                        546
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", where",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        319,
+                                        535,
+                                        329,
+                                        544
+                                    ],
+                                    "score": 0.85,
+                                    "content": "\\pi _ { i }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        330,
+                                        533,
+                                        440,
+                                        546
+                                    ],
+                                    "score": 1.0,
+                                    "content": "can permute depth of size",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        441,
+                                        535,
+                                        451,
+                                        544
+                                    ],
+                                    "score": 0.85,
+                                    "content": "n _ { i }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        451,
+                                        533,
+                                        483,
+                                        546
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Since",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        483,
+                                        534,
+                                        494,
+                                        544
+                                    ],
+                                    "score": 0.85,
+                                    "content": "\\pi _ { i }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        494,
+                                        533,
+                                        506,
+                                        546
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                544,
+                                505,
+                                557
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        544,
+                                        238,
+                                        557
+                                    ],
+                                    "score": 1.0,
+                                    "content": "a linear function, it follows that",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        238,
+                                        544,
+                                        311,
+                                        556
+                                    ],
+                                    "score": 0.92,
+                                    "content": "C _ { i } ^ { \\prime } = \\pi _ { i } ^ { - 1 } C _ { i } \\pi _ { i - 1 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        311,
+                                        544,
+                                        505,
+                                        557
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is a valid convolutional operator, which can be",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                554,
+                                506,
+                                567
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        554,
+                                        165,
+                                        567
+                                    ],
+                                    "score": 1.0,
+                                    "content": "obtained from",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        165,
+                                        556,
+                                        176,
+                                        566
+                                    ],
+                                    "score": 0.87,
+                                    "content": "C _ { i }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        177,
+                                        554,
+                                        315,
+                                        567
+                                    ],
+                                    "score": 1.0,
+                                    "content": "by permuting its bias according to",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        315,
+                                        556,
+                                        325,
+                                        566
+                                    ],
+                                    "score": 0.86,
+                                    "content": "\\pi _ { i }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        326,
+                                        554,
+                                        506,
+                                        567
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and its weight matrix along input and output",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                565,
+                                432,
+                                578
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        565,
+                                        206,
+                                        578
+                                    ],
+                                    "score": 1.0,
+                                    "content": "dimensions according to",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        207,
+                                        567,
+                                        227,
+                                        578
+                                    ],
+                                    "score": 0.89,
+                                    "content": "\\pi _ { i - 1 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        227,
+                                        565,
+                                        245,
+                                        578
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        245,
+                                        567,
+                                        255,
+                                        577
+                                    ],
+                                    "score": 0.85,
+                                    "content": "\\pi _ { i }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        255,
+                                        565,
+                                        432,
+                                        578
+                                    ],
+                                    "score": 1.0,
+                                    "content": "respectively. For a new network defined as:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        }
+                    ],
+                    "index": 30.5,
+                    "bbox_fs": [
+                        105,
+                        511,
+                        506,
+                        578
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        208,
+                        586,
+                        402,
+                        600
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                208,
+                                586,
+                                402,
+                                600
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        208,
+                                        586,
+                                        402,
+                                        600
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\mathcal { N } ^ { \\prime } ( I ) = S ^ { \\prime } \\circ \\sigma _ { l } \\circ C _ { l } ^ { \\prime } \\circ \\sigma _ { l - 1 } \\circ \\cdots \\circ \\sigma _ { 1 } \\circ C _ { 1 } ^ { \\prime } ( I ) ,",
+                                    "type": "interline_equation",
+                                    "image_path": "f9d507d6a130c4c9052e073184e7bcd8a6aa91888fa306ef72272247738d2624.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 34,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                208,
+                                586,
+                                402,
+                                600
+                            ],
+                            "spans": [],
+                            "index": 34
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        608,
+                        504,
+                        632
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                608,
+                                506,
+                                621
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        608,
+                                        134,
+                                        621
+                                    ],
+                                    "score": 1.0,
+                                    "content": "where",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        134,
+                                        611,
+                                        146,
+                                        620
+                                    ],
+                                    "score": 0.85,
+                                    "content": "\\pi _ { 0 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        146,
+                                        608,
+                                        259,
+                                        621
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is an identity operator and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        259,
+                                        609,
+                                        312,
+                                        620
+                                    ],
+                                    "score": 0.91,
+                                    "content": "S ^ { \\prime } \\equiv S \\circ \\pi _ { l }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        312,
+                                        608,
+                                        376,
+                                        621
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", we claim that",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        376,
+                                        608,
+                                        442,
+                                        621
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\mathcal { N } ^ { \\prime } ( I ) \\equiv \\mathcal { N } ( I )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        442,
+                                        608,
+                                        506,
+                                        621
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Indeed, since",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                618,
+                                406,
+                                634
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        618,
+                                        299,
+                                        634
+                                    ],
+                                    "score": 1.0,
+                                    "content": "nonlinearities do not apply cross-depth we have",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        299,
+                                        619,
+                                        363,
+                                        632
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\pi _ { n } \\sigma _ { n } \\pi _ { n } ^ { - 1 } \\equiv \\sigma _ { n }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        363,
+                                        618,
+                                        406,
+                                        634
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and thus:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        }
+                    ],
+                    "index": 35.5,
+                    "bbox_fs": [
+                        104,
+                        608,
+                        506,
+                        634
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        115,
+                        641,
+                        495,
+                        675
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                115,
+                                641,
+                                495,
+                                675
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        641,
+                                        495,
+                                        675
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\begin{array} { r l } & { \\mathscr { N } ^ { \\prime } ( I ) = S ^ { \\prime } \\circ \\sigma _ { l } \\circ C _ { l } ^ { \\prime } \\circ \\sigma _ { l - 1 } \\circ \\cdots \\circ \\sigma _ { 1 } \\circ C _ { 1 } ^ { \\prime } ( I ) = } \\\\ & { \\qquad = S \\circ \\pi _ { l } \\circ \\sigma _ { l } \\circ \\pi _ { l } ^ { - 1 } \\circ C _ { l } \\circ \\pi _ { l - 1 } \\circ \\cdots \\circ \\pi _ { 1 } \\circ \\sigma _ { 1 } \\circ \\pi _ { 1 } ^ { - 1 } \\circ C _ { 1 } ( I ) = \\mathscr { N } ( I ) . } \\end{array}",
+                                    "type": "interline_equation",
+                                    "image_path": "f024c18d6595d3e2ba9aa755d9d14deecbf1af5c05c6f9f759b65544ecec492d.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 38,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                115,
+                                641,
+                                495,
+                                652.3333333333334
+                            ],
+                            "spans": [],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                652.3333333333334,
+                                495,
+                                663.6666666666667
+                            ],
+                            "spans": [],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                663.6666666666667,
+                                495,
+                                675.0000000000001
+                            ],
+                            "spans": [],
+                            "index": 39
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        681,
+                        414,
+                        693
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                680,
+                                415,
+                                695
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        680,
+                                        415,
+                                        695
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Thus, any set of permutations on hidden units defines an equivalent network.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        }
+                    ],
+                    "index": 40,
+                    "bbox_fs": [
+                        106,
+                        680,
+                        415,
+                        695
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        698,
+                        504,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                698,
+                                505,
+                                712
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        698,
+                                        505,
+                                        712
+                                    ],
+                                    "score": 1.0,
+                                    "content": "It is obvious that sparse networks are much more immune to parameter permutation – indeed every",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                709,
+                                506,
+                                723
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        709,
+                                        173,
+                                        723
+                                    ],
+                                    "score": 1.0,
+                                    "content": "channel at layer",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        173,
+                                        711,
+                                        177,
+                                        720
+                                    ],
+                                    "score": 0.66,
+                                    "content": "l",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        178,
+                                        709,
+                                        506,
+                                        723
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is likely to have a unique tree describing its connection matrix all the way down.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                720,
+                                329,
+                                733
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        720,
+                                        329,
+                                        733
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Exploring this direction is an interesting open question.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        }
+                    ],
+                    "index": 42,
+                    "bbox_fs": [
+                        105,
+                        698,
+                        506,
+                        733
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        81,
+                        200,
+                        94
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                79,
+                                202,
+                                96
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        79,
+                                        202,
+                                        96
+                                    ],
+                                    "score": 1.0,
+                                    "content": "5 EXPERIMENTS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        }
+                    ],
+                    "index": 0
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        106,
+                        505,
+                        140
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                105,
+                                505,
+                                119
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        105,
+                                        505,
+                                        119
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In this section, we demonstrate the effectiveness of the sparse random approach by comparing it to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                116,
+                                505,
+                                131
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        116,
+                                        505,
+                                        131
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the depth multiplier approach at different compression rates. Moreover, we examine several settings",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                128,
+                                503,
+                                141
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        128,
+                                        503,
+                                        141
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in the incremental training where connections gradually become active during the training process.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        }
+                    ],
+                    "index": 2
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        153,
+                        160,
+                        164
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                152,
+                                161,
+                                166
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        152,
+                                        161,
+                                        166
+                                    ],
+                                    "score": 1.0,
+                                    "content": "5.1 SETUP",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        }
+                    ],
+                    "index": 4
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        176,
+                        504,
+                        198
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                175,
+                                505,
+                                189
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        175,
+                                        505,
+                                        189
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Networks and Datasets Our experiments are conducted on 4 networks for 3 different datasets.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                187,
+                                424,
+                                199
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        187,
+                                        424,
+                                        199
+                                    ],
+                                    "score": 1.0,
+                                    "content": "All our experiments use open-source TensorFlow networks Abadi et al. (2015).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        }
+                    ],
+                    "index": 5.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        212,
+                        504,
+                        245
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                211,
+                                506,
+                                225
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        211,
+                                        506,
+                                        225
+                                    ],
+                                    "score": 1.0,
+                                    "content": "MNIST AND CIFAR-10 We use standard networks provided by TensorFlow. For MNIST, it has",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                224,
+                                505,
+                                235
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        224,
+                                        273,
+                                        235
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3-layer convolutional layers and achieves",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        273,
+                                        224,
+                                        300,
+                                        234
+                                    ],
+                                    "score": 0.86,
+                                    "content": "9 9 . 5 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        301,
+                                        224,
+                                        505,
+                                        235
+                                    ],
+                                    "score": 1.0,
+                                    "content": "accuracy when fully trained. For CIFAR-10, it has",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                234,
+                                311,
+                                248
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        234,
+                                        250,
+                                        248
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2 convolutional layers and achieves",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        250,
+                                        235,
+                                        270,
+                                        245
+                                    ],
+                                    "score": 0.88,
+                                    "content": "87 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        270,
+                                        234,
+                                        311,
+                                        248
+                                    ],
+                                    "score": 1.0,
+                                    "content": "accuracy.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        }
+                    ],
+                    "index": 8
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        259,
+                        503,
+                        293
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                259,
+                                505,
+                                273
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        259,
+                                        505,
+                                        273
+                                    ],
+                                    "score": 1.0,
+                                    "content": "IMAGENET We use open source Inception-V3 (Szegedy et al., 2016) network and a slightly mod-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                270,
+                                505,
+                                284
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        270,
+                                        505,
+                                        284
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ified version of VGG-16 (Simonyan & Zisserman, 2015) called VGG-16n on ImageNet ILSVRC",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                281,
+                                313,
+                                295
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        281,
+                                        313,
+                                        295
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2012 (Deng et al., 2009; Russakovsky et al., 2015).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        }
+                    ],
+                    "index": 11
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        307,
+                        505,
+                        351
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                308,
+                                505,
+                                319
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        308,
+                                        505,
+                                        319
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Random connections Connections are activated according to their likelihood from the uniform",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                318,
+                                505,
+                                330
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        318,
+                                        505,
+                                        330
+                                    ],
+                                    "score": 1.0,
+                                    "content": "distribution. In addition, they are activated in such a way that there are no connections going in or",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                329,
+                                505,
+                                342
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        329,
+                                        505,
+                                        342
+                                    ],
+                                    "score": 1.0,
+                                    "content": "coming out of dead filters (i.e., any connection must have a path to input image and a path to the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                340,
+                                396,
+                                352
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        340,
+                                        396,
+                                        352
+                                    ],
+                                    "score": 1.0,
+                                    "content": "final prediction.). All connections in fully connected layers are retained.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        }
+                    ],
+                    "index": 14.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        365,
+                        504,
+                        409
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                366,
+                                505,
+                                377
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        366,
+                                        505,
+                                        377
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Implementation details All code is implemented in TensorFlow (Abadi et al., 2015). Deactivat-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                376,
+                                506,
+                                389
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        376,
+                                        506,
+                                        389
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ing connections is done by applying masks to parameter tensors. The Inception-v3 and VGG-16n",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                387,
+                                506,
+                                400
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        387,
+                                        506,
+                                        400
+                                    ],
+                                    "score": 1.0,
+                                    "content": "networks are trained on 8 Tesla K80 GPUs, each with batch size 256 (32 per gpu) and batch normal-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                398,
+                                242,
+                                410
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        398,
+                                        242,
+                                        410
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ization was used for all networks.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        }
+                    ],
+                    "index": 18.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        425,
+                        414,
+                        435
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                424,
+                                416,
+                                437
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        424,
+                                        416,
+                                        437
+                                    ],
+                                    "score": 1.0,
+                                    "content": "5.2 COMPARISON BETWEEN SPARSE RANDOM AND DEPTH MULTIPLIER",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        }
+                    ],
+                    "index": 21
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        446,
+                        244,
+                        458
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                446,
+                                244,
+                                459
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        446,
+                                        244,
+                                        459
+                                    ],
+                                    "score": 1.0,
+                                    "content": "5.2.1 MNIST AND CIFAR-10",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        }
+                    ],
+                    "index": 22
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        469,
+                        505,
+                        546
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                469,
+                                505,
+                                482
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        469,
+                                        505,
+                                        482
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We first compare depth multiplier and sparse random for the two small networks on MNIST and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                480,
+                                505,
+                                492
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        480,
+                                        505,
+                                        492
+                                    ],
+                                    "score": 1.0,
+                                    "content": "CIFAR-10. We compare the accuracy of the two approaches when the numbers of connections are",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                491,
+                                506,
+                                503
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        491,
+                                        288,
+                                        503
+                                    ],
+                                    "score": 1.0,
+                                    "content": "roughly the same, based on a hyperparameter",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        289,
+                                        493,
+                                        296,
+                                        501
+                                    ],
+                                    "score": 0.69,
+                                    "content": "\\alpha",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        297,
+                                        491,
+                                        479,
+                                        503
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". For dense convolutions, we pick a multiplier",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        479,
+                                        493,
+                                        487,
+                                        501
+                                    ],
+                                    "score": 0.78,
+                                    "content": "\\alpha",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        488,
+                                        491,
+                                        506,
+                                        503
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                502,
+                                505,
+                                515
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        502,
+                                        246,
+                                        515
+                                    ],
+                                    "score": 1.0,
+                                    "content": "each filter depth is scaled down by",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        247,
+                                        502,
+                                        263,
+                                        514
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\sqrt { \\alpha }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        263,
+                                        502,
+                                        485,
+                                        515
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and then rounded up. In sparse convolutions, a fraction",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        486,
+                                        504,
+                                        493,
+                                        512
+                                    ],
+                                    "score": 0.75,
+                                    "content": "\\alpha",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        493,
+                                        502,
+                                        505,
+                                        515
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                513,
+                                505,
+                                525
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        513,
+                                        505,
+                                        525
+                                    ],
+                                    "score": 1.0,
+                                    "content": "connections are randomly deactivated if those parameters connect at least two filters on each layer;",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                523,
+                                506,
+                                537
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        523,
+                                        201,
+                                        537
+                                    ],
+                                    "score": 1.0,
+                                    "content": "otherwise, a fraction of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        202,
+                                        524,
+                                        217,
+                                        536
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\sqrt { \\alpha }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        218,
+                                        523,
+                                        506,
+                                        537
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is used instead if the parameters connect layers with only one filter left.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                535,
+                                469,
+                                547
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        535,
+                                        469,
+                                        547
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The accuracy numbers are averaged over 5 rounds for MNIST and 2 rounds on CIFAR-10.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        }
+                    ],
+                    "index": 26
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        552,
+                        505,
+                        629
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                552,
+                                505,
+                                564
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        552,
+                                        505,
+                                        564
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We show in Fig. 2 and Fig. 3 that the sparse networks have comparable or higher accuracy for the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                563,
+                                506,
+                                576
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        563,
+                                        506,
+                                        576
+                                    ],
+                                    "score": 1.0,
+                                    "content": "same number of parameters, with comparable accuracy at higher density. We note however that",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                574,
+                                505,
+                                586
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        574,
+                                        505,
+                                        586
+                                    ],
+                                    "score": 1.0,
+                                    "content": "these networks are so small that at high compression rates most of operations are concentrated at the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                585,
+                                505,
+                                597
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        585,
+                                        505,
+                                        597
+                                    ],
+                                    "score": 1.0,
+                                    "content": "first layer, which is negligible for large networks. Moreover, in MNIST example, the size of network",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                596,
+                                505,
+                                608
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        596,
+                                        487,
+                                        608
+                                    ],
+                                    "score": 1.0,
+                                    "content": "changes most dramatically from 2000 to 2 million parameters, while affecting accuracy only by",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        487,
+                                        596,
+                                        501,
+                                        606
+                                    ],
+                                    "score": 0.86,
+                                    "content": "1 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        501,
+                                        596,
+                                        505,
+                                        608
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                607,
+                                505,
+                                619
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        607,
+                                        505,
+                                        619
+                                    ],
+                                    "score": 1.0,
+                                    "content": "This observation suggests that there might be benefits of maintaining the number of filters to be high",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                618,
+                                435,
+                                630
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        618,
+                                        435,
+                                        630
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and/or breaking the symmetry of connections. We explore this in the next section.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        }
+                    ],
+                    "index": 33
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        642,
+                        267,
+                        654
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                641,
+                                269,
+                                656
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        641,
+                                        269,
+                                        656
+                                    ],
+                                    "score": 1.0,
+                                    "content": "5.2.2 INCEPTION-V3 ON IMAGENET",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        }
+                    ],
+                    "index": 37
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        666,
+                        505,
+                        731
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                665,
+                                505,
+                                678
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        665,
+                                        505,
+                                        678
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We consider different values of sparsity ranging from 0.003 to 1, and depth multiplier from 0.05",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                677,
+                                506,
+                                690
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        677,
+                                        506,
+                                        690
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to 1. Our experiments show (see Table 1 and Fig. 4) significant advantage of sparse networks over",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                687,
+                                505,
+                                701
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        687,
+                                        505,
+                                        701
+                                    ],
+                                    "score": 1.0,
+                                    "content": "equivalently sized dense networks. We note that due to time constraints the reported quantitative",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                699,
+                                504,
+                                711
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        699,
+                                        504,
+                                        711
+                                    ],
+                                    "score": 1.0,
+                                    "content": "numbers are preliminary, as the networks have not finished converging. We expect the final numbers",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                709,
+                                506,
+                                722
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        709,
+                                        506,
+                                        722
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to match the reported number for Inception V3 Szegedy et al. (2016), and the smaller networks to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                721,
+                                234,
+                                732
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        721,
+                                        234,
+                                        732
+                                    ],
+                                    "score": 1.0,
+                                    "content": "have comparable improvement.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        }
+                    ],
+                    "index": 40.5
+                }
+            ],
+            "page_idx": 5,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        107,
+                        27,
+                        308,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                107,
+                                26,
+                                308,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        26,
+                                        308,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Under review as a conference paper at ICLR 2017",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        302,
+                        752,
+                        308,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                302,
+                                751,
+                                309,
+                                762
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        302,
+                                        751,
+                                        309,
+                                        762
+                                    ],
+                                    "score": 1.0,
+                                    "content": "6",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        81,
+                        200,
+                        94
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                79,
+                                202,
+                                96
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        79,
+                                        202,
+                                        96
+                                    ],
+                                    "score": 1.0,
+                                    "content": "5 EXPERIMENTS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        }
+                    ],
+                    "index": 0
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        106,
+                        505,
+                        140
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                105,
+                                505,
+                                119
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        105,
+                                        505,
+                                        119
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In this section, we demonstrate the effectiveness of the sparse random approach by comparing it to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                116,
+                                505,
+                                131
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        116,
+                                        505,
+                                        131
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the depth multiplier approach at different compression rates. Moreover, we examine several settings",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                128,
+                                503,
+                                141
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        128,
+                                        503,
+                                        141
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in the incremental training where connections gradually become active during the training process.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        }
+                    ],
+                    "index": 2,
+                    "bbox_fs": [
+                        105,
+                        105,
+                        505,
+                        141
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        153,
+                        160,
+                        164
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                152,
+                                161,
+                                166
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        152,
+                                        161,
+                                        166
+                                    ],
+                                    "score": 1.0,
+                                    "content": "5.1 SETUP",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        }
+                    ],
+                    "index": 4
+                },
+                {
+                    "type": "list",
+                    "bbox": [
+                        108,
+                        176,
+                        504,
+                        198
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                175,
+                                505,
+                                189
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        175,
+                                        505,
+                                        189
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Networks and Datasets Our experiments are conducted on 4 networks for 3 different datasets.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                187,
+                                424,
+                                199
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        187,
+                                        424,
+                                        199
+                                    ],
+                                    "score": 1.0,
+                                    "content": "All our experiments use open-source TensorFlow networks Abadi et al. (2015).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6,
+                            "is_list_start_line": true,
+                            "is_list_end_line": true
+                        }
+                    ],
+                    "index": 5.5,
+                    "bbox_fs": [
+                        105,
+                        175,
+                        505,
+                        199
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        212,
+                        504,
+                        245
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                211,
+                                506,
+                                225
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        211,
+                                        506,
+                                        225
+                                    ],
+                                    "score": 1.0,
+                                    "content": "MNIST AND CIFAR-10 We use standard networks provided by TensorFlow. For MNIST, it has",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                224,
+                                505,
+                                235
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        224,
+                                        273,
+                                        235
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3-layer convolutional layers and achieves",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        273,
+                                        224,
+                                        300,
+                                        234
+                                    ],
+                                    "score": 0.86,
+                                    "content": "9 9 . 5 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        301,
+                                        224,
+                                        505,
+                                        235
+                                    ],
+                                    "score": 1.0,
+                                    "content": "accuracy when fully trained. For CIFAR-10, it has",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                234,
+                                311,
+                                248
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        234,
+                                        250,
+                                        248
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2 convolutional layers and achieves",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        250,
+                                        235,
+                                        270,
+                                        245
+                                    ],
+                                    "score": 0.88,
+                                    "content": "87 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        270,
+                                        234,
+                                        311,
+                                        248
+                                    ],
+                                    "score": 1.0,
+                                    "content": "accuracy.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        }
+                    ],
+                    "index": 8,
+                    "bbox_fs": [
+                        105,
+                        211,
+                        506,
+                        248
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        259,
+                        503,
+                        293
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                259,
+                                505,
+                                273
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        259,
+                                        505,
+                                        273
+                                    ],
+                                    "score": 1.0,
+                                    "content": "IMAGENET We use open source Inception-V3 (Szegedy et al., 2016) network and a slightly mod-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                270,
+                                505,
+                                284
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        270,
+                                        505,
+                                        284
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ified version of VGG-16 (Simonyan & Zisserman, 2015) called VGG-16n on ImageNet ILSVRC",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                281,
+                                313,
+                                295
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        281,
+                                        313,
+                                        295
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2012 (Deng et al., 2009; Russakovsky et al., 2015).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        }
+                    ],
+                    "index": 11,
+                    "bbox_fs": [
+                        105,
+                        259,
+                        505,
+                        295
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        307,
+                        505,
+                        351
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                308,
+                                505,
+                                319
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        308,
+                                        505,
+                                        319
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Random connections Connections are activated according to their likelihood from the uniform",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                318,
+                                505,
+                                330
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        318,
+                                        505,
+                                        330
+                                    ],
+                                    "score": 1.0,
+                                    "content": "distribution. In addition, they are activated in such a way that there are no connections going in or",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                329,
+                                505,
+                                342
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        329,
+                                        505,
+                                        342
+                                    ],
+                                    "score": 1.0,
+                                    "content": "coming out of dead filters (i.e., any connection must have a path to input image and a path to the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                340,
+                                396,
+                                352
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        340,
+                                        396,
+                                        352
+                                    ],
+                                    "score": 1.0,
+                                    "content": "final prediction.). All connections in fully connected layers are retained.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        }
+                    ],
+                    "index": 14.5,
+                    "bbox_fs": [
+                        105,
+                        308,
+                        505,
+                        352
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        365,
+                        504,
+                        409
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                366,
+                                505,
+                                377
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        366,
+                                        505,
+                                        377
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Implementation details All code is implemented in TensorFlow (Abadi et al., 2015). Deactivat-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                376,
+                                506,
+                                389
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        376,
+                                        506,
+                                        389
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ing connections is done by applying masks to parameter tensors. The Inception-v3 and VGG-16n",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                387,
+                                506,
+                                400
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        387,
+                                        506,
+                                        400
+                                    ],
+                                    "score": 1.0,
+                                    "content": "networks are trained on 8 Tesla K80 GPUs, each with batch size 256 (32 per gpu) and batch normal-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                398,
+                                242,
+                                410
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        398,
+                                        242,
+                                        410
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ization was used for all networks.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        }
+                    ],
+                    "index": 18.5,
+                    "bbox_fs": [
+                        105,
+                        366,
+                        506,
+                        410
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        425,
+                        414,
+                        435
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                424,
+                                416,
+                                437
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        424,
+                                        416,
+                                        437
+                                    ],
+                                    "score": 1.0,
+                                    "content": "5.2 COMPARISON BETWEEN SPARSE RANDOM AND DEPTH MULTIPLIER",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        }
+                    ],
+                    "index": 21,
+                    "bbox_fs": [
+                        105,
+                        424,
+                        416,
+                        437
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        446,
+                        244,
+                        458
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                446,
+                                244,
+                                459
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        446,
+                                        244,
+                                        459
+                                    ],
+                                    "score": 1.0,
+                                    "content": "5.2.1 MNIST AND CIFAR-10",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        }
+                    ],
+                    "index": 22
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        469,
+                        505,
+                        546
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                469,
+                                505,
+                                482
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        469,
+                                        505,
+                                        482
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We first compare depth multiplier and sparse random for the two small networks on MNIST and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                480,
+                                505,
+                                492
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        480,
+                                        505,
+                                        492
+                                    ],
+                                    "score": 1.0,
+                                    "content": "CIFAR-10. We compare the accuracy of the two approaches when the numbers of connections are",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                491,
+                                506,
+                                503
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        491,
+                                        288,
+                                        503
+                                    ],
+                                    "score": 1.0,
+                                    "content": "roughly the same, based on a hyperparameter",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        289,
+                                        493,
+                                        296,
+                                        501
+                                    ],
+                                    "score": 0.69,
+                                    "content": "\\alpha",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        297,
+                                        491,
+                                        479,
+                                        503
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". For dense convolutions, we pick a multiplier",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        479,
+                                        493,
+                                        487,
+                                        501
+                                    ],
+                                    "score": 0.78,
+                                    "content": "\\alpha",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        488,
+                                        491,
+                                        506,
+                                        503
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                502,
+                                505,
+                                515
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        502,
+                                        246,
+                                        515
+                                    ],
+                                    "score": 1.0,
+                                    "content": "each filter depth is scaled down by",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        247,
+                                        502,
+                                        263,
+                                        514
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\sqrt { \\alpha }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        263,
+                                        502,
+                                        485,
+                                        515
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and then rounded up. In sparse convolutions, a fraction",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        486,
+                                        504,
+                                        493,
+                                        512
+                                    ],
+                                    "score": 0.75,
+                                    "content": "\\alpha",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        493,
+                                        502,
+                                        505,
+                                        515
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                513,
+                                505,
+                                525
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        513,
+                                        505,
+                                        525
+                                    ],
+                                    "score": 1.0,
+                                    "content": "connections are randomly deactivated if those parameters connect at least two filters on each layer;",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                523,
+                                506,
+                                537
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        523,
+                                        201,
+                                        537
+                                    ],
+                                    "score": 1.0,
+                                    "content": "otherwise, a fraction of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        202,
+                                        524,
+                                        217,
+                                        536
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\sqrt { \\alpha }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        218,
+                                        523,
+                                        506,
+                                        537
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is used instead if the parameters connect layers with only one filter left.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                535,
+                                469,
+                                547
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        535,
+                                        469,
+                                        547
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The accuracy numbers are averaged over 5 rounds for MNIST and 2 rounds on CIFAR-10.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        }
+                    ],
+                    "index": 26,
+                    "bbox_fs": [
+                        105,
+                        469,
+                        506,
+                        547
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        552,
+                        505,
+                        629
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                552,
+                                505,
+                                564
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        552,
+                                        505,
+                                        564
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We show in Fig. 2 and Fig. 3 that the sparse networks have comparable or higher accuracy for the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                563,
+                                506,
+                                576
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        563,
+                                        506,
+                                        576
+                                    ],
+                                    "score": 1.0,
+                                    "content": "same number of parameters, with comparable accuracy at higher density. We note however that",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                574,
+                                505,
+                                586
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        574,
+                                        505,
+                                        586
+                                    ],
+                                    "score": 1.0,
+                                    "content": "these networks are so small that at high compression rates most of operations are concentrated at the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                585,
+                                505,
+                                597
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        585,
+                                        505,
+                                        597
+                                    ],
+                                    "score": 1.0,
+                                    "content": "first layer, which is negligible for large networks. Moreover, in MNIST example, the size of network",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                596,
+                                505,
+                                608
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        596,
+                                        487,
+                                        608
+                                    ],
+                                    "score": 1.0,
+                                    "content": "changes most dramatically from 2000 to 2 million parameters, while affecting accuracy only by",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        487,
+                                        596,
+                                        501,
+                                        606
+                                    ],
+                                    "score": 0.86,
+                                    "content": "1 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        501,
+                                        596,
+                                        505,
+                                        608
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                607,
+                                505,
+                                619
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        607,
+                                        505,
+                                        619
+                                    ],
+                                    "score": 1.0,
+                                    "content": "This observation suggests that there might be benefits of maintaining the number of filters to be high",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                618,
+                                435,
+                                630
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        618,
+                                        435,
+                                        630
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and/or breaking the symmetry of connections. We explore this in the next section.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        }
+                    ],
+                    "index": 33,
+                    "bbox_fs": [
+                        105,
+                        552,
+                        506,
+                        630
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        642,
+                        267,
+                        654
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                641,
+                                269,
+                                656
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        641,
+                                        269,
+                                        656
+                                    ],
+                                    "score": 1.0,
+                                    "content": "5.2.2 INCEPTION-V3 ON IMAGENET",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        }
+                    ],
+                    "index": 37
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        666,
+                        505,
+                        731
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                665,
+                                505,
+                                678
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        665,
+                                        505,
+                                        678
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We consider different values of sparsity ranging from 0.003 to 1, and depth multiplier from 0.05",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                677,
+                                506,
+                                690
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        677,
+                                        506,
+                                        690
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to 1. Our experiments show (see Table 1 and Fig. 4) significant advantage of sparse networks over",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                687,
+                                505,
+                                701
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        687,
+                                        505,
+                                        701
+                                    ],
+                                    "score": 1.0,
+                                    "content": "equivalently sized dense networks. We note that due to time constraints the reported quantitative",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                699,
+                                504,
+                                711
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        699,
+                                        504,
+                                        711
+                                    ],
+                                    "score": 1.0,
+                                    "content": "numbers are preliminary, as the networks have not finished converging. We expect the final numbers",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                709,
+                                506,
+                                722
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        709,
+                                        506,
+                                        722
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to match the reported number for Inception V3 Szegedy et al. (2016), and the smaller networks to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                721,
+                                234,
+                                732
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        721,
+                                        234,
+                                        732
+                                    ],
+                                    "score": 1.0,
+                                    "content": "have comparable improvement.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        }
+                    ],
+                    "index": 40.5,
+                    "bbox_fs": [
+                        105,
+                        665,
+                        506,
+                        732
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "image",
+                    "bbox": [
+                        125,
+                        80,
+                        480,
+                        215
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                125,
+                                80,
+                                480,
+                                215
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        125,
+                                        80,
+                                        480,
+                                        215
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                125,
+                                                80,
+                                                480,
+                                                215
+                                            ],
+                                            "score": 0.969,
+                                            "type": "image",
+                                            "image_path": "65926ca6ffb9b876a5add68d622ccc4f1c7a0fd94934ef67cf453485cb422141.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 1,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        125,
+                                        80,
+                                        480,
+                                        125.0
+                                    ],
+                                    "spans": [],
+                                    "index": 0
+                                },
+                                {
+                                    "bbox": [
+                                        125,
+                                        125.0,
+                                        480,
+                                        170.0
+                                    ],
+                                    "spans": [],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        125,
+                                        170.0,
+                                        480,
+                                        215.0
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                106,
+                                217,
+                                506,
+                                261
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        216,
+                                        506,
+                                        229
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                216,
+                                                506,
+                                                229
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 2: Comparison of accuracy (averaged over 5 rounds) vs. Number of parameters/Number of",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 3
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        227,
+                                        506,
+                                        240
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                227,
+                                                506,
+                                                240
+                                            ],
+                                            "score": 1.0,
+                                            "content": "multiply-adds between dense and sparse convolutions on MNIST dataset. Note that though sparse",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 4
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        239,
+                                        505,
+                                        251
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                239,
+                                                505,
+                                                251
+                                            ],
+                                            "score": 1.0,
+                                            "content": "convolution result in better parameter trade-off curve, the multiply-add curve shows the opposite",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 5
+                                },
+                                {
+                                    "bbox": [
+                                        104,
+                                        249,
+                                        140,
+                                        262
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                104,
+                                                249,
+                                                140,
+                                                262
+                                            ],
+                                            "score": 1.0,
+                                            "content": "pattern.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 6
+                                }
+                            ],
+                            "index": 4.5
+                        }
+                    ],
+                    "index": 2.75
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        270,
+                        505,
+                        304
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                270,
+                                505,
+                                284
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        270,
+                                        505,
+                                        284
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Table 1: Inception V3: Preliminary quantitative results after 100 Epochs. Note the smallest sparse",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                282,
+                                505,
+                                294
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        282,
+                                        505,
+                                        294
+                                    ],
+                                    "score": 1.0,
+                                    "content": "network is actually a hybrid network - we used both depth multiplier (0.5) and sparsity (0.01). The",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                291,
+                                425,
+                                306
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        291,
+                                        425,
+                                        306
+                                    ],
+                                    "score": 1.0,
+                                    "content": "number of parameters is the number of parameters excluding the softmax layer.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        }
+                    ],
+                    "index": 8
+                },
+                {
+                    "type": "table",
+                    "bbox": [
+                        132,
+                        324,
+                        298,
+                        426
+                    ],
+                    "blocks": [
+                        {
+                            "type": "table_caption",
+                            "bbox": [
+                                149,
+                                314,
+                                283,
+                                324
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        147,
+                                        312,
+                                        283,
+                                        326
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                147,
+                                                312,
+                                                283,
+                                                326
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Accuracy for sparse convolutions",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 10
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "type": "table_body",
+                            "bbox": [
+                                132,
+                                324,
+                                298,
+                                426
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        132,
+                                        324,
+                                        298,
+                                        426
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                132,
+                                                324,
+                                                298,
+                                                426
+                                            ],
+                                            "score": 0.969,
+                                            "html": "<table><tr><td>Sparsity</td><td>MAdds Params</td><td>P@1</td></tr><tr><td>0.50/0.01 0.003</td><td>43.0M 90k 158k</td><td>40.3 46.1</td></tr><tr><td>0.01</td><td>82.0M 104M</td><td>52.3</td></tr><tr><td>0.03</td><td>287k 208M 724k</td><td></td></tr><tr><td>0.10</td><td></td><td>59.5</td></tr><tr><td>0.30</td><td>628M 2.3 M</td><td>67.2</td></tr><tr><td></td><td>1.80 B 6.6M</td><td>73</td></tr><tr><td>0.60</td><td>3.50B 13M 22 M</td><td>75</td></tr><tr><td>1.00</td><td>5.70 B</td><td>77</td></tr></table>",
+                                            "type": "table",
+                                            "image_path": "b34ddde6d213af5b60d226747df5ee3604eee0c1cb8d2d5689b2231902ee2059.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 14,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        132,
+                                        324,
+                                        298,
+                                        338.57142857142856
+                                    ],
+                                    "spans": [],
+                                    "index": 12
+                                },
+                                {
+                                    "bbox": [
+                                        132,
+                                        338.57142857142856,
+                                        298,
+                                        353.1428571428571
+                                    ],
+                                    "spans": [],
+                                    "index": 13
+                                },
+                                {
+                                    "bbox": [
+                                        132,
+                                        353.1428571428571,
+                                        298,
+                                        367.71428571428567
+                                    ],
+                                    "spans": [],
+                                    "index": 11
+                                },
+                                {
+                                    "bbox": [
+                                        132,
+                                        367.71428571428567,
+                                        298,
+                                        382.2857142857142
+                                    ],
+                                    "spans": [],
+                                    "index": 14
+                                },
+                                {
+                                    "bbox": [
+                                        132,
+                                        382.2857142857142,
+                                        298,
+                                        396.8571428571428
+                                    ],
+                                    "spans": [],
+                                    "index": 15
+                                },
+                                {
+                                    "bbox": [
+                                        132,
+                                        396.8571428571428,
+                                        298,
+                                        411.42857142857133
+                                    ],
+                                    "spans": [],
+                                    "index": 16
+                                },
+                                {
+                                    "bbox": [
+                                        132,
+                                        411.42857142857133,
+                                        298,
+                                        425.9999999999999
+                                    ],
+                                    "spans": [],
+                                    "index": 17
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 12.0
+                },
+                {
+                    "type": "table",
+                    "bbox": [
+                        327,
+                        335,
+                        499,
+                        415
+                    ],
+                    "blocks": [
+                        {
+                            "type": "table_caption",
+                            "bbox": [
+                                352,
+                                313,
+                                475,
+                                325
+                            ],
+                            "group_id": 1,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        351,
+                                        312,
+                                        477,
+                                        327
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                351,
+                                                312,
+                                                477,
+                                                327
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Accuracy for Depth Multiplier",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 18
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "type": "table_body",
+                            "bbox": [
+                                327,
+                                335,
+                                499,
+                                415
+                            ],
+                            "group_id": 1,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        327,
+                                        335,
+                                        499,
+                                        415
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                327,
+                                                335,
+                                                499,
+                                                415
+                                            ],
+                                            "score": 0.974,
+                                            "html": "<table><tr><td>Multiplier</td><td>MAdds</td><td>Params P@1</td></tr><tr><td>0.05</td><td>55.0M 56k</td><td>24.6</td></tr><tr><td>0.10</td><td>75.0M</td><td>170k 38.6</td></tr><tr><td>0.20</td><td>183M 718k</td><td>54.2</td></tr><tr><td>0.30</td><td>439M 1.8M</td><td>64.0</td></tr><tr><td>0.50</td><td>1.40B 5.4M</td><td>72.3</td></tr><tr><td>0.80</td><td>3.40B 13M</td><td>75.6</td></tr></table>",
+                                            "type": "table",
+                                            "image_path": "f13c73e145cd97e43929b7f6bff27130ece79ed9e82f3f37297cc3a19a01a5c9.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 21.5,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        327,
+                                        335,
+                                        499,
+                                        348.3333333333333
+                                    ],
+                                    "spans": [],
+                                    "index": 19
+                                },
+                                {
+                                    "bbox": [
+                                        327,
+                                        348.3333333333333,
+                                        499,
+                                        361.66666666666663
+                                    ],
+                                    "spans": [],
+                                    "index": 20
+                                },
+                                {
+                                    "bbox": [
+                                        327,
+                                        361.66666666666663,
+                                        499,
+                                        374.99999999999994
+                                    ],
+                                    "spans": [],
+                                    "index": 21
+                                },
+                                {
+                                    "bbox": [
+                                        327,
+                                        374.99999999999994,
+                                        499,
+                                        388.33333333333326
+                                    ],
+                                    "spans": [],
+                                    "index": 22
+                                },
+                                {
+                                    "bbox": [
+                                        327,
+                                        388.33333333333326,
+                                        499,
+                                        401.6666666666666
+                                    ],
+                                    "spans": [],
+                                    "index": 23
+                                },
+                                {
+                                    "bbox": [
+                                        327,
+                                        401.6666666666666,
+                                        499,
+                                        414.9999999999999
+                                    ],
+                                    "spans": [],
+                                    "index": 24
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 19.75
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        119,
+                        435,
+                        312,
+                        447
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                117,
+                                433,
+                                313,
+                                449
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        117,
+                                        433,
+                                        313,
+                                        449
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Original network: 5.70 B 22 M 77 (78.8)",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        }
+                    ],
+                    "index": 25
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        471,
+                        243,
+                        483
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                470,
+                                244,
+                                485
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        470,
+                                        244,
+                                        485
+                                    ],
+                                    "score": 1.0,
+                                    "content": "5.2.3 VGG-16 ON IMAGENET",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        }
+                    ],
+                    "index": 26
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        493,
+                        505,
+                        560
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                494,
+                                505,
+                                506
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        494,
+                                        505,
+                                        506
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In our experiments with the VGG-16 network (Simonyan & Zisserman, 2015), we modify the model",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                504,
+                                505,
+                                517
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        504,
+                                        505,
+                                        517
+                                    ],
+                                    "score": 1.0,
+                                    "content": "architecture (calling it VGG-16n) by removing the two fully-connected layers with depth 4096 and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                515,
+                                505,
+                                528
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        515,
+                                        196,
+                                        528
+                                    ],
+                                    "score": 1.0,
+                                    "content": "replacing them with a",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        197,
+                                        516,
+                                        221,
+                                        527
+                                    ],
+                                    "score": 0.88,
+                                    "content": "2 \\times 2",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        221,
+                                        515,
+                                        340,
+                                        528
+                                    ],
+                                    "score": 1.0,
+                                    "content": "maxpool layer followed by a",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        340,
+                                        516,
+                                        364,
+                                        527
+                                    ],
+                                    "score": 0.91,
+                                    "content": "3 \\times 3",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        365,
+                                        515,
+                                        505,
+                                        528
+                                    ],
+                                    "score": 1.0,
+                                    "content": "convolutional layer with the depth",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                526,
+                                506,
+                                540
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        526,
+                                        506,
+                                        540
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of 1024. This alone sped up our training significantly. The comparison between depth multiplier",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                538,
+                                506,
+                                550
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        538,
+                                        506,
+                                        550
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and sparse connection approaches is shown in Fig. 5. The modified VGG-16n network has about 7",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                550,
+                                373,
+                                561
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        550,
+                                        373,
+                                        561
+                                    ],
+                                    "score": 1.0,
+                                    "content": "times fewer parameters, but appears to have comparable precision.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        }
+                    ],
+                    "index": 29.5
+                },
+                {
+                    "type": "image",
+                    "bbox": [
+                        136,
+                        582,
+                        470,
+                        710
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                136,
+                                582,
+                                470,
+                                710
+                            ],
+                            "group_id": 1,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        136,
+                                        582,
+                                        470,
+                                        710
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                136,
+                                                582,
+                                                470,
+                                                710
+                                            ],
+                                            "score": 0.963,
+                                            "type": "image",
+                                            "image_path": "f4cf6c184c4020dc5874321b477ae62c97ddf25c9d68322157c1a605a04d32e7.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 34,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        136,
+                                        582,
+                                        470,
+                                        624.6666666666666
+                                    ],
+                                    "spans": [],
+                                    "index": 33
+                                },
+                                {
+                                    "bbox": [
+                                        136,
+                                        624.6666666666666,
+                                        470,
+                                        667.3333333333333
+                                    ],
+                                    "spans": [],
+                                    "index": 34
+                                },
+                                {
+                                    "bbox": [
+                                        136,
+                                        667.3333333333333,
+                                        470,
+                                        709.9999999999999
+                                    ],
+                                    "spans": [],
+                                    "index": 35
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                106,
+                                712,
+                                506,
+                                734
+                            ],
+                            "group_id": 1,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        711,
+                                        505,
+                                        723
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                711,
+                                                505,
+                                                723
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 3: Comparison of accuracy (averaged over 2 rounds) vs. Number of parameters/Number of",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 36
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        722,
+                                        413,
+                                        735
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                722,
+                                                413,
+                                                735
+                                            ],
+                                            "score": 1.0,
+                                            "content": "multiply-adds between dense and sparse convolutions on CIFAR-10 dataset.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 37
+                                }
+                            ],
+                            "index": 36.5
+                        }
+                    ],
+                    "index": 35.25
+                }
+            ],
+            "page_idx": 6,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        107,
+                        27,
+                        308,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                107,
+                                26,
+                                308,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        26,
+                                        308,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Under review as a conference paper at ICLR 2017",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        302,
+                        751,
+                        309,
+                        759
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                302,
+                                750,
+                                309,
+                                762
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        302,
+                                        750,
+                                        309,
+                                        762
+                                    ],
+                                    "score": 1.0,
+                                    "content": "7",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "image",
+                    "bbox": [
+                        125,
+                        80,
+                        480,
+                        215
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                125,
+                                80,
+                                480,
+                                215
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        125,
+                                        80,
+                                        480,
+                                        215
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                125,
+                                                80,
+                                                480,
+                                                215
+                                            ],
+                                            "score": 0.969,
+                                            "type": "image",
+                                            "image_path": "65926ca6ffb9b876a5add68d622ccc4f1c7a0fd94934ef67cf453485cb422141.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 1,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        125,
+                                        80,
+                                        480,
+                                        125.0
+                                    ],
+                                    "spans": [],
+                                    "index": 0
+                                },
+                                {
+                                    "bbox": [
+                                        125,
+                                        125.0,
+                                        480,
+                                        170.0
+                                    ],
+                                    "spans": [],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        125,
+                                        170.0,
+                                        480,
+                                        215.0
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                106,
+                                217,
+                                506,
+                                261
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        216,
+                                        506,
+                                        229
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                216,
+                                                506,
+                                                229
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 2: Comparison of accuracy (averaged over 5 rounds) vs. Number of parameters/Number of",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 3
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        227,
+                                        506,
+                                        240
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                227,
+                                                506,
+                                                240
+                                            ],
+                                            "score": 1.0,
+                                            "content": "multiply-adds between dense and sparse convolutions on MNIST dataset. Note that though sparse",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 4
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        239,
+                                        505,
+                                        251
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                239,
+                                                505,
+                                                251
+                                            ],
+                                            "score": 1.0,
+                                            "content": "convolution result in better parameter trade-off curve, the multiply-add curve shows the opposite",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 5
+                                },
+                                {
+                                    "bbox": [
+                                        104,
+                                        249,
+                                        140,
+                                        262
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                104,
+                                                249,
+                                                140,
+                                                262
+                                            ],
+                                            "score": 1.0,
+                                            "content": "pattern.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 6
+                                }
+                            ],
+                            "index": 4.5
+                        }
+                    ],
+                    "index": 2.75
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        270,
+                        505,
+                        304
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                270,
+                                505,
+                                284
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        270,
+                                        505,
+                                        284
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Table 1: Inception V3: Preliminary quantitative results after 100 Epochs. Note the smallest sparse",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                282,
+                                505,
+                                294
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        282,
+                                        505,
+                                        294
+                                    ],
+                                    "score": 1.0,
+                                    "content": "network is actually a hybrid network - we used both depth multiplier (0.5) and sparsity (0.01). The",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                291,
+                                425,
+                                306
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        291,
+                                        425,
+                                        306
+                                    ],
+                                    "score": 1.0,
+                                    "content": "number of parameters is the number of parameters excluding the softmax layer.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        }
+                    ],
+                    "index": 8,
+                    "bbox_fs": [
+                        105,
+                        270,
+                        505,
+                        306
+                    ]
+                },
+                {
+                    "type": "table",
+                    "bbox": [
+                        132,
+                        324,
+                        298,
+                        426
+                    ],
+                    "blocks": [
+                        {
+                            "type": "table_caption",
+                            "bbox": [
+                                149,
+                                314,
+                                283,
+                                324
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        147,
+                                        312,
+                                        283,
+                                        326
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                147,
+                                                312,
+                                                283,
+                                                326
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Accuracy for sparse convolutions",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 10
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "type": "table_body",
+                            "bbox": [
+                                132,
+                                324,
+                                298,
+                                426
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        132,
+                                        324,
+                                        298,
+                                        426
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                132,
+                                                324,
+                                                298,
+                                                426
+                                            ],
+                                            "score": 0.969,
+                                            "html": "<table><tr><td>Sparsity</td><td>MAdds Params</td><td>P@1</td></tr><tr><td>0.50/0.01 0.003</td><td>43.0M 90k 158k</td><td>40.3 46.1</td></tr><tr><td>0.01</td><td>82.0M 104M</td><td>52.3</td></tr><tr><td>0.03</td><td>287k 208M 724k</td><td></td></tr><tr><td>0.10</td><td></td><td>59.5</td></tr><tr><td>0.30</td><td>628M 2.3 M</td><td>67.2</td></tr><tr><td></td><td>1.80 B 6.6M</td><td>73</td></tr><tr><td>0.60</td><td>3.50B 13M 22 M</td><td>75</td></tr><tr><td>1.00</td><td>5.70 B</td><td>77</td></tr></table>",
+                                            "type": "table",
+                                            "image_path": "b34ddde6d213af5b60d226747df5ee3604eee0c1cb8d2d5689b2231902ee2059.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 14,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        132,
+                                        324,
+                                        298,
+                                        338.57142857142856
+                                    ],
+                                    "spans": [],
+                                    "index": 12
+                                },
+                                {
+                                    "bbox": [
+                                        132,
+                                        338.57142857142856,
+                                        298,
+                                        353.1428571428571
+                                    ],
+                                    "spans": [],
+                                    "index": 13
+                                },
+                                {
+                                    "bbox": [
+                                        132,
+                                        353.1428571428571,
+                                        298,
+                                        367.71428571428567
+                                    ],
+                                    "spans": [],
+                                    "index": 11
+                                },
+                                {
+                                    "bbox": [
+                                        132,
+                                        367.71428571428567,
+                                        298,
+                                        382.2857142857142
+                                    ],
+                                    "spans": [],
+                                    "index": 14
+                                },
+                                {
+                                    "bbox": [
+                                        132,
+                                        382.2857142857142,
+                                        298,
+                                        396.8571428571428
+                                    ],
+                                    "spans": [],
+                                    "index": 15
+                                },
+                                {
+                                    "bbox": [
+                                        132,
+                                        396.8571428571428,
+                                        298,
+                                        411.42857142857133
+                                    ],
+                                    "spans": [],
+                                    "index": 16
+                                },
+                                {
+                                    "bbox": [
+                                        132,
+                                        411.42857142857133,
+                                        298,
+                                        425.9999999999999
+                                    ],
+                                    "spans": [],
+                                    "index": 17
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 12.0
+                },
+                {
+                    "type": "table",
+                    "bbox": [
+                        327,
+                        335,
+                        499,
+                        415
+                    ],
+                    "blocks": [
+                        {
+                            "type": "table_caption",
+                            "bbox": [
+                                352,
+                                313,
+                                475,
+                                325
+                            ],
+                            "group_id": 1,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        351,
+                                        312,
+                                        477,
+                                        327
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                351,
+                                                312,
+                                                477,
+                                                327
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Accuracy for Depth Multiplier",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 18
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "type": "table_body",
+                            "bbox": [
+                                327,
+                                335,
+                                499,
+                                415
+                            ],
+                            "group_id": 1,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        327,
+                                        335,
+                                        499,
+                                        415
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                327,
+                                                335,
+                                                499,
+                                                415
+                                            ],
+                                            "score": 0.974,
+                                            "html": "<table><tr><td>Multiplier</td><td>MAdds</td><td>Params P@1</td></tr><tr><td>0.05</td><td>55.0M 56k</td><td>24.6</td></tr><tr><td>0.10</td><td>75.0M</td><td>170k 38.6</td></tr><tr><td>0.20</td><td>183M 718k</td><td>54.2</td></tr><tr><td>0.30</td><td>439M 1.8M</td><td>64.0</td></tr><tr><td>0.50</td><td>1.40B 5.4M</td><td>72.3</td></tr><tr><td>0.80</td><td>3.40B 13M</td><td>75.6</td></tr></table>",
+                                            "type": "table",
+                                            "image_path": "f13c73e145cd97e43929b7f6bff27130ece79ed9e82f3f37297cc3a19a01a5c9.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 21.5,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        327,
+                                        335,
+                                        499,
+                                        348.3333333333333
+                                    ],
+                                    "spans": [],
+                                    "index": 19
+                                },
+                                {
+                                    "bbox": [
+                                        327,
+                                        348.3333333333333,
+                                        499,
+                                        361.66666666666663
+                                    ],
+                                    "spans": [],
+                                    "index": 20
+                                },
+                                {
+                                    "bbox": [
+                                        327,
+                                        361.66666666666663,
+                                        499,
+                                        374.99999999999994
+                                    ],
+                                    "spans": [],
+                                    "index": 21
+                                },
+                                {
+                                    "bbox": [
+                                        327,
+                                        374.99999999999994,
+                                        499,
+                                        388.33333333333326
+                                    ],
+                                    "spans": [],
+                                    "index": 22
+                                },
+                                {
+                                    "bbox": [
+                                        327,
+                                        388.33333333333326,
+                                        499,
+                                        401.6666666666666
+                                    ],
+                                    "spans": [],
+                                    "index": 23
+                                },
+                                {
+                                    "bbox": [
+                                        327,
+                                        401.6666666666666,
+                                        499,
+                                        414.9999999999999
+                                    ],
+                                    "spans": [],
+                                    "index": 24
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 19.75
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        119,
+                        435,
+                        312,
+                        447
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                117,
+                                433,
+                                313,
+                                449
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        117,
+                                        433,
+                                        313,
+                                        449
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Original network: 5.70 B 22 M 77 (78.8)",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        }
+                    ],
+                    "index": 25,
+                    "bbox_fs": [
+                        117,
+                        433,
+                        313,
+                        449
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        471,
+                        243,
+                        483
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                470,
+                                244,
+                                485
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        470,
+                                        244,
+                                        485
+                                    ],
+                                    "score": 1.0,
+                                    "content": "5.2.3 VGG-16 ON IMAGENET",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        }
+                    ],
+                    "index": 26
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        493,
+                        505,
+                        560
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                494,
+                                505,
+                                506
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        494,
+                                        505,
+                                        506
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In our experiments with the VGG-16 network (Simonyan & Zisserman, 2015), we modify the model",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                504,
+                                505,
+                                517
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        504,
+                                        505,
+                                        517
+                                    ],
+                                    "score": 1.0,
+                                    "content": "architecture (calling it VGG-16n) by removing the two fully-connected layers with depth 4096 and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                515,
+                                505,
+                                528
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        515,
+                                        196,
+                                        528
+                                    ],
+                                    "score": 1.0,
+                                    "content": "replacing them with a",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        197,
+                                        516,
+                                        221,
+                                        527
+                                    ],
+                                    "score": 0.88,
+                                    "content": "2 \\times 2",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        221,
+                                        515,
+                                        340,
+                                        528
+                                    ],
+                                    "score": 1.0,
+                                    "content": "maxpool layer followed by a",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        340,
+                                        516,
+                                        364,
+                                        527
+                                    ],
+                                    "score": 0.91,
+                                    "content": "3 \\times 3",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        365,
+                                        515,
+                                        505,
+                                        528
+                                    ],
+                                    "score": 1.0,
+                                    "content": "convolutional layer with the depth",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                526,
+                                506,
+                                540
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        526,
+                                        506,
+                                        540
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of 1024. This alone sped up our training significantly. The comparison between depth multiplier",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                538,
+                                506,
+                                550
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        538,
+                                        506,
+                                        550
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and sparse connection approaches is shown in Fig. 5. The modified VGG-16n network has about 7",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                550,
+                                373,
+                                561
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        550,
+                                        373,
+                                        561
+                                    ],
+                                    "score": 1.0,
+                                    "content": "times fewer parameters, but appears to have comparable precision.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        }
+                    ],
+                    "index": 29.5,
+                    "bbox_fs": [
+                        105,
+                        494,
+                        506,
+                        561
+                    ]
+                },
+                {
+                    "type": "image",
+                    "bbox": [
+                        136,
+                        582,
+                        470,
+                        710
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                136,
+                                582,
+                                470,
+                                710
+                            ],
+                            "group_id": 1,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        136,
+                                        582,
+                                        470,
+                                        710
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                136,
+                                                582,
+                                                470,
+                                                710
+                                            ],
+                                            "score": 0.963,
+                                            "type": "image",
+                                            "image_path": "f4cf6c184c4020dc5874321b477ae62c97ddf25c9d68322157c1a605a04d32e7.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 34,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        136,
+                                        582,
+                                        470,
+                                        624.6666666666666
+                                    ],
+                                    "spans": [],
+                                    "index": 33
+                                },
+                                {
+                                    "bbox": [
+                                        136,
+                                        624.6666666666666,
+                                        470,
+                                        667.3333333333333
+                                    ],
+                                    "spans": [],
+                                    "index": 34
+                                },
+                                {
+                                    "bbox": [
+                                        136,
+                                        667.3333333333333,
+                                        470,
+                                        709.9999999999999
+                                    ],
+                                    "spans": [],
+                                    "index": 35
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                106,
+                                712,
+                                506,
+                                734
+                            ],
+                            "group_id": 1,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        711,
+                                        505,
+                                        723
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                711,
+                                                505,
+                                                723
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 3: Comparison of accuracy (averaged over 2 rounds) vs. Number of parameters/Number of",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 36
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        722,
+                                        413,
+                                        735
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                722,
+                                                413,
+                                                735
+                                            ],
+                                            "score": 1.0,
+                                            "content": "multiply-adds between dense and sparse convolutions on CIFAR-10 dataset.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 37
+                                }
+                            ],
+                            "index": 36.5
+                        }
+                    ],
+                    "index": 35.25
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "image",
+                    "bbox": [
+                        125,
+                        81,
+                        482,
+                        217
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                125,
+                                81,
+                                482,
+                                217
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        125,
+                                        81,
+                                        482,
+                                        217
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                125,
+                                                81,
+                                                482,
+                                                217
+                                            ],
+                                            "score": 0.964,
+                                            "type": "image",
+                                            "image_path": "1cf4cc4464141afef87dfda693d69954f6e151da17bb74348a13d18643a16d98.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 1,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        125,
+                                        81,
+                                        482,
+                                        126.33333333333334
+                                    ],
+                                    "spans": [],
+                                    "index": 0
+                                },
+                                {
+                                    "bbox": [
+                                        125,
+                                        126.33333333333334,
+                                        482,
+                                        171.66666666666669
+                                    ],
+                                    "spans": [],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        125,
+                                        171.66666666666669,
+                                        482,
+                                        217.00000000000003
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                105,
+                                218,
+                                505,
+                                252
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        217,
+                                        506,
+                                        231
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                217,
+                                                321,
+                                                231
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 4: Inception V3: Comparison of Precision",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                322,
+                                                219,
+                                                336,
+                                                228
+                                            ],
+                                            "score": 0.5,
+                                            "content": "@ 1",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                336,
+                                                217,
+                                                506,
+                                                231
+                                            ],
+                                            "score": 1.0,
+                                            "content": "vs. Number of parameters/Number of",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 3
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        229,
+                                        505,
+                                        241
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                229,
+                                                505,
+                                                241
+                                            ],
+                                            "score": 1.0,
+                                            "content": "multiply-adds between dense and sparse convolutions on ImageNet/Inception-V3. The full network",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 4
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        240,
+                                        302,
+                                        253
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                240,
+                                                302,
+                                                253
+                                            ],
+                                            "score": 1.0,
+                                            "content": "corresponds to the right-most point of the curve.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 5
+                                }
+                            ],
+                            "index": 4
+                        }
+                    ],
+                    "index": 2.5
+                },
+                {
+                    "type": "image",
+                    "bbox": [
+                        126,
+                        266,
+                        481,
+                        402
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                126,
+                                266,
+                                481,
+                                402
+                            ],
+                            "group_id": 1,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        266,
+                                        481,
+                                        402
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                126,
+                                                266,
+                                                481,
+                                                402
+                                            ],
+                                            "score": 0.968,
+                                            "type": "image",
+                                            "image_path": "a57e0b1ff2bb1e2c18547e1723279c8b784b21823938c56fcb8383c2d7879efa.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 7,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        266,
+                                        481,
+                                        311.3333333333333
+                                    ],
+                                    "spans": [],
+                                    "index": 6
+                                },
+                                {
+                                    "bbox": [
+                                        126,
+                                        311.3333333333333,
+                                        481,
+                                        356.66666666666663
+                                    ],
+                                    "spans": [],
+                                    "index": 7
+                                },
+                                {
+                                    "bbox": [
+                                        126,
+                                        356.66666666666663,
+                                        481,
+                                        401.99999999999994
+                                    ],
+                                    "spans": [],
+                                    "index": 8
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                106,
+                                404,
+                                506,
+                                459
+                            ],
+                            "group_id": 1,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        403,
+                                        506,
+                                        416
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                403,
+                                                427,
+                                                416
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 5: VGG 16: Preliminary Quantitative Results. Comparison of Precision",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                427,
+                                                404,
+                                                442,
+                                                414
+                                            ],
+                                            "score": 0.77,
+                                            "content": "@ 1",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                442,
+                                                403,
+                                                506,
+                                                416
+                                            ],
+                                            "score": 1.0,
+                                            "content": "vs. Number of",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 9
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        414,
+                                        505,
+                                        427
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                414,
+                                                505,
+                                                427
+                                            ],
+                                            "score": 1.0,
+                                            "content": "parameters/Number of multiply-adds between dense and sparse convolutions on ImageNet/VGG-",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 10
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        424,
+                                        505,
+                                        438
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                424,
+                                                505,
+                                                438
+                                            ],
+                                            "score": 1.0,
+                                            "content": "16n. The full network corresponds to the right-most point of the curve. Original VGG-16 as de-",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 11
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        436,
+                                        506,
+                                        449
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                436,
+                                                506,
+                                                449
+                                            ],
+                                            "score": 1.0,
+                                            "content": "scribed in Simonyan & Zisserman (2015) (blue star) and the same model trained by us from scratch",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 12
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        446,
+                                        215,
+                                        460
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                446,
+                                                215,
+                                                460
+                                            ],
+                                            "score": 1.0,
+                                            "content": "(red cross) are also shown.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 13
+                                }
+                            ],
+                            "index": 11
+                        }
+                    ],
+                    "index": 9.0
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        483,
+                        240,
+                        493
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                483,
+                                240,
+                                495
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        483,
+                                        240,
+                                        495
+                                    ],
+                                    "score": 1.0,
+                                    "content": "5.3 INCREMENTAL TRAINING",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        }
+                    ],
+                    "index": 14
+                },
+                {
+                    "type": "image",
+                    "bbox": [
+                        190,
+                        521,
+                        414,
+                        698
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                190,
+                                521,
+                                414,
+                                698
+                            ],
+                            "group_id": 2,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        190,
+                                        521,
+                                        414,
+                                        698
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                190,
+                                                521,
+                                                414,
+                                                698
+                                            ],
+                                            "score": 0.959,
+                                            "type": "image",
+                                            "image_path": "f704c8a1c6906e9b1cd74e0b702c578d2acd95067c474ee8b78c09edf9df95a7.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 21,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        190,
+                                        521,
+                                        414,
+                                        534.6153846153846
+                                    ],
+                                    "spans": [],
+                                    "index": 15
+                                },
+                                {
+                                    "bbox": [
+                                        190,
+                                        534.6153846153846,
+                                        414,
+                                        548.2307692307693
+                                    ],
+                                    "spans": [],
+                                    "index": 16
+                                },
+                                {
+                                    "bbox": [
+                                        190,
+                                        548.2307692307693,
+                                        414,
+                                        561.8461538461539
+                                    ],
+                                    "spans": [],
+                                    "index": 17
+                                },
+                                {
+                                    "bbox": [
+                                        190,
+                                        561.8461538461539,
+                                        414,
+                                        575.4615384615386
+                                    ],
+                                    "spans": [],
+                                    "index": 18
+                                },
+                                {
+                                    "bbox": [
+                                        190,
+                                        575.4615384615386,
+                                        414,
+                                        589.0769230769232
+                                    ],
+                                    "spans": [],
+                                    "index": 19
+                                },
+                                {
+                                    "bbox": [
+                                        190,
+                                        589.0769230769232,
+                                        414,
+                                        602.6923076923078
+                                    ],
+                                    "spans": [],
+                                    "index": 20
+                                },
+                                {
+                                    "bbox": [
+                                        190,
+                                        602.6923076923078,
+                                        414,
+                                        616.3076923076925
+                                    ],
+                                    "spans": [],
+                                    "index": 21
+                                },
+                                {
+                                    "bbox": [
+                                        190,
+                                        616.3076923076925,
+                                        414,
+                                        629.9230769230771
+                                    ],
+                                    "spans": [],
+                                    "index": 22
+                                },
+                                {
+                                    "bbox": [
+                                        190,
+                                        629.9230769230771,
+                                        414,
+                                        643.5384615384618
+                                    ],
+                                    "spans": [],
+                                    "index": 23
+                                },
+                                {
+                                    "bbox": [
+                                        190,
+                                        643.5384615384618,
+                                        414,
+                                        657.1538461538464
+                                    ],
+                                    "spans": [],
+                                    "index": 24
+                                },
+                                {
+                                    "bbox": [
+                                        190,
+                                        657.1538461538464,
+                                        414,
+                                        670.7692307692311
+                                    ],
+                                    "spans": [],
+                                    "index": 25
+                                },
+                                {
+                                    "bbox": [
+                                        190,
+                                        670.7692307692311,
+                                        414,
+                                        684.3846153846157
+                                    ],
+                                    "spans": [],
+                                    "index": 26
+                                },
+                                {
+                                    "bbox": [
+                                        190,
+                                        684.3846153846157,
+                                        414,
+                                        698.0000000000003
+                                    ],
+                                    "spans": [],
+                                    "index": 27
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                106,
+                                700,
+                                505,
+                                734
+                            ],
+                            "group_id": 2,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        699,
+                                        505,
+                                        713
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                699,
+                                                378,
+                                                713
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 6: Incremental Training Of Inception V3: We show Precision",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                379,
+                                                701,
+                                                393,
+                                                711
+                                            ],
+                                            "score": 0.74,
+                                            "content": "@ 1",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                394,
+                                                699,
+                                                505,
+                                                713
+                                            ],
+                                            "score": 1.0,
+                                            "content": "during the training process,",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 28
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        711,
+                                        505,
+                                        723
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                711,
+                                                505,
+                                                723
+                                            ],
+                                            "score": 1.0,
+                                            "content": "where the networks densify over time. The saturation points show where the networks actually reach",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 29
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        721,
+                                        177,
+                                        735
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                721,
+                                                177,
+                                                735
+                                            ],
+                                            "score": 1.0,
+                                            "content": "their full density.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 30
+                                }
+                            ],
+                            "index": 29
+                        }
+                    ],
+                    "index": 25.0
+                }
+            ],
+            "page_idx": 7,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        106,
+                        27,
+                        308,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                107,
+                                26,
+                                308,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        26,
+                                        308,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Under review as a conference paper at ICLR 2017",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        302,
+                        751,
+                        308,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                300,
+                                750,
+                                309,
+                                761
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        300,
+                                        750,
+                                        309,
+                                        761
+                                    ],
+                                    "score": 1.0,
+                                    "content": "8",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "image",
+                    "bbox": [
+                        125,
+                        81,
+                        482,
+                        217
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                125,
+                                81,
+                                482,
+                                217
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        125,
+                                        81,
+                                        482,
+                                        217
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                125,
+                                                81,
+                                                482,
+                                                217
+                                            ],
+                                            "score": 0.964,
+                                            "type": "image",
+                                            "image_path": "1cf4cc4464141afef87dfda693d69954f6e151da17bb74348a13d18643a16d98.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 1,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        125,
+                                        81,
+                                        482,
+                                        126.33333333333334
+                                    ],
+                                    "spans": [],
+                                    "index": 0
+                                },
+                                {
+                                    "bbox": [
+                                        125,
+                                        126.33333333333334,
+                                        482,
+                                        171.66666666666669
+                                    ],
+                                    "spans": [],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        125,
+                                        171.66666666666669,
+                                        482,
+                                        217.00000000000003
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                105,
+                                218,
+                                505,
+                                252
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        217,
+                                        506,
+                                        231
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                217,
+                                                321,
+                                                231
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 4: Inception V3: Comparison of Precision",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                322,
+                                                219,
+                                                336,
+                                                228
+                                            ],
+                                            "score": 0.5,
+                                            "content": "@ 1",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                336,
+                                                217,
+                                                506,
+                                                231
+                                            ],
+                                            "score": 1.0,
+                                            "content": "vs. Number of parameters/Number of",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 3
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        229,
+                                        505,
+                                        241
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                229,
+                                                505,
+                                                241
+                                            ],
+                                            "score": 1.0,
+                                            "content": "multiply-adds between dense and sparse convolutions on ImageNet/Inception-V3. The full network",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 4
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        240,
+                                        302,
+                                        253
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                240,
+                                                302,
+                                                253
+                                            ],
+                                            "score": 1.0,
+                                            "content": "corresponds to the right-most point of the curve.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 5
+                                }
+                            ],
+                            "index": 4
+                        }
+                    ],
+                    "index": 2.5
+                },
+                {
+                    "type": "image",
+                    "bbox": [
+                        126,
+                        266,
+                        481,
+                        402
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                126,
+                                266,
+                                481,
+                                402
+                            ],
+                            "group_id": 1,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        266,
+                                        481,
+                                        402
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                126,
+                                                266,
+                                                481,
+                                                402
+                                            ],
+                                            "score": 0.968,
+                                            "type": "image",
+                                            "image_path": "a57e0b1ff2bb1e2c18547e1723279c8b784b21823938c56fcb8383c2d7879efa.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 7,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        266,
+                                        481,
+                                        311.3333333333333
+                                    ],
+                                    "spans": [],
+                                    "index": 6
+                                },
+                                {
+                                    "bbox": [
+                                        126,
+                                        311.3333333333333,
+                                        481,
+                                        356.66666666666663
+                                    ],
+                                    "spans": [],
+                                    "index": 7
+                                },
+                                {
+                                    "bbox": [
+                                        126,
+                                        356.66666666666663,
+                                        481,
+                                        401.99999999999994
+                                    ],
+                                    "spans": [],
+                                    "index": 8
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                106,
+                                404,
+                                506,
+                                459
+                            ],
+                            "group_id": 1,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        403,
+                                        506,
+                                        416
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                403,
+                                                427,
+                                                416
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 5: VGG 16: Preliminary Quantitative Results. Comparison of Precision",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                427,
+                                                404,
+                                                442,
+                                                414
+                                            ],
+                                            "score": 0.77,
+                                            "content": "@ 1",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                442,
+                                                403,
+                                                506,
+                                                416
+                                            ],
+                                            "score": 1.0,
+                                            "content": "vs. Number of",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 9
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        414,
+                                        505,
+                                        427
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                414,
+                                                505,
+                                                427
+                                            ],
+                                            "score": 1.0,
+                                            "content": "parameters/Number of multiply-adds between dense and sparse convolutions on ImageNet/VGG-",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 10
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        424,
+                                        505,
+                                        438
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                424,
+                                                505,
+                                                438
+                                            ],
+                                            "score": 1.0,
+                                            "content": "16n. The full network corresponds to the right-most point of the curve. Original VGG-16 as de-",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 11
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        436,
+                                        506,
+                                        449
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                436,
+                                                506,
+                                                449
+                                            ],
+                                            "score": 1.0,
+                                            "content": "scribed in Simonyan & Zisserman (2015) (blue star) and the same model trained by us from scratch",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 12
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        446,
+                                        215,
+                                        460
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                446,
+                                                215,
+                                                460
+                                            ],
+                                            "score": 1.0,
+                                            "content": "(red cross) are also shown.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 13
+                                }
+                            ],
+                            "index": 11
+                        }
+                    ],
+                    "index": 9.0
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        483,
+                        240,
+                        493
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                483,
+                                240,
+                                495
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        483,
+                                        240,
+                                        495
+                                    ],
+                                    "score": 1.0,
+                                    "content": "5.3 INCREMENTAL TRAINING",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        }
+                    ],
+                    "index": 14
+                },
+                {
+                    "type": "image",
+                    "bbox": [
+                        190,
+                        521,
+                        414,
+                        698
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                190,
+                                521,
+                                414,
+                                698
+                            ],
+                            "group_id": 2,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        190,
+                                        521,
+                                        414,
+                                        698
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                190,
+                                                521,
+                                                414,
+                                                698
+                                            ],
+                                            "score": 0.959,
+                                            "type": "image",
+                                            "image_path": "f704c8a1c6906e9b1cd74e0b702c578d2acd95067c474ee8b78c09edf9df95a7.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 21,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        190,
+                                        521,
+                                        414,
+                                        534.6153846153846
+                                    ],
+                                    "spans": [],
+                                    "index": 15
+                                },
+                                {
+                                    "bbox": [
+                                        190,
+                                        534.6153846153846,
+                                        414,
+                                        548.2307692307693
+                                    ],
+                                    "spans": [],
+                                    "index": 16
+                                },
+                                {
+                                    "bbox": [
+                                        190,
+                                        548.2307692307693,
+                                        414,
+                                        561.8461538461539
+                                    ],
+                                    "spans": [],
+                                    "index": 17
+                                },
+                                {
+                                    "bbox": [
+                                        190,
+                                        561.8461538461539,
+                                        414,
+                                        575.4615384615386
+                                    ],
+                                    "spans": [],
+                                    "index": 18
+                                },
+                                {
+                                    "bbox": [
+                                        190,
+                                        575.4615384615386,
+                                        414,
+                                        589.0769230769232
+                                    ],
+                                    "spans": [],
+                                    "index": 19
+                                },
+                                {
+                                    "bbox": [
+                                        190,
+                                        589.0769230769232,
+                                        414,
+                                        602.6923076923078
+                                    ],
+                                    "spans": [],
+                                    "index": 20
+                                },
+                                {
+                                    "bbox": [
+                                        190,
+                                        602.6923076923078,
+                                        414,
+                                        616.3076923076925
+                                    ],
+                                    "spans": [],
+                                    "index": 21
+                                },
+                                {
+                                    "bbox": [
+                                        190,
+                                        616.3076923076925,
+                                        414,
+                                        629.9230769230771
+                                    ],
+                                    "spans": [],
+                                    "index": 22
+                                },
+                                {
+                                    "bbox": [
+                                        190,
+                                        629.9230769230771,
+                                        414,
+                                        643.5384615384618
+                                    ],
+                                    "spans": [],
+                                    "index": 23
+                                },
+                                {
+                                    "bbox": [
+                                        190,
+                                        643.5384615384618,
+                                        414,
+                                        657.1538461538464
+                                    ],
+                                    "spans": [],
+                                    "index": 24
+                                },
+                                {
+                                    "bbox": [
+                                        190,
+                                        657.1538461538464,
+                                        414,
+                                        670.7692307692311
+                                    ],
+                                    "spans": [],
+                                    "index": 25
+                                },
+                                {
+                                    "bbox": [
+                                        190,
+                                        670.7692307692311,
+                                        414,
+                                        684.3846153846157
+                                    ],
+                                    "spans": [],
+                                    "index": 26
+                                },
+                                {
+                                    "bbox": [
+                                        190,
+                                        684.3846153846157,
+                                        414,
+                                        698.0000000000003
+                                    ],
+                                    "spans": [],
+                                    "index": 27
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                106,
+                                700,
+                                505,
+                                734
+                            ],
+                            "group_id": 2,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        699,
+                                        505,
+                                        713
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                699,
+                                                378,
+                                                713
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 6: Incremental Training Of Inception V3: We show Precision",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                379,
+                                                701,
+                                                393,
+                                                711
+                                            ],
+                                            "score": 0.74,
+                                            "content": "@ 1",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                394,
+                                                699,
+                                                505,
+                                                713
+                                            ],
+                                            "score": 1.0,
+                                            "content": "during the training process,",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 28
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        711,
+                                        505,
+                                        723
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                711,
+                                                505,
+                                                723
+                                            ],
+                                            "score": 1.0,
+                                            "content": "where the networks densify over time. The saturation points show where the networks actually reach",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 29
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        721,
+                                        177,
+                                        735
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                721,
+                                                177,
+                                                735
+                                            ],
+                                            "score": 1.0,
+                                            "content": "their full density.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 30
+                                }
+                            ],
+                            "index": 29
+                        }
+                    ],
+                    "index": 25.0
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        82,
+                        505,
+                        137
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                82,
+                                506,
+                                95
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        82,
+                                        506,
+                                        95
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Finally, we show that incremental training is a promising direction. We start with a very sparse",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                93,
+                                506,
+                                106
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        93,
+                                        506,
+                                        106
+                                    ],
+                                    "score": 1.0,
+                                    "content": "model and increase its density over time, using the approach described in Sect. 3.2.1. We note that",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                105,
+                                505,
+                                117
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        105,
+                                        505,
+                                        117
+                                    ],
+                                    "score": 1.0,
+                                    "content": "a naive approach where we simply add filters results in training process basically equivalent to as if",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                115,
+                                505,
+                                128
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        115,
+                                        505,
+                                        128
+                                    ],
+                                    "score": 1.0,
+                                    "content": "it started from scratch in every step. On the other hand, when the network densifies over time, all",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                126,
+                                445,
+                                138
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        126,
+                                        445,
+                                        138
+                                    ],
+                                    "score": 1.0,
+                                    "content": "channels already possess some discriminative power and that information is utilized.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        }
+                    ],
+                    "index": 2
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        143,
+                        505,
+                        242
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                144,
+                                505,
+                                156
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        144,
+                                        391,
+                                        156
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In our experiments, we initially start training Inception-V3 with only",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        392,
+                                        144,
+                                        406,
+                                        154
+                                    ],
+                                    "score": 0.84,
+                                    "content": "1 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        406,
+                                        144,
+                                        419,
+                                        156
+                                    ],
+                                    "score": 1.0,
+                                    "content": "or",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        419,
+                                        144,
+                                        441,
+                                        154
+                                    ],
+                                    "score": 0.86,
+                                    "content": "0 . 1 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        442,
+                                        144,
+                                        505,
+                                        156
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of connections",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                154,
+                                505,
+                                167
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        154,
+                                        353,
+                                        167
+                                    ],
+                                    "score": 1.0,
+                                    "content": "enabled. Then, we double the number of connections every",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        354,
+                                        155,
+                                        362,
+                                        164
+                                    ],
+                                    "score": 0.78,
+                                    "content": "T",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        363,
+                                        154,
+                                        424,
+                                        167
+                                    ],
+                                    "score": 1.0,
+                                    "content": "steps. We use",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        424,
+                                        154,
+                                        479,
+                                        165
+                                    ],
+                                    "score": 0.46,
+                                    "content": "T = 1 0 , 0 0 0",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        479,
+                                        154,
+                                        483,
+                                        167
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        483,
+                                        154,
+                                        505,
+                                        165
+                                    ],
+                                    "score": 0.79,
+                                    "content": "T =",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                164,
+                                506,
+                                178
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        164,
+                                        154,
+                                        178
+                                    ],
+                                    "score": 1.0,
+                                    "content": "25, 000 and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        154,
+                                        165,
+                                        205,
+                                        176
+                                    ],
+                                    "score": 0.76,
+                                    "content": "T = 5 0 , 0 0 0",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        206,
+                                        164,
+                                        506,
+                                        178
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". The results are presented in Fig. 6. We show that the networks trained with",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                176,
+                                505,
+                                189
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        176,
+                                        505,
+                                        189
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the incremental approach regardless of the doubling period can catch up with the full Inception-V3",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                187,
+                                505,
+                                201
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        187,
+                                        505,
+                                        201
+                                    ],
+                                    "score": 1.0,
+                                    "content": "network (in some cases with small gains). Moreover, they recover very quickly from adding more",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                198,
+                                505,
+                                211
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        198,
+                                        505,
+                                        211
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(untrained) connections. In fact, the recovery is so fast that it is shorter than our saving interval",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                208,
+                                505,
+                                223
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        208,
+                                        505,
+                                        223
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for all the networks except for the network with 10K doubling period (resulting in the sharp drop).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                218,
+                                505,
+                                234
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        218,
+                                        505,
+                                        234
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We believe that incremental training is a promising direction to speeding up the training of large",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                230,
+                                488,
+                                244
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        230,
+                                        488,
+                                        244
+                                    ],
+                                    "score": 1.0,
+                                    "content": "convolutional neural networks since early stages of the training require much less computation.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        }
+                    ],
+                    "index": 9
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        254,
+                        195,
+                        267
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                252,
+                                197,
+                                270
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        252,
+                                        197,
+                                        270
+                                    ],
+                                    "score": 1.0,
+                                    "content": "6 CONCLUSION",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        }
+                    ],
+                    "index": 14
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        279,
+                        505,
+                        410
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                277,
+                                506,
+                                293
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        277,
+                                        506,
+                                        293
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We have proposed a new compression technique that uses a sparse random connection structure",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                289,
+                                506,
+                                303
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        289,
+                                        506,
+                                        303
+                                    ],
+                                    "score": 1.0,
+                                    "content": "between input-output filters in convolutional layers of CNNs. We fix this structure before training",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                301,
+                                505,
+                                312
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        301,
+                                        505,
+                                        312
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and use the same structure across spatial dimensions to harvest savings from modern hardware. We",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                312,
+                                505,
+                                324
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        312,
+                                        505,
+                                        324
+                                    ],
+                                    "score": 1.0,
+                                    "content": "show that this approach is especially useful at very high compression rates for large networks. For",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                322,
+                                505,
+                                335
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        322,
+                                        505,
+                                        335
+                                    ],
+                                    "score": 1.0,
+                                    "content": "example, this simple method when applied to Inception V3 (Fig. 4), achieves AlexNet-level accuracy",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                333,
+                                505,
+                                346
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        333,
+                                        505,
+                                        346
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(Krizhevsky et al., 2012) with fewer than 400K parameters and VGG-level one (Fig. 5) with roughly",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                344,
+                                505,
+                                357
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        344,
+                                        129,
+                                        356
+                                    ],
+                                    "score": 0.26,
+                                    "content": "3 . 5 \\mathrm { M }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        130,
+                                        345,
+                                        505,
+                                        357
+                                    ],
+                                    "score": 1.0,
+                                    "content": "parameters. The simplicity of our approach is instructive in that it establishes a strong baseline",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                354,
+                                505,
+                                369
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        354,
+                                        505,
+                                        369
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to compare against when developing more advanced techniques. On the other hand, the uncanny",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                366,
+                                506,
+                                380
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        366,
+                                        443,
+                                        380
+                                    ],
+                                    "score": 1.0,
+                                    "content": "match in performance of dense and equivalently-sized sparse networks with sparsity",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        443,
+                                        367,
+                                        468,
+                                        377
+                                    ],
+                                    "score": 0.86,
+                                    "content": "> 0 . 1",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        468,
+                                        366,
+                                        506,
+                                        380
+                                    ],
+                                    "score": 1.0,
+                                    "content": "suggests",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                378,
+                                505,
+                                390
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        378,
+                                        505,
+                                        390
+                                    ],
+                                    "score": 1.0,
+                                    "content": "that there might be some fundamental property of network architectures that is controlled by the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                389,
+                                505,
+                                401
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        389,
+                                        505,
+                                        401
+                                    ],
+                                    "score": 1.0,
+                                    "content": "number of parameters, regardless of how they are organized. Exploring this further might yield",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                400,
+                                322,
+                                411
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        400,
+                                        322,
+                                        411
+                                    ],
+                                    "score": 1.0,
+                                    "content": "additional insights on understanding neural networks.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        }
+                    ],
+                    "index": 20.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        416,
+                        505,
+                        471
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                416,
+                                505,
+                                428
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        416,
+                                        505,
+                                        428
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In addition, we show that our method leads to an interesting novel incremental training technique,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                426,
+                                505,
+                                439
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        426,
+                                        505,
+                                        439
+                                    ],
+                                    "score": 1.0,
+                                    "content": "where we take advantage of sparse (and smaller) models to build a dense network. One interesting",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                438,
+                                506,
+                                451
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        438,
+                                        506,
+                                        451
+                                    ],
+                                    "score": 1.0,
+                                    "content": "open direction is to enable incremental training not to simply densify the network over time, but also",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                449,
+                                505,
+                                461
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        449,
+                                        505,
+                                        461
+                                    ],
+                                    "score": 1.0,
+                                    "content": "increase the number of channels. This would allow us to grow the network without having to fix its",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                460,
+                                201,
+                                474
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        460,
+                                        201,
+                                        474
+                                    ],
+                                    "score": 1.0,
+                                    "content": "original shape in place.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        }
+                    ],
+                    "index": 29
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        484,
+                        175,
+                        496
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                484,
+                                176,
+                                498
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        484,
+                                        176,
+                                        498
+                                    ],
+                                    "score": 1.0,
+                                    "content": "REFERENCES",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        }
+                    ],
+                    "index": 32
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        507,
+                        505,
+                        588
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                507,
+                                505,
+                                519
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        507,
+                                        505,
+                                        519
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Mart´ın Abadi, Ashish Agarwal, Paul Barham, Eugene Brevdo, Zhifeng Chen, Craig Citro, Greg S. Corrado,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                516,
+                                505,
+                                531
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        516,
+                                        505,
+                                        531
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Andy Davis, Jeffrey Dean, Matthieu Devin, Sanjay Ghemawat, Ian Goodfellow, Andrew Harp, Geoffrey",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                114,
+                                525,
+                                506,
+                                541
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        114,
+                                        525,
+                                        506,
+                                        541
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Irving, Michael Isard, Yangqing Jia, Rafal Jozefowicz, Lukasz Kaiser, Manjunath Kudlur, Josh Levenberg,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                536,
+                                506,
+                                550
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        536,
+                                        506,
+                                        550
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Dan Mane, Rajat Monga, Sherry Moore, Derek Murray, Chris Olah, Mike Schuster, Jonathon Shlens, Benoit ´",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                547,
+                                505,
+                                560
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        547,
+                                        505,
+                                        560
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Steiner, Ilya Sutskever, Kunal Talwar, Paul Tucker, Vincent Vanhoucke, Vijay Vasudevan, Fernanda Viegas, ´",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                557,
+                                505,
+                                569
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        557,
+                                        505,
+                                        569
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Oriol Vinyals, Pete Warden, Martin Wattenberg, Martin Wicke, Yuan Yu, and Xiaoqiang Zheng. TensorFlow:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                568,
+                                505,
+                                579
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        568,
+                                        505,
+                                        579
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Large-scale machine learning on heterogeneous systems, 2015. URL http://tensorflow.org/. Soft-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                576,
+                                245,
+                                590
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        576,
+                                        245,
+                                        590
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ware available from tensorflow.org.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        }
+                    ],
+                    "index": 36.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        104,
+                        598,
+                        504,
+                        619
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                597,
+                                505,
+                                610
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        597,
+                                        505,
+                                        610
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Sajid Anwar, Kyuyeon Hwang, and Wonyong Sung. Fixed point optimization of deep convolutional neural",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                608,
+                                304,
+                                620
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        608,
+                                        304,
+                                        620
+                                    ],
+                                    "score": 1.0,
+                                    "content": "networks for object recognition. In ICASSP, 2015a.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        }
+                    ],
+                    "index": 41.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        628,
+                        503,
+                        649
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                628,
+                                505,
+                                641
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        628,
+                                        505,
+                                        641
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Sajid Anwar, Kyuyeon Hwang, and Wonyong Sung. Structured pruning of deep convolutional neural networks.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                639,
+                                267,
+                                650
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        639,
+                                        267,
+                                        650
+                                    ],
+                                    "score": 1.0,
+                                    "content": "arXiv preprint arXiv:1512.08571, 2015b.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        }
+                    ],
+                    "index": 43.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        659,
+                        503,
+                        680
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                659,
+                                504,
+                                672
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        659,
+                                        504,
+                                        672
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Sanjeev Arora, Aditya Bhaskara, Rong Ge, and Tengyu Ma. Provable bounds for learning some deep represen-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                669,
+                                204,
+                                680
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        669,
+                                        204,
+                                        680
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tations. In ICML, 2014.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        }
+                    ],
+                    "index": 45.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        690,
+                        406,
+                        701
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                689,
+                                406,
+                                703
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        689,
+                                        406,
+                                        703
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Jimmy Ba and Rich Caruana. Do deep nets really need to be deep? In NIPS, 2014.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 47
+                        }
+                    ],
+                    "index": 47
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        711,
+                        505,
+                        731
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                709,
+                                505,
+                                724
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        709,
+                                        505,
+                                        724
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Wenlin Chen, James T. Wilson, Stephen Tyree, Kilian Q. Weinberger, and Yixin Chen. Compressing neural",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 48
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                721,
+                                293,
+                                732
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        721,
+                                        293,
+                                        732
+                                    ],
+                                    "score": 1.0,
+                                    "content": "networks with the hashing trick. In ICML, 2015.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 49
+                        }
+                    ],
+                    "index": 48.5
+                }
+            ],
+            "page_idx": 8,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        107,
+                        27,
+                        308,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                107,
+                                26,
+                                308,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        26,
+                                        308,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Under review as a conference paper at ICLR 2017",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        302,
+                        751,
+                        308,
+                        759
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                302,
+                                751,
+                                309,
+                                762
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        302,
+                                        751,
+                                        309,
+                                        762
+                                    ],
+                                    "score": 1.0,
+                                    "content": "9",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        82,
+                        505,
+                        137
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                82,
+                                506,
+                                95
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        82,
+                                        506,
+                                        95
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Finally, we show that incremental training is a promising direction. We start with a very sparse",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                93,
+                                506,
+                                106
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        93,
+                                        506,
+                                        106
+                                    ],
+                                    "score": 1.0,
+                                    "content": "model and increase its density over time, using the approach described in Sect. 3.2.1. We note that",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                105,
+                                505,
+                                117
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        105,
+                                        505,
+                                        117
+                                    ],
+                                    "score": 1.0,
+                                    "content": "a naive approach where we simply add filters results in training process basically equivalent to as if",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                115,
+                                505,
+                                128
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        115,
+                                        505,
+                                        128
+                                    ],
+                                    "score": 1.0,
+                                    "content": "it started from scratch in every step. On the other hand, when the network densifies over time, all",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                126,
+                                445,
+                                138
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        126,
+                                        445,
+                                        138
+                                    ],
+                                    "score": 1.0,
+                                    "content": "channels already possess some discriminative power and that information is utilized.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        }
+                    ],
+                    "index": 2,
+                    "bbox_fs": [
+                        105,
+                        82,
+                        506,
+                        138
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        143,
+                        505,
+                        242
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                144,
+                                505,
+                                156
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        144,
+                                        391,
+                                        156
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In our experiments, we initially start training Inception-V3 with only",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        392,
+                                        144,
+                                        406,
+                                        154
+                                    ],
+                                    "score": 0.84,
+                                    "content": "1 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        406,
+                                        144,
+                                        419,
+                                        156
+                                    ],
+                                    "score": 1.0,
+                                    "content": "or",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        419,
+                                        144,
+                                        441,
+                                        154
+                                    ],
+                                    "score": 0.86,
+                                    "content": "0 . 1 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        442,
+                                        144,
+                                        505,
+                                        156
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of connections",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                154,
+                                505,
+                                167
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        154,
+                                        353,
+                                        167
+                                    ],
+                                    "score": 1.0,
+                                    "content": "enabled. Then, we double the number of connections every",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        354,
+                                        155,
+                                        362,
+                                        164
+                                    ],
+                                    "score": 0.78,
+                                    "content": "T",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        363,
+                                        154,
+                                        424,
+                                        167
+                                    ],
+                                    "score": 1.0,
+                                    "content": "steps. We use",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        424,
+                                        154,
+                                        479,
+                                        165
+                                    ],
+                                    "score": 0.46,
+                                    "content": "T = 1 0 , 0 0 0",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        479,
+                                        154,
+                                        483,
+                                        167
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        483,
+                                        154,
+                                        505,
+                                        165
+                                    ],
+                                    "score": 0.79,
+                                    "content": "T =",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                164,
+                                506,
+                                178
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        164,
+                                        154,
+                                        178
+                                    ],
+                                    "score": 1.0,
+                                    "content": "25, 000 and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        154,
+                                        165,
+                                        205,
+                                        176
+                                    ],
+                                    "score": 0.76,
+                                    "content": "T = 5 0 , 0 0 0",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        206,
+                                        164,
+                                        506,
+                                        178
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". The results are presented in Fig. 6. We show that the networks trained with",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                176,
+                                505,
+                                189
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        176,
+                                        505,
+                                        189
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the incremental approach regardless of the doubling period can catch up with the full Inception-V3",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                187,
+                                505,
+                                201
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        187,
+                                        505,
+                                        201
+                                    ],
+                                    "score": 1.0,
+                                    "content": "network (in some cases with small gains). Moreover, they recover very quickly from adding more",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                198,
+                                505,
+                                211
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        198,
+                                        505,
+                                        211
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(untrained) connections. In fact, the recovery is so fast that it is shorter than our saving interval",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                208,
+                                505,
+                                223
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        208,
+                                        505,
+                                        223
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for all the networks except for the network with 10K doubling period (resulting in the sharp drop).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                218,
+                                505,
+                                234
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        218,
+                                        505,
+                                        234
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We believe that incremental training is a promising direction to speeding up the training of large",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                230,
+                                488,
+                                244
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        230,
+                                        488,
+                                        244
+                                    ],
+                                    "score": 1.0,
+                                    "content": "convolutional neural networks since early stages of the training require much less computation.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        }
+                    ],
+                    "index": 9,
+                    "bbox_fs": [
+                        105,
+                        144,
+                        506,
+                        244
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        254,
+                        195,
+                        267
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                252,
+                                197,
+                                270
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        252,
+                                        197,
+                                        270
+                                    ],
+                                    "score": 1.0,
+                                    "content": "6 CONCLUSION",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        }
+                    ],
+                    "index": 14
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        279,
+                        505,
+                        410
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                277,
+                                506,
+                                293
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        277,
+                                        506,
+                                        293
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We have proposed a new compression technique that uses a sparse random connection structure",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                289,
+                                506,
+                                303
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        289,
+                                        506,
+                                        303
+                                    ],
+                                    "score": 1.0,
+                                    "content": "between input-output filters in convolutional layers of CNNs. We fix this structure before training",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                301,
+                                505,
+                                312
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        301,
+                                        505,
+                                        312
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and use the same structure across spatial dimensions to harvest savings from modern hardware. We",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                312,
+                                505,
+                                324
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        312,
+                                        505,
+                                        324
+                                    ],
+                                    "score": 1.0,
+                                    "content": "show that this approach is especially useful at very high compression rates for large networks. For",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                322,
+                                505,
+                                335
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        322,
+                                        505,
+                                        335
+                                    ],
+                                    "score": 1.0,
+                                    "content": "example, this simple method when applied to Inception V3 (Fig. 4), achieves AlexNet-level accuracy",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                333,
+                                505,
+                                346
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        333,
+                                        505,
+                                        346
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(Krizhevsky et al., 2012) with fewer than 400K parameters and VGG-level one (Fig. 5) with roughly",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                344,
+                                505,
+                                357
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        344,
+                                        129,
+                                        356
+                                    ],
+                                    "score": 0.26,
+                                    "content": "3 . 5 \\mathrm { M }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        130,
+                                        345,
+                                        505,
+                                        357
+                                    ],
+                                    "score": 1.0,
+                                    "content": "parameters. The simplicity of our approach is instructive in that it establishes a strong baseline",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                354,
+                                505,
+                                369
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        354,
+                                        505,
+                                        369
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to compare against when developing more advanced techniques. On the other hand, the uncanny",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                366,
+                                506,
+                                380
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        366,
+                                        443,
+                                        380
+                                    ],
+                                    "score": 1.0,
+                                    "content": "match in performance of dense and equivalently-sized sparse networks with sparsity",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        443,
+                                        367,
+                                        468,
+                                        377
+                                    ],
+                                    "score": 0.86,
+                                    "content": "> 0 . 1",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        468,
+                                        366,
+                                        506,
+                                        380
+                                    ],
+                                    "score": 1.0,
+                                    "content": "suggests",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                378,
+                                505,
+                                390
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        378,
+                                        505,
+                                        390
+                                    ],
+                                    "score": 1.0,
+                                    "content": "that there might be some fundamental property of network architectures that is controlled by the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                389,
+                                505,
+                                401
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        389,
+                                        505,
+                                        401
+                                    ],
+                                    "score": 1.0,
+                                    "content": "number of parameters, regardless of how they are organized. Exploring this further might yield",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                400,
+                                322,
+                                411
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        400,
+                                        322,
+                                        411
+                                    ],
+                                    "score": 1.0,
+                                    "content": "additional insights on understanding neural networks.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        }
+                    ],
+                    "index": 20.5,
+                    "bbox_fs": [
+                        104,
+                        277,
+                        506,
+                        411
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        416,
+                        505,
+                        471
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                416,
+                                505,
+                                428
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        416,
+                                        505,
+                                        428
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In addition, we show that our method leads to an interesting novel incremental training technique,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                426,
+                                505,
+                                439
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        426,
+                                        505,
+                                        439
+                                    ],
+                                    "score": 1.0,
+                                    "content": "where we take advantage of sparse (and smaller) models to build a dense network. One interesting",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                438,
+                                506,
+                                451
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        438,
+                                        506,
+                                        451
+                                    ],
+                                    "score": 1.0,
+                                    "content": "open direction is to enable incremental training not to simply densify the network over time, but also",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                449,
+                                505,
+                                461
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        449,
+                                        505,
+                                        461
+                                    ],
+                                    "score": 1.0,
+                                    "content": "increase the number of channels. This would allow us to grow the network without having to fix its",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                460,
+                                201,
+                                474
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        460,
+                                        201,
+                                        474
+                                    ],
+                                    "score": 1.0,
+                                    "content": "original shape in place.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        }
+                    ],
+                    "index": 29,
+                    "bbox_fs": [
+                        105,
+                        416,
+                        506,
+                        474
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        484,
+                        175,
+                        496
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                484,
+                                176,
+                                498
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        484,
+                                        176,
+                                        498
+                                    ],
+                                    "score": 1.0,
+                                    "content": "REFERENCES",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        }
+                    ],
+                    "index": 32
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        507,
+                        505,
+                        588
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                507,
+                                505,
+                                519
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        507,
+                                        505,
+                                        519
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Mart´ın Abadi, Ashish Agarwal, Paul Barham, Eugene Brevdo, Zhifeng Chen, Craig Citro, Greg S. Corrado,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                516,
+                                505,
+                                531
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        516,
+                                        505,
+                                        531
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Andy Davis, Jeffrey Dean, Matthieu Devin, Sanjay Ghemawat, Ian Goodfellow, Andrew Harp, Geoffrey",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                114,
+                                525,
+                                506,
+                                541
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        114,
+                                        525,
+                                        506,
+                                        541
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Irving, Michael Isard, Yangqing Jia, Rafal Jozefowicz, Lukasz Kaiser, Manjunath Kudlur, Josh Levenberg,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                536,
+                                506,
+                                550
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        536,
+                                        506,
+                                        550
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Dan Mane, Rajat Monga, Sherry Moore, Derek Murray, Chris Olah, Mike Schuster, Jonathon Shlens, Benoit ´",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                547,
+                                505,
+                                560
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        547,
+                                        505,
+                                        560
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Steiner, Ilya Sutskever, Kunal Talwar, Paul Tucker, Vincent Vanhoucke, Vijay Vasudevan, Fernanda Viegas, ´",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                557,
+                                505,
+                                569
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        557,
+                                        505,
+                                        569
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Oriol Vinyals, Pete Warden, Martin Wattenberg, Martin Wicke, Yuan Yu, and Xiaoqiang Zheng. TensorFlow:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                568,
+                                505,
+                                579
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        568,
+                                        505,
+                                        579
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Large-scale machine learning on heterogeneous systems, 2015. URL http://tensorflow.org/. Soft-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                576,
+                                245,
+                                590
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        576,
+                                        245,
+                                        590
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ware available from tensorflow.org.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        }
+                    ],
+                    "index": 36.5,
+                    "bbox_fs": [
+                        106,
+                        507,
+                        506,
+                        590
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        104,
+                        598,
+                        504,
+                        619
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                597,
+                                505,
+                                610
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        597,
+                                        505,
+                                        610
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Sajid Anwar, Kyuyeon Hwang, and Wonyong Sung. Fixed point optimization of deep convolutional neural",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                608,
+                                304,
+                                620
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        608,
+                                        304,
+                                        620
+                                    ],
+                                    "score": 1.0,
+                                    "content": "networks for object recognition. In ICASSP, 2015a.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        }
+                    ],
+                    "index": 41.5,
+                    "bbox_fs": [
+                        105,
+                        597,
+                        505,
+                        620
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        628,
+                        503,
+                        649
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                628,
+                                505,
+                                641
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        628,
+                                        505,
+                                        641
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Sajid Anwar, Kyuyeon Hwang, and Wonyong Sung. Structured pruning of deep convolutional neural networks.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                639,
+                                267,
+                                650
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        639,
+                                        267,
+                                        650
+                                    ],
+                                    "score": 1.0,
+                                    "content": "arXiv preprint arXiv:1512.08571, 2015b.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        }
+                    ],
+                    "index": 43.5,
+                    "bbox_fs": [
+                        105,
+                        628,
+                        505,
+                        650
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        659,
+                        503,
+                        680
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                659,
+                                504,
+                                672
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        659,
+                                        504,
+                                        672
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Sanjeev Arora, Aditya Bhaskara, Rong Ge, and Tengyu Ma. Provable bounds for learning some deep represen-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                669,
+                                204,
+                                680
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        669,
+                                        204,
+                                        680
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tations. In ICML, 2014.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        }
+                    ],
+                    "index": 45.5,
+                    "bbox_fs": [
+                        105,
+                        659,
+                        504,
+                        680
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        690,
+                        406,
+                        701
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                689,
+                                406,
+                                703
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        689,
+                                        406,
+                                        703
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Jimmy Ba and Rich Caruana. Do deep nets really need to be deep? In NIPS, 2014.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 47
+                        }
+                    ],
+                    "index": 47,
+                    "bbox_fs": [
+                        105,
+                        689,
+                        406,
+                        703
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        711,
+                        505,
+                        731
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                709,
+                                505,
+                                724
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        709,
+                                        505,
+                                        724
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Wenlin Chen, James T. Wilson, Stephen Tyree, Kilian Q. Weinberger, and Yixin Chen. Compressing neural",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 48
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                721,
+                                293,
+                                732
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        721,
+                                        293,
+                                        732
+                                    ],
+                                    "score": 1.0,
+                                    "content": "networks with the hashing trick. In ICML, 2015.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 49
+                        }
+                    ],
+                    "index": 48.5,
+                    "bbox_fs": [
+                        105,
+                        709,
+                        505,
+                        732
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        75,
+                        507,
+                        736
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                82,
+                                505,
+                                97
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        82,
+                                        505,
+                                        97
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Dan C. Cires¸an, Ueli Meier, Jonathan Masci, Luca M. Gambardella, and Jurgen Schmidhuber. High- ¨",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                114,
+                                93,
+                                479,
+                                106
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        114,
+                                        93,
+                                        479,
+                                        106
+                                    ],
+                                    "score": 1.0,
+                                    "content": "performance neural networks for visual object classification. arXiv preprint arXiv:1102.0183, 2011.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                112,
+                                505,
+                                124
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        112,
+                                        505,
+                                        124
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Matthieu Courbariaux, Yoshua Bengio, and Jean-Pierre David. Binaryconnect: Training deep neural networks",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                121,
+                                323,
+                                135
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        121,
+                                        323,
+                                        135
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with binary weights during propagations. In NIPS, 2015.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                139,
+                                506,
+                                153
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        139,
+                                        506,
+                                        153
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Matthieu Courbariaux, Itay Hubara, Daniel Soudry, Ran El-Yaniv, and Yoshua Bengio. Binarized neural net-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                150,
+                                506,
+                                163
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        150,
+                                        415,
+                                        163
+                                    ],
+                                    "score": 1.0,
+                                    "content": "works: Training deep neural networks with weights and activations constrained to",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        415,
+                                        151,
+                                        426,
+                                        160
+                                    ],
+                                    "score": 0.8,
+                                    "content": "+ 1",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        426,
+                                        150,
+                                        506,
+                                        163
+                                    ],
+                                    "score": 1.0,
+                                    "content": "or -1. arXiv preprint",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                160,
+                                209,
+                                172
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        160,
+                                        209,
+                                        172
+                                    ],
+                                    "score": 1.0,
+                                    "content": "arXiv:1602.02830, 2016.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                180,
+                                494,
+                                191
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        180,
+                                        494,
+                                        191
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Bert De Brabandere, Xu Jia, Tinne Tuytelaars, and Luc Van Gool. Dynamic filter networks. In NIPS, 2016.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                198,
+                                506,
+                                209
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        198,
+                                        506,
+                                        209
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Jia Deng, Wei Dong, Richard Socher, Li-Jia Li, Kai Li, and Li Fei-Fei. Imagenet: A large-scale hierarchical",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                114,
+                                207,
+                                237,
+                                220
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        114,
+                                        207,
+                                        237,
+                                        220
+                                    ],
+                                    "score": 1.0,
+                                    "content": "image database. In CVPR, 2009.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                227,
+                                506,
+                                238
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        227,
+                                        506,
+                                        238
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Misha Denil, Babak Shakibi, Laurent Dinh, Marc Aurelio Ranzato, and Nando de Freitas. Predicting parameters",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                113,
+                                235,
+                                237,
+                                248
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        113,
+                                        235,
+                                        237,
+                                        248
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in deep learning. In NIPS, 2013.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                255,
+                                506,
+                                267
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        255,
+                                        506,
+                                        267
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Emily L. Denton, Wojciech Zaremba, Joan Bruna, Yann LeCun, and Rob Fergus. Exploiting linear structure",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                265,
+                                370,
+                                277
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        265,
+                                        370,
+                                        277
+                                    ],
+                                    "score": 1.0,
+                                    "content": "within convolutional networks for efficient evaluation. In NIPS, 2014.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                283,
+                                478,
+                                295
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        283,
+                                        478,
+                                        295
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Jiashi Feng and Trevor Darrell. Learning the structure of deep convolutional networks. In ICCV, 2015.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                299,
+                                507,
+                                317
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        299,
+                                        507,
+                                        317
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Yunchao Gong, Liu Liu, Ming Yang, and Lubomir Bourdev. Compressing deep convolutional networks using",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                311,
+                                333,
+                                324
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        311,
+                                        333,
+                                        324
+                                    ],
+                                    "score": 1.0,
+                                    "content": "vector quantization. arXiv preprint arXiv:1412.6115, 2014.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                330,
+                                506,
+                                344
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        330,
+                                        506,
+                                        344
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Thomas L. Griffiths and Zoubin Ghahramani. The indian buffet process: An introduction and review. Journal",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                341,
+                                332,
+                                352
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        341,
+                                        332,
+                                        352
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of Machine Learning Research, 12(Apr):1185–1224, 2011.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                359,
+                                506,
+                                370
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        359,
+                                        506,
+                                        370
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Song Han, Jeff Pool, John Tran, and William Dally. Learning both weights and connections for efficient neural",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                369,
+                                207,
+                                380
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        369,
+                                        207,
+                                        380
+                                    ],
+                                    "score": 1.0,
+                                    "content": "network. In NIPS, 2015.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                387,
+                                506,
+                                401
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        387,
+                                        506,
+                                        401
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Song Han, Huizi Mao, and William J. Dally. Deep compression: Compressing deep neural network with",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                113,
+                                397,
+                                361,
+                                410
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        113,
+                                        397,
+                                        361,
+                                        410
+                                    ],
+                                    "score": 1.0,
+                                    "content": "pruning, trained quantization and huffman coding. In ICLR, 2016a.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                414,
+                                506,
+                                429
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        414,
+                                        506,
+                                        429
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Song Han, Jeff Pool, Sharan Narang, Huizi Mao, Shijian Tang, Erich Elsen, Bryan Catanzaro, John Tran, and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                425,
+                                506,
+                                439
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        425,
+                                        506,
+                                        439
+                                    ],
+                                    "score": 1.0,
+                                    "content": "William J. Dally. Dsd: Regularizing deep neural networks with dense-sparse-dense training flow. arXiv",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                113,
+                                435,
+                                246,
+                                448
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        113,
+                                        435,
+                                        246,
+                                        448
+                                    ],
+                                    "score": 1.0,
+                                    "content": "preprint arXiv:1607.04381, 2016b.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                455,
+                                505,
+                                466
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        455,
+                                        505,
+                                        466
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Stephen Jose Hanson and Lorien Y. Pratt. Comparing biases for minimal network construction with back- ´",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                114,
+                                465,
+                                221,
+                                477
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        114,
+                                        465,
+                                        221,
+                                        477
+                                    ],
+                                    "score": 1.0,
+                                    "content": "propagation. In NIPS, 1989.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                482,
+                                506,
+                                496
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        482,
+                                        506,
+                                        496
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Babak Hassibi and David G. Stork. Second order derivatives for network pruning: Optimal brain surgeon. In",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                114,
+                                492,
+                                162,
+                                504
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        114,
+                                        492,
+                                        162,
+                                        504
+                                    ],
+                                    "score": 1.0,
+                                    "content": "NIPS, 1993.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                511,
+                                506,
+                                525
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        511,
+                                        506,
+                                        525
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image recognition. In",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                522,
+                                171,
+                                534
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        522,
+                                        171,
+                                        534
+                                    ],
+                                    "score": 1.0,
+                                    "content": "CVPR, 2016a.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                540,
+                                506,
+                                553
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        540,
+                                        506,
+                                        553
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Identity mappings in deep residual networks. In",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                550,
+                                172,
+                                562
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        550,
+                                        172,
+                                        562
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ECCV, 2016b.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                567,
+                                506,
+                                582
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        567,
+                                        506,
+                                        582
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Geoffrey E. Hinton, Nitish Srivastava, Alex Krizhevsky, Ilya Sutskever, and Ruslan Salakhutdinov. Improving",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                114,
+                                578,
+                                496,
+                                592
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        114,
+                                        578,
+                                        496,
+                                        592
+                                    ],
+                                    "score": 1.0,
+                                    "content": "neural networks by preventing co-adaptation of feature detectors. arXiv preprint arXiv:1207.0580, 2012.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                597,
+                                506,
+                                609
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        597,
+                                        506,
+                                        609
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Andrew G. Howard. Mobilenets: Efficient convolutional neural networks for mobile vision applications. In",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                114,
+                                608,
+                                188,
+                                619
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        114,
+                                        608,
+                                        188,
+                                        619
+                                    ],
+                                    "score": 1.0,
+                                    "content": "forthcoming, 2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                624,
+                                506,
+                                640
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        624,
+                                        506,
+                                        640
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Hengyuan Hu, Rui Peng, Yu-Wing Tai, and Chi-Keung Tang. Network trimming: A data-driven neuron pruning",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                114,
+                                635,
+                                430,
+                                649
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        114,
+                                        635,
+                                        430,
+                                        649
+                                    ],
+                                    "score": 1.0,
+                                    "content": "approach towards efficient deep architectures. arXiv preprint arXiv:1607.03250, 2016.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                654,
+                                506,
+                                667
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        654,
+                                        506,
+                                        667
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Gao Huang, Zhuang Liu, and Kilian Q. Weinberger. Densely connected convolutional networks. arXiv preprint",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                114,
+                                663,
+                                215,
+                                677
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        114,
+                                        663,
+                                        215,
+                                        677
+                                    ],
+                                    "score": 1.0,
+                                    "content": "arXiv:1608.06993, 2016a.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                681,
+                                506,
+                                696
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        681,
+                                        506,
+                                        696
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Gao Huang, Yu Sun, Zhuang Liu, Daniel Sedra, and Kilian Weinberger. Deep networks with stochastic depth.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                693,
+                                268,
+                                707
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        693,
+                                        268,
+                                        707
+                                    ],
+                                    "score": 1.0,
+                                    "content": "arXiv preprint arXiv:1603.09382, 2016b.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                711,
+                                506,
+                                725
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        711,
+                                        490,
+                                        725
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Kyuyeon Hwang and Wonyong Sung. Fixed-point feedforward deep neural network design using weights",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        491,
+                                        712,
+                                        501,
+                                        721
+                                    ],
+                                    "score": 0.81,
+                                    "content": "+ 1",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        502,
+                                        711,
+                                        506,
+                                        725
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                722,
+                                381,
+                                734
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        722,
+                                        381,
+                                        734
+                                    ],
+                                    "score": 1.0,
+                                    "content": "0, and -1. In IEEE Workshop on Signal Processing Systems (SiPS), 2014.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        }
+                    ],
+                    "index": 22.5
+                }
+            ],
+            "page_idx": 9,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        106,
+                        27,
+                        308,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                26,
+                                308,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        26,
+                                        308,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Under review as a conference paper at ICLR 2017",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        300,
+                        751,
+                        311,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                299,
+                                750,
+                                313,
+                                764
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        299,
+                                        750,
+                                        313,
+                                        764
+                                    ],
+                                    "score": 1.0,
+                                    "content": "10",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "list",
+                    "bbox": [
+                        105,
+                        75,
+                        507,
+                        736
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                82,
+                                505,
+                                97
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        82,
+                                        505,
+                                        97
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Dan C. Cires¸an, Ueli Meier, Jonathan Masci, Luca M. Gambardella, and Jurgen Schmidhuber. High- ¨",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                114,
+                                93,
+                                479,
+                                106
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        114,
+                                        93,
+                                        479,
+                                        106
+                                    ],
+                                    "score": 1.0,
+                                    "content": "performance neural networks for visual object classification. arXiv preprint arXiv:1102.0183, 2011.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                112,
+                                505,
+                                124
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        112,
+                                        505,
+                                        124
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Matthieu Courbariaux, Yoshua Bengio, and Jean-Pierre David. Binaryconnect: Training deep neural networks",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                121,
+                                323,
+                                135
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        121,
+                                        323,
+                                        135
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with binary weights during propagations. In NIPS, 2015.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                139,
+                                506,
+                                153
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        139,
+                                        506,
+                                        153
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Matthieu Courbariaux, Itay Hubara, Daniel Soudry, Ran El-Yaniv, and Yoshua Bengio. Binarized neural net-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                150,
+                                506,
+                                163
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        150,
+                                        415,
+                                        163
+                                    ],
+                                    "score": 1.0,
+                                    "content": "works: Training deep neural networks with weights and activations constrained to",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        415,
+                                        151,
+                                        426,
+                                        160
+                                    ],
+                                    "score": 0.8,
+                                    "content": "+ 1",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        426,
+                                        150,
+                                        506,
+                                        163
+                                    ],
+                                    "score": 1.0,
+                                    "content": "or -1. arXiv preprint",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                160,
+                                209,
+                                172
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        160,
+                                        209,
+                                        172
+                                    ],
+                                    "score": 1.0,
+                                    "content": "arXiv:1602.02830, 2016.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                180,
+                                494,
+                                191
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        180,
+                                        494,
+                                        191
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Bert De Brabandere, Xu Jia, Tinne Tuytelaars, and Luc Van Gool. Dynamic filter networks. In NIPS, 2016.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                198,
+                                506,
+                                209
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        198,
+                                        506,
+                                        209
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Jia Deng, Wei Dong, Richard Socher, Li-Jia Li, Kai Li, and Li Fei-Fei. Imagenet: A large-scale hierarchical",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                114,
+                                207,
+                                237,
+                                220
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        114,
+                                        207,
+                                        237,
+                                        220
+                                    ],
+                                    "score": 1.0,
+                                    "content": "image database. In CVPR, 2009.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                227,
+                                506,
+                                238
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        227,
+                                        506,
+                                        238
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Misha Denil, Babak Shakibi, Laurent Dinh, Marc Aurelio Ranzato, and Nando de Freitas. Predicting parameters",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                113,
+                                235,
+                                237,
+                                248
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        113,
+                                        235,
+                                        237,
+                                        248
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in deep learning. In NIPS, 2013.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                255,
+                                506,
+                                267
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        255,
+                                        506,
+                                        267
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Emily L. Denton, Wojciech Zaremba, Joan Bruna, Yann LeCun, and Rob Fergus. Exploiting linear structure",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                265,
+                                370,
+                                277
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        265,
+                                        370,
+                                        277
+                                    ],
+                                    "score": 1.0,
+                                    "content": "within convolutional networks for efficient evaluation. In NIPS, 2014.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                283,
+                                478,
+                                295
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        283,
+                                        478,
+                                        295
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Jiashi Feng and Trevor Darrell. Learning the structure of deep convolutional networks. In ICCV, 2015.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14,
+                            "is_list_start_line": true,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                299,
+                                507,
+                                317
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        299,
+                                        507,
+                                        317
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Yunchao Gong, Liu Liu, Ming Yang, and Lubomir Bourdev. Compressing deep convolutional networks using",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                311,
+                                333,
+                                324
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        311,
+                                        333,
+                                        324
+                                    ],
+                                    "score": 1.0,
+                                    "content": "vector quantization. arXiv preprint arXiv:1412.6115, 2014.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                330,
+                                506,
+                                344
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        330,
+                                        506,
+                                        344
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Thomas L. Griffiths and Zoubin Ghahramani. The indian buffet process: An introduction and review. Journal",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                341,
+                                332,
+                                352
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        341,
+                                        332,
+                                        352
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of Machine Learning Research, 12(Apr):1185–1224, 2011.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                359,
+                                506,
+                                370
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        359,
+                                        506,
+                                        370
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Song Han, Jeff Pool, John Tran, and William Dally. Learning both weights and connections for efficient neural",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                369,
+                                207,
+                                380
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        369,
+                                        207,
+                                        380
+                                    ],
+                                    "score": 1.0,
+                                    "content": "network. In NIPS, 2015.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                387,
+                                506,
+                                401
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        387,
+                                        506,
+                                        401
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Song Han, Huizi Mao, and William J. Dally. Deep compression: Compressing deep neural network with",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                113,
+                                397,
+                                361,
+                                410
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        113,
+                                        397,
+                                        361,
+                                        410
+                                    ],
+                                    "score": 1.0,
+                                    "content": "pruning, trained quantization and huffman coding. In ICLR, 2016a.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                414,
+                                506,
+                                429
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        414,
+                                        506,
+                                        429
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Song Han, Jeff Pool, Sharan Narang, Huizi Mao, Shijian Tang, Erich Elsen, Bryan Catanzaro, John Tran, and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                425,
+                                506,
+                                439
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        425,
+                                        506,
+                                        439
+                                    ],
+                                    "score": 1.0,
+                                    "content": "William J. Dally. Dsd: Regularizing deep neural networks with dense-sparse-dense training flow. arXiv",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                113,
+                                435,
+                                246,
+                                448
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        113,
+                                        435,
+                                        246,
+                                        448
+                                    ],
+                                    "score": 1.0,
+                                    "content": "preprint arXiv:1607.04381, 2016b.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                455,
+                                505,
+                                466
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        455,
+                                        505,
+                                        466
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Stephen Jose Hanson and Lorien Y. Pratt. Comparing biases for minimal network construction with back- ´",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                114,
+                                465,
+                                221,
+                                477
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        114,
+                                        465,
+                                        221,
+                                        477
+                                    ],
+                                    "score": 1.0,
+                                    "content": "propagation. In NIPS, 1989.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                482,
+                                506,
+                                496
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        482,
+                                        506,
+                                        496
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Babak Hassibi and David G. Stork. Second order derivatives for network pruning: Optimal brain surgeon. In",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                114,
+                                492,
+                                162,
+                                504
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        114,
+                                        492,
+                                        162,
+                                        504
+                                    ],
+                                    "score": 1.0,
+                                    "content": "NIPS, 1993.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                511,
+                                506,
+                                525
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        511,
+                                        506,
+                                        525
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image recognition. In",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                522,
+                                171,
+                                534
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        522,
+                                        171,
+                                        534
+                                    ],
+                                    "score": 1.0,
+                                    "content": "CVPR, 2016a.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                540,
+                                506,
+                                553
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        540,
+                                        506,
+                                        553
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Identity mappings in deep residual networks. In",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                550,
+                                172,
+                                562
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        550,
+                                        172,
+                                        562
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ECCV, 2016b.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                567,
+                                506,
+                                582
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        567,
+                                        506,
+                                        582
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Geoffrey E. Hinton, Nitish Srivastava, Alex Krizhevsky, Ilya Sutskever, and Ruslan Salakhutdinov. Improving",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                114,
+                                578,
+                                496,
+                                592
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        114,
+                                        578,
+                                        496,
+                                        592
+                                    ],
+                                    "score": 1.0,
+                                    "content": "neural networks by preventing co-adaptation of feature detectors. arXiv preprint arXiv:1207.0580, 2012.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                597,
+                                506,
+                                609
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        597,
+                                        506,
+                                        609
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Andrew G. Howard. Mobilenets: Efficient convolutional neural networks for mobile vision applications. In",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                114,
+                                608,
+                                188,
+                                619
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        114,
+                                        608,
+                                        188,
+                                        619
+                                    ],
+                                    "score": 1.0,
+                                    "content": "forthcoming, 2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                624,
+                                506,
+                                640
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        624,
+                                        506,
+                                        640
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Hengyuan Hu, Rui Peng, Yu-Wing Tai, and Chi-Keung Tang. Network trimming: A data-driven neuron pruning",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                114,
+                                635,
+                                430,
+                                649
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        114,
+                                        635,
+                                        430,
+                                        649
+                                    ],
+                                    "score": 1.0,
+                                    "content": "approach towards efficient deep architectures. arXiv preprint arXiv:1607.03250, 2016.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                654,
+                                506,
+                                667
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        654,
+                                        506,
+                                        667
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Gao Huang, Zhuang Liu, and Kilian Q. Weinberger. Densely connected convolutional networks. arXiv preprint",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                114,
+                                663,
+                                215,
+                                677
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        114,
+                                        663,
+                                        215,
+                                        677
+                                    ],
+                                    "score": 1.0,
+                                    "content": "arXiv:1608.06993, 2016a.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                681,
+                                506,
+                                696
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        681,
+                                        506,
+                                        696
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Gao Huang, Yu Sun, Zhuang Liu, Daniel Sedra, and Kilian Weinberger. Deep networks with stochastic depth.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                693,
+                                268,
+                                707
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        693,
+                                        268,
+                                        707
+                                    ],
+                                    "score": 1.0,
+                                    "content": "arXiv preprint arXiv:1603.09382, 2016b.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                711,
+                                506,
+                                725
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        711,
+                                        490,
+                                        725
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Kyuyeon Hwang and Wonyong Sung. Fixed-point feedforward deep neural network design using weights",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        491,
+                                        712,
+                                        501,
+                                        721
+                                    ],
+                                    "score": 0.81,
+                                    "content": "+ 1",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        502,
+                                        711,
+                                        506,
+                                        725
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                722,
+                                381,
+                                734
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        722,
+                                        381,
+                                        734
+                                    ],
+                                    "score": 1.0,
+                                    "content": "0, and -1. In IEEE Workshop on Signal Processing Systems (SiPS), 2014.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45,
+                            "is_list_end_line": true
+                        }
+                    ],
+                    "index": 22.5,
+                    "bbox_fs": [
+                        104,
+                        82,
+                        507,
+                        734
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        82,
+                        503,
+                        104
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                82,
+                                504,
+                                95
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        82,
+                                        504,
+                                        95
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Yani Ioannou, Duncan Robertson, Roberto Cipolla, and Antonio Criminisi. Deep roots: Improving cnn effi-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                93,
+                                405,
+                                105
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        93,
+                                        405,
+                                        105
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ciency with hierarchical filter groups. arXiv preprint arXiv:1605.06489, 2016a.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        }
+                    ],
+                    "index": 0.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        111,
+                        503,
+                        132
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                107,
+                                111,
+                                504,
+                                122
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        111,
+                                        504,
+                                        122
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Yani Ioannou, Duncan Robertson, Jamie Shotton, Roberto Cipolla, and Antonio Criminisi. Training cnns with",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                121,
+                                354,
+                                132
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        121,
+                                        354,
+                                        132
+                                    ],
+                                    "score": 1.0,
+                                    "content": "low-rank filters for efficient image classification. In ICLR, 2016b.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        }
+                    ],
+                    "index": 2.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        138,
+                        503,
+                        160
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                139,
+                                504,
+                                151
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        139,
+                                        504,
+                                        151
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Max Jaderberg, Andrea Vedaldi, and Andrew Zisserman. Speeding up convolutional neural networks with low",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                149,
+                                241,
+                                160
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        149,
+                                        241,
+                                        160
+                                    ],
+                                    "score": 1.0,
+                                    "content": "rank expansions. In BMVC, 2014.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        }
+                    ],
+                    "index": 4.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        167,
+                        504,
+                        189
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                168,
+                                504,
+                                179
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        168,
+                                        504,
+                                        179
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Xiaojie Jin, Xiaotong Yuan, Jiashi Feng, and Shuicheng Yan. Training skinny deep neural networks with",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                177,
+                                395,
+                                189
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        177,
+                                        395,
+                                        189
+                                    ],
+                                    "score": 1.0,
+                                    "content": "iterative hard thresholding methods. arXiv preprint arXiv:1607.05423, 2016.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        }
+                    ],
+                    "index": 6.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        195,
+                        504,
+                        217
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                194,
+                                506,
+                                208
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        194,
+                                        506,
+                                        208
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Yong-Deok Kim, Eunhyeok Park, Sungjoo Yoo, Taelim Choi, Lu Yang, and Dongjun Shin. Compression of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                205,
+                                462,
+                                217
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        205,
+                                        462,
+                                        217
+                                    ],
+                                    "score": 1.0,
+                                    "content": "deep convolutional neural networks for fast and low power mobile applications. In ICLR, 2016.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        }
+                    ],
+                    "index": 8.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        223,
+                        504,
+                        244
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                223,
+                                505,
+                                235
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        223,
+                                        505,
+                                        235
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Alex Krizhevsky, Ilya Sutskever, and Geoffrey E. Hinton. Imagenet classification with deep convolutional",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                234,
+                                235,
+                                245
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        234,
+                                        235,
+                                        245
+                                    ],
+                                    "score": 1.0,
+                                    "content": "neural networks. In NIPS, 2012.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        }
+                    ],
+                    "index": 10.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        251,
+                        503,
+                        273
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                250,
+                                505,
+                                264
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        250,
+                                        505,
+                                        264
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Vadim Lebedev, Yaroslav Ganin, Maksim Rakhuba, Ivan Oseledets, and Victor Lempitsky. Speeding-up con-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                117,
+                                262,
+                                401,
+                                273
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        117,
+                                        262,
+                                        401,
+                                        273
+                                    ],
+                                    "score": 1.0,
+                                    "content": "volutional neural networks using fine-tuned cp-decomposition. In ICLR, 2015.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        }
+                    ],
+                    "index": 12.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        279,
+                        504,
+                        300
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                278,
+                                505,
+                                291
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        278,
+                                        505,
+                                        291
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Yann LeCun, John S. Denker, Sara A. Solla, Richard E. Howard, and Lawrence D. Jackel. Optimal brain",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                290,
+                                205,
+                                300
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        290,
+                                        205,
+                                        300
+                                    ],
+                                    "score": 1.0,
+                                    "content": "damage. In NIPS, 1989.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        }
+                    ],
+                    "index": 14.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        307,
+                        504,
+                        329
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                306,
+                                505,
+                                320
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        306,
+                                        505,
+                                        320
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Yann LeCun, Leon Bottou, Yoshua Bengio, and Patrick Haffner. Gradient-based learning applied to document ´",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                318,
+                                350,
+                                329
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        318,
+                                        350,
+                                        329
+                                    ],
+                                    "score": 1.0,
+                                    "content": "recognition. Proceedings of the IEEE, 86(11):2278–2324, 1998.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        }
+                    ],
+                    "index": 16.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        335,
+                        504,
+                        357
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                335,
+                                505,
+                                348
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        335,
+                                        505,
+                                        348
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Hao Li, Asim Kadav, Igor Durdanovic, Hanan Samet, and Hans Peter Graf. Pruning filters for efficient convnets.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                346,
+                                262,
+                                357
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        346,
+                                        262,
+                                        357
+                                    ],
+                                    "score": 1.0,
+                                    "content": "arXiv preprint arXiv:1608.08710, 2016.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        }
+                    ],
+                    "index": 18.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        363,
+                        503,
+                        385
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                364,
+                                505,
+                                375
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        364,
+                                        505,
+                                        375
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Baoyuan Liu, Min Wang, Hassan Foroosh, Marshall Tappen, and Marianna Pensky. Sparse convolutional neural",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                374,
+                                214,
+                                384
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        374,
+                                        214,
+                                        384
+                                    ],
+                                    "score": 1.0,
+                                    "content": "networks. In CVPR, 2015.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        }
+                    ],
+                    "index": 20.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        392,
+                        503,
+                        413
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                390,
+                                505,
+                                406
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        390,
+                                        505,
+                                        406
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Julien Mairal, Francis Bach, Jean Ponce, and Guillermo Sapiro. Online dictionary learning for sparse coding.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                402,
+                                175,
+                                413
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        402,
+                                        175,
+                                        413
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In ICML, 2009.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        }
+                    ],
+                    "index": 22.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        420,
+                        504,
+                        442
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                420,
+                                505,
+                                433
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        420,
+                                        505,
+                                        433
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Kenton Murray and David Chiang. Auto-sizing neural networks: With applications to n-gram language models.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                430,
+                                183,
+                                441
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        430,
+                                        183,
+                                        441
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In EMNLP, 2015.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        }
+                    ],
+                    "index": 24.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        448,
+                        503,
+                        469
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                448,
+                                505,
+                                460
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        448,
+                                        505,
+                                        460
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Alexander Novikov, Dmitrii Podoprikhin, Anton Osokin, and Dmitry P. Vetrov. Tensorizing neural networks.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                458,
+                                172,
+                                469
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        458,
+                                        172,
+                                        469
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In NIPS, 2015.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        }
+                    ],
+                    "index": 26.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        476,
+                        503,
+                        497
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                475,
+                                505,
+                                488
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        475,
+                                        505,
+                                        488
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Ivan V. Oseledets. Tensor-train decomposition. SIAM Journal on Scientific Computing, 33(5):2295–2317,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                485,
+                                141,
+                                498
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        485,
+                                        141,
+                                        498
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2011.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        }
+                    ],
+                    "index": 28.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        504,
+                        504,
+                        525
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                504,
+                                505,
+                                516
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        504,
+                                        505,
+                                        516
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Mohammad Rastegari, Vicente Ordonez, Joseph Redmon, and Ali Farhadi. Xnor-net: Imagenet classification",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                514,
+                                337,
+                                525
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        514,
+                                        337,
+                                        525
+                                    ],
+                                    "score": 1.0,
+                                    "content": "using binary convolutional neural networks. In ECCV, 2016.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        }
+                    ],
+                    "index": 30.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        532,
+                        503,
+                        564
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                532,
+                                504,
+                                545
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        532,
+                                        504,
+                                        545
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Olga Russakovsky, Jia Deng, Hao Su, Jonathan Krause, Sanjeev Satheesh, Sean Ma, Zhiheng Huang, An-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                541,
+                                505,
+                                555
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        541,
+                                        505,
+                                        555
+                                    ],
+                                    "score": 1.0,
+                                    "content": "drej Karpathy, Aditya Khosla, Michael Bernstein, et al. Imagenet large scale visual recognition challenge.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                552,
+                                355,
+                                564
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        552,
+                                        355,
+                                        564
+                                    ],
+                                    "score": 1.0,
+                                    "content": "International Journal of Computer Vision, 115(3):211–252, 2015.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        }
+                    ],
+                    "index": 33
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        570,
+                        504,
+                        591
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                569,
+                                504,
+                                583
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        569,
+                                        504,
+                                        583
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Karen Simonyan and Andrew Zisserman. Very deep convolutional networks for large-scale image recognition.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                580,
+                                173,
+                                591
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        580,
+                                        173,
+                                        591
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In ICLR, 2015.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        }
+                    ],
+                    "index": 35.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        598,
+                        503,
+                        630
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                598,
+                                505,
+                                610
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        598,
+                                        505,
+                                        610
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Nitish Srivastava, Geoffrey E. Hinton, Alex Krizhevsky, Ilya Sutskever, and Ruslan Salakhutdinov. Dropout:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                609,
+                                505,
+                                620
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        609,
+                                        505,
+                                        620
+                                    ],
+                                    "score": 1.0,
+                                    "content": "a simple way to prevent neural networks from overfitting. Journal of Machine Learning Research, 15(1):",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                619,
+                                184,
+                                630
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        619,
+                                        184,
+                                        630
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1929–1958, 2014.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        }
+                    ],
+                    "index": 38
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        636,
+                        492,
+                        648
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                635,
+                                492,
+                                649
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        635,
+                                        492,
+                                        649
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Rupesh K. Srivastava, Klaus Greff, and Jurgen Schmidhuber. Training very deep networks. In ¨ NIPS, 2015.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        }
+                    ],
+                    "index": 40
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        655,
+                        504,
+                        676
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                655,
+                                505,
+                                667
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        655,
+                                        505,
+                                        667
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Christian Szegedy, Vincent Vanhoucke, Sergey Ioffe, Jonathon Shlens, and Zbigniew Wojna. Rethinking the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                665,
+                                332,
+                                676
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        665,
+                                        332,
+                                        676
+                                    ],
+                                    "score": 1.0,
+                                    "content": "inception architecture for computer vision. In CVPR, 2016.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        }
+                    ],
+                    "index": 41.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        683,
+                        504,
+                        704
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                682,
+                                505,
+                                696
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        682,
+                                        505,
+                                        696
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Li Wan, Matthew Zeiler, Sixin Zhang, Yann LeCun, and Rob Fergus. Regularization of neural networks using",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                693,
+                                224,
+                                704
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        693,
+                                        224,
+                                        704
+                                    ],
+                                    "score": 1.0,
+                                    "content": "dropconnect. In ICML, 2013.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        }
+                    ],
+                    "index": 43.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        711,
+                        505,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                711,
+                                505,
+                                723
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        711,
+                                        505,
+                                        723
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Wei Wen, Chunpeng Wu, Yandan Wang, Yiran Chen, and Hai Li. Learning structured sparsity in deep neural",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                721,
+                                211,
+                                731
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        721,
+                                        211,
+                                        731
+                                    ],
+                                    "score": 1.0,
+                                    "content": "networks. In NIPS, 2016.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        }
+                    ],
+                    "index": 45.5
+                }
+            ],
+            "page_idx": 10,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        300,
+                        751,
+                        310,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                299,
+                                750,
+                                312,
+                                765
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        299,
+                                        750,
+                                        312,
+                                        765
+                                    ],
+                                    "score": 1.0,
+                                    "content": "",
+                                    "type": "text",
+                                    "height": 15,
+                                    "width": 13
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        106,
+                        27,
+                        308,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                26,
+                                308,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        26,
+                                        308,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Under review as a conference paper at ICLR 2017",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        82,
+                        503,
+                        104
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                82,
+                                504,
+                                95
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        82,
+                                        504,
+                                        95
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Yani Ioannou, Duncan Robertson, Roberto Cipolla, and Antonio Criminisi. Deep roots: Improving cnn effi-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                93,
+                                405,
+                                105
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        93,
+                                        405,
+                                        105
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ciency with hierarchical filter groups. arXiv preprint arXiv:1605.06489, 2016a.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        }
+                    ],
+                    "index": 0.5,
+                    "bbox_fs": [
+                        106,
+                        82,
+                        504,
+                        105
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        111,
+                        503,
+                        132
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                107,
+                                111,
+                                504,
+                                122
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        111,
+                                        504,
+                                        122
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Yani Ioannou, Duncan Robertson, Jamie Shotton, Roberto Cipolla, and Antonio Criminisi. Training cnns with",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                121,
+                                354,
+                                132
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        121,
+                                        354,
+                                        132
+                                    ],
+                                    "score": 1.0,
+                                    "content": "low-rank filters for efficient image classification. In ICLR, 2016b.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        }
+                    ],
+                    "index": 2.5,
+                    "bbox_fs": [
+                        107,
+                        111,
+                        504,
+                        132
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        138,
+                        503,
+                        160
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                139,
+                                504,
+                                151
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        139,
+                                        504,
+                                        151
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Max Jaderberg, Andrea Vedaldi, and Andrew Zisserman. Speeding up convolutional neural networks with low",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                149,
+                                241,
+                                160
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        149,
+                                        241,
+                                        160
+                                    ],
+                                    "score": 1.0,
+                                    "content": "rank expansions. In BMVC, 2014.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        }
+                    ],
+                    "index": 4.5,
+                    "bbox_fs": [
+                        106,
+                        139,
+                        504,
+                        160
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        167,
+                        504,
+                        189
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                168,
+                                504,
+                                179
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        168,
+                                        504,
+                                        179
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Xiaojie Jin, Xiaotong Yuan, Jiashi Feng, and Shuicheng Yan. Training skinny deep neural networks with",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                177,
+                                395,
+                                189
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        177,
+                                        395,
+                                        189
+                                    ],
+                                    "score": 1.0,
+                                    "content": "iterative hard thresholding methods. arXiv preprint arXiv:1607.05423, 2016.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        }
+                    ],
+                    "index": 6.5,
+                    "bbox_fs": [
+                        106,
+                        168,
+                        504,
+                        189
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        195,
+                        504,
+                        217
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                194,
+                                506,
+                                208
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        194,
+                                        506,
+                                        208
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Yong-Deok Kim, Eunhyeok Park, Sungjoo Yoo, Taelim Choi, Lu Yang, and Dongjun Shin. Compression of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                205,
+                                462,
+                                217
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        205,
+                                        462,
+                                        217
+                                    ],
+                                    "score": 1.0,
+                                    "content": "deep convolutional neural networks for fast and low power mobile applications. In ICLR, 2016.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        }
+                    ],
+                    "index": 8.5,
+                    "bbox_fs": [
+                        106,
+                        194,
+                        506,
+                        217
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        223,
+                        504,
+                        244
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                223,
+                                505,
+                                235
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        223,
+                                        505,
+                                        235
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Alex Krizhevsky, Ilya Sutskever, and Geoffrey E. Hinton. Imagenet classification with deep convolutional",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                234,
+                                235,
+                                245
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        234,
+                                        235,
+                                        245
+                                    ],
+                                    "score": 1.0,
+                                    "content": "neural networks. In NIPS, 2012.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        }
+                    ],
+                    "index": 10.5,
+                    "bbox_fs": [
+                        106,
+                        223,
+                        505,
+                        245
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        251,
+                        503,
+                        273
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                250,
+                                505,
+                                264
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        250,
+                                        505,
+                                        264
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Vadim Lebedev, Yaroslav Ganin, Maksim Rakhuba, Ivan Oseledets, and Victor Lempitsky. Speeding-up con-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                117,
+                                262,
+                                401,
+                                273
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        117,
+                                        262,
+                                        401,
+                                        273
+                                    ],
+                                    "score": 1.0,
+                                    "content": "volutional neural networks using fine-tuned cp-decomposition. In ICLR, 2015.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        }
+                    ],
+                    "index": 12.5,
+                    "bbox_fs": [
+                        106,
+                        250,
+                        505,
+                        273
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        279,
+                        504,
+                        300
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                278,
+                                505,
+                                291
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        278,
+                                        505,
+                                        291
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Yann LeCun, John S. Denker, Sara A. Solla, Richard E. Howard, and Lawrence D. Jackel. Optimal brain",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                290,
+                                205,
+                                300
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        290,
+                                        205,
+                                        300
+                                    ],
+                                    "score": 1.0,
+                                    "content": "damage. In NIPS, 1989.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        }
+                    ],
+                    "index": 14.5,
+                    "bbox_fs": [
+                        105,
+                        278,
+                        505,
+                        300
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        307,
+                        504,
+                        329
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                306,
+                                505,
+                                320
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        306,
+                                        505,
+                                        320
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Yann LeCun, Leon Bottou, Yoshua Bengio, and Patrick Haffner. Gradient-based learning applied to document ´",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                318,
+                                350,
+                                329
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        318,
+                                        350,
+                                        329
+                                    ],
+                                    "score": 1.0,
+                                    "content": "recognition. Proceedings of the IEEE, 86(11):2278–2324, 1998.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        }
+                    ],
+                    "index": 16.5,
+                    "bbox_fs": [
+                        106,
+                        306,
+                        505,
+                        329
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        335,
+                        504,
+                        357
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                335,
+                                505,
+                                348
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        335,
+                                        505,
+                                        348
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Hao Li, Asim Kadav, Igor Durdanovic, Hanan Samet, and Hans Peter Graf. Pruning filters for efficient convnets.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                346,
+                                262,
+                                357
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        346,
+                                        262,
+                                        357
+                                    ],
+                                    "score": 1.0,
+                                    "content": "arXiv preprint arXiv:1608.08710, 2016.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        }
+                    ],
+                    "index": 18.5,
+                    "bbox_fs": [
+                        105,
+                        335,
+                        505,
+                        357
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        363,
+                        503,
+                        385
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                364,
+                                505,
+                                375
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        364,
+                                        505,
+                                        375
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Baoyuan Liu, Min Wang, Hassan Foroosh, Marshall Tappen, and Marianna Pensky. Sparse convolutional neural",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                374,
+                                214,
+                                384
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        374,
+                                        214,
+                                        384
+                                    ],
+                                    "score": 1.0,
+                                    "content": "networks. In CVPR, 2015.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        }
+                    ],
+                    "index": 20.5,
+                    "bbox_fs": [
+                        106,
+                        364,
+                        505,
+                        384
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        392,
+                        503,
+                        413
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                390,
+                                505,
+                                406
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        390,
+                                        505,
+                                        406
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Julien Mairal, Francis Bach, Jean Ponce, and Guillermo Sapiro. Online dictionary learning for sparse coding.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                402,
+                                175,
+                                413
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        402,
+                                        175,
+                                        413
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In ICML, 2009.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        }
+                    ],
+                    "index": 22.5,
+                    "bbox_fs": [
+                        104,
+                        390,
+                        505,
+                        413
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        420,
+                        504,
+                        442
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                420,
+                                505,
+                                433
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        420,
+                                        505,
+                                        433
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Kenton Murray and David Chiang. Auto-sizing neural networks: With applications to n-gram language models.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                430,
+                                183,
+                                441
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        430,
+                                        183,
+                                        441
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In EMNLP, 2015.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        }
+                    ],
+                    "index": 24.5,
+                    "bbox_fs": [
+                        106,
+                        420,
+                        505,
+                        441
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        448,
+                        503,
+                        469
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                448,
+                                505,
+                                460
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        448,
+                                        505,
+                                        460
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Alexander Novikov, Dmitrii Podoprikhin, Anton Osokin, and Dmitry P. Vetrov. Tensorizing neural networks.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                458,
+                                172,
+                                469
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        458,
+                                        172,
+                                        469
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In NIPS, 2015.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        }
+                    ],
+                    "index": 26.5,
+                    "bbox_fs": [
+                        106,
+                        448,
+                        505,
+                        469
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        476,
+                        503,
+                        497
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                475,
+                                505,
+                                488
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        475,
+                                        505,
+                                        488
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Ivan V. Oseledets. Tensor-train decomposition. SIAM Journal on Scientific Computing, 33(5):2295–2317,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                485,
+                                141,
+                                498
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        485,
+                                        141,
+                                        498
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2011.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        }
+                    ],
+                    "index": 28.5,
+                    "bbox_fs": [
+                        105,
+                        475,
+                        505,
+                        498
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        504,
+                        504,
+                        525
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                504,
+                                505,
+                                516
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        504,
+                                        505,
+                                        516
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Mohammad Rastegari, Vicente Ordonez, Joseph Redmon, and Ali Farhadi. Xnor-net: Imagenet classification",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                514,
+                                337,
+                                525
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        514,
+                                        337,
+                                        525
+                                    ],
+                                    "score": 1.0,
+                                    "content": "using binary convolutional neural networks. In ECCV, 2016.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        }
+                    ],
+                    "index": 30.5,
+                    "bbox_fs": [
+                        105,
+                        504,
+                        505,
+                        525
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        532,
+                        503,
+                        564
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                532,
+                                504,
+                                545
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        532,
+                                        504,
+                                        545
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Olga Russakovsky, Jia Deng, Hao Su, Jonathan Krause, Sanjeev Satheesh, Sean Ma, Zhiheng Huang, An-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                541,
+                                505,
+                                555
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        541,
+                                        505,
+                                        555
+                                    ],
+                                    "score": 1.0,
+                                    "content": "drej Karpathy, Aditya Khosla, Michael Bernstein, et al. Imagenet large scale visual recognition challenge.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                552,
+                                355,
+                                564
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        552,
+                                        355,
+                                        564
+                                    ],
+                                    "score": 1.0,
+                                    "content": "International Journal of Computer Vision, 115(3):211–252, 2015.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        }
+                    ],
+                    "index": 33,
+                    "bbox_fs": [
+                        105,
+                        532,
+                        505,
+                        564
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        570,
+                        504,
+                        591
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                569,
+                                504,
+                                583
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        569,
+                                        504,
+                                        583
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Karen Simonyan and Andrew Zisserman. Very deep convolutional networks for large-scale image recognition.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                580,
+                                173,
+                                591
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        580,
+                                        173,
+                                        591
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In ICLR, 2015.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        }
+                    ],
+                    "index": 35.5,
+                    "bbox_fs": [
+                        105,
+                        569,
+                        504,
+                        591
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        598,
+                        503,
+                        630
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                598,
+                                505,
+                                610
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        598,
+                                        505,
+                                        610
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Nitish Srivastava, Geoffrey E. Hinton, Alex Krizhevsky, Ilya Sutskever, and Ruslan Salakhutdinov. Dropout:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                609,
+                                505,
+                                620
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        609,
+                                        505,
+                                        620
+                                    ],
+                                    "score": 1.0,
+                                    "content": "a simple way to prevent neural networks from overfitting. Journal of Machine Learning Research, 15(1):",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                619,
+                                184,
+                                630
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        619,
+                                        184,
+                                        630
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1929–1958, 2014.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        }
+                    ],
+                    "index": 38,
+                    "bbox_fs": [
+                        105,
+                        598,
+                        505,
+                        630
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        636,
+                        492,
+                        648
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                635,
+                                492,
+                                649
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        635,
+                                        492,
+                                        649
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Rupesh K. Srivastava, Klaus Greff, and Jurgen Schmidhuber. Training very deep networks. In ¨ NIPS, 2015.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        }
+                    ],
+                    "index": 40,
+                    "bbox_fs": [
+                        105,
+                        635,
+                        492,
+                        649
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        655,
+                        504,
+                        676
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                655,
+                                505,
+                                667
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        655,
+                                        505,
+                                        667
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Christian Szegedy, Vincent Vanhoucke, Sergey Ioffe, Jonathon Shlens, and Zbigniew Wojna. Rethinking the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                665,
+                                332,
+                                676
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        665,
+                                        332,
+                                        676
+                                    ],
+                                    "score": 1.0,
+                                    "content": "inception architecture for computer vision. In CVPR, 2016.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        }
+                    ],
+                    "index": 41.5,
+                    "bbox_fs": [
+                        106,
+                        655,
+                        505,
+                        676
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        683,
+                        504,
+                        704
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                682,
+                                505,
+                                696
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        682,
+                                        505,
+                                        696
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Li Wan, Matthew Zeiler, Sixin Zhang, Yann LeCun, and Rob Fergus. Regularization of neural networks using",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                693,
+                                224,
+                                704
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        693,
+                                        224,
+                                        704
+                                    ],
+                                    "score": 1.0,
+                                    "content": "dropconnect. In ICML, 2013.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        }
+                    ],
+                    "index": 43.5,
+                    "bbox_fs": [
+                        105,
+                        682,
+                        505,
+                        704
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        711,
+                        505,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                711,
+                                505,
+                                723
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        711,
+                                        505,
+                                        723
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Wei Wen, Chunpeng Wu, Yandan Wang, Yiran Chen, and Hai Li. Learning structured sparsity in deep neural",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                721,
+                                211,
+                                731
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        721,
+                                        211,
+                                        731
+                                    ],
+                                    "score": 1.0,
+                                    "content": "networks. In NIPS, 2016.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        }
+                    ],
+                    "index": 45.5,
+                    "bbox_fs": [
+                        106,
+                        711,
+                        505,
+                        731
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        83,
+                        504,
+                        104
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                83,
+                                505,
+                                95
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        83,
+                                        505,
+                                        95
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Xiangyu Zhang, Jianhua Zou, Kaiming He, and Jian Sun. Accelerating very deep convolutional networks for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                93,
+                                483,
+                                105
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        93,
+                                        483,
+                                        105
+                                    ],
+                                    "score": 1.0,
+                                    "content": "classification and detection. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2015a.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        }
+                    ],
+                    "index": 0.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        111,
+                        504,
+                        131
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                111,
+                                505,
+                                123
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        111,
+                                        505,
+                                        123
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Xiangyu Zhang, Jianhua Zou, Xiang Ming, Kaiming He, and Jian Sun. Efficient and accurate approximations",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                121,
+                                315,
+                                132
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        121,
+                                        315,
+                                        132
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of nonlinear convolutional networks. In CVPR, 2015b.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        }
+                    ],
+                    "index": 2.5
+                }
+            ],
+            "page_idx": 11,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        300,
+                        751,
+                        311,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                299,
+                                750,
+                                313,
+                                764
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        299,
+                                        750,
+                                        313,
+                                        764
+                                    ],
+                                    "score": 1.0,
+                                    "content": "12",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        107,
+                        27,
+                        308,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                107,
+                                26,
+                                308,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        26,
+                                        308,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Under review as a conference paper at ICLR 2017",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        83,
+                        504,
+                        104
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                83,
+                                505,
+                                95
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        83,
+                                        505,
+                                        95
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Xiangyu Zhang, Jianhua Zou, Kaiming He, and Jian Sun. Accelerating very deep convolutional networks for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                93,
+                                483,
+                                105
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        93,
+                                        483,
+                                        105
+                                    ],
+                                    "score": 1.0,
+                                    "content": "classification and detection. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2015a.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        }
+                    ],
+                    "index": 0.5,
+                    "bbox_fs": [
+                        106,
+                        83,
+                        505,
+                        105
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        111,
+                        504,
+                        131
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                111,
+                                505,
+                                123
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        111,
+                                        505,
+                                        123
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Xiangyu Zhang, Jianhua Zou, Xiang Ming, Kaiming He, and Jian Sun. Efficient and accurate approximations",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                121,
+                                315,
+                                132
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        121,
+                                        315,
+                                        132
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of nonlinear convolutional networks. In CVPR, 2015b.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        }
+                    ],
+                    "index": 2.5,
+                    "bbox_fs": [
+                        105,
+                        111,
+                        505,
+                        132
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        82,
+                        442,
+                        93
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                81,
+                                444,
+                                96
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        81,
+                                        444,
+                                        96
+                                    ],
+                                    "score": 1.0,
+                                    "content": "A ADDITIONAL DETAILS ON DENSE VS. SPARSE CONVOLUTIONS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        }
+                    ],
+                    "index": 0
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        104,
+                        505,
+                        149
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                105,
+                                505,
+                                117
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        105,
+                                        505,
+                                        117
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We contrast naive implementations of dense and sparse convolutions (cf. Sect. 3) in Algorithm. 1",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                115,
+                                505,
+                                129
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        115,
+                                        505,
+                                        129
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and Algorithm 2. We emphasize that we do not use sparse matrices and only introduce sparsity from",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                127,
+                                505,
+                                139
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        127,
+                                        505,
+                                        139
+                                    ],
+                                    "score": 1.0,
+                                    "content": "channel to channel. Thus, walltime will be mostly in terms of Multiply-Adds; the basic operation",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                137,
+                                423,
+                                150
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        137,
+                                        423,
+                                        150
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(convolving the entire image plane in Line 8 of both algorithms) is unchanged.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        }
+                    ],
+                    "index": 2.5
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        106,
+                        161,
+                        336,
+                        173
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                160,
+                                338,
+                                174
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        160,
+                                        338,
+                                        174
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Algorithm 1 Naive implementation of dense convolution",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        }
+                    ],
+                    "index": 5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        110,
+                        177,
+                        385,
+                        302
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                110,
+                                176,
+                                154,
+                                189
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        110,
+                                        176,
+                                        154,
+                                        189
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1: Inputs:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                111,
+                                187,
+                                207,
+                                199
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        111,
+                                        187,
+                                        207,
+                                        199
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2: - input: Data tensor",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                110,
+                                198,
+                                216,
+                                210
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        110,
+                                        198,
+                                        216,
+                                        210
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3: - W : Parameter tensor",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                110,
+                                209,
+                                312,
+                                222
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        110,
+                                        209,
+                                        312,
+                                        222
+                                    ],
+                                    "score": 1.0,
+                                    "content": "4: - input channels: Array of input channel IDs",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                110,
+                                220,
+                                321,
+                                232
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        110,
+                                        220,
+                                        321,
+                                        232
+                                    ],
+                                    "score": 1.0,
+                                    "content": "5: - output channels: Array of output channel IDs",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                112,
+                                235,
+                                237,
+                                246
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        112,
+                                        235,
+                                        138,
+                                        246
+                                    ],
+                                    "score": 1.0,
+                                    "content": "6: for",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        138,
+                                        236,
+                                        143,
+                                        245
+                                    ],
+                                    "score": 0.49,
+                                    "content": "i",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        144,
+                                        235,
+                                        237,
+                                        246
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in input channels do",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                110,
+                                245,
+                                258,
+                                258
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        110,
+                                        245,
+                                        122,
+                                        258
+                                    ],
+                                    "score": 1.0,
+                                    "content": "7:",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        137,
+                                        246,
+                                        153,
+                                        257
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        154,
+                                        248,
+                                        160,
+                                        256
+                                    ],
+                                    "score": 0.27,
+                                    "content": "o",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        160,
+                                        246,
+                                        258,
+                                        257
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in output channels do",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                110,
+                                257,
+                                384,
+                                269
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        110,
+                                        257,
+                                        122,
+                                        268
+                                    ],
+                                    "score": 1.0,
+                                    "content": "8:",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        153,
+                                        257,
+                                        384,
+                                        269
+                                    ],
+                                    "score": 0.39,
+                                    "content": "o u t p u t [ o ] \\gets o u t p u t [ o ] + \\mathrm { c o n v o l v e } ( i n p u t [ i ] , W [ i , o , . . . ] )",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                110,
+                                267,
+                                172,
+                                280
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        110,
+                                        268,
+                                        122,
+                                        280
+                                    ],
+                                    "score": 1.0,
+                                    "content": "9:",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        137,
+                                        267,
+                                        172,
+                                        279
+                                    ],
+                                    "score": 1.0,
+                                    "content": "end for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                278,
+                                158,
+                                290
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        278,
+                                        158,
+                                        290
+                                    ],
+                                    "score": 1.0,
+                                    "content": "10: end for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                108,
+                                290,
+                                184,
+                                302
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        108,
+                                        290,
+                                        184,
+                                        302
+                                    ],
+                                    "score": 1.0,
+                                    "content": "11: return output",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        }
+                    ],
+                    "index": 11
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        324,
+                        338,
+                        335
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                321,
+                                340,
+                                339
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        321,
+                                        340,
+                                        339
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Algorithm 2 Naive implementation of sparse convolution",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        }
+                    ],
+                    "index": 17
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        110,
+                        340,
+                        505,
+                        478
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                110,
+                                338,
+                                155,
+                                352
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        110,
+                                        338,
+                                        155,
+                                        352
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1: Inputs:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                110,
+                                349,
+                                207,
+                                362
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        110,
+                                        349,
+                                        207,
+                                        362
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2: - input: Data tensor",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                110,
+                                360,
+                                216,
+                                372
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        110,
+                                        360,
+                                        216,
+                                        372
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3: - W : Parameter tensor",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                110,
+                                371,
+                                312,
+                                384
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        110,
+                                        371,
+                                        312,
+                                        384
+                                    ],
+                                    "score": 1.0,
+                                    "content": "4: - input channels: Array of input channel IDs",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                109,
+                                381,
+                                505,
+                                396
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        109,
+                                        381,
+                                        505,
+                                        396
+                                    ],
+                                    "score": 1.0,
+                                    "content": "5: - output channels connected to i: Array of array of output channel IDs specifying connec-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                123,
+                                393,
+                                234,
+                                406
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        123,
+                                        393,
+                                        234,
+                                        406
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tions to each input channel",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                110,
+                                408,
+                                237,
+                                420
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        110,
+                                        408,
+                                        138,
+                                        420
+                                    ],
+                                    "score": 1.0,
+                                    "content": "6: for",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        138,
+                                        409,
+                                        143,
+                                        418
+                                    ],
+                                    "score": 0.29,
+                                    "content": "i",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        144,
+                                        408,
+                                        237,
+                                        420
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in input channels do",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                110,
+                                419,
+                                411,
+                                432
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        110,
+                                        419,
+                                        122,
+                                        431
+                                    ],
+                                    "score": 1.0,
+                                    "content": "7:",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        137,
+                                        419,
+                                        183,
+                                        432
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for index,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        183,
+                                        421,
+                                        190,
+                                        429
+                                    ],
+                                    "score": 0.33,
+                                    "content": "o",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        190,
+                                        419,
+                                        411,
+                                        432
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in enumerate(output channels connected to i[i]) do",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                110,
+                                429,
+                                406,
+                                444
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        110,
+                                        430,
+                                        122,
+                                        442
+                                    ],
+                                    "score": 1.0,
+                                    "content": "8:",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        151,
+                                        429,
+                                        193,
+                                        444
+                                    ],
+                                    "score": 1.0,
+                                    "content": "output[o]",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        194,
+                                        430,
+                                        207,
+                                        441
+                                    ],
+                                    "score": 0.44,
+                                    "content": "\\gets",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        208,
+                                        429,
+                                        248,
+                                        444
+                                    ],
+                                    "score": 1.0,
+                                    "content": "output[o]",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        248,
+                                        431,
+                                        257,
+                                        441
+                                    ],
+                                    "score": 0.63,
+                                    "content": "^ +",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        257,
+                                        429,
+                                        406,
+                                        444
+                                    ],
+                                    "score": 1.0,
+                                    "content": "convolve(input[i], W [i, index, . . .])",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                109,
+                                441,
+                                172,
+                                453
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        109,
+                                        441,
+                                        122,
+                                        453
+                                    ],
+                                    "score": 1.0,
+                                    "content": "9:",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        137,
+                                        441,
+                                        172,
+                                        453
+                                    ],
+                                    "score": 1.0,
+                                    "content": "end for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                107,
+                                452,
+                                157,
+                                464
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        452,
+                                        157,
+                                        464
+                                    ],
+                                    "score": 1.0,
+                                    "content": "10: end for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                108,
+                                463,
+                                183,
+                                474
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        108,
+                                        463,
+                                        183,
+                                        474
+                                    ],
+                                    "score": 1.0,
+                                    "content": "11: return output",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        }
+                    ],
+                    "index": 23.5
+                }
+            ],
+            "page_idx": 12,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        107,
+                        27,
+                        308,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                107,
+                                26,
+                                308,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        26,
+                                        308,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Under review as a conference paper at ICLR 2017",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        300,
+                        751,
+                        311,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                299,
+                                750,
+                                313,
+                                764
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        299,
+                                        750,
+                                        313,
+                                        764
+                                    ],
+                                    "score": 1.0,
+                                    "content": "13",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        82,
+                        442,
+                        93
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                81,
+                                444,
+                                96
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        81,
+                                        444,
+                                        96
+                                    ],
+                                    "score": 1.0,
+                                    "content": "A ADDITIONAL DETAILS ON DENSE VS. SPARSE CONVOLUTIONS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        }
+                    ],
+                    "index": 0
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        104,
+                        505,
+                        149
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                105,
+                                505,
+                                117
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        105,
+                                        505,
+                                        117
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We contrast naive implementations of dense and sparse convolutions (cf. Sect. 3) in Algorithm. 1",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                115,
+                                505,
+                                129
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        115,
+                                        505,
+                                        129
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and Algorithm 2. We emphasize that we do not use sparse matrices and only introduce sparsity from",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                127,
+                                505,
+                                139
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        127,
+                                        505,
+                                        139
+                                    ],
+                                    "score": 1.0,
+                                    "content": "channel to channel. Thus, walltime will be mostly in terms of Multiply-Adds; the basic operation",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                137,
+                                423,
+                                150
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        137,
+                                        423,
+                                        150
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(convolving the entire image plane in Line 8 of both algorithms) is unchanged.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        }
+                    ],
+                    "index": 2.5,
+                    "bbox_fs": [
+                        105,
+                        105,
+                        505,
+                        150
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        106,
+                        161,
+                        336,
+                        173
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                160,
+                                338,
+                                174
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        160,
+                                        338,
+                                        174
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Algorithm 1 Naive implementation of dense convolution",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        }
+                    ],
+                    "index": 5
+                },
+                {
+                    "type": "index",
+                    "bbox": [
+                        110,
+                        177,
+                        385,
+                        302
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                110,
+                                176,
+                                154,
+                                189
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        110,
+                                        176,
+                                        154,
+                                        189
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1: Inputs:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                111,
+                                187,
+                                207,
+                                199
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        111,
+                                        187,
+                                        207,
+                                        199
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2: - input: Data tensor",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                110,
+                                198,
+                                216,
+                                210
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        110,
+                                        198,
+                                        216,
+                                        210
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3: - W : Parameter tensor",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                110,
+                                209,
+                                312,
+                                222
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        110,
+                                        209,
+                                        312,
+                                        222
+                                    ],
+                                    "score": 1.0,
+                                    "content": "4: - input channels: Array of input channel IDs",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                110,
+                                220,
+                                321,
+                                232
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        110,
+                                        220,
+                                        321,
+                                        232
+                                    ],
+                                    "score": 1.0,
+                                    "content": "5: - output channels: Array of output channel IDs",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                112,
+                                235,
+                                237,
+                                246
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        112,
+                                        235,
+                                        138,
+                                        246
+                                    ],
+                                    "score": 1.0,
+                                    "content": "6: for",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        138,
+                                        236,
+                                        143,
+                                        245
+                                    ],
+                                    "score": 0.49,
+                                    "content": "i",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        144,
+                                        235,
+                                        237,
+                                        246
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in input channels do",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                110,
+                                245,
+                                258,
+                                258
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        110,
+                                        245,
+                                        122,
+                                        258
+                                    ],
+                                    "score": 1.0,
+                                    "content": "7:",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        137,
+                                        246,
+                                        153,
+                                        257
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        154,
+                                        248,
+                                        160,
+                                        256
+                                    ],
+                                    "score": 0.27,
+                                    "content": "o",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        160,
+                                        246,
+                                        258,
+                                        257
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in output channels do",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                110,
+                                257,
+                                384,
+                                269
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        110,
+                                        257,
+                                        122,
+                                        268
+                                    ],
+                                    "score": 1.0,
+                                    "content": "8:",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        153,
+                                        257,
+                                        384,
+                                        269
+                                    ],
+                                    "score": 0.39,
+                                    "content": "o u t p u t [ o ] \\gets o u t p u t [ o ] + \\mathrm { c o n v o l v e } ( i n p u t [ i ] , W [ i , o , . . . ] )",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 13,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                110,
+                                267,
+                                172,
+                                280
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        110,
+                                        268,
+                                        122,
+                                        280
+                                    ],
+                                    "score": 1.0,
+                                    "content": "9:",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        137,
+                                        267,
+                                        172,
+                                        279
+                                    ],
+                                    "score": 1.0,
+                                    "content": "end for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                278,
+                                158,
+                                290
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        278,
+                                        158,
+                                        290
+                                    ],
+                                    "score": 1.0,
+                                    "content": "10: end for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                108,
+                                290,
+                                184,
+                                302
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        108,
+                                        290,
+                                        184,
+                                        302
+                                    ],
+                                    "score": 1.0,
+                                    "content": "11: return output",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16,
+                            "is_list_start_line": true
+                        }
+                    ],
+                    "index": 11,
+                    "bbox_fs": [
+                        106,
+                        176,
+                        384,
+                        302
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        324,
+                        338,
+                        335
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                321,
+                                340,
+                                339
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        321,
+                                        340,
+                                        339
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Algorithm 2 Naive implementation of sparse convolution",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        }
+                    ],
+                    "index": 17,
+                    "bbox_fs": [
+                        105,
+                        321,
+                        340,
+                        339
+                    ]
+                },
+                {
+                    "type": "index",
+                    "bbox": [
+                        110,
+                        340,
+                        505,
+                        478
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                110,
+                                338,
+                                155,
+                                352
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        110,
+                                        338,
+                                        155,
+                                        352
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1: Inputs:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                110,
+                                349,
+                                207,
+                                362
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        110,
+                                        349,
+                                        207,
+                                        362
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2: - input: Data tensor",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                110,
+                                360,
+                                216,
+                                372
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        110,
+                                        360,
+                                        216,
+                                        372
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3: - W : Parameter tensor",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                110,
+                                371,
+                                312,
+                                384
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        110,
+                                        371,
+                                        312,
+                                        384
+                                    ],
+                                    "score": 1.0,
+                                    "content": "4: - input channels: Array of input channel IDs",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                109,
+                                381,
+                                505,
+                                396
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        109,
+                                        381,
+                                        505,
+                                        396
+                                    ],
+                                    "score": 1.0,
+                                    "content": "5: - output channels connected to i: Array of array of output channel IDs specifying connec-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                123,
+                                393,
+                                234,
+                                406
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        123,
+                                        393,
+                                        234,
+                                        406
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tions to each input channel",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                110,
+                                408,
+                                237,
+                                420
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        110,
+                                        408,
+                                        138,
+                                        420
+                                    ],
+                                    "score": 1.0,
+                                    "content": "6: for",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        138,
+                                        409,
+                                        143,
+                                        418
+                                    ],
+                                    "score": 0.29,
+                                    "content": "i",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        144,
+                                        408,
+                                        237,
+                                        420
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in input channels do",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                110,
+                                419,
+                                411,
+                                432
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        110,
+                                        419,
+                                        122,
+                                        431
+                                    ],
+                                    "score": 1.0,
+                                    "content": "7:",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        137,
+                                        419,
+                                        183,
+                                        432
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for index,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        183,
+                                        421,
+                                        190,
+                                        429
+                                    ],
+                                    "score": 0.33,
+                                    "content": "o",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        190,
+                                        419,
+                                        411,
+                                        432
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in enumerate(output channels connected to i[i]) do",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                110,
+                                429,
+                                406,
+                                444
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        110,
+                                        430,
+                                        122,
+                                        442
+                                    ],
+                                    "score": 1.0,
+                                    "content": "8:",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        151,
+                                        429,
+                                        193,
+                                        444
+                                    ],
+                                    "score": 1.0,
+                                    "content": "output[o]",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        194,
+                                        430,
+                                        207,
+                                        441
+                                    ],
+                                    "score": 0.44,
+                                    "content": "\\gets",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        208,
+                                        429,
+                                        248,
+                                        444
+                                    ],
+                                    "score": 1.0,
+                                    "content": "output[o]",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        248,
+                                        431,
+                                        257,
+                                        441
+                                    ],
+                                    "score": 0.63,
+                                    "content": "^ +",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        257,
+                                        429,
+                                        406,
+                                        444
+                                    ],
+                                    "score": 1.0,
+                                    "content": "convolve(input[i], W [i, index, . . .])",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                109,
+                                441,
+                                172,
+                                453
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        109,
+                                        441,
+                                        122,
+                                        453
+                                    ],
+                                    "score": 1.0,
+                                    "content": "9:",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        137,
+                                        441,
+                                        172,
+                                        453
+                                    ],
+                                    "score": 1.0,
+                                    "content": "end for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                107,
+                                452,
+                                157,
+                                464
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        452,
+                                        157,
+                                        464
+                                    ],
+                                    "score": 1.0,
+                                    "content": "10: end for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                108,
+                                463,
+                                183,
+                                474
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        108,
+                                        463,
+                                        183,
+                                        474
+                                    ],
+                                    "score": 1.0,
+                                    "content": "11: return output",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29,
+                            "is_list_start_line": true
+                        }
+                    ],
+                    "index": 23.5,
+                    "bbox_fs": [
+                        107,
+                        338,
+                        505,
+                        474
+                    ]
+                }
+            ]
+        }
+    ],
+    "_backend": "pipeline",
+    "_version_name": "2.2.2"
+}
\ No newline at end of file
diff --git a/parse/train/SkC_7v5gx/SkC_7v5gx_model.json b/parse/train/SkC_7v5gx/SkC_7v5gx_model.json
new file mode 100644
index 0000000000000000000000000000000000000000..a625ea5a108f5d97cca8bb057f13ffbe53a3044e
--- /dev/null
+++ b/parse/train/SkC_7v5gx/SkC_7v5gx_model.json
@@ -0,0 +1,17224 @@
+[
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    1189,
+                    1404,
+                    1189,
+                    1404,
+                    1554,
+                    299,
+                    1554
+                ],
+                "score": 0.985
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    398,
+                    680,
+                    1302,
+                    680,
+                    1302,
+                    1076,
+                    398,
+                    1076
+                ],
+                "score": 0.983
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1571,
+                    1403,
+                    1571,
+                    1403,
+                    1816,
+                    298,
+                    1816
+                ],
+                "score": 0.977
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    1831,
+                    1401,
+                    1831,
+                    1401,
+                    1985,
+                    299,
+                    1985
+                ],
+                "score": 0.97
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    298,
+                    221,
+                    1102,
+                    221,
+                    1102,
+                    324,
+                    298,
+                    324
+                ],
+                "score": 0.958
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    313,
+                    377,
+                    690,
+                    377,
+                    690,
+                    529,
+                    313,
+                    529
+                ],
+                "score": 0.911
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    327,
+                    2005,
+                    1160,
+                    2005,
+                    1160,
+                    2033,
+                    327,
+                    2033
+                ],
+                "score": 0.904
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    302,
+                    1123,
+                    573,
+                    1123,
+                    573,
+                    1158,
+                    302,
+                    1158
+                ],
+                "score": 0.893
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    298,
+                    75,
+                    857,
+                    75,
+                    857,
+                    104,
+                    298,
+                    104
+                ],
+                "score": 0.883
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    773,
+                    611,
+                    927,
+                    611,
+                    927,
+                    644,
+                    773,
+                    644
+                ],
+                "score": 0.87
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    842,
+                    2089,
+                    857,
+                    2089,
+                    857,
+                    2112,
+                    842,
+                    2112
+                ],
+                "score": 0.657
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    814,
+                    412,
+                    1294,
+                    412,
+                    1294,
+                    530,
+                    814,
+                    530
+                ],
+                "score": 0.519
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    814,
+                    378,
+                    1266,
+                    378,
+                    1266,
+                    408,
+                    814,
+                    408
+                ],
+                "score": 0.187
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    814,
+                    378,
+                    1266,
+                    378,
+                    1266,
+                    408,
+                    814,
+                    408
+                ],
+                "score": 0.127
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1000,
+                    1464,
+                    1042,
+                    1464,
+                    1042,
+                    1493,
+                    1000,
+                    1493
+                ],
+                "score": 0.86,
+                "latex": "5 \\%"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1243,
+                    1602,
+                    1285,
+                    1602,
+                    1285,
+                    1631,
+                    1243,
+                    1631
+                ],
+                "score": 0.85,
+                "latex": "5 \\%"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    217.0,
+                    891.0,
+                    217.0,
+                    891.0,
+                    270.0,
+                    296.0,
+                    270.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    271.0,
+                    1105.0,
+                    271.0,
+                    1105.0,
+                    328.0,
+                    296.0,
+                    328.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    328.0,
+                    1998.0,
+                    1163.0,
+                    1998.0,
+                    1163.0,
+                    2040.0,
+                    328.0,
+                    2040.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1120.0,
+                    579.0,
+                    1120.0,
+                    579.0,
+                    1167.0,
+                    294.0,
+                    1167.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    72.0,
+                    857.0,
+                    72.0,
+                    857.0,
+                    108.0,
+                    297.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    768.0,
+                    607.0,
+                    934.0,
+                    607.0,
+                    934.0,
+                    650.0,
+                    768.0,
+                    650.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    841.0,
+                    2088.0,
+                    860.0,
+                    2088.0,
+                    860.0,
+                    2118.0,
+                    841.0,
+                    2118.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    809.0,
+                    372.0,
+                    1268.0,
+                    372.0,
+                    1268.0,
+                    417.0,
+                    809.0,
+                    417.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1190.0,
+                    1405.0,
+                    1190.0,
+                    1405.0,
+                    1224.0,
+                    295.0,
+                    1224.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1220.0,
+                    1405.0,
+                    1220.0,
+                    1405.0,
+                    1254.0,
+                    295.0,
+                    1254.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1249.0,
+                    1405.0,
+                    1249.0,
+                    1405.0,
+                    1286.0,
+                    293.0,
+                    1286.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1281.0,
+                    1404.0,
+                    1281.0,
+                    1404.0,
+                    1315.0,
+                    294.0,
+                    1315.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1312.0,
+                    1404.0,
+                    1312.0,
+                    1404.0,
+                    1346.0,
+                    295.0,
+                    1346.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1340.0,
+                    1406.0,
+                    1340.0,
+                    1406.0,
+                    1376.0,
+                    294.0,
+                    1376.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1373.0,
+                    1406.0,
+                    1373.0,
+                    1406.0,
+                    1405.0,
+                    293.0,
+                    1405.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1403.0,
+                    1404.0,
+                    1403.0,
+                    1404.0,
+                    1434.0,
+                    293.0,
+                    1434.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1434.0,
+                    1403.0,
+                    1434.0,
+                    1403.0,
+                    1468.0,
+                    294.0,
+                    1468.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1462.0,
+                    999.0,
+                    1462.0,
+                    999.0,
+                    1497.0,
+                    294.0,
+                    1497.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1043.0,
+                    1462.0,
+                    1405.0,
+                    1462.0,
+                    1405.0,
+                    1497.0,
+                    1043.0,
+                    1497.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1496.0,
+                    1405.0,
+                    1496.0,
+                    1405.0,
+                    1526.0,
+                    295.0,
+                    1526.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1525.0,
+                    588.0,
+                    1525.0,
+                    588.0,
+                    1559.0,
+                    295.0,
+                    1559.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    394.0,
+                    681.0,
+                    1305.0,
+                    681.0,
+                    1305.0,
+                    716.0,
+                    394.0,
+                    716.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    394.0,
+                    712.0,
+                    1305.0,
+                    712.0,
+                    1305.0,
+                    746.0,
+                    394.0,
+                    746.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    394.0,
+                    740.0,
+                    1305.0,
+                    740.0,
+                    1305.0,
+                    776.0,
+                    394.0,
+                    776.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    394.0,
+                    774.0,
+                    1306.0,
+                    774.0,
+                    1306.0,
+                    804.0,
+                    394.0,
+                    804.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    393.0,
+                    802.0,
+                    1306.0,
+                    802.0,
+                    1306.0,
+                    837.0,
+                    393.0,
+                    837.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    393.0,
+                    833.0,
+                    1306.0,
+                    833.0,
+                    1306.0,
+                    868.0,
+                    393.0,
+                    868.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    393.0,
+                    862.0,
+                    1307.0,
+                    862.0,
+                    1307.0,
+                    900.0,
+                    393.0,
+                    900.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    395.0,
+                    896.0,
+                    1307.0,
+                    896.0,
+                    1307.0,
+                    928.0,
+                    395.0,
+                    928.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    393.0,
+                    923.0,
+                    1306.0,
+                    923.0,
+                    1306.0,
+                    962.0,
+                    393.0,
+                    962.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    393.0,
+                    954.0,
+                    1308.0,
+                    954.0,
+                    1308.0,
+                    990.0,
+                    393.0,
+                    990.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    394.0,
+                    985.0,
+                    1304.0,
+                    985.0,
+                    1304.0,
+                    1020.0,
+                    394.0,
+                    1020.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    393.0,
+                    1015.0,
+                    1306.0,
+                    1015.0,
+                    1306.0,
+                    1051.0,
+                    393.0,
+                    1051.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    393.0,
+                    1044.0,
+                    557.0,
+                    1044.0,
+                    557.0,
+                    1081.0,
+                    393.0,
+                    1081.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1570.0,
+                    1404.0,
+                    1570.0,
+                    1404.0,
+                    1606.0,
+                    293.0,
+                    1606.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1602.0,
+                    1242.0,
+                    1602.0,
+                    1242.0,
+                    1636.0,
+                    294.0,
+                    1636.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1286.0,
+                    1602.0,
+                    1405.0,
+                    1602.0,
+                    1405.0,
+                    1636.0,
+                    1286.0,
+                    1636.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1635.0,
+                    1404.0,
+                    1635.0,
+                    1404.0,
+                    1665.0,
+                    296.0,
+                    1665.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1661.0,
+                    1407.0,
+                    1661.0,
+                    1407.0,
+                    1700.0,
+                    292.0,
+                    1700.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1693.0,
+                    1407.0,
+                    1693.0,
+                    1407.0,
+                    1731.0,
+                    293.0,
+                    1731.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1725.0,
+                    1405.0,
+                    1725.0,
+                    1405.0,
+                    1758.0,
+                    294.0,
+                    1758.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1753.0,
+                    1407.0,
+                    1753.0,
+                    1407.0,
+                    1791.0,
+                    293.0,
+                    1791.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1785.0,
+                    1010.0,
+                    1785.0,
+                    1010.0,
+                    1819.0,
+                    293.0,
+                    1819.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    1833.0,
+                    1403.0,
+                    1833.0,
+                    1403.0,
+                    1866.0,
+                    297.0,
+                    1866.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1859.0,
+                    1405.0,
+                    1859.0,
+                    1405.0,
+                    1899.0,
+                    293.0,
+                    1899.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    290.0,
+                    1888.0,
+                    1406.0,
+                    1888.0,
+                    1406.0,
+                    1932.0,
+                    290.0,
+                    1932.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1921.0,
+                    1406.0,
+                    1921.0,
+                    1406.0,
+                    1960.0,
+                    293.0,
+                    1960.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1953.0,
+                    1405.0,
+                    1953.0,
+                    1405.0,
+                    1990.0,
+                    295.0,
+                    1990.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    312.0,
+                    375.0,
+                    559.0,
+                    375.0,
+                    559.0,
+                    412.0,
+                    312.0,
+                    412.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    311.0,
+                    406.0,
+                    692.0,
+                    406.0,
+                    692.0,
+                    442.0,
+                    311.0,
+                    442.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    312.0,
+                    437.0,
+                    692.0,
+                    437.0,
+                    692.0,
+                    469.0,
+                    312.0,
+                    469.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    312.0,
+                    469.0,
+                    656.0,
+                    469.0,
+                    656.0,
+                    500.0,
+                    312.0,
+                    500.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    314.0,
+                    500.0,
+                    586.0,
+                    500.0,
+                    586.0,
+                    532.0,
+                    314.0,
+                    532.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    560.0,
+                    380.0,
+                    569.0,
+                    380.0,
+                    569.0,
+                    393.0,
+                    560.0,
+                    393.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    810.0,
+                    406.0,
+                    949.0,
+                    406.0,
+                    949.0,
+                    441.0,
+                    810.0,
+                    441.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    810.0,
+                    432.0,
+                    1138.0,
+                    432.0,
+                    1138.0,
+                    474.0,
+                    810.0,
+                    474.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    810.0,
+                    468.0,
+                    1185.0,
+                    468.0,
+                    1185.0,
+                    499.0,
+                    810.0,
+                    499.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    812.0,
+                    499.0,
+                    1296.0,
+                    499.0,
+                    1296.0,
+                    533.0,
+                    812.0,
+                    533.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    809.0,
+                    372.0,
+                    1268.0,
+                    372.0,
+                    1268.0,
+                    417.0,
+                    809.0,
+                    417.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 0,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    445,
+                    1403,
+                    445,
+                    1403,
+                    719,
+                    298,
+                    719
+                ],
+                "score": 0.981
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1311,
+                    1403,
+                    1311,
+                    1403,
+                    1493,
+                    298,
+                    1493
+                ],
+                "score": 0.976
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    1128,
+                    1403,
+                    1128,
+                    1403,
+                    1281,
+                    299,
+                    1281
+                ],
+                "score": 0.975
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1508,
+                    1404,
+                    1508,
+                    1404,
+                    1786,
+                    298,
+                    1786
+                ],
+                "score": 0.975
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    300,
+                    875,
+                    1403,
+                    875,
+                    1403,
+                    966,
+                    300,
+                    966
+                ],
+                "score": 0.973
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    737,
+                    1403,
+                    737,
+                    1403,
+                    860,
+                    299,
+                    860
+                ],
+                "score": 0.972
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    300,
+                    229,
+                    1400,
+                    229,
+                    1400,
+                    323,
+                    300,
+                    323
+                ],
+                "score": 0.964
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    300,
+                    338,
+                    1399,
+                    338,
+                    1399,
+                    431,
+                    300,
+                    431
+                ],
+                "score": 0.96
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    296,
+                    1919,
+                    1403,
+                    1919,
+                    1403,
+                    2033,
+                    296,
+                    2033
+                ],
+                "score": 0.96
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    302,
+                    1800,
+                    1398,
+                    1800,
+                    1398,
+                    1893,
+                    302,
+                    1893
+                ],
+                "score": 0.947
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    301,
+                    1000,
+                    588,
+                    1000,
+                    588,
+                    1036,
+                    301,
+                    1036
+                ],
+                "score": 0.897
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    298,
+                    76,
+                    856,
+                    76,
+                    856,
+                    104,
+                    298,
+                    104
+                ],
+                "score": 0.859
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    299,
+                    1071,
+                    1000,
+                    1071,
+                    1000,
+                    1100,
+                    299,
+                    1100
+                ],
+                "score": 0.846
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    841,
+                    2088,
+                    859,
+                    2088,
+                    859,
+                    2112,
+                    841,
+                    2112
+                ],
+                "score": 0.736
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    300,
+                    76,
+                    854,
+                    76,
+                    854,
+                    104,
+                    300,
+                    104
+                ],
+                "score": 0.1
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1202,
+                    477,
+                    1257,
+                    477,
+                    1257,
+                    506,
+                    1202,
+                    506
+                ],
+                "score": 0.87,
+                "latex": "52 \\%"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1233,
+                    508,
+                    1289,
+                    508,
+                    1289,
+                    537,
+                    1233,
+                    537
+                ],
+                "score": 0.85,
+                "latex": "41 \\%"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    592,
+                    569,
+                    626,
+                    569,
+                    626,
+                    597,
+                    592,
+                    597
+                ],
+                "score": 0.68,
+                "latex": "7 \\mathbf { x }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    528,
+                    1463,
+                    558,
+                    1463,
+                    558,
+                    1490,
+                    528,
+                    1490
+                ],
+                "score": 0.62,
+                "latex": "\\ell 2"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    743,
+                    660,
+                    789,
+                    660,
+                    789,
+                    689,
+                    743,
+                    689
+                ],
+                "score": 0.5,
+                "latex": "3 0 \\mathrm { x }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1064,
+                    660,
+                    1123,
+                    660,
+                    1123,
+                    689,
+                    1064,
+                    689
+                ],
+                "score": 0.48,
+                "latex": "2 0 0 \\mathrm { x }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    467,
+                    1464,
+                    495,
+                    1464,
+                    495,
+                    1490,
+                    467,
+                    1490
+                ],
+                "score": 0.26,
+                "latex": "\\ell { 1 }"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    333.0,
+                    1914.0,
+                    1404.0,
+                    1914.0,
+                    1404.0,
+                    1953.0,
+                    333.0,
+                    1953.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1945.0,
+                    793.0,
+                    1945.0,
+                    793.0,
+                    1980.0,
+                    295.0,
+                    1980.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    327.0,
+                    1968.0,
+                    1407.0,
+                    1968.0,
+                    1407.0,
+                    2016.0,
+                    327.0,
+                    2016.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    2005.0,
+                    435.0,
+                    2005.0,
+                    435.0,
+                    2036.0,
+                    294.0,
+                    2036.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    996.0,
+                    594.0,
+                    996.0,
+                    594.0,
+                    1042.0,
+                    291.0,
+                    1042.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    298.0,
+                    73.0,
+                    858.0,
+                    73.0,
+                    858.0,
+                    108.0,
+                    298.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1068.0,
+                    1003.0,
+                    1068.0,
+                    1003.0,
+                    1104.0,
+                    295.0,
+                    1104.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    838.0,
+                    2085.0,
+                    862.0,
+                    2085.0,
+                    862.0,
+                    2120.0,
+                    838.0,
+                    2120.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    73.0,
+                    858.0,
+                    73.0,
+                    858.0,
+                    108.0,
+                    297.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    446.0,
+                    1405.0,
+                    446.0,
+                    1405.0,
+                    482.0,
+                    294.0,
+                    482.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    476.0,
+                    1201.0,
+                    476.0,
+                    1201.0,
+                    511.0,
+                    294.0,
+                    511.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1258.0,
+                    476.0,
+                    1402.0,
+                    476.0,
+                    1402.0,
+                    511.0,
+                    1258.0,
+                    511.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    505.0,
+                    1232.0,
+                    505.0,
+                    1232.0,
+                    544.0,
+                    292.0,
+                    544.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1290.0,
+                    505.0,
+                    1405.0,
+                    505.0,
+                    1405.0,
+                    544.0,
+                    1290.0,
+                    544.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    537.0,
+                    1405.0,
+                    537.0,
+                    1405.0,
+                    573.0,
+                    293.0,
+                    573.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    568.0,
+                    591.0,
+                    568.0,
+                    591.0,
+                    604.0,
+                    294.0,
+                    604.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    627.0,
+                    568.0,
+                    1407.0,
+                    568.0,
+                    1407.0,
+                    604.0,
+                    627.0,
+                    604.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    600.0,
+                    1404.0,
+                    600.0,
+                    1404.0,
+                    633.0,
+                    296.0,
+                    633.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    631.0,
+                    1404.0,
+                    631.0,
+                    1404.0,
+                    664.0,
+                    294.0,
+                    664.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    289.0,
+                    656.0,
+                    742.0,
+                    656.0,
+                    742.0,
+                    698.0,
+                    289.0,
+                    698.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    790.0,
+                    656.0,
+                    1063.0,
+                    656.0,
+                    1063.0,
+                    698.0,
+                    790.0,
+                    698.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1124.0,
+                    656.0,
+                    1407.0,
+                    656.0,
+                    1407.0,
+                    698.0,
+                    1124.0,
+                    698.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    686.0,
+                    418.0,
+                    686.0,
+                    418.0,
+                    726.0,
+                    294.0,
+                    726.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1309.0,
+                    1403.0,
+                    1309.0,
+                    1403.0,
+                    1345.0,
+                    293.0,
+                    1345.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1342.0,
+                    1403.0,
+                    1342.0,
+                    1403.0,
+                    1377.0,
+                    293.0,
+                    1377.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1372.0,
+                    1404.0,
+                    1372.0,
+                    1404.0,
+                    1407.0,
+                    294.0,
+                    1407.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1401.0,
+                    1403.0,
+                    1401.0,
+                    1403.0,
+                    1437.0,
+                    293.0,
+                    1437.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1431.0,
+                    1407.0,
+                    1431.0,
+                    1407.0,
+                    1470.0,
+                    293.0,
+                    1470.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1461.0,
+                    466.0,
+                    1461.0,
+                    466.0,
+                    1499.0,
+                    294.0,
+                    1499.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    496.0,
+                    1461.0,
+                    527.0,
+                    1461.0,
+                    527.0,
+                    1499.0,
+                    496.0,
+                    1499.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    559.0,
+                    1461.0,
+                    1128.0,
+                    1461.0,
+                    1128.0,
+                    1499.0,
+                    559.0,
+                    1499.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1127.0,
+                    1405.0,
+                    1127.0,
+                    1405.0,
+                    1164.0,
+                    295.0,
+                    1164.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    1160.0,
+                    1403.0,
+                    1160.0,
+                    1403.0,
+                    1193.0,
+                    297.0,
+                    1193.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    1190.0,
+                    1404.0,
+                    1190.0,
+                    1404.0,
+                    1223.0,
+                    297.0,
+                    1223.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1220.0,
+                    1404.0,
+                    1220.0,
+                    1404.0,
+                    1253.0,
+                    295.0,
+                    1253.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1252.0,
+                    1018.0,
+                    1252.0,
+                    1018.0,
+                    1285.0,
+                    295.0,
+                    1285.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1511.0,
+                    1404.0,
+                    1511.0,
+                    1404.0,
+                    1543.0,
+                    296.0,
+                    1543.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1538.0,
+                    1405.0,
+                    1538.0,
+                    1405.0,
+                    1575.0,
+                    294.0,
+                    1575.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1572.0,
+                    1405.0,
+                    1572.0,
+                    1405.0,
+                    1605.0,
+                    296.0,
+                    1605.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1600.0,
+                    1406.0,
+                    1600.0,
+                    1406.0,
+                    1636.0,
+                    294.0,
+                    1636.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1630.0,
+                    1408.0,
+                    1630.0,
+                    1408.0,
+                    1671.0,
+                    293.0,
+                    1671.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1659.0,
+                    1406.0,
+                    1659.0,
+                    1406.0,
+                    1700.0,
+                    295.0,
+                    1700.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1694.0,
+                    1405.0,
+                    1694.0,
+                    1405.0,
+                    1727.0,
+                    294.0,
+                    1727.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1720.0,
+                    1406.0,
+                    1720.0,
+                    1406.0,
+                    1760.0,
+                    292.0,
+                    1760.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1752.0,
+                    889.0,
+                    1752.0,
+                    889.0,
+                    1791.0,
+                    293.0,
+                    1791.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    875.0,
+                    1404.0,
+                    875.0,
+                    1404.0,
+                    909.0,
+                    296.0,
+                    909.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    906.0,
+                    1405.0,
+                    906.0,
+                    1405.0,
+                    940.0,
+                    296.0,
+                    940.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    937.0,
+                    675.0,
+                    937.0,
+                    675.0,
+                    968.0,
+                    295.0,
+                    968.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    735.0,
+                    1405.0,
+                    735.0,
+                    1405.0,
+                    773.0,
+                    293.0,
+                    773.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    767.0,
+                    1404.0,
+                    767.0,
+                    1404.0,
+                    802.0,
+                    294.0,
+                    802.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    796.0,
+                    1407.0,
+                    796.0,
+                    1407.0,
+                    833.0,
+                    293.0,
+                    833.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    825.0,
+                    605.0,
+                    825.0,
+                    605.0,
+                    866.0,
+                    293.0,
+                    866.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    231.0,
+                    1403.0,
+                    231.0,
+                    1403.0,
+                    261.0,
+                    296.0,
+                    261.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    260.0,
+                    1404.0,
+                    260.0,
+                    1404.0,
+                    295.0,
+                    294.0,
+                    295.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    292.0,
+                    701.0,
+                    292.0,
+                    701.0,
+                    326.0,
+                    296.0,
+                    326.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    337.0,
+                    1404.0,
+                    337.0,
+                    1404.0,
+                    375.0,
+                    296.0,
+                    375.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    368.0,
+                    1403.0,
+                    368.0,
+                    1403.0,
+                    404.0,
+                    294.0,
+                    404.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    398.0,
+                    1123.0,
+                    398.0,
+                    1123.0,
+                    435.0,
+                    295.0,
+                    435.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    1799.0,
+                    1403.0,
+                    1799.0,
+                    1403.0,
+                    1833.0,
+                    297.0,
+                    1833.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1831.0,
+                    1403.0,
+                    1831.0,
+                    1403.0,
+                    1865.0,
+                    296.0,
+                    1865.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1861.0,
+                    1403.0,
+                    1861.0,
+                    1403.0,
+                    1898.0,
+                    296.0,
+                    1898.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 1,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    230,
+                    1403,
+                    230,
+                    1403,
+                    658,
+                    298,
+                    658
+                ],
+                "score": 0.984
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1102,
+                    1404,
+                    1102,
+                    1404,
+                    1377,
+                    297,
+                    1377
+                ],
+                "score": 0.982
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    826,
+                    1403,
+                    826,
+                    1403,
+                    1072,
+                    298,
+                    1072
+                ],
+                "score": 0.981
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1467,
+                    1402,
+                    1467,
+                    1402,
+                    1682,
+                    298,
+                    1682
+                ],
+                "score": 0.979
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    672,
+                    1402,
+                    672,
+                    1402,
+                    796,
+                    299,
+                    796
+                ],
+                "score": 0.977
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    300,
+                    1911,
+                    1401,
+                    1911,
+                    1401,
+                    2033,
+                    300,
+                    2033
+                ],
+                "score": 0.975
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    1697,
+                    1402,
+                    1697,
+                    1402,
+                    1821,
+                    299,
+                    1821
+                ],
+                "score": 0.974
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    299,
+                    1408,
+                    873,
+                    1408,
+                    873,
+                    1439,
+                    299,
+                    1439
+                ],
+                "score": 0.905
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    301,
+                    1851,
+                    799,
+                    1851,
+                    799,
+                    1882,
+                    301,
+                    1882
+                ],
+                "score": 0.877
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    299,
+                    76,
+                    856,
+                    76,
+                    856,
+                    104,
+                    299,
+                    104
+                ],
+                "score": 0.802
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    842,
+                    2088,
+                    858,
+                    2088,
+                    858,
+                    2112,
+                    842,
+                    2112
+                ],
+                "score": 0.581
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    842,
+                    2088,
+                    859,
+                    2088,
+                    859,
+                    2111,
+                    842,
+                    2111
+                ],
+                "score": 0.42
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    300,
+                    76,
+                    854,
+                    76,
+                    854,
+                    104,
+                    300,
+                    104
+                ],
+                "score": 0.333
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1405.0,
+                    877.0,
+                    1405.0,
+                    877.0,
+                    1444.0,
+                    293.0,
+                    1444.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1850.0,
+                    806.0,
+                    1850.0,
+                    806.0,
+                    1886.0,
+                    295.0,
+                    1886.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    298.0,
+                    73.0,
+                    858.0,
+                    73.0,
+                    858.0,
+                    108.0,
+                    298.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    838.0,
+                    2085.0,
+                    862.0,
+                    2085.0,
+                    862.0,
+                    2118.0,
+                    838.0,
+                    2118.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    839.0,
+                    2086.0,
+                    860.0,
+                    2086.0,
+                    860.0,
+                    2117.0,
+                    839.0,
+                    2117.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    73.0,
+                    858.0,
+                    73.0,
+                    858.0,
+                    108.0,
+                    297.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    232.0,
+                    1404.0,
+                    232.0,
+                    1404.0,
+                    264.0,
+                    296.0,
+                    264.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    261.0,
+                    1405.0,
+                    261.0,
+                    1405.0,
+                    297.0,
+                    296.0,
+                    297.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    292.0,
+                    1404.0,
+                    292.0,
+                    1404.0,
+                    327.0,
+                    294.0,
+                    327.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    320.0,
+                    1405.0,
+                    320.0,
+                    1405.0,
+                    356.0,
+                    293.0,
+                    356.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    353.0,
+                    1408.0,
+                    353.0,
+                    1408.0,
+                    388.0,
+                    293.0,
+                    388.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    383.0,
+                    1407.0,
+                    383.0,
+                    1407.0,
+                    419.0,
+                    294.0,
+                    419.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    414.0,
+                    1407.0,
+                    414.0,
+                    1407.0,
+                    449.0,
+                    294.0,
+                    449.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    444.0,
+                    1405.0,
+                    444.0,
+                    1405.0,
+                    480.0,
+                    294.0,
+                    480.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    473.0,
+                    1406.0,
+                    473.0,
+                    1406.0,
+                    510.0,
+                    292.0,
+                    510.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    505.0,
+                    1407.0,
+                    505.0,
+                    1407.0,
+                    540.0,
+                    296.0,
+                    540.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    534.0,
+                    1405.0,
+                    534.0,
+                    1405.0,
+                    570.0,
+                    294.0,
+                    570.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    566.0,
+                    1405.0,
+                    566.0,
+                    1405.0,
+                    601.0,
+                    294.0,
+                    601.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    594.0,
+                    1405.0,
+                    594.0,
+                    1405.0,
+                    632.0,
+                    293.0,
+                    632.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    625.0,
+                    537.0,
+                    625.0,
+                    537.0,
+                    660.0,
+                    293.0,
+                    660.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1103.0,
+                    1405.0,
+                    1103.0,
+                    1405.0,
+                    1139.0,
+                    295.0,
+                    1139.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1134.0,
+                    1405.0,
+                    1134.0,
+                    1405.0,
+                    1167.0,
+                    296.0,
+                    1167.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1160.0,
+                    1406.0,
+                    1160.0,
+                    1406.0,
+                    1202.0,
+                    292.0,
+                    1202.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1187.0,
+                    1405.0,
+                    1187.0,
+                    1405.0,
+                    1233.0,
+                    292.0,
+                    1233.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1221.0,
+                    1408.0,
+                    1221.0,
+                    1408.0,
+                    1264.0,
+                    291.0,
+                    1264.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1253.0,
+                    1404.0,
+                    1253.0,
+                    1404.0,
+                    1289.0,
+                    295.0,
+                    1289.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1282.0,
+                    1406.0,
+                    1282.0,
+                    1406.0,
+                    1322.0,
+                    292.0,
+                    1322.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1315.0,
+                    1406.0,
+                    1315.0,
+                    1406.0,
+                    1351.0,
+                    295.0,
+                    1351.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1345.0,
+                    590.0,
+                    1345.0,
+                    590.0,
+                    1381.0,
+                    295.0,
+                    1381.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    825.0,
+                    1407.0,
+                    825.0,
+                    1407.0,
+                    862.0,
+                    296.0,
+                    862.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    857.0,
+                    1407.0,
+                    857.0,
+                    1407.0,
+                    893.0,
+                    292.0,
+                    893.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    887.0,
+                    1403.0,
+                    887.0,
+                    1403.0,
+                    923.0,
+                    293.0,
+                    923.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    919.0,
+                    1405.0,
+                    919.0,
+                    1405.0,
+                    953.0,
+                    294.0,
+                    953.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    949.0,
+                    1405.0,
+                    949.0,
+                    1405.0,
+                    983.0,
+                    294.0,
+                    983.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    975.0,
+                    1407.0,
+                    975.0,
+                    1407.0,
+                    1017.0,
+                    291.0,
+                    1017.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1008.0,
+                    1404.0,
+                    1008.0,
+                    1404.0,
+                    1048.0,
+                    292.0,
+                    1048.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1039.0,
+                    970.0,
+                    1039.0,
+                    970.0,
+                    1078.0,
+                    292.0,
+                    1078.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1467.0,
+                    1406.0,
+                    1467.0,
+                    1406.0,
+                    1506.0,
+                    293.0,
+                    1506.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1498.0,
+                    1404.0,
+                    1498.0,
+                    1404.0,
+                    1535.0,
+                    294.0,
+                    1535.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1528.0,
+                    1404.0,
+                    1528.0,
+                    1404.0,
+                    1565.0,
+                    294.0,
+                    1565.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1560.0,
+                    1404.0,
+                    1560.0,
+                    1404.0,
+                    1594.0,
+                    293.0,
+                    1594.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1589.0,
+                    1404.0,
+                    1589.0,
+                    1404.0,
+                    1624.0,
+                    293.0,
+                    1624.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1620.0,
+                    1404.0,
+                    1620.0,
+                    1404.0,
+                    1657.0,
+                    293.0,
+                    1657.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1652.0,
+                    751.0,
+                    1652.0,
+                    751.0,
+                    1686.0,
+                    294.0,
+                    1686.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    673.0,
+                    1402.0,
+                    673.0,
+                    1402.0,
+                    706.0,
+                    295.0,
+                    706.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    701.0,
+                    1406.0,
+                    701.0,
+                    1406.0,
+                    739.0,
+                    293.0,
+                    739.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    736.0,
+                    1404.0,
+                    736.0,
+                    1404.0,
+                    769.0,
+                    294.0,
+                    769.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    764.0,
+                    1097.0,
+                    764.0,
+                    1097.0,
+                    802.0,
+                    293.0,
+                    802.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1911.0,
+                    1405.0,
+                    1911.0,
+                    1405.0,
+                    1945.0,
+                    294.0,
+                    1945.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1943.0,
+                    1405.0,
+                    1943.0,
+                    1405.0,
+                    1975.0,
+                    295.0,
+                    1975.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1972.0,
+                    1405.0,
+                    1972.0,
+                    1405.0,
+                    2007.0,
+                    294.0,
+                    2007.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    2002.0,
+                    841.0,
+                    2002.0,
+                    841.0,
+                    2036.0,
+                    295.0,
+                    2036.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1695.0,
+                    1406.0,
+                    1695.0,
+                    1406.0,
+                    1732.0,
+                    293.0,
+                    1732.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1725.0,
+                    1406.0,
+                    1725.0,
+                    1406.0,
+                    1765.0,
+                    293.0,
+                    1765.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1759.0,
+                    1406.0,
+                    1759.0,
+                    1406.0,
+                    1792.0,
+                    293.0,
+                    1792.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1791.0,
+                    1056.0,
+                    1791.0,
+                    1056.0,
+                    1823.0,
+                    294.0,
+                    1823.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 2,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1406,
+                    1403,
+                    1406,
+                    1403,
+                    1621,
+                    298,
+                    1621
+                ],
+                "score": 0.981
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    481,
+                    1403,
+                    481,
+                    1403,
+                    664,
+                    299,
+                    664
+                ],
+                "score": 0.98
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    1636,
+                    1403,
+                    1636,
+                    1403,
+                    1794,
+                    299,
+                    1794
+                ],
+                "score": 0.978
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    1238,
+                    1403,
+                    1238,
+                    1403,
+                    1390,
+                    299,
+                    1390
+                ],
+                "score": 0.977
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    750,
+                    1403,
+                    750,
+                    1403,
+                    874,
+                    299,
+                    874
+                ],
+                "score": 0.976
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    1880,
+                    1402,
+                    1880,
+                    1402,
+                    2034,
+                    299,
+                    2034
+                ],
+                "score": 0.976
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    997,
+                    1404,
+                    997,
+                    1404,
+                    1150,
+                    299,
+                    1150
+                ],
+                "score": 0.976
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    229,
+                    1403,
+                    229,
+                    1403,
+                    384,
+                    299,
+                    384
+                ],
+                "score": 0.975
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    301,
+                    890,
+                    1398,
+                    890,
+                    1398,
+                    983,
+                    301,
+                    983
+                ],
+                "score": 0.963
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    300,
+                    1179,
+                    575,
+                    1179,
+                    575,
+                    1210,
+                    300,
+                    1210
+                ],
+                "score": 0.909
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    300,
+                    692,
+                    598,
+                    692,
+                    598,
+                    722,
+                    300,
+                    722
+                ],
+                "score": 0.898
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    299,
+                    415,
+                    509,
+                    415,
+                    509,
+                    450,
+                    299,
+                    450
+                ],
+                "score": 0.897
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    301,
+                    1822,
+                    688,
+                    1822,
+                    688,
+                    1854,
+                    301,
+                    1854
+                ],
+                "score": 0.895
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    298,
+                    75,
+                    856,
+                    75,
+                    856,
+                    104,
+                    298,
+                    104
+                ],
+                "score": 0.883
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    841,
+                    2089,
+                    858,
+                    2089,
+                    858,
+                    2111,
+                    841,
+                    2111
+                ],
+                "score": 0.752
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    297,
+                    1760,
+                    491,
+                    1760,
+                    491,
+                    1796,
+                    297,
+                    1796
+                ],
+                "score": 0.93,
+                "latex": "k ^ { 2 } \\times \\alpha n _ { l - 1 } \\times n _ { l }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    839,
+                    1729,
+                    919,
+                    1729,
+                    919,
+                    1763,
+                    839,
+                    1763
+                ],
+                "score": 0.92,
+                "latex": "W _ { s a _ { i j } j }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    604,
+                    920,
+                    700,
+                    920,
+                    700,
+                    953,
+                    604,
+                    953
+                ],
+                "score": 0.92,
+                "latex": "\\lceil \\alpha n _ { l - 1 } \\rceil"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    758,
+                    920,
+                    826,
+                    920,
+                    826,
+                    954,
+                    758,
+                    954
+                ],
+                "score": 0.92,
+                "latex": "\\lceil \\alpha n _ { l } \\rceil"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    453,
+                    782,
+                    573,
+                    782,
+                    573,
+                    816,
+                    453,
+                    816
+                ],
+                "score": 0.91,
+                "latex": "\\alpha \\in ( 0 , 1 ]"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1213,
+                    1696,
+                    1393,
+                    1696,
+                    1393,
+                    1730,
+                    1213,
+                    1730
+                ],
+                "score": 0.91,
+                "latex": "k ^ { 2 } \\times n _ { l - 1 } \\times n _ { l }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    883,
+                    996,
+                    944,
+                    996,
+                    944,
+                    1031,
+                    883,
+                    1031
+                ],
+                "score": 0.91,
+                "latex": "1 / \\alpha ^ { 2 }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    992,
+                    1698,
+                    1052,
+                    1698,
+                    1052,
+                    1732,
+                    992,
+                    1732
+                ],
+                "score": 0.9,
+                "latex": "W _ { s i j }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    527,
+                    1641,
+                    584,
+                    1641,
+                    584,
+                    1669,
+                    527,
+                    1669
+                ],
+                "score": 0.9,
+                "latex": "n _ { l - 1 }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1005,
+                    1735,
+                    1042,
+                    1735,
+                    1042,
+                    1763,
+                    1005,
+                    1763
+                ],
+                "score": 0.88,
+                "latex": "a _ { i j }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    361,
+                    926,
+                    418,
+                    926,
+                    418,
+                    953,
+                    361,
+                    953
+                ],
+                "score": 0.87,
+                "latex": "n _ { l - 1 }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    343,
+                    896,
+                    400,
+                    896,
+                    400,
+                    923,
+                    343,
+                    923
+                ],
+                "score": 0.87,
+                "latex": "n _ { l - 1 }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    582,
+                    952,
+                    646,
+                    952,
+                    646,
+                    980,
+                    582,
+                    980
+                ],
+                "score": 0.87,
+                "latex": "\\approx \\alpha ^ { 2 }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    297,
+                    1973,
+                    338,
+                    1973,
+                    338,
+                    2002,
+                    297,
+                    2002
+                ],
+                "score": 0.86,
+                "latex": "1 \\%"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    387,
+                    1973,
+                    450,
+                    1973,
+                    450,
+                    2003,
+                    387,
+                    2003
+                ],
+                "score": 0.85,
+                "latex": "0 . 1 \\%"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    635,
+                    1641,
+                    664,
+                    1641,
+                    664,
+                    1667,
+                    635,
+                    1667
+                ],
+                "score": 0.84,
+                "latex": "n _ { l }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    472,
+                    925,
+                    501,
+                    925,
+                    501,
+                    951,
+                    472,
+                    951
+                ],
+                "score": 0.83,
+                "latex": "n _ { l }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    835,
+                    1670,
+                    853,
+                    1670,
+                    853,
+                    1700,
+                    835,
+                    1700
+                ],
+                "score": 0.82,
+                "latex": "j"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    454,
+                    896,
+                    483,
+                    896,
+                    483,
+                    921,
+                    454,
+                    921
+                ],
+                "score": 0.82,
+                "latex": "n _ { l }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    543,
+                    1763,
+                    562,
+                    1763,
+                    562,
+                    1790,
+                    543,
+                    1790
+                ],
+                "score": 0.8,
+                "latex": "k"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1096,
+                    1674,
+                    1119,
+                    1674,
+                    1119,
+                    1695,
+                    1096,
+                    1695
+                ],
+                "score": 0.77,
+                "latex": "\\alpha"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    593,
+                    1674,
+                    614,
+                    1674,
+                    614,
+                    1696,
+                    593,
+                    1696
+                ],
+                "score": 0.74,
+                "latex": "\\alpha"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1062,
+                    1637,
+                    1122,
+                    1637,
+                    1122,
+                    1666,
+                    1062,
+                    1666
+                ],
+                "score": 0.73,
+                "latex": "l - 1"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    566,
+                    819,
+                    587,
+                    819,
+                    587,
+                    840,
+                    566,
+                    840
+                ],
+                "score": 0.72,
+                "latex": "\\alpha"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1051,
+                    892,
+                    1064,
+                    892,
+                    1064,
+                    918,
+                    1051,
+                    918
+                ],
+                "score": 0.57,
+                "latex": "l"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1237,
+                    1639,
+                    1249,
+                    1639,
+                    1249,
+                    1665,
+                    1237,
+                    1665
+                ],
+                "score": 0.49,
+                "latex": "l"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1176.0,
+                    579.0,
+                    1176.0,
+                    579.0,
+                    1215.0,
+                    294.0,
+                    1215.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    691.0,
+                    602.0,
+                    691.0,
+                    602.0,
+                    725.0,
+                    295.0,
+                    725.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    410.0,
+                    515.0,
+                    410.0,
+                    515.0,
+                    457.0,
+                    291.0,
+                    457.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1821.0,
+                    692.0,
+                    1821.0,
+                    692.0,
+                    1856.0,
+                    296.0,
+                    1856.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    298.0,
+                    72.0,
+                    857.0,
+                    72.0,
+                    857.0,
+                    108.0,
+                    298.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    838.0,
+                    2086.0,
+                    862.0,
+                    2086.0,
+                    862.0,
+                    2118.0,
+                    838.0,
+                    2118.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1408.0,
+                    1403.0,
+                    1408.0,
+                    1403.0,
+                    1442.0,
+                    296.0,
+                    1442.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1440.0,
+                    1405.0,
+                    1440.0,
+                    1405.0,
+                    1470.0,
+                    294.0,
+                    1470.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1468.0,
+                    1405.0,
+                    1468.0,
+                    1405.0,
+                    1502.0,
+                    293.0,
+                    1502.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1499.0,
+                    1403.0,
+                    1499.0,
+                    1403.0,
+                    1532.0,
+                    293.0,
+                    1532.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1528.0,
+                    1404.0,
+                    1528.0,
+                    1404.0,
+                    1563.0,
+                    293.0,
+                    1563.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1559.0,
+                    1404.0,
+                    1559.0,
+                    1404.0,
+                    1596.0,
+                    293.0,
+                    1596.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1591.0,
+                    1075.0,
+                    1591.0,
+                    1075.0,
+                    1625.0,
+                    296.0,
+                    1625.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    479.0,
+                    1405.0,
+                    479.0,
+                    1405.0,
+                    517.0,
+                    293.0,
+                    517.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    514.0,
+                    1404.0,
+                    514.0,
+                    1404.0,
+                    546.0,
+                    295.0,
+                    546.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    540.0,
+                    1405.0,
+                    540.0,
+                    1405.0,
+                    581.0,
+                    292.0,
+                    581.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    573.0,
+                    1403.0,
+                    573.0,
+                    1403.0,
+                    608.0,
+                    294.0,
+                    608.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    603.0,
+                    1405.0,
+                    603.0,
+                    1405.0,
+                    639.0,
+                    293.0,
+                    639.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    636.0,
+                    1325.0,
+                    636.0,
+                    1325.0,
+                    667.0,
+                    295.0,
+                    667.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1637.0,
+                    526.0,
+                    1637.0,
+                    526.0,
+                    1671.0,
+                    295.0,
+                    1671.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    585.0,
+                    1637.0,
+                    634.0,
+                    1637.0,
+                    634.0,
+                    1671.0,
+                    585.0,
+                    1671.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    665.0,
+                    1637.0,
+                    1061.0,
+                    1637.0,
+                    1061.0,
+                    1671.0,
+                    665.0,
+                    1671.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1123.0,
+                    1637.0,
+                    1236.0,
+                    1637.0,
+                    1236.0,
+                    1671.0,
+                    1123.0,
+                    1671.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1250.0,
+                    1637.0,
+                    1402.0,
+                    1637.0,
+                    1402.0,
+                    1671.0,
+                    1250.0,
+                    1671.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1668.0,
+                    592.0,
+                    1668.0,
+                    592.0,
+                    1702.0,
+                    295.0,
+                    1702.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    615.0,
+                    1668.0,
+                    834.0,
+                    1668.0,
+                    834.0,
+                    1702.0,
+                    615.0,
+                    1702.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    854.0,
+                    1668.0,
+                    1095.0,
+                    1668.0,
+                    1095.0,
+                    1702.0,
+                    854.0,
+                    1702.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1120.0,
+                    1668.0,
+                    1404.0,
+                    1668.0,
+                    1404.0,
+                    1702.0,
+                    1120.0,
+                    1702.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1694.0,
+                    991.0,
+                    1694.0,
+                    991.0,
+                    1735.0,
+                    292.0,
+                    1735.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1053.0,
+                    1694.0,
+                    1212.0,
+                    1694.0,
+                    1212.0,
+                    1735.0,
+                    1053.0,
+                    1735.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1394.0,
+                    1694.0,
+                    1408.0,
+                    1694.0,
+                    1408.0,
+                    1735.0,
+                    1394.0,
+                    1735.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1724.0,
+                    838.0,
+                    1724.0,
+                    838.0,
+                    1770.0,
+                    292.0,
+                    1770.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    920.0,
+                    1724.0,
+                    1004.0,
+                    1724.0,
+                    1004.0,
+                    1770.0,
+                    920.0,
+                    1770.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1043.0,
+                    1724.0,
+                    1408.0,
+                    1724.0,
+                    1408.0,
+                    1770.0,
+                    1043.0,
+                    1770.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1758.0,
+                    296.0,
+                    1758.0,
+                    296.0,
+                    1800.0,
+                    293.0,
+                    1800.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    492.0,
+                    1758.0,
+                    542.0,
+                    1758.0,
+                    542.0,
+                    1800.0,
+                    492.0,
+                    1800.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    563.0,
+                    1758.0,
+                    765.0,
+                    1758.0,
+                    765.0,
+                    1800.0,
+                    563.0,
+                    1800.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1239.0,
+                    1404.0,
+                    1239.0,
+                    1404.0,
+                    1272.0,
+                    295.0,
+                    1272.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1269.0,
+                    1404.0,
+                    1269.0,
+                    1404.0,
+                    1302.0,
+                    295.0,
+                    1302.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1297.0,
+                    1405.0,
+                    1297.0,
+                    1405.0,
+                    1336.0,
+                    293.0,
+                    1336.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1327.0,
+                    1405.0,
+                    1327.0,
+                    1405.0,
+                    1365.0,
+                    293.0,
+                    1365.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1356.0,
+                    570.0,
+                    1356.0,
+                    570.0,
+                    1396.0,
+                    293.0,
+                    1396.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    749.0,
+                    1404.0,
+                    749.0,
+                    1404.0,
+                    786.0,
+                    294.0,
+                    786.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    783.0,
+                    452.0,
+                    783.0,
+                    452.0,
+                    815.0,
+                    293.0,
+                    815.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    574.0,
+                    783.0,
+                    1407.0,
+                    783.0,
+                    1407.0,
+                    815.0,
+                    574.0,
+                    815.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    814.0,
+                    565.0,
+                    814.0,
+                    565.0,
+                    845.0,
+                    294.0,
+                    845.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    588.0,
+                    814.0,
+                    1405.0,
+                    814.0,
+                    1405.0,
+                    845.0,
+                    588.0,
+                    845.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    845.0,
+                    1067.0,
+                    845.0,
+                    1067.0,
+                    877.0,
+                    295.0,
+                    877.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1883.0,
+                    1407.0,
+                    1883.0,
+                    1407.0,
+                    1916.0,
+                    295.0,
+                    1916.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1911.0,
+                    1406.0,
+                    1911.0,
+                    1406.0,
+                    1946.0,
+                    294.0,
+                    1946.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1941.0,
+                    1404.0,
+                    1941.0,
+                    1404.0,
+                    1978.0,
+                    293.0,
+                    1978.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    339.0,
+                    1969.0,
+                    386.0,
+                    1969.0,
+                    386.0,
+                    2009.0,
+                    339.0,
+                    2009.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    451.0,
+                    1969.0,
+                    1406.0,
+                    1969.0,
+                    1406.0,
+                    2009.0,
+                    451.0,
+                    2009.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    2003.0,
+                    1404.0,
+                    2003.0,
+                    1404.0,
+                    2036.0,
+                    295.0,
+                    2036.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    998.0,
+                    882.0,
+                    998.0,
+                    882.0,
+                    1031.0,
+                    297.0,
+                    1031.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    945.0,
+                    998.0,
+                    1403.0,
+                    998.0,
+                    1403.0,
+                    1031.0,
+                    945.0,
+                    1031.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1029.0,
+                    1403.0,
+                    1029.0,
+                    1403.0,
+                    1062.0,
+                    295.0,
+                    1062.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1056.0,
+                    1405.0,
+                    1056.0,
+                    1405.0,
+                    1095.0,
+                    293.0,
+                    1095.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1088.0,
+                    1404.0,
+                    1088.0,
+                    1404.0,
+                    1125.0,
+                    295.0,
+                    1125.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1121.0,
+                    624.0,
+                    1121.0,
+                    624.0,
+                    1154.0,
+                    295.0,
+                    1154.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    228.0,
+                    1405.0,
+                    228.0,
+                    1405.0,
+                    267.0,
+                    294.0,
+                    267.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    260.0,
+                    1403.0,
+                    260.0,
+                    1403.0,
+                    297.0,
+                    295.0,
+                    297.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    291.0,
+                    1403.0,
+                    291.0,
+                    1403.0,
+                    328.0,
+                    293.0,
+                    328.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    322.0,
+                    1404.0,
+                    322.0,
+                    1404.0,
+                    356.0,
+                    295.0,
+                    356.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    352.0,
+                    1020.0,
+                    352.0,
+                    1020.0,
+                    389.0,
+                    294.0,
+                    389.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    887.0,
+                    342.0,
+                    887.0,
+                    342.0,
+                    927.0,
+                    292.0,
+                    927.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    401.0,
+                    887.0,
+                    453.0,
+                    887.0,
+                    453.0,
+                    927.0,
+                    401.0,
+                    927.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    484.0,
+                    887.0,
+                    1050.0,
+                    887.0,
+                    1050.0,
+                    927.0,
+                    484.0,
+                    927.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1065.0,
+                    887.0,
+                    1403.0,
+                    887.0,
+                    1403.0,
+                    927.0,
+                    1065.0,
+                    927.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    921.0,
+                    360.0,
+                    921.0,
+                    360.0,
+                    955.0,
+                    296.0,
+                    955.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    419.0,
+                    921.0,
+                    471.0,
+                    921.0,
+                    471.0,
+                    955.0,
+                    419.0,
+                    955.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    502.0,
+                    921.0,
+                    603.0,
+                    921.0,
+                    603.0,
+                    955.0,
+                    502.0,
+                    955.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    701.0,
+                    921.0,
+                    757.0,
+                    921.0,
+                    757.0,
+                    955.0,
+                    701.0,
+                    955.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    827.0,
+                    921.0,
+                    1404.0,
+                    921.0,
+                    1404.0,
+                    955.0,
+                    827.0,
+                    955.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    950.0,
+                    581.0,
+                    950.0,
+                    581.0,
+                    987.0,
+                    296.0,
+                    987.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    647.0,
+                    950.0,
+                    911.0,
+                    950.0,
+                    911.0,
+                    987.0,
+                    647.0,
+                    987.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 3,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    851,
+                    1404,
+                    851,
+                    1404,
+                    1037,
+                    297,
+                    1037
+                ],
+                "score": 0.981
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1419,
+                    1405,
+                    1419,
+                    1405,
+                    1605,
+                    297,
+                    1605
+                ],
+                "score": 0.98
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    621,
+                    1404,
+                    621,
+                    1404,
+                    747,
+                    298,
+                    747
+                ],
+                "score": 0.979
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1195,
+                    1404,
+                    1195,
+                    1404,
+                    1350,
+                    297,
+                    1350
+                ],
+                "score": 0.978
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    1941,
+                    1402,
+                    1941,
+                    1402,
+                    2036,
+                    299,
+                    2036
+                ],
+                "score": 0.971
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1067,
+                    1403,
+                    1067,
+                    1403,
+                    1167,
+                    297,
+                    1167
+                ],
+                "score": 0.971
+            },
+            {
+                "category_id": 4,
+                "poly": [
+                    296,
+                    414,
+                    1407,
+                    414,
+                    1407,
+                    567,
+                    296,
+                    567
+                ],
+                "score": 0.968
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1690,
+                    1402,
+                    1690,
+                    1402,
+                    1756,
+                    297,
+                    1756
+                ],
+                "score": 0.958
+            },
+            {
+                "category_id": 3,
+                "poly": [
+                    349,
+                    235,
+                    1333,
+                    235,
+                    1333,
+                    410,
+                    349,
+                    410
+                ],
+                "score": 0.956
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    579,
+                    1628,
+                    1119,
+                    1628,
+                    1119,
+                    1668,
+                    579,
+                    1668
+                ],
+                "score": 0.942
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    591,
+                    1362,
+                    1106,
+                    1362,
+                    1106,
+                    1400,
+                    591,
+                    1400
+                ],
+                "score": 0.938
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    301,
+                    1893,
+                    1151,
+                    1893,
+                    1151,
+                    1927,
+                    301,
+                    1927
+                ],
+                "score": 0.93
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    318,
+                    1785,
+                    1378,
+                    1785,
+                    1378,
+                    1872,
+                    318,
+                    1872
+                ],
+                "score": 0.929
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    299,
+                    785,
+                    497,
+                    785,
+                    497,
+                    818,
+                    299,
+                    818
+                ],
+                "score": 0.905
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    297,
+                    75,
+                    857,
+                    75,
+                    857,
+                    105,
+                    297,
+                    105
+                ],
+                "score": 0.89
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    1376,
+                    1896,
+                    1400,
+                    1896,
+                    1400,
+                    1922,
+                    1376,
+                    1922
+                ],
+                "score": 0.757
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    841,
+                    2088,
+                    858,
+                    2088,
+                    858,
+                    2112,
+                    841,
+                    2112
+                ],
+                "score": 0.757
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    833,
+                    1721,
+                    1010,
+                    1721,
+                    1010,
+                    1756,
+                    833,
+                    1756
+                ],
+                "score": 0.93,
+                "latex": "\\pi _ { n } \\sigma _ { n } \\pi _ { n } ^ { - 1 } \\equiv \\sigma _ { n }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1206,
+                    1098,
+                    1312,
+                    1098,
+                    1312,
+                    1140,
+                    1206,
+                    1140
+                ],
+                "score": 0.93,
+                "latex": "\\textstyle \\prod _ { i = 1 } ^ { l } n _ { i } !"
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    322,
+                    1783,
+                    1375,
+                    1783,
+                    1375,
+                    1875,
+                    322,
+                    1875
+                ],
+                "score": 0.92,
+                "latex": "\\begin{array} { r l } & { \\mathscr { N } ^ { \\prime } ( I ) = S ^ { \\prime } \\circ \\sigma _ { l } \\circ C _ { l } ^ { \\prime } \\circ \\sigma _ { l - 1 } \\circ \\cdots \\circ \\sigma _ { 1 } \\circ C _ { 1 } ^ { \\prime } ( I ) = } \\\\ & { \\qquad = S \\circ \\pi _ { l } \\circ \\sigma _ { l } \\circ \\pi _ { l } ^ { - 1 } \\circ C _ { l } \\circ \\pi _ { l - 1 } \\circ \\cdots \\circ \\pi _ { 1 } \\circ \\sigma _ { 1 } \\circ \\pi _ { 1 } ^ { - 1 } \\circ C _ { 1 } ( I ) = \\mathscr { N } ( I ) . } \\end{array}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1046,
+                    1691,
+                    1228,
+                    1691,
+                    1228,
+                    1726,
+                    1046,
+                    1726
+                ],
+                "score": 0.92,
+                "latex": "\\mathcal { N } ^ { \\prime } ( I ) \\equiv \\mathcal { N } ( I )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    663,
+                    1512,
+                    865,
+                    1512,
+                    865,
+                    1546,
+                    663,
+                    1546
+                ],
+                "score": 0.92,
+                "latex": "C _ { i } ^ { \\prime } = \\pi _ { i } ^ { - 1 } C _ { i } \\pi _ { i - 1 }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    721,
+                    1693,
+                    868,
+                    1693,
+                    868,
+                    1723,
+                    721,
+                    1723
+                ],
+                "score": 0.91,
+                "latex": "S ^ { \\prime } \\equiv S \\circ \\pi _ { l }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    465,
+                    445,
+                    564,
+                    445,
+                    564,
+                    475,
+                    465,
+                    475
+                ],
+                "score": 0.91,
+                "latex": "n _ { l } = 1 0"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    298,
+                    446,
+                    412,
+                    446,
+                    412,
+                    476,
+                    298,
+                    476
+                ],
+                "score": 0.91,
+                "latex": "n _ { l - 1 } = 5"
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    579,
+                    1629,
+                    1118,
+                    1629,
+                    1118,
+                    1667,
+                    579,
+                    1667
+                ],
+                "score": 0.91,
+                "latex": "\\mathcal { N } ^ { \\prime } ( I ) = S ^ { \\prime } \\circ \\sigma _ { l } \\circ C _ { l } ^ { \\prime } \\circ \\sigma _ { l - 1 } \\circ \\cdots \\circ \\sigma _ { 1 } \\circ C _ { 1 } ^ { \\prime } ( I ) ,"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    575,
+                    1577,
+                    631,
+                    1577,
+                    631,
+                    1606,
+                    575,
+                    1606
+                ],
+                "score": 0.89,
+                "latex": "\\pi _ { i - 1 }"
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    591,
+                    1361,
+                    1107,
+                    1361,
+                    1107,
+                    1399,
+                    591,
+                    1399
+                ],
+                "score": 0.89,
+                "latex": "\\mathcal { N } ( I ) \\equiv S \\circ \\sigma _ { l } \\circ C _ { l } \\circ \\sigma _ { l - 1 } \\circ \\cdots \\circ \\sigma _ { 1 } \\circ C _ { 1 } ( I )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    694,
+                    1455,
+                    751,
+                    1455,
+                    751,
+                    1484,
+                    694,
+                    1484
+                ],
+                "score": 0.88,
+                "latex": "n _ { i - 1 }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    398,
+                    1453,
+                    430,
+                    1453,
+                    430,
+                    1482,
+                    398,
+                    1482
+                ],
+                "score": 0.87,
+                "latex": "C _ { i }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    809,
+                    1198,
+                    840,
+                    1198,
+                    840,
+                    1228,
+                    809,
+                    1228
+                ],
+                "score": 0.87,
+                "latex": "C _ { i }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    461,
+                    1545,
+                    491,
+                    1545,
+                    491,
+                    1573,
+                    461,
+                    1573
+                ],
+                "score": 0.87,
+                "latex": "C _ { i }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    600,
+                    1262,
+                    628,
+                    1262,
+                    628,
+                    1289,
+                    600,
+                    1289
+                ],
+                "score": 0.86,
+                "latex": "\\sigma _ { i }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    877,
+                    1547,
+                    905,
+                    1547,
+                    905,
+                    1573,
+                    877,
+                    1573
+                ],
+                "score": 0.86,
+                "latex": "\\pi _ { i }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1188,
+                    1455,
+                    1217,
+                    1455,
+                    1217,
+                    1482,
+                    1188,
+                    1482
+                ],
+                "score": 0.85,
+                "latex": "n _ { i }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1225,
+                    1487,
+                    1253,
+                    1487,
+                    1253,
+                    1512,
+                    1225,
+                    1512
+                ],
+                "score": 0.85,
+                "latex": "n _ { i }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    887,
+                    1487,
+                    916,
+                    1487,
+                    916,
+                    1512,
+                    887,
+                    1512
+                ],
+                "score": 0.85,
+                "latex": "\\pi _ { i }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    682,
+                    1577,
+                    710,
+                    1577,
+                    710,
+                    1604,
+                    682,
+                    1604
+                ],
+                "score": 0.85,
+                "latex": "\\pi _ { i }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    374,
+                    1698,
+                    406,
+                    1698,
+                    406,
+                    1723,
+                    374,
+                    1723
+                ],
+                "score": 0.85,
+                "latex": "\\pi _ { 0 }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1188,
+                    1203,
+                    1217,
+                    1203,
+                    1217,
+                    1228,
+                    1188,
+                    1228
+                ],
+                "score": 0.85,
+                "latex": "\\sigma _ { i }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1344,
+                    1485,
+                    1373,
+                    1485,
+                    1373,
+                    1512,
+                    1344,
+                    1512
+                ],
+                "score": 0.85,
+                "latex": "\\pi _ { i }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    947,
+                    1229,
+                    969,
+                    1229,
+                    969,
+                    1256,
+                    947,
+                    1256
+                ],
+                "score": 0.81,
+                "latex": "S"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    769,
+                    1488,
+                    796,
+                    1488,
+                    796,
+                    1511,
+                    769,
+                    1511
+                ],
+                "score": 0.79,
+                "latex": "\\pi _ { i }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    893,
+                    1109,
+                    924,
+                    1109,
+                    924,
+                    1135,
+                    893,
+                    1135
+                ],
+                "score": 0.75,
+                "latex": "n _ { 1 }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    417,
+                    1199,
+                    436,
+                    1199,
+                    436,
+                    1225,
+                    417,
+                    1225
+                ],
+                "score": 0.74,
+                "latex": "I"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    653,
+                    1230,
+                    666,
+                    1230,
+                    666,
+                    1255,
+                    653,
+                    1255
+                ],
+                "score": 0.72,
+                "latex": "i"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    460,
+                    1427,
+                    480,
+                    1427,
+                    480,
+                    1448,
+                    460,
+                    1448
+                ],
+                "score": 0.68,
+                "latex": "\\circ"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1030,
+                    1109,
+                    1059,
+                    1109,
+                    1059,
+                    1135,
+                    1030,
+                    1135
+                ],
+                "score": 0.68,
+                "latex": "n _ { l }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    482,
+                    1975,
+                    494,
+                    1975,
+                    494,
+                    2000,
+                    482,
+                    2000
+                ],
+                "score": 0.66,
+                "latex": "l"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    937,
+                    1110,
+                    969,
+                    1110,
+                    969,
+                    1135,
+                    937,
+                    1135
+                ],
+                "score": 0.62,
+                "latex": "n _ { 2 }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    359,
+                    319,
+                    405,
+                    319,
+                    405,
+                    336,
+                    359,
+                    336
+                ],
+                "score": 0.4,
+                "latex": "n _ { l - 1 }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    554,
+                    379,
+                    581,
+                    379,
+                    581,
+                    402,
+                    554,
+                    402
+                ],
+                "score": 0.37,
+                "latex": "n _ { l }"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    414.0,
+                    1404.0,
+                    414.0,
+                    1404.0,
+                    447.0,
+                    295.0,
+                    447.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    443.0,
+                    297.0,
+                    443.0,
+                    297.0,
+                    479.0,
+                    293.0,
+                    479.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    413.0,
+                    443.0,
+                    464.0,
+                    443.0,
+                    464.0,
+                    479.0,
+                    413.0,
+                    479.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    565.0,
+                    443.0,
+                    1405.0,
+                    443.0,
+                    1405.0,
+                    479.0,
+                    565.0,
+                    479.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    473.0,
+                    1405.0,
+                    473.0,
+                    1405.0,
+                    509.0,
+                    293.0,
+                    509.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    504.0,
+                    1408.0,
+                    504.0,
+                    1408.0,
+                    542.0,
+                    291.0,
+                    542.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    539.0,
+                    421.0,
+                    539.0,
+                    421.0,
+                    568.0,
+                    295.0,
+                    568.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    345.0,
+                    281.0,
+                    414.0,
+                    281.0,
+                    414.0,
+                    319.0,
+                    345.0,
+                    319.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    847.0,
+                    289.0,
+                    914.0,
+                    289.0,
+                    914.0,
+                    323.0,
+                    847.0,
+                    323.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    855.0,
+                    312.0,
+                    913.0,
+                    312.0,
+                    913.0,
+                    347.0,
+                    855.0,
+                    347.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    473.0,
+                    369.0,
+                    553.0,
+                    369.0,
+                    553.0,
+                    408.0,
+                    473.0,
+                    408.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    582.0,
+                    369.0,
+                    587.0,
+                    369.0,
+                    587.0,
+                    408.0,
+                    582.0,
+                    408.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    993.0,
+                    374.0,
+                    1103.0,
+                    374.0,
+                    1103.0,
+                    409.0,
+                    993.0,
+                    409.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    350.75,
+                    307.0,
+                    410.75,
+                    307.0,
+                    410.75,
+                    341.0,
+                    350.75,
+                    341.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    692.0,
+                    355.5,
+                    852.0,
+                    355.5,
+                    852.0,
+                    385.0,
+                    692.0,
+                    385.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    747.0,
+                    378.0,
+                    824.0,
+                    378.0,
+                    824.0,
+                    403.5,
+                    747.0,
+                    403.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1189.0,
+                    353.0,
+                    1346.0,
+                    353.0,
+                    1346.0,
+                    384.5,
+                    1189.0,
+                    384.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1243.25,
+                    373.5,
+                    1325.25,
+                    373.5,
+                    1325.25,
+                    401.5,
+                    1243.25,
+                    401.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    780.0,
+                    502.0,
+                    780.0,
+                    502.0,
+                    825.0,
+                    292.0,
+                    825.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    73.0,
+                    858.0,
+                    73.0,
+                    858.0,
+                    108.0,
+                    297.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1386.0,
+                    1907.0,
+                    1394.0,
+                    1907.0,
+                    1394.0,
+                    1915.0,
+                    1386.0,
+                    1915.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    839.0,
+                    2085.0,
+                    861.0,
+                    2085.0,
+                    861.0,
+                    2120.0,
+                    839.0,
+                    2120.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    853.0,
+                    1404.0,
+                    853.0,
+                    1404.0,
+                    885.0,
+                    295.0,
+                    885.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    883.0,
+                    1405.0,
+                    883.0,
+                    1405.0,
+                    917.0,
+                    295.0,
+                    917.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    915.0,
+                    1405.0,
+                    915.0,
+                    1405.0,
+                    947.0,
+                    295.0,
+                    947.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    943.0,
+                    1406.0,
+                    943.0,
+                    1406.0,
+                    980.0,
+                    292.0,
+                    980.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    972.0,
+                    1404.0,
+                    972.0,
+                    1404.0,
+                    1010.0,
+                    292.0,
+                    1010.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1007.0,
+                    1275.0,
+                    1007.0,
+                    1275.0,
+                    1038.0,
+                    295.0,
+                    1038.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    1421.0,
+                    459.0,
+                    1421.0,
+                    459.0,
+                    1453.0,
+                    297.0,
+                    1453.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    481.0,
+                    1421.0,
+                    1403.0,
+                    1421.0,
+                    1403.0,
+                    1453.0,
+                    481.0,
+                    1453.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1452.0,
+                    397.0,
+                    1452.0,
+                    397.0,
+                    1486.0,
+                    294.0,
+                    1486.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    431.0,
+                    1452.0,
+                    693.0,
+                    1452.0,
+                    693.0,
+                    1486.0,
+                    431.0,
+                    1486.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    752.0,
+                    1452.0,
+                    1187.0,
+                    1452.0,
+                    1187.0,
+                    1486.0,
+                    752.0,
+                    1486.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1218.0,
+                    1452.0,
+                    1405.0,
+                    1452.0,
+                    1405.0,
+                    1486.0,
+                    1218.0,
+                    1486.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1481.0,
+                    768.0,
+                    1481.0,
+                    768.0,
+                    1517.0,
+                    295.0,
+                    1517.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    797.0,
+                    1481.0,
+                    886.0,
+                    1481.0,
+                    886.0,
+                    1517.0,
+                    797.0,
+                    1517.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    917.0,
+                    1481.0,
+                    1224.0,
+                    1481.0,
+                    1224.0,
+                    1517.0,
+                    917.0,
+                    1517.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1254.0,
+                    1481.0,
+                    1343.0,
+                    1481.0,
+                    1343.0,
+                    1517.0,
+                    1254.0,
+                    1517.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1374.0,
+                    1481.0,
+                    1406.0,
+                    1481.0,
+                    1406.0,
+                    1517.0,
+                    1374.0,
+                    1517.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1512.0,
+                    662.0,
+                    1512.0,
+                    662.0,
+                    1548.0,
+                    294.0,
+                    1548.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    866.0,
+                    1512.0,
+                    1405.0,
+                    1512.0,
+                    1405.0,
+                    1548.0,
+                    866.0,
+                    1548.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1540.0,
+                    460.0,
+                    1540.0,
+                    460.0,
+                    1577.0,
+                    294.0,
+                    1577.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    492.0,
+                    1540.0,
+                    876.0,
+                    1540.0,
+                    876.0,
+                    1577.0,
+                    492.0,
+                    1577.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    906.0,
+                    1540.0,
+                    1406.0,
+                    1540.0,
+                    1406.0,
+                    1577.0,
+                    906.0,
+                    1577.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1572.0,
+                    574.0,
+                    1572.0,
+                    574.0,
+                    1608.0,
+                    296.0,
+                    1608.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    632.0,
+                    1572.0,
+                    681.0,
+                    1572.0,
+                    681.0,
+                    1608.0,
+                    632.0,
+                    1608.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    711.0,
+                    1572.0,
+                    1202.0,
+                    1572.0,
+                    1202.0,
+                    1608.0,
+                    711.0,
+                    1608.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    624.0,
+                    1402.0,
+                    624.0,
+                    1402.0,
+                    657.0,
+                    296.0,
+                    657.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    656.0,
+                    1404.0,
+                    656.0,
+                    1404.0,
+                    689.0,
+                    294.0,
+                    689.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    685.0,
+                    1404.0,
+                    685.0,
+                    1404.0,
+                    718.0,
+                    294.0,
+                    718.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    717.0,
+                    753.0,
+                    717.0,
+                    753.0,
+                    749.0,
+                    293.0,
+                    749.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1194.0,
+                    416.0,
+                    1194.0,
+                    416.0,
+                    1233.0,
+                    294.0,
+                    1233.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    437.0,
+                    1194.0,
+                    808.0,
+                    1194.0,
+                    808.0,
+                    1233.0,
+                    437.0,
+                    1233.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    841.0,
+                    1194.0,
+                    1187.0,
+                    1194.0,
+                    1187.0,
+                    1233.0,
+                    841.0,
+                    1233.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1218.0,
+                    1194.0,
+                    1404.0,
+                    1194.0,
+                    1404.0,
+                    1233.0,
+                    1218.0,
+                    1233.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1226.0,
+                    652.0,
+                    1226.0,
+                    652.0,
+                    1263.0,
+                    295.0,
+                    1263.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    667.0,
+                    1226.0,
+                    946.0,
+                    1226.0,
+                    946.0,
+                    1263.0,
+                    667.0,
+                    1263.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    970.0,
+                    1226.0,
+                    1405.0,
+                    1226.0,
+                    1405.0,
+                    1263.0,
+                    970.0,
+                    1263.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1257.0,
+                    599.0,
+                    1257.0,
+                    599.0,
+                    1292.0,
+                    295.0,
+                    1292.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    629.0,
+                    1257.0,
+                    1405.0,
+                    1257.0,
+                    1405.0,
+                    1292.0,
+                    629.0,
+                    1292.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1287.0,
+                    1405.0,
+                    1287.0,
+                    1405.0,
+                    1323.0,
+                    292.0,
+                    1323.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1317.0,
+                    423.0,
+                    1317.0,
+                    423.0,
+                    1354.0,
+                    294.0,
+                    1354.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1940.0,
+                    1403.0,
+                    1940.0,
+                    1403.0,
+                    1980.0,
+                    292.0,
+                    1980.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1970.0,
+                    481.0,
+                    1970.0,
+                    481.0,
+                    2009.0,
+                    293.0,
+                    2009.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    495.0,
+                    1970.0,
+                    1406.0,
+                    1970.0,
+                    1406.0,
+                    2009.0,
+                    495.0,
+                    2009.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    2002.0,
+                    916.0,
+                    2002.0,
+                    916.0,
+                    2038.0,
+                    294.0,
+                    2038.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1068.0,
+                    1405.0,
+                    1068.0,
+                    1405.0,
+                    1103.0,
+                    295.0,
+                    1103.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1099.0,
+                    892.0,
+                    1099.0,
+                    892.0,
+                    1142.0,
+                    293.0,
+                    1142.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    925.0,
+                    1099.0,
+                    936.0,
+                    1099.0,
+                    936.0,
+                    1142.0,
+                    925.0,
+                    1142.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    970.0,
+                    1099.0,
+                    1029.0,
+                    1099.0,
+                    1029.0,
+                    1142.0,
+                    970.0,
+                    1142.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1060.0,
+                    1099.0,
+                    1205.0,
+                    1099.0,
+                    1205.0,
+                    1142.0,
+                    1060.0,
+                    1142.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1313.0,
+                    1099.0,
+                    1405.0,
+                    1099.0,
+                    1405.0,
+                    1142.0,
+                    1313.0,
+                    1142.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1136.0,
+                    883.0,
+                    1136.0,
+                    883.0,
+                    1171.0,
+                    296.0,
+                    1171.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1691.0,
+                    373.0,
+                    1691.0,
+                    373.0,
+                    1727.0,
+                    296.0,
+                    1727.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    407.0,
+                    1691.0,
+                    720.0,
+                    1691.0,
+                    720.0,
+                    1727.0,
+                    407.0,
+                    1727.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    869.0,
+                    1691.0,
+                    1045.0,
+                    1691.0,
+                    1045.0,
+                    1727.0,
+                    869.0,
+                    1727.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1229.0,
+                    1691.0,
+                    1406.0,
+                    1691.0,
+                    1406.0,
+                    1727.0,
+                    1229.0,
+                    1727.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    290.0,
+                    1718.0,
+                    832.0,
+                    1718.0,
+                    832.0,
+                    1762.0,
+                    290.0,
+                    1762.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1011.0,
+                    1718.0,
+                    1128.0,
+                    1718.0,
+                    1128.0,
+                    1762.0,
+                    1011.0,
+                    1762.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1890.0,
+                    1153.0,
+                    1890.0,
+                    1153.0,
+                    1933.0,
+                    296.0,
+                    1933.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 4,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    1534,
+                    1404,
+                    1534,
+                    1404,
+                    1749,
+                    299,
+                    1749
+                ],
+                "score": 0.98
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1303,
+                    1404,
+                    1303,
+                    1404,
+                    1518,
+                    298,
+                    1518
+                ],
+                "score": 0.978
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1850,
+                    1404,
+                    1850,
+                    1404,
+                    2033,
+                    298,
+                    2033
+                ],
+                "score": 0.978
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    855,
+                    1403,
+                    855,
+                    1403,
+                    977,
+                    299,
+                    977
+                ],
+                "score": 0.974
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    300,
+                    1016,
+                    1402,
+                    1016,
+                    1402,
+                    1138,
+                    300,
+                    1138
+                ],
+                "score": 0.972
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    296,
+                    1403,
+                    296,
+                    1403,
+                    389,
+                    299,
+                    389
+                ],
+                "score": 0.971
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    302,
+                    590,
+                    1401,
+                    590,
+                    1401,
+                    683,
+                    302,
+                    683
+                ],
+                "score": 0.965
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    300,
+                    489,
+                    1402,
+                    489,
+                    1402,
+                    552,
+                    300,
+                    552
+                ],
+                "score": 0.93
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    299,
+                    226,
+                    558,
+                    226,
+                    558,
+                    262,
+                    299,
+                    262
+                ],
+                "score": 0.906
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    298,
+                    75,
+                    856,
+                    75,
+                    856,
+                    105,
+                    298,
+                    105
+                ],
+                "score": 0.894
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    300,
+                    1786,
+                    743,
+                    1786,
+                    743,
+                    1818,
+                    300,
+                    1818
+                ],
+                "score": 0.887
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    299,
+                    426,
+                    445,
+                    426,
+                    445,
+                    458,
+                    299,
+                    458
+                ],
+                "score": 0.879
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    301,
+                    1239,
+                    678,
+                    1239,
+                    678,
+                    1273,
+                    301,
+                    1273
+                ],
+                "score": 0.862
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    301,
+                    722,
+                    1399,
+                    722,
+                    1399,
+                    816,
+                    301,
+                    816
+                ],
+                "score": 0.79
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    840,
+                    2089,
+                    858,
+                    2089,
+                    858,
+                    2112,
+                    840,
+                    2112
+                ],
+                "score": 0.76
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    300,
+                    1181,
+                    1150,
+                    1181,
+                    1150,
+                    1209,
+                    300,
+                    1209
+                ],
+                "score": 0.482
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    300,
+                    1181,
+                    1150,
+                    1181,
+                    1150,
+                    1209,
+                    300,
+                    1209
+                ],
+                "score": 0.444
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    562,
+                    1457,
+                    605,
+                    1457,
+                    605,
+                    1491,
+                    562,
+                    1491
+                ],
+                "score": 0.93,
+                "latex": "\\sqrt { \\alpha }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    687,
+                    1396,
+                    731,
+                    1396,
+                    731,
+                    1430,
+                    687,
+                    1430
+                ],
+                "score": 0.91,
+                "latex": "\\sqrt { \\alpha }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    696,
+                    653,
+                    751,
+                    653,
+                    751,
+                    682,
+                    696,
+                    682
+                ],
+                "score": 0.88,
+                "latex": "87 \\%"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1354,
+                    1657,
+                    1393,
+                    1657,
+                    1393,
+                    1686,
+                    1354,
+                    1686
+                ],
+                "score": 0.86,
+                "latex": "1 \\%"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    761,
+                    623,
+                    836,
+                    623,
+                    836,
+                    652,
+                    761,
+                    652
+                ],
+                "score": 0.86,
+                "latex": "9 9 . 5 \\%"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1333,
+                    1370,
+                    1355,
+                    1370,
+                    1355,
+                    1393,
+                    1333,
+                    1393
+                ],
+                "score": 0.78,
+                "latex": "\\alpha"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1350,
+                    1402,
+                    1371,
+                    1402,
+                    1371,
+                    1424,
+                    1350,
+                    1424
+                ],
+                "score": 0.75,
+                "latex": "\\alpha"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    803,
+                    1370,
+                    824,
+                    1370,
+                    824,
+                    1393,
+                    803,
+                    1393
+                ],
+                "score": 0.69,
+                "latex": "\\alpha"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    222.0,
+                    562.0,
+                    222.0,
+                    562.0,
+                    267.0,
+                    291.0,
+                    267.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    298.0,
+                    73.0,
+                    858.0,
+                    73.0,
+                    858.0,
+                    108.0,
+                    298.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1783.0,
+                    749.0,
+                    1783.0,
+                    749.0,
+                    1823.0,
+                    294.0,
+                    1823.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    424.0,
+                    449.0,
+                    424.0,
+                    449.0,
+                    462.0,
+                    294.0,
+                    462.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1239.0,
+                    680.0,
+                    1239.0,
+                    680.0,
+                    1276.0,
+                    296.0,
+                    1276.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    840.0,
+                    2087.0,
+                    861.0,
+                    2087.0,
+                    861.0,
+                    2118.0,
+                    840.0,
+                    2118.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1178.0,
+                    1157.0,
+                    1178.0,
+                    1157.0,
+                    1215.0,
+                    294.0,
+                    1215.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    1534.0,
+                    1403.0,
+                    1534.0,
+                    1403.0,
+                    1569.0,
+                    297.0,
+                    1569.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1565.0,
+                    1406.0,
+                    1565.0,
+                    1406.0,
+                    1600.0,
+                    294.0,
+                    1600.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1596.0,
+                    1405.0,
+                    1596.0,
+                    1405.0,
+                    1630.0,
+                    294.0,
+                    1630.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1626.0,
+                    1405.0,
+                    1626.0,
+                    1405.0,
+                    1660.0,
+                    295.0,
+                    1660.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1656.0,
+                    1353.0,
+                    1656.0,
+                    1353.0,
+                    1691.0,
+                    295.0,
+                    1691.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1394.0,
+                    1656.0,
+                    1404.0,
+                    1656.0,
+                    1404.0,
+                    1691.0,
+                    1394.0,
+                    1691.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    1687.0,
+                    1405.0,
+                    1687.0,
+                    1405.0,
+                    1721.0,
+                    297.0,
+                    1721.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1718.0,
+                    1209.0,
+                    1718.0,
+                    1209.0,
+                    1751.0,
+                    294.0,
+                    1751.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1305.0,
+                    1405.0,
+                    1305.0,
+                    1405.0,
+                    1339.0,
+                    296.0,
+                    1339.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1334.0,
+                    1405.0,
+                    1334.0,
+                    1405.0,
+                    1369.0,
+                    296.0,
+                    1369.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1366.0,
+                    802.0,
+                    1366.0,
+                    802.0,
+                    1398.0,
+                    293.0,
+                    1398.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    825.0,
+                    1366.0,
+                    1332.0,
+                    1366.0,
+                    1332.0,
+                    1398.0,
+                    825.0,
+                    1398.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1356.0,
+                    1366.0,
+                    1406.0,
+                    1366.0,
+                    1406.0,
+                    1398.0,
+                    1356.0,
+                    1398.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1397.0,
+                    686.0,
+                    1397.0,
+                    686.0,
+                    1431.0,
+                    294.0,
+                    1431.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    732.0,
+                    1397.0,
+                    1349.0,
+                    1397.0,
+                    1349.0,
+                    1431.0,
+                    732.0,
+                    1431.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1372.0,
+                    1397.0,
+                    1405.0,
+                    1397.0,
+                    1405.0,
+                    1431.0,
+                    1372.0,
+                    1431.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1426.0,
+                    1405.0,
+                    1426.0,
+                    1405.0,
+                    1461.0,
+                    294.0,
+                    1461.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1455.0,
+                    561.0,
+                    1455.0,
+                    561.0,
+                    1492.0,
+                    293.0,
+                    1492.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    606.0,
+                    1455.0,
+                    1406.0,
+                    1455.0,
+                    1406.0,
+                    1492.0,
+                    606.0,
+                    1492.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1487.0,
+                    1304.0,
+                    1487.0,
+                    1304.0,
+                    1521.0,
+                    294.0,
+                    1521.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1849.0,
+                    1404.0,
+                    1849.0,
+                    1404.0,
+                    1885.0,
+                    296.0,
+                    1885.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1881.0,
+                    1406.0,
+                    1881.0,
+                    1406.0,
+                    1918.0,
+                    292.0,
+                    1918.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1910.0,
+                    1404.0,
+                    1910.0,
+                    1404.0,
+                    1948.0,
+                    293.0,
+                    1948.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1944.0,
+                    1402.0,
+                    1944.0,
+                    1402.0,
+                    1976.0,
+                    296.0,
+                    1976.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1972.0,
+                    1406.0,
+                    1972.0,
+                    1406.0,
+                    2007.0,
+                    294.0,
+                    2007.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    2005.0,
+                    651.0,
+                    2005.0,
+                    651.0,
+                    2036.0,
+                    296.0,
+                    2036.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    856.0,
+                    1404.0,
+                    856.0,
+                    1404.0,
+                    888.0,
+                    295.0,
+                    888.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    885.0,
+                    1405.0,
+                    885.0,
+                    1405.0,
+                    919.0,
+                    294.0,
+                    919.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    916.0,
+                    1404.0,
+                    916.0,
+                    1404.0,
+                    952.0,
+                    294.0,
+                    952.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    947.0,
+                    1101.0,
+                    947.0,
+                    1101.0,
+                    980.0,
+                    294.0,
+                    980.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1017.0,
+                    1403.0,
+                    1017.0,
+                    1403.0,
+                    1049.0,
+                    296.0,
+                    1049.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1046.0,
+                    1407.0,
+                    1046.0,
+                    1407.0,
+                    1082.0,
+                    293.0,
+                    1082.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1077.0,
+                    1406.0,
+                    1077.0,
+                    1406.0,
+                    1113.0,
+                    294.0,
+                    1113.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1108.0,
+                    674.0,
+                    1108.0,
+                    674.0,
+                    1141.0,
+                    296.0,
+                    1141.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    293.0,
+                    1405.0,
+                    293.0,
+                    1405.0,
+                    333.0,
+                    294.0,
+                    333.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    324.0,
+                    1404.0,
+                    324.0,
+                    1404.0,
+                    364.0,
+                    293.0,
+                    364.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    356.0,
+                    1398.0,
+                    356.0,
+                    1398.0,
+                    394.0,
+                    292.0,
+                    394.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    587.0,
+                    1406.0,
+                    587.0,
+                    1406.0,
+                    627.0,
+                    293.0,
+                    627.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    624.0,
+                    760.0,
+                    624.0,
+                    760.0,
+                    655.0,
+                    297.0,
+                    655.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    837.0,
+                    624.0,
+                    1404.0,
+                    624.0,
+                    1404.0,
+                    655.0,
+                    837.0,
+                    655.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    652.0,
+                    695.0,
+                    652.0,
+                    695.0,
+                    690.0,
+                    294.0,
+                    690.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    752.0,
+                    652.0,
+                    864.0,
+                    652.0,
+                    864.0,
+                    690.0,
+                    752.0,
+                    690.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    487.0,
+                    1404.0,
+                    487.0,
+                    1404.0,
+                    526.0,
+                    294.0,
+                    526.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    521.0,
+                    1178.0,
+                    521.0,
+                    1178.0,
+                    555.0,
+                    295.0,
+                    555.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    722.0,
+                    1403.0,
+                    722.0,
+                    1403.0,
+                    759.0,
+                    295.0,
+                    759.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    751.0,
+                    1405.0,
+                    751.0,
+                    1405.0,
+                    790.0,
+                    293.0,
+                    790.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    781.0,
+                    872.0,
+                    781.0,
+                    872.0,
+                    820.0,
+                    295.0,
+                    820.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1178.0,
+                    1157.0,
+                    1178.0,
+                    1157.0,
+                    1215.0,
+                    294.0,
+                    1215.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 5,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 5,
+                "poly": [
+                    911,
+                    931,
+                    1387,
+                    931,
+                    1387,
+                    1154,
+                    911,
+                    1154
+                ],
+                "score": 0.974,
+                "html": "<table><tr><td>Multiplier</td><td>MAdds</td><td>Params P@1</td></tr><tr><td>0.05</td><td>55.0M 56k</td><td>24.6</td></tr><tr><td>0.10</td><td>75.0M</td><td>170k 38.6</td></tr><tr><td>0.20</td><td>183M 718k</td><td>54.2</td></tr><tr><td>0.30</td><td>439M 1.8M</td><td>64.0</td></tr><tr><td>0.50</td><td>1.40B 5.4M</td><td>72.3</td></tr><tr><td>0.80</td><td>3.40B 13M</td><td>75.6</td></tr></table>"
+            },
+            {
+                "category_id": 3,
+                "poly": [
+                    349,
+                    223,
+                    1335,
+                    223,
+                    1335,
+                    598,
+                    349,
+                    598
+                ],
+                "score": 0.969
+            },
+            {
+                "category_id": 5,
+                "poly": [
+                    367,
+                    902,
+                    829,
+                    902,
+                    829,
+                    1185,
+                    367,
+                    1185
+                ],
+                "score": 0.969,
+                "html": "<table><tr><td>Sparsity</td><td>MAdds Params</td><td>P@1</td></tr><tr><td>0.50/0.01 0.003</td><td>43.0M 90k 158k</td><td>40.3 46.1</td></tr><tr><td>0.01</td><td>82.0M 104M</td><td>52.3</td></tr><tr><td>0.03</td><td>287k 208M 724k</td><td></td></tr><tr><td>0.10</td><td></td><td>59.5</td></tr><tr><td>0.30</td><td>628M 2.3 M</td><td>67.2</td></tr><tr><td></td><td>1.80 B 6.6M</td><td>73</td></tr><tr><td>0.60</td><td>3.50B 13M 22 M</td><td>75</td></tr><tr><td>1.00</td><td>5.70 B</td><td>77</td></tr></table>"
+            },
+            {
+                "category_id": 3,
+                "poly": [
+                    380,
+                    1618,
+                    1307,
+                    1618,
+                    1307,
+                    1974,
+                    380,
+                    1974
+                ],
+                "score": 0.963
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    296,
+                    1372,
+                    1405,
+                    1372,
+                    1405,
+                    1558,
+                    296,
+                    1558
+                ],
+                "score": 0.96
+            },
+            {
+                "category_id": 4,
+                "poly": [
+                    296,
+                    603,
+                    1406,
+                    603,
+                    1406,
+                    725,
+                    296,
+                    725
+                ],
+                "score": 0.96
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    300,
+                    751,
+                    1404,
+                    751,
+                    1404,
+                    846,
+                    300,
+                    846
+                ],
+                "score": 0.942
+            },
+            {
+                "category_id": 4,
+                "poly": [
+                    295,
+                    1978,
+                    1408,
+                    1978,
+                    1408,
+                    2040,
+                    295,
+                    2040
+                ],
+                "score": 0.935
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    298,
+                    1311,
+                    677,
+                    1311,
+                    677,
+                    1343,
+                    298,
+                    1343
+                ],
+                "score": 0.915
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    298,
+                    76,
+                    857,
+                    76,
+                    857,
+                    104,
+                    298,
+                    104
+                ],
+                "score": 0.85
+            },
+            {
+                "category_id": 6,
+                "poly": [
+                    414,
+                    873,
+                    787,
+                    873,
+                    787,
+                    902,
+                    414,
+                    902
+                ],
+                "score": 0.847
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    841,
+                    2087,
+                    858,
+                    2087,
+                    858,
+                    2111,
+                    841,
+                    2111
+                ],
+                "score": 0.71
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    331,
+                    1209,
+                    869,
+                    1209,
+                    869,
+                    1243,
+                    331,
+                    1243
+                ],
+                "score": 0.682
+            },
+            {
+                "category_id": 6,
+                "poly": [
+                    979,
+                    872,
+                    1321,
+                    872,
+                    1321,
+                    904,
+                    979,
+                    904
+                ],
+                "score": 0.651
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    979,
+                    872,
+                    1321,
+                    872,
+                    1321,
+                    904,
+                    979,
+                    904
+                ],
+                "score": 0.356
+            },
+            {
+                "category_id": 7,
+                "poly": [
+                    331,
+                    1209,
+                    869,
+                    1209,
+                    869,
+                    1243,
+                    331,
+                    1243
+                ],
+                "score": 0.244
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    841,
+                    2087,
+                    859,
+                    2087,
+                    859,
+                    2111,
+                    841,
+                    2111
+                ],
+                "score": 0.169
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    299,
+                    76,
+                    854,
+                    76,
+                    854,
+                    104,
+                    299,
+                    104
+                ],
+                "score": 0.137
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    946,
+                    1435,
+                    1013,
+                    1435,
+                    1013,
+                    1464,
+                    946,
+                    1464
+                ],
+                "score": 0.91,
+                "latex": "3 \\times 3"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    548,
+                    1436,
+                    615,
+                    1436,
+                    615,
+                    1464,
+                    548,
+                    1464
+                ],
+                "score": 0.88,
+                "latex": "2 \\times 2"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    379.0,
+                    235.0,
+                    422.0,
+                    235.0,
+                    422.0,
+                    257.0,
+                    379.0,
+                    257.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    442.0,
+                    227.0,
+                    807.0,
+                    227.0,
+                    807.0,
+                    248.0,
+                    442.0,
+                    248.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    873.0,
+                    235.0,
+                    917.0,
+                    235.0,
+                    917.0,
+                    257.0,
+                    873.0,
+                    257.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    930.0,
+                    227.0,
+                    1305.0,
+                    227.0,
+                    1305.0,
+                    248.0,
+                    930.0,
+                    248.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    423.0,
+                    252.0,
+                    604.0,
+                    252.0,
+                    604.0,
+                    281.0,
+                    423.0,
+                    281.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    919.0,
+                    253.0,
+                    1098.0,
+                    253.0,
+                    1098.0,
+                    278.0,
+                    919.0,
+                    278.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    423.0,
+                    269.0,
+                    600.0,
+                    269.0,
+                    600.0,
+                    297.0,
+                    423.0,
+                    297.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    919.0,
+                    271.0,
+                    1093.0,
+                    271.0,
+                    1093.0,
+                    295.0,
+                    919.0,
+                    295.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    384.0,
+                    286.0,
+                    421.0,
+                    286.0,
+                    421.0,
+                    310.0,
+                    384.0,
+                    310.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    757.0,
+                    296.0,
+                    771.0,
+                    296.0,
+                    771.0,
+                    310.0,
+                    757.0,
+                    310.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    794.0,
+                    299.0,
+                    806.0,
+                    299.0,
+                    806.0,
+                    309.0,
+                    794.0,
+                    309.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    879.0,
+                    288.0,
+                    914.0,
+                    288.0,
+                    914.0,
+                    309.0,
+                    879.0,
+                    309.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1229.0,
+                    299.0,
+                    1238.0,
+                    299.0,
+                    1238.0,
+                    307.0,
+                    1229.0,
+                    307.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    348.0,
+                    324.0,
+                    387.0,
+                    324.0,
+                    387.0,
+                    485.0,
+                    348.0,
+                    485.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    842.0,
+                    324.0,
+                    882.0,
+                    324.0,
+                    882.0,
+                    485.0,
+                    842.0,
+                    485.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    625.0,
+                    325.0,
+                    635.0,
+                    325.0,
+                    635.0,
+                    333.0,
+                    625.0,
+                    333.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    386.0,
+                    340.0,
+                    420.0,
+                    340.0,
+                    420.0,
+                    360.0,
+                    386.0,
+                    360.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    879.0,
+                    340.0,
+                    914.0,
+                    340.0,
+                    914.0,
+                    360.0,
+                    879.0,
+                    360.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    556.0,
+                    365.0,
+                    566.0,
+                    365.0,
+                    566.0,
+                    376.0,
+                    556.0,
+                    376.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    386.0,
+                    392.0,
+                    420.0,
+                    392.0,
+                    420.0,
+                    412.0,
+                    386.0,
+                    412.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    879.0,
+                    392.0,
+                    913.0,
+                    392.0,
+                    913.0,
+                    412.0,
+                    879.0,
+                    412.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    386.0,
+                    444.0,
+                    420.0,
+                    444.0,
+                    420.0,
+                    464.0,
+                    386.0,
+                    464.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    879.0,
+                    445.0,
+                    914.0,
+                    445.0,
+                    914.0,
+                    464.0,
+                    879.0,
+                    464.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    946.0,
+                    456.0,
+                    955.0,
+                    456.0,
+                    955.0,
+                    466.0,
+                    946.0,
+                    466.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    385.0,
+                    494.0,
+                    421.0,
+                    494.0,
+                    421.0,
+                    518.0,
+                    385.0,
+                    518.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    484.0,
+                    484.0,
+                    497.0,
+                    484.0,
+                    497.0,
+                    511.0,
+                    484.0,
+                    511.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    878.0,
+                    494.0,
+                    916.0,
+                    494.0,
+                    916.0,
+                    518.0,
+                    878.0,
+                    518.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    382.0,
+                    541.0,
+                    439.0,
+                    541.0,
+                    439.0,
+                    584.0,
+                    382.0,
+                    584.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    462.0,
+                    558.0,
+                    489.0,
+                    558.0,
+                    489.0,
+                    578.0,
+                    462.0,
+                    578.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    521.0,
+                    558.0,
+                    554.0,
+                    558.0,
+                    554.0,
+                    577.0,
+                    521.0,
+                    577.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    576.0,
+                    558.0,
+                    611.0,
+                    558.0,
+                    611.0,
+                    577.0,
+                    576.0,
+                    577.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    636.0,
+                    558.0,
+                    677.0,
+                    558.0,
+                    677.0,
+                    577.0,
+                    636.0,
+                    577.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    692.0,
+                    558.0,
+                    732.0,
+                    558.0,
+                    732.0,
+                    577.0,
+                    692.0,
+                    577.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    758.0,
+                    558.0,
+                    789.0,
+                    558.0,
+                    789.0,
+                    578.0,
+                    758.0,
+                    578.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    812.0,
+                    558.0,
+                    844.0,
+                    558.0,
+                    844.0,
+                    576.0,
+                    812.0,
+                    576.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    874.0,
+                    540.0,
+                    938.0,
+                    540.0,
+                    938.0,
+                    585.0,
+                    874.0,
+                    585.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    964.0,
+                    558.0,
+                    1004.0,
+                    558.0,
+                    1004.0,
+                    578.0,
+                    964.0,
+                    578.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1030.0,
+                    558.0,
+                    1069.0,
+                    558.0,
+                    1069.0,
+                    577.0,
+                    1030.0,
+                    577.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1108.0,
+                    559.0,
+                    1135.0,
+                    559.0,
+                    1135.0,
+                    575.0,
+                    1108.0,
+                    575.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1172.0,
+                    559.0,
+                    1199.0,
+                    559.0,
+                    1199.0,
+                    576.0,
+                    1172.0,
+                    576.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1240.0,
+                    558.0,
+                    1276.0,
+                    558.0,
+                    1276.0,
+                    577.0,
+                    1240.0,
+                    577.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1306.0,
+                    557.0,
+                    1341.0,
+                    557.0,
+                    1341.0,
+                    577.0,
+                    1306.0,
+                    577.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    545.0,
+                    570.0,
+                    703.0,
+                    570.0,
+                    703.0,
+                    602.0,
+                    545.0,
+                    602.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1033.0,
+                    571.0,
+                    1203.0,
+                    571.0,
+                    1203.0,
+                    601.0,
+                    1033.0,
+                    601.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    927.0,
+                    482.5,
+                    930.0,
+                    482.5,
+                    930.0,
+                    506.0,
+                    927.0,
+                    506.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    441.0,
+                    1620.0,
+                    822.0,
+                    1620.0,
+                    822.0,
+                    1640.0,
+                    441.0,
+                    1640.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    906.0,
+                    1620.0,
+                    1296.0,
+                    1620.0,
+                    1296.0,
+                    1640.0,
+                    906.0,
+                    1640.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    443.0,
+                    1645.0,
+                    615.0,
+                    1645.0,
+                    615.0,
+                    1669.0,
+                    443.0,
+                    1669.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    910.0,
+                    1644.0,
+                    1083.0,
+                    1644.0,
+                    1083.0,
+                    1672.0,
+                    910.0,
+                    1672.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    405.0,
+                    1666.0,
+                    438.0,
+                    1666.0,
+                    438.0,
+                    1685.0,
+                    405.0,
+                    1685.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    443.0,
+                    1661.0,
+                    610.0,
+                    1661.0,
+                    610.0,
+                    1686.0,
+                    443.0,
+                    1686.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    873.0,
+                    1666.0,
+                    909.0,
+                    1666.0,
+                    909.0,
+                    1685.0,
+                    873.0,
+                    1685.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    910.0,
+                    1659.0,
+                    1080.0,
+                    1659.0,
+                    1080.0,
+                    1687.0,
+                    910.0,
+                    1687.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1249.0,
+                    1667.0,
+                    1260.0,
+                    1667.0,
+                    1260.0,
+                    1677.0,
+                    1249.0,
+                    1677.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    674.0,
+                    1680.0,
+                    685.0,
+                    1680.0,
+                    685.0,
+                    1690.0,
+                    674.0,
+                    1690.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1214.0,
+                    1680.0,
+                    1222.0,
+                    1680.0,
+                    1222.0,
+                    1689.0,
+                    1214.0,
+                    1689.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    404.0,
+                    1701.0,
+                    438.0,
+                    1701.0,
+                    438.0,
+                    1724.0,
+                    404.0,
+                    1724.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    873.0,
+                    1701.0,
+                    909.0,
+                    1701.0,
+                    909.0,
+                    1724.0,
+                    873.0,
+                    1724.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    839.0,
+                    1708.0,
+                    888.0,
+                    1708.0,
+                    888.0,
+                    1869.0,
+                    839.0,
+                    1869.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    373.0,
+                    1709.0,
+                    418.0,
+                    1709.0,
+                    418.0,
+                    1868.0,
+                    373.0,
+                    1868.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    872.0,
+                    1737.0,
+                    907.0,
+                    1737.0,
+                    907.0,
+                    1760.0,
+                    872.0,
+                    1760.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    406.0,
+                    1738.0,
+                    436.0,
+                    1738.0,
+                    436.0,
+                    1759.0,
+                    406.0,
+                    1759.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    406.0,
+                    1777.0,
+                    437.0,
+                    1777.0,
+                    437.0,
+                    1797.0,
+                    406.0,
+                    1797.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    873.0,
+                    1775.0,
+                    908.0,
+                    1775.0,
+                    908.0,
+                    1798.0,
+                    873.0,
+                    1798.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    406.0,
+                    1815.0,
+                    437.0,
+                    1815.0,
+                    437.0,
+                    1834.0,
+                    406.0,
+                    1834.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    875.0,
+                    1815.0,
+                    907.0,
+                    1815.0,
+                    907.0,
+                    1834.0,
+                    875.0,
+                    1834.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1003.0,
+                    1824.0,
+                    1011.0,
+                    1824.0,
+                    1011.0,
+                    1832.0,
+                    1003.0,
+                    1832.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    406.0,
+                    1851.0,
+                    437.0,
+                    1851.0,
+                    437.0,
+                    1872.0,
+                    406.0,
+                    1872.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    875.0,
+                    1851.0,
+                    907.0,
+                    1851.0,
+                    907.0,
+                    1871.0,
+                    875.0,
+                    1871.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    495.0,
+                    1879.0,
+                    502.0,
+                    1879.0,
+                    502.0,
+                    1888.0,
+                    495.0,
+                    1888.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    962.0,
+                    1884.0,
+                    968.0,
+                    1884.0,
+                    968.0,
+                    1889.0,
+                    962.0,
+                    1889.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    404.0,
+                    1887.0,
+                    439.0,
+                    1887.0,
+                    439.0,
+                    1910.0,
+                    404.0,
+                    1910.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    487.0,
+                    1889.0,
+                    497.0,
+                    1889.0,
+                    497.0,
+                    1900.0,
+                    487.0,
+                    1900.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    873.0,
+                    1887.0,
+                    909.0,
+                    1887.0,
+                    909.0,
+                    1910.0,
+                    873.0,
+                    1910.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    953.0,
+                    1891.0,
+                    962.0,
+                    1891.0,
+                    962.0,
+                    1901.0,
+                    953.0,
+                    1901.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    400.0,
+                    1920.0,
+                    463.0,
+                    1920.0,
+                    463.0,
+                    1959.0,
+                    400.0,
+                    1959.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    543.0,
+                    1933.0,
+                    584.0,
+                    1933.0,
+                    584.0,
+                    1955.0,
+                    543.0,
+                    1955.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    686.0,
+                    1934.0,
+                    716.0,
+                    1934.0,
+                    716.0,
+                    1955.0,
+                    686.0,
+                    1955.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    812.0,
+                    1934.0,
+                    842.0,
+                    1934.0,
+                    842.0,
+                    1955.0,
+                    812.0,
+                    1955.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    868.0,
+                    1920.0,
+                    931.0,
+                    1920.0,
+                    931.0,
+                    1959.0,
+                    868.0,
+                    1959.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    994.0,
+                    1936.0,
+                    1021.0,
+                    1936.0,
+                    1021.0,
+                    1953.0,
+                    994.0,
+                    1953.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1086.0,
+                    1935.0,
+                    1116.0,
+                    1935.0,
+                    1116.0,
+                    1954.0,
+                    1086.0,
+                    1954.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1186.0,
+                    1934.0,
+                    1221.0,
+                    1934.0,
+                    1221.0,
+                    1954.0,
+                    1186.0,
+                    1954.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1279.0,
+                    1934.0,
+                    1313.0,
+                    1934.0,
+                    1313.0,
+                    1954.0,
+                    1279.0,
+                    1954.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    556.0,
+                    1944.0,
+                    710.0,
+                    1944.0,
+                    710.0,
+                    1979.0,
+                    556.0,
+                    1979.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1019.0,
+                    1946.0,
+                    1184.0,
+                    1946.0,
+                    1184.0,
+                    1978.0,
+                    1019.0,
+                    1978.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    696.0,
+                    1667.0,
+                    717.0,
+                    1667.0,
+                    717.0,
+                    1676.5,
+                    696.0,
+                    1676.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    527.0,
+                    1780.5,
+                    571.0,
+                    1780.5,
+                    571.0,
+                    1791.0,
+                    527.0,
+                    1791.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    501.0,
+                    1825.5,
+                    543.0,
+                    1825.5,
+                    543.0,
+                    1834.5,
+                    501.0,
+                    1834.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    602.0,
+                    1408.0,
+                    602.0,
+                    1408.0,
+                    637.0,
+                    295.0,
+                    637.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    632.0,
+                    1406.0,
+                    632.0,
+                    1406.0,
+                    667.0,
+                    294.0,
+                    667.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    665.0,
+                    1404.0,
+                    665.0,
+                    1404.0,
+                    698.0,
+                    295.0,
+                    698.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    290.0,
+                    694.0,
+                    390.0,
+                    694.0,
+                    390.0,
+                    730.0,
+                    290.0,
+                    730.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1975.0,
+                    1405.0,
+                    1975.0,
+                    1405.0,
+                    2010.0,
+                    296.0,
+                    2010.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    2006.0,
+                    1148.0,
+                    2006.0,
+                    1148.0,
+                    2042.0,
+                    294.0,
+                    2042.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1306.0,
+                    680.0,
+                    1306.0,
+                    680.0,
+                    1348.0,
+                    294.0,
+                    1348.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    298.0,
+                    73.0,
+                    857.0,
+                    73.0,
+                    857.0,
+                    108.0,
+                    298.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    410.0,
+                    869.0,
+                    788.0,
+                    869.0,
+                    788.0,
+                    908.0,
+                    410.0,
+                    908.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    839.0,
+                    2086.0,
+                    860.0,
+                    2086.0,
+                    860.0,
+                    2117.0,
+                    839.0,
+                    2117.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    976.0,
+                    868.0,
+                    1325.0,
+                    868.0,
+                    1325.0,
+                    911.0,
+                    976.0,
+                    911.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    327.0,
+                    1203.0,
+                    870.0,
+                    1203.0,
+                    870.0,
+                    1249.0,
+                    327.0,
+                    1249.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    839.0,
+                    2086.0,
+                    860.0,
+                    2086.0,
+                    860.0,
+                    2118.0,
+                    839.0,
+                    2118.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    73.0,
+                    858.0,
+                    73.0,
+                    858.0,
+                    108.0,
+                    297.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1374.0,
+                    1403.0,
+                    1374.0,
+                    1403.0,
+                    1406.0,
+                    295.0,
+                    1406.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1402.0,
+                    1405.0,
+                    1402.0,
+                    1405.0,
+                    1438.0,
+                    294.0,
+                    1438.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1433.0,
+                    547.0,
+                    1433.0,
+                    547.0,
+                    1469.0,
+                    293.0,
+                    1469.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    616.0,
+                    1433.0,
+                    945.0,
+                    1433.0,
+                    945.0,
+                    1469.0,
+                    616.0,
+                    1469.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1014.0,
+                    1433.0,
+                    1405.0,
+                    1433.0,
+                    1405.0,
+                    1469.0,
+                    1014.0,
+                    1469.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1462.0,
+                    1406.0,
+                    1462.0,
+                    1406.0,
+                    1501.0,
+                    293.0,
+                    1501.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1496.0,
+                    1406.0,
+                    1496.0,
+                    1406.0,
+                    1529.0,
+                    293.0,
+                    1529.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1528.0,
+                    1038.0,
+                    1528.0,
+                    1038.0,
+                    1559.0,
+                    296.0,
+                    1559.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    751.0,
+                    1405.0,
+                    751.0,
+                    1405.0,
+                    789.0,
+                    294.0,
+                    789.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    784.0,
+                    1403.0,
+                    784.0,
+                    1403.0,
+                    818.0,
+                    295.0,
+                    818.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    811.0,
+                    1183.0,
+                    811.0,
+                    1183.0,
+                    851.0,
+                    294.0,
+                    851.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    327.0,
+                    1203.0,
+                    870.0,
+                    1203.0,
+                    870.0,
+                    1249.0,
+                    327.0,
+                    1249.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    976.0,
+                    868.0,
+                    1325.0,
+                    868.0,
+                    1325.0,
+                    911.0,
+                    976.0,
+                    911.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 6,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 3,
+                "poly": [
+                    350,
+                    739,
+                    1337,
+                    739,
+                    1337,
+                    1118,
+                    350,
+                    1118
+                ],
+                "score": 0.968
+            },
+            {
+                "category_id": 3,
+                "poly": [
+                    349,
+                    226,
+                    1339,
+                    226,
+                    1339,
+                    603,
+                    349,
+                    603
+                ],
+                "score": 0.964
+            },
+            {
+                "category_id": 4,
+                "poly": [
+                    296,
+                    1123,
+                    1406,
+                    1123,
+                    1406,
+                    1275,
+                    296,
+                    1275
+                ],
+                "score": 0.96
+            },
+            {
+                "category_id": 3,
+                "poly": [
+                    528,
+                    1448,
+                    1152,
+                    1448,
+                    1152,
+                    1939,
+                    528,
+                    1939
+                ],
+                "score": 0.959
+            },
+            {
+                "category_id": 4,
+                "poly": [
+                    294,
+                    607,
+                    1405,
+                    607,
+                    1405,
+                    700,
+                    294,
+                    700
+                ],
+                "score": 0.942
+            },
+            {
+                "category_id": 4,
+                "poly": [
+                    297,
+                    1946,
+                    1403,
+                    1946,
+                    1403,
+                    2040,
+                    297,
+                    2040
+                ],
+                "score": 0.937
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    297,
+                    76,
+                    856,
+                    76,
+                    856,
+                    104,
+                    297,
+                    104
+                ],
+                "score": 0.866
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    299,
+                    1343,
+                    667,
+                    1343,
+                    667,
+                    1372,
+                    299,
+                    1372
+                ],
+                "score": 0.835
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    840,
+                    2088,
+                    858,
+                    2088,
+                    858,
+                    2112,
+                    840,
+                    2112
+                ],
+                "score": 0.804
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1188,
+                    1124,
+                    1228,
+                    1124,
+                    1228,
+                    1151,
+                    1188,
+                    1151
+                ],
+                "score": 0.77,
+                "latex": "@ 1"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1053,
+                    1948,
+                    1094,
+                    1948,
+                    1094,
+                    1975,
+                    1053,
+                    1975
+                ],
+                "score": 0.74,
+                "latex": "@ 1"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1036,
+                    1804,
+                    1109,
+                    1804,
+                    1109,
+                    1826,
+                    1036,
+                    1826
+                ],
+                "score": 0.69,
+                "latex": "( \\alpha = 0 . 0 1 "
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1036,
+                    1747,
+                    1117,
+                    1747,
+                    1117,
+                    1768,
+                    1036,
+                    1768
+                ],
+                "score": 0.61,
+                "latex": "( \\alpha = 0 . 0 0 1"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1035,
+                    1776,
+                    1110,
+                    1776,
+                    1110,
+                    1797,
+                    1035,
+                    1797
+                ],
+                "score": 0.6,
+                "latex": "( \\alpha = 0 . 0 1 )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    895,
+                    609,
+                    935,
+                    609,
+                    935,
+                    636,
+                    895,
+                    636
+                ],
+                "score": 0.5,
+                "latex": "@ 1"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    378.0,
+                    750.0,
+                    408.0,
+                    750.0,
+                    408.0,
+                    772.0,
+                    378.0,
+                    772.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    430.0,
+                    739.0,
+                    797.0,
+                    739.0,
+                    797.0,
+                    764.0,
+                    430.0,
+                    764.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    880.0,
+                    750.0,
+                    908.0,
+                    750.0,
+                    908.0,
+                    771.0,
+                    880.0,
+                    771.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    925.0,
+                    739.0,
+                    1302.0,
+                    739.0,
+                    1302.0,
+                    764.0,
+                    925.0,
+                    764.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    430.0,
+                    768.0,
+                    594.0,
+                    768.0,
+                    594.0,
+                    790.0,
+                    430.0,
+                    790.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    920.0,
+                    768.0,
+                    1095.0,
+                    768.0,
+                    1095.0,
+                    790.0,
+                    920.0,
+                    790.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    414.0,
+                    786.0,
+                    591.0,
+                    786.0,
+                    591.0,
+                    811.0,
+                    414.0,
+                    811.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    915.0,
+                    786.0,
+                    1092.0,
+                    786.0,
+                    1092.0,
+                    810.0,
+                    915.0,
+                    810.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    377.0,
+                    801.0,
+                    409.0,
+                    801.0,
+                    409.0,
+                    826.0,
+                    377.0,
+                    826.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    879.0,
+                    802.0,
+                    907.0,
+                    802.0,
+                    907.0,
+                    824.0,
+                    879.0,
+                    824.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1269.0,
+                    801.0,
+                    1278.0,
+                    801.0,
+                    1278.0,
+                    809.0,
+                    1269.0,
+                    809.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    331.0,
+                    820.0,
+                    431.0,
+                    820.0,
+                    431.0,
+                    1021.0,
+                    331.0,
+                    1021.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    777.0,
+                    826.0,
+                    787.0,
+                    826.0,
+                    787.0,
+                    836.0,
+                    777.0,
+                    836.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    674.0,
+                    829.0,
+                    683.0,
+                    829.0,
+                    683.0,
+                    839.0,
+                    674.0,
+                    839.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    847.0,
+                    836.0,
+                    896.0,
+                    836.0,
+                    896.0,
+                    1006.0,
+                    847.0,
+                    1006.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    880.0,
+                    856.0,
+                    907.0,
+                    856.0,
+                    907.0,
+                    877.0,
+                    880.0,
+                    877.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    382.0,
+                    910.0,
+                    405.0,
+                    910.0,
+                    405.0,
+                    929.0,
+                    382.0,
+                    929.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    881.0,
+                    908.0,
+                    906.0,
+                    908.0,
+                    906.0,
+                    930.0,
+                    881.0,
+                    930.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    380.0,
+                    961.0,
+                    408.0,
+                    961.0,
+                    408.0,
+                    983.0,
+                    380.0,
+                    983.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    882.0,
+                    962.0,
+                    906.0,
+                    962.0,
+                    906.0,
+                    980.0,
+                    882.0,
+                    980.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1007.0,
+                    969.0,
+                    1018.0,
+                    969.0,
+                    1018.0,
+                    979.0,
+                    1007.0,
+                    979.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    378.0,
+                    1014.0,
+                    408.0,
+                    1014.0,
+                    408.0,
+                    1036.0,
+                    378.0,
+                    1036.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    880.0,
+                    1014.0,
+                    908.0,
+                    1014.0,
+                    908.0,
+                    1036.0,
+                    880.0,
+                    1036.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    942.0,
+                    1030.0,
+                    948.0,
+                    1030.0,
+                    948.0,
+                    1036.0,
+                    942.0,
+                    1036.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    428.0,
+                    1033.0,
+                    440.0,
+                    1033.0,
+                    440.0,
+                    1048.0,
+                    428.0,
+                    1048.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    927.0,
+                    1034.0,
+                    941.0,
+                    1034.0,
+                    941.0,
+                    1048.0,
+                    927.0,
+                    1048.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    376.0,
+                    1060.0,
+                    435.0,
+                    1060.0,
+                    435.0,
+                    1105.0,
+                    376.0,
+                    1105.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    443.0,
+                    1077.0,
+                    485.0,
+                    1077.0,
+                    485.0,
+                    1097.0,
+                    443.0,
+                    1097.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    513.0,
+                    1078.0,
+                    540.0,
+                    1078.0,
+                    540.0,
+                    1095.0,
+                    513.0,
+                    1095.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    567.0,
+                    1076.0,
+                    599.0,
+                    1076.0,
+                    599.0,
+                    1097.0,
+                    567.0,
+                    1097.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    628.0,
+                    1077.0,
+                    666.0,
+                    1077.0,
+                    666.0,
+                    1097.0,
+                    628.0,
+                    1097.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    685.0,
+                    1077.0,
+                    722.0,
+                    1077.0,
+                    722.0,
+                    1097.0,
+                    685.0,
+                    1097.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    744.0,
+                    1077.0,
+                    788.0,
+                    1077.0,
+                    788.0,
+                    1097.0,
+                    744.0,
+                    1097.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    799.0,
+                    1075.0,
+                    846.0,
+                    1075.0,
+                    846.0,
+                    1098.0,
+                    799.0,
+                    1098.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    875.0,
+                    1059.0,
+                    936.0,
+                    1059.0,
+                    936.0,
+                    1103.0,
+                    875.0,
+                    1103.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    965.0,
+                    1077.0,
+                    1009.0,
+                    1077.0,
+                    1009.0,
+                    1097.0,
+                    965.0,
+                    1097.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1063.0,
+                    1078.0,
+                    1088.0,
+                    1078.0,
+                    1088.0,
+                    1094.0,
+                    1063.0,
+                    1094.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1142.0,
+                    1078.0,
+                    1168.0,
+                    1078.0,
+                    1168.0,
+                    1094.0,
+                    1142.0,
+                    1094.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1227.0,
+                    1077.0,
+                    1261.0,
+                    1077.0,
+                    1261.0,
+                    1097.0,
+                    1227.0,
+                    1097.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1307.0,
+                    1076.0,
+                    1342.0,
+                    1076.0,
+                    1342.0,
+                    1097.0,
+                    1307.0,
+                    1097.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    535.0,
+                    1090.0,
+                    694.0,
+                    1090.0,
+                    694.0,
+                    1122.0,
+                    535.0,
+                    1122.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1027.0,
+                    1087.0,
+                    1203.0,
+                    1087.0,
+                    1203.0,
+                    1123.0,
+                    1027.0,
+                    1123.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1079.0,
+                    917.0,
+                    1095.0,
+                    917.0,
+                    1095.0,
+                    925.0,
+                    1079.0,
+                    925.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    528.75,
+                    908.5,
+                    582.75,
+                    908.5,
+                    582.75,
+                    923.0,
+                    528.75,
+                    923.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    969.0,
+                    995.0,
+                    994.0,
+                    995.0,
+                    994.0,
+                    1005.5,
+                    969.0,
+                    1005.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    941.0,
+                    1018.0,
+                    969.0,
+                    1018.0,
+                    969.0,
+                    1031.5,
+                    941.0,
+                    1031.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    438.0,
+                    260.0,
+                    445.0,
+                    260.0,
+                    445.0,
+                    270.0,
+                    438.0,
+                    270.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    448.0,
+                    255.0,
+                    593.0,
+                    255.0,
+                    593.0,
+                    277.0,
+                    448.0,
+                    277.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    918.0,
+                    261.0,
+                    927.0,
+                    261.0,
+                    927.0,
+                    269.0,
+                    918.0,
+                    269.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    934.0,
+                    260.0,
+                    946.0,
+                    260.0,
+                    946.0,
+                    269.0,
+                    934.0,
+                    269.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    947.0,
+                    255.0,
+                    1092.0,
+                    255.0,
+                    1092.0,
+                    273.0,
+                    947.0,
+                    273.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    415.0,
+                    276.0,
+                    448.0,
+                    276.0,
+                    448.0,
+                    292.0,
+                    415.0,
+                    292.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    449.0,
+                    273.0,
+                    591.0,
+                    273.0,
+                    591.0,
+                    295.0,
+                    449.0,
+                    295.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    914.0,
+                    273.0,
+                    1090.0,
+                    273.0,
+                    1090.0,
+                    295.0,
+                    914.0,
+                    295.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    378.0,
+                    288.0,
+                    408.0,
+                    288.0,
+                    408.0,
+                    310.0,
+                    378.0,
+                    310.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    878.0,
+                    288.0,
+                    908.0,
+                    288.0,
+                    908.0,
+                    310.0,
+                    878.0,
+                    310.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    346.0,
+                    327.0,
+                    398.0,
+                    327.0,
+                    398.0,
+                    494.0,
+                    346.0,
+                    494.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    847.0,
+                    329.0,
+                    895.0,
+                    329.0,
+                    895.0,
+                    492.0,
+                    847.0,
+                    492.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    358.0,
+                    333.0,
+                    377.0,
+                    333.0,
+                    377.0,
+                    346.0,
+                    358.0,
+                    346.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    378.0,
+                    341.0,
+                    408.0,
+                    341.0,
+                    408.0,
+                    363.0,
+                    378.0,
+                    363.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    878.0,
+                    342.0,
+                    906.0,
+                    342.0,
+                    906.0,
+                    363.0,
+                    878.0,
+                    363.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1031.0,
+                    381.0,
+                    1037.0,
+                    381.0,
+                    1037.0,
+                    387.0,
+                    1031.0,
+                    387.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    380.0,
+                    394.0,
+                    407.0,
+                    394.0,
+                    407.0,
+                    414.0,
+                    380.0,
+                    414.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    880.0,
+                    396.0,
+                    905.0,
+                    396.0,
+                    905.0,
+                    413.0,
+                    880.0,
+                    413.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    380.0,
+                    447.0,
+                    407.0,
+                    447.0,
+                    407.0,
+                    467.0,
+                    380.0,
+                    467.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    506.0,
+                    462.0,
+                    514.0,
+                    462.0,
+                    514.0,
+                    471.0,
+                    506.0,
+                    471.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    879.0,
+                    447.0,
+                    906.0,
+                    447.0,
+                    906.0,
+                    467.0,
+                    879.0,
+                    467.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    967.0,
+                    460.0,
+                    976.0,
+                    460.0,
+                    976.0,
+                    471.0,
+                    967.0,
+                    471.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    378.0,
+                    499.0,
+                    408.0,
+                    499.0,
+                    408.0,
+                    520.0,
+                    378.0,
+                    520.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    879.0,
+                    499.0,
+                    908.0,
+                    499.0,
+                    908.0,
+                    520.0,
+                    879.0,
+                    520.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    438.0,
+                    531.0,
+                    451.0,
+                    531.0,
+                    451.0,
+                    545.0,
+                    438.0,
+                    545.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    946.0,
+                    533.0,
+                    954.0,
+                    533.0,
+                    954.0,
+                    543.0,
+                    946.0,
+                    543.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    375.0,
+                    548.0,
+                    428.0,
+                    548.0,
+                    428.0,
+                    586.0,
+                    375.0,
+                    586.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    459.0,
+                    562.0,
+                    500.0,
+                    562.0,
+                    500.0,
+                    582.0,
+                    459.0,
+                    582.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    525.0,
+                    562.0,
+                    566.0,
+                    562.0,
+                    566.0,
+                    582.0,
+                    525.0,
+                    582.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    602.0,
+                    562.0,
+                    634.0,
+                    562.0,
+                    634.0,
+                    582.0,
+                    602.0,
+                    582.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    668.0,
+                    562.0,
+                    700.0,
+                    562.0,
+                    700.0,
+                    583.0,
+                    668.0,
+                    583.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    738.0,
+                    562.0,
+                    775.0,
+                    562.0,
+                    775.0,
+                    582.0,
+                    738.0,
+                    582.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    803.0,
+                    561.0,
+                    842.0,
+                    561.0,
+                    842.0,
+                    582.0,
+                    803.0,
+                    582.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    875.0,
+                    548.0,
+                    929.0,
+                    548.0,
+                    929.0,
+                    586.0,
+                    875.0,
+                    586.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    971.0,
+                    562.0,
+                    1014.0,
+                    562.0,
+                    1014.0,
+                    582.0,
+                    971.0,
+                    582.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1050.0,
+                    564.0,
+                    1090.0,
+                    564.0,
+                    1090.0,
+                    580.0,
+                    1050.0,
+                    580.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1144.0,
+                    564.0,
+                    1170.0,
+                    564.0,
+                    1170.0,
+                    580.0,
+                    1144.0,
+                    580.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1221.0,
+                    562.0,
+                    1250.0,
+                    562.0,
+                    1250.0,
+                    583.0,
+                    1221.0,
+                    583.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1305.0,
+                    561.0,
+                    1340.0,
+                    561.0,
+                    1340.0,
+                    582.0,
+                    1305.0,
+                    582.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    533.0,
+                    574.0,
+                    695.0,
+                    574.0,
+                    695.0,
+                    608.0,
+                    533.0,
+                    608.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1028.0,
+                    576.0,
+                    1201.0,
+                    576.0,
+                    1201.0,
+                    606.0,
+                    1028.0,
+                    606.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    370.0,
+                    215.5,
+                    842.0,
+                    215.5,
+                    842.0,
+                    259.5,
+                    370.0,
+                    259.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    871.0,
+                    215.5,
+                    1346.0,
+                    215.5,
+                    1346.0,
+                    259.5,
+                    871.0,
+                    259.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    753.0,
+                    262.0,
+                    789.0,
+                    262.0,
+                    789.0,
+                    280.0,
+                    753.0,
+                    280.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1236.0,
+                    266.5,
+                    1256.0,
+                    266.5,
+                    1256.0,
+                    275.5,
+                    1236.0,
+                    275.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    631.0,
+                    324.0,
+                    666.0,
+                    324.0,
+                    666.0,
+                    340.0,
+                    631.0,
+                    340.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1081.0,
+                    326.0,
+                    1110.0,
+                    326.0,
+                    1110.0,
+                    338.0,
+                    1081.0,
+                    338.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    615.0,
+                    346.5,
+                    637.0,
+                    346.5,
+                    637.0,
+                    356.5,
+                    615.0,
+                    356.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1122.0,
+                    1187.0,
+                    1122.0,
+                    1187.0,
+                    1158.0,
+                    294.0,
+                    1158.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1229.0,
+                    1122.0,
+                    1406.0,
+                    1122.0,
+                    1406.0,
+                    1158.0,
+                    1229.0,
+                    1158.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1152.0,
+                    1404.0,
+                    1152.0,
+                    1404.0,
+                    1187.0,
+                    293.0,
+                    1187.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1179.0,
+                    1404.0,
+                    1179.0,
+                    1404.0,
+                    1217.0,
+                    295.0,
+                    1217.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1212.0,
+                    1406.0,
+                    1212.0,
+                    1406.0,
+                    1249.0,
+                    294.0,
+                    1249.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1241.0,
+                    598.0,
+                    1241.0,
+                    598.0,
+                    1280.0,
+                    295.0,
+                    1280.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    698.0,
+                    1445.0,
+                    1033.0,
+                    1445.0,
+                    1033.0,
+                    1473.0,
+                    698.0,
+                    1473.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    560.0,
+                    1458.0,
+                    596.0,
+                    1458.0,
+                    596.0,
+                    1483.0,
+                    560.0,
+                    1483.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    561.0,
+                    1509.0,
+                    594.0,
+                    1509.0,
+                    594.0,
+                    1535.0,
+                    561.0,
+                    1535.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    560.0,
+                    1561.0,
+                    596.0,
+                    1561.0,
+                    596.0,
+                    1589.0,
+                    560.0,
+                    1589.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    529.0,
+                    1595.0,
+                    563.0,
+                    1595.0,
+                    563.0,
+                    1628.0,
+                    529.0,
+                    1628.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    559.0,
+                    1613.0,
+                    595.0,
+                    1613.0,
+                    595.0,
+                    1644.0,
+                    559.0,
+                    1644.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    530.0,
+                    1616.0,
+                    566.0,
+                    1616.0,
+                    566.0,
+                    1656.0,
+                    530.0,
+                    1656.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    651.0,
+                    1633.0,
+                    661.0,
+                    1633.0,
+                    661.0,
+                    1642.0,
+                    651.0,
+                    1642.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    527.0,
+                    1644.0,
+                    567.0,
+                    1644.0,
+                    567.0,
+                    1774.0,
+                    527.0,
+                    1774.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    642.0,
+                    1644.0,
+                    658.0,
+                    1644.0,
+                    658.0,
+                    1656.0,
+                    642.0,
+                    1656.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    643.0,
+                    1661.0,
+                    652.0,
+                    1661.0,
+                    652.0,
+                    1669.0,
+                    643.0,
+                    1669.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    626.0,
+                    1666.0,
+                    652.0,
+                    1666.0,
+                    652.0,
+                    1688.0,
+                    626.0,
+                    1688.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    560.0,
+                    1668.0,
+                    595.0,
+                    1668.0,
+                    595.0,
+                    1694.0,
+                    560.0,
+                    1694.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    613.0,
+                    1694.0,
+                    630.0,
+                    1694.0,
+                    630.0,
+                    1711.0,
+                    613.0,
+                    1711.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    627.0,
+                    1699.0,
+                    638.0,
+                    1699.0,
+                    638.0,
+                    1712.0,
+                    627.0,
+                    1712.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    608.0,
+                    1708.0,
+                    624.0,
+                    1708.0,
+                    624.0,
+                    1725.0,
+                    608.0,
+                    1725.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    620.0,
+                    1713.0,
+                    633.0,
+                    1713.0,
+                    633.0,
+                    1725.0,
+                    620.0,
+                    1725.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    560.0,
+                    1720.0,
+                    595.0,
+                    1720.0,
+                    595.0,
+                    1746.0,
+                    560.0,
+                    1746.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    606.0,
+                    1722.0,
+                    617.0,
+                    1722.0,
+                    617.0,
+                    1742.0,
+                    606.0,
+                    1742.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    618.0,
+                    1728.0,
+                    627.0,
+                    1728.0,
+                    627.0,
+                    1739.0,
+                    618.0,
+                    1739.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    619.0,
+                    1746.0,
+                    624.0,
+                    1746.0,
+                    624.0,
+                    1751.0,
+                    619.0,
+                    1751.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    908.0,
+                    1741.0,
+                    1035.0,
+                    1741.0,
+                    1035.0,
+                    1773.0,
+                    908.0,
+                    1773.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1118.0,
+                    1741.0,
+                    1127.0,
+                    1741.0,
+                    1127.0,
+                    1773.0,
+                    1118.0,
+                    1773.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    599.0,
+                    1749.0,
+                    625.0,
+                    1749.0,
+                    625.0,
+                    1768.0,
+                    599.0,
+                    1768.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    560.0,
+                    1773.0,
+                    595.0,
+                    1773.0,
+                    595.0,
+                    1798.0,
+                    560.0,
+                    1798.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    596.0,
+                    1763.0,
+                    628.0,
+                    1763.0,
+                    628.0,
+                    1798.0,
+                    596.0,
+                    1798.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    908.0,
+                    1770.0,
+                    1034.0,
+                    1770.0,
+                    1034.0,
+                    1800.0,
+                    908.0,
+                    1800.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1111.0,
+                    1770.0,
+                    1120.0,
+                    1770.0,
+                    1120.0,
+                    1800.0,
+                    1111.0,
+                    1800.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    598.0,
+                    1792.0,
+                    627.0,
+                    1792.0,
+                    627.0,
+                    1831.0,
+                    598.0,
+                    1831.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    907.0,
+                    1798.0,
+                    1035.0,
+                    1798.0,
+                    1035.0,
+                    1830.0,
+                    907.0,
+                    1830.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1110.0,
+                    1798.0,
+                    1119.0,
+                    1798.0,
+                    1119.0,
+                    1830.0,
+                    1110.0,
+                    1830.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    561.0,
+                    1825.0,
+                    594.0,
+                    1825.0,
+                    594.0,
+                    1852.0,
+                    561.0,
+                    1852.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    610.0,
+                    1839.0,
+                    621.0,
+                    1839.0,
+                    621.0,
+                    1853.0,
+                    610.0,
+                    1853.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    907.0,
+                    1827.0,
+                    1029.0,
+                    1827.0,
+                    1029.0,
+                    1857.0,
+                    907.0,
+                    1857.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    907.0,
+                    1854.0,
+                    1068.0,
+                    1854.0,
+                    1068.0,
+                    1880.0,
+                    907.0,
+                    1880.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    562.0,
+                    1879.0,
+                    594.0,
+                    1879.0,
+                    594.0,
+                    1902.0,
+                    562.0,
+                    1902.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    588.0,
+                    1894.0,
+                    603.0,
+                    1894.0,
+                    603.0,
+                    1911.0,
+                    588.0,
+                    1911.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    659.0,
+                    1890.0,
+                    712.0,
+                    1890.0,
+                    712.0,
+                    1914.0,
+                    659.0,
+                    1914.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    749.0,
+                    1890.0,
+                    803.0,
+                    1890.0,
+                    803.0,
+                    1914.0,
+                    749.0,
+                    1914.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    840.0,
+                    1889.0,
+                    894.0,
+                    1889.0,
+                    894.0,
+                    1914.0,
+                    840.0,
+                    1914.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    930.0,
+                    1889.0,
+                    984.0,
+                    1889.0,
+                    984.0,
+                    1914.0,
+                    930.0,
+                    1914.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1020.0,
+                    1889.0,
+                    1074.0,
+                    1889.0,
+                    1074.0,
+                    1914.0,
+                    1020.0,
+                    1914.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1111.0,
+                    1889.0,
+                    1160.0,
+                    1889.0,
+                    1160.0,
+                    1914.0,
+                    1111.0,
+                    1914.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    827.0,
+                    1909.0,
+                    906.0,
+                    1909.0,
+                    906.0,
+                    1946.0,
+                    827.0,
+                    1946.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    620.0,
+                    1680.0,
+                    642.0,
+                    1680.0,
+                    642.0,
+                    1698.0,
+                    620.0,
+                    1698.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    605.0,
+                    894.0,
+                    605.0,
+                    894.0,
+                    643.0,
+                    294.0,
+                    643.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    936.0,
+                    605.0,
+                    1406.0,
+                    605.0,
+                    1406.0,
+                    643.0,
+                    936.0,
+                    643.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    638.0,
+                    1405.0,
+                    638.0,
+                    1405.0,
+                    672.0,
+                    296.0,
+                    672.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    668.0,
+                    839.0,
+                    668.0,
+                    839.0,
+                    705.0,
+                    294.0,
+                    705.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1943.0,
+                    1052.0,
+                    1943.0,
+                    1052.0,
+                    1983.0,
+                    294.0,
+                    1983.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1095.0,
+                    1943.0,
+                    1404.0,
+                    1943.0,
+                    1404.0,
+                    1983.0,
+                    1095.0,
+                    1983.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1976.0,
+                    1404.0,
+                    1976.0,
+                    1404.0,
+                    2011.0,
+                    295.0,
+                    2011.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    2004.0,
+                    493.0,
+                    2004.0,
+                    493.0,
+                    2043.0,
+                    294.0,
+                    2043.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    298.0,
+                    73.0,
+                    858.0,
+                    73.0,
+                    858.0,
+                    108.0,
+                    298.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1342.0,
+                    667.0,
+                    1342.0,
+                    667.0,
+                    1375.0,
+                    296.0,
+                    1375.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    836.0,
+                    2085.0,
+                    859.0,
+                    2085.0,
+                    859.0,
+                    2116.0,
+                    836.0,
+                    2116.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 7,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    775,
+                    1404,
+                    775,
+                    1404,
+                    1141,
+                    297,
+                    1141
+                ],
+                "score": 0.983
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    399,
+                    1404,
+                    399,
+                    1404,
+                    674,
+                    298,
+                    674
+                ],
+                "score": 0.98
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1157,
+                    1403,
+                    1157,
+                    1403,
+                    1310,
+                    298,
+                    1310
+                ],
+                "score": 0.976
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    229,
+                    1403,
+                    229,
+                    1403,
+                    383,
+                    298,
+                    383
+                ],
+                "score": 0.973
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    300,
+                    708,
+                    543,
+                    708,
+                    543,
+                    743,
+                    300,
+                    743
+                ],
+                "score": 0.892
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    299,
+                    1347,
+                    487,
+                    1347,
+                    487,
+                    1379,
+                    299,
+                    1379
+                ],
+                "score": 0.89
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    299,
+                    76,
+                    856,
+                    76,
+                    856,
+                    104,
+                    299,
+                    104
+                ],
+                "score": 0.86
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    840,
+                    2088,
+                    858,
+                    2088,
+                    858,
+                    2111,
+                    840,
+                    2111
+                ],
+                "score": 0.78
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    296,
+                    1977,
+                    1405,
+                    1977,
+                    1405,
+                    2033,
+                    296,
+                    2033
+                ],
+                "score": 0.705
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    300,
+                    1411,
+                    1404,
+                    1411,
+                    1404,
+                    1634,
+                    300,
+                    1634
+                ],
+                "score": 0.642
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    294,
+                    1833,
+                    1398,
+                    1833,
+                    1398,
+                    1890,
+                    294,
+                    1890
+                ],
+                "score": 0.636
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1918,
+                    1128,
+                    1918,
+                    1128,
+                    1949,
+                    297,
+                    1949
+                ],
+                "score": 0.625
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    293,
+                    1747,
+                    1398,
+                    1747,
+                    1398,
+                    1805,
+                    293,
+                    1805
+                ],
+                "score": 0.595
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    289,
+                    1662,
+                    1401,
+                    1662,
+                    1401,
+                    1720,
+                    289,
+                    1720
+                ],
+                "score": 0.574
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    300,
+                    76,
+                    854,
+                    76,
+                    854,
+                    104,
+                    300,
+                    104
+                ],
+                "score": 0.108
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1166,
+                    400,
+                    1227,
+                    400,
+                    1227,
+                    428,
+                    1166,
+                    428
+                ],
+                "score": 0.86,
+                "latex": "0 . 1 \\%"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1233,
+                    1020,
+                    1300,
+                    1020,
+                    1300,
+                    1049,
+                    1233,
+                    1049
+                ],
+                "score": 0.86,
+                "latex": "> 0 . 1"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1089,
+                    400,
+                    1129,
+                    400,
+                    1129,
+                    428,
+                    1089,
+                    428
+                ],
+                "score": 0.84,
+                "latex": "1 \\%"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1343,
+                    430,
+                    1404,
+                    430,
+                    1404,
+                    459,
+                    1343,
+                    459
+                ],
+                "score": 0.79,
+                "latex": "T ="
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    984,
+                    431,
+                    1008,
+                    431,
+                    1008,
+                    457,
+                    984,
+                    457
+                ],
+                "score": 0.78,
+                "latex": "T"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    430,
+                    461,
+                    572,
+                    461,
+                    572,
+                    491,
+                    430,
+                    491
+                ],
+                "score": 0.76,
+                "latex": "T = 5 0 , 0 0 0"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1180,
+                    430,
+                    1331,
+                    430,
+                    1331,
+                    461,
+                    1180,
+                    461
+                ],
+                "score": 0.46,
+                "latex": "T = 1 0 , 0 0 0"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    296,
+                    958,
+                    361,
+                    958,
+                    361,
+                    989,
+                    296,
+                    989
+                ],
+                "score": 0.26,
+                "latex": "3 . 5 \\mathrm { M }"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    702.0,
+                    549.0,
+                    702.0,
+                    549.0,
+                    752.0,
+                    291.0,
+                    752.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1345.0,
+                    491.0,
+                    1345.0,
+                    491.0,
+                    1385.0,
+                    296.0,
+                    1385.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    298.0,
+                    73.0,
+                    858.0,
+                    73.0,
+                    858.0,
+                    108.0,
+                    298.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    839.0,
+                    2087.0,
+                    860.0,
+                    2087.0,
+                    860.0,
+                    2117.0,
+                    839.0,
+                    2117.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    73.0,
+                    858.0,
+                    73.0,
+                    858.0,
+                    108.0,
+                    297.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    772.0,
+                    1408.0,
+                    772.0,
+                    1408.0,
+                    814.0,
+                    292.0,
+                    814.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    804.0,
+                    1406.0,
+                    804.0,
+                    1406.0,
+                    843.0,
+                    291.0,
+                    843.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    837.0,
+                    1405.0,
+                    837.0,
+                    1405.0,
+                    869.0,
+                    294.0,
+                    869.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    867.0,
+                    1405.0,
+                    867.0,
+                    1405.0,
+                    901.0,
+                    295.0,
+                    901.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    895.0,
+                    1404.0,
+                    895.0,
+                    1404.0,
+                    933.0,
+                    292.0,
+                    933.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    926.0,
+                    1404.0,
+                    926.0,
+                    1404.0,
+                    963.0,
+                    294.0,
+                    963.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    362.0,
+                    959.0,
+                    1404.0,
+                    959.0,
+                    1404.0,
+                    993.0,
+                    362.0,
+                    993.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    986.0,
+                    1405.0,
+                    986.0,
+                    1405.0,
+                    1026.0,
+                    291.0,
+                    1026.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1017.0,
+                    1232.0,
+                    1017.0,
+                    1232.0,
+                    1056.0,
+                    291.0,
+                    1056.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1301.0,
+                    1017.0,
+                    1406.0,
+                    1017.0,
+                    1406.0,
+                    1056.0,
+                    1301.0,
+                    1056.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1050.0,
+                    1404.0,
+                    1050.0,
+                    1404.0,
+                    1084.0,
+                    295.0,
+                    1084.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1081.0,
+                    1404.0,
+                    1081.0,
+                    1404.0,
+                    1115.0,
+                    295.0,
+                    1115.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1113.0,
+                    896.0,
+                    1113.0,
+                    896.0,
+                    1144.0,
+                    296.0,
+                    1144.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    401.0,
+                    1088.0,
+                    401.0,
+                    1088.0,
+                    434.0,
+                    294.0,
+                    434.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1130.0,
+                    401.0,
+                    1165.0,
+                    401.0,
+                    1165.0,
+                    434.0,
+                    1130.0,
+                    434.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1228.0,
+                    401.0,
+                    1405.0,
+                    401.0,
+                    1405.0,
+                    434.0,
+                    1228.0,
+                    434.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    428.0,
+                    983.0,
+                    428.0,
+                    983.0,
+                    465.0,
+                    293.0,
+                    465.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1009.0,
+                    428.0,
+                    1179.0,
+                    428.0,
+                    1179.0,
+                    465.0,
+                    1009.0,
+                    465.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1332.0,
+                    428.0,
+                    1342.0,
+                    428.0,
+                    1342.0,
+                    465.0,
+                    1332.0,
+                    465.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    456.0,
+                    429.0,
+                    456.0,
+                    429.0,
+                    497.0,
+                    293.0,
+                    497.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    573.0,
+                    456.0,
+                    1406.0,
+                    456.0,
+                    1406.0,
+                    497.0,
+                    573.0,
+                    497.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    490.0,
+                    1404.0,
+                    490.0,
+                    1404.0,
+                    526.0,
+                    294.0,
+                    526.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    522.0,
+                    1405.0,
+                    522.0,
+                    1405.0,
+                    559.0,
+                    292.0,
+                    559.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    551.0,
+                    1405.0,
+                    551.0,
+                    1405.0,
+                    587.0,
+                    294.0,
+                    587.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    578.0,
+                    1404.0,
+                    578.0,
+                    1404.0,
+                    620.0,
+                    292.0,
+                    620.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    607.0,
+                    1405.0,
+                    607.0,
+                    1405.0,
+                    652.0,
+                    292.0,
+                    652.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    641.0,
+                    1356.0,
+                    641.0,
+                    1356.0,
+                    678.0,
+                    292.0,
+                    678.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1156.0,
+                    1404.0,
+                    1156.0,
+                    1404.0,
+                    1191.0,
+                    294.0,
+                    1191.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1185.0,
+                    1404.0,
+                    1185.0,
+                    1404.0,
+                    1222.0,
+                    294.0,
+                    1222.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1218.0,
+                    1407.0,
+                    1218.0,
+                    1407.0,
+                    1253.0,
+                    293.0,
+                    1253.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1249.0,
+                    1404.0,
+                    1249.0,
+                    1404.0,
+                    1282.0,
+                    294.0,
+                    1282.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1280.0,
+                    561.0,
+                    1280.0,
+                    561.0,
+                    1317.0,
+                    294.0,
+                    1317.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    229.0,
+                    1407.0,
+                    229.0,
+                    1407.0,
+                    266.0,
+                    294.0,
+                    266.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    260.0,
+                    1407.0,
+                    260.0,
+                    1407.0,
+                    295.0,
+                    293.0,
+                    295.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    293.0,
+                    1405.0,
+                    293.0,
+                    1405.0,
+                    326.0,
+                    293.0,
+                    326.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    321.0,
+                    1405.0,
+                    321.0,
+                    1405.0,
+                    358.0,
+                    293.0,
+                    358.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    352.0,
+                    1238.0,
+                    352.0,
+                    1238.0,
+                    385.0,
+                    294.0,
+                    385.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1972.0,
+                    1403.0,
+                    1972.0,
+                    1403.0,
+                    2012.0,
+                    294.0,
+                    2012.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    2005.0,
+                    815.0,
+                    2005.0,
+                    815.0,
+                    2034.0,
+                    323.0,
+                    2034.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1411.0,
+                    1405.0,
+                    1411.0,
+                    1405.0,
+                    1443.0,
+                    295.0,
+                    1443.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    1435.0,
+                    1404.0,
+                    1435.0,
+                    1404.0,
+                    1475.0,
+                    322.0,
+                    1475.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    319.0,
+                    1461.0,
+                    1408.0,
+                    1461.0,
+                    1408.0,
+                    1505.0,
+                    319.0,
+                    1505.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    320.0,
+                    1491.0,
+                    1406.0,
+                    1491.0,
+                    1406.0,
+                    1528.0,
+                    320.0,
+                    1528.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    321.0,
+                    1521.0,
+                    1404.0,
+                    1521.0,
+                    1404.0,
+                    1557.0,
+                    321.0,
+                    1557.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    1549.0,
+                    1403.0,
+                    1549.0,
+                    1403.0,
+                    1581.0,
+                    323.0,
+                    1581.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    1578.0,
+                    1404.0,
+                    1578.0,
+                    1404.0,
+                    1610.0,
+                    323.0,
+                    1610.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    1602.0,
+                    683.0,
+                    1602.0,
+                    683.0,
+                    1641.0,
+                    322.0,
+                    1641.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1833.0,
+                    1402.0,
+                    1833.0,
+                    1402.0,
+                    1868.0,
+                    294.0,
+                    1868.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    1860.0,
+                    569.0,
+                    1860.0,
+                    569.0,
+                    1890.0,
+                    322.0,
+                    1890.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1916.0,
+                    1130.0,
+                    1916.0,
+                    1130.0,
+                    1954.0,
+                    294.0,
+                    1954.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1746.0,
+                    1403.0,
+                    1746.0,
+                    1403.0,
+                    1781.0,
+                    294.0,
+                    1781.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    1776.0,
+                    743.0,
+                    1776.0,
+                    743.0,
+                    1807.0,
+                    323.0,
+                    1807.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1661.0,
+                    1404.0,
+                    1661.0,
+                    1404.0,
+                    1696.0,
+                    293.0,
+                    1696.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    320.0,
+                    1690.0,
+                    847.0,
+                    1690.0,
+                    847.0,
+                    1723.0,
+                    320.0,
+                    1723.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 8,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 2,
+                "poly": [
+                    297,
+                    75,
+                    857,
+                    75,
+                    857,
+                    105,
+                    297,
+                    105
+                ],
+                "score": 0.889
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    836,
+                    2088,
+                    866,
+                    2088,
+                    866,
+                    2113,
+                    836,
+                    2113
+                ],
+                "score": 0.831
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    292,
+                    209,
+                    1410,
+                    209,
+                    1410,
+                    2045,
+                    292,
+                    2045
+                ],
+                "score": 0.656
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1364,
+                    1979,
+                    1394,
+                    1979,
+                    1394,
+                    2004,
+                    1364,
+                    2004
+                ],
+                "score": 0.81,
+                "latex": "+ 1"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1154,
+                    421,
+                    1185,
+                    421,
+                    1185,
+                    445,
+                    1154,
+                    445
+                ],
+                "score": 0.8,
+                "latex": "+ 1"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    73.0,
+                    858.0,
+                    73.0,
+                    858.0,
+                    108.0,
+                    297.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    832.0,
+                    2085.0,
+                    870.0,
+                    2085.0,
+                    870.0,
+                    2123.0,
+                    832.0,
+                    2123.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    228.0,
+                    1404.0,
+                    228.0,
+                    1404.0,
+                    270.0,
+                    292.0,
+                    270.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    318.0,
+                    260.0,
+                    1332.0,
+                    260.0,
+                    1332.0,
+                    296.0,
+                    318.0,
+                    296.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    312.0,
+                    1404.0,
+                    312.0,
+                    1404.0,
+                    345.0,
+                    296.0,
+                    345.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    320.0,
+                    338.0,
+                    899.0,
+                    338.0,
+                    899.0,
+                    377.0,
+                    320.0,
+                    377.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    290.0,
+                    387.0,
+                    1408.0,
+                    387.0,
+                    1408.0,
+                    427.0,
+                    290.0,
+                    427.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    320.0,
+                    417.0,
+                    1153.0,
+                    417.0,
+                    1153.0,
+                    453.0,
+                    320.0,
+                    453.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1186.0,
+                    417.0,
+                    1406.0,
+                    417.0,
+                    1406.0,
+                    453.0,
+                    1186.0,
+                    453.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    447.0,
+                    583.0,
+                    447.0,
+                    583.0,
+                    480.0,
+                    322.0,
+                    480.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    500.0,
+                    1374.0,
+                    500.0,
+                    1374.0,
+                    532.0,
+                    296.0,
+                    532.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    550.0,
+                    1406.0,
+                    550.0,
+                    1406.0,
+                    583.0,
+                    294.0,
+                    583.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    318.0,
+                    577.0,
+                    659.0,
+                    577.0,
+                    659.0,
+                    613.0,
+                    318.0,
+                    613.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    631.0,
+                    1406.0,
+                    631.0,
+                    1406.0,
+                    663.0,
+                    294.0,
+                    663.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    316.0,
+                    655.0,
+                    659.0,
+                    655.0,
+                    659.0,
+                    691.0,
+                    316.0,
+                    691.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    710.0,
+                    1406.0,
+                    710.0,
+                    1406.0,
+                    742.0,
+                    296.0,
+                    742.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    324.0,
+                    738.0,
+                    1029.0,
+                    738.0,
+                    1029.0,
+                    770.0,
+                    324.0,
+                    770.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    788.0,
+                    1330.0,
+                    788.0,
+                    1330.0,
+                    820.0,
+                    296.0,
+                    820.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    833.0,
+                    1410.0,
+                    833.0,
+                    1410.0,
+                    881.0,
+                    292.0,
+                    881.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    320.0,
+                    865.0,
+                    927.0,
+                    865.0,
+                    927.0,
+                    901.0,
+                    320.0,
+                    901.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    917.0,
+                    1408.0,
+                    917.0,
+                    1408.0,
+                    956.0,
+                    294.0,
+                    956.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    948.0,
+                    923.0,
+                    948.0,
+                    923.0,
+                    980.0,
+                    322.0,
+                    980.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    998.0,
+                    1406.0,
+                    998.0,
+                    1406.0,
+                    1030.0,
+                    294.0,
+                    1030.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    1026.0,
+                    577.0,
+                    1026.0,
+                    577.0,
+                    1058.0,
+                    322.0,
+                    1058.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1077.0,
+                    1408.0,
+                    1077.0,
+                    1408.0,
+                    1115.0,
+                    294.0,
+                    1115.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    316.0,
+                    1105.0,
+                    1005.0,
+                    1105.0,
+                    1005.0,
+                    1141.0,
+                    316.0,
+                    1141.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    290.0,
+                    1151.0,
+                    1408.0,
+                    1151.0,
+                    1408.0,
+                    1192.0,
+                    290.0,
+                    1192.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    320.0,
+                    1181.0,
+                    1406.0,
+                    1181.0,
+                    1406.0,
+                    1222.0,
+                    320.0,
+                    1222.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    316.0,
+                    1210.0,
+                    684.0,
+                    1210.0,
+                    684.0,
+                    1246.0,
+                    316.0,
+                    1246.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1264.0,
+                    1404.0,
+                    1264.0,
+                    1404.0,
+                    1296.0,
+                    296.0,
+                    1296.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    318.0,
+                    1293.0,
+                    616.0,
+                    1293.0,
+                    616.0,
+                    1326.0,
+                    318.0,
+                    1326.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1341.0,
+                    1408.0,
+                    1341.0,
+                    1408.0,
+                    1379.0,
+                    292.0,
+                    1379.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    318.0,
+                    1369.0,
+                    452.0,
+                    1369.0,
+                    452.0,
+                    1401.0,
+                    318.0,
+                    1401.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1421.0,
+                    1408.0,
+                    1421.0,
+                    1408.0,
+                    1460.0,
+                    292.0,
+                    1460.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    324.0,
+                    1452.0,
+                    476.0,
+                    1452.0,
+                    476.0,
+                    1484.0,
+                    324.0,
+                    1484.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1500.0,
+                    1408.0,
+                    1500.0,
+                    1408.0,
+                    1538.0,
+                    294.0,
+                    1538.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    1530.0,
+                    480.0,
+                    1530.0,
+                    480.0,
+                    1563.0,
+                    322.0,
+                    1563.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1575.0,
+                    1406.0,
+                    1575.0,
+                    1406.0,
+                    1617.0,
+                    292.0,
+                    1617.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    318.0,
+                    1607.0,
+                    1380.0,
+                    1607.0,
+                    1380.0,
+                    1645.0,
+                    318.0,
+                    1645.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1661.0,
+                    1406.0,
+                    1661.0,
+                    1406.0,
+                    1694.0,
+                    296.0,
+                    1694.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    318.0,
+                    1690.0,
+                    524.0,
+                    1690.0,
+                    524.0,
+                    1722.0,
+                    318.0,
+                    1722.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    290.0,
+                    1734.0,
+                    1408.0,
+                    1734.0,
+                    1408.0,
+                    1778.0,
+                    290.0,
+                    1778.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    318.0,
+                    1766.0,
+                    1196.0,
+                    1766.0,
+                    1196.0,
+                    1803.0,
+                    318.0,
+                    1803.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1817.0,
+                    1406.0,
+                    1817.0,
+                    1406.0,
+                    1855.0,
+                    294.0,
+                    1855.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    318.0,
+                    1843.0,
+                    599.0,
+                    1843.0,
+                    599.0,
+                    1883.0,
+                    318.0,
+                    1883.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    290.0,
+                    1893.0,
+                    1406.0,
+                    1893.0,
+                    1406.0,
+                    1934.0,
+                    290.0,
+                    1934.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    320.0,
+                    1926.0,
+                    745.0,
+                    1926.0,
+                    745.0,
+                    1964.0,
+                    320.0,
+                    1964.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1976.0,
+                    1363.0,
+                    1976.0,
+                    1363.0,
+                    2014.0,
+                    294.0,
+                    2014.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1395.0,
+                    1976.0,
+                    1408.0,
+                    1976.0,
+                    1408.0,
+                    2014.0,
+                    1395.0,
+                    2014.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    320.0,
+                    2006.0,
+                    1061.0,
+                    2006.0,
+                    1061.0,
+                    2039.0,
+                    320.0,
+                    2039.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 9,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 2,
+                "poly": [
+                    836,
+                    2088,
+                    863,
+                    2088,
+                    863,
+                    2113,
+                    836,
+                    2113
+                ],
+                "score": 0.829
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    297,
+                    75,
+                    857,
+                    75,
+                    857,
+                    105,
+                    297,
+                    105
+                ],
+                "score": 0.799
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    296,
+                    1167,
+                    1400,
+                    1167,
+                    1400,
+                    1228,
+                    296,
+                    1228
+                ],
+                "score": 0.6
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    293,
+                    1089,
+                    1399,
+                    1089,
+                    1399,
+                    1149,
+                    293,
+                    1149
+                ],
+                "score": 0.569
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    295,
+                    1245,
+                    1399,
+                    1245,
+                    1399,
+                    1305,
+                    295,
+                    1305
+                ],
+                "score": 0.538
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    698,
+                    1399,
+                    698,
+                    1399,
+                    760,
+                    297,
+                    760
+                ],
+                "score": 0.493
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1011,
+                    1399,
+                    1011,
+                    1399,
+                    1072,
+                    297,
+                    1072
+                ],
+                "score": 0.491
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    776,
+                    1400,
+                    776,
+                    1400,
+                    836,
+                    298,
+                    836
+                ],
+                "score": 0.486
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    620,
+                    1401,
+                    620,
+                    1401,
+                    680,
+                    299,
+                    680
+                ],
+                "score": 0.477
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    292,
+                    1976,
+                    1405,
+                    1976,
+                    1405,
+                    2034,
+                    292,
+                    2034
+                ],
+                "score": 0.471
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    296,
+                    932,
+                    1401,
+                    932,
+                    1401,
+                    993,
+                    296,
+                    993
+                ],
+                "score": 0.471
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1323,
+                    1399,
+                    1323,
+                    1399,
+                    1383,
+                    297,
+                    1383
+                ],
+                "score": 0.449
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    855,
+                    1402,
+                    855,
+                    1402,
+                    915,
+                    299,
+                    915
+                ],
+                "score": 0.448
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    542,
+                    1402,
+                    542,
+                    1402,
+                    603,
+                    297,
+                    603
+                ],
+                "score": 0.44
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1479,
+                    1398,
+                    1479,
+                    1398,
+                    1567,
+                    298,
+                    1567
+                ],
+                "score": 0.437
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    294,
+                    230,
+                    1399,
+                    230,
+                    1399,
+                    291,
+                    294,
+                    291
+                ],
+                "score": 0.436
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1898,
+                    1400,
+                    1898,
+                    1400,
+                    1957,
+                    297,
+                    1957
+                ],
+                "score": 0.436
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    293,
+                    1820,
+                    1400,
+                    1820,
+                    1400,
+                    1879,
+                    293,
+                    1879
+                ],
+                "score": 0.426
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    295,
+                    1402,
+                    1401,
+                    1402,
+                    1401,
+                    1461,
+                    295,
+                    1461
+                ],
+                "score": 0.426
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1663,
+                    1398,
+                    1663,
+                    1398,
+                    1750,
+                    298,
+                    1750
+                ],
+                "score": 0.422
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    296,
+                    1584,
+                    1401,
+                    1584,
+                    1401,
+                    1644,
+                    296,
+                    1644
+                ],
+                "score": 0.421
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    464,
+                    1400,
+                    464,
+                    1400,
+                    525,
+                    299,
+                    525
+                ],
+                "score": 0.412
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    300,
+                    386,
+                    1398,
+                    386,
+                    1398,
+                    446,
+                    300,
+                    446
+                ],
+                "score": 0.404
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    309,
+                    1398,
+                    309,
+                    1398,
+                    369,
+                    299,
+                    369
+                ],
+                "score": 0.394
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    294,
+                    1768,
+                    1369,
+                    1768,
+                    1369,
+                    1802,
+                    294,
+                    1802
+                ],
+                "score": 0.346
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    298,
+                    75,
+                    857,
+                    75,
+                    857,
+                    105,
+                    298,
+                    105
+                ],
+                "score": 0.116
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    832.0,
+                    2085.0,
+                    868.0,
+                    2085.0,
+                    868.0,
+                    2125.0,
+                    832.0,
+                    2125.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    73.0,
+                    858.0,
+                    73.0,
+                    858.0,
+                    108.0,
+                    297.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    298.0,
+                    73.0,
+                    857.0,
+                    73.0,
+                    857.0,
+                    108.0,
+                    298.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1167.0,
+                    1404.0,
+                    1167.0,
+                    1404.0,
+                    1203.0,
+                    295.0,
+                    1203.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    1195.0,
+                    509.0,
+                    1195.0,
+                    509.0,
+                    1226.0,
+                    323.0,
+                    1226.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1084.0,
+                    1404.0,
+                    1084.0,
+                    1404.0,
+                    1129.0,
+                    291.0,
+                    1129.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    1118.0,
+                    488.0,
+                    1118.0,
+                    488.0,
+                    1149.0,
+                    322.0,
+                    1149.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1245.0,
+                    1403.0,
+                    1245.0,
+                    1403.0,
+                    1280.0,
+                    295.0,
+                    1280.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    1274.0,
+                    478.0,
+                    1274.0,
+                    478.0,
+                    1303.0,
+                    323.0,
+                    1303.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    695.0,
+                    1403.0,
+                    695.0,
+                    1403.0,
+                    735.0,
+                    295.0,
+                    735.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    325.0,
+                    728.0,
+                    1115.0,
+                    728.0,
+                    1115.0,
+                    759.0,
+                    325.0,
+                    759.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1013.0,
+                    1403.0,
+                    1013.0,
+                    1403.0,
+                    1044.0,
+                    296.0,
+                    1044.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    321.0,
+                    1039.0,
+                    596.0,
+                    1039.0,
+                    596.0,
+                    1069.0,
+                    321.0,
+                    1069.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    773.0,
+                    1405.0,
+                    773.0,
+                    1405.0,
+                    811.0,
+                    294.0,
+                    811.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    806.0,
+                    572.0,
+                    806.0,
+                    572.0,
+                    836.0,
+                    323.0,
+                    836.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    620.0,
+                    1405.0,
+                    620.0,
+                    1405.0,
+                    655.0,
+                    295.0,
+                    655.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    650.0,
+                    653.0,
+                    650.0,
+                    653.0,
+                    681.0,
+                    322.0,
+                    681.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1975.0,
+                    1403.0,
+                    1975.0,
+                    1403.0,
+                    2010.0,
+                    295.0,
+                    2010.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    320.0,
+                    2004.0,
+                    588.0,
+                    2004.0,
+                    588.0,
+                    2033.0,
+                    320.0,
+                    2033.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    931.0,
+                    1405.0,
+                    931.0,
+                    1405.0,
+                    968.0,
+                    294.0,
+                    968.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    963.0,
+                    730.0,
+                    963.0,
+                    730.0,
+                    993.0,
+                    322.0,
+                    993.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1320.0,
+                    1404.0,
+                    1320.0,
+                    1404.0,
+                    1358.0,
+                    292.0,
+                    1358.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    320.0,
+                    1349.0,
+                    393.0,
+                    1349.0,
+                    393.0,
+                    1384.0,
+                    320.0,
+                    1384.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    852.0,
+                    1404.0,
+                    852.0,
+                    1404.0,
+                    890.0,
+                    295.0,
+                    890.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    320.0,
+                    884.0,
+                    973.0,
+                    884.0,
+                    973.0,
+                    916.0,
+                    320.0,
+                    916.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    541.0,
+                    1407.0,
+                    541.0,
+                    1407.0,
+                    578.0,
+                    296.0,
+                    578.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    321.0,
+                    571.0,
+                    1285.0,
+                    571.0,
+                    1285.0,
+                    605.0,
+                    321.0,
+                    605.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1480.0,
+                    1401.0,
+                    1480.0,
+                    1401.0,
+                    1514.0,
+                    294.0,
+                    1514.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    1505.0,
+                    1404.0,
+                    1505.0,
+                    1404.0,
+                    1544.0,
+                    322.0,
+                    1544.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    1536.0,
+                    987.0,
+                    1536.0,
+                    987.0,
+                    1569.0,
+                    323.0,
+                    1569.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    230.0,
+                    1401.0,
+                    230.0,
+                    1401.0,
+                    264.0,
+                    297.0,
+                    264.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    259.0,
+                    1125.0,
+                    259.0,
+                    1125.0,
+                    293.0,
+                    322.0,
+                    293.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1895.0,
+                    1405.0,
+                    1895.0,
+                    1405.0,
+                    1934.0,
+                    292.0,
+                    1934.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    1925.0,
+                    624.0,
+                    1925.0,
+                    624.0,
+                    1956.0,
+                    322.0,
+                    1956.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1820.0,
+                    1403.0,
+                    1820.0,
+                    1403.0,
+                    1854.0,
+                    296.0,
+                    1854.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    1848.0,
+                    923.0,
+                    1848.0,
+                    923.0,
+                    1879.0,
+                    322.0,
+                    1879.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1402.0,
+                    1405.0,
+                    1402.0,
+                    1405.0,
+                    1436.0,
+                    294.0,
+                    1436.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    1430.0,
+                    937.0,
+                    1430.0,
+                    937.0,
+                    1460.0,
+                    323.0,
+                    1460.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1663.0,
+                    1403.0,
+                    1663.0,
+                    1403.0,
+                    1697.0,
+                    294.0,
+                    1697.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    1694.0,
+                    1403.0,
+                    1694.0,
+                    1403.0,
+                    1723.0,
+                    322.0,
+                    1723.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    1721.0,
+                    512.0,
+                    1721.0,
+                    512.0,
+                    1750.0,
+                    323.0,
+                    1750.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1581.0,
+                    1401.0,
+                    1581.0,
+                    1401.0,
+                    1620.0,
+                    294.0,
+                    1620.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    320.0,
+                    1613.0,
+                    482.0,
+                    1613.0,
+                    482.0,
+                    1643.0,
+                    320.0,
+                    1643.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    467.0,
+                    1402.0,
+                    467.0,
+                    1402.0,
+                    498.0,
+                    297.0,
+                    498.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    494.0,
+                    1098.0,
+                    494.0,
+                    1098.0,
+                    526.0,
+                    323.0,
+                    526.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    388.0,
+                    1402.0,
+                    388.0,
+                    1402.0,
+                    420.0,
+                    296.0,
+                    420.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    321.0,
+                    416.0,
+                    672.0,
+                    416.0,
+                    672.0,
+                    447.0,
+                    321.0,
+                    447.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    298.0,
+                    310.0,
+                    1402.0,
+                    310.0,
+                    1402.0,
+                    341.0,
+                    298.0,
+                    341.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    338.0,
+                    986.0,
+                    338.0,
+                    986.0,
+                    369.0,
+                    323.0,
+                    369.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1765.0,
+                    1369.0,
+                    1765.0,
+                    1369.0,
+                    1805.0,
+                    294.0,
+                    1805.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 10,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 2,
+                "poly": [
+                    836,
+                    2088,
+                    865,
+                    2088,
+                    865,
+                    2112,
+                    836,
+                    2112
+                ],
+                "score": 0.832
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    231,
+                    1402,
+                    231,
+                    1402,
+                    289,
+                    297,
+                    289
+                ],
+                "score": 0.818
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    310,
+                    1400,
+                    310,
+                    1400,
+                    366,
+                    298,
+                    366
+                ],
+                "score": 0.802
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    298,
+                    75,
+                    857,
+                    75,
+                    857,
+                    105,
+                    298,
+                    105
+                ],
+                "score": 0.522
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    298,
+                    75,
+                    857,
+                    75,
+                    857,
+                    105,
+                    298,
+                    105
+                ],
+                "score": 0.311
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    832.0,
+                    2085.0,
+                    870.0,
+                    2085.0,
+                    870.0,
+                    2124.0,
+                    832.0,
+                    2124.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    298.0,
+                    73.0,
+                    857.0,
+                    73.0,
+                    857.0,
+                    108.0,
+                    298.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    298.0,
+                    73.0,
+                    857.0,
+                    73.0,
+                    857.0,
+                    108.0,
+                    298.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    233.0,
+                    1404.0,
+                    233.0,
+                    1404.0,
+                    264.0,
+                    296.0,
+                    264.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    324.0,
+                    261.0,
+                    1344.0,
+                    261.0,
+                    1344.0,
+                    292.0,
+                    324.0,
+                    292.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    309.0,
+                    1404.0,
+                    309.0,
+                    1404.0,
+                    343.0,
+                    294.0,
+                    343.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    324.0,
+                    337.0,
+                    877.0,
+                    337.0,
+                    877.0,
+                    367.0,
+                    324.0,
+                    367.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 11,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    291,
+                    1403,
+                    291,
+                    1403,
+                    415,
+                    298,
+                    415
+                ],
+                "score": 0.966
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    307,
+                    947,
+                    1404,
+                    947,
+                    1404,
+                    1328,
+                    307,
+                    1328
+                ],
+                "score": 0.954
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    306,
+                    494,
+                    1071,
+                    494,
+                    1071,
+                    841,
+                    306,
+                    841
+                ],
+                "score": 0.937
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    298,
+                    75,
+                    857,
+                    75,
+                    857,
+                    105,
+                    298,
+                    105
+                ],
+                "score": 0.899
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    836,
+                    2088,
+                    865,
+                    2088,
+                    865,
+                    2112,
+                    836,
+                    2112
+                ],
+                "score": 0.837
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    901,
+                    940,
+                    901,
+                    940,
+                    933,
+                    299,
+                    933
+                ],
+                "score": 0.558
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    302,
+                    229,
+                    1230,
+                    229,
+                    1230,
+                    261,
+                    302,
+                    261
+                ],
+                "score": 0.443
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    299,
+                    901,
+                    940,
+                    901,
+                    940,
+                    933,
+                    299,
+                    933
+                ],
+                "score": 0.419
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    302,
+                    229,
+                    1230,
+                    229,
+                    1230,
+                    261,
+                    302,
+                    261
+                ],
+                "score": 0.391
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    296,
+                    449,
+                    936,
+                    449,
+                    936,
+                    481,
+                    296,
+                    481
+                ],
+                "score": 0.35
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    690,
+                    1199,
+                    715,
+                    1199,
+                    715,
+                    1225,
+                    690,
+                    1225
+                ],
+                "score": 0.63,
+                "latex": "^ +"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    385,
+                    657,
+                    399,
+                    657,
+                    399,
+                    682,
+                    385,
+                    682
+                ],
+                "score": 0.49,
+                "latex": "i"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    539,
+                    1197,
+                    577,
+                    1197,
+                    577,
+                    1226,
+                    539,
+                    1226
+                ],
+                "score": 0.44,
+                "latex": "\\gets"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    691,
+                    717,
+                    714,
+                    717,
+                    714,
+                    742,
+                    691,
+                    742
+                ],
+                "score": 0.39,
+                "latex": "^ +"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    425,
+                    714,
+                    1069,
+                    714,
+                    1069,
+                    748,
+                    425,
+                    748
+                ],
+                "score": 0.39,
+                "latex": "o u t p u t [ o ] \\gets o u t p u t [ o ] + \\mathrm { c o n v o l v e } ( i n p u t [ i ] , W [ i , o , . . . ] )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    511,
+                    1172,
+                    528,
+                    1172,
+                    528,
+                    1193,
+                    511,
+                    1193
+                ],
+                "score": 0.33,
+                "latex": "o"
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    425,
+                    714,
+                    1068,
+                    714,
+                    1068,
+                    749,
+                    425,
+                    749
+                ],
+                "score": 0.31,
+                "latex": "o u t p u t [ o ] \\gets o u t p u t [ o ] + \\mathrm { c o n v o l v e } ( i n p u t [ i ] , W [ i , o , . . . ] )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    386,
+                    1138,
+                    399,
+                    1138,
+                    399,
+                    1163,
+                    386,
+                    1163
+                ],
+                "score": 0.29,
+                "latex": "i"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    428,
+                    690,
+                    445,
+                    690,
+                    445,
+                    712,
+                    428,
+                    712
+                ],
+                "score": 0.27,
+                "latex": "o"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    298.0,
+                    73.0,
+                    857.0,
+                    73.0,
+                    857.0,
+                    108.0,
+                    298.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    832.0,
+                    2084.0,
+                    870.0,
+                    2084.0,
+                    870.0,
+                    2123.0,
+                    832.0,
+                    2123.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    226.0,
+                    1234.0,
+                    226.0,
+                    1234.0,
+                    267.0,
+                    294.0,
+                    267.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    893.0,
+                    946.0,
+                    893.0,
+                    946.0,
+                    942.0,
+                    293.0,
+                    942.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    446.0,
+                    939.0,
+                    446.0,
+                    939.0,
+                    486.0,
+                    297.0,
+                    486.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    293.0,
+                    1403.0,
+                    293.0,
+                    1403.0,
+                    326.0,
+                    296.0,
+                    326.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    320.0,
+                    1404.0,
+                    320.0,
+                    1404.0,
+                    359.0,
+                    292.0,
+                    359.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    353.0,
+                    1404.0,
+                    353.0,
+                    1404.0,
+                    388.0,
+                    293.0,
+                    388.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    383.0,
+                    1177.0,
+                    383.0,
+                    1177.0,
+                    419.0,
+                    294.0,
+                    419.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    306.0,
+                    939.0,
+                    431.0,
+                    939.0,
+                    431.0,
+                    979.0,
+                    306.0,
+                    979.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    307.0,
+                    972.0,
+                    577.0,
+                    972.0,
+                    577.0,
+                    1007.0,
+                    307.0,
+                    1007.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    307.0,
+                    1001.0,
+                    601.0,
+                    1001.0,
+                    601.0,
+                    1036.0,
+                    307.0,
+                    1036.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    307.0,
+                    1032.0,
+                    868.0,
+                    1032.0,
+                    868.0,
+                    1067.0,
+                    307.0,
+                    1067.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    304.0,
+                    1061.0,
+                    1403.0,
+                    1061.0,
+                    1403.0,
+                    1101.0,
+                    304.0,
+                    1101.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    342.0,
+                    1094.0,
+                    650.0,
+                    1094.0,
+                    650.0,
+                    1129.0,
+                    342.0,
+                    1129.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    306.0,
+                    1134.0,
+                    385.0,
+                    1134.0,
+                    385.0,
+                    1169.0,
+                    306.0,
+                    1169.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    400.0,
+                    1134.0,
+                    660.0,
+                    1134.0,
+                    660.0,
+                    1169.0,
+                    400.0,
+                    1169.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    307.0,
+                    1165.0,
+                    339.0,
+                    1165.0,
+                    339.0,
+                    1198.0,
+                    307.0,
+                    1198.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    382.0,
+                    1165.0,
+                    510.0,
+                    1165.0,
+                    510.0,
+                    1200.0,
+                    382.0,
+                    1200.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    529.0,
+                    1165.0,
+                    1144.0,
+                    1165.0,
+                    1144.0,
+                    1200.0,
+                    529.0,
+                    1200.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    307.0,
+                    1196.0,
+                    341.0,
+                    1196.0,
+                    341.0,
+                    1228.0,
+                    307.0,
+                    1228.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    422.0,
+                    1194.0,
+                    538.0,
+                    1194.0,
+                    538.0,
+                    1234.0,
+                    422.0,
+                    1234.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    578.0,
+                    1194.0,
+                    689.0,
+                    1194.0,
+                    689.0,
+                    1234.0,
+                    578.0,
+                    1234.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    716.0,
+                    1194.0,
+                    1129.0,
+                    1194.0,
+                    1129.0,
+                    1234.0,
+                    716.0,
+                    1234.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    304.0,
+                    1225.0,
+                    341.0,
+                    1225.0,
+                    341.0,
+                    1260.0,
+                    304.0,
+                    1260.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    383.0,
+                    1225.0,
+                    479.0,
+                    1225.0,
+                    479.0,
+                    1259.0,
+                    383.0,
+                    1259.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    299.0,
+                    1256.0,
+                    438.0,
+                    1256.0,
+                    438.0,
+                    1290.0,
+                    299.0,
+                    1290.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    302.0,
+                    1287.0,
+                    511.0,
+                    1287.0,
+                    511.0,
+                    1319.0,
+                    302.0,
+                    1319.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    308.0,
+                    490.0,
+                    429.0,
+                    490.0,
+                    429.0,
+                    526.0,
+                    308.0,
+                    526.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    309.0,
+                    522.0,
+                    575.0,
+                    522.0,
+                    575.0,
+                    554.0,
+                    309.0,
+                    554.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    308.0,
+                    551.0,
+                    600.0,
+                    551.0,
+                    600.0,
+                    585.0,
+                    308.0,
+                    585.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    307.0,
+                    582.0,
+                    867.0,
+                    582.0,
+                    867.0,
+                    617.0,
+                    307.0,
+                    617.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    308.0,
+                    613.0,
+                    894.0,
+                    613.0,
+                    894.0,
+                    647.0,
+                    308.0,
+                    647.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    312.0,
+                    655.0,
+                    384.0,
+                    655.0,
+                    384.0,
+                    685.0,
+                    312.0,
+                    685.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    400.0,
+                    655.0,
+                    659.0,
+                    655.0,
+                    659.0,
+                    685.0,
+                    400.0,
+                    685.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    308.0,
+                    682.0,
+                    341.0,
+                    682.0,
+                    341.0,
+                    718.0,
+                    308.0,
+                    718.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    383.0,
+                    684.0,
+                    427.0,
+                    684.0,
+                    427.0,
+                    715.0,
+                    383.0,
+                    715.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    446.0,
+                    684.0,
+                    718.0,
+                    684.0,
+                    718.0,
+                    715.0,
+                    446.0,
+                    715.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    307.0,
+                    714.0,
+                    340.0,
+                    714.0,
+                    340.0,
+                    747.0,
+                    307.0,
+                    747.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    306.0,
+                    746.0,
+                    340.0,
+                    746.0,
+                    340.0,
+                    778.0,
+                    306.0,
+                    778.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    381.0,
+                    744.0,
+                    480.0,
+                    744.0,
+                    480.0,
+                    777.0,
+                    381.0,
+                    777.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    773.0,
+                    439.0,
+                    773.0,
+                    439.0,
+                    808.0,
+                    297.0,
+                    808.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    300.0,
+                    807.0,
+                    512.0,
+                    807.0,
+                    512.0,
+                    839.0,
+                    300.0,
+                    839.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    893.0,
+                    946.0,
+                    893.0,
+                    946.0,
+                    942.0,
+                    293.0,
+                    942.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    226.0,
+                    1234.0,
+                    226.0,
+                    1234.0,
+                    267.0,
+                    294.0,
+                    267.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 12,
+            "width": 1700,
+            "height": 2200
+        }
+    }
+]
\ No newline at end of file
diff --git a/parse/train/lHmhW2zmVN/lHmhW2zmVN.md b/parse/train/lHmhW2zmVN/lHmhW2zmVN.md
new file mode 100644
index 0000000000000000000000000000000000000000..bdd1acd38efe42d6bfac2197fc08f964bfdc64f0
--- /dev/null
+++ b/parse/train/lHmhW2zmVN/lHmhW2zmVN.md
@@ -0,0 +1,248 @@
+# Attention over learned object embeddings enables complex visual reasoning
+
+David Ding Felix Hill Adam Santoro Malcolm Reynolds Matt Botvinick DeepMind London, United Kingdom   
+{fding, felixhill, adamsantoro, mareynolds, botvinick}@google.com
+
+# Abstract
+
+Neural networks have achieved success in a wide array of perceptual tasks but often fail at tasks involving both perception and higher-level reasoning. On these more challenging tasks, bespoke approaches (such as modular symbolic components, independent dynamics models or semantic parsers) targeted towards that specific type of task have typically performed better. The downside to these targeted approaches, however, is that they can be more brittle than general-purpose neural networks, requiring significant modification or even redesign according to the particular task at hand. Here, we propose a more general neural-network-based approach to dynamic visual reasoning problems that obtains state-of-the-art performance on three different domains, in each case outperforming bespoke modular approaches tailored specifically to the task. Our method relies on learned object-centric representations, self-attention and self-supervised dynamics learning, and all three elements together are required for strong performance to emerge. The success of this combination suggests that there may be no need to trade off flexibility for performance on problems involving spatio-temporal or causal-style reasoning. With the right soft biases and learning objectives in a neural network we may be able to attain the best of both worlds.
+
+# 1 Introduction
+
+Despite the popularity of artificial neural networks, a body of recent work has focused on their limitations as models of cognition and reasoning. Experiments with dynamical reasoning datasets such as CLEVRER [41], CATER [12], and ACRE [44] show that neural networks can fail to adequately reason about the spatio-temporal, compositional or causal structure of visual scenes. On CLEVRER, where models must answer questions about the dynamics of colliding objects, previous experiments show that neural networks can adequately describe the video, but fail when asked to predict, explain, or consider counterfactual possibilities. Similarly, on CATER, an object-tracking task, models have trouble tracking the movement of objects when they are hidden in a container. Finally, on ACRE, a dataset testing for causal inference, popular models only learned correlations between visual scenes and not the deeper causal logic.
+
+Failures such as these on reasoning (rather than perception) problems have motivated the adoption of pipeline-style approaches that combine a general purpose neural network (such as a convolutional block) with a task-specific module that builds in the core logic of the task. For example, on CLEVRER the NS-DR method [41] applies a hand-coded symbolic logic engine (that has the core logic of CLEVRER built-in) to the outputs of a “perceptual” neural front-end, achieving better results than neural network baselines, particularly on counterfactual and explanatory problems. One limitation of these pipeline approaches, however, is that they are typically created with a single problem or problem domain in mind, and may not apply out-of-the-box to other related problems. For example, to apply NS-DR to CATER, the entire symbolic module needs to be rewritten to handle the new interactions and task logic of CATER: the custom logic to handle collisions and object removal must be replaced with new custom logic to handle occlusions and grid-resolution, and these changes require further modifications to the perceptual front-end to output data in a new format. This brittleness is not exclusive to symbolic approaches. While Hungarian-matching between object embeddings may be well-suited for object-tracking tasks [45], it is not obvious how it would help for causal inference tasks.
+
+Here, we describe a more general neural-network-based approach to visual spatio-temporal reasoning problems, which does not rely on task-specific integration of modular components. In place of these components, our model relies on three key aspects:
+
+• Self-attention to effectively integrate information over time   
+• Soft-discretization of the input at the most informative level of abstraction – above pixels and local features, and below entire frames—corresponding approximately to ‘objects’   
+Self-supervised learning, i.e. requiring the model to infer masked out objects, to extract more information about dynamics from each sample.
+
+While many past models have applied each individual ingredient separately (including on the tasks we study), we show that it is the combination of all three ingredients in the right way that allows our model to succeed.
+
+The resulting model, which we call Aloe (Attention over Learned Object Embeddings), outperforms both pipeline and neural-network-based approaches on three different task domains designed to test physical and dynamical reasoning from pixel inputs. We highlight our key results here:
+
+• CLEVRER (explanatory, predictive, and counterfactual reasoning): Aloe achieves significantly higher accuracy than both more task-specific, modular approaches, and previous neural network methods on all question types. On counterfactual questions, thought to be most challenging for neural-only architectures, we achieve $75 \%$ vs $46 \%$ accuracy for more specialised methods. • CATER (object-permanence): Aloe achieves accuracy exceeding or matching other current models. Notably, the strongest alternative models were expressly designed for object-tracking, whereas our architecture is applicable without modification to other reasoning tasks as well. ACRE (causal-inference “beyond the simple strategy of inducing causal relationships by covariation” [44]): Overall, Aloe achieves $94 \%$ vs the $67 \%$ accuracy achieved by the top neuro-symbolic model. On the most challenging tasks, we achieve, for “backward-blocking” inference, $9 4 . 4 8 \%$ (vs $1 6 . 0 6 \%$ by the best modular, neuro-symbolic systems), and, for “screenoff” inference, $9 8 . 9 7 \%$ (vs ${ \bf0 . 0 0 \% }$ by a CNN-BERT baseline).
+
+As we have emphasized, the previous best performing models for each task all contain task-specific design elements, whereas Aloe can be applied to all the tasks without modification. On CLEVRER, we also show that Aloe matches the performance of the previous best models with $40 \%$ less training data, which demonstrates that our approach is data-efficient as well as performant.
+
+# 2 Methods
+
+A guiding motivation for the design of Aloe is the converging evidence for the value of self-attention mechanisms operating on a finite sequences of discrete entities. Written language is inherently discrete and hence is well-suited to self-attention-based approaches. In other domains, such as raw audio or vision, it is less clear how to leverage self-attention. We hypothesize that the application of self-attention-based models to visual tasks could benefit from an approximate ‘discretization’ process, and determining the right level of discretization is an important choice that can significantly affect model performance.
+
+At the finest level, data could simply be discretized into pixels (as is already the case for most machine-processed visual data). Pixels are too fine-grained for many applications, however—for one, the memory required to support self-attention across all pixels is prohibitive. Partly for this reason, coarser representations, such as the downsampled “hyper-pixel” outputs of a convolutional network, are often used instead (e.g. [27, 43]). In the case of videos, previous work considered even coarser discretization schemes, such as frame or subclip level representations [35].
+
+![](images/fb75c68d2f8c0da36c1f5ab1ea3c806bea16df8a72eccf44e82552a1d3195e0e.jpg)  
+Figure 1: A schematic of the model architecture. See the main text for details.
+
+The neuroscience literature, however, suggests that biological visual systems infer and exploit the existence of objects, rather than spatial or temporal blocks with artificial boundaries [5, 30, 32]. Because objects are the atomic units of physical interactions, it makes sense to discretize on the level of objects. Numerous object segmentation algorithms have been proposed [15, 19, 29]. We chose to use MONet, an unsupervised object segmentation algorithm [2]. Because MONet is unsupervised, we can train it directly in our domain of interest without the need for object segmentation labels. We emphasize that our choice of MONet is an implementation detail, and in Appendix B, we show that our framework of attention over learned object embeddings also works with other object-segmentation schemes. We also do not need to place strong demands on the object segmentation algorithm, e.g. for it to produce aligned output or to have a built-in dynamics model.
+
+To segment each frame into object representations, MONet uses a recurrent attention network to obtain a set of $N _ { o }$ “object attention masks” ( $\mathrm { \Delta } N _ { o }$ is a fixed parameter). Each attention mask represents the probability that any given pixel belongs to that mask’s object. The pixels assigned to the mask are encoded into latent variables with means $\mu _ { t i } \in \mathbb { R } ^ { d }$ , where $i$ indexes the object slot and $t$ the frame. These means are used as the object embeddings in Aloe. More details are provided in Appendix A.1.
+
+The self-attention component is a transformer model [37] operating on a sequence of vectors in $\mathbb { R } ^ { d }$ : the object representations $\mu _ { t i }$ for all $t$ and $i$ , a trainable vector $\mathbf { \bar { \mathit { C L S } } } \in \mathbf { \bar { \mathbb { R } } ^ { d } }$ used to generate classification results (analogous to the CLS token in BERT [9]), and (for CLEVRER) the embedded words $\mathbf { w } _ { i }$ from the question (and choice for multiple choice questions). For the object representations $\mu _ { t i }$ and word embeddings $\mathbf { w } _ { i }$ , we append a two-dimensional one-hot vector to $\mu _ { t i }$ and $\mathbf { w } _ { i }$ to indicate whether the input is a word or an object. Because the transformer is shared between the modalities, information can flow between objects and words to solve the task, as we show in Section 3.1.
+
+We pass this sequence of vectors through a transformer with $N _ { T }$ layers. All inputs are first projected (via a linear layer and ReLU activation) to $\mathbb { R } ^ { N _ { H } \times d }$ , where $N _ { H }$ is the number of self-attention heads. We add a relative sinusoidal positional encoding at each layer of the transformer to give the model knowledge of the word and frame order [7]. The transformed value of $C L S$ is passed through an MLP (with one hidden layer of size $N _ { H }$ ) to generate the final answer. A schema of our architecture is shown in Figure 1.
+
+Note that in the model presented above (which we call global attention), the transformer sees no distinction between objects of different frames (other than through the position encoding). Another intuitive choice, which we call hierarchical attention, is to have one transformer acting on the objects of each frame independently, and another transformer acting on the concatenated outputs of the first transformer (this temporal division of input data is commonly used, e.g. in [35]). In pseudo-code, global attention can be expressed as
+
+![](images/8b8974247844831b2889dcf142e48a4d89360d0c23450734c7b5e9c93722a7eb.jpg)  
+Figure 2: Different masking schemes for self-supervised learning applied to Aloe.
+
+out $=$ transformer(reshape(objects, [B, F \* N, D])
+
+and hiearchical attention as
+
+out $=$ transformer1(reshape(objects, [B \* F, N, D])) out $=$ transformer2(reshape(out, [B, F, N $\star$ D])) .
+
+We study the importance of global attention (objects as the atomic entities) vs hierarchical attention (objects, and subsequently frames as the atomic entities). The comparison is shown in Table 1.
+
+# 2.1 Self-supervised learning
+
+We explored whether self-supervised learning could improve the performance of Aloe beyond the benefits conveyed by object-level representation, i.e. in ways that support the model’s interpretation of scene dynamics rather than just via improved perception of static observations. Our approach is inspired by the loss used in BERT [9], where a transformer model is trained to predict certain words that are masked from the input. In our case, we mask object embeddings, and train the model to infer the content of the masked object representations using its knowledge of unmasked objects.
+
+Concretely, during training, we multiply each MONet latent $\mu _ { t i }$ by a masking indicator, $m _ { t i } \in \{ 0 , 1 \}$ . Let $\mu _ { t i } ^ { \prime }$ be the transformed value of $m _ { t i } \mu _ { t i }$ after passing through the transformer. We expect the transformer to understand the underlying dynamics of the video, so that the masked out slot $\mu _ { t i }$ could be predicted from $\mu _ { t i } ^ { \prime }$ . To guide the transformer in learning effective representations capable of this type of dynamics prediction, we add an auxiliary loss:
+
+$$
+{ \mathrm { a u x i l i a r y ~ l o s s } } = \sum _ { t , i } { \tau _ { t i } l \left( f ( \mu _ { t i } ^ { \prime } ) , \mu \right) } ,
+$$
+
+where $f$ is a learned linear mapping to $\mathbb { R } ^ { d }$ , $l$ a loss function, and $\tau _ { t i } \in \{ 0 , 1 \}$ are one-hot indicator variables identifying the prediction targets (not necessarily just the masked out entries, since the prediction targets could be a subset of the masked out entries). We propagate gradients only to the parameters of $f$ and the transformer and not to the learned word and $C L S$ embeddings. This auxiliary loss is added to the main classification loss with weighting $\lambda$ , and both losses are minimized simultaneously by the optimizer. We do not pretrain the model with only the auxiliary loss.
+
+We tested two different loss functions for $l$ , an L2 loss and a contrastive loss (formulas given in Appendix A.2), and six different masking schemes (settings of $m _ { t i }$ and $\tau _ { t i }$ ), as illustrated in Figure 2. This exploration was motivated by the observation that video inputs at adjacent timesteps are highly correlated in a way that adjacent words are not. We thus hypothesized that BERT-style prediction of adjacent words might not be optimal. A different masking strategy, in which prediction targets are separated from the context by more than a single timestep, may stimulate capacity in the network to acquire knowledge that permits context-based unrolls and better long-horizon predictions.
+
+The simplest approach would be to set $m _ { t i } = 1$ uniformly at random across $t$ and $i$ , fixing the expected proportion of the $m _ { t i }$ set to 1 (schema $^ b$ in Figure 2). The targets would simply be the unmasked slots, $\tau _ { t i } = 1 - m _ { t i }$ . One potential problem with this approach is that multiple objects could be masked out in a single frame. MONet can unpredictably switch object-to-slot assignments multiple times in a single video. If multiple slots are masked out, the transformer cannot determine with certainty which missing object to assign to each slot. Thus, the auxiliary loss could penalize the model even if it predicted all the objects correctly. To avoid this problem, we also try constraining the mask such that exactly one slot is masked out per frame (schema $a$ ).
+
+To pose harder prediction challenges, we can add a buffer between the context (where $m _ { t i } = 1$ ) and the infilling targets (where $\tau _ { t i } = 1$ ). For $t$ in this buffer zone, both $m _ { t i } = 0$ and $\tau _ { t i } = 0$ (schemas $c { - } f )$ . We choose a single cutoff $T$ randomly, and we set $m _ { t i } = 0$ for $t < T$ and $m _ { t i } = 1$ for $t \geq T$ . In the presence of this buffer, we compared prediction (where the context is strictly before the targets; schema $c , d$ ) versus infilling (where the context surrounds the targets; schema $e , f ,$ ). We also compared setting the targets as individual objects (schema $c , e$ ) versus targets as all objects in the scene (schema $d , f )$ . We visually inspect the efficacy of this self-supervised loss in encouraging better representations (beyond improvements of scores on tasks) in Appendix D.
+
+# 3 Experiments
+
+We tested Aloe on three datasets, CLEVRER [41], CATER [12], and ACRE [44]. For each dataset, we pretrained a MONet model on individual frames. More training details and a table of hyperparameters are given in Appendix A.3; these hyperparameters were obtained through a hyperparameter sweep. All error bars are standard deviations computed over at least 5 random seeds.
+
+# 3.1 CLEVRER
+
+CLEVRER features videos of CLEVR objects [21] that move and collide with each other. For each video, several questions are posed to test the model’s understanding of the scene. Unlike most other visual question answering datasets, which test for only descriptive understanding (“what happened?”), CLEVRER poses other more complex questions, including explanatory questions (“why did something happen?”), predictive questions (“what will happen next?”), and counterfactual questions (“what would happen in a unseen circumstance?”) [41].
+
+We compare Aloe to state-of-the-art models reported in the literature: MAC $\left( \mathrm { V } + \right)$ and NS-DR [41], as well as the DCL model [6] (simultaneous to our work). MAC $\left( \mathrm { V } + \right)$ (based on the MAC network [20]) is an end-to-end network augmented with object information and trained using ground truth labels for object segmentation masks and features (e.g. color, shape). NS-DR and DCL are hybrid models that apply a symbolic logic engine to outputs of various neural networks. The neural networks are used to detect objects, predict dynamics, and parse the question into a program, and the symbolic executor runs the parsed program to obtain the final output. NS-DR is trained using ground truth labels and ground truth parsed programs, while DCL requires only the ground truth parsed programs.
+
+Table 1 shows the result of Aloe compared to these models. Across all categories, Aloe significantly outperforms the previous best models. Moreover, compared to the other models, Aloe does not use any labeled data other than the correct answer for the questions, nor does it require pretraining on any other dataset. Aloe also was not specifically designed for this task, and it straightforwardly generalizes to other tasks as well, such as CATER [12] and ACRE [44]. We provide a few sample model classifications on a randomly selected set of videos and questions in Appendix E.1 and detailed analysis of counterfactual questions in Appendix C. These examples suggest qualitatively that, for most instances where the model was incorrect, humans would plausibly furnish the same answer.
+
+Attention analysis (More analyses are given in Appendix D) We analyzed the cross-modal attention between question-words and the MONet objects. For each word, we determined the object that attended to that word with highest weight (for one head in the last layer). In the visualization below, the bounding boxes show the objects found by MONet, and each word is colored according to the
+
+Table 1: Performance (per question accuracy) on CLEVRER of Aloe compared to results from literature and to ablations: 1) MLP instead of self-attention; 2) ResNet superpixels instead of MONet objects; 3) hierarchical frame-level and intra-frame attention instead of global cross-frame object attention; 4) no auxiliary loss.   
+
+<table><tr><td>Model</td><td>Descriptive</td><td>Explanatory</td><td>Predictive</td><td>Counterfactual</td></tr><tr><td>MAC (V+)</td><td>86.4</td><td>22.3</td><td>42.9</td><td>25.1</td></tr><tr><td>NS-DR</td><td>88.1</td><td>79.6</td><td>68.7</td><td>42.2</td></tr><tr><td>DCL</td><td>90.7</td><td>82.8</td><td>82.0</td><td>46.5</td></tr><tr><td>Aloe</td><td>94.0 ± 0.4</td><td>96.0 ± 0.6</td><td>87.5 ± 3.0</td><td>75.6 ± 3.8</td></tr><tr><td>Aloe- self-attention+MLP</td><td>45.4</td><td>16.0</td><td>27.7</td><td>9.9</td></tr><tr><td>Aloe- object-repr. +ResNet</td><td>74.9</td><td>66.1</td><td>58.3</td><td>32.4</td></tr><tr><td>Aloe - global + hierarchical attn.</td><td>80.6</td><td>87.4</td><td>73.5</td><td>55.1</td></tr><tr><td>Aloe- self-supervised loss</td><td>91.0</td><td>92.8</td><td>82.8</td><td>68.7</td></tr></table>
+
+object that attended to it with highest weight (black represents a MONet slot without any objects).   
+We observe that generally, objects attend heavily to the words that describe them.
+
+![](images/229b16c4fda57fa6be8830ae071ce6a0ec54820c6c821a7f96d21f78abffa24d.jpg)
+
+Q: If the cylinder is removed, which event will not happen?
+
+1. The brown object collides with the green object.   
+2. The yellow object and the metal cube collide.   
+3. The yellow cube collides with the green object.
+
+We also looked at the objects that were most heavily attended upon in determining the final answer. The image below illustrates the attention weights for the $C L S$ token attending on each object (for one head in the last layer), when the model is tasked with assessing the first choice of the question above. The bounding boxes show the two most heavily attended upon objects for one transformer head. We observe that this head focuses on the green and brown objects (asked about in choice 1), but switches its focus to the cyan cylinder when it looks like the cylinder might collide with the cubes and change the outcome.
+
+![](images/f9813d948a0c16a84c18a5a51d92a5b3b2434d16be376473d12c2adad84e34ba.jpg)
+
+Model ablation Table 1 shows the contributions of various components of Aloe. First, self-attention is necessary for solving this problem. For comparison, we replace Aloe’s transformer with four fully connected layers with 2048 units per layer1. We find that an MLP is unable to answer non-descriptive questions effectively, despite using more parameters (20M vs 15M parameters).
+
+Second, we verify that an object-based discretization scheme is essential to the performance of Aloe. We compare with a version of the architecture where the MONet object representations $\mu _ { t i }$ are replaced with ResNet hyperpixels as in Zambaldi et al. [43]. Concretely, we flatten the output of the final convolutional layer of the ResNet to obtain a sequence of feature vectors that is fed into the transformer as the discrete entities. To match MONet’s pretraining regimen, we pretrain the ResNet on CLEVR [21] by training an Aloe model (using a ResNet instead of MONet) on the CLEVR task and initializing the ResNet used in the CLEVRER task with these pre-trained weights. We find that an object level representation, such as one output by MONet, greatly outperforms the locality-aware but object-agnostic ResNet representation.
+
+We also observe the importance of global attention between all objects across all frames, compared to a hierarchical attention model where objects within a frame could attend to each other but frames could only attend to each other as an atomic entity. We hypothesize that global attention may be important because with hierarchical attention, objects in different frames can only attend to each other at the “frame” granularity. A cube attending to a cube in a different frame would then gather information about the other non-cube objects, muddling the resulting representation.
+
+![](images/3444a8a096a123941b20407702c07de6ac1b82d124cdf4ea73c3ff1606e480c9.jpg)  
+Figure 3: Accuracy with/without auxiliary loss for different proportions of CLEVRER (row 1) and CATER (row 2) training data. We also show comparisons with previous and concurrent work. For CLEVRER, the lighter yellow bar represents the best neurosymbolic model DCL, and the darker yellow bar represents the previous best distributed model, MAC $\left( \mathrm { V } + \right)$ . For CATER, the lighter yellow bar represents Hopper and the darker yellow bar represents $\mathrm { R 3 D + N L }$ , the best published results for the moving camera dataset.
+
+Finally, we see that an auxiliary self-supervised loss improves the performance of the model by between 4 and 6 percentage points, with the greatest improvement on the counterfactual questions.
+
+Self-supervision strategies We compared the various masking schemes and loss functions for our auxiliary loss; a detailed figure is provided in Appendix A (Figure 4). We find that for all question types in CLEVRER, an L2 loss performs better than a contrastive loss, and among the masking schemes, masking one object per frame is the most effective. This particular result runs counter to our hypothesis that predictions or infilling in which the target is temporally removed from the context could encourage the model to learn more about scene dynamics and object interactions than (BERT-style) local predictions of adjacent targets. Of course, there may be other settings or loss functions that reveal the benefits of non-local prediction or constrastive losses; we leave this investigation to future work.
+
+Data efficiency We investigated how model performance varies as a function of the number of labelled (question-answer) pairs it learns from. To do so, we train models on $N \%$ of the videos and their associated labeled data. We evaluate the effect of including the auxiliary self-supervised loss (applied to the entire dataset, not just the labelled portion) in this low data regime. This scenario, where unlabeled data is plentiful while labeled data is scarce, occurs frequently in practice, since collecting labeled data is much more expensive than collecting unlabeled data.
+
+Figure 3 shows that our best model reaches the approximate level of the previous state-of-the-art approaches using only $50 \% - 6 0 \%$ of the data. The self-supervised auxiliary loss makes a particular improvement to performance in low-data regimes. For instance, when trained on only $50 \%$ of the available labelled data, self-supervised learning enables the model to reach a performance of $37 \%$ on counterfactual questions (compared to $2 5 \%$ by MAC $( \mathrm { V } + )$ and $42 \%$ by NS-DR on the full dataset), while without self-supervision, the model only reaches a performance of $13 \%$ (compared to the $10 \%$ achieved by answering randomly [41]).
+
+Table 2: Performance on CATER of Aloe compared to the best results from literature. We report top 1 accuracy, top 5 accuracy, and L1 distance between the predicted grid cell and true grid cell. The labels (S) and (M) refer to static and moving cameras.   
+
+<table><tr><td>Model</td><td>Top 1 (S)</td><td>Top 5 (S)</td><td>L1(S)</td><td>Top 1 (M)</td><td>Top 5 (M)</td><td>L1 (M)</td></tr><tr><td>R3DLSTM R3D+NL LSTM OPNet</td><td>60.2 46.2 74.8 73.2</td><td>81.8 69.9 1 93.8</td><td>1.2 1.5 0.54 0.85</td><td>28.6 38.6 1 1</td><td>63.3 70.2 1 1</td><td>1.7 1.5 -</td></tr><tr><td>Hopper Aloe (no auxiliary) Aloe Aloe (with L1 loss)</td><td>60.5 70.6 74.0 ± 0.3</td><td>84.5 93.0 94.0 ± 0.4</td><td>0.90 0.53 0.44 ± 0.01</td><td>46.8 56.6 59.7 ± 0.5</td><td>75.1 87.0 90.1 ± 0.6</td><td>- 1.3 0.82 0.69 ± 0.01</td></tr></table>
+
+# 3.2 CATER
+
+In a second experiment, we tested Aloe on CATER, a widely-used object-tracking dataset [12, 14, 31, 45]. In CATER, objects from the CLEVR dataset [21] move and potentially occlude other objects, and the goal is to predict the location of a target object (called the snitch) in the final frame. Because the snitch could be occluded by multiple objects that could move in the meantime, a successful model must be sensitive to notions of object permanence. CATER also includes a moving camera variant, which introduces additional complexities for the model.
+
+Concretely, CATER is setup as a classification challenge. Objects are located in an xyz coordinate system, where x and y range from -3 to 3. The xy plane is divided into a 6 by 6 grid, and the task is to predict the grid index of the snitch in the final frame. For Aloe, we use a classification loss (cross entropy over the 36 possible grid indices) and an L1 loss (L1 distance between predicted grid cell and the true grid cell).
+
+Table 2 shows Aloe compared to state-of-the-art models in the literature on both static and moving camera videos. R3D and R3D NL are the strongest two models evaluated by Girdhar and Ramanan [12]. OPNet, or the Object Permanence Network [31], is an architecture with inductive biases designed for object tracking tasks; it was trained with extra supervised labels, namely the bounding boxes for all objects (including occluded ones). Hopper is a multi-hop transformer model developed simultaneously with this work [45]. One key component of Hopper is Hungarian matching between objects of different frames, a strong inductive bias for object tracking.
+
+We train Aloe simultaneously on both static and moving camera videos. Aloe outperforms the R3D models for both static and moving cameras. We also ran Aloe with an additional auxiliary loss consisting of the L1 distance between the predicted cell and the actual cell. With this additional loss, we get comparable results in the moving camera case as the R3D models for the static camera case. Moreover, we achieve comparable accuracy as OPNet for accuracy and L1 distance, despite requiring less supervision to train. Appendix E.2 gives a few sample outputs from Aloe; in particular we note that it is able to find the target object in several cases where the object was occluded, demonstrating that Aloe is able to do some level of object tracking. Finally,we find that an auxiliary self-supervised loss helps the model perform well in the low data regime for CATER as well, as shown in Figure 3.
+
+# 3.3 ACRE
+
+Finally, we measured Aloe’s performance on ACRE, a causal induction dataset inspired by the Blicket task from developmental psychology [13, 44]. ACRE is divided into a set of problems. In each problem, certain objects are chosen to be “Blickets”, and this assignment changes across problems. Each problem presents a context of six images to the model, where different objects are placed on a Blicket machine that lights up if one of those objects is a Blicket. The model is asked whether an unseen combination of objects will light up the Blicket machine. Besides “yes” and “no”, a third possible answer is “undetermined”, which is the case if it is impossible to determine for certain if the objects will light up the machine. Correct inference goes beyond mere correlation: even if every context scene involving object A has a lit-up machine, A’s Blicketness is still uncertain if each of those scenes can potentially be explained by another object (deduction of A’s Blicketness is backward-blocked).
+
+Table 3: Performance on ACRE of Aloe compared to the best results from Zhang et al. [44], split across inference type $\mathrm { { D . R = } } 1$ Direct, I.D $\Vdash$ Indirect, ${ \bf { S . O } } \mathrm { { = } } \mathrm { { ; } }$ Screen-Off, B.B $\ c =$ Backwards Blocking) and generalization type ( $\mathrm { C } =$ Compositional, ${ \bf S } =$ Systematic).   
+
+<table><tr><td>Model</td><td>All (C)</td><td>D.R.</td><td>I.D.</td><td>S.0.</td><td>B.B.</td><td>All (S)</td><td>D.R.</td><td>1.D.</td><td>S.0.</td><td>B.B</td></tr><tr><td>CNN-BERT</td><td>43.79</td><td>54.07</td><td>46.88</td><td>40.57</td><td>28.79</td><td>39.93</td><td>55.97</td><td>68.25</td><td>0.00</td><td>45.59</td></tr><tr><td>NS-OPT</td><td>69.04</td><td>92.5</td><td>76.05</td><td>88.33</td><td>13.48</td><td>67.44</td><td>94.73</td><td>88.38</td><td>82.76</td><td>16.06</td></tr><tr><td>Aloe</td><td>91.76</td><td>97.14</td><td>90.8</td><td>96.8</td><td>78.81</td><td>93.90</td><td>97.18</td><td>71.24</td><td>98.97</td><td>94.48</td></tr></table>
+
+Inference problems in ACRE are categorized by reasoning type: reasoning from direct evidence (one of the context frames show the query objects on a machine), reasoning from indirect evidence (Blicketness must be deduced by combining evidence from several frames), screened-off reasoning (presence of non-Blickets do not matter if a single Blicket is present), and backward-blocked reasoning (Blicketness cannot be deduced due to confounding variables). Please see Zhang et al. [44] for a more detailed discussion of these reasoning types.
+
+Table 3 show Aloe performance compared to a CNN-BERT baseline and to NS-OPT, a neuro-symbolic model introduced in Zhang et al. [44]. Aloe outperforms all extant models for almost all reasoning types and train-test splits. We did not need to do any tuning to apply our model to ACRE—settings from CATER yielded the reported results on the first attempt. Contrary to widely-held opinions that neural networks cannot generalize, Aloe generalizes in scenarios where the training and test sets contain different visual features (compositional split) or different numbers of activated machines in the context (systematic split). Moreover, Aloe achieved by far the best performance on the backwardblocking task, which requires the model to “go beyond the simple covariation strategy to discover the hidden causal relations” [44], dispelling the notion that neural networks can only find correlation. Comparison with NS-OPT (which uses object representations) and CNN-BERT (which uses attention) shows that neither object representations nor attention alone is sufficient for the task; combining these two ideas, as done in Aloe for instance, is essential for this complex reasoning task as well.
+
+# 4 Related work
+
+Self-attention for reasoning Various studies have shown that transformers [37] can manipulate symbolic data in a manner traditionally associated with symbolic computation. For example, in Lample and Charton [23], a transformer model learned to do symbolic integration and solve ordinary differential equations symbolically, tasks traditionally reserved for symbolic computer algebra systems. Similarly, in Hahn et al. [17], a transformer model learned to solve formulas in propositional logic and demonstrated some degree of generalization to out of distribution formulas. Finally, Brown et al. [1] showed that a transformer trained for language modeling can also do simple analogical reasoning tasks without explicit training. Although these models do not necessarily beat carefully tuned symbolic algorithms in all cases (especially on out of distribution data), they are an important motivation for our proposed recipe for attaining strong reasoning capabilities from self-attention-based models on visually grounded tasks.
+
+Object representations A wide body of research points to the importance of object segmentation and representation learning (see e.g. Garnelo and Shanahan [11] for a discussion). Various methods have been proposed for object detection and feature extraction [2, 10, 15, 19, 25, 26, 29]. Past research have also investigated using object based representations in downstream tasks [8, 28].
+
+Self-supervised learning Another line of research concerns learning good representations through self-supervised learning, with an unsupervised auxiliary loss to encourage the discovery of better representations. These better representations could lead to improved performance on supervised tasks, especially when labeled data is scarce. In Devlin et al. [9], for instance, an auxiliary infill loss allows the BERT model to benefit from pretraining on a large corpus of unlabeled data. Our approach to object-centric self-supervised learning is heavily inspired by the BERT infilling loss. Other studies have shown similar benefits to auxiliary learning in vision as well [4, 16, 18]. These works apply various forms of contrastive losses to predict scene dynamics, and the better representations that result carry downstream benefits to supervised and reinforcement learning tasks.
+
+Vision and language in self-attention models Recently, many works have emerged on applying transformer models to visual and multimodal data, for static images [24, 27, 33, 36] and videos [34, 35, 43]. These approaches combine the output of convolutional networks with language in various ways using self-attention. While these previous works focused on popular visual question answering tasks, which typically consist of descriptive questions only [41], we focus on understanding deeper causal dynamics of videos. Together with these works, we provide more evidence that self-attention between visual and language elements enables good performance on a diverse set of tasks.
+
+In addition, while the use of object representations for discretization in tasks involving static images is becoming more popular, the right way to discretize videos is less clear. We provide strong evidence in the form of ablation studies for architectural decisions that we claim are essential for higher reasoning for this type of data: visual elements should correspond to physical objects in the videos and inter-frame attention between sub-frame entities (as opposed to inter-frame attention of entire frames) is crucial. We also demonstrate the success of using unsupervised object segmentation methods as opposed to the supervised methods used in past work.
+
+# 5 Conclusion
+
+We have presented Aloe, a model that obtains state-of-the-art performance on three different task domains involving spatiotemporal reasoning about objects. In each of these tasks, previous state-ofthe-art results were established by models with modular, task-specific components. Aloe, by contrast, is a unified solution to all three domains. Its flexibility comes from a reliance on only soft biases and learning objectives: self-attention over learned object embeddings and self-supervised learning of dynamics. We believe the simplicity of this approach is its strength, and hope that this fact, together with the provided code, makes it easy for others to adopt and apply to arbitrary spatio-temporal reasoning problems.
+
+On many of these spatio-temporal reasoning problems, previous state-of-the-art was achieved by neuro-symbolic models [6, 11, 40, 41, 44]. Compared to neuro-symbolic models, Aloe can more easily be adapted to other tasks. Indeed, the symbolic components of neuro-symbolic models are often task-specific and not straightforwardly applicable to other tasks. Neuro-symbolic models do have a few advantages, however. First, they are often easier to interpret. Despite the insights that can be gleaned from Aloe’s attention weights, these soft computations are harder to interpret than the explicit symbolic computation found in neuro-symbolic models. Moreover, neuro-symbolic models can be structured in a more modular fashion, which can enable effective generalization to sub-tasks of the task on which the model was trained [6].
+
+Aloe also has some important limitations. First, it has only been applied to synthetic datasets. This limitation is mainly due to the lack of real-world datasets that test for higher-order spatiotemporal reasoning, although we are excited that new datasets such as Traffic QA will be released soon [38]. Second, while the domains where Aloe is applied have been widely adopted and well-received by the research community, it remains possible that they do not evaluate the capacities that they aim to evaluate because of hidden biases or other factors. Regardless, we hope that this work stimulates the design and development of more challenging tasks that more closely approximate the ultimate goal of human or super-human-level visual, spatiotemporal and causal reasoning. Finally, from an ethical point of view, our model may share the common drawback of deep-learning models in perpetuating biases found in the training data, especially when applied to real world data. Development of causal reasoning models could also invite problematic applications involving automated assignment of blame.
+
+References   
+[1] Tom B Brown, Benjamin Mann, Nick Ryder, Melanie Subbiah, Jared Kaplan, Prafulla Dhariwal, Arvind Neelakantan, Pranav Shyam, Girish Sastry, Amanda Askell, et al. Language models are few-shot learners. arXiv preprint arXiv:2005.14165, 2020.   
+[2] Chris P. Burgess, Loic Matthey, Nick Watters, Rishabh Kabra, Irina Higgins, Matt Botvinick, and Alex Lerchner. MONet: Unsupervised scene decomposition and representation. arXiv preprint arXiv:1901.11390, 2019.   
+[3] Nicolas Carion, Francisco Massa, Gabriel Synnaeve, Nicolas Usunier, Alexander Kirillov, and Sergey Zagoruyko. End-to-end object detection with transformers. In ECCV, 2020.   
+[4] Ting Chen, Simon Kornblith, Mohammad Norouzi, and Geoffrey Hinton. A simple framework for contrastive learning of visual representations. arXiv preprint arXiv:2002.05709, 2020.   
+[5] Zhe Chen. Object-based attention: A tutorial review. Attention, Perception, & Psychophysics, 74:784–802, 2012. URL https://doi.org/10.3758/s13414-012-0322-z.   
+[6] Zhenfang Chen, Jiayuan Mao, Jiajun Wu, Kwan-Yee Kenneth Wong, Joshua B. Tenenbaum, and Chuang Gan. Grounding physical object and event concepts through dynamic visual reasoning. In International Conference on Learning Representations, 2021. URL https: //arxiv.org/pdf/2103.16564.pdf.   
+[7] Zihang Dai, Zhilin Yang, Yiming Yang, William W Cohen, Jaime Carbonell, Quoc V Le, and Ruslan Salakhutdinov. Transformer-xl: Attentive language models beyond a fixed-length context. arXiv preprint arXiv:1901.02860, 2019.   
+[8] Mikyas T. Desta, Larry Chen, and Tomasz Kornuta. Object-based reasoning in VQA. CoRR, abs/1801.09718, 2018. URL http://arxiv.org/abs/1801.09718.   
+[9] Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. BERT: Pre-training of deep bidirectional transformers for language understanding. arXiv preprint arXiv:1810.04805, 2018.   
+[10] Yilun Du, Kevin Smith, Tomer Ulman, Joshua Tenenbaum, and Jiajun Wu. Unsupervised discovery of 3d physical objects from video. arXiv preprint arXiv:2007.12348, 2020.   
+[11] Marta Garnelo and Murray Shanahan. Reconciling deep learning with symbolic artificial intelligence: representing objects and relations. Current Opinion in Behavioral Sciences, 29: 17–23, 2019.   
+[12] Rohit Girdhar and Deva Ramanan. CATER: A diagnostic dataset for Compositional Actions and TEmporal Reasoning. In ICLR, 2020.   
+[13] Alison Gopnik and David Sobel. Detecting blickets: how young children use information about novel causal powers in categorization and induction. Child Dev, pages 1205–22, 2000. doi: 10.1111/1467-8624.00224.   
+[14] Anirudh Goyal, Aniket Didolkar, Alex Lamb, Kartikeya Badola, Nan Rosemary Ke, Nasim Rahaman, Jonathan Binas, Charles Blundell, Michael Mozer, and Yoshua Bengio. Coordination among neural modules through a shared global workspace. CoRR, abs/2103.01197, 2021. URL https://arxiv.org/abs/2103.01197.   
+[15] Klaus Greff, Raphaël Lopez Kaufman, Rishabh Kabra, Nick Watters, Christopher Burgess, Daniel Zoran, Loïc Matthey, Matthew Botvinick, and Alexander Lerchner. Multi-object representation learning with iterative variational inference. CoRR, abs/1903.00450, 2019. URL http://arxiv.org/abs/1903.00450.   
+[16] Karol Gregor, Danilo Jimenez Rezende, Frederic Besse, Yan Wu, Hamza Merzic, and Aaron van den Oord. Shaping belief states with generative environment models for rl. In H. Wallach, H. Larochelle, A. Beygelzimer, F. d’ Alché-Buc, E. Fox, and R. Garnett, editors, Advances in Neural Information Processing Systems 32, pages 13475–13487. Curran Associates, Inc., 2019. URL http://papers.nips.cc/paper/ 9503-shaping-belief-states-with-generative-environment-models-f pdf.
+
+[17] Christopher Hahn, Frederik Schmitt, Jens U. Kreber, Markus N. Rabe, and Bernd Finkbeiner. Transformers generalize to the semantics of logics. arXiv preprint arXiv:2003.04218, 2020.
+
+[18] Tengda Han, Weidi Xie, and Andrew Zisserman. Video representation learning by dense predictive coding. In Workshop on Large Scale Holistic Video Understanding, ICCV, 2019.
+
+[19] Kaiming He, Georgia Gkioxari, Piotr Dollár, and Ross B. Girshick. Mask R-CNN. CoRR, abs/1703.06870, 2017. URL http://arxiv.org/abs/1703.06870.
+
+[20] Drew A Hudson and Christopher D Manning. Compositional attention networks for machine reasoning. 2018.
+
+[21] Justin Johnson, Bharath Hariharan, Laurens van der Maaten, Li Fei-Fei, C. Lawrence Zitnick, and Ross B. Girshick. CLEVR: A diagnostic dataset for compositional language and elementary visual reasoning. CoRR, abs/1612.06890, 2016. URL http://arxiv.org/abs/1612. 06890.
+
+[22] Diederik P Kingma and Jimmy Ba. Adam: A method for stochastic optimization. arXiv preprint arXiv:1412.6980, 2014.
+
+[23] Guillaume Lample and François Charton. Deep learning for symbolic mathematics. In International Conference on Learning Representations, 2020. URL https://openreview.net/ forum?id ${ . } = { }$ S1eZYeHFDS.
+
+[24] Liunian Harold Li, Mark Yatskar, Da Yin, Cho-Jui Hsieh, and Kai-Wei Chang. Visualbert: A simple and performant baseline for vision and language. arXiv preprint arXiv:1908.03557, 2019.
+
+[25] Zhixuan Lin, Yi-Fu Wu, Skand Vishwanath Peri, Weihao Sun, Gautam Singh, Fei Deng, Jindong Jiang, and Sungjin Ahn. SPACE: Unsupervised object-oriented scene representation via spatial attention and decomposition. In International Conference on Learning Representations, 2020. URL https://openreview.net/forum?id $\underline { { \underline { { \mathbf { \Pi } } } } }$ rkl03ySYDH.
+
+[26] Francesco Locatello, Dirk Weissenborn, Thomas Unterthiner, Aravindh Mahendran, Georg Heigold, Jakob Uszkoreit, Alexey Dosovitskiy, and Thomas Kipf1. Object-Centric Learning with Slot Attention. In Advances in Neural Information Processing Systems (NIPS), 2020.
+
+[27] Jiasen Lu, Dhruv Batra, Devi Parikh, and Stefan Lee. Vilbert: Pretraining task-agnostic visiolinguistic representations for vision-and-language tasks. arXiv preprint arXiv:1908.02265, 2019.
+
+[28] David Raposo, Adam Santoro, David G. T. Barrett, Razvan Pascanu, Timothy P. Lillicrap, and Peter W. Battaglia. Discovering objects and their relations from entangled scene representations. CoRR, abs/1702.05068, 2017. URL http://arxiv.org/abs/1702.05068.
+
+[29] Shaoqing Ren, Kaiming He, Ross B. Girshick, and Jian Sun. Faster R-CNN: towards realtime object detection with region proposal networks. CoRR, abs/1506.01497, 2015. URL http://arxiv.org/abs/1506.01497.
+
+[30] Pieter R. Roelfsema, Victor A. Lamme, and Henk Spekreijse. Object-based attention in the primary visual cortex of the macaque monkey. Nature, 395:376–381, 1998. doi: 10.1038/26475.
+
+[31] Aviv Shamsian, Ofri Kleinfeld, Amir Globerson, and Gal Chechik. Learning object permanence from video. arXiv preprint arXiv:2003.10469, 2020.
+
+[32] Elizabeth Spelke. Core knowledge. The American psychologist, 55:1233–43, 12 2000. doi: 10.1037/0003-066X.55.11.1233.
+
+[33] Weijie Su, Xizhou Zhu, Yue Cao, Bin Li, Lewei Lu, Furu Wei, and Jifeng Dai. Vl-bert: Pretraining of generic visual-linguistic representations. In International Conference on Learning Representations, 2020. URL https://openreview.net/forum?id $\underline { { \underline { { \mathbf { \Pi } } } } } =$ SygXPaEYvH.
+
+[34] Chen Sun, Fabien Baradel, Kevin Murphy, and Cordelia Schmid. Contrastive bidirectional transformer for temporal representation learning. CoRR, abs/1906.05743, 2019. URL http: //arxiv.org/abs/1906.05743.
+
+[35] Chen Sun, Austin Myers, Carl Vondrick, Kevin Murphy, and Cordelia Schmid. Videobert: A joint model for video and language representation learning. In Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), October 2019.
+
+[36] Hao Tan and Mohit Bansal. Lxmert: Learning cross-modality encoder representations from transformers. In Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing, 2019.
+
+[37] Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N Gomez, Łukasz Kaiser, and Illia Polosukhin. Attention is all you need. In Advances in neural information processing systems, pages 5998–6008, 2017.
+
+[38] Li Xu, He Huang, and Jun Liu. Sutd-trafficqa: A question answering benchmark and an efficient network for video reasoning over traffic events. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2021.
+
+[39] Kexin Yi, Jiajun Wu, Chuang Gan, Antonio Torralba, Pushmeet Kohli, and Joshua B Tenenbaum. Neural-Symbolic VQA: Disentangling Reasoning from Vision and Language Understanding. In Advances in Neural Information Processing Systems (NIPS), 2018.
+
+[40] Kexin Yi, Jiajun Wu, Chuang Gan, Antonio Torralba, Pushmeet Kohli, and Joshua B. Tenenbaum. Neural-symbolic vqa: Disentangling reasoning from vision and language understanding. In Advances in Neural Information Processing Systems, pages 1039–1050, 2018.
+
+[41] Kexin Yi, Chuang Gan, Yunzhu Li, Pushmeet Kohli, Jiajun Wu, Antonio Torralba, and Joshua B. Tenenbaum. CLEVRER: Collision events for video representation and reasoning. In International Conference on Learning Representations, 2020. URL https: //openreview.net/forum?id $\underline { { \underline { { \mathbf { \Pi } } } } } =$ HkxYzANYDB.
+
+[42] Yang You, Jing Li, Sashank Reddi, Jonathan Hseu, Sanjiv Kumar, Srinadh Bhojanapalli, Xiaodan Song, James Demmel, and Cho-Jui Hsieh. Large batch optimization for deep learning: Training BERT in 76 minutes. Technical Report UCB/EECS-2019-103, EECS Department, University of California, Berkeley, Jun 2019. URL http://www2.eecs.berkeley.edu/Pubs/ TechRpts/2019/EECS-2019-103.html.
+
+[43] Vinicius Zambaldi, David Raposo, Adam Santoro, Victor Bapst, Yujia Li, Igor Babuschkin, Karl Tuyls, David Reichert, Timothy Lillicrap, Edward Lockhart, Murray Shanahan, Victoria Langston, Razvan Pascanu, Matthew Botvinick, Oriol Vinyals, and Peter Battaglia. Deep reinforcement learning with relational inductive biases. In International Conference on Learning Representations, 2019. URL https://openreview.net/forum?id $\underline { { \underline { { \mathbf { \Pi } } } } } =$ HkxaFoC9KQ.
+
+[44] Chi Zhang, Baoxiong Jia, Mark Edmonds, Song-Chun Zhu, and Yixin Zhu. Acre: Abstract causal reasoning beyond covariation. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2021.
+
+[45] Honglu Zhou, Asim Kadav, Farley Lai, Alexandru Niculescu-Mizil, Martin Renqiang Min, Mubbasir Kapadia, and Hans Peter Graf. Hopper: Multi-hop transformer for spatiotemporal reasoning. In International Conference on Learning Representations, 2021. URL https: //arxiv.org/abs/2103.10574.
\ No newline at end of file
diff --git a/parse/train/lHmhW2zmVN/lHmhW2zmVN_content_list.json b/parse/train/lHmhW2zmVN/lHmhW2zmVN_content_list.json
new file mode 100644
index 0000000000000000000000000000000000000000..7983bc88f7b05735ee7b7f9ec02afc8d4fad37bf
--- /dev/null
+++ b/parse/train/lHmhW2zmVN/lHmhW2zmVN_content_list.json
@@ -0,0 +1,1278 @@
+[
+    {
+        "type": "text",
+        "text": "Attention over learned object embeddings enables complex visual reasoning ",
+        "text_level": 1,
+        "bbox": [
+            197,
+            122,
+            802,
+            172
+        ],
+        "page_idx": 0
+    },
+    {
+        "type": "text",
+        "text": "David Ding Felix Hill Adam Santoro Malcolm Reynolds Matt Botvinick DeepMind London, United Kingdom   \n{fding, felixhill, adamsantoro, mareynolds, botvinick}@google.com ",
+        "bbox": [
+            181,
+            224,
+            821,
+            282
+        ],
+        "page_idx": 0
+    },
+    {
+        "type": "text",
+        "text": "Abstract ",
+        "text_level": 1,
+        "bbox": [
+            462,
+            318,
+            535,
+            334
+        ],
+        "page_idx": 0
+    },
+    {
+        "type": "text",
+        "text": "Neural networks have achieved success in a wide array of perceptual tasks but often fail at tasks involving both perception and higher-level reasoning. On these more challenging tasks, bespoke approaches (such as modular symbolic components, independent dynamics models or semantic parsers) targeted towards that specific type of task have typically performed better. The downside to these targeted approaches, however, is that they can be more brittle than general-purpose neural networks, requiring significant modification or even redesign according to the particular task at hand. Here, we propose a more general neural-network-based approach to dynamic visual reasoning problems that obtains state-of-the-art performance on three different domains, in each case outperforming bespoke modular approaches tailored specifically to the task. Our method relies on learned object-centric representations, self-attention and self-supervised dynamics learning, and all three elements together are required for strong performance to emerge. The success of this combination suggests that there may be no need to trade off flexibility for performance on problems involving spatio-temporal or causal-style reasoning. With the right soft biases and learning objectives in a neural network we may be able to attain the best of both worlds. ",
+        "bbox": [
+            233,
+            348,
+            766,
+            582
+        ],
+        "page_idx": 0
+    },
+    {
+        "type": "text",
+        "text": "1 Introduction ",
+        "text_level": 1,
+        "bbox": [
+            174,
+            607,
+            310,
+            625
+        ],
+        "page_idx": 0
+    },
+    {
+        "type": "text",
+        "text": "Despite the popularity of artificial neural networks, a body of recent work has focused on their limitations as models of cognition and reasoning. Experiments with dynamical reasoning datasets such as CLEVRER [41], CATER [12], and ACRE [44] show that neural networks can fail to adequately reason about the spatio-temporal, compositional or causal structure of visual scenes. On CLEVRER, where models must answer questions about the dynamics of colliding objects, previous experiments show that neural networks can adequately describe the video, but fail when asked to predict, explain, or consider counterfactual possibilities. Similarly, on CATER, an object-tracking task, models have trouble tracking the movement of objects when they are hidden in a container. Finally, on ACRE, a dataset testing for causal inference, popular models only learned correlations between visual scenes and not the deeper causal logic. ",
+        "bbox": [
+            174,
+            638,
+            825,
+            777
+        ],
+        "page_idx": 0
+    },
+    {
+        "type": "text",
+        "text": "Failures such as these on reasoning (rather than perception) problems have motivated the adoption of pipeline-style approaches that combine a general purpose neural network (such as a convolutional block) with a task-specific module that builds in the core logic of the task. For example, on CLEVRER the NS-DR method [41] applies a hand-coded symbolic logic engine (that has the core logic of CLEVRER built-in) to the outputs of a “perceptual” neural front-end, achieving better results than neural network baselines, particularly on counterfactual and explanatory problems. One limitation of these pipeline approaches, however, is that they are typically created with a single problem or problem domain in mind, and may not apply out-of-the-box to other related problems. For example, to apply NS-DR to CATER, the entire symbolic module needs to be rewritten to handle the new interactions and task logic of CATER: the custom logic to handle collisions and object removal must be replaced with new custom logic to handle occlusions and grid-resolution, and these changes require further modifications to the perceptual front-end to output data in a new format. This brittleness is not exclusive to symbolic approaches. While Hungarian-matching between object embeddings may be well-suited for object-tracking tasks [45], it is not obvious how it would help for causal inference tasks. ",
+        "bbox": [
+            174,
+            784,
+            825,
+            867
+        ],
+        "page_idx": 0
+    },
+    {
+        "type": "text",
+        "text": "",
+        "bbox": [
+            174,
+            90,
+            825,
+            215
+        ],
+        "page_idx": 1
+    },
+    {
+        "type": "text",
+        "text": "Here, we describe a more general neural-network-based approach to visual spatio-temporal reasoning problems, which does not rely on task-specific integration of modular components. In place of these components, our model relies on three key aspects: ",
+        "bbox": [
+            176,
+            222,
+            825,
+            263
+        ],
+        "page_idx": 1
+    },
+    {
+        "type": "text",
+        "text": "• Self-attention to effectively integrate information over time   \n• Soft-discretization of the input at the most informative level of abstraction – above pixels and local features, and below entire frames—corresponding approximately to ‘objects’   \nSelf-supervised learning, i.e. requiring the model to infer masked out objects, to extract more information about dynamics from each sample. ",
+        "bbox": [
+            192,
+            275,
+            825,
+            357
+        ],
+        "page_idx": 1
+    },
+    {
+        "type": "text",
+        "text": "While many past models have applied each individual ingredient separately (including on the tasks we study), we show that it is the combination of all three ingredients in the right way that allows our model to succeed. ",
+        "bbox": [
+            178,
+            368,
+            823,
+            410
+        ],
+        "page_idx": 1
+    },
+    {
+        "type": "text",
+        "text": "The resulting model, which we call Aloe (Attention over Learned Object Embeddings), outperforms both pipeline and neural-network-based approaches on three different task domains designed to test physical and dynamical reasoning from pixel inputs. We highlight our key results here: ",
+        "bbox": [
+            178,
+            416,
+            823,
+            458
+        ],
+        "page_idx": 1
+    },
+    {
+        "type": "text",
+        "text": "• CLEVRER (explanatory, predictive, and counterfactual reasoning): Aloe achieves significantly higher accuracy than both more task-specific, modular approaches, and previous neural network methods on all question types. On counterfactual questions, thought to be most challenging for neural-only architectures, we achieve $75 \\%$ vs $46 \\%$ accuracy for more specialised methods. • CATER (object-permanence): Aloe achieves accuracy exceeding or matching other current models. Notably, the strongest alternative models were expressly designed for object-tracking, whereas our architecture is applicable without modification to other reasoning tasks as well. ACRE (causal-inference “beyond the simple strategy of inducing causal relationships by covariation” [44]): Overall, Aloe achieves $94 \\%$ vs the $67 \\%$ accuracy achieved by the top neuro-symbolic model. On the most challenging tasks, we achieve, for “backward-blocking” inference, $9 4 . 4 8 \\%$ (vs $1 6 . 0 6 \\%$ by the best modular, neuro-symbolic systems), and, for “screenoff” inference, $9 8 . 9 7 \\%$ (vs ${ \\bf0 . 0 0 \\% }$ by a CNN-BERT baseline). ",
+        "bbox": [
+            192,
+            469,
+            826,
+            647
+        ],
+        "page_idx": 1
+    },
+    {
+        "type": "text",
+        "text": "As we have emphasized, the previous best performing models for each task all contain task-specific design elements, whereas Aloe can be applied to all the tasks without modification. On CLEVRER, we also show that Aloe matches the performance of the previous best models with $40 \\%$ less training data, which demonstrates that our approach is data-efficient as well as performant. ",
+        "bbox": [
+            176,
+            659,
+            825,
+            714
+        ],
+        "page_idx": 1
+    },
+    {
+        "type": "text",
+        "text": "2 Methods ",
+        "text_level": 1,
+        "bbox": [
+            174,
+            734,
+            279,
+            751
+        ],
+        "page_idx": 1
+    },
+    {
+        "type": "text",
+        "text": "A guiding motivation for the design of Aloe is the converging evidence for the value of self-attention mechanisms operating on a finite sequences of discrete entities. Written language is inherently discrete and hence is well-suited to self-attention-based approaches. In other domains, such as raw audio or vision, it is less clear how to leverage self-attention. We hypothesize that the application of self-attention-based models to visual tasks could benefit from an approximate ‘discretization’ process, and determining the right level of discretization is an important choice that can significantly affect model performance. ",
+        "bbox": [
+            174,
+            766,
+            825,
+            863
+        ],
+        "page_idx": 1
+    },
+    {
+        "type": "text",
+        "text": "At the finest level, data could simply be discretized into pixels (as is already the case for most machine-processed visual data). Pixels are too fine-grained for many applications, however—for one, the memory required to support self-attention across all pixels is prohibitive. Partly for this reason, coarser representations, such as the downsampled “hyper-pixel” outputs of a convolutional network, are often used instead (e.g. [27, 43]). In the case of videos, previous work considered even coarser discretization schemes, such as frame or subclip level representations [35]. ",
+        "bbox": [
+            176,
+            869,
+            825,
+            911
+        ],
+        "page_idx": 1
+    },
+    {
+        "type": "image",
+        "img_path": "images/fb75c68d2f8c0da36c1f5ab1ea3c806bea16df8a72eccf44e82552a1d3195e0e.jpg",
+        "image_caption": [
+            "Figure 1: A schematic of the model architecture. See the main text for details. "
+        ],
+        "image_footnote": [],
+        "bbox": [
+            251,
+            93,
+            761,
+            300
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "text",
+        "text": "",
+        "bbox": [
+            176,
+            366,
+            825,
+            409
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "text",
+        "text": "The neuroscience literature, however, suggests that biological visual systems infer and exploit the existence of objects, rather than spatial or temporal blocks with artificial boundaries [5, 30, 32]. Because objects are the atomic units of physical interactions, it makes sense to discretize on the level of objects. Numerous object segmentation algorithms have been proposed [15, 19, 29]. We chose to use MONet, an unsupervised object segmentation algorithm [2]. Because MONet is unsupervised, we can train it directly in our domain of interest without the need for object segmentation labels. We emphasize that our choice of MONet is an implementation detail, and in Appendix B, we show that our framework of attention over learned object embeddings also works with other object-segmentation schemes. We also do not need to place strong demands on the object segmentation algorithm, e.g. for it to produce aligned output or to have a built-in dynamics model. ",
+        "bbox": [
+            174,
+            415,
+            825,
+            553
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "text",
+        "text": "To segment each frame into object representations, MONet uses a recurrent attention network to obtain a set of $N _ { o }$ “object attention masks” ( $\\mathrm { \\Delta } N _ { o }$ is a fixed parameter). Each attention mask represents the probability that any given pixel belongs to that mask’s object. The pixels assigned to the mask are encoded into latent variables with means $\\mu _ { t i } \\in \\mathbb { R } ^ { d }$ , where $i$ indexes the object slot and $t$ the frame. These means are used as the object embeddings in Aloe. More details are provided in Appendix A.1. ",
+        "bbox": [
+            174,
+            559,
+            825,
+            628
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "text",
+        "text": "The self-attention component is a transformer model [37] operating on a sequence of vectors in $\\mathbb { R } ^ { d }$ : the object representations $\\mu _ { t i }$ for all $t$ and $i$ , a trainable vector $\\mathbf { \\bar { \\mathit { C L S } } } \\in \\mathbf { \\bar { \\mathbb { R } } ^ { d } }$ used to generate classification results (analogous to the CLS token in BERT [9]), and (for CLEVRER) the embedded words $\\mathbf { w } _ { i }$ from the question (and choice for multiple choice questions). For the object representations $\\mu _ { t i }$ and word embeddings $\\mathbf { w } _ { i }$ , we append a two-dimensional one-hot vector to $\\mu _ { t i }$ and $\\mathbf { w } _ { i }$ to indicate whether the input is a word or an object. Because the transformer is shared between the modalities, information can flow between objects and words to solve the task, as we show in Section 3.1. ",
+        "bbox": [
+            174,
+            635,
+            825,
+            732
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "text",
+        "text": "We pass this sequence of vectors through a transformer with $N _ { T }$ layers. All inputs are first projected (via a linear layer and ReLU activation) to $\\mathbb { R } ^ { N _ { H } \\times d }$ , where $N _ { H }$ is the number of self-attention heads. We add a relative sinusoidal positional encoding at each layer of the transformer to give the model knowledge of the word and frame order [7]. The transformed value of $C L S$ is passed through an MLP (with one hidden layer of size $N _ { H }$ ) to generate the final answer. A schema of our architecture is shown in Figure 1. ",
+        "bbox": [
+            174,
+            738,
+            825,
+            821
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "text",
+        "text": "Note that in the model presented above (which we call global attention), the transformer sees no distinction between objects of different frames (other than through the position encoding). Another intuitive choice, which we call hierarchical attention, is to have one transformer acting on the objects of each frame independently, and another transformer acting on the concatenated outputs of the first transformer (this temporal division of input data is commonly used, e.g. in [35]). In pseudo-code, global attention can be expressed as ",
+        "bbox": [
+            174,
+            828,
+            825,
+            911
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "image",
+        "img_path": "images/8b8974247844831b2889dcf142e48a4d89360d0c23450734c7b5e9c93722a7eb.jpg",
+        "image_caption": [
+            "Figure 2: Different masking schemes for self-supervised learning applied to Aloe. "
+        ],
+        "image_footnote": [],
+        "bbox": [
+            174,
+            98,
+            820,
+            325
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "out $=$ transformer(reshape(objects, [B, F \\* N, D]) ",
+        "bbox": [
+            207,
+            385,
+            642,
+            398
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "and hiearchical attention as ",
+        "bbox": [
+            174,
+            405,
+            354,
+            419
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "out $=$ transformer1(reshape(objects, [B \\* F, N, D])) out $=$ transformer2(reshape(out, [B, F, N $\\star$ D])) . ",
+        "bbox": [
+            207,
+            425,
+            655,
+            459
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "We study the importance of global attention (objects as the atomic entities) vs hierarchical attention (objects, and subsequently frames as the atomic entities). The comparison is shown in Table 1. ",
+        "bbox": [
+            171,
+            465,
+            825,
+            494
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "2.1 Self-supervised learning ",
+        "text_level": 1,
+        "bbox": [
+            174,
+            512,
+            380,
+            527
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "We explored whether self-supervised learning could improve the performance of Aloe beyond the benefits conveyed by object-level representation, i.e. in ways that support the model’s interpretation of scene dynamics rather than just via improved perception of static observations. Our approach is inspired by the loss used in BERT [9], where a transformer model is trained to predict certain words that are masked from the input. In our case, we mask object embeddings, and train the model to infer the content of the masked object representations using its knowledge of unmasked objects. ",
+        "bbox": [
+            173,
+            537,
+            825,
+            622
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "Concretely, during training, we multiply each MONet latent $\\mu _ { t i }$ by a masking indicator, $m _ { t i } \\in \\{ 0 , 1 \\}$ . Let $\\mu _ { t i } ^ { \\prime }$ be the transformed value of $m _ { t i } \\mu _ { t i }$ after passing through the transformer. We expect the transformer to understand the underlying dynamics of the video, so that the masked out slot $\\mu _ { t i }$ could be predicted from $\\mu _ { t i } ^ { \\prime }$ . To guide the transformer in learning effective representations capable of this type of dynamics prediction, we add an auxiliary loss: ",
+        "bbox": [
+            173,
+            627,
+            825,
+            698
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "equation",
+        "img_path": "images/ec34b10eb5dbd048dcd818839d9d4c52bdbb686a32df86c0e869749c1ec07022.jpg",
+        "text": "$$\n{ \\mathrm { a u x i l i a r y ~ l o s s } } = \\sum _ { t , i } { \\tau _ { t i } l \\left( f ( \\mu _ { t i } ^ { \\prime } ) , \\mu \\right) } ,\n$$",
+        "text_format": "latex",
+        "bbox": [
+            377,
+            707,
+            619,
+            742
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "where $f$ is a learned linear mapping to $\\mathbb { R } ^ { d }$ , $l$ a loss function, and $\\tau _ { t i } \\in \\{ 0 , 1 \\}$ are one-hot indicator variables identifying the prediction targets (not necessarily just the masked out entries, since the prediction targets could be a subset of the masked out entries). We propagate gradients only to the parameters of $f$ and the transformer and not to the learned word and $C L S$ embeddings. This auxiliary loss is added to the main classification loss with weighting $\\lambda$ , and both losses are minimized simultaneously by the optimizer. We do not pretrain the model with only the auxiliary loss. ",
+        "bbox": [
+            173,
+            751,
+            825,
+            837
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "We tested two different loss functions for $l$ , an L2 loss and a contrastive loss (formulas given in Appendix A.2), and six different masking schemes (settings of $m _ { t i }$ and $\\tau _ { t i }$ ), as illustrated in Figure 2. This exploration was motivated by the observation that video inputs at adjacent timesteps are highly correlated in a way that adjacent words are not. We thus hypothesized that BERT-style prediction of adjacent words might not be optimal. A different masking strategy, in which prediction targets are separated from the context by more than a single timestep, may stimulate capacity in the network to acquire knowledge that permits context-based unrolls and better long-horizon predictions. ",
+        "bbox": [
+            174,
+            842,
+            825,
+            911
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "",
+        "bbox": [
+            173,
+            92,
+            823,
+            119
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "text",
+        "text": "The simplest approach would be to set $m _ { t i } = 1$ uniformly at random across $t$ and $i$ , fixing the expected proportion of the $m _ { t i }$ set to 1 (schema $^ b$ in Figure 2). The targets would simply be the unmasked slots, $\\tau _ { t i } = 1 - m _ { t i }$ . One potential problem with this approach is that multiple objects could be masked out in a single frame. MONet can unpredictably switch object-to-slot assignments multiple times in a single video. If multiple slots are masked out, the transformer cannot determine with certainty which missing object to assign to each slot. Thus, the auxiliary loss could penalize the model even if it predicted all the objects correctly. To avoid this problem, we also try constraining the mask such that exactly one slot is masked out per frame (schema $a$ ). ",
+        "bbox": [
+            174,
+            126,
+            825,
+            237
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "text",
+        "text": "To pose harder prediction challenges, we can add a buffer between the context (where $m _ { t i } = 1$ ) and the infilling targets (where $\\tau _ { t i } = 1$ ). For $t$ in this buffer zone, both $m _ { t i } = 0$ and $\\tau _ { t i } = 0$ (schemas $c { - } f )$ . We choose a single cutoff $T$ randomly, and we set $m _ { t i } = 0$ for $t < T$ and $m _ { t i } = 1$ for $t \\geq T$ . In the presence of this buffer, we compared prediction (where the context is strictly before the targets; schema $c , d$ ) versus infilling (where the context surrounds the targets; schema $e , f ,$ ). We also compared setting the targets as individual objects (schema $c , e$ ) versus targets as all objects in the scene (schema $d , f )$ . We visually inspect the efficacy of this self-supervised loss in encouraging better representations (beyond improvements of scores on tasks) in Appendix D. ",
+        "bbox": [
+            174,
+            242,
+            825,
+            353
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "text",
+        "text": "3 Experiments ",
+        "text_level": 1,
+        "bbox": [
+            174,
+            377,
+            312,
+            395
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "text",
+        "text": "We tested Aloe on three datasets, CLEVRER [41], CATER [12], and ACRE [44]. For each dataset, we pretrained a MONet model on individual frames. More training details and a table of hyperparameters are given in Appendix A.3; these hyperparameters were obtained through a hyperparameter sweep. All error bars are standard deviations computed over at least 5 random seeds. ",
+        "bbox": [
+            174,
+            412,
+            825,
+            468
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "text",
+        "text": "3.1 CLEVRER ",
+        "text_level": 1,
+        "bbox": [
+            174,
+            489,
+            292,
+            505
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "text",
+        "text": "CLEVRER features videos of CLEVR objects [21] that move and collide with each other. For each video, several questions are posed to test the model’s understanding of the scene. Unlike most other visual question answering datasets, which test for only descriptive understanding (“what happened?”), CLEVRER poses other more complex questions, including explanatory questions (“why did something happen?”), predictive questions (“what will happen next?”), and counterfactual questions (“what would happen in a unseen circumstance?”) [41]. ",
+        "bbox": [
+            174,
+            517,
+            825,
+            602
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "text",
+        "text": "We compare Aloe to state-of-the-art models reported in the literature: MAC $\\left( \\mathrm { V } + \\right)$ and NS-DR [41], as well as the DCL model [6] (simultaneous to our work). MAC $\\left( \\mathrm { V } + \\right)$ (based on the MAC network [20]) is an end-to-end network augmented with object information and trained using ground truth labels for object segmentation masks and features (e.g. color, shape). NS-DR and DCL are hybrid models that apply a symbolic logic engine to outputs of various neural networks. The neural networks are used to detect objects, predict dynamics, and parse the question into a program, and the symbolic executor runs the parsed program to obtain the final output. NS-DR is trained using ground truth labels and ground truth parsed programs, while DCL requires only the ground truth parsed programs. ",
+        "bbox": [
+            174,
+            607,
+            825,
+            718
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "text",
+        "text": "Table 1 shows the result of Aloe compared to these models. Across all categories, Aloe significantly outperforms the previous best models. Moreover, compared to the other models, Aloe does not use any labeled data other than the correct answer for the questions, nor does it require pretraining on any other dataset. Aloe also was not specifically designed for this task, and it straightforwardly generalizes to other tasks as well, such as CATER [12] and ACRE [44]. We provide a few sample model classifications on a randomly selected set of videos and questions in Appendix E.1 and detailed analysis of counterfactual questions in Appendix C. These examples suggest qualitatively that, for most instances where the model was incorrect, humans would plausibly furnish the same answer. ",
+        "bbox": [
+            174,
+            724,
+            825,
+            835
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "text",
+        "text": "Attention analysis (More analyses are given in Appendix D) We analyzed the cross-modal attention between question-words and the MONet objects. For each word, we determined the object that attended to that word with highest weight (for one head in the last layer). In the visualization below, the bounding boxes show the objects found by MONet, and each word is colored according to the ",
+        "bbox": [
+            176,
+            856,
+            825,
+            911
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "table",
+        "img_path": "images/b17fb2d33ac7519789a92be95f5ba1468a0123fe73f580cd362649f95d93168e.jpg",
+        "table_caption": [
+            "Table 1: Performance (per question accuracy) on CLEVRER of Aloe compared to results from literature and to ablations: 1) MLP instead of self-attention; 2) ResNet superpixels instead of MONet objects; 3) hierarchical frame-level and intra-frame attention instead of global cross-frame object attention; 4) no auxiliary loss. "
+        ],
+        "table_footnote": [],
+        "table_body": "<table><tr><td>Model</td><td>Descriptive</td><td>Explanatory</td><td>Predictive</td><td>Counterfactual</td></tr><tr><td>MAC (V+)</td><td>86.4</td><td>22.3</td><td>42.9</td><td>25.1</td></tr><tr><td>NS-DR</td><td>88.1</td><td>79.6</td><td>68.7</td><td>42.2</td></tr><tr><td>DCL</td><td>90.7</td><td>82.8</td><td>82.0</td><td>46.5</td></tr><tr><td>Aloe</td><td>94.0 ± 0.4</td><td>96.0 ± 0.6</td><td>87.5 ± 3.0</td><td>75.6 ± 3.8</td></tr><tr><td>Aloe- self-attention+MLP</td><td>45.4</td><td>16.0</td><td>27.7</td><td>9.9</td></tr><tr><td>Aloe- object-repr. +ResNet</td><td>74.9</td><td>66.1</td><td>58.3</td><td>32.4</td></tr><tr><td>Aloe - global + hierarchical attn.</td><td>80.6</td><td>87.4</td><td>73.5</td><td>55.1</td></tr><tr><td>Aloe- self-supervised loss</td><td>91.0</td><td>92.8</td><td>82.8</td><td>68.7</td></tr></table>",
+        "bbox": [
+            178,
+            87,
+            818,
+            217
+        ],
+        "page_idx": 5
+    },
+    {
+        "type": "text",
+        "text": "object that attended to it with highest weight (black represents a MONet slot without any objects).   \nWe observe that generally, objects attend heavily to the words that describe them. ",
+        "bbox": [
+            173,
+            306,
+            823,
+            335
+        ],
+        "page_idx": 5
+    },
+    {
+        "type": "image",
+        "img_path": "images/229b16c4fda57fa6be8830ae071ce6a0ec54820c6c821a7f96d21f78abffa24d.jpg",
+        "image_caption": [],
+        "image_footnote": [],
+        "bbox": [
+            178,
+            343,
+            380,
+            449
+        ],
+        "page_idx": 5
+    },
+    {
+        "type": "text",
+        "text": "Q: If the cylinder is removed, which event will not happen? ",
+        "bbox": [
+            431,
+            353,
+            759,
+            381
+        ],
+        "page_idx": 5
+    },
+    {
+        "type": "text",
+        "text": "1. The brown object collides with the green object.   \n2. The yellow object and the metal cube collide.   \n3. The yellow cube collides with the green object. ",
+        "bbox": [
+            431,
+            387,
+            763,
+            439
+        ],
+        "page_idx": 5
+    },
+    {
+        "type": "text",
+        "text": "We also looked at the objects that were most heavily attended upon in determining the final answer. The image below illustrates the attention weights for the $C L S$ token attending on each object (for one head in the last layer), when the model is tasked with assessing the first choice of the question above. The bounding boxes show the two most heavily attended upon objects for one transformer head. We observe that this head focuses on the green and brown objects (asked about in choice 1), but switches its focus to the cyan cylinder when it looks like the cylinder might collide with the cubes and change the outcome. ",
+        "bbox": [
+            173,
+            457,
+            826,
+            554
+        ],
+        "page_idx": 5
+    },
+    {
+        "type": "image",
+        "img_path": "images/f9813d948a0c16a84c18a5a51d92a5b3b2434d16be376473d12c2adad84e34ba.jpg",
+        "image_caption": [],
+        "image_footnote": [],
+        "bbox": [
+            176,
+            560,
+            821,
+            618
+        ],
+        "page_idx": 5
+    },
+    {
+        "type": "text",
+        "text": "Model ablation Table 1 shows the contributions of various components of Aloe. First, self-attention is necessary for solving this problem. For comparison, we replace Aloe’s transformer with four fully connected layers with 2048 units per layer1. We find that an MLP is unable to answer non-descriptive questions effectively, despite using more parameters (20M vs 15M parameters). ",
+        "bbox": [
+            174,
+            637,
+            825,
+            693
+        ],
+        "page_idx": 5
+    },
+    {
+        "type": "text",
+        "text": "Second, we verify that an object-based discretization scheme is essential to the performance of Aloe. We compare with a version of the architecture where the MONet object representations $\\mu _ { t i }$ are replaced with ResNet hyperpixels as in Zambaldi et al. [43]. Concretely, we flatten the output of the final convolutional layer of the ResNet to obtain a sequence of feature vectors that is fed into the transformer as the discrete entities. To match MONet’s pretraining regimen, we pretrain the ResNet on CLEVR [21] by training an Aloe model (using a ResNet instead of MONet) on the CLEVR task and initializing the ResNet used in the CLEVRER task with these pre-trained weights. We find that an object level representation, such as one output by MONet, greatly outperforms the locality-aware but object-agnostic ResNet representation. ",
+        "bbox": [
+            173,
+            699,
+            825,
+            824
+        ],
+        "page_idx": 5
+    },
+    {
+        "type": "text",
+        "text": "We also observe the importance of global attention between all objects across all frames, compared to a hierarchical attention model where objects within a frame could attend to each other but frames could only attend to each other as an atomic entity. We hypothesize that global attention may be important because with hierarchical attention, objects in different frames can only attend to each other at the “frame” granularity. A cube attending to a cube in a different frame would then gather information about the other non-cube objects, muddling the resulting representation. ",
+        "bbox": [
+            176,
+            830,
+            821,
+            872
+        ],
+        "page_idx": 5
+    },
+    {
+        "type": "image",
+        "img_path": "images/3444a8a096a123941b20407702c07de6ac1b82d124cdf4ea73c3ff1606e480c9.jpg",
+        "image_caption": [
+            "Figure 3: Accuracy with/without auxiliary loss for different proportions of CLEVRER (row 1) and CATER (row 2) training data. We also show comparisons with previous and concurrent work. For CLEVRER, the lighter yellow bar represents the best neurosymbolic model DCL, and the darker yellow bar represents the previous best distributed model, MAC $\\left( \\mathrm { V } + \\right)$ . For CATER, the lighter yellow bar represents Hopper and the darker yellow bar represents $\\mathrm { R 3 D + N L }$ , the best published results for the moving camera dataset. "
+        ],
+        "image_footnote": [],
+        "bbox": [
+            173,
+            87,
+            825,
+            328
+        ],
+        "page_idx": 6
+    },
+    {
+        "type": "text",
+        "text": "",
+        "bbox": [
+            174,
+            460,
+            825,
+            502
+        ],
+        "page_idx": 6
+    },
+    {
+        "type": "text",
+        "text": "Finally, we see that an auxiliary self-supervised loss improves the performance of the model by between 4 and 6 percentage points, with the greatest improvement on the counterfactual questions. ",
+        "bbox": [
+            174,
+            508,
+            823,
+            537
+        ],
+        "page_idx": 6
+    },
+    {
+        "type": "text",
+        "text": "Self-supervision strategies We compared the various masking schemes and loss functions for our auxiliary loss; a detailed figure is provided in Appendix A (Figure 4). We find that for all question types in CLEVRER, an L2 loss performs better than a contrastive loss, and among the masking schemes, masking one object per frame is the most effective. This particular result runs counter to our hypothesis that predictions or infilling in which the target is temporally removed from the context could encourage the model to learn more about scene dynamics and object interactions than (BERT-style) local predictions of adjacent targets. Of course, there may be other settings or loss functions that reveal the benefits of non-local prediction or constrastive losses; we leave this investigation to future work. ",
+        "bbox": [
+            174,
+            568,
+            825,
+            693
+        ],
+        "page_idx": 6
+    },
+    {
+        "type": "text",
+        "text": "Data efficiency We investigated how model performance varies as a function of the number of labelled (question-answer) pairs it learns from. To do so, we train models on $N \\%$ of the videos and their associated labeled data. We evaluate the effect of including the auxiliary self-supervised loss (applied to the entire dataset, not just the labelled portion) in this low data regime. This scenario, where unlabeled data is plentiful while labeled data is scarce, occurs frequently in practice, since collecting labeled data is much more expensive than collecting unlabeled data. ",
+        "bbox": [
+            173,
+            724,
+            825,
+            808
+        ],
+        "page_idx": 6
+    },
+    {
+        "type": "text",
+        "text": "Figure 3 shows that our best model reaches the approximate level of the previous state-of-the-art approaches using only $50 \\% - 6 0 \\%$ of the data. The self-supervised auxiliary loss makes a particular improvement to performance in low-data regimes. For instance, when trained on only $50 \\%$ of the available labelled data, self-supervised learning enables the model to reach a performance of $37 \\%$ on counterfactual questions (compared to $2 5 \\%$ by MAC $( \\mathrm { V } + )$ and $42 \\%$ by NS-DR on the full dataset), while without self-supervision, the model only reaches a performance of $13 \\%$ (compared to the $10 \\%$ achieved by answering randomly [41]). ",
+        "bbox": [
+            174,
+            814,
+            825,
+            911
+        ],
+        "page_idx": 6
+    },
+    {
+        "type": "table",
+        "img_path": "images/e9161209b964b799a3fdbb5b6fd16ab97b2278252e29ae3b7dbfb3b5a8d70f15.jpg",
+        "table_caption": [
+            "Table 2: Performance on CATER of Aloe compared to the best results from literature. We report top 1 accuracy, top 5 accuracy, and L1 distance between the predicted grid cell and true grid cell. The labels (S) and (M) refer to static and moving cameras. "
+        ],
+        "table_footnote": [],
+        "table_body": "<table><tr><td>Model</td><td>Top 1 (S)</td><td>Top 5 (S)</td><td>L1(S)</td><td>Top 1 (M)</td><td>Top 5 (M)</td><td>L1 (M)</td></tr><tr><td>R3DLSTM R3D+NL LSTM OPNet</td><td>60.2 46.2 74.8 73.2</td><td>81.8 69.9 1 93.8</td><td>1.2 1.5 0.54 0.85</td><td>28.6 38.6 1 1</td><td>63.3 70.2 1 1</td><td>1.7 1.5 -</td></tr><tr><td>Hopper Aloe (no auxiliary) Aloe Aloe (with L1 loss)</td><td>60.5 70.6 74.0 ± 0.3</td><td>84.5 93.0 94.0 ± 0.4</td><td>0.90 0.53 0.44 ± 0.01</td><td>46.8 56.6 59.7 ± 0.5</td><td>75.1 87.0 90.1 ± 0.6</td><td>- 1.3 0.82 0.69 ± 0.01</td></tr></table>",
+        "bbox": [
+            147,
+            87,
+            849,
+            203
+        ],
+        "page_idx": 7
+    },
+    {
+        "type": "text",
+        "text": "3.2 CATER ",
+        "text_level": 1,
+        "bbox": [
+            173,
+            284,
+            267,
+            299
+        ],
+        "page_idx": 7
+    },
+    {
+        "type": "text",
+        "text": "In a second experiment, we tested Aloe on CATER, a widely-used object-tracking dataset [12, 14, 31, 45]. In CATER, objects from the CLEVR dataset [21] move and potentially occlude other objects, and the goal is to predict the location of a target object (called the snitch) in the final frame. Because the snitch could be occluded by multiple objects that could move in the meantime, a successful model must be sensitive to notions of object permanence. CATER also includes a moving camera variant, which introduces additional complexities for the model. ",
+        "bbox": [
+            174,
+            313,
+            825,
+            396
+        ],
+        "page_idx": 7
+    },
+    {
+        "type": "text",
+        "text": "Concretely, CATER is setup as a classification challenge. Objects are located in an xyz coordinate system, where x and y range from -3 to 3. The xy plane is divided into a 6 by 6 grid, and the task is to predict the grid index of the snitch in the final frame. For Aloe, we use a classification loss (cross entropy over the 36 possible grid indices) and an L1 loss (L1 distance between predicted grid cell and the true grid cell). ",
+        "bbox": [
+            174,
+            402,
+            825,
+            472
+        ],
+        "page_idx": 7
+    },
+    {
+        "type": "text",
+        "text": "Table 2 shows Aloe compared to state-of-the-art models in the literature on both static and moving camera videos. R3D and R3D NL are the strongest two models evaluated by Girdhar and Ramanan [12]. OPNet, or the Object Permanence Network [31], is an architecture with inductive biases designed for object tracking tasks; it was trained with extra supervised labels, namely the bounding boxes for all objects (including occluded ones). Hopper is a multi-hop transformer model developed simultaneously with this work [45]. One key component of Hopper is Hungarian matching between objects of different frames, a strong inductive bias for object tracking. ",
+        "bbox": [
+            174,
+            478,
+            825,
+            575
+        ],
+        "page_idx": 7
+    },
+    {
+        "type": "text",
+        "text": "We train Aloe simultaneously on both static and moving camera videos. Aloe outperforms the R3D models for both static and moving cameras. We also ran Aloe with an additional auxiliary loss consisting of the L1 distance between the predicted cell and the actual cell. With this additional loss, we get comparable results in the moving camera case as the R3D models for the static camera case. Moreover, we achieve comparable accuracy as OPNet for accuracy and L1 distance, despite requiring less supervision to train. Appendix E.2 gives a few sample outputs from Aloe; in particular we note that it is able to find the target object in several cases where the object was occluded, demonstrating that Aloe is able to do some level of object tracking. Finally,we find that an auxiliary self-supervised loss helps the model perform well in the low data regime for CATER as well, as shown in Figure 3. ",
+        "bbox": [
+            174,
+            582,
+            825,
+            707
+        ],
+        "page_idx": 7
+    },
+    {
+        "type": "text",
+        "text": "3.3 ACRE ",
+        "text_level": 1,
+        "bbox": [
+            174,
+            731,
+            258,
+            744
+        ],
+        "page_idx": 7
+    },
+    {
+        "type": "text",
+        "text": "Finally, we measured Aloe’s performance on ACRE, a causal induction dataset inspired by the Blicket task from developmental psychology [13, 44]. ACRE is divided into a set of problems. In each problem, certain objects are chosen to be “Blickets”, and this assignment changes across problems. Each problem presents a context of six images to the model, where different objects are placed on a Blicket machine that lights up if one of those objects is a Blicket. The model is asked whether an unseen combination of objects will light up the Blicket machine. Besides “yes” and “no”, a third possible answer is “undetermined”, which is the case if it is impossible to determine for certain if the objects will light up the machine. Correct inference goes beyond mere correlation: even if every context scene involving object A has a lit-up machine, A’s Blicketness is still uncertain if each of those scenes can potentially be explained by another object (deduction of A’s Blicketness is backward-blocked). ",
+        "bbox": [
+            174,
+            758,
+            825,
+            911
+        ],
+        "page_idx": 7
+    },
+    {
+        "type": "table",
+        "img_path": "images/5ddeb805759cc50f38656ec7ccfdc74c896c226a89b61f575f9dcf2d123229ee.jpg",
+        "table_caption": [
+            "Table 3: Performance on ACRE of Aloe compared to the best results from Zhang et al. [44], split across inference type $\\mathrm { { D . R = } } 1$ Direct, I.D $\\Vdash$ Indirect, ${ \\bf { S . O } } \\mathrm { { = } } \\mathrm { { ; } }$ Screen-Off, B.B $\\ c =$ Backwards Blocking) and generalization type ( $\\mathrm { C } =$ Compositional, ${ \\bf S } =$ Systematic). "
+        ],
+        "table_footnote": [],
+        "table_body": "<table><tr><td>Model</td><td>All (C)</td><td>D.R.</td><td>I.D.</td><td>S.0.</td><td>B.B.</td><td>All (S)</td><td>D.R.</td><td>1.D.</td><td>S.0.</td><td>B.B</td></tr><tr><td>CNN-BERT</td><td>43.79</td><td>54.07</td><td>46.88</td><td>40.57</td><td>28.79</td><td>39.93</td><td>55.97</td><td>68.25</td><td>0.00</td><td>45.59</td></tr><tr><td>NS-OPT</td><td>69.04</td><td>92.5</td><td>76.05</td><td>88.33</td><td>13.48</td><td>67.44</td><td>94.73</td><td>88.38</td><td>82.76</td><td>16.06</td></tr><tr><td>Aloe</td><td>91.76</td><td>97.14</td><td>90.8</td><td>96.8</td><td>78.81</td><td>93.90</td><td>97.18</td><td>71.24</td><td>98.97</td><td>94.48</td></tr></table>",
+        "bbox": [
+            156,
+            87,
+            841,
+            147
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "Inference problems in ACRE are categorized by reasoning type: reasoning from direct evidence (one of the context frames show the query objects on a machine), reasoning from indirect evidence (Blicketness must be deduced by combining evidence from several frames), screened-off reasoning (presence of non-Blickets do not matter if a single Blicket is present), and backward-blocked reasoning (Blicketness cannot be deduced due to confounding variables). Please see Zhang et al. [44] for a more detailed discussion of these reasoning types. ",
+        "bbox": [
+            174,
+            227,
+            825,
+            310
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "Table 3 show Aloe performance compared to a CNN-BERT baseline and to NS-OPT, a neuro-symbolic model introduced in Zhang et al. [44]. Aloe outperforms all extant models for almost all reasoning types and train-test splits. We did not need to do any tuning to apply our model to ACRE—settings from CATER yielded the reported results on the first attempt. Contrary to widely-held opinions that neural networks cannot generalize, Aloe generalizes in scenarios where the training and test sets contain different visual features (compositional split) or different numbers of activated machines in the context (systematic split). Moreover, Aloe achieved by far the best performance on the backwardblocking task, which requires the model to “go beyond the simple covariation strategy to discover the hidden causal relations” [44], dispelling the notion that neural networks can only find correlation. Comparison with NS-OPT (which uses object representations) and CNN-BERT (which uses attention) shows that neither object representations nor attention alone is sufficient for the task; combining these two ideas, as done in Aloe for instance, is essential for this complex reasoning task as well. ",
+        "bbox": [
+            174,
+            315,
+            825,
+            482
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "4 Related work ",
+        "text_level": 1,
+        "bbox": [
+            174,
+            506,
+            316,
+            522
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "Self-attention for reasoning Various studies have shown that transformers [37] can manipulate symbolic data in a manner traditionally associated with symbolic computation. For example, in Lample and Charton [23], a transformer model learned to do symbolic integration and solve ordinary differential equations symbolically, tasks traditionally reserved for symbolic computer algebra systems. Similarly, in Hahn et al. [17], a transformer model learned to solve formulas in propositional logic and demonstrated some degree of generalization to out of distribution formulas. Finally, Brown et al. [1] showed that a transformer trained for language modeling can also do simple analogical reasoning tasks without explicit training. Although these models do not necessarily beat carefully tuned symbolic algorithms in all cases (especially on out of distribution data), they are an important motivation for our proposed recipe for attaining strong reasoning capabilities from self-attention-based models on visually grounded tasks. ",
+        "bbox": [
+            174,
+            540,
+            825,
+            691
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "Object representations A wide body of research points to the importance of object segmentation and representation learning (see e.g. Garnelo and Shanahan [11] for a discussion). Various methods have been proposed for object detection and feature extraction [2, 10, 15, 19, 25, 26, 29]. Past research have also investigated using object based representations in downstream tasks [8, 28]. ",
+        "bbox": [
+            174,
+            712,
+            825,
+            767
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "Self-supervised learning Another line of research concerns learning good representations through self-supervised learning, with an unsupervised auxiliary loss to encourage the discovery of better representations. These better representations could lead to improved performance on supervised tasks, especially when labeled data is scarce. In Devlin et al. [9], for instance, an auxiliary infill loss allows the BERT model to benefit from pretraining on a large corpus of unlabeled data. Our approach to object-centric self-supervised learning is heavily inspired by the BERT infilling loss. Other studies have shown similar benefits to auxiliary learning in vision as well [4, 16, 18]. These works apply various forms of contrastive losses to predict scene dynamics, and the better representations that result carry downstream benefits to supervised and reinforcement learning tasks. ",
+        "bbox": [
+            174,
+            786,
+            825,
+            911
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "Vision and language in self-attention models Recently, many works have emerged on applying transformer models to visual and multimodal data, for static images [24, 27, 33, 36] and videos [34, 35, 43]. These approaches combine the output of convolutional networks with language in various ways using self-attention. While these previous works focused on popular visual question answering tasks, which typically consist of descriptive questions only [41], we focus on understanding deeper causal dynamics of videos. Together with these works, we provide more evidence that self-attention between visual and language elements enables good performance on a diverse set of tasks. ",
+        "bbox": [
+            174,
+            92,
+            825,
+            188
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "In addition, while the use of object representations for discretization in tasks involving static images is becoming more popular, the right way to discretize videos is less clear. We provide strong evidence in the form of ablation studies for architectural decisions that we claim are essential for higher reasoning for this type of data: visual elements should correspond to physical objects in the videos and inter-frame attention between sub-frame entities (as opposed to inter-frame attention of entire frames) is crucial. We also demonstrate the success of using unsupervised object segmentation methods as opposed to the supervised methods used in past work. ",
+        "bbox": [
+            174,
+            194,
+            825,
+            291
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "5 Conclusion ",
+        "text_level": 1,
+        "bbox": [
+            174,
+            310,
+            299,
+            328
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "We have presented Aloe, a model that obtains state-of-the-art performance on three different task domains involving spatiotemporal reasoning about objects. In each of these tasks, previous state-ofthe-art results were established by models with modular, task-specific components. Aloe, by contrast, is a unified solution to all three domains. Its flexibility comes from a reliance on only soft biases and learning objectives: self-attention over learned object embeddings and self-supervised learning of dynamics. We believe the simplicity of this approach is its strength, and hope that this fact, together with the provided code, makes it easy for others to adopt and apply to arbitrary spatio-temporal reasoning problems. ",
+        "bbox": [
+            174,
+            342,
+            825,
+            453
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "On many of these spatio-temporal reasoning problems, previous state-of-the-art was achieved by neuro-symbolic models [6, 11, 40, 41, 44]. Compared to neuro-symbolic models, Aloe can more easily be adapted to other tasks. Indeed, the symbolic components of neuro-symbolic models are often task-specific and not straightforwardly applicable to other tasks. Neuro-symbolic models do have a few advantages, however. First, they are often easier to interpret. Despite the insights that can be gleaned from Aloe’s attention weights, these soft computations are harder to interpret than the explicit symbolic computation found in neuro-symbolic models. Moreover, neuro-symbolic models can be structured in a more modular fashion, which can enable effective generalization to sub-tasks of the task on which the model was trained [6]. ",
+        "bbox": [
+            174,
+            459,
+            825,
+            584
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "Aloe also has some important limitations. First, it has only been applied to synthetic datasets. This limitation is mainly due to the lack of real-world datasets that test for higher-order spatiotemporal reasoning, although we are excited that new datasets such as Traffic QA will be released soon [38]. Second, while the domains where Aloe is applied have been widely adopted and well-received by the research community, it remains possible that they do not evaluate the capacities that they aim to evaluate because of hidden biases or other factors. Regardless, we hope that this work stimulates the design and development of more challenging tasks that more closely approximate the ultimate goal of human or super-human-level visual, spatiotemporal and causal reasoning. Finally, from an ethical point of view, our model may share the common drawback of deep-learning models in perpetuating biases found in the training data, especially when applied to real world data. Development of causal reasoning models could also invite problematic applications involving automated assignment of blame. ",
+        "bbox": [
+            174,
+            590,
+            825,
+            756
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "References   \n[1] Tom B Brown, Benjamin Mann, Nick Ryder, Melanie Subbiah, Jared Kaplan, Prafulla Dhariwal, Arvind Neelakantan, Pranav Shyam, Girish Sastry, Amanda Askell, et al. Language models are few-shot learners. arXiv preprint arXiv:2005.14165, 2020.   \n[2] Chris P. Burgess, Loic Matthey, Nick Watters, Rishabh Kabra, Irina Higgins, Matt Botvinick, and Alex Lerchner. MONet: Unsupervised scene decomposition and representation. arXiv preprint arXiv:1901.11390, 2019.   \n[3] Nicolas Carion, Francisco Massa, Gabriel Synnaeve, Nicolas Usunier, Alexander Kirillov, and Sergey Zagoruyko. End-to-end object detection with transformers. In ECCV, 2020.   \n[4] Ting Chen, Simon Kornblith, Mohammad Norouzi, and Geoffrey Hinton. A simple framework for contrastive learning of visual representations. arXiv preprint arXiv:2002.05709, 2020.   \n[5] Zhe Chen. Object-based attention: A tutorial review. Attention, Perception, & Psychophysics, 74:784–802, 2012. URL https://doi.org/10.3758/s13414-012-0322-z.   \n[6] Zhenfang Chen, Jiayuan Mao, Jiajun Wu, Kwan-Yee Kenneth Wong, Joshua B. Tenenbaum, and Chuang Gan. Grounding physical object and event concepts through dynamic visual reasoning. In International Conference on Learning Representations, 2021. URL https: //arxiv.org/pdf/2103.16564.pdf.   \n[7] Zihang Dai, Zhilin Yang, Yiming Yang, William W Cohen, Jaime Carbonell, Quoc V Le, and Ruslan Salakhutdinov. Transformer-xl: Attentive language models beyond a fixed-length context. arXiv preprint arXiv:1901.02860, 2019.   \n[8] Mikyas T. Desta, Larry Chen, and Tomasz Kornuta. Object-based reasoning in VQA. CoRR, abs/1801.09718, 2018. URL http://arxiv.org/abs/1801.09718.   \n[9] Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. BERT: Pre-training of deep bidirectional transformers for language understanding. arXiv preprint arXiv:1810.04805, 2018.   \n[10] Yilun Du, Kevin Smith, Tomer Ulman, Joshua Tenenbaum, and Jiajun Wu. Unsupervised discovery of 3d physical objects from video. arXiv preprint arXiv:2007.12348, 2020.   \n[11] Marta Garnelo and Murray Shanahan. Reconciling deep learning with symbolic artificial intelligence: representing objects and relations. Current Opinion in Behavioral Sciences, 29: 17–23, 2019.   \n[12] Rohit Girdhar and Deva Ramanan. CATER: A diagnostic dataset for Compositional Actions and TEmporal Reasoning. In ICLR, 2020.   \n[13] Alison Gopnik and David Sobel. Detecting blickets: how young children use information about novel causal powers in categorization and induction. Child Dev, pages 1205–22, 2000. doi: 10.1111/1467-8624.00224.   \n[14] Anirudh Goyal, Aniket Didolkar, Alex Lamb, Kartikeya Badola, Nan Rosemary Ke, Nasim Rahaman, Jonathan Binas, Charles Blundell, Michael Mozer, and Yoshua Bengio. Coordination among neural modules through a shared global workspace. CoRR, abs/2103.01197, 2021. URL https://arxiv.org/abs/2103.01197.   \n[15] Klaus Greff, Raphaël Lopez Kaufman, Rishabh Kabra, Nick Watters, Christopher Burgess, Daniel Zoran, Loïc Matthey, Matthew Botvinick, and Alexander Lerchner. Multi-object representation learning with iterative variational inference. CoRR, abs/1903.00450, 2019. URL http://arxiv.org/abs/1903.00450.   \n[16] Karol Gregor, Danilo Jimenez Rezende, Frederic Besse, Yan Wu, Hamza Merzic, and Aaron van den Oord. Shaping belief states with generative environment models for rl. In H. Wallach, H. Larochelle, A. Beygelzimer, F. d’ Alché-Buc, E. Fox, and R. Garnett, editors, Advances in Neural Information Processing Systems 32, pages 13475–13487. Curran Associates, Inc., 2019. URL http://papers.nips.cc/paper/ 9503-shaping-belief-states-with-generative-environment-models-f pdf. ",
+        "bbox": [
+            173,
+            75,
+            828,
+            916
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "[17] Christopher Hahn, Frederik Schmitt, Jens U. Kreber, Markus N. Rabe, and Bernd Finkbeiner. Transformers generalize to the semantics of logics. arXiv preprint arXiv:2003.04218, 2020. ",
+        "bbox": [
+            173,
+            90,
+            825,
+            121
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "[18] Tengda Han, Weidi Xie, and Andrew Zisserman. Video representation learning by dense predictive coding. In Workshop on Large Scale Holistic Video Understanding, ICCV, 2019. ",
+        "bbox": [
+            174,
+            127,
+            823,
+            157
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "[19] Kaiming He, Georgia Gkioxari, Piotr Dollár, and Ross B. Girshick. Mask R-CNN. CoRR, abs/1703.06870, 2017. URL http://arxiv.org/abs/1703.06870. ",
+        "bbox": [
+            176,
+            165,
+            825,
+            194
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "[20] Drew A Hudson and Christopher D Manning. Compositional attention networks for machine reasoning. 2018. ",
+        "bbox": [
+            174,
+            202,
+            825,
+            231
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "[21] Justin Johnson, Bharath Hariharan, Laurens van der Maaten, Li Fei-Fei, C. Lawrence Zitnick, and Ross B. Girshick. CLEVR: A diagnostic dataset for compositional language and elementary visual reasoning. CoRR, abs/1612.06890, 2016. URL http://arxiv.org/abs/1612. 06890. ",
+        "bbox": [
+            174,
+            238,
+            826,
+            295
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "[22] Diederik P Kingma and Jimmy Ba. Adam: A method for stochastic optimization. arXiv preprint arXiv:1412.6980, 2014. ",
+        "bbox": [
+            173,
+            303,
+            825,
+            332
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "[23] Guillaume Lample and François Charton. Deep learning for symbolic mathematics. In International Conference on Learning Representations, 2020. URL https://openreview.net/ forum?id ${ . } = { }$ S1eZYeHFDS. ",
+        "bbox": [
+            174,
+            339,
+            825,
+            382
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "[24] Liunian Harold Li, Mark Yatskar, Da Yin, Cho-Jui Hsieh, and Kai-Wei Chang. Visualbert: A simple and performant baseline for vision and language. arXiv preprint arXiv:1908.03557, 2019. ",
+        "bbox": [
+            174,
+            390,
+            825,
+            433
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "[25] Zhixuan Lin, Yi-Fu Wu, Skand Vishwanath Peri, Weihao Sun, Gautam Singh, Fei Deng, Jindong Jiang, and Sungjin Ahn. SPACE: Unsupervised object-oriented scene representation via spatial attention and decomposition. In International Conference on Learning Representations, 2020. URL https://openreview.net/forum?id $\\underline { { \\underline { { \\mathbf { \\Pi } } } } }$ rkl03ySYDH. ",
+        "bbox": [
+            174,
+            440,
+            825,
+            497
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "[26] Francesco Locatello, Dirk Weissenborn, Thomas Unterthiner, Aravindh Mahendran, Georg Heigold, Jakob Uszkoreit, Alexey Dosovitskiy, and Thomas Kipf1. Object-Centric Learning with Slot Attention. In Advances in Neural Information Processing Systems (NIPS), 2020. ",
+        "bbox": [
+            174,
+            505,
+            823,
+            547
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "[27] Jiasen Lu, Dhruv Batra, Devi Parikh, and Stefan Lee. Vilbert: Pretraining task-agnostic visiolinguistic representations for vision-and-language tasks. arXiv preprint arXiv:1908.02265, 2019. ",
+        "bbox": [
+            173,
+            556,
+            823,
+            598
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "[28] David Raposo, Adam Santoro, David G. T. Barrett, Razvan Pascanu, Timothy P. Lillicrap, and Peter W. Battaglia. Discovering objects and their relations from entangled scene representations. CoRR, abs/1702.05068, 2017. URL http://arxiv.org/abs/1702.05068. ",
+        "bbox": [
+            174,
+            606,
+            825,
+            650
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "[29] Shaoqing Ren, Kaiming He, Ross B. Girshick, and Jian Sun. Faster R-CNN: towards realtime object detection with region proposal networks. CoRR, abs/1506.01497, 2015. URL http://arxiv.org/abs/1506.01497. ",
+        "bbox": [
+            173,
+            656,
+            823,
+            700
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "[30] Pieter R. Roelfsema, Victor A. Lamme, and Henk Spekreijse. Object-based attention in the primary visual cortex of the macaque monkey. Nature, 395:376–381, 1998. doi: 10.1038/26475. ",
+        "bbox": [
+            173,
+            708,
+            823,
+            737
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "[31] Aviv Shamsian, Ofri Kleinfeld, Amir Globerson, and Gal Chechik. Learning object permanence from video. arXiv preprint arXiv:2003.10469, 2020. ",
+        "bbox": [
+            173,
+            744,
+            823,
+            773
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "[32] Elizabeth Spelke. Core knowledge. The American psychologist, 55:1233–43, 12 2000. doi: 10.1037/0003-066X.55.11.1233. ",
+        "bbox": [
+            174,
+            781,
+            823,
+            810
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "[33] Weijie Su, Xizhou Zhu, Yue Cao, Bin Li, Lewei Lu, Furu Wei, and Jifeng Dai. Vl-bert: Pretraining of generic visual-linguistic representations. In International Conference on Learning Representations, 2020. URL https://openreview.net/forum?id $\\underline { { \\underline { { \\mathbf { \\Pi } } } } } =$ SygXPaEYvH. ",
+        "bbox": [
+            173,
+            818,
+            823,
+            862
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "[34] Chen Sun, Fabien Baradel, Kevin Murphy, and Cordelia Schmid. Contrastive bidirectional transformer for temporal representation learning. CoRR, abs/1906.05743, 2019. URL http: //arxiv.org/abs/1906.05743. ",
+        "bbox": [
+            174,
+            869,
+            828,
+            911
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "[35] Chen Sun, Austin Myers, Carl Vondrick, Kevin Murphy, and Cordelia Schmid. Videobert: A joint model for video and language representation learning. In Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), October 2019. ",
+        "bbox": [
+            171,
+            90,
+            823,
+            133
+        ],
+        "page_idx": 12
+    },
+    {
+        "type": "text",
+        "text": "[36] Hao Tan and Mohit Bansal. Lxmert: Learning cross-modality encoder representations from transformers. In Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing, 2019. ",
+        "bbox": [
+            173,
+            142,
+            823,
+            185
+        ],
+        "page_idx": 12
+    },
+    {
+        "type": "text",
+        "text": "[37] Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N Gomez, Łukasz Kaiser, and Illia Polosukhin. Attention is all you need. In Advances in neural information processing systems, pages 5998–6008, 2017. ",
+        "bbox": [
+            173,
+            193,
+            823,
+            236
+        ],
+        "page_idx": 12
+    },
+    {
+        "type": "text",
+        "text": "[38] Li Xu, He Huang, and Jun Liu. Sutd-trafficqa: A question answering benchmark and an efficient network for video reasoning over traffic events. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2021. ",
+        "bbox": [
+            171,
+            244,
+            823,
+            287
+        ],
+        "page_idx": 12
+    },
+    {
+        "type": "text",
+        "text": "[39] Kexin Yi, Jiajun Wu, Chuang Gan, Antonio Torralba, Pushmeet Kohli, and Joshua B Tenenbaum. Neural-Symbolic VQA: Disentangling Reasoning from Vision and Language Understanding. In Advances in Neural Information Processing Systems (NIPS), 2018. ",
+        "bbox": [
+            171,
+            296,
+            825,
+            339
+        ],
+        "page_idx": 12
+    },
+    {
+        "type": "text",
+        "text": "[40] Kexin Yi, Jiajun Wu, Chuang Gan, Antonio Torralba, Pushmeet Kohli, and Joshua B. Tenenbaum. Neural-symbolic vqa: Disentangling reasoning from vision and language understanding. In Advances in Neural Information Processing Systems, pages 1039–1050, 2018. ",
+        "bbox": [
+            173,
+            347,
+            823,
+            391
+        ],
+        "page_idx": 12
+    },
+    {
+        "type": "text",
+        "text": "[41] Kexin Yi, Chuang Gan, Yunzhu Li, Pushmeet Kohli, Jiajun Wu, Antonio Torralba, and Joshua B. Tenenbaum. CLEVRER: Collision events for video representation and reasoning. In International Conference on Learning Representations, 2020. URL https: //openreview.net/forum?id $\\underline { { \\underline { { \\mathbf { \\Pi } } } } } =$ HkxYzANYDB. ",
+        "bbox": [
+            174,
+            398,
+            825,
+            455
+        ],
+        "page_idx": 12
+    },
+    {
+        "type": "text",
+        "text": "[42] Yang You, Jing Li, Sashank Reddi, Jonathan Hseu, Sanjiv Kumar, Srinadh Bhojanapalli, Xiaodan Song, James Demmel, and Cho-Jui Hsieh. Large batch optimization for deep learning: Training BERT in 76 minutes. Technical Report UCB/EECS-2019-103, EECS Department, University of California, Berkeley, Jun 2019. URL http://www2.eecs.berkeley.edu/Pubs/ TechRpts/2019/EECS-2019-103.html. ",
+        "bbox": [
+            174,
+            464,
+            826,
+            535
+        ],
+        "page_idx": 12
+    },
+    {
+        "type": "text",
+        "text": "[43] Vinicius Zambaldi, David Raposo, Adam Santoro, Victor Bapst, Yujia Li, Igor Babuschkin, Karl Tuyls, David Reichert, Timothy Lillicrap, Edward Lockhart, Murray Shanahan, Victoria Langston, Razvan Pascanu, Matthew Botvinick, Oriol Vinyals, and Peter Battaglia. Deep reinforcement learning with relational inductive biases. In International Conference on Learning Representations, 2019. URL https://openreview.net/forum?id $\\underline { { \\underline { { \\mathbf { \\Pi } } } } } =$ HkxaFoC9KQ. ",
+        "bbox": [
+            173,
+            542,
+            825,
+            613
+        ],
+        "page_idx": 12
+    },
+    {
+        "type": "text",
+        "text": "[44] Chi Zhang, Baoxiong Jia, Mark Edmonds, Song-Chun Zhu, and Yixin Zhu. Acre: Abstract causal reasoning beyond covariation. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2021. ",
+        "bbox": [
+            169,
+            622,
+            823,
+            665
+        ],
+        "page_idx": 12
+    },
+    {
+        "type": "text",
+        "text": "[45] Honglu Zhou, Asim Kadav, Farley Lai, Alexandru Niculescu-Mizil, Martin Renqiang Min, Mubbasir Kapadia, and Hans Peter Graf. Hopper: Multi-hop transformer for spatiotemporal reasoning. In International Conference on Learning Representations, 2021. URL https: //arxiv.org/abs/2103.10574. ",
+        "bbox": [
+            174,
+            674,
+            826,
+            731
+        ],
+        "page_idx": 12
+    }
+]
\ No newline at end of file
diff --git a/parse/train/lHmhW2zmVN/lHmhW2zmVN_middle.json b/parse/train/lHmhW2zmVN/lHmhW2zmVN_middle.json
new file mode 100644
index 0000000000000000000000000000000000000000..2dbabd386ed0496f66e3ef25b52b2a09f22fec42
--- /dev/null
+++ b/parse/train/lHmhW2zmVN/lHmhW2zmVN_middle.json
@@ -0,0 +1,33552 @@
+{
+    "pdf_info": [
+        {
+            "preproc_blocks": [
+                {
+                    "type": "title",
+                    "bbox": [
+                        121,
+                        97,
+                        491,
+                        137
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                119,
+                                96,
+                                491,
+                                119
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        119,
+                                        96,
+                                        491,
+                                        119
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Attention over learned object embeddings enables",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                210,
+                                117,
+                                401,
+                                140
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        210,
+                                        117,
+                                        401,
+                                        140
+                                    ],
+                                    "score": 1.0,
+                                    "content": "complex visual reasoning",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        }
+                    ],
+                    "index": 0.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        111,
+                        178,
+                        503,
+                        224
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                114,
+                                177,
+                                500,
+                                193
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        114,
+                                        177,
+                                        168,
+                                        193
+                                    ],
+                                    "score": 1.0,
+                                    "content": "David Ding",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        184,
+                                        178,
+                                        228,
+                                        191
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Felix Hill",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        246,
+                                        179,
+                                        313,
+                                        191
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Adam Santoro",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        331,
+                                        179,
+                                        415,
+                                        191
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Malcolm Reynolds",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        433,
+                                        179,
+                                        500,
+                                        190
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Matt Botvinick",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                284,
+                                189,
+                                332,
+                                203
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        284,
+                                        189,
+                                        332,
+                                        203
+                                    ],
+                                    "score": 1.0,
+                                    "content": "DeepMind",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                254,
+                                201,
+                                360,
+                                213
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        254,
+                                        201,
+                                        360,
+                                        213
+                                    ],
+                                    "score": 1.0,
+                                    "content": "London, United Kingdom",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                112,
+                                211,
+                                504,
+                                225
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        112,
+                                        211,
+                                        504,
+                                        225
+                                    ],
+                                    "score": 1.0,
+                                    "content": "{fding, felixhill, adamsantoro, mareynolds, botvinick}@google.com",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        }
+                    ],
+                    "index": 3.5
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        283,
+                        252,
+                        328,
+                        265
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                281,
+                                251,
+                                331,
+                                267
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        281,
+                                        251,
+                                        331,
+                                        267
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Abstract",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        }
+                    ],
+                    "index": 6
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        143,
+                        276,
+                        469,
+                        461
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                141,
+                                276,
+                                469,
+                                288
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        276,
+                                        469,
+                                        288
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Neural networks have achieved success in a wide array of perceptual tasks but often",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                287,
+                                469,
+                                300
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        287,
+                                        469,
+                                        300
+                                    ],
+                                    "score": 1.0,
+                                    "content": "fail at tasks involving both perception and higher-level reasoning. On these more",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                298,
+                                470,
+                                311
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        298,
+                                        470,
+                                        311
+                                    ],
+                                    "score": 1.0,
+                                    "content": "challenging tasks, bespoke approaches (such as modular symbolic components, in-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                309,
+                                470,
+                                322
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        309,
+                                        470,
+                                        322
+                                    ],
+                                    "score": 1.0,
+                                    "content": "dependent dynamics models or semantic parsers) targeted towards that specific type",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                320,
+                                470,
+                                333
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        320,
+                                        470,
+                                        333
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of task have typically performed better. The downside to these targeted approaches,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                331,
+                                470,
+                                343
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        331,
+                                        470,
+                                        343
+                                    ],
+                                    "score": 1.0,
+                                    "content": "however, is that they can be more brittle than general-purpose neural networks,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                342,
+                                470,
+                                354
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        342,
+                                        470,
+                                        354
+                                    ],
+                                    "score": 1.0,
+                                    "content": "requiring significant modification or even redesign according to the particular",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                353,
+                                470,
+                                366
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        353,
+                                        470,
+                                        366
+                                    ],
+                                    "score": 1.0,
+                                    "content": "task at hand. Here, we propose a more general neural-network-based approach to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                364,
+                                469,
+                                376
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        364,
+                                        469,
+                                        376
+                                    ],
+                                    "score": 1.0,
+                                    "content": "dynamic visual reasoning problems that obtains state-of-the-art performance on",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                374,
+                                469,
+                                387
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        374,
+                                        469,
+                                        387
+                                    ],
+                                    "score": 1.0,
+                                    "content": "three different domains, in each case outperforming bespoke modular approaches",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                384,
+                                470,
+                                399
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        384,
+                                        470,
+                                        399
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tailored specifically to the task. Our method relies on learned object-centric rep-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                396,
+                                470,
+                                410
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        396,
+                                        470,
+                                        410
+                                    ],
+                                    "score": 1.0,
+                                    "content": "resentations, self-attention and self-supervised dynamics learning, and all three",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                407,
+                                470,
+                                420
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        407,
+                                        470,
+                                        420
+                                    ],
+                                    "score": 1.0,
+                                    "content": "elements together are required for strong performance to emerge. The success",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                418,
+                                469,
+                                431
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        418,
+                                        469,
+                                        431
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of this combination suggests that there may be no need to trade off flexibility",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                428,
+                                470,
+                                443
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        428,
+                                        470,
+                                        443
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for performance on problems involving spatio-temporal or causal-style reasoning.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                439,
+                                469,
+                                452
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        439,
+                                        469,
+                                        452
+                                    ],
+                                    "score": 1.0,
+                                    "content": "With the right soft biases and learning objectives in a neural network we may be",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                142,
+                                451,
+                                292,
+                                462
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        142,
+                                        451,
+                                        292,
+                                        462
+                                    ],
+                                    "score": 1.0,
+                                    "content": "able to attain the best of both worlds.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        }
+                    ],
+                    "index": 15
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        481,
+                        190,
+                        495
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                480,
+                                192,
+                                497
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        480,
+                                        192,
+                                        497
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1 Introduction",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        }
+                    ],
+                    "index": 24
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        506,
+                        505,
+                        616
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                506,
+                                506,
+                                518
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        506,
+                                        506,
+                                        518
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Despite the popularity of artificial neural networks, a body of recent work has focused on their",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                516,
+                                506,
+                                530
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        516,
+                                        506,
+                                        530
+                                    ],
+                                    "score": 1.0,
+                                    "content": "limitations as models of cognition and reasoning. Experiments with dynamical reasoning datasets",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                529,
+                                505,
+                                540
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        529,
+                                        505,
+                                        540
+                                    ],
+                                    "score": 1.0,
+                                    "content": "such as CLEVRER [41], CATER [12], and ACRE [44] show that neural networks can fail to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                540,
+                                505,
+                                551
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        540,
+                                        505,
+                                        551
+                                    ],
+                                    "score": 1.0,
+                                    "content": "adequately reason about the spatio-temporal, compositional or causal structure of visual scenes. On",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                549,
+                                506,
+                                563
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        549,
+                                        506,
+                                        563
+                                    ],
+                                    "score": 1.0,
+                                    "content": "CLEVRER, where models must answer questions about the dynamics of colliding objects, previous",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                561,
+                                506,
+                                573
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        561,
+                                        506,
+                                        573
+                                    ],
+                                    "score": 1.0,
+                                    "content": "experiments show that neural networks can adequately describe the video, but fail when asked to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                571,
+                                506,
+                                586
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        571,
+                                        506,
+                                        586
+                                    ],
+                                    "score": 1.0,
+                                    "content": "predict, explain, or consider counterfactual possibilities. Similarly, on CATER, an object-tracking",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                582,
+                                507,
+                                595
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        582,
+                                        507,
+                                        595
+                                    ],
+                                    "score": 1.0,
+                                    "content": "task, models have trouble tracking the movement of objects when they are hidden in a container.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                593,
+                                506,
+                                606
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        593,
+                                        506,
+                                        606
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Finally, on ACRE, a dataset testing for causal inference, popular models only learned correlations",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                603,
+                                325,
+                                618
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        603,
+                                        325,
+                                        618
+                                    ],
+                                    "score": 1.0,
+                                    "content": "between visual scenes and not the deeper causal logic.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        }
+                    ],
+                    "index": 29.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        621,
+                        505,
+                        687
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                620,
+                                505,
+                                633
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        620,
+                                        505,
+                                        633
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Failures such as these on reasoning (rather than perception) problems have motivated the adoption of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                631,
+                                506,
+                                645
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        631,
+                                        506,
+                                        645
+                                    ],
+                                    "score": 1.0,
+                                    "content": "pipeline-style approaches that combine a general purpose neural network (such as a convolutional",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                642,
+                                506,
+                                655
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        642,
+                                        506,
+                                        655
+                                    ],
+                                    "score": 1.0,
+                                    "content": "block) with a task-specific module that builds in the core logic of the task. For example, on",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                652,
+                                506,
+                                667
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        652,
+                                        506,
+                                        667
+                                    ],
+                                    "score": 1.0,
+                                    "content": "CLEVRER the NS-DR method [41] applies a hand-coded symbolic logic engine (that has the core",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                665,
+                                505,
+                                676
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        665,
+                                        505,
+                                        676
+                                    ],
+                                    "score": 1.0,
+                                    "content": "logic of CLEVRER built-in) to the outputs of a “perceptual” neural front-end, achieving better",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                675,
+                                505,
+                                688
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        675,
+                                        505,
+                                        688
+                                    ],
+                                    "score": 1.0,
+                                    "content": "results than neural network baselines, particularly on counterfactual and explanatory problems. One",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        }
+                    ],
+                    "index": 37.5
+                }
+            ],
+            "page_idx": 0,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        107,
+                        694,
+                        504,
+                        714
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                122,
+                                692,
+                                505,
+                                705
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        122,
+                                        692,
+                                        505,
+                                        705
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Model Code: https://github.com/deepmind/deepmind-research/tree/master/",
+                                    "type": "text"
+                                }
+                            ]
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                703,
+                                273,
+                                715
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        703,
+                                        273,
+                                        715
+                                    ],
+                                    "score": 1.0,
+                                    "content": "object_attention_for_reasoning.",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        106,
+                        731,
+                        385,
+                        742
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                730,
+                                386,
+                                743
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        730,
+                                        386,
+                                        743
+                                    ],
+                                    "score": 1.0,
+                                    "content": "35th Conference on Neural Information Processing Systems (NeurIPS 2021).",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        118,
+                        77,
+                        491,
+                        83
+                    ],
+                    "lines": []
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "title",
+                    "bbox": [
+                        121,
+                        97,
+                        491,
+                        137
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                119,
+                                96,
+                                491,
+                                119
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        119,
+                                        96,
+                                        491,
+                                        119
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Attention over learned object embeddings enables",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                210,
+                                117,
+                                401,
+                                140
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        210,
+                                        117,
+                                        401,
+                                        140
+                                    ],
+                                    "score": 1.0,
+                                    "content": "complex visual reasoning",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        }
+                    ],
+                    "index": 0.5
+                },
+                {
+                    "type": "list",
+                    "bbox": [
+                        111,
+                        178,
+                        503,
+                        224
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                114,
+                                177,
+                                500,
+                                193
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        114,
+                                        177,
+                                        168,
+                                        193
+                                    ],
+                                    "score": 1.0,
+                                    "content": "David Ding",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        184,
+                                        178,
+                                        228,
+                                        191
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Felix Hill",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        246,
+                                        179,
+                                        313,
+                                        191
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Adam Santoro",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        331,
+                                        179,
+                                        415,
+                                        191
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Malcolm Reynolds",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        433,
+                                        179,
+                                        500,
+                                        190
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Matt Botvinick",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                284,
+                                189,
+                                332,
+                                203
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        284,
+                                        189,
+                                        332,
+                                        203
+                                    ],
+                                    "score": 1.0,
+                                    "content": "DeepMind",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                254,
+                                201,
+                                360,
+                                213
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        254,
+                                        201,
+                                        360,
+                                        213
+                                    ],
+                                    "score": 1.0,
+                                    "content": "London, United Kingdom",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                112,
+                                211,
+                                504,
+                                225
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        112,
+                                        211,
+                                        504,
+                                        225
+                                    ],
+                                    "score": 1.0,
+                                    "content": "{fding, felixhill, adamsantoro, mareynolds, botvinick}@google.com",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5,
+                            "is_list_start_line": true
+                        }
+                    ],
+                    "index": 3.5,
+                    "bbox_fs": [
+                        112,
+                        177,
+                        504,
+                        225
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        283,
+                        252,
+                        328,
+                        265
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                281,
+                                251,
+                                331,
+                                267
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        281,
+                                        251,
+                                        331,
+                                        267
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Abstract",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        }
+                    ],
+                    "index": 6
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        143,
+                        276,
+                        469,
+                        461
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                141,
+                                276,
+                                469,
+                                288
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        276,
+                                        469,
+                                        288
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Neural networks have achieved success in a wide array of perceptual tasks but often",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                287,
+                                469,
+                                300
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        287,
+                                        469,
+                                        300
+                                    ],
+                                    "score": 1.0,
+                                    "content": "fail at tasks involving both perception and higher-level reasoning. On these more",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                298,
+                                470,
+                                311
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        298,
+                                        470,
+                                        311
+                                    ],
+                                    "score": 1.0,
+                                    "content": "challenging tasks, bespoke approaches (such as modular symbolic components, in-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                309,
+                                470,
+                                322
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        309,
+                                        470,
+                                        322
+                                    ],
+                                    "score": 1.0,
+                                    "content": "dependent dynamics models or semantic parsers) targeted towards that specific type",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                320,
+                                470,
+                                333
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        320,
+                                        470,
+                                        333
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of task have typically performed better. The downside to these targeted approaches,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                331,
+                                470,
+                                343
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        331,
+                                        470,
+                                        343
+                                    ],
+                                    "score": 1.0,
+                                    "content": "however, is that they can be more brittle than general-purpose neural networks,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                342,
+                                470,
+                                354
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        342,
+                                        470,
+                                        354
+                                    ],
+                                    "score": 1.0,
+                                    "content": "requiring significant modification or even redesign according to the particular",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                353,
+                                470,
+                                366
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        353,
+                                        470,
+                                        366
+                                    ],
+                                    "score": 1.0,
+                                    "content": "task at hand. Here, we propose a more general neural-network-based approach to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                364,
+                                469,
+                                376
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        364,
+                                        469,
+                                        376
+                                    ],
+                                    "score": 1.0,
+                                    "content": "dynamic visual reasoning problems that obtains state-of-the-art performance on",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                374,
+                                469,
+                                387
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        374,
+                                        469,
+                                        387
+                                    ],
+                                    "score": 1.0,
+                                    "content": "three different domains, in each case outperforming bespoke modular approaches",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                384,
+                                470,
+                                399
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        384,
+                                        470,
+                                        399
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tailored specifically to the task. Our method relies on learned object-centric rep-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                396,
+                                470,
+                                410
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        396,
+                                        470,
+                                        410
+                                    ],
+                                    "score": 1.0,
+                                    "content": "resentations, self-attention and self-supervised dynamics learning, and all three",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                407,
+                                470,
+                                420
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        407,
+                                        470,
+                                        420
+                                    ],
+                                    "score": 1.0,
+                                    "content": "elements together are required for strong performance to emerge. The success",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                418,
+                                469,
+                                431
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        418,
+                                        469,
+                                        431
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of this combination suggests that there may be no need to trade off flexibility",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                428,
+                                470,
+                                443
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        428,
+                                        470,
+                                        443
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for performance on problems involving spatio-temporal or causal-style reasoning.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                439,
+                                469,
+                                452
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        439,
+                                        469,
+                                        452
+                                    ],
+                                    "score": 1.0,
+                                    "content": "With the right soft biases and learning objectives in a neural network we may be",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                142,
+                                451,
+                                292,
+                                462
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        142,
+                                        451,
+                                        292,
+                                        462
+                                    ],
+                                    "score": 1.0,
+                                    "content": "able to attain the best of both worlds.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        }
+                    ],
+                    "index": 15,
+                    "bbox_fs": [
+                        141,
+                        276,
+                        470,
+                        462
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        481,
+                        190,
+                        495
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                480,
+                                192,
+                                497
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        480,
+                                        192,
+                                        497
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1 Introduction",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        }
+                    ],
+                    "index": 24
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        506,
+                        505,
+                        616
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                506,
+                                506,
+                                518
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        506,
+                                        506,
+                                        518
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Despite the popularity of artificial neural networks, a body of recent work has focused on their",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                516,
+                                506,
+                                530
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        516,
+                                        506,
+                                        530
+                                    ],
+                                    "score": 1.0,
+                                    "content": "limitations as models of cognition and reasoning. Experiments with dynamical reasoning datasets",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                529,
+                                505,
+                                540
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        529,
+                                        505,
+                                        540
+                                    ],
+                                    "score": 1.0,
+                                    "content": "such as CLEVRER [41], CATER [12], and ACRE [44] show that neural networks can fail to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                540,
+                                505,
+                                551
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        540,
+                                        505,
+                                        551
+                                    ],
+                                    "score": 1.0,
+                                    "content": "adequately reason about the spatio-temporal, compositional or causal structure of visual scenes. On",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                549,
+                                506,
+                                563
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        549,
+                                        506,
+                                        563
+                                    ],
+                                    "score": 1.0,
+                                    "content": "CLEVRER, where models must answer questions about the dynamics of colliding objects, previous",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                561,
+                                506,
+                                573
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        561,
+                                        506,
+                                        573
+                                    ],
+                                    "score": 1.0,
+                                    "content": "experiments show that neural networks can adequately describe the video, but fail when asked to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                571,
+                                506,
+                                586
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        571,
+                                        506,
+                                        586
+                                    ],
+                                    "score": 1.0,
+                                    "content": "predict, explain, or consider counterfactual possibilities. Similarly, on CATER, an object-tracking",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                582,
+                                507,
+                                595
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        582,
+                                        507,
+                                        595
+                                    ],
+                                    "score": 1.0,
+                                    "content": "task, models have trouble tracking the movement of objects when they are hidden in a container.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                593,
+                                506,
+                                606
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        593,
+                                        506,
+                                        606
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Finally, on ACRE, a dataset testing for causal inference, popular models only learned correlations",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                603,
+                                325,
+                                618
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        603,
+                                        325,
+                                        618
+                                    ],
+                                    "score": 1.0,
+                                    "content": "between visual scenes and not the deeper causal logic.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        }
+                    ],
+                    "index": 29.5,
+                    "bbox_fs": [
+                        104,
+                        506,
+                        507,
+                        618
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        621,
+                        505,
+                        687
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                620,
+                                505,
+                                633
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        620,
+                                        505,
+                                        633
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Failures such as these on reasoning (rather than perception) problems have motivated the adoption of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                631,
+                                506,
+                                645
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        631,
+                                        506,
+                                        645
+                                    ],
+                                    "score": 1.0,
+                                    "content": "pipeline-style approaches that combine a general purpose neural network (such as a convolutional",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                642,
+                                506,
+                                655
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        642,
+                                        506,
+                                        655
+                                    ],
+                                    "score": 1.0,
+                                    "content": "block) with a task-specific module that builds in the core logic of the task. For example, on",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                652,
+                                506,
+                                667
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        652,
+                                        506,
+                                        667
+                                    ],
+                                    "score": 1.0,
+                                    "content": "CLEVRER the NS-DR method [41] applies a hand-coded symbolic logic engine (that has the core",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                665,
+                                505,
+                                676
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        665,
+                                        505,
+                                        676
+                                    ],
+                                    "score": 1.0,
+                                    "content": "logic of CLEVRER built-in) to the outputs of a “perceptual” neural front-end, achieving better",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                675,
+                                505,
+                                688
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        675,
+                                        505,
+                                        688
+                                    ],
+                                    "score": 1.0,
+                                    "content": "results than neural network baselines, particularly on counterfactual and explanatory problems. One",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                73,
+                                506,
+                                86
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        73,
+                                        506,
+                                        86
+                                    ],
+                                    "score": 1.0,
+                                    "content": "limitation of these pipeline approaches, however, is that they are typically created with a single",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                84,
+                                506,
+                                96
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        84,
+                                        506,
+                                        96
+                                    ],
+                                    "score": 1.0,
+                                    "content": "problem or problem domain in mind, and may not apply out-of-the-box to other related problems.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                95,
+                                505,
+                                106
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        95,
+                                        505,
+                                        106
+                                    ],
+                                    "score": 1.0,
+                                    "content": "For example, to apply NS-DR to CATER, the entire symbolic module needs to be rewritten to handle",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                105,
+                                506,
+                                118
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        105,
+                                        506,
+                                        118
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the new interactions and task logic of CATER: the custom logic to handle collisions and object",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                117,
+                                506,
+                                129
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        117,
+                                        506,
+                                        129
+                                    ],
+                                    "score": 1.0,
+                                    "content": "removal must be replaced with new custom logic to handle occlusions and grid-resolution, and these",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                128,
+                                506,
+                                139
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        128,
+                                        506,
+                                        139
+                                    ],
+                                    "score": 1.0,
+                                    "content": "changes require further modifications to the perceptual front-end to output data in a new format.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                137,
+                                506,
+                                151
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        137,
+                                        506,
+                                        151
+                                    ],
+                                    "score": 1.0,
+                                    "content": "This brittleness is not exclusive to symbolic approaches. While Hungarian-matching between object",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                150,
+                                505,
+                                161
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        150,
+                                        505,
+                                        161
+                                    ],
+                                    "score": 1.0,
+                                    "content": "embeddings may be well-suited for object-tracking tasks [45], it is not obvious how it would help for",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                160,
+                                198,
+                                172
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        160,
+                                        198,
+                                        172
+                                    ],
+                                    "score": 1.0,
+                                    "content": "causal inference tasks.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 8
+                        }
+                    ],
+                    "index": 37.5,
+                    "bbox_fs": [
+                        104,
+                        620,
+                        506,
+                        688
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        72,
+                        505,
+                        171
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                73,
+                                506,
+                                86
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        73,
+                                        506,
+                                        86
+                                    ],
+                                    "score": 1.0,
+                                    "content": "limitation of these pipeline approaches, however, is that they are typically created with a single",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                84,
+                                506,
+                                96
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        84,
+                                        506,
+                                        96
+                                    ],
+                                    "score": 1.0,
+                                    "content": "problem or problem domain in mind, and may not apply out-of-the-box to other related problems.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                95,
+                                505,
+                                106
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        95,
+                                        505,
+                                        106
+                                    ],
+                                    "score": 1.0,
+                                    "content": "For example, to apply NS-DR to CATER, the entire symbolic module needs to be rewritten to handle",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                105,
+                                506,
+                                118
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        105,
+                                        506,
+                                        118
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the new interactions and task logic of CATER: the custom logic to handle collisions and object",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                117,
+                                506,
+                                129
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        117,
+                                        506,
+                                        129
+                                    ],
+                                    "score": 1.0,
+                                    "content": "removal must be replaced with new custom logic to handle occlusions and grid-resolution, and these",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                128,
+                                506,
+                                139
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        128,
+                                        506,
+                                        139
+                                    ],
+                                    "score": 1.0,
+                                    "content": "changes require further modifications to the perceptual front-end to output data in a new format.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                137,
+                                506,
+                                151
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        137,
+                                        506,
+                                        151
+                                    ],
+                                    "score": 1.0,
+                                    "content": "This brittleness is not exclusive to symbolic approaches. While Hungarian-matching between object",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                150,
+                                505,
+                                161
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        150,
+                                        505,
+                                        161
+                                    ],
+                                    "score": 1.0,
+                                    "content": "embeddings may be well-suited for object-tracking tasks [45], it is not obvious how it would help for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                160,
+                                198,
+                                172
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        160,
+                                        198,
+                                        172
+                                    ],
+                                    "score": 1.0,
+                                    "content": "causal inference tasks.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        }
+                    ],
+                    "index": 4
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        176,
+                        505,
+                        209
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                174,
+                                505,
+                                191
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        174,
+                                        505,
+                                        191
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Here, we describe a more general neural-network-based approach to visual spatio-temporal reasoning",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                187,
+                                504,
+                                199
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        187,
+                                        504,
+                                        199
+                                    ],
+                                    "score": 1.0,
+                                    "content": "problems, which does not rely on task-specific integration of modular components. In place of these",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                199,
+                                312,
+                                210
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        199,
+                                        312,
+                                        210
+                                    ],
+                                    "score": 1.0,
+                                    "content": "components, our model relies on three key aspects:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        }
+                    ],
+                    "index": 10
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        118,
+                        218,
+                        505,
+                        283
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                118,
+                                218,
+                                367,
+                                231
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        118,
+                                        218,
+                                        367,
+                                        231
+                                    ],
+                                    "score": 1.0,
+                                    "content": "• Self-attention to effectively integrate information over time",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                118,
+                                233,
+                                505,
+                                246
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        118,
+                                        233,
+                                        505,
+                                        246
+                                    ],
+                                    "score": 1.0,
+                                    "content": "• Soft-discretization of the input at the most informative level of abstraction – above pixels and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                128,
+                                245,
+                                459,
+                                257
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        128,
+                                        245,
+                                        459,
+                                        257
+                                    ],
+                                    "score": 1.0,
+                                    "content": "local features, and below entire frames—corresponding approximately to ‘objects’",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                124,
+                                259,
+                                505,
+                                273
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        124,
+                                        259,
+                                        505,
+                                        273
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Self-supervised learning, i.e. requiring the model to infer masked out objects, to extract more",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                128,
+                                270,
+                                320,
+                                284
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        128,
+                                        270,
+                                        320,
+                                        284
+                                    ],
+                                    "score": 1.0,
+                                    "content": "information about dynamics from each sample.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        }
+                    ],
+                    "index": 14
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        109,
+                        292,
+                        504,
+                        325
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                291,
+                                505,
+                                305
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        291,
+                                        505,
+                                        305
+                                    ],
+                                    "score": 1.0,
+                                    "content": "While many past models have applied each individual ingredient separately (including on the tasks",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                302,
+                                506,
+                                316
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        302,
+                                        506,
+                                        316
+                                    ],
+                                    "score": 1.0,
+                                    "content": "we study), we show that it is the combination of all three ingredients in the right way that allows our",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                108,
+                                314,
+                                180,
+                                325
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        108,
+                                        314,
+                                        180,
+                                        325
+                                    ],
+                                    "score": 1.0,
+                                    "content": "model to succeed.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        }
+                    ],
+                    "index": 18
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        109,
+                        330,
+                        504,
+                        363
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                330,
+                                505,
+                                343
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        330,
+                                        505,
+                                        343
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The resulting model, which we call Aloe (Attention over Learned Object Embeddings), outperforms",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                340,
+                                506,
+                                353
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        340,
+                                        506,
+                                        353
+                                    ],
+                                    "score": 1.0,
+                                    "content": "both pipeline and neural-network-based approaches on three different task domains designed to test",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                107,
+                                352,
+                                455,
+                                364
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        352,
+                                        455,
+                                        364
+                                    ],
+                                    "score": 1.0,
+                                    "content": "physical and dynamical reasoning from pixel inputs. We highlight our key results here:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        }
+                    ],
+                    "index": 21
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        118,
+                        372,
+                        506,
+                        513
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                123,
+                                372,
+                                505,
+                                385
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        123,
+                                        372,
+                                        505,
+                                        385
+                                    ],
+                                    "score": 1.0,
+                                    "content": "• CLEVRER (explanatory, predictive, and counterfactual reasoning): Aloe achieves significantly",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                128,
+                                383,
+                                506,
+                                396
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        128,
+                                        383,
+                                        506,
+                                        396
+                                    ],
+                                    "score": 1.0,
+                                    "content": "higher accuracy than both more task-specific, modular approaches, and previous neural network",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                129,
+                                394,
+                                506,
+                                407
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        129,
+                                        394,
+                                        506,
+                                        407
+                                    ],
+                                    "score": 1.0,
+                                    "content": "methods on all question types. On counterfactual questions, thought to be most challenging for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                129,
+                                405,
+                                496,
+                                417
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        129,
+                                        406,
+                                        280,
+                                        417
+                                    ],
+                                    "score": 1.0,
+                                    "content": "neural-only architectures, we achieve",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        280,
+                                        405,
+                                        301,
+                                        416
+                                    ],
+                                    "score": 0.87,
+                                    "content": "75 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        301,
+                                        406,
+                                        313,
+                                        417
+                                    ],
+                                    "score": 1.0,
+                                    "content": "vs",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        314,
+                                        405,
+                                        335,
+                                        416
+                                    ],
+                                    "score": 0.85,
+                                    "content": "46 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        335,
+                                        406,
+                                        496,
+                                        417
+                                    ],
+                                    "score": 1.0,
+                                    "content": "accuracy for more specialised methods.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                123,
+                                420,
+                                506,
+                                433
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        123,
+                                        420,
+                                        506,
+                                        433
+                                    ],
+                                    "score": 1.0,
+                                    "content": "• CATER (object-permanence): Aloe achieves accuracy exceeding or matching other current",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                429,
+                                507,
+                                446
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        429,
+                                        507,
+                                        446
+                                    ],
+                                    "score": 1.0,
+                                    "content": "models. Notably, the strongest alternative models were expressly designed for object-tracking,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                128,
+                                442,
+                                497,
+                                455
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        128,
+                                        442,
+                                        497,
+                                        455
+                                    ],
+                                    "score": 1.0,
+                                    "content": "whereas our architecture is applicable without modification to other reasoning tasks as well.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                458,
+                                505,
+                                470
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        458,
+                                        505,
+                                        470
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ACRE (causal-inference “beyond the simple strategy of inducing causal relationships by",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                128,
+                                468,
+                                506,
+                                482
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        128,
+                                        468,
+                                        306,
+                                        482
+                                    ],
+                                    "score": 1.0,
+                                    "content": "covariation” [44]): Overall, Aloe achieves",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        306,
+                                        469,
+                                        327,
+                                        479
+                                    ],
+                                    "score": 0.85,
+                                    "content": "94 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        328,
+                                        468,
+                                        358,
+                                        482
+                                    ],
+                                    "score": 1.0,
+                                    "content": "vs the",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        358,
+                                        469,
+                                        379,
+                                        479
+                                    ],
+                                    "score": 0.86,
+                                    "content": "67 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        380,
+                                        468,
+                                        506,
+                                        482
+                                    ],
+                                    "score": 1.0,
+                                    "content": "accuracy achieved by the top",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                478,
+                                506,
+                                493
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        478,
+                                        506,
+                                        493
+                                    ],
+                                    "score": 1.0,
+                                    "content": "neuro-symbolic model. On the most challenging tasks, we achieve, for “backward-blocking”",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                128,
+                                490,
+                                506,
+                                503
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        128,
+                                        490,
+                                        170,
+                                        503
+                                    ],
+                                    "score": 1.0,
+                                    "content": "inference,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        170,
+                                        491,
+                                        204,
+                                        501
+                                    ],
+                                    "score": 0.88,
+                                    "content": "9 4 . 4 8 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        204,
+                                        490,
+                                        219,
+                                        503
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(vs",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        220,
+                                        491,
+                                        253,
+                                        501
+                                    ],
+                                    "score": 0.86,
+                                    "content": "1 6 . 0 6 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        253,
+                                        490,
+                                        506,
+                                        503
+                                    ],
+                                    "score": 1.0,
+                                    "content": "by the best modular, neuro-symbolic systems), and, for “screen-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                129,
+                                501,
+                                381,
+                                514
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        129,
+                                        501,
+                                        190,
+                                        514
+                                    ],
+                                    "score": 1.0,
+                                    "content": "off” inference,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        190,
+                                        502,
+                                        223,
+                                        512
+                                    ],
+                                    "score": 0.89,
+                                    "content": "9 8 . 9 7 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        223,
+                                        501,
+                                        239,
+                                        514
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(vs",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        240,
+                                        502,
+                                        268,
+                                        512
+                                    ],
+                                    "score": 0.89,
+                                    "content": "{ \\bf0 . 0 0 \\% }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        268,
+                                        501,
+                                        381,
+                                        514
+                                    ],
+                                    "score": 1.0,
+                                    "content": "by a CNN-BERT baseline).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        }
+                    ],
+                    "index": 28.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        522,
+                        505,
+                        566
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                522,
+                                505,
+                                534
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        522,
+                                        505,
+                                        534
+                                    ],
+                                    "score": 1.0,
+                                    "content": "As we have emphasized, the previous best performing models for each task all contain task-specific",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                533,
+                                506,
+                                545
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        533,
+                                        506,
+                                        545
+                                    ],
+                                    "score": 1.0,
+                                    "content": "design elements, whereas Aloe can be applied to all the tasks without modification. On CLEVRER,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                542,
+                                506,
+                                557
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        542,
+                                        433,
+                                        557
+                                    ],
+                                    "score": 1.0,
+                                    "content": "we also show that Aloe matches the performance of the previous best models with",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        433,
+                                        544,
+                                        453,
+                                        554
+                                    ],
+                                    "score": 0.85,
+                                    "content": "40 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        453,
+                                        542,
+                                        506,
+                                        557
+                                    ],
+                                    "score": 1.0,
+                                    "content": "less training",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                555,
+                                436,
+                                567
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        555,
+                                        436,
+                                        567
+                                    ],
+                                    "score": 1.0,
+                                    "content": "data, which demonstrates that our approach is data-efficient as well as performant.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        }
+                    ],
+                    "index": 36.5
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        582,
+                        171,
+                        595
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                579,
+                                172,
+                                598
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        579,
+                                        172,
+                                        598
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2 Methods",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        }
+                    ],
+                    "index": 39
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        607,
+                        505,
+                        684
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                606,
+                                505,
+                                620
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        606,
+                                        505,
+                                        620
+                                    ],
+                                    "score": 1.0,
+                                    "content": "A guiding motivation for the design of Aloe is the converging evidence for the value of self-attention",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                618,
+                                505,
+                                630
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        618,
+                                        505,
+                                        630
+                                    ],
+                                    "score": 1.0,
+                                    "content": "mechanisms operating on a finite sequences of discrete entities. Written language is inherently",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                628,
+                                505,
+                                641
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        628,
+                                        505,
+                                        641
+                                    ],
+                                    "score": 1.0,
+                                    "content": "discrete and hence is well-suited to self-attention-based approaches. In other domains, such as raw",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                640,
+                                506,
+                                652
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        640,
+                                        506,
+                                        652
+                                    ],
+                                    "score": 1.0,
+                                    "content": "audio or vision, it is less clear how to leverage self-attention. We hypothesize that the application of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                649,
+                                507,
+                                664
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        649,
+                                        507,
+                                        664
+                                    ],
+                                    "score": 1.0,
+                                    "content": "self-attention-based models to visual tasks could benefit from an approximate ‘discretization’ process,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                661,
+                                506,
+                                675
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        661,
+                                        506,
+                                        675
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and determining the right level of discretization is an important choice that can significantly affect",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                673,
+                                189,
+                                685
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        673,
+                                        189,
+                                        685
+                                    ],
+                                    "score": 1.0,
+                                    "content": "model performance.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        }
+                    ],
+                    "index": 43
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        689,
+                        505,
+                        722
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                689,
+                                505,
+                                701
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        689,
+                                        505,
+                                        701
+                                    ],
+                                    "score": 1.0,
+                                    "content": "At the finest level, data could simply be discretized into pixels (as is already the case for most",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 47
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                700,
+                                507,
+                                713
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        700,
+                                        507,
+                                        713
+                                    ],
+                                    "score": 1.0,
+                                    "content": "machine-processed visual data). Pixels are too fine-grained for many applications, however—for one,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 48
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                710,
+                                507,
+                                725
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        710,
+                                        507,
+                                        725
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the memory required to support self-attention across all pixels is prohibitive. Partly for this reason,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 49
+                        }
+                    ],
+                    "index": 48
+                }
+            ],
+            "page_idx": 1,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        302,
+                        742,
+                        309,
+                        750
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                301,
+                                741,
+                                310,
+                                753
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        301,
+                                        741,
+                                        310,
+                                        753
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        72,
+                        505,
+                        171
+                    ],
+                    "lines": [],
+                    "index": 4,
+                    "bbox_fs": [
+                        105,
+                        73,
+                        506,
+                        172
+                    ],
+                    "lines_deleted": true
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        176,
+                        505,
+                        209
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                174,
+                                505,
+                                191
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        174,
+                                        505,
+                                        191
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Here, we describe a more general neural-network-based approach to visual spatio-temporal reasoning",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                187,
+                                504,
+                                199
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        187,
+                                        504,
+                                        199
+                                    ],
+                                    "score": 1.0,
+                                    "content": "problems, which does not rely on task-specific integration of modular components. In place of these",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                199,
+                                312,
+                                210
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        199,
+                                        312,
+                                        210
+                                    ],
+                                    "score": 1.0,
+                                    "content": "components, our model relies on three key aspects:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        }
+                    ],
+                    "index": 10,
+                    "bbox_fs": [
+                        105,
+                        174,
+                        505,
+                        210
+                    ]
+                },
+                {
+                    "type": "list",
+                    "bbox": [
+                        118,
+                        218,
+                        505,
+                        283
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                118,
+                                218,
+                                367,
+                                231
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        118,
+                                        218,
+                                        367,
+                                        231
+                                    ],
+                                    "score": 1.0,
+                                    "content": "• Self-attention to effectively integrate information over time",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12,
+                            "is_list_start_line": true,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                118,
+                                233,
+                                505,
+                                246
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        118,
+                                        233,
+                                        505,
+                                        246
+                                    ],
+                                    "score": 1.0,
+                                    "content": "• Soft-discretization of the input at the most informative level of abstraction – above pixels and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                128,
+                                245,
+                                459,
+                                257
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        128,
+                                        245,
+                                        459,
+                                        257
+                                    ],
+                                    "score": 1.0,
+                                    "content": "local features, and below entire frames—corresponding approximately to ‘objects’",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                124,
+                                259,
+                                505,
+                                273
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        124,
+                                        259,
+                                        505,
+                                        273
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Self-supervised learning, i.e. requiring the model to infer masked out objects, to extract more",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                128,
+                                270,
+                                320,
+                                284
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        128,
+                                        270,
+                                        320,
+                                        284
+                                    ],
+                                    "score": 1.0,
+                                    "content": "information about dynamics from each sample.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16,
+                            "is_list_end_line": true
+                        }
+                    ],
+                    "index": 14,
+                    "bbox_fs": [
+                        118,
+                        218,
+                        505,
+                        284
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        109,
+                        292,
+                        504,
+                        325
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                291,
+                                505,
+                                305
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        291,
+                                        505,
+                                        305
+                                    ],
+                                    "score": 1.0,
+                                    "content": "While many past models have applied each individual ingredient separately (including on the tasks",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                302,
+                                506,
+                                316
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        302,
+                                        506,
+                                        316
+                                    ],
+                                    "score": 1.0,
+                                    "content": "we study), we show that it is the combination of all three ingredients in the right way that allows our",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                108,
+                                314,
+                                180,
+                                325
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        108,
+                                        314,
+                                        180,
+                                        325
+                                    ],
+                                    "score": 1.0,
+                                    "content": "model to succeed.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        }
+                    ],
+                    "index": 18,
+                    "bbox_fs": [
+                        106,
+                        291,
+                        506,
+                        325
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        109,
+                        330,
+                        504,
+                        363
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                330,
+                                505,
+                                343
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        330,
+                                        505,
+                                        343
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The resulting model, which we call Aloe (Attention over Learned Object Embeddings), outperforms",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                340,
+                                506,
+                                353
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        340,
+                                        506,
+                                        353
+                                    ],
+                                    "score": 1.0,
+                                    "content": "both pipeline and neural-network-based approaches on three different task domains designed to test",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                107,
+                                352,
+                                455,
+                                364
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        352,
+                                        455,
+                                        364
+                                    ],
+                                    "score": 1.0,
+                                    "content": "physical and dynamical reasoning from pixel inputs. We highlight our key results here:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        }
+                    ],
+                    "index": 21,
+                    "bbox_fs": [
+                        106,
+                        330,
+                        506,
+                        364
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        118,
+                        372,
+                        506,
+                        513
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                123,
+                                372,
+                                505,
+                                385
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        123,
+                                        372,
+                                        505,
+                                        385
+                                    ],
+                                    "score": 1.0,
+                                    "content": "• CLEVRER (explanatory, predictive, and counterfactual reasoning): Aloe achieves significantly",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                128,
+                                383,
+                                506,
+                                396
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        128,
+                                        383,
+                                        506,
+                                        396
+                                    ],
+                                    "score": 1.0,
+                                    "content": "higher accuracy than both more task-specific, modular approaches, and previous neural network",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                129,
+                                394,
+                                506,
+                                407
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        129,
+                                        394,
+                                        506,
+                                        407
+                                    ],
+                                    "score": 1.0,
+                                    "content": "methods on all question types. On counterfactual questions, thought to be most challenging for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                129,
+                                405,
+                                496,
+                                417
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        129,
+                                        406,
+                                        280,
+                                        417
+                                    ],
+                                    "score": 1.0,
+                                    "content": "neural-only architectures, we achieve",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        280,
+                                        405,
+                                        301,
+                                        416
+                                    ],
+                                    "score": 0.87,
+                                    "content": "75 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        301,
+                                        406,
+                                        313,
+                                        417
+                                    ],
+                                    "score": 1.0,
+                                    "content": "vs",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        314,
+                                        405,
+                                        335,
+                                        416
+                                    ],
+                                    "score": 0.85,
+                                    "content": "46 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        335,
+                                        406,
+                                        496,
+                                        417
+                                    ],
+                                    "score": 1.0,
+                                    "content": "accuracy for more specialised methods.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                123,
+                                420,
+                                506,
+                                433
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        123,
+                                        420,
+                                        506,
+                                        433
+                                    ],
+                                    "score": 1.0,
+                                    "content": "• CATER (object-permanence): Aloe achieves accuracy exceeding or matching other current",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                429,
+                                507,
+                                446
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        429,
+                                        507,
+                                        446
+                                    ],
+                                    "score": 1.0,
+                                    "content": "models. Notably, the strongest alternative models were expressly designed for object-tracking,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                128,
+                                442,
+                                497,
+                                455
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        128,
+                                        442,
+                                        497,
+                                        455
+                                    ],
+                                    "score": 1.0,
+                                    "content": "whereas our architecture is applicable without modification to other reasoning tasks as well.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                458,
+                                505,
+                                470
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        458,
+                                        505,
+                                        470
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ACRE (causal-inference “beyond the simple strategy of inducing causal relationships by",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                128,
+                                468,
+                                506,
+                                482
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        128,
+                                        468,
+                                        306,
+                                        482
+                                    ],
+                                    "score": 1.0,
+                                    "content": "covariation” [44]): Overall, Aloe achieves",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        306,
+                                        469,
+                                        327,
+                                        479
+                                    ],
+                                    "score": 0.85,
+                                    "content": "94 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        328,
+                                        468,
+                                        358,
+                                        482
+                                    ],
+                                    "score": 1.0,
+                                    "content": "vs the",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        358,
+                                        469,
+                                        379,
+                                        479
+                                    ],
+                                    "score": 0.86,
+                                    "content": "67 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        380,
+                                        468,
+                                        506,
+                                        482
+                                    ],
+                                    "score": 1.0,
+                                    "content": "accuracy achieved by the top",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                478,
+                                506,
+                                493
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        478,
+                                        506,
+                                        493
+                                    ],
+                                    "score": 1.0,
+                                    "content": "neuro-symbolic model. On the most challenging tasks, we achieve, for “backward-blocking”",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                128,
+                                490,
+                                506,
+                                503
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        128,
+                                        490,
+                                        170,
+                                        503
+                                    ],
+                                    "score": 1.0,
+                                    "content": "inference,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        170,
+                                        491,
+                                        204,
+                                        501
+                                    ],
+                                    "score": 0.88,
+                                    "content": "9 4 . 4 8 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        204,
+                                        490,
+                                        219,
+                                        503
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(vs",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        220,
+                                        491,
+                                        253,
+                                        501
+                                    ],
+                                    "score": 0.86,
+                                    "content": "1 6 . 0 6 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        253,
+                                        490,
+                                        506,
+                                        503
+                                    ],
+                                    "score": 1.0,
+                                    "content": "by the best modular, neuro-symbolic systems), and, for “screen-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                129,
+                                501,
+                                381,
+                                514
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        129,
+                                        501,
+                                        190,
+                                        514
+                                    ],
+                                    "score": 1.0,
+                                    "content": "off” inference,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        190,
+                                        502,
+                                        223,
+                                        512
+                                    ],
+                                    "score": 0.89,
+                                    "content": "9 8 . 9 7 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        223,
+                                        501,
+                                        239,
+                                        514
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(vs",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        240,
+                                        502,
+                                        268,
+                                        512
+                                    ],
+                                    "score": 0.89,
+                                    "content": "{ \\bf0 . 0 0 \\% }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        268,
+                                        501,
+                                        381,
+                                        514
+                                    ],
+                                    "score": 1.0,
+                                    "content": "by a CNN-BERT baseline).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        }
+                    ],
+                    "index": 28.5,
+                    "bbox_fs": [
+                        123,
+                        372,
+                        507,
+                        514
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        522,
+                        505,
+                        566
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                522,
+                                505,
+                                534
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        522,
+                                        505,
+                                        534
+                                    ],
+                                    "score": 1.0,
+                                    "content": "As we have emphasized, the previous best performing models for each task all contain task-specific",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                533,
+                                506,
+                                545
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        533,
+                                        506,
+                                        545
+                                    ],
+                                    "score": 1.0,
+                                    "content": "design elements, whereas Aloe can be applied to all the tasks without modification. On CLEVRER,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                542,
+                                506,
+                                557
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        542,
+                                        433,
+                                        557
+                                    ],
+                                    "score": 1.0,
+                                    "content": "we also show that Aloe matches the performance of the previous best models with",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        433,
+                                        544,
+                                        453,
+                                        554
+                                    ],
+                                    "score": 0.85,
+                                    "content": "40 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        453,
+                                        542,
+                                        506,
+                                        557
+                                    ],
+                                    "score": 1.0,
+                                    "content": "less training",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                555,
+                                436,
+                                567
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        555,
+                                        436,
+                                        567
+                                    ],
+                                    "score": 1.0,
+                                    "content": "data, which demonstrates that our approach is data-efficient as well as performant.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        }
+                    ],
+                    "index": 36.5,
+                    "bbox_fs": [
+                        105,
+                        522,
+                        506,
+                        567
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        582,
+                        171,
+                        595
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                579,
+                                172,
+                                598
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        579,
+                                        172,
+                                        598
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2 Methods",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        }
+                    ],
+                    "index": 39
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        607,
+                        505,
+                        684
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                606,
+                                505,
+                                620
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        606,
+                                        505,
+                                        620
+                                    ],
+                                    "score": 1.0,
+                                    "content": "A guiding motivation for the design of Aloe is the converging evidence for the value of self-attention",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                618,
+                                505,
+                                630
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        618,
+                                        505,
+                                        630
+                                    ],
+                                    "score": 1.0,
+                                    "content": "mechanisms operating on a finite sequences of discrete entities. Written language is inherently",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                628,
+                                505,
+                                641
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        628,
+                                        505,
+                                        641
+                                    ],
+                                    "score": 1.0,
+                                    "content": "discrete and hence is well-suited to self-attention-based approaches. In other domains, such as raw",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                640,
+                                506,
+                                652
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        640,
+                                        506,
+                                        652
+                                    ],
+                                    "score": 1.0,
+                                    "content": "audio or vision, it is less clear how to leverage self-attention. We hypothesize that the application of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                649,
+                                507,
+                                664
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        649,
+                                        507,
+                                        664
+                                    ],
+                                    "score": 1.0,
+                                    "content": "self-attention-based models to visual tasks could benefit from an approximate ‘discretization’ process,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                661,
+                                506,
+                                675
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        661,
+                                        506,
+                                        675
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and determining the right level of discretization is an important choice that can significantly affect",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                673,
+                                189,
+                                685
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        673,
+                                        189,
+                                        685
+                                    ],
+                                    "score": 1.0,
+                                    "content": "model performance.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        }
+                    ],
+                    "index": 43,
+                    "bbox_fs": [
+                        105,
+                        606,
+                        507,
+                        685
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        689,
+                        505,
+                        722
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                689,
+                                505,
+                                701
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        689,
+                                        505,
+                                        701
+                                    ],
+                                    "score": 1.0,
+                                    "content": "At the finest level, data could simply be discretized into pixels (as is already the case for most",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 47
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                700,
+                                507,
+                                713
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        700,
+                                        507,
+                                        713
+                                    ],
+                                    "score": 1.0,
+                                    "content": "machine-processed visual data). Pixels are too fine-grained for many applications, however—for one,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 48
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                710,
+                                507,
+                                725
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        710,
+                                        507,
+                                        725
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the memory required to support self-attention across all pixels is prohibitive. Partly for this reason,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 49
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                290,
+                                506,
+                                303
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        290,
+                                        506,
+                                        303
+                                    ],
+                                    "score": 1.0,
+                                    "content": "coarser representations, such as the downsampled “hyper-pixel” outputs of a convolutional network,",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                302,
+                                506,
+                                314
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        302,
+                                        506,
+                                        314
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are often used instead (e.g. [27, 43]). In the case of videos, previous work considered even coarser",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                312,
+                                406,
+                                325
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        312,
+                                        406,
+                                        325
+                                    ],
+                                    "score": 1.0,
+                                    "content": "discretization schemes, such as frame or subclip level representations [35].",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 6
+                        }
+                    ],
+                    "index": 48,
+                    "bbox_fs": [
+                        105,
+                        689,
+                        507,
+                        725
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "image",
+                    "bbox": [
+                        154,
+                        74,
+                        466,
+                        238
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                154,
+                                74,
+                                466,
+                                238
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        154,
+                                        74,
+                                        466,
+                                        238
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                154,
+                                                74,
+                                                466,
+                                                238
+                                            ],
+                                            "score": 0.972,
+                                            "type": "image",
+                                            "image_path": "fb75c68d2f8c0da36c1f5ab1ea3c806bea16df8a72eccf44e82552a1d3195e0e.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 1,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        154,
+                                        74,
+                                        466,
+                                        128.66666666666666
+                                    ],
+                                    "spans": [],
+                                    "index": 0
+                                },
+                                {
+                                    "bbox": [
+                                        154,
+                                        128.66666666666666,
+                                        466,
+                                        183.33333333333331
+                                    ],
+                                    "spans": [],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        154,
+                                        183.33333333333331,
+                                        466,
+                                        237.99999999999997
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                149,
+                                243,
+                                462,
+                                255
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        147,
+                                        243,
+                                        463,
+                                        256
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                147,
+                                                243,
+                                                463,
+                                                256
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 1: A schematic of the model architecture. See the main text for details.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 3
+                                }
+                            ],
+                            "index": 3
+                        }
+                    ],
+                    "index": 2.0
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        290,
+                        505,
+                        324
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                290,
+                                506,
+                                303
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        290,
+                                        506,
+                                        303
+                                    ],
+                                    "score": 1.0,
+                                    "content": "coarser representations, such as the downsampled “hyper-pixel” outputs of a convolutional network,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                302,
+                                506,
+                                314
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        302,
+                                        506,
+                                        314
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are often used instead (e.g. [27, 43]). In the case of videos, previous work considered even coarser",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                312,
+                                406,
+                                325
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        312,
+                                        406,
+                                        325
+                                    ],
+                                    "score": 1.0,
+                                    "content": "discretization schemes, such as frame or subclip level representations [35].",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        }
+                    ],
+                    "index": 5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        329,
+                        505,
+                        438
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                329,
+                                506,
+                                342
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        329,
+                                        506,
+                                        342
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The neuroscience literature, however, suggests that biological visual systems infer and exploit the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                339,
+                                506,
+                                352
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        339,
+                                        506,
+                                        352
+                                    ],
+                                    "score": 1.0,
+                                    "content": "existence of objects, rather than spatial or temporal blocks with artificial boundaries [5, 30, 32].",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                351,
+                                505,
+                                362
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        351,
+                                        505,
+                                        362
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Because objects are the atomic units of physical interactions, it makes sense to discretize on the level",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                362,
+                                506,
+                                374
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        362,
+                                        506,
+                                        374
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of objects. Numerous object segmentation algorithms have been proposed [15, 19, 29]. We chose to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                372,
+                                506,
+                                385
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        372,
+                                        506,
+                                        385
+                                    ],
+                                    "score": 1.0,
+                                    "content": "use MONet, an unsupervised object segmentation algorithm [2]. Because MONet is unsupervised,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                384,
+                                506,
+                                396
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        384,
+                                        506,
+                                        396
+                                    ],
+                                    "score": 1.0,
+                                    "content": "we can train it directly in our domain of interest without the need for object segmentation labels. We",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                395,
+                                506,
+                                406
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        395,
+                                        506,
+                                        406
+                                    ],
+                                    "score": 1.0,
+                                    "content": "emphasize that our choice of MONet is an implementation detail, and in Appendix B, we show that",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                405,
+                                505,
+                                417
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        405,
+                                        505,
+                                        417
+                                    ],
+                                    "score": 1.0,
+                                    "content": "our framework of attention over learned object embeddings also works with other object-segmentation",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                416,
+                                505,
+                                429
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        416,
+                                        505,
+                                        429
+                                    ],
+                                    "score": 1.0,
+                                    "content": "schemes. We also do not need to place strong demands on the object segmentation algorithm, e.g. for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                427,
+                                370,
+                                439
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        427,
+                                        370,
+                                        439
+                                    ],
+                                    "score": 1.0,
+                                    "content": "it to produce aligned output or to have a built-in dynamics model.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        }
+                    ],
+                    "index": 11.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        443,
+                        505,
+                        498
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                442,
+                                506,
+                                456
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        442,
+                                        506,
+                                        456
+                                    ],
+                                    "score": 1.0,
+                                    "content": "To segment each frame into object representations, MONet uses a recurrent attention network to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                453,
+                                505,
+                                467
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        453,
+                                        164,
+                                        467
+                                    ],
+                                    "score": 1.0,
+                                    "content": "obtain a set of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        164,
+                                        455,
+                                        178,
+                                        465
+                                    ],
+                                    "score": 0.88,
+                                    "content": "N _ { o }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        178,
+                                        453,
+                                        281,
+                                        467
+                                    ],
+                                    "score": 1.0,
+                                    "content": "“object attention masks” (",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        281,
+                                        455,
+                                        294,
+                                        465
+                                    ],
+                                    "score": 0.87,
+                                    "content": "\\mathrm { \\Delta } N _ { o }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        295,
+                                        453,
+                                        505,
+                                        467
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is a fixed parameter). Each attention mask represents",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                465,
+                                505,
+                                478
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        465,
+                                        505,
+                                        478
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the probability that any given pixel belongs to that mask’s object. The pixels assigned to the mask are",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                474,
+                                507,
+                                490
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        474,
+                                        273,
+                                        490
+                                    ],
+                                    "score": 1.0,
+                                    "content": "encoded into latent variables with means",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        273,
+                                        475,
+                                        311,
+                                        487
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\mu _ { t i } \\in \\mathbb { R } ^ { d }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        311,
+                                        474,
+                                        342,
+                                        490
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", where",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        342,
+                                        477,
+                                        347,
+                                        486
+                                    ],
+                                    "score": 0.75,
+                                    "content": "i",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        347,
+                                        474,
+                                        457,
+                                        490
+                                    ],
+                                    "score": 1.0,
+                                    "content": "indexes the object slot and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        457,
+                                        477,
+                                        462,
+                                        486
+                                    ],
+                                    "score": 0.67,
+                                    "content": "t",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        462,
+                                        474,
+                                        507,
+                                        490
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the frame.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                486,
+                                507,
+                                500
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        486,
+                                        507,
+                                        500
+                                    ],
+                                    "score": 1.0,
+                                    "content": "These means are used as the object embeddings in Aloe. More details are provided in Appendix A.1.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        }
+                    ],
+                    "index": 19
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        503,
+                        505,
+                        580
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                502,
+                                505,
+                                516
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        502,
+                                        505,
+                                        516
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The self-attention component is a transformer model [37] operating on a sequence of vectors in",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                513,
+                                506,
+                                528
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        513,
+                                        119,
+                                        524
+                                    ],
+                                    "score": 0.87,
+                                    "content": "\\mathbb { R } ^ { d }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        120,
+                                        513,
+                                        233,
+                                        528
+                                    ],
+                                    "score": 1.0,
+                                    "content": ": the object representations",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        233,
+                                        516,
+                                        247,
+                                        526
+                                    ],
+                                    "score": 0.87,
+                                    "content": "\\mu _ { t i }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        247,
+                                        513,
+                                        276,
+                                        528
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for all",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        277,
+                                        515,
+                                        282,
+                                        524
+                                    ],
+                                    "score": 0.74,
+                                    "content": "t",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        282,
+                                        513,
+                                        301,
+                                        528
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        301,
+                                        515,
+                                        306,
+                                        524
+                                    ],
+                                    "score": 0.68,
+                                    "content": "i",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        306,
+                                        513,
+                                        385,
+                                        528
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", a trainable vector",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        385,
+                                        514,
+                                        434,
+                                        524
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\mathbf { \\bar { \\mathit { C L S } } } \\in \\mathbf { \\bar { \\mathbb { R } } ^ { d } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        434,
+                                        513,
+                                        506,
+                                        528
+                                    ],
+                                    "score": 1.0,
+                                    "content": "used to generate",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                525,
+                                505,
+                                537
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        525,
+                                        505,
+                                        537
+                                    ],
+                                    "score": 1.0,
+                                    "content": "classification results (analogous to the CLS token in BERT [9]), and (for CLEVRER) the embedded",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                536,
+                                504,
+                                549
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        536,
+                                        132,
+                                        549
+                                    ],
+                                    "score": 1.0,
+                                    "content": "words",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        132,
+                                        537,
+                                        145,
+                                        547
+                                    ],
+                                    "score": 0.88,
+                                    "content": "\\mathbf { w } _ { i }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        145,
+                                        536,
+                                        504,
+                                        549
+                                    ],
+                                    "score": 1.0,
+                                    "content": "from the question (and choice for multiple choice questions). For the object representations",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                547,
+                                506,
+                                560
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        549,
+                                        120,
+                                        559
+                                    ],
+                                    "score": 0.84,
+                                    "content": "\\mu _ { t i }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        120,
+                                        547,
+                                        210,
+                                        560
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and word embeddings",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        210,
+                                        549,
+                                        223,
+                                        558
+                                    ],
+                                    "score": 0.87,
+                                    "content": "\\mathbf { w } _ { i }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        223,
+                                        547,
+                                        416,
+                                        560
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", we append a two-dimensional one-hot vector to",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        416,
+                                        549,
+                                        429,
+                                        558
+                                    ],
+                                    "score": 0.87,
+                                    "content": "\\mu _ { t i }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        430,
+                                        547,
+                                        447,
+                                        560
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        447,
+                                        548,
+                                        460,
+                                        558
+                                    ],
+                                    "score": 0.87,
+                                    "content": "\\mathbf { w } _ { i }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        461,
+                                        547,
+                                        506,
+                                        560
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to indicate",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                558,
+                                507,
+                                570
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        558,
+                                        507,
+                                        570
+                                    ],
+                                    "score": 1.0,
+                                    "content": "whether the input is a word or an object. Because the transformer is shared between the modalities,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                569,
+                                480,
+                                581
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        569,
+                                        480,
+                                        581
+                                    ],
+                                    "score": 1.0,
+                                    "content": "information can flow between objects and words to solve the task, as we show in Section 3.1.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        }
+                    ],
+                    "index": 25
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        585,
+                        505,
+                        651
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                585,
+                                506,
+                                597
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        585,
+                                        346,
+                                        597
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We pass this sequence of vectors through a transformer with",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        346,
+                                        586,
+                                        361,
+                                        596
+                                    ],
+                                    "score": 0.89,
+                                    "content": "N _ { T }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        362,
+                                        585,
+                                        506,
+                                        597
+                                    ],
+                                    "score": 1.0,
+                                    "content": "layers. All inputs are first projected",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                594,
+                                507,
+                                609
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        594,
+                                        276,
+                                        609
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(via a linear layer and ReLU activation) to",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        277,
+                                        595,
+                                        308,
+                                        606
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\mathbb { R } ^ { N _ { H } \\times d }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        309,
+                                        594,
+                                        339,
+                                        609
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", where",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        339,
+                                        597,
+                                        356,
+                                        607
+                                    ],
+                                    "score": 0.89,
+                                    "content": "N _ { H }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        356,
+                                        594,
+                                        507,
+                                        609
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is the number of self-attention heads.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                606,
+                                506,
+                                619
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        606,
+                                        506,
+                                        619
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We add a relative sinusoidal positional encoding at each layer of the transformer to give the model",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                616,
+                                505,
+                                631
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        616,
+                                        395,
+                                        631
+                                    ],
+                                    "score": 1.0,
+                                    "content": "knowledge of the word and frame order [7]. The transformed value of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        396,
+                                        618,
+                                        418,
+                                        628
+                                    ],
+                                    "score": 0.78,
+                                    "content": "C L S",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        418,
+                                        616,
+                                        505,
+                                        631
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is passed through an",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                628,
+                                505,
+                                642
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        628,
+                                        250,
+                                        642
+                                    ],
+                                    "score": 1.0,
+                                    "content": "MLP (with one hidden layer of size",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        251,
+                                        629,
+                                        267,
+                                        640
+                                    ],
+                                    "score": 0.86,
+                                    "content": "N _ { H }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        268,
+                                        628,
+                                        505,
+                                        642
+                                    ],
+                                    "score": 1.0,
+                                    "content": ") to generate the final answer. A schema of our architecture",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                640,
+                                193,
+                                652
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        640,
+                                        193,
+                                        652
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is shown in Figure 1.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        }
+                    ],
+                    "index": 31.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        656,
+                        505,
+                        722
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                655,
+                                505,
+                                669
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        655,
+                                        505,
+                                        669
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Note that in the model presented above (which we call global attention), the transformer sees no",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                666,
+                                506,
+                                679
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        666,
+                                        506,
+                                        679
+                                    ],
+                                    "score": 1.0,
+                                    "content": "distinction between objects of different frames (other than through the position encoding). Another",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                677,
+                                505,
+                                690
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        677,
+                                        505,
+                                        690
+                                    ],
+                                    "score": 1.0,
+                                    "content": "intuitive choice, which we call hierarchical attention, is to have one transformer acting on the objects",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                689,
+                                505,
+                                701
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        689,
+                                        505,
+                                        701
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of each frame independently, and another transformer acting on the concatenated outputs of the first",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                700,
+                                506,
+                                713
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        700,
+                                        506,
+                                        713
+                                    ],
+                                    "score": 1.0,
+                                    "content": "transformer (this temporal division of input data is commonly used, e.g. in [35]). In pseudo-code,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                711,
+                                252,
+                                723
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        711,
+                                        252,
+                                        723
+                                    ],
+                                    "score": 1.0,
+                                    "content": "global attention can be expressed as",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        }
+                    ],
+                    "index": 37.5
+                }
+            ],
+            "page_idx": 2,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        302,
+                        741,
+                        309,
+                        750
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                301,
+                                740,
+                                309,
+                                752
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        301,
+                                        740,
+                                        309,
+                                        752
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "image",
+                    "bbox": [
+                        154,
+                        74,
+                        466,
+                        238
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                154,
+                                74,
+                                466,
+                                238
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        154,
+                                        74,
+                                        466,
+                                        238
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                154,
+                                                74,
+                                                466,
+                                                238
+                                            ],
+                                            "score": 0.972,
+                                            "type": "image",
+                                            "image_path": "fb75c68d2f8c0da36c1f5ab1ea3c806bea16df8a72eccf44e82552a1d3195e0e.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 1,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        154,
+                                        74,
+                                        466,
+                                        128.66666666666666
+                                    ],
+                                    "spans": [],
+                                    "index": 0
+                                },
+                                {
+                                    "bbox": [
+                                        154,
+                                        128.66666666666666,
+                                        466,
+                                        183.33333333333331
+                                    ],
+                                    "spans": [],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        154,
+                                        183.33333333333331,
+                                        466,
+                                        237.99999999999997
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                149,
+                                243,
+                                462,
+                                255
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        147,
+                                        243,
+                                        463,
+                                        256
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                147,
+                                                243,
+                                                463,
+                                                256
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 1: A schematic of the model architecture. See the main text for details.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 3
+                                }
+                            ],
+                            "index": 3
+                        }
+                    ],
+                    "index": 2.0
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        290,
+                        505,
+                        324
+                    ],
+                    "lines": [],
+                    "index": 5,
+                    "bbox_fs": [
+                        105,
+                        290,
+                        506,
+                        325
+                    ],
+                    "lines_deleted": true
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        329,
+                        505,
+                        438
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                329,
+                                506,
+                                342
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        329,
+                                        506,
+                                        342
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The neuroscience literature, however, suggests that biological visual systems infer and exploit the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                339,
+                                506,
+                                352
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        339,
+                                        506,
+                                        352
+                                    ],
+                                    "score": 1.0,
+                                    "content": "existence of objects, rather than spatial or temporal blocks with artificial boundaries [5, 30, 32].",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                351,
+                                505,
+                                362
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        351,
+                                        505,
+                                        362
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Because objects are the atomic units of physical interactions, it makes sense to discretize on the level",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                362,
+                                506,
+                                374
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        362,
+                                        506,
+                                        374
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of objects. Numerous object segmentation algorithms have been proposed [15, 19, 29]. We chose to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                372,
+                                506,
+                                385
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        372,
+                                        506,
+                                        385
+                                    ],
+                                    "score": 1.0,
+                                    "content": "use MONet, an unsupervised object segmentation algorithm [2]. Because MONet is unsupervised,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                384,
+                                506,
+                                396
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        384,
+                                        506,
+                                        396
+                                    ],
+                                    "score": 1.0,
+                                    "content": "we can train it directly in our domain of interest without the need for object segmentation labels. We",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                395,
+                                506,
+                                406
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        395,
+                                        506,
+                                        406
+                                    ],
+                                    "score": 1.0,
+                                    "content": "emphasize that our choice of MONet is an implementation detail, and in Appendix B, we show that",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                405,
+                                505,
+                                417
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        405,
+                                        505,
+                                        417
+                                    ],
+                                    "score": 1.0,
+                                    "content": "our framework of attention over learned object embeddings also works with other object-segmentation",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                416,
+                                505,
+                                429
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        416,
+                                        505,
+                                        429
+                                    ],
+                                    "score": 1.0,
+                                    "content": "schemes. We also do not need to place strong demands on the object segmentation algorithm, e.g. for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                427,
+                                370,
+                                439
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        427,
+                                        370,
+                                        439
+                                    ],
+                                    "score": 1.0,
+                                    "content": "it to produce aligned output or to have a built-in dynamics model.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        }
+                    ],
+                    "index": 11.5,
+                    "bbox_fs": [
+                        105,
+                        329,
+                        506,
+                        439
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        443,
+                        505,
+                        498
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                442,
+                                506,
+                                456
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        442,
+                                        506,
+                                        456
+                                    ],
+                                    "score": 1.0,
+                                    "content": "To segment each frame into object representations, MONet uses a recurrent attention network to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                453,
+                                505,
+                                467
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        453,
+                                        164,
+                                        467
+                                    ],
+                                    "score": 1.0,
+                                    "content": "obtain a set of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        164,
+                                        455,
+                                        178,
+                                        465
+                                    ],
+                                    "score": 0.88,
+                                    "content": "N _ { o }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        178,
+                                        453,
+                                        281,
+                                        467
+                                    ],
+                                    "score": 1.0,
+                                    "content": "“object attention masks” (",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        281,
+                                        455,
+                                        294,
+                                        465
+                                    ],
+                                    "score": 0.87,
+                                    "content": "\\mathrm { \\Delta } N _ { o }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        295,
+                                        453,
+                                        505,
+                                        467
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is a fixed parameter). Each attention mask represents",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                465,
+                                505,
+                                478
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        465,
+                                        505,
+                                        478
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the probability that any given pixel belongs to that mask’s object. The pixels assigned to the mask are",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                474,
+                                507,
+                                490
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        474,
+                                        273,
+                                        490
+                                    ],
+                                    "score": 1.0,
+                                    "content": "encoded into latent variables with means",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        273,
+                                        475,
+                                        311,
+                                        487
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\mu _ { t i } \\in \\mathbb { R } ^ { d }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        311,
+                                        474,
+                                        342,
+                                        490
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", where",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        342,
+                                        477,
+                                        347,
+                                        486
+                                    ],
+                                    "score": 0.75,
+                                    "content": "i",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        347,
+                                        474,
+                                        457,
+                                        490
+                                    ],
+                                    "score": 1.0,
+                                    "content": "indexes the object slot and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        457,
+                                        477,
+                                        462,
+                                        486
+                                    ],
+                                    "score": 0.67,
+                                    "content": "t",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        462,
+                                        474,
+                                        507,
+                                        490
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the frame.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                486,
+                                507,
+                                500
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        486,
+                                        507,
+                                        500
+                                    ],
+                                    "score": 1.0,
+                                    "content": "These means are used as the object embeddings in Aloe. More details are provided in Appendix A.1.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        }
+                    ],
+                    "index": 19,
+                    "bbox_fs": [
+                        105,
+                        442,
+                        507,
+                        500
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        503,
+                        505,
+                        580
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                502,
+                                505,
+                                516
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        502,
+                                        505,
+                                        516
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The self-attention component is a transformer model [37] operating on a sequence of vectors in",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                513,
+                                506,
+                                528
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        513,
+                                        119,
+                                        524
+                                    ],
+                                    "score": 0.87,
+                                    "content": "\\mathbb { R } ^ { d }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        120,
+                                        513,
+                                        233,
+                                        528
+                                    ],
+                                    "score": 1.0,
+                                    "content": ": the object representations",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        233,
+                                        516,
+                                        247,
+                                        526
+                                    ],
+                                    "score": 0.87,
+                                    "content": "\\mu _ { t i }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        247,
+                                        513,
+                                        276,
+                                        528
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for all",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        277,
+                                        515,
+                                        282,
+                                        524
+                                    ],
+                                    "score": 0.74,
+                                    "content": "t",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        282,
+                                        513,
+                                        301,
+                                        528
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        301,
+                                        515,
+                                        306,
+                                        524
+                                    ],
+                                    "score": 0.68,
+                                    "content": "i",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        306,
+                                        513,
+                                        385,
+                                        528
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", a trainable vector",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        385,
+                                        514,
+                                        434,
+                                        524
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\mathbf { \\bar { \\mathit { C L S } } } \\in \\mathbf { \\bar { \\mathbb { R } } ^ { d } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        434,
+                                        513,
+                                        506,
+                                        528
+                                    ],
+                                    "score": 1.0,
+                                    "content": "used to generate",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                525,
+                                505,
+                                537
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        525,
+                                        505,
+                                        537
+                                    ],
+                                    "score": 1.0,
+                                    "content": "classification results (analogous to the CLS token in BERT [9]), and (for CLEVRER) the embedded",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                536,
+                                504,
+                                549
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        536,
+                                        132,
+                                        549
+                                    ],
+                                    "score": 1.0,
+                                    "content": "words",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        132,
+                                        537,
+                                        145,
+                                        547
+                                    ],
+                                    "score": 0.88,
+                                    "content": "\\mathbf { w } _ { i }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        145,
+                                        536,
+                                        504,
+                                        549
+                                    ],
+                                    "score": 1.0,
+                                    "content": "from the question (and choice for multiple choice questions). For the object representations",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                547,
+                                506,
+                                560
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        549,
+                                        120,
+                                        559
+                                    ],
+                                    "score": 0.84,
+                                    "content": "\\mu _ { t i }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        120,
+                                        547,
+                                        210,
+                                        560
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and word embeddings",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        210,
+                                        549,
+                                        223,
+                                        558
+                                    ],
+                                    "score": 0.87,
+                                    "content": "\\mathbf { w } _ { i }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        223,
+                                        547,
+                                        416,
+                                        560
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", we append a two-dimensional one-hot vector to",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        416,
+                                        549,
+                                        429,
+                                        558
+                                    ],
+                                    "score": 0.87,
+                                    "content": "\\mu _ { t i }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        430,
+                                        547,
+                                        447,
+                                        560
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        447,
+                                        548,
+                                        460,
+                                        558
+                                    ],
+                                    "score": 0.87,
+                                    "content": "\\mathbf { w } _ { i }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        461,
+                                        547,
+                                        506,
+                                        560
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to indicate",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                558,
+                                507,
+                                570
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        558,
+                                        507,
+                                        570
+                                    ],
+                                    "score": 1.0,
+                                    "content": "whether the input is a word or an object. Because the transformer is shared between the modalities,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                569,
+                                480,
+                                581
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        569,
+                                        480,
+                                        581
+                                    ],
+                                    "score": 1.0,
+                                    "content": "information can flow between objects and words to solve the task, as we show in Section 3.1.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        }
+                    ],
+                    "index": 25,
+                    "bbox_fs": [
+                        105,
+                        502,
+                        507,
+                        581
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        585,
+                        505,
+                        651
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                585,
+                                506,
+                                597
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        585,
+                                        346,
+                                        597
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We pass this sequence of vectors through a transformer with",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        346,
+                                        586,
+                                        361,
+                                        596
+                                    ],
+                                    "score": 0.89,
+                                    "content": "N _ { T }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        362,
+                                        585,
+                                        506,
+                                        597
+                                    ],
+                                    "score": 1.0,
+                                    "content": "layers. All inputs are first projected",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                594,
+                                507,
+                                609
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        594,
+                                        276,
+                                        609
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(via a linear layer and ReLU activation) to",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        277,
+                                        595,
+                                        308,
+                                        606
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\mathbb { R } ^ { N _ { H } \\times d }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        309,
+                                        594,
+                                        339,
+                                        609
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", where",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        339,
+                                        597,
+                                        356,
+                                        607
+                                    ],
+                                    "score": 0.89,
+                                    "content": "N _ { H }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        356,
+                                        594,
+                                        507,
+                                        609
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is the number of self-attention heads.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                606,
+                                506,
+                                619
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        606,
+                                        506,
+                                        619
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We add a relative sinusoidal positional encoding at each layer of the transformer to give the model",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                616,
+                                505,
+                                631
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        616,
+                                        395,
+                                        631
+                                    ],
+                                    "score": 1.0,
+                                    "content": "knowledge of the word and frame order [7]. The transformed value of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        396,
+                                        618,
+                                        418,
+                                        628
+                                    ],
+                                    "score": 0.78,
+                                    "content": "C L S",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        418,
+                                        616,
+                                        505,
+                                        631
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is passed through an",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                628,
+                                505,
+                                642
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        628,
+                                        250,
+                                        642
+                                    ],
+                                    "score": 1.0,
+                                    "content": "MLP (with one hidden layer of size",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        251,
+                                        629,
+                                        267,
+                                        640
+                                    ],
+                                    "score": 0.86,
+                                    "content": "N _ { H }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        268,
+                                        628,
+                                        505,
+                                        642
+                                    ],
+                                    "score": 1.0,
+                                    "content": ") to generate the final answer. A schema of our architecture",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                640,
+                                193,
+                                652
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        640,
+                                        193,
+                                        652
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is shown in Figure 1.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        }
+                    ],
+                    "index": 31.5,
+                    "bbox_fs": [
+                        104,
+                        585,
+                        507,
+                        652
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        656,
+                        505,
+                        722
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                655,
+                                505,
+                                669
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        655,
+                                        505,
+                                        669
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Note that in the model presented above (which we call global attention), the transformer sees no",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                666,
+                                506,
+                                679
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        666,
+                                        506,
+                                        679
+                                    ],
+                                    "score": 1.0,
+                                    "content": "distinction between objects of different frames (other than through the position encoding). Another",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                677,
+                                505,
+                                690
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        677,
+                                        505,
+                                        690
+                                    ],
+                                    "score": 1.0,
+                                    "content": "intuitive choice, which we call hierarchical attention, is to have one transformer acting on the objects",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                689,
+                                505,
+                                701
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        689,
+                                        505,
+                                        701
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of each frame independently, and another transformer acting on the concatenated outputs of the first",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                700,
+                                506,
+                                713
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        700,
+                                        506,
+                                        713
+                                    ],
+                                    "score": 1.0,
+                                    "content": "transformer (this temporal division of input data is commonly used, e.g. in [35]). In pseudo-code,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                711,
+                                252,
+                                723
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        711,
+                                        252,
+                                        723
+                                    ],
+                                    "score": 1.0,
+                                    "content": "global attention can be expressed as",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        }
+                    ],
+                    "index": 37.5,
+                    "bbox_fs": [
+                        105,
+                        655,
+                        506,
+                        723
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "image",
+                    "bbox": [
+                        107,
+                        78,
+                        502,
+                        258
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                107,
+                                78,
+                                502,
+                                258
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        78,
+                                        502,
+                                        258
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                107,
+                                                78,
+                                                502,
+                                                258
+                                            ],
+                                            "score": 0.97,
+                                            "type": "image",
+                                            "image_path": "8b8974247844831b2889dcf142e48a4d89360d0c23450734c7b5e9c93722a7eb.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 1,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        78,
+                                        502,
+                                        138.0
+                                    ],
+                                    "spans": [],
+                                    "index": 0
+                                },
+                                {
+                                    "bbox": [
+                                        107,
+                                        138.0,
+                                        502,
+                                        198.0
+                                    ],
+                                    "spans": [],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        107,
+                                        198.0,
+                                        502,
+                                        258.0
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                140,
+                                269,
+                                470,
+                                281
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        138,
+                                        267,
+                                        472,
+                                        283
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                138,
+                                                267,
+                                                472,
+                                                283
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 2: Different masking schemes for self-supervised learning applied to Aloe.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 3
+                                }
+                            ],
+                            "index": 3
+                        }
+                    ],
+                    "index": 2.0
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        127,
+                        305,
+                        393,
+                        316
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                128,
+                                304,
+                                393,
+                                317
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        128,
+                                        304,
+                                        148,
+                                        317
+                                    ],
+                                    "score": 1.0,
+                                    "content": "out",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        148,
+                                        306,
+                                        158,
+                                        314
+                                    ],
+                                    "score": 0.76,
+                                    "content": "=",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        158,
+                                        304,
+                                        393,
+                                        317
+                                    ],
+                                    "score": 1.0,
+                                    "content": "transformer(reshape(objects, [B, F * N, D])",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        }
+                    ],
+                    "index": 4
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        321,
+                        217,
+                        332
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                107,
+                                321,
+                                218,
+                                333
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        321,
+                                        218,
+                                        333
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and hiearchical attention as",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        }
+                    ],
+                    "index": 5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        127,
+                        337,
+                        401,
+                        364
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                128,
+                                337,
+                                403,
+                                348
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        128,
+                                        337,
+                                        148,
+                                        348
+                                    ],
+                                    "score": 1.0,
+                                    "content": "out",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        148,
+                                        339,
+                                        158,
+                                        347
+                                    ],
+                                    "score": 0.8,
+                                    "content": "=",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        158,
+                                        337,
+                                        403,
+                                        348
+                                    ],
+                                    "score": 1.0,
+                                    "content": "transformer1(reshape(objects, [B * F, N, D]))",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                353,
+                                392,
+                                365
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        353,
+                                        147,
+                                        365
+                                    ],
+                                    "score": 1.0,
+                                    "content": "out",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        148,
+                                        354,
+                                        158,
+                                        363
+                                    ],
+                                    "score": 0.8,
+                                    "content": "=",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        159,
+                                        353,
+                                        348,
+                                        365
+                                    ],
+                                    "score": 1.0,
+                                    "content": "transformer2(reshape(out, [B, F, N",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        349,
+                                        355,
+                                        356,
+                                        362
+                                    ],
+                                    "score": 0.25,
+                                    "content": "\\star",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        356,
+                                        353,
+                                        392,
+                                        365
+                                    ],
+                                    "score": 1.0,
+                                    "content": "D])) .",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        }
+                    ],
+                    "index": 6.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        369,
+                        505,
+                        392
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                369,
+                                505,
+                                381
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        369,
+                                        505,
+                                        381
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We study the importance of global attention (objects as the atomic entities) vs hierarchical attention",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                380,
+                                486,
+                                393
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        380,
+                                        486,
+                                        393
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(objects, and subsequently frames as the atomic entities). The comparison is shown in Table 1.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        }
+                    ],
+                    "index": 8.5
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        406,
+                        233,
+                        418
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                403,
+                                235,
+                                421
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        403,
+                                        235,
+                                        421
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2.1 Self-supervised learning",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        }
+                    ],
+                    "index": 10
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        426,
+                        505,
+                        493
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                427,
+                                505,
+                                440
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        427,
+                                        505,
+                                        440
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We explored whether self-supervised learning could improve the performance of Aloe beyond the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                438,
+                                506,
+                                451
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        438,
+                                        506,
+                                        451
+                                    ],
+                                    "score": 1.0,
+                                    "content": "benefits conveyed by object-level representation, i.e. in ways that support the model’s interpretation",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                448,
+                                505,
+                                462
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        448,
+                                        505,
+                                        462
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of scene dynamics rather than just via improved perception of static observations. Our approach is",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                460,
+                                505,
+                                472
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        460,
+                                        505,
+                                        472
+                                    ],
+                                    "score": 1.0,
+                                    "content": "inspired by the loss used in BERT [9], where a transformer model is trained to predict certain words",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                470,
+                                506,
+                                483
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        470,
+                                        506,
+                                        483
+                                    ],
+                                    "score": 1.0,
+                                    "content": "that are masked from the input. In our case, we mask object embeddings, and train the model to infer",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                480,
+                                470,
+                                494
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        480,
+                                        470,
+                                        494
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the content of the masked object representations using its knowledge of unmasked objects.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        }
+                    ],
+                    "index": 13.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        497,
+                        505,
+                        553
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                497,
+                                506,
+                                510
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        498,
+                                        342,
+                                        510
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Concretely, during training, we multiply each MONet latent",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        342,
+                                        500,
+                                        356,
+                                        509
+                                    ],
+                                    "score": 0.86,
+                                    "content": "\\mu _ { t i }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        356,
+                                        498,
+                                        450,
+                                        510
+                                    ],
+                                    "score": 1.0,
+                                    "content": "by a masking indicator,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        451,
+                                        497,
+                                        503,
+                                        510
+                                    ],
+                                    "score": 0.93,
+                                    "content": "m _ { t i } \\in \\{ 0 , 1 \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        503,
+                                        498,
+                                        506,
+                                        510
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                509,
+                                506,
+                                522
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        509,
+                                        123,
+                                        522
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Let",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        123,
+                                        509,
+                                        137,
+                                        521
+                                    ],
+                                    "score": 0.9,
+                                    "content": "\\mu _ { t i } ^ { \\prime }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        137,
+                                        509,
+                                        254,
+                                        522
+                                    ],
+                                    "score": 1.0,
+                                    "content": "be the transformed value of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        255,
+                                        510,
+                                        283,
+                                        520
+                                    ],
+                                    "score": 0.9,
+                                    "content": "m _ { t i } \\mu _ { t i }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        284,
+                                        509,
+                                        506,
+                                        522
+                                    ],
+                                    "score": 1.0,
+                                    "content": "after passing through the transformer. We expect the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                519,
+                                506,
+                                533
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        519,
+                                        466,
+                                        533
+                                    ],
+                                    "score": 1.0,
+                                    "content": "transformer to understand the underlying dynamics of the video, so that the masked out slot",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        466,
+                                        521,
+                                        480,
+                                        532
+                                    ],
+                                    "score": 0.86,
+                                    "content": "\\mu _ { t i }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        480,
+                                        519,
+                                        506,
+                                        533
+                                    ],
+                                    "score": 1.0,
+                                    "content": "could",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                531,
+                                506,
+                                544
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        531,
+                                        180,
+                                        544
+                                    ],
+                                    "score": 1.0,
+                                    "content": "be predicted from",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        180,
+                                        531,
+                                        193,
+                                        543
+                                    ],
+                                    "score": 0.89,
+                                    "content": "\\mu _ { t i } ^ { \\prime }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        194,
+                                        531,
+                                        506,
+                                        544
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". To guide the transformer in learning effective representations capable of this",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                541,
+                                325,
+                                555
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        541,
+                                        325,
+                                        555
+                                    ],
+                                    "score": 1.0,
+                                    "content": "type of dynamics prediction, we add an auxiliary loss:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        }
+                    ],
+                    "index": 19
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        231,
+                        560,
+                        379,
+                        588
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                231,
+                                560,
+                                379,
+                                588
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        231,
+                                        560,
+                                        379,
+                                        588
+                                    ],
+                                    "score": 0.94,
+                                    "content": "{ \\mathrm { a u x i l i a r y ~ l o s s } } = \\sum _ { t , i } { \\tau _ { t i } l \\left( f ( \\mu _ { t i } ^ { \\prime } ) , \\mu \\right) } ,",
+                                    "type": "interline_equation",
+                                    "image_path": "ec34b10eb5dbd048dcd818839d9d4c52bdbb686a32df86c0e869749c1ec07022.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 22.5,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                231,
+                                560,
+                                379,
+                                574.0
+                            ],
+                            "spans": [],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                231,
+                                574.0,
+                                379,
+                                588.0
+                            ],
+                            "spans": [],
+                            "index": 23
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        595,
+                        505,
+                        663
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                595,
+                                506,
+                                609
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        595,
+                                        133,
+                                        609
+                                    ],
+                                    "score": 1.0,
+                                    "content": "where",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        133,
+                                        597,
+                                        141,
+                                        608
+                                    ],
+                                    "score": 0.86,
+                                    "content": "f",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        141,
+                                        595,
+                                        264,
+                                        609
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is a learned linear mapping to",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        264,
+                                        595,
+                                        277,
+                                        606
+                                    ],
+                                    "score": 0.74,
+                                    "content": "\\mathbb { R } ^ { d }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        278,
+                                        595,
+                                        281,
+                                        609
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        281,
+                                        596,
+                                        286,
+                                        606
+                                    ],
+                                    "score": 0.4,
+                                    "content": "l",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        286,
+                                        595,
+                                        368,
+                                        609
+                                    ],
+                                    "score": 1.0,
+                                    "content": "a loss function, and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        369,
+                                        596,
+                                        417,
+                                        608
+                                    ],
+                                    "score": 0.94,
+                                    "content": "\\tau _ { t i } \\in \\{ 0 , 1 \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        417,
+                                        595,
+                                        506,
+                                        609
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are one-hot indicator",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                606,
+                                506,
+                                621
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        606,
+                                        506,
+                                        621
+                                    ],
+                                    "score": 1.0,
+                                    "content": "variables identifying the prediction targets (not necessarily just the masked out entries, since the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                618,
+                                506,
+                                631
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        618,
+                                        506,
+                                        631
+                                    ],
+                                    "score": 1.0,
+                                    "content": "prediction targets could be a subset of the masked out entries). We propagate gradients only to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                629,
+                                505,
+                                641
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        629,
+                                        180,
+                                        641
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the parameters of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        181,
+                                        629,
+                                        188,
+                                        641
+                                    ],
+                                    "score": 0.85,
+                                    "content": "f",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        188,
+                                        629,
+                                        405,
+                                        641
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and the transformer and not to the learned word and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        405,
+                                        629,
+                                        428,
+                                        639
+                                    ],
+                                    "score": 0.59,
+                                    "content": "C L S",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        428,
+                                        629,
+                                        505,
+                                        641
+                                    ],
+                                    "score": 1.0,
+                                    "content": "embeddings. This",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                640,
+                                506,
+                                652
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        640,
+                                        374,
+                                        652
+                                    ],
+                                    "score": 1.0,
+                                    "content": "auxiliary loss is added to the main classification loss with weighting",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        374,
+                                        640,
+                                        381,
+                                        650
+                                    ],
+                                    "score": 0.71,
+                                    "content": "\\lambda",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        381,
+                                        640,
+                                        506,
+                                        652
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", and both losses are minimized",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                650,
+                                472,
+                                663
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        650,
+                                        472,
+                                        663
+                                    ],
+                                    "score": 1.0,
+                                    "content": "simultaneously by the optimizer. We do not pretrain the model with only the auxiliary loss.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        }
+                    ],
+                    "index": 26.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        667,
+                        505,
+                        722
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                666,
+                                505,
+                                679
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        666,
+                                        280,
+                                        679
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We tested two different loss functions for",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        280,
+                                        668,
+                                        284,
+                                        677
+                                    ],
+                                    "score": 0.5,
+                                    "content": "l",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        285,
+                                        666,
+                                        505,
+                                        679
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", an L2 loss and a contrastive loss (formulas given in",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                678,
+                                506,
+                                690
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        678,
+                                        353,
+                                        690
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Appendix A.2), and six different masking schemes (settings of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        354,
+                                        679,
+                                        370,
+                                        689
+                                    ],
+                                    "score": 0.88,
+                                    "content": "m _ { t i }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        370,
+                                        678,
+                                        387,
+                                        690
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        388,
+                                        679,
+                                        400,
+                                        689
+                                    ],
+                                    "score": 0.85,
+                                    "content": "\\tau _ { t i }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        400,
+                                        678,
+                                        506,
+                                        690
+                                    ],
+                                    "score": 1.0,
+                                    "content": "), as illustrated in Figure 2.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                688,
+                                506,
+                                702
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        688,
+                                        506,
+                                        702
+                                    ],
+                                    "score": 1.0,
+                                    "content": "This exploration was motivated by the observation that video inputs at adjacent timesteps are highly",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                699,
+                                506,
+                                713
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        699,
+                                        506,
+                                        713
+                                    ],
+                                    "score": 1.0,
+                                    "content": "correlated in a way that adjacent words are not. We thus hypothesized that BERT-style prediction of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                710,
+                                505,
+                                724
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        710,
+                                        505,
+                                        724
+                                    ],
+                                    "score": 1.0,
+                                    "content": "adjacent words might not be optimal. A different masking strategy, in which prediction targets are",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        }
+                    ],
+                    "index": 32
+                }
+            ],
+            "page_idx": 3,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        302,
+                        741,
+                        308,
+                        750
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                302,
+                                741,
+                                310,
+                                752
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        302,
+                                        741,
+                                        310,
+                                        752
+                                    ],
+                                    "score": 1.0,
+                                    "content": "4",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "image",
+                    "bbox": [
+                        107,
+                        78,
+                        502,
+                        258
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                107,
+                                78,
+                                502,
+                                258
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        78,
+                                        502,
+                                        258
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                107,
+                                                78,
+                                                502,
+                                                258
+                                            ],
+                                            "score": 0.97,
+                                            "type": "image",
+                                            "image_path": "8b8974247844831b2889dcf142e48a4d89360d0c23450734c7b5e9c93722a7eb.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 1,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        78,
+                                        502,
+                                        138.0
+                                    ],
+                                    "spans": [],
+                                    "index": 0
+                                },
+                                {
+                                    "bbox": [
+                                        107,
+                                        138.0,
+                                        502,
+                                        198.0
+                                    ],
+                                    "spans": [],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        107,
+                                        198.0,
+                                        502,
+                                        258.0
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                140,
+                                269,
+                                470,
+                                281
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        138,
+                                        267,
+                                        472,
+                                        283
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                138,
+                                                267,
+                                                472,
+                                                283
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 2: Different masking schemes for self-supervised learning applied to Aloe.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 3
+                                }
+                            ],
+                            "index": 3
+                        }
+                    ],
+                    "index": 2.0
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        127,
+                        305,
+                        393,
+                        316
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                128,
+                                304,
+                                393,
+                                317
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        128,
+                                        304,
+                                        148,
+                                        317
+                                    ],
+                                    "score": 1.0,
+                                    "content": "out",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        148,
+                                        306,
+                                        158,
+                                        314
+                                    ],
+                                    "score": 0.76,
+                                    "content": "=",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        158,
+                                        304,
+                                        393,
+                                        317
+                                    ],
+                                    "score": 1.0,
+                                    "content": "transformer(reshape(objects, [B, F * N, D])",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        }
+                    ],
+                    "index": 4,
+                    "bbox_fs": [
+                        128,
+                        304,
+                        393,
+                        317
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        321,
+                        217,
+                        332
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                107,
+                                321,
+                                218,
+                                333
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        321,
+                                        218,
+                                        333
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and hiearchical attention as",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        }
+                    ],
+                    "index": 5,
+                    "bbox_fs": [
+                        107,
+                        321,
+                        218,
+                        333
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        127,
+                        337,
+                        401,
+                        364
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                128,
+                                337,
+                                403,
+                                348
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        128,
+                                        337,
+                                        148,
+                                        348
+                                    ],
+                                    "score": 1.0,
+                                    "content": "out",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        148,
+                                        339,
+                                        158,
+                                        347
+                                    ],
+                                    "score": 0.8,
+                                    "content": "=",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        158,
+                                        337,
+                                        403,
+                                        348
+                                    ],
+                                    "score": 1.0,
+                                    "content": "transformer1(reshape(objects, [B * F, N, D]))",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                353,
+                                392,
+                                365
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        353,
+                                        147,
+                                        365
+                                    ],
+                                    "score": 1.0,
+                                    "content": "out",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        148,
+                                        354,
+                                        158,
+                                        363
+                                    ],
+                                    "score": 0.8,
+                                    "content": "=",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        159,
+                                        353,
+                                        348,
+                                        365
+                                    ],
+                                    "score": 1.0,
+                                    "content": "transformer2(reshape(out, [B, F, N",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        349,
+                                        355,
+                                        356,
+                                        362
+                                    ],
+                                    "score": 0.25,
+                                    "content": "\\star",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        356,
+                                        353,
+                                        392,
+                                        365
+                                    ],
+                                    "score": 1.0,
+                                    "content": "D])) .",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        }
+                    ],
+                    "index": 6.5,
+                    "bbox_fs": [
+                        127,
+                        337,
+                        403,
+                        365
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        369,
+                        505,
+                        392
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                369,
+                                505,
+                                381
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        369,
+                                        505,
+                                        381
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We study the importance of global attention (objects as the atomic entities) vs hierarchical attention",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                380,
+                                486,
+                                393
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        380,
+                                        486,
+                                        393
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(objects, and subsequently frames as the atomic entities). The comparison is shown in Table 1.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        }
+                    ],
+                    "index": 8.5,
+                    "bbox_fs": [
+                        106,
+                        369,
+                        505,
+                        393
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        406,
+                        233,
+                        418
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                403,
+                                235,
+                                421
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        403,
+                                        235,
+                                        421
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2.1 Self-supervised learning",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        }
+                    ],
+                    "index": 10
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        426,
+                        505,
+                        493
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                427,
+                                505,
+                                440
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        427,
+                                        505,
+                                        440
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We explored whether self-supervised learning could improve the performance of Aloe beyond the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                438,
+                                506,
+                                451
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        438,
+                                        506,
+                                        451
+                                    ],
+                                    "score": 1.0,
+                                    "content": "benefits conveyed by object-level representation, i.e. in ways that support the model’s interpretation",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                448,
+                                505,
+                                462
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        448,
+                                        505,
+                                        462
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of scene dynamics rather than just via improved perception of static observations. Our approach is",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                460,
+                                505,
+                                472
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        460,
+                                        505,
+                                        472
+                                    ],
+                                    "score": 1.0,
+                                    "content": "inspired by the loss used in BERT [9], where a transformer model is trained to predict certain words",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                470,
+                                506,
+                                483
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        470,
+                                        506,
+                                        483
+                                    ],
+                                    "score": 1.0,
+                                    "content": "that are masked from the input. In our case, we mask object embeddings, and train the model to infer",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                480,
+                                470,
+                                494
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        480,
+                                        470,
+                                        494
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the content of the masked object representations using its knowledge of unmasked objects.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        }
+                    ],
+                    "index": 13.5,
+                    "bbox_fs": [
+                        105,
+                        427,
+                        506,
+                        494
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        497,
+                        505,
+                        553
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                497,
+                                506,
+                                510
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        498,
+                                        342,
+                                        510
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Concretely, during training, we multiply each MONet latent",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        342,
+                                        500,
+                                        356,
+                                        509
+                                    ],
+                                    "score": 0.86,
+                                    "content": "\\mu _ { t i }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        356,
+                                        498,
+                                        450,
+                                        510
+                                    ],
+                                    "score": 1.0,
+                                    "content": "by a masking indicator,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        451,
+                                        497,
+                                        503,
+                                        510
+                                    ],
+                                    "score": 0.93,
+                                    "content": "m _ { t i } \\in \\{ 0 , 1 \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        503,
+                                        498,
+                                        506,
+                                        510
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                509,
+                                506,
+                                522
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        509,
+                                        123,
+                                        522
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Let",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        123,
+                                        509,
+                                        137,
+                                        521
+                                    ],
+                                    "score": 0.9,
+                                    "content": "\\mu _ { t i } ^ { \\prime }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        137,
+                                        509,
+                                        254,
+                                        522
+                                    ],
+                                    "score": 1.0,
+                                    "content": "be the transformed value of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        255,
+                                        510,
+                                        283,
+                                        520
+                                    ],
+                                    "score": 0.9,
+                                    "content": "m _ { t i } \\mu _ { t i }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        284,
+                                        509,
+                                        506,
+                                        522
+                                    ],
+                                    "score": 1.0,
+                                    "content": "after passing through the transformer. We expect the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                519,
+                                506,
+                                533
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        519,
+                                        466,
+                                        533
+                                    ],
+                                    "score": 1.0,
+                                    "content": "transformer to understand the underlying dynamics of the video, so that the masked out slot",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        466,
+                                        521,
+                                        480,
+                                        532
+                                    ],
+                                    "score": 0.86,
+                                    "content": "\\mu _ { t i }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        480,
+                                        519,
+                                        506,
+                                        533
+                                    ],
+                                    "score": 1.0,
+                                    "content": "could",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                531,
+                                506,
+                                544
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        531,
+                                        180,
+                                        544
+                                    ],
+                                    "score": 1.0,
+                                    "content": "be predicted from",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        180,
+                                        531,
+                                        193,
+                                        543
+                                    ],
+                                    "score": 0.89,
+                                    "content": "\\mu _ { t i } ^ { \\prime }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        194,
+                                        531,
+                                        506,
+                                        544
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". To guide the transformer in learning effective representations capable of this",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                541,
+                                325,
+                                555
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        541,
+                                        325,
+                                        555
+                                    ],
+                                    "score": 1.0,
+                                    "content": "type of dynamics prediction, we add an auxiliary loss:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        }
+                    ],
+                    "index": 19,
+                    "bbox_fs": [
+                        105,
+                        497,
+                        506,
+                        555
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        231,
+                        560,
+                        379,
+                        588
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                231,
+                                560,
+                                379,
+                                588
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        231,
+                                        560,
+                                        379,
+                                        588
+                                    ],
+                                    "score": 0.94,
+                                    "content": "{ \\mathrm { a u x i l i a r y ~ l o s s } } = \\sum _ { t , i } { \\tau _ { t i } l \\left( f ( \\mu _ { t i } ^ { \\prime } ) , \\mu \\right) } ,",
+                                    "type": "interline_equation",
+                                    "image_path": "ec34b10eb5dbd048dcd818839d9d4c52bdbb686a32df86c0e869749c1ec07022.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 22.5,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                231,
+                                560,
+                                379,
+                                574.0
+                            ],
+                            "spans": [],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                231,
+                                574.0,
+                                379,
+                                588.0
+                            ],
+                            "spans": [],
+                            "index": 23
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        595,
+                        505,
+                        663
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                595,
+                                506,
+                                609
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        595,
+                                        133,
+                                        609
+                                    ],
+                                    "score": 1.0,
+                                    "content": "where",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        133,
+                                        597,
+                                        141,
+                                        608
+                                    ],
+                                    "score": 0.86,
+                                    "content": "f",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        141,
+                                        595,
+                                        264,
+                                        609
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is a learned linear mapping to",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        264,
+                                        595,
+                                        277,
+                                        606
+                                    ],
+                                    "score": 0.74,
+                                    "content": "\\mathbb { R } ^ { d }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        278,
+                                        595,
+                                        281,
+                                        609
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        281,
+                                        596,
+                                        286,
+                                        606
+                                    ],
+                                    "score": 0.4,
+                                    "content": "l",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        286,
+                                        595,
+                                        368,
+                                        609
+                                    ],
+                                    "score": 1.0,
+                                    "content": "a loss function, and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        369,
+                                        596,
+                                        417,
+                                        608
+                                    ],
+                                    "score": 0.94,
+                                    "content": "\\tau _ { t i } \\in \\{ 0 , 1 \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        417,
+                                        595,
+                                        506,
+                                        609
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are one-hot indicator",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                606,
+                                506,
+                                621
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        606,
+                                        506,
+                                        621
+                                    ],
+                                    "score": 1.0,
+                                    "content": "variables identifying the prediction targets (not necessarily just the masked out entries, since the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                618,
+                                506,
+                                631
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        618,
+                                        506,
+                                        631
+                                    ],
+                                    "score": 1.0,
+                                    "content": "prediction targets could be a subset of the masked out entries). We propagate gradients only to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                629,
+                                505,
+                                641
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        629,
+                                        180,
+                                        641
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the parameters of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        181,
+                                        629,
+                                        188,
+                                        641
+                                    ],
+                                    "score": 0.85,
+                                    "content": "f",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        188,
+                                        629,
+                                        405,
+                                        641
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and the transformer and not to the learned word and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        405,
+                                        629,
+                                        428,
+                                        639
+                                    ],
+                                    "score": 0.59,
+                                    "content": "C L S",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        428,
+                                        629,
+                                        505,
+                                        641
+                                    ],
+                                    "score": 1.0,
+                                    "content": "embeddings. This",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                640,
+                                506,
+                                652
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        640,
+                                        374,
+                                        652
+                                    ],
+                                    "score": 1.0,
+                                    "content": "auxiliary loss is added to the main classification loss with weighting",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        374,
+                                        640,
+                                        381,
+                                        650
+                                    ],
+                                    "score": 0.71,
+                                    "content": "\\lambda",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        381,
+                                        640,
+                                        506,
+                                        652
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", and both losses are minimized",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                650,
+                                472,
+                                663
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        650,
+                                        472,
+                                        663
+                                    ],
+                                    "score": 1.0,
+                                    "content": "simultaneously by the optimizer. We do not pretrain the model with only the auxiliary loss.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        }
+                    ],
+                    "index": 26.5,
+                    "bbox_fs": [
+                        105,
+                        595,
+                        506,
+                        663
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        667,
+                        505,
+                        722
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                666,
+                                505,
+                                679
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        666,
+                                        280,
+                                        679
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We tested two different loss functions for",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        280,
+                                        668,
+                                        284,
+                                        677
+                                    ],
+                                    "score": 0.5,
+                                    "content": "l",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        285,
+                                        666,
+                                        505,
+                                        679
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", an L2 loss and a contrastive loss (formulas given in",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                678,
+                                506,
+                                690
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        678,
+                                        353,
+                                        690
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Appendix A.2), and six different masking schemes (settings of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        354,
+                                        679,
+                                        370,
+                                        689
+                                    ],
+                                    "score": 0.88,
+                                    "content": "m _ { t i }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        370,
+                                        678,
+                                        387,
+                                        690
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        388,
+                                        679,
+                                        400,
+                                        689
+                                    ],
+                                    "score": 0.85,
+                                    "content": "\\tau _ { t i }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        400,
+                                        678,
+                                        506,
+                                        690
+                                    ],
+                                    "score": 1.0,
+                                    "content": "), as illustrated in Figure 2.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                688,
+                                506,
+                                702
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        688,
+                                        506,
+                                        702
+                                    ],
+                                    "score": 1.0,
+                                    "content": "This exploration was motivated by the observation that video inputs at adjacent timesteps are highly",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                699,
+                                506,
+                                713
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        699,
+                                        506,
+                                        713
+                                    ],
+                                    "score": 1.0,
+                                    "content": "correlated in a way that adjacent words are not. We thus hypothesized that BERT-style prediction of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                710,
+                                505,
+                                724
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        710,
+                                        505,
+                                        724
+                                    ],
+                                    "score": 1.0,
+                                    "content": "adjacent words might not be optimal. A different masking strategy, in which prediction targets are",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                72,
+                                506,
+                                86
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        72,
+                                        506,
+                                        86
+                                    ],
+                                    "score": 1.0,
+                                    "content": "separated from the context by more than a single timestep, may stimulate capacity in the network to",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                83,
+                                467,
+                                96
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        83,
+                                        467,
+                                        96
+                                    ],
+                                    "score": 1.0,
+                                    "content": "acquire knowledge that permits context-based unrolls and better long-horizon predictions.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 1
+                        }
+                    ],
+                    "index": 32,
+                    "bbox_fs": [
+                        105,
+                        666,
+                        506,
+                        724
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        73,
+                        504,
+                        95
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                72,
+                                506,
+                                86
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        72,
+                                        506,
+                                        86
+                                    ],
+                                    "score": 1.0,
+                                    "content": "separated from the context by more than a single timestep, may stimulate capacity in the network to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                83,
+                                467,
+                                96
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        83,
+                                        467,
+                                        96
+                                    ],
+                                    "score": 1.0,
+                                    "content": "acquire knowledge that permits context-based unrolls and better long-horizon predictions.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        }
+                    ],
+                    "index": 0.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        100,
+                        505,
+                        188
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                99,
+                                505,
+                                113
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        99,
+                                        270,
+                                        113
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The simplest approach would be to set",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        271,
+                                        101,
+                                        309,
+                                        111
+                                    ],
+                                    "score": 0.9,
+                                    "content": "m _ { t i } = 1",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        309,
+                                        99,
+                                        428,
+                                        113
+                                    ],
+                                    "score": 1.0,
+                                    "content": "uniformly at random across",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        428,
+                                        101,
+                                        434,
+                                        110
+                                    ],
+                                    "score": 0.73,
+                                    "content": "t",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        434,
+                                        99,
+                                        453,
+                                        113
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        453,
+                                        101,
+                                        458,
+                                        110
+                                    ],
+                                    "score": 0.73,
+                                    "content": "i",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        459,
+                                        99,
+                                        505,
+                                        113
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", fixing the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                111,
+                                505,
+                                123
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        111,
+                                        218,
+                                        123
+                                    ],
+                                    "score": 1.0,
+                                    "content": "expected proportion of the",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        218,
+                                        112,
+                                        234,
+                                        122
+                                    ],
+                                    "score": 0.86,
+                                    "content": "m _ { t i }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        235,
+                                        111,
+                                        306,
+                                        123
+                                    ],
+                                    "score": 1.0,
+                                    "content": "set to 1 (schema",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        306,
+                                        111,
+                                        313,
+                                        121
+                                    ],
+                                    "score": 0.75,
+                                    "content": "^ b",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        313,
+                                        111,
+                                        505,
+                                        123
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in Figure 2). The targets would simply be the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                122,
+                                505,
+                                134
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        122,
+                                        174,
+                                        134
+                                    ],
+                                    "score": 1.0,
+                                    "content": "unmasked slots,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        174,
+                                        122,
+                                        232,
+                                        133
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\tau _ { t i } = 1 - m _ { t i }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        233,
+                                        122,
+                                        505,
+                                        134
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". One potential problem with this approach is that multiple objects",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                133,
+                                505,
+                                145
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        133,
+                                        505,
+                                        145
+                                    ],
+                                    "score": 1.0,
+                                    "content": "could be masked out in a single frame. MONet can unpredictably switch object-to-slot assignments",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                144,
+                                505,
+                                156
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        144,
+                                        505,
+                                        156
+                                    ],
+                                    "score": 1.0,
+                                    "content": "multiple times in a single video. If multiple slots are masked out, the transformer cannot determine",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                155,
+                                505,
+                                167
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        155,
+                                        505,
+                                        167
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with certainty which missing object to assign to each slot. Thus, the auxiliary loss could penalize the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                165,
+                                505,
+                                179
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        165,
+                                        505,
+                                        179
+                                    ],
+                                    "score": 1.0,
+                                    "content": "model even if it predicted all the objects correctly. To avoid this problem, we also try constraining the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                176,
+                                380,
+                                189
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        176,
+                                        366,
+                                        189
+                                    ],
+                                    "score": 1.0,
+                                    "content": "mask such that exactly one slot is masked out per frame (schema",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        367,
+                                        179,
+                                        373,
+                                        186
+                                    ],
+                                    "score": 0.49,
+                                    "content": "a",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        374,
+                                        176,
+                                        380,
+                                        189
+                                    ],
+                                    "score": 1.0,
+                                    "content": ").",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        }
+                    ],
+                    "index": 5.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        192,
+                        505,
+                        280
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                193,
+                                505,
+                                205
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        193,
+                                        449,
+                                        205
+                                    ],
+                                    "score": 1.0,
+                                    "content": "To pose harder prediction challenges, we can add a buffer between the context (where",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        449,
+                                        194,
+                                        484,
+                                        204
+                                    ],
+                                    "score": 0.89,
+                                    "content": "m _ { t i } = 1",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        484,
+                                        193,
+                                        505,
+                                        205
+                                    ],
+                                    "score": 1.0,
+                                    "content": ") and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                203,
+                                505,
+                                216
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        203,
+                                        216,
+                                        216
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the infilling targets (where",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        217,
+                                        204,
+                                        248,
+                                        215
+                                    ],
+                                    "score": 0.9,
+                                    "content": "\\tau _ { t i } = 1",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        248,
+                                        203,
+                                        272,
+                                        216
+                                    ],
+                                    "score": 1.0,
+                                    "content": "). For",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        272,
+                                        204,
+                                        277,
+                                        213
+                                    ],
+                                    "score": 0.65,
+                                    "content": "t",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        278,
+                                        203,
+                                        378,
+                                        216
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in this buffer zone, both",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        379,
+                                        204,
+                                        414,
+                                        215
+                                    ],
+                                    "score": 0.91,
+                                    "content": "m _ { t i } = 0",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        414,
+                                        203,
+                                        432,
+                                        216
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        433,
+                                        204,
+                                        464,
+                                        215
+                                    ],
+                                    "score": 0.9,
+                                    "content": "\\tau _ { t i } = 0",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        464,
+                                        203,
+                                        505,
+                                        216
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(schemas",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                215,
+                                506,
+                                227
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        215,
+                                        124,
+                                        226
+                                    ],
+                                    "score": 0.85,
+                                    "content": "c { - } f )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        124,
+                                        215,
+                                        235,
+                                        227
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". We choose a single cutoff",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        236,
+                                        216,
+                                        244,
+                                        225
+                                    ],
+                                    "score": 0.8,
+                                    "content": "T",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        245,
+                                        215,
+                                        333,
+                                        227
+                                    ],
+                                    "score": 1.0,
+                                    "content": "randomly, and we set",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        334,
+                                        215,
+                                        369,
+                                        226
+                                    ],
+                                    "score": 0.91,
+                                    "content": "m _ { t i } = 0",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        369,
+                                        215,
+                                        384,
+                                        227
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        384,
+                                        215,
+                                        409,
+                                        225
+                                    ],
+                                    "score": 0.9,
+                                    "content": "t < T",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        410,
+                                        215,
+                                        427,
+                                        227
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        428,
+                                        216,
+                                        462,
+                                        226
+                                    ],
+                                    "score": 0.91,
+                                    "content": "m _ { t i } = 1",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        462,
+                                        215,
+                                        477,
+                                        227
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        478,
+                                        215,
+                                        503,
+                                        226
+                                    ],
+                                    "score": 0.91,
+                                    "content": "t \\geq T",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        503,
+                                        215,
+                                        506,
+                                        227
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                225,
+                                506,
+                                239
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        225,
+                                        506,
+                                        239
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In the presence of this buffer, we compared prediction (where the context is strictly before the targets;",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                237,
+                                505,
+                                249
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        237,
+                                        137,
+                                        249
+                                    ],
+                                    "score": 1.0,
+                                    "content": "schema",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        138,
+                                        237,
+                                        153,
+                                        247
+                                    ],
+                                    "score": 0.37,
+                                    "content": "c , d",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        154,
+                                        237,
+                                        409,
+                                        249
+                                    ],
+                                    "score": 1.0,
+                                    "content": ") versus infilling (where the context surrounds the targets; schema",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        409,
+                                        237,
+                                        425,
+                                        248
+                                    ],
+                                    "score": 0.84,
+                                    "content": "e , f ,",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        425,
+                                        237,
+                                        505,
+                                        249
+                                    ],
+                                    "score": 1.0,
+                                    "content": "). We also compared",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                247,
+                                506,
+                                261
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        247,
+                                        294,
+                                        261
+                                    ],
+                                    "score": 1.0,
+                                    "content": "setting the targets as individual objects (schema",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        294,
+                                        249,
+                                        310,
+                                        258
+                                    ],
+                                    "score": 0.35,
+                                    "content": "c , e",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        310,
+                                        247,
+                                        506,
+                                        261
+                                    ],
+                                    "score": 1.0,
+                                    "content": ") versus targets as all objects in the scene (schema",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                257,
+                                506,
+                                271
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        258,
+                                        124,
+                                        270
+                                    ],
+                                    "score": 0.82,
+                                    "content": "d , f )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        124,
+                                        257,
+                                        506,
+                                        271
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". We visually inspect the efficacy of this self-supervised loss in encouraging better representations",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                268,
+                                341,
+                                282
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        268,
+                                        341,
+                                        282
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(beyond improvements of scores on tasks) in Appendix D.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        }
+                    ],
+                    "index": 13.5
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        299,
+                        191,
+                        313
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                298,
+                                193,
+                                316
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        298,
+                                        193,
+                                        316
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3 Experiments",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        }
+                    ],
+                    "index": 18
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        327,
+                        505,
+                        371
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                327,
+                                505,
+                                339
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        327,
+                                        505,
+                                        339
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We tested Aloe on three datasets, CLEVRER [41], CATER [12], and ACRE [44]. For each dataset, we",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                338,
+                                506,
+                                351
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        338,
+                                        506,
+                                        351
+                                    ],
+                                    "score": 1.0,
+                                    "content": "pretrained a MONet model on individual frames. More training details and a table of hyperparameters",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                349,
+                                507,
+                                363
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        349,
+                                        507,
+                                        363
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are given in Appendix A.3; these hyperparameters were obtained through a hyperparameter sweep.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                360,
+                                416,
+                                372
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        360,
+                                        416,
+                                        372
+                                    ],
+                                    "score": 1.0,
+                                    "content": "All error bars are standard deviations computed over at least 5 random seeds.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        }
+                    ],
+                    "index": 20.5
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        388,
+                        179,
+                        400
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                387,
+                                181,
+                                403
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        387,
+                                        181,
+                                        403
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3.1 CLEVRER",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        }
+                    ],
+                    "index": 23
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        410,
+                        505,
+                        477
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                410,
+                                506,
+                                423
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        410,
+                                        506,
+                                        423
+                                    ],
+                                    "score": 1.0,
+                                    "content": "CLEVRER features videos of CLEVR objects [21] that move and collide with each other. For",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                422,
+                                505,
+                                434
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        422,
+                                        505,
+                                        434
+                                    ],
+                                    "score": 1.0,
+                                    "content": "each video, several questions are posed to test the model’s understanding of the scene. Unlike",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                432,
+                                506,
+                                445
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        432,
+                                        506,
+                                        445
+                                    ],
+                                    "score": 1.0,
+                                    "content": "most other visual question answering datasets, which test for only descriptive understanding (“what",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                443,
+                                505,
+                                456
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        443,
+                                        505,
+                                        456
+                                    ],
+                                    "score": 1.0,
+                                    "content": "happened?”), CLEVRER poses other more complex questions, including explanatory questions",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                453,
+                                505,
+                                467
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        453,
+                                        505,
+                                        467
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(“why did something happen?”), predictive questions (“what will happen next?”), and counterfactual",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                465,
+                                372,
+                                478
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        465,
+                                        372,
+                                        478
+                                    ],
+                                    "score": 1.0,
+                                    "content": "questions (“what would happen in a unseen circumstance?”) [41].",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        }
+                    ],
+                    "index": 26.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        481,
+                        505,
+                        569
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                480,
+                                506,
+                                494
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        480,
+                                        403,
+                                        494
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We compare Aloe to state-of-the-art models reported in the literature: MAC",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        403,
+                                        482,
+                                        423,
+                                        493
+                                    ],
+                                    "score": 0.78,
+                                    "content": "\\left( \\mathrm { V } + \\right)",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        423,
+                                        480,
+                                        506,
+                                        494
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and NS-DR [41], as",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                492,
+                                506,
+                                505
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        492,
+                                        349,
+                                        505
+                                    ],
+                                    "score": 1.0,
+                                    "content": "well as the DCL model [6] (simultaneous to our work). MAC",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        349,
+                                        493,
+                                        369,
+                                        504
+                                    ],
+                                    "score": 0.79,
+                                    "content": "\\left( \\mathrm { V } + \\right)",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        370,
+                                        492,
+                                        506,
+                                        505
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(based on the MAC network [20])",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                503,
+                                506,
+                                516
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        503,
+                                        506,
+                                        516
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is an end-to-end network augmented with object information and trained using ground truth labels for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                514,
+                                506,
+                                526
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        514,
+                                        506,
+                                        526
+                                    ],
+                                    "score": 1.0,
+                                    "content": "object segmentation masks and features (e.g. color, shape). NS-DR and DCL are hybrid models that",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                525,
+                                506,
+                                537
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        525,
+                                        506,
+                                        537
+                                    ],
+                                    "score": 1.0,
+                                    "content": "apply a symbolic logic engine to outputs of various neural networks. The neural networks are used to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                536,
+                                506,
+                                550
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        536,
+                                        506,
+                                        550
+                                    ],
+                                    "score": 1.0,
+                                    "content": "detect objects, predict dynamics, and parse the question into a program, and the symbolic executor",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                547,
+                                505,
+                                559
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        547,
+                                        505,
+                                        559
+                                    ],
+                                    "score": 1.0,
+                                    "content": "runs the parsed program to obtain the final output. NS-DR is trained using ground truth labels and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                558,
+                                470,
+                                570
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        558,
+                                        470,
+                                        570
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ground truth parsed programs, while DCL requires only the ground truth parsed programs.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        }
+                    ],
+                    "index": 33.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        574,
+                        505,
+                        662
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                574,
+                                505,
+                                587
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        574,
+                                        505,
+                                        587
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Table 1 shows the result of Aloe compared to these models. Across all categories, Aloe significantly",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                585,
+                                505,
+                                597
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        585,
+                                        505,
+                                        597
+                                    ],
+                                    "score": 1.0,
+                                    "content": "outperforms the previous best models. Moreover, compared to the other models, Aloe does not use",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                596,
+                                505,
+                                610
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        596,
+                                        505,
+                                        610
+                                    ],
+                                    "score": 1.0,
+                                    "content": "any labeled data other than the correct answer for the questions, nor does it require pretraining on",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                608,
+                                505,
+                                619
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        608,
+                                        505,
+                                        619
+                                    ],
+                                    "score": 1.0,
+                                    "content": "any other dataset. Aloe also was not specifically designed for this task, and it straightforwardly",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                618,
+                                506,
+                                631
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        618,
+                                        506,
+                                        631
+                                    ],
+                                    "score": 1.0,
+                                    "content": "generalizes to other tasks as well, such as CATER [12] and ACRE [44]. We provide a few sample",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                628,
+                                506,
+                                641
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        628,
+                                        506,
+                                        641
+                                    ],
+                                    "score": 1.0,
+                                    "content": "model classifications on a randomly selected set of videos and questions in Appendix E.1 and detailed",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                640,
+                                505,
+                                652
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        640,
+                                        505,
+                                        652
+                                    ],
+                                    "score": 1.0,
+                                    "content": "analysis of counterfactual questions in Appendix C. These examples suggest qualitatively that, for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                651,
+                                495,
+                                662
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        651,
+                                        495,
+                                        662
+                                    ],
+                                    "score": 1.0,
+                                    "content": "most instances where the model was incorrect, humans would plausibly furnish the same answer.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        }
+                    ],
+                    "index": 41.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        678,
+                        505,
+                        722
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                678,
+                                505,
+                                691
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        678,
+                                        505,
+                                        691
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Attention analysis (More analyses are given in Appendix D) We analyzed the cross-modal attention",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                689,
+                                505,
+                                700
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        689,
+                                        505,
+                                        700
+                                    ],
+                                    "score": 1.0,
+                                    "content": "between question-words and the MONet objects. For each word, we determined the object that",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 47
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                700,
+                                506,
+                                712
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        700,
+                                        506,
+                                        712
+                                    ],
+                                    "score": 1.0,
+                                    "content": "attended to that word with highest weight (for one head in the last layer). In the visualization below,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 48
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                711,
+                                505,
+                                723
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        711,
+                                        505,
+                                        723
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the bounding boxes show the objects found by MONet, and each word is colored according to the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 49
+                        }
+                    ],
+                    "index": 47.5
+                }
+            ],
+            "page_idx": 4,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        302,
+                        741,
+                        309,
+                        750
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                301,
+                                740,
+                                310,
+                                753
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        301,
+                                        740,
+                                        310,
+                                        753
+                                    ],
+                                    "score": 1.0,
+                                    "content": "5",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        73,
+                        504,
+                        95
+                    ],
+                    "lines": [],
+                    "index": 0.5,
+                    "bbox_fs": [
+                        105,
+                        72,
+                        506,
+                        96
+                    ],
+                    "lines_deleted": true
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        100,
+                        505,
+                        188
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                99,
+                                505,
+                                113
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        99,
+                                        270,
+                                        113
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The simplest approach would be to set",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        271,
+                                        101,
+                                        309,
+                                        111
+                                    ],
+                                    "score": 0.9,
+                                    "content": "m _ { t i } = 1",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        309,
+                                        99,
+                                        428,
+                                        113
+                                    ],
+                                    "score": 1.0,
+                                    "content": "uniformly at random across",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        428,
+                                        101,
+                                        434,
+                                        110
+                                    ],
+                                    "score": 0.73,
+                                    "content": "t",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        434,
+                                        99,
+                                        453,
+                                        113
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        453,
+                                        101,
+                                        458,
+                                        110
+                                    ],
+                                    "score": 0.73,
+                                    "content": "i",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        459,
+                                        99,
+                                        505,
+                                        113
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", fixing the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                111,
+                                505,
+                                123
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        111,
+                                        218,
+                                        123
+                                    ],
+                                    "score": 1.0,
+                                    "content": "expected proportion of the",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        218,
+                                        112,
+                                        234,
+                                        122
+                                    ],
+                                    "score": 0.86,
+                                    "content": "m _ { t i }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        235,
+                                        111,
+                                        306,
+                                        123
+                                    ],
+                                    "score": 1.0,
+                                    "content": "set to 1 (schema",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        306,
+                                        111,
+                                        313,
+                                        121
+                                    ],
+                                    "score": 0.75,
+                                    "content": "^ b",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        313,
+                                        111,
+                                        505,
+                                        123
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in Figure 2). The targets would simply be the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                122,
+                                505,
+                                134
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        122,
+                                        174,
+                                        134
+                                    ],
+                                    "score": 1.0,
+                                    "content": "unmasked slots,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        174,
+                                        122,
+                                        232,
+                                        133
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\tau _ { t i } = 1 - m _ { t i }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        233,
+                                        122,
+                                        505,
+                                        134
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". One potential problem with this approach is that multiple objects",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                133,
+                                505,
+                                145
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        133,
+                                        505,
+                                        145
+                                    ],
+                                    "score": 1.0,
+                                    "content": "could be masked out in a single frame. MONet can unpredictably switch object-to-slot assignments",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                144,
+                                505,
+                                156
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        144,
+                                        505,
+                                        156
+                                    ],
+                                    "score": 1.0,
+                                    "content": "multiple times in a single video. If multiple slots are masked out, the transformer cannot determine",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                155,
+                                505,
+                                167
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        155,
+                                        505,
+                                        167
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with certainty which missing object to assign to each slot. Thus, the auxiliary loss could penalize the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                165,
+                                505,
+                                179
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        165,
+                                        505,
+                                        179
+                                    ],
+                                    "score": 1.0,
+                                    "content": "model even if it predicted all the objects correctly. To avoid this problem, we also try constraining the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                176,
+                                380,
+                                189
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        176,
+                                        366,
+                                        189
+                                    ],
+                                    "score": 1.0,
+                                    "content": "mask such that exactly one slot is masked out per frame (schema",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        367,
+                                        179,
+                                        373,
+                                        186
+                                    ],
+                                    "score": 0.49,
+                                    "content": "a",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        374,
+                                        176,
+                                        380,
+                                        189
+                                    ],
+                                    "score": 1.0,
+                                    "content": ").",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        }
+                    ],
+                    "index": 5.5,
+                    "bbox_fs": [
+                        105,
+                        99,
+                        505,
+                        189
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        192,
+                        505,
+                        280
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                193,
+                                505,
+                                205
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        193,
+                                        449,
+                                        205
+                                    ],
+                                    "score": 1.0,
+                                    "content": "To pose harder prediction challenges, we can add a buffer between the context (where",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        449,
+                                        194,
+                                        484,
+                                        204
+                                    ],
+                                    "score": 0.89,
+                                    "content": "m _ { t i } = 1",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        484,
+                                        193,
+                                        505,
+                                        205
+                                    ],
+                                    "score": 1.0,
+                                    "content": ") and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                203,
+                                505,
+                                216
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        203,
+                                        216,
+                                        216
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the infilling targets (where",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        217,
+                                        204,
+                                        248,
+                                        215
+                                    ],
+                                    "score": 0.9,
+                                    "content": "\\tau _ { t i } = 1",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        248,
+                                        203,
+                                        272,
+                                        216
+                                    ],
+                                    "score": 1.0,
+                                    "content": "). For",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        272,
+                                        204,
+                                        277,
+                                        213
+                                    ],
+                                    "score": 0.65,
+                                    "content": "t",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        278,
+                                        203,
+                                        378,
+                                        216
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in this buffer zone, both",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        379,
+                                        204,
+                                        414,
+                                        215
+                                    ],
+                                    "score": 0.91,
+                                    "content": "m _ { t i } = 0",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        414,
+                                        203,
+                                        432,
+                                        216
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        433,
+                                        204,
+                                        464,
+                                        215
+                                    ],
+                                    "score": 0.9,
+                                    "content": "\\tau _ { t i } = 0",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        464,
+                                        203,
+                                        505,
+                                        216
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(schemas",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                215,
+                                506,
+                                227
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        215,
+                                        124,
+                                        226
+                                    ],
+                                    "score": 0.85,
+                                    "content": "c { - } f )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        124,
+                                        215,
+                                        235,
+                                        227
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". We choose a single cutoff",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        236,
+                                        216,
+                                        244,
+                                        225
+                                    ],
+                                    "score": 0.8,
+                                    "content": "T",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        245,
+                                        215,
+                                        333,
+                                        227
+                                    ],
+                                    "score": 1.0,
+                                    "content": "randomly, and we set",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        334,
+                                        215,
+                                        369,
+                                        226
+                                    ],
+                                    "score": 0.91,
+                                    "content": "m _ { t i } = 0",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        369,
+                                        215,
+                                        384,
+                                        227
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        384,
+                                        215,
+                                        409,
+                                        225
+                                    ],
+                                    "score": 0.9,
+                                    "content": "t < T",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        410,
+                                        215,
+                                        427,
+                                        227
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        428,
+                                        216,
+                                        462,
+                                        226
+                                    ],
+                                    "score": 0.91,
+                                    "content": "m _ { t i } = 1",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        462,
+                                        215,
+                                        477,
+                                        227
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        478,
+                                        215,
+                                        503,
+                                        226
+                                    ],
+                                    "score": 0.91,
+                                    "content": "t \\geq T",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        503,
+                                        215,
+                                        506,
+                                        227
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                225,
+                                506,
+                                239
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        225,
+                                        506,
+                                        239
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In the presence of this buffer, we compared prediction (where the context is strictly before the targets;",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                237,
+                                505,
+                                249
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        237,
+                                        137,
+                                        249
+                                    ],
+                                    "score": 1.0,
+                                    "content": "schema",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        138,
+                                        237,
+                                        153,
+                                        247
+                                    ],
+                                    "score": 0.37,
+                                    "content": "c , d",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        154,
+                                        237,
+                                        409,
+                                        249
+                                    ],
+                                    "score": 1.0,
+                                    "content": ") versus infilling (where the context surrounds the targets; schema",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        409,
+                                        237,
+                                        425,
+                                        248
+                                    ],
+                                    "score": 0.84,
+                                    "content": "e , f ,",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        425,
+                                        237,
+                                        505,
+                                        249
+                                    ],
+                                    "score": 1.0,
+                                    "content": "). We also compared",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                247,
+                                506,
+                                261
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        247,
+                                        294,
+                                        261
+                                    ],
+                                    "score": 1.0,
+                                    "content": "setting the targets as individual objects (schema",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        294,
+                                        249,
+                                        310,
+                                        258
+                                    ],
+                                    "score": 0.35,
+                                    "content": "c , e",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        310,
+                                        247,
+                                        506,
+                                        261
+                                    ],
+                                    "score": 1.0,
+                                    "content": ") versus targets as all objects in the scene (schema",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                257,
+                                506,
+                                271
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        258,
+                                        124,
+                                        270
+                                    ],
+                                    "score": 0.82,
+                                    "content": "d , f )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        124,
+                                        257,
+                                        506,
+                                        271
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". We visually inspect the efficacy of this self-supervised loss in encouraging better representations",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                268,
+                                341,
+                                282
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        268,
+                                        341,
+                                        282
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(beyond improvements of scores on tasks) in Appendix D.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        }
+                    ],
+                    "index": 13.5,
+                    "bbox_fs": [
+                        104,
+                        193,
+                        506,
+                        282
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        299,
+                        191,
+                        313
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                298,
+                                193,
+                                316
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        298,
+                                        193,
+                                        316
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3 Experiments",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        }
+                    ],
+                    "index": 18
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        327,
+                        505,
+                        371
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                327,
+                                505,
+                                339
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        327,
+                                        505,
+                                        339
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We tested Aloe on three datasets, CLEVRER [41], CATER [12], and ACRE [44]. For each dataset, we",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                338,
+                                506,
+                                351
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        338,
+                                        506,
+                                        351
+                                    ],
+                                    "score": 1.0,
+                                    "content": "pretrained a MONet model on individual frames. More training details and a table of hyperparameters",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                349,
+                                507,
+                                363
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        349,
+                                        507,
+                                        363
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are given in Appendix A.3; these hyperparameters were obtained through a hyperparameter sweep.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                360,
+                                416,
+                                372
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        360,
+                                        416,
+                                        372
+                                    ],
+                                    "score": 1.0,
+                                    "content": "All error bars are standard deviations computed over at least 5 random seeds.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        }
+                    ],
+                    "index": 20.5,
+                    "bbox_fs": [
+                        105,
+                        327,
+                        507,
+                        372
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        388,
+                        179,
+                        400
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                387,
+                                181,
+                                403
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        387,
+                                        181,
+                                        403
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3.1 CLEVRER",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        }
+                    ],
+                    "index": 23
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        410,
+                        505,
+                        477
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                410,
+                                506,
+                                423
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        410,
+                                        506,
+                                        423
+                                    ],
+                                    "score": 1.0,
+                                    "content": "CLEVRER features videos of CLEVR objects [21] that move and collide with each other. For",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                422,
+                                505,
+                                434
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        422,
+                                        505,
+                                        434
+                                    ],
+                                    "score": 1.0,
+                                    "content": "each video, several questions are posed to test the model’s understanding of the scene. Unlike",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                432,
+                                506,
+                                445
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        432,
+                                        506,
+                                        445
+                                    ],
+                                    "score": 1.0,
+                                    "content": "most other visual question answering datasets, which test for only descriptive understanding (“what",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                443,
+                                505,
+                                456
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        443,
+                                        505,
+                                        456
+                                    ],
+                                    "score": 1.0,
+                                    "content": "happened?”), CLEVRER poses other more complex questions, including explanatory questions",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                453,
+                                505,
+                                467
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        453,
+                                        505,
+                                        467
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(“why did something happen?”), predictive questions (“what will happen next?”), and counterfactual",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                465,
+                                372,
+                                478
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        465,
+                                        372,
+                                        478
+                                    ],
+                                    "score": 1.0,
+                                    "content": "questions (“what would happen in a unseen circumstance?”) [41].",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        }
+                    ],
+                    "index": 26.5,
+                    "bbox_fs": [
+                        105,
+                        410,
+                        506,
+                        478
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        481,
+                        505,
+                        569
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                480,
+                                506,
+                                494
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        480,
+                                        403,
+                                        494
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We compare Aloe to state-of-the-art models reported in the literature: MAC",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        403,
+                                        482,
+                                        423,
+                                        493
+                                    ],
+                                    "score": 0.78,
+                                    "content": "\\left( \\mathrm { V } + \\right)",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        423,
+                                        480,
+                                        506,
+                                        494
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and NS-DR [41], as",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                492,
+                                506,
+                                505
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        492,
+                                        349,
+                                        505
+                                    ],
+                                    "score": 1.0,
+                                    "content": "well as the DCL model [6] (simultaneous to our work). MAC",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        349,
+                                        493,
+                                        369,
+                                        504
+                                    ],
+                                    "score": 0.79,
+                                    "content": "\\left( \\mathrm { V } + \\right)",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        370,
+                                        492,
+                                        506,
+                                        505
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(based on the MAC network [20])",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                503,
+                                506,
+                                516
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        503,
+                                        506,
+                                        516
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is an end-to-end network augmented with object information and trained using ground truth labels for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                514,
+                                506,
+                                526
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        514,
+                                        506,
+                                        526
+                                    ],
+                                    "score": 1.0,
+                                    "content": "object segmentation masks and features (e.g. color, shape). NS-DR and DCL are hybrid models that",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                525,
+                                506,
+                                537
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        525,
+                                        506,
+                                        537
+                                    ],
+                                    "score": 1.0,
+                                    "content": "apply a symbolic logic engine to outputs of various neural networks. The neural networks are used to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                536,
+                                506,
+                                550
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        536,
+                                        506,
+                                        550
+                                    ],
+                                    "score": 1.0,
+                                    "content": "detect objects, predict dynamics, and parse the question into a program, and the symbolic executor",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                547,
+                                505,
+                                559
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        547,
+                                        505,
+                                        559
+                                    ],
+                                    "score": 1.0,
+                                    "content": "runs the parsed program to obtain the final output. NS-DR is trained using ground truth labels and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                558,
+                                470,
+                                570
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        558,
+                                        470,
+                                        570
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ground truth parsed programs, while DCL requires only the ground truth parsed programs.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        }
+                    ],
+                    "index": 33.5,
+                    "bbox_fs": [
+                        105,
+                        480,
+                        506,
+                        570
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        574,
+                        505,
+                        662
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                574,
+                                505,
+                                587
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        574,
+                                        505,
+                                        587
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Table 1 shows the result of Aloe compared to these models. Across all categories, Aloe significantly",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                585,
+                                505,
+                                597
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        585,
+                                        505,
+                                        597
+                                    ],
+                                    "score": 1.0,
+                                    "content": "outperforms the previous best models. Moreover, compared to the other models, Aloe does not use",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                596,
+                                505,
+                                610
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        596,
+                                        505,
+                                        610
+                                    ],
+                                    "score": 1.0,
+                                    "content": "any labeled data other than the correct answer for the questions, nor does it require pretraining on",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                608,
+                                505,
+                                619
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        608,
+                                        505,
+                                        619
+                                    ],
+                                    "score": 1.0,
+                                    "content": "any other dataset. Aloe also was not specifically designed for this task, and it straightforwardly",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                618,
+                                506,
+                                631
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        618,
+                                        506,
+                                        631
+                                    ],
+                                    "score": 1.0,
+                                    "content": "generalizes to other tasks as well, such as CATER [12] and ACRE [44]. We provide a few sample",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                628,
+                                506,
+                                641
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        628,
+                                        506,
+                                        641
+                                    ],
+                                    "score": 1.0,
+                                    "content": "model classifications on a randomly selected set of videos and questions in Appendix E.1 and detailed",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                640,
+                                505,
+                                652
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        640,
+                                        505,
+                                        652
+                                    ],
+                                    "score": 1.0,
+                                    "content": "analysis of counterfactual questions in Appendix C. These examples suggest qualitatively that, for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                651,
+                                495,
+                                662
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        651,
+                                        495,
+                                        662
+                                    ],
+                                    "score": 1.0,
+                                    "content": "most instances where the model was incorrect, humans would plausibly furnish the same answer.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        }
+                    ],
+                    "index": 41.5,
+                    "bbox_fs": [
+                        105,
+                        574,
+                        506,
+                        662
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        678,
+                        505,
+                        722
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                678,
+                                505,
+                                691
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        678,
+                                        505,
+                                        691
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Attention analysis (More analyses are given in Appendix D) We analyzed the cross-modal attention",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                689,
+                                505,
+                                700
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        689,
+                                        505,
+                                        700
+                                    ],
+                                    "score": 1.0,
+                                    "content": "between question-words and the MONet objects. For each word, we determined the object that",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 47
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                700,
+                                506,
+                                712
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        700,
+                                        506,
+                                        712
+                                    ],
+                                    "score": 1.0,
+                                    "content": "attended to that word with highest weight (for one head in the last layer). In the visualization below,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 48
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                711,
+                                505,
+                                723
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        711,
+                                        505,
+                                        723
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the bounding boxes show the objects found by MONet, and each word is colored according to the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 49
+                        }
+                    ],
+                    "index": 47.5,
+                    "bbox_fs": [
+                        105,
+                        678,
+                        506,
+                        723
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "table",
+                    "bbox": [
+                        109,
+                        69,
+                        501,
+                        172
+                    ],
+                    "blocks": [
+                        {
+                            "type": "table_body",
+                            "bbox": [
+                                109,
+                                69,
+                                501,
+                                172
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        109,
+                                        69,
+                                        501,
+                                        172
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                109,
+                                                69,
+                                                501,
+                                                172
+                                            ],
+                                            "score": 0.98,
+                                            "html": "<table><tr><td>Model</td><td>Descriptive</td><td>Explanatory</td><td>Predictive</td><td>Counterfactual</td></tr><tr><td>MAC (V+)</td><td>86.4</td><td>22.3</td><td>42.9</td><td>25.1</td></tr><tr><td>NS-DR</td><td>88.1</td><td>79.6</td><td>68.7</td><td>42.2</td></tr><tr><td>DCL</td><td>90.7</td><td>82.8</td><td>82.0</td><td>46.5</td></tr><tr><td>Aloe</td><td>94.0 ± 0.4</td><td>96.0 ± 0.6</td><td>87.5 ± 3.0</td><td>75.6 ± 3.8</td></tr><tr><td>Aloe- self-attention+MLP</td><td>45.4</td><td>16.0</td><td>27.7</td><td>9.9</td></tr><tr><td>Aloe- object-repr. +ResNet</td><td>74.9</td><td>66.1</td><td>58.3</td><td>32.4</td></tr><tr><td>Aloe - global + hierarchical attn.</td><td>80.6</td><td>87.4</td><td>73.5</td><td>55.1</td></tr><tr><td>Aloe- self-supervised loss</td><td>91.0</td><td>92.8</td><td>82.8</td><td>68.7</td></tr></table>",
+                                            "type": "table",
+                                            "image_path": "b17fb2d33ac7519789a92be95f5ba1468a0123fe73f580cd362649f95d93168e.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 1,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        109,
+                                        69,
+                                        501,
+                                        103.33333333333334
+                                    ],
+                                    "spans": [],
+                                    "index": 0
+                                },
+                                {
+                                    "bbox": [
+                                        109,
+                                        103.33333333333334,
+                                        501,
+                                        137.66666666666669
+                                    ],
+                                    "spans": [],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        109,
+                                        137.66666666666669,
+                                        501,
+                                        172.00000000000003
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                }
+                            ]
+                        },
+                        {
+                            "type": "table_caption",
+                            "bbox": [
+                                106,
+                                177,
+                                505,
+                                221
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        176,
+                                        505,
+                                        189
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                104,
+                                                176,
+                                                505,
+                                                189
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Table 1: Performance (per question accuracy) on CLEVRER of Aloe compared to results from",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 3
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        187,
+                                        506,
+                                        200
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                187,
+                                                506,
+                                                200
+                                            ],
+                                            "score": 1.0,
+                                            "content": "literature and to ablations: 1) MLP instead of self-attention; 2) ResNet superpixels instead of MONet",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 4
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        199,
+                                        505,
+                                        211
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                199,
+                                                505,
+                                                211
+                                            ],
+                                            "score": 1.0,
+                                            "content": "objects; 3) hierarchical frame-level and intra-frame attention instead of global cross-frame object",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 5
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        210,
+                                        228,
+                                        222
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                210,
+                                                228,
+                                                222
+                                            ],
+                                            "score": 1.0,
+                                            "content": "attention; 4) no auxiliary loss.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 6
+                                }
+                            ],
+                            "index": 4.5
+                        }
+                    ],
+                    "index": 2.75
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        243,
+                        504,
+                        266
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                243,
+                                505,
+                                256
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        243,
+                                        505,
+                                        256
+                                    ],
+                                    "score": 1.0,
+                                    "content": "object that attended to it with highest weight (black represents a MONet slot without any objects).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                254,
+                                433,
+                                267
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        254,
+                                        433,
+                                        267
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We observe that generally, objects attend heavily to the words that describe them.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        }
+                    ],
+                    "index": 7.5
+                },
+                {
+                    "type": "image",
+                    "bbox": [
+                        109,
+                        272,
+                        233,
+                        356
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                109,
+                                272,
+                                233,
+                                356
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        109,
+                                        272,
+                                        233,
+                                        356
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                109,
+                                                272,
+                                                233,
+                                                356
+                                            ],
+                                            "score": 0.964,
+                                            "type": "image",
+                                            "image_path": "229b16c4fda57fa6be8830ae071ce6a0ec54820c6c821a7f96d21f78abffa24d.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 9.5,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        109,
+                                        272,
+                                        233,
+                                        314.0
+                                    ],
+                                    "spans": [],
+                                    "index": 9
+                                },
+                                {
+                                    "bbox": [
+                                        109,
+                                        314.0,
+                                        233,
+                                        356.0
+                                    ],
+                                    "spans": [],
+                                    "index": 10
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 9.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        264,
+                        280,
+                        465,
+                        302
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                264,
+                                278,
+                                466,
+                                293
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        264,
+                                        278,
+                                        466,
+                                        293
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Q: If the cylinder is removed, which event will not",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                263,
+                                290,
+                                302,
+                                303
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        263,
+                                        290,
+                                        302,
+                                        303
+                                    ],
+                                    "score": 1.0,
+                                    "content": "happen?",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        }
+                    ],
+                    "index": 11.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        264,
+                        307,
+                        467,
+                        348
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                263,
+                                304,
+                                468,
+                                320
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        263,
+                                        304,
+                                        468,
+                                        320
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1. The brown object collides with the green object.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                263,
+                                321,
+                                461,
+                                334
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        263,
+                                        321,
+                                        461,
+                                        334
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2. The yellow object and the metal cube collide.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                263,
+                                336,
+                                467,
+                                349
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        263,
+                                        336,
+                                        467,
+                                        349
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3. The yellow cube collides with the green object.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        }
+                    ],
+                    "index": 14
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        362,
+                        506,
+                        439
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                361,
+                                506,
+                                374
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        361,
+                                        506,
+                                        374
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We also looked at the objects that were most heavily attended upon in determining the final answer.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                373,
+                                505,
+                                385
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        373,
+                                        337,
+                                        385
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The image below illustrates the attention weights for the",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        337,
+                                        373,
+                                        360,
+                                        383
+                                    ],
+                                    "score": 0.66,
+                                    "content": "C L S",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        360,
+                                        373,
+                                        505,
+                                        385
+                                    ],
+                                    "score": 1.0,
+                                    "content": "token attending on each object (for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                384,
+                                505,
+                                396
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        384,
+                                        505,
+                                        396
+                                    ],
+                                    "score": 1.0,
+                                    "content": "one head in the last layer), when the model is tasked with assessing the first choice of the question",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                396,
+                                505,
+                                406
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        396,
+                                        505,
+                                        406
+                                    ],
+                                    "score": 1.0,
+                                    "content": "above. The bounding boxes show the two most heavily attended upon objects for one transformer",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                405,
+                                506,
+                                418
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        405,
+                                        506,
+                                        418
+                                    ],
+                                    "score": 1.0,
+                                    "content": "head. We observe that this head focuses on the green and brown objects (asked about in choice 1),",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                416,
+                                506,
+                                429
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        416,
+                                        506,
+                                        429
+                                    ],
+                                    "score": 1.0,
+                                    "content": "but switches its focus to the cyan cylinder when it looks like the cylinder might collide with the cubes",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                426,
+                                208,
+                                441
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        426,
+                                        208,
+                                        441
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and change the outcome.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        }
+                    ],
+                    "index": 19
+                },
+                {
+                    "type": "image",
+                    "bbox": [
+                        108,
+                        444,
+                        503,
+                        490
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                108,
+                                444,
+                                503,
+                                490
+                            ],
+                            "group_id": 1,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        108,
+                                        444,
+                                        503,
+                                        490
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                108,
+                                                444,
+                                                503,
+                                                490
+                                            ],
+                                            "score": 0.949,
+                                            "type": "image",
+                                            "image_path": "f9813d948a0c16a84c18a5a51d92a5b3b2434d16be376473d12c2adad84e34ba.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 24,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        108,
+                                        444,
+                                        503,
+                                        459.3333333333333
+                                    ],
+                                    "spans": [],
+                                    "index": 23
+                                },
+                                {
+                                    "bbox": [
+                                        108,
+                                        459.3333333333333,
+                                        503,
+                                        474.66666666666663
+                                    ],
+                                    "spans": [],
+                                    "index": 24
+                                },
+                                {
+                                    "bbox": [
+                                        108,
+                                        474.66666666666663,
+                                        503,
+                                        489.99999999999994
+                                    ],
+                                    "spans": [],
+                                    "index": 25
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 24
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        505,
+                        505,
+                        549
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                505,
+                                505,
+                                517
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        505,
+                                        505,
+                                        517
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Model ablation Table 1 shows the contributions of various components of Aloe. First, self-attention",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                516,
+                                505,
+                                528
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        516,
+                                        505,
+                                        528
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is necessary for solving this problem. For comparison, we replace Aloe’s transformer with four fully",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                527,
+                                505,
+                                540
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        527,
+                                        505,
+                                        540
+                                    ],
+                                    "score": 1.0,
+                                    "content": "connected layers with 2048 units per layer1. We find that an MLP is unable to answer non-descriptive",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                538,
+                                426,
+                                550
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        538,
+                                        426,
+                                        550
+                                    ],
+                                    "score": 1.0,
+                                    "content": "questions effectively, despite using more parameters (20M vs 15M parameters).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        }
+                    ],
+                    "index": 27.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        554,
+                        505,
+                        653
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                555,
+                                505,
+                                567
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        555,
+                                        505,
+                                        567
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Second, we verify that an object-based discretization scheme is essential to the performance of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                564,
+                                504,
+                                579
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        564,
+                                        490,
+                                        579
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Aloe. We compare with a version of the architecture where the MONet object representations",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        490,
+                                        567,
+                                        504,
+                                        577
+                                    ],
+                                    "score": 0.82,
+                                    "content": "\\mu _ { t i }",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                576,
+                                506,
+                                589
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        576,
+                                        506,
+                                        589
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are replaced with ResNet hyperpixels as in Zambaldi et al. [43]. Concretely, we flatten the output of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                587,
+                                505,
+                                599
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        587,
+                                        505,
+                                        599
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the final convolutional layer of the ResNet to obtain a sequence of feature vectors that is fed into the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                598,
+                                506,
+                                610
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        598,
+                                        506,
+                                        610
+                                    ],
+                                    "score": 1.0,
+                                    "content": "transformer as the discrete entities. To match MONet’s pretraining regimen, we pretrain the ResNet",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                609,
+                                506,
+                                622
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        609,
+                                        506,
+                                        622
+                                    ],
+                                    "score": 1.0,
+                                    "content": "on CLEVR [21] by training an Aloe model (using a ResNet instead of MONet) on the CLEVR task",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                619,
+                                506,
+                                632
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        619,
+                                        506,
+                                        632
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and initializing the ResNet used in the CLEVRER task with these pre-trained weights. We find that",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                631,
+                                505,
+                                644
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        631,
+                                        505,
+                                        644
+                                    ],
+                                    "score": 1.0,
+                                    "content": "an object level representation, such as one output by MONet, greatly outperforms the locality-aware",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                641,
+                                278,
+                                654
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        641,
+                                        278,
+                                        654
+                                    ],
+                                    "score": 1.0,
+                                    "content": "but object-agnostic ResNet representation.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        }
+                    ],
+                    "index": 34
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        658,
+                        503,
+                        691
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                658,
+                                505,
+                                670
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        658,
+                                        505,
+                                        670
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We also observe the importance of global attention between all objects across all frames, compared to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                670,
+                                505,
+                                681
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        670,
+                                        505,
+                                        681
+                                    ],
+                                    "score": 1.0,
+                                    "content": "a hierarchical attention model where objects within a frame could attend to each other but frames",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                680,
+                                505,
+                                692
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        680,
+                                        505,
+                                        692
+                                    ],
+                                    "score": 1.0,
+                                    "content": "could only attend to each other as an atomic entity. We hypothesize that global attention may be",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        }
+                    ],
+                    "index": 40
+                }
+            ],
+            "page_idx": 5,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        106,
+                        701,
+                        505,
+                        722
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                119,
+                                700,
+                                505,
+                                713
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        119,
+                                        700,
+                                        505,
+                                        713
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1We also tried a bidirectional LSTM, which achieved even lower performance. This may be because the",
+                                    "type": "text"
+                                }
+                            ]
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                711,
+                                368,
+                                723
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        711,
+                                        368,
+                                        723
+                                    ],
+                                    "score": 1.0,
+                                    "content": "structure of our inputs requires the learning of long-range dependencies.",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        302,
+                        742,
+                        309,
+                        750
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                302,
+                                741,
+                                310,
+                                752
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        302,
+                                        741,
+                                        310,
+                                        752
+                                    ],
+                                    "score": 1.0,
+                                    "content": "6",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "table",
+                    "bbox": [
+                        109,
+                        69,
+                        501,
+                        172
+                    ],
+                    "blocks": [
+                        {
+                            "type": "table_body",
+                            "bbox": [
+                                109,
+                                69,
+                                501,
+                                172
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        109,
+                                        69,
+                                        501,
+                                        172
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                109,
+                                                69,
+                                                501,
+                                                172
+                                            ],
+                                            "score": 0.98,
+                                            "html": "<table><tr><td>Model</td><td>Descriptive</td><td>Explanatory</td><td>Predictive</td><td>Counterfactual</td></tr><tr><td>MAC (V+)</td><td>86.4</td><td>22.3</td><td>42.9</td><td>25.1</td></tr><tr><td>NS-DR</td><td>88.1</td><td>79.6</td><td>68.7</td><td>42.2</td></tr><tr><td>DCL</td><td>90.7</td><td>82.8</td><td>82.0</td><td>46.5</td></tr><tr><td>Aloe</td><td>94.0 ± 0.4</td><td>96.0 ± 0.6</td><td>87.5 ± 3.0</td><td>75.6 ± 3.8</td></tr><tr><td>Aloe- self-attention+MLP</td><td>45.4</td><td>16.0</td><td>27.7</td><td>9.9</td></tr><tr><td>Aloe- object-repr. +ResNet</td><td>74.9</td><td>66.1</td><td>58.3</td><td>32.4</td></tr><tr><td>Aloe - global + hierarchical attn.</td><td>80.6</td><td>87.4</td><td>73.5</td><td>55.1</td></tr><tr><td>Aloe- self-supervised loss</td><td>91.0</td><td>92.8</td><td>82.8</td><td>68.7</td></tr></table>",
+                                            "type": "table",
+                                            "image_path": "b17fb2d33ac7519789a92be95f5ba1468a0123fe73f580cd362649f95d93168e.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 1,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        109,
+                                        69,
+                                        501,
+                                        103.33333333333334
+                                    ],
+                                    "spans": [],
+                                    "index": 0
+                                },
+                                {
+                                    "bbox": [
+                                        109,
+                                        103.33333333333334,
+                                        501,
+                                        137.66666666666669
+                                    ],
+                                    "spans": [],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        109,
+                                        137.66666666666669,
+                                        501,
+                                        172.00000000000003
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                }
+                            ]
+                        },
+                        {
+                            "type": "table_caption",
+                            "bbox": [
+                                106,
+                                177,
+                                505,
+                                221
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        176,
+                                        505,
+                                        189
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                104,
+                                                176,
+                                                505,
+                                                189
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Table 1: Performance (per question accuracy) on CLEVRER of Aloe compared to results from",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 3
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        187,
+                                        506,
+                                        200
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                187,
+                                                506,
+                                                200
+                                            ],
+                                            "score": 1.0,
+                                            "content": "literature and to ablations: 1) MLP instead of self-attention; 2) ResNet superpixels instead of MONet",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 4
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        199,
+                                        505,
+                                        211
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                199,
+                                                505,
+                                                211
+                                            ],
+                                            "score": 1.0,
+                                            "content": "objects; 3) hierarchical frame-level and intra-frame attention instead of global cross-frame object",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 5
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        210,
+                                        228,
+                                        222
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                210,
+                                                228,
+                                                222
+                                            ],
+                                            "score": 1.0,
+                                            "content": "attention; 4) no auxiliary loss.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 6
+                                }
+                            ],
+                            "index": 4.5
+                        }
+                    ],
+                    "index": 2.75
+                },
+                {
+                    "type": "list",
+                    "bbox": [
+                        106,
+                        243,
+                        504,
+                        266
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                243,
+                                505,
+                                256
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        243,
+                                        505,
+                                        256
+                                    ],
+                                    "score": 1.0,
+                                    "content": "object that attended to it with highest weight (black represents a MONet slot without any objects).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                254,
+                                433,
+                                267
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        254,
+                                        433,
+                                        267
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We observe that generally, objects attend heavily to the words that describe them.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8,
+                            "is_list_start_line": true,
+                            "is_list_end_line": true
+                        }
+                    ],
+                    "index": 7.5,
+                    "bbox_fs": [
+                        106,
+                        243,
+                        505,
+                        267
+                    ]
+                },
+                {
+                    "type": "image",
+                    "bbox": [
+                        109,
+                        272,
+                        233,
+                        356
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                109,
+                                272,
+                                233,
+                                356
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        109,
+                                        272,
+                                        233,
+                                        356
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                109,
+                                                272,
+                                                233,
+                                                356
+                                            ],
+                                            "score": 0.964,
+                                            "type": "image",
+                                            "image_path": "229b16c4fda57fa6be8830ae071ce6a0ec54820c6c821a7f96d21f78abffa24d.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 9.5,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        109,
+                                        272,
+                                        233,
+                                        314.0
+                                    ],
+                                    "spans": [],
+                                    "index": 9
+                                },
+                                {
+                                    "bbox": [
+                                        109,
+                                        314.0,
+                                        233,
+                                        356.0
+                                    ],
+                                    "spans": [],
+                                    "index": 10
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 9.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        264,
+                        280,
+                        465,
+                        302
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                264,
+                                278,
+                                466,
+                                293
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        264,
+                                        278,
+                                        466,
+                                        293
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Q: If the cylinder is removed, which event will not",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                263,
+                                290,
+                                302,
+                                303
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        263,
+                                        290,
+                                        302,
+                                        303
+                                    ],
+                                    "score": 1.0,
+                                    "content": "happen?",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        }
+                    ],
+                    "index": 11.5,
+                    "bbox_fs": [
+                        263,
+                        278,
+                        466,
+                        303
+                    ]
+                },
+                {
+                    "type": "index",
+                    "bbox": [
+                        264,
+                        307,
+                        467,
+                        348
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                263,
+                                304,
+                                468,
+                                320
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        263,
+                                        304,
+                                        468,
+                                        320
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1. The brown object collides with the green object.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                263,
+                                321,
+                                461,
+                                334
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        263,
+                                        321,
+                                        461,
+                                        334
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2. The yellow object and the metal cube collide.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                263,
+                                336,
+                                467,
+                                349
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        263,
+                                        336,
+                                        467,
+                                        349
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3. The yellow cube collides with the green object.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15,
+                            "is_list_start_line": true
+                        }
+                    ],
+                    "index": 14,
+                    "bbox_fs": [
+                        263,
+                        304,
+                        468,
+                        349
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        362,
+                        506,
+                        439
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                361,
+                                506,
+                                374
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        361,
+                                        506,
+                                        374
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We also looked at the objects that were most heavily attended upon in determining the final answer.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                373,
+                                505,
+                                385
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        373,
+                                        337,
+                                        385
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The image below illustrates the attention weights for the",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        337,
+                                        373,
+                                        360,
+                                        383
+                                    ],
+                                    "score": 0.66,
+                                    "content": "C L S",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        360,
+                                        373,
+                                        505,
+                                        385
+                                    ],
+                                    "score": 1.0,
+                                    "content": "token attending on each object (for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                384,
+                                505,
+                                396
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        384,
+                                        505,
+                                        396
+                                    ],
+                                    "score": 1.0,
+                                    "content": "one head in the last layer), when the model is tasked with assessing the first choice of the question",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                396,
+                                505,
+                                406
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        396,
+                                        505,
+                                        406
+                                    ],
+                                    "score": 1.0,
+                                    "content": "above. The bounding boxes show the two most heavily attended upon objects for one transformer",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                405,
+                                506,
+                                418
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        405,
+                                        506,
+                                        418
+                                    ],
+                                    "score": 1.0,
+                                    "content": "head. We observe that this head focuses on the green and brown objects (asked about in choice 1),",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                416,
+                                506,
+                                429
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        416,
+                                        506,
+                                        429
+                                    ],
+                                    "score": 1.0,
+                                    "content": "but switches its focus to the cyan cylinder when it looks like the cylinder might collide with the cubes",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                426,
+                                208,
+                                441
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        426,
+                                        208,
+                                        441
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and change the outcome.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        }
+                    ],
+                    "index": 19,
+                    "bbox_fs": [
+                        105,
+                        361,
+                        506,
+                        441
+                    ]
+                },
+                {
+                    "type": "image",
+                    "bbox": [
+                        108,
+                        444,
+                        503,
+                        490
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                108,
+                                444,
+                                503,
+                                490
+                            ],
+                            "group_id": 1,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        108,
+                                        444,
+                                        503,
+                                        490
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                108,
+                                                444,
+                                                503,
+                                                490
+                                            ],
+                                            "score": 0.949,
+                                            "type": "image",
+                                            "image_path": "f9813d948a0c16a84c18a5a51d92a5b3b2434d16be376473d12c2adad84e34ba.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 24,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        108,
+                                        444,
+                                        503,
+                                        459.3333333333333
+                                    ],
+                                    "spans": [],
+                                    "index": 23
+                                },
+                                {
+                                    "bbox": [
+                                        108,
+                                        459.3333333333333,
+                                        503,
+                                        474.66666666666663
+                                    ],
+                                    "spans": [],
+                                    "index": 24
+                                },
+                                {
+                                    "bbox": [
+                                        108,
+                                        474.66666666666663,
+                                        503,
+                                        489.99999999999994
+                                    ],
+                                    "spans": [],
+                                    "index": 25
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 24
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        505,
+                        505,
+                        549
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                505,
+                                505,
+                                517
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        505,
+                                        505,
+                                        517
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Model ablation Table 1 shows the contributions of various components of Aloe. First, self-attention",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                516,
+                                505,
+                                528
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        516,
+                                        505,
+                                        528
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is necessary for solving this problem. For comparison, we replace Aloe’s transformer with four fully",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                527,
+                                505,
+                                540
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        527,
+                                        505,
+                                        540
+                                    ],
+                                    "score": 1.0,
+                                    "content": "connected layers with 2048 units per layer1. We find that an MLP is unable to answer non-descriptive",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                538,
+                                426,
+                                550
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        538,
+                                        426,
+                                        550
+                                    ],
+                                    "score": 1.0,
+                                    "content": "questions effectively, despite using more parameters (20M vs 15M parameters).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        }
+                    ],
+                    "index": 27.5,
+                    "bbox_fs": [
+                        105,
+                        505,
+                        505,
+                        550
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        554,
+                        505,
+                        653
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                555,
+                                505,
+                                567
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        555,
+                                        505,
+                                        567
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Second, we verify that an object-based discretization scheme is essential to the performance of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                564,
+                                504,
+                                579
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        564,
+                                        490,
+                                        579
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Aloe. We compare with a version of the architecture where the MONet object representations",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        490,
+                                        567,
+                                        504,
+                                        577
+                                    ],
+                                    "score": 0.82,
+                                    "content": "\\mu _ { t i }",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                576,
+                                506,
+                                589
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        576,
+                                        506,
+                                        589
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are replaced with ResNet hyperpixels as in Zambaldi et al. [43]. Concretely, we flatten the output of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                587,
+                                505,
+                                599
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        587,
+                                        505,
+                                        599
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the final convolutional layer of the ResNet to obtain a sequence of feature vectors that is fed into the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                598,
+                                506,
+                                610
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        598,
+                                        506,
+                                        610
+                                    ],
+                                    "score": 1.0,
+                                    "content": "transformer as the discrete entities. To match MONet’s pretraining regimen, we pretrain the ResNet",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                609,
+                                506,
+                                622
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        609,
+                                        506,
+                                        622
+                                    ],
+                                    "score": 1.0,
+                                    "content": "on CLEVR [21] by training an Aloe model (using a ResNet instead of MONet) on the CLEVR task",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                619,
+                                506,
+                                632
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        619,
+                                        506,
+                                        632
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and initializing the ResNet used in the CLEVRER task with these pre-trained weights. We find that",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                631,
+                                505,
+                                644
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        631,
+                                        505,
+                                        644
+                                    ],
+                                    "score": 1.0,
+                                    "content": "an object level representation, such as one output by MONet, greatly outperforms the locality-aware",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                641,
+                                278,
+                                654
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        641,
+                                        278,
+                                        654
+                                    ],
+                                    "score": 1.0,
+                                    "content": "but object-agnostic ResNet representation.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        }
+                    ],
+                    "index": 34,
+                    "bbox_fs": [
+                        104,
+                        555,
+                        506,
+                        654
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        658,
+                        503,
+                        691
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                658,
+                                505,
+                                670
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        658,
+                                        505,
+                                        670
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We also observe the importance of global attention between all objects across all frames, compared to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                670,
+                                505,
+                                681
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        670,
+                                        505,
+                                        681
+                                    ],
+                                    "score": 1.0,
+                                    "content": "a hierarchical attention model where objects within a frame could attend to each other but frames",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                680,
+                                505,
+                                692
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        680,
+                                        505,
+                                        692
+                                    ],
+                                    "score": 1.0,
+                                    "content": "could only attend to each other as an atomic entity. We hypothesize that global attention may be",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                366,
+                                505,
+                                377
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        366,
+                                        505,
+                                        377
+                                    ],
+                                    "score": 1.0,
+                                    "content": "important because with hierarchical attention, objects in different frames can only attend to each",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                376,
+                                505,
+                                388
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        376,
+                                        505,
+                                        388
+                                    ],
+                                    "score": 1.0,
+                                    "content": "other at the “frame” granularity. A cube attending to a cube in a different frame would then gather",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                386,
+                                445,
+                                399
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        386,
+                                        445,
+                                        399
+                                    ],
+                                    "score": 1.0,
+                                    "content": "information about the other non-cube objects, muddling the resulting representation.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 11
+                        }
+                    ],
+                    "index": 40,
+                    "bbox_fs": [
+                        106,
+                        658,
+                        505,
+                        692
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "image",
+                    "bbox": [
+                        106,
+                        69,
+                        505,
+                        260
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                106,
+                                69,
+                                505,
+                                260
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        69,
+                                        505,
+                                        260
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                69,
+                                                505,
+                                                260
+                                            ],
+                                            "score": 0.974,
+                                            "type": "image",
+                                            "image_path": "3444a8a096a123941b20407702c07de6ac1b82d124cdf4ea73c3ff1606e480c9.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 1,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        69,
+                                        505,
+                                        132.66666666666666
+                                    ],
+                                    "spans": [],
+                                    "index": 0
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        132.66666666666666,
+                                        505,
+                                        196.33333333333331
+                                    ],
+                                    "spans": [],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        196.33333333333331,
+                                        505,
+                                        260.0
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                106,
+                                265,
+                                505,
+                                331
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        266,
+                                        506,
+                                        278
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                266,
+                                                506,
+                                                278
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 3: Accuracy with/without auxiliary loss for different proportions of CLEVRER (row 1) and",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 3
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        277,
+                                        506,
+                                        290
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                277,
+                                                506,
+                                                290
+                                            ],
+                                            "score": 1.0,
+                                            "content": "CATER (row 2) training data. We also show comparisons with previous and concurrent work. For",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 4
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        288,
+                                        506,
+                                        300
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                288,
+                                                506,
+                                                300
+                                            ],
+                                            "score": 1.0,
+                                            "content": "CLEVRER, the lighter yellow bar represents the best neurosymbolic model DCL, and the darker",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 5
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        298,
+                                        505,
+                                        311
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                298,
+                                                358,
+                                                311
+                                            ],
+                                            "score": 1.0,
+                                            "content": "yellow bar represents the previous best distributed model, MAC",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                358,
+                                                299,
+                                                378,
+                                                309
+                                            ],
+                                            "score": 0.81,
+                                            "content": "\\left( \\mathrm { V } + \\right)",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                378,
+                                                298,
+                                                505,
+                                                311
+                                            ],
+                                            "score": 1.0,
+                                            "content": ". For CATER, the lighter yellow",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 6
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        309,
+                                        506,
+                                        322
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                309,
+                                                344,
+                                                322
+                                            ],
+                                            "score": 1.0,
+                                            "content": "bar represents Hopper and the darker yellow bar represents",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                345,
+                                                310,
+                                                384,
+                                                320
+                                            ],
+                                            "score": 0.78,
+                                            "content": "\\mathrm { R 3 D + N L }",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                384,
+                                                309,
+                                                506,
+                                                322
+                                            ],
+                                            "score": 1.0,
+                                            "content": ", the best published results for",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 7
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        321,
+                                        218,
+                                        334
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                321,
+                                                218,
+                                                334
+                                            ],
+                                            "score": 1.0,
+                                            "content": "the moving camera dataset.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 8
+                                }
+                            ],
+                            "index": 5.5
+                        }
+                    ],
+                    "index": 3.25
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        365,
+                        505,
+                        398
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                366,
+                                505,
+                                377
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        366,
+                                        505,
+                                        377
+                                    ],
+                                    "score": 1.0,
+                                    "content": "important because with hierarchical attention, objects in different frames can only attend to each",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                376,
+                                505,
+                                388
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        376,
+                                        505,
+                                        388
+                                    ],
+                                    "score": 1.0,
+                                    "content": "other at the “frame” granularity. A cube attending to a cube in a different frame would then gather",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                386,
+                                445,
+                                399
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        386,
+                                        445,
+                                        399
+                                    ],
+                                    "score": 1.0,
+                                    "content": "information about the other non-cube objects, muddling the resulting representation.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        }
+                    ],
+                    "index": 10
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        403,
+                        504,
+                        426
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                402,
+                                505,
+                                416
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        402,
+                                        505,
+                                        416
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Finally, we see that an auxiliary self-supervised loss improves the performance of the model by",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                415,
+                                501,
+                                426
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        415,
+                                        501,
+                                        426
+                                    ],
+                                    "score": 1.0,
+                                    "content": "between 4 and 6 percentage points, with the greatest improvement on the counterfactual questions.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        }
+                    ],
+                    "index": 12.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        450,
+                        505,
+                        549
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                451,
+                                506,
+                                464
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        451,
+                                        506,
+                                        464
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Self-supervision strategies We compared the various masking schemes and loss functions for our",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                462,
+                                505,
+                                475
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        462,
+                                        505,
+                                        475
+                                    ],
+                                    "score": 1.0,
+                                    "content": "auxiliary loss; a detailed figure is provided in Appendix A (Figure 4). We find that for all question",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                472,
+                                506,
+                                487
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        472,
+                                        506,
+                                        487
+                                    ],
+                                    "score": 1.0,
+                                    "content": "types in CLEVRER, an L2 loss performs better than a contrastive loss, and among the masking",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                483,
+                                506,
+                                496
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        483,
+                                        506,
+                                        496
+                                    ],
+                                    "score": 1.0,
+                                    "content": "schemes, masking one object per frame is the most effective. This particular result runs counter",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                495,
+                                505,
+                                507
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        495,
+                                        505,
+                                        507
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to our hypothesis that predictions or infilling in which the target is temporally removed from the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                506,
+                                505,
+                                517
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        506,
+                                        505,
+                                        517
+                                    ],
+                                    "score": 1.0,
+                                    "content": "context could encourage the model to learn more about scene dynamics and object interactions",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                516,
+                                506,
+                                529
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        516,
+                                        506,
+                                        529
+                                    ],
+                                    "score": 1.0,
+                                    "content": "than (BERT-style) local predictions of adjacent targets. Of course, there may be other settings or",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                528,
+                                505,
+                                539
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        528,
+                                        505,
+                                        539
+                                    ],
+                                    "score": 1.0,
+                                    "content": "loss functions that reveal the benefits of non-local prediction or constrastive losses; we leave this",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                538,
+                                221,
+                                550
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        538,
+                                        221,
+                                        550
+                                    ],
+                                    "score": 1.0,
+                                    "content": "investigation to future work.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        }
+                    ],
+                    "index": 18
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        574,
+                        505,
+                        640
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                574,
+                                506,
+                                586
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        574,
+                                        506,
+                                        586
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Data efficiency We investigated how model performance varies as a function of the number of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                585,
+                                506,
+                                597
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        585,
+                                        414,
+                                        597
+                                    ],
+                                    "score": 1.0,
+                                    "content": "labelled (question-answer) pairs it learns from. To do so, we train models on",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        414,
+                                        585,
+                                        433,
+                                        596
+                                    ],
+                                    "score": 0.9,
+                                    "content": "N \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        433,
+                                        585,
+                                        506,
+                                        597
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of the videos and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                596,
+                                505,
+                                609
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        596,
+                                        505,
+                                        609
+                                    ],
+                                    "score": 1.0,
+                                    "content": "their associated labeled data. We evaluate the effect of including the auxiliary self-supervised loss",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                607,
+                                506,
+                                620
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        607,
+                                        506,
+                                        620
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(applied to the entire dataset, not just the labelled portion) in this low data regime. This scenario,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                618,
+                                505,
+                                630
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        618,
+                                        505,
+                                        630
+                                    ],
+                                    "score": 1.0,
+                                    "content": "where unlabeled data is plentiful while labeled data is scarce, occurs frequently in practice, since",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                630,
+                                421,
+                                641
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        630,
+                                        421,
+                                        641
+                                    ],
+                                    "score": 1.0,
+                                    "content": "collecting labeled data is much more expensive than collecting unlabeled data.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        }
+                    ],
+                    "index": 25.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        645,
+                        505,
+                        722
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                645,
+                                505,
+                                658
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        645,
+                                        505,
+                                        658
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Figure 3 shows that our best model reaches the approximate level of the previous state-of-the-art",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                656,
+                                506,
+                                669
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        656,
+                                        199,
+                                        669
+                                    ],
+                                    "score": 1.0,
+                                    "content": "approaches using only",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        200,
+                                        656,
+                                        241,
+                                        667
+                                    ],
+                                    "score": 0.78,
+                                    "content": "50 \\% - 6 0 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        241,
+                                        656,
+                                        506,
+                                        669
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of the data. The self-supervised auxiliary loss makes a particular",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                667,
+                                506,
+                                680
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        667,
+                                        457,
+                                        680
+                                    ],
+                                    "score": 1.0,
+                                    "content": "improvement to performance in low-data regimes. For instance, when trained on only",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        458,
+                                        667,
+                                        478,
+                                        678
+                                    ],
+                                    "score": 0.88,
+                                    "content": "50 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        478,
+                                        667,
+                                        506,
+                                        680
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                678,
+                                506,
+                                690
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        678,
+                                        472,
+                                        690
+                                    ],
+                                    "score": 1.0,
+                                    "content": "available labelled data, self-supervised learning enables the model to reach a performance of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        472,
+                                        678,
+                                        492,
+                                        689
+                                    ],
+                                    "score": 0.87,
+                                    "content": "37 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        492,
+                                        678,
+                                        506,
+                                        690
+                                    ],
+                                    "score": 1.0,
+                                    "content": "on",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                688,
+                                506,
+                                702
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        688,
+                                        262,
+                                        702
+                                    ],
+                                    "score": 1.0,
+                                    "content": "counterfactual questions (compared to",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        262,
+                                        689,
+                                        282,
+                                        699
+                                    ],
+                                    "score": 0.87,
+                                    "content": "2 5 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        283,
+                                        688,
+                                        321,
+                                        702
+                                    ],
+                                    "score": 1.0,
+                                    "content": "by MAC",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        322,
+                                        689,
+                                        342,
+                                        700
+                                    ],
+                                    "score": 0.8,
+                                    "content": "( \\mathrm { V } + )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        342,
+                                        688,
+                                        360,
+                                        702
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        360,
+                                        689,
+                                        380,
+                                        699
+                                    ],
+                                    "score": 0.87,
+                                    "content": "42 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        380,
+                                        688,
+                                        506,
+                                        702
+                                    ],
+                                    "score": 1.0,
+                                    "content": "by NS-DR on the full dataset),",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                700,
+                                505,
+                                713
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        700,
+                                        395,
+                                        713
+                                    ],
+                                    "score": 1.0,
+                                    "content": "while without self-supervision, the model only reaches a performance of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        396,
+                                        700,
+                                        415,
+                                        711
+                                    ],
+                                    "score": 0.87,
+                                    "content": "13 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        415,
+                                        700,
+                                        485,
+                                        713
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(compared to the",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        486,
+                                        700,
+                                        505,
+                                        711
+                                    ],
+                                    "score": 0.86,
+                                    "content": "10 \\%",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                710,
+                                265,
+                                723
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        710,
+                                        265,
+                                        723
+                                    ],
+                                    "score": 1.0,
+                                    "content": "achieved by answering randomly [41]).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        }
+                    ],
+                    "index": 32
+                }
+            ],
+            "page_idx": 6,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        303,
+                        741,
+                        309,
+                        750
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                302,
+                                741,
+                                309,
+                                753
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        302,
+                                        741,
+                                        309,
+                                        753
+                                    ],
+                                    "score": 1.0,
+                                    "content": "7",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "image",
+                    "bbox": [
+                        106,
+                        69,
+                        505,
+                        260
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                106,
+                                69,
+                                505,
+                                260
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        69,
+                                        505,
+                                        260
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                69,
+                                                505,
+                                                260
+                                            ],
+                                            "score": 0.974,
+                                            "type": "image",
+                                            "image_path": "3444a8a096a123941b20407702c07de6ac1b82d124cdf4ea73c3ff1606e480c9.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 1,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        69,
+                                        505,
+                                        132.66666666666666
+                                    ],
+                                    "spans": [],
+                                    "index": 0
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        132.66666666666666,
+                                        505,
+                                        196.33333333333331
+                                    ],
+                                    "spans": [],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        196.33333333333331,
+                                        505,
+                                        260.0
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                106,
+                                265,
+                                505,
+                                331
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        266,
+                                        506,
+                                        278
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                266,
+                                                506,
+                                                278
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 3: Accuracy with/without auxiliary loss for different proportions of CLEVRER (row 1) and",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 3
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        277,
+                                        506,
+                                        290
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                277,
+                                                506,
+                                                290
+                                            ],
+                                            "score": 1.0,
+                                            "content": "CATER (row 2) training data. We also show comparisons with previous and concurrent work. For",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 4
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        288,
+                                        506,
+                                        300
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                288,
+                                                506,
+                                                300
+                                            ],
+                                            "score": 1.0,
+                                            "content": "CLEVRER, the lighter yellow bar represents the best neurosymbolic model DCL, and the darker",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 5
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        298,
+                                        505,
+                                        311
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                298,
+                                                358,
+                                                311
+                                            ],
+                                            "score": 1.0,
+                                            "content": "yellow bar represents the previous best distributed model, MAC",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                358,
+                                                299,
+                                                378,
+                                                309
+                                            ],
+                                            "score": 0.81,
+                                            "content": "\\left( \\mathrm { V } + \\right)",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                378,
+                                                298,
+                                                505,
+                                                311
+                                            ],
+                                            "score": 1.0,
+                                            "content": ". For CATER, the lighter yellow",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 6
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        309,
+                                        506,
+                                        322
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                309,
+                                                344,
+                                                322
+                                            ],
+                                            "score": 1.0,
+                                            "content": "bar represents Hopper and the darker yellow bar represents",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                345,
+                                                310,
+                                                384,
+                                                320
+                                            ],
+                                            "score": 0.78,
+                                            "content": "\\mathrm { R 3 D + N L }",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                384,
+                                                309,
+                                                506,
+                                                322
+                                            ],
+                                            "score": 1.0,
+                                            "content": ", the best published results for",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 7
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        321,
+                                        218,
+                                        334
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                321,
+                                                218,
+                                                334
+                                            ],
+                                            "score": 1.0,
+                                            "content": "the moving camera dataset.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 8
+                                }
+                            ],
+                            "index": 5.5
+                        }
+                    ],
+                    "index": 3.25
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        365,
+                        505,
+                        398
+                    ],
+                    "lines": [],
+                    "index": 10,
+                    "bbox_fs": [
+                        105,
+                        366,
+                        505,
+                        399
+                    ],
+                    "lines_deleted": true
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        403,
+                        504,
+                        426
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                402,
+                                505,
+                                416
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        402,
+                                        505,
+                                        416
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Finally, we see that an auxiliary self-supervised loss improves the performance of the model by",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                415,
+                                501,
+                                426
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        415,
+                                        501,
+                                        426
+                                    ],
+                                    "score": 1.0,
+                                    "content": "between 4 and 6 percentage points, with the greatest improvement on the counterfactual questions.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        }
+                    ],
+                    "index": 12.5,
+                    "bbox_fs": [
+                        105,
+                        402,
+                        505,
+                        426
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        450,
+                        505,
+                        549
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                451,
+                                506,
+                                464
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        451,
+                                        506,
+                                        464
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Self-supervision strategies We compared the various masking schemes and loss functions for our",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                462,
+                                505,
+                                475
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        462,
+                                        505,
+                                        475
+                                    ],
+                                    "score": 1.0,
+                                    "content": "auxiliary loss; a detailed figure is provided in Appendix A (Figure 4). We find that for all question",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                472,
+                                506,
+                                487
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        472,
+                                        506,
+                                        487
+                                    ],
+                                    "score": 1.0,
+                                    "content": "types in CLEVRER, an L2 loss performs better than a contrastive loss, and among the masking",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                483,
+                                506,
+                                496
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        483,
+                                        506,
+                                        496
+                                    ],
+                                    "score": 1.0,
+                                    "content": "schemes, masking one object per frame is the most effective. This particular result runs counter",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                495,
+                                505,
+                                507
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        495,
+                                        505,
+                                        507
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to our hypothesis that predictions or infilling in which the target is temporally removed from the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                506,
+                                505,
+                                517
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        506,
+                                        505,
+                                        517
+                                    ],
+                                    "score": 1.0,
+                                    "content": "context could encourage the model to learn more about scene dynamics and object interactions",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                516,
+                                506,
+                                529
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        516,
+                                        506,
+                                        529
+                                    ],
+                                    "score": 1.0,
+                                    "content": "than (BERT-style) local predictions of adjacent targets. Of course, there may be other settings or",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                528,
+                                505,
+                                539
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        528,
+                                        505,
+                                        539
+                                    ],
+                                    "score": 1.0,
+                                    "content": "loss functions that reveal the benefits of non-local prediction or constrastive losses; we leave this",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                538,
+                                221,
+                                550
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        538,
+                                        221,
+                                        550
+                                    ],
+                                    "score": 1.0,
+                                    "content": "investigation to future work.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        }
+                    ],
+                    "index": 18,
+                    "bbox_fs": [
+                        105,
+                        451,
+                        506,
+                        550
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        574,
+                        505,
+                        640
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                574,
+                                506,
+                                586
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        574,
+                                        506,
+                                        586
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Data efficiency We investigated how model performance varies as a function of the number of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                585,
+                                506,
+                                597
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        585,
+                                        414,
+                                        597
+                                    ],
+                                    "score": 1.0,
+                                    "content": "labelled (question-answer) pairs it learns from. To do so, we train models on",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        414,
+                                        585,
+                                        433,
+                                        596
+                                    ],
+                                    "score": 0.9,
+                                    "content": "N \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        433,
+                                        585,
+                                        506,
+                                        597
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of the videos and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                596,
+                                505,
+                                609
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        596,
+                                        505,
+                                        609
+                                    ],
+                                    "score": 1.0,
+                                    "content": "their associated labeled data. We evaluate the effect of including the auxiliary self-supervised loss",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                607,
+                                506,
+                                620
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        607,
+                                        506,
+                                        620
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(applied to the entire dataset, not just the labelled portion) in this low data regime. This scenario,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                618,
+                                505,
+                                630
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        618,
+                                        505,
+                                        630
+                                    ],
+                                    "score": 1.0,
+                                    "content": "where unlabeled data is plentiful while labeled data is scarce, occurs frequently in practice, since",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                630,
+                                421,
+                                641
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        630,
+                                        421,
+                                        641
+                                    ],
+                                    "score": 1.0,
+                                    "content": "collecting labeled data is much more expensive than collecting unlabeled data.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        }
+                    ],
+                    "index": 25.5,
+                    "bbox_fs": [
+                        105,
+                        574,
+                        506,
+                        641
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        645,
+                        505,
+                        722
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                645,
+                                505,
+                                658
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        645,
+                                        505,
+                                        658
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Figure 3 shows that our best model reaches the approximate level of the previous state-of-the-art",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                656,
+                                506,
+                                669
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        656,
+                                        199,
+                                        669
+                                    ],
+                                    "score": 1.0,
+                                    "content": "approaches using only",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        200,
+                                        656,
+                                        241,
+                                        667
+                                    ],
+                                    "score": 0.78,
+                                    "content": "50 \\% - 6 0 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        241,
+                                        656,
+                                        506,
+                                        669
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of the data. The self-supervised auxiliary loss makes a particular",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                667,
+                                506,
+                                680
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        667,
+                                        457,
+                                        680
+                                    ],
+                                    "score": 1.0,
+                                    "content": "improvement to performance in low-data regimes. For instance, when trained on only",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        458,
+                                        667,
+                                        478,
+                                        678
+                                    ],
+                                    "score": 0.88,
+                                    "content": "50 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        478,
+                                        667,
+                                        506,
+                                        680
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                678,
+                                506,
+                                690
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        678,
+                                        472,
+                                        690
+                                    ],
+                                    "score": 1.0,
+                                    "content": "available labelled data, self-supervised learning enables the model to reach a performance of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        472,
+                                        678,
+                                        492,
+                                        689
+                                    ],
+                                    "score": 0.87,
+                                    "content": "37 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        492,
+                                        678,
+                                        506,
+                                        690
+                                    ],
+                                    "score": 1.0,
+                                    "content": "on",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                688,
+                                506,
+                                702
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        688,
+                                        262,
+                                        702
+                                    ],
+                                    "score": 1.0,
+                                    "content": "counterfactual questions (compared to",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        262,
+                                        689,
+                                        282,
+                                        699
+                                    ],
+                                    "score": 0.87,
+                                    "content": "2 5 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        283,
+                                        688,
+                                        321,
+                                        702
+                                    ],
+                                    "score": 1.0,
+                                    "content": "by MAC",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        322,
+                                        689,
+                                        342,
+                                        700
+                                    ],
+                                    "score": 0.8,
+                                    "content": "( \\mathrm { V } + )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        342,
+                                        688,
+                                        360,
+                                        702
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        360,
+                                        689,
+                                        380,
+                                        699
+                                    ],
+                                    "score": 0.87,
+                                    "content": "42 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        380,
+                                        688,
+                                        506,
+                                        702
+                                    ],
+                                    "score": 1.0,
+                                    "content": "by NS-DR on the full dataset),",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                700,
+                                505,
+                                713
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        700,
+                                        395,
+                                        713
+                                    ],
+                                    "score": 1.0,
+                                    "content": "while without self-supervision, the model only reaches a performance of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        396,
+                                        700,
+                                        415,
+                                        711
+                                    ],
+                                    "score": 0.87,
+                                    "content": "13 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        415,
+                                        700,
+                                        485,
+                                        713
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(compared to the",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        486,
+                                        700,
+                                        505,
+                                        711
+                                    ],
+                                    "score": 0.86,
+                                    "content": "10 \\%",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                710,
+                                265,
+                                723
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        710,
+                                        265,
+                                        723
+                                    ],
+                                    "score": 1.0,
+                                    "content": "achieved by answering randomly [41]).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        }
+                    ],
+                    "index": 32,
+                    "bbox_fs": [
+                        105,
+                        645,
+                        506,
+                        723
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "table",
+                    "bbox": [
+                        90,
+                        69,
+                        520,
+                        161
+                    ],
+                    "blocks": [
+                        {
+                            "type": "table_body",
+                            "bbox": [
+                                90,
+                                69,
+                                520,
+                                161
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        90,
+                                        69,
+                                        520,
+                                        161
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                90,
+                                                69,
+                                                520,
+                                                161
+                                            ],
+                                            "score": 0.978,
+                                            "html": "<table><tr><td>Model</td><td>Top 1 (S)</td><td>Top 5 (S)</td><td>L1(S)</td><td>Top 1 (M)</td><td>Top 5 (M)</td><td>L1 (M)</td></tr><tr><td>R3DLSTM R3D+NL LSTM OPNet</td><td>60.2 46.2 74.8 73.2</td><td>81.8 69.9 1 93.8</td><td>1.2 1.5 0.54 0.85</td><td>28.6 38.6 1 1</td><td>63.3 70.2 1 1</td><td>1.7 1.5 -</td></tr><tr><td>Hopper Aloe (no auxiliary) Aloe Aloe (with L1 loss)</td><td>60.5 70.6 74.0 ± 0.3</td><td>84.5 93.0 94.0 ± 0.4</td><td>0.90 0.53 0.44 ± 0.01</td><td>46.8 56.6 59.7 ± 0.5</td><td>75.1 87.0 90.1 ± 0.6</td><td>- 1.3 0.82 0.69 ± 0.01</td></tr></table>",
+                                            "type": "table",
+                                            "image_path": "e9161209b964b799a3fdbb5b6fd16ab97b2278252e29ae3b7dbfb3b5a8d70f15.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 1,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        90,
+                                        69,
+                                        520,
+                                        99.66666666666667
+                                    ],
+                                    "spans": [],
+                                    "index": 0
+                                },
+                                {
+                                    "bbox": [
+                                        90,
+                                        99.66666666666667,
+                                        520,
+                                        130.33333333333334
+                                    ],
+                                    "spans": [],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        90,
+                                        130.33333333333334,
+                                        520,
+                                        161.0
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                }
+                            ]
+                        },
+                        {
+                            "type": "table_caption",
+                            "bbox": [
+                                106,
+                                165,
+                                506,
+                                199
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        163,
+                                        505,
+                                        179
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                163,
+                                                505,
+                                                179
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Table 2: Performance on CATER of Aloe compared to the best results from literature. We report top",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 3
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        176,
+                                        505,
+                                        189
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                176,
+                                                505,
+                                                189
+                                            ],
+                                            "score": 1.0,
+                                            "content": "1 accuracy, top 5 accuracy, and L1 distance between the predicted grid cell and true grid cell. The",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 4
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        187,
+                                        323,
+                                        200
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                187,
+                                                323,
+                                                200
+                                            ],
+                                            "score": 1.0,
+                                            "content": "labels (S) and (M) refer to static and moving cameras.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 5
+                                }
+                            ],
+                            "index": 4
+                        }
+                    ],
+                    "index": 2.5
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        106,
+                        225,
+                        164,
+                        237
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                224,
+                                165,
+                                239
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        224,
+                                        165,
+                                        239
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3.2 CATER",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        }
+                    ],
+                    "index": 6
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        248,
+                        505,
+                        314
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                249,
+                                506,
+                                260
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        249,
+                                        506,
+                                        260
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In a second experiment, we tested Aloe on CATER, a widely-used object-tracking dataset [12, 14, 31,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                258,
+                                507,
+                                273
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        258,
+                                        507,
+                                        273
+                                    ],
+                                    "score": 1.0,
+                                    "content": "45]. In CATER, objects from the CLEVR dataset [21] move and potentially occlude other objects,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                270,
+                                506,
+                                282
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        270,
+                                        506,
+                                        282
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and the goal is to predict the location of a target object (called the snitch) in the final frame. Because",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                281,
+                                505,
+                                294
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        281,
+                                        505,
+                                        294
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the snitch could be occluded by multiple objects that could move in the meantime, a successful model",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                291,
+                                506,
+                                304
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        291,
+                                        506,
+                                        304
+                                    ],
+                                    "score": 1.0,
+                                    "content": "must be sensitive to notions of object permanence. CATER also includes a moving camera variant,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                303,
+                                330,
+                                315
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        303,
+                                        330,
+                                        315
+                                    ],
+                                    "score": 1.0,
+                                    "content": "which introduces additional complexities for the model.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        }
+                    ],
+                    "index": 9.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        319,
+                        505,
+                        374
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                319,
+                                506,
+                                332
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        319,
+                                        506,
+                                        332
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Concretely, CATER is setup as a classification challenge. Objects are located in an xyz coordinate",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                330,
+                                506,
+                                342
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        330,
+                                        506,
+                                        342
+                                    ],
+                                    "score": 1.0,
+                                    "content": "system, where x and y range from -3 to 3. The xy plane is divided into a 6 by 6 grid, and the task is",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                341,
+                                506,
+                                354
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        341,
+                                        506,
+                                        354
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to predict the grid index of the snitch in the final frame. For Aloe, we use a classification loss (cross",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                352,
+                                505,
+                                365
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        352,
+                                        505,
+                                        365
+                                    ],
+                                    "score": 1.0,
+                                    "content": "entropy over the 36 possible grid indices) and an L1 loss (L1 distance between predicted grid cell and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                363,
+                                180,
+                                376
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        363,
+                                        180,
+                                        376
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the true grid cell).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        }
+                    ],
+                    "index": 15
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        379,
+                        505,
+                        456
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                378,
+                                505,
+                                393
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        378,
+                                        505,
+                                        393
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Table 2 shows Aloe compared to state-of-the-art models in the literature on both static and moving",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                390,
+                                505,
+                                403
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        390,
+                                        505,
+                                        403
+                                    ],
+                                    "score": 1.0,
+                                    "content": "camera videos. R3D and R3D NL are the strongest two models evaluated by Girdhar and Ramanan",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                400,
+                                506,
+                                414
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        400,
+                                        506,
+                                        414
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[12]. OPNet, or the Object Permanence Network [31], is an architecture with inductive biases",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                411,
+                                505,
+                                426
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        411,
+                                        505,
+                                        426
+                                    ],
+                                    "score": 1.0,
+                                    "content": "designed for object tracking tasks; it was trained with extra supervised labels, namely the bounding",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                423,
+                                505,
+                                436
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        423,
+                                        505,
+                                        436
+                                    ],
+                                    "score": 1.0,
+                                    "content": "boxes for all objects (including occluded ones). Hopper is a multi-hop transformer model developed",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                434,
+                                505,
+                                447
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        434,
+                                        505,
+                                        447
+                                    ],
+                                    "score": 1.0,
+                                    "content": "simultaneously with this work [45]. One key component of Hopper is Hungarian matching between",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                444,
+                                387,
+                                459
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        444,
+                                        387,
+                                        459
+                                    ],
+                                    "score": 1.0,
+                                    "content": "objects of different frames, a strong inductive bias for object tracking.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        }
+                    ],
+                    "index": 21
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        461,
+                        505,
+                        560
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                461,
+                                505,
+                                474
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        461,
+                                        505,
+                                        474
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We train Aloe simultaneously on both static and moving camera videos. Aloe outperforms the R3D",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                473,
+                                505,
+                                484
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        473,
+                                        505,
+                                        484
+                                    ],
+                                    "score": 1.0,
+                                    "content": "models for both static and moving cameras. We also ran Aloe with an additional auxiliary loss",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                483,
+                                507,
+                                496
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        483,
+                                        507,
+                                        496
+                                    ],
+                                    "score": 1.0,
+                                    "content": "consisting of the L1 distance between the predicted cell and the actual cell. With this additional loss,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                494,
+                                506,
+                                507
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        494,
+                                        506,
+                                        507
+                                    ],
+                                    "score": 1.0,
+                                    "content": "we get comparable results in the moving camera case as the R3D models for the static camera case.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                503,
+                                506,
+                                519
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        503,
+                                        506,
+                                        519
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Moreover, we achieve comparable accuracy as OPNet for accuracy and L1 distance, despite requiring",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                515,
+                                506,
+                                529
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        515,
+                                        506,
+                                        529
+                                    ],
+                                    "score": 1.0,
+                                    "content": "less supervision to train. Appendix E.2 gives a few sample outputs from Aloe; in particular we note",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                525,
+                                505,
+                                541
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        525,
+                                        505,
+                                        541
+                                    ],
+                                    "score": 1.0,
+                                    "content": "that it is able to find the target object in several cases where the object was occluded, demonstrating",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                538,
+                                504,
+                                550
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        538,
+                                        504,
+                                        550
+                                    ],
+                                    "score": 1.0,
+                                    "content": "that Aloe is able to do some level of object tracking. Finally,we find that an auxiliary self-supervised",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                548,
+                                505,
+                                561
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        548,
+                                        505,
+                                        561
+                                    ],
+                                    "score": 1.0,
+                                    "content": "loss helps the model perform well in the low data regime for CATER as well, as shown in Figure 3.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        }
+                    ],
+                    "index": 29
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        579,
+                        158,
+                        590
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                577,
+                                160,
+                                592
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        577,
+                                        160,
+                                        592
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3.3 ACRE",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        }
+                    ],
+                    "index": 34
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        601,
+                        505,
+                        722
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                602,
+                                505,
+                                613
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        602,
+                                        505,
+                                        613
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Finally, we measured Aloe’s performance on ACRE, a causal induction dataset inspired by the Blicket",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                613,
+                                505,
+                                624
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        613,
+                                        505,
+                                        624
+                                    ],
+                                    "score": 1.0,
+                                    "content": "task from developmental psychology [13, 44]. ACRE is divided into a set of problems. In each",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                623,
+                                507,
+                                637
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        623,
+                                        507,
+                                        637
+                                    ],
+                                    "score": 1.0,
+                                    "content": "problem, certain objects are chosen to be “Blickets”, and this assignment changes across problems.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                635,
+                                506,
+                                647
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        635,
+                                        506,
+                                        647
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Each problem presents a context of six images to the model, where different objects are placed on a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                644,
+                                506,
+                                658
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        644,
+                                        506,
+                                        658
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Blicket machine that lights up if one of those objects is a Blicket. The model is asked whether an",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                656,
+                                506,
+                                668
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        656,
+                                        506,
+                                        668
+                                    ],
+                                    "score": 1.0,
+                                    "content": "unseen combination of objects will light up the Blicket machine. Besides “yes” and “no”, a third",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                667,
+                                506,
+                                679
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        667,
+                                        506,
+                                        679
+                                    ],
+                                    "score": 1.0,
+                                    "content": "possible answer is “undetermined”, which is the case if it is impossible to determine for certain",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                678,
+                                506,
+                                690
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        678,
+                                        506,
+                                        690
+                                    ],
+                                    "score": 1.0,
+                                    "content": "if the objects will light up the machine. Correct inference goes beyond mere correlation: even if",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                689,
+                                506,
+                                700
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        689,
+                                        506,
+                                        700
+                                    ],
+                                    "score": 1.0,
+                                    "content": "every context scene involving object A has a lit-up machine, A’s Blicketness is still uncertain if",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                700,
+                                505,
+                                713
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        700,
+                                        505,
+                                        713
+                                    ],
+                                    "score": 1.0,
+                                    "content": "each of those scenes can potentially be explained by another object (deduction of A’s Blicketness is",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                711,
+                                187,
+                                722
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        711,
+                                        187,
+                                        722
+                                    ],
+                                    "score": 1.0,
+                                    "content": "backward-blocked).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        }
+                    ],
+                    "index": 40
+                }
+            ],
+            "page_idx": 7,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        302,
+                        742,
+                        308,
+                        750
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                301,
+                                740,
+                                310,
+                                752
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        301,
+                                        740,
+                                        310,
+                                        752
+                                    ],
+                                    "score": 1.0,
+                                    "content": "",
+                                    "type": "text",
+                                    "height": 12,
+                                    "width": 9
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "table",
+                    "bbox": [
+                        90,
+                        69,
+                        520,
+                        161
+                    ],
+                    "blocks": [
+                        {
+                            "type": "table_body",
+                            "bbox": [
+                                90,
+                                69,
+                                520,
+                                161
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        90,
+                                        69,
+                                        520,
+                                        161
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                90,
+                                                69,
+                                                520,
+                                                161
+                                            ],
+                                            "score": 0.978,
+                                            "html": "<table><tr><td>Model</td><td>Top 1 (S)</td><td>Top 5 (S)</td><td>L1(S)</td><td>Top 1 (M)</td><td>Top 5 (M)</td><td>L1 (M)</td></tr><tr><td>R3DLSTM R3D+NL LSTM OPNet</td><td>60.2 46.2 74.8 73.2</td><td>81.8 69.9 1 93.8</td><td>1.2 1.5 0.54 0.85</td><td>28.6 38.6 1 1</td><td>63.3 70.2 1 1</td><td>1.7 1.5 -</td></tr><tr><td>Hopper Aloe (no auxiliary) Aloe Aloe (with L1 loss)</td><td>60.5 70.6 74.0 ± 0.3</td><td>84.5 93.0 94.0 ± 0.4</td><td>0.90 0.53 0.44 ± 0.01</td><td>46.8 56.6 59.7 ± 0.5</td><td>75.1 87.0 90.1 ± 0.6</td><td>- 1.3 0.82 0.69 ± 0.01</td></tr></table>",
+                                            "type": "table",
+                                            "image_path": "e9161209b964b799a3fdbb5b6fd16ab97b2278252e29ae3b7dbfb3b5a8d70f15.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 1,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        90,
+                                        69,
+                                        520,
+                                        99.66666666666667
+                                    ],
+                                    "spans": [],
+                                    "index": 0
+                                },
+                                {
+                                    "bbox": [
+                                        90,
+                                        99.66666666666667,
+                                        520,
+                                        130.33333333333334
+                                    ],
+                                    "spans": [],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        90,
+                                        130.33333333333334,
+                                        520,
+                                        161.0
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                }
+                            ]
+                        },
+                        {
+                            "type": "table_caption",
+                            "bbox": [
+                                106,
+                                165,
+                                506,
+                                199
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        163,
+                                        505,
+                                        179
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                163,
+                                                505,
+                                                179
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Table 2: Performance on CATER of Aloe compared to the best results from literature. We report top",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 3
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        176,
+                                        505,
+                                        189
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                176,
+                                                505,
+                                                189
+                                            ],
+                                            "score": 1.0,
+                                            "content": "1 accuracy, top 5 accuracy, and L1 distance between the predicted grid cell and true grid cell. The",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 4
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        187,
+                                        323,
+                                        200
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                187,
+                                                323,
+                                                200
+                                            ],
+                                            "score": 1.0,
+                                            "content": "labels (S) and (M) refer to static and moving cameras.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 5
+                                }
+                            ],
+                            "index": 4
+                        }
+                    ],
+                    "index": 2.5
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        106,
+                        225,
+                        164,
+                        237
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                224,
+                                165,
+                                239
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        224,
+                                        165,
+                                        239
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3.2 CATER",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        }
+                    ],
+                    "index": 6
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        248,
+                        505,
+                        314
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                249,
+                                506,
+                                260
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        249,
+                                        506,
+                                        260
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In a second experiment, we tested Aloe on CATER, a widely-used object-tracking dataset [12, 14, 31,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                258,
+                                507,
+                                273
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        258,
+                                        507,
+                                        273
+                                    ],
+                                    "score": 1.0,
+                                    "content": "45]. In CATER, objects from the CLEVR dataset [21] move and potentially occlude other objects,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                270,
+                                506,
+                                282
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        270,
+                                        506,
+                                        282
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and the goal is to predict the location of a target object (called the snitch) in the final frame. Because",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                281,
+                                505,
+                                294
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        281,
+                                        505,
+                                        294
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the snitch could be occluded by multiple objects that could move in the meantime, a successful model",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                291,
+                                506,
+                                304
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        291,
+                                        506,
+                                        304
+                                    ],
+                                    "score": 1.0,
+                                    "content": "must be sensitive to notions of object permanence. CATER also includes a moving camera variant,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                303,
+                                330,
+                                315
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        303,
+                                        330,
+                                        315
+                                    ],
+                                    "score": 1.0,
+                                    "content": "which introduces additional complexities for the model.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        }
+                    ],
+                    "index": 9.5,
+                    "bbox_fs": [
+                        105,
+                        249,
+                        507,
+                        315
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        319,
+                        505,
+                        374
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                319,
+                                506,
+                                332
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        319,
+                                        506,
+                                        332
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Concretely, CATER is setup as a classification challenge. Objects are located in an xyz coordinate",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                330,
+                                506,
+                                342
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        330,
+                                        506,
+                                        342
+                                    ],
+                                    "score": 1.0,
+                                    "content": "system, where x and y range from -3 to 3. The xy plane is divided into a 6 by 6 grid, and the task is",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                341,
+                                506,
+                                354
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        341,
+                                        506,
+                                        354
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to predict the grid index of the snitch in the final frame. For Aloe, we use a classification loss (cross",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                352,
+                                505,
+                                365
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        352,
+                                        505,
+                                        365
+                                    ],
+                                    "score": 1.0,
+                                    "content": "entropy over the 36 possible grid indices) and an L1 loss (L1 distance between predicted grid cell and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                363,
+                                180,
+                                376
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        363,
+                                        180,
+                                        376
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the true grid cell).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        }
+                    ],
+                    "index": 15,
+                    "bbox_fs": [
+                        105,
+                        319,
+                        506,
+                        376
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        379,
+                        505,
+                        456
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                378,
+                                505,
+                                393
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        378,
+                                        505,
+                                        393
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Table 2 shows Aloe compared to state-of-the-art models in the literature on both static and moving",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                390,
+                                505,
+                                403
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        390,
+                                        505,
+                                        403
+                                    ],
+                                    "score": 1.0,
+                                    "content": "camera videos. R3D and R3D NL are the strongest two models evaluated by Girdhar and Ramanan",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                400,
+                                506,
+                                414
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        400,
+                                        506,
+                                        414
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[12]. OPNet, or the Object Permanence Network [31], is an architecture with inductive biases",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                411,
+                                505,
+                                426
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        411,
+                                        505,
+                                        426
+                                    ],
+                                    "score": 1.0,
+                                    "content": "designed for object tracking tasks; it was trained with extra supervised labels, namely the bounding",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                423,
+                                505,
+                                436
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        423,
+                                        505,
+                                        436
+                                    ],
+                                    "score": 1.0,
+                                    "content": "boxes for all objects (including occluded ones). Hopper is a multi-hop transformer model developed",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                434,
+                                505,
+                                447
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        434,
+                                        505,
+                                        447
+                                    ],
+                                    "score": 1.0,
+                                    "content": "simultaneously with this work [45]. One key component of Hopper is Hungarian matching between",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                444,
+                                387,
+                                459
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        444,
+                                        387,
+                                        459
+                                    ],
+                                    "score": 1.0,
+                                    "content": "objects of different frames, a strong inductive bias for object tracking.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        }
+                    ],
+                    "index": 21,
+                    "bbox_fs": [
+                        105,
+                        378,
+                        506,
+                        459
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        461,
+                        505,
+                        560
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                461,
+                                505,
+                                474
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        461,
+                                        505,
+                                        474
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We train Aloe simultaneously on both static and moving camera videos. Aloe outperforms the R3D",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                473,
+                                505,
+                                484
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        473,
+                                        505,
+                                        484
+                                    ],
+                                    "score": 1.0,
+                                    "content": "models for both static and moving cameras. We also ran Aloe with an additional auxiliary loss",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                483,
+                                507,
+                                496
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        483,
+                                        507,
+                                        496
+                                    ],
+                                    "score": 1.0,
+                                    "content": "consisting of the L1 distance between the predicted cell and the actual cell. With this additional loss,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                494,
+                                506,
+                                507
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        494,
+                                        506,
+                                        507
+                                    ],
+                                    "score": 1.0,
+                                    "content": "we get comparable results in the moving camera case as the R3D models for the static camera case.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                503,
+                                506,
+                                519
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        503,
+                                        506,
+                                        519
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Moreover, we achieve comparable accuracy as OPNet for accuracy and L1 distance, despite requiring",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                515,
+                                506,
+                                529
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        515,
+                                        506,
+                                        529
+                                    ],
+                                    "score": 1.0,
+                                    "content": "less supervision to train. Appendix E.2 gives a few sample outputs from Aloe; in particular we note",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                525,
+                                505,
+                                541
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        525,
+                                        505,
+                                        541
+                                    ],
+                                    "score": 1.0,
+                                    "content": "that it is able to find the target object in several cases where the object was occluded, demonstrating",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                538,
+                                504,
+                                550
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        538,
+                                        504,
+                                        550
+                                    ],
+                                    "score": 1.0,
+                                    "content": "that Aloe is able to do some level of object tracking. Finally,we find that an auxiliary self-supervised",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                548,
+                                505,
+                                561
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        548,
+                                        505,
+                                        561
+                                    ],
+                                    "score": 1.0,
+                                    "content": "loss helps the model perform well in the low data regime for CATER as well, as shown in Figure 3.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        }
+                    ],
+                    "index": 29,
+                    "bbox_fs": [
+                        105,
+                        461,
+                        507,
+                        561
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        579,
+                        158,
+                        590
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                577,
+                                160,
+                                592
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        577,
+                                        160,
+                                        592
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3.3 ACRE",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        }
+                    ],
+                    "index": 34
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        601,
+                        505,
+                        722
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                602,
+                                505,
+                                613
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        602,
+                                        505,
+                                        613
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Finally, we measured Aloe’s performance on ACRE, a causal induction dataset inspired by the Blicket",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                613,
+                                505,
+                                624
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        613,
+                                        505,
+                                        624
+                                    ],
+                                    "score": 1.0,
+                                    "content": "task from developmental psychology [13, 44]. ACRE is divided into a set of problems. In each",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                623,
+                                507,
+                                637
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        623,
+                                        507,
+                                        637
+                                    ],
+                                    "score": 1.0,
+                                    "content": "problem, certain objects are chosen to be “Blickets”, and this assignment changes across problems.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                635,
+                                506,
+                                647
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        635,
+                                        506,
+                                        647
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Each problem presents a context of six images to the model, where different objects are placed on a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                644,
+                                506,
+                                658
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        644,
+                                        506,
+                                        658
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Blicket machine that lights up if one of those objects is a Blicket. The model is asked whether an",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                656,
+                                506,
+                                668
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        656,
+                                        506,
+                                        668
+                                    ],
+                                    "score": 1.0,
+                                    "content": "unseen combination of objects will light up the Blicket machine. Besides “yes” and “no”, a third",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                667,
+                                506,
+                                679
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        667,
+                                        506,
+                                        679
+                                    ],
+                                    "score": 1.0,
+                                    "content": "possible answer is “undetermined”, which is the case if it is impossible to determine for certain",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                678,
+                                506,
+                                690
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        678,
+                                        506,
+                                        690
+                                    ],
+                                    "score": 1.0,
+                                    "content": "if the objects will light up the machine. Correct inference goes beyond mere correlation: even if",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                689,
+                                506,
+                                700
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        689,
+                                        506,
+                                        700
+                                    ],
+                                    "score": 1.0,
+                                    "content": "every context scene involving object A has a lit-up machine, A’s Blicketness is still uncertain if",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                700,
+                                505,
+                                713
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        700,
+                                        505,
+                                        713
+                                    ],
+                                    "score": 1.0,
+                                    "content": "each of those scenes can potentially be explained by another object (deduction of A’s Blicketness is",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                711,
+                                187,
+                                722
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        711,
+                                        187,
+                                        722
+                                    ],
+                                    "score": 1.0,
+                                    "content": "backward-blocked).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        }
+                    ],
+                    "index": 40,
+                    "bbox_fs": [
+                        105,
+                        602,
+                        507,
+                        722
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "table",
+                    "bbox": [
+                        96,
+                        69,
+                        515,
+                        117
+                    ],
+                    "blocks": [
+                        {
+                            "type": "table_body",
+                            "bbox": [
+                                96,
+                                69,
+                                515,
+                                117
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        96,
+                                        69,
+                                        515,
+                                        117
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                96,
+                                                69,
+                                                515,
+                                                117
+                                            ],
+                                            "score": 0.959,
+                                            "html": "<table><tr><td>Model</td><td>All (C)</td><td>D.R.</td><td>I.D.</td><td>S.0.</td><td>B.B.</td><td>All (S)</td><td>D.R.</td><td>1.D.</td><td>S.0.</td><td>B.B</td></tr><tr><td>CNN-BERT</td><td>43.79</td><td>54.07</td><td>46.88</td><td>40.57</td><td>28.79</td><td>39.93</td><td>55.97</td><td>68.25</td><td>0.00</td><td>45.59</td></tr><tr><td>NS-OPT</td><td>69.04</td><td>92.5</td><td>76.05</td><td>88.33</td><td>13.48</td><td>67.44</td><td>94.73</td><td>88.38</td><td>82.76</td><td>16.06</td></tr><tr><td>Aloe</td><td>91.76</td><td>97.14</td><td>90.8</td><td>96.8</td><td>78.81</td><td>93.90</td><td>97.18</td><td>71.24</td><td>98.97</td><td>94.48</td></tr></table>",
+                                            "type": "table",
+                                            "image_path": "5ddeb805759cc50f38656ec7ccfdc74c896c226a89b61f575f9dcf2d123229ee.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 1,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        96,
+                                        69,
+                                        515,
+                                        85.0
+                                    ],
+                                    "spans": [],
+                                    "index": 0
+                                },
+                                {
+                                    "bbox": [
+                                        96,
+                                        85.0,
+                                        515,
+                                        101.0
+                                    ],
+                                    "spans": [],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        96,
+                                        101.0,
+                                        515,
+                                        117.0
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                }
+                            ]
+                        },
+                        {
+                            "type": "table_caption",
+                            "bbox": [
+                                107,
+                                122,
+                                506,
+                                155
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        121,
+                                        505,
+                                        135
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                121,
+                                                505,
+                                                135
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Table 3: Performance on ACRE of Aloe compared to the best results from Zhang et al. [44], split",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 3
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        132,
+                                        506,
+                                        145
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                132,
+                                                197,
+                                                145
+                                            ],
+                                            "score": 1.0,
+                                            "content": "across inference type",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                197,
+                                                133,
+                                                222,
+                                                143
+                                            ],
+                                            "score": 0.32,
+                                            "content": "\\mathrm { { D . R = } } 1",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                222,
+                                                132,
+                                                263,
+                                                145
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Direct, I.D",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                263,
+                                                134,
+                                                270,
+                                                142
+                                            ],
+                                            "score": 0.33,
+                                            "content": "\\Vdash",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                270,
+                                                132,
+                                                305,
+                                                145
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Indirect,",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                305,
+                                                133,
+                                                329,
+                                                143
+                                            ],
+                                            "score": 0.31,
+                                            "content": "{ \\bf { S . O } } \\mathrm { { = } } \\mathrm { { ; } }",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                329,
+                                                132,
+                                                392,
+                                                145
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Screen-Off, B.B",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                392,
+                                                134,
+                                                399,
+                                                142
+                                            ],
+                                            "score": 0.43,
+                                            "content": "\\ c =",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                400,
+                                                132,
+                                                506,
+                                                145
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Backwards Blocking) and",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 4
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        144,
+                                        327,
+                                        156
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                144,
+                                                189,
+                                                156
+                                            ],
+                                            "score": 1.0,
+                                            "content": "generalization type (",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                189,
+                                                144,
+                                                202,
+                                                154
+                                            ],
+                                            "score": 0.65,
+                                            "content": "\\mathrm { C } =",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                202,
+                                                144,
+                                                264,
+                                                156
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Compositional,",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                264,
+                                                144,
+                                                277,
+                                                154
+                                            ],
+                                            "score": 0.66,
+                                            "content": "{ \\bf S } =",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                277,
+                                                144,
+                                                327,
+                                                156
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Systematic).",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 5
+                                }
+                            ],
+                            "index": 4
+                        }
+                    ],
+                    "index": 2.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        180,
+                        505,
+                        246
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                179,
+                                506,
+                                194
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        179,
+                                        506,
+                                        194
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Inference problems in ACRE are categorized by reasoning type: reasoning from direct evidence",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                190,
+                                505,
+                                204
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        190,
+                                        505,
+                                        204
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(one of the context frames show the query objects on a machine), reasoning from indirect evidence",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                200,
+                                506,
+                                216
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        200,
+                                        506,
+                                        216
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(Blicketness must be deduced by combining evidence from several frames), screened-off reasoning",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                211,
+                                506,
+                                227
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        211,
+                                        506,
+                                        227
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(presence of non-Blickets do not matter if a single Blicket is present), and backward-blocked reasoning",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                224,
+                                506,
+                                235
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        224,
+                                        506,
+                                        235
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(Blicketness cannot be deduced due to confounding variables). Please see Zhang et al. [44] for a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                234,
+                                308,
+                                247
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        234,
+                                        308,
+                                        247
+                                    ],
+                                    "score": 1.0,
+                                    "content": "more detailed discussion of these reasoning types.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        }
+                    ],
+                    "index": 8.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        250,
+                        505,
+                        382
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                251,
+                                505,
+                                263
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        251,
+                                        505,
+                                        263
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Table 3 show Aloe performance compared to a CNN-BERT baseline and to NS-OPT, a neuro-symbolic",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                261,
+                                505,
+                                275
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        261,
+                                        505,
+                                        275
+                                    ],
+                                    "score": 1.0,
+                                    "content": "model introduced in Zhang et al. [44]. Aloe outperforms all extant models for almost all reasoning",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                272,
+                                505,
+                                286
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        272,
+                                        505,
+                                        286
+                                    ],
+                                    "score": 1.0,
+                                    "content": "types and train-test splits. We did not need to do any tuning to apply our model to ACRE—settings",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                284,
+                                506,
+                                297
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        284,
+                                        506,
+                                        297
+                                    ],
+                                    "score": 1.0,
+                                    "content": "from CATER yielded the reported results on the first attempt. Contrary to widely-held opinions that",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                295,
+                                504,
+                                306
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        295,
+                                        504,
+                                        306
+                                    ],
+                                    "score": 1.0,
+                                    "content": "neural networks cannot generalize, Aloe generalizes in scenarios where the training and test sets",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                305,
+                                505,
+                                317
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        305,
+                                        505,
+                                        317
+                                    ],
+                                    "score": 1.0,
+                                    "content": "contain different visual features (compositional split) or different numbers of activated machines in",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                317,
+                                506,
+                                328
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        317,
+                                        506,
+                                        328
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the context (systematic split). Moreover, Aloe achieved by far the best performance on the backward-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                327,
+                                506,
+                                340
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        327,
+                                        506,
+                                        340
+                                    ],
+                                    "score": 1.0,
+                                    "content": "blocking task, which requires the model to “go beyond the simple covariation strategy to discover the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                337,
+                                507,
+                                351
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        337,
+                                        507,
+                                        351
+                                    ],
+                                    "score": 1.0,
+                                    "content": "hidden causal relations” [44], dispelling the notion that neural networks can only find correlation.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                348,
+                                506,
+                                362
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        348,
+                                        506,
+                                        362
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Comparison with NS-OPT (which uses object representations) and CNN-BERT (which uses attention)",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                360,
+                                505,
+                                372
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        360,
+                                        505,
+                                        372
+                                    ],
+                                    "score": 1.0,
+                                    "content": "shows that neither object representations nor attention alone is sufficient for the task; combining these",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                371,
+                                471,
+                                383
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        371,
+                                        471,
+                                        383
+                                    ],
+                                    "score": 1.0,
+                                    "content": "two ideas, as done in Aloe for instance, is essential for this complex reasoning task as well.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        }
+                    ],
+                    "index": 17.5
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        401,
+                        194,
+                        414
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                399,
+                                196,
+                                416
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        399,
+                                        196,
+                                        416
+                                    ],
+                                    "score": 1.0,
+                                    "content": "4 Related work",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        }
+                    ],
+                    "index": 24
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        428,
+                        505,
+                        548
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                428,
+                                505,
+                                441
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        428,
+                                        505,
+                                        441
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Self-attention for reasoning Various studies have shown that transformers [37] can manipulate",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                439,
+                                505,
+                                452
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        439,
+                                        505,
+                                        452
+                                    ],
+                                    "score": 1.0,
+                                    "content": "symbolic data in a manner traditionally associated with symbolic computation. For example, in",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                450,
+                                505,
+                                463
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        450,
+                                        505,
+                                        463
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Lample and Charton [23], a transformer model learned to do symbolic integration and solve ordinary",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                461,
+                                505,
+                                474
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        461,
+                                        505,
+                                        474
+                                    ],
+                                    "score": 1.0,
+                                    "content": "differential equations symbolically, tasks traditionally reserved for symbolic computer algebra",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                471,
+                                506,
+                                485
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        471,
+                                        506,
+                                        485
+                                    ],
+                                    "score": 1.0,
+                                    "content": "systems. Similarly, in Hahn et al. [17], a transformer model learned to solve formulas in propositional",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                483,
+                                506,
+                                496
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        483,
+                                        506,
+                                        496
+                                    ],
+                                    "score": 1.0,
+                                    "content": "logic and demonstrated some degree of generalization to out of distribution formulas. Finally, Brown",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                494,
+                                505,
+                                506
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        494,
+                                        505,
+                                        506
+                                    ],
+                                    "score": 1.0,
+                                    "content": "et al. [1] showed that a transformer trained for language modeling can also do simple analogical",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                505,
+                                505,
+                                517
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        505,
+                                        505,
+                                        517
+                                    ],
+                                    "score": 1.0,
+                                    "content": "reasoning tasks without explicit training. Although these models do not necessarily beat carefully",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                516,
+                                505,
+                                528
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        516,
+                                        505,
+                                        528
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tuned symbolic algorithms in all cases (especially on out of distribution data), they are an important",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                526,
+                                506,
+                                540
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        526,
+                                        506,
+                                        540
+                                    ],
+                                    "score": 1.0,
+                                    "content": "motivation for our proposed recipe for attaining strong reasoning capabilities from self-attention-based",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                538,
+                                249,
+                                549
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        538,
+                                        249,
+                                        549
+                                    ],
+                                    "score": 1.0,
+                                    "content": "models on visually grounded tasks.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        }
+                    ],
+                    "index": 30
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        564,
+                        505,
+                        608
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                564,
+                                505,
+                                577
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        564,
+                                        505,
+                                        577
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Object representations A wide body of research points to the importance of object segmentation",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                575,
+                                505,
+                                587
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        575,
+                                        505,
+                                        587
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and representation learning (see e.g. Garnelo and Shanahan [11] for a discussion). Various methods",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                586,
+                                505,
+                                598
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        586,
+                                        505,
+                                        598
+                                    ],
+                                    "score": 1.0,
+                                    "content": "have been proposed for object detection and feature extraction [2, 10, 15, 19, 25, 26, 29]. Past",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                597,
+                                486,
+                                609
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        597,
+                                        486,
+                                        609
+                                    ],
+                                    "score": 1.0,
+                                    "content": "research have also investigated using object based representations in downstream tasks [8, 28].",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        }
+                    ],
+                    "index": 37.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        623,
+                        505,
+                        722
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                623,
+                                505,
+                                636
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        623,
+                                        505,
+                                        636
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Self-supervised learning Another line of research concerns learning good representations through",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                635,
+                                506,
+                                647
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        635,
+                                        506,
+                                        647
+                                    ],
+                                    "score": 1.0,
+                                    "content": "self-supervised learning, with an unsupervised auxiliary loss to encourage the discovery of better",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                645,
+                                505,
+                                658
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        645,
+                                        505,
+                                        658
+                                    ],
+                                    "score": 1.0,
+                                    "content": "representations. These better representations could lead to improved performance on supervised",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                656,
+                                505,
+                                669
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        656,
+                                        505,
+                                        669
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tasks, especially when labeled data is scarce. In Devlin et al. [9], for instance, an auxiliary infill loss",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                667,
+                                505,
+                                680
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        667,
+                                        505,
+                                        680
+                                    ],
+                                    "score": 1.0,
+                                    "content": "allows the BERT model to benefit from pretraining on a large corpus of unlabeled data. Our approach",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                678,
+                                505,
+                                690
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        678,
+                                        505,
+                                        690
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to object-centric self-supervised learning is heavily inspired by the BERT infilling loss. Other studies",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                687,
+                                505,
+                                703
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        687,
+                                        505,
+                                        703
+                                    ],
+                                    "score": 1.0,
+                                    "content": "have shown similar benefits to auxiliary learning in vision as well [4, 16, 18]. These works apply",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                700,
+                                505,
+                                712
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        700,
+                                        505,
+                                        712
+                                    ],
+                                    "score": 1.0,
+                                    "content": "various forms of contrastive losses to predict scene dynamics, and the better representations that",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 47
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                711,
+                                429,
+                                723
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        711,
+                                        429,
+                                        723
+                                    ],
+                                    "score": 1.0,
+                                    "content": "result carry downstream benefits to supervised and reinforcement learning tasks.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 48
+                        }
+                    ],
+                    "index": 44
+                }
+            ],
+            "page_idx": 8,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        302,
+                        741,
+                        308,
+                        750
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                302,
+                                741,
+                                309,
+                                752
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        302,
+                                        741,
+                                        309,
+                                        752
+                                    ],
+                                    "score": 1.0,
+                                    "content": "9",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "table",
+                    "bbox": [
+                        96,
+                        69,
+                        515,
+                        117
+                    ],
+                    "blocks": [
+                        {
+                            "type": "table_body",
+                            "bbox": [
+                                96,
+                                69,
+                                515,
+                                117
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        96,
+                                        69,
+                                        515,
+                                        117
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                96,
+                                                69,
+                                                515,
+                                                117
+                                            ],
+                                            "score": 0.959,
+                                            "html": "<table><tr><td>Model</td><td>All (C)</td><td>D.R.</td><td>I.D.</td><td>S.0.</td><td>B.B.</td><td>All (S)</td><td>D.R.</td><td>1.D.</td><td>S.0.</td><td>B.B</td></tr><tr><td>CNN-BERT</td><td>43.79</td><td>54.07</td><td>46.88</td><td>40.57</td><td>28.79</td><td>39.93</td><td>55.97</td><td>68.25</td><td>0.00</td><td>45.59</td></tr><tr><td>NS-OPT</td><td>69.04</td><td>92.5</td><td>76.05</td><td>88.33</td><td>13.48</td><td>67.44</td><td>94.73</td><td>88.38</td><td>82.76</td><td>16.06</td></tr><tr><td>Aloe</td><td>91.76</td><td>97.14</td><td>90.8</td><td>96.8</td><td>78.81</td><td>93.90</td><td>97.18</td><td>71.24</td><td>98.97</td><td>94.48</td></tr></table>",
+                                            "type": "table",
+                                            "image_path": "5ddeb805759cc50f38656ec7ccfdc74c896c226a89b61f575f9dcf2d123229ee.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 1,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        96,
+                                        69,
+                                        515,
+                                        85.0
+                                    ],
+                                    "spans": [],
+                                    "index": 0
+                                },
+                                {
+                                    "bbox": [
+                                        96,
+                                        85.0,
+                                        515,
+                                        101.0
+                                    ],
+                                    "spans": [],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        96,
+                                        101.0,
+                                        515,
+                                        117.0
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                }
+                            ]
+                        },
+                        {
+                            "type": "table_caption",
+                            "bbox": [
+                                107,
+                                122,
+                                506,
+                                155
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        121,
+                                        505,
+                                        135
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                121,
+                                                505,
+                                                135
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Table 3: Performance on ACRE of Aloe compared to the best results from Zhang et al. [44], split",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 3
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        132,
+                                        506,
+                                        145
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                132,
+                                                197,
+                                                145
+                                            ],
+                                            "score": 1.0,
+                                            "content": "across inference type",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                197,
+                                                133,
+                                                222,
+                                                143
+                                            ],
+                                            "score": 0.32,
+                                            "content": "\\mathrm { { D . R = } } 1",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                222,
+                                                132,
+                                                263,
+                                                145
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Direct, I.D",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                263,
+                                                134,
+                                                270,
+                                                142
+                                            ],
+                                            "score": 0.33,
+                                            "content": "\\Vdash",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                270,
+                                                132,
+                                                305,
+                                                145
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Indirect,",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                305,
+                                                133,
+                                                329,
+                                                143
+                                            ],
+                                            "score": 0.31,
+                                            "content": "{ \\bf { S . O } } \\mathrm { { = } } \\mathrm { { ; } }",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                329,
+                                                132,
+                                                392,
+                                                145
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Screen-Off, B.B",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                392,
+                                                134,
+                                                399,
+                                                142
+                                            ],
+                                            "score": 0.43,
+                                            "content": "\\ c =",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                400,
+                                                132,
+                                                506,
+                                                145
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Backwards Blocking) and",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 4
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        144,
+                                        327,
+                                        156
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                144,
+                                                189,
+                                                156
+                                            ],
+                                            "score": 1.0,
+                                            "content": "generalization type (",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                189,
+                                                144,
+                                                202,
+                                                154
+                                            ],
+                                            "score": 0.65,
+                                            "content": "\\mathrm { C } =",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                202,
+                                                144,
+                                                264,
+                                                156
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Compositional,",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                264,
+                                                144,
+                                                277,
+                                                154
+                                            ],
+                                            "score": 0.66,
+                                            "content": "{ \\bf S } =",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                277,
+                                                144,
+                                                327,
+                                                156
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Systematic).",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 5
+                                }
+                            ],
+                            "index": 4
+                        }
+                    ],
+                    "index": 2.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        180,
+                        505,
+                        246
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                179,
+                                506,
+                                194
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        179,
+                                        506,
+                                        194
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Inference problems in ACRE are categorized by reasoning type: reasoning from direct evidence",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                190,
+                                505,
+                                204
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        190,
+                                        505,
+                                        204
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(one of the context frames show the query objects on a machine), reasoning from indirect evidence",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                200,
+                                506,
+                                216
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        200,
+                                        506,
+                                        216
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(Blicketness must be deduced by combining evidence from several frames), screened-off reasoning",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                211,
+                                506,
+                                227
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        211,
+                                        506,
+                                        227
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(presence of non-Blickets do not matter if a single Blicket is present), and backward-blocked reasoning",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                224,
+                                506,
+                                235
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        224,
+                                        506,
+                                        235
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(Blicketness cannot be deduced due to confounding variables). Please see Zhang et al. [44] for a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                234,
+                                308,
+                                247
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        234,
+                                        308,
+                                        247
+                                    ],
+                                    "score": 1.0,
+                                    "content": "more detailed discussion of these reasoning types.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        }
+                    ],
+                    "index": 8.5,
+                    "bbox_fs": [
+                        105,
+                        179,
+                        506,
+                        247
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        250,
+                        505,
+                        382
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                251,
+                                505,
+                                263
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        251,
+                                        505,
+                                        263
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Table 3 show Aloe performance compared to a CNN-BERT baseline and to NS-OPT, a neuro-symbolic",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                261,
+                                505,
+                                275
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        261,
+                                        505,
+                                        275
+                                    ],
+                                    "score": 1.0,
+                                    "content": "model introduced in Zhang et al. [44]. Aloe outperforms all extant models for almost all reasoning",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                272,
+                                505,
+                                286
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        272,
+                                        505,
+                                        286
+                                    ],
+                                    "score": 1.0,
+                                    "content": "types and train-test splits. We did not need to do any tuning to apply our model to ACRE—settings",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                284,
+                                506,
+                                297
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        284,
+                                        506,
+                                        297
+                                    ],
+                                    "score": 1.0,
+                                    "content": "from CATER yielded the reported results on the first attempt. Contrary to widely-held opinions that",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                295,
+                                504,
+                                306
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        295,
+                                        504,
+                                        306
+                                    ],
+                                    "score": 1.0,
+                                    "content": "neural networks cannot generalize, Aloe generalizes in scenarios where the training and test sets",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                305,
+                                505,
+                                317
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        305,
+                                        505,
+                                        317
+                                    ],
+                                    "score": 1.0,
+                                    "content": "contain different visual features (compositional split) or different numbers of activated machines in",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                317,
+                                506,
+                                328
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        317,
+                                        506,
+                                        328
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the context (systematic split). Moreover, Aloe achieved by far the best performance on the backward-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                327,
+                                506,
+                                340
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        327,
+                                        506,
+                                        340
+                                    ],
+                                    "score": 1.0,
+                                    "content": "blocking task, which requires the model to “go beyond the simple covariation strategy to discover the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                337,
+                                507,
+                                351
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        337,
+                                        507,
+                                        351
+                                    ],
+                                    "score": 1.0,
+                                    "content": "hidden causal relations” [44], dispelling the notion that neural networks can only find correlation.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                348,
+                                506,
+                                362
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        348,
+                                        506,
+                                        362
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Comparison with NS-OPT (which uses object representations) and CNN-BERT (which uses attention)",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                360,
+                                505,
+                                372
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        360,
+                                        505,
+                                        372
+                                    ],
+                                    "score": 1.0,
+                                    "content": "shows that neither object representations nor attention alone is sufficient for the task; combining these",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                371,
+                                471,
+                                383
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        371,
+                                        471,
+                                        383
+                                    ],
+                                    "score": 1.0,
+                                    "content": "two ideas, as done in Aloe for instance, is essential for this complex reasoning task as well.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        }
+                    ],
+                    "index": 17.5,
+                    "bbox_fs": [
+                        105,
+                        251,
+                        507,
+                        383
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        401,
+                        194,
+                        414
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                399,
+                                196,
+                                416
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        399,
+                                        196,
+                                        416
+                                    ],
+                                    "score": 1.0,
+                                    "content": "4 Related work",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        }
+                    ],
+                    "index": 24
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        428,
+                        505,
+                        548
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                428,
+                                505,
+                                441
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        428,
+                                        505,
+                                        441
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Self-attention for reasoning Various studies have shown that transformers [37] can manipulate",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                439,
+                                505,
+                                452
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        439,
+                                        505,
+                                        452
+                                    ],
+                                    "score": 1.0,
+                                    "content": "symbolic data in a manner traditionally associated with symbolic computation. For example, in",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                450,
+                                505,
+                                463
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        450,
+                                        505,
+                                        463
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Lample and Charton [23], a transformer model learned to do symbolic integration and solve ordinary",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                461,
+                                505,
+                                474
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        461,
+                                        505,
+                                        474
+                                    ],
+                                    "score": 1.0,
+                                    "content": "differential equations symbolically, tasks traditionally reserved for symbolic computer algebra",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                471,
+                                506,
+                                485
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        471,
+                                        506,
+                                        485
+                                    ],
+                                    "score": 1.0,
+                                    "content": "systems. Similarly, in Hahn et al. [17], a transformer model learned to solve formulas in propositional",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                483,
+                                506,
+                                496
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        483,
+                                        506,
+                                        496
+                                    ],
+                                    "score": 1.0,
+                                    "content": "logic and demonstrated some degree of generalization to out of distribution formulas. Finally, Brown",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                494,
+                                505,
+                                506
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        494,
+                                        505,
+                                        506
+                                    ],
+                                    "score": 1.0,
+                                    "content": "et al. [1] showed that a transformer trained for language modeling can also do simple analogical",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                505,
+                                505,
+                                517
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        505,
+                                        505,
+                                        517
+                                    ],
+                                    "score": 1.0,
+                                    "content": "reasoning tasks without explicit training. Although these models do not necessarily beat carefully",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                516,
+                                505,
+                                528
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        516,
+                                        505,
+                                        528
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tuned symbolic algorithms in all cases (especially on out of distribution data), they are an important",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                526,
+                                506,
+                                540
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        526,
+                                        506,
+                                        540
+                                    ],
+                                    "score": 1.0,
+                                    "content": "motivation for our proposed recipe for attaining strong reasoning capabilities from self-attention-based",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                538,
+                                249,
+                                549
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        538,
+                                        249,
+                                        549
+                                    ],
+                                    "score": 1.0,
+                                    "content": "models on visually grounded tasks.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        }
+                    ],
+                    "index": 30,
+                    "bbox_fs": [
+                        105,
+                        428,
+                        506,
+                        549
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        564,
+                        505,
+                        608
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                564,
+                                505,
+                                577
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        564,
+                                        505,
+                                        577
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Object representations A wide body of research points to the importance of object segmentation",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                575,
+                                505,
+                                587
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        575,
+                                        505,
+                                        587
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and representation learning (see e.g. Garnelo and Shanahan [11] for a discussion). Various methods",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                586,
+                                505,
+                                598
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        586,
+                                        505,
+                                        598
+                                    ],
+                                    "score": 1.0,
+                                    "content": "have been proposed for object detection and feature extraction [2, 10, 15, 19, 25, 26, 29]. Past",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                597,
+                                486,
+                                609
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        597,
+                                        486,
+                                        609
+                                    ],
+                                    "score": 1.0,
+                                    "content": "research have also investigated using object based representations in downstream tasks [8, 28].",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        }
+                    ],
+                    "index": 37.5,
+                    "bbox_fs": [
+                        105,
+                        564,
+                        505,
+                        609
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        623,
+                        505,
+                        722
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                623,
+                                505,
+                                636
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        623,
+                                        505,
+                                        636
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Self-supervised learning Another line of research concerns learning good representations through",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                635,
+                                506,
+                                647
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        635,
+                                        506,
+                                        647
+                                    ],
+                                    "score": 1.0,
+                                    "content": "self-supervised learning, with an unsupervised auxiliary loss to encourage the discovery of better",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                645,
+                                505,
+                                658
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        645,
+                                        505,
+                                        658
+                                    ],
+                                    "score": 1.0,
+                                    "content": "representations. These better representations could lead to improved performance on supervised",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                656,
+                                505,
+                                669
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        656,
+                                        505,
+                                        669
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tasks, especially when labeled data is scarce. In Devlin et al. [9], for instance, an auxiliary infill loss",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                667,
+                                505,
+                                680
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        667,
+                                        505,
+                                        680
+                                    ],
+                                    "score": 1.0,
+                                    "content": "allows the BERT model to benefit from pretraining on a large corpus of unlabeled data. Our approach",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                678,
+                                505,
+                                690
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        678,
+                                        505,
+                                        690
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to object-centric self-supervised learning is heavily inspired by the BERT infilling loss. Other studies",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                687,
+                                505,
+                                703
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        687,
+                                        505,
+                                        703
+                                    ],
+                                    "score": 1.0,
+                                    "content": "have shown similar benefits to auxiliary learning in vision as well [4, 16, 18]. These works apply",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                700,
+                                505,
+                                712
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        700,
+                                        505,
+                                        712
+                                    ],
+                                    "score": 1.0,
+                                    "content": "various forms of contrastive losses to predict scene dynamics, and the better representations that",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 47
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                711,
+                                429,
+                                723
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        711,
+                                        429,
+                                        723
+                                    ],
+                                    "score": 1.0,
+                                    "content": "result carry downstream benefits to supervised and reinforcement learning tasks.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 48
+                        }
+                    ],
+                    "index": 44,
+                    "bbox_fs": [
+                        105,
+                        623,
+                        506,
+                        723
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        73,
+                        505,
+                        149
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                71,
+                                505,
+                                86
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        71,
+                                        505,
+                                        86
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Vision and language in self-attention models Recently, many works have emerged on applying",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                84,
+                                505,
+                                96
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        84,
+                                        505,
+                                        96
+                                    ],
+                                    "score": 1.0,
+                                    "content": "transformer models to visual and multimodal data, for static images [24, 27, 33, 36] and videos",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                95,
+                                505,
+                                107
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        95,
+                                        505,
+                                        107
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[34, 35, 43]. These approaches combine the output of convolutional networks with language in various",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                105,
+                                506,
+                                119
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        105,
+                                        506,
+                                        119
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ways using self-attention. While these previous works focused on popular visual question answering",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                115,
+                                506,
+                                129
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        115,
+                                        506,
+                                        129
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tasks, which typically consist of descriptive questions only [41], we focus on understanding deeper",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                127,
+                                505,
+                                139
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        127,
+                                        505,
+                                        139
+                                    ],
+                                    "score": 1.0,
+                                    "content": "causal dynamics of videos. Together with these works, we provide more evidence that self-attention",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                138,
+                                469,
+                                151
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        138,
+                                        469,
+                                        151
+                                    ],
+                                    "score": 1.0,
+                                    "content": "between visual and language elements enables good performance on a diverse set of tasks.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        }
+                    ],
+                    "index": 3
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        154,
+                        505,
+                        231
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                153,
+                                505,
+                                168
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        153,
+                                        505,
+                                        168
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In addition, while the use of object representations for discretization in tasks involving static images",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                164,
+                                505,
+                                179
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        164,
+                                        505,
+                                        179
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is becoming more popular, the right way to discretize videos is less clear. We provide strong evidence",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                177,
+                                505,
+                                189
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        177,
+                                        505,
+                                        189
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in the form of ablation studies for architectural decisions that we claim are essential for higher",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                187,
+                                506,
+                                200
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        187,
+                                        506,
+                                        200
+                                    ],
+                                    "score": 1.0,
+                                    "content": "reasoning for this type of data: visual elements should correspond to physical objects in the videos",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                198,
+                                505,
+                                210
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        198,
+                                        505,
+                                        210
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and inter-frame attention between sub-frame entities (as opposed to inter-frame attention of entire",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                209,
+                                505,
+                                222
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        209,
+                                        505,
+                                        222
+                                    ],
+                                    "score": 1.0,
+                                    "content": "frames) is crucial. We also demonstrate the success of using unsupervised object segmentation",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                220,
+                                370,
+                                232
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        220,
+                                        370,
+                                        232
+                                    ],
+                                    "score": 1.0,
+                                    "content": "methods as opposed to the supervised methods used in past work.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        }
+                    ],
+                    "index": 10
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        246,
+                        183,
+                        260
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                244,
+                                185,
+                                263
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        244,
+                                        185,
+                                        263
+                                    ],
+                                    "score": 1.0,
+                                    "content": "5 Conclusion",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        }
+                    ],
+                    "index": 14
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        271,
+                        505,
+                        359
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                271,
+                                506,
+                                284
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        271,
+                                        506,
+                                        284
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We have presented Aloe, a model that obtains state-of-the-art performance on three different task",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                282,
+                                506,
+                                295
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        282,
+                                        506,
+                                        295
+                                    ],
+                                    "score": 1.0,
+                                    "content": "domains involving spatiotemporal reasoning about objects. In each of these tasks, previous state-of-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                293,
+                                506,
+                                306
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        293,
+                                        506,
+                                        306
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the-art results were established by models with modular, task-specific components. Aloe, by contrast,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                304,
+                                506,
+                                317
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        304,
+                                        506,
+                                        317
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is a unified solution to all three domains. Its flexibility comes from a reliance on only soft biases and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                315,
+                                506,
+                                327
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        315,
+                                        506,
+                                        327
+                                    ],
+                                    "score": 1.0,
+                                    "content": "learning objectives: self-attention over learned object embeddings and self-supervised learning of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                326,
+                                505,
+                                338
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        326,
+                                        505,
+                                        338
+                                    ],
+                                    "score": 1.0,
+                                    "content": "dynamics. We believe the simplicity of this approach is its strength, and hope that this fact, together",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                337,
+                                505,
+                                350
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        337,
+                                        505,
+                                        350
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with the provided code, makes it easy for others to adopt and apply to arbitrary spatio-temporal",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                349,
+                                189,
+                                360
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        349,
+                                        189,
+                                        360
+                                    ],
+                                    "score": 1.0,
+                                    "content": "reasoning problems.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        }
+                    ],
+                    "index": 18.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        364,
+                        505,
+                        463
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                364,
+                                505,
+                                377
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        364,
+                                        505,
+                                        377
+                                    ],
+                                    "score": 1.0,
+                                    "content": "On many of these spatio-temporal reasoning problems, previous state-of-the-art was achieved by",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                374,
+                                506,
+                                389
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        374,
+                                        506,
+                                        389
+                                    ],
+                                    "score": 1.0,
+                                    "content": "neuro-symbolic models [6, 11, 40, 41, 44]. Compared to neuro-symbolic models, Aloe can more",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                386,
+                                506,
+                                399
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        386,
+                                        506,
+                                        399
+                                    ],
+                                    "score": 1.0,
+                                    "content": "easily be adapted to other tasks. Indeed, the symbolic components of neuro-symbolic models are",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                396,
+                                505,
+                                409
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        396,
+                                        505,
+                                        409
+                                    ],
+                                    "score": 1.0,
+                                    "content": "often task-specific and not straightforwardly applicable to other tasks. Neuro-symbolic models do",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                406,
+                                506,
+                                422
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        406,
+                                        506,
+                                        422
+                                    ],
+                                    "score": 1.0,
+                                    "content": "have a few advantages, however. First, they are often easier to interpret. Despite the insights that can",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                418,
+                                505,
+                                432
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        418,
+                                        505,
+                                        432
+                                    ],
+                                    "score": 1.0,
+                                    "content": "be gleaned from Aloe’s attention weights, these soft computations are harder to interpret than the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                429,
+                                505,
+                                442
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        429,
+                                        505,
+                                        442
+                                    ],
+                                    "score": 1.0,
+                                    "content": "explicit symbolic computation found in neuro-symbolic models. Moreover, neuro-symbolic models",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                440,
+                                505,
+                                453
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        440,
+                                        505,
+                                        453
+                                    ],
+                                    "score": 1.0,
+                                    "content": "can be structured in a more modular fashion, which can enable effective generalization to sub-tasks",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                451,
+                                296,
+                                464
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        451,
+                                        296,
+                                        464
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of the task on which the model was trained [6].",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        }
+                    ],
+                    "index": 27
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        468,
+                        505,
+                        599
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                468,
+                                505,
+                                480
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        468,
+                                        505,
+                                        480
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Aloe also has some important limitations. First, it has only been applied to synthetic datasets. This",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                479,
+                                506,
+                                491
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        479,
+                                        506,
+                                        491
+                                    ],
+                                    "score": 1.0,
+                                    "content": "limitation is mainly due to the lack of real-world datasets that test for higher-order spatiotemporal",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                490,
+                                506,
+                                502
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        490,
+                                        506,
+                                        502
+                                    ],
+                                    "score": 1.0,
+                                    "content": "reasoning, although we are excited that new datasets such as Traffic QA will be released soon [38].",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                500,
+                                505,
+                                514
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        500,
+                                        505,
+                                        514
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Second, while the domains where Aloe is applied have been widely adopted and well-received by",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                511,
+                                506,
+                                524
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        511,
+                                        506,
+                                        524
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the research community, it remains possible that they do not evaluate the capacities that they aim to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                523,
+                                505,
+                                535
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        523,
+                                        505,
+                                        535
+                                    ],
+                                    "score": 1.0,
+                                    "content": "evaluate because of hidden biases or other factors. Regardless, we hope that this work stimulates the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                533,
+                                506,
+                                546
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        533,
+                                        506,
+                                        546
+                                    ],
+                                    "score": 1.0,
+                                    "content": "design and development of more challenging tasks that more closely approximate the ultimate goal",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                545,
+                                505,
+                                556
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        545,
+                                        505,
+                                        556
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of human or super-human-level visual, spatiotemporal and causal reasoning. Finally, from an ethical",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                554,
+                                505,
+                                569
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        554,
+                                        505,
+                                        569
+                                    ],
+                                    "score": 1.0,
+                                    "content": "point of view, our model may share the common drawback of deep-learning models in perpetuating",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                565,
+                                506,
+                                579
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        565,
+                                        506,
+                                        579
+                                    ],
+                                    "score": 1.0,
+                                    "content": "biases found in the training data, especially when applied to real world data. Development of causal",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                577,
+                                506,
+                                590
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        577,
+                                        506,
+                                        590
+                                    ],
+                                    "score": 1.0,
+                                    "content": "reasoning models could also invite problematic applications involving automated assignment of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                587,
+                                137,
+                                601
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        587,
+                                        137,
+                                        601
+                                    ],
+                                    "score": 1.0,
+                                    "content": "blame.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        }
+                    ],
+                    "index": 37.5
+                }
+            ],
+            "page_idx": 9,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        300,
+                        741,
+                        311,
+                        750
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                299,
+                                740,
+                                313,
+                                754
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        299,
+                                        740,
+                                        313,
+                                        754
+                                    ],
+                                    "score": 1.0,
+                                    "content": "10",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        73,
+                        505,
+                        149
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                71,
+                                505,
+                                86
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        71,
+                                        505,
+                                        86
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Vision and language in self-attention models Recently, many works have emerged on applying",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                84,
+                                505,
+                                96
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        84,
+                                        505,
+                                        96
+                                    ],
+                                    "score": 1.0,
+                                    "content": "transformer models to visual and multimodal data, for static images [24, 27, 33, 36] and videos",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                95,
+                                505,
+                                107
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        95,
+                                        505,
+                                        107
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[34, 35, 43]. These approaches combine the output of convolutional networks with language in various",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                105,
+                                506,
+                                119
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        105,
+                                        506,
+                                        119
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ways using self-attention. While these previous works focused on popular visual question answering",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                115,
+                                506,
+                                129
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        115,
+                                        506,
+                                        129
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tasks, which typically consist of descriptive questions only [41], we focus on understanding deeper",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                127,
+                                505,
+                                139
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        127,
+                                        505,
+                                        139
+                                    ],
+                                    "score": 1.0,
+                                    "content": "causal dynamics of videos. Together with these works, we provide more evidence that self-attention",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                138,
+                                469,
+                                151
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        138,
+                                        469,
+                                        151
+                                    ],
+                                    "score": 1.0,
+                                    "content": "between visual and language elements enables good performance on a diverse set of tasks.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        }
+                    ],
+                    "index": 3,
+                    "bbox_fs": [
+                        105,
+                        71,
+                        506,
+                        151
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        154,
+                        505,
+                        231
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                153,
+                                505,
+                                168
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        153,
+                                        505,
+                                        168
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In addition, while the use of object representations for discretization in tasks involving static images",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                164,
+                                505,
+                                179
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        164,
+                                        505,
+                                        179
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is becoming more popular, the right way to discretize videos is less clear. We provide strong evidence",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                177,
+                                505,
+                                189
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        177,
+                                        505,
+                                        189
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in the form of ablation studies for architectural decisions that we claim are essential for higher",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                187,
+                                506,
+                                200
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        187,
+                                        506,
+                                        200
+                                    ],
+                                    "score": 1.0,
+                                    "content": "reasoning for this type of data: visual elements should correspond to physical objects in the videos",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                198,
+                                505,
+                                210
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        198,
+                                        505,
+                                        210
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and inter-frame attention between sub-frame entities (as opposed to inter-frame attention of entire",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                209,
+                                505,
+                                222
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        209,
+                                        505,
+                                        222
+                                    ],
+                                    "score": 1.0,
+                                    "content": "frames) is crucial. We also demonstrate the success of using unsupervised object segmentation",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                220,
+                                370,
+                                232
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        220,
+                                        370,
+                                        232
+                                    ],
+                                    "score": 1.0,
+                                    "content": "methods as opposed to the supervised methods used in past work.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        }
+                    ],
+                    "index": 10,
+                    "bbox_fs": [
+                        104,
+                        153,
+                        506,
+                        232
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        246,
+                        183,
+                        260
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                244,
+                                185,
+                                263
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        244,
+                                        185,
+                                        263
+                                    ],
+                                    "score": 1.0,
+                                    "content": "5 Conclusion",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        }
+                    ],
+                    "index": 14
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        271,
+                        505,
+                        359
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                271,
+                                506,
+                                284
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        271,
+                                        506,
+                                        284
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We have presented Aloe, a model that obtains state-of-the-art performance on three different task",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                282,
+                                506,
+                                295
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        282,
+                                        506,
+                                        295
+                                    ],
+                                    "score": 1.0,
+                                    "content": "domains involving spatiotemporal reasoning about objects. In each of these tasks, previous state-of-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                293,
+                                506,
+                                306
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        293,
+                                        506,
+                                        306
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the-art results were established by models with modular, task-specific components. Aloe, by contrast,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                304,
+                                506,
+                                317
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        304,
+                                        506,
+                                        317
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is a unified solution to all three domains. Its flexibility comes from a reliance on only soft biases and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                315,
+                                506,
+                                327
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        315,
+                                        506,
+                                        327
+                                    ],
+                                    "score": 1.0,
+                                    "content": "learning objectives: self-attention over learned object embeddings and self-supervised learning of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                326,
+                                505,
+                                338
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        326,
+                                        505,
+                                        338
+                                    ],
+                                    "score": 1.0,
+                                    "content": "dynamics. We believe the simplicity of this approach is its strength, and hope that this fact, together",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                337,
+                                505,
+                                350
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        337,
+                                        505,
+                                        350
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with the provided code, makes it easy for others to adopt and apply to arbitrary spatio-temporal",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                349,
+                                189,
+                                360
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        349,
+                                        189,
+                                        360
+                                    ],
+                                    "score": 1.0,
+                                    "content": "reasoning problems.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        }
+                    ],
+                    "index": 18.5,
+                    "bbox_fs": [
+                        105,
+                        271,
+                        506,
+                        360
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        364,
+                        505,
+                        463
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                364,
+                                505,
+                                377
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        364,
+                                        505,
+                                        377
+                                    ],
+                                    "score": 1.0,
+                                    "content": "On many of these spatio-temporal reasoning problems, previous state-of-the-art was achieved by",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                374,
+                                506,
+                                389
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        374,
+                                        506,
+                                        389
+                                    ],
+                                    "score": 1.0,
+                                    "content": "neuro-symbolic models [6, 11, 40, 41, 44]. Compared to neuro-symbolic models, Aloe can more",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                386,
+                                506,
+                                399
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        386,
+                                        506,
+                                        399
+                                    ],
+                                    "score": 1.0,
+                                    "content": "easily be adapted to other tasks. Indeed, the symbolic components of neuro-symbolic models are",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                396,
+                                505,
+                                409
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        396,
+                                        505,
+                                        409
+                                    ],
+                                    "score": 1.0,
+                                    "content": "often task-specific and not straightforwardly applicable to other tasks. Neuro-symbolic models do",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                406,
+                                506,
+                                422
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        406,
+                                        506,
+                                        422
+                                    ],
+                                    "score": 1.0,
+                                    "content": "have a few advantages, however. First, they are often easier to interpret. Despite the insights that can",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                418,
+                                505,
+                                432
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        418,
+                                        505,
+                                        432
+                                    ],
+                                    "score": 1.0,
+                                    "content": "be gleaned from Aloe’s attention weights, these soft computations are harder to interpret than the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                429,
+                                505,
+                                442
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        429,
+                                        505,
+                                        442
+                                    ],
+                                    "score": 1.0,
+                                    "content": "explicit symbolic computation found in neuro-symbolic models. Moreover, neuro-symbolic models",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                440,
+                                505,
+                                453
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        440,
+                                        505,
+                                        453
+                                    ],
+                                    "score": 1.0,
+                                    "content": "can be structured in a more modular fashion, which can enable effective generalization to sub-tasks",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                451,
+                                296,
+                                464
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        451,
+                                        296,
+                                        464
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of the task on which the model was trained [6].",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        }
+                    ],
+                    "index": 27,
+                    "bbox_fs": [
+                        104,
+                        364,
+                        506,
+                        464
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        468,
+                        505,
+                        599
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                468,
+                                505,
+                                480
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        468,
+                                        505,
+                                        480
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Aloe also has some important limitations. First, it has only been applied to synthetic datasets. This",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                479,
+                                506,
+                                491
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        479,
+                                        506,
+                                        491
+                                    ],
+                                    "score": 1.0,
+                                    "content": "limitation is mainly due to the lack of real-world datasets that test for higher-order spatiotemporal",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                490,
+                                506,
+                                502
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        490,
+                                        506,
+                                        502
+                                    ],
+                                    "score": 1.0,
+                                    "content": "reasoning, although we are excited that new datasets such as Traffic QA will be released soon [38].",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                500,
+                                505,
+                                514
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        500,
+                                        505,
+                                        514
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Second, while the domains where Aloe is applied have been widely adopted and well-received by",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                511,
+                                506,
+                                524
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        511,
+                                        506,
+                                        524
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the research community, it remains possible that they do not evaluate the capacities that they aim to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                523,
+                                505,
+                                535
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        523,
+                                        505,
+                                        535
+                                    ],
+                                    "score": 1.0,
+                                    "content": "evaluate because of hidden biases or other factors. Regardless, we hope that this work stimulates the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                533,
+                                506,
+                                546
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        533,
+                                        506,
+                                        546
+                                    ],
+                                    "score": 1.0,
+                                    "content": "design and development of more challenging tasks that more closely approximate the ultimate goal",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                545,
+                                505,
+                                556
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        545,
+                                        505,
+                                        556
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of human or super-human-level visual, spatiotemporal and causal reasoning. Finally, from an ethical",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                554,
+                                505,
+                                569
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        554,
+                                        505,
+                                        569
+                                    ],
+                                    "score": 1.0,
+                                    "content": "point of view, our model may share the common drawback of deep-learning models in perpetuating",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                565,
+                                506,
+                                579
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        565,
+                                        506,
+                                        579
+                                    ],
+                                    "score": 1.0,
+                                    "content": "biases found in the training data, especially when applied to real world data. Development of causal",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                577,
+                                506,
+                                590
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        577,
+                                        506,
+                                        590
+                                    ],
+                                    "score": 1.0,
+                                    "content": "reasoning models could also invite problematic applications involving automated assignment of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                587,
+                                137,
+                                601
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        587,
+                                        137,
+                                        601
+                                    ],
+                                    "score": 1.0,
+                                    "content": "blame.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        }
+                    ],
+                    "index": 37.5,
+                    "bbox_fs": [
+                        105,
+                        468,
+                        506,
+                        601
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        60,
+                        507,
+                        726
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                70,
+                                165,
+                                86
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        70,
+                                        165,
+                                        86
+                                    ],
+                                    "score": 1.0,
+                                    "content": "References",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                109,
+                                89,
+                                506,
+                                103
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        109,
+                                        89,
+                                        506,
+                                        103
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[1] Tom B Brown, Benjamin Mann, Nick Ryder, Melanie Subbiah, Jared Kaplan, Prafulla Dhariwal,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                100,
+                                506,
+                                114
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        100,
+                                        506,
+                                        114
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Arvind Neelakantan, Pranav Shyam, Girish Sastry, Amanda Askell, et al. Language models are",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                111,
+                                365,
+                                125
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        111,
+                                        365,
+                                        125
+                                    ],
+                                    "score": 1.0,
+                                    "content": "few-shot learners. arXiv preprint arXiv:2005.14165, 2020.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                109,
+                                128,
+                                507,
+                                144
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        109,
+                                        128,
+                                        507,
+                                        144
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[2] Chris P. Burgess, Loic Matthey, Nick Watters, Rishabh Kabra, Irina Higgins, Matt Botvinick,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                140,
+                                506,
+                                154
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        140,
+                                        506,
+                                        154
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and Alex Lerchner. MONet: Unsupervised scene decomposition and representation. arXiv",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                125,
+                                150,
+                                267,
+                                164
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        125,
+                                        150,
+                                        267,
+                                        164
+                                    ],
+                                    "score": 1.0,
+                                    "content": "preprint arXiv:1901.11390, 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                109,
+                                168,
+                                506,
+                                181
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        109,
+                                        168,
+                                        506,
+                                        181
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[3] Nicolas Carion, Francisco Massa, Gabriel Synnaeve, Nicolas Usunier, Alexander Kirillov, and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                179,
+                                464,
+                                193
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        179,
+                                        464,
+                                        193
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Sergey Zagoruyko. End-to-end object detection with transformers. In ECCV, 2020.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                110,
+                                196,
+                                506,
+                                210
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        110,
+                                        196,
+                                        506,
+                                        210
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[4] Ting Chen, Simon Kornblith, Mohammad Norouzi, and Geoffrey Hinton. A simple framework",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                207,
+                                489,
+                                221
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        207,
+                                        489,
+                                        221
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for contrastive learning of visual representations. arXiv preprint arXiv:2002.05709, 2020.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                110,
+                                225,
+                                506,
+                                238
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        110,
+                                        225,
+                                        506,
+                                        238
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[5] Zhe Chen. Object-based attention: A tutorial review. Attention, Perception, & Psychophysics,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                235,
+                                480,
+                                249
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        235,
+                                        480,
+                                        249
+                                    ],
+                                    "score": 1.0,
+                                    "content": "74:784–802, 2012. URL https://doi.org/10.3758/s13414-012-0322-z.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                109,
+                                253,
+                                506,
+                                266
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        109,
+                                        253,
+                                        506,
+                                        266
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[6] Zhenfang Chen, Jiayuan Mao, Jiajun Wu, Kwan-Yee Kenneth Wong, Joshua B. Tenenbaum,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                263,
+                                505,
+                                277
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        263,
+                                        505,
+                                        277
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and Chuang Gan. Grounding physical object and event concepts through dynamic visual",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                273,
+                                509,
+                                290
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        273,
+                                        509,
+                                        290
+                                    ],
+                                    "score": 1.0,
+                                    "content": "reasoning. In International Conference on Learning Representations, 2021. URL https:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                286,
+                                312,
+                                300
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        286,
+                                        312,
+                                        300
+                                    ],
+                                    "score": 1.0,
+                                    "content": "//arxiv.org/pdf/2103.16564.pdf.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                109,
+                                303,
+                                506,
+                                317
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        109,
+                                        303,
+                                        506,
+                                        317
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[7] Zihang Dai, Zhilin Yang, Yiming Yang, William W Cohen, Jaime Carbonell, Quoc V Le,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                313,
+                                506,
+                                329
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        313,
+                                        506,
+                                        329
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and Ruslan Salakhutdinov. Transformer-xl: Attentive language models beyond a fixed-length",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                325,
+                                326,
+                                339
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        325,
+                                        326,
+                                        339
+                                    ],
+                                    "score": 1.0,
+                                    "content": "context. arXiv preprint arXiv:1901.02860, 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                110,
+                                342,
+                                506,
+                                356
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        110,
+                                        342,
+                                        506,
+                                        356
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[8] Mikyas T. Desta, Larry Chen, and Tomasz Kornuta. Object-based reasoning in VQA. CoRR,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                353,
+                                436,
+                                367
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        353,
+                                        436,
+                                        367
+                                    ],
+                                    "score": 1.0,
+                                    "content": "abs/1801.09718, 2018. URL http://arxiv.org/abs/1801.09718.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                110,
+                                371,
+                                506,
+                                385
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        110,
+                                        371,
+                                        506,
+                                        385
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[9] Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. BERT: Pre-training of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                381,
+                                506,
+                                396
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        381,
+                                        506,
+                                        396
+                                    ],
+                                    "score": 1.0,
+                                    "content": "deep bidirectional transformers for language understanding. arXiv preprint arXiv:1810.04805,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                391,
+                                155,
+                                406
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        391,
+                                        155,
+                                        406
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                410,
+                                506,
+                                423
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        410,
+                                        506,
+                                        423
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[10] Yilun Du, Kevin Smith, Tomer Ulman, Joshua Tenenbaum, and Jiajun Wu. Unsupervised",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                420,
+                                472,
+                                435
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        420,
+                                        472,
+                                        435
+                                    ],
+                                    "score": 1.0,
+                                    "content": "discovery of 3d physical objects from video. arXiv preprint arXiv:2007.12348, 2020.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                438,
+                                506,
+                                452
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        438,
+                                        506,
+                                        452
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[11] Marta Garnelo and Murray Shanahan. Reconciling deep learning with symbolic artificial",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                449,
+                                506,
+                                463
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        449,
+                                        506,
+                                        463
+                                    ],
+                                    "score": 1.0,
+                                    "content": "intelligence: representing objects and relations. Current Opinion in Behavioral Sciences, 29:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                461,
+                                182,
+                                472
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        461,
+                                        182,
+                                        472
+                                    ],
+                                    "score": 1.0,
+                                    "content": "17–23, 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                476,
+                                506,
+                                491
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        476,
+                                        506,
+                                        491
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[12] Rohit Girdhar and Deva Ramanan. CATER: A diagnostic dataset for Compositional Actions",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                487,
+                                299,
+                                501
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        487,
+                                        299,
+                                        501
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and TEmporal Reasoning. In ICLR, 2020.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                505,
+                                506,
+                                519
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        505,
+                                        506,
+                                        519
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[13] Alison Gopnik and David Sobel. Detecting blickets: how young children use information about",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                516,
+                                506,
+                                530
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        516,
+                                        506,
+                                        530
+                                    ],
+                                    "score": 1.0,
+                                    "content": "novel causal powers in categorization and induction. Child Dev, pages 1205–22, 2000. doi:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                528,
+                                240,
+                                540
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        528,
+                                        240,
+                                        540
+                                    ],
+                                    "score": 1.0,
+                                    "content": "10.1111/1467-8624.00224.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                545,
+                                506,
+                                558
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        545,
+                                        506,
+                                        558
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[14] Anirudh Goyal, Aniket Didolkar, Alex Lamb, Kartikeya Badola, Nan Rosemary Ke, Nasim",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                556,
+                                506,
+                                570
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        556,
+                                        506,
+                                        570
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Rahaman, Jonathan Binas, Charles Blundell, Michael Mozer, and Yoshua Bengio. Coordination",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                567,
+                                506,
+                                579
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        567,
+                                        506,
+                                        579
+                                    ],
+                                    "score": 1.0,
+                                    "content": "among neural modules through a shared global workspace. CoRR, abs/2103.01197, 2021. URL",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                578,
+                                324,
+                                591
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        578,
+                                        324,
+                                        591
+                                    ],
+                                    "score": 1.0,
+                                    "content": "https://arxiv.org/abs/2103.01197.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                593,
+                                507,
+                                610
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        593,
+                                        507,
+                                        610
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[15] Klaus Greff, Raphaël Lopez Kaufman, Rishabh Kabra, Nick Watters, Christopher Burgess,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                125,
+                                603,
+                                506,
+                                621
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        125,
+                                        603,
+                                        506,
+                                        621
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Daniel Zoran, Loïc Matthey, Matthew Botvinick, and Alexander Lerchner. Multi-object rep-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                617,
+                                506,
+                                630
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        617,
+                                        506,
+                                        630
+                                    ],
+                                    "score": 1.0,
+                                    "content": "resentation learning with iterative variational inference. CoRR, abs/1903.00450, 2019. URL",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                629,
+                                318,
+                                641
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        629,
+                                        318,
+                                        641
+                                    ],
+                                    "score": 1.0,
+                                    "content": "http://arxiv.org/abs/1903.00450.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                645,
+                                506,
+                                659
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        645,
+                                        506,
+                                        659
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[16] Karol Gregor, Danilo Jimenez Rezende, Frederic Besse, Yan Wu, Hamza Merzic,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                656,
+                                506,
+                                670
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        656,
+                                        506,
+                                        670
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and Aaron van den Oord. Shaping belief states with generative environment models",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                667,
+                                506,
+                                681
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        667,
+                                        506,
+                                        681
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for rl. In H. Wallach, H. Larochelle, A. Beygelzimer, F. d’ Alché-Buc, E. Fox, and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                125,
+                                676,
+                                506,
+                                693
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        125,
+                                        676,
+                                        506,
+                                        693
+                                    ],
+                                    "score": 1.0,
+                                    "content": "R. Garnett, editors, Advances in Neural Information Processing Systems 32, pages",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                686,
+                                507,
+                                704
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        686,
+                                        507,
+                                        704
+                                    ],
+                                    "score": 1.0,
+                                    "content": "13475–13487. Curran Associates, Inc., 2019. URL http://papers.nips.cc/paper/",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 47
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                700,
+                                511,
+                                714
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        700,
+                                        511,
+                                        714
+                                    ],
+                                    "score": 1.0,
+                                    "content": "9503-shaping-belief-states-with-generative-environment-models-f",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 48
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                711,
+                                153,
+                                725
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        711,
+                                        153,
+                                        725
+                                    ],
+                                    "score": 1.0,
+                                    "content": "pdf.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 49
+                        }
+                    ],
+                    "index": 24.5
+                }
+            ],
+            "page_idx": 10,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        300,
+                        741,
+                        310,
+                        750
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                299,
+                                740,
+                                312,
+                                755
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        299,
+                                        740,
+                                        312,
+                                        755
+                                    ],
+                                    "score": 1.0,
+                                    "content": "",
+                                    "type": "text",
+                                    "height": 15,
+                                    "width": 13
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "list",
+                    "bbox": [
+                        106,
+                        60,
+                        507,
+                        726
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                70,
+                                165,
+                                86
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        70,
+                                        165,
+                                        86
+                                    ],
+                                    "score": 1.0,
+                                    "content": "References",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0,
+                            "is_list_start_line": true,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                109,
+                                89,
+                                506,
+                                103
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        109,
+                                        89,
+                                        506,
+                                        103
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[1] Tom B Brown, Benjamin Mann, Nick Ryder, Melanie Subbiah, Jared Kaplan, Prafulla Dhariwal,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                100,
+                                506,
+                                114
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        100,
+                                        506,
+                                        114
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Arvind Neelakantan, Pranav Shyam, Girish Sastry, Amanda Askell, et al. Language models are",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                111,
+                                365,
+                                125
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        111,
+                                        365,
+                                        125
+                                    ],
+                                    "score": 1.0,
+                                    "content": "few-shot learners. arXiv preprint arXiv:2005.14165, 2020.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                109,
+                                128,
+                                507,
+                                144
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        109,
+                                        128,
+                                        507,
+                                        144
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[2] Chris P. Burgess, Loic Matthey, Nick Watters, Rishabh Kabra, Irina Higgins, Matt Botvinick,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                140,
+                                506,
+                                154
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        140,
+                                        506,
+                                        154
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and Alex Lerchner. MONet: Unsupervised scene decomposition and representation. arXiv",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                125,
+                                150,
+                                267,
+                                164
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        125,
+                                        150,
+                                        267,
+                                        164
+                                    ],
+                                    "score": 1.0,
+                                    "content": "preprint arXiv:1901.11390, 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                109,
+                                168,
+                                506,
+                                181
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        109,
+                                        168,
+                                        506,
+                                        181
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[3] Nicolas Carion, Francisco Massa, Gabriel Synnaeve, Nicolas Usunier, Alexander Kirillov, and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                179,
+                                464,
+                                193
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        179,
+                                        464,
+                                        193
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Sergey Zagoruyko. End-to-end object detection with transformers. In ECCV, 2020.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                110,
+                                196,
+                                506,
+                                210
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        110,
+                                        196,
+                                        506,
+                                        210
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[4] Ting Chen, Simon Kornblith, Mohammad Norouzi, and Geoffrey Hinton. A simple framework",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                207,
+                                489,
+                                221
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        207,
+                                        489,
+                                        221
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for contrastive learning of visual representations. arXiv preprint arXiv:2002.05709, 2020.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                110,
+                                225,
+                                506,
+                                238
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        110,
+                                        225,
+                                        506,
+                                        238
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[5] Zhe Chen. Object-based attention: A tutorial review. Attention, Perception, & Psychophysics,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                235,
+                                480,
+                                249
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        235,
+                                        480,
+                                        249
+                                    ],
+                                    "score": 1.0,
+                                    "content": "74:784–802, 2012. URL https://doi.org/10.3758/s13414-012-0322-z.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                109,
+                                253,
+                                506,
+                                266
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        109,
+                                        253,
+                                        506,
+                                        266
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[6] Zhenfang Chen, Jiayuan Mao, Jiajun Wu, Kwan-Yee Kenneth Wong, Joshua B. Tenenbaum,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                263,
+                                505,
+                                277
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        263,
+                                        505,
+                                        277
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and Chuang Gan. Grounding physical object and event concepts through dynamic visual",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                273,
+                                509,
+                                290
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        273,
+                                        509,
+                                        290
+                                    ],
+                                    "score": 1.0,
+                                    "content": "reasoning. In International Conference on Learning Representations, 2021. URL https:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                286,
+                                312,
+                                300
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        286,
+                                        312,
+                                        300
+                                    ],
+                                    "score": 1.0,
+                                    "content": "//arxiv.org/pdf/2103.16564.pdf.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                109,
+                                303,
+                                506,
+                                317
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        109,
+                                        303,
+                                        506,
+                                        317
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[7] Zihang Dai, Zhilin Yang, Yiming Yang, William W Cohen, Jaime Carbonell, Quoc V Le,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                313,
+                                506,
+                                329
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        313,
+                                        506,
+                                        329
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and Ruslan Salakhutdinov. Transformer-xl: Attentive language models beyond a fixed-length",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                325,
+                                326,
+                                339
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        325,
+                                        326,
+                                        339
+                                    ],
+                                    "score": 1.0,
+                                    "content": "context. arXiv preprint arXiv:1901.02860, 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                110,
+                                342,
+                                506,
+                                356
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        110,
+                                        342,
+                                        506,
+                                        356
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[8] Mikyas T. Desta, Larry Chen, and Tomasz Kornuta. Object-based reasoning in VQA. CoRR,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                353,
+                                436,
+                                367
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        353,
+                                        436,
+                                        367
+                                    ],
+                                    "score": 1.0,
+                                    "content": "abs/1801.09718, 2018. URL http://arxiv.org/abs/1801.09718.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                110,
+                                371,
+                                506,
+                                385
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        110,
+                                        371,
+                                        506,
+                                        385
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[9] Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. BERT: Pre-training of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                381,
+                                506,
+                                396
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        381,
+                                        506,
+                                        396
+                                    ],
+                                    "score": 1.0,
+                                    "content": "deep bidirectional transformers for language understanding. arXiv preprint arXiv:1810.04805,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                391,
+                                155,
+                                406
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        391,
+                                        155,
+                                        406
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                410,
+                                506,
+                                423
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        410,
+                                        506,
+                                        423
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[10] Yilun Du, Kevin Smith, Tomer Ulman, Joshua Tenenbaum, and Jiajun Wu. Unsupervised",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                420,
+                                472,
+                                435
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        420,
+                                        472,
+                                        435
+                                    ],
+                                    "score": 1.0,
+                                    "content": "discovery of 3d physical objects from video. arXiv preprint arXiv:2007.12348, 2020.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                438,
+                                506,
+                                452
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        438,
+                                        506,
+                                        452
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[11] Marta Garnelo and Murray Shanahan. Reconciling deep learning with symbolic artificial",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                449,
+                                506,
+                                463
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        449,
+                                        506,
+                                        463
+                                    ],
+                                    "score": 1.0,
+                                    "content": "intelligence: representing objects and relations. Current Opinion in Behavioral Sciences, 29:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                461,
+                                182,
+                                472
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        461,
+                                        182,
+                                        472
+                                    ],
+                                    "score": 1.0,
+                                    "content": "17–23, 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                476,
+                                506,
+                                491
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        476,
+                                        506,
+                                        491
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[12] Rohit Girdhar and Deva Ramanan. CATER: A diagnostic dataset for Compositional Actions",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                487,
+                                299,
+                                501
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        487,
+                                        299,
+                                        501
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and TEmporal Reasoning. In ICLR, 2020.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                505,
+                                506,
+                                519
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        505,
+                                        506,
+                                        519
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[13] Alison Gopnik and David Sobel. Detecting blickets: how young children use information about",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                516,
+                                506,
+                                530
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        516,
+                                        506,
+                                        530
+                                    ],
+                                    "score": 1.0,
+                                    "content": "novel causal powers in categorization and induction. Child Dev, pages 1205–22, 2000. doi:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                528,
+                                240,
+                                540
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        528,
+                                        240,
+                                        540
+                                    ],
+                                    "score": 1.0,
+                                    "content": "10.1111/1467-8624.00224.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                545,
+                                506,
+                                558
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        545,
+                                        506,
+                                        558
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[14] Anirudh Goyal, Aniket Didolkar, Alex Lamb, Kartikeya Badola, Nan Rosemary Ke, Nasim",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                556,
+                                506,
+                                570
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        556,
+                                        506,
+                                        570
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Rahaman, Jonathan Binas, Charles Blundell, Michael Mozer, and Yoshua Bengio. Coordination",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                567,
+                                506,
+                                579
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        567,
+                                        506,
+                                        579
+                                    ],
+                                    "score": 1.0,
+                                    "content": "among neural modules through a shared global workspace. CoRR, abs/2103.01197, 2021. URL",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                578,
+                                324,
+                                591
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        578,
+                                        324,
+                                        591
+                                    ],
+                                    "score": 1.0,
+                                    "content": "https://arxiv.org/abs/2103.01197.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                593,
+                                507,
+                                610
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        593,
+                                        507,
+                                        610
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[15] Klaus Greff, Raphaël Lopez Kaufman, Rishabh Kabra, Nick Watters, Christopher Burgess,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                125,
+                                603,
+                                506,
+                                621
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        125,
+                                        603,
+                                        506,
+                                        621
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Daniel Zoran, Loïc Matthey, Matthew Botvinick, and Alexander Lerchner. Multi-object rep-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                617,
+                                506,
+                                630
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        617,
+                                        506,
+                                        630
+                                    ],
+                                    "score": 1.0,
+                                    "content": "resentation learning with iterative variational inference. CoRR, abs/1903.00450, 2019. URL",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                629,
+                                318,
+                                641
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        629,
+                                        318,
+                                        641
+                                    ],
+                                    "score": 1.0,
+                                    "content": "http://arxiv.org/abs/1903.00450.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                645,
+                                506,
+                                659
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        645,
+                                        506,
+                                        659
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[16] Karol Gregor, Danilo Jimenez Rezende, Frederic Besse, Yan Wu, Hamza Merzic,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                656,
+                                506,
+                                670
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        656,
+                                        506,
+                                        670
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and Aaron van den Oord. Shaping belief states with generative environment models",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                667,
+                                506,
+                                681
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        667,
+                                        506,
+                                        681
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for rl. In H. Wallach, H. Larochelle, A. Beygelzimer, F. d’ Alché-Buc, E. Fox, and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                125,
+                                676,
+                                506,
+                                693
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        125,
+                                        676,
+                                        506,
+                                        693
+                                    ],
+                                    "score": 1.0,
+                                    "content": "R. Garnett, editors, Advances in Neural Information Processing Systems 32, pages",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                686,
+                                507,
+                                704
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        686,
+                                        507,
+                                        704
+                                    ],
+                                    "score": 1.0,
+                                    "content": "13475–13487. Curran Associates, Inc., 2019. URL http://papers.nips.cc/paper/",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 47
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                700,
+                                511,
+                                714
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        700,
+                                        511,
+                                        714
+                                    ],
+                                    "score": 1.0,
+                                    "content": "9503-shaping-belief-states-with-generative-environment-models-f",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 48
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                711,
+                                153,
+                                725
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        711,
+                                        153,
+                                        725
+                                    ],
+                                    "score": 1.0,
+                                    "content": "pdf.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 49,
+                            "is_list_end_line": true
+                        }
+                    ],
+                    "index": 24.5,
+                    "bbox_fs": [
+                        104,
+                        70,
+                        511,
+                        725
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        72,
+                        505,
+                        96
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                72,
+                                506,
+                                85
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        72,
+                                        506,
+                                        85
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[17] Christopher Hahn, Frederik Schmitt, Jens U. Kreber, Markus N. Rabe, and Bernd Finkbeiner.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                128,
+                                83,
+                                496,
+                                96
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        128,
+                                        83,
+                                        496,
+                                        96
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Transformers generalize to the semantics of logics. arXiv preprint arXiv:2003.04218, 2020.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        }
+                    ],
+                    "index": 0.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        101,
+                        504,
+                        125
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                101,
+                                505,
+                                115
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        101,
+                                        505,
+                                        115
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[18] Tengda Han, Weidi Xie, and Andrew Zisserman. Video representation learning by dense",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                112,
+                                495,
+                                126
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        112,
+                                        495,
+                                        126
+                                    ],
+                                    "score": 1.0,
+                                    "content": "predictive coding. In Workshop on Large Scale Holistic Video Understanding, ICCV, 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        }
+                    ],
+                    "index": 2.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        131,
+                        505,
+                        154
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                131,
+                                506,
+                                143
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        131,
+                                        506,
+                                        143
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[19] Kaiming He, Georgia Gkioxari, Piotr Dollár, and Ross B. Girshick. Mask R-CNN. CoRR,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                141,
+                                435,
+                                154
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        141,
+                                        435,
+                                        154
+                                    ],
+                                    "score": 1.0,
+                                    "content": "abs/1703.06870, 2017. URL http://arxiv.org/abs/1703.06870.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        }
+                    ],
+                    "index": 4.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        160,
+                        505,
+                        183
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                160,
+                                505,
+                                173
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        160,
+                                        505,
+                                        173
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[20] Drew A Hudson and Christopher D Manning. Compositional attention networks for machine",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                171,
+                                198,
+                                183
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        171,
+                                        198,
+                                        183
+                                    ],
+                                    "score": 1.0,
+                                    "content": "reasoning. 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        }
+                    ],
+                    "index": 6.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        189,
+                        506,
+                        234
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                189,
+                                507,
+                                201
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        189,
+                                        507,
+                                        201
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[21] Justin Johnson, Bharath Hariharan, Laurens van der Maaten, Li Fei-Fei, C. Lawrence Zitnick,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                199,
+                                506,
+                                213
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        199,
+                                        506,
+                                        213
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and Ross B. Girshick. CLEVR: A diagnostic dataset for compositional language and elementary",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                128,
+                                212,
+                                508,
+                                223
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        128,
+                                        212,
+                                        508,
+                                        223
+                                    ],
+                                    "score": 1.0,
+                                    "content": "visual reasoning. CoRR, abs/1612.06890, 2016. URL http://arxiv.org/abs/1612.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                222,
+                                164,
+                                234
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        222,
+                                        164,
+                                        234
+                                    ],
+                                    "score": 1.0,
+                                    "content": "06890.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        }
+                    ],
+                    "index": 9.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        240,
+                        505,
+                        263
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                239,
+                                505,
+                                254
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        239,
+                                        505,
+                                        254
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[22] Diederik P Kingma and Jimmy Ba. Adam: A method for stochastic optimization. arXiv preprint",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                251,
+                                226,
+                                262
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        251,
+                                        226,
+                                        262
+                                    ],
+                                    "score": 1.0,
+                                    "content": "arXiv:1412.6980, 2014.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        }
+                    ],
+                    "index": 12.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        269,
+                        505,
+                        303
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                269,
+                                506,
+                                282
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        269,
+                                        506,
+                                        282
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[23] Guillaume Lample and François Charton. Deep learning for symbolic mathematics. In Interna-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                280,
+                                505,
+                                293
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        280,
+                                        505,
+                                        293
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tional Conference on Learning Representations, 2020. URL https://openreview.net/",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                128,
+                                292,
+                                247,
+                                303
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        128,
+                                        292,
+                                        176,
+                                        303
+                                    ],
+                                    "score": 1.0,
+                                    "content": "forum?id",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        176,
+                                        293,
+                                        183,
+                                        300
+                                    ],
+                                    "score": 0.3,
+                                    "content": "{ . } = { }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        183,
+                                        292,
+                                        247,
+                                        303
+                                    ],
+                                    "score": 1.0,
+                                    "content": "S1eZYeHFDS.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        }
+                    ],
+                    "index": 15
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        309,
+                        505,
+                        343
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                309,
+                                506,
+                                322
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        309,
+                                        506,
+                                        322
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[24] Liunian Harold Li, Mark Yatskar, Da Yin, Cho-Jui Hsieh, and Kai-Wei Chang. Visualbert: A",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                320,
+                                506,
+                                333
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        320,
+                                        506,
+                                        333
+                                    ],
+                                    "score": 1.0,
+                                    "content": "simple and performant baseline for vision and language. arXiv preprint arXiv:1908.03557,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                330,
+                                154,
+                                344
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        330,
+                                        154,
+                                        344
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        }
+                    ],
+                    "index": 18
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        349,
+                        505,
+                        394
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                349,
+                                506,
+                                363
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        349,
+                                        506,
+                                        363
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[25] Zhixuan Lin, Yi-Fu Wu, Skand Vishwanath Peri, Weihao Sun, Gautam Singh, Fei Deng, Jindong",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                361,
+                                505,
+                                373
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        361,
+                                        505,
+                                        373
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Jiang, and Sungjin Ahn. SPACE: Unsupervised object-oriented scene representation via spatial",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                372,
+                                506,
+                                384
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        372,
+                                        506,
+                                        384
+                                    ],
+                                    "score": 1.0,
+                                    "content": "attention and decomposition. In International Conference on Learning Representations, 2020.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                383,
+                                407,
+                                395
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        383,
+                                        336,
+                                        395
+                                    ],
+                                    "score": 1.0,
+                                    "content": "URL https://openreview.net/forum?id",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        336,
+                                        384,
+                                        343,
+                                        392
+                                    ],
+                                    "score": 0.25,
+                                    "content": "\\underline { { \\underline { { \\mathbf { \\Pi } } } } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        343,
+                                        383,
+                                        407,
+                                        395
+                                    ],
+                                    "score": 1.0,
+                                    "content": "rkl03ySYDH.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        }
+                    ],
+                    "index": 21.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        400,
+                        504,
+                        434
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                399,
+                                506,
+                                415
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        399,
+                                        506,
+                                        415
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[26] Francesco Locatello, Dirk Weissenborn, Thomas Unterthiner, Aravindh Mahendran, Georg",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                410,
+                                506,
+                                426
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        410,
+                                        506,
+                                        426
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Heigold, Jakob Uszkoreit, Alexey Dosovitskiy, and Thomas Kipf1. Object-Centric Learning",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                423,
+                                489,
+                                435
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        423,
+                                        489,
+                                        435
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with Slot Attention. In Advances in Neural Information Processing Systems (NIPS), 2020.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        }
+                    ],
+                    "index": 25
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        441,
+                        504,
+                        474
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                439,
+                                506,
+                                454
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        439,
+                                        506,
+                                        454
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[27] Jiasen Lu, Dhruv Batra, Devi Parikh, and Stefan Lee. Vilbert: Pretraining task-agnostic",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                451,
+                                506,
+                                465
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        451,
+                                        506,
+                                        465
+                                    ],
+                                    "score": 1.0,
+                                    "content": "visiolinguistic representations for vision-and-language tasks. arXiv preprint arXiv:1908.02265,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                462,
+                                154,
+                                475
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        462,
+                                        154,
+                                        475
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        }
+                    ],
+                    "index": 28
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        480,
+                        505,
+                        515
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                480,
+                                506,
+                                493
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        480,
+                                        506,
+                                        493
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[28] David Raposo, Adam Santoro, David G. T. Barrett, Razvan Pascanu, Timothy P. Lillicrap, and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                492,
+                                507,
+                                504
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        492,
+                                        507,
+                                        504
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Peter W. Battaglia. Discovering objects and their relations from entangled scene representations.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                128,
+                                503,
+                                464,
+                                515
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        128,
+                                        503,
+                                        464,
+                                        515
+                                    ],
+                                    "score": 1.0,
+                                    "content": "CoRR, abs/1702.05068, 2017. URL http://arxiv.org/abs/1702.05068.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        }
+                    ],
+                    "index": 31
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        520,
+                        504,
+                        555
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                520,
+                                505,
+                                532
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        520,
+                                        505,
+                                        532
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[29] Shaoqing Ren, Kaiming He, Ross B. Girshick, and Jian Sun. Faster R-CNN: towards real-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                532,
+                                506,
+                                544
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        532,
+                                        506,
+                                        544
+                                    ],
+                                    "score": 1.0,
+                                    "content": "time object detection with region proposal networks. CoRR, abs/1506.01497, 2015. URL",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                543,
+                                319,
+                                555
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        543,
+                                        319,
+                                        555
+                                    ],
+                                    "score": 1.0,
+                                    "content": "http://arxiv.org/abs/1506.01497.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        }
+                    ],
+                    "index": 34
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        561,
+                        504,
+                        584
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                561,
+                                505,
+                                574
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        561,
+                                        505,
+                                        574
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[30] Pieter R. Roelfsema, Victor A. Lamme, and Henk Spekreijse. Object-based attention in the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                572,
+                                505,
+                                584
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        572,
+                                        505,
+                                        584
+                                    ],
+                                    "score": 1.0,
+                                    "content": "primary visual cortex of the macaque monkey. Nature, 395:376–381, 1998. doi: 10.1038/26475.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        }
+                    ],
+                    "index": 36.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        590,
+                        504,
+                        613
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                589,
+                                505,
+                                604
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        589,
+                                        505,
+                                        604
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[31] Aviv Shamsian, Ofri Kleinfeld, Amir Globerson, and Gal Chechik. Learning object permanence",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                601,
+                                340,
+                                613
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        601,
+                                        340,
+                                        613
+                                    ],
+                                    "score": 1.0,
+                                    "content": "from video. arXiv preprint arXiv:2003.10469, 2020.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        }
+                    ],
+                    "index": 38.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        619,
+                        504,
+                        642
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                619,
+                                507,
+                                633
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        619,
+                                        507,
+                                        633
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[32] Elizabeth Spelke. Core knowledge. The American psychologist, 55:1233–43, 12 2000. doi:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                128,
+                                630,
+                                261,
+                                642
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        128,
+                                        630,
+                                        261,
+                                        642
+                                    ],
+                                    "score": 1.0,
+                                    "content": "10.1037/0003-066X.55.11.1233.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        }
+                    ],
+                    "index": 40.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        648,
+                        504,
+                        683
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                648,
+                                505,
+                                662
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        648,
+                                        505,
+                                        662
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[33] Weijie Su, Xizhou Zhu, Yue Cao, Bin Li, Lewei Lu, Furu Wei, and Jifeng Dai. Vl-bert: Pre-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                659,
+                                505,
+                                674
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        659,
+                                        505,
+                                        674
+                                    ],
+                                    "score": 1.0,
+                                    "content": "training of generic visual-linguistic representations. In International Conference on Learning",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                670,
+                                502,
+                                684
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        670,
+                                        430,
+                                        684
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Representations, 2020. URL https://openreview.net/forum?id",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        431,
+                                        672,
+                                        437,
+                                        680
+                                    ],
+                                    "score": 0.44,
+                                    "content": "\\underline { { \\underline { { \\mathbf { \\Pi } } } } } =",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        438,
+                                        670,
+                                        502,
+                                        684
+                                    ],
+                                    "score": 1.0,
+                                    "content": "SygXPaEYvH.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        }
+                    ],
+                    "index": 43
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        689,
+                        507,
+                        722
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                688,
+                                506,
+                                701
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        688,
+                                        506,
+                                        701
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[34] Chen Sun, Fabien Baradel, Kevin Murphy, and Cordelia Schmid. Contrastive bidirectional",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                699,
+                                507,
+                                713
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        699,
+                                        507,
+                                        713
+                                    ],
+                                    "score": 1.0,
+                                    "content": "transformer for temporal representation learning. CoRR, abs/1906.05743, 2019. URL http:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        },
+                        {
+                            "bbox": [
+                                128,
+                                712,
+                                287,
+                                722
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        128,
+                                        712,
+                                        287,
+                                        722
+                                    ],
+                                    "score": 1.0,
+                                    "content": "//arxiv.org/abs/1906.05743.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 47
+                        }
+                    ],
+                    "index": 46
+                }
+            ],
+            "page_idx": 11,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        300,
+                        741,
+                        311,
+                        750
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                299,
+                                740,
+                                312,
+                                754
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        299,
+                                        740,
+                                        312,
+                                        754
+                                    ],
+                                    "score": 1.0,
+                                    "content": "12",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        72,
+                        505,
+                        96
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                72,
+                                506,
+                                85
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        72,
+                                        506,
+                                        85
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[17] Christopher Hahn, Frederik Schmitt, Jens U. Kreber, Markus N. Rabe, and Bernd Finkbeiner.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                128,
+                                83,
+                                496,
+                                96
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        128,
+                                        83,
+                                        496,
+                                        96
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Transformers generalize to the semantics of logics. arXiv preprint arXiv:2003.04218, 2020.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        }
+                    ],
+                    "index": 0.5,
+                    "bbox_fs": [
+                        106,
+                        72,
+                        506,
+                        96
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        101,
+                        504,
+                        125
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                101,
+                                505,
+                                115
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        101,
+                                        505,
+                                        115
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[18] Tengda Han, Weidi Xie, and Andrew Zisserman. Video representation learning by dense",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                112,
+                                495,
+                                126
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        112,
+                                        495,
+                                        126
+                                    ],
+                                    "score": 1.0,
+                                    "content": "predictive coding. In Workshop on Large Scale Holistic Video Understanding, ICCV, 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        }
+                    ],
+                    "index": 2.5,
+                    "bbox_fs": [
+                        105,
+                        101,
+                        505,
+                        126
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        131,
+                        505,
+                        154
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                131,
+                                506,
+                                143
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        131,
+                                        506,
+                                        143
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[19] Kaiming He, Georgia Gkioxari, Piotr Dollár, and Ross B. Girshick. Mask R-CNN. CoRR,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                141,
+                                435,
+                                154
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        141,
+                                        435,
+                                        154
+                                    ],
+                                    "score": 1.0,
+                                    "content": "abs/1703.06870, 2017. URL http://arxiv.org/abs/1703.06870.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        }
+                    ],
+                    "index": 4.5,
+                    "bbox_fs": [
+                        106,
+                        131,
+                        506,
+                        154
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        160,
+                        505,
+                        183
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                160,
+                                505,
+                                173
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        160,
+                                        505,
+                                        173
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[20] Drew A Hudson and Christopher D Manning. Compositional attention networks for machine",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                171,
+                                198,
+                                183
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        171,
+                                        198,
+                                        183
+                                    ],
+                                    "score": 1.0,
+                                    "content": "reasoning. 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        }
+                    ],
+                    "index": 6.5,
+                    "bbox_fs": [
+                        106,
+                        160,
+                        505,
+                        183
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        189,
+                        506,
+                        234
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                189,
+                                507,
+                                201
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        189,
+                                        507,
+                                        201
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[21] Justin Johnson, Bharath Hariharan, Laurens van der Maaten, Li Fei-Fei, C. Lawrence Zitnick,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                199,
+                                506,
+                                213
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        199,
+                                        506,
+                                        213
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and Ross B. Girshick. CLEVR: A diagnostic dataset for compositional language and elementary",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                128,
+                                212,
+                                508,
+                                223
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        128,
+                                        212,
+                                        508,
+                                        223
+                                    ],
+                                    "score": 1.0,
+                                    "content": "visual reasoning. CoRR, abs/1612.06890, 2016. URL http://arxiv.org/abs/1612.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                222,
+                                164,
+                                234
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        222,
+                                        164,
+                                        234
+                                    ],
+                                    "score": 1.0,
+                                    "content": "06890.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        }
+                    ],
+                    "index": 9.5,
+                    "bbox_fs": [
+                        105,
+                        189,
+                        508,
+                        234
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        240,
+                        505,
+                        263
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                239,
+                                505,
+                                254
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        239,
+                                        505,
+                                        254
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[22] Diederik P Kingma and Jimmy Ba. Adam: A method for stochastic optimization. arXiv preprint",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                251,
+                                226,
+                                262
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        251,
+                                        226,
+                                        262
+                                    ],
+                                    "score": 1.0,
+                                    "content": "arXiv:1412.6980, 2014.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        }
+                    ],
+                    "index": 12.5,
+                    "bbox_fs": [
+                        105,
+                        239,
+                        505,
+                        262
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        269,
+                        505,
+                        303
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                269,
+                                506,
+                                282
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        269,
+                                        506,
+                                        282
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[23] Guillaume Lample and François Charton. Deep learning for symbolic mathematics. In Interna-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                280,
+                                505,
+                                293
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        280,
+                                        505,
+                                        293
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tional Conference on Learning Representations, 2020. URL https://openreview.net/",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                128,
+                                292,
+                                247,
+                                303
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        128,
+                                        292,
+                                        176,
+                                        303
+                                    ],
+                                    "score": 1.0,
+                                    "content": "forum?id",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        176,
+                                        293,
+                                        183,
+                                        300
+                                    ],
+                                    "score": 0.3,
+                                    "content": "{ . } = { }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        183,
+                                        292,
+                                        247,
+                                        303
+                                    ],
+                                    "score": 1.0,
+                                    "content": "S1eZYeHFDS.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        }
+                    ],
+                    "index": 15,
+                    "bbox_fs": [
+                        106,
+                        269,
+                        506,
+                        303
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        309,
+                        505,
+                        343
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                309,
+                                506,
+                                322
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        309,
+                                        506,
+                                        322
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[24] Liunian Harold Li, Mark Yatskar, Da Yin, Cho-Jui Hsieh, and Kai-Wei Chang. Visualbert: A",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                320,
+                                506,
+                                333
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        320,
+                                        506,
+                                        333
+                                    ],
+                                    "score": 1.0,
+                                    "content": "simple and performant baseline for vision and language. arXiv preprint arXiv:1908.03557,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                330,
+                                154,
+                                344
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        330,
+                                        154,
+                                        344
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        }
+                    ],
+                    "index": 18,
+                    "bbox_fs": [
+                        106,
+                        309,
+                        506,
+                        344
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        349,
+                        505,
+                        394
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                349,
+                                506,
+                                363
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        349,
+                                        506,
+                                        363
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[25] Zhixuan Lin, Yi-Fu Wu, Skand Vishwanath Peri, Weihao Sun, Gautam Singh, Fei Deng, Jindong",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                361,
+                                505,
+                                373
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        361,
+                                        505,
+                                        373
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Jiang, and Sungjin Ahn. SPACE: Unsupervised object-oriented scene representation via spatial",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                372,
+                                506,
+                                384
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        372,
+                                        506,
+                                        384
+                                    ],
+                                    "score": 1.0,
+                                    "content": "attention and decomposition. In International Conference on Learning Representations, 2020.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                383,
+                                407,
+                                395
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        383,
+                                        336,
+                                        395
+                                    ],
+                                    "score": 1.0,
+                                    "content": "URL https://openreview.net/forum?id",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        336,
+                                        384,
+                                        343,
+                                        392
+                                    ],
+                                    "score": 0.25,
+                                    "content": "\\underline { { \\underline { { \\mathbf { \\Pi } } } } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        343,
+                                        383,
+                                        407,
+                                        395
+                                    ],
+                                    "score": 1.0,
+                                    "content": "rkl03ySYDH.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        }
+                    ],
+                    "index": 21.5,
+                    "bbox_fs": [
+                        105,
+                        349,
+                        506,
+                        395
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        400,
+                        504,
+                        434
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                399,
+                                506,
+                                415
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        399,
+                                        506,
+                                        415
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[26] Francesco Locatello, Dirk Weissenborn, Thomas Unterthiner, Aravindh Mahendran, Georg",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                410,
+                                506,
+                                426
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        410,
+                                        506,
+                                        426
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Heigold, Jakob Uszkoreit, Alexey Dosovitskiy, and Thomas Kipf1. Object-Centric Learning",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                423,
+                                489,
+                                435
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        423,
+                                        489,
+                                        435
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with Slot Attention. In Advances in Neural Information Processing Systems (NIPS), 2020.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        }
+                    ],
+                    "index": 25,
+                    "bbox_fs": [
+                        105,
+                        399,
+                        506,
+                        435
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        441,
+                        504,
+                        474
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                439,
+                                506,
+                                454
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        439,
+                                        506,
+                                        454
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[27] Jiasen Lu, Dhruv Batra, Devi Parikh, and Stefan Lee. Vilbert: Pretraining task-agnostic",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                451,
+                                506,
+                                465
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        451,
+                                        506,
+                                        465
+                                    ],
+                                    "score": 1.0,
+                                    "content": "visiolinguistic representations for vision-and-language tasks. arXiv preprint arXiv:1908.02265,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                462,
+                                154,
+                                475
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        462,
+                                        154,
+                                        475
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        }
+                    ],
+                    "index": 28,
+                    "bbox_fs": [
+                        106,
+                        439,
+                        506,
+                        475
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        480,
+                        505,
+                        515
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                480,
+                                506,
+                                493
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        480,
+                                        506,
+                                        493
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[28] David Raposo, Adam Santoro, David G. T. Barrett, Razvan Pascanu, Timothy P. Lillicrap, and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                492,
+                                507,
+                                504
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        492,
+                                        507,
+                                        504
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Peter W. Battaglia. Discovering objects and their relations from entangled scene representations.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                128,
+                                503,
+                                464,
+                                515
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        128,
+                                        503,
+                                        464,
+                                        515
+                                    ],
+                                    "score": 1.0,
+                                    "content": "CoRR, abs/1702.05068, 2017. URL http://arxiv.org/abs/1702.05068.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        }
+                    ],
+                    "index": 31,
+                    "bbox_fs": [
+                        105,
+                        480,
+                        507,
+                        515
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        520,
+                        504,
+                        555
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                520,
+                                505,
+                                532
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        520,
+                                        505,
+                                        532
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[29] Shaoqing Ren, Kaiming He, Ross B. Girshick, and Jian Sun. Faster R-CNN: towards real-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                532,
+                                506,
+                                544
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        532,
+                                        506,
+                                        544
+                                    ],
+                                    "score": 1.0,
+                                    "content": "time object detection with region proposal networks. CoRR, abs/1506.01497, 2015. URL",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                543,
+                                319,
+                                555
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        543,
+                                        319,
+                                        555
+                                    ],
+                                    "score": 1.0,
+                                    "content": "http://arxiv.org/abs/1506.01497.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        }
+                    ],
+                    "index": 34,
+                    "bbox_fs": [
+                        106,
+                        520,
+                        506,
+                        555
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        561,
+                        504,
+                        584
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                561,
+                                505,
+                                574
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        561,
+                                        505,
+                                        574
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[30] Pieter R. Roelfsema, Victor A. Lamme, and Henk Spekreijse. Object-based attention in the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                572,
+                                505,
+                                584
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        572,
+                                        505,
+                                        584
+                                    ],
+                                    "score": 1.0,
+                                    "content": "primary visual cortex of the macaque monkey. Nature, 395:376–381, 1998. doi: 10.1038/26475.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        }
+                    ],
+                    "index": 36.5,
+                    "bbox_fs": [
+                        106,
+                        561,
+                        505,
+                        584
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        590,
+                        504,
+                        613
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                589,
+                                505,
+                                604
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        589,
+                                        505,
+                                        604
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[31] Aviv Shamsian, Ofri Kleinfeld, Amir Globerson, and Gal Chechik. Learning object permanence",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                601,
+                                340,
+                                613
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        601,
+                                        340,
+                                        613
+                                    ],
+                                    "score": 1.0,
+                                    "content": "from video. arXiv preprint arXiv:2003.10469, 2020.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        }
+                    ],
+                    "index": 38.5,
+                    "bbox_fs": [
+                        105,
+                        589,
+                        505,
+                        613
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        619,
+                        504,
+                        642
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                619,
+                                507,
+                                633
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        619,
+                                        507,
+                                        633
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[32] Elizabeth Spelke. Core knowledge. The American psychologist, 55:1233–43, 12 2000. doi:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                128,
+                                630,
+                                261,
+                                642
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        128,
+                                        630,
+                                        261,
+                                        642
+                                    ],
+                                    "score": 1.0,
+                                    "content": "10.1037/0003-066X.55.11.1233.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        }
+                    ],
+                    "index": 40.5,
+                    "bbox_fs": [
+                        105,
+                        619,
+                        507,
+                        642
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        648,
+                        504,
+                        683
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                648,
+                                505,
+                                662
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        648,
+                                        505,
+                                        662
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[33] Weijie Su, Xizhou Zhu, Yue Cao, Bin Li, Lewei Lu, Furu Wei, and Jifeng Dai. Vl-bert: Pre-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                659,
+                                505,
+                                674
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        659,
+                                        505,
+                                        674
+                                    ],
+                                    "score": 1.0,
+                                    "content": "training of generic visual-linguistic representations. In International Conference on Learning",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                670,
+                                502,
+                                684
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        670,
+                                        430,
+                                        684
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Representations, 2020. URL https://openreview.net/forum?id",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        431,
+                                        672,
+                                        437,
+                                        680
+                                    ],
+                                    "score": 0.44,
+                                    "content": "\\underline { { \\underline { { \\mathbf { \\Pi } } } } } =",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        438,
+                                        670,
+                                        502,
+                                        684
+                                    ],
+                                    "score": 1.0,
+                                    "content": "SygXPaEYvH.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        }
+                    ],
+                    "index": 43,
+                    "bbox_fs": [
+                        106,
+                        648,
+                        505,
+                        684
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        689,
+                        507,
+                        722
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                688,
+                                506,
+                                701
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        688,
+                                        506,
+                                        701
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[34] Chen Sun, Fabien Baradel, Kevin Murphy, and Cordelia Schmid. Contrastive bidirectional",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                699,
+                                507,
+                                713
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        699,
+                                        507,
+                                        713
+                                    ],
+                                    "score": 1.0,
+                                    "content": "transformer for temporal representation learning. CoRR, abs/1906.05743, 2019. URL http:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        },
+                        {
+                            "bbox": [
+                                128,
+                                712,
+                                287,
+                                722
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        128,
+                                        712,
+                                        287,
+                                        722
+                                    ],
+                                    "score": 1.0,
+                                    "content": "//arxiv.org/abs/1906.05743.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 47
+                        }
+                    ],
+                    "index": 46,
+                    "bbox_fs": [
+                        105,
+                        688,
+                        507,
+                        722
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        72,
+                        504,
+                        106
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                72,
+                                506,
+                                85
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        72,
+                                        506,
+                                        85
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[35] Chen Sun, Austin Myers, Carl Vondrick, Kevin Murphy, and Cordelia Schmid. Videobert: A",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                83,
+                                506,
+                                96
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        83,
+                                        506,
+                                        96
+                                    ],
+                                    "score": 1.0,
+                                    "content": "joint model for video and language representation learning. In Proceedings of the IEEE/CVF",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                128,
+                                95,
+                                407,
+                                107
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        128,
+                                        95,
+                                        407,
+                                        107
+                                    ],
+                                    "score": 1.0,
+                                    "content": "International Conference on Computer Vision (ICCV), October 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        }
+                    ],
+                    "index": 1
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        113,
+                        504,
+                        147
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                113,
+                                505,
+                                126
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        113,
+                                        505,
+                                        126
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[36] Hao Tan and Mohit Bansal. Lxmert: Learning cross-modality encoder representations from",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                123,
+                                505,
+                                137
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        123,
+                                        505,
+                                        137
+                                    ],
+                                    "score": 1.0,
+                                    "content": "transformers. In Proceedings of the 2019 Conference on Empirical Methods in Natural Language",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                135,
+                                202,
+                                147
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        135,
+                                        202,
+                                        147
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Processing, 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        }
+                    ],
+                    "index": 4
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        153,
+                        504,
+                        187
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                153,
+                                506,
+                                167
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        153,
+                                        506,
+                                        167
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[37] Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N Gomez,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                164,
+                                506,
+                                177
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        164,
+                                        506,
+                                        177
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Łukasz Kaiser, and Illia Polosukhin. Attention is all you need. In Advances in neural information",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                177,
+                                308,
+                                187
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        177,
+                                        308,
+                                        187
+                                    ],
+                                    "score": 1.0,
+                                    "content": "processing systems, pages 5998–6008, 2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        }
+                    ],
+                    "index": 7
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        194,
+                        504,
+                        228
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                195,
+                                505,
+                                207
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        195,
+                                        505,
+                                        207
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[38] Li Xu, He Huang, and Jun Liu. Sutd-trafficqa: A question answering benchmark and an efficient",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                205,
+                                506,
+                                219
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        205,
+                                        506,
+                                        219
+                                    ],
+                                    "score": 1.0,
+                                    "content": "network for video reasoning over traffic events. In Proceedings of the IEEE Conference on",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                215,
+                                325,
+                                229
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        215,
+                                        325,
+                                        229
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Computer Vision and Pattern Recognition, 2021.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        }
+                    ],
+                    "index": 10
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        235,
+                        505,
+                        269
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                235,
+                                507,
+                                248
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        235,
+                                        507,
+                                        248
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[39] Kexin Yi, Jiajun Wu, Chuang Gan, Antonio Torralba, Pushmeet Kohli, and Joshua B Tenenbaum.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                245,
+                                507,
+                                260
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        245,
+                                        507,
+                                        260
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Neural-Symbolic VQA: Disentangling Reasoning from Vision and Language Understanding.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                257,
+                                408,
+                                270
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        257,
+                                        408,
+                                        270
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In Advances in Neural Information Processing Systems (NIPS), 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        }
+                    ],
+                    "index": 13
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        275,
+                        504,
+                        310
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                275,
+                                506,
+                                289
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        275,
+                                        506,
+                                        289
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[40] Kexin Yi, Jiajun Wu, Chuang Gan, Antonio Torralba, Pushmeet Kohli, and Joshua B. Tenenbaum.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                286,
+                                505,
+                                300
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        286,
+                                        505,
+                                        300
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Neural-symbolic vqa: Disentangling reasoning from vision and language understanding. In",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                298,
+                                441,
+                                311
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        298,
+                                        441,
+                                        311
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Advances in Neural Information Processing Systems, pages 1039–1050, 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        }
+                    ],
+                    "index": 16
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        316,
+                        505,
+                        361
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                317,
+                                505,
+                                330
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        317,
+                                        505,
+                                        330
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[41] Kexin Yi, Chuang Gan, Yunzhu Li, Pushmeet Kohli, Jiajun Wu, Antonio Torralba, and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                327,
+                                507,
+                                340
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        327,
+                                        507,
+                                        340
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Joshua B. Tenenbaum. CLEVRER: Collision events for video representation and rea-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                339,
+                                507,
+                                352
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        339,
+                                        507,
+                                        352
+                                    ],
+                                    "score": 1.0,
+                                    "content": "soning. In International Conference on Learning Representations, 2020. URL https:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                350,
+                                347,
+                                362
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        350,
+                                        277,
+                                        362
+                                    ],
+                                    "score": 1.0,
+                                    "content": "//openreview.net/forum?id",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        277,
+                                        351,
+                                        284,
+                                        359
+                                    ],
+                                    "score": 0.57,
+                                    "content": "\\underline { { \\underline { { \\mathbf { \\Pi } } } } } =",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        285,
+                                        350,
+                                        347,
+                                        362
+                                    ],
+                                    "score": 1.0,
+                                    "content": "HkxYzANYDB.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        }
+                    ],
+                    "index": 19.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        368,
+                        506,
+                        424
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                368,
+                                505,
+                                381
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        368,
+                                        505,
+                                        381
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[42] Yang You, Jing Li, Sashank Reddi, Jonathan Hseu, Sanjiv Kumar, Srinadh Bhojanapalli, Xiaodan",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                378,
+                                505,
+                                393
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        378,
+                                        505,
+                                        393
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Song, James Demmel, and Cho-Jui Hsieh. Large batch optimization for deep learning: Training",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                389,
+                                505,
+                                404
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        389,
+                                        505,
+                                        404
+                                    ],
+                                    "score": 1.0,
+                                    "content": "BERT in 76 minutes. Technical Report UCB/EECS-2019-103, EECS Department, University",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                400,
+                                505,
+                                414
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        400,
+                                        505,
+                                        414
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of California, Berkeley, Jun 2019. URL http://www2.eecs.berkeley.edu/Pubs/",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                128,
+                                412,
+                                325,
+                                424
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        128,
+                                        412,
+                                        325,
+                                        424
+                                    ],
+                                    "score": 1.0,
+                                    "content": "TechRpts/2019/EECS-2019-103.html.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        }
+                    ],
+                    "index": 24
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        430,
+                        505,
+                        486
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                431,
+                                506,
+                                443
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        431,
+                                        506,
+                                        443
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[43] Vinicius Zambaldi, David Raposo, Adam Santoro, Victor Bapst, Yujia Li, Igor Babuschkin,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                441,
+                                506,
+                                455
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        441,
+                                        506,
+                                        455
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Karl Tuyls, David Reichert, Timothy Lillicrap, Edward Lockhart, Murray Shanahan, Victoria",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                451,
+                                506,
+                                468
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        451,
+                                        506,
+                                        468
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Langston, Razvan Pascanu, Matthew Botvinick, Oriol Vinyals, and Peter Battaglia. Deep",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                462,
+                                505,
+                                478
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        462,
+                                        505,
+                                        478
+                                    ],
+                                    "score": 1.0,
+                                    "content": "reinforcement learning with relational inductive biases. In International Conference on Learning",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                474,
+                                501,
+                                488
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        474,
+                                        430,
+                                        488
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Representations, 2019. URL https://openreview.net/forum?id",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        431,
+                                        476,
+                                        438,
+                                        484
+                                    ],
+                                    "score": 0.65,
+                                    "content": "\\underline { { \\underline { { \\mathbf { \\Pi } } } } } =",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        438,
+                                        474,
+                                        501,
+                                        488
+                                    ],
+                                    "score": 1.0,
+                                    "content": "HkxaFoC9KQ.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        }
+                    ],
+                    "index": 29
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        104,
+                        493,
+                        504,
+                        527
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                493,
+                                506,
+                                506
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        493,
+                                        506,
+                                        506
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[44] Chi Zhang, Baoxiong Jia, Mark Edmonds, Song-Chun Zhu, and Yixin Zhu. Acre: Abstract",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                505,
+                                505,
+                                517
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        505,
+                                        505,
+                                        517
+                                    ],
+                                    "score": 1.0,
+                                    "content": "causal reasoning beyond covariation. In Proceedings of the IEEE Conference on Computer",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                514,
+                                315,
+                                528
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        514,
+                                        315,
+                                        528
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Vision and Pattern Recognition (CVPR), 2021.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        }
+                    ],
+                    "index": 33
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        534,
+                        506,
+                        579
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                534,
+                                506,
+                                547
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        534,
+                                        506,
+                                        547
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[45] Honglu Zhou, Asim Kadav, Farley Lai, Alexandru Niculescu-Mizil, Martin Renqiang Min,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                545,
+                                506,
+                                558
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        545,
+                                        506,
+                                        558
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Mubbasir Kapadia, and Hans Peter Graf. Hopper: Multi-hop transformer for spatiotemporal",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                556,
+                                507,
+                                569
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        556,
+                                        507,
+                                        569
+                                    ],
+                                    "score": 1.0,
+                                    "content": "reasoning. In International Conference on Learning Representations, 2021. URL https:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                568,
+                                289,
+                                579
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        568,
+                                        289,
+                                        579
+                                    ],
+                                    "score": 1.0,
+                                    "content": "//arxiv.org/abs/2103.10574.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        }
+                    ],
+                    "index": 36.5
+                }
+            ],
+            "page_idx": 12,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        300,
+                        741,
+                        311,
+                        750
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                299,
+                                740,
+                                313,
+                                754
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        299,
+                                        740,
+                                        313,
+                                        754
+                                    ],
+                                    "score": 1.0,
+                                    "content": "13",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        72,
+                        504,
+                        106
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                72,
+                                506,
+                                85
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        72,
+                                        506,
+                                        85
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[35] Chen Sun, Austin Myers, Carl Vondrick, Kevin Murphy, and Cordelia Schmid. Videobert: A",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                83,
+                                506,
+                                96
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        83,
+                                        506,
+                                        96
+                                    ],
+                                    "score": 1.0,
+                                    "content": "joint model for video and language representation learning. In Proceedings of the IEEE/CVF",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                128,
+                                95,
+                                407,
+                                107
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        128,
+                                        95,
+                                        407,
+                                        107
+                                    ],
+                                    "score": 1.0,
+                                    "content": "International Conference on Computer Vision (ICCV), October 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        }
+                    ],
+                    "index": 1,
+                    "bbox_fs": [
+                        105,
+                        72,
+                        506,
+                        107
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        113,
+                        504,
+                        147
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                113,
+                                505,
+                                126
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        113,
+                                        505,
+                                        126
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[36] Hao Tan and Mohit Bansal. Lxmert: Learning cross-modality encoder representations from",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                123,
+                                505,
+                                137
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        123,
+                                        505,
+                                        137
+                                    ],
+                                    "score": 1.0,
+                                    "content": "transformers. In Proceedings of the 2019 Conference on Empirical Methods in Natural Language",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                135,
+                                202,
+                                147
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        135,
+                                        202,
+                                        147
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Processing, 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        }
+                    ],
+                    "index": 4,
+                    "bbox_fs": [
+                        105,
+                        113,
+                        505,
+                        147
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        153,
+                        504,
+                        187
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                153,
+                                506,
+                                167
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        153,
+                                        506,
+                                        167
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[37] Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N Gomez,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                164,
+                                506,
+                                177
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        164,
+                                        506,
+                                        177
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Łukasz Kaiser, and Illia Polosukhin. Attention is all you need. In Advances in neural information",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                177,
+                                308,
+                                187
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        177,
+                                        308,
+                                        187
+                                    ],
+                                    "score": 1.0,
+                                    "content": "processing systems, pages 5998–6008, 2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        }
+                    ],
+                    "index": 7,
+                    "bbox_fs": [
+                        105,
+                        153,
+                        506,
+                        187
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        194,
+                        504,
+                        228
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                195,
+                                505,
+                                207
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        195,
+                                        505,
+                                        207
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[38] Li Xu, He Huang, and Jun Liu. Sutd-trafficqa: A question answering benchmark and an efficient",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                205,
+                                506,
+                                219
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        205,
+                                        506,
+                                        219
+                                    ],
+                                    "score": 1.0,
+                                    "content": "network for video reasoning over traffic events. In Proceedings of the IEEE Conference on",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                215,
+                                325,
+                                229
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        215,
+                                        325,
+                                        229
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Computer Vision and Pattern Recognition, 2021.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        }
+                    ],
+                    "index": 10,
+                    "bbox_fs": [
+                        106,
+                        195,
+                        506,
+                        229
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        235,
+                        505,
+                        269
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                235,
+                                507,
+                                248
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        235,
+                                        507,
+                                        248
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[39] Kexin Yi, Jiajun Wu, Chuang Gan, Antonio Torralba, Pushmeet Kohli, and Joshua B Tenenbaum.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                245,
+                                507,
+                                260
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        245,
+                                        507,
+                                        260
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Neural-Symbolic VQA: Disentangling Reasoning from Vision and Language Understanding.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                257,
+                                408,
+                                270
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        257,
+                                        408,
+                                        270
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In Advances in Neural Information Processing Systems (NIPS), 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        }
+                    ],
+                    "index": 13,
+                    "bbox_fs": [
+                        106,
+                        235,
+                        507,
+                        270
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        275,
+                        504,
+                        310
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                275,
+                                506,
+                                289
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        275,
+                                        506,
+                                        289
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[40] Kexin Yi, Jiajun Wu, Chuang Gan, Antonio Torralba, Pushmeet Kohli, and Joshua B. Tenenbaum.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                286,
+                                505,
+                                300
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        286,
+                                        505,
+                                        300
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Neural-symbolic vqa: Disentangling reasoning from vision and language understanding. In",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                298,
+                                441,
+                                311
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        298,
+                                        441,
+                                        311
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Advances in Neural Information Processing Systems, pages 1039–1050, 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        }
+                    ],
+                    "index": 16,
+                    "bbox_fs": [
+                        105,
+                        275,
+                        506,
+                        311
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        316,
+                        505,
+                        361
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                317,
+                                505,
+                                330
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        317,
+                                        505,
+                                        330
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[41] Kexin Yi, Chuang Gan, Yunzhu Li, Pushmeet Kohli, Jiajun Wu, Antonio Torralba, and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                327,
+                                507,
+                                340
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        327,
+                                        507,
+                                        340
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Joshua B. Tenenbaum. CLEVRER: Collision events for video representation and rea-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                339,
+                                507,
+                                352
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        339,
+                                        507,
+                                        352
+                                    ],
+                                    "score": 1.0,
+                                    "content": "soning. In International Conference on Learning Representations, 2020. URL https:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                350,
+                                347,
+                                362
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        350,
+                                        277,
+                                        362
+                                    ],
+                                    "score": 1.0,
+                                    "content": "//openreview.net/forum?id",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        277,
+                                        351,
+                                        284,
+                                        359
+                                    ],
+                                    "score": 0.57,
+                                    "content": "\\underline { { \\underline { { \\mathbf { \\Pi } } } } } =",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        285,
+                                        350,
+                                        347,
+                                        362
+                                    ],
+                                    "score": 1.0,
+                                    "content": "HkxYzANYDB.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        }
+                    ],
+                    "index": 19.5,
+                    "bbox_fs": [
+                        106,
+                        317,
+                        507,
+                        362
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        368,
+                        506,
+                        424
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                368,
+                                505,
+                                381
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        368,
+                                        505,
+                                        381
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[42] Yang You, Jing Li, Sashank Reddi, Jonathan Hseu, Sanjiv Kumar, Srinadh Bhojanapalli, Xiaodan",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                378,
+                                505,
+                                393
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        378,
+                                        505,
+                                        393
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Song, James Demmel, and Cho-Jui Hsieh. Large batch optimization for deep learning: Training",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                389,
+                                505,
+                                404
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        389,
+                                        505,
+                                        404
+                                    ],
+                                    "score": 1.0,
+                                    "content": "BERT in 76 minutes. Technical Report UCB/EECS-2019-103, EECS Department, University",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                400,
+                                505,
+                                414
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        400,
+                                        505,
+                                        414
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of California, Berkeley, Jun 2019. URL http://www2.eecs.berkeley.edu/Pubs/",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                128,
+                                412,
+                                325,
+                                424
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        128,
+                                        412,
+                                        325,
+                                        424
+                                    ],
+                                    "score": 1.0,
+                                    "content": "TechRpts/2019/EECS-2019-103.html.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        }
+                    ],
+                    "index": 24,
+                    "bbox_fs": [
+                        105,
+                        368,
+                        505,
+                        424
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        430,
+                        505,
+                        486
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                431,
+                                506,
+                                443
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        431,
+                                        506,
+                                        443
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[43] Vinicius Zambaldi, David Raposo, Adam Santoro, Victor Bapst, Yujia Li, Igor Babuschkin,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                441,
+                                506,
+                                455
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        441,
+                                        506,
+                                        455
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Karl Tuyls, David Reichert, Timothy Lillicrap, Edward Lockhart, Murray Shanahan, Victoria",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                451,
+                                506,
+                                468
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        451,
+                                        506,
+                                        468
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Langston, Razvan Pascanu, Matthew Botvinick, Oriol Vinyals, and Peter Battaglia. Deep",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                462,
+                                505,
+                                478
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        462,
+                                        505,
+                                        478
+                                    ],
+                                    "score": 1.0,
+                                    "content": "reinforcement learning with relational inductive biases. In International Conference on Learning",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                474,
+                                501,
+                                488
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        474,
+                                        430,
+                                        488
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Representations, 2019. URL https://openreview.net/forum?id",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        431,
+                                        476,
+                                        438,
+                                        484
+                                    ],
+                                    "score": 0.65,
+                                    "content": "\\underline { { \\underline { { \\mathbf { \\Pi } } } } } =",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        438,
+                                        474,
+                                        501,
+                                        488
+                                    ],
+                                    "score": 1.0,
+                                    "content": "HkxaFoC9KQ.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        }
+                    ],
+                    "index": 29,
+                    "bbox_fs": [
+                        106,
+                        431,
+                        506,
+                        488
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        104,
+                        493,
+                        504,
+                        527
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                493,
+                                506,
+                                506
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        493,
+                                        506,
+                                        506
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[44] Chi Zhang, Baoxiong Jia, Mark Edmonds, Song-Chun Zhu, and Yixin Zhu. Acre: Abstract",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                505,
+                                505,
+                                517
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        505,
+                                        505,
+                                        517
+                                    ],
+                                    "score": 1.0,
+                                    "content": "causal reasoning beyond covariation. In Proceedings of the IEEE Conference on Computer",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                514,
+                                315,
+                                528
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        514,
+                                        315,
+                                        528
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Vision and Pattern Recognition (CVPR), 2021.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        }
+                    ],
+                    "index": 33,
+                    "bbox_fs": [
+                        106,
+                        493,
+                        506,
+                        528
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        534,
+                        506,
+                        579
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                534,
+                                506,
+                                547
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        534,
+                                        506,
+                                        547
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[45] Honglu Zhou, Asim Kadav, Farley Lai, Alexandru Niculescu-Mizil, Martin Renqiang Min,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                545,
+                                506,
+                                558
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        545,
+                                        506,
+                                        558
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Mubbasir Kapadia, and Hans Peter Graf. Hopper: Multi-hop transformer for spatiotemporal",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                556,
+                                507,
+                                569
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        556,
+                                        507,
+                                        569
+                                    ],
+                                    "score": 1.0,
+                                    "content": "reasoning. In International Conference on Learning Representations, 2021. URL https:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                568,
+                                289,
+                                579
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        568,
+                                        289,
+                                        579
+                                    ],
+                                    "score": 1.0,
+                                    "content": "//arxiv.org/abs/2103.10574.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        }
+                    ],
+                    "index": 36.5,
+                    "bbox_fs": [
+                        106,
+                        534,
+                        507,
+                        579
+                    ]
+                }
+            ]
+        }
+    ],
+    "_backend": "pipeline",
+    "_version_name": "2.2.2"
+}
\ No newline at end of file
diff --git a/parse/train/lHmhW2zmVN/lHmhW2zmVN_model.json b/parse/train/lHmhW2zmVN/lHmhW2zmVN_model.json
new file mode 100644
index 0000000000000000000000000000000000000000..c632160de6803e7dfc62ef4e7421d23218ca02ec
--- /dev/null
+++ b/parse/train/lHmhW2zmVN/lHmhW2zmVN_model.json
@@ -0,0 +1,16658 @@
+[
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    398,
+                    768,
+                    1305,
+                    768,
+                    1305,
+                    1283,
+                    398,
+                    1283
+                ],
+                "score": 0.983
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1407,
+                    1404,
+                    1407,
+                    1404,
+                    1712,
+                    298,
+                    1712
+                ],
+                "score": 0.981
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1725,
+                    1404,
+                    1725,
+                    1404,
+                    1909,
+                    298,
+                    1909
+                ],
+                "score": 0.971
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    338,
+                    272,
+                    1364,
+                    272,
+                    1364,
+                    383,
+                    338,
+                    383
+                ],
+                "score": 0.96
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    299,
+                    1338,
+                    530,
+                    1338,
+                    530,
+                    1376,
+                    299,
+                    1376
+                ],
+                "score": 0.903
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    787,
+                    701,
+                    913,
+                    701,
+                    913,
+                    737,
+                    787,
+                    737
+                ],
+                "score": 0.894
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    299,
+                    1928,
+                    1401,
+                    1928,
+                    1401,
+                    1984,
+                    299,
+                    1984
+                ],
+                "score": 0.889
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    297,
+                    2033,
+                    1071,
+                    2033,
+                    1071,
+                    2062,
+                    297,
+                    2062
+                ],
+                "score": 0.882
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    310,
+                    497,
+                    1398,
+                    497,
+                    1398,
+                    623,
+                    310,
+                    623
+                ],
+                "score": 0.57
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    328,
+                    216,
+                    1366,
+                    216,
+                    1366,
+                    233,
+                    328,
+                    233
+                ],
+                "score": 0.097
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    333.0,
+                    269.0,
+                    1364.0,
+                    269.0,
+                    1364.0,
+                    332.0,
+                    333.0,
+                    332.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    585.0,
+                    325.0,
+                    1116.0,
+                    325.0,
+                    1116.0,
+                    391.0,
+                    585.0,
+                    391.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1335.0,
+                    536.0,
+                    1335.0,
+                    536.0,
+                    1383.0,
+                    292.0,
+                    1383.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    782.0,
+                    698.0,
+                    920.0,
+                    698.0,
+                    920.0,
+                    742.0,
+                    782.0,
+                    742.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    339.0,
+                    1924.0,
+                    1405.0,
+                    1924.0,
+                    1405.0,
+                    1960.0,
+                    339.0,
+                    1960.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1954.0,
+                    759.0,
+                    1954.0,
+                    759.0,
+                    1988.0,
+                    295.0,
+                    1988.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    2028.0,
+                    1073.0,
+                    2028.0,
+                    1073.0,
+                    2066.0,
+                    294.0,
+                    2066.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    394.0,
+                    769.0,
+                    1304.0,
+                    769.0,
+                    1304.0,
+                    802.0,
+                    394.0,
+                    802.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    393.0,
+                    798.0,
+                    1305.0,
+                    798.0,
+                    1305.0,
+                    834.0,
+                    393.0,
+                    834.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    394.0,
+                    828.0,
+                    1308.0,
+                    828.0,
+                    1308.0,
+                    866.0,
+                    394.0,
+                    866.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    393.0,
+                    859.0,
+                    1306.0,
+                    859.0,
+                    1306.0,
+                    896.0,
+                    393.0,
+                    896.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    393.0,
+                    890.0,
+                    1308.0,
+                    890.0,
+                    1308.0,
+                    925.0,
+                    393.0,
+                    925.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    394.0,
+                    922.0,
+                    1308.0,
+                    922.0,
+                    1308.0,
+                    955.0,
+                    394.0,
+                    955.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    393.0,
+                    952.0,
+                    1306.0,
+                    952.0,
+                    1306.0,
+                    985.0,
+                    393.0,
+                    985.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    393.0,
+                    982.0,
+                    1306.0,
+                    982.0,
+                    1306.0,
+                    1018.0,
+                    393.0,
+                    1018.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    394.0,
+                    1012.0,
+                    1305.0,
+                    1012.0,
+                    1305.0,
+                    1047.0,
+                    394.0,
+                    1047.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    393.0,
+                    1040.0,
+                    1305.0,
+                    1040.0,
+                    1305.0,
+                    1077.0,
+                    393.0,
+                    1077.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    392.0,
+                    1067.0,
+                    1308.0,
+                    1067.0,
+                    1308.0,
+                    1111.0,
+                    392.0,
+                    1111.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    393.0,
+                    1102.0,
+                    1306.0,
+                    1102.0,
+                    1306.0,
+                    1139.0,
+                    393.0,
+                    1139.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    393.0,
+                    1132.0,
+                    1306.0,
+                    1132.0,
+                    1306.0,
+                    1168.0,
+                    393.0,
+                    1168.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    394.0,
+                    1162.0,
+                    1305.0,
+                    1162.0,
+                    1305.0,
+                    1198.0,
+                    394.0,
+                    1198.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    392.0,
+                    1191.0,
+                    1308.0,
+                    1191.0,
+                    1308.0,
+                    1232.0,
+                    392.0,
+                    1232.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    393.0,
+                    1222.0,
+                    1305.0,
+                    1222.0,
+                    1305.0,
+                    1258.0,
+                    393.0,
+                    1258.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    395.0,
+                    1255.0,
+                    812.0,
+                    1255.0,
+                    812.0,
+                    1285.0,
+                    395.0,
+                    1285.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1406.0,
+                    1406.0,
+                    1406.0,
+                    1406.0,
+                    1441.0,
+                    293.0,
+                    1441.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1435.0,
+                    1406.0,
+                    1435.0,
+                    1406.0,
+                    1473.0,
+                    293.0,
+                    1473.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1470.0,
+                    1405.0,
+                    1470.0,
+                    1405.0,
+                    1501.0,
+                    296.0,
+                    1501.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1500.0,
+                    1404.0,
+                    1500.0,
+                    1404.0,
+                    1532.0,
+                    296.0,
+                    1532.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1527.0,
+                    1406.0,
+                    1527.0,
+                    1406.0,
+                    1565.0,
+                    293.0,
+                    1565.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1560.0,
+                    1406.0,
+                    1560.0,
+                    1406.0,
+                    1594.0,
+                    293.0,
+                    1594.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1587.0,
+                    1408.0,
+                    1587.0,
+                    1408.0,
+                    1628.0,
+                    291.0,
+                    1628.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1618.0,
+                    1409.0,
+                    1618.0,
+                    1409.0,
+                    1654.0,
+                    293.0,
+                    1654.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1648.0,
+                    1406.0,
+                    1648.0,
+                    1406.0,
+                    1686.0,
+                    292.0,
+                    1686.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1676.0,
+                    904.0,
+                    1676.0,
+                    904.0,
+                    1719.0,
+                    293.0,
+                    1719.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1723.0,
+                    1405.0,
+                    1723.0,
+                    1405.0,
+                    1759.0,
+                    293.0,
+                    1759.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1755.0,
+                    1407.0,
+                    1755.0,
+                    1407.0,
+                    1792.0,
+                    291.0,
+                    1792.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1785.0,
+                    1406.0,
+                    1785.0,
+                    1406.0,
+                    1821.0,
+                    293.0,
+                    1821.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1813.0,
+                    1406.0,
+                    1813.0,
+                    1406.0,
+                    1853.0,
+                    293.0,
+                    1853.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1849.0,
+                    1405.0,
+                    1849.0,
+                    1405.0,
+                    1880.0,
+                    294.0,
+                    1880.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1875.0,
+                    1404.0,
+                    1875.0,
+                    1404.0,
+                    1913.0,
+                    292.0,
+                    1913.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    317.0,
+                    492.0,
+                    467.0,
+                    492.0,
+                    467.0,
+                    537.0,
+                    317.0,
+                    537.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    512.0,
+                    496.0,
+                    636.0,
+                    496.0,
+                    636.0,
+                    531.0,
+                    512.0,
+                    531.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    686.0,
+                    498.0,
+                    871.0,
+                    498.0,
+                    871.0,
+                    532.0,
+                    686.0,
+                    532.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    920.0,
+                    498.0,
+                    1153.0,
+                    498.0,
+                    1153.0,
+                    532.0,
+                    920.0,
+                    532.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1203.0,
+                    499.0,
+                    1391.0,
+                    499.0,
+                    1391.0,
+                    529.0,
+                    1203.0,
+                    529.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    790.0,
+                    526.0,
+                    923.0,
+                    526.0,
+                    923.0,
+                    564.0,
+                    790.0,
+                    564.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    707.0,
+                    560.0,
+                    1002.0,
+                    560.0,
+                    1002.0,
+                    594.0,
+                    707.0,
+                    594.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    313.0,
+                    588.0,
+                    1402.0,
+                    588.0,
+                    1402.0,
+                    626.0,
+                    313.0,
+                    626.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 0,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    1687,
+                    1404,
+                    1687,
+                    1404,
+                    1900,
+                    299,
+                    1900
+                ],
+                "score": 0.976
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    202,
+                    1405,
+                    202,
+                    1405,
+                    476,
+                    298,
+                    476
+                ],
+                "score": 0.975
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    300,
+                    1451,
+                    1405,
+                    1451,
+                    1405,
+                    1574,
+                    300,
+                    1574
+                ],
+                "score": 0.97
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    300,
+                    1914,
+                    1404,
+                    1914,
+                    1404,
+                    2007,
+                    300,
+                    2007
+                ],
+                "score": 0.965
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    301,
+                    491,
+                    1405,
+                    491,
+                    1405,
+                    583,
+                    301,
+                    583
+                ],
+                "score": 0.962
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    305,
+                    812,
+                    1400,
+                    812,
+                    1400,
+                    903,
+                    305,
+                    903
+                ],
+                "score": 0.956
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    303,
+                    917,
+                    1402,
+                    917,
+                    1402,
+                    1011,
+                    303,
+                    1011
+                ],
+                "score": 0.954
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    330,
+                    608,
+                    1403,
+                    608,
+                    1403,
+                    787,
+                    330,
+                    787
+                ],
+                "score": 0.936
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    329,
+                    1036,
+                    1406,
+                    1036,
+                    1406,
+                    1425,
+                    329,
+                    1425
+                ],
+                "score": 0.927
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    299,
+                    1617,
+                    475,
+                    1617,
+                    475,
+                    1654,
+                    299,
+                    1654
+                ],
+                "score": 0.882
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    841,
+                    2062,
+                    858,
+                    2062,
+                    858,
+                    2084,
+                    841,
+                    2084
+                ],
+                "score": 0.726
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    841,
+                    2062,
+                    859,
+                    2062,
+                    859,
+                    2084,
+                    841,
+                    2084
+                ],
+                "score": 0.103
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    667,
+                    1395,
+                    745,
+                    1395,
+                    745,
+                    1424,
+                    667,
+                    1424
+                ],
+                "score": 0.89,
+                "latex": "{ \\bf0 . 0 0 \\% }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    529,
+                    1395,
+                    621,
+                    1395,
+                    621,
+                    1424,
+                    529,
+                    1424
+                ],
+                "score": 0.89,
+                "latex": "9 8 . 9 7 \\%"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    474,
+                    1364,
+                    567,
+                    1364,
+                    567,
+                    1393,
+                    474,
+                    1393
+                ],
+                "score": 0.88,
+                "latex": "9 4 . 4 8 \\%"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    779,
+                    1127,
+                    837,
+                    1127,
+                    837,
+                    1156,
+                    779,
+                    1156
+                ],
+                "score": 0.87,
+                "latex": "75 \\%"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    612,
+                    1364,
+                    703,
+                    1364,
+                    703,
+                    1393,
+                    612,
+                    1393
+                ],
+                "score": 0.86,
+                "latex": "1 6 . 0 6 \\%"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    996,
+                    1303,
+                    1055,
+                    1303,
+                    1055,
+                    1333,
+                    996,
+                    1333
+                ],
+                "score": 0.86,
+                "latex": "67 \\%"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1205,
+                    1512,
+                    1260,
+                    1512,
+                    1260,
+                    1541,
+                    1205,
+                    1541
+                ],
+                "score": 0.85,
+                "latex": "40 \\%"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    873,
+                    1127,
+                    931,
+                    1127,
+                    931,
+                    1156,
+                    873,
+                    1156
+                ],
+                "score": 0.85,
+                "latex": "46 \\%"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    852,
+                    1304,
+                    911,
+                    1304,
+                    911,
+                    1333,
+                    852,
+                    1333
+                ],
+                "score": 0.85,
+                "latex": "94 \\%"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1611.0,
+                    479.0,
+                    1611.0,
+                    479.0,
+                    1662.0,
+                    291.0,
+                    1662.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    839.0,
+                    2059.0,
+                    862.0,
+                    2059.0,
+                    862.0,
+                    2093.0,
+                    839.0,
+                    2093.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    838.0,
+                    2059.0,
+                    862.0,
+                    2059.0,
+                    862.0,
+                    2093.0,
+                    838.0,
+                    2093.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1686.0,
+                    1404.0,
+                    1686.0,
+                    1404.0,
+                    1723.0,
+                    294.0,
+                    1723.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1718.0,
+                    1404.0,
+                    1718.0,
+                    1404.0,
+                    1752.0,
+                    294.0,
+                    1752.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1747.0,
+                    1405.0,
+                    1747.0,
+                    1405.0,
+                    1783.0,
+                    294.0,
+                    1783.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1779.0,
+                    1406.0,
+                    1779.0,
+                    1406.0,
+                    1813.0,
+                    294.0,
+                    1813.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1804.0,
+                    1409.0,
+                    1804.0,
+                    1409.0,
+                    1846.0,
+                    292.0,
+                    1846.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1838.0,
+                    1408.0,
+                    1838.0,
+                    1408.0,
+                    1877.0,
+                    293.0,
+                    1877.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    1872.0,
+                    526.0,
+                    1872.0,
+                    526.0,
+                    1904.0,
+                    297.0,
+                    1904.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    203.0,
+                    1406.0,
+                    203.0,
+                    1406.0,
+                    239.0,
+                    295.0,
+                    239.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    235.0,
+                    1407.0,
+                    235.0,
+                    1407.0,
+                    268.0,
+                    296.0,
+                    268.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    265.0,
+                    1405.0,
+                    265.0,
+                    1405.0,
+                    297.0,
+                    295.0,
+                    297.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    294.0,
+                    1406.0,
+                    294.0,
+                    1406.0,
+                    330.0,
+                    293.0,
+                    330.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    325.0,
+                    1406.0,
+                    325.0,
+                    1406.0,
+                    361.0,
+                    292.0,
+                    361.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    356.0,
+                    1407.0,
+                    356.0,
+                    1407.0,
+                    388.0,
+                    296.0,
+                    388.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    382.0,
+                    1407.0,
+                    382.0,
+                    1407.0,
+                    421.0,
+                    293.0,
+                    421.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    417.0,
+                    1403.0,
+                    417.0,
+                    1403.0,
+                    449.0,
+                    296.0,
+                    449.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    447.0,
+                    552.0,
+                    447.0,
+                    552.0,
+                    479.0,
+                    295.0,
+                    479.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1452.0,
+                    1404.0,
+                    1452.0,
+                    1404.0,
+                    1485.0,
+                    295.0,
+                    1485.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1482.0,
+                    1407.0,
+                    1482.0,
+                    1407.0,
+                    1515.0,
+                    295.0,
+                    1515.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1508.0,
+                    1204.0,
+                    1508.0,
+                    1204.0,
+                    1549.0,
+                    292.0,
+                    1549.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1261.0,
+                    1508.0,
+                    1406.0,
+                    1508.0,
+                    1406.0,
+                    1549.0,
+                    1261.0,
+                    1549.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1542.0,
+                    1213.0,
+                    1542.0,
+                    1213.0,
+                    1576.0,
+                    294.0,
+                    1576.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1915.0,
+                    1405.0,
+                    1915.0,
+                    1405.0,
+                    1948.0,
+                    295.0,
+                    1948.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1945.0,
+                    1409.0,
+                    1945.0,
+                    1409.0,
+                    1981.0,
+                    294.0,
+                    1981.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1973.0,
+                    1409.0,
+                    1973.0,
+                    1409.0,
+                    2014.0,
+                    294.0,
+                    2014.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    485.0,
+                    1405.0,
+                    485.0,
+                    1405.0,
+                    531.0,
+                    292.0,
+                    531.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    521.0,
+                    1402.0,
+                    521.0,
+                    1402.0,
+                    555.0,
+                    296.0,
+                    555.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    553.0,
+                    868.0,
+                    553.0,
+                    868.0,
+                    586.0,
+                    296.0,
+                    586.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    811.0,
+                    1405.0,
+                    811.0,
+                    1405.0,
+                    848.0,
+                    297.0,
+                    848.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    841.0,
+                    1406.0,
+                    841.0,
+                    1406.0,
+                    879.0,
+                    296.0,
+                    879.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    301.0,
+                    873.0,
+                    501.0,
+                    873.0,
+                    501.0,
+                    903.0,
+                    301.0,
+                    903.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    917.0,
+                    1405.0,
+                    917.0,
+                    1405.0,
+                    954.0,
+                    297.0,
+                    954.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    947.0,
+                    1406.0,
+                    947.0,
+                    1406.0,
+                    983.0,
+                    296.0,
+                    983.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    298.0,
+                    979.0,
+                    1266.0,
+                    979.0,
+                    1266.0,
+                    1012.0,
+                    298.0,
+                    1012.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    330.0,
+                    607.0,
+                    1022.0,
+                    607.0,
+                    1022.0,
+                    643.0,
+                    330.0,
+                    643.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    329.0,
+                    648.0,
+                    1404.0,
+                    648.0,
+                    1404.0,
+                    685.0,
+                    329.0,
+                    685.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    357.0,
+                    681.0,
+                    1276.0,
+                    681.0,
+                    1276.0,
+                    716.0,
+                    357.0,
+                    716.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    346.0,
+                    722.0,
+                    1403.0,
+                    722.0,
+                    1403.0,
+                    760.0,
+                    346.0,
+                    760.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    357.0,
+                    752.0,
+                    890.0,
+                    752.0,
+                    890.0,
+                    790.0,
+                    357.0,
+                    790.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    344.0,
+                    1034.0,
+                    1403.0,
+                    1034.0,
+                    1403.0,
+                    1071.0,
+                    344.0,
+                    1071.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    357.0,
+                    1066.0,
+                    1406.0,
+                    1066.0,
+                    1406.0,
+                    1102.0,
+                    357.0,
+                    1102.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    359.0,
+                    1097.0,
+                    1406.0,
+                    1097.0,
+                    1406.0,
+                    1132.0,
+                    359.0,
+                    1132.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    359.0,
+                    1129.0,
+                    778.0,
+                    1129.0,
+                    778.0,
+                    1160.0,
+                    359.0,
+                    1160.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    838.0,
+                    1129.0,
+                    872.0,
+                    1129.0,
+                    872.0,
+                    1160.0,
+                    838.0,
+                    1160.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    932.0,
+                    1129.0,
+                    1379.0,
+                    1129.0,
+                    1379.0,
+                    1160.0,
+                    932.0,
+                    1160.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    344.0,
+                    1169.0,
+                    1406.0,
+                    1169.0,
+                    1406.0,
+                    1205.0,
+                    344.0,
+                    1205.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    355.0,
+                    1194.0,
+                    1409.0,
+                    1194.0,
+                    1409.0,
+                    1239.0,
+                    355.0,
+                    1239.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    357.0,
+                    1230.0,
+                    1381.0,
+                    1230.0,
+                    1381.0,
+                    1266.0,
+                    357.0,
+                    1266.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    353.0,
+                    1273.0,
+                    1405.0,
+                    1273.0,
+                    1405.0,
+                    1308.0,
+                    353.0,
+                    1308.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    356.0,
+                    1302.0,
+                    851.0,
+                    1302.0,
+                    851.0,
+                    1340.0,
+                    356.0,
+                    1340.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    912.0,
+                    1302.0,
+                    995.0,
+                    1302.0,
+                    995.0,
+                    1340.0,
+                    912.0,
+                    1340.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1056.0,
+                    1302.0,
+                    1406.0,
+                    1302.0,
+                    1406.0,
+                    1340.0,
+                    1056.0,
+                    1340.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    355.0,
+                    1330.0,
+                    1406.0,
+                    1330.0,
+                    1406.0,
+                    1372.0,
+                    355.0,
+                    1372.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    356.0,
+                    1362.0,
+                    473.0,
+                    1362.0,
+                    473.0,
+                    1399.0,
+                    356.0,
+                    1399.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    568.0,
+                    1362.0,
+                    611.0,
+                    1362.0,
+                    611.0,
+                    1399.0,
+                    568.0,
+                    1399.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    704.0,
+                    1362.0,
+                    1408.0,
+                    1362.0,
+                    1408.0,
+                    1399.0,
+                    704.0,
+                    1399.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    359.0,
+                    1393.0,
+                    528.0,
+                    1393.0,
+                    528.0,
+                    1428.0,
+                    359.0,
+                    1428.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    622.0,
+                    1393.0,
+                    666.0,
+                    1393.0,
+                    666.0,
+                    1428.0,
+                    622.0,
+                    1428.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    746.0,
+                    1393.0,
+                    1059.0,
+                    1393.0,
+                    1059.0,
+                    1428.0,
+                    746.0,
+                    1428.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 1,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    914,
+                    1405,
+                    914,
+                    1405,
+                    1219,
+                    298,
+                    1219
+                ],
+                "score": 0.982
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1626,
+                    1404,
+                    1626,
+                    1404,
+                    1810,
+                    298,
+                    1810
+                ],
+                "score": 0.978
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1399,
+                    1404,
+                    1399,
+                    1404,
+                    1613,
+                    298,
+                    1613
+                ],
+                "score": 0.977
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    1823,
+                    1403,
+                    1823,
+                    1403,
+                    2007,
+                    299,
+                    2007
+                ],
+                "score": 0.977
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1232,
+                    1404,
+                    1232,
+                    1404,
+                    1386,
+                    298,
+                    1386
+                ],
+                "score": 0.974
+            },
+            {
+                "category_id": 3,
+                "poly": [
+                    429,
+                    207,
+                    1297,
+                    207,
+                    1297,
+                    662,
+                    429,
+                    662
+                ],
+                "score": 0.972
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    300,
+                    807,
+                    1404,
+                    807,
+                    1404,
+                    901,
+                    300,
+                    901
+                ],
+                "score": 0.963
+            },
+            {
+                "category_id": 4,
+                "poly": [
+                    414,
+                    677,
+                    1286,
+                    677,
+                    1286,
+                    710,
+                    414,
+                    710
+                ],
+                "score": 0.923
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    841,
+                    2061,
+                    858,
+                    2061,
+                    858,
+                    2084,
+                    841,
+                    2084
+                ],
+                "score": 0.646
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    841,
+                    2061,
+                    859,
+                    2061,
+                    859,
+                    2084,
+                    841,
+                    2084
+                ],
+                "score": 0.317
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    761,
+                    1322,
+                    865,
+                    1322,
+                    865,
+                    1355,
+                    761,
+                    1355
+                ],
+                "score": 0.92,
+                "latex": "\\mu _ { t i } \\in \\mathbb { R } ^ { d }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    770,
+                    1655,
+                    858,
+                    1655,
+                    858,
+                    1686,
+                    770,
+                    1686
+                ],
+                "score": 0.92,
+                "latex": "\\mathbb { R } ^ { N _ { H } \\times d }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1072,
+                    1428,
+                    1206,
+                    1428,
+                    1206,
+                    1458,
+                    1072,
+                    1458
+                ],
+                "score": 0.91,
+                "latex": "\\mathbf { \\bar { \\mathit { C L S } } } \\in \\mathbf { \\bar { \\mathbb { R } } ^ { d } }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    963,
+                    1628,
+                    1005,
+                    1628,
+                    1005,
+                    1657,
+                    963,
+                    1657
+                ],
+                "score": 0.89,
+                "latex": "N _ { T }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    943,
+                    1659,
+                    989,
+                    1659,
+                    989,
+                    1688,
+                    943,
+                    1688
+                ],
+                "score": 0.89,
+                "latex": "N _ { H }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    458,
+                    1264,
+                    495,
+                    1264,
+                    495,
+                    1294,
+                    458,
+                    1294
+                ],
+                "score": 0.88,
+                "latex": "N _ { o }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    368,
+                    1494,
+                    404,
+                    1494,
+                    404,
+                    1521,
+                    368,
+                    1521
+                ],
+                "score": 0.88,
+                "latex": "\\mathbf { w } _ { i }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1244,
+                    1524,
+                    1280,
+                    1524,
+                    1280,
+                    1551,
+                    1244,
+                    1551
+                ],
+                "score": 0.87,
+                "latex": "\\mathbf { w } _ { i }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    782,
+                    1264,
+                    819,
+                    1264,
+                    819,
+                    1294,
+                    782,
+                    1294
+                ],
+                "score": 0.87,
+                "latex": "\\mathrm { \\Delta } N _ { o }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    649,
+                    1434,
+                    687,
+                    1434,
+                    687,
+                    1462,
+                    649,
+                    1462
+                ],
+                "score": 0.87,
+                "latex": "\\mu _ { t i }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    297,
+                    1427,
+                    333,
+                    1427,
+                    333,
+                    1457,
+                    297,
+                    1457
+                ],
+                "score": 0.87,
+                "latex": "\\mathbb { R } ^ { d }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1157,
+                    1525,
+                    1194,
+                    1525,
+                    1194,
+                    1552,
+                    1157,
+                    1552
+                ],
+                "score": 0.87,
+                "latex": "\\mu _ { t i }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    585,
+                    1525,
+                    620,
+                    1525,
+                    620,
+                    1551,
+                    585,
+                    1551
+                ],
+                "score": 0.87,
+                "latex": "\\mathbf { w } _ { i }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    698,
+                    1749,
+                    744,
+                    1749,
+                    744,
+                    1779,
+                    698,
+                    1779
+                ],
+                "score": 0.86,
+                "latex": "N _ { H }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    297,
+                    1525,
+                    335,
+                    1525,
+                    335,
+                    1553,
+                    297,
+                    1553
+                ],
+                "score": 0.84,
+                "latex": "\\mu _ { t i }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1100,
+                    1718,
+                    1162,
+                    1718,
+                    1162,
+                    1746,
+                    1100,
+                    1746
+                ],
+                "score": 0.78,
+                "latex": "C L S"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    951,
+                    1326,
+                    964,
+                    1326,
+                    964,
+                    1351,
+                    951,
+                    1351
+                ],
+                "score": 0.75,
+                "latex": "i"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    770,
+                    1432,
+                    785,
+                    1432,
+                    785,
+                    1457,
+                    770,
+                    1457
+                ],
+                "score": 0.74,
+                "latex": "t"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    838,
+                    1432,
+                    851,
+                    1432,
+                    851,
+                    1457,
+                    838,
+                    1457
+                ],
+                "score": 0.68,
+                "latex": "i"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1271,
+                    1326,
+                    1285,
+                    1326,
+                    1285,
+                    1351,
+                    1271,
+                    1351
+                ],
+                "score": 0.67,
+                "latex": "t"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    476.0,
+                    204.0,
+                    739.0,
+                    204.0,
+                    739.0,
+                    243.0,
+                    476.0,
+                    243.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1052.0,
+                    204.0,
+                    1267.0,
+                    204.0,
+                    1267.0,
+                    243.0,
+                    1052.0,
+                    243.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    559.0,
+                    247.0,
+                    565.0,
+                    247.0,
+                    565.0,
+                    252.0,
+                    559.0,
+                    252.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    590.0,
+                    304.0,
+                    653.0,
+                    304.0,
+                    653.0,
+                    331.0,
+                    590.0,
+                    331.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1097.0,
+                    304.0,
+                    1246.0,
+                    304.0,
+                    1246.0,
+                    330.0,
+                    1097.0,
+                    330.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    646.0,
+                    403.0,
+                    1063.0,
+                    403.0,
+                    1063.0,
+                    446.0,
+                    646.0,
+                    446.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    463.0,
+                    491.0,
+                    483.0,
+                    491.0,
+                    483.0,
+                    501.0,
+                    463.0,
+                    501.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    529.0,
+                    522.0,
+                    613.0,
+                    522.0,
+                    613.0,
+                    556.0,
+                    529.0,
+                    556.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    947.0,
+                    522.0,
+                    1142.0,
+                    522.0,
+                    1142.0,
+                    557.0,
+                    947.0,
+                    557.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    784.0,
+                    604.0,
+                    1281.0,
+                    604.0,
+                    1281.0,
+                    634.0,
+                    784.0,
+                    634.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1138.0,
+                    240.5,
+                    1188.0,
+                    240.5,
+                    1188.0,
+                    270.0,
+                    1138.0,
+                    270.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    409.0,
+                    675.0,
+                    1288.0,
+                    675.0,
+                    1288.0,
+                    712.0,
+                    409.0,
+                    712.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    838.0,
+                    2058.0,
+                    861.0,
+                    2058.0,
+                    861.0,
+                    2091.0,
+                    838.0,
+                    2091.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    839.0,
+                    2059.0,
+                    860.0,
+                    2059.0,
+                    860.0,
+                    2090.0,
+                    839.0,
+                    2090.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    915.0,
+                    1406.0,
+                    915.0,
+                    1406.0,
+                    950.0,
+                    295.0,
+                    950.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    944.0,
+                    1408.0,
+                    944.0,
+                    1408.0,
+                    979.0,
+                    295.0,
+                    979.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    977.0,
+                    1405.0,
+                    977.0,
+                    1405.0,
+                    1008.0,
+                    295.0,
+                    1008.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1006.0,
+                    1407.0,
+                    1006.0,
+                    1407.0,
+                    1041.0,
+                    295.0,
+                    1041.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1036.0,
+                    1408.0,
+                    1036.0,
+                    1408.0,
+                    1072.0,
+                    295.0,
+                    1072.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1067.0,
+                    1407.0,
+                    1067.0,
+                    1407.0,
+                    1101.0,
+                    292.0,
+                    1101.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1098.0,
+                    1406.0,
+                    1098.0,
+                    1406.0,
+                    1130.0,
+                    296.0,
+                    1130.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1126.0,
+                    1403.0,
+                    1126.0,
+                    1403.0,
+                    1160.0,
+                    295.0,
+                    1160.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1158.0,
+                    1405.0,
+                    1158.0,
+                    1405.0,
+                    1193.0,
+                    295.0,
+                    1193.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1187.0,
+                    1030.0,
+                    1187.0,
+                    1030.0,
+                    1221.0,
+                    292.0,
+                    1221.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1625.0,
+                    962.0,
+                    1625.0,
+                    962.0,
+                    1661.0,
+                    294.0,
+                    1661.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1006.0,
+                    1625.0,
+                    1406.0,
+                    1625.0,
+                    1406.0,
+                    1661.0,
+                    1006.0,
+                    1661.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1651.0,
+                    769.0,
+                    1651.0,
+                    769.0,
+                    1694.0,
+                    291.0,
+                    1694.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    859.0,
+                    1651.0,
+                    942.0,
+                    1651.0,
+                    942.0,
+                    1694.0,
+                    859.0,
+                    1694.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    990.0,
+                    1651.0,
+                    1410.0,
+                    1651.0,
+                    1410.0,
+                    1694.0,
+                    990.0,
+                    1694.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1686.0,
+                    1406.0,
+                    1686.0,
+                    1406.0,
+                    1722.0,
+                    293.0,
+                    1722.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1713.0,
+                    1099.0,
+                    1713.0,
+                    1099.0,
+                    1755.0,
+                    291.0,
+                    1755.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1163.0,
+                    1713.0,
+                    1405.0,
+                    1713.0,
+                    1405.0,
+                    1755.0,
+                    1163.0,
+                    1755.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1746.0,
+                    697.0,
+                    1746.0,
+                    697.0,
+                    1784.0,
+                    292.0,
+                    1784.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    745.0,
+                    1746.0,
+                    1405.0,
+                    1746.0,
+                    1405.0,
+                    1784.0,
+                    745.0,
+                    1784.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1778.0,
+                    538.0,
+                    1778.0,
+                    538.0,
+                    1813.0,
+                    292.0,
+                    1813.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1397.0,
+                    1404.0,
+                    1397.0,
+                    1404.0,
+                    1435.0,
+                    293.0,
+                    1435.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1425.0,
+                    296.0,
+                    1425.0,
+                    296.0,
+                    1467.0,
+                    293.0,
+                    1467.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    334.0,
+                    1425.0,
+                    648.0,
+                    1425.0,
+                    648.0,
+                    1467.0,
+                    334.0,
+                    1467.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    688.0,
+                    1425.0,
+                    769.0,
+                    1425.0,
+                    769.0,
+                    1467.0,
+                    688.0,
+                    1467.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    786.0,
+                    1425.0,
+                    837.0,
+                    1425.0,
+                    837.0,
+                    1467.0,
+                    786.0,
+                    1467.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    852.0,
+                    1425.0,
+                    1071.0,
+                    1425.0,
+                    1071.0,
+                    1467.0,
+                    852.0,
+                    1467.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1207.0,
+                    1425.0,
+                    1406.0,
+                    1425.0,
+                    1406.0,
+                    1467.0,
+                    1207.0,
+                    1467.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1459.0,
+                    1405.0,
+                    1459.0,
+                    1405.0,
+                    1494.0,
+                    293.0,
+                    1494.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1491.0,
+                    367.0,
+                    1491.0,
+                    367.0,
+                    1526.0,
+                    294.0,
+                    1526.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    405.0,
+                    1491.0,
+                    1402.0,
+                    1491.0,
+                    1402.0,
+                    1526.0,
+                    405.0,
+                    1526.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1520.0,
+                    296.0,
+                    1520.0,
+                    296.0,
+                    1556.0,
+                    292.0,
+                    1556.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    336.0,
+                    1520.0,
+                    584.0,
+                    1520.0,
+                    584.0,
+                    1556.0,
+                    336.0,
+                    1556.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    621.0,
+                    1520.0,
+                    1156.0,
+                    1520.0,
+                    1156.0,
+                    1556.0,
+                    621.0,
+                    1556.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1195.0,
+                    1520.0,
+                    1243.0,
+                    1520.0,
+                    1243.0,
+                    1556.0,
+                    1195.0,
+                    1556.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1281.0,
+                    1520.0,
+                    1406.0,
+                    1520.0,
+                    1406.0,
+                    1556.0,
+                    1281.0,
+                    1556.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1551.0,
+                    1410.0,
+                    1551.0,
+                    1410.0,
+                    1586.0,
+                    294.0,
+                    1586.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1581.0,
+                    1336.0,
+                    1581.0,
+                    1336.0,
+                    1615.0,
+                    294.0,
+                    1615.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1820.0,
+                    1405.0,
+                    1820.0,
+                    1405.0,
+                    1859.0,
+                    293.0,
+                    1859.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1852.0,
+                    1407.0,
+                    1852.0,
+                    1407.0,
+                    1888.0,
+                    294.0,
+                    1888.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1883.0,
+                    1405.0,
+                    1883.0,
+                    1405.0,
+                    1919.0,
+                    295.0,
+                    1919.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1914.0,
+                    1405.0,
+                    1914.0,
+                    1405.0,
+                    1948.0,
+                    294.0,
+                    1948.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1945.0,
+                    1408.0,
+                    1945.0,
+                    1408.0,
+                    1981.0,
+                    295.0,
+                    1981.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1976.0,
+                    702.0,
+                    1976.0,
+                    702.0,
+                    2011.0,
+                    295.0,
+                    2011.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1230.0,
+                    1406.0,
+                    1230.0,
+                    1406.0,
+                    1268.0,
+                    293.0,
+                    1268.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1260.0,
+                    457.0,
+                    1260.0,
+                    457.0,
+                    1298.0,
+                    293.0,
+                    1298.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    496.0,
+                    1260.0,
+                    781.0,
+                    1260.0,
+                    781.0,
+                    1298.0,
+                    496.0,
+                    1298.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    820.0,
+                    1260.0,
+                    1405.0,
+                    1260.0,
+                    1405.0,
+                    1298.0,
+                    820.0,
+                    1298.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1293.0,
+                    1405.0,
+                    1293.0,
+                    1405.0,
+                    1330.0,
+                    294.0,
+                    1330.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1318.0,
+                    760.0,
+                    1318.0,
+                    760.0,
+                    1362.0,
+                    292.0,
+                    1362.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    866.0,
+                    1318.0,
+                    950.0,
+                    1318.0,
+                    950.0,
+                    1362.0,
+                    866.0,
+                    1362.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    965.0,
+                    1318.0,
+                    1270.0,
+                    1318.0,
+                    1270.0,
+                    1362.0,
+                    965.0,
+                    1362.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1286.0,
+                    1318.0,
+                    1409.0,
+                    1318.0,
+                    1409.0,
+                    1362.0,
+                    1286.0,
+                    1362.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1351.0,
+                    1409.0,
+                    1351.0,
+                    1409.0,
+                    1391.0,
+                    293.0,
+                    1391.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    807.0,
+                    1408.0,
+                    807.0,
+                    1408.0,
+                    844.0,
+                    294.0,
+                    844.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    839.0,
+                    1406.0,
+                    839.0,
+                    1406.0,
+                    873.0,
+                    295.0,
+                    873.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    867.0,
+                    1130.0,
+                    867.0,
+                    1130.0,
+                    904.0,
+                    295.0,
+                    904.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 2,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1186,
+                    1405,
+                    1186,
+                    1405,
+                    1371,
+                    297,
+                    1371
+                ],
+                "score": 0.981
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1654,
+                    1404,
+                    1654,
+                    1404,
+                    1842,
+                    297,
+                    1842
+                ],
+                "score": 0.979
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    296,
+                    1383,
+                    1405,
+                    1383,
+                    1405,
+                    1537,
+                    296,
+                    1537
+                ],
+                "score": 0.979
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1853,
+                    1404,
+                    1853,
+                    1404,
+                    2008,
+                    298,
+                    2008
+                ],
+                "score": 0.978
+            },
+            {
+                "category_id": 3,
+                "poly": [
+                    299,
+                    217,
+                    1397,
+                    217,
+                    1397,
+                    717,
+                    299,
+                    717
+                ],
+                "score": 0.97
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    643,
+                    1557,
+                    1055,
+                    1557,
+                    1055,
+                    1631,
+                    643,
+                    1631
+                ],
+                "score": 0.949
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    292,
+                    1026,
+                    1403,
+                    1026,
+                    1403,
+                    1089,
+                    292,
+                    1089
+                ],
+                "score": 0.947
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    299,
+                    1128,
+                    649,
+                    1128,
+                    649,
+                    1162,
+                    299,
+                    1162
+                ],
+                "score": 0.922
+            },
+            {
+                "category_id": 4,
+                "poly": [
+                    391,
+                    749,
+                    1306,
+                    749,
+                    1306,
+                    783,
+                    391,
+                    783
+                ],
+                "score": 0.912
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    841,
+                    2061,
+                    858,
+                    2061,
+                    858,
+                    2084,
+                    841,
+                    2084
+                ],
+                "score": 0.769
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    354,
+                    937,
+                    1115,
+                    937,
+                    1115,
+                    1013,
+                    354,
+                    1013
+                ],
+                "score": 0.723
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    354,
+                    848,
+                    1092,
+                    848,
+                    1092,
+                    879,
+                    354,
+                    879
+                ],
+                "score": 0.459
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    893,
+                    604,
+                    893,
+                    604,
+                    923,
+                    298,
+                    923
+                ],
+                "score": 0.362
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    354,
+                    848,
+                    1092,
+                    848,
+                    1092,
+                    879,
+                    354,
+                    879
+                ],
+                "score": 0.177
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    298,
+                    893,
+                    604,
+                    893,
+                    604,
+                    923,
+                    298,
+                    923
+                ],
+                "score": 0.169
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    354,
+                    937,
+                    1115,
+                    937,
+                    1115,
+                    1013,
+                    354,
+                    1013
+                ],
+                "score": 0.115
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1025,
+                    1657,
+                    1159,
+                    1657,
+                    1159,
+                    1691,
+                    1025,
+                    1691
+                ],
+                "score": 0.94,
+                "latex": "\\tau _ { t i } \\in \\{ 0 , 1 \\}"
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    643,
+                    1556,
+                    1053,
+                    1556,
+                    1053,
+                    1634,
+                    643,
+                    1634
+                ],
+                "score": 0.94,
+                "latex": "{ \\mathrm { a u x i l i a r y ~ l o s s } } = \\sum _ { t , i } { \\tau _ { t i } l \\left( f ( \\mu _ { t i } ^ { \\prime } ) , \\mu \\right) } ,"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1253,
+                    1383,
+                    1398,
+                    1383,
+                    1398,
+                    1418,
+                    1253,
+                    1418
+                ],
+                "score": 0.93,
+                "latex": "m _ { t i } \\in \\{ 0 , 1 \\}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    709,
+                    1419,
+                    788,
+                    1419,
+                    788,
+                    1447,
+                    709,
+                    1447
+                ],
+                "score": 0.9,
+                "latex": "m _ { t i } \\mu _ { t i }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    343,
+                    1414,
+                    381,
+                    1414,
+                    381,
+                    1448,
+                    343,
+                    1448
+                ],
+                "score": 0.9,
+                "latex": "\\mu _ { t i } ^ { \\prime }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    502,
+                    1475,
+                    538,
+                    1475,
+                    538,
+                    1509,
+                    502,
+                    1509
+                ],
+                "score": 0.89,
+                "latex": "\\mu _ { t i } ^ { \\prime }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    984,
+                    1888,
+                    1028,
+                    1888,
+                    1028,
+                    1915,
+                    984,
+                    1915
+                ],
+                "score": 0.88,
+                "latex": "m _ { t i }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    952,
+                    1389,
+                    989,
+                    1389,
+                    989,
+                    1416,
+                    952,
+                    1416
+                ],
+                "score": 0.86,
+                "latex": "\\mu _ { t i }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1296,
+                    1449,
+                    1334,
+                    1449,
+                    1334,
+                    1478,
+                    1296,
+                    1478
+                ],
+                "score": 0.86,
+                "latex": "\\mu _ { t i }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    372,
+                    1659,
+                    392,
+                    1659,
+                    392,
+                    1690,
+                    372,
+                    1690
+                ],
+                "score": 0.86,
+                "latex": "f"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1078,
+                    1888,
+                    1112,
+                    1888,
+                    1112,
+                    1915,
+                    1078,
+                    1915
+                ],
+                "score": 0.85,
+                "latex": "\\tau _ { t i }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    503,
+                    1749,
+                    523,
+                    1749,
+                    523,
+                    1781,
+                    503,
+                    1781
+                ],
+                "score": 0.85,
+                "latex": "f"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    413,
+                    942,
+                    440,
+                    942,
+                    440,
+                    964,
+                    413,
+                    964
+                ],
+                "score": 0.8,
+                "latex": "="
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    412,
+                    986,
+                    441,
+                    986,
+                    441,
+                    1009,
+                    412,
+                    1009
+                ],
+                "score": 0.8,
+                "latex": "="
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    413,
+                    852,
+                    440,
+                    852,
+                    440,
+                    874,
+                    413,
+                    874
+                ],
+                "score": 0.76,
+                "latex": "="
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    736,
+                    1655,
+                    772,
+                    1655,
+                    772,
+                    1686,
+                    736,
+                    1686
+                ],
+                "score": 0.74,
+                "latex": "\\mathbb { R } ^ { d }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1040,
+                    1780,
+                    1060,
+                    1780,
+                    1060,
+                    1807,
+                    1040,
+                    1807
+                ],
+                "score": 0.71,
+                "latex": "\\lambda"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1126,
+                    1749,
+                    1189,
+                    1749,
+                    1189,
+                    1777,
+                    1126,
+                    1777
+                ],
+                "score": 0.59,
+                "latex": "C L S"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    780,
+                    1857,
+                    791,
+                    1857,
+                    791,
+                    1882,
+                    780,
+                    1882
+                ],
+                "score": 0.5,
+                "latex": "l"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    782,
+                    1657,
+                    796,
+                    1657,
+                    796,
+                    1686,
+                    782,
+                    1686
+                ],
+                "score": 0.4,
+                "latex": "l"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    970,
+                    988,
+                    990,
+                    988,
+                    990,
+                    1007,
+                    970,
+                    1007
+                ],
+                "score": 0.25,
+                "latex": "\\star"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    311.0,
+                    221.0,
+                    550.0,
+                    221.0,
+                    550.0,
+                    252.0,
+                    311.0,
+                    252.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    690.0,
+                    217.0,
+                    875.0,
+                    217.0,
+                    875.0,
+                    250.0,
+                    690.0,
+                    250.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1067.0,
+                    220.0,
+                    1208.0,
+                    220.0,
+                    1208.0,
+                    249.0,
+                    1067.0,
+                    249.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    304.0,
+                    244.0,
+                    360.0,
+                    244.0,
+                    360.0,
+                    271.0,
+                    304.0,
+                    271.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    316.0,
+                    267.0,
+                    359.0,
+                    267.0,
+                    359.0,
+                    308.0,
+                    316.0,
+                    308.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    394.0,
+                    262.0,
+                    429.0,
+                    262.0,
+                    429.0,
+                    307.0,
+                    394.0,
+                    307.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    462.0,
+                    270.0,
+                    489.0,
+                    270.0,
+                    489.0,
+                    311.0,
+                    462.0,
+                    311.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    516.0,
+                    273.0,
+                    550.0,
+                    273.0,
+                    550.0,
+                    316.0,
+                    516.0,
+                    316.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    590.0,
+                    262.0,
+                    612.0,
+                    262.0,
+                    612.0,
+                    302.0,
+                    590.0,
+                    302.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    967.0,
+                    263.0,
+                    988.0,
+                    263.0,
+                    988.0,
+                    296.0,
+                    967.0,
+                    296.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1222.0,
+                    289.0,
+                    1242.0,
+                    289.0,
+                    1242.0,
+                    309.0,
+                    1222.0,
+                    309.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    326.0,
+                    308.0,
+                    349.0,
+                    308.0,
+                    349.0,
+                    327.0,
+                    326.0,
+                    327.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    304.0,
+                    326.0,
+                    353.0,
+                    326.0,
+                    353.0,
+                    347.0,
+                    304.0,
+                    347.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    325.0,
+                    352.0,
+                    344.0,
+                    352.0,
+                    344.0,
+                    367.0,
+                    325.0,
+                    367.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    406.0,
+                    346.0,
+                    416.0,
+                    346.0,
+                    416.0,
+                    358.0,
+                    406.0,
+                    358.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    466.0,
+                    353.0,
+                    476.0,
+                    353.0,
+                    476.0,
+                    365.0,
+                    466.0,
+                    365.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    706.0,
+                    348.0,
+                    716.0,
+                    348.0,
+                    716.0,
+                    359.0,
+                    706.0,
+                    359.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    782.0,
+                    339.0,
+                    797.0,
+                    339.0,
+                    797.0,
+                    355.0,
+                    782.0,
+                    355.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    843.0,
+                    347.0,
+                    853.0,
+                    347.0,
+                    853.0,
+                    357.0,
+                    843.0,
+                    357.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    969.0,
+                    344.0,
+                    984.0,
+                    344.0,
+                    984.0,
+                    359.0,
+                    969.0,
+                    359.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1081.0,
+                    350.0,
+                    1096.0,
+                    350.0,
+                    1096.0,
+                    366.0,
+                    1081.0,
+                    366.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1219.0,
+                    350.0,
+                    1234.0,
+                    350.0,
+                    1234.0,
+                    366.0,
+                    1219.0,
+                    366.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1277.0,
+                    352.0,
+                    1296.0,
+                    352.0,
+                    1296.0,
+                    366.0,
+                    1277.0,
+                    366.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1343.0,
+                    348.0,
+                    1367.0,
+                    348.0,
+                    1367.0,
+                    382.0,
+                    1343.0,
+                    382.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    338.0,
+                    368.0,
+                    359.0,
+                    368.0,
+                    359.0,
+                    388.0,
+                    338.0,
+                    388.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    395.0,
+                    368.0,
+                    417.0,
+                    368.0,
+                    417.0,
+                    387.0,
+                    395.0,
+                    387.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    590.0,
+                    363.0,
+                    612.0,
+                    363.0,
+                    612.0,
+                    382.0,
+                    590.0,
+                    382.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    713.0,
+                    363.0,
+                    736.0,
+                    363.0,
+                    736.0,
+                    382.0,
+                    713.0,
+                    382.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    771.0,
+                    362.0,
+                    794.0,
+                    362.0,
+                    794.0,
+                    383.0,
+                    771.0,
+                    383.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    848.0,
+                    376.0,
+                    859.0,
+                    376.0,
+                    859.0,
+                    385.0,
+                    848.0,
+                    385.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    900.0,
+                    373.0,
+                    910.0,
+                    373.0,
+                    910.0,
+                    385.0,
+                    900.0,
+                    385.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1094.0,
+                    370.0,
+                    1111.0,
+                    370.0,
+                    1111.0,
+                    386.0,
+                    1094.0,
+                    386.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1150.0,
+                    367.0,
+                    1171.0,
+                    367.0,
+                    1171.0,
+                    386.0,
+                    1150.0,
+                    386.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1217.0,
+                    365.0,
+                    1244.0,
+                    365.0,
+                    1244.0,
+                    393.0,
+                    1217.0,
+                    393.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    470.0,
+                    381.0,
+                    481.0,
+                    381.0,
+                    481.0,
+                    390.0,
+                    470.0,
+                    390.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    542.0,
+                    386.0,
+                    555.0,
+                    386.0,
+                    555.0,
+                    396.0,
+                    542.0,
+                    396.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    417.0,
+                    383.0,
+                    417.0,
+                    383.0,
+                    463.0,
+                    294.0,
+                    463.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    721.0,
+                    416.0,
+                    789.0,
+                    416.0,
+                    789.0,
+                    461.0,
+                    721.0,
+                    461.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    844.0,
+                    413.0,
+                    899.0,
+                    413.0,
+                    899.0,
+                    460.0,
+                    844.0,
+                    460.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    938.0,
+                    409.0,
+                    1024.0,
+                    409.0,
+                    1024.0,
+                    457.0,
+                    938.0,
+                    457.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1060.0,
+                    405.0,
+                    1129.0,
+                    405.0,
+                    1129.0,
+                    451.0,
+                    1060.0,
+                    451.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1133.0,
+                    405.0,
+                    1317.0,
+                    405.0,
+                    1317.0,
+                    460.0,
+                    1133.0,
+                    460.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1325.0,
+                    405.0,
+                    1393.0,
+                    405.0,
+                    1393.0,
+                    429.0,
+                    1325.0,
+                    429.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    313.0,
+                    468.0,
+                    578.0,
+                    468.0,
+                    578.0,
+                    500.0,
+                    313.0,
+                    500.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    697.0,
+                    465.0,
+                    878.0,
+                    465.0,
+                    878.0,
+                    497.0,
+                    697.0,
+                    497.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1067.0,
+                    467.0,
+                    1201.0,
+                    467.0,
+                    1201.0,
+                    497.0,
+                    1067.0,
+                    497.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    303.0,
+                    496.0,
+                    358.0,
+                    496.0,
+                    358.0,
+                    521.0,
+                    303.0,
+                    521.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    320.0,
+                    519.0,
+                    349.0,
+                    519.0,
+                    349.0,
+                    544.0,
+                    320.0,
+                    544.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    393.0,
+                    519.0,
+                    421.0,
+                    519.0,
+                    421.0,
+                    557.0,
+                    393.0,
+                    557.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    590.0,
+                    523.0,
+                    607.0,
+                    523.0,
+                    607.0,
+                    540.0,
+                    590.0,
+                    540.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    970.0,
+                    525.0,
+                    983.0,
+                    525.0,
+                    983.0,
+                    536.0,
+                    970.0,
+                    536.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    968.0,
+                    532.0,
+                    988.0,
+                    532.0,
+                    988.0,
+                    553.0,
+                    968.0,
+                    553.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    324.0,
+                    542.0,
+                    359.0,
+                    542.0,
+                    359.0,
+                    578.0,
+                    324.0,
+                    578.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    393.0,
+                    563.0,
+                    412.0,
+                    563.0,
+                    412.0,
+                    576.0,
+                    393.0,
+                    576.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    465.0,
+                    553.0,
+                    489.0,
+                    553.0,
+                    489.0,
+                    575.0,
+                    465.0,
+                    575.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    520.0,
+                    546.0,
+                    550.0,
+                    546.0,
+                    550.0,
+                    568.0,
+                    520.0,
+                    568.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    590.0,
+                    561.0,
+                    610.0,
+                    561.0,
+                    610.0,
+                    576.0,
+                    590.0,
+                    576.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    300.0,
+                    576.0,
+                    353.0,
+                    576.0,
+                    353.0,
+                    601.0,
+                    300.0,
+                    601.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    325.0,
+                    605.0,
+                    343.0,
+                    605.0,
+                    343.0,
+                    621.0,
+                    325.0,
+                    621.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    395.0,
+                    603.0,
+                    416.0,
+                    603.0,
+                    416.0,
+                    622.0,
+                    395.0,
+                    622.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    593.0,
+                    607.0,
+                    600.0,
+                    607.0,
+                    600.0,
+                    613.0,
+                    593.0,
+                    613.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    704.0,
+                    605.0,
+                    719.0,
+                    605.0,
+                    719.0,
+                    619.0,
+                    704.0,
+                    619.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1084.0,
+                    608.0,
+                    1095.0,
+                    608.0,
+                    1095.0,
+                    618.0,
+                    1084.0,
+                    618.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1343.0,
+                    603.0,
+                    1366.0,
+                    603.0,
+                    1366.0,
+                    637.0,
+                    1343.0,
+                    637.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    585.0,
+                    623.0,
+                    600.0,
+                    623.0,
+                    600.0,
+                    638.0,
+                    585.0,
+                    638.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    713.0,
+                    619.0,
+                    736.0,
+                    619.0,
+                    736.0,
+                    641.0,
+                    713.0,
+                    641.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    772.0,
+                    621.0,
+                    794.0,
+                    621.0,
+                    794.0,
+                    641.0,
+                    772.0,
+                    641.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1093.0,
+                    623.0,
+                    1114.0,
+                    623.0,
+                    1114.0,
+                    642.0,
+                    1093.0,
+                    642.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    326.0,
+                    645.0,
+                    336.0,
+                    645.0,
+                    336.0,
+                    655.0,
+                    326.0,
+                    655.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1084.0,
+                    646.0,
+                    1100.0,
+                    646.0,
+                    1100.0,
+                    656.0,
+                    1084.0,
+                    656.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1345.0,
+                    643.0,
+                    1361.0,
+                    643.0,
+                    1361.0,
+                    657.0,
+                    1345.0,
+                    657.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    668.0,
+                    382.0,
+                    668.0,
+                    382.0,
+                    715.0,
+                    295.0,
+                    715.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    722.0,
+                    675.0,
+                    787.0,
+                    675.0,
+                    787.0,
+                    718.0,
+                    722.0,
+                    718.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    844.0,
+                    673.0,
+                    899.0,
+                    673.0,
+                    899.0,
+                    718.0,
+                    844.0,
+                    718.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    939.0,
+                    668.0,
+                    1025.0,
+                    668.0,
+                    1025.0,
+                    714.0,
+                    939.0,
+                    714.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1060.0,
+                    659.0,
+                    1129.0,
+                    659.0,
+                    1129.0,
+                    706.0,
+                    1060.0,
+                    706.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1134.0,
+                    659.0,
+                    1267.0,
+                    659.0,
+                    1267.0,
+                    716.0,
+                    1134.0,
+                    716.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1269.0,
+                    659.0,
+                    1316.0,
+                    659.0,
+                    1316.0,
+                    685.0,
+                    1269.0,
+                    685.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1327.0,
+                    659.0,
+                    1395.0,
+                    659.0,
+                    1395.0,
+                    685.0,
+                    1327.0,
+                    685.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1268.0,
+                    367.0,
+                    1316.0,
+                    367.0,
+                    1316.0,
+                    394.5,
+                    1268.0,
+                    394.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    338.0,
+                    630.0,
+                    351.0,
+                    630.0,
+                    351.0,
+                    640.0,
+                    338.0,
+                    640.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1122.0,
+                    654.0,
+                    1122.0,
+                    654.0,
+                    1171.0,
+                    291.0,
+                    1171.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    386.0,
+                    744.0,
+                    1312.0,
+                    744.0,
+                    1312.0,
+                    788.0,
+                    386.0,
+                    788.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    839.0,
+                    2059.0,
+                    862.0,
+                    2059.0,
+                    862.0,
+                    2091.0,
+                    839.0,
+                    2091.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    298.0,
+                    893.0,
+                    606.0,
+                    893.0,
+                    606.0,
+                    925.0,
+                    298.0,
+                    925.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1187.0,
+                    1405.0,
+                    1187.0,
+                    1405.0,
+                    1223.0,
+                    295.0,
+                    1223.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1217.0,
+                    1406.0,
+                    1217.0,
+                    1406.0,
+                    1253.0,
+                    294.0,
+                    1253.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1247.0,
+                    1405.0,
+                    1247.0,
+                    1405.0,
+                    1284.0,
+                    294.0,
+                    1284.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1280.0,
+                    1403.0,
+                    1280.0,
+                    1403.0,
+                    1312.0,
+                    296.0,
+                    1312.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1308.0,
+                    1407.0,
+                    1308.0,
+                    1407.0,
+                    1343.0,
+                    294.0,
+                    1343.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1336.0,
+                    1307.0,
+                    1336.0,
+                    1307.0,
+                    1374.0,
+                    294.0,
+                    1374.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1653.0,
+                    371.0,
+                    1653.0,
+                    371.0,
+                    1694.0,
+                    292.0,
+                    1694.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    393.0,
+                    1653.0,
+                    735.0,
+                    1653.0,
+                    735.0,
+                    1694.0,
+                    393.0,
+                    1694.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    773.0,
+                    1653.0,
+                    781.0,
+                    1653.0,
+                    781.0,
+                    1694.0,
+                    773.0,
+                    1694.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    797.0,
+                    1653.0,
+                    1024.0,
+                    1653.0,
+                    1024.0,
+                    1694.0,
+                    797.0,
+                    1694.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1160.0,
+                    1653.0,
+                    1407.0,
+                    1653.0,
+                    1407.0,
+                    1694.0,
+                    1160.0,
+                    1694.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1684.0,
+                    1406.0,
+                    1684.0,
+                    1406.0,
+                    1725.0,
+                    294.0,
+                    1725.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1717.0,
+                    1406.0,
+                    1717.0,
+                    1406.0,
+                    1753.0,
+                    294.0,
+                    1753.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1749.0,
+                    502.0,
+                    1749.0,
+                    502.0,
+                    1781.0,
+                    296.0,
+                    1781.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    524.0,
+                    1749.0,
+                    1125.0,
+                    1749.0,
+                    1125.0,
+                    1781.0,
+                    524.0,
+                    1781.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1190.0,
+                    1749.0,
+                    1405.0,
+                    1749.0,
+                    1405.0,
+                    1781.0,
+                    1190.0,
+                    1781.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1778.0,
+                    1039.0,
+                    1778.0,
+                    1039.0,
+                    1812.0,
+                    294.0,
+                    1812.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1061.0,
+                    1778.0,
+                    1406.0,
+                    1778.0,
+                    1406.0,
+                    1812.0,
+                    1061.0,
+                    1812.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1808.0,
+                    1312.0,
+                    1808.0,
+                    1312.0,
+                    1844.0,
+                    295.0,
+                    1844.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1385.0,
+                    951.0,
+                    1385.0,
+                    951.0,
+                    1418.0,
+                    296.0,
+                    1418.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    990.0,
+                    1385.0,
+                    1252.0,
+                    1385.0,
+                    1252.0,
+                    1418.0,
+                    990.0,
+                    1418.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1399.0,
+                    1385.0,
+                    1406.0,
+                    1385.0,
+                    1406.0,
+                    1418.0,
+                    1399.0,
+                    1418.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1414.0,
+                    342.0,
+                    1414.0,
+                    342.0,
+                    1451.0,
+                    294.0,
+                    1451.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    382.0,
+                    1414.0,
+                    708.0,
+                    1414.0,
+                    708.0,
+                    1451.0,
+                    382.0,
+                    1451.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    789.0,
+                    1414.0,
+                    1406.0,
+                    1414.0,
+                    1406.0,
+                    1451.0,
+                    789.0,
+                    1451.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1444.0,
+                    1295.0,
+                    1444.0,
+                    1295.0,
+                    1481.0,
+                    294.0,
+                    1481.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1335.0,
+                    1444.0,
+                    1406.0,
+                    1444.0,
+                    1406.0,
+                    1481.0,
+                    1335.0,
+                    1481.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1475.0,
+                    501.0,
+                    1475.0,
+                    501.0,
+                    1512.0,
+                    294.0,
+                    1512.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    539.0,
+                    1475.0,
+                    1406.0,
+                    1475.0,
+                    1406.0,
+                    1512.0,
+                    539.0,
+                    1512.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1504.0,
+                    903.0,
+                    1504.0,
+                    903.0,
+                    1542.0,
+                    294.0,
+                    1542.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1852.0,
+                    779.0,
+                    1852.0,
+                    779.0,
+                    1888.0,
+                    295.0,
+                    1888.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    792.0,
+                    1852.0,
+                    1404.0,
+                    1852.0,
+                    1404.0,
+                    1888.0,
+                    792.0,
+                    1888.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1885.0,
+                    983.0,
+                    1885.0,
+                    983.0,
+                    1919.0,
+                    296.0,
+                    1919.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1029.0,
+                    1885.0,
+                    1077.0,
+                    1885.0,
+                    1077.0,
+                    1919.0,
+                    1029.0,
+                    1919.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1113.0,
+                    1885.0,
+                    1406.0,
+                    1885.0,
+                    1406.0,
+                    1919.0,
+                    1113.0,
+                    1919.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1913.0,
+                    1406.0,
+                    1913.0,
+                    1406.0,
+                    1950.0,
+                    294.0,
+                    1950.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1944.0,
+                    1406.0,
+                    1944.0,
+                    1406.0,
+                    1981.0,
+                    294.0,
+                    1981.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1974.0,
+                    1405.0,
+                    1974.0,
+                    1405.0,
+                    2013.0,
+                    295.0,
+                    2013.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1026.0,
+                    1405.0,
+                    1026.0,
+                    1405.0,
+                    1061.0,
+                    295.0,
+                    1061.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1056.0,
+                    1351.0,
+                    1056.0,
+                    1351.0,
+                    1092.0,
+                    295.0,
+                    1092.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    356.0,
+                    937.0,
+                    412.0,
+                    937.0,
+                    412.0,
+                    969.0,
+                    356.0,
+                    969.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    441.0,
+                    937.0,
+                    1121.0,
+                    937.0,
+                    1121.0,
+                    969.0,
+                    441.0,
+                    969.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    355.0,
+                    982.0,
+                    411.0,
+                    982.0,
+                    411.0,
+                    1015.0,
+                    355.0,
+                    1015.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    442.0,
+                    982.0,
+                    969.0,
+                    982.0,
+                    969.0,
+                    1015.0,
+                    442.0,
+                    1015.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    991.0,
+                    982.0,
+                    1089.0,
+                    982.0,
+                    1089.0,
+                    1015.0,
+                    991.0,
+                    1015.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    356.0,
+                    847.0,
+                    412.0,
+                    847.0,
+                    412.0,
+                    882.0,
+                    356.0,
+                    882.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    441.0,
+                    847.0,
+                    1092.0,
+                    847.0,
+                    1092.0,
+                    882.0,
+                    441.0,
+                    882.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    298.0,
+                    893.0,
+                    606.0,
+                    893.0,
+                    606.0,
+                    925.0,
+                    298.0,
+                    925.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 3,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    1338,
+                    1403,
+                    1338,
+                    1403,
+                    1583,
+                    299,
+                    1583
+                ],
+                "score": 0.984
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1595,
+                    1404,
+                    1595,
+                    1404,
+                    1840,
+                    298,
+                    1840
+                ],
+                "score": 0.981
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    279,
+                    1403,
+                    279,
+                    1403,
+                    523,
+                    298,
+                    523
+                ],
+                "score": 0.98
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    535,
+                    1405,
+                    535,
+                    1405,
+                    780,
+                    298,
+                    780
+                ],
+                "score": 0.98
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1140,
+                    1404,
+                    1140,
+                    1404,
+                    1325,
+                    298,
+                    1325
+                ],
+                "score": 0.979
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    909,
+                    1405,
+                    909,
+                    1405,
+                    1032,
+                    299,
+                    1032
+                ],
+                "score": 0.977
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    300,
+                    1884,
+                    1403,
+                    1884,
+                    1403,
+                    2007,
+                    300,
+                    2007
+                ],
+                "score": 0.976
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    295,
+                    203,
+                    1400,
+                    203,
+                    1400,
+                    264,
+                    295,
+                    264
+                ],
+                "score": 0.952
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    298,
+                    833,
+                    532,
+                    833,
+                    532,
+                    872,
+                    298,
+                    872
+                ],
+                "score": 0.909
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    298,
+                    1079,
+                    498,
+                    1079,
+                    498,
+                    1112,
+                    298,
+                    1112
+                ],
+                "score": 0.904
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    841,
+                    2062,
+                    858,
+                    2062,
+                    858,
+                    2085,
+                    841,
+                    2085
+                ],
+                "score": 0.694
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    841,
+                    2061,
+                    859,
+                    2061,
+                    859,
+                    2085,
+                    841,
+                    2085
+                ],
+                "score": 0.123
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    486,
+                    341,
+                    647,
+                    341,
+                    647,
+                    371,
+                    486,
+                    371
+                ],
+                "score": 0.92,
+                "latex": "\\tau _ { t i } = 1 - m _ { t i }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    928,
+                    599,
+                    1025,
+                    599,
+                    1025,
+                    628,
+                    928,
+                    628
+                ],
+                "score": 0.91,
+                "latex": "m _ { t i } = 0"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1328,
+                    598,
+                    1398,
+                    598,
+                    1398,
+                    628,
+                    1328,
+                    628
+                ],
+                "score": 0.91,
+                "latex": "t \\geq T"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1189,
+                    600,
+                    1285,
+                    600,
+                    1285,
+                    628,
+                    1189,
+                    628
+                ],
+                "score": 0.91,
+                "latex": "m _ { t i } = 1"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1053,
+                    568,
+                    1151,
+                    568,
+                    1151,
+                    598,
+                    1053,
+                    598
+                ],
+                "score": 0.91,
+                "latex": "m _ { t i } = 0"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    603,
+                    568,
+                    690,
+                    568,
+                    690,
+                    598,
+                    603,
+                    598
+                ],
+                "score": 0.9,
+                "latex": "\\tau _ { t i } = 1"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    753,
+                    281,
+                    859,
+                    281,
+                    859,
+                    310,
+                    753,
+                    310
+                ],
+                "score": 0.9,
+                "latex": "m _ { t i } = 1"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1068,
+                    599,
+                    1138,
+                    599,
+                    1138,
+                    626,
+                    1068,
+                    626
+                ],
+                "score": 0.9,
+                "latex": "t < T"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1203,
+                    569,
+                    1289,
+                    569,
+                    1289,
+                    598,
+                    1203,
+                    598
+                ],
+                "score": 0.9,
+                "latex": "\\tau _ { t i } = 0"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1249,
+                    539,
+                    1346,
+                    539,
+                    1346,
+                    567,
+                    1249,
+                    567
+                ],
+                "score": 0.89,
+                "latex": "m _ { t i } = 1"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    607,
+                    313,
+                    652,
+                    313,
+                    652,
+                    340,
+                    607,
+                    340
+                ],
+                "score": 0.86,
+                "latex": "m _ { t i }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    297,
+                    598,
+                    346,
+                    598,
+                    346,
+                    629,
+                    297,
+                    629
+                ],
+                "score": 0.85,
+                "latex": "c { - } f )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1138,
+                    660,
+                    1181,
+                    660,
+                    1181,
+                    690,
+                    1138,
+                    690
+                ],
+                "score": 0.84,
+                "latex": "e , f ,"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    297,
+                    719,
+                    346,
+                    719,
+                    346,
+                    751,
+                    297,
+                    751
+                ],
+                "score": 0.82,
+                "latex": "d , f )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    656,
+                    600,
+                    680,
+                    600,
+                    680,
+                    625,
+                    656,
+                    625
+                ],
+                "score": 0.8,
+                "latex": "T"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    972,
+                    1371,
+                    1027,
+                    1371,
+                    1027,
+                    1401,
+                    972,
+                    1401
+                ],
+                "score": 0.79,
+                "latex": "\\left( \\mathrm { V } + \\right)"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1122,
+                    1339,
+                    1176,
+                    1339,
+                    1176,
+                    1370,
+                    1122,
+                    1370
+                ],
+                "score": 0.78,
+                "latex": "\\left( \\mathrm { V } + \\right)"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    851,
+                    311,
+                    870,
+                    311,
+                    870,
+                    338,
+                    851,
+                    338
+                ],
+                "score": 0.75,
+                "latex": "^ b"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1261,
+                    282,
+                    1274,
+                    282,
+                    1274,
+                    308,
+                    1261,
+                    308
+                ],
+                "score": 0.73,
+                "latex": "i"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1191,
+                    282,
+                    1206,
+                    282,
+                    1206,
+                    307,
+                    1191,
+                    307
+                ],
+                "score": 0.73,
+                "latex": "t"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    758,
+                    569,
+                    772,
+                    569,
+                    772,
+                    594,
+                    758,
+                    594
+                ],
+                "score": 0.65,
+                "latex": "t"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1020,
+                    498,
+                    1038,
+                    498,
+                    1038,
+                    519,
+                    1020,
+                    519
+                ],
+                "score": 0.49,
+                "latex": "a"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    384,
+                    660,
+                    427,
+                    660,
+                    427,
+                    688,
+                    384,
+                    688
+                ],
+                "score": 0.37,
+                "latex": "c , d"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    819,
+                    693,
+                    862,
+                    693,
+                    862,
+                    718,
+                    819,
+                    718
+                ],
+                "score": 0.35,
+                "latex": "c , e"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    289.0,
+                    828.0,
+                    537.0,
+                    828.0,
+                    537.0,
+                    880.0,
+                    289.0,
+                    880.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1077.0,
+                    503.0,
+                    1077.0,
+                    503.0,
+                    1120.0,
+                    292.0,
+                    1120.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    839.0,
+                    2060.0,
+                    860.0,
+                    2060.0,
+                    860.0,
+                    2091.0,
+                    839.0,
+                    2091.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    838.0,
+                    2058.0,
+                    862.0,
+                    2058.0,
+                    862.0,
+                    2092.0,
+                    838.0,
+                    2092.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1336.0,
+                    1121.0,
+                    1336.0,
+                    1121.0,
+                    1373.0,
+                    294.0,
+                    1373.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1177.0,
+                    1336.0,
+                    1407.0,
+                    1336.0,
+                    1407.0,
+                    1373.0,
+                    1177.0,
+                    1373.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1369.0,
+                    971.0,
+                    1369.0,
+                    971.0,
+                    1403.0,
+                    294.0,
+                    1403.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1028.0,
+                    1369.0,
+                    1407.0,
+                    1369.0,
+                    1407.0,
+                    1403.0,
+                    1028.0,
+                    1403.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1398.0,
+                    1407.0,
+                    1398.0,
+                    1407.0,
+                    1434.0,
+                    293.0,
+                    1434.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1429.0,
+                    1407.0,
+                    1429.0,
+                    1407.0,
+                    1463.0,
+                    294.0,
+                    1463.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1459.0,
+                    1407.0,
+                    1459.0,
+                    1407.0,
+                    1494.0,
+                    293.0,
+                    1494.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1490.0,
+                    1408.0,
+                    1490.0,
+                    1408.0,
+                    1528.0,
+                    293.0,
+                    1528.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1522.0,
+                    1405.0,
+                    1522.0,
+                    1405.0,
+                    1555.0,
+                    294.0,
+                    1555.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1550.0,
+                    1306.0,
+                    1550.0,
+                    1306.0,
+                    1586.0,
+                    293.0,
+                    1586.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1595.0,
+                    1404.0,
+                    1595.0,
+                    1404.0,
+                    1631.0,
+                    293.0,
+                    1631.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1626.0,
+                    1405.0,
+                    1626.0,
+                    1405.0,
+                    1661.0,
+                    293.0,
+                    1661.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1656.0,
+                    1405.0,
+                    1656.0,
+                    1405.0,
+                    1695.0,
+                    295.0,
+                    1695.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1689.0,
+                    1404.0,
+                    1689.0,
+                    1404.0,
+                    1722.0,
+                    294.0,
+                    1722.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1718.0,
+                    1406.0,
+                    1718.0,
+                    1406.0,
+                    1753.0,
+                    292.0,
+                    1753.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1746.0,
+                    1406.0,
+                    1746.0,
+                    1406.0,
+                    1782.0,
+                    292.0,
+                    1782.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1780.0,
+                    1405.0,
+                    1780.0,
+                    1405.0,
+                    1813.0,
+                    294.0,
+                    1813.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1811.0,
+                    1376.0,
+                    1811.0,
+                    1376.0,
+                    1841.0,
+                    296.0,
+                    1841.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    277.0,
+                    752.0,
+                    277.0,
+                    752.0,
+                    315.0,
+                    292.0,
+                    315.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    860.0,
+                    277.0,
+                    1190.0,
+                    277.0,
+                    1190.0,
+                    315.0,
+                    860.0,
+                    315.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1207.0,
+                    277.0,
+                    1260.0,
+                    277.0,
+                    1260.0,
+                    315.0,
+                    1207.0,
+                    315.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1275.0,
+                    277.0,
+                    1405.0,
+                    277.0,
+                    1405.0,
+                    315.0,
+                    1275.0,
+                    315.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    310.0,
+                    606.0,
+                    310.0,
+                    606.0,
+                    344.0,
+                    293.0,
+                    344.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    653.0,
+                    310.0,
+                    850.0,
+                    310.0,
+                    850.0,
+                    344.0,
+                    653.0,
+                    344.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    871.0,
+                    310.0,
+                    1404.0,
+                    310.0,
+                    1404.0,
+                    344.0,
+                    871.0,
+                    344.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    340.0,
+                    485.0,
+                    340.0,
+                    485.0,
+                    374.0,
+                    294.0,
+                    374.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    648.0,
+                    340.0,
+                    1405.0,
+                    340.0,
+                    1405.0,
+                    374.0,
+                    648.0,
+                    374.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    370.0,
+                    1404.0,
+                    370.0,
+                    1404.0,
+                    403.0,
+                    294.0,
+                    403.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    401.0,
+                    1403.0,
+                    401.0,
+                    1403.0,
+                    434.0,
+                    294.0,
+                    434.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    431.0,
+                    1403.0,
+                    431.0,
+                    1403.0,
+                    464.0,
+                    294.0,
+                    464.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    461.0,
+                    1404.0,
+                    461.0,
+                    1404.0,
+                    498.0,
+                    293.0,
+                    498.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    491.0,
+                    1019.0,
+                    491.0,
+                    1019.0,
+                    526.0,
+                    293.0,
+                    526.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1039.0,
+                    491.0,
+                    1056.0,
+                    491.0,
+                    1056.0,
+                    526.0,
+                    1039.0,
+                    526.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    537.0,
+                    1248.0,
+                    537.0,
+                    1248.0,
+                    570.0,
+                    295.0,
+                    570.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1347.0,
+                    537.0,
+                    1405.0,
+                    537.0,
+                    1405.0,
+                    570.0,
+                    1347.0,
+                    570.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    566.0,
+                    602.0,
+                    566.0,
+                    602.0,
+                    600.0,
+                    293.0,
+                    600.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    691.0,
+                    566.0,
+                    757.0,
+                    566.0,
+                    757.0,
+                    600.0,
+                    691.0,
+                    600.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    773.0,
+                    566.0,
+                    1052.0,
+                    566.0,
+                    1052.0,
+                    600.0,
+                    773.0,
+                    600.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1152.0,
+                    566.0,
+                    1202.0,
+                    566.0,
+                    1202.0,
+                    600.0,
+                    1152.0,
+                    600.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1290.0,
+                    566.0,
+                    1405.0,
+                    566.0,
+                    1405.0,
+                    600.0,
+                    1290.0,
+                    600.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    598.0,
+                    296.0,
+                    598.0,
+                    296.0,
+                    631.0,
+                    293.0,
+                    631.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    347.0,
+                    598.0,
+                    655.0,
+                    598.0,
+                    655.0,
+                    631.0,
+                    347.0,
+                    631.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    681.0,
+                    598.0,
+                    927.0,
+                    598.0,
+                    927.0,
+                    631.0,
+                    681.0,
+                    631.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1026.0,
+                    598.0,
+                    1067.0,
+                    598.0,
+                    1067.0,
+                    631.0,
+                    1026.0,
+                    631.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1139.0,
+                    598.0,
+                    1188.0,
+                    598.0,
+                    1188.0,
+                    631.0,
+                    1139.0,
+                    631.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1286.0,
+                    598.0,
+                    1327.0,
+                    598.0,
+                    1327.0,
+                    631.0,
+                    1286.0,
+                    631.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1399.0,
+                    598.0,
+                    1407.0,
+                    598.0,
+                    1407.0,
+                    631.0,
+                    1399.0,
+                    631.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    625.0,
+                    1408.0,
+                    625.0,
+                    1408.0,
+                    664.0,
+                    291.0,
+                    664.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    659.0,
+                    383.0,
+                    659.0,
+                    383.0,
+                    692.0,
+                    295.0,
+                    692.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    428.0,
+                    659.0,
+                    1137.0,
+                    659.0,
+                    1137.0,
+                    692.0,
+                    428.0,
+                    692.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1182.0,
+                    659.0,
+                    1405.0,
+                    659.0,
+                    1405.0,
+                    692.0,
+                    1182.0,
+                    692.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    687.0,
+                    818.0,
+                    687.0,
+                    818.0,
+                    725.0,
+                    293.0,
+                    725.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    863.0,
+                    687.0,
+                    1406.0,
+                    687.0,
+                    1406.0,
+                    725.0,
+                    863.0,
+                    725.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    716.0,
+                    296.0,
+                    716.0,
+                    296.0,
+                    755.0,
+                    292.0,
+                    755.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    347.0,
+                    716.0,
+                    1406.0,
+                    716.0,
+                    1406.0,
+                    755.0,
+                    347.0,
+                    755.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    747.0,
+                    948.0,
+                    747.0,
+                    948.0,
+                    784.0,
+                    295.0,
+                    784.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1141.0,
+                    1406.0,
+                    1141.0,
+                    1406.0,
+                    1177.0,
+                    296.0,
+                    1177.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1173.0,
+                    1405.0,
+                    1173.0,
+                    1405.0,
+                    1207.0,
+                    293.0,
+                    1207.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1202.0,
+                    1406.0,
+                    1202.0,
+                    1406.0,
+                    1238.0,
+                    293.0,
+                    1238.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1232.0,
+                    1404.0,
+                    1232.0,
+                    1404.0,
+                    1268.0,
+                    293.0,
+                    1268.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1259.0,
+                    1405.0,
+                    1259.0,
+                    1405.0,
+                    1299.0,
+                    293.0,
+                    1299.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1294.0,
+                    1036.0,
+                    1294.0,
+                    1036.0,
+                    1329.0,
+                    293.0,
+                    1329.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    911.0,
+                    1403.0,
+                    911.0,
+                    1403.0,
+                    944.0,
+                    295.0,
+                    944.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    941.0,
+                    1406.0,
+                    941.0,
+                    1406.0,
+                    977.0,
+                    294.0,
+                    977.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    970.0,
+                    1410.0,
+                    970.0,
+                    1410.0,
+                    1009.0,
+                    293.0,
+                    1009.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1002.0,
+                    1156.0,
+                    1002.0,
+                    1156.0,
+                    1034.0,
+                    295.0,
+                    1034.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1884.0,
+                    1405.0,
+                    1884.0,
+                    1405.0,
+                    1920.0,
+                    295.0,
+                    1920.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1914.0,
+                    1405.0,
+                    1914.0,
+                    1405.0,
+                    1947.0,
+                    295.0,
+                    1947.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1945.0,
+                    1407.0,
+                    1945.0,
+                    1407.0,
+                    1979.0,
+                    294.0,
+                    1979.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1977.0,
+                    1404.0,
+                    1977.0,
+                    1404.0,
+                    2009.0,
+                    295.0,
+                    2009.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    202.0,
+                    1406.0,
+                    202.0,
+                    1406.0,
+                    240.0,
+                    293.0,
+                    240.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    233.0,
+                    1298.0,
+                    233.0,
+                    1298.0,
+                    269.0,
+                    295.0,
+                    269.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 4,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 5,
+                "poly": [
+                    305,
+                    194,
+                    1394,
+                    194,
+                    1394,
+                    478,
+                    305,
+                    478
+                ],
+                "score": 0.98,
+                "html": "<table><tr><td>Model</td><td>Descriptive</td><td>Explanatory</td><td>Predictive</td><td>Counterfactual</td></tr><tr><td>MAC (V+)</td><td>86.4</td><td>22.3</td><td>42.9</td><td>25.1</td></tr><tr><td>NS-DR</td><td>88.1</td><td>79.6</td><td>68.7</td><td>42.2</td></tr><tr><td>DCL</td><td>90.7</td><td>82.8</td><td>82.0</td><td>46.5</td></tr><tr><td>Aloe</td><td>94.0 ± 0.4</td><td>96.0 ± 0.6</td><td>87.5 ± 3.0</td><td>75.6 ± 3.8</td></tr><tr><td>Aloe- self-attention+MLP</td><td>45.4</td><td>16.0</td><td>27.7</td><td>9.9</td></tr><tr><td>Aloe- object-repr. +ResNet</td><td>74.9</td><td>66.1</td><td>58.3</td><td>32.4</td></tr><tr><td>Aloe - global + hierarchical attn.</td><td>80.6</td><td>87.4</td><td>73.5</td><td>55.1</td></tr><tr><td>Aloe- self-supervised loss</td><td>91.0</td><td>92.8</td><td>82.8</td><td>68.7</td></tr></table>"
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    296,
+                    1540,
+                    1404,
+                    1540,
+                    1404,
+                    1815,
+                    296,
+                    1815
+                ],
+                "score": 0.98
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1006,
+                    1407,
+                    1006,
+                    1407,
+                    1220,
+                    297,
+                    1220
+                ],
+                "score": 0.978
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1403,
+                    1403,
+                    1403,
+                    1403,
+                    1527,
+                    298,
+                    1527
+                ],
+                "score": 0.967
+            },
+            {
+                "category_id": 3,
+                "poly": [
+                    303,
+                    758,
+                    649,
+                    758,
+                    649,
+                    991,
+                    303,
+                    991
+                ],
+                "score": 0.964
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    301,
+                    1829,
+                    1399,
+                    1829,
+                    1399,
+                    1921,
+                    301,
+                    1921
+                ],
+                "score": 0.957
+            },
+            {
+                "category_id": 3,
+                "poly": [
+                    301,
+                    1235,
+                    1399,
+                    1235,
+                    1399,
+                    1362,
+                    301,
+                    1362
+                ],
+                "score": 0.949
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    677,
+                    1401,
+                    677,
+                    1401,
+                    741,
+                    297,
+                    741
+                ],
+                "score": 0.947
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    297,
+                    1949,
+                    1403,
+                    1949,
+                    1403,
+                    2007,
+                    297,
+                    2007
+                ],
+                "score": 0.931
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    735,
+                    778,
+                    1293,
+                    778,
+                    1293,
+                    841,
+                    735,
+                    841
+                ],
+                "score": 0.863
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    840,
+                    2062,
+                    859,
+                    2062,
+                    859,
+                    2085,
+                    840,
+                    2085
+                ],
+                "score": 0.802
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    735,
+                    853,
+                    1298,
+                    853,
+                    1298,
+                    969,
+                    735,
+                    969
+                ],
+                "score": 0.692
+            },
+            {
+                "category_id": 6,
+                "poly": [
+                    296,
+                    492,
+                    1405,
+                    492,
+                    1405,
+                    616,
+                    296,
+                    616
+                ],
+                "score": 0.667
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    296,
+                    492,
+                    1405,
+                    492,
+                    1405,
+                    616,
+                    296,
+                    616
+                ],
+                "score": 0.145
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1363,
+                    1576,
+                    1400,
+                    1576,
+                    1400,
+                    1604,
+                    1363,
+                    1604
+                ],
+                "score": 0.82,
+                "latex": "\\mu _ { t i }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    938,
+                    1037,
+                    1001,
+                    1037,
+                    1001,
+                    1065,
+                    938,
+                    1065
+                ],
+                "score": 0.66,
+                "latex": "C L S"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    484,
+                    416,
+                    507,
+                    416,
+                    507,
+                    440,
+                    484,
+                    440
+                ],
+                "score": 0.59,
+                "latex": "^ +"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    587,
+                    358,
+                    609,
+                    358,
+                    609,
+                    379,
+                    587,
+                    379
+                ],
+                "score": 0.52,
+                "latex": "^ +"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    517,
+                    229,
+                    574,
+                    229,
+                    574,
+                    260,
+                    517,
+                    260
+                ],
+                "score": 0.37,
+                "latex": "\\overline { { ( \\mathrm { V } + ) } }"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    473.0,
+                    1244.0,
+                    486.0,
+                    1244.0,
+                    486.0,
+                    1254.0,
+                    473.0,
+                    1254.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    820.0,
+                    1258.0,
+                    838.0,
+                    1258.0,
+                    838.0,
+                    1274.0,
+                    820.0,
+                    1274.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    331.0,
+                    1945.0,
+                    1404.0,
+                    1945.0,
+                    1404.0,
+                    1982.0,
+                    331.0,
+                    1982.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1977.0,
+                    1023.0,
+                    1977.0,
+                    1023.0,
+                    2011.0,
+                    295.0,
+                    2011.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    840.0,
+                    2060.0,
+                    862.0,
+                    2060.0,
+                    862.0,
+                    2091.0,
+                    840.0,
+                    2091.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    490.0,
+                    1405.0,
+                    490.0,
+                    1405.0,
+                    527.0,
+                    291.0,
+                    527.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    521.0,
+                    1406.0,
+                    521.0,
+                    1406.0,
+                    556.0,
+                    293.0,
+                    556.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    554.0,
+                    1405.0,
+                    554.0,
+                    1405.0,
+                    587.0,
+                    296.0,
+                    587.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    586.0,
+                    634.0,
+                    586.0,
+                    634.0,
+                    619.0,
+                    295.0,
+                    619.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1542.0,
+                    1405.0,
+                    1542.0,
+                    1405.0,
+                    1575.0,
+                    296.0,
+                    1575.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    290.0,
+                    1568.0,
+                    1362.0,
+                    1568.0,
+                    1362.0,
+                    1610.0,
+                    290.0,
+                    1610.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1401.0,
+                    1568.0,
+                    1406.0,
+                    1568.0,
+                    1406.0,
+                    1610.0,
+                    1401.0,
+                    1610.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1601.0,
+                    1406.0,
+                    1601.0,
+                    1406.0,
+                    1637.0,
+                    294.0,
+                    1637.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1633.0,
+                    1405.0,
+                    1633.0,
+                    1405.0,
+                    1666.0,
+                    295.0,
+                    1666.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1662.0,
+                    1407.0,
+                    1662.0,
+                    1407.0,
+                    1697.0,
+                    293.0,
+                    1697.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1692.0,
+                    1406.0,
+                    1692.0,
+                    1406.0,
+                    1728.0,
+                    294.0,
+                    1728.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1722.0,
+                    1406.0,
+                    1722.0,
+                    1406.0,
+                    1758.0,
+                    295.0,
+                    1758.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1753.0,
+                    1404.0,
+                    1753.0,
+                    1404.0,
+                    1789.0,
+                    294.0,
+                    1789.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1783.0,
+                    773.0,
+                    1783.0,
+                    773.0,
+                    1819.0,
+                    295.0,
+                    1819.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1004.0,
+                    1408.0,
+                    1004.0,
+                    1408.0,
+                    1041.0,
+                    294.0,
+                    1041.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1037.0,
+                    937.0,
+                    1037.0,
+                    937.0,
+                    1071.0,
+                    295.0,
+                    1071.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1002.0,
+                    1037.0,
+                    1405.0,
+                    1037.0,
+                    1405.0,
+                    1071.0,
+                    1002.0,
+                    1071.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1068.0,
+                    1405.0,
+                    1068.0,
+                    1405.0,
+                    1102.0,
+                    295.0,
+                    1102.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1100.0,
+                    1405.0,
+                    1100.0,
+                    1405.0,
+                    1130.0,
+                    296.0,
+                    1130.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1125.0,
+                    1407.0,
+                    1125.0,
+                    1407.0,
+                    1162.0,
+                    292.0,
+                    1162.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1156.0,
+                    1407.0,
+                    1156.0,
+                    1407.0,
+                    1193.0,
+                    292.0,
+                    1193.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1186.0,
+                    580.0,
+                    1186.0,
+                    580.0,
+                    1225.0,
+                    294.0,
+                    1225.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1404.0,
+                    1405.0,
+                    1404.0,
+                    1405.0,
+                    1437.0,
+                    294.0,
+                    1437.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1436.0,
+                    1405.0,
+                    1436.0,
+                    1405.0,
+                    1468.0,
+                    294.0,
+                    1468.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1464.0,
+                    1404.0,
+                    1464.0,
+                    1404.0,
+                    1501.0,
+                    293.0,
+                    1501.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1497.0,
+                    1186.0,
+                    1497.0,
+                    1186.0,
+                    1530.0,
+                    293.0,
+                    1530.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1829.0,
+                    1404.0,
+                    1829.0,
+                    1404.0,
+                    1863.0,
+                    296.0,
+                    1863.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1862.0,
+                    1403.0,
+                    1862.0,
+                    1403.0,
+                    1892.0,
+                    296.0,
+                    1892.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    1891.0,
+                    1403.0,
+                    1891.0,
+                    1403.0,
+                    1924.0,
+                    297.0,
+                    1924.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    677.0,
+                    1405.0,
+                    677.0,
+                    1405.0,
+                    713.0,
+                    296.0,
+                    713.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    708.0,
+                    1203.0,
+                    708.0,
+                    1203.0,
+                    744.0,
+                    295.0,
+                    744.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    734.0,
+                    774.0,
+                    1296.0,
+                    774.0,
+                    1296.0,
+                    814.0,
+                    734.0,
+                    814.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    733.0,
+                    807.0,
+                    839.0,
+                    807.0,
+                    839.0,
+                    844.0,
+                    733.0,
+                    844.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    733.0,
+                    847.0,
+                    1300.0,
+                    847.0,
+                    1300.0,
+                    890.0,
+                    733.0,
+                    890.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    733.0,
+                    893.0,
+                    1282.0,
+                    893.0,
+                    1282.0,
+                    929.0,
+                    733.0,
+                    929.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    732.0,
+                    934.0,
+                    1298.0,
+                    934.0,
+                    1298.0,
+                    972.0,
+                    732.0,
+                    972.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    490.0,
+                    1405.0,
+                    490.0,
+                    1405.0,
+                    527.0,
+                    291.0,
+                    527.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    521.0,
+                    1406.0,
+                    521.0,
+                    1406.0,
+                    556.0,
+                    293.0,
+                    556.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    554.0,
+                    1405.0,
+                    554.0,
+                    1405.0,
+                    587.0,
+                    296.0,
+                    587.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    586.0,
+                    634.0,
+                    586.0,
+                    634.0,
+                    619.0,
+                    295.0,
+                    619.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 5,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1252,
+                    1404,
+                    1252,
+                    1404,
+                    1526,
+                    298,
+                    1526
+                ],
+                "score": 0.983
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1793,
+                    1404,
+                    1793,
+                    1404,
+                    2008,
+                    298,
+                    2008
+                ],
+                "score": 0.98
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1595,
+                    1404,
+                    1595,
+                    1404,
+                    1780,
+                    297,
+                    1780
+                ],
+                "score": 0.978
+            },
+            {
+                "category_id": 3,
+                "poly": [
+                    295,
+                    192,
+                    1403,
+                    192,
+                    1403,
+                    724,
+                    295,
+                    724
+                ],
+                "score": 0.974
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    1015,
+                    1404,
+                    1015,
+                    1404,
+                    1107,
+                    299,
+                    1107
+                ],
+                "score": 0.97
+            },
+            {
+                "category_id": 4,
+                "poly": [
+                    297,
+                    738,
+                    1405,
+                    738,
+                    1405,
+                    922,
+                    297,
+                    922
+                ],
+                "score": 0.965
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1122,
+                    1402,
+                    1122,
+                    1402,
+                    1184,
+                    298,
+                    1184
+                ],
+                "score": 0.948
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    842,
+                    2061,
+                    858,
+                    2061,
+                    858,
+                    2084,
+                    842,
+                    2084
+                ],
+                "score": 0.695
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    842,
+                    2061,
+                    859,
+                    2061,
+                    859,
+                    2084,
+                    842,
+                    2084
+                ],
+                "score": 0.124
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1152,
+                    1626,
+                    1204,
+                    1626,
+                    1204,
+                    1656,
+                    1152,
+                    1656
+                ],
+                "score": 0.9,
+                "latex": "N \\%"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1273,
+                    1855,
+                    1328,
+                    1855,
+                    1328,
+                    1884,
+                    1273,
+                    1884
+                ],
+                "score": 0.88,
+                "latex": "50 \\%"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    730,
+                    1915,
+                    786,
+                    1915,
+                    786,
+                    1944,
+                    730,
+                    1944
+                ],
+                "score": 0.87,
+                "latex": "2 5 \\%"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1001,
+                    1915,
+                    1057,
+                    1915,
+                    1057,
+                    1944,
+                    1001,
+                    1944
+                ],
+                "score": 0.87,
+                "latex": "42 \\%"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1313,
+                    1885,
+                    1367,
+                    1885,
+                    1367,
+                    1914,
+                    1313,
+                    1914
+                ],
+                "score": 0.87,
+                "latex": "37 \\%"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1100,
+                    1946,
+                    1153,
+                    1946,
+                    1153,
+                    1975,
+                    1100,
+                    1975
+                ],
+                "score": 0.87,
+                "latex": "13 \\%"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1350,
+                    1946,
+                    1403,
+                    1946,
+                    1403,
+                    1975,
+                    1350,
+                    1975
+                ],
+                "score": 0.86,
+                "latex": "10 \\%"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    997,
+                    832,
+                    1051,
+                    832,
+                    1051,
+                    861,
+                    997,
+                    861
+                ],
+                "score": 0.81,
+                "latex": "\\left( \\mathrm { V } + \\right)"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    895,
+                    1916,
+                    951,
+                    1916,
+                    951,
+                    1946,
+                    895,
+                    1946
+                ],
+                "score": 0.8,
+                "latex": "( \\mathrm { V } + )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    959,
+                    862,
+                    1067,
+                    862,
+                    1067,
+                    890,
+                    959,
+                    890
+                ],
+                "score": 0.78,
+                "latex": "\\mathrm { R 3 D + N L }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    556,
+                    1824,
+                    671,
+                    1824,
+                    671,
+                    1854,
+                    556,
+                    1854
+                ],
+                "score": 0.78,
+                "latex": "50 \\% - 6 0 \\%"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    415.0,
+                    192.0,
+                    533.0,
+                    192.0,
+                    533.0,
+                    224.0,
+                    415.0,
+                    224.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    675.0,
+                    194.0,
+                    799.0,
+                    194.0,
+                    799.0,
+                    225.0,
+                    675.0,
+                    225.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    949.0,
+                    192.0,
+                    1053.0,
+                    192.0,
+                    1053.0,
+                    224.0,
+                    949.0,
+                    224.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1189.0,
+                    195.0,
+                    1336.0,
+                    195.0,
+                    1336.0,
+                    222.0,
+                    1189.0,
+                    222.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    314.0,
+                    209.0,
+                    349.0,
+                    209.0,
+                    349.0,
+                    232.0,
+                    314.0,
+                    232.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    537.0,
+                    233.0,
+                    583.0,
+                    233.0,
+                    583.0,
+                    251.0,
+                    537.0,
+                    251.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    840.0,
+                    228.0,
+                    863.0,
+                    228.0,
+                    863.0,
+                    242.0,
+                    840.0,
+                    242.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1143.0,
+                    225.0,
+                    1314.0,
+                    225.0,
+                    1314.0,
+                    252.0,
+                    1143.0,
+                    252.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    314.0,
+                    247.0,
+                    351.0,
+                    247.0,
+                    351.0,
+                    274.0,
+                    314.0,
+                    274.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1159.0,
+                    249.0,
+                    1169.0,
+                    249.0,
+                    1169.0,
+                    259.0,
+                    1159.0,
+                    259.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1180.0,
+                    246.0,
+                    1327.0,
+                    246.0,
+                    1327.0,
+                    273.0,
+                    1180.0,
+                    273.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    298.0,
+                    276.0,
+                    349.0,
+                    276.0,
+                    349.0,
+                    370.0,
+                    298.0,
+                    370.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    623.0,
+                    297.0,
+                    786.0,
+                    297.0,
+                    786.0,
+                    325.0,
+                    623.0,
+                    325.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    313.0,
+                    327.0,
+                    349.0,
+                    327.0,
+                    349.0,
+                    357.0,
+                    313.0,
+                    357.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    367.0,
+                    327.0,
+                    378.0,
+                    327.0,
+                    378.0,
+                    338.0,
+                    367.0,
+                    338.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    648.0,
+                    319.0,
+                    800.0,
+                    319.0,
+                    800.0,
+                    346.0,
+                    648.0,
+                    346.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    313.0,
+                    370.0,
+                    351.0,
+                    370.0,
+                    351.0,
+                    398.0,
+                    313.0,
+                    398.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    353.0,
+                    372.0,
+                    523.0,
+                    372.0,
+                    523.0,
+                    395.0,
+                    353.0,
+                    395.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    916.0,
+                    373.0,
+                    1079.0,
+                    373.0,
+                    1079.0,
+                    396.0,
+                    916.0,
+                    396.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    362.0,
+                    395.0,
+                    380.0,
+                    395.0,
+                    380.0,
+                    409.0,
+                    362.0,
+                    409.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    389.0,
+                    390.0,
+                    536.0,
+                    390.0,
+                    536.0,
+                    417.0,
+                    389.0,
+                    417.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    910.0,
+                    387.0,
+                    1095.0,
+                    387.0,
+                    1095.0,
+                    417.0,
+                    910.0,
+                    417.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    314.0,
+                    412.0,
+                    349.0,
+                    412.0,
+                    349.0,
+                    436.0,
+                    314.0,
+                    436.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    378.0,
+                    421.0,
+                    414.0,
+                    421.0,
+                    414.0,
+                    443.0,
+                    378.0,
+                    443.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    427.0,
+                    421.0,
+                    464.0,
+                    421.0,
+                    464.0,
+                    443.0,
+                    427.0,
+                    443.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    478.0,
+                    421.0,
+                    516.0,
+                    421.0,
+                    516.0,
+                    443.0,
+                    478.0,
+                    443.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    527.0,
+                    421.0,
+                    565.0,
+                    421.0,
+                    565.0,
+                    443.0,
+                    527.0,
+                    443.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    581.0,
+                    421.0,
+                    614.0,
+                    421.0,
+                    614.0,
+                    444.0,
+                    581.0,
+                    444.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    643.0,
+                    421.0,
+                    677.0,
+                    421.0,
+                    677.0,
+                    443.0,
+                    643.0,
+                    443.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    690.0,
+                    421.0,
+                    727.0,
+                    421.0,
+                    727.0,
+                    443.0,
+                    690.0,
+                    443.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    741.0,
+                    421.0,
+                    779.0,
+                    421.0,
+                    779.0,
+                    443.0,
+                    741.0,
+                    443.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    791.0,
+                    421.0,
+                    828.0,
+                    421.0,
+                    828.0,
+                    443.0,
+                    791.0,
+                    443.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    844.0,
+                    421.0,
+                    878.0,
+                    421.0,
+                    878.0,
+                    443.0,
+                    844.0,
+                    443.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    906.0,
+                    421.0,
+                    940.0,
+                    421.0,
+                    940.0,
+                    443.0,
+                    906.0,
+                    443.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    955.0,
+                    420.0,
+                    990.0,
+                    420.0,
+                    990.0,
+                    443.0,
+                    955.0,
+                    443.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1005.0,
+                    421.0,
+                    1042.0,
+                    421.0,
+                    1042.0,
+                    443.0,
+                    1005.0,
+                    443.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1055.0,
+                    421.0,
+                    1091.0,
+                    421.0,
+                    1091.0,
+                    443.0,
+                    1055.0,
+                    443.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1107.0,
+                    421.0,
+                    1141.0,
+                    421.0,
+                    1141.0,
+                    443.0,
+                    1107.0,
+                    443.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1169.0,
+                    421.0,
+                    1203.0,
+                    421.0,
+                    1203.0,
+                    443.0,
+                    1169.0,
+                    443.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1218.0,
+                    421.0,
+                    1253.0,
+                    421.0,
+                    1253.0,
+                    443.0,
+                    1218.0,
+                    443.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1268.0,
+                    421.0,
+                    1303.0,
+                    421.0,
+                    1303.0,
+                    443.0,
+                    1268.0,
+                    443.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1318.0,
+                    421.0,
+                    1354.0,
+                    421.0,
+                    1354.0,
+                    443.0,
+                    1318.0,
+                    443.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1371.0,
+                    421.0,
+                    1404.0,
+                    421.0,
+                    1404.0,
+                    443.0,
+                    1371.0,
+                    443.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    377.0,
+                    436.0,
+                    571.0,
+                    436.0,
+                    571.0,
+                    463.0,
+                    377.0,
+                    463.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    640.0,
+                    436.0,
+                    834.0,
+                    436.0,
+                    834.0,
+                    463.0,
+                    640.0,
+                    463.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    903.0,
+                    436.0,
+                    1097.0,
+                    436.0,
+                    1097.0,
+                    463.0,
+                    903.0,
+                    463.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1167.0,
+                    436.0,
+                    1360.0,
+                    436.0,
+                    1360.0,
+                    463.0,
+                    1167.0,
+                    463.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    423.0,
+                    454.0,
+                    526.0,
+                    454.0,
+                    526.0,
+                    489.0,
+                    423.0,
+                    489.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    649.0,
+                    454.0,
+                    825.0,
+                    454.0,
+                    825.0,
+                    489.0,
+                    649.0,
+                    489.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    950.0,
+                    457.0,
+                    1049.0,
+                    457.0,
+                    1049.0,
+                    485.0,
+                    950.0,
+                    485.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1177.0,
+                    454.0,
+                    1350.0,
+                    454.0,
+                    1350.0,
+                    489.0,
+                    1177.0,
+                    489.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    313.0,
+                    469.0,
+                    351.0,
+                    469.0,
+                    351.0,
+                    498.0,
+                    313.0,
+                    498.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    422.0,
+                    488.0,
+                    583.0,
+                    488.0,
+                    583.0,
+                    515.0,
+                    422.0,
+                    515.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    683.0,
+                    486.0,
+                    847.0,
+                    486.0,
+                    847.0,
+                    516.0,
+                    683.0,
+                    516.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    313.0,
+                    510.0,
+                    351.0,
+                    510.0,
+                    351.0,
+                    538.0,
+                    313.0,
+                    538.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    445.0,
+                    508.0,
+                    596.0,
+                    508.0,
+                    596.0,
+                    536.0,
+                    445.0,
+                    536.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    708.0,
+                    508.0,
+                    858.0,
+                    508.0,
+                    858.0,
+                    536.0,
+                    708.0,
+                    536.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    286.0,
+                    525.0,
+                    365.0,
+                    525.0,
+                    365.0,
+                    644.0,
+                    286.0,
+                    644.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    565.0,
+                    558.0,
+                    580.0,
+                    558.0,
+                    580.0,
+                    567.0,
+                    565.0,
+                    567.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    314.0,
+                    591.0,
+                    349.0,
+                    591.0,
+                    349.0,
+                    618.0,
+                    314.0,
+                    618.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    830.0,
+                    585.0,
+                    852.0,
+                    585.0,
+                    852.0,
+                    597.0,
+                    830.0,
+                    597.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1179.0,
+                    588.0,
+                    1189.0,
+                    588.0,
+                    1189.0,
+                    599.0,
+                    1179.0,
+                    599.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    313.0,
+                    633.0,
+                    349.0,
+                    633.0,
+                    349.0,
+                    660.0,
+                    313.0,
+                    660.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    915.0,
+                    632.0,
+                    1049.0,
+                    632.0,
+                    1049.0,
+                    659.0,
+                    915.0,
+                    659.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1150.0,
+                    633.0,
+                    1314.0,
+                    633.0,
+                    1314.0,
+                    661.0,
+                    1150.0,
+                    661.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    916.0,
+                    656.0,
+                    1063.0,
+                    656.0,
+                    1063.0,
+                    679.0,
+                    916.0,
+                    679.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1180.0,
+                    656.0,
+                    1326.0,
+                    656.0,
+                    1326.0,
+                    679.0,
+                    1180.0,
+                    679.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    315.0,
+                    675.0,
+                    349.0,
+                    675.0,
+                    349.0,
+                    699.0,
+                    315.0,
+                    699.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    378.0,
+                    681.0,
+                    414.0,
+                    681.0,
+                    414.0,
+                    709.0,
+                    378.0,
+                    709.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    429.0,
+                    684.0,
+                    463.0,
+                    684.0,
+                    463.0,
+                    706.0,
+                    429.0,
+                    706.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    478.0,
+                    684.0,
+                    514.0,
+                    684.0,
+                    514.0,
+                    706.0,
+                    478.0,
+                    706.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    528.0,
+                    684.0,
+                    565.0,
+                    684.0,
+                    565.0,
+                    706.0,
+                    528.0,
+                    706.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    581.0,
+                    684.0,
+                    614.0,
+                    684.0,
+                    614.0,
+                    706.0,
+                    581.0,
+                    706.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    643.0,
+                    683.0,
+                    675.0,
+                    683.0,
+                    675.0,
+                    706.0,
+                    643.0,
+                    706.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    692.0,
+                    684.0,
+                    728.0,
+                    684.0,
+                    728.0,
+                    706.0,
+                    692.0,
+                    706.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    741.0,
+                    684.0,
+                    777.0,
+                    684.0,
+                    777.0,
+                    706.0,
+                    741.0,
+                    706.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    791.0,
+                    684.0,
+                    828.0,
+                    684.0,
+                    828.0,
+                    706.0,
+                    791.0,
+                    706.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    845.0,
+                    684.0,
+                    877.0,
+                    684.0,
+                    877.0,
+                    706.0,
+                    845.0,
+                    706.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    904.0,
+                    681.0,
+                    941.0,
+                    681.0,
+                    941.0,
+                    709.0,
+                    904.0,
+                    709.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    955.0,
+                    683.0,
+                    991.0,
+                    683.0,
+                    991.0,
+                    706.0,
+                    955.0,
+                    706.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1005.0,
+                    683.0,
+                    1040.0,
+                    683.0,
+                    1040.0,
+                    706.0,
+                    1005.0,
+                    706.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1055.0,
+                    684.0,
+                    1092.0,
+                    684.0,
+                    1092.0,
+                    706.0,
+                    1055.0,
+                    706.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1108.0,
+                    684.0,
+                    1141.0,
+                    684.0,
+                    1141.0,
+                    706.0,
+                    1108.0,
+                    706.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1169.0,
+                    683.0,
+                    1203.0,
+                    683.0,
+                    1203.0,
+                    706.0,
+                    1169.0,
+                    706.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1218.0,
+                    683.0,
+                    1253.0,
+                    683.0,
+                    1253.0,
+                    706.0,
+                    1218.0,
+                    706.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1268.0,
+                    684.0,
+                    1305.0,
+                    684.0,
+                    1305.0,
+                    706.0,
+                    1268.0,
+                    706.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1318.0,
+                    684.0,
+                    1355.0,
+                    684.0,
+                    1355.0,
+                    706.0,
+                    1318.0,
+                    706.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1371.0,
+                    684.0,
+                    1404.0,
+                    684.0,
+                    1404.0,
+                    706.0,
+                    1371.0,
+                    706.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    377.0,
+                    699.0,
+                    570.0,
+                    699.0,
+                    570.0,
+                    726.0,
+                    377.0,
+                    726.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    638.0,
+                    695.0,
+                    835.0,
+                    695.0,
+                    835.0,
+                    728.0,
+                    638.0,
+                    728.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    901.0,
+                    695.0,
+                    1100.0,
+                    695.0,
+                    1100.0,
+                    728.0,
+                    901.0,
+                    728.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1167.0,
+                    699.0,
+                    1361.0,
+                    699.0,
+                    1361.0,
+                    726.0,
+                    1167.0,
+                    726.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    739.0,
+                    1406.0,
+                    739.0,
+                    1406.0,
+                    774.0,
+                    294.0,
+                    774.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    770.0,
+                    1406.0,
+                    770.0,
+                    1406.0,
+                    806.0,
+                    295.0,
+                    806.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    800.0,
+                    1406.0,
+                    800.0,
+                    1406.0,
+                    835.0,
+                    295.0,
+                    835.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    830.0,
+                    996.0,
+                    830.0,
+                    996.0,
+                    865.0,
+                    292.0,
+                    865.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1052.0,
+                    830.0,
+                    1405.0,
+                    830.0,
+                    1405.0,
+                    865.0,
+                    1052.0,
+                    865.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    860.0,
+                    958.0,
+                    860.0,
+                    958.0,
+                    896.0,
+                    295.0,
+                    896.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1068.0,
+                    860.0,
+                    1406.0,
+                    860.0,
+                    1406.0,
+                    896.0,
+                    1068.0,
+                    896.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    892.0,
+                    607.0,
+                    892.0,
+                    607.0,
+                    928.0,
+                    295.0,
+                    928.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    840.0,
+                    2059.0,
+                    860.0,
+                    2059.0,
+                    860.0,
+                    2092.0,
+                    840.0,
+                    2092.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    840.0,
+                    2060.0,
+                    861.0,
+                    2060.0,
+                    861.0,
+                    2092.0,
+                    840.0,
+                    2092.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1253.0,
+                    1406.0,
+                    1253.0,
+                    1406.0,
+                    1289.0,
+                    294.0,
+                    1289.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1284.0,
+                    1405.0,
+                    1284.0,
+                    1405.0,
+                    1320.0,
+                    294.0,
+                    1320.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1312.0,
+                    1408.0,
+                    1312.0,
+                    1408.0,
+                    1353.0,
+                    293.0,
+                    1353.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1344.0,
+                    1406.0,
+                    1344.0,
+                    1406.0,
+                    1380.0,
+                    294.0,
+                    1380.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1375.0,
+                    1405.0,
+                    1375.0,
+                    1405.0,
+                    1411.0,
+                    294.0,
+                    1411.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1406.0,
+                    1404.0,
+                    1406.0,
+                    1404.0,
+                    1438.0,
+                    294.0,
+                    1438.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1434.0,
+                    1406.0,
+                    1434.0,
+                    1406.0,
+                    1472.0,
+                    293.0,
+                    1472.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1467.0,
+                    1405.0,
+                    1467.0,
+                    1405.0,
+                    1499.0,
+                    296.0,
+                    1499.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1497.0,
+                    616.0,
+                    1497.0,
+                    616.0,
+                    1529.0,
+                    296.0,
+                    1529.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1793.0,
+                    1405.0,
+                    1793.0,
+                    1405.0,
+                    1828.0,
+                    296.0,
+                    1828.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1823.0,
+                    555.0,
+                    1823.0,
+                    555.0,
+                    1859.0,
+                    293.0,
+                    1859.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    672.0,
+                    1823.0,
+                    1406.0,
+                    1823.0,
+                    1406.0,
+                    1859.0,
+                    672.0,
+                    1859.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1853.0,
+                    1272.0,
+                    1853.0,
+                    1272.0,
+                    1891.0,
+                    293.0,
+                    1891.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1329.0,
+                    1853.0,
+                    1406.0,
+                    1853.0,
+                    1406.0,
+                    1891.0,
+                    1329.0,
+                    1891.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1885.0,
+                    1312.0,
+                    1885.0,
+                    1312.0,
+                    1919.0,
+                    294.0,
+                    1919.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1368.0,
+                    1885.0,
+                    1406.0,
+                    1885.0,
+                    1406.0,
+                    1919.0,
+                    1368.0,
+                    1919.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1913.0,
+                    729.0,
+                    1913.0,
+                    729.0,
+                    1951.0,
+                    293.0,
+                    1951.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    787.0,
+                    1913.0,
+                    894.0,
+                    1913.0,
+                    894.0,
+                    1951.0,
+                    787.0,
+                    1951.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    952.0,
+                    1913.0,
+                    1000.0,
+                    1913.0,
+                    1000.0,
+                    1951.0,
+                    952.0,
+                    1951.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1058.0,
+                    1913.0,
+                    1408.0,
+                    1913.0,
+                    1408.0,
+                    1951.0,
+                    1058.0,
+                    1951.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1945.0,
+                    1099.0,
+                    1945.0,
+                    1099.0,
+                    1983.0,
+                    293.0,
+                    1983.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1154.0,
+                    1945.0,
+                    1349.0,
+                    1945.0,
+                    1349.0,
+                    1983.0,
+                    1154.0,
+                    1983.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1404.0,
+                    1945.0,
+                    1409.0,
+                    1945.0,
+                    1409.0,
+                    1983.0,
+                    1404.0,
+                    1983.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1974.0,
+                    738.0,
+                    1974.0,
+                    738.0,
+                    2011.0,
+                    295.0,
+                    2011.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1595.0,
+                    1406.0,
+                    1595.0,
+                    1406.0,
+                    1629.0,
+                    292.0,
+                    1629.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1625.0,
+                    1151.0,
+                    1625.0,
+                    1151.0,
+                    1661.0,
+                    295.0,
+                    1661.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1205.0,
+                    1625.0,
+                    1406.0,
+                    1625.0,
+                    1406.0,
+                    1661.0,
+                    1205.0,
+                    1661.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1656.0,
+                    1405.0,
+                    1656.0,
+                    1405.0,
+                    1693.0,
+                    292.0,
+                    1693.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1687.0,
+                    1407.0,
+                    1687.0,
+                    1407.0,
+                    1723.0,
+                    294.0,
+                    1723.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1719.0,
+                    1405.0,
+                    1719.0,
+                    1405.0,
+                    1751.0,
+                    295.0,
+                    1751.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1750.0,
+                    1170.0,
+                    1750.0,
+                    1170.0,
+                    1782.0,
+                    296.0,
+                    1782.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1017.0,
+                    1404.0,
+                    1017.0,
+                    1404.0,
+                    1049.0,
+                    294.0,
+                    1049.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1047.0,
+                    1405.0,
+                    1047.0,
+                    1405.0,
+                    1080.0,
+                    295.0,
+                    1080.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1074.0,
+                    1237.0,
+                    1074.0,
+                    1237.0,
+                    1111.0,
+                    294.0,
+                    1111.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1119.0,
+                    1403.0,
+                    1119.0,
+                    1403.0,
+                    1158.0,
+                    294.0,
+                    1158.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    1153.0,
+                    1393.0,
+                    1153.0,
+                    1393.0,
+                    1185.0,
+                    297.0,
+                    1185.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 6,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1672,
+                    1405,
+                    1672,
+                    1405,
+                    2006,
+                    298,
+                    2006
+                ],
+                "score": 0.982
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1282,
+                    1405,
+                    1282,
+                    1405,
+                    1556,
+                    298,
+                    1556
+                ],
+                "score": 0.979
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1055,
+                    1403,
+                    1055,
+                    1403,
+                    1268,
+                    298,
+                    1268
+                ],
+                "score": 0.979
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    691,
+                    1405,
+                    691,
+                    1405,
+                    873,
+                    298,
+                    873
+                ],
+                "score": 0.978
+            },
+            {
+                "category_id": 5,
+                "poly": [
+                    252,
+                    192,
+                    1445,
+                    192,
+                    1445,
+                    449,
+                    252,
+                    449
+                ],
+                "score": 0.978,
+                "html": "<table><tr><td>Model</td><td>Top 1 (S)</td><td>Top 5 (S)</td><td>L1(S)</td><td>Top 1 (M)</td><td>Top 5 (M)</td><td>L1 (M)</td></tr><tr><td>R3DLSTM R3D+NL LSTM OPNet</td><td>60.2 46.2 74.8 73.2</td><td>81.8 69.9 1 93.8</td><td>1.2 1.5 0.54 0.85</td><td>28.6 38.6 1 1</td><td>63.3 70.2 1 1</td><td>1.7 1.5 -</td></tr><tr><td>Hopper Aloe (no auxiliary) Aloe Aloe (with L1 loss)</td><td>60.5 70.6 74.0 ± 0.3</td><td>84.5 93.0 94.0 ± 0.4</td><td>0.90 0.53 0.44 ± 0.01</td><td>46.8 56.6 59.7 ± 0.5</td><td>75.1 87.0 90.1 ± 0.6</td><td>- 1.3 0.82 0.69 ± 0.01</td></tr></table>"
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    887,
+                    1404,
+                    887,
+                    1404,
+                    1041,
+                    298,
+                    1041
+                ],
+                "score": 0.976
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    297,
+                    627,
+                    457,
+                    627,
+                    457,
+                    659,
+                    297,
+                    659
+                ],
+                "score": 0.902
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    299,
+                    1609,
+                    440,
+                    1609,
+                    440,
+                    1641,
+                    299,
+                    1641
+                ],
+                "score": 0.897
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    840,
+                    2062,
+                    858,
+                    2062,
+                    858,
+                    2085,
+                    840,
+                    2085
+                ],
+                "score": 0.799
+            },
+            {
+                "category_id": 6,
+                "poly": [
+                    296,
+                    460,
+                    1407,
+                    460,
+                    1407,
+                    554,
+                    296,
+                    554
+                ],
+                "score": 0.492
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    507,
+                    411,
+                    641,
+                    411,
+                    641,
+                    446,
+                    507,
+                    446
+                ],
+                "score": 0.66,
+                "latex": "7 4 . 0 \\pm 0 . 3"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    669,
+                    411,
+                    794,
+                    411,
+                    794,
+                    445,
+                    669,
+                    445
+                ],
+                "score": 0.44,
+                "latex": "{ \\bf 9 4 . 0 \\pm 0 . 4 }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1143,
+                    411,
+                    1269,
+                    411,
+                    1269,
+                    445,
+                    1143,
+                    445
+                ],
+                "score": 0.34,
+                "latex": "{ \\bf 9 0 . 1 \\pm 0 . 6 }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    991,
+                    411,
+                    1116,
+                    411,
+                    1116,
+                    445,
+                    991,
+                    445
+                ],
+                "score": 0.33,
+                "latex": "{ \\bf 5 9 . 7 \\pm 0 . 5 }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1295,
+                    410,
+                    1432,
+                    410,
+                    1432,
+                    445,
+                    1295,
+                    445
+                ],
+                "score": 0.33,
+                "latex": "{ \\bf 0 . 6 9 \\pm 0 . 0 1 }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    820,
+                    410,
+                    958,
+                    410,
+                    958,
+                    445,
+                    820,
+                    445
+                ],
+                "score": 0.28,
+                "latex": "{ \\bf 0 . 4 4 \\pm 0 . 0 1 }"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    624.0,
+                    461.0,
+                    624.0,
+                    461.0,
+                    665.0,
+                    293.0,
+                    665.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1605.0,
+                    445.0,
+                    1605.0,
+                    445.0,
+                    1645.0,
+                    291.0,
+                    1645.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    838.0,
+                    2058.0,
+                    862.0,
+                    2058.0,
+                    862.0,
+                    2090.0,
+                    838.0,
+                    2090.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    455.0,
+                    1405.0,
+                    455.0,
+                    1405.0,
+                    499.0,
+                    292.0,
+                    499.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    491.0,
+                    1405.0,
+                    491.0,
+                    1405.0,
+                    525.0,
+                    293.0,
+                    525.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    522.0,
+                    899.0,
+                    522.0,
+                    899.0,
+                    556.0,
+                    294.0,
+                    556.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1674.0,
+                    1403.0,
+                    1674.0,
+                    1403.0,
+                    1705.0,
+                    296.0,
+                    1705.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1705.0,
+                    1403.0,
+                    1705.0,
+                    1403.0,
+                    1735.0,
+                    296.0,
+                    1735.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1732.0,
+                    1410.0,
+                    1732.0,
+                    1410.0,
+                    1770.0,
+                    295.0,
+                    1770.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1764.0,
+                    1406.0,
+                    1764.0,
+                    1406.0,
+                    1798.0,
+                    295.0,
+                    1798.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1790.0,
+                    1407.0,
+                    1790.0,
+                    1407.0,
+                    1829.0,
+                    292.0,
+                    1829.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1824.0,
+                    1406.0,
+                    1824.0,
+                    1406.0,
+                    1856.0,
+                    293.0,
+                    1856.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1854.0,
+                    1406.0,
+                    1854.0,
+                    1406.0,
+                    1888.0,
+                    293.0,
+                    1888.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1885.0,
+                    1407.0,
+                    1885.0,
+                    1407.0,
+                    1919.0,
+                    295.0,
+                    1919.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1915.0,
+                    1407.0,
+                    1915.0,
+                    1407.0,
+                    1947.0,
+                    292.0,
+                    1947.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1946.0,
+                    1405.0,
+                    1946.0,
+                    1405.0,
+                    1981.0,
+                    295.0,
+                    1981.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1977.0,
+                    522.0,
+                    1977.0,
+                    522.0,
+                    2008.0,
+                    296.0,
+                    2008.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1281.0,
+                    1405.0,
+                    1281.0,
+                    1405.0,
+                    1318.0,
+                    296.0,
+                    1318.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1314.0,
+                    1403.0,
+                    1314.0,
+                    1403.0,
+                    1346.0,
+                    296.0,
+                    1346.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1342.0,
+                    1410.0,
+                    1342.0,
+                    1410.0,
+                    1379.0,
+                    295.0,
+                    1379.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1374.0,
+                    1408.0,
+                    1374.0,
+                    1408.0,
+                    1410.0,
+                    293.0,
+                    1410.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1399.0,
+                    1406.0,
+                    1399.0,
+                    1406.0,
+                    1442.0,
+                    292.0,
+                    1442.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1432.0,
+                    1406.0,
+                    1432.0,
+                    1406.0,
+                    1470.0,
+                    293.0,
+                    1470.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1459.0,
+                    1405.0,
+                    1459.0,
+                    1405.0,
+                    1503.0,
+                    293.0,
+                    1503.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1496.0,
+                    1402.0,
+                    1496.0,
+                    1402.0,
+                    1528.0,
+                    296.0,
+                    1528.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1524.0,
+                    1403.0,
+                    1524.0,
+                    1403.0,
+                    1561.0,
+                    295.0,
+                    1561.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1051.0,
+                    1404.0,
+                    1051.0,
+                    1404.0,
+                    1093.0,
+                    292.0,
+                    1093.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1086.0,
+                    1405.0,
+                    1086.0,
+                    1405.0,
+                    1120.0,
+                    293.0,
+                    1120.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1113.0,
+                    1407.0,
+                    1113.0,
+                    1407.0,
+                    1152.0,
+                    292.0,
+                    1152.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1142.0,
+                    1404.0,
+                    1142.0,
+                    1404.0,
+                    1184.0,
+                    292.0,
+                    1184.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1175.0,
+                    1404.0,
+                    1175.0,
+                    1404.0,
+                    1213.0,
+                    293.0,
+                    1213.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1206.0,
+                    1405.0,
+                    1206.0,
+                    1405.0,
+                    1242.0,
+                    293.0,
+                    1242.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1234.0,
+                    1076.0,
+                    1234.0,
+                    1076.0,
+                    1275.0,
+                    293.0,
+                    1275.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    693.0,
+                    1406.0,
+                    693.0,
+                    1406.0,
+                    724.0,
+                    295.0,
+                    724.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    718.0,
+                    1410.0,
+                    718.0,
+                    1410.0,
+                    759.0,
+                    292.0,
+                    759.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    751.0,
+                    1406.0,
+                    751.0,
+                    1406.0,
+                    786.0,
+                    295.0,
+                    786.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    782.0,
+                    1405.0,
+                    782.0,
+                    1405.0,
+                    817.0,
+                    295.0,
+                    817.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    811.0,
+                    1408.0,
+                    811.0,
+                    1408.0,
+                    847.0,
+                    293.0,
+                    847.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    844.0,
+                    919.0,
+                    844.0,
+                    919.0,
+                    875.0,
+                    296.0,
+                    875.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    887.0,
+                    1406.0,
+                    887.0,
+                    1406.0,
+                    924.0,
+                    296.0,
+                    924.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    918.0,
+                    1406.0,
+                    918.0,
+                    1406.0,
+                    952.0,
+                    293.0,
+                    952.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    948.0,
+                    1406.0,
+                    948.0,
+                    1406.0,
+                    984.0,
+                    293.0,
+                    984.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    979.0,
+                    1405.0,
+                    979.0,
+                    1405.0,
+                    1014.0,
+                    293.0,
+                    1014.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1010.0,
+                    502.0,
+                    1010.0,
+                    502.0,
+                    1045.0,
+                    296.0,
+                    1045.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 7,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1190,
+                    1404,
+                    1190,
+                    1404,
+                    1524,
+                    298,
+                    1524
+                ],
+                "score": 0.984
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    697,
+                    1405,
+                    697,
+                    1405,
+                    1062,
+                    298,
+                    1062
+                ],
+                "score": 0.982
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1732,
+                    1404,
+                    1732,
+                    1404,
+                    2007,
+                    298,
+                    2007
+                ],
+                "score": 0.98
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    1567,
+                    1403,
+                    1567,
+                    1403,
+                    1689,
+                    299,
+                    1689
+                ],
+                "score": 0.976
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    500,
+                    1403,
+                    500,
+                    1403,
+                    684,
+                    298,
+                    684
+                ],
+                "score": 0.976
+            },
+            {
+                "category_id": 5,
+                "poly": [
+                    267,
+                    192,
+                    1433,
+                    192,
+                    1433,
+                    327,
+                    267,
+                    327
+                ],
+                "score": 0.959,
+                "html": "<table><tr><td>Model</td><td>All (C)</td><td>D.R.</td><td>I.D.</td><td>S.0.</td><td>B.B.</td><td>All (S)</td><td>D.R.</td><td>1.D.</td><td>S.0.</td><td>B.B</td></tr><tr><td>CNN-BERT</td><td>43.79</td><td>54.07</td><td>46.88</td><td>40.57</td><td>28.79</td><td>39.93</td><td>55.97</td><td>68.25</td><td>0.00</td><td>45.59</td></tr><tr><td>NS-OPT</td><td>69.04</td><td>92.5</td><td>76.05</td><td>88.33</td><td>13.48</td><td>67.44</td><td>94.73</td><td>88.38</td><td>82.76</td><td>16.06</td></tr><tr><td>Aloe</td><td>91.76</td><td>97.14</td><td>90.8</td><td>96.8</td><td>78.81</td><td>93.90</td><td>97.18</td><td>71.24</td><td>98.97</td><td>94.48</td></tr></table>"
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    299,
+                    1115,
+                    540,
+                    1115,
+                    540,
+                    1152,
+                    299,
+                    1152
+                ],
+                "score": 0.915
+            },
+            {
+                "category_id": 6,
+                "poly": [
+                    298,
+                    339,
+                    1407,
+                    339,
+                    1407,
+                    433,
+                    298,
+                    433
+                ],
+                "score": 0.892
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    840,
+                    2061,
+                    858,
+                    2061,
+                    858,
+                    2084,
+                    840,
+                    2084
+                ],
+                "score": 0.798
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    735,
+                    401,
+                    770,
+                    401,
+                    770,
+                    430,
+                    735,
+                    430
+                ],
+                "score": 0.66,
+                "latex": "{ \\bf S } ="
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    526,
+                    402,
+                    562,
+                    402,
+                    562,
+                    429,
+                    526,
+                    429
+                ],
+                "score": 0.65,
+                "latex": "\\mathrm { C } ="
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1091,
+                    373,
+                    1111,
+                    373,
+                    1111,
+                    396,
+                    1091,
+                    396
+                ],
+                "score": 0.43,
+                "latex": "\\ c ="
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    733,
+                    374,
+                    751,
+                    374,
+                    751,
+                    395,
+                    733,
+                    395
+                ],
+                "score": 0.33,
+                "latex": "\\Vdash"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    549,
+                    371,
+                    618,
+                    371,
+                    618,
+                    399,
+                    549,
+                    399
+                ],
+                "score": 0.32,
+                "latex": "\\mathrm { { D . R = } } 1"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    849,
+                    370,
+                    914,
+                    370,
+                    914,
+                    399,
+                    849,
+                    399
+                ],
+                "score": 0.31,
+                "latex": "{ \\bf { S . O } } \\mathrm { { = } } \\mathrm { { ; } }"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1110.0,
+                    546.0,
+                    1110.0,
+                    546.0,
+                    1157.0,
+                    292.0,
+                    1157.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    337.0,
+                    1405.0,
+                    337.0,
+                    1405.0,
+                    375.0,
+                    293.0,
+                    375.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    369.0,
+                    548.0,
+                    369.0,
+                    548.0,
+                    405.0,
+                    292.0,
+                    405.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    619.0,
+                    369.0,
+                    732.0,
+                    369.0,
+                    732.0,
+                    405.0,
+                    619.0,
+                    405.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    752.0,
+                    369.0,
+                    848.0,
+                    369.0,
+                    848.0,
+                    405.0,
+                    752.0,
+                    405.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    915.0,
+                    369.0,
+                    1090.0,
+                    369.0,
+                    1090.0,
+                    405.0,
+                    915.0,
+                    405.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1112.0,
+                    369.0,
+                    1407.0,
+                    369.0,
+                    1407.0,
+                    405.0,
+                    1112.0,
+                    405.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    402.0,
+                    525.0,
+                    402.0,
+                    525.0,
+                    434.0,
+                    295.0,
+                    434.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    563.0,
+                    402.0,
+                    734.0,
+                    402.0,
+                    734.0,
+                    434.0,
+                    563.0,
+                    434.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    771.0,
+                    402.0,
+                    910.0,
+                    402.0,
+                    910.0,
+                    434.0,
+                    771.0,
+                    434.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    839.0,
+                    2060.0,
+                    860.0,
+                    2060.0,
+                    860.0,
+                    2090.0,
+                    839.0,
+                    2090.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1189.0,
+                    1404.0,
+                    1189.0,
+                    1404.0,
+                    1225.0,
+                    293.0,
+                    1225.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1221.0,
+                    1405.0,
+                    1221.0,
+                    1405.0,
+                    1256.0,
+                    294.0,
+                    1256.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1250.0,
+                    1404.0,
+                    1250.0,
+                    1404.0,
+                    1288.0,
+                    293.0,
+                    1288.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1281.0,
+                    1405.0,
+                    1281.0,
+                    1405.0,
+                    1318.0,
+                    293.0,
+                    1318.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1310.0,
+                    1406.0,
+                    1310.0,
+                    1406.0,
+                    1348.0,
+                    293.0,
+                    1348.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1343.0,
+                    1406.0,
+                    1343.0,
+                    1406.0,
+                    1378.0,
+                    294.0,
+                    1378.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1373.0,
+                    1404.0,
+                    1373.0,
+                    1404.0,
+                    1408.0,
+                    293.0,
+                    1408.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1403.0,
+                    1404.0,
+                    1403.0,
+                    1404.0,
+                    1437.0,
+                    293.0,
+                    1437.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1436.0,
+                    1405.0,
+                    1436.0,
+                    1405.0,
+                    1467.0,
+                    296.0,
+                    1467.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1463.0,
+                    1406.0,
+                    1463.0,
+                    1406.0,
+                    1500.0,
+                    292.0,
+                    1500.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1496.0,
+                    693.0,
+                    1496.0,
+                    693.0,
+                    1527.0,
+                    296.0,
+                    1527.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    698.0,
+                    1405.0,
+                    698.0,
+                    1405.0,
+                    732.0,
+                    292.0,
+                    732.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    727.0,
+                    1405.0,
+                    727.0,
+                    1405.0,
+                    764.0,
+                    292.0,
+                    764.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    758.0,
+                    1405.0,
+                    758.0,
+                    1405.0,
+                    796.0,
+                    295.0,
+                    796.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    789.0,
+                    1406.0,
+                    789.0,
+                    1406.0,
+                    825.0,
+                    293.0,
+                    825.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    821.0,
+                    1402.0,
+                    821.0,
+                    1402.0,
+                    852.0,
+                    296.0,
+                    852.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    849.0,
+                    1405.0,
+                    849.0,
+                    1405.0,
+                    883.0,
+                    295.0,
+                    883.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    881.0,
+                    1408.0,
+                    881.0,
+                    1408.0,
+                    913.0,
+                    293.0,
+                    913.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    910.0,
+                    1406.0,
+                    910.0,
+                    1406.0,
+                    947.0,
+                    293.0,
+                    947.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    938.0,
+                    1410.0,
+                    938.0,
+                    1410.0,
+                    976.0,
+                    292.0,
+                    976.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    968.0,
+                    1407.0,
+                    968.0,
+                    1407.0,
+                    1007.0,
+                    292.0,
+                    1007.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1002.0,
+                    1405.0,
+                    1002.0,
+                    1405.0,
+                    1036.0,
+                    296.0,
+                    1036.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1032.0,
+                    1309.0,
+                    1032.0,
+                    1309.0,
+                    1066.0,
+                    295.0,
+                    1066.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1733.0,
+                    1404.0,
+                    1733.0,
+                    1404.0,
+                    1769.0,
+                    294.0,
+                    1769.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1765.0,
+                    1406.0,
+                    1765.0,
+                    1406.0,
+                    1798.0,
+                    296.0,
+                    1798.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1794.0,
+                    1405.0,
+                    1794.0,
+                    1405.0,
+                    1829.0,
+                    293.0,
+                    1829.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1823.0,
+                    1405.0,
+                    1823.0,
+                    1405.0,
+                    1859.0,
+                    294.0,
+                    1859.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1853.0,
+                    1405.0,
+                    1853.0,
+                    1405.0,
+                    1890.0,
+                    293.0,
+                    1890.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1884.0,
+                    1405.0,
+                    1884.0,
+                    1405.0,
+                    1919.0,
+                    293.0,
+                    1919.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1910.0,
+                    1405.0,
+                    1910.0,
+                    1405.0,
+                    1953.0,
+                    292.0,
+                    1953.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1946.0,
+                    1405.0,
+                    1946.0,
+                    1405.0,
+                    1979.0,
+                    296.0,
+                    1979.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1975.0,
+                    1194.0,
+                    1975.0,
+                    1194.0,
+                    2011.0,
+                    294.0,
+                    2011.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1567.0,
+                    1404.0,
+                    1567.0,
+                    1404.0,
+                    1603.0,
+                    295.0,
+                    1603.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1599.0,
+                    1404.0,
+                    1599.0,
+                    1404.0,
+                    1631.0,
+                    294.0,
+                    1631.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1630.0,
+                    1405.0,
+                    1630.0,
+                    1405.0,
+                    1663.0,
+                    295.0,
+                    1663.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1659.0,
+                    1350.0,
+                    1659.0,
+                    1350.0,
+                    1692.0,
+                    294.0,
+                    1692.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    499.0,
+                    1407.0,
+                    499.0,
+                    1407.0,
+                    539.0,
+                    293.0,
+                    539.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    530.0,
+                    1405.0,
+                    530.0,
+                    1405.0,
+                    567.0,
+                    293.0,
+                    567.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    556.0,
+                    1407.0,
+                    556.0,
+                    1407.0,
+                    601.0,
+                    292.0,
+                    601.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    588.0,
+                    1407.0,
+                    588.0,
+                    1407.0,
+                    631.0,
+                    292.0,
+                    631.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    624.0,
+                    1407.0,
+                    624.0,
+                    1407.0,
+                    655.0,
+                    296.0,
+                    655.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    650.0,
+                    858.0,
+                    650.0,
+                    858.0,
+                    688.0,
+                    292.0,
+                    688.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 8,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1301,
+                    1404,
+                    1301,
+                    1404,
+                    1665,
+                    298,
+                    1665
+                ],
+                "score": 0.986
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    430,
+                    1403,
+                    430,
+                    1403,
+                    644,
+                    298,
+                    644
+                ],
+                "score": 0.98
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    755,
+                    1405,
+                    755,
+                    1405,
+                    999,
+                    298,
+                    999
+                ],
+                "score": 0.98
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1013,
+                    1404,
+                    1013,
+                    1404,
+                    1287,
+                    298,
+                    1287
+                ],
+                "score": 0.979
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    203,
+                    1403,
+                    203,
+                    1403,
+                    416,
+                    298,
+                    416
+                ],
+                "score": 0.979
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    298,
+                    685,
+                    509,
+                    685,
+                    509,
+                    723,
+                    298,
+                    723
+                ],
+                "score": 0.903
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    836,
+                    2061,
+                    865,
+                    2061,
+                    865,
+                    2085,
+                    836,
+                    2085
+                ],
+                "score": 0.849
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    289.0,
+                    680.0,
+                    515.0,
+                    680.0,
+                    515.0,
+                    731.0,
+                    289.0,
+                    731.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    832.0,
+                    2057.0,
+                    871.0,
+                    2057.0,
+                    871.0,
+                    2096.0,
+                    832.0,
+                    2096.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1302.0,
+                    1405.0,
+                    1302.0,
+                    1405.0,
+                    1336.0,
+                    293.0,
+                    1336.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1332.0,
+                    1406.0,
+                    1332.0,
+                    1406.0,
+                    1366.0,
+                    294.0,
+                    1366.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1362.0,
+                    1408.0,
+                    1362.0,
+                    1408.0,
+                    1396.0,
+                    294.0,
+                    1396.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1389.0,
+                    1405.0,
+                    1389.0,
+                    1405.0,
+                    1429.0,
+                    292.0,
+                    1429.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1421.0,
+                    1406.0,
+                    1421.0,
+                    1406.0,
+                    1458.0,
+                    293.0,
+                    1458.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1453.0,
+                    1405.0,
+                    1453.0,
+                    1405.0,
+                    1487.0,
+                    294.0,
+                    1487.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1481.0,
+                    1406.0,
+                    1481.0,
+                    1406.0,
+                    1519.0,
+                    292.0,
+                    1519.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1516.0,
+                    1404.0,
+                    1516.0,
+                    1404.0,
+                    1546.0,
+                    296.0,
+                    1546.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1541.0,
+                    1405.0,
+                    1541.0,
+                    1405.0,
+                    1583.0,
+                    292.0,
+                    1583.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1572.0,
+                    1406.0,
+                    1572.0,
+                    1406.0,
+                    1610.0,
+                    293.0,
+                    1610.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1604.0,
+                    1408.0,
+                    1604.0,
+                    1408.0,
+                    1641.0,
+                    292.0,
+                    1641.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1632.0,
+                    383.0,
+                    1632.0,
+                    383.0,
+                    1670.0,
+                    293.0,
+                    1670.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    425.0,
+                    1405.0,
+                    425.0,
+                    1405.0,
+                    468.0,
+                    292.0,
+                    468.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    458.0,
+                    1405.0,
+                    458.0,
+                    1405.0,
+                    498.0,
+                    291.0,
+                    498.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    492.0,
+                    1405.0,
+                    492.0,
+                    1405.0,
+                    526.0,
+                    293.0,
+                    526.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    521.0,
+                    1407.0,
+                    521.0,
+                    1407.0,
+                    557.0,
+                    292.0,
+                    557.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    552.0,
+                    1403.0,
+                    552.0,
+                    1403.0,
+                    586.0,
+                    294.0,
+                    586.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    581.0,
+                    1404.0,
+                    581.0,
+                    1404.0,
+                    618.0,
+                    293.0,
+                    618.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    613.0,
+                    1030.0,
+                    613.0,
+                    1030.0,
+                    647.0,
+                    294.0,
+                    647.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    755.0,
+                    1406.0,
+                    755.0,
+                    1406.0,
+                    789.0,
+                    296.0,
+                    789.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    786.0,
+                    1408.0,
+                    786.0,
+                    1408.0,
+                    820.0,
+                    296.0,
+                    820.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    815.0,
+                    1407.0,
+                    815.0,
+                    1407.0,
+                    851.0,
+                    293.0,
+                    851.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    847.0,
+                    1407.0,
+                    847.0,
+                    1407.0,
+                    881.0,
+                    293.0,
+                    881.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    877.0,
+                    1406.0,
+                    877.0,
+                    1406.0,
+                    910.0,
+                    295.0,
+                    910.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    907.0,
+                    1405.0,
+                    907.0,
+                    1405.0,
+                    940.0,
+                    295.0,
+                    940.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    937.0,
+                    1405.0,
+                    937.0,
+                    1405.0,
+                    973.0,
+                    293.0,
+                    973.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    970.0,
+                    527.0,
+                    970.0,
+                    527.0,
+                    1001.0,
+                    296.0,
+                    1001.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1012.0,
+                    1404.0,
+                    1012.0,
+                    1404.0,
+                    1049.0,
+                    296.0,
+                    1049.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1041.0,
+                    1406.0,
+                    1041.0,
+                    1406.0,
+                    1081.0,
+                    293.0,
+                    1081.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1073.0,
+                    1406.0,
+                    1073.0,
+                    1406.0,
+                    1110.0,
+                    294.0,
+                    1110.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1102.0,
+                    1405.0,
+                    1102.0,
+                    1405.0,
+                    1138.0,
+                    292.0,
+                    1138.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1130.0,
+                    1406.0,
+                    1130.0,
+                    1406.0,
+                    1173.0,
+                    291.0,
+                    1173.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1163.0,
+                    1404.0,
+                    1163.0,
+                    1404.0,
+                    1201.0,
+                    293.0,
+                    1201.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1193.0,
+                    1405.0,
+                    1193.0,
+                    1405.0,
+                    1230.0,
+                    292.0,
+                    1230.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1224.0,
+                    1405.0,
+                    1224.0,
+                    1405.0,
+                    1260.0,
+                    294.0,
+                    1260.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1254.0,
+                    823.0,
+                    1254.0,
+                    823.0,
+                    1289.0,
+                    295.0,
+                    1289.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    199.0,
+                    1404.0,
+                    199.0,
+                    1404.0,
+                    241.0,
+                    293.0,
+                    241.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    234.0,
+                    1405.0,
+                    234.0,
+                    1405.0,
+                    268.0,
+                    294.0,
+                    268.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    264.0,
+                    1405.0,
+                    264.0,
+                    1405.0,
+                    298.0,
+                    294.0,
+                    298.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    294.0,
+                    1407.0,
+                    294.0,
+                    1407.0,
+                    333.0,
+                    292.0,
+                    333.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    322.0,
+                    1407.0,
+                    322.0,
+                    1407.0,
+                    360.0,
+                    293.0,
+                    360.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    355.0,
+                    1405.0,
+                    355.0,
+                    1405.0,
+                    388.0,
+                    293.0,
+                    388.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    385.0,
+                    1304.0,
+                    385.0,
+                    1304.0,
+                    420.0,
+                    294.0,
+                    420.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 9,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 0,
+                "poly": [
+                    299,
+                    200,
+                    455,
+                    200,
+                    455,
+                    235,
+                    299,
+                    235
+                ],
+                "score": 0.897
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    295,
+                    169,
+                    1410,
+                    169,
+                    1410,
+                    2019,
+                    295,
+                    2019
+                ],
+                "score": 0.783
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    836,
+                    2061,
+                    863,
+                    2061,
+                    863,
+                    2086,
+                    836,
+                    2086
+                ],
+                "score": 0.704
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    836,
+                    2061,
+                    863,
+                    2061,
+                    863,
+                    2086,
+                    836,
+                    2086
+                ],
+                "score": 0.301
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    196.0,
+                    460.0,
+                    196.0,
+                    460.0,
+                    241.0,
+                    295.0,
+                    241.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    832.0,
+                    2058.0,
+                    868.0,
+                    2058.0,
+                    868.0,
+                    2098.0,
+                    832.0,
+                    2098.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    832.0,
+                    2058.0,
+                    868.0,
+                    2058.0,
+                    868.0,
+                    2098.0,
+                    832.0,
+                    2098.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    196.0,
+                    463.0,
+                    196.0,
+                    463.0,
+                    241.0,
+                    294.0,
+                    241.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    305.0,
+                    249.0,
+                    1408.0,
+                    249.0,
+                    1408.0,
+                    288.0,
+                    305.0,
+                    288.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    353.0,
+                    279.0,
+                    1406.0,
+                    279.0,
+                    1406.0,
+                    318.0,
+                    353.0,
+                    318.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    353.0,
+                    310.0,
+                    1016.0,
+                    310.0,
+                    1016.0,
+                    349.0,
+                    353.0,
+                    349.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    305.0,
+                    357.0,
+                    1410.0,
+                    357.0,
+                    1410.0,
+                    401.0,
+                    305.0,
+                    401.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    353.0,
+                    389.0,
+                    1406.0,
+                    389.0,
+                    1406.0,
+                    428.0,
+                    353.0,
+                    428.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    349.0,
+                    418.0,
+                    744.0,
+                    418.0,
+                    744.0,
+                    458.0,
+                    349.0,
+                    458.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    305.0,
+                    468.0,
+                    1406.0,
+                    468.0,
+                    1406.0,
+                    503.0,
+                    305.0,
+                    503.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    353.0,
+                    499.0,
+                    1290.0,
+                    499.0,
+                    1290.0,
+                    537.0,
+                    353.0,
+                    537.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    307.0,
+                    546.0,
+                    1406.0,
+                    546.0,
+                    1406.0,
+                    584.0,
+                    307.0,
+                    584.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    351.0,
+                    576.0,
+                    1360.0,
+                    576.0,
+                    1360.0,
+                    615.0,
+                    351.0,
+                    615.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    307.0,
+                    625.0,
+                    1408.0,
+                    625.0,
+                    1408.0,
+                    663.0,
+                    307.0,
+                    663.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    353.0,
+                    655.0,
+                    1334.0,
+                    655.0,
+                    1334.0,
+                    694.0,
+                    353.0,
+                    694.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    305.0,
+                    704.0,
+                    1408.0,
+                    704.0,
+                    1408.0,
+                    741.0,
+                    305.0,
+                    741.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    351.0,
+                    732.0,
+                    1404.0,
+                    732.0,
+                    1404.0,
+                    771.0,
+                    351.0,
+                    771.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    351.0,
+                    761.0,
+                    1414.0,
+                    761.0,
+                    1414.0,
+                    806.0,
+                    351.0,
+                    806.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    353.0,
+                    795.0,
+                    868.0,
+                    795.0,
+                    868.0,
+                    834.0,
+                    353.0,
+                    834.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    305.0,
+                    842.0,
+                    1408.0,
+                    842.0,
+                    1408.0,
+                    881.0,
+                    305.0,
+                    881.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    351.0,
+                    871.0,
+                    1406.0,
+                    871.0,
+                    1406.0,
+                    915.0,
+                    351.0,
+                    915.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    353.0,
+                    903.0,
+                    908.0,
+                    903.0,
+                    908.0,
+                    942.0,
+                    353.0,
+                    942.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    307.0,
+                    952.0,
+                    1408.0,
+                    952.0,
+                    1408.0,
+                    990.0,
+                    307.0,
+                    990.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    353.0,
+                    982.0,
+                    1212.0,
+                    982.0,
+                    1212.0,
+                    1021.0,
+                    353.0,
+                    1021.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    307.0,
+                    1031.0,
+                    1408.0,
+                    1031.0,
+                    1408.0,
+                    1070.0,
+                    307.0,
+                    1070.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    353.0,
+                    1061.0,
+                    1406.0,
+                    1061.0,
+                    1406.0,
+                    1100.0,
+                    353.0,
+                    1100.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    351.0,
+                    1088.0,
+                    431.0,
+                    1088.0,
+                    431.0,
+                    1129.0,
+                    351.0,
+                    1129.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1139.0,
+                    1406.0,
+                    1139.0,
+                    1406.0,
+                    1177.0,
+                    295.0,
+                    1177.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    351.0,
+                    1167.0,
+                    1312.0,
+                    1167.0,
+                    1312.0,
+                    1210.0,
+                    351.0,
+                    1210.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1218.0,
+                    1406.0,
+                    1218.0,
+                    1406.0,
+                    1256.0,
+                    295.0,
+                    1256.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    353.0,
+                    1248.0,
+                    1408.0,
+                    1248.0,
+                    1408.0,
+                    1287.0,
+                    353.0,
+                    1287.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    355.0,
+                    1281.0,
+                    507.0,
+                    1281.0,
+                    507.0,
+                    1313.0,
+                    355.0,
+                    1313.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1324.0,
+                    1406.0,
+                    1324.0,
+                    1406.0,
+                    1366.0,
+                    291.0,
+                    1366.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    353.0,
+                    1354.0,
+                    831.0,
+                    1354.0,
+                    831.0,
+                    1393.0,
+                    353.0,
+                    1393.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1405.0,
+                    1406.0,
+                    1405.0,
+                    1406.0,
+                    1443.0,
+                    295.0,
+                    1443.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    353.0,
+                    1435.0,
+                    1408.0,
+                    1435.0,
+                    1408.0,
+                    1474.0,
+                    353.0,
+                    1474.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    355.0,
+                    1468.0,
+                    667.0,
+                    1468.0,
+                    667.0,
+                    1500.0,
+                    355.0,
+                    1500.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1514.0,
+                    1406.0,
+                    1514.0,
+                    1406.0,
+                    1551.0,
+                    293.0,
+                    1551.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    353.0,
+                    1545.0,
+                    1406.0,
+                    1545.0,
+                    1406.0,
+                    1584.0,
+                    353.0,
+                    1584.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    351.0,
+                    1575.0,
+                    1406.0,
+                    1575.0,
+                    1406.0,
+                    1610.0,
+                    351.0,
+                    1610.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    351.0,
+                    1606.0,
+                    902.0,
+                    1606.0,
+                    902.0,
+                    1642.0,
+                    351.0,
+                    1642.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1649.0,
+                    1410.0,
+                    1649.0,
+                    1410.0,
+                    1695.0,
+                    291.0,
+                    1695.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    349.0,
+                    1677.0,
+                    1408.0,
+                    1677.0,
+                    1408.0,
+                    1726.0,
+                    349.0,
+                    1726.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    351.0,
+                    1714.0,
+                    1406.0,
+                    1714.0,
+                    1406.0,
+                    1750.0,
+                    351.0,
+                    1750.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    355.0,
+                    1748.0,
+                    884.0,
+                    1748.0,
+                    884.0,
+                    1781.0,
+                    355.0,
+                    1781.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1793.0,
+                    1408.0,
+                    1793.0,
+                    1408.0,
+                    1831.0,
+                    295.0,
+                    1831.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    353.0,
+                    1823.0,
+                    1408.0,
+                    1823.0,
+                    1408.0,
+                    1862.0,
+                    353.0,
+                    1862.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    353.0,
+                    1854.0,
+                    1406.0,
+                    1854.0,
+                    1406.0,
+                    1892.0,
+                    353.0,
+                    1892.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    349.0,
+                    1878.0,
+                    1408.0,
+                    1878.0,
+                    1408.0,
+                    1927.0,
+                    349.0,
+                    1927.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    351.0,
+                    1906.0,
+                    1410.0,
+                    1906.0,
+                    1410.0,
+                    1957.0,
+                    351.0,
+                    1957.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    353.0,
+                    1945.0,
+                    1420.0,
+                    1945.0,
+                    1420.0,
+                    1984.0,
+                    353.0,
+                    1984.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    351.0,
+                    1976.0,
+                    427.0,
+                    1976.0,
+                    427.0,
+                    2016.0,
+                    351.0,
+                    2016.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 10,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 2,
+                "poly": [
+                    836,
+                    2061,
+                    866,
+                    2061,
+                    866,
+                    2086,
+                    836,
+                    2086
+                ],
+                "score": 0.823
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    859,
+                    1403,
+                    859,
+                    1403,
+                    953,
+                    298,
+                    953
+                ],
+                "score": 0.507
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    526,
+                    1407,
+                    526,
+                    1407,
+                    650,
+                    299,
+                    650
+                ],
+                "score": 0.497
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    749,
+                    1403,
+                    749,
+                    1403,
+                    843,
+                    299,
+                    843
+                ],
+                "score": 0.493
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    1112,
+                    1401,
+                    1112,
+                    1401,
+                    1208,
+                    299,
+                    1208
+                ],
+                "score": 0.488
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    295,
+                    202,
+                    1405,
+                    202,
+                    1405,
+                    267,
+                    295,
+                    267
+                ],
+                "score": 0.488
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    1335,
+                    1403,
+                    1335,
+                    1403,
+                    1432,
+                    299,
+                    1432
+                ],
+                "score": 0.485
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1225,
+                    1401,
+                    1225,
+                    1401,
+                    1318,
+                    297,
+                    1318
+                ],
+                "score": 0.481
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1447,
+                    1400,
+                    1447,
+                    1400,
+                    1543,
+                    297,
+                    1543
+                ],
+                "score": 0.474
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    445,
+                    1403,
+                    445,
+                    1403,
+                    510,
+                    298,
+                    510
+                ],
+                "score": 0.469
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    295,
+                    667,
+                    1403,
+                    667,
+                    1403,
+                    733,
+                    295,
+                    733
+                ],
+                "score": 0.468
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    283,
+                    1400,
+                    283,
+                    1400,
+                    348,
+                    298,
+                    348
+                ],
+                "score": 0.455
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    302,
+                    364,
+                    1403,
+                    364,
+                    1403,
+                    429,
+                    302,
+                    429
+                ],
+                "score": 0.445
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    296,
+                    1559,
+                    1400,
+                    1559,
+                    1400,
+                    1624,
+                    296,
+                    1624
+                ],
+                "score": 0.435
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    295,
+                    1640,
+                    1401,
+                    1640,
+                    1401,
+                    1704,
+                    295,
+                    1704
+                ],
+                "score": 0.429
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    296,
+                    1801,
+                    1400,
+                    1801,
+                    1400,
+                    1898,
+                    296,
+                    1898
+                ],
+                "score": 0.388
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1721,
+                    1402,
+                    1721,
+                    1402,
+                    1785,
+                    298,
+                    1785
+                ],
+                "score": 0.378
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    972,
+                    1405,
+                    972,
+                    1405,
+                    1095,
+                    298,
+                    1095
+                ],
+                "score": 0.378
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1914,
+                    1409,
+                    1914,
+                    1409,
+                    2008,
+                    298,
+                    2008
+                ],
+                "score": 0.095
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1198,
+                    1869,
+                    1216,
+                    1869,
+                    1216,
+                    1891,
+                    1198,
+                    1891
+                ],
+                "score": 0.44,
+                "latex": "\\underline { { \\underline { { \\mathbf { \\Pi } } } } } ="
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    491,
+                    814,
+                    510,
+                    814,
+                    510,
+                    836,
+                    491,
+                    836
+                ],
+                "score": 0.3,
+                "latex": "{ . } = { }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    936,
+                    1068,
+                    954,
+                    1068,
+                    954,
+                    1089,
+                    936,
+                    1089
+                ],
+                "score": 0.25,
+                "latex": "\\underline { { \\underline { { \\mathbf { \\Pi } } } } }"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    832.0,
+                    2058.0,
+                    869.0,
+                    2058.0,
+                    869.0,
+                    2097.0,
+                    832.0,
+                    2097.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    859.0,
+                    1407.0,
+                    859.0,
+                    1407.0,
+                    896.0,
+                    296.0,
+                    896.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    353.0,
+                    891.0,
+                    1408.0,
+                    891.0,
+                    1408.0,
+                    925.0,
+                    353.0,
+                    925.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    351.0,
+                    918.0,
+                    430.0,
+                    918.0,
+                    430.0,
+                    956.0,
+                    351.0,
+                    956.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    526.0,
+                    1409.0,
+                    526.0,
+                    1409.0,
+                    561.0,
+                    294.0,
+                    561.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    352.0,
+                    555.0,
+                    1407.0,
+                    555.0,
+                    1407.0,
+                    594.0,
+                    352.0,
+                    594.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    356.0,
+                    589.0,
+                    1412.0,
+                    589.0,
+                    1412.0,
+                    622.0,
+                    356.0,
+                    622.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    355.0,
+                    618.0,
+                    457.0,
+                    618.0,
+                    457.0,
+                    650.0,
+                    355.0,
+                    650.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    749.0,
+                    1408.0,
+                    749.0,
+                    1408.0,
+                    786.0,
+                    295.0,
+                    786.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    353.0,
+                    780.0,
+                    1405.0,
+                    780.0,
+                    1405.0,
+                    815.0,
+                    353.0,
+                    815.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    358.0,
+                    812.0,
+                    490.0,
+                    812.0,
+                    490.0,
+                    842.0,
+                    358.0,
+                    842.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    511.0,
+                    812.0,
+                    688.0,
+                    812.0,
+                    688.0,
+                    842.0,
+                    511.0,
+                    842.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1111.0,
+                    1406.0,
+                    1111.0,
+                    1406.0,
+                    1153.0,
+                    294.0,
+                    1153.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    352.0,
+                    1140.0,
+                    1407.0,
+                    1140.0,
+                    1407.0,
+                    1186.0,
+                    352.0,
+                    1186.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    354.0,
+                    1176.0,
+                    1360.0,
+                    1176.0,
+                    1360.0,
+                    1210.0,
+                    354.0,
+                    1210.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    202.0,
+                    1408.0,
+                    202.0,
+                    1408.0,
+                    238.0,
+                    295.0,
+                    238.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    356.0,
+                    233.0,
+                    1379.0,
+                    233.0,
+                    1379.0,
+                    269.0,
+                    356.0,
+                    269.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1335.0,
+                    1407.0,
+                    1335.0,
+                    1407.0,
+                    1372.0,
+                    294.0,
+                    1372.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    354.0,
+                    1367.0,
+                    1409.0,
+                    1367.0,
+                    1409.0,
+                    1402.0,
+                    354.0,
+                    1402.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    356.0,
+                    1398.0,
+                    1291.0,
+                    1398.0,
+                    1291.0,
+                    1433.0,
+                    356.0,
+                    1433.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1222.0,
+                    1406.0,
+                    1222.0,
+                    1406.0,
+                    1263.0,
+                    295.0,
+                    1263.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    352.0,
+                    1255.0,
+                    1408.0,
+                    1255.0,
+                    1408.0,
+                    1292.0,
+                    352.0,
+                    1292.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    350.0,
+                    1284.0,
+                    429.0,
+                    1284.0,
+                    429.0,
+                    1320.0,
+                    350.0,
+                    1320.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1447.0,
+                    1405.0,
+                    1447.0,
+                    1405.0,
+                    1480.0,
+                    296.0,
+                    1480.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    355.0,
+                    1479.0,
+                    1406.0,
+                    1479.0,
+                    1406.0,
+                    1512.0,
+                    355.0,
+                    1512.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    355.0,
+                    1510.0,
+                    887.0,
+                    1510.0,
+                    887.0,
+                    1544.0,
+                    355.0,
+                    1544.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    445.0,
+                    1404.0,
+                    445.0,
+                    1404.0,
+                    481.0,
+                    297.0,
+                    481.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    352.0,
+                    476.0,
+                    551.0,
+                    476.0,
+                    551.0,
+                    509.0,
+                    352.0,
+                    509.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    665.0,
+                    1405.0,
+                    665.0,
+                    1405.0,
+                    707.0,
+                    294.0,
+                    707.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    354.0,
+                    699.0,
+                    630.0,
+                    699.0,
+                    630.0,
+                    730.0,
+                    354.0,
+                    730.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    282.0,
+                    1405.0,
+                    282.0,
+                    1405.0,
+                    322.0,
+                    294.0,
+                    322.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    352.0,
+                    313.0,
+                    1376.0,
+                    313.0,
+                    1376.0,
+                    350.0,
+                    352.0,
+                    350.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    364.0,
+                    1408.0,
+                    364.0,
+                    1408.0,
+                    399.0,
+                    295.0,
+                    399.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    355.0,
+                    393.0,
+                    1210.0,
+                    393.0,
+                    1210.0,
+                    429.0,
+                    355.0,
+                    429.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1559.0,
+                    1404.0,
+                    1559.0,
+                    1404.0,
+                    1595.0,
+                    296.0,
+                    1595.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    354.0,
+                    1590.0,
+                    1404.0,
+                    1590.0,
+                    1404.0,
+                    1624.0,
+                    354.0,
+                    1624.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1637.0,
+                    1405.0,
+                    1637.0,
+                    1405.0,
+                    1678.0,
+                    294.0,
+                    1678.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    354.0,
+                    1671.0,
+                    946.0,
+                    1671.0,
+                    946.0,
+                    1705.0,
+                    354.0,
+                    1705.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1801.0,
+                    1405.0,
+                    1801.0,
+                    1405.0,
+                    1840.0,
+                    295.0,
+                    1840.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    353.0,
+                    1831.0,
+                    1405.0,
+                    1831.0,
+                    1405.0,
+                    1873.0,
+                    353.0,
+                    1873.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    353.0,
+                    1862.0,
+                    1197.0,
+                    1862.0,
+                    1197.0,
+                    1902.0,
+                    353.0,
+                    1902.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1217.0,
+                    1862.0,
+                    1396.0,
+                    1862.0,
+                    1396.0,
+                    1902.0,
+                    1217.0,
+                    1902.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1720.0,
+                    1409.0,
+                    1720.0,
+                    1409.0,
+                    1759.0,
+                    294.0,
+                    1759.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    356.0,
+                    1752.0,
+                    725.0,
+                    1752.0,
+                    725.0,
+                    1784.0,
+                    356.0,
+                    1784.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    971.0,
+                    1407.0,
+                    971.0,
+                    1407.0,
+                    1011.0,
+                    292.0,
+                    1011.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    355.0,
+                    1004.0,
+                    1405.0,
+                    1004.0,
+                    1405.0,
+                    1037.0,
+                    355.0,
+                    1037.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    352.0,
+                    1034.0,
+                    1408.0,
+                    1034.0,
+                    1408.0,
+                    1068.0,
+                    352.0,
+                    1068.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    355.0,
+                    1064.0,
+                    935.0,
+                    1064.0,
+                    935.0,
+                    1099.0,
+                    355.0,
+                    1099.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    955.0,
+                    1064.0,
+                    1131.0,
+                    1064.0,
+                    1131.0,
+                    1099.0,
+                    955.0,
+                    1099.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1913.0,
+                    1406.0,
+                    1913.0,
+                    1406.0,
+                    1948.0,
+                    293.0,
+                    1948.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    354.0,
+                    1944.0,
+                    1410.0,
+                    1944.0,
+                    1410.0,
+                    1982.0,
+                    354.0,
+                    1982.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    358.0,
+                    1978.0,
+                    799.0,
+                    1978.0,
+                    799.0,
+                    2008.0,
+                    358.0,
+                    2008.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 11,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 2,
+                "poly": [
+                    836,
+                    2061,
+                    865,
+                    2061,
+                    865,
+                    2085,
+                    836,
+                    2085
+                ],
+                "score": 0.826
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1484,
+                    1407,
+                    1484,
+                    1407,
+                    1609,
+                    298,
+                    1609
+                ],
+                "score": 0.592
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    295,
+                    766,
+                    1401,
+                    766,
+                    1401,
+                    862,
+                    295,
+                    862
+                ],
+                "score": 0.576
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    879,
+                    1405,
+                    879,
+                    1405,
+                    1004,
+                    298,
+                    1004
+                ],
+                "score": 0.573
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    294,
+                    202,
+                    1402,
+                    202,
+                    1402,
+                    296,
+                    294,
+                    296
+                ],
+                "score": 0.567
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    294,
+                    653,
+                    1404,
+                    653,
+                    1404,
+                    748,
+                    294,
+                    748
+                ],
+                "score": 0.549
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1023,
+                    1407,
+                    1023,
+                    1407,
+                    1178,
+                    298,
+                    1178
+                ],
+                "score": 0.535
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1197,
+                    1405,
+                    1197,
+                    1405,
+                    1352,
+                    297,
+                    1352
+                ],
+                "score": 0.515
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    293,
+                    541,
+                    1402,
+                    541,
+                    1402,
+                    636,
+                    293,
+                    636
+                ],
+                "score": 0.509
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    295,
+                    427,
+                    1402,
+                    427,
+                    1402,
+                    522,
+                    295,
+                    522
+                ],
+                "score": 0.475
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    295,
+                    314,
+                    1401,
+                    314,
+                    1401,
+                    409,
+                    295,
+                    409
+                ],
+                "score": 0.455
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    289,
+                    1371,
+                    1402,
+                    1371,
+                    1402,
+                    1465,
+                    289,
+                    1465
+                ],
+                "score": 0.447
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1198,
+                    1324,
+                    1217,
+                    1324,
+                    1217,
+                    1346,
+                    1198,
+                    1346
+                ],
+                "score": 0.65,
+                "latex": "\\underline { { \\underline { { \\mathbf { \\Pi } } } } } ="
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    772,
+                    977,
+                    791,
+                    977,
+                    791,
+                    999,
+                    772,
+                    999
+                ],
+                "score": 0.57,
+                "latex": "\\underline { { \\underline { { \\mathbf { \\Pi } } } } } ="
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    832.0,
+                    2057.0,
+                    870.0,
+                    2057.0,
+                    870.0,
+                    2096.0,
+                    832.0,
+                    2096.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1484.0,
+                    1408.0,
+                    1484.0,
+                    1408.0,
+                    1521.0,
+                    295.0,
+                    1521.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    354.0,
+                    1514.0,
+                    1407.0,
+                    1514.0,
+                    1407.0,
+                    1551.0,
+                    354.0,
+                    1551.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    354.0,
+                    1545.0,
+                    1410.0,
+                    1545.0,
+                    1410.0,
+                    1582.0,
+                    354.0,
+                    1582.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    355.0,
+                    1578.0,
+                    803.0,
+                    1578.0,
+                    803.0,
+                    1611.0,
+                    355.0,
+                    1611.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    766.0,
+                    1406.0,
+                    766.0,
+                    1406.0,
+                    803.0,
+                    294.0,
+                    803.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    353.0,
+                    797.0,
+                    1405.0,
+                    797.0,
+                    1405.0,
+                    834.0,
+                    353.0,
+                    834.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    352.0,
+                    830.0,
+                    1226.0,
+                    830.0,
+                    1226.0,
+                    864.0,
+                    352.0,
+                    864.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    881.0,
+                    1405.0,
+                    881.0,
+                    1405.0,
+                    917.0,
+                    295.0,
+                    917.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    352.0,
+                    910.0,
+                    1409.0,
+                    910.0,
+                    1409.0,
+                    945.0,
+                    352.0,
+                    945.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    354.0,
+                    943.0,
+                    1411.0,
+                    943.0,
+                    1411.0,
+                    979.0,
+                    354.0,
+                    979.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    355.0,
+                    974.0,
+                    771.0,
+                    974.0,
+                    771.0,
+                    1007.0,
+                    355.0,
+                    1007.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    792.0,
+                    974.0,
+                    966.0,
+                    974.0,
+                    966.0,
+                    1007.0,
+                    792.0,
+                    1007.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    202.0,
+                    1407.0,
+                    202.0,
+                    1407.0,
+                    237.0,
+                    294.0,
+                    237.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    351.0,
+                    233.0,
+                    1407.0,
+                    233.0,
+                    1407.0,
+                    268.0,
+                    351.0,
+                    268.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    356.0,
+                    264.0,
+                    1132.0,
+                    264.0,
+                    1132.0,
+                    298.0,
+                    356.0,
+                    298.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    653.0,
+                    1409.0,
+                    653.0,
+                    1409.0,
+                    691.0,
+                    296.0,
+                    691.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    352.0,
+                    681.0,
+                    1409.0,
+                    681.0,
+                    1409.0,
+                    724.0,
+                    352.0,
+                    724.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    354.0,
+                    714.0,
+                    1135.0,
+                    714.0,
+                    1135.0,
+                    750.0,
+                    354.0,
+                    750.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1023.0,
+                    1405.0,
+                    1023.0,
+                    1405.0,
+                    1059.0,
+                    293.0,
+                    1059.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    354.0,
+                    1051.0,
+                    1405.0,
+                    1051.0,
+                    1405.0,
+                    1093.0,
+                    354.0,
+                    1093.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    353.0,
+                    1081.0,
+                    1405.0,
+                    1081.0,
+                    1405.0,
+                    1123.0,
+                    353.0,
+                    1123.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    354.0,
+                    1112.0,
+                    1405.0,
+                    1112.0,
+                    1405.0,
+                    1152.0,
+                    354.0,
+                    1152.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    356.0,
+                    1147.0,
+                    904.0,
+                    1147.0,
+                    904.0,
+                    1180.0,
+                    356.0,
+                    1180.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1199.0,
+                    1407.0,
+                    1199.0,
+                    1407.0,
+                    1232.0,
+                    296.0,
+                    1232.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    353.0,
+                    1225.0,
+                    1406.0,
+                    1225.0,
+                    1406.0,
+                    1264.0,
+                    353.0,
+                    1264.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    352.0,
+                    1253.0,
+                    1406.0,
+                    1253.0,
+                    1406.0,
+                    1300.0,
+                    352.0,
+                    1300.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    352.0,
+                    1284.0,
+                    1405.0,
+                    1284.0,
+                    1405.0,
+                    1329.0,
+                    352.0,
+                    1329.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    354.0,
+                    1318.0,
+                    1197.0,
+                    1318.0,
+                    1197.0,
+                    1356.0,
+                    354.0,
+                    1356.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1218.0,
+                    1318.0,
+                    1393.0,
+                    1318.0,
+                    1393.0,
+                    1356.0,
+                    1218.0,
+                    1356.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    542.0,
+                    1405.0,
+                    542.0,
+                    1405.0,
+                    576.0,
+                    296.0,
+                    576.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    351.0,
+                    570.0,
+                    1407.0,
+                    570.0,
+                    1407.0,
+                    609.0,
+                    351.0,
+                    609.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    355.0,
+                    599.0,
+                    903.0,
+                    599.0,
+                    903.0,
+                    638.0,
+                    355.0,
+                    638.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    426.0,
+                    1408.0,
+                    426.0,
+                    1408.0,
+                    466.0,
+                    294.0,
+                    466.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    352.0,
+                    457.0,
+                    1406.0,
+                    457.0,
+                    1406.0,
+                    494.0,
+                    352.0,
+                    494.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    353.0,
+                    492.0,
+                    858.0,
+                    492.0,
+                    858.0,
+                    522.0,
+                    353.0,
+                    522.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    314.0,
+                    1405.0,
+                    314.0,
+                    1405.0,
+                    352.0,
+                    294.0,
+                    352.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    353.0,
+                    344.0,
+                    1405.0,
+                    344.0,
+                    1405.0,
+                    382.0,
+                    353.0,
+                    382.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    353.0,
+                    375.0,
+                    562.0,
+                    375.0,
+                    562.0,
+                    411.0,
+                    353.0,
+                    411.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1370.0,
+                    1406.0,
+                    1370.0,
+                    1406.0,
+                    1408.0,
+                    295.0,
+                    1408.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    355.0,
+                    1403.0,
+                    1405.0,
+                    1403.0,
+                    1405.0,
+                    1437.0,
+                    355.0,
+                    1437.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    355.0,
+                    1430.0,
+                    877.0,
+                    1430.0,
+                    877.0,
+                    1467.0,
+                    355.0,
+                    1467.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 12,
+            "width": 1700,
+            "height": 2200
+        }
+    }
+]
\ No newline at end of file
diff --git a/parse/train/r1My6sR9tX/r1My6sR9tX.md b/parse/train/r1My6sR9tX/r1My6sR9tX.md
new file mode 100644
index 0000000000000000000000000000000000000000..08f920f1a693ee519c7837b62e58569cac18a44c
--- /dev/null
+++ b/parse/train/r1My6sR9tX/r1My6sR9tX.md
@@ -0,0 +1,388 @@
+# UNSUPERVISED LEARNING VIA META-LEARNING
+
+Kyle Hsu†   
+University of Toronto   
+kyle.hsu@mail.utoronto.ca
+
+Sergey Levine, Chelsea Finn University of California, Berkeley {svlevine,cbfinn}@eecs.berkeley.edu
+
+# ABSTRACT
+
+A central goal of unsupervised learning is to acquire representations from unlabeled data or experience that can be used for more effective learning of downstream tasks from modest amounts of labeled data. Many prior unsupervised learning works aim to do so by developing proxy objectives based on reconstruction, disentanglement, prediction, and other metrics. Instead, we develop an unsupervised meta-learning method that explicitly optimizes for the ability to learn a variety of tasks from small amounts of data. To do so, we construct tasks from unlabeled data in an automatic way and run meta-learning over the constructed tasks. Surprisingly, we find that, when integrated with meta-learning, relatively simple task construction mechanisms, such as clustering embeddings, lead to good performance on a variety of downstream, human-specified tasks. Our experiments across four image datasets indicate that our unsupervised meta-learning approach acquires a learning algorithm without any labeled data that is applicable to a wide range of downstream classification tasks, improving upon the embedding learned by four prior unsupervised learning methods.
+
+# 1 INTRODUCTION
+
+Unsupervised learning is a fundamental, unsolved problem (Hastie et al., 2009) and has seen promising results in domains such as image recognition (Le et al., 2013) and natural language understanding (Ramachandran et al., 2017). A central use case of unsupervised learning methods is enabling better or more efficient learning of downstream tasks by training on top of unsupervised representations (Reed et al., 2014; Cheung et al., 2015; Chen et al., 2016) or fine-tuning a learned model (Erhan et al., 2010). However, since the downstream objective requires access to supervision, the objectives used for unsupervised learning are only a rough proxy for downstream performance. If a central goal of unsupervised learning is to learn useful representations, can we derive an unsupervised learning objective that explicitly takes into account how the representation will be used?
+
+The use of unsupervised representations for downstream tasks is closely related to the objective of meta-learning techniques: finding a learning procedure that is more efficient and effective than learning from scratch. However, unlike unsupervised learning methods, meta-learning methods require large, labeled datasets and hand-specified task distributions. These dependencies are major obstacles to widespread use of these methods for few-shot classification.
+
+To begin addressing these problems, we propose an unsupervised meta-learning method: one which aims to learn a learning procedure, without supervision, that is useful for solving a wide range of new, human-specified tasks. With only raw, unlabeled observations, our model’s goal is to learn a useful prior such that, after meta-training, when presented with a modestly-sized dataset for a human-specified task, the model can transfer its prior experience to efficiently learn to perform the new task. If we can build such an algorithm, we can enable few-shot learning of new tasks without needing any labeled data nor any pre-defined tasks.
+
+To perform unsupervised meta-learning, we need to automatically construct tasks from unlabeled data. We study several options for how this can be done. We find that a good task distribution should be diverse, but also not too difficult: na¨ıve random approaches for task generation produce tasks that contain insufficient regularity to enable useful meta-learning. To that end, our method proposes tasks by first leveraging prior unsupervised learning algorithms to learn an embedding of the input data, and then performing an overcomplete partitioning of the dataset to construct numerous categorizations of the data. We show how we can derive classification tasks from these categorizations for use with meta-learning algorithms. Surprisingly, even with simple mechanisms for partitioning the embedding space, such as $k$ -means clustering, we find that meta-learning acquires priors that, when used to learn new, human-designed tasks, learn those tasks more effectively than methods that directly learn on the embedding. That is, the learning algorithm acquired through unsupervised meta-learning achieves better downstream performance than the original representation used to derive meta-training tasks, without introducing any additional assumptions or supervision. See Figure 1 for an illustration of the complete approach.
+
+The core idea in this paper is that we can leverage unsupervised embeddings to propose tasks for a meta-learning algorithm, leading to an unsupervised meta-learning algorithm that is particularly effective as pre-training for human-specified downstream tasks. In the following sections, we formalize our problem assumptions and goal, which match those of unsupervised learning, and discuss several options for automatically deriving tasks from embeddings. We instantiate our method with two meta-learning algorithms and compare to prior state-of-the-art unsupervised learning methods. Across four image datasets (MNIST, Omniglot, miniImageNet, and CelebA), we find that our method consistently leads to effective downstream learning of a variety of human-specified tasks, including character recognition tasks, object classification tasks, and facial attribute discrimination tasks, without requiring any labels or hand-designed tasks during meta-learning and where key hyperparameters of our method are held constant across all domains. We show that, even though our unsupervised meta-learning algorithm trains for one-shot generalization, one instantiation of our approach performs well not only on few-shot learning, but also when learning downstream tasks with up to 50 training examples per class. In fact, some of our results begin to approach the performance of fully-supervised meta-learning techniques trained with fully-specified task distributions.
+
+![](images/2aa42c4a43d437245b1ba217e8b16d1cdbed8dd1add1db4a9609cd8a7a78f543.jpg)  
+Figure 1: Illustration of the proposed unsupervised meta-learning procedure. Embeddings of raw observations are clustered with $k$ -means to construct partitions, which give rise to classification tasks. Each task involves distinguishing between examples from $N = 2$ clusters, with $K _ { \mathrm { m - t r } } = 1$ example from each cluster being a training input. The meta-learner’s aim is to produce a learning procedure that successfully solves these tasks.
+
+# 2 UNSUPERVISED META-LEARNING
+
+In this section, we describe our problem setting in relation to that of unsupervised and semisupervised learning, provide necessary preliminaries, and present our approach.
+
+# 2.1 PROBLEM STATEMENT
+
+Our goal is to leverage unlabeled data for the efficient learning of a range of human-specified downstream tasks. We only assume access to an unlabeled dataset $\mathcal { D } = \{ \mathbf { x } _ { i } \}$ during meta-training. After learning from the unlabeled data, which we will refer to as unsupervised meta-training, we want to apply what was learned towards learning a variety of downstream, human-specified tasks from a modest amount of labeled data, potentially as few as a single example per class. These downstream tasks may, in general, have different underlying classes or attributes (in contrast to typical semi-supervised problem assumptions), but are assumed to have inputs from the same distribution as the one from which datapoints in $\mathcal { D }$ are drawn. Concretely, we assume that downstream tasks are $M$ -way classification tasks, and that the goal is to learn an accurate classifier using $K$ labeled datapoints $\left( \mathbf { x } _ { k } , \mathbf { y } _ { k } \right)$ from each of the $M$ classes, where $K$ is relatively small (i.e. between 1 and 50).
+
+The unsupervised meta-training phase aligns with the unsupervised learning problem in that it involves no access to information about the downstream tasks, other than the fact that they are $M$ -way classification tasks, for variable $M$ upper-bounded by $N$ . The upper bound $N$ is assumed to be known during unsupervised meta-training, but otherwise, the values of $M$ and $K$ are not known $a$ priori. As a result, the unsupervised meta-training phase needs to acquire a sufficiently general prior for applicability to a range of classification tasks with variable quantities of data and classes. This problem definition is our prototype for a practical use-case in which a user would like to train an application-specific image classifier, but does not have an abundance of labeled data.
+
+# 2.2 PRELIMINARIES
+
+Unsupervised embedding learning. An unsupervised embedding learning algorithm $\mathcal { E }$ is a procedure that takes as input an unlabeled dataset $\mathcal { D } = \left\{ \mathbf { x } _ { i } \right\}$ and outputs a mapping from $\{ { \bf { x } } _ { i } \}$ to embeddings $\left\{ \mathbf { z } _ { i } \right\}$ . These embedded points are typically lower-dimensional and arranged such that distances correspond to meaningful differences between inputs, in contrast to distances between the original inputs, such as image pixels, which are not meaningful measures of image similarity.
+
+Task. An $M$ -way $K$ -shot classification task $\tau$ consists of $K$ training datapoints and labels $\{ ( \mathbf { x } _ { k } , \ell _ { k } ) \}$ per class, which are used for learning a classifier, and $Q$ query datapoints and labels per class, on which the learned classifier is evaluated. That is, in a task there are $K + Q = R$ datapoints and labels for each of the $M$ classes.
+
+Meta-learning. A supervised meta-learning algorithm $\mathcal { M } ( \cdot )$ takes as input a set of supervised metatraining tasks $\{ \mathcal { T } _ { t } \}$ . It produces a learning procedure $\mathcal F ( \cdot )$ , which, in turn, ingests the supervised training data of a task to produce a classifier $f ( \cdot )$ . The goal of $\mathcal { M }$ is to learn $\mathcal { F }$ such that, when faced with a meta-test time task $\mathcal { T } _ { t ^ { \prime } }$ held-out from $\{ \mathcal { T } _ { t } \}$ , $\mathcal { F }$ can learn a $f _ { t ^ { \prime } }$ that accomplishes $\mathcal { T } _ { t ^ { \prime } }$ . At a high level, the quintessential meta-learning strategy is to have $\mathcal { M }$ iterate over $\{ \mathcal { T } _ { t } \}$ , cycling between applying the current form of $\mathcal { F } _ { t }$ on training data from $\mathcal { T } _ { t }$ to learn $f _ { t }$ , assessing its performance by calculating some meta-loss $\mathcal { L }$ on held-out data from the task, and optimizing $\mathcal { L }$ to improve the learning procedure.
+
+We build upon two meta-learning algorithms: model agnostic meta-learning (MAML) (Finn et al., 2017) and prototypical networks (ProtoNets) (Snell et al., 2017). MAML aims to learn the initial parameters of a deep network such that one or a few gradient steps leads to effective generalization; it specifies $\mathcal { F }$ as gradient descent starting from the meta-learned parameters. ProtoNets aim to metalearn a representation in which a class is effectively identified by its prototype, defined to be the mean of the class’ training examples in the meta-learned space; $\mathcal { F }$ is the computation of these class prototypes, and $f$ is a linear classifier that predicts the class whose prototype is closest in Euclidean distance to the query’s representation.
+
+Task generation for meta-learning. We briefly summarize how tasks are typically generated from labeled datasets $\left\{ \left( \mathbf { x } _ { i } , \mathbf { y } _ { i } \right) \right\}$ for supervised meta-learning, as introduced by Santoro et al. (2016). For simplicity, consider the case where the labels are discrete scalar values $y _ { i }$ . To construct an $N$ -way classification task $\tau$ (assuming $N$ is not greater than the number of unique $y _ { i }$ ), we can sample $N$ classes, sample $R$ datapoints $\{ { \bf { x } } _ { r } \} _ { n }$ for each of the $N$ classes, and sample a permutation of $N$ distinct one-hot vectors $( \ell _ { n } )$ to serve as task-specific labels of the $N$ sampled classes. The task is then defined as $\mathcal { T } = \{ ( \mathbf { x } _ { n , r } , \ell _ { n } ) ~ | ~ \mathbf { x } _ { n , r } \in \{ \mathbf { x } _ { r } \} _ { n } \}$ . Of course, this procedure is only possible with labeled data; in the next section, we discuss how we can construct tasks without ground-truth labels.
+
+# 2.3 UNSUPERVISED META-LEARNING WITH AUTOMATICALLY CONSTRUCTED TASKS
+
+We approach our problem from a meta-learning perspective, framing the problem as the acquisition, from unlabeled data, of an efficient learning procedure that is transferable to human-designed tasks. In particular, we aim to construct classification tasks from the unlabeled data and then learn how to efficiently learn these tasks. If such tasks are adequately diverse and structured, then metalearning these tasks should enable fast learning of new, human-provided tasks. A key question, then, is how to automatically construct such tasks from unlabeled data $\mathcal { D } = \left\{ \mathbf { x } _ { i } \right\}$ . Notice that in the supervised meta-learning task generation procedure detailed in Section 2.2, the labels $y _ { i }$ induce a partition $\mathcal { P } = \{ \mathcal { C } _ { c } \}$ over $\{ { \bf { x } } _ { i } \}$ by assigning all datapoints with label $y _ { c }$ to subset $\mathcal { C } _ { c }$ . Once a partition is obtained, task generation is simple; we can reduce the problem of constructing tasks to that of constructing a partition over $\{ { \bf { x } } _ { i } \}$ . All that’s left is to find a principled alternative to human labels for defining the partition.
+
+A na¨ıve approach is to randomly partition the data $\mathcal { D }$ . While such a scheme introduces diverse tasks, there is no structure; that is, there is no consistency between a task’s training data and query data, and hence nothing to be learned during each task, let alone across tasks. As seen in Table 3, providing a meta-learner with purely random tasks results in failed meta-learning.
+
+To construct tasks with structure that resembles that of human-specified labels, we need to group datapoints into consistent and distinct subsets based on salient features. With this motivation in mind, we propose to use $k$ -means clustering. Consider the partition $\mathcal { P } = \{ \mathcal { C } _ { c } \}$ learned by $k$ -means as a simplification of a Gaussian mixture model $p ( \mathbf { x } | c ) p ( \bar { c } )$ . If the clusters can recover a semblance of the true class-conditional generative distributions $p ( \mathbf { x } | c )$ , creating tasks based on treating these clusters as classes should result in useful unsupervised meta-training. However, the result of $k$ -means is critically dependent on the metric space on which its objective is defined. Clustering in pixel-space is unappealing for two reasons: (1) distance in pixel-space correlates poorly with semantic meaning, and (2) the high dimensionality of raw images renders clustering difficult in practice. We empirically show in Table 3 that meta-learning with tasks defined by pixel-space clusters, with preprocessing as directed by Coates & $\mathrm { N g }$ (2012), also fails.
+
+We are now motivated to cluster in spaces in which common distance functions correlate to semantic meaning. However, we must satisfy the constraints of our problem statement in the process of learning such spaces. To these ends, we use state-of-the-art unsupervised learning methods to produce useful embedding spaces. For qualitative evidence in the unsupervised learning literature that such embedding spaces exhibit semantic meaning, see Cheung et al. (2015); Bojanowski & Joulin (2017); Donahue et al. (2017). We note that while a given embedding space may not be directly suitable for highly-efficient learning of new tasks (which would require the embedding space to be precisely aligned or adaptable to the classes of those tasks), we can still leverage it for the construction of structured tasks, a process for which requirements are less strict.
+
+Thus, we first run an out-of-the-box unsupervised embedding learning algorithm $\mathcal { E }$ on $\mathcal { D }$ , then map the data $\{ { \bf { x } } _ { i } \}$ into the embedding space $\mathcal { Z }$ , producing $\left\{ \mathbf { z } _ { i } \right\}$ . To produce a diverse task set, we generate $P$ partitions $\{ \mathcal P _ { p } \}$ by running clustering $P$ times, applying random scaling to the dimensions of $\mathcal { Z }$ to induce a different metric, represented by diagonal matrix A, for each run of clustering. With $\pmb { \mu } _ { c }$ denoting the learned centroid of cluster $\mathcal { C } _ { c }$ , a single run of clustering can be summarized with
+
+$$
+\mathcal { P } , \{ \pmb { \mu } _ { c } \} = \underset { \{ \mathcal { C } _ { c } \} , \{ \pmb { \mu } _ { c } \} } { \arg \operatorname* { m i n } } \sum _ { c = 1 } ^ { k } \sum _ { \mathbf { z } \in \mathcal { C } _ { c } } \| \mathbf { z } - \pmb { \mu } _ { c } \| _ { \mathbf { A } } ^ { 2 }
+$$
+
+We derive tasks for meta-learning from the partitions using the procedure detailed in Section 2.2, except we begin the construction of each task by sampling a partition from the uniform distribution $\mathcal { U } ( \mathcal { P } )$ , and for ${ \bf x } _ { i } \in \mathcal { C } _ { c }$ , specify $y _ { i } = c$ . To avoid imbalanced clusters dominating the meta-training tasks, we opt not to sample from $p ( c ) \propto | \mathcal { C } _ { c } |$ , but instead sample $N$ clusters uniformly without replacement for each task. We note that Caron et al. (2018) are similarly motivated in their design decision of sampling data from a uniform distribution over clusters.
+
+With the partitions being constructed over $\left\{ \mathbf { z } _ { i } \right\}$ , we have one more design decision to make: should we perform meta-learning on embeddings or images? We consider that, to successfully solve new tasks at meta-test time, a learning procedure $\mathcal { F }$ that takes embeddings as input would depend on the embedding function’s ability to generalize to out-of-distribution observations. On the other hand, by meta-learning on images, $\mathcal { F }$ can separately adapt $f$ to each evaluation task from the rawest level of representation. Thus, we choose to meta-learn on images.
+
+We call our method clustering to automatically construct tasks for unsupervised meta-learning (CACTUs). We detail the task construction algorithm in Algorithm 1, and provide an illustration of the complete unsupervised meta-learning approach for classification in Figure 1.
+
+# 3 RELATED WORK
+
+The method we propose aims to address the unsupervised learning problem (Hastie et al., 2009; Le et al., 2013), namely acquiring a transferable learning procedure without labels. We show that our
+
+# Algorithm 1 CACTUs for classification
+
+<table><tr><td colspan="2">1: procedure CACTUs(ε,D,P,k,T,N,Km-tr, Q)</td><td colspan="4"></td></tr><tr><td>2:</td><td colspan="4">Run embedding learning algorithm ε on D and produce embeddings {zi} from observations {xi}.</td></tr><tr><td>3:</td><td colspan="4">Run k-means on {zi} P times (with random scaling) to generate a set of partitions {Pp = {Cc}p}.</td></tr><tr><td>4:</td><td colspan="4">for t from 1 to the number of desired tasks T do</td></tr><tr><td>5:</td><td colspan="4">Sample a partition P uniformly at random from the set of partitions {Pp}.</td></tr><tr><td>6:</td><td colspan="4">Sample a cluster Cn uniformly without replacement from P for each of the N classes desired for a task.</td></tr><tr><td>7:</td><td colspan="4">Sample an embedding Zr without replacement from Cn for each of the R= Km-tr+Q training and</td></tr><tr><td>8:</td><td colspan="4">query examples desired for each class, and record the corresponding datapoint Xn,r. Sample a permutation(ln) of N one-hot labels.</td></tr><tr><td>9:</td><td colspan="4">Construct Tt ={(xn,r,ln)}.</td></tr><tr><td>10:</td><td colspan="4">return {Tt}</td></tr></table>
+
+method is complementary to a number of unsupervised learning methods, including ACAI (Berthelot et al., 2018), BiGAN (Donahue et al., 2017; Dumoulin et al., 2017), DeepCluster (Caron et al., 2018), and InfoGAN (Chen et al., 2016): we leverage these prior methods to learn embeddings used for constructing meta-learning tasks, and demonstrate that our method learns a more useful representation than the embeddings. The ability to use what was learned during unsupervised pretraining to better or more efficiently learn a variety of downstream tasks is arguably one of the most practical applications of unsupervised learning methods, which has a long history in neural network training (Hinton et al., 2006; Bengio et al., 2007; Ranzato et al., 2006; Vincent et al., 2008; Erhan et al., 2010). Unsupervised pre-training has demonstrated success in a number of domains, including speech recognition (Yu et al., 2010), image classification (Zhang et al., 2017), machine translation (Ramachandran et al., 2017), and text classification (Dai & Le, 2015; Howard & Ruder, 2018; Radford et al., 2018). Our approach, unsupervised meta-learning, can be viewed as an unsupervised learning algorithm that explicitly optimizes for few-shot transferability. As a result, we can expect it to better learn human-specified downstream tasks, compared to unsupervised learning methods that optimize for other metrics, such as reconstruction (Vincent et al., 2010; Higgins et al., 2017), fidelity of constructed images (Radford et al., 2016; Salimans et al., 2016; Donahue et al., 2017; Dumoulin et al., 2017), representation interpolation (Berthelot et al., 2018), disentanglement (Bengio et al., 2013; Reed et al., 2014; Cheung et al., 2015; Chen et al., 2016; Mathieu et al., 2016; Denton & Birodkar, 2017), and clustering (Coates & Ng, 2012; Krahenb ¨ uhl et al., 2016; Bojanowski & Joulin, ¨ 2017; Caron et al., 2018). We empirically evaluate this hypothesis in the next section. In contrast to many previous evaluations of unsupervised pre-training, we focus on settings in which only a small amount of data for the downstream tasks is available, since this is where the unlabeled data can be maximally useful.
+
+Unsupervised pre-training followed by supervised learning can be viewed as a special case of the semi-supervised learning problem (Zhu, 2011; Kingma et al., 2014; Rasmus et al., 2015; Oliver et al., 2018). However, in contrast to our problem statement, semi-supervised learning methods assume that a significant proportion of the unlabeled data, if not all of it, shares underlying labels with the labeled data. Additionally, our approach and other unsupervised learning methods are wellsuited for transferring their learned representation to many possible downstream tasks or labelings, whereas semi-supervised learning methods typically optimize for performance on a single task, with respect to a single labeling of the data.
+
+Our method builds upon the ideas of meta-learning (Schmidhuber, 1987; Bengio et al., 1991; Naik & Mammone, 1992) and few-shot learning (Santoro et al., 2016; Vinyals et al., 2016; Ravi & Larochelle, 2017; Munkhdalai & Yu, 2017; Snell et al., 2017). We apply two meta-learning algorithms, model-agnostic meta-learning (Finn et al., 2017) and prototypical networks (Snell et al., 2017), to tasks constructed in an unsupervised manner. Similar to our problem setting, some prior works have aimed to learn an unsupervised learning procedure with supervised data (Garg & Kalai, 2017; Metz et al., 2018). Instead, we consider a problem setting that is entirely unsupervised, aiming to learn efficient learning algorithms using unlabeled datasets. Our problem setting is similar to that considered by Gupta et al. (2018), but we develop an approach that is suitable for supervised downstream tasks, rather than reinforcement learning problems, and demonstrate our algorithm on problems with high-dimensional visual observations.
+
+# 4 EXPERIMENTS
+
+We begin the experimental section by presenting our research questions and how our experiments are designed to address them. Links to code for the experiments can be found at https://sites. google.com/view/unsupervised-via-meta.
+
+Benefit of meta-learning. Is there any significant benefit to doing meta-learning on tasks derived from embeddings, or is the embedding function already sufficient for downstream supervised learning of new tasks? To investigate this, we run MAML and ProtoNets on tasks generated via CACTUs (CACTUs-MAML, CACTUs-ProtoNets). We compare to five alternate algorithms, with four being supervised learning methods on top of the embedding function. i) Embedding $k _ { \mathrm { n n } }$ -nearest neighbors first infers the embeddings of the downstream task images. For a query test image, it predicts the plurality vote of the labels of the $k _ { \mathrm { n n } }$ training images that are closest in the embedding space to the query’s embedding. ii) Embedding linear classifier also begins by inferring the embeddings of the downstream task images. It then fits a linear classifier using the $N K$ training embeddings and labels, and predicts labels for the query embeddings using the classifier. iii) Embedding multilayer perceptron instead uses a network with one hidden layer of 128 units and tuned dropout (Srivastava et al., 2014). iv) To isolate the effect of meta-learning on images, we also compare to embedding cluster matching, i.e. directly using the meta-training clusters for classification by labeling clusters with a task’s training data via plurality vote. If a query datapoint maps to an unlabeled cluster, the closest labeled cluster is used. v) As a baseline, we forgo any unsupervised pre-training and train a model with the MAML architecture from standard random network initialization via gradient descent separately for each evaluation task.
+
+Different embedding spaces. Does CACTUs result in successful meta-learning for many distinct task-generating embeddings? To investigate this, we run unsupervised meta-learning using four embedding learning algorithms: ACAI (Berthelot et al., 2018), BiGAN (Donahue et al., 2017), DeepCluster (Caron et al., 2018), and InfoGAN (Chen et al., 2016). These four approaches collectively cover the following range of objectives and frameworks in the unsupervised learning literature: generative modeling, two-player games, reconstruction, representation interpolation, discriminative clustering, and information maximization. We describe these methods in more detail in Appendix A.
+
+Applicability to different tasks. Can unsupervised meta-learning yield a good prior for a variety of task types? In other words, can unsupervised meta-learning yield a good representation for tasks that assess the ability to distinguish between features on different scales, or tasks with various amounts of supervision signal? To investigate this, we evaluate our procedure on tasks assessing recognition of character identity, object identity, and facial attributes. For this purpose we choose to use the existing Omniglot (Santoro et al., 2016) and miniImageNet (Ravi & Larochelle, 2017) datasets and few-shot classification tasks and, inspired by Finn et al. (2018), also construct a new few-shot classification benchmark based on the CelebA dataset and its binary attribute annotations. For miniImageNet, we consider both few-shot downstream tasks and tasks involving larger datasets (up to 50-shot). Specifics on the datasets and human-designed tasks are presented in Appendix B.
+
+Oracle. How does the performance of our unsupervised meta-learning method compare to supervised meta-learning with a human-specified, near-optimal task distribution derived from a labeled dataset? To investigate this, we use labeled versions of the meta-training datasets to run MAML and ProtoNets as supervised meta-learning algorithms (Oracle-MAML, Oracle-ProtoNets). To facilitate fair comparison with the unsupervised variants, we control for the relevant hyperparameters.
+
+Task construction ablation. How do the alternatives for constructing tasks from the embeddings compare? To investigate this, we run MAML on tasks constructed via clustering (CACTUs-MAML) and MAML on tasks constructed via random hyperplane slices of the embedding space with varying margin (Hyperplanes-MAML). The latter partitioning procedure is detailed in Appendix C. For the experiments where tasks are constructed via clustering, we also investigate the effect of sampling based on a single partition versus multiple partitions. We additionally experiment with tasks based on random assignments of images to “clusters” (Random-MAML) and tasks based on pixel-space clusters (Pixels CACTUs-MAML) with the Omniglot dataset.
+
+To investigate the limitations of our method, we also consider an easier version of our problem statement where the data distributions at meta-training and meta-test time perfectly overlap, i.e. the images share a common set of underlying labels (Appendix D). Finally, we present results on miniImageNet after unsupervised meta-learning on most of ILSVRC 2012 (Appendix G).
+
+![](images/da015e73e499847b381f6ba35cc8ae17e2cc6b8b08026a9f2a626ccbe7e0ef3c.jpg)  
+Figure 2: Examples of three DeepCluster-embedding cluster-based classes (a) and a 2-way 5-shot test task (b) for two datasets. (a) Some of the clusters correspond well to unseen labels (top left, bottom left). Others exhibit semantic meaning despite members not being grouped as such in the labeled version of the dataset (top middle: pair of objects, bottom middle: white hat). Still others are uninterpretable (top right) or are based on image artifacts (bottom right). (b) We evaluate unsupervised pre-training based on the ability to learn downstream, human-designed tasks with held-out images and underlying classes.
+
+# 4.1 EXPERIMENTAL PROTOCOL SUMMARY
+
+As discussed by Oliver et al. (2018), keeping proper experimental protocol is particularly important when evaluating unsupervised and semi-supervised learning algorithms. Our foremost concern is to avoid falsely embellishing the capabilities of our approach by overfitting to the specific datasets and task types that we consider. To this end, we adhere to two key principles. We do not perform any architecture engineering: we use architectures from prior work as-is, or lightly adapt them to our needs if necessary. We also keep hyperparameters related to the unsupervised meta-learning stage as constant as possible across all experiments, including the MAML and ProtoNets model architectures. Details on hyperparameters and architectures are presented in Appendix E. We assume knowledge of an upper bound on the number of classes $N$ present in each downstream meta-testing task for each dataset. However, regardless of the number of shots $K$ , we do not assume knowledge of $K$ during unsupervised meta-learning. We use $N$ -way 1-shot tasks during meta-training, but test on larger values of $K$ during meta-testing.
+
+We partition each dataset into meta-training, meta-validation, and meta-testing splits. For Omniglot and miniImageNet, these splits contain disjoint sets of classes. For all algorithms, we run unsupervised pre-training on the unlabeled meta-training split and report performance on downstream tasks dictated by the labeled data of the meta-testing split, generated using the procedure from prior work recounted in Section 2.2. For the supervised meta-learning oracles, meta-training tasks are constructed in the same manner but from the dataset’s meta-training split. See Figure 2 for illustrative examples of embedding-derived clusters and human-designed test tasks.
+
+To facilitate analysis on meta-overfitting, we use the labels of the meta-validation split (instead of clustering embeddings) to construct tasks for meta-validation. However, because our aim is to perform meta-learning without supervision, we do not tune hyperparameters on this labeled data. We use a fixed number of meta-training iterations, since there is no suitable criterion for early stopping.
+
+When we experiment with the embedding-plus-supervised-learning methods used as fair comparisons to unsupervised meta-learning, we err on the side of providing more supervision and data than technically allowed. Specifically, we separately tune the supervised learning hyperparameters for each dataset and each task difficulty on the labeled version of the meta-validation split. With DeepCluster embeddings, we also use the entire meta-testing split’s statistics to perform dimensionality reduction (via PCA) and whitening, which is unfair as this shares information across tasks.
+
+# 4.2 RESULTS
+
+Our primary results are summarized in Tables 1 and 2. Task construction ablations are summarized in Tables 3 and 4.
+
+Benefit of meta-learning. CACTUs-MAML consistently yields a learning procedure that results in more successful downstream task performance than all other unsupervised methods, including those that learn on top of the embedding that generated meta-training tasks for MAML. We find the same result for CACTUs-ProtoNets for 1-shot downstream tasks. However, as noted by Snell et al. (2017), ProtoNets perform best when meta-training shot and meta-testing shot are matched; this characteristic prevents ProtoNets from improving upon ACAI for 20-way 5-shot Omniglot and upon DeepCluster for 50-shot miniImageNet. We attribute the success of CACTUs-based meta-learning over the embedding-based methods to two factors: its practice in distinguishing between many distinct sets of clusters from modest amounts of signal, and the underlying classes of the meta-testing split data being out-of-distribution. In principle, the latter factor is solely responsible for the success over embedding cluster matching, since this algorithm can be viewed as a meta-learner on embeddings that trivially obtains perfect accuracy (via memorization) on the meta-training tasks. The same factor also helps explain why training from standard network initialization is, in general, competitive with directly using the task-generating embedding as a representation. On the other hand, the MNIST results (Table 7 in Appendix F) suggest that when the meta-training and meta-testing data distributions have perfect overlap and the embedding is well-suited enough that embedding cluster matching can already achieve high performance, CACTUs-MAML yields only a small benefit, as we would expect.
+
+Table 1: Results of unsupervised learning on Omniglot images, averaged over 1000 downstream character recognition tasks. CACTUs experiments use $k = 5 0 0$ clusters for each of $P = 1 0 0$ partitions. Embedding cluster matching uses the same $k$ . For complete results with confidence intervals, see Table 8 in Appendix F.   
+
+<table><tr><td>Algorithm</td><td>(way, shot)</td><td>(5,1)</td><td>(5,5)</td><td>(20,1)</td><td>(20,5)</td></tr><tr><td>Training from scratch</td><td>52.50%</td><td></td><td>74.78%</td><td>24.91%</td><td>47.62%</td></tr><tr><td>ACAI knn-nearest neighbors</td><td></td><td>57.46%</td><td>81.16%</td><td>39.73%</td><td>66.38%</td></tr><tr><td>ACAIlinear classifier</td><td></td><td>61.08%</td><td>81.82%</td><td>43.20%</td><td>66.33%</td></tr><tr><td>ACAI MLP with dropout</td><td></td><td>51.95%</td><td>77.20%</td><td>30.65%</td><td>58.62%</td></tr><tr><td>ACAI cluster matching</td><td></td><td>54.94%</td><td>71.09%</td><td>32.19%</td><td>45.93%</td></tr><tr><td>ACAI CACTUs-MAML (ours)</td><td>68.84%</td><td></td><td>87.78%</td><td>48.09%</td><td>73.36%</td></tr><tr><td>ACAI CACTUs-ProtoNets (ours)</td><td>68.12%</td><td></td><td>83.58%</td><td>47.75%</td><td>66.27%</td></tr><tr><td>BiGAN knn-nearest neighbors</td><td>49.55%</td><td></td><td>68.06%</td><td>27.37%</td><td>46.70%</td></tr><tr><td>BiGAN linearclassifier</td><td>48.28%</td><td></td><td>68.72%</td><td>27.80%</td><td>45.82%</td></tr><tr><td>BiGAN MLP with dropout</td><td>40.54%</td><td></td><td>62.56%</td><td>19.92%</td><td>40.71%</td></tr><tr><td>BiGAN cluster matching</td><td>43.96%</td><td></td><td>58.62%</td><td>21.54%</td><td>31.06%</td></tr><tr><td>BiGAN CACTUs-MAML (ours)</td><td>58.18%</td><td></td><td>78.66%</td><td>35.56%</td><td>58.62%</td></tr><tr><td>BiGAN CACTUs-ProtoNets (ours)</td><td>54.74%</td><td></td><td>71.69%</td><td>33.40%</td><td>50.62%</td></tr><tr><td>Oracle-MAML (control)</td><td>94.46%</td><td></td><td>98.83%</td><td>84.60%</td><td>96.29%</td></tr><tr><td>Oracle-ProtoNets (control)</td><td>98.35%</td><td></td><td>99.58%</td><td>95.31%</td><td>98.81%</td></tr></table>
+
+Different embedding spaces. CACTUs is effective for a variety of embedding learning methods used for task generation. The performance of unsupervised meta-learning can largely be predicted by the performance of the embedding-based non-meta-learning methods. For example, the ACAI embedding does well with Omniglot, leading to the best unsupervised results with ACAI CACTUsMAML. Likewise, on miniImageNet, the best performing prior embedding (DeepCluster) also corresponds to the best performing unsupervised meta-learner (DeepCluster CACTUs-MAML).
+
+Applicability to different tasks. CACTUs-MAML learns an effective prior for a variety of task types. This can be attributed to the application-agnostic task-generation process and the expressive power of MAML (Finn & Levine, 2018). We also observe that, despite all meta-learning models being trained for $N$ -way 1-shot classification of unsupervised tasks, the models work well for a variety of $M$ -way $K$ -shot tasks, where $M \ \leq \ N$ and $K \ \leq \ 5 0$ . As mentioned previously, the representation that CACTUs-ProtoNets learns is best suited for downstream tasks which match the single shot used for meta-training.
+
+Oracle. The penalty for not having ground truth labels to construct near-optimal tasks ranges from substantial to severe, depending on the difficulty of the downstream task. Easier downstream tasks (which have fewer classes and/or more supervision) incur less of a penalty. We conjecture that with such tasks, the difference in the usefulness of the priors matters less since the downstream task-specific evidence has more power to shape the posterior.
+
+Table 2: Results of unsupervised learning on miniImageNet and CelebA images, averaged over 1000 downstream human-designed tasks. CACTUs experiments use $k = 5 0 0$ for each of $P = 5 0$ partitions. Embedding cluster matching uses the same $k$ . For complete results with confidence intervals, see Tables 9 and 10 in Appendix F.   
+
+<table><tr><td></td><td colspan="4">miniImageNet</td><td>CelebA</td></tr><tr><td></td><td>(5,1)</td><td>(5,5)</td><td>(5,20)</td><td>(5,50)</td><td>(2,5)</td></tr><tr><td>Training from scratch</td><td>27.59%</td><td>38.48%</td><td>51.53%</td><td>59.63%</td><td>63.19%</td></tr><tr><td>BiGAN knn-nearest neighbors</td><td>25.56%</td><td>31.10%</td><td>37.31%</td><td>43.60%</td><td>56.15%</td></tr><tr><td>BiGAN linear classifier</td><td>27.08%</td><td>33.91%</td><td>44.00%</td><td>50.41%</td><td>58.44%</td></tr><tr><td>BiGAN MLP with dropout</td><td>22.91%</td><td>29.06%</td><td>40.06%</td><td>48.36%</td><td>56.26%</td></tr><tr><td>BiGAN cluster matching</td><td>24.63%</td><td>29.49%</td><td>33.89%</td><td>36.13%</td><td>56.20%</td></tr><tr><td>BiGAN CACTUs-MAML (ours)</td><td>36.24%</td><td>51.28%</td><td>61.33%</td><td>66.91%</td><td>74.98%</td></tr><tr><td>BiGAN CACTUs-ProtoNets (ours)</td><td>36.62%</td><td>50.16%</td><td>59.56%</td><td>63.27%</td><td>65.58%</td></tr><tr><td>DeepCluster knn-nearest neighbors</td><td>28.90%</td><td>42.25%</td><td>56.44%</td><td>63.90%</td><td>61.47%</td></tr><tr><td>DeepCluster linear classifier</td><td>29.44%</td><td>39.79%</td><td>56.19%</td><td>65.28%</td><td>59.57%</td></tr><tr><td>DeepCluster MLP with dropout</td><td>29.03%</td><td>39.67%</td><td>52.71%</td><td>60.95%</td><td>60.65%</td></tr><tr><td>DeepCluster cluster matching</td><td>22.20%</td><td>23.50%</td><td>24.97%</td><td>26.87%</td><td>51.51%</td></tr><tr><td>DeepCluster CACTUs-MAML (ours)</td><td>39.90%</td><td>53.97%</td><td>63.84%</td><td>69.64%</td><td>73.79%</td></tr><tr><td>DeepCluster CACTUs-ProtoNets (ours)</td><td>39.18%</td><td>53.36%</td><td>61.54%</td><td>63.55%</td><td>74.15%</td></tr><tr><td>Oracle-MAML (control)</td><td>46.81%</td><td>62.13%</td><td>71.03%</td><td>75.54%</td><td>87.10%</td></tr><tr><td>Oracle-ProtoNets (control)</td><td>46.56%</td><td>62.29%</td><td>70.05%</td><td>72.04%</td><td>85.13%</td></tr></table>
+
+Task construction ablation. As seen in Tables 3 and 4, CACTUs-MAML consistently outperforms Hyperplanes-MAML with any margin. We hypothesize that this is due to the issues with zero-margin Hyperplanes-MAML pointed out in Appendix C, and the fact that nonzero-margin HyperplanesMAML is able to use less of the meta-training split to generate tasks than CACTUs-MAML is. We find that using multiple partitions for CACTUs-MAML, while beneficial, is not strictly necessary. Using non-zero margin with Hyperplanes-MAML is crucial for miniImageNet, but not for Omniglot. We conjecture that the enforced degree of separation between classes is needed for miniImageNet because of the dataset’s high diversity. Meta-learning on random tasks or tasks derived from pixel-space clustering (Table 3) results in a prior that is much less useful than any other considered algorithm, including a random network initialization; evidently, practicing badly is worse than not practicing at all.
+
+Note on overfitting. Because of the combinatorially many unsupervised tasks we can create from multiple partitions of the dataset, we do not observe substantial overfitting to the unsupervised metatraining tasks. However, we observe that meta-training performance is sometimes worse than metatest time performance, which is likely due to a portion of the automatically generated tasks being based on nonsensical clusters (for examples, see Figure 2). Additionally, we find that, with a few exceptions, using multiple partitions has a regularizing effect on the meta-learner: a diverse task set reduces overfitting to the meta-training tasks and increases the applicability of the learned prior.
+
+Table 3: Ablation study of task construction methods on Omniglot. For a more complete set of results with confidence intervals, see Table 8 in Appendix F.   
+
+<table><tr><td>Algorithm</td><td>(5,1)</td><td>(5,5)</td><td>(20,1)</td><td>(20,5)</td></tr><tr><td>Random-MAML,P= 2400,k = 500</td><td>25.99%</td><td>25.74%</td><td>6.51%</td><td>6.74%</td></tr><tr><td>Pixels CACTUs-MAML,P=1, k = 500</td><td>30.55%</td><td>40.19%</td><td>12.05%</td><td>19.01%</td></tr><tr><td>ACAIHyperplanes-MAML,P= 2400,m= 0</td><td>62.34%</td><td>81.81%</td><td>39.30%</td><td>63.18%</td></tr><tr><td>ACAI Hyperplanes-MAML,P= 2400,m =1.2</td><td>62.44%</td><td>83.20%</td><td>41.86%</td><td>65.23%</td></tr><tr><td>ACAI CACTUs-MAML,P=1,k = 500</td><td>66.49%</td><td>85.60%</td><td>45.04%</td><td>69.14%</td></tr><tr><td>ACAICACTUs-MAML,P=100,k = 500</td><td>68.84%</td><td>87.78%</td><td>48.09%</td><td>73.36%</td></tr><tr><td>BiGANHyperplanes-MAML,P= 2400,m = 0</td><td>53.60%</td><td>74.60%</td><td>29.02%</td><td>50.77%</td></tr><tr><td>BiGANHyperplanes-MAML,P= 2400,m = 0.5</td><td>53.18%</td><td>73.55%</td><td>29.98%</td><td>50.14%</td></tr><tr><td>BiGANCACTUs-MAML,P=1,k= 500</td><td>55.92%</td><td>76.28%</td><td>32.44%</td><td>54.22%</td></tr><tr><td>BiGANCACTUs-MAML,P=100,k= 500</td><td>58.18%</td><td>78.66%</td><td>35.56%</td><td>58.62%</td></tr></table>
+
+Table 4: Ablation study of task construction methods on miniImageNet. For a more complete set of results with confidence intervals, see Table 9 in Appendix F.   
+
+<table><tr><td>Algorithm</td><td>(5,1)</td><td>(5,5)</td><td>(5,20)</td><td>(5,50)</td></tr><tr><td>BiGANHyperplanes-MAML,P= 4800,m =0</td><td>20.00%</td><td>20.00%</td><td>20.00%</td><td>20.00%</td></tr><tr><td>BiGAN Hyperplanes-MAML,P= 4800,m = 0.9</td><td>29.67%</td><td>41.92%</td><td>51.32%</td><td>54.72%</td></tr><tr><td>BiGAN CACTUs-MAML,P=1,k= 500</td><td>37.75%</td><td>52.59%</td><td>62.70%</td><td>67.98%</td></tr><tr><td>BiGANCACTUs-MAML,P= 50,k= 500</td><td>36.24%</td><td>51.28%</td><td>61.33%</td><td>66.91%</td></tr><tr><td>DeepCluster Hyperplanes-MAML,P= 4800,m = 0</td><td>20.02%</td><td>20.01%</td><td>20.00%</td><td>20.01%</td></tr><tr><td>DeepClusterHyperplanes-MAML,P= 4800,m= 0.1</td><td>35.85%</td><td>49.54%</td><td>60.68%</td><td>65.55%</td></tr><tr><td>DeepCluster CACTUs-MAML,P= 1,k = 500</td><td>38.75%</td><td>52.73%</td><td>62.72%</td><td>67.77%</td></tr><tr><td>DeepCluster CACTUs-MAML,P= 50,k = 500</td><td>39.90%</td><td>53.97%</td><td>63.84%</td><td>69.64%</td></tr></table>
+
+# 5 DISCUSSION
+
+We demonstrate that meta-learning on tasks produced using simple mechanisms based on embeddings improves upon the utility of these representations in learning downstream, human-specified tasks. We empirically show that this holds across benchmark datasets and tasks in the few-shot classification literature (Santoro et al., 2016; Ravi & Larochelle, 2017; Finn et al., 2018), task difficulties, and embedding learning methods while fixing key hyperparameters across all experiments.
+
+In a sense, CACTUs can be seen as a facilitating interface between an embedding learning method and a meta-learning algorithm. As shown in the results, the meta-learner’s performance significantly depends on the nature and quality of the task-generating embeddings. We can expect our method to yield better performance as the methods that produce these embedding functions improve, becoming better suited for generating diverse yet distinctive clusterings of the data. However, the gap between unsupervised and supervised meta-learning will likely persist because, with the latter, the meta-training task distribution is human-designed to mimic the expected evaluation task distribution as much as possible. Indeed, to some extent, supervised meta-learning algorithms offload the effort of designing and tuning algorithms onto the effort of designing and tuning task distributions. With its evaluation-agnostic task generation, CACTUs-based meta-learning trades off performance in specific use-cases for broad applicability and the ability to train on unlabeled data. In principle, CACTUs-based meta-learning may outperform supervised meta-learning when the latter is trained on a misaligned task distribution. We leave this investigation to future work.
+
+While we have demonstrated that $k$ -means is a broadly useful mechanism for constructing tasks from embeddings, it is unlikely that combinations of $k$ -means clusters in learned embedding spaces are universal approximations of arbitrary class definitions. An important direction for future work is to find examples of datasets and human-designed tasks for which CACTUs-based meta-learning results in ineffective downstream learning. This will result in better understanding of the practical scope of applicability for our method, and spur further development in automatic task construction mechanisms for unsupervised meta-learning.
+
+A potential concern of our experimental evaluation is that MNIST, Omniglot, and miniImageNet exhibit particular structure in the underlying class distribution (i.e., perfectly balanced classes), since they were designed to be supervised learning benchmarks. In more practical applications of machine learning, such structure would likely not exist. Our CelebA results indicate that CACTUs is effective even in the case of a dataset without neatly balanced classes or attributes. An interesting direction for future work is to better characterize the performance of CACTUs and other unsupervised pretraining methods with highly-unstructured, unlabeled datasets.
+
+Since MAML and ProtoNets produce nothing more than a learned representation, our method can be viewed as deriving, from a previous unsupervised representation, a new representation particularly suited for learning downstream tasks. Beyond visual classification tasks, the notion of using unsupervised pre-training is generally applicable to a wide range of domains, including regression, speech (Oord et al., 2018), language (Howard & Ruder, 2018), and reinforcement learning (Shelhamer et al., 2017). Hence, our unsupervised meta-learning approach has the potential to improve unsupervised representations for a variety of such domains, an exciting avenue for future work.
+
+# ACKNOWLEDGMENTS
+
+We thank Kelvin Xu, Richard Zhang, Brian Cheung, Ben Poole, Aaron van den Oord, Luke Metz, ¨ Siddharth Reddy, and the anonymous reviewers for feedback on an early draft of this paper.
+
+# REFERENCES
+
+Yoshua Bengio, Samy Bengio, and Jocelyn Cloutier. Learning a synaptic learning rule. In International Joint Conference on Neural Networks (IJCNN), 1991.
+
+Yoshua Bengio, Pascal Lamblin, Dan Popovici, and Hugo Larochelle. Greedy layer-wise training of deep networks. In Advances in Neural Information Processing Systems (NIPS), 2007.
+
+Yoshua Bengio, Aaron Courville, and Pascal Vincent. Representation learning: A review and new perspectives. IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI), 2013.
+
+David Berthelot, Colin Raffel, Aurko Roy, and Ian Goodfellow. Understanding and improving interpolation in autoencoders via an adversarial regularizer. arXiv preprint arXiv:1807.07543, 2018.
+
+Piotr Bojanowski and Armand Joulin. Unsupervised learning by predicting noise. In International Conference on Machine Learning (ICML), 2017.
+
+Mathilde Caron, Piotr Bojanowski, Armand Joulin, and Matthijs Douze. Deep clustering for unsupervised learning of visual features. In European Conference on Computer Vision (ECCV), 2018.
+
+Xi Chen, Yan Duan, Rein Houthooft, John Schulman, Ilya Sutskever, and Pieter Abbeel. InfoGAN: Interpretable representation learning by information maximizing generative adversarial nets. In Neural Information Processing Systems (NIPS), 2016.
+
+Brian Cheung, Jesse A Livezey, Arjun K Bansal, and Bruno A Olshausen. Discovering hidden factors of variation in deep networks. In International Conference on Learning Representations (ICLR), 2015.
+
+Adam Coates and Andrew Y Ng. Learning feature representations with k-means. In Neural Networks: Tricks of the Trade. Springer, 2012.
+
+Andrew M Dai and Quoc V Le. Semi-supervised sequence learning. In Neural Information Processing Systems (NIPS), 2015.
+
+Emily L Denton and Vighnesh Birodkar. Unsupervised learning of disentangled representations from video. In Neural Information Processing Systems (NIPS), 2017.
+
+Jeff Donahue, Philipp Krahenb ¨ uhl, and Trevor Darrell. Adversarial feature learning. In ¨ International Conference on Learning Representations (ICLR), 2017.
+
+Vincent Dumoulin, Ishmael Belghazi, Ben Poole, Olivier Mastropietro, Alex Lamb, Martin Arjovsky, and Aaron Courville. Adversarially learned inference. In International Conference on Learning Representations (ICLR), 2017.
+
+Dumitru Erhan, Yoshua Bengio, Aaron Courville, Pierre-Antoine Manzagol, Pascal Vincent, and Samy Bengio. Why does unsupervised pre-training help deep learning? Journal of Machine Learning Research (JMLR), 2010.
+
+Chelsea Finn and Sergey Levine. Meta-learning and universality: Deep representations and gradient descent can approximate any learning algorithm. In International Conference on Learning Representations (ICLR), 2018.
+
+Chelsea Finn, Pieter Abbeel, and Sergey Levine. Model-agnostic meta-learning for fast adaptation of deep networks. In International Conference on Machine Learning (ICML), 2017.
+
+Chelsea Finn, Kelvin Xu, and Sergey Levine. Probabilistic model-agnostic meta-learning. In Neural Information Processing Systems (NIPS), 2018.
+
+Vikas K Garg and Adam Kalai. Supervising unsupervised learning. arXiv preprint arXiv:1709.05262, 2017.
+
+Abhishek Gupta, Benjamin Eysenbach, Chelsea Finn, and Sergey Levine. Unsupervised metalearning for reinforcement learning. arXiv preprint arXiv:1806.04640, 2018.
+
+Trevor Hastie, Robert Tibshirani, and Jerome Friedman. Unsupervised learning. In The Elements of Statistical Learning. Springer, 2009.
+
+Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Identity mappings in deep residual networks. European Conference on Computer Vision (ECCV), 2016.
+
+Irina Higgins, Loic Matthey, Arka Pal, Christopher Burgess, Xavier Glorot, Matthew Botvinick, Shakir Mohamed, and Alexander Lerchner. $\beta$ -VAE: Learning basic visual concepts with a constrained variational framework. In International Conference on Learning Representations (ICLR), 2017.
+
+Geoffrey E Hinton, Simon Osindero, and Yee-Whye Teh. A fast learning algorithm for deep belief nets. Neural Computation, 2006.
+
+Jeremy Howard and Sebastian Ruder. Universal language model fine-tuning for text classification. In Association for Computational Linguistics (ACL), 2018.
+
+Sergey Ioffe and Christian Szegedy. Batch normalization: Accelerating deep network training by reducing internal covariate shift. In International Conference on Machine Learning (ICML), 2015.
+
+Jaehong Kim, Sangyeul Lee, Sungwan Kim, Moonsu Cha, Jung Kwon Lee, Youngduck Choi, Yongseok Choi, Dong-Yeon Cho, and Jiwon Kim. Auto-Meta: Automated gradient based meta learner search. arXiv preprint arXiv:1806.06927, 2018.
+
+Diederik Kingma and Jimmy Ba. Adam: A method for stochastic optimization. In Proceedings of the 3rd International Conference on Learning Representations (ICLR), 2014.
+
+Diederik P Kingma, Shakir Mohamed, Danilo Jimenez Rezende, and Max Welling. Semi-supervised learning with deep generative models. In Neural Information Processing Systems (NIPS), 2014.
+
+Philipp Krahenb ¨ uhl, Carl Doersch, Jeff Donahue, and Trevor Darrell. Data-dependent initializa- ¨ tions of convolutional neural networks. In International Conference on Learning Representations (ICLR), 2016.
+
+Quoc V Le, Marc’Aurelio Ranzato, Rajat Monga, Matthieu Devin, Gregory S. Corrado, Kai Chen, Jeffrey Dean, and Andrew Y Ng. Building high-level features using large scale unsupervised learning. In International Conference on Acoustics, Speech and Signal Processing (ICASSP), 2013.
+
+Michael F Mathieu, Junbo Jake Zhao, Junbo Zhao, Aditya Ramesh, Pablo Sprechmann, and Yann LeCun. Disentangling factors of variation in deep representation using adversarial training. In Neural Information Processing Systems (NIPS), 2016.
+
+Luke Metz, Niru Maheswaranathan, Brian Cheung, and Jascha Sohl-Dickstein. Learning to learn without labels. In International Conference on Learning Representations (ICLR), 2018.
+
+Tsendsuren Munkhdalai and Hong Yu. Meta networks. In International Conference on Machine Learning (ICML), 2017.
+
+Devang K Naik and RJ Mammone. Meta-neural networks that learn by learning. In International Joint Conference on Neural Networks (IJCNN), 1992.
+
+Avital Oliver, Augustus Odena, Colin Raffel, Ekin D Cubuk, and Ian J Goodfellow. Realistic evaluation of deep semi-supervised learning algorithms. In International Conference on Learning Representations (ICLR), 2018.
+
+Aaron van den Oord, Yazhe Li, and Oriol Vinyals. Representation learning with contrastive predictive coding. arXiv preprint arXiv:1807.03748, 2018.
+
+Alec Radford, Luke Metz, and Soumith Chintala. Unsupervised representation learning with deep convolutional generative adversarial networks. In International Conference on Learning Representations (ICLR), 2016.
+
+Alec Radford, Karthik Narasimhan, Tim Salimans, and Ilya Sutskever. Improving language understanding by generative pre-training. Preprint, 2018.
+
+Prajit Ramachandran, Peter J Liu, and Quoc V Le. Unsupervised pretraining for sequence to sequence learning. In Empirical Methods in Natural Language Processing (EMNLP), 2017.
+
+Marc’Aurelio Ranzato, Christopher Poultney, Sumit Chopra, and Yann LeCun. Efficient learning of sparse representations with an energy-based model. In Neural Information Processing Systems (NIPS), 2006.
+
+Antti Rasmus, Mathias Berglund, Mikko Honkala, Harri Valpola, and Tapani Raiko. Semisupervised learning with ladder networks. In Neural Information Processing Systems (NIPS), 2015.
+
+Sachin Ravi and Hugo Larochelle. Optimization as a model for few-shot learning. In International Conference on Learning Representations (ICLR), 2017.
+
+Scott Reed, Kihyuk Sohn, Yuting Zhang, and Honglak Lee. Learning to disentangle factors of variation with manifold interaction. In International Conference on Machine Learning (ICML), 2014.
+
+Olga Russakovsky, Jia Deng, Hao Su, Jonathan Krause, Sanjeev Satheesh, Sean Ma, Zhiheng Huang, Andrej Karpathy, Aditya Khosla, Michael Bernstein, Alexander C. Berg, and Fei-Fei Li. ImageNet large scale visual recognition challenge. International Journal of Computer Vision (IJCV), 2015.
+
+Tim Salimans, Ian Goodfellow, Wojciech Zaremba, Vicki Cheung, Alec Radford, and Xi Chen. Improved techniques for training GANs. In Neural Information Processing Systems (NIPS), 2016.
+
+Adam Santoro, Sergey Bartunov, Matthew Botvinick, Daan Wierstra, and Timothy Lillicrap. Metalearning with memory-augmented neural networks. In International Conference on Machine Learning (ICML), 2016.
+
+Jurgen Schmidhuber. ¨ Evolutionary principles in self-referential learning. PhD thesis, Institut fur¨ Informatik, Technische Universitat M¨ unchen, 1987.¨
+
+Evan Shelhamer, Parsa Mahmoudieh, Max Argus, and Trevor Darrell. Loss is its own reward: Selfsupervision for reinforcement learning. In International Conference on Learning Representations (ICLR), 2017.
+
+Jake Snell, Kevin Swersky, and Richard S Zemel. Prototypical networks for few-shot learning. In Neural Information Processing Systems (NIPS), 2017.
+
+Nitish Srivastava, Geoffrey Hinton, Alex Krizhevsky, Ilya Sutskever, and Ruslan Salakhutdinov. Dropout: A simple way to prevent neural networks from overfitting. Journal of Machine Learning Research (JMLR), 2014.
+
+Pascal Vincent, Hugo Larochelle, Yoshua Bengio, and Pierre-Antoine Manzagol. Extracting and composing robust features with denoising autoencoders. In International Conference on Machine Learning (ICML), 2008.
+
+Pascal Vincent, Hugo Larochelle, Isabelle Lajoie, Yoshua Bengio, and Pierre-Antoine Manzagol. Stacked denoising autoencoders: Learning useful representations in a deep network with a local denoising criterion. Journal of Machine Learning Research (JMLR), 2010.
+
+Oriol Vinyals, Charles Blundell, Timothy Lillicrap, Koray Kavukcuoglu, and Daan Wierstra. Matching networks for one shot learning. In Neural Information Processing Systems (NIPS), 2016.
+
+Dong Yu, Li Deng, and George Dahl. Roles of pre-training and fine-tuning in context-dependent DBN-HMMs for real-world speech recognition. In NIPS Workshop on Deep Learning and Unsupervised Feature Learning, 2010.
+
+Richard Zhang, Phillip Isola, and Alexei A Efros. Split-brain autoencoders: Unsupervised learning by cross-channel prediction. In Computer Vision and Pattern Recognition (CVPR), 2017.
+
+Xiaojin Zhu. Semi-supervised learning. In Encyclopedia of Machine Learning. Springer, 2011.
+
+# APPENDIX A THE EMBEDDING LEARNING ZOO
+
+We evaluate four distinct methods from prior work for learning the task-generating embeddings.
+
+In adversarially constrained autoencoder interpolation (ACAI), a convolutional autoencoder’s pixelwise $L ^ { 2 }$ loss is regularized with a term encouraging meaningful interpolations in the latent space (Berthelot et al., 2018). Specifically, a critic network takes as input a synthetic image generated from a convex combination of the latents of two dataset samples, and regresses to the mixing factor. The decoder of the autoencoder and the generator for the critic are one and the same. The regularization term is minimized when the autoencoder fools the critic into predicting that the synthetic image is a real sample.
+
+The bidirectional GAN (BiGAN) is an instance of a generative-adversarial framework in which the generator produces both synthetic image and embedding from real embedding and image, respectively (Donahue et al., 2017; Dumoulin et al., 2017). Discrimination is done in joint imageembedding space.
+
+The DeepCluster method does discriminative clustering by alternating between clustering the features of a convolutional neural network and using the clusters as labels to optimize the network weights via backpropagating a standard classification loss (Caron et al., 2018).
+
+The InfoGAN framework conceptually decomposes the generator’s input into a latent code and incompressible noise (Chen et al., 2016). The structure of the latent code is hand-specified based on knowledge of the dataset. The canonical GAN minimax objective is regularized with a mutual information term between the code and the generated image. In practice, this term is optimized using variational inference, involving the approximation of the posterior with an auxiliary distribution $Q$ (code|image) parameterized by a recognition network.
+
+Whereas ACAI explicitly optimizes pixel-wise reconstruction error, BiGAN only encourages the fidelity of generated image and latent samples with respect to their respective prior distributions. While InfoGAN also encourages the fidelity of generated images, it leverages domain-specific knowledge to impose a favorable structure on the embedding space and information-theoretic methods for optimization. DeepCluster departs from the aforementioned methods in that it is not concerned with generation or decoding, and only seeks to learn general-purpose visual features by way of end-to-end discriminative clustering.
+
+# APPENDIX B DATASET INFORMATION
+
+The Omniglot dataset consists of 1623 characters each with 20 hand-drawn examples. Ignoring the alphabets from which the characters originate, we use 1100, 100, and 423 characters for our meta-training, meta-validation, and meta-testing splits. The miniImageNet dataset consists of 100 classes each with 600 examples. The images are predominantly natural and realistic. We use the same meta-training/meta-validation/meta-testing splits of 64/16/20 classes as proposed by Ravi & Larochelle (2017). The CelebA dataset includes 202,599 facial images of celebrities and 40 binary attributes that annotate every image. We follow the prescribed 162,770/19,867/19,962 data split.
+
+For Omniglot and miniImageNet, supervised meta-learning tasks and evaluation tasks are constructed exactly as detailed in Section 2.2: for an $N$ -way $K$ -shot task with $Q$ queries per class, we sample $N$ classes from the data split and $K + Q$ datapoints per class, labeling the task’s data with a random permutation of $N$ one-hot vectors.
+
+For CelebA, we consider binary classification tasks (i.e., 2-way), each defined by 3 attributes and an ordering of 3 Booleans, one for each attribute. Every image in a task-specific class shares all task-specific attributes with each other and none with images in the other class. For example, the task illustrated in Figure 2 involves distinguishing between images whose subjects satisfy not Sideburns, Straight Hair, and not Young, and those whose subjects satisfy Sideburns, not Straight Hair, and Young. To keep with the idea of having distinct classes for meta-training and meta-testing, we split the task-defining attributes. For the supervised meta-learning oracle, we construct meta-training tasks from the first 20 attributes (when alphabetically ordered), meta-validation tasks from the next 10, and meta-testing tasks from the last 10. Discarding tasks with too few examples in either class, this results in 4287, 391, and 402 task prototypes (but many more possible tasks). We use the same meta-test time tasks to evaluate the unsupervised methods. We only consider assessment with 5-shot tasks because, given that there are multiple attributes other than the task-defining ones, any 1-shot task is likely to be ill-defined.
+
+# APPENDIX C TASK CONSTRUCTION VIA RANDOM HYPERPLANES
+
+Given a set of embedding points $\left\{ \mathbf { z } _ { i } \right\}$ in a space $\mathcal { Z }$ , a simple way of defining a partition $\mathcal { P } = \{ \mathcal { C } _ { c } \}$ on $\left\{ \mathbf { z } _ { i } \right\}$ is to use random hyperplanes to slice $\mathcal { Z }$ into subspaces and assign the embeddings that lie in the $c$ -th subspace to subset $\mathcal { C } _ { c }$ . However, a hyperplane slicing can group together two arbitrarily far embeddings, or separate two arbitrarily close ones; given our assumption that good embedding spaces have a semantically meaningful metric, this creates ill-defined classes. This problem can be partially alleviated by extending the hyperplane boundaries with a non-zero margin, as empirically shown in Section 4.2.
+
+We now describe how to generate tasks via random hyperplanes in the embedding space. We first describe a procedure to generate a partition $\mathcal { P }$ of the set of embeddings $\left\{ \mathbf { z } _ { i } \right\}$ for constructing metatraining tasks. A given hyperplane slices the embedding space into two, so for an $N$ -way task, we need $\bar { H } = \lceil \log _ { 2 } \bar { N } \rceil$ hyperplanes to define sufficiently many subsets/classes for a task. To randomly define a hyperplane in $d$ -dimensional embedding space, we sample a normal vector $\mathbf { n }$ and a point on the plane $\mathbf { z } _ { 0 }$ , each with $d$ elements. For an embedding point $\mathbf { z }$ , the signed point-plane distance is given by $\frac { \mathbf { n } } { \left| \mathbf { n } \right| _ { 2 } } \cdot \left( \mathbf { z } - \mathbf { z } _ { 0 } \right)$ . Defining $H$ hyperplanes in this manner, we discard embeddings for which the signed point-plane distance to any of the $H$ hyperplanes lies within $( - m , m )$ , where $m$ is a desired margin. The $H$ hyperplanes collectively define $\mathbf { \bar { 2 } } ^ { H }$ subspaces. We assign embedding points in the $c$ -th subspace to subset $\mathcal { C } _ { c }$ . We define the partition as $\mathcal { P } = \{ \mathcal { C } _ { c } \}$ . We prune subsets that do not have at least $R = K _ { \mathrm { m - t r } } + Q$ members, and check that the partition has at least $N$ remaining subsets; if not, we reject the partition and restart the procedure. After obtaining partitions $\{ \mathcal P _ { p } \}$ , meta-training tasks can be generated by following Algorithm 1 from Line 4.
+
+In terms of practical implementation, we pre-compute 1000 hyperplanes and pruned pairs of subsets of $\left\{ \mathbf { z } _ { i } \right\}$ . We generate partitions by sampling combinations of the hyperplanes and taking intersections of their associated subsets to define the elements of the partition. We determine the number of partitions needed for a given Hyperplanes-MAML run by the number of meta-training tasks desired for the meta-learner: we fix 100 tasks per partition.
+
+# APPENDIX D MNIST EXPERIMENTS
+
+The MNIST dataset consists of 70,000 hand-drawn examples of the 10 numerical digits. Our split respects the original MNIST 60,000/10,000 training/testing split. We assess on 10-way classification tasks. This setup results in examples from all 10 digits being present for both meta-training and meta-testing, making the probem setting essentially equivalent to that of semi-supervised learning sans a fixed permutation of the labels. The MNIST scenario is thus a special case of the problem setting considered in the rest of the paper. For MNIST, we only experiment with MAML as the meta-learning algorithm.
+
+For ACAI and InfoGAN we constructed the meta-validation split from the last 5,000 examples of the meta-training split; for BiGAN this figure was 10,000. After training the ACAI model and inferring embeddings, manually assigning labels to 10 clusters by inspection results in a classification accuracy of $9 6 . 0 0 \%$ on the testing split. As the ACAI authors observe, we found it important to whiten the ACAI embeddings before clustering. The same metric for the InfoGAN embedding (taking an argmax over the categorical dimensions instead of actually running clustering) is $9 6 . 8 3 \%$ . Note that these results are an upper-bound for embedding cluster matching. To see this, consider the 10-way 1-shot scenario. 1 example sampled from each cluster is insufficient to guarantee the optimal label for that cluster; 1 example sampled from each label is not guaranteed to each end up in the optimal category.
+
+Aside from CACTUs-MAML, embedding $k _ { \mathrm { n n } }$ -nearest neighbors, embedding linear classifier, and embedding direct clustering, we also ran CACTUs-MAML on embeddings instead of raw images, using a simple model with 2 hidden layers with 64 units each and ReLU activation, and all other MAML hyperparameters being the same as in Table 5.
+
+Departing from the fixed $k = 5 0 0$ used for all other datasets, we deliberately use $k = 1 0$ to better understand the limitations of CACTUs-MAML. The results can be seen in Table 7 in Appendix B. In brief, with the better embeddings (ACAI and InfoGAN), there is only little benefit of CACTUsMAML over embedding cluster matching. Additionally, even in the best cases, CACTUs-MAML falls short of state-of-the-art semi-supervised learning methods.
+
+# APPENDIX E HYPERPARAMETERS AND ARCHITECTURES
+
+# E.1 MAML
+
+Table 5: MAML hyperparameter summary.   
+
+<table><tr><td>Hyperparameter</td><td>MNIST</td><td>Omniglot</td><td>miniImageNet</td><td>CelebA</td></tr><tr><td>Input size</td><td>28×28</td><td>28×28</td><td>84 ×84×3</td><td>84 ×84×3</td></tr><tr><td>Outer (meta) learning rate</td><td>0.001</td><td>0.001</td><td>0.001</td><td>0.001</td></tr><tr><td>Inner learning rate</td><td>0.05</td><td>0.05</td><td>0.05</td><td>0.05</td></tr><tr><td>Task batch size</td><td>8</td><td>8</td><td>8</td><td>8</td></tr><tr><td>Inner adaptation steps (meta-training)</td><td>5</td><td></td><td>5</td><td>5</td></tr><tr><td>Meta-training iterations</td><td>30,000</td><td>5 30,000</td><td>60,000</td><td>60,000</td></tr><tr><td>Adaptation steps (evaluation)</td><td>50</td><td>50</td><td>50</td><td>50</td></tr><tr><td>Classes per task (meta-training)</td><td>10</td><td>20</td><td>5</td><td>2</td></tr><tr><td>Shots per class (meta-training)</td><td>1</td><td>1</td><td>1</td><td>1</td></tr><tr><td>Queries per class</td><td>5</td><td>5</td><td>5</td><td>5</td></tr></table>
+
+For MNIST and Omniglot we use the same 4-block convolutional architecture as used by Finn et al. (2017) for their Omniglot experiments, but with 32 filters (instead of 64) for each convolutional layer for consistency with the model used for miniImageNet and CelebA, which is the same as what Finn et al. (2017) used for their miniImageNet experiments. When evaluating the meta-learned 20-way Omniglot model with 5-way tasks, we prune the unused output dimensions. The outer optimizer is Adam (Kingma & Ba, 2014), and the inner optimizer is SGD. We build on the authors’ publicly available codebase found at https://github.com/cbfinn/maml.
+
+When using batch normalization (Ioffe & Szegedy, 2015) to process a task’s training or query inputs, we observe that using only 1 query datapoint per class can allow the model to exploit batch statistics, learning a strategy analogous to a process of elimination that causes significant, but spurious, improvement in accuracy. To mitigate this, we fix 5 queries per class for every task’s evaluation phase, meta-training or meta-testing.
+
+# E.2 PROTONETS
+
+Table 6: ProtoNets hyperparameter summary.   
+
+<table><tr><td>Hyperparameter</td><td>Omniglot</td><td>miniImageNet</td><td>CelebA</td></tr><tr><td>Input size</td><td>28×28</td><td>84×84×3</td><td>84×84×3</td></tr><tr><td>Learning rate</td><td>0.001</td><td>0.001</td><td>0.001</td></tr><tr><td>Task batch size</td><td>1</td><td>1</td><td>1</td></tr><tr><td>Training iterations</td><td>30,000</td><td>60,000</td><td>60,000</td></tr><tr><td>Classes per task (meta-training)</td><td>20</td><td>5</td><td>2</td></tr><tr><td>Shots per class (meta-training)</td><td>1</td><td>1</td><td>1</td></tr><tr><td>Queries per class (meta-training/meta-testing)</td><td>15/5</td><td>15/5</td><td>15/5</td></tr></table>
+
+For the three considered datasets we use the same architecture as used by Snell et al. (2017) for their Omniglot and miniImageNet experiments. This is a 4-block convolutional architecture with each block consisting of a convolutional layer with $6 4 \ 3 \times 3$ filters, stride 1, and padding 1, followed by BatchNorm, ReLU activation, and $2 \times 2$ MaxPooling. The ProtoNets embedding is simply the flattened output of the last block. We follow the authors and use the Adam optimizer, but do not use a learning rate scheduler. We build upon the authors’ publicly available codebase found at https://github.com/jakesnell/prototypical-networks.
+
+# E.3 CACTUS
+
+For Omniglot, miniImageNet, and CelebA we fix the number of clusters $k$ to be 500. For Omniglot we choose the number of partitions $P = 1 0 0$ , but in the interest of keeping runtime manageable, choose $P = 5 0$ for miniImageNet and CelebA.
+
+# E.4 USE OF UNSUPERVISED LEARNING METHODS
+
+ACAI (Berthelot et al., 2018): We run ACAI for MNIST and Omniglot. We pad the images by 2 and use the authors’ architecture. We use a 256-dimensional embedding for all datasets. We build upon the authors’ publicly available codebase found at https://github.com/ brain-research/acai.
+
+We unsuccessfully try running ACAI on $6 4 \times 6 4$ miniImageNet and CelebA. To facilitate this input size, we add one block consisting of two convolutional layers (512 filters each) and one downsampling/upsampling layer to the encoder and decoder. However, because of ACAI’s pixel-wise reconstruction loss, for these datasets the ACAI embedding prioritizes information about the few “features” that dominate the reconstruction pixel count, resulting in clusters that only corresponded to a limited range of factors, such as background color and pose. For curiosity’s sake, we tried running meta-learning on tasks derived from these uninteresting clusters anyways, and found that the meta-learner quickly produced a learning procedure that obtained high accuracy on the meta-training tasks. However, this learned prior was not useful for solving downstream tasks.
+
+BiGAN (Donahue et al., 2017): For MNIST, we follow the BiGAN authors and specify a uniform 50-dimensional prior on the unit hypercube for the latent. The BiGAN authors use a 200- dimensional version of the same prior for their ImageNet experiments, so we follow suit for Omniglot, miniImageNet, and CelebA. For MNIST and Omniglot, we use the permutation-invariant architecture (i.e. fully connected layers only) used by the authors for their MNIST results; for miniImageNet and CelebA, we randomly crop to $6 4 \times 6 4$ and use the AlexNet-inspired architecture used by Donahue et al. (2017) for their ImageNet results. We build upon the authors’ publicly available codebase found at https://github.com/jeffdonahue/bigan.
+
+DeepCluster (Caron et al., 2018): We run DeepCluster for miniImageNet and CelebA, which we respectively randomly crop and resize to $6 4 \times 6 4$ . We modify the first layer of the AlexNet architecture used by the authors to accommodate this input size. We follow the authors and use the input to the (linear) output layer as the embedding. These are 4096-dimensional, so we follow the authors and apply PCA to reduce the dimensionality to 256, followed by whitening. We build upon the authors’ publicly available codebase found at https://github.com/facebookresearch/ deepcluster.
+
+InfoGAN (Chen et al., 2016): We only run InfoGAN for MNIST. We follow the InfoGAN authors and specify the product of a 10-way categorical distribution and a 2-dimensional uniform distribution as the latent code. We use the authors’ architecture. Given an image, we use the recognition network to obtain its embedding. We build upon the authors’ publicly available codebase found at https: //github.com/openai/InfoGAN.
+
+# APPENDIX F EXPERIMENTAL RESULTS
+
+This section containsfull experimental results for the MNIST, Omniglot, miniImageNet, and CelebA datasets, including consolidated versions of the tables found in the main text. The metric is classification accuracy averaged over 1000 tasks based on human-specified labels of the testing split, with $9 5 \%$ confidence intervals. $d$ : dimensionality of embedding, $h$ : number of hidden units in a fully connected layer, $k$ : number of clusters in a partition, $P$ : number of partitions used during meta-learning, $m$ : margin on boundary-defining hyperplanes.
+
+<table><tr><td rowspan=1 colspan=1>(0101)</td><td rowspan=4 colspan=1>% LI&#x27;0干06&#x27;16% 01.0581555% 71&#x27;0 干 10&#x27;96% 710 67&#x27;96% II&#x27;0干08&#x27;96% L70干08&#x27;88% 11011.55% 910 76.56% 71&#x27;0干80&#x27;96% 110 干 9596% 780 6444% 18031.95% 7109 8535% 40.0 99.56% 870 LL&#x27;46C[ = y[= T (sani) sSuiPPq Tt Tr-shyt[ = y[ = T (sA) P T T-SAT r rrrrret gaasetr Srrpeirrlrer iaerrertrrnrrltgFaraeregarrgrrirrettA4=P444</td><td rowspan=2 colspan=1>% 0509 86:10% 1£&#x27;0干0L09%50055672% 67089&#x27;89% 0000658% 104530% 01004535% 44061.44% 141 65:14%70015:35%03.0335.55% 0ε0干 S0&#x27;LL% 51081:14%41044:44</td><td rowspan=1 colspan=1>% 700 47.99% 11&#x27;0干 58&#x27;96% 01&#x27;0干 66&#x27;96% 010 干 87&#x27; L6% 600 干13&#x27; L6% 01756% 010  6</td><td rowspan=1 colspan=1>% L0&#x27;0 干 1S&#x27;86</td></tr><tr><td rowspan=2 colspan=1>(s01)</td><td rowspan=2 colspan=1>% LI&#x27;0干 S0&#x27;96% 170干19&#x27;96% 110T 6696% 010干8126% 010776% 010干£1L6% 010干 80&#x27;L6</td><td rowspan=2 colspan=1>% L00干1S&#x27;86</td></tr><tr><td rowspan=1 colspan=1>% 44061.44% 141 65:14%70015:35%03.0335.55% 0ε0干 S0&#x27;LL% 51081:14%41044:44</td></tr><tr><td rowspan=1 colspan=1>(101)(tous em)1oila</td><td rowspan=1 colspan=1>%230325:65% 680 F980%57025224%944144% 14011.64%85033335% 3：271:31[[ = y O = T‘(sIni) suipP-qs u Tr-t[ = y  = (sani) supPqa uu Trn-st[[ = y OII = T (I1I) SSA- V T-NC[ = yI=T sa   -T=grrrraree geasepr rripegahrlerserrerarrnriirenggareregaerrriipetP=P-14</td><td rowspan=1 colspan=1>% 155311:55% 7408126% 31.074.16%050333:5% 610干80&#x27;96% LI0干 6996% L10 干8596C = y[= T(sa) supq c -[[ = y O = T‘(sni) sYuiPPqs uu Tr-sntC = y O = T () P 1 TI-sC=y[= T(I)A T T-sAg rrirpre aaetr rrirearrger serrerarrnriirertggaraerrarrrriiiertt=p 4</td><td rowspan=1 colspan=1>% L1&#x27;0干Iε&#x27;L6Surreend pasiersnnGOLnaier Giii-larei</td></tr></table>
+
+ers per convolutional layer, 3×data augmentation, and folded the validation set   
+  
+
+<table><tr><td rowspan=7 colspan=1>(s 0)(107)(s‘s)(1‘s)(ous em)oiiral</td><td rowspan=7 colspan=1>% 944% 810 449% 67&#x27;0干 10&#x27;61% 110013.55% 810 11&#x27;9%1102555% 6908&#x27;44% 690441% 1∠&#x27;06100%20035174% 32056577%33357530PPT =γ[= TIIII-SALAI SI11.1ay = y  = 111A1--4111rres ionnaieraBeesees</td><td rowspan=4 colspan=1>% 90.038399% 33.333339% 350 79:11  %80085.8  % 990 41.69 % 1333 3334% L00干 L7&#x27;996%83033139% 670227/4%3330730 % 30033331% 8109835% 4404 44.14  % 1 1124</td><td rowspan=1 colspan=1>% 90001.95% 2102 78.14% 401141% L£&#x27;0干90% 6£&#x27;0干 LL&#x27;0  %37:227755% 8057788</td><td rowspan=1 colspan=2>% 0105 6796% L0&#x27;0干18&#x27;86</td></tr><tr><td rowspan=1 colspan=1>% 90001.95% 2102 78.14% 401141% L£&#x27;0干90% 6£&#x27;0干 LL&#x27;0  %37:227755% 8057788% 30:0377.3050</td><td rowspan=2 colspan=2>% 0105 6796% L0&#x27;0干18&#x27;86% 700339:55% 810533.55</td></tr><tr><td rowspan=2 colspan=1>%11033135% 190 08:27% 770955.66%71224115% 210370.65% 200586.655210224555%335395559% 33033151</td></tr><tr><td></td><td rowspan=1 colspan=1>土+</td></tr><tr><td rowspan=3 colspan=1>% LS0干9118%8107818%S9007&#x27;LL% LL&#x27;0干60&#x27;IL% 09&#x27;0干18&#x27;18%8703355% 353355.55% 0S&#x27;0 干8L&#x27;L8% 3903 8138% 1113 1511% 711 819% 7.0 16:14  %7.055  % 08&#x27;0 干 64&#x27;99  % 480  71896260 =   =  () T-       CCT = y TT=1 (In) s11T1II-SSCC = γ OII = 1 ‘(S.I1I) T1I1-SALOCY =γ[=T (SIAI) TIII-SAI8 = Y inodop p PT SuippggregreemysppegarierereereirrA44=P4444</td><td rowspan=1 colspan=1></td><td></td></tr><tr><td rowspan=1 colspan=1>% I∠&#x27;0干 90&#x27;89% 99&#x27;0 干7L&#x27;89% 6/0 9579%8/07985% 69&#x27;009&#x27;L% 6905555% 19085.94% S9&#x27;0 干99&#x27;8L% L0干69</td><td rowspan=2 colspan=2>% 710£8&#x27;86% 60&#x27;0干85&#x27;66% 11.03.9555% 773523:32Surureree peseeedn  GDi Drirr eoeiiioriireri（OLDLier Girr-aarei</td></tr><tr><td rowspan=1 colspan=1>% 71 1171% 1742 1185% 6104544% 080 964% 5039.02 % 0807650% 18081&#x27;89% 71:044440 =  O = P (I) T-sdHP&#x27; =   = P(s) -s-L CC =y OII = 1 (SIAN) TSII-SA/S0C = y =  (i) sTod-snt8 = Y inidoip pm d SuippqC =yI =1 (SIA) T-SAagrrpaee easeta Srrplgrhgier neeennr neipreggarerrgrereippegPP =P 4400</td></tr></table>
+
+denotes a 95% confidence interval. d: dimensionality of embedding,
+
+<table><tr><td rowspan=1 colspan=1>(0s&#x27;s)(2 &#x27;0)(s‘s)(1‘s)(ous Xem)Woirai</td><td rowspan=1 colspan=1>% 5/.05 39.61% 533112% 9908589% 67003735Yrirlrriiir reirraperreeees</td><td rowspan=1 colspan=1>% 230309.3% 31:15:11% 1/.0 91&#x27;81%55333.99% 00&#x27;0干00&#x27;  % 89&#x27;0干 86&#x27;L% 89&#x27;0干 16&#x27;99% L9:0 17:99546%344231.21% 100404% 7/0 9000     %8900L&#x27;79%57003515% 5703.31%£9090:65%81096565% 0000000 %15457171% 850 87.15%05099511% 3513 95:75% 1 8 2 470036.74     %990干8000干（09.4士22.5519&#x27;6725&#x27;330 = w Xs =  (s) Ti-H6&#x27;0 = 1 O = P (s) T--H CC1 = yO1= P (SIA4） PIII-SA/CT1 =yFT= (sT) s1TI-sg = Y iniip pi P grippC =y[=1 (SIA) TI-SAS =yrrrpaee gaisnpa Seippegger nsereny iprgasereresteippeggPP=P4440</td><td rowspan=1 colspan=1>% 8030639%5500854% £9&#x27;0干 56:09% 1555585% 70&#x27;0干I0% 99055% 79&#x27;0LL% £90干 5969% 553515:396566% 1440444.94%5506554  %41:045.44       % 0105809% 89045:1989&#x27;0921% 42:015%0706/:66 277333755 %1/044:64% 71172% 010 1690% 3303.73.51% 17:5--55.85%%   %%119.03 30.67990干  5503066120.00555555830 =  Os = P (sI1I) Tt--AdL1HT&#x27;0 = 1s= P sa) T-H C=yE1=PIA S-ACC = y I= P （(I) sTTd-SHg = Y iniip pi P grippaggC1 =yT=1A） T-SNarrpiereisepaSerpggrhler errrairrireigrrsesereesseippeggz =pashisea4</td><td rowspan=1 colspan=1>% 77:045:14% 09044% 69000.14% 590F 50.00% 7/0511.79% 1/0 6777% LL&#x27;0干 18&#x27;94% 9/0 95.94Surererrer lseeetdnnODTLr) Giii-rriOrnerer eeeinerrceeret</td></tr></table>
+
+denotes a 95% confidence interval. d: dimensionality of embeddi
+
+Table 10: CelebA facial attribute classification results averaged over 1000 tasks. $\pm$ denotes a $9 5 \%$ confidence interval. $d$ : dimensionality of embedding, $h$ : number of hidden units in a fully connected layer, $k$ : number of clusters in a partition, $P$ : number of partitions used during meta-learning.   
+
+<table><tr><td>Algorithm</td><td>(2,5)</td></tr><tr><td>Baselines Training from scratch</td><td>63.19 ± 1.06 %</td></tr><tr><td>BiGAN,d= 200</td><td></td></tr><tr><td>Embedding knn-nearest neighbors Embedding linearclassifier</td><td>56.15 ± 0.89 % 58.44 ± 0.90 %</td></tr><tr><td>Embedding MLP with dropout,h = 128</td><td>56.26 ± 0.94 %</td></tr><tr><td>Embedding cluster matching,k = 500</td><td>56.20 ±1.00 %</td></tr><tr><td>CACTUs-MAML(ours),P= 50,k= 500 CACTUs-ProtoNets (ours),P= 50,k = 500</td><td>74.98 ± 1.02 %</td></tr><tr><td></td><td>65.58 ± 1.04 %</td></tr><tr><td>DeepCluster,d= 256 Embedding knn-nearest neighbors</td><td></td></tr><tr><td>Embedding linear classifier</td><td>61.47 ± 0.99 % 59.57 ± 0.98 %</td></tr><tr><td>Embedding MLP with dropout,h = 128</td><td>60.65 ± 0.98 %</td></tr><tr><td>Embedding cluster matching,k = 500</td><td>51.51 ± 0.89 %</td></tr><tr><td>CACTUs-MAML (ours),P= 50,k = 500</td><td>73.79 ± 1.01 %</td></tr><tr><td>CACTUs-ProtoNets (ours),P= 50,k = 500</td><td>74.15 ± 1.02 %</td></tr><tr><td></td><td></td></tr><tr><td>Supervised meta-learning</td><td></td></tr><tr><td>Oracle-MAML (control)</td><td>87.10 ± 0.85 %</td></tr><tr><td></td><td></td></tr><tr><td>Oracle-ProtoNets (control)</td><td>85.13 ± 0.92 %</td></tr></table>
+
+# APPENDIX G IMAGENET EXPERIMENTS
+
+We investigate unsupervised meta-learning in the context of a larger unsupervised meta-training dataset by using the ILSVRC 2012 dataset’s training split (Russakovsky et al., 2015), which is a superset of the miniImageNet dataset (including meta-validation and meta-testing data) consisting of 1000 classes and over 1,200,000 images. To facilitate comparison to the previous miniImageNet experiments, for meta-validation and meta-test we use the miniImageNet meta-validation and metatest splits. To avoid task leakage, we hold out all data from these 36 underlying classes from the rest of the data to construct the meta-training split.
+
+For CACTUs, we use the best-performing unsupervised learning method from the previous experiments, DeepCluster, to obtain the embeddings. Following Caron et al. (2018), we run DeepCluster using the VGG-16 architecture with a 256-dimensional feature space and 10,000 clusters on the meta-training data until the normalized mutual information between the data-cluster mappings of two consecutive epochs converges. To our knowledge, no prior works have yet been published on using MAML for ImageNet-sized meta-learning. We extend the standard convolutional neural network model class with residual connections (He et al., 2016), validate hyperparameters with supervised meta-learning, then use it for unsupervised meta-learning without further tuning. See Table 11 for MAML hyperparameters. The training from scratch, embedding $k _ { \mathrm { n n } }$ -nearest neighbors, and embedding linear classifier algorithms are the same as they were in the previous sets of experiments. For Oracle-MAML, we generated tasks using the ground-truth 964 ImageNet meta-training classes. We also run semi-supervised MAML, with the meta-training tasks consisting of CACTUs-based tasks as well as tasks constructed from the 64 miniImageNet meta-training classes. The unsupervised/supervised task proportion split was fixed according to the ratio of the number of data available to each task proposal method. As before, the meta-learning methods only meta-learned on 1-shot tasks.
+
+Table 11: MAML hyperparameter summary for ImageNet.   
+
+<table><tr><td>Hyperparameter</td><td>Value</td></tr><tr><td>Input size</td><td>224× 224</td></tr><tr><td>Outer (meta) learning rate</td><td>0.0001</td></tr><tr><td>Inner learning rate</td><td>0.001</td></tr><tr><td>Task batch size</td><td>3</td></tr><tr><td>Inner adaptation steps (meta-training)</td><td>5</td></tr><tr><td>Meta-training iterations</td><td>240,000</td></tr><tr><td>Adaptation steps (evaluation)</td><td>100</td></tr><tr><td>Classes per task (meta-training)</td><td>5</td></tr><tr><td>Shots per class (meta-training)</td><td>1</td></tr><tr><td>Queries per class</td><td>5</td></tr><tr><td>Residual blocks</td><td>5</td></tr><tr><td>Layers per residual block</td><td>2</td></tr></table>
+
+We find that the vastly increased amount of unlabeled meta-training data (in comparison to miniImageNet) results in significant increases for all methods over their counterparts in Table 9 (other than training from scratch, which does not use this data). We find that CACTUs-MAML slightly outperforms embedding linear classifier for the 1-shot test tasks, but that the linear classifier on top of the unsupervised embedding becomes better as the amount of test time supervision increases. Augmenting the unsupervised tasks with (a small number of) supervised tasks during meta-training results in slight improvement for the 1-shot test tasks. The lackluster performance of CACTUs-MAML is unsurprising insofar as meta-learning with large task spaces is still an open problem: higher shot Oracle-MAML only marginally stays ahead of the embedding linear classifier, which is not the case in the other, smaller-scale experiments. We expect that using a larger architecture in conjunction with MAML (such as Kim et al. (2018)) would result in increased performance for all methods based on MAML. Further, given the extensive degree to which unsupervised learning methods have been studied, we suspect that unsupervised task construction coupled with better meta-learning algorithms and architectures will result in improved performance on the entire unsupervised learning problem. We leave such investigation to future work.
+
+denotes a 95% confidence interval. d: dimensionality of   
+
+<table><tr><td>(os&#x27;s) （07&#x27;s) (s‘s) (1‘s) (ous xem) 1loirrra</td><td>% LL0干 89&#x27;6S% SL0干 L9&#x27;εS % 3901/30% 11:059:85 Ppt =pessstsesg Yrrlrriiir liiiiera reneseg</td><td>% 41:0 4244 % L70 干 0106 % 3705 75.77 % 040 18.61 % 130 11:11 % 990F 5878 % 17034355 % 00 1064 % 1/0494 % I9&#x27;0 干 LI&#x27;01 % 7703181 % 4101109 [[V[ = γ[=1 ‘(SJ1) TII-SALO0 ohhgrer serrerarg nrippegg garreerrnrilpestg</td><td>% 190  L728 % 690£8&#x27;78 %0/0£494 % SL&#x27;0 干 SL&#x27;19 Sururersritu prseetdeesntts Jra psrsN-</td><td>% 1/0 £7&#x27;98 % L/071:4 Srinerniar peseadn DEDLLrr Tirr-larrt</td><td>% 990 干 90℃6% 69&#x27;0干4116</td></tr></table>
\ No newline at end of file
diff --git a/parse/train/r1My6sR9tX/r1My6sR9tX_content_list.json b/parse/train/r1My6sR9tX/r1My6sR9tX_content_list.json
new file mode 100644
index 0000000000000000000000000000000000000000..0d40910e65338be53470ac8f7c6059cee0c67182
--- /dev/null
+++ b/parse/train/r1My6sR9tX/r1My6sR9tX_content_list.json
@@ -0,0 +1,2115 @@
+[
+    {
+        "type": "text",
+        "text": "UNSUPERVISED LEARNING VIA META-LEARNING ",
+        "text_level": 1,
+        "bbox": [
+            173,
+            98,
+            771,
+            121
+        ],
+        "page_idx": 0
+    },
+    {
+        "type": "text",
+        "text": "Kyle Hsu†   \nUniversity of Toronto   \nkyle.hsu@mail.utoronto.ca ",
+        "bbox": [
+            184,
+            143,
+            429,
+            186
+        ],
+        "page_idx": 0
+    },
+    {
+        "type": "text",
+        "text": "Sergey Levine, Chelsea Finn University of California, Berkeley {svlevine,cbfinn}@eecs.berkeley.edu ",
+        "bbox": [
+            459,
+            145,
+            800,
+            186
+        ],
+        "page_idx": 0
+    },
+    {
+        "type": "text",
+        "text": "ABSTRACT ",
+        "text_level": 1,
+        "bbox": [
+            454,
+            223,
+            544,
+            238
+        ],
+        "page_idx": 0
+    },
+    {
+        "type": "text",
+        "text": "A central goal of unsupervised learning is to acquire representations from unlabeled data or experience that can be used for more effective learning of downstream tasks from modest amounts of labeled data. Many prior unsupervised learning works aim to do so by developing proxy objectives based on reconstruction, disentanglement, prediction, and other metrics. Instead, we develop an unsupervised meta-learning method that explicitly optimizes for the ability to learn a variety of tasks from small amounts of data. To do so, we construct tasks from unlabeled data in an automatic way and run meta-learning over the constructed tasks. Surprisingly, we find that, when integrated with meta-learning, relatively simple task construction mechanisms, such as clustering embeddings, lead to good performance on a variety of downstream, human-specified tasks. Our experiments across four image datasets indicate that our unsupervised meta-learning approach acquires a learning algorithm without any labeled data that is applicable to a wide range of downstream classification tasks, improving upon the embedding learned by four prior unsupervised learning methods. ",
+        "bbox": [
+            233,
+            255,
+            764,
+            462
+        ],
+        "page_idx": 0
+    },
+    {
+        "type": "text",
+        "text": "1 INTRODUCTION ",
+        "text_level": 1,
+        "bbox": [
+            176,
+            489,
+            336,
+            506
+        ],
+        "page_idx": 0
+    },
+    {
+        "type": "text",
+        "text": "Unsupervised learning is a fundamental, unsolved problem (Hastie et al., 2009) and has seen promising results in domains such as image recognition (Le et al., 2013) and natural language understanding (Ramachandran et al., 2017). A central use case of unsupervised learning methods is enabling better or more efficient learning of downstream tasks by training on top of unsupervised representations (Reed et al., 2014; Cheung et al., 2015; Chen et al., 2016) or fine-tuning a learned model (Erhan et al., 2010). However, since the downstream objective requires access to supervision, the objectives used for unsupervised learning are only a rough proxy for downstream performance. If a central goal of unsupervised learning is to learn useful representations, can we derive an unsupervised learning objective that explicitly takes into account how the representation will be used? ",
+        "bbox": [
+            173,
+            517,
+            825,
+            642
+        ],
+        "page_idx": 0
+    },
+    {
+        "type": "text",
+        "text": "The use of unsupervised representations for downstream tasks is closely related to the objective of meta-learning techniques: finding a learning procedure that is more efficient and effective than learning from scratch. However, unlike unsupervised learning methods, meta-learning methods require large, labeled datasets and hand-specified task distributions. These dependencies are major obstacles to widespread use of these methods for few-shot classification. ",
+        "bbox": [
+            174,
+            650,
+            825,
+            718
+        ],
+        "page_idx": 0
+    },
+    {
+        "type": "text",
+        "text": "To begin addressing these problems, we propose an unsupervised meta-learning method: one which aims to learn a learning procedure, without supervision, that is useful for solving a wide range of new, human-specified tasks. With only raw, unlabeled observations, our model’s goal is to learn a useful prior such that, after meta-training, when presented with a modestly-sized dataset for a human-specified task, the model can transfer its prior experience to efficiently learn to perform the new task. If we can build such an algorithm, we can enable few-shot learning of new tasks without needing any labeled data nor any pre-defined tasks. ",
+        "bbox": [
+            174,
+            727,
+            825,
+            823
+        ],
+        "page_idx": 0
+    },
+    {
+        "type": "text",
+        "text": "To perform unsupervised meta-learning, we need to automatically construct tasks from unlabeled data. We study several options for how this can be done. We find that a good task distribution should be diverse, but also not too difficult: na¨ıve random approaches for task generation produce tasks that contain insufficient regularity to enable useful meta-learning. To that end, our method proposes tasks by first leveraging prior unsupervised learning algorithms to learn an embedding of the input data, and then performing an overcomplete partitioning of the dataset to construct numerous categorizations of the data. We show how we can derive classification tasks from these categorizations for use with meta-learning algorithms. Surprisingly, even with simple mechanisms for partitioning the embedding space, such as $k$ -means clustering, we find that meta-learning acquires priors that, when used to learn new, human-designed tasks, learn those tasks more effectively than methods that directly learn on the embedding. That is, the learning algorithm acquired through unsupervised meta-learning achieves better downstream performance than the original representation used to derive meta-training tasks, without introducing any additional assumptions or supervision. See Figure 1 for an illustration of the complete approach. ",
+        "bbox": [
+            174,
+            830,
+            823,
+            901
+        ],
+        "page_idx": 0
+    },
+    {
+        "type": "text",
+        "text": "",
+        "bbox": [
+            174,
+            103,
+            823,
+            229
+        ],
+        "page_idx": 1
+    },
+    {
+        "type": "text",
+        "text": "The core idea in this paper is that we can leverage unsupervised embeddings to propose tasks for a meta-learning algorithm, leading to an unsupervised meta-learning algorithm that is particularly effective as pre-training for human-specified downstream tasks. In the following sections, we formalize our problem assumptions and goal, which match those of unsupervised learning, and discuss several options for automatically deriving tasks from embeddings. We instantiate our method with two meta-learning algorithms and compare to prior state-of-the-art unsupervised learning methods. Across four image datasets (MNIST, Omniglot, miniImageNet, and CelebA), we find that our method consistently leads to effective downstream learning of a variety of human-specified tasks, including character recognition tasks, object classification tasks, and facial attribute discrimination tasks, without requiring any labels or hand-designed tasks during meta-learning and where key hyperparameters of our method are held constant across all domains. We show that, even though our unsupervised meta-learning algorithm trains for one-shot generalization, one instantiation of our approach performs well not only on few-shot learning, but also when learning downstream tasks with up to 50 training examples per class. In fact, some of our results begin to approach the performance of fully-supervised meta-learning techniques trained with fully-specified task distributions. ",
+        "bbox": [
+            173,
+            236,
+            825,
+            444
+        ],
+        "page_idx": 1
+    },
+    {
+        "type": "image",
+        "img_path": "images/2aa42c4a43d437245b1ba217e8b16d1cdbed8dd1add1db4a9609cd8a7a78f543.jpg",
+        "image_caption": [
+            "Figure 1: Illustration of the proposed unsupervised meta-learning procedure. Embeddings of raw observations are clustered with $k$ -means to construct partitions, which give rise to classification tasks. Each task involves distinguishing between examples from $N = 2$ clusters, with $K _ { \\mathrm { m - t r } } = 1$ example from each cluster being a training input. The meta-learner’s aim is to produce a learning procedure that successfully solves these tasks. "
+        ],
+        "image_footnote": [],
+        "bbox": [
+            192,
+            462,
+            797,
+            592
+        ],
+        "page_idx": 1
+    },
+    {
+        "type": "text",
+        "text": "2 UNSUPERVISED META-LEARNING ",
+        "text_level": 1,
+        "bbox": [
+            174,
+            684,
+            488,
+            699
+        ],
+        "page_idx": 1
+    },
+    {
+        "type": "text",
+        "text": "In this section, we describe our problem setting in relation to that of unsupervised and semisupervised learning, provide necessary preliminaries, and present our approach. ",
+        "bbox": [
+            169,
+            714,
+            821,
+            742
+        ],
+        "page_idx": 1
+    },
+    {
+        "type": "text",
+        "text": "2.1 PROBLEM STATEMENT ",
+        "text_level": 1,
+        "bbox": [
+            176,
+            761,
+            370,
+            773
+        ],
+        "page_idx": 1
+    },
+    {
+        "type": "text",
+        "text": "Our goal is to leverage unlabeled data for the efficient learning of a range of human-specified downstream tasks. We only assume access to an unlabeled dataset $\\mathcal { D } = \\{ \\mathbf { x } _ { i } \\}$ during meta-training. After learning from the unlabeled data, which we will refer to as unsupervised meta-training, we want to apply what was learned towards learning a variety of downstream, human-specified tasks from a modest amount of labeled data, potentially as few as a single example per class. These downstream tasks may, in general, have different underlying classes or attributes (in contrast to typical semi-supervised problem assumptions), but are assumed to have inputs from the same distribution as the one from which datapoints in $\\mathcal { D }$ are drawn. Concretely, we assume that downstream tasks are $M$ -way classification tasks, and that the goal is to learn an accurate classifier using $K$ labeled datapoints $\\left( \\mathbf { x } _ { k } , \\mathbf { y } _ { k } \\right)$ from each of the $M$ classes, where $K$ is relatively small (i.e. between 1 and 50). ",
+        "bbox": [
+            174,
+            785,
+            825,
+            924
+        ],
+        "page_idx": 1
+    },
+    {
+        "type": "text",
+        "text": "The unsupervised meta-training phase aligns with the unsupervised learning problem in that it involves no access to information about the downstream tasks, other than the fact that they are $M$ -way classification tasks, for variable $M$ upper-bounded by $N$ . The upper bound $N$ is assumed to be known during unsupervised meta-training, but otherwise, the values of $M$ and $K$ are not known $a$ priori. As a result, the unsupervised meta-training phase needs to acquire a sufficiently general prior for applicability to a range of classification tasks with variable quantities of data and classes. This problem definition is our prototype for a practical use-case in which a user would like to train an application-specific image classifier, but does not have an abundance of labeled data. ",
+        "bbox": [
+            174,
+            103,
+            825,
+            215
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "text",
+        "text": "2.2 PRELIMINARIES ",
+        "text_level": 1,
+        "bbox": [
+            174,
+            232,
+            326,
+            246
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "text",
+        "text": "Unsupervised embedding learning. An unsupervised embedding learning algorithm $\\mathcal { E }$ is a procedure that takes as input an unlabeled dataset $\\mathcal { D } = \\left\\{ \\mathbf { x } _ { i } \\right\\}$ and outputs a mapping from $\\{ { \\bf { x } } _ { i } \\}$ to embeddings $\\left\\{ \\mathbf { z } _ { i } \\right\\}$ . These embedded points are typically lower-dimensional and arranged such that distances correspond to meaningful differences between inputs, in contrast to distances between the original inputs, such as image pixels, which are not meaningful measures of image similarity. ",
+        "bbox": [
+            174,
+            256,
+            823,
+            325
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "text",
+        "text": "Task. An $M$ -way $K$ -shot classification task $\\tau$ consists of $K$ training datapoints and labels $\\{ ( \\mathbf { x } _ { k } , \\ell _ { k } ) \\}$ per class, which are used for learning a classifier, and $Q$ query datapoints and labels per class, on which the learned classifier is evaluated. That is, in a task there are $K + Q = R$ datapoints and labels for each of the $M$ classes. ",
+        "bbox": [
+            174,
+            332,
+            825,
+            388
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "text",
+        "text": "Meta-learning. A supervised meta-learning algorithm $\\mathcal { M } ( \\cdot )$ takes as input a set of supervised metatraining tasks $\\{ \\mathcal { T } _ { t } \\}$ . It produces a learning procedure $\\mathcal F ( \\cdot )$ , which, in turn, ingests the supervised training data of a task to produce a classifier $f ( \\cdot )$ . The goal of $\\mathcal { M }$ is to learn $\\mathcal { F }$ such that, when faced with a meta-test time task $\\mathcal { T } _ { t ^ { \\prime } }$ held-out from $\\{ \\mathcal { T } _ { t } \\}$ , $\\mathcal { F }$ can learn a $f _ { t ^ { \\prime } }$ that accomplishes $\\mathcal { T } _ { t ^ { \\prime } }$ . At a high level, the quintessential meta-learning strategy is to have $\\mathcal { M }$ iterate over $\\{ \\mathcal { T } _ { t } \\}$ , cycling between applying the current form of $\\mathcal { F } _ { t }$ on training data from $\\mathcal { T } _ { t }$ to learn $f _ { t }$ , assessing its performance by calculating some meta-loss $\\mathcal { L }$ on held-out data from the task, and optimizing $\\mathcal { L }$ to improve the learning procedure. ",
+        "bbox": [
+            174,
+            395,
+            825,
+            507
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "text",
+        "text": "We build upon two meta-learning algorithms: model agnostic meta-learning (MAML) (Finn et al., 2017) and prototypical networks (ProtoNets) (Snell et al., 2017). MAML aims to learn the initial parameters of a deep network such that one or a few gradient steps leads to effective generalization; it specifies $\\mathcal { F }$ as gradient descent starting from the meta-learned parameters. ProtoNets aim to metalearn a representation in which a class is effectively identified by its prototype, defined to be the mean of the class’ training examples in the meta-learned space; $\\mathcal { F }$ is the computation of these class prototypes, and $f$ is a linear classifier that predicts the class whose prototype is closest in Euclidean distance to the query’s representation. ",
+        "bbox": [
+            174,
+            513,
+            825,
+            626
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "text",
+        "text": "Task generation for meta-learning. We briefly summarize how tasks are typically generated from labeled datasets $\\left\\{ \\left( \\mathbf { x } _ { i } , \\mathbf { y } _ { i } \\right) \\right\\}$ for supervised meta-learning, as introduced by Santoro et al. (2016). For simplicity, consider the case where the labels are discrete scalar values $y _ { i }$ . To construct an $N$ -way classification task $\\tau$ (assuming $N$ is not greater than the number of unique $y _ { i }$ ), we can sample $N$ classes, sample $R$ datapoints $\\{ { \\bf { x } } _ { r } \\} _ { n }$ for each of the $N$ classes, and sample a permutation of $N$ distinct one-hot vectors $( \\ell _ { n } )$ to serve as task-specific labels of the $N$ sampled classes. The task is then defined as $\\mathcal { T } = \\{ ( \\mathbf { x } _ { n , r } , \\ell _ { n } ) ~ | ~ \\mathbf { x } _ { n , r } \\in \\{ \\mathbf { x } _ { r } \\} _ { n } \\}$ . Of course, this procedure is only possible with labeled data; in the next section, we discuss how we can construct tasks without ground-truth labels. ",
+        "bbox": [
+            174,
+            632,
+            825,
+            743
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "text",
+        "text": "2.3 UNSUPERVISED META-LEARNING WITH AUTOMATICALLY CONSTRUCTED TASKS",
+        "text_level": 1,
+        "bbox": [
+            176,
+            775,
+            776,
+            789
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "text",
+        "text": "We approach our problem from a meta-learning perspective, framing the problem as the acquisition, from unlabeled data, of an efficient learning procedure that is transferable to human-designed tasks. In particular, we aim to construct classification tasks from the unlabeled data and then learn how to efficiently learn these tasks. If such tasks are adequately diverse and structured, then metalearning these tasks should enable fast learning of new, human-provided tasks. A key question, then, is how to automatically construct such tasks from unlabeled data $\\mathcal { D } = \\left\\{ \\mathbf { x } _ { i } \\right\\}$ . Notice that in the supervised meta-learning task generation procedure detailed in Section 2.2, the labels $y _ { i }$ induce a partition $\\mathcal { P } = \\{ \\mathcal { C } _ { c } \\}$ over $\\{ { \\bf { x } } _ { i } \\}$ by assigning all datapoints with label $y _ { c }$ to subset $\\mathcal { C } _ { c }$ . Once a partition is obtained, task generation is simple; we can reduce the problem of constructing tasks to that of constructing a partition over $\\{ { \\bf { x } } _ { i } \\}$ . All that’s left is to find a principled alternative to human labels for defining the partition. ",
+        "bbox": [
+            174,
+            797,
+            825,
+            924
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "text",
+        "text": "",
+        "bbox": [
+            173,
+            103,
+            823,
+            132
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "A na¨ıve approach is to randomly partition the data $\\mathcal { D }$ . While such a scheme introduces diverse tasks, there is no structure; that is, there is no consistency between a task’s training data and query data, and hence nothing to be learned during each task, let alone across tasks. As seen in Table 3, providing a meta-learner with purely random tasks results in failed meta-learning. ",
+        "bbox": [
+            174,
+            138,
+            823,
+            194
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "To construct tasks with structure that resembles that of human-specified labels, we need to group datapoints into consistent and distinct subsets based on salient features. With this motivation in mind, we propose to use $k$ -means clustering. Consider the partition $\\mathcal { P } = \\{ \\mathcal { C } _ { c } \\}$ learned by $k$ -means as a simplification of a Gaussian mixture model $p ( \\mathbf { x } | c ) p ( \\bar { c } )$ . If the clusters can recover a semblance of the true class-conditional generative distributions $p ( \\mathbf { x } | c )$ , creating tasks based on treating these clusters as classes should result in useful unsupervised meta-training. However, the result of $k$ -means is critically dependent on the metric space on which its objective is defined. Clustering in pixel-space is unappealing for two reasons: (1) distance in pixel-space correlates poorly with semantic meaning, and (2) the high dimensionality of raw images renders clustering difficult in practice. We empirically show in Table 3 that meta-learning with tasks defined by pixel-space clusters, with preprocessing as directed by Coates & $\\mathrm { N g }$ (2012), also fails. ",
+        "bbox": [
+            174,
+            202,
+            825,
+            354
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "We are now motivated to cluster in spaces in which common distance functions correlate to semantic meaning. However, we must satisfy the constraints of our problem statement in the process of learning such spaces. To these ends, we use state-of-the-art unsupervised learning methods to produce useful embedding spaces. For qualitative evidence in the unsupervised learning literature that such embedding spaces exhibit semantic meaning, see Cheung et al. (2015); Bojanowski & Joulin (2017); Donahue et al. (2017). We note that while a given embedding space may not be directly suitable for highly-efficient learning of new tasks (which would require the embedding space to be precisely aligned or adaptable to the classes of those tasks), we can still leverage it for the construction of structured tasks, a process for which requirements are less strict. ",
+        "bbox": [
+            174,
+            361,
+            825,
+            487
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "Thus, we first run an out-of-the-box unsupervised embedding learning algorithm $\\mathcal { E }$ on $\\mathcal { D }$ , then map the data $\\{ { \\bf { x } } _ { i } \\}$ into the embedding space $\\mathcal { Z }$ , producing $\\left\\{ \\mathbf { z } _ { i } \\right\\}$ . To produce a diverse task set, we generate $P$ partitions $\\{ \\mathcal P _ { p } \\}$ by running clustering $P$ times, applying random scaling to the dimensions of $\\mathcal { Z }$ to induce a different metric, represented by diagonal matrix A, for each run of clustering. With $\\pmb { \\mu } _ { c }$ denoting the learned centroid of cluster $\\mathcal { C } _ { c }$ , a single run of clustering can be summarized with ",
+        "bbox": [
+            174,
+            493,
+            825,
+            564
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "equation",
+        "img_path": "images/6cb4a83a12fefaed35c95d6609ab822a3a5d0ea1fb314437bd6dd6eda06f39ea.jpg",
+        "text": "$$\n\\mathcal { P } , \\{ \\pmb { \\mu } _ { c } \\} = \\underset { \\{ \\mathcal { C } _ { c } \\} , \\{ \\pmb { \\mu } _ { c } \\} } { \\arg \\operatorname* { m i n } } \\sum _ { c = 1 } ^ { k } \\sum _ { \\mathbf { z } \\in \\mathcal { C } _ { c } } \\| \\mathbf { z } - \\pmb { \\mu } _ { c } \\| _ { \\mathbf { A } } ^ { 2 }\n$$",
+        "text_format": "latex",
+        "bbox": [
+            362,
+            571,
+            633,
+            617
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "We derive tasks for meta-learning from the partitions using the procedure detailed in Section 2.2, except we begin the construction of each task by sampling a partition from the uniform distribution $\\mathcal { U } ( \\mathcal { P } )$ , and for ${ \\bf x } _ { i } \\in \\mathcal { C } _ { c }$ , specify $y _ { i } = c$ . To avoid imbalanced clusters dominating the meta-training tasks, we opt not to sample from $p ( c ) \\propto | \\mathcal { C } _ { c } |$ , but instead sample $N$ clusters uniformly without replacement for each task. We note that Caron et al. (2018) are similarly motivated in their design decision of sampling data from a uniform distribution over clusters. ",
+        "bbox": [
+            173,
+            625,
+            825,
+            709
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "With the partitions being constructed over $\\left\\{ \\mathbf { z } _ { i } \\right\\}$ , we have one more design decision to make: should we perform meta-learning on embeddings or images? We consider that, to successfully solve new tasks at meta-test time, a learning procedure $\\mathcal { F }$ that takes embeddings as input would depend on the embedding function’s ability to generalize to out-of-distribution observations. On the other hand, by meta-learning on images, $\\mathcal { F }$ can separately adapt $f$ to each evaluation task from the rawest level of representation. Thus, we choose to meta-learn on images. ",
+        "bbox": [
+            174,
+            715,
+            825,
+            800
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "We call our method clustering to automatically construct tasks for unsupervised meta-learning (CACTUs). We detail the task construction algorithm in Algorithm 1, and provide an illustration of the complete unsupervised meta-learning approach for classification in Figure 1. ",
+        "bbox": [
+            176,
+            806,
+            825,
+            849
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "3 RELATED WORK ",
+        "text_level": 1,
+        "bbox": [
+            176,
+            867,
+            344,
+            883
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "The method we propose aims to address the unsupervised learning problem (Hastie et al., 2009; Le et al., 2013), namely acquiring a transferable learning procedure without labels. We show that our ",
+        "bbox": [
+            173,
+            895,
+            823,
+            924
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "Algorithm 1 CACTUs for classification ",
+        "text_level": 1,
+        "bbox": [
+            176,
+            103,
+            437,
+            118
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "table",
+        "img_path": "images/3c47f37c28699f9f6647b45003f86db9df2ebf5ba64e6abf56db7f4740de6eb0.jpg",
+        "table_caption": [],
+        "table_footnote": [],
+        "table_body": "<table><tr><td colspan=\"2\">1: procedure CACTUs(ε,D,P,k,T,N,Km-tr, Q)</td><td colspan=\"4\"></td></tr><tr><td>2:</td><td colspan=\"4\">Run embedding learning algorithm ε on D and produce embeddings {zi} from observations {xi}.</td></tr><tr><td>3:</td><td colspan=\"4\">Run k-means on {zi} P times (with random scaling) to generate a set of partitions {Pp = {Cc}p}.</td></tr><tr><td>4:</td><td colspan=\"4\">for t from 1 to the number of desired tasks T do</td></tr><tr><td>5:</td><td colspan=\"4\">Sample a partition P uniformly at random from the set of partitions {Pp}.</td></tr><tr><td>6:</td><td colspan=\"4\">Sample a cluster Cn uniformly without replacement from P for each of the N classes desired for a task.</td></tr><tr><td>7:</td><td colspan=\"4\">Sample an embedding Zr without replacement from Cn for each of the R= Km-tr+Q training and</td></tr><tr><td>8:</td><td colspan=\"4\">query examples desired for each class, and record the corresponding datapoint Xn,r. Sample a permutation(ln) of N one-hot labels.</td></tr><tr><td>9:</td><td colspan=\"4\">Construct Tt ={(xn,r,ln)}.</td></tr><tr><td>10:</td><td colspan=\"4\">return {Tt}</td></tr></table>",
+        "bbox": [
+            176,
+            121,
+            823,
+            284
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "text",
+        "text": "method is complementary to a number of unsupervised learning methods, including ACAI (Berthelot et al., 2018), BiGAN (Donahue et al., 2017; Dumoulin et al., 2017), DeepCluster (Caron et al., 2018), and InfoGAN (Chen et al., 2016): we leverage these prior methods to learn embeddings used for constructing meta-learning tasks, and demonstrate that our method learns a more useful representation than the embeddings. The ability to use what was learned during unsupervised pretraining to better or more efficiently learn a variety of downstream tasks is arguably one of the most practical applications of unsupervised learning methods, which has a long history in neural network training (Hinton et al., 2006; Bengio et al., 2007; Ranzato et al., 2006; Vincent et al., 2008; Erhan et al., 2010). Unsupervised pre-training has demonstrated success in a number of domains, including speech recognition (Yu et al., 2010), image classification (Zhang et al., 2017), machine translation (Ramachandran et al., 2017), and text classification (Dai & Le, 2015; Howard & Ruder, 2018; Radford et al., 2018). Our approach, unsupervised meta-learning, can be viewed as an unsupervised learning algorithm that explicitly optimizes for few-shot transferability. As a result, we can expect it to better learn human-specified downstream tasks, compared to unsupervised learning methods that optimize for other metrics, such as reconstruction (Vincent et al., 2010; Higgins et al., 2017), fidelity of constructed images (Radford et al., 2016; Salimans et al., 2016; Donahue et al., 2017; Dumoulin et al., 2017), representation interpolation (Berthelot et al., 2018), disentanglement (Bengio et al., 2013; Reed et al., 2014; Cheung et al., 2015; Chen et al., 2016; Mathieu et al., 2016; Denton & Birodkar, 2017), and clustering (Coates & Ng, 2012; Krahenb ¨ uhl et al., 2016; Bojanowski & Joulin, ¨ 2017; Caron et al., 2018). We empirically evaluate this hypothesis in the next section. In contrast to many previous evaluations of unsupervised pre-training, we focus on settings in which only a small amount of data for the downstream tasks is available, since this is where the unlabeled data can be maximally useful. ",
+        "bbox": [
+            173,
+            328,
+            825,
+            646
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "text",
+        "text": "Unsupervised pre-training followed by supervised learning can be viewed as a special case of the semi-supervised learning problem (Zhu, 2011; Kingma et al., 2014; Rasmus et al., 2015; Oliver et al., 2018). However, in contrast to our problem statement, semi-supervised learning methods assume that a significant proportion of the unlabeled data, if not all of it, shares underlying labels with the labeled data. Additionally, our approach and other unsupervised learning methods are wellsuited for transferring their learned representation to many possible downstream tasks or labelings, whereas semi-supervised learning methods typically optimize for performance on a single task, with respect to a single labeling of the data. ",
+        "bbox": [
+            173,
+            654,
+            825,
+            765
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "text",
+        "text": "Our method builds upon the ideas of meta-learning (Schmidhuber, 1987; Bengio et al., 1991; Naik & Mammone, 1992) and few-shot learning (Santoro et al., 2016; Vinyals et al., 2016; Ravi & Larochelle, 2017; Munkhdalai & Yu, 2017; Snell et al., 2017). We apply two meta-learning algorithms, model-agnostic meta-learning (Finn et al., 2017) and prototypical networks (Snell et al., 2017), to tasks constructed in an unsupervised manner. Similar to our problem setting, some prior works have aimed to learn an unsupervised learning procedure with supervised data (Garg & Kalai, 2017; Metz et al., 2018). Instead, we consider a problem setting that is entirely unsupervised, aiming to learn efficient learning algorithms using unlabeled datasets. Our problem setting is similar to that considered by Gupta et al. (2018), but we develop an approach that is suitable for supervised downstream tasks, rather than reinforcement learning problems, and demonstrate our algorithm on problems with high-dimensional visual observations. ",
+        "bbox": [
+            173,
+            771,
+            825,
+            924
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "text",
+        "text": "4 EXPERIMENTS ",
+        "text_level": 1,
+        "bbox": [
+            176,
+            102,
+            326,
+            117
+        ],
+        "page_idx": 5
+    },
+    {
+        "type": "text",
+        "text": "We begin the experimental section by presenting our research questions and how our experiments are designed to address them. Links to code for the experiments can be found at https://sites. google.com/view/unsupervised-via-meta. ",
+        "bbox": [
+            174,
+            131,
+            823,
+            172
+        ],
+        "page_idx": 5
+    },
+    {
+        "type": "text",
+        "text": "Benefit of meta-learning. Is there any significant benefit to doing meta-learning on tasks derived from embeddings, or is the embedding function already sufficient for downstream supervised learning of new tasks? To investigate this, we run MAML and ProtoNets on tasks generated via CACTUs (CACTUs-MAML, CACTUs-ProtoNets). We compare to five alternate algorithms, with four being supervised learning methods on top of the embedding function. i) Embedding $k _ { \\mathrm { n n } }$ -nearest neighbors first infers the embeddings of the downstream task images. For a query test image, it predicts the plurality vote of the labels of the $k _ { \\mathrm { n n } }$ training images that are closest in the embedding space to the query’s embedding. ii) Embedding linear classifier also begins by inferring the embeddings of the downstream task images. It then fits a linear classifier using the $N K$ training embeddings and labels, and predicts labels for the query embeddings using the classifier. iii) Embedding multilayer perceptron instead uses a network with one hidden layer of 128 units and tuned dropout (Srivastava et al., 2014). iv) To isolate the effect of meta-learning on images, we also compare to embedding cluster matching, i.e. directly using the meta-training clusters for classification by labeling clusters with a task’s training data via plurality vote. If a query datapoint maps to an unlabeled cluster, the closest labeled cluster is used. v) As a baseline, we forgo any unsupervised pre-training and train a model with the MAML architecture from standard random network initialization via gradient descent separately for each evaluation task. ",
+        "bbox": [
+            174,
+            181,
+            825,
+            416
+        ],
+        "page_idx": 5
+    },
+    {
+        "type": "text",
+        "text": "Different embedding spaces. Does CACTUs result in successful meta-learning for many distinct task-generating embeddings? To investigate this, we run unsupervised meta-learning using four embedding learning algorithms: ACAI (Berthelot et al., 2018), BiGAN (Donahue et al., 2017), DeepCluster (Caron et al., 2018), and InfoGAN (Chen et al., 2016). These four approaches collectively cover the following range of objectives and frameworks in the unsupervised learning literature: generative modeling, two-player games, reconstruction, representation interpolation, discriminative clustering, and information maximization. We describe these methods in more detail in Appendix A. ",
+        "bbox": [
+            173,
+            422,
+            825,
+            521
+        ],
+        "page_idx": 5
+    },
+    {
+        "type": "text",
+        "text": "Applicability to different tasks. Can unsupervised meta-learning yield a good prior for a variety of task types? In other words, can unsupervised meta-learning yield a good representation for tasks that assess the ability to distinguish between features on different scales, or tasks with various amounts of supervision signal? To investigate this, we evaluate our procedure on tasks assessing recognition of character identity, object identity, and facial attributes. For this purpose we choose to use the existing Omniglot (Santoro et al., 2016) and miniImageNet (Ravi & Larochelle, 2017) datasets and few-shot classification tasks and, inspired by Finn et al. (2018), also construct a new few-shot classification benchmark based on the CelebA dataset and its binary attribute annotations. For miniImageNet, we consider both few-shot downstream tasks and tasks involving larger datasets (up to 50-shot). Specifics on the datasets and human-designed tasks are presented in Appendix B. ",
+        "bbox": [
+            173,
+            527,
+            825,
+            666
+        ],
+        "page_idx": 5
+    },
+    {
+        "type": "text",
+        "text": "Oracle. How does the performance of our unsupervised meta-learning method compare to supervised meta-learning with a human-specified, near-optimal task distribution derived from a labeled dataset? To investigate this, we use labeled versions of the meta-training datasets to run MAML and ProtoNets as supervised meta-learning algorithms (Oracle-MAML, Oracle-ProtoNets). To facilitate fair comparison with the unsupervised variants, we control for the relevant hyperparameters. ",
+        "bbox": [
+            174,
+            674,
+            823,
+            743
+        ],
+        "page_idx": 5
+    },
+    {
+        "type": "text",
+        "text": "Task construction ablation. How do the alternatives for constructing tasks from the embeddings compare? To investigate this, we run MAML on tasks constructed via clustering (CACTUs-MAML) and MAML on tasks constructed via random hyperplane slices of the embedding space with varying margin (Hyperplanes-MAML). The latter partitioning procedure is detailed in Appendix C. For the experiments where tasks are constructed via clustering, we also investigate the effect of sampling based on a single partition versus multiple partitions. We additionally experiment with tasks based on random assignments of images to “clusters” (Random-MAML) and tasks based on pixel-space clusters (Pixels CACTUs-MAML) with the Omniglot dataset. ",
+        "bbox": [
+            173,
+            750,
+            825,
+            862
+        ],
+        "page_idx": 5
+    },
+    {
+        "type": "text",
+        "text": "To investigate the limitations of our method, we also consider an easier version of our problem statement where the data distributions at meta-training and meta-test time perfectly overlap, i.e. the images share a common set of underlying labels (Appendix D). Finally, we present results on miniImageNet after unsupervised meta-learning on most of ILSVRC 2012 (Appendix G). ",
+        "bbox": [
+            174,
+            869,
+            823,
+            924
+        ],
+        "page_idx": 5
+    },
+    {
+        "type": "image",
+        "img_path": "images/da015e73e499847b381f6ba35cc8ae17e2cc6b8b08026a9f2a626ccbe7e0ef3c.jpg",
+        "image_caption": [
+            "Figure 2: Examples of three DeepCluster-embedding cluster-based classes (a) and a 2-way 5-shot test task (b) for two datasets. (a) Some of the clusters correspond well to unseen labels (top left, bottom left). Others exhibit semantic meaning despite members not being grouped as such in the labeled version of the dataset (top middle: pair of objects, bottom middle: white hat). Still others are uninterpretable (top right) or are based on image artifacts (bottom right). (b) We evaluate unsupervised pre-training based on the ability to learn downstream, human-designed tasks with held-out images and underlying classes. "
+        ],
+        "image_footnote": [],
+        "bbox": [
+            176,
+            94,
+            820,
+            284
+        ],
+        "page_idx": 6
+    },
+    {
+        "type": "text",
+        "text": "4.1 EXPERIMENTAL PROTOCOL SUMMARY ",
+        "text_level": 1,
+        "bbox": [
+            176,
+            393,
+            482,
+            407
+        ],
+        "page_idx": 6
+    },
+    {
+        "type": "text",
+        "text": "As discussed by Oliver et al. (2018), keeping proper experimental protocol is particularly important when evaluating unsupervised and semi-supervised learning algorithms. Our foremost concern is to avoid falsely embellishing the capabilities of our approach by overfitting to the specific datasets and task types that we consider. To this end, we adhere to two key principles. We do not perform any architecture engineering: we use architectures from prior work as-is, or lightly adapt them to our needs if necessary. We also keep hyperparameters related to the unsupervised meta-learning stage as constant as possible across all experiments, including the MAML and ProtoNets model architectures. Details on hyperparameters and architectures are presented in Appendix E. We assume knowledge of an upper bound on the number of classes $N$ present in each downstream meta-testing task for each dataset. However, regardless of the number of shots $K$ , we do not assume knowledge of $K$ during unsupervised meta-learning. We use $N$ -way 1-shot tasks during meta-training, but test on larger values of $K$ during meta-testing. ",
+        "bbox": [
+            174,
+            420,
+            825,
+            587
+        ],
+        "page_idx": 6
+    },
+    {
+        "type": "text",
+        "text": "We partition each dataset into meta-training, meta-validation, and meta-testing splits. For Omniglot and miniImageNet, these splits contain disjoint sets of classes. For all algorithms, we run unsupervised pre-training on the unlabeled meta-training split and report performance on downstream tasks dictated by the labeled data of the meta-testing split, generated using the procedure from prior work recounted in Section 2.2. For the supervised meta-learning oracles, meta-training tasks are constructed in the same manner but from the dataset’s meta-training split. See Figure 2 for illustrative examples of embedding-derived clusters and human-designed test tasks. ",
+        "bbox": [
+            174,
+            594,
+            825,
+            691
+        ],
+        "page_idx": 6
+    },
+    {
+        "type": "text",
+        "text": "To facilitate analysis on meta-overfitting, we use the labels of the meta-validation split (instead of clustering embeddings) to construct tasks for meta-validation. However, because our aim is to perform meta-learning without supervision, we do not tune hyperparameters on this labeled data. We use a fixed number of meta-training iterations, since there is no suitable criterion for early stopping. ",
+        "bbox": [
+            174,
+            698,
+            823,
+            755
+        ],
+        "page_idx": 6
+    },
+    {
+        "type": "text",
+        "text": "When we experiment with the embedding-plus-supervised-learning methods used as fair comparisons to unsupervised meta-learning, we err on the side of providing more supervision and data than technically allowed. Specifically, we separately tune the supervised learning hyperparameters for each dataset and each task difficulty on the labeled version of the meta-validation split. With DeepCluster embeddings, we also use the entire meta-testing split’s statistics to perform dimensionality reduction (via PCA) and whitening, which is unfair as this shares information across tasks. ",
+        "bbox": [
+            174,
+            761,
+            823,
+            844
+        ],
+        "page_idx": 6
+    },
+    {
+        "type": "text",
+        "text": "4.2 RESULTS ",
+        "text_level": 1,
+        "bbox": [
+            176,
+            869,
+            277,
+            882
+        ],
+        "page_idx": 6
+    },
+    {
+        "type": "text",
+        "text": "Our primary results are summarized in Tables 1 and 2. Task construction ablations are summarized in Tables 3 and 4. ",
+        "bbox": [
+            176,
+            895,
+            821,
+            922
+        ],
+        "page_idx": 6
+    },
+    {
+        "type": "text",
+        "text": "Benefit of meta-learning. CACTUs-MAML consistently yields a learning procedure that results in more successful downstream task performance than all other unsupervised methods, including those that learn on top of the embedding that generated meta-training tasks for MAML. We find the same result for CACTUs-ProtoNets for 1-shot downstream tasks. However, as noted by Snell et al. (2017), ProtoNets perform best when meta-training shot and meta-testing shot are matched; this characteristic prevents ProtoNets from improving upon ACAI for 20-way 5-shot Omniglot and upon DeepCluster for 50-shot miniImageNet. We attribute the success of CACTUs-based meta-learning over the embedding-based methods to two factors: its practice in distinguishing between many distinct sets of clusters from modest amounts of signal, and the underlying classes of the meta-testing split data being out-of-distribution. In principle, the latter factor is solely responsible for the success over embedding cluster matching, since this algorithm can be viewed as a meta-learner on embeddings that trivially obtains perfect accuracy (via memorization) on the meta-training tasks. The same factor also helps explain why training from standard network initialization is, in general, competitive with directly using the task-generating embedding as a representation. On the other hand, the MNIST results (Table 7 in Appendix F) suggest that when the meta-training and meta-testing data distributions have perfect overlap and the embedding is well-suited enough that embedding cluster matching can already achieve high performance, CACTUs-MAML yields only a small benefit, as we would expect. ",
+        "bbox": [
+            173,
+            104,
+            825,
+            353
+        ],
+        "page_idx": 7
+    },
+    {
+        "type": "table",
+        "img_path": "images/07ac8e953e870b7ab2d216bc12e1a6585a87a844939cdc0ab6c24a82e821f314.jpg",
+        "table_caption": [
+            "Table 1: Results of unsupervised learning on Omniglot images, averaged over 1000 downstream character recognition tasks. CACTUs experiments use $k = 5 0 0$ clusters for each of $P = 1 0 0$ partitions. Embedding cluster matching uses the same $k$ . For complete results with confidence intervals, see Table 8 in Appendix F. "
+        ],
+        "table_footnote": [],
+        "table_body": "<table><tr><td>Algorithm</td><td>(way, shot)</td><td>(5,1)</td><td>(5,5)</td><td>(20,1)</td><td>(20,5)</td></tr><tr><td>Training from scratch</td><td>52.50%</td><td></td><td>74.78%</td><td>24.91%</td><td>47.62%</td></tr><tr><td>ACAI knn-nearest neighbors</td><td></td><td>57.46%</td><td>81.16%</td><td>39.73%</td><td>66.38%</td></tr><tr><td>ACAIlinear classifier</td><td></td><td>61.08%</td><td>81.82%</td><td>43.20%</td><td>66.33%</td></tr><tr><td>ACAI MLP with dropout</td><td></td><td>51.95%</td><td>77.20%</td><td>30.65%</td><td>58.62%</td></tr><tr><td>ACAI cluster matching</td><td></td><td>54.94%</td><td>71.09%</td><td>32.19%</td><td>45.93%</td></tr><tr><td>ACAI CACTUs-MAML (ours)</td><td>68.84%</td><td></td><td>87.78%</td><td>48.09%</td><td>73.36%</td></tr><tr><td>ACAI CACTUs-ProtoNets (ours)</td><td>68.12%</td><td></td><td>83.58%</td><td>47.75%</td><td>66.27%</td></tr><tr><td>BiGAN knn-nearest neighbors</td><td>49.55%</td><td></td><td>68.06%</td><td>27.37%</td><td>46.70%</td></tr><tr><td>BiGAN linearclassifier</td><td>48.28%</td><td></td><td>68.72%</td><td>27.80%</td><td>45.82%</td></tr><tr><td>BiGAN MLP with dropout</td><td>40.54%</td><td></td><td>62.56%</td><td>19.92%</td><td>40.71%</td></tr><tr><td>BiGAN cluster matching</td><td>43.96%</td><td></td><td>58.62%</td><td>21.54%</td><td>31.06%</td></tr><tr><td>BiGAN CACTUs-MAML (ours)</td><td>58.18%</td><td></td><td>78.66%</td><td>35.56%</td><td>58.62%</td></tr><tr><td>BiGAN CACTUs-ProtoNets (ours)</td><td>54.74%</td><td></td><td>71.69%</td><td>33.40%</td><td>50.62%</td></tr><tr><td>Oracle-MAML (control)</td><td>94.46%</td><td></td><td>98.83%</td><td>84.60%</td><td>96.29%</td></tr><tr><td>Oracle-ProtoNets (control)</td><td>98.35%</td><td></td><td>99.58%</td><td>95.31%</td><td>98.81%</td></tr></table>",
+        "bbox": [
+            256,
+            412,
+            740,
+            648
+        ],
+        "page_idx": 7
+    },
+    {
+        "type": "text",
+        "text": "Different embedding spaces. CACTUs is effective for a variety of embedding learning methods used for task generation. The performance of unsupervised meta-learning can largely be predicted by the performance of the embedding-based non-meta-learning methods. For example, the ACAI embedding does well with Omniglot, leading to the best unsupervised results with ACAI CACTUsMAML. Likewise, on miniImageNet, the best performing prior embedding (DeepCluster) also corresponds to the best performing unsupervised meta-learner (DeepCluster CACTUs-MAML). ",
+        "bbox": [
+            173,
+            659,
+            825,
+            742
+        ],
+        "page_idx": 7
+    },
+    {
+        "type": "text",
+        "text": "Applicability to different tasks. CACTUs-MAML learns an effective prior for a variety of task types. This can be attributed to the application-agnostic task-generation process and the expressive power of MAML (Finn & Levine, 2018). We also observe that, despite all meta-learning models being trained for $N$ -way 1-shot classification of unsupervised tasks, the models work well for a variety of $M$ -way $K$ -shot tasks, where $M \\ \\leq \\ N$ and $K \\ \\leq \\ 5 0$ . As mentioned previously, the representation that CACTUs-ProtoNets learns is best suited for downstream tasks which match the single shot used for meta-training. ",
+        "bbox": [
+            174,
+            750,
+            825,
+            848
+        ],
+        "page_idx": 7
+    },
+    {
+        "type": "text",
+        "text": "Oracle. The penalty for not having ground truth labels to construct near-optimal tasks ranges from substantial to severe, depending on the difficulty of the downstream task. Easier downstream tasks (which have fewer classes and/or more supervision) incur less of a penalty. We conjecture that with such tasks, the difference in the usefulness of the priors matters less since the downstream task-specific evidence has more power to shape the posterior. ",
+        "bbox": [
+            174,
+            854,
+            823,
+            924
+        ],
+        "page_idx": 7
+    },
+    {
+        "type": "table",
+        "img_path": "images/b3cfea21fbe70207bc75235f3f925ee17c07ae5244ac6a3361c724c2f235ba9a.jpg",
+        "table_caption": [
+            "Table 2: Results of unsupervised learning on miniImageNet and CelebA images, averaged over 1000 downstream human-designed tasks. CACTUs experiments use $k = 5 0 0$ for each of $P = 5 0$ partitions. Embedding cluster matching uses the same $k$ . For complete results with confidence intervals, see Tables 9 and 10 in Appendix F. "
+        ],
+        "table_footnote": [],
+        "table_body": "<table><tr><td></td><td colspan=\"4\">miniImageNet</td><td>CelebA</td></tr><tr><td></td><td>(5,1)</td><td>(5,5)</td><td>(5,20)</td><td>(5,50)</td><td>(2,5)</td></tr><tr><td>Training from scratch</td><td>27.59%</td><td>38.48%</td><td>51.53%</td><td>59.63%</td><td>63.19%</td></tr><tr><td>BiGAN knn-nearest neighbors</td><td>25.56%</td><td>31.10%</td><td>37.31%</td><td>43.60%</td><td>56.15%</td></tr><tr><td>BiGAN linear classifier</td><td>27.08%</td><td>33.91%</td><td>44.00%</td><td>50.41%</td><td>58.44%</td></tr><tr><td>BiGAN MLP with dropout</td><td>22.91%</td><td>29.06%</td><td>40.06%</td><td>48.36%</td><td>56.26%</td></tr><tr><td>BiGAN cluster matching</td><td>24.63%</td><td>29.49%</td><td>33.89%</td><td>36.13%</td><td>56.20%</td></tr><tr><td>BiGAN CACTUs-MAML (ours)</td><td>36.24%</td><td>51.28%</td><td>61.33%</td><td>66.91%</td><td>74.98%</td></tr><tr><td>BiGAN CACTUs-ProtoNets (ours)</td><td>36.62%</td><td>50.16%</td><td>59.56%</td><td>63.27%</td><td>65.58%</td></tr><tr><td>DeepCluster knn-nearest neighbors</td><td>28.90%</td><td>42.25%</td><td>56.44%</td><td>63.90%</td><td>61.47%</td></tr><tr><td>DeepCluster linear classifier</td><td>29.44%</td><td>39.79%</td><td>56.19%</td><td>65.28%</td><td>59.57%</td></tr><tr><td>DeepCluster MLP with dropout</td><td>29.03%</td><td>39.67%</td><td>52.71%</td><td>60.95%</td><td>60.65%</td></tr><tr><td>DeepCluster cluster matching</td><td>22.20%</td><td>23.50%</td><td>24.97%</td><td>26.87%</td><td>51.51%</td></tr><tr><td>DeepCluster CACTUs-MAML (ours)</td><td>39.90%</td><td>53.97%</td><td>63.84%</td><td>69.64%</td><td>73.79%</td></tr><tr><td>DeepCluster CACTUs-ProtoNets (ours)</td><td>39.18%</td><td>53.36%</td><td>61.54%</td><td>63.55%</td><td>74.15%</td></tr><tr><td>Oracle-MAML (control)</td><td>46.81%</td><td>62.13%</td><td>71.03%</td><td>75.54%</td><td>87.10%</td></tr><tr><td>Oracle-ProtoNets (control)</td><td>46.56%</td><td>62.29%</td><td>70.05%</td><td>72.04%</td><td>85.13%</td></tr></table>",
+        "bbox": [
+            210,
+            159,
+            787,
+            407
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "Task construction ablation. As seen in Tables 3 and 4, CACTUs-MAML consistently outperforms Hyperplanes-MAML with any margin. We hypothesize that this is due to the issues with zero-margin Hyperplanes-MAML pointed out in Appendix C, and the fact that nonzero-margin HyperplanesMAML is able to use less of the meta-training split to generate tasks than CACTUs-MAML is. We find that using multiple partitions for CACTUs-MAML, while beneficial, is not strictly necessary. Using non-zero margin with Hyperplanes-MAML is crucial for miniImageNet, but not for Omniglot. We conjecture that the enforced degree of separation between classes is needed for miniImageNet because of the dataset’s high diversity. Meta-learning on random tasks or tasks derived from pixel-space clustering (Table 3) results in a prior that is much less useful than any other considered algorithm, including a random network initialization; evidently, practicing badly is worse than not practicing at all. ",
+        "bbox": [
+            173,
+            433,
+            825,
+            585
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "Note on overfitting. Because of the combinatorially many unsupervised tasks we can create from multiple partitions of the dataset, we do not observe substantial overfitting to the unsupervised metatraining tasks. However, we observe that meta-training performance is sometimes worse than metatest time performance, which is likely due to a portion of the automatically generated tasks being based on nonsensical clusters (for examples, see Figure 2). Additionally, we find that, with a few exceptions, using multiple partitions has a regularizing effect on the meta-learner: a diverse task set reduces overfitting to the meta-training tasks and increases the applicability of the learned prior. ",
+        "bbox": [
+            173,
+            592,
+            825,
+            690
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "table",
+        "img_path": "images/e280051126fea76121cf052c226a6d15683009963fc30dd690979c1412e9f902.jpg",
+        "table_caption": [
+            "Table 3: Ablation study of task construction methods on Omniglot. For a more complete set of results with confidence intervals, see Table 8 in Appendix F. "
+        ],
+        "table_footnote": [],
+        "table_body": "<table><tr><td>Algorithm</td><td>(5,1)</td><td>(5,5)</td><td>(20,1)</td><td>(20,5)</td></tr><tr><td>Random-MAML,P= 2400,k = 500</td><td>25.99%</td><td>25.74%</td><td>6.51%</td><td>6.74%</td></tr><tr><td>Pixels CACTUs-MAML,P=1, k = 500</td><td>30.55%</td><td>40.19%</td><td>12.05%</td><td>19.01%</td></tr><tr><td>ACAIHyperplanes-MAML,P= 2400,m= 0</td><td>62.34%</td><td>81.81%</td><td>39.30%</td><td>63.18%</td></tr><tr><td>ACAI Hyperplanes-MAML,P= 2400,m =1.2</td><td>62.44%</td><td>83.20%</td><td>41.86%</td><td>65.23%</td></tr><tr><td>ACAI CACTUs-MAML,P=1,k = 500</td><td>66.49%</td><td>85.60%</td><td>45.04%</td><td>69.14%</td></tr><tr><td>ACAICACTUs-MAML,P=100,k = 500</td><td>68.84%</td><td>87.78%</td><td>48.09%</td><td>73.36%</td></tr><tr><td>BiGANHyperplanes-MAML,P= 2400,m = 0</td><td>53.60%</td><td>74.60%</td><td>29.02%</td><td>50.77%</td></tr><tr><td>BiGANHyperplanes-MAML,P= 2400,m = 0.5</td><td>53.18%</td><td>73.55%</td><td>29.98%</td><td>50.14%</td></tr><tr><td>BiGANCACTUs-MAML,P=1,k= 500</td><td>55.92%</td><td>76.28%</td><td>32.44%</td><td>54.22%</td></tr><tr><td>BiGANCACTUs-MAML,P=100,k= 500</td><td>58.18%</td><td>78.66%</td><td>35.56%</td><td>58.62%</td></tr></table>",
+        "bbox": [
+            210,
+            739,
+            787,
+            906
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "table",
+        "img_path": "images/0a65d9b00a7e20305e684fb38d2993d23ba0551b26d704c18abfd4d99ddb29c3.jpg",
+        "table_caption": [
+            "Table 4: Ablation study of task construction methods on miniImageNet. For a more complete set of results with confidence intervals, see Table 9 in Appendix F. "
+        ],
+        "table_footnote": [],
+        "table_body": "<table><tr><td>Algorithm</td><td>(5,1)</td><td>(5,5)</td><td>(5,20)</td><td>(5,50)</td></tr><tr><td>BiGANHyperplanes-MAML,P= 4800,m =0</td><td>20.00%</td><td>20.00%</td><td>20.00%</td><td>20.00%</td></tr><tr><td>BiGAN Hyperplanes-MAML,P= 4800,m = 0.9</td><td>29.67%</td><td>41.92%</td><td>51.32%</td><td>54.72%</td></tr><tr><td>BiGAN CACTUs-MAML,P=1,k= 500</td><td>37.75%</td><td>52.59%</td><td>62.70%</td><td>67.98%</td></tr><tr><td>BiGANCACTUs-MAML,P= 50,k= 500</td><td>36.24%</td><td>51.28%</td><td>61.33%</td><td>66.91%</td></tr><tr><td>DeepCluster Hyperplanes-MAML,P= 4800,m = 0</td><td>20.02%</td><td>20.01%</td><td>20.00%</td><td>20.01%</td></tr><tr><td>DeepClusterHyperplanes-MAML,P= 4800,m= 0.1</td><td>35.85%</td><td>49.54%</td><td>60.68%</td><td>65.55%</td></tr><tr><td>DeepCluster CACTUs-MAML,P= 1,k = 500</td><td>38.75%</td><td>52.73%</td><td>62.72%</td><td>67.77%</td></tr><tr><td>DeepCluster CACTUs-MAML,P= 50,k = 500</td><td>39.90%</td><td>53.97%</td><td>63.84%</td><td>69.64%</td></tr></table>",
+        "bbox": [
+            196,
+            133,
+            802,
+            270
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "5 DISCUSSION ",
+        "text_level": 1,
+        "bbox": [
+            174,
+            310,
+            310,
+            327
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "We demonstrate that meta-learning on tasks produced using simple mechanisms based on embeddings improves upon the utility of these representations in learning downstream, human-specified tasks. We empirically show that this holds across benchmark datasets and tasks in the few-shot classification literature (Santoro et al., 2016; Ravi & Larochelle, 2017; Finn et al., 2018), task difficulties, and embedding learning methods while fixing key hyperparameters across all experiments. ",
+        "bbox": [
+            173,
+            353,
+            825,
+            424
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "In a sense, CACTUs can be seen as a facilitating interface between an embedding learning method and a meta-learning algorithm. As shown in the results, the meta-learner’s performance significantly depends on the nature and quality of the task-generating embeddings. We can expect our method to yield better performance as the methods that produce these embedding functions improve, becoming better suited for generating diverse yet distinctive clusterings of the data. However, the gap between unsupervised and supervised meta-learning will likely persist because, with the latter, the meta-training task distribution is human-designed to mimic the expected evaluation task distribution as much as possible. Indeed, to some extent, supervised meta-learning algorithms offload the effort of designing and tuning algorithms onto the effort of designing and tuning task distributions. With its evaluation-agnostic task generation, CACTUs-based meta-learning trades off performance in specific use-cases for broad applicability and the ability to train on unlabeled data. In principle, CACTUs-based meta-learning may outperform supervised meta-learning when the latter is trained on a misaligned task distribution. We leave this investigation to future work. ",
+        "bbox": [
+            174,
+            431,
+            825,
+            611
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "While we have demonstrated that $k$ -means is a broadly useful mechanism for constructing tasks from embeddings, it is unlikely that combinations of $k$ -means clusters in learned embedding spaces are universal approximations of arbitrary class definitions. An important direction for future work is to find examples of datasets and human-designed tasks for which CACTUs-based meta-learning results in ineffective downstream learning. This will result in better understanding of the practical scope of applicability for our method, and spur further development in automatic task construction mechanisms for unsupervised meta-learning. ",
+        "bbox": [
+            174,
+            618,
+            825,
+            715
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "A potential concern of our experimental evaluation is that MNIST, Omniglot, and miniImageNet exhibit particular structure in the underlying class distribution (i.e., perfectly balanced classes), since they were designed to be supervised learning benchmarks. In more practical applications of machine learning, such structure would likely not exist. Our CelebA results indicate that CACTUs is effective even in the case of a dataset without neatly balanced classes or attributes. An interesting direction for future work is to better characterize the performance of CACTUs and other unsupervised pretraining methods with highly-unstructured, unlabeled datasets. ",
+        "bbox": [
+            174,
+            723,
+            825,
+            820
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "Since MAML and ProtoNets produce nothing more than a learned representation, our method can be viewed as deriving, from a previous unsupervised representation, a new representation particularly suited for learning downstream tasks. Beyond visual classification tasks, the notion of using unsupervised pre-training is generally applicable to a wide range of domains, including regression, speech (Oord et al., 2018), language (Howard & Ruder, 2018), and reinforcement learning (Shelhamer et al., 2017). Hence, our unsupervised meta-learning approach has the potential to improve unsupervised representations for a variety of such domains, an exciting avenue for future work. ",
+        "bbox": [
+            174,
+            827,
+            825,
+            924
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "ACKNOWLEDGMENTS ",
+        "text_level": 1,
+        "bbox": [
+            176,
+            103,
+            356,
+            118
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "We thank Kelvin Xu, Richard Zhang, Brian Cheung, Ben Poole, Aaron van den Oord, Luke Metz, ¨ Siddharth Reddy, and the anonymous reviewers for feedback on an early draft of this paper. ",
+        "bbox": [
+            173,
+            128,
+            823,
+            159
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "REFERENCES ",
+        "text_level": 1,
+        "bbox": [
+            174,
+            179,
+            287,
+            194
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Yoshua Bengio, Samy Bengio, and Jocelyn Cloutier. Learning a synaptic learning rule. In International Joint Conference on Neural Networks (IJCNN), 1991. ",
+        "bbox": [
+            176,
+            202,
+            823,
+            231
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Yoshua Bengio, Pascal Lamblin, Dan Popovici, and Hugo Larochelle. Greedy layer-wise training of deep networks. In Advances in Neural Information Processing Systems (NIPS), 2007. ",
+        "bbox": [
+            173,
+            238,
+            823,
+            268
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Yoshua Bengio, Aaron Courville, and Pascal Vincent. Representation learning: A review and new perspectives. IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI), 2013. ",
+        "bbox": [
+            174,
+            275,
+            823,
+            306
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "David Berthelot, Colin Raffel, Aurko Roy, and Ian Goodfellow. Understanding and improving interpolation in autoencoders via an adversarial regularizer. arXiv preprint arXiv:1807.07543, 2018. ",
+        "bbox": [
+            174,
+            313,
+            825,
+            356
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Piotr Bojanowski and Armand Joulin. Unsupervised learning by predicting noise. In International Conference on Machine Learning (ICML), 2017. ",
+        "bbox": [
+            173,
+            364,
+            825,
+            393
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Mathilde Caron, Piotr Bojanowski, Armand Joulin, and Matthijs Douze. Deep clustering for unsupervised learning of visual features. In European Conference on Computer Vision (ECCV), 2018. ",
+        "bbox": [
+            174,
+            401,
+            823,
+            444
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Xi Chen, Yan Duan, Rein Houthooft, John Schulman, Ilya Sutskever, and Pieter Abbeel. InfoGAN: Interpretable representation learning by information maximizing generative adversarial nets. In Neural Information Processing Systems (NIPS), 2016. ",
+        "bbox": [
+            176,
+            452,
+            823,
+            496
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Brian Cheung, Jesse A Livezey, Arjun K Bansal, and Bruno A Olshausen. Discovering hidden factors of variation in deep networks. In International Conference on Learning Representations (ICLR), 2015. ",
+        "bbox": [
+            174,
+            503,
+            823,
+            547
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Adam Coates and Andrew Y Ng. Learning feature representations with k-means. In Neural Networks: Tricks of the Trade. Springer, 2012. ",
+        "bbox": [
+            169,
+            554,
+            823,
+            585
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Andrew M Dai and Quoc V Le. Semi-supervised sequence learning. In Neural Information Processing Systems (NIPS), 2015. ",
+        "bbox": [
+            171,
+            592,
+            823,
+            622
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Emily L Denton and Vighnesh Birodkar. Unsupervised learning of disentangled representations from video. In Neural Information Processing Systems (NIPS), 2017. ",
+        "bbox": [
+            171,
+            628,
+            823,
+            660
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Jeff Donahue, Philipp Krahenb ¨ uhl, and Trevor Darrell. Adversarial feature learning. In ¨ International Conference on Learning Representations (ICLR), 2017. ",
+        "bbox": [
+            173,
+            666,
+            823,
+            696
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Vincent Dumoulin, Ishmael Belghazi, Ben Poole, Olivier Mastropietro, Alex Lamb, Martin Arjovsky, and Aaron Courville. Adversarially learned inference. In International Conference on Learning Representations (ICLR), 2017. ",
+        "bbox": [
+            176,
+            704,
+            823,
+            747
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Dumitru Erhan, Yoshua Bengio, Aaron Courville, Pierre-Antoine Manzagol, Pascal Vincent, and Samy Bengio. Why does unsupervised pre-training help deep learning? Journal of Machine Learning Research (JMLR), 2010. ",
+        "bbox": [
+            176,
+            755,
+            825,
+            797
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Chelsea Finn and Sergey Levine. Meta-learning and universality: Deep representations and gradient descent can approximate any learning algorithm. In International Conference on Learning Representations (ICLR), 2018. ",
+        "bbox": [
+            173,
+            806,
+            825,
+            849
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Chelsea Finn, Pieter Abbeel, and Sergey Levine. Model-agnostic meta-learning for fast adaptation of deep networks. In International Conference on Machine Learning (ICML), 2017. ",
+        "bbox": [
+            173,
+            858,
+            823,
+            887
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Chelsea Finn, Kelvin Xu, and Sergey Levine. Probabilistic model-agnostic meta-learning. In Neural Information Processing Systems (NIPS), 2018. ",
+        "bbox": [
+            176,
+            895,
+            823,
+            924
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Vikas K Garg and Adam Kalai. Supervising unsupervised learning. arXiv preprint arXiv:1709.05262, 2017. ",
+        "bbox": [
+            171,
+            103,
+            823,
+            132
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "Abhishek Gupta, Benjamin Eysenbach, Chelsea Finn, and Sergey Levine. Unsupervised metalearning for reinforcement learning. arXiv preprint arXiv:1806.04640, 2018. ",
+        "bbox": [
+            174,
+            140,
+            821,
+            170
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "Trevor Hastie, Robert Tibshirani, and Jerome Friedman. Unsupervised learning. In The Elements of Statistical Learning. Springer, 2009. ",
+        "bbox": [
+            174,
+            179,
+            823,
+            208
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Identity mappings in deep residual networks. European Conference on Computer Vision (ECCV), 2016. ",
+        "bbox": [
+            173,
+            217,
+            825,
+            246
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "Irina Higgins, Loic Matthey, Arka Pal, Christopher Burgess, Xavier Glorot, Matthew Botvinick, Shakir Mohamed, and Alexander Lerchner. $\\beta$ -VAE: Learning basic visual concepts with a constrained variational framework. In International Conference on Learning Representations (ICLR), 2017. ",
+        "bbox": [
+            173,
+            253,
+            825,
+            310
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "Geoffrey E Hinton, Simon Osindero, and Yee-Whye Teh. A fast learning algorithm for deep belief nets. Neural Computation, 2006. ",
+        "bbox": [
+            173,
+            319,
+            823,
+            349
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "Jeremy Howard and Sebastian Ruder. Universal language model fine-tuning for text classification. In Association for Computational Linguistics (ACL), 2018. ",
+        "bbox": [
+            173,
+            357,
+            821,
+            387
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "Sergey Ioffe and Christian Szegedy. Batch normalization: Accelerating deep network training by reducing internal covariate shift. In International Conference on Machine Learning (ICML), 2015. ",
+        "bbox": [
+            174,
+            395,
+            823,
+            425
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "Jaehong Kim, Sangyeul Lee, Sungwan Kim, Moonsu Cha, Jung Kwon Lee, Youngduck Choi, Yongseok Choi, Dong-Yeon Cho, and Jiwon Kim. Auto-Meta: Automated gradient based meta learner search. arXiv preprint arXiv:1806.06927, 2018. ",
+        "bbox": [
+            173,
+            433,
+            825,
+            477
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "Diederik Kingma and Jimmy Ba. Adam: A method for stochastic optimization. In Proceedings of the 3rd International Conference on Learning Representations (ICLR), 2014. ",
+        "bbox": [
+            174,
+            484,
+            823,
+            515
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "Diederik P Kingma, Shakir Mohamed, Danilo Jimenez Rezende, and Max Welling. Semi-supervised learning with deep generative models. In Neural Information Processing Systems (NIPS), 2014. ",
+        "bbox": [
+            173,
+            522,
+            823,
+            553
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "Philipp Krahenb ¨ uhl, Carl Doersch, Jeff Donahue, and Trevor Darrell. Data-dependent initializa- ¨ tions of convolutional neural networks. In International Conference on Learning Representations (ICLR), 2016. ",
+        "bbox": [
+            174,
+            560,
+            825,
+            603
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "Quoc V Le, Marc’Aurelio Ranzato, Rajat Monga, Matthieu Devin, Gregory S. Corrado, Kai Chen, Jeffrey Dean, and Andrew Y Ng. Building high-level features using large scale unsupervised learning. In International Conference on Acoustics, Speech and Signal Processing (ICASSP), 2013. ",
+        "bbox": [
+            174,
+            612,
+            825,
+            669
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "Michael F Mathieu, Junbo Jake Zhao, Junbo Zhao, Aditya Ramesh, Pablo Sprechmann, and Yann LeCun. Disentangling factors of variation in deep representation using adversarial training. In Neural Information Processing Systems (NIPS), 2016. ",
+        "bbox": [
+            178,
+            678,
+            821,
+            722
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "Luke Metz, Niru Maheswaranathan, Brian Cheung, and Jascha Sohl-Dickstein. Learning to learn without labels. In International Conference on Learning Representations (ICLR), 2018. ",
+        "bbox": [
+            171,
+            729,
+            823,
+            758
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "Tsendsuren Munkhdalai and Hong Yu. Meta networks. In International Conference on Machine Learning (ICML), 2017. ",
+        "bbox": [
+            173,
+            767,
+            823,
+            796
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "Devang K Naik and RJ Mammone. Meta-neural networks that learn by learning. In International Joint Conference on Neural Networks (IJCNN), 1992. ",
+        "bbox": [
+            169,
+            805,
+            823,
+            834
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "Avital Oliver, Augustus Odena, Colin Raffel, Ekin D Cubuk, and Ian J Goodfellow. Realistic evaluation of deep semi-supervised learning algorithms. In International Conference on Learning Representations (ICLR), 2018. ",
+        "bbox": [
+            174,
+            843,
+            821,
+            886
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "Aaron van den Oord, Yazhe Li, and Oriol Vinyals. Representation learning with contrastive predictive coding. arXiv preprint arXiv:1807.03748, 2018. ",
+        "bbox": [
+            173,
+            895,
+            820,
+            924
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "Alec Radford, Luke Metz, and Soumith Chintala. Unsupervised representation learning with deep convolutional generative adversarial networks. In International Conference on Learning Representations (ICLR), 2016. ",
+        "bbox": [
+            174,
+            103,
+            823,
+            146
+        ],
+        "page_idx": 12
+    },
+    {
+        "type": "text",
+        "text": "Alec Radford, Karthik Narasimhan, Tim Salimans, and Ilya Sutskever. Improving language understanding by generative pre-training. Preprint, 2018. ",
+        "bbox": [
+            171,
+            154,
+            821,
+            183
+        ],
+        "page_idx": 12
+    },
+    {
+        "type": "text",
+        "text": "Prajit Ramachandran, Peter J Liu, and Quoc V Le. Unsupervised pretraining for sequence to sequence learning. In Empirical Methods in Natural Language Processing (EMNLP), 2017. ",
+        "bbox": [
+            171,
+            190,
+            821,
+            220
+        ],
+        "page_idx": 12
+    },
+    {
+        "type": "text",
+        "text": "Marc’Aurelio Ranzato, Christopher Poultney, Sumit Chopra, and Yann LeCun. Efficient learning of sparse representations with an energy-based model. In Neural Information Processing Systems (NIPS), 2006. ",
+        "bbox": [
+            174,
+            227,
+            823,
+            270
+        ],
+        "page_idx": 12
+    },
+    {
+        "type": "text",
+        "text": "Antti Rasmus, Mathias Berglund, Mikko Honkala, Harri Valpola, and Tapani Raiko. Semisupervised learning with ladder networks. In Neural Information Processing Systems (NIPS), 2015. ",
+        "bbox": [
+            174,
+            277,
+            823,
+            320
+        ],
+        "page_idx": 12
+    },
+    {
+        "type": "text",
+        "text": "Sachin Ravi and Hugo Larochelle. Optimization as a model for few-shot learning. In International Conference on Learning Representations (ICLR), 2017. ",
+        "bbox": [
+            171,
+            329,
+            825,
+            358
+        ],
+        "page_idx": 12
+    },
+    {
+        "type": "text",
+        "text": "Scott Reed, Kihyuk Sohn, Yuting Zhang, and Honglak Lee. Learning to disentangle factors of variation with manifold interaction. In International Conference on Machine Learning (ICML), 2014. ",
+        "bbox": [
+            176,
+            364,
+            823,
+            409
+        ],
+        "page_idx": 12
+    },
+    {
+        "type": "text",
+        "text": "Olga Russakovsky, Jia Deng, Hao Su, Jonathan Krause, Sanjeev Satheesh, Sean Ma, Zhiheng Huang, Andrej Karpathy, Aditya Khosla, Michael Bernstein, Alexander C. Berg, and Fei-Fei Li. ImageNet large scale visual recognition challenge. International Journal of Computer Vision (IJCV), 2015. ",
+        "bbox": [
+            173,
+            416,
+            825,
+            473
+        ],
+        "page_idx": 12
+    },
+    {
+        "type": "text",
+        "text": "Tim Salimans, Ian Goodfellow, Wojciech Zaremba, Vicki Cheung, Alec Radford, and Xi Chen. Improved techniques for training GANs. In Neural Information Processing Systems (NIPS), 2016. ",
+        "bbox": [
+            171,
+            481,
+            823,
+            510
+        ],
+        "page_idx": 12
+    },
+    {
+        "type": "text",
+        "text": "Adam Santoro, Sergey Bartunov, Matthew Botvinick, Daan Wierstra, and Timothy Lillicrap. Metalearning with memory-augmented neural networks. In International Conference on Machine Learning (ICML), 2016. ",
+        "bbox": [
+            173,
+            517,
+            823,
+            560
+        ],
+        "page_idx": 12
+    },
+    {
+        "type": "text",
+        "text": "Jurgen Schmidhuber. ¨ Evolutionary principles in self-referential learning. PhD thesis, Institut fur¨ Informatik, Technische Universitat M¨ unchen, 1987.¨ ",
+        "bbox": [
+            171,
+            568,
+            823,
+            597
+        ],
+        "page_idx": 12
+    },
+    {
+        "type": "text",
+        "text": "Evan Shelhamer, Parsa Mahmoudieh, Max Argus, and Trevor Darrell. Loss is its own reward: Selfsupervision for reinforcement learning. In International Conference on Learning Representations (ICLR), 2017. ",
+        "bbox": [
+            173,
+            604,
+            825,
+            647
+        ],
+        "page_idx": 12
+    },
+    {
+        "type": "text",
+        "text": "Jake Snell, Kevin Swersky, and Richard S Zemel. Prototypical networks for few-shot learning. In Neural Information Processing Systems (NIPS), 2017. ",
+        "bbox": [
+            169,
+            655,
+            823,
+            685
+        ],
+        "page_idx": 12
+    },
+    {
+        "type": "text",
+        "text": "Nitish Srivastava, Geoffrey Hinton, Alex Krizhevsky, Ilya Sutskever, and Ruslan Salakhutdinov. Dropout: A simple way to prevent neural networks from overfitting. Journal of Machine Learning Research (JMLR), 2014. ",
+        "bbox": [
+            174,
+            691,
+            825,
+            736
+        ],
+        "page_idx": 12
+    },
+    {
+        "type": "text",
+        "text": "Pascal Vincent, Hugo Larochelle, Yoshua Bengio, and Pierre-Antoine Manzagol. Extracting and composing robust features with denoising autoencoders. In International Conference on Machine Learning (ICML), 2008. ",
+        "bbox": [
+            173,
+            742,
+            825,
+            786
+        ],
+        "page_idx": 12
+    },
+    {
+        "type": "text",
+        "text": "Pascal Vincent, Hugo Larochelle, Isabelle Lajoie, Yoshua Bengio, and Pierre-Antoine Manzagol. Stacked denoising autoencoders: Learning useful representations in a deep network with a local denoising criterion. Journal of Machine Learning Research (JMLR), 2010. ",
+        "bbox": [
+            173,
+            792,
+            825,
+            837
+        ],
+        "page_idx": 12
+    },
+    {
+        "type": "text",
+        "text": "Oriol Vinyals, Charles Blundell, Timothy Lillicrap, Koray Kavukcuoglu, and Daan Wierstra. Matching networks for one shot learning. In Neural Information Processing Systems (NIPS), 2016. ",
+        "bbox": [
+            173,
+            844,
+            821,
+            875
+        ],
+        "page_idx": 12
+    },
+    {
+        "type": "text",
+        "text": "Dong Yu, Li Deng, and George Dahl. Roles of pre-training and fine-tuning in context-dependent DBN-HMMs for real-world speech recognition. In NIPS Workshop on Deep Learning and Unsupervised Feature Learning, 2010. ",
+        "bbox": [
+            174,
+            881,
+            823,
+            924
+        ],
+        "page_idx": 12
+    },
+    {
+        "type": "text",
+        "text": "Richard Zhang, Phillip Isola, and Alexei A Efros. Split-brain autoencoders: Unsupervised learning by cross-channel prediction. In Computer Vision and Pattern Recognition (CVPR), 2017. ",
+        "bbox": [
+            173,
+            103,
+            825,
+            132
+        ],
+        "page_idx": 13
+    },
+    {
+        "type": "text",
+        "text": "Xiaojin Zhu. Semi-supervised learning. In Encyclopedia of Machine Learning. Springer, 2011. ",
+        "bbox": [
+            176,
+            141,
+            799,
+            156
+        ],
+        "page_idx": 13
+    },
+    {
+        "type": "text",
+        "text": "APPENDIX A THE EMBEDDING LEARNING ZOO ",
+        "text_level": 1,
+        "bbox": [
+            176,
+            102,
+            584,
+            118
+        ],
+        "page_idx": 14
+    },
+    {
+        "type": "text",
+        "text": "We evaluate four distinct methods from prior work for learning the task-generating embeddings. ",
+        "bbox": [
+            173,
+            135,
+            802,
+            150
+        ],
+        "page_idx": 14
+    },
+    {
+        "type": "text",
+        "text": "In adversarially constrained autoencoder interpolation (ACAI), a convolutional autoencoder’s pixelwise $L ^ { 2 }$ loss is regularized with a term encouraging meaningful interpolations in the latent space (Berthelot et al., 2018). Specifically, a critic network takes as input a synthetic image generated from a convex combination of the latents of two dataset samples, and regresses to the mixing factor. The decoder of the autoencoder and the generator for the critic are one and the same. The regularization term is minimized when the autoencoder fools the critic into predicting that the synthetic image is a real sample. ",
+        "bbox": [
+            174,
+            156,
+            825,
+            253
+        ],
+        "page_idx": 14
+    },
+    {
+        "type": "text",
+        "text": "The bidirectional GAN (BiGAN) is an instance of a generative-adversarial framework in which the generator produces both synthetic image and embedding from real embedding and image, respectively (Donahue et al., 2017; Dumoulin et al., 2017). Discrimination is done in joint imageembedding space. ",
+        "bbox": [
+            176,
+            261,
+            823,
+            316
+        ],
+        "page_idx": 14
+    },
+    {
+        "type": "text",
+        "text": "The DeepCluster method does discriminative clustering by alternating between clustering the features of a convolutional neural network and using the clusters as labels to optimize the network weights via backpropagating a standard classification loss (Caron et al., 2018). ",
+        "bbox": [
+            174,
+            324,
+            823,
+            366
+        ],
+        "page_idx": 14
+    },
+    {
+        "type": "text",
+        "text": "The InfoGAN framework conceptually decomposes the generator’s input into a latent code and incompressible noise (Chen et al., 2016). The structure of the latent code is hand-specified based on knowledge of the dataset. The canonical GAN minimax objective is regularized with a mutual information term between the code and the generated image. In practice, this term is optimized using variational inference, involving the approximation of the posterior with an auxiliary distribution $Q$ (code|image) parameterized by a recognition network. ",
+        "bbox": [
+            174,
+            373,
+            825,
+            457
+        ],
+        "page_idx": 14
+    },
+    {
+        "type": "text",
+        "text": "Whereas ACAI explicitly optimizes pixel-wise reconstruction error, BiGAN only encourages the fidelity of generated image and latent samples with respect to their respective prior distributions. While InfoGAN also encourages the fidelity of generated images, it leverages domain-specific knowledge to impose a favorable structure on the embedding space and information-theoretic methods for optimization. DeepCluster departs from the aforementioned methods in that it is not concerned with generation or decoding, and only seeks to learn general-purpose visual features by way of end-to-end discriminative clustering. ",
+        "bbox": [
+            174,
+            463,
+            825,
+            560
+        ],
+        "page_idx": 14
+    },
+    {
+        "type": "text",
+        "text": "APPENDIX B DATASET INFORMATION ",
+        "text_level": 1,
+        "bbox": [
+            174,
+            584,
+            503,
+            601
+        ],
+        "page_idx": 14
+    },
+    {
+        "type": "text",
+        "text": "The Omniglot dataset consists of 1623 characters each with 20 hand-drawn examples. Ignoring the alphabets from which the characters originate, we use 1100, 100, and 423 characters for our meta-training, meta-validation, and meta-testing splits. The miniImageNet dataset consists of 100 classes each with 600 examples. The images are predominantly natural and realistic. We use the same meta-training/meta-validation/meta-testing splits of 64/16/20 classes as proposed by Ravi & Larochelle (2017). The CelebA dataset includes 202,599 facial images of celebrities and 40 binary attributes that annotate every image. We follow the prescribed 162,770/19,867/19,962 data split. ",
+        "bbox": [
+            174,
+            617,
+            825,
+            715
+        ],
+        "page_idx": 14
+    },
+    {
+        "type": "text",
+        "text": "For Omniglot and miniImageNet, supervised meta-learning tasks and evaluation tasks are constructed exactly as detailed in Section 2.2: for an $N$ -way $K$ -shot task with $Q$ queries per class, we sample $N$ classes from the data split and $K + Q$ datapoints per class, labeling the task’s data with a random permutation of $N$ one-hot vectors. ",
+        "bbox": [
+            174,
+            722,
+            823,
+            777
+        ],
+        "page_idx": 14
+    },
+    {
+        "type": "text",
+        "text": "For CelebA, we consider binary classification tasks (i.e., 2-way), each defined by 3 attributes and an ordering of 3 Booleans, one for each attribute. Every image in a task-specific class shares all task-specific attributes with each other and none with images in the other class. For example, the task illustrated in Figure 2 involves distinguishing between images whose subjects satisfy not Sideburns, Straight Hair, and not Young, and those whose subjects satisfy Sideburns, not Straight Hair, and Young. To keep with the idea of having distinct classes for meta-training and meta-testing, we split the task-defining attributes. For the supervised meta-learning oracle, we construct meta-training tasks from the first 20 attributes (when alphabetically ordered), meta-validation tasks from the next 10, and meta-testing tasks from the last 10. Discarding tasks with too few examples in either class, this results in 4287, 391, and 402 task prototypes (but many more possible tasks). We use the same meta-test time tasks to evaluate the unsupervised methods. We only consider assessment with 5-shot tasks because, given that there are multiple attributes other than the task-defining ones, any 1-shot task is likely to be ill-defined. ",
+        "bbox": [
+            174,
+            785,
+            823,
+            922
+        ],
+        "page_idx": 14
+    },
+    {
+        "type": "text",
+        "text": "",
+        "bbox": [
+            174,
+            103,
+            825,
+            146
+        ],
+        "page_idx": 15
+    },
+    {
+        "type": "text",
+        "text": "APPENDIX C TASK CONSTRUCTION VIA RANDOM HYPERPLANES ",
+        "text_level": 1,
+        "bbox": [
+            178,
+            166,
+            732,
+            183
+        ],
+        "page_idx": 15
+    },
+    {
+        "type": "text",
+        "text": "Given a set of embedding points $\\left\\{ \\mathbf { z } _ { i } \\right\\}$ in a space $\\mathcal { Z }$ , a simple way of defining a partition $\\mathcal { P } = \\{ \\mathcal { C } _ { c } \\}$ on $\\left\\{ \\mathbf { z } _ { i } \\right\\}$ is to use random hyperplanes to slice $\\mathcal { Z }$ into subspaces and assign the embeddings that lie in the $c$ -th subspace to subset $\\mathcal { C } _ { c }$ . However, a hyperplane slicing can group together two arbitrarily far embeddings, or separate two arbitrarily close ones; given our assumption that good embedding spaces have a semantically meaningful metric, this creates ill-defined classes. This problem can be partially alleviated by extending the hyperplane boundaries with a non-zero margin, as empirically shown in Section 4.2. ",
+        "bbox": [
+            174,
+            198,
+            825,
+            296
+        ],
+        "page_idx": 15
+    },
+    {
+        "type": "text",
+        "text": "We now describe how to generate tasks via random hyperplanes in the embedding space. We first describe a procedure to generate a partition $\\mathcal { P }$ of the set of embeddings $\\left\\{ \\mathbf { z } _ { i } \\right\\}$ for constructing metatraining tasks. A given hyperplane slices the embedding space into two, so for an $N$ -way task, we need $\\bar { H } = \\lceil \\log _ { 2 } \\bar { N } \\rceil$ hyperplanes to define sufficiently many subsets/classes for a task. To randomly define a hyperplane in $d$ -dimensional embedding space, we sample a normal vector $\\mathbf { n }$ and a point on the plane $\\mathbf { z } _ { 0 }$ , each with $d$ elements. For an embedding point $\\mathbf { z }$ , the signed point-plane distance is given by $\\frac { \\mathbf { n } } { \\left| \\mathbf { n } \\right| _ { 2 } } \\cdot \\left( \\mathbf { z } - \\mathbf { z } _ { 0 } \\right)$ . Defining $H$ hyperplanes in this manner, we discard embeddings for which the signed point-plane distance to any of the $H$ hyperplanes lies within $( - m , m )$ , where $m$ is a desired margin. The $H$ hyperplanes collectively define $\\mathbf { \\bar { 2 } } ^ { H }$ subspaces. We assign embedding points in the $c$ -th subspace to subset $\\mathcal { C } _ { c }$ . We define the partition as $\\mathcal { P } = \\{ \\mathcal { C } _ { c } \\}$ . We prune subsets that do not have at least $R = K _ { \\mathrm { m - t r } } + Q$ members, and check that the partition has at least $N$ remaining subsets; if not, we reject the partition and restart the procedure. After obtaining partitions $\\{ \\mathcal P _ { p } \\}$ , meta-training tasks can be generated by following Algorithm 1 from Line 4. ",
+        "bbox": [
+            174,
+            304,
+            825,
+            488
+        ],
+        "page_idx": 15
+    },
+    {
+        "type": "text",
+        "text": "In terms of practical implementation, we pre-compute 1000 hyperplanes and pruned pairs of subsets of $\\left\\{ \\mathbf { z } _ { i } \\right\\}$ . We generate partitions by sampling combinations of the hyperplanes and taking intersections of their associated subsets to define the elements of the partition. We determine the number of partitions needed for a given Hyperplanes-MAML run by the number of meta-training tasks desired for the meta-learner: we fix 100 tasks per partition. ",
+        "bbox": [
+            174,
+            494,
+            825,
+            564
+        ],
+        "page_idx": 15
+    },
+    {
+        "type": "text",
+        "text": "APPENDIX D MNIST EXPERIMENTS ",
+        "text_level": 1,
+        "bbox": [
+            176,
+            585,
+            495,
+            602
+        ],
+        "page_idx": 15
+    },
+    {
+        "type": "text",
+        "text": "The MNIST dataset consists of 70,000 hand-drawn examples of the 10 numerical digits. Our split respects the original MNIST 60,000/10,000 training/testing split. We assess on 10-way classification tasks. This setup results in examples from all 10 digits being present for both meta-training and meta-testing, making the probem setting essentially equivalent to that of semi-supervised learning sans a fixed permutation of the labels. The MNIST scenario is thus a special case of the problem setting considered in the rest of the paper. For MNIST, we only experiment with MAML as the meta-learning algorithm. ",
+        "bbox": [
+            174,
+            617,
+            825,
+            715
+        ],
+        "page_idx": 15
+    },
+    {
+        "type": "text",
+        "text": "For ACAI and InfoGAN we constructed the meta-validation split from the last 5,000 examples of the meta-training split; for BiGAN this figure was 10,000. After training the ACAI model and inferring embeddings, manually assigning labels to 10 clusters by inspection results in a classification accuracy of $9 6 . 0 0 \\%$ on the testing split. As the ACAI authors observe, we found it important to whiten the ACAI embeddings before clustering. The same metric for the InfoGAN embedding (taking an argmax over the categorical dimensions instead of actually running clustering) is $9 6 . 8 3 \\%$ . Note that these results are an upper-bound for embedding cluster matching. To see this, consider the 10-way 1-shot scenario. 1 example sampled from each cluster is insufficient to guarantee the optimal label for that cluster; 1 example sampled from each label is not guaranteed to each end up in the optimal category. ",
+        "bbox": [
+            174,
+            722,
+            825,
+            861
+        ],
+        "page_idx": 15
+    },
+    {
+        "type": "text",
+        "text": "Aside from CACTUs-MAML, embedding $k _ { \\mathrm { n n } }$ -nearest neighbors, embedding linear classifier, and embedding direct clustering, we also ran CACTUs-MAML on embeddings instead of raw images, using a simple model with 2 hidden layers with 64 units each and ReLU activation, and all other MAML hyperparameters being the same as in Table 5. ",
+        "bbox": [
+            176,
+            867,
+            823,
+            924
+        ],
+        "page_idx": 15
+    },
+    {
+        "type": "text",
+        "text": "Departing from the fixed $k = 5 0 0$ used for all other datasets, we deliberately use $k = 1 0$ to better understand the limitations of CACTUs-MAML. The results can be seen in Table 7 in Appendix B. In brief, with the better embeddings (ACAI and InfoGAN), there is only little benefit of CACTUsMAML over embedding cluster matching. Additionally, even in the best cases, CACTUs-MAML falls short of state-of-the-art semi-supervised learning methods. ",
+        "bbox": [
+            174,
+            103,
+            825,
+            174
+        ],
+        "page_idx": 16
+    },
+    {
+        "type": "text",
+        "text": "APPENDIX E HYPERPARAMETERS AND ARCHITECTURES ",
+        "text_level": 1,
+        "bbox": [
+            178,
+            194,
+            661,
+            210
+        ],
+        "page_idx": 16
+    },
+    {
+        "type": "text",
+        "text": "E.1 MAML ",
+        "text_level": 1,
+        "bbox": [
+            173,
+            224,
+            269,
+            239
+        ],
+        "page_idx": 16
+    },
+    {
+        "type": "table",
+        "img_path": "images/2ec3224bb93dcc723ac1de7bc1bd6a5ca9baee579e7397416bb952c4c0951f97.jpg",
+        "table_caption": [
+            "Table 5: MAML hyperparameter summary. "
+        ],
+        "table_footnote": [],
+        "table_body": "<table><tr><td>Hyperparameter</td><td>MNIST</td><td>Omniglot</td><td>miniImageNet</td><td>CelebA</td></tr><tr><td>Input size</td><td>28×28</td><td>28×28</td><td>84 ×84×3</td><td>84 ×84×3</td></tr><tr><td>Outer (meta) learning rate</td><td>0.001</td><td>0.001</td><td>0.001</td><td>0.001</td></tr><tr><td>Inner learning rate</td><td>0.05</td><td>0.05</td><td>0.05</td><td>0.05</td></tr><tr><td>Task batch size</td><td>8</td><td>8</td><td>8</td><td>8</td></tr><tr><td>Inner adaptation steps (meta-training)</td><td>5</td><td></td><td>5</td><td>5</td></tr><tr><td>Meta-training iterations</td><td>30,000</td><td>5 30,000</td><td>60,000</td><td>60,000</td></tr><tr><td>Adaptation steps (evaluation)</td><td>50</td><td>50</td><td>50</td><td>50</td></tr><tr><td>Classes per task (meta-training)</td><td>10</td><td>20</td><td>5</td><td>2</td></tr><tr><td>Shots per class (meta-training)</td><td>1</td><td>1</td><td>1</td><td>1</td></tr><tr><td>Queries per class</td><td>5</td><td>5</td><td>5</td><td>5</td></tr></table>",
+        "bbox": [
+            189,
+            286,
+            810,
+            455
+        ],
+        "page_idx": 16
+    },
+    {
+        "type": "text",
+        "text": "For MNIST and Omniglot we use the same 4-block convolutional architecture as used by Finn et al. (2017) for their Omniglot experiments, but with 32 filters (instead of 64) for each convolutional layer for consistency with the model used for miniImageNet and CelebA, which is the same as what Finn et al. (2017) used for their miniImageNet experiments. When evaluating the meta-learned 20-way Omniglot model with 5-way tasks, we prune the unused output dimensions. The outer optimizer is Adam (Kingma & Ba, 2014), and the inner optimizer is SGD. We build on the authors’ publicly available codebase found at https://github.com/cbfinn/maml. ",
+        "bbox": [
+            173,
+            484,
+            825,
+            583
+        ],
+        "page_idx": 16
+    },
+    {
+        "type": "text",
+        "text": "When using batch normalization (Ioffe & Szegedy, 2015) to process a task’s training or query inputs, we observe that using only 1 query datapoint per class can allow the model to exploit batch statistics, learning a strategy analogous to a process of elimination that causes significant, but spurious, improvement in accuracy. To mitigate this, we fix 5 queries per class for every task’s evaluation phase, meta-training or meta-testing. ",
+        "bbox": [
+            174,
+            589,
+            825,
+            660
+        ],
+        "page_idx": 16
+    },
+    {
+        "type": "text",
+        "text": "E.2 PROTONETS ",
+        "text_level": 1,
+        "bbox": [
+            174,
+            676,
+            302,
+            690
+        ],
+        "page_idx": 16
+    },
+    {
+        "type": "table",
+        "img_path": "images/05251c8e5fc194d39ee7bd093f158e80b90d1f4e442d42b8db485abe6931ad93.jpg",
+        "table_caption": [
+            "Table 6: ProtoNets hyperparameter summary. "
+        ],
+        "table_footnote": [],
+        "table_body": "<table><tr><td>Hyperparameter</td><td>Omniglot</td><td>miniImageNet</td><td>CelebA</td></tr><tr><td>Input size</td><td>28×28</td><td>84×84×3</td><td>84×84×3</td></tr><tr><td>Learning rate</td><td>0.001</td><td>0.001</td><td>0.001</td></tr><tr><td>Task batch size</td><td>1</td><td>1</td><td>1</td></tr><tr><td>Training iterations</td><td>30,000</td><td>60,000</td><td>60,000</td></tr><tr><td>Classes per task (meta-training)</td><td>20</td><td>5</td><td>2</td></tr><tr><td>Shots per class (meta-training)</td><td>1</td><td>1</td><td>1</td></tr><tr><td>Queries per class (meta-training/meta-testing)</td><td>15/5</td><td>15/5</td><td>15/5</td></tr></table>",
+        "bbox": [
+            199,
+            738,
+            799,
+            866
+        ],
+        "page_idx": 16
+    },
+    {
+        "type": "text",
+        "text": "For the three considered datasets we use the same architecture as used by Snell et al. (2017) for their Omniglot and miniImageNet experiments. This is a 4-block convolutional architecture with each block consisting of a convolutional layer with $6 4 \\ 3 \\times 3$ filters, stride 1, and padding 1, followed by BatchNorm, ReLU activation, and $2 \\times 2$ MaxPooling. The ProtoNets embedding is simply the flattened output of the last block. We follow the authors and use the Adam optimizer, but do not use a learning rate scheduler. We build upon the authors’ publicly available codebase found at https://github.com/jakesnell/prototypical-networks. ",
+        "bbox": [
+            176,
+            895,
+            823,
+            924
+        ],
+        "page_idx": 16
+    },
+    {
+        "type": "text",
+        "text": "",
+        "bbox": [
+            174,
+            103,
+            825,
+            174
+        ],
+        "page_idx": 17
+    },
+    {
+        "type": "text",
+        "text": "E.3 CACTUS ",
+        "text_level": 1,
+        "bbox": [
+            174,
+            189,
+            284,
+            204
+        ],
+        "page_idx": 17
+    },
+    {
+        "type": "text",
+        "text": "For Omniglot, miniImageNet, and CelebA we fix the number of clusters $k$ to be 500. For Omniglot we choose the number of partitions $P = 1 0 0$ , but in the interest of keeping runtime manageable, choose $P = 5 0$ for miniImageNet and CelebA. ",
+        "bbox": [
+            176,
+            217,
+            825,
+            258
+        ],
+        "page_idx": 17
+    },
+    {
+        "type": "text",
+        "text": "E.4 USE OF UNSUPERVISED LEARNING METHODS ",
+        "text_level": 1,
+        "bbox": [
+            174,
+            275,
+            535,
+            289
+        ],
+        "page_idx": 17
+    },
+    {
+        "type": "text",
+        "text": "ACAI (Berthelot et al., 2018): We run ACAI for MNIST and Omniglot. We pad the images by 2 and use the authors’ architecture. We use a 256-dimensional embedding for all datasets. We build upon the authors’ publicly available codebase found at https://github.com/ brain-research/acai. ",
+        "bbox": [
+            174,
+            300,
+            823,
+            357
+        ],
+        "page_idx": 17
+    },
+    {
+        "type": "text",
+        "text": "We unsuccessfully try running ACAI on $6 4 \\times 6 4$ miniImageNet and CelebA. To facilitate this input size, we add one block consisting of two convolutional layers (512 filters each) and one downsampling/upsampling layer to the encoder and decoder. However, because of ACAI’s pixel-wise reconstruction loss, for these datasets the ACAI embedding prioritizes information about the few “features” that dominate the reconstruction pixel count, resulting in clusters that only corresponded to a limited range of factors, such as background color and pose. For curiosity’s sake, we tried running meta-learning on tasks derived from these uninteresting clusters anyways, and found that the meta-learner quickly produced a learning procedure that obtained high accuracy on the meta-training tasks. However, this learned prior was not useful for solving downstream tasks. ",
+        "bbox": [
+            174,
+            363,
+            825,
+            489
+        ],
+        "page_idx": 17
+    },
+    {
+        "type": "text",
+        "text": "BiGAN (Donahue et al., 2017): For MNIST, we follow the BiGAN authors and specify a uniform 50-dimensional prior on the unit hypercube for the latent. The BiGAN authors use a 200- dimensional version of the same prior for their ImageNet experiments, so we follow suit for Omniglot, miniImageNet, and CelebA. For MNIST and Omniglot, we use the permutation-invariant architecture (i.e. fully connected layers only) used by the authors for their MNIST results; for miniImageNet and CelebA, we randomly crop to $6 4 \\times 6 4$ and use the AlexNet-inspired architecture used by Donahue et al. (2017) for their ImageNet results. We build upon the authors’ publicly available codebase found at https://github.com/jeffdonahue/bigan. ",
+        "bbox": [
+            174,
+            496,
+            825,
+            608
+        ],
+        "page_idx": 17
+    },
+    {
+        "type": "text",
+        "text": "DeepCluster (Caron et al., 2018): We run DeepCluster for miniImageNet and CelebA, which we respectively randomly crop and resize to $6 4 \\times 6 4$ . We modify the first layer of the AlexNet architecture used by the authors to accommodate this input size. We follow the authors and use the input to the (linear) output layer as the embedding. These are 4096-dimensional, so we follow the authors and apply PCA to reduce the dimensionality to 256, followed by whitening. We build upon the authors’ publicly available codebase found at https://github.com/facebookresearch/ deepcluster. ",
+        "bbox": [
+            174,
+            614,
+            825,
+            712
+        ],
+        "page_idx": 17
+    },
+    {
+        "type": "text",
+        "text": "InfoGAN (Chen et al., 2016): We only run InfoGAN for MNIST. We follow the InfoGAN authors and specify the product of a 10-way categorical distribution and a 2-dimensional uniform distribution as the latent code. We use the authors’ architecture. Given an image, we use the recognition network to obtain its embedding. We build upon the authors’ publicly available codebase found at https: //github.com/openai/InfoGAN. ",
+        "bbox": [
+            174,
+            718,
+            825,
+            787
+        ],
+        "page_idx": 17
+    },
+    {
+        "type": "text",
+        "text": "APPENDIX F EXPERIMENTAL RESULTS ",
+        "text_level": 1,
+        "bbox": [
+            176,
+            809,
+            513,
+            824
+        ],
+        "page_idx": 17
+    },
+    {
+        "type": "text",
+        "text": "This section containsfull experimental results for the MNIST, Omniglot, miniImageNet, and CelebA datasets, including consolidated versions of the tables found in the main text. The metric is classification accuracy averaged over 1000 tasks based on human-specified labels of the testing split, with $9 5 \\%$ confidence intervals. $d$ : dimensionality of embedding, $h$ : number of hidden units in a fully connected layer, $k$ : number of clusters in a partition, $P$ : number of partitions used during meta-learning, $m$ : margin on boundary-defining hyperplanes. ",
+        "bbox": [
+            174,
+            839,
+            825,
+            924
+        ],
+        "page_idx": 17
+    },
+    {
+        "type": "table",
+        "img_path": "images/51d7959ad0634b2257179895a73cb6b153b95b71e9943386f23a9e1ebc7580a9.jpg",
+        "table_caption": [],
+        "table_footnote": [],
+        "table_body": "<table><tr><td rowspan=1 colspan=1>(0101)</td><td rowspan=4 colspan=1>% LI&#x27;0干06&#x27;16% 01.0581555% 71&#x27;0 干 10&#x27;96% 710 67&#x27;96% II&#x27;0干08&#x27;96% L70干08&#x27;88% 11011.55% 910 76.56% 71&#x27;0干80&#x27;96% 110 干 9596% 780 6444% 18031.95% 7109 8535% 40.0 99.56% 870 LL&#x27;46C[ = y[= T (sani) sSuiPPq Tt Tr-shyt[ = y[ = T (sA) P T T-SAT r rrrrret gaasetr Srrpeirrlrer iaerrertrrnrrltgFaraeregarrgrrirrettA4=P444</td><td rowspan=2 colspan=1>% 0509 86:10% 1£&#x27;0干0L09%50055672% 67089&#x27;89% 0000658% 104530% 01004535% 44061.44% 141 65:14%70015:35%03.0335.55% 0ε0干 S0&#x27;LL% 51081:14%41044:44</td><td rowspan=1 colspan=1>% 700 47.99% 11&#x27;0干 58&#x27;96% 01&#x27;0干 66&#x27;96% 010 干 87&#x27; L6% 600 干13&#x27; L6% 01756% 010  6</td><td rowspan=1 colspan=1>% L0&#x27;0 干 1S&#x27;86</td></tr><tr><td rowspan=2 colspan=1>(s01)</td><td rowspan=2 colspan=1>% LI&#x27;0干 S0&#x27;96% 170干19&#x27;96% 110T 6696% 010干8126% 010776% 010干£1L6% 010干 80&#x27;L6</td><td rowspan=2 colspan=1>% L00干1S&#x27;86</td></tr><tr><td rowspan=1 colspan=1>% 44061.44% 141 65:14%70015:35%03.0335.55% 0ε0干 S0&#x27;LL% 51081:14%41044:44</td></tr><tr><td rowspan=1 colspan=1>(101)(tous em)1oila</td><td rowspan=1 colspan=1>%230325:65% 680 F980%57025224%944144% 14011.64%85033335% 3：271:31[[ = y O = T‘(sIni) suipP-qs u Tr-t[ = y  = (sani) supPqa uu Trn-st[[ = y OII = T (I1I) SSA- V T-NC[ = yI=T sa   -T=grrrraree geasepr rripegahrlerserrerarrnriirenggareregaerrriipetP=P-14</td><td rowspan=1 colspan=1>% 155311:55% 7408126% 31.074.16%050333:5% 610干80&#x27;96% LI0干 6996% L10 干8596C = y[= T(sa) supq c -[[ = y O = T‘(sni) sYuiPPqs uu Tr-sntC = y O = T () P 1 TI-sC=y[= T(I)A T T-sAg rrirpre aaetr rrirearrger serrerarrnriirertggaraerrarrrriiiertt=p 4</td><td rowspan=1 colspan=1>% L1&#x27;0干Iε&#x27;L6Surreend pasiersnnGOLnaier Giii-larei</td></tr></table>",
+        "bbox": [
+            295,
+            239,
+            746,
+            785
+        ],
+        "page_idx": 18
+    },
+    {
+        "type": "table",
+        "img_path": "images/694c397fac710ab1a4f54d056b99bd24fdc34ed1a1fa932f558420a2417f3576.jpg",
+        "table_caption": [
+            "ers per convolutional layer, 3×data augmentation, and folded the validation set ",
+            ""
+        ],
+        "table_footnote": [
+            "denotes a 95% confidence interval. d: dimensionality of embedding, "
+        ],
+        "table_body": "<table><tr><td rowspan=7 colspan=1>(s 0)(107)(s‘s)(1‘s)(ous em)oiiral</td><td rowspan=7 colspan=1>% 944% 810 449% 67&#x27;0干 10&#x27;61% 110013.55% 810 11&#x27;9%1102555% 6908&#x27;44% 690441% 1∠&#x27;06100%20035174% 32056577%33357530PPT =γ[= TIIII-SALAI SI11.1ay = y  = 111A1--4111rres ionnaieraBeesees</td><td rowspan=4 colspan=1>% 90.038399% 33.333339% 350 79:11  %80085.8  % 990 41.69 % 1333 3334% L00干 L7&#x27;996%83033139% 670227/4%3330730 % 30033331% 8109835% 4404 44.14  % 1 1124</td><td rowspan=1 colspan=1>% 90001.95% 2102 78.14% 401141% L£&#x27;0干90% 6£&#x27;0干 LL&#x27;0  %37:227755% 8057788</td><td rowspan=1 colspan=2>% 0105 6796% L0&#x27;0干18&#x27;86</td></tr><tr><td rowspan=1 colspan=1>% 90001.95% 2102 78.14% 401141% L£&#x27;0干90% 6£&#x27;0干 LL&#x27;0  %37:227755% 8057788% 30:0377.3050</td><td rowspan=2 colspan=2>% 0105 6796% L0&#x27;0干18&#x27;86% 700339:55% 810533.55</td></tr><tr><td rowspan=2 colspan=1>%11033135% 190 08:27% 770955.66%71224115% 210370.65% 200586.655210224555%335395559% 33033151</td></tr><tr><td></td><td rowspan=1 colspan=1>土+</td></tr><tr><td rowspan=3 colspan=1>% LS0干9118%8107818%S9007&#x27;LL% LL&#x27;0干60&#x27;IL% 09&#x27;0干18&#x27;18%8703355% 353355.55% 0S&#x27;0 干8L&#x27;L8% 3903 8138% 1113 1511% 711 819% 7.0 16:14  %7.055  % 08&#x27;0 干 64&#x27;99  % 480  71896260 =   =  () T-       CCT = y TT=1 (In) s11T1II-SSCC = γ OII = 1 ‘(S.I1I) T1I1-SALOCY =γ[=T (SIAI) TIII-SAI8 = Y inodop p PT SuippggregreemysppegarierereereirrA44=P4444</td><td rowspan=1 colspan=1></td><td></td></tr><tr><td rowspan=1 colspan=1>% I∠&#x27;0干 90&#x27;89% 99&#x27;0 干7L&#x27;89% 6/0 9579%8/07985% 69&#x27;009&#x27;L% 6905555% 19085.94% S9&#x27;0 干99&#x27;8L% L0干69</td><td rowspan=2 colspan=2>% 710£8&#x27;86% 60&#x27;0干85&#x27;66% 11.03.9555% 773523:32Surureree peseeedn  GDi Drirr eoeiiioriireri（OLDLier Girr-aarei</td></tr><tr><td rowspan=1 colspan=1>% 71 1171% 1742 1185% 6104544% 080 964% 5039.02 % 0807650% 18081&#x27;89% 71:044440 =  O = P (I) T-sdHP&#x27; =   = P(s) -s-L CC =y OII = 1 (SIAN) TSII-SA/S0C = y =  (i) sTod-snt8 = Y inidoip pm d SuippqC =yI =1 (SIA) T-SAagrrpaee easeta Srrplgrhgier neeennr neipreggarerrgrereippegPP =P 4400</td></tr></table>",
+        "bbox": [
+            251,
+            212,
+            750,
+            810
+        ],
+        "page_idx": 19
+    },
+    {
+        "type": "table",
+        "img_path": "images/80d028f66a313fb206276b503e8a1b093d0c96a5856ce2c601050fb245cae4f2.jpg",
+        "table_caption": [],
+        "table_footnote": [
+            "denotes a 95% confidence interval. d: dimensionality of embeddi "
+        ],
+        "table_body": "<table><tr><td rowspan=1 colspan=1>(0s&#x27;s)(2 &#x27;0)(s‘s)(1‘s)(ous Xem)Woirai</td><td rowspan=1 colspan=1>% 5/.05 39.61% 533112% 9908589% 67003735Yrirlrriiir reirraperreeees</td><td rowspan=1 colspan=1>% 230309.3% 31:15:11% 1/.0 91&#x27;81%55333.99% 00&#x27;0干00&#x27;  % 89&#x27;0干 86&#x27;L% 89&#x27;0干 16&#x27;99% L9:0 17:99546%344231.21% 100404% 7/0 9000     %8900L&#x27;79%57003515% 5703.31%£9090:65%81096565% 0000000 %15457171% 850 87.15%05099511% 3513 95:75% 1 8 2 470036.74     %990干8000干（09.4士22.5519&#x27;6725&#x27;330 = w Xs =  (s) Ti-H6&#x27;0 = 1 O = P (s) T--H CC1 = yO1= P (SIA4） PIII-SA/CT1 =yFT= (sT) s1TI-sg = Y iniip pi P grippC =y[=1 (SIA) TI-SAS =yrrrpaee gaisnpa Seippegger nsereny iprgasereresteippeggPP=P4440</td><td rowspan=1 colspan=1>% 8030639%5500854% £9&#x27;0干 56:09% 1555585% 70&#x27;0干I0% 99055% 79&#x27;0LL% £90干 5969% 553515:396566% 1440444.94%5506554  %41:045.44       % 0105809% 89045:1989&#x27;0921% 42:015%0706/:66 277333755 %1/044:64% 71172% 010 1690% 3303.73.51% 17:5--55.85%%   %%119.03 30.67990干  5503066120.00555555830 =  Os = P (sI1I) Tt--AdL1HT&#x27;0 = 1s= P sa) T-H C=yE1=PIA S-ACC = y I= P （(I) sTTd-SHg = Y iniip pi P grippaggC1 =yT=1A） T-SNarrpiereisepaSerpggrhler errrairrireigrrsesereesseippeggz =pashisea4</td><td rowspan=1 colspan=1>% 77:045:14% 09044% 69000.14% 590F 50.00% 7/0511.79% 1/0 6777% LL&#x27;0干 18&#x27;94% 9/0 95.94Surererrer lseeetdnnODTLr) Giii-rriOrnerer eeeinerrceeret</td></tr></table>",
+        "bbox": [
+            289,
+            214,
+            753,
+            813
+        ],
+        "page_idx": 20
+    },
+    {
+        "type": "table",
+        "img_path": "images/660b21b7784e16644ff9108f25cb77fb320be6baea7dd2154f3fcf8d42bd91fb.jpg",
+        "table_caption": [
+            "Table 10: CelebA facial attribute classification results averaged over 1000 tasks. $\\pm$ denotes a $9 5 \\%$ confidence interval. $d$ : dimensionality of embedding, $h$ : number of hidden units in a fully connected layer, $k$ : number of clusters in a partition, $P$ : number of partitions used during meta-learning. "
+        ],
+        "table_footnote": [],
+        "table_body": "<table><tr><td>Algorithm</td><td>(2,5)</td></tr><tr><td>Baselines Training from scratch</td><td>63.19 ± 1.06 %</td></tr><tr><td>BiGAN,d= 200</td><td></td></tr><tr><td>Embedding knn-nearest neighbors Embedding linearclassifier</td><td>56.15 ± 0.89 % 58.44 ± 0.90 %</td></tr><tr><td>Embedding MLP with dropout,h = 128</td><td>56.26 ± 0.94 %</td></tr><tr><td>Embedding cluster matching,k = 500</td><td>56.20 ±1.00 %</td></tr><tr><td>CACTUs-MAML(ours),P= 50,k= 500 CACTUs-ProtoNets (ours),P= 50,k = 500</td><td>74.98 ± 1.02 %</td></tr><tr><td></td><td>65.58 ± 1.04 %</td></tr><tr><td>DeepCluster,d= 256 Embedding knn-nearest neighbors</td><td></td></tr><tr><td>Embedding linear classifier</td><td>61.47 ± 0.99 % 59.57 ± 0.98 %</td></tr><tr><td>Embedding MLP with dropout,h = 128</td><td>60.65 ± 0.98 %</td></tr><tr><td>Embedding cluster matching,k = 500</td><td>51.51 ± 0.89 %</td></tr><tr><td>CACTUs-MAML (ours),P= 50,k = 500</td><td>73.79 ± 1.01 %</td></tr><tr><td>CACTUs-ProtoNets (ours),P= 50,k = 500</td><td>74.15 ± 1.02 %</td></tr><tr><td></td><td></td></tr><tr><td>Supervised meta-learning</td><td></td></tr><tr><td>Oracle-MAML (control)</td><td>87.10 ± 0.85 %</td></tr><tr><td></td><td></td></tr><tr><td>Oracle-ProtoNets (control)</td><td>85.13 ± 0.92 %</td></tr></table>",
+        "bbox": [
+            297,
+            392,
+            699,
+            680
+        ],
+        "page_idx": 21
+    },
+    {
+        "type": "text",
+        "text": "APPENDIX G IMAGENET EXPERIMENTS ",
+        "text_level": 1,
+        "bbox": [
+            176,
+            102,
+            521,
+            118
+        ],
+        "page_idx": 22
+    },
+    {
+        "type": "text",
+        "text": "We investigate unsupervised meta-learning in the context of a larger unsupervised meta-training dataset by using the ILSVRC 2012 dataset’s training split (Russakovsky et al., 2015), which is a superset of the miniImageNet dataset (including meta-validation and meta-testing data) consisting of 1000 classes and over 1,200,000 images. To facilitate comparison to the previous miniImageNet experiments, for meta-validation and meta-test we use the miniImageNet meta-validation and metatest splits. To avoid task leakage, we hold out all data from these 36 underlying classes from the rest of the data to construct the meta-training split. ",
+        "bbox": [
+            174,
+            133,
+            825,
+            231
+        ],
+        "page_idx": 22
+    },
+    {
+        "type": "text",
+        "text": "For CACTUs, we use the best-performing unsupervised learning method from the previous experiments, DeepCluster, to obtain the embeddings. Following Caron et al. (2018), we run DeepCluster using the VGG-16 architecture with a 256-dimensional feature space and 10,000 clusters on the meta-training data until the normalized mutual information between the data-cluster mappings of two consecutive epochs converges. To our knowledge, no prior works have yet been published on using MAML for ImageNet-sized meta-learning. We extend the standard convolutional neural network model class with residual connections (He et al., 2016), validate hyperparameters with supervised meta-learning, then use it for unsupervised meta-learning without further tuning. See Table 11 for MAML hyperparameters. The training from scratch, embedding $k _ { \\mathrm { n n } }$ -nearest neighbors, and embedding linear classifier algorithms are the same as they were in the previous sets of experiments. For Oracle-MAML, we generated tasks using the ground-truth 964 ImageNet meta-training classes. We also run semi-supervised MAML, with the meta-training tasks consisting of CACTUs-based tasks as well as tasks constructed from the 64 miniImageNet meta-training classes. The unsupervised/supervised task proportion split was fixed according to the ratio of the number of data available to each task proposal method. As before, the meta-learning methods only meta-learned on 1-shot tasks. ",
+        "bbox": [
+            173,
+            238,
+            825,
+            459
+        ],
+        "page_idx": 22
+    },
+    {
+        "type": "table",
+        "img_path": "images/080631d0533a0a33810053381fc1eed4a2b60fc845717b42c6eee8481e50188e.jpg",
+        "table_caption": [
+            "Table 11: MAML hyperparameter summary for ImageNet. "
+        ],
+        "table_footnote": [],
+        "table_body": "<table><tr><td>Hyperparameter</td><td>Value</td></tr><tr><td>Input size</td><td>224× 224</td></tr><tr><td>Outer (meta) learning rate</td><td>0.0001</td></tr><tr><td>Inner learning rate</td><td>0.001</td></tr><tr><td>Task batch size</td><td>3</td></tr><tr><td>Inner adaptation steps (meta-training)</td><td>5</td></tr><tr><td>Meta-training iterations</td><td>240,000</td></tr><tr><td>Adaptation steps (evaluation)</td><td>100</td></tr><tr><td>Classes per task (meta-training)</td><td>5</td></tr><tr><td>Shots per class (meta-training)</td><td>1</td></tr><tr><td>Queries per class</td><td>5</td></tr><tr><td>Residual blocks</td><td>5</td></tr><tr><td>Layers per residual block</td><td>2</td></tr></table>",
+        "bbox": [
+            325,
+            492,
+            674,
+            689
+        ],
+        "page_idx": 22
+    },
+    {
+        "type": "text",
+        "text": "We find that the vastly increased amount of unlabeled meta-training data (in comparison to miniImageNet) results in significant increases for all methods over their counterparts in Table 9 (other than training from scratch, which does not use this data). We find that CACTUs-MAML slightly outperforms embedding linear classifier for the 1-shot test tasks, but that the linear classifier on top of the unsupervised embedding becomes better as the amount of test time supervision increases. Augmenting the unsupervised tasks with (a small number of) supervised tasks during meta-training results in slight improvement for the 1-shot test tasks. The lackluster performance of CACTUs-MAML is unsurprising insofar as meta-learning with large task spaces is still an open problem: higher shot Oracle-MAML only marginally stays ahead of the embedding linear classifier, which is not the case in the other, smaller-scale experiments. We expect that using a larger architecture in conjunction with MAML (such as Kim et al. (2018)) would result in increased performance for all methods based on MAML. Further, given the extensive degree to which unsupervised learning methods have been studied, we suspect that unsupervised task construction coupled with better meta-learning algorithms and architectures will result in improved performance on the entire unsupervised learning problem. We leave such investigation to future work. ",
+        "bbox": [
+            173,
+            708,
+            825,
+            916
+        ],
+        "page_idx": 22
+    },
+    {
+        "type": "table",
+        "img_path": "images/246c7ecf053e28afcdae6dee0f1be59ff6fc7f6487369b9082e633dbd7cb3d14.jpg",
+        "table_caption": [
+            "denotes a 95% confidence interval. d: dimensionality of "
+        ],
+        "table_footnote": [],
+        "table_body": "<table><tr><td>(os&#x27;s) （07&#x27;s) (s‘s) (1‘s) (ous xem) 1loirrra</td><td>% LL0干 89&#x27;6S% SL0干 L9&#x27;εS % 3901/30% 11:059:85 Ppt =pessstsesg Yrrlrriiir liiiiera reneseg</td><td>% 41:0 4244 % L70 干 0106 % 3705 75.77 % 040 18.61 % 130 11:11 % 990F 5878 % 17034355 % 00 1064 % 1/0494 % I9&#x27;0 干 LI&#x27;01 % 7703181 % 4101109 [[V[ = γ[=1 ‘(SJ1) TII-SALO0 ohhgrer serrerarg nrippegg garreerrnrilpestg</td><td>% 190  L728 % 690£8&#x27;78 %0/0£494 % SL&#x27;0 干 SL&#x27;19 Sururersritu prseetdeesntts Jra psrsN-</td><td>% 1/0 £7&#x27;98 % L/071:4 Srinerniar peseadn DEDLLrr Tirr-larrt</td><td>% 990 干 90℃6% 69&#x27;0干4116</td></tr></table>",
+        "bbox": [
+            400,
+            199,
+            640,
+            830
+        ],
+        "page_idx": 23
+    }
+]
\ No newline at end of file
diff --git a/parse/train/r1My6sR9tX/r1My6sR9tX_middle.json b/parse/train/r1My6sR9tX/r1My6sR9tX_middle.json
new file mode 100644
index 0000000000000000000000000000000000000000..b82717c775704213995a79875147df72f2fd9b0d
--- /dev/null
+++ b/parse/train/r1My6sR9tX/r1My6sR9tX_middle.json
@@ -0,0 +1,54134 @@
+{
+    "pdf_info": [
+        {
+            "preproc_blocks": [
+                {
+                    "type": "title",
+                    "bbox": [
+                        106,
+                        78,
+                        472,
+                        96
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                78,
+                                474,
+                                99
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        78,
+                                        474,
+                                        99
+                                    ],
+                                    "score": 1.0,
+                                    "content": "UNSUPERVISED LEARNING VIA META-LEARNING",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        }
+                    ],
+                    "index": 0
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        113,
+                        114,
+                        263,
+                        148
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                111,
+                                114,
+                                158,
+                                127
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        111,
+                                        114,
+                                        158,
+                                        127
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Kyle Hsu†",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                111,
+                                125,
+                                202,
+                                138
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        111,
+                                        125,
+                                        202,
+                                        138
+                                    ],
+                                    "score": 1.0,
+                                    "content": "University of Toronto",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                112,
+                                137,
+                                264,
+                                149
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        112,
+                                        137,
+                                        264,
+                                        149
+                                    ],
+                                    "score": 1.0,
+                                    "content": "kyle.hsu@mail.utoronto.ca",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        }
+                    ],
+                    "index": 3
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        281,
+                        115,
+                        490,
+                        148
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                281,
+                                115,
+                                404,
+                                127
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        281,
+                                        115,
+                                        404,
+                                        127
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Sergey Levine, Chelsea Finn",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                281,
+                                125,
+                                420,
+                                138
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        281,
+                                        125,
+                                        420,
+                                        138
+                                    ],
+                                    "score": 1.0,
+                                    "content": "University of California, Berkeley",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                282,
+                                137,
+                                491,
+                                150
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        282,
+                                        137,
+                                        491,
+                                        150
+                                    ],
+                                    "score": 1.0,
+                                    "content": "{svlevine,cbfinn}@eecs.berkeley.edu",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        }
+                    ],
+                    "index": 4
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        278,
+                        177,
+                        333,
+                        189
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                276,
+                                177,
+                                335,
+                                190
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        276,
+                                        177,
+                                        335,
+                                        190
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ABSTRACT",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        }
+                    ],
+                    "index": 7
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        143,
+                        202,
+                        468,
+                        366
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                142,
+                                202,
+                                469,
+                                214
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        142,
+                                        202,
+                                        469,
+                                        214
+                                    ],
+                                    "score": 1.0,
+                                    "content": "A central goal of unsupervised learning is to acquire representations from unla-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                212,
+                                469,
+                                225
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        212,
+                                        469,
+                                        225
+                                    ],
+                                    "score": 1.0,
+                                    "content": "beled data or experience that can be used for more effective learning of down-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                223,
+                                470,
+                                236
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        223,
+                                        470,
+                                        236
+                                    ],
+                                    "score": 1.0,
+                                    "content": "stream tasks from modest amounts of labeled data. Many prior unsupervised",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                235,
+                                470,
+                                247
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        235,
+                                        470,
+                                        247
+                                    ],
+                                    "score": 1.0,
+                                    "content": "learning works aim to do so by developing proxy objectives based on reconstruc-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                246,
+                                469,
+                                258
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        246,
+                                        469,
+                                        258
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tion, disentanglement, prediction, and other metrics. Instead, we develop an unsu-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                257,
+                                470,
+                                269
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        257,
+                                        470,
+                                        269
+                                    ],
+                                    "score": 1.0,
+                                    "content": "pervised meta-learning method that explicitly optimizes for the ability to learn a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                142,
+                                268,
+                                470,
+                                279
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        142,
+                                        268,
+                                        470,
+                                        279
+                                    ],
+                                    "score": 1.0,
+                                    "content": "variety of tasks from small amounts of data. To do so, we construct tasks from un-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                278,
+                                470,
+                                291
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        278,
+                                        470,
+                                        291
+                                    ],
+                                    "score": 1.0,
+                                    "content": "labeled data in an automatic way and run meta-learning over the constructed tasks.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                290,
+                                470,
+                                302
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        290,
+                                        470,
+                                        302
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Surprisingly, we find that, when integrated with meta-learning, relatively simple",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                300,
+                                470,
+                                313
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        300,
+                                        470,
+                                        313
+                                    ],
+                                    "score": 1.0,
+                                    "content": "task construction mechanisms, such as clustering embeddings, lead to good per-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                142,
+                                312,
+                                469,
+                                323
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        142,
+                                        312,
+                                        469,
+                                        323
+                                    ],
+                                    "score": 1.0,
+                                    "content": "formance on a variety of downstream, human-specified tasks. Our experiments",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                322,
+                                470,
+                                335
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        322,
+                                        470,
+                                        335
+                                    ],
+                                    "score": 1.0,
+                                    "content": "across four image datasets indicate that our unsupervised meta-learning approach",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                333,
+                                470,
+                                345
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        333,
+                                        470,
+                                        345
+                                    ],
+                                    "score": 1.0,
+                                    "content": "acquires a learning algorithm without any labeled data that is applicable to a wide",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                344,
+                                470,
+                                357
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        344,
+                                        470,
+                                        357
+                                    ],
+                                    "score": 1.0,
+                                    "content": "range of downstream classification tasks, improving upon the embedding learned",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                355,
+                                325,
+                                368
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        355,
+                                        325,
+                                        368
+                                    ],
+                                    "score": 1.0,
+                                    "content": "by four prior unsupervised learning methods.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        }
+                    ],
+                    "index": 15
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        388,
+                        206,
+                        401
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                387,
+                                208,
+                                403
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        387,
+                                        208,
+                                        403
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1 INTRODUCTION",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        }
+                    ],
+                    "index": 23
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        410,
+                        505,
+                        509
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                409,
+                                505,
+                                423
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        409,
+                                        505,
+                                        423
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Unsupervised learning is a fundamental, unsolved problem (Hastie et al., 2009) and has seen promis-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                421,
+                                505,
+                                434
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        421,
+                                        505,
+                                        434
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ing results in domains such as image recognition (Le et al., 2013) and natural language understand-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                431,
+                                506,
+                                446
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        431,
+                                        506,
+                                        446
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ing (Ramachandran et al., 2017). A central use case of unsupervised learning methods is enabling",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                443,
+                                505,
+                                456
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        443,
+                                        505,
+                                        456
+                                    ],
+                                    "score": 1.0,
+                                    "content": "better or more efficient learning of downstream tasks by training on top of unsupervised representa-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                454,
+                                505,
+                                467
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        454,
+                                        505,
+                                        467
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tions (Reed et al., 2014; Cheung et al., 2015; Chen et al., 2016) or fine-tuning a learned model (Erhan",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                465,
+                                505,
+                                478
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        465,
+                                        505,
+                                        478
+                                    ],
+                                    "score": 1.0,
+                                    "content": "et al., 2010). However, since the downstream objective requires access to supervision, the objectives",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                476,
+                                505,
+                                489
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        476,
+                                        505,
+                                        489
+                                    ],
+                                    "score": 1.0,
+                                    "content": "used for unsupervised learning are only a rough proxy for downstream performance. If a central goal",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                485,
+                                505,
+                                501
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        485,
+                                        505,
+                                        501
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of unsupervised learning is to learn useful representations, can we derive an unsupervised learning",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                498,
+                                425,
+                                511
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        498,
+                                        425,
+                                        511
+                                    ],
+                                    "score": 1.0,
+                                    "content": "objective that explicitly takes into account how the representation will be used?",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        }
+                    ],
+                    "index": 28
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        515,
+                        505,
+                        569
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                515,
+                                505,
+                                527
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        515,
+                                        505,
+                                        527
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The use of unsupervised representations for downstream tasks is closely related to the objective",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                525,
+                                505,
+                                538
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        525,
+                                        505,
+                                        538
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of meta-learning techniques: finding a learning procedure that is more efficient and effective than",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                537,
+                                505,
+                                550
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        537,
+                                        505,
+                                        550
+                                    ],
+                                    "score": 1.0,
+                                    "content": "learning from scratch. However, unlike unsupervised learning methods, meta-learning methods",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                546,
+                                506,
+                                561
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        546,
+                                        506,
+                                        561
+                                    ],
+                                    "score": 1.0,
+                                    "content": "require large, labeled datasets and hand-specified task distributions. These dependencies are major",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                559,
+                                397,
+                                571
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        559,
+                                        397,
+                                        571
+                                    ],
+                                    "score": 1.0,
+                                    "content": "obstacles to widespread use of these methods for few-shot classification.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        }
+                    ],
+                    "index": 35
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        576,
+                        505,
+                        652
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                574,
+                                506,
+                                590
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        574,
+                                        506,
+                                        590
+                                    ],
+                                    "score": 1.0,
+                                    "content": "To begin addressing these problems, we propose an unsupervised meta-learning method: one which",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                586,
+                                505,
+                                599
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        586,
+                                        505,
+                                        599
+                                    ],
+                                    "score": 1.0,
+                                    "content": "aims to learn a learning procedure, without supervision, that is useful for solving a wide range of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                597,
+                                506,
+                                610
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        597,
+                                        506,
+                                        610
+                                    ],
+                                    "score": 1.0,
+                                    "content": "new, human-specified tasks. With only raw, unlabeled observations, our model’s goal is to learn",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                609,
+                                505,
+                                620
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        609,
+                                        505,
+                                        620
+                                    ],
+                                    "score": 1.0,
+                                    "content": "a useful prior such that, after meta-training, when presented with a modestly-sized dataset for a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                619,
+                                505,
+                                632
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        619,
+                                        505,
+                                        632
+                                    ],
+                                    "score": 1.0,
+                                    "content": "human-specified task, the model can transfer its prior experience to efficiently learn to perform the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                631,
+                                505,
+                                642
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        631,
+                                        505,
+                                        642
+                                    ],
+                                    "score": 1.0,
+                                    "content": "new task. If we can build such an algorithm, we can enable few-shot learning of new tasks without",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                642,
+                                313,
+                                653
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        642,
+                                        313,
+                                        653
+                                    ],
+                                    "score": 1.0,
+                                    "content": "needing any labeled data nor any pre-defined tasks.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        }
+                    ],
+                    "index": 41
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        658,
+                        504,
+                        714
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                658,
+                                506,
+                                670
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        658,
+                                        506,
+                                        670
+                                    ],
+                                    "score": 1.0,
+                                    "content": "To perform unsupervised meta-learning, we need to automatically construct tasks from unlabeled",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                668,
+                                505,
+                                682
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        668,
+                                        505,
+                                        682
+                                    ],
+                                    "score": 1.0,
+                                    "content": "data. We study several options for how this can be done. We find that a good task distribution",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                679,
+                                506,
+                                693
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        679,
+                                        506,
+                                        693
+                                    ],
+                                    "score": 1.0,
+                                    "content": "should be diverse, but also not too difficult: na¨ıve random approaches for task generation produce",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 47
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                689,
+                                506,
+                                705
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        689,
+                                        506,
+                                        705
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tasks that contain insufficient regularity to enable useful meta-learning. To that end, our method pro-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 48
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                702,
+                                506,
+                                714
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        702,
+                                        506,
+                                        714
+                                    ],
+                                    "score": 1.0,
+                                    "content": "poses tasks by first leveraging prior unsupervised learning algorithms to learn an embedding of the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 49
+                        }
+                    ],
+                    "index": 47
+                }
+            ],
+            "page_idx": 0,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        118,
+                        721,
+                        424,
+                        731
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                118,
+                                718,
+                                425,
+                                735
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        118,
+                                        718,
+                                        425,
+                                        735
+                                    ],
+                                    "score": 1.0,
+                                    "content": "†Work done as a visiting student researcher at the University of California, Berkeley.",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        107,
+                        27,
+                        293,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                26,
+                                294,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        26,
+                                        294,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Published as a conference paper at ICLR 2019",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        303,
+                        752,
+                        308,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                302,
+                                751,
+                                309,
+                                762
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        302,
+                                        751,
+                                        309,
+                                        762
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "title",
+                    "bbox": [
+                        106,
+                        78,
+                        472,
+                        96
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                78,
+                                474,
+                                99
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        78,
+                                        474,
+                                        99
+                                    ],
+                                    "score": 1.0,
+                                    "content": "UNSUPERVISED LEARNING VIA META-LEARNING",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        }
+                    ],
+                    "index": 0
+                },
+                {
+                    "type": "list",
+                    "bbox": [
+                        113,
+                        114,
+                        263,
+                        148
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                111,
+                                114,
+                                158,
+                                127
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        111,
+                                        114,
+                                        158,
+                                        127
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Kyle Hsu†",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                111,
+                                125,
+                                202,
+                                138
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        111,
+                                        125,
+                                        202,
+                                        138
+                                    ],
+                                    "score": 1.0,
+                                    "content": "University of Toronto",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                112,
+                                137,
+                                264,
+                                149
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        112,
+                                        137,
+                                        264,
+                                        149
+                                    ],
+                                    "score": 1.0,
+                                    "content": "kyle.hsu@mail.utoronto.ca",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5,
+                            "is_list_start_line": true
+                        }
+                    ],
+                    "index": 3,
+                    "bbox_fs": [
+                        111,
+                        114,
+                        264,
+                        149
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        281,
+                        115,
+                        490,
+                        148
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                281,
+                                115,
+                                404,
+                                127
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        281,
+                                        115,
+                                        404,
+                                        127
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Sergey Levine, Chelsea Finn",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                281,
+                                125,
+                                420,
+                                138
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        281,
+                                        125,
+                                        420,
+                                        138
+                                    ],
+                                    "score": 1.0,
+                                    "content": "University of California, Berkeley",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                282,
+                                137,
+                                491,
+                                150
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        282,
+                                        137,
+                                        491,
+                                        150
+                                    ],
+                                    "score": 1.0,
+                                    "content": "{svlevine,cbfinn}@eecs.berkeley.edu",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        }
+                    ],
+                    "index": 4,
+                    "bbox_fs": [
+                        281,
+                        115,
+                        491,
+                        150
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        278,
+                        177,
+                        333,
+                        189
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                276,
+                                177,
+                                335,
+                                190
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        276,
+                                        177,
+                                        335,
+                                        190
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ABSTRACT",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        }
+                    ],
+                    "index": 7
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        143,
+                        202,
+                        468,
+                        366
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                142,
+                                202,
+                                469,
+                                214
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        142,
+                                        202,
+                                        469,
+                                        214
+                                    ],
+                                    "score": 1.0,
+                                    "content": "A central goal of unsupervised learning is to acquire representations from unla-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                212,
+                                469,
+                                225
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        212,
+                                        469,
+                                        225
+                                    ],
+                                    "score": 1.0,
+                                    "content": "beled data or experience that can be used for more effective learning of down-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                223,
+                                470,
+                                236
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        223,
+                                        470,
+                                        236
+                                    ],
+                                    "score": 1.0,
+                                    "content": "stream tasks from modest amounts of labeled data. Many prior unsupervised",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                235,
+                                470,
+                                247
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        235,
+                                        470,
+                                        247
+                                    ],
+                                    "score": 1.0,
+                                    "content": "learning works aim to do so by developing proxy objectives based on reconstruc-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                246,
+                                469,
+                                258
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        246,
+                                        469,
+                                        258
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tion, disentanglement, prediction, and other metrics. Instead, we develop an unsu-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                257,
+                                470,
+                                269
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        257,
+                                        470,
+                                        269
+                                    ],
+                                    "score": 1.0,
+                                    "content": "pervised meta-learning method that explicitly optimizes for the ability to learn a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                142,
+                                268,
+                                470,
+                                279
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        142,
+                                        268,
+                                        470,
+                                        279
+                                    ],
+                                    "score": 1.0,
+                                    "content": "variety of tasks from small amounts of data. To do so, we construct tasks from un-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                278,
+                                470,
+                                291
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        278,
+                                        470,
+                                        291
+                                    ],
+                                    "score": 1.0,
+                                    "content": "labeled data in an automatic way and run meta-learning over the constructed tasks.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                290,
+                                470,
+                                302
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        290,
+                                        470,
+                                        302
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Surprisingly, we find that, when integrated with meta-learning, relatively simple",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                300,
+                                470,
+                                313
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        300,
+                                        470,
+                                        313
+                                    ],
+                                    "score": 1.0,
+                                    "content": "task construction mechanisms, such as clustering embeddings, lead to good per-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                142,
+                                312,
+                                469,
+                                323
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        142,
+                                        312,
+                                        469,
+                                        323
+                                    ],
+                                    "score": 1.0,
+                                    "content": "formance on a variety of downstream, human-specified tasks. Our experiments",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                322,
+                                470,
+                                335
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        322,
+                                        470,
+                                        335
+                                    ],
+                                    "score": 1.0,
+                                    "content": "across four image datasets indicate that our unsupervised meta-learning approach",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                333,
+                                470,
+                                345
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        333,
+                                        470,
+                                        345
+                                    ],
+                                    "score": 1.0,
+                                    "content": "acquires a learning algorithm without any labeled data that is applicable to a wide",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                344,
+                                470,
+                                357
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        344,
+                                        470,
+                                        357
+                                    ],
+                                    "score": 1.0,
+                                    "content": "range of downstream classification tasks, improving upon the embedding learned",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                355,
+                                325,
+                                368
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        355,
+                                        325,
+                                        368
+                                    ],
+                                    "score": 1.0,
+                                    "content": "by four prior unsupervised learning methods.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        }
+                    ],
+                    "index": 15,
+                    "bbox_fs": [
+                        141,
+                        202,
+                        470,
+                        368
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        388,
+                        206,
+                        401
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                387,
+                                208,
+                                403
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        387,
+                                        208,
+                                        403
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1 INTRODUCTION",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        }
+                    ],
+                    "index": 23
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        410,
+                        505,
+                        509
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                409,
+                                505,
+                                423
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        409,
+                                        505,
+                                        423
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Unsupervised learning is a fundamental, unsolved problem (Hastie et al., 2009) and has seen promis-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                421,
+                                505,
+                                434
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        421,
+                                        505,
+                                        434
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ing results in domains such as image recognition (Le et al., 2013) and natural language understand-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                431,
+                                506,
+                                446
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        431,
+                                        506,
+                                        446
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ing (Ramachandran et al., 2017). A central use case of unsupervised learning methods is enabling",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                443,
+                                505,
+                                456
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        443,
+                                        505,
+                                        456
+                                    ],
+                                    "score": 1.0,
+                                    "content": "better or more efficient learning of downstream tasks by training on top of unsupervised representa-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                454,
+                                505,
+                                467
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        454,
+                                        505,
+                                        467
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tions (Reed et al., 2014; Cheung et al., 2015; Chen et al., 2016) or fine-tuning a learned model (Erhan",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                465,
+                                505,
+                                478
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        465,
+                                        505,
+                                        478
+                                    ],
+                                    "score": 1.0,
+                                    "content": "et al., 2010). However, since the downstream objective requires access to supervision, the objectives",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                476,
+                                505,
+                                489
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        476,
+                                        505,
+                                        489
+                                    ],
+                                    "score": 1.0,
+                                    "content": "used for unsupervised learning are only a rough proxy for downstream performance. If a central goal",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                485,
+                                505,
+                                501
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        485,
+                                        505,
+                                        501
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of unsupervised learning is to learn useful representations, can we derive an unsupervised learning",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                498,
+                                425,
+                                511
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        498,
+                                        425,
+                                        511
+                                    ],
+                                    "score": 1.0,
+                                    "content": "objective that explicitly takes into account how the representation will be used?",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        }
+                    ],
+                    "index": 28,
+                    "bbox_fs": [
+                        104,
+                        409,
+                        506,
+                        511
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        515,
+                        505,
+                        569
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                515,
+                                505,
+                                527
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        515,
+                                        505,
+                                        527
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The use of unsupervised representations for downstream tasks is closely related to the objective",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                525,
+                                505,
+                                538
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        525,
+                                        505,
+                                        538
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of meta-learning techniques: finding a learning procedure that is more efficient and effective than",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                537,
+                                505,
+                                550
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        537,
+                                        505,
+                                        550
+                                    ],
+                                    "score": 1.0,
+                                    "content": "learning from scratch. However, unlike unsupervised learning methods, meta-learning methods",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                546,
+                                506,
+                                561
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        546,
+                                        506,
+                                        561
+                                    ],
+                                    "score": 1.0,
+                                    "content": "require large, labeled datasets and hand-specified task distributions. These dependencies are major",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                559,
+                                397,
+                                571
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        559,
+                                        397,
+                                        571
+                                    ],
+                                    "score": 1.0,
+                                    "content": "obstacles to widespread use of these methods for few-shot classification.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        }
+                    ],
+                    "index": 35,
+                    "bbox_fs": [
+                        104,
+                        515,
+                        506,
+                        571
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        576,
+                        505,
+                        652
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                574,
+                                506,
+                                590
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        574,
+                                        506,
+                                        590
+                                    ],
+                                    "score": 1.0,
+                                    "content": "To begin addressing these problems, we propose an unsupervised meta-learning method: one which",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                586,
+                                505,
+                                599
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        586,
+                                        505,
+                                        599
+                                    ],
+                                    "score": 1.0,
+                                    "content": "aims to learn a learning procedure, without supervision, that is useful for solving a wide range of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                597,
+                                506,
+                                610
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        597,
+                                        506,
+                                        610
+                                    ],
+                                    "score": 1.0,
+                                    "content": "new, human-specified tasks. With only raw, unlabeled observations, our model’s goal is to learn",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                609,
+                                505,
+                                620
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        609,
+                                        505,
+                                        620
+                                    ],
+                                    "score": 1.0,
+                                    "content": "a useful prior such that, after meta-training, when presented with a modestly-sized dataset for a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                619,
+                                505,
+                                632
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        619,
+                                        505,
+                                        632
+                                    ],
+                                    "score": 1.0,
+                                    "content": "human-specified task, the model can transfer its prior experience to efficiently learn to perform the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                631,
+                                505,
+                                642
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        631,
+                                        505,
+                                        642
+                                    ],
+                                    "score": 1.0,
+                                    "content": "new task. If we can build such an algorithm, we can enable few-shot learning of new tasks without",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                642,
+                                313,
+                                653
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        642,
+                                        313,
+                                        653
+                                    ],
+                                    "score": 1.0,
+                                    "content": "needing any labeled data nor any pre-defined tasks.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        }
+                    ],
+                    "index": 41,
+                    "bbox_fs": [
+                        105,
+                        574,
+                        506,
+                        653
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        658,
+                        504,
+                        714
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                658,
+                                506,
+                                670
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        658,
+                                        506,
+                                        670
+                                    ],
+                                    "score": 1.0,
+                                    "content": "To perform unsupervised meta-learning, we need to automatically construct tasks from unlabeled",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                668,
+                                505,
+                                682
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        668,
+                                        505,
+                                        682
+                                    ],
+                                    "score": 1.0,
+                                    "content": "data. We study several options for how this can be done. We find that a good task distribution",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                679,
+                                506,
+                                693
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        679,
+                                        506,
+                                        693
+                                    ],
+                                    "score": 1.0,
+                                    "content": "should be diverse, but also not too difficult: na¨ıve random approaches for task generation produce",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 47
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                689,
+                                506,
+                                705
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        689,
+                                        506,
+                                        705
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tasks that contain insufficient regularity to enable useful meta-learning. To that end, our method pro-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 48
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                702,
+                                506,
+                                714
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        702,
+                                        506,
+                                        714
+                                    ],
+                                    "score": 1.0,
+                                    "content": "poses tasks by first leveraging prior unsupervised learning algorithms to learn an embedding of the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 49
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                82,
+                                506,
+                                96
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        82,
+                                        506,
+                                        96
+                                    ],
+                                    "score": 1.0,
+                                    "content": "input data, and then performing an overcomplete partitioning of the dataset to construct numerous",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                93,
+                                506,
+                                107
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        93,
+                                        506,
+                                        107
+                                    ],
+                                    "score": 1.0,
+                                    "content": "categorizations of the data. We show how we can derive classification tasks from these catego-",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                105,
+                                505,
+                                117
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        105,
+                                        505,
+                                        117
+                                    ],
+                                    "score": 1.0,
+                                    "content": "rizations for use with meta-learning algorithms. Surprisingly, even with simple mechanisms for",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                115,
+                                506,
+                                128
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        115,
+                                        278,
+                                        128
+                                    ],
+                                    "score": 1.0,
+                                    "content": "partitioning the embedding space, such as",
+                                    "type": "text",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        278,
+                                        116,
+                                        285,
+                                        125
+                                    ],
+                                    "score": 0.82,
+                                    "content": "k",
+                                    "type": "inline_equation",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        285,
+                                        115,
+                                        506,
+                                        128
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-means clustering, we find that meta-learning acquires",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                126,
+                                505,
+                                139
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        126,
+                                        505,
+                                        139
+                                    ],
+                                    "score": 1.0,
+                                    "content": "priors that, when used to learn new, human-designed tasks, learn those tasks more effectively than",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                137,
+                                505,
+                                151
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        137,
+                                        505,
+                                        151
+                                    ],
+                                    "score": 1.0,
+                                    "content": "methods that directly learn on the embedding. That is, the learning algorithm acquired through un-",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                148,
+                                505,
+                                162
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        148,
+                                        505,
+                                        162
+                                    ],
+                                    "score": 1.0,
+                                    "content": "supervised meta-learning achieves better downstream performance than the original representation",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                159,
+                                505,
+                                172
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        159,
+                                        505,
+                                        172
+                                    ],
+                                    "score": 1.0,
+                                    "content": "used to derive meta-training tasks, without introducing any additional assumptions or supervision.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                171,
+                                336,
+                                183
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        171,
+                                        336,
+                                        183
+                                    ],
+                                    "score": 1.0,
+                                    "content": "See Figure 1 for an illustration of the complete approach.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 8
+                        }
+                    ],
+                    "index": 47,
+                    "bbox_fs": [
+                        105,
+                        658,
+                        506,
+                        714
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        82,
+                        504,
+                        182
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                82,
+                                506,
+                                96
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        82,
+                                        506,
+                                        96
+                                    ],
+                                    "score": 1.0,
+                                    "content": "input data, and then performing an overcomplete partitioning of the dataset to construct numerous",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                93,
+                                506,
+                                107
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        93,
+                                        506,
+                                        107
+                                    ],
+                                    "score": 1.0,
+                                    "content": "categorizations of the data. We show how we can derive classification tasks from these catego-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                105,
+                                505,
+                                117
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        105,
+                                        505,
+                                        117
+                                    ],
+                                    "score": 1.0,
+                                    "content": "rizations for use with meta-learning algorithms. Surprisingly, even with simple mechanisms for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                115,
+                                506,
+                                128
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        115,
+                                        278,
+                                        128
+                                    ],
+                                    "score": 1.0,
+                                    "content": "partitioning the embedding space, such as",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        278,
+                                        116,
+                                        285,
+                                        125
+                                    ],
+                                    "score": 0.82,
+                                    "content": "k",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        285,
+                                        115,
+                                        506,
+                                        128
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-means clustering, we find that meta-learning acquires",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                126,
+                                505,
+                                139
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        126,
+                                        505,
+                                        139
+                                    ],
+                                    "score": 1.0,
+                                    "content": "priors that, when used to learn new, human-designed tasks, learn those tasks more effectively than",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                137,
+                                505,
+                                151
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        137,
+                                        505,
+                                        151
+                                    ],
+                                    "score": 1.0,
+                                    "content": "methods that directly learn on the embedding. That is, the learning algorithm acquired through un-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                148,
+                                505,
+                                162
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        148,
+                                        505,
+                                        162
+                                    ],
+                                    "score": 1.0,
+                                    "content": "supervised meta-learning achieves better downstream performance than the original representation",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                159,
+                                505,
+                                172
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        159,
+                                        505,
+                                        172
+                                    ],
+                                    "score": 1.0,
+                                    "content": "used to derive meta-training tasks, without introducing any additional assumptions or supervision.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                171,
+                                336,
+                                183
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        171,
+                                        336,
+                                        183
+                                    ],
+                                    "score": 1.0,
+                                    "content": "See Figure 1 for an illustration of the complete approach.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        }
+                    ],
+                    "index": 4
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        187,
+                        505,
+                        352
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                187,
+                                505,
+                                200
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        187,
+                                        505,
+                                        200
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The core idea in this paper is that we can leverage unsupervised embeddings to propose tasks for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                198,
+                                505,
+                                212
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        198,
+                                        505,
+                                        212
+                                    ],
+                                    "score": 1.0,
+                                    "content": "a meta-learning algorithm, leading to an unsupervised meta-learning algorithm that is particularly",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                209,
+                                505,
+                                222
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        209,
+                                        505,
+                                        222
+                                    ],
+                                    "score": 1.0,
+                                    "content": "effective as pre-training for human-specified downstream tasks. In the following sections, we for-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                220,
+                                505,
+                                232
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        220,
+                                        505,
+                                        232
+                                    ],
+                                    "score": 1.0,
+                                    "content": "malize our problem assumptions and goal, which match those of unsupervised learning, and discuss",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                230,
+                                505,
+                                243
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        230,
+                                        505,
+                                        243
+                                    ],
+                                    "score": 1.0,
+                                    "content": "several options for automatically deriving tasks from embeddings. We instantiate our method with",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                242,
+                                505,
+                                254
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        242,
+                                        505,
+                                        254
+                                    ],
+                                    "score": 1.0,
+                                    "content": "two meta-learning algorithms and compare to prior state-of-the-art unsupervised learning meth-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                252,
+                                505,
+                                266
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        252,
+                                        505,
+                                        266
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ods. Across four image datasets (MNIST, Omniglot, miniImageNet, and CelebA), we find that our",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                263,
+                                505,
+                                277
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        263,
+                                        505,
+                                        277
+                                    ],
+                                    "score": 1.0,
+                                    "content": "method consistently leads to effective downstream learning of a variety of human-specified tasks,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                274,
+                                505,
+                                288
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        274,
+                                        505,
+                                        288
+                                    ],
+                                    "score": 1.0,
+                                    "content": "including character recognition tasks, object classification tasks, and facial attribute discrimination",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                286,
+                                505,
+                                299
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        286,
+                                        505,
+                                        299
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tasks, without requiring any labels or hand-designed tasks during meta-learning and where key hy-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                296,
+                                505,
+                                310
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        296,
+                                        505,
+                                        310
+                                    ],
+                                    "score": 1.0,
+                                    "content": "perparameters of our method are held constant across all domains. We show that, even though our",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                307,
+                                505,
+                                321
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        307,
+                                        505,
+                                        321
+                                    ],
+                                    "score": 1.0,
+                                    "content": "unsupervised meta-learning algorithm trains for one-shot generalization, one instantiation of our ap-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                318,
+                                505,
+                                331
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        318,
+                                        505,
+                                        331
+                                    ],
+                                    "score": 1.0,
+                                    "content": "proach performs well not only on few-shot learning, but also when learning downstream tasks with",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                329,
+                                506,
+                                343
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        329,
+                                        506,
+                                        343
+                                    ],
+                                    "score": 1.0,
+                                    "content": "up to 50 training examples per class. In fact, some of our results begin to approach the performance",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                339,
+                                471,
+                                354
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        339,
+                                        471,
+                                        354
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of fully-supervised meta-learning techniques trained with fully-specified task distributions.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        }
+                    ],
+                    "index": 16
+                },
+                {
+                    "type": "image",
+                    "bbox": [
+                        118,
+                        366,
+                        488,
+                        469
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                118,
+                                366,
+                                488,
+                                469
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        118,
+                                        366,
+                                        488,
+                                        469
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                118,
+                                                366,
+                                                488,
+                                                469
+                                            ],
+                                            "score": 0.949,
+                                            "type": "image",
+                                            "image_path": "2aa42c4a43d437245b1ba217e8b16d1cdbed8dd1add1db4a9609cd8a7a78f543.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 25,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        118,
+                                        366,
+                                        488,
+                                        400.3333333333333
+                                    ],
+                                    "spans": [],
+                                    "index": 24
+                                },
+                                {
+                                    "bbox": [
+                                        118,
+                                        400.3333333333333,
+                                        488,
+                                        434.66666666666663
+                                    ],
+                                    "spans": [],
+                                    "index": 25
+                                },
+                                {
+                                    "bbox": [
+                                        118,
+                                        434.66666666666663,
+                                        488,
+                                        468.99999999999994
+                                    ],
+                                    "spans": [],
+                                    "index": 26
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                106,
+                                474,
+                                505,
+                                515
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        475,
+                                        505,
+                                        486
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                107,
+                                                475,
+                                                505,
+                                                486
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 1: Illustration of the proposed unsupervised meta-learning procedure. Embeddings of raw observations",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 27
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        484,
+                                        505,
+                                        496
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                484,
+                                                174,
+                                                496
+                                            ],
+                                            "score": 1.0,
+                                            "content": "are clustered with",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                174,
+                                                485,
+                                                180,
+                                                493
+                                            ],
+                                            "score": 0.82,
+                                            "content": "k",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                181,
+                                                484,
+                                                505,
+                                                496
+                                            ],
+                                            "score": 1.0,
+                                            "content": "-means to construct partitions, which give rise to classification tasks. Each task involves",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 28
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        493,
+                                        506,
+                                        507
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                493,
+                                                250,
+                                                507
+                                            ],
+                                            "score": 1.0,
+                                            "content": "distinguishing between examples from",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                251,
+                                                495,
+                                                280,
+                                                504
+                                            ],
+                                            "score": 0.9,
+                                            "content": "N = 2",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                281,
+                                                493,
+                                                334,
+                                                507
+                                            ],
+                                            "score": 1.0,
+                                            "content": "clusters, with",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                334,
+                                                495,
+                                                374,
+                                                505
+                                            ],
+                                            "score": 0.92,
+                                            "content": "K _ { \\mathrm { m - t r } } = 1",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                374,
+                                                493,
+                                                506,
+                                                507
+                                            ],
+                                            "score": 1.0,
+                                            "content": "example from each cluster being a",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 29
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        505,
+                                        499,
+                                        515
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                505,
+                                                499,
+                                                515
+                                            ],
+                                            "score": 1.0,
+                                            "content": "training input. The meta-learner’s aim is to produce a learning procedure that successfully solves these tasks.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 30
+                                }
+                            ],
+                            "index": 28.5
+                        }
+                    ],
+                    "index": 26.75
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        542,
+                        299,
+                        554
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                540,
+                                300,
+                                557
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        540,
+                                        300,
+                                        557
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2 UNSUPERVISED META-LEARNING",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        }
+                    ],
+                    "index": 31
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        104,
+                        566,
+                        503,
+                        588
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                566,
+                                504,
+                                578
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        566,
+                                        504,
+                                        578
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In this section, we describe our problem setting in relation to that of unsupervised and semi-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                577,
+                                426,
+                                590
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        577,
+                                        426,
+                                        590
+                                    ],
+                                    "score": 1.0,
+                                    "content": "supervised learning, provide necessary preliminaries, and present our approach.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        }
+                    ],
+                    "index": 32.5
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        603,
+                        227,
+                        613
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                602,
+                                228,
+                                615
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        602,
+                                        228,
+                                        615
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2.1 PROBLEM STATEMENT",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        }
+                    ],
+                    "index": 34
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        622,
+                        505,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                621,
+                                504,
+                                634
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        621,
+                                        504,
+                                        634
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Our goal is to leverage unlabeled data for the efficient learning of a range of human-specified down-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                632,
+                                505,
+                                646
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        632,
+                                        350,
+                                        646
+                                    ],
+                                    "score": 1.0,
+                                    "content": "stream tasks. We only assume access to an unlabeled dataset",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        351,
+                                        633,
+                                        392,
+                                        645
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\mathcal { D } = \\{ \\mathbf { x } _ { i } \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        393,
+                                        632,
+                                        505,
+                                        646
+                                    ],
+                                    "score": 1.0,
+                                    "content": "during meta-training. After",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                643,
+                                505,
+                                657
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        643,
+                                        505,
+                                        657
+                                    ],
+                                    "score": 1.0,
+                                    "content": "learning from the unlabeled data, which we will refer to as unsupervised meta-training, we want",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                655,
+                                505,
+                                668
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        655,
+                                        505,
+                                        668
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to apply what was learned towards learning a variety of downstream, human-specified tasks from",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                666,
+                                505,
+                                678
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        666,
+                                        505,
+                                        678
+                                    ],
+                                    "score": 1.0,
+                                    "content": "a modest amount of labeled data, potentially as few as a single example per class. These down-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                678,
+                                505,
+                                689
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        678,
+                                        505,
+                                        689
+                                    ],
+                                    "score": 1.0,
+                                    "content": "stream tasks may, in general, have different underlying classes or attributes (in contrast to typical",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                687,
+                                506,
+                                700
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        687,
+                                        506,
+                                        700
+                                    ],
+                                    "score": 1.0,
+                                    "content": "semi-supervised problem assumptions), but are assumed to have inputs from the same distribution",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                699,
+                                505,
+                                711
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        699,
+                                        257,
+                                        711
+                                    ],
+                                    "score": 1.0,
+                                    "content": "as the one from which datapoints in",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        257,
+                                        700,
+                                        266,
+                                        708
+                                    ],
+                                    "score": 0.83,
+                                    "content": "\\mathcal { D }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        266,
+                                        699,
+                                        505,
+                                        711
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are drawn. Concretely, we assume that downstream tasks",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                709,
+                                505,
+                                721
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        709,
+                                        122,
+                                        721
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        122,
+                                        710,
+                                        134,
+                                        720
+                                    ],
+                                    "score": 0.87,
+                                    "content": "M",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        134,
+                                        709,
+                                        461,
+                                        721
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-way classification tasks, and that the goal is to learn an accurate classifier using",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        461,
+                                        710,
+                                        472,
+                                        720
+                                    ],
+                                    "score": 0.79,
+                                    "content": "K",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        473,
+                                        709,
+                                        505,
+                                        721
+                                    ],
+                                    "score": 1.0,
+                                    "content": "labeled",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                720,
+                                505,
+                                733
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        720,
+                                        150,
+                                        733
+                                    ],
+                                    "score": 1.0,
+                                    "content": "datapoints",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        150,
+                                        721,
+                                        185,
+                                        732
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\left( \\mathbf { x } _ { k } , \\mathbf { y } _ { k } \\right)",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        185,
+                                        720,
+                                        253,
+                                        733
+                                    ],
+                                    "score": 1.0,
+                                    "content": "from each of the",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        253,
+                                        721,
+                                        265,
+                                        730
+                                    ],
+                                    "score": 0.8,
+                                    "content": "M",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        266,
+                                        720,
+                                        325,
+                                        733
+                                    ],
+                                    "score": 1.0,
+                                    "content": "classes, where",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        326,
+                                        723,
+                                        335,
+                                        730
+                                    ],
+                                    "score": 0.87,
+                                    "content": "K",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        336,
+                                        720,
+                                        505,
+                                        733
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is relatively small (i.e. between 1 and 50).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        }
+                    ],
+                    "index": 39.5
+                }
+            ],
+            "page_idx": 1,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        108,
+                        27,
+                        293,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                26,
+                                294,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        26,
+                                        294,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Published as a conference paper at ICLR 2019",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        302,
+                        751,
+                        309,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                301,
+                                750,
+                                310,
+                                763
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        301,
+                                        750,
+                                        310,
+                                        763
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        82,
+                        504,
+                        182
+                    ],
+                    "lines": [],
+                    "index": 4,
+                    "bbox_fs": [
+                        105,
+                        82,
+                        506,
+                        183
+                    ],
+                    "lines_deleted": true
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        187,
+                        505,
+                        352
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                187,
+                                505,
+                                200
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        187,
+                                        505,
+                                        200
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The core idea in this paper is that we can leverage unsupervised embeddings to propose tasks for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                198,
+                                505,
+                                212
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        198,
+                                        505,
+                                        212
+                                    ],
+                                    "score": 1.0,
+                                    "content": "a meta-learning algorithm, leading to an unsupervised meta-learning algorithm that is particularly",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                209,
+                                505,
+                                222
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        209,
+                                        505,
+                                        222
+                                    ],
+                                    "score": 1.0,
+                                    "content": "effective as pre-training for human-specified downstream tasks. In the following sections, we for-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                220,
+                                505,
+                                232
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        220,
+                                        505,
+                                        232
+                                    ],
+                                    "score": 1.0,
+                                    "content": "malize our problem assumptions and goal, which match those of unsupervised learning, and discuss",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                230,
+                                505,
+                                243
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        230,
+                                        505,
+                                        243
+                                    ],
+                                    "score": 1.0,
+                                    "content": "several options for automatically deriving tasks from embeddings. We instantiate our method with",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                242,
+                                505,
+                                254
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        242,
+                                        505,
+                                        254
+                                    ],
+                                    "score": 1.0,
+                                    "content": "two meta-learning algorithms and compare to prior state-of-the-art unsupervised learning meth-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                252,
+                                505,
+                                266
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        252,
+                                        505,
+                                        266
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ods. Across four image datasets (MNIST, Omniglot, miniImageNet, and CelebA), we find that our",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                263,
+                                505,
+                                277
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        263,
+                                        505,
+                                        277
+                                    ],
+                                    "score": 1.0,
+                                    "content": "method consistently leads to effective downstream learning of a variety of human-specified tasks,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                274,
+                                505,
+                                288
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        274,
+                                        505,
+                                        288
+                                    ],
+                                    "score": 1.0,
+                                    "content": "including character recognition tasks, object classification tasks, and facial attribute discrimination",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                286,
+                                505,
+                                299
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        286,
+                                        505,
+                                        299
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tasks, without requiring any labels or hand-designed tasks during meta-learning and where key hy-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                296,
+                                505,
+                                310
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        296,
+                                        505,
+                                        310
+                                    ],
+                                    "score": 1.0,
+                                    "content": "perparameters of our method are held constant across all domains. We show that, even though our",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                307,
+                                505,
+                                321
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        307,
+                                        505,
+                                        321
+                                    ],
+                                    "score": 1.0,
+                                    "content": "unsupervised meta-learning algorithm trains for one-shot generalization, one instantiation of our ap-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                318,
+                                505,
+                                331
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        318,
+                                        505,
+                                        331
+                                    ],
+                                    "score": 1.0,
+                                    "content": "proach performs well not only on few-shot learning, but also when learning downstream tasks with",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                329,
+                                506,
+                                343
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        329,
+                                        506,
+                                        343
+                                    ],
+                                    "score": 1.0,
+                                    "content": "up to 50 training examples per class. In fact, some of our results begin to approach the performance",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                339,
+                                471,
+                                354
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        339,
+                                        471,
+                                        354
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of fully-supervised meta-learning techniques trained with fully-specified task distributions.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        }
+                    ],
+                    "index": 16,
+                    "bbox_fs": [
+                        104,
+                        187,
+                        506,
+                        354
+                    ]
+                },
+                {
+                    "type": "image",
+                    "bbox": [
+                        118,
+                        366,
+                        488,
+                        469
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                118,
+                                366,
+                                488,
+                                469
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        118,
+                                        366,
+                                        488,
+                                        469
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                118,
+                                                366,
+                                                488,
+                                                469
+                                            ],
+                                            "score": 0.949,
+                                            "type": "image",
+                                            "image_path": "2aa42c4a43d437245b1ba217e8b16d1cdbed8dd1add1db4a9609cd8a7a78f543.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 25,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        118,
+                                        366,
+                                        488,
+                                        400.3333333333333
+                                    ],
+                                    "spans": [],
+                                    "index": 24
+                                },
+                                {
+                                    "bbox": [
+                                        118,
+                                        400.3333333333333,
+                                        488,
+                                        434.66666666666663
+                                    ],
+                                    "spans": [],
+                                    "index": 25
+                                },
+                                {
+                                    "bbox": [
+                                        118,
+                                        434.66666666666663,
+                                        488,
+                                        468.99999999999994
+                                    ],
+                                    "spans": [],
+                                    "index": 26
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                106,
+                                474,
+                                505,
+                                515
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        475,
+                                        505,
+                                        486
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                107,
+                                                475,
+                                                505,
+                                                486
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 1: Illustration of the proposed unsupervised meta-learning procedure. Embeddings of raw observations",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 27
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        484,
+                                        505,
+                                        496
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                484,
+                                                174,
+                                                496
+                                            ],
+                                            "score": 1.0,
+                                            "content": "are clustered with",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                174,
+                                                485,
+                                                180,
+                                                493
+                                            ],
+                                            "score": 0.82,
+                                            "content": "k",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                181,
+                                                484,
+                                                505,
+                                                496
+                                            ],
+                                            "score": 1.0,
+                                            "content": "-means to construct partitions, which give rise to classification tasks. Each task involves",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 28
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        493,
+                                        506,
+                                        507
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                493,
+                                                250,
+                                                507
+                                            ],
+                                            "score": 1.0,
+                                            "content": "distinguishing between examples from",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                251,
+                                                495,
+                                                280,
+                                                504
+                                            ],
+                                            "score": 0.9,
+                                            "content": "N = 2",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                281,
+                                                493,
+                                                334,
+                                                507
+                                            ],
+                                            "score": 1.0,
+                                            "content": "clusters, with",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                334,
+                                                495,
+                                                374,
+                                                505
+                                            ],
+                                            "score": 0.92,
+                                            "content": "K _ { \\mathrm { m - t r } } = 1",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                374,
+                                                493,
+                                                506,
+                                                507
+                                            ],
+                                            "score": 1.0,
+                                            "content": "example from each cluster being a",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 29
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        505,
+                                        499,
+                                        515
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                505,
+                                                499,
+                                                515
+                                            ],
+                                            "score": 1.0,
+                                            "content": "training input. The meta-learner’s aim is to produce a learning procedure that successfully solves these tasks.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 30
+                                }
+                            ],
+                            "index": 28.5
+                        }
+                    ],
+                    "index": 26.75
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        542,
+                        299,
+                        554
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                540,
+                                300,
+                                557
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        540,
+                                        300,
+                                        557
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2 UNSUPERVISED META-LEARNING",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        }
+                    ],
+                    "index": 31
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        104,
+                        566,
+                        503,
+                        588
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                566,
+                                504,
+                                578
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        566,
+                                        504,
+                                        578
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In this section, we describe our problem setting in relation to that of unsupervised and semi-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                577,
+                                426,
+                                590
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        577,
+                                        426,
+                                        590
+                                    ],
+                                    "score": 1.0,
+                                    "content": "supervised learning, provide necessary preliminaries, and present our approach.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        }
+                    ],
+                    "index": 32.5,
+                    "bbox_fs": [
+                        105,
+                        566,
+                        504,
+                        590
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        603,
+                        227,
+                        613
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                602,
+                                228,
+                                615
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        602,
+                                        228,
+                                        615
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2.1 PROBLEM STATEMENT",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        }
+                    ],
+                    "index": 34
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        622,
+                        505,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                621,
+                                504,
+                                634
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        621,
+                                        504,
+                                        634
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Our goal is to leverage unlabeled data for the efficient learning of a range of human-specified down-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                632,
+                                505,
+                                646
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        632,
+                                        350,
+                                        646
+                                    ],
+                                    "score": 1.0,
+                                    "content": "stream tasks. We only assume access to an unlabeled dataset",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        351,
+                                        633,
+                                        392,
+                                        645
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\mathcal { D } = \\{ \\mathbf { x } _ { i } \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        393,
+                                        632,
+                                        505,
+                                        646
+                                    ],
+                                    "score": 1.0,
+                                    "content": "during meta-training. After",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                643,
+                                505,
+                                657
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        643,
+                                        505,
+                                        657
+                                    ],
+                                    "score": 1.0,
+                                    "content": "learning from the unlabeled data, which we will refer to as unsupervised meta-training, we want",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                655,
+                                505,
+                                668
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        655,
+                                        505,
+                                        668
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to apply what was learned towards learning a variety of downstream, human-specified tasks from",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                666,
+                                505,
+                                678
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        666,
+                                        505,
+                                        678
+                                    ],
+                                    "score": 1.0,
+                                    "content": "a modest amount of labeled data, potentially as few as a single example per class. These down-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                678,
+                                505,
+                                689
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        678,
+                                        505,
+                                        689
+                                    ],
+                                    "score": 1.0,
+                                    "content": "stream tasks may, in general, have different underlying classes or attributes (in contrast to typical",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                687,
+                                506,
+                                700
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        687,
+                                        506,
+                                        700
+                                    ],
+                                    "score": 1.0,
+                                    "content": "semi-supervised problem assumptions), but are assumed to have inputs from the same distribution",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                699,
+                                505,
+                                711
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        699,
+                                        257,
+                                        711
+                                    ],
+                                    "score": 1.0,
+                                    "content": "as the one from which datapoints in",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        257,
+                                        700,
+                                        266,
+                                        708
+                                    ],
+                                    "score": 0.83,
+                                    "content": "\\mathcal { D }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        266,
+                                        699,
+                                        505,
+                                        711
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are drawn. Concretely, we assume that downstream tasks",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                709,
+                                505,
+                                721
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        709,
+                                        122,
+                                        721
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        122,
+                                        710,
+                                        134,
+                                        720
+                                    ],
+                                    "score": 0.87,
+                                    "content": "M",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        134,
+                                        709,
+                                        461,
+                                        721
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-way classification tasks, and that the goal is to learn an accurate classifier using",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        461,
+                                        710,
+                                        472,
+                                        720
+                                    ],
+                                    "score": 0.79,
+                                    "content": "K",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        473,
+                                        709,
+                                        505,
+                                        721
+                                    ],
+                                    "score": 1.0,
+                                    "content": "labeled",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                720,
+                                505,
+                                733
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        720,
+                                        150,
+                                        733
+                                    ],
+                                    "score": 1.0,
+                                    "content": "datapoints",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        150,
+                                        721,
+                                        185,
+                                        732
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\left( \\mathbf { x } _ { k } , \\mathbf { y } _ { k } \\right)",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        185,
+                                        720,
+                                        253,
+                                        733
+                                    ],
+                                    "score": 1.0,
+                                    "content": "from each of the",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        253,
+                                        721,
+                                        265,
+                                        730
+                                    ],
+                                    "score": 0.8,
+                                    "content": "M",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        266,
+                                        720,
+                                        325,
+                                        733
+                                    ],
+                                    "score": 1.0,
+                                    "content": "classes, where",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        326,
+                                        723,
+                                        335,
+                                        730
+                                    ],
+                                    "score": 0.87,
+                                    "content": "K",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        336,
+                                        720,
+                                        505,
+                                        733
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is relatively small (i.e. between 1 and 50).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        }
+                    ],
+                    "index": 39.5,
+                    "bbox_fs": [
+                        105,
+                        621,
+                        506,
+                        733
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        82,
+                        505,
+                        171
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                82,
+                                505,
+                                95
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        82,
+                                        505,
+                                        95
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The unsupervised meta-training phase aligns with the unsupervised learning problem in that it in-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                92,
+                                505,
+                                107
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        92,
+                                        472,
+                                        107
+                                    ],
+                                    "score": 1.0,
+                                    "content": "volves no access to information about the downstream tasks, other than the fact that they are",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        472,
+                                        94,
+                                        484,
+                                        104
+                                    ],
+                                    "score": 0.78,
+                                    "content": "M",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        484,
+                                        92,
+                                        505,
+                                        107
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-way",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                105,
+                                505,
+                                117
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        105,
+                                        238,
+                                        117
+                                    ],
+                                    "score": 1.0,
+                                    "content": "classification tasks, for variable",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        239,
+                                        105,
+                                        251,
+                                        115
+                                    ],
+                                    "score": 0.72,
+                                    "content": "M",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        252,
+                                        105,
+                                        330,
+                                        117
+                                    ],
+                                    "score": 1.0,
+                                    "content": "upper-bounded by",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        330,
+                                        105,
+                                        340,
+                                        114
+                                    ],
+                                    "score": 0.75,
+                                    "content": "N",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        341,
+                                        105,
+                                        421,
+                                        117
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". The upper bound",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        421,
+                                        105,
+                                        432,
+                                        114
+                                    ],
+                                    "score": 0.82,
+                                    "content": "N",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        432,
+                                        105,
+                                        505,
+                                        117
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is assumed to be",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                115,
+                                504,
+                                128
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        115,
+                                        394,
+                                        128
+                                    ],
+                                    "score": 1.0,
+                                    "content": "known during unsupervised meta-training, but otherwise, the values of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        394,
+                                        116,
+                                        406,
+                                        125
+                                    ],
+                                    "score": 0.8,
+                                    "content": "M",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        406,
+                                        115,
+                                        425,
+                                        128
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        425,
+                                        116,
+                                        435,
+                                        126
+                                    ],
+                                    "score": 0.81,
+                                    "content": "K",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        436,
+                                        115,
+                                        497,
+                                        128
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are not known",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        497,
+                                        117,
+                                        504,
+                                        125
+                                    ],
+                                    "score": 0.46,
+                                    "content": "a",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                127,
+                                505,
+                                139
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        127,
+                                        505,
+                                        139
+                                    ],
+                                    "score": 1.0,
+                                    "content": "priori. As a result, the unsupervised meta-training phase needs to acquire a sufficiently general prior",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                137,
+                                505,
+                                149
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        137,
+                                        505,
+                                        149
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for applicability to a range of classification tasks with variable quantities of data and classes. This",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                148,
+                                506,
+                                161
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        148,
+                                        506,
+                                        161
+                                    ],
+                                    "score": 1.0,
+                                    "content": "problem definition is our prototype for a practical use-case in which a user would like to train an",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                160,
+                                447,
+                                172
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        160,
+                                        447,
+                                        172
+                                    ],
+                                    "score": 1.0,
+                                    "content": "application-specific image classifier, but does not have an abundance of labeled data.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        }
+                    ],
+                    "index": 3.5
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        184,
+                        200,
+                        195
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                183,
+                                201,
+                                197
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        183,
+                                        201,
+                                        197
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2.2 PRELIMINARIES",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        }
+                    ],
+                    "index": 8
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        203,
+                        504,
+                        258
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                107,
+                                204,
+                                505,
+                                216
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        204,
+                                        459,
+                                        216
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Unsupervised embedding learning. An unsupervised embedding learning algorithm",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        459,
+                                        204,
+                                        466,
+                                        213
+                                    ],
+                                    "score": 0.78,
+                                    "content": "\\mathcal { E }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        466,
+                                        204,
+                                        505,
+                                        216
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is a pro-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                213,
+                                506,
+                                227
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        213,
+                                        302,
+                                        227
+                                    ],
+                                    "score": 1.0,
+                                    "content": "cedure that takes as input an unlabeled dataset",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        302,
+                                        214,
+                                        349,
+                                        226
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\mathcal { D } = \\left\\{ \\mathbf { x } _ { i } \\right\\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        349,
+                                        213,
+                                        471,
+                                        227
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and outputs a mapping from",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        472,
+                                        214,
+                                        492,
+                                        226
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\{ { \\bf { x } } _ { i } \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        493,
+                                        213,
+                                        506,
+                                        227
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                225,
+                                506,
+                                238
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        225,
+                                        158,
+                                        238
+                                    ],
+                                    "score": 1.0,
+                                    "content": "embeddings",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        158,
+                                        225,
+                                        177,
+                                        237
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\left\\{ \\mathbf { z } _ { i } \\right\\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        177,
+                                        225,
+                                        506,
+                                        238
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". These embedded points are typically lower-dimensional and arranged such that",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                236,
+                                506,
+                                249
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        236,
+                                        506,
+                                        249
+                                    ],
+                                    "score": 1.0,
+                                    "content": "distances correspond to meaningful differences between inputs, in contrast to distances between the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                246,
+                                480,
+                                260
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        246,
+                                        480,
+                                        260
+                                    ],
+                                    "score": 1.0,
+                                    "content": "original inputs, such as image pixels, which are not meaningful measures of image similarity.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        }
+                    ],
+                    "index": 11
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        263,
+                        505,
+                        308
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                263,
+                                505,
+                                276
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        263,
+                                        156,
+                                        276
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Task. An",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        156,
+                                        264,
+                                        168,
+                                        274
+                                    ],
+                                    "score": 0.76,
+                                    "content": "M",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        168,
+                                        263,
+                                        191,
+                                        276
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-way",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        191,
+                                        264,
+                                        201,
+                                        274
+                                    ],
+                                    "score": 0.8,
+                                    "content": "K",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        202,
+                                        263,
+                                        302,
+                                        276
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-shot classification task",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        303,
+                                        264,
+                                        313,
+                                        274
+                                    ],
+                                    "score": 0.77,
+                                    "content": "\\tau",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        313,
+                                        263,
+                                        365,
+                                        276
+                                    ],
+                                    "score": 1.0,
+                                    "content": "consists of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        365,
+                                        264,
+                                        375,
+                                        274
+                                    ],
+                                    "score": 0.81,
+                                    "content": "K",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        376,
+                                        263,
+                                        505,
+                                        276
+                                    ],
+                                    "score": 1.0,
+                                    "content": "training datapoints and labels",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                107,
+                                275,
+                                506,
+                                288
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        275,
+                                        151,
+                                        287
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\{ ( \\mathbf { x } _ { k } , \\ell _ { k } ) \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        151,
+                                        275,
+                                        379,
+                                        288
+                                    ],
+                                    "score": 1.0,
+                                    "content": "per class, which are used for learning a classifier, and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        379,
+                                        275,
+                                        388,
+                                        286
+                                    ],
+                                    "score": 0.83,
+                                    "content": "Q",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        389,
+                                        275,
+                                        506,
+                                        288
+                                    ],
+                                    "score": 1.0,
+                                    "content": "query datapoints and labels",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                286,
+                                504,
+                                298
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        286,
+                                        447,
+                                        298
+                                    ],
+                                    "score": 1.0,
+                                    "content": "per class, on which the learned classifier is evaluated. That is, in a task there are",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        447,
+                                        286,
+                                        504,
+                                        297
+                                    ],
+                                    "score": 0.91,
+                                    "content": "K + Q = R",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                297,
+                                299,
+                                309
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        297,
+                                        253,
+                                        309
+                                    ],
+                                    "score": 1.0,
+                                    "content": "datapoints and labels for each of the",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        253,
+                                        297,
+                                        265,
+                                        307
+                                    ],
+                                    "score": 0.68,
+                                    "content": "M",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        265,
+                                        297,
+                                        299,
+                                        309
+                                    ],
+                                    "score": 1.0,
+                                    "content": "classes.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        }
+                    ],
+                    "index": 15.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        313,
+                        505,
+                        402
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                313,
+                                505,
+                                327
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        313,
+                                        325,
+                                        327
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Meta-learning. A supervised meta-learning algorithm",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        325,
+                                        314,
+                                        348,
+                                        325
+                                    ],
+                                    "score": 0.89,
+                                    "content": "\\mathcal { M } ( \\cdot )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        349,
+                                        313,
+                                        505,
+                                        327
+                                    ],
+                                    "score": 1.0,
+                                    "content": "takes as input a set of supervised meta-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                324,
+                                506,
+                                338
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        324,
+                                        164,
+                                        338
+                                    ],
+                                    "score": 1.0,
+                                    "content": "training tasks",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        164,
+                                        325,
+                                        184,
+                                        337
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\{ \\mathcal { T } _ { t } \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        184,
+                                        324,
+                                        326,
+                                        338
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". It produces a learning procedure",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        326,
+                                        325,
+                                        345,
+                                        336
+                                    ],
+                                    "score": 0.9,
+                                    "content": "\\mathcal F ( \\cdot )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        346,
+                                        324,
+                                        506,
+                                        338
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", which, in turn, ingests the supervised",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                336,
+                                505,
+                                348
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        336,
+                                        282,
+                                        348
+                                    ],
+                                    "score": 1.0,
+                                    "content": "training data of a task to produce a classifier",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        282,
+                                        336,
+                                        299,
+                                        347
+                                    ],
+                                    "score": 0.89,
+                                    "content": "f ( \\cdot )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        300,
+                                        336,
+                                        352,
+                                        348
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". The goal of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        352,
+                                        336,
+                                        365,
+                                        346
+                                    ],
+                                    "score": 0.84,
+                                    "content": "\\mathcal { M }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        365,
+                                        336,
+                                        407,
+                                        348
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is to learn",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        407,
+                                        336,
+                                        416,
+                                        346
+                                    ],
+                                    "score": 0.83,
+                                    "content": "\\mathcal { F }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        417,
+                                        336,
+                                        505,
+                                        348
+                                    ],
+                                    "score": 1.0,
+                                    "content": "such that, when faced",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                346,
+                                506,
+                                360
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        346,
+                                        216,
+                                        360
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with a meta-test time task",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        216,
+                                        347,
+                                        228,
+                                        358
+                                    ],
+                                    "score": 0.88,
+                                    "content": "\\mathcal { T } _ { t ^ { \\prime } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        229,
+                                        346,
+                                        290,
+                                        360
+                                    ],
+                                    "score": 1.0,
+                                    "content": "held-out from",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        290,
+                                        347,
+                                        310,
+                                        359
+                                    ],
+                                    "score": 0.89,
+                                    "content": "\\{ \\mathcal { T } _ { t } \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        310,
+                                        346,
+                                        315,
+                                        360
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        315,
+                                        347,
+                                        325,
+                                        357
+                                    ],
+                                    "score": 0.78,
+                                    "content": "\\mathcal { F }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        325,
+                                        346,
+                                        375,
+                                        360
+                                    ],
+                                    "score": 1.0,
+                                    "content": "can learn a",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        375,
+                                        347,
+                                        387,
+                                        358
+                                    ],
+                                    "score": 0.88,
+                                    "content": "f _ { t ^ { \\prime } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        387,
+                                        346,
+                                        464,
+                                        360
+                                    ],
+                                    "score": 1.0,
+                                    "content": "that accomplishes",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        465,
+                                        347,
+                                        477,
+                                        358
+                                    ],
+                                    "score": 0.87,
+                                    "content": "\\mathcal { T } _ { t ^ { \\prime } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        478,
+                                        346,
+                                        506,
+                                        360
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". At a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                357,
+                                505,
+                                370
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        357,
+                                        353,
+                                        370
+                                    ],
+                                    "score": 1.0,
+                                    "content": "high level, the quintessential meta-learning strategy is to have",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        353,
+                                        358,
+                                        366,
+                                        368
+                                    ],
+                                    "score": 0.84,
+                                    "content": "\\mathcal { M }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        366,
+                                        357,
+                                        414,
+                                        370
+                                    ],
+                                    "score": 1.0,
+                                    "content": "iterate over",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        415,
+                                        357,
+                                        434,
+                                        370
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\{ \\mathcal { T } _ { t } \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        434,
+                                        357,
+                                        505,
+                                        370
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", cycling between",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                368,
+                                506,
+                                382
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        368,
+                                        228,
+                                        382
+                                    ],
+                                    "score": 1.0,
+                                    "content": "applying the current form of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        228,
+                                        369,
+                                        240,
+                                        380
+                                    ],
+                                    "score": 0.88,
+                                    "content": "\\mathcal { F } _ { t }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        241,
+                                        368,
+                                        335,
+                                        382
+                                    ],
+                                    "score": 1.0,
+                                    "content": "on training data from",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        335,
+                                        369,
+                                        345,
+                                        380
+                                    ],
+                                    "score": 0.85,
+                                    "content": "\\mathcal { T } _ { t }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        345,
+                                        368,
+                                        383,
+                                        382
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to learn",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        383,
+                                        369,
+                                        393,
+                                        380
+                                    ],
+                                    "score": 0.88,
+                                    "content": "f _ { t }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        393,
+                                        368,
+                                        506,
+                                        382
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", assessing its performance",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                380,
+                                505,
+                                393
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        380,
+                                        231,
+                                        393
+                                    ],
+                                    "score": 1.0,
+                                    "content": "by calculating some meta-loss",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        232,
+                                        380,
+                                        240,
+                                        389
+                                    ],
+                                    "score": 0.8,
+                                    "content": "\\mathcal { L }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        240,
+                                        380,
+                                        433,
+                                        393
+                                    ],
+                                    "score": 1.0,
+                                    "content": "on held-out data from the task, and optimizing",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        434,
+                                        380,
+                                        442,
+                                        389
+                                    ],
+                                    "score": 0.8,
+                                    "content": "\\mathcal { L }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        442,
+                                        380,
+                                        505,
+                                        393
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to improve the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                391,
+                                186,
+                                403
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        391,
+                                        186,
+                                        403
+                                    ],
+                                    "score": 1.0,
+                                    "content": "learning procedure.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        }
+                    ],
+                    "index": 21.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        407,
+                        505,
+                        496
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                407,
+                                505,
+                                421
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        407,
+                                        505,
+                                        421
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We build upon two meta-learning algorithms: model agnostic meta-learning (MAML) (Finn et al.,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                419,
+                                505,
+                                430
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        419,
+                                        505,
+                                        430
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2017) and prototypical networks (ProtoNets) (Snell et al., 2017). MAML aims to learn the initial",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                430,
+                                505,
+                                442
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        430,
+                                        505,
+                                        442
+                                    ],
+                                    "score": 1.0,
+                                    "content": "parameters of a deep network such that one or a few gradient steps leads to effective generalization;",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                441,
+                                505,
+                                453
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        441,
+                                        151,
+                                        453
+                                    ],
+                                    "score": 1.0,
+                                    "content": "it specifies",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        151,
+                                        441,
+                                        160,
+                                        450
+                                    ],
+                                    "score": 0.82,
+                                    "content": "\\mathcal { F }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        161,
+                                        441,
+                                        505,
+                                        453
+                                    ],
+                                    "score": 1.0,
+                                    "content": "as gradient descent starting from the meta-learned parameters. ProtoNets aim to meta-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                452,
+                                505,
+                                464
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        452,
+                                        505,
+                                        464
+                                    ],
+                                    "score": 1.0,
+                                    "content": "learn a representation in which a class is effectively identified by its prototype, defined to be the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                462,
+                                505,
+                                475
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        462,
+                                        362,
+                                        475
+                                    ],
+                                    "score": 1.0,
+                                    "content": "mean of the class’ training examples in the meta-learned space;",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        362,
+                                        463,
+                                        372,
+                                        473
+                                    ],
+                                    "score": 0.83,
+                                    "content": "\\mathcal { F }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        372,
+                                        462,
+                                        505,
+                                        475
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is the computation of these class",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                473,
+                                505,
+                                486
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        473,
+                                        170,
+                                        486
+                                    ],
+                                    "score": 1.0,
+                                    "content": "prototypes, and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        171,
+                                        474,
+                                        178,
+                                        484
+                                    ],
+                                    "score": 0.85,
+                                    "content": "f",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        178,
+                                        473,
+                                        505,
+                                        486
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is a linear classifier that predicts the class whose prototype is closest in Euclidean",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                485,
+                                259,
+                                497
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        485,
+                                        259,
+                                        497
+                                    ],
+                                    "score": 1.0,
+                                    "content": "distance to the query’s representation.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        }
+                    ],
+                    "index": 29.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        501,
+                        505,
+                        589
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                500,
+                                506,
+                                514
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        500,
+                                        506,
+                                        514
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Task generation for meta-learning. We briefly summarize how tasks are typically generated from",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                511,
+                                506,
+                                526
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        511,
+                                        172,
+                                        526
+                                    ],
+                                    "score": 1.0,
+                                    "content": "labeled datasets",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        172,
+                                        512,
+                                        213,
+                                        524
+                                    ],
+                                    "score": 0.94,
+                                    "content": "\\left\\{ \\left( \\mathbf { x } _ { i } , \\mathbf { y } _ { i } \\right) \\right\\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        214,
+                                        511,
+                                        506,
+                                        526
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for supervised meta-learning, as introduced by Santoro et al. (2016). For",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                523,
+                                505,
+                                536
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        523,
+                                        393,
+                                        536
+                                    ],
+                                    "score": 1.0,
+                                    "content": "simplicity, consider the case where the labels are discrete scalar values",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        394,
+                                        525,
+                                        403,
+                                        535
+                                    ],
+                                    "score": 0.84,
+                                    "content": "y _ { i }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        404,
+                                        523,
+                                        473,
+                                        536
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". To construct an",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        474,
+                                        524,
+                                        484,
+                                        533
+                                    ],
+                                    "score": 0.81,
+                                    "content": "N",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        484,
+                                        523,
+                                        505,
+                                        536
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-way",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                533,
+                                505,
+                                546
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        533,
+                                        182,
+                                        546
+                                    ],
+                                    "score": 1.0,
+                                    "content": "classification task",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        182,
+                                        534,
+                                        192,
+                                        544
+                                    ],
+                                    "score": 0.82,
+                                    "content": "\\tau",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        192,
+                                        533,
+                                        238,
+                                        546
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(assuming",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        239,
+                                        534,
+                                        249,
+                                        544
+                                    ],
+                                    "score": 0.81,
+                                    "content": "N",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        249,
+                                        533,
+                                        423,
+                                        546
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is not greater than the number of unique",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        423,
+                                        535,
+                                        433,
+                                        545
+                                    ],
+                                    "score": 0.81,
+                                    "content": "y _ { i }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        433,
+                                        533,
+                                        505,
+                                        546
+                                    ],
+                                    "score": 1.0,
+                                    "content": "), we can sample",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                107,
+                                544,
+                                504,
+                                557
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        545,
+                                        117,
+                                        555
+                                    ],
+                                    "score": 0.79,
+                                    "content": "N",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        117,
+                                        544,
+                                        183,
+                                        557
+                                    ],
+                                    "score": 1.0,
+                                    "content": "classes, sample",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        183,
+                                        546,
+                                        192,
+                                        555
+                                    ],
+                                    "score": 0.76,
+                                    "content": "R",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        193,
+                                        544,
+                                        238,
+                                        557
+                                    ],
+                                    "score": 1.0,
+                                    "content": "datapoints",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        238,
+                                        545,
+                                        265,
+                                        557
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\{ { \\bf { x } } _ { r } \\} _ { n }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        265,
+                                        544,
+                                        329,
+                                        557
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for each of the",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        329,
+                                        545,
+                                        340,
+                                        555
+                                    ],
+                                    "score": 0.78,
+                                    "content": "N",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        340,
+                                        544,
+                                        493,
+                                        557
+                                    ],
+                                    "score": 1.0,
+                                    "content": "classes, and sample a permutation of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        494,
+                                        546,
+                                        504,
+                                        555
+                                    ],
+                                    "score": 0.82,
+                                    "content": "N",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                555,
+                                506,
+                                569
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        555,
+                                        204,
+                                        569
+                                    ],
+                                    "score": 1.0,
+                                    "content": "distinct one-hot vectors",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        204,
+                                        556,
+                                        222,
+                                        568
+                                    ],
+                                    "score": 0.86,
+                                    "content": "( \\ell _ { n } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        223,
+                                        555,
+                                        375,
+                                        569
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to serve as task-specific labels of the",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        376,
+                                        556,
+                                        386,
+                                        566
+                                    ],
+                                    "score": 0.77,
+                                    "content": "N",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        387,
+                                        555,
+                                        506,
+                                        569
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sampled classes. The task is",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                567,
+                                506,
+                                580
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        567,
+                                        170,
+                                        580
+                                    ],
+                                    "score": 1.0,
+                                    "content": "then defined as",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        170,
+                                        567,
+                                        310,
+                                        579
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\mathcal { T } = \\{ ( \\mathbf { x } _ { n , r } , \\ell _ { n } ) ~ | ~ \\mathbf { x } _ { n , r } \\in \\{ \\mathbf { x } _ { r } \\} _ { n } \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        310,
+                                        567,
+                                        506,
+                                        580
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Of course, this procedure is only possible with",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                577,
+                                505,
+                                590
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        577,
+                                        505,
+                                        590
+                                    ],
+                                    "score": 1.0,
+                                    "content": "labeled data; in the next section, we discuss how we can construct tasks without ground-truth labels.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        }
+                    ],
+                    "index": 37.5
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        614,
+                        475,
+                        625
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                613,
+                                477,
+                                626
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        613,
+                                        477,
+                                        626
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2.3 UNSUPERVISED META-LEARNING WITH AUTOMATICALLY CONSTRUCTED TASKS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        }
+                    ],
+                    "index": 42
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        632,
+                        505,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                633,
+                                505,
+                                646
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        633,
+                                        505,
+                                        646
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We approach our problem from a meta-learning perspective, framing the problem as the acquisition,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                644,
+                                504,
+                                656
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        644,
+                                        504,
+                                        656
+                                    ],
+                                    "score": 1.0,
+                                    "content": "from unlabeled data, of an efficient learning procedure that is transferable to human-designed tasks.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                655,
+                                504,
+                                667
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        655,
+                                        504,
+                                        667
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In particular, we aim to construct classification tasks from the unlabeled data and then learn how",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                665,
+                                505,
+                                678
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        665,
+                                        505,
+                                        678
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to efficiently learn these tasks. If such tasks are adequately diverse and structured, then meta-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                677,
+                                505,
+                                689
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        677,
+                                        505,
+                                        689
+                                    ],
+                                    "score": 1.0,
+                                    "content": "learning these tasks should enable fast learning of new, human-provided tasks. A key question, then,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 47
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                687,
+                                505,
+                                700
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        687,
+                                        377,
+                                        700
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is how to automatically construct such tasks from unlabeled data",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        377,
+                                        687,
+                                        423,
+                                        700
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\mathcal { D } = \\left\\{ \\mathbf { x } _ { i } \\right\\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        423,
+                                        687,
+                                        505,
+                                        700
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Notice that in the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 48
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                698,
+                                506,
+                                711
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        698,
+                                        456,
+                                        711
+                                    ],
+                                    "score": 1.0,
+                                    "content": "supervised meta-learning task generation procedure detailed in Section 2.2, the labels",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        457,
+                                        700,
+                                        466,
+                                        710
+                                    ],
+                                    "score": 0.83,
+                                    "content": "y _ { i }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        467,
+                                        698,
+                                        506,
+                                        711
+                                    ],
+                                    "score": 1.0,
+                                    "content": "induce a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 49
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                709,
+                                506,
+                                722
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        709,
+                                        142,
+                                        722
+                                    ],
+                                    "score": 1.0,
+                                    "content": "partition",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        142,
+                                        709,
+                                        184,
+                                        722
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\mathcal { P } = \\{ \\mathcal { C } _ { c } \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        185,
+                                        709,
+                                        205,
+                                        722
+                                    ],
+                                    "score": 1.0,
+                                    "content": "over",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        205,
+                                        710,
+                                        225,
+                                        722
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\{ { \\bf { x } } _ { i } \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        226,
+                                        709,
+                                        375,
+                                        722
+                                    ],
+                                    "score": 1.0,
+                                    "content": "by assigning all datapoints with label",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        375,
+                                        711,
+                                        386,
+                                        721
+                                    ],
+                                    "score": 0.86,
+                                    "content": "y _ { c }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        386,
+                                        709,
+                                        424,
+                                        722
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to subset",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        424,
+                                        710,
+                                        434,
+                                        721
+                                    ],
+                                    "score": 0.86,
+                                    "content": "\\mathcal { C } _ { c }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        435,
+                                        709,
+                                        506,
+                                        722
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Once a partition",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 50
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                720,
+                                506,
+                                733
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        720,
+                                        506,
+                                        733
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is obtained, task generation is simple; we can reduce the problem of constructing tasks to that of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 51
+                        }
+                    ],
+                    "index": 47
+                }
+            ],
+            "page_idx": 2,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        108,
+                        27,
+                        293,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                26,
+                                294,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        26,
+                                        294,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Published as a conference paper at ICLR 2019",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        302,
+                        751,
+                        309,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                301,
+                                750,
+                                309,
+                                762
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        301,
+                                        750,
+                                        309,
+                                        762
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        82,
+                        505,
+                        171
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                82,
+                                505,
+                                95
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        82,
+                                        505,
+                                        95
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The unsupervised meta-training phase aligns with the unsupervised learning problem in that it in-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                92,
+                                505,
+                                107
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        92,
+                                        472,
+                                        107
+                                    ],
+                                    "score": 1.0,
+                                    "content": "volves no access to information about the downstream tasks, other than the fact that they are",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        472,
+                                        94,
+                                        484,
+                                        104
+                                    ],
+                                    "score": 0.78,
+                                    "content": "M",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        484,
+                                        92,
+                                        505,
+                                        107
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-way",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                105,
+                                505,
+                                117
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        105,
+                                        238,
+                                        117
+                                    ],
+                                    "score": 1.0,
+                                    "content": "classification tasks, for variable",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        239,
+                                        105,
+                                        251,
+                                        115
+                                    ],
+                                    "score": 0.72,
+                                    "content": "M",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        252,
+                                        105,
+                                        330,
+                                        117
+                                    ],
+                                    "score": 1.0,
+                                    "content": "upper-bounded by",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        330,
+                                        105,
+                                        340,
+                                        114
+                                    ],
+                                    "score": 0.75,
+                                    "content": "N",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        341,
+                                        105,
+                                        421,
+                                        117
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". The upper bound",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        421,
+                                        105,
+                                        432,
+                                        114
+                                    ],
+                                    "score": 0.82,
+                                    "content": "N",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        432,
+                                        105,
+                                        505,
+                                        117
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is assumed to be",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                115,
+                                504,
+                                128
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        115,
+                                        394,
+                                        128
+                                    ],
+                                    "score": 1.0,
+                                    "content": "known during unsupervised meta-training, but otherwise, the values of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        394,
+                                        116,
+                                        406,
+                                        125
+                                    ],
+                                    "score": 0.8,
+                                    "content": "M",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        406,
+                                        115,
+                                        425,
+                                        128
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        425,
+                                        116,
+                                        435,
+                                        126
+                                    ],
+                                    "score": 0.81,
+                                    "content": "K",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        436,
+                                        115,
+                                        497,
+                                        128
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are not known",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        497,
+                                        117,
+                                        504,
+                                        125
+                                    ],
+                                    "score": 0.46,
+                                    "content": "a",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                127,
+                                505,
+                                139
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        127,
+                                        505,
+                                        139
+                                    ],
+                                    "score": 1.0,
+                                    "content": "priori. As a result, the unsupervised meta-training phase needs to acquire a sufficiently general prior",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                137,
+                                505,
+                                149
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        137,
+                                        505,
+                                        149
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for applicability to a range of classification tasks with variable quantities of data and classes. This",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                148,
+                                506,
+                                161
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        148,
+                                        506,
+                                        161
+                                    ],
+                                    "score": 1.0,
+                                    "content": "problem definition is our prototype for a practical use-case in which a user would like to train an",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                160,
+                                447,
+                                172
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        160,
+                                        447,
+                                        172
+                                    ],
+                                    "score": 1.0,
+                                    "content": "application-specific image classifier, but does not have an abundance of labeled data.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        }
+                    ],
+                    "index": 3.5,
+                    "bbox_fs": [
+                        105,
+                        82,
+                        506,
+                        172
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        184,
+                        200,
+                        195
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                183,
+                                201,
+                                197
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        183,
+                                        201,
+                                        197
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2.2 PRELIMINARIES",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        }
+                    ],
+                    "index": 8
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        203,
+                        504,
+                        258
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                107,
+                                204,
+                                505,
+                                216
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        204,
+                                        459,
+                                        216
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Unsupervised embedding learning. An unsupervised embedding learning algorithm",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        459,
+                                        204,
+                                        466,
+                                        213
+                                    ],
+                                    "score": 0.78,
+                                    "content": "\\mathcal { E }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        466,
+                                        204,
+                                        505,
+                                        216
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is a pro-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                213,
+                                506,
+                                227
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        213,
+                                        302,
+                                        227
+                                    ],
+                                    "score": 1.0,
+                                    "content": "cedure that takes as input an unlabeled dataset",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        302,
+                                        214,
+                                        349,
+                                        226
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\mathcal { D } = \\left\\{ \\mathbf { x } _ { i } \\right\\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        349,
+                                        213,
+                                        471,
+                                        227
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and outputs a mapping from",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        472,
+                                        214,
+                                        492,
+                                        226
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\{ { \\bf { x } } _ { i } \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        493,
+                                        213,
+                                        506,
+                                        227
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                225,
+                                506,
+                                238
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        225,
+                                        158,
+                                        238
+                                    ],
+                                    "score": 1.0,
+                                    "content": "embeddings",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        158,
+                                        225,
+                                        177,
+                                        237
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\left\\{ \\mathbf { z } _ { i } \\right\\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        177,
+                                        225,
+                                        506,
+                                        238
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". These embedded points are typically lower-dimensional and arranged such that",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                236,
+                                506,
+                                249
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        236,
+                                        506,
+                                        249
+                                    ],
+                                    "score": 1.0,
+                                    "content": "distances correspond to meaningful differences between inputs, in contrast to distances between the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                246,
+                                480,
+                                260
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        246,
+                                        480,
+                                        260
+                                    ],
+                                    "score": 1.0,
+                                    "content": "original inputs, such as image pixels, which are not meaningful measures of image similarity.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        }
+                    ],
+                    "index": 11,
+                    "bbox_fs": [
+                        105,
+                        204,
+                        506,
+                        260
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        263,
+                        505,
+                        308
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                263,
+                                505,
+                                276
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        263,
+                                        156,
+                                        276
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Task. An",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        156,
+                                        264,
+                                        168,
+                                        274
+                                    ],
+                                    "score": 0.76,
+                                    "content": "M",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        168,
+                                        263,
+                                        191,
+                                        276
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-way",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        191,
+                                        264,
+                                        201,
+                                        274
+                                    ],
+                                    "score": 0.8,
+                                    "content": "K",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        202,
+                                        263,
+                                        302,
+                                        276
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-shot classification task",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        303,
+                                        264,
+                                        313,
+                                        274
+                                    ],
+                                    "score": 0.77,
+                                    "content": "\\tau",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        313,
+                                        263,
+                                        365,
+                                        276
+                                    ],
+                                    "score": 1.0,
+                                    "content": "consists of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        365,
+                                        264,
+                                        375,
+                                        274
+                                    ],
+                                    "score": 0.81,
+                                    "content": "K",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        376,
+                                        263,
+                                        505,
+                                        276
+                                    ],
+                                    "score": 1.0,
+                                    "content": "training datapoints and labels",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                107,
+                                275,
+                                506,
+                                288
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        275,
+                                        151,
+                                        287
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\{ ( \\mathbf { x } _ { k } , \\ell _ { k } ) \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        151,
+                                        275,
+                                        379,
+                                        288
+                                    ],
+                                    "score": 1.0,
+                                    "content": "per class, which are used for learning a classifier, and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        379,
+                                        275,
+                                        388,
+                                        286
+                                    ],
+                                    "score": 0.83,
+                                    "content": "Q",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        389,
+                                        275,
+                                        506,
+                                        288
+                                    ],
+                                    "score": 1.0,
+                                    "content": "query datapoints and labels",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                286,
+                                504,
+                                298
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        286,
+                                        447,
+                                        298
+                                    ],
+                                    "score": 1.0,
+                                    "content": "per class, on which the learned classifier is evaluated. That is, in a task there are",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        447,
+                                        286,
+                                        504,
+                                        297
+                                    ],
+                                    "score": 0.91,
+                                    "content": "K + Q = R",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                297,
+                                299,
+                                309
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        297,
+                                        253,
+                                        309
+                                    ],
+                                    "score": 1.0,
+                                    "content": "datapoints and labels for each of the",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        253,
+                                        297,
+                                        265,
+                                        307
+                                    ],
+                                    "score": 0.68,
+                                    "content": "M",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        265,
+                                        297,
+                                        299,
+                                        309
+                                    ],
+                                    "score": 1.0,
+                                    "content": "classes.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        }
+                    ],
+                    "index": 15.5,
+                    "bbox_fs": [
+                        105,
+                        263,
+                        506,
+                        309
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        313,
+                        505,
+                        402
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                313,
+                                505,
+                                327
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        313,
+                                        325,
+                                        327
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Meta-learning. A supervised meta-learning algorithm",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        325,
+                                        314,
+                                        348,
+                                        325
+                                    ],
+                                    "score": 0.89,
+                                    "content": "\\mathcal { M } ( \\cdot )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        349,
+                                        313,
+                                        505,
+                                        327
+                                    ],
+                                    "score": 1.0,
+                                    "content": "takes as input a set of supervised meta-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                324,
+                                506,
+                                338
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        324,
+                                        164,
+                                        338
+                                    ],
+                                    "score": 1.0,
+                                    "content": "training tasks",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        164,
+                                        325,
+                                        184,
+                                        337
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\{ \\mathcal { T } _ { t } \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        184,
+                                        324,
+                                        326,
+                                        338
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". It produces a learning procedure",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        326,
+                                        325,
+                                        345,
+                                        336
+                                    ],
+                                    "score": 0.9,
+                                    "content": "\\mathcal F ( \\cdot )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        346,
+                                        324,
+                                        506,
+                                        338
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", which, in turn, ingests the supervised",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                336,
+                                505,
+                                348
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        336,
+                                        282,
+                                        348
+                                    ],
+                                    "score": 1.0,
+                                    "content": "training data of a task to produce a classifier",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        282,
+                                        336,
+                                        299,
+                                        347
+                                    ],
+                                    "score": 0.89,
+                                    "content": "f ( \\cdot )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        300,
+                                        336,
+                                        352,
+                                        348
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". The goal of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        352,
+                                        336,
+                                        365,
+                                        346
+                                    ],
+                                    "score": 0.84,
+                                    "content": "\\mathcal { M }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        365,
+                                        336,
+                                        407,
+                                        348
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is to learn",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        407,
+                                        336,
+                                        416,
+                                        346
+                                    ],
+                                    "score": 0.83,
+                                    "content": "\\mathcal { F }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        417,
+                                        336,
+                                        505,
+                                        348
+                                    ],
+                                    "score": 1.0,
+                                    "content": "such that, when faced",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                346,
+                                506,
+                                360
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        346,
+                                        216,
+                                        360
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with a meta-test time task",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        216,
+                                        347,
+                                        228,
+                                        358
+                                    ],
+                                    "score": 0.88,
+                                    "content": "\\mathcal { T } _ { t ^ { \\prime } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        229,
+                                        346,
+                                        290,
+                                        360
+                                    ],
+                                    "score": 1.0,
+                                    "content": "held-out from",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        290,
+                                        347,
+                                        310,
+                                        359
+                                    ],
+                                    "score": 0.89,
+                                    "content": "\\{ \\mathcal { T } _ { t } \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        310,
+                                        346,
+                                        315,
+                                        360
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        315,
+                                        347,
+                                        325,
+                                        357
+                                    ],
+                                    "score": 0.78,
+                                    "content": "\\mathcal { F }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        325,
+                                        346,
+                                        375,
+                                        360
+                                    ],
+                                    "score": 1.0,
+                                    "content": "can learn a",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        375,
+                                        347,
+                                        387,
+                                        358
+                                    ],
+                                    "score": 0.88,
+                                    "content": "f _ { t ^ { \\prime } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        387,
+                                        346,
+                                        464,
+                                        360
+                                    ],
+                                    "score": 1.0,
+                                    "content": "that accomplishes",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        465,
+                                        347,
+                                        477,
+                                        358
+                                    ],
+                                    "score": 0.87,
+                                    "content": "\\mathcal { T } _ { t ^ { \\prime } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        478,
+                                        346,
+                                        506,
+                                        360
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". At a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                357,
+                                505,
+                                370
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        357,
+                                        353,
+                                        370
+                                    ],
+                                    "score": 1.0,
+                                    "content": "high level, the quintessential meta-learning strategy is to have",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        353,
+                                        358,
+                                        366,
+                                        368
+                                    ],
+                                    "score": 0.84,
+                                    "content": "\\mathcal { M }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        366,
+                                        357,
+                                        414,
+                                        370
+                                    ],
+                                    "score": 1.0,
+                                    "content": "iterate over",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        415,
+                                        357,
+                                        434,
+                                        370
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\{ \\mathcal { T } _ { t } \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        434,
+                                        357,
+                                        505,
+                                        370
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", cycling between",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                368,
+                                506,
+                                382
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        368,
+                                        228,
+                                        382
+                                    ],
+                                    "score": 1.0,
+                                    "content": "applying the current form of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        228,
+                                        369,
+                                        240,
+                                        380
+                                    ],
+                                    "score": 0.88,
+                                    "content": "\\mathcal { F } _ { t }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        241,
+                                        368,
+                                        335,
+                                        382
+                                    ],
+                                    "score": 1.0,
+                                    "content": "on training data from",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        335,
+                                        369,
+                                        345,
+                                        380
+                                    ],
+                                    "score": 0.85,
+                                    "content": "\\mathcal { T } _ { t }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        345,
+                                        368,
+                                        383,
+                                        382
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to learn",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        383,
+                                        369,
+                                        393,
+                                        380
+                                    ],
+                                    "score": 0.88,
+                                    "content": "f _ { t }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        393,
+                                        368,
+                                        506,
+                                        382
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", assessing its performance",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                380,
+                                505,
+                                393
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        380,
+                                        231,
+                                        393
+                                    ],
+                                    "score": 1.0,
+                                    "content": "by calculating some meta-loss",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        232,
+                                        380,
+                                        240,
+                                        389
+                                    ],
+                                    "score": 0.8,
+                                    "content": "\\mathcal { L }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        240,
+                                        380,
+                                        433,
+                                        393
+                                    ],
+                                    "score": 1.0,
+                                    "content": "on held-out data from the task, and optimizing",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        434,
+                                        380,
+                                        442,
+                                        389
+                                    ],
+                                    "score": 0.8,
+                                    "content": "\\mathcal { L }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        442,
+                                        380,
+                                        505,
+                                        393
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to improve the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                391,
+                                186,
+                                403
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        391,
+                                        186,
+                                        403
+                                    ],
+                                    "score": 1.0,
+                                    "content": "learning procedure.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        }
+                    ],
+                    "index": 21.5,
+                    "bbox_fs": [
+                        104,
+                        313,
+                        506,
+                        403
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        407,
+                        505,
+                        496
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                407,
+                                505,
+                                421
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        407,
+                                        505,
+                                        421
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We build upon two meta-learning algorithms: model agnostic meta-learning (MAML) (Finn et al.,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                419,
+                                505,
+                                430
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        419,
+                                        505,
+                                        430
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2017) and prototypical networks (ProtoNets) (Snell et al., 2017). MAML aims to learn the initial",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                430,
+                                505,
+                                442
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        430,
+                                        505,
+                                        442
+                                    ],
+                                    "score": 1.0,
+                                    "content": "parameters of a deep network such that one or a few gradient steps leads to effective generalization;",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                441,
+                                505,
+                                453
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        441,
+                                        151,
+                                        453
+                                    ],
+                                    "score": 1.0,
+                                    "content": "it specifies",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        151,
+                                        441,
+                                        160,
+                                        450
+                                    ],
+                                    "score": 0.82,
+                                    "content": "\\mathcal { F }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        161,
+                                        441,
+                                        505,
+                                        453
+                                    ],
+                                    "score": 1.0,
+                                    "content": "as gradient descent starting from the meta-learned parameters. ProtoNets aim to meta-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                452,
+                                505,
+                                464
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        452,
+                                        505,
+                                        464
+                                    ],
+                                    "score": 1.0,
+                                    "content": "learn a representation in which a class is effectively identified by its prototype, defined to be the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                462,
+                                505,
+                                475
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        462,
+                                        362,
+                                        475
+                                    ],
+                                    "score": 1.0,
+                                    "content": "mean of the class’ training examples in the meta-learned space;",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        362,
+                                        463,
+                                        372,
+                                        473
+                                    ],
+                                    "score": 0.83,
+                                    "content": "\\mathcal { F }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        372,
+                                        462,
+                                        505,
+                                        475
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is the computation of these class",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                473,
+                                505,
+                                486
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        473,
+                                        170,
+                                        486
+                                    ],
+                                    "score": 1.0,
+                                    "content": "prototypes, and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        171,
+                                        474,
+                                        178,
+                                        484
+                                    ],
+                                    "score": 0.85,
+                                    "content": "f",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        178,
+                                        473,
+                                        505,
+                                        486
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is a linear classifier that predicts the class whose prototype is closest in Euclidean",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                485,
+                                259,
+                                497
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        485,
+                                        259,
+                                        497
+                                    ],
+                                    "score": 1.0,
+                                    "content": "distance to the query’s representation.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        }
+                    ],
+                    "index": 29.5,
+                    "bbox_fs": [
+                        105,
+                        407,
+                        505,
+                        497
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        501,
+                        505,
+                        589
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                500,
+                                506,
+                                514
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        500,
+                                        506,
+                                        514
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Task generation for meta-learning. We briefly summarize how tasks are typically generated from",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                511,
+                                506,
+                                526
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        511,
+                                        172,
+                                        526
+                                    ],
+                                    "score": 1.0,
+                                    "content": "labeled datasets",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        172,
+                                        512,
+                                        213,
+                                        524
+                                    ],
+                                    "score": 0.94,
+                                    "content": "\\left\\{ \\left( \\mathbf { x } _ { i } , \\mathbf { y } _ { i } \\right) \\right\\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        214,
+                                        511,
+                                        506,
+                                        526
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for supervised meta-learning, as introduced by Santoro et al. (2016). For",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                523,
+                                505,
+                                536
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        523,
+                                        393,
+                                        536
+                                    ],
+                                    "score": 1.0,
+                                    "content": "simplicity, consider the case where the labels are discrete scalar values",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        394,
+                                        525,
+                                        403,
+                                        535
+                                    ],
+                                    "score": 0.84,
+                                    "content": "y _ { i }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        404,
+                                        523,
+                                        473,
+                                        536
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". To construct an",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        474,
+                                        524,
+                                        484,
+                                        533
+                                    ],
+                                    "score": 0.81,
+                                    "content": "N",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        484,
+                                        523,
+                                        505,
+                                        536
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-way",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                533,
+                                505,
+                                546
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        533,
+                                        182,
+                                        546
+                                    ],
+                                    "score": 1.0,
+                                    "content": "classification task",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        182,
+                                        534,
+                                        192,
+                                        544
+                                    ],
+                                    "score": 0.82,
+                                    "content": "\\tau",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        192,
+                                        533,
+                                        238,
+                                        546
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(assuming",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        239,
+                                        534,
+                                        249,
+                                        544
+                                    ],
+                                    "score": 0.81,
+                                    "content": "N",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        249,
+                                        533,
+                                        423,
+                                        546
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is not greater than the number of unique",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        423,
+                                        535,
+                                        433,
+                                        545
+                                    ],
+                                    "score": 0.81,
+                                    "content": "y _ { i }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        433,
+                                        533,
+                                        505,
+                                        546
+                                    ],
+                                    "score": 1.0,
+                                    "content": "), we can sample",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                107,
+                                544,
+                                504,
+                                557
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        545,
+                                        117,
+                                        555
+                                    ],
+                                    "score": 0.79,
+                                    "content": "N",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        117,
+                                        544,
+                                        183,
+                                        557
+                                    ],
+                                    "score": 1.0,
+                                    "content": "classes, sample",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        183,
+                                        546,
+                                        192,
+                                        555
+                                    ],
+                                    "score": 0.76,
+                                    "content": "R",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        193,
+                                        544,
+                                        238,
+                                        557
+                                    ],
+                                    "score": 1.0,
+                                    "content": "datapoints",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        238,
+                                        545,
+                                        265,
+                                        557
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\{ { \\bf { x } } _ { r } \\} _ { n }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        265,
+                                        544,
+                                        329,
+                                        557
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for each of the",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        329,
+                                        545,
+                                        340,
+                                        555
+                                    ],
+                                    "score": 0.78,
+                                    "content": "N",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        340,
+                                        544,
+                                        493,
+                                        557
+                                    ],
+                                    "score": 1.0,
+                                    "content": "classes, and sample a permutation of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        494,
+                                        546,
+                                        504,
+                                        555
+                                    ],
+                                    "score": 0.82,
+                                    "content": "N",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                555,
+                                506,
+                                569
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        555,
+                                        204,
+                                        569
+                                    ],
+                                    "score": 1.0,
+                                    "content": "distinct one-hot vectors",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        204,
+                                        556,
+                                        222,
+                                        568
+                                    ],
+                                    "score": 0.86,
+                                    "content": "( \\ell _ { n } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        223,
+                                        555,
+                                        375,
+                                        569
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to serve as task-specific labels of the",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        376,
+                                        556,
+                                        386,
+                                        566
+                                    ],
+                                    "score": 0.77,
+                                    "content": "N",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        387,
+                                        555,
+                                        506,
+                                        569
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sampled classes. The task is",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                567,
+                                506,
+                                580
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        567,
+                                        170,
+                                        580
+                                    ],
+                                    "score": 1.0,
+                                    "content": "then defined as",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        170,
+                                        567,
+                                        310,
+                                        579
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\mathcal { T } = \\{ ( \\mathbf { x } _ { n , r } , \\ell _ { n } ) ~ | ~ \\mathbf { x } _ { n , r } \\in \\{ \\mathbf { x } _ { r } \\} _ { n } \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        310,
+                                        567,
+                                        506,
+                                        580
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Of course, this procedure is only possible with",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                577,
+                                505,
+                                590
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        577,
+                                        505,
+                                        590
+                                    ],
+                                    "score": 1.0,
+                                    "content": "labeled data; in the next section, we discuss how we can construct tasks without ground-truth labels.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        }
+                    ],
+                    "index": 37.5,
+                    "bbox_fs": [
+                        105,
+                        500,
+                        506,
+                        590
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        614,
+                        475,
+                        625
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                613,
+                                477,
+                                626
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        613,
+                                        477,
+                                        626
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2.3 UNSUPERVISED META-LEARNING WITH AUTOMATICALLY CONSTRUCTED TASKS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        }
+                    ],
+                    "index": 42
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        632,
+                        505,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                633,
+                                505,
+                                646
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        633,
+                                        505,
+                                        646
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We approach our problem from a meta-learning perspective, framing the problem as the acquisition,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                644,
+                                504,
+                                656
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        644,
+                                        504,
+                                        656
+                                    ],
+                                    "score": 1.0,
+                                    "content": "from unlabeled data, of an efficient learning procedure that is transferable to human-designed tasks.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                655,
+                                504,
+                                667
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        655,
+                                        504,
+                                        667
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In particular, we aim to construct classification tasks from the unlabeled data and then learn how",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                665,
+                                505,
+                                678
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        665,
+                                        505,
+                                        678
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to efficiently learn these tasks. If such tasks are adequately diverse and structured, then meta-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                677,
+                                505,
+                                689
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        677,
+                                        505,
+                                        689
+                                    ],
+                                    "score": 1.0,
+                                    "content": "learning these tasks should enable fast learning of new, human-provided tasks. A key question, then,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 47
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                687,
+                                505,
+                                700
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        687,
+                                        377,
+                                        700
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is how to automatically construct such tasks from unlabeled data",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        377,
+                                        687,
+                                        423,
+                                        700
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\mathcal { D } = \\left\\{ \\mathbf { x } _ { i } \\right\\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        423,
+                                        687,
+                                        505,
+                                        700
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Notice that in the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 48
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                698,
+                                506,
+                                711
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        698,
+                                        456,
+                                        711
+                                    ],
+                                    "score": 1.0,
+                                    "content": "supervised meta-learning task generation procedure detailed in Section 2.2, the labels",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        457,
+                                        700,
+                                        466,
+                                        710
+                                    ],
+                                    "score": 0.83,
+                                    "content": "y _ { i }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        467,
+                                        698,
+                                        506,
+                                        711
+                                    ],
+                                    "score": 1.0,
+                                    "content": "induce a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 49
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                709,
+                                506,
+                                722
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        709,
+                                        142,
+                                        722
+                                    ],
+                                    "score": 1.0,
+                                    "content": "partition",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        142,
+                                        709,
+                                        184,
+                                        722
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\mathcal { P } = \\{ \\mathcal { C } _ { c } \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        185,
+                                        709,
+                                        205,
+                                        722
+                                    ],
+                                    "score": 1.0,
+                                    "content": "over",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        205,
+                                        710,
+                                        225,
+                                        722
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\{ { \\bf { x } } _ { i } \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        226,
+                                        709,
+                                        375,
+                                        722
+                                    ],
+                                    "score": 1.0,
+                                    "content": "by assigning all datapoints with label",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        375,
+                                        711,
+                                        386,
+                                        721
+                                    ],
+                                    "score": 0.86,
+                                    "content": "y _ { c }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        386,
+                                        709,
+                                        424,
+                                        722
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to subset",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        424,
+                                        710,
+                                        434,
+                                        721
+                                    ],
+                                    "score": 0.86,
+                                    "content": "\\mathcal { C } _ { c }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        435,
+                                        709,
+                                        506,
+                                        722
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Once a partition",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 50
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                720,
+                                506,
+                                733
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        720,
+                                        506,
+                                        733
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is obtained, task generation is simple; we can reduce the problem of constructing tasks to that of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 51
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                82,
+                                505,
+                                95
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        82,
+                                        223,
+                                        95
+                                    ],
+                                    "score": 1.0,
+                                    "content": "constructing a partition over",
+                                    "type": "text",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        223,
+                                        82,
+                                        243,
+                                        95
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\{ { \\bf { x } } _ { i } \\}",
+                                    "type": "inline_equation",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        244,
+                                        82,
+                                        505,
+                                        95
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". All that’s left is to find a principled alternative to human labels",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                93,
+                                209,
+                                106
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        93,
+                                        209,
+                                        106
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for defining the partition.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 1
+                        }
+                    ],
+                    "index": 47,
+                    "bbox_fs": [
+                        105,
+                        633,
+                        506,
+                        733
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        82,
+                        504,
+                        105
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                82,
+                                505,
+                                95
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        82,
+                                        223,
+                                        95
+                                    ],
+                                    "score": 1.0,
+                                    "content": "constructing a partition over",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        223,
+                                        82,
+                                        243,
+                                        95
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\{ { \\bf { x } } _ { i } \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        244,
+                                        82,
+                                        505,
+                                        95
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". All that’s left is to find a principled alternative to human labels",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                93,
+                                209,
+                                106
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        93,
+                                        209,
+                                        106
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for defining the partition.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        }
+                    ],
+                    "index": 0.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        110,
+                        504,
+                        154
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                110,
+                                506,
+                                123
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        110,
+                                        320,
+                                        123
+                                    ],
+                                    "score": 1.0,
+                                    "content": "A na¨ıve approach is to randomly partition the data",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        320,
+                                        111,
+                                        329,
+                                        120
+                                    ],
+                                    "score": 0.76,
+                                    "content": "\\mathcal { D }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        329,
+                                        110,
+                                        506,
+                                        123
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". While such a scheme introduces diverse",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                120,
+                                506,
+                                134
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        120,
+                                        506,
+                                        134
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tasks, there is no structure; that is, there is no consistency between a task’s training data and query",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                133,
+                                506,
+                                144
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        133,
+                                        506,
+                                        144
+                                    ],
+                                    "score": 1.0,
+                                    "content": "data, and hence nothing to be learned during each task, let alone across tasks. As seen in Table 3,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                144,
+                                435,
+                                156
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        144,
+                                        435,
+                                        156
+                                    ],
+                                    "score": 1.0,
+                                    "content": "providing a meta-learner with purely random tasks results in failed meta-learning.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        }
+                    ],
+                    "index": 3.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        160,
+                        505,
+                        281
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                160,
+                                505,
+                                173
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        160,
+                                        505,
+                                        173
+                                    ],
+                                    "score": 1.0,
+                                    "content": "To construct tasks with structure that resembles that of human-specified labels, we need to group dat-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                171,
+                                505,
+                                183
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        171,
+                                        505,
+                                        183
+                                    ],
+                                    "score": 1.0,
+                                    "content": "apoints into consistent and distinct subsets based on salient features. With this motivation in mind,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                181,
+                                507,
+                                195
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        181,
+                                        182,
+                                        195
+                                    ],
+                                    "score": 1.0,
+                                    "content": "we propose to use",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        183,
+                                        183,
+                                        190,
+                                        192
+                                    ],
+                                    "score": 0.83,
+                                    "content": "k",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        190,
+                                        181,
+                                        358,
+                                        195
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-means clustering. Consider the partition",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        358,
+                                        182,
+                                        402,
+                                        194
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\mathcal { P } = \\{ \\mathcal { C } _ { c } \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        402,
+                                        181,
+                                        449,
+                                        195
+                                    ],
+                                    "score": 1.0,
+                                    "content": "learned by",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        449,
+                                        183,
+                                        456,
+                                        192
+                                    ],
+                                    "score": 0.83,
+                                    "content": "k",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        456,
+                                        181,
+                                        507,
+                                        195
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-means as a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                193,
+                                505,
+                                205
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        194,
+                                        284,
+                                        205
+                                    ],
+                                    "score": 1.0,
+                                    "content": "simplification of a Gaussian mixture model",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        284,
+                                        193,
+                                        328,
+                                        205
+                                    ],
+                                    "score": 0.92,
+                                    "content": "p ( \\mathbf { x } | c ) p ( \\bar { c } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        329,
+                                        194,
+                                        505,
+                                        205
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". If the clusters can recover a semblance of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                203,
+                                505,
+                                217
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        203,
+                                        303,
+                                        217
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the true class-conditional generative distributions",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        304,
+                                        205,
+                                        330,
+                                        216
+                                    ],
+                                    "score": 0.92,
+                                    "content": "p ( \\mathbf { x } | c )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        331,
+                                        203,
+                                        505,
+                                        217
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", creating tasks based on treating these clus-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                216,
+                                505,
+                                227
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        216,
+                                        459,
+                                        227
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ters as classes should result in useful unsupervised meta-training. However, the result of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        459,
+                                        216,
+                                        466,
+                                        225
+                                    ],
+                                    "score": 0.82,
+                                    "content": "k",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        466,
+                                        216,
+                                        505,
+                                        227
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-means is",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                225,
+                                505,
+                                240
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        225,
+                                        505,
+                                        240
+                                    ],
+                                    "score": 1.0,
+                                    "content": "critically dependent on the metric space on which its objective is defined. Clustering in pixel-space",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                236,
+                                506,
+                                251
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        236,
+                                        506,
+                                        251
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is unappealing for two reasons: (1) distance in pixel-space correlates poorly with semantic meaning,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                248,
+                                504,
+                                260
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        248,
+                                        504,
+                                        260
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and (2) the high dimensionality of raw images renders clustering difficult in practice. We empirically",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                258,
+                                505,
+                                272
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        258,
+                                        505,
+                                        272
+                                    ],
+                                    "score": 1.0,
+                                    "content": "show in Table 3 that meta-learning with tasks defined by pixel-space clusters, with preprocessing as",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                270,
+                                280,
+                                282
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        271,
+                                        193,
+                                        282
+                                    ],
+                                    "score": 1.0,
+                                    "content": "directed by Coates &",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        193,
+                                        270,
+                                        207,
+                                        281
+                                    ],
+                                    "score": 0.44,
+                                    "content": "\\mathrm { N g }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        208,
+                                        271,
+                                        280,
+                                        282
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(2012), also fails.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        }
+                    ],
+                    "index": 11
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        286,
+                        505,
+                        386
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                107,
+                                287,
+                                505,
+                                299
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        287,
+                                        505,
+                                        299
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We are now motivated to cluster in spaces in which common distance functions correlate to semantic",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                297,
+                                505,
+                                311
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        297,
+                                        505,
+                                        311
+                                    ],
+                                    "score": 1.0,
+                                    "content": "meaning. However, we must satisfy the constraints of our problem statement in the process of learn-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                309,
+                                505,
+                                321
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        309,
+                                        505,
+                                        321
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ing such spaces. To these ends, we use state-of-the-art unsupervised learning methods to produce",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                320,
+                                505,
+                                333
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        320,
+                                        505,
+                                        333
+                                    ],
+                                    "score": 1.0,
+                                    "content": "useful embedding spaces. For qualitative evidence in the unsupervised learning literature that such",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                330,
+                                505,
+                                343
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        330,
+                                        505,
+                                        343
+                                    ],
+                                    "score": 1.0,
+                                    "content": "embedding spaces exhibit semantic meaning, see Cheung et al. (2015); Bojanowski & Joulin (2017);",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                341,
+                                505,
+                                355
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        341,
+                                        505,
+                                        355
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Donahue et al. (2017). We note that while a given embedding space may not be directly suitable",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                352,
+                                505,
+                                366
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        352,
+                                        505,
+                                        366
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for highly-efficient learning of new tasks (which would require the embedding space to be precisely",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                364,
+                                505,
+                                375
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        364,
+                                        505,
+                                        375
+                                    ],
+                                    "score": 1.0,
+                                    "content": "aligned or adaptable to the classes of those tasks), we can still leverage it for the construction of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                375,
+                                365,
+                                387
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        375,
+                                        365,
+                                        387
+                                    ],
+                                    "score": 1.0,
+                                    "content": "structured tasks, a process for which requirements are less strict.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        }
+                    ],
+                    "index": 21
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        391,
+                        505,
+                        447
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                390,
+                                505,
+                                405
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        390,
+                                        431,
+                                        405
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Thus, we first run an out-of-the-box unsupervised embedding learning algorithm",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        431,
+                                        392,
+                                        439,
+                                        401
+                                    ],
+                                    "score": 0.76,
+                                    "content": "\\mathcal { E }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        439,
+                                        390,
+                                        452,
+                                        405
+                                    ],
+                                    "score": 1.0,
+                                    "content": "on",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        452,
+                                        392,
+                                        461,
+                                        401
+                                    ],
+                                    "score": 0.79,
+                                    "content": "\\mathcal { D }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        462,
+                                        390,
+                                        505,
+                                        405
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", then map",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                402,
+                                506,
+                                415
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        402,
+                                        140,
+                                        415
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the data",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        140,
+                                        402,
+                                        160,
+                                        414
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\{ { \\bf { x } } _ { i } \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        161,
+                                        402,
+                                        265,
+                                        415
+                                    ],
+                                    "score": 1.0,
+                                    "content": "into the embedding space",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        265,
+                                        403,
+                                        274,
+                                        412
+                                    ],
+                                    "score": 0.8,
+                                    "content": "\\mathcal { Z }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        274,
+                                        402,
+                                        321,
+                                        415
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", producing",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        321,
+                                        402,
+                                        340,
+                                        414
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\left\\{ \\mathbf { z } _ { i } \\right\\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        340,
+                                        402,
+                                        506,
+                                        415
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". To produce a diverse task set, we gener-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                413,
+                                506,
+                                426
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        413,
+                                        120,
+                                        426
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ate",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        121,
+                                        414,
+                                        130,
+                                        423
+                                    ],
+                                    "score": 0.82,
+                                    "content": "P",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        130,
+                                        413,
+                                        171,
+                                        426
+                                    ],
+                                    "score": 1.0,
+                                    "content": "partitions",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        172,
+                                        413,
+                                        194,
+                                        425
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\{ \\mathcal P _ { p } \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        194,
+                                        413,
+                                        283,
+                                        426
+                                    ],
+                                    "score": 1.0,
+                                    "content": "by running clustering",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        283,
+                                        414,
+                                        292,
+                                        423
+                                    ],
+                                    "score": 0.84,
+                                    "content": "P",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        293,
+                                        413,
+                                        506,
+                                        426
+                                    ],
+                                    "score": 1.0,
+                                    "content": "times, applying random scaling to the dimensions of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                107,
+                                424,
+                                505,
+                                437
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        425,
+                                        116,
+                                        434
+                                    ],
+                                    "score": 0.84,
+                                    "content": "\\mathcal { Z }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        116,
+                                        424,
+                                        505,
+                                        437
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to induce a different metric, represented by diagonal matrix A, for each run of clustering. With",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                435,
+                                495,
+                                448
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        437,
+                                        119,
+                                        447
+                                    ],
+                                    "score": 0.85,
+                                    "content": "\\pmb { \\mu } _ { c }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        119,
+                                        435,
+                                        279,
+                                        448
+                                    ],
+                                    "score": 1.0,
+                                    "content": "denoting the learned centroid of cluster",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        279,
+                                        436,
+                                        289,
+                                        446
+                                    ],
+                                    "score": 0.87,
+                                    "content": "\\mathcal { C } _ { c }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        290,
+                                        435,
+                                        495,
+                                        448
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", a single run of clustering can be summarized with",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        }
+                    ],
+                    "index": 28
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        222,
+                        453,
+                        388,
+                        489
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                222,
+                                453,
+                                388,
+                                489
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        222,
+                                        453,
+                                        388,
+                                        489
+                                    ],
+                                    "score": 0.95,
+                                    "content": "\\mathcal { P } , \\{ \\pmb { \\mu } _ { c } \\} = \\underset { \\{ \\mathcal { C } _ { c } \\} , \\{ \\pmb { \\mu } _ { c } \\} } { \\arg \\operatorname* { m i n } } \\sum _ { c = 1 } ^ { k } \\sum _ { \\mathbf { z } \\in \\mathcal { C } _ { c } } \\| \\mathbf { z } - \\pmb { \\mu } _ { c } \\| _ { \\mathbf { A } } ^ { 2 }",
+                                    "type": "interline_equation",
+                                    "image_path": "6cb4a83a12fefaed35c95d6609ab822a3a5d0ea1fb314437bd6dd6eda06f39ea.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 31.5,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                222,
+                                453,
+                                388,
+                                471.0
+                            ],
+                            "spans": [],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                222,
+                                471.0,
+                                388,
+                                489.0
+                            ],
+                            "spans": [],
+                            "index": 32
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        495,
+                        505,
+                        562
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                495,
+                                505,
+                                508
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        495,
+                                        505,
+                                        508
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We derive tasks for meta-learning from the partitions using the procedure detailed in Section 2.2,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                507,
+                                505,
+                                519
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        507,
+                                        505,
+                                        519
+                                    ],
+                                    "score": 1.0,
+                                    "content": "except we begin the construction of each task by sampling a partition from the uniform distribution",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                516,
+                                506,
+                                532
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        518,
+                                        130,
+                                        530
+                                    ],
+                                    "score": 0.9,
+                                    "content": "\\mathcal { U } ( \\mathcal { P } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        131,
+                                        516,
+                                        165,
+                                        532
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", and for",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        166,
+                                        518,
+                                        198,
+                                        529
+                                    ],
+                                    "score": 0.92,
+                                    "content": "{ \\bf x } _ { i } \\in \\mathcal { C } _ { c }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        199,
+                                        516,
+                                        234,
+                                        532
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", specify",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        234,
+                                        519,
+                                        262,
+                                        529
+                                    ],
+                                    "score": 0.87,
+                                    "content": "y _ { i } = c",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        262,
+                                        516,
+                                        506,
+                                        532
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". To avoid imbalanced clusters dominating the meta-training",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                528,
+                                505,
+                                541
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        529,
+                                        246,
+                                        541
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tasks, we opt not to sample from",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        247,
+                                        528,
+                                        297,
+                                        541
+                                    ],
+                                    "score": 0.92,
+                                    "content": "p ( c ) \\propto | \\mathcal { C } _ { c } |",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        297,
+                                        529,
+                                        382,
+                                        541
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", but instead sample",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        382,
+                                        529,
+                                        392,
+                                        538
+                                    ],
+                                    "score": 0.78,
+                                    "content": "N",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        393,
+                                        529,
+                                        505,
+                                        541
+                                    ],
+                                    "score": 1.0,
+                                    "content": "clusters uniformly without",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                539,
+                                505,
+                                552
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        539,
+                                        505,
+                                        552
+                                    ],
+                                    "score": 1.0,
+                                    "content": "replacement for each task. We note that Caron et al. (2018) are similarly motivated in their design",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                551,
+                                378,
+                                562
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        551,
+                                        378,
+                                        562
+                                    ],
+                                    "score": 1.0,
+                                    "content": "decision of sampling data from a uniform distribution over clusters.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        }
+                    ],
+                    "index": 35.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        567,
+                        505,
+                        634
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                567,
+                                506,
+                                581
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        567,
+                                        276,
+                                        581
+                                    ],
+                                    "score": 1.0,
+                                    "content": "With the partitions being constructed over",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        277,
+                                        567,
+                                        295,
+                                        579
+                                    ],
+                                    "score": 0.94,
+                                    "content": "\\left\\{ \\mathbf { z } _ { i } \\right\\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        296,
+                                        567,
+                                        506,
+                                        581
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", we have one more design decision to make: should",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                577,
+                                506,
+                                592
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        577,
+                                        506,
+                                        592
+                                    ],
+                                    "score": 1.0,
+                                    "content": "we perform meta-learning on embeddings or images? We consider that, to successfully solve new",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                590,
+                                505,
+                                601
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        590,
+                                        284,
+                                        601
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tasks at meta-test time, a learning procedure",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        285,
+                                        590,
+                                        294,
+                                        599
+                                    ],
+                                    "score": 0.83,
+                                    "content": "\\mathcal { F }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        294,
+                                        590,
+                                        505,
+                                        601
+                                    ],
+                                    "score": 1.0,
+                                    "content": "that takes embeddings as input would depend on the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                600,
+                                505,
+                                613
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        600,
+                                        505,
+                                        613
+                                    ],
+                                    "score": 1.0,
+                                    "content": "embedding function’s ability to generalize to out-of-distribution observations. On the other hand, by",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                612,
+                                505,
+                                623
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        612,
+                                        210,
+                                        623
+                                    ],
+                                    "score": 1.0,
+                                    "content": "meta-learning on images,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        210,
+                                        612,
+                                        220,
+                                        622
+                                    ],
+                                    "score": 0.83,
+                                    "content": "\\mathcal { F }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        221,
+                                        612,
+                                        305,
+                                        623
+                                    ],
+                                    "score": 1.0,
+                                    "content": "can separately adapt",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        305,
+                                        612,
+                                        312,
+                                        623
+                                    ],
+                                    "score": 0.85,
+                                    "content": "f",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        313,
+                                        612,
+                                        505,
+                                        623
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to each evaluation task from the rawest level of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                622,
+                                338,
+                                635
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        622,
+                                        338,
+                                        635
+                                    ],
+                                    "score": 1.0,
+                                    "content": "representation. Thus, we choose to meta-learn on images.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        }
+                    ],
+                    "index": 41.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        639,
+                        505,
+                        673
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                637,
+                                505,
+                                653
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        637,
+                                        505,
+                                        653
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We call our method clustering to automatically construct tasks for unsupervised meta-learning",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                650,
+                                505,
+                                662
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        650,
+                                        505,
+                                        662
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(CACTUs). We detail the task construction algorithm in Algorithm 1, and provide an illustration of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                660,
+                                430,
+                                674
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        660,
+                                        430,
+                                        674
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the complete unsupervised meta-learning approach for classification in Figure 1.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 47
+                        }
+                    ],
+                    "index": 46
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        687,
+                        211,
+                        700
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                686,
+                                213,
+                                703
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        686,
+                                        213,
+                                        703
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3 RELATED WORK",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 48
+                        }
+                    ],
+                    "index": 48
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        709,
+                        504,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                709,
+                                505,
+                                722
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        709,
+                                        505,
+                                        722
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The method we propose aims to address the unsupervised learning problem (Hastie et al., 2009; Le",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 49
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                720,
+                                506,
+                                733
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        720,
+                                        506,
+                                        733
+                                    ],
+                                    "score": 1.0,
+                                    "content": "et al., 2013), namely acquiring a transferable learning procedure without labels. We show that our",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 50
+                        }
+                    ],
+                    "index": 49.5
+                }
+            ],
+            "page_idx": 3,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        108,
+                        27,
+                        293,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                26,
+                                294,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        26,
+                                        294,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Published as a conference paper at ICLR 2019",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        302,
+                        752,
+                        308,
+                        759
+                    ],
+                    "lines": []
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        82,
+                        504,
+                        105
+                    ],
+                    "lines": [],
+                    "index": 0.5,
+                    "bbox_fs": [
+                        106,
+                        82,
+                        505,
+                        106
+                    ],
+                    "lines_deleted": true
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        110,
+                        504,
+                        154
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                110,
+                                506,
+                                123
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        110,
+                                        320,
+                                        123
+                                    ],
+                                    "score": 1.0,
+                                    "content": "A na¨ıve approach is to randomly partition the data",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        320,
+                                        111,
+                                        329,
+                                        120
+                                    ],
+                                    "score": 0.76,
+                                    "content": "\\mathcal { D }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        329,
+                                        110,
+                                        506,
+                                        123
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". While such a scheme introduces diverse",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                120,
+                                506,
+                                134
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        120,
+                                        506,
+                                        134
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tasks, there is no structure; that is, there is no consistency between a task’s training data and query",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                133,
+                                506,
+                                144
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        133,
+                                        506,
+                                        144
+                                    ],
+                                    "score": 1.0,
+                                    "content": "data, and hence nothing to be learned during each task, let alone across tasks. As seen in Table 3,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                144,
+                                435,
+                                156
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        144,
+                                        435,
+                                        156
+                                    ],
+                                    "score": 1.0,
+                                    "content": "providing a meta-learner with purely random tasks results in failed meta-learning.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        }
+                    ],
+                    "index": 3.5,
+                    "bbox_fs": [
+                        105,
+                        110,
+                        506,
+                        156
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        160,
+                        505,
+                        281
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                160,
+                                505,
+                                173
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        160,
+                                        505,
+                                        173
+                                    ],
+                                    "score": 1.0,
+                                    "content": "To construct tasks with structure that resembles that of human-specified labels, we need to group dat-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                171,
+                                505,
+                                183
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        171,
+                                        505,
+                                        183
+                                    ],
+                                    "score": 1.0,
+                                    "content": "apoints into consistent and distinct subsets based on salient features. With this motivation in mind,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                181,
+                                507,
+                                195
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        181,
+                                        182,
+                                        195
+                                    ],
+                                    "score": 1.0,
+                                    "content": "we propose to use",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        183,
+                                        183,
+                                        190,
+                                        192
+                                    ],
+                                    "score": 0.83,
+                                    "content": "k",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        190,
+                                        181,
+                                        358,
+                                        195
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-means clustering. Consider the partition",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        358,
+                                        182,
+                                        402,
+                                        194
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\mathcal { P } = \\{ \\mathcal { C } _ { c } \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        402,
+                                        181,
+                                        449,
+                                        195
+                                    ],
+                                    "score": 1.0,
+                                    "content": "learned by",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        449,
+                                        183,
+                                        456,
+                                        192
+                                    ],
+                                    "score": 0.83,
+                                    "content": "k",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        456,
+                                        181,
+                                        507,
+                                        195
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-means as a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                193,
+                                505,
+                                205
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        194,
+                                        284,
+                                        205
+                                    ],
+                                    "score": 1.0,
+                                    "content": "simplification of a Gaussian mixture model",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        284,
+                                        193,
+                                        328,
+                                        205
+                                    ],
+                                    "score": 0.92,
+                                    "content": "p ( \\mathbf { x } | c ) p ( \\bar { c } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        329,
+                                        194,
+                                        505,
+                                        205
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". If the clusters can recover a semblance of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                203,
+                                505,
+                                217
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        203,
+                                        303,
+                                        217
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the true class-conditional generative distributions",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        304,
+                                        205,
+                                        330,
+                                        216
+                                    ],
+                                    "score": 0.92,
+                                    "content": "p ( \\mathbf { x } | c )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        331,
+                                        203,
+                                        505,
+                                        217
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", creating tasks based on treating these clus-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                216,
+                                505,
+                                227
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        216,
+                                        459,
+                                        227
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ters as classes should result in useful unsupervised meta-training. However, the result of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        459,
+                                        216,
+                                        466,
+                                        225
+                                    ],
+                                    "score": 0.82,
+                                    "content": "k",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        466,
+                                        216,
+                                        505,
+                                        227
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-means is",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                225,
+                                505,
+                                240
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        225,
+                                        505,
+                                        240
+                                    ],
+                                    "score": 1.0,
+                                    "content": "critically dependent on the metric space on which its objective is defined. Clustering in pixel-space",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                236,
+                                506,
+                                251
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        236,
+                                        506,
+                                        251
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is unappealing for two reasons: (1) distance in pixel-space correlates poorly with semantic meaning,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                248,
+                                504,
+                                260
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        248,
+                                        504,
+                                        260
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and (2) the high dimensionality of raw images renders clustering difficult in practice. We empirically",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                258,
+                                505,
+                                272
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        258,
+                                        505,
+                                        272
+                                    ],
+                                    "score": 1.0,
+                                    "content": "show in Table 3 that meta-learning with tasks defined by pixel-space clusters, with preprocessing as",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                270,
+                                280,
+                                282
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        271,
+                                        193,
+                                        282
+                                    ],
+                                    "score": 1.0,
+                                    "content": "directed by Coates &",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        193,
+                                        270,
+                                        207,
+                                        281
+                                    ],
+                                    "score": 0.44,
+                                    "content": "\\mathrm { N g }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        208,
+                                        271,
+                                        280,
+                                        282
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(2012), also fails.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        }
+                    ],
+                    "index": 11,
+                    "bbox_fs": [
+                        105,
+                        160,
+                        507,
+                        282
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        286,
+                        505,
+                        386
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                107,
+                                287,
+                                505,
+                                299
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        287,
+                                        505,
+                                        299
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We are now motivated to cluster in spaces in which common distance functions correlate to semantic",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                297,
+                                505,
+                                311
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        297,
+                                        505,
+                                        311
+                                    ],
+                                    "score": 1.0,
+                                    "content": "meaning. However, we must satisfy the constraints of our problem statement in the process of learn-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                309,
+                                505,
+                                321
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        309,
+                                        505,
+                                        321
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ing such spaces. To these ends, we use state-of-the-art unsupervised learning methods to produce",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                320,
+                                505,
+                                333
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        320,
+                                        505,
+                                        333
+                                    ],
+                                    "score": 1.0,
+                                    "content": "useful embedding spaces. For qualitative evidence in the unsupervised learning literature that such",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                330,
+                                505,
+                                343
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        330,
+                                        505,
+                                        343
+                                    ],
+                                    "score": 1.0,
+                                    "content": "embedding spaces exhibit semantic meaning, see Cheung et al. (2015); Bojanowski & Joulin (2017);",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                341,
+                                505,
+                                355
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        341,
+                                        505,
+                                        355
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Donahue et al. (2017). We note that while a given embedding space may not be directly suitable",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                352,
+                                505,
+                                366
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        352,
+                                        505,
+                                        366
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for highly-efficient learning of new tasks (which would require the embedding space to be precisely",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                364,
+                                505,
+                                375
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        364,
+                                        505,
+                                        375
+                                    ],
+                                    "score": 1.0,
+                                    "content": "aligned or adaptable to the classes of those tasks), we can still leverage it for the construction of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                375,
+                                365,
+                                387
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        375,
+                                        365,
+                                        387
+                                    ],
+                                    "score": 1.0,
+                                    "content": "structured tasks, a process for which requirements are less strict.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        }
+                    ],
+                    "index": 21,
+                    "bbox_fs": [
+                        105,
+                        287,
+                        505,
+                        387
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        391,
+                        505,
+                        447
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                390,
+                                505,
+                                405
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        390,
+                                        431,
+                                        405
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Thus, we first run an out-of-the-box unsupervised embedding learning algorithm",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        431,
+                                        392,
+                                        439,
+                                        401
+                                    ],
+                                    "score": 0.76,
+                                    "content": "\\mathcal { E }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        439,
+                                        390,
+                                        452,
+                                        405
+                                    ],
+                                    "score": 1.0,
+                                    "content": "on",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        452,
+                                        392,
+                                        461,
+                                        401
+                                    ],
+                                    "score": 0.79,
+                                    "content": "\\mathcal { D }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        462,
+                                        390,
+                                        505,
+                                        405
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", then map",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                402,
+                                506,
+                                415
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        402,
+                                        140,
+                                        415
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the data",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        140,
+                                        402,
+                                        160,
+                                        414
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\{ { \\bf { x } } _ { i } \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        161,
+                                        402,
+                                        265,
+                                        415
+                                    ],
+                                    "score": 1.0,
+                                    "content": "into the embedding space",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        265,
+                                        403,
+                                        274,
+                                        412
+                                    ],
+                                    "score": 0.8,
+                                    "content": "\\mathcal { Z }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        274,
+                                        402,
+                                        321,
+                                        415
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", producing",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        321,
+                                        402,
+                                        340,
+                                        414
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\left\\{ \\mathbf { z } _ { i } \\right\\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        340,
+                                        402,
+                                        506,
+                                        415
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". To produce a diverse task set, we gener-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                413,
+                                506,
+                                426
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        413,
+                                        120,
+                                        426
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ate",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        121,
+                                        414,
+                                        130,
+                                        423
+                                    ],
+                                    "score": 0.82,
+                                    "content": "P",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        130,
+                                        413,
+                                        171,
+                                        426
+                                    ],
+                                    "score": 1.0,
+                                    "content": "partitions",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        172,
+                                        413,
+                                        194,
+                                        425
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\{ \\mathcal P _ { p } \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        194,
+                                        413,
+                                        283,
+                                        426
+                                    ],
+                                    "score": 1.0,
+                                    "content": "by running clustering",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        283,
+                                        414,
+                                        292,
+                                        423
+                                    ],
+                                    "score": 0.84,
+                                    "content": "P",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        293,
+                                        413,
+                                        506,
+                                        426
+                                    ],
+                                    "score": 1.0,
+                                    "content": "times, applying random scaling to the dimensions of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                107,
+                                424,
+                                505,
+                                437
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        425,
+                                        116,
+                                        434
+                                    ],
+                                    "score": 0.84,
+                                    "content": "\\mathcal { Z }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        116,
+                                        424,
+                                        505,
+                                        437
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to induce a different metric, represented by diagonal matrix A, for each run of clustering. With",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                435,
+                                495,
+                                448
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        437,
+                                        119,
+                                        447
+                                    ],
+                                    "score": 0.85,
+                                    "content": "\\pmb { \\mu } _ { c }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        119,
+                                        435,
+                                        279,
+                                        448
+                                    ],
+                                    "score": 1.0,
+                                    "content": "denoting the learned centroid of cluster",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        279,
+                                        436,
+                                        289,
+                                        446
+                                    ],
+                                    "score": 0.87,
+                                    "content": "\\mathcal { C } _ { c }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        290,
+                                        435,
+                                        495,
+                                        448
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", a single run of clustering can be summarized with",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        }
+                    ],
+                    "index": 28,
+                    "bbox_fs": [
+                        105,
+                        390,
+                        506,
+                        448
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        222,
+                        453,
+                        388,
+                        489
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                222,
+                                453,
+                                388,
+                                489
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        222,
+                                        453,
+                                        388,
+                                        489
+                                    ],
+                                    "score": 0.95,
+                                    "content": "\\mathcal { P } , \\{ \\pmb { \\mu } _ { c } \\} = \\underset { \\{ \\mathcal { C } _ { c } \\} , \\{ \\pmb { \\mu } _ { c } \\} } { \\arg \\operatorname* { m i n } } \\sum _ { c = 1 } ^ { k } \\sum _ { \\mathbf { z } \\in \\mathcal { C } _ { c } } \\| \\mathbf { z } - \\pmb { \\mu } _ { c } \\| _ { \\mathbf { A } } ^ { 2 }",
+                                    "type": "interline_equation",
+                                    "image_path": "6cb4a83a12fefaed35c95d6609ab822a3a5d0ea1fb314437bd6dd6eda06f39ea.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 31.5,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                222,
+                                453,
+                                388,
+                                471.0
+                            ],
+                            "spans": [],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                222,
+                                471.0,
+                                388,
+                                489.0
+                            ],
+                            "spans": [],
+                            "index": 32
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        495,
+                        505,
+                        562
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                495,
+                                505,
+                                508
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        495,
+                                        505,
+                                        508
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We derive tasks for meta-learning from the partitions using the procedure detailed in Section 2.2,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                507,
+                                505,
+                                519
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        507,
+                                        505,
+                                        519
+                                    ],
+                                    "score": 1.0,
+                                    "content": "except we begin the construction of each task by sampling a partition from the uniform distribution",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                516,
+                                506,
+                                532
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        518,
+                                        130,
+                                        530
+                                    ],
+                                    "score": 0.9,
+                                    "content": "\\mathcal { U } ( \\mathcal { P } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        131,
+                                        516,
+                                        165,
+                                        532
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", and for",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        166,
+                                        518,
+                                        198,
+                                        529
+                                    ],
+                                    "score": 0.92,
+                                    "content": "{ \\bf x } _ { i } \\in \\mathcal { C } _ { c }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        199,
+                                        516,
+                                        234,
+                                        532
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", specify",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        234,
+                                        519,
+                                        262,
+                                        529
+                                    ],
+                                    "score": 0.87,
+                                    "content": "y _ { i } = c",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        262,
+                                        516,
+                                        506,
+                                        532
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". To avoid imbalanced clusters dominating the meta-training",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                528,
+                                505,
+                                541
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        529,
+                                        246,
+                                        541
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tasks, we opt not to sample from",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        247,
+                                        528,
+                                        297,
+                                        541
+                                    ],
+                                    "score": 0.92,
+                                    "content": "p ( c ) \\propto | \\mathcal { C } _ { c } |",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        297,
+                                        529,
+                                        382,
+                                        541
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", but instead sample",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        382,
+                                        529,
+                                        392,
+                                        538
+                                    ],
+                                    "score": 0.78,
+                                    "content": "N",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        393,
+                                        529,
+                                        505,
+                                        541
+                                    ],
+                                    "score": 1.0,
+                                    "content": "clusters uniformly without",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                539,
+                                505,
+                                552
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        539,
+                                        505,
+                                        552
+                                    ],
+                                    "score": 1.0,
+                                    "content": "replacement for each task. We note that Caron et al. (2018) are similarly motivated in their design",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                551,
+                                378,
+                                562
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        551,
+                                        378,
+                                        562
+                                    ],
+                                    "score": 1.0,
+                                    "content": "decision of sampling data from a uniform distribution over clusters.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        }
+                    ],
+                    "index": 35.5,
+                    "bbox_fs": [
+                        105,
+                        495,
+                        506,
+                        562
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        567,
+                        505,
+                        634
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                567,
+                                506,
+                                581
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        567,
+                                        276,
+                                        581
+                                    ],
+                                    "score": 1.0,
+                                    "content": "With the partitions being constructed over",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        277,
+                                        567,
+                                        295,
+                                        579
+                                    ],
+                                    "score": 0.94,
+                                    "content": "\\left\\{ \\mathbf { z } _ { i } \\right\\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        296,
+                                        567,
+                                        506,
+                                        581
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", we have one more design decision to make: should",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                577,
+                                506,
+                                592
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        577,
+                                        506,
+                                        592
+                                    ],
+                                    "score": 1.0,
+                                    "content": "we perform meta-learning on embeddings or images? We consider that, to successfully solve new",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                590,
+                                505,
+                                601
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        590,
+                                        284,
+                                        601
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tasks at meta-test time, a learning procedure",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        285,
+                                        590,
+                                        294,
+                                        599
+                                    ],
+                                    "score": 0.83,
+                                    "content": "\\mathcal { F }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        294,
+                                        590,
+                                        505,
+                                        601
+                                    ],
+                                    "score": 1.0,
+                                    "content": "that takes embeddings as input would depend on the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                600,
+                                505,
+                                613
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        600,
+                                        505,
+                                        613
+                                    ],
+                                    "score": 1.0,
+                                    "content": "embedding function’s ability to generalize to out-of-distribution observations. On the other hand, by",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                612,
+                                505,
+                                623
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        612,
+                                        210,
+                                        623
+                                    ],
+                                    "score": 1.0,
+                                    "content": "meta-learning on images,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        210,
+                                        612,
+                                        220,
+                                        622
+                                    ],
+                                    "score": 0.83,
+                                    "content": "\\mathcal { F }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        221,
+                                        612,
+                                        305,
+                                        623
+                                    ],
+                                    "score": 1.0,
+                                    "content": "can separately adapt",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        305,
+                                        612,
+                                        312,
+                                        623
+                                    ],
+                                    "score": 0.85,
+                                    "content": "f",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        313,
+                                        612,
+                                        505,
+                                        623
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to each evaluation task from the rawest level of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                622,
+                                338,
+                                635
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        622,
+                                        338,
+                                        635
+                                    ],
+                                    "score": 1.0,
+                                    "content": "representation. Thus, we choose to meta-learn on images.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        }
+                    ],
+                    "index": 41.5,
+                    "bbox_fs": [
+                        105,
+                        567,
+                        506,
+                        635
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        639,
+                        505,
+                        673
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                637,
+                                505,
+                                653
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        637,
+                                        505,
+                                        653
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We call our method clustering to automatically construct tasks for unsupervised meta-learning",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                650,
+                                505,
+                                662
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        650,
+                                        505,
+                                        662
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(CACTUs). We detail the task construction algorithm in Algorithm 1, and provide an illustration of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                660,
+                                430,
+                                674
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        660,
+                                        430,
+                                        674
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the complete unsupervised meta-learning approach for classification in Figure 1.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 47
+                        }
+                    ],
+                    "index": 46,
+                    "bbox_fs": [
+                        105,
+                        637,
+                        505,
+                        674
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        687,
+                        211,
+                        700
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                686,
+                                213,
+                                703
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        686,
+                                        213,
+                                        703
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3 RELATED WORK",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 48
+                        }
+                    ],
+                    "index": 48
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        709,
+                        504,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                709,
+                                505,
+                                722
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        709,
+                                        505,
+                                        722
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The method we propose aims to address the unsupervised learning problem (Hastie et al., 2009; Le",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 49
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                720,
+                                506,
+                                733
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        720,
+                                        506,
+                                        733
+                                    ],
+                                    "score": 1.0,
+                                    "content": "et al., 2013), namely acquiring a transferable learning procedure without labels. We show that our",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 50
+                        }
+                    ],
+                    "index": 49.5,
+                    "bbox_fs": [
+                        106,
+                        709,
+                        506,
+                        733
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        82,
+                        268,
+                        94
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                81,
+                                270,
+                                95
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        81,
+                                        270,
+                                        95
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Algorithm 1 CACTUs for classification",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        }
+                    ],
+                    "index": 0
+                },
+                {
+                    "type": "table",
+                    "bbox": [
+                        108,
+                        96,
+                        504,
+                        225
+                    ],
+                    "blocks": [
+                        {
+                            "type": "table_body",
+                            "bbox": [
+                                108,
+                                96,
+                                504,
+                                225
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        108,
+                                        96,
+                                        504,
+                                        225
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                108,
+                                                96,
+                                                504,
+                                                225
+                                            ],
+                                            "score": 0.879,
+                                            "html": "<table><tr><td colspan=\"2\">1: procedure CACTUs(ε,D,P,k,T,N,Km-tr, Q)</td><td colspan=\"4\"></td></tr><tr><td>2:</td><td colspan=\"4\">Run embedding learning algorithm ε on D and produce embeddings {zi} from observations {xi}.</td></tr><tr><td>3:</td><td colspan=\"4\">Run k-means on {zi} P times (with random scaling) to generate a set of partitions {Pp = {Cc}p}.</td></tr><tr><td>4:</td><td colspan=\"4\">for t from 1 to the number of desired tasks T do</td></tr><tr><td>5:</td><td colspan=\"4\">Sample a partition P uniformly at random from the set of partitions {Pp}.</td></tr><tr><td>6:</td><td colspan=\"4\">Sample a cluster Cn uniformly without replacement from P for each of the N classes desired for a task.</td></tr><tr><td>7:</td><td colspan=\"4\">Sample an embedding Zr without replacement from Cn for each of the R= Km-tr+Q training and</td></tr><tr><td>8:</td><td colspan=\"4\">query examples desired for each class, and record the corresponding datapoint Xn,r. Sample a permutation(ln) of N one-hot labels.</td></tr><tr><td>9:</td><td colspan=\"4\">Construct Tt ={(xn,r,ln)}.</td></tr><tr><td>10:</td><td colspan=\"4\">return {Tt}</td></tr></table>",
+                                            "type": "table",
+                                            "image_path": "3c47f37c28699f9f6647b45003f86db9df2ebf5ba64e6abf56db7f4740de6eb0.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 2,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        108,
+                                        96,
+                                        504,
+                                        139.0
+                                    ],
+                                    "spans": [],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        108,
+                                        139.0,
+                                        504,
+                                        182.0
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                },
+                                {
+                                    "bbox": [
+                                        108,
+                                        182.0,
+                                        504,
+                                        225.0
+                                    ],
+                                    "spans": [],
+                                    "index": 3
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 2
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        260,
+                        505,
+                        512
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                260,
+                                505,
+                                272
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        260,
+                                        505,
+                                        272
+                                    ],
+                                    "score": 1.0,
+                                    "content": "method is complementary to a number of unsupervised learning methods, including ACAI (Berth-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                270,
+                                505,
+                                283
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        270,
+                                        505,
+                                        283
+                                    ],
+                                    "score": 1.0,
+                                    "content": "elot et al., 2018), BiGAN (Donahue et al., 2017; Dumoulin et al., 2017), DeepCluster (Caron et al.,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                281,
+                                505,
+                                294
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        281,
+                                        505,
+                                        294
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2018), and InfoGAN (Chen et al., 2016): we leverage these prior methods to learn embeddings",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                293,
+                                505,
+                                305
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        293,
+                                        505,
+                                        305
+                                    ],
+                                    "score": 1.0,
+                                    "content": "used for constructing meta-learning tasks, and demonstrate that our method learns a more useful",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                303,
+                                505,
+                                317
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        303,
+                                        505,
+                                        317
+                                    ],
+                                    "score": 1.0,
+                                    "content": "representation than the embeddings. The ability to use what was learned during unsupervised pre-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                315,
+                                506,
+                                328
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        315,
+                                        506,
+                                        328
+                                    ],
+                                    "score": 1.0,
+                                    "content": "training to better or more efficiently learn a variety of downstream tasks is arguably one of the most",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                326,
+                                506,
+                                339
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        326,
+                                        506,
+                                        339
+                                    ],
+                                    "score": 1.0,
+                                    "content": "practical applications of unsupervised learning methods, which has a long history in neural network",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                337,
+                                505,
+                                349
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        337,
+                                        505,
+                                        349
+                                    ],
+                                    "score": 1.0,
+                                    "content": "training (Hinton et al., 2006; Bengio et al., 2007; Ranzato et al., 2006; Vincent et al., 2008; Erhan",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                348,
+                                505,
+                                360
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        348,
+                                        505,
+                                        360
+                                    ],
+                                    "score": 1.0,
+                                    "content": "et al., 2010). Unsupervised pre-training has demonstrated success in a number of domains, includ-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                358,
+                                505,
+                                371
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        358,
+                                        505,
+                                        371
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ing speech recognition (Yu et al., 2010), image classification (Zhang et al., 2017), machine transla-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                369,
+                                505,
+                                382
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        369,
+                                        505,
+                                        382
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tion (Ramachandran et al., 2017), and text classification (Dai & Le, 2015; Howard & Ruder, 2018;",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                379,
+                                506,
+                                394
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        379,
+                                        506,
+                                        394
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Radford et al., 2018). Our approach, unsupervised meta-learning, can be viewed as an unsupervised",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                392,
+                                506,
+                                404
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        392,
+                                        506,
+                                        404
+                                    ],
+                                    "score": 1.0,
+                                    "content": "learning algorithm that explicitly optimizes for few-shot transferability. As a result, we can expect it",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                403,
+                                506,
+                                415
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        403,
+                                        506,
+                                        415
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to better learn human-specified downstream tasks, compared to unsupervised learning methods that",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                414,
+                                505,
+                                426
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        414,
+                                        505,
+                                        426
+                                    ],
+                                    "score": 1.0,
+                                    "content": "optimize for other metrics, such as reconstruction (Vincent et al., 2010; Higgins et al., 2017), fidelity",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                424,
+                                505,
+                                437
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        424,
+                                        505,
+                                        437
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of constructed images (Radford et al., 2016; Salimans et al., 2016; Donahue et al., 2017; Dumoulin",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                435,
+                                505,
+                                448
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        435,
+                                        505,
+                                        448
+                                    ],
+                                    "score": 1.0,
+                                    "content": "et al., 2017), representation interpolation (Berthelot et al., 2018), disentanglement (Bengio et al.,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                445,
+                                506,
+                                459
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        445,
+                                        506,
+                                        459
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2013; Reed et al., 2014; Cheung et al., 2015; Chen et al., 2016; Mathieu et al., 2016; Denton &",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                456,
+                                506,
+                                470
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        456,
+                                        506,
+                                        470
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Birodkar, 2017), and clustering (Coates & Ng, 2012; Krahenb ¨ uhl et al., 2016; Bojanowski & Joulin, ¨",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                468,
+                                505,
+                                480
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        468,
+                                        505,
+                                        480
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2017; Caron et al., 2018). We empirically evaluate this hypothesis in the next section. In contrast to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                480,
+                                505,
+                                491
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        480,
+                                        505,
+                                        491
+                                    ],
+                                    "score": 1.0,
+                                    "content": "many previous evaluations of unsupervised pre-training, we focus on settings in which only a small",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                490,
+                                505,
+                                503
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        490,
+                                        505,
+                                        503
+                                    ],
+                                    "score": 1.0,
+                                    "content": "amount of data for the downstream tasks is available, since this is where the unlabeled data can be",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                501,
+                                180,
+                                513
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        501,
+                                        180,
+                                        513
+                                    ],
+                                    "score": 1.0,
+                                    "content": "maximally useful.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        }
+                    ],
+                    "index": 15
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        518,
+                        505,
+                        606
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                518,
+                                505,
+                                530
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        518,
+                                        505,
+                                        530
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Unsupervised pre-training followed by supervised learning can be viewed as a special case of the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                529,
+                                506,
+                                541
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        529,
+                                        506,
+                                        541
+                                    ],
+                                    "score": 1.0,
+                                    "content": "semi-supervised learning problem (Zhu, 2011; Kingma et al., 2014; Rasmus et al., 2015; Oliver",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                540,
+                                505,
+                                552
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        540,
+                                        505,
+                                        552
+                                    ],
+                                    "score": 1.0,
+                                    "content": "et al., 2018). However, in contrast to our problem statement, semi-supervised learning methods",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                551,
+                                505,
+                                563
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        551,
+                                        505,
+                                        563
+                                    ],
+                                    "score": 1.0,
+                                    "content": "assume that a significant proportion of the unlabeled data, if not all of it, shares underlying labels",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                561,
+                                505,
+                                573
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        561,
+                                        505,
+                                        573
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with the labeled data. Additionally, our approach and other unsupervised learning methods are well-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                572,
+                                505,
+                                586
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        572,
+                                        505,
+                                        586
+                                    ],
+                                    "score": 1.0,
+                                    "content": "suited for transferring their learned representation to many possible downstream tasks or labelings,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                584,
+                                505,
+                                596
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        584,
+                                        505,
+                                        596
+                                    ],
+                                    "score": 1.0,
+                                    "content": "whereas semi-supervised learning methods typically optimize for performance on a single task, with",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                595,
+                                262,
+                                607
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        595,
+                                        262,
+                                        607
+                                    ],
+                                    "score": 1.0,
+                                    "content": "respect to a single labeling of the data.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        }
+                    ],
+                    "index": 30.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        611,
+                        505,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                612,
+                                506,
+                                624
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        612,
+                                        506,
+                                        624
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Our method builds upon the ideas of meta-learning (Schmidhuber, 1987; Bengio et al., 1991; Naik",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                622,
+                                505,
+                                635
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        622,
+                                        505,
+                                        635
+                                    ],
+                                    "score": 1.0,
+                                    "content": "& Mammone, 1992) and few-shot learning (Santoro et al., 2016; Vinyals et al., 2016; Ravi &",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                633,
+                                505,
+                                646
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        633,
+                                        505,
+                                        646
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Larochelle, 2017; Munkhdalai & Yu, 2017; Snell et al., 2017). We apply two meta-learning al-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                644,
+                                506,
+                                658
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        644,
+                                        506,
+                                        658
+                                    ],
+                                    "score": 1.0,
+                                    "content": "gorithms, model-agnostic meta-learning (Finn et al., 2017) and prototypical networks (Snell et al.,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                655,
+                                505,
+                                669
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        655,
+                                        505,
+                                        669
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2017), to tasks constructed in an unsupervised manner. Similar to our problem setting, some prior",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                666,
+                                505,
+                                680
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        666,
+                                        505,
+                                        680
+                                    ],
+                                    "score": 1.0,
+                                    "content": "works have aimed to learn an unsupervised learning procedure with supervised data (Garg & Kalai,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                677,
+                                505,
+                                690
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        677,
+                                        505,
+                                        690
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2017; Metz et al., 2018). Instead, we consider a problem setting that is entirely unsupervised, aim-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                688,
+                                505,
+                                701
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        688,
+                                        505,
+                                        701
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ing to learn efficient learning algorithms using unlabeled datasets. Our problem setting is similar to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                699,
+                                505,
+                                712
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        699,
+                                        505,
+                                        712
+                                    ],
+                                    "score": 1.0,
+                                    "content": "that considered by Gupta et al. (2018), but we develop an approach that is suitable for supervised",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                711,
+                                505,
+                                723
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        711,
+                                        505,
+                                        723
+                                    ],
+                                    "score": 1.0,
+                                    "content": "downstream tasks, rather than reinforcement learning problems, and demonstrate our algorithm on",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                721,
+                                320,
+                                734
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        721,
+                                        320,
+                                        734
+                                    ],
+                                    "score": 1.0,
+                                    "content": "problems with high-dimensional visual observations.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        }
+                    ],
+                    "index": 40
+                }
+            ],
+            "page_idx": 4,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        108,
+                        27,
+                        293,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                26,
+                                294,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        26,
+                                        294,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Published as a conference paper at ICLR 2019",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        302,
+                        751,
+                        308,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                302,
+                                750,
+                                309,
+                                763
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        302,
+                                        750,
+                                        309,
+                                        763
+                                    ],
+                                    "score": 1.0,
+                                    "content": "5",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        82,
+                        268,
+                        94
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                81,
+                                270,
+                                95
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        81,
+                                        270,
+                                        95
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Algorithm 1 CACTUs for classification",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        }
+                    ],
+                    "index": 0
+                },
+                {
+                    "type": "table",
+                    "bbox": [
+                        108,
+                        96,
+                        504,
+                        225
+                    ],
+                    "blocks": [
+                        {
+                            "type": "table_body",
+                            "bbox": [
+                                108,
+                                96,
+                                504,
+                                225
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        108,
+                                        96,
+                                        504,
+                                        225
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                108,
+                                                96,
+                                                504,
+                                                225
+                                            ],
+                                            "score": 0.879,
+                                            "html": "<table><tr><td colspan=\"2\">1: procedure CACTUs(ε,D,P,k,T,N,Km-tr, Q)</td><td colspan=\"4\"></td></tr><tr><td>2:</td><td colspan=\"4\">Run embedding learning algorithm ε on D and produce embeddings {zi} from observations {xi}.</td></tr><tr><td>3:</td><td colspan=\"4\">Run k-means on {zi} P times (with random scaling) to generate a set of partitions {Pp = {Cc}p}.</td></tr><tr><td>4:</td><td colspan=\"4\">for t from 1 to the number of desired tasks T do</td></tr><tr><td>5:</td><td colspan=\"4\">Sample a partition P uniformly at random from the set of partitions {Pp}.</td></tr><tr><td>6:</td><td colspan=\"4\">Sample a cluster Cn uniformly without replacement from P for each of the N classes desired for a task.</td></tr><tr><td>7:</td><td colspan=\"4\">Sample an embedding Zr without replacement from Cn for each of the R= Km-tr+Q training and</td></tr><tr><td>8:</td><td colspan=\"4\">query examples desired for each class, and record the corresponding datapoint Xn,r. Sample a permutation(ln) of N one-hot labels.</td></tr><tr><td>9:</td><td colspan=\"4\">Construct Tt ={(xn,r,ln)}.</td></tr><tr><td>10:</td><td colspan=\"4\">return {Tt}</td></tr></table>",
+                                            "type": "table",
+                                            "image_path": "3c47f37c28699f9f6647b45003f86db9df2ebf5ba64e6abf56db7f4740de6eb0.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 2,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        108,
+                                        96,
+                                        504,
+                                        139.0
+                                    ],
+                                    "spans": [],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        108,
+                                        139.0,
+                                        504,
+                                        182.0
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                },
+                                {
+                                    "bbox": [
+                                        108,
+                                        182.0,
+                                        504,
+                                        225.0
+                                    ],
+                                    "spans": [],
+                                    "index": 3
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 2
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        260,
+                        505,
+                        512
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                260,
+                                505,
+                                272
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        260,
+                                        505,
+                                        272
+                                    ],
+                                    "score": 1.0,
+                                    "content": "method is complementary to a number of unsupervised learning methods, including ACAI (Berth-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                270,
+                                505,
+                                283
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        270,
+                                        505,
+                                        283
+                                    ],
+                                    "score": 1.0,
+                                    "content": "elot et al., 2018), BiGAN (Donahue et al., 2017; Dumoulin et al., 2017), DeepCluster (Caron et al.,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                281,
+                                505,
+                                294
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        281,
+                                        505,
+                                        294
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2018), and InfoGAN (Chen et al., 2016): we leverage these prior methods to learn embeddings",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                293,
+                                505,
+                                305
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        293,
+                                        505,
+                                        305
+                                    ],
+                                    "score": 1.0,
+                                    "content": "used for constructing meta-learning tasks, and demonstrate that our method learns a more useful",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                303,
+                                505,
+                                317
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        303,
+                                        505,
+                                        317
+                                    ],
+                                    "score": 1.0,
+                                    "content": "representation than the embeddings. The ability to use what was learned during unsupervised pre-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                315,
+                                506,
+                                328
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        315,
+                                        506,
+                                        328
+                                    ],
+                                    "score": 1.0,
+                                    "content": "training to better or more efficiently learn a variety of downstream tasks is arguably one of the most",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                326,
+                                506,
+                                339
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        326,
+                                        506,
+                                        339
+                                    ],
+                                    "score": 1.0,
+                                    "content": "practical applications of unsupervised learning methods, which has a long history in neural network",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                337,
+                                505,
+                                349
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        337,
+                                        505,
+                                        349
+                                    ],
+                                    "score": 1.0,
+                                    "content": "training (Hinton et al., 2006; Bengio et al., 2007; Ranzato et al., 2006; Vincent et al., 2008; Erhan",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                348,
+                                505,
+                                360
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        348,
+                                        505,
+                                        360
+                                    ],
+                                    "score": 1.0,
+                                    "content": "et al., 2010). Unsupervised pre-training has demonstrated success in a number of domains, includ-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                358,
+                                505,
+                                371
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        358,
+                                        505,
+                                        371
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ing speech recognition (Yu et al., 2010), image classification (Zhang et al., 2017), machine transla-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                369,
+                                505,
+                                382
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        369,
+                                        505,
+                                        382
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tion (Ramachandran et al., 2017), and text classification (Dai & Le, 2015; Howard & Ruder, 2018;",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                379,
+                                506,
+                                394
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        379,
+                                        506,
+                                        394
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Radford et al., 2018). Our approach, unsupervised meta-learning, can be viewed as an unsupervised",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                392,
+                                506,
+                                404
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        392,
+                                        506,
+                                        404
+                                    ],
+                                    "score": 1.0,
+                                    "content": "learning algorithm that explicitly optimizes for few-shot transferability. As a result, we can expect it",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                403,
+                                506,
+                                415
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        403,
+                                        506,
+                                        415
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to better learn human-specified downstream tasks, compared to unsupervised learning methods that",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                414,
+                                505,
+                                426
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        414,
+                                        505,
+                                        426
+                                    ],
+                                    "score": 1.0,
+                                    "content": "optimize for other metrics, such as reconstruction (Vincent et al., 2010; Higgins et al., 2017), fidelity",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                424,
+                                505,
+                                437
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        424,
+                                        505,
+                                        437
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of constructed images (Radford et al., 2016; Salimans et al., 2016; Donahue et al., 2017; Dumoulin",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                435,
+                                505,
+                                448
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        435,
+                                        505,
+                                        448
+                                    ],
+                                    "score": 1.0,
+                                    "content": "et al., 2017), representation interpolation (Berthelot et al., 2018), disentanglement (Bengio et al.,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                445,
+                                506,
+                                459
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        445,
+                                        506,
+                                        459
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2013; Reed et al., 2014; Cheung et al., 2015; Chen et al., 2016; Mathieu et al., 2016; Denton &",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                456,
+                                506,
+                                470
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        456,
+                                        506,
+                                        470
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Birodkar, 2017), and clustering (Coates & Ng, 2012; Krahenb ¨ uhl et al., 2016; Bojanowski & Joulin, ¨",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                468,
+                                505,
+                                480
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        468,
+                                        505,
+                                        480
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2017; Caron et al., 2018). We empirically evaluate this hypothesis in the next section. In contrast to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                480,
+                                505,
+                                491
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        480,
+                                        505,
+                                        491
+                                    ],
+                                    "score": 1.0,
+                                    "content": "many previous evaluations of unsupervised pre-training, we focus on settings in which only a small",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                490,
+                                505,
+                                503
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        490,
+                                        505,
+                                        503
+                                    ],
+                                    "score": 1.0,
+                                    "content": "amount of data for the downstream tasks is available, since this is where the unlabeled data can be",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                501,
+                                180,
+                                513
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        501,
+                                        180,
+                                        513
+                                    ],
+                                    "score": 1.0,
+                                    "content": "maximally useful.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        }
+                    ],
+                    "index": 15,
+                    "bbox_fs": [
+                        104,
+                        260,
+                        506,
+                        513
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        518,
+                        505,
+                        606
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                518,
+                                505,
+                                530
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        518,
+                                        505,
+                                        530
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Unsupervised pre-training followed by supervised learning can be viewed as a special case of the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                529,
+                                506,
+                                541
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        529,
+                                        506,
+                                        541
+                                    ],
+                                    "score": 1.0,
+                                    "content": "semi-supervised learning problem (Zhu, 2011; Kingma et al., 2014; Rasmus et al., 2015; Oliver",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                540,
+                                505,
+                                552
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        540,
+                                        505,
+                                        552
+                                    ],
+                                    "score": 1.0,
+                                    "content": "et al., 2018). However, in contrast to our problem statement, semi-supervised learning methods",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                551,
+                                505,
+                                563
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        551,
+                                        505,
+                                        563
+                                    ],
+                                    "score": 1.0,
+                                    "content": "assume that a significant proportion of the unlabeled data, if not all of it, shares underlying labels",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                561,
+                                505,
+                                573
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        561,
+                                        505,
+                                        573
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with the labeled data. Additionally, our approach and other unsupervised learning methods are well-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                572,
+                                505,
+                                586
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        572,
+                                        505,
+                                        586
+                                    ],
+                                    "score": 1.0,
+                                    "content": "suited for transferring their learned representation to many possible downstream tasks or labelings,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                584,
+                                505,
+                                596
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        584,
+                                        505,
+                                        596
+                                    ],
+                                    "score": 1.0,
+                                    "content": "whereas semi-supervised learning methods typically optimize for performance on a single task, with",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                595,
+                                262,
+                                607
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        595,
+                                        262,
+                                        607
+                                    ],
+                                    "score": 1.0,
+                                    "content": "respect to a single labeling of the data.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        }
+                    ],
+                    "index": 30.5,
+                    "bbox_fs": [
+                        105,
+                        518,
+                        506,
+                        607
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        611,
+                        505,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                612,
+                                506,
+                                624
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        612,
+                                        506,
+                                        624
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Our method builds upon the ideas of meta-learning (Schmidhuber, 1987; Bengio et al., 1991; Naik",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                622,
+                                505,
+                                635
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        622,
+                                        505,
+                                        635
+                                    ],
+                                    "score": 1.0,
+                                    "content": "& Mammone, 1992) and few-shot learning (Santoro et al., 2016; Vinyals et al., 2016; Ravi &",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                633,
+                                505,
+                                646
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        633,
+                                        505,
+                                        646
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Larochelle, 2017; Munkhdalai & Yu, 2017; Snell et al., 2017). We apply two meta-learning al-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                644,
+                                506,
+                                658
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        644,
+                                        506,
+                                        658
+                                    ],
+                                    "score": 1.0,
+                                    "content": "gorithms, model-agnostic meta-learning (Finn et al., 2017) and prototypical networks (Snell et al.,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                655,
+                                505,
+                                669
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        655,
+                                        505,
+                                        669
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2017), to tasks constructed in an unsupervised manner. Similar to our problem setting, some prior",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                666,
+                                505,
+                                680
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        666,
+                                        505,
+                                        680
+                                    ],
+                                    "score": 1.0,
+                                    "content": "works have aimed to learn an unsupervised learning procedure with supervised data (Garg & Kalai,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                677,
+                                505,
+                                690
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        677,
+                                        505,
+                                        690
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2017; Metz et al., 2018). Instead, we consider a problem setting that is entirely unsupervised, aim-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                688,
+                                505,
+                                701
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        688,
+                                        505,
+                                        701
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ing to learn efficient learning algorithms using unlabeled datasets. Our problem setting is similar to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                699,
+                                505,
+                                712
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        699,
+                                        505,
+                                        712
+                                    ],
+                                    "score": 1.0,
+                                    "content": "that considered by Gupta et al. (2018), but we develop an approach that is suitable for supervised",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                711,
+                                505,
+                                723
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        711,
+                                        505,
+                                        723
+                                    ],
+                                    "score": 1.0,
+                                    "content": "downstream tasks, rather than reinforcement learning problems, and demonstrate our algorithm on",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                721,
+                                320,
+                                734
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        721,
+                                        320,
+                                        734
+                                    ],
+                                    "score": 1.0,
+                                    "content": "problems with high-dimensional visual observations.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        }
+                    ],
+                    "index": 40,
+                    "bbox_fs": [
+                        105,
+                        612,
+                        506,
+                        734
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        81,
+                        200,
+                        93
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                80,
+                                201,
+                                96
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        80,
+                                        201,
+                                        96
+                                    ],
+                                    "score": 1.0,
+                                    "content": "4 EXPERIMENTS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        }
+                    ],
+                    "index": 0
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        104,
+                        504,
+                        137
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                104,
+                                506,
+                                118
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        104,
+                                        506,
+                                        118
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We begin the experimental section by presenting our research questions and how our experiments are",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                115,
+                                504,
+                                128
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        115,
+                                        504,
+                                        128
+                                    ],
+                                    "score": 1.0,
+                                    "content": "designed to address them. Links to code for the experiments can be found at https://sites.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                127,
+                                335,
+                                140
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        127,
+                                        335,
+                                        140
+                                    ],
+                                    "score": 1.0,
+                                    "content": "google.com/view/unsupervised-via-meta.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        }
+                    ],
+                    "index": 2
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        144,
+                        505,
+                        330
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                143,
+                                505,
+                                156
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        143,
+                                        505,
+                                        156
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Benefit of meta-learning. Is there any significant benefit to doing meta-learning on tasks derived",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                154,
+                                506,
+                                168
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        154,
+                                        506,
+                                        168
+                                    ],
+                                    "score": 1.0,
+                                    "content": "from embeddings, or is the embedding function already sufficient for downstream supervised learn-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                165,
+                                505,
+                                178
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        165,
+                                        505,
+                                        178
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ing of new tasks? To investigate this, we run MAML and ProtoNets on tasks generated via CACTUs",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                174,
+                                506,
+                                190
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        174,
+                                        506,
+                                        190
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(CACTUs-MAML, CACTUs-ProtoNets). We compare to five alternate algorithms, with four being",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                187,
+                                505,
+                                200
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        188,
+                                        416,
+                                        200
+                                    ],
+                                    "score": 1.0,
+                                    "content": "supervised learning methods on top of the embedding function. i) Embedding",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        417,
+                                        187,
+                                        430,
+                                        199
+                                    ],
+                                    "score": 0.88,
+                                    "content": "k _ { \\mathrm { n n } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        431,
+                                        188,
+                                        505,
+                                        200
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-nearest neighbors",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                198,
+                                505,
+                                211
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        198,
+                                        505,
+                                        211
+                                    ],
+                                    "score": 1.0,
+                                    "content": "first infers the embeddings of the downstream task images. For a query test image, it predicts the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                209,
+                                506,
+                                222
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        209,
+                                        246,
+                                        222
+                                    ],
+                                    "score": 1.0,
+                                    "content": "plurality vote of the labels of the",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        246,
+                                        209,
+                                        261,
+                                        220
+                                    ],
+                                    "score": 0.88,
+                                    "content": "k _ { \\mathrm { n n } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        261,
+                                        209,
+                                        506,
+                                        222
+                                    ],
+                                    "score": 1.0,
+                                    "content": "training images that are closest in the embedding space to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                220,
+                                506,
+                                233
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        220,
+                                        506,
+                                        233
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the query’s embedding. ii) Embedding linear classifier also begins by inferring the embeddings of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                231,
+                                506,
+                                245
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        231,
+                                        382,
+                                        245
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the downstream task images. It then fits a linear classifier using the",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        382,
+                                        231,
+                                        402,
+                                        241
+                                    ],
+                                    "score": 0.7,
+                                    "content": "N K",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        402,
+                                        231,
+                                        506,
+                                        245
+                                    ],
+                                    "score": 1.0,
+                                    "content": "training embeddings and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                240,
+                                505,
+                                256
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        240,
+                                        505,
+                                        256
+                                    ],
+                                    "score": 1.0,
+                                    "content": "labels, and predicts labels for the query embeddings using the classifier. iii) Embedding multilayer",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                253,
+                                506,
+                                267
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        253,
+                                        506,
+                                        267
+                                    ],
+                                    "score": 1.0,
+                                    "content": "perceptron instead uses a network with one hidden layer of 128 units and tuned dropout (Srivastava",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                263,
+                                505,
+                                277
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        263,
+                                        505,
+                                        277
+                                    ],
+                                    "score": 1.0,
+                                    "content": "et al., 2014). iv) To isolate the effect of meta-learning on images, we also compare to embedding",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                275,
+                                505,
+                                288
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        275,
+                                        505,
+                                        288
+                                    ],
+                                    "score": 1.0,
+                                    "content": "cluster matching, i.e. directly using the meta-training clusters for classification by labeling clusters",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                285,
+                                505,
+                                299
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        285,
+                                        505,
+                                        299
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with a task’s training data via plurality vote. If a query datapoint maps to an unlabeled cluster,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                296,
+                                505,
+                                310
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        296,
+                                        505,
+                                        310
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the closest labeled cluster is used. v) As a baseline, we forgo any unsupervised pre-training and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                307,
+                                505,
+                                321
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        307,
+                                        505,
+                                        321
+                                    ],
+                                    "score": 1.0,
+                                    "content": "train a model with the MAML architecture from standard random network initialization via gradient",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                320,
+                                281,
+                                331
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        320,
+                                        281,
+                                        331
+                                    ],
+                                    "score": 1.0,
+                                    "content": "descent separately for each evaluation task.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        }
+                    ],
+                    "index": 12
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        335,
+                        505,
+                        413
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                334,
+                                506,
+                                349
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        334,
+                                        506,
+                                        349
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Different embedding spaces. Does CACTUs result in successful meta-learning for many distinct",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                346,
+                                505,
+                                360
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        346,
+                                        505,
+                                        360
+                                    ],
+                                    "score": 1.0,
+                                    "content": "task-generating embeddings? To investigate this, we run unsupervised meta-learning using four",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                357,
+                                505,
+                                370
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        357,
+                                        505,
+                                        370
+                                    ],
+                                    "score": 1.0,
+                                    "content": "embedding learning algorithms: ACAI (Berthelot et al., 2018), BiGAN (Donahue et al., 2017),",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                369,
+                                504,
+                                381
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        369,
+                                        504,
+                                        381
+                                    ],
+                                    "score": 1.0,
+                                    "content": "DeepCluster (Caron et al., 2018), and InfoGAN (Chen et al., 2016). These four approaches collec-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                380,
+                                505,
+                                392
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        380,
+                                        505,
+                                        392
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tively cover the following range of objectives and frameworks in the unsupervised learning literature:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                390,
+                                506,
+                                403
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        390,
+                                        506,
+                                        403
+                                    ],
+                                    "score": 1.0,
+                                    "content": "generative modeling, two-player games, reconstruction, representation interpolation, discriminative",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                402,
+                                505,
+                                414
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        402,
+                                        505,
+                                        414
+                                    ],
+                                    "score": 1.0,
+                                    "content": "clustering, and information maximization. We describe these methods in more detail in Appendix A.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        }
+                    ],
+                    "index": 24
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        418,
+                        505,
+                        528
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                417,
+                                506,
+                                431
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        417,
+                                        506,
+                                        431
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Applicability to different tasks. Can unsupervised meta-learning yield a good prior for a variety of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                429,
+                                506,
+                                441
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        429,
+                                        506,
+                                        441
+                                    ],
+                                    "score": 1.0,
+                                    "content": "task types? In other words, can unsupervised meta-learning yield a good representation for tasks that",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                441,
+                                506,
+                                453
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        441,
+                                        506,
+                                        453
+                                    ],
+                                    "score": 1.0,
+                                    "content": "assess the ability to distinguish between features on different scales, or tasks with various amounts of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                451,
+                                506,
+                                463
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        451,
+                                        506,
+                                        463
+                                    ],
+                                    "score": 1.0,
+                                    "content": "supervision signal? To investigate this, we evaluate our procedure on tasks assessing recognition of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                461,
+                                506,
+                                476
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        461,
+                                        506,
+                                        476
+                                    ],
+                                    "score": 1.0,
+                                    "content": "character identity, object identity, and facial attributes. For this purpose we choose to use the existing",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                474,
+                                505,
+                                485
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        474,
+                                        505,
+                                        485
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Omniglot (Santoro et al., 2016) and miniImageNet (Ravi & Larochelle, 2017) datasets and few-shot",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                485,
+                                505,
+                                496
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        485,
+                                        505,
+                                        496
+                                    ],
+                                    "score": 1.0,
+                                    "content": "classification tasks and, inspired by Finn et al. (2018), also construct a new few-shot classification",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                494,
+                                505,
+                                507
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        494,
+                                        505,
+                                        507
+                                    ],
+                                    "score": 1.0,
+                                    "content": "benchmark based on the CelebA dataset and its binary attribute annotations. For miniImageNet,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                506,
+                                505,
+                                519
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        506,
+                                        505,
+                                        519
+                                    ],
+                                    "score": 1.0,
+                                    "content": "we consider both few-shot downstream tasks and tasks involving larger datasets (up to 50-shot).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                517,
+                                433,
+                                530
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        517,
+                                        433,
+                                        530
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Specifics on the datasets and human-designed tasks are presented in Appendix B.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        }
+                    ],
+                    "index": 32.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        534,
+                        504,
+                        589
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                534,
+                                505,
+                                546
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        534,
+                                        505,
+                                        546
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Oracle. How does the performance of our unsupervised meta-learning method compare to super-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                545,
+                                505,
+                                556
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        545,
+                                        505,
+                                        556
+                                    ],
+                                    "score": 1.0,
+                                    "content": "vised meta-learning with a human-specified, near-optimal task distribution derived from a labeled",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                556,
+                                505,
+                                568
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        556,
+                                        505,
+                                        568
+                                    ],
+                                    "score": 1.0,
+                                    "content": "dataset? To investigate this, we use labeled versions of the meta-training datasets to run MAML and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                567,
+                                505,
+                                579
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        567,
+                                        505,
+                                        579
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ProtoNets as supervised meta-learning algorithms (Oracle-MAML, Oracle-ProtoNets). To facilitate",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                578,
+                                477,
+                                590
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        578,
+                                        477,
+                                        590
+                                    ],
+                                    "score": 1.0,
+                                    "content": "fair comparison with the unsupervised variants, we control for the relevant hyperparameters.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        }
+                    ],
+                    "index": 40
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        594,
+                        505,
+                        683
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                593,
+                                505,
+                                608
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        593,
+                                        505,
+                                        608
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Task construction ablation. How do the alternatives for constructing tasks from the embeddings",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                605,
+                                505,
+                                619
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        605,
+                                        505,
+                                        619
+                                    ],
+                                    "score": 1.0,
+                                    "content": "compare? To investigate this, we run MAML on tasks constructed via clustering (CACTUs-MAML)",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                615,
+                                505,
+                                630
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        615,
+                                        505,
+                                        630
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and MAML on tasks constructed via random hyperplane slices of the embedding space with varying",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                628,
+                                505,
+                                640
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        628,
+                                        505,
+                                        640
+                                    ],
+                                    "score": 1.0,
+                                    "content": "margin (Hyperplanes-MAML). The latter partitioning procedure is detailed in Appendix C. For the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                638,
+                                506,
+                                653
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        638,
+                                        506,
+                                        653
+                                    ],
+                                    "score": 1.0,
+                                    "content": "experiments where tasks are constructed via clustering, we also investigate the effect of sampling",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 47
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                649,
+                                506,
+                                662
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        649,
+                                        506,
+                                        662
+                                    ],
+                                    "score": 1.0,
+                                    "content": "based on a single partition versus multiple partitions. We additionally experiment with tasks based",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 48
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                660,
+                                506,
+                                673
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        660,
+                                        506,
+                                        673
+                                    ],
+                                    "score": 1.0,
+                                    "content": "on random assignments of images to “clusters” (Random-MAML) and tasks based on pixel-space",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 49
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                671,
+                                355,
+                                684
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        671,
+                                        355,
+                                        684
+                                    ],
+                                    "score": 1.0,
+                                    "content": "clusters (Pixels CACTUs-MAML) with the Omniglot dataset.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 50
+                        }
+                    ],
+                    "index": 46.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        689,
+                        504,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                689,
+                                505,
+                                700
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        689,
+                                        505,
+                                        700
+                                    ],
+                                    "score": 1.0,
+                                    "content": "To investigate the limitations of our method, we also consider an easier version of our problem",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 51
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                699,
+                                506,
+                                712
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        699,
+                                        506,
+                                        712
+                                    ],
+                                    "score": 1.0,
+                                    "content": "statement where the data distributions at meta-training and meta-test time perfectly overlap, i.e.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 52
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                711,
+                                506,
+                                723
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        711,
+                                        506,
+                                        723
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the images share a common set of underlying labels (Appendix D). Finally, we present results on",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 53
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                721,
+                                465,
+                                733
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        721,
+                                        465,
+                                        733
+                                    ],
+                                    "score": 1.0,
+                                    "content": "miniImageNet after unsupervised meta-learning on most of ILSVRC 2012 (Appendix G).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 54
+                        }
+                    ],
+                    "index": 52.5
+                }
+            ],
+            "page_idx": 5,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        108,
+                        27,
+                        293,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                26,
+                                294,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        26,
+                                        294,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Published as a conference paper at ICLR 2019",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        302,
+                        752,
+                        308,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                302,
+                                751,
+                                309,
+                                762
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        302,
+                                        751,
+                                        309,
+                                        762
+                                    ],
+                                    "score": 1.0,
+                                    "content": "6",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        81,
+                        200,
+                        93
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                80,
+                                201,
+                                96
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        80,
+                                        201,
+                                        96
+                                    ],
+                                    "score": 1.0,
+                                    "content": "4 EXPERIMENTS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        }
+                    ],
+                    "index": 0
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        104,
+                        504,
+                        137
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                104,
+                                506,
+                                118
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        104,
+                                        506,
+                                        118
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We begin the experimental section by presenting our research questions and how our experiments are",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                115,
+                                504,
+                                128
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        115,
+                                        504,
+                                        128
+                                    ],
+                                    "score": 1.0,
+                                    "content": "designed to address them. Links to code for the experiments can be found at https://sites.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                127,
+                                335,
+                                140
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        127,
+                                        335,
+                                        140
+                                    ],
+                                    "score": 1.0,
+                                    "content": "google.com/view/unsupervised-via-meta.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        }
+                    ],
+                    "index": 2,
+                    "bbox_fs": [
+                        105,
+                        104,
+                        506,
+                        140
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        144,
+                        505,
+                        330
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                143,
+                                505,
+                                156
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        143,
+                                        505,
+                                        156
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Benefit of meta-learning. Is there any significant benefit to doing meta-learning on tasks derived",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                154,
+                                506,
+                                168
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        154,
+                                        506,
+                                        168
+                                    ],
+                                    "score": 1.0,
+                                    "content": "from embeddings, or is the embedding function already sufficient for downstream supervised learn-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                165,
+                                505,
+                                178
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        165,
+                                        505,
+                                        178
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ing of new tasks? To investigate this, we run MAML and ProtoNets on tasks generated via CACTUs",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                174,
+                                506,
+                                190
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        174,
+                                        506,
+                                        190
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(CACTUs-MAML, CACTUs-ProtoNets). We compare to five alternate algorithms, with four being",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                187,
+                                505,
+                                200
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        188,
+                                        416,
+                                        200
+                                    ],
+                                    "score": 1.0,
+                                    "content": "supervised learning methods on top of the embedding function. i) Embedding",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        417,
+                                        187,
+                                        430,
+                                        199
+                                    ],
+                                    "score": 0.88,
+                                    "content": "k _ { \\mathrm { n n } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        431,
+                                        188,
+                                        505,
+                                        200
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-nearest neighbors",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                198,
+                                505,
+                                211
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        198,
+                                        505,
+                                        211
+                                    ],
+                                    "score": 1.0,
+                                    "content": "first infers the embeddings of the downstream task images. For a query test image, it predicts the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                209,
+                                506,
+                                222
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        209,
+                                        246,
+                                        222
+                                    ],
+                                    "score": 1.0,
+                                    "content": "plurality vote of the labels of the",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        246,
+                                        209,
+                                        261,
+                                        220
+                                    ],
+                                    "score": 0.88,
+                                    "content": "k _ { \\mathrm { n n } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        261,
+                                        209,
+                                        506,
+                                        222
+                                    ],
+                                    "score": 1.0,
+                                    "content": "training images that are closest in the embedding space to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                220,
+                                506,
+                                233
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        220,
+                                        506,
+                                        233
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the query’s embedding. ii) Embedding linear classifier also begins by inferring the embeddings of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                231,
+                                506,
+                                245
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        231,
+                                        382,
+                                        245
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the downstream task images. It then fits a linear classifier using the",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        382,
+                                        231,
+                                        402,
+                                        241
+                                    ],
+                                    "score": 0.7,
+                                    "content": "N K",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        402,
+                                        231,
+                                        506,
+                                        245
+                                    ],
+                                    "score": 1.0,
+                                    "content": "training embeddings and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                240,
+                                505,
+                                256
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        240,
+                                        505,
+                                        256
+                                    ],
+                                    "score": 1.0,
+                                    "content": "labels, and predicts labels for the query embeddings using the classifier. iii) Embedding multilayer",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                253,
+                                506,
+                                267
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        253,
+                                        506,
+                                        267
+                                    ],
+                                    "score": 1.0,
+                                    "content": "perceptron instead uses a network with one hidden layer of 128 units and tuned dropout (Srivastava",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                263,
+                                505,
+                                277
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        263,
+                                        505,
+                                        277
+                                    ],
+                                    "score": 1.0,
+                                    "content": "et al., 2014). iv) To isolate the effect of meta-learning on images, we also compare to embedding",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                275,
+                                505,
+                                288
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        275,
+                                        505,
+                                        288
+                                    ],
+                                    "score": 1.0,
+                                    "content": "cluster matching, i.e. directly using the meta-training clusters for classification by labeling clusters",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                285,
+                                505,
+                                299
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        285,
+                                        505,
+                                        299
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with a task’s training data via plurality vote. If a query datapoint maps to an unlabeled cluster,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                296,
+                                505,
+                                310
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        296,
+                                        505,
+                                        310
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the closest labeled cluster is used. v) As a baseline, we forgo any unsupervised pre-training and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                307,
+                                505,
+                                321
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        307,
+                                        505,
+                                        321
+                                    ],
+                                    "score": 1.0,
+                                    "content": "train a model with the MAML architecture from standard random network initialization via gradient",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                320,
+                                281,
+                                331
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        320,
+                                        281,
+                                        331
+                                    ],
+                                    "score": 1.0,
+                                    "content": "descent separately for each evaluation task.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        }
+                    ],
+                    "index": 12,
+                    "bbox_fs": [
+                        104,
+                        143,
+                        506,
+                        331
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        335,
+                        505,
+                        413
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                334,
+                                506,
+                                349
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        334,
+                                        506,
+                                        349
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Different embedding spaces. Does CACTUs result in successful meta-learning for many distinct",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                346,
+                                505,
+                                360
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        346,
+                                        505,
+                                        360
+                                    ],
+                                    "score": 1.0,
+                                    "content": "task-generating embeddings? To investigate this, we run unsupervised meta-learning using four",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                357,
+                                505,
+                                370
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        357,
+                                        505,
+                                        370
+                                    ],
+                                    "score": 1.0,
+                                    "content": "embedding learning algorithms: ACAI (Berthelot et al., 2018), BiGAN (Donahue et al., 2017),",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                369,
+                                504,
+                                381
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        369,
+                                        504,
+                                        381
+                                    ],
+                                    "score": 1.0,
+                                    "content": "DeepCluster (Caron et al., 2018), and InfoGAN (Chen et al., 2016). These four approaches collec-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                380,
+                                505,
+                                392
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        380,
+                                        505,
+                                        392
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tively cover the following range of objectives and frameworks in the unsupervised learning literature:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                390,
+                                506,
+                                403
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        390,
+                                        506,
+                                        403
+                                    ],
+                                    "score": 1.0,
+                                    "content": "generative modeling, two-player games, reconstruction, representation interpolation, discriminative",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                402,
+                                505,
+                                414
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        402,
+                                        505,
+                                        414
+                                    ],
+                                    "score": 1.0,
+                                    "content": "clustering, and information maximization. We describe these methods in more detail in Appendix A.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        }
+                    ],
+                    "index": 24,
+                    "bbox_fs": [
+                        105,
+                        334,
+                        506,
+                        414
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        418,
+                        505,
+                        528
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                417,
+                                506,
+                                431
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        417,
+                                        506,
+                                        431
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Applicability to different tasks. Can unsupervised meta-learning yield a good prior for a variety of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                429,
+                                506,
+                                441
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        429,
+                                        506,
+                                        441
+                                    ],
+                                    "score": 1.0,
+                                    "content": "task types? In other words, can unsupervised meta-learning yield a good representation for tasks that",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                441,
+                                506,
+                                453
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        441,
+                                        506,
+                                        453
+                                    ],
+                                    "score": 1.0,
+                                    "content": "assess the ability to distinguish between features on different scales, or tasks with various amounts of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                451,
+                                506,
+                                463
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        451,
+                                        506,
+                                        463
+                                    ],
+                                    "score": 1.0,
+                                    "content": "supervision signal? To investigate this, we evaluate our procedure on tasks assessing recognition of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                461,
+                                506,
+                                476
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        461,
+                                        506,
+                                        476
+                                    ],
+                                    "score": 1.0,
+                                    "content": "character identity, object identity, and facial attributes. For this purpose we choose to use the existing",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                474,
+                                505,
+                                485
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        474,
+                                        505,
+                                        485
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Omniglot (Santoro et al., 2016) and miniImageNet (Ravi & Larochelle, 2017) datasets and few-shot",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                485,
+                                505,
+                                496
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        485,
+                                        505,
+                                        496
+                                    ],
+                                    "score": 1.0,
+                                    "content": "classification tasks and, inspired by Finn et al. (2018), also construct a new few-shot classification",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                494,
+                                505,
+                                507
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        494,
+                                        505,
+                                        507
+                                    ],
+                                    "score": 1.0,
+                                    "content": "benchmark based on the CelebA dataset and its binary attribute annotations. For miniImageNet,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                506,
+                                505,
+                                519
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        506,
+                                        505,
+                                        519
+                                    ],
+                                    "score": 1.0,
+                                    "content": "we consider both few-shot downstream tasks and tasks involving larger datasets (up to 50-shot).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                517,
+                                433,
+                                530
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        517,
+                                        433,
+                                        530
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Specifics on the datasets and human-designed tasks are presented in Appendix B.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        }
+                    ],
+                    "index": 32.5,
+                    "bbox_fs": [
+                        105,
+                        417,
+                        506,
+                        530
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        534,
+                        504,
+                        589
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                534,
+                                505,
+                                546
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        534,
+                                        505,
+                                        546
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Oracle. How does the performance of our unsupervised meta-learning method compare to super-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                545,
+                                505,
+                                556
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        545,
+                                        505,
+                                        556
+                                    ],
+                                    "score": 1.0,
+                                    "content": "vised meta-learning with a human-specified, near-optimal task distribution derived from a labeled",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                556,
+                                505,
+                                568
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        556,
+                                        505,
+                                        568
+                                    ],
+                                    "score": 1.0,
+                                    "content": "dataset? To investigate this, we use labeled versions of the meta-training datasets to run MAML and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                567,
+                                505,
+                                579
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        567,
+                                        505,
+                                        579
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ProtoNets as supervised meta-learning algorithms (Oracle-MAML, Oracle-ProtoNets). To facilitate",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                578,
+                                477,
+                                590
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        578,
+                                        477,
+                                        590
+                                    ],
+                                    "score": 1.0,
+                                    "content": "fair comparison with the unsupervised variants, we control for the relevant hyperparameters.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        }
+                    ],
+                    "index": 40,
+                    "bbox_fs": [
+                        105,
+                        534,
+                        505,
+                        590
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        594,
+                        505,
+                        683
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                593,
+                                505,
+                                608
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        593,
+                                        505,
+                                        608
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Task construction ablation. How do the alternatives for constructing tasks from the embeddings",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                605,
+                                505,
+                                619
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        605,
+                                        505,
+                                        619
+                                    ],
+                                    "score": 1.0,
+                                    "content": "compare? To investigate this, we run MAML on tasks constructed via clustering (CACTUs-MAML)",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                615,
+                                505,
+                                630
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        615,
+                                        505,
+                                        630
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and MAML on tasks constructed via random hyperplane slices of the embedding space with varying",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                628,
+                                505,
+                                640
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        628,
+                                        505,
+                                        640
+                                    ],
+                                    "score": 1.0,
+                                    "content": "margin (Hyperplanes-MAML). The latter partitioning procedure is detailed in Appendix C. For the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                638,
+                                506,
+                                653
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        638,
+                                        506,
+                                        653
+                                    ],
+                                    "score": 1.0,
+                                    "content": "experiments where tasks are constructed via clustering, we also investigate the effect of sampling",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 47
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                649,
+                                506,
+                                662
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        649,
+                                        506,
+                                        662
+                                    ],
+                                    "score": 1.0,
+                                    "content": "based on a single partition versus multiple partitions. We additionally experiment with tasks based",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 48
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                660,
+                                506,
+                                673
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        660,
+                                        506,
+                                        673
+                                    ],
+                                    "score": 1.0,
+                                    "content": "on random assignments of images to “clusters” (Random-MAML) and tasks based on pixel-space",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 49
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                671,
+                                355,
+                                684
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        671,
+                                        355,
+                                        684
+                                    ],
+                                    "score": 1.0,
+                                    "content": "clusters (Pixels CACTUs-MAML) with the Omniglot dataset.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 50
+                        }
+                    ],
+                    "index": 46.5,
+                    "bbox_fs": [
+                        104,
+                        593,
+                        506,
+                        684
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        689,
+                        504,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                689,
+                                505,
+                                700
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        689,
+                                        505,
+                                        700
+                                    ],
+                                    "score": 1.0,
+                                    "content": "To investigate the limitations of our method, we also consider an easier version of our problem",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 51
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                699,
+                                506,
+                                712
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        699,
+                                        506,
+                                        712
+                                    ],
+                                    "score": 1.0,
+                                    "content": "statement where the data distributions at meta-training and meta-test time perfectly overlap, i.e.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 52
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                711,
+                                506,
+                                723
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        711,
+                                        506,
+                                        723
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the images share a common set of underlying labels (Appendix D). Finally, we present results on",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 53
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                721,
+                                465,
+                                733
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        721,
+                                        465,
+                                        733
+                                    ],
+                                    "score": 1.0,
+                                    "content": "miniImageNet after unsupervised meta-learning on most of ILSVRC 2012 (Appendix G).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 54
+                        }
+                    ],
+                    "index": 52.5,
+                    "bbox_fs": [
+                        105,
+                        689,
+                        506,
+                        733
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "image",
+                    "bbox": [
+                        108,
+                        75,
+                        502,
+                        225
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                108,
+                                75,
+                                502,
+                                225
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        108,
+                                        75,
+                                        502,
+                                        225
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                108,
+                                                75,
+                                                502,
+                                                225
+                                            ],
+                                            "score": 0.976,
+                                            "type": "image",
+                                            "image_path": "da015e73e499847b381f6ba35cc8ae17e2cc6b8b08026a9f2a626ccbe7e0ef3c.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 1,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        108,
+                                        75,
+                                        502,
+                                        125.0
+                                    ],
+                                    "spans": [],
+                                    "index": 0
+                                },
+                                {
+                                    "bbox": [
+                                        108,
+                                        125.0,
+                                        502,
+                                        175.0
+                                    ],
+                                    "spans": [],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        108,
+                                        175.0,
+                                        502,
+                                        225.0
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                106,
+                                228,
+                                505,
+                                289
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        228,
+                                        505,
+                                        240
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                228,
+                                                505,
+                                                240
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 2: Examples of three DeepCluster-embedding cluster-based classes (a) and a 2-way 5-shot test task (b)",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 3
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        239,
+                                        506,
+                                        250
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                239,
+                                                506,
+                                                250
+                                            ],
+                                            "score": 1.0,
+                                            "content": "for two datasets. (a) Some of the clusters correspond well to unseen labels (top left, bottom left). Others exhibit",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 4
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        249,
+                                        505,
+                                        260
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                249,
+                                                505,
+                                                260
+                                            ],
+                                            "score": 1.0,
+                                            "content": "semantic meaning despite members not being grouped as such in the labeled version of the dataset (top middle:",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 5
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        259,
+                                        505,
+                                        271
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                259,
+                                                505,
+                                                271
+                                            ],
+                                            "score": 1.0,
+                                            "content": "pair of objects, bottom middle: white hat). Still others are uninterpretable (top right) or are based on image",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 6
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        269,
+                                        505,
+                                        280
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                269,
+                                                505,
+                                                280
+                                            ],
+                                            "score": 1.0,
+                                            "content": "artifacts (bottom right). (b) We evaluate unsupervised pre-training based on the ability to learn downstream,",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 7
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        279,
+                                        352,
+                                        290
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                279,
+                                                352,
+                                                290
+                                            ],
+                                            "score": 1.0,
+                                            "content": "human-designed tasks with held-out images and underlying classes.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 8
+                                }
+                            ],
+                            "index": 5.5
+                        }
+                    ],
+                    "index": 3.25
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        312,
+                        295,
+                        323
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                312,
+                                296,
+                                325
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        312,
+                                        296,
+                                        325
+                                    ],
+                                    "score": 1.0,
+                                    "content": "4.1 EXPERIMENTAL PROTOCOL SUMMARY",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        }
+                    ],
+                    "index": 9
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        333,
+                        505,
+                        465
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                333,
+                                506,
+                                348
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        333,
+                                        506,
+                                        348
+                                    ],
+                                    "score": 1.0,
+                                    "content": "As discussed by Oliver et al. (2018), keeping proper experimental protocol is particularly important",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                344,
+                                506,
+                                357
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        344,
+                                        506,
+                                        357
+                                    ],
+                                    "score": 1.0,
+                                    "content": "when evaluating unsupervised and semi-supervised learning algorithms. Our foremost concern is",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                355,
+                                505,
+                                368
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        355,
+                                        505,
+                                        368
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to avoid falsely embellishing the capabilities of our approach by overfitting to the specific datasets",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                366,
+                                505,
+                                379
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        366,
+                                        505,
+                                        379
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and task types that we consider. To this end, we adhere to two key principles. We do not perform",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                378,
+                                505,
+                                390
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        378,
+                                        505,
+                                        390
+                                    ],
+                                    "score": 1.0,
+                                    "content": "any architecture engineering: we use architectures from prior work as-is, or lightly adapt them to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                389,
+                                505,
+                                401
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        389,
+                                        505,
+                                        401
+                                    ],
+                                    "score": 1.0,
+                                    "content": "our needs if necessary. We also keep hyperparameters related to the unsupervised meta-learning",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                399,
+                                505,
+                                412
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        399,
+                                        505,
+                                        412
+                                    ],
+                                    "score": 1.0,
+                                    "content": "stage as constant as possible across all experiments, including the MAML and ProtoNets model",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                410,
+                                505,
+                                423
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        410,
+                                        505,
+                                        423
+                                    ],
+                                    "score": 1.0,
+                                    "content": "architectures. Details on hyperparameters and architectures are presented in Appendix E. We assume",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                420,
+                                505,
+                                435
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        420,
+                                        328,
+                                        435
+                                    ],
+                                    "score": 1.0,
+                                    "content": "knowledge of an upper bound on the number of classes",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        329,
+                                        421,
+                                        339,
+                                        431
+                                    ],
+                                    "score": 0.79,
+                                    "content": "N",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        339,
+                                        420,
+                                        505,
+                                        435
+                                    ],
+                                    "score": 1.0,
+                                    "content": "present in each downstream meta-testing",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                431,
+                                505,
+                                445
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        431,
+                                        370,
+                                        445
+                                    ],
+                                    "score": 1.0,
+                                    "content": "task for each dataset. However, regardless of the number of shots",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        370,
+                                        433,
+                                        381,
+                                        442
+                                    ],
+                                    "score": 0.8,
+                                    "content": "K",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        381,
+                                        431,
+                                        505,
+                                        445
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", we do not assume knowledge",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                443,
+                                505,
+                                456
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        443,
+                                        117,
+                                        456
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        118,
+                                        443,
+                                        128,
+                                        453
+                                    ],
+                                    "score": 0.81,
+                                    "content": "K",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        128,
+                                        443,
+                                        305,
+                                        456
+                                    ],
+                                    "score": 1.0,
+                                    "content": "during unsupervised meta-learning. We use",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        305,
+                                        443,
+                                        315,
+                                        453
+                                    ],
+                                    "score": 0.81,
+                                    "content": "N",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        316,
+                                        443,
+                                        505,
+                                        456
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-way 1-shot tasks during meta-training, but test",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                453,
+                                277,
+                                467
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        453,
+                                        183,
+                                        467
+                                    ],
+                                    "score": 1.0,
+                                    "content": "on larger values of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        183,
+                                        454,
+                                        194,
+                                        464
+                                    ],
+                                    "score": 0.8,
+                                    "content": "K",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        194,
+                                        453,
+                                        277,
+                                        467
+                                    ],
+                                    "score": 1.0,
+                                    "content": "during meta-testing.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        }
+                    ],
+                    "index": 15.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        471,
+                        505,
+                        548
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                470,
+                                505,
+                                483
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        470,
+                                        505,
+                                        483
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We partition each dataset into meta-training, meta-validation, and meta-testing splits. For Omniglot",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                481,
+                                505,
+                                495
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        481,
+                                        505,
+                                        495
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and miniImageNet, these splits contain disjoint sets of classes. For all algorithms, we run unsuper-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                492,
+                                506,
+                                506
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        492,
+                                        506,
+                                        506
+                                    ],
+                                    "score": 1.0,
+                                    "content": "vised pre-training on the unlabeled meta-training split and report performance on downstream tasks",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                503,
+                                505,
+                                516
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        503,
+                                        505,
+                                        516
+                                    ],
+                                    "score": 1.0,
+                                    "content": "dictated by the labeled data of the meta-testing split, generated using the procedure from prior work",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                515,
+                                505,
+                                528
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        515,
+                                        505,
+                                        528
+                                    ],
+                                    "score": 1.0,
+                                    "content": "recounted in Section 2.2. For the supervised meta-learning oracles, meta-training tasks are con-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                526,
+                                505,
+                                538
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        526,
+                                        505,
+                                        538
+                                    ],
+                                    "score": 1.0,
+                                    "content": "structed in the same manner but from the dataset’s meta-training split. See Figure 2 for illustrative",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                537,
+                                396,
+                                549
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        537,
+                                        396,
+                                        549
+                                    ],
+                                    "score": 1.0,
+                                    "content": "examples of embedding-derived clusters and human-designed test tasks.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        }
+                    ],
+                    "index": 25
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        553,
+                        504,
+                        598
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                553,
+                                506,
+                                565
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        553,
+                                        506,
+                                        565
+                                    ],
+                                    "score": 1.0,
+                                    "content": "To facilitate analysis on meta-overfitting, we use the labels of the meta-validation split (instead",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                564,
+                                505,
+                                576
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        564,
+                                        505,
+                                        576
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of clustering embeddings) to construct tasks for meta-validation. However, because our aim is to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                576,
+                                506,
+                                587
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        576,
+                                        506,
+                                        587
+                                    ],
+                                    "score": 1.0,
+                                    "content": "perform meta-learning without supervision, we do not tune hyperparameters on this labeled data. We",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                585,
+                                506,
+                                601
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        585,
+                                        506,
+                                        601
+                                    ],
+                                    "score": 1.0,
+                                    "content": "use a fixed number of meta-training iterations, since there is no suitable criterion for early stopping.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        }
+                    ],
+                    "index": 30.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        603,
+                        504,
+                        669
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                603,
+                                506,
+                                616
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        603,
+                                        506,
+                                        616
+                                    ],
+                                    "score": 1.0,
+                                    "content": "When we experiment with the embedding-plus-supervised-learning methods used as fair compar-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                614,
+                                506,
+                                627
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        614,
+                                        506,
+                                        627
+                                    ],
+                                    "score": 1.0,
+                                    "content": "isons to unsupervised meta-learning, we err on the side of providing more supervision and data than",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                625,
+                                506,
+                                638
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        625,
+                                        506,
+                                        638
+                                    ],
+                                    "score": 1.0,
+                                    "content": "technically allowed. Specifically, we separately tune the supervised learning hyperparameters for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                634,
+                                506,
+                                650
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        634,
+                                        506,
+                                        650
+                                    ],
+                                    "score": 1.0,
+                                    "content": "each dataset and each task difficulty on the labeled version of the meta-validation split. With Deep-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                647,
+                                505,
+                                660
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        647,
+                                        505,
+                                        660
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Cluster embeddings, we also use the entire meta-testing split’s statistics to perform dimensionality",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                659,
+                                470,
+                                670
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        659,
+                                        470,
+                                        670
+                                    ],
+                                    "score": 1.0,
+                                    "content": "reduction (via PCA) and whitening, which is unfair as this shares information across tasks.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        }
+                    ],
+                    "index": 35.5
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        689,
+                        170,
+                        699
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                687,
+                                172,
+                                701
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        687,
+                                        172,
+                                        701
+                                    ],
+                                    "score": 1.0,
+                                    "content": "4.2 RESULTS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        }
+                    ],
+                    "index": 39
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        709,
+                        503,
+                        731
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                709,
+                                505,
+                                722
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        709,
+                                        505,
+                                        722
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Our primary results are summarized in Tables 1 and 2. Task construction ablations are summarized",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                720,
+                                179,
+                                731
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        720,
+                                        179,
+                                        731
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in Tables 3 and 4.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        }
+                    ],
+                    "index": 40.5
+                }
+            ],
+            "page_idx": 6,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        107,
+                        27,
+                        293,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                26,
+                                293,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        26,
+                                        293,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Published as a conference paper at ICLR 2019",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        303,
+                        751,
+                        308,
+                        759
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                302,
+                                751,
+                                309,
+                                762
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        302,
+                                        751,
+                                        309,
+                                        762
+                                    ],
+                                    "score": 1.0,
+                                    "content": "7",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "image",
+                    "bbox": [
+                        108,
+                        75,
+                        502,
+                        225
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                108,
+                                75,
+                                502,
+                                225
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        108,
+                                        75,
+                                        502,
+                                        225
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                108,
+                                                75,
+                                                502,
+                                                225
+                                            ],
+                                            "score": 0.976,
+                                            "type": "image",
+                                            "image_path": "da015e73e499847b381f6ba35cc8ae17e2cc6b8b08026a9f2a626ccbe7e0ef3c.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 1,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        108,
+                                        75,
+                                        502,
+                                        125.0
+                                    ],
+                                    "spans": [],
+                                    "index": 0
+                                },
+                                {
+                                    "bbox": [
+                                        108,
+                                        125.0,
+                                        502,
+                                        175.0
+                                    ],
+                                    "spans": [],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        108,
+                                        175.0,
+                                        502,
+                                        225.0
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                106,
+                                228,
+                                505,
+                                289
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        228,
+                                        505,
+                                        240
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                228,
+                                                505,
+                                                240
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 2: Examples of three DeepCluster-embedding cluster-based classes (a) and a 2-way 5-shot test task (b)",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 3
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        239,
+                                        506,
+                                        250
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                239,
+                                                506,
+                                                250
+                                            ],
+                                            "score": 1.0,
+                                            "content": "for two datasets. (a) Some of the clusters correspond well to unseen labels (top left, bottom left). Others exhibit",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 4
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        249,
+                                        505,
+                                        260
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                249,
+                                                505,
+                                                260
+                                            ],
+                                            "score": 1.0,
+                                            "content": "semantic meaning despite members not being grouped as such in the labeled version of the dataset (top middle:",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 5
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        259,
+                                        505,
+                                        271
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                259,
+                                                505,
+                                                271
+                                            ],
+                                            "score": 1.0,
+                                            "content": "pair of objects, bottom middle: white hat). Still others are uninterpretable (top right) or are based on image",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 6
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        269,
+                                        505,
+                                        280
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                269,
+                                                505,
+                                                280
+                                            ],
+                                            "score": 1.0,
+                                            "content": "artifacts (bottom right). (b) We evaluate unsupervised pre-training based on the ability to learn downstream,",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 7
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        279,
+                                        352,
+                                        290
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                279,
+                                                352,
+                                                290
+                                            ],
+                                            "score": 1.0,
+                                            "content": "human-designed tasks with held-out images and underlying classes.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 8
+                                }
+                            ],
+                            "index": 5.5
+                        }
+                    ],
+                    "index": 3.25
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        312,
+                        295,
+                        323
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                312,
+                                296,
+                                325
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        312,
+                                        296,
+                                        325
+                                    ],
+                                    "score": 1.0,
+                                    "content": "4.1 EXPERIMENTAL PROTOCOL SUMMARY",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        }
+                    ],
+                    "index": 9
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        333,
+                        505,
+                        465
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                333,
+                                506,
+                                348
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        333,
+                                        506,
+                                        348
+                                    ],
+                                    "score": 1.0,
+                                    "content": "As discussed by Oliver et al. (2018), keeping proper experimental protocol is particularly important",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                344,
+                                506,
+                                357
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        344,
+                                        506,
+                                        357
+                                    ],
+                                    "score": 1.0,
+                                    "content": "when evaluating unsupervised and semi-supervised learning algorithms. Our foremost concern is",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                355,
+                                505,
+                                368
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        355,
+                                        505,
+                                        368
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to avoid falsely embellishing the capabilities of our approach by overfitting to the specific datasets",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                366,
+                                505,
+                                379
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        366,
+                                        505,
+                                        379
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and task types that we consider. To this end, we adhere to two key principles. We do not perform",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                378,
+                                505,
+                                390
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        378,
+                                        505,
+                                        390
+                                    ],
+                                    "score": 1.0,
+                                    "content": "any architecture engineering: we use architectures from prior work as-is, or lightly adapt them to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                389,
+                                505,
+                                401
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        389,
+                                        505,
+                                        401
+                                    ],
+                                    "score": 1.0,
+                                    "content": "our needs if necessary. We also keep hyperparameters related to the unsupervised meta-learning",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                399,
+                                505,
+                                412
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        399,
+                                        505,
+                                        412
+                                    ],
+                                    "score": 1.0,
+                                    "content": "stage as constant as possible across all experiments, including the MAML and ProtoNets model",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                410,
+                                505,
+                                423
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        410,
+                                        505,
+                                        423
+                                    ],
+                                    "score": 1.0,
+                                    "content": "architectures. Details on hyperparameters and architectures are presented in Appendix E. We assume",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                420,
+                                505,
+                                435
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        420,
+                                        328,
+                                        435
+                                    ],
+                                    "score": 1.0,
+                                    "content": "knowledge of an upper bound on the number of classes",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        329,
+                                        421,
+                                        339,
+                                        431
+                                    ],
+                                    "score": 0.79,
+                                    "content": "N",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        339,
+                                        420,
+                                        505,
+                                        435
+                                    ],
+                                    "score": 1.0,
+                                    "content": "present in each downstream meta-testing",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                431,
+                                505,
+                                445
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        431,
+                                        370,
+                                        445
+                                    ],
+                                    "score": 1.0,
+                                    "content": "task for each dataset. However, regardless of the number of shots",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        370,
+                                        433,
+                                        381,
+                                        442
+                                    ],
+                                    "score": 0.8,
+                                    "content": "K",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        381,
+                                        431,
+                                        505,
+                                        445
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", we do not assume knowledge",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                443,
+                                505,
+                                456
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        443,
+                                        117,
+                                        456
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        118,
+                                        443,
+                                        128,
+                                        453
+                                    ],
+                                    "score": 0.81,
+                                    "content": "K",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        128,
+                                        443,
+                                        305,
+                                        456
+                                    ],
+                                    "score": 1.0,
+                                    "content": "during unsupervised meta-learning. We use",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        305,
+                                        443,
+                                        315,
+                                        453
+                                    ],
+                                    "score": 0.81,
+                                    "content": "N",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        316,
+                                        443,
+                                        505,
+                                        456
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-way 1-shot tasks during meta-training, but test",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                453,
+                                277,
+                                467
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        453,
+                                        183,
+                                        467
+                                    ],
+                                    "score": 1.0,
+                                    "content": "on larger values of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        183,
+                                        454,
+                                        194,
+                                        464
+                                    ],
+                                    "score": 0.8,
+                                    "content": "K",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        194,
+                                        453,
+                                        277,
+                                        467
+                                    ],
+                                    "score": 1.0,
+                                    "content": "during meta-testing.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        }
+                    ],
+                    "index": 15.5,
+                    "bbox_fs": [
+                        105,
+                        333,
+                        506,
+                        467
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        471,
+                        505,
+                        548
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                470,
+                                505,
+                                483
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        470,
+                                        505,
+                                        483
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We partition each dataset into meta-training, meta-validation, and meta-testing splits. For Omniglot",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                481,
+                                505,
+                                495
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        481,
+                                        505,
+                                        495
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and miniImageNet, these splits contain disjoint sets of classes. For all algorithms, we run unsuper-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                492,
+                                506,
+                                506
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        492,
+                                        506,
+                                        506
+                                    ],
+                                    "score": 1.0,
+                                    "content": "vised pre-training on the unlabeled meta-training split and report performance on downstream tasks",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                503,
+                                505,
+                                516
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        503,
+                                        505,
+                                        516
+                                    ],
+                                    "score": 1.0,
+                                    "content": "dictated by the labeled data of the meta-testing split, generated using the procedure from prior work",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                515,
+                                505,
+                                528
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        515,
+                                        505,
+                                        528
+                                    ],
+                                    "score": 1.0,
+                                    "content": "recounted in Section 2.2. For the supervised meta-learning oracles, meta-training tasks are con-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                526,
+                                505,
+                                538
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        526,
+                                        505,
+                                        538
+                                    ],
+                                    "score": 1.0,
+                                    "content": "structed in the same manner but from the dataset’s meta-training split. See Figure 2 for illustrative",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                537,
+                                396,
+                                549
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        537,
+                                        396,
+                                        549
+                                    ],
+                                    "score": 1.0,
+                                    "content": "examples of embedding-derived clusters and human-designed test tasks.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        }
+                    ],
+                    "index": 25,
+                    "bbox_fs": [
+                        105,
+                        470,
+                        506,
+                        549
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        553,
+                        504,
+                        598
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                553,
+                                506,
+                                565
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        553,
+                                        506,
+                                        565
+                                    ],
+                                    "score": 1.0,
+                                    "content": "To facilitate analysis on meta-overfitting, we use the labels of the meta-validation split (instead",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                564,
+                                505,
+                                576
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        564,
+                                        505,
+                                        576
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of clustering embeddings) to construct tasks for meta-validation. However, because our aim is to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                576,
+                                506,
+                                587
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        576,
+                                        506,
+                                        587
+                                    ],
+                                    "score": 1.0,
+                                    "content": "perform meta-learning without supervision, we do not tune hyperparameters on this labeled data. We",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                585,
+                                506,
+                                601
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        585,
+                                        506,
+                                        601
+                                    ],
+                                    "score": 1.0,
+                                    "content": "use a fixed number of meta-training iterations, since there is no suitable criterion for early stopping.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        }
+                    ],
+                    "index": 30.5,
+                    "bbox_fs": [
+                        104,
+                        553,
+                        506,
+                        601
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        603,
+                        504,
+                        669
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                603,
+                                506,
+                                616
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        603,
+                                        506,
+                                        616
+                                    ],
+                                    "score": 1.0,
+                                    "content": "When we experiment with the embedding-plus-supervised-learning methods used as fair compar-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                614,
+                                506,
+                                627
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        614,
+                                        506,
+                                        627
+                                    ],
+                                    "score": 1.0,
+                                    "content": "isons to unsupervised meta-learning, we err on the side of providing more supervision and data than",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                625,
+                                506,
+                                638
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        625,
+                                        506,
+                                        638
+                                    ],
+                                    "score": 1.0,
+                                    "content": "technically allowed. Specifically, we separately tune the supervised learning hyperparameters for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                634,
+                                506,
+                                650
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        634,
+                                        506,
+                                        650
+                                    ],
+                                    "score": 1.0,
+                                    "content": "each dataset and each task difficulty on the labeled version of the meta-validation split. With Deep-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                647,
+                                505,
+                                660
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        647,
+                                        505,
+                                        660
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Cluster embeddings, we also use the entire meta-testing split’s statistics to perform dimensionality",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                659,
+                                470,
+                                670
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        659,
+                                        470,
+                                        670
+                                    ],
+                                    "score": 1.0,
+                                    "content": "reduction (via PCA) and whitening, which is unfair as this shares information across tasks.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        }
+                    ],
+                    "index": 35.5,
+                    "bbox_fs": [
+                        105,
+                        603,
+                        506,
+                        670
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        689,
+                        170,
+                        699
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                687,
+                                172,
+                                701
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        687,
+                                        172,
+                                        701
+                                    ],
+                                    "score": 1.0,
+                                    "content": "4.2 RESULTS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        }
+                    ],
+                    "index": 39
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        709,
+                        503,
+                        731
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                709,
+                                505,
+                                722
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        709,
+                                        505,
+                                        722
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Our primary results are summarized in Tables 1 and 2. Task construction ablations are summarized",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                720,
+                                179,
+                                731
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        720,
+                                        179,
+                                        731
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in Tables 3 and 4.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        }
+                    ],
+                    "index": 40.5,
+                    "bbox_fs": [
+                        106,
+                        709,
+                        505,
+                        731
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        83,
+                        505,
+                        280
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                82,
+                                504,
+                                94
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        82,
+                                        504,
+                                        94
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Benefit of meta-learning. CACTUs-MAML consistently yields a learning procedure that results",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                93,
+                                505,
+                                106
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        93,
+                                        505,
+                                        106
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in more successful downstream task performance than all other unsupervised methods, including",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                104,
+                                506,
+                                117
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        104,
+                                        506,
+                                        117
+                                    ],
+                                    "score": 1.0,
+                                    "content": "those that learn on top of the embedding that generated meta-training tasks for MAML. We find the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                115,
+                                505,
+                                128
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        115,
+                                        505,
+                                        128
+                                    ],
+                                    "score": 1.0,
+                                    "content": "same result for CACTUs-ProtoNets for 1-shot downstream tasks. However, as noted by Snell et al.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                126,
+                                505,
+                                140
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        126,
+                                        505,
+                                        140
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(2017), ProtoNets perform best when meta-training shot and meta-testing shot are matched; this",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                137,
+                                505,
+                                150
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        137,
+                                        505,
+                                        150
+                                    ],
+                                    "score": 1.0,
+                                    "content": "characteristic prevents ProtoNets from improving upon ACAI for 20-way 5-shot Omniglot and upon",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                147,
+                                506,
+                                162
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        147,
+                                        506,
+                                        162
+                                    ],
+                                    "score": 1.0,
+                                    "content": "DeepCluster for 50-shot miniImageNet. We attribute the success of CACTUs-based meta-learning",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                159,
+                                505,
+                                172
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        159,
+                                        505,
+                                        172
+                                    ],
+                                    "score": 1.0,
+                                    "content": "over the embedding-based methods to two factors: its practice in distinguishing between many dis-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                169,
+                                506,
+                                184
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        169,
+                                        506,
+                                        184
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tinct sets of clusters from modest amounts of signal, and the underlying classes of the meta-testing",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                181,
+                                506,
+                                194
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        181,
+                                        506,
+                                        194
+                                    ],
+                                    "score": 1.0,
+                                    "content": "split data being out-of-distribution. In principle, the latter factor is solely responsible for the success",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                192,
+                                505,
+                                205
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        192,
+                                        505,
+                                        205
+                                    ],
+                                    "score": 1.0,
+                                    "content": "over embedding cluster matching, since this algorithm can be viewed as a meta-learner on embed-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                202,
+                                506,
+                                216
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        202,
+                                        506,
+                                        216
+                                    ],
+                                    "score": 1.0,
+                                    "content": "dings that trivially obtains perfect accuracy (via memorization) on the meta-training tasks. The same",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                214,
+                                505,
+                                226
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        214,
+                                        505,
+                                        226
+                                    ],
+                                    "score": 1.0,
+                                    "content": "factor also helps explain why training from standard network initialization is, in general, competi-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                225,
+                                505,
+                                238
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        225,
+                                        505,
+                                        238
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tive with directly using the task-generating embedding as a representation. On the other hand, the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                235,
+                                505,
+                                249
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        235,
+                                        505,
+                                        249
+                                    ],
+                                    "score": 1.0,
+                                    "content": "MNIST results (Table 7 in Appendix F) suggest that when the meta-training and meta-testing data",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                247,
+                                505,
+                                260
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        247,
+                                        505,
+                                        260
+                                    ],
+                                    "score": 1.0,
+                                    "content": "distributions have perfect overlap and the embedding is well-suited enough that embedding cluster",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                258,
+                                505,
+                                271
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        258,
+                                        505,
+                                        271
+                                    ],
+                                    "score": 1.0,
+                                    "content": "matching can already achieve high performance, CACTUs-MAML yields only a small benefit, as",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                270,
+                                178,
+                                281
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        270,
+                                        178,
+                                        281
+                                    ],
+                                    "score": 1.0,
+                                    "content": "we would expect.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        }
+                    ],
+                    "index": 8.5
+                },
+                {
+                    "type": "table",
+                    "bbox": [
+                        157,
+                        327,
+                        453,
+                        514
+                    ],
+                    "blocks": [
+                        {
+                            "type": "table_caption",
+                            "bbox": [
+                                107,
+                                291,
+                                503,
+                                322
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        290,
+                                        505,
+                                        303
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                290,
+                                                505,
+                                                303
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Table 1: Results of unsupervised learning on Omniglot images, averaged over 1000 downstream character",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 18
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        301,
+                                        505,
+                                        313
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                301,
+                                                272,
+                                                313
+                                            ],
+                                            "score": 1.0,
+                                            "content": "recognition tasks. CACTUs experiments use",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                272,
+                                                302,
+                                                308,
+                                                311
+                                            ],
+                                            "score": 0.9,
+                                            "content": "k = 5 0 0",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                308,
+                                                301,
+                                                382,
+                                                313
+                                            ],
+                                            "score": 1.0,
+                                            "content": "clusters for each of",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                382,
+                                                302,
+                                                419,
+                                                311
+                                            ],
+                                            "score": 0.9,
+                                            "content": "P = 1 0 0",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                420,
+                                                301,
+                                                505,
+                                                313
+                                            ],
+                                            "score": 1.0,
+                                            "content": "partitions. Embedding",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 19
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        311,
+                                        499,
+                                        323
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                311,
+                                                219,
+                                                323
+                                            ],
+                                            "score": 1.0,
+                                            "content": "cluster matching uses the same",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                220,
+                                                312,
+                                                226,
+                                                320
+                                            ],
+                                            "score": 0.72,
+                                            "content": "k",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                226,
+                                                311,
+                                                499,
+                                                323
+                                            ],
+                                            "score": 1.0,
+                                            "content": ". For complete results with confidence intervals, see Table 8 in Appendix F.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 20
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "type": "table_body",
+                            "bbox": [
+                                157,
+                                327,
+                                453,
+                                514
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        157,
+                                        327,
+                                        453,
+                                        514
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                157,
+                                                327,
+                                                453,
+                                                514
+                                            ],
+                                            "score": 0.982,
+                                            "html": "<table><tr><td>Algorithm</td><td>(way, shot)</td><td>(5,1)</td><td>(5,5)</td><td>(20,1)</td><td>(20,5)</td></tr><tr><td>Training from scratch</td><td>52.50%</td><td></td><td>74.78%</td><td>24.91%</td><td>47.62%</td></tr><tr><td>ACAI knn-nearest neighbors</td><td></td><td>57.46%</td><td>81.16%</td><td>39.73%</td><td>66.38%</td></tr><tr><td>ACAIlinear classifier</td><td></td><td>61.08%</td><td>81.82%</td><td>43.20%</td><td>66.33%</td></tr><tr><td>ACAI MLP with dropout</td><td></td><td>51.95%</td><td>77.20%</td><td>30.65%</td><td>58.62%</td></tr><tr><td>ACAI cluster matching</td><td></td><td>54.94%</td><td>71.09%</td><td>32.19%</td><td>45.93%</td></tr><tr><td>ACAI CACTUs-MAML (ours)</td><td>68.84%</td><td></td><td>87.78%</td><td>48.09%</td><td>73.36%</td></tr><tr><td>ACAI CACTUs-ProtoNets (ours)</td><td>68.12%</td><td></td><td>83.58%</td><td>47.75%</td><td>66.27%</td></tr><tr><td>BiGAN knn-nearest neighbors</td><td>49.55%</td><td></td><td>68.06%</td><td>27.37%</td><td>46.70%</td></tr><tr><td>BiGAN linearclassifier</td><td>48.28%</td><td></td><td>68.72%</td><td>27.80%</td><td>45.82%</td></tr><tr><td>BiGAN MLP with dropout</td><td>40.54%</td><td></td><td>62.56%</td><td>19.92%</td><td>40.71%</td></tr><tr><td>BiGAN cluster matching</td><td>43.96%</td><td></td><td>58.62%</td><td>21.54%</td><td>31.06%</td></tr><tr><td>BiGAN CACTUs-MAML (ours)</td><td>58.18%</td><td></td><td>78.66%</td><td>35.56%</td><td>58.62%</td></tr><tr><td>BiGAN CACTUs-ProtoNets (ours)</td><td>54.74%</td><td></td><td>71.69%</td><td>33.40%</td><td>50.62%</td></tr><tr><td>Oracle-MAML (control)</td><td>94.46%</td><td></td><td>98.83%</td><td>84.60%</td><td>96.29%</td></tr><tr><td>Oracle-ProtoNets (control)</td><td>98.35%</td><td></td><td>99.58%</td><td>95.31%</td><td>98.81%</td></tr></table>",
+                                            "type": "table",
+                                            "image_path": "07ac8e953e870b7ab2d216bc12e1a6585a87a844939cdc0ab6c24a82e821f314.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 22,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        157,
+                                        327,
+                                        453,
+                                        389.3333333333333
+                                    ],
+                                    "spans": [],
+                                    "index": 21
+                                },
+                                {
+                                    "bbox": [
+                                        157,
+                                        389.3333333333333,
+                                        453,
+                                        451.66666666666663
+                                    ],
+                                    "spans": [],
+                                    "index": 22
+                                },
+                                {
+                                    "bbox": [
+                                        157,
+                                        451.66666666666663,
+                                        453,
+                                        514.0
+                                    ],
+                                    "spans": [],
+                                    "index": 23
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 20.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        522,
+                        505,
+                        588
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                522,
+                                505,
+                                535
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        522,
+                                        505,
+                                        535
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Different embedding spaces. CACTUs is effective for a variety of embedding learning methods",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                533,
+                                505,
+                                546
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        533,
+                                        505,
+                                        546
+                                    ],
+                                    "score": 1.0,
+                                    "content": "used for task generation. The performance of unsupervised meta-learning can largely be predicted",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                544,
+                                505,
+                                557
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        544,
+                                        505,
+                                        557
+                                    ],
+                                    "score": 1.0,
+                                    "content": "by the performance of the embedding-based non-meta-learning methods. For example, the ACAI",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                555,
+                                505,
+                                568
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        555,
+                                        505,
+                                        568
+                                    ],
+                                    "score": 1.0,
+                                    "content": "embedding does well with Omniglot, leading to the best unsupervised results with ACAI CACTUs-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                565,
+                                505,
+                                579
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        565,
+                                        505,
+                                        579
+                                    ],
+                                    "score": 1.0,
+                                    "content": "MAML. Likewise, on miniImageNet, the best performing prior embedding (DeepCluster) also cor-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                577,
+                                478,
+                                590
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        577,
+                                        478,
+                                        590
+                                    ],
+                                    "score": 1.0,
+                                    "content": "responds to the best performing unsupervised meta-learner (DeepCluster CACTUs-MAML).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        }
+                    ],
+                    "index": 26.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        594,
+                        505,
+                        672
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                594,
+                                505,
+                                606
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        594,
+                                        505,
+                                        606
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Applicability to different tasks. CACTUs-MAML learns an effective prior for a variety of task",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                605,
+                                506,
+                                618
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        605,
+                                        506,
+                                        618
+                                    ],
+                                    "score": 1.0,
+                                    "content": "types. This can be attributed to the application-agnostic task-generation process and the expressive",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                615,
+                                506,
+                                628
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        615,
+                                        506,
+                                        628
+                                    ],
+                                    "score": 1.0,
+                                    "content": "power of MAML (Finn & Levine, 2018). We also observe that, despite all meta-learning models",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                627,
+                                506,
+                                639
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        627,
+                                        178,
+                                        639
+                                    ],
+                                    "score": 1.0,
+                                    "content": "being trained for",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        179,
+                                        627,
+                                        189,
+                                        637
+                                    ],
+                                    "score": 0.82,
+                                    "content": "N",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        189,
+                                        627,
+                                        506,
+                                        639
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-way 1-shot classification of unsupervised tasks, the models work well for a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                637,
+                                505,
+                                650
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        637,
+                                        150,
+                                        650
+                                    ],
+                                    "score": 1.0,
+                                    "content": "variety of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        150,
+                                        638,
+                                        162,
+                                        648
+                                    ],
+                                    "score": 0.79,
+                                    "content": "M",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        162,
+                                        637,
+                                        184,
+                                        650
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-way",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        185,
+                                        639,
+                                        195,
+                                        648
+                                    ],
+                                    "score": 0.82,
+                                    "content": "K",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        195,
+                                        637,
+                                        272,
+                                        650
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-shot tasks, where",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        273,
+                                        638,
+                                        313,
+                                        649
+                                    ],
+                                    "score": 0.92,
+                                    "content": "M \\ \\leq \\ N",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        313,
+                                        637,
+                                        333,
+                                        650
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        334,
+                                        638,
+                                        372,
+                                        649
+                                    ],
+                                    "score": 0.91,
+                                    "content": "K \\ \\leq \\ 5 0",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        373,
+                                        637,
+                                        505,
+                                        650
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". As mentioned previously, the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                649,
+                                505,
+                                660
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        649,
+                                        505,
+                                        660
+                                    ],
+                                    "score": 1.0,
+                                    "content": "representation that CACTUs-ProtoNets learns is best suited for downstream tasks which match the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                660,
+                                245,
+                                673
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        660,
+                                        245,
+                                        673
+                                    ],
+                                    "score": 1.0,
+                                    "content": "single shot used for meta-training.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        }
+                    ],
+                    "index": 33
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        677,
+                        504,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                676,
+                                506,
+                                690
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        676,
+                                        506,
+                                        690
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Oracle. The penalty for not having ground truth labels to construct near-optimal tasks ranges from",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                687,
+                                505,
+                                699
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        687,
+                                        505,
+                                        699
+                                    ],
+                                    "score": 1.0,
+                                    "content": "substantial to severe, depending on the difficulty of the downstream task. Easier downstream tasks",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                699,
+                                505,
+                                711
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        699,
+                                        505,
+                                        711
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(which have fewer classes and/or more supervision) incur less of a penalty. We conjecture that",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                708,
+                                506,
+                                723
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        708,
+                                        506,
+                                        723
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with such tasks, the difference in the usefulness of the priors matters less since the downstream",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                721,
+                                352,
+                                733
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        721,
+                                        352,
+                                        733
+                                    ],
+                                    "score": 1.0,
+                                    "content": "task-specific evidence has more power to shape the posterior.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        }
+                    ],
+                    "index": 39
+                }
+            ],
+            "page_idx": 7,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        108,
+                        27,
+                        293,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                26,
+                                294,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        26,
+                                        294,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Published as a conference paper at ICLR 2019",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        302,
+                        752,
+                        308,
+                        759
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                302,
+                                750,
+                                309,
+                                761
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        302,
+                                        750,
+                                        309,
+                                        761
+                                    ],
+                                    "score": 1.0,
+                                    "content": "8",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        83,
+                        505,
+                        280
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                82,
+                                504,
+                                94
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        82,
+                                        504,
+                                        94
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Benefit of meta-learning. CACTUs-MAML consistently yields a learning procedure that results",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                93,
+                                505,
+                                106
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        93,
+                                        505,
+                                        106
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in more successful downstream task performance than all other unsupervised methods, including",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                104,
+                                506,
+                                117
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        104,
+                                        506,
+                                        117
+                                    ],
+                                    "score": 1.0,
+                                    "content": "those that learn on top of the embedding that generated meta-training tasks for MAML. We find the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                115,
+                                505,
+                                128
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        115,
+                                        505,
+                                        128
+                                    ],
+                                    "score": 1.0,
+                                    "content": "same result for CACTUs-ProtoNets for 1-shot downstream tasks. However, as noted by Snell et al.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                126,
+                                505,
+                                140
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        126,
+                                        505,
+                                        140
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(2017), ProtoNets perform best when meta-training shot and meta-testing shot are matched; this",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                137,
+                                505,
+                                150
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        137,
+                                        505,
+                                        150
+                                    ],
+                                    "score": 1.0,
+                                    "content": "characteristic prevents ProtoNets from improving upon ACAI for 20-way 5-shot Omniglot and upon",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                147,
+                                506,
+                                162
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        147,
+                                        506,
+                                        162
+                                    ],
+                                    "score": 1.0,
+                                    "content": "DeepCluster for 50-shot miniImageNet. We attribute the success of CACTUs-based meta-learning",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                159,
+                                505,
+                                172
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        159,
+                                        505,
+                                        172
+                                    ],
+                                    "score": 1.0,
+                                    "content": "over the embedding-based methods to two factors: its practice in distinguishing between many dis-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                169,
+                                506,
+                                184
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        169,
+                                        506,
+                                        184
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tinct sets of clusters from modest amounts of signal, and the underlying classes of the meta-testing",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                181,
+                                506,
+                                194
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        181,
+                                        506,
+                                        194
+                                    ],
+                                    "score": 1.0,
+                                    "content": "split data being out-of-distribution. In principle, the latter factor is solely responsible for the success",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                192,
+                                505,
+                                205
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        192,
+                                        505,
+                                        205
+                                    ],
+                                    "score": 1.0,
+                                    "content": "over embedding cluster matching, since this algorithm can be viewed as a meta-learner on embed-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                202,
+                                506,
+                                216
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        202,
+                                        506,
+                                        216
+                                    ],
+                                    "score": 1.0,
+                                    "content": "dings that trivially obtains perfect accuracy (via memorization) on the meta-training tasks. The same",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                214,
+                                505,
+                                226
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        214,
+                                        505,
+                                        226
+                                    ],
+                                    "score": 1.0,
+                                    "content": "factor also helps explain why training from standard network initialization is, in general, competi-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                225,
+                                505,
+                                238
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        225,
+                                        505,
+                                        238
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tive with directly using the task-generating embedding as a representation. On the other hand, the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                235,
+                                505,
+                                249
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        235,
+                                        505,
+                                        249
+                                    ],
+                                    "score": 1.0,
+                                    "content": "MNIST results (Table 7 in Appendix F) suggest that when the meta-training and meta-testing data",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                247,
+                                505,
+                                260
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        247,
+                                        505,
+                                        260
+                                    ],
+                                    "score": 1.0,
+                                    "content": "distributions have perfect overlap and the embedding is well-suited enough that embedding cluster",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                258,
+                                505,
+                                271
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        258,
+                                        505,
+                                        271
+                                    ],
+                                    "score": 1.0,
+                                    "content": "matching can already achieve high performance, CACTUs-MAML yields only a small benefit, as",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                270,
+                                178,
+                                281
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        270,
+                                        178,
+                                        281
+                                    ],
+                                    "score": 1.0,
+                                    "content": "we would expect.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        }
+                    ],
+                    "index": 8.5,
+                    "bbox_fs": [
+                        104,
+                        82,
+                        506,
+                        281
+                    ]
+                },
+                {
+                    "type": "table",
+                    "bbox": [
+                        157,
+                        327,
+                        453,
+                        514
+                    ],
+                    "blocks": [
+                        {
+                            "type": "table_caption",
+                            "bbox": [
+                                107,
+                                291,
+                                503,
+                                322
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        290,
+                                        505,
+                                        303
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                290,
+                                                505,
+                                                303
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Table 1: Results of unsupervised learning on Omniglot images, averaged over 1000 downstream character",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 18
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        301,
+                                        505,
+                                        313
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                301,
+                                                272,
+                                                313
+                                            ],
+                                            "score": 1.0,
+                                            "content": "recognition tasks. CACTUs experiments use",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                272,
+                                                302,
+                                                308,
+                                                311
+                                            ],
+                                            "score": 0.9,
+                                            "content": "k = 5 0 0",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                308,
+                                                301,
+                                                382,
+                                                313
+                                            ],
+                                            "score": 1.0,
+                                            "content": "clusters for each of",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                382,
+                                                302,
+                                                419,
+                                                311
+                                            ],
+                                            "score": 0.9,
+                                            "content": "P = 1 0 0",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                420,
+                                                301,
+                                                505,
+                                                313
+                                            ],
+                                            "score": 1.0,
+                                            "content": "partitions. Embedding",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 19
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        311,
+                                        499,
+                                        323
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                311,
+                                                219,
+                                                323
+                                            ],
+                                            "score": 1.0,
+                                            "content": "cluster matching uses the same",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                220,
+                                                312,
+                                                226,
+                                                320
+                                            ],
+                                            "score": 0.72,
+                                            "content": "k",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                226,
+                                                311,
+                                                499,
+                                                323
+                                            ],
+                                            "score": 1.0,
+                                            "content": ". For complete results with confidence intervals, see Table 8 in Appendix F.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 20
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "type": "table_body",
+                            "bbox": [
+                                157,
+                                327,
+                                453,
+                                514
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        157,
+                                        327,
+                                        453,
+                                        514
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                157,
+                                                327,
+                                                453,
+                                                514
+                                            ],
+                                            "score": 0.982,
+                                            "html": "<table><tr><td>Algorithm</td><td>(way, shot)</td><td>(5,1)</td><td>(5,5)</td><td>(20,1)</td><td>(20,5)</td></tr><tr><td>Training from scratch</td><td>52.50%</td><td></td><td>74.78%</td><td>24.91%</td><td>47.62%</td></tr><tr><td>ACAI knn-nearest neighbors</td><td></td><td>57.46%</td><td>81.16%</td><td>39.73%</td><td>66.38%</td></tr><tr><td>ACAIlinear classifier</td><td></td><td>61.08%</td><td>81.82%</td><td>43.20%</td><td>66.33%</td></tr><tr><td>ACAI MLP with dropout</td><td></td><td>51.95%</td><td>77.20%</td><td>30.65%</td><td>58.62%</td></tr><tr><td>ACAI cluster matching</td><td></td><td>54.94%</td><td>71.09%</td><td>32.19%</td><td>45.93%</td></tr><tr><td>ACAI CACTUs-MAML (ours)</td><td>68.84%</td><td></td><td>87.78%</td><td>48.09%</td><td>73.36%</td></tr><tr><td>ACAI CACTUs-ProtoNets (ours)</td><td>68.12%</td><td></td><td>83.58%</td><td>47.75%</td><td>66.27%</td></tr><tr><td>BiGAN knn-nearest neighbors</td><td>49.55%</td><td></td><td>68.06%</td><td>27.37%</td><td>46.70%</td></tr><tr><td>BiGAN linearclassifier</td><td>48.28%</td><td></td><td>68.72%</td><td>27.80%</td><td>45.82%</td></tr><tr><td>BiGAN MLP with dropout</td><td>40.54%</td><td></td><td>62.56%</td><td>19.92%</td><td>40.71%</td></tr><tr><td>BiGAN cluster matching</td><td>43.96%</td><td></td><td>58.62%</td><td>21.54%</td><td>31.06%</td></tr><tr><td>BiGAN CACTUs-MAML (ours)</td><td>58.18%</td><td></td><td>78.66%</td><td>35.56%</td><td>58.62%</td></tr><tr><td>BiGAN CACTUs-ProtoNets (ours)</td><td>54.74%</td><td></td><td>71.69%</td><td>33.40%</td><td>50.62%</td></tr><tr><td>Oracle-MAML (control)</td><td>94.46%</td><td></td><td>98.83%</td><td>84.60%</td><td>96.29%</td></tr><tr><td>Oracle-ProtoNets (control)</td><td>98.35%</td><td></td><td>99.58%</td><td>95.31%</td><td>98.81%</td></tr></table>",
+                                            "type": "table",
+                                            "image_path": "07ac8e953e870b7ab2d216bc12e1a6585a87a844939cdc0ab6c24a82e821f314.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 22,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        157,
+                                        327,
+                                        453,
+                                        389.3333333333333
+                                    ],
+                                    "spans": [],
+                                    "index": 21
+                                },
+                                {
+                                    "bbox": [
+                                        157,
+                                        389.3333333333333,
+                                        453,
+                                        451.66666666666663
+                                    ],
+                                    "spans": [],
+                                    "index": 22
+                                },
+                                {
+                                    "bbox": [
+                                        157,
+                                        451.66666666666663,
+                                        453,
+                                        514.0
+                                    ],
+                                    "spans": [],
+                                    "index": 23
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 20.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        522,
+                        505,
+                        588
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                522,
+                                505,
+                                535
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        522,
+                                        505,
+                                        535
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Different embedding spaces. CACTUs is effective for a variety of embedding learning methods",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                533,
+                                505,
+                                546
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        533,
+                                        505,
+                                        546
+                                    ],
+                                    "score": 1.0,
+                                    "content": "used for task generation. The performance of unsupervised meta-learning can largely be predicted",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                544,
+                                505,
+                                557
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        544,
+                                        505,
+                                        557
+                                    ],
+                                    "score": 1.0,
+                                    "content": "by the performance of the embedding-based non-meta-learning methods. For example, the ACAI",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                555,
+                                505,
+                                568
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        555,
+                                        505,
+                                        568
+                                    ],
+                                    "score": 1.0,
+                                    "content": "embedding does well with Omniglot, leading to the best unsupervised results with ACAI CACTUs-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                565,
+                                505,
+                                579
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        565,
+                                        505,
+                                        579
+                                    ],
+                                    "score": 1.0,
+                                    "content": "MAML. Likewise, on miniImageNet, the best performing prior embedding (DeepCluster) also cor-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                577,
+                                478,
+                                590
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        577,
+                                        478,
+                                        590
+                                    ],
+                                    "score": 1.0,
+                                    "content": "responds to the best performing unsupervised meta-learner (DeepCluster CACTUs-MAML).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        }
+                    ],
+                    "index": 26.5,
+                    "bbox_fs": [
+                        105,
+                        522,
+                        505,
+                        590
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        594,
+                        505,
+                        672
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                594,
+                                505,
+                                606
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        594,
+                                        505,
+                                        606
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Applicability to different tasks. CACTUs-MAML learns an effective prior for a variety of task",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                605,
+                                506,
+                                618
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        605,
+                                        506,
+                                        618
+                                    ],
+                                    "score": 1.0,
+                                    "content": "types. This can be attributed to the application-agnostic task-generation process and the expressive",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                615,
+                                506,
+                                628
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        615,
+                                        506,
+                                        628
+                                    ],
+                                    "score": 1.0,
+                                    "content": "power of MAML (Finn & Levine, 2018). We also observe that, despite all meta-learning models",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                627,
+                                506,
+                                639
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        627,
+                                        178,
+                                        639
+                                    ],
+                                    "score": 1.0,
+                                    "content": "being trained for",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        179,
+                                        627,
+                                        189,
+                                        637
+                                    ],
+                                    "score": 0.82,
+                                    "content": "N",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        189,
+                                        627,
+                                        506,
+                                        639
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-way 1-shot classification of unsupervised tasks, the models work well for a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                637,
+                                505,
+                                650
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        637,
+                                        150,
+                                        650
+                                    ],
+                                    "score": 1.0,
+                                    "content": "variety of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        150,
+                                        638,
+                                        162,
+                                        648
+                                    ],
+                                    "score": 0.79,
+                                    "content": "M",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        162,
+                                        637,
+                                        184,
+                                        650
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-way",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        185,
+                                        639,
+                                        195,
+                                        648
+                                    ],
+                                    "score": 0.82,
+                                    "content": "K",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        195,
+                                        637,
+                                        272,
+                                        650
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-shot tasks, where",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        273,
+                                        638,
+                                        313,
+                                        649
+                                    ],
+                                    "score": 0.92,
+                                    "content": "M \\ \\leq \\ N",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        313,
+                                        637,
+                                        333,
+                                        650
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        334,
+                                        638,
+                                        372,
+                                        649
+                                    ],
+                                    "score": 0.91,
+                                    "content": "K \\ \\leq \\ 5 0",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        373,
+                                        637,
+                                        505,
+                                        650
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". As mentioned previously, the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                649,
+                                505,
+                                660
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        649,
+                                        505,
+                                        660
+                                    ],
+                                    "score": 1.0,
+                                    "content": "representation that CACTUs-ProtoNets learns is best suited for downstream tasks which match the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                660,
+                                245,
+                                673
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        660,
+                                        245,
+                                        673
+                                    ],
+                                    "score": 1.0,
+                                    "content": "single shot used for meta-training.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        }
+                    ],
+                    "index": 33,
+                    "bbox_fs": [
+                        105,
+                        594,
+                        506,
+                        673
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        677,
+                        504,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                676,
+                                506,
+                                690
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        676,
+                                        506,
+                                        690
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Oracle. The penalty for not having ground truth labels to construct near-optimal tasks ranges from",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                687,
+                                505,
+                                699
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        687,
+                                        505,
+                                        699
+                                    ],
+                                    "score": 1.0,
+                                    "content": "substantial to severe, depending on the difficulty of the downstream task. Easier downstream tasks",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                699,
+                                505,
+                                711
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        699,
+                                        505,
+                                        711
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(which have fewer classes and/or more supervision) incur less of a penalty. We conjecture that",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                708,
+                                506,
+                                723
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        708,
+                                        506,
+                                        723
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with such tasks, the difference in the usefulness of the priors matters less since the downstream",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                721,
+                                352,
+                                733
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        721,
+                                        352,
+                                        733
+                                    ],
+                                    "score": 1.0,
+                                    "content": "task-specific evidence has more power to shape the posterior.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        }
+                    ],
+                    "index": 39,
+                    "bbox_fs": [
+                        105,
+                        676,
+                        506,
+                        733
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "table",
+                    "bbox": [
+                        129,
+                        126,
+                        482,
+                        323
+                    ],
+                    "blocks": [
+                        {
+                            "type": "table_caption",
+                            "bbox": [
+                                107,
+                                80,
+                                505,
+                                121
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        79,
+                                        505,
+                                        92
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                79,
+                                                505,
+                                                92
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Table 2: Results of unsupervised learning on miniImageNet and CelebA images, averaged over 1000 down-",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 0
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        88,
+                                        505,
+                                        103
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                88,
+                                                314,
+                                                103
+                                            ],
+                                            "score": 1.0,
+                                            "content": "stream human-designed tasks. CACTUs experiments use",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                314,
+                                                90,
+                                                348,
+                                                100
+                                            ],
+                                            "score": 0.9,
+                                            "content": "k = 5 0 0",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                348,
+                                                88,
+                                                390,
+                                                103
+                                            ],
+                                            "score": 1.0,
+                                            "content": "for each of",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                390,
+                                                90,
+                                                421,
+                                                100
+                                            ],
+                                            "score": 0.9,
+                                            "content": "P = 5 0",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                421,
+                                                88,
+                                                505,
+                                                103
+                                            ],
+                                            "score": 1.0,
+                                            "content": "partitions. Embedding",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        100,
+                                        505,
+                                        111
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                100,
+                                                226,
+                                                111
+                                            ],
+                                            "score": 1.0,
+                                            "content": "cluster matching uses the same",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                226,
+                                                101,
+                                                232,
+                                                109
+                                            ],
+                                            "score": 0.71,
+                                            "content": "k",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                232,
+                                                100,
+                                                505,
+                                                111
+                                            ],
+                                            "score": 1.0,
+                                            "content": ". For complete results with confidence intervals, see Tables 9 and 10 in",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 2
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        110,
+                                        154,
+                                        122
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                110,
+                                                154,
+                                                122
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Appendix F.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 3
+                                }
+                            ],
+                            "index": 1.5
+                        },
+                        {
+                            "type": "table_body",
+                            "bbox": [
+                                129,
+                                126,
+                                482,
+                                323
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        129,
+                                        126,
+                                        482,
+                                        323
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                129,
+                                                126,
+                                                482,
+                                                323
+                                            ],
+                                            "score": 0.984,
+                                            "html": "<table><tr><td></td><td colspan=\"4\">miniImageNet</td><td>CelebA</td></tr><tr><td></td><td>(5,1)</td><td>(5,5)</td><td>(5,20)</td><td>(5,50)</td><td>(2,5)</td></tr><tr><td>Training from scratch</td><td>27.59%</td><td>38.48%</td><td>51.53%</td><td>59.63%</td><td>63.19%</td></tr><tr><td>BiGAN knn-nearest neighbors</td><td>25.56%</td><td>31.10%</td><td>37.31%</td><td>43.60%</td><td>56.15%</td></tr><tr><td>BiGAN linear classifier</td><td>27.08%</td><td>33.91%</td><td>44.00%</td><td>50.41%</td><td>58.44%</td></tr><tr><td>BiGAN MLP with dropout</td><td>22.91%</td><td>29.06%</td><td>40.06%</td><td>48.36%</td><td>56.26%</td></tr><tr><td>BiGAN cluster matching</td><td>24.63%</td><td>29.49%</td><td>33.89%</td><td>36.13%</td><td>56.20%</td></tr><tr><td>BiGAN CACTUs-MAML (ours)</td><td>36.24%</td><td>51.28%</td><td>61.33%</td><td>66.91%</td><td>74.98%</td></tr><tr><td>BiGAN CACTUs-ProtoNets (ours)</td><td>36.62%</td><td>50.16%</td><td>59.56%</td><td>63.27%</td><td>65.58%</td></tr><tr><td>DeepCluster knn-nearest neighbors</td><td>28.90%</td><td>42.25%</td><td>56.44%</td><td>63.90%</td><td>61.47%</td></tr><tr><td>DeepCluster linear classifier</td><td>29.44%</td><td>39.79%</td><td>56.19%</td><td>65.28%</td><td>59.57%</td></tr><tr><td>DeepCluster MLP with dropout</td><td>29.03%</td><td>39.67%</td><td>52.71%</td><td>60.95%</td><td>60.65%</td></tr><tr><td>DeepCluster cluster matching</td><td>22.20%</td><td>23.50%</td><td>24.97%</td><td>26.87%</td><td>51.51%</td></tr><tr><td>DeepCluster CACTUs-MAML (ours)</td><td>39.90%</td><td>53.97%</td><td>63.84%</td><td>69.64%</td><td>73.79%</td></tr><tr><td>DeepCluster CACTUs-ProtoNets (ours)</td><td>39.18%</td><td>53.36%</td><td>61.54%</td><td>63.55%</td><td>74.15%</td></tr><tr><td>Oracle-MAML (control)</td><td>46.81%</td><td>62.13%</td><td>71.03%</td><td>75.54%</td><td>87.10%</td></tr><tr><td>Oracle-ProtoNets (control)</td><td>46.56%</td><td>62.29%</td><td>70.05%</td><td>72.04%</td><td>85.13%</td></tr></table>",
+                                            "type": "table",
+                                            "image_path": "b3cfea21fbe70207bc75235f3f925ee17c07ae5244ac6a3361c724c2f235ba9a.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 5,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        129,
+                                        126,
+                                        482,
+                                        191.66666666666669
+                                    ],
+                                    "spans": [],
+                                    "index": 4
+                                },
+                                {
+                                    "bbox": [
+                                        129,
+                                        191.66666666666669,
+                                        482,
+                                        257.33333333333337
+                                    ],
+                                    "spans": [],
+                                    "index": 5
+                                },
+                                {
+                                    "bbox": [
+                                        129,
+                                        257.33333333333337,
+                                        482,
+                                        323.00000000000006
+                                    ],
+                                    "spans": [],
+                                    "index": 6
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 3.25
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        343,
+                        505,
+                        464
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                343,
+                                505,
+                                356
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        343,
+                                        505,
+                                        356
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Task construction ablation. As seen in Tables 3 and 4, CACTUs-MAML consistently outperforms",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                354,
+                                505,
+                                367
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        354,
+                                        505,
+                                        367
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Hyperplanes-MAML with any margin. We hypothesize that this is due to the issues with zero-margin",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                365,
+                                505,
+                                378
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        365,
+                                        505,
+                                        378
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Hyperplanes-MAML pointed out in Appendix C, and the fact that nonzero-margin Hyperplanes-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                375,
+                                505,
+                                389
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        375,
+                                        505,
+                                        389
+                                    ],
+                                    "score": 1.0,
+                                    "content": "MAML is able to use less of the meta-training split to generate tasks than CACTUs-MAML is.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                385,
+                                506,
+                                400
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        385,
+                                        506,
+                                        400
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We find that using multiple partitions for CACTUs-MAML, while beneficial, is not strictly neces-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                398,
+                                505,
+                                410
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        398,
+                                        505,
+                                        410
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sary. Using non-zero margin with Hyperplanes-MAML is crucial for miniImageNet, but not for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                408,
+                                505,
+                                421
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        408,
+                                        505,
+                                        421
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Omniglot. We conjecture that the enforced degree of separation between classes is needed for mini-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                420,
+                                505,
+                                432
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        420,
+                                        505,
+                                        432
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ImageNet because of the dataset’s high diversity. Meta-learning on random tasks or tasks derived",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                430,
+                                505,
+                                443
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        430,
+                                        505,
+                                        443
+                                    ],
+                                    "score": 1.0,
+                                    "content": "from pixel-space clustering (Table 3) results in a prior that is much less useful than any other con-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                441,
+                                505,
+                                455
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        441,
+                                        505,
+                                        455
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sidered algorithm, including a random network initialization; evidently, practicing badly is worse",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                452,
+                                208,
+                                465
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        452,
+                                        208,
+                                        465
+                                    ],
+                                    "score": 1.0,
+                                    "content": "than not practicing at all.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        }
+                    ],
+                    "index": 12
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        469,
+                        505,
+                        547
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                468,
+                                505,
+                                482
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        468,
+                                        505,
+                                        482
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Note on overfitting. Because of the combinatorially many unsupervised tasks we can create from",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                480,
+                                505,
+                                493
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        480,
+                                        505,
+                                        493
+                                    ],
+                                    "score": 1.0,
+                                    "content": "multiple partitions of the dataset, we do not observe substantial overfitting to the unsupervised meta-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                492,
+                                505,
+                                504
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        492,
+                                        505,
+                                        504
+                                    ],
+                                    "score": 1.0,
+                                    "content": "training tasks. However, we observe that meta-training performance is sometimes worse than meta-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                501,
+                                505,
+                                516
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        501,
+                                        505,
+                                        516
+                                    ],
+                                    "score": 1.0,
+                                    "content": "test time performance, which is likely due to a portion of the automatically generated tasks being",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                513,
+                                504,
+                                525
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        513,
+                                        504,
+                                        525
+                                    ],
+                                    "score": 1.0,
+                                    "content": "based on nonsensical clusters (for examples, see Figure 2). Additionally, we find that, with a few",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                525,
+                                506,
+                                536
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        525,
+                                        506,
+                                        536
+                                    ],
+                                    "score": 1.0,
+                                    "content": "exceptions, using multiple partitions has a regularizing effect on the meta-learner: a diverse task set",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                536,
+                                489,
+                                547
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        536,
+                                        489,
+                                        547
+                                    ],
+                                    "score": 1.0,
+                                    "content": "reduces overfitting to the meta-training tasks and increases the applicability of the learned prior.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        }
+                    ],
+                    "index": 21
+                },
+                {
+                    "type": "table",
+                    "bbox": [
+                        129,
+                        586,
+                        482,
+                        718
+                    ],
+                    "blocks": [
+                        {
+                            "type": "table_caption",
+                            "bbox": [
+                                106,
+                                560,
+                                504,
+                                581
+                            ],
+                            "group_id": 1,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        560,
+                                        505,
+                                        572
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                560,
+                                                505,
+                                                572
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Table 3: Ablation study of task construction methods on Omniglot. For a more complete set of results with",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 25
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        570,
+                                        282,
+                                        581
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                570,
+                                                282,
+                                                581
+                                            ],
+                                            "score": 1.0,
+                                            "content": "confidence intervals, see Table 8 in Appendix F.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 26
+                                }
+                            ],
+                            "index": 25.5
+                        },
+                        {
+                            "type": "table_body",
+                            "bbox": [
+                                129,
+                                586,
+                                482,
+                                718
+                            ],
+                            "group_id": 1,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        129,
+                                        586,
+                                        482,
+                                        718
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                129,
+                                                586,
+                                                482,
+                                                718
+                                            ],
+                                            "score": 0.984,
+                                            "html": "<table><tr><td>Algorithm</td><td>(5,1)</td><td>(5,5)</td><td>(20,1)</td><td>(20,5)</td></tr><tr><td>Random-MAML,P= 2400,k = 500</td><td>25.99%</td><td>25.74%</td><td>6.51%</td><td>6.74%</td></tr><tr><td>Pixels CACTUs-MAML,P=1, k = 500</td><td>30.55%</td><td>40.19%</td><td>12.05%</td><td>19.01%</td></tr><tr><td>ACAIHyperplanes-MAML,P= 2400,m= 0</td><td>62.34%</td><td>81.81%</td><td>39.30%</td><td>63.18%</td></tr><tr><td>ACAI Hyperplanes-MAML,P= 2400,m =1.2</td><td>62.44%</td><td>83.20%</td><td>41.86%</td><td>65.23%</td></tr><tr><td>ACAI CACTUs-MAML,P=1,k = 500</td><td>66.49%</td><td>85.60%</td><td>45.04%</td><td>69.14%</td></tr><tr><td>ACAICACTUs-MAML,P=100,k = 500</td><td>68.84%</td><td>87.78%</td><td>48.09%</td><td>73.36%</td></tr><tr><td>BiGANHyperplanes-MAML,P= 2400,m = 0</td><td>53.60%</td><td>74.60%</td><td>29.02%</td><td>50.77%</td></tr><tr><td>BiGANHyperplanes-MAML,P= 2400,m = 0.5</td><td>53.18%</td><td>73.55%</td><td>29.98%</td><td>50.14%</td></tr><tr><td>BiGANCACTUs-MAML,P=1,k= 500</td><td>55.92%</td><td>76.28%</td><td>32.44%</td><td>54.22%</td></tr><tr><td>BiGANCACTUs-MAML,P=100,k= 500</td><td>58.18%</td><td>78.66%</td><td>35.56%</td><td>58.62%</td></tr></table>",
+                                            "type": "table",
+                                            "image_path": "e280051126fea76121cf052c226a6d15683009963fc30dd690979c1412e9f902.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 28,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        129,
+                                        586,
+                                        482,
+                                        630.0
+                                    ],
+                                    "spans": [],
+                                    "index": 27
+                                },
+                                {
+                                    "bbox": [
+                                        129,
+                                        630.0,
+                                        482,
+                                        674.0
+                                    ],
+                                    "spans": [],
+                                    "index": 28
+                                },
+                                {
+                                    "bbox": [
+                                        129,
+                                        674.0,
+                                        482,
+                                        718.0
+                                    ],
+                                    "spans": [],
+                                    "index": 29
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 26.75
+                }
+            ],
+            "page_idx": 8,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        108,
+                        27,
+                        293,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                26,
+                                294,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        26,
+                                        294,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Published as a conference paper at ICLR 2019",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        302,
+                        751,
+                        309,
+                        759
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                302,
+                                751,
+                                309,
+                                762
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        302,
+                                        751,
+                                        309,
+                                        762
+                                    ],
+                                    "score": 1.0,
+                                    "content": "9",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "table",
+                    "bbox": [
+                        129,
+                        126,
+                        482,
+                        323
+                    ],
+                    "blocks": [
+                        {
+                            "type": "table_caption",
+                            "bbox": [
+                                107,
+                                80,
+                                505,
+                                121
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        79,
+                                        505,
+                                        92
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                79,
+                                                505,
+                                                92
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Table 2: Results of unsupervised learning on miniImageNet and CelebA images, averaged over 1000 down-",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 0
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        88,
+                                        505,
+                                        103
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                88,
+                                                314,
+                                                103
+                                            ],
+                                            "score": 1.0,
+                                            "content": "stream human-designed tasks. CACTUs experiments use",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                314,
+                                                90,
+                                                348,
+                                                100
+                                            ],
+                                            "score": 0.9,
+                                            "content": "k = 5 0 0",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                348,
+                                                88,
+                                                390,
+                                                103
+                                            ],
+                                            "score": 1.0,
+                                            "content": "for each of",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                390,
+                                                90,
+                                                421,
+                                                100
+                                            ],
+                                            "score": 0.9,
+                                            "content": "P = 5 0",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                421,
+                                                88,
+                                                505,
+                                                103
+                                            ],
+                                            "score": 1.0,
+                                            "content": "partitions. Embedding",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        100,
+                                        505,
+                                        111
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                100,
+                                                226,
+                                                111
+                                            ],
+                                            "score": 1.0,
+                                            "content": "cluster matching uses the same",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                226,
+                                                101,
+                                                232,
+                                                109
+                                            ],
+                                            "score": 0.71,
+                                            "content": "k",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                232,
+                                                100,
+                                                505,
+                                                111
+                                            ],
+                                            "score": 1.0,
+                                            "content": ". For complete results with confidence intervals, see Tables 9 and 10 in",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 2
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        110,
+                                        154,
+                                        122
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                110,
+                                                154,
+                                                122
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Appendix F.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 3
+                                }
+                            ],
+                            "index": 1.5
+                        },
+                        {
+                            "type": "table_body",
+                            "bbox": [
+                                129,
+                                126,
+                                482,
+                                323
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        129,
+                                        126,
+                                        482,
+                                        323
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                129,
+                                                126,
+                                                482,
+                                                323
+                                            ],
+                                            "score": 0.984,
+                                            "html": "<table><tr><td></td><td colspan=\"4\">miniImageNet</td><td>CelebA</td></tr><tr><td></td><td>(5,1)</td><td>(5,5)</td><td>(5,20)</td><td>(5,50)</td><td>(2,5)</td></tr><tr><td>Training from scratch</td><td>27.59%</td><td>38.48%</td><td>51.53%</td><td>59.63%</td><td>63.19%</td></tr><tr><td>BiGAN knn-nearest neighbors</td><td>25.56%</td><td>31.10%</td><td>37.31%</td><td>43.60%</td><td>56.15%</td></tr><tr><td>BiGAN linear classifier</td><td>27.08%</td><td>33.91%</td><td>44.00%</td><td>50.41%</td><td>58.44%</td></tr><tr><td>BiGAN MLP with dropout</td><td>22.91%</td><td>29.06%</td><td>40.06%</td><td>48.36%</td><td>56.26%</td></tr><tr><td>BiGAN cluster matching</td><td>24.63%</td><td>29.49%</td><td>33.89%</td><td>36.13%</td><td>56.20%</td></tr><tr><td>BiGAN CACTUs-MAML (ours)</td><td>36.24%</td><td>51.28%</td><td>61.33%</td><td>66.91%</td><td>74.98%</td></tr><tr><td>BiGAN CACTUs-ProtoNets (ours)</td><td>36.62%</td><td>50.16%</td><td>59.56%</td><td>63.27%</td><td>65.58%</td></tr><tr><td>DeepCluster knn-nearest neighbors</td><td>28.90%</td><td>42.25%</td><td>56.44%</td><td>63.90%</td><td>61.47%</td></tr><tr><td>DeepCluster linear classifier</td><td>29.44%</td><td>39.79%</td><td>56.19%</td><td>65.28%</td><td>59.57%</td></tr><tr><td>DeepCluster MLP with dropout</td><td>29.03%</td><td>39.67%</td><td>52.71%</td><td>60.95%</td><td>60.65%</td></tr><tr><td>DeepCluster cluster matching</td><td>22.20%</td><td>23.50%</td><td>24.97%</td><td>26.87%</td><td>51.51%</td></tr><tr><td>DeepCluster CACTUs-MAML (ours)</td><td>39.90%</td><td>53.97%</td><td>63.84%</td><td>69.64%</td><td>73.79%</td></tr><tr><td>DeepCluster CACTUs-ProtoNets (ours)</td><td>39.18%</td><td>53.36%</td><td>61.54%</td><td>63.55%</td><td>74.15%</td></tr><tr><td>Oracle-MAML (control)</td><td>46.81%</td><td>62.13%</td><td>71.03%</td><td>75.54%</td><td>87.10%</td></tr><tr><td>Oracle-ProtoNets (control)</td><td>46.56%</td><td>62.29%</td><td>70.05%</td><td>72.04%</td><td>85.13%</td></tr></table>",
+                                            "type": "table",
+                                            "image_path": "b3cfea21fbe70207bc75235f3f925ee17c07ae5244ac6a3361c724c2f235ba9a.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 5,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        129,
+                                        126,
+                                        482,
+                                        191.66666666666669
+                                    ],
+                                    "spans": [],
+                                    "index": 4
+                                },
+                                {
+                                    "bbox": [
+                                        129,
+                                        191.66666666666669,
+                                        482,
+                                        257.33333333333337
+                                    ],
+                                    "spans": [],
+                                    "index": 5
+                                },
+                                {
+                                    "bbox": [
+                                        129,
+                                        257.33333333333337,
+                                        482,
+                                        323.00000000000006
+                                    ],
+                                    "spans": [],
+                                    "index": 6
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 3.25
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        343,
+                        505,
+                        464
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                343,
+                                505,
+                                356
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        343,
+                                        505,
+                                        356
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Task construction ablation. As seen in Tables 3 and 4, CACTUs-MAML consistently outperforms",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                354,
+                                505,
+                                367
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        354,
+                                        505,
+                                        367
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Hyperplanes-MAML with any margin. We hypothesize that this is due to the issues with zero-margin",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                365,
+                                505,
+                                378
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        365,
+                                        505,
+                                        378
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Hyperplanes-MAML pointed out in Appendix C, and the fact that nonzero-margin Hyperplanes-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                375,
+                                505,
+                                389
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        375,
+                                        505,
+                                        389
+                                    ],
+                                    "score": 1.0,
+                                    "content": "MAML is able to use less of the meta-training split to generate tasks than CACTUs-MAML is.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                385,
+                                506,
+                                400
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        385,
+                                        506,
+                                        400
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We find that using multiple partitions for CACTUs-MAML, while beneficial, is not strictly neces-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                398,
+                                505,
+                                410
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        398,
+                                        505,
+                                        410
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sary. Using non-zero margin with Hyperplanes-MAML is crucial for miniImageNet, but not for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                408,
+                                505,
+                                421
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        408,
+                                        505,
+                                        421
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Omniglot. We conjecture that the enforced degree of separation between classes is needed for mini-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                420,
+                                505,
+                                432
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        420,
+                                        505,
+                                        432
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ImageNet because of the dataset’s high diversity. Meta-learning on random tasks or tasks derived",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                430,
+                                505,
+                                443
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        430,
+                                        505,
+                                        443
+                                    ],
+                                    "score": 1.0,
+                                    "content": "from pixel-space clustering (Table 3) results in a prior that is much less useful than any other con-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                441,
+                                505,
+                                455
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        441,
+                                        505,
+                                        455
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sidered algorithm, including a random network initialization; evidently, practicing badly is worse",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                452,
+                                208,
+                                465
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        452,
+                                        208,
+                                        465
+                                    ],
+                                    "score": 1.0,
+                                    "content": "than not practicing at all.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        }
+                    ],
+                    "index": 12,
+                    "bbox_fs": [
+                        105,
+                        343,
+                        506,
+                        465
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        469,
+                        505,
+                        547
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                468,
+                                505,
+                                482
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        468,
+                                        505,
+                                        482
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Note on overfitting. Because of the combinatorially many unsupervised tasks we can create from",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                480,
+                                505,
+                                493
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        480,
+                                        505,
+                                        493
+                                    ],
+                                    "score": 1.0,
+                                    "content": "multiple partitions of the dataset, we do not observe substantial overfitting to the unsupervised meta-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                492,
+                                505,
+                                504
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        492,
+                                        505,
+                                        504
+                                    ],
+                                    "score": 1.0,
+                                    "content": "training tasks. However, we observe that meta-training performance is sometimes worse than meta-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                501,
+                                505,
+                                516
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        501,
+                                        505,
+                                        516
+                                    ],
+                                    "score": 1.0,
+                                    "content": "test time performance, which is likely due to a portion of the automatically generated tasks being",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                513,
+                                504,
+                                525
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        513,
+                                        504,
+                                        525
+                                    ],
+                                    "score": 1.0,
+                                    "content": "based on nonsensical clusters (for examples, see Figure 2). Additionally, we find that, with a few",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                525,
+                                506,
+                                536
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        525,
+                                        506,
+                                        536
+                                    ],
+                                    "score": 1.0,
+                                    "content": "exceptions, using multiple partitions has a regularizing effect on the meta-learner: a diverse task set",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                536,
+                                489,
+                                547
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        536,
+                                        489,
+                                        547
+                                    ],
+                                    "score": 1.0,
+                                    "content": "reduces overfitting to the meta-training tasks and increases the applicability of the learned prior.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        }
+                    ],
+                    "index": 21,
+                    "bbox_fs": [
+                        104,
+                        468,
+                        506,
+                        547
+                    ]
+                },
+                {
+                    "type": "table",
+                    "bbox": [
+                        129,
+                        586,
+                        482,
+                        718
+                    ],
+                    "blocks": [
+                        {
+                            "type": "table_caption",
+                            "bbox": [
+                                106,
+                                560,
+                                504,
+                                581
+                            ],
+                            "group_id": 1,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        560,
+                                        505,
+                                        572
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                560,
+                                                505,
+                                                572
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Table 3: Ablation study of task construction methods on Omniglot. For a more complete set of results with",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 25
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        570,
+                                        282,
+                                        581
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                570,
+                                                282,
+                                                581
+                                            ],
+                                            "score": 1.0,
+                                            "content": "confidence intervals, see Table 8 in Appendix F.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 26
+                                }
+                            ],
+                            "index": 25.5
+                        },
+                        {
+                            "type": "table_body",
+                            "bbox": [
+                                129,
+                                586,
+                                482,
+                                718
+                            ],
+                            "group_id": 1,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        129,
+                                        586,
+                                        482,
+                                        718
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                129,
+                                                586,
+                                                482,
+                                                718
+                                            ],
+                                            "score": 0.984,
+                                            "html": "<table><tr><td>Algorithm</td><td>(5,1)</td><td>(5,5)</td><td>(20,1)</td><td>(20,5)</td></tr><tr><td>Random-MAML,P= 2400,k = 500</td><td>25.99%</td><td>25.74%</td><td>6.51%</td><td>6.74%</td></tr><tr><td>Pixels CACTUs-MAML,P=1, k = 500</td><td>30.55%</td><td>40.19%</td><td>12.05%</td><td>19.01%</td></tr><tr><td>ACAIHyperplanes-MAML,P= 2400,m= 0</td><td>62.34%</td><td>81.81%</td><td>39.30%</td><td>63.18%</td></tr><tr><td>ACAI Hyperplanes-MAML,P= 2400,m =1.2</td><td>62.44%</td><td>83.20%</td><td>41.86%</td><td>65.23%</td></tr><tr><td>ACAI CACTUs-MAML,P=1,k = 500</td><td>66.49%</td><td>85.60%</td><td>45.04%</td><td>69.14%</td></tr><tr><td>ACAICACTUs-MAML,P=100,k = 500</td><td>68.84%</td><td>87.78%</td><td>48.09%</td><td>73.36%</td></tr><tr><td>BiGANHyperplanes-MAML,P= 2400,m = 0</td><td>53.60%</td><td>74.60%</td><td>29.02%</td><td>50.77%</td></tr><tr><td>BiGANHyperplanes-MAML,P= 2400,m = 0.5</td><td>53.18%</td><td>73.55%</td><td>29.98%</td><td>50.14%</td></tr><tr><td>BiGANCACTUs-MAML,P=1,k= 500</td><td>55.92%</td><td>76.28%</td><td>32.44%</td><td>54.22%</td></tr><tr><td>BiGANCACTUs-MAML,P=100,k= 500</td><td>58.18%</td><td>78.66%</td><td>35.56%</td><td>58.62%</td></tr></table>",
+                                            "type": "table",
+                                            "image_path": "e280051126fea76121cf052c226a6d15683009963fc30dd690979c1412e9f902.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 28,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        129,
+                                        586,
+                                        482,
+                                        630.0
+                                    ],
+                                    "spans": [],
+                                    "index": 27
+                                },
+                                {
+                                    "bbox": [
+                                        129,
+                                        630.0,
+                                        482,
+                                        674.0
+                                    ],
+                                    "spans": [],
+                                    "index": 28
+                                },
+                                {
+                                    "bbox": [
+                                        129,
+                                        674.0,
+                                        482,
+                                        718.0
+                                    ],
+                                    "spans": [],
+                                    "index": 29
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 26.75
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "table",
+                    "bbox": [
+                        120,
+                        106,
+                        491,
+                        214
+                    ],
+                    "blocks": [
+                        {
+                            "type": "table_caption",
+                            "bbox": [
+                                107,
+                                80,
+                                503,
+                                101
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        81,
+                                        504,
+                                        91
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                81,
+                                                504,
+                                                91
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Table 4: Ablation study of task construction methods on miniImageNet. For a more complete set of results",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 0
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        90,
+                                        300,
+                                        101
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                90,
+                                                300,
+                                                101
+                                            ],
+                                            "score": 1.0,
+                                            "content": "with confidence intervals, see Table 9 in Appendix F.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 1
+                                }
+                            ],
+                            "index": 0.5
+                        },
+                        {
+                            "type": "table_body",
+                            "bbox": [
+                                120,
+                                106,
+                                491,
+                                214
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        120,
+                                        106,
+                                        491,
+                                        214
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                120,
+                                                106,
+                                                491,
+                                                214
+                                            ],
+                                            "score": 0.983,
+                                            "html": "<table><tr><td>Algorithm</td><td>(5,1)</td><td>(5,5)</td><td>(5,20)</td><td>(5,50)</td></tr><tr><td>BiGANHyperplanes-MAML,P= 4800,m =0</td><td>20.00%</td><td>20.00%</td><td>20.00%</td><td>20.00%</td></tr><tr><td>BiGAN Hyperplanes-MAML,P= 4800,m = 0.9</td><td>29.67%</td><td>41.92%</td><td>51.32%</td><td>54.72%</td></tr><tr><td>BiGAN CACTUs-MAML,P=1,k= 500</td><td>37.75%</td><td>52.59%</td><td>62.70%</td><td>67.98%</td></tr><tr><td>BiGANCACTUs-MAML,P= 50,k= 500</td><td>36.24%</td><td>51.28%</td><td>61.33%</td><td>66.91%</td></tr><tr><td>DeepCluster Hyperplanes-MAML,P= 4800,m = 0</td><td>20.02%</td><td>20.01%</td><td>20.00%</td><td>20.01%</td></tr><tr><td>DeepClusterHyperplanes-MAML,P= 4800,m= 0.1</td><td>35.85%</td><td>49.54%</td><td>60.68%</td><td>65.55%</td></tr><tr><td>DeepCluster CACTUs-MAML,P= 1,k = 500</td><td>38.75%</td><td>52.73%</td><td>62.72%</td><td>67.77%</td></tr><tr><td>DeepCluster CACTUs-MAML,P= 50,k = 500</td><td>39.90%</td><td>53.97%</td><td>63.84%</td><td>69.64%</td></tr></table>",
+                                            "type": "table",
+                                            "image_path": "0a65d9b00a7e20305e684fb38d2993d23ba0551b26d704c18abfd4d99ddb29c3.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 3,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        120,
+                                        106,
+                                        491,
+                                        142.0
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                },
+                                {
+                                    "bbox": [
+                                        120,
+                                        142.0,
+                                        491,
+                                        178.0
+                                    ],
+                                    "spans": [],
+                                    "index": 3
+                                },
+                                {
+                                    "bbox": [
+                                        120,
+                                        178.0,
+                                        491,
+                                        214.0
+                                    ],
+                                    "spans": [],
+                                    "index": 4
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 1.75
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        246,
+                        190,
+                        259
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                246,
+                                192,
+                                262
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        246,
+                                        192,
+                                        262
+                                    ],
+                                    "score": 1.0,
+                                    "content": "5 DISCUSSION",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        }
+                    ],
+                    "index": 5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        280,
+                        505,
+                        336
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                280,
+                                505,
+                                294
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        280,
+                                        505,
+                                        294
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We demonstrate that meta-learning on tasks produced using simple mechanisms based on embed-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                292,
+                                505,
+                                304
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        292,
+                                        505,
+                                        304
+                                    ],
+                                    "score": 1.0,
+                                    "content": "dings improves upon the utility of these representations in learning downstream, human-specified",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                303,
+                                505,
+                                315
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        303,
+                                        505,
+                                        315
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tasks. We empirically show that this holds across benchmark datasets and tasks in the few-shot",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                314,
+                                504,
+                                326
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        314,
+                                        504,
+                                        326
+                                    ],
+                                    "score": 1.0,
+                                    "content": "classification literature (Santoro et al., 2016; Ravi & Larochelle, 2017; Finn et al., 2018), task diffi-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                325,
+                                501,
+                                337
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        325,
+                                        501,
+                                        337
+                                    ],
+                                    "score": 1.0,
+                                    "content": "culties, and embedding learning methods while fixing key hyperparameters across all experiments.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        }
+                    ],
+                    "index": 8
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        342,
+                        505,
+                        484
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                341,
+                                505,
+                                354
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        341,
+                                        505,
+                                        354
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In a sense, CACTUs can be seen as a facilitating interface between an embedding learning method",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                352,
+                                505,
+                                365
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        352,
+                                        505,
+                                        365
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and a meta-learning algorithm. As shown in the results, the meta-learner’s performance significantly",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                365,
+                                505,
+                                376
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        365,
+                                        505,
+                                        376
+                                    ],
+                                    "score": 1.0,
+                                    "content": "depends on the nature and quality of the task-generating embeddings. We can expect our method",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                374,
+                                505,
+                                387
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        374,
+                                        505,
+                                        387
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to yield better performance as the methods that produce these embedding functions improve, be-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                385,
+                                504,
+                                398
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        385,
+                                        504,
+                                        398
+                                    ],
+                                    "score": 1.0,
+                                    "content": "coming better suited for generating diverse yet distinctive clusterings of the data. However, the gap",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                397,
+                                505,
+                                408
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        397,
+                                        505,
+                                        408
+                                    ],
+                                    "score": 1.0,
+                                    "content": "between unsupervised and supervised meta-learning will likely persist because, with the latter, the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                408,
+                                505,
+                                420
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        408,
+                                        505,
+                                        420
+                                    ],
+                                    "score": 1.0,
+                                    "content": "meta-training task distribution is human-designed to mimic the expected evaluation task distribu-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                419,
+                                505,
+                                432
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        419,
+                                        505,
+                                        432
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tion as much as possible. Indeed, to some extent, supervised meta-learning algorithms offload the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                429,
+                                505,
+                                442
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        429,
+                                        505,
+                                        442
+                                    ],
+                                    "score": 1.0,
+                                    "content": "effort of designing and tuning algorithms onto the effort of designing and tuning task distributions.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                439,
+                                505,
+                                454
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        439,
+                                        505,
+                                        454
+                                    ],
+                                    "score": 1.0,
+                                    "content": "With its evaluation-agnostic task generation, CACTUs-based meta-learning trades off performance",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                451,
+                                504,
+                                464
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        451,
+                                        504,
+                                        464
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in specific use-cases for broad applicability and the ability to train on unlabeled data. In principle,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                461,
+                                506,
+                                476
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        461,
+                                        506,
+                                        476
+                                    ],
+                                    "score": 1.0,
+                                    "content": "CACTUs-based meta-learning may outperform supervised meta-learning when the latter is trained",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                474,
+                                412,
+                                486
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        474,
+                                        412,
+                                        486
+                                    ],
+                                    "score": 1.0,
+                                    "content": "on a misaligned task distribution. We leave this investigation to future work.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        }
+                    ],
+                    "index": 17
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        490,
+                        505,
+                        567
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                489,
+                                505,
+                                502
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        489,
+                                        247,
+                                        502
+                                    ],
+                                    "score": 1.0,
+                                    "content": "While we have demonstrated that",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        248,
+                                        491,
+                                        254,
+                                        500
+                                    ],
+                                    "score": 0.82,
+                                    "content": "k",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        254,
+                                        489,
+                                        505,
+                                        502
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-means is a broadly useful mechanism for constructing tasks",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                501,
+                                505,
+                                513
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        501,
+                                        318,
+                                        513
+                                    ],
+                                    "score": 1.0,
+                                    "content": "from embeddings, it is unlikely that combinations of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        318,
+                                        502,
+                                        325,
+                                        511
+                                    ],
+                                    "score": 0.83,
+                                    "content": "k",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        325,
+                                        501,
+                                        505,
+                                        513
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-means clusters in learned embedding spaces",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                512,
+                                505,
+                                524
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        512,
+                                        505,
+                                        524
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are universal approximations of arbitrary class definitions. An important direction for future work",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                522,
+                                505,
+                                537
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        522,
+                                        505,
+                                        537
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is to find examples of datasets and human-designed tasks for which CACTUs-based meta-learning",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                534,
+                                505,
+                                546
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        534,
+                                        505,
+                                        546
+                                    ],
+                                    "score": 1.0,
+                                    "content": "results in ineffective downstream learning. This will result in better understanding of the practical",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                545,
+                                505,
+                                557
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        545,
+                                        505,
+                                        557
+                                    ],
+                                    "score": 1.0,
+                                    "content": "scope of applicability for our method, and spur further development in automatic task construction",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                555,
+                                287,
+                                569
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        555,
+                                        287,
+                                        569
+                                    ],
+                                    "score": 1.0,
+                                    "content": "mechanisms for unsupervised meta-learning.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        }
+                    ],
+                    "index": 27
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        573,
+                        505,
+                        650
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                573,
+                                505,
+                                585
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        573,
+                                        505,
+                                        585
+                                    ],
+                                    "score": 1.0,
+                                    "content": "A potential concern of our experimental evaluation is that MNIST, Omniglot, and miniImageNet",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                584,
+                                505,
+                                596
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        584,
+                                        505,
+                                        596
+                                    ],
+                                    "score": 1.0,
+                                    "content": "exhibit particular structure in the underlying class distribution (i.e., perfectly balanced classes), since",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                595,
+                                505,
+                                607
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        595,
+                                        505,
+                                        607
+                                    ],
+                                    "score": 1.0,
+                                    "content": "they were designed to be supervised learning benchmarks. In more practical applications of machine",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                606,
+                                505,
+                                618
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        606,
+                                        505,
+                                        618
+                                    ],
+                                    "score": 1.0,
+                                    "content": "learning, such structure would likely not exist. Our CelebA results indicate that CACTUs is effective",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                616,
+                                505,
+                                629
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        616,
+                                        505,
+                                        629
+                                    ],
+                                    "score": 1.0,
+                                    "content": "even in the case of a dataset without neatly balanced classes or attributes. An interesting direction",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                627,
+                                505,
+                                641
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        627,
+                                        505,
+                                        641
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for future work is to better characterize the performance of CACTUs and other unsupervised pre-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                639,
+                                358,
+                                651
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        639,
+                                        358,
+                                        651
+                                    ],
+                                    "score": 1.0,
+                                    "content": "training methods with highly-unstructured, unlabeled datasets.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        }
+                    ],
+                    "index": 34
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        655,
+                        505,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                655,
+                                505,
+                                668
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        655,
+                                        505,
+                                        668
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Since MAML and ProtoNets produce nothing more than a learned representation, our method can",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                666,
+                                505,
+                                680
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        666,
+                                        505,
+                                        680
+                                    ],
+                                    "score": 1.0,
+                                    "content": "be viewed as deriving, from a previous unsupervised representation, a new representation particu-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                677,
+                                505,
+                                690
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        677,
+                                        505,
+                                        690
+                                    ],
+                                    "score": 1.0,
+                                    "content": "larly suited for learning downstream tasks. Beyond visual classification tasks, the notion of using",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                688,
+                                505,
+                                702
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        688,
+                                        505,
+                                        702
+                                    ],
+                                    "score": 1.0,
+                                    "content": "unsupervised pre-training is generally applicable to a wide range of domains, including regression,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                699,
+                                506,
+                                712
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        699,
+                                        506,
+                                        712
+                                    ],
+                                    "score": 1.0,
+                                    "content": "speech (Oord et al., 2018), language (Howard & Ruder, 2018), and reinforcement learning (Shel-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                710,
+                                506,
+                                724
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        710,
+                                        506,
+                                        724
+                                    ],
+                                    "score": 1.0,
+                                    "content": "hamer et al., 2017). Hence, our unsupervised meta-learning approach has the potential to improve",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                721,
+                                488,
+                                734
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        721,
+                                        488,
+                                        734
+                                    ],
+                                    "score": 1.0,
+                                    "content": "unsupervised representations for a variety of such domains, an exciting avenue for future work.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        }
+                    ],
+                    "index": 41
+                }
+            ],
+            "page_idx": 9,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        108,
+                        27,
+                        293,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                26,
+                                294,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        26,
+                                        294,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Published as a conference paper at ICLR 2019",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        301,
+                        751,
+                        311,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                299,
+                                750,
+                                313,
+                                765
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        299,
+                                        750,
+                                        313,
+                                        765
+                                    ],
+                                    "score": 1.0,
+                                    "content": "",
+                                    "type": "text",
+                                    "height": 15,
+                                    "width": 14
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "table",
+                    "bbox": [
+                        120,
+                        106,
+                        491,
+                        214
+                    ],
+                    "blocks": [
+                        {
+                            "type": "table_caption",
+                            "bbox": [
+                                107,
+                                80,
+                                503,
+                                101
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        81,
+                                        504,
+                                        91
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                81,
+                                                504,
+                                                91
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Table 4: Ablation study of task construction methods on miniImageNet. For a more complete set of results",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 0
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        90,
+                                        300,
+                                        101
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                90,
+                                                300,
+                                                101
+                                            ],
+                                            "score": 1.0,
+                                            "content": "with confidence intervals, see Table 9 in Appendix F.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 1
+                                }
+                            ],
+                            "index": 0.5
+                        },
+                        {
+                            "type": "table_body",
+                            "bbox": [
+                                120,
+                                106,
+                                491,
+                                214
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        120,
+                                        106,
+                                        491,
+                                        214
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                120,
+                                                106,
+                                                491,
+                                                214
+                                            ],
+                                            "score": 0.983,
+                                            "html": "<table><tr><td>Algorithm</td><td>(5,1)</td><td>(5,5)</td><td>(5,20)</td><td>(5,50)</td></tr><tr><td>BiGANHyperplanes-MAML,P= 4800,m =0</td><td>20.00%</td><td>20.00%</td><td>20.00%</td><td>20.00%</td></tr><tr><td>BiGAN Hyperplanes-MAML,P= 4800,m = 0.9</td><td>29.67%</td><td>41.92%</td><td>51.32%</td><td>54.72%</td></tr><tr><td>BiGAN CACTUs-MAML,P=1,k= 500</td><td>37.75%</td><td>52.59%</td><td>62.70%</td><td>67.98%</td></tr><tr><td>BiGANCACTUs-MAML,P= 50,k= 500</td><td>36.24%</td><td>51.28%</td><td>61.33%</td><td>66.91%</td></tr><tr><td>DeepCluster Hyperplanes-MAML,P= 4800,m = 0</td><td>20.02%</td><td>20.01%</td><td>20.00%</td><td>20.01%</td></tr><tr><td>DeepClusterHyperplanes-MAML,P= 4800,m= 0.1</td><td>35.85%</td><td>49.54%</td><td>60.68%</td><td>65.55%</td></tr><tr><td>DeepCluster CACTUs-MAML,P= 1,k = 500</td><td>38.75%</td><td>52.73%</td><td>62.72%</td><td>67.77%</td></tr><tr><td>DeepCluster CACTUs-MAML,P= 50,k = 500</td><td>39.90%</td><td>53.97%</td><td>63.84%</td><td>69.64%</td></tr></table>",
+                                            "type": "table",
+                                            "image_path": "0a65d9b00a7e20305e684fb38d2993d23ba0551b26d704c18abfd4d99ddb29c3.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 3,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        120,
+                                        106,
+                                        491,
+                                        142.0
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                },
+                                {
+                                    "bbox": [
+                                        120,
+                                        142.0,
+                                        491,
+                                        178.0
+                                    ],
+                                    "spans": [],
+                                    "index": 3
+                                },
+                                {
+                                    "bbox": [
+                                        120,
+                                        178.0,
+                                        491,
+                                        214.0
+                                    ],
+                                    "spans": [],
+                                    "index": 4
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 1.75
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        246,
+                        190,
+                        259
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                246,
+                                192,
+                                262
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        246,
+                                        192,
+                                        262
+                                    ],
+                                    "score": 1.0,
+                                    "content": "5 DISCUSSION",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        }
+                    ],
+                    "index": 5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        280,
+                        505,
+                        336
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                280,
+                                505,
+                                294
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        280,
+                                        505,
+                                        294
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We demonstrate that meta-learning on tasks produced using simple mechanisms based on embed-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                292,
+                                505,
+                                304
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        292,
+                                        505,
+                                        304
+                                    ],
+                                    "score": 1.0,
+                                    "content": "dings improves upon the utility of these representations in learning downstream, human-specified",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                303,
+                                505,
+                                315
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        303,
+                                        505,
+                                        315
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tasks. We empirically show that this holds across benchmark datasets and tasks in the few-shot",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                314,
+                                504,
+                                326
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        314,
+                                        504,
+                                        326
+                                    ],
+                                    "score": 1.0,
+                                    "content": "classification literature (Santoro et al., 2016; Ravi & Larochelle, 2017; Finn et al., 2018), task diffi-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                325,
+                                501,
+                                337
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        325,
+                                        501,
+                                        337
+                                    ],
+                                    "score": 1.0,
+                                    "content": "culties, and embedding learning methods while fixing key hyperparameters across all experiments.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        }
+                    ],
+                    "index": 8,
+                    "bbox_fs": [
+                        106,
+                        280,
+                        505,
+                        337
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        342,
+                        505,
+                        484
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                341,
+                                505,
+                                354
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        341,
+                                        505,
+                                        354
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In a sense, CACTUs can be seen as a facilitating interface between an embedding learning method",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                352,
+                                505,
+                                365
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        352,
+                                        505,
+                                        365
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and a meta-learning algorithm. As shown in the results, the meta-learner’s performance significantly",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                365,
+                                505,
+                                376
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        365,
+                                        505,
+                                        376
+                                    ],
+                                    "score": 1.0,
+                                    "content": "depends on the nature and quality of the task-generating embeddings. We can expect our method",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                374,
+                                505,
+                                387
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        374,
+                                        505,
+                                        387
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to yield better performance as the methods that produce these embedding functions improve, be-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                385,
+                                504,
+                                398
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        385,
+                                        504,
+                                        398
+                                    ],
+                                    "score": 1.0,
+                                    "content": "coming better suited for generating diverse yet distinctive clusterings of the data. However, the gap",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                397,
+                                505,
+                                408
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        397,
+                                        505,
+                                        408
+                                    ],
+                                    "score": 1.0,
+                                    "content": "between unsupervised and supervised meta-learning will likely persist because, with the latter, the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                408,
+                                505,
+                                420
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        408,
+                                        505,
+                                        420
+                                    ],
+                                    "score": 1.0,
+                                    "content": "meta-training task distribution is human-designed to mimic the expected evaluation task distribu-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                419,
+                                505,
+                                432
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        419,
+                                        505,
+                                        432
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tion as much as possible. Indeed, to some extent, supervised meta-learning algorithms offload the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                429,
+                                505,
+                                442
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        429,
+                                        505,
+                                        442
+                                    ],
+                                    "score": 1.0,
+                                    "content": "effort of designing and tuning algorithms onto the effort of designing and tuning task distributions.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                439,
+                                505,
+                                454
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        439,
+                                        505,
+                                        454
+                                    ],
+                                    "score": 1.0,
+                                    "content": "With its evaluation-agnostic task generation, CACTUs-based meta-learning trades off performance",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                451,
+                                504,
+                                464
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        451,
+                                        504,
+                                        464
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in specific use-cases for broad applicability and the ability to train on unlabeled data. In principle,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                461,
+                                506,
+                                476
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        461,
+                                        506,
+                                        476
+                                    ],
+                                    "score": 1.0,
+                                    "content": "CACTUs-based meta-learning may outperform supervised meta-learning when the latter is trained",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                474,
+                                412,
+                                486
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        474,
+                                        412,
+                                        486
+                                    ],
+                                    "score": 1.0,
+                                    "content": "on a misaligned task distribution. We leave this investigation to future work.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        }
+                    ],
+                    "index": 17,
+                    "bbox_fs": [
+                        105,
+                        341,
+                        506,
+                        486
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        490,
+                        505,
+                        567
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                489,
+                                505,
+                                502
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        489,
+                                        247,
+                                        502
+                                    ],
+                                    "score": 1.0,
+                                    "content": "While we have demonstrated that",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        248,
+                                        491,
+                                        254,
+                                        500
+                                    ],
+                                    "score": 0.82,
+                                    "content": "k",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        254,
+                                        489,
+                                        505,
+                                        502
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-means is a broadly useful mechanism for constructing tasks",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                501,
+                                505,
+                                513
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        501,
+                                        318,
+                                        513
+                                    ],
+                                    "score": 1.0,
+                                    "content": "from embeddings, it is unlikely that combinations of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        318,
+                                        502,
+                                        325,
+                                        511
+                                    ],
+                                    "score": 0.83,
+                                    "content": "k",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        325,
+                                        501,
+                                        505,
+                                        513
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-means clusters in learned embedding spaces",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                512,
+                                505,
+                                524
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        512,
+                                        505,
+                                        524
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are universal approximations of arbitrary class definitions. An important direction for future work",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                522,
+                                505,
+                                537
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        522,
+                                        505,
+                                        537
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is to find examples of datasets and human-designed tasks for which CACTUs-based meta-learning",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                534,
+                                505,
+                                546
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        534,
+                                        505,
+                                        546
+                                    ],
+                                    "score": 1.0,
+                                    "content": "results in ineffective downstream learning. This will result in better understanding of the practical",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                545,
+                                505,
+                                557
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        545,
+                                        505,
+                                        557
+                                    ],
+                                    "score": 1.0,
+                                    "content": "scope of applicability for our method, and spur further development in automatic task construction",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                555,
+                                287,
+                                569
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        555,
+                                        287,
+                                        569
+                                    ],
+                                    "score": 1.0,
+                                    "content": "mechanisms for unsupervised meta-learning.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        }
+                    ],
+                    "index": 27,
+                    "bbox_fs": [
+                        105,
+                        489,
+                        505,
+                        569
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        573,
+                        505,
+                        650
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                573,
+                                505,
+                                585
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        573,
+                                        505,
+                                        585
+                                    ],
+                                    "score": 1.0,
+                                    "content": "A potential concern of our experimental evaluation is that MNIST, Omniglot, and miniImageNet",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                584,
+                                505,
+                                596
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        584,
+                                        505,
+                                        596
+                                    ],
+                                    "score": 1.0,
+                                    "content": "exhibit particular structure in the underlying class distribution (i.e., perfectly balanced classes), since",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                595,
+                                505,
+                                607
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        595,
+                                        505,
+                                        607
+                                    ],
+                                    "score": 1.0,
+                                    "content": "they were designed to be supervised learning benchmarks. In more practical applications of machine",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                606,
+                                505,
+                                618
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        606,
+                                        505,
+                                        618
+                                    ],
+                                    "score": 1.0,
+                                    "content": "learning, such structure would likely not exist. Our CelebA results indicate that CACTUs is effective",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                616,
+                                505,
+                                629
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        616,
+                                        505,
+                                        629
+                                    ],
+                                    "score": 1.0,
+                                    "content": "even in the case of a dataset without neatly balanced classes or attributes. An interesting direction",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                627,
+                                505,
+                                641
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        627,
+                                        505,
+                                        641
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for future work is to better characterize the performance of CACTUs and other unsupervised pre-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                639,
+                                358,
+                                651
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        639,
+                                        358,
+                                        651
+                                    ],
+                                    "score": 1.0,
+                                    "content": "training methods with highly-unstructured, unlabeled datasets.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        }
+                    ],
+                    "index": 34,
+                    "bbox_fs": [
+                        105,
+                        573,
+                        505,
+                        651
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        655,
+                        505,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                655,
+                                505,
+                                668
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        655,
+                                        505,
+                                        668
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Since MAML and ProtoNets produce nothing more than a learned representation, our method can",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                666,
+                                505,
+                                680
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        666,
+                                        505,
+                                        680
+                                    ],
+                                    "score": 1.0,
+                                    "content": "be viewed as deriving, from a previous unsupervised representation, a new representation particu-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                677,
+                                505,
+                                690
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        677,
+                                        505,
+                                        690
+                                    ],
+                                    "score": 1.0,
+                                    "content": "larly suited for learning downstream tasks. Beyond visual classification tasks, the notion of using",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                688,
+                                505,
+                                702
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        688,
+                                        505,
+                                        702
+                                    ],
+                                    "score": 1.0,
+                                    "content": "unsupervised pre-training is generally applicable to a wide range of domains, including regression,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                699,
+                                506,
+                                712
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        699,
+                                        506,
+                                        712
+                                    ],
+                                    "score": 1.0,
+                                    "content": "speech (Oord et al., 2018), language (Howard & Ruder, 2018), and reinforcement learning (Shel-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                710,
+                                506,
+                                724
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        710,
+                                        506,
+                                        724
+                                    ],
+                                    "score": 1.0,
+                                    "content": "hamer et al., 2017). Hence, our unsupervised meta-learning approach has the potential to improve",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                721,
+                                488,
+                                734
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        721,
+                                        488,
+                                        734
+                                    ],
+                                    "score": 1.0,
+                                    "content": "unsupervised representations for a variety of such domains, an exciting avenue for future work.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        }
+                    ],
+                    "index": 41,
+                    "bbox_fs": [
+                        105,
+                        655,
+                        506,
+                        734
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        82,
+                        218,
+                        94
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                107,
+                                81,
+                                219,
+                                96
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        81,
+                                        219,
+                                        96
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ACKNOWLEDGMENTS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        }
+                    ],
+                    "index": 0
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        102,
+                        504,
+                        126
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                101,
+                                505,
+                                116
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        101,
+                                        505,
+                                        116
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We thank Kelvin Xu, Richard Zhang, Brian Cheung, Ben Poole, Aaron van den Oord, Luke Metz, ¨",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                113,
+                                474,
+                                128
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        113,
+                                        474,
+                                        128
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Siddharth Reddy, and the anonymous reviewers for feedback on an early draft of this paper.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        }
+                    ],
+                    "index": 1.5
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        142,
+                        176,
+                        154
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                143,
+                                176,
+                                155
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        143,
+                                        176,
+                                        155
+                                    ],
+                                    "score": 1.0,
+                                    "content": "REFERENCES",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        }
+                    ],
+                    "index": 3
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        160,
+                        504,
+                        183
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                158,
+                                505,
+                                174
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        158,
+                                        505,
+                                        174
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Yoshua Bengio, Samy Bengio, and Jocelyn Cloutier. Learning a synaptic learning rule. In Interna-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                171,
+                                360,
+                                183
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        171,
+                                        360,
+                                        183
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tional Joint Conference on Neural Networks (IJCNN), 1991.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        }
+                    ],
+                    "index": 4.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        189,
+                        504,
+                        213
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                188,
+                                506,
+                                204
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        188,
+                                        506,
+                                        204
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Yoshua Bengio, Pascal Lamblin, Dan Popovici, and Hugo Larochelle. Greedy layer-wise training",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                200,
+                                471,
+                                214
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        200,
+                                        471,
+                                        214
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of deep networks. In Advances in Neural Information Processing Systems (NIPS), 2007.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        }
+                    ],
+                    "index": 6.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        218,
+                        504,
+                        243
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                218,
+                                505,
+                                232
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        218,
+                                        505,
+                                        232
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Yoshua Bengio, Aaron Courville, and Pascal Vincent. Representation learning: A review and new",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                230,
+                                499,
+                                243
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        230,
+                                        499,
+                                        243
+                                    ],
+                                    "score": 1.0,
+                                    "content": "perspectives. IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI), 2013.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        }
+                    ],
+                    "index": 8.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        248,
+                        505,
+                        282
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                247,
+                                505,
+                                262
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        247,
+                                        505,
+                                        262
+                                    ],
+                                    "score": 1.0,
+                                    "content": "David Berthelot, Colin Raffel, Aurko Roy, and Ian Goodfellow. Understanding and improving",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                260,
+                                505,
+                                272
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        260,
+                                        505,
+                                        272
+                                    ],
+                                    "score": 1.0,
+                                    "content": "interpolation in autoencoders via an adversarial regularizer. arXiv preprint arXiv:1807.07543,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                270,
+                                143,
+                                283
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        270,
+                                        143,
+                                        283
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        }
+                    ],
+                    "index": 11
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        289,
+                        505,
+                        312
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                289,
+                                506,
+                                303
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        289,
+                                        506,
+                                        303
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Piotr Bojanowski and Armand Joulin. Unsupervised learning by predicting noise. In International",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                300,
+                                314,
+                                312
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        300,
+                                        314,
+                                        312
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Conference on Machine Learning (ICML), 2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        }
+                    ],
+                    "index": 13.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        318,
+                        504,
+                        352
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                317,
+                                504,
+                                332
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        317,
+                                        504,
+                                        332
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Mathilde Caron, Piotr Bojanowski, Armand Joulin, and Matthijs Douze. Deep clustering for un-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                330,
+                                505,
+                                343
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        330,
+                                        505,
+                                        343
+                                    ],
+                                    "score": 1.0,
+                                    "content": "supervised learning of visual features. In European Conference on Computer Vision (ECCV),",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                114,
+                                339,
+                                142,
+                                353
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        114,
+                                        339,
+                                        142,
+                                        353
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        }
+                    ],
+                    "index": 16
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        358,
+                        504,
+                        393
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                359,
+                                505,
+                                371
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        359,
+                                        505,
+                                        371
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Xi Chen, Yan Duan, Rein Houthooft, John Schulman, Ilya Sutskever, and Pieter Abbeel. InfoGAN:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                371,
+                                505,
+                                383
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        371,
+                                        505,
+                                        383
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Interpretable representation learning by information maximizing generative adversarial nets. In",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                381,
+                                334,
+                                394
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        381,
+                                        334,
+                                        394
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Neural Information Processing Systems (NIPS), 2016.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        }
+                    ],
+                    "index": 19
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        399,
+                        504,
+                        434
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                399,
+                                505,
+                                413
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        399,
+                                        505,
+                                        413
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Brian Cheung, Jesse A Livezey, Arjun K Bansal, and Bruno A Olshausen. Discovering hidden",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                411,
+                                505,
+                                423
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        411,
+                                        505,
+                                        423
+                                    ],
+                                    "score": 1.0,
+                                    "content": "factors of variation in deep networks. In International Conference on Learning Representations",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                422,
+                                174,
+                                434
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        422,
+                                        174,
+                                        434
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(ICLR), 2015.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        }
+                    ],
+                    "index": 22
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        104,
+                        439,
+                        504,
+                        464
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                440,
+                                505,
+                                453
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        440,
+                                        505,
+                                        453
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Adam Coates and Andrew Y Ng. Learning feature representations with k-means. In Neural Net-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                451,
+                                290,
+                                463
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        451,
+                                        290,
+                                        463
+                                    ],
+                                    "score": 1.0,
+                                    "content": "works: Tricks of the Trade. Springer, 2012.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        }
+                    ],
+                    "index": 24.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        469,
+                        504,
+                        493
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                469,
+                                505,
+                                483
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        469,
+                                        505,
+                                        483
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Andrew M Dai and Quoc V Le. Semi-supervised sequence learning. In Neural Information Pro-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                481,
+                                240,
+                                493
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        481,
+                                        240,
+                                        493
+                                    ],
+                                    "score": 1.0,
+                                    "content": "cessing Systems (NIPS), 2015.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        }
+                    ],
+                    "index": 26.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        498,
+                        504,
+                        523
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                497,
+                                505,
+                                513
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        497,
+                                        505,
+                                        513
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Emily L Denton and Vighnesh Birodkar. Unsupervised learning of disentangled representations",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                510,
+                                396,
+                                523
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        510,
+                                        396,
+                                        523
+                                    ],
+                                    "score": 1.0,
+                                    "content": "from video. In Neural Information Processing Systems (NIPS), 2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        }
+                    ],
+                    "index": 28.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        528,
+                        504,
+                        552
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                527,
+                                506,
+                                542
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        527,
+                                        506,
+                                        542
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Jeff Donahue, Philipp Krahenb ¨ uhl, and Trevor Darrell. Adversarial feature learning. In ¨ International",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                540,
+                                341,
+                                552
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        540,
+                                        341,
+                                        552
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Conference on Learning Representations (ICLR), 2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        }
+                    ],
+                    "index": 30.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        558,
+                        504,
+                        592
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                107,
+                                559,
+                                504,
+                                571
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        559,
+                                        504,
+                                        571
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Vincent Dumoulin, Ishmael Belghazi, Ben Poole, Olivier Mastropietro, Alex Lamb, Martin Ar-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                569,
+                                506,
+                                583
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        569,
+                                        506,
+                                        583
+                                    ],
+                                    "score": 1.0,
+                                    "content": "jovsky, and Aaron Courville. Adversarially learned inference. In International Conference on",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                581,
+                                280,
+                                593
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        581,
+                                        280,
+                                        593
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Learning Representations (ICLR), 2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        }
+                    ],
+                    "index": 33
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        598,
+                        505,
+                        632
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                598,
+                                505,
+                                612
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        598,
+                                        505,
+                                        612
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Dumitru Erhan, Yoshua Bengio, Aaron Courville, Pierre-Antoine Manzagol, Pascal Vincent, and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                610,
+                                505,
+                                623
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        610,
+                                        505,
+                                        623
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Samy Bengio. Why does unsupervised pre-training help deep learning? Journal of Machine",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                621,
+                                255,
+                                633
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        621,
+                                        255,
+                                        633
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Learning Research (JMLR), 2010.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        }
+                    ],
+                    "index": 36
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        639,
+                        505,
+                        673
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                639,
+                                504,
+                                652
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        639,
+                                        504,
+                                        652
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Chelsea Finn and Sergey Levine. Meta-learning and universality: Deep representations and gradi-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                650,
+                                505,
+                                664
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        650,
+                                        505,
+                                        664
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ent descent can approximate any learning algorithm. In International Conference on Learning",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                662,
+                                241,
+                                674
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        662,
+                                        241,
+                                        674
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Representations (ICLR), 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        }
+                    ],
+                    "index": 39
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        680,
+                        504,
+                        703
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                679,
+                                505,
+                                693
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        679,
+                                        505,
+                                        693
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Chelsea Finn, Pieter Abbeel, and Sergey Levine. Model-agnostic meta-learning for fast adaptation",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                691,
+                                454,
+                                704
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        691,
+                                        454,
+                                        704
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of deep networks. In International Conference on Machine Learning (ICML), 2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        }
+                    ],
+                    "index": 41.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        709,
+                        504,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                708,
+                                505,
+                                723
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        708,
+                                        505,
+                                        723
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Chelsea Finn, Kelvin Xu, and Sergey Levine. Probabilistic model-agnostic meta-learning. In Neural",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                720,
+                                304,
+                                732
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        720,
+                                        304,
+                                        732
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Information Processing Systems (NIPS), 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        }
+                    ],
+                    "index": 43.5
+                }
+            ],
+            "page_idx": 10,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        107,
+                        27,
+                        293,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                26,
+                                294,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        26,
+                                        294,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Published as a conference paper at ICLR 2019",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        300,
+                        751,
+                        310,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                299,
+                                750,
+                                312,
+                                765
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        299,
+                                        750,
+                                        312,
+                                        765
+                                    ],
+                                    "score": 1.0,
+                                    "content": "",
+                                    "type": "text",
+                                    "height": 15,
+                                    "width": 13
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        82,
+                        218,
+                        94
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                107,
+                                81,
+                                219,
+                                96
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        81,
+                                        219,
+                                        96
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ACKNOWLEDGMENTS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        }
+                    ],
+                    "index": 0
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        102,
+                        504,
+                        126
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                101,
+                                505,
+                                116
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        101,
+                                        505,
+                                        116
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We thank Kelvin Xu, Richard Zhang, Brian Cheung, Ben Poole, Aaron van den Oord, Luke Metz, ¨",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                113,
+                                474,
+                                128
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        113,
+                                        474,
+                                        128
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Siddharth Reddy, and the anonymous reviewers for feedback on an early draft of this paper.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        }
+                    ],
+                    "index": 1.5,
+                    "bbox_fs": [
+                        105,
+                        101,
+                        505,
+                        128
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        142,
+                        176,
+                        154
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                143,
+                                176,
+                                155
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        143,
+                                        176,
+                                        155
+                                    ],
+                                    "score": 1.0,
+                                    "content": "REFERENCES",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        }
+                    ],
+                    "index": 3
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        160,
+                        504,
+                        183
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                158,
+                                505,
+                                174
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        158,
+                                        505,
+                                        174
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Yoshua Bengio, Samy Bengio, and Jocelyn Cloutier. Learning a synaptic learning rule. In Interna-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                171,
+                                360,
+                                183
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        171,
+                                        360,
+                                        183
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tional Joint Conference on Neural Networks (IJCNN), 1991.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        }
+                    ],
+                    "index": 4.5,
+                    "bbox_fs": [
+                        106,
+                        158,
+                        505,
+                        183
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        189,
+                        504,
+                        213
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                188,
+                                506,
+                                204
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        188,
+                                        506,
+                                        204
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Yoshua Bengio, Pascal Lamblin, Dan Popovici, and Hugo Larochelle. Greedy layer-wise training",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                200,
+                                471,
+                                214
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        200,
+                                        471,
+                                        214
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of deep networks. In Advances in Neural Information Processing Systems (NIPS), 2007.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        }
+                    ],
+                    "index": 6.5,
+                    "bbox_fs": [
+                        105,
+                        188,
+                        506,
+                        214
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        218,
+                        504,
+                        243
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                218,
+                                505,
+                                232
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        218,
+                                        505,
+                                        232
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Yoshua Bengio, Aaron Courville, and Pascal Vincent. Representation learning: A review and new",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                230,
+                                499,
+                                243
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        230,
+                                        499,
+                                        243
+                                    ],
+                                    "score": 1.0,
+                                    "content": "perspectives. IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI), 2013.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        }
+                    ],
+                    "index": 8.5,
+                    "bbox_fs": [
+                        106,
+                        218,
+                        505,
+                        243
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        248,
+                        505,
+                        282
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                247,
+                                505,
+                                262
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        247,
+                                        505,
+                                        262
+                                    ],
+                                    "score": 1.0,
+                                    "content": "David Berthelot, Colin Raffel, Aurko Roy, and Ian Goodfellow. Understanding and improving",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                260,
+                                505,
+                                272
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        260,
+                                        505,
+                                        272
+                                    ],
+                                    "score": 1.0,
+                                    "content": "interpolation in autoencoders via an adversarial regularizer. arXiv preprint arXiv:1807.07543,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                270,
+                                143,
+                                283
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        270,
+                                        143,
+                                        283
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        }
+                    ],
+                    "index": 11,
+                    "bbox_fs": [
+                        105,
+                        247,
+                        505,
+                        283
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        289,
+                        505,
+                        312
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                289,
+                                506,
+                                303
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        289,
+                                        506,
+                                        303
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Piotr Bojanowski and Armand Joulin. Unsupervised learning by predicting noise. In International",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                300,
+                                314,
+                                312
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        300,
+                                        314,
+                                        312
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Conference on Machine Learning (ICML), 2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        }
+                    ],
+                    "index": 13.5,
+                    "bbox_fs": [
+                        105,
+                        289,
+                        506,
+                        312
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        318,
+                        504,
+                        352
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                317,
+                                504,
+                                332
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        317,
+                                        504,
+                                        332
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Mathilde Caron, Piotr Bojanowski, Armand Joulin, and Matthijs Douze. Deep clustering for un-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                330,
+                                505,
+                                343
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        330,
+                                        505,
+                                        343
+                                    ],
+                                    "score": 1.0,
+                                    "content": "supervised learning of visual features. In European Conference on Computer Vision (ECCV),",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                114,
+                                339,
+                                142,
+                                353
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        114,
+                                        339,
+                                        142,
+                                        353
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        }
+                    ],
+                    "index": 16,
+                    "bbox_fs": [
+                        105,
+                        317,
+                        505,
+                        353
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        358,
+                        504,
+                        393
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                359,
+                                505,
+                                371
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        359,
+                                        505,
+                                        371
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Xi Chen, Yan Duan, Rein Houthooft, John Schulman, Ilya Sutskever, and Pieter Abbeel. InfoGAN:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                371,
+                                505,
+                                383
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        371,
+                                        505,
+                                        383
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Interpretable representation learning by information maximizing generative adversarial nets. In",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                381,
+                                334,
+                                394
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        381,
+                                        334,
+                                        394
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Neural Information Processing Systems (NIPS), 2016.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        }
+                    ],
+                    "index": 19,
+                    "bbox_fs": [
+                        106,
+                        359,
+                        505,
+                        394
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        399,
+                        504,
+                        434
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                399,
+                                505,
+                                413
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        399,
+                                        505,
+                                        413
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Brian Cheung, Jesse A Livezey, Arjun K Bansal, and Bruno A Olshausen. Discovering hidden",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                411,
+                                505,
+                                423
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        411,
+                                        505,
+                                        423
+                                    ],
+                                    "score": 1.0,
+                                    "content": "factors of variation in deep networks. In International Conference on Learning Representations",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                422,
+                                174,
+                                434
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        422,
+                                        174,
+                                        434
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(ICLR), 2015.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        }
+                    ],
+                    "index": 22,
+                    "bbox_fs": [
+                        105,
+                        399,
+                        505,
+                        434
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        104,
+                        439,
+                        504,
+                        464
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                440,
+                                505,
+                                453
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        440,
+                                        505,
+                                        453
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Adam Coates and Andrew Y Ng. Learning feature representations with k-means. In Neural Net-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                451,
+                                290,
+                                463
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        451,
+                                        290,
+                                        463
+                                    ],
+                                    "score": 1.0,
+                                    "content": "works: Tricks of the Trade. Springer, 2012.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        }
+                    ],
+                    "index": 24.5,
+                    "bbox_fs": [
+                        106,
+                        440,
+                        505,
+                        463
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        469,
+                        504,
+                        493
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                469,
+                                505,
+                                483
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        469,
+                                        505,
+                                        483
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Andrew M Dai and Quoc V Le. Semi-supervised sequence learning. In Neural Information Pro-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                481,
+                                240,
+                                493
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        481,
+                                        240,
+                                        493
+                                    ],
+                                    "score": 1.0,
+                                    "content": "cessing Systems (NIPS), 2015.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        }
+                    ],
+                    "index": 26.5,
+                    "bbox_fs": [
+                        106,
+                        469,
+                        505,
+                        493
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        498,
+                        504,
+                        523
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                497,
+                                505,
+                                513
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        497,
+                                        505,
+                                        513
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Emily L Denton and Vighnesh Birodkar. Unsupervised learning of disentangled representations",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                510,
+                                396,
+                                523
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        510,
+                                        396,
+                                        523
+                                    ],
+                                    "score": 1.0,
+                                    "content": "from video. In Neural Information Processing Systems (NIPS), 2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        }
+                    ],
+                    "index": 28.5,
+                    "bbox_fs": [
+                        105,
+                        497,
+                        505,
+                        523
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        528,
+                        504,
+                        552
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                527,
+                                506,
+                                542
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        527,
+                                        506,
+                                        542
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Jeff Donahue, Philipp Krahenb ¨ uhl, and Trevor Darrell. Adversarial feature learning. In ¨ International",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                540,
+                                341,
+                                552
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        540,
+                                        341,
+                                        552
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Conference on Learning Representations (ICLR), 2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        }
+                    ],
+                    "index": 30.5,
+                    "bbox_fs": [
+                        105,
+                        527,
+                        506,
+                        552
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        558,
+                        504,
+                        592
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                107,
+                                559,
+                                504,
+                                571
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        559,
+                                        504,
+                                        571
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Vincent Dumoulin, Ishmael Belghazi, Ben Poole, Olivier Mastropietro, Alex Lamb, Martin Ar-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                569,
+                                506,
+                                583
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        569,
+                                        506,
+                                        583
+                                    ],
+                                    "score": 1.0,
+                                    "content": "jovsky, and Aaron Courville. Adversarially learned inference. In International Conference on",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                581,
+                                280,
+                                593
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        581,
+                                        280,
+                                        593
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Learning Representations (ICLR), 2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        }
+                    ],
+                    "index": 33,
+                    "bbox_fs": [
+                        107,
+                        559,
+                        506,
+                        593
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        598,
+                        505,
+                        632
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                598,
+                                505,
+                                612
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        598,
+                                        505,
+                                        612
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Dumitru Erhan, Yoshua Bengio, Aaron Courville, Pierre-Antoine Manzagol, Pascal Vincent, and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                610,
+                                505,
+                                623
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        610,
+                                        505,
+                                        623
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Samy Bengio. Why does unsupervised pre-training help deep learning? Journal of Machine",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                621,
+                                255,
+                                633
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        621,
+                                        255,
+                                        633
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Learning Research (JMLR), 2010.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        }
+                    ],
+                    "index": 36,
+                    "bbox_fs": [
+                        105,
+                        598,
+                        505,
+                        633
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        639,
+                        505,
+                        673
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                639,
+                                504,
+                                652
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        639,
+                                        504,
+                                        652
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Chelsea Finn and Sergey Levine. Meta-learning and universality: Deep representations and gradi-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                650,
+                                505,
+                                664
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        650,
+                                        505,
+                                        664
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ent descent can approximate any learning algorithm. In International Conference on Learning",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                662,
+                                241,
+                                674
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        662,
+                                        241,
+                                        674
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Representations (ICLR), 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        }
+                    ],
+                    "index": 39,
+                    "bbox_fs": [
+                        106,
+                        639,
+                        505,
+                        674
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        680,
+                        504,
+                        703
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                679,
+                                505,
+                                693
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        679,
+                                        505,
+                                        693
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Chelsea Finn, Pieter Abbeel, and Sergey Levine. Model-agnostic meta-learning for fast adaptation",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                691,
+                                454,
+                                704
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        691,
+                                        454,
+                                        704
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of deep networks. In International Conference on Machine Learning (ICML), 2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        }
+                    ],
+                    "index": 41.5,
+                    "bbox_fs": [
+                        106,
+                        679,
+                        505,
+                        704
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        709,
+                        504,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                708,
+                                505,
+                                723
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        708,
+                                        505,
+                                        723
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Chelsea Finn, Kelvin Xu, and Sergey Levine. Probabilistic model-agnostic meta-learning. In Neural",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                720,
+                                304,
+                                732
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        720,
+                                        304,
+                                        732
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Information Processing Systems (NIPS), 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        }
+                    ],
+                    "index": 43.5,
+                    "bbox_fs": [
+                        106,
+                        708,
+                        505,
+                        732
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        82,
+                        504,
+                        105
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                81,
+                                506,
+                                97
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        81,
+                                        259,
+                                        95
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Vikas K Garg and Adam Kalai.",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        270,
+                                        81,
+                                        506,
+                                        97
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Supervising unsupervised learning. arXiv preprint",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                93,
+                                220,
+                                105
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        93,
+                                        220,
+                                        105
+                                    ],
+                                    "score": 1.0,
+                                    "content": "arXiv:1709.05262, 2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        }
+                    ],
+                    "index": 0.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        111,
+                        503,
+                        135
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                111,
+                                505,
+                                125
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        111,
+                                        505,
+                                        125
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Abhishek Gupta, Benjamin Eysenbach, Chelsea Finn, and Sergey Levine. Unsupervised meta-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                123,
+                                425,
+                                136
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        123,
+                                        425,
+                                        136
+                                    ],
+                                    "score": 1.0,
+                                    "content": "learning for reinforcement learning. arXiv preprint arXiv:1806.04640, 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        }
+                    ],
+                    "index": 2.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        142,
+                        504,
+                        165
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                141,
+                                506,
+                                156
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        141,
+                                        506,
+                                        156
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Trevor Hastie, Robert Tibshirani, and Jerome Friedman. Unsupervised learning. In The Elements of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                153,
+                                264,
+                                165
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        153,
+                                        264,
+                                        165
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Statistical Learning. Springer, 2009.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        }
+                    ],
+                    "index": 4.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        172,
+                        505,
+                        195
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                172,
+                                505,
+                                185
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        172,
+                                        505,
+                                        185
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Identity mappings in deep residual",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                183,
+                                393,
+                                195
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        183,
+                                        393,
+                                        195
+                                    ],
+                                    "score": 1.0,
+                                    "content": "networks. European Conference on Computer Vision (ECCV), 2016.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        }
+                    ],
+                    "index": 6.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        201,
+                        505,
+                        246
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                203,
+                                505,
+                                214
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        203,
+                                        505,
+                                        214
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Irina Higgins, Loic Matthey, Arka Pal, Christopher Burgess, Xavier Glorot, Matthew Botvinick,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                213,
+                                505,
+                                226
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        213,
+                                        295,
+                                        226
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Shakir Mohamed, and Alexander Lerchner.",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        296,
+                                        214,
+                                        303,
+                                        225
+                                    ],
+                                    "score": 0.84,
+                                    "content": "\\beta",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        303,
+                                        213,
+                                        505,
+                                        226
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-VAE: Learning basic visual concepts with a con-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                223,
+                                505,
+                                237
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        223,
+                                        505,
+                                        237
+                                    ],
+                                    "score": 1.0,
+                                    "content": "strained variational framework. In International Conference on Learning Representations (ICLR),",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                235,
+                                142,
+                                247
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        235,
+                                        142,
+                                        247
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        }
+                    ],
+                    "index": 9.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        253,
+                        504,
+                        277
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                254,
+                                505,
+                                267
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        254,
+                                        505,
+                                        267
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Geoffrey E Hinton, Simon Osindero, and Yee-Whye Teh. A fast learning algorithm for deep belief",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                265,
+                                250,
+                                277
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        265,
+                                        250,
+                                        277
+                                    ],
+                                    "score": 1.0,
+                                    "content": "nets. Neural Computation, 2006.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        }
+                    ],
+                    "index": 12.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        283,
+                        503,
+                        307
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                282,
+                                505,
+                                298
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        282,
+                                        505,
+                                        298
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Jeremy Howard and Sebastian Ruder. Universal language model fine-tuning for text classification.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                295,
+                                353,
+                                307
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        295,
+                                        353,
+                                        307
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In Association for Computational Linguistics (ACL), 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        }
+                    ],
+                    "index": 14.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        313,
+                        504,
+                        337
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                313,
+                                505,
+                                328
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        313,
+                                        505,
+                                        328
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Sergey Ioffe and Christian Szegedy. Batch normalization: Accelerating deep network training by",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                325,
+                                505,
+                                338
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        325,
+                                        505,
+                                        338
+                                    ],
+                                    "score": 1.0,
+                                    "content": "reducing internal covariate shift. In International Conference on Machine Learning (ICML), 2015.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        }
+                    ],
+                    "index": 16.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        343,
+                        505,
+                        378
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                343,
+                                505,
+                                357
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        343,
+                                        505,
+                                        357
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Jaehong Kim, Sangyeul Lee, Sungwan Kim, Moonsu Cha, Jung Kwon Lee, Youngduck Choi,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                355,
+                                505,
+                                368
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        355,
+                                        505,
+                                        368
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Yongseok Choi, Dong-Yeon Cho, and Jiwon Kim. Auto-Meta: Automated gradient based meta",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                366,
+                                340,
+                                378
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        366,
+                                        340,
+                                        378
+                                    ],
+                                    "score": 1.0,
+                                    "content": "learner search. arXiv preprint arXiv:1806.06927, 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        }
+                    ],
+                    "index": 19
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        384,
+                        504,
+                        408
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                384,
+                                506,
+                                399
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        384,
+                                        506,
+                                        399
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Diederik Kingma and Jimmy Ba. Adam: A method for stochastic optimization. In Proceedings of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                395,
+                                426,
+                                408
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        395,
+                                        426,
+                                        408
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the 3rd International Conference on Learning Representations (ICLR), 2014.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        }
+                    ],
+                    "index": 21.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        414,
+                        504,
+                        438
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                415,
+                                505,
+                                428
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        415,
+                                        505,
+                                        428
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Diederik P Kingma, Shakir Mohamed, Danilo Jimenez Rezende, and Max Welling. Semi-supervised",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                426,
+                                501,
+                                439
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        426,
+                                        501,
+                                        439
+                                    ],
+                                    "score": 1.0,
+                                    "content": "learning with deep generative models. In Neural Information Processing Systems (NIPS), 2014.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        }
+                    ],
+                    "index": 23.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        444,
+                        505,
+                        478
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                444,
+                                505,
+                                458
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        444,
+                                        505,
+                                        458
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Philipp Krahenb ¨ uhl, Carl Doersch, Jeff Donahue, and Trevor Darrell. Data-dependent initializa- ¨",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                455,
+                                505,
+                                468
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        455,
+                                        505,
+                                        468
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tions of convolutional neural networks. In International Conference on Learning Representations",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                467,
+                                175,
+                                479
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        467,
+                                        175,
+                                        479
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(ICLR), 2016.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        }
+                    ],
+                    "index": 26
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        485,
+                        505,
+                        530
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                486,
+                                505,
+                                499
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        486,
+                                        505,
+                                        499
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Quoc V Le, Marc’Aurelio Ranzato, Rajat Monga, Matthieu Devin, Gregory S. Corrado, Kai Chen,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                497,
+                                506,
+                                510
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        497,
+                                        506,
+                                        510
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Jeffrey Dean, and Andrew Y Ng. Building high-level features using large scale unsupervised",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                114,
+                                507,
+                                506,
+                                522
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        114,
+                                        507,
+                                        506,
+                                        522
+                                    ],
+                                    "score": 1.0,
+                                    "content": "learning. In International Conference on Acoustics, Speech and Signal Processing (ICASSP),",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                518,
+                                143,
+                                531
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        518,
+                                        143,
+                                        531
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2013.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        }
+                    ],
+                    "index": 29.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        109,
+                        537,
+                        503,
+                        572
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                107,
+                                537,
+                                505,
+                                550
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        537,
+                                        505,
+                                        550
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Michael F Mathieu, Junbo Jake Zhao, Junbo Zhao, Aditya Ramesh, Pablo Sprechmann, and Yann",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                549,
+                                505,
+                                561
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        549,
+                                        505,
+                                        561
+                                    ],
+                                    "score": 1.0,
+                                    "content": "LeCun. Disentangling factors of variation in deep representation using adversarial training. In",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                559,
+                                334,
+                                572
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        559,
+                                        334,
+                                        572
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Neural Information Processing Systems (NIPS), 2016.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        }
+                    ],
+                    "index": 33
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        578,
+                        504,
+                        601
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                578,
+                                505,
+                                592
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        578,
+                                        505,
+                                        592
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Luke Metz, Niru Maheswaranathan, Brian Cheung, and Jascha Sohl-Dickstein. Learning to learn",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                589,
+                                469,
+                                602
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        589,
+                                        469,
+                                        602
+                                    ],
+                                    "score": 1.0,
+                                    "content": "without labels. In International Conference on Learning Representations (ICLR), 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        }
+                    ],
+                    "index": 35.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        608,
+                        504,
+                        631
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                608,
+                                505,
+                                621
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        608,
+                                        505,
+                                        621
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Tsendsuren Munkhdalai and Hong Yu. Meta networks. In International Conference on Machine",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                619,
+                                216,
+                                632
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        619,
+                                        216,
+                                        632
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Learning (ICML), 2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        }
+                    ],
+                    "index": 37.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        104,
+                        638,
+                        504,
+                        661
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                638,
+                                505,
+                                651
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        638,
+                                        505,
+                                        651
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Devang K Naik and RJ Mammone. Meta-neural networks that learn by learning. In International",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                649,
+                                333,
+                                662
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        649,
+                                        333,
+                                        662
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Joint Conference on Neural Networks (IJCNN), 1992.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        }
+                    ],
+                    "index": 39.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        668,
+                        503,
+                        702
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                668,
+                                505,
+                                681
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        668,
+                                        505,
+                                        681
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Avital Oliver, Augustus Odena, Colin Raffel, Ekin D Cubuk, and Ian J Goodfellow. Realistic eval-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                679,
+                                505,
+                                693
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        679,
+                                        505,
+                                        693
+                                    ],
+                                    "score": 1.0,
+                                    "content": "uation of deep semi-supervised learning algorithms. In International Conference on Learning",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                690,
+                                241,
+                                703
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        690,
+                                        241,
+                                        703
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Representations (ICLR), 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        }
+                    ],
+                    "index": 42
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        709,
+                        502,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                709,
+                                504,
+                                722
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        709,
+                                        504,
+                                        722
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Aaron van den Oord, Yazhe Li, and Oriol Vinyals. Representation learning with contrastive predic-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                721,
+                                329,
+                                732
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        721,
+                                        329,
+                                        732
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tive coding. arXiv preprint arXiv:1807.03748, 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        }
+                    ],
+                    "index": 44.5
+                }
+            ],
+            "page_idx": 11,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        108,
+                        27,
+                        293,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                26,
+                                294,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        26,
+                                        294,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Published as a conference paper at ICLR 2019",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        300,
+                        751,
+                        311,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                299,
+                                750,
+                                313,
+                                764
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        299,
+                                        750,
+                                        313,
+                                        764
+                                    ],
+                                    "score": 1.0,
+                                    "content": "12",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        82,
+                        504,
+                        105
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                81,
+                                506,
+                                97
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        81,
+                                        259,
+                                        95
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Vikas K Garg and Adam Kalai.",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        270,
+                                        81,
+                                        506,
+                                        97
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Supervising unsupervised learning. arXiv preprint",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                93,
+                                220,
+                                105
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        93,
+                                        220,
+                                        105
+                                    ],
+                                    "score": 1.0,
+                                    "content": "arXiv:1709.05262, 2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        }
+                    ],
+                    "index": 0.5,
+                    "bbox_fs": [
+                        106,
+                        81,
+                        506,
+                        105
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        111,
+                        503,
+                        135
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                111,
+                                505,
+                                125
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        111,
+                                        505,
+                                        125
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Abhishek Gupta, Benjamin Eysenbach, Chelsea Finn, and Sergey Levine. Unsupervised meta-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                123,
+                                425,
+                                136
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        123,
+                                        425,
+                                        136
+                                    ],
+                                    "score": 1.0,
+                                    "content": "learning for reinforcement learning. arXiv preprint arXiv:1806.04640, 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        }
+                    ],
+                    "index": 2.5,
+                    "bbox_fs": [
+                        105,
+                        111,
+                        505,
+                        136
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        142,
+                        504,
+                        165
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                141,
+                                506,
+                                156
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        141,
+                                        506,
+                                        156
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Trevor Hastie, Robert Tibshirani, and Jerome Friedman. Unsupervised learning. In The Elements of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                153,
+                                264,
+                                165
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        153,
+                                        264,
+                                        165
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Statistical Learning. Springer, 2009.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        }
+                    ],
+                    "index": 4.5,
+                    "bbox_fs": [
+                        105,
+                        141,
+                        506,
+                        165
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        172,
+                        505,
+                        195
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                172,
+                                505,
+                                185
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        172,
+                                        505,
+                                        185
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Identity mappings in deep residual",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                183,
+                                393,
+                                195
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        183,
+                                        393,
+                                        195
+                                    ],
+                                    "score": 1.0,
+                                    "content": "networks. European Conference on Computer Vision (ECCV), 2016.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        }
+                    ],
+                    "index": 6.5,
+                    "bbox_fs": [
+                        106,
+                        172,
+                        505,
+                        195
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        201,
+                        505,
+                        246
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                203,
+                                505,
+                                214
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        203,
+                                        505,
+                                        214
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Irina Higgins, Loic Matthey, Arka Pal, Christopher Burgess, Xavier Glorot, Matthew Botvinick,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                213,
+                                505,
+                                226
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        213,
+                                        295,
+                                        226
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Shakir Mohamed, and Alexander Lerchner.",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        296,
+                                        214,
+                                        303,
+                                        225
+                                    ],
+                                    "score": 0.84,
+                                    "content": "\\beta",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        303,
+                                        213,
+                                        505,
+                                        226
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-VAE: Learning basic visual concepts with a con-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                223,
+                                505,
+                                237
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        223,
+                                        505,
+                                        237
+                                    ],
+                                    "score": 1.0,
+                                    "content": "strained variational framework. In International Conference on Learning Representations (ICLR),",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                235,
+                                142,
+                                247
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        235,
+                                        142,
+                                        247
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        }
+                    ],
+                    "index": 9.5,
+                    "bbox_fs": [
+                        106,
+                        203,
+                        505,
+                        247
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        253,
+                        504,
+                        277
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                254,
+                                505,
+                                267
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        254,
+                                        505,
+                                        267
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Geoffrey E Hinton, Simon Osindero, and Yee-Whye Teh. A fast learning algorithm for deep belief",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                265,
+                                250,
+                                277
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        265,
+                                        250,
+                                        277
+                                    ],
+                                    "score": 1.0,
+                                    "content": "nets. Neural Computation, 2006.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        }
+                    ],
+                    "index": 12.5,
+                    "bbox_fs": [
+                        106,
+                        254,
+                        505,
+                        277
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        283,
+                        503,
+                        307
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                282,
+                                505,
+                                298
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        282,
+                                        505,
+                                        298
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Jeremy Howard and Sebastian Ruder. Universal language model fine-tuning for text classification.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                295,
+                                353,
+                                307
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        295,
+                                        353,
+                                        307
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In Association for Computational Linguistics (ACL), 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        }
+                    ],
+                    "index": 14.5,
+                    "bbox_fs": [
+                        105,
+                        282,
+                        505,
+                        307
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        313,
+                        504,
+                        337
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                313,
+                                505,
+                                328
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        313,
+                                        505,
+                                        328
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Sergey Ioffe and Christian Szegedy. Batch normalization: Accelerating deep network training by",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                325,
+                                505,
+                                338
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        325,
+                                        505,
+                                        338
+                                    ],
+                                    "score": 1.0,
+                                    "content": "reducing internal covariate shift. In International Conference on Machine Learning (ICML), 2015.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        }
+                    ],
+                    "index": 16.5,
+                    "bbox_fs": [
+                        105,
+                        313,
+                        505,
+                        338
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        343,
+                        505,
+                        378
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                343,
+                                505,
+                                357
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        343,
+                                        505,
+                                        357
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Jaehong Kim, Sangyeul Lee, Sungwan Kim, Moonsu Cha, Jung Kwon Lee, Youngduck Choi,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                355,
+                                505,
+                                368
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        355,
+                                        505,
+                                        368
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Yongseok Choi, Dong-Yeon Cho, and Jiwon Kim. Auto-Meta: Automated gradient based meta",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                366,
+                                340,
+                                378
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        366,
+                                        340,
+                                        378
+                                    ],
+                                    "score": 1.0,
+                                    "content": "learner search. arXiv preprint arXiv:1806.06927, 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        }
+                    ],
+                    "index": 19,
+                    "bbox_fs": [
+                        105,
+                        343,
+                        505,
+                        378
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        384,
+                        504,
+                        408
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                384,
+                                506,
+                                399
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        384,
+                                        506,
+                                        399
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Diederik Kingma and Jimmy Ba. Adam: A method for stochastic optimization. In Proceedings of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                395,
+                                426,
+                                408
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        395,
+                                        426,
+                                        408
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the 3rd International Conference on Learning Representations (ICLR), 2014.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        }
+                    ],
+                    "index": 21.5,
+                    "bbox_fs": [
+                        105,
+                        384,
+                        506,
+                        408
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        414,
+                        504,
+                        438
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                415,
+                                505,
+                                428
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        415,
+                                        505,
+                                        428
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Diederik P Kingma, Shakir Mohamed, Danilo Jimenez Rezende, and Max Welling. Semi-supervised",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                426,
+                                501,
+                                439
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        426,
+                                        501,
+                                        439
+                                    ],
+                                    "score": 1.0,
+                                    "content": "learning with deep generative models. In Neural Information Processing Systems (NIPS), 2014.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        }
+                    ],
+                    "index": 23.5,
+                    "bbox_fs": [
+                        106,
+                        415,
+                        505,
+                        439
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        444,
+                        505,
+                        478
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                444,
+                                505,
+                                458
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        444,
+                                        505,
+                                        458
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Philipp Krahenb ¨ uhl, Carl Doersch, Jeff Donahue, and Trevor Darrell. Data-dependent initializa- ¨",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                455,
+                                505,
+                                468
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        455,
+                                        505,
+                                        468
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tions of convolutional neural networks. In International Conference on Learning Representations",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                467,
+                                175,
+                                479
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        467,
+                                        175,
+                                        479
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(ICLR), 2016.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        }
+                    ],
+                    "index": 26,
+                    "bbox_fs": [
+                        105,
+                        444,
+                        505,
+                        479
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        485,
+                        505,
+                        530
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                486,
+                                505,
+                                499
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        486,
+                                        505,
+                                        499
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Quoc V Le, Marc’Aurelio Ranzato, Rajat Monga, Matthieu Devin, Gregory S. Corrado, Kai Chen,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                497,
+                                506,
+                                510
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        497,
+                                        506,
+                                        510
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Jeffrey Dean, and Andrew Y Ng. Building high-level features using large scale unsupervised",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                114,
+                                507,
+                                506,
+                                522
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        114,
+                                        507,
+                                        506,
+                                        522
+                                    ],
+                                    "score": 1.0,
+                                    "content": "learning. In International Conference on Acoustics, Speech and Signal Processing (ICASSP),",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                518,
+                                143,
+                                531
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        518,
+                                        143,
+                                        531
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2013.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        }
+                    ],
+                    "index": 29.5,
+                    "bbox_fs": [
+                        106,
+                        486,
+                        506,
+                        531
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        109,
+                        537,
+                        503,
+                        572
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                107,
+                                537,
+                                505,
+                                550
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        537,
+                                        505,
+                                        550
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Michael F Mathieu, Junbo Jake Zhao, Junbo Zhao, Aditya Ramesh, Pablo Sprechmann, and Yann",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                549,
+                                505,
+                                561
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        549,
+                                        505,
+                                        561
+                                    ],
+                                    "score": 1.0,
+                                    "content": "LeCun. Disentangling factors of variation in deep representation using adversarial training. In",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                559,
+                                334,
+                                572
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        559,
+                                        334,
+                                        572
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Neural Information Processing Systems (NIPS), 2016.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        }
+                    ],
+                    "index": 33,
+                    "bbox_fs": [
+                        107,
+                        537,
+                        505,
+                        572
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        578,
+                        504,
+                        601
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                578,
+                                505,
+                                592
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        578,
+                                        505,
+                                        592
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Luke Metz, Niru Maheswaranathan, Brian Cheung, and Jascha Sohl-Dickstein. Learning to learn",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                589,
+                                469,
+                                602
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        589,
+                                        469,
+                                        602
+                                    ],
+                                    "score": 1.0,
+                                    "content": "without labels. In International Conference on Learning Representations (ICLR), 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        }
+                    ],
+                    "index": 35.5,
+                    "bbox_fs": [
+                        105,
+                        578,
+                        505,
+                        602
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        608,
+                        504,
+                        631
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                608,
+                                505,
+                                621
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        608,
+                                        505,
+                                        621
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Tsendsuren Munkhdalai and Hong Yu. Meta networks. In International Conference on Machine",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                619,
+                                216,
+                                632
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        619,
+                                        216,
+                                        632
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Learning (ICML), 2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        }
+                    ],
+                    "index": 37.5,
+                    "bbox_fs": [
+                        106,
+                        608,
+                        505,
+                        632
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        104,
+                        638,
+                        504,
+                        661
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                638,
+                                505,
+                                651
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        638,
+                                        505,
+                                        651
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Devang K Naik and RJ Mammone. Meta-neural networks that learn by learning. In International",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                649,
+                                333,
+                                662
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        649,
+                                        333,
+                                        662
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Joint Conference on Neural Networks (IJCNN), 1992.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        }
+                    ],
+                    "index": 39.5,
+                    "bbox_fs": [
+                        106,
+                        638,
+                        505,
+                        662
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        668,
+                        503,
+                        702
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                668,
+                                505,
+                                681
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        668,
+                                        505,
+                                        681
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Avital Oliver, Augustus Odena, Colin Raffel, Ekin D Cubuk, and Ian J Goodfellow. Realistic eval-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                679,
+                                505,
+                                693
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        679,
+                                        505,
+                                        693
+                                    ],
+                                    "score": 1.0,
+                                    "content": "uation of deep semi-supervised learning algorithms. In International Conference on Learning",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                690,
+                                241,
+                                703
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        690,
+                                        241,
+                                        703
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Representations (ICLR), 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        }
+                    ],
+                    "index": 42,
+                    "bbox_fs": [
+                        106,
+                        668,
+                        505,
+                        703
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        709,
+                        502,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                709,
+                                504,
+                                722
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        709,
+                                        504,
+                                        722
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Aaron van den Oord, Yazhe Li, and Oriol Vinyals. Representation learning with contrastive predic-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                721,
+                                329,
+                                732
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        721,
+                                        329,
+                                        732
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tive coding. arXiv preprint arXiv:1807.03748, 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        }
+                    ],
+                    "index": 44.5,
+                    "bbox_fs": [
+                        106,
+                        709,
+                        504,
+                        732
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        82,
+                        504,
+                        116
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                81,
+                                505,
+                                96
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        81,
+                                        505,
+                                        96
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Alec Radford, Luke Metz, and Soumith Chintala. Unsupervised representation learning with deep",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                93,
+                                505,
+                                107
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        93,
+                                        505,
+                                        107
+                                    ],
+                                    "score": 1.0,
+                                    "content": "convolutional generative adversarial networks. In International Conference on Learning Repre-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                105,
+                                218,
+                                117
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        105,
+                                        218,
+                                        117
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sentations (ICLR), 2016.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        }
+                    ],
+                    "index": 1
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        122,
+                        503,
+                        145
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                121,
+                                504,
+                                136
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        121,
+                                        504,
+                                        136
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Alec Radford, Karthik Narasimhan, Tim Salimans, and Ilya Sutskever. Improving language under-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                133,
+                                325,
+                                146
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        133,
+                                        325,
+                                        146
+                                    ],
+                                    "score": 1.0,
+                                    "content": "standing by generative pre-training. Preprint, 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        }
+                    ],
+                    "index": 3.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        151,
+                        503,
+                        175
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                150,
+                                505,
+                                165
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        150,
+                                        505,
+                                        165
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Prajit Ramachandran, Peter J Liu, and Quoc V Le. Unsupervised pretraining for sequence to se-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                162,
+                                477,
+                                176
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        162,
+                                        477,
+                                        176
+                                    ],
+                                    "score": 1.0,
+                                    "content": "quence learning. In Empirical Methods in Natural Language Processing (EMNLP), 2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        }
+                    ],
+                    "index": 5.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        180,
+                        504,
+                        214
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                181,
+                                506,
+                                193
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        181,
+                                        506,
+                                        193
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Marc’Aurelio Ranzato, Christopher Poultney, Sumit Chopra, and Yann LeCun. Efficient learning of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                191,
+                                506,
+                                205
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        191,
+                                        506,
+                                        205
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sparse representations with an energy-based model. In Neural Information Processing Systems",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                203,
+                                174,
+                                215
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        203,
+                                        174,
+                                        215
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(NIPS), 2006.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        }
+                    ],
+                    "index": 8
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        220,
+                        504,
+                        254
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                221,
+                                505,
+                                233
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        221,
+                                        505,
+                                        233
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Antti Rasmus, Mathias Berglund, Mikko Honkala, Harri Valpola, and Tapani Raiko. Semi-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                231,
+                                505,
+                                245
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        231,
+                                        505,
+                                        245
+                                    ],
+                                    "score": 1.0,
+                                    "content": "supervised learning with ladder networks. In Neural Information Processing Systems (NIPS),",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                114,
+                                241,
+                                143,
+                                255
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        114,
+                                        241,
+                                        143,
+                                        255
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2015.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        }
+                    ],
+                    "index": 11
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        261,
+                        505,
+                        284
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                261,
+                                505,
+                                273
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        261,
+                                        505,
+                                        273
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Sachin Ravi and Hugo Larochelle. Optimization as a model for few-shot learning. In International",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                117,
+                                272,
+                                341,
+                                284
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        117,
+                                        272,
+                                        341,
+                                        284
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Conference on Learning Representations (ICLR), 2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        }
+                    ],
+                    "index": 13.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        289,
+                        504,
+                        324
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                289,
+                                506,
+                                303
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        289,
+                                        506,
+                                        303
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Scott Reed, Kihyuk Sohn, Yuting Zhang, and Honglak Lee. Learning to disentangle factors of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                300,
+                                505,
+                                314
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        300,
+                                        505,
+                                        314
+                                    ],
+                                    "score": 1.0,
+                                    "content": "variation with manifold interaction. In International Conference on Machine Learning (ICML),",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                114,
+                                311,
+                                143,
+                                324
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        114,
+                                        311,
+                                        143,
+                                        324
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2014.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        }
+                    ],
+                    "index": 16
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        330,
+                        505,
+                        375
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                330,
+                                506,
+                                343
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        330,
+                                        506,
+                                        343
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Olga Russakovsky, Jia Deng, Hao Su, Jonathan Krause, Sanjeev Satheesh, Sean Ma, Zhiheng",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                340,
+                                505,
+                                353
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        340,
+                                        505,
+                                        353
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Huang, Andrej Karpathy, Aditya Khosla, Michael Bernstein, Alexander C. Berg, and Fei-Fei",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                352,
+                                506,
+                                366
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        352,
+                                        506,
+                                        366
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Li. ImageNet large scale visual recognition challenge. International Journal of Computer Vision",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                363,
+                                174,
+                                375
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        363,
+                                        174,
+                                        375
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(IJCV), 2015.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        }
+                    ],
+                    "index": 19.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        381,
+                        504,
+                        404
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                380,
+                                504,
+                                394
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        380,
+                                        504,
+                                        394
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Tim Salimans, Ian Goodfellow, Wojciech Zaremba, Vicki Cheung, Alec Radford, and Xi Chen.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                117,
+                                392,
+                                504,
+                                404
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        117,
+                                        392,
+                                        504,
+                                        404
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Improved techniques for training GANs. In Neural Information Processing Systems (NIPS), 2016.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        }
+                    ],
+                    "index": 22.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        410,
+                        504,
+                        444
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                409,
+                                505,
+                                424
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        409,
+                                        505,
+                                        424
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Adam Santoro, Sergey Bartunov, Matthew Botvinick, Daan Wierstra, and Timothy Lillicrap. Meta-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                117,
+                                423,
+                                505,
+                                433
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        117,
+                                        423,
+                                        505,
+                                        433
+                                    ],
+                                    "score": 1.0,
+                                    "content": "learning with memory-augmented neural networks. In International Conference on Machine",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                432,
+                                216,
+                                444
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        432,
+                                        216,
+                                        444
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Learning (ICML), 2016.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        }
+                    ],
+                    "index": 25
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        450,
+                        504,
+                        473
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                450,
+                                505,
+                                463
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        450,
+                                        505,
+                                        463
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Jurgen Schmidhuber. ¨ Evolutionary principles in self-referential learning. PhD thesis, Institut fur¨",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                461,
+                                325,
+                                472
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        461,
+                                        325,
+                                        472
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Informatik, Technische Universitat M¨ unchen, 1987.¨",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        }
+                    ],
+                    "index": 27.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        479,
+                        505,
+                        513
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                479,
+                                504,
+                                492
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        479,
+                                        504,
+                                        492
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Evan Shelhamer, Parsa Mahmoudieh, Max Argus, and Trevor Darrell. Loss is its own reward: Self-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                491,
+                                505,
+                                503
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        491,
+                                        505,
+                                        503
+                                    ],
+                                    "score": 1.0,
+                                    "content": "supervision for reinforcement learning. In International Conference on Learning Representations",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                501,
+                                175,
+                                514
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        501,
+                                        175,
+                                        514
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(ICLR), 2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        }
+                    ],
+                    "index": 30
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        104,
+                        519,
+                        504,
+                        543
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                519,
+                                505,
+                                532
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        519,
+                                        505,
+                                        532
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Jake Snell, Kevin Swersky, and Richard S Zemel. Prototypical networks for few-shot learning. In",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                531,
+                                335,
+                                543
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        531,
+                                        335,
+                                        543
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Neural Information Processing Systems (NIPS), 2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        }
+                    ],
+                    "index": 32.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        548,
+                        505,
+                        583
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                548,
+                                505,
+                                562
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        548,
+                                        505,
+                                        562
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Nitish Srivastava, Geoffrey Hinton, Alex Krizhevsky, Ilya Sutskever, and Ruslan Salakhutdinov.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                559,
+                                505,
+                                574
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        559,
+                                        505,
+                                        574
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Dropout: A simple way to prevent neural networks from overfitting. Journal of Machine Learning",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                572,
+                                217,
+                                583
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        572,
+                                        217,
+                                        583
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Research (JMLR), 2014.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        }
+                    ],
+                    "index": 35
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        588,
+                        505,
+                        623
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                588,
+                                505,
+                                602
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        588,
+                                        505,
+                                        602
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Pascal Vincent, Hugo Larochelle, Yoshua Bengio, and Pierre-Antoine Manzagol. Extracting and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                600,
+                                505,
+                                613
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        600,
+                                        505,
+                                        613
+                                    ],
+                                    "score": 1.0,
+                                    "content": "composing robust features with denoising autoencoders. In International Conference on Machine",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                611,
+                                216,
+                                624
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        611,
+                                        216,
+                                        624
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Learning (ICML), 2008.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        }
+                    ],
+                    "index": 38
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        628,
+                        505,
+                        663
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                628,
+                                505,
+                                642
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        628,
+                                        505,
+                                        642
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Pascal Vincent, Hugo Larochelle, Isabelle Lajoie, Yoshua Bengio, and Pierre-Antoine Manzagol.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                640,
+                                506,
+                                654
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        640,
+                                        506,
+                                        654
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Stacked denoising autoencoders: Learning useful representations in a deep network with a local",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                651,
+                                417,
+                                664
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        651,
+                                        417,
+                                        664
+                                    ],
+                                    "score": 1.0,
+                                    "content": "denoising criterion. Journal of Machine Learning Research (JMLR), 2010.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        }
+                    ],
+                    "index": 41
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        669,
+                        503,
+                        693
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                670,
+                                504,
+                                682
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        670,
+                                        504,
+                                        682
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Oriol Vinyals, Charles Blundell, Timothy Lillicrap, Koray Kavukcuoglu, and Daan Wierstra. Match-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                680,
+                                488,
+                                693
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        680,
+                                        488,
+                                        693
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ing networks for one shot learning. In Neural Information Processing Systems (NIPS), 2016.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        }
+                    ],
+                    "index": 43.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        698,
+                        504,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                698,
+                                506,
+                                712
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        698,
+                                        506,
+                                        712
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Dong Yu, Li Deng, and George Dahl. Roles of pre-training and fine-tuning in context-dependent",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                709,
+                                504,
+                                722
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        709,
+                                        504,
+                                        722
+                                    ],
+                                    "score": 1.0,
+                                    "content": "DBN-HMMs for real-world speech recognition. In NIPS Workshop on Deep Learning and Unsu-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        },
+                        {
+                            "bbox": [
+                                114,
+                                721,
+                                252,
+                                733
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        114,
+                                        721,
+                                        252,
+                                        733
+                                    ],
+                                    "score": 1.0,
+                                    "content": "pervised Feature Learning, 2010.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 47
+                        }
+                    ],
+                    "index": 46
+                }
+            ],
+            "page_idx": 12,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        108,
+                        27,
+                        293,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                26,
+                                294,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        26,
+                                        294,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Published as a conference paper at ICLR 2019",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        300,
+                        751,
+                        311,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                299,
+                                750,
+                                312,
+                                764
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        299,
+                                        750,
+                                        312,
+                                        764
+                                    ],
+                                    "score": 1.0,
+                                    "content": "13",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        82,
+                        504,
+                        116
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                81,
+                                505,
+                                96
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        81,
+                                        505,
+                                        96
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Alec Radford, Luke Metz, and Soumith Chintala. Unsupervised representation learning with deep",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                93,
+                                505,
+                                107
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        93,
+                                        505,
+                                        107
+                                    ],
+                                    "score": 1.0,
+                                    "content": "convolutional generative adversarial networks. In International Conference on Learning Repre-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                105,
+                                218,
+                                117
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        105,
+                                        218,
+                                        117
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sentations (ICLR), 2016.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        }
+                    ],
+                    "index": 1,
+                    "bbox_fs": [
+                        105,
+                        81,
+                        505,
+                        117
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        122,
+                        503,
+                        145
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                121,
+                                504,
+                                136
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        121,
+                                        504,
+                                        136
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Alec Radford, Karthik Narasimhan, Tim Salimans, and Ilya Sutskever. Improving language under-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                133,
+                                325,
+                                146
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        133,
+                                        325,
+                                        146
+                                    ],
+                                    "score": 1.0,
+                                    "content": "standing by generative pre-training. Preprint, 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        }
+                    ],
+                    "index": 3.5,
+                    "bbox_fs": [
+                        105,
+                        121,
+                        504,
+                        146
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        151,
+                        503,
+                        175
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                150,
+                                505,
+                                165
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        150,
+                                        505,
+                                        165
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Prajit Ramachandran, Peter J Liu, and Quoc V Le. Unsupervised pretraining for sequence to se-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                162,
+                                477,
+                                176
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        162,
+                                        477,
+                                        176
+                                    ],
+                                    "score": 1.0,
+                                    "content": "quence learning. In Empirical Methods in Natural Language Processing (EMNLP), 2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        }
+                    ],
+                    "index": 5.5,
+                    "bbox_fs": [
+                        105,
+                        150,
+                        505,
+                        176
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        180,
+                        504,
+                        214
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                181,
+                                506,
+                                193
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        181,
+                                        506,
+                                        193
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Marc’Aurelio Ranzato, Christopher Poultney, Sumit Chopra, and Yann LeCun. Efficient learning of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                191,
+                                506,
+                                205
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        191,
+                                        506,
+                                        205
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sparse representations with an energy-based model. In Neural Information Processing Systems",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                203,
+                                174,
+                                215
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        203,
+                                        174,
+                                        215
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(NIPS), 2006.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        }
+                    ],
+                    "index": 8,
+                    "bbox_fs": [
+                        106,
+                        181,
+                        506,
+                        215
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        220,
+                        504,
+                        254
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                221,
+                                505,
+                                233
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        221,
+                                        505,
+                                        233
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Antti Rasmus, Mathias Berglund, Mikko Honkala, Harri Valpola, and Tapani Raiko. Semi-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                231,
+                                505,
+                                245
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        231,
+                                        505,
+                                        245
+                                    ],
+                                    "score": 1.0,
+                                    "content": "supervised learning with ladder networks. In Neural Information Processing Systems (NIPS),",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                114,
+                                241,
+                                143,
+                                255
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        114,
+                                        241,
+                                        143,
+                                        255
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2015.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        }
+                    ],
+                    "index": 11,
+                    "bbox_fs": [
+                        105,
+                        221,
+                        505,
+                        255
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        261,
+                        505,
+                        284
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                261,
+                                505,
+                                273
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        261,
+                                        505,
+                                        273
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Sachin Ravi and Hugo Larochelle. Optimization as a model for few-shot learning. In International",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                117,
+                                272,
+                                341,
+                                284
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        117,
+                                        272,
+                                        341,
+                                        284
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Conference on Learning Representations (ICLR), 2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        }
+                    ],
+                    "index": 13.5,
+                    "bbox_fs": [
+                        106,
+                        261,
+                        505,
+                        284
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        289,
+                        504,
+                        324
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                289,
+                                506,
+                                303
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        289,
+                                        506,
+                                        303
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Scott Reed, Kihyuk Sohn, Yuting Zhang, and Honglak Lee. Learning to disentangle factors of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                300,
+                                505,
+                                314
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        300,
+                                        505,
+                                        314
+                                    ],
+                                    "score": 1.0,
+                                    "content": "variation with manifold interaction. In International Conference on Machine Learning (ICML),",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                114,
+                                311,
+                                143,
+                                324
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        114,
+                                        311,
+                                        143,
+                                        324
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2014.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        }
+                    ],
+                    "index": 16,
+                    "bbox_fs": [
+                        105,
+                        289,
+                        506,
+                        324
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        330,
+                        505,
+                        375
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                330,
+                                506,
+                                343
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        330,
+                                        506,
+                                        343
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Olga Russakovsky, Jia Deng, Hao Su, Jonathan Krause, Sanjeev Satheesh, Sean Ma, Zhiheng",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                340,
+                                505,
+                                353
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        340,
+                                        505,
+                                        353
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Huang, Andrej Karpathy, Aditya Khosla, Michael Bernstein, Alexander C. Berg, and Fei-Fei",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                352,
+                                506,
+                                366
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        352,
+                                        506,
+                                        366
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Li. ImageNet large scale visual recognition challenge. International Journal of Computer Vision",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                363,
+                                174,
+                                375
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        363,
+                                        174,
+                                        375
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(IJCV), 2015.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        }
+                    ],
+                    "index": 19.5,
+                    "bbox_fs": [
+                        105,
+                        330,
+                        506,
+                        375
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        381,
+                        504,
+                        404
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                380,
+                                504,
+                                394
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        380,
+                                        504,
+                                        394
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Tim Salimans, Ian Goodfellow, Wojciech Zaremba, Vicki Cheung, Alec Radford, and Xi Chen.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                117,
+                                392,
+                                504,
+                                404
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        117,
+                                        392,
+                                        504,
+                                        404
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Improved techniques for training GANs. In Neural Information Processing Systems (NIPS), 2016.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        }
+                    ],
+                    "index": 22.5,
+                    "bbox_fs": [
+                        106,
+                        380,
+                        504,
+                        404
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        410,
+                        504,
+                        444
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                409,
+                                505,
+                                424
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        409,
+                                        505,
+                                        424
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Adam Santoro, Sergey Bartunov, Matthew Botvinick, Daan Wierstra, and Timothy Lillicrap. Meta-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                117,
+                                423,
+                                505,
+                                433
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        117,
+                                        423,
+                                        505,
+                                        433
+                                    ],
+                                    "score": 1.0,
+                                    "content": "learning with memory-augmented neural networks. In International Conference on Machine",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                432,
+                                216,
+                                444
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        432,
+                                        216,
+                                        444
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Learning (ICML), 2016.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        }
+                    ],
+                    "index": 25,
+                    "bbox_fs": [
+                        105,
+                        409,
+                        505,
+                        444
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        450,
+                        504,
+                        473
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                450,
+                                505,
+                                463
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        450,
+                                        505,
+                                        463
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Jurgen Schmidhuber. ¨ Evolutionary principles in self-referential learning. PhD thesis, Institut fur¨",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                461,
+                                325,
+                                472
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        461,
+                                        325,
+                                        472
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Informatik, Technische Universitat M¨ unchen, 1987.¨",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        }
+                    ],
+                    "index": 27.5,
+                    "bbox_fs": [
+                        105,
+                        450,
+                        505,
+                        472
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        479,
+                        505,
+                        513
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                479,
+                                504,
+                                492
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        479,
+                                        504,
+                                        492
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Evan Shelhamer, Parsa Mahmoudieh, Max Argus, and Trevor Darrell. Loss is its own reward: Self-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                491,
+                                505,
+                                503
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        491,
+                                        505,
+                                        503
+                                    ],
+                                    "score": 1.0,
+                                    "content": "supervision for reinforcement learning. In International Conference on Learning Representations",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                501,
+                                175,
+                                514
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        501,
+                                        175,
+                                        514
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(ICLR), 2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        }
+                    ],
+                    "index": 30,
+                    "bbox_fs": [
+                        106,
+                        479,
+                        505,
+                        514
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        104,
+                        519,
+                        504,
+                        543
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                519,
+                                505,
+                                532
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        519,
+                                        505,
+                                        532
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Jake Snell, Kevin Swersky, and Richard S Zemel. Prototypical networks for few-shot learning. In",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                531,
+                                335,
+                                543
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        531,
+                                        335,
+                                        543
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Neural Information Processing Systems (NIPS), 2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        }
+                    ],
+                    "index": 32.5,
+                    "bbox_fs": [
+                        105,
+                        519,
+                        505,
+                        543
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        548,
+                        505,
+                        583
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                548,
+                                505,
+                                562
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        548,
+                                        505,
+                                        562
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Nitish Srivastava, Geoffrey Hinton, Alex Krizhevsky, Ilya Sutskever, and Ruslan Salakhutdinov.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                559,
+                                505,
+                                574
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        559,
+                                        505,
+                                        574
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Dropout: A simple way to prevent neural networks from overfitting. Journal of Machine Learning",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                572,
+                                217,
+                                583
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        572,
+                                        217,
+                                        583
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Research (JMLR), 2014.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        }
+                    ],
+                    "index": 35,
+                    "bbox_fs": [
+                        105,
+                        548,
+                        505,
+                        583
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        588,
+                        505,
+                        623
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                588,
+                                505,
+                                602
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        588,
+                                        505,
+                                        602
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Pascal Vincent, Hugo Larochelle, Yoshua Bengio, and Pierre-Antoine Manzagol. Extracting and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                600,
+                                505,
+                                613
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        600,
+                                        505,
+                                        613
+                                    ],
+                                    "score": 1.0,
+                                    "content": "composing robust features with denoising autoencoders. In International Conference on Machine",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                611,
+                                216,
+                                624
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        611,
+                                        216,
+                                        624
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Learning (ICML), 2008.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        }
+                    ],
+                    "index": 38,
+                    "bbox_fs": [
+                        105,
+                        588,
+                        505,
+                        624
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        628,
+                        505,
+                        663
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                628,
+                                505,
+                                642
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        628,
+                                        505,
+                                        642
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Pascal Vincent, Hugo Larochelle, Isabelle Lajoie, Yoshua Bengio, and Pierre-Antoine Manzagol.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                640,
+                                506,
+                                654
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        640,
+                                        506,
+                                        654
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Stacked denoising autoencoders: Learning useful representations in a deep network with a local",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                651,
+                                417,
+                                664
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        651,
+                                        417,
+                                        664
+                                    ],
+                                    "score": 1.0,
+                                    "content": "denoising criterion. Journal of Machine Learning Research (JMLR), 2010.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        }
+                    ],
+                    "index": 41,
+                    "bbox_fs": [
+                        105,
+                        628,
+                        506,
+                        664
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        669,
+                        503,
+                        693
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                670,
+                                504,
+                                682
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        670,
+                                        504,
+                                        682
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Oriol Vinyals, Charles Blundell, Timothy Lillicrap, Koray Kavukcuoglu, and Daan Wierstra. Match-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                680,
+                                488,
+                                693
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        680,
+                                        488,
+                                        693
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ing networks for one shot learning. In Neural Information Processing Systems (NIPS), 2016.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        }
+                    ],
+                    "index": 43.5,
+                    "bbox_fs": [
+                        106,
+                        670,
+                        504,
+                        693
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        698,
+                        504,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                698,
+                                506,
+                                712
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        698,
+                                        506,
+                                        712
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Dong Yu, Li Deng, and George Dahl. Roles of pre-training and fine-tuning in context-dependent",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                709,
+                                504,
+                                722
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        709,
+                                        504,
+                                        722
+                                    ],
+                                    "score": 1.0,
+                                    "content": "DBN-HMMs for real-world speech recognition. In NIPS Workshop on Deep Learning and Unsu-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        },
+                        {
+                            "bbox": [
+                                114,
+                                721,
+                                252,
+                                733
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        114,
+                                        721,
+                                        252,
+                                        733
+                                    ],
+                                    "score": 1.0,
+                                    "content": "pervised Feature Learning, 2010.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 47
+                        }
+                    ],
+                    "index": 46,
+                    "bbox_fs": [
+                        105,
+                        698,
+                        506,
+                        733
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        82,
+                        505,
+                        105
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                80,
+                                506,
+                                97
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        80,
+                                        506,
+                                        97
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Richard Zhang, Phillip Isola, and Alexei A Efros. Split-brain autoencoders: Unsupervised learning",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                93,
+                                474,
+                                106
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        93,
+                                        474,
+                                        106
+                                    ],
+                                    "score": 1.0,
+                                    "content": "by cross-channel prediction. In Computer Vision and Pattern Recognition (CVPR), 2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        }
+                    ],
+                    "index": 0.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        112,
+                        489,
+                        124
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                111,
+                                488,
+                                126
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        111,
+                                        488,
+                                        126
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Xiaojin Zhu. Semi-supervised learning. In Encyclopedia of Machine Learning. Springer, 2011.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        }
+                    ],
+                    "index": 2
+                }
+            ],
+            "page_idx": 13,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        107,
+                        26,
+                        293,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                25,
+                                294,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        25,
+                                        294,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Published as a conference paper at ICLR 2019",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        300,
+                        751,
+                        311,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                299,
+                                750,
+                                313,
+                                763
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        299,
+                                        750,
+                                        313,
+                                        763
+                                    ],
+                                    "score": 1.0,
+                                    "content": "14",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        82,
+                        505,
+                        105
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                80,
+                                506,
+                                97
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        80,
+                                        506,
+                                        97
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Richard Zhang, Phillip Isola, and Alexei A Efros. Split-brain autoencoders: Unsupervised learning",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                93,
+                                474,
+                                106
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        93,
+                                        474,
+                                        106
+                                    ],
+                                    "score": 1.0,
+                                    "content": "by cross-channel prediction. In Computer Vision and Pattern Recognition (CVPR), 2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        }
+                    ],
+                    "index": 0.5,
+                    "bbox_fs": [
+                        104,
+                        80,
+                        506,
+                        106
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        112,
+                        489,
+                        124
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                111,
+                                488,
+                                126
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        111,
+                                        488,
+                                        126
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Xiaojin Zhu. Semi-supervised learning. In Encyclopedia of Machine Learning. Springer, 2011.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        }
+                    ],
+                    "index": 2,
+                    "bbox_fs": [
+                        106,
+                        111,
+                        488,
+                        126
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        81,
+                        358,
+                        94
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                80,
+                                360,
+                                96
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        80,
+                                        360,
+                                        96
+                                    ],
+                                    "score": 1.0,
+                                    "content": "APPENDIX A THE EMBEDDING LEARNING ZOO",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        }
+                    ],
+                    "index": 0
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        107,
+                        491,
+                        119
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                105,
+                                492,
+                                122
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        105,
+                                        492,
+                                        122
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We evaluate four distinct methods from prior work for learning the task-generating embeddings.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        }
+                    ],
+                    "index": 1
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        124,
+                        505,
+                        201
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                124,
+                                505,
+                                137
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        124,
+                                        505,
+                                        137
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In adversarially constrained autoencoder interpolation (ACAI), a convolutional autoencoder’s pixel-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                134,
+                                506,
+                                149
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        134,
+                                        129,
+                                        149
+                                    ],
+                                    "score": 1.0,
+                                    "content": "wise",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        130,
+                                        135,
+                                        142,
+                                        145
+                                    ],
+                                    "score": 0.86,
+                                    "content": "L ^ { 2 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        143,
+                                        134,
+                                        506,
+                                        149
+                                    ],
+                                    "score": 1.0,
+                                    "content": "loss is regularized with a term encouraging meaningful interpolations in the latent",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                145,
+                                506,
+                                160
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        145,
+                                        506,
+                                        160
+                                    ],
+                                    "score": 1.0,
+                                    "content": "space (Berthelot et al., 2018). Specifically, a critic network takes as input a synthetic image gener-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                157,
+                                505,
+                                171
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        157,
+                                        505,
+                                        171
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ated from a convex combination of the latents of two dataset samples, and regresses to the mixing",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                166,
+                                505,
+                                182
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        166,
+                                        505,
+                                        182
+                                    ],
+                                    "score": 1.0,
+                                    "content": "factor. The decoder of the autoencoder and the generator for the critic are one and the same. The reg-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                179,
+                                505,
+                                192
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        179,
+                                        505,
+                                        192
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ularization term is minimized when the autoencoder fools the critic into predicting that the synthetic",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                190,
+                                200,
+                                203
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        190,
+                                        200,
+                                        203
+                                    ],
+                                    "score": 1.0,
+                                    "content": "image is a real sample.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        }
+                    ],
+                    "index": 5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        207,
+                        504,
+                        251
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                207,
+                                506,
+                                219
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        207,
+                                        506,
+                                        219
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The bidirectional GAN (BiGAN) is an instance of a generative-adversarial framework in which",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                218,
+                                506,
+                                231
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        218,
+                                        506,
+                                        231
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the generator produces both synthetic image and embedding from real embedding and image, re-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                228,
+                                506,
+                                243
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        228,
+                                        506,
+                                        243
+                                    ],
+                                    "score": 1.0,
+                                    "content": "spectively (Donahue et al., 2017; Dumoulin et al., 2017). Discrimination is done in joint image-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                239,
+                                181,
+                                254
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        239,
+                                        181,
+                                        254
+                                    ],
+                                    "score": 1.0,
+                                    "content": "embedding space.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        }
+                    ],
+                    "index": 10.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        257,
+                        504,
+                        290
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                257,
+                                504,
+                                269
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        257,
+                                        504,
+                                        269
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The DeepCluster method does discriminative clustering by alternating between clustering the fea-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                269,
+                                505,
+                                280
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        269,
+                                        505,
+                                        280
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tures of a convolutional neural network and using the clusters as labels to optimize the network",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                279,
+                                422,
+                                291
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        279,
+                                        422,
+                                        291
+                                    ],
+                                    "score": 1.0,
+                                    "content": "weights via backpropagating a standard classification loss (Caron et al., 2018).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        }
+                    ],
+                    "index": 14
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        296,
+                        505,
+                        362
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                295,
+                                505,
+                                308
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        295,
+                                        505,
+                                        308
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The InfoGAN framework conceptually decomposes the generator’s input into a latent code and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                307,
+                                505,
+                                318
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        307,
+                                        505,
+                                        318
+                                    ],
+                                    "score": 1.0,
+                                    "content": "incompressible noise (Chen et al., 2016). The structure of the latent code is hand-specified based",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                318,
+                                505,
+                                329
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        318,
+                                        505,
+                                        329
+                                    ],
+                                    "score": 1.0,
+                                    "content": "on knowledge of the dataset. The canonical GAN minimax objective is regularized with a mutual",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                327,
+                                506,
+                                343
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        327,
+                                        506,
+                                        343
+                                    ],
+                                    "score": 1.0,
+                                    "content": "information term between the code and the generated image. In practice, this term is optimized using",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                339,
+                                505,
+                                352
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        339,
+                                        505,
+                                        352
+                                    ],
+                                    "score": 1.0,
+                                    "content": "variational inference, involving the approximation of the posterior with an auxiliary distribution",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                351,
+                                334,
+                                363
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        351,
+                                        115,
+                                        363
+                                    ],
+                                    "score": 0.51,
+                                    "content": "Q",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        116,
+                                        351,
+                                        334,
+                                        363
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(code|image) parameterized by a recognition network.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        }
+                    ],
+                    "index": 18.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        367,
+                        505,
+                        444
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                367,
+                                505,
+                                380
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        367,
+                                        505,
+                                        380
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Whereas ACAI explicitly optimizes pixel-wise reconstruction error, BiGAN only encourages the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                378,
+                                505,
+                                392
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        378,
+                                        505,
+                                        392
+                                    ],
+                                    "score": 1.0,
+                                    "content": "fidelity of generated image and latent samples with respect to their respective prior distributions.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                388,
+                                505,
+                                403
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        388,
+                                        505,
+                                        403
+                                    ],
+                                    "score": 1.0,
+                                    "content": "While InfoGAN also encourages the fidelity of generated images, it leverages domain-specific",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                400,
+                                505,
+                                412
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        400,
+                                        505,
+                                        412
+                                    ],
+                                    "score": 1.0,
+                                    "content": "knowledge to impose a favorable structure on the embedding space and information-theoretic meth-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                412,
+                                505,
+                                424
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        412,
+                                        505,
+                                        424
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ods for optimization. DeepCluster departs from the aforementioned methods in that it is not con-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                422,
+                                505,
+                                436
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        422,
+                                        505,
+                                        436
+                                    ],
+                                    "score": 1.0,
+                                    "content": "cerned with generation or decoding, and only seeks to learn general-purpose visual features by way",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                432,
+                                266,
+                                447
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        432,
+                                        266,
+                                        447
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of end-to-end discriminative clustering.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        }
+                    ],
+                    "index": 25
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        463,
+                        308,
+                        476
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                462,
+                                310,
+                                478
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        462,
+                                        310,
+                                        478
+                                    ],
+                                    "score": 1.0,
+                                    "content": "APPENDIX B DATASET INFORMATION",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        }
+                    ],
+                    "index": 29
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        489,
+                        505,
+                        567
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                488,
+                                506,
+                                503
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        488,
+                                        506,
+                                        503
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The Omniglot dataset consists of 1623 characters each with 20 hand-drawn examples. Ignoring",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                501,
+                                505,
+                                513
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        501,
+                                        505,
+                                        513
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the alphabets from which the characters originate, we use 1100, 100, and 423 characters for our",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                511,
+                                505,
+                                523
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        511,
+                                        505,
+                                        523
+                                    ],
+                                    "score": 1.0,
+                                    "content": "meta-training, meta-validation, and meta-testing splits. The miniImageNet dataset consists of 100",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                522,
+                                506,
+                                535
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        522,
+                                        506,
+                                        535
+                                    ],
+                                    "score": 1.0,
+                                    "content": "classes each with 600 examples. The images are predominantly natural and realistic. We use the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                533,
+                                505,
+                                546
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        533,
+                                        505,
+                                        546
+                                    ],
+                                    "score": 1.0,
+                                    "content": "same meta-training/meta-validation/meta-testing splits of 64/16/20 classes as proposed by Ravi &",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                543,
+                                505,
+                                558
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        543,
+                                        505,
+                                        558
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Larochelle (2017). The CelebA dataset includes 202,599 facial images of celebrities and 40 binary",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                555,
+                                492,
+                                568
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        555,
+                                        492,
+                                        568
+                                    ],
+                                    "score": 1.0,
+                                    "content": "attributes that annotate every image. We follow the prescribed 162,770/19,867/19,962 data split.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        }
+                    ],
+                    "index": 33
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        572,
+                        504,
+                        616
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                571,
+                                504,
+                                585
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        571,
+                                        504,
+                                        585
+                                    ],
+                                    "score": 1.0,
+                                    "content": "For Omniglot and miniImageNet, supervised meta-learning tasks and evaluation tasks are con-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                582,
+                                506,
+                                596
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        582,
+                                        314,
+                                        596
+                                    ],
+                                    "score": 1.0,
+                                    "content": "structed exactly as detailed in Section 2.2: for an",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        315,
+                                        583,
+                                        325,
+                                        593
+                                    ],
+                                    "score": 0.83,
+                                    "content": "N",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        325,
+                                        582,
+                                        347,
+                                        596
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-way",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        347,
+                                        583,
+                                        357,
+                                        593
+                                    ],
+                                    "score": 0.82,
+                                    "content": "K",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        358,
+                                        582,
+                                        420,
+                                        596
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-shot task with",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        420,
+                                        583,
+                                        430,
+                                        594
+                                    ],
+                                    "score": 0.85,
+                                    "content": "Q",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        430,
+                                        582,
+                                        506,
+                                        596
+                                    ],
+                                    "score": 1.0,
+                                    "content": "queries per class,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                594,
+                                505,
+                                606
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        595,
+                                        152,
+                                        606
+                                    ],
+                                    "score": 1.0,
+                                    "content": "we sample",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        153,
+                                        594,
+                                        163,
+                                        604
+                                    ],
+                                    "score": 0.77,
+                                    "content": "N",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        164,
+                                        595,
+                                        291,
+                                        606
+                                    ],
+                                    "score": 1.0,
+                                    "content": "classes from the data split and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        291,
+                                        594,
+                                        322,
+                                        606
+                                    ],
+                                    "score": 0.92,
+                                    "content": "K + Q",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        323,
+                                        595,
+                                        505,
+                                        606
+                                    ],
+                                    "score": 1.0,
+                                    "content": "datapoints per class, labeling the task’s data",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                605,
+                                307,
+                                617
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        605,
+                                        228,
+                                        617
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with a random permutation of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        228,
+                                        605,
+                                        238,
+                                        615
+                                    ],
+                                    "score": 0.81,
+                                    "content": "N",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        239,
+                                        605,
+                                        307,
+                                        617
+                                    ],
+                                    "score": 1.0,
+                                    "content": "one-hot vectors.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        }
+                    ],
+                    "index": 38.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        622,
+                        504,
+                        731
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                621,
+                                506,
+                                634
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        621,
+                                        506,
+                                        634
+                                    ],
+                                    "score": 1.0,
+                                    "content": "For CelebA, we consider binary classification tasks (i.e., 2-way), each defined by 3 attributes and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                632,
+                                505,
+                                645
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        632,
+                                        505,
+                                        645
+                                    ],
+                                    "score": 1.0,
+                                    "content": "an ordering of 3 Booleans, one for each attribute. Every image in a task-specific class shares",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                643,
+                                506,
+                                656
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        643,
+                                        506,
+                                        656
+                                    ],
+                                    "score": 1.0,
+                                    "content": "all task-specific attributes with each other and none with images in the other class. For exam-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                655,
+                                506,
+                                667
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        655,
+                                        506,
+                                        667
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ple, the task illustrated in Figure 2 involves distinguishing between images whose subjects sat-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                666,
+                                505,
+                                678
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        666,
+                                        505,
+                                        678
+                                    ],
+                                    "score": 1.0,
+                                    "content": "isfy not Sideburns, Straight Hair, and not Young, and those whose subjects satisfy",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                677,
+                                506,
+                                689
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        677,
+                                        506,
+                                        689
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Sideburns, not Straight Hair, and Young. To keep with the idea of having distinct",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                687,
+                                506,
+                                700
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        687,
+                                        506,
+                                        700
+                                    ],
+                                    "score": 1.0,
+                                    "content": "classes for meta-training and meta-testing, we split the task-defining attributes. For the supervised",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 47
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                699,
+                                506,
+                                711
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        699,
+                                        506,
+                                        711
+                                    ],
+                                    "score": 1.0,
+                                    "content": "meta-learning oracle, we construct meta-training tasks from the first 20 attributes (when alphabet-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 48
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                709,
+                                505,
+                                722
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        709,
+                                        505,
+                                        722
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ically ordered), meta-validation tasks from the next 10, and meta-testing tasks from the last 10.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 49
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                720,
+                                506,
+                                733
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        720,
+                                        506,
+                                        733
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Discarding tasks with too few examples in either class, this results in 4287, 391, and 402 task",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 50
+                        }
+                    ],
+                    "index": 45.5
+                }
+            ],
+            "page_idx": 14,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        108,
+                        27,
+                        293,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                26,
+                                294,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        26,
+                                        294,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Published as a conference paper at ICLR 2019",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        301,
+                        751,
+                        311,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                299,
+                                750,
+                                312,
+                                764
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        299,
+                                        750,
+                                        312,
+                                        764
+                                    ],
+                                    "score": 1.0,
+                                    "content": "",
+                                    "type": "text",
+                                    "height": 14,
+                                    "width": 13
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        81,
+                        358,
+                        94
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                80,
+                                360,
+                                96
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        80,
+                                        360,
+                                        96
+                                    ],
+                                    "score": 1.0,
+                                    "content": "APPENDIX A THE EMBEDDING LEARNING ZOO",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        }
+                    ],
+                    "index": 0
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        107,
+                        491,
+                        119
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                105,
+                                492,
+                                122
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        105,
+                                        492,
+                                        122
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We evaluate four distinct methods from prior work for learning the task-generating embeddings.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        }
+                    ],
+                    "index": 1,
+                    "bbox_fs": [
+                        105,
+                        105,
+                        492,
+                        122
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        124,
+                        505,
+                        201
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                124,
+                                505,
+                                137
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        124,
+                                        505,
+                                        137
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In adversarially constrained autoencoder interpolation (ACAI), a convolutional autoencoder’s pixel-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                134,
+                                506,
+                                149
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        134,
+                                        129,
+                                        149
+                                    ],
+                                    "score": 1.0,
+                                    "content": "wise",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        130,
+                                        135,
+                                        142,
+                                        145
+                                    ],
+                                    "score": 0.86,
+                                    "content": "L ^ { 2 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        143,
+                                        134,
+                                        506,
+                                        149
+                                    ],
+                                    "score": 1.0,
+                                    "content": "loss is regularized with a term encouraging meaningful interpolations in the latent",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                145,
+                                506,
+                                160
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        145,
+                                        506,
+                                        160
+                                    ],
+                                    "score": 1.0,
+                                    "content": "space (Berthelot et al., 2018). Specifically, a critic network takes as input a synthetic image gener-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                157,
+                                505,
+                                171
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        157,
+                                        505,
+                                        171
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ated from a convex combination of the latents of two dataset samples, and regresses to the mixing",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                166,
+                                505,
+                                182
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        166,
+                                        505,
+                                        182
+                                    ],
+                                    "score": 1.0,
+                                    "content": "factor. The decoder of the autoencoder and the generator for the critic are one and the same. The reg-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                179,
+                                505,
+                                192
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        179,
+                                        505,
+                                        192
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ularization term is minimized when the autoencoder fools the critic into predicting that the synthetic",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                190,
+                                200,
+                                203
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        190,
+                                        200,
+                                        203
+                                    ],
+                                    "score": 1.0,
+                                    "content": "image is a real sample.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        }
+                    ],
+                    "index": 5,
+                    "bbox_fs": [
+                        105,
+                        124,
+                        506,
+                        203
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        207,
+                        504,
+                        251
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                207,
+                                506,
+                                219
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        207,
+                                        506,
+                                        219
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The bidirectional GAN (BiGAN) is an instance of a generative-adversarial framework in which",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                218,
+                                506,
+                                231
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        218,
+                                        506,
+                                        231
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the generator produces both synthetic image and embedding from real embedding and image, re-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                228,
+                                506,
+                                243
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        228,
+                                        506,
+                                        243
+                                    ],
+                                    "score": 1.0,
+                                    "content": "spectively (Donahue et al., 2017; Dumoulin et al., 2017). Discrimination is done in joint image-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                239,
+                                181,
+                                254
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        239,
+                                        181,
+                                        254
+                                    ],
+                                    "score": 1.0,
+                                    "content": "embedding space.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        }
+                    ],
+                    "index": 10.5,
+                    "bbox_fs": [
+                        105,
+                        207,
+                        506,
+                        254
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        257,
+                        504,
+                        290
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                257,
+                                504,
+                                269
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        257,
+                                        504,
+                                        269
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The DeepCluster method does discriminative clustering by alternating between clustering the fea-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                269,
+                                505,
+                                280
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        269,
+                                        505,
+                                        280
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tures of a convolutional neural network and using the clusters as labels to optimize the network",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                279,
+                                422,
+                                291
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        279,
+                                        422,
+                                        291
+                                    ],
+                                    "score": 1.0,
+                                    "content": "weights via backpropagating a standard classification loss (Caron et al., 2018).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        }
+                    ],
+                    "index": 14,
+                    "bbox_fs": [
+                        106,
+                        257,
+                        505,
+                        291
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        296,
+                        505,
+                        362
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                295,
+                                505,
+                                308
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        295,
+                                        505,
+                                        308
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The InfoGAN framework conceptually decomposes the generator’s input into a latent code and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                307,
+                                505,
+                                318
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        307,
+                                        505,
+                                        318
+                                    ],
+                                    "score": 1.0,
+                                    "content": "incompressible noise (Chen et al., 2016). The structure of the latent code is hand-specified based",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                318,
+                                505,
+                                329
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        318,
+                                        505,
+                                        329
+                                    ],
+                                    "score": 1.0,
+                                    "content": "on knowledge of the dataset. The canonical GAN minimax objective is regularized with a mutual",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                327,
+                                506,
+                                343
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        327,
+                                        506,
+                                        343
+                                    ],
+                                    "score": 1.0,
+                                    "content": "information term between the code and the generated image. In practice, this term is optimized using",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                339,
+                                505,
+                                352
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        339,
+                                        505,
+                                        352
+                                    ],
+                                    "score": 1.0,
+                                    "content": "variational inference, involving the approximation of the posterior with an auxiliary distribution",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                351,
+                                334,
+                                363
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        351,
+                                        115,
+                                        363
+                                    ],
+                                    "score": 0.51,
+                                    "content": "Q",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        116,
+                                        351,
+                                        334,
+                                        363
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(code|image) parameterized by a recognition network.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        }
+                    ],
+                    "index": 18.5,
+                    "bbox_fs": [
+                        105,
+                        295,
+                        506,
+                        363
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        367,
+                        505,
+                        444
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                367,
+                                505,
+                                380
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        367,
+                                        505,
+                                        380
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Whereas ACAI explicitly optimizes pixel-wise reconstruction error, BiGAN only encourages the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                378,
+                                505,
+                                392
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        378,
+                                        505,
+                                        392
+                                    ],
+                                    "score": 1.0,
+                                    "content": "fidelity of generated image and latent samples with respect to their respective prior distributions.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                388,
+                                505,
+                                403
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        388,
+                                        505,
+                                        403
+                                    ],
+                                    "score": 1.0,
+                                    "content": "While InfoGAN also encourages the fidelity of generated images, it leverages domain-specific",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                400,
+                                505,
+                                412
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        400,
+                                        505,
+                                        412
+                                    ],
+                                    "score": 1.0,
+                                    "content": "knowledge to impose a favorable structure on the embedding space and information-theoretic meth-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                412,
+                                505,
+                                424
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        412,
+                                        505,
+                                        424
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ods for optimization. DeepCluster departs from the aforementioned methods in that it is not con-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                422,
+                                505,
+                                436
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        422,
+                                        505,
+                                        436
+                                    ],
+                                    "score": 1.0,
+                                    "content": "cerned with generation or decoding, and only seeks to learn general-purpose visual features by way",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                432,
+                                266,
+                                447
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        432,
+                                        266,
+                                        447
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of end-to-end discriminative clustering.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        }
+                    ],
+                    "index": 25,
+                    "bbox_fs": [
+                        105,
+                        367,
+                        505,
+                        447
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        463,
+                        308,
+                        476
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                462,
+                                310,
+                                478
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        462,
+                                        310,
+                                        478
+                                    ],
+                                    "score": 1.0,
+                                    "content": "APPENDIX B DATASET INFORMATION",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        }
+                    ],
+                    "index": 29
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        489,
+                        505,
+                        567
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                488,
+                                506,
+                                503
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        488,
+                                        506,
+                                        503
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The Omniglot dataset consists of 1623 characters each with 20 hand-drawn examples. Ignoring",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                501,
+                                505,
+                                513
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        501,
+                                        505,
+                                        513
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the alphabets from which the characters originate, we use 1100, 100, and 423 characters for our",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                511,
+                                505,
+                                523
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        511,
+                                        505,
+                                        523
+                                    ],
+                                    "score": 1.0,
+                                    "content": "meta-training, meta-validation, and meta-testing splits. The miniImageNet dataset consists of 100",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                522,
+                                506,
+                                535
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        522,
+                                        506,
+                                        535
+                                    ],
+                                    "score": 1.0,
+                                    "content": "classes each with 600 examples. The images are predominantly natural and realistic. We use the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                533,
+                                505,
+                                546
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        533,
+                                        505,
+                                        546
+                                    ],
+                                    "score": 1.0,
+                                    "content": "same meta-training/meta-validation/meta-testing splits of 64/16/20 classes as proposed by Ravi &",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                543,
+                                505,
+                                558
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        543,
+                                        505,
+                                        558
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Larochelle (2017). The CelebA dataset includes 202,599 facial images of celebrities and 40 binary",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                555,
+                                492,
+                                568
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        555,
+                                        492,
+                                        568
+                                    ],
+                                    "score": 1.0,
+                                    "content": "attributes that annotate every image. We follow the prescribed 162,770/19,867/19,962 data split.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        }
+                    ],
+                    "index": 33,
+                    "bbox_fs": [
+                        105,
+                        488,
+                        506,
+                        568
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        572,
+                        504,
+                        616
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                571,
+                                504,
+                                585
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        571,
+                                        504,
+                                        585
+                                    ],
+                                    "score": 1.0,
+                                    "content": "For Omniglot and miniImageNet, supervised meta-learning tasks and evaluation tasks are con-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                582,
+                                506,
+                                596
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        582,
+                                        314,
+                                        596
+                                    ],
+                                    "score": 1.0,
+                                    "content": "structed exactly as detailed in Section 2.2: for an",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        315,
+                                        583,
+                                        325,
+                                        593
+                                    ],
+                                    "score": 0.83,
+                                    "content": "N",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        325,
+                                        582,
+                                        347,
+                                        596
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-way",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        347,
+                                        583,
+                                        357,
+                                        593
+                                    ],
+                                    "score": 0.82,
+                                    "content": "K",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        358,
+                                        582,
+                                        420,
+                                        596
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-shot task with",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        420,
+                                        583,
+                                        430,
+                                        594
+                                    ],
+                                    "score": 0.85,
+                                    "content": "Q",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        430,
+                                        582,
+                                        506,
+                                        596
+                                    ],
+                                    "score": 1.0,
+                                    "content": "queries per class,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                594,
+                                505,
+                                606
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        595,
+                                        152,
+                                        606
+                                    ],
+                                    "score": 1.0,
+                                    "content": "we sample",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        153,
+                                        594,
+                                        163,
+                                        604
+                                    ],
+                                    "score": 0.77,
+                                    "content": "N",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        164,
+                                        595,
+                                        291,
+                                        606
+                                    ],
+                                    "score": 1.0,
+                                    "content": "classes from the data split and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        291,
+                                        594,
+                                        322,
+                                        606
+                                    ],
+                                    "score": 0.92,
+                                    "content": "K + Q",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        323,
+                                        595,
+                                        505,
+                                        606
+                                    ],
+                                    "score": 1.0,
+                                    "content": "datapoints per class, labeling the task’s data",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                605,
+                                307,
+                                617
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        605,
+                                        228,
+                                        617
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with a random permutation of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        228,
+                                        605,
+                                        238,
+                                        615
+                                    ],
+                                    "score": 0.81,
+                                    "content": "N",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        239,
+                                        605,
+                                        307,
+                                        617
+                                    ],
+                                    "score": 1.0,
+                                    "content": "one-hot vectors.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        }
+                    ],
+                    "index": 38.5,
+                    "bbox_fs": [
+                        105,
+                        571,
+                        506,
+                        617
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        622,
+                        504,
+                        731
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                621,
+                                506,
+                                634
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        621,
+                                        506,
+                                        634
+                                    ],
+                                    "score": 1.0,
+                                    "content": "For CelebA, we consider binary classification tasks (i.e., 2-way), each defined by 3 attributes and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                632,
+                                505,
+                                645
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        632,
+                                        505,
+                                        645
+                                    ],
+                                    "score": 1.0,
+                                    "content": "an ordering of 3 Booleans, one for each attribute. Every image in a task-specific class shares",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                643,
+                                506,
+                                656
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        643,
+                                        506,
+                                        656
+                                    ],
+                                    "score": 1.0,
+                                    "content": "all task-specific attributes with each other and none with images in the other class. For exam-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                655,
+                                506,
+                                667
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        655,
+                                        506,
+                                        667
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ple, the task illustrated in Figure 2 involves distinguishing between images whose subjects sat-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                666,
+                                505,
+                                678
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        666,
+                                        505,
+                                        678
+                                    ],
+                                    "score": 1.0,
+                                    "content": "isfy not Sideburns, Straight Hair, and not Young, and those whose subjects satisfy",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                677,
+                                506,
+                                689
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        677,
+                                        506,
+                                        689
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Sideburns, not Straight Hair, and Young. To keep with the idea of having distinct",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                687,
+                                506,
+                                700
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        687,
+                                        506,
+                                        700
+                                    ],
+                                    "score": 1.0,
+                                    "content": "classes for meta-training and meta-testing, we split the task-defining attributes. For the supervised",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 47
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                699,
+                                506,
+                                711
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        699,
+                                        506,
+                                        711
+                                    ],
+                                    "score": 1.0,
+                                    "content": "meta-learning oracle, we construct meta-training tasks from the first 20 attributes (when alphabet-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 48
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                709,
+                                505,
+                                722
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        709,
+                                        505,
+                                        722
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ically ordered), meta-validation tasks from the next 10, and meta-testing tasks from the last 10.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 49
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                720,
+                                506,
+                                733
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        720,
+                                        506,
+                                        733
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Discarding tasks with too few examples in either class, this results in 4287, 391, and 402 task",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 50
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                82,
+                                506,
+                                95
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        82,
+                                        506,
+                                        95
+                                    ],
+                                    "score": 1.0,
+                                    "content": "prototypes (but many more possible tasks). We use the same meta-test time tasks to evaluate the",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                94,
+                                505,
+                                106
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        94,
+                                        505,
+                                        106
+                                    ],
+                                    "score": 1.0,
+                                    "content": "unsupervised methods. We only consider assessment with 5-shot tasks because, given that there are",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                104,
+                                482,
+                                118
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        104,
+                                        482,
+                                        118
+                                    ],
+                                    "score": 1.0,
+                                    "content": "multiple attributes other than the task-defining ones, any 1-shot task is likely to be ill-defined.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 2
+                        }
+                    ],
+                    "index": 45.5,
+                    "bbox_fs": [
+                        105,
+                        621,
+                        506,
+                        733
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        82,
+                        505,
+                        116
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                82,
+                                506,
+                                95
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        82,
+                                        506,
+                                        95
+                                    ],
+                                    "score": 1.0,
+                                    "content": "prototypes (but many more possible tasks). We use the same meta-test time tasks to evaluate the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                94,
+                                505,
+                                106
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        94,
+                                        505,
+                                        106
+                                    ],
+                                    "score": 1.0,
+                                    "content": "unsupervised methods. We only consider assessment with 5-shot tasks because, given that there are",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                104,
+                                482,
+                                118
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        104,
+                                        482,
+                                        118
+                                    ],
+                                    "score": 1.0,
+                                    "content": "multiple attributes other than the task-defining ones, any 1-shot task is likely to be ill-defined.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        }
+                    ],
+                    "index": 1
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        109,
+                        132,
+                        448,
+                        145
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                108,
+                                132,
+                                451,
+                                147
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        108,
+                                        132,
+                                        451,
+                                        147
+                                    ],
+                                    "score": 1.0,
+                                    "content": "APPENDIX C TASK CONSTRUCTION VIA RANDOM HYPERPLANES",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        }
+                    ],
+                    "index": 3
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        157,
+                        505,
+                        235
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                156,
+                                504,
+                                172
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        156,
+                                        239,
+                                        172
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Given a set of embedding points",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        239,
+                                        158,
+                                        259,
+                                        170
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\left\\{ \\mathbf { z } _ { i } \\right\\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        259,
+                                        156,
+                                        302,
+                                        172
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in a space",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        303,
+                                        158,
+                                        312,
+                                        168
+                                    ],
+                                    "score": 0.77,
+                                    "content": "\\mathcal { Z }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        312,
+                                        156,
+                                        461,
+                                        172
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", a simple way of defining a partition",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        462,
+                                        157,
+                                        504,
+                                        170
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\mathcal { P } = \\{ \\mathcal { C } _ { c } \\}",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                168,
+                                506,
+                                182
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        168,
+                                        119,
+                                        182
+                                    ],
+                                    "score": 1.0,
+                                    "content": "on",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        120,
+                                        169,
+                                        139,
+                                        181
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\left\\{ \\mathbf { z } _ { i } \\right\\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        139,
+                                        168,
+                                        292,
+                                        182
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is to use random hyperplanes to slice",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        292,
+                                        169,
+                                        302,
+                                        179
+                                    ],
+                                    "score": 0.8,
+                                    "content": "\\mathcal { Z }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        302,
+                                        168,
+                                        506,
+                                        182
+                                    ],
+                                    "score": 1.0,
+                                    "content": "into subspaces and assign the embeddings that lie",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                180,
+                                505,
+                                192
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        180,
+                                        132,
+                                        192
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in the",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        132,
+                                        182,
+                                        138,
+                                        190
+                                    ],
+                                    "score": 0.71,
+                                    "content": "c",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        138,
+                                        180,
+                                        227,
+                                        192
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-th subspace to subset",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        227,
+                                        180,
+                                        237,
+                                        191
+                                    ],
+                                    "score": 0.87,
+                                    "content": "\\mathcal { C } _ { c }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        238,
+                                        180,
+                                        505,
+                                        192
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". However, a hyperplane slicing can group together two arbitrarily",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                190,
+                                506,
+                                204
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        190,
+                                        506,
+                                        204
+                                    ],
+                                    "score": 1.0,
+                                    "content": "far embeddings, or separate two arbitrarily close ones; given our assumption that good embedding",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                201,
+                                506,
+                                214
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        201,
+                                        506,
+                                        214
+                                    ],
+                                    "score": 1.0,
+                                    "content": "spaces have a semantically meaningful metric, this creates ill-defined classes. This problem can be",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                213,
+                                505,
+                                226
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        213,
+                                        505,
+                                        226
+                                    ],
+                                    "score": 1.0,
+                                    "content": "partially alleviated by extending the hyperplane boundaries with a non-zero margin, as empirically",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                224,
+                                194,
+                                234
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        224,
+                                        194,
+                                        234
+                                    ],
+                                    "score": 1.0,
+                                    "content": "shown in Section 4.2.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        }
+                    ],
+                    "index": 7
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        241,
+                        505,
+                        387
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                241,
+                                505,
+                                253
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        241,
+                                        505,
+                                        253
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We now describe how to generate tasks via random hyperplanes in the embedding space. We first",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                251,
+                                505,
+                                265
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        251,
+                                        281,
+                                        265
+                                    ],
+                                    "score": 1.0,
+                                    "content": "describe a procedure to generate a partition",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        282,
+                                        252,
+                                        291,
+                                        262
+                                    ],
+                                    "score": 0.81,
+                                    "content": "\\mathcal { P }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        291,
+                                        251,
+                                        393,
+                                        265
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of the set of embeddings",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        393,
+                                        251,
+                                        412,
+                                        263
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\left\\{ \\mathbf { z } _ { i } \\right\\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        413,
+                                        251,
+                                        505,
+                                        265
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for constructing meta-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                262,
+                                505,
+                                276
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        262,
+                                        437,
+                                        276
+                                    ],
+                                    "score": 1.0,
+                                    "content": "training tasks. A given hyperplane slices the embedding space into two, so for an",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        438,
+                                        263,
+                                        448,
+                                        272
+                                    ],
+                                    "score": 0.82,
+                                    "content": "N",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        448,
+                                        262,
+                                        505,
+                                        276
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-way task, we",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                273,
+                                505,
+                                286
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        273,
+                                        127,
+                                        286
+                                    ],
+                                    "score": 1.0,
+                                    "content": "need",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        128,
+                                        273,
+                                        188,
+                                        285
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\bar { H } = \\lceil \\log _ { 2 } \\bar { N } \\rceil",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        188,
+                                        273,
+                                        505,
+                                        286
+                                    ],
+                                    "score": 1.0,
+                                    "content": "hyperplanes to define sufficiently many subsets/classes for a task. To randomly",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                284,
+                                506,
+                                298
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        284,
+                                        199,
+                                        298
+                                    ],
+                                    "score": 1.0,
+                                    "content": "define a hyperplane in",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        200,
+                                        285,
+                                        206,
+                                        294
+                                    ],
+                                    "score": 0.81,
+                                    "content": "d",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        207,
+                                        284,
+                                        447,
+                                        298
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-dimensional embedding space, we sample a normal vector",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        447,
+                                        286,
+                                        456,
+                                        295
+                                    ],
+                                    "score": 0.37,
+                                    "content": "\\mathbf { n }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        456,
+                                        284,
+                                        506,
+                                        298
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and a point",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                295,
+                                505,
+                                308
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        295,
+                                        158,
+                                        308
+                                    ],
+                                    "score": 1.0,
+                                    "content": "on the plane",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        159,
+                                        297,
+                                        170,
+                                        306
+                                    ],
+                                    "score": 0.85,
+                                    "content": "\\mathbf { z } _ { 0 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        170,
+                                        295,
+                                        216,
+                                        308
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", each with",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        216,
+                                        296,
+                                        223,
+                                        305
+                                    ],
+                                    "score": 0.78,
+                                    "content": "d",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        223,
+                                        295,
+                                        366,
+                                        308
+                                    ],
+                                    "score": 1.0,
+                                    "content": "elements. For an embedding point",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        367,
+                                        297,
+                                        373,
+                                        305
+                                    ],
+                                    "score": 0.57,
+                                    "content": "\\mathbf { z }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        373,
+                                        295,
+                                        505,
+                                        308
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", the signed point-plane distance",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                103,
+                                305,
+                                507,
+                                323
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        103,
+                                        305,
+                                        156,
+                                        323
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is given by",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        156,
+                                        307,
+                                        217,
+                                        321
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\frac { \\mathbf { n } } { \\left| \\mathbf { n } \\right| _ { 2 } } \\cdot \\left( \\mathbf { z } - \\mathbf { z } _ { 0 } \\right)",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        217,
+                                        305,
+                                        263,
+                                        323
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Defining",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        263,
+                                        307,
+                                        273,
+                                        317
+                                    ],
+                                    "score": 0.79,
+                                    "content": "H",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        274,
+                                        305,
+                                        507,
+                                        323
+                                    ],
+                                    "score": 1.0,
+                                    "content": "hyperplanes in this manner, we discard embeddings for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                319,
+                                504,
+                                334
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        319,
+                                        315,
+                                        334
+                                    ],
+                                    "score": 1.0,
+                                    "content": "which the signed point-plane distance to any of the",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        316,
+                                        321,
+                                        326,
+                                        330
+                                    ],
+                                    "score": 0.76,
+                                    "content": "H",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        326,
+                                        319,
+                                        424,
+                                        334
+                                    ],
+                                    "score": 1.0,
+                                    "content": "hyperplanes lies within",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        424,
+                                        320,
+                                        462,
+                                        333
+                                    ],
+                                    "score": 0.92,
+                                    "content": "( - m , m )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        462,
+                                        319,
+                                        493,
+                                        334
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", where",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        494,
+                                        322,
+                                        504,
+                                        330
+                                    ],
+                                    "score": 0.66,
+                                    "content": "m",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                103,
+                                329,
+                                506,
+                                346
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        103,
+                                        329,
+                                        210,
+                                        346
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is a desired margin. The",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        210,
+                                        331,
+                                        221,
+                                        342
+                                    ],
+                                    "score": 0.8,
+                                    "content": "H",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        221,
+                                        329,
+                                        350,
+                                        346
+                                    ],
+                                    "score": 1.0,
+                                    "content": "hyperplanes collectively define",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        351,
+                                        331,
+                                        364,
+                                        342
+                                    ],
+                                    "score": 0.87,
+                                    "content": "\\mathbf { \\bar { 2 } } ^ { H }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        365,
+                                        329,
+                                        506,
+                                        346
+                                    ],
+                                    "score": 1.0,
+                                    "content": "subspaces. We assign embedding",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                342,
+                                506,
+                                355
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        342,
+                                        158,
+                                        355
+                                    ],
+                                    "score": 1.0,
+                                    "content": "points in the",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        158,
+                                        344,
+                                        164,
+                                        352
+                                    ],
+                                    "score": 0.75,
+                                    "content": "c",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        164,
+                                        342,
+                                        251,
+                                        355
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-th subspace to subset",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        252,
+                                        343,
+                                        262,
+                                        353
+                                    ],
+                                    "score": 0.88,
+                                    "content": "\\mathcal { C } _ { c }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        262,
+                                        342,
+                                        370,
+                                        355
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". We define the partition as",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        370,
+                                        342,
+                                        411,
+                                        354
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\mathcal { P } = \\{ \\mathcal { C } _ { c } \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        412,
+                                        342,
+                                        506,
+                                        355
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". We prune subsets that",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                352,
+                                505,
+                                366
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        352,
+                                        186,
+                                        366
+                                    ],
+                                    "score": 1.0,
+                                    "content": "do not have at least",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        186,
+                                        353,
+                                        250,
+                                        365
+                                    ],
+                                    "score": 0.93,
+                                    "content": "R = K _ { \\mathrm { m - t r } } + Q",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        250,
+                                        352,
+                                        450,
+                                        366
+                                    ],
+                                    "score": 1.0,
+                                    "content": "members, and check that the partition has at least",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        451,
+                                        354,
+                                        461,
+                                        363
+                                    ],
+                                    "score": 0.82,
+                                    "content": "N",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        461,
+                                        352,
+                                        505,
+                                        366
+                                    ],
+                                    "score": 1.0,
+                                    "content": "remaining",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                364,
+                                505,
+                                378
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        364,
+                                        479,
+                                        378
+                                    ],
+                                    "score": 1.0,
+                                    "content": "subsets; if not, we reject the partition and restart the procedure. After obtaining partitions",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        479,
+                                        364,
+                                        501,
+                                        377
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\{ \\mathcal P _ { p } \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        501,
+                                        364,
+                                        505,
+                                        378
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                375,
+                                412,
+                                388
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        375,
+                                        412,
+                                        388
+                                    ],
+                                    "score": 1.0,
+                                    "content": "meta-training tasks can be generated by following Algorithm 1 from Line 4.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        }
+                    ],
+                    "index": 17
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        392,
+                        505,
+                        447
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                392,
+                                505,
+                                405
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        392,
+                                        505,
+                                        405
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In terms of practical implementation, we pre-compute 1000 hyperplanes and pruned pairs of subsets",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                403,
+                                505,
+                                416
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        403,
+                                        118,
+                                        416
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        118,
+                                        403,
+                                        137,
+                                        415
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\left\\{ \\mathbf { z } _ { i } \\right\\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        137,
+                                        403,
+                                        505,
+                                        416
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". We generate partitions by sampling combinations of the hyperplanes and taking intersec-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                414,
+                                505,
+                                426
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        414,
+                                        505,
+                                        426
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tions of their associated subsets to define the elements of the partition. We determine the number of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                425,
+                                505,
+                                437
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        425,
+                                        505,
+                                        437
+                                    ],
+                                    "score": 1.0,
+                                    "content": "partitions needed for a given Hyperplanes-MAML run by the number of meta-training tasks desired",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                435,
+                                312,
+                                448
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        435,
+                                        312,
+                                        448
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for the meta-learner: we fix 100 tasks per partition.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        }
+                    ],
+                    "index": 26
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        464,
+                        303,
+                        477
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                463,
+                                304,
+                                479
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        463,
+                                        304,
+                                        479
+                                    ],
+                                    "score": 1.0,
+                                    "content": "APPENDIX D MNIST EXPERIMENTS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        }
+                    ],
+                    "index": 29
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        489,
+                        505,
+                        567
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                488,
+                                505,
+                                502
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        488,
+                                        505,
+                                        502
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The MNIST dataset consists of 70,000 hand-drawn examples of the 10 numerical digits. Our split",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                501,
+                                505,
+                                513
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        501,
+                                        505,
+                                        513
+                                    ],
+                                    "score": 1.0,
+                                    "content": "respects the original MNIST 60,000/10,000 training/testing split. We assess on 10-way classification",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                511,
+                                506,
+                                524
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        511,
+                                        506,
+                                        524
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tasks. This setup results in examples from all 10 digits being present for both meta-training and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                521,
+                                505,
+                                537
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        521,
+                                        505,
+                                        537
+                                    ],
+                                    "score": 1.0,
+                                    "content": "meta-testing, making the probem setting essentially equivalent to that of semi-supervised learning",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                533,
+                                505,
+                                545
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        533,
+                                        505,
+                                        545
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sans a fixed permutation of the labels. The MNIST scenario is thus a special case of the problem",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                544,
+                                506,
+                                558
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        544,
+                                        506,
+                                        558
+                                    ],
+                                    "score": 1.0,
+                                    "content": "setting considered in the rest of the paper. For MNIST, we only experiment with MAML as the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                556,
+                                207,
+                                568
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        556,
+                                        207,
+                                        568
+                                    ],
+                                    "score": 1.0,
+                                    "content": "meta-learning algorithm.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        }
+                    ],
+                    "index": 33
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        572,
+                        505,
+                        682
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                572,
+                                505,
+                                584
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        572,
+                                        505,
+                                        584
+                                    ],
+                                    "score": 1.0,
+                                    "content": "For ACAI and InfoGAN we constructed the meta-validation split from the last 5,000 examples of the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                582,
+                                506,
+                                597
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        582,
+                                        506,
+                                        597
+                                    ],
+                                    "score": 1.0,
+                                    "content": "meta-training split; for BiGAN this figure was 10,000. After training the ACAI model and inferring",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                595,
+                                505,
+                                606
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        595,
+                                        505,
+                                        606
+                                    ],
+                                    "score": 1.0,
+                                    "content": "embeddings, manually assigning labels to 10 clusters by inspection results in a classification accu-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                605,
+                                505,
+                                617
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        605,
+                                        137,
+                                        617
+                                    ],
+                                    "score": 1.0,
+                                    "content": "racy of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        137,
+                                        605,
+                                        169,
+                                        615
+                                    ],
+                                    "score": 0.88,
+                                    "content": "9 6 . 0 0 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        170,
+                                        605,
+                                        505,
+                                        617
+                                    ],
+                                    "score": 1.0,
+                                    "content": "on the testing split. As the ACAI authors observe, we found it important to whiten",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                614,
+                                506,
+                                630
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        614,
+                                        506,
+                                        630
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the ACAI embeddings before clustering. The same metric for the InfoGAN embedding (taking an",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                627,
+                                506,
+                                640
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        627,
+                                        429,
+                                        640
+                                    ],
+                                    "score": 1.0,
+                                    "content": "argmax over the categorical dimensions instead of actually running clustering) is",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        429,
+                                        627,
+                                        461,
+                                        637
+                                    ],
+                                    "score": 0.88,
+                                    "content": "9 6 . 8 3 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        461,
+                                        627,
+                                        506,
+                                        640
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Note that",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                636,
+                                505,
+                                651
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        636,
+                                        505,
+                                        651
+                                    ],
+                                    "score": 1.0,
+                                    "content": "these results are an upper-bound for embedding cluster matching. To see this, consider the 10-way",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                648,
+                                505,
+                                661
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        648,
+                                        505,
+                                        661
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1-shot scenario. 1 example sampled from each cluster is insufficient to guarantee the optimal label",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                658,
+                                506,
+                                673
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        658,
+                                        506,
+                                        673
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for that cluster; 1 example sampled from each label is not guaranteed to each end up in the optimal",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                670,
+                                146,
+                                685
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        670,
+                                        146,
+                                        685
+                                    ],
+                                    "score": 1.0,
+                                    "content": "category.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        }
+                    ],
+                    "index": 41.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        687,
+                        504,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                687,
+                                506,
+                                700
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        687,
+                                        280,
+                                        700
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Aside from CACTUs-MAML, embedding",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        281,
+                                        688,
+                                        294,
+                                        699
+                                    ],
+                                    "score": 0.88,
+                                    "content": "k _ { \\mathrm { n n } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        295,
+                                        687,
+                                        506,
+                                        700
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-nearest neighbors, embedding linear classifier, and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 47
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                697,
+                                506,
+                                713
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        697,
+                                        506,
+                                        713
+                                    ],
+                                    "score": 1.0,
+                                    "content": "embedding direct clustering, we also ran CACTUs-MAML on embeddings instead of raw images,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 48
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                710,
+                                506,
+                                721
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        710,
+                                        506,
+                                        721
+                                    ],
+                                    "score": 1.0,
+                                    "content": "using a simple model with 2 hidden layers with 64 units each and ReLU activation, and all other",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 49
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                720,
+                                327,
+                                733
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        720,
+                                        327,
+                                        733
+                                    ],
+                                    "score": 1.0,
+                                    "content": "MAML hyperparameters being the same as in Table 5.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 50
+                        }
+                    ],
+                    "index": 48.5
+                }
+            ],
+            "page_idx": 15,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        108,
+                        27,
+                        293,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                26,
+                                294,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        26,
+                                        294,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Published as a conference paper at ICLR 2019",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        301,
+                        752,
+                        311,
+                        760
+                    ],
+                    "lines": []
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        82,
+                        505,
+                        116
+                    ],
+                    "lines": [],
+                    "index": 1,
+                    "bbox_fs": [
+                        105,
+                        82,
+                        506,
+                        118
+                    ],
+                    "lines_deleted": true
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        109,
+                        132,
+                        448,
+                        145
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                108,
+                                132,
+                                451,
+                                147
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        108,
+                                        132,
+                                        451,
+                                        147
+                                    ],
+                                    "score": 1.0,
+                                    "content": "APPENDIX C TASK CONSTRUCTION VIA RANDOM HYPERPLANES",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        }
+                    ],
+                    "index": 3
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        157,
+                        505,
+                        235
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                156,
+                                504,
+                                172
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        156,
+                                        239,
+                                        172
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Given a set of embedding points",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        239,
+                                        158,
+                                        259,
+                                        170
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\left\\{ \\mathbf { z } _ { i } \\right\\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        259,
+                                        156,
+                                        302,
+                                        172
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in a space",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        303,
+                                        158,
+                                        312,
+                                        168
+                                    ],
+                                    "score": 0.77,
+                                    "content": "\\mathcal { Z }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        312,
+                                        156,
+                                        461,
+                                        172
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", a simple way of defining a partition",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        462,
+                                        157,
+                                        504,
+                                        170
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\mathcal { P } = \\{ \\mathcal { C } _ { c } \\}",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                168,
+                                506,
+                                182
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        168,
+                                        119,
+                                        182
+                                    ],
+                                    "score": 1.0,
+                                    "content": "on",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        120,
+                                        169,
+                                        139,
+                                        181
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\left\\{ \\mathbf { z } _ { i } \\right\\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        139,
+                                        168,
+                                        292,
+                                        182
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is to use random hyperplanes to slice",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        292,
+                                        169,
+                                        302,
+                                        179
+                                    ],
+                                    "score": 0.8,
+                                    "content": "\\mathcal { Z }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        302,
+                                        168,
+                                        506,
+                                        182
+                                    ],
+                                    "score": 1.0,
+                                    "content": "into subspaces and assign the embeddings that lie",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                180,
+                                505,
+                                192
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        180,
+                                        132,
+                                        192
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in the",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        132,
+                                        182,
+                                        138,
+                                        190
+                                    ],
+                                    "score": 0.71,
+                                    "content": "c",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        138,
+                                        180,
+                                        227,
+                                        192
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-th subspace to subset",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        227,
+                                        180,
+                                        237,
+                                        191
+                                    ],
+                                    "score": 0.87,
+                                    "content": "\\mathcal { C } _ { c }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        238,
+                                        180,
+                                        505,
+                                        192
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". However, a hyperplane slicing can group together two arbitrarily",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                190,
+                                506,
+                                204
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        190,
+                                        506,
+                                        204
+                                    ],
+                                    "score": 1.0,
+                                    "content": "far embeddings, or separate two arbitrarily close ones; given our assumption that good embedding",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                201,
+                                506,
+                                214
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        201,
+                                        506,
+                                        214
+                                    ],
+                                    "score": 1.0,
+                                    "content": "spaces have a semantically meaningful metric, this creates ill-defined classes. This problem can be",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                213,
+                                505,
+                                226
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        213,
+                                        505,
+                                        226
+                                    ],
+                                    "score": 1.0,
+                                    "content": "partially alleviated by extending the hyperplane boundaries with a non-zero margin, as empirically",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                224,
+                                194,
+                                234
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        224,
+                                        194,
+                                        234
+                                    ],
+                                    "score": 1.0,
+                                    "content": "shown in Section 4.2.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        }
+                    ],
+                    "index": 7,
+                    "bbox_fs": [
+                        105,
+                        156,
+                        506,
+                        234
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        241,
+                        505,
+                        387
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                241,
+                                505,
+                                253
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        241,
+                                        505,
+                                        253
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We now describe how to generate tasks via random hyperplanes in the embedding space. We first",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                251,
+                                505,
+                                265
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        251,
+                                        281,
+                                        265
+                                    ],
+                                    "score": 1.0,
+                                    "content": "describe a procedure to generate a partition",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        282,
+                                        252,
+                                        291,
+                                        262
+                                    ],
+                                    "score": 0.81,
+                                    "content": "\\mathcal { P }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        291,
+                                        251,
+                                        393,
+                                        265
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of the set of embeddings",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        393,
+                                        251,
+                                        412,
+                                        263
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\left\\{ \\mathbf { z } _ { i } \\right\\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        413,
+                                        251,
+                                        505,
+                                        265
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for constructing meta-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                262,
+                                505,
+                                276
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        262,
+                                        437,
+                                        276
+                                    ],
+                                    "score": 1.0,
+                                    "content": "training tasks. A given hyperplane slices the embedding space into two, so for an",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        438,
+                                        263,
+                                        448,
+                                        272
+                                    ],
+                                    "score": 0.82,
+                                    "content": "N",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        448,
+                                        262,
+                                        505,
+                                        276
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-way task, we",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                273,
+                                505,
+                                286
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        273,
+                                        127,
+                                        286
+                                    ],
+                                    "score": 1.0,
+                                    "content": "need",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        128,
+                                        273,
+                                        188,
+                                        285
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\bar { H } = \\lceil \\log _ { 2 } \\bar { N } \\rceil",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        188,
+                                        273,
+                                        505,
+                                        286
+                                    ],
+                                    "score": 1.0,
+                                    "content": "hyperplanes to define sufficiently many subsets/classes for a task. To randomly",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                284,
+                                506,
+                                298
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        284,
+                                        199,
+                                        298
+                                    ],
+                                    "score": 1.0,
+                                    "content": "define a hyperplane in",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        200,
+                                        285,
+                                        206,
+                                        294
+                                    ],
+                                    "score": 0.81,
+                                    "content": "d",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        207,
+                                        284,
+                                        447,
+                                        298
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-dimensional embedding space, we sample a normal vector",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        447,
+                                        286,
+                                        456,
+                                        295
+                                    ],
+                                    "score": 0.37,
+                                    "content": "\\mathbf { n }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        456,
+                                        284,
+                                        506,
+                                        298
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and a point",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                295,
+                                505,
+                                308
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        295,
+                                        158,
+                                        308
+                                    ],
+                                    "score": 1.0,
+                                    "content": "on the plane",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        159,
+                                        297,
+                                        170,
+                                        306
+                                    ],
+                                    "score": 0.85,
+                                    "content": "\\mathbf { z } _ { 0 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        170,
+                                        295,
+                                        216,
+                                        308
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", each with",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        216,
+                                        296,
+                                        223,
+                                        305
+                                    ],
+                                    "score": 0.78,
+                                    "content": "d",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        223,
+                                        295,
+                                        366,
+                                        308
+                                    ],
+                                    "score": 1.0,
+                                    "content": "elements. For an embedding point",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        367,
+                                        297,
+                                        373,
+                                        305
+                                    ],
+                                    "score": 0.57,
+                                    "content": "\\mathbf { z }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        373,
+                                        295,
+                                        505,
+                                        308
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", the signed point-plane distance",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                103,
+                                305,
+                                507,
+                                323
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        103,
+                                        305,
+                                        156,
+                                        323
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is given by",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        156,
+                                        307,
+                                        217,
+                                        321
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\frac { \\mathbf { n } } { \\left| \\mathbf { n } \\right| _ { 2 } } \\cdot \\left( \\mathbf { z } - \\mathbf { z } _ { 0 } \\right)",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        217,
+                                        305,
+                                        263,
+                                        323
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Defining",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        263,
+                                        307,
+                                        273,
+                                        317
+                                    ],
+                                    "score": 0.79,
+                                    "content": "H",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        274,
+                                        305,
+                                        507,
+                                        323
+                                    ],
+                                    "score": 1.0,
+                                    "content": "hyperplanes in this manner, we discard embeddings for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                319,
+                                504,
+                                334
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        319,
+                                        315,
+                                        334
+                                    ],
+                                    "score": 1.0,
+                                    "content": "which the signed point-plane distance to any of the",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        316,
+                                        321,
+                                        326,
+                                        330
+                                    ],
+                                    "score": 0.76,
+                                    "content": "H",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        326,
+                                        319,
+                                        424,
+                                        334
+                                    ],
+                                    "score": 1.0,
+                                    "content": "hyperplanes lies within",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        424,
+                                        320,
+                                        462,
+                                        333
+                                    ],
+                                    "score": 0.92,
+                                    "content": "( - m , m )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        462,
+                                        319,
+                                        493,
+                                        334
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", where",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        494,
+                                        322,
+                                        504,
+                                        330
+                                    ],
+                                    "score": 0.66,
+                                    "content": "m",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                103,
+                                329,
+                                506,
+                                346
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        103,
+                                        329,
+                                        210,
+                                        346
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is a desired margin. The",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        210,
+                                        331,
+                                        221,
+                                        342
+                                    ],
+                                    "score": 0.8,
+                                    "content": "H",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        221,
+                                        329,
+                                        350,
+                                        346
+                                    ],
+                                    "score": 1.0,
+                                    "content": "hyperplanes collectively define",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        351,
+                                        331,
+                                        364,
+                                        342
+                                    ],
+                                    "score": 0.87,
+                                    "content": "\\mathbf { \\bar { 2 } } ^ { H }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        365,
+                                        329,
+                                        506,
+                                        346
+                                    ],
+                                    "score": 1.0,
+                                    "content": "subspaces. We assign embedding",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                342,
+                                506,
+                                355
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        342,
+                                        158,
+                                        355
+                                    ],
+                                    "score": 1.0,
+                                    "content": "points in the",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        158,
+                                        344,
+                                        164,
+                                        352
+                                    ],
+                                    "score": 0.75,
+                                    "content": "c",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        164,
+                                        342,
+                                        251,
+                                        355
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-th subspace to subset",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        252,
+                                        343,
+                                        262,
+                                        353
+                                    ],
+                                    "score": 0.88,
+                                    "content": "\\mathcal { C } _ { c }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        262,
+                                        342,
+                                        370,
+                                        355
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". We define the partition as",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        370,
+                                        342,
+                                        411,
+                                        354
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\mathcal { P } = \\{ \\mathcal { C } _ { c } \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        412,
+                                        342,
+                                        506,
+                                        355
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". We prune subsets that",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                352,
+                                505,
+                                366
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        352,
+                                        186,
+                                        366
+                                    ],
+                                    "score": 1.0,
+                                    "content": "do not have at least",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        186,
+                                        353,
+                                        250,
+                                        365
+                                    ],
+                                    "score": 0.93,
+                                    "content": "R = K _ { \\mathrm { m - t r } } + Q",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        250,
+                                        352,
+                                        450,
+                                        366
+                                    ],
+                                    "score": 1.0,
+                                    "content": "members, and check that the partition has at least",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        451,
+                                        354,
+                                        461,
+                                        363
+                                    ],
+                                    "score": 0.82,
+                                    "content": "N",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        461,
+                                        352,
+                                        505,
+                                        366
+                                    ],
+                                    "score": 1.0,
+                                    "content": "remaining",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                364,
+                                505,
+                                378
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        364,
+                                        479,
+                                        378
+                                    ],
+                                    "score": 1.0,
+                                    "content": "subsets; if not, we reject the partition and restart the procedure. After obtaining partitions",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        479,
+                                        364,
+                                        501,
+                                        377
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\{ \\mathcal P _ { p } \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        501,
+                                        364,
+                                        505,
+                                        378
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                375,
+                                412,
+                                388
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        375,
+                                        412,
+                                        388
+                                    ],
+                                    "score": 1.0,
+                                    "content": "meta-training tasks can be generated by following Algorithm 1 from Line 4.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        }
+                    ],
+                    "index": 17,
+                    "bbox_fs": [
+                        103,
+                        241,
+                        507,
+                        388
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        392,
+                        505,
+                        447
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                392,
+                                505,
+                                405
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        392,
+                                        505,
+                                        405
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In terms of practical implementation, we pre-compute 1000 hyperplanes and pruned pairs of subsets",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                403,
+                                505,
+                                416
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        403,
+                                        118,
+                                        416
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        118,
+                                        403,
+                                        137,
+                                        415
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\left\\{ \\mathbf { z } _ { i } \\right\\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        137,
+                                        403,
+                                        505,
+                                        416
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". We generate partitions by sampling combinations of the hyperplanes and taking intersec-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                414,
+                                505,
+                                426
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        414,
+                                        505,
+                                        426
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tions of their associated subsets to define the elements of the partition. We determine the number of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                425,
+                                505,
+                                437
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        425,
+                                        505,
+                                        437
+                                    ],
+                                    "score": 1.0,
+                                    "content": "partitions needed for a given Hyperplanes-MAML run by the number of meta-training tasks desired",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                435,
+                                312,
+                                448
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        435,
+                                        312,
+                                        448
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for the meta-learner: we fix 100 tasks per partition.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        }
+                    ],
+                    "index": 26,
+                    "bbox_fs": [
+                        105,
+                        392,
+                        505,
+                        448
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        464,
+                        303,
+                        477
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                463,
+                                304,
+                                479
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        463,
+                                        304,
+                                        479
+                                    ],
+                                    "score": 1.0,
+                                    "content": "APPENDIX D MNIST EXPERIMENTS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        }
+                    ],
+                    "index": 29
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        489,
+                        505,
+                        567
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                488,
+                                505,
+                                502
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        488,
+                                        505,
+                                        502
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The MNIST dataset consists of 70,000 hand-drawn examples of the 10 numerical digits. Our split",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                501,
+                                505,
+                                513
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        501,
+                                        505,
+                                        513
+                                    ],
+                                    "score": 1.0,
+                                    "content": "respects the original MNIST 60,000/10,000 training/testing split. We assess on 10-way classification",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                511,
+                                506,
+                                524
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        511,
+                                        506,
+                                        524
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tasks. This setup results in examples from all 10 digits being present for both meta-training and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                521,
+                                505,
+                                537
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        521,
+                                        505,
+                                        537
+                                    ],
+                                    "score": 1.0,
+                                    "content": "meta-testing, making the probem setting essentially equivalent to that of semi-supervised learning",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                533,
+                                505,
+                                545
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        533,
+                                        505,
+                                        545
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sans a fixed permutation of the labels. The MNIST scenario is thus a special case of the problem",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                544,
+                                506,
+                                558
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        544,
+                                        506,
+                                        558
+                                    ],
+                                    "score": 1.0,
+                                    "content": "setting considered in the rest of the paper. For MNIST, we only experiment with MAML as the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                556,
+                                207,
+                                568
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        556,
+                                        207,
+                                        568
+                                    ],
+                                    "score": 1.0,
+                                    "content": "meta-learning algorithm.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        }
+                    ],
+                    "index": 33,
+                    "bbox_fs": [
+                        105,
+                        488,
+                        506,
+                        568
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        572,
+                        505,
+                        682
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                572,
+                                505,
+                                584
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        572,
+                                        505,
+                                        584
+                                    ],
+                                    "score": 1.0,
+                                    "content": "For ACAI and InfoGAN we constructed the meta-validation split from the last 5,000 examples of the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                582,
+                                506,
+                                597
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        582,
+                                        506,
+                                        597
+                                    ],
+                                    "score": 1.0,
+                                    "content": "meta-training split; for BiGAN this figure was 10,000. After training the ACAI model and inferring",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                595,
+                                505,
+                                606
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        595,
+                                        505,
+                                        606
+                                    ],
+                                    "score": 1.0,
+                                    "content": "embeddings, manually assigning labels to 10 clusters by inspection results in a classification accu-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                605,
+                                505,
+                                617
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        605,
+                                        137,
+                                        617
+                                    ],
+                                    "score": 1.0,
+                                    "content": "racy of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        137,
+                                        605,
+                                        169,
+                                        615
+                                    ],
+                                    "score": 0.88,
+                                    "content": "9 6 . 0 0 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        170,
+                                        605,
+                                        505,
+                                        617
+                                    ],
+                                    "score": 1.0,
+                                    "content": "on the testing split. As the ACAI authors observe, we found it important to whiten",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                614,
+                                506,
+                                630
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        614,
+                                        506,
+                                        630
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the ACAI embeddings before clustering. The same metric for the InfoGAN embedding (taking an",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                627,
+                                506,
+                                640
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        627,
+                                        429,
+                                        640
+                                    ],
+                                    "score": 1.0,
+                                    "content": "argmax over the categorical dimensions instead of actually running clustering) is",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        429,
+                                        627,
+                                        461,
+                                        637
+                                    ],
+                                    "score": 0.88,
+                                    "content": "9 6 . 8 3 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        461,
+                                        627,
+                                        506,
+                                        640
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Note that",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                636,
+                                505,
+                                651
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        636,
+                                        505,
+                                        651
+                                    ],
+                                    "score": 1.0,
+                                    "content": "these results are an upper-bound for embedding cluster matching. To see this, consider the 10-way",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                648,
+                                505,
+                                661
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        648,
+                                        505,
+                                        661
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1-shot scenario. 1 example sampled from each cluster is insufficient to guarantee the optimal label",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                658,
+                                506,
+                                673
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        658,
+                                        506,
+                                        673
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for that cluster; 1 example sampled from each label is not guaranteed to each end up in the optimal",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                670,
+                                146,
+                                685
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        670,
+                                        146,
+                                        685
+                                    ],
+                                    "score": 1.0,
+                                    "content": "category.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        }
+                    ],
+                    "index": 41.5,
+                    "bbox_fs": [
+                        104,
+                        572,
+                        506,
+                        685
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        687,
+                        504,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                687,
+                                506,
+                                700
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        687,
+                                        280,
+                                        700
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Aside from CACTUs-MAML, embedding",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        281,
+                                        688,
+                                        294,
+                                        699
+                                    ],
+                                    "score": 0.88,
+                                    "content": "k _ { \\mathrm { n n } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        295,
+                                        687,
+                                        506,
+                                        700
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-nearest neighbors, embedding linear classifier, and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 47
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                697,
+                                506,
+                                713
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        697,
+                                        506,
+                                        713
+                                    ],
+                                    "score": 1.0,
+                                    "content": "embedding direct clustering, we also ran CACTUs-MAML on embeddings instead of raw images,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 48
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                710,
+                                506,
+                                721
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        710,
+                                        506,
+                                        721
+                                    ],
+                                    "score": 1.0,
+                                    "content": "using a simple model with 2 hidden layers with 64 units each and ReLU activation, and all other",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 49
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                720,
+                                327,
+                                733
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        720,
+                                        327,
+                                        733
+                                    ],
+                                    "score": 1.0,
+                                    "content": "MAML hyperparameters being the same as in Table 5.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 50
+                        }
+                    ],
+                    "index": 48.5,
+                    "bbox_fs": [
+                        105,
+                        687,
+                        506,
+                        733
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        82,
+                        505,
+                        138
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                82,
+                                505,
+                                94
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        82,
+                                        209,
+                                        94
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Departing from the fixed",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        209,
+                                        83,
+                                        246,
+                                        93
+                                    ],
+                                    "score": 0.92,
+                                    "content": "k = 5 0 0",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        246,
+                                        82,
+                                        436,
+                                        94
+                                    ],
+                                    "score": 1.0,
+                                    "content": "used for all other datasets, we deliberately use",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        436,
+                                        83,
+                                        468,
+                                        93
+                                    ],
+                                    "score": 0.91,
+                                    "content": "k = 1 0",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        468,
+                                        82,
+                                        505,
+                                        94
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to better",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                93,
+                                505,
+                                106
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        93,
+                                        505,
+                                        106
+                                    ],
+                                    "score": 1.0,
+                                    "content": "understand the limitations of CACTUs-MAML. The results can be seen in Table 7 in Appendix B.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                105,
+                                505,
+                                117
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        105,
+                                        505,
+                                        117
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In brief, with the better embeddings (ACAI and InfoGAN), there is only little benefit of CACTUs-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                115,
+                                505,
+                                128
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        115,
+                                        505,
+                                        128
+                                    ],
+                                    "score": 1.0,
+                                    "content": "MAML over embedding cluster matching. Additionally, even in the best cases, CACTUs-MAML",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                125,
+                                362,
+                                139
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        125,
+                                        362,
+                                        139
+                                    ],
+                                    "score": 1.0,
+                                    "content": "falls short of state-of-the-art semi-supervised learning methods.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        }
+                    ],
+                    "index": 2
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        109,
+                        154,
+                        405,
+                        167
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                107,
+                                153,
+                                406,
+                                168
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        153,
+                                        406,
+                                        168
+                                    ],
+                                    "score": 1.0,
+                                    "content": "APPENDIX E HYPERPARAMETERS AND ARCHITECTURES",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        }
+                    ],
+                    "index": 5
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        106,
+                        178,
+                        165,
+                        190
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                178,
+                                167,
+                                191
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        178,
+                                        167,
+                                        191
+                                    ],
+                                    "score": 1.0,
+                                    "content": "E.1 MAML",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        }
+                    ],
+                    "index": 6
+                },
+                {
+                    "type": "table",
+                    "bbox": [
+                        116,
+                        227,
+                        496,
+                        361
+                    ],
+                    "blocks": [
+                        {
+                            "type": "table_caption",
+                            "bbox": [
+                                227,
+                                202,
+                                383,
+                                214
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        226,
+                                        201,
+                                        384,
+                                        215
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                226,
+                                                201,
+                                                384,
+                                                215
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Table 5: MAML hyperparameter summary.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 7
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "type": "table_body",
+                            "bbox": [
+                                116,
+                                227,
+                                496,
+                                361
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        227,
+                                        496,
+                                        361
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                116,
+                                                227,
+                                                496,
+                                                361
+                                            ],
+                                            "score": 0.985,
+                                            "html": "<table><tr><td>Hyperparameter</td><td>MNIST</td><td>Omniglot</td><td>miniImageNet</td><td>CelebA</td></tr><tr><td>Input size</td><td>28×28</td><td>28×28</td><td>84 ×84×3</td><td>84 ×84×3</td></tr><tr><td>Outer (meta) learning rate</td><td>0.001</td><td>0.001</td><td>0.001</td><td>0.001</td></tr><tr><td>Inner learning rate</td><td>0.05</td><td>0.05</td><td>0.05</td><td>0.05</td></tr><tr><td>Task batch size</td><td>8</td><td>8</td><td>8</td><td>8</td></tr><tr><td>Inner adaptation steps (meta-training)</td><td>5</td><td></td><td>5</td><td>5</td></tr><tr><td>Meta-training iterations</td><td>30,000</td><td>5 30,000</td><td>60,000</td><td>60,000</td></tr><tr><td>Adaptation steps (evaluation)</td><td>50</td><td>50</td><td>50</td><td>50</td></tr><tr><td>Classes per task (meta-training)</td><td>10</td><td>20</td><td>5</td><td>2</td></tr><tr><td>Shots per class (meta-training)</td><td>1</td><td>1</td><td>1</td><td>1</td></tr><tr><td>Queries per class</td><td>5</td><td>5</td><td>5</td><td>5</td></tr></table>",
+                                            "type": "table",
+                                            "image_path": "2ec3224bb93dcc723ac1de7bc1bd6a5ca9baee579e7397416bb952c4c0951f97.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 9,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        227,
+                                        496,
+                                        271.6666666666667
+                                    ],
+                                    "spans": [],
+                                    "index": 8
+                                },
+                                {
+                                    "bbox": [
+                                        116,
+                                        271.6666666666667,
+                                        496,
+                                        316.33333333333337
+                                    ],
+                                    "spans": [],
+                                    "index": 9
+                                },
+                                {
+                                    "bbox": [
+                                        116,
+                                        316.33333333333337,
+                                        496,
+                                        361.00000000000006
+                                    ],
+                                    "spans": [],
+                                    "index": 10
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 8.0
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        384,
+                        505,
+                        462
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                384,
+                                505,
+                                397
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        384,
+                                        505,
+                                        397
+                                    ],
+                                    "score": 1.0,
+                                    "content": "For MNIST and Omniglot we use the same 4-block convolutional architecture as used by Finn et al.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                394,
+                                506,
+                                409
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        394,
+                                        506,
+                                        409
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(2017) for their Omniglot experiments, but with 32 filters (instead of 64) for each convolutional layer",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                407,
+                                505,
+                                419
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        407,
+                                        505,
+                                        419
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for consistency with the model used for miniImageNet and CelebA, which is the same as what Finn",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                416,
+                                505,
+                                431
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        416,
+                                        505,
+                                        431
+                                    ],
+                                    "score": 1.0,
+                                    "content": "et al. (2017) used for their miniImageNet experiments. When evaluating the meta-learned 20-way",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                428,
+                                505,
+                                442
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        428,
+                                        505,
+                                        442
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Omniglot model with 5-way tasks, we prune the unused output dimensions. The outer optimizer",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                438,
+                                505,
+                                453
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        438,
+                                        505,
+                                        453
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is Adam (Kingma & Ba, 2014), and the inner optimizer is SGD. We build on the authors’ publicly",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                450,
+                                403,
+                                463
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        450,
+                                        403,
+                                        463
+                                    ],
+                                    "score": 1.0,
+                                    "content": "available codebase found at https://github.com/cbfinn/maml.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        }
+                    ],
+                    "index": 14
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        467,
+                        505,
+                        523
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                468,
+                                504,
+                                480
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        468,
+                                        504,
+                                        480
+                                    ],
+                                    "score": 1.0,
+                                    "content": "When using batch normalization (Ioffe & Szegedy, 2015) to process a task’s training or query in-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                478,
+                                505,
+                                491
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        478,
+                                        505,
+                                        491
+                                    ],
+                                    "score": 1.0,
+                                    "content": "puts, we observe that using only 1 query datapoint per class can allow the model to exploit batch",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                489,
+                                505,
+                                503
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        489,
+                                        505,
+                                        503
+                                    ],
+                                    "score": 1.0,
+                                    "content": "statistics, learning a strategy analogous to a process of elimination that causes significant, but spuri-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                501,
+                                505,
+                                513
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        501,
+                                        505,
+                                        513
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ous, improvement in accuracy. To mitigate this, we fix 5 queries per class for every task’s evaluation",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                511,
+                                255,
+                                525
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        511,
+                                        255,
+                                        525
+                                    ],
+                                    "score": 1.0,
+                                    "content": "phase, meta-training or meta-testing.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        }
+                    ],
+                    "index": 20
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        536,
+                        185,
+                        547
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                534,
+                                186,
+                                549
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        534,
+                                        186,
+                                        549
+                                    ],
+                                    "score": 1.0,
+                                    "content": "E.2 PROTONETS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        }
+                    ],
+                    "index": 23
+                },
+                {
+                    "type": "table",
+                    "bbox": [
+                        122,
+                        585,
+                        489,
+                        686
+                    ],
+                    "blocks": [
+                        {
+                            "type": "table_caption",
+                            "bbox": [
+                                223,
+                                560,
+                                387,
+                                571
+                            ],
+                            "group_id": 1,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        222,
+                                        559,
+                                        388,
+                                        573
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                222,
+                                                559,
+                                                388,
+                                                573
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Table 6: ProtoNets hyperparameter summary.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 24
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "type": "table_body",
+                            "bbox": [
+                                122,
+                                585,
+                                489,
+                                686
+                            ],
+                            "group_id": 1,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        122,
+                                        585,
+                                        489,
+                                        686
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                122,
+                                                585,
+                                                489,
+                                                686
+                                            ],
+                                            "score": 0.984,
+                                            "html": "<table><tr><td>Hyperparameter</td><td>Omniglot</td><td>miniImageNet</td><td>CelebA</td></tr><tr><td>Input size</td><td>28×28</td><td>84×84×3</td><td>84×84×3</td></tr><tr><td>Learning rate</td><td>0.001</td><td>0.001</td><td>0.001</td></tr><tr><td>Task batch size</td><td>1</td><td>1</td><td>1</td></tr><tr><td>Training iterations</td><td>30,000</td><td>60,000</td><td>60,000</td></tr><tr><td>Classes per task (meta-training)</td><td>20</td><td>5</td><td>2</td></tr><tr><td>Shots per class (meta-training)</td><td>1</td><td>1</td><td>1</td></tr><tr><td>Queries per class (meta-training/meta-testing)</td><td>15/5</td><td>15/5</td><td>15/5</td></tr></table>",
+                                            "type": "table",
+                                            "image_path": "05251c8e5fc194d39ee7bd093f158e80b90d1f4e442d42b8db485abe6931ad93.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 26,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        122,
+                                        585,
+                                        489,
+                                        618.6666666666666
+                                    ],
+                                    "spans": [],
+                                    "index": 25
+                                },
+                                {
+                                    "bbox": [
+                                        122,
+                                        618.6666666666666,
+                                        489,
+                                        652.3333333333333
+                                    ],
+                                    "spans": [],
+                                    "index": 26
+                                },
+                                {
+                                    "bbox": [
+                                        122,
+                                        652.3333333333333,
+                                        489,
+                                        685.9999999999999
+                                    ],
+                                    "spans": [],
+                                    "index": 27
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 25.0
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        709,
+                        504,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                709,
+                                505,
+                                722
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        709,
+                                        505,
+                                        722
+                                    ],
+                                    "score": 1.0,
+                                    "content": "For the three considered datasets we use the same architecture as used by Snell et al. (2017) for their",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                720,
+                                505,
+                                732
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        720,
+                                        505,
+                                        732
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Omniglot and miniImageNet experiments. This is a 4-block convolutional architecture with each",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        }
+                    ],
+                    "index": 28.5
+                }
+            ],
+            "page_idx": 16,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        107,
+                        27,
+                        293,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                26,
+                                294,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        26,
+                                        294,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Published as a conference paper at ICLR 2019",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        300,
+                        751,
+                        311,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                299,
+                                750,
+                                312,
+                                764
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        299,
+                                        750,
+                                        312,
+                                        764
+                                    ],
+                                    "score": 1.0,
+                                    "content": "17",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        82,
+                        505,
+                        138
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                82,
+                                505,
+                                94
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        82,
+                                        209,
+                                        94
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Departing from the fixed",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        209,
+                                        83,
+                                        246,
+                                        93
+                                    ],
+                                    "score": 0.92,
+                                    "content": "k = 5 0 0",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        246,
+                                        82,
+                                        436,
+                                        94
+                                    ],
+                                    "score": 1.0,
+                                    "content": "used for all other datasets, we deliberately use",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        436,
+                                        83,
+                                        468,
+                                        93
+                                    ],
+                                    "score": 0.91,
+                                    "content": "k = 1 0",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        468,
+                                        82,
+                                        505,
+                                        94
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to better",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                93,
+                                505,
+                                106
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        93,
+                                        505,
+                                        106
+                                    ],
+                                    "score": 1.0,
+                                    "content": "understand the limitations of CACTUs-MAML. The results can be seen in Table 7 in Appendix B.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                105,
+                                505,
+                                117
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        105,
+                                        505,
+                                        117
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In brief, with the better embeddings (ACAI and InfoGAN), there is only little benefit of CACTUs-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                115,
+                                505,
+                                128
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        115,
+                                        505,
+                                        128
+                                    ],
+                                    "score": 1.0,
+                                    "content": "MAML over embedding cluster matching. Additionally, even in the best cases, CACTUs-MAML",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                125,
+                                362,
+                                139
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        125,
+                                        362,
+                                        139
+                                    ],
+                                    "score": 1.0,
+                                    "content": "falls short of state-of-the-art semi-supervised learning methods.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        }
+                    ],
+                    "index": 2,
+                    "bbox_fs": [
+                        105,
+                        82,
+                        505,
+                        139
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        109,
+                        154,
+                        405,
+                        167
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                107,
+                                153,
+                                406,
+                                168
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        153,
+                                        406,
+                                        168
+                                    ],
+                                    "score": 1.0,
+                                    "content": "APPENDIX E HYPERPARAMETERS AND ARCHITECTURES",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        }
+                    ],
+                    "index": 5
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        106,
+                        178,
+                        165,
+                        190
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                178,
+                                167,
+                                191
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        178,
+                                        167,
+                                        191
+                                    ],
+                                    "score": 1.0,
+                                    "content": "E.1 MAML",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        }
+                    ],
+                    "index": 6
+                },
+                {
+                    "type": "table",
+                    "bbox": [
+                        116,
+                        227,
+                        496,
+                        361
+                    ],
+                    "blocks": [
+                        {
+                            "type": "table_caption",
+                            "bbox": [
+                                227,
+                                202,
+                                383,
+                                214
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        226,
+                                        201,
+                                        384,
+                                        215
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                226,
+                                                201,
+                                                384,
+                                                215
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Table 5: MAML hyperparameter summary.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 7
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "type": "table_body",
+                            "bbox": [
+                                116,
+                                227,
+                                496,
+                                361
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        227,
+                                        496,
+                                        361
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                116,
+                                                227,
+                                                496,
+                                                361
+                                            ],
+                                            "score": 0.985,
+                                            "html": "<table><tr><td>Hyperparameter</td><td>MNIST</td><td>Omniglot</td><td>miniImageNet</td><td>CelebA</td></tr><tr><td>Input size</td><td>28×28</td><td>28×28</td><td>84 ×84×3</td><td>84 ×84×3</td></tr><tr><td>Outer (meta) learning rate</td><td>0.001</td><td>0.001</td><td>0.001</td><td>0.001</td></tr><tr><td>Inner learning rate</td><td>0.05</td><td>0.05</td><td>0.05</td><td>0.05</td></tr><tr><td>Task batch size</td><td>8</td><td>8</td><td>8</td><td>8</td></tr><tr><td>Inner adaptation steps (meta-training)</td><td>5</td><td></td><td>5</td><td>5</td></tr><tr><td>Meta-training iterations</td><td>30,000</td><td>5 30,000</td><td>60,000</td><td>60,000</td></tr><tr><td>Adaptation steps (evaluation)</td><td>50</td><td>50</td><td>50</td><td>50</td></tr><tr><td>Classes per task (meta-training)</td><td>10</td><td>20</td><td>5</td><td>2</td></tr><tr><td>Shots per class (meta-training)</td><td>1</td><td>1</td><td>1</td><td>1</td></tr><tr><td>Queries per class</td><td>5</td><td>5</td><td>5</td><td>5</td></tr></table>",
+                                            "type": "table",
+                                            "image_path": "2ec3224bb93dcc723ac1de7bc1bd6a5ca9baee579e7397416bb952c4c0951f97.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 9,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        227,
+                                        496,
+                                        271.6666666666667
+                                    ],
+                                    "spans": [],
+                                    "index": 8
+                                },
+                                {
+                                    "bbox": [
+                                        116,
+                                        271.6666666666667,
+                                        496,
+                                        316.33333333333337
+                                    ],
+                                    "spans": [],
+                                    "index": 9
+                                },
+                                {
+                                    "bbox": [
+                                        116,
+                                        316.33333333333337,
+                                        496,
+                                        361.00000000000006
+                                    ],
+                                    "spans": [],
+                                    "index": 10
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 8.0
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        384,
+                        505,
+                        462
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                384,
+                                505,
+                                397
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        384,
+                                        505,
+                                        397
+                                    ],
+                                    "score": 1.0,
+                                    "content": "For MNIST and Omniglot we use the same 4-block convolutional architecture as used by Finn et al.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                394,
+                                506,
+                                409
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        394,
+                                        506,
+                                        409
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(2017) for their Omniglot experiments, but with 32 filters (instead of 64) for each convolutional layer",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                407,
+                                505,
+                                419
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        407,
+                                        505,
+                                        419
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for consistency with the model used for miniImageNet and CelebA, which is the same as what Finn",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                416,
+                                505,
+                                431
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        416,
+                                        505,
+                                        431
+                                    ],
+                                    "score": 1.0,
+                                    "content": "et al. (2017) used for their miniImageNet experiments. When evaluating the meta-learned 20-way",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                428,
+                                505,
+                                442
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        428,
+                                        505,
+                                        442
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Omniglot model with 5-way tasks, we prune the unused output dimensions. The outer optimizer",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                438,
+                                505,
+                                453
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        438,
+                                        505,
+                                        453
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is Adam (Kingma & Ba, 2014), and the inner optimizer is SGD. We build on the authors’ publicly",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                450,
+                                403,
+                                463
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        450,
+                                        403,
+                                        463
+                                    ],
+                                    "score": 1.0,
+                                    "content": "available codebase found at https://github.com/cbfinn/maml.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        }
+                    ],
+                    "index": 14,
+                    "bbox_fs": [
+                        104,
+                        384,
+                        506,
+                        463
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        467,
+                        505,
+                        523
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                468,
+                                504,
+                                480
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        468,
+                                        504,
+                                        480
+                                    ],
+                                    "score": 1.0,
+                                    "content": "When using batch normalization (Ioffe & Szegedy, 2015) to process a task’s training or query in-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                478,
+                                505,
+                                491
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        478,
+                                        505,
+                                        491
+                                    ],
+                                    "score": 1.0,
+                                    "content": "puts, we observe that using only 1 query datapoint per class can allow the model to exploit batch",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                489,
+                                505,
+                                503
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        489,
+                                        505,
+                                        503
+                                    ],
+                                    "score": 1.0,
+                                    "content": "statistics, learning a strategy analogous to a process of elimination that causes significant, but spuri-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                501,
+                                505,
+                                513
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        501,
+                                        505,
+                                        513
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ous, improvement in accuracy. To mitigate this, we fix 5 queries per class for every task’s evaluation",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                511,
+                                255,
+                                525
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        511,
+                                        255,
+                                        525
+                                    ],
+                                    "score": 1.0,
+                                    "content": "phase, meta-training or meta-testing.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        }
+                    ],
+                    "index": 20,
+                    "bbox_fs": [
+                        105,
+                        468,
+                        505,
+                        525
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        536,
+                        185,
+                        547
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                534,
+                                186,
+                                549
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        534,
+                                        186,
+                                        549
+                                    ],
+                                    "score": 1.0,
+                                    "content": "E.2 PROTONETS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        }
+                    ],
+                    "index": 23
+                },
+                {
+                    "type": "table",
+                    "bbox": [
+                        122,
+                        585,
+                        489,
+                        686
+                    ],
+                    "blocks": [
+                        {
+                            "type": "table_caption",
+                            "bbox": [
+                                223,
+                                560,
+                                387,
+                                571
+                            ],
+                            "group_id": 1,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        222,
+                                        559,
+                                        388,
+                                        573
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                222,
+                                                559,
+                                                388,
+                                                573
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Table 6: ProtoNets hyperparameter summary.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 24
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "type": "table_body",
+                            "bbox": [
+                                122,
+                                585,
+                                489,
+                                686
+                            ],
+                            "group_id": 1,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        122,
+                                        585,
+                                        489,
+                                        686
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                122,
+                                                585,
+                                                489,
+                                                686
+                                            ],
+                                            "score": 0.984,
+                                            "html": "<table><tr><td>Hyperparameter</td><td>Omniglot</td><td>miniImageNet</td><td>CelebA</td></tr><tr><td>Input size</td><td>28×28</td><td>84×84×3</td><td>84×84×3</td></tr><tr><td>Learning rate</td><td>0.001</td><td>0.001</td><td>0.001</td></tr><tr><td>Task batch size</td><td>1</td><td>1</td><td>1</td></tr><tr><td>Training iterations</td><td>30,000</td><td>60,000</td><td>60,000</td></tr><tr><td>Classes per task (meta-training)</td><td>20</td><td>5</td><td>2</td></tr><tr><td>Shots per class (meta-training)</td><td>1</td><td>1</td><td>1</td></tr><tr><td>Queries per class (meta-training/meta-testing)</td><td>15/5</td><td>15/5</td><td>15/5</td></tr></table>",
+                                            "type": "table",
+                                            "image_path": "05251c8e5fc194d39ee7bd093f158e80b90d1f4e442d42b8db485abe6931ad93.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 26,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        122,
+                                        585,
+                                        489,
+                                        618.6666666666666
+                                    ],
+                                    "spans": [],
+                                    "index": 25
+                                },
+                                {
+                                    "bbox": [
+                                        122,
+                                        618.6666666666666,
+                                        489,
+                                        652.3333333333333
+                                    ],
+                                    "spans": [],
+                                    "index": 26
+                                },
+                                {
+                                    "bbox": [
+                                        122,
+                                        652.3333333333333,
+                                        489,
+                                        685.9999999999999
+                                    ],
+                                    "spans": [],
+                                    "index": 27
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 25.0
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        709,
+                        504,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                709,
+                                505,
+                                722
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        709,
+                                        505,
+                                        722
+                                    ],
+                                    "score": 1.0,
+                                    "content": "For the three considered datasets we use the same architecture as used by Snell et al. (2017) for their",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                720,
+                                505,
+                                732
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        720,
+                                        505,
+                                        732
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Omniglot and miniImageNet experiments. This is a 4-block convolutional architecture with each",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                81,
+                                505,
+                                95
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        81,
+                                        297,
+                                        95
+                                    ],
+                                    "score": 1.0,
+                                    "content": "block consisting of a convolutional layer with",
+                                    "type": "text",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        297,
+                                        83,
+                                        336,
+                                        93
+                                    ],
+                                    "score": 0.88,
+                                    "content": "6 4 \\ 3 \\times 3",
+                                    "type": "inline_equation",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        336,
+                                        81,
+                                        505,
+                                        95
+                                    ],
+                                    "score": 1.0,
+                                    "content": "filters, stride 1, and padding 1, followed",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                93,
+                                505,
+                                106
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        93,
+                                        260,
+                                        106
+                                    ],
+                                    "score": 1.0,
+                                    "content": "by BatchNorm, ReLU activation, and",
+                                    "type": "text",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        261,
+                                        94,
+                                        285,
+                                        104
+                                    ],
+                                    "score": 0.89,
+                                    "content": "2 \\times 2",
+                                    "type": "inline_equation",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        285,
+                                        93,
+                                        505,
+                                        106
+                                    ],
+                                    "score": 1.0,
+                                    "content": "MaxPooling. The ProtoNets embedding is simply the",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                105,
+                                505,
+                                117
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        105,
+                                        505,
+                                        117
+                                    ],
+                                    "score": 1.0,
+                                    "content": "flattened output of the last block. We follow the authors and use the Adam optimizer, but do not",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                115,
+                                506,
+                                128
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        115,
+                                        506,
+                                        128
+                                    ],
+                                    "score": 1.0,
+                                    "content": "use a learning rate scheduler. We build upon the authors’ publicly available codebase found at",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                126,
+                                411,
+                                140
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        126,
+                                        411,
+                                        140
+                                    ],
+                                    "score": 1.0,
+                                    "content": "https://github.com/jakesnell/prototypical-networks.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 4
+                        }
+                    ],
+                    "index": 28.5,
+                    "bbox_fs": [
+                        106,
+                        709,
+                        505,
+                        732
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        82,
+                        505,
+                        138
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                81,
+                                505,
+                                95
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        81,
+                                        297,
+                                        95
+                                    ],
+                                    "score": 1.0,
+                                    "content": "block consisting of a convolutional layer with",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        297,
+                                        83,
+                                        336,
+                                        93
+                                    ],
+                                    "score": 0.88,
+                                    "content": "6 4 \\ 3 \\times 3",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        336,
+                                        81,
+                                        505,
+                                        95
+                                    ],
+                                    "score": 1.0,
+                                    "content": "filters, stride 1, and padding 1, followed",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                93,
+                                505,
+                                106
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        93,
+                                        260,
+                                        106
+                                    ],
+                                    "score": 1.0,
+                                    "content": "by BatchNorm, ReLU activation, and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        261,
+                                        94,
+                                        285,
+                                        104
+                                    ],
+                                    "score": 0.89,
+                                    "content": "2 \\times 2",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        285,
+                                        93,
+                                        505,
+                                        106
+                                    ],
+                                    "score": 1.0,
+                                    "content": "MaxPooling. The ProtoNets embedding is simply the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                105,
+                                505,
+                                117
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        105,
+                                        505,
+                                        117
+                                    ],
+                                    "score": 1.0,
+                                    "content": "flattened output of the last block. We follow the authors and use the Adam optimizer, but do not",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                115,
+                                506,
+                                128
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        115,
+                                        506,
+                                        128
+                                    ],
+                                    "score": 1.0,
+                                    "content": "use a learning rate scheduler. We build upon the authors’ publicly available codebase found at",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                126,
+                                411,
+                                140
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        126,
+                                        411,
+                                        140
+                                    ],
+                                    "score": 1.0,
+                                    "content": "https://github.com/jakesnell/prototypical-networks.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        }
+                    ],
+                    "index": 2
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        150,
+                        174,
+                        162
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                149,
+                                176,
+                                165
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        149,
+                                        176,
+                                        165
+                                    ],
+                                    "score": 1.0,
+                                    "content": "E.3 CACTUS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        }
+                    ],
+                    "index": 5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        172,
+                        505,
+                        205
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                171,
+                                505,
+                                184
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        171,
+                                        397,
+                                        184
+                                    ],
+                                    "score": 1.0,
+                                    "content": "For Omniglot, miniImageNet, and CelebA we fix the number of clusters",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        397,
+                                        172,
+                                        404,
+                                        182
+                                    ],
+                                    "score": 0.82,
+                                    "content": "k",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        405,
+                                        171,
+                                        505,
+                                        184
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to be 500. For Omniglot",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                182,
+                                505,
+                                195
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        182,
+                                        253,
+                                        195
+                                    ],
+                                    "score": 1.0,
+                                    "content": "we choose the number of partitions",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        253,
+                                        183,
+                                        293,
+                                        193
+                                    ],
+                                    "score": 0.9,
+                                    "content": "P = 1 0 0",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        293,
+                                        182,
+                                        505,
+                                        195
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", but in the interest of keeping runtime manageable,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                194,
+                                296,
+                                206
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        194,
+                                        136,
+                                        206
+                                    ],
+                                    "score": 1.0,
+                                    "content": "choose",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        137,
+                                        194,
+                                        169,
+                                        204
+                                    ],
+                                    "score": 0.9,
+                                    "content": "P = 5 0",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        169,
+                                        194,
+                                        296,
+                                        206
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for miniImageNet and CelebA.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        }
+                    ],
+                    "index": 7
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        218,
+                        328,
+                        229
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                217,
+                                329,
+                                231
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        217,
+                                        329,
+                                        231
+                                    ],
+                                    "score": 1.0,
+                                    "content": "E.4 USE OF UNSUPERVISED LEARNING METHODS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        }
+                    ],
+                    "index": 9
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        238,
+                        504,
+                        283
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                237,
+                                506,
+                                253
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        237,
+                                        506,
+                                        253
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ACAI (Berthelot et al., 2018): We run ACAI for MNIST and Omniglot. We pad the images",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                250,
+                                505,
+                                262
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        250,
+                                        505,
+                                        262
+                                    ],
+                                    "score": 1.0,
+                                    "content": "by 2 and use the authors’ architecture. We use a 256-dimensional embedding for all datasets.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                260,
+                                505,
+                                274
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        260,
+                                        505,
+                                        274
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We build upon the authors’ publicly available codebase found at https://github.com/",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                273,
+                                226,
+                                283
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        273,
+                                        226,
+                                        283
+                                    ],
+                                    "score": 1.0,
+                                    "content": "brain-research/acai.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        }
+                    ],
+                    "index": 11.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        288,
+                        505,
+                        388
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                289,
+                                505,
+                                301
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        289,
+                                        268,
+                                        301
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We unsuccessfully try running ACAI on",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        269,
+                                        289,
+                                        302,
+                                        299
+                                    ],
+                                    "score": 0.88,
+                                    "content": "6 4 \\times 6 4",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        303,
+                                        289,
+                                        505,
+                                        301
+                                    ],
+                                    "score": 1.0,
+                                    "content": "miniImageNet and CelebA. To facilitate this input",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                300,
+                                505,
+                                312
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        300,
+                                        505,
+                                        312
+                                    ],
+                                    "score": 1.0,
+                                    "content": "size, we add one block consisting of two convolutional layers (512 filters each) and one down-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                311,
+                                505,
+                                323
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        311,
+                                        505,
+                                        323
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sampling/upsampling layer to the encoder and decoder. However, because of ACAI’s pixel-wise",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                321,
+                                505,
+                                334
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        321,
+                                        505,
+                                        334
+                                    ],
+                                    "score": 1.0,
+                                    "content": "reconstruction loss, for these datasets the ACAI embedding prioritizes information about the few",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                332,
+                                505,
+                                346
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        332,
+                                        505,
+                                        346
+                                    ],
+                                    "score": 1.0,
+                                    "content": "“features” that dominate the reconstruction pixel count, resulting in clusters that only corresponded",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                343,
+                                505,
+                                357
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        343,
+                                        505,
+                                        357
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to a limited range of factors, such as background color and pose. For curiosity’s sake, we tried run-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                354,
+                                505,
+                                368
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        354,
+                                        505,
+                                        368
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ning meta-learning on tasks derived from these uninteresting clusters anyways, and found that the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                365,
+                                505,
+                                379
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        365,
+                                        505,
+                                        379
+                                    ],
+                                    "score": 1.0,
+                                    "content": "meta-learner quickly produced a learning procedure that obtained high accuracy on the meta-training",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                377,
+                                424,
+                                389
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        377,
+                                        424,
+                                        389
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tasks. However, this learned prior was not useful for solving downstream tasks.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        }
+                    ],
+                    "index": 18
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        393,
+                        505,
+                        482
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                393,
+                                505,
+                                405
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        393,
+                                        505,
+                                        405
+                                    ],
+                                    "score": 1.0,
+                                    "content": "BiGAN (Donahue et al., 2017): For MNIST, we follow the BiGAN authors and specify a uni-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                405,
+                                505,
+                                416
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        405,
+                                        505,
+                                        416
+                                    ],
+                                    "score": 1.0,
+                                    "content": "form 50-dimensional prior on the unit hypercube for the latent. The BiGAN authors use a 200-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                414,
+                                505,
+                                428
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        414,
+                                        505,
+                                        428
+                                    ],
+                                    "score": 1.0,
+                                    "content": "dimensional version of the same prior for their ImageNet experiments, so we follow suit for Om-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                426,
+                                505,
+                                439
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        426,
+                                        505,
+                                        439
+                                    ],
+                                    "score": 1.0,
+                                    "content": "niglot, miniImageNet, and CelebA. For MNIST and Omniglot, we use the permutation-invariant",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                437,
+                                506,
+                                450
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        437,
+                                        506,
+                                        450
+                                    ],
+                                    "score": 1.0,
+                                    "content": "architecture (i.e. fully connected layers only) used by the authors for their MNIST results; for mini-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                448,
+                                505,
+                                461
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        448,
+                                        284,
+                                        461
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ImageNet and CelebA, we randomly crop to",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        284,
+                                        448,
+                                        316,
+                                        459
+                                    ],
+                                    "score": 0.91,
+                                    "content": "6 4 \\times 6 4",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        317,
+                                        448,
+                                        505,
+                                        461
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and use the AlexNet-inspired architecture used",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                459,
+                                505,
+                                471
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        459,
+                                        505,
+                                        471
+                                    ],
+                                    "score": 1.0,
+                                    "content": "by Donahue et al. (2017) for their ImageNet results. We build upon the authors’ publicly available",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                471,
+                                402,
+                                483
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        471,
+                                        402,
+                                        483
+                                    ],
+                                    "score": 1.0,
+                                    "content": "codebase found at https://github.com/jeffdonahue/bigan.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        }
+                    ],
+                    "index": 26.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        487,
+                        505,
+                        564
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                487,
+                                505,
+                                500
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        487,
+                                        505,
+                                        500
+                                    ],
+                                    "score": 1.0,
+                                    "content": "DeepCluster (Caron et al., 2018): We run DeepCluster for miniImageNet and CelebA, which we",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                498,
+                                505,
+                                510
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        498,
+                                        272,
+                                        510
+                                    ],
+                                    "score": 1.0,
+                                    "content": "respectively randomly crop and resize to",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        273,
+                                        498,
+                                        307,
+                                        509
+                                    ],
+                                    "score": 0.9,
+                                    "content": "6 4 \\times 6 4",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        307,
+                                        498,
+                                        505,
+                                        510
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". We modify the first layer of the AlexNet archi-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                508,
+                                505,
+                                522
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        508,
+                                        505,
+                                        522
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tecture used by the authors to accommodate this input size. We follow the authors and use the input",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                520,
+                                505,
+                                532
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        520,
+                                        505,
+                                        532
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to the (linear) output layer as the embedding. These are 4096-dimensional, so we follow the au-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                531,
+                                505,
+                                543
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        531,
+                                        505,
+                                        543
+                                    ],
+                                    "score": 1.0,
+                                    "content": "thors and apply PCA to reduce the dimensionality to 256, followed by whitening. We build upon the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                542,
+                                505,
+                                555
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        542,
+                                        505,
+                                        555
+                                    ],
+                                    "score": 1.0,
+                                    "content": "authors’ publicly available codebase found at https://github.com/facebookresearch/",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                553,
+                                177,
+                                566
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        553,
+                                        177,
+                                        566
+                                    ],
+                                    "score": 1.0,
+                                    "content": "deepcluster.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        }
+                    ],
+                    "index": 34
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        569,
+                        505,
+                        624
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                569,
+                                505,
+                                582
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        569,
+                                        505,
+                                        582
+                                    ],
+                                    "score": 1.0,
+                                    "content": "InfoGAN (Chen et al., 2016): We only run InfoGAN for MNIST. We follow the InfoGAN authors",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                581,
+                                505,
+                                593
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        581,
+                                        505,
+                                        593
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and specify the product of a 10-way categorical distribution and a 2-dimensional uniform distribution",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                591,
+                                505,
+                                604
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        591,
+                                        505,
+                                        604
+                                    ],
+                                    "score": 1.0,
+                                    "content": "as the latent code. We use the authors’ architecture. Given an image, we use the recognition network",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                601,
+                                505,
+                                617
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        601,
+                                        505,
+                                        617
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to obtain its embedding. We build upon the authors’ publicly available codebase found at https:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                613,
+                                273,
+                                627
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        613,
+                                        273,
+                                        627
+                                    ],
+                                    "score": 1.0,
+                                    "content": "//github.com/openai/InfoGAN.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        }
+                    ],
+                    "index": 40
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        641,
+                        314,
+                        653
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                641,
+                                315,
+                                655
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        641,
+                                        315,
+                                        655
+                                    ],
+                                    "score": 1.0,
+                                    "content": "APPENDIX F EXPERIMENTAL RESULTS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        }
+                    ],
+                    "index": 43
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        665,
+                        505,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                666,
+                                505,
+                                677
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        666,
+                                        505,
+                                        677
+                                    ],
+                                    "score": 1.0,
+                                    "content": "This section containsfull experimental results for the MNIST, Omniglot, miniImageNet, and CelebA",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                676,
+                                505,
+                                689
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        676,
+                                        505,
+                                        689
+                                    ],
+                                    "score": 1.0,
+                                    "content": "datasets, including consolidated versions of the tables found in the main text. The metric is clas-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                687,
+                                505,
+                                700
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        687,
+                                        505,
+                                        700
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sification accuracy averaged over 1000 tasks based on human-specified labels of the testing split,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                698,
+                                506,
+                                711
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        698,
+                                        127,
+                                        711
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        127,
+                                        699,
+                                        147,
+                                        709
+                                    ],
+                                    "score": 0.86,
+                                    "content": "9 5 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        147,
+                                        698,
+                                        237,
+                                        711
+                                    ],
+                                    "score": 1.0,
+                                    "content": "confidence intervals.",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        238,
+                                        699,
+                                        244,
+                                        709
+                                    ],
+                                    "score": 0.46,
+                                    "content": "d",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        245,
+                                        698,
+                                        375,
+                                        711
+                                    ],
+                                    "score": 1.0,
+                                    "content": ": dimensionality of embedding,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        375,
+                                        699,
+                                        382,
+                                        709
+                                    ],
+                                    "score": 0.56,
+                                    "content": "h",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        383,
+                                        698,
+                                        506,
+                                        711
+                                    ],
+                                    "score": 1.0,
+                                    "content": ": number of hidden units in a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 47
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                708,
+                                506,
+                                724
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        708,
+                                        198,
+                                        724
+                                    ],
+                                    "score": 1.0,
+                                    "content": "fully connected layer,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        199,
+                                        710,
+                                        205,
+                                        720
+                                    ],
+                                    "score": 0.7,
+                                    "content": "k",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        206,
+                                        708,
+                                        352,
+                                        724
+                                    ],
+                                    "score": 1.0,
+                                    "content": ": number of clusters in a partition,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        352,
+                                        711,
+                                        361,
+                                        720
+                                    ],
+                                    "score": 0.82,
+                                    "content": "P",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        361,
+                                        708,
+                                        506,
+                                        724
+                                    ],
+                                    "score": 1.0,
+                                    "content": ": number of partitions used during",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 48
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                720,
+                                354,
+                                733
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        720,
+                                        167,
+                                        733
+                                    ],
+                                    "score": 1.0,
+                                    "content": "meta-learning,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        167,
+                                        722,
+                                        177,
+                                        730
+                                    ],
+                                    "score": 0.67,
+                                    "content": "m",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        177,
+                                        720,
+                                        354,
+                                        733
+                                    ],
+                                    "score": 1.0,
+                                    "content": ": margin on boundary-defining hyperplanes.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 49
+                        }
+                    ],
+                    "index": 46.5
+                }
+            ],
+            "page_idx": 17,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        108,
+                        27,
+                        293,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                26,
+                                294,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        26,
+                                        294,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Published as a conference paper at ICLR 2019",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        301,
+                        751,
+                        311,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                299,
+                                750,
+                                312,
+                                763
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        299,
+                                        750,
+                                        312,
+                                        763
+                                    ],
+                                    "score": 1.0,
+                                    "content": "18",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        82,
+                        505,
+                        138
+                    ],
+                    "lines": [],
+                    "index": 2,
+                    "bbox_fs": [
+                        105,
+                        81,
+                        506,
+                        140
+                    ],
+                    "lines_deleted": true
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        150,
+                        174,
+                        162
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                149,
+                                176,
+                                165
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        149,
+                                        176,
+                                        165
+                                    ],
+                                    "score": 1.0,
+                                    "content": "E.3 CACTUS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        }
+                    ],
+                    "index": 5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        172,
+                        505,
+                        205
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                171,
+                                505,
+                                184
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        171,
+                                        397,
+                                        184
+                                    ],
+                                    "score": 1.0,
+                                    "content": "For Omniglot, miniImageNet, and CelebA we fix the number of clusters",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        397,
+                                        172,
+                                        404,
+                                        182
+                                    ],
+                                    "score": 0.82,
+                                    "content": "k",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        405,
+                                        171,
+                                        505,
+                                        184
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to be 500. For Omniglot",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                182,
+                                505,
+                                195
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        182,
+                                        253,
+                                        195
+                                    ],
+                                    "score": 1.0,
+                                    "content": "we choose the number of partitions",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        253,
+                                        183,
+                                        293,
+                                        193
+                                    ],
+                                    "score": 0.9,
+                                    "content": "P = 1 0 0",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        293,
+                                        182,
+                                        505,
+                                        195
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", but in the interest of keeping runtime manageable,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                194,
+                                296,
+                                206
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        194,
+                                        136,
+                                        206
+                                    ],
+                                    "score": 1.0,
+                                    "content": "choose",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        137,
+                                        194,
+                                        169,
+                                        204
+                                    ],
+                                    "score": 0.9,
+                                    "content": "P = 5 0",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        169,
+                                        194,
+                                        296,
+                                        206
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for miniImageNet and CelebA.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        }
+                    ],
+                    "index": 7,
+                    "bbox_fs": [
+                        106,
+                        171,
+                        505,
+                        206
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        218,
+                        328,
+                        229
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                217,
+                                329,
+                                231
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        217,
+                                        329,
+                                        231
+                                    ],
+                                    "score": 1.0,
+                                    "content": "E.4 USE OF UNSUPERVISED LEARNING METHODS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        }
+                    ],
+                    "index": 9
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        238,
+                        504,
+                        283
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                237,
+                                506,
+                                253
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        237,
+                                        506,
+                                        253
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ACAI (Berthelot et al., 2018): We run ACAI for MNIST and Omniglot. We pad the images",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                250,
+                                505,
+                                262
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        250,
+                                        505,
+                                        262
+                                    ],
+                                    "score": 1.0,
+                                    "content": "by 2 and use the authors’ architecture. We use a 256-dimensional embedding for all datasets.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                260,
+                                505,
+                                274
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        260,
+                                        505,
+                                        274
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We build upon the authors’ publicly available codebase found at https://github.com/",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                273,
+                                226,
+                                283
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        273,
+                                        226,
+                                        283
+                                    ],
+                                    "score": 1.0,
+                                    "content": "brain-research/acai.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        }
+                    ],
+                    "index": 11.5,
+                    "bbox_fs": [
+                        104,
+                        237,
+                        506,
+                        283
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        288,
+                        505,
+                        388
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                289,
+                                505,
+                                301
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        289,
+                                        268,
+                                        301
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We unsuccessfully try running ACAI on",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        269,
+                                        289,
+                                        302,
+                                        299
+                                    ],
+                                    "score": 0.88,
+                                    "content": "6 4 \\times 6 4",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        303,
+                                        289,
+                                        505,
+                                        301
+                                    ],
+                                    "score": 1.0,
+                                    "content": "miniImageNet and CelebA. To facilitate this input",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                300,
+                                505,
+                                312
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        300,
+                                        505,
+                                        312
+                                    ],
+                                    "score": 1.0,
+                                    "content": "size, we add one block consisting of two convolutional layers (512 filters each) and one down-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                311,
+                                505,
+                                323
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        311,
+                                        505,
+                                        323
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sampling/upsampling layer to the encoder and decoder. However, because of ACAI’s pixel-wise",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                321,
+                                505,
+                                334
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        321,
+                                        505,
+                                        334
+                                    ],
+                                    "score": 1.0,
+                                    "content": "reconstruction loss, for these datasets the ACAI embedding prioritizes information about the few",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                332,
+                                505,
+                                346
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        332,
+                                        505,
+                                        346
+                                    ],
+                                    "score": 1.0,
+                                    "content": "“features” that dominate the reconstruction pixel count, resulting in clusters that only corresponded",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                343,
+                                505,
+                                357
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        343,
+                                        505,
+                                        357
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to a limited range of factors, such as background color and pose. For curiosity’s sake, we tried run-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                354,
+                                505,
+                                368
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        354,
+                                        505,
+                                        368
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ning meta-learning on tasks derived from these uninteresting clusters anyways, and found that the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                365,
+                                505,
+                                379
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        365,
+                                        505,
+                                        379
+                                    ],
+                                    "score": 1.0,
+                                    "content": "meta-learner quickly produced a learning procedure that obtained high accuracy on the meta-training",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                377,
+                                424,
+                                389
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        377,
+                                        424,
+                                        389
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tasks. However, this learned prior was not useful for solving downstream tasks.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        }
+                    ],
+                    "index": 18,
+                    "bbox_fs": [
+                        105,
+                        289,
+                        505,
+                        389
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        393,
+                        505,
+                        482
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                393,
+                                505,
+                                405
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        393,
+                                        505,
+                                        405
+                                    ],
+                                    "score": 1.0,
+                                    "content": "BiGAN (Donahue et al., 2017): For MNIST, we follow the BiGAN authors and specify a uni-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                405,
+                                505,
+                                416
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        405,
+                                        505,
+                                        416
+                                    ],
+                                    "score": 1.0,
+                                    "content": "form 50-dimensional prior on the unit hypercube for the latent. The BiGAN authors use a 200-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                414,
+                                505,
+                                428
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        414,
+                                        505,
+                                        428
+                                    ],
+                                    "score": 1.0,
+                                    "content": "dimensional version of the same prior for their ImageNet experiments, so we follow suit for Om-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                426,
+                                505,
+                                439
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        426,
+                                        505,
+                                        439
+                                    ],
+                                    "score": 1.0,
+                                    "content": "niglot, miniImageNet, and CelebA. For MNIST and Omniglot, we use the permutation-invariant",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                437,
+                                506,
+                                450
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        437,
+                                        506,
+                                        450
+                                    ],
+                                    "score": 1.0,
+                                    "content": "architecture (i.e. fully connected layers only) used by the authors for their MNIST results; for mini-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                448,
+                                505,
+                                461
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        448,
+                                        284,
+                                        461
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ImageNet and CelebA, we randomly crop to",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        284,
+                                        448,
+                                        316,
+                                        459
+                                    ],
+                                    "score": 0.91,
+                                    "content": "6 4 \\times 6 4",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        317,
+                                        448,
+                                        505,
+                                        461
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and use the AlexNet-inspired architecture used",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                459,
+                                505,
+                                471
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        459,
+                                        505,
+                                        471
+                                    ],
+                                    "score": 1.0,
+                                    "content": "by Donahue et al. (2017) for their ImageNet results. We build upon the authors’ publicly available",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                471,
+                                402,
+                                483
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        471,
+                                        402,
+                                        483
+                                    ],
+                                    "score": 1.0,
+                                    "content": "codebase found at https://github.com/jeffdonahue/bigan.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        }
+                    ],
+                    "index": 26.5,
+                    "bbox_fs": [
+                        105,
+                        393,
+                        506,
+                        483
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        487,
+                        505,
+                        564
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                487,
+                                505,
+                                500
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        487,
+                                        505,
+                                        500
+                                    ],
+                                    "score": 1.0,
+                                    "content": "DeepCluster (Caron et al., 2018): We run DeepCluster for miniImageNet and CelebA, which we",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                498,
+                                505,
+                                510
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        498,
+                                        272,
+                                        510
+                                    ],
+                                    "score": 1.0,
+                                    "content": "respectively randomly crop and resize to",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        273,
+                                        498,
+                                        307,
+                                        509
+                                    ],
+                                    "score": 0.9,
+                                    "content": "6 4 \\times 6 4",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        307,
+                                        498,
+                                        505,
+                                        510
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". We modify the first layer of the AlexNet archi-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                508,
+                                505,
+                                522
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        508,
+                                        505,
+                                        522
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tecture used by the authors to accommodate this input size. We follow the authors and use the input",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                520,
+                                505,
+                                532
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        520,
+                                        505,
+                                        532
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to the (linear) output layer as the embedding. These are 4096-dimensional, so we follow the au-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                531,
+                                505,
+                                543
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        531,
+                                        505,
+                                        543
+                                    ],
+                                    "score": 1.0,
+                                    "content": "thors and apply PCA to reduce the dimensionality to 256, followed by whitening. We build upon the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                542,
+                                505,
+                                555
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        542,
+                                        505,
+                                        555
+                                    ],
+                                    "score": 1.0,
+                                    "content": "authors’ publicly available codebase found at https://github.com/facebookresearch/",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                553,
+                                177,
+                                566
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        553,
+                                        177,
+                                        566
+                                    ],
+                                    "score": 1.0,
+                                    "content": "deepcluster.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        }
+                    ],
+                    "index": 34,
+                    "bbox_fs": [
+                        105,
+                        487,
+                        505,
+                        566
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        569,
+                        505,
+                        624
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                569,
+                                505,
+                                582
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        569,
+                                        505,
+                                        582
+                                    ],
+                                    "score": 1.0,
+                                    "content": "InfoGAN (Chen et al., 2016): We only run InfoGAN for MNIST. We follow the InfoGAN authors",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                581,
+                                505,
+                                593
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        581,
+                                        505,
+                                        593
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and specify the product of a 10-way categorical distribution and a 2-dimensional uniform distribution",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                591,
+                                505,
+                                604
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        591,
+                                        505,
+                                        604
+                                    ],
+                                    "score": 1.0,
+                                    "content": "as the latent code. We use the authors’ architecture. Given an image, we use the recognition network",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                601,
+                                505,
+                                617
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        601,
+                                        505,
+                                        617
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to obtain its embedding. We build upon the authors’ publicly available codebase found at https:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                613,
+                                273,
+                                627
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        613,
+                                        273,
+                                        627
+                                    ],
+                                    "score": 1.0,
+                                    "content": "//github.com/openai/InfoGAN.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        }
+                    ],
+                    "index": 40,
+                    "bbox_fs": [
+                        105,
+                        569,
+                        505,
+                        627
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        641,
+                        314,
+                        653
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                641,
+                                315,
+                                655
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        641,
+                                        315,
+                                        655
+                                    ],
+                                    "score": 1.0,
+                                    "content": "APPENDIX F EXPERIMENTAL RESULTS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        }
+                    ],
+                    "index": 43
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        665,
+                        505,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                666,
+                                505,
+                                677
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        666,
+                                        505,
+                                        677
+                                    ],
+                                    "score": 1.0,
+                                    "content": "This section containsfull experimental results for the MNIST, Omniglot, miniImageNet, and CelebA",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                676,
+                                505,
+                                689
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        676,
+                                        505,
+                                        689
+                                    ],
+                                    "score": 1.0,
+                                    "content": "datasets, including consolidated versions of the tables found in the main text. The metric is clas-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                687,
+                                505,
+                                700
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        687,
+                                        505,
+                                        700
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sification accuracy averaged over 1000 tasks based on human-specified labels of the testing split,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                698,
+                                506,
+                                711
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        698,
+                                        127,
+                                        711
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        127,
+                                        699,
+                                        147,
+                                        709
+                                    ],
+                                    "score": 0.86,
+                                    "content": "9 5 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        147,
+                                        698,
+                                        237,
+                                        711
+                                    ],
+                                    "score": 1.0,
+                                    "content": "confidence intervals.",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        238,
+                                        699,
+                                        244,
+                                        709
+                                    ],
+                                    "score": 0.46,
+                                    "content": "d",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        245,
+                                        698,
+                                        375,
+                                        711
+                                    ],
+                                    "score": 1.0,
+                                    "content": ": dimensionality of embedding,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        375,
+                                        699,
+                                        382,
+                                        709
+                                    ],
+                                    "score": 0.56,
+                                    "content": "h",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        383,
+                                        698,
+                                        506,
+                                        711
+                                    ],
+                                    "score": 1.0,
+                                    "content": ": number of hidden units in a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 47
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                708,
+                                506,
+                                724
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        708,
+                                        198,
+                                        724
+                                    ],
+                                    "score": 1.0,
+                                    "content": "fully connected layer,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        199,
+                                        710,
+                                        205,
+                                        720
+                                    ],
+                                    "score": 0.7,
+                                    "content": "k",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        206,
+                                        708,
+                                        352,
+                                        724
+                                    ],
+                                    "score": 1.0,
+                                    "content": ": number of clusters in a partition,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        352,
+                                        711,
+                                        361,
+                                        720
+                                    ],
+                                    "score": 0.82,
+                                    "content": "P",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        361,
+                                        708,
+                                        506,
+                                        724
+                                    ],
+                                    "score": 1.0,
+                                    "content": ": number of partitions used during",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 48
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                720,
+                                354,
+                                733
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        720,
+                                        167,
+                                        733
+                                    ],
+                                    "score": 1.0,
+                                    "content": "meta-learning,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        167,
+                                        722,
+                                        177,
+                                        730
+                                    ],
+                                    "score": 0.67,
+                                    "content": "m",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        177,
+                                        720,
+                                        354,
+                                        733
+                                    ],
+                                    "score": 1.0,
+                                    "content": ": margin on boundary-defining hyperplanes.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 49
+                        }
+                    ],
+                    "index": 46.5,
+                    "bbox_fs": [
+                        105,
+                        666,
+                        506,
+                        733
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "table",
+                    "bbox": [
+                        181,
+                        190,
+                        457,
+                        622
+                    ],
+                    "blocks": [
+                        {
+                            "type": "table_body",
+                            "bbox": [
+                                181,
+                                190,
+                                457,
+                                622
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        181,
+                                        190,
+                                        457,
+                                        622
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                181,
+                                                190,
+                                                457,
+                                                622
+                                            ],
+                                            "score": 0.965,
+                                            "html": "<table><tr><td rowspan=1 colspan=1>(0101)</td><td rowspan=4 colspan=1>% LI&#x27;0干06&#x27;16% 01.0581555% 71&#x27;0 干 10&#x27;96% 710 67&#x27;96% II&#x27;0干08&#x27;96% L70干08&#x27;88% 11011.55% 910 76.56% 71&#x27;0干80&#x27;96% 110 干 9596% 780 6444% 18031.95% 7109 8535% 40.0 99.56% 870 LL&#x27;46C[ = y[= T (sani) sSuiPPq Tt Tr-shyt[ = y[ = T (sA) P T T-SAT r rrrrret gaasetr Srrpeirrlrer iaerrertrrnrrltgFaraeregarrgrrirrettA4=P444</td><td rowspan=2 colspan=1>% 0509 86:10% 1£&#x27;0干0L09%50055672% 67089&#x27;89% 0000658% 104530% 01004535% 44061.44% 141 65:14%70015:35%03.0335.55% 0ε0干 S0&#x27;LL% 51081:14%41044:44</td><td rowspan=1 colspan=1>% 700 47.99% 11&#x27;0干 58&#x27;96% 01&#x27;0干 66&#x27;96% 010 干 87&#x27; L6% 600 干13&#x27; L6% 01756% 010  6</td><td rowspan=1 colspan=1>% L0&#x27;0 干 1S&#x27;86</td></tr><tr><td rowspan=2 colspan=1>(s01)</td><td rowspan=2 colspan=1>% LI&#x27;0干 S0&#x27;96% 170干19&#x27;96% 110T 6696% 010干8126% 010776% 010干£1L6% 010干 80&#x27;L6</td><td rowspan=2 colspan=1>% L00干1S&#x27;86</td></tr><tr><td rowspan=1 colspan=1>% 44061.44% 141 65:14%70015:35%03.0335.55% 0ε0干 S0&#x27;LL% 51081:14%41044:44</td></tr><tr><td rowspan=1 colspan=1>(101)(tous em)1oila</td><td rowspan=1 colspan=1>%230325:65% 680 F980%57025224%944144% 14011.64%85033335% 3：271:31[[ = y O = T‘(sIni) suipP-qs u Tr-t[ = y  = (sani) supPqa uu Trn-st[[ = y OII = T (I1I) SSA- V T-NC[ = yI=T sa   -T=grrrraree geasepr rripegahrlerserrerarrnriirenggareregaerrriipetP=P-14</td><td rowspan=1 colspan=1>% 155311:55% 7408126% 31.074.16%050333:5% 610干80&#x27;96% LI0干 6996% L10 干8596C = y[= T(sa) supq c -[[ = y O = T‘(sni) sYuiPPqs uu Tr-sntC = y O = T () P 1 TI-sC=y[= T(I)A T T-sAg rrirpre aaetr rrirearrger serrerarrnriirertggaraerrarrrriiiertt=p 4</td><td rowspan=1 colspan=1>% L1&#x27;0干Iε&#x27;L6Surreend pasiersnnGOLnaier Giii-larei</td></tr></table>",
+                                            "type": "table",
+                                            "image_path": "51d7959ad0634b2257179895a73cb6b153b95b71e9943386f23a9e1ebc7580a9.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 21,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        181,
+                                        190,
+                                        457,
+                                        200.04651162790697
+                                    ],
+                                    "spans": [],
+                                    "index": 0
+                                },
+                                {
+                                    "bbox": [
+                                        181,
+                                        200.04651162790697,
+                                        457,
+                                        210.09302325581393
+                                    ],
+                                    "spans": [],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        181,
+                                        210.09302325581393,
+                                        457,
+                                        220.1395348837209
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                },
+                                {
+                                    "bbox": [
+                                        181,
+                                        220.1395348837209,
+                                        457,
+                                        230.18604651162786
+                                    ],
+                                    "spans": [],
+                                    "index": 3
+                                },
+                                {
+                                    "bbox": [
+                                        181,
+                                        230.18604651162786,
+                                        457,
+                                        240.23255813953483
+                                    ],
+                                    "spans": [],
+                                    "index": 4
+                                },
+                                {
+                                    "bbox": [
+                                        181,
+                                        240.23255813953483,
+                                        457,
+                                        250.2790697674418
+                                    ],
+                                    "spans": [],
+                                    "index": 5
+                                },
+                                {
+                                    "bbox": [
+                                        181,
+                                        250.2790697674418,
+                                        457,
+                                        260.32558139534876
+                                    ],
+                                    "spans": [],
+                                    "index": 6
+                                },
+                                {
+                                    "bbox": [
+                                        181,
+                                        260.32558139534876,
+                                        457,
+                                        270.37209302325573
+                                    ],
+                                    "spans": [],
+                                    "index": 7
+                                },
+                                {
+                                    "bbox": [
+                                        181,
+                                        270.37209302325573,
+                                        457,
+                                        280.4186046511627
+                                    ],
+                                    "spans": [],
+                                    "index": 8
+                                },
+                                {
+                                    "bbox": [
+                                        181,
+                                        280.4186046511627,
+                                        457,
+                                        290.46511627906966
+                                    ],
+                                    "spans": [],
+                                    "index": 9
+                                },
+                                {
+                                    "bbox": [
+                                        181,
+                                        290.46511627906966,
+                                        457,
+                                        300.5116279069766
+                                    ],
+                                    "spans": [],
+                                    "index": 10
+                                },
+                                {
+                                    "bbox": [
+                                        181,
+                                        300.5116279069766,
+                                        457,
+                                        310.5581395348836
+                                    ],
+                                    "spans": [],
+                                    "index": 11
+                                },
+                                {
+                                    "bbox": [
+                                        181,
+                                        310.5581395348836,
+                                        457,
+                                        320.60465116279056
+                                    ],
+                                    "spans": [],
+                                    "index": 12
+                                },
+                                {
+                                    "bbox": [
+                                        181,
+                                        320.60465116279056,
+                                        457,
+                                        330.6511627906975
+                                    ],
+                                    "spans": [],
+                                    "index": 13
+                                },
+                                {
+                                    "bbox": [
+                                        181,
+                                        330.6511627906975,
+                                        457,
+                                        340.6976744186045
+                                    ],
+                                    "spans": [],
+                                    "index": 14
+                                },
+                                {
+                                    "bbox": [
+                                        181,
+                                        340.6976744186045,
+                                        457,
+                                        350.74418604651146
+                                    ],
+                                    "spans": [],
+                                    "index": 15
+                                },
+                                {
+                                    "bbox": [
+                                        181,
+                                        350.74418604651146,
+                                        457,
+                                        360.7906976744184
+                                    ],
+                                    "spans": [],
+                                    "index": 16
+                                },
+                                {
+                                    "bbox": [
+                                        181,
+                                        360.7906976744184,
+                                        457,
+                                        370.8372093023254
+                                    ],
+                                    "spans": [],
+                                    "index": 17
+                                },
+                                {
+                                    "bbox": [
+                                        181,
+                                        370.8372093023254,
+                                        457,
+                                        380.88372093023236
+                                    ],
+                                    "spans": [],
+                                    "index": 18
+                                },
+                                {
+                                    "bbox": [
+                                        181,
+                                        380.88372093023236,
+                                        457,
+                                        390.9302325581393
+                                    ],
+                                    "spans": [],
+                                    "index": 19
+                                },
+                                {
+                                    "bbox": [
+                                        181,
+                                        390.9302325581393,
+                                        457,
+                                        400.9767441860463
+                                    ],
+                                    "spans": [],
+                                    "index": 20
+                                },
+                                {
+                                    "bbox": [
+                                        181,
+                                        400.9767441860463,
+                                        457,
+                                        411.02325581395326
+                                    ],
+                                    "spans": [],
+                                    "index": 21
+                                },
+                                {
+                                    "bbox": [
+                                        181,
+                                        411.02325581395326,
+                                        457,
+                                        421.0697674418602
+                                    ],
+                                    "spans": [],
+                                    "index": 22
+                                },
+                                {
+                                    "bbox": [
+                                        181,
+                                        421.0697674418602,
+                                        457,
+                                        431.1162790697672
+                                    ],
+                                    "spans": [],
+                                    "index": 23
+                                },
+                                {
+                                    "bbox": [
+                                        181,
+                                        431.1162790697672,
+                                        457,
+                                        441.16279069767415
+                                    ],
+                                    "spans": [],
+                                    "index": 24
+                                },
+                                {
+                                    "bbox": [
+                                        181,
+                                        441.16279069767415,
+                                        457,
+                                        451.2093023255811
+                                    ],
+                                    "spans": [],
+                                    "index": 25
+                                },
+                                {
+                                    "bbox": [
+                                        181,
+                                        451.2093023255811,
+                                        457,
+                                        461.2558139534881
+                                    ],
+                                    "spans": [],
+                                    "index": 26
+                                },
+                                {
+                                    "bbox": [
+                                        181,
+                                        461.2558139534881,
+                                        457,
+                                        471.30232558139505
+                                    ],
+                                    "spans": [],
+                                    "index": 27
+                                },
+                                {
+                                    "bbox": [
+                                        181,
+                                        471.30232558139505,
+                                        457,
+                                        481.348837209302
+                                    ],
+                                    "spans": [],
+                                    "index": 28
+                                },
+                                {
+                                    "bbox": [
+                                        181,
+                                        481.348837209302,
+                                        457,
+                                        491.395348837209
+                                    ],
+                                    "spans": [],
+                                    "index": 29
+                                },
+                                {
+                                    "bbox": [
+                                        181,
+                                        491.395348837209,
+                                        457,
+                                        501.44186046511595
+                                    ],
+                                    "spans": [],
+                                    "index": 30
+                                },
+                                {
+                                    "bbox": [
+                                        181,
+                                        501.44186046511595,
+                                        457,
+                                        511.4883720930229
+                                    ],
+                                    "spans": [],
+                                    "index": 31
+                                },
+                                {
+                                    "bbox": [
+                                        181,
+                                        511.4883720930229,
+                                        457,
+                                        521.5348837209299
+                                    ],
+                                    "spans": [],
+                                    "index": 32
+                                },
+                                {
+                                    "bbox": [
+                                        181,
+                                        521.5348837209299,
+                                        457,
+                                        531.5813953488368
+                                    ],
+                                    "spans": [],
+                                    "index": 33
+                                },
+                                {
+                                    "bbox": [
+                                        181,
+                                        531.5813953488368,
+                                        457,
+                                        541.6279069767438
+                                    ],
+                                    "spans": [],
+                                    "index": 34
+                                },
+                                {
+                                    "bbox": [
+                                        181,
+                                        541.6279069767438,
+                                        457,
+                                        551.6744186046508
+                                    ],
+                                    "spans": [],
+                                    "index": 35
+                                },
+                                {
+                                    "bbox": [
+                                        181,
+                                        551.6744186046508,
+                                        457,
+                                        561.7209302325577
+                                    ],
+                                    "spans": [],
+                                    "index": 36
+                                },
+                                {
+                                    "bbox": [
+                                        181,
+                                        561.7209302325577,
+                                        457,
+                                        571.7674418604647
+                                    ],
+                                    "spans": [],
+                                    "index": 37
+                                },
+                                {
+                                    "bbox": [
+                                        181,
+                                        571.7674418604647,
+                                        457,
+                                        581.8139534883717
+                                    ],
+                                    "spans": [],
+                                    "index": 38
+                                },
+                                {
+                                    "bbox": [
+                                        181,
+                                        581.8139534883717,
+                                        457,
+                                        591.8604651162786
+                                    ],
+                                    "spans": [],
+                                    "index": 39
+                                },
+                                {
+                                    "bbox": [
+                                        181,
+                                        591.8604651162786,
+                                        457,
+                                        601.9069767441856
+                                    ],
+                                    "spans": [],
+                                    "index": 40
+                                },
+                                {
+                                    "bbox": [
+                                        181,
+                                        601.9069767441856,
+                                        457,
+                                        611.9534883720926
+                                    ],
+                                    "spans": [],
+                                    "index": 41
+                                },
+                                {
+                                    "bbox": [
+                                        181,
+                                        611.9534883720926,
+                                        457,
+                                        621.9999999999995
+                                    ],
+                                    "spans": [],
+                                    "index": 42
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 21
+                }
+            ],
+            "page_idx": 18,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        107,
+                        27,
+                        293,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                26,
+                                294,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        26,
+                                        294,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Published as a conference paper at ICLR 2019",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        300,
+                        751,
+                        311,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                299,
+                                750,
+                                313,
+                                764
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        299,
+                                        750,
+                                        313,
+                                        764
+                                    ],
+                                    "score": 1.0,
+                                    "content": "19",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "table",
+                    "bbox": [
+                        181,
+                        190,
+                        457,
+                        622
+                    ],
+                    "blocks": [
+                        {
+                            "type": "table_body",
+                            "bbox": [
+                                181,
+                                190,
+                                457,
+                                622
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        181,
+                                        190,
+                                        457,
+                                        622
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                181,
+                                                190,
+                                                457,
+                                                622
+                                            ],
+                                            "score": 0.965,
+                                            "html": "<table><tr><td rowspan=1 colspan=1>(0101)</td><td rowspan=4 colspan=1>% LI&#x27;0干06&#x27;16% 01.0581555% 71&#x27;0 干 10&#x27;96% 710 67&#x27;96% II&#x27;0干08&#x27;96% L70干08&#x27;88% 11011.55% 910 76.56% 71&#x27;0干80&#x27;96% 110 干 9596% 780 6444% 18031.95% 7109 8535% 40.0 99.56% 870 LL&#x27;46C[ = y[= T (sani) sSuiPPq Tt Tr-shyt[ = y[ = T (sA) P T T-SAT r rrrrret gaasetr Srrpeirrlrer iaerrertrrnrrltgFaraeregarrgrrirrettA4=P444</td><td rowspan=2 colspan=1>% 0509 86:10% 1£&#x27;0干0L09%50055672% 67089&#x27;89% 0000658% 104530% 01004535% 44061.44% 141 65:14%70015:35%03.0335.55% 0ε0干 S0&#x27;LL% 51081:14%41044:44</td><td rowspan=1 colspan=1>% 700 47.99% 11&#x27;0干 58&#x27;96% 01&#x27;0干 66&#x27;96% 010 干 87&#x27; L6% 600 干13&#x27; L6% 01756% 010  6</td><td rowspan=1 colspan=1>% L0&#x27;0 干 1S&#x27;86</td></tr><tr><td rowspan=2 colspan=1>(s01)</td><td rowspan=2 colspan=1>% LI&#x27;0干 S0&#x27;96% 170干19&#x27;96% 110T 6696% 010干8126% 010776% 010干£1L6% 010干 80&#x27;L6</td><td rowspan=2 colspan=1>% L00干1S&#x27;86</td></tr><tr><td rowspan=1 colspan=1>% 44061.44% 141 65:14%70015:35%03.0335.55% 0ε0干 S0&#x27;LL% 51081:14%41044:44</td></tr><tr><td rowspan=1 colspan=1>(101)(tous em)1oila</td><td rowspan=1 colspan=1>%230325:65% 680 F980%57025224%944144% 14011.64%85033335% 3：271:31[[ = y O = T‘(sIni) suipP-qs u Tr-t[ = y  = (sani) supPqa uu Trn-st[[ = y OII = T (I1I) SSA- V T-NC[ = yI=T sa   -T=grrrraree geasepr rripegahrlerserrerarrnriirenggareregaerrriipetP=P-14</td><td rowspan=1 colspan=1>% 155311:55% 7408126% 31.074.16%050333:5% 610干80&#x27;96% LI0干 6996% L10 干8596C = y[= T(sa) supq c -[[ = y O = T‘(sni) sYuiPPqs uu Tr-sntC = y O = T () P 1 TI-sC=y[= T(I)A T T-sAg rrirpre aaetr rrirearrger serrerarrnriirertggaraerrarrrriiiertt=p 4</td><td rowspan=1 colspan=1>% L1&#x27;0干Iε&#x27;L6Surreend pasiersnnGOLnaier Giii-larei</td></tr></table>",
+                                            "type": "table",
+                                            "image_path": "51d7959ad0634b2257179895a73cb6b153b95b71e9943386f23a9e1ebc7580a9.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 21,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        181,
+                                        190,
+                                        457,
+                                        200.04651162790697
+                                    ],
+                                    "spans": [],
+                                    "index": 0
+                                },
+                                {
+                                    "bbox": [
+                                        181,
+                                        200.04651162790697,
+                                        457,
+                                        210.09302325581393
+                                    ],
+                                    "spans": [],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        181,
+                                        210.09302325581393,
+                                        457,
+                                        220.1395348837209
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                },
+                                {
+                                    "bbox": [
+                                        181,
+                                        220.1395348837209,
+                                        457,
+                                        230.18604651162786
+                                    ],
+                                    "spans": [],
+                                    "index": 3
+                                },
+                                {
+                                    "bbox": [
+                                        181,
+                                        230.18604651162786,
+                                        457,
+                                        240.23255813953483
+                                    ],
+                                    "spans": [],
+                                    "index": 4
+                                },
+                                {
+                                    "bbox": [
+                                        181,
+                                        240.23255813953483,
+                                        457,
+                                        250.2790697674418
+                                    ],
+                                    "spans": [],
+                                    "index": 5
+                                },
+                                {
+                                    "bbox": [
+                                        181,
+                                        250.2790697674418,
+                                        457,
+                                        260.32558139534876
+                                    ],
+                                    "spans": [],
+                                    "index": 6
+                                },
+                                {
+                                    "bbox": [
+                                        181,
+                                        260.32558139534876,
+                                        457,
+                                        270.37209302325573
+                                    ],
+                                    "spans": [],
+                                    "index": 7
+                                },
+                                {
+                                    "bbox": [
+                                        181,
+                                        270.37209302325573,
+                                        457,
+                                        280.4186046511627
+                                    ],
+                                    "spans": [],
+                                    "index": 8
+                                },
+                                {
+                                    "bbox": [
+                                        181,
+                                        280.4186046511627,
+                                        457,
+                                        290.46511627906966
+                                    ],
+                                    "spans": [],
+                                    "index": 9
+                                },
+                                {
+                                    "bbox": [
+                                        181,
+                                        290.46511627906966,
+                                        457,
+                                        300.5116279069766
+                                    ],
+                                    "spans": [],
+                                    "index": 10
+                                },
+                                {
+                                    "bbox": [
+                                        181,
+                                        300.5116279069766,
+                                        457,
+                                        310.5581395348836
+                                    ],
+                                    "spans": [],
+                                    "index": 11
+                                },
+                                {
+                                    "bbox": [
+                                        181,
+                                        310.5581395348836,
+                                        457,
+                                        320.60465116279056
+                                    ],
+                                    "spans": [],
+                                    "index": 12
+                                },
+                                {
+                                    "bbox": [
+                                        181,
+                                        320.60465116279056,
+                                        457,
+                                        330.6511627906975
+                                    ],
+                                    "spans": [],
+                                    "index": 13
+                                },
+                                {
+                                    "bbox": [
+                                        181,
+                                        330.6511627906975,
+                                        457,
+                                        340.6976744186045
+                                    ],
+                                    "spans": [],
+                                    "index": 14
+                                },
+                                {
+                                    "bbox": [
+                                        181,
+                                        340.6976744186045,
+                                        457,
+                                        350.74418604651146
+                                    ],
+                                    "spans": [],
+                                    "index": 15
+                                },
+                                {
+                                    "bbox": [
+                                        181,
+                                        350.74418604651146,
+                                        457,
+                                        360.7906976744184
+                                    ],
+                                    "spans": [],
+                                    "index": 16
+                                },
+                                {
+                                    "bbox": [
+                                        181,
+                                        360.7906976744184,
+                                        457,
+                                        370.8372093023254
+                                    ],
+                                    "spans": [],
+                                    "index": 17
+                                },
+                                {
+                                    "bbox": [
+                                        181,
+                                        370.8372093023254,
+                                        457,
+                                        380.88372093023236
+                                    ],
+                                    "spans": [],
+                                    "index": 18
+                                },
+                                {
+                                    "bbox": [
+                                        181,
+                                        380.88372093023236,
+                                        457,
+                                        390.9302325581393
+                                    ],
+                                    "spans": [],
+                                    "index": 19
+                                },
+                                {
+                                    "bbox": [
+                                        181,
+                                        390.9302325581393,
+                                        457,
+                                        400.9767441860463
+                                    ],
+                                    "spans": [],
+                                    "index": 20
+                                },
+                                {
+                                    "bbox": [
+                                        181,
+                                        400.9767441860463,
+                                        457,
+                                        411.02325581395326
+                                    ],
+                                    "spans": [],
+                                    "index": 21
+                                },
+                                {
+                                    "bbox": [
+                                        181,
+                                        411.02325581395326,
+                                        457,
+                                        421.0697674418602
+                                    ],
+                                    "spans": [],
+                                    "index": 22
+                                },
+                                {
+                                    "bbox": [
+                                        181,
+                                        421.0697674418602,
+                                        457,
+                                        431.1162790697672
+                                    ],
+                                    "spans": [],
+                                    "index": 23
+                                },
+                                {
+                                    "bbox": [
+                                        181,
+                                        431.1162790697672,
+                                        457,
+                                        441.16279069767415
+                                    ],
+                                    "spans": [],
+                                    "index": 24
+                                },
+                                {
+                                    "bbox": [
+                                        181,
+                                        441.16279069767415,
+                                        457,
+                                        451.2093023255811
+                                    ],
+                                    "spans": [],
+                                    "index": 25
+                                },
+                                {
+                                    "bbox": [
+                                        181,
+                                        451.2093023255811,
+                                        457,
+                                        461.2558139534881
+                                    ],
+                                    "spans": [],
+                                    "index": 26
+                                },
+                                {
+                                    "bbox": [
+                                        181,
+                                        461.2558139534881,
+                                        457,
+                                        471.30232558139505
+                                    ],
+                                    "spans": [],
+                                    "index": 27
+                                },
+                                {
+                                    "bbox": [
+                                        181,
+                                        471.30232558139505,
+                                        457,
+                                        481.348837209302
+                                    ],
+                                    "spans": [],
+                                    "index": 28
+                                },
+                                {
+                                    "bbox": [
+                                        181,
+                                        481.348837209302,
+                                        457,
+                                        491.395348837209
+                                    ],
+                                    "spans": [],
+                                    "index": 29
+                                },
+                                {
+                                    "bbox": [
+                                        181,
+                                        491.395348837209,
+                                        457,
+                                        501.44186046511595
+                                    ],
+                                    "spans": [],
+                                    "index": 30
+                                },
+                                {
+                                    "bbox": [
+                                        181,
+                                        501.44186046511595,
+                                        457,
+                                        511.4883720930229
+                                    ],
+                                    "spans": [],
+                                    "index": 31
+                                },
+                                {
+                                    "bbox": [
+                                        181,
+                                        511.4883720930229,
+                                        457,
+                                        521.5348837209299
+                                    ],
+                                    "spans": [],
+                                    "index": 32
+                                },
+                                {
+                                    "bbox": [
+                                        181,
+                                        521.5348837209299,
+                                        457,
+                                        531.5813953488368
+                                    ],
+                                    "spans": [],
+                                    "index": 33
+                                },
+                                {
+                                    "bbox": [
+                                        181,
+                                        531.5813953488368,
+                                        457,
+                                        541.6279069767438
+                                    ],
+                                    "spans": [],
+                                    "index": 34
+                                },
+                                {
+                                    "bbox": [
+                                        181,
+                                        541.6279069767438,
+                                        457,
+                                        551.6744186046508
+                                    ],
+                                    "spans": [],
+                                    "index": 35
+                                },
+                                {
+                                    "bbox": [
+                                        181,
+                                        551.6744186046508,
+                                        457,
+                                        561.7209302325577
+                                    ],
+                                    "spans": [],
+                                    "index": 36
+                                },
+                                {
+                                    "bbox": [
+                                        181,
+                                        561.7209302325577,
+                                        457,
+                                        571.7674418604647
+                                    ],
+                                    "spans": [],
+                                    "index": 37
+                                },
+                                {
+                                    "bbox": [
+                                        181,
+                                        571.7674418604647,
+                                        457,
+                                        581.8139534883717
+                                    ],
+                                    "spans": [],
+                                    "index": 38
+                                },
+                                {
+                                    "bbox": [
+                                        181,
+                                        581.8139534883717,
+                                        457,
+                                        591.8604651162786
+                                    ],
+                                    "spans": [],
+                                    "index": 39
+                                },
+                                {
+                                    "bbox": [
+                                        181,
+                                        591.8604651162786,
+                                        457,
+                                        601.9069767441856
+                                    ],
+                                    "spans": [],
+                                    "index": 40
+                                },
+                                {
+                                    "bbox": [
+                                        181,
+                                        601.9069767441856,
+                                        457,
+                                        611.9534883720926
+                                    ],
+                                    "spans": [],
+                                    "index": 41
+                                },
+                                {
+                                    "bbox": [
+                                        181,
+                                        611.9534883720926,
+                                        457,
+                                        621.9999999999995
+                                    ],
+                                    "spans": [],
+                                    "index": 42
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 21
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "table",
+                    "bbox": [
+                        154,
+                        168,
+                        459,
+                        642
+                    ],
+                    "blocks": [
+                        {
+                            "type": "table_footnote",
+                            "bbox": [
+                                119,
+                                92,
+                                144,
+                                727
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        161,
+                                        145,
+                                        731
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                127,
+                                                161,
+                                                145,
+                                                731
+                                            ],
+                                            "score": 0.0,
+                                            "content": "",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        118,
+                                        87,
+                                        134,
+                                        731
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                118,
+                                                87,
+                                                134,
+                                                731
+                                            ],
+                                            "score": 1.0,
+                                            "content": "denotes a 95% confidence interval. d: dimensionality of embedding,",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 0
+                                }
+                            ],
+                            "index": 0.5
+                        },
+                        {
+                            "type": "table_body",
+                            "bbox": [
+                                154,
+                                168,
+                                459,
+                                642
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        154,
+                                        168,
+                                        459,
+                                        642
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                154,
+                                                168,
+                                                459,
+                                                642
+                                            ],
+                                            "score": 0.966,
+                                            "html": "<table><tr><td rowspan=7 colspan=1>(s 0)(107)(s‘s)(1‘s)(ous em)oiiral</td><td rowspan=7 colspan=1>% 944% 810 449% 67&#x27;0干 10&#x27;61% 110013.55% 810 11&#x27;9%1102555% 6908&#x27;44% 690441% 1∠&#x27;06100%20035174% 32056577%33357530PPT =γ[= TIIII-SALAI SI11.1ay = y  = 111A1--4111rres ionnaieraBeesees</td><td rowspan=4 colspan=1>% 90.038399% 33.333339% 350 79:11  %80085.8  % 990 41.69 % 1333 3334% L00干 L7&#x27;996%83033139% 670227/4%3330730 % 30033331% 8109835% 4404 44.14  % 1 1124</td><td rowspan=1 colspan=1>% 90001.95% 2102 78.14% 401141% L£&#x27;0干90% 6£&#x27;0干 LL&#x27;0  %37:227755% 8057788</td><td rowspan=1 colspan=2>% 0105 6796% L0&#x27;0干18&#x27;86</td></tr><tr><td rowspan=1 colspan=1>% 90001.95% 2102 78.14% 401141% L£&#x27;0干90% 6£&#x27;0干 LL&#x27;0  %37:227755% 8057788% 30:0377.3050</td><td rowspan=2 colspan=2>% 0105 6796% L0&#x27;0干18&#x27;86% 700339:55% 810533.55</td></tr><tr><td rowspan=2 colspan=1>%11033135% 190 08:27% 770955.66%71224115% 210370.65% 200586.655210224555%335395559% 33033151</td></tr><tr><td></td><td rowspan=1 colspan=1>土+</td></tr><tr><td rowspan=3 colspan=1>% LS0干9118%8107818%S9007&#x27;LL% LL&#x27;0干60&#x27;IL% 09&#x27;0干18&#x27;18%8703355% 353355.55% 0S&#x27;0 干8L&#x27;L8% 3903 8138% 1113 1511% 711 819% 7.0 16:14  %7.055  % 08&#x27;0 干 64&#x27;99  % 480  71896260 =   =  () T-       CCT = y TT=1 (In) s11T1II-SSCC = γ OII = 1 ‘(S.I1I) T1I1-SALOCY =γ[=T (SIAI) TIII-SAI8 = Y inodop p PT SuippggregreemysppegarierereereirrA44=P4444</td><td rowspan=1 colspan=1></td><td></td></tr><tr><td rowspan=1 colspan=1>% I∠&#x27;0干 90&#x27;89% 99&#x27;0 干7L&#x27;89% 6/0 9579%8/07985% 69&#x27;009&#x27;L% 6905555% 19085.94% S9&#x27;0 干99&#x27;8L% L0干69</td><td rowspan=2 colspan=2>% 710£8&#x27;86% 60&#x27;0干85&#x27;66% 11.03.9555% 773523:32Surureree peseeedn  GDi Drirr eoeiiioriireri（OLDLier Girr-aarei</td></tr><tr><td rowspan=1 colspan=1>% 71 1171% 1742 1185% 6104544% 080 964% 5039.02 % 0807650% 18081&#x27;89% 71:044440 =  O = P (I) T-sdHP&#x27; =   = P(s) -s-L CC =y OII = 1 (SIAN) TSII-SA/S0C = y =  (i) sTod-snt8 = Y inidoip pm d SuippqC =yI =1 (SIA) T-SAagrrpaee easeta Srrplgrhgier neeennr neipreggarerrgrereippegPP =P 4400</td></tr></table>",
+                                            "type": "table",
+                                            "image_path": "694c397fac710ab1a4f54d056b99bd24fdc34ed1a1fa932f558420a2417f3576.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 2,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        154,
+                                        168,
+                                        459,
+                                        642
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                }
+                            ]
+                        },
+                        {
+                            "type": "table_caption",
+                            "bbox": [
+                                463,
+                                190,
+                                478,
+                                637
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        462,
+                                        198,
+                                        479,
+                                        634
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                462,
+                                                198,
+                                                479,
+                                                634
+                                            ],
+                                            "score": 1.0,
+                                            "content": "ers per convolutional layer, 3×data augmentation, and folded the validation set",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 3
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "type": "table_caption",
+                            "bbox": [
+                                477,
+                                516,
+                                484,
+                                619
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        474,
+                                        520,
+                                        486,
+                                        622
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                474,
+                                                520,
+                                                486,
+                                                622
+                                            ],
+                                            "score": 0.0,
+                                            "content": "",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 4
+                                }
+                            ],
+                            "index": 4
+                        }
+                    ],
+                    "index": 2.5
+                }
+            ],
+            "page_idx": 19,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        107,
+                        27,
+                        293,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                26,
+                                293,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        26,
+                                        293,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Published as a conference paper at ICLR 2019",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        300,
+                        751,
+                        311,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                298,
+                                749,
+                                313,
+                                764
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        298,
+                                        749,
+                                        313,
+                                        764
+                                    ],
+                                    "score": 1.0,
+                                    "content": "",
+                                    "type": "text",
+                                    "height": 15,
+                                    "width": 15
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "table",
+                    "bbox": [
+                        154,
+                        168,
+                        459,
+                        642
+                    ],
+                    "blocks": [
+                        {
+                            "type": "table_footnote",
+                            "bbox": [
+                                119,
+                                92,
+                                144,
+                                727
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        161,
+                                        145,
+                                        731
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                127,
+                                                161,
+                                                145,
+                                                731
+                                            ],
+                                            "score": 0.0,
+                                            "content": "",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        118,
+                                        87,
+                                        134,
+                                        731
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                118,
+                                                87,
+                                                134,
+                                                731
+                                            ],
+                                            "score": 1.0,
+                                            "content": "denotes a 95% confidence interval. d: dimensionality of embedding,",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 0
+                                }
+                            ],
+                            "index": 0.5
+                        },
+                        {
+                            "type": "table_body",
+                            "bbox": [
+                                154,
+                                168,
+                                459,
+                                642
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        154,
+                                        168,
+                                        459,
+                                        642
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                154,
+                                                168,
+                                                459,
+                                                642
+                                            ],
+                                            "score": 0.966,
+                                            "html": "<table><tr><td rowspan=7 colspan=1>(s 0)(107)(s‘s)(1‘s)(ous em)oiiral</td><td rowspan=7 colspan=1>% 944% 810 449% 67&#x27;0干 10&#x27;61% 110013.55% 810 11&#x27;9%1102555% 6908&#x27;44% 690441% 1∠&#x27;06100%20035174% 32056577%33357530PPT =γ[= TIIII-SALAI SI11.1ay = y  = 111A1--4111rres ionnaieraBeesees</td><td rowspan=4 colspan=1>% 90.038399% 33.333339% 350 79:11  %80085.8  % 990 41.69 % 1333 3334% L00干 L7&#x27;996%83033139% 670227/4%3330730 % 30033331% 8109835% 4404 44.14  % 1 1124</td><td rowspan=1 colspan=1>% 90001.95% 2102 78.14% 401141% L£&#x27;0干90% 6£&#x27;0干 LL&#x27;0  %37:227755% 8057788</td><td rowspan=1 colspan=2>% 0105 6796% L0&#x27;0干18&#x27;86</td></tr><tr><td rowspan=1 colspan=1>% 90001.95% 2102 78.14% 401141% L£&#x27;0干90% 6£&#x27;0干 LL&#x27;0  %37:227755% 8057788% 30:0377.3050</td><td rowspan=2 colspan=2>% 0105 6796% L0&#x27;0干18&#x27;86% 700339:55% 810533.55</td></tr><tr><td rowspan=2 colspan=1>%11033135% 190 08:27% 770955.66%71224115% 210370.65% 200586.655210224555%335395559% 33033151</td></tr><tr><td></td><td rowspan=1 colspan=1>土+</td></tr><tr><td rowspan=3 colspan=1>% LS0干9118%8107818%S9007&#x27;LL% LL&#x27;0干60&#x27;IL% 09&#x27;0干18&#x27;18%8703355% 353355.55% 0S&#x27;0 干8L&#x27;L8% 3903 8138% 1113 1511% 711 819% 7.0 16:14  %7.055  % 08&#x27;0 干 64&#x27;99  % 480  71896260 =   =  () T-       CCT = y TT=1 (In) s11T1II-SSCC = γ OII = 1 ‘(S.I1I) T1I1-SALOCY =γ[=T (SIAI) TIII-SAI8 = Y inodop p PT SuippggregreemysppegarierereereirrA44=P4444</td><td rowspan=1 colspan=1></td><td></td></tr><tr><td rowspan=1 colspan=1>% I∠&#x27;0干 90&#x27;89% 99&#x27;0 干7L&#x27;89% 6/0 9579%8/07985% 69&#x27;009&#x27;L% 6905555% 19085.94% S9&#x27;0 干99&#x27;8L% L0干69</td><td rowspan=2 colspan=2>% 710£8&#x27;86% 60&#x27;0干85&#x27;66% 11.03.9555% 773523:32Surureree peseeedn  GDi Drirr eoeiiioriireri（OLDLier Girr-aarei</td></tr><tr><td rowspan=1 colspan=1>% 71 1171% 1742 1185% 6104544% 080 964% 5039.02 % 0807650% 18081&#x27;89% 71:044440 =  O = P (I) T-sdHP&#x27; =   = P(s) -s-L CC =y OII = 1 (SIAN) TSII-SA/S0C = y =  (i) sTod-snt8 = Y inidoip pm d SuippqC =yI =1 (SIA) T-SAagrrpaee easeta Srrplgrhgier neeennr neipreggarerrgrereippegPP =P 4400</td></tr></table>",
+                                            "type": "table",
+                                            "image_path": "694c397fac710ab1a4f54d056b99bd24fdc34ed1a1fa932f558420a2417f3576.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 2,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        154,
+                                        168,
+                                        459,
+                                        642
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                }
+                            ]
+                        },
+                        {
+                            "type": "table_caption",
+                            "bbox": [
+                                463,
+                                190,
+                                478,
+                                637
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        462,
+                                        198,
+                                        479,
+                                        634
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                462,
+                                                198,
+                                                479,
+                                                634
+                                            ],
+                                            "score": 1.0,
+                                            "content": "ers per convolutional layer, 3×data augmentation, and folded the validation set",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 3
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "type": "table_caption",
+                            "bbox": [
+                                477,
+                                516,
+                                484,
+                                619
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        474,
+                                        520,
+                                        486,
+                                        622
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                474,
+                                                520,
+                                                486,
+                                                622
+                                            ],
+                                            "score": 0.0,
+                                            "content": "",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 4
+                                }
+                            ],
+                            "index": 4
+                        }
+                    ],
+                    "index": 2.5
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "table",
+                    "bbox": [
+                        177,
+                        170,
+                        461,
+                        644
+                    ],
+                    "blocks": [
+                        {
+                            "type": "table_footnote",
+                            "bbox": [
+                                140,
+                                99,
+                                165,
+                                729
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        150,
+                                        162,
+                                        167,
+                                        733
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                150,
+                                                162,
+                                                167,
+                                                733
+                                            ],
+                                            "score": 0.0,
+                                            "content": "",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        140,
+                                        95,
+                                        156,
+                                        733
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                140,
+                                                95,
+                                                156,
+                                                733
+                                            ],
+                                            "score": 1.0,
+                                            "content": "denotes a 95% confidence interval. d: dimensionality of embeddi",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 0
+                                }
+                            ],
+                            "index": 0.5
+                        },
+                        {
+                            "type": "table_body",
+                            "bbox": [
+                                177,
+                                170,
+                                461,
+                                644
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        177,
+                                        170,
+                                        461,
+                                        644
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                177,
+                                                170,
+                                                461,
+                                                644
+                                            ],
+                                            "score": 0.969,
+                                            "html": "<table><tr><td rowspan=1 colspan=1>(0s&#x27;s)(2 &#x27;0)(s‘s)(1‘s)(ous Xem)Woirai</td><td rowspan=1 colspan=1>% 5/.05 39.61% 533112% 9908589% 67003735Yrirlrriiir reirraperreeees</td><td rowspan=1 colspan=1>% 230309.3% 31:15:11% 1/.0 91&#x27;81%55333.99% 00&#x27;0干00&#x27;  % 89&#x27;0干 86&#x27;L% 89&#x27;0干 16&#x27;99% L9:0 17:99546%344231.21% 100404% 7/0 9000     %8900L&#x27;79%57003515% 5703.31%£9090:65%81096565% 0000000 %15457171% 850 87.15%05099511% 3513 95:75% 1 8 2 470036.74     %990干8000干（09.4士22.5519&#x27;6725&#x27;330 = w Xs =  (s) Ti-H6&#x27;0 = 1 O = P (s) T--H CC1 = yO1= P (SIA4） PIII-SA/CT1 =yFT= (sT) s1TI-sg = Y iniip pi P grippC =y[=1 (SIA) TI-SAS =yrrrpaee gaisnpa Seippegger nsereny iprgasereresteippeggPP=P4440</td><td rowspan=1 colspan=1>% 8030639%5500854% £9&#x27;0干 56:09% 1555585% 70&#x27;0干I0% 99055% 79&#x27;0LL% £90干 5969% 553515:396566% 1440444.94%5506554  %41:045.44       % 0105809% 89045:1989&#x27;0921% 42:015%0706/:66 277333755 %1/044:64% 71172% 010 1690% 3303.73.51% 17:5--55.85%%   %%119.03 30.67990干  5503066120.00555555830 =  Os = P (sI1I) Tt--AdL1HT&#x27;0 = 1s= P sa) T-H C=yE1=PIA S-ACC = y I= P （(I) sTTd-SHg = Y iniip pi P grippaggC1 =yT=1A） T-SNarrpiereisepaSerpggrhler errrairrireigrrsesereesseippeggz =pashisea4</td><td rowspan=1 colspan=1>% 77:045:14% 09044% 69000.14% 590F 50.00% 7/0511.79% 1/0 6777% LL&#x27;0干 18&#x27;94% 9/0 95.94Surererrer lseeetdnnODTLr) Giii-rriOrnerer eeeinerrceeret</td></tr></table>",
+                                            "type": "table",
+                                            "image_path": "80d028f66a313fb206276b503e8a1b093d0c96a5856ce2c601050fb245cae4f2.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 2,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        177,
+                                        170,
+                                        461,
+                                        644
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 1.25
+                }
+            ],
+            "page_idx": 20,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        107,
+                        27,
+                        293,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                26,
+                                294,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        26,
+                                        294,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Published as a conference paper at ICLR 2019",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        300,
+                        751,
+                        310,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                298,
+                                750,
+                                312,
+                                764
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        298,
+                                        750,
+                                        312,
+                                        764
+                                    ],
+                                    "score": 1.0,
+                                    "content": "",
+                                    "type": "text",
+                                    "height": 14,
+                                    "width": 14
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "table",
+                    "bbox": [
+                        177,
+                        170,
+                        461,
+                        644
+                    ],
+                    "blocks": [
+                        {
+                            "type": "table_footnote",
+                            "bbox": [
+                                140,
+                                99,
+                                165,
+                                729
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        150,
+                                        162,
+                                        167,
+                                        733
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                150,
+                                                162,
+                                                167,
+                                                733
+                                            ],
+                                            "score": 0.0,
+                                            "content": "",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        140,
+                                        95,
+                                        156,
+                                        733
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                140,
+                                                95,
+                                                156,
+                                                733
+                                            ],
+                                            "score": 1.0,
+                                            "content": "denotes a 95% confidence interval. d: dimensionality of embeddi",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 0
+                                }
+                            ],
+                            "index": 0.5
+                        },
+                        {
+                            "type": "table_body",
+                            "bbox": [
+                                177,
+                                170,
+                                461,
+                                644
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        177,
+                                        170,
+                                        461,
+                                        644
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                177,
+                                                170,
+                                                461,
+                                                644
+                                            ],
+                                            "score": 0.969,
+                                            "html": "<table><tr><td rowspan=1 colspan=1>(0s&#x27;s)(2 &#x27;0)(s‘s)(1‘s)(ous Xem)Woirai</td><td rowspan=1 colspan=1>% 5/.05 39.61% 533112% 9908589% 67003735Yrirlrriiir reirraperreeees</td><td rowspan=1 colspan=1>% 230309.3% 31:15:11% 1/.0 91&#x27;81%55333.99% 00&#x27;0干00&#x27;  % 89&#x27;0干 86&#x27;L% 89&#x27;0干 16&#x27;99% L9:0 17:99546%344231.21% 100404% 7/0 9000     %8900L&#x27;79%57003515% 5703.31%£9090:65%81096565% 0000000 %15457171% 850 87.15%05099511% 3513 95:75% 1 8 2 470036.74     %990干8000干（09.4士22.5519&#x27;6725&#x27;330 = w Xs =  (s) Ti-H6&#x27;0 = 1 O = P (s) T--H CC1 = yO1= P (SIA4） PIII-SA/CT1 =yFT= (sT) s1TI-sg = Y iniip pi P grippC =y[=1 (SIA) TI-SAS =yrrrpaee gaisnpa Seippegger nsereny iprgasereresteippeggPP=P4440</td><td rowspan=1 colspan=1>% 8030639%5500854% £9&#x27;0干 56:09% 1555585% 70&#x27;0干I0% 99055% 79&#x27;0LL% £90干 5969% 553515:396566% 1440444.94%5506554  %41:045.44       % 0105809% 89045:1989&#x27;0921% 42:015%0706/:66 277333755 %1/044:64% 71172% 010 1690% 3303.73.51% 17:5--55.85%%   %%119.03 30.67990干  5503066120.00555555830 =  Os = P (sI1I) Tt--AdL1HT&#x27;0 = 1s= P sa) T-H C=yE1=PIA S-ACC = y I= P （(I) sTTd-SHg = Y iniip pi P grippaggC1 =yT=1A） T-SNarrpiereisepaSerpggrhler errrairrireigrrsesereesseippeggz =pashisea4</td><td rowspan=1 colspan=1>% 77:045:14% 09044% 69000.14% 590F 50.00% 7/0511.79% 1/0 6777% LL&#x27;0干 18&#x27;94% 9/0 95.94Surererrer lseeetdnnODTLr) Giii-rriOrnerer eeeinerrceeret</td></tr></table>",
+                                            "type": "table",
+                                            "image_path": "80d028f66a313fb206276b503e8a1b093d0c96a5856ce2c601050fb245cae4f2.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 2,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        177,
+                                        170,
+                                        461,
+                                        644
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 1.25
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "table",
+                    "bbox": [
+                        182,
+                        311,
+                        428,
+                        539
+                    ],
+                    "blocks": [
+                        {
+                            "type": "table_caption",
+                            "bbox": [
+                                107,
+                                266,
+                                505,
+                                297
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        266,
+                                        505,
+                                        278
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                266,
+                                                398,
+                                                278
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Table 10: CelebA facial attribute classification results averaged over 1000 tasks.",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                398,
+                                                267,
+                                                408,
+                                                277
+                                            ],
+                                            "score": 0.73,
+                                            "content": "\\pm",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                409,
+                                                266,
+                                                444,
+                                                278
+                                            ],
+                                            "score": 1.0,
+                                            "content": "denotes a",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                445,
+                                                267,
+                                                462,
+                                                276
+                                            ],
+                                            "score": 0.86,
+                                            "content": "9 5 \\%",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                463,
+                                                266,
+                                                505,
+                                                278
+                                            ],
+                                            "score": 1.0,
+                                            "content": "confidence",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 0
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        276,
+                                        505,
+                                        287
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                276,
+                                                139,
+                                                287
+                                            ],
+                                            "score": 1.0,
+                                            "content": "interval.",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                140,
+                                                277,
+                                                146,
+                                                286
+                                            ],
+                                            "score": 0.59,
+                                            "content": "d",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                146,
+                                                276,
+                                                260,
+                                                287
+                                            ],
+                                            "score": 1.0,
+                                            "content": ": dimensionality of embedding,",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                261,
+                                                277,
+                                                267,
+                                                286
+                                            ],
+                                            "score": 0.62,
+                                            "content": "h",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                267,
+                                                276,
+                                                454,
+                                                287
+                                            ],
+                                            "score": 1.0,
+                                            "content": ": number of hidden units in a fully connected layer,",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                455,
+                                                277,
+                                                461,
+                                                286
+                                            ],
+                                            "score": 0.58,
+                                            "content": "k",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                461,
+                                                276,
+                                                505,
+                                                287
+                                            ],
+                                            "score": 1.0,
+                                            "content": ": number of",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        286,
+                                        373,
+                                        299
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                286,
+                                                186,
+                                                299
+                                            ],
+                                            "score": 1.0,
+                                            "content": "clusters in a partition,",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                187,
+                                                287,
+                                                195,
+                                                296
+                                            ],
+                                            "score": 0.76,
+                                            "content": "P",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                195,
+                                                286,
+                                                373,
+                                                299
+                                            ],
+                                            "score": 1.0,
+                                            "content": ": number of partitions used during meta-learning.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 2
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "type": "table_body",
+                            "bbox": [
+                                182,
+                                311,
+                                428,
+                                539
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        182,
+                                        311,
+                                        428,
+                                        539
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                182,
+                                                311,
+                                                428,
+                                                539
+                                            ],
+                                            "score": 0.979,
+                                            "html": "<table><tr><td>Algorithm</td><td>(2,5)</td></tr><tr><td>Baselines Training from scratch</td><td>63.19 ± 1.06 %</td></tr><tr><td>BiGAN,d= 200</td><td></td></tr><tr><td>Embedding knn-nearest neighbors Embedding linearclassifier</td><td>56.15 ± 0.89 % 58.44 ± 0.90 %</td></tr><tr><td>Embedding MLP with dropout,h = 128</td><td>56.26 ± 0.94 %</td></tr><tr><td>Embedding cluster matching,k = 500</td><td>56.20 ±1.00 %</td></tr><tr><td>CACTUs-MAML(ours),P= 50,k= 500 CACTUs-ProtoNets (ours),P= 50,k = 500</td><td>74.98 ± 1.02 %</td></tr><tr><td></td><td>65.58 ± 1.04 %</td></tr><tr><td>DeepCluster,d= 256 Embedding knn-nearest neighbors</td><td></td></tr><tr><td>Embedding linear classifier</td><td>61.47 ± 0.99 % 59.57 ± 0.98 %</td></tr><tr><td>Embedding MLP with dropout,h = 128</td><td>60.65 ± 0.98 %</td></tr><tr><td>Embedding cluster matching,k = 500</td><td>51.51 ± 0.89 %</td></tr><tr><td>CACTUs-MAML (ours),P= 50,k = 500</td><td>73.79 ± 1.01 %</td></tr><tr><td>CACTUs-ProtoNets (ours),P= 50,k = 500</td><td>74.15 ± 1.02 %</td></tr><tr><td></td><td></td></tr><tr><td>Supervised meta-learning</td><td></td></tr><tr><td>Oracle-MAML (control)</td><td>87.10 ± 0.85 %</td></tr><tr><td></td><td></td></tr><tr><td>Oracle-ProtoNets (control)</td><td>85.13 ± 0.92 %</td></tr></table>",
+                                            "type": "table",
+                                            "image_path": "660b21b7784e16644ff9108f25cb77fb320be6baea7dd2154f3fcf8d42bd91fb.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 12,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        182,
+                                        311,
+                                        428,
+                                        323.0
+                                    ],
+                                    "spans": [],
+                                    "index": 3
+                                },
+                                {
+                                    "bbox": [
+                                        182,
+                                        323.0,
+                                        428,
+                                        335.0
+                                    ],
+                                    "spans": [],
+                                    "index": 4
+                                },
+                                {
+                                    "bbox": [
+                                        182,
+                                        335.0,
+                                        428,
+                                        347.0
+                                    ],
+                                    "spans": [],
+                                    "index": 5
+                                },
+                                {
+                                    "bbox": [
+                                        182,
+                                        347.0,
+                                        428,
+                                        359.0
+                                    ],
+                                    "spans": [],
+                                    "index": 6
+                                },
+                                {
+                                    "bbox": [
+                                        182,
+                                        359.0,
+                                        428,
+                                        371.0
+                                    ],
+                                    "spans": [],
+                                    "index": 7
+                                },
+                                {
+                                    "bbox": [
+                                        182,
+                                        371.0,
+                                        428,
+                                        383.0
+                                    ],
+                                    "spans": [],
+                                    "index": 8
+                                },
+                                {
+                                    "bbox": [
+                                        182,
+                                        383.0,
+                                        428,
+                                        395.0
+                                    ],
+                                    "spans": [],
+                                    "index": 9
+                                },
+                                {
+                                    "bbox": [
+                                        182,
+                                        395.0,
+                                        428,
+                                        407.0
+                                    ],
+                                    "spans": [],
+                                    "index": 10
+                                },
+                                {
+                                    "bbox": [
+                                        182,
+                                        407.0,
+                                        428,
+                                        419.0
+                                    ],
+                                    "spans": [],
+                                    "index": 11
+                                },
+                                {
+                                    "bbox": [
+                                        182,
+                                        419.0,
+                                        428,
+                                        431.0
+                                    ],
+                                    "spans": [],
+                                    "index": 12
+                                },
+                                {
+                                    "bbox": [
+                                        182,
+                                        431.0,
+                                        428,
+                                        443.0
+                                    ],
+                                    "spans": [],
+                                    "index": 13
+                                },
+                                {
+                                    "bbox": [
+                                        182,
+                                        443.0,
+                                        428,
+                                        455.0
+                                    ],
+                                    "spans": [],
+                                    "index": 14
+                                },
+                                {
+                                    "bbox": [
+                                        182,
+                                        455.0,
+                                        428,
+                                        467.0
+                                    ],
+                                    "spans": [],
+                                    "index": 15
+                                },
+                                {
+                                    "bbox": [
+                                        182,
+                                        467.0,
+                                        428,
+                                        479.0
+                                    ],
+                                    "spans": [],
+                                    "index": 16
+                                },
+                                {
+                                    "bbox": [
+                                        182,
+                                        479.0,
+                                        428,
+                                        491.0
+                                    ],
+                                    "spans": [],
+                                    "index": 17
+                                },
+                                {
+                                    "bbox": [
+                                        182,
+                                        491.0,
+                                        428,
+                                        503.0
+                                    ],
+                                    "spans": [],
+                                    "index": 18
+                                },
+                                {
+                                    "bbox": [
+                                        182,
+                                        503.0,
+                                        428,
+                                        515.0
+                                    ],
+                                    "spans": [],
+                                    "index": 19
+                                },
+                                {
+                                    "bbox": [
+                                        182,
+                                        515.0,
+                                        428,
+                                        527.0
+                                    ],
+                                    "spans": [],
+                                    "index": 20
+                                },
+                                {
+                                    "bbox": [
+                                        182,
+                                        527.0,
+                                        428,
+                                        539.0
+                                    ],
+                                    "spans": [],
+                                    "index": 21
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 6.5
+                }
+            ],
+            "page_idx": 21,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        107,
+                        27,
+                        293,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                26,
+                                294,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        26,
+                                        294,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Published as a conference paper at ICLR 2019",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        300,
+                        751,
+                        311,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                298,
+                                750,
+                                313,
+                                765
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        298,
+                                        750,
+                                        313,
+                                        765
+                                    ],
+                                    "score": 1.0,
+                                    "content": "",
+                                    "type": "text",
+                                    "height": 15,
+                                    "width": 15
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "table",
+                    "bbox": [
+                        182,
+                        311,
+                        428,
+                        539
+                    ],
+                    "blocks": [
+                        {
+                            "type": "table_caption",
+                            "bbox": [
+                                107,
+                                266,
+                                505,
+                                297
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        266,
+                                        505,
+                                        278
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                266,
+                                                398,
+                                                278
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Table 10: CelebA facial attribute classification results averaged over 1000 tasks.",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                398,
+                                                267,
+                                                408,
+                                                277
+                                            ],
+                                            "score": 0.73,
+                                            "content": "\\pm",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                409,
+                                                266,
+                                                444,
+                                                278
+                                            ],
+                                            "score": 1.0,
+                                            "content": "denotes a",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                445,
+                                                267,
+                                                462,
+                                                276
+                                            ],
+                                            "score": 0.86,
+                                            "content": "9 5 \\%",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                463,
+                                                266,
+                                                505,
+                                                278
+                                            ],
+                                            "score": 1.0,
+                                            "content": "confidence",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 0
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        276,
+                                        505,
+                                        287
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                276,
+                                                139,
+                                                287
+                                            ],
+                                            "score": 1.0,
+                                            "content": "interval.",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                140,
+                                                277,
+                                                146,
+                                                286
+                                            ],
+                                            "score": 0.59,
+                                            "content": "d",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                146,
+                                                276,
+                                                260,
+                                                287
+                                            ],
+                                            "score": 1.0,
+                                            "content": ": dimensionality of embedding,",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                261,
+                                                277,
+                                                267,
+                                                286
+                                            ],
+                                            "score": 0.62,
+                                            "content": "h",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                267,
+                                                276,
+                                                454,
+                                                287
+                                            ],
+                                            "score": 1.0,
+                                            "content": ": number of hidden units in a fully connected layer,",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                455,
+                                                277,
+                                                461,
+                                                286
+                                            ],
+                                            "score": 0.58,
+                                            "content": "k",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                461,
+                                                276,
+                                                505,
+                                                287
+                                            ],
+                                            "score": 1.0,
+                                            "content": ": number of",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        286,
+                                        373,
+                                        299
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                286,
+                                                186,
+                                                299
+                                            ],
+                                            "score": 1.0,
+                                            "content": "clusters in a partition,",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                187,
+                                                287,
+                                                195,
+                                                296
+                                            ],
+                                            "score": 0.76,
+                                            "content": "P",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                195,
+                                                286,
+                                                373,
+                                                299
+                                            ],
+                                            "score": 1.0,
+                                            "content": ": number of partitions used during meta-learning.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 2
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "type": "table_body",
+                            "bbox": [
+                                182,
+                                311,
+                                428,
+                                539
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        182,
+                                        311,
+                                        428,
+                                        539
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                182,
+                                                311,
+                                                428,
+                                                539
+                                            ],
+                                            "score": 0.979,
+                                            "html": "<table><tr><td>Algorithm</td><td>(2,5)</td></tr><tr><td>Baselines Training from scratch</td><td>63.19 ± 1.06 %</td></tr><tr><td>BiGAN,d= 200</td><td></td></tr><tr><td>Embedding knn-nearest neighbors Embedding linearclassifier</td><td>56.15 ± 0.89 % 58.44 ± 0.90 %</td></tr><tr><td>Embedding MLP with dropout,h = 128</td><td>56.26 ± 0.94 %</td></tr><tr><td>Embedding cluster matching,k = 500</td><td>56.20 ±1.00 %</td></tr><tr><td>CACTUs-MAML(ours),P= 50,k= 500 CACTUs-ProtoNets (ours),P= 50,k = 500</td><td>74.98 ± 1.02 %</td></tr><tr><td></td><td>65.58 ± 1.04 %</td></tr><tr><td>DeepCluster,d= 256 Embedding knn-nearest neighbors</td><td></td></tr><tr><td>Embedding linear classifier</td><td>61.47 ± 0.99 % 59.57 ± 0.98 %</td></tr><tr><td>Embedding MLP with dropout,h = 128</td><td>60.65 ± 0.98 %</td></tr><tr><td>Embedding cluster matching,k = 500</td><td>51.51 ± 0.89 %</td></tr><tr><td>CACTUs-MAML (ours),P= 50,k = 500</td><td>73.79 ± 1.01 %</td></tr><tr><td>CACTUs-ProtoNets (ours),P= 50,k = 500</td><td>74.15 ± 1.02 %</td></tr><tr><td></td><td></td></tr><tr><td>Supervised meta-learning</td><td></td></tr><tr><td>Oracle-MAML (control)</td><td>87.10 ± 0.85 %</td></tr><tr><td></td><td></td></tr><tr><td>Oracle-ProtoNets (control)</td><td>85.13 ± 0.92 %</td></tr></table>",
+                                            "type": "table",
+                                            "image_path": "660b21b7784e16644ff9108f25cb77fb320be6baea7dd2154f3fcf8d42bd91fb.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 12,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        182,
+                                        311,
+                                        428,
+                                        323.0
+                                    ],
+                                    "spans": [],
+                                    "index": 3
+                                },
+                                {
+                                    "bbox": [
+                                        182,
+                                        323.0,
+                                        428,
+                                        335.0
+                                    ],
+                                    "spans": [],
+                                    "index": 4
+                                },
+                                {
+                                    "bbox": [
+                                        182,
+                                        335.0,
+                                        428,
+                                        347.0
+                                    ],
+                                    "spans": [],
+                                    "index": 5
+                                },
+                                {
+                                    "bbox": [
+                                        182,
+                                        347.0,
+                                        428,
+                                        359.0
+                                    ],
+                                    "spans": [],
+                                    "index": 6
+                                },
+                                {
+                                    "bbox": [
+                                        182,
+                                        359.0,
+                                        428,
+                                        371.0
+                                    ],
+                                    "spans": [],
+                                    "index": 7
+                                },
+                                {
+                                    "bbox": [
+                                        182,
+                                        371.0,
+                                        428,
+                                        383.0
+                                    ],
+                                    "spans": [],
+                                    "index": 8
+                                },
+                                {
+                                    "bbox": [
+                                        182,
+                                        383.0,
+                                        428,
+                                        395.0
+                                    ],
+                                    "spans": [],
+                                    "index": 9
+                                },
+                                {
+                                    "bbox": [
+                                        182,
+                                        395.0,
+                                        428,
+                                        407.0
+                                    ],
+                                    "spans": [],
+                                    "index": 10
+                                },
+                                {
+                                    "bbox": [
+                                        182,
+                                        407.0,
+                                        428,
+                                        419.0
+                                    ],
+                                    "spans": [],
+                                    "index": 11
+                                },
+                                {
+                                    "bbox": [
+                                        182,
+                                        419.0,
+                                        428,
+                                        431.0
+                                    ],
+                                    "spans": [],
+                                    "index": 12
+                                },
+                                {
+                                    "bbox": [
+                                        182,
+                                        431.0,
+                                        428,
+                                        443.0
+                                    ],
+                                    "spans": [],
+                                    "index": 13
+                                },
+                                {
+                                    "bbox": [
+                                        182,
+                                        443.0,
+                                        428,
+                                        455.0
+                                    ],
+                                    "spans": [],
+                                    "index": 14
+                                },
+                                {
+                                    "bbox": [
+                                        182,
+                                        455.0,
+                                        428,
+                                        467.0
+                                    ],
+                                    "spans": [],
+                                    "index": 15
+                                },
+                                {
+                                    "bbox": [
+                                        182,
+                                        467.0,
+                                        428,
+                                        479.0
+                                    ],
+                                    "spans": [],
+                                    "index": 16
+                                },
+                                {
+                                    "bbox": [
+                                        182,
+                                        479.0,
+                                        428,
+                                        491.0
+                                    ],
+                                    "spans": [],
+                                    "index": 17
+                                },
+                                {
+                                    "bbox": [
+                                        182,
+                                        491.0,
+                                        428,
+                                        503.0
+                                    ],
+                                    "spans": [],
+                                    "index": 18
+                                },
+                                {
+                                    "bbox": [
+                                        182,
+                                        503.0,
+                                        428,
+                                        515.0
+                                    ],
+                                    "spans": [],
+                                    "index": 19
+                                },
+                                {
+                                    "bbox": [
+                                        182,
+                                        515.0,
+                                        428,
+                                        527.0
+                                    ],
+                                    "spans": [],
+                                    "index": 20
+                                },
+                                {
+                                    "bbox": [
+                                        182,
+                                        527.0,
+                                        428,
+                                        539.0
+                                    ],
+                                    "spans": [],
+                                    "index": 21
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 6.5
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        81,
+                        319,
+                        94
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                79,
+                                321,
+                                96
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        79,
+                                        321,
+                                        96
+                                    ],
+                                    "score": 1.0,
+                                    "content": "APPENDIX G IMAGENET EXPERIMENTS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        }
+                    ],
+                    "index": 0
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        106,
+                        505,
+                        183
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                105,
+                                506,
+                                120
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        105,
+                                        506,
+                                        120
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We investigate unsupervised meta-learning in the context of a larger unsupervised meta-training",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                117,
+                                506,
+                                129
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        117,
+                                        506,
+                                        129
+                                    ],
+                                    "score": 1.0,
+                                    "content": "dataset by using the ILSVRC 2012 dataset’s training split (Russakovsky et al., 2015), which is a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                127,
+                                505,
+                                142
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        127,
+                                        505,
+                                        142
+                                    ],
+                                    "score": 1.0,
+                                    "content": "superset of the miniImageNet dataset (including meta-validation and meta-testing data) consisting",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                139,
+                                505,
+                                152
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        139,
+                                        505,
+                                        152
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of 1000 classes and over 1,200,000 images. To facilitate comparison to the previous miniImageNet",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                150,
+                                505,
+                                163
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        150,
+                                        505,
+                                        163
+                                    ],
+                                    "score": 1.0,
+                                    "content": "experiments, for meta-validation and meta-test we use the miniImageNet meta-validation and meta-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                162,
+                                505,
+                                173
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        162,
+                                        505,
+                                        173
+                                    ],
+                                    "score": 1.0,
+                                    "content": "test splits. To avoid task leakage, we hold out all data from these 36 underlying classes from the rest",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                171,
+                                293,
+                                184
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        171,
+                                        293,
+                                        184
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of the data to construct the meta-training split.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        }
+                    ],
+                    "index": 4
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        189,
+                        505,
+                        364
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                188,
+                                504,
+                                202
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        188,
+                                        504,
+                                        202
+                                    ],
+                                    "score": 1.0,
+                                    "content": "For CACTUs, we use the best-performing unsupervised learning method from the previous experi-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                200,
+                                505,
+                                212
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        200,
+                                        505,
+                                        212
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ments, DeepCluster, to obtain the embeddings. Following Caron et al. (2018), we run DeepCluster",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                210,
+                                506,
+                                223
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        210,
+                                        506,
+                                        223
+                                    ],
+                                    "score": 1.0,
+                                    "content": "using the VGG-16 architecture with a 256-dimensional feature space and 10,000 clusters on the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                222,
+                                506,
+                                235
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        222,
+                                        506,
+                                        235
+                                    ],
+                                    "score": 1.0,
+                                    "content": "meta-training data until the normalized mutual information between the data-cluster mappings of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                233,
+                                505,
+                                245
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        233,
+                                        505,
+                                        245
+                                    ],
+                                    "score": 1.0,
+                                    "content": "two consecutive epochs converges. To our knowledge, no prior works have yet been published on",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                243,
+                                504,
+                                255
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        243,
+                                        504,
+                                        255
+                                    ],
+                                    "score": 1.0,
+                                    "content": "using MAML for ImageNet-sized meta-learning. We extend the standard convolutional neural net-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                254,
+                                505,
+                                268
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        254,
+                                        505,
+                                        268
+                                    ],
+                                    "score": 1.0,
+                                    "content": "work model class with residual connections (He et al., 2016), validate hyperparameters with super-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                266,
+                                504,
+                                278
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        266,
+                                        504,
+                                        278
+                                    ],
+                                    "score": 1.0,
+                                    "content": "vised meta-learning, then use it for unsupervised meta-learning without further tuning. See Table 11",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                277,
+                                504,
+                                289
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        277,
+                                        379,
+                                        289
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for MAML hyperparameters. The training from scratch, embedding",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        379,
+                                        277,
+                                        393,
+                                        288
+                                    ],
+                                    "score": 0.89,
+                                    "content": "k _ { \\mathrm { n n } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        393,
+                                        277,
+                                        504,
+                                        289
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-nearest neighbors, and em-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                287,
+                                505,
+                                300
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        287,
+                                        505,
+                                        300
+                                    ],
+                                    "score": 1.0,
+                                    "content": "bedding linear classifier algorithms are the same as they were in the previous sets of experiments.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                298,
+                                505,
+                                312
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        298,
+                                        505,
+                                        312
+                                    ],
+                                    "score": 1.0,
+                                    "content": "For Oracle-MAML, we generated tasks using the ground-truth 964 ImageNet meta-training classes.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                309,
+                                505,
+                                322
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        309,
+                                        505,
+                                        322
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We also run semi-supervised MAML, with the meta-training tasks consisting of CACTUs-based",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                320,
+                                505,
+                                333
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        320,
+                                        505,
+                                        333
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tasks as well as tasks constructed from the 64 miniImageNet meta-training classes. The unsuper-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                331,
+                                506,
+                                344
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        331,
+                                        506,
+                                        344
+                                    ],
+                                    "score": 1.0,
+                                    "content": "vised/supervised task proportion split was fixed according to the ratio of the number of data available",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                343,
+                                505,
+                                354
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        343,
+                                        505,
+                                        354
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to each task proposal method. As before, the meta-learning methods only meta-learned on 1-shot",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                353,
+                                133,
+                                365
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        353,
+                                        133,
+                                        365
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tasks.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        }
+                    ],
+                    "index": 15.5
+                },
+                {
+                    "type": "table",
+                    "bbox": [
+                        199,
+                        390,
+                        413,
+                        546
+                    ],
+                    "blocks": [
+                        {
+                            "type": "table_caption",
+                            "bbox": [
+                                199,
+                                374,
+                                411,
+                                385
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        198,
+                                        372,
+                                        412,
+                                        387
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                198,
+                                                372,
+                                                412,
+                                                387
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Table 11: MAML hyperparameter summary for ImageNet.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 24
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "type": "table_body",
+                            "bbox": [
+                                199,
+                                390,
+                                413,
+                                546
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        199,
+                                        390,
+                                        413,
+                                        546
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                199,
+                                                390,
+                                                413,
+                                                546
+                                            ],
+                                            "score": 0.981,
+                                            "html": "<table><tr><td>Hyperparameter</td><td>Value</td></tr><tr><td>Input size</td><td>224× 224</td></tr><tr><td>Outer (meta) learning rate</td><td>0.0001</td></tr><tr><td>Inner learning rate</td><td>0.001</td></tr><tr><td>Task batch size</td><td>3</td></tr><tr><td>Inner adaptation steps (meta-training)</td><td>5</td></tr><tr><td>Meta-training iterations</td><td>240,000</td></tr><tr><td>Adaptation steps (evaluation)</td><td>100</td></tr><tr><td>Classes per task (meta-training)</td><td>5</td></tr><tr><td>Shots per class (meta-training)</td><td>1</td></tr><tr><td>Queries per class</td><td>5</td></tr><tr><td>Residual blocks</td><td>5</td></tr><tr><td>Layers per residual block</td><td>2</td></tr></table>",
+                                            "type": "table",
+                                            "image_path": "080631d0533a0a33810053381fc1eed4a2b60fc845717b42c6eee8481e50188e.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 30.5,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        199,
+                                        390,
+                                        413,
+                                        403.0
+                                    ],
+                                    "spans": [],
+                                    "index": 25
+                                },
+                                {
+                                    "bbox": [
+                                        199,
+                                        403.0,
+                                        413,
+                                        416.0
+                                    ],
+                                    "spans": [],
+                                    "index": 26
+                                },
+                                {
+                                    "bbox": [
+                                        199,
+                                        416.0,
+                                        413,
+                                        429.0
+                                    ],
+                                    "spans": [],
+                                    "index": 27
+                                },
+                                {
+                                    "bbox": [
+                                        199,
+                                        429.0,
+                                        413,
+                                        442.0
+                                    ],
+                                    "spans": [],
+                                    "index": 28
+                                },
+                                {
+                                    "bbox": [
+                                        199,
+                                        442.0,
+                                        413,
+                                        455.0
+                                    ],
+                                    "spans": [],
+                                    "index": 29
+                                },
+                                {
+                                    "bbox": [
+                                        199,
+                                        455.0,
+                                        413,
+                                        468.0
+                                    ],
+                                    "spans": [],
+                                    "index": 30
+                                },
+                                {
+                                    "bbox": [
+                                        199,
+                                        468.0,
+                                        413,
+                                        481.0
+                                    ],
+                                    "spans": [],
+                                    "index": 31
+                                },
+                                {
+                                    "bbox": [
+                                        199,
+                                        481.0,
+                                        413,
+                                        494.0
+                                    ],
+                                    "spans": [],
+                                    "index": 32
+                                },
+                                {
+                                    "bbox": [
+                                        199,
+                                        494.0,
+                                        413,
+                                        507.0
+                                    ],
+                                    "spans": [],
+                                    "index": 33
+                                },
+                                {
+                                    "bbox": [
+                                        199,
+                                        507.0,
+                                        413,
+                                        520.0
+                                    ],
+                                    "spans": [],
+                                    "index": 34
+                                },
+                                {
+                                    "bbox": [
+                                        199,
+                                        520.0,
+                                        413,
+                                        533.0
+                                    ],
+                                    "spans": [],
+                                    "index": 35
+                                },
+                                {
+                                    "bbox": [
+                                        199,
+                                        533.0,
+                                        413,
+                                        546.0
+                                    ],
+                                    "spans": [],
+                                    "index": 36
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 27.25
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        561,
+                        505,
+                        726
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                561,
+                                505,
+                                574
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        561,
+                                        505,
+                                        574
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We find that the vastly increased amount of unlabeled meta-training data (in comparison to miniIm-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                572,
+                                506,
+                                585
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        572,
+                                        506,
+                                        585
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ageNet) results in significant increases for all methods over their counterparts in Table 9 (other than",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                582,
+                                505,
+                                597
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        582,
+                                        505,
+                                        597
+                                    ],
+                                    "score": 1.0,
+                                    "content": "training from scratch, which does not use this data). We find that CACTUs-MAML slightly outper-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                595,
+                                504,
+                                606
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        595,
+                                        504,
+                                        606
+                                    ],
+                                    "score": 1.0,
+                                    "content": "forms embedding linear classifier for the 1-shot test tasks, but that the linear classifier on top of the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                606,
+                                505,
+                                618
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        606,
+                                        505,
+                                        618
+                                    ],
+                                    "score": 1.0,
+                                    "content": "unsupervised embedding becomes better as the amount of test time supervision increases. Augment-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                617,
+                                505,
+                                629
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        617,
+                                        505,
+                                        629
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ing the unsupervised tasks with (a small number of) supervised tasks during meta-training results",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                627,
+                                506,
+                                640
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        627,
+                                        506,
+                                        640
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in slight improvement for the 1-shot test tasks. The lackluster performance of CACTUs-MAML is",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                639,
+                                505,
+                                651
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        639,
+                                        505,
+                                        651
+                                    ],
+                                    "score": 1.0,
+                                    "content": "unsurprising insofar as meta-learning with large task spaces is still an open problem: higher shot",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                648,
+                                506,
+                                663
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        648,
+                                        506,
+                                        663
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Oracle-MAML only marginally stays ahead of the embedding linear classifier, which is not the case",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                660,
+                                506,
+                                673
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        660,
+                                        506,
+                                        673
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in the other, smaller-scale experiments. We expect that using a larger architecture in conjunction",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                672,
+                                505,
+                                683
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        672,
+                                        505,
+                                        683
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with MAML (such as Kim et al. (2018)) would result in increased performance for all methods",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 47
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                682,
+                                505,
+                                694
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        682,
+                                        505,
+                                        694
+                                    ],
+                                    "score": 1.0,
+                                    "content": "based on MAML. Further, given the extensive degree to which unsupervised learning methods have",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 48
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                694,
+                                504,
+                                705
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        694,
+                                        504,
+                                        705
+                                    ],
+                                    "score": 1.0,
+                                    "content": "been studied, we suspect that unsupervised task construction coupled with better meta-learning al-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 49
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                704,
+                                505,
+                                718
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        704,
+                                        505,
+                                        718
+                                    ],
+                                    "score": 1.0,
+                                    "content": "gorithms and architectures will result in improved performance on the entire unsupervised learning",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 50
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                715,
+                                320,
+                                727
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        715,
+                                        320,
+                                        727
+                                    ],
+                                    "score": 1.0,
+                                    "content": "problem. We leave such investigation to future work.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 51
+                        }
+                    ],
+                    "index": 44
+                }
+            ],
+            "page_idx": 22,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        107,
+                        27,
+                        293,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                26,
+                                294,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        26,
+                                        294,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Published as a conference paper at ICLR 2019",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        300,
+                        751,
+                        311,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                298,
+                                750,
+                                313,
+                                764
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        298,
+                                        750,
+                                        313,
+                                        764
+                                    ],
+                                    "score": 1.0,
+                                    "content": "",
+                                    "type": "text",
+                                    "height": 14,
+                                    "width": 15
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        81,
+                        319,
+                        94
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                79,
+                                321,
+                                96
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        79,
+                                        321,
+                                        96
+                                    ],
+                                    "score": 1.0,
+                                    "content": "APPENDIX G IMAGENET EXPERIMENTS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        }
+                    ],
+                    "index": 0
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        106,
+                        505,
+                        183
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                105,
+                                506,
+                                120
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        105,
+                                        506,
+                                        120
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We investigate unsupervised meta-learning in the context of a larger unsupervised meta-training",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                117,
+                                506,
+                                129
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        117,
+                                        506,
+                                        129
+                                    ],
+                                    "score": 1.0,
+                                    "content": "dataset by using the ILSVRC 2012 dataset’s training split (Russakovsky et al., 2015), which is a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                127,
+                                505,
+                                142
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        127,
+                                        505,
+                                        142
+                                    ],
+                                    "score": 1.0,
+                                    "content": "superset of the miniImageNet dataset (including meta-validation and meta-testing data) consisting",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                139,
+                                505,
+                                152
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        139,
+                                        505,
+                                        152
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of 1000 classes and over 1,200,000 images. To facilitate comparison to the previous miniImageNet",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                150,
+                                505,
+                                163
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        150,
+                                        505,
+                                        163
+                                    ],
+                                    "score": 1.0,
+                                    "content": "experiments, for meta-validation and meta-test we use the miniImageNet meta-validation and meta-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                162,
+                                505,
+                                173
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        162,
+                                        505,
+                                        173
+                                    ],
+                                    "score": 1.0,
+                                    "content": "test splits. To avoid task leakage, we hold out all data from these 36 underlying classes from the rest",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                171,
+                                293,
+                                184
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        171,
+                                        293,
+                                        184
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of the data to construct the meta-training split.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        }
+                    ],
+                    "index": 4,
+                    "bbox_fs": [
+                        105,
+                        105,
+                        506,
+                        184
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        189,
+                        505,
+                        364
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                188,
+                                504,
+                                202
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        188,
+                                        504,
+                                        202
+                                    ],
+                                    "score": 1.0,
+                                    "content": "For CACTUs, we use the best-performing unsupervised learning method from the previous experi-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                200,
+                                505,
+                                212
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        200,
+                                        505,
+                                        212
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ments, DeepCluster, to obtain the embeddings. Following Caron et al. (2018), we run DeepCluster",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                210,
+                                506,
+                                223
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        210,
+                                        506,
+                                        223
+                                    ],
+                                    "score": 1.0,
+                                    "content": "using the VGG-16 architecture with a 256-dimensional feature space and 10,000 clusters on the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                222,
+                                506,
+                                235
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        222,
+                                        506,
+                                        235
+                                    ],
+                                    "score": 1.0,
+                                    "content": "meta-training data until the normalized mutual information between the data-cluster mappings of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                233,
+                                505,
+                                245
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        233,
+                                        505,
+                                        245
+                                    ],
+                                    "score": 1.0,
+                                    "content": "two consecutive epochs converges. To our knowledge, no prior works have yet been published on",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                243,
+                                504,
+                                255
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        243,
+                                        504,
+                                        255
+                                    ],
+                                    "score": 1.0,
+                                    "content": "using MAML for ImageNet-sized meta-learning. We extend the standard convolutional neural net-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                254,
+                                505,
+                                268
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        254,
+                                        505,
+                                        268
+                                    ],
+                                    "score": 1.0,
+                                    "content": "work model class with residual connections (He et al., 2016), validate hyperparameters with super-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                266,
+                                504,
+                                278
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        266,
+                                        504,
+                                        278
+                                    ],
+                                    "score": 1.0,
+                                    "content": "vised meta-learning, then use it for unsupervised meta-learning without further tuning. See Table 11",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                277,
+                                504,
+                                289
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        277,
+                                        379,
+                                        289
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for MAML hyperparameters. The training from scratch, embedding",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        379,
+                                        277,
+                                        393,
+                                        288
+                                    ],
+                                    "score": 0.89,
+                                    "content": "k _ { \\mathrm { n n } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        393,
+                                        277,
+                                        504,
+                                        289
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-nearest neighbors, and em-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                287,
+                                505,
+                                300
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        287,
+                                        505,
+                                        300
+                                    ],
+                                    "score": 1.0,
+                                    "content": "bedding linear classifier algorithms are the same as they were in the previous sets of experiments.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                298,
+                                505,
+                                312
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        298,
+                                        505,
+                                        312
+                                    ],
+                                    "score": 1.0,
+                                    "content": "For Oracle-MAML, we generated tasks using the ground-truth 964 ImageNet meta-training classes.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                309,
+                                505,
+                                322
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        309,
+                                        505,
+                                        322
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We also run semi-supervised MAML, with the meta-training tasks consisting of CACTUs-based",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                320,
+                                505,
+                                333
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        320,
+                                        505,
+                                        333
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tasks as well as tasks constructed from the 64 miniImageNet meta-training classes. The unsuper-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                331,
+                                506,
+                                344
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        331,
+                                        506,
+                                        344
+                                    ],
+                                    "score": 1.0,
+                                    "content": "vised/supervised task proportion split was fixed according to the ratio of the number of data available",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                343,
+                                505,
+                                354
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        343,
+                                        505,
+                                        354
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to each task proposal method. As before, the meta-learning methods only meta-learned on 1-shot",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                353,
+                                133,
+                                365
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        353,
+                                        133,
+                                        365
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tasks.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        }
+                    ],
+                    "index": 15.5,
+                    "bbox_fs": [
+                        104,
+                        188,
+                        506,
+                        365
+                    ]
+                },
+                {
+                    "type": "table",
+                    "bbox": [
+                        199,
+                        390,
+                        413,
+                        546
+                    ],
+                    "blocks": [
+                        {
+                            "type": "table_caption",
+                            "bbox": [
+                                199,
+                                374,
+                                411,
+                                385
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        198,
+                                        372,
+                                        412,
+                                        387
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                198,
+                                                372,
+                                                412,
+                                                387
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Table 11: MAML hyperparameter summary for ImageNet.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 24
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "type": "table_body",
+                            "bbox": [
+                                199,
+                                390,
+                                413,
+                                546
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        199,
+                                        390,
+                                        413,
+                                        546
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                199,
+                                                390,
+                                                413,
+                                                546
+                                            ],
+                                            "score": 0.981,
+                                            "html": "<table><tr><td>Hyperparameter</td><td>Value</td></tr><tr><td>Input size</td><td>224× 224</td></tr><tr><td>Outer (meta) learning rate</td><td>0.0001</td></tr><tr><td>Inner learning rate</td><td>0.001</td></tr><tr><td>Task batch size</td><td>3</td></tr><tr><td>Inner adaptation steps (meta-training)</td><td>5</td></tr><tr><td>Meta-training iterations</td><td>240,000</td></tr><tr><td>Adaptation steps (evaluation)</td><td>100</td></tr><tr><td>Classes per task (meta-training)</td><td>5</td></tr><tr><td>Shots per class (meta-training)</td><td>1</td></tr><tr><td>Queries per class</td><td>5</td></tr><tr><td>Residual blocks</td><td>5</td></tr><tr><td>Layers per residual block</td><td>2</td></tr></table>",
+                                            "type": "table",
+                                            "image_path": "080631d0533a0a33810053381fc1eed4a2b60fc845717b42c6eee8481e50188e.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 30.5,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        199,
+                                        390,
+                                        413,
+                                        403.0
+                                    ],
+                                    "spans": [],
+                                    "index": 25
+                                },
+                                {
+                                    "bbox": [
+                                        199,
+                                        403.0,
+                                        413,
+                                        416.0
+                                    ],
+                                    "spans": [],
+                                    "index": 26
+                                },
+                                {
+                                    "bbox": [
+                                        199,
+                                        416.0,
+                                        413,
+                                        429.0
+                                    ],
+                                    "spans": [],
+                                    "index": 27
+                                },
+                                {
+                                    "bbox": [
+                                        199,
+                                        429.0,
+                                        413,
+                                        442.0
+                                    ],
+                                    "spans": [],
+                                    "index": 28
+                                },
+                                {
+                                    "bbox": [
+                                        199,
+                                        442.0,
+                                        413,
+                                        455.0
+                                    ],
+                                    "spans": [],
+                                    "index": 29
+                                },
+                                {
+                                    "bbox": [
+                                        199,
+                                        455.0,
+                                        413,
+                                        468.0
+                                    ],
+                                    "spans": [],
+                                    "index": 30
+                                },
+                                {
+                                    "bbox": [
+                                        199,
+                                        468.0,
+                                        413,
+                                        481.0
+                                    ],
+                                    "spans": [],
+                                    "index": 31
+                                },
+                                {
+                                    "bbox": [
+                                        199,
+                                        481.0,
+                                        413,
+                                        494.0
+                                    ],
+                                    "spans": [],
+                                    "index": 32
+                                },
+                                {
+                                    "bbox": [
+                                        199,
+                                        494.0,
+                                        413,
+                                        507.0
+                                    ],
+                                    "spans": [],
+                                    "index": 33
+                                },
+                                {
+                                    "bbox": [
+                                        199,
+                                        507.0,
+                                        413,
+                                        520.0
+                                    ],
+                                    "spans": [],
+                                    "index": 34
+                                },
+                                {
+                                    "bbox": [
+                                        199,
+                                        520.0,
+                                        413,
+                                        533.0
+                                    ],
+                                    "spans": [],
+                                    "index": 35
+                                },
+                                {
+                                    "bbox": [
+                                        199,
+                                        533.0,
+                                        413,
+                                        546.0
+                                    ],
+                                    "spans": [],
+                                    "index": 36
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 27.25
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        561,
+                        505,
+                        726
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                561,
+                                505,
+                                574
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        561,
+                                        505,
+                                        574
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We find that the vastly increased amount of unlabeled meta-training data (in comparison to miniIm-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                572,
+                                506,
+                                585
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        572,
+                                        506,
+                                        585
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ageNet) results in significant increases for all methods over their counterparts in Table 9 (other than",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                582,
+                                505,
+                                597
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        582,
+                                        505,
+                                        597
+                                    ],
+                                    "score": 1.0,
+                                    "content": "training from scratch, which does not use this data). We find that CACTUs-MAML slightly outper-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                595,
+                                504,
+                                606
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        595,
+                                        504,
+                                        606
+                                    ],
+                                    "score": 1.0,
+                                    "content": "forms embedding linear classifier for the 1-shot test tasks, but that the linear classifier on top of the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                606,
+                                505,
+                                618
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        606,
+                                        505,
+                                        618
+                                    ],
+                                    "score": 1.0,
+                                    "content": "unsupervised embedding becomes better as the amount of test time supervision increases. Augment-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                617,
+                                505,
+                                629
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        617,
+                                        505,
+                                        629
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ing the unsupervised tasks with (a small number of) supervised tasks during meta-training results",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                627,
+                                506,
+                                640
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        627,
+                                        506,
+                                        640
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in slight improvement for the 1-shot test tasks. The lackluster performance of CACTUs-MAML is",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                639,
+                                505,
+                                651
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        639,
+                                        505,
+                                        651
+                                    ],
+                                    "score": 1.0,
+                                    "content": "unsurprising insofar as meta-learning with large task spaces is still an open problem: higher shot",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                648,
+                                506,
+                                663
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        648,
+                                        506,
+                                        663
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Oracle-MAML only marginally stays ahead of the embedding linear classifier, which is not the case",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                660,
+                                506,
+                                673
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        660,
+                                        506,
+                                        673
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in the other, smaller-scale experiments. We expect that using a larger architecture in conjunction",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                672,
+                                505,
+                                683
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        672,
+                                        505,
+                                        683
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with MAML (such as Kim et al. (2018)) would result in increased performance for all methods",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 47
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                682,
+                                505,
+                                694
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        682,
+                                        505,
+                                        694
+                                    ],
+                                    "score": 1.0,
+                                    "content": "based on MAML. Further, given the extensive degree to which unsupervised learning methods have",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 48
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                694,
+                                504,
+                                705
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        694,
+                                        504,
+                                        705
+                                    ],
+                                    "score": 1.0,
+                                    "content": "been studied, we suspect that unsupervised task construction coupled with better meta-learning al-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 49
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                704,
+                                505,
+                                718
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        704,
+                                        505,
+                                        718
+                                    ],
+                                    "score": 1.0,
+                                    "content": "gorithms and architectures will result in improved performance on the entire unsupervised learning",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 50
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                715,
+                                320,
+                                727
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        715,
+                                        320,
+                                        727
+                                    ],
+                                    "score": 1.0,
+                                    "content": "problem. We leave such investigation to future work.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 51
+                        }
+                    ],
+                    "index": 44,
+                    "bbox_fs": [
+                        105,
+                        561,
+                        506,
+                        727
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "table",
+                    "bbox": [
+                        245,
+                        158,
+                        392,
+                        658
+                    ],
+                    "blocks": [
+                        {
+                            "type": "table_caption",
+                            "bbox": [
+                                209,
+                                108,
+                                233,
+                                743
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        219,
+                                        365,
+                                        235,
+                                        733
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                219,
+                                                365,
+                                                235,
+                                                733
+                                            ],
+                                            "score": 0.0,
+                                            "content": "",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 2
+                                },
+                                {
+                                    "bbox": [
+                                        210,
+                                        105,
+                                        223,
+                                        732
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                210,
+                                                105,
+                                                223,
+                                                732
+                                            ],
+                                            "score": 1.0,
+                                            "content": " denotes a 95% confidence interval. d: dimensionality of\r\n",
+                                            "type": "text",
+                                            "height": 627,
+                                            "width": 13
+                                        }
+                                    ],
+                                    "index": 0
+                                }
+                            ],
+                            "index": 1.0
+                        },
+                        {
+                            "type": "table_body",
+                            "bbox": [
+                                245,
+                                158,
+                                392,
+                                658
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        245,
+                                        158,
+                                        392,
+                                        658
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                245,
+                                                158,
+                                                392,
+                                                658
+                                            ],
+                                            "score": 0.925,
+                                            "html": "<table><tr><td>(os&#x27;s) （07&#x27;s) (s‘s) (1‘s) (ous xem) 1loirrra</td><td>% LL0干 89&#x27;6S% SL0干 L9&#x27;εS % 3901/30% 11:059:85 Ppt =pessstsesg Yrrlrriiir liiiiera reneseg</td><td>% 41:0 4244 % L70 干 0106 % 3705 75.77 % 040 18.61 % 130 11:11 % 990F 5878 % 17034355 % 00 1064 % 1/0494 % I9&#x27;0 干 LI&#x27;01 % 7703181 % 4101109 [[V[ = γ[=1 ‘(SJ1) TII-SALO0 ohhgrer serrerarg nrippegg garreerrnrilpestg</td><td>% 190  L728 % 690£8&#x27;78 %0/0£494 % SL&#x27;0 干 SL&#x27;19 Sururersritu prseetdeesntts Jra psrsN-</td><td>% 1/0 £7&#x27;98 % L/071:4 Srinerniar peseadn DEDLLrr Tirr-larrt</td><td>% 990 干 90℃6% 69&#x27;0干4116</td></tr></table>",
+                                            "type": "table",
+                                            "image_path": "246c7ecf053e28afcdae6dee0f1be59ff6fc7f6487369b9082e633dbd7cb3d14.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 1,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        245,
+                                        158,
+                                        392,
+                                        658
+                                    ],
+                                    "spans": [],
+                                    "index": 1
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 1.0
+                }
+            ],
+            "page_idx": 23,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        107,
+                        27,
+                        293,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                26,
+                                293,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        26,
+                                        293,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Published as a conference paper at ICLR 2019",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        300,
+                        751,
+                        311,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                298,
+                                750,
+                                313,
+                                764
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        298,
+                                        750,
+                                        313,
+                                        764
+                                    ],
+                                    "score": 1.0,
+                                    "content": "",
+                                    "type": "text",
+                                    "height": 14,
+                                    "width": 15
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "table",
+                    "bbox": [
+                        245,
+                        158,
+                        392,
+                        658
+                    ],
+                    "blocks": [
+                        {
+                            "type": "table_caption",
+                            "bbox": [
+                                209,
+                                108,
+                                233,
+                                743
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        219,
+                                        365,
+                                        235,
+                                        733
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                219,
+                                                365,
+                                                235,
+                                                733
+                                            ],
+                                            "score": 0.0,
+                                            "content": "",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 2
+                                },
+                                {
+                                    "bbox": [
+                                        210,
+                                        105,
+                                        223,
+                                        732
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                210,
+                                                105,
+                                                223,
+                                                732
+                                            ],
+                                            "score": 1.0,
+                                            "content": " denotes a 95% confidence interval. d: dimensionality of\r\n",
+                                            "type": "text",
+                                            "height": 627,
+                                            "width": 13
+                                        }
+                                    ],
+                                    "index": 0
+                                }
+                            ],
+                            "index": 1.0
+                        },
+                        {
+                            "type": "table_body",
+                            "bbox": [
+                                245,
+                                158,
+                                392,
+                                658
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        245,
+                                        158,
+                                        392,
+                                        658
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                245,
+                                                158,
+                                                392,
+                                                658
+                                            ],
+                                            "score": 0.925,
+                                            "html": "<table><tr><td>(os&#x27;s) （07&#x27;s) (s‘s) (1‘s) (ous xem) 1loirrra</td><td>% LL0干 89&#x27;6S% SL0干 L9&#x27;εS % 3901/30% 11:059:85 Ppt =pessstsesg Yrrlrriiir liiiiera reneseg</td><td>% 41:0 4244 % L70 干 0106 % 3705 75.77 % 040 18.61 % 130 11:11 % 990F 5878 % 17034355 % 00 1064 % 1/0494 % I9&#x27;0 干 LI&#x27;01 % 7703181 % 4101109 [[V[ = γ[=1 ‘(SJ1) TII-SALO0 ohhgrer serrerarg nrippegg garreerrnrilpestg</td><td>% 190  L728 % 690£8&#x27;78 %0/0£494 % SL&#x27;0 干 SL&#x27;19 Sururersritu prseetdeesntts Jra psrsN-</td><td>% 1/0 £7&#x27;98 % L/071:4 Srinerniar peseadn DEDLLrr Tirr-larrt</td><td>% 990 干 90℃6% 69&#x27;0干4116</td></tr></table>",
+                                            "type": "table",
+                                            "image_path": "246c7ecf053e28afcdae6dee0f1be59ff6fc7f6487369b9082e633dbd7cb3d14.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 1,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        245,
+                                        158,
+                                        392,
+                                        658
+                                    ],
+                                    "spans": [],
+                                    "index": 1
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 1.0
+                }
+            ]
+        }
+    ],
+    "_backend": "pipeline",
+    "_version_name": "2.2.2"
+}
\ No newline at end of file
diff --git a/parse/train/r1My6sR9tX/r1My6sR9tX_model.json b/parse/train/r1My6sR9tX/r1My6sR9tX_model.json
new file mode 100644
index 0000000000000000000000000000000000000000..3f6817d84718ee821e182f75268dfa450ba30fe0
--- /dev/null
+++ b/parse/train/r1My6sR9tX/r1My6sR9tX_model.json
@@ -0,0 +1,24421 @@
+[
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1140,
+                    1403,
+                    1140,
+                    1403,
+                    1415,
+                    297,
+                    1415
+                ],
+                "score": 0.982
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    398,
+                    563,
+                    1302,
+                    563,
+                    1302,
+                    1018,
+                    398,
+                    1018
+                ],
+                "score": 0.98
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1432,
+                    1403,
+                    1432,
+                    1403,
+                    1583,
+                    298,
+                    1583
+                ],
+                "score": 0.977
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1601,
+                    1403,
+                    1601,
+                    1403,
+                    1813,
+                    298,
+                    1813
+                ],
+                "score": 0.975
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    1830,
+                    1402,
+                    1830,
+                    1402,
+                    1984,
+                    299,
+                    1984
+                ],
+                "score": 0.97
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    314,
+                    319,
+                    732,
+                    319,
+                    732,
+                    412,
+                    314,
+                    412
+                ],
+                "score": 0.943
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    782,
+                    321,
+                    1363,
+                    321,
+                    1363,
+                    413,
+                    782,
+                    413
+                ],
+                "score": 0.941
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    297,
+                    219,
+                    1312,
+                    219,
+                    1312,
+                    268,
+                    297,
+                    268
+                ],
+                "score": 0.928
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    328,
+                    2005,
+                    1179,
+                    2005,
+                    1179,
+                    2033,
+                    328,
+                    2033
+                ],
+                "score": 0.904
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    302,
+                    1080,
+                    573,
+                    1080,
+                    573,
+                    1114,
+                    302,
+                    1114
+                ],
+                "score": 0.898
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    773,
+                    494,
+                    926,
+                    494,
+                    926,
+                    527,
+                    773,
+                    527
+                ],
+                "score": 0.878
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    299,
+                    76,
+                    816,
+                    76,
+                    816,
+                    104,
+                    299,
+                    104
+                ],
+                "score": 0.874
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    842,
+                    2089,
+                    857,
+                    2089,
+                    857,
+                    2112,
+                    842,
+                    2112
+                ],
+                "score": 0.668
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    217.0,
+                    1317.0,
+                    217.0,
+                    1317.0,
+                    275.0,
+                    295.0,
+                    275.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    329.0,
+                    1997.0,
+                    1183.0,
+                    1997.0,
+                    1183.0,
+                    2042.0,
+                    329.0,
+                    2042.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1076.0,
+                    579.0,
+                    1076.0,
+                    579.0,
+                    1122.0,
+                    294.0,
+                    1122.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    769.0,
+                    493.0,
+                    932.0,
+                    493.0,
+                    932.0,
+                    530.0,
+                    769.0,
+                    530.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    73.0,
+                    817.0,
+                    73.0,
+                    817.0,
+                    108.0,
+                    295.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    841.0,
+                    2088.0,
+                    860.0,
+                    2088.0,
+                    860.0,
+                    2118.0,
+                    841.0,
+                    2118.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1138.0,
+                    1403.0,
+                    1138.0,
+                    1403.0,
+                    1176.0,
+                    295.0,
+                    1176.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1171.0,
+                    1403.0,
+                    1171.0,
+                    1403.0,
+                    1206.0,
+                    293.0,
+                    1206.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1198.0,
+                    1407.0,
+                    1198.0,
+                    1407.0,
+                    1241.0,
+                    291.0,
+                    1241.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1232.0,
+                    1403.0,
+                    1232.0,
+                    1403.0,
+                    1268.0,
+                    293.0,
+                    1268.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1262.0,
+                    1405.0,
+                    1262.0,
+                    1405.0,
+                    1298.0,
+                    293.0,
+                    1298.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1292.0,
+                    1405.0,
+                    1292.0,
+                    1405.0,
+                    1328.0,
+                    293.0,
+                    1328.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1323.0,
+                    1405.0,
+                    1323.0,
+                    1405.0,
+                    1360.0,
+                    293.0,
+                    1360.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1349.0,
+                    1405.0,
+                    1349.0,
+                    1405.0,
+                    1393.0,
+                    291.0,
+                    1393.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1384.0,
+                    1182.0,
+                    1384.0,
+                    1182.0,
+                    1421.0,
+                    295.0,
+                    1421.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    395.0,
+                    562.0,
+                    1304.0,
+                    562.0,
+                    1304.0,
+                    597.0,
+                    395.0,
+                    597.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    393.0,
+                    591.0,
+                    1304.0,
+                    591.0,
+                    1304.0,
+                    627.0,
+                    393.0,
+                    627.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    394.0,
+                    622.0,
+                    1306.0,
+                    622.0,
+                    1306.0,
+                    657.0,
+                    394.0,
+                    657.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    393.0,
+                    653.0,
+                    1306.0,
+                    653.0,
+                    1306.0,
+                    687.0,
+                    393.0,
+                    687.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    394.0,
+                    685.0,
+                    1305.0,
+                    685.0,
+                    1305.0,
+                    717.0,
+                    394.0,
+                    717.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    392.0,
+                    714.0,
+                    1307.0,
+                    714.0,
+                    1307.0,
+                    749.0,
+                    392.0,
+                    749.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    395.0,
+                    746.0,
+                    1306.0,
+                    746.0,
+                    1306.0,
+                    777.0,
+                    395.0,
+                    777.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    393.0,
+                    774.0,
+                    1306.0,
+                    774.0,
+                    1306.0,
+                    809.0,
+                    393.0,
+                    809.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    394.0,
+                    806.0,
+                    1306.0,
+                    806.0,
+                    1306.0,
+                    840.0,
+                    394.0,
+                    840.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    392.0,
+                    834.0,
+                    1306.0,
+                    834.0,
+                    1306.0,
+                    871.0,
+                    392.0,
+                    871.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    395.0,
+                    867.0,
+                    1304.0,
+                    867.0,
+                    1304.0,
+                    899.0,
+                    395.0,
+                    899.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    394.0,
+                    897.0,
+                    1306.0,
+                    897.0,
+                    1306.0,
+                    931.0,
+                    394.0,
+                    931.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    393.0,
+                    926.0,
+                    1306.0,
+                    926.0,
+                    1306.0,
+                    961.0,
+                    393.0,
+                    961.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    393.0,
+                    958.0,
+                    1306.0,
+                    958.0,
+                    1306.0,
+                    993.0,
+                    393.0,
+                    993.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    394.0,
+                    988.0,
+                    904.0,
+                    988.0,
+                    904.0,
+                    1023.0,
+                    394.0,
+                    1023.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    1432.0,
+                    1404.0,
+                    1432.0,
+                    1404.0,
+                    1464.0,
+                    297.0,
+                    1464.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1460.0,
+                    1405.0,
+                    1460.0,
+                    1405.0,
+                    1497.0,
+                    294.0,
+                    1497.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1492.0,
+                    1405.0,
+                    1492.0,
+                    1405.0,
+                    1528.0,
+                    293.0,
+                    1528.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1519.0,
+                    1407.0,
+                    1519.0,
+                    1407.0,
+                    1561.0,
+                    291.0,
+                    1561.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1553.0,
+                    1104.0,
+                    1553.0,
+                    1104.0,
+                    1587.0,
+                    294.0,
+                    1587.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1597.0,
+                    1408.0,
+                    1597.0,
+                    1408.0,
+                    1639.0,
+                    292.0,
+                    1639.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1630.0,
+                    1404.0,
+                    1630.0,
+                    1404.0,
+                    1664.0,
+                    294.0,
+                    1664.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1659.0,
+                    1407.0,
+                    1659.0,
+                    1407.0,
+                    1697.0,
+                    292.0,
+                    1697.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1694.0,
+                    1405.0,
+                    1694.0,
+                    1405.0,
+                    1724.0,
+                    293.0,
+                    1724.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1722.0,
+                    1404.0,
+                    1722.0,
+                    1404.0,
+                    1756.0,
+                    294.0,
+                    1756.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1755.0,
+                    1404.0,
+                    1755.0,
+                    1404.0,
+                    1785.0,
+                    296.0,
+                    1785.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1785.0,
+                    871.0,
+                    1785.0,
+                    871.0,
+                    1815.0,
+                    296.0,
+                    1815.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1828.0,
+                    1406.0,
+                    1828.0,
+                    1406.0,
+                    1863.0,
+                    294.0,
+                    1863.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1857.0,
+                    1404.0,
+                    1857.0,
+                    1404.0,
+                    1896.0,
+                    293.0,
+                    1896.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1887.0,
+                    1406.0,
+                    1887.0,
+                    1406.0,
+                    1927.0,
+                    293.0,
+                    1927.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1916.0,
+                    1406.0,
+                    1916.0,
+                    1406.0,
+                    1960.0,
+                    292.0,
+                    1960.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1952.0,
+                    1406.0,
+                    1952.0,
+                    1406.0,
+                    1986.0,
+                    293.0,
+                    1986.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    309.0,
+                    317.0,
+                    440.0,
+                    317.0,
+                    440.0,
+                    353.0,
+                    309.0,
+                    353.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    311.0,
+                    348.0,
+                    562.0,
+                    348.0,
+                    562.0,
+                    386.0,
+                    311.0,
+                    386.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    313.0,
+                    383.0,
+                    735.0,
+                    383.0,
+                    735.0,
+                    415.0,
+                    313.0,
+                    415.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    782.0,
+                    321.0,
+                    1124.0,
+                    321.0,
+                    1124.0,
+                    354.0,
+                    782.0,
+                    354.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    781.0,
+                    348.0,
+                    1168.0,
+                    348.0,
+                    1168.0,
+                    386.0,
+                    781.0,
+                    386.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    784.0,
+                    382.0,
+                    1365.0,
+                    382.0,
+                    1365.0,
+                    417.0,
+                    784.0,
+                    417.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 0,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    522,
+                    1404,
+                    522,
+                    1404,
+                    979,
+                    297,
+                    979
+                ],
+                "score": 0.983
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1728,
+                    1403,
+                    1728,
+                    1403,
+                    2034,
+                    298,
+                    2034
+                ],
+                "score": 0.982
+            },
+            {
+                "category_id": 4,
+                "poly": [
+                    296,
+                    1318,
+                    1405,
+                    1318,
+                    1405,
+                    1432,
+                    296,
+                    1432
+                ],
+                "score": 0.953
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    291,
+                    1573,
+                    1399,
+                    1573,
+                    1399,
+                    1636,
+                    291,
+                    1636
+                ],
+                "score": 0.95
+            },
+            {
+                "category_id": 3,
+                "poly": [
+                    328,
+                    1017,
+                    1358,
+                    1017,
+                    1358,
+                    1304,
+                    328,
+                    1304
+                ],
+                "score": 0.949
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    229,
+                    1402,
+                    229,
+                    1402,
+                    506,
+                    298,
+                    506
+                ],
+                "score": 0.947
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    299,
+                    1506,
+                    833,
+                    1506,
+                    833,
+                    1541,
+                    299,
+                    1541
+                ],
+                "score": 0.926
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    300,
+                    76,
+                    815,
+                    76,
+                    815,
+                    104,
+                    300,
+                    104
+                ],
+                "score": 0.893
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    300,
+                    1675,
+                    632,
+                    1675,
+                    632,
+                    1705,
+                    300,
+                    1705
+                ],
+                "score": 0.892
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    841,
+                    2088,
+                    858,
+                    2088,
+                    858,
+                    2112,
+                    841,
+                    2112
+                ],
+                "score": 0.702
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    841,
+                    2088,
+                    859,
+                    2088,
+                    859,
+                    2112,
+                    841,
+                    2112
+                ],
+                "score": 0.176
+            },
+            {
+                "category_id": 4,
+                "poly": [
+                    681,
+                    1022,
+                    1135,
+                    1022,
+                    1135,
+                    1045,
+                    681,
+                    1045
+                ],
+                "score": 0.122
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    419,
+                    2003,
+                    515,
+                    2003,
+                    515,
+                    2036,
+                    419,
+                    2036
+                ],
+                "score": 0.93,
+                "latex": "\\left( \\mathbf { x } _ { k } , \\mathbf { y } _ { k } \\right)"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    975,
+                    1759,
+                    1091,
+                    1759,
+                    1091,
+                    1793,
+                    975,
+                    1793
+                ],
+                "score": 0.93,
+                "latex": "\\mathcal { D } = \\{ \\mathbf { x } _ { i } \\}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    929,
+                    1375,
+                    1039,
+                    1375,
+                    1039,
+                    1403,
+                    929,
+                    1403
+                ],
+                "score": 0.92,
+                "latex": "K _ { \\mathrm { m - t r } } = 1"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    698,
+                    1375,
+                    780,
+                    1375,
+                    780,
+                    1400,
+                    698,
+                    1400
+                ],
+                "score": 0.9,
+                "latex": "N = 2"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    906,
+                    2009,
+                    933,
+                    2009,
+                    933,
+                    2030,
+                    906,
+                    2030
+                ],
+                "score": 0.87,
+                "latex": "K"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    340,
+                    1973,
+                    373,
+                    1973,
+                    373,
+                    2001,
+                    340,
+                    2001
+                ],
+                "score": 0.87,
+                "latex": "M"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    715,
+                    1945,
+                    740,
+                    1945,
+                    740,
+                    1969,
+                    715,
+                    1969
+                ],
+                "score": 0.83,
+                "latex": "\\mathcal { D }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    774,
+                    323,
+                    793,
+                    323,
+                    793,
+                    349,
+                    774,
+                    349
+                ],
+                "score": 0.82,
+                "latex": "k"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    485,
+                    1348,
+                    502,
+                    1348,
+                    502,
+                    1372,
+                    485,
+                    1372
+                ],
+                "score": 0.82,
+                "latex": "k"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    705,
+                    2004,
+                    738,
+                    2004,
+                    738,
+                    2030,
+                    705,
+                    2030
+                ],
+                "score": 0.8,
+                "latex": "M"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1283,
+                    1974,
+                    1313,
+                    1974,
+                    1313,
+                    2001,
+                    1283,
+                    2001
+                ],
+                "score": 0.79,
+                "latex": "K"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    298.0,
+                    1320.0,
+                    1403.0,
+                    1320.0,
+                    1403.0,
+                    1350.0,
+                    298.0,
+                    1350.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1346.0,
+                    484.0,
+                    1346.0,
+                    484.0,
+                    1379.0,
+                    295.0,
+                    1379.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    503.0,
+                    1346.0,
+                    1405.0,
+                    1346.0,
+                    1405.0,
+                    1379.0,
+                    503.0,
+                    1379.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1370.0,
+                    697.0,
+                    1370.0,
+                    697.0,
+                    1410.0,
+                    293.0,
+                    1410.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    781.0,
+                    1370.0,
+                    928.0,
+                    1370.0,
+                    928.0,
+                    1410.0,
+                    781.0,
+                    1410.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1040.0,
+                    1370.0,
+                    1407.0,
+                    1370.0,
+                    1407.0,
+                    1410.0,
+                    1040.0,
+                    1410.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1403.0,
+                    1387.0,
+                    1403.0,
+                    1387.0,
+                    1433.0,
+                    296.0,
+                    1433.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    325.0,
+                    1014.0,
+                    558.0,
+                    1014.0,
+                    558.0,
+                    1049.0,
+                    325.0,
+                    1049.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    682.0,
+                    1019.0,
+                    1138.0,
+                    1019.0,
+                    1138.0,
+                    1046.0,
+                    682.0,
+                    1046.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    324.0,
+                    1053.0,
+                    596.0,
+                    1053.0,
+                    596.0,
+                    1131.0,
+                    324.0,
+                    1131.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    608.0,
+                    1058.0,
+                    638.0,
+                    1058.0,
+                    638.0,
+                    1087.0,
+                    608.0,
+                    1087.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1196.0,
+                    1094.0,
+                    1223.0,
+                    1094.0,
+                    1223.0,
+                    1111.0,
+                    1196.0,
+                    1111.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    344.0,
+                    1121.0,
+                    402.0,
+                    1121.0,
+                    402.0,
+                    1159.0,
+                    344.0,
+                    1159.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    598.0,
+                    1119.0,
+                    657.0,
+                    1119.0,
+                    657.0,
+                    1159.0,
+                    598.0,
+                    1159.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    708.0,
+                    1113.0,
+                    723.0,
+                    1113.0,
+                    723.0,
+                    1127.0,
+                    708.0,
+                    1127.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    807.0,
+                    1104.0,
+                    857.0,
+                    1104.0,
+                    857.0,
+                    1140.0,
+                    807.0,
+                    1140.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    325.0,
+                    1153.0,
+                    658.0,
+                    1153.0,
+                    658.0,
+                    1183.0,
+                    325.0,
+                    1183.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    677.0,
+                    1154.0,
+                    937.0,
+                    1154.0,
+                    937.0,
+                    1185.0,
+                    677.0,
+                    1185.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    335.0,
+                    1185.0,
+                    440.0,
+                    1185.0,
+                    440.0,
+                    1285.0,
+                    335.0,
+                    1285.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    450.0,
+                    1225.0,
+                    459.0,
+                    1225.0,
+                    459.0,
+                    1234.0,
+                    450.0,
+                    1234.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    476.0,
+                    1227.0,
+                    502.0,
+                    1227.0,
+                    502.0,
+                    1269.0,
+                    476.0,
+                    1269.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    510.0,
+                    1187.0,
+                    575.0,
+                    1187.0,
+                    575.0,
+                    1238.0,
+                    510.0,
+                    1238.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    802.0,
+                    1225.0,
+                    812.0,
+                    1225.0,
+                    812.0,
+                    1234.0,
+                    802.0,
+                    1234.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    834.0,
+                    1223.0,
+                    872.0,
+                    1223.0,
+                    872.0,
+                    1248.0,
+                    834.0,
+                    1248.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    927.0,
+                    1221.0,
+                    941.0,
+                    1221.0,
+                    941.0,
+                    1235.0,
+                    927.0,
+                    1235.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    944.0,
+                    1221.0,
+                    984.0,
+                    1221.0,
+                    984.0,
+                    1238.0,
+                    944.0,
+                    1238.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    986.0,
+                    1207.0,
+                    1142.0,
+                    1207.0,
+                    1142.0,
+                    1235.0,
+                    986.0,
+                    1235.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1158.0,
+                    1210.0,
+                    1355.0,
+                    1210.0,
+                    1355.0,
+                    1242.0,
+                    1158.0,
+                    1242.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    377.0,
+                    1275.0,
+                    407.0,
+                    1275.0,
+                    407.0,
+                    1303.0,
+                    377.0,
+                    1303.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    506.0,
+                    1275.0,
+                    543.0,
+                    1275.0,
+                    543.0,
+                    1308.0,
+                    506.0,
+                    1308.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    726.0,
+                    1275.0,
+                    761.0,
+                    1275.0,
+                    761.0,
+                    1309.0,
+                    726.0,
+                    1309.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    854.0,
+                    1275.0,
+                    890.0,
+                    1275.0,
+                    890.0,
+                    1309.0,
+                    854.0,
+                    1309.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    591.25,
+                    1078.5,
+                    660.25,
+                    1078.5,
+                    660.25,
+                    1117.5,
+                    591.25,
+                    1117.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1501.0,
+                    836.0,
+                    1501.0,
+                    836.0,
+                    1548.0,
+                    291.0,
+                    1548.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    73.0,
+                    817.0,
+                    73.0,
+                    817.0,
+                    108.0,
+                    296.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1674.0,
+                    636.0,
+                    1674.0,
+                    636.0,
+                    1709.0,
+                    294.0,
+                    1709.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    839.0,
+                    2085.0,
+                    862.0,
+                    2085.0,
+                    862.0,
+                    2120.0,
+                    839.0,
+                    2120.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    838.0,
+                    2085.0,
+                    862.0,
+                    2085.0,
+                    862.0,
+                    2120.0,
+                    838.0,
+                    2120.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    680.0,
+                    1017.0,
+                    1138.0,
+                    1017.0,
+                    1138.0,
+                    1049.0,
+                    680.0,
+                    1049.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    522.0,
+                    1405.0,
+                    522.0,
+                    1405.0,
+                    557.0,
+                    295.0,
+                    557.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    550.0,
+                    1405.0,
+                    550.0,
+                    1405.0,
+                    589.0,
+                    291.0,
+                    589.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    583.0,
+                    1404.0,
+                    583.0,
+                    1404.0,
+                    617.0,
+                    295.0,
+                    617.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    612.0,
+                    1404.0,
+                    612.0,
+                    1404.0,
+                    647.0,
+                    295.0,
+                    647.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    641.0,
+                    1404.0,
+                    641.0,
+                    1404.0,
+                    677.0,
+                    292.0,
+                    677.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    673.0,
+                    1404.0,
+                    673.0,
+                    1404.0,
+                    708.0,
+                    292.0,
+                    708.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    702.0,
+                    1405.0,
+                    702.0,
+                    1405.0,
+                    741.0,
+                    294.0,
+                    741.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    733.0,
+                    1405.0,
+                    733.0,
+                    1405.0,
+                    770.0,
+                    292.0,
+                    770.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    763.0,
+                    1405.0,
+                    763.0,
+                    1405.0,
+                    800.0,
+                    294.0,
+                    800.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    797.0,
+                    1404.0,
+                    797.0,
+                    1404.0,
+                    831.0,
+                    295.0,
+                    831.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    824.0,
+                    1405.0,
+                    824.0,
+                    1405.0,
+                    862.0,
+                    291.0,
+                    862.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    855.0,
+                    1404.0,
+                    855.0,
+                    1404.0,
+                    893.0,
+                    294.0,
+                    893.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    886.0,
+                    1404.0,
+                    886.0,
+                    1404.0,
+                    921.0,
+                    292.0,
+                    921.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    916.0,
+                    1407.0,
+                    916.0,
+                    1407.0,
+                    953.0,
+                    292.0,
+                    953.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    944.0,
+                    1311.0,
+                    944.0,
+                    1311.0,
+                    984.0,
+                    294.0,
+                    984.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1726.0,
+                    1402.0,
+                    1726.0,
+                    1402.0,
+                    1762.0,
+                    296.0,
+                    1762.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1757.0,
+                    974.0,
+                    1757.0,
+                    974.0,
+                    1795.0,
+                    292.0,
+                    1795.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1092.0,
+                    1757.0,
+                    1405.0,
+                    1757.0,
+                    1405.0,
+                    1795.0,
+                    1092.0,
+                    1795.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1787.0,
+                    1405.0,
+                    1787.0,
+                    1405.0,
+                    1825.0,
+                    293.0,
+                    1825.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1820.0,
+                    1405.0,
+                    1820.0,
+                    1405.0,
+                    1856.0,
+                    294.0,
+                    1856.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1852.0,
+                    1404.0,
+                    1852.0,
+                    1404.0,
+                    1884.0,
+                    293.0,
+                    1884.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1884.0,
+                    1404.0,
+                    1884.0,
+                    1404.0,
+                    1915.0,
+                    294.0,
+                    1915.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1911.0,
+                    1407.0,
+                    1911.0,
+                    1407.0,
+                    1946.0,
+                    293.0,
+                    1946.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1942.0,
+                    714.0,
+                    1942.0,
+                    714.0,
+                    1977.0,
+                    294.0,
+                    1977.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    741.0,
+                    1942.0,
+                    1405.0,
+                    1942.0,
+                    1405.0,
+                    1977.0,
+                    741.0,
+                    1977.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1971.0,
+                    339.0,
+                    1971.0,
+                    339.0,
+                    2005.0,
+                    293.0,
+                    2005.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    374.0,
+                    1971.0,
+                    1282.0,
+                    1971.0,
+                    1282.0,
+                    2005.0,
+                    374.0,
+                    2005.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1314.0,
+                    1971.0,
+                    1405.0,
+                    1971.0,
+                    1405.0,
+                    2005.0,
+                    1314.0,
+                    2005.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    2000.0,
+                    418.0,
+                    2000.0,
+                    418.0,
+                    2038.0,
+                    293.0,
+                    2038.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    516.0,
+                    2000.0,
+                    704.0,
+                    2000.0,
+                    704.0,
+                    2038.0,
+                    516.0,
+                    2038.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    739.0,
+                    2000.0,
+                    905.0,
+                    2000.0,
+                    905.0,
+                    2038.0,
+                    739.0,
+                    2038.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    934.0,
+                    2000.0,
+                    1404.0,
+                    2000.0,
+                    1404.0,
+                    2038.0,
+                    934.0,
+                    2038.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1573.0,
+                    1402.0,
+                    1573.0,
+                    1402.0,
+                    1608.0,
+                    294.0,
+                    1608.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1605.0,
+                    1185.0,
+                    1605.0,
+                    1185.0,
+                    1639.0,
+                    295.0,
+                    1639.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    230.0,
+                    1407.0,
+                    230.0,
+                    1407.0,
+                    267.0,
+                    294.0,
+                    267.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    261.0,
+                    1406.0,
+                    261.0,
+                    1406.0,
+                    298.0,
+                    294.0,
+                    298.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    293.0,
+                    1404.0,
+                    293.0,
+                    1404.0,
+                    326.0,
+                    296.0,
+                    326.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    322.0,
+                    773.0,
+                    322.0,
+                    773.0,
+                    358.0,
+                    294.0,
+                    358.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    794.0,
+                    322.0,
+                    1406.0,
+                    322.0,
+                    1406.0,
+                    358.0,
+                    794.0,
+                    358.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    352.0,
+                    1404.0,
+                    352.0,
+                    1404.0,
+                    388.0,
+                    294.0,
+                    388.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    381.0,
+                    1404.0,
+                    381.0,
+                    1404.0,
+                    420.0,
+                    292.0,
+                    420.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    413.0,
+                    1404.0,
+                    413.0,
+                    1404.0,
+                    450.0,
+                    294.0,
+                    450.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    442.0,
+                    1404.0,
+                    442.0,
+                    1404.0,
+                    480.0,
+                    293.0,
+                    480.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    476.0,
+                    935.0,
+                    476.0,
+                    935.0,
+                    509.0,
+                    296.0,
+                    509.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 1,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    1393,
+                    1403,
+                    1393,
+                    1403,
+                    1638,
+                    299,
+                    1638
+                ],
+                "score": 0.984
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1758,
+                    1404,
+                    1758,
+                    1404,
+                    2034,
+                    298,
+                    2034
+                ],
+                "score": 0.982
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    229,
+                    1403,
+                    229,
+                    1403,
+                    475,
+                    298,
+                    475
+                ],
+                "score": 0.982
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    872,
+                    1404,
+                    872,
+                    1404,
+                    1117,
+                    298,
+                    1117
+                ],
+                "score": 0.98
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1132,
+                    1403,
+                    1132,
+                    1403,
+                    1378,
+                    298,
+                    1378
+                ],
+                "score": 0.98
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    566,
+                    1402,
+                    566,
+                    1402,
+                    719,
+                    299,
+                    719
+                ],
+                "score": 0.978
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    733,
+                    1403,
+                    733,
+                    1403,
+                    856,
+                    299,
+                    856
+                ],
+                "score": 0.977
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    300,
+                    76,
+                    815,
+                    76,
+                    815,
+                    104,
+                    300,
+                    104
+                ],
+                "score": 0.892
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    299,
+                    513,
+                    556,
+                    513,
+                    556,
+                    544,
+                    299,
+                    544
+                ],
+                "score": 0.887
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    300,
+                    1706,
+                    1322,
+                    1706,
+                    1322,
+                    1737,
+                    300,
+                    1737
+                ],
+                "score": 0.724
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    841,
+                    2088,
+                    858,
+                    2088,
+                    858,
+                    2112,
+                    841,
+                    2112
+                ],
+                "score": 0.596
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    842,
+                    2088,
+                    859,
+                    2088,
+                    859,
+                    2112,
+                    842,
+                    2112
+                ],
+                "score": 0.431
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    300,
+                    1706,
+                    1322,
+                    1706,
+                    1322,
+                    1737,
+                    300,
+                    1737
+                ],
+                "score": 0.157
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    479,
+                    1423,
+                    594,
+                    1423,
+                    594,
+                    1457,
+                    479,
+                    1457
+                ],
+                "score": 0.94,
+                "latex": "\\left\\{ \\left( \\mathbf { x } _ { i } , \\mathbf { y } _ { i } \\right) \\right\\}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    841,
+                    596,
+                    970,
+                    596,
+                    970,
+                    630,
+                    841,
+                    630
+                ],
+                "score": 0.93,
+                "latex": "\\mathcal { D } = \\left\\{ \\mathbf { x } _ { i } \\right\\}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    397,
+                    1972,
+                    513,
+                    1972,
+                    513,
+                    2006,
+                    397,
+                    2006
+                ],
+                "score": 0.93,
+                "latex": "\\mathcal { P } = \\{ \\mathcal { C } _ { c } \\}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    298,
+                    765,
+                    420,
+                    765,
+                    420,
+                    799,
+                    298,
+                    799
+                ],
+                "score": 0.93,
+                "latex": "\\{ ( \\mathbf { x } _ { k } , \\ell _ { k } ) \\}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1049,
+                    1911,
+                    1175,
+                    1911,
+                    1175,
+                    1945,
+                    1049,
+                    1945
+                ],
+                "score": 0.93,
+                "latex": "\\mathcal { D } = \\left\\{ \\mathbf { x } _ { i } \\right\\}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    663,
+                    1515,
+                    737,
+                    1515,
+                    737,
+                    1549,
+                    663,
+                    1549
+                ],
+                "score": 0.93,
+                "latex": "\\{ { \\bf { x } } _ { r } \\} _ { n }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    457,
+                    904,
+                    512,
+                    904,
+                    512,
+                    937,
+                    457,
+                    937
+                ],
+                "score": 0.92,
+                "latex": "\\{ \\mathcal { T } _ { t } \\}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    440,
+                    626,
+                    493,
+                    626,
+                    493,
+                    660,
+                    440,
+                    660
+                ],
+                "score": 0.92,
+                "latex": "\\left\\{ \\mathbf { z } _ { i } \\right\\}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    571,
+                    1973,
+                    627,
+                    1973,
+                    627,
+                    2006,
+                    571,
+                    2006
+                ],
+                "score": 0.92,
+                "latex": "\\{ { \\bf { x } } _ { i } \\}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1153,
+                    994,
+                    1207,
+                    994,
+                    1207,
+                    1028,
+                    1153,
+                    1028
+                ],
+                "score": 0.92,
+                "latex": "\\{ \\mathcal { T } _ { t } \\}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    474,
+                    1576,
+                    862,
+                    1576,
+                    862,
+                    1611,
+                    474,
+                    1611
+                ],
+                "score": 0.92,
+                "latex": "\\mathcal { T } = \\{ ( \\mathbf { x } _ { n , r } , \\ell _ { n } ) ~ | ~ \\mathbf { x } _ { n , r } \\in \\{ \\mathbf { x } _ { r } \\} _ { n } \\}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1312,
+                    597,
+                    1369,
+                    597,
+                    1369,
+                    630,
+                    1312,
+                    630
+                ],
+                "score": 0.92,
+                "latex": "\\{ { \\bf { x } } _ { i } \\}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1243,
+                    796,
+                    1402,
+                    796,
+                    1402,
+                    827,
+                    1243,
+                    827
+                ],
+                "score": 0.91,
+                "latex": "K + Q = R"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    907,
+                    905,
+                    961,
+                    905,
+                    961,
+                    936,
+                    907,
+                    936
+                ],
+                "score": 0.9,
+                "latex": "\\mathcal F ( \\cdot )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    905,
+                    873,
+                    969,
+                    873,
+                    969,
+                    905,
+                    905,
+                    905
+                ],
+                "score": 0.89,
+                "latex": "\\mathcal { M } ( \\cdot )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    808,
+                    966,
+                    862,
+                    966,
+                    862,
+                    998,
+                    808,
+                    998
+                ],
+                "score": 0.89,
+                "latex": "\\{ \\mathcal { T } _ { t } \\}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    786,
+                    934,
+                    833,
+                    934,
+                    833,
+                    966,
+                    786,
+                    966
+                ],
+                "score": 0.89,
+                "latex": "f ( \\cdot )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1043,
+                    966,
+                    1076,
+                    966,
+                    1076,
+                    996,
+                    1043,
+                    996
+                ],
+                "score": 0.88,
+                "latex": "f _ { t ^ { \\prime } }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    601,
+                    965,
+                    636,
+                    965,
+                    636,
+                    995,
+                    601,
+                    995
+                ],
+                "score": 0.88,
+                "latex": "\\mathcal { T } _ { t ^ { \\prime } }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1065,
+                    1027,
+                    1092,
+                    1027,
+                    1092,
+                    1057,
+                    1065,
+                    1057
+                ],
+                "score": 0.88,
+                "latex": "f _ { t }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    636,
+                    1026,
+                    669,
+                    1026,
+                    669,
+                    1056,
+                    636,
+                    1056
+                ],
+                "score": 0.88,
+                "latex": "\\mathcal { F } _ { t }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1292,
+                    965,
+                    1327,
+                    965,
+                    1327,
+                    995,
+                    1292,
+                    995
+                ],
+                "score": 0.87,
+                "latex": "\\mathcal { T } _ { t ^ { \\prime } }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    569,
+                    1546,
+                    619,
+                    1546,
+                    619,
+                    1578,
+                    569,
+                    1578
+                ],
+                "score": 0.86,
+                "latex": "( \\ell _ { n } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1179,
+                    1974,
+                    1208,
+                    1974,
+                    1208,
+                    2003,
+                    1179,
+                    2003
+                ],
+                "score": 0.86,
+                "latex": "\\mathcal { C } _ { c }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1044,
+                    1976,
+                    1073,
+                    1976,
+                    1073,
+                    2004,
+                    1044,
+                    2004
+                ],
+                "score": 0.86,
+                "latex": "y _ { c }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    932,
+                    1026,
+                    960,
+                    1026,
+                    960,
+                    1056,
+                    932,
+                    1056
+                ],
+                "score": 0.85,
+                "latex": "\\mathcal { T } _ { t }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    475,
+                    1317,
+                    495,
+                    1317,
+                    495,
+                    1347,
+                    475,
+                    1347
+                ],
+                "score": 0.85,
+                "latex": "f"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    982,
+                    996,
+                    1018,
+                    996,
+                    1018,
+                    1023,
+                    982,
+                    1023
+                ],
+                "score": 0.84,
+                "latex": "\\mathcal { M }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    980,
+                    936,
+                    1015,
+                    936,
+                    1015,
+                    963,
+                    980,
+                    963
+                ],
+                "score": 0.84,
+                "latex": "\\mathcal { M }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1095,
+                    1460,
+                    1122,
+                    1460,
+                    1122,
+                    1487,
+                    1095,
+                    1487
+                ],
+                "score": 0.84,
+                "latex": "y _ { i }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1054,
+                    766,
+                    1080,
+                    766,
+                    1080,
+                    797,
+                    1054,
+                    797
+                ],
+                "score": 0.83,
+                "latex": "Q"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1270,
+                    1947,
+                    1297,
+                    1947,
+                    1297,
+                    1974,
+                    1270,
+                    1974
+                ],
+                "score": 0.83,
+                "latex": "y _ { i }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1008,
+                    1287,
+                    1034,
+                    1287,
+                    1034,
+                    1314,
+                    1008,
+                    1314
+                ],
+                "score": 0.83,
+                "latex": "\\mathcal { F }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1132,
+                    936,
+                    1158,
+                    936,
+                    1158,
+                    962,
+                    1132,
+                    962
+                ],
+                "score": 0.83,
+                "latex": "\\mathcal { F }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    421,
+                    1226,
+                    447,
+                    1226,
+                    447,
+                    1252,
+                    421,
+                    1252
+                ],
+                "score": 0.82,
+                "latex": "\\mathcal { F }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    507,
+                    1485,
+                    535,
+                    1485,
+                    535,
+                    1513,
+                    507,
+                    1513
+                ],
+                "score": 0.82,
+                "latex": "\\tau"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1172,
+                    293,
+                    1201,
+                    293,
+                    1201,
+                    319,
+                    1172,
+                    319
+                ],
+                "score": 0.82,
+                "latex": "N"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1373,
+                    1517,
+                    1401,
+                    1517,
+                    1401,
+                    1543,
+                    1373,
+                    1543
+                ],
+                "score": 0.82,
+                "latex": "N"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    664,
+                    1486,
+                    693,
+                    1486,
+                    693,
+                    1512,
+                    664,
+                    1512
+                ],
+                "score": 0.81,
+                "latex": "N"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1015,
+                    736,
+                    1044,
+                    736,
+                    1044,
+                    762,
+                    1015,
+                    762
+                ],
+                "score": 0.81,
+                "latex": "K"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1177,
+                    1488,
+                    1203,
+                    1488,
+                    1203,
+                    1516,
+                    1177,
+                    1516
+                ],
+                "score": 0.81,
+                "latex": "y _ { i }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1182,
+                    323,
+                    1211,
+                    323,
+                    1211,
+                    350,
+                    1182,
+                    350
+                ],
+                "score": 0.81,
+                "latex": "K"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1317,
+                    1456,
+                    1345,
+                    1456,
+                    1345,
+                    1482,
+                    1317,
+                    1482
+                ],
+                "score": 0.81,
+                "latex": "N"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    532,
+                    735,
+                    561,
+                    735,
+                    561,
+                    762,
+                    532,
+                    762
+                ],
+                "score": 0.8,
+                "latex": "K"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1206,
+                    1057,
+                    1229,
+                    1057,
+                    1229,
+                    1083,
+                    1206,
+                    1083
+                ],
+                "score": 0.8,
+                "latex": "\\mathcal { L }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    645,
+                    1058,
+                    668,
+                    1058,
+                    668,
+                    1083,
+                    645,
+                    1083
+                ],
+                "score": 0.8,
+                "latex": "\\mathcal { L }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1096,
+                    323,
+                    1129,
+                    323,
+                    1129,
+                    349,
+                    1096,
+                    349
+                ],
+                "score": 0.8,
+                "latex": "M"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    298,
+                    1516,
+                    326,
+                    1516,
+                    326,
+                    1543,
+                    298,
+                    1543
+                ],
+                "score": 0.79,
+                "latex": "N"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    916,
+                    1516,
+                    945,
+                    1516,
+                    945,
+                    1543,
+                    916,
+                    1543
+                ],
+                "score": 0.78,
+                "latex": "N"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1276,
+                    568,
+                    1296,
+                    568,
+                    1296,
+                    594,
+                    1276,
+                    594
+                ],
+                "score": 0.78,
+                "latex": "\\mathcal { E }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1313,
+                    262,
+                    1345,
+                    262,
+                    1345,
+                    290,
+                    1313,
+                    290
+                ],
+                "score": 0.78,
+                "latex": "M"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    876,
+                    966,
+                    904,
+                    966,
+                    904,
+                    993,
+                    876,
+                    993
+                ],
+                "score": 0.78,
+                "latex": "\\mathcal { F }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1045,
+                    1547,
+                    1074,
+                    1547,
+                    1074,
+                    1573,
+                    1045,
+                    1573
+                ],
+                "score": 0.77,
+                "latex": "N"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    842,
+                    735,
+                    870,
+                    735,
+                    870,
+                    763,
+                    842,
+                    763
+                ],
+                "score": 0.77,
+                "latex": "\\tau"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    511,
+                    1517,
+                    536,
+                    1517,
+                    536,
+                    1544,
+                    511,
+                    1544
+                ],
+                "score": 0.76,
+                "latex": "R"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    435,
+                    735,
+                    468,
+                    735,
+                    468,
+                    763,
+                    435,
+                    763
+                ],
+                "score": 0.76,
+                "latex": "M"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    918,
+                    293,
+                    947,
+                    293,
+                    947,
+                    319,
+                    918,
+                    319
+                ],
+                "score": 0.75,
+                "latex": "N"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    664,
+                    293,
+                    699,
+                    293,
+                    699,
+                    320,
+                    664,
+                    320
+                ],
+                "score": 0.72,
+                "latex": "M"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    704,
+                    827,
+                    737,
+                    827,
+                    737,
+                    854,
+                    704,
+                    854
+                ],
+                "score": 0.68,
+                "latex": "M"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1383,
+                    327,
+                    1402,
+                    327,
+                    1402,
+                    349,
+                    1383,
+                    349
+                ],
+                "score": 0.46,
+                "latex": "a"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    73.0,
+                    817.0,
+                    73.0,
+                    817.0,
+                    108.0,
+                    296.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    510.0,
+                    559.0,
+                    510.0,
+                    559.0,
+                    548.0,
+                    294.0,
+                    548.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1704.0,
+                    1326.0,
+                    1704.0,
+                    1326.0,
+                    1741.0,
+                    293.0,
+                    1741.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    839.0,
+                    2085.0,
+                    860.0,
+                    2085.0,
+                    860.0,
+                    2116.0,
+                    839.0,
+                    2116.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    838.0,
+                    2085.0,
+                    861.0,
+                    2085.0,
+                    861.0,
+                    2118.0,
+                    838.0,
+                    2118.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1390.0,
+                    1407.0,
+                    1390.0,
+                    1407.0,
+                    1429.0,
+                    293.0,
+                    1429.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1421.0,
+                    478.0,
+                    1421.0,
+                    478.0,
+                    1462.0,
+                    293.0,
+                    1462.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    595.0,
+                    1421.0,
+                    1408.0,
+                    1421.0,
+                    1408.0,
+                    1462.0,
+                    595.0,
+                    1462.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1453.0,
+                    1094.0,
+                    1453.0,
+                    1094.0,
+                    1489.0,
+                    294.0,
+                    1489.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1123.0,
+                    1453.0,
+                    1316.0,
+                    1453.0,
+                    1316.0,
+                    1489.0,
+                    1123.0,
+                    1489.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1346.0,
+                    1453.0,
+                    1403.0,
+                    1453.0,
+                    1403.0,
+                    1489.0,
+                    1346.0,
+                    1489.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1483.0,
+                    506.0,
+                    1483.0,
+                    506.0,
+                    1519.0,
+                    294.0,
+                    1519.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    536.0,
+                    1483.0,
+                    663.0,
+                    1483.0,
+                    663.0,
+                    1519.0,
+                    536.0,
+                    1519.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    694.0,
+                    1483.0,
+                    1176.0,
+                    1483.0,
+                    1176.0,
+                    1519.0,
+                    694.0,
+                    1519.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1204.0,
+                    1483.0,
+                    1405.0,
+                    1483.0,
+                    1405.0,
+                    1519.0,
+                    1204.0,
+                    1519.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1513.0,
+                    297.0,
+                    1513.0,
+                    297.0,
+                    1549.0,
+                    293.0,
+                    1549.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    327.0,
+                    1513.0,
+                    510.0,
+                    1513.0,
+                    510.0,
+                    1549.0,
+                    327.0,
+                    1549.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    537.0,
+                    1513.0,
+                    662.0,
+                    1513.0,
+                    662.0,
+                    1549.0,
+                    537.0,
+                    1549.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    738.0,
+                    1513.0,
+                    915.0,
+                    1513.0,
+                    915.0,
+                    1549.0,
+                    738.0,
+                    1549.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    946.0,
+                    1513.0,
+                    1372.0,
+                    1513.0,
+                    1372.0,
+                    1549.0,
+                    946.0,
+                    1549.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1402.0,
+                    1513.0,
+                    1407.0,
+                    1513.0,
+                    1407.0,
+                    1549.0,
+                    1402.0,
+                    1549.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1544.0,
+                    568.0,
+                    1544.0,
+                    568.0,
+                    1581.0,
+                    293.0,
+                    1581.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    620.0,
+                    1544.0,
+                    1044.0,
+                    1544.0,
+                    1044.0,
+                    1581.0,
+                    620.0,
+                    1581.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1075.0,
+                    1544.0,
+                    1407.0,
+                    1544.0,
+                    1407.0,
+                    1581.0,
+                    1075.0,
+                    1581.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1575.0,
+                    473.0,
+                    1575.0,
+                    473.0,
+                    1613.0,
+                    293.0,
+                    1613.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    863.0,
+                    1575.0,
+                    1407.0,
+                    1575.0,
+                    1407.0,
+                    1613.0,
+                    863.0,
+                    1613.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1604.0,
+                    1404.0,
+                    1604.0,
+                    1404.0,
+                    1640.0,
+                    294.0,
+                    1640.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1759.0,
+                    1404.0,
+                    1759.0,
+                    1404.0,
+                    1795.0,
+                    296.0,
+                    1795.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1790.0,
+                    1402.0,
+                    1790.0,
+                    1402.0,
+                    1823.0,
+                    294.0,
+                    1823.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1820.0,
+                    1402.0,
+                    1820.0,
+                    1402.0,
+                    1853.0,
+                    294.0,
+                    1853.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1849.0,
+                    1404.0,
+                    1849.0,
+                    1404.0,
+                    1886.0,
+                    293.0,
+                    1886.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1883.0,
+                    1404.0,
+                    1883.0,
+                    1404.0,
+                    1916.0,
+                    294.0,
+                    1916.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1909.0,
+                    1048.0,
+                    1909.0,
+                    1048.0,
+                    1947.0,
+                    292.0,
+                    1947.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1176.0,
+                    1909.0,
+                    1405.0,
+                    1909.0,
+                    1405.0,
+                    1947.0,
+                    1176.0,
+                    1947.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1940.0,
+                    1269.0,
+                    1940.0,
+                    1269.0,
+                    1977.0,
+                    293.0,
+                    1977.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1298.0,
+                    1940.0,
+                    1406.0,
+                    1940.0,
+                    1406.0,
+                    1977.0,
+                    1298.0,
+                    1977.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1970.0,
+                    396.0,
+                    1970.0,
+                    396.0,
+                    2008.0,
+                    292.0,
+                    2008.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    514.0,
+                    1970.0,
+                    570.0,
+                    1970.0,
+                    570.0,
+                    2008.0,
+                    514.0,
+                    2008.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    628.0,
+                    1970.0,
+                    1043.0,
+                    1970.0,
+                    1043.0,
+                    2008.0,
+                    628.0,
+                    2008.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1074.0,
+                    1970.0,
+                    1178.0,
+                    1970.0,
+                    1178.0,
+                    2008.0,
+                    1074.0,
+                    2008.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1209.0,
+                    1970.0,
+                    1407.0,
+                    1970.0,
+                    1407.0,
+                    2008.0,
+                    1209.0,
+                    2008.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    2002.0,
+                    1406.0,
+                    2002.0,
+                    1406.0,
+                    2037.0,
+                    292.0,
+                    2037.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    229.0,
+                    1403.0,
+                    229.0,
+                    1403.0,
+                    264.0,
+                    294.0,
+                    264.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    258.0,
+                    1312.0,
+                    258.0,
+                    1312.0,
+                    298.0,
+                    293.0,
+                    298.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1346.0,
+                    258.0,
+                    1405.0,
+                    258.0,
+                    1405.0,
+                    298.0,
+                    1346.0,
+                    298.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    292.0,
+                    663.0,
+                    292.0,
+                    663.0,
+                    326.0,
+                    294.0,
+                    326.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    700.0,
+                    292.0,
+                    917.0,
+                    292.0,
+                    917.0,
+                    326.0,
+                    700.0,
+                    326.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    948.0,
+                    292.0,
+                    1171.0,
+                    292.0,
+                    1171.0,
+                    326.0,
+                    948.0,
+                    326.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1202.0,
+                    292.0,
+                    1405.0,
+                    292.0,
+                    1405.0,
+                    326.0,
+                    1202.0,
+                    326.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    322.0,
+                    1095.0,
+                    322.0,
+                    1095.0,
+                    356.0,
+                    294.0,
+                    356.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1130.0,
+                    322.0,
+                    1181.0,
+                    322.0,
+                    1181.0,
+                    356.0,
+                    1130.0,
+                    356.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1212.0,
+                    322.0,
+                    1382.0,
+                    322.0,
+                    1382.0,
+                    356.0,
+                    1212.0,
+                    356.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1403.0,
+                    322.0,
+                    1407.0,
+                    322.0,
+                    1407.0,
+                    356.0,
+                    1403.0,
+                    356.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    353.0,
+                    1405.0,
+                    353.0,
+                    1405.0,
+                    387.0,
+                    292.0,
+                    387.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    383.0,
+                    1405.0,
+                    383.0,
+                    1405.0,
+                    416.0,
+                    293.0,
+                    416.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    413.0,
+                    1408.0,
+                    413.0,
+                    1408.0,
+                    448.0,
+                    292.0,
+                    448.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    445.0,
+                    1242.0,
+                    445.0,
+                    1242.0,
+                    478.0,
+                    293.0,
+                    478.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    872.0,
+                    904.0,
+                    872.0,
+                    904.0,
+                    909.0,
+                    293.0,
+                    909.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    970.0,
+                    872.0,
+                    1405.0,
+                    872.0,
+                    1405.0,
+                    909.0,
+                    970.0,
+                    909.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    902.0,
+                    456.0,
+                    902.0,
+                    456.0,
+                    939.0,
+                    293.0,
+                    939.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    513.0,
+                    902.0,
+                    906.0,
+                    902.0,
+                    906.0,
+                    939.0,
+                    513.0,
+                    939.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    962.0,
+                    902.0,
+                    1406.0,
+                    902.0,
+                    1406.0,
+                    939.0,
+                    962.0,
+                    939.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    935.0,
+                    785.0,
+                    935.0,
+                    785.0,
+                    968.0,
+                    293.0,
+                    968.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    834.0,
+                    935.0,
+                    979.0,
+                    935.0,
+                    979.0,
+                    968.0,
+                    834.0,
+                    968.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1016.0,
+                    935.0,
+                    1131.0,
+                    935.0,
+                    1131.0,
+                    968.0,
+                    1016.0,
+                    968.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1159.0,
+                    935.0,
+                    1405.0,
+                    935.0,
+                    1405.0,
+                    968.0,
+                    1159.0,
+                    968.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    963.0,
+                    600.0,
+                    963.0,
+                    600.0,
+                    1000.0,
+                    293.0,
+                    1000.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    637.0,
+                    963.0,
+                    807.0,
+                    963.0,
+                    807.0,
+                    1000.0,
+                    637.0,
+                    1000.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    863.0,
+                    963.0,
+                    875.0,
+                    963.0,
+                    875.0,
+                    1000.0,
+                    863.0,
+                    1000.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    905.0,
+                    963.0,
+                    1042.0,
+                    963.0,
+                    1042.0,
+                    1000.0,
+                    905.0,
+                    1000.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1077.0,
+                    963.0,
+                    1291.0,
+                    963.0,
+                    1291.0,
+                    1000.0,
+                    1077.0,
+                    1000.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1328.0,
+                    963.0,
+                    1406.0,
+                    963.0,
+                    1406.0,
+                    1000.0,
+                    1328.0,
+                    1000.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    992.0,
+                    981.0,
+                    992.0,
+                    981.0,
+                    1029.0,
+                    293.0,
+                    1029.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1019.0,
+                    992.0,
+                    1152.0,
+                    992.0,
+                    1152.0,
+                    1029.0,
+                    1019.0,
+                    1029.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1208.0,
+                    992.0,
+                    1404.0,
+                    992.0,
+                    1404.0,
+                    1029.0,
+                    1208.0,
+                    1029.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1023.0,
+                    635.0,
+                    1023.0,
+                    635.0,
+                    1063.0,
+                    291.0,
+                    1063.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    670.0,
+                    1023.0,
+                    931.0,
+                    1023.0,
+                    931.0,
+                    1063.0,
+                    670.0,
+                    1063.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    961.0,
+                    1023.0,
+                    1064.0,
+                    1023.0,
+                    1064.0,
+                    1063.0,
+                    961.0,
+                    1063.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1093.0,
+                    1023.0,
+                    1406.0,
+                    1023.0,
+                    1406.0,
+                    1063.0,
+                    1093.0,
+                    1063.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1056.0,
+                    644.0,
+                    1056.0,
+                    644.0,
+                    1093.0,
+                    293.0,
+                    1093.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    669.0,
+                    1056.0,
+                    1205.0,
+                    1056.0,
+                    1205.0,
+                    1093.0,
+                    669.0,
+                    1093.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1230.0,
+                    1056.0,
+                    1405.0,
+                    1056.0,
+                    1405.0,
+                    1093.0,
+                    1230.0,
+                    1093.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1087.0,
+                    518.0,
+                    1087.0,
+                    518.0,
+                    1121.0,
+                    294.0,
+                    1121.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1132.0,
+                    1405.0,
+                    1132.0,
+                    1405.0,
+                    1170.0,
+                    294.0,
+                    1170.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1164.0,
+                    1405.0,
+                    1164.0,
+                    1405.0,
+                    1197.0,
+                    294.0,
+                    1197.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1196.0,
+                    1405.0,
+                    1196.0,
+                    1405.0,
+                    1229.0,
+                    292.0,
+                    1229.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1225.0,
+                    420.0,
+                    1225.0,
+                    420.0,
+                    1259.0,
+                    294.0,
+                    1259.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    448.0,
+                    1225.0,
+                    1405.0,
+                    1225.0,
+                    1405.0,
+                    1259.0,
+                    448.0,
+                    1259.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1256.0,
+                    1405.0,
+                    1256.0,
+                    1405.0,
+                    1290.0,
+                    294.0,
+                    1290.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1286.0,
+                    1007.0,
+                    1286.0,
+                    1007.0,
+                    1320.0,
+                    294.0,
+                    1320.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1035.0,
+                    1286.0,
+                    1405.0,
+                    1286.0,
+                    1405.0,
+                    1320.0,
+                    1035.0,
+                    1320.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1315.0,
+                    474.0,
+                    1315.0,
+                    474.0,
+                    1351.0,
+                    292.0,
+                    1351.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    496.0,
+                    1315.0,
+                    1405.0,
+                    1315.0,
+                    1405.0,
+                    1351.0,
+                    496.0,
+                    1351.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1348.0,
+                    721.0,
+                    1348.0,
+                    721.0,
+                    1381.0,
+                    294.0,
+                    1381.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    298.0,
+                    567.0,
+                    1275.0,
+                    567.0,
+                    1275.0,
+                    600.0,
+                    298.0,
+                    600.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1297.0,
+                    567.0,
+                    1403.0,
+                    567.0,
+                    1403.0,
+                    600.0,
+                    1297.0,
+                    600.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    594.0,
+                    840.0,
+                    594.0,
+                    840.0,
+                    632.0,
+                    294.0,
+                    632.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    971.0,
+                    594.0,
+                    1311.0,
+                    594.0,
+                    1311.0,
+                    632.0,
+                    971.0,
+                    632.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1370.0,
+                    594.0,
+                    1406.0,
+                    594.0,
+                    1406.0,
+                    632.0,
+                    1370.0,
+                    632.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    625.0,
+                    439.0,
+                    625.0,
+                    439.0,
+                    662.0,
+                    294.0,
+                    662.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    494.0,
+                    625.0,
+                    1406.0,
+                    625.0,
+                    1406.0,
+                    662.0,
+                    494.0,
+                    662.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    656.0,
+                    1406.0,
+                    656.0,
+                    1406.0,
+                    693.0,
+                    294.0,
+                    693.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    686.0,
+                    1334.0,
+                    686.0,
+                    1334.0,
+                    723.0,
+                    295.0,
+                    723.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    731.0,
+                    434.0,
+                    731.0,
+                    434.0,
+                    769.0,
+                    293.0,
+                    769.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    469.0,
+                    731.0,
+                    531.0,
+                    731.0,
+                    531.0,
+                    769.0,
+                    469.0,
+                    769.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    562.0,
+                    731.0,
+                    841.0,
+                    731.0,
+                    841.0,
+                    769.0,
+                    562.0,
+                    769.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    871.0,
+                    731.0,
+                    1014.0,
+                    731.0,
+                    1014.0,
+                    769.0,
+                    871.0,
+                    769.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1045.0,
+                    731.0,
+                    1405.0,
+                    731.0,
+                    1405.0,
+                    769.0,
+                    1045.0,
+                    769.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    421.0,
+                    764.0,
+                    1053.0,
+                    764.0,
+                    1053.0,
+                    800.0,
+                    421.0,
+                    800.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1081.0,
+                    764.0,
+                    1407.0,
+                    764.0,
+                    1407.0,
+                    800.0,
+                    1081.0,
+                    800.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    795.0,
+                    1242.0,
+                    795.0,
+                    1242.0,
+                    828.0,
+                    293.0,
+                    828.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1403.0,
+                    795.0,
+                    1408.0,
+                    795.0,
+                    1408.0,
+                    828.0,
+                    1403.0,
+                    828.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    825.0,
+                    703.0,
+                    825.0,
+                    703.0,
+                    860.0,
+                    294.0,
+                    860.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    738.0,
+                    825.0,
+                    831.0,
+                    825.0,
+                    831.0,
+                    860.0,
+                    738.0,
+                    860.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1704.0,
+                    1326.0,
+                    1704.0,
+                    1326.0,
+                    1741.0,
+                    293.0,
+                    1741.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 2,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    445,
+                    1404,
+                    445,
+                    1404,
+                    782,
+                    298,
+                    782
+                ],
+                "score": 0.983
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    796,
+                    1404,
+                    796,
+                    1404,
+                    1073,
+                    298,
+                    1073
+                ],
+                "score": 0.982
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1576,
+                    1403,
+                    1576,
+                    1403,
+                    1762,
+                    298,
+                    1762
+                ],
+                "score": 0.98
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1377,
+                    1403,
+                    1377,
+                    1403,
+                    1563,
+                    297,
+                    1563
+                ],
+                "score": 0.98
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    1087,
+                    1404,
+                    1087,
+                    1404,
+                    1243,
+                    299,
+                    1243
+                ],
+                "score": 0.978
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    307,
+                    1402,
+                    307,
+                    1402,
+                    430,
+                    299,
+                    430
+                ],
+                "score": 0.976
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    300,
+                    1776,
+                    1403,
+                    1776,
+                    1403,
+                    1870,
+                    300,
+                    1870
+                ],
+                "score": 0.974
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    619,
+                    1266,
+                    1078,
+                    1266,
+                    1078,
+                    1359,
+                    619,
+                    1359
+                ],
+                "score": 0.961
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    296,
+                    228,
+                    1401,
+                    228,
+                    1401,
+                    293,
+                    296,
+                    293
+                ],
+                "score": 0.951
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1972,
+                    1401,
+                    1972,
+                    1401,
+                    2035,
+                    297,
+                    2035
+                ],
+                "score": 0.947
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    301,
+                    1911,
+                    588,
+                    1911,
+                    588,
+                    1947,
+                    301,
+                    1947
+                ],
+                "score": 0.914
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    300,
+                    76,
+                    815,
+                    76,
+                    815,
+                    104,
+                    300,
+                    104
+                ],
+                "score": 0.9
+            },
+            {
+                "category_id": 9,
+                "poly": [
+                    1366,
+                    1294,
+                    1400,
+                    1294,
+                    1400,
+                    1324,
+                    1366,
+                    1324
+                ],
+                "score": 0.886
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    841,
+                    2089,
+                    858,
+                    2089,
+                    858,
+                    2111,
+                    841,
+                    2111
+                ],
+                "score": 0.764
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    619,
+                    1260,
+                    1079,
+                    1260,
+                    1079,
+                    1360,
+                    619,
+                    1360
+                ],
+                "score": 0.95,
+                "latex": "\\mathcal { P } , \\{ \\pmb { \\mu } _ { c } \\} = \\underset { \\{ \\mathcal { C } _ { c } \\} , \\{ \\pmb { \\mu } _ { c } \\} } { \\arg \\operatorname* { m i n } } \\sum _ { c = 1 } ^ { k } \\sum _ { \\mathbf { z } \\in \\mathcal { C } _ { c } } \\| \\mathbf { z } - \\pmb { \\mu } _ { c } \\| _ { \\mathbf { A } } ^ { 2 }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    770,
+                    1577,
+                    822,
+                    1577,
+                    822,
+                    1611,
+                    770,
+                    1611
+                ],
+                "score": 0.94,
+                "latex": "\\left\\{ \\mathbf { z } _ { i } \\right\\}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    997,
+                    507,
+                    1118,
+                    507,
+                    1118,
+                    541,
+                    997,
+                    541
+                ],
+                "score": 0.93,
+                "latex": "\\mathcal { P } = \\{ \\mathcal { C } _ { c } \\}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    791,
+                    537,
+                    913,
+                    537,
+                    913,
+                    570,
+                    791,
+                    570
+                ],
+                "score": 0.92,
+                "latex": "p ( \\mathbf { x } | c ) p ( \\bar { c } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    391,
+                    1119,
+                    447,
+                    1119,
+                    447,
+                    1152,
+                    391,
+                    1152
+                ],
+                "score": 0.92,
+                "latex": "\\{ { \\bf { x } } _ { i } \\}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    478,
+                    1149,
+                    539,
+                    1149,
+                    539,
+                    1183,
+                    478,
+                    1183
+                ],
+                "score": 0.92,
+                "latex": "\\{ \\mathcal P _ { p } \\}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    845,
+                    570,
+                    919,
+                    570,
+                    919,
+                    601,
+                    845,
+                    601
+                ],
+                "score": 0.92,
+                "latex": "p ( \\mathbf { x } | c )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    687,
+                    1469,
+                    826,
+                    1469,
+                    826,
+                    1503,
+                    687,
+                    1503
+                ],
+                "score": 0.92,
+                "latex": "p ( c ) \\propto | \\mathcal { C } _ { c } |"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    462,
+                    1440,
+                    552,
+                    1440,
+                    552,
+                    1471,
+                    462,
+                    1471
+                ],
+                "score": 0.92,
+                "latex": "{ \\bf x } _ { i } \\in \\mathcal { C } _ { c }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    621,
+                    230,
+                    677,
+                    230,
+                    677,
+                    264,
+                    621,
+                    264
+                ],
+                "score": 0.92,
+                "latex": "\\{ { \\bf { x } } _ { i } \\}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    893,
+                    1118,
+                    946,
+                    1118,
+                    946,
+                    1152,
+                    893,
+                    1152
+                ],
+                "score": 0.91,
+                "latex": "\\left\\{ \\mathbf { z } _ { i } \\right\\}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    296,
+                    1439,
+                    363,
+                    1439,
+                    363,
+                    1473,
+                    296,
+                    1473
+                ],
+                "score": 0.9,
+                "latex": "\\mathcal { U } ( \\mathcal { P } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    652,
+                    1442,
+                    729,
+                    1442,
+                    729,
+                    1471,
+                    652,
+                    1471
+                ],
+                "score": 0.87,
+                "latex": "y _ { i } = c"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    776,
+                    1212,
+                    805,
+                    1212,
+                    805,
+                    1241,
+                    776,
+                    1241
+                ],
+                "score": 0.87,
+                "latex": "\\mathcal { C } _ { c }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    849,
+                    1701,
+                    869,
+                    1701,
+                    869,
+                    1732,
+                    849,
+                    1732
+                ],
+                "score": 0.85,
+                "latex": "f"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    297,
+                    1215,
+                    332,
+                    1215,
+                    332,
+                    1243,
+                    297,
+                    1243
+                ],
+                "score": 0.85,
+                "latex": "\\pmb { \\mu } _ { c }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    298,
+                    1181,
+                    323,
+                    1181,
+                    323,
+                    1207,
+                    298,
+                    1207
+                ],
+                "score": 0.84,
+                "latex": "\\mathcal { Z }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    788,
+                    1151,
+                    813,
+                    1151,
+                    813,
+                    1177,
+                    788,
+                    1177
+                ],
+                "score": 0.84,
+                "latex": "P"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1249,
+                    509,
+                    1268,
+                    509,
+                    1268,
+                    535,
+                    1249,
+                    535
+                ],
+                "score": 0.83,
+                "latex": "k"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    509,
+                    509,
+                    528,
+                    509,
+                    528,
+                    535,
+                    509,
+                    535
+                ],
+                "score": 0.83,
+                "latex": "k"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    792,
+                    1640,
+                    818,
+                    1640,
+                    818,
+                    1666,
+                    792,
+                    1666
+                ],
+                "score": 0.83,
+                "latex": "\\mathcal { F }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    586,
+                    1701,
+                    613,
+                    1701,
+                    613,
+                    1728,
+                    586,
+                    1728
+                ],
+                "score": 0.83,
+                "latex": "\\mathcal { F }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    337,
+                    1151,
+                    362,
+                    1151,
+                    362,
+                    1177,
+                    337,
+                    1177
+                ],
+                "score": 0.82,
+                "latex": "P"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1277,
+                    600,
+                    1295,
+                    600,
+                    1295,
+                    626,
+                    1277,
+                    626
+                ],
+                "score": 0.82,
+                "latex": "k"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    738,
+                    1120,
+                    762,
+                    1120,
+                    762,
+                    1146,
+                    738,
+                    1146
+                ],
+                "score": 0.8,
+                "latex": "\\mathcal { Z }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1258,
+                    1090,
+                    1283,
+                    1090,
+                    1283,
+                    1116,
+                    1258,
+                    1116
+                ],
+                "score": 0.79,
+                "latex": "\\mathcal { D }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1063,
+                    1471,
+                    1091,
+                    1471,
+                    1091,
+                    1497,
+                    1063,
+                    1497
+                ],
+                "score": 0.78,
+                "latex": "N"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1199,
+                    1090,
+                    1220,
+                    1090,
+                    1220,
+                    1116,
+                    1199,
+                    1116
+                ],
+                "score": 0.76,
+                "latex": "\\mathcal { E }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    890,
+                    310,
+                    915,
+                    310,
+                    915,
+                    336,
+                    890,
+                    336
+                ],
+                "score": 0.76,
+                "latex": "\\mathcal { D }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    538,
+                    752,
+                    577,
+                    752,
+                    577,
+                    783,
+                    538,
+                    783
+                ],
+                "score": 0.44,
+                "latex": "\\mathrm { N g }"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1907.0,
+                    594.0,
+                    1907.0,
+                    594.0,
+                    1953.0,
+                    291.0,
+                    1953.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    73.0,
+                    817.0,
+                    73.0,
+                    817.0,
+                    108.0,
+                    296.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    838.0,
+                    2086.0,
+                    862.0,
+                    2086.0,
+                    862.0,
+                    2118.0,
+                    838.0,
+                    2118.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    445.0,
+                    1404.0,
+                    445.0,
+                    1404.0,
+                    481.0,
+                    295.0,
+                    481.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    477.0,
+                    1404.0,
+                    477.0,
+                    1404.0,
+                    510.0,
+                    294.0,
+                    510.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    505.0,
+                    508.0,
+                    505.0,
+                    508.0,
+                    544.0,
+                    293.0,
+                    544.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    529.0,
+                    505.0,
+                    996.0,
+                    505.0,
+                    996.0,
+                    544.0,
+                    529.0,
+                    544.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1119.0,
+                    505.0,
+                    1248.0,
+                    505.0,
+                    1248.0,
+                    544.0,
+                    1119.0,
+                    544.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1269.0,
+                    505.0,
+                    1409.0,
+                    505.0,
+                    1409.0,
+                    544.0,
+                    1269.0,
+                    544.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    539.0,
+                    790.0,
+                    539.0,
+                    790.0,
+                    570.0,
+                    296.0,
+                    570.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    914.0,
+                    539.0,
+                    1404.0,
+                    539.0,
+                    1404.0,
+                    570.0,
+                    914.0,
+                    570.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    566.0,
+                    844.0,
+                    566.0,
+                    844.0,
+                    604.0,
+                    293.0,
+                    604.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    920.0,
+                    566.0,
+                    1404.0,
+                    566.0,
+                    1404.0,
+                    604.0,
+                    920.0,
+                    604.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    601.0,
+                    1276.0,
+                    601.0,
+                    1276.0,
+                    632.0,
+                    296.0,
+                    632.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1296.0,
+                    601.0,
+                    1405.0,
+                    601.0,
+                    1405.0,
+                    632.0,
+                    1296.0,
+                    632.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    627.0,
+                    1404.0,
+                    627.0,
+                    1404.0,
+                    667.0,
+                    293.0,
+                    667.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    658.0,
+                    1406.0,
+                    658.0,
+                    1406.0,
+                    698.0,
+                    292.0,
+                    698.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    689.0,
+                    1402.0,
+                    689.0,
+                    1402.0,
+                    724.0,
+                    294.0,
+                    724.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    719.0,
+                    1405.0,
+                    719.0,
+                    1405.0,
+                    758.0,
+                    293.0,
+                    758.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    753.0,
+                    537.0,
+                    753.0,
+                    537.0,
+                    784.0,
+                    297.0,
+                    784.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    578.0,
+                    753.0,
+                    778.0,
+                    753.0,
+                    778.0,
+                    784.0,
+                    578.0,
+                    784.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    298.0,
+                    798.0,
+                    1404.0,
+                    798.0,
+                    1404.0,
+                    831.0,
+                    298.0,
+                    831.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    827.0,
+                    1404.0,
+                    827.0,
+                    1404.0,
+                    864.0,
+                    294.0,
+                    864.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    859.0,
+                    1404.0,
+                    859.0,
+                    1404.0,
+                    894.0,
+                    293.0,
+                    894.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    889.0,
+                    1405.0,
+                    889.0,
+                    1405.0,
+                    925.0,
+                    294.0,
+                    925.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    919.0,
+                    1404.0,
+                    919.0,
+                    1404.0,
+                    955.0,
+                    294.0,
+                    955.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    948.0,
+                    1405.0,
+                    948.0,
+                    1405.0,
+                    987.0,
+                    293.0,
+                    987.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    978.0,
+                    1404.0,
+                    978.0,
+                    1404.0,
+                    1017.0,
+                    293.0,
+                    1017.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1012.0,
+                    1405.0,
+                    1012.0,
+                    1405.0,
+                    1044.0,
+                    296.0,
+                    1044.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1043.0,
+                    1016.0,
+                    1043.0,
+                    1016.0,
+                    1076.0,
+                    296.0,
+                    1076.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1575.0,
+                    769.0,
+                    1575.0,
+                    769.0,
+                    1614.0,
+                    294.0,
+                    1614.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    823.0,
+                    1575.0,
+                    1407.0,
+                    1575.0,
+                    1407.0,
+                    1614.0,
+                    823.0,
+                    1614.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1605.0,
+                    1407.0,
+                    1605.0,
+                    1407.0,
+                    1645.0,
+                    294.0,
+                    1645.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1640.0,
+                    791.0,
+                    1640.0,
+                    791.0,
+                    1672.0,
+                    296.0,
+                    1672.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    819.0,
+                    1640.0,
+                    1403.0,
+                    1640.0,
+                    1403.0,
+                    1672.0,
+                    819.0,
+                    1672.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1668.0,
+                    1405.0,
+                    1668.0,
+                    1405.0,
+                    1703.0,
+                    294.0,
+                    1703.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1700.0,
+                    585.0,
+                    1700.0,
+                    585.0,
+                    1732.0,
+                    294.0,
+                    1732.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    614.0,
+                    1700.0,
+                    848.0,
+                    1700.0,
+                    848.0,
+                    1732.0,
+                    614.0,
+                    1732.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    870.0,
+                    1700.0,
+                    1405.0,
+                    1700.0,
+                    1405.0,
+                    1732.0,
+                    870.0,
+                    1732.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1729.0,
+                    941.0,
+                    1729.0,
+                    941.0,
+                    1765.0,
+                    294.0,
+                    1765.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1376.0,
+                    1405.0,
+                    1376.0,
+                    1405.0,
+                    1413.0,
+                    295.0,
+                    1413.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1410.0,
+                    1405.0,
+                    1410.0,
+                    1405.0,
+                    1442.0,
+                    295.0,
+                    1442.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1435.0,
+                    295.0,
+                    1435.0,
+                    295.0,
+                    1478.0,
+                    291.0,
+                    1478.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    364.0,
+                    1435.0,
+                    461.0,
+                    1435.0,
+                    461.0,
+                    1478.0,
+                    364.0,
+                    1478.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    553.0,
+                    1435.0,
+                    651.0,
+                    1435.0,
+                    651.0,
+                    1478.0,
+                    553.0,
+                    1478.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    730.0,
+                    1435.0,
+                    1407.0,
+                    1435.0,
+                    1407.0,
+                    1478.0,
+                    730.0,
+                    1478.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1471.0,
+                    686.0,
+                    1471.0,
+                    686.0,
+                    1503.0,
+                    296.0,
+                    1503.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    827.0,
+                    1471.0,
+                    1062.0,
+                    1471.0,
+                    1062.0,
+                    1503.0,
+                    827.0,
+                    1503.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1092.0,
+                    1471.0,
+                    1405.0,
+                    1471.0,
+                    1405.0,
+                    1503.0,
+                    1092.0,
+                    1503.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1499.0,
+                    1405.0,
+                    1499.0,
+                    1405.0,
+                    1536.0,
+                    292.0,
+                    1536.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1531.0,
+                    1051.0,
+                    1531.0,
+                    1051.0,
+                    1563.0,
+                    296.0,
+                    1563.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1085.0,
+                    1198.0,
+                    1085.0,
+                    1198.0,
+                    1125.0,
+                    294.0,
+                    1125.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1221.0,
+                    1085.0,
+                    1257.0,
+                    1085.0,
+                    1257.0,
+                    1125.0,
+                    1221.0,
+                    1125.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1284.0,
+                    1085.0,
+                    1405.0,
+                    1085.0,
+                    1405.0,
+                    1125.0,
+                    1284.0,
+                    1125.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1117.0,
+                    390.0,
+                    1117.0,
+                    390.0,
+                    1154.0,
+                    295.0,
+                    1154.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    448.0,
+                    1117.0,
+                    737.0,
+                    1117.0,
+                    737.0,
+                    1154.0,
+                    448.0,
+                    1154.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    763.0,
+                    1117.0,
+                    892.0,
+                    1117.0,
+                    892.0,
+                    1154.0,
+                    763.0,
+                    1154.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    947.0,
+                    1117.0,
+                    1406.0,
+                    1117.0,
+                    1406.0,
+                    1154.0,
+                    947.0,
+                    1154.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1149.0,
+                    336.0,
+                    1149.0,
+                    336.0,
+                    1186.0,
+                    294.0,
+                    1186.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    363.0,
+                    1149.0,
+                    477.0,
+                    1149.0,
+                    477.0,
+                    1186.0,
+                    363.0,
+                    1186.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    540.0,
+                    1149.0,
+                    787.0,
+                    1149.0,
+                    787.0,
+                    1186.0,
+                    540.0,
+                    1186.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    814.0,
+                    1149.0,
+                    1406.0,
+                    1149.0,
+                    1406.0,
+                    1186.0,
+                    814.0,
+                    1186.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    324.0,
+                    1178.0,
+                    1405.0,
+                    1178.0,
+                    1405.0,
+                    1216.0,
+                    324.0,
+                    1216.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1210.0,
+                    296.0,
+                    1210.0,
+                    296.0,
+                    1245.0,
+                    293.0,
+                    1245.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    333.0,
+                    1210.0,
+                    775.0,
+                    1210.0,
+                    775.0,
+                    1245.0,
+                    333.0,
+                    1245.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    806.0,
+                    1210.0,
+                    1376.0,
+                    1210.0,
+                    1376.0,
+                    1245.0,
+                    806.0,
+                    1245.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    307.0,
+                    889.0,
+                    307.0,
+                    889.0,
+                    342.0,
+                    293.0,
+                    342.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    916.0,
+                    307.0,
+                    1407.0,
+                    307.0,
+                    1407.0,
+                    342.0,
+                    916.0,
+                    342.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    336.0,
+                    1406.0,
+                    336.0,
+                    1406.0,
+                    374.0,
+                    292.0,
+                    374.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    370.0,
+                    1406.0,
+                    370.0,
+                    1406.0,
+                    402.0,
+                    295.0,
+                    402.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    400.0,
+                    1210.0,
+                    400.0,
+                    1210.0,
+                    436.0,
+                    294.0,
+                    436.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1772.0,
+                    1404.0,
+                    1772.0,
+                    1404.0,
+                    1815.0,
+                    294.0,
+                    1815.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1807.0,
+                    1405.0,
+                    1807.0,
+                    1405.0,
+                    1841.0,
+                    296.0,
+                    1841.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1836.0,
+                    1196.0,
+                    1836.0,
+                    1196.0,
+                    1874.0,
+                    295.0,
+                    1874.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    228.0,
+                    620.0,
+                    228.0,
+                    620.0,
+                    264.0,
+                    295.0,
+                    264.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    678.0,
+                    228.0,
+                    1405.0,
+                    228.0,
+                    1405.0,
+                    264.0,
+                    678.0,
+                    264.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    260.0,
+                    582.0,
+                    260.0,
+                    582.0,
+                    297.0,
+                    295.0,
+                    297.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1972.0,
+                    1405.0,
+                    1972.0,
+                    1405.0,
+                    2007.0,
+                    296.0,
+                    2007.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    2002.0,
+                    1406.0,
+                    2002.0,
+                    1406.0,
+                    2038.0,
+                    295.0,
+                    2038.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 3,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    723,
+                    1405,
+                    723,
+                    1405,
+                    1424,
+                    297,
+                    1424
+                ],
+                "score": 0.983
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1699,
+                    1405,
+                    1699,
+                    1405,
+                    2036,
+                    297,
+                    2036
+                ],
+                "score": 0.981
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1439,
+                    1404,
+                    1439,
+                    1404,
+                    1685,
+                    297,
+                    1685
+                ],
+                "score": 0.981
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    300,
+                    76,
+                    815,
+                    76,
+                    815,
+                    104,
+                    300,
+                    104
+                ],
+                "score": 0.893
+            },
+            {
+                "category_id": 5,
+                "poly": [
+                    302,
+                    269,
+                    1402,
+                    269,
+                    1402,
+                    625,
+                    302,
+                    625
+                ],
+                "score": 0.879,
+                "html": "<table><tr><td colspan=\"2\">1: procedure CACTUs(ε,D,P,k,T,N,Km-tr, Q)</td><td colspan=\"4\"></td></tr><tr><td>2:</td><td colspan=\"4\">Run embedding learning algorithm ε on D and produce embeddings {zi} from observations {xi}.</td></tr><tr><td>3:</td><td colspan=\"4\">Run k-means on {zi} P times (with random scaling) to generate a set of partitions {Pp = {Cc}p}.</td></tr><tr><td>4:</td><td colspan=\"4\">for t from 1 to the number of desired tasks T do</td></tr><tr><td>5:</td><td colspan=\"4\">Sample a partition P uniformly at random from the set of partitions {Pp}.</td></tr><tr><td>6:</td><td colspan=\"4\">Sample a cluster Cn uniformly without replacement from P for each of the N classes desired for a task.</td></tr><tr><td>7:</td><td colspan=\"4\">Sample an embedding Zr without replacement from Cn for each of the R= Km-tr+Q training and</td></tr><tr><td>8:</td><td colspan=\"4\">query examples desired for each class, and record the corresponding datapoint Xn,r. Sample a permutation(ln) of N one-hot labels.</td></tr><tr><td>9:</td><td colspan=\"4\">Construct Tt ={(xn,r,ln)}.</td></tr><tr><td>10:</td><td colspan=\"4\">return {Tt}</td></tr></table>"
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    841,
+                    2088,
+                    858,
+                    2088,
+                    858,
+                    2112,
+                    841,
+                    2112
+                ],
+                "score": 0.722
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    300,
+                    229,
+                    746,
+                    229,
+                    746,
+                    262,
+                    300,
+                    262
+                ],
+                "score": 0.604
+            },
+            {
+                "category_id": 6,
+                "poly": [
+                    300,
+                    229,
+                    746,
+                    229,
+                    746,
+                    262,
+                    300,
+                    262
+                ],
+                "score": 0.111
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    302,
+                    269,
+                    1402,
+                    269,
+                    1402,
+                    625,
+                    302,
+                    625
+                ],
+                "score": 0.105
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    518,
+                    552,
+                    707,
+                    552,
+                    707,
+                    583,
+                    518,
+                    583
+                ],
+                "score": 0.93,
+                "latex": "\\mathcal { T } _ { t } = \\{ ( \\mathbf { x } _ { n , r } , \\pmb { \\ell } _ { n } ) \\}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1116,
+                    466,
+                    1271,
+                    466,
+                    1271,
+                    494,
+                    1116,
+                    494
+                ],
+                "score": 0.93,
+                "latex": "R = K _ { \\mathrm { m - t r } } + Q"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1210,
+                    328,
+                    1369,
+                    328,
+                    1369,
+                    359,
+                    1210,
+                    359
+                ],
+                "score": 0.92,
+                "latex": "\\{ \\mathcal { P } _ { p } = \\{ { \\mathcal C } _ { c } \\} _ { p } \\}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1068,
+                    299,
+                    1118,
+                    299,
+                    1118,
+                    330,
+                    1068,
+                    330
+                ],
+                "score": 0.92,
+                "latex": "\\left\\{ \\mathbf { z } _ { i } \\right\\}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1095,
+                    382,
+                    1152,
+                    382,
+                    1152,
+                    413,
+                    1095,
+                    413
+                ],
+                "score": 0.91,
+                "latex": "\\{ \\mathcal P _ { p } \\}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1308,
+                    299,
+                    1360,
+                    299,
+                    1360,
+                    329,
+                    1308,
+                    329
+                ],
+                "score": 0.89,
+                "latex": "\\left\\{ { { \\bf { x } } _ { i } } \\right\\}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1200,
+                    499,
+                    1248,
+                    499,
+                    1248,
+                    524,
+                    1200,
+                    524
+                ],
+                "score": 0.89,
+                "latex": "{ \\bf x } _ { n , r }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    933,
+                    467,
+                    964,
+                    467,
+                    964,
+                    493,
+                    933,
+                    493
+                ],
+                "score": 0.87,
+                "latex": "{ \\mathcal { C } } _ { n }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    584,
+                    412,
+                    615,
+                    412,
+                    615,
+                    437,
+                    584,
+                    437
+                ],
+                "score": 0.87,
+                "latex": "\\scriptstyle { { \\mathcal { C } } _ { n } }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    549,
+                    327,
+                    598,
+                    327,
+                    598,
+                    358,
+                    549,
+                    358
+                ],
+                "score": 0.86,
+                "latex": "\\left\\{ \\mathbf { z } _ { i } \\right\\}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    639,
+                    470,
+                    667,
+                    470,
+                    667,
+                    493,
+                    639,
+                    493
+                ],
+                "score": 0.86,
+                "latex": "\\mathbf { z } _ { r }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    642,
+                    523,
+                    688,
+                    523,
+                    688,
+                    552,
+                    642,
+                    552
+                ],
+                "score": 0.84,
+                "latex": "( \\ell _ { n } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    988,
+                    412,
+                    1011,
+                    412,
+                    1011,
+                    436,
+                    988,
+                    436
+                ],
+                "score": 0.83,
+                "latex": "\\mathcal { P }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    603,
+                    384,
+                    626,
+                    384,
+                    626,
+                    409,
+                    603,
+                    409
+                ],
+                "score": 0.82,
+                "latex": "\\mathcal { P }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    719,
+                    524,
+                    745,
+                    524,
+                    745,
+                    549,
+                    719,
+                    549
+                ],
+                "score": 0.82,
+                "latex": "N"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1164,
+                    412,
+                    1190,
+                    412,
+                    1190,
+                    436,
+                    1164,
+                    436
+                ],
+                "score": 0.81,
+                "latex": "N"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    517,
+                    272,
+                    833,
+                    272,
+                    833,
+                    301,
+                    517,
+                    301
+                ],
+                "score": 0.79,
+                "latex": "\\Gamma \\mathrm { U s } ( \\mathcal { E } , \\mathcal { D } , P , k , T , N , K _ { \\mathrm { m - t r } } , Q )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    786,
+                    302,
+                    810,
+                    302,
+                    810,
+                    325,
+                    786,
+                    325
+                ],
+                "score": 0.77,
+                "latex": "\\mathcal { D }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    733,
+                    302,
+                    752,
+                    302,
+                    752,
+                    325,
+                    733,
+                    325
+                ],
+                "score": 0.76,
+                "latex": "\\varepsilon"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    809,
+                    356,
+                    831,
+                    356,
+                    831,
+                    380,
+                    809,
+                    380
+                ],
+                "score": 0.73,
+                "latex": "T"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    425,
+                    329,
+                    442,
+                    329,
+                    442,
+                    353,
+                    425,
+                    353
+                ],
+                "score": 0.73,
+                "latex": "k"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    414,
+                    357,
+                    428,
+                    357,
+                    428,
+                    380,
+                    414,
+                    380
+                ],
+                "score": 0.58,
+                "latex": "t"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    602,
+                    329,
+                    625,
+                    329,
+                    625,
+                    354,
+                    602,
+                    354
+                ],
+                "score": 0.51,
+                "latex": "P"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    454,
+                    583,
+                    507,
+                    583,
+                    507,
+                    616,
+                    454,
+                    616
+                ],
+                "score": 0.5,
+                "latex": "\\{ \\mathcal { T } _ { t } \\}"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    73.0,
+                    817.0,
+                    73.0,
+                    817.0,
+                    108.0,
+                    296.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    839.0,
+                    2085.0,
+                    861.0,
+                    2085.0,
+                    861.0,
+                    2120.0,
+                    839.0,
+                    2120.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    227.0,
+                    750.0,
+                    227.0,
+                    750.0,
+                    266.0,
+                    296.0,
+                    266.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    227.0,
+                    750.0,
+                    227.0,
+                    750.0,
+                    266.0,
+                    296.0,
+                    266.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    723.0,
+                    1403.0,
+                    723.0,
+                    1403.0,
+                    758.0,
+                    294.0,
+                    758.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    752.0,
+                    1403.0,
+                    752.0,
+                    1403.0,
+                    788.0,
+                    292.0,
+                    788.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    782.0,
+                    1403.0,
+                    782.0,
+                    1403.0,
+                    819.0,
+                    294.0,
+                    819.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    814.0,
+                    1403.0,
+                    814.0,
+                    1403.0,
+                    849.0,
+                    295.0,
+                    849.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    844.0,
+                    1405.0,
+                    844.0,
+                    1405.0,
+                    882.0,
+                    291.0,
+                    882.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    877.0,
+                    1406.0,
+                    877.0,
+                    1406.0,
+                    912.0,
+                    295.0,
+                    912.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    907.0,
+                    1406.0,
+                    907.0,
+                    1406.0,
+                    942.0,
+                    295.0,
+                    942.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    937.0,
+                    1405.0,
+                    937.0,
+                    1405.0,
+                    970.0,
+                    294.0,
+                    970.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    967.0,
+                    1403.0,
+                    967.0,
+                    1403.0,
+                    1002.0,
+                    294.0,
+                    1002.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    997.0,
+                    1405.0,
+                    997.0,
+                    1405.0,
+                    1032.0,
+                    291.0,
+                    1032.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1027.0,
+                    1405.0,
+                    1027.0,
+                    1405.0,
+                    1062.0,
+                    294.0,
+                    1062.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1055.0,
+                    1407.0,
+                    1055.0,
+                    1407.0,
+                    1095.0,
+                    291.0,
+                    1095.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1089.0,
+                    1407.0,
+                    1089.0,
+                    1407.0,
+                    1124.0,
+                    294.0,
+                    1124.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1120.0,
+                    1406.0,
+                    1120.0,
+                    1406.0,
+                    1155.0,
+                    294.0,
+                    1155.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1150.0,
+                    1403.0,
+                    1150.0,
+                    1403.0,
+                    1185.0,
+                    295.0,
+                    1185.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1180.0,
+                    1405.0,
+                    1180.0,
+                    1405.0,
+                    1215.0,
+                    295.0,
+                    1215.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1210.0,
+                    1405.0,
+                    1210.0,
+                    1405.0,
+                    1245.0,
+                    294.0,
+                    1245.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1237.0,
+                    1406.0,
+                    1237.0,
+                    1406.0,
+                    1275.0,
+                    294.0,
+                    1275.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1269.0,
+                    1406.0,
+                    1269.0,
+                    1406.0,
+                    1306.0,
+                    294.0,
+                    1306.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1301.0,
+                    1405.0,
+                    1301.0,
+                    1405.0,
+                    1336.0,
+                    295.0,
+                    1336.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1334.0,
+                    1405.0,
+                    1334.0,
+                    1405.0,
+                    1366.0,
+                    292.0,
+                    1366.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1363.0,
+                    1405.0,
+                    1363.0,
+                    1405.0,
+                    1398.0,
+                    295.0,
+                    1398.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1394.0,
+                    502.0,
+                    1394.0,
+                    502.0,
+                    1426.0,
+                    295.0,
+                    1426.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1700.0,
+                    1406.0,
+                    1700.0,
+                    1406.0,
+                    1735.0,
+                    295.0,
+                    1735.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1728.0,
+                    1405.0,
+                    1728.0,
+                    1405.0,
+                    1766.0,
+                    292.0,
+                    1766.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1761.0,
+                    1403.0,
+                    1761.0,
+                    1403.0,
+                    1795.0,
+                    294.0,
+                    1795.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1791.0,
+                    1406.0,
+                    1791.0,
+                    1406.0,
+                    1828.0,
+                    292.0,
+                    1828.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1820.0,
+                    1405.0,
+                    1820.0,
+                    1405.0,
+                    1859.0,
+                    294.0,
+                    1859.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1851.0,
+                    1405.0,
+                    1851.0,
+                    1405.0,
+                    1889.0,
+                    294.0,
+                    1889.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1881.0,
+                    1403.0,
+                    1881.0,
+                    1403.0,
+                    1918.0,
+                    294.0,
+                    1918.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1913.0,
+                    1405.0,
+                    1913.0,
+                    1405.0,
+                    1948.0,
+                    294.0,
+                    1948.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1943.0,
+                    1405.0,
+                    1943.0,
+                    1405.0,
+                    1978.0,
+                    294.0,
+                    1978.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1976.0,
+                    1405.0,
+                    1976.0,
+                    1405.0,
+                    2010.0,
+                    295.0,
+                    2010.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    2005.0,
+                    890.0,
+                    2005.0,
+                    890.0,
+                    2039.0,
+                    294.0,
+                    2039.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1441.0,
+                    1405.0,
+                    1441.0,
+                    1405.0,
+                    1474.0,
+                    296.0,
+                    1474.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1471.0,
+                    1406.0,
+                    1471.0,
+                    1406.0,
+                    1504.0,
+                    295.0,
+                    1504.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1502.0,
+                    1404.0,
+                    1502.0,
+                    1404.0,
+                    1536.0,
+                    294.0,
+                    1536.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1532.0,
+                    1404.0,
+                    1532.0,
+                    1404.0,
+                    1566.0,
+                    295.0,
+                    1566.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1561.0,
+                    1404.0,
+                    1561.0,
+                    1404.0,
+                    1594.0,
+                    295.0,
+                    1594.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1590.0,
+                    1404.0,
+                    1590.0,
+                    1404.0,
+                    1629.0,
+                    292.0,
+                    1629.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1623.0,
+                    1404.0,
+                    1623.0,
+                    1404.0,
+                    1657.0,
+                    295.0,
+                    1657.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1655.0,
+                    730.0,
+                    1655.0,
+                    730.0,
+                    1688.0,
+                    293.0,
+                    1688.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    309.0,
+                    268.0,
+                    516.0,
+                    268.0,
+                    516.0,
+                    306.0,
+                    309.0,
+                    306.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    834.0,
+                    268.0,
+                    837.0,
+                    268.0,
+                    837.0,
+                    306.0,
+                    834.0,
+                    306.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    301.0,
+                    293.0,
+                    338.0,
+                    293.0,
+                    338.0,
+                    389.0,
+                    301.0,
+                    389.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    373.0,
+                    297.0,
+                    732.0,
+                    297.0,
+                    732.0,
+                    333.0,
+                    373.0,
+                    333.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    753.0,
+                    297.0,
+                    785.0,
+                    297.0,
+                    785.0,
+                    333.0,
+                    753.0,
+                    333.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    811.0,
+                    297.0,
+                    1067.0,
+                    297.0,
+                    1067.0,
+                    333.0,
+                    811.0,
+                    333.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1119.0,
+                    297.0,
+                    1307.0,
+                    297.0,
+                    1307.0,
+                    333.0,
+                    1119.0,
+                    333.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1361.0,
+                    297.0,
+                    1371.0,
+                    297.0,
+                    1371.0,
+                    333.0,
+                    1361.0,
+                    333.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    372.0,
+                    323.0,
+                    424.0,
+                    323.0,
+                    424.0,
+                    363.0,
+                    372.0,
+                    363.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    443.0,
+                    323.0,
+                    548.0,
+                    323.0,
+                    548.0,
+                    363.0,
+                    443.0,
+                    363.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    626.0,
+                    323.0,
+                    1209.0,
+                    323.0,
+                    1209.0,
+                    363.0,
+                    626.0,
+                    363.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1370.0,
+                    323.0,
+                    1380.0,
+                    323.0,
+                    1380.0,
+                    363.0,
+                    1370.0,
+                    363.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    375.0,
+                    354.0,
+                    413.0,
+                    354.0,
+                    413.0,
+                    385.0,
+                    375.0,
+                    385.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    429.0,
+                    354.0,
+                    808.0,
+                    354.0,
+                    808.0,
+                    385.0,
+                    429.0,
+                    385.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    832.0,
+                    354.0,
+                    867.0,
+                    354.0,
+                    867.0,
+                    385.0,
+                    832.0,
+                    385.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    304.0,
+                    380.0,
+                    336.0,
+                    380.0,
+                    336.0,
+                    413.0,
+                    304.0,
+                    413.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    412.0,
+                    380.0,
+                    602.0,
+                    380.0,
+                    602.0,
+                    416.0,
+                    412.0,
+                    416.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    627.0,
+                    380.0,
+                    1094.0,
+                    380.0,
+                    1094.0,
+                    416.0,
+                    627.0,
+                    416.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1153.0,
+                    380.0,
+                    1162.0,
+                    380.0,
+                    1162.0,
+                    416.0,
+                    1153.0,
+                    416.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    306.0,
+                    409.0,
+                    336.0,
+                    409.0,
+                    336.0,
+                    444.0,
+                    306.0,
+                    444.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    411.0,
+                    407.0,
+                    583.0,
+                    407.0,
+                    583.0,
+                    443.0,
+                    411.0,
+                    443.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    616.0,
+                    407.0,
+                    987.0,
+                    407.0,
+                    987.0,
+                    443.0,
+                    616.0,
+                    443.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1012.0,
+                    407.0,
+                    1163.0,
+                    407.0,
+                    1163.0,
+                    443.0,
+                    1012.0,
+                    443.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1191.0,
+                    407.0,
+                    1401.0,
+                    407.0,
+                    1401.0,
+                    443.0,
+                    1191.0,
+                    443.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    413.0,
+                    439.0,
+                    468.0,
+                    439.0,
+                    468.0,
+                    466.0,
+                    413.0,
+                    466.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    304.0,
+                    463.0,
+                    336.0,
+                    463.0,
+                    336.0,
+                    499.0,
+                    304.0,
+                    499.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    412.0,
+                    463.0,
+                    638.0,
+                    463.0,
+                    638.0,
+                    499.0,
+                    412.0,
+                    499.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    668.0,
+                    463.0,
+                    932.0,
+                    463.0,
+                    932.0,
+                    499.0,
+                    668.0,
+                    499.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    965.0,
+                    463.0,
+                    1115.0,
+                    463.0,
+                    1115.0,
+                    499.0,
+                    965.0,
+                    499.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1272.0,
+                    463.0,
+                    1400.0,
+                    463.0,
+                    1400.0,
+                    499.0,
+                    1272.0,
+                    499.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    408.0,
+                    490.0,
+                    1199.0,
+                    490.0,
+                    1199.0,
+                    533.0,
+                    408.0,
+                    533.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1249.0,
+                    490.0,
+                    1261.0,
+                    490.0,
+                    1261.0,
+                    533.0,
+                    1249.0,
+                    533.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    304.0,
+                    520.0,
+                    336.0,
+                    520.0,
+                    336.0,
+                    555.0,
+                    304.0,
+                    555.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    412.0,
+                    522.0,
+                    641.0,
+                    522.0,
+                    641.0,
+                    555.0,
+                    412.0,
+                    555.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    689.0,
+                    522.0,
+                    718.0,
+                    522.0,
+                    718.0,
+                    555.0,
+                    689.0,
+                    555.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    746.0,
+                    522.0,
+                    901.0,
+                    522.0,
+                    901.0,
+                    555.0,
+                    746.0,
+                    555.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    303.0,
+                    549.0,
+                    336.0,
+                    549.0,
+                    336.0,
+                    583.0,
+                    303.0,
+                    583.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    411.0,
+                    549.0,
+                    517.0,
+                    549.0,
+                    517.0,
+                    584.0,
+                    411.0,
+                    584.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    708.0,
+                    549.0,
+                    717.0,
+                    549.0,
+                    717.0,
+                    584.0,
+                    708.0,
+                    584.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    582.0,
+                    338.0,
+                    582.0,
+                    338.0,
+                    617.0,
+                    296.0,
+                    617.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    374.0,
+                    577.0,
+                    453.0,
+                    577.0,
+                    453.0,
+                    620.0,
+                    374.0,
+                    620.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 4,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    400,
+                    1404,
+                    400,
+                    1404,
+                    917,
+                    298,
+                    917
+                ],
+                "score": 0.985
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1163,
+                    1404,
+                    1163,
+                    1404,
+                    1469,
+                    297,
+                    1469
+                ],
+                "score": 0.981
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1652,
+                    1404,
+                    1652,
+                    1404,
+                    1898,
+                    297,
+                    1898
+                ],
+                "score": 0.98
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1484,
+                    1402,
+                    1484,
+                    1402,
+                    1637,
+                    298,
+                    1637
+                ],
+                "score": 0.978
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    932,
+                    1403,
+                    932,
+                    1403,
+                    1148,
+                    297,
+                    1148
+                ],
+                "score": 0.978
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    1914,
+                    1402,
+                    1914,
+                    1402,
+                    2036,
+                    299,
+                    2036
+                ],
+                "score": 0.973
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    291,
+                    1402,
+                    291,
+                    1402,
+                    383,
+                    299,
+                    383
+                ],
+                "score": 0.971
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    300,
+                    226,
+                    557,
+                    226,
+                    557,
+                    261,
+                    300,
+                    261
+                ],
+                "score": 0.903
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    300,
+                    76,
+                    815,
+                    76,
+                    815,
+                    104,
+                    300,
+                    104
+                ],
+                "score": 0.89
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    840,
+                    2089,
+                    858,
+                    2089,
+                    858,
+                    2112,
+                    840,
+                    2112
+                ],
+                "score": 0.764
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1159,
+                    522,
+                    1197,
+                    522,
+                    1197,
+                    553,
+                    1159,
+                    553
+                ],
+                "score": 0.88,
+                "latex": "k _ { \\mathrm { n n } }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    685,
+                    583,
+                    725,
+                    583,
+                    725,
+                    613,
+                    685,
+                    613
+                ],
+                "score": 0.88,
+                "latex": "k _ { \\mathrm { n n } }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1063,
+                    644,
+                    1117,
+                    644,
+                    1117,
+                    672,
+                    1063,
+                    672
+                ],
+                "score": 0.7,
+                "latex": "N K"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    223.0,
+                    561.0,
+                    223.0,
+                    561.0,
+                    267.0,
+                    293.0,
+                    267.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    73.0,
+                    817.0,
+                    73.0,
+                    817.0,
+                    108.0,
+                    296.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    840.0,
+                    2087.0,
+                    861.0,
+                    2087.0,
+                    861.0,
+                    2118.0,
+                    840.0,
+                    2118.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    398.0,
+                    1405.0,
+                    398.0,
+                    1405.0,
+                    435.0,
+                    293.0,
+                    435.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    428.0,
+                    1406.0,
+                    428.0,
+                    1406.0,
+                    467.0,
+                    293.0,
+                    467.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    461.0,
+                    1405.0,
+                    461.0,
+                    1405.0,
+                    495.0,
+                    293.0,
+                    495.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    486.0,
+                    1406.0,
+                    486.0,
+                    1406.0,
+                    530.0,
+                    292.0,
+                    530.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    524.0,
+                    1158.0,
+                    524.0,
+                    1158.0,
+                    556.0,
+                    294.0,
+                    556.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1198.0,
+                    524.0,
+                    1404.0,
+                    524.0,
+                    1404.0,
+                    556.0,
+                    1198.0,
+                    556.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    552.0,
+                    1405.0,
+                    552.0,
+                    1405.0,
+                    588.0,
+                    294.0,
+                    588.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    583.0,
+                    684.0,
+                    583.0,
+                    684.0,
+                    619.0,
+                    294.0,
+                    619.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    726.0,
+                    583.0,
+                    1406.0,
+                    583.0,
+                    1406.0,
+                    619.0,
+                    726.0,
+                    619.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    612.0,
+                    1406.0,
+                    612.0,
+                    1406.0,
+                    648.0,
+                    294.0,
+                    648.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    642.0,
+                    1062.0,
+                    642.0,
+                    1062.0,
+                    682.0,
+                    293.0,
+                    682.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1118.0,
+                    642.0,
+                    1406.0,
+                    642.0,
+                    1406.0,
+                    682.0,
+                    1118.0,
+                    682.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    669.0,
+                    1405.0,
+                    669.0,
+                    1405.0,
+                    713.0,
+                    291.0,
+                    713.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    704.0,
+                    1406.0,
+                    704.0,
+                    1406.0,
+                    742.0,
+                    292.0,
+                    742.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    733.0,
+                    1404.0,
+                    733.0,
+                    1404.0,
+                    772.0,
+                    292.0,
+                    772.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    765.0,
+                    1405.0,
+                    765.0,
+                    1405.0,
+                    801.0,
+                    293.0,
+                    801.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    793.0,
+                    1404.0,
+                    793.0,
+                    1404.0,
+                    832.0,
+                    293.0,
+                    832.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    824.0,
+                    1404.0,
+                    824.0,
+                    1404.0,
+                    862.0,
+                    293.0,
+                    862.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    854.0,
+                    1405.0,
+                    854.0,
+                    1405.0,
+                    892.0,
+                    293.0,
+                    892.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    889.0,
+                    781.0,
+                    889.0,
+                    781.0,
+                    921.0,
+                    296.0,
+                    921.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1161.0,
+                    1407.0,
+                    1161.0,
+                    1407.0,
+                    1199.0,
+                    294.0,
+                    1199.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1194.0,
+                    1407.0,
+                    1194.0,
+                    1407.0,
+                    1227.0,
+                    293.0,
+                    1227.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1225.0,
+                    1407.0,
+                    1225.0,
+                    1407.0,
+                    1259.0,
+                    292.0,
+                    1259.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1255.0,
+                    1407.0,
+                    1255.0,
+                    1407.0,
+                    1288.0,
+                    293.0,
+                    1288.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1282.0,
+                    1406.0,
+                    1282.0,
+                    1406.0,
+                    1324.0,
+                    292.0,
+                    1324.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1317.0,
+                    1405.0,
+                    1317.0,
+                    1405.0,
+                    1349.0,
+                    296.0,
+                    1349.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1348.0,
+                    1404.0,
+                    1348.0,
+                    1404.0,
+                    1379.0,
+                    296.0,
+                    1379.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1374.0,
+                    1405.0,
+                    1374.0,
+                    1405.0,
+                    1411.0,
+                    292.0,
+                    1411.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1407.0,
+                    1405.0,
+                    1407.0,
+                    1405.0,
+                    1443.0,
+                    295.0,
+                    1443.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1438.0,
+                    1205.0,
+                    1438.0,
+                    1205.0,
+                    1473.0,
+                    295.0,
+                    1473.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1649.0,
+                    1405.0,
+                    1649.0,
+                    1405.0,
+                    1690.0,
+                    292.0,
+                    1690.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1681.0,
+                    1405.0,
+                    1681.0,
+                    1405.0,
+                    1720.0,
+                    292.0,
+                    1720.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1711.0,
+                    1405.0,
+                    1711.0,
+                    1405.0,
+                    1752.0,
+                    292.0,
+                    1752.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1745.0,
+                    1405.0,
+                    1745.0,
+                    1405.0,
+                    1779.0,
+                    294.0,
+                    1779.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1773.0,
+                    1406.0,
+                    1773.0,
+                    1406.0,
+                    1815.0,
+                    291.0,
+                    1815.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1804.0,
+                    1406.0,
+                    1804.0,
+                    1406.0,
+                    1841.0,
+                    292.0,
+                    1841.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1834.0,
+                    1406.0,
+                    1834.0,
+                    1406.0,
+                    1872.0,
+                    291.0,
+                    1872.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1865.0,
+                    988.0,
+                    1865.0,
+                    988.0,
+                    1901.0,
+                    294.0,
+                    1901.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    1485.0,
+                    1403.0,
+                    1485.0,
+                    1403.0,
+                    1518.0,
+                    297.0,
+                    1518.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1514.0,
+                    1404.0,
+                    1514.0,
+                    1404.0,
+                    1547.0,
+                    294.0,
+                    1547.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1545.0,
+                    1403.0,
+                    1545.0,
+                    1403.0,
+                    1578.0,
+                    296.0,
+                    1578.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1576.0,
+                    1404.0,
+                    1576.0,
+                    1404.0,
+                    1609.0,
+                    296.0,
+                    1609.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1608.0,
+                    1325.0,
+                    1608.0,
+                    1325.0,
+                    1641.0,
+                    294.0,
+                    1641.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    930.0,
+                    1407.0,
+                    930.0,
+                    1407.0,
+                    970.0,
+                    292.0,
+                    970.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    963.0,
+                    1405.0,
+                    963.0,
+                    1405.0,
+                    1001.0,
+                    295.0,
+                    1001.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    993.0,
+                    1405.0,
+                    993.0,
+                    1405.0,
+                    1030.0,
+                    292.0,
+                    1030.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1025.0,
+                    1401.0,
+                    1025.0,
+                    1401.0,
+                    1060.0,
+                    295.0,
+                    1060.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1056.0,
+                    1404.0,
+                    1056.0,
+                    1404.0,
+                    1091.0,
+                    295.0,
+                    1091.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1086.0,
+                    1406.0,
+                    1086.0,
+                    1406.0,
+                    1121.0,
+                    292.0,
+                    1121.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1117.0,
+                    1403.0,
+                    1117.0,
+                    1403.0,
+                    1151.0,
+                    295.0,
+                    1151.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1914.0,
+                    1403.0,
+                    1914.0,
+                    1403.0,
+                    1946.0,
+                    295.0,
+                    1946.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1944.0,
+                    1406.0,
+                    1944.0,
+                    1406.0,
+                    1980.0,
+                    293.0,
+                    1980.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1975.0,
+                    1407.0,
+                    1975.0,
+                    1407.0,
+                    2011.0,
+                    294.0,
+                    2011.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    2005.0,
+                    1293.0,
+                    2005.0,
+                    1293.0,
+                    2037.0,
+                    295.0,
+                    2037.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    289.0,
+                    1407.0,
+                    289.0,
+                    1407.0,
+                    329.0,
+                    294.0,
+                    329.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    321.0,
+                    1401.0,
+                    321.0,
+                    1401.0,
+                    357.0,
+                    294.0,
+                    357.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    353.0,
+                    931.0,
+                    353.0,
+                    931.0,
+                    389.0,
+                    293.0,
+                    389.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 5,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    927,
+                    1404,
+                    927,
+                    1404,
+                    1294,
+                    298,
+                    1294
+                ],
+                "score": 0.982
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1310,
+                    1403,
+                    1310,
+                    1403,
+                    1523,
+                    298,
+                    1523
+                ],
+                "score": 0.979
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1676,
+                    1402,
+                    1676,
+                    1402,
+                    1861,
+                    298,
+                    1861
+                ],
+                "score": 0.979
+            },
+            {
+                "category_id": 3,
+                "poly": [
+                    302,
+                    209,
+                    1395,
+                    209,
+                    1395,
+                    627,
+                    302,
+                    627
+                ],
+                "score": 0.976
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1538,
+                    1402,
+                    1538,
+                    1402,
+                    1662,
+                    298,
+                    1662
+                ],
+                "score": 0.974
+            },
+            {
+                "category_id": 4,
+                "poly": [
+                    297,
+                    636,
+                    1404,
+                    636,
+                    1404,
+                    804,
+                    297,
+                    804
+                ],
+                "score": 0.969
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    301,
+                    1972,
+                    1399,
+                    1972,
+                    1399,
+                    2032,
+                    301,
+                    2032
+                ],
+                "score": 0.936
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    299,
+                    76,
+                    815,
+                    76,
+                    815,
+                    104,
+                    299,
+                    104
+                ],
+                "score": 0.888
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    300,
+                    1914,
+                    473,
+                    1914,
+                    473,
+                    1944,
+                    300,
+                    1944
+                ],
+                "score": 0.886
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    301,
+                    869,
+                    820,
+                    869,
+                    820,
+                    899,
+                    301,
+                    899
+                ],
+                "score": 0.751
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    842,
+                    2088,
+                    858,
+                    2088,
+                    858,
+                    2110,
+                    842,
+                    2110
+                ],
+                "score": 0.625
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    842,
+                    2088,
+                    858,
+                    2088,
+                    858,
+                    2110,
+                    842,
+                    2110
+                ],
+                "score": 0.229
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    301,
+                    869,
+                    820,
+                    869,
+                    820,
+                    899,
+                    301,
+                    899
+                ],
+                "score": 0.16
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    328,
+                    1233,
+                    357,
+                    1233,
+                    357,
+                    1259,
+                    328,
+                    1259
+                ],
+                "score": 0.81,
+                "latex": "K"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    849,
+                    1233,
+                    877,
+                    1233,
+                    877,
+                    1259,
+                    849,
+                    1259
+                ],
+                "score": 0.81,
+                "latex": "N"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    510,
+                    1263,
+                    539,
+                    1263,
+                    539,
+                    1290,
+                    510,
+                    1290
+                ],
+                "score": 0.8,
+                "latex": "K"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1030,
+                    1203,
+                    1059,
+                    1203,
+                    1059,
+                    1229,
+                    1030,
+                    1229
+                ],
+                "score": 0.8,
+                "latex": "K"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    914,
+                    1172,
+                    943,
+                    1172,
+                    943,
+                    1199,
+                    914,
+                    1199
+                ],
+                "score": 0.79,
+                "latex": "N"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1106.0,
+                    205.0,
+                    1159.0,
+                    205.0,
+                    1159.0,
+                    242.0,
+                    1106.0,
+                    242.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1348.0,
+                    221.0,
+                    1365.0,
+                    221.0,
+                    1365.0,
+                    230.0,
+                    1348.0,
+                    230.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    299.0,
+                    225.0,
+                    525.0,
+                    225.0,
+                    525.0,
+                    453.0,
+                    299.0,
+                    453.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    951.0,
+                    257.0,
+                    1183.0,
+                    257.0,
+                    1183.0,
+                    330.0,
+                    951.0,
+                    330.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1209.0,
+                    291.0,
+                    1219.0,
+                    291.0,
+                    1219.0,
+                    300.0,
+                    1209.0,
+                    300.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    574.0,
+                    309.0,
+                    681.0,
+                    309.0,
+                    681.0,
+                    364.0,
+                    574.0,
+                    364.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1003.0,
+                    324.0,
+                    1395.0,
+                    324.0,
+                    1395.0,
+                    432.0,
+                    1003.0,
+                    432.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    305.0,
+                    482.0,
+                    335.0,
+                    482.0,
+                    335.0,
+                    552.0,
+                    305.0,
+                    552.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    935.0,
+                    427.0,
+                    1395.0,
+                    427.0,
+                    1395.0,
+                    604.0,
+                    935.0,
+                    604.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    590.0,
+                    602.0,
+                    624.0,
+                    602.0,
+                    624.0,
+                    633.0,
+                    590.0,
+                    633.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1150.0,
+                    599.0,
+                    1188.0,
+                    599.0,
+                    1188.0,
+                    633.0,
+                    1150.0,
+                    633.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    635.0,
+                    1405.0,
+                    635.0,
+                    1405.0,
+                    669.0,
+                    295.0,
+                    669.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    664.0,
+                    1406.0,
+                    664.0,
+                    1406.0,
+                    696.0,
+                    292.0,
+                    696.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    693.0,
+                    1405.0,
+                    693.0,
+                    1405.0,
+                    724.0,
+                    292.0,
+                    724.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    720.0,
+                    1404.0,
+                    720.0,
+                    1404.0,
+                    754.0,
+                    294.0,
+                    754.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    749.0,
+                    1405.0,
+                    749.0,
+                    1405.0,
+                    779.0,
+                    296.0,
+                    779.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    777.0,
+                    979.0,
+                    777.0,
+                    979.0,
+                    807.0,
+                    295.0,
+                    807.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    73.0,
+                    816.0,
+                    73.0,
+                    816.0,
+                    108.0,
+                    295.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1911.0,
+                    478.0,
+                    1911.0,
+                    478.0,
+                    1949.0,
+                    294.0,
+                    1949.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    868.0,
+                    824.0,
+                    868.0,
+                    824.0,
+                    903.0,
+                    295.0,
+                    903.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    840.0,
+                    2087.0,
+                    860.0,
+                    2087.0,
+                    860.0,
+                    2118.0,
+                    840.0,
+                    2118.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    840.0,
+                    2087.0,
+                    860.0,
+                    2087.0,
+                    860.0,
+                    2118.0,
+                    840.0,
+                    2118.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    926.0,
+                    1406.0,
+                    926.0,
+                    1406.0,
+                    967.0,
+                    293.0,
+                    967.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    958.0,
+                    1406.0,
+                    958.0,
+                    1406.0,
+                    993.0,
+                    296.0,
+                    993.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    988.0,
+                    1404.0,
+                    988.0,
+                    1404.0,
+                    1024.0,
+                    293.0,
+                    1024.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1019.0,
+                    1404.0,
+                    1019.0,
+                    1404.0,
+                    1053.0,
+                    294.0,
+                    1053.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1051.0,
+                    1405.0,
+                    1051.0,
+                    1405.0,
+                    1084.0,
+                    293.0,
+                    1084.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1081.0,
+                    1404.0,
+                    1081.0,
+                    1404.0,
+                    1115.0,
+                    294.0,
+                    1115.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1111.0,
+                    1405.0,
+                    1111.0,
+                    1405.0,
+                    1145.0,
+                    294.0,
+                    1145.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1141.0,
+                    1404.0,
+                    1141.0,
+                    1404.0,
+                    1175.0,
+                    294.0,
+                    1175.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1168.0,
+                    913.0,
+                    1168.0,
+                    913.0,
+                    1209.0,
+                    292.0,
+                    1209.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    944.0,
+                    1168.0,
+                    1405.0,
+                    1168.0,
+                    1405.0,
+                    1209.0,
+                    944.0,
+                    1209.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1199.0,
+                    1029.0,
+                    1199.0,
+                    1029.0,
+                    1238.0,
+                    292.0,
+                    1238.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1060.0,
+                    1199.0,
+                    1405.0,
+                    1199.0,
+                    1405.0,
+                    1238.0,
+                    1060.0,
+                    1238.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1231.0,
+                    327.0,
+                    1231.0,
+                    327.0,
+                    1267.0,
+                    293.0,
+                    1267.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    358.0,
+                    1231.0,
+                    848.0,
+                    1231.0,
+                    848.0,
+                    1267.0,
+                    358.0,
+                    1267.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    878.0,
+                    1231.0,
+                    1405.0,
+                    1231.0,
+                    1405.0,
+                    1267.0,
+                    878.0,
+                    1267.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1261.0,
+                    509.0,
+                    1261.0,
+                    509.0,
+                    1299.0,
+                    295.0,
+                    1299.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    540.0,
+                    1261.0,
+                    772.0,
+                    1261.0,
+                    772.0,
+                    1299.0,
+                    540.0,
+                    1299.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1308.0,
+                    1404.0,
+                    1308.0,
+                    1404.0,
+                    1344.0,
+                    294.0,
+                    1344.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1338.0,
+                    1403.0,
+                    1338.0,
+                    1403.0,
+                    1375.0,
+                    293.0,
+                    1375.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1368.0,
+                    1407.0,
+                    1368.0,
+                    1407.0,
+                    1407.0,
+                    293.0,
+                    1407.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1399.0,
+                    1405.0,
+                    1399.0,
+                    1405.0,
+                    1435.0,
+                    293.0,
+                    1435.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1432.0,
+                    1403.0,
+                    1432.0,
+                    1403.0,
+                    1467.0,
+                    293.0,
+                    1467.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1462.0,
+                    1405.0,
+                    1462.0,
+                    1405.0,
+                    1496.0,
+                    294.0,
+                    1496.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1492.0,
+                    1102.0,
+                    1492.0,
+                    1102.0,
+                    1527.0,
+                    294.0,
+                    1527.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1676.0,
+                    1406.0,
+                    1676.0,
+                    1406.0,
+                    1713.0,
+                    294.0,
+                    1713.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1707.0,
+                    1406.0,
+                    1707.0,
+                    1406.0,
+                    1743.0,
+                    293.0,
+                    1743.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1737.0,
+                    1406.0,
+                    1737.0,
+                    1406.0,
+                    1774.0,
+                    293.0,
+                    1774.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1762.0,
+                    1406.0,
+                    1762.0,
+                    1406.0,
+                    1807.0,
+                    292.0,
+                    1807.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1799.0,
+                    1404.0,
+                    1799.0,
+                    1404.0,
+                    1835.0,
+                    294.0,
+                    1835.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1831.0,
+                    1307.0,
+                    1831.0,
+                    1307.0,
+                    1863.0,
+                    296.0,
+                    1863.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1537.0,
+                    1406.0,
+                    1537.0,
+                    1406.0,
+                    1572.0,
+                    293.0,
+                    1572.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1568.0,
+                    1404.0,
+                    1568.0,
+                    1404.0,
+                    1601.0,
+                    294.0,
+                    1601.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1601.0,
+                    1406.0,
+                    1601.0,
+                    1406.0,
+                    1633.0,
+                    293.0,
+                    1633.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1626.0,
+                    1406.0,
+                    1626.0,
+                    1406.0,
+                    1670.0,
+                    291.0,
+                    1670.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1971.0,
+                    1404.0,
+                    1971.0,
+                    1404.0,
+                    2006.0,
+                    296.0,
+                    2006.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    2002.0,
+                    499.0,
+                    2002.0,
+                    499.0,
+                    2033.0,
+                    295.0,
+                    2033.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    868.0,
+                    824.0,
+                    868.0,
+                    824.0,
+                    903.0,
+                    295.0,
+                    903.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 6,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    231,
+                    1404,
+                    231,
+                    1404,
+                    779,
+                    297,
+                    779
+                ],
+                "score": 0.985
+            },
+            {
+                "category_id": 5,
+                "poly": [
+                    438,
+                    911,
+                    1261,
+                    911,
+                    1261,
+                    1428,
+                    438,
+                    1428
+                ],
+                "score": 0.982,
+                "html": "<table><tr><td>Algorithm</td><td>(way, shot)</td><td>(5,1)</td><td>(5,5)</td><td>(20,1)</td><td>(20,5)</td></tr><tr><td>Training from scratch</td><td>52.50%</td><td></td><td>74.78%</td><td>24.91%</td><td>47.62%</td></tr><tr><td>ACAI knn-nearest neighbors</td><td></td><td>57.46%</td><td>81.16%</td><td>39.73%</td><td>66.38%</td></tr><tr><td>ACAIlinear classifier</td><td></td><td>61.08%</td><td>81.82%</td><td>43.20%</td><td>66.33%</td></tr><tr><td>ACAI MLP with dropout</td><td></td><td>51.95%</td><td>77.20%</td><td>30.65%</td><td>58.62%</td></tr><tr><td>ACAI cluster matching</td><td></td><td>54.94%</td><td>71.09%</td><td>32.19%</td><td>45.93%</td></tr><tr><td>ACAI CACTUs-MAML (ours)</td><td>68.84%</td><td></td><td>87.78%</td><td>48.09%</td><td>73.36%</td></tr><tr><td>ACAI CACTUs-ProtoNets (ours)</td><td>68.12%</td><td></td><td>83.58%</td><td>47.75%</td><td>66.27%</td></tr><tr><td>BiGAN knn-nearest neighbors</td><td>49.55%</td><td></td><td>68.06%</td><td>27.37%</td><td>46.70%</td></tr><tr><td>BiGAN linearclassifier</td><td>48.28%</td><td></td><td>68.72%</td><td>27.80%</td><td>45.82%</td></tr><tr><td>BiGAN MLP with dropout</td><td>40.54%</td><td></td><td>62.56%</td><td>19.92%</td><td>40.71%</td></tr><tr><td>BiGAN cluster matching</td><td>43.96%</td><td></td><td>58.62%</td><td>21.54%</td><td>31.06%</td></tr><tr><td>BiGAN CACTUs-MAML (ours)</td><td>58.18%</td><td></td><td>78.66%</td><td>35.56%</td><td>58.62%</td></tr><tr><td>BiGAN CACTUs-ProtoNets (ours)</td><td>54.74%</td><td></td><td>71.69%</td><td>33.40%</td><td>50.62%</td></tr><tr><td>Oracle-MAML (control)</td><td>94.46%</td><td></td><td>98.83%</td><td>84.60%</td><td>96.29%</td></tr><tr><td>Oracle-ProtoNets (control)</td><td>98.35%</td><td></td><td>99.58%</td><td>95.31%</td><td>98.81%</td></tr></table>"
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1650,
+                    1403,
+                    1650,
+                    1403,
+                    1867,
+                    298,
+                    1867
+                ],
+                "score": 0.981
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    1881,
+                    1402,
+                    1881,
+                    1402,
+                    2034,
+                    299,
+                    2034
+                ],
+                "score": 0.977
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1451,
+                    1403,
+                    1451,
+                    1403,
+                    1636,
+                    297,
+                    1636
+                ],
+                "score": 0.97
+            },
+            {
+                "category_id": 6,
+                "poly": [
+                    299,
+                    809,
+                    1399,
+                    809,
+                    1399,
+                    896,
+                    299,
+                    896
+                ],
+                "score": 0.94
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    300,
+                    76,
+                    815,
+                    76,
+                    815,
+                    104,
+                    300,
+                    104
+                ],
+                "score": 0.895
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    841,
+                    2089,
+                    858,
+                    2089,
+                    858,
+                    2111,
+                    841,
+                    2111
+                ],
+                "score": 0.794
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    759,
+                    1774,
+                    870,
+                    1774,
+                    870,
+                    1804,
+                    759,
+                    1804
+                ],
+                "score": 0.92,
+                "latex": "M \\ \\leq \\ N"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    928,
+                    1774,
+                    1036,
+                    1774,
+                    1036,
+                    1804,
+                    928,
+                    1804
+                ],
+                "score": 0.91,
+                "latex": "K \\ \\leq \\ 5 0"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    758,
+                    839,
+                    856,
+                    839,
+                    856,
+                    865,
+                    758,
+                    865
+                ],
+                "score": 0.9,
+                "latex": "k = 5 0 0"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1063,
+                    839,
+                    1166,
+                    839,
+                    1166,
+                    865,
+                    1063,
+                    865
+                ],
+                "score": 0.9,
+                "latex": "P = 1 0 0"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    514,
+                    1775,
+                    543,
+                    1775,
+                    543,
+                    1801,
+                    514,
+                    1801
+                ],
+                "score": 0.82,
+                "latex": "K"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    498,
+                    1744,
+                    526,
+                    1744,
+                    526,
+                    1770,
+                    498,
+                    1770
+                ],
+                "score": 0.82,
+                "latex": "N"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    527,
+                    1183,
+                    562,
+                    1183,
+                    562,
+                    1209,
+                    527,
+                    1209
+                ],
+                "score": 0.81,
+                "latex": "k _ { \\mathrm { n n } }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    418,
+                    1774,
+                    451,
+                    1774,
+                    451,
+                    1801,
+                    418,
+                    1801
+                ],
+                "score": 0.79,
+                "latex": "M"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    612,
+                    867,
+                    629,
+                    867,
+                    629,
+                    891,
+                    612,
+                    891
+                ],
+                "score": 0.72,
+                "latex": "k"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    510,
+                    1005,
+                    544,
+                    1005,
+                    544,
+                    1030,
+                    510,
+                    1030
+                ],
+                "score": 0.6,
+                "latex": "k _ { \\mathrm { n n } }"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    808.0,
+                    1405.0,
+                    808.0,
+                    1405.0,
+                    843.0,
+                    294.0,
+                    843.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    837.0,
+                    757.0,
+                    837.0,
+                    757.0,
+                    871.0,
+                    292.0,
+                    871.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    857.0,
+                    837.0,
+                    1062.0,
+                    837.0,
+                    1062.0,
+                    871.0,
+                    857.0,
+                    871.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1167.0,
+                    837.0,
+                    1404.0,
+                    837.0,
+                    1404.0,
+                    871.0,
+                    1167.0,
+                    871.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    865.0,
+                    611.0,
+                    865.0,
+                    611.0,
+                    898.0,
+                    295.0,
+                    898.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    630.0,
+                    865.0,
+                    1388.0,
+                    865.0,
+                    1388.0,
+                    898.0,
+                    630.0,
+                    898.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    73.0,
+                    817.0,
+                    73.0,
+                    817.0,
+                    108.0,
+                    296.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    839.0,
+                    2086.0,
+                    860.0,
+                    2086.0,
+                    860.0,
+                    2116.0,
+                    839.0,
+                    2116.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    228.0,
+                    1402.0,
+                    228.0,
+                    1402.0,
+                    263.0,
+                    295.0,
+                    263.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    259.0,
+                    1405.0,
+                    259.0,
+                    1405.0,
+                    297.0,
+                    291.0,
+                    297.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    291.0,
+                    1406.0,
+                    291.0,
+                    1406.0,
+                    327.0,
+                    294.0,
+                    327.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    320.0,
+                    1404.0,
+                    320.0,
+                    1404.0,
+                    356.0,
+                    294.0,
+                    356.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    351.0,
+                    1405.0,
+                    351.0,
+                    1405.0,
+                    389.0,
+                    294.0,
+                    389.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    381.0,
+                    1405.0,
+                    381.0,
+                    1405.0,
+                    419.0,
+                    294.0,
+                    419.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    410.0,
+                    1406.0,
+                    410.0,
+                    1406.0,
+                    452.0,
+                    291.0,
+                    452.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    442.0,
+                    1404.0,
+                    442.0,
+                    1404.0,
+                    480.0,
+                    292.0,
+                    480.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    471.0,
+                    1406.0,
+                    471.0,
+                    1406.0,
+                    513.0,
+                    291.0,
+                    513.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    505.0,
+                    1406.0,
+                    505.0,
+                    1406.0,
+                    540.0,
+                    295.0,
+                    540.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    534.0,
+                    1404.0,
+                    534.0,
+                    1404.0,
+                    570.0,
+                    294.0,
+                    570.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    563.0,
+                    1406.0,
+                    563.0,
+                    1406.0,
+                    601.0,
+                    294.0,
+                    601.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    595.0,
+                    1404.0,
+                    595.0,
+                    1404.0,
+                    630.0,
+                    294.0,
+                    630.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    627.0,
+                    1404.0,
+                    627.0,
+                    1404.0,
+                    662.0,
+                    294.0,
+                    662.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    653.0,
+                    1405.0,
+                    653.0,
+                    1405.0,
+                    694.0,
+                    292.0,
+                    694.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    687.0,
+                    1405.0,
+                    687.0,
+                    1405.0,
+                    723.0,
+                    295.0,
+                    723.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    718.0,
+                    1404.0,
+                    718.0,
+                    1404.0,
+                    753.0,
+                    295.0,
+                    753.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    751.0,
+                    496.0,
+                    751.0,
+                    496.0,
+                    783.0,
+                    296.0,
+                    783.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1651.0,
+                    1405.0,
+                    1651.0,
+                    1405.0,
+                    1685.0,
+                    296.0,
+                    1685.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1681.0,
+                    1407.0,
+                    1681.0,
+                    1407.0,
+                    1718.0,
+                    292.0,
+                    1718.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1711.0,
+                    1407.0,
+                    1711.0,
+                    1407.0,
+                    1747.0,
+                    292.0,
+                    1747.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1742.0,
+                    497.0,
+                    1742.0,
+                    497.0,
+                    1777.0,
+                    296.0,
+                    1777.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    527.0,
+                    1742.0,
+                    1408.0,
+                    1742.0,
+                    1408.0,
+                    1777.0,
+                    527.0,
+                    1777.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1770.0,
+                    417.0,
+                    1770.0,
+                    417.0,
+                    1808.0,
+                    293.0,
+                    1808.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    452.0,
+                    1770.0,
+                    513.0,
+                    1770.0,
+                    513.0,
+                    1808.0,
+                    452.0,
+                    1808.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    544.0,
+                    1770.0,
+                    758.0,
+                    1770.0,
+                    758.0,
+                    1808.0,
+                    544.0,
+                    1808.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    871.0,
+                    1770.0,
+                    927.0,
+                    1770.0,
+                    927.0,
+                    1808.0,
+                    871.0,
+                    1808.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1037.0,
+                    1770.0,
+                    1405.0,
+                    1770.0,
+                    1405.0,
+                    1808.0,
+                    1037.0,
+                    1808.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1803.0,
+                    1405.0,
+                    1803.0,
+                    1405.0,
+                    1836.0,
+                    293.0,
+                    1836.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1834.0,
+                    682.0,
+                    1834.0,
+                    682.0,
+                    1871.0,
+                    293.0,
+                    1871.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1878.0,
+                    1406.0,
+                    1878.0,
+                    1406.0,
+                    1917.0,
+                    294.0,
+                    1917.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1911.0,
+                    1404.0,
+                    1911.0,
+                    1404.0,
+                    1944.0,
+                    295.0,
+                    1944.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    1943.0,
+                    1404.0,
+                    1943.0,
+                    1404.0,
+                    1976.0,
+                    297.0,
+                    1976.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1969.0,
+                    1406.0,
+                    1969.0,
+                    1406.0,
+                    2009.0,
+                    294.0,
+                    2009.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    2004.0,
+                    978.0,
+                    2004.0,
+                    978.0,
+                    2037.0,
+                    297.0,
+                    2037.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1452.0,
+                    1405.0,
+                    1452.0,
+                    1405.0,
+                    1488.0,
+                    293.0,
+                    1488.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1482.0,
+                    1405.0,
+                    1482.0,
+                    1405.0,
+                    1518.0,
+                    295.0,
+                    1518.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1513.0,
+                    1405.0,
+                    1513.0,
+                    1405.0,
+                    1549.0,
+                    295.0,
+                    1549.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1542.0,
+                    1405.0,
+                    1542.0,
+                    1405.0,
+                    1578.0,
+                    295.0,
+                    1578.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1571.0,
+                    1405.0,
+                    1571.0,
+                    1405.0,
+                    1610.0,
+                    292.0,
+                    1610.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1605.0,
+                    1330.0,
+                    1605.0,
+                    1330.0,
+                    1640.0,
+                    293.0,
+                    1640.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 7,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 5,
+                "poly": [
+                    359,
+                    1629,
+                    1341,
+                    1629,
+                    1341,
+                    1997,
+                    359,
+                    1997
+                ],
+                "score": 0.984,
+                "html": "<table><tr><td>Algorithm</td><td>(5,1)</td><td>(5,5)</td><td>(20,1)</td><td>(20,5)</td></tr><tr><td>Random-MAML,P= 2400,k = 500</td><td>25.99%</td><td>25.74%</td><td>6.51%</td><td>6.74%</td></tr><tr><td>Pixels CACTUs-MAML,P=1, k = 500</td><td>30.55%</td><td>40.19%</td><td>12.05%</td><td>19.01%</td></tr><tr><td>ACAIHyperplanes-MAML,P= 2400,m= 0</td><td>62.34%</td><td>81.81%</td><td>39.30%</td><td>63.18%</td></tr><tr><td>ACAI Hyperplanes-MAML,P= 2400,m =1.2</td><td>62.44%</td><td>83.20%</td><td>41.86%</td><td>65.23%</td></tr><tr><td>ACAI CACTUs-MAML,P=1,k = 500</td><td>66.49%</td><td>85.60%</td><td>45.04%</td><td>69.14%</td></tr><tr><td>ACAICACTUs-MAML,P=100,k = 500</td><td>68.84%</td><td>87.78%</td><td>48.09%</td><td>73.36%</td></tr><tr><td>BiGANHyperplanes-MAML,P= 2400,m = 0</td><td>53.60%</td><td>74.60%</td><td>29.02%</td><td>50.77%</td></tr><tr><td>BiGANHyperplanes-MAML,P= 2400,m = 0.5</td><td>53.18%</td><td>73.55%</td><td>29.98%</td><td>50.14%</td></tr><tr><td>BiGANCACTUs-MAML,P=1,k= 500</td><td>55.92%</td><td>76.28%</td><td>32.44%</td><td>54.22%</td></tr><tr><td>BiGANCACTUs-MAML,P=100,k= 500</td><td>58.18%</td><td>78.66%</td><td>35.56%</td><td>58.62%</td></tr></table>"
+            },
+            {
+                "category_id": 5,
+                "poly": [
+                    361,
+                    352,
+                    1340,
+                    352,
+                    1340,
+                    899,
+                    361,
+                    899
+                ],
+                "score": 0.984,
+                "html": "<table><tr><td></td><td colspan=\"4\">miniImageNet</td><td>CelebA</td></tr><tr><td></td><td>(5,1)</td><td>(5,5)</td><td>(5,20)</td><td>(5,50)</td><td>(2,5)</td></tr><tr><td>Training from scratch</td><td>27.59%</td><td>38.48%</td><td>51.53%</td><td>59.63%</td><td>63.19%</td></tr><tr><td>BiGAN knn-nearest neighbors</td><td>25.56%</td><td>31.10%</td><td>37.31%</td><td>43.60%</td><td>56.15%</td></tr><tr><td>BiGAN linear classifier</td><td>27.08%</td><td>33.91%</td><td>44.00%</td><td>50.41%</td><td>58.44%</td></tr><tr><td>BiGAN MLP with dropout</td><td>22.91%</td><td>29.06%</td><td>40.06%</td><td>48.36%</td><td>56.26%</td></tr><tr><td>BiGAN cluster matching</td><td>24.63%</td><td>29.49%</td><td>33.89%</td><td>36.13%</td><td>56.20%</td></tr><tr><td>BiGAN CACTUs-MAML (ours)</td><td>36.24%</td><td>51.28%</td><td>61.33%</td><td>66.91%</td><td>74.98%</td></tr><tr><td>BiGAN CACTUs-ProtoNets (ours)</td><td>36.62%</td><td>50.16%</td><td>59.56%</td><td>63.27%</td><td>65.58%</td></tr><tr><td>DeepCluster knn-nearest neighbors</td><td>28.90%</td><td>42.25%</td><td>56.44%</td><td>63.90%</td><td>61.47%</td></tr><tr><td>DeepCluster linear classifier</td><td>29.44%</td><td>39.79%</td><td>56.19%</td><td>65.28%</td><td>59.57%</td></tr><tr><td>DeepCluster MLP with dropout</td><td>29.03%</td><td>39.67%</td><td>52.71%</td><td>60.95%</td><td>60.65%</td></tr><tr><td>DeepCluster cluster matching</td><td>22.20%</td><td>23.50%</td><td>24.97%</td><td>26.87%</td><td>51.51%</td></tr><tr><td>DeepCluster CACTUs-MAML (ours)</td><td>39.90%</td><td>53.97%</td><td>63.84%</td><td>69.64%</td><td>73.79%</td></tr><tr><td>DeepCluster CACTUs-ProtoNets (ours)</td><td>39.18%</td><td>53.36%</td><td>61.54%</td><td>63.55%</td><td>74.15%</td></tr><tr><td>Oracle-MAML (control)</td><td>46.81%</td><td>62.13%</td><td>71.03%</td><td>75.54%</td><td>87.10%</td></tr><tr><td>Oracle-ProtoNets (control)</td><td>46.56%</td><td>62.29%</td><td>70.05%</td><td>72.04%</td><td>85.13%</td></tr></table>"
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    953,
+                    1404,
+                    953,
+                    1404,
+                    1290,
+                    297,
+                    1290
+                ],
+                "score": 0.981
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1305,
+                    1404,
+                    1305,
+                    1404,
+                    1521,
+                    297,
+                    1521
+                ],
+                "score": 0.975
+            },
+            {
+                "category_id": 6,
+                "poly": [
+                    298,
+                    224,
+                    1403,
+                    224,
+                    1403,
+                    337,
+                    298,
+                    337
+                ],
+                "score": 0.964
+            },
+            {
+                "category_id": 6,
+                "poly": [
+                    297,
+                    1558,
+                    1401,
+                    1558,
+                    1401,
+                    1615,
+                    297,
+                    1615
+                ],
+                "score": 0.922
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    300,
+                    76,
+                    815,
+                    76,
+                    815,
+                    104,
+                    300,
+                    104
+                ],
+                "score": 0.899
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    840,
+                    2088,
+                    859,
+                    2088,
+                    859,
+                    2111,
+                    840,
+                    2111
+                ],
+                "score": 0.811
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    874,
+                    252,
+                    967,
+                    252,
+                    967,
+                    278,
+                    874,
+                    278
+                ],
+                "score": 0.9,
+                "latex": "k = 5 0 0"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1085,
+                    252,
+                    1170,
+                    252,
+                    1170,
+                    278,
+                    1085,
+                    278
+                ],
+                "score": 0.9,
+                "latex": "P = 5 0"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    493,
+                    652,
+                    529,
+                    652,
+                    529,
+                    678,
+                    493,
+                    678
+                ],
+                "score": 0.84,
+                "latex": "k _ { \\mathrm { n n } } ."
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    447,
+                    471,
+                    482,
+                    471,
+                    482,
+                    500,
+                    447,
+                    500
+                ],
+                "score": 0.79,
+                "latex": "k _ { \\mathrm { n n } } ."
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    629,
+                    281,
+                    646,
+                    281,
+                    646,
+                    305,
+                    629,
+                    305
+                ],
+                "score": 0.71,
+                "latex": "k"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    544,
+                    1683,
+                    652,
+                    1683,
+                    652,
+                    1709,
+                    544,
+                    1709
+                ],
+                "score": 0.68,
+                "latex": "P = 2 4 0 0"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    662,
+                    1684,
+                    753,
+                    1684,
+                    753,
+                    1709,
+                    662,
+                    1709
+                ],
+                "score": 0.48,
+                "latex": "k = 5 0 0"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    619,
+                    1712,
+                    689,
+                    1712,
+                    689,
+                    1737,
+                    619,
+                    1737
+                ],
+                "score": 0.37,
+                "latex": "P = 1"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    619,
+                    1808,
+                    688,
+                    1808,
+                    688,
+                    1832,
+                    619,
+                    1832
+                ],
+                "score": 0.33,
+                "latex": "P = 1"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    788,
+                    1908,
+                    879,
+                    1908,
+                    879,
+                    1928,
+                    788,
+                    1928
+                ],
+                "score": 0.32,
+                "latex": "m = 0 . 5"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    619,
+                    1834,
+                    714,
+                    1834,
+                    714,
+                    1859,
+                    619,
+                    1859
+                ],
+                "score": 0.28,
+                "latex": "P = 1 0 0"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    222.0,
+                    1404.0,
+                    222.0,
+                    1404.0,
+                    258.0,
+                    294.0,
+                    258.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    247.0,
+                    873.0,
+                    247.0,
+                    873.0,
+                    287.0,
+                    292.0,
+                    287.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    968.0,
+                    247.0,
+                    1084.0,
+                    247.0,
+                    1084.0,
+                    287.0,
+                    968.0,
+                    287.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1171.0,
+                    247.0,
+                    1405.0,
+                    247.0,
+                    1405.0,
+                    287.0,
+                    1171.0,
+                    287.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    278.0,
+                    628.0,
+                    278.0,
+                    628.0,
+                    311.0,
+                    294.0,
+                    311.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    647.0,
+                    278.0,
+                    1405.0,
+                    278.0,
+                    1405.0,
+                    311.0,
+                    647.0,
+                    311.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    306.0,
+                    430.0,
+                    306.0,
+                    430.0,
+                    340.0,
+                    294.0,
+                    340.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1558.0,
+                    1403.0,
+                    1558.0,
+                    1403.0,
+                    1589.0,
+                    296.0,
+                    1589.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1585.0,
+                    784.0,
+                    1585.0,
+                    784.0,
+                    1616.0,
+                    295.0,
+                    1616.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    73.0,
+                    817.0,
+                    73.0,
+                    817.0,
+                    108.0,
+                    296.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    839.0,
+                    2087.0,
+                    861.0,
+                    2087.0,
+                    861.0,
+                    2117.0,
+                    839.0,
+                    2117.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    954.0,
+                    1404.0,
+                    954.0,
+                    1404.0,
+                    989.0,
+                    295.0,
+                    989.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    985.0,
+                    1404.0,
+                    985.0,
+                    1404.0,
+                    1020.0,
+                    295.0,
+                    1020.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1016.0,
+                    1404.0,
+                    1016.0,
+                    1404.0,
+                    1051.0,
+                    295.0,
+                    1051.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1043.0,
+                    1404.0,
+                    1043.0,
+                    1404.0,
+                    1081.0,
+                    292.0,
+                    1081.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1072.0,
+                    1406.0,
+                    1072.0,
+                    1406.0,
+                    1113.0,
+                    292.0,
+                    1113.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1107.0,
+                    1405.0,
+                    1107.0,
+                    1405.0,
+                    1140.0,
+                    292.0,
+                    1140.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1136.0,
+                    1404.0,
+                    1136.0,
+                    1404.0,
+                    1171.0,
+                    295.0,
+                    1171.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1167.0,
+                    1405.0,
+                    1167.0,
+                    1405.0,
+                    1202.0,
+                    295.0,
+                    1202.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1197.0,
+                    1404.0,
+                    1197.0,
+                    1404.0,
+                    1232.0,
+                    294.0,
+                    1232.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1227.0,
+                    1405.0,
+                    1227.0,
+                    1405.0,
+                    1264.0,
+                    294.0,
+                    1264.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1258.0,
+                    580.0,
+                    1258.0,
+                    580.0,
+                    1293.0,
+                    293.0,
+                    1293.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1302.0,
+                    1404.0,
+                    1302.0,
+                    1404.0,
+                    1341.0,
+                    294.0,
+                    1341.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1335.0,
+                    1404.0,
+                    1335.0,
+                    1404.0,
+                    1372.0,
+                    294.0,
+                    1372.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1367.0,
+                    1404.0,
+                    1367.0,
+                    1404.0,
+                    1402.0,
+                    295.0,
+                    1402.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1393.0,
+                    1405.0,
+                    1393.0,
+                    1405.0,
+                    1435.0,
+                    291.0,
+                    1435.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1427.0,
+                    1402.0,
+                    1427.0,
+                    1402.0,
+                    1461.0,
+                    294.0,
+                    1461.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1459.0,
+                    1406.0,
+                    1459.0,
+                    1406.0,
+                    1491.0,
+                    293.0,
+                    1491.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1491.0,
+                    1361.0,
+                    1491.0,
+                    1361.0,
+                    1522.0,
+                    296.0,
+                    1522.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 8,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    950,
+                    1404,
+                    950,
+                    1404,
+                    1346,
+                    298,
+                    1346
+                ],
+                "score": 0.984
+            },
+            {
+                "category_id": 5,
+                "poly": [
+                    334,
+                    296,
+                    1364,
+                    296,
+                    1364,
+                    596,
+                    334,
+                    596
+                ],
+                "score": 0.983,
+                "html": "<table><tr><td>Algorithm</td><td>(5,1)</td><td>(5,5)</td><td>(5,20)</td><td>(5,50)</td></tr><tr><td>BiGANHyperplanes-MAML,P= 4800,m =0</td><td>20.00%</td><td>20.00%</td><td>20.00%</td><td>20.00%</td></tr><tr><td>BiGAN Hyperplanes-MAML,P= 4800,m = 0.9</td><td>29.67%</td><td>41.92%</td><td>51.32%</td><td>54.72%</td></tr><tr><td>BiGAN CACTUs-MAML,P=1,k= 500</td><td>37.75%</td><td>52.59%</td><td>62.70%</td><td>67.98%</td></tr><tr><td>BiGANCACTUs-MAML,P= 50,k= 500</td><td>36.24%</td><td>51.28%</td><td>61.33%</td><td>66.91%</td></tr><tr><td>DeepCluster Hyperplanes-MAML,P= 4800,m = 0</td><td>20.02%</td><td>20.01%</td><td>20.00%</td><td>20.01%</td></tr><tr><td>DeepClusterHyperplanes-MAML,P= 4800,m= 0.1</td><td>35.85%</td><td>49.54%</td><td>60.68%</td><td>65.55%</td></tr><tr><td>DeepCluster CACTUs-MAML,P= 1,k = 500</td><td>38.75%</td><td>52.73%</td><td>62.72%</td><td>67.77%</td></tr><tr><td>DeepCluster CACTUs-MAML,P= 50,k = 500</td><td>39.90%</td><td>53.97%</td><td>63.84%</td><td>69.64%</td></tr></table>"
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1592,
+                    1403,
+                    1592,
+                    1403,
+                    1807,
+                    298,
+                    1807
+                ],
+                "score": 0.981
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1362,
+                    1403,
+                    1362,
+                    1403,
+                    1577,
+                    298,
+                    1577
+                ],
+                "score": 0.98
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1822,
+                    1403,
+                    1822,
+                    1403,
+                    2036,
+                    298,
+                    2036
+                ],
+                "score": 0.979
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    780,
+                    1403,
+                    780,
+                    1403,
+                    936,
+                    297,
+                    936
+                ],
+                "score": 0.975
+            },
+            {
+                "category_id": 6,
+                "poly": [
+                    299,
+                    224,
+                    1399,
+                    224,
+                    1399,
+                    282,
+                    299,
+                    282
+                ],
+                "score": 0.934
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    298,
+                    685,
+                    530,
+                    685,
+                    530,
+                    721,
+                    298,
+                    721
+                ],
+                "score": 0.914
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    300,
+                    75,
+                    815,
+                    75,
+                    815,
+                    105,
+                    300,
+                    105
+                ],
+                "score": 0.897
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    837,
+                    2088,
+                    865,
+                    2088,
+                    865,
+                    2112,
+                    837,
+                    2112
+                ],
+                "score": 0.852
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    885,
+                    1395,
+                    904,
+                    1395,
+                    904,
+                    1421,
+                    885,
+                    1421
+                ],
+                "score": 0.83,
+                "latex": "k"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    689,
+                    1364,
+                    707,
+                    1364,
+                    707,
+                    1390,
+                    689,
+                    1390
+                ],
+                "score": 0.82,
+                "latex": "k"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    613,
+                    434,
+                    696,
+                    434,
+                    696,
+                    459,
+                    613,
+                    459
+                ],
+                "score": 0.55,
+                "latex": "P = 5 0"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    661,
+                    529,
+                    729,
+                    529,
+                    729,
+                    554,
+                    661,
+                    554
+                ],
+                "score": 0.52,
+                "latex": "P = 1"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    613,
+                    406,
+                    683,
+                    406,
+                    683,
+                    431,
+                    613,
+                    431
+                ],
+                "score": 0.46,
+                "latex": "P = 1"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    646,
+                    349,
+                    753,
+                    349,
+                    753,
+                    375,
+                    646,
+                    375
+                ],
+                "score": 0.35,
+                "latex": "P = 4 8 0 0"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    693,
+                    501,
+                    801,
+                    501,
+                    801,
+                    526,
+                    693,
+                    526
+                ],
+                "score": 0.34,
+                "latex": "P = 4 8 0 0"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    740,
+                    530,
+                    830,
+                    530,
+                    830,
+                    553,
+                    740,
+                    553
+                ],
+                "score": 0.26,
+                "latex": "k = 5 0 0"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    225.0,
+                    1402.0,
+                    225.0,
+                    1402.0,
+                    255.0,
+                    296.0,
+                    255.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    251.0,
+                    834.0,
+                    251.0,
+                    834.0,
+                    281.0,
+                    296.0,
+                    281.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    684.0,
+                    534.0,
+                    684.0,
+                    534.0,
+                    729.0,
+                    292.0,
+                    729.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    73.0,
+                    817.0,
+                    73.0,
+                    817.0,
+                    108.0,
+                    296.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    831.0,
+                    2084.0,
+                    871.0,
+                    2084.0,
+                    871.0,
+                    2125.0,
+                    831.0,
+                    2125.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    949.0,
+                    1405.0,
+                    949.0,
+                    1405.0,
+                    985.0,
+                    294.0,
+                    985.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    979.0,
+                    1404.0,
+                    979.0,
+                    1404.0,
+                    1016.0,
+                    293.0,
+                    1016.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1014.0,
+                    1404.0,
+                    1014.0,
+                    1404.0,
+                    1046.0,
+                    296.0,
+                    1046.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1041.0,
+                    1404.0,
+                    1041.0,
+                    1404.0,
+                    1077.0,
+                    293.0,
+                    1077.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1072.0,
+                    1402.0,
+                    1072.0,
+                    1402.0,
+                    1108.0,
+                    293.0,
+                    1108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1104.0,
+                    1404.0,
+                    1104.0,
+                    1404.0,
+                    1136.0,
+                    294.0,
+                    1136.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1135.0,
+                    1404.0,
+                    1135.0,
+                    1404.0,
+                    1167.0,
+                    294.0,
+                    1167.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1164.0,
+                    1404.0,
+                    1164.0,
+                    1404.0,
+                    1200.0,
+                    294.0,
+                    1200.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1192.0,
+                    1404.0,
+                    1192.0,
+                    1404.0,
+                    1229.0,
+                    292.0,
+                    1229.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1220.0,
+                    1405.0,
+                    1220.0,
+                    1405.0,
+                    1262.0,
+                    292.0,
+                    1262.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1255.0,
+                    1402.0,
+                    1255.0,
+                    1402.0,
+                    1291.0,
+                    293.0,
+                    1291.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1281.0,
+                    1406.0,
+                    1281.0,
+                    1406.0,
+                    1324.0,
+                    292.0,
+                    1324.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1317.0,
+                    1146.0,
+                    1317.0,
+                    1146.0,
+                    1350.0,
+                    294.0,
+                    1350.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1592.0,
+                    1405.0,
+                    1592.0,
+                    1405.0,
+                    1627.0,
+                    294.0,
+                    1627.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1623.0,
+                    1405.0,
+                    1623.0,
+                    1405.0,
+                    1658.0,
+                    294.0,
+                    1658.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1654.0,
+                    1405.0,
+                    1654.0,
+                    1405.0,
+                    1688.0,
+                    294.0,
+                    1688.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1684.0,
+                    1405.0,
+                    1684.0,
+                    1405.0,
+                    1718.0,
+                    294.0,
+                    1718.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1713.0,
+                    1403.0,
+                    1713.0,
+                    1403.0,
+                    1749.0,
+                    292.0,
+                    1749.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1744.0,
+                    1403.0,
+                    1744.0,
+                    1403.0,
+                    1781.0,
+                    293.0,
+                    1781.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1776.0,
+                    996.0,
+                    1776.0,
+                    996.0,
+                    1810.0,
+                    294.0,
+                    1810.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1360.0,
+                    688.0,
+                    1360.0,
+                    688.0,
+                    1396.0,
+                    294.0,
+                    1396.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    708.0,
+                    1360.0,
+                    1404.0,
+                    1360.0,
+                    1404.0,
+                    1396.0,
+                    708.0,
+                    1396.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1394.0,
+                    884.0,
+                    1394.0,
+                    884.0,
+                    1425.0,
+                    296.0,
+                    1425.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    905.0,
+                    1394.0,
+                    1403.0,
+                    1394.0,
+                    1403.0,
+                    1425.0,
+                    905.0,
+                    1425.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1424.0,
+                    1405.0,
+                    1424.0,
+                    1405.0,
+                    1458.0,
+                    293.0,
+                    1458.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1451.0,
+                    1404.0,
+                    1451.0,
+                    1404.0,
+                    1492.0,
+                    292.0,
+                    1492.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1484.0,
+                    1404.0,
+                    1484.0,
+                    1404.0,
+                    1519.0,
+                    294.0,
+                    1519.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1515.0,
+                    1405.0,
+                    1515.0,
+                    1405.0,
+                    1549.0,
+                    294.0,
+                    1549.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1544.0,
+                    799.0,
+                    1544.0,
+                    799.0,
+                    1583.0,
+                    293.0,
+                    1583.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1822.0,
+                    1404.0,
+                    1822.0,
+                    1404.0,
+                    1857.0,
+                    296.0,
+                    1857.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1852.0,
+                    1403.0,
+                    1852.0,
+                    1403.0,
+                    1889.0,
+                    293.0,
+                    1889.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1882.0,
+                    1404.0,
+                    1882.0,
+                    1404.0,
+                    1919.0,
+                    293.0,
+                    1919.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1913.0,
+                    1405.0,
+                    1913.0,
+                    1405.0,
+                    1950.0,
+                    294.0,
+                    1950.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1943.0,
+                    1407.0,
+                    1943.0,
+                    1407.0,
+                    1979.0,
+                    293.0,
+                    1979.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1974.0,
+                    1407.0,
+                    1974.0,
+                    1407.0,
+                    2012.0,
+                    294.0,
+                    2012.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    2005.0,
+                    1358.0,
+                    2005.0,
+                    1358.0,
+                    2039.0,
+                    293.0,
+                    2039.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    778.0,
+                    1404.0,
+                    778.0,
+                    1404.0,
+                    817.0,
+                    295.0,
+                    817.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    813.0,
+                    1404.0,
+                    813.0,
+                    1404.0,
+                    846.0,
+                    296.0,
+                    846.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    843.0,
+                    1405.0,
+                    843.0,
+                    1405.0,
+                    877.0,
+                    295.0,
+                    877.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    873.0,
+                    1401.0,
+                    873.0,
+                    1401.0,
+                    906.0,
+                    295.0,
+                    906.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    903.0,
+                    1394.0,
+                    903.0,
+                    1394.0,
+                    937.0,
+                    296.0,
+                    937.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 9,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    286,
+                    1401,
+                    286,
+                    1401,
+                    350,
+                    297,
+                    350
+                ],
+                "score": 0.924
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    299,
+                    75,
+                    816,
+                    75,
+                    816,
+                    105,
+                    299,
+                    105
+                ],
+                "score": 0.884
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    300,
+                    1663,
+                    1403,
+                    1663,
+                    1403,
+                    1758,
+                    300,
+                    1758
+                ],
+                "score": 0.816
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    291,
+                    1222,
+                    1401,
+                    1222,
+                    1401,
+                    1289,
+                    291,
+                    1289
+                ],
+                "score": 0.81
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    836,
+                    2088,
+                    863,
+                    2088,
+                    863,
+                    2113,
+                    836,
+                    2113
+                ],
+                "score": 0.806
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    293,
+                    1304,
+                    1400,
+                    1304,
+                    1400,
+                    1370,
+                    293,
+                    1370
+                ],
+                "score": 0.803
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    295,
+                    804,
+                    1403,
+                    804,
+                    1403,
+                    869,
+                    295,
+                    869
+                ],
+                "score": 0.803
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    300,
+                    445,
+                    1400,
+                    445,
+                    1400,
+                    510,
+                    300,
+                    510
+                ],
+                "score": 0.799
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    885,
+                    1400,
+                    885,
+                    1400,
+                    979,
+                    299,
+                    979
+                ],
+                "score": 0.794
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    1110,
+                    1402,
+                    1110,
+                    1402,
+                    1206,
+                    299,
+                    1206
+                ],
+                "score": 0.788
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    292,
+                    1386,
+                    1401,
+                    1386,
+                    1401,
+                    1453,
+                    292,
+                    1453
+                ],
+                "score": 0.759
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    302,
+                    1550,
+                    1402,
+                    1550,
+                    1402,
+                    1646,
+                    302,
+                    1646
+                ],
+                "score": 0.749
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    295,
+                    1468,
+                    1402,
+                    1468,
+                    1402,
+                    1534,
+                    295,
+                    1534
+                ],
+                "score": 0.748
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    302,
+                    997,
+                    1402,
+                    997,
+                    1402,
+                    1094,
+                    302,
+                    1094
+                ],
+                "score": 0.739
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    690,
+                    1403,
+                    690,
+                    1403,
+                    784,
+                    298,
+                    784
+                ],
+                "score": 0.732
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    608,
+                    1400,
+                    608,
+                    1400,
+                    675,
+                    298,
+                    675
+                ],
+                "score": 0.728
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    296,
+                    526,
+                    1401,
+                    526,
+                    1401,
+                    593,
+                    296,
+                    593
+                ],
+                "score": 0.726
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    301,
+                    228,
+                    608,
+                    228,
+                    608,
+                    262,
+                    301,
+                    262
+                ],
+                "score": 0.722
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    300,
+                    1971,
+                    1400,
+                    1971,
+                    1400,
+                    2034,
+                    300,
+                    2034
+                ],
+                "score": 0.719
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    299,
+                    395,
+                    489,
+                    395,
+                    489,
+                    430,
+                    299,
+                    430
+                ],
+                "score": 0.691
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1889,
+                    1400,
+                    1889,
+                    1400,
+                    1953,
+                    297,
+                    1953
+                ],
+                "score": 0.689
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1776,
+                    1404,
+                    1776,
+                    1404,
+                    1870,
+                    297,
+                    1870
+                ],
+                "score": 0.68
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    395,
+                    489,
+                    395,
+                    489,
+                    430,
+                    299,
+                    430
+                ],
+                "score": 0.195
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    301,
+                    228,
+                    608,
+                    228,
+                    608,
+                    262,
+                    301,
+                    262
+                ],
+                "score": 0.151
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    73.0,
+                    817.0,
+                    73.0,
+                    817.0,
+                    108.0,
+                    295.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    832.0,
+                    2085.0,
+                    868.0,
+                    2085.0,
+                    868.0,
+                    2125.0,
+                    832.0,
+                    2125.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    298.0,
+                    227.0,
+                    610.0,
+                    227.0,
+                    610.0,
+                    267.0,
+                    298.0,
+                    267.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    398.0,
+                    490.0,
+                    398.0,
+                    490.0,
+                    431.0,
+                    297.0,
+                    431.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    283.0,
+                    1405.0,
+                    283.0,
+                    1405.0,
+                    323.0,
+                    295.0,
+                    323.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    314.0,
+                    1317.0,
+                    314.0,
+                    1317.0,
+                    356.0,
+                    293.0,
+                    356.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1662.0,
+                    1405.0,
+                    1662.0,
+                    1405.0,
+                    1701.0,
+                    294.0,
+                    1701.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    321.0,
+                    1695.0,
+                    1405.0,
+                    1695.0,
+                    1405.0,
+                    1732.0,
+                    321.0,
+                    1732.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    321.0,
+                    1725.0,
+                    710.0,
+                    1725.0,
+                    710.0,
+                    1759.0,
+                    321.0,
+                    1759.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1223.0,
+                    1404.0,
+                    1223.0,
+                    1404.0,
+                    1259.0,
+                    295.0,
+                    1259.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    1254.0,
+                    808.0,
+                    1254.0,
+                    808.0,
+                    1288.0,
+                    322.0,
+                    1288.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1303.0,
+                    1403.0,
+                    1303.0,
+                    1403.0,
+                    1343.0,
+                    295.0,
+                    1343.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    1337.0,
+                    669.0,
+                    1337.0,
+                    669.0,
+                    1371.0,
+                    322.0,
+                    1371.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    803.0,
+                    1406.0,
+                    803.0,
+                    1406.0,
+                    843.0,
+                    294.0,
+                    843.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    324.0,
+                    834.0,
+                    874.0,
+                    834.0,
+                    874.0,
+                    869.0,
+                    324.0,
+                    869.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    441.0,
+                    1404.0,
+                    441.0,
+                    1404.0,
+                    485.0,
+                    295.0,
+                    485.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    321.0,
+                    476.0,
+                    1000.0,
+                    476.0,
+                    1000.0,
+                    511.0,
+                    321.0,
+                    511.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    883.0,
+                    1402.0,
+                    883.0,
+                    1402.0,
+                    924.0,
+                    294.0,
+                    924.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    919.0,
+                    1404.0,
+                    919.0,
+                    1404.0,
+                    953.0,
+                    322.0,
+                    953.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    319.0,
+                    944.0,
+                    397.0,
+                    944.0,
+                    397.0,
+                    982.0,
+                    319.0,
+                    982.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1109.0,
+                    1404.0,
+                    1109.0,
+                    1404.0,
+                    1148.0,
+                    293.0,
+                    1148.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    1142.0,
+                    1403.0,
+                    1142.0,
+                    1403.0,
+                    1176.0,
+                    323.0,
+                    1176.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    1173.0,
+                    486.0,
+                    1173.0,
+                    486.0,
+                    1208.0,
+                    322.0,
+                    1208.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1383.0,
+                    1404.0,
+                    1383.0,
+                    1404.0,
+                    1427.0,
+                    292.0,
+                    1427.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    1418.0,
+                    1101.0,
+                    1418.0,
+                    1101.0,
+                    1455.0,
+                    323.0,
+                    1455.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    298.0,
+                    1553.0,
+                    1402.0,
+                    1553.0,
+                    1402.0,
+                    1587.0,
+                    298.0,
+                    1587.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    321.0,
+                    1582.0,
+                    1406.0,
+                    1582.0,
+                    1406.0,
+                    1620.0,
+                    321.0,
+                    1620.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    1614.0,
+                    778.0,
+                    1614.0,
+                    778.0,
+                    1648.0,
+                    323.0,
+                    1648.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1466.0,
+                    1407.0,
+                    1466.0,
+                    1407.0,
+                    1508.0,
+                    292.0,
+                    1508.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    324.0,
+                    1500.0,
+                    948.0,
+                    1500.0,
+                    948.0,
+                    1535.0,
+                    324.0,
+                    1535.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    999.0,
+                    1405.0,
+                    999.0,
+                    1405.0,
+                    1033.0,
+                    296.0,
+                    1033.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    1031.0,
+                    1405.0,
+                    1031.0,
+                    1405.0,
+                    1065.0,
+                    323.0,
+                    1065.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    1060.0,
+                    930.0,
+                    1060.0,
+                    930.0,
+                    1095.0,
+                    323.0,
+                    1095.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    687.0,
+                    1404.0,
+                    687.0,
+                    1404.0,
+                    730.0,
+                    292.0,
+                    730.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    724.0,
+                    1405.0,
+                    724.0,
+                    1405.0,
+                    758.0,
+                    322.0,
+                    758.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    320.0,
+                    750.0,
+                    399.0,
+                    750.0,
+                    399.0,
+                    787.0,
+                    320.0,
+                    787.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    608.0,
+                    1404.0,
+                    608.0,
+                    1404.0,
+                    645.0,
+                    296.0,
+                    645.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    321.0,
+                    640.0,
+                    1387.0,
+                    640.0,
+                    1387.0,
+                    677.0,
+                    321.0,
+                    677.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    523.0,
+                    1406.0,
+                    523.0,
+                    1406.0,
+                    567.0,
+                    294.0,
+                    567.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    324.0,
+                    558.0,
+                    1309.0,
+                    558.0,
+                    1309.0,
+                    595.0,
+                    324.0,
+                    595.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1969.0,
+                    1405.0,
+                    1969.0,
+                    1405.0,
+                    2009.0,
+                    295.0,
+                    2009.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    2000.0,
+                    846.0,
+                    2000.0,
+                    846.0,
+                    2036.0,
+                    322.0,
+                    2036.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1888.0,
+                    1404.0,
+                    1888.0,
+                    1404.0,
+                    1925.0,
+                    295.0,
+                    1925.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    1920.0,
+                    1263.0,
+                    1920.0,
+                    1263.0,
+                    1956.0,
+                    323.0,
+                    1956.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1776.0,
+                    1402.0,
+                    1776.0,
+                    1402.0,
+                    1813.0,
+                    296.0,
+                    1813.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    321.0,
+                    1806.0,
+                    1405.0,
+                    1806.0,
+                    1405.0,
+                    1846.0,
+                    321.0,
+                    1846.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    324.0,
+                    1839.0,
+                    672.0,
+                    1839.0,
+                    672.0,
+                    1873.0,
+                    324.0,
+                    1873.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    398.0,
+                    490.0,
+                    398.0,
+                    490.0,
+                    431.0,
+                    297.0,
+                    431.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    298.0,
+                    227.0,
+                    610.0,
+                    227.0,
+                    610.0,
+                    267.0,
+                    298.0,
+                    267.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 10,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 2,
+                "poly": [
+                    300,
+                    75,
+                    816,
+                    75,
+                    816,
+                    105,
+                    300,
+                    105
+                ],
+                "score": 0.881
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    836,
+                    2088,
+                    865,
+                    2088,
+                    865,
+                    2113,
+                    836,
+                    2113
+                ],
+                "score": 0.822
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    705,
+                    1400,
+                    705,
+                    1400,
+                    771,
+                    297,
+                    771
+                ],
+                "score": 0.739
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    872,
+                    1401,
+                    872,
+                    1401,
+                    938,
+                    298,
+                    938
+                ],
+                "score": 0.738
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    788,
+                    1398,
+                    788,
+                    1398,
+                    855,
+                    297,
+                    855
+                ],
+                "score": 0.738
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1152,
+                    1400,
+                    1152,
+                    1400,
+                    1217,
+                    297,
+                    1217
+                ],
+                "score": 0.731
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1069,
+                    1400,
+                    1069,
+                    1400,
+                    1134,
+                    298,
+                    1134
+                ],
+                "score": 0.716
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1349,
+                    1405,
+                    1349,
+                    1405,
+                    1473,
+                    298,
+                    1473
+                ],
+                "score": 0.706
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    305,
+                    1493,
+                    1399,
+                    1493,
+                    1399,
+                    1590,
+                    305,
+                    1590
+                ],
+                "score": 0.704
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    296,
+                    478,
+                    1403,
+                    478,
+                    1403,
+                    543,
+                    296,
+                    543
+                ],
+                "score": 0.676
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    296,
+                    1690,
+                    1400,
+                    1690,
+                    1400,
+                    1754,
+                    296,
+                    1754
+                ],
+                "score": 0.674
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    293,
+                    1607,
+                    1401,
+                    1607,
+                    1401,
+                    1672,
+                    293,
+                    1672
+                ],
+                "score": 0.65
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1234,
+                    1403,
+                    1234,
+                    1403,
+                    1330,
+                    298,
+                    1330
+                ],
+                "score": 0.643
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    292,
+                    228,
+                    1402,
+                    228,
+                    1402,
+                    294,
+                    292,
+                    294
+                ],
+                "score": 0.599
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    395,
+                    1402,
+                    395,
+                    1402,
+                    460,
+                    299,
+                    460
+                ],
+                "score": 0.593
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    290,
+                    1774,
+                    1402,
+                    1774,
+                    1402,
+                    1838,
+                    290,
+                    1838
+                ],
+                "score": 0.59
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    296,
+                    561,
+                    1405,
+                    561,
+                    1405,
+                    686,
+                    296,
+                    686
+                ],
+                "score": 0.586
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    311,
+                    1399,
+                    311,
+                    1399,
+                    377,
+                    298,
+                    377
+                ],
+                "score": 0.565
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    296,
+                    955,
+                    1405,
+                    955,
+                    1405,
+                    1051,
+                    296,
+                    1051
+                ],
+                "score": 0.518
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1971,
+                    1397,
+                    1971,
+                    1397,
+                    2034,
+                    297,
+                    2034
+                ],
+                "score": 0.483
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1856,
+                    1399,
+                    1856,
+                    1399,
+                    1952,
+                    298,
+                    1952
+                ],
+                "score": 0.477
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    823,
+                    595,
+                    843,
+                    595,
+                    843,
+                    626,
+                    823,
+                    626
+                ],
+                "score": 0.84,
+                "latex": "\\beta"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    73.0,
+                    817.0,
+                    73.0,
+                    817.0,
+                    108.0,
+                    296.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    832.0,
+                    2085.0,
+                    870.0,
+                    2085.0,
+                    870.0,
+                    2124.0,
+                    832.0,
+                    2124.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    706.0,
+                    1405.0,
+                    706.0,
+                    1405.0,
+                    742.0,
+                    296.0,
+                    742.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    738.0,
+                    697.0,
+                    738.0,
+                    697.0,
+                    772.0,
+                    322.0,
+                    772.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    871.0,
+                    1405.0,
+                    871.0,
+                    1405.0,
+                    913.0,
+                    293.0,
+                    913.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    904.0,
+                    1403.0,
+                    904.0,
+                    1403.0,
+                    940.0,
+                    322.0,
+                    940.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    786.0,
+                    1403.0,
+                    786.0,
+                    1403.0,
+                    828.0,
+                    292.0,
+                    828.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    321.0,
+                    820.0,
+                    983.0,
+                    820.0,
+                    983.0,
+                    855.0,
+                    321.0,
+                    855.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1154.0,
+                    1404.0,
+                    1154.0,
+                    1404.0,
+                    1190.0,
+                    296.0,
+                    1190.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    1184.0,
+                    1392.0,
+                    1184.0,
+                    1392.0,
+                    1221.0,
+                    323.0,
+                    1221.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1067.0,
+                    1406.0,
+                    1067.0,
+                    1406.0,
+                    1109.0,
+                    293.0,
+                    1109.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    324.0,
+                    1099.0,
+                    1185.0,
+                    1099.0,
+                    1185.0,
+                    1135.0,
+                    324.0,
+                    1135.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1350.0,
+                    1403.0,
+                    1350.0,
+                    1403.0,
+                    1387.0,
+                    296.0,
+                    1387.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    320.0,
+                    1381.0,
+                    1406.0,
+                    1381.0,
+                    1406.0,
+                    1417.0,
+                    320.0,
+                    1417.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    317.0,
+                    1409.0,
+                    1406.0,
+                    1409.0,
+                    1406.0,
+                    1450.0,
+                    317.0,
+                    1450.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    320.0,
+                    1440.0,
+                    398.0,
+                    1440.0,
+                    398.0,
+                    1476.0,
+                    320.0,
+                    1476.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    299.0,
+                    1492.0,
+                    1403.0,
+                    1492.0,
+                    1403.0,
+                    1529.0,
+                    299.0,
+                    1529.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    1525.0,
+                    1404.0,
+                    1525.0,
+                    1404.0,
+                    1560.0,
+                    322.0,
+                    1560.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    1554.0,
+                    930.0,
+                    1554.0,
+                    930.0,
+                    1591.0,
+                    322.0,
+                    1591.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    480.0,
+                    1404.0,
+                    480.0,
+                    1404.0,
+                    516.0,
+                    296.0,
+                    516.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    321.0,
+                    511.0,
+                    1093.0,
+                    511.0,
+                    1093.0,
+                    544.0,
+                    321.0,
+                    544.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1691.0,
+                    1404.0,
+                    1691.0,
+                    1404.0,
+                    1727.0,
+                    296.0,
+                    1727.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    321.0,
+                    1721.0,
+                    602.0,
+                    1721.0,
+                    602.0,
+                    1757.0,
+                    321.0,
+                    1757.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1606.0,
+                    1404.0,
+                    1606.0,
+                    1404.0,
+                    1645.0,
+                    293.0,
+                    1645.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    324.0,
+                    1638.0,
+                    1305.0,
+                    1638.0,
+                    1305.0,
+                    1674.0,
+                    324.0,
+                    1674.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1234.0,
+                    1403.0,
+                    1234.0,
+                    1403.0,
+                    1273.0,
+                    292.0,
+                    1273.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    321.0,
+                    1266.0,
+                    1403.0,
+                    1266.0,
+                    1403.0,
+                    1302.0,
+                    321.0,
+                    1302.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    1298.0,
+                    488.0,
+                    1298.0,
+                    488.0,
+                    1333.0,
+                    323.0,
+                    1333.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    226.0,
+                    721.0,
+                    226.0,
+                    721.0,
+                    265.0,
+                    296.0,
+                    265.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    750.0,
+                    226.0,
+                    1408.0,
+                    226.0,
+                    1408.0,
+                    270.0,
+                    750.0,
+                    270.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    320.0,
+                    259.0,
+                    612.0,
+                    259.0,
+                    612.0,
+                    293.0,
+                    320.0,
+                    293.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    393.0,
+                    1407.0,
+                    393.0,
+                    1407.0,
+                    434.0,
+                    294.0,
+                    434.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    426.0,
+                    734.0,
+                    426.0,
+                    734.0,
+                    461.0,
+                    323.0,
+                    461.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1774.0,
+                    1405.0,
+                    1774.0,
+                    1405.0,
+                    1810.0,
+                    296.0,
+                    1810.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    321.0,
+                    1803.0,
+                    927.0,
+                    1803.0,
+                    927.0,
+                    1839.0,
+                    321.0,
+                    1839.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    564.0,
+                    1405.0,
+                    564.0,
+                    1405.0,
+                    597.0,
+                    295.0,
+                    597.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    592.0,
+                    822.0,
+                    592.0,
+                    822.0,
+                    628.0,
+                    323.0,
+                    628.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    844.0,
+                    592.0,
+                    1403.0,
+                    592.0,
+                    1403.0,
+                    628.0,
+                    844.0,
+                    628.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    622.0,
+                    1403.0,
+                    622.0,
+                    1403.0,
+                    660.0,
+                    322.0,
+                    660.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    654.0,
+                    397.0,
+                    654.0,
+                    397.0,
+                    687.0,
+                    322.0,
+                    687.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    310.0,
+                    1403.0,
+                    310.0,
+                    1403.0,
+                    349.0,
+                    294.0,
+                    349.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    343.0,
+                    1182.0,
+                    343.0,
+                    1182.0,
+                    378.0,
+                    323.0,
+                    378.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    955.0,
+                    1405.0,
+                    955.0,
+                    1405.0,
+                    993.0,
+                    294.0,
+                    993.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    324.0,
+                    987.0,
+                    1405.0,
+                    987.0,
+                    1405.0,
+                    1023.0,
+                    324.0,
+                    1023.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    1018.0,
+                    947.0,
+                    1018.0,
+                    947.0,
+                    1052.0,
+                    322.0,
+                    1052.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    1972.0,
+                    1400.0,
+                    1972.0,
+                    1400.0,
+                    2008.0,
+                    297.0,
+                    2008.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    2003.0,
+                    916.0,
+                    2003.0,
+                    916.0,
+                    2036.0,
+                    322.0,
+                    2036.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1857.0,
+                    1403.0,
+                    1857.0,
+                    1403.0,
+                    1892.0,
+                    296.0,
+                    1892.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    1887.0,
+                    1405.0,
+                    1887.0,
+                    1405.0,
+                    1927.0,
+                    322.0,
+                    1927.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    1919.0,
+                    672.0,
+                    1919.0,
+                    672.0,
+                    1953.0,
+                    323.0,
+                    1953.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 11,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 2,
+                "poly": [
+                    300,
+                    75,
+                    816,
+                    75,
+                    816,
+                    105,
+                    300,
+                    105
+                ],
+                "score": 0.884
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    295,
+                    1636,
+                    1403,
+                    1636,
+                    1403,
+                    1731,
+                    295,
+                    1731
+                ],
+                "score": 0.884
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    613,
+                    1401,
+                    613,
+                    1401,
+                    707,
+                    298,
+                    707
+                ],
+                "score": 0.868
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    229,
+                    1400,
+                    229,
+                    1400,
+                    323,
+                    299,
+                    323
+                ],
+                "score": 0.858
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    294,
+                    725,
+                    1403,
+                    725,
+                    1403,
+                    790,
+                    294,
+                    790
+                ],
+                "score": 0.854
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    292,
+                    1251,
+                    1402,
+                    1251,
+                    1402,
+                    1316,
+                    292,
+                    1316
+                ],
+                "score": 0.852
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    301,
+                    805,
+                    1402,
+                    805,
+                    1402,
+                    900,
+                    301,
+                    900
+                ],
+                "score": 0.851
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    294,
+                    340,
+                    1399,
+                    340,
+                    1399,
+                    405,
+                    294,
+                    405
+                ],
+                "score": 0.849
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    296,
+                    1139,
+                    1401,
+                    1139,
+                    1401,
+                    1235,
+                    296,
+                    1235
+                ],
+                "score": 0.848
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    1524,
+                    1403,
+                    1524,
+                    1403,
+                    1620,
+                    299,
+                    1620
+                ],
+                "score": 0.841
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    836,
+                    2088,
+                    864,
+                    2088,
+                    864,
+                    2113,
+                    836,
+                    2113
+                ],
+                "score": 0.83
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    290,
+                    1443,
+                    1402,
+                    1443,
+                    1402,
+                    1509,
+                    290,
+                    1509
+                ],
+                "score": 0.828
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1332,
+                    1403,
+                    1332,
+                    1403,
+                    1427,
+                    297,
+                    1427
+                ],
+                "score": 0.824
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    502,
+                    1402,
+                    502,
+                    1402,
+                    596,
+                    299,
+                    596
+                ],
+                "score": 0.822
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    917,
+                    1405,
+                    917,
+                    1405,
+                    1042,
+                    297,
+                    1042
+                ],
+                "score": 0.818
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    293,
+                    1059,
+                    1402,
+                    1059,
+                    1402,
+                    1124,
+                    293,
+                    1124
+                ],
+                "score": 0.816
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    294,
+                    420,
+                    1399,
+                    420,
+                    1399,
+                    487,
+                    294,
+                    487
+                ],
+                "score": 0.807
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    1941,
+                    1402,
+                    1941,
+                    1402,
+                    2034,
+                    299,
+                    2034
+                ],
+                "score": 0.749
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    296,
+                    1859,
+                    1398,
+                    1859,
+                    1398,
+                    1925,
+                    296,
+                    1925
+                ],
+                "score": 0.725
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    296,
+                    1747,
+                    1404,
+                    1747,
+                    1404,
+                    1843,
+                    296,
+                    1843
+                ],
+                "score": 0.705
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    73.0,
+                    817.0,
+                    73.0,
+                    817.0,
+                    108.0,
+                    296.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    832.0,
+                    2085.0,
+                    869.0,
+                    2085.0,
+                    869.0,
+                    2123.0,
+                    832.0,
+                    2123.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1635.0,
+                    1405.0,
+                    1635.0,
+                    1405.0,
+                    1674.0,
+                    293.0,
+                    1674.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    320.0,
+                    1668.0,
+                    1405.0,
+                    1668.0,
+                    1405.0,
+                    1704.0,
+                    320.0,
+                    1704.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    1699.0,
+                    601.0,
+                    1699.0,
+                    601.0,
+                    1735.0,
+                    323.0,
+                    1735.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    614.0,
+                    1403.0,
+                    614.0,
+                    1403.0,
+                    648.0,
+                    294.0,
+                    648.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    321.0,
+                    644.0,
+                    1403.0,
+                    644.0,
+                    1403.0,
+                    681.0,
+                    321.0,
+                    681.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    319.0,
+                    672.0,
+                    398.0,
+                    672.0,
+                    398.0,
+                    710.0,
+                    319.0,
+                    710.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    226.0,
+                    1405.0,
+                    226.0,
+                    1405.0,
+                    268.0,
+                    294.0,
+                    268.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    260.0,
+                    1404.0,
+                    260.0,
+                    1404.0,
+                    298.0,
+                    323.0,
+                    298.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    292.0,
+                    606.0,
+                    292.0,
+                    606.0,
+                    326.0,
+                    322.0,
+                    326.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    725.0,
+                    1405.0,
+                    725.0,
+                    1405.0,
+                    761.0,
+                    296.0,
+                    761.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    325.0,
+                    756.0,
+                    948.0,
+                    756.0,
+                    948.0,
+                    791.0,
+                    325.0,
+                    791.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1251.0,
+                    1405.0,
+                    1251.0,
+                    1405.0,
+                    1287.0,
+                    294.0,
+                    1287.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    324.0,
+                    1281.0,
+                    904.0,
+                    1281.0,
+                    904.0,
+                    1313.0,
+                    324.0,
+                    1313.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    804.0,
+                    1407.0,
+                    804.0,
+                    1407.0,
+                    843.0,
+                    294.0,
+                    843.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    321.0,
+                    835.0,
+                    1405.0,
+                    835.0,
+                    1405.0,
+                    874.0,
+                    321.0,
+                    874.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    319.0,
+                    865.0,
+                    399.0,
+                    865.0,
+                    399.0,
+                    901.0,
+                    319.0,
+                    901.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    338.0,
+                    1401.0,
+                    338.0,
+                    1401.0,
+                    378.0,
+                    294.0,
+                    378.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    371.0,
+                    904.0,
+                    371.0,
+                    904.0,
+                    407.0,
+                    323.0,
+                    407.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1137.0,
+                    1403.0,
+                    1137.0,
+                    1403.0,
+                    1178.0,
+                    294.0,
+                    1178.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    325.0,
+                    1175.0,
+                    1403.0,
+                    1175.0,
+                    1403.0,
+                    1205.0,
+                    325.0,
+                    1205.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    1202.0,
+                    601.0,
+                    1202.0,
+                    601.0,
+                    1236.0,
+                    323.0,
+                    1236.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1524.0,
+                    1404.0,
+                    1524.0,
+                    1404.0,
+                    1562.0,
+                    293.0,
+                    1562.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    321.0,
+                    1555.0,
+                    1405.0,
+                    1555.0,
+                    1405.0,
+                    1595.0,
+                    321.0,
+                    1595.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    324.0,
+                    1589.0,
+                    604.0,
+                    1589.0,
+                    604.0,
+                    1620.0,
+                    324.0,
+                    1620.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1444.0,
+                    1404.0,
+                    1444.0,
+                    1404.0,
+                    1480.0,
+                    293.0,
+                    1480.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    1475.0,
+                    932.0,
+                    1475.0,
+                    932.0,
+                    1511.0,
+                    322.0,
+                    1511.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1333.0,
+                    1401.0,
+                    1333.0,
+                    1401.0,
+                    1367.0,
+                    295.0,
+                    1367.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    1365.0,
+                    1404.0,
+                    1365.0,
+                    1404.0,
+                    1399.0,
+                    322.0,
+                    1399.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    324.0,
+                    1394.0,
+                    488.0,
+                    1394.0,
+                    488.0,
+                    1429.0,
+                    324.0,
+                    1429.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    503.0,
+                    1406.0,
+                    503.0,
+                    1406.0,
+                    537.0,
+                    295.0,
+                    537.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    320.0,
+                    533.0,
+                    1406.0,
+                    533.0,
+                    1406.0,
+                    571.0,
+                    320.0,
+                    571.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    324.0,
+                    564.0,
+                    485.0,
+                    564.0,
+                    485.0,
+                    599.0,
+                    324.0,
+                    599.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    917.0,
+                    1406.0,
+                    917.0,
+                    1406.0,
+                    955.0,
+                    294.0,
+                    955.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    321.0,
+                    947.0,
+                    1405.0,
+                    947.0,
+                    1405.0,
+                    983.0,
+                    321.0,
+                    983.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    320.0,
+                    978.0,
+                    1406.0,
+                    978.0,
+                    1406.0,
+                    1017.0,
+                    320.0,
+                    1017.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    320.0,
+                    1009.0,
+                    484.0,
+                    1009.0,
+                    484.0,
+                    1044.0,
+                    320.0,
+                    1044.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1058.0,
+                    1402.0,
+                    1058.0,
+                    1402.0,
+                    1097.0,
+                    295.0,
+                    1097.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    325.0,
+                    1091.0,
+                    1402.0,
+                    1091.0,
+                    1402.0,
+                    1124.0,
+                    325.0,
+                    1124.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    418.0,
+                    1403.0,
+                    418.0,
+                    1403.0,
+                    459.0,
+                    294.0,
+                    459.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    452.0,
+                    1327.0,
+                    452.0,
+                    1327.0,
+                    489.0,
+                    323.0,
+                    489.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1940.0,
+                    1406.0,
+                    1940.0,
+                    1406.0,
+                    1978.0,
+                    294.0,
+                    1978.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    1971.0,
+                    1402.0,
+                    1971.0,
+                    1402.0,
+                    2008.0,
+                    322.0,
+                    2008.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    319.0,
+                    2003.0,
+                    702.0,
+                    2003.0,
+                    702.0,
+                    2038.0,
+                    319.0,
+                    2038.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    1862.0,
+                    1400.0,
+                    1862.0,
+                    1400.0,
+                    1895.0,
+                    297.0,
+                    1895.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    321.0,
+                    1891.0,
+                    1357.0,
+                    1891.0,
+                    1357.0,
+                    1926.0,
+                    321.0,
+                    1926.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1747.0,
+                    1404.0,
+                    1747.0,
+                    1404.0,
+                    1785.0,
+                    294.0,
+                    1785.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    1779.0,
+                    1406.0,
+                    1779.0,
+                    1406.0,
+                    1817.0,
+                    323.0,
+                    1817.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    324.0,
+                    1811.0,
+                    1161.0,
+                    1811.0,
+                    1161.0,
+                    1845.0,
+                    324.0,
+                    1845.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 12,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    229,
+                    1403,
+                    229,
+                    1403,
+                    293,
+                    297,
+                    293
+                ],
+                "score": 0.846
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    299,
+                    74,
+                    816,
+                    74,
+                    816,
+                    105,
+                    299,
+                    105
+                ],
+                "score": 0.84
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    836,
+                    2088,
+                    865,
+                    2088,
+                    865,
+                    2112,
+                    836,
+                    2112
+                ],
+                "score": 0.829
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    302,
+                    312,
+                    1359,
+                    312,
+                    1359,
+                    346,
+                    302,
+                    346
+                ],
+                "score": 0.677
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    72.0,
+                    817.0,
+                    72.0,
+                    817.0,
+                    108.0,
+                    295.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    832.0,
+                    2084.0,
+                    871.0,
+                    2084.0,
+                    871.0,
+                    2122.0,
+                    832.0,
+                    2122.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    223.0,
+                    1408.0,
+                    223.0,
+                    1408.0,
+                    271.0,
+                    291.0,
+                    271.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    261.0,
+                    1319.0,
+                    261.0,
+                    1319.0,
+                    296.0,
+                    322.0,
+                    296.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    311.0,
+                    1358.0,
+                    311.0,
+                    1358.0,
+                    350.0,
+                    297.0,
+                    350.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 13,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    1022,
+                    1403,
+                    1022,
+                    1403,
+                    1236,
+                    299,
+                    1236
+                ],
+                "score": 0.981
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1728,
+                    1402,
+                    1728,
+                    1402,
+                    2033,
+                    298,
+                    2033
+                ],
+                "score": 0.98
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1359,
+                    1403,
+                    1359,
+                    1403,
+                    1575,
+                    298,
+                    1575
+                ],
+                "score": 0.978
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    823,
+                    1403,
+                    823,
+                    1403,
+                    1007,
+                    298,
+                    1007
+                ],
+                "score": 0.977
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    347,
+                    1403,
+                    347,
+                    1403,
+                    561,
+                    298,
+                    561
+                ],
+                "score": 0.976
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    715,
+                    1402,
+                    715,
+                    1402,
+                    807,
+                    299,
+                    807
+                ],
+                "score": 0.972
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    300,
+                    576,
+                    1402,
+                    576,
+                    1402,
+                    699,
+                    300,
+                    699
+                ],
+                "score": 0.971
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    1590,
+                    1402,
+                    1590,
+                    1402,
+                    1712,
+                    299,
+                    1712
+                ],
+                "score": 0.971
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    300,
+                    225,
+                    997,
+                    225,
+                    997,
+                    262,
+                    300,
+                    262
+                ],
+                "score": 0.91
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    299,
+                    1287,
+                    857,
+                    1287,
+                    857,
+                    1323,
+                    299,
+                    1323
+                ],
+                "score": 0.903
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    299,
+                    1364,
+                    299,
+                    1364,
+                    332,
+                    297,
+                    332
+                ],
+                "score": 0.895
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    300,
+                    76,
+                    816,
+                    76,
+                    816,
+                    104,
+                    300,
+                    104
+                ],
+                "score": 0.884
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    837,
+                    2088,
+                    864,
+                    2088,
+                    864,
+                    2112,
+                    837,
+                    2112
+                ],
+                "score": 0.846
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    810,
+                    1652,
+                    897,
+                    1652,
+                    897,
+                    1684,
+                    810,
+                    1684
+                ],
+                "score": 0.92,
+                "latex": "K + Q"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    362,
+                    376,
+                    397,
+                    376,
+                    397,
+                    405,
+                    362,
+                    405
+                ],
+                "score": 0.86,
+                "latex": "L ^ { 2 }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1169,
+                    1621,
+                    1196,
+                    1621,
+                    1196,
+                    1652,
+                    1169,
+                    1652
+                ],
+                "score": 0.85,
+                "latex": "Q"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    875,
+                    1622,
+                    903,
+                    1622,
+                    903,
+                    1649,
+                    875,
+                    1649
+                ],
+                "score": 0.83,
+                "latex": "N"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    965,
+                    1622,
+                    994,
+                    1622,
+                    994,
+                    1649,
+                    965,
+                    1649
+                ],
+                "score": 0.82,
+                "latex": "K"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    635,
+                    1683,
+                    663,
+                    1683,
+                    663,
+                    1709,
+                    635,
+                    1709
+                ],
+                "score": 0.81,
+                "latex": "N"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    425,
+                    1652,
+                    455,
+                    1652,
+                    455,
+                    1679,
+                    425,
+                    1679
+                ],
+                "score": 0.77,
+                "latex": "N"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    297,
+                    976,
+                    322,
+                    976,
+                    322,
+                    1009,
+                    297,
+                    1009
+                ],
+                "score": 0.51,
+                "latex": "Q"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    223.0,
+                    1001.0,
+                    223.0,
+                    1001.0,
+                    267.0,
+                    297.0,
+                    267.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    1285.0,
+                    862.0,
+                    1285.0,
+                    862.0,
+                    1328.0,
+                    297.0,
+                    1328.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    73.0,
+                    817.0,
+                    73.0,
+                    817.0,
+                    108.0,
+                    296.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    833.0,
+                    2085.0,
+                    869.0,
+                    2085.0,
+                    869.0,
+                    2124.0,
+                    833.0,
+                    2124.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1021.0,
+                    1404.0,
+                    1021.0,
+                    1404.0,
+                    1058.0,
+                    295.0,
+                    1058.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1052.0,
+                    1404.0,
+                    1052.0,
+                    1404.0,
+                    1089.0,
+                    294.0,
+                    1089.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1080.0,
+                    1403.0,
+                    1080.0,
+                    1403.0,
+                    1120.0,
+                    294.0,
+                    1120.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1113.0,
+                    1403.0,
+                    1113.0,
+                    1403.0,
+                    1147.0,
+                    293.0,
+                    1147.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1145.0,
+                    1403.0,
+                    1145.0,
+                    1403.0,
+                    1179.0,
+                    294.0,
+                    1179.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1174.0,
+                    1404.0,
+                    1174.0,
+                    1404.0,
+                    1212.0,
+                    293.0,
+                    1212.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1202.0,
+                    741.0,
+                    1202.0,
+                    741.0,
+                    1243.0,
+                    293.0,
+                    1243.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1727.0,
+                    1406.0,
+                    1727.0,
+                    1406.0,
+                    1763.0,
+                    293.0,
+                    1763.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1758.0,
+                    1404.0,
+                    1758.0,
+                    1404.0,
+                    1793.0,
+                    293.0,
+                    1793.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1788.0,
+                    1406.0,
+                    1788.0,
+                    1406.0,
+                    1824.0,
+                    294.0,
+                    1824.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1820.0,
+                    1406.0,
+                    1820.0,
+                    1406.0,
+                    1855.0,
+                    293.0,
+                    1855.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1850.0,
+                    1404.0,
+                    1850.0,
+                    1404.0,
+                    1886.0,
+                    294.0,
+                    1886.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1883.0,
+                    1406.0,
+                    1883.0,
+                    1406.0,
+                    1915.0,
+                    296.0,
+                    1915.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1911.0,
+                    1406.0,
+                    1911.0,
+                    1406.0,
+                    1946.0,
+                    294.0,
+                    1946.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1943.0,
+                    1406.0,
+                    1943.0,
+                    1406.0,
+                    1977.0,
+                    293.0,
+                    1977.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1972.0,
+                    1404.0,
+                    1972.0,
+                    1404.0,
+                    2007.0,
+                    294.0,
+                    2007.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    2002.0,
+                    1407.0,
+                    2002.0,
+                    1407.0,
+                    2037.0,
+                    294.0,
+                    2037.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1356.0,
+                    1407.0,
+                    1356.0,
+                    1407.0,
+                    1399.0,
+                    293.0,
+                    1399.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1392.0,
+                    1405.0,
+                    1392.0,
+                    1405.0,
+                    1426.0,
+                    294.0,
+                    1426.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1421.0,
+                    1405.0,
+                    1421.0,
+                    1405.0,
+                    1455.0,
+                    292.0,
+                    1455.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1451.0,
+                    1407.0,
+                    1451.0,
+                    1407.0,
+                    1488.0,
+                    292.0,
+                    1488.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1483.0,
+                    1405.0,
+                    1483.0,
+                    1405.0,
+                    1518.0,
+                    293.0,
+                    1518.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1510.0,
+                    1404.0,
+                    1510.0,
+                    1404.0,
+                    1551.0,
+                    292.0,
+                    1551.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1544.0,
+                    1369.0,
+                    1544.0,
+                    1369.0,
+                    1578.0,
+                    294.0,
+                    1578.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    822.0,
+                    1405.0,
+                    822.0,
+                    1405.0,
+                    858.0,
+                    294.0,
+                    858.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    854.0,
+                    1404.0,
+                    854.0,
+                    1404.0,
+                    886.0,
+                    296.0,
+                    886.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    885.0,
+                    1404.0,
+                    885.0,
+                    1404.0,
+                    916.0,
+                    296.0,
+                    916.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    909.0,
+                    1407.0,
+                    909.0,
+                    1407.0,
+                    953.0,
+                    292.0,
+                    953.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    944.0,
+                    1405.0,
+                    944.0,
+                    1405.0,
+                    980.0,
+                    294.0,
+                    980.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    976.0,
+                    930.0,
+                    976.0,
+                    930.0,
+                    1011.0,
+                    323.0,
+                    1011.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    347.0,
+                    1403.0,
+                    347.0,
+                    1403.0,
+                    382.0,
+                    294.0,
+                    382.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    374.0,
+                    361.0,
+                    374.0,
+                    361.0,
+                    415.0,
+                    292.0,
+                    415.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    398.0,
+                    374.0,
+                    1407.0,
+                    374.0,
+                    1407.0,
+                    415.0,
+                    398.0,
+                    415.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    405.0,
+                    1407.0,
+                    405.0,
+                    1407.0,
+                    447.0,
+                    292.0,
+                    447.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    437.0,
+                    1404.0,
+                    437.0,
+                    1404.0,
+                    475.0,
+                    293.0,
+                    475.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    463.0,
+                    1403.0,
+                    463.0,
+                    1403.0,
+                    507.0,
+                    292.0,
+                    507.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    499.0,
+                    1403.0,
+                    499.0,
+                    1403.0,
+                    534.0,
+                    294.0,
+                    534.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    530.0,
+                    557.0,
+                    530.0,
+                    557.0,
+                    566.0,
+                    294.0,
+                    566.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    715.0,
+                    1402.0,
+                    715.0,
+                    1402.0,
+                    749.0,
+                    295.0,
+                    749.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    748.0,
+                    1403.0,
+                    748.0,
+                    1403.0,
+                    778.0,
+                    297.0,
+                    778.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    777.0,
+                    1173.0,
+                    777.0,
+                    1173.0,
+                    810.0,
+                    295.0,
+                    810.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    575.0,
+                    1406.0,
+                    575.0,
+                    1406.0,
+                    611.0,
+                    295.0,
+                    611.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    606.0,
+                    1406.0,
+                    606.0,
+                    1406.0,
+                    643.0,
+                    293.0,
+                    643.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    636.0,
+                    1406.0,
+                    636.0,
+                    1406.0,
+                    676.0,
+                    293.0,
+                    676.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    666.0,
+                    503.0,
+                    666.0,
+                    503.0,
+                    707.0,
+                    296.0,
+                    707.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1587.0,
+                    1402.0,
+                    1587.0,
+                    1402.0,
+                    1626.0,
+                    294.0,
+                    1626.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1619.0,
+                    874.0,
+                    1619.0,
+                    874.0,
+                    1656.0,
+                    293.0,
+                    1656.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    904.0,
+                    1619.0,
+                    964.0,
+                    1619.0,
+                    964.0,
+                    1656.0,
+                    904.0,
+                    1656.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    995.0,
+                    1619.0,
+                    1168.0,
+                    1619.0,
+                    1168.0,
+                    1656.0,
+                    995.0,
+                    1656.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1197.0,
+                    1619.0,
+                    1406.0,
+                    1619.0,
+                    1406.0,
+                    1656.0,
+                    1197.0,
+                    1656.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1653.0,
+                    424.0,
+                    1653.0,
+                    424.0,
+                    1686.0,
+                    295.0,
+                    1686.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    456.0,
+                    1653.0,
+                    809.0,
+                    1653.0,
+                    809.0,
+                    1686.0,
+                    456.0,
+                    1686.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    898.0,
+                    1653.0,
+                    1404.0,
+                    1653.0,
+                    1404.0,
+                    1686.0,
+                    898.0,
+                    1686.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1682.0,
+                    634.0,
+                    1682.0,
+                    634.0,
+                    1715.0,
+                    295.0,
+                    1715.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    664.0,
+                    1682.0,
+                    853.0,
+                    1682.0,
+                    853.0,
+                    1715.0,
+                    664.0,
+                    1715.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    293.0,
+                    1367.0,
+                    293.0,
+                    1367.0,
+                    340.0,
+                    293.0,
+                    340.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 14,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1589,
+                    1403,
+                    1589,
+                    1403,
+                    1896,
+                    298,
+                    1896
+                ],
+                "score": 0.982
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    670,
+                    1404,
+                    670,
+                    1404,
+                    1075,
+                    298,
+                    1075
+                ],
+                "score": 0.982
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    437,
+                    1403,
+                    437,
+                    1403,
+                    653,
+                    298,
+                    653
+                ],
+                "score": 0.979
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1359,
+                    1404,
+                    1359,
+                    1404,
+                    1575,
+                    298,
+                    1575
+                ],
+                "score": 0.979
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1091,
+                    1403,
+                    1091,
+                    1403,
+                    1244,
+                    298,
+                    1244
+                ],
+                "score": 0.977
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    300,
+                    1911,
+                    1402,
+                    1911,
+                    1402,
+                    2034,
+                    300,
+                    2034
+                ],
+                "score": 0.973
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    230,
+                    1403,
+                    230,
+                    1403,
+                    323,
+                    299,
+                    323
+                ],
+                "score": 0.957
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    300,
+                    76,
+                    815,
+                    76,
+                    815,
+                    104,
+                    300,
+                    104
+                ],
+                "score": 0.891
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    305,
+                    369,
+                    1247,
+                    369,
+                    1247,
+                    405,
+                    305,
+                    405
+                ],
+                "score": 0.876
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    837,
+                    2089,
+                    865,
+                    2089,
+                    865,
+                    2112,
+                    837,
+                    2112
+                ],
+                "score": 0.844
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    300,
+                    1290,
+                    842,
+                    1290,
+                    842,
+                    1326,
+                    300,
+                    1326
+                ],
+                "score": 0.563
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    301,
+                    1290,
+                    842,
+                    1290,
+                    842,
+                    1325,
+                    301,
+                    1325
+                ],
+                "score": 0.29
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    356,
+                    760,
+                    523,
+                    760,
+                    523,
+                    794,
+                    356,
+                    794
+                ],
+                "score": 0.93,
+                "latex": "\\bar { H } = \\lceil \\log _ { 2 } \\bar { N } \\rceil"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1029,
+                    952,
+                    1144,
+                    952,
+                    1144,
+                    986,
+                    1029,
+                    986
+                ],
+                "score": 0.93,
+                "latex": "\\mathcal { P } = \\{ \\mathcal { C } _ { c } \\}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    519,
+                    983,
+                    695,
+                    983,
+                    695,
+                    1014,
+                    519,
+                    1014
+                ],
+                "score": 0.93,
+                "latex": "R = K _ { \\mathrm { m - t r } } + Q"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    334,
+                    470,
+                    387,
+                    470,
+                    387,
+                    504,
+                    334,
+                    504
+                ],
+                "score": 0.92,
+                "latex": "\\left\\{ \\mathbf { z } _ { i } \\right\\}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    666,
+                    439,
+                    720,
+                    439,
+                    720,
+                    473,
+                    666,
+                    473
+                ],
+                "score": 0.92,
+                "latex": "\\left\\{ \\mathbf { z } _ { i } \\right\\}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1093,
+                    699,
+                    1147,
+                    699,
+                    1147,
+                    733,
+                    1093,
+                    733
+                ],
+                "score": 0.92,
+                "latex": "\\left\\{ \\mathbf { z } _ { i } \\right\\}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1284,
+                    438,
+                    1401,
+                    438,
+                    1401,
+                    473,
+                    1284,
+                    473
+                ],
+                "score": 0.92,
+                "latex": "\\mathcal { P } = \\{ \\mathcal { C } _ { c } \\}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    435,
+                    853,
+                    604,
+                    853,
+                    604,
+                    894,
+                    435,
+                    894
+                ],
+                "score": 0.92,
+                "latex": "\\frac { \\mathbf { n } } { \\left| \\mathbf { n } \\right| _ { 2 } } \\cdot \\left( \\mathbf { z } - \\mathbf { z } _ { 0 } \\right)"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    329,
+                    1121,
+                    382,
+                    1121,
+                    382,
+                    1155,
+                    329,
+                    1155
+                ],
+                "score": 0.92,
+                "latex": "\\left\\{ \\mathbf { z } _ { i } \\right\\}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1180,
+                    891,
+                    1285,
+                    891,
+                    1285,
+                    925,
+                    1180,
+                    925
+                ],
+                "score": 0.92,
+                "latex": "( - m , m )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1332,
+                    1013,
+                    1393,
+                    1013,
+                    1393,
+                    1048,
+                    1332,
+                    1048
+                ],
+                "score": 0.92,
+                "latex": "\\{ \\mathcal P _ { p } \\}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    781,
+                    1913,
+                    819,
+                    1913,
+                    819,
+                    1942,
+                    781,
+                    1942
+                ],
+                "score": 0.88,
+                "latex": "k _ { \\mathrm { n n } }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    382,
+                    1682,
+                    472,
+                    1682,
+                    472,
+                    1711,
+                    382,
+                    1711
+                ],
+                "score": 0.88,
+                "latex": "9 6 . 0 0 \\%"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1194,
+                    1743,
+                    1282,
+                    1743,
+                    1282,
+                    1772,
+                    1194,
+                    1772
+                ],
+                "score": 0.88,
+                "latex": "9 6 . 8 3 \\%"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    700,
+                    953,
+                    729,
+                    953,
+                    729,
+                    983,
+                    700,
+                    983
+                ],
+                "score": 0.88,
+                "latex": "\\mathcal { C } _ { c }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    975,
+                    920,
+                    1013,
+                    920,
+                    1013,
+                    950,
+                    975,
+                    950
+                ],
+                "score": 0.87,
+                "latex": "\\mathbf { \\bar { 2 } } ^ { H }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    632,
+                    502,
+                    661,
+                    502,
+                    661,
+                    531,
+                    632,
+                    531
+                ],
+                "score": 0.87,
+                "latex": "\\mathcal { C } _ { c }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    442,
+                    826,
+                    473,
+                    826,
+                    473,
+                    852,
+                    442,
+                    852
+                ],
+                "score": 0.85,
+                "latex": "\\mathbf { z } _ { 0 }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1253,
+                    984,
+                    1281,
+                    984,
+                    1281,
+                    1010,
+                    1253,
+                    1010
+                ],
+                "score": 0.82,
+                "latex": "N"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1217,
+                    732,
+                    1246,
+                    732,
+                    1246,
+                    758,
+                    1217,
+                    758
+                ],
+                "score": 0.82,
+                "latex": "N"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    784,
+                    701,
+                    809,
+                    701,
+                    809,
+                    728,
+                    784,
+                    728
+                ],
+                "score": 0.81,
+                "latex": "\\mathcal { P }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    556,
+                    793,
+                    574,
+                    793,
+                    574,
+                    819,
+                    556,
+                    819
+                ],
+                "score": 0.81,
+                "latex": "d"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    813,
+                    472,
+                    839,
+                    472,
+                    839,
+                    498,
+                    813,
+                    498
+                ],
+                "score": 0.8,
+                "latex": "\\mathcal { Z }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    586,
+                    922,
+                    615,
+                    922,
+                    615,
+                    950,
+                    586,
+                    950
+                ],
+                "score": 0.8,
+                "latex": "H"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    732,
+                    853,
+                    761,
+                    853,
+                    761,
+                    881,
+                    732,
+                    881
+                ],
+                "score": 0.79,
+                "latex": "H"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    602,
+                    823,
+                    621,
+                    823,
+                    621,
+                    849,
+                    602,
+                    849
+                ],
+                "score": 0.78,
+                "latex": "d"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    842,
+                    441,
+                    867,
+                    441,
+                    867,
+                    468,
+                    842,
+                    468
+                ],
+                "score": 0.77,
+                "latex": "\\mathcal { Z }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    878,
+                    892,
+                    907,
+                    892,
+                    907,
+                    919,
+                    878,
+                    919
+                ],
+                "score": 0.76,
+                "latex": "H"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    440,
+                    957,
+                    456,
+                    957,
+                    456,
+                    980,
+                    440,
+                    980
+                ],
+                "score": 0.75,
+                "latex": "c"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    368,
+                    507,
+                    384,
+                    507,
+                    384,
+                    528,
+                    368,
+                    528
+                ],
+                "score": 0.71,
+                "latex": "c"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1373,
+                    897,
+                    1401,
+                    897,
+                    1401,
+                    919,
+                    1373,
+                    919
+                ],
+                "score": 0.66,
+                "latex": "m"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1020,
+                    827,
+                    1037,
+                    827,
+                    1037,
+                    849,
+                    1020,
+                    849
+                ],
+                "score": 0.57,
+                "latex": "\\mathbf { z }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1244,
+                    797,
+                    1267,
+                    797,
+                    1267,
+                    820,
+                    1244,
+                    820
+                ],
+                "score": 0.37,
+                "latex": "\\mathbf { n }"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    73.0,
+                    817.0,
+                    73.0,
+                    817.0,
+                    108.0,
+                    296.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    300.0,
+                    367.0,
+                    1254.0,
+                    367.0,
+                    1254.0,
+                    410.0,
+                    300.0,
+                    410.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    831.0,
+                    2084.0,
+                    871.0,
+                    2084.0,
+                    871.0,
+                    2125.0,
+                    831.0,
+                    2125.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1288.0,
+                    846.0,
+                    1288.0,
+                    846.0,
+                    1331.0,
+                    296.0,
+                    1331.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1289.0,
+                    846.0,
+                    1289.0,
+                    846.0,
+                    1332.0,
+                    296.0,
+                    1332.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1591.0,
+                    1404.0,
+                    1591.0,
+                    1404.0,
+                    1623.0,
+                    296.0,
+                    1623.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1618.0,
+                    1407.0,
+                    1618.0,
+                    1407.0,
+                    1659.0,
+                    291.0,
+                    1659.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1653.0,
+                    1404.0,
+                    1653.0,
+                    1404.0,
+                    1685.0,
+                    296.0,
+                    1685.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1683.0,
+                    381.0,
+                    1683.0,
+                    381.0,
+                    1715.0,
+                    294.0,
+                    1715.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    473.0,
+                    1683.0,
+                    1404.0,
+                    1683.0,
+                    1404.0,
+                    1715.0,
+                    473.0,
+                    1715.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1708.0,
+                    1407.0,
+                    1708.0,
+                    1407.0,
+                    1750.0,
+                    292.0,
+                    1750.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1743.0,
+                    1193.0,
+                    1743.0,
+                    1193.0,
+                    1778.0,
+                    294.0,
+                    1778.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1283.0,
+                    1743.0,
+                    1407.0,
+                    1743.0,
+                    1407.0,
+                    1778.0,
+                    1283.0,
+                    1778.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1769.0,
+                    1404.0,
+                    1769.0,
+                    1404.0,
+                    1811.0,
+                    292.0,
+                    1811.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1801.0,
+                    1405.0,
+                    1801.0,
+                    1405.0,
+                    1838.0,
+                    294.0,
+                    1838.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1830.0,
+                    1407.0,
+                    1830.0,
+                    1407.0,
+                    1871.0,
+                    291.0,
+                    1871.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1863.0,
+                    406.0,
+                    1863.0,
+                    406.0,
+                    1904.0,
+                    293.0,
+                    1904.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    670.0,
+                    1405.0,
+                    670.0,
+                    1405.0,
+                    704.0,
+                    296.0,
+                    704.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    699.0,
+                    783.0,
+                    699.0,
+                    783.0,
+                    737.0,
+                    293.0,
+                    737.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    810.0,
+                    699.0,
+                    1092.0,
+                    699.0,
+                    1092.0,
+                    737.0,
+                    810.0,
+                    737.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1148.0,
+                    699.0,
+                    1405.0,
+                    699.0,
+                    1405.0,
+                    737.0,
+                    1148.0,
+                    737.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    729.0,
+                    1216.0,
+                    729.0,
+                    1216.0,
+                    767.0,
+                    293.0,
+                    767.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1247.0,
+                    729.0,
+                    1405.0,
+                    729.0,
+                    1405.0,
+                    767.0,
+                    1247.0,
+                    767.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    760.0,
+                    355.0,
+                    760.0,
+                    355.0,
+                    797.0,
+                    292.0,
+                    797.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    524.0,
+                    760.0,
+                    1404.0,
+                    760.0,
+                    1404.0,
+                    797.0,
+                    524.0,
+                    797.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    789.0,
+                    555.0,
+                    789.0,
+                    555.0,
+                    828.0,
+                    292.0,
+                    828.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    575.0,
+                    789.0,
+                    1243.0,
+                    789.0,
+                    1243.0,
+                    828.0,
+                    575.0,
+                    828.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1268.0,
+                    789.0,
+                    1406.0,
+                    789.0,
+                    1406.0,
+                    828.0,
+                    1268.0,
+                    828.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    822.0,
+                    441.0,
+                    822.0,
+                    441.0,
+                    856.0,
+                    293.0,
+                    856.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    474.0,
+                    822.0,
+                    601.0,
+                    822.0,
+                    601.0,
+                    856.0,
+                    474.0,
+                    856.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    622.0,
+                    822.0,
+                    1019.0,
+                    822.0,
+                    1019.0,
+                    856.0,
+                    622.0,
+                    856.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1038.0,
+                    822.0,
+                    1404.0,
+                    822.0,
+                    1404.0,
+                    856.0,
+                    1038.0,
+                    856.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    287.0,
+                    848.0,
+                    434.0,
+                    848.0,
+                    434.0,
+                    898.0,
+                    287.0,
+                    898.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    605.0,
+                    848.0,
+                    731.0,
+                    848.0,
+                    731.0,
+                    898.0,
+                    605.0,
+                    898.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    762.0,
+                    848.0,
+                    1410.0,
+                    848.0,
+                    1410.0,
+                    898.0,
+                    762.0,
+                    898.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    888.0,
+                    877.0,
+                    888.0,
+                    877.0,
+                    928.0,
+                    292.0,
+                    928.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    908.0,
+                    888.0,
+                    1179.0,
+                    888.0,
+                    1179.0,
+                    928.0,
+                    908.0,
+                    928.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1286.0,
+                    888.0,
+                    1372.0,
+                    888.0,
+                    1372.0,
+                    928.0,
+                    1286.0,
+                    928.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1402.0,
+                    888.0,
+                    1406.0,
+                    888.0,
+                    1406.0,
+                    928.0,
+                    1402.0,
+                    928.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    288.0,
+                    914.0,
+                    585.0,
+                    914.0,
+                    585.0,
+                    962.0,
+                    288.0,
+                    962.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    616.0,
+                    914.0,
+                    974.0,
+                    914.0,
+                    974.0,
+                    962.0,
+                    616.0,
+                    962.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1014.0,
+                    914.0,
+                    1408.0,
+                    914.0,
+                    1408.0,
+                    962.0,
+                    1014.0,
+                    962.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    951.0,
+                    439.0,
+                    951.0,
+                    439.0,
+                    988.0,
+                    293.0,
+                    988.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    457.0,
+                    951.0,
+                    699.0,
+                    951.0,
+                    699.0,
+                    988.0,
+                    457.0,
+                    988.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    730.0,
+                    951.0,
+                    1028.0,
+                    951.0,
+                    1028.0,
+                    988.0,
+                    730.0,
+                    988.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1145.0,
+                    951.0,
+                    1406.0,
+                    951.0,
+                    1406.0,
+                    988.0,
+                    1145.0,
+                    988.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    979.0,
+                    518.0,
+                    979.0,
+                    518.0,
+                    1019.0,
+                    292.0,
+                    1019.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    696.0,
+                    979.0,
+                    1252.0,
+                    979.0,
+                    1252.0,
+                    1019.0,
+                    696.0,
+                    1019.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1282.0,
+                    979.0,
+                    1405.0,
+                    979.0,
+                    1405.0,
+                    1019.0,
+                    1282.0,
+                    1019.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1013.0,
+                    1331.0,
+                    1013.0,
+                    1331.0,
+                    1051.0,
+                    293.0,
+                    1051.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1394.0,
+                    1013.0,
+                    1405.0,
+                    1013.0,
+                    1405.0,
+                    1051.0,
+                    1394.0,
+                    1051.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1044.0,
+                    1146.0,
+                    1044.0,
+                    1146.0,
+                    1078.0,
+                    294.0,
+                    1078.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    436.0,
+                    665.0,
+                    436.0,
+                    665.0,
+                    478.0,
+                    293.0,
+                    478.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    721.0,
+                    436.0,
+                    841.0,
+                    436.0,
+                    841.0,
+                    478.0,
+                    721.0,
+                    478.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    868.0,
+                    436.0,
+                    1283.0,
+                    436.0,
+                    1283.0,
+                    478.0,
+                    868.0,
+                    478.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1402.0,
+                    436.0,
+                    1405.0,
+                    436.0,
+                    1405.0,
+                    478.0,
+                    1402.0,
+                    478.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    467.0,
+                    333.0,
+                    467.0,
+                    333.0,
+                    507.0,
+                    292.0,
+                    507.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    388.0,
+                    467.0,
+                    812.0,
+                    467.0,
+                    812.0,
+                    507.0,
+                    388.0,
+                    507.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    840.0,
+                    467.0,
+                    1407.0,
+                    467.0,
+                    1407.0,
+                    507.0,
+                    840.0,
+                    507.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    501.0,
+                    367.0,
+                    501.0,
+                    367.0,
+                    535.0,
+                    294.0,
+                    535.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    385.0,
+                    501.0,
+                    631.0,
+                    501.0,
+                    631.0,
+                    535.0,
+                    385.0,
+                    535.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    662.0,
+                    501.0,
+                    1403.0,
+                    501.0,
+                    1403.0,
+                    535.0,
+                    662.0,
+                    535.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    530.0,
+                    1407.0,
+                    530.0,
+                    1407.0,
+                    568.0,
+                    293.0,
+                    568.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    560.0,
+                    1406.0,
+                    560.0,
+                    1406.0,
+                    597.0,
+                    292.0,
+                    597.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    593.0,
+                    1403.0,
+                    593.0,
+                    1403.0,
+                    628.0,
+                    294.0,
+                    628.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    623.0,
+                    541.0,
+                    623.0,
+                    541.0,
+                    652.0,
+                    294.0,
+                    652.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1357.0,
+                    1404.0,
+                    1357.0,
+                    1404.0,
+                    1397.0,
+                    293.0,
+                    1397.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1392.0,
+                    1405.0,
+                    1392.0,
+                    1405.0,
+                    1426.0,
+                    293.0,
+                    1426.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1420.0,
+                    1406.0,
+                    1420.0,
+                    1406.0,
+                    1458.0,
+                    293.0,
+                    1458.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1448.0,
+                    1405.0,
+                    1448.0,
+                    1405.0,
+                    1492.0,
+                    292.0,
+                    1492.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1483.0,
+                    1404.0,
+                    1483.0,
+                    1404.0,
+                    1516.0,
+                    293.0,
+                    1516.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1513.0,
+                    1406.0,
+                    1513.0,
+                    1406.0,
+                    1551.0,
+                    293.0,
+                    1551.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1545.0,
+                    577.0,
+                    1545.0,
+                    577.0,
+                    1579.0,
+                    294.0,
+                    1579.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1090.0,
+                    1405.0,
+                    1090.0,
+                    1405.0,
+                    1127.0,
+                    294.0,
+                    1127.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1120.0,
+                    328.0,
+                    1120.0,
+                    328.0,
+                    1157.0,
+                    294.0,
+                    1157.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    383.0,
+                    1120.0,
+                    1405.0,
+                    1120.0,
+                    1405.0,
+                    1157.0,
+                    383.0,
+                    1157.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1152.0,
+                    1405.0,
+                    1152.0,
+                    1405.0,
+                    1185.0,
+                    296.0,
+                    1185.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1183.0,
+                    1404.0,
+                    1183.0,
+                    1404.0,
+                    1216.0,
+                    296.0,
+                    1216.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1211.0,
+                    868.0,
+                    1211.0,
+                    868.0,
+                    1247.0,
+                    293.0,
+                    1247.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1909.0,
+                    780.0,
+                    1909.0,
+                    780.0,
+                    1947.0,
+                    294.0,
+                    1947.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    820.0,
+                    1909.0,
+                    1407.0,
+                    1909.0,
+                    1407.0,
+                    1947.0,
+                    820.0,
+                    1947.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1938.0,
+                    1407.0,
+                    1938.0,
+                    1407.0,
+                    1981.0,
+                    294.0,
+                    1981.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1974.0,
+                    1406.0,
+                    1974.0,
+                    1406.0,
+                    2005.0,
+                    295.0,
+                    2005.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    2002.0,
+                    910.0,
+                    2002.0,
+                    910.0,
+                    2038.0,
+                    294.0,
+                    2038.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    229.0,
+                    1407.0,
+                    229.0,
+                    1407.0,
+                    264.0,
+                    293.0,
+                    264.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    262.0,
+                    1405.0,
+                    262.0,
+                    1405.0,
+                    296.0,
+                    295.0,
+                    296.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    291.0,
+                    1339.0,
+                    291.0,
+                    1339.0,
+                    328.0,
+                    295.0,
+                    328.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 15,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 5,
+                "poly": [
+                    323,
+                    631,
+                    1378,
+                    631,
+                    1378,
+                    1003,
+                    323,
+                    1003
+                ],
+                "score": 0.985,
+                "html": "<table><tr><td>Hyperparameter</td><td>MNIST</td><td>Omniglot</td><td>miniImageNet</td><td>CelebA</td></tr><tr><td>Input size</td><td>28×28</td><td>28×28</td><td>84 ×84×3</td><td>84 ×84×3</td></tr><tr><td>Outer (meta) learning rate</td><td>0.001</td><td>0.001</td><td>0.001</td><td>0.001</td></tr><tr><td>Inner learning rate</td><td>0.05</td><td>0.05</td><td>0.05</td><td>0.05</td></tr><tr><td>Task batch size</td><td>8</td><td>8</td><td>8</td><td>8</td></tr><tr><td>Inner adaptation steps (meta-training)</td><td>5</td><td></td><td>5</td><td>5</td></tr><tr><td>Meta-training iterations</td><td>30,000</td><td>5 30,000</td><td>60,000</td><td>60,000</td></tr><tr><td>Adaptation steps (evaluation)</td><td>50</td><td>50</td><td>50</td><td>50</td></tr><tr><td>Classes per task (meta-training)</td><td>10</td><td>20</td><td>5</td><td>2</td></tr><tr><td>Shots per class (meta-training)</td><td>1</td><td>1</td><td>1</td><td>1</td></tr><tr><td>Queries per class</td><td>5</td><td>5</td><td>5</td><td>5</td></tr></table>"
+            },
+            {
+                "category_id": 5,
+                "poly": [
+                    339,
+                    1625,
+                    1361,
+                    1625,
+                    1361,
+                    1907,
+                    339,
+                    1907
+                ],
+                "score": 0.984,
+                "html": "<table><tr><td>Hyperparameter</td><td>Omniglot</td><td>miniImageNet</td><td>CelebA</td></tr><tr><td>Input size</td><td>28×28</td><td>84×84×3</td><td>84×84×3</td></tr><tr><td>Learning rate</td><td>0.001</td><td>0.001</td><td>0.001</td></tr><tr><td>Task batch size</td><td>1</td><td>1</td><td>1</td></tr><tr><td>Training iterations</td><td>30,000</td><td>60,000</td><td>60,000</td></tr><tr><td>Classes per task (meta-training)</td><td>20</td><td>5</td><td>2</td></tr><tr><td>Shots per class (meta-training)</td><td>1</td><td>1</td><td>1</td></tr><tr><td>Queries per class (meta-training/meta-testing)</td><td>15/5</td><td>15/5</td><td>15/5</td></tr></table>"
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1069,
+                    1404,
+                    1069,
+                    1404,
+                    1285,
+                    297,
+                    1285
+                ],
+                "score": 0.98
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1299,
+                    1403,
+                    1299,
+                    1403,
+                    1454,
+                    298,
+                    1454
+                ],
+                "score": 0.978
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    229,
+                    1403,
+                    229,
+                    1403,
+                    384,
+                    298,
+                    384
+                ],
+                "score": 0.976
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    300,
+                    1972,
+                    1401,
+                    1972,
+                    1401,
+                    2034,
+                    300,
+                    2034
+                ],
+                "score": 0.949
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    303,
+                    428,
+                    1126,
+                    428,
+                    1126,
+                    464,
+                    303,
+                    464
+                ],
+                "score": 0.906
+            },
+            {
+                "category_id": 6,
+                "poly": [
+                    632,
+                    563,
+                    1066,
+                    563,
+                    1066,
+                    595,
+                    632,
+                    595
+                ],
+                "score": 0.9
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    299,
+                    75,
+                    815,
+                    75,
+                    815,
+                    105,
+                    299,
+                    105
+                ],
+                "score": 0.899
+            },
+            {
+                "category_id": 6,
+                "poly": [
+                    622,
+                    1557,
+                    1077,
+                    1557,
+                    1077,
+                    1588,
+                    622,
+                    1588
+                ],
+                "score": 0.893
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    299,
+                    1489,
+                    515,
+                    1489,
+                    515,
+                    1521,
+                    299,
+                    1521
+                ],
+                "score": 0.891
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    836,
+                    2088,
+                    864,
+                    2088,
+                    864,
+                    2112,
+                    836,
+                    2112
+                ],
+                "score": 0.864
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    297,
+                    496,
+                    461,
+                    496,
+                    461,
+                    529,
+                    297,
+                    529
+                ],
+                "score": 0.811
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    582,
+                    231,
+                    684,
+                    231,
+                    684,
+                    260,
+                    582,
+                    260
+                ],
+                "score": 0.92,
+                "latex": "k = 5 0 0"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1213,
+                    231,
+                    1301,
+                    231,
+                    1301,
+                    259,
+                    1213,
+                    259
+                ],
+                "score": 0.91,
+                "latex": "k = 1 0"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    298.0,
+                    426.0,
+                    1130.0,
+                    426.0,
+                    1130.0,
+                    469.0,
+                    298.0,
+                    469.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    628.0,
+                    560.0,
+                    1069.0,
+                    560.0,
+                    1069.0,
+                    599.0,
+                    628.0,
+                    599.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    73.0,
+                    817.0,
+                    73.0,
+                    817.0,
+                    108.0,
+                    295.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    618.0,
+                    1554.0,
+                    1080.0,
+                    1554.0,
+                    1080.0,
+                    1592.0,
+                    618.0,
+                    1592.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1486.0,
+                    519.0,
+                    1486.0,
+                    519.0,
+                    1526.0,
+                    293.0,
+                    1526.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    832.0,
+                    2085.0,
+                    869.0,
+                    2085.0,
+                    869.0,
+                    2124.0,
+                    832.0,
+                    2124.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    495.0,
+                    465.0,
+                    495.0,
+                    465.0,
+                    531.0,
+                    294.0,
+                    531.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1068.0,
+                    1405.0,
+                    1068.0,
+                    1405.0,
+                    1104.0,
+                    294.0,
+                    1104.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1097.0,
+                    1406.0,
+                    1097.0,
+                    1406.0,
+                    1138.0,
+                    292.0,
+                    1138.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1131.0,
+                    1405.0,
+                    1131.0,
+                    1405.0,
+                    1166.0,
+                    295.0,
+                    1166.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1158.0,
+                    1405.0,
+                    1158.0,
+                    1405.0,
+                    1199.0,
+                    291.0,
+                    1199.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1189.0,
+                    1405.0,
+                    1189.0,
+                    1405.0,
+                    1229.0,
+                    291.0,
+                    1229.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1219.0,
+                    1405.0,
+                    1219.0,
+                    1405.0,
+                    1260.0,
+                    291.0,
+                    1260.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1251.0,
+                    1122.0,
+                    1251.0,
+                    1122.0,
+                    1287.0,
+                    292.0,
+                    1287.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    1301.0,
+                    1402.0,
+                    1301.0,
+                    1402.0,
+                    1334.0,
+                    297.0,
+                    1334.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1330.0,
+                    1405.0,
+                    1330.0,
+                    1405.0,
+                    1365.0,
+                    293.0,
+                    1365.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1361.0,
+                    1405.0,
+                    1361.0,
+                    1405.0,
+                    1398.0,
+                    294.0,
+                    1398.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1392.0,
+                    1404.0,
+                    1392.0,
+                    1404.0,
+                    1426.0,
+                    296.0,
+                    1426.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1420.0,
+                    709.0,
+                    1420.0,
+                    709.0,
+                    1460.0,
+                    293.0,
+                    1460.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    229.0,
+                    581.0,
+                    229.0,
+                    581.0,
+                    263.0,
+                    296.0,
+                    263.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    685.0,
+                    229.0,
+                    1212.0,
+                    229.0,
+                    1212.0,
+                    263.0,
+                    685.0,
+                    263.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1302.0,
+                    229.0,
+                    1404.0,
+                    229.0,
+                    1404.0,
+                    263.0,
+                    1302.0,
+                    263.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    260.0,
+                    1405.0,
+                    260.0,
+                    1405.0,
+                    297.0,
+                    294.0,
+                    297.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    292.0,
+                    1404.0,
+                    292.0,
+                    1404.0,
+                    325.0,
+                    294.0,
+                    325.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    320.0,
+                    1405.0,
+                    320.0,
+                    1405.0,
+                    357.0,
+                    293.0,
+                    357.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    349.0,
+                    1006.0,
+                    349.0,
+                    1006.0,
+                    388.0,
+                    292.0,
+                    388.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1970.0,
+                    1403.0,
+                    1970.0,
+                    1403.0,
+                    2006.0,
+                    295.0,
+                    2006.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    2002.0,
+                    1405.0,
+                    2002.0,
+                    1405.0,
+                    2036.0,
+                    295.0,
+                    2036.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 16,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    802,
+                    1403,
+                    802,
+                    1403,
+                    1079,
+                    298,
+                    1079
+                ],
+                "score": 0.98
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1354,
+                    1403,
+                    1354,
+                    1403,
+                    1567,
+                    298,
+                    1567
+                ],
+                "score": 0.979
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1849,
+                    1403,
+                    1849,
+                    1403,
+                    2034,
+                    298,
+                    2034
+                ],
+                "score": 0.978
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1093,
+                    1403,
+                    1093,
+                    1403,
+                    1339,
+                    298,
+                    1339
+                ],
+                "score": 0.978
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    1582,
+                    1403,
+                    1582,
+                    1403,
+                    1736,
+                    299,
+                    1736
+                ],
+                "score": 0.975
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    663,
+                    1402,
+                    663,
+                    1402,
+                    787,
+                    298,
+                    787
+                ],
+                "score": 0.973
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    229,
+                    1403,
+                    229,
+                    1403,
+                    384,
+                    298,
+                    384
+                ],
+                "score": 0.971
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    301,
+                    478,
+                    1403,
+                    478,
+                    1403,
+                    570,
+                    301,
+                    570
+                ],
+                "score": 0.956
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    300,
+                    1781,
+                    874,
+                    1781,
+                    874,
+                    1816,
+                    300,
+                    1816
+                ],
+                "score": 0.897
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    299,
+                    607,
+                    913,
+                    607,
+                    913,
+                    638,
+                    299,
+                    638
+                ],
+                "score": 0.896
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    300,
+                    76,
+                    815,
+                    76,
+                    815,
+                    104,
+                    300,
+                    104
+                ],
+                "score": 0.888
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    299,
+                    419,
+                    484,
+                    419,
+                    484,
+                    452,
+                    299,
+                    452
+                ],
+                "score": 0.886
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    837,
+                    2088,
+                    864,
+                    2088,
+                    864,
+                    2112,
+                    837,
+                    2112
+                ],
+                "score": 0.838
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    791,
+                    1247,
+                    880,
+                    1247,
+                    880,
+                    1276,
+                    791,
+                    1276
+                ],
+                "score": 0.91,
+                "latex": "6 4 \\times 6 4"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    759,
+                    1386,
+                    854,
+                    1386,
+                    854,
+                    1414,
+                    759,
+                    1414
+                ],
+                "score": 0.9,
+                "latex": "6 4 \\times 6 4"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    381,
+                    540,
+                    471,
+                    540,
+                    471,
+                    568,
+                    381,
+                    568
+                ],
+                "score": 0.9,
+                "latex": "P = 5 0"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    705,
+                    509,
+                    815,
+                    509,
+                    815,
+                    537,
+                    705,
+                    537
+                ],
+                "score": 0.9,
+                "latex": "P = 1 0 0"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    725,
+                    262,
+                    792,
+                    262,
+                    792,
+                    291,
+                    725,
+                    291
+                ],
+                "score": 0.89,
+                "latex": "2 \\times 2"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    827,
+                    231,
+                    934,
+                    231,
+                    934,
+                    260,
+                    827,
+                    260
+                ],
+                "score": 0.88,
+                "latex": "6 4 \\ 3 \\times 3"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    748,
+                    804,
+                    841,
+                    804,
+                    841,
+                    833,
+                    748,
+                    833
+                ],
+                "score": 0.88,
+                "latex": "6 4 \\times 6 4"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    355,
+                    1942,
+                    410,
+                    1942,
+                    410,
+                    1972,
+                    355,
+                    1972
+                ],
+                "score": 0.86,
+                "latex": "9 5 \\%"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1105,
+                    480,
+                    1124,
+                    480,
+                    1124,
+                    506,
+                    1105,
+                    506
+                ],
+                "score": 0.82,
+                "latex": "k"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    979,
+                    1975,
+                    1003,
+                    1975,
+                    1003,
+                    2000,
+                    979,
+                    2000
+                ],
+                "score": 0.82,
+                "latex": "P"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    553,
+                    1974,
+                    572,
+                    1974,
+                    572,
+                    2000,
+                    553,
+                    2000
+                ],
+                "score": 0.7,
+                "latex": "k"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    465,
+                    2008,
+                    492,
+                    2008,
+                    492,
+                    2030,
+                    465,
+                    2030
+                ],
+                "score": 0.67,
+                "latex": "m"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1043,
+                    1944,
+                    1063,
+                    1944,
+                    1063,
+                    1970,
+                    1043,
+                    1970
+                ],
+                "score": 0.56,
+                "latex": "h"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    662,
+                    1944,
+                    680,
+                    1944,
+                    680,
+                    1970,
+                    662,
+                    1970
+                ],
+                "score": 0.46,
+                "latex": "d"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    1781.0,
+                    877.0,
+                    1781.0,
+                    877.0,
+                    1821.0,
+                    297.0,
+                    1821.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    605.0,
+                    915.0,
+                    605.0,
+                    915.0,
+                    642.0,
+                    294.0,
+                    642.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    73.0,
+                    817.0,
+                    73.0,
+                    817.0,
+                    108.0,
+                    296.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    414.0,
+                    489.0,
+                    414.0,
+                    489.0,
+                    460.0,
+                    292.0,
+                    460.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    832.0,
+                    2085.0,
+                    869.0,
+                    2085.0,
+                    869.0,
+                    2122.0,
+                    832.0,
+                    2122.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    804.0,
+                    747.0,
+                    804.0,
+                    747.0,
+                    837.0,
+                    297.0,
+                    837.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    842.0,
+                    804.0,
+                    1404.0,
+                    804.0,
+                    1404.0,
+                    837.0,
+                    842.0,
+                    837.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    834.0,
+                    1404.0,
+                    834.0,
+                    1404.0,
+                    867.0,
+                    296.0,
+                    867.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    866.0,
+                    1404.0,
+                    866.0,
+                    1404.0,
+                    899.0,
+                    294.0,
+                    899.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    893.0,
+                    1405.0,
+                    893.0,
+                    1405.0,
+                    930.0,
+                    293.0,
+                    930.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    923.0,
+                    1405.0,
+                    923.0,
+                    1405.0,
+                    963.0,
+                    294.0,
+                    963.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    955.0,
+                    1405.0,
+                    955.0,
+                    1405.0,
+                    992.0,
+                    294.0,
+                    992.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    986.0,
+                    1405.0,
+                    986.0,
+                    1405.0,
+                    1023.0,
+                    294.0,
+                    1023.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1014.0,
+                    1405.0,
+                    1014.0,
+                    1405.0,
+                    1054.0,
+                    293.0,
+                    1054.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1049.0,
+                    1179.0,
+                    1049.0,
+                    1179.0,
+                    1082.0,
+                    296.0,
+                    1082.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1353.0,
+                    1405.0,
+                    1353.0,
+                    1405.0,
+                    1391.0,
+                    294.0,
+                    1391.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1385.0,
+                    758.0,
+                    1385.0,
+                    758.0,
+                    1419.0,
+                    293.0,
+                    1419.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    855.0,
+                    1385.0,
+                    1404.0,
+                    1385.0,
+                    1404.0,
+                    1419.0,
+                    855.0,
+                    1419.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1413.0,
+                    1405.0,
+                    1413.0,
+                    1405.0,
+                    1452.0,
+                    293.0,
+                    1452.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1446.0,
+                    1405.0,
+                    1446.0,
+                    1405.0,
+                    1479.0,
+                    292.0,
+                    1479.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1476.0,
+                    1404.0,
+                    1476.0,
+                    1404.0,
+                    1511.0,
+                    294.0,
+                    1511.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1506.0,
+                    1403.0,
+                    1506.0,
+                    1403.0,
+                    1544.0,
+                    293.0,
+                    1544.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1538.0,
+                    493.0,
+                    1538.0,
+                    493.0,
+                    1573.0,
+                    296.0,
+                    1573.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    1851.0,
+                    1404.0,
+                    1851.0,
+                    1404.0,
+                    1883.0,
+                    297.0,
+                    1883.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1880.0,
+                    1403.0,
+                    1880.0,
+                    1403.0,
+                    1915.0,
+                    293.0,
+                    1915.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1910.0,
+                    1404.0,
+                    1910.0,
+                    1404.0,
+                    1947.0,
+                    293.0,
+                    1947.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1941.0,
+                    354.0,
+                    1941.0,
+                    354.0,
+                    1977.0,
+                    294.0,
+                    1977.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    411.0,
+                    1941.0,
+                    661.0,
+                    1941.0,
+                    661.0,
+                    1977.0,
+                    411.0,
+                    1977.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    681.0,
+                    1941.0,
+                    1042.0,
+                    1941.0,
+                    1042.0,
+                    1977.0,
+                    681.0,
+                    1977.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1064.0,
+                    1941.0,
+                    1408.0,
+                    1941.0,
+                    1408.0,
+                    1977.0,
+                    1064.0,
+                    1977.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1967.0,
+                    552.0,
+                    1967.0,
+                    552.0,
+                    2012.0,
+                    292.0,
+                    2012.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    573.0,
+                    1967.0,
+                    978.0,
+                    1967.0,
+                    978.0,
+                    2012.0,
+                    573.0,
+                    2012.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1004.0,
+                    1967.0,
+                    1407.0,
+                    1967.0,
+                    1407.0,
+                    2012.0,
+                    1004.0,
+                    2012.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    2001.0,
+                    464.0,
+                    2001.0,
+                    464.0,
+                    2038.0,
+                    293.0,
+                    2038.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    493.0,
+                    2001.0,
+                    984.0,
+                    2001.0,
+                    984.0,
+                    2038.0,
+                    493.0,
+                    2038.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1093.0,
+                    1403.0,
+                    1093.0,
+                    1403.0,
+                    1127.0,
+                    294.0,
+                    1127.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1125.0,
+                    1403.0,
+                    1125.0,
+                    1403.0,
+                    1158.0,
+                    294.0,
+                    1158.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1151.0,
+                    1403.0,
+                    1151.0,
+                    1403.0,
+                    1190.0,
+                    293.0,
+                    1190.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1186.0,
+                    1405.0,
+                    1186.0,
+                    1405.0,
+                    1220.0,
+                    294.0,
+                    1220.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1214.0,
+                    1407.0,
+                    1214.0,
+                    1407.0,
+                    1250.0,
+                    292.0,
+                    1250.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1245.0,
+                    790.0,
+                    1245.0,
+                    790.0,
+                    1281.0,
+                    293.0,
+                    1281.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    881.0,
+                    1245.0,
+                    1405.0,
+                    1245.0,
+                    1405.0,
+                    1281.0,
+                    881.0,
+                    1281.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1277.0,
+                    1405.0,
+                    1277.0,
+                    1405.0,
+                    1311.0,
+                    296.0,
+                    1311.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1309.0,
+                    1117.0,
+                    1309.0,
+                    1117.0,
+                    1342.0,
+                    294.0,
+                    1342.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1582.0,
+                    1405.0,
+                    1582.0,
+                    1405.0,
+                    1619.0,
+                    294.0,
+                    1619.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1615.0,
+                    1404.0,
+                    1615.0,
+                    1404.0,
+                    1648.0,
+                    295.0,
+                    1648.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1644.0,
+                    1403.0,
+                    1644.0,
+                    1403.0,
+                    1678.0,
+                    294.0,
+                    1678.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1671.0,
+                    1403.0,
+                    1671.0,
+                    1403.0,
+                    1715.0,
+                    292.0,
+                    1715.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1705.0,
+                    759.0,
+                    1705.0,
+                    759.0,
+                    1742.0,
+                    295.0,
+                    1742.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    660.0,
+                    1407.0,
+                    660.0,
+                    1407.0,
+                    703.0,
+                    291.0,
+                    703.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    696.0,
+                    1404.0,
+                    696.0,
+                    1404.0,
+                    728.0,
+                    294.0,
+                    728.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    724.0,
+                    1403.0,
+                    724.0,
+                    1403.0,
+                    762.0,
+                    292.0,
+                    762.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    759.0,
+                    628.0,
+                    759.0,
+                    628.0,
+                    788.0,
+                    296.0,
+                    788.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    227.0,
+                    826.0,
+                    227.0,
+                    826.0,
+                    265.0,
+                    293.0,
+                    265.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    935.0,
+                    227.0,
+                    1405.0,
+                    227.0,
+                    1405.0,
+                    265.0,
+                    935.0,
+                    265.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    260.0,
+                    724.0,
+                    260.0,
+                    724.0,
+                    297.0,
+                    294.0,
+                    297.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    793.0,
+                    260.0,
+                    1405.0,
+                    260.0,
+                    1405.0,
+                    297.0,
+                    793.0,
+                    297.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    292.0,
+                    1405.0,
+                    292.0,
+                    1405.0,
+                    325.0,
+                    296.0,
+                    325.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    320.0,
+                    1408.0,
+                    320.0,
+                    1408.0,
+                    357.0,
+                    294.0,
+                    357.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    350.0,
+                    1143.0,
+                    350.0,
+                    1143.0,
+                    389.0,
+                    294.0,
+                    389.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    476.0,
+                    1104.0,
+                    476.0,
+                    1104.0,
+                    513.0,
+                    295.0,
+                    513.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1125.0,
+                    476.0,
+                    1405.0,
+                    476.0,
+                    1405.0,
+                    513.0,
+                    1125.0,
+                    513.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    507.0,
+                    704.0,
+                    507.0,
+                    704.0,
+                    544.0,
+                    295.0,
+                    544.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    816.0,
+                    507.0,
+                    1405.0,
+                    507.0,
+                    1405.0,
+                    544.0,
+                    816.0,
+                    544.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    540.0,
+                    380.0,
+                    540.0,
+                    380.0,
+                    573.0,
+                    297.0,
+                    573.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    472.0,
+                    540.0,
+                    824.0,
+                    540.0,
+                    824.0,
+                    573.0,
+                    472.0,
+                    573.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 17,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 5,
+                "poly": [
+                    503,
+                    528,
+                    1270,
+                    528,
+                    1270,
+                    1728,
+                    503,
+                    1728
+                ],
+                "score": 0.965,
+                "html": "<table><tr><td rowspan=1 colspan=1>(0101)</td><td rowspan=4 colspan=1>% LI&#x27;0干06&#x27;16% 01.0581555% 71&#x27;0 干 10&#x27;96% 710 67&#x27;96% II&#x27;0干08&#x27;96% L70干08&#x27;88% 11011.55% 910 76.56% 71&#x27;0干80&#x27;96% 110 干 9596% 780 6444% 18031.95% 7109 8535% 40.0 99.56% 870 LL&#x27;46C[ = y[= T (sani) sSuiPPq Tt Tr-shyt[ = y[ = T (sA) P T T-SAT r rrrrret gaasetr Srrpeirrlrer iaerrertrrnrrltgFaraeregarrgrrirrettA4=P444</td><td rowspan=2 colspan=1>% 0509 86:10% 1£&#x27;0干0L09%50055672% 67089&#x27;89% 0000658% 104530% 01004535% 44061.44% 141 65:14%70015:35%03.0335.55% 0ε0干 S0&#x27;LL% 51081:14%41044:44</td><td rowspan=1 colspan=1>% 700 47.99% 11&#x27;0干 58&#x27;96% 01&#x27;0干 66&#x27;96% 010 干 87&#x27; L6% 600 干13&#x27; L6% 01756% 010  6</td><td rowspan=1 colspan=1>% L0&#x27;0 干 1S&#x27;86</td></tr><tr><td rowspan=2 colspan=1>(s01)</td><td rowspan=2 colspan=1>% LI&#x27;0干 S0&#x27;96% 170干19&#x27;96% 110T 6696% 010干8126% 010776% 010干£1L6% 010干 80&#x27;L6</td><td rowspan=2 colspan=1>% L00干1S&#x27;86</td></tr><tr><td rowspan=1 colspan=1>% 44061.44% 141 65:14%70015:35%03.0335.55% 0ε0干 S0&#x27;LL% 51081:14%41044:44</td></tr><tr><td rowspan=1 colspan=1>(101)(tous em)1oila</td><td rowspan=1 colspan=1>%230325:65% 680 F980%57025224%944144% 14011.64%85033335% 3：271:31[[ = y O = T‘(sIni) suipP-qs u Tr-t[ = y  = (sani) supPqa uu Trn-st[[ = y OII = T (I1I) SSA- V T-NC[ = yI=T sa   -T=grrrraree geasepr rripegahrlerserrerarrnriirenggareregaerrriipetP=P-14</td><td rowspan=1 colspan=1>% 155311:55% 7408126% 31.074.16%050333:5% 610干80&#x27;96% LI0干 6996% L10 干8596C = y[= T(sa) supq c -[[ = y O = T‘(sni) sYuiPPqs uu Tr-sntC = y O = T () P 1 TI-sC=y[= T(I)A T T-sAg rrirpre aaetr rrirearrger serrerarrnriirertggaraerrarrrriiiertt=p 4</td><td rowspan=1 colspan=1>% L1&#x27;0干Iε&#x27;L6Surreend pasiersnnGOLnaier Giii-larei</td></tr></table>"
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    298,
+                    76,
+                    815,
+                    76,
+                    815,
+                    104,
+                    298,
+                    104
+                ],
+                "score": 0.887
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    836,
+                    2088,
+                    865,
+                    2088,
+                    865,
+                    2112,
+                    836,
+                    2112
+                ],
+                "score": 0.836
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    73.0,
+                    817.0,
+                    73.0,
+                    817.0,
+                    108.0,
+                    296.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    832.0,
+                    2085.0,
+                    870.0,
+                    2085.0,
+                    870.0,
+                    2124.0,
+                    832.0,
+                    2124.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 18,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 5,
+                "poly": [
+                    429,
+                    467,
+                    1275,
+                    467,
+                    1275,
+                    1786,
+                    429,
+                    1786
+                ],
+                "score": 0.966,
+                "html": "<table><tr><td rowspan=7 colspan=1>(s 0)(107)(s‘s)(1‘s)(ous em)oiiral</td><td rowspan=7 colspan=1>% 944% 810 449% 67&#x27;0干 10&#x27;61% 110013.55% 810 11&#x27;9%1102555% 6908&#x27;44% 690441% 1∠&#x27;06100%20035174% 32056577%33357530PPT =γ[= TIIII-SALAI SI11.1ay = y  = 111A1--4111rres ionnaieraBeesees</td><td rowspan=4 colspan=1>% 90.038399% 33.333339% 350 79:11  %80085.8  % 990 41.69 % 1333 3334% L00干 L7&#x27;996%83033139% 670227/4%3330730 % 30033331% 8109835% 4404 44.14  % 1 1124</td><td rowspan=1 colspan=1>% 90001.95% 2102 78.14% 401141% L£&#x27;0干90% 6£&#x27;0干 LL&#x27;0  %37:227755% 8057788</td><td rowspan=1 colspan=2>% 0105 6796% L0&#x27;0干18&#x27;86</td></tr><tr><td rowspan=1 colspan=1>% 90001.95% 2102 78.14% 401141% L£&#x27;0干90% 6£&#x27;0干 LL&#x27;0  %37:227755% 8057788% 30:0377.3050</td><td rowspan=2 colspan=2>% 0105 6796% L0&#x27;0干18&#x27;86% 700339:55% 810533.55</td></tr><tr><td rowspan=2 colspan=1>%11033135% 190 08:27% 770955.66%71224115% 210370.65% 200586.655210224555%335395559% 33033151</td></tr><tr><td></td><td rowspan=1 colspan=1>土+</td></tr><tr><td rowspan=3 colspan=1>% LS0干9118%8107818%S9007&#x27;LL% LL&#x27;0干60&#x27;IL% 09&#x27;0干18&#x27;18%8703355% 353355.55% 0S&#x27;0 干8L&#x27;L8% 3903 8138% 1113 1511% 711 819% 7.0 16:14  %7.055  % 08&#x27;0 干 64&#x27;99  % 480  71896260 =   =  () T-       CCT = y TT=1 (In) s11T1II-SSCC = γ OII = 1 ‘(S.I1I) T1I1-SALOCY =γ[=T (SIAI) TIII-SAI8 = Y inodop p PT SuippggregreemysppegarierereereirrA44=P4444</td><td rowspan=1 colspan=1></td><td></td></tr><tr><td rowspan=1 colspan=1>% I∠&#x27;0干 90&#x27;89% 99&#x27;0 干7L&#x27;89% 6/0 9579%8/07985% 69&#x27;009&#x27;L% 6905555% 19085.94% S9&#x27;0 干99&#x27;8L% L0干69</td><td rowspan=2 colspan=2>% 710£8&#x27;86% 60&#x27;0干85&#x27;66% 11.03.9555% 773523:32Surureree peseeedn  GDi Drirr eoeiiioriireri（OLDLier Girr-aarei</td></tr><tr><td rowspan=1 colspan=1>% 71 1171% 1742 1185% 6104544% 080 964% 5039.02 % 0807650% 18081&#x27;89% 71:044440 =  O = P (I) T-sdHP&#x27; =   = P(s) -s-L CC =y OII = 1 (SIAN) TSII-SA/S0C = y =  (i) sTod-snt8 = Y inidoip pm d SuippqC =yI =1 (SIA) T-SAagrrpaee easeta Srrplgrhgier neeennr neipreggarerrgrereippegPP =P 4400</td></tr></table>"
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    299,
+                    76,
+                    815,
+                    76,
+                    815,
+                    104,
+                    299,
+                    104
+                ],
+                "score": 0.884
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    835,
+                    2088,
+                    865,
+                    2088,
+                    865,
+                    2113,
+                    835,
+                    2113
+                ],
+                "score": 0.846
+            },
+            {
+                "category_id": 7,
+                "poly": [
+                    331,
+                    257,
+                    400,
+                    257,
+                    400,
+                    2020,
+                    331,
+                    2020
+                ],
+                "score": 0.513
+            },
+            {
+                "category_id": 6,
+                "poly": [
+                    1288,
+                    530,
+                    1328,
+                    530,
+                    1328,
+                    1772,
+                    1288,
+                    1772
+                ],
+                "score": 0.182
+            },
+            {
+                "category_id": 6,
+                "poly": [
+                    1325,
+                    1436,
+                    1347,
+                    1436,
+                    1347,
+                    1720,
+                    1325,
+                    1720
+                ],
+                "score": 0.107
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    73.0,
+                    816.0,
+                    73.0,
+                    816.0,
+                    108.0,
+                    295.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    829.0,
+                    2083.0,
+                    871.0,
+                    2083.0,
+                    871.0,
+                    2123.0,
+                    829.0,
+                    2123.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    329.0,
+                    244.0,
+                    373.0,
+                    244.0,
+                    373.0,
+                    2031.0,
+                    329.0,
+                    2031.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    355.0,
+                    448.0,
+                    404.0,
+                    448.0,
+                    404.0,
+                    2033.0,
+                    355.0,
+                    2033.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1285.0,
+                    550.0,
+                    1331.0,
+                    550.0,
+                    1331.0,
+                    1762.0,
+                    1285.0,
+                    1762.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1323.0,
+                    1709.0,
+                    1334.0,
+                    1709.0,
+                    1334.0,
+                    1724.0,
+                    1323.0,
+                    1724.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1319.0,
+                    1445.0,
+                    1352.0,
+                    1445.0,
+                    1352.0,
+                    1728.0,
+                    1319.0,
+                    1728.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 19,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 5,
+                "poly": [
+                    492,
+                    473,
+                    1283,
+                    473,
+                    1283,
+                    1789,
+                    492,
+                    1789
+                ],
+                "score": 0.969,
+                "html": "<table><tr><td rowspan=1 colspan=1>(0s&#x27;s)(2 &#x27;0)(s‘s)(1‘s)(ous Xem)Woirai</td><td rowspan=1 colspan=1>% 5/.05 39.61% 533112% 9908589% 67003735Yrirlrriiir reirraperreeees</td><td rowspan=1 colspan=1>% 230309.3% 31:15:11% 1/.0 91&#x27;81%55333.99% 00&#x27;0干00&#x27;  % 89&#x27;0干 86&#x27;L% 89&#x27;0干 16&#x27;99% L9:0 17:99546%344231.21% 100404% 7/0 9000     %8900L&#x27;79%57003515% 5703.31%£9090:65%81096565% 0000000 %15457171% 850 87.15%05099511% 3513 95:75% 1 8 2 470036.74     %990干8000干（09.4士22.5519&#x27;6725&#x27;330 = w Xs =  (s) Ti-H6&#x27;0 = 1 O = P (s) T--H CC1 = yO1= P (SIA4） PIII-SA/CT1 =yFT= (sT) s1TI-sg = Y iniip pi P grippC =y[=1 (SIA) TI-SAS =yrrrpaee gaisnpa Seippegger nsereny iprgasereresteippeggPP=P4440</td><td rowspan=1 colspan=1>% 8030639%5500854% £9&#x27;0干 56:09% 1555585% 70&#x27;0干I0% 99055% 79&#x27;0LL% £90干 5969% 553515:396566% 1440444.94%5506554  %41:045.44       % 0105809% 89045:1989&#x27;0921% 42:015%0706/:66 277333755 %1/044:64% 71172% 010 1690% 3303.73.51% 17:5--55.85%%   %%119.03 30.67990干  5503066120.00555555830 =  Os = P (sI1I) Tt--AdL1HT&#x27;0 = 1s= P sa) T-H C=yE1=PIA S-ACC = y I= P （(I) sTTd-SHg = Y iniip pi P grippaggC1 =yT=1A） T-SNarrpiereisepaSerpggrhler errrairrireigrrsesereesseippeggz =pashisea4</td><td rowspan=1 colspan=1>% 77:045:14% 09044% 69000.14% 590F 50.00% 7/0511.79% 1/0 6777% LL&#x27;0干 18&#x27;94% 9/0 95.94Surererrer lseeetdnnODTLr) Giii-rriOrnerer eeeinerrceeret</td></tr></table>"
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    298,
+                    76,
+                    815,
+                    76,
+                    815,
+                    104,
+                    298,
+                    104
+                ],
+                "score": 0.885
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    834,
+                    2088,
+                    863,
+                    2088,
+                    863,
+                    2113,
+                    834,
+                    2113
+                ],
+                "score": 0.833
+            },
+            {
+                "category_id": 7,
+                "poly": [
+                    391,
+                    277,
+                    459,
+                    277,
+                    459,
+                    2027,
+                    391,
+                    2027
+                ],
+                "score": 0.555
+            },
+            {
+                "category_id": 7,
+                "poly": [
+                    1196,
+                    1497,
+                    1283,
+                    1497,
+                    1283,
+                    1768,
+                    1196,
+                    1768
+                ],
+                "score": 0.204
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    73.0,
+                    817.0,
+                    73.0,
+                    817.0,
+                    108.0,
+                    296.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    830.0,
+                    2085.0,
+                    868.0,
+                    2085.0,
+                    868.0,
+                    2124.0,
+                    830.0,
+                    2124.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    390.0,
+                    264.0,
+                    434.0,
+                    264.0,
+                    434.0,
+                    2037.0,
+                    390.0,
+                    2037.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    418.0,
+                    451.0,
+                    464.0,
+                    451.0,
+                    464.0,
+                    2038.0,
+                    418.0,
+                    2038.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1251.0,
+                    1494.0,
+                    1281.0,
+                    1494.0,
+                    1281.0,
+                    1774.0,
+                    1251.0,
+                    1774.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1194.0,
+                    1504.0,
+                    1228.0,
+                    1504.0,
+                    1228.0,
+                    1774.0,
+                    1194.0,
+                    1774.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1224.0,
+                    1513.0,
+                    1252.0,
+                    1513.0,
+                    1252.0,
+                    1773.0,
+                    1224.0,
+                    1773.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 20,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 5,
+                "poly": [
+                    508,
+                    865,
+                    1189,
+                    865,
+                    1189,
+                    1498,
+                    508,
+                    1498
+                ],
+                "score": 0.979,
+                "html": "<table><tr><td>Algorithm</td><td>(2,5)</td></tr><tr><td>Baselines Training from scratch</td><td>63.19 ± 1.06 %</td></tr><tr><td>BiGAN,d= 200</td><td></td></tr><tr><td>Embedding knn-nearest neighbors Embedding linearclassifier</td><td>56.15 ± 0.89 % 58.44 ± 0.90 %</td></tr><tr><td>Embedding MLP with dropout,h = 128</td><td>56.26 ± 0.94 %</td></tr><tr><td>Embedding cluster matching,k = 500</td><td>56.20 ±1.00 %</td></tr><tr><td>CACTUs-MAML(ours),P= 50,k= 500 CACTUs-ProtoNets (ours),P= 50,k = 500</td><td>74.98 ± 1.02 %</td></tr><tr><td></td><td>65.58 ± 1.04 %</td></tr><tr><td>DeepCluster,d= 256 Embedding knn-nearest neighbors</td><td></td></tr><tr><td>Embedding linear classifier</td><td>61.47 ± 0.99 % 59.57 ± 0.98 %</td></tr><tr><td>Embedding MLP with dropout,h = 128</td><td>60.65 ± 0.98 %</td></tr><tr><td>Embedding cluster matching,k = 500</td><td>51.51 ± 0.89 %</td></tr><tr><td>CACTUs-MAML (ours),P= 50,k = 500</td><td>73.79 ± 1.01 %</td></tr><tr><td>CACTUs-ProtoNets (ours),P= 50,k = 500</td><td>74.15 ± 1.02 %</td></tr><tr><td></td><td></td></tr><tr><td>Supervised meta-learning</td><td></td></tr><tr><td>Oracle-MAML (control)</td><td>87.10 ± 0.85 %</td></tr><tr><td></td><td></td></tr><tr><td>Oracle-ProtoNets (control)</td><td>85.13 ± 0.92 %</td></tr></table>"
+            },
+            {
+                "category_id": 6,
+                "poly": [
+                    299,
+                    741,
+                    1404,
+                    741,
+                    1404,
+                    827,
+                    299,
+                    827
+                ],
+                "score": 0.923
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    299,
+                    75,
+                    816,
+                    75,
+                    816,
+                    105,
+                    299,
+                    105
+                ],
+                "score": 0.878
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    834,
+                    2088,
+                    865,
+                    2088,
+                    865,
+                    2113,
+                    834,
+                    2113
+                ],
+                "score": 0.842
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1237,
+                    743,
+                    1286,
+                    743,
+                    1286,
+                    769,
+                    1237,
+                    769
+                ],
+                "score": 0.86,
+                "latex": "9 5 \\%"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    633,
+                    1224,
+                    667,
+                    1224,
+                    667,
+                    1250,
+                    633,
+                    1250
+                ],
+                "score": 0.86,
+                "latex": "k _ { \\mathrm { n n } }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    633,
+                    1017,
+                    667,
+                    1017,
+                    667,
+                    1042,
+                    633,
+                    1042
+                ],
+                "score": 0.84,
+                "latex": "k _ { \\mathrm { n n } }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    650,
+                    1195,
+                    741,
+                    1195,
+                    741,
+                    1221,
+                    650,
+                    1221
+                ],
+                "score": 0.77,
+                "latex": "d = 2 5 6"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    520,
+                    799,
+                    542,
+                    799,
+                    542,
+                    823,
+                    520,
+                    823
+                ],
+                "score": 0.76,
+                "latex": "P"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    597,
+                    988,
+                    686,
+                    988,
+                    686,
+                    1014,
+                    597,
+                    1014
+                ],
+                "score": 0.73,
+                "latex": "d = 2 0 0"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1108,
+                    744,
+                    1136,
+                    744,
+                    1136,
+                    770,
+                    1108,
+                    770
+                ],
+                "score": 0.73,
+                "latex": "\\pm"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    831,
+                    1279,
+                    922,
+                    1279,
+                    922,
+                    1304,
+                    831,
+                    1304
+                ],
+                "score": 0.66,
+                "latex": "h = 1 2 8"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    725,
+                    771,
+                    742,
+                    771,
+                    742,
+                    795,
+                    725,
+                    795
+                ],
+                "score": 0.62,
+                "latex": "h"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    389,
+                    771,
+                    406,
+                    771,
+                    406,
+                    795,
+                    389,
+                    795
+                ],
+                "score": 0.59,
+                "latex": "d"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1264,
+                    772,
+                    1281,
+                    772,
+                    1281,
+                    795,
+                    1264,
+                    795
+                ],
+                "score": 0.58,
+                "latex": "k"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    831,
+                    1070,
+                    923,
+                    1070,
+                    923,
+                    1097,
+                    831,
+                    1097
+                ],
+                "score": 0.56,
+                "latex": "h = 1 2 8"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    769,
+                    1127,
+                    851,
+                    1127,
+                    851,
+                    1152,
+                    769,
+                    1152
+                ],
+                "score": 0.54,
+                "latex": "P = 5 0"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    811,
+                    1306,
+                    902,
+                    1306,
+                    902,
+                    1332,
+                    811,
+                    1332
+                ],
+                "score": 0.49,
+                "latex": "k = 5 0 0"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    769,
+                    1335,
+                    852,
+                    1335,
+                    852,
+                    1360,
+                    769,
+                    1360
+                ],
+                "score": 0.44,
+                "latex": "P = 5 0"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    791,
+                    1155,
+                    873,
+                    1155,
+                    873,
+                    1181,
+                    791,
+                    1181
+                ],
+                "score": 0.39,
+                "latex": "P = 5 0"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    811,
+                    1099,
+                    902,
+                    1099,
+                    902,
+                    1125,
+                    811,
+                    1125
+                ],
+                "score": 0.36,
+                "latex": "k = 5 0 0"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    791,
+                    1362,
+                    873,
+                    1362,
+                    873,
+                    1389,
+                    791,
+                    1389
+                ],
+                "score": 0.35,
+                "latex": "P = 5 0"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    883,
+                    1154,
+                    974,
+                    1154,
+                    974,
+                    1183,
+                    883,
+                    1183
+                ],
+                "score": 0.34,
+                "latex": "k = 5 0 0"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    861,
+                    1126,
+                    952,
+                    1126,
+                    952,
+                    1152,
+                    861,
+                    1152
+                ],
+                "score": 0.29,
+                "latex": "k = 5 0 0"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    741.0,
+                    1107.0,
+                    741.0,
+                    1107.0,
+                    774.0,
+                    294.0,
+                    774.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1137.0,
+                    741.0,
+                    1236.0,
+                    741.0,
+                    1236.0,
+                    774.0,
+                    1137.0,
+                    774.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1287.0,
+                    741.0,
+                    1403.0,
+                    741.0,
+                    1403.0,
+                    774.0,
+                    1287.0,
+                    774.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    769.0,
+                    388.0,
+                    769.0,
+                    388.0,
+                    798.0,
+                    295.0,
+                    798.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    407.0,
+                    769.0,
+                    724.0,
+                    769.0,
+                    724.0,
+                    798.0,
+                    407.0,
+                    798.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    743.0,
+                    769.0,
+                    1263.0,
+                    769.0,
+                    1263.0,
+                    798.0,
+                    743.0,
+                    798.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1282.0,
+                    769.0,
+                    1404.0,
+                    769.0,
+                    1404.0,
+                    798.0,
+                    1282.0,
+                    798.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    796.0,
+                    519.0,
+                    796.0,
+                    519.0,
+                    832.0,
+                    294.0,
+                    832.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    543.0,
+                    796.0,
+                    1037.0,
+                    796.0,
+                    1037.0,
+                    832.0,
+                    543.0,
+                    832.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    73.0,
+                    817.0,
+                    73.0,
+                    817.0,
+                    108.0,
+                    295.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    829.0,
+                    2084.0,
+                    871.0,
+                    2084.0,
+                    871.0,
+                    2125.0,
+                    829.0,
+                    2125.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 21,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    525,
+                    1404,
+                    525,
+                    1404,
+                    1013,
+                    297,
+                    1013
+                ],
+                "score": 0.985
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1561,
+                    1404,
+                    1561,
+                    1404,
+                    2019,
+                    297,
+                    2019
+                ],
+                "score": 0.983
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    296,
+                    1404,
+                    296,
+                    1404,
+                    510,
+                    298,
+                    510
+                ],
+                "score": 0.981
+            },
+            {
+                "category_id": 5,
+                "poly": [
+                    554,
+                    1084,
+                    1148,
+                    1084,
+                    1148,
+                    1517,
+                    554,
+                    1517
+                ],
+                "score": 0.981,
+                "html": "<table><tr><td>Hyperparameter</td><td>Value</td></tr><tr><td>Input size</td><td>224× 224</td></tr><tr><td>Outer (meta) learning rate</td><td>0.0001</td></tr><tr><td>Inner learning rate</td><td>0.001</td></tr><tr><td>Task batch size</td><td>3</td></tr><tr><td>Inner adaptation steps (meta-training)</td><td>5</td></tr><tr><td>Meta-training iterations</td><td>240,000</td></tr><tr><td>Adaptation steps (evaluation)</td><td>100</td></tr><tr><td>Classes per task (meta-training)</td><td>5</td></tr><tr><td>Shots per class (meta-training)</td><td>1</td></tr><tr><td>Queries per class</td><td>5</td></tr><tr><td>Residual blocks</td><td>5</td></tr><tr><td>Layers per residual block</td><td>2</td></tr></table>"
+            },
+            {
+                "category_id": 6,
+                "poly": [
+                    555,
+                    1040,
+                    1144,
+                    1040,
+                    1144,
+                    1071,
+                    555,
+                    1071
+                ],
+                "score": 0.919
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    300,
+                    225,
+                    888,
+                    225,
+                    888,
+                    262,
+                    300,
+                    262
+                ],
+                "score": 0.901
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    299,
+                    76,
+                    816,
+                    76,
+                    816,
+                    104,
+                    299,
+                    104
+                ],
+                "score": 0.884
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    835,
+                    2088,
+                    864,
+                    2088,
+                    864,
+                    2112,
+                    835,
+                    2112
+                ],
+                "score": 0.851
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1054,
+                    771,
+                    1093,
+                    771,
+                    1093,
+                    800,
+                    1054,
+                    800
+                ],
+                "score": 0.89,
+                "latex": "k _ { \\mathrm { n n } }"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    551.0,
+                    1036.0,
+                    1146.0,
+                    1036.0,
+                    1146.0,
+                    1076.0,
+                    551.0,
+                    1076.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    222.0,
+                    894.0,
+                    222.0,
+                    894.0,
+                    267.0,
+                    295.0,
+                    267.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    73.0,
+                    817.0,
+                    73.0,
+                    817.0,
+                    108.0,
+                    295.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    828.0,
+                    2084.0,
+                    871.0,
+                    2084.0,
+                    871.0,
+                    2123.0,
+                    828.0,
+                    2123.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    524.0,
+                    1402.0,
+                    524.0,
+                    1402.0,
+                    562.0,
+                    294.0,
+                    562.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    556.0,
+                    1404.0,
+                    556.0,
+                    1404.0,
+                    590.0,
+                    295.0,
+                    590.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    585.0,
+                    1406.0,
+                    585.0,
+                    1406.0,
+                    621.0,
+                    292.0,
+                    621.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    618.0,
+                    1406.0,
+                    618.0,
+                    1406.0,
+                    653.0,
+                    295.0,
+                    653.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    648.0,
+                    1404.0,
+                    648.0,
+                    1404.0,
+                    682.0,
+                    295.0,
+                    682.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    677.0,
+                    1402.0,
+                    677.0,
+                    1402.0,
+                    711.0,
+                    295.0,
+                    711.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    708.0,
+                    1404.0,
+                    708.0,
+                    1404.0,
+                    745.0,
+                    294.0,
+                    745.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    740.0,
+                    1402.0,
+                    740.0,
+                    1402.0,
+                    774.0,
+                    295.0,
+                    774.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    770.0,
+                    1053.0,
+                    770.0,
+                    1053.0,
+                    805.0,
+                    295.0,
+                    805.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1094.0,
+                    770.0,
+                    1402.0,
+                    770.0,
+                    1402.0,
+                    805.0,
+                    1094.0,
+                    805.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    798.0,
+                    1404.0,
+                    798.0,
+                    1404.0,
+                    835.0,
+                    292.0,
+                    835.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    828.0,
+                    1405.0,
+                    828.0,
+                    1405.0,
+                    868.0,
+                    291.0,
+                    868.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    861.0,
+                    1405.0,
+                    861.0,
+                    1405.0,
+                    895.0,
+                    295.0,
+                    895.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    890.0,
+                    1404.0,
+                    890.0,
+                    1404.0,
+                    926.0,
+                    295.0,
+                    926.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    922.0,
+                    1406.0,
+                    922.0,
+                    1406.0,
+                    956.0,
+                    296.0,
+                    956.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    955.0,
+                    1405.0,
+                    955.0,
+                    1405.0,
+                    986.0,
+                    295.0,
+                    986.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    982.0,
+                    370.0,
+                    982.0,
+                    370.0,
+                    1015.0,
+                    292.0,
+                    1015.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1560.0,
+                    1404.0,
+                    1560.0,
+                    1404.0,
+                    1596.0,
+                    295.0,
+                    1596.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1591.0,
+                    1406.0,
+                    1591.0,
+                    1406.0,
+                    1627.0,
+                    293.0,
+                    1627.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1619.0,
+                    1405.0,
+                    1619.0,
+                    1405.0,
+                    1659.0,
+                    292.0,
+                    1659.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1654.0,
+                    1402.0,
+                    1654.0,
+                    1402.0,
+                    1685.0,
+                    296.0,
+                    1685.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1684.0,
+                    1404.0,
+                    1684.0,
+                    1404.0,
+                    1719.0,
+                    295.0,
+                    1719.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1714.0,
+                    1404.0,
+                    1714.0,
+                    1404.0,
+                    1749.0,
+                    294.0,
+                    1749.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1744.0,
+                    1406.0,
+                    1744.0,
+                    1406.0,
+                    1779.0,
+                    294.0,
+                    1779.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1775.0,
+                    1405.0,
+                    1775.0,
+                    1405.0,
+                    1810.0,
+                    295.0,
+                    1810.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1801.0,
+                    1408.0,
+                    1801.0,
+                    1408.0,
+                    1842.0,
+                    292.0,
+                    1842.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1836.0,
+                    1406.0,
+                    1836.0,
+                    1406.0,
+                    1871.0,
+                    294.0,
+                    1871.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1867.0,
+                    1404.0,
+                    1867.0,
+                    1404.0,
+                    1898.0,
+                    296.0,
+                    1898.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1895.0,
+                    1405.0,
+                    1895.0,
+                    1405.0,
+                    1929.0,
+                    295.0,
+                    1929.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1928.0,
+                    1402.0,
+                    1928.0,
+                    1402.0,
+                    1959.0,
+                    296.0,
+                    1959.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1956.0,
+                    1405.0,
+                    1956.0,
+                    1405.0,
+                    1996.0,
+                    292.0,
+                    1996.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1988.0,
+                    889.0,
+                    1988.0,
+                    889.0,
+                    2022.0,
+                    295.0,
+                    2022.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    293.0,
+                    1406.0,
+                    293.0,
+                    1406.0,
+                    334.0,
+                    293.0,
+                    334.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    327.0,
+                    1408.0,
+                    327.0,
+                    1408.0,
+                    361.0,
+                    294.0,
+                    361.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    355.0,
+                    1405.0,
+                    355.0,
+                    1405.0,
+                    396.0,
+                    292.0,
+                    396.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    387.0,
+                    1404.0,
+                    387.0,
+                    1404.0,
+                    423.0,
+                    295.0,
+                    423.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    418.0,
+                    1405.0,
+                    418.0,
+                    1405.0,
+                    455.0,
+                    292.0,
+                    455.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    451.0,
+                    1405.0,
+                    451.0,
+                    1405.0,
+                    481.0,
+                    296.0,
+                    481.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    477.0,
+                    815.0,
+                    477.0,
+                    815.0,
+                    513.0,
+                    293.0,
+                    513.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 22,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 5,
+                "poly": [
+                    681,
+                    441,
+                    1089,
+                    441,
+                    1089,
+                    1829,
+                    681,
+                    1829
+                ],
+                "score": 0.925,
+                "html": "<table><tr><td>(os&#x27;s) （07&#x27;s) (s‘s) (1‘s) (ous xem) 1loirrra</td><td>% LL0干 89&#x27;6S% SL0干 L9&#x27;εS % 3901/30% 11:059:85 Ppt =pessstsesg Yrrlrriiir liiiiera reneseg</td><td>% 41:0 4244 % L70 干 0106 % 3705 75.77 % 040 18.61 % 130 11:11 % 990F 5878 % 17034355 % 00 1064 % 1/0494 % I9&#x27;0 干 LI&#x27;01 % 7703181 % 4101109 [[V[ = γ[=1 ‘(SJ1) TII-SALO0 ohhgrer serrerarg nrippegg garreerrnrilpestg</td><td>% 190  L728 % 690£8&#x27;78 %0/0£494 % SL&#x27;0 干 SL&#x27;19 Sururersritu prseetdeesntts Jra psrsN-</td><td>% 1/0 £7&#x27;98 % L/071:4 Srinerniar peseadn DEDLLrr Tirr-larrt</td><td>% 990 干 90℃6% 69&#x27;0干4116</td></tr></table>"
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    299,
+                    76,
+                    815,
+                    76,
+                    815,
+                    104,
+                    299,
+                    104
+                ],
+                "score": 0.888
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    834,
+                    2088,
+                    865,
+                    2088,
+                    865,
+                    2112,
+                    834,
+                    2112
+                ],
+                "score": 0.845
+            },
+            {
+                "category_id": 6,
+                "poly": [
+                    583,
+                    301,
+                    648,
+                    301,
+                    648,
+                    2064,
+                    583,
+                    2064
+                ],
+                "score": 0.342
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    73.0,
+                    816.0,
+                    73.0,
+                    816.0,
+                    108.0,
+                    295.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    829.0,
+                    2084.0,
+                    872.0,
+                    2084.0,
+                    872.0,
+                    2124.0,
+                    829.0,
+                    2124.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    585.0,
+                    292.0,
+                    622.0,
+                    292.0,
+                    622.0,
+                    2034.0,
+                    585.0,
+                    2034.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    610.0,
+                    1016.0,
+                    655.0,
+                    1016.0,
+                    655.0,
+                    2038.0,
+                    610.0,
+                    2038.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 23,
+            "width": 1700,
+            "height": 2200
+        }
+    }
+]
\ No newline at end of file
diff --git a/parse/train/ryg7vA4tPB/ryg7vA4tPB.md b/parse/train/ryg7vA4tPB/ryg7vA4tPB.md
new file mode 100644
index 0000000000000000000000000000000000000000..22f8cbc5f02e39c1f51ac712abbde8e65ae2e1df
--- /dev/null
+++ b/parse/train/ryg7vA4tPB/ryg7vA4tPB.md
@@ -0,0 +1,363 @@
+# RIGGING THE LOTTERY: MAKING ALL TICKETS WINNERS
+
+Anonymous authors Paper under double-blind review
+
+# ABSTRACT
+
+Sparse neural networks have been shown to be more parameter and compute efficient compared to dense networks and in some cases they are even successfully used to decrease wall clock inference times. There is a large body of work on training dense networks to yield sparse networks for inference (Molchanov et al., 2017; Zhu & Gupta, 2018; Louizos et al., 2017; Li et al., 2016; Guo et al., 2016). This limits the size of the largest trainable sparse model to that of the largest trainable dense model. In this paper we introduce a method to train sparse neural networks with a fixed parameter count and a fixed computational cost throughout training, without sacrificing accuracy relative to existing dense-to-sparse training methods. Our method updates the topology of the network during training by using parameter magnitudes and infrequent gradient calculations. We show that this approach requires fewer floating-point operations (FLOPs) to achieve a given level of accuracy compared to prior techniques. We demonstrate state-of-the-art sparse training results with ResNet-50, MobileNet v1 and MobileNet v2 on the ImageNet-2012 dataset, WideResNets on the CIFAR-10 dataset and RNNs on the WikiText-103 dataset. Finally, we provide some insights into why allowing the topology to change during the optimization can overcome local minima encountered when the topology remains static.
+
+# 1 INTRODUCTION
+
+The parameter and floating point operation (FLOP) efficiency of sparse neural networks is now well demonstrated on a variety of problems (Han et al., 2015; Srinivas et al., 2017). Some work has even shown inference time speedups are possible on Recurrent Neural Networks (RNNs) (Kalchbrenner et al., 2018) and Convolutional Neural Networks (ConvNets) (Park et al., 2016). Currently, the most accurate sparse models are obtained with techniques that require, at a minimum, the cost of training a dense model in terms of memory and FLOPs (Zhu & Gupta, 2018; Guo et al., 2016), and sometimes significantly more (Molchanov et al., 2017). This paradigm has two main limitations:
+
+1. The maximum size of sparse models is limited to the largest dense model that can be trained. Even if sparse models are more parameter efficient, we can’t use pruning to train models that are larger and more accurate than the largest possible dense models.   
+2. It is inefficient. Large amounts of computation must be performed for parameters that are zero valued or that will be zero during inference.
+
+Additionally, it remains unknown if the performance of the current best pruning algorithms are an upper bound on the quality of sparse models. Gale et al. (2019) found that three different dense-tosparse training algorithms all achieve about the same sparsity / accuracy trade-off. However, this is far from conclusive proof that no better performance is possible. In this work we show the surprising result that dynamic sparse training, which includes the method we introduce below, can find more accurate models than the current best approaches to pruning initially dense networks. Importantly, our method does not change the FLOPs required to execute the model during training, allowing one to decide on a specific inference cost prior to training.
+
+The Lottery Ticket Hypothesis (Frankle & Carbin, 2019) hypothesized that if we can find a sparse neural network with iterative pruning, then we can train that sparse network from scratch, to the same level of accuracy, by starting from the original initial conditions. In this paper we introduce a new method for training sparse models without the need of a “lucky” initialization; for this reason, we call our method “The Rigged Lottery” or $R i g L ^ { * }$ . We show that this method is:
+
+• Memory efficient: It requires memory only proportional to the size of the sparse model. It never requires storing quantities that are the size of the dense model. This is in contrast to Dettmers & Zettlemoyer (2019) which requires storing the momentum for all parameters, even those that are zero valued.   
+• Computationally efficient: The amount of computation required to train the model is proportional to the number of nonzero parameters in the model.   
+• Accurate: The performance achieved by the method matches and sometimes exceeds the performance of pruning based approaches.
+
+Our method works by infrequently using instantaneous gradient information to inform a re-wiring of the network. We show that this allows the optimization to escape local minima where it would otherwise become trapped if the sparsity pattern were to remain static. Crucially, as long as the full gradient information is needed less than every $\frac { 1 } { 1 - s p a r s i t y }$ iterations, then the overall work remains proportional to the model sparsity.
+
+# 2 RELATED WORK
+
+Research on finding sparse neural networks dates back decades; for example, at least to Thimm & Fiesler (1995) who concluded that pruning weights based on magnitude was a simple and powerful technique. Strom (1997) later introduced the idea of retraining the previously pruned network to ¨ increase accuracy. Han et al. (2016b) went further and introduced multiple rounds of magnitude pruning and retraining. This is, however, relatively inefficient, requiring ten rounds of retraining when removing $2 0 \%$ of the connections to reach a final sparsity of $9 0 \%$ . To overcome this problem, Narang et al. (2017) introduced gradual pruning, where connections are slowly removed over the course of a single round of training. Zhu & Gupta (2018) refined the technique to minimize the amount of hyper-parameter selection required.
+
+A diversity of approaches not based on magnitude based pruning have also been proposed. LeCun et al. (1990) and Hassibi & Stork (1993) are some early examples, but impractical for modern neural networks as they use information from the Hessian to prune a trained network. More recent work includes $L _ { 0 }$ Regularization (Christos Louizos, 2018), Variational Dropout (Molchanov et al., 2017), Dynamic Network Surgery (Guo et al., 2016) and Sensitivity Driven Regularization (Tartaglione et al., 2018). Gale et al. (2019) examined magnitude pruning, $L _ { 0 }$ Regularization and Variational Dropout and concluded that they all achieve about the same accuracy versus sparsity trade-off on ResNet-50 and Transformer architectures.
+
+Training techniques that allow for sparsity throughout the entire training process were, to our knowledge, first introduced in Deep Rewiring (DeepR) (Bellec et al., 2017). In DeepR, the standard Stochastic Gradient Descent (SGD) optimizer is augmented with a random walk in parameter space. Additionally, connections have a pre-defined sign assigned at random; when the optimizer would normally flip the sign, the weight is set to 0 instead and new weights are activated at random.
+
+Sparse Evolutionary Training (SET) (Mocanu et al., 2018) proposed a simpler scheme where weights are pruned according to the standard magnitude criterion used in pruning and are added back at random. The method is simple and achieves reasonable performance in practice. Dynamic Sparse Reparameterization (DSR) (Mostafa & Wang, 2019) introduced the idea of allowing the parameter budget to shift between different layers of the model, allowing for non-uniform sparsity. This allows the model to distribute parameters where they are most effective. Unfortunately, the models under consideration are mostly convolutional networks, so the result of this parameter reallocation (which is to decrease the sparsity of early layers and increase the sparsity of later layers) has the overall effect of increasing the FLOP count because the spatial size is largest at the beginning. Sparse Networks from Scratch (SNFS) (Dettmers & Zettlemoyer, 2019) introduces the idea of using the momentum of each parameter as the criterion to be used for growing weights and demonstrates it leads to an improvement in test accuracy. Like DSR, they allow the sparsity of each layer to change and focus on a constant parameter, not FLOP, budget. Importantly, the method requires computing gradients and updating the momentum for every parameter in the model, even those that are zero, at every iteration. This can result in a significant amount of overall computation. Additionally, depending on the model and training setup, the required storage for the full momentum tensor could be prohibitive. Single-Shot Network Pruning (SNIP) (Lee et al., 2019) attempts to find an initial mask with one-shot pruning and uses the saliency score of parameters to decide which parameters to keep. After pruning training proceeds with this static sparse network. Properties of the different sparse training techniques are summarized in Table 1.
+
+Table 1: Comparison of different sparse training techniques. Drop and Grow columns correspond to the strategies used during the mask update. Selectable $F L O P s$ is possible if the cost of training the model is fixed at the beginning of training.   
+
+<table><tr><td>Method</td><td>Drop</td><td>Grow</td><td>Selectable FLOPs</td><td>Space &amp;FLOPs X</td></tr><tr><td>SNIP</td><td>min(10 * VL(0)I)</td><td>none</td><td>yes</td><td>sparse</td></tr><tr><td>DeepR</td><td>stochastic</td><td>random</td><td>yes</td><td>sparse</td></tr><tr><td>SET</td><td>min(l0))</td><td>random</td><td>yes</td><td>sparse</td></tr><tr><td>DSR</td><td>min(le)</td><td>random</td><td>no</td><td>sparse</td></tr><tr><td>SNFS</td><td>min(l0))</td><td>momentum</td><td>no</td><td>dense</td></tr><tr><td>RigL (ours)</td><td>min(l0)</td><td>gradient</td><td>yes</td><td>sparse</td></tr></table>
+
+There has also been a line of work investigating the Lottery Ticket Hypothesis (Frankle & Carbin, 2019). Frankle et al. (2019) showed that the formulation must be weakened to apply to larger networks such as ResNet-50 (He et al., 2015). In large networks, instead of the original initialization, the values after thousands of optimization steps must be used for initialization. Zhou et al. (2019) showed that lottery tickets obtain non-random accuracies even before the training has started. Though the possibility of training sparse neural networks with a fixed sparsity mask using lottery tickets is intriguing, it remains unclear whether it is possible to generate such initializations – for both masks and parameters – de novo.
+
+# 3 RIGGING THE LOTTERY
+
+Our method, RigL, is illustrated in Figure 1. At regularly spaced intervals our method removes a fraction of connections based on weight magnitudes and activates new ones using instantaneous gradient information. After updating the connectivity, training continues with the updated network until the next update. The main parts of our algorithm, Sparsity Distribution, Update Schedule, Drop Criterion, Grow Criterion, and the various options we considered for each, are explained below. The improved performance of $R i g L$ is due to two reasons: the use of a new method for activating connections that is efficient and more effective than choosing at random, and the use of a natural extension to an existing method for distributing parameters statically among convolutional layers.
+
+(0) Notation. Given a dataset $D$ with individual inputs $x _ { i }$ and targets $y _ { i }$ , one can train a neural network to minimize the loss function $\textstyle \sum _ { i } L ( f _ { \theta } ( x _ { i } ) , y _ { i } )$ , where $f _ { \boldsymbol { \theta } } ( \boldsymbol { x } )$ is the neural network with parameters $\theta$ of length $N$ . The vector $\theta$ can be decomposed into parameters $\theta ^ { l }$ , of length $N ^ { l }$ , for each layer $l$ . A sparse network keeps only a fraction $D \in ( 0 , 1 )$ of all connections, resulting in a sparsity of $S = 1 - D$ . More precisely, denoting the sparsity of individual layers with $s ^ { l }$ , the total parameter count of the sparse neural network satisfies $\begin{array} { r } { \dot { \sum _ { l } } ( 1 - s ^ { l } ) N ^ { l } = ( 1 - \bar { S } ) * N } \end{array}$ .
+
+![](images/c4f8c255dd49151f0525882b1c8f56ef8e1197a1ff58bb53e4c69623eb1ec0b1.jpg)  
+Figure 1: Dynamic sparse training aims to change connectivity during training to help out optimzation.
+
+(1) Sparsity Distribution. There are many ways of distributing the non-zero weights across the layers while satisfying the equality above. We avoid re-allocating parameters between layers during the training process as it makes it difficult to target a specific final FLOP budget, which is important for many inference applications. We consider the following three strategies:
+
+1. Uniform: The sparsity $s ^ { l }$ of each individual layer is the same as the total sparsity $S$ . We keep the first layer dense $\boldsymbol s ^ { 0 } = 0 ,$ ), since it has negligible number of parameters.   
+2. Erdos-R ˝ enyi: ´ As introduced in Mocanu et al. (2018), $s ^ { l }$ scales with n +n nl−1∗nl , where nl denotes number of neurons at layer l. This enables the number of connections in a sparse layer to scale with the sum of the number of output and input channels.   
+3. Erdos-R ˝ enyi-Kernel (ERK): ´ This method modifies the original Erdos-R ˝ enyi formulation ´ by including the kernel dimensions in the scaling factors. In other words, the number of parameters of the sparse convolutional layers are scaled proportional to 1− nl−1+nl+wl+hlnl−1∗nl∗wl∗hl , where $w ^ { l }$ and $h ^ { l }$ are the width and the height of the $\mathbf { \nabla } _ { l } ,$ ’th convolutional kernel. Sparsity of the fully connected layers scale as in the original Erdos-R ˝ enyi formulation. Similar to Erd ´ os- ˝ Renyi, ERK allocates higher sparsities to the layers with more parameters while allocating ´ lower sparsities to the smaller ones.
+
+In all methods, the bias and batch-norm parameters are kept dense.
+
+(2) Update Schedule. The update schedule is defined by the following parameters: (1) the number of iterations between sparse connectivity updates $( \Delta T )$ , (2) the iteration at which to stop updating the sparse connectivity $( T _ { e n d } )$ , (3) the initial fraction of connections updated $( \alpha )$ and (4) a function $f _ { d e c a y }$ , invoked every $\Delta T$ iterations until $T _ { e n d }$ , possibly decaying the fraction of updated connections over time. For the latter we choose to use cosine annealing, as we find it slightly outperforms the other methods considered.
+
+$$
+f _ { d e c a y } ( t ) = \frac { \alpha } { 2 } \left( 1 + c o s \left( \frac { t \pi } { T _ { e n d } } \right) \right)
+$$
+
+Alternatives to cosine annealing like a constant schedule and inverse power annealing are studied in the Appendix F.
+
+(3) Drop criterion. Over the course of training, we drop the lowest magnitude weights according to the update schedule since they are expected to effect the training loss least. Specifically, we drop the connections given by
+
+$T o p K ( - | \theta ^ { l } | , f _ { d e c a y } ( t ) \dot { ( } 1 - s ^ { l } ) N ^ { l } ) ^ { \ddag } .$
+
+(4) Grow criterion. The novelty of our method lies in how we grow new connections. We grow the connections with highest magnitude gradients, $T o p K _ { w \notin \theta _ { a c t i v e } ^ { l } } \breve { ( | { g r a d ( \theta ^ { l } ) } | } , f _ { d e c a y } ( t ) ( 1 - s ^ { l } ) \breve { N } ^ { l } ) .$ , where $\theta _ { a c t i v e } ^ { l }$ is the set of active connections after the drop step. Newly activated connections are initialized to zero and therefore don’t effect the output of the network. However they are expected to receive gradients with high magnitudes in the next iteration and therefore reduce the loss fastest. This procedure can be applied to each layer in sequence and the dense gradients can be discarded immediately after selecting the top connections. If a layer is too large to materialize the full gradient with respect to the weights, then we can further reduce the memory requirements by performing an iterative calculation:
+
+1. Initialize the set $T K = \{ \}$ .   
+2. Materialize a subset of size M of the full gradient, which we denote $G _ { i : i + M }$ .   
+3. Update $T K$ to contain the Top- $\mathbf { \nabla } \cdot \mathbf { K }$ elements of $G _ { i : i + M }$ concatenated with $T K$ .   
+4. Repeat steps 1 through 3 until all of the gradients have been materialized. The final set $T K$   
+contains the connections we wish to grow.
+
+![](images/dfa7237fe18adc99e80cc225a8c395a896800044e15fa934ef6beb417685da6e.jpg)  
+Figure 2: (left) Performance of various dynamic sparse training methods on ImageNet-2012 classification task. We use $80 \%$ sparse ResNet-50 architecture with uniform sparsity distribution. Points at each curve correspond to the individual training runs with training multipliers from 1 to 5 (except pruning which is scaled between 0.5 and 2). We repeat training 3 times at every multiplier and report the mean accuracies. The number of FLOPs required to train a standard dense Resnet-50 along with its performance is indicated with a dashed red line. (right) Performance of RigL at different sparsity levels with extended training. Results are averaged over 3 runs.
+
+As long as $\Delta T > \frac { 1 } { 1 - s }$ the total work in calculating dense gradients is amortized and still proportional to $1 - S$ . This is in contrast to the method of Dettmers & Zettlemoyer (2019), which requires calculating and storing the full gradients at each optimization step.
+
+# 4 EMPIRICAL EVALUATION
+
+Our experiments include image classification using CNNs on the ImageNet-2012 (Russakovsky et al., 2015) and CIFAR-10 (Krizhevsky et al.) datasets and character based language modelling using RNNs with the WikiText-103 dataset (Merity et al., 2016). We use the TensorFlow Model Pruning library (Zhu & Gupta, 2018) for our pruning baselines. A Tensorflow (Abadi et al., 2015) implementation of our method along with three other baselines (SET, SNFS, SNIP) will be open sourced. When we increase the training steps by a factor $M$ , the anchor epochs of the learning rate schedule and the end iteration of the mask update schedule are also scaled by the same factor; we indicate this scaling with a subscript (e.g. ${ \mathrm { R i g L } } _ { M \times } ,$ ).
+
+# 4.1 IMAGENET-2012 DATASET
+
+In all experiments in this section, we use SGD with momentum as our optimizer. We set the momentum coefficient of the optimizer to 0.9, $L _ { 2 }$ regularization coefficient to 0.0001, and label smoothing (Szegedy et al., 2016) to 0.1. The learning rate schedule starts with a linear warm up reaching its maximum value of 1.6 at epoch 5 which is then dropped by a factor of 10 at epochs 30, 70 and 90. We train our networks with a batch size of 4096 for 32000 steps which roughly corresponds to 100 epochs of training. Our training pipeline uses standard data augmentation, which includes random flips and crops.
+
+# 4.1.1 RESNET-50
+
+Figure 2-left summarizes the performance of various methods on training an $80 \%$ sparse ResNet-50. We also train small dense networks with equivalent parameter count. All sparse networks use the constant sparsity distribution and a cosine update schedule $\mathrm { ~ \ : ( \alpha ~ ) = ~ 0 . 3 ~ }$ , $\Delta T = 1 0 0$ ). Overall, we observe that the performance of all methods improves with training time; thus, for each method we run extended training with up to $5 \times$ the training steps of the original.
+
+As noted by Gale et al. (2019), Evci et al. (2019), Frankle et al. (2019), and Mostafa & Wang (2019), training a network with fixed sparsity from scratch (Static) leads to inferior performance. Training a small dense network with the same number of parameters gets better results than Static, but fails to match the performance of dynamic sparse models. Similarly $S E T$ improves the performance over Small-Dense, however saturates around $7 5 \%$ accuracy indicating the limits of growing new connec
+
+<table><tr><td>Method</td><td>Top-1 Accuracy</td><td>FLOPs (Train)</td><td>FLOPs (Test)</td><td>Top-1 Accuracy</td><td>FLOPs (Train)</td><td>FLOPs (Test)</td></tr><tr><td>Dense</td><td>76.8±0.09</td><td>1x (3.2e18)</td><td>1x (8.2e9)</td><td colspan="3"></td></tr><tr><td>Static</td><td colspan="3">S=0.8</td><td colspan="3">S=0.9</td></tr><tr><td>SNIP</td><td>70.6±0.06 72.0±0.10</td><td>0.23x 0.23x</td><td>0.23x 0.23x</td><td>65.8±0.04 67.2±0.12</td><td>0.10x 0.10x</td><td>0.10x 0.10x</td></tr><tr><td>Small-Dense</td><td>72.1±0.12</td><td>0.20x</td><td>0.20x</td><td>68.9±0.10</td><td>0.12x</td><td>0.12x</td></tr><tr><td>SET</td><td>72.9±0.39</td><td>0.23x</td><td>0.23x</td><td>69.6±0.23</td><td>0.10x</td><td>0.10x</td></tr><tr><td>RigL</td><td>74.6±0.06</td><td>0.23x</td><td>0.23x</td><td>72.0±0.05</td><td>0.10x</td><td>0.10x</td></tr><tr><td>Small-Dense5x RigL5x</td><td>73.9±0.07</td><td>1.01x</td><td>0.20x</td><td>71.3±0.10</td><td>0.60x</td><td>0.12x</td></tr><tr><td>Static (ERK)</td><td>76.6±0.06</td><td>1.14x</td><td>0.23x</td><td>75.7±0.06</td><td>0.52x</td><td>0.10x</td></tr><tr><td>DSR*</td><td>72.1±0.04 73.3</td><td>0.42x 0.40x</td><td>0.42x 0.40x</td><td>67.7±0.12</td><td>0.24x</td><td>0.24x</td></tr><tr><td></td><td></td><td></td><td></td><td>71.6</td><td>0.30x</td><td>0.30x</td></tr><tr><td>RigL (ERK)</td><td>75.1±0.05</td><td>0.42x</td><td>0.42x</td><td>73.0±0.04</td><td>0.25x</td><td>0.24x</td></tr><tr><td>RigL5x (ERK)</td><td>77.1±0.06</td><td>2.09x</td><td>0.42x</td><td>76.4±0.05</td><td>1.23x</td><td>0.24x</td></tr><tr><td>SNFS*</td><td>74.2</td><td>n/a</td><td>n/a</td><td>72.3</td><td>n/a</td><td>n/a</td></tr><tr><td>SNFS (ERK)</td><td>75.2±0.11</td><td>0.61x</td><td>0.42x</td><td>72.9±0.06</td><td>0.50x</td><td>0.24x</td></tr><tr><td>Pruning* (Zhu)</td><td>73.2</td><td>1.00x</td><td>0.23x</td><td>70.3</td><td>1.00x</td><td>0.10x</td></tr><tr><td>Pruning* (Gale)</td><td>75.6</td><td>1.00x</td><td>0.23x</td><td>73.9</td><td>1.00x</td><td>0.10x</td></tr><tr><td>Pruning1.5× (Gale)</td><td>76.5</td><td>1.50x</td><td>0.23x</td><td></td><td>1.50x</td><td></td></tr><tr><td></td><td></td><td></td><td></td><td>75.2</td><td></td><td>0.10x</td></tr></table>
+
+Table 2: Performance and cost of sparse training methods on training $80 \%$ and $90 \%$ sparse ResNet50s. FLOPs needed for training and test are normalized with the FLOPs of a dense model (see Appendix G for details on how FLOPs are calculated). Methods with a subscript indicate a rescaled training time, whereas ‘\*’ indicates reported results. (ERK) corresponds to the sparse networks with Erdos-Renyi-Kernel sparsity distribution.˝ $\mathrm { R i g L } _ { \mathrm { 5 } \times }$ (ERK) achieves $7 7 . 1 \%$ Top-1 Accuracy using only $20 \%$ of the parameters of a dense model and $42 \%$ of its FLOPs.
+
+tions randomly. Methods that use gradient information to grow new connections (RigL and SNFS) obtain higher accuracies, but RigL achieves the highest accuracy and does so while consistently requiring fewer FLOPs than the other methods.
+
+Given that different applications or scenarios might require a limit on the number of FLOPs for inference, we investigate the performance of our method at various sparsity levels. As mentioned previously, one strength of our method is that its resource requirements are constant throughout training and we can choose the level of sparsity that fits our training and/or inference constraints. In Figure 2-right we show the performance of our method at different sparsities and compare them with the pruning results of Gale et al. (2019), which uses $1 . 5 \mathrm { x }$ training steps, relative to the original 32k iterations. To make a fair comparison with regards to FLOPs, we scale the learning schedule of all other methods by ${ 5 } \mathbf { x }$ . Note that even after extending the training, it takes less FLOPs to train sparse networks using rigL (except for the $80 \%$ sparse RigL-ERK) compared to the pruning method.
+
+RigL, our method with constant sparsity distribution, exceeds the performance of magnitude based iterative pruning in all sparsity levels while requiring less FLOPs to train. Sparse networks that use Erdos-Renyi-Kernel (ERK) ˝ sparsity distribution obtains even greater performance. For example ResNet-50 with $9 6 . 5 \%$ sparsity achieves a remarkable $7 2 . 7 5 \%$ Top-1 Accuracy, around $3 . 5 \%$ higher than the extended magnitude pruning results reported by Gale et al. (2019). As observed earlier, smaller dense models (with the same number of parameters) or sparse models with a static connectivity can not perform at a comparable level.
+
+A more fine grained comparison of sparse training methods is presented in Table 2. Methods using uniform sparsity distribution and whose FLOP/memory footprint scales directly with (1-S) are placed in the first sub-group of the table. The second sub-group includes DSR and networks with ERK sparsity distribution which require a higher number of FLOPs for inference with same parameter count. The final sub-group includes methods that require the space and the work proportional to training a dense model.
+
+![](images/8ad076c5d43264442c51ef52152072882fb12f6803ab3f5d550b0ad517566e72.jpg)  
+Figure 3: (left) RigL significantly improves the performance of Sparse MobileNets on ImageNet2012 dataset and exceeds the pruning results reported by Zhu & Gupta (2018). Performance of the dense MobileNets are indicated with red lines. (right) Performance of sparse MobileNet-v1 architectures presented with their inference FLOPs. Networks with ERK distribution get better performance with the same number of parameters but take more FLOPs to run. Training wider sparse models with RigL (Big-Sparse) yields a significant performance improvement over the dense model.
+
+<table><tr><td rowspan=1 colspan=1>S</td><td rowspan=1 colspan=1>Method</td><td rowspan=1 colspan=1>Top-1</td><td rowspan=1 colspan=1>FLOPs</td></tr><tr><td rowspan=1 colspan=1>0.75</td><td rowspan=1 colspan=1>Small-Dense5×Pruning (Zhu)RigL5×RigL5x(ERK)</td><td rowspan=1 colspan=1>66.0±0.1167.771.5±0.0671.9±0.01</td><td rowspan=1 colspan=1>0.23x0.27x0.27x0.52x</td></tr><tr><td rowspan=1 colspan=1>0.90</td><td rowspan=1 colspan=1>Small-Dense5xPruning (Zhu)RigL5×RigL5x(ERK)</td><td rowspan=1 colspan=1>57.7±0.3461.867.0±0.1768.1±0.11</td><td rowspan=1 colspan=1>0.09x0.12x0.12x0.27x</td></tr><tr><td rowspan=1 colspan=1></td><td rowspan=1 colspan=1>Dense</td><td rowspan=1 colspan=1>72.1±0.17</td><td rowspan=1 colspan=1>1x (1.1e9)</td></tr><tr><td rowspan=1 colspan=1>0.75</td><td rowspan=1 colspan=1>Big-Sparse5×Big-Sparse5x(ERK)</td><td rowspan=1 colspan=1>76.4±0.0577.0±0.08</td><td rowspan=1 colspan=1>0.98x1.91x</td></tr></table>
+
+# 4.1.2 MOBILENET
+
+MobileNet is a compact architecture that performs remarkably well in resource constrained settings. Due to its compact nature with separable convolutions it is known to be difficult to sparsify (Zhu & Gupta, 2018). In this section we apply our method to MobileNet-v1 (Howard et al., 2017) and MobileNet-v2 (Sandler et al., 2018). Due to its low parameter count we keep the first layer dense, and use ERK and Uniform sparsity distributions to sparsify the remaining layers.
+
+The performance of sparse MobileNets trained with RigL as well as the baselines are shown in Figure 3. We do extended training (5x of the original number of steps) for all runs in this section. Although MobileNets are more sensitive to sparsity compared to the ResNet-50 architecture, RigL successfully trains sparse MobileNets at high sparsities and exceeds the performance of previously reported pruning results.
+
+To demonstrate the advantages of sparse models, next, we train wider MobileNets while keeping the FLOPs and total number of parameters the same as the dense baseline using sparsity. A sparse MobileNet-v1 with width multiplier 1.98 and constant $7 5 \%$ sparsity has the same FLOPs and parameter count as the dense baseline. Training this network with RigL yields an impressive $4 . 3 \%$ absolute improvement in Top-1 Accuracy.
+
+# 4.2 CHARACTER LEVEL LANGUAGE MODELLING
+
+Most prior work has only examined sparse training on vision networks [the exception is the earliest work - Deep Rewiring (Bellec et al., 2017) which trained an LSTM (Hochreiter & Schmidhuber, 1997) on the TIMIT (Garofolo et al., 1993) dataset]. To fully understand these techniques it is important to examine different architectures on different datasets. Kalchbrenner et al. (2018) found sparse GRUs (Cho et al., 2014) to be very effective at modeling speech, however the dataset they used is not available. We choose a proxy task with similar characteristics (dataset size and vocabulary size are approximately the same) - character level language modeling on the publicly available WikiText-103 (Merity et al., 2016) dataset.
+
+Our network consists of a shared embedding with dimensionality 128, a vocabulary size of 256, a GRU with a state size of 512, a readout from the GRU state consisting of two linear layers with 256 units and 128 units respectively. We train the next step prediction task with the standard cross entropy loss, the Adam optimizer, a learning rate of $7 e - 4$ , an L2 regularization coefficient of $5 e { - 4 }$ , a sequence length of 512, a batch size of 32 and gradient absolute value clipping of values larger (in magnitude) than 10. Baseline training length is 200,000 iterations. When inducing sparsity with magnitude pruning (Zhu & Gupta, 2018), we perform pruning between iterations 50,000 and
+
+![](images/f16a2ec71b43df85e1a6ff38138e0337a2c670eede9c85bef31c854aa486fd2f.jpg)  
+Figure 4: (left) Final validation loss of various sparse training methods on character level language modelling task. Cross entropy loss is converted to bits (from nats). Performance and the training cost of a dense model is indicated with dashed red lines. (right) Test accuracies of sparse WideResNet22-2’s on CIFAR-10 task.
+
+150,000 with a pruning frequency of 1,000. We initialize sparse networks with a uniform sparsity distribution and use a cosine update schedule with $\alpha = 0 . 1$ and $\Delta T = 1 0 0$ . Unlike the previous experiments we keep updating the mask until the end of the training; we observed this performed slightly better than stopping at iteration 150,000.
+
+In Figure 4-left we report the validation loss of various solutions at the end of the training. For each method we perform extended runs to see how they scale with increasing training time. As observed before, SET performs worst than the other dynamic training methods and its performance improves only slightly with increased training time. On the other hand the performance of RigL and SNFS improves constantly with more training steps. Both of these methods falls short of matching the pruning performance.
+
+# 4.3 WIDERESNET-22-2 ON CIFAR-10
+
+In this section, we evaluate the performance of RigL on CIFAR-10 image classification benchmark. We train Wide Residual Network’s (Zagoruyko & Komodakis, 2016) with 22 layers using a width multiplier of 2 for 250 epochs (97656 steps). Learning rate starts at 0.1 and scaled down by a factor of 5 every 30,000 iterations. We use an L2 regularization coefficient of 5e-4, a batch size of 128 and a momentum coefficient of 0.9. We keep the hyper-parameters specific to RigL same as the ImageNet experiments, except the final iteration for mask updates; which is adjusted to 75000. Results with different mask update intervals can be found in Appendix H.
+
+Performance of RigL across different sparsity levels is presented in Figure 4-right. Corresponding final training losses of the trained networks can be found in Appendix H. The dense baseline obtains $9 4 . 1 \%$ test accuracy. Networks with half of the connections removed $50 \%$ sparsity) achieves roughly the same accuracy as the dense baseline. Surprisingly, some of the networks at this sparsity level generalize better than the dense baseline demonstrating the regularization aspect of using sparsity. With increased sparsity, we start to see a performance gap between the Static and Pruning solutions. Training static RigL networks longer seems to have limited effect on the final performance. On the other hand, RigL, matches the performance of pruning using only a fraction of resources needed for training a dense network.
+
+# 4.4 ANALYZING THE PERFORMANCE OF RigL
+
+In this section we study the effect of sparsity distributions, update schedules, and dynamic connections on the performance of our method. The results for SET and SNFS are similar and are discussed in Appendices B and E.
+
+Effect of Mask Initialization: Figure 5-left shows how the sparsity distribution affects the final test accuracy of sparse ResNet-50s trained with RigL. Erdos-R ˝ enyi-Kernel (ERK) performs consis- ´ tently better than the other two distributions. ERK automatically allocates more parameters to the layers with few parameters by decreasing their sparsities†. This reallocation seems to be crucial for preserving the capacity of the network at high sparsity levels where ERK outperforms other distributions by a greater margin. Though it performs better, the ERK distribution requires approximately twice as many FLOPs compared to a uniform distribution. This highlights an interesting trade-off between accuracy and computational efficiency even though both models have the same number of parameters.
+
+![](images/5800160114d0737b3e2907dc82a457c52f3b8ccbcf1ecf67ee038a009c690904.jpg)  
+Figure 5: (left) Performance of RigL at different sparsities using different sparsity masks (right) Ablation study on cosine schedule. Other methods are in the appendix.
+
+Effect of Update Schedule and Frequency: In Figure 5-right, we evaluate the performance of our method on update intervals $\Delta T \in [ 5 0 , 1 0 0 , 5 0 0 , 1 0 0 0 ]$ and initial drop fractions $\alpha \in [ 0 . 1 , 0 . 3 , 0 . 5$ ]. The best accuracies are obtained when the mask is updated every 100 iterations with an initial drop fraction of 0.3 or 0.5. Notably, even with frequent update intervals (e.g. every 1000 iterations), RigL performs above $7 3 . 5 \%$ .
+
+Effect of Dynamic connections: Frankle et al. (2019) and Mostafa & Wang (2019) observed that static sparse training converges to a solution with a higher loss than dynamic sparse training. In Figure 6-left we examine the loss landscape lying between a solution found via static sparse training and a solution found via pruning to understand whether former lies in a basin isolated from the latter. Performing a linear interpolation between the two reveals the expected result – high-loss barrier – demonstrating that the loss landscape is not trivially connected. However, this is only one of infinitely many paths between the two points optimization can be used to find parametric curves that connects solutions (Garipov et al., 2018; Draxler et al., 2018) subject to constraints. For example Garipov et al. (2018) showed different dense solutions lie in the same basin by finding 2nd order Bezier curves with low energy between the two solutions. Following their method, we attempt ´ to find quadratic and cubic Bezier curves between the two sparse solutions. Surprisingly, even with a ´ cubic curve, we fail to find a path without a high-loss barrier. These results suggest that static sparse training can get stuck at local minima that are isolated from improved solutions. On the other hand, when we optimize the quadratic Bezier curve across the full ´ dense space we find a near-monotonic path to the improved solution, suggesting that allowing new connections to grow lends dynamic sparse training greater flexibility in navigating the loss landscape.
+
+![](images/96e95693d928d87ad902426e986b917f4232a0de25740f3cd1cb80a5c1a6297c.jpg)  
+Figure 6: (left) Training loss evaluated at various points on interpolation curves between a magnitude pruning model (0.0) and a model trained with static sparsity (1.0). (right) Training loss of RigL and Static methods starting from the static sparse solution, and their final accuracies.
+
+In Figure 6-right we train RigL starting from the sub-optimal solution found by static sparse training, demonstrating that it is able to escape the local minimum, whereas re-training with static sparse training cannot. RigL first removes connections with the smallest magnitudes since removing these connections have been shown to have a minimal effect on the loss (Han et al., 2015; Evci, 2018). Next, it activates connections with the high gradients, since these connections are expected to decrease the loss fastest. We hypothesize in Appendix A that RigL escapes bad critical points by replacing saddle directions with high gradient dimensions.
+
+# 5 DISCUSSION & CONCLUSION
+
+In this work we introduced ‘Rigged Lottery’ or RigL, an algorithm for training sparse neural networks efficiently. For a given computational budget RigL achieves higher accuracies than existing dense-to-sparse and sparse-to-sparse training algorithms. RigL is useful in three different scenarios: (1) To improve the accuracy of sparse models intended for deployment; (2) To improve the accuracy of large sparse models which can only be trained for a limited number of iterations; and (3) Combined with sparse primitives to enable training of extremely large sparse models which otherwise would not be possible.
+
+The third scenario is unexplored due to the lack of hardware and software support for sparsity. Nonetheless, work continues to improve the performance of sparse networks on current hardware (Hong et al., 2019; Merrill & Garland, 2016), and new types of hardware accelerators will have better support for parameter sparsity (Wang et al., 2018; Mike Ashby, 2019; Liu et al., 2018; Han et al., 2016a; Chen et al., 2019). RigL provides the tools to take advantage of, and motivation for, such advances.
+
+# REFERENCES
+
+Mart´ın Abadi, Ashish Agarwal, Paul Barham, Eugene Brevdo, Zhifeng Chen, Craig Citro, Greg S. Corrado, Andy Davis, Jeffrey Dean, Matthieu Devin, Sanjay Ghemawat, Ian Goodfellow, Andrew Harp, Geoffrey Irving, Michael Isard, Yangqing Jia, Rafal Jozefowicz, Lukasz Kaiser, Manjunath Kudlur, Josh Levenberg, Dan Mane, Rajat Monga, Sherry Moore, Derek Murray, Chris Olah, ´ Mike Schuster, Jonathon Shlens, Benoit Steiner, Ilya Sutskever, Kunal Talwar, Paul Tucker, Vincent Vanhoucke, Vijay Vasudevan, Fernanda Viegas, Oriol Vinyals, Pete Warden, Martin Watten- ´ berg, Martin Wicke, Yuan Yu, and Xiaoqiang Zheng. TensorFlow: Large-scale machine learning on heterogeneous systems, 2015. URL http://tensorflow.org/.
+
+Guillaume Bellec, David Kappel, Wolfgang Maass, and Robert A. Legenstein. Deep rewiring: Training very sparse deep networks. CoRR, abs/1711.05136, 2017. URL http://arxiv. org/abs/1711.05136.
+
+Y. Chen, T. Yang, J. Emer, and V. Sze. Eyeriss v2: A flexible accelerator for emerging deep neural networks on mobile devices. IEEE Journal on Emerging and Selected Topics in Circuits and Systems, 9(2):292–308, June 2019. doi: 10.1109/JETCAS.2019.2910232.
+
+Kyunghyun Cho, B van Merrienboer, Caglar Gulcehre, F Bougares, H Schwenk, and Yoshua Bengio. Learning phrase representations using rnn encoder-decoder for statistical machine translation. In Conference on Empirical Methods in Natural Language Processing (EMNLP 2014), 2014.
+
+Diederik P. Kingma Christos Louizos, Max Welling. Learning sparse neural networks through $l _ { 0 }$ regularization. In International Conference on Learning Representations, 2018.
+
+Tim Dettmers and Luke Zettlemoyer. Sparse networks from scratch: Faster training without losing performance. ArXiv, 2019. URL http://arxiv.org/abs/1907.04840.
+
+Felix Draxler, Kambis Veschgini, Manfred Salmhofer, and Fred A Hamprecht. Essentially No Barriers in Neural Network Energy Landscape. In International Conference on Machine Learning, 2018.
+
+Utku Evci. Detecting dead weights and units in neural networks. CoRR, 2018. URL http: //arxiv.org/abs/1806.06068.
+
+Utku Evci, Fabian Pedregosa, Aidan N. Gomez, and Erich Elsen. The difficulty of training sparse neural networks. ArXiv, 2019. URL http://arxiv.org/abs/1906.10732.
+
+Jonathan Frankle and Michael Carbin. The lottery ticket hypothesis: Finding sparse, trainable neural networks. In 7th International Conference on Learning Representations, ICLR 2019, New Orleans, LA, USA, May 6-9, 2019, 2019. URL https://openreview.net/forum?id $=$ rJl-b3RcF7.
+
+Jonathan Frankle, Gintare Karolina Dziugaite, Daniel M. Roy, and Michael Carbin. The lottery ticket hypothesis at scale. ArXiv, 2019. URL http://arxiv.org/abs/1903.01611.
+
+Trevor Gale, Erich Elsen, and Sara Hooker. The state of sparsity in deep neural networks. CoRR, abs/1902.09574, 2019. URL http://arxiv.org/abs/1902.09574.
+
+Timur Garipov, Pavel Izmailov, Dmitrii Podoprikhin, Dmitry P Vetrov, and Andrew Gordon Wilson. Loss surfaces, mode connectivity, and fast ensembling of dnns. In Advances in Neural Information Processing Systems, 2018.
+
+J. S. Garofolo, L. F. Lamel, W. M. Fisher, J. G. Fiscus, D. S. Pallett, and N. L. Dahlgren. Darpa timit acoustic phonetic continuous speech corpus cdrom, 1993.
+
+Yiwen Guo, Anbang Yao, and Yurong Chen. Dynamic network surgery for efficient DNNs. CoRR, abs/1608.04493, 2016. URL http://arxiv.org/abs/1608.04493.
+
+Song Han, Jeff Pool, John Tran, and William Dally. Learning both weights and connections for efficient neural network. In Advances in neural information processing systems, 2015.
+
+Song Han, Xingyu Liu, Huizi Mao, Jing Pu, Ardavan Pedram, Mark A. Horowitz, and William J. Dally. EIE: Efficient Inference Engine on compressed deep neural network. In Proceedings of the 43rd International Symposium on Computer Architecture, 2016a.
+
+Song Han, Huizi Mao, and William J. Dally. Deep compression: Compressing deep neural network with pruning, trained quantization and huffman coding. In 4th International Conference on Learning Representations, ICLR 2016, San Juan, Puerto Rico, May 2-4, 2016, Conference Track Proceedings, 2016b. URL http://arxiv.org/abs/1510.00149.
+
+B. Hassibi and D. Stork. Second order derivatives for network pruning: Optimal Brain Surgeon. 1993.
+
+Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Delving deep into rectifiers: Surpassing human-level performance on imagenet classification. In Proceedings of the 2015 IEEE International Conference on Computer Vision (ICCV), 2015.
+
+Sepp Hochreiter and Jurgen Schmidhuber. Long short-term memory. ¨ Neural Comput., 9(8):1735– 1780, November 1997. ISSN 0899-7667. doi: 10.1162/neco.1997.9.8.1735. URL http://dx. doi.org/10.1162/neco.1997.9.8.1735.
+
+Changwan Hong, Aravind Sukumaran-Rajam, Israt Nisa, Kunal Singh, and P. Sadayappan. Adaptive sparse tiling for sparse matrix multiplication. In Proceedings of the 24th Symposium on Principles and Practice of Parallel Programming, PPoPP ’19, pp. 300–314, New York, NY, USA, 2019. ACM. ISBN 978-1-4503-6225-2. doi: 10.1145/3293883.3295712. URL http://doi.acm. org/10.1145/3293883.3295712.
+
+Andrew G. Howard, Menglong Zhu, Bo Chen, Dmitry Kalenichenko, Weijun Wang, Tobias Weyand, Marco Andreetto, and Hartwig Adam. Mobilenets: Efficient convolutional neural networks for mobile vision applications. CoRR, abs/1704.04861, 2017. URL http://arxiv.org/abs/ 1704.04861.
+
+Nal Kalchbrenner, Erich Elsen, Karen Simonyan, Seb Noury, Norman Casagrande, Edward Lockhart, Florian Stimberg, Aaron Oord, Sander Dieleman, and Koray Kavukcuoglu. Efficient neural audio synthesis. In International Conference on Machine Learning (ICML), 2018.
+
+Alex Krizhevsky, Vinod Nair, and Geoffrey Hinton. Cifar-10 (canadian institute for advanced research). URL http://www.cs.toronto.edu/˜kriz/cifar.html.
+
+Yann LeCun, John S. Denker, and Sara A. Solla. Optimal Brain Damage. In Advances in Neural Information Processing Systems, 1990.
+
+Namhoon Lee, Thalaiyasingam Ajanthan, and Philip H. S. Torr. SNIP: Single-shot Network Pruning based on Connection Sensitivity. In International Conference on Learning Representations (ICLR), 2019, 2019.
+
+Hao Li, Asim Kadav, Igor Durdanovic, Hanan Samet, and Hans Peter Graf. Pruning filters for efficient convnets. In International Conference on Learning Representations, 2016.
+
+Chen Liu, Guillaume Bellec, Bernhard Vogginger, David Kappel, Johannes Partzsch, Felix Neumaerker, Sebastian Hoppner, Wolfgang Maass, Stephen B. Furber, Robert A. Legenstein, and ¨ Christian Mayr. Memory-efficient deep learning on a spinnaker 2 prototype. In Front. Neurosci., 2018.
+
+Christos Louizos, Karen Ullrich, and Max Welling. Bayesian compression for deep learning. In Advances in Neural Information Processing Systems, 2017.
+
+Stephen Merity, Caiming Xiong, James Bradbury, and Richard Socher. Pointer sentinel mixture models. ArXiv, 2016. URL http://arxiv.org/abs/1609.07843.
+
+Duane Merrill and Michael Garland. Merge-based sparse matrix-vector multiplication (spmv) using the csr storage format. In Proceedings of the 21st ACM SIGPLAN Symposium on Principles and Practice of Parallel Programming, PPoPP ’16, pp. 43:1–43:2, New York, NY, USA, 2016. ACM. ISBN 978-1-4503-4092-2. doi: 10.1145/2851141.2851190. URL http://doi.acm.org/ 10.1145/2851141.2851190.
+
+Peter Baldwin Martijn Bastiaan Oliver Bunting Aiken Cairncross Christopher Chalmers Liz Corrigan Sam Davis Nathan van Doorn Jon Fowler Graham Hazel Basile Henry David Page Jonny Shipton Shaun Steenkamp Mike Ashby, Christiaan Baaij. Exploiting unstructured sparsity on next-generation datacenter hardware. 2019. URL https://myrtle.ai/wp-content/ uploads/2019/06/IEEEformatMyrtle.ai_.21.06.19_b.pdf.
+
+Decebal Constantin Mocanu, Elena Mocanu, Peter Stone, Phuong H Nguyen, Madeleine Gibescu, and Antonio Liotta. Scalable training of artificial neural networks with adaptive sparse connectivity inspired by network science. Nature Communications, 2018.
+
+Dmitry Molchanov, Arsenii Ashukha, and Dmitry P. Vetrov. Variational Dropout Sparsifies Deep Neural Networks. In Proceedings of the 34th International Conference on Machine Learning, ICML 2017, Sydney, NSW, Australia, 6-11 August 2017, pp. 2498–2507, 2017.
+
+Pavlo Molchanov, Stephen Tyree, Tero Karras, Timo Aila, and Jan Kautz. Pruning Convolutional Neural Networks for Resource Efficient Transfer Learning. CoRR, abs/1611.06440, 2016.
+
+Hesham Mostafa and Xin Wang. Parameter efficient training of deep convolutional neural networks by dynamic sparse reparameterization. In Proceedings of the 36th International Conference on Machine Learning, ICML 2019, 9-15 June 2019, Long Beach, California, USA, pp. 4646–4655, 2019. URL http://proceedings.mlr.press/v97/mostafa19a.html.
+
+Sharan Narang, Greg Diamos, Shubho Sengupta, and Erich Elsen. Exploring sparsity in recurrent neural networks. In 5th International Conference on Learning Representations, ICLR 2017, Toulon, France, April 24-26, 2017, Conference Track Proceedings, 2017. URL https: //openreview.net/forum?id=BylSPv9gx.
+
+Jongsoo Park, Sheng R. Li, Wei Wen, Hai Li, Yiran Chen, and Pradeep Dubey. Holistic SparseCNN: Forging the trident of accuracy, speed, and size. CoRR, abs/1608.01409, 2016. URL http: //arxiv.org/abs/1608.01409.
+
+Olga Russakovsky, Jia Deng, Hao Su, Jonathan Krause, Sanjeev Satheesh, Sean Ma, Zhiheng Huang, Andrej Karpathy, Aditya Khosla, Michael Bernstein, Alexander C. Berg, and Li FeiFei. Imagenet large scale visual recognition challenge. International Journal of Computer Vision (IJCV), 2015.
+
+M. Sandler, A. Howard, M. Zhu, A. Zhmoginov, and L. Chen. Mobilenetv2: Inverted residuals and linear bottlenecks. In 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp. 4510–4520, June 2018. doi: 10.1109/CVPR.2018.00474.
+
+S. Srinivas, A. Subramanya, and R. V. Babu. Training sparse neural networks. In 2017 IEEE Conference on Computer Vision and Pattern Recognition Workshops (CVPRW), pp. 455–462, July 2017. doi: 10.1109/CVPRW.2017.61.
+
+Nikko Strom. Sparse Connection and Pruning in Large Dynamic Artificial Neural Networks. In ¨ EUROSPEECH, 1997.
+
+Christian Szegedy, Vincent Vanhoucke, Sergey Ioffe, Jonathon Shlens, and Zbigniew Wojna. Rethinking the inception architecture for computer vision. In Proceedings of IEEE Conference on Computer Vision and Pattern Recognition,, 2016. URL http://arxiv.org/abs/1512. 00567.
+
+Enzo Tartaglione, Skjalg Lepsøy, Attilio Fiandrotti, and Gianluca Francini. Learning sparse neural networks via sensitivity-driven regularization. In Proceedings of the 32nd International Conference on Neural Information Processing Systems, NIPS’18, pp. 3882–3892, USA, 2018. Curran Associates Inc. URL http://dl.acm.org/citation.cfm?id $\underline { { \underline { { \mathbf { \Pi } } } } } =$ 3327144.3327303.
+
+Georg Thimm and Emile Fiesler. Evaluating pruning methods. In National Chiao-Tung University, pp. 2, 1995.
+
+Peiqi Wang, Yu Ji, Chi Hong, Yongqiang Lyu, Dongsheng Wang, and Yuan Xie. Snrram: An efficient sparse neural network computation architecture based on resistive random-access memory. In Proceedings of the 55th Annual Design Automation Conference, DAC ’18, pp. 106:1–106:6, New York, NY, USA, 2018. ACM. ISBN 978-1-4503-5700-5. doi: 10.1145/3195970.3196116. URL http://doi.acm.org/10.1145/3195970.3196116.
+
+Sergey Zagoruyko and Nikos Komodakis. Wide residual networks. In BMVC, 2016.
+
+Hattie Zhou, Janice Lan, Rosanne Liu, and Jason Yosinski. Deconstructing Lottery Tickets: Zeros, Signs, and the Supermask. ArXiv, 2019.
+
+Michael Zhu and Suyog Gupta. To Prune, or Not to Prune: Exploring the Efficacy of Pruning for Model Compression. In International Conference on Learning Representations Workshop, 2018.
+
+# A EFFECT OF MASK UPDATES ON THE ENERGY LANDSCAPE
+
+To update the connectivity of our sparse network, we first need to drop a fraction $d$ of the existing connections for each layer independently to create a budget for growing new connections. Following the recipe of magnitude based pruning(Han et al., 2015), we order parameters at layer $i$ by magnitude $| \theta _ { i } |$ and drop the $N * ( 1 - S ) * d$ parameters with lowest magnitude. The effectiveness of this simple criteria can be explained through the first order Taylor approximation of the loss $L$ around the current set of parameters $\theta$ .
+
+$$
+\Delta L = L ( \theta + \Delta \theta ) - L ( \theta ) = \nabla _ { \theta } L ( \theta ) \Delta \theta + R ( | | \Delta \theta | | _ { 2 } ^ { 2 } )
+$$
+
+The main goal of dropping connections is to remove parameters with minimal impact on the neural network and therefore on its performance. Since removing the connection $\theta _ { i }$ corresponds to setting it to zero, it incurs a change of $\Delta \theta ~ = ~ - \theta _ { i }$ in that direction and a change of $\begin{array} { r l } { \Delta L _ { i } } & { { } = } \end{array}$ $- \nabla _ { \boldsymbol { \theta } _ { i } } \mathbf { \bar { L } } ( \boldsymbol { \theta } ) \boldsymbol { \theta } _ { i } + R ( \boldsymbol { \theta } _ { i } ^ { 2 } )$ in the loss, where the first term is usually defined as the saliency of a connection. Though using saliency to remove connections has been used as a criteria for removing connections (Molchanov et al., 2016), it has been shown to produce inferior results compared to magnitude based removal, especially when used to remove multiple connections at once (Evci, 2018). In contrast, picking the lowest magnitude connections ensures a small remainder term in addition to a low saliency, limiting the damage we make when we drop connections. Additionally, we note that connections with small magnitude can only remain small if the gradient is also small, meaning that the saliency is likely small when the parameter itself is small. Therefore we argue that the connections removed by RigL are likely to be saddle directions of the energy landscape.
+
+After the removal of insignificant connections, we enable new connections that have the highest expected gradients. Since we initialize these new connections to zero, they are guaranteed to have high gradients in the proceeding iteration and therefore to reduce the loss quickly. By definition a direction with high magnitude gradient is not a saddle direction. Combining this observation with the previous ( $R i g L$ is likely to remove saddle directions) and the results in Section 4.4 we suggest that RigL improves the energy landscape of the optimization by replacing saddle directions with the ones with high gradient. This helps the optimization procedure to escape bad critical points and find solutions with higher quality.
+
+# B EFFECT OF SPARSITY DISTRIBUTION ON OTHER METHODS
+
+In Figure 7-left we show the effect of sparsity distribution choice on 4 different sparse training methods. ERK distribution performs better than other distributions for each training method.
+
+# C EFFECT OF MOMENTUM COEFFICIENT FOR SNFS
+
+In Figure 7 right we show the effect of the momentum coefficient on the performance of SNFS. Our results shows that using a coefficient of 0.99 brings the best performance. On the other hand using the most recent gradient only (coefficient of 0) performs as good as using a coefficient of 0.9. This result might be due to the large batch size we are using (4096), but it still motivates using RigL and instantaneous gradient information only when needed, instead of accumulating them.
+
+# D EXISTENCE OF LOTTERY TICKETS
+
+We perform the following experiment to see whether Lottery Tickets exist in our setting. We take the sparse network found by RigL and restart training using original initialization, both with RigL and with fixed topology as in the original Lottery Ticket Hypothesis. Results in table 3 demonstrate that training with a fixed topology is significantly worse than training with RigL and that RigL does not benefit from starting again with the final topology and the original initialization - training for twice as long instead of rewiring is more effective. In short, there are no special tickets, with RigL all tickets seems to win.
+
+![](images/f433bd443059bd764822e9e7a3060af7bbc37812c757ca9ad010a470abc0a839.jpg)  
+Figure 7: (left) Effect of sparsity distribution choice on sparse training methods at different sparsity levels. We average over 3 runs and report the standard deviations for each. (right) Effect of momentum value on the performance of SNFS algorithm. Setting the momentum coefficient of the SNFS algorithm to 0 seems to perform best, suggesting the accumulated values are not important.
+
+Table 3: Effect of lottery ticket initialization on the final performance. There are no special tickets and dynamic connectivity provided by RigL is critical for good performance.   
+
+<table><tr><td>Initialization</td><td>Training Method</td><td>Test Accuracy</td><td>Training FLOPs</td></tr><tr><td>Lottery</td><td>Static</td><td>70.82±0.07</td><td>0.46x</td></tr><tr><td>Lottery</td><td>RigL</td><td>73.93±0.09</td><td>0.46x</td></tr><tr><td>Random</td><td>RigL</td><td>74.55±0.06</td><td>0.23x</td></tr><tr><td>Random</td><td>RigL2x</td><td>76.06±0.09</td><td>0.46x</td></tr></table>
+
+# E EFFECT OF UPDATE SCHEDULES ON OTHER DYNAMIC SPARSE METHODS
+
+In Figure 8 we repeat the hyper-parameter sweep done for RigL in Figure 5-right, using SET and SNFS. Cosine schedule with $\Delta T = 5 0$ and $\alpha = 0 . 1$ seems to work best across all methods. An interesting observation is that higher drop fractions $( \alpha )$ seem to work better with longer intervals $\Delta T$ . For example, SET with $\Delta T = 1 0 0 0$ seems to work best with $\alpha = 0 . 5$ .
+
+# F ALTERNATIVE UPDATE SCHEDULES
+
+In Figure 9, we share the performance of two alternative annealing functions:
+
+1. Constant: $f _ { d e c a y } ( t ) = \alpha$ .   
+2. Inverse Power: The fraction of weights updated decreases similarly to the schedule used in Zhu & Gupta (2018) for iterative pruning: $\begin{array} { r } { f _ { d e c a y } ( t ) = \alpha ( 1 - \frac { t } { T _ { e n d } } ) ^ { k } } \end{array}$ )k. In our experiments we tried $k = 1$ which is the linear decay and their default $k = 3$ .
+
+Constant seems to perform well with low initial drop fractions like $\alpha = 0 . 1$ , but it starts to perform worse with increasing $\alpha$ . Inverse Power for ${ \bf k } = 3$ and ${ \mathrm { k } } { = } 1$ (Linear) seems to perform similarly for low $\alpha$ values. However the performance drops noticeably for ${ \bf k } = 3$ when we increase the update interval. As reported by Dettmers & Zettlemoyer (2019) linear $( \mathrm { k } { = } 1$ ) seems to provide similar results as the cosine schedule.
+
+# G CALCULATING FLOPS OF MODELS AND METHODS
+
+In order to calculate FLOPs needed for a single forward pass of a sparse model, we count the total number of multiplications and additions layer by layer for a given layer sparsity $s ^ { l }$ . The total FLOPs is then obtained by summing up all of these multiply and adds.
+
+Different sparsity distributions require different number of FLOPs to compute a single prediction. For example Erdos-Renyi-Kernel ˝ distributions usually cause earlier layers to be less sparse than the later layers (see Appendix I). The inputs of earlier layers have greater spatial dimensions, so a convolutional kernel that works on such inputs will require more FLOPs to compute the output features compared to later layers. Thus, having earlier layers which are less sparse results in a higher total number of FLOPs required by a model.
+
+![](images/b3e467a52c2f5b11283887318ea99ba50224d604ca1a45394151b61438546ee9.jpg)  
+Figure 8: Cosine update schedule hyper-parameter sweep done using dynamic sparse training methods SET (left) and SNFS (right).
+
+Training a neural network consists of 2 main steps:
+
+1. forward pass: Calculating the loss of the current set of parameters on a given batch of data. During this process layer activations are calculated in sequence using the previous activations and the parameters of the layer. Activation of layers are stored in memory for the backward pass.   
+2. backward pass: Using the loss value as the initial error signal, we back-propagate the error signal while calculating the gradient of parameters. During the backward pass each layer calculates 2 quantities: the gradient of the activations of the previous layer and the gradient of its parameters. Therefore in our calculations we count backward passes as two times the computational expense of the forward pass. We omit the FLOPs needed for batch normalization and cross entropy.
+
+Dynamic sparse training methods require some extra FLOPs to update the connectivity of the neural network. We omit FLOPs needed for dropping the lowest magnitude connections in our calculations. For a given dense architecture with FLOPs $f _ { D }$ and a sparse version with FLOPs $f _ { S }$ , the total FLOPs required to calculate the gradient on a single sample is computed as follows:
+
+• Static Sparse and Dense. Scales with $3 * f _ { S }$ and $3 * f _ { D }$ FLOPs, respectively.   
+• Snip. We omit the initial dense gradient calculation since it is negligible, which means Snip scales in the same way as Static methods: $3 * f _ { S }$ FLOPs.   
+• SET. We omit the extra FLOPs needed for growing random connections, since this operation can be done on chip efficiently. Therefore, the total FLOPs for SET scales with $3 * f _ { S }$ . SNFS. Forward pass and back-propagating the error signal needs $2 * f _ { S }$ FLOPs. However, the dense gradient needs to be calculated at every iteration. Thus, the total number of FLOPs scales with $2 * f _ { S } + f _ { D }$ . RigL. Iterations with no connection updates need $3 * f _ { S }$ FLOPs. However, at every $\Delta T$ iteration we need to calculate the dense gradients. This results in the average FLOPs for RigL given by $\frac { ( 3 * f _ { S } * \Delta T + 2 * f _ { S } + f _ { D } ) } { ( \Delta T + 1 ) }$ .
+
+# H ADDITIONAL PLOTS AND EXPERIMENTS FOR CIFAR-10
+
+In Figure 10-left, we plot the final training loss of experiments presented in Section 4.3 to investigate the generalization properties of the algorithms considered. Poor performance of Static reflects itself in training loss clearly across all sparsity levels. RigL achieves similar final loss as the pruning, despite having around half percent less accuracy. Training longer with RigL decreases the final loss further and the test accuracies start matching pruning (see Figure 4-right) performance. These results show that RigL improves the optimization as promised, however generalizes slightly worse than pruning.
+
+![](images/bf63372d44ae944bab2fb94af7feac16fac25a9bc8f8d3818b33dd4260972182.jpg)  
+Figure 9: Using other update schedules with RigL: (left) Constant (middle) Exponential $\left( \mathrm { k } \mathrm { = } 3 \right)$ and (right) Linear
+
+![](images/e207781847e04baa8da603b4cc537a5073cb27d1d766bb864de9512780d702bc.jpg)  
+Figure 10: Final training loss of sparse models (left) and performance of $R i g L$ at different mask update intervals (right).
+
+In Figure 10-right, we sweep mask update interval $\Delta T$ and plot the final test accuracies. We fix initial drop fraction $\alpha$ to 0.3 and evaluate two different sparsity distributions: Uniform and $E R K$ . Both curves follow a similar pattern as in Imagenet-2012 sweeps (see Figure 8) and best results are obtained when $\Delta T = 1 0 0$ .
+
+# I SPARSITY OF INDIVIDUAL LAYERS FOR SPARSE RESNET-50
+
+Sparsity of ResNet-50 layers given by the Erdos-R ˝ enyi-Kernel sparsity distribution plotted in Figure ´ 11.
+
+# J BUGS DISCOVERED DURING EXPERIMENTS
+
+Our initial implementations contained some subtle bugs, which while not affecting the general conclusion that RigL is more effective than other techniques, did result in lower accuracy for all sparse training techniques. We detail these issues here with the hope that others may learn from our mistakes.
+
+1. Random operations on multiple replicas. We use data parallelism to split a mini-batch among multiple replicas. Each replica independently calculates the gradients using a different sub-mini-batch of data. The gradients are aggregated using an ALL-REDUCE operation before the optimizer update. Our implementation of SET, SNFS and RigL depended on each replica independently choosing to drop and grow the same connections. However, due to the nature of random operations in Tensorflow, this did not happen. Instead, different replicas diverged after the first drop/grow step. This was most pronounced in SET where each replica chose at random and much less so for SNFS and RigL where randomness is only needed to break ties. If left unchecked this might be expected to be catastrophic, but due to the behavior of Estimators and/or TF-replicator, the values on the first replica are broadcast to the others periodically (every approximately 1000 steps in our case).
+
+We fixed this bug by using stateless random operations. As a result the performance of SET improved slightly $0 . 1 \%$ higher on Table 2).
+
+2. Synchronization between replicas. RigL and SNFS depend on calculating dense gradients with respect to the masked parameters. However, as explained above, in the multiple replica setting these gradients need to be aggregated. Normally this aggregation is automatically done by the optimizer, but in our case, this does not happen (only the gradients with respect to the unmasked parameters are aggregated automatically). This bug affected SNFS and RigL, but not SET since SET does not rely on the gradients to grow connections. Again, the synchronization of the parameters from the first replica every approximately 1000 steps masked this bug.
+
+We fixed this bug by explicitly calling ALL-REDUCE on the gradients with respect to the masked parameters. With this fix, the performance of RigL and SNFS improved significantly, particularly for default training lengths (around $0 . 5 – 1 \%$ improvement).
+
+![](images/0b1f9b452aac571c6f076bd13ec4d1e1e1f13be162984229dcfda1d2c8d58426.jpg)  
+Figure 11: Sparsities of individual layers of the ResNet-50.
+
+3. SNIP Experiments. Our first implementation of SNIP used the gradient magnitudes to decide which connections to keep causing its performance to be worse than static. Upon our discussions with the authors of SNIP, we realized that the correct metric is the saliency (gradient times parameter magnitude). With this correction SNIP performance improved dramatically to better than random (Static) even at Resnet-50/ImageNet scale. It is surprising that picking connections with the highest gradient magnitudes can be so detrimental to training (it resulted in much worse than random performance).
\ No newline at end of file
diff --git a/parse/train/ryg7vA4tPB/ryg7vA4tPB_content_list.json b/parse/train/ryg7vA4tPB/ryg7vA4tPB_content_list.json
new file mode 100644
index 0000000000000000000000000000000000000000..47e1bbd9b2d7bb6bdda475ba265b73a55c785401
--- /dev/null
+++ b/parse/train/ryg7vA4tPB/ryg7vA4tPB_content_list.json
@@ -0,0 +1,1993 @@
+[
+    {
+        "type": "text",
+        "text": "RIGGING THE LOTTERY: MAKING ALL TICKETS WINNERS ",
+        "text_level": 1,
+        "bbox": [
+            174,
+            99,
+            578,
+            146
+        ],
+        "page_idx": 0
+    },
+    {
+        "type": "text",
+        "text": "Anonymous authors Paper under double-blind review ",
+        "bbox": [
+            183,
+            170,
+            400,
+            198
+        ],
+        "page_idx": 0
+    },
+    {
+        "type": "text",
+        "text": "ABSTRACT ",
+        "text_level": 1,
+        "bbox": [
+            454,
+            234,
+            544,
+            251
+        ],
+        "page_idx": 0
+    },
+    {
+        "type": "text",
+        "text": "Sparse neural networks have been shown to be more parameter and compute efficient compared to dense networks and in some cases they are even successfully used to decrease wall clock inference times. There is a large body of work on training dense networks to yield sparse networks for inference (Molchanov et al., 2017; Zhu & Gupta, 2018; Louizos et al., 2017; Li et al., 2016; Guo et al., 2016). This limits the size of the largest trainable sparse model to that of the largest trainable dense model. In this paper we introduce a method to train sparse neural networks with a fixed parameter count and a fixed computational cost throughout training, without sacrificing accuracy relative to existing dense-to-sparse training methods. Our method updates the topology of the network during training by using parameter magnitudes and infrequent gradient calculations. We show that this approach requires fewer floating-point operations (FLOPs) to achieve a given level of accuracy compared to prior techniques. We demonstrate state-of-the-art sparse training results with ResNet-50, MobileNet v1 and MobileNet v2 on the ImageNet-2012 dataset, WideResNets on the CIFAR-10 dataset and RNNs on the WikiText-103 dataset. Finally, we provide some insights into why allowing the topology to change during the optimization can overcome local minima encountered when the topology remains static. ",
+        "bbox": [
+            233,
+            266,
+            764,
+            517
+        ],
+        "page_idx": 0
+    },
+    {
+        "type": "text",
+        "text": "1 INTRODUCTION ",
+        "text_level": 1,
+        "bbox": [
+            176,
+            545,
+            336,
+            560
+        ],
+        "page_idx": 0
+    },
+    {
+        "type": "text",
+        "text": "The parameter and floating point operation (FLOP) efficiency of sparse neural networks is now well demonstrated on a variety of problems (Han et al., 2015; Srinivas et al., 2017). Some work has even shown inference time speedups are possible on Recurrent Neural Networks (RNNs) (Kalchbrenner et al., 2018) and Convolutional Neural Networks (ConvNets) (Park et al., 2016). Currently, the most accurate sparse models are obtained with techniques that require, at a minimum, the cost of training a dense model in terms of memory and FLOPs (Zhu & Gupta, 2018; Guo et al., 2016), and sometimes significantly more (Molchanov et al., 2017). This paradigm has two main limitations: ",
+        "bbox": [
+            174,
+            577,
+            825,
+            674
+        ],
+        "page_idx": 0
+    },
+    {
+        "type": "text",
+        "text": "1. The maximum size of sparse models is limited to the largest dense model that can be trained. Even if sparse models are more parameter efficient, we can’t use pruning to train models that are larger and more accurate than the largest possible dense models.   \n2. It is inefficient. Large amounts of computation must be performed for parameters that are zero valued or that will be zero during inference. ",
+        "bbox": [
+            210,
+            688,
+            825,
+            763
+        ],
+        "page_idx": 0
+    },
+    {
+        "type": "text",
+        "text": "Additionally, it remains unknown if the performance of the current best pruning algorithms are an upper bound on the quality of sparse models. Gale et al. (2019) found that three different dense-tosparse training algorithms all achieve about the same sparsity / accuracy trade-off. However, this is far from conclusive proof that no better performance is possible. In this work we show the surprising result that dynamic sparse training, which includes the method we introduce below, can find more accurate models than the current best approaches to pruning initially dense networks. Importantly, our method does not change the FLOPs required to execute the model during training, allowing one to decide on a specific inference cost prior to training. ",
+        "bbox": [
+            174,
+            777,
+            825,
+            888
+        ],
+        "page_idx": 0
+    },
+    {
+        "type": "text",
+        "text": "The Lottery Ticket Hypothesis (Frankle & Carbin, 2019) hypothesized that if we can find a sparse neural network with iterative pruning, then we can train that sparse network from scratch, to the same level of accuracy, by starting from the original initial conditions. In this paper we introduce a new method for training sparse models without the need of a “lucky” initialization; for this reason, we call our method “The Rigged Lottery” or $R i g L ^ { * }$ . We show that this method is: ",
+        "bbox": [
+            174,
+            895,
+            823,
+            924
+        ],
+        "page_idx": 0
+    },
+    {
+        "type": "text",
+        "text": "",
+        "bbox": [
+            176,
+            103,
+            823,
+            146
+        ],
+        "page_idx": 1
+    },
+    {
+        "type": "text",
+        "text": "• Memory efficient: It requires memory only proportional to the size of the sparse model. It never requires storing quantities that are the size of the dense model. This is in contrast to Dettmers & Zettlemoyer (2019) which requires storing the momentum for all parameters, even those that are zero valued.   \n• Computationally efficient: The amount of computation required to train the model is proportional to the number of nonzero parameters in the model.   \n• Accurate: The performance achieved by the method matches and sometimes exceeds the performance of pruning based approaches. ",
+        "bbox": [
+            215,
+            160,
+            823,
+            282
+        ],
+        "page_idx": 1
+    },
+    {
+        "type": "text",
+        "text": "Our method works by infrequently using instantaneous gradient information to inform a re-wiring of the network. We show that this allows the optimization to escape local minima where it would otherwise become trapped if the sparsity pattern were to remain static. Crucially, as long as the full gradient information is needed less than every $\\frac { 1 } { 1 - s p a r s i t y }$ iterations, then the overall work remains proportional to the model sparsity. ",
+        "bbox": [
+            174,
+            296,
+            825,
+            367
+        ],
+        "page_idx": 1
+    },
+    {
+        "type": "text",
+        "text": "2 RELATED WORK ",
+        "text_level": 1,
+        "bbox": [
+            176,
+            387,
+            344,
+            404
+        ],
+        "page_idx": 1
+    },
+    {
+        "type": "text",
+        "text": "Research on finding sparse neural networks dates back decades; for example, at least to Thimm & Fiesler (1995) who concluded that pruning weights based on magnitude was a simple and powerful technique. Strom (1997) later introduced the idea of retraining the previously pruned network to ¨ increase accuracy. Han et al. (2016b) went further and introduced multiple rounds of magnitude pruning and retraining. This is, however, relatively inefficient, requiring ten rounds of retraining when removing $2 0 \\%$ of the connections to reach a final sparsity of $9 0 \\%$ . To overcome this problem, Narang et al. (2017) introduced gradual pruning, where connections are slowly removed over the course of a single round of training. Zhu & Gupta (2018) refined the technique to minimize the amount of hyper-parameter selection required. ",
+        "bbox": [
+            174,
+            420,
+            825,
+            545
+        ],
+        "page_idx": 1
+    },
+    {
+        "type": "text",
+        "text": "A diversity of approaches not based on magnitude based pruning have also been proposed. LeCun et al. (1990) and Hassibi & Stork (1993) are some early examples, but impractical for modern neural networks as they use information from the Hessian to prune a trained network. More recent work includes $L _ { 0 }$ Regularization (Christos Louizos, 2018), Variational Dropout (Molchanov et al., 2017), Dynamic Network Surgery (Guo et al., 2016) and Sensitivity Driven Regularization (Tartaglione et al., 2018). Gale et al. (2019) examined magnitude pruning, $L _ { 0 }$ Regularization and Variational Dropout and concluded that they all achieve about the same accuracy versus sparsity trade-off on ResNet-50 and Transformer architectures. ",
+        "bbox": [
+            174,
+            553,
+            825,
+            662
+        ],
+        "page_idx": 1
+    },
+    {
+        "type": "text",
+        "text": "Training techniques that allow for sparsity throughout the entire training process were, to our knowledge, first introduced in Deep Rewiring (DeepR) (Bellec et al., 2017). In DeepR, the standard Stochastic Gradient Descent (SGD) optimizer is augmented with a random walk in parameter space. Additionally, connections have a pre-defined sign assigned at random; when the optimizer would normally flip the sign, the weight is set to 0 instead and new weights are activated at random. ",
+        "bbox": [
+            174,
+            670,
+            825,
+            741
+        ],
+        "page_idx": 1
+    },
+    {
+        "type": "text",
+        "text": "Sparse Evolutionary Training (SET) (Mocanu et al., 2018) proposed a simpler scheme where weights are pruned according to the standard magnitude criterion used in pruning and are added back at random. The method is simple and achieves reasonable performance in practice. Dynamic Sparse Reparameterization (DSR) (Mostafa & Wang, 2019) introduced the idea of allowing the parameter budget to shift between different layers of the model, allowing for non-uniform sparsity. This allows the model to distribute parameters where they are most effective. Unfortunately, the models under consideration are mostly convolutional networks, so the result of this parameter reallocation (which is to decrease the sparsity of early layers and increase the sparsity of later layers) has the overall effect of increasing the FLOP count because the spatial size is largest at the beginning. Sparse Networks from Scratch (SNFS) (Dettmers & Zettlemoyer, 2019) introduces the idea of using the momentum of each parameter as the criterion to be used for growing weights and demonstrates it leads to an improvement in test accuracy. Like DSR, they allow the sparsity of each layer to change and focus on a constant parameter, not FLOP, budget. Importantly, the method requires computing gradients and updating the momentum for every parameter in the model, even those that are zero, at every iteration. This can result in a significant amount of overall computation. Additionally, depending on the model and training setup, the required storage for the full momentum tensor could be prohibitive. Single-Shot Network Pruning (SNIP) (Lee et al., 2019) attempts to find an initial mask with one-shot pruning and uses the saliency score of parameters to decide which parameters to keep. After pruning training proceeds with this static sparse network. Properties of the different sparse training techniques are summarized in Table 1. ",
+        "bbox": [
+            174,
+            747,
+            825,
+            900
+        ],
+        "page_idx": 1
+    },
+    {
+        "type": "table",
+        "img_path": "images/2dd036930c9611b0039842e97192636016f15c49e7d59ff72eef1af9e125a6c0.jpg",
+        "table_caption": [
+            "Table 1: Comparison of different sparse training techniques. Drop and Grow columns correspond to the strategies used during the mask update. Selectable $F L O P s$ is possible if the cost of training the model is fixed at the beginning of training. "
+        ],
+        "table_footnote": [],
+        "table_body": "<table><tr><td>Method</td><td>Drop</td><td>Grow</td><td>Selectable FLOPs</td><td>Space &amp;FLOPs X</td></tr><tr><td>SNIP</td><td>min(10 * VL(0)I)</td><td>none</td><td>yes</td><td>sparse</td></tr><tr><td>DeepR</td><td>stochastic</td><td>random</td><td>yes</td><td>sparse</td></tr><tr><td>SET</td><td>min(l0))</td><td>random</td><td>yes</td><td>sparse</td></tr><tr><td>DSR</td><td>min(le)</td><td>random</td><td>no</td><td>sparse</td></tr><tr><td>SNFS</td><td>min(l0))</td><td>momentum</td><td>no</td><td>dense</td></tr><tr><td>RigL (ours)</td><td>min(l0)</td><td>gradient</td><td>yes</td><td>sparse</td></tr></table>",
+        "bbox": [
+            187,
+            99,
+            810,
+            202
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "text",
+        "text": "",
+        "bbox": [
+            174,
+            282,
+            825,
+            409
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "text",
+        "text": "There has also been a line of work investigating the Lottery Ticket Hypothesis (Frankle & Carbin, 2019). Frankle et al. (2019) showed that the formulation must be weakened to apply to larger networks such as ResNet-50 (He et al., 2015). In large networks, instead of the original initialization, the values after thousands of optimization steps must be used for initialization. Zhou et al. (2019) showed that lottery tickets obtain non-random accuracies even before the training has started. Though the possibility of training sparse neural networks with a fixed sparsity mask using lottery tickets is intriguing, it remains unclear whether it is possible to generate such initializations – for both masks and parameters – de novo. ",
+        "bbox": [
+            173,
+            415,
+            825,
+            526
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "text",
+        "text": "3 RIGGING THE LOTTERY ",
+        "text_level": 1,
+        "bbox": [
+            176,
+            550,
+            405,
+            566
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "text",
+        "text": "Our method, RigL, is illustrated in Figure 1. At regularly spaced intervals our method removes a fraction of connections based on weight magnitudes and activates new ones using instantaneous gradient information. After updating the connectivity, training continues with the updated network until the next update. The main parts of our algorithm, Sparsity Distribution, Update Schedule, Drop Criterion, Grow Criterion, and the various options we considered for each, are explained below. The improved performance of $R i g L$ is due to two reasons: the use of a new method for activating connections that is efficient and more effective than choosing at random, and the use of a natural extension to an existing method for distributing parameters statically among convolutional layers. ",
+        "bbox": [
+            173,
+            583,
+            825,
+            709
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "text",
+        "text": "(0) Notation. Given a dataset $D$ with individual inputs $x _ { i }$ and targets $y _ { i }$ , one can train a neural network to minimize the loss function $\\textstyle \\sum _ { i } L ( f _ { \\theta } ( x _ { i } ) , y _ { i } )$ , where $f _ { \\boldsymbol { \\theta } } ( \\boldsymbol { x } )$ is the neural network with parameters $\\theta$ of length $N$ . The vector $\\theta$ can be decomposed into parameters $\\theta ^ { l }$ , of length $N ^ { l }$ , for each layer $l$ . A sparse network keeps only a fraction $D \\in ( 0 , 1 )$ of all connections, resulting in a sparsity of $S = 1 - D$ . More precisely, denoting the sparsity of individual layers with $s ^ { l }$ , the total parameter count of the sparse neural network satisfies $\\begin{array} { r } { \\dot { \\sum _ { l } } ( 1 - s ^ { l } ) N ^ { l } = ( 1 - \\bar { S } ) * N } \\end{array}$ . ",
+        "bbox": [
+            171,
+            715,
+            823,
+            746
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "image",
+        "img_path": "images/c4f8c255dd49151f0525882b1c8f56ef8e1197a1ff58bb53e4c69623eb1ec0b1.jpg",
+        "image_caption": [
+            "Figure 1: Dynamic sparse training aims to change connectivity during training to help out optimzation. "
+        ],
+        "image_footnote": [],
+        "bbox": [
+            204,
+            775,
+            794,
+            900
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "text",
+        "text": "",
+        "bbox": [
+            174,
+            103,
+            825,
+            161
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "(1) Sparsity Distribution. There are many ways of distributing the non-zero weights across the layers while satisfying the equality above. We avoid re-allocating parameters between layers during the training process as it makes it difficult to target a specific final FLOP budget, which is important for many inference applications. We consider the following three strategies: ",
+        "bbox": [
+            174,
+            166,
+            825,
+            223
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "1. Uniform: The sparsity $s ^ { l }$ of each individual layer is the same as the total sparsity $S$ . We keep the first layer dense $\\boldsymbol s ^ { 0 } = 0 ,$ ), since it has negligible number of parameters.   \n2. Erdos-R ˝ enyi: ´ As introduced in Mocanu et al. (2018), $s ^ { l }$ scales with n +n nl−1∗nl , where nl denotes number of neurons at layer l. This enables the number of connections in a sparse layer to scale with the sum of the number of output and input channels.   \n3. Erdos-R ˝ enyi-Kernel (ERK): ´ This method modifies the original Erdos-R ˝ enyi formulation ´ by including the kernel dimensions in the scaling factors. In other words, the number of parameters of the sparse convolutional layers are scaled proportional to 1− nl−1+nl+wl+hlnl−1∗nl∗wl∗hl , where $w ^ { l }$ and $h ^ { l }$ are the width and the height of the $\\mathbf { \\nabla } _ { l } ,$ ’th convolutional kernel. Sparsity of the fully connected layers scale as in the original Erdos-R ˝ enyi formulation. Similar to Erd ´ os- ˝ Renyi, ERK allocates higher sparsities to the layers with more parameters while allocating ´ lower sparsities to the smaller ones. ",
+        "bbox": [
+            210,
+            234,
+            825,
+            420
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "In all methods, the bias and batch-norm parameters are kept dense. ",
+        "bbox": [
+            173,
+            433,
+            611,
+            446
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "(2) Update Schedule. The update schedule is defined by the following parameters: (1) the number of iterations between sparse connectivity updates $( \\Delta T )$ , (2) the iteration at which to stop updating the sparse connectivity $( T _ { e n d } )$ , (3) the initial fraction of connections updated $( \\alpha )$ and (4) a function $f _ { d e c a y }$ , invoked every $\\Delta T$ iterations until $T _ { e n d }$ , possibly decaying the fraction of updated connections over time. For the latter we choose to use cosine annealing, as we find it slightly outperforms the other methods considered. ",
+        "bbox": [
+            173,
+            453,
+            825,
+            536
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "equation",
+        "img_path": "images/2875bcc5bfcbfd95d66fe9aaef666e78bcbfca0191f7386b9395ba33e82f35c9.jpg",
+        "text": "$$\nf _ { d e c a y } ( t ) = \\frac { \\alpha } { 2 } \\left( 1 + c o s \\left( \\frac { t \\pi } { T _ { e n d } } \\right) \\right)\n$$",
+        "text_format": "latex",
+        "bbox": [
+            379,
+            540,
+            619,
+            574
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "Alternatives to cosine annealing like a constant schedule and inverse power annealing are studied in the Appendix F. ",
+        "bbox": [
+            176,
+            578,
+            825,
+            607
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "(3) Drop criterion. Over the course of training, we drop the lowest magnitude weights according to the update schedule since they are expected to effect the training loss least. Specifically, we drop the connections given by ",
+        "bbox": [
+            176,
+            613,
+            823,
+            655
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "$T o p K ( - | \\theta ^ { l } | , f _ { d e c a y } ( t ) \\dot { ( } 1 - s ^ { l } ) N ^ { l } ) ^ { \\ddag } .$ ",
+        "bbox": [
+            176,
+            655,
+            418,
+            671
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "(4) Grow criterion. The novelty of our method lies in how we grow new connections. We grow the connections with highest magnitude gradients, $T o p K _ { w \\notin \\theta _ { a c t i v e } ^ { l } } \\breve { ( | { g r a d ( \\theta ^ { l } ) } | } , f _ { d e c a y } ( t ) ( 1 - s ^ { l } ) \\breve { N } ^ { l } ) .$ , where $\\theta _ { a c t i v e } ^ { l }$ is the set of active connections after the drop step. Newly activated connections are initialized to zero and therefore don’t effect the output of the network. However they are expected to receive gradients with high magnitudes in the next iteration and therefore reduce the loss fastest. This procedure can be applied to each layer in sequence and the dense gradients can be discarded immediately after selecting the top connections. If a layer is too large to materialize the full gradient with respect to the weights, then we can further reduce the memory requirements by performing an iterative calculation: ",
+        "bbox": [
+            173,
+            675,
+            825,
+            805
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "1. Initialize the set $T K = \\{ \\}$ .   \n2. Materialize a subset of size M of the full gradient, which we denote $G _ { i : i + M }$ .   \n3. Update $T K$ to contain the Top- $\\mathbf { \\nabla } \\cdot \\mathbf { K }$ elements of $G _ { i : i + M }$ concatenated with $T K$ .   \n4. Repeat steps 1 through 3 until all of the gradients have been materialized. The final set $T K$   \ncontains the connections we wish to grow. ",
+        "bbox": [
+            209,
+            815,
+            826,
+            901
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "image",
+        "img_path": "images/dfa7237fe18adc99e80cc225a8c395a896800044e15fa934ef6beb417685da6e.jpg",
+        "image_caption": [
+            "Figure 2: (left) Performance of various dynamic sparse training methods on ImageNet-2012 classification task. We use $80 \\%$ sparse ResNet-50 architecture with uniform sparsity distribution. Points at each curve correspond to the individual training runs with training multipliers from 1 to 5 (except pruning which is scaled between 0.5 and 2). We repeat training 3 times at every multiplier and report the mean accuracies. The number of FLOPs required to train a standard dense Resnet-50 along with its performance is indicated with a dashed red line. (right) Performance of RigL at different sparsity levels with extended training. Results are averaged over 3 runs. "
+        ],
+        "image_footnote": [],
+        "bbox": [
+            215,
+            102,
+            777,
+            266
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "text",
+        "text": "As long as $\\Delta T > \\frac { 1 } { 1 - s }$ the total work in calculating dense gradients is amortized and still proportional to $1 - S$ . This is in contrast to the method of Dettmers & Zettlemoyer (2019), which requires calculating and storing the full gradients at each optimization step. ",
+        "bbox": [
+            174,
+            386,
+            825,
+            431
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "text",
+        "text": "4 EMPIRICAL EVALUATION ",
+        "text_level": 1,
+        "bbox": [
+            176,
+            452,
+            415,
+            467
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "text",
+        "text": "Our experiments include image classification using CNNs on the ImageNet-2012 (Russakovsky et al., 2015) and CIFAR-10 (Krizhevsky et al.) datasets and character based language modelling using RNNs with the WikiText-103 dataset (Merity et al., 2016). We use the TensorFlow Model Pruning library (Zhu & Gupta, 2018) for our pruning baselines. A Tensorflow (Abadi et al., 2015) implementation of our method along with three other baselines (SET, SNFS, SNIP) will be open sourced. When we increase the training steps by a factor $M$ , the anchor epochs of the learning rate schedule and the end iteration of the mask update schedule are also scaled by the same factor; we indicate this scaling with a subscript (e.g. ${ \\mathrm { R i g L } } _ { M \\times } ,$ ). ",
+        "bbox": [
+            174,
+            483,
+            825,
+            595
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "text",
+        "text": "4.1 IMAGENET-2012 DATASET ",
+        "text_level": 1,
+        "bbox": [
+            176,
+            613,
+            400,
+            626
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "text",
+        "text": "In all experiments in this section, we use SGD with momentum as our optimizer. We set the momentum coefficient of the optimizer to 0.9, $L _ { 2 }$ regularization coefficient to 0.0001, and label smoothing (Szegedy et al., 2016) to 0.1. The learning rate schedule starts with a linear warm up reaching its maximum value of 1.6 at epoch 5 which is then dropped by a factor of 10 at epochs 30, 70 and 90. We train our networks with a batch size of 4096 for 32000 steps which roughly corresponds to 100 epochs of training. Our training pipeline uses standard data augmentation, which includes random flips and crops. ",
+        "bbox": [
+            174,
+            638,
+            825,
+            736
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "text",
+        "text": "4.1.1 RESNET-50 ",
+        "text_level": 1,
+        "bbox": [
+            174,
+            752,
+            312,
+            767
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "text",
+        "text": "Figure 2-left summarizes the performance of various methods on training an $80 \\%$ sparse ResNet-50. We also train small dense networks with equivalent parameter count. All sparse networks use the constant sparsity distribution and a cosine update schedule $\\mathrm { ~ \\ : ( \\alpha ~ ) = ~ 0 . 3 ~ }$ , $\\Delta T = 1 0 0$ ). Overall, we observe that the performance of all methods improves with training time; thus, for each method we run extended training with up to $5 \\times$ the training steps of the original. ",
+        "bbox": [
+            174,
+            776,
+            825,
+            847
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "text",
+        "text": "As noted by Gale et al. (2019), Evci et al. (2019), Frankle et al. (2019), and Mostafa & Wang (2019), training a network with fixed sparsity from scratch (Static) leads to inferior performance. Training a small dense network with the same number of parameters gets better results than Static, but fails to match the performance of dynamic sparse models. Similarly $S E T$ improves the performance over Small-Dense, however saturates around $7 5 \\%$ accuracy indicating the limits of growing new connec",
+        "bbox": [
+            174,
+            853,
+            823,
+            924
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "table",
+        "img_path": "images/6456898747c525bc970340414a9c373557a13cad7006f88915dcf071144cbb2e.jpg",
+        "table_caption": [],
+        "table_footnote": [],
+        "table_body": "<table><tr><td>Method</td><td>Top-1 Accuracy</td><td>FLOPs (Train)</td><td>FLOPs (Test)</td><td>Top-1 Accuracy</td><td>FLOPs (Train)</td><td>FLOPs (Test)</td></tr><tr><td>Dense</td><td>76.8±0.09</td><td>1x (3.2e18)</td><td>1x (8.2e9)</td><td colspan=\"3\"></td></tr><tr><td>Static</td><td colspan=\"3\">S=0.8</td><td colspan=\"3\">S=0.9</td></tr><tr><td>SNIP</td><td>70.6±0.06 72.0±0.10</td><td>0.23x 0.23x</td><td>0.23x 0.23x</td><td>65.8±0.04 67.2±0.12</td><td>0.10x 0.10x</td><td>0.10x 0.10x</td></tr><tr><td>Small-Dense</td><td>72.1±0.12</td><td>0.20x</td><td>0.20x</td><td>68.9±0.10</td><td>0.12x</td><td>0.12x</td></tr><tr><td>SET</td><td>72.9±0.39</td><td>0.23x</td><td>0.23x</td><td>69.6±0.23</td><td>0.10x</td><td>0.10x</td></tr><tr><td>RigL</td><td>74.6±0.06</td><td>0.23x</td><td>0.23x</td><td>72.0±0.05</td><td>0.10x</td><td>0.10x</td></tr><tr><td>Small-Dense5x RigL5x</td><td>73.9±0.07</td><td>1.01x</td><td>0.20x</td><td>71.3±0.10</td><td>0.60x</td><td>0.12x</td></tr><tr><td>Static (ERK)</td><td>76.6±0.06</td><td>1.14x</td><td>0.23x</td><td>75.7±0.06</td><td>0.52x</td><td>0.10x</td></tr><tr><td>DSR*</td><td>72.1±0.04 73.3</td><td>0.42x 0.40x</td><td>0.42x 0.40x</td><td>67.7±0.12</td><td>0.24x</td><td>0.24x</td></tr><tr><td></td><td></td><td></td><td></td><td>71.6</td><td>0.30x</td><td>0.30x</td></tr><tr><td>RigL (ERK)</td><td>75.1±0.05</td><td>0.42x</td><td>0.42x</td><td>73.0±0.04</td><td>0.25x</td><td>0.24x</td></tr><tr><td>RigL5x (ERK)</td><td>77.1±0.06</td><td>2.09x</td><td>0.42x</td><td>76.4±0.05</td><td>1.23x</td><td>0.24x</td></tr><tr><td>SNFS*</td><td>74.2</td><td>n/a</td><td>n/a</td><td>72.3</td><td>n/a</td><td>n/a</td></tr><tr><td>SNFS (ERK)</td><td>75.2±0.11</td><td>0.61x</td><td>0.42x</td><td>72.9±0.06</td><td>0.50x</td><td>0.24x</td></tr><tr><td>Pruning* (Zhu)</td><td>73.2</td><td>1.00x</td><td>0.23x</td><td>70.3</td><td>1.00x</td><td>0.10x</td></tr><tr><td>Pruning* (Gale)</td><td>75.6</td><td>1.00x</td><td>0.23x</td><td>73.9</td><td>1.00x</td><td>0.10x</td></tr><tr><td>Pruning1.5× (Gale)</td><td>76.5</td><td>1.50x</td><td>0.23x</td><td></td><td>1.50x</td><td></td></tr><tr><td></td><td></td><td></td><td></td><td>75.2</td><td></td><td>0.10x</td></tr></table>",
+        "bbox": [
+            199,
+            101,
+            799,
+            414
+        ],
+        "page_idx": 5
+    },
+    {
+        "type": "text",
+        "text": "Table 2: Performance and cost of sparse training methods on training $80 \\%$ and $90 \\%$ sparse ResNet50s. FLOPs needed for training and test are normalized with the FLOPs of a dense model (see Appendix G for details on how FLOPs are calculated). Methods with a subscript indicate a rescaled training time, whereas ‘\\*’ indicates reported results. (ERK) corresponds to the sparse networks with Erdos-Renyi-Kernel sparsity distribution.˝ $\\mathrm { R i g L } _ { \\mathrm { 5 } \\times }$ (ERK) achieves $7 7 . 1 \\%$ Top-1 Accuracy using only $20 \\%$ of the parameters of a dense model and $42 \\%$ of its FLOPs. ",
+        "bbox": [
+            173,
+            424,
+            825,
+            507
+        ],
+        "page_idx": 5
+    },
+    {
+        "type": "text",
+        "text": "tions randomly. Methods that use gradient information to grow new connections (RigL and SNFS) obtain higher accuracies, but RigL achieves the highest accuracy and does so while consistently requiring fewer FLOPs than the other methods. ",
+        "bbox": [
+            174,
+            518,
+            825,
+            560
+        ],
+        "page_idx": 5
+    },
+    {
+        "type": "text",
+        "text": "Given that different applications or scenarios might require a limit on the number of FLOPs for inference, we investigate the performance of our method at various sparsity levels. As mentioned previously, one strength of our method is that its resource requirements are constant throughout training and we can choose the level of sparsity that fits our training and/or inference constraints. In Figure 2-right we show the performance of our method at different sparsities and compare them with the pruning results of Gale et al. (2019), which uses $1 . 5 \\mathrm { x }$ training steps, relative to the original 32k iterations. To make a fair comparison with regards to FLOPs, we scale the learning schedule of all other methods by ${ 5 } \\mathbf { x }$ . Note that even after extending the training, it takes less FLOPs to train sparse networks using rigL (except for the $80 \\%$ sparse RigL-ERK) compared to the pruning method. ",
+        "bbox": [
+            174,
+            568,
+            825,
+            693
+        ],
+        "page_idx": 5
+    },
+    {
+        "type": "text",
+        "text": "RigL, our method with constant sparsity distribution, exceeds the performance of magnitude based iterative pruning in all sparsity levels while requiring less FLOPs to train. Sparse networks that use Erdos-Renyi-Kernel (ERK) ˝ sparsity distribution obtains even greater performance. For example ResNet-50 with $9 6 . 5 \\%$ sparsity achieves a remarkable $7 2 . 7 5 \\%$ Top-1 Accuracy, around $3 . 5 \\%$ higher than the extended magnitude pruning results reported by Gale et al. (2019). As observed earlier, smaller dense models (with the same number of parameters) or sparse models with a static connectivity can not perform at a comparable level. ",
+        "bbox": [
+            173,
+            699,
+            825,
+            797
+        ],
+        "page_idx": 5
+    },
+    {
+        "type": "text",
+        "text": "A more fine grained comparison of sparse training methods is presented in Table 2. Methods using uniform sparsity distribution and whose FLOP/memory footprint scales directly with (1-S) are placed in the first sub-group of the table. The second sub-group includes DSR and networks with ERK sparsity distribution which require a higher number of FLOPs for inference with same parameter count. The final sub-group includes methods that require the space and the work proportional to training a dense model. ",
+        "bbox": [
+            174,
+            804,
+            823,
+            888
+        ],
+        "page_idx": 5
+    },
+    {
+        "type": "image",
+        "img_path": "images/8ad076c5d43264442c51ef52152072882fb12f6803ab3f5d550b0ad517566e72.jpg",
+        "image_caption": [
+            "Figure 3: (left) RigL significantly improves the performance of Sparse MobileNets on ImageNet2012 dataset and exceeds the pruning results reported by Zhu & Gupta (2018). Performance of the dense MobileNets are indicated with red lines. (right) Performance of sparse MobileNet-v1 architectures presented with their inference FLOPs. Networks with ERK distribution get better performance with the same number of parameters but take more FLOPs to run. Training wider sparse models with RigL (Big-Sparse) yields a significant performance improvement over the dense model. "
+        ],
+        "image_footnote": [],
+        "bbox": [
+            184,
+            106,
+            450,
+            267
+        ],
+        "page_idx": 6
+    },
+    {
+        "type": "table",
+        "img_path": "images/29d17293935e4ad2e65d8ff2ae86da99bee4c3ed75e6bf42fab1681aa0fc4a94.jpg",
+        "table_caption": [],
+        "table_footnote": [],
+        "table_body": "<table><tr><td rowspan=1 colspan=1>S</td><td rowspan=1 colspan=1>Method</td><td rowspan=1 colspan=1>Top-1</td><td rowspan=1 colspan=1>FLOPs</td></tr><tr><td rowspan=1 colspan=1>0.75</td><td rowspan=1 colspan=1>Small-Dense5×Pruning (Zhu)RigL5×RigL5x(ERK)</td><td rowspan=1 colspan=1>66.0±0.1167.771.5±0.0671.9±0.01</td><td rowspan=1 colspan=1>0.23x0.27x0.27x0.52x</td></tr><tr><td rowspan=1 colspan=1>0.90</td><td rowspan=1 colspan=1>Small-Dense5xPruning (Zhu)RigL5×RigL5x(ERK)</td><td rowspan=1 colspan=1>57.7±0.3461.867.0±0.1768.1±0.11</td><td rowspan=1 colspan=1>0.09x0.12x0.12x0.27x</td></tr><tr><td rowspan=1 colspan=1></td><td rowspan=1 colspan=1>Dense</td><td rowspan=1 colspan=1>72.1±0.17</td><td rowspan=1 colspan=1>1x (1.1e9)</td></tr><tr><td rowspan=1 colspan=1>0.75</td><td rowspan=1 colspan=1>Big-Sparse5×Big-Sparse5x(ERK)</td><td rowspan=1 colspan=1>76.4±0.0577.0±0.08</td><td rowspan=1 colspan=1>0.98x1.91x</td></tr></table>",
+        "bbox": [
+            457,
+            99,
+            841,
+            275
+        ],
+        "page_idx": 6
+    },
+    {
+        "type": "text",
+        "text": "4.1.2 MOBILENET ",
+        "text_level": 1,
+        "bbox": [
+            176,
+            411,
+            316,
+            426
+        ],
+        "page_idx": 6
+    },
+    {
+        "type": "text",
+        "text": "MobileNet is a compact architecture that performs remarkably well in resource constrained settings. Due to its compact nature with separable convolutions it is known to be difficult to sparsify (Zhu & Gupta, 2018). In this section we apply our method to MobileNet-v1 (Howard et al., 2017) and MobileNet-v2 (Sandler et al., 2018). Due to its low parameter count we keep the first layer dense, and use ERK and Uniform sparsity distributions to sparsify the remaining layers. ",
+        "bbox": [
+            174,
+            438,
+            825,
+            507
+        ],
+        "page_idx": 6
+    },
+    {
+        "type": "text",
+        "text": "The performance of sparse MobileNets trained with RigL as well as the baselines are shown in Figure 3. We do extended training (5x of the original number of steps) for all runs in this section. Although MobileNets are more sensitive to sparsity compared to the ResNet-50 architecture, RigL successfully trains sparse MobileNets at high sparsities and exceeds the performance of previously reported pruning results. ",
+        "bbox": [
+            174,
+            513,
+            823,
+            584
+        ],
+        "page_idx": 6
+    },
+    {
+        "type": "text",
+        "text": "To demonstrate the advantages of sparse models, next, we train wider MobileNets while keeping the FLOPs and total number of parameters the same as the dense baseline using sparsity. A sparse MobileNet-v1 with width multiplier 1.98 and constant $7 5 \\%$ sparsity has the same FLOPs and parameter count as the dense baseline. Training this network with RigL yields an impressive $4 . 3 \\%$ absolute improvement in Top-1 Accuracy. ",
+        "bbox": [
+            174,
+            590,
+            823,
+            661
+        ],
+        "page_idx": 6
+    },
+    {
+        "type": "text",
+        "text": "4.2 CHARACTER LEVEL LANGUAGE MODELLING ",
+        "text_level": 1,
+        "bbox": [
+            174,
+            681,
+            531,
+            695
+        ],
+        "page_idx": 6
+    },
+    {
+        "type": "text",
+        "text": "Most prior work has only examined sparse training on vision networks [the exception is the earliest work - Deep Rewiring (Bellec et al., 2017) which trained an LSTM (Hochreiter & Schmidhuber, 1997) on the TIMIT (Garofolo et al., 1993) dataset]. To fully understand these techniques it is important to examine different architectures on different datasets. Kalchbrenner et al. (2018) found sparse GRUs (Cho et al., 2014) to be very effective at modeling speech, however the dataset they used is not available. We choose a proxy task with similar characteristics (dataset size and vocabulary size are approximately the same) - character level language modeling on the publicly available WikiText-103 (Merity et al., 2016) dataset. ",
+        "bbox": [
+            174,
+            708,
+            823,
+            819
+        ],
+        "page_idx": 6
+    },
+    {
+        "type": "text",
+        "text": "Our network consists of a shared embedding with dimensionality 128, a vocabulary size of 256, a GRU with a state size of 512, a readout from the GRU state consisting of two linear layers with 256 units and 128 units respectively. We train the next step prediction task with the standard cross entropy loss, the Adam optimizer, a learning rate of $7 e - 4$ , an L2 regularization coefficient of $5 e { - 4 }$ , a sequence length of 512, a batch size of 32 and gradient absolute value clipping of values larger (in magnitude) than 10. Baseline training length is 200,000 iterations. When inducing sparsity with magnitude pruning (Zhu & Gupta, 2018), we perform pruning between iterations 50,000 and ",
+        "bbox": [
+            174,
+            825,
+            825,
+            924
+        ],
+        "page_idx": 6
+    },
+    {
+        "type": "image",
+        "img_path": "images/f16a2ec71b43df85e1a6ff38138e0337a2c670eede9c85bef31c854aa486fd2f.jpg",
+        "image_caption": [
+            "Figure 4: (left) Final validation loss of various sparse training methods on character level language modelling task. Cross entropy loss is converted to bits (from nats). Performance and the training cost of a dense model is indicated with dashed red lines. (right) Test accuracies of sparse WideResNet22-2’s on CIFAR-10 task. "
+        ],
+        "image_footnote": [],
+        "bbox": [
+            183,
+            102,
+            807,
+            285
+        ],
+        "page_idx": 7
+    },
+    {
+        "type": "text",
+        "text": "150,000 with a pruning frequency of 1,000. We initialize sparse networks with a uniform sparsity distribution and use a cosine update schedule with $\\alpha = 0 . 1$ and $\\Delta T = 1 0 0$ . Unlike the previous experiments we keep updating the mask until the end of the training; we observed this performed slightly better than stopping at iteration 150,000. ",
+        "bbox": [
+            174,
+            364,
+            825,
+            420
+        ],
+        "page_idx": 7
+    },
+    {
+        "type": "text",
+        "text": "In Figure 4-left we report the validation loss of various solutions at the end of the training. For each method we perform extended runs to see how they scale with increasing training time. As observed before, SET performs worst than the other dynamic training methods and its performance improves only slightly with increased training time. On the other hand the performance of RigL and SNFS improves constantly with more training steps. Both of these methods falls short of matching the pruning performance. ",
+        "bbox": [
+            174,
+            428,
+            825,
+            511
+        ],
+        "page_idx": 7
+    },
+    {
+        "type": "text",
+        "text": "4.3 WIDERESNET-22-2 ON CIFAR-10 ",
+        "text_level": 1,
+        "bbox": [
+            178,
+            530,
+            455,
+            545
+        ],
+        "page_idx": 7
+    },
+    {
+        "type": "text",
+        "text": "In this section, we evaluate the performance of RigL on CIFAR-10 image classification benchmark. We train Wide Residual Network’s (Zagoruyko & Komodakis, 2016) with 22 layers using a width multiplier of 2 for 250 epochs (97656 steps). Learning rate starts at 0.1 and scaled down by a factor of 5 every 30,000 iterations. We use an L2 regularization coefficient of 5e-4, a batch size of 128 and a momentum coefficient of 0.9. We keep the hyper-parameters specific to RigL same as the ImageNet experiments, except the final iteration for mask updates; which is adjusted to 75000. Results with different mask update intervals can be found in Appendix H. ",
+        "bbox": [
+            174,
+            556,
+            825,
+            655
+        ],
+        "page_idx": 7
+    },
+    {
+        "type": "text",
+        "text": "Performance of RigL across different sparsity levels is presented in Figure 4-right. Corresponding final training losses of the trained networks can be found in Appendix H. The dense baseline obtains $9 4 . 1 \\%$ test accuracy. Networks with half of the connections removed $50 \\%$ sparsity) achieves roughly the same accuracy as the dense baseline. Surprisingly, some of the networks at this sparsity level generalize better than the dense baseline demonstrating the regularization aspect of using sparsity. With increased sparsity, we start to see a performance gap between the Static and Pruning solutions. Training static RigL networks longer seems to have limited effect on the final performance. On the other hand, RigL, matches the performance of pruning using only a fraction of resources needed for training a dense network. ",
+        "bbox": [
+            174,
+            661,
+            825,
+            786
+        ],
+        "page_idx": 7
+    },
+    {
+        "type": "text",
+        "text": "4.4 ANALYZING THE PERFORMANCE OF RigL ",
+        "text_level": 1,
+        "bbox": [
+            178,
+            806,
+            498,
+            820
+        ],
+        "page_idx": 7
+    },
+    {
+        "type": "text",
+        "text": "In this section we study the effect of sparsity distributions, update schedules, and dynamic connections on the performance of our method. The results for SET and SNFS are similar and are discussed in Appendices B and E. ",
+        "bbox": [
+            176,
+            833,
+            825,
+            875
+        ],
+        "page_idx": 7
+    },
+    {
+        "type": "text",
+        "text": "Effect of Mask Initialization: Figure 5-left shows how the sparsity distribution affects the final test accuracy of sparse ResNet-50s trained with RigL. Erdos-R ˝ enyi-Kernel (ERK) performs consis- ´ tently better than the other two distributions. ERK automatically allocates more parameters to the layers with few parameters by decreasing their sparsities†. This reallocation seems to be crucial for preserving the capacity of the network at high sparsity levels where ERK outperforms other distributions by a greater margin. Though it performs better, the ERK distribution requires approximately twice as many FLOPs compared to a uniform distribution. This highlights an interesting trade-off between accuracy and computational efficiency even though both models have the same number of parameters. ",
+        "bbox": [
+            174,
+            882,
+            823,
+            924
+        ],
+        "page_idx": 7
+    },
+    {
+        "type": "image",
+        "img_path": "images/5800160114d0737b3e2907dc82a457c52f3b8ccbcf1ecf67ee038a009c690904.jpg",
+        "image_caption": [
+            "Figure 5: (left) Performance of RigL at different sparsities using different sparsity masks (right) Ablation study on cosine schedule. Other methods are in the appendix. "
+        ],
+        "image_footnote": [],
+        "bbox": [
+            184,
+            102,
+            805,
+            285
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "",
+        "bbox": [
+            174,
+            335,
+            825,
+            420
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "Effect of Update Schedule and Frequency: In Figure 5-right, we evaluate the performance of our method on update intervals $\\Delta T \\in [ 5 0 , 1 0 0 , 5 0 0 , 1 0 0 0 ]$ and initial drop fractions $\\alpha \\in [ 0 . 1 , 0 . 3 , 0 . 5$ ]. The best accuracies are obtained when the mask is updated every 100 iterations with an initial drop fraction of 0.3 or 0.5. Notably, even with frequent update intervals (e.g. every 1000 iterations), RigL performs above $7 3 . 5 \\%$ . ",
+        "bbox": [
+            174,
+            426,
+            825,
+            496
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "Effect of Dynamic connections: Frankle et al. (2019) and Mostafa & Wang (2019) observed that static sparse training converges to a solution with a higher loss than dynamic sparse training. In Figure 6-left we examine the loss landscape lying between a solution found via static sparse training and a solution found via pruning to understand whether former lies in a basin isolated from the latter. Performing a linear interpolation between the two reveals the expected result – high-loss barrier – demonstrating that the loss landscape is not trivially connected. However, this is only one of infinitely many paths between the two points optimization can be used to find parametric curves that connects solutions (Garipov et al., 2018; Draxler et al., 2018) subject to constraints. For example Garipov et al. (2018) showed different dense solutions lie in the same basin by finding 2nd order Bezier curves with low energy between the two solutions. Following their method, we attempt ´ to find quadratic and cubic Bezier curves between the two sparse solutions. Surprisingly, even with a ´ cubic curve, we fail to find a path without a high-loss barrier. These results suggest that static sparse training can get stuck at local minima that are isolated from improved solutions. On the other hand, when we optimize the quadratic Bezier curve across the full ´ dense space we find a near-monotonic path to the improved solution, suggesting that allowing new connections to grow lends dynamic sparse training greater flexibility in navigating the loss landscape. ",
+        "bbox": [
+            173,
+            503,
+            825,
+            643
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "image",
+        "img_path": "images/96e95693d928d87ad902426e986b917f4232a0de25740f3cd1cb80a5c1a6297c.jpg",
+        "image_caption": [
+            "Figure 6: (left) Training loss evaluated at various points on interpolation curves between a magnitude pruning model (0.0) and a model trained with static sparsity (1.0). (right) Training loss of RigL and Static methods starting from the static sparse solution, and their final accuracies. "
+        ],
+        "image_footnote": [],
+        "bbox": [
+            179,
+            695,
+            816,
+            882
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "",
+        "bbox": [
+            174,
+            103,
+            825,
+            188
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "In Figure 6-right we train RigL starting from the sub-optimal solution found by static sparse training, demonstrating that it is able to escape the local minimum, whereas re-training with static sparse training cannot. RigL first removes connections with the smallest magnitudes since removing these connections have been shown to have a minimal effect on the loss (Han et al., 2015; Evci, 2018). Next, it activates connections with the high gradients, since these connections are expected to decrease the loss fastest. We hypothesize in Appendix A that RigL escapes bad critical points by replacing saddle directions with high gradient dimensions. ",
+        "bbox": [
+            174,
+            194,
+            825,
+            291
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "5 DISCUSSION & CONCLUSION ",
+        "text_level": 1,
+        "bbox": [
+            176,
+            313,
+            447,
+            328
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "In this work we introduced ‘Rigged Lottery’ or RigL, an algorithm for training sparse neural networks efficiently. For a given computational budget RigL achieves higher accuracies than existing dense-to-sparse and sparse-to-sparse training algorithms. RigL is useful in three different scenarios: (1) To improve the accuracy of sparse models intended for deployment; (2) To improve the accuracy of large sparse models which can only be trained for a limited number of iterations; and (3) Combined with sparse primitives to enable training of extremely large sparse models which otherwise would not be possible. ",
+        "bbox": [
+            174,
+            344,
+            825,
+            441
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "The third scenario is unexplored due to the lack of hardware and software support for sparsity. Nonetheless, work continues to improve the performance of sparse networks on current hardware (Hong et al., 2019; Merrill & Garland, 2016), and new types of hardware accelerators will have better support for parameter sparsity (Wang et al., 2018; Mike Ashby, 2019; Liu et al., 2018; Han et al., 2016a; Chen et al., 2019). RigL provides the tools to take advantage of, and motivation for, such advances. ",
+        "bbox": [
+            174,
+            449,
+            825,
+            532
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "REFERENCES ",
+        "text_level": 1,
+        "bbox": [
+            174,
+            554,
+            285,
+            569
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "Mart´ın Abadi, Ashish Agarwal, Paul Barham, Eugene Brevdo, Zhifeng Chen, Craig Citro, Greg S. Corrado, Andy Davis, Jeffrey Dean, Matthieu Devin, Sanjay Ghemawat, Ian Goodfellow, Andrew Harp, Geoffrey Irving, Michael Isard, Yangqing Jia, Rafal Jozefowicz, Lukasz Kaiser, Manjunath Kudlur, Josh Levenberg, Dan Mane, Rajat Monga, Sherry Moore, Derek Murray, Chris Olah, ´ Mike Schuster, Jonathon Shlens, Benoit Steiner, Ilya Sutskever, Kunal Talwar, Paul Tucker, Vincent Vanhoucke, Vijay Vasudevan, Fernanda Viegas, Oriol Vinyals, Pete Warden, Martin Watten- ´ berg, Martin Wicke, Yuan Yu, and Xiaoqiang Zheng. TensorFlow: Large-scale machine learning on heterogeneous systems, 2015. URL http://tensorflow.org/. ",
+        "bbox": [
+            176,
+            578,
+            825,
+            689
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "Guillaume Bellec, David Kappel, Wolfgang Maass, and Robert A. Legenstein. Deep rewiring: Training very sparse deep networks. CoRR, abs/1711.05136, 2017. URL http://arxiv. org/abs/1711.05136. ",
+        "bbox": [
+            174,
+            699,
+            825,
+            741
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "Y. Chen, T. Yang, J. Emer, and V. Sze. Eyeriss v2: A flexible accelerator for emerging deep neural networks on mobile devices. IEEE Journal on Emerging and Selected Topics in Circuits and Systems, 9(2):292–308, June 2019. doi: 10.1109/JETCAS.2019.2910232. ",
+        "bbox": [
+            176,
+            751,
+            823,
+            794
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "Kyunghyun Cho, B van Merrienboer, Caglar Gulcehre, F Bougares, H Schwenk, and Yoshua Bengio. Learning phrase representations using rnn encoder-decoder for statistical machine translation. In Conference on Empirical Methods in Natural Language Processing (EMNLP 2014), 2014. ",
+        "bbox": [
+            176,
+            804,
+            821,
+            847
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "Diederik P. Kingma Christos Louizos, Max Welling. Learning sparse neural networks through $l _ { 0 }$ regularization. In International Conference on Learning Representations, 2018. ",
+        "bbox": [
+            173,
+            856,
+            818,
+            886
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "Tim Dettmers and Luke Zettlemoyer. Sparse networks from scratch: Faster training without losing performance. ArXiv, 2019. URL http://arxiv.org/abs/1907.04840. ",
+        "bbox": [
+            176,
+            895,
+            821,
+            924
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "Felix Draxler, Kambis Veschgini, Manfred Salmhofer, and Fred A Hamprecht. Essentially No Barriers in Neural Network Energy Landscape. In International Conference on Machine Learning, 2018. ",
+        "bbox": [
+            174,
+            103,
+            823,
+            146
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Utku Evci. Detecting dead weights and units in neural networks. CoRR, 2018. URL http: //arxiv.org/abs/1806.06068. ",
+        "bbox": [
+            171,
+            155,
+            820,
+            183
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Utku Evci, Fabian Pedregosa, Aidan N. Gomez, and Erich Elsen. The difficulty of training sparse neural networks. ArXiv, 2019. URL http://arxiv.org/abs/1906.10732. ",
+        "bbox": [
+            171,
+            191,
+            823,
+            220
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Jonathan Frankle and Michael Carbin. The lottery ticket hypothesis: Finding sparse, trainable neural networks. In 7th International Conference on Learning Representations, ICLR 2019, New Orleans, LA, USA, May 6-9, 2019, 2019. URL https://openreview.net/forum?id $=$ rJl-b3RcF7. ",
+        "bbox": [
+            173,
+            229,
+            825,
+            286
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Jonathan Frankle, Gintare Karolina Dziugaite, Daniel M. Roy, and Michael Carbin. The lottery ticket hypothesis at scale. ArXiv, 2019. URL http://arxiv.org/abs/1903.01611. ",
+        "bbox": [
+            171,
+            294,
+            823,
+            324
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Trevor Gale, Erich Elsen, and Sara Hooker. The state of sparsity in deep neural networks. CoRR, abs/1902.09574, 2019. URL http://arxiv.org/abs/1902.09574. ",
+        "bbox": [
+            173,
+            332,
+            821,
+            361
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Timur Garipov, Pavel Izmailov, Dmitrii Podoprikhin, Dmitry P Vetrov, and Andrew Gordon Wilson. Loss surfaces, mode connectivity, and fast ensembling of dnns. In Advances in Neural Information Processing Systems, 2018. ",
+        "bbox": [
+            173,
+            369,
+            825,
+            411
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "J. S. Garofolo, L. F. Lamel, W. M. Fisher, J. G. Fiscus, D. S. Pallett, and N. L. Dahlgren. Darpa timit acoustic phonetic continuous speech corpus cdrom, 1993. ",
+        "bbox": [
+            171,
+            420,
+            823,
+            449
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Yiwen Guo, Anbang Yao, and Yurong Chen. Dynamic network surgery for efficient DNNs. CoRR, abs/1608.04493, 2016. URL http://arxiv.org/abs/1608.04493. ",
+        "bbox": [
+            174,
+            458,
+            821,
+            487
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Song Han, Jeff Pool, John Tran, and William Dally. Learning both weights and connections for efficient neural network. In Advances in neural information processing systems, 2015. ",
+        "bbox": [
+            173,
+            494,
+            823,
+            525
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Song Han, Xingyu Liu, Huizi Mao, Jing Pu, Ardavan Pedram, Mark A. Horowitz, and William J. Dally. EIE: Efficient Inference Engine on compressed deep neural network. In Proceedings of the 43rd International Symposium on Computer Architecture, 2016a. ",
+        "bbox": [
+            178,
+            531,
+            823,
+            575
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Song Han, Huizi Mao, and William J. Dally. Deep compression: Compressing deep neural network with pruning, trained quantization and huffman coding. In 4th International Conference on Learning Representations, ICLR 2016, San Juan, Puerto Rico, May 2-4, 2016, Conference Track Proceedings, 2016b. URL http://arxiv.org/abs/1510.00149. ",
+        "bbox": [
+            173,
+            583,
+            825,
+            640
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "B. Hassibi and D. Stork. Second order derivatives for network pruning: Optimal Brain Surgeon. 1993. ",
+        "bbox": [
+            174,
+            648,
+            823,
+            676
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Delving deep into rectifiers: Surpassing human-level performance on imagenet classification. In Proceedings of the 2015 IEEE International Conference on Computer Vision (ICCV), 2015. ",
+        "bbox": [
+            173,
+            685,
+            823,
+            729
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Sepp Hochreiter and Jurgen Schmidhuber. Long short-term memory. ¨ Neural Comput., 9(8):1735– 1780, November 1997. ISSN 0899-7667. doi: 10.1162/neco.1997.9.8.1735. URL http://dx. doi.org/10.1162/neco.1997.9.8.1735. ",
+        "bbox": [
+            176,
+            738,
+            823,
+            780
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Changwan Hong, Aravind Sukumaran-Rajam, Israt Nisa, Kunal Singh, and P. Sadayappan. Adaptive sparse tiling for sparse matrix multiplication. In Proceedings of the 24th Symposium on Principles and Practice of Parallel Programming, PPoPP ’19, pp. 300–314, New York, NY, USA, 2019. ACM. ISBN 978-1-4503-6225-2. doi: 10.1145/3293883.3295712. URL http://doi.acm. org/10.1145/3293883.3295712. ",
+        "bbox": [
+            174,
+            789,
+            825,
+            858
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Andrew G. Howard, Menglong Zhu, Bo Chen, Dmitry Kalenichenko, Weijun Wang, Tobias Weyand, Marco Andreetto, and Hartwig Adam. Mobilenets: Efficient convolutional neural networks for mobile vision applications. CoRR, abs/1704.04861, 2017. URL http://arxiv.org/abs/ 1704.04861. ",
+        "bbox": [
+            174,
+            867,
+            825,
+            922
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "Nal Kalchbrenner, Erich Elsen, Karen Simonyan, Seb Noury, Norman Casagrande, Edward Lockhart, Florian Stimberg, Aaron Oord, Sander Dieleman, and Koray Kavukcuoglu. Efficient neural audio synthesis. In International Conference on Machine Learning (ICML), 2018. ",
+        "bbox": [
+            176,
+            103,
+            821,
+            146
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "Alex Krizhevsky, Vinod Nair, and Geoffrey Hinton. Cifar-10 (canadian institute for advanced research). URL http://www.cs.toronto.edu/˜kriz/cifar.html. ",
+        "bbox": [
+            173,
+            154,
+            821,
+            183
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "Yann LeCun, John S. Denker, and Sara A. Solla. Optimal Brain Damage. In Advances in Neural Information Processing Systems, 1990. ",
+        "bbox": [
+            173,
+            190,
+            823,
+            220
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "Namhoon Lee, Thalaiyasingam Ajanthan, and Philip H. S. Torr. SNIP: Single-shot Network Pruning based on Connection Sensitivity. In International Conference on Learning Representations (ICLR), 2019, 2019. ",
+        "bbox": [
+            173,
+            228,
+            823,
+            271
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "Hao Li, Asim Kadav, Igor Durdanovic, Hanan Samet, and Hans Peter Graf. Pruning filters for efficient convnets. In International Conference on Learning Representations, 2016. ",
+        "bbox": [
+            173,
+            279,
+            823,
+            309
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "Chen Liu, Guillaume Bellec, Bernhard Vogginger, David Kappel, Johannes Partzsch, Felix Neumaerker, Sebastian Hoppner, Wolfgang Maass, Stephen B. Furber, Robert A. Legenstein, and ¨ Christian Mayr. Memory-efficient deep learning on a spinnaker 2 prototype. In Front. Neurosci., 2018. ",
+        "bbox": [
+            173,
+            315,
+            825,
+            372
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "Christos Louizos, Karen Ullrich, and Max Welling. Bayesian compression for deep learning. In Advances in Neural Information Processing Systems, 2017. ",
+        "bbox": [
+            174,
+            381,
+            821,
+            410
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "Stephen Merity, Caiming Xiong, James Bradbury, and Richard Socher. Pointer sentinel mixture models. ArXiv, 2016. URL http://arxiv.org/abs/1609.07843. ",
+        "bbox": [
+            173,
+            417,
+            823,
+            448
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "Duane Merrill and Michael Garland. Merge-based sparse matrix-vector multiplication (spmv) using the csr storage format. In Proceedings of the 21st ACM SIGPLAN Symposium on Principles and Practice of Parallel Programming, PPoPP ’16, pp. 43:1–43:2, New York, NY, USA, 2016. ACM. ISBN 978-1-4503-4092-2. doi: 10.1145/2851141.2851190. URL http://doi.acm.org/ 10.1145/2851141.2851190. ",
+        "bbox": [
+            173,
+            455,
+            825,
+            526
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "Peter Baldwin Martijn Bastiaan Oliver Bunting Aiken Cairncross Christopher Chalmers Liz Corrigan Sam Davis Nathan van Doorn Jon Fowler Graham Hazel Basile Henry David Page Jonny Shipton Shaun Steenkamp Mike Ashby, Christiaan Baaij. Exploiting unstructured sparsity on next-generation datacenter hardware. 2019. URL https://myrtle.ai/wp-content/ uploads/2019/06/IEEEformatMyrtle.ai_.21.06.19_b.pdf. ",
+        "bbox": [
+            173,
+            535,
+            825,
+            606
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "Decebal Constantin Mocanu, Elena Mocanu, Peter Stone, Phuong H Nguyen, Madeleine Gibescu, and Antonio Liotta. Scalable training of artificial neural networks with adaptive sparse connectivity inspired by network science. Nature Communications, 2018. ",
+        "bbox": [
+            174,
+            613,
+            823,
+            655
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "Dmitry Molchanov, Arsenii Ashukha, and Dmitry P. Vetrov. Variational Dropout Sparsifies Deep Neural Networks. In Proceedings of the 34th International Conference on Machine Learning, ICML 2017, Sydney, NSW, Australia, 6-11 August 2017, pp. 2498–2507, 2017. ",
+        "bbox": [
+            174,
+            662,
+            821,
+            707
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "Pavlo Molchanov, Stephen Tyree, Tero Karras, Timo Aila, and Jan Kautz. Pruning Convolutional Neural Networks for Resource Efficient Transfer Learning. CoRR, abs/1611.06440, 2016. ",
+        "bbox": [
+            169,
+            714,
+            823,
+            743
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "Hesham Mostafa and Xin Wang. Parameter efficient training of deep convolutional neural networks by dynamic sparse reparameterization. In Proceedings of the 36th International Conference on Machine Learning, ICML 2019, 9-15 June 2019, Long Beach, California, USA, pp. 4646–4655, 2019. URL http://proceedings.mlr.press/v97/mostafa19a.html. ",
+        "bbox": [
+            173,
+            751,
+            825,
+            809
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "Sharan Narang, Greg Diamos, Shubho Sengupta, and Erich Elsen. Exploring sparsity in recurrent neural networks. In 5th International Conference on Learning Representations, ICLR 2017, Toulon, France, April 24-26, 2017, Conference Track Proceedings, 2017. URL https: //openreview.net/forum?id=BylSPv9gx. ",
+        "bbox": [
+            174,
+            816,
+            825,
+            873
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "Jongsoo Park, Sheng R. Li, Wei Wen, Hai Li, Yiran Chen, and Pradeep Dubey. Holistic SparseCNN: Forging the trident of accuracy, speed, and size. CoRR, abs/1608.01409, 2016. URL http: //arxiv.org/abs/1608.01409. ",
+        "bbox": [
+            174,
+            882,
+            823,
+            924
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "Olga Russakovsky, Jia Deng, Hao Su, Jonathan Krause, Sanjeev Satheesh, Sean Ma, Zhiheng Huang, Andrej Karpathy, Aditya Khosla, Michael Bernstein, Alexander C. Berg, and Li FeiFei. Imagenet large scale visual recognition challenge. International Journal of Computer Vision (IJCV), 2015. ",
+        "bbox": [
+            174,
+            103,
+            825,
+            160
+        ],
+        "page_idx": 12
+    },
+    {
+        "type": "text",
+        "text": "M. Sandler, A. Howard, M. Zhu, A. Zhmoginov, and L. Chen. Mobilenetv2: Inverted residuals and linear bottlenecks. In 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp. 4510–4520, June 2018. doi: 10.1109/CVPR.2018.00474. ",
+        "bbox": [
+            171,
+            169,
+            823,
+            212
+        ],
+        "page_idx": 12
+    },
+    {
+        "type": "text",
+        "text": "S. Srinivas, A. Subramanya, and R. V. Babu. Training sparse neural networks. In 2017 IEEE Conference on Computer Vision and Pattern Recognition Workshops (CVPRW), pp. 455–462, July 2017. doi: 10.1109/CVPRW.2017.61. ",
+        "bbox": [
+            173,
+            220,
+            826,
+            263
+        ],
+        "page_idx": 12
+    },
+    {
+        "type": "text",
+        "text": "Nikko Strom. Sparse Connection and Pruning in Large Dynamic Artificial Neural Networks. In ¨ EUROSPEECH, 1997. ",
+        "bbox": [
+            173,
+            272,
+            823,
+            301
+        ],
+        "page_idx": 12
+    },
+    {
+        "type": "text",
+        "text": "Christian Szegedy, Vincent Vanhoucke, Sergey Ioffe, Jonathon Shlens, and Zbigniew Wojna. Rethinking the inception architecture for computer vision. In Proceedings of IEEE Conference on Computer Vision and Pattern Recognition,, 2016. URL http://arxiv.org/abs/1512. 00567. ",
+        "bbox": [
+            173,
+            310,
+            825,
+            366
+        ],
+        "page_idx": 12
+    },
+    {
+        "type": "text",
+        "text": "Enzo Tartaglione, Skjalg Lepsøy, Attilio Fiandrotti, and Gianluca Francini. Learning sparse neural networks via sensitivity-driven regularization. In Proceedings of the 32nd International Conference on Neural Information Processing Systems, NIPS’18, pp. 3882–3892, USA, 2018. Curran Associates Inc. URL http://dl.acm.org/citation.cfm?id $\\underline { { \\underline { { \\mathbf { \\Pi } } } } } =$ 3327144.3327303. ",
+        "bbox": [
+            174,
+            376,
+            825,
+            433
+        ],
+        "page_idx": 12
+    },
+    {
+        "type": "text",
+        "text": "Georg Thimm and Emile Fiesler. Evaluating pruning methods. In National Chiao-Tung University, pp. 2, 1995. ",
+        "bbox": [
+            171,
+            441,
+            823,
+            470
+        ],
+        "page_idx": 12
+    },
+    {
+        "type": "text",
+        "text": "Peiqi Wang, Yu Ji, Chi Hong, Yongqiang Lyu, Dongsheng Wang, and Yuan Xie. Snrram: An efficient sparse neural network computation architecture based on resistive random-access memory. In Proceedings of the 55th Annual Design Automation Conference, DAC ’18, pp. 106:1–106:6, New York, NY, USA, 2018. ACM. ISBN 978-1-4503-5700-5. doi: 10.1145/3195970.3196116. URL http://doi.acm.org/10.1145/3195970.3196116. ",
+        "bbox": [
+            173,
+            479,
+            825,
+            551
+        ],
+        "page_idx": 12
+    },
+    {
+        "type": "text",
+        "text": "Sergey Zagoruyko and Nikos Komodakis. Wide residual networks. In BMVC, 2016. ",
+        "bbox": [
+            176,
+            559,
+            725,
+            574
+        ],
+        "page_idx": 12
+    },
+    {
+        "type": "text",
+        "text": "Hattie Zhou, Janice Lan, Rosanne Liu, and Jason Yosinski. Deconstructing Lottery Tickets: Zeros, Signs, and the Supermask. ArXiv, 2019. ",
+        "bbox": [
+            176,
+            583,
+            820,
+            612
+        ],
+        "page_idx": 12
+    },
+    {
+        "type": "text",
+        "text": "Michael Zhu and Suyog Gupta. To Prune, or Not to Prune: Exploring the Efficacy of Pruning for Model Compression. In International Conference on Learning Representations Workshop, 2018. ",
+        "bbox": [
+            174,
+            621,
+            825,
+            651
+        ],
+        "page_idx": 12
+    },
+    {
+        "type": "text",
+        "text": "A EFFECT OF MASK UPDATES ON THE ENERGY LANDSCAPE ",
+        "text_level": 1,
+        "bbox": [
+            174,
+            676,
+            692,
+            694
+        ],
+        "page_idx": 12
+    },
+    {
+        "type": "text",
+        "text": "To update the connectivity of our sparse network, we first need to drop a fraction $d$ of the existing connections for each layer independently to create a budget for growing new connections. Following the recipe of magnitude based pruning(Han et al., 2015), we order parameters at layer $i$ by magnitude $| \\theta _ { i } |$ and drop the $N * ( 1 - S ) * d$ parameters with lowest magnitude. The effectiveness of this simple criteria can be explained through the first order Taylor approximation of the loss $L$ around the current set of parameters $\\theta$ . ",
+        "bbox": [
+            174,
+            708,
+            826,
+            792
+        ],
+        "page_idx": 12
+    },
+    {
+        "type": "equation",
+        "img_path": "images/0ec8b0e97fecff408ba0a2770a49d10b1475a273f85c0d31deec99d36e6e0140.jpg",
+        "text": "$$\n\\Delta L = L ( \\theta + \\Delta \\theta ) - L ( \\theta ) = \\nabla _ { \\theta } L ( \\theta ) \\Delta \\theta + R ( | | \\Delta \\theta | | _ { 2 } ^ { 2 } )\n$$",
+        "text_format": "latex",
+        "bbox": [
+            316,
+            811,
+            681,
+            830
+        ],
+        "page_idx": 12
+    },
+    {
+        "type": "text",
+        "text": "The main goal of dropping connections is to remove parameters with minimal impact on the neural network and therefore on its performance. Since removing the connection $\\theta _ { i }$ corresponds to setting it to zero, it incurs a change of $\\Delta \\theta ~ = ~ - \\theta _ { i }$ in that direction and a change of $\\begin{array} { r l } { \\Delta L _ { i } } & { { } = } \\end{array}$ $- \\nabla _ { \\boldsymbol { \\theta } _ { i } } \\mathbf { \\bar { L } } ( \\boldsymbol { \\theta } ) \\boldsymbol { \\theta } _ { i } + R ( \\boldsymbol { \\theta } _ { i } ^ { 2 } )$ in the loss, where the first term is usually defined as the saliency of a connection. Though using saliency to remove connections has been used as a criteria for removing connections (Molchanov et al., 2016), it has been shown to produce inferior results compared to magnitude based removal, especially when used to remove multiple connections at once (Evci, 2018). In contrast, picking the lowest magnitude connections ensures a small remainder term in addition to a low saliency, limiting the damage we make when we drop connections. Additionally, we note that connections with small magnitude can only remain small if the gradient is also small, meaning that the saliency is likely small when the parameter itself is small. Therefore we argue that the connections removed by RigL are likely to be saddle directions of the energy landscape. ",
+        "bbox": [
+            173,
+            839,
+            825,
+            924
+        ],
+        "page_idx": 12
+    },
+    {
+        "type": "text",
+        "text": "",
+        "bbox": [
+            174,
+            103,
+            823,
+            188
+        ],
+        "page_idx": 13
+    },
+    {
+        "type": "text",
+        "text": "After the removal of insignificant connections, we enable new connections that have the highest expected gradients. Since we initialize these new connections to zero, they are guaranteed to have high gradients in the proceeding iteration and therefore to reduce the loss quickly. By definition a direction with high magnitude gradient is not a saddle direction. Combining this observation with the previous ( $R i g L$ is likely to remove saddle directions) and the results in Section 4.4 we suggest that RigL improves the energy landscape of the optimization by replacing saddle directions with the ones with high gradient. This helps the optimization procedure to escape bad critical points and find solutions with higher quality. ",
+        "bbox": [
+            174,
+            194,
+            825,
+            306
+        ],
+        "page_idx": 13
+    },
+    {
+        "type": "text",
+        "text": "B EFFECT OF SPARSITY DISTRIBUTION ON OTHER METHODS ",
+        "text_level": 1,
+        "bbox": [
+            173,
+            327,
+            699,
+            343
+        ],
+        "page_idx": 13
+    },
+    {
+        "type": "text",
+        "text": "In Figure 7-left we show the effect of sparsity distribution choice on 4 different sparse training methods. ERK distribution performs better than other distributions for each training method. ",
+        "bbox": [
+            173,
+            359,
+            825,
+            387
+        ],
+        "page_idx": 13
+    },
+    {
+        "type": "text",
+        "text": "C EFFECT OF MOMENTUM COEFFICIENT FOR SNFS ",
+        "text_level": 1,
+        "bbox": [
+            173,
+            409,
+            620,
+            425
+        ],
+        "page_idx": 13
+    },
+    {
+        "type": "text",
+        "text": "In Figure 7 right we show the effect of the momentum coefficient on the performance of SNFS. Our results shows that using a coefficient of 0.99 brings the best performance. On the other hand using the most recent gradient only (coefficient of 0) performs as good as using a coefficient of 0.9. This result might be due to the large batch size we are using (4096), but it still motivates using RigL and instantaneous gradient information only when needed, instead of accumulating them. ",
+        "bbox": [
+            174,
+            441,
+            825,
+            511
+        ],
+        "page_idx": 13
+    },
+    {
+        "type": "text",
+        "text": "D EXISTENCE OF LOTTERY TICKETS ",
+        "text_level": 1,
+        "bbox": [
+            174,
+            534,
+            495,
+            549
+        ],
+        "page_idx": 13
+    },
+    {
+        "type": "text",
+        "text": "We perform the following experiment to see whether Lottery Tickets exist in our setting. We take the sparse network found by RigL and restart training using original initialization, both with RigL and with fixed topology as in the original Lottery Ticket Hypothesis. Results in table 3 demonstrate that training with a fixed topology is significantly worse than training with RigL and that RigL does not benefit from starting again with the final topology and the original initialization - training for twice as long instead of rewiring is more effective. In short, there are no special tickets, with RigL all tickets seems to win. ",
+        "bbox": [
+            173,
+            565,
+            825,
+            662
+        ],
+        "page_idx": 13
+    },
+    {
+        "type": "image",
+        "img_path": "images/f433bd443059bd764822e9e7a3060af7bbc37812c757ca9ad010a470abc0a839.jpg",
+        "image_caption": [
+            "Figure 7: (left) Effect of sparsity distribution choice on sparse training methods at different sparsity levels. We average over 3 runs and report the standard deviations for each. (right) Effect of momentum value on the performance of SNFS algorithm. Setting the momentum coefficient of the SNFS algorithm to 0 seems to perform best, suggesting the accumulated values are not important. "
+        ],
+        "image_footnote": [],
+        "bbox": [
+            183,
+            685,
+            808,
+            868
+        ],
+        "page_idx": 13
+    },
+    {
+        "type": "table",
+        "img_path": "images/94d62b6ac1e54ad04744876e63f71ea149d6ef8741dcfc583da03af321e74f12.jpg",
+        "table_caption": [
+            "Table 3: Effect of lottery ticket initialization on the final performance. There are no special tickets and dynamic connectivity provided by RigL is critical for good performance. "
+        ],
+        "table_footnote": [],
+        "table_body": "<table><tr><td>Initialization</td><td>Training Method</td><td>Test Accuracy</td><td>Training FLOPs</td></tr><tr><td>Lottery</td><td>Static</td><td>70.82±0.07</td><td>0.46x</td></tr><tr><td>Lottery</td><td>RigL</td><td>73.93±0.09</td><td>0.46x</td></tr><tr><td>Random</td><td>RigL</td><td>74.55±0.06</td><td>0.23x</td></tr><tr><td>Random</td><td>RigL2x</td><td>76.06±0.09</td><td>0.46x</td></tr></table>",
+        "bbox": [
+            261,
+            99,
+            738,
+            174
+        ],
+        "page_idx": 14
+    },
+    {
+        "type": "text",
+        "text": "E EFFECT OF UPDATE SCHEDULES ON OTHER DYNAMIC SPARSE METHODS ",
+        "text_level": 1,
+        "bbox": [
+            171,
+            236,
+            818,
+            253
+        ],
+        "page_idx": 14
+    },
+    {
+        "type": "text",
+        "text": "In Figure 8 we repeat the hyper-parameter sweep done for RigL in Figure 5-right, using SET and SNFS. Cosine schedule with $\\Delta T = 5 0$ and $\\alpha = 0 . 1$ seems to work best across all methods. An interesting observation is that higher drop fractions $( \\alpha )$ seem to work better with longer intervals $\\Delta T$ . For example, SET with $\\Delta T = 1 0 0 0$ seems to work best with $\\alpha = 0 . 5$ . ",
+        "bbox": [
+            174,
+            267,
+            825,
+            324
+        ],
+        "page_idx": 14
+    },
+    {
+        "type": "text",
+        "text": "F ALTERNATIVE UPDATE SCHEDULES ",
+        "text_level": 1,
+        "bbox": [
+            173,
+            344,
+            504,
+            361
+        ],
+        "page_idx": 14
+    },
+    {
+        "type": "text",
+        "text": "In Figure 9, we share the performance of two alternative annealing functions: ",
+        "bbox": [
+            171,
+            376,
+            679,
+            391
+        ],
+        "page_idx": 14
+    },
+    {
+        "type": "text",
+        "text": "1. Constant: $f _ { d e c a y } ( t ) = \\alpha$ .   \n2. Inverse Power: The fraction of weights updated decreases similarly to the schedule used in Zhu & Gupta (2018) for iterative pruning: $\\begin{array} { r } { f _ { d e c a y } ( t ) = \\alpha ( 1 - \\frac { t } { T _ { e n d } } ) ^ { k } } \\end{array}$ )k. In our experiments we tried $k = 1$ which is the linear decay and their default $k = 3$ . ",
+        "bbox": [
+            212,
+            401,
+            825,
+            465
+        ],
+        "page_idx": 14
+    },
+    {
+        "type": "text",
+        "text": "Constant seems to perform well with low initial drop fractions like $\\alpha = 0 . 1$ , but it starts to perform worse with increasing $\\alpha$ . Inverse Power for ${ \\bf k } = 3$ and ${ \\mathrm { k } } { = } 1$ (Linear) seems to perform similarly for low $\\alpha$ values. However the performance drops noticeably for ${ \\bf k } = 3$ when we increase the update interval. As reported by Dettmers & Zettlemoyer (2019) linear $( \\mathrm { k } { = } 1$ ) seems to provide similar results as the cosine schedule. ",
+        "bbox": [
+            174,
+            477,
+            825,
+            546
+        ],
+        "page_idx": 14
+    },
+    {
+        "type": "text",
+        "text": "G CALCULATING FLOPS OF MODELS AND METHODS ",
+        "text_level": 1,
+        "bbox": [
+            174,
+            568,
+            630,
+            583
+        ],
+        "page_idx": 14
+    },
+    {
+        "type": "text",
+        "text": "In order to calculate FLOPs needed for a single forward pass of a sparse model, we count the total number of multiplications and additions layer by layer for a given layer sparsity $s ^ { l }$ . The total FLOPs is then obtained by summing up all of these multiply and adds. ",
+        "bbox": [
+            174,
+            598,
+            825,
+            641
+        ],
+        "page_idx": 14
+    },
+    {
+        "type": "text",
+        "text": "Different sparsity distributions require different number of FLOPs to compute a single prediction. For example Erdos-Renyi-Kernel ˝ distributions usually cause earlier layers to be less sparse than the later layers (see Appendix I). The inputs of earlier layers have greater spatial dimensions, so a convolutional kernel that works on such inputs will require more FLOPs to compute the output features compared to later layers. Thus, having earlier layers which are less sparse results in a higher total number of FLOPs required by a model. ",
+        "bbox": [
+            174,
+            647,
+            825,
+            690
+        ],
+        "page_idx": 14
+    },
+    {
+        "type": "image",
+        "img_path": "images/b3e467a52c2f5b11283887318ea99ba50224d604ca1a45394151b61438546ee9.jpg",
+        "image_caption": [
+            "Figure 8: Cosine update schedule hyper-parameter sweep done using dynamic sparse training methods SET (left) and SNFS (right). "
+        ],
+        "image_footnote": [],
+        "bbox": [
+            183,
+            714,
+            805,
+            895
+        ],
+        "page_idx": 14
+    },
+    {
+        "type": "text",
+        "text": "",
+        "bbox": [
+            176,
+            103,
+            825,
+            146
+        ],
+        "page_idx": 15
+    },
+    {
+        "type": "text",
+        "text": "Training a neural network consists of 2 main steps: ",
+        "bbox": [
+            176,
+            152,
+            508,
+            167
+        ],
+        "page_idx": 15
+    },
+    {
+        "type": "text",
+        "text": "1. forward pass: Calculating the loss of the current set of parameters on a given batch of data. During this process layer activations are calculated in sequence using the previous activations and the parameters of the layer. Activation of layers are stored in memory for the backward pass.   \n2. backward pass: Using the loss value as the initial error signal, we back-propagate the error signal while calculating the gradient of parameters. During the backward pass each layer calculates 2 quantities: the gradient of the activations of the previous layer and the gradient of its parameters. Therefore in our calculations we count backward passes as two times the computational expense of the forward pass. We omit the FLOPs needed for batch normalization and cross entropy. ",
+        "bbox": [
+            210,
+            179,
+            825,
+            325
+        ],
+        "page_idx": 15
+    },
+    {
+        "type": "text",
+        "text": "Dynamic sparse training methods require some extra FLOPs to update the connectivity of the neural network. We omit FLOPs needed for dropping the lowest magnitude connections in our calculations. For a given dense architecture with FLOPs $f _ { D }$ and a sparse version with FLOPs $f _ { S }$ , the total FLOPs required to calculate the gradient on a single sample is computed as follows: ",
+        "bbox": [
+            176,
+            337,
+            825,
+            393
+        ],
+        "page_idx": 15
+    },
+    {
+        "type": "text",
+        "text": "• Static Sparse and Dense. Scales with $3 * f _ { S }$ and $3 * f _ { D }$ FLOPs, respectively.   \n• Snip. We omit the initial dense gradient calculation since it is negligible, which means Snip scales in the same way as Static methods: $3 * f _ { S }$ FLOPs.   \n• SET. We omit the extra FLOPs needed for growing random connections, since this operation can be done on chip efficiently. Therefore, the total FLOPs for SET scales with $3 * f _ { S }$ . SNFS. Forward pass and back-propagating the error signal needs $2 * f _ { S }$ FLOPs. However, the dense gradient needs to be calculated at every iteration. Thus, the total number of FLOPs scales with $2 * f _ { S } + f _ { D }$ . RigL. Iterations with no connection updates need $3 * f _ { S }$ FLOPs. However, at every $\\Delta T$ iteration we need to calculate the dense gradients. This results in the average FLOPs for RigL given by $\\frac { ( 3 * f _ { S } * \\Delta T + 2 * f _ { S } + f _ { D } ) } { ( \\Delta T + 1 ) }$ . ",
+        "bbox": [
+            215,
+            405,
+            825,
+            603
+        ],
+        "page_idx": 15
+    },
+    {
+        "type": "text",
+        "text": "H ADDITIONAL PLOTS AND EXPERIMENTS FOR CIFAR-10 ",
+        "text_level": 1,
+        "bbox": [
+            176,
+            621,
+            679,
+            637
+        ],
+        "page_idx": 15
+    },
+    {
+        "type": "text",
+        "text": "In Figure 10-left, we plot the final training loss of experiments presented in Section 4.3 to investigate the generalization properties of the algorithms considered. Poor performance of Static reflects itself in training loss clearly across all sparsity levels. RigL achieves similar final loss as the pruning, despite having around half percent less accuracy. Training longer with RigL decreases the final loss further and the test accuracies start matching pruning (see Figure 4-right) performance. These results show that RigL improves the optimization as promised, however generalizes slightly worse than pruning. ",
+        "bbox": [
+            173,
+            654,
+            825,
+            752
+        ],
+        "page_idx": 15
+    },
+    {
+        "type": "image",
+        "img_path": "images/bf63372d44ae944bab2fb94af7feac16fac25a9bc8f8d3818b33dd4260972182.jpg",
+        "image_caption": [
+            "Figure 9: Using other update schedules with RigL: (left) Constant (middle) Exponential $\\left( \\mathrm { k } \\mathrm { = } 3 \\right)$ and (right) Linear "
+        ],
+        "image_footnote": [],
+        "bbox": [
+            181,
+            776,
+            812,
+            897
+        ],
+        "page_idx": 15
+    },
+    {
+        "type": "image",
+        "img_path": "images/e207781847e04baa8da603b4cc537a5073cb27d1d766bb864de9512780d702bc.jpg",
+        "image_caption": [
+            "Figure 10: Final training loss of sparse models (left) and performance of $R i g L$ at different mask update intervals (right). "
+        ],
+        "image_footnote": [],
+        "bbox": [
+            192,
+            102,
+            807,
+            275
+        ],
+        "page_idx": 16
+    },
+    {
+        "type": "text",
+        "text": "In Figure 10-right, we sweep mask update interval $\\Delta T$ and plot the final test accuracies. We fix initial drop fraction $\\alpha$ to 0.3 and evaluate two different sparsity distributions: Uniform and $E R K$ . Both curves follow a similar pattern as in Imagenet-2012 sweeps (see Figure 8) and best results are obtained when $\\Delta T = 1 0 0$ . ",
+        "bbox": [
+            174,
+            325,
+            825,
+            381
+        ],
+        "page_idx": 16
+    },
+    {
+        "type": "text",
+        "text": "I SPARSITY OF INDIVIDUAL LAYERS FOR SPARSE RESNET-50 ",
+        "text_level": 1,
+        "bbox": [
+            174,
+            401,
+            702,
+            417
+        ],
+        "page_idx": 16
+    },
+    {
+        "type": "text",
+        "text": "Sparsity of ResNet-50 layers given by the Erdos-R ˝ enyi-Kernel sparsity distribution plotted in Figure ´ 11. ",
+        "bbox": [
+            173,
+            433,
+            821,
+            460
+        ],
+        "page_idx": 16
+    },
+    {
+        "type": "text",
+        "text": "J BUGS DISCOVERED DURING EXPERIMENTS ",
+        "text_level": 1,
+        "bbox": [
+            174,
+            481,
+            566,
+            497
+        ],
+        "page_idx": 16
+    },
+    {
+        "type": "text",
+        "text": "Our initial implementations contained some subtle bugs, which while not affecting the general conclusion that RigL is more effective than other techniques, did result in lower accuracy for all sparse training techniques. We detail these issues here with the hope that others may learn from our mistakes. ",
+        "bbox": [
+            174,
+            512,
+            825,
+            568
+        ],
+        "page_idx": 16
+    },
+    {
+        "type": "text",
+        "text": "1. Random operations on multiple replicas. We use data parallelism to split a mini-batch among multiple replicas. Each replica independently calculates the gradients using a different sub-mini-batch of data. The gradients are aggregated using an ALL-REDUCE operation before the optimizer update. Our implementation of SET, SNFS and RigL depended on each replica independently choosing to drop and grow the same connections. However, due to the nature of random operations in Tensorflow, this did not happen. Instead, different replicas diverged after the first drop/grow step. This was most pronounced in SET where each replica chose at random and much less so for SNFS and RigL where randomness is only needed to break ties. If left unchecked this might be expected to be catastrophic, but due to the behavior of Estimators and/or TF-replicator, the values on the first replica are broadcast to the others periodically (every approximately 1000 steps in our case). ",
+        "bbox": [
+            214,
+            582,
+            825,
+            734
+        ],
+        "page_idx": 16
+    },
+    {
+        "type": "text",
+        "text": "We fixed this bug by using stateless random operations. As a result the performance of SET improved slightly $0 . 1 \\%$ higher on Table 2). ",
+        "bbox": [
+            228,
+            736,
+            820,
+            765
+        ],
+        "page_idx": 16
+    },
+    {
+        "type": "text",
+        "text": "2. Synchronization between replicas. RigL and SNFS depend on calculating dense gradients with respect to the masked parameters. However, as explained above, in the multiple replica setting these gradients need to be aggregated. Normally this aggregation is automatically done by the optimizer, but in our case, this does not happen (only the gradients with respect to the unmasked parameters are aggregated automatically). This bug affected SNFS and RigL, but not SET since SET does not rely on the gradients to grow connections. Again, the synchronization of the parameters from the first replica every approximately 1000 steps masked this bug. ",
+        "bbox": [
+            214,
+            768,
+            825,
+            880
+        ],
+        "page_idx": 16
+    },
+    {
+        "type": "text",
+        "text": "We fixed this bug by explicitly calling ALL-REDUCE on the gradients with respect to the masked parameters. With this fix, the performance of RigL and SNFS improved significantly, particularly for default training lengths (around $0 . 5 – 1 \\%$ improvement). ",
+        "bbox": [
+            233,
+            882,
+            823,
+            924
+        ],
+        "page_idx": 16
+    },
+    {
+        "type": "image",
+        "img_path": "images/0b1f9b452aac571c6f076bd13ec4d1e1e1f13be162984229dcfda1d2c8d58426.jpg",
+        "image_caption": [
+            "Figure 11: Sparsities of individual layers of the ResNet-50. "
+        ],
+        "image_footnote": [],
+        "bbox": [
+            243,
+            209,
+            753,
+            815
+        ],
+        "page_idx": 17
+    },
+    {
+        "type": "text",
+        "text": "3. SNIP Experiments. Our first implementation of SNIP used the gradient magnitudes to decide which connections to keep causing its performance to be worse than static. Upon our discussions with the authors of SNIP, we realized that the correct metric is the saliency (gradient times parameter magnitude). With this correction SNIP performance improved dramatically to better than random (Static) even at Resnet-50/ImageNet scale. It is surprising that picking connections with the highest gradient magnitudes can be so detrimental to training (it resulted in much worse than random performance). ",
+        "bbox": [
+            214,
+            103,
+            825,
+            202
+        ],
+        "page_idx": 18
+    }
+]
\ No newline at end of file
diff --git a/parse/train/ryg7vA4tPB/ryg7vA4tPB_middle.json b/parse/train/ryg7vA4tPB/ryg7vA4tPB_middle.json
new file mode 100644
index 0000000000000000000000000000000000000000..5d25a3a07a8c4a1c124466d46090a118f559661a
--- /dev/null
+++ b/parse/train/ryg7vA4tPB/ryg7vA4tPB_middle.json
@@ -0,0 +1,45234 @@
+{
+    "pdf_info": [
+        {
+            "preproc_blocks": [
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        79,
+                        354,
+                        116
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                78,
+                                287,
+                                97
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        78,
+                                        287,
+                                        97
+                                    ],
+                                    "score": 1.0,
+                                    "content": "RIGGING THE LOTTERY:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                97,
+                                355,
+                                117
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        97,
+                                        355,
+                                        117
+                                    ],
+                                    "score": 1.0,
+                                    "content": "MAKING ALL TICKETS WINNERS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        }
+                    ],
+                    "index": 0.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        112,
+                        135,
+                        245,
+                        157
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                113,
+                                135,
+                                201,
+                                147
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        113,
+                                        135,
+                                        201,
+                                        147
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Anonymous authors",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                111,
+                                146,
+                                245,
+                                158
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        111,
+                                        146,
+                                        245,
+                                        158
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Paper under double-blind review",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        }
+                    ],
+                    "index": 2.5
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        278,
+                        186,
+                        333,
+                        199
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                276,
+                                185,
+                                336,
+                                200
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        276,
+                                        185,
+                                        336,
+                                        200
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ABSTRACT",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        }
+                    ],
+                    "index": 4
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        143,
+                        211,
+                        468,
+                        410
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                141,
+                                211,
+                                469,
+                                225
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        211,
+                                        469,
+                                        225
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Sparse neural networks have been shown to be more parameter and compute ef-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                223,
+                                469,
+                                236
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        223,
+                                        469,
+                                        236
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ficient compared to dense networks and in some cases they are even successfully",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                142,
+                                234,
+                                469,
+                                246
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        142,
+                                        234,
+                                        469,
+                                        246
+                                    ],
+                                    "score": 1.0,
+                                    "content": "used to decrease wall clock inference times. There is a large body of work on",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                245,
+                                469,
+                                257
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        245,
+                                        469,
+                                        257
+                                    ],
+                                    "score": 1.0,
+                                    "content": "training dense networks to yield sparse networks for inference (Molchanov et al.,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                255,
+                                469,
+                                268
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        255,
+                                        469,
+                                        268
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2017; Zhu & Gupta, 2018; Louizos et al., 2017; Li et al., 2016; Guo et al., 2016).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                266,
+                                470,
+                                280
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        266,
+                                        470,
+                                        280
+                                    ],
+                                    "score": 1.0,
+                                    "content": "This limits the size of the largest trainable sparse model to that of the largest train-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                277,
+                                470,
+                                290
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        277,
+                                        470,
+                                        290
+                                    ],
+                                    "score": 1.0,
+                                    "content": "able dense model. In this paper we introduce a method to train sparse neural",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                289,
+                                470,
+                                301
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        289,
+                                        470,
+                                        301
+                                    ],
+                                    "score": 1.0,
+                                    "content": "networks with a fixed parameter count and a fixed computational cost throughout",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                300,
+                                469,
+                                312
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        300,
+                                        469,
+                                        312
+                                    ],
+                                    "score": 1.0,
+                                    "content": "training, without sacrificing accuracy relative to existing dense-to-sparse training",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                310,
+                                469,
+                                324
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        310,
+                                        469,
+                                        324
+                                    ],
+                                    "score": 1.0,
+                                    "content": "methods. Our method updates the topology of the network during training by",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                321,
+                                470,
+                                334
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        321,
+                                        470,
+                                        334
+                                    ],
+                                    "score": 1.0,
+                                    "content": "using parameter magnitudes and infrequent gradient calculations. We show that",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                332,
+                                470,
+                                346
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        332,
+                                        470,
+                                        346
+                                    ],
+                                    "score": 1.0,
+                                    "content": "this approach requires fewer floating-point operations (FLOPs) to achieve a given",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                343,
+                                470,
+                                357
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        343,
+                                        470,
+                                        357
+                                    ],
+                                    "score": 1.0,
+                                    "content": "level of accuracy compared to prior techniques. We demonstrate state-of-the-art",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                355,
+                                470,
+                                366
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        355,
+                                        470,
+                                        366
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sparse training results with ResNet-50, MobileNet v1 and MobileNet v2 on the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                365,
+                                469,
+                                378
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        365,
+                                        469,
+                                        378
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ImageNet-2012 dataset, WideResNets on the CIFAR-10 dataset and RNNs on the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                376,
+                                469,
+                                389
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        376,
+                                        469,
+                                        389
+                                    ],
+                                    "score": 1.0,
+                                    "content": "WikiText-103 dataset. Finally, we provide some insights into why allowing the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                387,
+                                470,
+                                401
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        387,
+                                        470,
+                                        401
+                                    ],
+                                    "score": 1.0,
+                                    "content": "topology to change during the optimization can overcome local minima encoun-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                398,
+                                302,
+                                411
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        398,
+                                        302,
+                                        411
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tered when the topology remains static.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        }
+                    ],
+                    "index": 13.5
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        432,
+                        206,
+                        444
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                430,
+                                208,
+                                447
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        430,
+                                        208,
+                                        447
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1 INTRODUCTION",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        }
+                    ],
+                    "index": 23
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        457,
+                        505,
+                        534
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                457,
+                                506,
+                                471
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        457,
+                                        506,
+                                        471
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The parameter and floating point operation (FLOP) efficiency of sparse neural networks is now well",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                467,
+                                506,
+                                481
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        467,
+                                        506,
+                                        481
+                                    ],
+                                    "score": 1.0,
+                                    "content": "demonstrated on a variety of problems (Han et al., 2015; Srinivas et al., 2017). Some work has even",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                479,
+                                505,
+                                491
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        479,
+                                        505,
+                                        491
+                                    ],
+                                    "score": 1.0,
+                                    "content": "shown inference time speedups are possible on Recurrent Neural Networks (RNNs) (Kalchbrenner",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                490,
+                                505,
+                                502
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        490,
+                                        505,
+                                        502
+                                    ],
+                                    "score": 1.0,
+                                    "content": "et al., 2018) and Convolutional Neural Networks (ConvNets) (Park et al., 2016). Currently, the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                501,
+                                506,
+                                513
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        501,
+                                        506,
+                                        513
+                                    ],
+                                    "score": 1.0,
+                                    "content": "most accurate sparse models are obtained with techniques that require, at a minimum, the cost of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                512,
+                                506,
+                                525
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        512,
+                                        506,
+                                        525
+                                    ],
+                                    "score": 1.0,
+                                    "content": "training a dense model in terms of memory and FLOPs (Zhu & Gupta, 2018; Guo et al., 2016), and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                523,
+                                495,
+                                535
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        523,
+                                        495,
+                                        535
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sometimes significantly more (Molchanov et al., 2017). This paradigm has two main limitations:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        }
+                    ],
+                    "index": 27
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        129,
+                        545,
+                        505,
+                        605
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                129,
+                                545,
+                                505,
+                                558
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        129,
+                                        545,
+                                        505,
+                                        558
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1. The maximum size of sparse models is limited to the largest dense model that can be",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                142,
+                                556,
+                                505,
+                                569
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        142,
+                                        556,
+                                        505,
+                                        569
+                                    ],
+                                    "score": 1.0,
+                                    "content": "trained. Even if sparse models are more parameter efficient, we can’t use pruning to train",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                142,
+                                568,
+                                464,
+                                580
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        142,
+                                        568,
+                                        464,
+                                        580
+                                    ],
+                                    "score": 1.0,
+                                    "content": "models that are larger and more accurate than the largest possible dense models.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                128,
+                                581,
+                                505,
+                                596
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        128,
+                                        581,
+                                        505,
+                                        596
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2. It is inefficient. Large amounts of computation must be performed for parameters that are",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                142,
+                                594,
+                                339,
+                                606
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        142,
+                                        594,
+                                        339,
+                                        606
+                                    ],
+                                    "score": 1.0,
+                                    "content": "zero valued or that will be zero during inference.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        }
+                    ],
+                    "index": 33
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        616,
+                        505,
+                        704
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                615,
+                                505,
+                                628
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        615,
+                                        505,
+                                        628
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Additionally, it remains unknown if the performance of the current best pruning algorithms are an",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                627,
+                                505,
+                                639
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        627,
+                                        505,
+                                        639
+                                    ],
+                                    "score": 1.0,
+                                    "content": "upper bound on the quality of sparse models. Gale et al. (2019) found that three different dense-to-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                638,
+                                505,
+                                650
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        638,
+                                        505,
+                                        650
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sparse training algorithms all achieve about the same sparsity / accuracy trade-off. However, this is",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                648,
+                                505,
+                                663
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        648,
+                                        505,
+                                        663
+                                    ],
+                                    "score": 1.0,
+                                    "content": "far from conclusive proof that no better performance is possible. In this work we show the surprising",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                660,
+                                505,
+                                672
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        660,
+                                        505,
+                                        672
+                                    ],
+                                    "score": 1.0,
+                                    "content": "result that dynamic sparse training, which includes the method we introduce below, can find more",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                670,
+                                505,
+                                684
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        670,
+                                        505,
+                                        684
+                                    ],
+                                    "score": 1.0,
+                                    "content": "accurate models than the current best approaches to pruning initially dense networks. Importantly,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                681,
+                                505,
+                                694
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        681,
+                                        505,
+                                        694
+                                    ],
+                                    "score": 1.0,
+                                    "content": "our method does not change the FLOPs required to execute the model during training, allowing one",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                692,
+                                323,
+                                706
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        692,
+                                        323,
+                                        706
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to decide on a specific inference cost prior to training.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        }
+                    ],
+                    "index": 39.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        709,
+                        504,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                709,
+                                505,
+                                722
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        709,
+                                        505,
+                                        722
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The Lottery Ticket Hypothesis (Frankle & Carbin, 2019) hypothesized that if we can find a sparse",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                721,
+                                505,
+                                733
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        721,
+                                        505,
+                                        733
+                                    ],
+                                    "score": 1.0,
+                                    "content": "neural network with iterative pruning, then we can train that sparse network from scratch, to the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        }
+                    ],
+                    "index": 44.5
+                }
+            ],
+            "page_idx": 0,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        107,
+                        27,
+                        308,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                107,
+                                26,
+                                308,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        26,
+                                        308,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Under review as a conference paper at ICLR 2020",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        303,
+                        751,
+                        308,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                302,
+                                751,
+                                308,
+                                761
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        302,
+                                        751,
+                                        308,
+                                        761
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        79,
+                        354,
+                        116
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                78,
+                                287,
+                                97
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        78,
+                                        287,
+                                        97
+                                    ],
+                                    "score": 1.0,
+                                    "content": "RIGGING THE LOTTERY:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                97,
+                                355,
+                                117
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        97,
+                                        355,
+                                        117
+                                    ],
+                                    "score": 1.0,
+                                    "content": "MAKING ALL TICKETS WINNERS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        }
+                    ],
+                    "index": 0.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        112,
+                        135,
+                        245,
+                        157
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                113,
+                                135,
+                                201,
+                                147
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        113,
+                                        135,
+                                        201,
+                                        147
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Anonymous authors",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                111,
+                                146,
+                                245,
+                                158
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        111,
+                                        146,
+                                        245,
+                                        158
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Paper under double-blind review",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        }
+                    ],
+                    "index": 2.5,
+                    "bbox_fs": [
+                        111,
+                        135,
+                        245,
+                        158
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        278,
+                        186,
+                        333,
+                        199
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                276,
+                                185,
+                                336,
+                                200
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        276,
+                                        185,
+                                        336,
+                                        200
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ABSTRACT",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        }
+                    ],
+                    "index": 4
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        143,
+                        211,
+                        468,
+                        410
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                141,
+                                211,
+                                469,
+                                225
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        211,
+                                        469,
+                                        225
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Sparse neural networks have been shown to be more parameter and compute ef-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                223,
+                                469,
+                                236
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        223,
+                                        469,
+                                        236
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ficient compared to dense networks and in some cases they are even successfully",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                142,
+                                234,
+                                469,
+                                246
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        142,
+                                        234,
+                                        469,
+                                        246
+                                    ],
+                                    "score": 1.0,
+                                    "content": "used to decrease wall clock inference times. There is a large body of work on",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                245,
+                                469,
+                                257
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        245,
+                                        469,
+                                        257
+                                    ],
+                                    "score": 1.0,
+                                    "content": "training dense networks to yield sparse networks for inference (Molchanov et al.,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                255,
+                                469,
+                                268
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        255,
+                                        469,
+                                        268
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2017; Zhu & Gupta, 2018; Louizos et al., 2017; Li et al., 2016; Guo et al., 2016).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                266,
+                                470,
+                                280
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        266,
+                                        470,
+                                        280
+                                    ],
+                                    "score": 1.0,
+                                    "content": "This limits the size of the largest trainable sparse model to that of the largest train-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                277,
+                                470,
+                                290
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        277,
+                                        470,
+                                        290
+                                    ],
+                                    "score": 1.0,
+                                    "content": "able dense model. In this paper we introduce a method to train sparse neural",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                289,
+                                470,
+                                301
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        289,
+                                        470,
+                                        301
+                                    ],
+                                    "score": 1.0,
+                                    "content": "networks with a fixed parameter count and a fixed computational cost throughout",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                300,
+                                469,
+                                312
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        300,
+                                        469,
+                                        312
+                                    ],
+                                    "score": 1.0,
+                                    "content": "training, without sacrificing accuracy relative to existing dense-to-sparse training",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                310,
+                                469,
+                                324
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        310,
+                                        469,
+                                        324
+                                    ],
+                                    "score": 1.0,
+                                    "content": "methods. Our method updates the topology of the network during training by",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                321,
+                                470,
+                                334
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        321,
+                                        470,
+                                        334
+                                    ],
+                                    "score": 1.0,
+                                    "content": "using parameter magnitudes and infrequent gradient calculations. We show that",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                332,
+                                470,
+                                346
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        332,
+                                        470,
+                                        346
+                                    ],
+                                    "score": 1.0,
+                                    "content": "this approach requires fewer floating-point operations (FLOPs) to achieve a given",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                343,
+                                470,
+                                357
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        343,
+                                        470,
+                                        357
+                                    ],
+                                    "score": 1.0,
+                                    "content": "level of accuracy compared to prior techniques. We demonstrate state-of-the-art",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                355,
+                                470,
+                                366
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        355,
+                                        470,
+                                        366
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sparse training results with ResNet-50, MobileNet v1 and MobileNet v2 on the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                365,
+                                469,
+                                378
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        365,
+                                        469,
+                                        378
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ImageNet-2012 dataset, WideResNets on the CIFAR-10 dataset and RNNs on the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                376,
+                                469,
+                                389
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        376,
+                                        469,
+                                        389
+                                    ],
+                                    "score": 1.0,
+                                    "content": "WikiText-103 dataset. Finally, we provide some insights into why allowing the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                387,
+                                470,
+                                401
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        387,
+                                        470,
+                                        401
+                                    ],
+                                    "score": 1.0,
+                                    "content": "topology to change during the optimization can overcome local minima encoun-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                398,
+                                302,
+                                411
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        398,
+                                        302,
+                                        411
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tered when the topology remains static.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        }
+                    ],
+                    "index": 13.5,
+                    "bbox_fs": [
+                        141,
+                        211,
+                        470,
+                        411
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        432,
+                        206,
+                        444
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                430,
+                                208,
+                                447
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        430,
+                                        208,
+                                        447
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1 INTRODUCTION",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        }
+                    ],
+                    "index": 23
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        457,
+                        505,
+                        534
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                457,
+                                506,
+                                471
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        457,
+                                        506,
+                                        471
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The parameter and floating point operation (FLOP) efficiency of sparse neural networks is now well",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                467,
+                                506,
+                                481
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        467,
+                                        506,
+                                        481
+                                    ],
+                                    "score": 1.0,
+                                    "content": "demonstrated on a variety of problems (Han et al., 2015; Srinivas et al., 2017). Some work has even",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                479,
+                                505,
+                                491
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        479,
+                                        505,
+                                        491
+                                    ],
+                                    "score": 1.0,
+                                    "content": "shown inference time speedups are possible on Recurrent Neural Networks (RNNs) (Kalchbrenner",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                490,
+                                505,
+                                502
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        490,
+                                        505,
+                                        502
+                                    ],
+                                    "score": 1.0,
+                                    "content": "et al., 2018) and Convolutional Neural Networks (ConvNets) (Park et al., 2016). Currently, the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                501,
+                                506,
+                                513
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        501,
+                                        506,
+                                        513
+                                    ],
+                                    "score": 1.0,
+                                    "content": "most accurate sparse models are obtained with techniques that require, at a minimum, the cost of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                512,
+                                506,
+                                525
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        512,
+                                        506,
+                                        525
+                                    ],
+                                    "score": 1.0,
+                                    "content": "training a dense model in terms of memory and FLOPs (Zhu & Gupta, 2018; Guo et al., 2016), and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                523,
+                                495,
+                                535
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        523,
+                                        495,
+                                        535
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sometimes significantly more (Molchanov et al., 2017). This paradigm has two main limitations:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        }
+                    ],
+                    "index": 27,
+                    "bbox_fs": [
+                        105,
+                        457,
+                        506,
+                        535
+                    ]
+                },
+                {
+                    "type": "list",
+                    "bbox": [
+                        129,
+                        545,
+                        505,
+                        605
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                129,
+                                545,
+                                505,
+                                558
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        129,
+                                        545,
+                                        505,
+                                        558
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1. The maximum size of sparse models is limited to the largest dense model that can be",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                142,
+                                556,
+                                505,
+                                569
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        142,
+                                        556,
+                                        505,
+                                        569
+                                    ],
+                                    "score": 1.0,
+                                    "content": "trained. Even if sparse models are more parameter efficient, we can’t use pruning to train",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                142,
+                                568,
+                                464,
+                                580
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        142,
+                                        568,
+                                        464,
+                                        580
+                                    ],
+                                    "score": 1.0,
+                                    "content": "models that are larger and more accurate than the largest possible dense models.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                128,
+                                581,
+                                505,
+                                596
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        128,
+                                        581,
+                                        505,
+                                        596
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2. It is inefficient. Large amounts of computation must be performed for parameters that are",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                142,
+                                594,
+                                339,
+                                606
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        142,
+                                        594,
+                                        339,
+                                        606
+                                    ],
+                                    "score": 1.0,
+                                    "content": "zero valued or that will be zero during inference.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35,
+                            "is_list_end_line": true
+                        }
+                    ],
+                    "index": 33,
+                    "bbox_fs": [
+                        128,
+                        545,
+                        505,
+                        606
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        616,
+                        505,
+                        704
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                615,
+                                505,
+                                628
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        615,
+                                        505,
+                                        628
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Additionally, it remains unknown if the performance of the current best pruning algorithms are an",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                627,
+                                505,
+                                639
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        627,
+                                        505,
+                                        639
+                                    ],
+                                    "score": 1.0,
+                                    "content": "upper bound on the quality of sparse models. Gale et al. (2019) found that three different dense-to-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                638,
+                                505,
+                                650
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        638,
+                                        505,
+                                        650
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sparse training algorithms all achieve about the same sparsity / accuracy trade-off. However, this is",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                648,
+                                505,
+                                663
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        648,
+                                        505,
+                                        663
+                                    ],
+                                    "score": 1.0,
+                                    "content": "far from conclusive proof that no better performance is possible. In this work we show the surprising",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                660,
+                                505,
+                                672
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        660,
+                                        505,
+                                        672
+                                    ],
+                                    "score": 1.0,
+                                    "content": "result that dynamic sparse training, which includes the method we introduce below, can find more",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                670,
+                                505,
+                                684
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        670,
+                                        505,
+                                        684
+                                    ],
+                                    "score": 1.0,
+                                    "content": "accurate models than the current best approaches to pruning initially dense networks. Importantly,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                681,
+                                505,
+                                694
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        681,
+                                        505,
+                                        694
+                                    ],
+                                    "score": 1.0,
+                                    "content": "our method does not change the FLOPs required to execute the model during training, allowing one",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                692,
+                                323,
+                                706
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        692,
+                                        323,
+                                        706
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to decide on a specific inference cost prior to training.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        }
+                    ],
+                    "index": 39.5,
+                    "bbox_fs": [
+                        104,
+                        615,
+                        505,
+                        706
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        709,
+                        504,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                709,
+                                505,
+                                722
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        709,
+                                        505,
+                                        722
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The Lottery Ticket Hypothesis (Frankle & Carbin, 2019) hypothesized that if we can find a sparse",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                721,
+                                505,
+                                733
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        721,
+                                        505,
+                                        733
+                                    ],
+                                    "score": 1.0,
+                                    "content": "neural network with iterative pruning, then we can train that sparse network from scratch, to the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                83,
+                                506,
+                                95
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        83,
+                                        506,
+                                        95
+                                    ],
+                                    "score": 1.0,
+                                    "content": "same level of accuracy, by starting from the original initial conditions. In this paper we introduce a",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                93,
+                                506,
+                                107
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        93,
+                                        506,
+                                        107
+                                    ],
+                                    "score": 1.0,
+                                    "content": "new method for training sparse models without the need of a “lucky” initialization; for this reason,",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                104,
+                                433,
+                                117
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        104,
+                                        286,
+                                        117
+                                    ],
+                                    "score": 1.0,
+                                    "content": "we call our method “The Rigged Lottery” or",
+                                    "type": "text",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        286,
+                                        105,
+                                        311,
+                                        116
+                                    ],
+                                    "score": 0.86,
+                                    "content": "R i g L ^ { * }",
+                                    "type": "inline_equation",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        312,
+                                        104,
+                                        433,
+                                        117
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". We show that this method is:",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 2
+                        }
+                    ],
+                    "index": 44.5,
+                    "bbox_fs": [
+                        106,
+                        709,
+                        505,
+                        733
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        82,
+                        504,
+                        116
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                83,
+                                506,
+                                95
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        83,
+                                        506,
+                                        95
+                                    ],
+                                    "score": 1.0,
+                                    "content": "same level of accuracy, by starting from the original initial conditions. In this paper we introduce a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                93,
+                                506,
+                                107
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        93,
+                                        506,
+                                        107
+                                    ],
+                                    "score": 1.0,
+                                    "content": "new method for training sparse models without the need of a “lucky” initialization; for this reason,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                104,
+                                433,
+                                117
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        104,
+                                        286,
+                                        117
+                                    ],
+                                    "score": 1.0,
+                                    "content": "we call our method “The Rigged Lottery” or",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        286,
+                                        105,
+                                        311,
+                                        116
+                                    ],
+                                    "score": 0.86,
+                                    "content": "R i g L ^ { * }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        312,
+                                        104,
+                                        433,
+                                        117
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". We show that this method is:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        }
+                    ],
+                    "index": 1
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        132,
+                        127,
+                        504,
+                        224
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                132,
+                                127,
+                                505,
+                                140
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        132,
+                                        127,
+                                        505,
+                                        140
+                                    ],
+                                    "score": 1.0,
+                                    "content": "• Memory efficient: It requires memory only proportional to the size of the sparse model.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                138,
+                                505,
+                                149
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        138,
+                                        505,
+                                        149
+                                    ],
+                                    "score": 1.0,
+                                    "content": "It never requires storing quantities that are the size of the dense model. This is in contrast",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                149,
+                                505,
+                                162
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        149,
+                                        505,
+                                        162
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to Dettmers & Zettlemoyer (2019) which requires storing the momentum for all parameters,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                142,
+                                160,
+                                270,
+                                172
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        142,
+                                        160,
+                                        270,
+                                        172
+                                    ],
+                                    "score": 1.0,
+                                    "content": "even those that are zero valued.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                134,
+                                174,
+                                505,
+                                188
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        134,
+                                        174,
+                                        505,
+                                        188
+                                    ],
+                                    "score": 1.0,
+                                    "content": "• Computationally efficient: The amount of computation required to train the model is pro-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                186,
+                                384,
+                                198
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        186,
+                                        384,
+                                        198
+                                    ],
+                                    "score": 1.0,
+                                    "content": "portional to the number of nonzero parameters in the model.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                132,
+                                201,
+                                505,
+                                213
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        132,
+                                        201,
+                                        505,
+                                        213
+                                    ],
+                                    "score": 1.0,
+                                    "content": "• Accurate: The performance achieved by the method matches and sometimes exceeds the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                142,
+                                213,
+                                314,
+                                225
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        142,
+                                        213,
+                                        314,
+                                        225
+                                    ],
+                                    "score": 1.0,
+                                    "content": "performance of pruning based approaches.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        }
+                    ],
+                    "index": 6.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        235,
+                        505,
+                        291
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                233,
+                                505,
+                                248
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        233,
+                                        505,
+                                        248
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Our method works by infrequently using instantaneous gradient information to inform a re-wiring",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                246,
+                                505,
+                                258
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        246,
+                                        505,
+                                        258
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of the network. We show that this allows the optimization to escape local minima where it would",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                256,
+                                505,
+                                269
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        256,
+                                        505,
+                                        269
+                                    ],
+                                    "score": 1.0,
+                                    "content": "otherwise become trapped if the sparsity pattern were to remain static. Crucially, as long as the full",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                263,
+                                505,
+                                285
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        263,
+                                        295,
+                                        285
+                                    ],
+                                    "score": 1.0,
+                                    "content": "gradient information is needed less than every",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        295,
+                                        267,
+                                        339,
+                                        282
+                                    ],
+                                    "score": 0.94,
+                                    "content": "\\frac { 1 } { 1 - s p a r s i t y }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        339,
+                                        263,
+                                        505,
+                                        285
+                                    ],
+                                    "score": 1.0,
+                                    "content": "iterations, then the overall work remains",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                280,
+                                245,
+                                294
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        280,
+                                        245,
+                                        294
+                                    ],
+                                    "score": 1.0,
+                                    "content": "proportional to the model sparsity.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        }
+                    ],
+                    "index": 13
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        307,
+                        211,
+                        320
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                306,
+                                213,
+                                323
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        306,
+                                        213,
+                                        323
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2 RELATED WORK",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        }
+                    ],
+                    "index": 16
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        333,
+                        505,
+                        432
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                333,
+                                505,
+                                345
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        333,
+                                        505,
+                                        345
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Research on finding sparse neural networks dates back decades; for example, at least to Thimm &",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                344,
+                                505,
+                                357
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        344,
+                                        505,
+                                        357
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Fiesler (1995) who concluded that pruning weights based on magnitude was a simple and powerful",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                355,
+                                506,
+                                369
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        355,
+                                        506,
+                                        369
+                                    ],
+                                    "score": 1.0,
+                                    "content": "technique. Strom (1997) later introduced the idea of retraining the previously pruned network to ¨",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                367,
+                                505,
+                                378
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        367,
+                                        505,
+                                        378
+                                    ],
+                                    "score": 1.0,
+                                    "content": "increase accuracy. Han et al. (2016b) went further and introduced multiple rounds of magnitude",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                376,
+                                506,
+                                391
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        376,
+                                        506,
+                                        391
+                                    ],
+                                    "score": 1.0,
+                                    "content": "pruning and retraining. This is, however, relatively inefficient, requiring ten rounds of retraining",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                387,
+                                505,
+                                401
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        387,
+                                        171,
+                                        401
+                                    ],
+                                    "score": 1.0,
+                                    "content": "when removing",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        171,
+                                        388,
+                                        190,
+                                        399
+                                    ],
+                                    "score": 0.88,
+                                    "content": "2 0 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        191,
+                                        387,
+                                        372,
+                                        401
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of the connections to reach a final sparsity of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        372,
+                                        388,
+                                        392,
+                                        399
+                                    ],
+                                    "score": 0.89,
+                                    "content": "9 0 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        392,
+                                        387,
+                                        505,
+                                        401
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". To overcome this problem,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                399,
+                                505,
+                                412
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        399,
+                                        505,
+                                        412
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Narang et al. (2017) introduced gradual pruning, where connections are slowly removed over the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                410,
+                                505,
+                                423
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        410,
+                                        505,
+                                        423
+                                    ],
+                                    "score": 1.0,
+                                    "content": "course of a single round of training. Zhu & Gupta (2018) refined the technique to minimize the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                421,
+                                294,
+                                434
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        421,
+                                        294,
+                                        434
+                                    ],
+                                    "score": 1.0,
+                                    "content": "amount of hyper-parameter selection required.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        }
+                    ],
+                    "index": 21
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        438,
+                        505,
+                        525
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                438,
+                                505,
+                                451
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        438,
+                                        505,
+                                        451
+                                    ],
+                                    "score": 1.0,
+                                    "content": "A diversity of approaches not based on magnitude based pruning have also been proposed. LeCun",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                449,
+                                505,
+                                461
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        449,
+                                        505,
+                                        461
+                                    ],
+                                    "score": 1.0,
+                                    "content": "et al. (1990) and Hassibi & Stork (1993) are some early examples, but impractical for modern neural",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                460,
+                                505,
+                                472
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        460,
+                                        505,
+                                        472
+                                    ],
+                                    "score": 1.0,
+                                    "content": "networks as they use information from the Hessian to prune a trained network. More recent work",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                471,
+                                505,
+                                483
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        471,
+                                        142,
+                                        483
+                                    ],
+                                    "score": 1.0,
+                                    "content": "includes",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        142,
+                                        471,
+                                        154,
+                                        482
+                                    ],
+                                    "score": 0.91,
+                                    "content": "L _ { 0 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        155,
+                                        471,
+                                        505,
+                                        483
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Regularization (Christos Louizos, 2018), Variational Dropout (Molchanov et al., 2017),",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                482,
+                                504,
+                                494
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        482,
+                                        504,
+                                        494
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Dynamic Network Surgery (Guo et al., 2016) and Sensitivity Driven Regularization (Tartaglione",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                492,
+                                505,
+                                506
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        492,
+                                        364,
+                                        506
+                                    ],
+                                    "score": 1.0,
+                                    "content": "et al., 2018). Gale et al. (2019) examined magnitude pruning,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        364,
+                                        493,
+                                        377,
+                                        504
+                                    ],
+                                    "score": 0.87,
+                                    "content": "L _ { 0 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        378,
+                                        492,
+                                        505,
+                                        506
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Regularization and Variational",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                503,
+                                506,
+                                517
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        503,
+                                        506,
+                                        517
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Dropout and concluded that they all achieve about the same accuracy versus sparsity trade-off on",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                514,
+                                276,
+                                526
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        514,
+                                        276,
+                                        526
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ResNet-50 and Transformer architectures.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        }
+                    ],
+                    "index": 29.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        531,
+                        505,
+                        587
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                531,
+                                505,
+                                545
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        531,
+                                        505,
+                                        545
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Training techniques that allow for sparsity throughout the entire training process were, to our knowl-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                542,
+                                505,
+                                555
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        542,
+                                        505,
+                                        555
+                                    ],
+                                    "score": 1.0,
+                                    "content": "edge, first introduced in Deep Rewiring (DeepR) (Bellec et al., 2017). In DeepR, the standard",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                552,
+                                505,
+                                567
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        552,
+                                        505,
+                                        567
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Stochastic Gradient Descent (SGD) optimizer is augmented with a random walk in parameter space.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                564,
+                                505,
+                                578
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        564,
+                                        505,
+                                        578
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Additionally, connections have a pre-defined sign assigned at random; when the optimizer would",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                575,
+                                479,
+                                588
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        575,
+                                        479,
+                                        588
+                                    ],
+                                    "score": 1.0,
+                                    "content": "normally flip the sign, the weight is set to 0 instead and new weights are activated at random.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        }
+                    ],
+                    "index": 36
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        592,
+                        505,
+                        713
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                592,
+                                505,
+                                605
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        592,
+                                        505,
+                                        605
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Sparse Evolutionary Training (SET) (Mocanu et al., 2018) proposed a simpler scheme where weights",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                604,
+                                506,
+                                616
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        604,
+                                        506,
+                                        616
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are pruned according to the standard magnitude criterion used in pruning and are added back at",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                613,
+                                506,
+                                627
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        613,
+                                        506,
+                                        627
+                                    ],
+                                    "score": 1.0,
+                                    "content": "random. The method is simple and achieves reasonable performance in practice. Dynamic Sparse",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                625,
+                                505,
+                                638
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        625,
+                                        505,
+                                        638
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Reparameterization (DSR) (Mostafa & Wang, 2019) introduced the idea of allowing the parameter",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                635,
+                                505,
+                                649
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        635,
+                                        505,
+                                        649
+                                    ],
+                                    "score": 1.0,
+                                    "content": "budget to shift between different layers of the model, allowing for non-uniform sparsity. This allows",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                647,
+                                505,
+                                660
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        647,
+                                        505,
+                                        660
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the model to distribute parameters where they are most effective. Unfortunately, the models under",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                658,
+                                505,
+                                671
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        658,
+                                        505,
+                                        671
+                                    ],
+                                    "score": 1.0,
+                                    "content": "consideration are mostly convolutional networks, so the result of this parameter reallocation (which",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                669,
+                                505,
+                                682
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        669,
+                                        505,
+                                        682
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is to decrease the sparsity of early layers and increase the sparsity of later layers) has the overall",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                680,
+                                505,
+                                693
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        680,
+                                        505,
+                                        693
+                                    ],
+                                    "score": 1.0,
+                                    "content": "effect of increasing the FLOP count because the spatial size is largest at the beginning. Sparse",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 47
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                691,
+                                505,
+                                704
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        691,
+                                        505,
+                                        704
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Networks from Scratch (SNFS) (Dettmers & Zettlemoyer, 2019) introduces the idea of using the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 48
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                703,
+                                506,
+                                714
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        703,
+                                        506,
+                                        714
+                                    ],
+                                    "score": 1.0,
+                                    "content": "momentum of each parameter as the criterion to be used for growing weights and demonstrates it",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 49
+                        }
+                    ],
+                    "index": 44
+                }
+            ],
+            "page_idx": 1,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        121,
+                        722,
+                        208,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                120,
+                                720,
+                                209,
+                                733
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        120,
+                                        720,
+                                        209,
+                                        733
+                                    ],
+                                    "score": 1.0,
+                                    "content": "∗Pronounced ”wriggle”.",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        107,
+                        27,
+                        308,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                25,
+                                308,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        25,
+                                        308,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Under review as a conference paper at ICLR 2020",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        302,
+                        751,
+                        309,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                301,
+                                750,
+                                310,
+                                763
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        301,
+                                        750,
+                                        310,
+                                        763
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        82,
+                        504,
+                        116
+                    ],
+                    "lines": [],
+                    "index": 1,
+                    "bbox_fs": [
+                        105,
+                        83,
+                        506,
+                        117
+                    ],
+                    "lines_deleted": true
+                },
+                {
+                    "type": "list",
+                    "bbox": [
+                        132,
+                        127,
+                        504,
+                        224
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                132,
+                                127,
+                                505,
+                                140
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        132,
+                                        127,
+                                        505,
+                                        140
+                                    ],
+                                    "score": 1.0,
+                                    "content": "• Memory efficient: It requires memory only proportional to the size of the sparse model.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                138,
+                                505,
+                                149
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        138,
+                                        505,
+                                        149
+                                    ],
+                                    "score": 1.0,
+                                    "content": "It never requires storing quantities that are the size of the dense model. This is in contrast",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                149,
+                                505,
+                                162
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        149,
+                                        505,
+                                        162
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to Dettmers & Zettlemoyer (2019) which requires storing the momentum for all parameters,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                142,
+                                160,
+                                270,
+                                172
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        142,
+                                        160,
+                                        270,
+                                        172
+                                    ],
+                                    "score": 1.0,
+                                    "content": "even those that are zero valued.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                134,
+                                174,
+                                505,
+                                188
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        134,
+                                        174,
+                                        505,
+                                        188
+                                    ],
+                                    "score": 1.0,
+                                    "content": "• Computationally efficient: The amount of computation required to train the model is pro-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                186,
+                                384,
+                                198
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        186,
+                                        384,
+                                        198
+                                    ],
+                                    "score": 1.0,
+                                    "content": "portional to the number of nonzero parameters in the model.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                132,
+                                201,
+                                505,
+                                213
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        132,
+                                        201,
+                                        505,
+                                        213
+                                    ],
+                                    "score": 1.0,
+                                    "content": "• Accurate: The performance achieved by the method matches and sometimes exceeds the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                142,
+                                213,
+                                314,
+                                225
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        142,
+                                        213,
+                                        314,
+                                        225
+                                    ],
+                                    "score": 1.0,
+                                    "content": "performance of pruning based approaches.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10,
+                            "is_list_end_line": true
+                        }
+                    ],
+                    "index": 6.5,
+                    "bbox_fs": [
+                        132,
+                        127,
+                        505,
+                        225
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        235,
+                        505,
+                        291
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                233,
+                                505,
+                                248
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        233,
+                                        505,
+                                        248
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Our method works by infrequently using instantaneous gradient information to inform a re-wiring",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                246,
+                                505,
+                                258
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        246,
+                                        505,
+                                        258
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of the network. We show that this allows the optimization to escape local minima where it would",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                256,
+                                505,
+                                269
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        256,
+                                        505,
+                                        269
+                                    ],
+                                    "score": 1.0,
+                                    "content": "otherwise become trapped if the sparsity pattern were to remain static. Crucially, as long as the full",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                263,
+                                505,
+                                285
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        263,
+                                        295,
+                                        285
+                                    ],
+                                    "score": 1.0,
+                                    "content": "gradient information is needed less than every",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        295,
+                                        267,
+                                        339,
+                                        282
+                                    ],
+                                    "score": 0.94,
+                                    "content": "\\frac { 1 } { 1 - s p a r s i t y }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        339,
+                                        263,
+                                        505,
+                                        285
+                                    ],
+                                    "score": 1.0,
+                                    "content": "iterations, then the overall work remains",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                280,
+                                245,
+                                294
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        280,
+                                        245,
+                                        294
+                                    ],
+                                    "score": 1.0,
+                                    "content": "proportional to the model sparsity.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        }
+                    ],
+                    "index": 13,
+                    "bbox_fs": [
+                        104,
+                        233,
+                        505,
+                        294
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        307,
+                        211,
+                        320
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                306,
+                                213,
+                                323
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        306,
+                                        213,
+                                        323
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2 RELATED WORK",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        }
+                    ],
+                    "index": 16
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        333,
+                        505,
+                        432
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                333,
+                                505,
+                                345
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        333,
+                                        505,
+                                        345
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Research on finding sparse neural networks dates back decades; for example, at least to Thimm &",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                344,
+                                505,
+                                357
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        344,
+                                        505,
+                                        357
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Fiesler (1995) who concluded that pruning weights based on magnitude was a simple and powerful",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                355,
+                                506,
+                                369
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        355,
+                                        506,
+                                        369
+                                    ],
+                                    "score": 1.0,
+                                    "content": "technique. Strom (1997) later introduced the idea of retraining the previously pruned network to ¨",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                367,
+                                505,
+                                378
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        367,
+                                        505,
+                                        378
+                                    ],
+                                    "score": 1.0,
+                                    "content": "increase accuracy. Han et al. (2016b) went further and introduced multiple rounds of magnitude",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                376,
+                                506,
+                                391
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        376,
+                                        506,
+                                        391
+                                    ],
+                                    "score": 1.0,
+                                    "content": "pruning and retraining. This is, however, relatively inefficient, requiring ten rounds of retraining",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                387,
+                                505,
+                                401
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        387,
+                                        171,
+                                        401
+                                    ],
+                                    "score": 1.0,
+                                    "content": "when removing",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        171,
+                                        388,
+                                        190,
+                                        399
+                                    ],
+                                    "score": 0.88,
+                                    "content": "2 0 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        191,
+                                        387,
+                                        372,
+                                        401
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of the connections to reach a final sparsity of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        372,
+                                        388,
+                                        392,
+                                        399
+                                    ],
+                                    "score": 0.89,
+                                    "content": "9 0 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        392,
+                                        387,
+                                        505,
+                                        401
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". To overcome this problem,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                399,
+                                505,
+                                412
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        399,
+                                        505,
+                                        412
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Narang et al. (2017) introduced gradual pruning, where connections are slowly removed over the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                410,
+                                505,
+                                423
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        410,
+                                        505,
+                                        423
+                                    ],
+                                    "score": 1.0,
+                                    "content": "course of a single round of training. Zhu & Gupta (2018) refined the technique to minimize the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                421,
+                                294,
+                                434
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        421,
+                                        294,
+                                        434
+                                    ],
+                                    "score": 1.0,
+                                    "content": "amount of hyper-parameter selection required.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        }
+                    ],
+                    "index": 21,
+                    "bbox_fs": [
+                        105,
+                        333,
+                        506,
+                        434
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        438,
+                        505,
+                        525
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                438,
+                                505,
+                                451
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        438,
+                                        505,
+                                        451
+                                    ],
+                                    "score": 1.0,
+                                    "content": "A diversity of approaches not based on magnitude based pruning have also been proposed. LeCun",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                449,
+                                505,
+                                461
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        449,
+                                        505,
+                                        461
+                                    ],
+                                    "score": 1.0,
+                                    "content": "et al. (1990) and Hassibi & Stork (1993) are some early examples, but impractical for modern neural",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                460,
+                                505,
+                                472
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        460,
+                                        505,
+                                        472
+                                    ],
+                                    "score": 1.0,
+                                    "content": "networks as they use information from the Hessian to prune a trained network. More recent work",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                471,
+                                505,
+                                483
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        471,
+                                        142,
+                                        483
+                                    ],
+                                    "score": 1.0,
+                                    "content": "includes",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        142,
+                                        471,
+                                        154,
+                                        482
+                                    ],
+                                    "score": 0.91,
+                                    "content": "L _ { 0 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        155,
+                                        471,
+                                        505,
+                                        483
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Regularization (Christos Louizos, 2018), Variational Dropout (Molchanov et al., 2017),",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                482,
+                                504,
+                                494
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        482,
+                                        504,
+                                        494
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Dynamic Network Surgery (Guo et al., 2016) and Sensitivity Driven Regularization (Tartaglione",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                492,
+                                505,
+                                506
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        492,
+                                        364,
+                                        506
+                                    ],
+                                    "score": 1.0,
+                                    "content": "et al., 2018). Gale et al. (2019) examined magnitude pruning,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        364,
+                                        493,
+                                        377,
+                                        504
+                                    ],
+                                    "score": 0.87,
+                                    "content": "L _ { 0 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        378,
+                                        492,
+                                        505,
+                                        506
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Regularization and Variational",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                503,
+                                506,
+                                517
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        503,
+                                        506,
+                                        517
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Dropout and concluded that they all achieve about the same accuracy versus sparsity trade-off on",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                514,
+                                276,
+                                526
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        514,
+                                        276,
+                                        526
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ResNet-50 and Transformer architectures.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        }
+                    ],
+                    "index": 29.5,
+                    "bbox_fs": [
+                        104,
+                        438,
+                        506,
+                        526
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        531,
+                        505,
+                        587
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                531,
+                                505,
+                                545
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        531,
+                                        505,
+                                        545
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Training techniques that allow for sparsity throughout the entire training process were, to our knowl-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                542,
+                                505,
+                                555
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        542,
+                                        505,
+                                        555
+                                    ],
+                                    "score": 1.0,
+                                    "content": "edge, first introduced in Deep Rewiring (DeepR) (Bellec et al., 2017). In DeepR, the standard",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                552,
+                                505,
+                                567
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        552,
+                                        505,
+                                        567
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Stochastic Gradient Descent (SGD) optimizer is augmented with a random walk in parameter space.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                564,
+                                505,
+                                578
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        564,
+                                        505,
+                                        578
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Additionally, connections have a pre-defined sign assigned at random; when the optimizer would",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                575,
+                                479,
+                                588
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        575,
+                                        479,
+                                        588
+                                    ],
+                                    "score": 1.0,
+                                    "content": "normally flip the sign, the weight is set to 0 instead and new weights are activated at random.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        }
+                    ],
+                    "index": 36,
+                    "bbox_fs": [
+                        105,
+                        531,
+                        505,
+                        588
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        592,
+                        505,
+                        713
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                592,
+                                505,
+                                605
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        592,
+                                        505,
+                                        605
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Sparse Evolutionary Training (SET) (Mocanu et al., 2018) proposed a simpler scheme where weights",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                604,
+                                506,
+                                616
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        604,
+                                        506,
+                                        616
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are pruned according to the standard magnitude criterion used in pruning and are added back at",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                613,
+                                506,
+                                627
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        613,
+                                        506,
+                                        627
+                                    ],
+                                    "score": 1.0,
+                                    "content": "random. The method is simple and achieves reasonable performance in practice. Dynamic Sparse",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                625,
+                                505,
+                                638
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        625,
+                                        505,
+                                        638
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Reparameterization (DSR) (Mostafa & Wang, 2019) introduced the idea of allowing the parameter",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                635,
+                                505,
+                                649
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        635,
+                                        505,
+                                        649
+                                    ],
+                                    "score": 1.0,
+                                    "content": "budget to shift between different layers of the model, allowing for non-uniform sparsity. This allows",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                647,
+                                505,
+                                660
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        647,
+                                        505,
+                                        660
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the model to distribute parameters where they are most effective. Unfortunately, the models under",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                658,
+                                505,
+                                671
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        658,
+                                        505,
+                                        671
+                                    ],
+                                    "score": 1.0,
+                                    "content": "consideration are mostly convolutional networks, so the result of this parameter reallocation (which",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                669,
+                                505,
+                                682
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        669,
+                                        505,
+                                        682
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is to decrease the sparsity of early layers and increase the sparsity of later layers) has the overall",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                680,
+                                505,
+                                693
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        680,
+                                        505,
+                                        693
+                                    ],
+                                    "score": 1.0,
+                                    "content": "effect of increasing the FLOP count because the spatial size is largest at the beginning. Sparse",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 47
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                691,
+                                505,
+                                704
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        691,
+                                        505,
+                                        704
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Networks from Scratch (SNFS) (Dettmers & Zettlemoyer, 2019) introduces the idea of using the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 48
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                703,
+                                506,
+                                714
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        703,
+                                        506,
+                                        714
+                                    ],
+                                    "score": 1.0,
+                                    "content": "momentum of each parameter as the criterion to be used for growing weights and demonstrates it",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 49
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                224,
+                                505,
+                                238
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        224,
+                                        505,
+                                        238
+                                    ],
+                                    "score": 1.0,
+                                    "content": "leads to an improvement in test accuracy. Like DSR, they allow the sparsity of each layer to change",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                234,
+                                505,
+                                250
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        234,
+                                        505,
+                                        250
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and focus on a constant parameter, not FLOP, budget. Importantly, the method requires computing",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                245,
+                                505,
+                                260
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        245,
+                                        505,
+                                        260
+                                    ],
+                                    "score": 1.0,
+                                    "content": "gradients and updating the momentum for every parameter in the model, even those that are zero,",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                258,
+                                504,
+                                270
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        258,
+                                        504,
+                                        270
+                                    ],
+                                    "score": 1.0,
+                                    "content": "at every iteration. This can result in a significant amount of overall computation. Additionally,",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                269,
+                                505,
+                                282
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        269,
+                                        505,
+                                        282
+                                    ],
+                                    "score": 1.0,
+                                    "content": "depending on the model and training setup, the required storage for the full momentum tensor could",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                279,
+                                505,
+                                292
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        279,
+                                        505,
+                                        292
+                                    ],
+                                    "score": 1.0,
+                                    "content": "be prohibitive. Single-Shot Network Pruning (SNIP) (Lee et al., 2019) attempts to find an initial",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                290,
+                                505,
+                                304
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        290,
+                                        505,
+                                        304
+                                    ],
+                                    "score": 1.0,
+                                    "content": "mask with one-shot pruning and uses the saliency score of parameters to decide which parameters",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                301,
+                                505,
+                                315
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        301,
+                                        505,
+                                        315
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to keep. After pruning training proceeds with this static sparse network. Properties of the different",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                313,
+                                323,
+                                325
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        313,
+                                        323,
+                                        325
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sparse training techniques are summarized in Table 1.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 14
+                        }
+                    ],
+                    "index": 44,
+                    "bbox_fs": [
+                        105,
+                        592,
+                        506,
+                        714
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "table",
+                    "bbox": [
+                        115,
+                        79,
+                        496,
+                        160
+                    ],
+                    "blocks": [
+                        {
+                            "type": "table_body",
+                            "bbox": [
+                                115,
+                                79,
+                                496,
+                                160
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        79,
+                                        496,
+                                        160
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                115,
+                                                79,
+                                                496,
+                                                160
+                                            ],
+                                            "score": 0.975,
+                                            "html": "<table><tr><td>Method</td><td>Drop</td><td>Grow</td><td>Selectable FLOPs</td><td>Space &amp;FLOPs X</td></tr><tr><td>SNIP</td><td>min(10 * VL(0)I)</td><td>none</td><td>yes</td><td>sparse</td></tr><tr><td>DeepR</td><td>stochastic</td><td>random</td><td>yes</td><td>sparse</td></tr><tr><td>SET</td><td>min(l0))</td><td>random</td><td>yes</td><td>sparse</td></tr><tr><td>DSR</td><td>min(le)</td><td>random</td><td>no</td><td>sparse</td></tr><tr><td>SNFS</td><td>min(l0))</td><td>momentum</td><td>no</td><td>dense</td></tr><tr><td>RigL (ours)</td><td>min(l0)</td><td>gradient</td><td>yes</td><td>sparse</td></tr></table>",
+                                            "type": "table",
+                                            "image_path": "2dd036930c9611b0039842e97192636016f15c49e7d59ff72eef1af9e125a6c0.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 1,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        79,
+                                        496,
+                                        106.0
+                                    ],
+                                    "spans": [],
+                                    "index": 0
+                                },
+                                {
+                                    "bbox": [
+                                        115,
+                                        106.0,
+                                        496,
+                                        133.0
+                                    ],
+                                    "spans": [],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        115,
+                                        133.0,
+                                        496,
+                                        160.0
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                }
+                            ]
+                        },
+                        {
+                            "type": "table_caption",
+                            "bbox": [
+                                107,
+                                167,
+                                505,
+                                201
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        166,
+                                        505,
+                                        181
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                166,
+                                                505,
+                                                181
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Table 1: Comparison of different sparse training techniques. Drop and Grow columns correspond to",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 3
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        178,
+                                        505,
+                                        191
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                178,
+                                                326,
+                                                191
+                                            ],
+                                            "score": 1.0,
+                                            "content": "the strategies used during the mask update. Selectable",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                326,
+                                                179,
+                                                357,
+                                                189
+                                            ],
+                                            "score": 0.41,
+                                            "content": "F L O P s",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                358,
+                                                178,
+                                                505,
+                                                191
+                                            ],
+                                            "score": 1.0,
+                                            "content": "is possible if the cost of training the",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 4
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        189,
+                                        279,
+                                        202
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                189,
+                                                279,
+                                                202
+                                            ],
+                                            "score": 1.0,
+                                            "content": "model is fixed at the beginning of training.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 5
+                                }
+                            ],
+                            "index": 4
+                        }
+                    ],
+                    "index": 2.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        224,
+                        505,
+                        324
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                224,
+                                505,
+                                238
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        224,
+                                        505,
+                                        238
+                                    ],
+                                    "score": 1.0,
+                                    "content": "leads to an improvement in test accuracy. Like DSR, they allow the sparsity of each layer to change",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                234,
+                                505,
+                                250
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        234,
+                                        505,
+                                        250
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and focus on a constant parameter, not FLOP, budget. Importantly, the method requires computing",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                245,
+                                505,
+                                260
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        245,
+                                        505,
+                                        260
+                                    ],
+                                    "score": 1.0,
+                                    "content": "gradients and updating the momentum for every parameter in the model, even those that are zero,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                258,
+                                504,
+                                270
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        258,
+                                        504,
+                                        270
+                                    ],
+                                    "score": 1.0,
+                                    "content": "at every iteration. This can result in a significant amount of overall computation. Additionally,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                269,
+                                505,
+                                282
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        269,
+                                        505,
+                                        282
+                                    ],
+                                    "score": 1.0,
+                                    "content": "depending on the model and training setup, the required storage for the full momentum tensor could",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                279,
+                                505,
+                                292
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        279,
+                                        505,
+                                        292
+                                    ],
+                                    "score": 1.0,
+                                    "content": "be prohibitive. Single-Shot Network Pruning (SNIP) (Lee et al., 2019) attempts to find an initial",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                290,
+                                505,
+                                304
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        290,
+                                        505,
+                                        304
+                                    ],
+                                    "score": 1.0,
+                                    "content": "mask with one-shot pruning and uses the saliency score of parameters to decide which parameters",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                301,
+                                505,
+                                315
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        301,
+                                        505,
+                                        315
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to keep. After pruning training proceeds with this static sparse network. Properties of the different",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                313,
+                                323,
+                                325
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        313,
+                                        323,
+                                        325
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sparse training techniques are summarized in Table 1.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        }
+                    ],
+                    "index": 10
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        329,
+                        505,
+                        417
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                329,
+                                505,
+                                342
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        329,
+                                        505,
+                                        342
+                                    ],
+                                    "score": 1.0,
+                                    "content": "There has also been a line of work investigating the Lottery Ticket Hypothesis (Frankle & Carbin,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                339,
+                                505,
+                                352
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        339,
+                                        505,
+                                        352
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2019). Frankle et al. (2019) showed that the formulation must be weakened to apply to larger",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                351,
+                                505,
+                                364
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        351,
+                                        505,
+                                        364
+                                    ],
+                                    "score": 1.0,
+                                    "content": "networks such as ResNet-50 (He et al., 2015). In large networks, instead of the original initial-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                362,
+                                505,
+                                375
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        362,
+                                        505,
+                                        375
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ization, the values after thousands of optimization steps must be used for initialization. Zhou et al.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                373,
+                                504,
+                                385
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        373,
+                                        504,
+                                        385
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(2019) showed that lottery tickets obtain non-random accuracies even before the training has started.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                383,
+                                505,
+                                398
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        383,
+                                        505,
+                                        398
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Though the possibility of training sparse neural networks with a fixed sparsity mask using lottery",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                395,
+                                505,
+                                407
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        395,
+                                        505,
+                                        407
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tickets is intriguing, it remains unclear whether it is possible to generate such initializations – for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                406,
+                                261,
+                                418
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        406,
+                                        261,
+                                        418
+                                    ],
+                                    "score": 1.0,
+                                    "content": "both masks and parameters – de novo.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        }
+                    ],
+                    "index": 18.5
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        436,
+                        248,
+                        449
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                434,
+                                250,
+                                452
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        434,
+                                        250,
+                                        452
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3 RIGGING THE LOTTERY",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        }
+                    ],
+                    "index": 23
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        462,
+                        505,
+                        562
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                463,
+                                505,
+                                475
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        463,
+                                        505,
+                                        475
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Our method, RigL, is illustrated in Figure 1. At regularly spaced intervals our method removes",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                473,
+                                505,
+                                487
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        473,
+                                        505,
+                                        487
+                                    ],
+                                    "score": 1.0,
+                                    "content": "a fraction of connections based on weight magnitudes and activates new ones using instantaneous",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                484,
+                                506,
+                                498
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        484,
+                                        506,
+                                        498
+                                    ],
+                                    "score": 1.0,
+                                    "content": "gradient information. After updating the connectivity, training continues with the updated network",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                496,
+                                505,
+                                508
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        496,
+                                        505,
+                                        508
+                                    ],
+                                    "score": 1.0,
+                                    "content": "until the next update. The main parts of our algorithm, Sparsity Distribution, Update Schedule,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                506,
+                                505,
+                                519
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        506,
+                                        505,
+                                        519
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Drop Criterion, Grow Criterion, and the various options we considered for each, are explained",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                518,
+                                506,
+                                531
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        518,
+                                        265,
+                                        531
+                                    ],
+                                    "score": 1.0,
+                                    "content": "below. The improved performance of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        265,
+                                        518,
+                                        287,
+                                        529
+                                    ],
+                                    "score": 0.43,
+                                    "content": "R i g L",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        287,
+                                        518,
+                                        506,
+                                        531
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is due to two reasons: the use of a new method for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                528,
+                                506,
+                                541
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        528,
+                                        506,
+                                        541
+                                    ],
+                                    "score": 1.0,
+                                    "content": "activating connections that is efficient and more effective than choosing at random, and the use of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                540,
+                                505,
+                                553
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        540,
+                                        505,
+                                        553
+                                    ],
+                                    "score": 1.0,
+                                    "content": "a natural extension to an existing method for distributing parameters statically among convolutional",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                550,
+                                137,
+                                565
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        550,
+                                        137,
+                                        565
+                                    ],
+                                    "score": 1.0,
+                                    "content": "layers.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        }
+                    ],
+                    "index": 28
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        567,
+                        504,
+                        591
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                566,
+                                505,
+                                580
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        566,
+                                        234,
+                                        580
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(0) Notation. Given a dataset",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        234,
+                                        568,
+                                        244,
+                                        577
+                                    ],
+                                    "score": 0.76,
+                                    "content": "D",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        244,
+                                        566,
+                                        338,
+                                        580
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with individual inputs",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        338,
+                                        569,
+                                        348,
+                                        579
+                                    ],
+                                    "score": 0.85,
+                                    "content": "x _ { i }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        349,
+                                        566,
+                                        398,
+                                        580
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and targets",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        398,
+                                        570,
+                                        407,
+                                        579
+                                    ],
+                                    "score": 0.84,
+                                    "content": "y _ { i }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        408,
+                                        566,
+                                        505,
+                                        580
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", one can train a neural",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                578,
+                                505,
+                                591
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        578,
+                                        266,
+                                        591
+                                    ],
+                                    "score": 1.0,
+                                    "content": "network to minimize the loss function",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        266,
+                                        578,
+                                        335,
+                                        591
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\textstyle \\sum _ { i } L ( f _ { \\theta } ( x _ { i } ) , y _ { i } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        336,
+                                        578,
+                                        368,
+                                        591
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", where",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        369,
+                                        579,
+                                        393,
+                                        591
+                                    ],
+                                    "score": 0.92,
+                                    "content": "f _ { \\boldsymbol { \\theta } } ( \\boldsymbol { x } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        393,
+                                        578,
+                                        505,
+                                        591
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is the neural network with",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        }
+                    ],
+                    "index": 33.5
+                },
+                {
+                    "type": "image",
+                    "bbox": [
+                        125,
+                        614,
+                        486,
+                        713
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                125,
+                                614,
+                                486,
+                                713
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        125,
+                                        614,
+                                        486,
+                                        713
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                125,
+                                                614,
+                                                486,
+                                                713
+                                            ],
+                                            "score": 0.969,
+                                            "type": "image",
+                                            "image_path": "c4f8c255dd49151f0525882b1c8f56ef8e1197a1ff58bb53e4c69623eb1ec0b1.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 36,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        125,
+                                        614,
+                                        486,
+                                        647.0
+                                    ],
+                                    "spans": [],
+                                    "index": 35
+                                },
+                                {
+                                    "bbox": [
+                                        125,
+                                        647.0,
+                                        486,
+                                        680.0
+                                    ],
+                                    "spans": [],
+                                    "index": 36
+                                },
+                                {
+                                    "bbox": [
+                                        125,
+                                        680.0,
+                                        486,
+                                        713.0
+                                    ],
+                                    "spans": [],
+                                    "index": 37
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                105,
+                                721,
+                                505,
+                                743
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        719,
+                                        504,
+                                        734
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                719,
+                                                504,
+                                                734
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 1: Dynamic sparse training aims to change connectivity during training to help out optimza-",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 38
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        731,
+                                        129,
+                                        744
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                731,
+                                                129,
+                                                744
+                                            ],
+                                            "score": 1.0,
+                                            "content": "tion.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 39
+                                }
+                            ],
+                            "index": 38.5
+                        }
+                    ],
+                    "index": 37.25
+                }
+            ],
+            "page_idx": 2,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        107,
+                        27,
+                        308,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                107,
+                                26,
+                                308,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        26,
+                                        308,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Under review as a conference paper at ICLR 2020",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        302,
+                        751,
+                        309,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                301,
+                                750,
+                                310,
+                                762
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        301,
+                                        750,
+                                        310,
+                                        762
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "table",
+                    "bbox": [
+                        115,
+                        79,
+                        496,
+                        160
+                    ],
+                    "blocks": [
+                        {
+                            "type": "table_body",
+                            "bbox": [
+                                115,
+                                79,
+                                496,
+                                160
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        79,
+                                        496,
+                                        160
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                115,
+                                                79,
+                                                496,
+                                                160
+                                            ],
+                                            "score": 0.975,
+                                            "html": "<table><tr><td>Method</td><td>Drop</td><td>Grow</td><td>Selectable FLOPs</td><td>Space &amp;FLOPs X</td></tr><tr><td>SNIP</td><td>min(10 * VL(0)I)</td><td>none</td><td>yes</td><td>sparse</td></tr><tr><td>DeepR</td><td>stochastic</td><td>random</td><td>yes</td><td>sparse</td></tr><tr><td>SET</td><td>min(l0))</td><td>random</td><td>yes</td><td>sparse</td></tr><tr><td>DSR</td><td>min(le)</td><td>random</td><td>no</td><td>sparse</td></tr><tr><td>SNFS</td><td>min(l0))</td><td>momentum</td><td>no</td><td>dense</td></tr><tr><td>RigL (ours)</td><td>min(l0)</td><td>gradient</td><td>yes</td><td>sparse</td></tr></table>",
+                                            "type": "table",
+                                            "image_path": "2dd036930c9611b0039842e97192636016f15c49e7d59ff72eef1af9e125a6c0.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 1,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        79,
+                                        496,
+                                        106.0
+                                    ],
+                                    "spans": [],
+                                    "index": 0
+                                },
+                                {
+                                    "bbox": [
+                                        115,
+                                        106.0,
+                                        496,
+                                        133.0
+                                    ],
+                                    "spans": [],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        115,
+                                        133.0,
+                                        496,
+                                        160.0
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                }
+                            ]
+                        },
+                        {
+                            "type": "table_caption",
+                            "bbox": [
+                                107,
+                                167,
+                                505,
+                                201
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        166,
+                                        505,
+                                        181
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                166,
+                                                505,
+                                                181
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Table 1: Comparison of different sparse training techniques. Drop and Grow columns correspond to",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 3
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        178,
+                                        505,
+                                        191
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                178,
+                                                326,
+                                                191
+                                            ],
+                                            "score": 1.0,
+                                            "content": "the strategies used during the mask update. Selectable",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                326,
+                                                179,
+                                                357,
+                                                189
+                                            ],
+                                            "score": 0.41,
+                                            "content": "F L O P s",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                358,
+                                                178,
+                                                505,
+                                                191
+                                            ],
+                                            "score": 1.0,
+                                            "content": "is possible if the cost of training the",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 4
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        189,
+                                        279,
+                                        202
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                189,
+                                                279,
+                                                202
+                                            ],
+                                            "score": 1.0,
+                                            "content": "model is fixed at the beginning of training.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 5
+                                }
+                            ],
+                            "index": 4
+                        }
+                    ],
+                    "index": 2.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        224,
+                        505,
+                        324
+                    ],
+                    "lines": [],
+                    "index": 10,
+                    "bbox_fs": [
+                        105,
+                        224,
+                        505,
+                        325
+                    ],
+                    "lines_deleted": true
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        329,
+                        505,
+                        417
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                329,
+                                505,
+                                342
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        329,
+                                        505,
+                                        342
+                                    ],
+                                    "score": 1.0,
+                                    "content": "There has also been a line of work investigating the Lottery Ticket Hypothesis (Frankle & Carbin,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                339,
+                                505,
+                                352
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        339,
+                                        505,
+                                        352
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2019). Frankle et al. (2019) showed that the formulation must be weakened to apply to larger",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                351,
+                                505,
+                                364
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        351,
+                                        505,
+                                        364
+                                    ],
+                                    "score": 1.0,
+                                    "content": "networks such as ResNet-50 (He et al., 2015). In large networks, instead of the original initial-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                362,
+                                505,
+                                375
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        362,
+                                        505,
+                                        375
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ization, the values after thousands of optimization steps must be used for initialization. Zhou et al.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                373,
+                                504,
+                                385
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        373,
+                                        504,
+                                        385
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(2019) showed that lottery tickets obtain non-random accuracies even before the training has started.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                383,
+                                505,
+                                398
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        383,
+                                        505,
+                                        398
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Though the possibility of training sparse neural networks with a fixed sparsity mask using lottery",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                395,
+                                505,
+                                407
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        395,
+                                        505,
+                                        407
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tickets is intriguing, it remains unclear whether it is possible to generate such initializations – for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                406,
+                                261,
+                                418
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        406,
+                                        261,
+                                        418
+                                    ],
+                                    "score": 1.0,
+                                    "content": "both masks and parameters – de novo.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        }
+                    ],
+                    "index": 18.5,
+                    "bbox_fs": [
+                        105,
+                        329,
+                        505,
+                        418
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        436,
+                        248,
+                        449
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                434,
+                                250,
+                                452
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        434,
+                                        250,
+                                        452
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3 RIGGING THE LOTTERY",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        }
+                    ],
+                    "index": 23
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        462,
+                        505,
+                        562
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                463,
+                                505,
+                                475
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        463,
+                                        505,
+                                        475
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Our method, RigL, is illustrated in Figure 1. At regularly spaced intervals our method removes",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                473,
+                                505,
+                                487
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        473,
+                                        505,
+                                        487
+                                    ],
+                                    "score": 1.0,
+                                    "content": "a fraction of connections based on weight magnitudes and activates new ones using instantaneous",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                484,
+                                506,
+                                498
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        484,
+                                        506,
+                                        498
+                                    ],
+                                    "score": 1.0,
+                                    "content": "gradient information. After updating the connectivity, training continues with the updated network",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                496,
+                                505,
+                                508
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        496,
+                                        505,
+                                        508
+                                    ],
+                                    "score": 1.0,
+                                    "content": "until the next update. The main parts of our algorithm, Sparsity Distribution, Update Schedule,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                506,
+                                505,
+                                519
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        506,
+                                        505,
+                                        519
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Drop Criterion, Grow Criterion, and the various options we considered for each, are explained",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                518,
+                                506,
+                                531
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        518,
+                                        265,
+                                        531
+                                    ],
+                                    "score": 1.0,
+                                    "content": "below. The improved performance of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        265,
+                                        518,
+                                        287,
+                                        529
+                                    ],
+                                    "score": 0.43,
+                                    "content": "R i g L",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        287,
+                                        518,
+                                        506,
+                                        531
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is due to two reasons: the use of a new method for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                528,
+                                506,
+                                541
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        528,
+                                        506,
+                                        541
+                                    ],
+                                    "score": 1.0,
+                                    "content": "activating connections that is efficient and more effective than choosing at random, and the use of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                540,
+                                505,
+                                553
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        540,
+                                        505,
+                                        553
+                                    ],
+                                    "score": 1.0,
+                                    "content": "a natural extension to an existing method for distributing parameters statically among convolutional",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                550,
+                                137,
+                                565
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        550,
+                                        137,
+                                        565
+                                    ],
+                                    "score": 1.0,
+                                    "content": "layers.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        }
+                    ],
+                    "index": 28,
+                    "bbox_fs": [
+                        104,
+                        463,
+                        506,
+                        565
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        567,
+                        504,
+                        591
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                566,
+                                505,
+                                580
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        566,
+                                        234,
+                                        580
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(0) Notation. Given a dataset",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        234,
+                                        568,
+                                        244,
+                                        577
+                                    ],
+                                    "score": 0.76,
+                                    "content": "D",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        244,
+                                        566,
+                                        338,
+                                        580
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with individual inputs",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        338,
+                                        569,
+                                        348,
+                                        579
+                                    ],
+                                    "score": 0.85,
+                                    "content": "x _ { i }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        349,
+                                        566,
+                                        398,
+                                        580
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and targets",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        398,
+                                        570,
+                                        407,
+                                        579
+                                    ],
+                                    "score": 0.84,
+                                    "content": "y _ { i }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        408,
+                                        566,
+                                        505,
+                                        580
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", one can train a neural",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                578,
+                                505,
+                                591
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        578,
+                                        266,
+                                        591
+                                    ],
+                                    "score": 1.0,
+                                    "content": "network to minimize the loss function",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        266,
+                                        578,
+                                        335,
+                                        591
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\textstyle \\sum _ { i } L ( f _ { \\theta } ( x _ { i } ) , y _ { i } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        336,
+                                        578,
+                                        368,
+                                        591
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", where",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        369,
+                                        579,
+                                        393,
+                                        591
+                                    ],
+                                    "score": 0.92,
+                                    "content": "f _ { \\boldsymbol { \\theta } } ( \\boldsymbol { x } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        393,
+                                        578,
+                                        505,
+                                        591
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is the neural network with",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                82,
+                                505,
+                                94
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        82,
+                                        153,
+                                        94
+                                    ],
+                                    "score": 1.0,
+                                    "content": "parameters",
+                                    "type": "text",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        154,
+                                        83,
+                                        160,
+                                        92
+                                    ],
+                                    "score": 0.81,
+                                    "content": "\\theta",
+                                    "type": "inline_equation",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        160,
+                                        82,
+                                        200,
+                                        94
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of length",
+                                    "type": "text",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        201,
+                                        83,
+                                        211,
+                                        93
+                                    ],
+                                    "score": 0.73,
+                                    "content": "N",
+                                    "type": "inline_equation",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        212,
+                                        82,
+                                        263,
+                                        94
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". The vector",
+                                    "type": "text",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        264,
+                                        83,
+                                        271,
+                                        92
+                                    ],
+                                    "score": 0.78,
+                                    "content": "\\theta",
+                                    "type": "inline_equation",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        271,
+                                        82,
+                                        419,
+                                        94
+                                    ],
+                                    "score": 1.0,
+                                    "content": "can be decomposed into parameters",
+                                    "type": "text",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        420,
+                                        82,
+                                        429,
+                                        93
+                                    ],
+                                    "score": 0.86,
+                                    "content": "\\theta ^ { l }",
+                                    "type": "inline_equation",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        429,
+                                        82,
+                                        473,
+                                        94
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", of length",
+                                    "type": "text",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        473,
+                                        82,
+                                        486,
+                                        93
+                                    ],
+                                    "score": 0.88,
+                                    "content": "N ^ { l }",
+                                    "type": "inline_equation",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        487,
+                                        82,
+                                        505,
+                                        94
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", for",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                93,
+                                506,
+                                106
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        93,
+                                        151,
+                                        106
+                                    ],
+                                    "score": 1.0,
+                                    "content": "each layer",
+                                    "type": "text",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        151,
+                                        94,
+                                        155,
+                                        104
+                                    ],
+                                    "score": 0.57,
+                                    "content": "l",
+                                    "type": "inline_equation",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        155,
+                                        93,
+                                        324,
+                                        106
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". A sparse network keeps only a fraction",
+                                    "type": "text",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        324,
+                                        93,
+                                        369,
+                                        106
+                                    ],
+                                    "score": 0.92,
+                                    "content": "D \\in ( 0 , 1 )",
+                                    "type": "inline_equation",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        370,
+                                        93,
+                                        506,
+                                        106
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of all connections, resulting in a",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                104,
+                                505,
+                                117
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        104,
+                                        154,
+                                        117
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sparsity of",
+                                    "type": "text",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        154,
+                                        105,
+                                        201,
+                                        115
+                                    ],
+                                    "score": 0.91,
+                                    "content": "S = 1 - D",
+                                    "type": "inline_equation",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        202,
+                                        104,
+                                        457,
+                                        117
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". More precisely, denoting the sparsity of individual layers with",
+                                    "type": "text",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        457,
+                                        104,
+                                        466,
+                                        114
+                                    ],
+                                    "score": 0.85,
+                                    "content": "s ^ { l }",
+                                    "type": "inline_equation",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        466,
+                                        104,
+                                        505,
+                                        117
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", the total",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                114,
+                                450,
+                                128
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        115,
+                                        322,
+                                        128
+                                    ],
+                                    "score": 1.0,
+                                    "content": "parameter count of the sparse neural network satisfies",
+                                    "type": "text",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        323,
+                                        114,
+                                        446,
+                                        128
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\begin{array} { r } { \\dot { \\sum _ { l } } ( 1 - s ^ { l } ) N ^ { l } = ( 1 - \\bar { S } ) * N } \\end{array}",
+                                    "type": "inline_equation",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        446,
+                                        115,
+                                        450,
+                                        128
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 3
+                        }
+                    ],
+                    "index": 33.5,
+                    "bbox_fs": [
+                        105,
+                        566,
+                        505,
+                        591
+                    ]
+                },
+                {
+                    "type": "image",
+                    "bbox": [
+                        125,
+                        614,
+                        486,
+                        713
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                125,
+                                614,
+                                486,
+                                713
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        125,
+                                        614,
+                                        486,
+                                        713
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                125,
+                                                614,
+                                                486,
+                                                713
+                                            ],
+                                            "score": 0.969,
+                                            "type": "image",
+                                            "image_path": "c4f8c255dd49151f0525882b1c8f56ef8e1197a1ff58bb53e4c69623eb1ec0b1.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 36,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        125,
+                                        614,
+                                        486,
+                                        647.0
+                                    ],
+                                    "spans": [],
+                                    "index": 35
+                                },
+                                {
+                                    "bbox": [
+                                        125,
+                                        647.0,
+                                        486,
+                                        680.0
+                                    ],
+                                    "spans": [],
+                                    "index": 36
+                                },
+                                {
+                                    "bbox": [
+                                        125,
+                                        680.0,
+                                        486,
+                                        713.0
+                                    ],
+                                    "spans": [],
+                                    "index": 37
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                105,
+                                721,
+                                505,
+                                743
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        719,
+                                        504,
+                                        734
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                719,
+                                                504,
+                                                734
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 1: Dynamic sparse training aims to change connectivity during training to help out optimza-",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 38
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        731,
+                                        129,
+                                        744
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                731,
+                                                129,
+                                                744
+                                            ],
+                                            "score": 1.0,
+                                            "content": "tion.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 39
+                                }
+                            ],
+                            "index": 38.5
+                        }
+                    ],
+                    "index": 37.25
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        82,
+                        505,
+                        128
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                82,
+                                505,
+                                94
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        82,
+                                        153,
+                                        94
+                                    ],
+                                    "score": 1.0,
+                                    "content": "parameters",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        154,
+                                        83,
+                                        160,
+                                        92
+                                    ],
+                                    "score": 0.81,
+                                    "content": "\\theta",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        160,
+                                        82,
+                                        200,
+                                        94
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of length",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        201,
+                                        83,
+                                        211,
+                                        93
+                                    ],
+                                    "score": 0.73,
+                                    "content": "N",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        212,
+                                        82,
+                                        263,
+                                        94
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". The vector",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        264,
+                                        83,
+                                        271,
+                                        92
+                                    ],
+                                    "score": 0.78,
+                                    "content": "\\theta",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        271,
+                                        82,
+                                        419,
+                                        94
+                                    ],
+                                    "score": 1.0,
+                                    "content": "can be decomposed into parameters",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        420,
+                                        82,
+                                        429,
+                                        93
+                                    ],
+                                    "score": 0.86,
+                                    "content": "\\theta ^ { l }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        429,
+                                        82,
+                                        473,
+                                        94
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", of length",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        473,
+                                        82,
+                                        486,
+                                        93
+                                    ],
+                                    "score": 0.88,
+                                    "content": "N ^ { l }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        487,
+                                        82,
+                                        505,
+                                        94
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                93,
+                                506,
+                                106
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        93,
+                                        151,
+                                        106
+                                    ],
+                                    "score": 1.0,
+                                    "content": "each layer",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        151,
+                                        94,
+                                        155,
+                                        104
+                                    ],
+                                    "score": 0.57,
+                                    "content": "l",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        155,
+                                        93,
+                                        324,
+                                        106
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". A sparse network keeps only a fraction",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        324,
+                                        93,
+                                        369,
+                                        106
+                                    ],
+                                    "score": 0.92,
+                                    "content": "D \\in ( 0 , 1 )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        370,
+                                        93,
+                                        506,
+                                        106
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of all connections, resulting in a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                104,
+                                505,
+                                117
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        104,
+                                        154,
+                                        117
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sparsity of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        154,
+                                        105,
+                                        201,
+                                        115
+                                    ],
+                                    "score": 0.91,
+                                    "content": "S = 1 - D",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        202,
+                                        104,
+                                        457,
+                                        117
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". More precisely, denoting the sparsity of individual layers with",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        457,
+                                        104,
+                                        466,
+                                        114
+                                    ],
+                                    "score": 0.85,
+                                    "content": "s ^ { l }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        466,
+                                        104,
+                                        505,
+                                        117
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", the total",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                114,
+                                450,
+                                128
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        115,
+                                        322,
+                                        128
+                                    ],
+                                    "score": 1.0,
+                                    "content": "parameter count of the sparse neural network satisfies",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        323,
+                                        114,
+                                        446,
+                                        128
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\begin{array} { r } { \\dot { \\sum _ { l } } ( 1 - s ^ { l } ) N ^ { l } = ( 1 - \\bar { S } ) * N } \\end{array}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        446,
+                                        115,
+                                        450,
+                                        128
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        }
+                    ],
+                    "index": 1.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        132,
+                        505,
+                        177
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                132,
+                                505,
+                                145
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        132,
+                                        505,
+                                        145
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(1) Sparsity Distribution. There are many ways of distributing the non-zero weights across the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                142,
+                                505,
+                                157
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        142,
+                                        505,
+                                        157
+                                    ],
+                                    "score": 1.0,
+                                    "content": "layers while satisfying the equality above. We avoid re-allocating parameters between layers during",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                154,
+                                506,
+                                167
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        154,
+                                        506,
+                                        167
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the training process as it makes it difficult to target a specific final FLOP budget, which is important",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                165,
+                                411,
+                                178
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        165,
+                                        411,
+                                        178
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for many inference applications. We consider the following three strategies:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        }
+                    ],
+                    "index": 5.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        129,
+                        186,
+                        505,
+                        333
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                130,
+                                186,
+                                505,
+                                198
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        130,
+                                        186,
+                                        236,
+                                        198
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1. Uniform: The sparsity",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        236,
+                                        186,
+                                        245,
+                                        196
+                                    ],
+                                    "score": 0.86,
+                                    "content": "s ^ { l }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        245,
+                                        186,
+                                        475,
+                                        198
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of each individual layer is the same as the total sparsity",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        476,
+                                        187,
+                                        483,
+                                        196
+                                    ],
+                                    "score": 0.78,
+                                    "content": "S",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        484,
+                                        186,
+                                        505,
+                                        198
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". We",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                197,
+                                464,
+                                210
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        198,
+                                        246,
+                                        210
+                                    ],
+                                    "score": 1.0,
+                                    "content": "keep the first layer dense",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        247,
+                                        197,
+                                        277,
+                                        208
+                                    ],
+                                    "score": 0.88,
+                                    "content": "\\boldsymbol s ^ { 0 } = 0 ,",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        277,
+                                        198,
+                                        464,
+                                        210
+                                    ],
+                                    "score": 1.0,
+                                    "content": "), since it has negligible number of parameters.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                210,
+                                507,
+                                230
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        210,
+                                        355,
+                                        227
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2. Erdos-R ˝ enyi: ´ As introduced in Mocanu et al. (2018),",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        355,
+                                        213,
+                                        364,
+                                        225
+                                    ],
+                                    "score": 0.83,
+                                    "content": "s ^ { l }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        365,
+                                        210,
+                                        412,
+                                        227
+                                    ],
+                                    "score": 1.0,
+                                    "content": "scales with",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        427,
+                                        213,
+                                        507,
+                                        230
+                                    ],
+                                    "score": 1.0,
+                                    "content": "n +n nl−1∗nl , where nl",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                225,
+                                506,
+                                238
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        225,
+                                        506,
+                                        238
+                                    ],
+                                    "score": 1.0,
+                                    "content": "denotes number of neurons at layer l. This enables the number of connections in a sparse",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                142,
+                                237,
+                                428,
+                                249
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        142,
+                                        237,
+                                        428,
+                                        249
+                                    ],
+                                    "score": 1.0,
+                                    "content": "layer to scale with the sum of the number of output and input channels.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                129,
+                                251,
+                                505,
+                                263
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        129,
+                                        251,
+                                        505,
+                                        263
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3. Erdos-R ˝ enyi-Kernel (ERK): ´ This method modifies the original Erdos-R ˝ enyi formulation ´",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                262,
+                                506,
+                                274
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        262,
+                                        506,
+                                        274
+                                    ],
+                                    "score": 1.0,
+                                    "content": "by including the kernel dimensions in the scaling factors. In other words, the number of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                138,
+                                271,
+                                506,
+                                291
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        138,
+                                        271,
+                                        506,
+                                        291
+                                    ],
+                                    "score": 1.0,
+                                    "content": "parameters of the sparse convolutional layers are scaled proportional to 1− nl−1+nl+wl+hlnl−1∗nl∗wl∗hl ,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                288,
+                                505,
+                                301
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        288,
+                                        168,
+                                        301
+                                    ],
+                                    "score": 1.0,
+                                    "content": "where",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        169,
+                                        288,
+                                        180,
+                                        299
+                                    ],
+                                    "score": 0.88,
+                                    "content": "w ^ { l }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        181,
+                                        288,
+                                        197,
+                                        301
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        198,
+                                        288,
+                                        207,
+                                        299
+                                    ],
+                                    "score": 0.87,
+                                    "content": "h ^ { l }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        208,
+                                        288,
+                                        343,
+                                        301
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are the width and the height of the",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        343,
+                                        289,
+                                        349,
+                                        299
+                                    ],
+                                    "score": 0.57,
+                                    "content": "\\mathbf { \\nabla } _ { l } ,",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        349,
+                                        288,
+                                        505,
+                                        301
+                                    ],
+                                    "score": 1.0,
+                                    "content": "’th convolutional kernel. Sparsity of the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                299,
+                                505,
+                                312
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        299,
+                                        505,
+                                        312
+                                    ],
+                                    "score": 1.0,
+                                    "content": "fully connected layers scale as in the original Erdos-R ˝ enyi formulation. Similar to Erd ´ os- ˝",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                310,
+                                505,
+                                324
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        310,
+                                        505,
+                                        324
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Renyi, ERK allocates higher sparsities to the layers with more parameters while allocating ´",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                321,
+                                287,
+                                334
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        321,
+                                        287,
+                                        334
+                                    ],
+                                    "score": 1.0,
+                                    "content": "lower sparsities to the smaller ones.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        }
+                    ],
+                    "index": 13.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        343,
+                        374,
+                        354
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                341,
+                                375,
+                                356
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        341,
+                                        375,
+                                        356
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In all methods, the bias and batch-norm parameters are kept dense.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        }
+                    ],
+                    "index": 20
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        359,
+                        505,
+                        425
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                360,
+                                505,
+                                372
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        360,
+                                        505,
+                                        372
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(2) Update Schedule. The update schedule is defined by the following parameters: (1) the number",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                369,
+                                506,
+                                384
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        369,
+                                        307,
+                                        384
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of iterations between sparse connectivity updates",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        307,
+                                        371,
+                                        330,
+                                        382
+                                    ],
+                                    "score": 0.8,
+                                    "content": "( \\Delta T )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        330,
+                                        369,
+                                        506,
+                                        384
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", (2) the iteration at which to stop updating",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                382,
+                                505,
+                                394
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        382,
+                                        201,
+                                        394
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the sparse connectivity",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        201,
+                                        382,
+                                        227,
+                                        393
+                                    ],
+                                    "score": 0.89,
+                                    "content": "( T _ { e n d } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        227,
+                                        382,
+                                        415,
+                                        394
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", (3) the initial fraction of connections updated",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        416,
+                                        382,
+                                        430,
+                                        393
+                                    ],
+                                    "score": 0.73,
+                                    "content": "( \\alpha )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        430,
+                                        382,
+                                        505,
+                                        394
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and (4) a function",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                107,
+                                392,
+                                505,
+                                406
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        393,
+                                        133,
+                                        405
+                                    ],
+                                    "score": 0.59,
+                                    "content": "f _ { d e c a y }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        133,
+                                        392,
+                                        196,
+                                        406
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", invoked every",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        197,
+                                        393,
+                                        213,
+                                        403
+                                    ],
+                                    "score": 0.86,
+                                    "content": "\\Delta T",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        214,
+                                        392,
+                                        276,
+                                        406
+                                    ],
+                                    "score": 1.0,
+                                    "content": "iterations until",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        276,
+                                        393,
+                                        297,
+                                        404
+                                    ],
+                                    "score": 0.9,
+                                    "content": "T _ { e n d }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        297,
+                                        392,
+                                        505,
+                                        406
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", possibly decaying the fraction of updated connec-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                404,
+                                505,
+                                416
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        404,
+                                        505,
+                                        416
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tions over time. For the latter we choose to use cosine annealing, as we find it slightly outperforms",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                415,
+                                228,
+                                426
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        415,
+                                        228,
+                                        426
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the other methods considered.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        }
+                    ],
+                    "index": 23.5
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        232,
+                        428,
+                        379,
+                        455
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                232,
+                                428,
+                                379,
+                                455
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        232,
+                                        428,
+                                        379,
+                                        455
+                                    ],
+                                    "score": 0.94,
+                                    "content": "f _ { d e c a y } ( t ) = \\frac { \\alpha } { 2 } \\left( 1 + c o s \\left( \\frac { t \\pi } { T _ { e n d } } \\right) \\right)",
+                                    "type": "interline_equation",
+                                    "image_path": "2875bcc5bfcbfd95d66fe9aaef666e78bcbfca0191f7386b9395ba33e82f35c9.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 27.5,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                232,
+                                428,
+                                379,
+                                441.5
+                            ],
+                            "spans": [],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                232,
+                                441.5,
+                                379,
+                                455.0
+                            ],
+                            "spans": [],
+                            "index": 28
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        458,
+                        505,
+                        481
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                457,
+                                505,
+                                471
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        457,
+                                        505,
+                                        471
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Alternatives to cosine annealing like a constant schedule and inverse power annealing are studied in",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                468,
+                                174,
+                                482
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        468,
+                                        174,
+                                        482
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the Appendix F.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        }
+                    ],
+                    "index": 29.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        486,
+                        504,
+                        519
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                486,
+                                505,
+                                499
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        486,
+                                        505,
+                                        499
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(3) Drop criterion. Over the course of training, we drop the lowest magnitude weights according",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                496,
+                                505,
+                                511
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        496,
+                                        505,
+                                        511
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to the update schedule since they are expected to effect the training loss least. Specifically, we drop",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                507,
+                                208,
+                                522
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        507,
+                                        208,
+                                        522
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the connections given by",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        }
+                    ],
+                    "index": 32
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        519,
+                        256,
+                        532
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                518,
+                                255,
+                                532
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        518,
+                                        255,
+                                        532
+                                    ],
+                                    "score": 0.86,
+                                    "content": "T o p K ( - | \\theta ^ { l } | , f _ { d e c a y } ( t ) \\dot { ( } 1 - s ^ { l } ) N ^ { l } ) ^ { \\ddag } .",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 34
+                        }
+                    ],
+                    "index": 34
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        535,
+                        505,
+                        638
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                535,
+                                505,
+                                549
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        535,
+                                        505,
+                                        549
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(4) Grow criterion. The novelty of our method lies in how we grow new connections. We grow",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                543,
+                                505,
+                                563
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        543,
+                                        305,
+                                        563
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the connections with highest magnitude gradients,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        306,
+                                        546,
+                                        502,
+                                        561
+                                    ],
+                                    "score": 0.89,
+                                    "content": "T o p K _ { w \\notin \\theta _ { a c t i v e } ^ { l } } \\breve { ( | { g r a d ( \\theta ^ { l } ) } | } , f _ { d e c a y } ( t ) ( 1 - s ^ { l } ) \\breve { N } ^ { l } ) .",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        502,
+                                        546,
+                                        505,
+                                        560
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                560,
+                                505,
+                                573
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        560,
+                                        133,
+                                        573
+                                    ],
+                                    "score": 1.0,
+                                    "content": "where",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        134,
+                                        560,
+                                        162,
+                                        573
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\theta _ { a c t i v e } ^ { l }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        162,
+                                        561,
+                                        505,
+                                        573
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is the set of active connections after the drop step. Newly activated connections are",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                572,
+                                505,
+                                584
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        572,
+                                        505,
+                                        584
+                                    ],
+                                    "score": 1.0,
+                                    "content": "initialized to zero and therefore don’t effect the output of the network. However they are expected",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                583,
+                                505,
+                                595
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        583,
+                                        505,
+                                        595
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to receive gradients with high magnitudes in the next iteration and therefore reduce the loss fastest.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                594,
+                                506,
+                                607
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        594,
+                                        506,
+                                        607
+                                    ],
+                                    "score": 1.0,
+                                    "content": "This procedure can be applied to each layer in sequence and the dense gradients can be discarded",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                605,
+                                505,
+                                617
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        605,
+                                        505,
+                                        617
+                                    ],
+                                    "score": 1.0,
+                                    "content": "immediately after selecting the top connections. If a layer is too large to materialize the full gradient",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                616,
+                                505,
+                                629
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        616,
+                                        505,
+                                        629
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with respect to the weights, then we can further reduce the memory requirements by performing an",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                626,
+                                190,
+                                639
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        626,
+                                        190,
+                                        639
+                                    ],
+                                    "score": 1.0,
+                                    "content": "iterative calculation:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        }
+                    ],
+                    "index": 39
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        128,
+                        646,
+                        506,
+                        714
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                129,
+                                645,
+                                253,
+                                660
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        129,
+                                        645,
+                                        208,
+                                        660
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1. Initialize the set",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        208,
+                                        647,
+                                        249,
+                                        659
+                                    ],
+                                    "score": 0.92,
+                                    "content": "T K = \\{ \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        249,
+                                        645,
+                                        253,
+                                        660
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                128,
+                                659,
+                                451,
+                                676
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        128,
+                                        659,
+                                        414,
+                                        676
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2. Materialize a subset of size M of the full gradient, which we denote",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        414,
+                                        662,
+                                        446,
+                                        674
+                                    ],
+                                    "score": 0.92,
+                                    "content": "G _ { i : i + M }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        446,
+                                        659,
+                                        451,
+                                        676
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                129,
+                                676,
+                                455,
+                                689
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        129,
+                                        676,
+                                        173,
+                                        689
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3. Update",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        174,
+                                        676,
+                                        191,
+                                        687
+                                    ],
+                                    "score": 0.7,
+                                    "content": "T K",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        192,
+                                        676,
+                                        268,
+                                        689
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to contain the Top-",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        268,
+                                        677,
+                                        277,
+                                        687
+                                    ],
+                                    "score": 0.25,
+                                    "content": "\\mathbf { \\nabla } \\cdot \\mathbf { K }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        277,
+                                        676,
+                                        326,
+                                        689
+                                    ],
+                                    "score": 1.0,
+                                    "content": "elements of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        326,
+                                        677,
+                                        358,
+                                        688
+                                    ],
+                                    "score": 0.92,
+                                    "content": "G _ { i : i + M }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        358,
+                                        676,
+                                        434,
+                                        689
+                                    ],
+                                    "score": 1.0,
+                                    "content": "concatenated with",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        434,
+                                        676,
+                                        452,
+                                        687
+                                    ],
+                                    "score": 0.8,
+                                    "content": "T K",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        452,
+                                        676,
+                                        455,
+                                        689
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        },
+                        {
+                            "bbox": [
+                                129,
+                                691,
+                                504,
+                                703
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        129,
+                                        691,
+                                        486,
+                                        703
+                                    ],
+                                    "score": 1.0,
+                                    "content": "4. Repeat steps 1 through 3 until all of the gradients have been materialized. The final set",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        486,
+                                        691,
+                                        504,
+                                        702
+                                    ],
+                                    "score": 0.74,
+                                    "content": "T K",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 47
+                        },
+                        {
+                            "bbox": [
+                                142,
+                                703,
+                                313,
+                                714
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        142,
+                                        703,
+                                        313,
+                                        714
+                                    ],
+                                    "score": 1.0,
+                                    "content": "contains the connections we wish to grow.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 48
+                        }
+                    ],
+                    "index": 46
+                }
+            ],
+            "page_idx": 3,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        107,
+                        27,
+                        308,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                107,
+                                26,
+                                308,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        26,
+                                        308,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Under review as a conference paper at ICLR 2020",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        122,
+                        721,
+                        402,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                121,
+                                719,
+                                403,
+                                734
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        121,
+                                        720,
+                                        171,
+                                        732
+                                    ],
+                                    "score": 0.87,
+                                    "content": "^ { \\ddagger } T o p K ( v , k )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        171,
+                                        719,
+                                        317,
+                                        734
+                                    ],
+                                    "score": 1.0,
+                                    "content": "returns the indices and values of the top-",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        318,
+                                        722,
+                                        324,
+                                        730
+                                    ],
+                                    "score": 0.82,
+                                    "content": "k",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        324,
+                                        719,
+                                        393,
+                                        734
+                                    ],
+                                    "score": 1.0,
+                                    "content": "elements of vector",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        393,
+                                        724,
+                                        399,
+                                        730
+                                    ],
+                                    "score": 0.73,
+                                    "content": "v",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        399,
+                                        719,
+                                        403,
+                                        734
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        302,
+                        752,
+                        308,
+                        759
+                    ],
+                    "lines": []
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        82,
+                        505,
+                        128
+                    ],
+                    "lines": [],
+                    "index": 1.5,
+                    "bbox_fs": [
+                        105,
+                        82,
+                        506,
+                        128
+                    ],
+                    "lines_deleted": true
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        132,
+                        505,
+                        177
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                132,
+                                505,
+                                145
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        132,
+                                        505,
+                                        145
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(1) Sparsity Distribution. There are many ways of distributing the non-zero weights across the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                142,
+                                505,
+                                157
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        142,
+                                        505,
+                                        157
+                                    ],
+                                    "score": 1.0,
+                                    "content": "layers while satisfying the equality above. We avoid re-allocating parameters between layers during",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                154,
+                                506,
+                                167
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        154,
+                                        506,
+                                        167
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the training process as it makes it difficult to target a specific final FLOP budget, which is important",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                165,
+                                411,
+                                178
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        165,
+                                        411,
+                                        178
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for many inference applications. We consider the following three strategies:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        }
+                    ],
+                    "index": 5.5,
+                    "bbox_fs": [
+                        105,
+                        132,
+                        506,
+                        178
+                    ]
+                },
+                {
+                    "type": "list",
+                    "bbox": [
+                        129,
+                        186,
+                        505,
+                        333
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                130,
+                                186,
+                                505,
+                                198
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        130,
+                                        186,
+                                        236,
+                                        198
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1. Uniform: The sparsity",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        236,
+                                        186,
+                                        245,
+                                        196
+                                    ],
+                                    "score": 0.86,
+                                    "content": "s ^ { l }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        245,
+                                        186,
+                                        475,
+                                        198
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of each individual layer is the same as the total sparsity",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        476,
+                                        187,
+                                        483,
+                                        196
+                                    ],
+                                    "score": 0.78,
+                                    "content": "S",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        484,
+                                        186,
+                                        505,
+                                        198
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". We",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                197,
+                                464,
+                                210
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        198,
+                                        246,
+                                        210
+                                    ],
+                                    "score": 1.0,
+                                    "content": "keep the first layer dense",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        247,
+                                        197,
+                                        277,
+                                        208
+                                    ],
+                                    "score": 0.88,
+                                    "content": "\\boldsymbol s ^ { 0 } = 0 ,",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        277,
+                                        198,
+                                        464,
+                                        210
+                                    ],
+                                    "score": 1.0,
+                                    "content": "), since it has negligible number of parameters.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                210,
+                                507,
+                                230
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        210,
+                                        355,
+                                        227
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2. Erdos-R ˝ enyi: ´ As introduced in Mocanu et al. (2018),",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        355,
+                                        213,
+                                        364,
+                                        225
+                                    ],
+                                    "score": 0.83,
+                                    "content": "s ^ { l }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        365,
+                                        210,
+                                        412,
+                                        227
+                                    ],
+                                    "score": 1.0,
+                                    "content": "scales with",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        427,
+                                        213,
+                                        507,
+                                        230
+                                    ],
+                                    "score": 1.0,
+                                    "content": "n +n nl−1∗nl , where nl",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                225,
+                                506,
+                                238
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        225,
+                                        506,
+                                        238
+                                    ],
+                                    "score": 1.0,
+                                    "content": "denotes number of neurons at layer l. This enables the number of connections in a sparse",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                142,
+                                237,
+                                428,
+                                249
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        142,
+                                        237,
+                                        428,
+                                        249
+                                    ],
+                                    "score": 1.0,
+                                    "content": "layer to scale with the sum of the number of output and input channels.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                129,
+                                251,
+                                505,
+                                263
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        129,
+                                        251,
+                                        505,
+                                        263
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3. Erdos-R ˝ enyi-Kernel (ERK): ´ This method modifies the original Erdos-R ˝ enyi formulation ´",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                262,
+                                506,
+                                274
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        262,
+                                        506,
+                                        274
+                                    ],
+                                    "score": 1.0,
+                                    "content": "by including the kernel dimensions in the scaling factors. In other words, the number of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                138,
+                                271,
+                                506,
+                                291
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        138,
+                                        271,
+                                        506,
+                                        291
+                                    ],
+                                    "score": 1.0,
+                                    "content": "parameters of the sparse convolutional layers are scaled proportional to 1− nl−1+nl+wl+hlnl−1∗nl∗wl∗hl ,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                288,
+                                505,
+                                301
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        288,
+                                        168,
+                                        301
+                                    ],
+                                    "score": 1.0,
+                                    "content": "where",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        169,
+                                        288,
+                                        180,
+                                        299
+                                    ],
+                                    "score": 0.88,
+                                    "content": "w ^ { l }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        181,
+                                        288,
+                                        197,
+                                        301
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        198,
+                                        288,
+                                        207,
+                                        299
+                                    ],
+                                    "score": 0.87,
+                                    "content": "h ^ { l }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        208,
+                                        288,
+                                        343,
+                                        301
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are the width and the height of the",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        343,
+                                        289,
+                                        349,
+                                        299
+                                    ],
+                                    "score": 0.57,
+                                    "content": "\\mathbf { \\nabla } _ { l } ,",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        349,
+                                        288,
+                                        505,
+                                        301
+                                    ],
+                                    "score": 1.0,
+                                    "content": "’th convolutional kernel. Sparsity of the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                299,
+                                505,
+                                312
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        299,
+                                        505,
+                                        312
+                                    ],
+                                    "score": 1.0,
+                                    "content": "fully connected layers scale as in the original Erdos-R ˝ enyi formulation. Similar to Erd ´ os- ˝",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                310,
+                                505,
+                                324
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        310,
+                                        505,
+                                        324
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Renyi, ERK allocates higher sparsities to the layers with more parameters while allocating ´",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                321,
+                                287,
+                                334
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        321,
+                                        287,
+                                        334
+                                    ],
+                                    "score": 1.0,
+                                    "content": "lower sparsities to the smaller ones.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19,
+                            "is_list_end_line": true
+                        }
+                    ],
+                    "index": 13.5,
+                    "bbox_fs": [
+                        127,
+                        186,
+                        507,
+                        334
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        343,
+                        374,
+                        354
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                341,
+                                375,
+                                356
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        341,
+                                        375,
+                                        356
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In all methods, the bias and batch-norm parameters are kept dense.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        }
+                    ],
+                    "index": 20,
+                    "bbox_fs": [
+                        105,
+                        341,
+                        375,
+                        356
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        359,
+                        505,
+                        425
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                360,
+                                505,
+                                372
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        360,
+                                        505,
+                                        372
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(2) Update Schedule. The update schedule is defined by the following parameters: (1) the number",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                369,
+                                506,
+                                384
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        369,
+                                        307,
+                                        384
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of iterations between sparse connectivity updates",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        307,
+                                        371,
+                                        330,
+                                        382
+                                    ],
+                                    "score": 0.8,
+                                    "content": "( \\Delta T )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        330,
+                                        369,
+                                        506,
+                                        384
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", (2) the iteration at which to stop updating",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                382,
+                                505,
+                                394
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        382,
+                                        201,
+                                        394
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the sparse connectivity",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        201,
+                                        382,
+                                        227,
+                                        393
+                                    ],
+                                    "score": 0.89,
+                                    "content": "( T _ { e n d } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        227,
+                                        382,
+                                        415,
+                                        394
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", (3) the initial fraction of connections updated",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        416,
+                                        382,
+                                        430,
+                                        393
+                                    ],
+                                    "score": 0.73,
+                                    "content": "( \\alpha )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        430,
+                                        382,
+                                        505,
+                                        394
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and (4) a function",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                107,
+                                392,
+                                505,
+                                406
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        393,
+                                        133,
+                                        405
+                                    ],
+                                    "score": 0.59,
+                                    "content": "f _ { d e c a y }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        133,
+                                        392,
+                                        196,
+                                        406
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", invoked every",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        197,
+                                        393,
+                                        213,
+                                        403
+                                    ],
+                                    "score": 0.86,
+                                    "content": "\\Delta T",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        214,
+                                        392,
+                                        276,
+                                        406
+                                    ],
+                                    "score": 1.0,
+                                    "content": "iterations until",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        276,
+                                        393,
+                                        297,
+                                        404
+                                    ],
+                                    "score": 0.9,
+                                    "content": "T _ { e n d }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        297,
+                                        392,
+                                        505,
+                                        406
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", possibly decaying the fraction of updated connec-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                404,
+                                505,
+                                416
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        404,
+                                        505,
+                                        416
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tions over time. For the latter we choose to use cosine annealing, as we find it slightly outperforms",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                415,
+                                228,
+                                426
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        415,
+                                        228,
+                                        426
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the other methods considered.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        }
+                    ],
+                    "index": 23.5,
+                    "bbox_fs": [
+                        105,
+                        360,
+                        506,
+                        426
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        232,
+                        428,
+                        379,
+                        455
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                232,
+                                428,
+                                379,
+                                455
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        232,
+                                        428,
+                                        379,
+                                        455
+                                    ],
+                                    "score": 0.94,
+                                    "content": "f _ { d e c a y } ( t ) = \\frac { \\alpha } { 2 } \\left( 1 + c o s \\left( \\frac { t \\pi } { T _ { e n d } } \\right) \\right)",
+                                    "type": "interline_equation",
+                                    "image_path": "2875bcc5bfcbfd95d66fe9aaef666e78bcbfca0191f7386b9395ba33e82f35c9.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 27.5,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                232,
+                                428,
+                                379,
+                                441.5
+                            ],
+                            "spans": [],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                232,
+                                441.5,
+                                379,
+                                455.0
+                            ],
+                            "spans": [],
+                            "index": 28
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        458,
+                        505,
+                        481
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                457,
+                                505,
+                                471
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        457,
+                                        505,
+                                        471
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Alternatives to cosine annealing like a constant schedule and inverse power annealing are studied in",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                468,
+                                174,
+                                482
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        468,
+                                        174,
+                                        482
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the Appendix F.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        }
+                    ],
+                    "index": 29.5,
+                    "bbox_fs": [
+                        105,
+                        457,
+                        505,
+                        482
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        486,
+                        504,
+                        519
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                486,
+                                505,
+                                499
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        486,
+                                        505,
+                                        499
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(3) Drop criterion. Over the course of training, we drop the lowest magnitude weights according",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                496,
+                                505,
+                                511
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        496,
+                                        505,
+                                        511
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to the update schedule since they are expected to effect the training loss least. Specifically, we drop",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                507,
+                                208,
+                                522
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        507,
+                                        208,
+                                        522
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the connections given by",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        }
+                    ],
+                    "index": 32,
+                    "bbox_fs": [
+                        105,
+                        486,
+                        505,
+                        522
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        519,
+                        256,
+                        532
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                518,
+                                255,
+                                532
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        518,
+                                        255,
+                                        532
+                                    ],
+                                    "score": 0.86,
+                                    "content": "T o p K ( - | \\theta ^ { l } | , f _ { d e c a y } ( t ) \\dot { ( } 1 - s ^ { l } ) N ^ { l } ) ^ { \\ddag } .",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 34
+                        }
+                    ],
+                    "index": 34,
+                    "bbox_fs": [
+                        106,
+                        518,
+                        255,
+                        532
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        535,
+                        505,
+                        638
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                535,
+                                505,
+                                549
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        535,
+                                        505,
+                                        549
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(4) Grow criterion. The novelty of our method lies in how we grow new connections. We grow",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                543,
+                                505,
+                                563
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        543,
+                                        305,
+                                        563
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the connections with highest magnitude gradients,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        306,
+                                        546,
+                                        502,
+                                        561
+                                    ],
+                                    "score": 0.89,
+                                    "content": "T o p K _ { w \\notin \\theta _ { a c t i v e } ^ { l } } \\breve { ( | { g r a d ( \\theta ^ { l } ) } | } , f _ { d e c a y } ( t ) ( 1 - s ^ { l } ) \\breve { N } ^ { l } ) .",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        502,
+                                        546,
+                                        505,
+                                        560
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                560,
+                                505,
+                                573
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        560,
+                                        133,
+                                        573
+                                    ],
+                                    "score": 1.0,
+                                    "content": "where",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        134,
+                                        560,
+                                        162,
+                                        573
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\theta _ { a c t i v e } ^ { l }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        162,
+                                        561,
+                                        505,
+                                        573
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is the set of active connections after the drop step. Newly activated connections are",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                572,
+                                505,
+                                584
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        572,
+                                        505,
+                                        584
+                                    ],
+                                    "score": 1.0,
+                                    "content": "initialized to zero and therefore don’t effect the output of the network. However they are expected",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                583,
+                                505,
+                                595
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        583,
+                                        505,
+                                        595
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to receive gradients with high magnitudes in the next iteration and therefore reduce the loss fastest.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                594,
+                                506,
+                                607
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        594,
+                                        506,
+                                        607
+                                    ],
+                                    "score": 1.0,
+                                    "content": "This procedure can be applied to each layer in sequence and the dense gradients can be discarded",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                605,
+                                505,
+                                617
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        605,
+                                        505,
+                                        617
+                                    ],
+                                    "score": 1.0,
+                                    "content": "immediately after selecting the top connections. If a layer is too large to materialize the full gradient",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                616,
+                                505,
+                                629
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        616,
+                                        505,
+                                        629
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with respect to the weights, then we can further reduce the memory requirements by performing an",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                626,
+                                190,
+                                639
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        626,
+                                        190,
+                                        639
+                                    ],
+                                    "score": 1.0,
+                                    "content": "iterative calculation:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        }
+                    ],
+                    "index": 39,
+                    "bbox_fs": [
+                        104,
+                        535,
+                        506,
+                        639
+                    ]
+                },
+                {
+                    "type": "index",
+                    "bbox": [
+                        128,
+                        646,
+                        506,
+                        714
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                129,
+                                645,
+                                253,
+                                660
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        129,
+                                        645,
+                                        208,
+                                        660
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1. Initialize the set",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        208,
+                                        647,
+                                        249,
+                                        659
+                                    ],
+                                    "score": 0.92,
+                                    "content": "T K = \\{ \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        249,
+                                        645,
+                                        253,
+                                        660
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                128,
+                                659,
+                                451,
+                                676
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        128,
+                                        659,
+                                        414,
+                                        676
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2. Materialize a subset of size M of the full gradient, which we denote",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        414,
+                                        662,
+                                        446,
+                                        674
+                                    ],
+                                    "score": 0.92,
+                                    "content": "G _ { i : i + M }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        446,
+                                        659,
+                                        451,
+                                        676
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                129,
+                                676,
+                                455,
+                                689
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        129,
+                                        676,
+                                        173,
+                                        689
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3. Update",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        174,
+                                        676,
+                                        191,
+                                        687
+                                    ],
+                                    "score": 0.7,
+                                    "content": "T K",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        192,
+                                        676,
+                                        268,
+                                        689
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to contain the Top-",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        268,
+                                        677,
+                                        277,
+                                        687
+                                    ],
+                                    "score": 0.25,
+                                    "content": "\\mathbf { \\nabla } \\cdot \\mathbf { K }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        277,
+                                        676,
+                                        326,
+                                        689
+                                    ],
+                                    "score": 1.0,
+                                    "content": "elements of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        326,
+                                        677,
+                                        358,
+                                        688
+                                    ],
+                                    "score": 0.92,
+                                    "content": "G _ { i : i + M }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        358,
+                                        676,
+                                        434,
+                                        689
+                                    ],
+                                    "score": 1.0,
+                                    "content": "concatenated with",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        434,
+                                        676,
+                                        452,
+                                        687
+                                    ],
+                                    "score": 0.8,
+                                    "content": "T K",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        452,
+                                        676,
+                                        455,
+                                        689
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                129,
+                                691,
+                                504,
+                                703
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        129,
+                                        691,
+                                        486,
+                                        703
+                                    ],
+                                    "score": 1.0,
+                                    "content": "4. Repeat steps 1 through 3 until all of the gradients have been materialized. The final set",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        486,
+                                        691,
+                                        504,
+                                        702
+                                    ],
+                                    "score": 0.74,
+                                    "content": "T K",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 47,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                142,
+                                703,
+                                313,
+                                714
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        142,
+                                        703,
+                                        313,
+                                        714
+                                    ],
+                                    "score": 1.0,
+                                    "content": "contains the connections we wish to grow.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 48,
+                            "is_list_start_line": true
+                        }
+                    ],
+                    "index": 46,
+                    "bbox_fs": [
+                        128,
+                        645,
+                        504,
+                        714
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "image",
+                    "bbox": [
+                        132,
+                        81,
+                        476,
+                        211
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                132,
+                                81,
+                                476,
+                                211
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        132,
+                                        81,
+                                        476,
+                                        211
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                132,
+                                                81,
+                                                476,
+                                                211
+                                            ],
+                                            "score": 0.972,
+                                            "type": "image",
+                                            "image_path": "dfa7237fe18adc99e80cc225a8c395a896800044e15fa934ef6beb417685da6e.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 1,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        132,
+                                        81,
+                                        476,
+                                        124.33333333333334
+                                    ],
+                                    "spans": [],
+                                    "index": 0
+                                },
+                                {
+                                    "bbox": [
+                                        132,
+                                        124.33333333333334,
+                                        476,
+                                        167.66666666666669
+                                    ],
+                                    "spans": [],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        132,
+                                        167.66666666666669,
+                                        476,
+                                        211.00000000000003
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                106,
+                                221,
+                                505,
+                                299
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        221,
+                                        506,
+                                        235
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                221,
+                                                506,
+                                                235
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 2: (left) Performance of various dynamic sparse training methods on ImageNet-2012 classi-",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 3
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        232,
+                                        506,
+                                        245
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                232,
+                                                192,
+                                                245
+                                            ],
+                                            "score": 1.0,
+                                            "content": "fication task. We use",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                193,
+                                                233,
+                                                213,
+                                                243
+                                            ],
+                                            "score": 0.86,
+                                            "content": "80 \\%",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                213,
+                                                232,
+                                                506,
+                                                245
+                                            ],
+                                            "score": 1.0,
+                                            "content": "sparse ResNet-50 architecture with uniform sparsity distribution. Points",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 4
+                                },
+                                {
+                                    "bbox": [
+                                        104,
+                                        242,
+                                        506,
+                                        257
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                104,
+                                                242,
+                                                506,
+                                                257
+                                            ],
+                                            "score": 1.0,
+                                            "content": "at each curve correspond to the individual training runs with training multipliers from 1 to 5 (except",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 5
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        254,
+                                        506,
+                                        268
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                254,
+                                                506,
+                                                268
+                                            ],
+                                            "score": 1.0,
+                                            "content": "pruning which is scaled between 0.5 and 2). We repeat training 3 times at every multiplier and report",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 6
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        265,
+                                        505,
+                                        278
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                265,
+                                                505,
+                                                278
+                                            ],
+                                            "score": 1.0,
+                                            "content": "the mean accuracies. The number of FLOPs required to train a standard dense Resnet-50 along with",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 7
+                                },
+                                {
+                                    "bbox": [
+                                        104,
+                                        275,
+                                        505,
+                                        290
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                104,
+                                                275,
+                                                505,
+                                                290
+                                            ],
+                                            "score": 1.0,
+                                            "content": "its performance is indicated with a dashed red line. (right) Performance of RigL at different sparsity",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 8
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        287,
+                                        361,
+                                        300
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                287,
+                                                361,
+                                                300
+                                            ],
+                                            "score": 1.0,
+                                            "content": "levels with extended training. Results are averaged over 3 runs.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 9
+                                }
+                            ],
+                            "index": 6
+                        }
+                    ],
+                    "index": 3.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        306,
+                        505,
+                        342
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                304,
+                                506,
+                                323
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        304,
+                                        153,
+                                        323
+                                    ],
+                                    "score": 1.0,
+                                    "content": "As long as",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        153,
+                                        306,
+                                        201,
+                                        320
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\Delta T > \\frac { 1 } { 1 - s }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        201,
+                                        304,
+                                        506,
+                                        323
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the total work in calculating dense gradients is amortized and still propor-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                318,
+                                505,
+                                331
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        318,
+                                        142,
+                                        331
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tional to",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        142,
+                                        319,
+                                        167,
+                                        329
+                                    ],
+                                    "score": 0.78,
+                                    "content": "1 - S",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        167,
+                                        318,
+                                        505,
+                                        331
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". This is in contrast to the method of Dettmers & Zettlemoyer (2019), which requires",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                329,
+                                374,
+                                344
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        329,
+                                        374,
+                                        344
+                                    ],
+                                    "score": 1.0,
+                                    "content": "calculating and storing the full gradients at each optimization step.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        }
+                    ],
+                    "index": 11
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        358,
+                        254,
+                        370
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                356,
+                                256,
+                                372
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        356,
+                                        256,
+                                        372
+                                    ],
+                                    "score": 1.0,
+                                    "content": "4 EMPIRICAL EVALUATION",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        }
+                    ],
+                    "index": 13
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        383,
+                        505,
+                        472
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                382,
+                                505,
+                                397
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        382,
+                                        505,
+                                        397
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Our experiments include image classification using CNNs on the ImageNet-2012 (Russakovsky",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                393,
+                                506,
+                                407
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        393,
+                                        506,
+                                        407
+                                    ],
+                                    "score": 1.0,
+                                    "content": "et al., 2015) and CIFAR-10 (Krizhevsky et al.) datasets and character based language modelling",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                405,
+                                506,
+                                417
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        405,
+                                        506,
+                                        417
+                                    ],
+                                    "score": 1.0,
+                                    "content": "using RNNs with the WikiText-103 dataset (Merity et al., 2016). We use the TensorFlow Model",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                416,
+                                505,
+                                428
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        416,
+                                        505,
+                                        428
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Pruning library (Zhu & Gupta, 2018) for our pruning baselines. A Tensorflow (Abadi et al., 2015)",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                427,
+                                506,
+                                440
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        427,
+                                        506,
+                                        440
+                                    ],
+                                    "score": 1.0,
+                                    "content": "implementation of our method along with three other baselines (SET, SNFS, SNIP) will be open",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                438,
+                                505,
+                                451
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        438,
+                                        335,
+                                        451
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sourced. When we increase the training steps by a factor",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        336,
+                                        438,
+                                        348,
+                                        448
+                                    ],
+                                    "score": 0.76,
+                                    "content": "M",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        348,
+                                        438,
+                                        505,
+                                        451
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", the anchor epochs of the learning rate",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                449,
+                                505,
+                                462
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        449,
+                                        505,
+                                        462
+                                    ],
+                                    "score": 1.0,
+                                    "content": "schedule and the end iteration of the mask update schedule are also scaled by the same factor; we",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                459,
+                                318,
+                                474
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        459,
+                                        274,
+                                        474
+                                    ],
+                                    "score": 1.0,
+                                    "content": "indicate this scaling with a subscript (e.g.",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        274,
+                                        460,
+                                        312,
+                                        472
+                                    ],
+                                    "score": 0.67,
+                                    "content": "{ \\mathrm { R i g L } } _ { M \\times } ,",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        313,
+                                        459,
+                                        318,
+                                        474
+                                    ],
+                                    "score": 1.0,
+                                    "content": ").",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        }
+                    ],
+                    "index": 17.5
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        486,
+                        245,
+                        496
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                485,
+                                247,
+                                498
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        485,
+                                        247,
+                                        498
+                                    ],
+                                    "score": 1.0,
+                                    "content": "4.1 IMAGENET-2012 DATASET",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        }
+                    ],
+                    "index": 22
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        506,
+                        505,
+                        583
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                506,
+                                505,
+                                519
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        506,
+                                        505,
+                                        519
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In all experiments in this section, we use SGD with momentum as our optimizer. We set the momen-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                516,
+                                505,
+                                530
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        516,
+                                        263,
+                                        530
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tum coefficient of the optimizer to 0.9,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        263,
+                                        518,
+                                        276,
+                                        529
+                                    ],
+                                    "score": 0.85,
+                                    "content": "L _ { 2 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        276,
+                                        516,
+                                        505,
+                                        530
+                                    ],
+                                    "score": 1.0,
+                                    "content": "regularization coefficient to 0.0001, and label smoothing",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                529,
+                                505,
+                                541
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        529,
+                                        505,
+                                        541
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(Szegedy et al., 2016) to 0.1. The learning rate schedule starts with a linear warm up reaching its",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                540,
+                                504,
+                                551
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        540,
+                                        504,
+                                        551
+                                    ],
+                                    "score": 1.0,
+                                    "content": "maximum value of 1.6 at epoch 5 which is then dropped by a factor of 10 at epochs 30, 70 and 90.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                550,
+                                505,
+                                563
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        550,
+                                        505,
+                                        563
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We train our networks with a batch size of 4096 for 32000 steps which roughly corresponds to 100",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                561,
+                                505,
+                                574
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        561,
+                                        505,
+                                        574
+                                    ],
+                                    "score": 1.0,
+                                    "content": "epochs of training. Our training pipeline uses standard data augmentation, which includes random",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                571,
+                                170,
+                                586
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        571,
+                                        170,
+                                        586
+                                    ],
+                                    "score": 1.0,
+                                    "content": "flips and crops.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        }
+                    ],
+                    "index": 26
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        596,
+                        191,
+                        608
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                595,
+                                191,
+                                609
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        595,
+                                        191,
+                                        609
+                                    ],
+                                    "score": 1.0,
+                                    "content": "4.1.1 RESNET-50",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        }
+                    ],
+                    "index": 30
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        615,
+                        505,
+                        671
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                616,
+                                504,
+                                628
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        616,
+                                        410,
+                                        628
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Figure 2-left summarizes the performance of various methods on training an",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        410,
+                                        616,
+                                        429,
+                                        627
+                                    ],
+                                    "score": 0.85,
+                                    "content": "80 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        430,
+                                        616,
+                                        504,
+                                        628
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sparse ResNet-50.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                627,
+                                505,
+                                639
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        627,
+                                        505,
+                                        639
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We also train small dense networks with equivalent parameter count. All sparse networks use the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                637,
+                                505,
+                                650
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        637,
+                                        353,
+                                        650
+                                    ],
+                                    "score": 1.0,
+                                    "content": "constant sparsity distribution and a cosine update schedule",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        353,
+                                        638,
+                                        391,
+                                        649
+                                    ],
+                                    "score": 0.8,
+                                    "content": "\\mathrm { ~ \\ : ( \\alpha ~ ) = ~ 0 . 3 ~ }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        391,
+                                        637,
+                                        396,
+                                        650
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        396,
+                                        638,
+                                        445,
+                                        649
+                                    ],
+                                    "score": 0.8,
+                                    "content": "\\Delta T = 1 0 0",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        445,
+                                        637,
+                                        505,
+                                        650
+                                    ],
+                                    "score": 1.0,
+                                    "content": "). Overall, we",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                649,
+                                505,
+                                661
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        649,
+                                        505,
+                                        661
+                                    ],
+                                    "score": 1.0,
+                                    "content": "observe that the performance of all methods improves with training time; thus, for each method we",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                659,
+                                384,
+                                673
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        659,
+                                        236,
+                                        673
+                                    ],
+                                    "score": 1.0,
+                                    "content": "run extended training with up to",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        237,
+                                        660,
+                                        251,
+                                        671
+                                    ],
+                                    "score": 0.86,
+                                    "content": "5 \\times",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        251,
+                                        659,
+                                        384,
+                                        673
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the training steps of the original.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        }
+                    ],
+                    "index": 33
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        676,
+                        504,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                676,
+                                505,
+                                690
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        676,
+                                        505,
+                                        690
+                                    ],
+                                    "score": 1.0,
+                                    "content": "As noted by Gale et al. (2019), Evci et al. (2019), Frankle et al. (2019), and Mostafa & Wang (2019),",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                686,
+                                506,
+                                702
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        686,
+                                        506,
+                                        702
+                                    ],
+                                    "score": 1.0,
+                                    "content": "training a network with fixed sparsity from scratch (Static) leads to inferior performance. Training",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                699,
+                                505,
+                                711
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        699,
+                                        505,
+                                        711
+                                    ],
+                                    "score": 1.0,
+                                    "content": "a small dense network with the same number of parameters gets better results than Static, but fails",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                709,
+                                506,
+                                723
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        709,
+                                        358,
+                                        723
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to match the performance of dynamic sparse models. Similarly",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        359,
+                                        710,
+                                        378,
+                                        720
+                                    ],
+                                    "score": 0.28,
+                                    "content": "S E T",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        378,
+                                        709,
+                                        506,
+                                        723
+                                    ],
+                                    "score": 1.0,
+                                    "content": "improves the performance over",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                720,
+                                505,
+                                733
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        720,
+                                        266,
+                                        733
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Small-Dense, however saturates around",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        266,
+                                        721,
+                                        285,
+                                        731
+                                    ],
+                                    "score": 0.87,
+                                    "content": "7 5 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        286,
+                                        720,
+                                        505,
+                                        733
+                                    ],
+                                    "score": 1.0,
+                                    "content": "accuracy indicating the limits of growing new connec-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        }
+                    ],
+                    "index": 38
+                }
+            ],
+            "page_idx": 4,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        106,
+                        27,
+                        308,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                26,
+                                308,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        26,
+                                        308,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Under review as a conference paper at ICLR 2020",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        302,
+                        751,
+                        308,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                302,
+                                750,
+                                309,
+                                763
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        302,
+                                        750,
+                                        309,
+                                        763
+                                    ],
+                                    "score": 1.0,
+                                    "content": "5",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "image",
+                    "bbox": [
+                        132,
+                        81,
+                        476,
+                        211
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                132,
+                                81,
+                                476,
+                                211
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        132,
+                                        81,
+                                        476,
+                                        211
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                132,
+                                                81,
+                                                476,
+                                                211
+                                            ],
+                                            "score": 0.972,
+                                            "type": "image",
+                                            "image_path": "dfa7237fe18adc99e80cc225a8c395a896800044e15fa934ef6beb417685da6e.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 1,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        132,
+                                        81,
+                                        476,
+                                        124.33333333333334
+                                    ],
+                                    "spans": [],
+                                    "index": 0
+                                },
+                                {
+                                    "bbox": [
+                                        132,
+                                        124.33333333333334,
+                                        476,
+                                        167.66666666666669
+                                    ],
+                                    "spans": [],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        132,
+                                        167.66666666666669,
+                                        476,
+                                        211.00000000000003
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                106,
+                                221,
+                                505,
+                                299
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        221,
+                                        506,
+                                        235
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                221,
+                                                506,
+                                                235
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 2: (left) Performance of various dynamic sparse training methods on ImageNet-2012 classi-",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 3
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        232,
+                                        506,
+                                        245
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                232,
+                                                192,
+                                                245
+                                            ],
+                                            "score": 1.0,
+                                            "content": "fication task. We use",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                193,
+                                                233,
+                                                213,
+                                                243
+                                            ],
+                                            "score": 0.86,
+                                            "content": "80 \\%",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                213,
+                                                232,
+                                                506,
+                                                245
+                                            ],
+                                            "score": 1.0,
+                                            "content": "sparse ResNet-50 architecture with uniform sparsity distribution. Points",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 4
+                                },
+                                {
+                                    "bbox": [
+                                        104,
+                                        242,
+                                        506,
+                                        257
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                104,
+                                                242,
+                                                506,
+                                                257
+                                            ],
+                                            "score": 1.0,
+                                            "content": "at each curve correspond to the individual training runs with training multipliers from 1 to 5 (except",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 5
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        254,
+                                        506,
+                                        268
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                254,
+                                                506,
+                                                268
+                                            ],
+                                            "score": 1.0,
+                                            "content": "pruning which is scaled between 0.5 and 2). We repeat training 3 times at every multiplier and report",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 6
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        265,
+                                        505,
+                                        278
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                265,
+                                                505,
+                                                278
+                                            ],
+                                            "score": 1.0,
+                                            "content": "the mean accuracies. The number of FLOPs required to train a standard dense Resnet-50 along with",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 7
+                                },
+                                {
+                                    "bbox": [
+                                        104,
+                                        275,
+                                        505,
+                                        290
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                104,
+                                                275,
+                                                505,
+                                                290
+                                            ],
+                                            "score": 1.0,
+                                            "content": "its performance is indicated with a dashed red line. (right) Performance of RigL at different sparsity",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 8
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        287,
+                                        361,
+                                        300
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                287,
+                                                361,
+                                                300
+                                            ],
+                                            "score": 1.0,
+                                            "content": "levels with extended training. Results are averaged over 3 runs.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 9
+                                }
+                            ],
+                            "index": 6
+                        }
+                    ],
+                    "index": 3.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        306,
+                        505,
+                        342
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                304,
+                                506,
+                                323
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        304,
+                                        153,
+                                        323
+                                    ],
+                                    "score": 1.0,
+                                    "content": "As long as",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        153,
+                                        306,
+                                        201,
+                                        320
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\Delta T > \\frac { 1 } { 1 - s }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        201,
+                                        304,
+                                        506,
+                                        323
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the total work in calculating dense gradients is amortized and still propor-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                318,
+                                505,
+                                331
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        318,
+                                        142,
+                                        331
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tional to",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        142,
+                                        319,
+                                        167,
+                                        329
+                                    ],
+                                    "score": 0.78,
+                                    "content": "1 - S",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        167,
+                                        318,
+                                        505,
+                                        331
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". This is in contrast to the method of Dettmers & Zettlemoyer (2019), which requires",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                329,
+                                374,
+                                344
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        329,
+                                        374,
+                                        344
+                                    ],
+                                    "score": 1.0,
+                                    "content": "calculating and storing the full gradients at each optimization step.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        }
+                    ],
+                    "index": 11,
+                    "bbox_fs": [
+                        104,
+                        304,
+                        506,
+                        344
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        358,
+                        254,
+                        370
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                356,
+                                256,
+                                372
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        356,
+                                        256,
+                                        372
+                                    ],
+                                    "score": 1.0,
+                                    "content": "4 EMPIRICAL EVALUATION",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        }
+                    ],
+                    "index": 13
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        383,
+                        505,
+                        472
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                382,
+                                505,
+                                397
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        382,
+                                        505,
+                                        397
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Our experiments include image classification using CNNs on the ImageNet-2012 (Russakovsky",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                393,
+                                506,
+                                407
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        393,
+                                        506,
+                                        407
+                                    ],
+                                    "score": 1.0,
+                                    "content": "et al., 2015) and CIFAR-10 (Krizhevsky et al.) datasets and character based language modelling",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                405,
+                                506,
+                                417
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        405,
+                                        506,
+                                        417
+                                    ],
+                                    "score": 1.0,
+                                    "content": "using RNNs with the WikiText-103 dataset (Merity et al., 2016). We use the TensorFlow Model",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                416,
+                                505,
+                                428
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        416,
+                                        505,
+                                        428
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Pruning library (Zhu & Gupta, 2018) for our pruning baselines. A Tensorflow (Abadi et al., 2015)",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                427,
+                                506,
+                                440
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        427,
+                                        506,
+                                        440
+                                    ],
+                                    "score": 1.0,
+                                    "content": "implementation of our method along with three other baselines (SET, SNFS, SNIP) will be open",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                438,
+                                505,
+                                451
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        438,
+                                        335,
+                                        451
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sourced. When we increase the training steps by a factor",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        336,
+                                        438,
+                                        348,
+                                        448
+                                    ],
+                                    "score": 0.76,
+                                    "content": "M",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        348,
+                                        438,
+                                        505,
+                                        451
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", the anchor epochs of the learning rate",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                449,
+                                505,
+                                462
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        449,
+                                        505,
+                                        462
+                                    ],
+                                    "score": 1.0,
+                                    "content": "schedule and the end iteration of the mask update schedule are also scaled by the same factor; we",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                459,
+                                318,
+                                474
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        459,
+                                        274,
+                                        474
+                                    ],
+                                    "score": 1.0,
+                                    "content": "indicate this scaling with a subscript (e.g.",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        274,
+                                        460,
+                                        312,
+                                        472
+                                    ],
+                                    "score": 0.67,
+                                    "content": "{ \\mathrm { R i g L } } _ { M \\times } ,",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        313,
+                                        459,
+                                        318,
+                                        474
+                                    ],
+                                    "score": 1.0,
+                                    "content": ").",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        }
+                    ],
+                    "index": 17.5,
+                    "bbox_fs": [
+                        104,
+                        382,
+                        506,
+                        474
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        486,
+                        245,
+                        496
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                485,
+                                247,
+                                498
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        485,
+                                        247,
+                                        498
+                                    ],
+                                    "score": 1.0,
+                                    "content": "4.1 IMAGENET-2012 DATASET",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        }
+                    ],
+                    "index": 22
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        506,
+                        505,
+                        583
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                506,
+                                505,
+                                519
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        506,
+                                        505,
+                                        519
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In all experiments in this section, we use SGD with momentum as our optimizer. We set the momen-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                516,
+                                505,
+                                530
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        516,
+                                        263,
+                                        530
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tum coefficient of the optimizer to 0.9,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        263,
+                                        518,
+                                        276,
+                                        529
+                                    ],
+                                    "score": 0.85,
+                                    "content": "L _ { 2 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        276,
+                                        516,
+                                        505,
+                                        530
+                                    ],
+                                    "score": 1.0,
+                                    "content": "regularization coefficient to 0.0001, and label smoothing",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                529,
+                                505,
+                                541
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        529,
+                                        505,
+                                        541
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(Szegedy et al., 2016) to 0.1. The learning rate schedule starts with a linear warm up reaching its",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                540,
+                                504,
+                                551
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        540,
+                                        504,
+                                        551
+                                    ],
+                                    "score": 1.0,
+                                    "content": "maximum value of 1.6 at epoch 5 which is then dropped by a factor of 10 at epochs 30, 70 and 90.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                550,
+                                505,
+                                563
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        550,
+                                        505,
+                                        563
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We train our networks with a batch size of 4096 for 32000 steps which roughly corresponds to 100",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                561,
+                                505,
+                                574
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        561,
+                                        505,
+                                        574
+                                    ],
+                                    "score": 1.0,
+                                    "content": "epochs of training. Our training pipeline uses standard data augmentation, which includes random",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                571,
+                                170,
+                                586
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        571,
+                                        170,
+                                        586
+                                    ],
+                                    "score": 1.0,
+                                    "content": "flips and crops.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        }
+                    ],
+                    "index": 26,
+                    "bbox_fs": [
+                        105,
+                        506,
+                        505,
+                        586
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        596,
+                        191,
+                        608
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                595,
+                                191,
+                                609
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        595,
+                                        191,
+                                        609
+                                    ],
+                                    "score": 1.0,
+                                    "content": "4.1.1 RESNET-50",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        }
+                    ],
+                    "index": 30
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        615,
+                        505,
+                        671
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                616,
+                                504,
+                                628
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        616,
+                                        410,
+                                        628
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Figure 2-left summarizes the performance of various methods on training an",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        410,
+                                        616,
+                                        429,
+                                        627
+                                    ],
+                                    "score": 0.85,
+                                    "content": "80 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        430,
+                                        616,
+                                        504,
+                                        628
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sparse ResNet-50.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                627,
+                                505,
+                                639
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        627,
+                                        505,
+                                        639
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We also train small dense networks with equivalent parameter count. All sparse networks use the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                637,
+                                505,
+                                650
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        637,
+                                        353,
+                                        650
+                                    ],
+                                    "score": 1.0,
+                                    "content": "constant sparsity distribution and a cosine update schedule",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        353,
+                                        638,
+                                        391,
+                                        649
+                                    ],
+                                    "score": 0.8,
+                                    "content": "\\mathrm { ~ \\ : ( \\alpha ~ ) = ~ 0 . 3 ~ }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        391,
+                                        637,
+                                        396,
+                                        650
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        396,
+                                        638,
+                                        445,
+                                        649
+                                    ],
+                                    "score": 0.8,
+                                    "content": "\\Delta T = 1 0 0",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        445,
+                                        637,
+                                        505,
+                                        650
+                                    ],
+                                    "score": 1.0,
+                                    "content": "). Overall, we",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                649,
+                                505,
+                                661
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        649,
+                                        505,
+                                        661
+                                    ],
+                                    "score": 1.0,
+                                    "content": "observe that the performance of all methods improves with training time; thus, for each method we",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                659,
+                                384,
+                                673
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        659,
+                                        236,
+                                        673
+                                    ],
+                                    "score": 1.0,
+                                    "content": "run extended training with up to",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        237,
+                                        660,
+                                        251,
+                                        671
+                                    ],
+                                    "score": 0.86,
+                                    "content": "5 \\times",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        251,
+                                        659,
+                                        384,
+                                        673
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the training steps of the original.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        }
+                    ],
+                    "index": 33,
+                    "bbox_fs": [
+                        105,
+                        616,
+                        505,
+                        673
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        676,
+                        504,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                676,
+                                505,
+                                690
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        676,
+                                        505,
+                                        690
+                                    ],
+                                    "score": 1.0,
+                                    "content": "As noted by Gale et al. (2019), Evci et al. (2019), Frankle et al. (2019), and Mostafa & Wang (2019),",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                686,
+                                506,
+                                702
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        686,
+                                        506,
+                                        702
+                                    ],
+                                    "score": 1.0,
+                                    "content": "training a network with fixed sparsity from scratch (Static) leads to inferior performance. Training",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                699,
+                                505,
+                                711
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        699,
+                                        505,
+                                        711
+                                    ],
+                                    "score": 1.0,
+                                    "content": "a small dense network with the same number of parameters gets better results than Static, but fails",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                709,
+                                506,
+                                723
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        709,
+                                        358,
+                                        723
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to match the performance of dynamic sparse models. Similarly",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        359,
+                                        710,
+                                        378,
+                                        720
+                                    ],
+                                    "score": 0.28,
+                                    "content": "S E T",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        378,
+                                        709,
+                                        506,
+                                        723
+                                    ],
+                                    "score": 1.0,
+                                    "content": "improves the performance over",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                720,
+                                505,
+                                733
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        720,
+                                        266,
+                                        733
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Small-Dense, however saturates around",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        266,
+                                        721,
+                                        285,
+                                        731
+                                    ],
+                                    "score": 0.87,
+                                    "content": "7 5 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        286,
+                                        720,
+                                        505,
+                                        733
+                                    ],
+                                    "score": 1.0,
+                                    "content": "accuracy indicating the limits of growing new connec-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        }
+                    ],
+                    "index": 38,
+                    "bbox_fs": [
+                        105,
+                        676,
+                        506,
+                        733
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "table",
+                    "bbox": [
+                        122,
+                        80,
+                        489,
+                        328
+                    ],
+                    "blocks": [
+                        {
+                            "type": "table_body",
+                            "bbox": [
+                                122,
+                                80,
+                                489,
+                                328
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        122,
+                                        80,
+                                        489,
+                                        328
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                122,
+                                                80,
+                                                489,
+                                                328
+                                            ],
+                                            "score": 0.986,
+                                            "html": "<table><tr><td>Method</td><td>Top-1 Accuracy</td><td>FLOPs (Train)</td><td>FLOPs (Test)</td><td>Top-1 Accuracy</td><td>FLOPs (Train)</td><td>FLOPs (Test)</td></tr><tr><td>Dense</td><td>76.8±0.09</td><td>1x (3.2e18)</td><td>1x (8.2e9)</td><td colspan=\"3\"></td></tr><tr><td>Static</td><td colspan=\"3\">S=0.8</td><td colspan=\"3\">S=0.9</td></tr><tr><td>SNIP</td><td>70.6±0.06 72.0±0.10</td><td>0.23x 0.23x</td><td>0.23x 0.23x</td><td>65.8±0.04 67.2±0.12</td><td>0.10x 0.10x</td><td>0.10x 0.10x</td></tr><tr><td>Small-Dense</td><td>72.1±0.12</td><td>0.20x</td><td>0.20x</td><td>68.9±0.10</td><td>0.12x</td><td>0.12x</td></tr><tr><td>SET</td><td>72.9±0.39</td><td>0.23x</td><td>0.23x</td><td>69.6±0.23</td><td>0.10x</td><td>0.10x</td></tr><tr><td>RigL</td><td>74.6±0.06</td><td>0.23x</td><td>0.23x</td><td>72.0±0.05</td><td>0.10x</td><td>0.10x</td></tr><tr><td>Small-Dense5x RigL5x</td><td>73.9±0.07</td><td>1.01x</td><td>0.20x</td><td>71.3±0.10</td><td>0.60x</td><td>0.12x</td></tr><tr><td>Static (ERK)</td><td>76.6±0.06</td><td>1.14x</td><td>0.23x</td><td>75.7±0.06</td><td>0.52x</td><td>0.10x</td></tr><tr><td>DSR*</td><td>72.1±0.04 73.3</td><td>0.42x 0.40x</td><td>0.42x 0.40x</td><td>67.7±0.12</td><td>0.24x</td><td>0.24x</td></tr><tr><td></td><td></td><td></td><td></td><td>71.6</td><td>0.30x</td><td>0.30x</td></tr><tr><td>RigL (ERK)</td><td>75.1±0.05</td><td>0.42x</td><td>0.42x</td><td>73.0±0.04</td><td>0.25x</td><td>0.24x</td></tr><tr><td>RigL5x (ERK)</td><td>77.1±0.06</td><td>2.09x</td><td>0.42x</td><td>76.4±0.05</td><td>1.23x</td><td>0.24x</td></tr><tr><td>SNFS*</td><td>74.2</td><td>n/a</td><td>n/a</td><td>72.3</td><td>n/a</td><td>n/a</td></tr><tr><td>SNFS (ERK)</td><td>75.2±0.11</td><td>0.61x</td><td>0.42x</td><td>72.9±0.06</td><td>0.50x</td><td>0.24x</td></tr><tr><td>Pruning* (Zhu)</td><td>73.2</td><td>1.00x</td><td>0.23x</td><td>70.3</td><td>1.00x</td><td>0.10x</td></tr><tr><td>Pruning* (Gale)</td><td>75.6</td><td>1.00x</td><td>0.23x</td><td>73.9</td><td>1.00x</td><td>0.10x</td></tr><tr><td>Pruning1.5× (Gale)</td><td>76.5</td><td>1.50x</td><td>0.23x</td><td></td><td>1.50x</td><td></td></tr><tr><td></td><td></td><td></td><td></td><td>75.2</td><td></td><td>0.10x</td></tr></table>",
+                                            "type": "table",
+                                            "image_path": "6456898747c525bc970340414a9c373557a13cad7006f88915dcf071144cbb2e.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 1,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        122,
+                                        80,
+                                        489,
+                                        162.66666666666669
+                                    ],
+                                    "spans": [],
+                                    "index": 0
+                                },
+                                {
+                                    "bbox": [
+                                        122,
+                                        162.66666666666669,
+                                        489,
+                                        245.33333333333337
+                                    ],
+                                    "spans": [],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        122,
+                                        245.33333333333337,
+                                        489,
+                                        328.00000000000006
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 1
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        336,
+                        505,
+                        402
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                335,
+                                505,
+                                348
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        335,
+                                        384,
+                                        348
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Table 2: Performance and cost of sparse training methods on training",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        384,
+                                        336,
+                                        404,
+                                        347
+                                    ],
+                                    "score": 0.87,
+                                    "content": "80 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        404,
+                                        335,
+                                        421,
+                                        348
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        422,
+                                        336,
+                                        442,
+                                        347
+                                    ],
+                                    "score": 0.88,
+                                    "content": "90 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        442,
+                                        335,
+                                        505,
+                                        348
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sparse ResNet-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                347,
+                                505,
+                                360
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        347,
+                                        505,
+                                        360
+                                    ],
+                                    "score": 1.0,
+                                    "content": "50s. FLOPs needed for training and test are normalized with the FLOPs of a dense model (see",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                358,
+                                505,
+                                370
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        358,
+                                        505,
+                                        370
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Appendix G for details on how FLOPs are calculated). Methods with a subscript indicate a rescaled",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                369,
+                                505,
+                                382
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        369,
+                                        505,
+                                        382
+                                    ],
+                                    "score": 1.0,
+                                    "content": "training time, whereas ‘*’ indicates reported results. (ERK) corresponds to the sparse networks with",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                378,
+                                506,
+                                394
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        378,
+                                        279,
+                                        394
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Erdos-Renyi-Kernel sparsity distribution.˝",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        280,
+                                        380,
+                                        312,
+                                        392
+                                    ],
+                                    "score": 0.86,
+                                    "content": "\\mathrm { R i g L } _ { \\mathrm { 5 } \\times }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        313,
+                                        378,
+                                        382,
+                                        394
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(ERK) achieves",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        382,
+                                        380,
+                                        410,
+                                        391
+                                    ],
+                                    "score": 0.86,
+                                    "content": "7 7 . 1 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        410,
+                                        378,
+                                        506,
+                                        394
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Top-1 Accuracy using",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                391,
+                                383,
+                                403
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        391,
+                                        126,
+                                        403
+                                    ],
+                                    "score": 1.0,
+                                    "content": "only",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        127,
+                                        391,
+                                        146,
+                                        402
+                                    ],
+                                    "score": 0.87,
+                                    "content": "20 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        147,
+                                        391,
+                                        306,
+                                        403
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of the parameters of a dense model and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        306,
+                                        392,
+                                        326,
+                                        401
+                                    ],
+                                    "score": 0.86,
+                                    "content": "42 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        326,
+                                        391,
+                                        383,
+                                        403
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of its FLOPs.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        }
+                    ],
+                    "index": 5.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        411,
+                        505,
+                        444
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                411,
+                                505,
+                                424
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        411,
+                                        505,
+                                        424
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tions randomly. Methods that use gradient information to grow new connections (RigL and SNFS)",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                422,
+                                504,
+                                434
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        422,
+                                        504,
+                                        434
+                                    ],
+                                    "score": 1.0,
+                                    "content": "obtain higher accuracies, but RigL achieves the highest accuracy and does so while consistently",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                434,
+                                297,
+                                445
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        434,
+                                        297,
+                                        445
+                                    ],
+                                    "score": 1.0,
+                                    "content": "requiring fewer FLOPs than the other methods.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        }
+                    ],
+                    "index": 10
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        450,
+                        505,
+                        549
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                450,
+                                506,
+                                463
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        450,
+                                        506,
+                                        463
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Given that different applications or scenarios might require a limit on the number of FLOPs for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                461,
+                                505,
+                                473
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        461,
+                                        505,
+                                        473
+                                    ],
+                                    "score": 1.0,
+                                    "content": "inference, we investigate the performance of our method at various sparsity levels. As mentioned",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                473,
+                                506,
+                                485
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        473,
+                                        506,
+                                        485
+                                    ],
+                                    "score": 1.0,
+                                    "content": "previously, one strength of our method is that its resource requirements are constant throughout",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                483,
+                                505,
+                                496
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        483,
+                                        505,
+                                        496
+                                    ],
+                                    "score": 1.0,
+                                    "content": "training and we can choose the level of sparsity that fits our training and/or inference constraints.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                495,
+                                505,
+                                507
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        495,
+                                        505,
+                                        507
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In Figure 2-right we show the performance of our method at different sparsities and compare them",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                505,
+                                505,
+                                518
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        505,
+                                        336,
+                                        518
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with the pruning results of Gale et al. (2019), which uses",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        336,
+                                        505,
+                                        355,
+                                        516
+                                    ],
+                                    "score": 0.69,
+                                    "content": "1 . 5 \\mathrm { x }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        355,
+                                        505,
+                                        505,
+                                        518
+                                    ],
+                                    "score": 1.0,
+                                    "content": "training steps, relative to the original",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                515,
+                                506,
+                                529
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        515,
+                                        506,
+                                        529
+                                    ],
+                                    "score": 1.0,
+                                    "content": "32k iterations. To make a fair comparison with regards to FLOPs, we scale the learning schedule",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                527,
+                                505,
+                                540
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        527,
+                                        202,
+                                        540
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of all other methods by",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        202,
+                                        528,
+                                        214,
+                                        537
+                                    ],
+                                    "score": 0.34,
+                                    "content": "{ 5 } \\mathbf { x }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        214,
+                                        527,
+                                        505,
+                                        540
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Note that even after extending the training, it takes less FLOPs to train",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                538,
+                                505,
+                                551
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        538,
+                                        275,
+                                        551
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sparse networks using rigL (except for the",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        275,
+                                        538,
+                                        295,
+                                        549
+                                    ],
+                                    "score": 0.86,
+                                    "content": "80 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        295,
+                                        538,
+                                        505,
+                                        551
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sparse RigL-ERK) compared to the pruning method.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        }
+                    ],
+                    "index": 16
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        554,
+                        505,
+                        632
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                554,
+                                506,
+                                567
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        554,
+                                        506,
+                                        567
+                                    ],
+                                    "score": 1.0,
+                                    "content": "RigL, our method with constant sparsity distribution, exceeds the performance of magnitude based",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                566,
+                                505,
+                                578
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        566,
+                                        505,
+                                        578
+                                    ],
+                                    "score": 1.0,
+                                    "content": "iterative pruning in all sparsity levels while requiring less FLOPs to train. Sparse networks that",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                577,
+                                504,
+                                590
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        577,
+                                        504,
+                                        590
+                                    ],
+                                    "score": 1.0,
+                                    "content": "use Erdos-Renyi-Kernel (ERK) ˝ sparsity distribution obtains even greater performance. For exam-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                588,
+                                504,
+                                600
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        588,
+                                        189,
+                                        600
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ple ResNet-50 with",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        189,
+                                        588,
+                                        217,
+                                        599
+                                    ],
+                                    "score": 0.88,
+                                    "content": "9 6 . 5 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        217,
+                                        588,
+                                        345,
+                                        600
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sparsity achieves a remarkable",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        346,
+                                        588,
+                                        378,
+                                        599
+                                    ],
+                                    "score": 0.88,
+                                    "content": "7 2 . 7 5 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        379,
+                                        588,
+                                        481,
+                                        600
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Top-1 Accuracy, around",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        482,
+                                        588,
+                                        504,
+                                        599
+                                    ],
+                                    "score": 0.85,
+                                    "content": "3 . 5 \\%",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                599,
+                                505,
+                                612
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        599,
+                                        505,
+                                        612
+                                    ],
+                                    "score": 1.0,
+                                    "content": "higher than the extended magnitude pruning results reported by Gale et al. (2019). As observed",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                610,
+                                505,
+                                622
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        610,
+                                        505,
+                                        622
+                                    ],
+                                    "score": 1.0,
+                                    "content": "earlier, smaller dense models (with the same number of parameters) or sparse models with a static",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                621,
+                                314,
+                                633
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        621,
+                                        314,
+                                        633
+                                    ],
+                                    "score": 1.0,
+                                    "content": "connectivity can not perform at a comparable level.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        }
+                    ],
+                    "index": 24
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        637,
+                        504,
+                        704
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                637,
+                                506,
+                                650
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        637,
+                                        506,
+                                        650
+                                    ],
+                                    "score": 1.0,
+                                    "content": "A more fine grained comparison of sparse training methods is presented in Table 2. Methods us-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                649,
+                                506,
+                                662
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        649,
+                                        506,
+                                        662
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ing uniform sparsity distribution and whose FLOP/memory footprint scales directly with (1-S) are",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                660,
+                                506,
+                                672
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        660,
+                                        506,
+                                        672
+                                    ],
+                                    "score": 1.0,
+                                    "content": "placed in the first sub-group of the table. The second sub-group includes DSR and networks with",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                670,
+                                506,
+                                683
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        670,
+                                        506,
+                                        683
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ERK sparsity distribution which require a higher number of FLOPs for inference with same param-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                681,
+                                505,
+                                694
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        681,
+                                        505,
+                                        694
+                                    ],
+                                    "score": 1.0,
+                                    "content": "eter count. The final sub-group includes methods that require the space and the work proportional",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                693,
+                                213,
+                                704
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        693,
+                                        213,
+                                        704
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to training a dense model.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        }
+                    ],
+                    "index": 30.5
+                }
+            ],
+            "page_idx": 5,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        107,
+                        27,
+                        308,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                107,
+                                26,
+                                308,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        26,
+                                        308,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Under review as a conference paper at ICLR 2020",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        122,
+                        722,
+                        175,
+                        731
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                120,
+                                720,
+                                177,
+                                732
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        120,
+                                        720,
+                                        177,
+                                        732
+                                    ],
+                                    "score": 1.0,
+                                    "content": "∗approximated",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        302,
+                        752,
+                        309,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                302,
+                                751,
+                                310,
+                                762
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        302,
+                                        751,
+                                        310,
+                                        762
+                                    ],
+                                    "score": 1.0,
+                                    "content": "6",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "table",
+                    "bbox": [
+                        122,
+                        80,
+                        489,
+                        328
+                    ],
+                    "blocks": [
+                        {
+                            "type": "table_body",
+                            "bbox": [
+                                122,
+                                80,
+                                489,
+                                328
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        122,
+                                        80,
+                                        489,
+                                        328
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                122,
+                                                80,
+                                                489,
+                                                328
+                                            ],
+                                            "score": 0.986,
+                                            "html": "<table><tr><td>Method</td><td>Top-1 Accuracy</td><td>FLOPs (Train)</td><td>FLOPs (Test)</td><td>Top-1 Accuracy</td><td>FLOPs (Train)</td><td>FLOPs (Test)</td></tr><tr><td>Dense</td><td>76.8±0.09</td><td>1x (3.2e18)</td><td>1x (8.2e9)</td><td colspan=\"3\"></td></tr><tr><td>Static</td><td colspan=\"3\">S=0.8</td><td colspan=\"3\">S=0.9</td></tr><tr><td>SNIP</td><td>70.6±0.06 72.0±0.10</td><td>0.23x 0.23x</td><td>0.23x 0.23x</td><td>65.8±0.04 67.2±0.12</td><td>0.10x 0.10x</td><td>0.10x 0.10x</td></tr><tr><td>Small-Dense</td><td>72.1±0.12</td><td>0.20x</td><td>0.20x</td><td>68.9±0.10</td><td>0.12x</td><td>0.12x</td></tr><tr><td>SET</td><td>72.9±0.39</td><td>0.23x</td><td>0.23x</td><td>69.6±0.23</td><td>0.10x</td><td>0.10x</td></tr><tr><td>RigL</td><td>74.6±0.06</td><td>0.23x</td><td>0.23x</td><td>72.0±0.05</td><td>0.10x</td><td>0.10x</td></tr><tr><td>Small-Dense5x RigL5x</td><td>73.9±0.07</td><td>1.01x</td><td>0.20x</td><td>71.3±0.10</td><td>0.60x</td><td>0.12x</td></tr><tr><td>Static (ERK)</td><td>76.6±0.06</td><td>1.14x</td><td>0.23x</td><td>75.7±0.06</td><td>0.52x</td><td>0.10x</td></tr><tr><td>DSR*</td><td>72.1±0.04 73.3</td><td>0.42x 0.40x</td><td>0.42x 0.40x</td><td>67.7±0.12</td><td>0.24x</td><td>0.24x</td></tr><tr><td></td><td></td><td></td><td></td><td>71.6</td><td>0.30x</td><td>0.30x</td></tr><tr><td>RigL (ERK)</td><td>75.1±0.05</td><td>0.42x</td><td>0.42x</td><td>73.0±0.04</td><td>0.25x</td><td>0.24x</td></tr><tr><td>RigL5x (ERK)</td><td>77.1±0.06</td><td>2.09x</td><td>0.42x</td><td>76.4±0.05</td><td>1.23x</td><td>0.24x</td></tr><tr><td>SNFS*</td><td>74.2</td><td>n/a</td><td>n/a</td><td>72.3</td><td>n/a</td><td>n/a</td></tr><tr><td>SNFS (ERK)</td><td>75.2±0.11</td><td>0.61x</td><td>0.42x</td><td>72.9±0.06</td><td>0.50x</td><td>0.24x</td></tr><tr><td>Pruning* (Zhu)</td><td>73.2</td><td>1.00x</td><td>0.23x</td><td>70.3</td><td>1.00x</td><td>0.10x</td></tr><tr><td>Pruning* (Gale)</td><td>75.6</td><td>1.00x</td><td>0.23x</td><td>73.9</td><td>1.00x</td><td>0.10x</td></tr><tr><td>Pruning1.5× (Gale)</td><td>76.5</td><td>1.50x</td><td>0.23x</td><td></td><td>1.50x</td><td></td></tr><tr><td></td><td></td><td></td><td></td><td>75.2</td><td></td><td>0.10x</td></tr></table>",
+                                            "type": "table",
+                                            "image_path": "6456898747c525bc970340414a9c373557a13cad7006f88915dcf071144cbb2e.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 1,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        122,
+                                        80,
+                                        489,
+                                        162.66666666666669
+                                    ],
+                                    "spans": [],
+                                    "index": 0
+                                },
+                                {
+                                    "bbox": [
+                                        122,
+                                        162.66666666666669,
+                                        489,
+                                        245.33333333333337
+                                    ],
+                                    "spans": [],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        122,
+                                        245.33333333333337,
+                                        489,
+                                        328.00000000000006
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 1
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        336,
+                        505,
+                        402
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                335,
+                                505,
+                                348
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        335,
+                                        384,
+                                        348
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Table 2: Performance and cost of sparse training methods on training",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        384,
+                                        336,
+                                        404,
+                                        347
+                                    ],
+                                    "score": 0.87,
+                                    "content": "80 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        404,
+                                        335,
+                                        421,
+                                        348
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        422,
+                                        336,
+                                        442,
+                                        347
+                                    ],
+                                    "score": 0.88,
+                                    "content": "90 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        442,
+                                        335,
+                                        505,
+                                        348
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sparse ResNet-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                347,
+                                505,
+                                360
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        347,
+                                        505,
+                                        360
+                                    ],
+                                    "score": 1.0,
+                                    "content": "50s. FLOPs needed for training and test are normalized with the FLOPs of a dense model (see",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                358,
+                                505,
+                                370
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        358,
+                                        505,
+                                        370
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Appendix G for details on how FLOPs are calculated). Methods with a subscript indicate a rescaled",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                369,
+                                505,
+                                382
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        369,
+                                        505,
+                                        382
+                                    ],
+                                    "score": 1.0,
+                                    "content": "training time, whereas ‘*’ indicates reported results. (ERK) corresponds to the sparse networks with",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                378,
+                                506,
+                                394
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        378,
+                                        279,
+                                        394
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Erdos-Renyi-Kernel sparsity distribution.˝",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        280,
+                                        380,
+                                        312,
+                                        392
+                                    ],
+                                    "score": 0.86,
+                                    "content": "\\mathrm { R i g L } _ { \\mathrm { 5 } \\times }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        313,
+                                        378,
+                                        382,
+                                        394
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(ERK) achieves",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        382,
+                                        380,
+                                        410,
+                                        391
+                                    ],
+                                    "score": 0.86,
+                                    "content": "7 7 . 1 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        410,
+                                        378,
+                                        506,
+                                        394
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Top-1 Accuracy using",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                391,
+                                383,
+                                403
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        391,
+                                        126,
+                                        403
+                                    ],
+                                    "score": 1.0,
+                                    "content": "only",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        127,
+                                        391,
+                                        146,
+                                        402
+                                    ],
+                                    "score": 0.87,
+                                    "content": "20 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        147,
+                                        391,
+                                        306,
+                                        403
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of the parameters of a dense model and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        306,
+                                        392,
+                                        326,
+                                        401
+                                    ],
+                                    "score": 0.86,
+                                    "content": "42 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        326,
+                                        391,
+                                        383,
+                                        403
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of its FLOPs.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        }
+                    ],
+                    "index": 5.5,
+                    "bbox_fs": [
+                        105,
+                        335,
+                        506,
+                        403
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        411,
+                        505,
+                        444
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                411,
+                                505,
+                                424
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        411,
+                                        505,
+                                        424
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tions randomly. Methods that use gradient information to grow new connections (RigL and SNFS)",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                422,
+                                504,
+                                434
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        422,
+                                        504,
+                                        434
+                                    ],
+                                    "score": 1.0,
+                                    "content": "obtain higher accuracies, but RigL achieves the highest accuracy and does so while consistently",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                434,
+                                297,
+                                445
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        434,
+                                        297,
+                                        445
+                                    ],
+                                    "score": 1.0,
+                                    "content": "requiring fewer FLOPs than the other methods.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        }
+                    ],
+                    "index": 10,
+                    "bbox_fs": [
+                        105,
+                        411,
+                        505,
+                        445
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        450,
+                        505,
+                        549
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                450,
+                                506,
+                                463
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        450,
+                                        506,
+                                        463
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Given that different applications or scenarios might require a limit on the number of FLOPs for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                461,
+                                505,
+                                473
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        461,
+                                        505,
+                                        473
+                                    ],
+                                    "score": 1.0,
+                                    "content": "inference, we investigate the performance of our method at various sparsity levels. As mentioned",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                473,
+                                506,
+                                485
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        473,
+                                        506,
+                                        485
+                                    ],
+                                    "score": 1.0,
+                                    "content": "previously, one strength of our method is that its resource requirements are constant throughout",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                483,
+                                505,
+                                496
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        483,
+                                        505,
+                                        496
+                                    ],
+                                    "score": 1.0,
+                                    "content": "training and we can choose the level of sparsity that fits our training and/or inference constraints.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                495,
+                                505,
+                                507
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        495,
+                                        505,
+                                        507
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In Figure 2-right we show the performance of our method at different sparsities and compare them",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                505,
+                                505,
+                                518
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        505,
+                                        336,
+                                        518
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with the pruning results of Gale et al. (2019), which uses",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        336,
+                                        505,
+                                        355,
+                                        516
+                                    ],
+                                    "score": 0.69,
+                                    "content": "1 . 5 \\mathrm { x }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        355,
+                                        505,
+                                        505,
+                                        518
+                                    ],
+                                    "score": 1.0,
+                                    "content": "training steps, relative to the original",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                515,
+                                506,
+                                529
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        515,
+                                        506,
+                                        529
+                                    ],
+                                    "score": 1.0,
+                                    "content": "32k iterations. To make a fair comparison with regards to FLOPs, we scale the learning schedule",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                527,
+                                505,
+                                540
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        527,
+                                        202,
+                                        540
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of all other methods by",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        202,
+                                        528,
+                                        214,
+                                        537
+                                    ],
+                                    "score": 0.34,
+                                    "content": "{ 5 } \\mathbf { x }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        214,
+                                        527,
+                                        505,
+                                        540
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Note that even after extending the training, it takes less FLOPs to train",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                538,
+                                505,
+                                551
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        538,
+                                        275,
+                                        551
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sparse networks using rigL (except for the",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        275,
+                                        538,
+                                        295,
+                                        549
+                                    ],
+                                    "score": 0.86,
+                                    "content": "80 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        295,
+                                        538,
+                                        505,
+                                        551
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sparse RigL-ERK) compared to the pruning method.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        }
+                    ],
+                    "index": 16,
+                    "bbox_fs": [
+                        105,
+                        450,
+                        506,
+                        551
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        554,
+                        505,
+                        632
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                554,
+                                506,
+                                567
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        554,
+                                        506,
+                                        567
+                                    ],
+                                    "score": 1.0,
+                                    "content": "RigL, our method with constant sparsity distribution, exceeds the performance of magnitude based",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                566,
+                                505,
+                                578
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        566,
+                                        505,
+                                        578
+                                    ],
+                                    "score": 1.0,
+                                    "content": "iterative pruning in all sparsity levels while requiring less FLOPs to train. Sparse networks that",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                577,
+                                504,
+                                590
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        577,
+                                        504,
+                                        590
+                                    ],
+                                    "score": 1.0,
+                                    "content": "use Erdos-Renyi-Kernel (ERK) ˝ sparsity distribution obtains even greater performance. For exam-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                588,
+                                504,
+                                600
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        588,
+                                        189,
+                                        600
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ple ResNet-50 with",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        189,
+                                        588,
+                                        217,
+                                        599
+                                    ],
+                                    "score": 0.88,
+                                    "content": "9 6 . 5 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        217,
+                                        588,
+                                        345,
+                                        600
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sparsity achieves a remarkable",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        346,
+                                        588,
+                                        378,
+                                        599
+                                    ],
+                                    "score": 0.88,
+                                    "content": "7 2 . 7 5 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        379,
+                                        588,
+                                        481,
+                                        600
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Top-1 Accuracy, around",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        482,
+                                        588,
+                                        504,
+                                        599
+                                    ],
+                                    "score": 0.85,
+                                    "content": "3 . 5 \\%",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                599,
+                                505,
+                                612
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        599,
+                                        505,
+                                        612
+                                    ],
+                                    "score": 1.0,
+                                    "content": "higher than the extended magnitude pruning results reported by Gale et al. (2019). As observed",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                610,
+                                505,
+                                622
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        610,
+                                        505,
+                                        622
+                                    ],
+                                    "score": 1.0,
+                                    "content": "earlier, smaller dense models (with the same number of parameters) or sparse models with a static",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                621,
+                                314,
+                                633
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        621,
+                                        314,
+                                        633
+                                    ],
+                                    "score": 1.0,
+                                    "content": "connectivity can not perform at a comparable level.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        }
+                    ],
+                    "index": 24,
+                    "bbox_fs": [
+                        105,
+                        554,
+                        506,
+                        633
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        637,
+                        504,
+                        704
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                637,
+                                506,
+                                650
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        637,
+                                        506,
+                                        650
+                                    ],
+                                    "score": 1.0,
+                                    "content": "A more fine grained comparison of sparse training methods is presented in Table 2. Methods us-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                649,
+                                506,
+                                662
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        649,
+                                        506,
+                                        662
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ing uniform sparsity distribution and whose FLOP/memory footprint scales directly with (1-S) are",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                660,
+                                506,
+                                672
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        660,
+                                        506,
+                                        672
+                                    ],
+                                    "score": 1.0,
+                                    "content": "placed in the first sub-group of the table. The second sub-group includes DSR and networks with",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                670,
+                                506,
+                                683
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        670,
+                                        506,
+                                        683
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ERK sparsity distribution which require a higher number of FLOPs for inference with same param-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                681,
+                                505,
+                                694
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        681,
+                                        505,
+                                        694
+                                    ],
+                                    "score": 1.0,
+                                    "content": "eter count. The final sub-group includes methods that require the space and the work proportional",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                693,
+                                213,
+                                704
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        693,
+                                        213,
+                                        704
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to training a dense model.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        }
+                    ],
+                    "index": 30.5,
+                    "bbox_fs": [
+                        105,
+                        637,
+                        506,
+                        704
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "image",
+                    "bbox": [
+                        113,
+                        84,
+                        276,
+                        212
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                113,
+                                84,
+                                276,
+                                212
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        113,
+                                        84,
+                                        276,
+                                        212
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                113,
+                                                84,
+                                                276,
+                                                212
+                                            ],
+                                            "score": 0.84,
+                                            "type": "image",
+                                            "image_path": "8ad076c5d43264442c51ef52152072882fb12f6803ab3f5d550b0ad517566e72.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 4,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        113,
+                                        84,
+                                        276,
+                                        98.22222222222223
+                                    ],
+                                    "spans": [],
+                                    "index": 0
+                                },
+                                {
+                                    "bbox": [
+                                        113,
+                                        98.22222222222223,
+                                        276,
+                                        112.44444444444446
+                                    ],
+                                    "spans": [],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        113,
+                                        112.44444444444446,
+                                        276,
+                                        126.66666666666669
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                },
+                                {
+                                    "bbox": [
+                                        113,
+                                        126.66666666666669,
+                                        276,
+                                        140.8888888888889
+                                    ],
+                                    "spans": [],
+                                    "index": 3
+                                },
+                                {
+                                    "bbox": [
+                                        113,
+                                        140.8888888888889,
+                                        276,
+                                        155.11111111111114
+                                    ],
+                                    "spans": [],
+                                    "index": 4
+                                },
+                                {
+                                    "bbox": [
+                                        113,
+                                        155.11111111111114,
+                                        276,
+                                        169.33333333333337
+                                    ],
+                                    "spans": [],
+                                    "index": 5
+                                },
+                                {
+                                    "bbox": [
+                                        113,
+                                        169.33333333333337,
+                                        276,
+                                        183.5555555555556
+                                    ],
+                                    "spans": [],
+                                    "index": 6
+                                },
+                                {
+                                    "bbox": [
+                                        113,
+                                        183.5555555555556,
+                                        276,
+                                        197.77777777777783
+                                    ],
+                                    "spans": [],
+                                    "index": 7
+                                },
+                                {
+                                    "bbox": [
+                                        113,
+                                        197.77777777777783,
+                                        276,
+                                        212.00000000000006
+                                    ],
+                                    "spans": [],
+                                    "index": 8
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                106,
+                                226,
+                                506,
+                                303
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        226,
+                                        505,
+                                        238
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                226,
+                                                505,
+                                                238
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 3: (left) RigL significantly improves the performance of Sparse MobileNets on ImageNet-",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 19
+                                },
+                                {
+                                    "bbox": [
+                                        104,
+                                        235,
+                                        506,
+                                        250
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                104,
+                                                235,
+                                                506,
+                                                250
+                                            ],
+                                            "score": 1.0,
+                                            "content": "2012 dataset and exceeds the pruning results reported by Zhu & Gupta (2018). Performance of",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 20
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        248,
+                                        505,
+                                        260
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                248,
+                                                505,
+                                                260
+                                            ],
+                                            "score": 1.0,
+                                            "content": "the dense MobileNets are indicated with red lines. (right) Performance of sparse MobileNet-v1",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 21
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        258,
+                                        506,
+                                        271
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                258,
+                                                506,
+                                                271
+                                            ],
+                                            "score": 1.0,
+                                            "content": "architectures presented with their inference FLOPs. Networks with ERK distribution get better",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 22
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        269,
+                                        506,
+                                        283
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                269,
+                                                506,
+                                                283
+                                            ],
+                                            "score": 1.0,
+                                            "content": "performance with the same number of parameters but take more FLOPs to run. Training wider",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 23
+                                },
+                                {
+                                    "bbox": [
+                                        104,
+                                        280,
+                                        506,
+                                        293
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                104,
+                                                280,
+                                                506,
+                                                293
+                                            ],
+                                            "score": 1.0,
+                                            "content": "sparse models with RigL (Big-Sparse) yields a significant performance improvement over the dense",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 24
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        291,
+                                        137,
+                                        302
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                291,
+                                                137,
+                                                302
+                                            ],
+                                            "score": 1.0,
+                                            "content": "model.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 25
+                                }
+                            ],
+                            "index": 22
+                        }
+                    ],
+                    "index": 13.0
+                },
+                {
+                    "type": "table",
+                    "bbox": [
+                        280,
+                        79,
+                        515,
+                        218
+                    ],
+                    "blocks": [
+                        {
+                            "type": "table_body",
+                            "bbox": [
+                                280,
+                                79,
+                                515,
+                                218
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        280,
+                                        79,
+                                        515,
+                                        218
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                280,
+                                                79,
+                                                515,
+                                                218
+                                            ],
+                                            "score": 0.757,
+                                            "html": "<table><tr><td rowspan=1 colspan=1>S</td><td rowspan=1 colspan=1>Method</td><td rowspan=1 colspan=1>Top-1</td><td rowspan=1 colspan=1>FLOPs</td></tr><tr><td rowspan=1 colspan=1>0.75</td><td rowspan=1 colspan=1>Small-Dense5×Pruning (Zhu)RigL5×RigL5x(ERK)</td><td rowspan=1 colspan=1>66.0±0.1167.771.5±0.0671.9±0.01</td><td rowspan=1 colspan=1>0.23x0.27x0.27x0.52x</td></tr><tr><td rowspan=1 colspan=1>0.90</td><td rowspan=1 colspan=1>Small-Dense5xPruning (Zhu)RigL5×RigL5x(ERK)</td><td rowspan=1 colspan=1>57.7±0.3461.867.0±0.1768.1±0.11</td><td rowspan=1 colspan=1>0.09x0.12x0.12x0.27x</td></tr><tr><td rowspan=1 colspan=1></td><td rowspan=1 colspan=1>Dense</td><td rowspan=1 colspan=1>72.1±0.17</td><td rowspan=1 colspan=1>1x (1.1e9)</td></tr><tr><td rowspan=1 colspan=1>0.75</td><td rowspan=1 colspan=1>Big-Sparse5×Big-Sparse5x(ERK)</td><td rowspan=1 colspan=1>76.4±0.0577.0±0.08</td><td rowspan=1 colspan=1>0.98x1.91x</td></tr></table>",
+                                            "type": "table",
+                                            "image_path": "29d17293935e4ad2e65d8ff2ae86da99bee4c3ed75e6bf42fab1681aa0fc4a94.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 13.5,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        280,
+                                        79,
+                                        515,
+                                        92.9
+                                    ],
+                                    "spans": [],
+                                    "index": 9
+                                },
+                                {
+                                    "bbox": [
+                                        280,
+                                        92.9,
+                                        515,
+                                        106.80000000000001
+                                    ],
+                                    "spans": [],
+                                    "index": 10
+                                },
+                                {
+                                    "bbox": [
+                                        280,
+                                        106.80000000000001,
+                                        515,
+                                        120.70000000000002
+                                    ],
+                                    "spans": [],
+                                    "index": 11
+                                },
+                                {
+                                    "bbox": [
+                                        280,
+                                        120.70000000000002,
+                                        515,
+                                        134.60000000000002
+                                    ],
+                                    "spans": [],
+                                    "index": 12
+                                },
+                                {
+                                    "bbox": [
+                                        280,
+                                        134.60000000000002,
+                                        515,
+                                        148.50000000000003
+                                    ],
+                                    "spans": [],
+                                    "index": 13
+                                },
+                                {
+                                    "bbox": [
+                                        280,
+                                        148.50000000000003,
+                                        515,
+                                        162.40000000000003
+                                    ],
+                                    "spans": [],
+                                    "index": 14
+                                },
+                                {
+                                    "bbox": [
+                                        280,
+                                        162.40000000000003,
+                                        515,
+                                        176.30000000000004
+                                    ],
+                                    "spans": [],
+                                    "index": 15
+                                },
+                                {
+                                    "bbox": [
+                                        280,
+                                        176.30000000000004,
+                                        515,
+                                        190.20000000000005
+                                    ],
+                                    "spans": [],
+                                    "index": 16
+                                },
+                                {
+                                    "bbox": [
+                                        280,
+                                        190.20000000000005,
+                                        515,
+                                        204.10000000000005
+                                    ],
+                                    "spans": [],
+                                    "index": 17
+                                },
+                                {
+                                    "bbox": [
+                                        280,
+                                        204.10000000000005,
+                                        515,
+                                        218.00000000000006
+                                    ],
+                                    "spans": [],
+                                    "index": 18
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 13.5
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        326,
+                        194,
+                        338
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                324,
+                                196,
+                                339
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        324,
+                                        196,
+                                        339
+                                    ],
+                                    "score": 1.0,
+                                    "content": "4.1.2 MOBILENET",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        }
+                    ],
+                    "index": 26
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        347,
+                        505,
+                        402
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                346,
+                                505,
+                                360
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        346,
+                                        505,
+                                        360
+                                    ],
+                                    "score": 1.0,
+                                    "content": "MobileNet is a compact architecture that performs remarkably well in resource constrained settings.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                358,
+                                505,
+                                370
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        358,
+                                        505,
+                                        370
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Due to its compact nature with separable convolutions it is known to be difficult to sparsify (Zhu",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                368,
+                                505,
+                                381
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        368,
+                                        505,
+                                        381
+                                    ],
+                                    "score": 1.0,
+                                    "content": "& Gupta, 2018). In this section we apply our method to MobileNet-v1 (Howard et al., 2017) and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                378,
+                                506,
+                                394
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        378,
+                                        506,
+                                        394
+                                    ],
+                                    "score": 1.0,
+                                    "content": "MobileNet-v2 (Sandler et al., 2018). Due to its low parameter count we keep the first layer dense,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                390,
+                                432,
+                                404
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        390,
+                                        432,
+                                        404
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and use ERK and Uniform sparsity distributions to sparsify the remaining layers.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        }
+                    ],
+                    "index": 29
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        407,
+                        504,
+                        463
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                408,
+                                505,
+                                420
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        408,
+                                        505,
+                                        420
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The performance of sparse MobileNets trained with RigL as well as the baselines are shown in",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                419,
+                                505,
+                                431
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        419,
+                                        505,
+                                        431
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Figure 3. We do extended training (5x of the original number of steps) for all runs in this section.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                429,
+                                506,
+                                443
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        429,
+                                        506,
+                                        443
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Although MobileNets are more sensitive to sparsity compared to the ResNet-50 architecture, RigL",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                440,
+                                505,
+                                453
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        440,
+                                        505,
+                                        453
+                                    ],
+                                    "score": 1.0,
+                                    "content": "successfully trains sparse MobileNets at high sparsities and exceeds the performance of previously",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                452,
+                                207,
+                                465
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        452,
+                                        207,
+                                        465
+                                    ],
+                                    "score": 1.0,
+                                    "content": "reported pruning results.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        }
+                    ],
+                    "index": 34
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        468,
+                        504,
+                        524
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                467,
+                                506,
+                                482
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        467,
+                                        506,
+                                        482
+                                    ],
+                                    "score": 1.0,
+                                    "content": "To demonstrate the advantages of sparse models, next, we train wider MobileNets while keeping",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                477,
+                                506,
+                                493
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        477,
+                                        506,
+                                        493
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the FLOPs and total number of parameters the same as the dense baseline using sparsity. A sparse",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                488,
+                                506,
+                                504
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        488,
+                                        329,
+                                        504
+                                    ],
+                                    "score": 1.0,
+                                    "content": "MobileNet-v1 with width multiplier 1.98 and constant",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        330,
+                                        491,
+                                        350,
+                                        501
+                                    ],
+                                    "score": 0.85,
+                                    "content": "7 5 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        350,
+                                        488,
+                                        506,
+                                        504
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sparsity has the same FLOPs and pa-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                501,
+                                503,
+                                514
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        501,
+                                        479,
+                                        514
+                                    ],
+                                    "score": 1.0,
+                                    "content": "rameter count as the dense baseline. Training this network with RigL yields an impressive",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        480,
+                                        501,
+                                        503,
+                                        512
+                                    ],
+                                    "score": 0.85,
+                                    "content": "4 . 3 \\%",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                511,
+                                282,
+                                526
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        511,
+                                        282,
+                                        526
+                                    ],
+                                    "score": 1.0,
+                                    "content": "absolute improvement in Top-1 Accuracy.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        }
+                    ],
+                    "index": 39
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        540,
+                        325,
+                        551
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                539,
+                                325,
+                                552
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        539,
+                                        325,
+                                        552
+                                    ],
+                                    "score": 1.0,
+                                    "content": "4.2 CHARACTER LEVEL LANGUAGE MODELLING",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        }
+                    ],
+                    "index": 42
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        561,
+                        504,
+                        649
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                561,
+                                506,
+                                574
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        561,
+                                        506,
+                                        574
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Most prior work has only examined sparse training on vision networks [the exception is the earliest",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                572,
+                                506,
+                                584
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        572,
+                                        506,
+                                        584
+                                    ],
+                                    "score": 1.0,
+                                    "content": "work - Deep Rewiring (Bellec et al., 2017) which trained an LSTM (Hochreiter & Schmidhuber,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                582,
+                                505,
+                                595
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        582,
+                                        505,
+                                        595
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1997) on the TIMIT (Garofolo et al., 1993) dataset]. To fully understand these techniques it is im-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                594,
+                                506,
+                                606
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        594,
+                                        506,
+                                        606
+                                    ],
+                                    "score": 1.0,
+                                    "content": "portant to examine different architectures on different datasets. Kalchbrenner et al. (2018) found",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                604,
+                                506,
+                                619
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        604,
+                                        506,
+                                        619
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sparse GRUs (Cho et al., 2014) to be very effective at modeling speech, however the dataset they",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 47
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                615,
+                                506,
+                                628
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        615,
+                                        506,
+                                        628
+                                    ],
+                                    "score": 1.0,
+                                    "content": "used is not available. We choose a proxy task with similar characteristics (dataset size and vocabu-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 48
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                627,
+                                505,
+                                639
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        627,
+                                        505,
+                                        639
+                                    ],
+                                    "score": 1.0,
+                                    "content": "lary size are approximately the same) - character level language modeling on the publicly available",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 49
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                637,
+                                280,
+                                651
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        637,
+                                        280,
+                                        651
+                                    ],
+                                    "score": 1.0,
+                                    "content": "WikiText-103 (Merity et al., 2016) dataset.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 50
+                        }
+                    ],
+                    "index": 46.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        654,
+                        505,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                654,
+                                506,
+                                667
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        654,
+                                        506,
+                                        667
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Our network consists of a shared embedding with dimensionality 128, a vocabulary size of 256, a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 51
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                665,
+                                505,
+                                678
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        665,
+                                        505,
+                                        678
+                                    ],
+                                    "score": 1.0,
+                                    "content": "GRU with a state size of 512, a readout from the GRU state consisting of two linear layers with",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 52
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                676,
+                                505,
+                                689
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        676,
+                                        505,
+                                        689
+                                    ],
+                                    "score": 1.0,
+                                    "content": "256 units and 128 units respectively. We train the next step prediction task with the standard cross",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 53
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                688,
+                                505,
+                                700
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        688,
+                                        311,
+                                        700
+                                    ],
+                                    "score": 1.0,
+                                    "content": "entropy loss, the Adam optimizer, a learning rate of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        311,
+                                        688,
+                                        336,
+                                        698
+                                    ],
+                                    "score": 0.81,
+                                    "content": "7 e - 4",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        336,
+                                        688,
+                                        476,
+                                        700
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", an L2 regularization coefficient of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        476,
+                                        688,
+                                        501,
+                                        699
+                                    ],
+                                    "score": 0.74,
+                                    "content": "5 e { - 4 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        502,
+                                        688,
+                                        505,
+                                        700
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 54
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                698,
+                                506,
+                                712
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        698,
+                                        506,
+                                        712
+                                    ],
+                                    "score": 1.0,
+                                    "content": "a sequence length of 512, a batch size of 32 and gradient absolute value clipping of values larger",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 55
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                710,
+                                505,
+                                722
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        710,
+                                        505,
+                                        722
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(in magnitude) than 10. Baseline training length is 200,000 iterations. When inducing sparsity",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 56
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                720,
+                                505,
+                                733
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        720,
+                                        505,
+                                        733
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with magnitude pruning (Zhu & Gupta, 2018), we perform pruning between iterations 50,000 and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 57
+                        }
+                    ],
+                    "index": 54
+                }
+            ],
+            "page_idx": 6,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        106,
+                        27,
+                        308,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                26,
+                                308,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        26,
+                                        308,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Under review as a conference paper at ICLR 2020",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        303,
+                        751,
+                        309,
+                        759
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                302,
+                                750,
+                                309,
+                                762
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        302,
+                                        750,
+                                        309,
+                                        762
+                                    ],
+                                    "score": 1.0,
+                                    "content": "7",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "image",
+                    "bbox": [
+                        113,
+                        84,
+                        276,
+                        212
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                113,
+                                84,
+                                276,
+                                212
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        113,
+                                        84,
+                                        276,
+                                        212
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                113,
+                                                84,
+                                                276,
+                                                212
+                                            ],
+                                            "score": 0.84,
+                                            "type": "image",
+                                            "image_path": "8ad076c5d43264442c51ef52152072882fb12f6803ab3f5d550b0ad517566e72.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 4,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        113,
+                                        84,
+                                        276,
+                                        98.22222222222223
+                                    ],
+                                    "spans": [],
+                                    "index": 0
+                                },
+                                {
+                                    "bbox": [
+                                        113,
+                                        98.22222222222223,
+                                        276,
+                                        112.44444444444446
+                                    ],
+                                    "spans": [],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        113,
+                                        112.44444444444446,
+                                        276,
+                                        126.66666666666669
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                },
+                                {
+                                    "bbox": [
+                                        113,
+                                        126.66666666666669,
+                                        276,
+                                        140.8888888888889
+                                    ],
+                                    "spans": [],
+                                    "index": 3
+                                },
+                                {
+                                    "bbox": [
+                                        113,
+                                        140.8888888888889,
+                                        276,
+                                        155.11111111111114
+                                    ],
+                                    "spans": [],
+                                    "index": 4
+                                },
+                                {
+                                    "bbox": [
+                                        113,
+                                        155.11111111111114,
+                                        276,
+                                        169.33333333333337
+                                    ],
+                                    "spans": [],
+                                    "index": 5
+                                },
+                                {
+                                    "bbox": [
+                                        113,
+                                        169.33333333333337,
+                                        276,
+                                        183.5555555555556
+                                    ],
+                                    "spans": [],
+                                    "index": 6
+                                },
+                                {
+                                    "bbox": [
+                                        113,
+                                        183.5555555555556,
+                                        276,
+                                        197.77777777777783
+                                    ],
+                                    "spans": [],
+                                    "index": 7
+                                },
+                                {
+                                    "bbox": [
+                                        113,
+                                        197.77777777777783,
+                                        276,
+                                        212.00000000000006
+                                    ],
+                                    "spans": [],
+                                    "index": 8
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                106,
+                                226,
+                                506,
+                                303
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        226,
+                                        505,
+                                        238
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                226,
+                                                505,
+                                                238
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 3: (left) RigL significantly improves the performance of Sparse MobileNets on ImageNet-",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 19
+                                },
+                                {
+                                    "bbox": [
+                                        104,
+                                        235,
+                                        506,
+                                        250
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                104,
+                                                235,
+                                                506,
+                                                250
+                                            ],
+                                            "score": 1.0,
+                                            "content": "2012 dataset and exceeds the pruning results reported by Zhu & Gupta (2018). Performance of",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 20
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        248,
+                                        505,
+                                        260
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                248,
+                                                505,
+                                                260
+                                            ],
+                                            "score": 1.0,
+                                            "content": "the dense MobileNets are indicated with red lines. (right) Performance of sparse MobileNet-v1",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 21
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        258,
+                                        506,
+                                        271
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                258,
+                                                506,
+                                                271
+                                            ],
+                                            "score": 1.0,
+                                            "content": "architectures presented with their inference FLOPs. Networks with ERK distribution get better",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 22
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        269,
+                                        506,
+                                        283
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                269,
+                                                506,
+                                                283
+                                            ],
+                                            "score": 1.0,
+                                            "content": "performance with the same number of parameters but take more FLOPs to run. Training wider",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 23
+                                },
+                                {
+                                    "bbox": [
+                                        104,
+                                        280,
+                                        506,
+                                        293
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                104,
+                                                280,
+                                                506,
+                                                293
+                                            ],
+                                            "score": 1.0,
+                                            "content": "sparse models with RigL (Big-Sparse) yields a significant performance improvement over the dense",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 24
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        291,
+                                        137,
+                                        302
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                291,
+                                                137,
+                                                302
+                                            ],
+                                            "score": 1.0,
+                                            "content": "model.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 25
+                                }
+                            ],
+                            "index": 22
+                        }
+                    ],
+                    "index": 13.0
+                },
+                {
+                    "type": "table",
+                    "bbox": [
+                        280,
+                        79,
+                        515,
+                        218
+                    ],
+                    "blocks": [
+                        {
+                            "type": "table_body",
+                            "bbox": [
+                                280,
+                                79,
+                                515,
+                                218
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        280,
+                                        79,
+                                        515,
+                                        218
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                280,
+                                                79,
+                                                515,
+                                                218
+                                            ],
+                                            "score": 0.757,
+                                            "html": "<table><tr><td rowspan=1 colspan=1>S</td><td rowspan=1 colspan=1>Method</td><td rowspan=1 colspan=1>Top-1</td><td rowspan=1 colspan=1>FLOPs</td></tr><tr><td rowspan=1 colspan=1>0.75</td><td rowspan=1 colspan=1>Small-Dense5×Pruning (Zhu)RigL5×RigL5x(ERK)</td><td rowspan=1 colspan=1>66.0±0.1167.771.5±0.0671.9±0.01</td><td rowspan=1 colspan=1>0.23x0.27x0.27x0.52x</td></tr><tr><td rowspan=1 colspan=1>0.90</td><td rowspan=1 colspan=1>Small-Dense5xPruning (Zhu)RigL5×RigL5x(ERK)</td><td rowspan=1 colspan=1>57.7±0.3461.867.0±0.1768.1±0.11</td><td rowspan=1 colspan=1>0.09x0.12x0.12x0.27x</td></tr><tr><td rowspan=1 colspan=1></td><td rowspan=1 colspan=1>Dense</td><td rowspan=1 colspan=1>72.1±0.17</td><td rowspan=1 colspan=1>1x (1.1e9)</td></tr><tr><td rowspan=1 colspan=1>0.75</td><td rowspan=1 colspan=1>Big-Sparse5×Big-Sparse5x(ERK)</td><td rowspan=1 colspan=1>76.4±0.0577.0±0.08</td><td rowspan=1 colspan=1>0.98x1.91x</td></tr></table>",
+                                            "type": "table",
+                                            "image_path": "29d17293935e4ad2e65d8ff2ae86da99bee4c3ed75e6bf42fab1681aa0fc4a94.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 13.5,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        280,
+                                        79,
+                                        515,
+                                        92.9
+                                    ],
+                                    "spans": [],
+                                    "index": 9
+                                },
+                                {
+                                    "bbox": [
+                                        280,
+                                        92.9,
+                                        515,
+                                        106.80000000000001
+                                    ],
+                                    "spans": [],
+                                    "index": 10
+                                },
+                                {
+                                    "bbox": [
+                                        280,
+                                        106.80000000000001,
+                                        515,
+                                        120.70000000000002
+                                    ],
+                                    "spans": [],
+                                    "index": 11
+                                },
+                                {
+                                    "bbox": [
+                                        280,
+                                        120.70000000000002,
+                                        515,
+                                        134.60000000000002
+                                    ],
+                                    "spans": [],
+                                    "index": 12
+                                },
+                                {
+                                    "bbox": [
+                                        280,
+                                        134.60000000000002,
+                                        515,
+                                        148.50000000000003
+                                    ],
+                                    "spans": [],
+                                    "index": 13
+                                },
+                                {
+                                    "bbox": [
+                                        280,
+                                        148.50000000000003,
+                                        515,
+                                        162.40000000000003
+                                    ],
+                                    "spans": [],
+                                    "index": 14
+                                },
+                                {
+                                    "bbox": [
+                                        280,
+                                        162.40000000000003,
+                                        515,
+                                        176.30000000000004
+                                    ],
+                                    "spans": [],
+                                    "index": 15
+                                },
+                                {
+                                    "bbox": [
+                                        280,
+                                        176.30000000000004,
+                                        515,
+                                        190.20000000000005
+                                    ],
+                                    "spans": [],
+                                    "index": 16
+                                },
+                                {
+                                    "bbox": [
+                                        280,
+                                        190.20000000000005,
+                                        515,
+                                        204.10000000000005
+                                    ],
+                                    "spans": [],
+                                    "index": 17
+                                },
+                                {
+                                    "bbox": [
+                                        280,
+                                        204.10000000000005,
+                                        515,
+                                        218.00000000000006
+                                    ],
+                                    "spans": [],
+                                    "index": 18
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 13.5
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        326,
+                        194,
+                        338
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                324,
+                                196,
+                                339
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        324,
+                                        196,
+                                        339
+                                    ],
+                                    "score": 1.0,
+                                    "content": "4.1.2 MOBILENET",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        }
+                    ],
+                    "index": 26
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        347,
+                        505,
+                        402
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                346,
+                                505,
+                                360
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        346,
+                                        505,
+                                        360
+                                    ],
+                                    "score": 1.0,
+                                    "content": "MobileNet is a compact architecture that performs remarkably well in resource constrained settings.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                358,
+                                505,
+                                370
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        358,
+                                        505,
+                                        370
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Due to its compact nature with separable convolutions it is known to be difficult to sparsify (Zhu",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                368,
+                                505,
+                                381
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        368,
+                                        505,
+                                        381
+                                    ],
+                                    "score": 1.0,
+                                    "content": "& Gupta, 2018). In this section we apply our method to MobileNet-v1 (Howard et al., 2017) and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                378,
+                                506,
+                                394
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        378,
+                                        506,
+                                        394
+                                    ],
+                                    "score": 1.0,
+                                    "content": "MobileNet-v2 (Sandler et al., 2018). Due to its low parameter count we keep the first layer dense,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                390,
+                                432,
+                                404
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        390,
+                                        432,
+                                        404
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and use ERK and Uniform sparsity distributions to sparsify the remaining layers.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        }
+                    ],
+                    "index": 29,
+                    "bbox_fs": [
+                        105,
+                        346,
+                        506,
+                        404
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        407,
+                        504,
+                        463
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                408,
+                                505,
+                                420
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        408,
+                                        505,
+                                        420
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The performance of sparse MobileNets trained with RigL as well as the baselines are shown in",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                419,
+                                505,
+                                431
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        419,
+                                        505,
+                                        431
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Figure 3. We do extended training (5x of the original number of steps) for all runs in this section.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                429,
+                                506,
+                                443
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        429,
+                                        506,
+                                        443
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Although MobileNets are more sensitive to sparsity compared to the ResNet-50 architecture, RigL",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                440,
+                                505,
+                                453
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        440,
+                                        505,
+                                        453
+                                    ],
+                                    "score": 1.0,
+                                    "content": "successfully trains sparse MobileNets at high sparsities and exceeds the performance of previously",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                452,
+                                207,
+                                465
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        452,
+                                        207,
+                                        465
+                                    ],
+                                    "score": 1.0,
+                                    "content": "reported pruning results.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        }
+                    ],
+                    "index": 34,
+                    "bbox_fs": [
+                        105,
+                        408,
+                        506,
+                        465
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        468,
+                        504,
+                        524
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                467,
+                                506,
+                                482
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        467,
+                                        506,
+                                        482
+                                    ],
+                                    "score": 1.0,
+                                    "content": "To demonstrate the advantages of sparse models, next, we train wider MobileNets while keeping",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                477,
+                                506,
+                                493
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        477,
+                                        506,
+                                        493
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the FLOPs and total number of parameters the same as the dense baseline using sparsity. A sparse",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                488,
+                                506,
+                                504
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        488,
+                                        329,
+                                        504
+                                    ],
+                                    "score": 1.0,
+                                    "content": "MobileNet-v1 with width multiplier 1.98 and constant",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        330,
+                                        491,
+                                        350,
+                                        501
+                                    ],
+                                    "score": 0.85,
+                                    "content": "7 5 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        350,
+                                        488,
+                                        506,
+                                        504
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sparsity has the same FLOPs and pa-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                501,
+                                503,
+                                514
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        501,
+                                        479,
+                                        514
+                                    ],
+                                    "score": 1.0,
+                                    "content": "rameter count as the dense baseline. Training this network with RigL yields an impressive",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        480,
+                                        501,
+                                        503,
+                                        512
+                                    ],
+                                    "score": 0.85,
+                                    "content": "4 . 3 \\%",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                511,
+                                282,
+                                526
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        511,
+                                        282,
+                                        526
+                                    ],
+                                    "score": 1.0,
+                                    "content": "absolute improvement in Top-1 Accuracy.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        }
+                    ],
+                    "index": 39,
+                    "bbox_fs": [
+                        105,
+                        467,
+                        506,
+                        526
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        540,
+                        325,
+                        551
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                539,
+                                325,
+                                552
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        539,
+                                        325,
+                                        552
+                                    ],
+                                    "score": 1.0,
+                                    "content": "4.2 CHARACTER LEVEL LANGUAGE MODELLING",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        }
+                    ],
+                    "index": 42
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        561,
+                        504,
+                        649
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                561,
+                                506,
+                                574
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        561,
+                                        506,
+                                        574
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Most prior work has only examined sparse training on vision networks [the exception is the earliest",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                572,
+                                506,
+                                584
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        572,
+                                        506,
+                                        584
+                                    ],
+                                    "score": 1.0,
+                                    "content": "work - Deep Rewiring (Bellec et al., 2017) which trained an LSTM (Hochreiter & Schmidhuber,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                582,
+                                505,
+                                595
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        582,
+                                        505,
+                                        595
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1997) on the TIMIT (Garofolo et al., 1993) dataset]. To fully understand these techniques it is im-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                594,
+                                506,
+                                606
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        594,
+                                        506,
+                                        606
+                                    ],
+                                    "score": 1.0,
+                                    "content": "portant to examine different architectures on different datasets. Kalchbrenner et al. (2018) found",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                604,
+                                506,
+                                619
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        604,
+                                        506,
+                                        619
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sparse GRUs (Cho et al., 2014) to be very effective at modeling speech, however the dataset they",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 47
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                615,
+                                506,
+                                628
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        615,
+                                        506,
+                                        628
+                                    ],
+                                    "score": 1.0,
+                                    "content": "used is not available. We choose a proxy task with similar characteristics (dataset size and vocabu-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 48
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                627,
+                                505,
+                                639
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        627,
+                                        505,
+                                        639
+                                    ],
+                                    "score": 1.0,
+                                    "content": "lary size are approximately the same) - character level language modeling on the publicly available",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 49
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                637,
+                                280,
+                                651
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        637,
+                                        280,
+                                        651
+                                    ],
+                                    "score": 1.0,
+                                    "content": "WikiText-103 (Merity et al., 2016) dataset.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 50
+                        }
+                    ],
+                    "index": 46.5,
+                    "bbox_fs": [
+                        105,
+                        561,
+                        506,
+                        651
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        654,
+                        505,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                654,
+                                506,
+                                667
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        654,
+                                        506,
+                                        667
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Our network consists of a shared embedding with dimensionality 128, a vocabulary size of 256, a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 51
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                665,
+                                505,
+                                678
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        665,
+                                        505,
+                                        678
+                                    ],
+                                    "score": 1.0,
+                                    "content": "GRU with a state size of 512, a readout from the GRU state consisting of two linear layers with",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 52
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                676,
+                                505,
+                                689
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        676,
+                                        505,
+                                        689
+                                    ],
+                                    "score": 1.0,
+                                    "content": "256 units and 128 units respectively. We train the next step prediction task with the standard cross",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 53
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                688,
+                                505,
+                                700
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        688,
+                                        311,
+                                        700
+                                    ],
+                                    "score": 1.0,
+                                    "content": "entropy loss, the Adam optimizer, a learning rate of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        311,
+                                        688,
+                                        336,
+                                        698
+                                    ],
+                                    "score": 0.81,
+                                    "content": "7 e - 4",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        336,
+                                        688,
+                                        476,
+                                        700
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", an L2 regularization coefficient of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        476,
+                                        688,
+                                        501,
+                                        699
+                                    ],
+                                    "score": 0.74,
+                                    "content": "5 e { - 4 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        502,
+                                        688,
+                                        505,
+                                        700
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 54
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                698,
+                                506,
+                                712
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        698,
+                                        506,
+                                        712
+                                    ],
+                                    "score": 1.0,
+                                    "content": "a sequence length of 512, a batch size of 32 and gradient absolute value clipping of values larger",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 55
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                710,
+                                505,
+                                722
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        710,
+                                        505,
+                                        722
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(in magnitude) than 10. Baseline training length is 200,000 iterations. When inducing sparsity",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 56
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                720,
+                                505,
+                                733
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        720,
+                                        505,
+                                        733
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with magnitude pruning (Zhu & Gupta, 2018), we perform pruning between iterations 50,000 and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 57
+                        }
+                    ],
+                    "index": 54,
+                    "bbox_fs": [
+                        105,
+                        654,
+                        506,
+                        733
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "image",
+                    "bbox": [
+                        112,
+                        81,
+                        494,
+                        226
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                112,
+                                81,
+                                494,
+                                226
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        112,
+                                        81,
+                                        494,
+                                        226
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                112,
+                                                81,
+                                                494,
+                                                226
+                                            ],
+                                            "score": 0.969,
+                                            "type": "image",
+                                            "image_path": "f16a2ec71b43df85e1a6ff38138e0337a2c670eede9c85bef31c854aa486fd2f.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 1,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        112,
+                                        81,
+                                        494,
+                                        129.33333333333334
+                                    ],
+                                    "spans": [],
+                                    "index": 0
+                                },
+                                {
+                                    "bbox": [
+                                        112,
+                                        129.33333333333334,
+                                        494,
+                                        177.66666666666669
+                                    ],
+                                    "spans": [],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        112,
+                                        177.66666666666669,
+                                        494,
+                                        226.00000000000003
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                106,
+                                236,
+                                505,
+                                280
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        235,
+                                        505,
+                                        249
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                235,
+                                                505,
+                                                249
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 4: (left) Final validation loss of various sparse training methods on character level language",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 3
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        247,
+                                        506,
+                                        260
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                247,
+                                                506,
+                                                260
+                                            ],
+                                            "score": 1.0,
+                                            "content": "modelling task. Cross entropy loss is converted to bits (from nats). Performance and the training cost",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 4
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        258,
+                                        505,
+                                        271
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                258,
+                                                505,
+                                                271
+                                            ],
+                                            "score": 1.0,
+                                            "content": "of a dense model is indicated with dashed red lines. (right) Test accuracies of sparse WideResNet-",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 5
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        268,
+                                        212,
+                                        280
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                268,
+                                                212,
+                                                280
+                                            ],
+                                            "score": 1.0,
+                                            "content": "22-2’s on CIFAR-10 task.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 6
+                                }
+                            ],
+                            "index": 4.5
+                        }
+                    ],
+                    "index": 2.75
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        289,
+                        505,
+                        333
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                288,
+                                505,
+                                302
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        288,
+                                        505,
+                                        302
+                                    ],
+                                    "score": 1.0,
+                                    "content": "150,000 with a pruning frequency of 1,000. We initialize sparse networks with a uniform sparsity",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                299,
+                                505,
+                                312
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        299,
+                                        313,
+                                        312
+                                    ],
+                                    "score": 1.0,
+                                    "content": "distribution and use a cosine update schedule with",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        314,
+                                        300,
+                                        350,
+                                        311
+                                    ],
+                                    "score": 0.89,
+                                    "content": "\\alpha = 0 . 1",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        351,
+                                        299,
+                                        369,
+                                        312
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        370,
+                                        300,
+                                        417,
+                                        311
+                                    ],
+                                    "score": 0.89,
+                                    "content": "\\Delta T = 1 0 0",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        417,
+                                        299,
+                                        505,
+                                        312
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Unlike the previous",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                312,
+                                505,
+                                323
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        312,
+                                        505,
+                                        323
+                                    ],
+                                    "score": 1.0,
+                                    "content": "experiments we keep updating the mask until the end of the training; we observed this performed",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                322,
+                                303,
+                                334
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        322,
+                                        303,
+                                        334
+                                    ],
+                                    "score": 1.0,
+                                    "content": "slightly better than stopping at iteration 150,000.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        }
+                    ],
+                    "index": 8.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        339,
+                        505,
+                        405
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                339,
+                                505,
+                                352
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        339,
+                                        505,
+                                        352
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In Figure 4-left we report the validation loss of various solutions at the end of the training. For each",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                350,
+                                505,
+                                362
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        350,
+                                        505,
+                                        362
+                                    ],
+                                    "score": 1.0,
+                                    "content": "method we perform extended runs to see how they scale with increasing training time. As observed",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                361,
+                                505,
+                                374
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        361,
+                                        505,
+                                        374
+                                    ],
+                                    "score": 1.0,
+                                    "content": "before, SET performs worst than the other dynamic training methods and its performance improves",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                372,
+                                505,
+                                384
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        372,
+                                        505,
+                                        384
+                                    ],
+                                    "score": 1.0,
+                                    "content": "only slightly with increased training time. On the other hand the performance of RigL and SNFS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                383,
+                                505,
+                                396
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        383,
+                                        505,
+                                        396
+                                    ],
+                                    "score": 1.0,
+                                    "content": "improves constantly with more training steps. Both of these methods falls short of matching the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                394,
+                                195,
+                                407
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        394,
+                                        195,
+                                        407
+                                    ],
+                                    "score": 1.0,
+                                    "content": "pruning performance.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        }
+                    ],
+                    "index": 13.5
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        109,
+                        420,
+                        279,
+                        432
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                420,
+                                280,
+                                433
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        420,
+                                        280,
+                                        433
+                                    ],
+                                    "score": 1.0,
+                                    "content": "4.3 WIDERESNET-22-2 ON CIFAR-10",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        }
+                    ],
+                    "index": 17
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        441,
+                        505,
+                        519
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                442,
+                                504,
+                                454
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        442,
+                                        504,
+                                        454
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In this section, we evaluate the performance of RigL on CIFAR-10 image classification benchmark.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                452,
+                                505,
+                                464
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        452,
+                                        505,
+                                        464
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We train Wide Residual Network’s (Zagoruyko & Komodakis, 2016) with 22 layers using a width",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                464,
+                                505,
+                                476
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        464,
+                                        505,
+                                        476
+                                    ],
+                                    "score": 1.0,
+                                    "content": "multiplier of 2 for 250 epochs (97656 steps). Learning rate starts at 0.1 and scaled down by a factor",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                475,
+                                506,
+                                487
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        475,
+                                        506,
+                                        487
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of 5 every 30,000 iterations. We use an L2 regularization coefficient of 5e-4, a batch size of 128 and a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                486,
+                                505,
+                                498
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        486,
+                                        505,
+                                        498
+                                    ],
+                                    "score": 1.0,
+                                    "content": "momentum coefficient of 0.9. We keep the hyper-parameters specific to RigL same as the ImageNet",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                497,
+                                505,
+                                508
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        497,
+                                        505,
+                                        508
+                                    ],
+                                    "score": 1.0,
+                                    "content": "experiments, except the final iteration for mask updates; which is adjusted to 75000. Results with",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                507,
+                                351,
+                                521
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        507,
+                                        351,
+                                        521
+                                    ],
+                                    "score": 1.0,
+                                    "content": "different mask update intervals can be found in Appendix H.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        }
+                    ],
+                    "index": 21
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        524,
+                        505,
+                        623
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                523,
+                                505,
+                                537
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        523,
+                                        505,
+                                        537
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Performance of RigL across different sparsity levels is presented in Figure 4-right. Corresponding",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                536,
+                                504,
+                                547
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        536,
+                                        504,
+                                        547
+                                    ],
+                                    "score": 1.0,
+                                    "content": "final training losses of the trained networks can be found in Appendix H. The dense baseline ob-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                546,
+                                505,
+                                559
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        546,
+                                        128,
+                                        559
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tains",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        128,
+                                        546,
+                                        155,
+                                        557
+                                    ],
+                                    "score": 0.87,
+                                    "content": "9 4 . 1 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        155,
+                                        546,
+                                        410,
+                                        559
+                                    ],
+                                    "score": 1.0,
+                                    "content": "test accuracy. Networks with half of the connections removed",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        411,
+                                        546,
+                                        430,
+                                        557
+                                    ],
+                                    "score": 0.85,
+                                    "content": "50 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        431,
+                                        546,
+                                        505,
+                                        559
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sparsity) achieves",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                556,
+                                505,
+                                570
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        556,
+                                        505,
+                                        570
+                                    ],
+                                    "score": 1.0,
+                                    "content": "roughly the same accuracy as the dense baseline. Surprisingly, some of the networks at this spar-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                568,
+                                505,
+                                581
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        568,
+                                        505,
+                                        581
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sity level generalize better than the dense baseline demonstrating the regularization aspect of using",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                579,
+                                505,
+                                593
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        579,
+                                        505,
+                                        593
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sparsity. With increased sparsity, we start to see a performance gap between the Static and Pruning",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                590,
+                                505,
+                                602
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        590,
+                                        505,
+                                        602
+                                    ],
+                                    "score": 1.0,
+                                    "content": "solutions. Training static RigL networks longer seems to have limited effect on the final perfor-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                600,
+                                506,
+                                615
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        600,
+                                        506,
+                                        615
+                                    ],
+                                    "score": 1.0,
+                                    "content": "mance. On the other hand, RigL, matches the performance of pruning using only a fraction of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                612,
+                                293,
+                                624
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        612,
+                                        293,
+                                        624
+                                    ],
+                                    "score": 1.0,
+                                    "content": "resources needed for training a dense network.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        }
+                    ],
+                    "index": 29
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        109,
+                        639,
+                        305,
+                        650
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                638,
+                                307,
+                                651
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        638,
+                                        307,
+                                        651
+                                    ],
+                                    "score": 1.0,
+                                    "content": "4.4 ANALYZING THE PERFORMANCE OF RigL",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        }
+                    ],
+                    "index": 34
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        660,
+                        505,
+                        693
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                659,
+                                505,
+                                672
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        659,
+                                        505,
+                                        672
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In this section we study the effect of sparsity distributions, update schedules, and dynamic connec-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                671,
+                                505,
+                                682
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        671,
+                                        505,
+                                        682
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tions on the performance of our method. The results for SET and SNFS are similar and are discussed",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                681,
+                                204,
+                                693
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        681,
+                                        204,
+                                        693
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in Appendices B and E.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        }
+                    ],
+                    "index": 36
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        699,
+                        504,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                698,
+                                506,
+                                712
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        698,
+                                        506,
+                                        712
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Effect of Mask Initialization: Figure 5-left shows how the sparsity distribution affects the final",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                711,
+                                505,
+                                721
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        711,
+                                        505,
+                                        721
+                                    ],
+                                    "score": 1.0,
+                                    "content": "test accuracy of sparse ResNet-50s trained with RigL. Erdos-R ˝ enyi-Kernel (ERK) performs consis- ´",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                721,
+                                505,
+                                733
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        721,
+                                        505,
+                                        733
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tently better than the other two distributions. ERK automatically allocates more parameters to the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        }
+                    ],
+                    "index": 39
+                }
+            ],
+            "page_idx": 7,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        108,
+                        27,
+                        307,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                26,
+                                308,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        26,
+                                        308,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Under review as a conference paper at ICLR 2020",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        302,
+                        752,
+                        308,
+                        759
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                302,
+                                750,
+                                309,
+                                761
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        302,
+                                        750,
+                                        309,
+                                        761
+                                    ],
+                                    "score": 1.0,
+                                    "content": "8",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "image",
+                    "bbox": [
+                        112,
+                        81,
+                        494,
+                        226
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                112,
+                                81,
+                                494,
+                                226
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        112,
+                                        81,
+                                        494,
+                                        226
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                112,
+                                                81,
+                                                494,
+                                                226
+                                            ],
+                                            "score": 0.969,
+                                            "type": "image",
+                                            "image_path": "f16a2ec71b43df85e1a6ff38138e0337a2c670eede9c85bef31c854aa486fd2f.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 1,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        112,
+                                        81,
+                                        494,
+                                        129.33333333333334
+                                    ],
+                                    "spans": [],
+                                    "index": 0
+                                },
+                                {
+                                    "bbox": [
+                                        112,
+                                        129.33333333333334,
+                                        494,
+                                        177.66666666666669
+                                    ],
+                                    "spans": [],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        112,
+                                        177.66666666666669,
+                                        494,
+                                        226.00000000000003
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                106,
+                                236,
+                                505,
+                                280
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        235,
+                                        505,
+                                        249
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                235,
+                                                505,
+                                                249
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 4: (left) Final validation loss of various sparse training methods on character level language",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 3
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        247,
+                                        506,
+                                        260
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                247,
+                                                506,
+                                                260
+                                            ],
+                                            "score": 1.0,
+                                            "content": "modelling task. Cross entropy loss is converted to bits (from nats). Performance and the training cost",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 4
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        258,
+                                        505,
+                                        271
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                258,
+                                                505,
+                                                271
+                                            ],
+                                            "score": 1.0,
+                                            "content": "of a dense model is indicated with dashed red lines. (right) Test accuracies of sparse WideResNet-",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 5
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        268,
+                                        212,
+                                        280
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                268,
+                                                212,
+                                                280
+                                            ],
+                                            "score": 1.0,
+                                            "content": "22-2’s on CIFAR-10 task.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 6
+                                }
+                            ],
+                            "index": 4.5
+                        }
+                    ],
+                    "index": 2.75
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        289,
+                        505,
+                        333
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                288,
+                                505,
+                                302
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        288,
+                                        505,
+                                        302
+                                    ],
+                                    "score": 1.0,
+                                    "content": "150,000 with a pruning frequency of 1,000. We initialize sparse networks with a uniform sparsity",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                299,
+                                505,
+                                312
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        299,
+                                        313,
+                                        312
+                                    ],
+                                    "score": 1.0,
+                                    "content": "distribution and use a cosine update schedule with",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        314,
+                                        300,
+                                        350,
+                                        311
+                                    ],
+                                    "score": 0.89,
+                                    "content": "\\alpha = 0 . 1",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        351,
+                                        299,
+                                        369,
+                                        312
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        370,
+                                        300,
+                                        417,
+                                        311
+                                    ],
+                                    "score": 0.89,
+                                    "content": "\\Delta T = 1 0 0",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        417,
+                                        299,
+                                        505,
+                                        312
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Unlike the previous",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                312,
+                                505,
+                                323
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        312,
+                                        505,
+                                        323
+                                    ],
+                                    "score": 1.0,
+                                    "content": "experiments we keep updating the mask until the end of the training; we observed this performed",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                322,
+                                303,
+                                334
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        322,
+                                        303,
+                                        334
+                                    ],
+                                    "score": 1.0,
+                                    "content": "slightly better than stopping at iteration 150,000.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        }
+                    ],
+                    "index": 8.5,
+                    "bbox_fs": [
+                        105,
+                        288,
+                        505,
+                        334
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        339,
+                        505,
+                        405
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                339,
+                                505,
+                                352
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        339,
+                                        505,
+                                        352
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In Figure 4-left we report the validation loss of various solutions at the end of the training. For each",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                350,
+                                505,
+                                362
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        350,
+                                        505,
+                                        362
+                                    ],
+                                    "score": 1.0,
+                                    "content": "method we perform extended runs to see how they scale with increasing training time. As observed",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                361,
+                                505,
+                                374
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        361,
+                                        505,
+                                        374
+                                    ],
+                                    "score": 1.0,
+                                    "content": "before, SET performs worst than the other dynamic training methods and its performance improves",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                372,
+                                505,
+                                384
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        372,
+                                        505,
+                                        384
+                                    ],
+                                    "score": 1.0,
+                                    "content": "only slightly with increased training time. On the other hand the performance of RigL and SNFS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                383,
+                                505,
+                                396
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        383,
+                                        505,
+                                        396
+                                    ],
+                                    "score": 1.0,
+                                    "content": "improves constantly with more training steps. Both of these methods falls short of matching the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                394,
+                                195,
+                                407
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        394,
+                                        195,
+                                        407
+                                    ],
+                                    "score": 1.0,
+                                    "content": "pruning performance.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        }
+                    ],
+                    "index": 13.5,
+                    "bbox_fs": [
+                        105,
+                        339,
+                        505,
+                        407
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        109,
+                        420,
+                        279,
+                        432
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                420,
+                                280,
+                                433
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        420,
+                                        280,
+                                        433
+                                    ],
+                                    "score": 1.0,
+                                    "content": "4.3 WIDERESNET-22-2 ON CIFAR-10",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        }
+                    ],
+                    "index": 17
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        441,
+                        505,
+                        519
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                442,
+                                504,
+                                454
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        442,
+                                        504,
+                                        454
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In this section, we evaluate the performance of RigL on CIFAR-10 image classification benchmark.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                452,
+                                505,
+                                464
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        452,
+                                        505,
+                                        464
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We train Wide Residual Network’s (Zagoruyko & Komodakis, 2016) with 22 layers using a width",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                464,
+                                505,
+                                476
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        464,
+                                        505,
+                                        476
+                                    ],
+                                    "score": 1.0,
+                                    "content": "multiplier of 2 for 250 epochs (97656 steps). Learning rate starts at 0.1 and scaled down by a factor",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                475,
+                                506,
+                                487
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        475,
+                                        506,
+                                        487
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of 5 every 30,000 iterations. We use an L2 regularization coefficient of 5e-4, a batch size of 128 and a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                486,
+                                505,
+                                498
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        486,
+                                        505,
+                                        498
+                                    ],
+                                    "score": 1.0,
+                                    "content": "momentum coefficient of 0.9. We keep the hyper-parameters specific to RigL same as the ImageNet",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                497,
+                                505,
+                                508
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        497,
+                                        505,
+                                        508
+                                    ],
+                                    "score": 1.0,
+                                    "content": "experiments, except the final iteration for mask updates; which is adjusted to 75000. Results with",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                507,
+                                351,
+                                521
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        507,
+                                        351,
+                                        521
+                                    ],
+                                    "score": 1.0,
+                                    "content": "different mask update intervals can be found in Appendix H.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        }
+                    ],
+                    "index": 21,
+                    "bbox_fs": [
+                        105,
+                        442,
+                        506,
+                        521
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        524,
+                        505,
+                        623
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                523,
+                                505,
+                                537
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        523,
+                                        505,
+                                        537
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Performance of RigL across different sparsity levels is presented in Figure 4-right. Corresponding",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                536,
+                                504,
+                                547
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        536,
+                                        504,
+                                        547
+                                    ],
+                                    "score": 1.0,
+                                    "content": "final training losses of the trained networks can be found in Appendix H. The dense baseline ob-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                546,
+                                505,
+                                559
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        546,
+                                        128,
+                                        559
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tains",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        128,
+                                        546,
+                                        155,
+                                        557
+                                    ],
+                                    "score": 0.87,
+                                    "content": "9 4 . 1 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        155,
+                                        546,
+                                        410,
+                                        559
+                                    ],
+                                    "score": 1.0,
+                                    "content": "test accuracy. Networks with half of the connections removed",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        411,
+                                        546,
+                                        430,
+                                        557
+                                    ],
+                                    "score": 0.85,
+                                    "content": "50 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        431,
+                                        546,
+                                        505,
+                                        559
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sparsity) achieves",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                556,
+                                505,
+                                570
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        556,
+                                        505,
+                                        570
+                                    ],
+                                    "score": 1.0,
+                                    "content": "roughly the same accuracy as the dense baseline. Surprisingly, some of the networks at this spar-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                568,
+                                505,
+                                581
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        568,
+                                        505,
+                                        581
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sity level generalize better than the dense baseline demonstrating the regularization aspect of using",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                579,
+                                505,
+                                593
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        579,
+                                        505,
+                                        593
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sparsity. With increased sparsity, we start to see a performance gap between the Static and Pruning",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                590,
+                                505,
+                                602
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        590,
+                                        505,
+                                        602
+                                    ],
+                                    "score": 1.0,
+                                    "content": "solutions. Training static RigL networks longer seems to have limited effect on the final perfor-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                600,
+                                506,
+                                615
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        600,
+                                        506,
+                                        615
+                                    ],
+                                    "score": 1.0,
+                                    "content": "mance. On the other hand, RigL, matches the performance of pruning using only a fraction of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                612,
+                                293,
+                                624
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        612,
+                                        293,
+                                        624
+                                    ],
+                                    "score": 1.0,
+                                    "content": "resources needed for training a dense network.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        }
+                    ],
+                    "index": 29,
+                    "bbox_fs": [
+                        105,
+                        523,
+                        506,
+                        624
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        109,
+                        639,
+                        305,
+                        650
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                638,
+                                307,
+                                651
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        638,
+                                        307,
+                                        651
+                                    ],
+                                    "score": 1.0,
+                                    "content": "4.4 ANALYZING THE PERFORMANCE OF RigL",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        }
+                    ],
+                    "index": 34
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        660,
+                        505,
+                        693
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                659,
+                                505,
+                                672
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        659,
+                                        505,
+                                        672
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In this section we study the effect of sparsity distributions, update schedules, and dynamic connec-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                671,
+                                505,
+                                682
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        671,
+                                        505,
+                                        682
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tions on the performance of our method. The results for SET and SNFS are similar and are discussed",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                681,
+                                204,
+                                693
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        681,
+                                        204,
+                                        693
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in Appendices B and E.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        }
+                    ],
+                    "index": 36,
+                    "bbox_fs": [
+                        105,
+                        659,
+                        505,
+                        693
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        699,
+                        504,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                698,
+                                506,
+                                712
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        698,
+                                        506,
+                                        712
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Effect of Mask Initialization: Figure 5-left shows how the sparsity distribution affects the final",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                711,
+                                505,
+                                721
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        711,
+                                        505,
+                                        721
+                                    ],
+                                    "score": 1.0,
+                                    "content": "test accuracy of sparse ResNet-50s trained with RigL. Erdos-R ˝ enyi-Kernel (ERK) performs consis- ´",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                721,
+                                505,
+                                733
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        721,
+                                        505,
+                                        733
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tently better than the other two distributions. ERK automatically allocates more parameters to the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                266,
+                                505,
+                                279
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        266,
+                                        505,
+                                        279
+                                    ],
+                                    "score": 1.0,
+                                    "content": "layers with few parameters by decreasing their sparsities†. This reallocation seems to be crucial for",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                278,
+                                505,
+                                290
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        278,
+                                        505,
+                                        290
+                                    ],
+                                    "score": 1.0,
+                                    "content": "preserving the capacity of the network at high sparsity levels where ERK outperforms other distri-",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                289,
+                                505,
+                                302
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        289,
+                                        505,
+                                        302
+                                    ],
+                                    "score": 1.0,
+                                    "content": "butions by a greater margin. Though it performs better, the ERK distribution requires approximately",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                300,
+                                506,
+                                313
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        300,
+                                        506,
+                                        313
+                                    ],
+                                    "score": 1.0,
+                                    "content": "twice as many FLOPs compared to a uniform distribution. This highlights an interesting trade-off",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                310,
+                                506,
+                                323
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        310,
+                                        506,
+                                        323
+                                    ],
+                                    "score": 1.0,
+                                    "content": "between accuracy and computational efficiency even though both models have the same number of",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                324,
+                                155,
+                                333
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        324,
+                                        155,
+                                        333
+                                    ],
+                                    "score": 1.0,
+                                    "content": "parameters.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 10
+                        }
+                    ],
+                    "index": 39,
+                    "bbox_fs": [
+                        105,
+                        698,
+                        506,
+                        733
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "image",
+                    "bbox": [
+                        113,
+                        81,
+                        493,
+                        226
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                113,
+                                81,
+                                493,
+                                226
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        113,
+                                        81,
+                                        493,
+                                        226
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                113,
+                                                81,
+                                                493,
+                                                226
+                                            ],
+                                            "score": 0.967,
+                                            "type": "image",
+                                            "image_path": "5800160114d0737b3e2907dc82a457c52f3b8ccbcf1ecf67ee038a009c690904.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 1,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        113,
+                                        81,
+                                        493,
+                                        129.33333333333334
+                                    ],
+                                    "spans": [],
+                                    "index": 0
+                                },
+                                {
+                                    "bbox": [
+                                        113,
+                                        129.33333333333334,
+                                        493,
+                                        177.66666666666669
+                                    ],
+                                    "spans": [],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        113,
+                                        177.66666666666669,
+                                        493,
+                                        226.00000000000003
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                106,
+                                236,
+                                504,
+                                259
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        234,
+                                        505,
+                                        250
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                234,
+                                                505,
+                                                250
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 5: (left) Performance of RigL at different sparsities using different sparsity masks (right)",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 3
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        246,
+                                        391,
+                                        259
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                246,
+                                                391,
+                                                259
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Ablation study on cosine schedule. Other methods are in the appendix.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 4
+                                }
+                            ],
+                            "index": 3.5
+                        }
+                    ],
+                    "index": 2.25
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        266,
+                        505,
+                        333
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                266,
+                                505,
+                                279
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        266,
+                                        505,
+                                        279
+                                    ],
+                                    "score": 1.0,
+                                    "content": "layers with few parameters by decreasing their sparsities†. This reallocation seems to be crucial for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                278,
+                                505,
+                                290
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        278,
+                                        505,
+                                        290
+                                    ],
+                                    "score": 1.0,
+                                    "content": "preserving the capacity of the network at high sparsity levels where ERK outperforms other distri-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                289,
+                                505,
+                                302
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        289,
+                                        505,
+                                        302
+                                    ],
+                                    "score": 1.0,
+                                    "content": "butions by a greater margin. Though it performs better, the ERK distribution requires approximately",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                300,
+                                506,
+                                313
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        300,
+                                        506,
+                                        313
+                                    ],
+                                    "score": 1.0,
+                                    "content": "twice as many FLOPs compared to a uniform distribution. This highlights an interesting trade-off",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                310,
+                                506,
+                                323
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        310,
+                                        506,
+                                        323
+                                    ],
+                                    "score": 1.0,
+                                    "content": "between accuracy and computational efficiency even though both models have the same number of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                324,
+                                155,
+                                333
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        324,
+                                        155,
+                                        333
+                                    ],
+                                    "score": 1.0,
+                                    "content": "parameters.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        }
+                    ],
+                    "index": 7.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        338,
+                        505,
+                        393
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                338,
+                                505,
+                                352
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        338,
+                                        505,
+                                        352
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Effect of Update Schedule and Frequency: In Figure 5-right, we evaluate the performance of our",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                349,
+                                505,
+                                362
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        349,
+                                        217,
+                                        362
+                                    ],
+                                    "score": 1.0,
+                                    "content": "method on update intervals",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        218,
+                                        350,
+                                        325,
+                                        362
+                                    ],
+                                    "score": 0.83,
+                                    "content": "\\Delta T \\in [ 5 0 , 1 0 0 , 5 0 0 , 1 0 0 0 ]",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        325,
+                                        349,
+                                        428,
+                                        362
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and initial drop fractions",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        429,
+                                        350,
+                                        498,
+                                        362
+                                    ],
+                                    "score": 0.77,
+                                    "content": "\\alpha \\in [ 0 . 1 , 0 . 3 , 0 . 5",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        499,
+                                        349,
+                                        505,
+                                        362
+                                    ],
+                                    "score": 1.0,
+                                    "content": "].",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                358,
+                                506,
+                                375
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        358,
+                                        506,
+                                        375
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The best accuracies are obtained when the mask is updated every 100 iterations with an initial drop",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                370,
+                                506,
+                                385
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        370,
+                                        506,
+                                        385
+                                    ],
+                                    "score": 1.0,
+                                    "content": "fraction of 0.3 or 0.5. Notably, even with frequent update intervals (e.g. every 1000 iterations), RigL",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                382,
+                                202,
+                                393
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        382,
+                                        171,
+                                        393
+                                    ],
+                                    "score": 1.0,
+                                    "content": "performs above",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        171,
+                                        383,
+                                        198,
+                                        393
+                                    ],
+                                    "score": 0.87,
+                                    "content": "7 3 . 5 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        198,
+                                        382,
+                                        202,
+                                        393
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        }
+                    ],
+                    "index": 13
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        399,
+                        505,
+                        510
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                399,
+                                506,
+                                412
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        399,
+                                        506,
+                                        412
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Effect of Dynamic connections: Frankle et al. (2019) and Mostafa & Wang (2019) observed that",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                410,
+                                505,
+                                424
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        410,
+                                        505,
+                                        424
+                                    ],
+                                    "score": 1.0,
+                                    "content": "static sparse training converges to a solution with a higher loss than dynamic sparse training. In",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                419,
+                                506,
+                                435
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        419,
+                                        506,
+                                        435
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Figure 6-left we examine the loss landscape lying between a solution found via static sparse training",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                432,
+                                506,
+                                445
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        432,
+                                        506,
+                                        445
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and a solution found via pruning to understand whether former lies in a basin isolated from the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                443,
+                                505,
+                                456
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        443,
+                                        505,
+                                        456
+                                    ],
+                                    "score": 1.0,
+                                    "content": "latter. Performing a linear interpolation between the two reveals the expected result – high-loss",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                455,
+                                505,
+                                468
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        455,
+                                        505,
+                                        468
+                                    ],
+                                    "score": 1.0,
+                                    "content": "barrier – demonstrating that the loss landscape is not trivially connected. However, this is only",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                465,
+                                505,
+                                477
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        465,
+                                        505,
+                                        477
+                                    ],
+                                    "score": 1.0,
+                                    "content": "one of infinitely many paths between the two points optimization can be used to find parametric",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                476,
+                                506,
+                                489
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        476,
+                                        506,
+                                        489
+                                    ],
+                                    "score": 1.0,
+                                    "content": "curves that connects solutions (Garipov et al., 2018; Draxler et al., 2018) subject to constraints. For",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                487,
+                                506,
+                                500
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        487,
+                                        506,
+                                        500
+                                    ],
+                                    "score": 1.0,
+                                    "content": "example Garipov et al. (2018) showed different dense solutions lie in the same basin by finding 2nd",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                497,
+                                505,
+                                511
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        497,
+                                        505,
+                                        511
+                                    ],
+                                    "score": 1.0,
+                                    "content": "order Bezier curves with low energy between the two solutions. Following their method, we attempt ´",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        }
+                    ],
+                    "index": 20.5
+                },
+                {
+                    "type": "image",
+                    "bbox": [
+                        110,
+                        551,
+                        500,
+                        699
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                110,
+                                551,
+                                500,
+                                699
+                            ],
+                            "group_id": 1,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        110,
+                                        551,
+                                        500,
+                                        699
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                110,
+                                                551,
+                                                500,
+                                                699
+                                            ],
+                                            "score": 0.966,
+                                            "type": "image",
+                                            "image_path": "96e95693d928d87ad902426e986b917f4232a0de25740f3cd1cb80a5c1a6297c.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 27,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        110,
+                                        551,
+                                        500,
+                                        600.3333333333334
+                                    ],
+                                    "spans": [],
+                                    "index": 26
+                                },
+                                {
+                                    "bbox": [
+                                        110,
+                                        600.3333333333334,
+                                        500,
+                                        649.6666666666667
+                                    ],
+                                    "spans": [],
+                                    "index": 27
+                                },
+                                {
+                                    "bbox": [
+                                        110,
+                                        649.6666666666667,
+                                        500,
+                                        699.0000000000001
+                                    ],
+                                    "spans": [],
+                                    "index": 28
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                107,
+                                709,
+                                505,
+                                743
+                            ],
+                            "group_id": 1,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        710,
+                                        505,
+                                        722
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                710,
+                                                505,
+                                                722
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 6: (left) Training loss evaluated at various points on interpolation curves between a magnitude",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 29
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        721,
+                                        506,
+                                        733
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                721,
+                                                506,
+                                                733
+                                            ],
+                                            "score": 1.0,
+                                            "content": "pruning model (0.0) and a model trained with static sparsity (1.0). (right) Training loss of RigL and",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 30
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        732,
+                                        429,
+                                        744
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                732,
+                                                429,
+                                                744
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Static methods starting from the static sparse solution, and their final accuracies.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 31
+                                }
+                            ],
+                            "index": 30
+                        }
+                    ],
+                    "index": 28.5
+                }
+            ],
+            "page_idx": 8,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        115,
+                        519,
+                        481,
+                        531
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                120,
+                                518,
+                                482,
+                                532
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        120,
+                                        518,
+                                        355,
+                                        532
+                                    ],
+                                    "score": 1.0,
+                                    "content": "†see Appendix I for exact layer-wise sparsities given by ERK for",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        356,
+                                        520,
+                                        373,
+                                        529
+                                    ],
+                                    "score": 0.86,
+                                    "content": "80 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        374,
+                                        518,
+                                        389,
+                                        532
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        389,
+                                        520,
+                                        407,
+                                        529
+                                    ],
+                                    "score": 0.86,
+                                    "content": "90 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        407,
+                                        518,
+                                        482,
+                                        532
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sparse ResNet-50’s.",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        107,
+                        27,
+                        308,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                25,
+                                309,
+                                39
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        25,
+                                        309,
+                                        39
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Under review as a conference paper at ICLR 2020",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        302,
+                        751,
+                        309,
+                        759
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                302,
+                                750,
+                                309,
+                                761
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        302,
+                                        750,
+                                        309,
+                                        761
+                                    ],
+                                    "score": 1.0,
+                                    "content": "9",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "image",
+                    "bbox": [
+                        113,
+                        81,
+                        493,
+                        226
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                113,
+                                81,
+                                493,
+                                226
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        113,
+                                        81,
+                                        493,
+                                        226
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                113,
+                                                81,
+                                                493,
+                                                226
+                                            ],
+                                            "score": 0.967,
+                                            "type": "image",
+                                            "image_path": "5800160114d0737b3e2907dc82a457c52f3b8ccbcf1ecf67ee038a009c690904.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 1,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        113,
+                                        81,
+                                        493,
+                                        129.33333333333334
+                                    ],
+                                    "spans": [],
+                                    "index": 0
+                                },
+                                {
+                                    "bbox": [
+                                        113,
+                                        129.33333333333334,
+                                        493,
+                                        177.66666666666669
+                                    ],
+                                    "spans": [],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        113,
+                                        177.66666666666669,
+                                        493,
+                                        226.00000000000003
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                106,
+                                236,
+                                504,
+                                259
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        234,
+                                        505,
+                                        250
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                234,
+                                                505,
+                                                250
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 5: (left) Performance of RigL at different sparsities using different sparsity masks (right)",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 3
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        246,
+                                        391,
+                                        259
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                246,
+                                                391,
+                                                259
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Ablation study on cosine schedule. Other methods are in the appendix.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 4
+                                }
+                            ],
+                            "index": 3.5
+                        }
+                    ],
+                    "index": 2.25
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        266,
+                        505,
+                        333
+                    ],
+                    "lines": [],
+                    "index": 7.5,
+                    "bbox_fs": [
+                        105,
+                        266,
+                        506,
+                        333
+                    ],
+                    "lines_deleted": true
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        338,
+                        505,
+                        393
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                338,
+                                505,
+                                352
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        338,
+                                        505,
+                                        352
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Effect of Update Schedule and Frequency: In Figure 5-right, we evaluate the performance of our",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                349,
+                                505,
+                                362
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        349,
+                                        217,
+                                        362
+                                    ],
+                                    "score": 1.0,
+                                    "content": "method on update intervals",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        218,
+                                        350,
+                                        325,
+                                        362
+                                    ],
+                                    "score": 0.83,
+                                    "content": "\\Delta T \\in [ 5 0 , 1 0 0 , 5 0 0 , 1 0 0 0 ]",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        325,
+                                        349,
+                                        428,
+                                        362
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and initial drop fractions",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        429,
+                                        350,
+                                        498,
+                                        362
+                                    ],
+                                    "score": 0.77,
+                                    "content": "\\alpha \\in [ 0 . 1 , 0 . 3 , 0 . 5",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        499,
+                                        349,
+                                        505,
+                                        362
+                                    ],
+                                    "score": 1.0,
+                                    "content": "].",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                358,
+                                506,
+                                375
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        358,
+                                        506,
+                                        375
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The best accuracies are obtained when the mask is updated every 100 iterations with an initial drop",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                370,
+                                506,
+                                385
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        370,
+                                        506,
+                                        385
+                                    ],
+                                    "score": 1.0,
+                                    "content": "fraction of 0.3 or 0.5. Notably, even with frequent update intervals (e.g. every 1000 iterations), RigL",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                382,
+                                202,
+                                393
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        382,
+                                        171,
+                                        393
+                                    ],
+                                    "score": 1.0,
+                                    "content": "performs above",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        171,
+                                        383,
+                                        198,
+                                        393
+                                    ],
+                                    "score": 0.87,
+                                    "content": "7 3 . 5 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        198,
+                                        382,
+                                        202,
+                                        393
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        }
+                    ],
+                    "index": 13,
+                    "bbox_fs": [
+                        105,
+                        338,
+                        506,
+                        393
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        399,
+                        505,
+                        510
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                399,
+                                506,
+                                412
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        399,
+                                        506,
+                                        412
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Effect of Dynamic connections: Frankle et al. (2019) and Mostafa & Wang (2019) observed that",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                410,
+                                505,
+                                424
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        410,
+                                        505,
+                                        424
+                                    ],
+                                    "score": 1.0,
+                                    "content": "static sparse training converges to a solution with a higher loss than dynamic sparse training. In",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                419,
+                                506,
+                                435
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        419,
+                                        506,
+                                        435
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Figure 6-left we examine the loss landscape lying between a solution found via static sparse training",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                432,
+                                506,
+                                445
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        432,
+                                        506,
+                                        445
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and a solution found via pruning to understand whether former lies in a basin isolated from the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                443,
+                                505,
+                                456
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        443,
+                                        505,
+                                        456
+                                    ],
+                                    "score": 1.0,
+                                    "content": "latter. Performing a linear interpolation between the two reveals the expected result – high-loss",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                455,
+                                505,
+                                468
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        455,
+                                        505,
+                                        468
+                                    ],
+                                    "score": 1.0,
+                                    "content": "barrier – demonstrating that the loss landscape is not trivially connected. However, this is only",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                465,
+                                505,
+                                477
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        465,
+                                        505,
+                                        477
+                                    ],
+                                    "score": 1.0,
+                                    "content": "one of infinitely many paths between the two points optimization can be used to find parametric",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                476,
+                                506,
+                                489
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        476,
+                                        506,
+                                        489
+                                    ],
+                                    "score": 1.0,
+                                    "content": "curves that connects solutions (Garipov et al., 2018; Draxler et al., 2018) subject to constraints. For",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                487,
+                                506,
+                                500
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        487,
+                                        506,
+                                        500
+                                    ],
+                                    "score": 1.0,
+                                    "content": "example Garipov et al. (2018) showed different dense solutions lie in the same basin by finding 2nd",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                497,
+                                505,
+                                511
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        497,
+                                        505,
+                                        511
+                                    ],
+                                    "score": 1.0,
+                                    "content": "order Bezier curves with low energy between the two solutions. Following their method, we attempt ´",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                83,
+                                505,
+                                94
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        83,
+                                        505,
+                                        94
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to find quadratic and cubic Bezier curves between the two sparse solutions. Surprisingly, even with a ´",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                94,
+                                505,
+                                106
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        94,
+                                        505,
+                                        106
+                                    ],
+                                    "score": 1.0,
+                                    "content": "cubic curve, we fail to find a path without a high-loss barrier. These results suggest that static sparse",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                104,
+                                505,
+                                117
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        104,
+                                        505,
+                                        117
+                                    ],
+                                    "score": 1.0,
+                                    "content": "training can get stuck at local minima that are isolated from improved solutions. On the other hand,",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                116,
+                                505,
+                                127
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        116,
+                                        505,
+                                        127
+                                    ],
+                                    "score": 1.0,
+                                    "content": "when we optimize the quadratic Bezier curve across the full ´ dense space we find a near-monotonic",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                127,
+                                505,
+                                138
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        127,
+                                        505,
+                                        138
+                                    ],
+                                    "score": 1.0,
+                                    "content": "path to the improved solution, suggesting that allowing new connections to grow lends dynamic",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                137,
+                                370,
+                                151
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        137,
+                                        370,
+                                        151
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sparse training greater flexibility in navigating the loss landscape.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 5
+                        }
+                    ],
+                    "index": 20.5,
+                    "bbox_fs": [
+                        104,
+                        399,
+                        506,
+                        511
+                    ]
+                },
+                {
+                    "type": "image",
+                    "bbox": [
+                        110,
+                        551,
+                        500,
+                        699
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                110,
+                                551,
+                                500,
+                                699
+                            ],
+                            "group_id": 1,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        110,
+                                        551,
+                                        500,
+                                        699
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                110,
+                                                551,
+                                                500,
+                                                699
+                                            ],
+                                            "score": 0.966,
+                                            "type": "image",
+                                            "image_path": "96e95693d928d87ad902426e986b917f4232a0de25740f3cd1cb80a5c1a6297c.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 27,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        110,
+                                        551,
+                                        500,
+                                        600.3333333333334
+                                    ],
+                                    "spans": [],
+                                    "index": 26
+                                },
+                                {
+                                    "bbox": [
+                                        110,
+                                        600.3333333333334,
+                                        500,
+                                        649.6666666666667
+                                    ],
+                                    "spans": [],
+                                    "index": 27
+                                },
+                                {
+                                    "bbox": [
+                                        110,
+                                        649.6666666666667,
+                                        500,
+                                        699.0000000000001
+                                    ],
+                                    "spans": [],
+                                    "index": 28
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                107,
+                                709,
+                                505,
+                                743
+                            ],
+                            "group_id": 1,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        710,
+                                        505,
+                                        722
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                710,
+                                                505,
+                                                722
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 6: (left) Training loss evaluated at various points on interpolation curves between a magnitude",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 29
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        721,
+                                        506,
+                                        733
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                721,
+                                                506,
+                                                733
+                                            ],
+                                            "score": 1.0,
+                                            "content": "pruning model (0.0) and a model trained with static sparsity (1.0). (right) Training loss of RigL and",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 30
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        732,
+                                        429,
+                                        744
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                732,
+                                                429,
+                                                744
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Static methods starting from the static sparse solution, and their final accuracies.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 31
+                                }
+                            ],
+                            "index": 30
+                        }
+                    ],
+                    "index": 28.5
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        82,
+                        505,
+                        149
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                83,
+                                505,
+                                94
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        83,
+                                        505,
+                                        94
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to find quadratic and cubic Bezier curves between the two sparse solutions. Surprisingly, even with a ´",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                94,
+                                505,
+                                106
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        94,
+                                        505,
+                                        106
+                                    ],
+                                    "score": 1.0,
+                                    "content": "cubic curve, we fail to find a path without a high-loss barrier. These results suggest that static sparse",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                104,
+                                505,
+                                117
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        104,
+                                        505,
+                                        117
+                                    ],
+                                    "score": 1.0,
+                                    "content": "training can get stuck at local minima that are isolated from improved solutions. On the other hand,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                116,
+                                505,
+                                127
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        116,
+                                        505,
+                                        127
+                                    ],
+                                    "score": 1.0,
+                                    "content": "when we optimize the quadratic Bezier curve across the full ´ dense space we find a near-monotonic",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                127,
+                                505,
+                                138
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        127,
+                                        505,
+                                        138
+                                    ],
+                                    "score": 1.0,
+                                    "content": "path to the improved solution, suggesting that allowing new connections to grow lends dynamic",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                137,
+                                370,
+                                151
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        137,
+                                        370,
+                                        151
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sparse training greater flexibility in navigating the loss landscape.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        }
+                    ],
+                    "index": 2.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        154,
+                        505,
+                        231
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                154,
+                                505,
+                                167
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        154,
+                                        505,
+                                        167
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In Figure 6-right we train RigL starting from the sub-optimal solution found by static sparse train-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                165,
+                                505,
+                                178
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        165,
+                                        505,
+                                        178
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ing, demonstrating that it is able to escape the local minimum, whereas re-training with static sparse",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                176,
+                                505,
+                                189
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        176,
+                                        505,
+                                        189
+                                    ],
+                                    "score": 1.0,
+                                    "content": "training cannot. RigL first removes connections with the smallest magnitudes since removing these",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                187,
+                                505,
+                                200
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        187,
+                                        505,
+                                        200
+                                    ],
+                                    "score": 1.0,
+                                    "content": "connections have been shown to have a minimal effect on the loss (Han et al., 2015; Evci, 2018).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                198,
+                                505,
+                                210
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        198,
+                                        505,
+                                        210
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Next, it activates connections with the high gradients, since these connections are expected to de-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                208,
+                                505,
+                                223
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        208,
+                                        505,
+                                        223
+                                    ],
+                                    "score": 1.0,
+                                    "content": "crease the loss fastest. We hypothesize in Appendix A that RigL escapes bad critical points by",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                221,
+                                342,
+                                233
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        221,
+                                        342,
+                                        233
+                                    ],
+                                    "score": 1.0,
+                                    "content": "replacing saddle directions with high gradient dimensions.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        }
+                    ],
+                    "index": 9
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        248,
+                        274,
+                        260
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                245,
+                                277,
+                                263
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        245,
+                                        277,
+                                        263
+                                    ],
+                                    "score": 1.0,
+                                    "content": "5 DISCUSSION & CONCLUSION",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        }
+                    ],
+                    "index": 13
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        273,
+                        505,
+                        350
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                272,
+                                506,
+                                286
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        272,
+                                        506,
+                                        286
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In this work we introduced ‘Rigged Lottery’ or RigL, an algorithm for training sparse neural net-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                284,
+                                505,
+                                297
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        284,
+                                        505,
+                                        297
+                                    ],
+                                    "score": 1.0,
+                                    "content": "works efficiently. For a given computational budget RigL achieves higher accuracies than existing",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                295,
+                                505,
+                                308
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        295,
+                                        505,
+                                        308
+                                    ],
+                                    "score": 1.0,
+                                    "content": "dense-to-sparse and sparse-to-sparse training algorithms. RigL is useful in three different scenarios:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                305,
+                                505,
+                                320
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        305,
+                                        505,
+                                        320
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(1) To improve the accuracy of sparse models intended for deployment; (2) To improve the accuracy",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                316,
+                                505,
+                                330
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        316,
+                                        505,
+                                        330
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of large sparse models which can only be trained for a limited number of iterations; and (3) Com-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                328,
+                                506,
+                                342
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        328,
+                                        506,
+                                        342
+                                    ],
+                                    "score": 1.0,
+                                    "content": "bined with sparse primitives to enable training of extremely large sparse models which otherwise",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                339,
+                                199,
+                                352
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        339,
+                                        199,
+                                        352
+                                    ],
+                                    "score": 1.0,
+                                    "content": "would not be possible.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        }
+                    ],
+                    "index": 17
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        356,
+                        505,
+                        422
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                354,
+                                505,
+                                368
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        354,
+                                        505,
+                                        368
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The third scenario is unexplored due to the lack of hardware and software support for sparsity.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                366,
+                                505,
+                                379
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        366,
+                                        505,
+                                        379
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Nonetheless, work continues to improve the performance of sparse networks on current hard-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                378,
+                                505,
+                                390
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        378,
+                                        505,
+                                        390
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ware (Hong et al., 2019; Merrill & Garland, 2016), and new types of hardware accelerators will",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                388,
+                                506,
+                                402
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        388,
+                                        506,
+                                        402
+                                    ],
+                                    "score": 1.0,
+                                    "content": "have better support for parameter sparsity (Wang et al., 2018; Mike Ashby, 2019; Liu et al., 2018;",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                399,
+                                505,
+                                412
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        399,
+                                        505,
+                                        412
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Han et al., 2016a; Chen et al., 2019). RigL provides the tools to take advantage of, and motivation",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                410,
+                                184,
+                                423
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        410,
+                                        184,
+                                        423
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for, such advances.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        }
+                    ],
+                    "index": 23.5
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        439,
+                        175,
+                        451
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                439,
+                                176,
+                                452
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        439,
+                                        176,
+                                        452
+                                    ],
+                                    "score": 1.0,
+                                    "content": "REFERENCES",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        }
+                    ],
+                    "index": 27
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        458,
+                        505,
+                        546
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                457,
+                                505,
+                                471
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        457,
+                                        505,
+                                        471
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Mart´ın Abadi, Ashish Agarwal, Paul Barham, Eugene Brevdo, Zhifeng Chen, Craig Citro, Greg S.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                469,
+                                505,
+                                481
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        469,
+                                        505,
+                                        481
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Corrado, Andy Davis, Jeffrey Dean, Matthieu Devin, Sanjay Ghemawat, Ian Goodfellow, Andrew",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                479,
+                                505,
+                                493
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        479,
+                                        505,
+                                        493
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Harp, Geoffrey Irving, Michael Isard, Yangqing Jia, Rafal Jozefowicz, Lukasz Kaiser, Manjunath",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                490,
+                                505,
+                                504
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        490,
+                                        505,
+                                        504
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Kudlur, Josh Levenberg, Dan Mane, Rajat Monga, Sherry Moore, Derek Murray, Chris Olah, ´",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                502,
+                                505,
+                                514
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        502,
+                                        505,
+                                        514
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Mike Schuster, Jonathon Shlens, Benoit Steiner, Ilya Sutskever, Kunal Talwar, Paul Tucker, Vin-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                513,
+                                505,
+                                525
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        513,
+                                        505,
+                                        525
+                                    ],
+                                    "score": 1.0,
+                                    "content": "cent Vanhoucke, Vijay Vasudevan, Fernanda Viegas, Oriol Vinyals, Pete Warden, Martin Watten- ´",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                523,
+                                505,
+                                537
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        523,
+                                        505,
+                                        537
+                                    ],
+                                    "score": 1.0,
+                                    "content": "berg, Martin Wicke, Yuan Yu, and Xiaoqiang Zheng. TensorFlow: Large-scale machine learning",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                535,
+                                410,
+                                547
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        535,
+                                        410,
+                                        547
+                                    ],
+                                    "score": 1.0,
+                                    "content": "on heterogeneous systems, 2015. URL http://tensorflow.org/.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        }
+                    ],
+                    "index": 31.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        554,
+                        505,
+                        587
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                552,
+                                505,
+                                568
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        552,
+                                        505,
+                                        568
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Guillaume Bellec, David Kappel, Wolfgang Maass, and Robert A. Legenstein. Deep rewiring:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                565,
+                                505,
+                                578
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        565,
+                                        505,
+                                        578
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Training very sparse deep networks. CoRR, abs/1711.05136, 2017. URL http://arxiv.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                576,
+                                230,
+                                587
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        576,
+                                        230,
+                                        587
+                                    ],
+                                    "score": 1.0,
+                                    "content": "org/abs/1711.05136.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        }
+                    ],
+                    "index": 37
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        595,
+                        504,
+                        629
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                595,
+                                505,
+                                609
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        595,
+                                        505,
+                                        609
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Y. Chen, T. Yang, J. Emer, and V. Sze. Eyeriss v2: A flexible accelerator for emerging deep neural",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                607,
+                                505,
+                                619
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        607,
+                                        505,
+                                        619
+                                    ],
+                                    "score": 1.0,
+                                    "content": "networks on mobile devices. IEEE Journal on Emerging and Selected Topics in Circuits and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                618,
+                                413,
+                                629
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        618,
+                                        413,
+                                        629
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Systems, 9(2):292–308, June 2019. doi: 10.1109/JETCAS.2019.2910232.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        }
+                    ],
+                    "index": 40
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        637,
+                        503,
+                        671
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                637,
+                                505,
+                                650
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        637,
+                                        505,
+                                        650
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Kyunghyun Cho, B van Merrienboer, Caglar Gulcehre, F Bougares, H Schwenk, and Yoshua Ben-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                648,
+                                505,
+                                661
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        648,
+                                        505,
+                                        661
+                                    ],
+                                    "score": 1.0,
+                                    "content": "gio. Learning phrase representations using rnn encoder-decoder for statistical machine translation.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                659,
+                                490,
+                                672
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        659,
+                                        490,
+                                        672
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In Conference on Empirical Methods in Natural Language Processing (EMNLP 2014), 2014.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        }
+                    ],
+                    "index": 43
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        678,
+                        501,
+                        702
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                678,
+                                504,
+                                692
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        678,
+                                        495,
+                                        692
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Diederik P. Kingma Christos Louizos, Max Welling. Learning sparse neural networks through",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        495,
+                                        680,
+                                        504,
+                                        690
+                                    ],
+                                    "score": 0.81,
+                                    "content": "l _ { 0 }",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                690,
+                                437,
+                                702
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        690,
+                                        437,
+                                        702
+                                    ],
+                                    "score": 1.0,
+                                    "content": "regularization. In International Conference on Learning Representations, 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        }
+                    ],
+                    "index": 45.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        709,
+                        503,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                707,
+                                505,
+                                723
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        707,
+                                        505,
+                                        723
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Tim Dettmers and Luke Zettlemoyer. Sparse networks from scratch: Faster training without losing",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 47
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                720,
+                                441,
+                                732
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        720,
+                                        441,
+                                        732
+                                    ],
+                                    "score": 1.0,
+                                    "content": "performance. ArXiv, 2019. URL http://arxiv.org/abs/1907.04840.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 48
+                        }
+                    ],
+                    "index": 47.5
+                }
+            ],
+            "page_idx": 9,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        107,
+                        27,
+                        308,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                107,
+                                26,
+                                308,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        26,
+                                        308,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Under review as a conference paper at ICLR 2020",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        301,
+                        752,
+                        311,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                299,
+                                750,
+                                313,
+                                764
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        299,
+                                        750,
+                                        313,
+                                        764
+                                    ],
+                                    "score": 1.0,
+                                    "content": "",
+                                    "type": "text",
+                                    "height": 14,
+                                    "width": 14
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        82,
+                        505,
+                        149
+                    ],
+                    "lines": [],
+                    "index": 2.5,
+                    "bbox_fs": [
+                        105,
+                        83,
+                        505,
+                        151
+                    ],
+                    "lines_deleted": true
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        154,
+                        505,
+                        231
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                154,
+                                505,
+                                167
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        154,
+                                        505,
+                                        167
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In Figure 6-right we train RigL starting from the sub-optimal solution found by static sparse train-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                165,
+                                505,
+                                178
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        165,
+                                        505,
+                                        178
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ing, demonstrating that it is able to escape the local minimum, whereas re-training with static sparse",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                176,
+                                505,
+                                189
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        176,
+                                        505,
+                                        189
+                                    ],
+                                    "score": 1.0,
+                                    "content": "training cannot. RigL first removes connections with the smallest magnitudes since removing these",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                187,
+                                505,
+                                200
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        187,
+                                        505,
+                                        200
+                                    ],
+                                    "score": 1.0,
+                                    "content": "connections have been shown to have a minimal effect on the loss (Han et al., 2015; Evci, 2018).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                198,
+                                505,
+                                210
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        198,
+                                        505,
+                                        210
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Next, it activates connections with the high gradients, since these connections are expected to de-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                208,
+                                505,
+                                223
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        208,
+                                        505,
+                                        223
+                                    ],
+                                    "score": 1.0,
+                                    "content": "crease the loss fastest. We hypothesize in Appendix A that RigL escapes bad critical points by",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                221,
+                                342,
+                                233
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        221,
+                                        342,
+                                        233
+                                    ],
+                                    "score": 1.0,
+                                    "content": "replacing saddle directions with high gradient dimensions.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        }
+                    ],
+                    "index": 9,
+                    "bbox_fs": [
+                        105,
+                        154,
+                        505,
+                        233
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        248,
+                        274,
+                        260
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                245,
+                                277,
+                                263
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        245,
+                                        277,
+                                        263
+                                    ],
+                                    "score": 1.0,
+                                    "content": "5 DISCUSSION & CONCLUSION",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        }
+                    ],
+                    "index": 13
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        273,
+                        505,
+                        350
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                272,
+                                506,
+                                286
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        272,
+                                        506,
+                                        286
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In this work we introduced ‘Rigged Lottery’ or RigL, an algorithm for training sparse neural net-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                284,
+                                505,
+                                297
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        284,
+                                        505,
+                                        297
+                                    ],
+                                    "score": 1.0,
+                                    "content": "works efficiently. For a given computational budget RigL achieves higher accuracies than existing",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                295,
+                                505,
+                                308
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        295,
+                                        505,
+                                        308
+                                    ],
+                                    "score": 1.0,
+                                    "content": "dense-to-sparse and sparse-to-sparse training algorithms. RigL is useful in three different scenarios:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                305,
+                                505,
+                                320
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        305,
+                                        505,
+                                        320
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(1) To improve the accuracy of sparse models intended for deployment; (2) To improve the accuracy",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                316,
+                                505,
+                                330
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        316,
+                                        505,
+                                        330
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of large sparse models which can only be trained for a limited number of iterations; and (3) Com-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                328,
+                                506,
+                                342
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        328,
+                                        506,
+                                        342
+                                    ],
+                                    "score": 1.0,
+                                    "content": "bined with sparse primitives to enable training of extremely large sparse models which otherwise",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                339,
+                                199,
+                                352
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        339,
+                                        199,
+                                        352
+                                    ],
+                                    "score": 1.0,
+                                    "content": "would not be possible.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        }
+                    ],
+                    "index": 17,
+                    "bbox_fs": [
+                        105,
+                        272,
+                        506,
+                        352
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        356,
+                        505,
+                        422
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                354,
+                                505,
+                                368
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        354,
+                                        505,
+                                        368
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The third scenario is unexplored due to the lack of hardware and software support for sparsity.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                366,
+                                505,
+                                379
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        366,
+                                        505,
+                                        379
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Nonetheless, work continues to improve the performance of sparse networks on current hard-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                378,
+                                505,
+                                390
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        378,
+                                        505,
+                                        390
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ware (Hong et al., 2019; Merrill & Garland, 2016), and new types of hardware accelerators will",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                388,
+                                506,
+                                402
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        388,
+                                        506,
+                                        402
+                                    ],
+                                    "score": 1.0,
+                                    "content": "have better support for parameter sparsity (Wang et al., 2018; Mike Ashby, 2019; Liu et al., 2018;",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                399,
+                                505,
+                                412
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        399,
+                                        505,
+                                        412
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Han et al., 2016a; Chen et al., 2019). RigL provides the tools to take advantage of, and motivation",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                410,
+                                184,
+                                423
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        410,
+                                        184,
+                                        423
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for, such advances.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        }
+                    ],
+                    "index": 23.5,
+                    "bbox_fs": [
+                        105,
+                        354,
+                        506,
+                        423
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        439,
+                        175,
+                        451
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                439,
+                                176,
+                                452
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        439,
+                                        176,
+                                        452
+                                    ],
+                                    "score": 1.0,
+                                    "content": "REFERENCES",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        }
+                    ],
+                    "index": 27
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        458,
+                        505,
+                        546
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                457,
+                                505,
+                                471
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        457,
+                                        505,
+                                        471
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Mart´ın Abadi, Ashish Agarwal, Paul Barham, Eugene Brevdo, Zhifeng Chen, Craig Citro, Greg S.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                469,
+                                505,
+                                481
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        469,
+                                        505,
+                                        481
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Corrado, Andy Davis, Jeffrey Dean, Matthieu Devin, Sanjay Ghemawat, Ian Goodfellow, Andrew",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                479,
+                                505,
+                                493
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        479,
+                                        505,
+                                        493
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Harp, Geoffrey Irving, Michael Isard, Yangqing Jia, Rafal Jozefowicz, Lukasz Kaiser, Manjunath",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                490,
+                                505,
+                                504
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        490,
+                                        505,
+                                        504
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Kudlur, Josh Levenberg, Dan Mane, Rajat Monga, Sherry Moore, Derek Murray, Chris Olah, ´",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                502,
+                                505,
+                                514
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        502,
+                                        505,
+                                        514
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Mike Schuster, Jonathon Shlens, Benoit Steiner, Ilya Sutskever, Kunal Talwar, Paul Tucker, Vin-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                513,
+                                505,
+                                525
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        513,
+                                        505,
+                                        525
+                                    ],
+                                    "score": 1.0,
+                                    "content": "cent Vanhoucke, Vijay Vasudevan, Fernanda Viegas, Oriol Vinyals, Pete Warden, Martin Watten- ´",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                523,
+                                505,
+                                537
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        523,
+                                        505,
+                                        537
+                                    ],
+                                    "score": 1.0,
+                                    "content": "berg, Martin Wicke, Yuan Yu, and Xiaoqiang Zheng. TensorFlow: Large-scale machine learning",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                535,
+                                410,
+                                547
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        535,
+                                        410,
+                                        547
+                                    ],
+                                    "score": 1.0,
+                                    "content": "on heterogeneous systems, 2015. URL http://tensorflow.org/.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        }
+                    ],
+                    "index": 31.5,
+                    "bbox_fs": [
+                        106,
+                        457,
+                        505,
+                        547
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        554,
+                        505,
+                        587
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                552,
+                                505,
+                                568
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        552,
+                                        505,
+                                        568
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Guillaume Bellec, David Kappel, Wolfgang Maass, and Robert A. Legenstein. Deep rewiring:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                565,
+                                505,
+                                578
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        565,
+                                        505,
+                                        578
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Training very sparse deep networks. CoRR, abs/1711.05136, 2017. URL http://arxiv.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                576,
+                                230,
+                                587
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        576,
+                                        230,
+                                        587
+                                    ],
+                                    "score": 1.0,
+                                    "content": "org/abs/1711.05136.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        }
+                    ],
+                    "index": 37,
+                    "bbox_fs": [
+                        106,
+                        552,
+                        505,
+                        587
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        595,
+                        504,
+                        629
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                595,
+                                505,
+                                609
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        595,
+                                        505,
+                                        609
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Y. Chen, T. Yang, J. Emer, and V. Sze. Eyeriss v2: A flexible accelerator for emerging deep neural",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                607,
+                                505,
+                                619
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        607,
+                                        505,
+                                        619
+                                    ],
+                                    "score": 1.0,
+                                    "content": "networks on mobile devices. IEEE Journal on Emerging and Selected Topics in Circuits and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                618,
+                                413,
+                                629
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        618,
+                                        413,
+                                        629
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Systems, 9(2):292–308, June 2019. doi: 10.1109/JETCAS.2019.2910232.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        }
+                    ],
+                    "index": 40,
+                    "bbox_fs": [
+                        105,
+                        595,
+                        505,
+                        629
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        637,
+                        503,
+                        671
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                637,
+                                505,
+                                650
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        637,
+                                        505,
+                                        650
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Kyunghyun Cho, B van Merrienboer, Caglar Gulcehre, F Bougares, H Schwenk, and Yoshua Ben-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                648,
+                                505,
+                                661
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        648,
+                                        505,
+                                        661
+                                    ],
+                                    "score": 1.0,
+                                    "content": "gio. Learning phrase representations using rnn encoder-decoder for statistical machine translation.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                659,
+                                490,
+                                672
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        659,
+                                        490,
+                                        672
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In Conference on Empirical Methods in Natural Language Processing (EMNLP 2014), 2014.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        }
+                    ],
+                    "index": 43,
+                    "bbox_fs": [
+                        106,
+                        637,
+                        505,
+                        672
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        678,
+                        501,
+                        702
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                678,
+                                504,
+                                692
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        678,
+                                        495,
+                                        692
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Diederik P. Kingma Christos Louizos, Max Welling. Learning sparse neural networks through",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        495,
+                                        680,
+                                        504,
+                                        690
+                                    ],
+                                    "score": 0.81,
+                                    "content": "l _ { 0 }",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                690,
+                                437,
+                                702
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        690,
+                                        437,
+                                        702
+                                    ],
+                                    "score": 1.0,
+                                    "content": "regularization. In International Conference on Learning Representations, 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        }
+                    ],
+                    "index": 45.5,
+                    "bbox_fs": [
+                        105,
+                        678,
+                        504,
+                        702
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        709,
+                        503,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                707,
+                                505,
+                                723
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        707,
+                                        505,
+                                        723
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Tim Dettmers and Luke Zettlemoyer. Sparse networks from scratch: Faster training without losing",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 47
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                720,
+                                441,
+                                732
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        720,
+                                        441,
+                                        732
+                                    ],
+                                    "score": 1.0,
+                                    "content": "performance. ArXiv, 2019. URL http://arxiv.org/abs/1907.04840.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 48
+                        }
+                    ],
+                    "index": 47.5,
+                    "bbox_fs": [
+                        106,
+                        707,
+                        505,
+                        732
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        82,
+                        504,
+                        116
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                82,
+                                504,
+                                95
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        82,
+                                        504,
+                                        95
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Felix Draxler, Kambis Veschgini, Manfred Salmhofer, and Fred A Hamprecht. Essentially No Bar-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                114,
+                                92,
+                                505,
+                                108
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        114,
+                                        92,
+                                        505,
+                                        108
+                                    ],
+                                    "score": 1.0,
+                                    "content": "riers in Neural Network Energy Landscape. In International Conference on Machine Learning,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                104,
+                                144,
+                                117
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        104,
+                                        144,
+                                        117
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        }
+                    ],
+                    "index": 1
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        123,
+                        502,
+                        145
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                121,
+                                505,
+                                137
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        121,
+                                        505,
+                                        137
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Utku Evci. Detecting dead weights and units in neural networks. CoRR, 2018. URL http:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                117,
+                                135,
+                                277,
+                                146
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        117,
+                                        135,
+                                        277,
+                                        146
+                                    ],
+                                    "score": 1.0,
+                                    "content": "//arxiv.org/abs/1806.06068.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        }
+                    ],
+                    "index": 3.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        152,
+                        504,
+                        175
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                151,
+                                505,
+                                165
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        151,
+                                        505,
+                                        165
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Utku Evci, Fabian Pedregosa, Aidan N. Gomez, and Erich Elsen. The difficulty of training sparse",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                163,
+                                454,
+                                176
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        163,
+                                        454,
+                                        176
+                                    ],
+                                    "score": 1.0,
+                                    "content": "neural networks. ArXiv, 2019. URL http://arxiv.org/abs/1906.10732.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        }
+                    ],
+                    "index": 5.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        182,
+                        505,
+                        227
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                181,
+                                505,
+                                195
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        181,
+                                        505,
+                                        195
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Jonathan Frankle and Michael Carbin. The lottery ticket hypothesis: Finding sparse, trainable neu-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                194,
+                                506,
+                                205
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        194,
+                                        506,
+                                        205
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ral networks. In 7th International Conference on Learning Representations, ICLR 2019, New",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                204,
+                                505,
+                                217
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        204,
+                                        497,
+                                        217
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Orleans, LA, USA, May 6-9, 2019, 2019. URL https://openreview.net/forum?id",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        497,
+                                        205,
+                                        505,
+                                        214
+                                    ],
+                                    "score": 0.34,
+                                    "content": "=",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                216,
+                                182,
+                                227
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        216,
+                                        182,
+                                        227
+                                    ],
+                                    "score": 1.0,
+                                    "content": "rJl-b3RcF7.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        }
+                    ],
+                    "index": 8.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        233,
+                        504,
+                        257
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                232,
+                                505,
+                                247
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        232,
+                                        505,
+                                        247
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Jonathan Frankle, Gintare Karolina Dziugaite, Daniel M. Roy, and Michael Carbin. The lottery",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                244,
+                                488,
+                                258
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        244,
+                                        488,
+                                        258
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ticket hypothesis at scale. ArXiv, 2019. URL http://arxiv.org/abs/1903.01611.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        }
+                    ],
+                    "index": 11.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        263,
+                        503,
+                        286
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                262,
+                                505,
+                                276
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        262,
+                                        505,
+                                        276
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Trevor Gale, Erich Elsen, and Sara Hooker. The state of sparsity in deep neural networks. CoRR,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                273,
+                                424,
+                                287
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        273,
+                                        424,
+                                        287
+                                    ],
+                                    "score": 1.0,
+                                    "content": "abs/1902.09574, 2019. URL http://arxiv.org/abs/1902.09574.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        }
+                    ],
+                    "index": 13.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        293,
+                        505,
+                        326
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                292,
+                                504,
+                                305
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        292,
+                                        504,
+                                        305
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Timur Garipov, Pavel Izmailov, Dmitrii Podoprikhin, Dmitry P Vetrov, and Andrew Gordon Wilson.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                304,
+                                505,
+                                316
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        304,
+                                        505,
+                                        316
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Loss surfaces, mode connectivity, and fast ensembling of dnns. In Advances in Neural Information",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                315,
+                                225,
+                                327
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        315,
+                                        225,
+                                        327
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Processing Systems, 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        }
+                    ],
+                    "index": 16
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        333,
+                        504,
+                        356
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                331,
+                                505,
+                                347
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        331,
+                                        505,
+                                        347
+                                    ],
+                                    "score": 1.0,
+                                    "content": "J. S. Garofolo, L. F. Lamel, W. M. Fisher, J. G. Fiscus, D. S. Pallett, and N. L. Dahlgren. Darpa",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                344,
+                                370,
+                                357
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        344,
+                                        370,
+                                        357
+                                    ],
+                                    "score": 1.0,
+                                    "content": "timit acoustic phonetic continuous speech corpus cdrom, 1993.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        }
+                    ],
+                    "index": 18.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        363,
+                        503,
+                        386
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                107,
+                                363,
+                                505,
+                                376
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        363,
+                                        505,
+                                        376
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Yiwen Guo, Anbang Yao, and Yurong Chen. Dynamic network surgery for efficient DNNs. CoRR,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                374,
+                                424,
+                                387
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        374,
+                                        424,
+                                        387
+                                    ],
+                                    "score": 1.0,
+                                    "content": "abs/1608.04493, 2016. URL http://arxiv.org/abs/1608.04493.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        }
+                    ],
+                    "index": 20.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        392,
+                        504,
+                        416
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                392,
+                                505,
+                                405
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        392,
+                                        505,
+                                        405
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Song Han, Jeff Pool, John Tran, and William Dally. Learning both weights and connections for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                403,
+                                463,
+                                416
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        403,
+                                        463,
+                                        416
+                                    ],
+                                    "score": 1.0,
+                                    "content": "efficient neural network. In Advances in neural information processing systems, 2015.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        }
+                    ],
+                    "index": 22.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        109,
+                        421,
+                        504,
+                        456
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                421,
+                                505,
+                                435
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        421,
+                                        505,
+                                        435
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Song Han, Xingyu Liu, Huizi Mao, Jing Pu, Ardavan Pedram, Mark A. Horowitz, and William J.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                432,
+                                505,
+                                446
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        432,
+                                        505,
+                                        446
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Dally. EIE: Efficient Inference Engine on compressed deep neural network. In Proceedings of the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                443,
+                                379,
+                                457
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        443,
+                                        379,
+                                        457
+                                    ],
+                                    "score": 1.0,
+                                    "content": "43rd International Symposium on Computer Architecture, 2016a.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        }
+                    ],
+                    "index": 25
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        462,
+                        505,
+                        507
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                462,
+                                505,
+                                476
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        462,
+                                        505,
+                                        476
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Song Han, Huizi Mao, and William J. Dally. Deep compression: Compressing deep neural net-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                473,
+                                506,
+                                487
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        473,
+                                        506,
+                                        487
+                                    ],
+                                    "score": 1.0,
+                                    "content": "work with pruning, trained quantization and huffman coding. In 4th International Conference on",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                486,
+                                506,
+                                497
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        486,
+                                        506,
+                                        497
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Learning Representations, ICLR 2016, San Juan, Puerto Rico, May 2-4, 2016, Conference Track",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                496,
+                                415,
+                                508
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        496,
+                                        415,
+                                        508
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Proceedings, 2016b. URL http://arxiv.org/abs/1510.00149.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        }
+                    ],
+                    "index": 28.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        514,
+                        504,
+                        536
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                512,
+                                504,
+                                528
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        512,
+                                        504,
+                                        528
+                                    ],
+                                    "score": 1.0,
+                                    "content": "B. Hassibi and D. Stork. Second order derivatives for network pruning: Optimal Brain Surgeon.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                524,
+                                143,
+                                538
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        524,
+                                        143,
+                                        538
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1993.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        }
+                    ],
+                    "index": 31.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        543,
+                        504,
+                        578
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                542,
+                                506,
+                                558
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        542,
+                                        506,
+                                        558
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Delving deep into rectifiers: Surpassing",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                555,
+                                505,
+                                568
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        555,
+                                        505,
+                                        568
+                                    ],
+                                    "score": 1.0,
+                                    "content": "human-level performance on imagenet classification. In Proceedings of the 2015 IEEE Interna-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                566,
+                                332,
+                                578
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        566,
+                                        332,
+                                        578
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tional Conference on Computer Vision (ICCV), 2015.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        }
+                    ],
+                    "index": 34
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        585,
+                        504,
+                        618
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                584,
+                                505,
+                                597
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        584,
+                                        505,
+                                        597
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Sepp Hochreiter and Jurgen Schmidhuber. Long short-term memory. ¨ Neural Comput., 9(8):1735–",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                594,
+                                504,
+                                608
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        594,
+                                        504,
+                                        608
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1780, November 1997. ISSN 0899-7667. doi: 10.1162/neco.1997.9.8.1735. URL http://dx.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                117,
+                                607,
+                                325,
+                                619
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        117,
+                                        607,
+                                        325,
+                                        619
+                                    ],
+                                    "score": 1.0,
+                                    "content": "doi.org/10.1162/neco.1997.9.8.1735.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        }
+                    ],
+                    "index": 37
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        625,
+                        505,
+                        680
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                626,
+                                505,
+                                638
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        626,
+                                        505,
+                                        638
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Changwan Hong, Aravind Sukumaran-Rajam, Israt Nisa, Kunal Singh, and P. Sadayappan. Adaptive",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                636,
+                                506,
+                                649
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        636,
+                                        506,
+                                        649
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sparse tiling for sparse matrix multiplication. In Proceedings of the 24th Symposium on Principles",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                647,
+                                505,
+                                660
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        647,
+                                        505,
+                                        660
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and Practice of Parallel Programming, PPoPP ’19, pp. 300–314, New York, NY, USA, 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                656,
+                                505,
+                                672
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        656,
+                                        505,
+                                        672
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ACM. ISBN 978-1-4503-6225-2. doi: 10.1145/3293883.3295712. URL http://doi.acm.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                669,
+                                283,
+                                681
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        669,
+                                        283,
+                                        681
+                                    ],
+                                    "score": 1.0,
+                                    "content": "org/10.1145/3293883.3295712.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        }
+                    ],
+                    "index": 41
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        687,
+                        505,
+                        731
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                687,
+                                505,
+                                700
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        687,
+                                        505,
+                                        700
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Andrew G. Howard, Menglong Zhu, Bo Chen, Dmitry Kalenichenko, Weijun Wang, Tobias Weyand,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                699,
+                                505,
+                                711
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        699,
+                                        505,
+                                        711
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Marco Andreetto, and Hartwig Adam. Mobilenets: Efficient convolutional neural networks for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                710,
+                                505,
+                                722
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        710,
+                                        505,
+                                        722
+                                    ],
+                                    "score": 1.0,
+                                    "content": "mobile vision applications. CoRR, abs/1704.04861, 2017. URL http://arxiv.org/abs/",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                721,
+                                182,
+                                732
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        721,
+                                        182,
+                                        732
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1704.04861.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 47
+                        }
+                    ],
+                    "index": 45.5
+                }
+            ],
+            "page_idx": 10,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        107,
+                        27,
+                        308,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                107,
+                                26,
+                                308,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        26,
+                                        308,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Under review as a conference paper at ICLR 2020",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        300,
+                        751,
+                        310,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                299,
+                                750,
+                                312,
+                                765
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        299,
+                                        750,
+                                        312,
+                                        765
+                                    ],
+                                    "score": 1.0,
+                                    "content": "",
+                                    "type": "text",
+                                    "height": 15,
+                                    "width": 13
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        82,
+                        504,
+                        116
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                82,
+                                504,
+                                95
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        82,
+                                        504,
+                                        95
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Felix Draxler, Kambis Veschgini, Manfred Salmhofer, and Fred A Hamprecht. Essentially No Bar-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                114,
+                                92,
+                                505,
+                                108
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        114,
+                                        92,
+                                        505,
+                                        108
+                                    ],
+                                    "score": 1.0,
+                                    "content": "riers in Neural Network Energy Landscape. In International Conference on Machine Learning,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                104,
+                                144,
+                                117
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        104,
+                                        144,
+                                        117
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        }
+                    ],
+                    "index": 1,
+                    "bbox_fs": [
+                        105,
+                        82,
+                        505,
+                        117
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        123,
+                        502,
+                        145
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                121,
+                                505,
+                                137
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        121,
+                                        505,
+                                        137
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Utku Evci. Detecting dead weights and units in neural networks. CoRR, 2018. URL http:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                117,
+                                135,
+                                277,
+                                146
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        117,
+                                        135,
+                                        277,
+                                        146
+                                    ],
+                                    "score": 1.0,
+                                    "content": "//arxiv.org/abs/1806.06068.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        }
+                    ],
+                    "index": 3.5,
+                    "bbox_fs": [
+                        105,
+                        121,
+                        505,
+                        146
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        152,
+                        504,
+                        175
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                151,
+                                505,
+                                165
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        151,
+                                        505,
+                                        165
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Utku Evci, Fabian Pedregosa, Aidan N. Gomez, and Erich Elsen. The difficulty of training sparse",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                163,
+                                454,
+                                176
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        163,
+                                        454,
+                                        176
+                                    ],
+                                    "score": 1.0,
+                                    "content": "neural networks. ArXiv, 2019. URL http://arxiv.org/abs/1906.10732.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        }
+                    ],
+                    "index": 5.5,
+                    "bbox_fs": [
+                        106,
+                        151,
+                        505,
+                        176
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        182,
+                        505,
+                        227
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                181,
+                                505,
+                                195
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        181,
+                                        505,
+                                        195
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Jonathan Frankle and Michael Carbin. The lottery ticket hypothesis: Finding sparse, trainable neu-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                194,
+                                506,
+                                205
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        194,
+                                        506,
+                                        205
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ral networks. In 7th International Conference on Learning Representations, ICLR 2019, New",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                204,
+                                505,
+                                217
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        204,
+                                        497,
+                                        217
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Orleans, LA, USA, May 6-9, 2019, 2019. URL https://openreview.net/forum?id",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        497,
+                                        205,
+                                        505,
+                                        214
+                                    ],
+                                    "score": 0.34,
+                                    "content": "=",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                216,
+                                182,
+                                227
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        216,
+                                        182,
+                                        227
+                                    ],
+                                    "score": 1.0,
+                                    "content": "rJl-b3RcF7.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        }
+                    ],
+                    "index": 8.5,
+                    "bbox_fs": [
+                        105,
+                        181,
+                        506,
+                        227
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        233,
+                        504,
+                        257
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                232,
+                                505,
+                                247
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        232,
+                                        505,
+                                        247
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Jonathan Frankle, Gintare Karolina Dziugaite, Daniel M. Roy, and Michael Carbin. The lottery",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                244,
+                                488,
+                                258
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        244,
+                                        488,
+                                        258
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ticket hypothesis at scale. ArXiv, 2019. URL http://arxiv.org/abs/1903.01611.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        }
+                    ],
+                    "index": 11.5,
+                    "bbox_fs": [
+                        105,
+                        232,
+                        505,
+                        258
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        263,
+                        503,
+                        286
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                262,
+                                505,
+                                276
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        262,
+                                        505,
+                                        276
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Trevor Gale, Erich Elsen, and Sara Hooker. The state of sparsity in deep neural networks. CoRR,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                273,
+                                424,
+                                287
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        273,
+                                        424,
+                                        287
+                                    ],
+                                    "score": 1.0,
+                                    "content": "abs/1902.09574, 2019. URL http://arxiv.org/abs/1902.09574.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        }
+                    ],
+                    "index": 13.5,
+                    "bbox_fs": [
+                        106,
+                        262,
+                        505,
+                        287
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        293,
+                        505,
+                        326
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                292,
+                                504,
+                                305
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        292,
+                                        504,
+                                        305
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Timur Garipov, Pavel Izmailov, Dmitrii Podoprikhin, Dmitry P Vetrov, and Andrew Gordon Wilson.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                304,
+                                505,
+                                316
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        304,
+                                        505,
+                                        316
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Loss surfaces, mode connectivity, and fast ensembling of dnns. In Advances in Neural Information",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                315,
+                                225,
+                                327
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        315,
+                                        225,
+                                        327
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Processing Systems, 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        }
+                    ],
+                    "index": 16,
+                    "bbox_fs": [
+                        106,
+                        292,
+                        505,
+                        327
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        333,
+                        504,
+                        356
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                331,
+                                505,
+                                347
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        331,
+                                        505,
+                                        347
+                                    ],
+                                    "score": 1.0,
+                                    "content": "J. S. Garofolo, L. F. Lamel, W. M. Fisher, J. G. Fiscus, D. S. Pallett, and N. L. Dahlgren. Darpa",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                344,
+                                370,
+                                357
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        344,
+                                        370,
+                                        357
+                                    ],
+                                    "score": 1.0,
+                                    "content": "timit acoustic phonetic continuous speech corpus cdrom, 1993.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        }
+                    ],
+                    "index": 18.5,
+                    "bbox_fs": [
+                        104,
+                        331,
+                        505,
+                        357
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        363,
+                        503,
+                        386
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                107,
+                                363,
+                                505,
+                                376
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        363,
+                                        505,
+                                        376
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Yiwen Guo, Anbang Yao, and Yurong Chen. Dynamic network surgery for efficient DNNs. CoRR,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                374,
+                                424,
+                                387
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        374,
+                                        424,
+                                        387
+                                    ],
+                                    "score": 1.0,
+                                    "content": "abs/1608.04493, 2016. URL http://arxiv.org/abs/1608.04493.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        }
+                    ],
+                    "index": 20.5,
+                    "bbox_fs": [
+                        107,
+                        363,
+                        505,
+                        387
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        392,
+                        504,
+                        416
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                392,
+                                505,
+                                405
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        392,
+                                        505,
+                                        405
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Song Han, Jeff Pool, John Tran, and William Dally. Learning both weights and connections for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                403,
+                                463,
+                                416
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        403,
+                                        463,
+                                        416
+                                    ],
+                                    "score": 1.0,
+                                    "content": "efficient neural network. In Advances in neural information processing systems, 2015.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        }
+                    ],
+                    "index": 22.5,
+                    "bbox_fs": [
+                        106,
+                        392,
+                        505,
+                        416
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        109,
+                        421,
+                        504,
+                        456
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                421,
+                                505,
+                                435
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        421,
+                                        505,
+                                        435
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Song Han, Xingyu Liu, Huizi Mao, Jing Pu, Ardavan Pedram, Mark A. Horowitz, and William J.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                432,
+                                505,
+                                446
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        432,
+                                        505,
+                                        446
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Dally. EIE: Efficient Inference Engine on compressed deep neural network. In Proceedings of the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                443,
+                                379,
+                                457
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        443,
+                                        379,
+                                        457
+                                    ],
+                                    "score": 1.0,
+                                    "content": "43rd International Symposium on Computer Architecture, 2016a.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        }
+                    ],
+                    "index": 25,
+                    "bbox_fs": [
+                        106,
+                        421,
+                        505,
+                        457
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        462,
+                        505,
+                        507
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                462,
+                                505,
+                                476
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        462,
+                                        505,
+                                        476
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Song Han, Huizi Mao, and William J. Dally. Deep compression: Compressing deep neural net-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                473,
+                                506,
+                                487
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        473,
+                                        506,
+                                        487
+                                    ],
+                                    "score": 1.0,
+                                    "content": "work with pruning, trained quantization and huffman coding. In 4th International Conference on",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                486,
+                                506,
+                                497
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        486,
+                                        506,
+                                        497
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Learning Representations, ICLR 2016, San Juan, Puerto Rico, May 2-4, 2016, Conference Track",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                496,
+                                415,
+                                508
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        496,
+                                        415,
+                                        508
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Proceedings, 2016b. URL http://arxiv.org/abs/1510.00149.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        }
+                    ],
+                    "index": 28.5,
+                    "bbox_fs": [
+                        105,
+                        462,
+                        506,
+                        508
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        514,
+                        504,
+                        536
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                512,
+                                504,
+                                528
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        512,
+                                        504,
+                                        528
+                                    ],
+                                    "score": 1.0,
+                                    "content": "B. Hassibi and D. Stork. Second order derivatives for network pruning: Optimal Brain Surgeon.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                524,
+                                143,
+                                538
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        524,
+                                        143,
+                                        538
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1993.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        }
+                    ],
+                    "index": 31.5,
+                    "bbox_fs": [
+                        104,
+                        512,
+                        504,
+                        538
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        543,
+                        504,
+                        578
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                542,
+                                506,
+                                558
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        542,
+                                        506,
+                                        558
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Delving deep into rectifiers: Surpassing",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                555,
+                                505,
+                                568
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        555,
+                                        505,
+                                        568
+                                    ],
+                                    "score": 1.0,
+                                    "content": "human-level performance on imagenet classification. In Proceedings of the 2015 IEEE Interna-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                566,
+                                332,
+                                578
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        566,
+                                        332,
+                                        578
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tional Conference on Computer Vision (ICCV), 2015.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        }
+                    ],
+                    "index": 34,
+                    "bbox_fs": [
+                        105,
+                        542,
+                        506,
+                        578
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        585,
+                        504,
+                        618
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                584,
+                                505,
+                                597
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        584,
+                                        505,
+                                        597
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Sepp Hochreiter and Jurgen Schmidhuber. Long short-term memory. ¨ Neural Comput., 9(8):1735–",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                594,
+                                504,
+                                608
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        594,
+                                        504,
+                                        608
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1780, November 1997. ISSN 0899-7667. doi: 10.1162/neco.1997.9.8.1735. URL http://dx.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                117,
+                                607,
+                                325,
+                                619
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        117,
+                                        607,
+                                        325,
+                                        619
+                                    ],
+                                    "score": 1.0,
+                                    "content": "doi.org/10.1162/neco.1997.9.8.1735.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        }
+                    ],
+                    "index": 37,
+                    "bbox_fs": [
+                        106,
+                        584,
+                        505,
+                        619
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        625,
+                        505,
+                        680
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                626,
+                                505,
+                                638
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        626,
+                                        505,
+                                        638
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Changwan Hong, Aravind Sukumaran-Rajam, Israt Nisa, Kunal Singh, and P. Sadayappan. Adaptive",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                636,
+                                506,
+                                649
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        636,
+                                        506,
+                                        649
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sparse tiling for sparse matrix multiplication. In Proceedings of the 24th Symposium on Principles",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                647,
+                                505,
+                                660
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        647,
+                                        505,
+                                        660
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and Practice of Parallel Programming, PPoPP ’19, pp. 300–314, New York, NY, USA, 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                656,
+                                505,
+                                672
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        656,
+                                        505,
+                                        672
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ACM. ISBN 978-1-4503-6225-2. doi: 10.1145/3293883.3295712. URL http://doi.acm.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                669,
+                                283,
+                                681
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        669,
+                                        283,
+                                        681
+                                    ],
+                                    "score": 1.0,
+                                    "content": "org/10.1145/3293883.3295712.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        }
+                    ],
+                    "index": 41,
+                    "bbox_fs": [
+                        106,
+                        626,
+                        506,
+                        681
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        687,
+                        505,
+                        731
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                687,
+                                505,
+                                700
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        687,
+                                        505,
+                                        700
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Andrew G. Howard, Menglong Zhu, Bo Chen, Dmitry Kalenichenko, Weijun Wang, Tobias Weyand,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                699,
+                                505,
+                                711
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        699,
+                                        505,
+                                        711
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Marco Andreetto, and Hartwig Adam. Mobilenets: Efficient convolutional neural networks for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                710,
+                                505,
+                                722
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        710,
+                                        505,
+                                        722
+                                    ],
+                                    "score": 1.0,
+                                    "content": "mobile vision applications. CoRR, abs/1704.04861, 2017. URL http://arxiv.org/abs/",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                721,
+                                182,
+                                732
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        721,
+                                        182,
+                                        732
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1704.04861.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 47
+                        }
+                    ],
+                    "index": 45.5,
+                    "bbox_fs": [
+                        105,
+                        687,
+                        505,
+                        732
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        82,
+                        503,
+                        116
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                82,
+                                505,
+                                96
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        82,
+                                        505,
+                                        96
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Nal Kalchbrenner, Erich Elsen, Karen Simonyan, Seb Noury, Norman Casagrande, Edward Lock-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                94,
+                                505,
+                                106
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        94,
+                                        505,
+                                        106
+                                    ],
+                                    "score": 1.0,
+                                    "content": "hart, Florian Stimberg, Aaron Oord, Sander Dieleman, and Koray Kavukcuoglu. Efficient neural",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                104,
+                                447,
+                                118
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        104,
+                                        447,
+                                        118
+                                    ],
+                                    "score": 1.0,
+                                    "content": "audio synthesis. In International Conference on Machine Learning (ICML), 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        }
+                    ],
+                    "index": 1
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        122,
+                        503,
+                        145
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                122,
+                                505,
+                                135
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        122,
+                                        505,
+                                        135
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Alex Krizhevsky, Vinod Nair, and Geoffrey Hinton. Cifar-10 (canadian institute for advanced re-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                134,
+                                430,
+                                146
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        134,
+                                        430,
+                                        146
+                                    ],
+                                    "score": 1.0,
+                                    "content": "search). URL http://www.cs.toronto.edu/˜kriz/cifar.html.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        }
+                    ],
+                    "index": 3.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        151,
+                        504,
+                        175
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                151,
+                                506,
+                                164
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        151,
+                                        506,
+                                        164
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Yann LeCun, John S. Denker, and Sara A. Solla. Optimal Brain Damage. In Advances in Neural",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                163,
+                                274,
+                                175
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        163,
+                                        274,
+                                        175
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Information Processing Systems, 1990.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        }
+                    ],
+                    "index": 5.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        181,
+                        504,
+                        215
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                181,
+                                504,
+                                194
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        181,
+                                        504,
+                                        194
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Namhoon Lee, Thalaiyasingam Ajanthan, and Philip H. S. Torr. SNIP: Single-shot Network Prun-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                192,
+                                505,
+                                205
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        192,
+                                        505,
+                                        205
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ing based on Connection Sensitivity. In International Conference on Learning Representations",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                204,
+                                199,
+                                215
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        204,
+                                        199,
+                                        215
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(ICLR), 2019, 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        }
+                    ],
+                    "index": 8
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        221,
+                        504,
+                        245
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                221,
+                                505,
+                                234
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        221,
+                                        505,
+                                        234
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Hao Li, Asim Kadav, Igor Durdanovic, Hanan Samet, and Hans Peter Graf. Pruning filters for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                232,
+                                451,
+                                245
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        232,
+                                        451,
+                                        245
+                                    ],
+                                    "score": 1.0,
+                                    "content": "efficient convnets. In International Conference on Learning Representations, 2016.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        }
+                    ],
+                    "index": 10.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        250,
+                        505,
+                        295
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                251,
+                                505,
+                                264
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        251,
+                                        505,
+                                        264
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Chen Liu, Guillaume Bellec, Bernhard Vogginger, David Kappel, Johannes Partzsch, Felix Neu-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                263,
+                                505,
+                                275
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        263,
+                                        505,
+                                        275
+                                    ],
+                                    "score": 1.0,
+                                    "content": "maerker, Sebastian Hoppner, Wolfgang Maass, Stephen B. Furber, Robert A. Legenstein, and ¨",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                273,
+                                506,
+                                286
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        273,
+                                        506,
+                                        286
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Christian Mayr. Memory-efficient deep learning on a spinnaker 2 prototype. In Front. Neurosci.,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                283,
+                                143,
+                                296
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        283,
+                                        143,
+                                        296
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        }
+                    ],
+                    "index": 13.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        302,
+                        503,
+                        325
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                302,
+                                505,
+                                316
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        302,
+                                        505,
+                                        316
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Christos Louizos, Karen Ullrich, and Max Welling. Bayesian compression for deep learning. In",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                313,
+                                356,
+                                326
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        313,
+                                        356,
+                                        326
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Advances in Neural Information Processing Systems, 2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        }
+                    ],
+                    "index": 16.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        331,
+                        504,
+                        355
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                331,
+                                505,
+                                344
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        331,
+                                        505,
+                                        344
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Stephen Merity, Caiming Xiong, James Bradbury, and Richard Socher. Pointer sentinel mixture",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                343,
+                                419,
+                                356
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        343,
+                                        419,
+                                        356
+                                    ],
+                                    "score": 1.0,
+                                    "content": "models. ArXiv, 2016. URL http://arxiv.org/abs/1609.07843.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        }
+                    ],
+                    "index": 18.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        361,
+                        505,
+                        417
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                360,
+                                505,
+                                375
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        360,
+                                        505,
+                                        375
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Duane Merrill and Michael Garland. Merge-based sparse matrix-vector multiplication (spmv) using",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                372,
+                                505,
+                                385
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        372,
+                                        505,
+                                        385
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the csr storage format. In Proceedings of the 21st ACM SIGPLAN Symposium on Principles and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                383,
+                                505,
+                                397
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        383,
+                                        505,
+                                        397
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Practice of Parallel Programming, PPoPP ’16, pp. 43:1–43:2, New York, NY, USA, 2016. ACM.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                393,
+                                505,
+                                407
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        393,
+                                        505,
+                                        407
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ISBN 978-1-4503-4092-2. doi: 10.1145/2851141.2851190. URL http://doi.acm.org/",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                405,
+                                259,
+                                417
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        405,
+                                        259,
+                                        417
+                                    ],
+                                    "score": 1.0,
+                                    "content": "10.1145/2851141.2851190.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        }
+                    ],
+                    "index": 22
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        424,
+                        505,
+                        480
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                424,
+                                505,
+                                436
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        424,
+                                        505,
+                                        436
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Peter Baldwin Martijn Bastiaan Oliver Bunting Aiken Cairncross Christopher Chalmers Liz Cor-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                433,
+                                505,
+                                449
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        433,
+                                        505,
+                                        449
+                                    ],
+                                    "score": 1.0,
+                                    "content": "rigan Sam Davis Nathan van Doorn Jon Fowler Graham Hazel Basile Henry David Page Jonny",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                445,
+                                505,
+                                459
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        445,
+                                        505,
+                                        459
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Shipton Shaun Steenkamp Mike Ashby, Christiaan Baaij. Exploiting unstructured sparsity on",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                456,
+                                505,
+                                469
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        456,
+                                        505,
+                                        469
+                                    ],
+                                    "score": 1.0,
+                                    "content": "next-generation datacenter hardware. 2019. URL https://myrtle.ai/wp-content/",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                468,
+                                426,
+                                481
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        468,
+                                        426,
+                                        481
+                                    ],
+                                    "score": 1.0,
+                                    "content": "uploads/2019/06/IEEEformatMyrtle.ai_.21.06.19_b.pdf.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        }
+                    ],
+                    "index": 27
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        486,
+                        504,
+                        519
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                485,
+                                505,
+                                500
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        485,
+                                        505,
+                                        500
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Decebal Constantin Mocanu, Elena Mocanu, Peter Stone, Phuong H Nguyen, Madeleine Gibescu,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                496,
+                                505,
+                                510
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        496,
+                                        505,
+                                        510
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and Antonio Liotta. Scalable training of artificial neural networks with adaptive sparse connec-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                508,
+                                385,
+                                520
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        508,
+                                        385,
+                                        520
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tivity inspired by network science. Nature Communications, 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        }
+                    ],
+                    "index": 31
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        525,
+                        503,
+                        560
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                525,
+                                505,
+                                540
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        525,
+                                        505,
+                                        540
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Dmitry Molchanov, Arsenii Ashukha, and Dmitry P. Vetrov. Variational Dropout Sparsifies Deep",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                536,
+                                505,
+                                551
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        536,
+                                        505,
+                                        551
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Neural Networks. In Proceedings of the 34th International Conference on Machine Learning,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                548,
+                                431,
+                                561
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        548,
+                                        431,
+                                        561
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ICML 2017, Sydney, NSW, Australia, 6-11 August 2017, pp. 2498–2507, 2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        }
+                    ],
+                    "index": 34
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        104,
+                        566,
+                        504,
+                        589
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                566,
+                                505,
+                                579
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        566,
+                                        505,
+                                        579
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Pavlo Molchanov, Stephen Tyree, Tero Karras, Timo Aila, and Jan Kautz. Pruning Convolutional",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                578,
+                                477,
+                                589
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        578,
+                                        477,
+                                        589
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Neural Networks for Resource Efficient Transfer Learning. CoRR, abs/1611.06440, 2016.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        }
+                    ],
+                    "index": 36.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        595,
+                        505,
+                        641
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                596,
+                                505,
+                                608
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        596,
+                                        505,
+                                        608
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Hesham Mostafa and Xin Wang. Parameter efficient training of deep convolutional neural networks",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                608,
+                                505,
+                                619
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        608,
+                                        505,
+                                        619
+                                    ],
+                                    "score": 1.0,
+                                    "content": "by dynamic sparse reparameterization. In Proceedings of the 36th International Conference on",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                618,
+                                505,
+                                630
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        618,
+                                        505,
+                                        630
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Machine Learning, ICML 2019, 9-15 June 2019, Long Beach, California, USA, pp. 4646–4655,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                630,
+                                457,
+                                641
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        630,
+                                        457,
+                                        641
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2019. URL http://proceedings.mlr.press/v97/mostafa19a.html.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        }
+                    ],
+                    "index": 39.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        647,
+                        505,
+                        692
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                646,
+                                505,
+                                660
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        646,
+                                        505,
+                                        660
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Sharan Narang, Greg Diamos, Shubho Sengupta, and Erich Elsen. Exploring sparsity in re-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                659,
+                                505,
+                                671
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        659,
+                                        505,
+                                        671
+                                    ],
+                                    "score": 1.0,
+                                    "content": "current neural networks. In 5th International Conference on Learning Representations, ICLR",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                668,
+                                505,
+                                683
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        668,
+                                        505,
+                                        683
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2017, Toulon, France, April 24-26, 2017, Conference Track Proceedings, 2017. URL https:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                680,
+                                331,
+                                694
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        680,
+                                        331,
+                                        694
+                                    ],
+                                    "score": 1.0,
+                                    "content": "//openreview.net/forum?id=BylSPv9gx.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        }
+                    ],
+                    "index": 43.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        699,
+                        504,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                698,
+                                505,
+                                711
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        698,
+                                        505,
+                                        711
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Jongsoo Park, Sheng R. Li, Wei Wen, Hai Li, Yiran Chen, and Pradeep Dubey. Holistic SparseCNN:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                709,
+                                504,
+                                722
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        709,
+                                        504,
+                                        722
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Forging the trident of accuracy, speed, and size. CoRR, abs/1608.01409, 2016. URL http:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 47
+                        },
+                        {
+                            "bbox": [
+                                117,
+                                721,
+                                277,
+                                732
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        117,
+                                        721,
+                                        277,
+                                        732
+                                    ],
+                                    "score": 1.0,
+                                    "content": "//arxiv.org/abs/1608.01409.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 48
+                        }
+                    ],
+                    "index": 47
+                }
+            ],
+            "page_idx": 11,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        107,
+                        27,
+                        308,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                107,
+                                26,
+                                308,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        26,
+                                        308,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Under review as a conference paper at ICLR 2020",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        300,
+                        751,
+                        311,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                299,
+                                750,
+                                313,
+                                764
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        299,
+                                        750,
+                                        313,
+                                        764
+                                    ],
+                                    "score": 1.0,
+                                    "content": "12",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        82,
+                        503,
+                        116
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                82,
+                                505,
+                                96
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        82,
+                                        505,
+                                        96
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Nal Kalchbrenner, Erich Elsen, Karen Simonyan, Seb Noury, Norman Casagrande, Edward Lock-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                94,
+                                505,
+                                106
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        94,
+                                        505,
+                                        106
+                                    ],
+                                    "score": 1.0,
+                                    "content": "hart, Florian Stimberg, Aaron Oord, Sander Dieleman, and Koray Kavukcuoglu. Efficient neural",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                104,
+                                447,
+                                118
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        104,
+                                        447,
+                                        118
+                                    ],
+                                    "score": 1.0,
+                                    "content": "audio synthesis. In International Conference on Machine Learning (ICML), 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        }
+                    ],
+                    "index": 1,
+                    "bbox_fs": [
+                        105,
+                        82,
+                        505,
+                        118
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        122,
+                        503,
+                        145
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                122,
+                                505,
+                                135
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        122,
+                                        505,
+                                        135
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Alex Krizhevsky, Vinod Nair, and Geoffrey Hinton. Cifar-10 (canadian institute for advanced re-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                134,
+                                430,
+                                146
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        134,
+                                        430,
+                                        146
+                                    ],
+                                    "score": 1.0,
+                                    "content": "search). URL http://www.cs.toronto.edu/˜kriz/cifar.html.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        }
+                    ],
+                    "index": 3.5,
+                    "bbox_fs": [
+                        106,
+                        122,
+                        505,
+                        146
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        151,
+                        504,
+                        175
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                151,
+                                506,
+                                164
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        151,
+                                        506,
+                                        164
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Yann LeCun, John S. Denker, and Sara A. Solla. Optimal Brain Damage. In Advances in Neural",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                163,
+                                274,
+                                175
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        163,
+                                        274,
+                                        175
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Information Processing Systems, 1990.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        }
+                    ],
+                    "index": 5.5,
+                    "bbox_fs": [
+                        106,
+                        151,
+                        506,
+                        175
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        181,
+                        504,
+                        215
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                181,
+                                504,
+                                194
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        181,
+                                        504,
+                                        194
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Namhoon Lee, Thalaiyasingam Ajanthan, and Philip H. S. Torr. SNIP: Single-shot Network Prun-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                192,
+                                505,
+                                205
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        192,
+                                        505,
+                                        205
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ing based on Connection Sensitivity. In International Conference on Learning Representations",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                204,
+                                199,
+                                215
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        204,
+                                        199,
+                                        215
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(ICLR), 2019, 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        }
+                    ],
+                    "index": 8,
+                    "bbox_fs": [
+                        106,
+                        181,
+                        505,
+                        215
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        221,
+                        504,
+                        245
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                221,
+                                505,
+                                234
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        221,
+                                        505,
+                                        234
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Hao Li, Asim Kadav, Igor Durdanovic, Hanan Samet, and Hans Peter Graf. Pruning filters for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                232,
+                                451,
+                                245
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        232,
+                                        451,
+                                        245
+                                    ],
+                                    "score": 1.0,
+                                    "content": "efficient convnets. In International Conference on Learning Representations, 2016.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        }
+                    ],
+                    "index": 10.5,
+                    "bbox_fs": [
+                        106,
+                        221,
+                        505,
+                        245
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        250,
+                        505,
+                        295
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                251,
+                                505,
+                                264
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        251,
+                                        505,
+                                        264
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Chen Liu, Guillaume Bellec, Bernhard Vogginger, David Kappel, Johannes Partzsch, Felix Neu-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                263,
+                                505,
+                                275
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        263,
+                                        505,
+                                        275
+                                    ],
+                                    "score": 1.0,
+                                    "content": "maerker, Sebastian Hoppner, Wolfgang Maass, Stephen B. Furber, Robert A. Legenstein, and ¨",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                273,
+                                506,
+                                286
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        273,
+                                        506,
+                                        286
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Christian Mayr. Memory-efficient deep learning on a spinnaker 2 prototype. In Front. Neurosci.,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                283,
+                                143,
+                                296
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        283,
+                                        143,
+                                        296
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        }
+                    ],
+                    "index": 13.5,
+                    "bbox_fs": [
+                        106,
+                        251,
+                        506,
+                        296
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        302,
+                        503,
+                        325
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                302,
+                                505,
+                                316
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        302,
+                                        505,
+                                        316
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Christos Louizos, Karen Ullrich, and Max Welling. Bayesian compression for deep learning. In",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                313,
+                                356,
+                                326
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        313,
+                                        356,
+                                        326
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Advances in Neural Information Processing Systems, 2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        }
+                    ],
+                    "index": 16.5,
+                    "bbox_fs": [
+                        105,
+                        302,
+                        505,
+                        326
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        331,
+                        504,
+                        355
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                331,
+                                505,
+                                344
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        331,
+                                        505,
+                                        344
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Stephen Merity, Caiming Xiong, James Bradbury, and Richard Socher. Pointer sentinel mixture",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                343,
+                                419,
+                                356
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        343,
+                                        419,
+                                        356
+                                    ],
+                                    "score": 1.0,
+                                    "content": "models. ArXiv, 2016. URL http://arxiv.org/abs/1609.07843.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        }
+                    ],
+                    "index": 18.5,
+                    "bbox_fs": [
+                        106,
+                        331,
+                        505,
+                        356
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        361,
+                        505,
+                        417
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                360,
+                                505,
+                                375
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        360,
+                                        505,
+                                        375
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Duane Merrill and Michael Garland. Merge-based sparse matrix-vector multiplication (spmv) using",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                372,
+                                505,
+                                385
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        372,
+                                        505,
+                                        385
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the csr storage format. In Proceedings of the 21st ACM SIGPLAN Symposium on Principles and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                383,
+                                505,
+                                397
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        383,
+                                        505,
+                                        397
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Practice of Parallel Programming, PPoPP ’16, pp. 43:1–43:2, New York, NY, USA, 2016. ACM.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                393,
+                                505,
+                                407
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        393,
+                                        505,
+                                        407
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ISBN 978-1-4503-4092-2. doi: 10.1145/2851141.2851190. URL http://doi.acm.org/",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                405,
+                                259,
+                                417
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        405,
+                                        259,
+                                        417
+                                    ],
+                                    "score": 1.0,
+                                    "content": "10.1145/2851141.2851190.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        }
+                    ],
+                    "index": 22,
+                    "bbox_fs": [
+                        105,
+                        360,
+                        505,
+                        417
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        424,
+                        505,
+                        480
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                424,
+                                505,
+                                436
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        424,
+                                        505,
+                                        436
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Peter Baldwin Martijn Bastiaan Oliver Bunting Aiken Cairncross Christopher Chalmers Liz Cor-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                433,
+                                505,
+                                449
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        433,
+                                        505,
+                                        449
+                                    ],
+                                    "score": 1.0,
+                                    "content": "rigan Sam Davis Nathan van Doorn Jon Fowler Graham Hazel Basile Henry David Page Jonny",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                445,
+                                505,
+                                459
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        445,
+                                        505,
+                                        459
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Shipton Shaun Steenkamp Mike Ashby, Christiaan Baaij. Exploiting unstructured sparsity on",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                456,
+                                505,
+                                469
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        456,
+                                        505,
+                                        469
+                                    ],
+                                    "score": 1.0,
+                                    "content": "next-generation datacenter hardware. 2019. URL https://myrtle.ai/wp-content/",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                468,
+                                426,
+                                481
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        468,
+                                        426,
+                                        481
+                                    ],
+                                    "score": 1.0,
+                                    "content": "uploads/2019/06/IEEEformatMyrtle.ai_.21.06.19_b.pdf.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        }
+                    ],
+                    "index": 27,
+                    "bbox_fs": [
+                        106,
+                        424,
+                        505,
+                        481
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        486,
+                        504,
+                        519
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                485,
+                                505,
+                                500
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        485,
+                                        505,
+                                        500
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Decebal Constantin Mocanu, Elena Mocanu, Peter Stone, Phuong H Nguyen, Madeleine Gibescu,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                496,
+                                505,
+                                510
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        496,
+                                        505,
+                                        510
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and Antonio Liotta. Scalable training of artificial neural networks with adaptive sparse connec-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                508,
+                                385,
+                                520
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        508,
+                                        385,
+                                        520
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tivity inspired by network science. Nature Communications, 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        }
+                    ],
+                    "index": 31,
+                    "bbox_fs": [
+                        105,
+                        485,
+                        505,
+                        520
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        525,
+                        503,
+                        560
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                525,
+                                505,
+                                540
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        525,
+                                        505,
+                                        540
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Dmitry Molchanov, Arsenii Ashukha, and Dmitry P. Vetrov. Variational Dropout Sparsifies Deep",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                536,
+                                505,
+                                551
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        536,
+                                        505,
+                                        551
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Neural Networks. In Proceedings of the 34th International Conference on Machine Learning,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                548,
+                                431,
+                                561
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        548,
+                                        431,
+                                        561
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ICML 2017, Sydney, NSW, Australia, 6-11 August 2017, pp. 2498–2507, 2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        }
+                    ],
+                    "index": 34,
+                    "bbox_fs": [
+                        105,
+                        525,
+                        505,
+                        561
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        104,
+                        566,
+                        504,
+                        589
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                566,
+                                505,
+                                579
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        566,
+                                        505,
+                                        579
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Pavlo Molchanov, Stephen Tyree, Tero Karras, Timo Aila, and Jan Kautz. Pruning Convolutional",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                578,
+                                477,
+                                589
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        578,
+                                        477,
+                                        589
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Neural Networks for Resource Efficient Transfer Learning. CoRR, abs/1611.06440, 2016.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        }
+                    ],
+                    "index": 36.5,
+                    "bbox_fs": [
+                        105,
+                        566,
+                        505,
+                        589
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        595,
+                        505,
+                        641
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                596,
+                                505,
+                                608
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        596,
+                                        505,
+                                        608
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Hesham Mostafa and Xin Wang. Parameter efficient training of deep convolutional neural networks",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                608,
+                                505,
+                                619
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        608,
+                                        505,
+                                        619
+                                    ],
+                                    "score": 1.0,
+                                    "content": "by dynamic sparse reparameterization. In Proceedings of the 36th International Conference on",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                618,
+                                505,
+                                630
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        618,
+                                        505,
+                                        630
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Machine Learning, ICML 2019, 9-15 June 2019, Long Beach, California, USA, pp. 4646–4655,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                630,
+                                457,
+                                641
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        630,
+                                        457,
+                                        641
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2019. URL http://proceedings.mlr.press/v97/mostafa19a.html.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        }
+                    ],
+                    "index": 39.5,
+                    "bbox_fs": [
+                        106,
+                        596,
+                        505,
+                        641
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        647,
+                        505,
+                        692
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                646,
+                                505,
+                                660
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        646,
+                                        505,
+                                        660
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Sharan Narang, Greg Diamos, Shubho Sengupta, and Erich Elsen. Exploring sparsity in re-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                659,
+                                505,
+                                671
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        659,
+                                        505,
+                                        671
+                                    ],
+                                    "score": 1.0,
+                                    "content": "current neural networks. In 5th International Conference on Learning Representations, ICLR",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                668,
+                                505,
+                                683
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        668,
+                                        505,
+                                        683
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2017, Toulon, France, April 24-26, 2017, Conference Track Proceedings, 2017. URL https:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                680,
+                                331,
+                                694
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        680,
+                                        331,
+                                        694
+                                    ],
+                                    "score": 1.0,
+                                    "content": "//openreview.net/forum?id=BylSPv9gx.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        }
+                    ],
+                    "index": 43.5,
+                    "bbox_fs": [
+                        105,
+                        646,
+                        505,
+                        694
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        699,
+                        504,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                698,
+                                505,
+                                711
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        698,
+                                        505,
+                                        711
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Jongsoo Park, Sheng R. Li, Wei Wen, Hai Li, Yiran Chen, and Pradeep Dubey. Holistic SparseCNN:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                709,
+                                504,
+                                722
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        709,
+                                        504,
+                                        722
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Forging the trident of accuracy, speed, and size. CoRR, abs/1608.01409, 2016. URL http:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 47
+                        },
+                        {
+                            "bbox": [
+                                117,
+                                721,
+                                277,
+                                732
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        117,
+                                        721,
+                                        277,
+                                        732
+                                    ],
+                                    "score": 1.0,
+                                    "content": "//arxiv.org/abs/1608.01409.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 48
+                        }
+                    ],
+                    "index": 47,
+                    "bbox_fs": [
+                        105,
+                        698,
+                        505,
+                        732
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        82,
+                        505,
+                        127
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                81,
+                                505,
+                                96
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        81,
+                                        505,
+                                        96
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Olga Russakovsky, Jia Deng, Hao Su, Jonathan Krause, Sanjeev Satheesh, Sean Ma, Zhiheng",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                94,
+                                505,
+                                106
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        94,
+                                        505,
+                                        106
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Huang, Andrej Karpathy, Aditya Khosla, Michael Bernstein, Alexander C. Berg, and Li Fei-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                104,
+                                505,
+                                118
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        104,
+                                        505,
+                                        118
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Fei. Imagenet large scale visual recognition challenge. International Journal of Computer Vision",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                115,
+                                174,
+                                127
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        115,
+                                        174,
+                                        127
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(IJCV), 2015.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        }
+                    ],
+                    "index": 1.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        134,
+                        504,
+                        168
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                135,
+                                505,
+                                146
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        135,
+                                        505,
+                                        146
+                                    ],
+                                    "score": 1.0,
+                                    "content": "M. Sandler, A. Howard, M. Zhu, A. Zhmoginov, and L. Chen. Mobilenetv2: Inverted residuals and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                145,
+                                505,
+                                158
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        145,
+                                        505,
+                                        158
+                                    ],
+                                    "score": 1.0,
+                                    "content": "linear bottlenecks. In 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                156,
+                                363,
+                                168
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        156,
+                                        363,
+                                        168
+                                    ],
+                                    "score": 1.0,
+                                    "content": "pp. 4510–4520, June 2018. doi: 10.1109/CVPR.2018.00474.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        }
+                    ],
+                    "index": 5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        175,
+                        506,
+                        209
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                175,
+                                506,
+                                189
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        175,
+                                        506,
+                                        189
+                                    ],
+                                    "score": 1.0,
+                                    "content": "S. Srinivas, A. Subramanya, and R. V. Babu. Training sparse neural networks. In 2017 IEEE",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                186,
+                                505,
+                                199
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        186,
+                                        505,
+                                        199
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Conference on Computer Vision and Pattern Recognition Workshops (CVPRW), pp. 455–462,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                197,
+                                290,
+                                209
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        197,
+                                        290,
+                                        209
+                                    ],
+                                    "score": 1.0,
+                                    "content": "July 2017. doi: 10.1109/CVPRW.2017.61.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        }
+                    ],
+                    "index": 8
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        216,
+                        504,
+                        239
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                216,
+                                506,
+                                230
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        216,
+                                        506,
+                                        230
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Nikko Strom. Sparse Connection and Pruning in Large Dynamic Artificial Neural Networks. In ¨",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                227,
+                                210,
+                                240
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        227,
+                                        210,
+                                        240
+                                    ],
+                                    "score": 1.0,
+                                    "content": "EUROSPEECH, 1997.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        }
+                    ],
+                    "index": 10.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        246,
+                        505,
+                        290
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                246,
+                                505,
+                                259
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        246,
+                                        505,
+                                        259
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Christian Szegedy, Vincent Vanhoucke, Sergey Ioffe, Jonathon Shlens, and Zbigniew Wojna. Re-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                258,
+                                505,
+                                270
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        258,
+                                        505,
+                                        270
+                                    ],
+                                    "score": 1.0,
+                                    "content": "thinking the inception architecture for computer vision. In Proceedings of IEEE Conference on",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                268,
+                                504,
+                                281
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        268,
+                                        504,
+                                        281
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Computer Vision and Pattern Recognition,, 2016. URL http://arxiv.org/abs/1512.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                280,
+                                152,
+                                291
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        280,
+                                        152,
+                                        291
+                                    ],
+                                    "score": 1.0,
+                                    "content": "00567.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        }
+                    ],
+                    "index": 13.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        298,
+                        505,
+                        343
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                299,
+                                505,
+                                311
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        299,
+                                        505,
+                                        311
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Enzo Tartaglione, Skjalg Lepsøy, Attilio Fiandrotti, and Gianluca Francini. Learning sparse neural",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                310,
+                                505,
+                                322
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        310,
+                                        505,
+                                        322
+                                    ],
+                                    "score": 1.0,
+                                    "content": "networks via sensitivity-driven regularization. In Proceedings of the 32nd International Confer-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                321,
+                                505,
+                                333
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        321,
+                                        505,
+                                        333
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ence on Neural Information Processing Systems, NIPS’18, pp. 3882–3892, USA, 2018. Curran",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                331,
+                                502,
+                                344
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        331,
+                                        400,
+                                        344
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Associates Inc. URL http://dl.acm.org/citation.cfm?id",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        401,
+                                        333,
+                                        407,
+                                        341
+                                    ],
+                                    "score": 0.27,
+                                    "content": "\\underline { { \\underline { { \\mathbf { \\Pi } } } } } =",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        407,
+                                        331,
+                                        502,
+                                        344
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3327144.3327303.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        }
+                    ],
+                    "index": 17.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        350,
+                        504,
+                        373
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                349,
+                                505,
+                                364
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        349,
+                                        505,
+                                        364
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Georg Thimm and Emile Fiesler. Evaluating pruning methods. In National Chiao-Tung University,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                362,
+                                169,
+                                374
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        362,
+                                        169,
+                                        374
+                                    ],
+                                    "score": 1.0,
+                                    "content": "pp. 2, 1995.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        }
+                    ],
+                    "index": 20.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        380,
+                        505,
+                        437
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                381,
+                                506,
+                                394
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        381,
+                                        506,
+                                        394
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Peiqi Wang, Yu Ji, Chi Hong, Yongqiang Lyu, Dongsheng Wang, and Yuan Xie. Snrram: An efficient",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                392,
+                                506,
+                                405
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        392,
+                                        506,
+                                        405
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sparse neural network computation architecture based on resistive random-access memory. In",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                403,
+                                505,
+                                416
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        403,
+                                        505,
+                                        416
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Proceedings of the 55th Annual Design Automation Conference, DAC ’18, pp. 106:1–106:6, New",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                117,
+                                414,
+                                505,
+                                426
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        117,
+                                        414,
+                                        505,
+                                        426
+                                    ],
+                                    "score": 1.0,
+                                    "content": "York, NY, USA, 2018. ACM. ISBN 978-1-4503-5700-5. doi: 10.1145/3195970.3196116. URL",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                425,
+                                373,
+                                437
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        425,
+                                        373,
+                                        437
+                                    ],
+                                    "score": 1.0,
+                                    "content": "http://doi.acm.org/10.1145/3195970.3196116.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        }
+                    ],
+                    "index": 24
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        443,
+                        444,
+                        455
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                442,
+                                447,
+                                457
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        442,
+                                        447,
+                                        457
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Sergey Zagoruyko and Nikos Komodakis. Wide residual networks. In BMVC, 2016.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        }
+                    ],
+                    "index": 27
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        462,
+                        502,
+                        485
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                461,
+                                505,
+                                477
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        461,
+                                        505,
+                                        477
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Hattie Zhou, Janice Lan, Rosanne Liu, and Jason Yosinski. Deconstructing Lottery Tickets: Zeros,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                474,
+                                279,
+                                485
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        474,
+                                        279,
+                                        485
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Signs, and the Supermask. ArXiv, 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        }
+                    ],
+                    "index": 28.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        492,
+                        505,
+                        516
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                493,
+                                505,
+                                506
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        493,
+                                        505,
+                                        506
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Michael Zhu and Suyog Gupta. To Prune, or Not to Prune: Exploring the Efficacy of Pruning for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                503,
+                                504,
+                                516
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        503,
+                                        504,
+                                        516
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Model Compression. In International Conference on Learning Representations Workshop, 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        }
+                    ],
+                    "index": 30.5
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        536,
+                        424,
+                        550
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                536,
+                                425,
+                                551
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        536,
+                                        425,
+                                        551
+                                    ],
+                                    "score": 1.0,
+                                    "content": "A EFFECT OF MASK UPDATES ON THE ENERGY LANDSCAPE",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        }
+                    ],
+                    "index": 32
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        561,
+                        506,
+                        628
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                560,
+                                506,
+                                576
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        560,
+                                        437,
+                                        576
+                                    ],
+                                    "score": 1.0,
+                                    "content": "To update the connectivity of our sparse network, we first need to drop a fraction",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        437,
+                                        563,
+                                        444,
+                                        572
+                                    ],
+                                    "score": 0.78,
+                                    "content": "d",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        444,
+                                        560,
+                                        506,
+                                        576
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of the existing",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                572,
+                                506,
+                                587
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        572,
+                                        506,
+                                        587
+                                    ],
+                                    "score": 1.0,
+                                    "content": "connections for each layer independently to create a budget for growing new connections. Following",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                583,
+                                506,
+                                596
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        583,
+                                        443,
+                                        596
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the recipe of magnitude based pruning(Han et al., 2015), we order parameters at layer",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        443,
+                                        585,
+                                        448,
+                                        594
+                                    ],
+                                    "score": 0.77,
+                                    "content": "i",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        448,
+                                        583,
+                                        506,
+                                        596
+                                    ],
+                                    "score": 1.0,
+                                    "content": "by magnitude",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                107,
+                                594,
+                                506,
+                                608
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        596,
+                                        121,
+                                        607
+                                    ],
+                                    "score": 0.89,
+                                    "content": "| \\theta _ { i } |",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        122,
+                                        594,
+                                        174,
+                                        608
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and drop the",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        174,
+                                        595,
+                                        235,
+                                        606
+                                    ],
+                                    "score": 0.92,
+                                    "content": "N * ( 1 - S ) * d",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        235,
+                                        594,
+                                        506,
+                                        608
+                                    ],
+                                    "score": 1.0,
+                                    "content": "parameters with lowest magnitude. The effectiveness of this simple",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                606,
+                                506,
+                                618
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        606,
+                                        421,
+                                        618
+                                    ],
+                                    "score": 1.0,
+                                    "content": "criteria can be explained through the first order Taylor approximation of the loss",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        422,
+                                        606,
+                                        430,
+                                        615
+                                    ],
+                                    "score": 0.83,
+                                    "content": "L",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        430,
+                                        606,
+                                        506,
+                                        618
+                                    ],
+                                    "score": 1.0,
+                                    "content": "around the current",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                617,
+                                187,
+                                629
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        617,
+                                        177,
+                                        629
+                                    ],
+                                    "score": 1.0,
+                                    "content": "set of parameters",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        177,
+                                        618,
+                                        183,
+                                        626
+                                    ],
+                                    "score": 0.81,
+                                    "content": "\\theta",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        183,
+                                        617,
+                                        187,
+                                        629
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        }
+                    ],
+                    "index": 35.5
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        194,
+                        643,
+                        417,
+                        658
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                194,
+                                643,
+                                417,
+                                658
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        194,
+                                        643,
+                                        417,
+                                        658
+                                    ],
+                                    "score": 0.88,
+                                    "content": "\\Delta L = L ( \\theta + \\Delta \\theta ) - L ( \\theta ) = \\nabla _ { \\theta } L ( \\theta ) \\Delta \\theta + R ( | | \\Delta \\theta | | _ { 2 } ^ { 2 } )",
+                                    "type": "interline_equation",
+                                    "image_path": "0ec8b0e97fecff408ba0a2770a49d10b1475a273f85c0d31deec99d36e6e0140.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 39,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                194,
+                                643,
+                                417,
+                                658
+                            ],
+                            "spans": [],
+                            "index": 39
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        665,
+                        505,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                665,
+                                505,
+                                678
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        665,
+                                        505,
+                                        678
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The main goal of dropping connections is to remove parameters with minimal impact on the neu-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                677,
+                                505,
+                                689
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        677,
+                                        431,
+                                        689
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ral network and therefore on its performance. Since removing the connection",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        431,
+                                        677,
+                                        441,
+                                        688
+                                    ],
+                                    "score": 0.86,
+                                    "content": "\\theta _ { i }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        441,
+                                        677,
+                                        505,
+                                        689
+                                    ],
+                                    "score": 1.0,
+                                    "content": "corresponds to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                686,
+                                505,
+                                701
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        686,
+                                        275,
+                                        701
+                                    ],
+                                    "score": 1.0,
+                                    "content": "setting it to zero, it incurs a change of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        275,
+                                        688,
+                                        325,
+                                        699
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\Delta \\theta ~ = ~ - \\theta _ { i }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        325,
+                                        686,
+                                        470,
+                                        701
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in that direction and a change of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        470,
+                                        688,
+                                        505,
+                                        699
+                                    ],
+                                    "score": 0.9,
+                                    "content": "\\begin{array} { r l } { \\Delta L _ { i } } & { { } = } \\end{array}",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                698,
+                                506,
+                                711
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        698,
+                                        195,
+                                        711
+                                    ],
+                                    "score": 0.91,
+                                    "content": "- \\nabla _ { \\boldsymbol { \\theta } _ { i } } \\mathbf { \\bar { L } } ( \\boldsymbol { \\theta } ) \\boldsymbol { \\theta } _ { i } + R ( \\boldsymbol { \\theta } _ { i } ^ { 2 } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        195,
+                                        698,
+                                        506,
+                                        711
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in the loss, where the first term is usually defined as the saliency of a connec-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                709,
+                                505,
+                                722
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        709,
+                                        505,
+                                        722
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tion. Though using saliency to remove connections has been used as a criteria for removing connec-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                720,
+                                505,
+                                733
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        720,
+                                        505,
+                                        733
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tions (Molchanov et al., 2016), it has been shown to produce inferior results compared to magnitude",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        }
+                    ],
+                    "index": 42.5
+                }
+            ],
+            "page_idx": 12,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        300,
+                        751,
+                        311,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                299,
+                                750,
+                                312,
+                                764
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        299,
+                                        750,
+                                        312,
+                                        764
+                                    ],
+                                    "score": 1.0,
+                                    "content": "13",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        107,
+                        27,
+                        308,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                107,
+                                26,
+                                308,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        26,
+                                        308,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Under review as a conference paper at ICLR 2020",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        82,
+                        505,
+                        127
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                81,
+                                505,
+                                96
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        81,
+                                        505,
+                                        96
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Olga Russakovsky, Jia Deng, Hao Su, Jonathan Krause, Sanjeev Satheesh, Sean Ma, Zhiheng",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                94,
+                                505,
+                                106
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        94,
+                                        505,
+                                        106
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Huang, Andrej Karpathy, Aditya Khosla, Michael Bernstein, Alexander C. Berg, and Li Fei-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                104,
+                                505,
+                                118
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        104,
+                                        505,
+                                        118
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Fei. Imagenet large scale visual recognition challenge. International Journal of Computer Vision",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                115,
+                                174,
+                                127
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        115,
+                                        174,
+                                        127
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(IJCV), 2015.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        }
+                    ],
+                    "index": 1.5,
+                    "bbox_fs": [
+                        106,
+                        81,
+                        505,
+                        127
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        134,
+                        504,
+                        168
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                135,
+                                505,
+                                146
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        135,
+                                        505,
+                                        146
+                                    ],
+                                    "score": 1.0,
+                                    "content": "M. Sandler, A. Howard, M. Zhu, A. Zhmoginov, and L. Chen. Mobilenetv2: Inverted residuals and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                145,
+                                505,
+                                158
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        145,
+                                        505,
+                                        158
+                                    ],
+                                    "score": 1.0,
+                                    "content": "linear bottlenecks. In 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                156,
+                                363,
+                                168
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        156,
+                                        363,
+                                        168
+                                    ],
+                                    "score": 1.0,
+                                    "content": "pp. 4510–4520, June 2018. doi: 10.1109/CVPR.2018.00474.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        }
+                    ],
+                    "index": 5,
+                    "bbox_fs": [
+                        106,
+                        135,
+                        505,
+                        168
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        175,
+                        506,
+                        209
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                175,
+                                506,
+                                189
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        175,
+                                        506,
+                                        189
+                                    ],
+                                    "score": 1.0,
+                                    "content": "S. Srinivas, A. Subramanya, and R. V. Babu. Training sparse neural networks. In 2017 IEEE",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                186,
+                                505,
+                                199
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        186,
+                                        505,
+                                        199
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Conference on Computer Vision and Pattern Recognition Workshops (CVPRW), pp. 455–462,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                197,
+                                290,
+                                209
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        197,
+                                        290,
+                                        209
+                                    ],
+                                    "score": 1.0,
+                                    "content": "July 2017. doi: 10.1109/CVPRW.2017.61.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        }
+                    ],
+                    "index": 8,
+                    "bbox_fs": [
+                        105,
+                        175,
+                        506,
+                        209
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        216,
+                        504,
+                        239
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                216,
+                                506,
+                                230
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        216,
+                                        506,
+                                        230
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Nikko Strom. Sparse Connection and Pruning in Large Dynamic Artificial Neural Networks. In ¨",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                227,
+                                210,
+                                240
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        227,
+                                        210,
+                                        240
+                                    ],
+                                    "score": 1.0,
+                                    "content": "EUROSPEECH, 1997.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        }
+                    ],
+                    "index": 10.5,
+                    "bbox_fs": [
+                        105,
+                        216,
+                        506,
+                        240
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        246,
+                        505,
+                        290
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                246,
+                                505,
+                                259
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        246,
+                                        505,
+                                        259
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Christian Szegedy, Vincent Vanhoucke, Sergey Ioffe, Jonathon Shlens, and Zbigniew Wojna. Re-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                258,
+                                505,
+                                270
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        258,
+                                        505,
+                                        270
+                                    ],
+                                    "score": 1.0,
+                                    "content": "thinking the inception architecture for computer vision. In Proceedings of IEEE Conference on",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                268,
+                                504,
+                                281
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        268,
+                                        504,
+                                        281
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Computer Vision and Pattern Recognition,, 2016. URL http://arxiv.org/abs/1512.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                280,
+                                152,
+                                291
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        280,
+                                        152,
+                                        291
+                                    ],
+                                    "score": 1.0,
+                                    "content": "00567.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        }
+                    ],
+                    "index": 13.5,
+                    "bbox_fs": [
+                        106,
+                        246,
+                        505,
+                        291
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        298,
+                        505,
+                        343
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                299,
+                                505,
+                                311
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        299,
+                                        505,
+                                        311
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Enzo Tartaglione, Skjalg Lepsøy, Attilio Fiandrotti, and Gianluca Francini. Learning sparse neural",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                310,
+                                505,
+                                322
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        310,
+                                        505,
+                                        322
+                                    ],
+                                    "score": 1.0,
+                                    "content": "networks via sensitivity-driven regularization. In Proceedings of the 32nd International Confer-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                321,
+                                505,
+                                333
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        321,
+                                        505,
+                                        333
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ence on Neural Information Processing Systems, NIPS’18, pp. 3882–3892, USA, 2018. Curran",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                331,
+                                502,
+                                344
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        331,
+                                        400,
+                                        344
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Associates Inc. URL http://dl.acm.org/citation.cfm?id",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        401,
+                                        333,
+                                        407,
+                                        341
+                                    ],
+                                    "score": 0.27,
+                                    "content": "\\underline { { \\underline { { \\mathbf { \\Pi } } } } } =",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        407,
+                                        331,
+                                        502,
+                                        344
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3327144.3327303.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        }
+                    ],
+                    "index": 17.5,
+                    "bbox_fs": [
+                        106,
+                        299,
+                        505,
+                        344
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        350,
+                        504,
+                        373
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                349,
+                                505,
+                                364
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        349,
+                                        505,
+                                        364
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Georg Thimm and Emile Fiesler. Evaluating pruning methods. In National Chiao-Tung University,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                362,
+                                169,
+                                374
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        362,
+                                        169,
+                                        374
+                                    ],
+                                    "score": 1.0,
+                                    "content": "pp. 2, 1995.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        }
+                    ],
+                    "index": 20.5,
+                    "bbox_fs": [
+                        105,
+                        349,
+                        505,
+                        374
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        380,
+                        505,
+                        437
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                381,
+                                506,
+                                394
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        381,
+                                        506,
+                                        394
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Peiqi Wang, Yu Ji, Chi Hong, Yongqiang Lyu, Dongsheng Wang, and Yuan Xie. Snrram: An efficient",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                115,
+                                392,
+                                506,
+                                405
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        115,
+                                        392,
+                                        506,
+                                        405
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sparse neural network computation architecture based on resistive random-access memory. In",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                403,
+                                505,
+                                416
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        403,
+                                        505,
+                                        416
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Proceedings of the 55th Annual Design Automation Conference, DAC ’18, pp. 106:1–106:6, New",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                117,
+                                414,
+                                505,
+                                426
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        117,
+                                        414,
+                                        505,
+                                        426
+                                    ],
+                                    "score": 1.0,
+                                    "content": "York, NY, USA, 2018. ACM. ISBN 978-1-4503-5700-5. doi: 10.1145/3195970.3196116. URL",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                425,
+                                373,
+                                437
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        425,
+                                        373,
+                                        437
+                                    ],
+                                    "score": 1.0,
+                                    "content": "http://doi.acm.org/10.1145/3195970.3196116.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        }
+                    ],
+                    "index": 24,
+                    "bbox_fs": [
+                        105,
+                        381,
+                        506,
+                        437
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        443,
+                        444,
+                        455
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                442,
+                                447,
+                                457
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        442,
+                                        447,
+                                        457
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Sergey Zagoruyko and Nikos Komodakis. Wide residual networks. In BMVC, 2016.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        }
+                    ],
+                    "index": 27,
+                    "bbox_fs": [
+                        105,
+                        442,
+                        447,
+                        457
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        462,
+                        502,
+                        485
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                461,
+                                505,
+                                477
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        461,
+                                        505,
+                                        477
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Hattie Zhou, Janice Lan, Rosanne Liu, and Jason Yosinski. Deconstructing Lottery Tickets: Zeros,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                474,
+                                279,
+                                485
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        474,
+                                        279,
+                                        485
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Signs, and the Supermask. ArXiv, 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        }
+                    ],
+                    "index": 28.5,
+                    "bbox_fs": [
+                        105,
+                        461,
+                        505,
+                        485
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        492,
+                        505,
+                        516
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                493,
+                                505,
+                                506
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        493,
+                                        505,
+                                        506
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Michael Zhu and Suyog Gupta. To Prune, or Not to Prune: Exploring the Efficacy of Pruning for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                503,
+                                504,
+                                516
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        503,
+                                        504,
+                                        516
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Model Compression. In International Conference on Learning Representations Workshop, 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        }
+                    ],
+                    "index": 30.5,
+                    "bbox_fs": [
+                        106,
+                        493,
+                        505,
+                        516
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        536,
+                        424,
+                        550
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                536,
+                                425,
+                                551
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        536,
+                                        425,
+                                        551
+                                    ],
+                                    "score": 1.0,
+                                    "content": "A EFFECT OF MASK UPDATES ON THE ENERGY LANDSCAPE",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        }
+                    ],
+                    "index": 32
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        561,
+                        506,
+                        628
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                560,
+                                506,
+                                576
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        560,
+                                        437,
+                                        576
+                                    ],
+                                    "score": 1.0,
+                                    "content": "To update the connectivity of our sparse network, we first need to drop a fraction",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        437,
+                                        563,
+                                        444,
+                                        572
+                                    ],
+                                    "score": 0.78,
+                                    "content": "d",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        444,
+                                        560,
+                                        506,
+                                        576
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of the existing",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                572,
+                                506,
+                                587
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        572,
+                                        506,
+                                        587
+                                    ],
+                                    "score": 1.0,
+                                    "content": "connections for each layer independently to create a budget for growing new connections. Following",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                583,
+                                506,
+                                596
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        583,
+                                        443,
+                                        596
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the recipe of magnitude based pruning(Han et al., 2015), we order parameters at layer",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        443,
+                                        585,
+                                        448,
+                                        594
+                                    ],
+                                    "score": 0.77,
+                                    "content": "i",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        448,
+                                        583,
+                                        506,
+                                        596
+                                    ],
+                                    "score": 1.0,
+                                    "content": "by magnitude",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                107,
+                                594,
+                                506,
+                                608
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        596,
+                                        121,
+                                        607
+                                    ],
+                                    "score": 0.89,
+                                    "content": "| \\theta _ { i } |",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        122,
+                                        594,
+                                        174,
+                                        608
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and drop the",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        174,
+                                        595,
+                                        235,
+                                        606
+                                    ],
+                                    "score": 0.92,
+                                    "content": "N * ( 1 - S ) * d",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        235,
+                                        594,
+                                        506,
+                                        608
+                                    ],
+                                    "score": 1.0,
+                                    "content": "parameters with lowest magnitude. The effectiveness of this simple",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                606,
+                                506,
+                                618
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        606,
+                                        421,
+                                        618
+                                    ],
+                                    "score": 1.0,
+                                    "content": "criteria can be explained through the first order Taylor approximation of the loss",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        422,
+                                        606,
+                                        430,
+                                        615
+                                    ],
+                                    "score": 0.83,
+                                    "content": "L",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        430,
+                                        606,
+                                        506,
+                                        618
+                                    ],
+                                    "score": 1.0,
+                                    "content": "around the current",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                617,
+                                187,
+                                629
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        617,
+                                        177,
+                                        629
+                                    ],
+                                    "score": 1.0,
+                                    "content": "set of parameters",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        177,
+                                        618,
+                                        183,
+                                        626
+                                    ],
+                                    "score": 0.81,
+                                    "content": "\\theta",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        183,
+                                        617,
+                                        187,
+                                        629
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        }
+                    ],
+                    "index": 35.5,
+                    "bbox_fs": [
+                        104,
+                        560,
+                        506,
+                        629
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        194,
+                        643,
+                        417,
+                        658
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                194,
+                                643,
+                                417,
+                                658
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        194,
+                                        643,
+                                        417,
+                                        658
+                                    ],
+                                    "score": 0.88,
+                                    "content": "\\Delta L = L ( \\theta + \\Delta \\theta ) - L ( \\theta ) = \\nabla _ { \\theta } L ( \\theta ) \\Delta \\theta + R ( | | \\Delta \\theta | | _ { 2 } ^ { 2 } )",
+                                    "type": "interline_equation",
+                                    "image_path": "0ec8b0e97fecff408ba0a2770a49d10b1475a273f85c0d31deec99d36e6e0140.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 39,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                194,
+                                643,
+                                417,
+                                658
+                            ],
+                            "spans": [],
+                            "index": 39
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        665,
+                        505,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                665,
+                                505,
+                                678
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        665,
+                                        505,
+                                        678
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The main goal of dropping connections is to remove parameters with minimal impact on the neu-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                677,
+                                505,
+                                689
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        677,
+                                        431,
+                                        689
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ral network and therefore on its performance. Since removing the connection",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        431,
+                                        677,
+                                        441,
+                                        688
+                                    ],
+                                    "score": 0.86,
+                                    "content": "\\theta _ { i }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        441,
+                                        677,
+                                        505,
+                                        689
+                                    ],
+                                    "score": 1.0,
+                                    "content": "corresponds to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                686,
+                                505,
+                                701
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        686,
+                                        275,
+                                        701
+                                    ],
+                                    "score": 1.0,
+                                    "content": "setting it to zero, it incurs a change of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        275,
+                                        688,
+                                        325,
+                                        699
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\Delta \\theta ~ = ~ - \\theta _ { i }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        325,
+                                        686,
+                                        470,
+                                        701
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in that direction and a change of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        470,
+                                        688,
+                                        505,
+                                        699
+                                    ],
+                                    "score": 0.9,
+                                    "content": "\\begin{array} { r l } { \\Delta L _ { i } } & { { } = } \\end{array}",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                698,
+                                506,
+                                711
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        698,
+                                        195,
+                                        711
+                                    ],
+                                    "score": 0.91,
+                                    "content": "- \\nabla _ { \\boldsymbol { \\theta } _ { i } } \\mathbf { \\bar { L } } ( \\boldsymbol { \\theta } ) \\boldsymbol { \\theta } _ { i } + R ( \\boldsymbol { \\theta } _ { i } ^ { 2 } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        195,
+                                        698,
+                                        506,
+                                        711
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in the loss, where the first term is usually defined as the saliency of a connec-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                709,
+                                505,
+                                722
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        709,
+                                        505,
+                                        722
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tion. Though using saliency to remove connections has been used as a criteria for removing connec-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                720,
+                                505,
+                                733
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        720,
+                                        505,
+                                        733
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tions (Molchanov et al., 2016), it has been shown to produce inferior results compared to magnitude",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                83,
+                                505,
+                                94
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        83,
+                                        505,
+                                        94
+                                    ],
+                                    "score": 1.0,
+                                    "content": "based removal, especially when used to remove multiple connections at once (Evci, 2018). In con-",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                94,
+                                505,
+                                106
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        94,
+                                        505,
+                                        106
+                                    ],
+                                    "score": 1.0,
+                                    "content": "trast, picking the lowest magnitude connections ensures a small remainder term in addition to a low",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                104,
+                                506,
+                                117
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        104,
+                                        506,
+                                        117
+                                    ],
+                                    "score": 1.0,
+                                    "content": "saliency, limiting the damage we make when we drop connections. Additionally, we note that con-",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                116,
+                                505,
+                                127
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        116,
+                                        505,
+                                        127
+                                    ],
+                                    "score": 1.0,
+                                    "content": "nections with small magnitude can only remain small if the gradient is also small, meaning that the",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                126,
+                                506,
+                                139
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        126,
+                                        506,
+                                        139
+                                    ],
+                                    "score": 1.0,
+                                    "content": "saliency is likely small when the parameter itself is small. Therefore we argue that the connections",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                137,
+                                408,
+                                150
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        137,
+                                        408,
+                                        150
+                                    ],
+                                    "score": 1.0,
+                                    "content": "removed by RigL are likely to be saddle directions of the energy landscape.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 5
+                        }
+                    ],
+                    "index": 42.5,
+                    "bbox_fs": [
+                        105,
+                        665,
+                        506,
+                        733
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        82,
+                        504,
+                        149
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                83,
+                                505,
+                                94
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        83,
+                                        505,
+                                        94
+                                    ],
+                                    "score": 1.0,
+                                    "content": "based removal, especially when used to remove multiple connections at once (Evci, 2018). In con-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                94,
+                                505,
+                                106
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        94,
+                                        505,
+                                        106
+                                    ],
+                                    "score": 1.0,
+                                    "content": "trast, picking the lowest magnitude connections ensures a small remainder term in addition to a low",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                104,
+                                506,
+                                117
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        104,
+                                        506,
+                                        117
+                                    ],
+                                    "score": 1.0,
+                                    "content": "saliency, limiting the damage we make when we drop connections. Additionally, we note that con-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                116,
+                                505,
+                                127
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        116,
+                                        505,
+                                        127
+                                    ],
+                                    "score": 1.0,
+                                    "content": "nections with small magnitude can only remain small if the gradient is also small, meaning that the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                126,
+                                506,
+                                139
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        126,
+                                        506,
+                                        139
+                                    ],
+                                    "score": 1.0,
+                                    "content": "saliency is likely small when the parameter itself is small. Therefore we argue that the connections",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                137,
+                                408,
+                                150
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        137,
+                                        408,
+                                        150
+                                    ],
+                                    "score": 1.0,
+                                    "content": "removed by RigL are likely to be saddle directions of the energy landscape.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        }
+                    ],
+                    "index": 2.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        154,
+                        505,
+                        243
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                154,
+                                506,
+                                166
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        154,
+                                        506,
+                                        166
+                                    ],
+                                    "score": 1.0,
+                                    "content": "After the removal of insignificant connections, we enable new connections that have the highest",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                166,
+                                506,
+                                178
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        166,
+                                        506,
+                                        178
+                                    ],
+                                    "score": 1.0,
+                                    "content": "expected gradients. Since we initialize these new connections to zero, they are guaranteed to have",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                176,
+                                506,
+                                188
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        176,
+                                        506,
+                                        188
+                                    ],
+                                    "score": 1.0,
+                                    "content": "high gradients in the proceeding iteration and therefore to reduce the loss quickly. By definition a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                187,
+                                505,
+                                200
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        187,
+                                        505,
+                                        200
+                                    ],
+                                    "score": 1.0,
+                                    "content": "direction with high magnitude gradient is not a saddle direction. Combining this observation with",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                198,
+                                506,
+                                211
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        198,
+                                        162,
+                                        211
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the previous (",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        162,
+                                        199,
+                                        183,
+                                        209
+                                    ],
+                                    "score": 0.36,
+                                    "content": "R i g L",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        184,
+                                        198,
+                                        506,
+                                        211
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is likely to remove saddle directions) and the results in Section 4.4 we suggest",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                209,
+                                505,
+                                222
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        209,
+                                        505,
+                                        222
+                                    ],
+                                    "score": 1.0,
+                                    "content": "that RigL improves the energy landscape of the optimization by replacing saddle directions with the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                220,
+                                506,
+                                233
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        220,
+                                        506,
+                                        233
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ones with high gradient. This helps the optimization procedure to escape bad critical points and find",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                231,
+                                225,
+                                244
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        231,
+                                        225,
+                                        244
+                                    ],
+                                    "score": 1.0,
+                                    "content": "solutions with higher quality.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        }
+                    ],
+                    "index": 9.5
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        106,
+                        259,
+                        428,
+                        272
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                259,
+                                429,
+                                273
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        259,
+                                        429,
+                                        273
+                                    ],
+                                    "score": 1.0,
+                                    "content": "B EFFECT OF SPARSITY DISTRIBUTION ON OTHER METHODS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        }
+                    ],
+                    "index": 14
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        285,
+                        505,
+                        307
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                284,
+                                506,
+                                299
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        284,
+                                        506,
+                                        299
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In Figure 7-left we show the effect of sparsity distribution choice on 4 different sparse training",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                296,
+                                478,
+                                308
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        296,
+                                        478,
+                                        308
+                                    ],
+                                    "score": 1.0,
+                                    "content": "methods. ERK distribution performs better than other distributions for each training method.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        }
+                    ],
+                    "index": 15.5
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        106,
+                        324,
+                        380,
+                        337
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                323,
+                                380,
+                                339
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        323,
+                                        380,
+                                        339
+                                    ],
+                                    "score": 1.0,
+                                    "content": "C EFFECT OF MOMENTUM COEFFICIENT FOR SNFS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        }
+                    ],
+                    "index": 17
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        350,
+                        505,
+                        405
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                351,
+                                505,
+                                362
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        351,
+                                        505,
+                                        362
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In Figure 7 right we show the effect of the momentum coefficient on the performance of SNFS. Our",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                360,
+                                505,
+                                374
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        360,
+                                        505,
+                                        374
+                                    ],
+                                    "score": 1.0,
+                                    "content": "results shows that using a coefficient of 0.99 brings the best performance. On the other hand using",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                372,
+                                506,
+                                385
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        372,
+                                        506,
+                                        385
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the most recent gradient only (coefficient of 0) performs as good as using a coefficient of 0.9. This",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                383,
+                                505,
+                                396
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        383,
+                                        505,
+                                        396
+                                    ],
+                                    "score": 1.0,
+                                    "content": "result might be due to the large batch size we are using (4096), but it still motivates using RigL and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                395,
+                                447,
+                                407
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        395,
+                                        447,
+                                        407
+                                    ],
+                                    "score": 1.0,
+                                    "content": "instantaneous gradient information only when needed, instead of accumulating them.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        }
+                    ],
+                    "index": 20
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        423,
+                        303,
+                        435
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                422,
+                                304,
+                                437
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        422,
+                                        304,
+                                        437
+                                    ],
+                                    "score": 1.0,
+                                    "content": "D EXISTENCE OF LOTTERY TICKETS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        }
+                    ],
+                    "index": 23
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        448,
+                        505,
+                        525
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                447,
+                                506,
+                                461
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        447,
+                                        506,
+                                        461
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We perform the following experiment to see whether Lottery Tickets exist in our setting. We take",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                459,
+                                505,
+                                472
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        459,
+                                        505,
+                                        472
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the sparse network found by RigL and restart training using original initialization, both with RigL",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                470,
+                                505,
+                                483
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        470,
+                                        505,
+                                        483
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and with fixed topology as in the original Lottery Ticket Hypothesis. Results in table 3 demonstrate",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                481,
+                                505,
+                                493
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        481,
+                                        505,
+                                        493
+                                    ],
+                                    "score": 1.0,
+                                    "content": "that training with a fixed topology is significantly worse than training with RigL and that RigL does",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                492,
+                                506,
+                                505
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        492,
+                                        506,
+                                        505
+                                    ],
+                                    "score": 1.0,
+                                    "content": "not benefit from starting again with the final topology and the original initialization - training for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                503,
+                                505,
+                                515
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        503,
+                                        505,
+                                        515
+                                    ],
+                                    "score": 1.0,
+                                    "content": "twice as long instead of rewiring is more effective. In short, there are no special tickets, with RigL",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                514,
+                                204,
+                                526
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        514,
+                                        204,
+                                        526
+                                    ],
+                                    "score": 1.0,
+                                    "content": "all tickets seems to win.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        }
+                    ],
+                    "index": 27
+                },
+                {
+                    "type": "image",
+                    "bbox": [
+                        112,
+                        543,
+                        495,
+                        688
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                112,
+                                543,
+                                495,
+                                688
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        112,
+                                        543,
+                                        495,
+                                        688
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                112,
+                                                543,
+                                                495,
+                                                688
+                                            ],
+                                            "score": 0.969,
+                                            "type": "image",
+                                            "image_path": "f433bd443059bd764822e9e7a3060af7bbc37812c757ca9ad010a470abc0a839.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 32,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        112,
+                                        543,
+                                        495,
+                                        591.3333333333334
+                                    ],
+                                    "spans": [],
+                                    "index": 31
+                                },
+                                {
+                                    "bbox": [
+                                        112,
+                                        591.3333333333334,
+                                        495,
+                                        639.6666666666667
+                                    ],
+                                    "spans": [],
+                                    "index": 32
+                                },
+                                {
+                                    "bbox": [
+                                        112,
+                                        639.6666666666667,
+                                        495,
+                                        688.0000000000001
+                                    ],
+                                    "spans": [],
+                                    "index": 33
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                106,
+                                698,
+                                505,
+                                743
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        698,
+                                        505,
+                                        712
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                698,
+                                                505,
+                                                712
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 7: (left) Effect of sparsity distribution choice on sparse training methods at different sparsity",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 34
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        709,
+                                        505,
+                                        722
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                709,
+                                                505,
+                                                722
+                                            ],
+                                            "score": 1.0,
+                                            "content": "levels. We average over 3 runs and report the standard deviations for each. (right) Effect of momen-",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 35
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        721,
+                                        505,
+                                        732
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                721,
+                                                505,
+                                                732
+                                            ],
+                                            "score": 1.0,
+                                            "content": "tum value on the performance of SNFS algorithm. Setting the momentum coefficient of the SNFS",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 36
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        732,
+                                        472,
+                                        745
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                732,
+                                                472,
+                                                745
+                                            ],
+                                            "score": 1.0,
+                                            "content": "algorithm to 0 seems to perform best, suggesting the accumulated values are not important.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 37
+                                }
+                            ],
+                            "index": 35.5
+                        }
+                    ],
+                    "index": 33.75
+                }
+            ],
+            "page_idx": 13,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        107,
+                        27,
+                        308,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                107,
+                                26,
+                                308,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        26,
+                                        308,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Under review as a conference paper at ICLR 2020",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        300,
+                        751,
+                        311,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                299,
+                                750,
+                                313,
+                                764
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        299,
+                                        750,
+                                        313,
+                                        764
+                                    ],
+                                    "score": 1.0,
+                                    "content": "14",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        82,
+                        504,
+                        149
+                    ],
+                    "lines": [],
+                    "index": 2.5,
+                    "bbox_fs": [
+                        105,
+                        83,
+                        506,
+                        150
+                    ],
+                    "lines_deleted": true
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        154,
+                        505,
+                        243
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                154,
+                                506,
+                                166
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        154,
+                                        506,
+                                        166
+                                    ],
+                                    "score": 1.0,
+                                    "content": "After the removal of insignificant connections, we enable new connections that have the highest",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                166,
+                                506,
+                                178
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        166,
+                                        506,
+                                        178
+                                    ],
+                                    "score": 1.0,
+                                    "content": "expected gradients. Since we initialize these new connections to zero, they are guaranteed to have",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                176,
+                                506,
+                                188
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        176,
+                                        506,
+                                        188
+                                    ],
+                                    "score": 1.0,
+                                    "content": "high gradients in the proceeding iteration and therefore to reduce the loss quickly. By definition a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                187,
+                                505,
+                                200
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        187,
+                                        505,
+                                        200
+                                    ],
+                                    "score": 1.0,
+                                    "content": "direction with high magnitude gradient is not a saddle direction. Combining this observation with",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                198,
+                                506,
+                                211
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        198,
+                                        162,
+                                        211
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the previous (",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        162,
+                                        199,
+                                        183,
+                                        209
+                                    ],
+                                    "score": 0.36,
+                                    "content": "R i g L",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        184,
+                                        198,
+                                        506,
+                                        211
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is likely to remove saddle directions) and the results in Section 4.4 we suggest",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                209,
+                                505,
+                                222
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        209,
+                                        505,
+                                        222
+                                    ],
+                                    "score": 1.0,
+                                    "content": "that RigL improves the energy landscape of the optimization by replacing saddle directions with the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                220,
+                                506,
+                                233
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        220,
+                                        506,
+                                        233
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ones with high gradient. This helps the optimization procedure to escape bad critical points and find",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                231,
+                                225,
+                                244
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        231,
+                                        225,
+                                        244
+                                    ],
+                                    "score": 1.0,
+                                    "content": "solutions with higher quality.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        }
+                    ],
+                    "index": 9.5,
+                    "bbox_fs": [
+                        105,
+                        154,
+                        506,
+                        244
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        106,
+                        259,
+                        428,
+                        272
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                259,
+                                429,
+                                273
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        259,
+                                        429,
+                                        273
+                                    ],
+                                    "score": 1.0,
+                                    "content": "B EFFECT OF SPARSITY DISTRIBUTION ON OTHER METHODS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        }
+                    ],
+                    "index": 14
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        285,
+                        505,
+                        307
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                284,
+                                506,
+                                299
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        284,
+                                        506,
+                                        299
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In Figure 7-left we show the effect of sparsity distribution choice on 4 different sparse training",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                296,
+                                478,
+                                308
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        296,
+                                        478,
+                                        308
+                                    ],
+                                    "score": 1.0,
+                                    "content": "methods. ERK distribution performs better than other distributions for each training method.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        }
+                    ],
+                    "index": 15.5,
+                    "bbox_fs": [
+                        104,
+                        284,
+                        506,
+                        308
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        106,
+                        324,
+                        380,
+                        337
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                323,
+                                380,
+                                339
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        323,
+                                        380,
+                                        339
+                                    ],
+                                    "score": 1.0,
+                                    "content": "C EFFECT OF MOMENTUM COEFFICIENT FOR SNFS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        }
+                    ],
+                    "index": 17
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        350,
+                        505,
+                        405
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                351,
+                                505,
+                                362
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        351,
+                                        505,
+                                        362
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In Figure 7 right we show the effect of the momentum coefficient on the performance of SNFS. Our",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                360,
+                                505,
+                                374
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        360,
+                                        505,
+                                        374
+                                    ],
+                                    "score": 1.0,
+                                    "content": "results shows that using a coefficient of 0.99 brings the best performance. On the other hand using",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                372,
+                                506,
+                                385
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        372,
+                                        506,
+                                        385
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the most recent gradient only (coefficient of 0) performs as good as using a coefficient of 0.9. This",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                383,
+                                505,
+                                396
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        383,
+                                        505,
+                                        396
+                                    ],
+                                    "score": 1.0,
+                                    "content": "result might be due to the large batch size we are using (4096), but it still motivates using RigL and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                395,
+                                447,
+                                407
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        395,
+                                        447,
+                                        407
+                                    ],
+                                    "score": 1.0,
+                                    "content": "instantaneous gradient information only when needed, instead of accumulating them.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        }
+                    ],
+                    "index": 20,
+                    "bbox_fs": [
+                        105,
+                        351,
+                        506,
+                        407
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        423,
+                        303,
+                        435
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                422,
+                                304,
+                                437
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        422,
+                                        304,
+                                        437
+                                    ],
+                                    "score": 1.0,
+                                    "content": "D EXISTENCE OF LOTTERY TICKETS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        }
+                    ],
+                    "index": 23
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        448,
+                        505,
+                        525
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                447,
+                                506,
+                                461
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        447,
+                                        506,
+                                        461
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We perform the following experiment to see whether Lottery Tickets exist in our setting. We take",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                459,
+                                505,
+                                472
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        459,
+                                        505,
+                                        472
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the sparse network found by RigL and restart training using original initialization, both with RigL",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                470,
+                                505,
+                                483
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        470,
+                                        505,
+                                        483
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and with fixed topology as in the original Lottery Ticket Hypothesis. Results in table 3 demonstrate",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                481,
+                                505,
+                                493
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        481,
+                                        505,
+                                        493
+                                    ],
+                                    "score": 1.0,
+                                    "content": "that training with a fixed topology is significantly worse than training with RigL and that RigL does",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                492,
+                                506,
+                                505
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        492,
+                                        506,
+                                        505
+                                    ],
+                                    "score": 1.0,
+                                    "content": "not benefit from starting again with the final topology and the original initialization - training for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                503,
+                                505,
+                                515
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        503,
+                                        505,
+                                        515
+                                    ],
+                                    "score": 1.0,
+                                    "content": "twice as long instead of rewiring is more effective. In short, there are no special tickets, with RigL",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                514,
+                                204,
+                                526
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        514,
+                                        204,
+                                        526
+                                    ],
+                                    "score": 1.0,
+                                    "content": "all tickets seems to win.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        }
+                    ],
+                    "index": 27,
+                    "bbox_fs": [
+                        105,
+                        447,
+                        506,
+                        526
+                    ]
+                },
+                {
+                    "type": "image",
+                    "bbox": [
+                        112,
+                        543,
+                        495,
+                        688
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                112,
+                                543,
+                                495,
+                                688
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        112,
+                                        543,
+                                        495,
+                                        688
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                112,
+                                                543,
+                                                495,
+                                                688
+                                            ],
+                                            "score": 0.969,
+                                            "type": "image",
+                                            "image_path": "f433bd443059bd764822e9e7a3060af7bbc37812c757ca9ad010a470abc0a839.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 32,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        112,
+                                        543,
+                                        495,
+                                        591.3333333333334
+                                    ],
+                                    "spans": [],
+                                    "index": 31
+                                },
+                                {
+                                    "bbox": [
+                                        112,
+                                        591.3333333333334,
+                                        495,
+                                        639.6666666666667
+                                    ],
+                                    "spans": [],
+                                    "index": 32
+                                },
+                                {
+                                    "bbox": [
+                                        112,
+                                        639.6666666666667,
+                                        495,
+                                        688.0000000000001
+                                    ],
+                                    "spans": [],
+                                    "index": 33
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                106,
+                                698,
+                                505,
+                                743
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        698,
+                                        505,
+                                        712
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                698,
+                                                505,
+                                                712
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 7: (left) Effect of sparsity distribution choice on sparse training methods at different sparsity",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 34
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        709,
+                                        505,
+                                        722
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                709,
+                                                505,
+                                                722
+                                            ],
+                                            "score": 1.0,
+                                            "content": "levels. We average over 3 runs and report the standard deviations for each. (right) Effect of momen-",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 35
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        721,
+                                        505,
+                                        732
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                721,
+                                                505,
+                                                732
+                                            ],
+                                            "score": 1.0,
+                                            "content": "tum value on the performance of SNFS algorithm. Setting the momentum coefficient of the SNFS",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 36
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        732,
+                                        472,
+                                        745
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                732,
+                                                472,
+                                                745
+                                            ],
+                                            "score": 1.0,
+                                            "content": "algorithm to 0 seems to perform best, suggesting the accumulated values are not important.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 37
+                                }
+                            ],
+                            "index": 35.5
+                        }
+                    ],
+                    "index": 33.75
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "table",
+                    "bbox": [
+                        160,
+                        79,
+                        452,
+                        138
+                    ],
+                    "blocks": [
+                        {
+                            "type": "table_body",
+                            "bbox": [
+                                160,
+                                79,
+                                452,
+                                138
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        160,
+                                        79,
+                                        452,
+                                        138
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                160,
+                                                79,
+                                                452,
+                                                138
+                                            ],
+                                            "score": 0.978,
+                                            "html": "<table><tr><td>Initialization</td><td>Training Method</td><td>Test Accuracy</td><td>Training FLOPs</td></tr><tr><td>Lottery</td><td>Static</td><td>70.82±0.07</td><td>0.46x</td></tr><tr><td>Lottery</td><td>RigL</td><td>73.93±0.09</td><td>0.46x</td></tr><tr><td>Random</td><td>RigL</td><td>74.55±0.06</td><td>0.23x</td></tr><tr><td>Random</td><td>RigL2x</td><td>76.06±0.09</td><td>0.46x</td></tr></table>",
+                                            "type": "table",
+                                            "image_path": "94d62b6ac1e54ad04744876e63f71ea149d6ef8741dcfc583da03af321e74f12.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 1,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        160,
+                                        79,
+                                        452,
+                                        98.66666666666667
+                                    ],
+                                    "spans": [],
+                                    "index": 0
+                                },
+                                {
+                                    "bbox": [
+                                        160,
+                                        98.66666666666667,
+                                        452,
+                                        118.33333333333334
+                                    ],
+                                    "spans": [],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        160,
+                                        118.33333333333334,
+                                        452,
+                                        138.0
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                }
+                            ]
+                        },
+                        {
+                            "type": "table_caption",
+                            "bbox": [
+                                105,
+                                145,
+                                502,
+                                168
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        144,
+                                        505,
+                                        158
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                144,
+                                                505,
+                                                158
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Table 3: Effect of lottery ticket initialization on the final performance. There are no special tickets",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 3
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        156,
+                                        415,
+                                        169
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                156,
+                                                415,
+                                                169
+                                            ],
+                                            "score": 1.0,
+                                            "content": "and dynamic connectivity provided by RigL is critical for good performance.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 4
+                                }
+                            ],
+                            "index": 3.5
+                        }
+                    ],
+                    "index": 2.25
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        105,
+                        187,
+                        501,
+                        201
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                187,
+                                502,
+                                202
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        187,
+                                        502,
+                                        202
+                                    ],
+                                    "score": 1.0,
+                                    "content": "E EFFECT OF UPDATE SCHEDULES ON OTHER DYNAMIC SPARSE METHODS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        }
+                    ],
+                    "index": 5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        212,
+                        505,
+                        257
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                212,
+                                506,
+                                226
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        212,
+                                        506,
+                                        226
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In Figure 8 we repeat the hyper-parameter sweep done for RigL in Figure 5-right, using SET and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                224,
+                                505,
+                                236
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        224,
+                                        226,
+                                        236
+                                    ],
+                                    "score": 1.0,
+                                    "content": "SNFS. Cosine schedule with",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        226,
+                                        224,
+                                        269,
+                                        234
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\Delta T = 5 0",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        269,
+                                        224,
+                                        288,
+                                        236
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        288,
+                                        225,
+                                        325,
+                                        235
+                                    ],
+                                    "score": 0.9,
+                                    "content": "\\alpha = 0 . 1",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        325,
+                                        224,
+                                        505,
+                                        236
+                                    ],
+                                    "score": 1.0,
+                                    "content": "seems to work best across all methods. An",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                235,
+                                505,
+                                247
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        235,
+                                        318,
+                                        247
+                                    ],
+                                    "score": 1.0,
+                                    "content": "interesting observation is that higher drop fractions",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        318,
+                                        236,
+                                        332,
+                                        246
+                                    ],
+                                    "score": 0.74,
+                                    "content": "( \\alpha )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        333,
+                                        235,
+                                        505,
+                                        247
+                                    ],
+                                    "score": 1.0,
+                                    "content": "seem to work better with longer intervals",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                107,
+                                246,
+                                410,
+                                258
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        246,
+                                        123,
+                                        256
+                                    ],
+                                    "score": 0.75,
+                                    "content": "\\Delta T",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        124,
+                                        246,
+                                        222,
+                                        258
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". For example, SET with",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        223,
+                                        246,
+                                        273,
+                                        256
+                                    ],
+                                    "score": 0.9,
+                                    "content": "\\Delta T = 1 0 0 0",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        273,
+                                        246,
+                                        372,
+                                        258
+                                    ],
+                                    "score": 1.0,
+                                    "content": "seems to work best with",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        372,
+                                        246,
+                                        406,
+                                        256
+                                    ],
+                                    "score": 0.89,
+                                    "content": "\\alpha = 0 . 5",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        407,
+                                        246,
+                                        410,
+                                        258
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        }
+                    ],
+                    "index": 7.5
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        106,
+                        273,
+                        309,
+                        286
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                272,
+                                310,
+                                288
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        272,
+                                        310,
+                                        288
+                                    ],
+                                    "score": 1.0,
+                                    "content": "F ALTERNATIVE UPDATE SCHEDULES",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        }
+                    ],
+                    "index": 10
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        298,
+                        416,
+                        310
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                297,
+                                417,
+                                312
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        297,
+                                        417,
+                                        312
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In Figure 9, we share the performance of two alternative annealing functions:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        }
+                    ],
+                    "index": 11
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        130,
+                        318,
+                        505,
+                        369
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                129,
+                                317,
+                                246,
+                                333
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        129,
+                                        317,
+                                        185,
+                                        333
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1. Constant:",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        185,
+                                        319,
+                                        243,
+                                        332
+                                    ],
+                                    "score": 0.92,
+                                    "content": "f _ { d e c a y } ( t ) = \\alpha",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        243,
+                                        317,
+                                        246,
+                                        333
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                129,
+                                333,
+                                505,
+                                346
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        129,
+                                        333,
+                                        505,
+                                        346
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2. Inverse Power: The fraction of weights updated decreases similarly to the schedule used",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                343,
+                                506,
+                                359
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        344,
+                                        320,
+                                        358
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in Zhu & Gupta (2018) for iterative pruning:",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        321,
+                                        344,
+                                        424,
+                                        358
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\begin{array} { r } { f _ { d e c a y } ( t ) = \\alpha ( 1 - \\frac { t } { T _ { e n d } } ) ^ { k } } \\end{array}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        413,
+                                        343,
+                                        506,
+                                        359
+                                    ],
+                                    "score": 1.0,
+                                    "content": ")k. In our experiments",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                142,
+                                358,
+                                402,
+                                369
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        142,
+                                        358,
+                                        177,
+                                        369
+                                    ],
+                                    "score": 1.0,
+                                    "content": "we tried",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        177,
+                                        358,
+                                        202,
+                                        367
+                                    ],
+                                    "score": 0.9,
+                                    "content": "k = 1",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        203,
+                                        358,
+                                        372,
+                                        369
+                                    ],
+                                    "score": 1.0,
+                                    "content": "which is the linear decay and their default",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        373,
+                                        358,
+                                        398,
+                                        367
+                                    ],
+                                    "score": 0.87,
+                                    "content": "k = 3",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        398,
+                                        358,
+                                        402,
+                                        369
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        }
+                    ],
+                    "index": 13.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        378,
+                        505,
+                        433
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                378,
+                                505,
+                                391
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        378,
+                                        375,
+                                        391
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Constant seems to perform well with low initial drop fractions like",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        376,
+                                        379,
+                                        409,
+                                        389
+                                    ],
+                                    "score": 0.89,
+                                    "content": "\\alpha = 0 . 1",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        410,
+                                        378,
+                                        505,
+                                        391
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", but it starts to perform",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                389,
+                                505,
+                                402
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        389,
+                                        195,
+                                        402
+                                    ],
+                                    "score": 1.0,
+                                    "content": "worse with increasing",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        195,
+                                        391,
+                                        203,
+                                        399
+                                    ],
+                                    "score": 0.62,
+                                    "content": "\\alpha",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        203,
+                                        389,
+                                        278,
+                                        402
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Inverse Power for",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        279,
+                                        390,
+                                        296,
+                                        399
+                                    ],
+                                    "score": 0.79,
+                                    "content": "{ \\bf k } = 3",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        296,
+                                        389,
+                                        312,
+                                        402
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        313,
+                                        390,
+                                        330,
+                                        400
+                                    ],
+                                    "score": 0.8,
+                                    "content": "{ \\mathrm { k } } { = } 1",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        330,
+                                        389,
+                                        505,
+                                        402
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(Linear) seems to perform similarly for low",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                107,
+                                399,
+                                504,
+                                412
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        402,
+                                        114,
+                                        410
+                                    ],
+                                    "score": 0.72,
+                                    "content": "\\alpha",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        115,
+                                        399,
+                                        334,
+                                        412
+                                    ],
+                                    "score": 1.0,
+                                    "content": "values. However the performance drops noticeably for",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        334,
+                                        401,
+                                        352,
+                                        411
+                                    ],
+                                    "score": 0.75,
+                                    "content": "{ \\bf k } = 3",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        352,
+                                        399,
+                                        504,
+                                        412
+                                    ],
+                                    "score": 1.0,
+                                    "content": "when we increase the update interval.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                411,
+                                505,
+                                424
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        411,
+                                        326,
+                                        424
+                                    ],
+                                    "score": 1.0,
+                                    "content": "As reported by Dettmers & Zettlemoyer (2019) linear",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        327,
+                                        412,
+                                        345,
+                                        422
+                                    ],
+                                    "score": 0.73,
+                                    "content": "( \\mathrm { k } { = } 1",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        345,
+                                        411,
+                                        505,
+                                        424
+                                    ],
+                                    "score": 1.0,
+                                    "content": ") seems to provide similar results as the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                422,
+                                173,
+                                434
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        422,
+                                        173,
+                                        434
+                                    ],
+                                    "score": 1.0,
+                                    "content": "cosine schedule.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        }
+                    ],
+                    "index": 18
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        450,
+                        386,
+                        462
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                448,
+                                387,
+                                464
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        448,
+                                        387,
+                                        464
+                                    ],
+                                    "score": 1.0,
+                                    "content": "G CALCULATING FLOPS OF MODELS AND METHODS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        }
+                    ],
+                    "index": 21
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        474,
+                        505,
+                        508
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                474,
+                                505,
+                                487
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        474,
+                                        505,
+                                        487
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In order to calculate FLOPs needed for a single forward pass of a sparse model, we count the total",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                485,
+                                505,
+                                498
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        486,
+                                        423,
+                                        498
+                                    ],
+                                    "score": 1.0,
+                                    "content": "number of multiplications and additions layer by layer for a given layer sparsity",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        424,
+                                        485,
+                                        432,
+                                        496
+                                    ],
+                                    "score": 0.86,
+                                    "content": "s ^ { l }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        433,
+                                        486,
+                                        505,
+                                        498
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". The total FLOPs",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                497,
+                                360,
+                                509
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        497,
+                                        360,
+                                        509
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is then obtained by summing up all of these multiply and adds.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        }
+                    ],
+                    "index": 23
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        513,
+                        505,
+                        547
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                513,
+                                505,
+                                527
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        513,
+                                        505,
+                                        527
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Different sparsity distributions require different number of FLOPs to compute a single prediction.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                524,
+                                505,
+                                537
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        524,
+                                        505,
+                                        537
+                                    ],
+                                    "score": 1.0,
+                                    "content": "For example Erdos-Renyi-Kernel ˝ distributions usually cause earlier layers to be less sparse than",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                535,
+                                505,
+                                549
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        535,
+                                        505,
+                                        549
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the later layers (see Appendix I). The inputs of earlier layers have greater spatial dimensions, so",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        }
+                    ],
+                    "index": 26
+                },
+                {
+                    "type": "image",
+                    "bbox": [
+                        112,
+                        566,
+                        493,
+                        709
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                112,
+                                566,
+                                493,
+                                709
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        112,
+                                        566,
+                                        493,
+                                        709
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                112,
+                                                566,
+                                                493,
+                                                709
+                                            ],
+                                            "score": 0.966,
+                                            "type": "image",
+                                            "image_path": "b3e467a52c2f5b11283887318ea99ba50224d604ca1a45394151b61438546ee9.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 29,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        112,
+                                        566,
+                                        493,
+                                        613.6666666666666
+                                    ],
+                                    "spans": [],
+                                    "index": 28
+                                },
+                                {
+                                    "bbox": [
+                                        112,
+                                        613.6666666666666,
+                                        493,
+                                        661.3333333333333
+                                    ],
+                                    "spans": [],
+                                    "index": 29
+                                },
+                                {
+                                    "bbox": [
+                                        112,
+                                        661.3333333333333,
+                                        493,
+                                        708.9999999999999
+                                    ],
+                                    "spans": [],
+                                    "index": 30
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                105,
+                                721,
+                                505,
+                                743
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        721,
+                                        504,
+                                        733
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                721,
+                                                504,
+                                                733
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 8: Cosine update schedule hyper-parameter sweep done using dynamic sparse training meth-",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 31
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        731,
+                                        242,
+                                        744
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                731,
+                                                242,
+                                                744
+                                            ],
+                                            "score": 1.0,
+                                            "content": "ods SET (left) and SNFS (right).",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 32
+                                }
+                            ],
+                            "index": 31.5
+                        }
+                    ],
+                    "index": 30.25
+                }
+            ],
+            "page_idx": 14,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        106,
+                        26,
+                        308,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                25,
+                                308,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        25,
+                                        308,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Under review as a conference paper at ICLR 2020",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        300,
+                        751,
+                        311,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                299,
+                                750,
+                                313,
+                                764
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        299,
+                                        750,
+                                        313,
+                                        764
+                                    ],
+                                    "score": 1.0,
+                                    "content": "15",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "table",
+                    "bbox": [
+                        160,
+                        79,
+                        452,
+                        138
+                    ],
+                    "blocks": [
+                        {
+                            "type": "table_body",
+                            "bbox": [
+                                160,
+                                79,
+                                452,
+                                138
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        160,
+                                        79,
+                                        452,
+                                        138
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                160,
+                                                79,
+                                                452,
+                                                138
+                                            ],
+                                            "score": 0.978,
+                                            "html": "<table><tr><td>Initialization</td><td>Training Method</td><td>Test Accuracy</td><td>Training FLOPs</td></tr><tr><td>Lottery</td><td>Static</td><td>70.82±0.07</td><td>0.46x</td></tr><tr><td>Lottery</td><td>RigL</td><td>73.93±0.09</td><td>0.46x</td></tr><tr><td>Random</td><td>RigL</td><td>74.55±0.06</td><td>0.23x</td></tr><tr><td>Random</td><td>RigL2x</td><td>76.06±0.09</td><td>0.46x</td></tr></table>",
+                                            "type": "table",
+                                            "image_path": "94d62b6ac1e54ad04744876e63f71ea149d6ef8741dcfc583da03af321e74f12.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 1,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        160,
+                                        79,
+                                        452,
+                                        98.66666666666667
+                                    ],
+                                    "spans": [],
+                                    "index": 0
+                                },
+                                {
+                                    "bbox": [
+                                        160,
+                                        98.66666666666667,
+                                        452,
+                                        118.33333333333334
+                                    ],
+                                    "spans": [],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        160,
+                                        118.33333333333334,
+                                        452,
+                                        138.0
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                }
+                            ]
+                        },
+                        {
+                            "type": "table_caption",
+                            "bbox": [
+                                105,
+                                145,
+                                502,
+                                168
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        144,
+                                        505,
+                                        158
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                144,
+                                                505,
+                                                158
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Table 3: Effect of lottery ticket initialization on the final performance. There are no special tickets",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 3
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        156,
+                                        415,
+                                        169
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                156,
+                                                415,
+                                                169
+                                            ],
+                                            "score": 1.0,
+                                            "content": "and dynamic connectivity provided by RigL is critical for good performance.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 4
+                                }
+                            ],
+                            "index": 3.5
+                        }
+                    ],
+                    "index": 2.25
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        105,
+                        187,
+                        501,
+                        201
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                187,
+                                502,
+                                202
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        187,
+                                        502,
+                                        202
+                                    ],
+                                    "score": 1.0,
+                                    "content": "E EFFECT OF UPDATE SCHEDULES ON OTHER DYNAMIC SPARSE METHODS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        }
+                    ],
+                    "index": 5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        212,
+                        505,
+                        257
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                212,
+                                506,
+                                226
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        212,
+                                        506,
+                                        226
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In Figure 8 we repeat the hyper-parameter sweep done for RigL in Figure 5-right, using SET and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                224,
+                                505,
+                                236
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        224,
+                                        226,
+                                        236
+                                    ],
+                                    "score": 1.0,
+                                    "content": "SNFS. Cosine schedule with",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        226,
+                                        224,
+                                        269,
+                                        234
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\Delta T = 5 0",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        269,
+                                        224,
+                                        288,
+                                        236
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        288,
+                                        225,
+                                        325,
+                                        235
+                                    ],
+                                    "score": 0.9,
+                                    "content": "\\alpha = 0 . 1",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        325,
+                                        224,
+                                        505,
+                                        236
+                                    ],
+                                    "score": 1.0,
+                                    "content": "seems to work best across all methods. An",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                235,
+                                505,
+                                247
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        235,
+                                        318,
+                                        247
+                                    ],
+                                    "score": 1.0,
+                                    "content": "interesting observation is that higher drop fractions",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        318,
+                                        236,
+                                        332,
+                                        246
+                                    ],
+                                    "score": 0.74,
+                                    "content": "( \\alpha )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        333,
+                                        235,
+                                        505,
+                                        247
+                                    ],
+                                    "score": 1.0,
+                                    "content": "seem to work better with longer intervals",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                107,
+                                246,
+                                410,
+                                258
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        246,
+                                        123,
+                                        256
+                                    ],
+                                    "score": 0.75,
+                                    "content": "\\Delta T",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        124,
+                                        246,
+                                        222,
+                                        258
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". For example, SET with",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        223,
+                                        246,
+                                        273,
+                                        256
+                                    ],
+                                    "score": 0.9,
+                                    "content": "\\Delta T = 1 0 0 0",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        273,
+                                        246,
+                                        372,
+                                        258
+                                    ],
+                                    "score": 1.0,
+                                    "content": "seems to work best with",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        372,
+                                        246,
+                                        406,
+                                        256
+                                    ],
+                                    "score": 0.89,
+                                    "content": "\\alpha = 0 . 5",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        407,
+                                        246,
+                                        410,
+                                        258
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        }
+                    ],
+                    "index": 7.5,
+                    "bbox_fs": [
+                        105,
+                        212,
+                        506,
+                        258
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        106,
+                        273,
+                        309,
+                        286
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                272,
+                                310,
+                                288
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        272,
+                                        310,
+                                        288
+                                    ],
+                                    "score": 1.0,
+                                    "content": "F ALTERNATIVE UPDATE SCHEDULES",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        }
+                    ],
+                    "index": 10
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        298,
+                        416,
+                        310
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                297,
+                                417,
+                                312
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        297,
+                                        417,
+                                        312
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In Figure 9, we share the performance of two alternative annealing functions:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        }
+                    ],
+                    "index": 11,
+                    "bbox_fs": [
+                        106,
+                        297,
+                        417,
+                        312
+                    ]
+                },
+                {
+                    "type": "list",
+                    "bbox": [
+                        130,
+                        318,
+                        505,
+                        369
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                129,
+                                317,
+                                246,
+                                333
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        129,
+                                        317,
+                                        185,
+                                        333
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1. Constant:",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        185,
+                                        319,
+                                        243,
+                                        332
+                                    ],
+                                    "score": 0.92,
+                                    "content": "f _ { d e c a y } ( t ) = \\alpha",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        243,
+                                        317,
+                                        246,
+                                        333
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12,
+                            "is_list_start_line": true,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                129,
+                                333,
+                                505,
+                                346
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        129,
+                                        333,
+                                        505,
+                                        346
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2. Inverse Power: The fraction of weights updated decreases similarly to the schedule used",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                343,
+                                506,
+                                359
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        344,
+                                        320,
+                                        358
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in Zhu & Gupta (2018) for iterative pruning:",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        321,
+                                        344,
+                                        424,
+                                        358
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\begin{array} { r } { f _ { d e c a y } ( t ) = \\alpha ( 1 - \\frac { t } { T _ { e n d } } ) ^ { k } } \\end{array}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        413,
+                                        343,
+                                        506,
+                                        359
+                                    ],
+                                    "score": 1.0,
+                                    "content": ")k. In our experiments",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                142,
+                                358,
+                                402,
+                                369
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        142,
+                                        358,
+                                        177,
+                                        369
+                                    ],
+                                    "score": 1.0,
+                                    "content": "we tried",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        177,
+                                        358,
+                                        202,
+                                        367
+                                    ],
+                                    "score": 0.9,
+                                    "content": "k = 1",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        203,
+                                        358,
+                                        372,
+                                        369
+                                    ],
+                                    "score": 1.0,
+                                    "content": "which is the linear decay and their default",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        373,
+                                        358,
+                                        398,
+                                        367
+                                    ],
+                                    "score": 0.87,
+                                    "content": "k = 3",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        398,
+                                        358,
+                                        402,
+                                        369
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15,
+                            "is_list_end_line": true
+                        }
+                    ],
+                    "index": 13.5,
+                    "bbox_fs": [
+                        129,
+                        317,
+                        506,
+                        369
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        378,
+                        505,
+                        433
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                378,
+                                505,
+                                391
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        378,
+                                        375,
+                                        391
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Constant seems to perform well with low initial drop fractions like",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        376,
+                                        379,
+                                        409,
+                                        389
+                                    ],
+                                    "score": 0.89,
+                                    "content": "\\alpha = 0 . 1",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        410,
+                                        378,
+                                        505,
+                                        391
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", but it starts to perform",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                389,
+                                505,
+                                402
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        389,
+                                        195,
+                                        402
+                                    ],
+                                    "score": 1.0,
+                                    "content": "worse with increasing",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        195,
+                                        391,
+                                        203,
+                                        399
+                                    ],
+                                    "score": 0.62,
+                                    "content": "\\alpha",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        203,
+                                        389,
+                                        278,
+                                        402
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Inverse Power for",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        279,
+                                        390,
+                                        296,
+                                        399
+                                    ],
+                                    "score": 0.79,
+                                    "content": "{ \\bf k } = 3",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        296,
+                                        389,
+                                        312,
+                                        402
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        313,
+                                        390,
+                                        330,
+                                        400
+                                    ],
+                                    "score": 0.8,
+                                    "content": "{ \\mathrm { k } } { = } 1",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        330,
+                                        389,
+                                        505,
+                                        402
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(Linear) seems to perform similarly for low",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                107,
+                                399,
+                                504,
+                                412
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        402,
+                                        114,
+                                        410
+                                    ],
+                                    "score": 0.72,
+                                    "content": "\\alpha",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        115,
+                                        399,
+                                        334,
+                                        412
+                                    ],
+                                    "score": 1.0,
+                                    "content": "values. However the performance drops noticeably for",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        334,
+                                        401,
+                                        352,
+                                        411
+                                    ],
+                                    "score": 0.75,
+                                    "content": "{ \\bf k } = 3",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        352,
+                                        399,
+                                        504,
+                                        412
+                                    ],
+                                    "score": 1.0,
+                                    "content": "when we increase the update interval.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                411,
+                                505,
+                                424
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        411,
+                                        326,
+                                        424
+                                    ],
+                                    "score": 1.0,
+                                    "content": "As reported by Dettmers & Zettlemoyer (2019) linear",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        327,
+                                        412,
+                                        345,
+                                        422
+                                    ],
+                                    "score": 0.73,
+                                    "content": "( \\mathrm { k } { = } 1",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        345,
+                                        411,
+                                        505,
+                                        424
+                                    ],
+                                    "score": 1.0,
+                                    "content": ") seems to provide similar results as the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                422,
+                                173,
+                                434
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        422,
+                                        173,
+                                        434
+                                    ],
+                                    "score": 1.0,
+                                    "content": "cosine schedule.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        }
+                    ],
+                    "index": 18,
+                    "bbox_fs": [
+                        106,
+                        378,
+                        505,
+                        434
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        450,
+                        386,
+                        462
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                448,
+                                387,
+                                464
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        448,
+                                        387,
+                                        464
+                                    ],
+                                    "score": 1.0,
+                                    "content": "G CALCULATING FLOPS OF MODELS AND METHODS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        }
+                    ],
+                    "index": 21
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        474,
+                        505,
+                        508
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                474,
+                                505,
+                                487
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        474,
+                                        505,
+                                        487
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In order to calculate FLOPs needed for a single forward pass of a sparse model, we count the total",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                485,
+                                505,
+                                498
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        486,
+                                        423,
+                                        498
+                                    ],
+                                    "score": 1.0,
+                                    "content": "number of multiplications and additions layer by layer for a given layer sparsity",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        424,
+                                        485,
+                                        432,
+                                        496
+                                    ],
+                                    "score": 0.86,
+                                    "content": "s ^ { l }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        433,
+                                        486,
+                                        505,
+                                        498
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". The total FLOPs",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                497,
+                                360,
+                                509
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        497,
+                                        360,
+                                        509
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is then obtained by summing up all of these multiply and adds.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        }
+                    ],
+                    "index": 23,
+                    "bbox_fs": [
+                        105,
+                        474,
+                        505,
+                        509
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        513,
+                        505,
+                        547
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                513,
+                                505,
+                                527
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        513,
+                                        505,
+                                        527
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Different sparsity distributions require different number of FLOPs to compute a single prediction.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                524,
+                                505,
+                                537
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        524,
+                                        505,
+                                        537
+                                    ],
+                                    "score": 1.0,
+                                    "content": "For example Erdos-Renyi-Kernel ˝ distributions usually cause earlier layers to be less sparse than",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                535,
+                                505,
+                                549
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        535,
+                                        505,
+                                        549
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the later layers (see Appendix I). The inputs of earlier layers have greater spatial dimensions, so",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                82,
+                                505,
+                                95
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        82,
+                                        505,
+                                        95
+                                    ],
+                                    "score": 1.0,
+                                    "content": "a convolutional kernel that works on such inputs will require more FLOPs to compute the output",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                93,
+                                506,
+                                106
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        93,
+                                        506,
+                                        106
+                                    ],
+                                    "score": 1.0,
+                                    "content": "features compared to later layers. Thus, having earlier layers which are less sparse results in a",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                105,
+                                314,
+                                117
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        105,
+                                        314,
+                                        117
+                                    ],
+                                    "score": 1.0,
+                                    "content": "higher total number of FLOPs required by a model.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 2
+                        }
+                    ],
+                    "index": 26,
+                    "bbox_fs": [
+                        105,
+                        513,
+                        505,
+                        549
+                    ]
+                },
+                {
+                    "type": "image",
+                    "bbox": [
+                        112,
+                        566,
+                        493,
+                        709
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                112,
+                                566,
+                                493,
+                                709
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        112,
+                                        566,
+                                        493,
+                                        709
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                112,
+                                                566,
+                                                493,
+                                                709
+                                            ],
+                                            "score": 0.966,
+                                            "type": "image",
+                                            "image_path": "b3e467a52c2f5b11283887318ea99ba50224d604ca1a45394151b61438546ee9.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 29,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        112,
+                                        566,
+                                        493,
+                                        613.6666666666666
+                                    ],
+                                    "spans": [],
+                                    "index": 28
+                                },
+                                {
+                                    "bbox": [
+                                        112,
+                                        613.6666666666666,
+                                        493,
+                                        661.3333333333333
+                                    ],
+                                    "spans": [],
+                                    "index": 29
+                                },
+                                {
+                                    "bbox": [
+                                        112,
+                                        661.3333333333333,
+                                        493,
+                                        708.9999999999999
+                                    ],
+                                    "spans": [],
+                                    "index": 30
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                105,
+                                721,
+                                505,
+                                743
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        721,
+                                        504,
+                                        733
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                721,
+                                                504,
+                                                733
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 8: Cosine update schedule hyper-parameter sweep done using dynamic sparse training meth-",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 31
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        731,
+                                        242,
+                                        744
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                731,
+                                                242,
+                                                744
+                                            ],
+                                            "score": 1.0,
+                                            "content": "ods SET (left) and SNFS (right).",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 32
+                                }
+                            ],
+                            "index": 31.5
+                        }
+                    ],
+                    "index": 30.25
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        82,
+                        505,
+                        116
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                82,
+                                505,
+                                95
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        82,
+                                        505,
+                                        95
+                                    ],
+                                    "score": 1.0,
+                                    "content": "a convolutional kernel that works on such inputs will require more FLOPs to compute the output",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                93,
+                                506,
+                                106
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        93,
+                                        506,
+                                        106
+                                    ],
+                                    "score": 1.0,
+                                    "content": "features compared to later layers. Thus, having earlier layers which are less sparse results in a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                105,
+                                314,
+                                117
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        105,
+                                        314,
+                                        117
+                                    ],
+                                    "score": 1.0,
+                                    "content": "higher total number of FLOPs required by a model.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        }
+                    ],
+                    "index": 1
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        121,
+                        311,
+                        133
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                120,
+                                312,
+                                135
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        120,
+                                        312,
+                                        135
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Training a neural network consists of 2 main steps:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        }
+                    ],
+                    "index": 3
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        129,
+                        142,
+                        505,
+                        258
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                129,
+                                142,
+                                506,
+                                156
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        129,
+                                        142,
+                                        506,
+                                        156
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1. forward pass: Calculating the loss of the current set of parameters on a given batch of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                154,
+                                505,
+                                166
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        154,
+                                        505,
+                                        166
+                                    ],
+                                    "score": 1.0,
+                                    "content": "data. During this process layer activations are calculated in sequence using the previous",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                164,
+                                506,
+                                178
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        164,
+                                        506,
+                                        178
+                                    ],
+                                    "score": 1.0,
+                                    "content": "activations and the parameters of the layer. Activation of layers are stored in memory for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                175,
+                                221,
+                                189
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        175,
+                                        221,
+                                        189
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the backward pass.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                128,
+                                191,
+                                505,
+                                204
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        128,
+                                        191,
+                                        505,
+                                        204
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2. backward pass: Using the loss value as the initial error signal, we back-propagate the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                202,
+                                506,
+                                215
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        202,
+                                        506,
+                                        215
+                                    ],
+                                    "score": 1.0,
+                                    "content": "error signal while calculating the gradient of parameters. During the backward pass each",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                214,
+                                505,
+                                226
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        214,
+                                        505,
+                                        226
+                                    ],
+                                    "score": 1.0,
+                                    "content": "layer calculates 2 quantities: the gradient of the activations of the previous layer and the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                224,
+                                506,
+                                237
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        224,
+                                        506,
+                                        237
+                                    ],
+                                    "score": 1.0,
+                                    "content": "gradient of its parameters. Therefore in our calculations we count backward passes as two",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                142,
+                                236,
+                                505,
+                                248
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        142,
+                                        236,
+                                        505,
+                                        248
+                                    ],
+                                    "score": 1.0,
+                                    "content": "times the computational expense of the forward pass. We omit the FLOPs needed for batch",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                246,
+                                275,
+                                260
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        246,
+                                        275,
+                                        260
+                                    ],
+                                    "score": 1.0,
+                                    "content": "normalization and cross entropy.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        }
+                    ],
+                    "index": 8.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        267,
+                        505,
+                        312
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                267,
+                                505,
+                                280
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        267,
+                                        505,
+                                        280
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Dynamic sparse training methods require some extra FLOPs to update the connectivity of the neural",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                279,
+                                505,
+                                291
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        279,
+                                        505,
+                                        291
+                                    ],
+                                    "score": 1.0,
+                                    "content": "network. We omit FLOPs needed for dropping the lowest magnitude connections in our calculations.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                290,
+                                506,
+                                303
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        290,
+                                        278,
+                                        303
+                                    ],
+                                    "score": 1.0,
+                                    "content": "For a given dense architecture with FLOPs",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        278,
+                                        290,
+                                        291,
+                                        301
+                                    ],
+                                    "score": 0.89,
+                                    "content": "f _ { D }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        291,
+                                        290,
+                                        424,
+                                        303
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and a sparse version with FLOPs",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        424,
+                                        290,
+                                        436,
+                                        301
+                                    ],
+                                    "score": 0.86,
+                                    "content": "f _ { S }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        437,
+                                        290,
+                                        506,
+                                        303
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", the total FLOPs",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                300,
+                                413,
+                                313
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        300,
+                                        413,
+                                        313
+                                    ],
+                                    "score": 1.0,
+                                    "content": "required to calculate the gradient on a single sample is computed as follows:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        }
+                    ],
+                    "index": 15.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        132,
+                        321,
+                        505,
+                        478
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                131,
+                                321,
+                                456,
+                                335
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        131,
+                                        321,
+                                        298,
+                                        335
+                                    ],
+                                    "score": 1.0,
+                                    "content": "• Static Sparse and Dense. Scales with",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        298,
+                                        322,
+                                        324,
+                                        334
+                                    ],
+                                    "score": 0.91,
+                                    "content": "3 * f _ { S }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        325,
+                                        321,
+                                        342,
+                                        335
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        343,
+                                        322,
+                                        370,
+                                        334
+                                    ],
+                                    "score": 0.92,
+                                    "content": "3 * f _ { D }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        370,
+                                        321,
+                                        456,
+                                        335
+                                    ],
+                                    "score": 1.0,
+                                    "content": "FLOPs, respectively.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                132,
+                                338,
+                                505,
+                                350
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        132,
+                                        338,
+                                        505,
+                                        350
+                                    ],
+                                    "score": 1.0,
+                                    "content": "• Snip. We omit the initial dense gradient calculation since it is negligible, which means",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                349,
+                                393,
+                                361
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        349,
+                                        331,
+                                        361
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Snip scales in the same way as Static methods:",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        332,
+                                        349,
+                                        358,
+                                        360
+                                    ],
+                                    "score": 0.9,
+                                    "content": "3 * f _ { S }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        358,
+                                        349,
+                                        393,
+                                        361
+                                    ],
+                                    "score": 1.0,
+                                    "content": "FLOPs.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                132,
+                                363,
+                                504,
+                                377
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        132,
+                                        363,
+                                        504,
+                                        377
+                                    ],
+                                    "score": 1.0,
+                                    "content": "• SET. We omit the extra FLOPs needed for growing random connections, since this op-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                142,
+                                376,
+                                505,
+                                387
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        142,
+                                        376,
+                                        505,
+                                        387
+                                    ],
+                                    "score": 1.0,
+                                    "content": "eration can be done on chip efficiently. Therefore, the total FLOPs for SET scales with",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                142,
+                                385,
+                                173,
+                                399
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        142,
+                                        387,
+                                        168,
+                                        398
+                                    ],
+                                    "score": 0.9,
+                                    "content": "3 * f _ { S }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        169,
+                                        385,
+                                        173,
+                                        399
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                140,
+                                401,
+                                505,
+                                415
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        140,
+                                        401,
+                                        403,
+                                        415
+                                    ],
+                                    "score": 1.0,
+                                    "content": "SNFS. Forward pass and back-propagating the error signal needs",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        404,
+                                        403,
+                                        429,
+                                        414
+                                    ],
+                                    "score": 0.91,
+                                    "content": "2 * f _ { S }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        430,
+                                        401,
+                                        505,
+                                        415
+                                    ],
+                                    "score": 1.0,
+                                    "content": "FLOPs. However,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                142,
+                                413,
+                                505,
+                                425
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        142,
+                                        413,
+                                        505,
+                                        425
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the dense gradient needs to be calculated at every iteration. Thus, the total number of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                142,
+                                424,
+                                273,
+                                436
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        142,
+                                        424,
+                                        220,
+                                        436
+                                    ],
+                                    "score": 1.0,
+                                    "content": "FLOPs scales with",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        220,
+                                        424,
+                                        270,
+                                        436
+                                    ],
+                                    "score": 0.94,
+                                    "content": "2 * f _ { S } + f _ { D }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        270,
+                                        424,
+                                        273,
+                                        436
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                140,
+                                440,
+                                504,
+                                452
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        140,
+                                        440,
+                                        346,
+                                        452
+                                    ],
+                                    "score": 1.0,
+                                    "content": "RigL. Iterations with no connection updates need",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        346,
+                                        440,
+                                        373,
+                                        451
+                                    ],
+                                    "score": 0.9,
+                                    "content": "3 * f _ { S }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        374,
+                                        440,
+                                        487,
+                                        452
+                                    ],
+                                    "score": 1.0,
+                                    "content": "FLOPs. However, at every",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        487,
+                                        440,
+                                        504,
+                                        450
+                                    ],
+                                    "score": 0.81,
+                                    "content": "\\Delta T",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                451,
+                                505,
+                                463
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        451,
+                                        505,
+                                        463
+                                    ],
+                                    "score": 1.0,
+                                    "content": "iteration we need to calculate the dense gradients. This results in the average FLOPs for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                461,
+                                286,
+                                479
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        461,
+                                        201,
+                                        479
+                                    ],
+                                    "score": 1.0,
+                                    "content": "RigL given by",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        202,
+                                        461,
+                                        283,
+                                        479
+                                    ],
+                                    "score": 0.94,
+                                    "content": "\\frac { ( 3 * f _ { S } * \\Delta T + 2 * f _ { S } + f _ { D } ) } { ( \\Delta T + 1 ) }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        283,
+                                        461,
+                                        286,
+                                        479
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        }
+                    ],
+                    "index": 23.5
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        492,
+                        416,
+                        505
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                492,
+                                417,
+                                507
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        492,
+                                        417,
+                                        507
+                                    ],
+                                    "score": 1.0,
+                                    "content": "H ADDITIONAL PLOTS AND EXPERIMENTS FOR CIFAR-10",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        }
+                    ],
+                    "index": 30
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        518,
+                        505,
+                        596
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                518,
+                                505,
+                                531
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        518,
+                                        505,
+                                        531
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In Figure 10-left, we plot the final training loss of experiments presented in Section 4.3 to investigate",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                529,
+                                506,
+                                542
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        529,
+                                        506,
+                                        542
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the generalization properties of the algorithms considered. Poor performance of Static reflects itself",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                539,
+                                505,
+                                554
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        539,
+                                        505,
+                                        554
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in training loss clearly across all sparsity levels. RigL achieves similar final loss as the pruning,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                551,
+                                505,
+                                564
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        551,
+                                        505,
+                                        564
+                                    ],
+                                    "score": 1.0,
+                                    "content": "despite having around half percent less accuracy. Training longer with RigL decreases the final",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                561,
+                                506,
+                                576
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        561,
+                                        506,
+                                        576
+                                    ],
+                                    "score": 1.0,
+                                    "content": "loss further and the test accuracies start matching pruning (see Figure 4-right) performance. These",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                573,
+                                505,
+                                586
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        573,
+                                        505,
+                                        586
+                                    ],
+                                    "score": 1.0,
+                                    "content": "results show that RigL improves the optimization as promised, however generalizes slightly worse",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                582,
+                                162,
+                                598
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        582,
+                                        162,
+                                        598
+                                    ],
+                                    "score": 1.0,
+                                    "content": "than pruning.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        }
+                    ],
+                    "index": 34
+                },
+                {
+                    "type": "image",
+                    "bbox": [
+                        111,
+                        615,
+                        497,
+                        711
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                111,
+                                615,
+                                497,
+                                711
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        111,
+                                        615,
+                                        497,
+                                        711
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                111,
+                                                615,
+                                                497,
+                                                711
+                                            ],
+                                            "score": 0.95,
+                                            "type": "image",
+                                            "image_path": "bf63372d44ae944bab2fb94af7feac16fac25a9bc8f8d3818b33dd4260972182.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 39,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        111,
+                                        615,
+                                        497,
+                                        647.0
+                                    ],
+                                    "spans": [],
+                                    "index": 38
+                                },
+                                {
+                                    "bbox": [
+                                        111,
+                                        647.0,
+                                        497,
+                                        679.0
+                                    ],
+                                    "spans": [],
+                                    "index": 39
+                                },
+                                {
+                                    "bbox": [
+                                        111,
+                                        679.0,
+                                        497,
+                                        711.0
+                                    ],
+                                    "spans": [],
+                                    "index": 40
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                105,
+                                721,
+                                506,
+                                743
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        719,
+                                        505,
+                                        733
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                719,
+                                                464,
+                                                733
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 9: Using other update schedules with RigL: (left) Constant (middle) Exponential",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                464,
+                                                721,
+                                                485,
+                                                731
+                                            ],
+                                            "score": 0.76,
+                                            "content": "\\left( \\mathrm { k } \\mathrm { = } 3 \\right)",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                486,
+                                                719,
+                                                505,
+                                                733
+                                            ],
+                                            "score": 1.0,
+                                            "content": "and",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 41
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        731,
+                                        166,
+                                        744
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                731,
+                                                166,
+                                                744
+                                            ],
+                                            "score": 1.0,
+                                            "content": "(right) Linear",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 42
+                                }
+                            ],
+                            "index": 41.5
+                        }
+                    ],
+                    "index": 40.25
+                }
+            ],
+            "page_idx": 15,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        107,
+                        27,
+                        308,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                107,
+                                26,
+                                308,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        26,
+                                        308,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Under review as a conference paper at ICLR 2020",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        300,
+                        751,
+                        311,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                299,
+                                750,
+                                313,
+                                764
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        299,
+                                        750,
+                                        313,
+                                        764
+                                    ],
+                                    "score": 1.0,
+                                    "content": "16",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        82,
+                        505,
+                        116
+                    ],
+                    "lines": [],
+                    "index": 1,
+                    "bbox_fs": [
+                        105,
+                        82,
+                        506,
+                        117
+                    ],
+                    "lines_deleted": true
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        121,
+                        311,
+                        133
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                120,
+                                312,
+                                135
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        120,
+                                        312,
+                                        135
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Training a neural network consists of 2 main steps:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        }
+                    ],
+                    "index": 3,
+                    "bbox_fs": [
+                        106,
+                        120,
+                        312,
+                        135
+                    ]
+                },
+                {
+                    "type": "list",
+                    "bbox": [
+                        129,
+                        142,
+                        505,
+                        258
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                129,
+                                142,
+                                506,
+                                156
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        129,
+                                        142,
+                                        506,
+                                        156
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1. forward pass: Calculating the loss of the current set of parameters on a given batch of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                154,
+                                505,
+                                166
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        154,
+                                        505,
+                                        166
+                                    ],
+                                    "score": 1.0,
+                                    "content": "data. During this process layer activations are calculated in sequence using the previous",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                164,
+                                506,
+                                178
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        164,
+                                        506,
+                                        178
+                                    ],
+                                    "score": 1.0,
+                                    "content": "activations and the parameters of the layer. Activation of layers are stored in memory for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                175,
+                                221,
+                                189
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        175,
+                                        221,
+                                        189
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the backward pass.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                128,
+                                191,
+                                505,
+                                204
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        128,
+                                        191,
+                                        505,
+                                        204
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2. backward pass: Using the loss value as the initial error signal, we back-propagate the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                202,
+                                506,
+                                215
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        202,
+                                        506,
+                                        215
+                                    ],
+                                    "score": 1.0,
+                                    "content": "error signal while calculating the gradient of parameters. During the backward pass each",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                214,
+                                505,
+                                226
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        214,
+                                        505,
+                                        226
+                                    ],
+                                    "score": 1.0,
+                                    "content": "layer calculates 2 quantities: the gradient of the activations of the previous layer and the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                224,
+                                506,
+                                237
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        224,
+                                        506,
+                                        237
+                                    ],
+                                    "score": 1.0,
+                                    "content": "gradient of its parameters. Therefore in our calculations we count backward passes as two",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                142,
+                                236,
+                                505,
+                                248
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        142,
+                                        236,
+                                        505,
+                                        248
+                                    ],
+                                    "score": 1.0,
+                                    "content": "times the computational expense of the forward pass. We omit the FLOPs needed for batch",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                246,
+                                275,
+                                260
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        246,
+                                        275,
+                                        260
+                                    ],
+                                    "score": 1.0,
+                                    "content": "normalization and cross entropy.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13,
+                            "is_list_end_line": true
+                        }
+                    ],
+                    "index": 8.5,
+                    "bbox_fs": [
+                        128,
+                        142,
+                        506,
+                        260
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        267,
+                        505,
+                        312
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                267,
+                                505,
+                                280
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        267,
+                                        505,
+                                        280
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Dynamic sparse training methods require some extra FLOPs to update the connectivity of the neural",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                279,
+                                505,
+                                291
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        279,
+                                        505,
+                                        291
+                                    ],
+                                    "score": 1.0,
+                                    "content": "network. We omit FLOPs needed for dropping the lowest magnitude connections in our calculations.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                290,
+                                506,
+                                303
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        290,
+                                        278,
+                                        303
+                                    ],
+                                    "score": 1.0,
+                                    "content": "For a given dense architecture with FLOPs",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        278,
+                                        290,
+                                        291,
+                                        301
+                                    ],
+                                    "score": 0.89,
+                                    "content": "f _ { D }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        291,
+                                        290,
+                                        424,
+                                        303
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and a sparse version with FLOPs",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        424,
+                                        290,
+                                        436,
+                                        301
+                                    ],
+                                    "score": 0.86,
+                                    "content": "f _ { S }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        437,
+                                        290,
+                                        506,
+                                        303
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", the total FLOPs",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                300,
+                                413,
+                                313
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        300,
+                                        413,
+                                        313
+                                    ],
+                                    "score": 1.0,
+                                    "content": "required to calculate the gradient on a single sample is computed as follows:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        }
+                    ],
+                    "index": 15.5,
+                    "bbox_fs": [
+                        105,
+                        267,
+                        506,
+                        313
+                    ]
+                },
+                {
+                    "type": "list",
+                    "bbox": [
+                        132,
+                        321,
+                        505,
+                        478
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                131,
+                                321,
+                                456,
+                                335
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        131,
+                                        321,
+                                        298,
+                                        335
+                                    ],
+                                    "score": 1.0,
+                                    "content": "• Static Sparse and Dense. Scales with",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        298,
+                                        322,
+                                        324,
+                                        334
+                                    ],
+                                    "score": 0.91,
+                                    "content": "3 * f _ { S }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        325,
+                                        321,
+                                        342,
+                                        335
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        343,
+                                        322,
+                                        370,
+                                        334
+                                    ],
+                                    "score": 0.92,
+                                    "content": "3 * f _ { D }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        370,
+                                        321,
+                                        456,
+                                        335
+                                    ],
+                                    "score": 1.0,
+                                    "content": "FLOPs, respectively.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18,
+                            "is_list_start_line": true,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                132,
+                                338,
+                                505,
+                                350
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        132,
+                                        338,
+                                        505,
+                                        350
+                                    ],
+                                    "score": 1.0,
+                                    "content": "• Snip. We omit the initial dense gradient calculation since it is negligible, which means",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                349,
+                                393,
+                                361
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        349,
+                                        331,
+                                        361
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Snip scales in the same way as Static methods:",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        332,
+                                        349,
+                                        358,
+                                        360
+                                    ],
+                                    "score": 0.9,
+                                    "content": "3 * f _ { S }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        358,
+                                        349,
+                                        393,
+                                        361
+                                    ],
+                                    "score": 1.0,
+                                    "content": "FLOPs.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                132,
+                                363,
+                                504,
+                                377
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        132,
+                                        363,
+                                        504,
+                                        377
+                                    ],
+                                    "score": 1.0,
+                                    "content": "• SET. We omit the extra FLOPs needed for growing random connections, since this op-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                142,
+                                376,
+                                505,
+                                387
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        142,
+                                        376,
+                                        505,
+                                        387
+                                    ],
+                                    "score": 1.0,
+                                    "content": "eration can be done on chip efficiently. Therefore, the total FLOPs for SET scales with",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                142,
+                                385,
+                                173,
+                                399
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        142,
+                                        387,
+                                        168,
+                                        398
+                                    ],
+                                    "score": 0.9,
+                                    "content": "3 * f _ { S }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        169,
+                                        385,
+                                        173,
+                                        399
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                140,
+                                401,
+                                505,
+                                415
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        140,
+                                        401,
+                                        403,
+                                        415
+                                    ],
+                                    "score": 1.0,
+                                    "content": "SNFS. Forward pass and back-propagating the error signal needs",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        404,
+                                        403,
+                                        429,
+                                        414
+                                    ],
+                                    "score": 0.91,
+                                    "content": "2 * f _ { S }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        430,
+                                        401,
+                                        505,
+                                        415
+                                    ],
+                                    "score": 1.0,
+                                    "content": "FLOPs. However,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                142,
+                                413,
+                                505,
+                                425
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        142,
+                                        413,
+                                        505,
+                                        425
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the dense gradient needs to be calculated at every iteration. Thus, the total number of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                142,
+                                424,
+                                273,
+                                436
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        142,
+                                        424,
+                                        220,
+                                        436
+                                    ],
+                                    "score": 1.0,
+                                    "content": "FLOPs scales with",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        220,
+                                        424,
+                                        270,
+                                        436
+                                    ],
+                                    "score": 0.94,
+                                    "content": "2 * f _ { S } + f _ { D }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        270,
+                                        424,
+                                        273,
+                                        436
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                140,
+                                440,
+                                504,
+                                452
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        140,
+                                        440,
+                                        346,
+                                        452
+                                    ],
+                                    "score": 1.0,
+                                    "content": "RigL. Iterations with no connection updates need",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        346,
+                                        440,
+                                        373,
+                                        451
+                                    ],
+                                    "score": 0.9,
+                                    "content": "3 * f _ { S }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        374,
+                                        440,
+                                        487,
+                                        452
+                                    ],
+                                    "score": 1.0,
+                                    "content": "FLOPs. However, at every",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        487,
+                                        440,
+                                        504,
+                                        450
+                                    ],
+                                    "score": 0.81,
+                                    "content": "\\Delta T",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                451,
+                                505,
+                                463
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        451,
+                                        505,
+                                        463
+                                    ],
+                                    "score": 1.0,
+                                    "content": "iteration we need to calculate the dense gradients. This results in the average FLOPs for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                461,
+                                286,
+                                479
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        461,
+                                        201,
+                                        479
+                                    ],
+                                    "score": 1.0,
+                                    "content": "RigL given by",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        202,
+                                        461,
+                                        283,
+                                        479
+                                    ],
+                                    "score": 0.94,
+                                    "content": "\\frac { ( 3 * f _ { S } * \\Delta T + 2 * f _ { S } + f _ { D } ) } { ( \\Delta T + 1 ) }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        283,
+                                        461,
+                                        286,
+                                        479
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29,
+                            "is_list_end_line": true
+                        }
+                    ],
+                    "index": 23.5,
+                    "bbox_fs": [
+                        131,
+                        321,
+                        505,
+                        479
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        492,
+                        416,
+                        505
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                492,
+                                417,
+                                507
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        492,
+                                        417,
+                                        507
+                                    ],
+                                    "score": 1.0,
+                                    "content": "H ADDITIONAL PLOTS AND EXPERIMENTS FOR CIFAR-10",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        }
+                    ],
+                    "index": 30
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        518,
+                        505,
+                        596
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                518,
+                                505,
+                                531
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        518,
+                                        505,
+                                        531
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In Figure 10-left, we plot the final training loss of experiments presented in Section 4.3 to investigate",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                529,
+                                506,
+                                542
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        529,
+                                        506,
+                                        542
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the generalization properties of the algorithms considered. Poor performance of Static reflects itself",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                539,
+                                505,
+                                554
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        539,
+                                        505,
+                                        554
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in training loss clearly across all sparsity levels. RigL achieves similar final loss as the pruning,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                551,
+                                505,
+                                564
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        551,
+                                        505,
+                                        564
+                                    ],
+                                    "score": 1.0,
+                                    "content": "despite having around half percent less accuracy. Training longer with RigL decreases the final",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                561,
+                                506,
+                                576
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        561,
+                                        506,
+                                        576
+                                    ],
+                                    "score": 1.0,
+                                    "content": "loss further and the test accuracies start matching pruning (see Figure 4-right) performance. These",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                573,
+                                505,
+                                586
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        573,
+                                        505,
+                                        586
+                                    ],
+                                    "score": 1.0,
+                                    "content": "results show that RigL improves the optimization as promised, however generalizes slightly worse",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                582,
+                                162,
+                                598
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        582,
+                                        162,
+                                        598
+                                    ],
+                                    "score": 1.0,
+                                    "content": "than pruning.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        }
+                    ],
+                    "index": 34,
+                    "bbox_fs": [
+                        104,
+                        518,
+                        506,
+                        598
+                    ]
+                },
+                {
+                    "type": "image",
+                    "bbox": [
+                        111,
+                        615,
+                        497,
+                        711
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                111,
+                                615,
+                                497,
+                                711
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        111,
+                                        615,
+                                        497,
+                                        711
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                111,
+                                                615,
+                                                497,
+                                                711
+                                            ],
+                                            "score": 0.95,
+                                            "type": "image",
+                                            "image_path": "bf63372d44ae944bab2fb94af7feac16fac25a9bc8f8d3818b33dd4260972182.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 39,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        111,
+                                        615,
+                                        497,
+                                        647.0
+                                    ],
+                                    "spans": [],
+                                    "index": 38
+                                },
+                                {
+                                    "bbox": [
+                                        111,
+                                        647.0,
+                                        497,
+                                        679.0
+                                    ],
+                                    "spans": [],
+                                    "index": 39
+                                },
+                                {
+                                    "bbox": [
+                                        111,
+                                        679.0,
+                                        497,
+                                        711.0
+                                    ],
+                                    "spans": [],
+                                    "index": 40
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                105,
+                                721,
+                                506,
+                                743
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        719,
+                                        505,
+                                        733
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                719,
+                                                464,
+                                                733
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 9: Using other update schedules with RigL: (left) Constant (middle) Exponential",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                464,
+                                                721,
+                                                485,
+                                                731
+                                            ],
+                                            "score": 0.76,
+                                            "content": "\\left( \\mathrm { k } \\mathrm { = } 3 \\right)",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                486,
+                                                719,
+                                                505,
+                                                733
+                                            ],
+                                            "score": 1.0,
+                                            "content": "and",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 41
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        731,
+                                        166,
+                                        744
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                731,
+                                                166,
+                                                744
+                                            ],
+                                            "score": 1.0,
+                                            "content": "(right) Linear",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 42
+                                }
+                            ],
+                            "index": 41.5
+                        }
+                    ],
+                    "index": 40.25
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "image",
+                    "bbox": [
+                        118,
+                        81,
+                        494,
+                        218
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                118,
+                                81,
+                                494,
+                                218
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        118,
+                                        81,
+                                        494,
+                                        218
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                118,
+                                                81,
+                                                494,
+                                                218
+                                            ],
+                                            "score": 0.969,
+                                            "type": "image",
+                                            "image_path": "e207781847e04baa8da603b4cc537a5073cb27d1d766bb864de9512780d702bc.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 1,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        118,
+                                        81,
+                                        494,
+                                        126.66666666666666
+                                    ],
+                                    "spans": [],
+                                    "index": 0
+                                },
+                                {
+                                    "bbox": [
+                                        118,
+                                        126.66666666666666,
+                                        494,
+                                        172.33333333333331
+                                    ],
+                                    "spans": [],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        118,
+                                        172.33333333333331,
+                                        494,
+                                        217.99999999999997
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                107,
+                                228,
+                                503,
+                                251
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        228,
+                                        505,
+                                        241
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                228,
+                                                410,
+                                                241
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 10: Final training loss of sparse models (left) and performance of",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                410,
+                                                229,
+                                                432,
+                                                240
+                                            ],
+                                            "score": 0.42,
+                                            "content": "R i g L",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                432,
+                                                228,
+                                                505,
+                                                241
+                                            ],
+                                            "score": 1.0,
+                                            "content": "at different mask",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 3
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        240,
+                                        205,
+                                        253
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                240,
+                                                205,
+                                                253
+                                            ],
+                                            "score": 1.0,
+                                            "content": "update intervals (right).",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 4
+                                }
+                            ],
+                            "index": 3.5
+                        }
+                    ],
+                    "index": 2.25
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        258,
+                        505,
+                        302
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                258,
+                                504,
+                                270
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        258,
+                                        316,
+                                        270
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In Figure 10-right, we sweep mask update interval",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        317,
+                                        258,
+                                        334,
+                                        268
+                                    ],
+                                    "score": 0.69,
+                                    "content": "\\Delta T",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        334,
+                                        258,
+                                        504,
+                                        270
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and plot the final test accuracies. We fix",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                268,
+                                505,
+                                281
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        268,
+                                        189,
+                                        281
+                                    ],
+                                    "score": 1.0,
+                                    "content": "initial drop fraction",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        189,
+                                        271,
+                                        197,
+                                        279
+                                    ],
+                                    "score": 0.77,
+                                    "content": "\\alpha",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        197,
+                                        268,
+                                        481,
+                                        281
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to 0.3 and evaluate two different sparsity distributions: Uniform and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        481,
+                                        269,
+                                        502,
+                                        279
+                                    ],
+                                    "score": 0.28,
+                                    "content": "E R K",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        502,
+                                        268,
+                                        505,
+                                        281
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                279,
+                                505,
+                                293
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        279,
+                                        505,
+                                        293
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Both curves follow a similar pattern as in Imagenet-2012 sweeps (see Figure 8) and best results are",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                291,
+                                216,
+                                303
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        291,
+                                        167,
+                                        303
+                                    ],
+                                    "score": 1.0,
+                                    "content": "obtained when",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        168,
+                                        291,
+                                        212,
+                                        301
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\Delta T = 1 0 0",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        213,
+                                        291,
+                                        216,
+                                        303
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        }
+                    ],
+                    "index": 6.5
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        318,
+                        430,
+                        331
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                318,
+                                430,
+                                332
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        318,
+                                        430,
+                                        332
+                                    ],
+                                    "score": 1.0,
+                                    "content": "I SPARSITY OF INDIVIDUAL LAYERS FOR SPARSE RESNET-50",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        }
+                    ],
+                    "index": 9
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        343,
+                        503,
+                        365
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                342,
+                                505,
+                                357
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        342,
+                                        505,
+                                        357
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Sparsity of ResNet-50 layers given by the Erdos-R ˝ enyi-Kernel sparsity distribution plotted in Figure ´",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                353,
+                                122,
+                                366
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        353,
+                                        122,
+                                        366
+                                    ],
+                                    "score": 1.0,
+                                    "content": "11.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        }
+                    ],
+                    "index": 10.5
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        381,
+                        347,
+                        394
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                380,
+                                349,
+                                396
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        380,
+                                        349,
+                                        396
+                                    ],
+                                    "score": 1.0,
+                                    "content": "J BUGS DISCOVERED DURING EXPERIMENTS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        }
+                    ],
+                    "index": 12
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        406,
+                        505,
+                        450
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                405,
+                                505,
+                                419
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        405,
+                                        505,
+                                        419
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Our initial implementations contained some subtle bugs, which while not affecting the general con-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                417,
+                                505,
+                                429
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        417,
+                                        505,
+                                        429
+                                    ],
+                                    "score": 1.0,
+                                    "content": "clusion that RigL is more effective than other techniques, did result in lower accuracy for all sparse",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                429,
+                                505,
+                                441
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        429,
+                                        505,
+                                        441
+                                    ],
+                                    "score": 1.0,
+                                    "content": "training techniques. We detail these issues here with the hope that others may learn from our mis-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                438,
+                                133,
+                                452
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        438,
+                                        133,
+                                        452
+                                    ],
+                                    "score": 1.0,
+                                    "content": "takes.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        }
+                    ],
+                    "index": 14.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        131,
+                        461,
+                        505,
+                        582
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                129,
+                                461,
+                                505,
+                                473
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        129,
+                                        461,
+                                        505,
+                                        473
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1. Random operations on multiple replicas. We use data parallelism to split a mini-batch",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                472,
+                                505,
+                                484
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        472,
+                                        505,
+                                        484
+                                    ],
+                                    "score": 1.0,
+                                    "content": "among multiple replicas. Each replica independently calculates the gradients using a dif-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                483,
+                                505,
+                                496
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        483,
+                                        505,
+                                        496
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ferent sub-mini-batch of data. The gradients are aggregated using an ALL-REDUCE oper-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                142,
+                                494,
+                                505,
+                                506
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        142,
+                                        494,
+                                        505,
+                                        506
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ation before the optimizer update. Our implementation of SET, SNFS and RigL depended",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                504,
+                                506,
+                                518
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        504,
+                                        506,
+                                        518
+                                    ],
+                                    "score": 1.0,
+                                    "content": "on each replica independently choosing to drop and grow the same connections. However,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                515,
+                                505,
+                                528
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        515,
+                                        505,
+                                        528
+                                    ],
+                                    "score": 1.0,
+                                    "content": "due to the nature of random operations in Tensorflow, this did not happen. Instead, different",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                527,
+                                505,
+                                539
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        527,
+                                        505,
+                                        539
+                                    ],
+                                    "score": 1.0,
+                                    "content": "replicas diverged after the first drop/grow step. This was most pronounced in SET where",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                142,
+                                538,
+                                505,
+                                549
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        142,
+                                        538,
+                                        505,
+                                        549
+                                    ],
+                                    "score": 1.0,
+                                    "content": "each replica chose at random and much less so for SNFS and RigL where randomness is",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                548,
+                                505,
+                                560
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        548,
+                                        505,
+                                        560
+                                    ],
+                                    "score": 1.0,
+                                    "content": "only needed to break ties. If left unchecked this might be expected to be catastrophic, but",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                559,
+                                506,
+                                572
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        559,
+                                        506,
+                                        572
+                                    ],
+                                    "score": 1.0,
+                                    "content": "due to the behavior of Estimators and/or TF-replicator, the values on the first replica are",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                570,
+                                469,
+                                583
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        570,
+                                        469,
+                                        583
+                                    ],
+                                    "score": 1.0,
+                                    "content": "broadcast to the others periodically (every approximately 1000 steps in our case).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        }
+                    ],
+                    "index": 22
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        140,
+                        583,
+                        502,
+                        606
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                142,
+                                582,
+                                504,
+                                595
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        142,
+                                        582,
+                                        504,
+                                        595
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We fixed this bug by using stateless random operations. As a result the performance of SET",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                594,
+                                337,
+                                607
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        594,
+                                        218,
+                                        607
+                                    ],
+                                    "score": 1.0,
+                                    "content": "improved slightly",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        218,
+                                        595,
+                                        259,
+                                        605
+                                    ],
+                                    "score": 0.8,
+                                    "content": "0 . 1 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        259,
+                                        594,
+                                        337,
+                                        607
+                                    ],
+                                    "score": 1.0,
+                                    "content": "higher on Table 2).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        }
+                    ],
+                    "index": 28.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        131,
+                        609,
+                        505,
+                        697
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                129,
+                                609,
+                                505,
+                                622
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        129,
+                                        609,
+                                        505,
+                                        622
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2. Synchronization between replicas. RigL and SNFS depend on calculating dense gradients",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                620,
+                                506,
+                                633
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        620,
+                                        506,
+                                        633
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with respect to the masked parameters. However, as explained above, in the multiple replica",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                142,
+                                631,
+                                506,
+                                644
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        142,
+                                        631,
+                                        506,
+                                        644
+                                    ],
+                                    "score": 1.0,
+                                    "content": "setting these gradients need to be aggregated. Normally this aggregation is automatically",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                642,
+                                506,
+                                655
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        642,
+                                        506,
+                                        655
+                                    ],
+                                    "score": 1.0,
+                                    "content": "done by the optimizer, but in our case, this does not happen (only the gradients with respect",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                652,
+                                506,
+                                667
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        652,
+                                        506,
+                                        667
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to the unmasked parameters are aggregated automatically). This bug affected SNFS and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                663,
+                                505,
+                                677
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        663,
+                                        505,
+                                        677
+                                    ],
+                                    "score": 1.0,
+                                    "content": "RigL, but not SET since SET does not rely on the gradients to grow connections. Again,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                673,
+                                506,
+                                689
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        673,
+                                        506,
+                                        689
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the synchronization of the parameters from the first replica every approximately 1000 steps",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                684,
+                                212,
+                                699
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        684,
+                                        212,
+                                        699
+                                    ],
+                                    "score": 1.0,
+                                    "content": "masked this bug.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        }
+                    ],
+                    "index": 33.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        143,
+                        699,
+                        504,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                141,
+                                698,
+                                505,
+                                712
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        698,
+                                        505,
+                                        712
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We fixed this bug by explicitly calling ALL-REDUCE on the gradients with respect to the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                709,
+                                505,
+                                723
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        709,
+                                        505,
+                                        723
+                                    ],
+                                    "score": 1.0,
+                                    "content": "masked parameters. With this fix, the performance of RigL and SNFS improved signifi-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                142,
+                                720,
+                                454,
+                                734
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        142,
+                                        720,
+                                        362,
+                                        734
+                                    ],
+                                    "score": 1.0,
+                                    "content": "cantly, particularly for default training lengths (around",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        362,
+                                        721,
+                                        392,
+                                        731
+                                    ],
+                                    "score": 0.82,
+                                    "content": "0 . 5 – 1 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        393,
+                                        720,
+                                        454,
+                                        734
+                                    ],
+                                    "score": 1.0,
+                                    "content": "improvement).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        }
+                    ],
+                    "index": 39
+                }
+            ],
+            "page_idx": 16,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        300,
+                        751,
+                        311,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                299,
+                                750,
+                                312,
+                                764
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        299,
+                                        750,
+                                        312,
+                                        764
+                                    ],
+                                    "score": 1.0,
+                                    "content": "17",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        107,
+                        27,
+                        308,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                25,
+                                309,
+                                39
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        25,
+                                        309,
+                                        39
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Under review as a conference paper at ICLR 2020",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "image",
+                    "bbox": [
+                        118,
+                        81,
+                        494,
+                        218
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                118,
+                                81,
+                                494,
+                                218
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        118,
+                                        81,
+                                        494,
+                                        218
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                118,
+                                                81,
+                                                494,
+                                                218
+                                            ],
+                                            "score": 0.969,
+                                            "type": "image",
+                                            "image_path": "e207781847e04baa8da603b4cc537a5073cb27d1d766bb864de9512780d702bc.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 1,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        118,
+                                        81,
+                                        494,
+                                        126.66666666666666
+                                    ],
+                                    "spans": [],
+                                    "index": 0
+                                },
+                                {
+                                    "bbox": [
+                                        118,
+                                        126.66666666666666,
+                                        494,
+                                        172.33333333333331
+                                    ],
+                                    "spans": [],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        118,
+                                        172.33333333333331,
+                                        494,
+                                        217.99999999999997
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                107,
+                                228,
+                                503,
+                                251
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        228,
+                                        505,
+                                        241
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                228,
+                                                410,
+                                                241
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 10: Final training loss of sparse models (left) and performance of",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                410,
+                                                229,
+                                                432,
+                                                240
+                                            ],
+                                            "score": 0.42,
+                                            "content": "R i g L",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                432,
+                                                228,
+                                                505,
+                                                241
+                                            ],
+                                            "score": 1.0,
+                                            "content": "at different mask",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 3
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        240,
+                                        205,
+                                        253
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                240,
+                                                205,
+                                                253
+                                            ],
+                                            "score": 1.0,
+                                            "content": "update intervals (right).",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 4
+                                }
+                            ],
+                            "index": 3.5
+                        }
+                    ],
+                    "index": 2.25
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        258,
+                        505,
+                        302
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                258,
+                                504,
+                                270
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        258,
+                                        316,
+                                        270
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In Figure 10-right, we sweep mask update interval",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        317,
+                                        258,
+                                        334,
+                                        268
+                                    ],
+                                    "score": 0.69,
+                                    "content": "\\Delta T",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        334,
+                                        258,
+                                        504,
+                                        270
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and plot the final test accuracies. We fix",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                268,
+                                505,
+                                281
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        268,
+                                        189,
+                                        281
+                                    ],
+                                    "score": 1.0,
+                                    "content": "initial drop fraction",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        189,
+                                        271,
+                                        197,
+                                        279
+                                    ],
+                                    "score": 0.77,
+                                    "content": "\\alpha",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        197,
+                                        268,
+                                        481,
+                                        281
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to 0.3 and evaluate two different sparsity distributions: Uniform and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        481,
+                                        269,
+                                        502,
+                                        279
+                                    ],
+                                    "score": 0.28,
+                                    "content": "E R K",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        502,
+                                        268,
+                                        505,
+                                        281
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                279,
+                                505,
+                                293
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        279,
+                                        505,
+                                        293
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Both curves follow a similar pattern as in Imagenet-2012 sweeps (see Figure 8) and best results are",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                291,
+                                216,
+                                303
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        291,
+                                        167,
+                                        303
+                                    ],
+                                    "score": 1.0,
+                                    "content": "obtained when",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        168,
+                                        291,
+                                        212,
+                                        301
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\Delta T = 1 0 0",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        213,
+                                        291,
+                                        216,
+                                        303
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        }
+                    ],
+                    "index": 6.5,
+                    "bbox_fs": [
+                        105,
+                        258,
+                        505,
+                        303
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        318,
+                        430,
+                        331
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                318,
+                                430,
+                                332
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        318,
+                                        430,
+                                        332
+                                    ],
+                                    "score": 1.0,
+                                    "content": "I SPARSITY OF INDIVIDUAL LAYERS FOR SPARSE RESNET-50",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        }
+                    ],
+                    "index": 9
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        343,
+                        503,
+                        365
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                342,
+                                505,
+                                357
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        342,
+                                        505,
+                                        357
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Sparsity of ResNet-50 layers given by the Erdos-R ˝ enyi-Kernel sparsity distribution plotted in Figure ´",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                353,
+                                122,
+                                366
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        353,
+                                        122,
+                                        366
+                                    ],
+                                    "score": 1.0,
+                                    "content": "11.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        }
+                    ],
+                    "index": 10.5,
+                    "bbox_fs": [
+                        105,
+                        342,
+                        505,
+                        366
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        381,
+                        347,
+                        394
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                380,
+                                349,
+                                396
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        380,
+                                        349,
+                                        396
+                                    ],
+                                    "score": 1.0,
+                                    "content": "J BUGS DISCOVERED DURING EXPERIMENTS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        }
+                    ],
+                    "index": 12
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        406,
+                        505,
+                        450
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                405,
+                                505,
+                                419
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        405,
+                                        505,
+                                        419
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Our initial implementations contained some subtle bugs, which while not affecting the general con-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                417,
+                                505,
+                                429
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        417,
+                                        505,
+                                        429
+                                    ],
+                                    "score": 1.0,
+                                    "content": "clusion that RigL is more effective than other techniques, did result in lower accuracy for all sparse",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                429,
+                                505,
+                                441
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        429,
+                                        505,
+                                        441
+                                    ],
+                                    "score": 1.0,
+                                    "content": "training techniques. We detail these issues here with the hope that others may learn from our mis-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                438,
+                                133,
+                                452
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        438,
+                                        133,
+                                        452
+                                    ],
+                                    "score": 1.0,
+                                    "content": "takes.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        }
+                    ],
+                    "index": 14.5,
+                    "bbox_fs": [
+                        105,
+                        405,
+                        505,
+                        452
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        131,
+                        461,
+                        505,
+                        582
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                129,
+                                461,
+                                505,
+                                473
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        129,
+                                        461,
+                                        505,
+                                        473
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1. Random operations on multiple replicas. We use data parallelism to split a mini-batch",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                472,
+                                505,
+                                484
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        472,
+                                        505,
+                                        484
+                                    ],
+                                    "score": 1.0,
+                                    "content": "among multiple replicas. Each replica independently calculates the gradients using a dif-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                483,
+                                505,
+                                496
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        483,
+                                        505,
+                                        496
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ferent sub-mini-batch of data. The gradients are aggregated using an ALL-REDUCE oper-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                142,
+                                494,
+                                505,
+                                506
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        142,
+                                        494,
+                                        505,
+                                        506
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ation before the optimizer update. Our implementation of SET, SNFS and RigL depended",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                504,
+                                506,
+                                518
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        504,
+                                        506,
+                                        518
+                                    ],
+                                    "score": 1.0,
+                                    "content": "on each replica independently choosing to drop and grow the same connections. However,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                515,
+                                505,
+                                528
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        515,
+                                        505,
+                                        528
+                                    ],
+                                    "score": 1.0,
+                                    "content": "due to the nature of random operations in Tensorflow, this did not happen. Instead, different",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                527,
+                                505,
+                                539
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        527,
+                                        505,
+                                        539
+                                    ],
+                                    "score": 1.0,
+                                    "content": "replicas diverged after the first drop/grow step. This was most pronounced in SET where",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                142,
+                                538,
+                                505,
+                                549
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        142,
+                                        538,
+                                        505,
+                                        549
+                                    ],
+                                    "score": 1.0,
+                                    "content": "each replica chose at random and much less so for SNFS and RigL where randomness is",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                548,
+                                505,
+                                560
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        548,
+                                        505,
+                                        560
+                                    ],
+                                    "score": 1.0,
+                                    "content": "only needed to break ties. If left unchecked this might be expected to be catastrophic, but",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                559,
+                                506,
+                                572
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        559,
+                                        506,
+                                        572
+                                    ],
+                                    "score": 1.0,
+                                    "content": "due to the behavior of Estimators and/or TF-replicator, the values on the first replica are",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                570,
+                                469,
+                                583
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        570,
+                                        469,
+                                        583
+                                    ],
+                                    "score": 1.0,
+                                    "content": "broadcast to the others periodically (every approximately 1000 steps in our case).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        }
+                    ],
+                    "index": 22,
+                    "bbox_fs": [
+                        129,
+                        461,
+                        506,
+                        583
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        140,
+                        583,
+                        502,
+                        606
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                142,
+                                582,
+                                504,
+                                595
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        142,
+                                        582,
+                                        504,
+                                        595
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We fixed this bug by using stateless random operations. As a result the performance of SET",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                594,
+                                337,
+                                607
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        594,
+                                        218,
+                                        607
+                                    ],
+                                    "score": 1.0,
+                                    "content": "improved slightly",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        218,
+                                        595,
+                                        259,
+                                        605
+                                    ],
+                                    "score": 0.8,
+                                    "content": "0 . 1 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        259,
+                                        594,
+                                        337,
+                                        607
+                                    ],
+                                    "score": 1.0,
+                                    "content": "higher on Table 2).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        }
+                    ],
+                    "index": 28.5,
+                    "bbox_fs": [
+                        141,
+                        582,
+                        504,
+                        607
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        131,
+                        609,
+                        505,
+                        697
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                129,
+                                609,
+                                505,
+                                622
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        129,
+                                        609,
+                                        505,
+                                        622
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2. Synchronization between replicas. RigL and SNFS depend on calculating dense gradients",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                620,
+                                506,
+                                633
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        620,
+                                        506,
+                                        633
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with respect to the masked parameters. However, as explained above, in the multiple replica",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                142,
+                                631,
+                                506,
+                                644
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        142,
+                                        631,
+                                        506,
+                                        644
+                                    ],
+                                    "score": 1.0,
+                                    "content": "setting these gradients need to be aggregated. Normally this aggregation is automatically",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                642,
+                                506,
+                                655
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        642,
+                                        506,
+                                        655
+                                    ],
+                                    "score": 1.0,
+                                    "content": "done by the optimizer, but in our case, this does not happen (only the gradients with respect",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                652,
+                                506,
+                                667
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        652,
+                                        506,
+                                        667
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to the unmasked parameters are aggregated automatically). This bug affected SNFS and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                663,
+                                505,
+                                677
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        663,
+                                        505,
+                                        677
+                                    ],
+                                    "score": 1.0,
+                                    "content": "RigL, but not SET since SET does not rely on the gradients to grow connections. Again,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                673,
+                                506,
+                                689
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        673,
+                                        506,
+                                        689
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the synchronization of the parameters from the first replica every approximately 1000 steps",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                684,
+                                212,
+                                699
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        684,
+                                        212,
+                                        699
+                                    ],
+                                    "score": 1.0,
+                                    "content": "masked this bug.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        }
+                    ],
+                    "index": 33.5,
+                    "bbox_fs": [
+                        129,
+                        609,
+                        506,
+                        699
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        143,
+                        699,
+                        504,
+                        732
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                141,
+                                698,
+                                505,
+                                712
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        698,
+                                        505,
+                                        712
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We fixed this bug by explicitly calling ALL-REDUCE on the gradients with respect to the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                709,
+                                505,
+                                723
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        709,
+                                        505,
+                                        723
+                                    ],
+                                    "score": 1.0,
+                                    "content": "masked parameters. With this fix, the performance of RigL and SNFS improved signifi-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                142,
+                                720,
+                                454,
+                                734
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        142,
+                                        720,
+                                        362,
+                                        734
+                                    ],
+                                    "score": 1.0,
+                                    "content": "cantly, particularly for default training lengths (around",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        362,
+                                        721,
+                                        392,
+                                        731
+                                    ],
+                                    "score": 0.82,
+                                    "content": "0 . 5 – 1 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        393,
+                                        720,
+                                        454,
+                                        734
+                                    ],
+                                    "score": 1.0,
+                                    "content": "improvement).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        }
+                    ],
+                    "index": 39,
+                    "bbox_fs": [
+                        141,
+                        698,
+                        505,
+                        734
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "image",
+                    "bbox": [
+                        149,
+                        166,
+                        461,
+                        646
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                149,
+                                166,
+                                461,
+                                646
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        149,
+                                        167,
+                                        461,
+                                        646
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                149,
+                                                167,
+                                                461,
+                                                646
+                                            ],
+                                            "score": 0.967,
+                                            "type": "image",
+                                            "image_path": "0b1f9b452aac571c6f076bd13ec4d1e1e1f13be162984229dcfda1d2c8d58426.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 1,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        149,
+                                        166,
+                                        461,
+                                        326.0
+                                    ],
+                                    "spans": [],
+                                    "index": 0
+                                },
+                                {
+                                    "bbox": [
+                                        149,
+                                        326.0,
+                                        461,
+                                        486.0
+                                    ],
+                                    "spans": [],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        149,
+                                        486.0,
+                                        461,
+                                        646.0
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                185,
+                                648,
+                                425,
+                                661
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        186,
+                                        648,
+                                        424,
+                                        662
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                186,
+                                                648,
+                                                424,
+                                                662
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 11: Sparsities of individual layers of the ResNet-50.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 3
+                                }
+                            ],
+                            "index": 3
+                        }
+                    ],
+                    "index": 2.0
+                }
+            ],
+            "page_idx": 17,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        106,
+                        27,
+                        309,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                26,
+                                308,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        26,
+                                        308,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Under review as a conference paper at ICLR 2020",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        300,
+                        751,
+                        311,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                299,
+                                750,
+                                313,
+                                763
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        299,
+                                        750,
+                                        313,
+                                        763
+                                    ],
+                                    "score": 1.0,
+                                    "content": "18",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "image",
+                    "bbox": [
+                        149,
+                        166,
+                        461,
+                        646
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                149,
+                                166,
+                                461,
+                                646
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        149,
+                                        167,
+                                        461,
+                                        646
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                149,
+                                                167,
+                                                461,
+                                                646
+                                            ],
+                                            "score": 0.967,
+                                            "type": "image",
+                                            "image_path": "0b1f9b452aac571c6f076bd13ec4d1e1e1f13be162984229dcfda1d2c8d58426.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 1,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        149,
+                                        166,
+                                        461,
+                                        326.0
+                                    ],
+                                    "spans": [],
+                                    "index": 0
+                                },
+                                {
+                                    "bbox": [
+                                        149,
+                                        326.0,
+                                        461,
+                                        486.0
+                                    ],
+                                    "spans": [],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        149,
+                                        486.0,
+                                        461,
+                                        646.0
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                185,
+                                648,
+                                425,
+                                661
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        186,
+                                        648,
+                                        424,
+                                        662
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                186,
+                                                648,
+                                                424,
+                                                662
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 11: Sparsities of individual layers of the ResNet-50.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 3
+                                }
+                            ],
+                            "index": 3
+                        }
+                    ],
+                    "index": 2.0
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        131,
+                        82,
+                        505,
+                        160
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                128,
+                                81,
+                                506,
+                                96
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        128,
+                                        81,
+                                        506,
+                                        96
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3. SNIP Experiments. Our first implementation of SNIP used the gradient magnitudes to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                93,
+                                506,
+                                107
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        93,
+                                        506,
+                                        107
+                                    ],
+                                    "score": 1.0,
+                                    "content": "decide which connections to keep causing its performance to be worse than static. Upon",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                104,
+                                505,
+                                118
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        104,
+                                        505,
+                                        118
+                                    ],
+                                    "score": 1.0,
+                                    "content": "our discussions with the authors of SNIP, we realized that the correct metric is the saliency",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                115,
+                                505,
+                                128
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        115,
+                                        505,
+                                        128
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(gradient times parameter magnitude). With this correction SNIP performance improved",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                127,
+                                505,
+                                139
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        127,
+                                        505,
+                                        139
+                                    ],
+                                    "score": 1.0,
+                                    "content": "dramatically to better than random (Static) even at Resnet-50/ImageNet scale. It is surpris-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                137,
+                                506,
+                                150
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        137,
+                                        506,
+                                        150
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ing that picking connections with the highest gradient magnitudes can be so detrimental to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                142,
+                                149,
+                                394,
+                                161
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        142,
+                                        149,
+                                        394,
+                                        161
+                                    ],
+                                    "score": 1.0,
+                                    "content": "training (it resulted in much worse than random performance).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        }
+                    ],
+                    "index": 3
+                }
+            ],
+            "page_idx": 18,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        107,
+                        27,
+                        308,
+                        37
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                25,
+                                308,
+                                38
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        25,
+                                        308,
+                                        38
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Under review as a conference paper at ICLR 2020",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        301,
+                        752,
+                        311,
+                        760
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                299,
+                                750,
+                                312,
+                                765
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        299,
+                                        750,
+                                        312,
+                                        765
+                                    ],
+                                    "score": 1.0,
+                                    "content": "",
+                                    "type": "text",
+                                    "height": 15,
+                                    "width": 13
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        131,
+                        82,
+                        505,
+                        160
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                128,
+                                81,
+                                506,
+                                96
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        128,
+                                        81,
+                                        506,
+                                        96
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3. SNIP Experiments. Our first implementation of SNIP used the gradient magnitudes to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                93,
+                                506,
+                                107
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        93,
+                                        506,
+                                        107
+                                    ],
+                                    "score": 1.0,
+                                    "content": "decide which connections to keep causing its performance to be worse than static. Upon",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                104,
+                                505,
+                                118
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        104,
+                                        505,
+                                        118
+                                    ],
+                                    "score": 1.0,
+                                    "content": "our discussions with the authors of SNIP, we realized that the correct metric is the saliency",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                115,
+                                505,
+                                128
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        115,
+                                        505,
+                                        128
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(gradient times parameter magnitude). With this correction SNIP performance improved",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                127,
+                                505,
+                                139
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        127,
+                                        505,
+                                        139
+                                    ],
+                                    "score": 1.0,
+                                    "content": "dramatically to better than random (Static) even at Resnet-50/ImageNet scale. It is surpris-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                137,
+                                506,
+                                150
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        137,
+                                        506,
+                                        150
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ing that picking connections with the highest gradient magnitudes can be so detrimental to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                142,
+                                149,
+                                394,
+                                161
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        142,
+                                        149,
+                                        394,
+                                        161
+                                    ],
+                                    "score": 1.0,
+                                    "content": "training (it resulted in much worse than random performance).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        }
+                    ],
+                    "index": 3,
+                    "bbox_fs": [
+                        128,
+                        81,
+                        506,
+                        161
+                    ]
+                }
+            ]
+        }
+    ],
+    "_backend": "pipeline",
+    "_version_name": "2.2.2"
+}
\ No newline at end of file
diff --git a/parse/train/ryg7vA4tPB/ryg7vA4tPB_model.json b/parse/train/ryg7vA4tPB/ryg7vA4tPB_model.json
new file mode 100644
index 0000000000000000000000000000000000000000..803c88fee78bcebca62a61bd8036935d5b677bb1
--- /dev/null
+++ b/parse/train/ryg7vA4tPB/ryg7vA4tPB_model.json
@@ -0,0 +1,25935 @@
+[
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    398,
+                    587,
+                    1302,
+                    587,
+                    1302,
+                    1139,
+                    398,
+                    1139
+                ],
+                "score": 0.983
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1712,
+                    1404,
+                    1712,
+                    1404,
+                    1957,
+                    298,
+                    1957
+                ],
+                "score": 0.977
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1270,
+                    1404,
+                    1270,
+                    1404,
+                    1485,
+                    298,
+                    1485
+                ],
+                "score": 0.973
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    298,
+                    220,
+                    984,
+                    220,
+                    984,
+                    324,
+                    298,
+                    324
+                ],
+                "score": 0.961
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    1971,
+                    1400,
+                    1971,
+                    1400,
+                    2034,
+                    299,
+                    2034
+                ],
+                "score": 0.937
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    360,
+                    1515,
+                    1403,
+                    1515,
+                    1403,
+                    1682,
+                    360,
+                    1682
+                ],
+                "score": 0.925
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    302,
+                    1200,
+                    573,
+                    1200,
+                    573,
+                    1235,
+                    302,
+                    1235
+                ],
+                "score": 0.897
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    298,
+                    75,
+                    858,
+                    75,
+                    858,
+                    105,
+                    298,
+                    105
+                ],
+                "score": 0.894
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    773,
+                    518,
+                    927,
+                    518,
+                    927,
+                    553,
+                    773,
+                    553
+                ],
+                "score": 0.882
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    313,
+                    376,
+                    681,
+                    376,
+                    681,
+                    437,
+                    313,
+                    437
+                ],
+                "score": 0.805
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    842,
+                    2088,
+                    857,
+                    2088,
+                    857,
+                    2112,
+                    842,
+                    2112
+                ],
+                "score": 0.641
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    217.0,
+                    798.0,
+                    217.0,
+                    798.0,
+                    270.0,
+                    295.0,
+                    270.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    271.0,
+                    987.0,
+                    271.0,
+                    987.0,
+                    327.0,
+                    294.0,
+                    327.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1197.0,
+                    579.0,
+                    1197.0,
+                    579.0,
+                    1244.0,
+                    294.0,
+                    1244.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    298.0,
+                    73.0,
+                    858.0,
+                    73.0,
+                    858.0,
+                    108.0,
+                    298.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    768.0,
+                    515.0,
+                    934.0,
+                    515.0,
+                    934.0,
+                    558.0,
+                    768.0,
+                    558.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    841.0,
+                    2088.0,
+                    858.0,
+                    2088.0,
+                    858.0,
+                    2116.0,
+                    841.0,
+                    2116.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    393.0,
+                    588.0,
+                    1305.0,
+                    588.0,
+                    1305.0,
+                    625.0,
+                    393.0,
+                    625.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    394.0,
+                    621.0,
+                    1304.0,
+                    621.0,
+                    1304.0,
+                    657.0,
+                    394.0,
+                    657.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    395.0,
+                    652.0,
+                    1305.0,
+                    652.0,
+                    1305.0,
+                    685.0,
+                    395.0,
+                    685.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    394.0,
+                    683.0,
+                    1305.0,
+                    683.0,
+                    1305.0,
+                    716.0,
+                    394.0,
+                    716.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    393.0,
+                    710.0,
+                    1305.0,
+                    710.0,
+                    1305.0,
+                    746.0,
+                    393.0,
+                    746.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    392.0,
+                    740.0,
+                    1306.0,
+                    740.0,
+                    1306.0,
+                    778.0,
+                    392.0,
+                    778.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    393.0,
+                    770.0,
+                    1307.0,
+                    770.0,
+                    1307.0,
+                    808.0,
+                    393.0,
+                    808.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    394.0,
+                    804.0,
+                    1306.0,
+                    804.0,
+                    1306.0,
+                    837.0,
+                    394.0,
+                    837.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    394.0,
+                    834.0,
+                    1305.0,
+                    834.0,
+                    1305.0,
+                    869.0,
+                    394.0,
+                    869.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    394.0,
+                    862.0,
+                    1303.0,
+                    862.0,
+                    1303.0,
+                    901.0,
+                    394.0,
+                    901.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    394.0,
+                    893.0,
+                    1306.0,
+                    893.0,
+                    1306.0,
+                    929.0,
+                    394.0,
+                    929.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    392.0,
+                    923.0,
+                    1307.0,
+                    923.0,
+                    1307.0,
+                    962.0,
+                    392.0,
+                    962.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    393.0,
+                    955.0,
+                    1307.0,
+                    955.0,
+                    1307.0,
+                    992.0,
+                    393.0,
+                    992.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    393.0,
+                    987.0,
+                    1306.0,
+                    987.0,
+                    1306.0,
+                    1017.0,
+                    393.0,
+                    1017.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    394.0,
+                    1016.0,
+                    1305.0,
+                    1016.0,
+                    1305.0,
+                    1050.0,
+                    394.0,
+                    1050.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    394.0,
+                    1045.0,
+                    1305.0,
+                    1045.0,
+                    1305.0,
+                    1082.0,
+                    394.0,
+                    1082.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    394.0,
+                    1077.0,
+                    1307.0,
+                    1077.0,
+                    1307.0,
+                    1114.0,
+                    394.0,
+                    1114.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    394.0,
+                    1107.0,
+                    839.0,
+                    1107.0,
+                    839.0,
+                    1142.0,
+                    394.0,
+                    1142.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1711.0,
+                    1404.0,
+                    1711.0,
+                    1404.0,
+                    1747.0,
+                    293.0,
+                    1747.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1743.0,
+                    1404.0,
+                    1743.0,
+                    1404.0,
+                    1776.0,
+                    293.0,
+                    1776.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1773.0,
+                    1405.0,
+                    1773.0,
+                    1405.0,
+                    1808.0,
+                    293.0,
+                    1808.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1800.0,
+                    1405.0,
+                    1800.0,
+                    1405.0,
+                    1842.0,
+                    292.0,
+                    1842.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1834.0,
+                    1404.0,
+                    1834.0,
+                    1404.0,
+                    1868.0,
+                    294.0,
+                    1868.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1863.0,
+                    1404.0,
+                    1863.0,
+                    1404.0,
+                    1902.0,
+                    291.0,
+                    1902.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1894.0,
+                    1404.0,
+                    1894.0,
+                    1404.0,
+                    1930.0,
+                    293.0,
+                    1930.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1924.0,
+                    899.0,
+                    1924.0,
+                    899.0,
+                    1963.0,
+                    293.0,
+                    1963.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1270.0,
+                    1406.0,
+                    1270.0,
+                    1406.0,
+                    1309.0,
+                    294.0,
+                    1309.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1299.0,
+                    1406.0,
+                    1299.0,
+                    1406.0,
+                    1337.0,
+                    293.0,
+                    1337.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1332.0,
+                    1405.0,
+                    1332.0,
+                    1405.0,
+                    1366.0,
+                    294.0,
+                    1366.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1363.0,
+                    1405.0,
+                    1363.0,
+                    1405.0,
+                    1397.0,
+                    293.0,
+                    1397.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1392.0,
+                    1406.0,
+                    1392.0,
+                    1406.0,
+                    1427.0,
+                    293.0,
+                    1427.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1424.0,
+                    1406.0,
+                    1424.0,
+                    1406.0,
+                    1459.0,
+                    294.0,
+                    1459.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1455.0,
+                    1376.0,
+                    1455.0,
+                    1376.0,
+                    1488.0,
+                    293.0,
+                    1488.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1971.0,
+                    1405.0,
+                    1971.0,
+                    1405.0,
+                    2008.0,
+                    295.0,
+                    2008.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    2003.0,
+                    1404.0,
+                    2003.0,
+                    1404.0,
+                    2038.0,
+                    295.0,
+                    2038.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    361.0,
+                    1514.0,
+                    1403.0,
+                    1514.0,
+                    1403.0,
+                    1550.0,
+                    361.0,
+                    1550.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    396.0,
+                    1547.0,
+                    1404.0,
+                    1547.0,
+                    1404.0,
+                    1581.0,
+                    396.0,
+                    1581.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    396.0,
+                    1578.0,
+                    1290.0,
+                    1578.0,
+                    1290.0,
+                    1612.0,
+                    396.0,
+                    1612.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    358.0,
+                    1616.0,
+                    1405.0,
+                    1616.0,
+                    1405.0,
+                    1656.0,
+                    358.0,
+                    1656.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    395.0,
+                    1650.0,
+                    943.0,
+                    1650.0,
+                    943.0,
+                    1686.0,
+                    395.0,
+                    1686.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    315.0,
+                    377.0,
+                    560.0,
+                    377.0,
+                    560.0,
+                    410.0,
+                    315.0,
+                    410.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    311.0,
+                    406.0,
+                    682.0,
+                    406.0,
+                    682.0,
+                    439.0,
+                    311.0,
+                    439.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 0,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    926,
+                    1404,
+                    926,
+                    1404,
+                    1202,
+                    298,
+                    1202
+                ],
+                "score": 0.981
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1217,
+                    1404,
+                    1217,
+                    1404,
+                    1461,
+                    298,
+                    1461
+                ],
+                "score": 0.977
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    653,
+                    1404,
+                    653,
+                    1404,
+                    811,
+                    299,
+                    811
+                ],
+                "score": 0.976
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1645,
+                    1404,
+                    1645,
+                    1404,
+                    1983,
+                    298,
+                    1983
+                ],
+                "score": 0.976
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    1477,
+                    1403,
+                    1477,
+                    1403,
+                    1631,
+                    299,
+                    1631
+                ],
+                "score": 0.973
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    368,
+                    353,
+                    1402,
+                    353,
+                    1402,
+                    623,
+                    368,
+                    623
+                ],
+                "score": 0.959
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    300,
+                    229,
+                    1402,
+                    229,
+                    1402,
+                    323,
+                    300,
+                    323
+                ],
+                "score": 0.954
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    301,
+                    855,
+                    587,
+                    855,
+                    587,
+                    891,
+                    301,
+                    891
+                ],
+                "score": 0.905
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    338,
+                    2006,
+                    579,
+                    2006,
+                    579,
+                    2034,
+                    338,
+                    2034
+                ],
+                "score": 0.895
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    298,
+                    75,
+                    858,
+                    75,
+                    858,
+                    104,
+                    298,
+                    104
+                ],
+                "score": 0.889
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    841,
+                    2088,
+                    859,
+                    2088,
+                    859,
+                    2112,
+                    841,
+                    2112
+                ],
+                "score": 0.725
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    822,
+                    742,
+                    942,
+                    742,
+                    942,
+                    784,
+                    822,
+                    784
+                ],
+                "score": 0.94,
+                "latex": "\\frac { 1 } { 1 - s p a r s i t y }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    396,
+                    1310,
+                    430,
+                    1310,
+                    430,
+                    1340,
+                    396,
+                    1340
+                ],
+                "score": 0.91,
+                "latex": "L _ { 0 }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1035,
+                    1079,
+                    1089,
+                    1079,
+                    1089,
+                    1109,
+                    1035,
+                    1109
+                ],
+                "score": 0.89,
+                "latex": "9 0 \\%"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    476,
+                    1079,
+                    530,
+                    1079,
+                    530,
+                    1109,
+                    476,
+                    1109
+                ],
+                "score": 0.88,
+                "latex": "2 0 \\%"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1013,
+                    1371,
+                    1049,
+                    1371,
+                    1049,
+                    1402,
+                    1013,
+                    1402
+                ],
+                "score": 0.87,
+                "latex": "L _ { 0 }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    796,
+                    293,
+                    866,
+                    293,
+                    866,
+                    323,
+                    796,
+                    323
+                ],
+                "score": 0.86,
+                "latex": "R i g L ^ { * }"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    852.0,
+                    594.0,
+                    852.0,
+                    594.0,
+                    899.0,
+                    291.0,
+                    899.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    336.0,
+                    2002.0,
+                    582.0,
+                    2002.0,
+                    582.0,
+                    2037.0,
+                    336.0,
+                    2037.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    72.0,
+                    858.0,
+                    72.0,
+                    858.0,
+                    108.0,
+                    297.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    838.0,
+                    2085.0,
+                    862.0,
+                    2085.0,
+                    862.0,
+                    2120.0,
+                    838.0,
+                    2120.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    926.0,
+                    1404.0,
+                    926.0,
+                    1404.0,
+                    961.0,
+                    293.0,
+                    961.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    957.0,
+                    1405.0,
+                    957.0,
+                    1405.0,
+                    994.0,
+                    294.0,
+                    994.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    988.0,
+                    1406.0,
+                    988.0,
+                    1406.0,
+                    1025.0,
+                    293.0,
+                    1025.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1020.0,
+                    1405.0,
+                    1020.0,
+                    1405.0,
+                    1052.0,
+                    294.0,
+                    1052.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1047.0,
+                    1406.0,
+                    1047.0,
+                    1406.0,
+                    1088.0,
+                    293.0,
+                    1088.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1077.0,
+                    475.0,
+                    1077.0,
+                    475.0,
+                    1116.0,
+                    293.0,
+                    1116.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    531.0,
+                    1077.0,
+                    1034.0,
+                    1077.0,
+                    1034.0,
+                    1116.0,
+                    531.0,
+                    1116.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1090.0,
+                    1077.0,
+                    1404.0,
+                    1077.0,
+                    1404.0,
+                    1116.0,
+                    1090.0,
+                    1116.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1110.0,
+                    1405.0,
+                    1110.0,
+                    1405.0,
+                    1146.0,
+                    294.0,
+                    1146.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1140.0,
+                    1405.0,
+                    1140.0,
+                    1405.0,
+                    1176.0,
+                    293.0,
+                    1176.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1171.0,
+                    817.0,
+                    1171.0,
+                    817.0,
+                    1206.0,
+                    293.0,
+                    1206.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1217.0,
+                    1405.0,
+                    1217.0,
+                    1405.0,
+                    1254.0,
+                    293.0,
+                    1254.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1249.0,
+                    1405.0,
+                    1249.0,
+                    1405.0,
+                    1283.0,
+                    293.0,
+                    1283.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1279.0,
+                    1404.0,
+                    1279.0,
+                    1404.0,
+                    1313.0,
+                    293.0,
+                    1313.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1309.0,
+                    395.0,
+                    1309.0,
+                    395.0,
+                    1344.0,
+                    293.0,
+                    1344.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    431.0,
+                    1309.0,
+                    1404.0,
+                    1309.0,
+                    1404.0,
+                    1344.0,
+                    431.0,
+                    1344.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1340.0,
+                    1402.0,
+                    1340.0,
+                    1402.0,
+                    1374.0,
+                    294.0,
+                    1374.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1369.0,
+                    1012.0,
+                    1369.0,
+                    1012.0,
+                    1406.0,
+                    292.0,
+                    1406.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1050.0,
+                    1369.0,
+                    1405.0,
+                    1369.0,
+                    1405.0,
+                    1406.0,
+                    1050.0,
+                    1406.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1399.0,
+                    1406.0,
+                    1399.0,
+                    1406.0,
+                    1437.0,
+                    291.0,
+                    1437.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1430.0,
+                    768.0,
+                    1430.0,
+                    768.0,
+                    1463.0,
+                    295.0,
+                    1463.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    649.0,
+                    1405.0,
+                    649.0,
+                    1405.0,
+                    690.0,
+                    294.0,
+                    690.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    684.0,
+                    1403.0,
+                    684.0,
+                    1403.0,
+                    717.0,
+                    295.0,
+                    717.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    712.0,
+                    1404.0,
+                    712.0,
+                    1404.0,
+                    749.0,
+                    293.0,
+                    749.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    290.0,
+                    733.0,
+                    821.0,
+                    733.0,
+                    821.0,
+                    794.0,
+                    290.0,
+                    794.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    943.0,
+                    733.0,
+                    1404.0,
+                    733.0,
+                    1404.0,
+                    794.0,
+                    943.0,
+                    794.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    779.0,
+                    683.0,
+                    779.0,
+                    683.0,
+                    817.0,
+                    294.0,
+                    817.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1645.0,
+                    1405.0,
+                    1645.0,
+                    1405.0,
+                    1683.0,
+                    294.0,
+                    1683.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1678.0,
+                    1406.0,
+                    1678.0,
+                    1406.0,
+                    1713.0,
+                    294.0,
+                    1713.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1704.0,
+                    1406.0,
+                    1704.0,
+                    1406.0,
+                    1744.0,
+                    293.0,
+                    1744.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1737.0,
+                    1405.0,
+                    1737.0,
+                    1405.0,
+                    1774.0,
+                    293.0,
+                    1774.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1765.0,
+                    1405.0,
+                    1765.0,
+                    1405.0,
+                    1805.0,
+                    292.0,
+                    1805.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1799.0,
+                    1405.0,
+                    1799.0,
+                    1405.0,
+                    1834.0,
+                    294.0,
+                    1834.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1829.0,
+                    1405.0,
+                    1829.0,
+                    1405.0,
+                    1864.0,
+                    293.0,
+                    1864.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1860.0,
+                    1404.0,
+                    1860.0,
+                    1404.0,
+                    1895.0,
+                    293.0,
+                    1895.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1891.0,
+                    1404.0,
+                    1891.0,
+                    1404.0,
+                    1926.0,
+                    294.0,
+                    1926.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1921.0,
+                    1405.0,
+                    1921.0,
+                    1405.0,
+                    1956.0,
+                    294.0,
+                    1956.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1953.0,
+                    1406.0,
+                    1953.0,
+                    1406.0,
+                    1986.0,
+                    293.0,
+                    1986.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1475.0,
+                    1404.0,
+                    1475.0,
+                    1404.0,
+                    1516.0,
+                    294.0,
+                    1516.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1506.0,
+                    1403.0,
+                    1506.0,
+                    1403.0,
+                    1543.0,
+                    295.0,
+                    1543.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1535.0,
+                    1405.0,
+                    1535.0,
+                    1405.0,
+                    1575.0,
+                    293.0,
+                    1575.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1569.0,
+                    1405.0,
+                    1569.0,
+                    1405.0,
+                    1606.0,
+                    294.0,
+                    1606.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1599.0,
+                    1331.0,
+                    1599.0,
+                    1331.0,
+                    1635.0,
+                    293.0,
+                    1635.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    368.0,
+                    353.0,
+                    1404.0,
+                    353.0,
+                    1404.0,
+                    389.0,
+                    368.0,
+                    389.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    392.0,
+                    384.0,
+                    1405.0,
+                    384.0,
+                    1405.0,
+                    416.0,
+                    392.0,
+                    416.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    394.0,
+                    415.0,
+                    1405.0,
+                    415.0,
+                    1405.0,
+                    450.0,
+                    394.0,
+                    450.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    395.0,
+                    445.0,
+                    750.0,
+                    445.0,
+                    750.0,
+                    478.0,
+                    395.0,
+                    478.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    374.0,
+                    486.0,
+                    1404.0,
+                    486.0,
+                    1404.0,
+                    523.0,
+                    374.0,
+                    523.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    394.0,
+                    519.0,
+                    1069.0,
+                    519.0,
+                    1069.0,
+                    550.0,
+                    394.0,
+                    550.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    368.0,
+                    561.0,
+                    1404.0,
+                    561.0,
+                    1404.0,
+                    593.0,
+                    368.0,
+                    593.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    395.0,
+                    593.0,
+                    873.0,
+                    593.0,
+                    873.0,
+                    626.0,
+                    395.0,
+                    626.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    231.0,
+                    1406.0,
+                    231.0,
+                    1406.0,
+                    265.0,
+                    296.0,
+                    265.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    259.0,
+                    1407.0,
+                    259.0,
+                    1407.0,
+                    298.0,
+                    293.0,
+                    298.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    289.0,
+                    795.0,
+                    289.0,
+                    795.0,
+                    326.0,
+                    295.0,
+                    326.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    867.0,
+                    289.0,
+                    1205.0,
+                    289.0,
+                    1205.0,
+                    326.0,
+                    867.0,
+                    326.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 1,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    915,
+                    1404,
+                    915,
+                    1404,
+                    1161,
+                    297,
+                    1161
+                ],
+                "score": 0.98
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1286,
+                    1404,
+                    1286,
+                    1404,
+                    1563,
+                    297,
+                    1563
+                ],
+                "score": 0.98
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    623,
+                    1404,
+                    623,
+                    1404,
+                    901,
+                    298,
+                    901
+                ],
+                "score": 0.979
+            },
+            {
+                "category_id": 5,
+                "poly": [
+                    321,
+                    221,
+                    1379,
+                    221,
+                    1379,
+                    446,
+                    321,
+                    446
+                ],
+                "score": 0.975,
+                "html": "<table><tr><td>Method</td><td>Drop</td><td>Grow</td><td>Selectable FLOPs</td><td>Space &amp;FLOPs X</td></tr><tr><td>SNIP</td><td>min(10 * VL(0)I)</td><td>none</td><td>yes</td><td>sparse</td></tr><tr><td>DeepR</td><td>stochastic</td><td>random</td><td>yes</td><td>sparse</td></tr><tr><td>SET</td><td>min(l0))</td><td>random</td><td>yes</td><td>sparse</td></tr><tr><td>DSR</td><td>min(le)</td><td>random</td><td>no</td><td>sparse</td></tr><tr><td>SNFS</td><td>min(l0))</td><td>momentum</td><td>no</td><td>dense</td></tr><tr><td>RigL (ours)</td><td>min(l0)</td><td>gradient</td><td>yes</td><td>sparse</td></tr></table>"
+            },
+            {
+                "category_id": 3,
+                "poly": [
+                    349,
+                    1707,
+                    1352,
+                    1707,
+                    1352,
+                    1981,
+                    349,
+                    1981
+                ],
+                "score": 0.969
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    294,
+                    1576,
+                    1402,
+                    1576,
+                    1402,
+                    1643,
+                    294,
+                    1643
+                ],
+                "score": 0.951
+            },
+            {
+                "category_id": 4,
+                "poly": [
+                    294,
+                    2003,
+                    1404,
+                    2003,
+                    1404,
+                    2066,
+                    294,
+                    2066
+                ],
+                "score": 0.93
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    298,
+                    75,
+                    858,
+                    75,
+                    858,
+                    105,
+                    298,
+                    105
+                ],
+                "score": 0.91
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    302,
+                    1213,
+                    689,
+                    1213,
+                    689,
+                    1249,
+                    302,
+                    1249
+                ],
+                "score": 0.901
+            },
+            {
+                "category_id": 6,
+                "poly": [
+                    299,
+                    466,
+                    1405,
+                    466,
+                    1405,
+                    559,
+                    299,
+                    559
+                ],
+                "score": 0.656
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    841,
+                    2088,
+                    859,
+                    2088,
+                    859,
+                    2112,
+                    841,
+                    2112
+                ],
+                "score": 0.615
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    841,
+                    2088,
+                    859,
+                    2088,
+                    859,
+                    2112,
+                    841,
+                    2112
+                ],
+                "score": 0.557
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    740,
+                    1608,
+                    933,
+                    1608,
+                    933,
+                    1643,
+                    740,
+                    1643
+                ],
+                "score": 0.93,
+                "latex": "\\textstyle \\sum _ { i } L ( f _ { \\theta } ( x _ { i } ) , y _ { i } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1025,
+                    1609,
+                    1092,
+                    1609,
+                    1092,
+                    1642,
+                    1025,
+                    1642
+                ],
+                "score": 0.92,
+                "latex": "f _ { \\boldsymbol { \\theta } } ( \\boldsymbol { x } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    941,
+                    1583,
+                    969,
+                    1583,
+                    969,
+                    1609,
+                    941,
+                    1609
+                ],
+                "score": 0.85,
+                "latex": "x _ { i }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    503,
+                    256,
+                    723,
+                    256,
+                    723,
+                    289,
+                    503,
+                    289
+                ],
+                "score": 0.85,
+                "latex": "m i n ( | \\theta * \\nabla _ { \\theta } L ( \\theta ) | )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1107,
+                    1584,
+                    1133,
+                    1584,
+                    1133,
+                    1609,
+                    1107,
+                    1609
+                ],
+                "score": 0.84,
+                "latex": "y _ { i }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    651,
+                    1579,
+                    678,
+                    1579,
+                    678,
+                    1605,
+                    651,
+                    1605
+                ],
+                "score": 0.76,
+                "latex": "D"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    561,
+                    350,
+                    664,
+                    350,
+                    664,
+                    380,
+                    561,
+                    380
+                ],
+                "score": 0.66,
+                "latex": "m i n ( | \\theta | )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    561,
+                    410,
+                    664,
+                    410,
+                    664,
+                    441,
+                    561,
+                    441
+                ],
+                "score": 0.59,
+                "latex": "m i n ( | \\theta | )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    561,
+                    380,
+                    664,
+                    380,
+                    664,
+                    410,
+                    561,
+                    410
+                ],
+                "score": 0.56,
+                "latex": "m i n ( | \\theta | )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    560,
+                    317,
+                    664,
+                    317,
+                    664,
+                    349,
+                    560,
+                    349
+                ],
+                "score": 0.51,
+                "latex": "m i n ( | \\theta | )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    738,
+                    1441,
+                    798,
+                    1441,
+                    798,
+                    1471,
+                    738,
+                    1471
+                ],
+                "score": 0.43,
+                "latex": "R i g L"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    908,
+                    498,
+                    994,
+                    498,
+                    994,
+                    526,
+                    908,
+                    526
+                ],
+                "score": 0.41,
+                "latex": "F L O P s"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    991.0,
+                    1705.0,
+                    1069.0,
+                    1705.0,
+                    1069.0,
+                    1739.0,
+                    991.0,
+                    1739.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1183.0,
+                    1707.0,
+                    1264.0,
+                    1707.0,
+                    1264.0,
+                    1738.0,
+                    1183.0,
+                    1738.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    806.0,
+                    1735.0,
+                    902.0,
+                    1735.0,
+                    902.0,
+                    1766.0,
+                    806.0,
+                    1766.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    810.0,
+                    1756.0,
+                    897.0,
+                    1756.0,
+                    897.0,
+                    1787.0,
+                    810.0,
+                    1787.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    371.0,
+                    1792.0,
+                    473.0,
+                    1792.0,
+                    473.0,
+                    1827.0,
+                    371.0,
+                    1827.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    494.0,
+                    1797.0,
+                    579.0,
+                    1797.0,
+                    579.0,
+                    1821.0,
+                    494.0,
+                    1821.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    821.0,
+                    1794.0,
+                    905.0,
+                    1794.0,
+                    905.0,
+                    1821.0,
+                    821.0,
+                    1821.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    370.0,
+                    1818.0,
+                    476.0,
+                    1818.0,
+                    476.0,
+                    1845.0,
+                    370.0,
+                    1845.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    822.0,
+                    1815.0,
+                    903.0,
+                    1815.0,
+                    903.0,
+                    1842.0,
+                    822.0,
+                    1842.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    924.0,
+                    1826.0,
+                    960.0,
+                    1826.0,
+                    960.0,
+                    1846.0,
+                    924.0,
+                    1846.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    867.0,
+                    1862.0,
+                    899.0,
+                    1862.0,
+                    899.0,
+                    1887.0,
+                    867.0,
+                    1887.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    760.0,
+                    1880.0,
+                    825.0,
+                    1880.0,
+                    825.0,
+                    1907.0,
+                    760.0,
+                    1907.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    759.0,
+                    1899.0,
+                    831.0,
+                    1899.0,
+                    831.0,
+                    1930.0,
+                    759.0,
+                    1930.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    758.0,
+                    1918.5,
+                    807.0,
+                    1918.5,
+                    807.0,
+                    1951.5,
+                    758.0,
+                    1951.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1999.0,
+                    1402.0,
+                    1999.0,
+                    1402.0,
+                    2041.0,
+                    292.0,
+                    2041.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    2032.0,
+                    359.0,
+                    2032.0,
+                    359.0,
+                    2068.0,
+                    293.0,
+                    2068.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    298.0,
+                    73.0,
+                    858.0,
+                    73.0,
+                    858.0,
+                    108.0,
+                    298.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1208.0,
+                    696.0,
+                    1208.0,
+                    696.0,
+                    1257.0,
+                    291.0,
+                    1257.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    463.0,
+                    1405.0,
+                    463.0,
+                    1405.0,
+                    503.0,
+                    293.0,
+                    503.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    497.0,
+                    907.0,
+                    497.0,
+                    907.0,
+                    531.0,
+                    295.0,
+                    531.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    995.0,
+                    497.0,
+                    1403.0,
+                    497.0,
+                    1403.0,
+                    531.0,
+                    995.0,
+                    531.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    526.0,
+                    775.0,
+                    526.0,
+                    775.0,
+                    563.0,
+                    294.0,
+                    563.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    838.0,
+                    2085.0,
+                    862.0,
+                    2085.0,
+                    862.0,
+                    2118.0,
+                    838.0,
+                    2118.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    838.0,
+                    2085.0,
+                    862.0,
+                    2085.0,
+                    862.0,
+                    2118.0,
+                    838.0,
+                    2118.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    914.0,
+                    1404.0,
+                    914.0,
+                    1404.0,
+                    950.0,
+                    294.0,
+                    950.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    944.0,
+                    1405.0,
+                    944.0,
+                    1405.0,
+                    980.0,
+                    294.0,
+                    980.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    976.0,
+                    1404.0,
+                    976.0,
+                    1404.0,
+                    1012.0,
+                    292.0,
+                    1012.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1008.0,
+                    1404.0,
+                    1008.0,
+                    1404.0,
+                    1042.0,
+                    294.0,
+                    1042.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1038.0,
+                    1402.0,
+                    1038.0,
+                    1402.0,
+                    1072.0,
+                    295.0,
+                    1072.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1066.0,
+                    1405.0,
+                    1066.0,
+                    1405.0,
+                    1107.0,
+                    292.0,
+                    1107.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1099.0,
+                    1405.0,
+                    1099.0,
+                    1405.0,
+                    1133.0,
+                    294.0,
+                    1133.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1130.0,
+                    725.0,
+                    1130.0,
+                    725.0,
+                    1163.0,
+                    294.0,
+                    1163.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1288.0,
+                    1405.0,
+                    1288.0,
+                    1405.0,
+                    1321.0,
+                    296.0,
+                    1321.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1316.0,
+                    1405.0,
+                    1316.0,
+                    1405.0,
+                    1354.0,
+                    292.0,
+                    1354.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1347.0,
+                    1407.0,
+                    1347.0,
+                    1407.0,
+                    1385.0,
+                    291.0,
+                    1385.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1380.0,
+                    1404.0,
+                    1380.0,
+                    1404.0,
+                    1413.0,
+                    295.0,
+                    1413.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1407.0,
+                    1405.0,
+                    1407.0,
+                    1405.0,
+                    1444.0,
+                    294.0,
+                    1444.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1439.0,
+                    737.0,
+                    1439.0,
+                    737.0,
+                    1476.0,
+                    294.0,
+                    1476.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    799.0,
+                    1439.0,
+                    1406.0,
+                    1439.0,
+                    1406.0,
+                    1476.0,
+                    799.0,
+                    1476.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1469.0,
+                    1406.0,
+                    1469.0,
+                    1406.0,
+                    1503.0,
+                    292.0,
+                    1503.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1500.0,
+                    1405.0,
+                    1500.0,
+                    1405.0,
+                    1537.0,
+                    292.0,
+                    1537.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1529.0,
+                    381.0,
+                    1529.0,
+                    381.0,
+                    1572.0,
+                    291.0,
+                    1572.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    623.0,
+                    1404.0,
+                    623.0,
+                    1404.0,
+                    662.0,
+                    293.0,
+                    662.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    652.0,
+                    1405.0,
+                    652.0,
+                    1405.0,
+                    696.0,
+                    292.0,
+                    696.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    683.0,
+                    1405.0,
+                    683.0,
+                    1405.0,
+                    724.0,
+                    293.0,
+                    724.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    719.0,
+                    1402.0,
+                    719.0,
+                    1402.0,
+                    751.0,
+                    294.0,
+                    751.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    748.0,
+                    1405.0,
+                    748.0,
+                    1405.0,
+                    784.0,
+                    294.0,
+                    784.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    776.0,
+                    1405.0,
+                    776.0,
+                    1405.0,
+                    812.0,
+                    293.0,
+                    812.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    808.0,
+                    1405.0,
+                    808.0,
+                    1405.0,
+                    845.0,
+                    293.0,
+                    845.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    838.0,
+                    1405.0,
+                    838.0,
+                    1405.0,
+                    875.0,
+                    293.0,
+                    875.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    871.0,
+                    899.0,
+                    871.0,
+                    899.0,
+                    904.0,
+                    294.0,
+                    904.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1574.0,
+                    650.0,
+                    1574.0,
+                    650.0,
+                    1613.0,
+                    294.0,
+                    1613.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    679.0,
+                    1574.0,
+                    940.0,
+                    1574.0,
+                    940.0,
+                    1613.0,
+                    679.0,
+                    1613.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    970.0,
+                    1574.0,
+                    1106.0,
+                    1574.0,
+                    1106.0,
+                    1613.0,
+                    970.0,
+                    1613.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1134.0,
+                    1574.0,
+                    1404.0,
+                    1574.0,
+                    1404.0,
+                    1613.0,
+                    1134.0,
+                    1613.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1607.0,
+                    739.0,
+                    1607.0,
+                    739.0,
+                    1643.0,
+                    293.0,
+                    1643.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    934.0,
+                    1607.0,
+                    1024.0,
+                    1607.0,
+                    1024.0,
+                    1643.0,
+                    934.0,
+                    1643.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1093.0,
+                    1607.0,
+                    1404.0,
+                    1607.0,
+                    1404.0,
+                    1643.0,
+                    1093.0,
+                    1643.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 2,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    999,
+                    1405,
+                    999,
+                    1405,
+                    1183,
+                    297,
+                    1183
+                ],
+                "score": 0.972
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    228,
+                    1403,
+                    228,
+                    1403,
+                    356,
+                    298,
+                    356
+                ],
+                "score": 0.968
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    367,
+                    1405,
+                    367,
+                    1405,
+                    493,
+                    299,
+                    493
+                ],
+                "score": 0.957
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1488,
+                    1404,
+                    1488,
+                    1404,
+                    1774,
+                    297,
+                    1774
+                ],
+                "score": 0.955
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    301,
+                    1274,
+                    1403,
+                    1274,
+                    1403,
+                    1337,
+                    301,
+                    1337
+                ],
+                "score": 0.946
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    646,
+                    1191,
+                    1051,
+                    1191,
+                    1051,
+                    1265,
+                    646,
+                    1265
+                ],
+                "score": 0.94
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    359,
+                    518,
+                    1404,
+                    518,
+                    1404,
+                    927,
+                    359,
+                    927
+                ],
+                "score": 0.933
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    953,
+                    1040,
+                    953,
+                    1040,
+                    986,
+                    297,
+                    986
+                ],
+                "score": 0.92
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    298,
+                    75,
+                    858,
+                    75,
+                    858,
+                    105,
+                    298,
+                    105
+                ],
+                "score": 0.911
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    356,
+                    1796,
+                    1406,
+                    1796,
+                    1406,
+                    1984,
+                    356,
+                    1984
+                ],
+                "score": 0.9
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    300,
+                    1350,
+                    1401,
+                    1350,
+                    1401,
+                    1442,
+                    300,
+                    1442
+                ],
+                "score": 0.873
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    339,
+                    2004,
+                    1118,
+                    2004,
+                    1118,
+                    2035,
+                    339,
+                    2035
+                ],
+                "score": 0.856
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    841,
+                    2089,
+                    858,
+                    2089,
+                    858,
+                    2111,
+                    841,
+                    2111
+                ],
+                "score": 0.773
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    300,
+                    1443,
+                    712,
+                    1443,
+                    712,
+                    1478,
+                    300,
+                    1478
+                ],
+                "score": 0.293
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    300,
+                    1443,
+                    712,
+                    1443,
+                    712,
+                    1478,
+                    300,
+                    1478
+                ],
+                "score": 0.141
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    646,
+                    1189,
+                    1054,
+                    1189,
+                    1054,
+                    1266,
+                    646,
+                    1266
+                ],
+                "score": 0.94,
+                "latex": "f _ { d e c a y } ( t ) = \\frac { \\alpha } { 2 } \\left( 1 + c o s \\left( \\frac { t \\pi } { T _ { e n d } } \\right) \\right)"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1182,
+                    760,
+                    1394,
+                    760,
+                    1394,
+                    803,
+                    1182,
+                    803
+                ],
+                "score": 0.93,
+                "latex": "\\begin{array} { r } { 1 - \\frac { n ^ { l - 1 } + n ^ { l } + w ^ { l } + h ^ { l } } { n ^ { l - 1 } * n ^ { l } * w ^ { l } * h ^ { l } } } \\end{array}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    901,
+                    261,
+                    1027,
+                    261,
+                    1027,
+                    295,
+                    901,
+                    295
+                ],
+                "score": 0.92,
+                "latex": "D \\in ( 0 , 1 )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1147,
+                    589,
+                    1287,
+                    589,
+                    1287,
+                    634,
+                    1147,
+                    634
+                ],
+                "score": 0.92,
+                "latex": "\\begin{array} { r } { 1 - \\frac { n ^ { l - 1 } + n ^ { l } } { n ^ { l - 1 } * n ^ { l } } } \\end{array}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    580,
+                    1798,
+                    693,
+                    1798,
+                    693,
+                    1833,
+                    580,
+                    1833
+                ],
+                "score": 0.92,
+                "latex": "T K = \\{ \\}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    373,
+                    1556,
+                    451,
+                    1556,
+                    451,
+                    1594,
+                    373,
+                    1594
+                ],
+                "score": 0.92,
+                "latex": "\\theta _ { a c t i v e } ^ { l }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1152,
+                    1841,
+                    1239,
+                    1841,
+                    1239,
+                    1873,
+                    1152,
+                    1873
+                ],
+                "score": 0.92,
+                "latex": "G _ { i : i + M }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    898,
+                    319,
+                    1240,
+                    319,
+                    1240,
+                    357,
+                    898,
+                    357
+                ],
+                "score": 0.92,
+                "latex": "\\begin{array} { r } { \\dot { \\sum _ { l } } ( 1 - s ^ { l } ) N ^ { l } = ( 1 - \\bar { S } ) * N } \\end{array}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    908,
+                    1881,
+                    995,
+                    1881,
+                    995,
+                    1913,
+                    908,
+                    1913
+                ],
+                "score": 0.92,
+                "latex": "G _ { i : i + M }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    429,
+                    292,
+                    561,
+                    292,
+                    561,
+                    320,
+                    429,
+                    320
+                ],
+                "score": 0.91,
+                "latex": "S = 1 - D"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    769,
+                    1092,
+                    825,
+                    1092,
+                    825,
+                    1123,
+                    769,
+                    1123
+                ],
+                "score": 0.9,
+                "latex": "T _ { e n d }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    561,
+                    1062,
+                    632,
+                    1062,
+                    632,
+                    1092,
+                    561,
+                    1092
+                ],
+                "score": 0.89,
+                "latex": "( T _ { e n d } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    850,
+                    1518,
+                    1396,
+                    1518,
+                    1396,
+                    1560,
+                    850,
+                    1560
+                ],
+                "score": 0.89,
+                "latex": "T o p K _ { w \\notin \\theta _ { a c t i v e } ^ { l } } \\breve { ( | { g r a d ( \\theta ^ { l } ) } | } , f _ { d e c a y } ( t ) ( 1 - s ^ { l } ) \\breve { N } ^ { l } ) ."
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1315,
+                    228,
+                    1352,
+                    228,
+                    1352,
+                    259,
+                    1315,
+                    259
+                ],
+                "score": 0.88,
+                "latex": "N ^ { l }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    470,
+                    801,
+                    502,
+                    801,
+                    502,
+                    832,
+                    470,
+                    832
+                ],
+                "score": 0.88,
+                "latex": "w ^ { l }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    687,
+                    548,
+                    770,
+                    548,
+                    770,
+                    579,
+                    687,
+                    579
+                ],
+                "score": 0.88,
+                "latex": "\\boldsymbol s ^ { 0 } = 0 ,"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    338,
+                    2001,
+                    476,
+                    2001,
+                    476,
+                    2035,
+                    338,
+                    2035
+                ],
+                "score": 0.87,
+                "latex": "^ { \\ddagger } T o p K ( v , k )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    550,
+                    801,
+                    577,
+                    801,
+                    577,
+                    832,
+                    550,
+                    832
+                ],
+                "score": 0.87,
+                "latex": "h ^ { l }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1167,
+                    228,
+                    1192,
+                    228,
+                    1192,
+                    259,
+                    1167,
+                    259
+                ],
+                "score": 0.86,
+                "latex": "\\theta ^ { l }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    548,
+                    1094,
+                    594,
+                    1094,
+                    594,
+                    1120,
+                    548,
+                    1120
+                ],
+                "score": 0.86,
+                "latex": "\\Delta T"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    658,
+                    517,
+                    682,
+                    517,
+                    682,
+                    547,
+                    658,
+                    547
+                ],
+                "score": 0.86,
+                "latex": "s ^ { l }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    297,
+                    1440,
+                    709,
+                    1440,
+                    709,
+                    1479,
+                    297,
+                    1479
+                ],
+                "score": 0.86,
+                "latex": "T o p K ( - | \\theta ^ { l } | , f _ { d e c a y } ( t ) \\dot { ( } 1 - s ^ { l } ) N ^ { l } ) ^ { \\ddag } ."
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1272,
+                    290,
+                    1295,
+                    290,
+                    1295,
+                    319,
+                    1272,
+                    319
+                ],
+                "score": 0.85,
+                "latex": "s ^ { l }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1372,
+                    596,
+                    1400,
+                    596,
+                    1400,
+                    625,
+                    1372,
+                    625
+                ],
+                "score": 0.84,
+                "latex": "n ^ { l }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    988,
+                    594,
+                    1013,
+                    594,
+                    1013,
+                    626,
+                    988,
+                    626
+                ],
+                "score": 0.83,
+                "latex": "s ^ { l }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    884,
+                    2007,
+                    900,
+                    2007,
+                    900,
+                    2030,
+                    884,
+                    2030
+                ],
+                "score": 0.82,
+                "latex": "k"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    428,
+                    232,
+                    445,
+                    232,
+                    445,
+                    258,
+                    428,
+                    258
+                ],
+                "score": 0.81,
+                "latex": "\\theta"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    855,
+                    1031,
+                    918,
+                    1031,
+                    918,
+                    1062,
+                    855,
+                    1062
+                ],
+                "score": 0.8,
+                "latex": "( \\Delta T )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1207,
+                    1880,
+                    1256,
+                    1880,
+                    1256,
+                    1909,
+                    1207,
+                    1909
+                ],
+                "score": 0.8,
+                "latex": "T K"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1323,
+                    521,
+                    1344,
+                    521,
+                    1344,
+                    547,
+                    1323,
+                    547
+                ],
+                "score": 0.78,
+                "latex": "S"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    734,
+                    232,
+                    753,
+                    232,
+                    753,
+                    258,
+                    734,
+                    258
+                ],
+                "score": 0.78,
+                "latex": "\\theta"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1352,
+                    1921,
+                    1402,
+                    1921,
+                    1402,
+                    1950,
+                    1352,
+                    1950
+                ],
+                "score": 0.74,
+                "latex": "T K"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1094,
+                    2013,
+                    1109,
+                    2013,
+                    1109,
+                    2029,
+                    1094,
+                    2029
+                ],
+                "score": 0.73,
+                "latex": "v"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    559,
+                    232,
+                    588,
+                    232,
+                    588,
+                    259,
+                    559,
+                    259
+                ],
+                "score": 0.73,
+                "latex": "N"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1156,
+                    1063,
+                    1195,
+                    1063,
+                    1195,
+                    1092,
+                    1156,
+                    1092
+                ],
+                "score": 0.73,
+                "latex": "( \\alpha )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    484,
+                    1880,
+                    533,
+                    1880,
+                    533,
+                    1909,
+                    484,
+                    1909
+                ],
+                "score": 0.7,
+                "latex": "T K"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    298,
+                    1092,
+                    371,
+                    1092,
+                    371,
+                    1126,
+                    298,
+                    1126
+                ],
+                "score": 0.59,
+                "latex": "f _ { d e c a y }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    421,
+                    263,
+                    432,
+                    263,
+                    432,
+                    289,
+                    421,
+                    289
+                ],
+                "score": 0.57,
+                "latex": "l"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    954,
+                    805,
+                    971,
+                    805,
+                    971,
+                    832,
+                    954,
+                    832
+                ],
+                "score": 0.57,
+                "latex": "\\mathbf { \\nabla } _ { l } ,"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    746,
+                    1881,
+                    770,
+                    1881,
+                    770,
+                    1909,
+                    746,
+                    1909
+                ],
+                "score": 0.25,
+                "latex": "\\mathbf { \\nabla } \\cdot \\mathbf { K }"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    298.0,
+                    73.0,
+                    858.0,
+                    73.0,
+                    858.0,
+                    108.0,
+                    298.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    334.0,
+                    1998.0,
+                    337.0,
+                    1998.0,
+                    337.0,
+                    2040.0,
+                    334.0,
+                    2040.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    477.0,
+                    1998.0,
+                    883.0,
+                    1998.0,
+                    883.0,
+                    2040.0,
+                    477.0,
+                    2040.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    901.0,
+                    1998.0,
+                    1093.0,
+                    1998.0,
+                    1093.0,
+                    2040.0,
+                    901.0,
+                    2040.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1110.0,
+                    1998.0,
+                    1121.0,
+                    1998.0,
+                    1121.0,
+                    2040.0,
+                    1110.0,
+                    2040.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    838.0,
+                    2086.0,
+                    862.0,
+                    2086.0,
+                    862.0,
+                    2118.0,
+                    838.0,
+                    2118.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1000.0,
+                    1405.0,
+                    1000.0,
+                    1405.0,
+                    1035.0,
+                    295.0,
+                    1035.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1026.0,
+                    854.0,
+                    1026.0,
+                    854.0,
+                    1068.0,
+                    292.0,
+                    1068.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    919.0,
+                    1026.0,
+                    1406.0,
+                    1026.0,
+                    1406.0,
+                    1068.0,
+                    919.0,
+                    1068.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1062.0,
+                    560.0,
+                    1062.0,
+                    560.0,
+                    1097.0,
+                    295.0,
+                    1097.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    633.0,
+                    1062.0,
+                    1155.0,
+                    1062.0,
+                    1155.0,
+                    1097.0,
+                    633.0,
+                    1097.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1196.0,
+                    1062.0,
+                    1405.0,
+                    1062.0,
+                    1405.0,
+                    1097.0,
+                    1196.0,
+                    1097.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1091.0,
+                    297.0,
+                    1091.0,
+                    297.0,
+                    1128.0,
+                    292.0,
+                    1128.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    372.0,
+                    1091.0,
+                    547.0,
+                    1091.0,
+                    547.0,
+                    1128.0,
+                    372.0,
+                    1128.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    595.0,
+                    1091.0,
+                    768.0,
+                    1091.0,
+                    768.0,
+                    1128.0,
+                    595.0,
+                    1128.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    826.0,
+                    1091.0,
+                    1405.0,
+                    1091.0,
+                    1405.0,
+                    1128.0,
+                    826.0,
+                    1128.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1123.0,
+                    1403.0,
+                    1123.0,
+                    1403.0,
+                    1158.0,
+                    294.0,
+                    1158.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1153.0,
+                    636.0,
+                    1153.0,
+                    636.0,
+                    1185.0,
+                    296.0,
+                    1185.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    229.0,
+                    427.0,
+                    229.0,
+                    427.0,
+                    263.0,
+                    292.0,
+                    263.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    446.0,
+                    229.0,
+                    558.0,
+                    229.0,
+                    558.0,
+                    263.0,
+                    446.0,
+                    263.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    589.0,
+                    229.0,
+                    733.0,
+                    229.0,
+                    733.0,
+                    263.0,
+                    589.0,
+                    263.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    754.0,
+                    229.0,
+                    1166.0,
+                    229.0,
+                    1166.0,
+                    263.0,
+                    754.0,
+                    263.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1193.0,
+                    229.0,
+                    1314.0,
+                    229.0,
+                    1314.0,
+                    263.0,
+                    1193.0,
+                    263.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1353.0,
+                    229.0,
+                    1405.0,
+                    229.0,
+                    1405.0,
+                    263.0,
+                    1353.0,
+                    263.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    260.0,
+                    420.0,
+                    260.0,
+                    420.0,
+                    297.0,
+                    293.0,
+                    297.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    433.0,
+                    260.0,
+                    900.0,
+                    260.0,
+                    900.0,
+                    297.0,
+                    433.0,
+                    297.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1028.0,
+                    260.0,
+                    1408.0,
+                    260.0,
+                    1408.0,
+                    297.0,
+                    1028.0,
+                    297.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    291.0,
+                    428.0,
+                    291.0,
+                    428.0,
+                    326.0,
+                    292.0,
+                    326.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    562.0,
+                    291.0,
+                    1271.0,
+                    291.0,
+                    1271.0,
+                    326.0,
+                    562.0,
+                    326.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1296.0,
+                    291.0,
+                    1405.0,
+                    291.0,
+                    1405.0,
+                    326.0,
+                    1296.0,
+                    326.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    322.0,
+                    897.0,
+                    322.0,
+                    897.0,
+                    356.0,
+                    293.0,
+                    356.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1241.0,
+                    322.0,
+                    1251.0,
+                    322.0,
+                    1251.0,
+                    356.0,
+                    1241.0,
+                    356.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    368.0,
+                    1403.0,
+                    368.0,
+                    1403.0,
+                    404.0,
+                    294.0,
+                    404.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    397.0,
+                    1405.0,
+                    397.0,
+                    1405.0,
+                    437.0,
+                    293.0,
+                    437.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    430.0,
+                    1406.0,
+                    430.0,
+                    1406.0,
+                    466.0,
+                    294.0,
+                    466.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    459.0,
+                    1143.0,
+                    459.0,
+                    1143.0,
+                    497.0,
+                    293.0,
+                    497.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1487.0,
+                    1405.0,
+                    1487.0,
+                    1405.0,
+                    1526.0,
+                    294.0,
+                    1526.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    290.0,
+                    1510.0,
+                    849.0,
+                    1510.0,
+                    849.0,
+                    1565.0,
+                    290.0,
+                    1565.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1397.0,
+                    1518.0,
+                    1405.0,
+                    1518.0,
+                    1405.0,
+                    1557.0,
+                    1397.0,
+                    1557.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1556.0,
+                    372.0,
+                    1556.0,
+                    372.0,
+                    1594.0,
+                    295.0,
+                    1594.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    452.0,
+                    1560.0,
+                    1405.0,
+                    1560.0,
+                    1405.0,
+                    1594.0,
+                    452.0,
+                    1594.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1589.0,
+                    1405.0,
+                    1589.0,
+                    1405.0,
+                    1624.0,
+                    294.0,
+                    1624.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1620.0,
+                    1404.0,
+                    1620.0,
+                    1404.0,
+                    1653.0,
+                    294.0,
+                    1653.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1650.0,
+                    1406.0,
+                    1650.0,
+                    1406.0,
+                    1688.0,
+                    294.0,
+                    1688.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1682.0,
+                    1405.0,
+                    1682.0,
+                    1405.0,
+                    1715.0,
+                    295.0,
+                    1715.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1712.0,
+                    1404.0,
+                    1712.0,
+                    1404.0,
+                    1748.0,
+                    295.0,
+                    1748.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1741.0,
+                    529.0,
+                    1741.0,
+                    529.0,
+                    1775.0,
+                    295.0,
+                    1775.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1270.0,
+                    1405.0,
+                    1270.0,
+                    1405.0,
+                    1311.0,
+                    294.0,
+                    1311.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1302.0,
+                    485.0,
+                    1302.0,
+                    485.0,
+                    1339.0,
+                    296.0,
+                    1339.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    363.0,
+                    518.0,
+                    657.0,
+                    518.0,
+                    657.0,
+                    552.0,
+                    363.0,
+                    552.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    683.0,
+                    518.0,
+                    1322.0,
+                    518.0,
+                    1322.0,
+                    552.0,
+                    683.0,
+                    552.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1345.0,
+                    518.0,
+                    1404.0,
+                    518.0,
+                    1404.0,
+                    552.0,
+                    1345.0,
+                    552.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    394.0,
+                    550.0,
+                    686.0,
+                    550.0,
+                    686.0,
+                    584.0,
+                    394.0,
+                    584.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    771.0,
+                    550.0,
+                    1291.0,
+                    550.0,
+                    1291.0,
+                    584.0,
+                    771.0,
+                    584.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    355.0,
+                    586.0,
+                    987.0,
+                    586.0,
+                    987.0,
+                    632.0,
+                    355.0,
+                    632.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1014.0,
+                    586.0,
+                    1146.0,
+                    586.0,
+                    1146.0,
+                    632.0,
+                    1014.0,
+                    632.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    393.0,
+                    626.0,
+                    1406.0,
+                    626.0,
+                    1406.0,
+                    663.0,
+                    393.0,
+                    663.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    395.0,
+                    659.0,
+                    1190.0,
+                    659.0,
+                    1190.0,
+                    693.0,
+                    395.0,
+                    693.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    359.0,
+                    698.0,
+                    1405.0,
+                    698.0,
+                    1405.0,
+                    732.0,
+                    359.0,
+                    732.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    392.0,
+                    730.0,
+                    1406.0,
+                    730.0,
+                    1406.0,
+                    763.0,
+                    392.0,
+                    763.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1225.0,
+                    779.0,
+                    1404.0,
+                    779.0,
+                    1404.0,
+                    809.0,
+                    1225.0,
+                    809.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    394.0,
+                    800.0,
+                    469.0,
+                    800.0,
+                    469.0,
+                    838.0,
+                    394.0,
+                    838.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    503.0,
+                    800.0,
+                    549.0,
+                    800.0,
+                    549.0,
+                    838.0,
+                    503.0,
+                    838.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    578.0,
+                    800.0,
+                    953.0,
+                    800.0,
+                    953.0,
+                    838.0,
+                    578.0,
+                    838.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    972.0,
+                    800.0,
+                    1405.0,
+                    800.0,
+                    1405.0,
+                    838.0,
+                    972.0,
+                    838.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    394.0,
+                    831.0,
+                    1404.0,
+                    831.0,
+                    1404.0,
+                    867.0,
+                    394.0,
+                    867.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    394.0,
+                    862.0,
+                    1405.0,
+                    862.0,
+                    1405.0,
+                    900.0,
+                    394.0,
+                    900.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    394.0,
+                    894.0,
+                    799.0,
+                    894.0,
+                    799.0,
+                    929.0,
+                    394.0,
+                    929.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1186.75,
+                    593.5,
+                    1409.75,
+                    593.5,
+                    1409.75,
+                    641.0,
+                    1186.75,
+                    641.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    386.0,
+                    755.5,
+                    1408.0,
+                    755.5,
+                    1408.0,
+                    809.5,
+                    386.0,
+                    809.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    949.0,
+                    1043.0,
+                    949.0,
+                    1043.0,
+                    990.0,
+                    294.0,
+                    990.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    359.0,
+                    1792.0,
+                    579.0,
+                    1792.0,
+                    579.0,
+                    1836.0,
+                    359.0,
+                    1836.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    694.0,
+                    1792.0,
+                    705.0,
+                    1792.0,
+                    705.0,
+                    1836.0,
+                    694.0,
+                    1836.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    356.0,
+                    1833.0,
+                    1151.0,
+                    1833.0,
+                    1151.0,
+                    1878.0,
+                    356.0,
+                    1878.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1240.0,
+                    1833.0,
+                    1255.0,
+                    1833.0,
+                    1255.0,
+                    1878.0,
+                    1240.0,
+                    1878.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    359.0,
+                    1879.0,
+                    483.0,
+                    1879.0,
+                    483.0,
+                    1914.0,
+                    359.0,
+                    1914.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    534.0,
+                    1879.0,
+                    745.0,
+                    1879.0,
+                    745.0,
+                    1914.0,
+                    534.0,
+                    1914.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    771.0,
+                    1879.0,
+                    907.0,
+                    1879.0,
+                    907.0,
+                    1914.0,
+                    771.0,
+                    1914.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    996.0,
+                    1879.0,
+                    1206.0,
+                    1879.0,
+                    1206.0,
+                    1914.0,
+                    996.0,
+                    1914.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1257.0,
+                    1879.0,
+                    1265.0,
+                    1879.0,
+                    1265.0,
+                    1914.0,
+                    1257.0,
+                    1914.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    359.0,
+                    1920.0,
+                    1351.0,
+                    1920.0,
+                    1351.0,
+                    1955.0,
+                    359.0,
+                    1955.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    395.0,
+                    1954.0,
+                    870.0,
+                    1954.0,
+                    870.0,
+                    1986.0,
+                    395.0,
+                    1986.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1350.0,
+                    1405.0,
+                    1350.0,
+                    1405.0,
+                    1388.0,
+                    294.0,
+                    1388.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1379.0,
+                    1405.0,
+                    1379.0,
+                    1405.0,
+                    1420.0,
+                    293.0,
+                    1420.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1409.0,
+                    579.0,
+                    1409.0,
+                    579.0,
+                    1450.0,
+                    294.0,
+                    1450.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1435.0,
+                    296.0,
+                    1435.0,
+                    296.0,
+                    1484.0,
+                    293.0,
+                    1484.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    710.0,
+                    1435.0,
+                    717.0,
+                    1435.0,
+                    717.0,
+                    1484.0,
+                    710.0,
+                    1484.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 3,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1407,
+                    1404,
+                    1407,
+                    1404,
+                    1622,
+                    298,
+                    1622
+                ],
+                "score": 0.982
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1711,
+                    1403,
+                    1711,
+                    1403,
+                    1866,
+                    298,
+                    1866
+                ],
+                "score": 0.979
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    1880,
+                    1402,
+                    1880,
+                    1402,
+                    2034,
+                    299,
+                    2034
+                ],
+                "score": 0.976
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    851,
+                    1403,
+                    851,
+                    1403,
+                    950,
+                    299,
+                    950
+                ],
+                "score": 0.972
+            },
+            {
+                "category_id": 3,
+                "poly": [
+                    368,
+                    227,
+                    1323,
+                    227,
+                    1323,
+                    587,
+                    368,
+                    587
+                ],
+                "score": 0.972
+            },
+            {
+                "category_id": 4,
+                "poly": [
+                    296,
+                    616,
+                    1404,
+                    616,
+                    1404,
+                    831,
+                    296,
+                    831
+                ],
+                "score": 0.966
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    299,
+                    1657,
+                    531,
+                    1657,
+                    531,
+                    1689,
+                    299,
+                    1689
+                ],
+                "score": 0.897
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    300,
+                    1350,
+                    683,
+                    1350,
+                    683,
+                    1380,
+                    300,
+                    1380
+                ],
+                "score": 0.89
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    302,
+                    996,
+                    707,
+                    996,
+                    707,
+                    1030,
+                    302,
+                    1030
+                ],
+                "score": 0.89
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    297,
+                    76,
+                    858,
+                    76,
+                    858,
+                    104,
+                    297,
+                    104
+                ],
+                "score": 0.864
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1065,
+                    1403,
+                    1065,
+                    1403,
+                    1313,
+                    298,
+                    1313
+                ],
+                "score": 0.862
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    841,
+                    2088,
+                    858,
+                    2088,
+                    858,
+                    2112,
+                    841,
+                    2112
+                ],
+                "score": 0.724
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    426,
+                    851,
+                    559,
+                    851,
+                    559,
+                    890,
+                    426,
+                    890
+                ],
+                "score": 0.92,
+                "latex": "\\Delta T > \\frac { 1 } { 1 - s }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    740,
+                    2003,
+                    794,
+                    2003,
+                    794,
+                    2032,
+                    740,
+                    2032
+                ],
+                "score": 0.87,
+                "latex": "7 5 \\%"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    537,
+                    648,
+                    592,
+                    648,
+                    592,
+                    677,
+                    537,
+                    677
+                ],
+                "score": 0.86,
+                "latex": "80 \\%"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    659,
+                    1836,
+                    698,
+                    1836,
+                    698,
+                    1864,
+                    659,
+                    1864
+                ],
+                "score": 0.86,
+                "latex": "5 \\times"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1140,
+                    1713,
+                    1194,
+                    1713,
+                    1194,
+                    1742,
+                    1140,
+                    1742
+                ],
+                "score": 0.85,
+                "latex": "80 \\%"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    732,
+                    1440,
+                    767,
+                    1440,
+                    767,
+                    1470,
+                    732,
+                    1470
+                ],
+                "score": 0.85,
+                "latex": "L _ { 2 }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    983,
+                    1774,
+                    1087,
+                    1774,
+                    1087,
+                    1803,
+                    983,
+                    1803
+                ],
+                "score": 0.8,
+                "latex": "\\mathrm { ~ \\ : ( \\alpha ~ ) = ~ 0 . 3 ~ }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1101,
+                    1774,
+                    1237,
+                    1774,
+                    1237,
+                    1803,
+                    1101,
+                    1803
+                ],
+                "score": 0.8,
+                "latex": "\\Delta T = 1 0 0"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    396,
+                    888,
+                    464,
+                    888,
+                    464,
+                    916,
+                    396,
+                    916
+                ],
+                "score": 0.78,
+                "latex": "1 - S"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    934,
+                    1219,
+                    967,
+                    1219,
+                    967,
+                    1246,
+                    934,
+                    1246
+                ],
+                "score": 0.76,
+                "latex": "M"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    763,
+                    1280,
+                    869,
+                    1280,
+                    869,
+                    1313,
+                    763,
+                    1313
+                ],
+                "score": 0.67,
+                "latex": "{ \\mathrm { R i g L } } _ { M \\times } ,"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    823,
+                    1283,
+                    866,
+                    1283,
+                    866,
+                    1312,
+                    823,
+                    1312
+                ],
+                "score": 0.58,
+                "latex": "M \\times"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    998,
+                    1973,
+                    1051,
+                    1973,
+                    1051,
+                    2001,
+                    998,
+                    2001
+                ],
+                "score": 0.28,
+                "latex": "S E T"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    442.0,
+                    226.0,
+                    828.0,
+                    226.0,
+                    828.0,
+                    260.0,
+                    442.0,
+                    260.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1038.0,
+                    231.0,
+                    1216.0,
+                    231.0,
+                    1216.0,
+                    271.0,
+                    1038.0,
+                    271.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    398.0,
+                    264.0,
+                    440.0,
+                    264.0,
+                    440.0,
+                    293.0,
+                    398.0,
+                    293.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    368.0,
+                    298.0,
+                    411.0,
+                    298.0,
+                    411.0,
+                    496.0,
+                    368.0,
+                    496.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    906.0,
+                    302.0,
+                    940.0,
+                    302.0,
+                    940.0,
+                    322.0,
+                    906.0,
+                    322.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    399.0,
+                    303.0,
+                    438.0,
+                    303.0,
+                    438.0,
+                    327.0,
+                    399.0,
+                    327.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    878.0,
+                    309.0,
+                    914.0,
+                    309.0,
+                    914.0,
+                    491.0,
+                    878.0,
+                    491.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    400.0,
+                    340.0,
+                    438.0,
+                    340.0,
+                    438.0,
+                    364.0,
+                    400.0,
+                    364.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    564.0,
+                    347.0,
+                    573.0,
+                    347.0,
+                    573.0,
+                    355.0,
+                    564.0,
+                    355.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    463.0,
+                    365.0,
+                    473.0,
+                    365.0,
+                    473.0,
+                    376.0,
+                    463.0,
+                    376.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    709.0,
+                    357.0,
+                    786.0,
+                    357.0,
+                    786.0,
+                    392.0,
+                    709.0,
+                    392.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    400.0,
+                    378.0,
+                    438.0,
+                    378.0,
+                    438.0,
+                    398.0,
+                    400.0,
+                    398.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    622.0,
+                    374.0,
+                    637.0,
+                    374.0,
+                    637.0,
+                    386.0,
+                    622.0,
+                    386.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    706.0,
+                    381.0,
+                    758.0,
+                    381.0,
+                    758.0,
+                    422.0,
+                    706.0,
+                    422.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    906.0,
+                    374.0,
+                    942.0,
+                    374.0,
+                    942.0,
+                    394.0,
+                    906.0,
+                    394.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    399.0,
+                    414.0,
+                    436.0,
+                    414.0,
+                    436.0,
+                    434.0,
+                    399.0,
+                    434.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    695.0,
+                    413.0,
+                    753.0,
+                    413.0,
+                    753.0,
+                    443.0,
+                    695.0,
+                    443.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    996.0,
+                    408.0,
+                    1065.0,
+                    408.0,
+                    1065.0,
+                    448.0,
+                    996.0,
+                    448.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    399.0,
+                    449.0,
+                    438.0,
+                    449.0,
+                    438.0,
+                    473.0,
+                    399.0,
+                    473.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    562.0,
+                    459.0,
+                    574.0,
+                    459.0,
+                    574.0,
+                    468.0,
+                    562.0,
+                    468.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    706.0,
+                    440.0,
+                    823.0,
+                    440.0,
+                    823.0,
+                    499.0,
+                    706.0,
+                    499.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    906.0,
+                    447.0,
+                    941.0,
+                    447.0,
+                    941.0,
+                    467.0,
+                    906.0,
+                    467.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    993.0,
+                    437.0,
+                    1124.0,
+                    437.0,
+                    1124.0,
+                    496.0,
+                    993.0,
+                    496.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    399.0,
+                    487.0,
+                    437.0,
+                    487.0,
+                    437.0,
+                    511.0,
+                    399.0,
+                    511.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    707.0,
+                    493.0,
+                    768.0,
+                    493.0,
+                    768.0,
+                    526.0,
+                    707.0,
+                    526.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    400.0,
+                    522.0,
+                    440.0,
+                    522.0,
+                    440.0,
+                    546.0,
+                    400.0,
+                    546.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1036.0,
+                    526.0,
+                    1075.0,
+                    526.0,
+                    1075.0,
+                    551.0,
+                    1036.0,
+                    551.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1148.0,
+                    526.0,
+                    1186.0,
+                    526.0,
+                    1186.0,
+                    551.0,
+                    1148.0,
+                    551.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1260.0,
+                    528.0,
+                    1297.0,
+                    528.0,
+                    1297.0,
+                    551.0,
+                    1260.0,
+                    551.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    536.0,
+                    547.0,
+                    733.0,
+                    547.0,
+                    733.0,
+                    594.0,
+                    536.0,
+                    594.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1072.0,
+                    542.0,
+                    1186.0,
+                    542.0,
+                    1186.0,
+                    586.0,
+                    1072.0,
+                    586.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    997.75,
+                    387.5,
+                    1098.75,
+                    387.5,
+                    1098.75,
+                    417.5,
+                    997.75,
+                    417.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    998.0,
+                    489.5,
+                    1069.0,
+                    489.5,
+                    1069.0,
+                    520.5,
+                    998.0,
+                    520.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    910.75,
+                    515.0,
+                    957.75,
+                    515.0,
+                    957.75,
+                    551.5,
+                    910.75,
+                    551.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    491.0,
+                    532.5,
+                    533.0,
+                    532.5,
+                    533.0,
+                    556.5,
+                    491.0,
+                    556.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    616.0,
+                    1406.0,
+                    616.0,
+                    1406.0,
+                    655.0,
+                    294.0,
+                    655.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    646.0,
+                    536.0,
+                    646.0,
+                    536.0,
+                    682.0,
+                    294.0,
+                    682.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    593.0,
+                    646.0,
+                    1406.0,
+                    646.0,
+                    1406.0,
+                    682.0,
+                    593.0,
+                    682.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    674.0,
+                    1406.0,
+                    674.0,
+                    1406.0,
+                    715.0,
+                    291.0,
+                    715.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    708.0,
+                    1407.0,
+                    708.0,
+                    1407.0,
+                    746.0,
+                    294.0,
+                    746.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    738.0,
+                    1405.0,
+                    738.0,
+                    1405.0,
+                    773.0,
+                    295.0,
+                    773.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    765.0,
+                    1404.0,
+                    765.0,
+                    1404.0,
+                    807.0,
+                    291.0,
+                    807.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    799.0,
+                    1005.0,
+                    799.0,
+                    1005.0,
+                    836.0,
+                    293.0,
+                    836.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1654.0,
+                    533.0,
+                    1654.0,
+                    533.0,
+                    1694.0,
+                    294.0,
+                    1694.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1349.0,
+                    687.0,
+                    1349.0,
+                    687.0,
+                    1384.0,
+                    295.0,
+                    1384.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    991.0,
+                    713.0,
+                    991.0,
+                    713.0,
+                    1036.0,
+                    293.0,
+                    1036.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    73.0,
+                    858.0,
+                    73.0,
+                    858.0,
+                    108.0,
+                    297.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    839.0,
+                    2085.0,
+                    861.0,
+                    2085.0,
+                    861.0,
+                    2120.0,
+                    839.0,
+                    2120.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1407.0,
+                    1405.0,
+                    1407.0,
+                    1405.0,
+                    1443.0,
+                    293.0,
+                    1443.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1436.0,
+                    731.0,
+                    1436.0,
+                    731.0,
+                    1474.0,
+                    293.0,
+                    1474.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    768.0,
+                    1436.0,
+                    1404.0,
+                    1436.0,
+                    1404.0,
+                    1474.0,
+                    768.0,
+                    1474.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1470.0,
+                    1404.0,
+                    1470.0,
+                    1404.0,
+                    1505.0,
+                    296.0,
+                    1505.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1501.0,
+                    1401.0,
+                    1501.0,
+                    1401.0,
+                    1532.0,
+                    294.0,
+                    1532.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1529.0,
+                    1405.0,
+                    1529.0,
+                    1405.0,
+                    1564.0,
+                    296.0,
+                    1564.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1561.0,
+                    1405.0,
+                    1561.0,
+                    1405.0,
+                    1596.0,
+                    294.0,
+                    1596.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1587.0,
+                    474.0,
+                    1587.0,
+                    474.0,
+                    1629.0,
+                    292.0,
+                    1629.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1713.0,
+                    1139.0,
+                    1713.0,
+                    1139.0,
+                    1746.0,
+                    296.0,
+                    1746.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1195.0,
+                    1713.0,
+                    1402.0,
+                    1713.0,
+                    1402.0,
+                    1746.0,
+                    1195.0,
+                    1746.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    1743.0,
+                    1404.0,
+                    1743.0,
+                    1404.0,
+                    1777.0,
+                    297.0,
+                    1777.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1771.0,
+                    982.0,
+                    1771.0,
+                    982.0,
+                    1808.0,
+                    293.0,
+                    1808.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1088.0,
+                    1771.0,
+                    1100.0,
+                    1771.0,
+                    1100.0,
+                    1808.0,
+                    1088.0,
+                    1808.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1238.0,
+                    1771.0,
+                    1405.0,
+                    1771.0,
+                    1405.0,
+                    1808.0,
+                    1238.0,
+                    1808.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1804.0,
+                    1404.0,
+                    1804.0,
+                    1404.0,
+                    1838.0,
+                    296.0,
+                    1838.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1832.0,
+                    658.0,
+                    1832.0,
+                    658.0,
+                    1872.0,
+                    292.0,
+                    1872.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    699.0,
+                    1832.0,
+                    1068.0,
+                    1832.0,
+                    1068.0,
+                    1872.0,
+                    699.0,
+                    1872.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1879.0,
+                    1404.0,
+                    1879.0,
+                    1404.0,
+                    1917.0,
+                    295.0,
+                    1917.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1907.0,
+                    1406.0,
+                    1907.0,
+                    1406.0,
+                    1950.0,
+                    293.0,
+                    1950.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1944.0,
+                    1404.0,
+                    1944.0,
+                    1404.0,
+                    1977.0,
+                    293.0,
+                    1977.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1972.0,
+                    997.0,
+                    1972.0,
+                    997.0,
+                    2009.0,
+                    294.0,
+                    2009.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1052.0,
+                    1972.0,
+                    1406.0,
+                    1972.0,
+                    1406.0,
+                    2009.0,
+                    1052.0,
+                    2009.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    2002.0,
+                    739.0,
+                    2002.0,
+                    739.0,
+                    2038.0,
+                    294.0,
+                    2038.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    795.0,
+                    2002.0,
+                    1403.0,
+                    2002.0,
+                    1403.0,
+                    2038.0,
+                    795.0,
+                    2038.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    290.0,
+                    847.0,
+                    425.0,
+                    847.0,
+                    425.0,
+                    898.0,
+                    290.0,
+                    898.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    560.0,
+                    847.0,
+                    1408.0,
+                    847.0,
+                    1408.0,
+                    898.0,
+                    560.0,
+                    898.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    884.0,
+                    395.0,
+                    884.0,
+                    395.0,
+                    922.0,
+                    294.0,
+                    922.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    465.0,
+                    884.0,
+                    1404.0,
+                    884.0,
+                    1404.0,
+                    922.0,
+                    465.0,
+                    922.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    915.0,
+                    1040.0,
+                    915.0,
+                    1040.0,
+                    957.0,
+                    293.0,
+                    957.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1063.0,
+                    1404.0,
+                    1063.0,
+                    1404.0,
+                    1103.0,
+                    293.0,
+                    1103.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1094.0,
+                    1407.0,
+                    1094.0,
+                    1407.0,
+                    1133.0,
+                    291.0,
+                    1133.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1127.0,
+                    1406.0,
+                    1127.0,
+                    1406.0,
+                    1161.0,
+                    293.0,
+                    1161.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1158.0,
+                    1405.0,
+                    1158.0,
+                    1405.0,
+                    1191.0,
+                    293.0,
+                    1191.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1187.0,
+                    1407.0,
+                    1187.0,
+                    1407.0,
+                    1224.0,
+                    293.0,
+                    1224.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1219.0,
+                    933.0,
+                    1219.0,
+                    933.0,
+                    1253.0,
+                    294.0,
+                    1253.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    968.0,
+                    1219.0,
+                    1405.0,
+                    1219.0,
+                    1405.0,
+                    1253.0,
+                    968.0,
+                    1253.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1248.0,
+                    1405.0,
+                    1248.0,
+                    1405.0,
+                    1286.0,
+                    293.0,
+                    1286.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1277.0,
+                    762.0,
+                    1277.0,
+                    762.0,
+                    1317.0,
+                    293.0,
+                    1317.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    870.0,
+                    1277.0,
+                    886.0,
+                    1277.0,
+                    886.0,
+                    1317.0,
+                    870.0,
+                    1317.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 4,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 5,
+                "poly": [
+                    339,
+                    224,
+                    1359,
+                    224,
+                    1359,
+                    913,
+                    339,
+                    913
+                ],
+                "score": 0.986,
+                "html": "<table><tr><td>Method</td><td>Top-1 Accuracy</td><td>FLOPs (Train)</td><td>FLOPs (Test)</td><td>Top-1 Accuracy</td><td>FLOPs (Train)</td><td>FLOPs (Test)</td></tr><tr><td>Dense</td><td>76.8±0.09</td><td>1x (3.2e18)</td><td>1x (8.2e9)</td><td colspan=\"3\"></td></tr><tr><td>Static</td><td colspan=\"3\">S=0.8</td><td colspan=\"3\">S=0.9</td></tr><tr><td>SNIP</td><td>70.6±0.06 72.0±0.10</td><td>0.23x 0.23x</td><td>0.23x 0.23x</td><td>65.8±0.04 67.2±0.12</td><td>0.10x 0.10x</td><td>0.10x 0.10x</td></tr><tr><td>Small-Dense</td><td>72.1±0.12</td><td>0.20x</td><td>0.20x</td><td>68.9±0.10</td><td>0.12x</td><td>0.12x</td></tr><tr><td>SET</td><td>72.9±0.39</td><td>0.23x</td><td>0.23x</td><td>69.6±0.23</td><td>0.10x</td><td>0.10x</td></tr><tr><td>RigL</td><td>74.6±0.06</td><td>0.23x</td><td>0.23x</td><td>72.0±0.05</td><td>0.10x</td><td>0.10x</td></tr><tr><td>Small-Dense5x RigL5x</td><td>73.9±0.07</td><td>1.01x</td><td>0.20x</td><td>71.3±0.10</td><td>0.60x</td><td>0.12x</td></tr><tr><td>Static (ERK)</td><td>76.6±0.06</td><td>1.14x</td><td>0.23x</td><td>75.7±0.06</td><td>0.52x</td><td>0.10x</td></tr><tr><td>DSR*</td><td>72.1±0.04 73.3</td><td>0.42x 0.40x</td><td>0.42x 0.40x</td><td>67.7±0.12</td><td>0.24x</td><td>0.24x</td></tr><tr><td></td><td></td><td></td><td></td><td>71.6</td><td>0.30x</td><td>0.30x</td></tr><tr><td>RigL (ERK)</td><td>75.1±0.05</td><td>0.42x</td><td>0.42x</td><td>73.0±0.04</td><td>0.25x</td><td>0.24x</td></tr><tr><td>RigL5x (ERK)</td><td>77.1±0.06</td><td>2.09x</td><td>0.42x</td><td>76.4±0.05</td><td>1.23x</td><td>0.24x</td></tr><tr><td>SNFS*</td><td>74.2</td><td>n/a</td><td>n/a</td><td>72.3</td><td>n/a</td><td>n/a</td></tr><tr><td>SNFS (ERK)</td><td>75.2±0.11</td><td>0.61x</td><td>0.42x</td><td>72.9±0.06</td><td>0.50x</td><td>0.24x</td></tr><tr><td>Pruning* (Zhu)</td><td>73.2</td><td>1.00x</td><td>0.23x</td><td>70.3</td><td>1.00x</td><td>0.10x</td></tr><tr><td>Pruning* (Gale)</td><td>75.6</td><td>1.00x</td><td>0.23x</td><td>73.9</td><td>1.00x</td><td>0.10x</td></tr><tr><td>Pruning1.5× (Gale)</td><td>76.5</td><td>1.50x</td><td>0.23x</td><td></td><td>1.50x</td><td></td></tr><tr><td></td><td></td><td></td><td></td><td>75.2</td><td></td><td>0.10x</td></tr></table>"
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1252,
+                    1404,
+                    1252,
+                    1404,
+                    1527,
+                    298,
+                    1527
+                ],
+                "score": 0.981
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1541,
+                    1403,
+                    1541,
+                    1403,
+                    1757,
+                    297,
+                    1757
+                ],
+                "score": 0.978
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1772,
+                    1402,
+                    1772,
+                    1402,
+                    1956,
+                    298,
+                    1956
+                ],
+                "score": 0.975
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    1144,
+                    1403,
+                    1144,
+                    1403,
+                    1236,
+                    299,
+                    1236
+                ],
+                "score": 0.974
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    298,
+                    75,
+                    858,
+                    75,
+                    858,
+                    105,
+                    298,
+                    105
+                ],
+                "score": 0.907
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    340,
+                    2007,
+                    488,
+                    2007,
+                    488,
+                    2033,
+                    340,
+                    2033
+                ],
+                "score": 0.855
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    840,
+                    2089,
+                    859,
+                    2089,
+                    859,
+                    2112,
+                    840,
+                    2112
+                ],
+                "score": 0.777
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    934,
+                    1404,
+                    934,
+                    1404,
+                    1119,
+                    297,
+                    1119
+                ],
+                "score": 0.758
+            },
+            {
+                "category_id": 6,
+                "poly": [
+                    297,
+                    934,
+                    1404,
+                    934,
+                    1404,
+                    1119,
+                    297,
+                    1119
+                ],
+                "score": 0.138
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    962,
+                    1635,
+                    1052,
+                    1635,
+                    1052,
+                    1664,
+                    962,
+                    1664
+                ],
+                "score": 0.88,
+                "latex": "7 2 . 7 5 \\%"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    527,
+                    1635,
+                    603,
+                    1635,
+                    603,
+                    1664,
+                    527,
+                    1664
+                ],
+                "score": 0.88,
+                "latex": "9 6 . 5 \\%"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1173,
+                    936,
+                    1228,
+                    936,
+                    1228,
+                    965,
+                    1173,
+                    965
+                ],
+                "score": 0.88,
+                "latex": "90 \\%"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1068,
+                    936,
+                    1123,
+                    936,
+                    1123,
+                    965,
+                    1068,
+                    965
+                ],
+                "score": 0.87,
+                "latex": "80 \\%"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    353,
+                    1088,
+                    408,
+                    1088,
+                    408,
+                    1117,
+                    353,
+                    1117
+                ],
+                "score": 0.87,
+                "latex": "20 \\%"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    852,
+                    1089,
+                    906,
+                    1089,
+                    906,
+                    1116,
+                    852,
+                    1116
+                ],
+                "score": 0.86,
+                "latex": "42 \\%"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    778,
+                    1058,
+                    869,
+                    1058,
+                    869,
+                    1090,
+                    778,
+                    1090
+                ],
+                "score": 0.86,
+                "latex": "\\mathrm { R i g L } _ { \\mathrm { 5 } \\times }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1063,
+                    1058,
+                    1140,
+                    1058,
+                    1140,
+                    1088,
+                    1063,
+                    1088
+                ],
+                "score": 0.86,
+                "latex": "7 7 . 1 \\%"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    766,
+                    1496,
+                    820,
+                    1496,
+                    820,
+                    1525,
+                    766,
+                    1525
+                ],
+                "score": 0.86,
+                "latex": "80 \\%"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1339,
+                    1634,
+                    1401,
+                    1634,
+                    1401,
+                    1664,
+                    1339,
+                    1664
+                ],
+                "score": 0.85,
+                "latex": "3 . 5 \\%"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    936,
+                    1405,
+                    987,
+                    1405,
+                    987,
+                    1434,
+                    936,
+                    1434
+                ],
+                "score": 0.69,
+                "latex": "1 . 5 \\mathrm { x }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    604,
+                    780,
+                    720,
+                    780,
+                    720,
+                    809,
+                    604,
+                    809
+                ],
+                "score": 0.36,
+                "latex": "7 5 . 2 { \\pm } 0 . 1 1 "
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    563,
+                    1467,
+                    595,
+                    1467,
+                    595,
+                    1494,
+                    563,
+                    1494
+                ],
+                "score": 0.34,
+                "latex": "{ 5 } \\mathbf { x }"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    298.0,
+                    73.0,
+                    858.0,
+                    73.0,
+                    858.0,
+                    108.0,
+                    298.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    334.0,
+                    2002.0,
+                    493.0,
+                    2002.0,
+                    493.0,
+                    2036.0,
+                    334.0,
+                    2036.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    840.0,
+                    2087.0,
+                    862.0,
+                    2087.0,
+                    862.0,
+                    2118.0,
+                    840.0,
+                    2118.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    932.0,
+                    1067.0,
+                    932.0,
+                    1067.0,
+                    969.0,
+                    294.0,
+                    969.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1124.0,
+                    932.0,
+                    1172.0,
+                    932.0,
+                    1172.0,
+                    969.0,
+                    1124.0,
+                    969.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1229.0,
+                    932.0,
+                    1405.0,
+                    932.0,
+                    1405.0,
+                    969.0,
+                    1229.0,
+                    969.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    965.0,
+                    1405.0,
+                    965.0,
+                    1405.0,
+                    1001.0,
+                    295.0,
+                    1001.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    996.0,
+                    1405.0,
+                    996.0,
+                    1405.0,
+                    1030.0,
+                    294.0,
+                    1030.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1026.0,
+                    1404.0,
+                    1026.0,
+                    1404.0,
+                    1062.0,
+                    295.0,
+                    1062.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1052.0,
+                    777.0,
+                    1052.0,
+                    777.0,
+                    1097.0,
+                    292.0,
+                    1097.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    870.0,
+                    1052.0,
+                    1062.0,
+                    1052.0,
+                    1062.0,
+                    1097.0,
+                    870.0,
+                    1097.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1141.0,
+                    1052.0,
+                    1406.0,
+                    1052.0,
+                    1406.0,
+                    1097.0,
+                    1141.0,
+                    1097.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1088.0,
+                    352.0,
+                    1088.0,
+                    352.0,
+                    1121.0,
+                    295.0,
+                    1121.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    409.0,
+                    1088.0,
+                    851.0,
+                    1088.0,
+                    851.0,
+                    1121.0,
+                    409.0,
+                    1121.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    907.0,
+                    1088.0,
+                    1065.0,
+                    1088.0,
+                    1065.0,
+                    1121.0,
+                    907.0,
+                    1121.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1251.0,
+                    1406.0,
+                    1251.0,
+                    1406.0,
+                    1287.0,
+                    294.0,
+                    1287.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1283.0,
+                    1404.0,
+                    1283.0,
+                    1404.0,
+                    1315.0,
+                    296.0,
+                    1315.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1314.0,
+                    1406.0,
+                    1314.0,
+                    1406.0,
+                    1349.0,
+                    293.0,
+                    1349.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1343.0,
+                    1404.0,
+                    1343.0,
+                    1404.0,
+                    1379.0,
+                    294.0,
+                    1379.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1376.0,
+                    1404.0,
+                    1376.0,
+                    1404.0,
+                    1409.0,
+                    294.0,
+                    1409.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1403.0,
+                    935.0,
+                    1403.0,
+                    935.0,
+                    1440.0,
+                    293.0,
+                    1440.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    988.0,
+                    1403.0,
+                    1404.0,
+                    1403.0,
+                    1404.0,
+                    1440.0,
+                    988.0,
+                    1440.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1432.0,
+                    1406.0,
+                    1432.0,
+                    1406.0,
+                    1470.0,
+                    293.0,
+                    1470.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1465.0,
+                    562.0,
+                    1465.0,
+                    562.0,
+                    1501.0,
+                    293.0,
+                    1501.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    596.0,
+                    1465.0,
+                    1404.0,
+                    1465.0,
+                    1404.0,
+                    1501.0,
+                    596.0,
+                    1501.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1496.0,
+                    765.0,
+                    1496.0,
+                    765.0,
+                    1531.0,
+                    293.0,
+                    1531.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    821.0,
+                    1496.0,
+                    1404.0,
+                    1496.0,
+                    1404.0,
+                    1531.0,
+                    821.0,
+                    1531.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1540.0,
+                    1406.0,
+                    1540.0,
+                    1406.0,
+                    1577.0,
+                    292.0,
+                    1577.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1574.0,
+                    1405.0,
+                    1574.0,
+                    1405.0,
+                    1608.0,
+                    293.0,
+                    1608.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1605.0,
+                    1401.0,
+                    1605.0,
+                    1401.0,
+                    1639.0,
+                    293.0,
+                    1639.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1634.0,
+                    526.0,
+                    1634.0,
+                    526.0,
+                    1669.0,
+                    293.0,
+                    1669.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    604.0,
+                    1634.0,
+                    961.0,
+                    1634.0,
+                    961.0,
+                    1669.0,
+                    604.0,
+                    1669.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1053.0,
+                    1634.0,
+                    1338.0,
+                    1634.0,
+                    1338.0,
+                    1669.0,
+                    1053.0,
+                    1669.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1665.0,
+                    1405.0,
+                    1665.0,
+                    1405.0,
+                    1700.0,
+                    293.0,
+                    1700.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1696.0,
+                    1405.0,
+                    1696.0,
+                    1405.0,
+                    1729.0,
+                    293.0,
+                    1729.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1726.0,
+                    874.0,
+                    1726.0,
+                    874.0,
+                    1760.0,
+                    293.0,
+                    1760.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1771.0,
+                    1406.0,
+                    1771.0,
+                    1406.0,
+                    1807.0,
+                    293.0,
+                    1807.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1803.0,
+                    1406.0,
+                    1803.0,
+                    1406.0,
+                    1839.0,
+                    294.0,
+                    1839.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1834.0,
+                    1406.0,
+                    1834.0,
+                    1406.0,
+                    1867.0,
+                    292.0,
+                    1867.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1863.0,
+                    1406.0,
+                    1863.0,
+                    1406.0,
+                    1898.0,
+                    293.0,
+                    1898.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1894.0,
+                    1404.0,
+                    1894.0,
+                    1404.0,
+                    1930.0,
+                    293.0,
+                    1930.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1925.0,
+                    593.0,
+                    1925.0,
+                    593.0,
+                    1958.0,
+                    292.0,
+                    1958.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1143.0,
+                    1405.0,
+                    1143.0,
+                    1405.0,
+                    1180.0,
+                    294.0,
+                    1180.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1174.0,
+                    1400.0,
+                    1174.0,
+                    1400.0,
+                    1208.0,
+                    295.0,
+                    1208.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1207.0,
+                    827.0,
+                    1207.0,
+                    827.0,
+                    1238.0,
+                    294.0,
+                    1238.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    932.0,
+                    1067.0,
+                    932.0,
+                    1067.0,
+                    969.0,
+                    294.0,
+                    969.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1124.0,
+                    932.0,
+                    1172.0,
+                    932.0,
+                    1172.0,
+                    969.0,
+                    1124.0,
+                    969.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1229.0,
+                    932.0,
+                    1405.0,
+                    932.0,
+                    1405.0,
+                    969.0,
+                    1229.0,
+                    969.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    965.0,
+                    1405.0,
+                    965.0,
+                    1405.0,
+                    1001.0,
+                    295.0,
+                    1001.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    996.0,
+                    1405.0,
+                    996.0,
+                    1405.0,
+                    1030.0,
+                    294.0,
+                    1030.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1026.0,
+                    1404.0,
+                    1026.0,
+                    1404.0,
+                    1062.0,
+                    295.0,
+                    1062.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1052.0,
+                    777.0,
+                    1052.0,
+                    777.0,
+                    1097.0,
+                    292.0,
+                    1097.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    870.0,
+                    1052.0,
+                    1062.0,
+                    1052.0,
+                    1062.0,
+                    1097.0,
+                    870.0,
+                    1097.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1141.0,
+                    1052.0,
+                    1406.0,
+                    1052.0,
+                    1406.0,
+                    1097.0,
+                    1141.0,
+                    1097.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1088.0,
+                    352.0,
+                    1088.0,
+                    352.0,
+                    1121.0,
+                    295.0,
+                    1121.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    409.0,
+                    1088.0,
+                    851.0,
+                    1088.0,
+                    851.0,
+                    1121.0,
+                    409.0,
+                    1121.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    907.0,
+                    1088.0,
+                    1065.0,
+                    1088.0,
+                    1065.0,
+                    1121.0,
+                    907.0,
+                    1121.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 5,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    1559,
+                    1402,
+                    1559,
+                    1402,
+                    1804,
+                    299,
+                    1804
+                ],
+                "score": 0.984
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1819,
+                    1403,
+                    1819,
+                    1403,
+                    2035,
+                    298,
+                    2035
+                ],
+                "score": 0.978
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1302,
+                    1402,
+                    1302,
+                    1402,
+                    1456,
+                    298,
+                    1456
+                ],
+                "score": 0.977
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    964,
+                    1403,
+                    964,
+                    1403,
+                    1118,
+                    299,
+                    1118
+                ],
+                "score": 0.976
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    1132,
+                    1402,
+                    1132,
+                    1402,
+                    1287,
+                    299,
+                    1287
+                ],
+                "score": 0.975
+            },
+            {
+                "category_id": 4,
+                "poly": [
+                    296,
+                    628,
+                    1406,
+                    628,
+                    1406,
+                    842,
+                    296,
+                    842
+                ],
+                "score": 0.924
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    300,
+                    907,
+                    540,
+                    907,
+                    540,
+                    939,
+                    300,
+                    939
+                ],
+                "score": 0.903
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    297,
+                    76,
+                    858,
+                    76,
+                    858,
+                    104,
+                    297,
+                    104
+                ],
+                "score": 0.879
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    299,
+                    1500,
+                    903,
+                    1500,
+                    903,
+                    1531,
+                    299,
+                    1531
+                ],
+                "score": 0.872
+            },
+            {
+                "category_id": 3,
+                "poly": [
+                    314,
+                    234,
+                    767,
+                    234,
+                    767,
+                    591,
+                    314,
+                    591
+                ],
+                "score": 0.84
+            },
+            {
+                "category_id": 5,
+                "poly": [
+                    779,
+                    222,
+                    1431,
+                    222,
+                    1431,
+                    607,
+                    779,
+                    607
+                ],
+                "score": 0.757,
+                "html": "<table><tr><td rowspan=1 colspan=1>S</td><td rowspan=1 colspan=1>Method</td><td rowspan=1 colspan=1>Top-1</td><td rowspan=1 colspan=1>FLOPs</td></tr><tr><td rowspan=1 colspan=1>0.75</td><td rowspan=1 colspan=1>Small-Dense5×Pruning (Zhu)RigL5×RigL5x(ERK)</td><td rowspan=1 colspan=1>66.0±0.1167.771.5±0.0671.9±0.01</td><td rowspan=1 colspan=1>0.23x0.27x0.27x0.52x</td></tr><tr><td rowspan=1 colspan=1>0.90</td><td rowspan=1 colspan=1>Small-Dense5xPruning (Zhu)RigL5×RigL5x(ERK)</td><td rowspan=1 colspan=1>57.7±0.3461.867.0±0.1768.1±0.11</td><td rowspan=1 colspan=1>0.09x0.12x0.12x0.27x</td></tr><tr><td rowspan=1 colspan=1></td><td rowspan=1 colspan=1>Dense</td><td rowspan=1 colspan=1>72.1±0.17</td><td rowspan=1 colspan=1>1x (1.1e9)</td></tr><tr><td rowspan=1 colspan=1>0.75</td><td rowspan=1 colspan=1>Big-Sparse5×Big-Sparse5x(ERK)</td><td rowspan=1 colspan=1>76.4±0.0577.0±0.08</td><td rowspan=1 colspan=1>0.98x1.91x</td></tr></table>"
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    842,
+                    2088,
+                    858,
+                    2088,
+                    858,
+                    2111,
+                    842,
+                    2111
+                ],
+                "score": 0.641
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    842,
+                    2088,
+                    859,
+                    2088,
+                    859,
+                    2111,
+                    842,
+                    2111
+                ],
+                "score": 0.248
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    917,
+                    1364,
+                    973,
+                    1364,
+                    973,
+                    1393,
+                    917,
+                    1393
+                ],
+                "score": 0.85,
+                "latex": "7 5 \\%"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1334,
+                    1394,
+                    1399,
+                    1394,
+                    1399,
+                    1423,
+                    1334,
+                    1423
+                ],
+                "score": 0.85,
+                "latex": "4 . 3 \\%"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    866,
+                    1913,
+                    935,
+                    1913,
+                    935,
+                    1941,
+                    866,
+                    1941
+                ],
+                "score": 0.81,
+                "latex": "7 e - 4"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1324,
+                    1912,
+                    1394,
+                    1912,
+                    1394,
+                    1942,
+                    1324,
+                    1942
+                ],
+                "score": 0.74,
+                "latex": "5 e { - 4 }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1146,
+                    571,
+                    1267,
+                    571,
+                    1267,
+                    604,
+                    1146,
+                    604
+                ],
+                "score": 0.31,
+                "latex": "{ \\bf 7 7 . 0 { \\pm 0 . 0 8 } }"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    628.0,
+                    1404.0,
+                    628.0,
+                    1404.0,
+                    663.0,
+                    294.0,
+                    663.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    655.0,
+                    1407.0,
+                    655.0,
+                    1407.0,
+                    695.0,
+                    291.0,
+                    695.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    690.0,
+                    1403.0,
+                    690.0,
+                    1403.0,
+                    724.0,
+                    295.0,
+                    724.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    719.0,
+                    1406.0,
+                    719.0,
+                    1406.0,
+                    755.0,
+                    293.0,
+                    755.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    748.0,
+                    1407.0,
+                    748.0,
+                    1407.0,
+                    787.0,
+                    293.0,
+                    787.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    780.0,
+                    1407.0,
+                    780.0,
+                    1407.0,
+                    816.0,
+                    291.0,
+                    816.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    810.0,
+                    381.0,
+                    810.0,
+                    381.0,
+                    841.0,
+                    292.0,
+                    841.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    902.0,
+                    545.0,
+                    902.0,
+                    545.0,
+                    944.0,
+                    294.0,
+                    944.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    73.0,
+                    858.0,
+                    73.0,
+                    858.0,
+                    108.0,
+                    297.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1498.0,
+                    905.0,
+                    1498.0,
+                    905.0,
+                    1534.0,
+                    295.0,
+                    1534.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    456.0,
+                    236.0,
+                    680.0,
+                    236.0,
+                    680.0,
+                    269.0,
+                    456.0,
+                    269.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    344.0,
+                    291.0,
+                    381.0,
+                    291.0,
+                    381.0,
+                    313.0,
+                    344.0,
+                    313.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    315.0,
+                    311.0,
+                    352.0,
+                    311.0,
+                    352.0,
+                    497.0,
+                    315.0,
+                    497.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    343.0,
+                    344.0,
+                    381.0,
+                    344.0,
+                    381.0,
+                    366.0,
+                    343.0,
+                    366.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    345.0,
+                    400.0,
+                    381.0,
+                    400.0,
+                    381.0,
+                    419.0,
+                    345.0,
+                    419.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    423.0,
+                    391.0,
+                    516.0,
+                    391.0,
+                    516.0,
+                    423.0,
+                    423.0,
+                    423.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    436.0,
+                    417.0,
+                    551.0,
+                    417.0,
+                    551.0,
+                    450.0,
+                    436.0,
+                    450.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    344.0,
+                    451.0,
+                    386.0,
+                    451.0,
+                    386.0,
+                    474.0,
+                    344.0,
+                    474.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    426.0,
+                    445.0,
+                    569.0,
+                    445.0,
+                    569.0,
+                    476.0,
+                    426.0,
+                    476.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    437.0,
+                    470.0,
+                    563.0,
+                    470.0,
+                    563.0,
+                    502.0,
+                    437.0,
+                    502.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    344.0,
+                    505.0,
+                    382.0,
+                    505.0,
+                    382.0,
+                    528.0,
+                    344.0,
+                    528.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    432.0,
+                    500.0,
+                    555.0,
+                    500.0,
+                    555.0,
+                    524.0,
+                    432.0,
+                    524.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    363.0,
+                    536.0,
+                    403.0,
+                    536.0,
+                    403.0,
+                    561.0,
+                    363.0,
+                    561.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    455.0,
+                    536.0,
+                    497.0,
+                    536.0,
+                    497.0,
+                    560.0,
+                    455.0,
+                    560.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    548.0,
+                    535.0,
+                    589.0,
+                    535.0,
+                    589.0,
+                    560.0,
+                    548.0,
+                    560.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    641.0,
+                    538.0,
+                    680.0,
+                    538.0,
+                    680.0,
+                    559.0,
+                    641.0,
+                    559.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    734.0,
+                    537.0,
+                    773.0,
+                    537.0,
+                    773.0,
+                    560.0,
+                    734.0,
+                    560.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    510.0,
+                    550.0,
+                    627.0,
+                    550.0,
+                    627.0,
+                    597.0,
+                    510.0,
+                    597.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    571.75,
+                    399.5,
+                    694.75,
+                    399.5,
+                    694.75,
+                    431.5,
+                    571.75,
+                    431.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    840.0,
+                    2086.0,
+                    860.0,
+                    2086.0,
+                    860.0,
+                    2119.0,
+                    840.0,
+                    2119.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    840.0,
+                    2087.0,
+                    861.0,
+                    2087.0,
+                    861.0,
+                    2119.0,
+                    840.0,
+                    2119.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1559.0,
+                    1406.0,
+                    1559.0,
+                    1406.0,
+                    1596.0,
+                    294.0,
+                    1596.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1590.0,
+                    1406.0,
+                    1590.0,
+                    1406.0,
+                    1624.0,
+                    295.0,
+                    1624.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1619.0,
+                    1404.0,
+                    1619.0,
+                    1404.0,
+                    1654.0,
+                    294.0,
+                    1654.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1652.0,
+                    1407.0,
+                    1652.0,
+                    1407.0,
+                    1685.0,
+                    292.0,
+                    1685.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1679.0,
+                    1407.0,
+                    1679.0,
+                    1407.0,
+                    1720.0,
+                    292.0,
+                    1720.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1711.0,
+                    1406.0,
+                    1711.0,
+                    1406.0,
+                    1747.0,
+                    292.0,
+                    1747.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1744.0,
+                    1404.0,
+                    1744.0,
+                    1404.0,
+                    1777.0,
+                    294.0,
+                    1777.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1772.0,
+                    778.0,
+                    1772.0,
+                    778.0,
+                    1809.0,
+                    295.0,
+                    1809.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1818.0,
+                    1407.0,
+                    1818.0,
+                    1407.0,
+                    1854.0,
+                    294.0,
+                    1854.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1848.0,
+                    1405.0,
+                    1848.0,
+                    1405.0,
+                    1885.0,
+                    293.0,
+                    1885.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1879.0,
+                    1405.0,
+                    1879.0,
+                    1405.0,
+                    1916.0,
+                    293.0,
+                    1916.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1912.0,
+                    865.0,
+                    1912.0,
+                    865.0,
+                    1947.0,
+                    294.0,
+                    1947.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    936.0,
+                    1912.0,
+                    1323.0,
+                    1912.0,
+                    1323.0,
+                    1947.0,
+                    936.0,
+                    1947.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1395.0,
+                    1912.0,
+                    1405.0,
+                    1912.0,
+                    1405.0,
+                    1947.0,
+                    1395.0,
+                    1947.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1941.0,
+                    1407.0,
+                    1941.0,
+                    1407.0,
+                    1979.0,
+                    292.0,
+                    1979.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1973.0,
+                    1403.0,
+                    1973.0,
+                    1403.0,
+                    2007.0,
+                    296.0,
+                    2007.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    2002.0,
+                    1405.0,
+                    2002.0,
+                    1405.0,
+                    2037.0,
+                    294.0,
+                    2037.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1298.0,
+                    1406.0,
+                    1298.0,
+                    1406.0,
+                    1339.0,
+                    294.0,
+                    1339.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1327.0,
+                    1406.0,
+                    1327.0,
+                    1406.0,
+                    1370.0,
+                    293.0,
+                    1370.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1356.0,
+                    916.0,
+                    1356.0,
+                    916.0,
+                    1402.0,
+                    292.0,
+                    1402.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    974.0,
+                    1356.0,
+                    1406.0,
+                    1356.0,
+                    1406.0,
+                    1402.0,
+                    974.0,
+                    1402.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1392.0,
+                    1333.0,
+                    1392.0,
+                    1333.0,
+                    1428.0,
+                    292.0,
+                    1428.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1400.0,
+                    1392.0,
+                    1403.0,
+                    1392.0,
+                    1403.0,
+                    1428.0,
+                    1400.0,
+                    1428.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1421.0,
+                    785.0,
+                    1421.0,
+                    785.0,
+                    1462.0,
+                    293.0,
+                    1462.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    962.0,
+                    1405.0,
+                    962.0,
+                    1405.0,
+                    1000.0,
+                    293.0,
+                    1000.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    996.0,
+                    1404.0,
+                    996.0,
+                    1404.0,
+                    1029.0,
+                    295.0,
+                    1029.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1024.0,
+                    1405.0,
+                    1024.0,
+                    1405.0,
+                    1061.0,
+                    294.0,
+                    1061.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1050.0,
+                    1407.0,
+                    1050.0,
+                    1407.0,
+                    1095.0,
+                    293.0,
+                    1095.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1085.0,
+                    1200.0,
+                    1085.0,
+                    1200.0,
+                    1123.0,
+                    293.0,
+                    1123.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    1134.0,
+                    1404.0,
+                    1134.0,
+                    1404.0,
+                    1167.0,
+                    297.0,
+                    1167.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    1166.0,
+                    1404.0,
+                    1166.0,
+                    1404.0,
+                    1199.0,
+                    297.0,
+                    1199.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1194.0,
+                    1406.0,
+                    1194.0,
+                    1406.0,
+                    1231.0,
+                    295.0,
+                    1231.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1224.0,
+                    1403.0,
+                    1224.0,
+                    1403.0,
+                    1261.0,
+                    294.0,
+                    1261.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1257.0,
+                    576.0,
+                    1257.0,
+                    576.0,
+                    1292.0,
+                    294.0,
+                    1292.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 6,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1456,
+                    1404,
+                    1456,
+                    1404,
+                    1732,
+                    298,
+                    1732
+                ],
+                "score": 0.981
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1227,
+                    1403,
+                    1227,
+                    1403,
+                    1442,
+                    298,
+                    1442
+                ],
+                "score": 0.98
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    943,
+                    1403,
+                    943,
+                    1403,
+                    1127,
+                    299,
+                    1127
+                ],
+                "score": 0.976
+            },
+            {
+                "category_id": 3,
+                "poly": [
+                    313,
+                    226,
+                    1373,
+                    226,
+                    1373,
+                    628,
+                    313,
+                    628
+                ],
+                "score": 0.969
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    300,
+                    1834,
+                    1403,
+                    1834,
+                    1403,
+                    1925,
+                    300,
+                    1925
+                ],
+                "score": 0.969
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    803,
+                    1403,
+                    803,
+                    1403,
+                    927,
+                    299,
+                    927
+                ],
+                "score": 0.966
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    1942,
+                    1402,
+                    1942,
+                    1402,
+                    2034,
+                    299,
+                    2034
+                ],
+                "score": 0.966
+            },
+            {
+                "category_id": 4,
+                "poly": [
+                    296,
+                    657,
+                    1404,
+                    657,
+                    1404,
+                    779,
+                    296,
+                    779
+                ],
+                "score": 0.965
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    303,
+                    1168,
+                    776,
+                    1168,
+                    776,
+                    1200,
+                    303,
+                    1200
+                ],
+                "score": 0.898
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    303,
+                    1776,
+                    849,
+                    1776,
+                    849,
+                    1806,
+                    303,
+                    1806
+                ],
+                "score": 0.849
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    300,
+                    76,
+                    855,
+                    76,
+                    855,
+                    104,
+                    300,
+                    104
+                ],
+                "score": 0.806
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    841,
+                    2089,
+                    858,
+                    2089,
+                    858,
+                    2111,
+                    841,
+                    2111
+                ],
+                "score": 0.784
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    873,
+                    836,
+                    974,
+                    836,
+                    974,
+                    865,
+                    873,
+                    865
+                ],
+                "score": 0.89,
+                "latex": "\\alpha = 0 . 1"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1028,
+                    836,
+                    1159,
+                    836,
+                    1159,
+                    864,
+                    1028,
+                    864
+                ],
+                "score": 0.89,
+                "latex": "\\Delta T = 1 0 0"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    357,
+                    1519,
+                    432,
+                    1519,
+                    432,
+                    1548,
+                    357,
+                    1548
+                ],
+                "score": 0.87,
+                "latex": "9 4 . 1 \\%"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1142,
+                    1519,
+                    1197,
+                    1519,
+                    1197,
+                    1548,
+                    1142,
+                    1548
+                ],
+                "score": 0.85,
+                "latex": "50 \\%"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    416.0,
+                    227.0,
+                    806.0,
+                    227.0,
+                    806.0,
+                    260.0,
+                    416.0,
+                    260.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    982.0,
+                    236.0,
+                    1336.0,
+                    236.0,
+                    1336.0,
+                    271.0,
+                    982.0,
+                    271.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    911.0,
+                    259.0,
+                    954.0,
+                    259.0,
+                    954.0,
+                    286.0,
+                    911.0,
+                    286.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    401.0,
+                    297.0,
+                    410.0,
+                    297.0,
+                    410.0,
+                    306.0,
+                    401.0,
+                    306.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    911.0,
+                    300.0,
+                    952.0,
+                    300.0,
+                    952.0,
+                    327.0,
+                    911.0,
+                    327.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    628.0,
+                    314.0,
+                    638.0,
+                    314.0,
+                    638.0,
+                    323.0,
+                    628.0,
+                    323.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    880.0,
+                    316.0,
+                    922.0,
+                    316.0,
+                    922.0,
+                    523.0,
+                    880.0,
+                    523.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    318.0,
+                    333.0,
+                    356.0,
+                    333.0,
+                    356.0,
+                    382.0,
+                    318.0,
+                    382.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    912.0,
+                    341.0,
+                    953.0,
+                    341.0,
+                    953.0,
+                    368.0,
+                    912.0,
+                    368.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    404.0,
+                    353.0,
+                    410.0,
+                    353.0,
+                    410.0,
+                    359.0,
+                    404.0,
+                    359.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    315.0,
+                    365.0,
+                    351.0,
+                    365.0,
+                    351.0,
+                    399.0,
+                    315.0,
+                    399.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    316.0,
+                    390.0,
+                    339.0,
+                    390.0,
+                    339.0,
+                    412.0,
+                    316.0,
+                    412.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    912.0,
+                    384.0,
+                    952.0,
+                    384.0,
+                    952.0,
+                    409.0,
+                    912.0,
+                    409.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    319.0,
+                    402.0,
+                    396.0,
+                    402.0,
+                    396.0,
+                    433.0,
+                    319.0,
+                    433.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1013.0,
+                    405.0,
+                    1143.0,
+                    405.0,
+                    1143.0,
+                    437.0,
+                    1013.0,
+                    437.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    313.0,
+                    425.0,
+                    355.0,
+                    425.0,
+                    355.0,
+                    517.0,
+                    313.0,
+                    517.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    455.0,
+                    430.0,
+                    512.0,
+                    430.0,
+                    512.0,
+                    497.0,
+                    455.0,
+                    497.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    911.0,
+                    425.0,
+                    952.0,
+                    425.0,
+                    952.0,
+                    490.0,
+                    911.0,
+                    490.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1004.0,
+                    433.0,
+                    1138.0,
+                    433.0,
+                    1138.0,
+                    496.0,
+                    1004.0,
+                    496.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    349.0,
+                    478.0,
+                    395.0,
+                    478.0,
+                    395.0,
+                    508.0,
+                    349.0,
+                    508.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    458.0,
+                    493.0,
+                    523.0,
+                    493.0,
+                    523.0,
+                    556.0,
+                    458.0,
+                    556.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    911.0,
+                    508.0,
+                    958.0,
+                    508.0,
+                    958.0,
+                    533.0,
+                    911.0,
+                    533.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1013.0,
+                    488.0,
+                    1147.0,
+                    488.0,
+                    1147.0,
+                    554.0,
+                    1013.0,
+                    554.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    350.0,
+                    556.0,
+                    395.0,
+                    556.0,
+                    395.0,
+                    581.0,
+                    350.0,
+                    581.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    503.0,
+                    568.0,
+                    739.0,
+                    568.0,
+                    739.0,
+                    635.0,
+                    503.0,
+                    635.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1027.0,
+                    560.0,
+                    1063.0,
+                    560.0,
+                    1063.0,
+                    586.0,
+                    1027.0,
+                    586.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1096.0,
+                    560.0,
+                    1244.0,
+                    560.0,
+                    1244.0,
+                    626.0,
+                    1096.0,
+                    626.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1302.0,
+                    562.0,
+                    1335.0,
+                    562.0,
+                    1335.0,
+                    585.0,
+                    1302.0,
+                    585.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    319.75,
+                    321.0,
+                    393.75,
+                    321.0,
+                    393.75,
+                    354.0,
+                    319.75,
+                    354.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    455.0,
+                    404.5,
+                    541.0,
+                    404.5,
+                    541.0,
+                    437.0,
+                    455.0,
+                    437.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    914.0,
+                    546.0,
+                    968.0,
+                    546.0,
+                    968.0,
+                    586.0,
+                    914.0,
+                    586.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    654.0,
+                    1405.0,
+                    654.0,
+                    1405.0,
+                    694.0,
+                    294.0,
+                    694.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    687.0,
+                    1406.0,
+                    687.0,
+                    1406.0,
+                    723.0,
+                    294.0,
+                    723.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    718.0,
+                    1405.0,
+                    718.0,
+                    1405.0,
+                    754.0,
+                    294.0,
+                    754.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    747.0,
+                    590.0,
+                    747.0,
+                    590.0,
+                    779.0,
+                    295.0,
+                    779.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    1168.0,
+                    778.0,
+                    1168.0,
+                    778.0,
+                    1204.0,
+                    297.0,
+                    1204.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1774.0,
+                    853.0,
+                    1774.0,
+                    853.0,
+                    1811.0,
+                    295.0,
+                    1811.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    73.0,
+                    858.0,
+                    73.0,
+                    858.0,
+                    108.0,
+                    297.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    839.0,
+                    2086.0,
+                    860.0,
+                    2086.0,
+                    860.0,
+                    2116.0,
+                    839.0,
+                    2116.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1454.0,
+                    1404.0,
+                    1454.0,
+                    1404.0,
+                    1494.0,
+                    292.0,
+                    1494.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1490.0,
+                    1401.0,
+                    1490.0,
+                    1401.0,
+                    1522.0,
+                    294.0,
+                    1522.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1517.0,
+                    356.0,
+                    1517.0,
+                    356.0,
+                    1555.0,
+                    292.0,
+                    1555.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    433.0,
+                    1517.0,
+                    1141.0,
+                    1517.0,
+                    1141.0,
+                    1555.0,
+                    433.0,
+                    1555.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1198.0,
+                    1517.0,
+                    1405.0,
+                    1517.0,
+                    1405.0,
+                    1555.0,
+                    1198.0,
+                    1555.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1547.0,
+                    1404.0,
+                    1547.0,
+                    1404.0,
+                    1586.0,
+                    293.0,
+                    1586.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1578.0,
+                    1404.0,
+                    1578.0,
+                    1404.0,
+                    1615.0,
+                    294.0,
+                    1615.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1611.0,
+                    1405.0,
+                    1611.0,
+                    1405.0,
+                    1648.0,
+                    294.0,
+                    1648.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1641.0,
+                    1405.0,
+                    1641.0,
+                    1405.0,
+                    1674.0,
+                    293.0,
+                    1674.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1668.0,
+                    1408.0,
+                    1668.0,
+                    1408.0,
+                    1709.0,
+                    292.0,
+                    1709.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1702.0,
+                    816.0,
+                    1702.0,
+                    816.0,
+                    1735.0,
+                    293.0,
+                    1735.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1229.0,
+                    1402.0,
+                    1229.0,
+                    1402.0,
+                    1263.0,
+                    293.0,
+                    1263.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1257.0,
+                    1403.0,
+                    1257.0,
+                    1403.0,
+                    1291.0,
+                    296.0,
+                    1291.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1289.0,
+                    1405.0,
+                    1289.0,
+                    1405.0,
+                    1323.0,
+                    294.0,
+                    1323.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1320.0,
+                    1407.0,
+                    1320.0,
+                    1407.0,
+                    1354.0,
+                    294.0,
+                    1354.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1350.0,
+                    1405.0,
+                    1350.0,
+                    1405.0,
+                    1385.0,
+                    293.0,
+                    1385.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1381.0,
+                    1404.0,
+                    1381.0,
+                    1404.0,
+                    1413.0,
+                    293.0,
+                    1413.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1410.0,
+                    976.0,
+                    1410.0,
+                    976.0,
+                    1448.0,
+                    293.0,
+                    1448.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    942.0,
+                    1405.0,
+                    942.0,
+                    1405.0,
+                    978.0,
+                    294.0,
+                    978.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    974.0,
+                    1403.0,
+                    974.0,
+                    1403.0,
+                    1006.0,
+                    295.0,
+                    1006.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1005.0,
+                    1405.0,
+                    1005.0,
+                    1405.0,
+                    1040.0,
+                    294.0,
+                    1040.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1034.0,
+                    1404.0,
+                    1034.0,
+                    1404.0,
+                    1068.0,
+                    294.0,
+                    1068.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1065.0,
+                    1405.0,
+                    1065.0,
+                    1405.0,
+                    1100.0,
+                    293.0,
+                    1100.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1097.0,
+                    544.0,
+                    1097.0,
+                    544.0,
+                    1133.0,
+                    294.0,
+                    1133.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1832.0,
+                    1403.0,
+                    1832.0,
+                    1403.0,
+                    1869.0,
+                    294.0,
+                    1869.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1866.0,
+                    1404.0,
+                    1866.0,
+                    1404.0,
+                    1896.0,
+                    295.0,
+                    1896.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1894.0,
+                    567.0,
+                    1894.0,
+                    567.0,
+                    1927.0,
+                    294.0,
+                    1927.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    802.0,
+                    1403.0,
+                    802.0,
+                    1403.0,
+                    839.0,
+                    294.0,
+                    839.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    832.0,
+                    872.0,
+                    832.0,
+                    872.0,
+                    869.0,
+                    294.0,
+                    869.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    975.0,
+                    832.0,
+                    1027.0,
+                    832.0,
+                    1027.0,
+                    869.0,
+                    975.0,
+                    869.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1160.0,
+                    832.0,
+                    1403.0,
+                    832.0,
+                    1403.0,
+                    869.0,
+                    1160.0,
+                    869.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    867.0,
+                    1404.0,
+                    867.0,
+                    1404.0,
+                    899.0,
+                    294.0,
+                    899.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    897.0,
+                    843.0,
+                    897.0,
+                    843.0,
+                    929.0,
+                    294.0,
+                    929.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1941.0,
+                    1406.0,
+                    1941.0,
+                    1406.0,
+                    1978.0,
+                    294.0,
+                    1978.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    1975.0,
+                    1403.0,
+                    1975.0,
+                    1403.0,
+                    2005.0,
+                    297.0,
+                    2005.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    2004.0,
+                    1404.0,
+                    2004.0,
+                    1404.0,
+                    2037.0,
+                    295.0,
+                    2037.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 7,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    741,
+                    1404,
+                    741,
+                    1404,
+                    925,
+                    298,
+                    925
+                ],
+                "score": 0.976
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    941,
+                    1403,
+                    941,
+                    1403,
+                    1093,
+                    298,
+                    1093
+                ],
+                "score": 0.974
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    296,
+                    1111,
+                    1405,
+                    1111,
+                    1405,
+                    1417,
+                    296,
+                    1417
+                ],
+                "score": 0.972
+            },
+            {
+                "category_id": 3,
+                "poly": [
+                    314,
+                    225,
+                    1370,
+                    225,
+                    1370,
+                    629,
+                    314,
+                    629
+                ],
+                "score": 0.967
+            },
+            {
+                "category_id": 3,
+                "poly": [
+                    308,
+                    1531,
+                    1390,
+                    1531,
+                    1390,
+                    1943,
+                    308,
+                    1943
+                ],
+                "score": 0.966
+            },
+            {
+                "category_id": 4,
+                "poly": [
+                    297,
+                    657,
+                    1400,
+                    657,
+                    1400,
+                    721,
+                    297,
+                    721
+                ],
+                "score": 0.952
+            },
+            {
+                "category_id": 4,
+                "poly": [
+                    298,
+                    1972,
+                    1405,
+                    1972,
+                    1405,
+                    2066,
+                    298,
+                    2066
+                ],
+                "score": 0.948
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    322,
+                    1444,
+                    1338,
+                    1444,
+                    1338,
+                    1475,
+                    322,
+                    1475
+                ],
+                "score": 0.863
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    298,
+                    76,
+                    858,
+                    76,
+                    858,
+                    104,
+                    298,
+                    104
+                ],
+                "score": 0.844
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    840,
+                    2087,
+                    859,
+                    2087,
+                    859,
+                    2111,
+                    840,
+                    2111
+                ],
+                "score": 0.794
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    299,
+                    76,
+                    855,
+                    76,
+                    855,
+                    104,
+                    299,
+                    104
+                ],
+                "score": 0.16
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    476,
+                    1064,
+                    551,
+                    1064,
+                    551,
+                    1093,
+                    476,
+                    1093
+                ],
+                "score": 0.87,
+                "latex": "7 3 . 5 \\%"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    989,
+                    1445,
+                    1038,
+                    1445,
+                    1038,
+                    1471,
+                    989,
+                    1471
+                ],
+                "score": 0.86,
+                "latex": "80 \\%"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1083,
+                    1445,
+                    1132,
+                    1445,
+                    1132,
+                    1472,
+                    1083,
+                    1472
+                ],
+                "score": 0.86,
+                "latex": "90 \\%"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    606,
+                    973,
+                    904,
+                    973,
+                    904,
+                    1006,
+                    606,
+                    1006
+                ],
+                "score": 0.83,
+                "latex": "\\Delta T \\in [ 5 0 , 1 0 0 , 5 0 0 , 1 0 0 0 ]"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1192,
+                    973,
+                    1386,
+                    973,
+                    1386,
+                    1006,
+                    1192,
+                    1006
+                ],
+                "score": 0.77,
+                "latex": "\\alpha \\in [ 0 . 1 , 0 . 3 , 0 . 5"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    469.0,
+                    226.0,
+                    755.0,
+                    226.0,
+                    755.0,
+                    265.0,
+                    469.0,
+                    265.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1002.0,
+                    237.0,
+                    1310.0,
+                    237.0,
+                    1310.0,
+                    272.0,
+                    1002.0,
+                    272.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    635.0,
+                    276.0,
+                    817.0,
+                    276.0,
+                    817.0,
+                    307.0,
+                    635.0,
+                    307.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    910.0,
+                    286.0,
+                    964.0,
+                    286.0,
+                    964.0,
+                    320.0,
+                    910.0,
+                    320.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    314.0,
+                    298.0,
+                    359.0,
+                    298.0,
+                    359.0,
+                    518.0,
+                    314.0,
+                    518.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    349.0,
+                    299.0,
+                    392.0,
+                    299.0,
+                    392.0,
+                    325.0,
+                    349.0,
+                    325.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    632.0,
+                    299.0,
+                    757.0,
+                    299.0,
+                    757.0,
+                    343.0,
+                    632.0,
+                    343.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    880.0,
+                    315.0,
+                    922.0,
+                    315.0,
+                    922.0,
+                    522.0,
+                    880.0,
+                    522.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    632.0,
+                    335.0,
+                    724.0,
+                    335.0,
+                    724.0,
+                    371.0,
+                    632.0,
+                    371.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    909.0,
+                    342.0,
+                    964.0,
+                    342.0,
+                    964.0,
+                    367.0,
+                    909.0,
+                    367.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    351.0,
+                    346.0,
+                    392.0,
+                    346.0,
+                    392.0,
+                    372.0,
+                    351.0,
+                    372.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    349.0,
+                    394.0,
+                    394.0,
+                    394.0,
+                    394.0,
+                    421.0,
+                    349.0,
+                    421.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    911.0,
+                    394.0,
+                    961.0,
+                    394.0,
+                    961.0,
+                    420.0,
+                    911.0,
+                    420.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    348.0,
+                    442.0,
+                    393.0,
+                    442.0,
+                    393.0,
+                    468.0,
+                    348.0,
+                    468.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    911.0,
+                    446.0,
+                    962.0,
+                    446.0,
+                    962.0,
+                    472.0,
+                    911.0,
+                    472.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1000.0,
+                    431.0,
+                    1033.0,
+                    431.0,
+                    1033.0,
+                    461.0,
+                    1000.0,
+                    461.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1021.0,
+                    461.0,
+                    1062.0,
+                    461.0,
+                    1062.0,
+                    494.0,
+                    1021.0,
+                    494.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    348.0,
+                    490.0,
+                    395.0,
+                    490.0,
+                    395.0,
+                    518.0,
+                    348.0,
+                    518.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    911.0,
+                    497.0,
+                    962.0,
+                    497.0,
+                    962.0,
+                    523.0,
+                    911.0,
+                    523.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1020.0,
+                    489.0,
+                    1065.0,
+                    489.0,
+                    1065.0,
+                    521.0,
+                    1020.0,
+                    521.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1017.0,
+                    519.0,
+                    1065.0,
+                    519.0,
+                    1065.0,
+                    550.0,
+                    1017.0,
+                    550.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    348.0,
+                    535.0,
+                    395.0,
+                    535.0,
+                    395.0,
+                    565.0,
+                    348.0,
+                    565.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    427.0,
+                    551.0,
+                    463.0,
+                    551.0,
+                    463.0,
+                    585.0,
+                    427.0,
+                    585.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    535.0,
+                    551.0,
+                    573.0,
+                    551.0,
+                    573.0,
+                    586.0,
+                    535.0,
+                    586.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    909.0,
+                    548.0,
+                    964.0,
+                    548.0,
+                    964.0,
+                    573.0,
+                    909.0,
+                    573.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1044.0,
+                    559.0,
+                    1079.0,
+                    559.0,
+                    1079.0,
+                    586.0,
+                    1044.0,
+                    586.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1355.0,
+                    566.0,
+                    1366.0,
+                    566.0,
+                    1366.0,
+                    575.0,
+                    1355.0,
+                    575.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    545.0,
+                    580.0,
+                    679.0,
+                    580.0,
+                    679.0,
+                    638.0,
+                    545.0,
+                    638.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1045.0,
+                    577.0,
+                    1267.0,
+                    577.0,
+                    1267.0,
+                    617.0,
+                    1045.0,
+                    617.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    632.75,
+                    553.5,
+                    694.75,
+                    553.5,
+                    694.75,
+                    584.5,
+                    632.75,
+                    584.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    740.25,
+                    557.0,
+                    807.25,
+                    557.0,
+                    807.25,
+                    586.0,
+                    740.25,
+                    586.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    328.0,
+                    1529.0,
+                    733.0,
+                    1529.0,
+                    733.0,
+                    1568.0,
+                    328.0,
+                    1568.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1327.0,
+                    1545.0,
+                    1376.0,
+                    1545.0,
+                    1376.0,
+                    1576.0,
+                    1327.0,
+                    1576.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    405.0,
+                    1575.0,
+                    477.0,
+                    1575.0,
+                    477.0,
+                    1610.0,
+                    405.0,
+                    1610.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    589.0,
+                    1577.0,
+                    643.0,
+                    1577.0,
+                    643.0,
+                    1607.0,
+                    589.0,
+                    1607.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    787.0,
+                    1578.0,
+                    822.0,
+                    1578.0,
+                    822.0,
+                    1605.0,
+                    787.0,
+                    1605.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1132.0,
+                    1582.0,
+                    1193.0,
+                    1582.0,
+                    1193.0,
+                    1623.0,
+                    1132.0,
+                    1623.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    855.0,
+                    1615.0,
+                    872.0,
+                    1615.0,
+                    872.0,
+                    1629.0,
+                    855.0,
+                    1629.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1135.0,
+                    1615.0,
+                    1204.0,
+                    1615.0,
+                    1204.0,
+                    1650.0,
+                    1135.0,
+                    1650.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    753.0,
+                    1624.0,
+                    806.0,
+                    1624.0,
+                    806.0,
+                    1815.0,
+                    753.0,
+                    1815.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    904.0,
+                    1624.0,
+                    925.0,
+                    1624.0,
+                    925.0,
+                    1636.0,
+                    904.0,
+                    1636.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    343.0,
+                    1641.0,
+                    354.0,
+                    1641.0,
+                    354.0,
+                    1653.0,
+                    343.0,
+                    1653.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    308.0,
+                    1643.0,
+                    353.0,
+                    1643.0,
+                    353.0,
+                    1810.0,
+                    308.0,
+                    1810.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    789.0,
+                    1646.0,
+                    819.0,
+                    1646.0,
+                    819.0,
+                    1671.0,
+                    789.0,
+                    1671.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    519.0,
+                    1690.0,
+                    576.0,
+                    1690.0,
+                    576.0,
+                    1732.0,
+                    519.0,
+                    1732.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    790.0,
+                    1712.0,
+                    819.0,
+                    1712.0,
+                    819.0,
+                    1736.0,
+                    790.0,
+                    1736.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1248.0,
+                    1698.0,
+                    1298.0,
+                    1698.0,
+                    1298.0,
+                    1728.0,
+                    1248.0,
+                    1728.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    518.0,
+                    1727.0,
+                    576.0,
+                    1727.0,
+                    576.0,
+                    1752.0,
+                    518.0,
+                    1752.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    518.0,
+                    1746.0,
+                    570.0,
+                    1746.0,
+                    570.0,
+                    1771.0,
+                    518.0,
+                    1771.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1137.0,
+                    1821.0,
+                    1164.0,
+                    1821.0,
+                    1164.0,
+                    1838.0,
+                    1137.0,
+                    1838.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    791.0,
+                    1847.0,
+                    817.0,
+                    1847.0,
+                    817.0,
+                    1865.0,
+                    791.0,
+                    1865.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    998.0,
+                    1877.0,
+                    1014.0,
+                    1877.0,
+                    1014.0,
+                    1895.0,
+                    998.0,
+                    1895.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1062.0,
+                    1877.0,
+                    1078.0,
+                    1877.0,
+                    1078.0,
+                    1895.0,
+                    1062.0,
+                    1895.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1122.0,
+                    1876.0,
+                    1142.0,
+                    1876.0,
+                    1142.0,
+                    1897.0,
+                    1122.0,
+                    1897.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1188.0,
+                    1879.0,
+                    1204.0,
+                    1879.0,
+                    1204.0,
+                    1895.0,
+                    1188.0,
+                    1895.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1244.0,
+                    1875.0,
+                    1377.0,
+                    1875.0,
+                    1377.0,
+                    1915.0,
+                    1244.0,
+                    1915.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    390.0,
+                    1896.0,
+                    419.0,
+                    1896.0,
+                    419.0,
+                    1918.0,
+                    390.0,
+                    1918.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    441.0,
+                    1898.0,
+                    467.0,
+                    1898.0,
+                    467.0,
+                    1915.0,
+                    441.0,
+                    1915.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    491.0,
+                    1898.0,
+                    518.0,
+                    1898.0,
+                    518.0,
+                    1915.0,
+                    491.0,
+                    1915.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    541.0,
+                    1898.0,
+                    568.0,
+                    1898.0,
+                    568.0,
+                    1917.0,
+                    541.0,
+                    1917.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    589.0,
+                    1896.0,
+                    619.0,
+                    1896.0,
+                    619.0,
+                    1918.0,
+                    589.0,
+                    1918.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    641.0,
+                    1896.0,
+                    670.0,
+                    1896.0,
+                    670.0,
+                    1918.0,
+                    641.0,
+                    1918.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    693.0,
+                    1898.0,
+                    717.0,
+                    1898.0,
+                    717.0,
+                    1915.0,
+                    693.0,
+                    1915.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    956.0,
+                    1892.0,
+                    1079.0,
+                    1892.0,
+                    1079.0,
+                    1929.0,
+                    956.0,
+                    1929.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1157.0,
+                    1886.0,
+                    1225.0,
+                    1886.0,
+                    1225.0,
+                    1930.0,
+                    1157.0,
+                    1930.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    386.0,
+                    1911.0,
+                    674.0,
+                    1911.0,
+                    674.0,
+                    1945.0,
+                    386.0,
+                    1945.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    344.0,
+                    1892.0,
+                    367.0,
+                    1892.0,
+                    367.0,
+                    1916.0,
+                    344.0,
+                    1916.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    652.0,
+                    1404.0,
+                    652.0,
+                    1404.0,
+                    695.0,
+                    293.0,
+                    695.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    686.0,
+                    1087.0,
+                    686.0,
+                    1087.0,
+                    722.0,
+                    296.0,
+                    722.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1973.0,
+                    1403.0,
+                    1973.0,
+                    1403.0,
+                    2007.0,
+                    296.0,
+                    2007.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    2003.0,
+                    1406.0,
+                    2003.0,
+                    1406.0,
+                    2038.0,
+                    292.0,
+                    2038.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    2034.0,
+                    1192.0,
+                    2034.0,
+                    1192.0,
+                    2068.0,
+                    295.0,
+                    2068.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    336.0,
+                    1441.0,
+                    988.0,
+                    1441.0,
+                    988.0,
+                    1479.0,
+                    336.0,
+                    1479.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1039.0,
+                    1441.0,
+                    1082.0,
+                    1441.0,
+                    1082.0,
+                    1479.0,
+                    1039.0,
+                    1479.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1133.0,
+                    1441.0,
+                    1341.0,
+                    1441.0,
+                    1341.0,
+                    1479.0,
+                    1133.0,
+                    1479.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    70.0,
+                    859.0,
+                    70.0,
+                    859.0,
+                    109.0,
+                    296.0,
+                    109.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    839.0,
+                    2086.0,
+                    861.0,
+                    2086.0,
+                    861.0,
+                    2116.0,
+                    839.0,
+                    2116.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    73.0,
+                    859.0,
+                    73.0,
+                    859.0,
+                    108.0,
+                    297.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    740.0,
+                    1405.0,
+                    740.0,
+                    1405.0,
+                    776.0,
+                    293.0,
+                    776.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    773.0,
+                    1404.0,
+                    773.0,
+                    1404.0,
+                    808.0,
+                    292.0,
+                    808.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    804.0,
+                    1405.0,
+                    804.0,
+                    1405.0,
+                    839.0,
+                    294.0,
+                    839.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    834.0,
+                    1406.0,
+                    834.0,
+                    1406.0,
+                    870.0,
+                    294.0,
+                    870.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    863.0,
+                    1406.0,
+                    863.0,
+                    1406.0,
+                    899.0,
+                    292.0,
+                    899.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    900.0,
+                    431.0,
+                    900.0,
+                    431.0,
+                    927.0,
+                    294.0,
+                    927.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    941.0,
+                    1405.0,
+                    941.0,
+                    1405.0,
+                    978.0,
+                    294.0,
+                    978.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    971.0,
+                    605.0,
+                    971.0,
+                    605.0,
+                    1008.0,
+                    294.0,
+                    1008.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    905.0,
+                    971.0,
+                    1191.0,
+                    971.0,
+                    1191.0,
+                    1008.0,
+                    905.0,
+                    1008.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1387.0,
+                    971.0,
+                    1404.0,
+                    971.0,
+                    1404.0,
+                    1008.0,
+                    1387.0,
+                    1008.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    997.0,
+                    1407.0,
+                    997.0,
+                    1407.0,
+                    1042.0,
+                    292.0,
+                    1042.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1030.0,
+                    1407.0,
+                    1030.0,
+                    1407.0,
+                    1070.0,
+                    293.0,
+                    1070.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1063.0,
+                    475.0,
+                    1063.0,
+                    475.0,
+                    1094.0,
+                    293.0,
+                    1094.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    552.0,
+                    1063.0,
+                    562.0,
+                    1063.0,
+                    562.0,
+                    1094.0,
+                    552.0,
+                    1094.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1110.0,
+                    1406.0,
+                    1110.0,
+                    1406.0,
+                    1146.0,
+                    294.0,
+                    1146.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1140.0,
+                    1405.0,
+                    1140.0,
+                    1405.0,
+                    1178.0,
+                    293.0,
+                    1178.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1166.0,
+                    1406.0,
+                    1166.0,
+                    1406.0,
+                    1211.0,
+                    291.0,
+                    1211.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1202.0,
+                    1406.0,
+                    1202.0,
+                    1406.0,
+                    1237.0,
+                    294.0,
+                    1237.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1231.0,
+                    1405.0,
+                    1231.0,
+                    1405.0,
+                    1267.0,
+                    294.0,
+                    1267.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1264.0,
+                    1405.0,
+                    1264.0,
+                    1405.0,
+                    1300.0,
+                    294.0,
+                    1300.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1292.0,
+                    1405.0,
+                    1292.0,
+                    1405.0,
+                    1327.0,
+                    294.0,
+                    1327.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1324.0,
+                    1407.0,
+                    1324.0,
+                    1407.0,
+                    1359.0,
+                    293.0,
+                    1359.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1354.0,
+                    1406.0,
+                    1354.0,
+                    1406.0,
+                    1390.0,
+                    294.0,
+                    1390.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1383.0,
+                    1405.0,
+                    1383.0,
+                    1405.0,
+                    1421.0,
+                    294.0,
+                    1421.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 8,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    429,
+                    1403,
+                    429,
+                    1403,
+                    644,
+                    298,
+                    644
+                ],
+                "score": 0.977
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    989,
+                    1403,
+                    989,
+                    1403,
+                    1174,
+                    298,
+                    1174
+                ],
+                "score": 0.973
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    229,
+                    1403,
+                    229,
+                    1403,
+                    414,
+                    299,
+                    414
+                ],
+                "score": 0.973
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    760,
+                    1403,
+                    760,
+                    1403,
+                    973,
+                    298,
+                    973
+                ],
+                "score": 0.971
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    302,
+                    689,
+                    763,
+                    689,
+                    763,
+                    724,
+                    302,
+                    724
+                ],
+                "score": 0.892
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    299,
+                    1221,
+                    488,
+                    1221,
+                    488,
+                    1254,
+                    299,
+                    1254
+                ],
+                "score": 0.881
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    298,
+                    76,
+                    857,
+                    76,
+                    857,
+                    104,
+                    298,
+                    104
+                ],
+                "score": 0.875
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    837,
+                    2089,
+                    865,
+                    2089,
+                    865,
+                    2112,
+                    837,
+                    2112
+                ],
+                "score": 0.839
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    301,
+                    1273,
+                    1404,
+                    1273,
+                    1404,
+                    1518,
+                    301,
+                    1518
+                ],
+                "score": 0.827
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    302,
+                    1655,
+                    1400,
+                    1655,
+                    1400,
+                    1749,
+                    302,
+                    1749
+                ],
+                "score": 0.732
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1541,
+                    1405,
+                    1541,
+                    1405,
+                    1632,
+                    298,
+                    1632
+                ],
+                "score": 0.702
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    302,
+                    1972,
+                    1398,
+                    1972,
+                    1398,
+                    2034,
+                    302,
+                    2034
+                ],
+                "score": 0.68
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    301,
+                    1770,
+                    1398,
+                    1770,
+                    1398,
+                    1865,
+                    301,
+                    1865
+                ],
+                "score": 0.579
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    296,
+                    1886,
+                    1393,
+                    1886,
+                    1393,
+                    1950,
+                    296,
+                    1950
+                ],
+                "score": 0.437
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1376,
+                    1889,
+                    1401,
+                    1889,
+                    1401,
+                    1919,
+                    1376,
+                    1919
+                ],
+                "score": 0.81,
+                "latex": "l _ { 0 }"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    683.0,
+                    770.0,
+                    683.0,
+                    770.0,
+                    733.0,
+                    291.0,
+                    733.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    1221.0,
+                    490.0,
+                    1221.0,
+                    490.0,
+                    1258.0,
+                    297.0,
+                    1258.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    298.0,
+                    73.0,
+                    858.0,
+                    73.0,
+                    858.0,
+                    108.0,
+                    298.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    832.0,
+                    2085.0,
+                    870.0,
+                    2085.0,
+                    870.0,
+                    2124.0,
+                    832.0,
+                    2124.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    428.0,
+                    1404.0,
+                    428.0,
+                    1404.0,
+                    466.0,
+                    293.0,
+                    466.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    460.0,
+                    1404.0,
+                    460.0,
+                    1404.0,
+                    495.0,
+                    294.0,
+                    495.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    491.0,
+                    1405.0,
+                    491.0,
+                    1405.0,
+                    525.0,
+                    294.0,
+                    525.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    520.0,
+                    1404.0,
+                    520.0,
+                    1404.0,
+                    556.0,
+                    292.0,
+                    556.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    550.0,
+                    1405.0,
+                    550.0,
+                    1405.0,
+                    586.0,
+                    292.0,
+                    586.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    580.0,
+                    1404.0,
+                    580.0,
+                    1404.0,
+                    620.0,
+                    292.0,
+                    620.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    614.0,
+                    951.0,
+                    614.0,
+                    951.0,
+                    648.0,
+                    294.0,
+                    648.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    986.0,
+                    1404.0,
+                    986.0,
+                    1404.0,
+                    1024.0,
+                    294.0,
+                    1024.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1019.0,
+                    1405.0,
+                    1019.0,
+                    1405.0,
+                    1055.0,
+                    294.0,
+                    1055.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1050.0,
+                    1405.0,
+                    1050.0,
+                    1405.0,
+                    1085.0,
+                    294.0,
+                    1085.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1079.0,
+                    1407.0,
+                    1079.0,
+                    1407.0,
+                    1118.0,
+                    293.0,
+                    1118.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1109.0,
+                    1405.0,
+                    1109.0,
+                    1405.0,
+                    1146.0,
+                    292.0,
+                    1146.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1141.0,
+                    513.0,
+                    1141.0,
+                    513.0,
+                    1177.0,
+                    293.0,
+                    1177.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    231.0,
+                    1404.0,
+                    231.0,
+                    1404.0,
+                    263.0,
+                    295.0,
+                    263.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    263.0,
+                    1404.0,
+                    263.0,
+                    1404.0,
+                    295.0,
+                    295.0,
+                    295.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    291.0,
+                    1404.0,
+                    291.0,
+                    1404.0,
+                    325.0,
+                    293.0,
+                    325.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    323.0,
+                    1403.0,
+                    323.0,
+                    1403.0,
+                    355.0,
+                    297.0,
+                    355.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    355.0,
+                    1404.0,
+                    355.0,
+                    1404.0,
+                    386.0,
+                    295.0,
+                    386.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    381.0,
+                    1029.0,
+                    381.0,
+                    1029.0,
+                    421.0,
+                    294.0,
+                    421.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    758.0,
+                    1407.0,
+                    758.0,
+                    1407.0,
+                    797.0,
+                    292.0,
+                    797.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    791.0,
+                    1404.0,
+                    791.0,
+                    1404.0,
+                    825.0,
+                    294.0,
+                    825.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    820.0,
+                    1404.0,
+                    820.0,
+                    1404.0,
+                    857.0,
+                    293.0,
+                    857.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    849.0,
+                    1404.0,
+                    849.0,
+                    1404.0,
+                    890.0,
+                    292.0,
+                    890.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    879.0,
+                    1404.0,
+                    879.0,
+                    1404.0,
+                    919.0,
+                    292.0,
+                    919.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    912.0,
+                    1407.0,
+                    912.0,
+                    1407.0,
+                    950.0,
+                    293.0,
+                    950.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    944.0,
+                    553.0,
+                    944.0,
+                    553.0,
+                    978.0,
+                    294.0,
+                    978.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1272.0,
+                    1404.0,
+                    1272.0,
+                    1404.0,
+                    1309.0,
+                    295.0,
+                    1309.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    324.0,
+                    1304.0,
+                    1405.0,
+                    1304.0,
+                    1405.0,
+                    1338.0,
+                    324.0,
+                    1338.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    321.0,
+                    1333.0,
+                    1405.0,
+                    1333.0,
+                    1405.0,
+                    1370.0,
+                    321.0,
+                    1370.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    320.0,
+                    1362.0,
+                    1405.0,
+                    1362.0,
+                    1405.0,
+                    1400.0,
+                    320.0,
+                    1400.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    1395.0,
+                    1404.0,
+                    1395.0,
+                    1404.0,
+                    1429.0,
+                    322.0,
+                    1429.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    1425.0,
+                    1404.0,
+                    1425.0,
+                    1404.0,
+                    1459.0,
+                    322.0,
+                    1459.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    320.0,
+                    1453.0,
+                    1405.0,
+                    1453.0,
+                    1405.0,
+                    1493.0,
+                    320.0,
+                    1493.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    1488.0,
+                    1141.0,
+                    1488.0,
+                    1141.0,
+                    1521.0,
+                    322.0,
+                    1521.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1653.0,
+                    1405.0,
+                    1653.0,
+                    1405.0,
+                    1692.0,
+                    294.0,
+                    1692.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    1687.0,
+                    1405.0,
+                    1687.0,
+                    1405.0,
+                    1721.0,
+                    322.0,
+                    1721.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    1718.0,
+                    1148.0,
+                    1718.0,
+                    1148.0,
+                    1749.0,
+                    322.0,
+                    1749.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1536.0,
+                    1403.0,
+                    1536.0,
+                    1403.0,
+                    1578.0,
+                    295.0,
+                    1578.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    1570.0,
+                    1403.0,
+                    1570.0,
+                    1403.0,
+                    1607.0,
+                    323.0,
+                    1607.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    1602.0,
+                    639.0,
+                    1602.0,
+                    639.0,
+                    1633.0,
+                    323.0,
+                    1633.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1966.0,
+                    1404.0,
+                    1966.0,
+                    1404.0,
+                    2011.0,
+                    296.0,
+                    2011.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    2002.0,
+                    1226.0,
+                    2002.0,
+                    1226.0,
+                    2036.0,
+                    322.0,
+                    2036.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1770.0,
+                    1403.0,
+                    1770.0,
+                    1403.0,
+                    1807.0,
+                    295.0,
+                    1807.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    320.0,
+                    1802.0,
+                    1404.0,
+                    1802.0,
+                    1404.0,
+                    1838.0,
+                    320.0,
+                    1838.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    1833.0,
+                    1363.0,
+                    1833.0,
+                    1363.0,
+                    1867.0,
+                    323.0,
+                    1867.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1885.0,
+                    1375.0,
+                    1885.0,
+                    1375.0,
+                    1923.0,
+                    293.0,
+                    1923.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    1917.0,
+                    1216.0,
+                    1917.0,
+                    1216.0,
+                    1951.0,
+                    322.0,
+                    1951.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 9,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 2,
+                "poly": [
+                    298,
+                    75,
+                    858,
+                    75,
+                    858,
+                    105,
+                    298,
+                    105
+                ],
+                "score": 0.895
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    836,
+                    2088,
+                    863,
+                    2088,
+                    863,
+                    2113,
+                    836,
+                    2113
+                ],
+                "score": 0.817
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1429,
+                    1401,
+                    1429,
+                    1401,
+                    1491,
+                    298,
+                    1491
+                ],
+                "score": 0.762
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    1009,
+                    1399,
+                    1009,
+                    1399,
+                    1074,
+                    299,
+                    1074
+                ],
+                "score": 0.761
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    293,
+                    926,
+                    1400,
+                    926,
+                    1400,
+                    991,
+                    293,
+                    991
+                ],
+                "score": 0.754
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    293,
+                    424,
+                    1402,
+                    424,
+                    1402,
+                    488,
+                    293,
+                    488
+                ],
+                "score": 0.72
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    229,
+                    1400,
+                    229,
+                    1400,
+                    324,
+                    298,
+                    324
+                ],
+                "score": 0.719
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1090,
+                    1401,
+                    1090,
+                    1401,
+                    1156,
+                    297,
+                    1156
+                ],
+                "score": 0.709
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    1911,
+                    1404,
+                    1911,
+                    1404,
+                    2033,
+                    299,
+                    2033
+                ],
+                "score": 0.706
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    732,
+                    1399,
+                    732,
+                    1399,
+                    797,
+                    297,
+                    797
+                ],
+                "score": 0.688
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1285,
+                    1404,
+                    1285,
+                    1404,
+                    1411,
+                    297,
+                    1411
+                ],
+                "score": 0.686
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    303,
+                    1172,
+                    1401,
+                    1172,
+                    1401,
+                    1268,
+                    303,
+                    1268
+                ],
+                "score": 0.67
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    294,
+                    648,
+                    1400,
+                    648,
+                    1400,
+                    715,
+                    294,
+                    715
+                ],
+                "score": 0.663
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    300,
+                    1625,
+                    1402,
+                    1625,
+                    1402,
+                    1718,
+                    300,
+                    1718
+                ],
+                "score": 0.662
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    293,
+                    342,
+                    1397,
+                    342,
+                    1397,
+                    405,
+                    293,
+                    405
+                ],
+                "score": 0.637
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    295,
+                    1511,
+                    1402,
+                    1511,
+                    1402,
+                    1606,
+                    295,
+                    1606
+                ],
+                "score": 0.631
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1737,
+                    1404,
+                    1737,
+                    1404,
+                    1890,
+                    298,
+                    1890
+                ],
+                "score": 0.596
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    296,
+                    506,
+                    1405,
+                    506,
+                    1405,
+                    631,
+                    296,
+                    631
+                ],
+                "score": 0.595
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    814,
+                    1403,
+                    814,
+                    1403,
+                    908,
+                    297,
+                    908
+                ],
+                "score": 0.588
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1383,
+                    572,
+                    1403,
+                    572,
+                    1403,
+                    595,
+                    1383,
+                    595
+                ],
+                "score": 0.34,
+                "latex": "="
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    298.0,
+                    73.0,
+                    858.0,
+                    73.0,
+                    858.0,
+                    108.0,
+                    298.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    832.0,
+                    2085.0,
+                    868.0,
+                    2085.0,
+                    868.0,
+                    2125.0,
+                    832.0,
+                    2125.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1423.0,
+                    1402.0,
+                    1423.0,
+                    1402.0,
+                    1468.0,
+                    291.0,
+                    1468.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    321.0,
+                    1456.0,
+                    398.0,
+                    1456.0,
+                    398.0,
+                    1495.0,
+                    321.0,
+                    1495.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    298.0,
+                    1009.0,
+                    1404.0,
+                    1009.0,
+                    1404.0,
+                    1045.0,
+                    298.0,
+                    1045.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    324.0,
+                    1039.0,
+                    1179.0,
+                    1039.0,
+                    1179.0,
+                    1075.0,
+                    324.0,
+                    1075.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    921.0,
+                    1405.0,
+                    921.0,
+                    1405.0,
+                    966.0,
+                    291.0,
+                    966.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    957.0,
+                    1030.0,
+                    957.0,
+                    1030.0,
+                    993.0,
+                    323.0,
+                    993.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    421.0,
+                    1404.0,
+                    421.0,
+                    1404.0,
+                    461.0,
+                    295.0,
+                    461.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    324.0,
+                    455.0,
+                    1263.0,
+                    455.0,
+                    1263.0,
+                    491.0,
+                    324.0,
+                    491.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    228.0,
+                    1402.0,
+                    228.0,
+                    1402.0,
+                    264.0,
+                    293.0,
+                    264.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    319.0,
+                    257.0,
+                    1405.0,
+                    257.0,
+                    1405.0,
+                    300.0,
+                    319.0,
+                    300.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    321.0,
+                    289.0,
+                    401.0,
+                    289.0,
+                    401.0,
+                    325.0,
+                    321.0,
+                    325.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1089.0,
+                    1403.0,
+                    1089.0,
+                    1403.0,
+                    1126.0,
+                    295.0,
+                    1126.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    324.0,
+                    1120.0,
+                    1287.0,
+                    1120.0,
+                    1287.0,
+                    1157.0,
+                    324.0,
+                    1157.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1910.0,
+                    1405.0,
+                    1910.0,
+                    1405.0,
+                    1945.0,
+                    294.0,
+                    1945.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    1943.0,
+                    1405.0,
+                    1943.0,
+                    1405.0,
+                    1975.0,
+                    323.0,
+                    1975.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    1973.0,
+                    1403.0,
+                    1973.0,
+                    1403.0,
+                    2006.0,
+                    322.0,
+                    2006.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    2004.0,
+                    506.0,
+                    2004.0,
+                    506.0,
+                    2034.0,
+                    323.0,
+                    2034.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    730.0,
+                    1404.0,
+                    730.0,
+                    1404.0,
+                    768.0,
+                    295.0,
+                    768.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    761.0,
+                    1179.0,
+                    761.0,
+                    1179.0,
+                    798.0,
+                    322.0,
+                    798.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1286.0,
+                    1405.0,
+                    1286.0,
+                    1405.0,
+                    1323.0,
+                    292.0,
+                    1323.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    321.0,
+                    1316.0,
+                    1406.0,
+                    1316.0,
+                    1406.0,
+                    1354.0,
+                    321.0,
+                    1354.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    324.0,
+                    1350.0,
+                    1406.0,
+                    1350.0,
+                    1406.0,
+                    1383.0,
+                    324.0,
+                    1383.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    324.0,
+                    1380.0,
+                    1155.0,
+                    1380.0,
+                    1155.0,
+                    1413.0,
+                    324.0,
+                    1413.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1172.0,
+                    1405.0,
+                    1172.0,
+                    1405.0,
+                    1210.0,
+                    295.0,
+                    1210.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    320.0,
+                    1202.0,
+                    1404.0,
+                    1202.0,
+                    1404.0,
+                    1241.0,
+                    320.0,
+                    1241.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    1233.0,
+                    1053.0,
+                    1233.0,
+                    1053.0,
+                    1270.0,
+                    322.0,
+                    1270.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    645.0,
+                    1404.0,
+                    645.0,
+                    1404.0,
+                    688.0,
+                    294.0,
+                    688.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    324.0,
+                    680.0,
+                    1358.0,
+                    680.0,
+                    1358.0,
+                    717.0,
+                    324.0,
+                    717.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1623.0,
+                    1404.0,
+                    1623.0,
+                    1404.0,
+                    1661.0,
+                    295.0,
+                    1661.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    1651.0,
+                    1402.0,
+                    1651.0,
+                    1402.0,
+                    1691.0,
+                    323.0,
+                    1691.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    325.0,
+                    1687.0,
+                    904.0,
+                    1687.0,
+                    904.0,
+                    1721.0,
+                    325.0,
+                    1721.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    337.0,
+                    1403.0,
+                    337.0,
+                    1403.0,
+                    381.0,
+                    293.0,
+                    381.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    326.0,
+                    375.0,
+                    771.0,
+                    375.0,
+                    771.0,
+                    407.0,
+                    326.0,
+                    407.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1507.0,
+                    1406.0,
+                    1507.0,
+                    1406.0,
+                    1552.0,
+                    293.0,
+                    1552.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    1544.0,
+                    1404.0,
+                    1544.0,
+                    1404.0,
+                    1578.0,
+                    322.0,
+                    1578.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    1573.0,
+                    923.0,
+                    1573.0,
+                    923.0,
+                    1607.0,
+                    322.0,
+                    1607.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    1740.0,
+                    1404.0,
+                    1740.0,
+                    1404.0,
+                    1773.0,
+                    297.0,
+                    1773.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    320.0,
+                    1769.0,
+                    1406.0,
+                    1769.0,
+                    1406.0,
+                    1804.0,
+                    320.0,
+                    1804.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    321.0,
+                    1798.0,
+                    1404.0,
+                    1798.0,
+                    1404.0,
+                    1835.0,
+                    321.0,
+                    1835.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    320.0,
+                    1823.0,
+                    1404.0,
+                    1823.0,
+                    1404.0,
+                    1869.0,
+                    320.0,
+                    1869.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    1861.0,
+                    787.0,
+                    1861.0,
+                    787.0,
+                    1894.0,
+                    323.0,
+                    1894.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    504.0,
+                    1405.0,
+                    504.0,
+                    1405.0,
+                    544.0,
+                    293.0,
+                    544.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    539.0,
+                    1406.0,
+                    539.0,
+                    1406.0,
+                    572.0,
+                    323.0,
+                    572.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    567.0,
+                    1382.0,
+                    567.0,
+                    1382.0,
+                    604.0,
+                    323.0,
+                    604.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    324.0,
+                    601.0,
+                    508.0,
+                    601.0,
+                    508.0,
+                    631.0,
+                    324.0,
+                    631.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    812.0,
+                    1402.0,
+                    812.0,
+                    1402.0,
+                    849.0,
+                    295.0,
+                    849.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    845.0,
+                    1404.0,
+                    845.0,
+                    1404.0,
+                    879.0,
+                    322.0,
+                    879.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    876.0,
+                    625.0,
+                    876.0,
+                    625.0,
+                    909.0,
+                    322.0,
+                    909.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 10,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 2,
+                "poly": [
+                    298,
+                    75,
+                    858,
+                    75,
+                    858,
+                    105,
+                    298,
+                    105
+                ],
+                "score": 0.895
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1004,
+                    1405,
+                    1004,
+                    1405,
+                    1159,
+                    297,
+                    1159
+                ],
+                "score": 0.87
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    697,
+                    1405,
+                    697,
+                    1405,
+                    821,
+                    297,
+                    821
+                ],
+                "score": 0.864
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    295,
+                    616,
+                    1401,
+                    616,
+                    1401,
+                    681,
+                    295,
+                    681
+                ],
+                "score": 0.843
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    1942,
+                    1400,
+                    1942,
+                    1400,
+                    2034,
+                    299,
+                    2034
+                ],
+                "score": 0.839
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    295,
+                    503,
+                    1402,
+                    503,
+                    1402,
+                    598,
+                    295,
+                    598
+                ],
+                "score": 0.838
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    296,
+                    422,
+                    1402,
+                    422,
+                    1402,
+                    487,
+                    296,
+                    487
+                ],
+                "score": 0.83
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    836,
+                    2088,
+                    865,
+                    2088,
+                    865,
+                    2113,
+                    836,
+                    2113
+                ],
+                "score": 0.829
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    296,
+                    1178,
+                    1404,
+                    1178,
+                    1404,
+                    1334,
+                    296,
+                    1334
+                ],
+                "score": 0.827
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1350,
+                    1401,
+                    1350,
+                    1401,
+                    1444,
+                    298,
+                    1444
+                ],
+                "score": 0.824
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    290,
+                    1574,
+                    1400,
+                    1574,
+                    1400,
+                    1638,
+                    290,
+                    1638
+                ],
+                "score": 0.814
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    295,
+                    922,
+                    1400,
+                    922,
+                    1400,
+                    987,
+                    295,
+                    987
+                ],
+                "score": 0.797
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1461,
+                    1399,
+                    1461,
+                    1399,
+                    1558,
+                    298,
+                    1558
+                ],
+                "score": 0.779
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1798,
+                    1404,
+                    1798,
+                    1404,
+                    1923,
+                    298,
+                    1923
+                ],
+                "score": 0.773
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1655,
+                    1404,
+                    1655,
+                    1404,
+                    1781,
+                    297,
+                    1781
+                ],
+                "score": 0.768
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    300,
+                    229,
+                    1399,
+                    229,
+                    1399,
+                    324,
+                    300,
+                    324
+                ],
+                "score": 0.751
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    296,
+                    340,
+                    1399,
+                    340,
+                    1399,
+                    405,
+                    296,
+                    405
+                ],
+                "score": 0.741
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    841,
+                    1398,
+                    841,
+                    1398,
+                    905,
+                    298,
+                    905
+                ],
+                "score": 0.713
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    298.0,
+                    73.0,
+                    858.0,
+                    73.0,
+                    858.0,
+                    108.0,
+                    298.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    832.0,
+                    2085.0,
+                    870.0,
+                    2085.0,
+                    870.0,
+                    2124.0,
+                    832.0,
+                    2124.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1002.0,
+                    1405.0,
+                    1002.0,
+                    1405.0,
+                    1044.0,
+                    294.0,
+                    1044.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    324.0,
+                    1036.0,
+                    1405.0,
+                    1036.0,
+                    1405.0,
+                    1070.0,
+                    324.0,
+                    1070.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    1066.0,
+                    1405.0,
+                    1066.0,
+                    1405.0,
+                    1103.0,
+                    323.0,
+                    1103.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    320.0,
+                    1092.0,
+                    1403.0,
+                    1092.0,
+                    1403.0,
+                    1133.0,
+                    320.0,
+                    1133.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    321.0,
+                    1127.0,
+                    720.0,
+                    1127.0,
+                    720.0,
+                    1161.0,
+                    321.0,
+                    1161.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    698.0,
+                    1403.0,
+                    698.0,
+                    1403.0,
+                    735.0,
+                    295.0,
+                    735.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    324.0,
+                    731.0,
+                    1405.0,
+                    731.0,
+                    1405.0,
+                    764.0,
+                    324.0,
+                    764.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    321.0,
+                    759.0,
+                    1406.0,
+                    759.0,
+                    1406.0,
+                    796.0,
+                    321.0,
+                    796.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    320.0,
+                    788.0,
+                    398.0,
+                    788.0,
+                    398.0,
+                    824.0,
+                    320.0,
+                    824.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    616.0,
+                    1405.0,
+                    616.0,
+                    1405.0,
+                    652.0,
+                    295.0,
+                    652.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    324.0,
+                    647.0,
+                    1255.0,
+                    647.0,
+                    1255.0,
+                    683.0,
+                    324.0,
+                    683.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1940.0,
+                    1405.0,
+                    1940.0,
+                    1405.0,
+                    1977.0,
+                    294.0,
+                    1977.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    1971.0,
+                    1401.0,
+                    1971.0,
+                    1401.0,
+                    2008.0,
+                    322.0,
+                    2008.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    327.0,
+                    2005.0,
+                    771.0,
+                    2005.0,
+                    771.0,
+                    2035.0,
+                    327.0,
+                    2035.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    505.0,
+                    1400.0,
+                    505.0,
+                    1400.0,
+                    540.0,
+                    295.0,
+                    540.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    536.0,
+                    1403.0,
+                    536.0,
+                    1403.0,
+                    570.0,
+                    323.0,
+                    570.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    568.0,
+                    554.0,
+                    568.0,
+                    554.0,
+                    599.0,
+                    323.0,
+                    599.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    422.0,
+                    1406.0,
+                    422.0,
+                    1406.0,
+                    458.0,
+                    296.0,
+                    458.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    453.0,
+                    762.0,
+                    453.0,
+                    762.0,
+                    488.0,
+                    323.0,
+                    488.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1179.0,
+                    1404.0,
+                    1179.0,
+                    1404.0,
+                    1212.0,
+                    295.0,
+                    1212.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    320.0,
+                    1204.0,
+                    1405.0,
+                    1204.0,
+                    1405.0,
+                    1248.0,
+                    320.0,
+                    1248.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    1237.0,
+                    1405.0,
+                    1237.0,
+                    1405.0,
+                    1276.0,
+                    323.0,
+                    1276.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    1268.0,
+                    1404.0,
+                    1268.0,
+                    1404.0,
+                    1305.0,
+                    322.0,
+                    1305.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    324.0,
+                    1300.0,
+                    1186.0,
+                    1300.0,
+                    1186.0,
+                    1337.0,
+                    324.0,
+                    1337.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1348.0,
+                    1405.0,
+                    1348.0,
+                    1405.0,
+                    1389.0,
+                    293.0,
+                    1389.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    1379.0,
+                    1403.0,
+                    1379.0,
+                    1403.0,
+                    1419.0,
+                    322.0,
+                    1419.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    324.0,
+                    1413.0,
+                    1070.0,
+                    1413.0,
+                    1070.0,
+                    1447.0,
+                    324.0,
+                    1447.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1574.0,
+                    1405.0,
+                    1574.0,
+                    1405.0,
+                    1610.0,
+                    294.0,
+                    1610.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    324.0,
+                    1606.0,
+                    1326.0,
+                    1606.0,
+                    1326.0,
+                    1638.0,
+                    324.0,
+                    1638.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    922.0,
+                    1404.0,
+                    922.0,
+                    1404.0,
+                    958.0,
+                    295.0,
+                    958.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    953.0,
+                    1164.0,
+                    953.0,
+                    1164.0,
+                    989.0,
+                    323.0,
+                    989.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1459.0,
+                    1404.0,
+                    1459.0,
+                    1404.0,
+                    1501.0,
+                    293.0,
+                    1501.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    321.0,
+                    1490.0,
+                    1403.0,
+                    1490.0,
+                    1403.0,
+                    1532.0,
+                    321.0,
+                    1532.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    1524.0,
+                    1199.0,
+                    1524.0,
+                    1199.0,
+                    1559.0,
+                    323.0,
+                    1559.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1797.0,
+                    1405.0,
+                    1797.0,
+                    1405.0,
+                    1835.0,
+                    293.0,
+                    1835.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    1831.0,
+                    1404.0,
+                    1831.0,
+                    1404.0,
+                    1864.0,
+                    322.0,
+                    1864.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    321.0,
+                    1857.0,
+                    1404.0,
+                    1857.0,
+                    1404.0,
+                    1898.0,
+                    321.0,
+                    1898.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    1889.0,
+                    920.0,
+                    1889.0,
+                    920.0,
+                    1929.0,
+                    322.0,
+                    1929.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1657.0,
+                    1404.0,
+                    1657.0,
+                    1404.0,
+                    1690.0,
+                    295.0,
+                    1690.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    324.0,
+                    1689.0,
+                    1405.0,
+                    1689.0,
+                    1405.0,
+                    1722.0,
+                    324.0,
+                    1722.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    1719.0,
+                    1405.0,
+                    1719.0,
+                    1405.0,
+                    1752.0,
+                    322.0,
+                    1752.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    1750.0,
+                    1270.0,
+                    1750.0,
+                    1270.0,
+                    1783.0,
+                    322.0,
+                    1783.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    229.0,
+                    1403.0,
+                    229.0,
+                    1403.0,
+                    267.0,
+                    294.0,
+                    267.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    262.0,
+                    1404.0,
+                    262.0,
+                    1404.0,
+                    296.0,
+                    322.0,
+                    296.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    324.0,
+                    291.0,
+                    1242.0,
+                    291.0,
+                    1242.0,
+                    329.0,
+                    324.0,
+                    329.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    340.0,
+                    1403.0,
+                    340.0,
+                    1403.0,
+                    376.0,
+                    296.0,
+                    376.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    373.0,
+                    1196.0,
+                    373.0,
+                    1196.0,
+                    407.0,
+                    322.0,
+                    407.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    840.0,
+                    1404.0,
+                    840.0,
+                    1404.0,
+                    878.0,
+                    294.0,
+                    878.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    872.0,
+                    990.0,
+                    872.0,
+                    990.0,
+                    906.0,
+                    322.0,
+                    906.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 11,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1849,
+                    1404,
+                    1849,
+                    1404,
+                    2035,
+                    297,
+                    2035
+                ],
+                "score": 0.979
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1560,
+                    1406,
+                    1560,
+                    1406,
+                    1746,
+                    298,
+                    1746
+                ],
+                "score": 0.978
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    540,
+                    1786,
+                    1159,
+                    1786,
+                    1159,
+                    1828,
+                    540,
+                    1828
+                ],
+                "score": 0.923
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    829,
+                    1403,
+                    829,
+                    1403,
+                    955,
+                    298,
+                    955
+                ],
+                "score": 0.869
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    836,
+                    2088,
+                    864,
+                    2088,
+                    864,
+                    2112,
+                    836,
+                    2112
+                ],
+                "score": 0.859
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    298,
+                    75,
+                    858,
+                    75,
+                    858,
+                    105,
+                    298,
+                    105
+                ],
+                "score": 0.852
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    294,
+                    974,
+                    1402,
+                    974,
+                    1402,
+                    1038,
+                    294,
+                    1038
+                ],
+                "score": 0.848
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1058,
+                    1405,
+                    1058,
+                    1405,
+                    1214,
+                    297,
+                    1214
+                ],
+                "score": 0.839
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    684,
+                    1404,
+                    684,
+                    1404,
+                    808,
+                    297,
+                    808
+                ],
+                "score": 0.839
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    295,
+                    601,
+                    1402,
+                    601,
+                    1402,
+                    665,
+                    295,
+                    665
+                ],
+                "score": 0.838
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    300,
+                    1233,
+                    1236,
+                    1233,
+                    1236,
+                    1266,
+                    300,
+                    1266
+                ],
+                "score": 0.834
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    1369,
+                    1404,
+                    1369,
+                    1404,
+                    1434,
+                    299,
+                    1434
+                ],
+                "score": 0.808
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    302,
+                    1286,
+                    1396,
+                    1286,
+                    1396,
+                    1349,
+                    302,
+                    1349
+                ],
+                "score": 0.789
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    296,
+                    488,
+                    1407,
+                    488,
+                    1407,
+                    581,
+                    296,
+                    581
+                ],
+                "score": 0.786
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    229,
+                    1404,
+                    229,
+                    1404,
+                    353,
+                    298,
+                    353
+                ],
+                "score": 0.776
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    299,
+                    1491,
+                    1179,
+                    1491,
+                    1179,
+                    1529,
+                    299,
+                    1529
+                ],
+                "score": 0.767
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    293,
+                    374,
+                    1400,
+                    374,
+                    1400,
+                    467,
+                    293,
+                    467
+                ],
+                "score": 0.705
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    1491,
+                    1179,
+                    1491,
+                    1179,
+                    1529,
+                    299,
+                    1529
+                ],
+                "score": 0.146
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    485,
+                    1654,
+                    654,
+                    1654,
+                    654,
+                    1686,
+                    485,
+                    1686
+                ],
+                "score": 0.92,
+                "latex": "N * ( 1 - S ) * d"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    766,
+                    1912,
+                    904,
+                    1912,
+                    904,
+                    1942,
+                    766,
+                    1942
+                ],
+                "score": 0.92,
+                "latex": "\\Delta \\theta ~ = ~ - \\theta _ { i }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    297,
+                    1941,
+                    542,
+                    1941,
+                    542,
+                    1976,
+                    297,
+                    1976
+                ],
+                "score": 0.91,
+                "latex": "- \\nabla _ { \\boldsymbol { \\theta } _ { i } } \\mathbf { \\bar { L } } ( \\boldsymbol { \\theta } ) \\boldsymbol { \\theta } _ { i } + R ( \\boldsymbol { \\theta } _ { i } ^ { 2 } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1308,
+                    1912,
+                    1403,
+                    1912,
+                    1403,
+                    1943,
+                    1308,
+                    1943
+                ],
+                "score": 0.9,
+                "latex": "\\begin{array} { r l } { \\Delta L _ { i } } & { { } = } \\end{array}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    299,
+                    1656,
+                    338,
+                    1656,
+                    338,
+                    1687,
+                    299,
+                    1687
+                ],
+                "score": 0.89,
+                "latex": "| \\theta _ { i } |"
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    541,
+                    1787,
+                    1159,
+                    1787,
+                    1159,
+                    1828,
+                    541,
+                    1828
+                ],
+                "score": 0.88,
+                "latex": "\\Delta L = L ( \\theta + \\Delta \\theta ) - L ( \\theta ) = \\nabla _ { \\theta } L ( \\theta ) \\Delta \\theta + R ( | | \\Delta \\theta | | _ { 2 } ^ { 2 } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1199,
+                    1883,
+                    1225,
+                    1883,
+                    1225,
+                    1912,
+                    1199,
+                    1912
+                ],
+                "score": 0.86,
+                "latex": "\\theta _ { i }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1173,
+                    1685,
+                    1195,
+                    1685,
+                    1195,
+                    1711,
+                    1173,
+                    1711
+                ],
+                "score": 0.83,
+                "latex": "L"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    494,
+                    1717,
+                    510,
+                    1717,
+                    510,
+                    1741,
+                    494,
+                    1741
+                ],
+                "score": 0.81,
+                "latex": "\\theta"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1215,
+                    1564,
+                    1234,
+                    1564,
+                    1234,
+                    1590,
+                    1215,
+                    1590
+                ],
+                "score": 0.78,
+                "latex": "d"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1233,
+                    1625,
+                    1245,
+                    1625,
+                    1245,
+                    1650,
+                    1233,
+                    1650
+                ],
+                "score": 0.77,
+                "latex": "i"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1114,
+                    926,
+                    1132,
+                    926,
+                    1132,
+                    948,
+                    1114,
+                    948
+                ],
+                "score": 0.27,
+                "latex": "\\underline { { \\underline { { \\mathbf { \\Pi } } } } } ="
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    832.0,
+                    2085.0,
+                    869.0,
+                    2085.0,
+                    869.0,
+                    2123.0,
+                    832.0,
+                    2123.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    298.0,
+                    73.0,
+                    858.0,
+                    73.0,
+                    858.0,
+                    108.0,
+                    298.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1491.0,
+                    1183.0,
+                    1491.0,
+                    1183.0,
+                    1532.0,
+                    295.0,
+                    1532.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1849.0,
+                    1405.0,
+                    1849.0,
+                    1405.0,
+                    1886.0,
+                    294.0,
+                    1886.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1881.0,
+                    1198.0,
+                    1881.0,
+                    1198.0,
+                    1915.0,
+                    294.0,
+                    1915.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1226.0,
+                    1881.0,
+                    1405.0,
+                    1881.0,
+                    1405.0,
+                    1915.0,
+                    1226.0,
+                    1915.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1908.0,
+                    765.0,
+                    1908.0,
+                    765.0,
+                    1948.0,
+                    292.0,
+                    1948.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    905.0,
+                    1908.0,
+                    1307.0,
+                    1908.0,
+                    1307.0,
+                    1948.0,
+                    905.0,
+                    1948.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    543.0,
+                    1941.0,
+                    1406.0,
+                    1941.0,
+                    1406.0,
+                    1976.0,
+                    543.0,
+                    1976.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1970.0,
+                    1404.0,
+                    1970.0,
+                    1404.0,
+                    2007.0,
+                    294.0,
+                    2007.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    2001.0,
+                    1404.0,
+                    2001.0,
+                    1404.0,
+                    2038.0,
+                    294.0,
+                    2038.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1557.0,
+                    1214.0,
+                    1557.0,
+                    1214.0,
+                    1601.0,
+                    293.0,
+                    1601.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1235.0,
+                    1557.0,
+                    1407.0,
+                    1557.0,
+                    1407.0,
+                    1601.0,
+                    1235.0,
+                    1601.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1589.0,
+                    1407.0,
+                    1589.0,
+                    1407.0,
+                    1631.0,
+                    291.0,
+                    1631.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1622.0,
+                    1232.0,
+                    1622.0,
+                    1232.0,
+                    1658.0,
+                    295.0,
+                    1658.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1246.0,
+                    1622.0,
+                    1406.0,
+                    1622.0,
+                    1406.0,
+                    1658.0,
+                    1246.0,
+                    1658.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1652.0,
+                    298.0,
+                    1652.0,
+                    298.0,
+                    1691.0,
+                    295.0,
+                    1691.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    339.0,
+                    1652.0,
+                    484.0,
+                    1652.0,
+                    484.0,
+                    1691.0,
+                    339.0,
+                    1691.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    655.0,
+                    1652.0,
+                    1406.0,
+                    1652.0,
+                    1406.0,
+                    1691.0,
+                    655.0,
+                    1691.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1684.0,
+                    1172.0,
+                    1684.0,
+                    1172.0,
+                    1719.0,
+                    295.0,
+                    1719.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1196.0,
+                    1684.0,
+                    1406.0,
+                    1684.0,
+                    1406.0,
+                    1719.0,
+                    1196.0,
+                    1719.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1714.0,
+                    493.0,
+                    1714.0,
+                    493.0,
+                    1749.0,
+                    293.0,
+                    1749.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    511.0,
+                    1714.0,
+                    522.0,
+                    1714.0,
+                    522.0,
+                    1749.0,
+                    511.0,
+                    1749.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    832.0,
+                    1404.0,
+                    832.0,
+                    1404.0,
+                    865.0,
+                    296.0,
+                    865.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    863.0,
+                    1405.0,
+                    863.0,
+                    1405.0,
+                    896.0,
+                    323.0,
+                    896.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    893.0,
+                    1405.0,
+                    893.0,
+                    1405.0,
+                    926.0,
+                    322.0,
+                    926.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    921.0,
+                    1113.0,
+                    921.0,
+                    1113.0,
+                    957.0,
+                    323.0,
+                    957.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1133.0,
+                    921.0,
+                    1397.0,
+                    921.0,
+                    1397.0,
+                    957.0,
+                    1133.0,
+                    957.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    970.0,
+                    1405.0,
+                    970.0,
+                    1405.0,
+                    1012.0,
+                    293.0,
+                    1012.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    320.0,
+                    1008.0,
+                    470.0,
+                    1008.0,
+                    470.0,
+                    1039.0,
+                    320.0,
+                    1039.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1059.0,
+                    1406.0,
+                    1059.0,
+                    1406.0,
+                    1096.0,
+                    294.0,
+                    1096.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    321.0,
+                    1089.0,
+                    1406.0,
+                    1089.0,
+                    1406.0,
+                    1125.0,
+                    321.0,
+                    1125.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    1120.0,
+                    1405.0,
+                    1120.0,
+                    1405.0,
+                    1157.0,
+                    323.0,
+                    1157.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    325.0,
+                    1151.0,
+                    1405.0,
+                    1151.0,
+                    1405.0,
+                    1184.0,
+                    325.0,
+                    1184.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    1183.0,
+                    1037.0,
+                    1183.0,
+                    1037.0,
+                    1216.0,
+                    323.0,
+                    1216.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    685.0,
+                    1405.0,
+                    685.0,
+                    1405.0,
+                    721.0,
+                    296.0,
+                    721.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    321.0,
+                    717.0,
+                    1404.0,
+                    717.0,
+                    1404.0,
+                    752.0,
+                    321.0,
+                    752.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    747.0,
+                    1402.0,
+                    747.0,
+                    1402.0,
+                    783.0,
+                    322.0,
+                    783.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    324.0,
+                    779.0,
+                    424.0,
+                    779.0,
+                    424.0,
+                    810.0,
+                    324.0,
+                    810.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    600.0,
+                    1407.0,
+                    600.0,
+                    1407.0,
+                    639.0,
+                    293.0,
+                    639.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    632.0,
+                    586.0,
+                    632.0,
+                    586.0,
+                    668.0,
+                    323.0,
+                    668.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1230.0,
+                    1242.0,
+                    1230.0,
+                    1242.0,
+                    1271.0,
+                    292.0,
+                    1271.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1371.0,
+                    1404.0,
+                    1371.0,
+                    1404.0,
+                    1407.0,
+                    295.0,
+                    1407.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    324.0,
+                    1399.0,
+                    1401.0,
+                    1399.0,
+                    1401.0,
+                    1435.0,
+                    324.0,
+                    1435.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1282.0,
+                    1403.0,
+                    1282.0,
+                    1403.0,
+                    1325.0,
+                    294.0,
+                    1325.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    1318.0,
+                    776.0,
+                    1318.0,
+                    776.0,
+                    1349.0,
+                    323.0,
+                    1349.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    487.0,
+                    1407.0,
+                    487.0,
+                    1407.0,
+                    525.0,
+                    294.0,
+                    525.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    324.0,
+                    518.0,
+                    1404.0,
+                    518.0,
+                    1404.0,
+                    554.0,
+                    324.0,
+                    554.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    323.0,
+                    549.0,
+                    806.0,
+                    549.0,
+                    806.0,
+                    583.0,
+                    323.0,
+                    583.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    227.0,
+                    1405.0,
+                    227.0,
+                    1405.0,
+                    267.0,
+                    295.0,
+                    267.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    262.0,
+                    1404.0,
+                    262.0,
+                    1404.0,
+                    295.0,
+                    322.0,
+                    295.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    320.0,
+                    290.0,
+                    1405.0,
+                    290.0,
+                    1405.0,
+                    328.0,
+                    320.0,
+                    328.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    321.0,
+                    320.0,
+                    484.0,
+                    320.0,
+                    484.0,
+                    355.0,
+                    321.0,
+                    355.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    375.0,
+                    1403.0,
+                    375.0,
+                    1403.0,
+                    408.0,
+                    295.0,
+                    408.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    321.0,
+                    404.0,
+                    1405.0,
+                    404.0,
+                    1405.0,
+                    441.0,
+                    321.0,
+                    441.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    436.0,
+                    1010.0,
+                    436.0,
+                    1010.0,
+                    468.0,
+                    322.0,
+                    468.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1491.0,
+                    1183.0,
+                    1491.0,
+                    1183.0,
+                    1532.0,
+                    295.0,
+                    1532.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 12,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    429,
+                    1404,
+                    429,
+                    1404,
+                    675,
+                    298,
+                    675
+                ],
+                "score": 0.98
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    973,
+                    1404,
+                    973,
+                    1404,
+                    1127,
+                    298,
+                    1127
+                ],
+                "score": 0.977
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    229,
+                    1402,
+                    229,
+                    1402,
+                    414,
+                    299,
+                    414
+                ],
+                "score": 0.976
+            },
+            {
+                "category_id": 3,
+                "poly": [
+                    313,
+                    1511,
+                    1375,
+                    1511,
+                    1375,
+                    1913,
+                    313,
+                    1913
+                ],
+                "score": 0.969
+            },
+            {
+                "category_id": 4,
+                "poly": [
+                    297,
+                    1941,
+                    1404,
+                    1941,
+                    1404,
+                    2066,
+                    297,
+                    2066
+                ],
+                "score": 0.953
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    793,
+                    1404,
+                    793,
+                    1404,
+                    855,
+                    297,
+                    855
+                ],
+                "score": 0.945
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1246,
+                    1405,
+                    1246,
+                    1405,
+                    1461,
+                    297,
+                    1461
+                ],
+                "score": 0.928
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    297,
+                    721,
+                    1189,
+                    721,
+                    1189,
+                    757,
+                    297,
+                    757
+                ],
+                "score": 0.919
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    298,
+                    1175,
+                    844,
+                    1175,
+                    844,
+                    1211,
+                    298,
+                    1211
+                ],
+                "score": 0.899
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    298,
+                    76,
+                    857,
+                    76,
+                    857,
+                    104,
+                    298,
+                    104
+                ],
+                "score": 0.873
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    297,
+                    901,
+                    1056,
+                    901,
+                    1056,
+                    938,
+                    297,
+                    938
+                ],
+                "score": 0.856
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    836,
+                    2088,
+                    864,
+                    2088,
+                    864,
+                    2112,
+                    836,
+                    2112
+                ],
+                "score": 0.854
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1300,
+                    1841,
+                    1330,
+                    1841,
+                    1330,
+                    1865,
+                    1300,
+                    1865
+                ],
+                "score": 0.66,
+                "latex": "\\phantom { - } 0 . 9 9"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1201,
+                    1841,
+                    1225,
+                    1841,
+                    1225,
+                    1861,
+                    1201,
+                    1861
+                ],
+                "score": 0.62,
+                "latex": "0 . 9"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    997,
+                    1841,
+                    1021,
+                    1841,
+                    1021,
+                    1860,
+                    997,
+                    1860
+                ],
+                "score": 0.45,
+                "latex": "{ \\boldsymbol \\updownarrow } ^ { \\mathrm { 0 } }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1099,
+                    1839,
+                    1123,
+                    1839,
+                    1123,
+                    1861,
+                    1099,
+                    1861
+                ],
+                "score": 0.37,
+                "latex": "0 . 5"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    452,
+                    553,
+                    511,
+                    553,
+                    511,
+                    583,
+                    452,
+                    583
+                ],
+                "score": 0.36,
+                "latex": "R i g L"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    347.0,
+                    1531.0,
+                    400.0,
+                    1531.0,
+                    400.0,
+                    1565.0,
+                    347.0,
+                    1565.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    411.0,
+                    1507.0,
+                    810.0,
+                    1507.0,
+                    810.0,
+                    1548.0,
+                    411.0,
+                    1548.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    909.0,
+                    1519.0,
+                    1364.0,
+                    1519.0,
+                    1364.0,
+                    1571.0,
+                    909.0,
+                    1571.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    348.0,
+                    1568.0,
+                    394.0,
+                    1568.0,
+                    394.0,
+                    1600.0,
+                    348.0,
+                    1600.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    632.0,
+                    1559.0,
+                    820.0,
+                    1559.0,
+                    820.0,
+                    1594.0,
+                    632.0,
+                    1594.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    315.0,
+                    1584.0,
+                    361.0,
+                    1584.0,
+                    361.0,
+                    1805.0,
+                    315.0,
+                    1805.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    632.0,
+                    1585.0,
+                    756.0,
+                    1585.0,
+                    756.0,
+                    1626.0,
+                    632.0,
+                    1626.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    349.0,
+                    1605.0,
+                    395.0,
+                    1605.0,
+                    395.0,
+                    1635.0,
+                    349.0,
+                    1635.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    632.0,
+                    1619.0,
+                    723.0,
+                    1619.0,
+                    723.0,
+                    1654.0,
+                    632.0,
+                    1654.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    350.0,
+                    1643.0,
+                    393.0,
+                    1643.0,
+                    393.0,
+                    1670.0,
+                    350.0,
+                    1670.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    641.0,
+                    1652.0,
+                    653.0,
+                    1652.0,
+                    653.0,
+                    1663.0,
+                    641.0,
+                    1663.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    909.0,
+                    1653.0,
+                    963.0,
+                    1653.0,
+                    963.0,
+                    1682.0,
+                    909.0,
+                    1682.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    349.0,
+                    1679.0,
+                    394.0,
+                    1679.0,
+                    394.0,
+                    1706.0,
+                    349.0,
+                    1706.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    350.0,
+                    1716.0,
+                    394.0,
+                    1716.0,
+                    394.0,
+                    1742.0,
+                    350.0,
+                    1742.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    909.0,
+                    1708.0,
+                    966.0,
+                    1708.0,
+                    966.0,
+                    1737.0,
+                    909.0,
+                    1737.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    349.0,
+                    1752.0,
+                    393.0,
+                    1752.0,
+                    393.0,
+                    1777.0,
+                    349.0,
+                    1777.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    909.0,
+                    1764.0,
+                    964.0,
+                    1764.0,
+                    964.0,
+                    1793.0,
+                    909.0,
+                    1793.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    348.0,
+                    1784.0,
+                    393.0,
+                    1784.0,
+                    393.0,
+                    1816.0,
+                    348.0,
+                    1816.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    348.0,
+                    1819.0,
+                    396.0,
+                    1819.0,
+                    396.0,
+                    1851.0,
+                    348.0,
+                    1851.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    909.0,
+                    1819.0,
+                    965.0,
+                    1819.0,
+                    965.0,
+                    1848.0,
+                    909.0,
+                    1848.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    535.0,
+                    1836.0,
+                    573.0,
+                    1836.0,
+                    573.0,
+                    1872.0,
+                    535.0,
+                    1872.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1088.0,
+                    1837.0,
+                    1098.0,
+                    1837.0,
+                    1098.0,
+                    1863.0,
+                    1088.0,
+                    1863.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1124.0,
+                    1837.0,
+                    1127.0,
+                    1837.0,
+                    1127.0,
+                    1863.0,
+                    1124.0,
+                    1863.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1195.0,
+                    1833.0,
+                    1200.0,
+                    1833.0,
+                    1200.0,
+                    1866.0,
+                    1195.0,
+                    1866.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1226.0,
+                    1833.0,
+                    1231.0,
+                    1833.0,
+                    1231.0,
+                    1866.0,
+                    1226.0,
+                    1866.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    545.0,
+                    1869.0,
+                    678.0,
+                    1869.0,
+                    678.0,
+                    1922.0,
+                    545.0,
+                    1922.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1099.0,
+                    1860.0,
+                    1229.0,
+                    1860.0,
+                    1229.0,
+                    1912.0,
+                    1099.0,
+                    1912.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    987.0,
+                    1833.5,
+                    1030.0,
+                    1833.5,
+                    1030.0,
+                    1862.5,
+                    987.0,
+                    1862.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    418.0,
+                    1840.5,
+                    471.0,
+                    1840.5,
+                    471.0,
+                    1872.0,
+                    418.0,
+                    1872.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    630.75,
+                    1839.0,
+                    697.75,
+                    1839.0,
+                    697.75,
+                    1873.0,
+                    630.75,
+                    1873.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    742.0,
+                    1842.5,
+                    808.0,
+                    1842.5,
+                    808.0,
+                    1872.0,
+                    742.0,
+                    1872.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1288.0,
+                    1838.5,
+                    1342.0,
+                    1838.5,
+                    1342.0,
+                    1864.5,
+                    1288.0,
+                    1864.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1940.0,
+                    1404.0,
+                    1940.0,
+                    1404.0,
+                    1979.0,
+                    292.0,
+                    1979.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1972.0,
+                    1405.0,
+                    1972.0,
+                    1405.0,
+                    2008.0,
+                    294.0,
+                    2008.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    2005.0,
+                    1404.0,
+                    2005.0,
+                    1404.0,
+                    2036.0,
+                    293.0,
+                    2036.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    2034.0,
+                    1313.0,
+                    2034.0,
+                    1313.0,
+                    2070.0,
+                    295.0,
+                    2070.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    720.0,
+                    1194.0,
+                    720.0,
+                    1194.0,
+                    760.0,
+                    293.0,
+                    760.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1174.0,
+                    847.0,
+                    1174.0,
+                    847.0,
+                    1214.0,
+                    294.0,
+                    1214.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    298.0,
+                    73.0,
+                    858.0,
+                    73.0,
+                    858.0,
+                    108.0,
+                    298.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    899.0,
+                    1058.0,
+                    899.0,
+                    1058.0,
+                    943.0,
+                    297.0,
+                    943.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    832.0,
+                    2085.0,
+                    870.0,
+                    2085.0,
+                    870.0,
+                    2123.0,
+                    832.0,
+                    2123.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    428.0,
+                    1406.0,
+                    428.0,
+                    1406.0,
+                    463.0,
+                    293.0,
+                    463.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    462.0,
+                    1406.0,
+                    462.0,
+                    1406.0,
+                    496.0,
+                    294.0,
+                    496.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    491.0,
+                    1406.0,
+                    491.0,
+                    1406.0,
+                    524.0,
+                    294.0,
+                    524.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    522.0,
+                    1404.0,
+                    522.0,
+                    1404.0,
+                    556.0,
+                    294.0,
+                    556.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    550.0,
+                    451.0,
+                    550.0,
+                    451.0,
+                    588.0,
+                    292.0,
+                    588.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    512.0,
+                    550.0,
+                    1408.0,
+                    550.0,
+                    1408.0,
+                    588.0,
+                    512.0,
+                    588.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    582.0,
+                    1405.0,
+                    582.0,
+                    1405.0,
+                    619.0,
+                    293.0,
+                    619.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    612.0,
+                    1406.0,
+                    612.0,
+                    1406.0,
+                    649.0,
+                    293.0,
+                    649.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    643.0,
+                    626.0,
+                    643.0,
+                    626.0,
+                    678.0,
+                    294.0,
+                    678.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    975.0,
+                    1405.0,
+                    975.0,
+                    1405.0,
+                    1008.0,
+                    296.0,
+                    1008.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1001.0,
+                    1405.0,
+                    1001.0,
+                    1405.0,
+                    1041.0,
+                    292.0,
+                    1041.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1034.0,
+                    1406.0,
+                    1034.0,
+                    1406.0,
+                    1070.0,
+                    293.0,
+                    1070.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1065.0,
+                    1405.0,
+                    1065.0,
+                    1405.0,
+                    1102.0,
+                    293.0,
+                    1102.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1098.0,
+                    1243.0,
+                    1098.0,
+                    1243.0,
+                    1131.0,
+                    296.0,
+                    1131.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    231.0,
+                    1403.0,
+                    231.0,
+                    1403.0,
+                    263.0,
+                    297.0,
+                    263.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    263.0,
+                    1404.0,
+                    263.0,
+                    1404.0,
+                    295.0,
+                    297.0,
+                    295.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    291.0,
+                    1406.0,
+                    291.0,
+                    1406.0,
+                    327.0,
+                    295.0,
+                    327.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    323.0,
+                    1404.0,
+                    323.0,
+                    1404.0,
+                    355.0,
+                    297.0,
+                    355.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    352.0,
+                    1406.0,
+                    352.0,
+                    1406.0,
+                    388.0,
+                    295.0,
+                    388.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    383.0,
+                    1135.0,
+                    383.0,
+                    1135.0,
+                    418.0,
+                    294.0,
+                    418.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    789.0,
+                    1406.0,
+                    789.0,
+                    1406.0,
+                    832.0,
+                    291.0,
+                    832.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    823.0,
+                    1328.0,
+                    823.0,
+                    1328.0,
+                    858.0,
+                    295.0,
+                    858.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1243.0,
+                    1406.0,
+                    1243.0,
+                    1406.0,
+                    1282.0,
+                    294.0,
+                    1282.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1277.0,
+                    1403.0,
+                    1277.0,
+                    1403.0,
+                    1312.0,
+                    295.0,
+                    1312.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1308.0,
+                    1405.0,
+                    1308.0,
+                    1405.0,
+                    1342.0,
+                    295.0,
+                    1342.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1338.0,
+                    1405.0,
+                    1338.0,
+                    1405.0,
+                    1372.0,
+                    294.0,
+                    1372.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1367.0,
+                    1406.0,
+                    1367.0,
+                    1406.0,
+                    1405.0,
+                    292.0,
+                    1405.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1399.0,
+                    1405.0,
+                    1399.0,
+                    1405.0,
+                    1433.0,
+                    295.0,
+                    1433.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1428.0,
+                    568.0,
+                    1428.0,
+                    568.0,
+                    1463.0,
+                    295.0,
+                    1463.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 13,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 5,
+                "poly": [
+                    445,
+                    222,
+                    1256,
+                    222,
+                    1256,
+                    385,
+                    445,
+                    385
+                ],
+                "score": 0.978,
+                "html": "<table><tr><td>Initialization</td><td>Training Method</td><td>Test Accuracy</td><td>Training FLOPs</td></tr><tr><td>Lottery</td><td>Static</td><td>70.82±0.07</td><td>0.46x</td></tr><tr><td>Lottery</td><td>RigL</td><td>73.93±0.09</td><td>0.46x</td></tr><tr><td>Random</td><td>RigL</td><td>74.55±0.06</td><td>0.23x</td></tr><tr><td>Random</td><td>RigL2x</td><td>76.06±0.09</td><td>0.46x</td></tr></table>"
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    591,
+                    1404,
+                    591,
+                    1404,
+                    716,
+                    298,
+                    716
+                ],
+                "score": 0.974
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1051,
+                    1405,
+                    1051,
+                    1405,
+                    1204,
+                    298,
+                    1204
+                ],
+                "score": 0.973
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1426,
+                    1404,
+                    1426,
+                    1404,
+                    1521,
+                    298,
+                    1521
+                ],
+                "score": 0.972
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    1318,
+                    1404,
+                    1318,
+                    1404,
+                    1412,
+                    299,
+                    1412
+                ],
+                "score": 0.972
+            },
+            {
+                "category_id": 3,
+                "poly": [
+                    313,
+                    1573,
+                    1370,
+                    1573,
+                    1370,
+                    1972,
+                    313,
+                    1972
+                ],
+                "score": 0.966
+            },
+            {
+                "category_id": 4,
+                "poly": [
+                    292,
+                    2003,
+                    1404,
+                    2003,
+                    1404,
+                    2066,
+                    292,
+                    2066
+                ],
+                "score": 0.929
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    294,
+                    829,
+                    1156,
+                    829,
+                    1156,
+                    863,
+                    294,
+                    863
+                ],
+                "score": 0.917
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    299,
+                    1250,
+                    1073,
+                    1250,
+                    1073,
+                    1286,
+                    299,
+                    1286
+                ],
+                "score": 0.911
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    297,
+                    74,
+                    858,
+                    74,
+                    858,
+                    105,
+                    297,
+                    105
+                ],
+                "score": 0.911
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    297,
+                    760,
+                    860,
+                    760,
+                    860,
+                    796,
+                    297,
+                    796
+                ],
+                "score": 0.911
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    362,
+                    886,
+                    1403,
+                    886,
+                    1403,
+                    1026,
+                    362,
+                    1026
+                ],
+                "score": 0.876
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    292,
+                    522,
+                    1392,
+                    522,
+                    1392,
+                    559,
+                    292,
+                    559
+                ],
+                "score": 0.875
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    836,
+                    2088,
+                    864,
+                    2088,
+                    864,
+                    2113,
+                    836,
+                    2113
+                ],
+                "score": 0.864
+            },
+            {
+                "category_id": 6,
+                "poly": [
+                    293,
+                    405,
+                    1397,
+                    405,
+                    1397,
+                    469,
+                    293,
+                    469
+                ],
+                "score": 0.738
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    892,
+                    958,
+                    1178,
+                    958,
+                    1178,
+                    997,
+                    892,
+                    997
+                ],
+                "score": 0.92,
+                "latex": "\\begin{array} { r } { f _ { d e c a y } ( t ) = \\alpha ( 1 - \\frac { t } { T _ { e n d } } ) ^ { k } } \\end{array}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    515,
+                    887,
+                    675,
+                    887,
+                    675,
+                    923,
+                    515,
+                    923
+                ],
+                "score": 0.92,
+                "latex": "f _ { d e c a y } ( t ) = \\alpha"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    629,
+                    624,
+                    748,
+                    624,
+                    748,
+                    652,
+                    629,
+                    652
+                ],
+                "score": 0.91,
+                "latex": "\\Delta T = 5 0"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    494,
+                    995,
+                    563,
+                    995,
+                    563,
+                    1022,
+                    494,
+                    1022
+                ],
+                "score": 0.9,
+                "latex": "k = 1"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    801,
+                    625,
+                    903,
+                    625,
+                    903,
+                    653,
+                    801,
+                    653
+                ],
+                "score": 0.9,
+                "latex": "\\alpha = 0 . 1"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    620,
+                    685,
+                    760,
+                    685,
+                    760,
+                    713,
+                    620,
+                    713
+                ],
+                "score": 0.9,
+                "latex": "\\Delta T = 1 0 0 0"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1036,
+                    685,
+                    1130,
+                    685,
+                    1130,
+                    713,
+                    1036,
+                    713
+                ],
+                "score": 0.89,
+                "latex": "\\alpha = 0 . 5"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1045,
+                    1054,
+                    1138,
+                    1054,
+                    1138,
+                    1082,
+                    1045,
+                    1082
+                ],
+                "score": 0.89,
+                "latex": "\\alpha = 0 . 1"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1037,
+                    997,
+                    1106,
+                    997,
+                    1106,
+                    1022,
+                    1037,
+                    1022
+                ],
+                "score": 0.87,
+                "latex": "k = 3"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1178,
+                    1349,
+                    1202,
+                    1349,
+                    1202,
+                    1378,
+                    1178,
+                    1378
+                ],
+                "score": 0.86,
+                "latex": "s ^ { l }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    870,
+                    1084,
+                    917,
+                    1084,
+                    917,
+                    1112,
+                    870,
+                    1112
+                ],
+                "score": 0.8,
+                "latex": "{ \\mathrm { k } } { = } 1"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    775,
+                    1084,
+                    823,
+                    1084,
+                    823,
+                    1111,
+                    775,
+                    1111
+                ],
+                "score": 0.79,
+                "latex": "{ \\bf k } = 3"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    298,
+                    685,
+                    344,
+                    685,
+                    344,
+                    713,
+                    298,
+                    713
+                ],
+                "score": 0.75,
+                "latex": "\\Delta T"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    930,
+                    1115,
+                    978,
+                    1115,
+                    978,
+                    1142,
+                    930,
+                    1142
+                ],
+                "score": 0.75,
+                "latex": "{ \\bf k } = 3"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    886,
+                    656,
+                    924,
+                    656,
+                    924,
+                    684,
+                    886,
+                    684
+                ],
+                "score": 0.74,
+                "latex": "( \\alpha )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    909,
+                    1146,
+                    960,
+                    1146,
+                    960,
+                    1173,
+                    909,
+                    1173
+                ],
+                "score": 0.73,
+                "latex": "( \\mathrm { k } { = } 1"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    298,
+                    1119,
+                    319,
+                    1119,
+                    319,
+                    1141,
+                    298,
+                    1141
+                ],
+                "score": 0.72,
+                "latex": "\\alpha"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    543,
+                    1088,
+                    564,
+                    1088,
+                    564,
+                    1110,
+                    543,
+                    1110
+                ],
+                "score": 0.62,
+                "latex": "\\alpha"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    533.0,
+                    1571.0,
+                    686.0,
+                    1571.0,
+                    686.0,
+                    1609.0,
+                    533.0,
+                    1609.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1074.0,
+                    1581.0,
+                    1242.0,
+                    1581.0,
+                    1242.0,
+                    1619.0,
+                    1074.0,
+                    1619.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    346.0,
+                    1626.0,
+                    405.0,
+                    1626.0,
+                    405.0,
+                    1660.0,
+                    346.0,
+                    1660.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    909.0,
+                    1635.0,
+                    963.0,
+                    1635.0,
+                    963.0,
+                    1665.0,
+                    909.0,
+                    1665.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    315.0,
+                    1654.0,
+                    362.0,
+                    1654.0,
+                    362.0,
+                    1875.0,
+                    315.0,
+                    1875.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    878.0,
+                    1659.0,
+                    924.0,
+                    1659.0,
+                    924.0,
+                    1869.0,
+                    878.0,
+                    1869.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    347.0,
+                    1683.0,
+                    405.0,
+                    1683.0,
+                    405.0,
+                    1713.0,
+                    347.0,
+                    1713.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    910.0,
+                    1687.0,
+                    964.0,
+                    1687.0,
+                    964.0,
+                    1716.0,
+                    910.0,
+                    1716.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    349.0,
+                    1738.0,
+                    404.0,
+                    1738.0,
+                    404.0,
+                    1767.0,
+                    349.0,
+                    1767.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    911.0,
+                    1740.0,
+                    962.0,
+                    1740.0,
+                    962.0,
+                    1765.0,
+                    911.0,
+                    1765.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    444.0,
+                    1777.0,
+                    478.0,
+                    1777.0,
+                    478.0,
+                    1807.0,
+                    444.0,
+                    1807.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    999.0,
+                    1776.0,
+                    1033.0,
+                    1776.0,
+                    1033.0,
+                    1806.0,
+                    999.0,
+                    1806.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    347.0,
+                    1792.0,
+                    404.0,
+                    1792.0,
+                    404.0,
+                    1820.0,
+                    347.0,
+                    1820.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    910.0,
+                    1792.0,
+                    962.0,
+                    1792.0,
+                    962.0,
+                    1817.0,
+                    910.0,
+                    1817.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    465.0,
+                    1808.0,
+                    511.0,
+                    1808.0,
+                    511.0,
+                    1844.0,
+                    465.0,
+                    1844.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1021.0,
+                    1807.0,
+                    1064.0,
+                    1807.0,
+                    1064.0,
+                    1840.0,
+                    1021.0,
+                    1840.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    347.0,
+                    1846.0,
+                    404.0,
+                    1846.0,
+                    404.0,
+                    1876.0,
+                    347.0,
+                    1876.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    449.0,
+                    1836.0,
+                    514.0,
+                    1836.0,
+                    514.0,
+                    1874.0,
+                    449.0,
+                    1874.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    909.0,
+                    1842.0,
+                    963.0,
+                    1842.0,
+                    963.0,
+                    1871.0,
+                    909.0,
+                    1871.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1016.0,
+                    1834.0,
+                    1067.0,
+                    1834.0,
+                    1067.0,
+                    1868.0,
+                    1016.0,
+                    1868.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    464.0,
+                    1868.0,
+                    514.0,
+                    1868.0,
+                    514.0,
+                    1904.0,
+                    464.0,
+                    1904.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1020.0,
+                    1861.0,
+                    1068.0,
+                    1861.0,
+                    1068.0,
+                    1897.0,
+                    1020.0,
+                    1897.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    349.0,
+                    1900.0,
+                    406.0,
+                    1900.0,
+                    406.0,
+                    1928.0,
+                    349.0,
+                    1928.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    490.0,
+                    1913.0,
+                    526.0,
+                    1913.0,
+                    526.0,
+                    1940.0,
+                    490.0,
+                    1940.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    801.0,
+                    1913.0,
+                    837.0,
+                    1913.0,
+                    837.0,
+                    1939.0,
+                    801.0,
+                    1939.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    911.0,
+                    1895.0,
+                    964.0,
+                    1895.0,
+                    964.0,
+                    1920.0,
+                    911.0,
+                    1920.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1046.0,
+                    1908.0,
+                    1078.0,
+                    1908.0,
+                    1078.0,
+                    1930.0,
+                    1046.0,
+                    1930.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    491.0,
+                    1931.0,
+                    726.0,
+                    1931.0,
+                    726.0,
+                    1974.0,
+                    491.0,
+                    1974.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1045.0,
+                    1924.0,
+                    1268.0,
+                    1924.0,
+                    1268.0,
+                    1966.0,
+                    1045.0,
+                    1966.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1346.0,
+                    1911.0,
+                    1370.0,
+                    1911.0,
+                    1370.0,
+                    1924.0,
+                    1346.0,
+                    1924.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    2003.0,
+                    1401.0,
+                    2003.0,
+                    1401.0,
+                    2038.0,
+                    295.0,
+                    2038.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    2033.0,
+                    673.0,
+                    2033.0,
+                    673.0,
+                    2069.0,
+                    295.0,
+                    2069.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1247.0,
+                    1077.0,
+                    1247.0,
+                    1077.0,
+                    1291.0,
+                    294.0,
+                    1291.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    72.0,
+                    858.0,
+                    72.0,
+                    858.0,
+                    108.0,
+                    297.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    756.0,
+                    862.0,
+                    756.0,
+                    862.0,
+                    801.0,
+                    291.0,
+                    801.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    521.0,
+                    1395.0,
+                    521.0,
+                    1395.0,
+                    562.0,
+                    293.0,
+                    562.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    832.0,
+                    2085.0,
+                    870.0,
+                    2085.0,
+                    870.0,
+                    2124.0,
+                    832.0,
+                    2124.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    402.0,
+                    1403.0,
+                    402.0,
+                    1403.0,
+                    441.0,
+                    294.0,
+                    441.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    436.0,
+                    1155.0,
+                    436.0,
+                    1155.0,
+                    472.0,
+                    294.0,
+                    472.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    591.0,
+                    1406.0,
+                    591.0,
+                    1406.0,
+                    628.0,
+                    293.0,
+                    628.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    623.0,
+                    628.0,
+                    623.0,
+                    628.0,
+                    656.0,
+                    296.0,
+                    656.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    749.0,
+                    623.0,
+                    800.0,
+                    623.0,
+                    800.0,
+                    656.0,
+                    749.0,
+                    656.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    904.0,
+                    623.0,
+                    1404.0,
+                    623.0,
+                    1404.0,
+                    656.0,
+                    904.0,
+                    656.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    655.0,
+                    885.0,
+                    655.0,
+                    885.0,
+                    687.0,
+                    294.0,
+                    687.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    925.0,
+                    655.0,
+                    1404.0,
+                    655.0,
+                    1404.0,
+                    687.0,
+                    925.0,
+                    687.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    345.0,
+                    685.0,
+                    619.0,
+                    685.0,
+                    619.0,
+                    718.0,
+                    345.0,
+                    718.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    761.0,
+                    685.0,
+                    1035.0,
+                    685.0,
+                    1035.0,
+                    718.0,
+                    761.0,
+                    718.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1131.0,
+                    685.0,
+                    1141.0,
+                    685.0,
+                    1141.0,
+                    718.0,
+                    1131.0,
+                    718.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1051.0,
+                    1044.0,
+                    1051.0,
+                    1044.0,
+                    1087.0,
+                    296.0,
+                    1087.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1139.0,
+                    1051.0,
+                    1405.0,
+                    1051.0,
+                    1405.0,
+                    1087.0,
+                    1139.0,
+                    1087.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1081.0,
+                    542.0,
+                    1081.0,
+                    542.0,
+                    1118.0,
+                    295.0,
+                    1118.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    565.0,
+                    1081.0,
+                    774.0,
+                    1081.0,
+                    774.0,
+                    1118.0,
+                    565.0,
+                    1118.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    824.0,
+                    1081.0,
+                    869.0,
+                    1081.0,
+                    869.0,
+                    1118.0,
+                    824.0,
+                    1118.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    918.0,
+                    1081.0,
+                    1405.0,
+                    1081.0,
+                    1405.0,
+                    1118.0,
+                    918.0,
+                    1118.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1110.0,
+                    297.0,
+                    1110.0,
+                    297.0,
+                    1147.0,
+                    293.0,
+                    1147.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    320.0,
+                    1110.0,
+                    929.0,
+                    1110.0,
+                    929.0,
+                    1147.0,
+                    320.0,
+                    1147.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    979.0,
+                    1110.0,
+                    1402.0,
+                    1110.0,
+                    1402.0,
+                    1147.0,
+                    979.0,
+                    1147.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1142.0,
+                    908.0,
+                    1142.0,
+                    908.0,
+                    1179.0,
+                    295.0,
+                    1179.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    961.0,
+                    1142.0,
+                    1405.0,
+                    1142.0,
+                    1405.0,
+                    1179.0,
+                    961.0,
+                    1179.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1174.0,
+                    483.0,
+                    1174.0,
+                    483.0,
+                    1207.0,
+                    295.0,
+                    1207.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1426.0,
+                    1404.0,
+                    1426.0,
+                    1404.0,
+                    1464.0,
+                    294.0,
+                    1464.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1458.0,
+                    1404.0,
+                    1458.0,
+                    1404.0,
+                    1493.0,
+                    293.0,
+                    1493.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1488.0,
+                    1404.0,
+                    1488.0,
+                    1404.0,
+                    1525.0,
+                    293.0,
+                    1525.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1319.0,
+                    1404.0,
+                    1319.0,
+                    1404.0,
+                    1353.0,
+                    295.0,
+                    1353.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1350.0,
+                    1177.0,
+                    1350.0,
+                    1177.0,
+                    1384.0,
+                    295.0,
+                    1384.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1203.0,
+                    1350.0,
+                    1404.0,
+                    1350.0,
+                    1404.0,
+                    1384.0,
+                    1203.0,
+                    1384.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1381.0,
+                    1000.0,
+                    1381.0,
+                    1000.0,
+                    1415.0,
+                    294.0,
+                    1415.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    827.0,
+                    1159.0,
+                    827.0,
+                    1159.0,
+                    867.0,
+                    295.0,
+                    867.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    360.0,
+                    883.0,
+                    514.0,
+                    883.0,
+                    514.0,
+                    926.0,
+                    360.0,
+                    926.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    676.0,
+                    883.0,
+                    686.0,
+                    883.0,
+                    686.0,
+                    926.0,
+                    676.0,
+                    926.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    360.0,
+                    927.0,
+                    1404.0,
+                    927.0,
+                    1404.0,
+                    962.0,
+                    360.0,
+                    962.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    392.0,
+                    956.0,
+                    891.0,
+                    956.0,
+                    891.0,
+                    997.0,
+                    392.0,
+                    997.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    396.0,
+                    995.0,
+                    493.0,
+                    995.0,
+                    493.0,
+                    1026.0,
+                    396.0,
+                    1026.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    564.0,
+                    995.0,
+                    1036.0,
+                    995.0,
+                    1036.0,
+                    1026.0,
+                    564.0,
+                    1026.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1107.0,
+                    995.0,
+                    1117.0,
+                    995.0,
+                    1117.0,
+                    1026.0,
+                    1107.0,
+                    1026.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1115.25,
+                    954.5,
+                    1140.25,
+                    954.5,
+                    1140.25,
+                    977.5,
+                    1115.25,
+                    977.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1148.75,
+                    953.5,
+                    1405.75,
+                    953.5,
+                    1405.75,
+                    997.5,
+                    1148.75,
+                    997.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1098.25,
+                    973.5,
+                    1153.25,
+                    973.5,
+                    1153.25,
+                    1003.5,
+                    1098.25,
+                    1003.5
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 14,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1439,
+                    1405,
+                    1439,
+                    1405,
+                    1656,
+                    297,
+                    1656
+                ],
+                "score": 0.975
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    368,
+                    894,
+                    1404,
+                    894,
+                    1404,
+                    1330,
+                    368,
+                    1330
+                ],
+                "score": 0.962
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    301,
+                    744,
+                    1404,
+                    744,
+                    1404,
+                    867,
+                    301,
+                    867
+                ],
+                "score": 0.959
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    300,
+                    229,
+                    1404,
+                    229,
+                    1404,
+                    323,
+                    300,
+                    323
+                ],
+                "score": 0.955
+            },
+            {
+                "category_id": 3,
+                "poly": [
+                    309,
+                    1709,
+                    1381,
+                    1709,
+                    1381,
+                    1976,
+                    309,
+                    1976
+                ],
+                "score": 0.95
+            },
+            {
+                "category_id": 4,
+                "poly": [
+                    293,
+                    2003,
+                    1407,
+                    2003,
+                    1407,
+                    2065,
+                    293,
+                    2065
+                ],
+                "score": 0.925
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    300,
+                    1369,
+                    1156,
+                    1369,
+                    1156,
+                    1405,
+                    300,
+                    1405
+                ],
+                "score": 0.908
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    298,
+                    75,
+                    858,
+                    75,
+                    858,
+                    105,
+                    298,
+                    105
+                ],
+                "score": 0.907
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    301,
+                    338,
+                    866,
+                    338,
+                    866,
+                    370,
+                    301,
+                    370
+                ],
+                "score": 0.889
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    361,
+                    397,
+                    1404,
+                    397,
+                    1404,
+                    718,
+                    361,
+                    718
+                ],
+                "score": 0.852
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    836,
+                    2088,
+                    865,
+                    2088,
+                    865,
+                    2113,
+                    836,
+                    2113
+                ],
+                "score": 0.851
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    613,
+                    1180,
+                    750,
+                    1180,
+                    750,
+                    1212,
+                    613,
+                    1212
+                ],
+                "score": 0.94,
+                "latex": "2 * f _ { S } + f _ { D }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    562,
+                    1282,
+                    787,
+                    1282,
+                    787,
+                    1331,
+                    562,
+                    1331
+                ],
+                "score": 0.94,
+                "latex": "\\frac { ( 3 * f _ { S } * \\Delta T + 2 * f _ { S } + f _ { D } ) } { ( \\Delta T + 1 ) }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    953,
+                    896,
+                    1029,
+                    896,
+                    1029,
+                    928,
+                    953,
+                    928
+                ],
+                "score": 0.92,
+                "latex": "3 * f _ { D }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    830,
+                    896,
+                    902,
+                    896,
+                    902,
+                    928,
+                    830,
+                    928
+                ],
+                "score": 0.91,
+                "latex": "3 * f _ { S }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1123,
+                    1120,
+                    1194,
+                    1120,
+                    1194,
+                    1150,
+                    1123,
+                    1150
+                ],
+                "score": 0.91,
+                "latex": "2 * f _ { S }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    923,
+                    970,
+                    996,
+                    970,
+                    996,
+                    1002,
+                    923,
+                    1002
+                ],
+                "score": 0.9,
+                "latex": "3 * f _ { S }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    396,
+                    1075,
+                    469,
+                    1075,
+                    469,
+                    1107,
+                    396,
+                    1107
+                ],
+                "score": 0.9,
+                "latex": "3 * f _ { S }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    963,
+                    1224,
+                    1038,
+                    1224,
+                    1038,
+                    1255,
+                    963,
+                    1255
+                ],
+                "score": 0.9,
+                "latex": "3 * f _ { S }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    774,
+                    807,
+                    809,
+                    807,
+                    809,
+                    837,
+                    774,
+                    837
+                ],
+                "score": 0.89,
+                "latex": "f _ { D }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1180,
+                    806,
+                    1213,
+                    806,
+                    1213,
+                    838,
+                    1180,
+                    838
+                ],
+                "score": 0.86,
+                "latex": "f _ { S }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1355,
+                    1224,
+                    1402,
+                    1224,
+                    1402,
+                    1251,
+                    1355,
+                    1251
+                ],
+                "score": 0.81,
+                "latex": "\\Delta T"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1291,
+                    2005,
+                    1349,
+                    2005,
+                    1349,
+                    2033,
+                    1291,
+                    2033
+                ],
+                "score": 0.76,
+                "latex": "\\left( \\mathrm { k } \\mathrm { = } 3 \\right)"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    907,
+                    1719,
+                    949,
+                    1719,
+                    949,
+                    1740,
+                    907,
+                    1740
+                ],
+                "score": 0.49,
+                "latex": "( k = 3 )"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    462.0,
+                    1707.0,
+                    551.0,
+                    1707.0,
+                    551.0,
+                    1737.0,
+                    462.0,
+                    1737.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    784.0,
+                    1715.0,
+                    906.0,
+                    1715.0,
+                    906.0,
+                    1743.0,
+                    784.0,
+                    1743.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1206.0,
+                    1711.0,
+                    1270.0,
+                    1711.0,
+                    1270.0,
+                    1745.0,
+                    1206.0,
+                    1745.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    332.0,
+                    1746.0,
+                    371.0,
+                    1746.0,
+                    371.0,
+                    1767.0,
+                    332.0,
+                    1767.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    703.0,
+                    1751.0,
+                    740.0,
+                    1751.0,
+                    740.0,
+                    1772.0,
+                    703.0,
+                    1772.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1073.0,
+                    1751.0,
+                    1110.0,
+                    1751.0,
+                    1110.0,
+                    1772.0,
+                    1073.0,
+                    1772.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    311.0,
+                    1763.0,
+                    346.0,
+                    1763.0,
+                    346.0,
+                    1911.0,
+                    311.0,
+                    1911.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    681.0,
+                    1766.0,
+                    717.0,
+                    1766.0,
+                    717.0,
+                    1908.0,
+                    681.0,
+                    1908.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    742.0,
+                    1770.0,
+                    751.0,
+                    1770.0,
+                    751.0,
+                    1781.0,
+                    742.0,
+                    1781.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1050.0,
+                    1766.0,
+                    1087.0,
+                    1766.0,
+                    1087.0,
+                    1908.0,
+                    1050.0,
+                    1908.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    330.0,
+                    1782.0,
+                    371.0,
+                    1782.0,
+                    371.0,
+                    1801.0,
+                    330.0,
+                    1801.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    703.0,
+                    1785.0,
+                    741.0,
+                    1785.0,
+                    741.0,
+                    1807.0,
+                    703.0,
+                    1807.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1073.0,
+                    1785.0,
+                    1110.0,
+                    1785.0,
+                    1110.0,
+                    1807.0,
+                    1073.0,
+                    1807.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    332.0,
+                    1817.0,
+                    371.0,
+                    1817.0,
+                    371.0,
+                    1840.0,
+                    332.0,
+                    1840.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    705.0,
+                    1818.0,
+                    740.0,
+                    1818.0,
+                    740.0,
+                    1840.0,
+                    705.0,
+                    1840.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1073.0,
+                    1818.0,
+                    1109.0,
+                    1818.0,
+                    1109.0,
+                    1840.0,
+                    1073.0,
+                    1840.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    333.0,
+                    1854.0,
+                    371.0,
+                    1854.0,
+                    371.0,
+                    1876.0,
+                    333.0,
+                    1876.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    395.0,
+                    1844.0,
+                    420.0,
+                    1844.0,
+                    420.0,
+                    1869.0,
+                    395.0,
+                    1869.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    702.0,
+                    1853.0,
+                    741.0,
+                    1853.0,
+                    741.0,
+                    1874.0,
+                    702.0,
+                    1874.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    763.0,
+                    1845.0,
+                    788.0,
+                    1845.0,
+                    788.0,
+                    1865.0,
+                    763.0,
+                    1865.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1073.0,
+                    1853.0,
+                    1109.0,
+                    1853.0,
+                    1109.0,
+                    1874.0,
+                    1073.0,
+                    1874.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1133.0,
+                    1845.0,
+                    1156.0,
+                    1845.0,
+                    1156.0,
+                    1867.0,
+                    1133.0,
+                    1867.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    409.0,
+                    1865.0,
+                    441.0,
+                    1865.0,
+                    441.0,
+                    1890.0,
+                    409.0,
+                    1890.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    775.0,
+                    1864.0,
+                    808.0,
+                    1864.0,
+                    808.0,
+                    1888.0,
+                    775.0,
+                    1888.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1145.0,
+                    1864.0,
+                    1180.0,
+                    1864.0,
+                    1180.0,
+                    1888.0,
+                    1145.0,
+                    1888.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    332.0,
+                    1890.0,
+                    370.0,
+                    1890.0,
+                    370.0,
+                    1911.0,
+                    332.0,
+                    1911.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    404.0,
+                    1886.0,
+                    445.0,
+                    1886.0,
+                    445.0,
+                    1909.0,
+                    404.0,
+                    1909.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    702.0,
+                    1887.0,
+                    740.0,
+                    1887.0,
+                    740.0,
+                    1909.0,
+                    702.0,
+                    1909.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    771.0,
+                    1883.0,
+                    808.0,
+                    1883.0,
+                    808.0,
+                    1906.0,
+                    771.0,
+                    1906.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1073.0,
+                    1887.0,
+                    1110.0,
+                    1887.0,
+                    1110.0,
+                    1909.0,
+                    1073.0,
+                    1909.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1140.0,
+                    1883.0,
+                    1180.0,
+                    1883.0,
+                    1180.0,
+                    1906.0,
+                    1140.0,
+                    1906.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    408.0,
+                    1904.0,
+                    445.0,
+                    1904.0,
+                    445.0,
+                    1932.0,
+                    408.0,
+                    1932.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    774.0,
+                    1902.0,
+                    807.0,
+                    1902.0,
+                    807.0,
+                    1927.0,
+                    774.0,
+                    1927.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1140.0,
+                    1900.0,
+                    1180.0,
+                    1900.0,
+                    1180.0,
+                    1927.0,
+                    1140.0,
+                    1927.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    332.0,
+                    1925.0,
+                    373.0,
+                    1925.0,
+                    373.0,
+                    1947.0,
+                    332.0,
+                    1947.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    427.0,
+                    1934.0,
+                    449.0,
+                    1934.0,
+                    449.0,
+                    1953.0,
+                    427.0,
+                    1953.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    633.0,
+                    1934.0,
+                    656.0,
+                    1934.0,
+                    656.0,
+                    1953.0,
+                    633.0,
+                    1953.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    703.0,
+                    1920.0,
+                    741.0,
+                    1920.0,
+                    741.0,
+                    1942.0,
+                    703.0,
+                    1942.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    793.0,
+                    1930.0,
+                    816.0,
+                    1930.0,
+                    816.0,
+                    1950.0,
+                    793.0,
+                    1950.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    989.0,
+                    1932.0,
+                    1007.0,
+                    1932.0,
+                    1007.0,
+                    1946.0,
+                    989.0,
+                    1946.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1073.0,
+                    1920.0,
+                    1110.0,
+                    1920.0,
+                    1110.0,
+                    1943.0,
+                    1073.0,
+                    1943.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1164.0,
+                    1930.0,
+                    1186.0,
+                    1930.0,
+                    1186.0,
+                    1950.0,
+                    1164.0,
+                    1950.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1361.0,
+                    1934.0,
+                    1375.0,
+                    1934.0,
+                    1375.0,
+                    1944.0,
+                    1361.0,
+                    1944.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    426.0,
+                    1946.0,
+                    585.0,
+                    1946.0,
+                    585.0,
+                    1979.0,
+                    426.0,
+                    1979.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    791.0,
+                    1941.0,
+                    942.0,
+                    1941.0,
+                    942.0,
+                    1971.0,
+                    791.0,
+                    1971.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1161.0,
+                    1940.0,
+                    1312.0,
+                    1940.0,
+                    1312.0,
+                    1970.0,
+                    1161.0,
+                    1970.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1999.0,
+                    1290.0,
+                    1999.0,
+                    1290.0,
+                    2038.0,
+                    294.0,
+                    2038.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1350.0,
+                    1999.0,
+                    1405.0,
+                    1999.0,
+                    1405.0,
+                    2038.0,
+                    1350.0,
+                    2038.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    2031.0,
+                    462.0,
+                    2031.0,
+                    462.0,
+                    2068.0,
+                    295.0,
+                    2068.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1367.0,
+                    1159.0,
+                    1367.0,
+                    1159.0,
+                    1409.0,
+                    293.0,
+                    1409.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    298.0,
+                    73.0,
+                    858.0,
+                    73.0,
+                    858.0,
+                    108.0,
+                    298.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    832.0,
+                    2085.0,
+                    871.0,
+                    2085.0,
+                    871.0,
+                    2123.0,
+                    832.0,
+                    2123.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1439.0,
+                    1405.0,
+                    1439.0,
+                    1405.0,
+                    1477.0,
+                    294.0,
+                    1477.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1470.0,
+                    1408.0,
+                    1470.0,
+                    1408.0,
+                    1507.0,
+                    292.0,
+                    1507.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1499.0,
+                    1405.0,
+                    1499.0,
+                    1405.0,
+                    1539.0,
+                    291.0,
+                    1539.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1533.0,
+                    1405.0,
+                    1533.0,
+                    1405.0,
+                    1567.0,
+                    295.0,
+                    1567.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1561.0,
+                    1406.0,
+                    1561.0,
+                    1406.0,
+                    1600.0,
+                    292.0,
+                    1600.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1593.0,
+                    1405.0,
+                    1593.0,
+                    1405.0,
+                    1628.0,
+                    294.0,
+                    1628.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1619.0,
+                    452.0,
+                    1619.0,
+                    452.0,
+                    1663.0,
+                    293.0,
+                    1663.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    365.0,
+                    893.0,
+                    829.0,
+                    893.0,
+                    829.0,
+                    931.0,
+                    365.0,
+                    931.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    903.0,
+                    893.0,
+                    952.0,
+                    893.0,
+                    952.0,
+                    931.0,
+                    903.0,
+                    931.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1030.0,
+                    893.0,
+                    1269.0,
+                    893.0,
+                    1269.0,
+                    931.0,
+                    1030.0,
+                    931.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    368.0,
+                    940.0,
+                    1404.0,
+                    940.0,
+                    1404.0,
+                    973.0,
+                    368.0,
+                    973.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    394.0,
+                    971.0,
+                    922.0,
+                    971.0,
+                    922.0,
+                    1004.0,
+                    394.0,
+                    1004.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    997.0,
+                    971.0,
+                    1092.0,
+                    971.0,
+                    1092.0,
+                    1004.0,
+                    997.0,
+                    1004.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    367.0,
+                    1011.0,
+                    1402.0,
+                    1011.0,
+                    1402.0,
+                    1049.0,
+                    367.0,
+                    1049.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    395.0,
+                    1046.0,
+                    1403.0,
+                    1046.0,
+                    1403.0,
+                    1076.0,
+                    395.0,
+                    1076.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    391.0,
+                    1072.0,
+                    395.0,
+                    1072.0,
+                    395.0,
+                    1110.0,
+                    391.0,
+                    1110.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    470.0,
+                    1072.0,
+                    483.0,
+                    1072.0,
+                    483.0,
+                    1110.0,
+                    470.0,
+                    1110.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    390.0,
+                    1116.0,
+                    1122.0,
+                    1116.0,
+                    1122.0,
+                    1154.0,
+                    390.0,
+                    1154.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1195.0,
+                    1116.0,
+                    1404.0,
+                    1116.0,
+                    1404.0,
+                    1154.0,
+                    1195.0,
+                    1154.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    395.0,
+                    1149.0,
+                    1405.0,
+                    1149.0,
+                    1405.0,
+                    1183.0,
+                    395.0,
+                    1183.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    395.0,
+                    1179.0,
+                    612.0,
+                    1179.0,
+                    612.0,
+                    1213.0,
+                    395.0,
+                    1213.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    751.0,
+                    1179.0,
+                    759.0,
+                    1179.0,
+                    759.0,
+                    1213.0,
+                    751.0,
+                    1213.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    390.0,
+                    1223.0,
+                    962.0,
+                    1223.0,
+                    962.0,
+                    1257.0,
+                    390.0,
+                    1257.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1039.0,
+                    1223.0,
+                    1354.0,
+                    1223.0,
+                    1354.0,
+                    1257.0,
+                    1039.0,
+                    1257.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    394.0,
+                    1254.0,
+                    1404.0,
+                    1254.0,
+                    1404.0,
+                    1288.0,
+                    394.0,
+                    1288.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    392.0,
+                    1281.0,
+                    561.0,
+                    1281.0,
+                    561.0,
+                    1333.0,
+                    392.0,
+                    1333.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    788.0,
+                    1281.0,
+                    795.0,
+                    1281.0,
+                    795.0,
+                    1333.0,
+                    788.0,
+                    1333.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    744.0,
+                    1405.0,
+                    744.0,
+                    1405.0,
+                    779.0,
+                    294.0,
+                    779.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    775.0,
+                    1405.0,
+                    775.0,
+                    1405.0,
+                    811.0,
+                    294.0,
+                    811.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    806.0,
+                    773.0,
+                    806.0,
+                    773.0,
+                    842.0,
+                    294.0,
+                    842.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    810.0,
+                    806.0,
+                    1179.0,
+                    806.0,
+                    1179.0,
+                    842.0,
+                    810.0,
+                    842.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1214.0,
+                    806.0,
+                    1406.0,
+                    806.0,
+                    1406.0,
+                    842.0,
+                    1214.0,
+                    842.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    836.0,
+                    1149.0,
+                    836.0,
+                    1149.0,
+                    872.0,
+                    295.0,
+                    872.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    229.0,
+                    1405.0,
+                    229.0,
+                    1405.0,
+                    265.0,
+                    294.0,
+                    265.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    261.0,
+                    1408.0,
+                    261.0,
+                    1408.0,
+                    295.0,
+                    295.0,
+                    295.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    292.0,
+                    873.0,
+                    292.0,
+                    873.0,
+                    326.0,
+                    296.0,
+                    326.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    335.0,
+                    869.0,
+                    335.0,
+                    869.0,
+                    376.0,
+                    296.0,
+                    376.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    361.0,
+                    397.0,
+                    1408.0,
+                    397.0,
+                    1408.0,
+                    434.0,
+                    361.0,
+                    434.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    394.0,
+                    430.0,
+                    1405.0,
+                    430.0,
+                    1405.0,
+                    463.0,
+                    394.0,
+                    463.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    393.0,
+                    458.0,
+                    1406.0,
+                    458.0,
+                    1406.0,
+                    495.0,
+                    393.0,
+                    495.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    394.0,
+                    487.0,
+                    614.0,
+                    487.0,
+                    614.0,
+                    525.0,
+                    394.0,
+                    525.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    357.0,
+                    531.0,
+                    1405.0,
+                    531.0,
+                    1405.0,
+                    569.0,
+                    357.0,
+                    569.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    393.0,
+                    562.0,
+                    1406.0,
+                    562.0,
+                    1406.0,
+                    599.0,
+                    393.0,
+                    599.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    394.0,
+                    595.0,
+                    1405.0,
+                    595.0,
+                    1405.0,
+                    628.0,
+                    394.0,
+                    628.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    392.0,
+                    624.0,
+                    1406.0,
+                    624.0,
+                    1406.0,
+                    660.0,
+                    392.0,
+                    660.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    396.0,
+                    656.0,
+                    1405.0,
+                    656.0,
+                    1405.0,
+                    689.0,
+                    396.0,
+                    689.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    393.0,
+                    684.0,
+                    765.0,
+                    684.0,
+                    765.0,
+                    723.0,
+                    393.0,
+                    723.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 15,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    1130,
+                    1403,
+                    1130,
+                    1403,
+                    1251,
+                    299,
+                    1251
+                ],
+                "score": 0.971
+            },
+            {
+                "category_id": 3,
+                "poly": [
+                    328,
+                    225,
+                    1373,
+                    225,
+                    1373,
+                    608,
+                    328,
+                    608
+                ],
+                "score": 0.969
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    717,
+                    1404,
+                    717,
+                    1404,
+                    839,
+                    298,
+                    839
+                ],
+                "score": 0.968
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    366,
+                    1694,
+                    1403,
+                    1694,
+                    1403,
+                    1938,
+                    366,
+                    1938
+                ],
+                "score": 0.958
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    398,
+                    1943,
+                    1400,
+                    1943,
+                    1400,
+                    2034,
+                    398,
+                    2034
+                ],
+                "score": 0.956
+            },
+            {
+                "category_id": 4,
+                "poly": [
+                    299,
+                    635,
+                    1399,
+                    635,
+                    1399,
+                    699,
+                    299,
+                    699
+                ],
+                "score": 0.947
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    953,
+                    1398,
+                    953,
+                    1398,
+                    1015,
+                    297,
+                    1015
+                ],
+                "score": 0.934
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    298,
+                    1060,
+                    966,
+                    1060,
+                    966,
+                    1096,
+                    298,
+                    1096
+                ],
+                "score": 0.908
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    298,
+                    885,
+                    1195,
+                    885,
+                    1195,
+                    921,
+                    298,
+                    921
+                ],
+                "score": 0.883
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    836,
+                    2088,
+                    864,
+                    2088,
+                    864,
+                    2112,
+                    836,
+                    2112
+                ],
+                "score": 0.843
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    298,
+                    76,
+                    858,
+                    76,
+                    858,
+                    104,
+                    298,
+                    104
+                ],
+                "score": 0.836
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    365,
+                    1282,
+                    1403,
+                    1282,
+                    1403,
+                    1617,
+                    365,
+                    1617
+                ],
+                "score": 0.834
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    389,
+                    1622,
+                    1397,
+                    1622,
+                    1397,
+                    1684,
+                    389,
+                    1684
+                ],
+                "score": 0.756
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    300,
+                    76,
+                    855,
+                    76,
+                    855,
+                    104,
+                    300,
+                    104
+                ],
+                "score": 0.166
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    467,
+                    809,
+                    591,
+                    809,
+                    591,
+                    838,
+                    467,
+                    838
+                ],
+                "score": 0.91,
+                "latex": "\\Delta T = 1 0 0"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1007,
+                    2004,
+                    1091,
+                    2004,
+                    1091,
+                    2032,
+                    1007,
+                    2032
+                ],
+                "score": 0.82,
+                "latex": "0 . 5 – 1 \\%"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    607,
+                    1653,
+                    720,
+                    1653,
+                    720,
+                    1683,
+                    607,
+                    1683
+                ],
+                "score": 0.8,
+                "latex": "0 . 1 \\%"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    526,
+                    754,
+                    548,
+                    754,
+                    548,
+                    776,
+                    526,
+                    776
+                ],
+                "score": 0.77,
+                "latex": "\\alpha"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    881,
+                    718,
+                    929,
+                    718,
+                    929,
+                    746,
+                    881,
+                    746
+                ],
+                "score": 0.69,
+                "latex": "\\Delta T"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1141,
+                    638,
+                    1201,
+                    638,
+                    1201,
+                    668,
+                    1141,
+                    668
+                ],
+                "score": 0.42,
+                "latex": "R i g L"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1338,
+                    749,
+                    1395,
+                    749,
+                    1395,
+                    777,
+                    1338,
+                    777
+                ],
+                "score": 0.28,
+                "latex": "E R K"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    431.0,
+                    225.0,
+                    786.0,
+                    225.0,
+                    786.0,
+                    262.0,
+                    431.0,
+                    262.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    955.0,
+                    225.0,
+                    1371.0,
+                    225.0,
+                    1371.0,
+                    261.0,
+                    955.0,
+                    261.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    358.0,
+                    247.0,
+                    412.0,
+                    247.0,
+                    412.0,
+                    276.0,
+                    358.0,
+                    276.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    460.0,
+                    270.0,
+                    594.0,
+                    270.0,
+                    594.0,
+                    304.0,
+                    460.0,
+                    304.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    357.0,
+                    277.0,
+                    414.0,
+                    277.0,
+                    414.0,
+                    310.0,
+                    357.0,
+                    310.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    910.0,
+                    286.0,
+                    965.0,
+                    286.0,
+                    965.0,
+                    315.0,
+                    910.0,
+                    315.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    310.0,
+                    295.0,
+                    430.0,
+                    295.0,
+                    430.0,
+                    402.0,
+                    310.0,
+                    402.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    460.0,
+                    299.0,
+                    567.0,
+                    299.0,
+                    567.0,
+                    334.0,
+                    460.0,
+                    334.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    878.0,
+                    307.0,
+                    924.0,
+                    307.0,
+                    924.0,
+                    510.0,
+                    878.0,
+                    510.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    909.0,
+                    322.0,
+                    963.0,
+                    322.0,
+                    963.0,
+                    354.0,
+                    909.0,
+                    354.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    459.0,
+                    325.0,
+                    588.0,
+                    325.0,
+                    588.0,
+                    362.0,
+                    459.0,
+                    362.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    358.0,
+                    346.0,
+                    404.0,
+                    346.0,
+                    404.0,
+                    372.0,
+                    358.0,
+                    372.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    463.0,
+                    357.0,
+                    577.0,
+                    357.0,
+                    577.0,
+                    391.0,
+                    463.0,
+                    391.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    911.0,
+                    365.0,
+                    961.0,
+                    365.0,
+                    961.0,
+                    390.0,
+                    911.0,
+                    390.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    358.0,
+                    411.0,
+                    404.0,
+                    411.0,
+                    404.0,
+                    437.0,
+                    358.0,
+                    437.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    460.0,
+                    384.0,
+                    598.0,
+                    384.0,
+                    598.0,
+                    421.0,
+                    460.0,
+                    421.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    911.0,
+                    404.0,
+                    961.0,
+                    404.0,
+                    961.0,
+                    428.0,
+                    911.0,
+                    428.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    360.0,
+                    443.0,
+                    403.0,
+                    443.0,
+                    403.0,
+                    468.0,
+                    360.0,
+                    468.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    909.0,
+                    441.0,
+                    963.0,
+                    441.0,
+                    963.0,
+                    469.0,
+                    909.0,
+                    469.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    361.0,
+                    472.0,
+                    404.0,
+                    472.0,
+                    404.0,
+                    501.0,
+                    361.0,
+                    501.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    909.0,
+                    479.0,
+                    965.0,
+                    479.0,
+                    965.0,
+                    507.0,
+                    909.0,
+                    507.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1019.0,
+                    477.0,
+                    1201.0,
+                    477.0,
+                    1201.0,
+                    513.0,
+                    1019.0,
+                    513.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    359.0,
+                    505.0,
+                    404.0,
+                    505.0,
+                    404.0,
+                    534.0,
+                    359.0,
+                    534.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1018.0,
+                    504.0,
+                    1110.0,
+                    504.0,
+                    1110.0,
+                    542.0,
+                    1018.0,
+                    542.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    910.0,
+                    518.0,
+                    964.0,
+                    518.0,
+                    964.0,
+                    547.0,
+                    910.0,
+                    547.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    477.0,
+                    551.0,
+                    513.0,
+                    551.0,
+                    513.0,
+                    577.0,
+                    477.0,
+                    577.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    569.0,
+                    550.0,
+                    604.0,
+                    550.0,
+                    604.0,
+                    577.0,
+                    569.0,
+                    577.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    660.0,
+                    550.0,
+                    695.0,
+                    550.0,
+                    695.0,
+                    577.0,
+                    660.0,
+                    577.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    750.0,
+                    550.0,
+                    786.0,
+                    550.0,
+                    786.0,
+                    577.0,
+                    750.0,
+                    577.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    953.0,
+                    549.0,
+                    988.0,
+                    549.0,
+                    988.0,
+                    574.0,
+                    953.0,
+                    574.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    545.0,
+                    566.0,
+                    674.0,
+                    566.0,
+                    674.0,
+                    617.0,
+                    545.0,
+                    617.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1051.0,
+                    567.0,
+                    1275.0,
+                    567.0,
+                    1275.0,
+                    608.0,
+                    1051.0,
+                    608.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    361.75,
+                    532.5,
+                    418.75,
+                    532.5,
+                    418.75,
+                    578.5,
+                    361.75,
+                    578.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1172.0,
+                    549.0,
+                    1214.0,
+                    549.0,
+                    1214.0,
+                    574.0,
+                    1172.0,
+                    574.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    635.0,
+                    1140.0,
+                    635.0,
+                    1140.0,
+                    671.0,
+                    295.0,
+                    671.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1202.0,
+                    635.0,
+                    1404.0,
+                    635.0,
+                    1404.0,
+                    671.0,
+                    1202.0,
+                    671.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    667.0,
+                    571.0,
+                    667.0,
+                    571.0,
+                    703.0,
+                    295.0,
+                    703.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1057.0,
+                    970.0,
+                    1057.0,
+                    970.0,
+                    1101.0,
+                    291.0,
+                    1101.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    884.0,
+                    1197.0,
+                    884.0,
+                    1197.0,
+                    924.0,
+                    293.0,
+                    924.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    832.0,
+                    2085.0,
+                    869.0,
+                    2085.0,
+                    869.0,
+                    2124.0,
+                    832.0,
+                    2124.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    70.0,
+                    859.0,
+                    70.0,
+                    859.0,
+                    109.0,
+                    296.0,
+                    109.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    73.0,
+                    858.0,
+                    73.0,
+                    858.0,
+                    108.0,
+                    297.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1127.0,
+                    1404.0,
+                    1127.0,
+                    1404.0,
+                    1165.0,
+                    293.0,
+                    1165.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1159.0,
+                    1404.0,
+                    1159.0,
+                    1404.0,
+                    1194.0,
+                    294.0,
+                    1194.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1193.0,
+                    1405.0,
+                    1193.0,
+                    1405.0,
+                    1225.0,
+                    295.0,
+                    1225.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1218.0,
+                    371.0,
+                    1218.0,
+                    371.0,
+                    1257.0,
+                    292.0,
+                    1257.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    718.0,
+                    880.0,
+                    718.0,
+                    880.0,
+                    750.0,
+                    294.0,
+                    750.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    930.0,
+                    718.0,
+                    1402.0,
+                    718.0,
+                    1402.0,
+                    750.0,
+                    930.0,
+                    750.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    747.0,
+                    525.0,
+                    747.0,
+                    525.0,
+                    781.0,
+                    293.0,
+                    781.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    549.0,
+                    747.0,
+                    1337.0,
+                    747.0,
+                    1337.0,
+                    781.0,
+                    549.0,
+                    781.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1396.0,
+                    747.0,
+                    1404.0,
+                    747.0,
+                    1404.0,
+                    781.0,
+                    1396.0,
+                    781.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    776.0,
+                    1405.0,
+                    776.0,
+                    1405.0,
+                    814.0,
+                    292.0,
+                    814.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    809.0,
+                    466.0,
+                    809.0,
+                    466.0,
+                    842.0,
+                    294.0,
+                    842.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    592.0,
+                    809.0,
+                    601.0,
+                    809.0,
+                    601.0,
+                    842.0,
+                    592.0,
+                    842.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    359.0,
+                    1693.0,
+                    1404.0,
+                    1693.0,
+                    1404.0,
+                    1730.0,
+                    359.0,
+                    1730.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    394.0,
+                    1723.0,
+                    1406.0,
+                    1723.0,
+                    1406.0,
+                    1760.0,
+                    394.0,
+                    1760.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    395.0,
+                    1754.0,
+                    1406.0,
+                    1754.0,
+                    1406.0,
+                    1791.0,
+                    395.0,
+                    1791.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    394.0,
+                    1784.0,
+                    1407.0,
+                    1784.0,
+                    1407.0,
+                    1821.0,
+                    394.0,
+                    1821.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    392.0,
+                    1812.0,
+                    1407.0,
+                    1812.0,
+                    1407.0,
+                    1853.0,
+                    392.0,
+                    1853.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    392.0,
+                    1843.0,
+                    1404.0,
+                    1843.0,
+                    1404.0,
+                    1883.0,
+                    392.0,
+                    1883.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    392.0,
+                    1872.0,
+                    1406.0,
+                    1872.0,
+                    1406.0,
+                    1915.0,
+                    392.0,
+                    1915.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    392.0,
+                    1902.0,
+                    591.0,
+                    1902.0,
+                    591.0,
+                    1944.0,
+                    392.0,
+                    1944.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    394.0,
+                    1940.0,
+                    1404.0,
+                    1940.0,
+                    1404.0,
+                    1979.0,
+                    394.0,
+                    1979.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    394.0,
+                    1972.0,
+                    1403.0,
+                    1972.0,
+                    1403.0,
+                    2009.0,
+                    394.0,
+                    2009.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    395.0,
+                    2002.0,
+                    1006.0,
+                    2002.0,
+                    1006.0,
+                    2039.0,
+                    395.0,
+                    2039.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1092.0,
+                    2002.0,
+                    1263.0,
+                    2002.0,
+                    1263.0,
+                    2039.0,
+                    1092.0,
+                    2039.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    950.0,
+                    1403.0,
+                    950.0,
+                    1403.0,
+                    992.0,
+                    293.0,
+                    992.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    983.0,
+                    341.0,
+                    983.0,
+                    341.0,
+                    1017.0,
+                    295.0,
+                    1017.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    361.0,
+                    1281.0,
+                    1404.0,
+                    1281.0,
+                    1404.0,
+                    1316.0,
+                    361.0,
+                    1316.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    393.0,
+                    1312.0,
+                    1404.0,
+                    1312.0,
+                    1404.0,
+                    1347.0,
+                    393.0,
+                    1347.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    393.0,
+                    1342.0,
+                    1404.0,
+                    1342.0,
+                    1404.0,
+                    1378.0,
+                    393.0,
+                    1378.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    396.0,
+                    1373.0,
+                    1404.0,
+                    1373.0,
+                    1404.0,
+                    1407.0,
+                    396.0,
+                    1407.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    392.0,
+                    1401.0,
+                    1406.0,
+                    1401.0,
+                    1406.0,
+                    1441.0,
+                    392.0,
+                    1441.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    393.0,
+                    1432.0,
+                    1404.0,
+                    1432.0,
+                    1404.0,
+                    1468.0,
+                    393.0,
+                    1468.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    394.0,
+                    1464.0,
+                    1404.0,
+                    1464.0,
+                    1404.0,
+                    1499.0,
+                    394.0,
+                    1499.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    396.0,
+                    1496.0,
+                    1403.0,
+                    1496.0,
+                    1403.0,
+                    1527.0,
+                    396.0,
+                    1527.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    394.0,
+                    1524.0,
+                    1404.0,
+                    1524.0,
+                    1404.0,
+                    1558.0,
+                    394.0,
+                    1558.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    394.0,
+                    1555.0,
+                    1406.0,
+                    1555.0,
+                    1406.0,
+                    1590.0,
+                    394.0,
+                    1590.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    394.0,
+                    1584.0,
+                    1304.0,
+                    1584.0,
+                    1304.0,
+                    1621.0,
+                    394.0,
+                    1621.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    397.0,
+                    1619.0,
+                    1400.0,
+                    1619.0,
+                    1400.0,
+                    1654.0,
+                    397.0,
+                    1654.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    394.0,
+                    1652.0,
+                    606.0,
+                    1652.0,
+                    606.0,
+                    1687.0,
+                    394.0,
+                    1687.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    721.0,
+                    1652.0,
+                    938.0,
+                    1652.0,
+                    938.0,
+                    1687.0,
+                    721.0,
+                    1687.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 16,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 3,
+                "poly": [
+                    414,
+                    464,
+                    1283,
+                    464,
+                    1283,
+                    1797,
+                    414,
+                    1797
+                ],
+                "score": 0.967
+            },
+            {
+                "category_id": 4,
+                "poly": [
+                    515,
+                    1802,
+                    1181,
+                    1802,
+                    1181,
+                    1837,
+                    515,
+                    1837
+                ],
+                "score": 0.933
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    297,
+                    75,
+                    859,
+                    75,
+                    859,
+                    105,
+                    297,
+                    105
+                ],
+                "score": 0.906
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    836,
+                    2088,
+                    865,
+                    2088,
+                    865,
+                    2113,
+                    836,
+                    2113
+                ],
+                "score": 0.86
+            },
+            {
+                "category_id": 4,
+                "poly": [
+                    585,
+                    463,
+                    790,
+                    463,
+                    790,
+                    491,
+                    585,
+                    491
+                ],
+                "score": 0.102
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    582.0,
+                    459.0,
+                    794.0,
+                    459.0,
+                    794.0,
+                    496.0,
+                    582.0,
+                    496.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    975.0,
+                    456.0,
+                    1188.0,
+                    456.0,
+                    1188.0,
+                    498.0,
+                    975.0,
+                    498.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    471.0,
+                    489.0,
+                    523.0,
+                    489.0,
+                    523.0,
+                    514.0,
+                    471.0,
+                    514.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    561.0,
+                    487.0,
+                    608.0,
+                    487.0,
+                    608.0,
+                    513.0,
+                    561.0,
+                    513.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1106.0,
+                    487.0,
+                    1150.0,
+                    487.0,
+                    1150.0,
+                    513.0,
+                    1106.0,
+                    513.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    414.0,
+                    510.0,
+                    561.0,
+                    510.0,
+                    561.0,
+                    535.0,
+                    414.0,
+                    535.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    908.0,
+                    510.0,
+                    953.0,
+                    510.0,
+                    953.0,
+                    535.0,
+                    908.0,
+                    535.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    412.0,
+                    535.0,
+                    521.0,
+                    535.0,
+                    521.0,
+                    559.0,
+                    412.0,
+                    559.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    730.0,
+                    536.0,
+                    773.0,
+                    536.0,
+                    773.0,
+                    557.0,
+                    730.0,
+                    557.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1189.0,
+                    536.0,
+                    1233.0,
+                    536.0,
+                    1233.0,
+                    557.0,
+                    1189.0,
+                    557.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    415.0,
+                    556.0,
+                    561.0,
+                    556.0,
+                    561.0,
+                    583.0,
+                    415.0,
+                    583.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    906.0,
+                    554.0,
+                    967.0,
+                    554.0,
+                    967.0,
+                    584.0,
+                    906.0,
+                    584.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    420.0,
+                    580.0,
+                    561.0,
+                    580.0,
+                    561.0,
+                    605.0,
+                    420.0,
+                    605.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    906.0,
+                    578.0,
+                    967.0,
+                    578.0,
+                    967.0,
+                    608.0,
+                    906.0,
+                    608.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    412.0,
+                    604.0,
+                    561.0,
+                    604.0,
+                    561.0,
+                    629.0,
+                    412.0,
+                    629.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    905.0,
+                    602.0,
+                    967.0,
+                    602.0,
+                    967.0,
+                    632.0,
+                    905.0,
+                    632.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    411.0,
+                    628.0,
+                    521.0,
+                    628.0,
+                    521.0,
+                    648.0,
+                    411.0,
+                    648.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    729.0,
+                    626.0,
+                    774.0,
+                    626.0,
+                    774.0,
+                    651.0,
+                    729.0,
+                    651.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1188.0,
+                    626.0,
+                    1235.0,
+                    626.0,
+                    1235.0,
+                    651.0,
+                    1188.0,
+                    651.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    414.0,
+                    648.0,
+                    561.0,
+                    648.0,
+                    561.0,
+                    675.0,
+                    414.0,
+                    675.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    906.0,
+                    647.0,
+                    967.0,
+                    647.0,
+                    967.0,
+                    678.0,
+                    906.0,
+                    678.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    415.0,
+                    672.0,
+                    561.0,
+                    672.0,
+                    561.0,
+                    698.0,
+                    415.0,
+                    698.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    905.0,
+                    673.0,
+                    964.0,
+                    673.0,
+                    964.0,
+                    699.0,
+                    905.0,
+                    699.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    414.0,
+                    696.0,
+                    521.0,
+                    696.0,
+                    521.0,
+                    720.0,
+                    414.0,
+                    720.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    729.0,
+                    698.0,
+                    773.0,
+                    698.0,
+                    773.0,
+                    719.0,
+                    729.0,
+                    719.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1189.0,
+                    698.0,
+                    1232.0,
+                    698.0,
+                    1232.0,
+                    719.0,
+                    1189.0,
+                    719.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    415.0,
+                    719.0,
+                    561.0,
+                    719.0,
+                    561.0,
+                    744.0,
+                    415.0,
+                    744.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    909.0,
+                    719.0,
+                    965.0,
+                    719.0,
+                    965.0,
+                    744.0,
+                    909.0,
+                    744.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    412.0,
+                    741.0,
+                    559.0,
+                    741.0,
+                    559.0,
+                    769.0,
+                    412.0,
+                    769.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1050.0,
+                    742.0,
+                    1098.0,
+                    742.0,
+                    1098.0,
+                    768.0,
+                    1050.0,
+                    768.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    412.0,
+                    766.0,
+                    520.0,
+                    766.0,
+                    520.0,
+                    790.0,
+                    412.0,
+                    790.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    785.0,
+                    765.0,
+                    836.0,
+                    765.0,
+                    836.0,
+                    790.0,
+                    785.0,
+                    790.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1216.0,
+                    768.0,
+                    1265.0,
+                    768.0,
+                    1265.0,
+                    789.0,
+                    1216.0,
+                    789.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    414.0,
+                    789.0,
+                    568.0,
+                    789.0,
+                    568.0,
+                    814.0,
+                    414.0,
+                    814.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1085.0,
+                    787.0,
+                    1132.0,
+                    787.0,
+                    1132.0,
+                    814.0,
+                    1085.0,
+                    814.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    420.0,
+                    813.0,
+                    520.0,
+                    813.0,
+                    520.0,
+                    836.0,
+                    420.0,
+                    836.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    653.0,
+                    814.0,
+                    699.0,
+                    814.0,
+                    699.0,
+                    835.0,
+                    653.0,
+                    835.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1150.0,
+                    814.0,
+                    1195.0,
+                    814.0,
+                    1195.0,
+                    835.0,
+                    1150.0,
+                    835.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    414.0,
+                    835.0,
+                    568.0,
+                    835.0,
+                    568.0,
+                    860.0,
+                    414.0,
+                    860.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1086.0,
+                    833.0,
+                    1132.0,
+                    833.0,
+                    1132.0,
+                    860.0,
+                    1086.0,
+                    860.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    412.0,
+                    859.0,
+                    521.0,
+                    859.0,
+                    521.0,
+                    883.0,
+                    412.0,
+                    883.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    791.0,
+                    859.0,
+                    836.0,
+                    859.0,
+                    836.0,
+                    884.0,
+                    791.0,
+                    884.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1218.0,
+                    859.0,
+                    1266.0,
+                    859.0,
+                    1266.0,
+                    884.0,
+                    1218.0,
+                    884.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    414.0,
+                    878.0,
+                    570.0,
+                    878.0,
+                    570.0,
+                    910.0,
+                    414.0,
+                    910.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1085.0,
+                    880.0,
+                    1132.0,
+                    880.0,
+                    1132.0,
+                    907.0,
+                    1085.0,
+                    907.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    414.0,
+                    902.0,
+                    570.0,
+                    902.0,
+                    570.0,
+                    932.0,
+                    414.0,
+                    932.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1085.0,
+                    904.0,
+                    1132.0,
+                    904.0,
+                    1132.0,
+                    929.0,
+                    1085.0,
+                    929.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    412.0,
+                    928.0,
+                    521.0,
+                    928.0,
+                    521.0,
+                    951.0,
+                    412.0,
+                    951.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    788.0,
+                    927.0,
+                    836.0,
+                    927.0,
+                    836.0,
+                    953.0,
+                    788.0,
+                    953.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1218.0,
+                    927.0,
+                    1265.0,
+                    927.0,
+                    1265.0,
+                    953.0,
+                    1218.0,
+                    953.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    415.0,
+                    948.0,
+                    570.0,
+                    948.0,
+                    570.0,
+                    978.0,
+                    415.0,
+                    978.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1085.0,
+                    950.0,
+                    1132.0,
+                    950.0,
+                    1132.0,
+                    977.0,
+                    1085.0,
+                    977.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    412.0,
+                    974.0,
+                    568.0,
+                    974.0,
+                    568.0,
+                    999.0,
+                    412.0,
+                    999.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1085.0,
+                    972.0,
+                    1132.0,
+                    972.0,
+                    1132.0,
+                    999.0,
+                    1085.0,
+                    999.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    412.0,
+                    998.0,
+                    521.0,
+                    998.0,
+                    521.0,
+                    1022.0,
+                    412.0,
+                    1022.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    789.0,
+                    998.0,
+                    836.0,
+                    998.0,
+                    836.0,
+                    1023.0,
+                    789.0,
+                    1023.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1219.0,
+                    999.0,
+                    1265.0,
+                    999.0,
+                    1265.0,
+                    1020.0,
+                    1219.0,
+                    1020.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    414.0,
+                    1020.0,
+                    568.0,
+                    1020.0,
+                    568.0,
+                    1045.0,
+                    414.0,
+                    1045.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1086.0,
+                    1020.0,
+                    1132.0,
+                    1020.0,
+                    1132.0,
+                    1045.0,
+                    1086.0,
+                    1045.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    414.0,
+                    1044.0,
+                    520.0,
+                    1044.0,
+                    520.0,
+                    1068.0,
+                    414.0,
+                    1068.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    653.0,
+                    1045.0,
+                    699.0,
+                    1045.0,
+                    699.0,
+                    1066.0,
+                    653.0,
+                    1066.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1151.0,
+                    1045.0,
+                    1195.0,
+                    1045.0,
+                    1195.0,
+                    1066.0,
+                    1151.0,
+                    1066.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    411.0,
+                    1065.0,
+                    529.0,
+                    1065.0,
+                    529.0,
+                    1090.0,
+                    411.0,
+                    1090.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    821.0,
+                    1066.0,
+                    867.0,
+                    1066.0,
+                    867.0,
+                    1092.0,
+                    821.0,
+                    1092.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1233.0,
+                    1066.0,
+                    1278.0,
+                    1066.0,
+                    1278.0,
+                    1092.0,
+                    1233.0,
+                    1092.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    415.0,
+                    1090.0,
+                    520.0,
+                    1090.0,
+                    520.0,
+                    1114.0,
+                    415.0,
+                    1114.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    688.0,
+                    1092.0,
+                    732.0,
+                    1092.0,
+                    732.0,
+                    1113.0,
+                    688.0,
+                    1113.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1166.0,
+                    1089.0,
+                    1212.0,
+                    1089.0,
+                    1212.0,
+                    1114.0,
+                    1166.0,
+                    1114.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    422.0,
+                    1114.0,
+                    520.0,
+                    1114.0,
+                    520.0,
+                    1138.0,
+                    422.0,
+                    1138.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    755.0,
+                    1116.0,
+                    796.0,
+                    1116.0,
+                    796.0,
+                    1135.0,
+                    755.0,
+                    1135.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1201.0,
+                    1113.0,
+                    1244.0,
+                    1113.0,
+                    1244.0,
+                    1138.0,
+                    1201.0,
+                    1138.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    414.0,
+                    1136.0,
+                    520.0,
+                    1136.0,
+                    520.0,
+                    1160.0,
+                    414.0,
+                    1160.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    688.0,
+                    1138.0,
+                    732.0,
+                    1138.0,
+                    732.0,
+                    1159.0,
+                    688.0,
+                    1159.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1165.0,
+                    1135.0,
+                    1210.0,
+                    1135.0,
+                    1210.0,
+                    1162.0,
+                    1165.0,
+                    1162.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    412.0,
+                    1160.0,
+                    520.0,
+                    1160.0,
+                    520.0,
+                    1184.0,
+                    412.0,
+                    1184.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    820.0,
+                    1159.0,
+                    867.0,
+                    1159.0,
+                    867.0,
+                    1184.0,
+                    820.0,
+                    1184.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1233.0,
+                    1159.0,
+                    1280.0,
+                    1159.0,
+                    1280.0,
+                    1184.0,
+                    1233.0,
+                    1184.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    414.0,
+                    1183.0,
+                    520.0,
+                    1183.0,
+                    520.0,
+                    1207.0,
+                    414.0,
+                    1207.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    686.0,
+                    1183.0,
+                    733.0,
+                    1183.0,
+                    733.0,
+                    1208.0,
+                    686.0,
+                    1208.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1166.0,
+                    1183.0,
+                    1212.0,
+                    1183.0,
+                    1212.0,
+                    1208.0,
+                    1166.0,
+                    1208.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    417.0,
+                    1208.0,
+                    518.0,
+                    1208.0,
+                    518.0,
+                    1228.0,
+                    417.0,
+                    1228.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    686.0,
+                    1205.0,
+                    733.0,
+                    1205.0,
+                    733.0,
+                    1231.0,
+                    686.0,
+                    1231.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1166.0,
+                    1205.0,
+                    1212.0,
+                    1205.0,
+                    1212.0,
+                    1231.0,
+                    1166.0,
+                    1231.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    414.0,
+                    1229.0,
+                    520.0,
+                    1229.0,
+                    520.0,
+                    1253.0,
+                    414.0,
+                    1253.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    821.0,
+                    1229.0,
+                    867.0,
+                    1229.0,
+                    867.0,
+                    1254.0,
+                    821.0,
+                    1254.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1235.0,
+                    1231.0,
+                    1278.0,
+                    1231.0,
+                    1278.0,
+                    1251.0,
+                    1235.0,
+                    1251.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    414.0,
+                    1250.0,
+                    522.0,
+                    1250.0,
+                    522.0,
+                    1275.0,
+                    414.0,
+                    1275.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    686.0,
+                    1251.0,
+                    733.0,
+                    1251.0,
+                    733.0,
+                    1277.0,
+                    686.0,
+                    1277.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1166.0,
+                    1251.0,
+                    1212.0,
+                    1251.0,
+                    1212.0,
+                    1277.0,
+                    1166.0,
+                    1277.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    412.0,
+                    1275.0,
+                    520.0,
+                    1275.0,
+                    520.0,
+                    1299.0,
+                    412.0,
+                    1299.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    686.0,
+                    1274.0,
+                    733.0,
+                    1274.0,
+                    733.0,
+                    1299.0,
+                    686.0,
+                    1299.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1166.0,
+                    1274.0,
+                    1212.0,
+                    1274.0,
+                    1212.0,
+                    1299.0,
+                    1166.0,
+                    1299.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    411.0,
+                    1298.0,
+                    521.0,
+                    1298.0,
+                    521.0,
+                    1322.0,
+                    411.0,
+                    1322.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    820.0,
+                    1298.0,
+                    867.0,
+                    1298.0,
+                    867.0,
+                    1323.0,
+                    820.0,
+                    1323.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1235.0,
+                    1301.0,
+                    1278.0,
+                    1301.0,
+                    1278.0,
+                    1322.0,
+                    1235.0,
+                    1322.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    414.0,
+                    1322.0,
+                    521.0,
+                    1322.0,
+                    521.0,
+                    1345.0,
+                    414.0,
+                    1345.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    685.0,
+                    1322.0,
+                    733.0,
+                    1322.0,
+                    733.0,
+                    1347.0,
+                    685.0,
+                    1347.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1165.0,
+                    1322.0,
+                    1212.0,
+                    1322.0,
+                    1212.0,
+                    1347.0,
+                    1165.0,
+                    1347.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    414.0,
+                    1345.0,
+                    520.0,
+                    1345.0,
+                    520.0,
+                    1369.0,
+                    414.0,
+                    1369.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    685.0,
+                    1344.0,
+                    733.0,
+                    1344.0,
+                    733.0,
+                    1369.0,
+                    685.0,
+                    1369.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1165.0,
+                    1344.0,
+                    1212.0,
+                    1344.0,
+                    1212.0,
+                    1369.0,
+                    1165.0,
+                    1369.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    412.0,
+                    1368.0,
+                    521.0,
+                    1368.0,
+                    521.0,
+                    1392.0,
+                    412.0,
+                    1392.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    820.0,
+                    1368.0,
+                    867.0,
+                    1368.0,
+                    867.0,
+                    1393.0,
+                    820.0,
+                    1393.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1235.0,
+                    1369.0,
+                    1278.0,
+                    1369.0,
+                    1278.0,
+                    1390.0,
+                    1235.0,
+                    1390.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    412.0,
+                    1390.0,
+                    521.0,
+                    1390.0,
+                    521.0,
+                    1414.0,
+                    412.0,
+                    1414.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    686.0,
+                    1390.0,
+                    733.0,
+                    1390.0,
+                    733.0,
+                    1417.0,
+                    686.0,
+                    1417.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1166.0,
+                    1390.0,
+                    1212.0,
+                    1390.0,
+                    1212.0,
+                    1416.0,
+                    1166.0,
+                    1416.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    417.0,
+                    1414.0,
+                    520.0,
+                    1414.0,
+                    520.0,
+                    1438.0,
+                    417.0,
+                    1438.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    686.0,
+                    1413.0,
+                    733.0,
+                    1413.0,
+                    733.0,
+                    1439.0,
+                    686.0,
+                    1439.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1166.0,
+                    1413.0,
+                    1212.0,
+                    1413.0,
+                    1212.0,
+                    1438.0,
+                    1166.0,
+                    1438.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    414.0,
+                    1435.0,
+                    523.0,
+                    1435.0,
+                    523.0,
+                    1463.0,
+                    414.0,
+                    1463.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    820.0,
+                    1437.0,
+                    867.0,
+                    1437.0,
+                    867.0,
+                    1463.0,
+                    820.0,
+                    1463.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1235.0,
+                    1439.0,
+                    1278.0,
+                    1439.0,
+                    1278.0,
+                    1459.0,
+                    1235.0,
+                    1459.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    419.0,
+                    1460.0,
+                    520.0,
+                    1460.0,
+                    520.0,
+                    1484.0,
+                    419.0,
+                    1484.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    686.0,
+                    1460.0,
+                    733.0,
+                    1460.0,
+                    733.0,
+                    1486.0,
+                    686.0,
+                    1486.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1166.0,
+                    1460.0,
+                    1210.0,
+                    1460.0,
+                    1210.0,
+                    1486.0,
+                    1166.0,
+                    1486.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    414.0,
+                    1484.0,
+                    521.0,
+                    1484.0,
+                    521.0,
+                    1508.0,
+                    414.0,
+                    1508.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    752.0,
+                    1483.0,
+                    799.0,
+                    1483.0,
+                    799.0,
+                    1508.0,
+                    752.0,
+                    1508.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1197.0,
+                    1483.0,
+                    1245.0,
+                    1483.0,
+                    1245.0,
+                    1508.0,
+                    1197.0,
+                    1508.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    412.0,
+                    1507.0,
+                    521.0,
+                    1507.0,
+                    521.0,
+                    1531.0,
+                    412.0,
+                    1531.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    835.0,
+                    1507.0,
+                    880.0,
+                    1507.0,
+                    880.0,
+                    1532.0,
+                    835.0,
+                    1532.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1241.0,
+                    1508.0,
+                    1286.0,
+                    1508.0,
+                    1286.0,
+                    1529.0,
+                    1241.0,
+                    1529.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    415.0,
+                    1531.0,
+                    523.0,
+                    1531.0,
+                    523.0,
+                    1554.0,
+                    415.0,
+                    1554.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    767.0,
+                    1529.0,
+                    814.0,
+                    1529.0,
+                    814.0,
+                    1554.0,
+                    767.0,
+                    1554.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1207.0,
+                    1529.0,
+                    1251.0,
+                    1529.0,
+                    1251.0,
+                    1554.0,
+                    1207.0,
+                    1554.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    420.0,
+                    1553.0,
+                    520.0,
+                    1553.0,
+                    520.0,
+                    1577.0,
+                    420.0,
+                    1577.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    802.0,
+                    1553.0,
+                    847.0,
+                    1553.0,
+                    847.0,
+                    1578.0,
+                    802.0,
+                    1578.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1226.0,
+                    1554.0,
+                    1269.0,
+                    1554.0,
+                    1269.0,
+                    1575.0,
+                    1226.0,
+                    1575.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    411.0,
+                    1574.0,
+                    522.0,
+                    1574.0,
+                    522.0,
+                    1602.0,
+                    411.0,
+                    1602.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    768.0,
+                    1574.0,
+                    814.0,
+                    1574.0,
+                    814.0,
+                    1601.0,
+                    768.0,
+                    1601.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1209.0,
+                    1575.0,
+                    1251.0,
+                    1575.0,
+                    1251.0,
+                    1601.0,
+                    1209.0,
+                    1601.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    412.0,
+                    1599.0,
+                    521.0,
+                    1599.0,
+                    521.0,
+                    1623.0,
+                    412.0,
+                    1623.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    836.0,
+                    1601.0,
+                    879.0,
+                    1601.0,
+                    879.0,
+                    1622.0,
+                    836.0,
+                    1622.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1241.0,
+                    1599.0,
+                    1288.0,
+                    1599.0,
+                    1288.0,
+                    1625.0,
+                    1241.0,
+                    1625.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    412.0,
+                    1623.0,
+                    521.0,
+                    1623.0,
+                    521.0,
+                    1647.0,
+                    412.0,
+                    1647.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    768.0,
+                    1620.0,
+                    814.0,
+                    1620.0,
+                    814.0,
+                    1647.0,
+                    768.0,
+                    1647.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1209.0,
+                    1622.0,
+                    1251.0,
+                    1622.0,
+                    1251.0,
+                    1647.0,
+                    1209.0,
+                    1647.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    415.0,
+                    1645.0,
+                    521.0,
+                    1645.0,
+                    521.0,
+                    1669.0,
+                    415.0,
+                    1669.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    768.0,
+                    1644.0,
+                    814.0,
+                    1644.0,
+                    814.0,
+                    1671.0,
+                    768.0,
+                    1671.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1209.0,
+                    1645.0,
+                    1251.0,
+                    1645.0,
+                    1251.0,
+                    1671.0,
+                    1209.0,
+                    1671.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    412.0,
+                    1668.0,
+                    520.0,
+                    1668.0,
+                    520.0,
+                    1692.0,
+                    412.0,
+                    1692.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    835.0,
+                    1668.0,
+                    880.0,
+                    1668.0,
+                    880.0,
+                    1693.0,
+                    835.0,
+                    1693.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1239.0,
+                    1668.0,
+                    1288.0,
+                    1668.0,
+                    1288.0,
+                    1693.0,
+                    1239.0,
+                    1693.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    415.0,
+                    1688.0,
+                    523.0,
+                    1688.0,
+                    523.0,
+                    1717.0,
+                    415.0,
+                    1717.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    768.0,
+                    1692.0,
+                    814.0,
+                    1692.0,
+                    814.0,
+                    1717.0,
+                    768.0,
+                    1717.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1207.0,
+                    1692.0,
+                    1251.0,
+                    1692.0,
+                    1251.0,
+                    1717.0,
+                    1207.0,
+                    1717.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    467.0,
+                    1714.0,
+                    532.0,
+                    1714.0,
+                    532.0,
+                    1740.0,
+                    467.0,
+                    1740.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    797.0,
+                    1714.0,
+                    845.0,
+                    1714.0,
+                    845.0,
+                    1740.0,
+                    797.0,
+                    1740.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1221.0,
+                    1714.0,
+                    1269.0,
+                    1714.0,
+                    1269.0,
+                    1740.0,
+                    1221.0,
+                    1740.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    635.0,
+                    1732.0,
+                    743.0,
+                    1732.0,
+                    743.0,
+                    1776.0,
+                    635.0,
+                    1776.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1025.0,
+                    1732.0,
+                    1135.0,
+                    1732.0,
+                    1135.0,
+                    1776.0,
+                    1025.0,
+                    1776.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    518.0,
+                    1801.0,
+                    1180.0,
+                    1801.0,
+                    1180.0,
+                    1841.0,
+                    518.0,
+                    1841.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    73.0,
+                    858.0,
+                    73.0,
+                    858.0,
+                    108.0,
+                    297.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    832.0,
+                    2085.0,
+                    870.0,
+                    2085.0,
+                    870.0,
+                    2122.0,
+                    832.0,
+                    2122.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    586.0,
+                    461.0,
+                    792.0,
+                    461.0,
+                    792.0,
+                    496.0,
+                    586.0,
+                    496.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 17,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    365,
+                    229,
+                    1403,
+                    229,
+                    1403,
+                    445,
+                    365,
+                    445
+                ],
+                "score": 0.975
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    298,
+                    75,
+                    858,
+                    75,
+                    858,
+                    104,
+                    298,
+                    104
+                ],
+                "score": 0.855
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    837,
+                    2089,
+                    865,
+                    2089,
+                    865,
+                    2112,
+                    837,
+                    2112
+                ],
+                "score": 0.833
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    300,
+                    76,
+                    856,
+                    76,
+                    856,
+                    104,
+                    300,
+                    104
+                ],
+                "score": 0.091
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    72.0,
+                    858.0,
+                    72.0,
+                    858.0,
+                    108.0,
+                    297.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    832.0,
+                    2086.0,
+                    869.0,
+                    2086.0,
+                    869.0,
+                    2125.0,
+                    832.0,
+                    2125.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    73.0,
+                    859.0,
+                    73.0,
+                    859.0,
+                    108.0,
+                    297.0,
+                    108.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    357.0,
+                    227.0,
+                    1407.0,
+                    227.0,
+                    1407.0,
+                    267.0,
+                    357.0,
+                    267.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    393.0,
+                    260.0,
+                    1406.0,
+                    260.0,
+                    1406.0,
+                    299.0,
+                    393.0,
+                    299.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    392.0,
+                    289.0,
+                    1404.0,
+                    289.0,
+                    1404.0,
+                    330.0,
+                    392.0,
+                    330.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    394.0,
+                    322.0,
+                    1404.0,
+                    322.0,
+                    1404.0,
+                    357.0,
+                    394.0,
+                    357.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    394.0,
+                    353.0,
+                    1404.0,
+                    353.0,
+                    1404.0,
+                    388.0,
+                    394.0,
+                    388.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    393.0,
+                    383.0,
+                    1406.0,
+                    383.0,
+                    1406.0,
+                    417.0,
+                    393.0,
+                    417.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    396.0,
+                    415.0,
+                    1095.0,
+                    415.0,
+                    1095.0,
+                    449.0,
+                    396.0,
+                    449.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 18,
+            "width": 1700,
+            "height": 2200
+        }
+    }
+]
\ No newline at end of file
diff --git a/parse/train/uVPZCMVtsSG/uVPZCMVtsSG.md b/parse/train/uVPZCMVtsSG/uVPZCMVtsSG.md
new file mode 100644
index 0000000000000000000000000000000000000000..7d5a6dd653e7d1b0ac56138fa574e5b23d7248e3
Binary files /dev/null and b/parse/train/uVPZCMVtsSG/uVPZCMVtsSG.md differ
diff --git a/parse/train/uVPZCMVtsSG/uVPZCMVtsSG_content_list.json b/parse/train/uVPZCMVtsSG/uVPZCMVtsSG_content_list.json
new file mode 100644
index 0000000000000000000000000000000000000000..da21c26b395a73c3141ac834364edb25d64381cd
--- /dev/null
+++ b/parse/train/uVPZCMVtsSG/uVPZCMVtsSG_content_list.json
@@ -0,0 +1,1736 @@
+[
+    {
+        "type": "text",
+        "text": "Weisfeiler and Lehman Go Cellular: CW Networks ",
+        "text_level": 1,
+        "bbox": [
+            186,
+            122,
+            808,
+            147
+        ],
+        "page_idx": 0
+    },
+    {
+        "type": "text",
+        "text": "Cristian Bodnar⇤ University of Cambridge cb2015@cam.ac.uk ",
+        "bbox": [
+            254,
+            196,
+            421,
+            238
+        ],
+        "page_idx": 0
+    },
+    {
+        "type": "text",
+        "text": "Fabrizio Frasca⇤ Imperial College London & Twitter ffrasca@twitter.com ",
+        "bbox": [
+            506,
+            196,
+            743,
+            238
+        ],
+        "page_idx": 0
+    },
+    {
+        "type": "text",
+        "text": "Nina Otter UCLA otter@math.ucla.edu ",
+        "bbox": [
+            263,
+            261,
+            429,
+            301
+        ],
+        "page_idx": 0
+    },
+    {
+        "type": "text",
+        "text": "Yu Guang Wang MPI-MIS, SJTU & UNSW yuguang.wang@unsw.edu.au ",
+        "bbox": [
+            524,
+            260,
+            735,
+            303
+        ],
+        "page_idx": 0
+    },
+    {
+        "type": "text",
+        "text": "Pietro Liò University of Cambridge pl219@cam.ac.uk ",
+        "bbox": [
+            184,
+            323,
+            349,
+            364
+        ],
+        "page_idx": 0
+    },
+    {
+        "type": "text",
+        "text": "Guido Montúfar MPI-MIS & UCLA montufar@math.ucla.edu ",
+        "bbox": [
+            367,
+            323,
+            560,
+            364
+        ],
+        "page_idx": 0
+    },
+    {
+        "type": "text",
+        "text": "Michael Bronstein Imperial College London & Twitter mbronstein@twitter.com ",
+        "bbox": [
+            576,
+            323,
+            812,
+            364
+        ],
+        "page_idx": 0
+    },
+    {
+        "type": "text",
+        "text": "Abstract ",
+        "text_level": 1,
+        "bbox": [
+            462,
+            401,
+            535,
+            417
+        ],
+        "page_idx": 0
+    },
+    {
+        "type": "text",
+        "text": "Graph Neural Networks (GNNs) are limited in their expressive power, struggle with long-range interactions and lack a principled way to model higher-order structures. These problems can be attributed to the strong coupling between the computational graph and the input graph structure. The recently proposed Message Passing Simplicial Networks naturally decouple these elements by performing message passing on the clique complex of the graph. Nevertheless, these models can be severely constrained by the rigid combinatorial structure of Simplicial Complexes (SCs). In this work, we extend recent theoretical results on SCs to regular Cell Complexes, topological objects that flexibly subsume SCs and graphs. We show that this generalisation provides a powerful set of graph “lifting” transformations, each leading to a unique hierarchical message passing procedure. The resulting methods, which we collectively call CW Networks (CWNs), are strictly more powerful than the WL test and not less powerful than the 3-WL test. In particular, we demonstrate the effectiveness of one such scheme, based on rings, when applied to molecular graph problems. The proposed architecture benefits from provably larger expressivity than commonly used GNNs, principled modelling of higherorder signals and from compressing the distances between nodes. We demonstrate that our model achieves state-of-the-art results on a variety of molecular datasets. ",
+        "bbox": [
+            232,
+            433,
+            766,
+            680
+        ],
+        "page_idx": 0
+    },
+    {
+        "type": "text",
+        "text": "1 Introduction ",
+        "text_level": 1,
+        "bbox": [
+            174,
+            707,
+            310,
+            724
+        ],
+        "page_idx": 0
+    },
+    {
+        "type": "text",
+        "text": "The operations performed by message passing Graph Neural Networks (GNNs) emulate the structure of the input graph. While this property has clear computational advantages, it brings with it a series of fundamental limitations. As observed by Xu et al. $\\dot { \\lVert \\mathcal { I } \\dot { 4 } \\rVert }$ and Morris et al. [55] the local neighbourhood aggregations used by GNNs are at most as powerful as the Weisfeiler-Lehman (WL) test $\\pmb { \\mathbb { Z } } \\mathbf { \\mathbb { 1 } }$ in distinguishing non-isomorphic graphs. Therefore, GNNs fail to detect certain higer-order meso-scale structures such as cliques or (induced) cycles [2, 15], which are particularly important in applications dealing with social and biological networks or molecular graphs. At the same time, many such layers have to be stacked to make long-range interactions in the graph possible. Besides the computational burden this incurs, deep GNNs typically come with additional problems such as over-smoothing [51] and over-squashing [1] of the node representations. ",
+        "bbox": [
+            174,
+            738,
+            825,
+            877
+        ],
+        "page_idx": 0
+    },
+    {
+        "type": "text",
+        "text": "To address these problems, we propose a novel message passing procedure based on (regular) cell complexes, also known as CW complexes2, topological objects that form the building block of algebraic topology [38]. When paired with a theoretically-justified “lifting” transformation augmenting the graph with higher-dimensional constructs called “cells”, our method results in a multi-dimensional and hierarchical message passing procedure over the input graph. Our approach generalises and subsumes the recently proposed Message Passing Simplicial Networks (MPSNs) [8], which operate on simplicial complexes (SCs), topological generalisations of graphs. However, SCs have a rigid combinatorial structure that significantly limits the range of lifting transformations one could use to meaningfully modulate the message passing procedure. In contrast, we show that cell complexes, which in turn generalise simplicial complexes and come with additional flexibility, allow one to construct new and better ways of decoupling the input and computational graphs. ",
+        "bbox": [
+            174,
+            92,
+            825,
+            243
+        ],
+        "page_idx": 1
+    },
+    {
+        "type": "text",
+        "text": "Main Contributions To summarise, we propose a message passing scheme operating on regular cell complexes. We call this family of models CW Networks (CWNs) and study their expressive power using a cellular version of the WL test. We show that for an entire class of “lifting” transformations CWNs are at least as powerful as the WL test. Furthermore, we prove that for some of the maps in this class, CWNs can be strictly more powerful than WL, Simplicial WL (SWL) and also not less powerful than 3-WL. We also express the fundamental symmetries of these models and show how they can be seen as generalised convolutional operators on cell complexes. Experimentally, we focus our attention on a particular “lifting” map based on induced cycles. When applied to molecular graphs, it leads to an intuitive hierarchical message passing procedure involving the atoms, the bonds between them and the chemical rings of the molecules. We demonstrate that this provably powerful approach obtains state-of-the-art results on popular large-scale molecular graph datasets and other related tasks. To the best of our knowledge, this is the first work proposing a cell complex representation for molecules. Our code is available at https://github.com/twitter-research/cwn. ",
+        "bbox": [
+            173,
+            257,
+            825,
+            439
+        ],
+        "page_idx": 1
+    },
+    {
+        "type": "text",
+        "text": "2 Background ",
+        "text_level": 1,
+        "bbox": [
+            174,
+            455,
+            308,
+            473
+        ],
+        "page_idx": 1
+    },
+    {
+        "type": "text",
+        "text": "Definition 1 (Hansen and Ghrist $\\begin{array} { r l } {  { \\bigl \\| \\widehat { 3 6 } \\bigr \\| } } & { { } } \\end{array}$ ). A regular cell complex (Figure $\\mathbb { D }$ is a topological space $X$ together with a partition $\\{ X _ { \\sigma } \\} _ { \\sigma \\in P _ { X } }$ of subspaces $X _ { \\sigma }$ of $X$ called cells, and such that ",
+        "bbox": [
+            174,
+            486,
+            821,
+            516
+        ],
+        "page_idx": 1
+    },
+    {
+        "type": "text",
+        "text": "1. For each $x \\in X$ there exists an open neighbordhood of $x$ that intersects finitely many cells. ",
+        "bbox": [
+            173,
+            520,
+            787,
+            535
+        ],
+        "page_idx": 1
+    },
+    {
+        "type": "text",
+        "text": "2. For all $\\sigma , \\tau$ we have that ${ \\cal X } _ { \\tau } \\cap \\overline { { { \\cal X } _ { \\sigma } } } \\ne \\emptyset$ iff $X _ { \\tau } \\subseteq { \\overline { { X _ { \\sigma } } } }$ , where $\\overline { { X _ { \\sigma } } }$ is the closure of a cell. ",
+        "bbox": [
+            174,
+            537,
+            766,
+            554
+        ],
+        "page_idx": 1
+    },
+    {
+        "type": "text",
+        "text": "3. Every cell is homeomorphic to $\\mathbb { R } ^ { n }$ for some $n$ ",
+        "bbox": [
+            178,
+            556,
+            496,
+            571
+        ],
+        "page_idx": 1
+    },
+    {
+        "type": "text",
+        "text": "4. (Regularity) For every $\\sigma \\in P _ { X }$ there is a homeomorphism $\\phi$ of a closed ball in $\\mathbb { R } ^ { n _ { \\sigma } }$ to $\\overline { { X _ { \\sigma } } }$ such that the restriction of $\\phi$ to the interior of the ball is a homeomorphism onto $X _ { \\sigma }$ . ",
+        "bbox": [
+            176,
+            574,
+            821,
+            603
+        ],
+        "page_idx": 1
+    },
+    {
+        "type": "text",
+        "text": "We note that by condition (2) the indexing set $P _ { X }$ has a poset structure $\\tau \\leq \\sigma \\Leftrightarrow X _ { \\tau } \\subseteq \\overline { { X _ { \\sigma } } }$ , while condition (4) guarantees that this poset structure encodes all the topological information about $X$ . Thus, we can identify a regular cell complex $X$ with this poset, called face poset of $X$ . We also use $\\tau < \\sigma$ for the strict version of this partial order. ",
+        "bbox": [
+            174,
+            612,
+            581,
+            696
+        ],
+        "page_idx": 1
+    },
+    {
+        "type": "text",
+        "text": "Intuitively, one constructs a cell complex through a hierarchical gluing procedure. One starts with a set of vertices (0-cells). Then edges (1-cells) are attached to these by gluing the endpoints of closed line segments to them. We have now only described a (multi) graph. However, one can generalise this even further by taking a two-dimensional closed disk and glue its boundary (i.e. a circle) to any simple cycle in the (multi) graph previously built as in Figure $2 .$ While we are generally not concerned with dimensions above two, this can be further generalised by gluing the boundary of $n$ -dimensional balls to certain $( n - 1 )$ \u0000cells in the complex. ",
+        "bbox": [
+            174,
+            704,
+            583,
+            832
+        ],
+        "page_idx": 1
+    },
+    {
+        "type": "image",
+        "img_path": "images/eefbc01081706e78f452ed80568013eb2d16d347cad7530a3a1b1de0d350067b.jpg",
+        "image_caption": [
+            "Figure 1: A cell complex $X$ and the corresponding homeomorphisms to the closed balls for three cells of different dimensions in the complex. "
+        ],
+        "image_footnote": [],
+        "bbox": [
+            596,
+            609,
+            823,
+            761
+        ],
+        "page_idx": 1
+    },
+    {
+        "type": "text",
+        "text": "",
+        "bbox": [
+            181,
+            828,
+            820,
+            842
+        ],
+        "page_idx": 1
+    },
+    {
+        "type": "text",
+        "text": "Consider the examples in Figure $\\textcircled { 3 }$ The shown sphere is a cell complex obtained from two 0-cells (i.e. vertices), to which two 1-cells (i.e. edges), which form the equator, were attached. The boundary of two 2-dimensional disks (i.e. the two hemispheres) were glued to the equator to form a sphere. The second example is a tetrahedron with empty interior. It is a particular type of cell complex called a simplicial complex (SC). The only 2-cells it allows are triangle-shaped. More generally, the $n$ -dimensional cells of SCs are $n$ -simplices, which makes them slightly more rigid structures. ",
+        "bbox": [
+            174,
+            848,
+            825,
+            876
+        ],
+        "page_idx": 1
+    },
+    {
+        "type": "image",
+        "img_path": "images/a55263d0b752b3bc5d806b470b3759825fb4eb9c7d2ae5d1f90223fef4cc5a08.jpg",
+        "image_caption": [
+            "Figure 2: Closed two-dimensional disks are glued to the boundary of the rings present in the graph (left). The result is a 2D regular cell complex (right). "
+        ],
+        "image_footnote": [],
+        "bbox": [
+            174,
+            88,
+            584,
+            151
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "image",
+        "img_path": "images/0b72066b35dac85fbfc07fd956eee7a5fee05fd86266a0ea383297003ddffad9.jpg",
+        "image_caption": [
+            "Figure 3: A sphere and an empty tetrahedron. The latter is also a simplicial complex. "
+        ],
+        "image_footnote": [],
+        "bbox": [
+            622,
+            89,
+            800,
+            151
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "text",
+        "text": "",
+        "bbox": [
+            173,
+            217,
+            826,
+            272
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "text",
+        "text": "Definition 2. The k-skeleton of a cell complex $X$ , denoted $X ^ { ( k ) }$ , is the subcomplex of $X$ consisting of cells of dimension at most $k$ . ",
+        "bbox": [
+            171,
+            277,
+            823,
+            306
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "text",
+        "text": "This definition is useful for referring for certain parts of the complex. For instance, $X ^ { ( 0 ) }$ contains the vertices in the complex, while $X ^ { ( 1 ) }$ contains the vertices and the edges (i.e. the underlying graph). ",
+        "bbox": [
+            174,
+            319,
+            823,
+            349
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "text",
+        "text": "The combinatorial structure of the complex can be more compactly described by an incidence relation we call the boundary relation, whose reflexive and transitive closure gives the partial order defined above. The boundary relation describes what cells are on the boundary of other cells. For instance, the edges of the sphere in Figure $\\bigtriangledown$ are on the boundary of the 2-cells forming the two hemispheres. ",
+        "bbox": [
+            173,
+            354,
+            825,
+            412
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "text",
+        "text": "Definition 3. We have the boundary relation $\\sigma \\prec \\tau$ iff $\\sigma < \\tau$ and there is no cell $\\delta$ such that $\\sigma < \\delta < \\tau$ . ",
+        "bbox": [
+            171,
+            416,
+            823,
+            445
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "text",
+        "text": "We can use this to define the four types of (local) adjacencies present in cell complexes. These adjacencies will be the fundamental building block of our message passing procedure. To explain these in more familiar terms, for each adjacency, we exemplify how it shows up in graphs. ",
+        "bbox": [
+            173,
+            457,
+            826,
+            500
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "text",
+        "text": "Definition 4 (Cell complex adjacencies). For a cell complex $X$ and a cell $\\sigma \\in P _ { X }$ , we define: ",
+        "bbox": [
+            173,
+            503,
+            792,
+            520
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "text",
+        "text": "1. The boundary adjacent cells $\\boldsymbol { B } ( \\sigma ) = \\{ \\tau | \\tau \\prec \\sigma \\}$ . These are the lower-dimensional cells on the boundary of $\\sigma$ . For instance, the boundary cells of an edge are its vertices.   \n2. The co-boundary adjacent cell $\\mathcal { C } ( \\sigma ) = \\{ \\tau \\ | \\ \\sigma \\prec \\tau \\}$ . These are the higher-dimensional cells with \u0000 on their boundary. For instance, the co-boundary cells of a vertex are the edges it is part of.   \n3. The lower adjacent cells $\\mathcal { N } _ { \\downarrow } ( \\sigma ) = \\{ \\tau \\ | \\ \\exists \\delta$ such that $\\delta \\prec \\sigma$ and $\\delta \\prec \\tau \\}$ . These are the cells of the same dimension as $\\sigma$ that share a lower dimensional cell on their boundary. The line graph adjacencies between the edges are a classic example of this.   \n4. The upper adjacent cells $\\mathcal { N } _ { \\uparrow } ( \\sigma ) = \\{ \\tau \\mid \\exists i$ such that $\\sigma \\prec \\delta$ and $\\tau \\prec \\delta \\}$ . These are the cells of the same dimension as $\\sigma$ that are on the boundary of the same higher-dimensional cell as $\\sigma$ . The typical graph adjacencies between vertices are the canonical example here. ",
+        "bbox": [
+            173,
+            523,
+            828,
+            676
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "text",
+        "text": "3 Cellular Weisfeiler Lehman ",
+        "text_level": 1,
+        "bbox": [
+            174,
+            696,
+            436,
+            714
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "text",
+        "text": "Overview The results in this section show how one can transform graphs into higher-dimensional cell complexes in such a way that performing colour refinement on the resulting cell complexes makes it easier to test their isomorphism. The message passing model from Section $\\bar { 4 }$ will take advantage of these theoretical results. All proofs can be found in Appendix A. ",
+        "bbox": [
+            173,
+            729,
+            825,
+            786
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "text",
+        "text": "Definition 5. Let c be a colouring of the cells in a complex $X$ with $c _ { \\sigma }$ denoting the colour assigned to cell $\\sigma \\in P _ { X }$ . Define $B ( \\sigma , \\tau ) : = B ( \\sigma ) \\cap B ( \\tau )$ and $\\mathcal { C } ( \\sigma , \\tau ) : = \\mathcal { C } ( \\sigma ) \\cap \\mathcal { C } ( \\tau )$ . We define the following multi-sets of colours: ",
+        "bbox": [
+            173,
+            791,
+            825,
+            833
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "text",
+        "text": "1. The colours of the boundary cells of $\\sigma$ : $c _ { B } ( \\sigma ) = \\{ \\{ c _ { \\tau } \\mid \\tau \\in B ( \\sigma ) \\} \\}$ .   \n2. The colours of the co-boundary cells of $\\sigma$ : $c _ { \\mathcal { C } } ( \\sigma ) = \\{ \\{ c _ { \\tau } ~ | ~ \\tau \\in \\mathcal { C } ( \\sigma ) \\} \\}$ .   \n3. The lower adjacent colours of $\\sigma$ : $c _ { \\downarrow } ( \\sigma ) = \\{ \\{ ( c _ { \\tau } , c _ { \\delta } ) \\mid \\tau \\in \\mathcal { N } _ { \\downarrow } ( \\sigma )$ and $\\delta \\in B ( \\sigma , \\tau ) \\}$ .   \n4. The upper adjacent colours of $\\sigma$ : $c _ { \\uparrow } ( \\sigma ) = \\{ \\{ ( c _ { \\tau } , c _ { \\delta } ) \\mid \\tau \\in \\mathcal { N } _ { \\uparrow } ( \\sigma )$ and $\\delta \\in \\mathcal { C } ( \\sigma , \\tau ) \\big \\}$ . ",
+        "bbox": [
+            171,
+            838,
+            746,
+            910
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "image",
+        "img_path": "images/f132e64580feb98e0dddba934ea49e01c6fa4a4cc004b0c67d64ceaec82753eb.jpg",
+        "image_caption": [
+            "Figure 4: The CWL colouring procedure for the yellow edge of the cell complex. All cells have been assigned unique colours to aid the visualisation of the adjacencies. Note that the yellow edge aggregates long-range information from the light green edge. "
+        ],
+        "image_footnote": [],
+        "bbox": [
+            209,
+            89,
+            792,
+            194
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "Note that unlike in graphs and simplicial complexes, the sets $B ( \\sigma , \\tau )$ and $\\mathcal { C } ( \\sigma , \\tau )$ can have more than one element. For instance, two (closed) 2-cells might intersect in more than one edge (e.g. the two hemispheres in Figure $\\textcircled { 3 }$ , and conversely, two edges might be on the boundary of the same two 2-cells. This illustrates the more flexible combinatorial structure of cell complexes. ",
+        "bbox": [
+            174,
+            268,
+            825,
+            324
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "Cellular WL (CWL) We consider CWL, a colour refinement scheme for cell complexes that generalises the Simplicial WL $\\textcircled { 8 } \\textcircled { 1 8 }$ and WL $\\mathbb { \\ m }$ tests. We use $c _ { \\sigma } ^ { t }$ to refer to the colour assigned by CWL to cell $\\sigma$ at iteration $t$ of the algorithm. When the input is a simplicial complex, this recovers the SWL algorithm. A step of the algorithm is graphically depicted in Figure $^ 4$ for a single cell. ",
+        "bbox": [
+            174,
+            338,
+            825,
+            395
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "1. Given a regular cell complex $X$ , all the cells $\\sigma$ are initialised with the same colour.   \n2. Given the colour $c _ { \\sigma } ^ { t }$ of cell $\\sigma$ at iteration $t$ , we compute the colour of cell $\\sigma$ at the next iteration $c _ { \\sigma } ^ { t + 1 }$ by injectively mapping the multi-sets of colours belonging to the adjacent cells of $\\sigma$ using a perfect HASH function: $c _ { \\sigma } ^ { t + 1 } = \\mathrm { H A S H } \\big ( c _ { \\sigma } ^ { t } , c _ { \\mathcal { B } } ^ { t } ( \\sigma ) , c _ { \\mathcal { C } } ^ { t } ( \\sigma ) , c _ { \\mathcal { A } } ^ { t } ( \\sigma ) , c _ { \\uparrow } ^ { t } ( \\sigma ) \\big )$ .   \n3. The algorithm stops when a stable colouring is reached. Two cell complexes are considered non-isomorphic if their colour histograms are different. Otherwise, the test is inconclusive. ",
+        "bbox": [
+            173,
+            402,
+            830,
+            497
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "First, we state the following theorem from Bodnar et al. $\\textcircled { 8 } \\textcircled { 1 8 }$ involving SWL and simplicial complexes. This theorem shows that on simplicial complexes, certain adjacencies can be pruned without affecting the non-isomorphic SCs that can be distinguished. This has important computational implications. ",
+        "bbox": [
+            176,
+            506,
+            823,
+            549
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "Theorem 6. SWL without coboundary and lower-adjacencies has the same expressive power in distinguishing non-isomorphic simplicial complexes as SWL with the complete set of adjacencies. ",
+        "bbox": [
+            173,
+            551,
+            823,
+            579
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "It is not immediately clear whether an equivalent theorem would also hold for cell complexes. This is because cells, unlike simplices, can have widely different shapes and, as described above, the adjacencies between them take more complicated forms. Nevertheless, we show that a positive result can be obtained. ",
+        "bbox": [
+            174,
+            588,
+            825,
+            643
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "Theorem 7. CWL without coboundary and lower-adjacencies has the same expressive power in distinguishing non-isomorphic cell complexes as CWL with the complete set of adjacencies. ",
+        "bbox": [
+            173,
+            646,
+            823,
+            675
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "We note this does not mean that the removed adjacencies are completely redundant in practice. Even if they are not needed from a (theoretical) colour refinment perspective, they might still include important inductive biases that make them suitable for certain tasks. ",
+        "bbox": [
+            174,
+            684,
+            825,
+            726
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "We are now interested in examining various procedures for mapping, or “lifting”, graphs into the space of regular cell complexes. Such a procedure can be used to test the isomorphism of two graphs by performing colour refinement on the cell complexes they are mapped to. The hope is that CWL applied to these cell complexes is more powerful than WL applied to the initial graphs. We will later show that for a wide range of transformations, this is indeed the case. We start by rigorously defining what we mean by a “lifting”. ",
+        "bbox": [
+            174,
+            732,
+            825,
+            815
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "Definition 8. A cellular lifting map is a function $f : \\mathcal G  \\mathcal { X }$ from the space of graphs $\\mathcal { G }$ to the space of regular cell complexes $\\mathcal { X }$ with the property that two graphs $G _ { 1 } , G _ { 2 }$ are isomorphic iff the cell complexes $f ( G _ { 1 } ) , f ( \\mathbf { \\bar { G } } _ { 2 } )$ are isomorphic. ",
+        "bbox": [
+            174,
+            819,
+            825,
+            861
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "This property ensures that testing the isomorphism of the two cell complexes is equivalent to testing the isomorphism in the input graphs. This would not be the case if two non-isomoprhic graphs were mapped to the same cell complex. ",
+        "bbox": [
+            174,
+            869,
+            825,
+            911
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "Example 9. It can be verified that the function mapping each graph to its clique complex (i.e. every $( k + 1 )$ -clique in the graph becomes a $k$ -simplex) is a cellular lifting map. ",
+        "bbox": [
+            171,
+            92,
+            823,
+            119
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "text",
+        "text": "The clique complex lifting map from Example 9 has been used by Bodnar et al. $\\pmb { \\mathbb { B } } ] \\mathbf { l }$ to show that SWL is strictly more powerful than WL. We restate this result: ",
+        "bbox": [
+            169,
+            127,
+            825,
+            156
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "text",
+        "text": "Theorem 10. SWL with clique complex lifting is strictly more powerful than WL. ",
+        "bbox": [
+            174,
+            160,
+            707,
+            174
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "text",
+        "text": "A natural question is what other lifting transformations make CWL strictly more powerful than WL? We first describe a space of lifting transformations that make CWL at least as powerful as WL. ",
+        "bbox": [
+            173,
+            183,
+            820,
+            212
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "text",
+        "text": "Definition 11. A lifting map is skeleton-preserving if for any graph $G$ , the 1-skeleton of $f ( G )$ and $G$ are isomorphic as (multi) graphs. ",
+        "bbox": [
+            176,
+            214,
+            825,
+            242
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "text",
+        "text": "Intuitively, skeleton-preserving liftings ensure that the additional structure added by the lifting map comes from attaching cells of dimension at least two to the graph. These mappings keep the 0-cells and 1-cells intact and are, therefore, restricted from making modifications to the input graph structure. An important remark is that for simplicial complexes, attaching simplices based on cliques present in the graph is the only possible skeleton preserving transformation. Once again, this illustrates the limitations of simplicial complexes for adding useful higher-dimensional structures to the graph. ",
+        "bbox": [
+            174,
+            251,
+            825,
+            335
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "text",
+        "text": "Example 12. The function from Example 9 is also skeleton-preserving because the 1-skeleton of the clique complex of a graph is trivially isomorphic to the graph. A lifting function mapping each graph to a multi-graph where each edge is doubled by a parallel edge is not skeleton-preserving (Figure 5) ",
+        "bbox": [
+            174,
+            337,
+            823,
+            380
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "text",
+        "text": "We now show that all the maps in the skeleton-preserving class have the following desirable property: ",
+        "bbox": [
+            173,
+            388,
+            581,
+            417
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "text",
+        "text": "Theorem 13. Let $f$ be a skeleton-preserving lifting map. Then $C W L ( f )$ (i.e. CWL using lifting $f$ ) is at least as powerful as WL in distinguishing non-isomorphic graphs. ",
+        "bbox": [
+            173,
+            419,
+            581,
+            462
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "image",
+        "img_path": "images/fb663b1a2f9a5a4fcc9886ee88e4877662331081ae9bf2749c10d4b141fb7fb5.jpg",
+        "image_caption": [
+            "Figure 5: A graph, its clique complex and the graph with duplicated edges. The first map is skeletonpreserving, while the second is not. "
+        ],
+        "image_footnote": [],
+        "bbox": [
+            594,
+            391,
+            823,
+            453
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "text",
+        "text": "To prove that some of these make CWL strictly more powerful than WL, it is sufficient to find a pair of graphs that cannot be distinguished by WL, but can be distinguished by CWL. The following result gives examples of such maps. ",
+        "bbox": [
+            173,
+            470,
+            581,
+            526
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "text",
+        "text": "Definition 14. Let $k$ -CL, $k$ -IC, $k$ -C be the lifting maps attaching cells to all the cliques, induced cycles and simple cycles, respectively, of size at most $k$ . ",
+        "bbox": [
+            174,
+            529,
+            823,
+            558
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "text",
+        "text": "Corollary 15. For all $k \\geq 3$ , CWL( $k$ -CL), CWL $k$ -IC) and $C W L ( k { \\mathrm { - C } } )$ are strictly more powerful than WL. ",
+        "bbox": [
+            176,
+            559,
+            821,
+            587
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "text",
+        "text": "We note that this is not a complete list. For instance, the result can also be extended to combinations of the above or other transformations. We can also relate CWL to the higher-order 3-WL test. ",
+        "bbox": [
+            174,
+            597,
+            821,
+            626
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "text",
+        "text": "Theorem 16. There exists a pair of graphs indistinguishable by 3-WL but distinguishable by CWL(kCL) with $k \\geq 4$ , CWL( $k$ -IC) with $k \\geq 4$ and $C W L ( k { \\mathrm { - C } } )$ with $k \\geq 8 .$ . ",
+        "bbox": [
+            171,
+            627,
+            825,
+            656
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "text",
+        "text": "Finally, we conclude this section by showing how CWL can achieve a superior expressive power compared to SWL. This result is proven by Corollary $3 1$ in the Appendix. ",
+        "bbox": [
+            173,
+            665,
+            823,
+            694
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "text",
+        "text": "Theorem 17. Let $k$ -CL [ $k$ -IC and $k { \\mathrm { - C L } } \\cup k $ -C denote combined liftings attaching cells to the union of the specified substructures. $C W L ( k _ { 1 } { \\mathrm { - C L } } \\cup k _ { 2 } { \\mathrm { - I C } } )$ and $C W L ( k _ { 1 } { \\mathrm { - C L } } \\cup k _ { 2 } { \\mathrm { - C } } )$ are strictly more powerful than SWL $k _ { 1 }$ -CL) for all $k _ { 2 } \\geq 5$ . ",
+        "bbox": [
+            174,
+            695,
+            825,
+            738
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "text",
+        "text": "4 Molecular Message Passing with CW Networks ",
+        "text_level": 1,
+        "bbox": [
+            174,
+            756,
+            599,
+            773
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "text",
+        "text": "We now describe CW Networks with an applied focus on molecular graphs to ground the discussion. Therefore, from now on we assume the use of the skeleton-preserving lifting transformation that attaches 2-cells to all the induced cycles (i.e. chordless cycles) in the graph as in Figure $2 .$ This leads to a message passing procedure involving atoms (vertices / 0-cells), the bonds between atoms (edges / 1-cells) and chemical rings (induced cycles / 2-cells). Additionally, in virtue of Theorem $\\underline { { \\overline { { 7 } } } } ,$ , we consider only the boundary and upper adjacencies between these cells without sacrificing the expressive power. The equations for the other adjacencies, which we do not use, can be found in Appendix A. We note however, that the theoretical results in this section are general and not particular to these specific choices of adjacencies and lifting transformation. ",
+        "bbox": [
+            173,
+            786,
+            825,
+            911
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "text",
+        "text": "Molecular Message Passing The cells in our CW Network receive two types of messages: ",
+        "bbox": [
+            169,
+            90,
+            781,
+            106
+        ],
+        "page_idx": 5
+    },
+    {
+        "type": "equation",
+        "img_path": "images/12808d6a3dee80c3fb7dc29e8406e3d8e577d1a1735f53ecc892c97dd6105a11.jpg",
+        "text": "$$\n\\begin{array} { r } { n _ { \\mathrm { { B } } } ^ { t + 1 } ( \\sigma ) = \\mathrm { A G G } _ { \\tau \\in \\mathcal { B } ( \\sigma ) } \\left( M _ { B } \\big ( h _ { \\sigma } ^ { t } , h _ { \\tau } ^ { t } \\big ) \\right) \\qquad m _ { \\uparrow } ^ { t + 1 } ( \\sigma ) = \\mathrm { A G G } _ { \\tau \\in \\mathcal { N } _ { \\uparrow } ( \\sigma ) , \\delta \\in \\mathcal { C } ( \\sigma , \\tau ) } \\Big ( M _ { \\uparrow } \\big ( h _ { \\sigma } ^ { t } , h _ { \\tau } ^ { t } , h _ { \\delta } ^ { t } \\big ) \\Big ) . } \\end{array}\n$$",
+        "text_format": "latex",
+        "bbox": [
+            181,
+            111,
+            826,
+            138
+        ],
+        "page_idx": 5
+    },
+    {
+        "type": "text",
+        "text": "The first specifies messages from atoms to bonds and from bonds to rings. The second type of message, specifies messages between atoms connected by a bond and messages between bonds that are part of the same ring (Figure $\\textcircled{6}$ . Note that for the second type of adjacency, when two atoms communicate, we include the features of the bond between them. Similarly, when two bonds communicate, we include the features of the ring they communicate through. The update operation takes into account these two types of incoming messages and updates the features of the cells: ",
+        "bbox": [
+            173,
+            143,
+            825,
+            228
+        ],
+        "page_idx": 5
+    },
+    {
+        "type": "equation",
+        "img_path": "images/223e4f9af63809eeefc62a6d123be7d6e8e636e54958be1c6047f1bfa8df0c96.jpg",
+        "text": "$$\nh _ { \\sigma } ^ { t + 1 } = U \\Big ( h _ { \\sigma } ^ { t } , m _ { B } ^ { t } ( \\sigma ) , m _ { \\uparrow } ^ { t + 1 } ( \\sigma ) \\Big ) .\n$$",
+        "text_format": "latex",
+        "bbox": [
+            382,
+            234,
+            614,
+            261
+        ],
+        "page_idx": 5
+    },
+    {
+        "type": "text",
+        "text": "To obtain a global embedding for a cell complex $X$ from a model with $L$ layers, the readout function takes as input the separate multi-sets of features corresponding to the atoms, bonds and the rings: ",
+        "bbox": [
+            173,
+            266,
+            825,
+            295
+        ],
+        "page_idx": 5
+    },
+    {
+        "type": "equation",
+        "img_path": "images/43496a4cccd9f2ca75248a3ad48ca01ed697d3230c15cf436492f4d1f4fa723d.jpg",
+        "text": "$$\nh _ { X } = \\mathrm { R E A D O U T } ( \\{ h _ { \\sigma } ^ { L } \\} | _ { d i m ( \\sigma ) = 0 } , \\{ \\{ h _ { \\sigma } ^ { L } \\} \\} _ { d i m ( \\sigma ) = 1 } , \\{ \\{ h _ { \\sigma } ^ { L } \\} \\} _ { d i m ( \\sigma ) = 2 } ) .\n$$",
+        "text_format": "latex",
+        "bbox": [
+            267,
+            301,
+            728,
+            321
+        ],
+        "page_idx": 5
+    },
+    {
+        "type": "text",
+        "text": "Expressivity Naturally, the ability of CWNs to distinguish non-isomorphic regular cell complexes is bounded by CWL. Similarly to GNNs and WL, CWNs can also be shown to be as powerful as CWL as long as they are equipped with a sufficient number of layers and the parametric local aggregators they use can learn to be injective. Multiple such multi-set aggregators [20, $\\textcircled { 7 4 } |$ are known to exist and can be directly employed in our model. ",
+        "bbox": [
+            173,
+            332,
+            666,
+            416
+        ],
+        "page_idx": 5
+    },
+    {
+        "type": "text",
+        "text": "Theorem 18. CW Networks are at most as powerful as CWL. Additionally, when using injective neighbourhood aggregators and a sufficient number of layers, CWNs are as powerful as CWL. ",
+        "bbox": [
+            173,
+            421,
+            666,
+            463
+        ],
+        "page_idx": 5
+    },
+    {
+        "type": "text",
+        "text": "Corollary $\\boxed { 1 5 }$ states that CWL is strictly more powerful than the standard WL when the lifting procedure attaches 2-cells to induced cycles of size $k \\geq 3$ . As a consequence of Theorem 18, this result also holds for molecular message passing CWNs equipped with injective aggregators. In practice, $k$ is to be considered as a standard hyperparameter, and its choice can either be driven by validation set performance, or by domain knowledge (if available). ",
+        "bbox": [
+            173,
+            473,
+            666,
+            558
+        ],
+        "page_idx": 5
+    },
+    {
+        "type": "image",
+        "img_path": "images/5dfc800ff4d91f10c1b21ac965ab4a1939dc1993dca160cda884fff19197fd1f.jpg",
+        "image_caption": [
+            "Figure 6: Hierarchical depiction of the message passing procedure. Orange arrows indicate boundary messages received by cells $\\sigma$ and $\\tau$ , while blue ones show upper messages received by cells $\\tau$ and $\\delta$ . "
+        ],
+        "image_footnote": [],
+        "bbox": [
+            678,
+            335,
+            825,
+            549
+        ],
+        "page_idx": 5
+    },
+    {
+        "type": "text",
+        "text": "Symmetries Given a graph $G$ with adjacency matrix $\\pmb { A }$ and feature matrix $\\boldsymbol { X }$ , a function $f$ is (node) permutation equivariant if $P f ( A , X ) =$ $f ( P A P ^ { T } , P X )$ , for any permutation matrix $_ { r }$ . GNN layers respect this equation, which ensures they compute the same functions up to a permutation (i.e. relabeling) of the nodes. Similarly, it can be shown that CW Networks are equivariant with respect to permutations of the cells and corresponding permutations of the boundary relations $\\sigma \\prec \\tau$ between cells. We define this notion of equivariance more formally in Appendix C. ",
+        "bbox": [
+            174,
+            571,
+            665,
+            684
+        ],
+        "page_idx": 5
+    },
+    {
+        "type": "text",
+        "text": "Theorem 19. CW Network layers are cell permutation equivariant. ",
+        "bbox": [
+            173,
+            688,
+            617,
+            702
+        ],
+        "page_idx": 5
+    },
+    {
+        "type": "text",
+        "text": "Long-Range Interactions Several graph-related tasks require the ability to capture long-range interactions between nodes. For instance, certain molecular properties depend on atoms placed on the opposite sides of a ring [31, 60]. As a consequence of the coupling between the input and computational graphs, $L$ message passing operations are necessary in GNNs to let a node receive information from an $L$ -hops distant node. In contrast, our hierarchical message passing scheme requires at most $L$ layers since 2-cells create shortcuts. For example, a constant number of CWN layers (3) is enough to capture dependencies between atoms on the opposite sides of a ring, independently of the ring size. In Section $5 . { \\overset { \\cdot } { 1 } }$ we verify this in a controlled scenario. Additional experiments on real world graphs in Section $5 . 2$ confirm that it can achieve state-of-the-art performance with a limited number of layers. ",
+        "bbox": [
+            173,
+            715,
+            826,
+            842
+        ],
+        "page_idx": 5
+    },
+    {
+        "type": "text",
+        "text": "Anisotropic Filters Due to the lack of a canonical ordering between neighbours, many common GNNs use symmetric convolutional kernels, resulting in isotropic filters treating neighbours equally. Recent works have proposed to address this limitation by employing additional structural information [6, 10]. CWNs also implicitly achieve this form of anisotropy by integrating information from ",
+        "bbox": [
+            176,
+            856,
+            825,
+            911
+        ],
+        "page_idx": 5
+    },
+    {
+        "type": "image",
+        "img_path": "images/e7e513d4a4a37f8f0a963dcd756c8d2c678cb9d6e65f18f363336e6494cbaf6d.jpg",
+        "image_caption": [],
+        "image_footnote": [],
+        "bbox": [
+            176,
+            90,
+            459,
+            262
+        ],
+        "page_idx": 6
+    },
+    {
+        "type": "text",
+        "text": "(a) RingTransfer Results. Accuracy is over 5 balanced classes. A score of 0.2 is equivalent to a random guess. Error bars show the min and max. Our model obtains high-scores in average even for large rings despite using only three layers. ",
+        "bbox": [
+            173,
+            270,
+            462,
+            334
+        ],
+        "page_idx": 6
+    },
+    {
+        "type": "image",
+        "img_path": "images/61b45b8b73568ed72007633ef6dafbc69a26a971d31f6e09c325a41d0ecbc3e1.jpg",
+        "image_caption": [],
+        "image_footnote": [],
+        "bbox": [
+            482,
+            90,
+            821,
+            287
+        ],
+        "page_idx": 6
+    },
+    {
+        "type": "text",
+        "text": "(b) Failure rates on the SR isomorphism task, the smaller the better (mean and std-error over 5 runs). In parantheses, for each model, the maximum size $k$ of rings lifted to 2-cells. ",
+        "bbox": [
+            477,
+            296,
+            825,
+            334
+        ],
+        "page_idx": 6
+    },
+    {
+        "type": "text",
+        "text": "Figure 7: Results on the RingTransfer and SR synthetic benchmarks. ",
+        "bbox": [
+            272,
+            343,
+            722,
+            358
+        ],
+        "page_idx": 6
+    },
+    {
+        "type": "text",
+        "text": "the higher-order cells and their associated substructures into the message passing procedure. For instance, bond features can learn to encode their membership to a ring and also communicate directly with other bonds present in the ring. Consequently, the messages between atoms connected through these bonds are modulated by the presence of the ring as well as by the presence of other nodes and bonds part of that ring. ",
+        "bbox": [
+            174,
+            378,
+            825,
+            449
+        ],
+        "page_idx": 6
+    },
+    {
+        "type": "text",
+        "text": "CWNs as Generalised Convolutions Our message passing scheme can be seen a (non-linear) generalisation of linear diffusion operators on cell complexes. Recent works $\\pm \\boxed { 1 3 } \\boxed { 2 5 } \\parallel$ have introduced convolutional operators on SCs by employing the Hodge Laplacian $\\pmb { \\mathbb { \\left\\| 6 3 \\right\\| } }$ , a generalisation of the graph Laplacian. By leveraging on the cellular Sheaf Laplacian $\\lVert \\dot { 3 6 } \\rVert$ , a similar construction can be extended to cell complexes to define cellular convolutional operators. In Appendix D we discuss this approach and show that our cellular message passing scheme subsumes it. This represents a promising avenue for studying CWNs from a spectral perspective, an endeavour we leave for future work. ",
+        "bbox": [
+            173,
+            473,
+            825,
+            570
+        ],
+        "page_idx": 6
+    },
+    {
+        "type": "text",
+        "text": "Computational Complexity When considering cells of a constant maximum dimension and boundary size, the computational complexity of the message passing scheme is linear in the size of the input complex. For the molecular applications we are interested in, the average number of rings per molecule is upper bounded by a small constant (e.g. three for MOLHIV), so the size of the complex is approximately the same as the size of the graph. Therefore, in this setting, the computational complexity of the model is similar to that of message passing GNNs. Separately of this, the one-time preprocessing step of computing the lifting of the graphs should also be considered. The $C$ induced cycles in a graph can be listed in $\\mathcal { O } \\big ( ( | E | ^ { - } | V | \\bar { C } \\big )$ polylog $| V | )$ time $\\pmb { \\Vert 2 6 \\Vert }$ . Again, given that $C$ is upper bounded by a small constant for the molecular datasets of interest in this work, the complexity of the lifting procedure is also almost linear in the size of the graph. A more detailed analysis backed up by wall-clock time experiments is given in Appendix B. ",
+        "bbox": [
+            173,
+            594,
+            825,
+            748
+        ],
+        "page_idx": 6
+    },
+    {
+        "type": "text",
+        "text": "5 Experiments ",
+        "text_level": 1,
+        "bbox": [
+            174,
+            776,
+            312,
+            794
+        ],
+        "page_idx": 6
+    },
+    {
+        "type": "text",
+        "text": "In this section we validate the theoretical and empirical properties of our proposed message passing scheme in controlled scenarios as well as in real-world graph classification problems, with a focus on large scale molecular benchmarks. For simplicity, in all experiments we employ a model which stacks CWN layers with local aggregators as in GIN $\\pmb { \\Vert 7 4 \\Vert }$ . We name our architecture “Cell Isomorphism Network” (CIN). 0-cells are always endowed with the original node features; higher-dimensional cells are populated in a benchmark specific manner. See Appendix $\\mathrm { ~ E ~ }$ for details on feature initialisation, message passing and readout operations, hyperparameters, implementation and benchmark statistics. ",
+        "bbox": [
+            174,
+            814,
+            825,
+            911
+        ],
+        "page_idx": 6
+    },
+    {
+        "type": "text",
+        "text": "5.1 Synthetic Benchmarks ",
+        "text_level": 1,
+        "bbox": [
+            174,
+            92,
+            370,
+            106
+        ],
+        "page_idx": 7
+    },
+    {
+        "type": "text",
+        "text": "CSL Circular Skip Link dataset was first introduced in $ { \\mathbb { B } } 5 7  { \\mathbb { I } }$ and has been recently adopted as a reference benchmark to test the expressivity of GNNs $\\mathbb { \\left[ \\left[ 2 4 \\right] \\right] }$ . It consists of 150 4-regular graphs from 10 different isomorphism classes, which we need to predict. Unsolvable by the WL test and message passing approaches [14, 57], we use it to validate the expressive power of CWNs. ",
+        "bbox": [
+            173,
+            117,
+            825,
+            172
+        ],
+        "page_idx": 7
+    },
+    {
+        "type": "table",
+        "img_path": "images/1dbe2e32b816a9c594dadd34b5912275a1943a122f54f248f6bb3cd4dbbd0f5f.jpg",
+        "table_caption": [
+            "Table 1: Classification accuracy on CSL. "
+        ],
+        "table_footnote": [],
+        "table_body": "<table><tr><td>Method</td><td>Mean</td><td>Min</td><td>Max</td></tr><tr><td>MP-GNNs</td><td>10.000±0.000</td><td>10.000</td><td>10.000</td></tr><tr><td>RingGNN</td><td>10.000±0.000</td><td>10.000</td><td>10.000</td></tr><tr><td>3WLGNN</td><td>97.800±10.916</td><td>30.000</td><td>100.000</td></tr><tr><td>CIN (Ours)</td><td>100.000±0.000</td><td>100.000</td><td>100.000</td></tr></table>",
+        "bbox": [
+            176,
+            202,
+            452,
+            273
+        ],
+        "page_idx": 7
+    },
+    {
+        "type": "text",
+        "text": "We follow the same evaluation setting as Dwivedi et al. [24]: 5-fold cross validation procedure and 20 different random weight initialisations. For our model, we set the maximum ring size $k = 8$ . In Table 1 we follow the common practice on this dataset and report the mean, minimum and maximum test accuracy obtained by CIN over the 100 runs, along with the results by the baselines presented in Dwivedi et al. [24]. MP-GNNs, that is classic message passing GNNs (GAT $\\mathbb { \\lVert \\rVert }$ , MoNet [54], GIN $\\pmb { \\Vert 7 4 \\Vert }$ , etc.), and RingGNN [14] perform as random guessers. In contrast, our model is able to identify the isomorphism class of each test graph in every run while featuring only a fraction of the computational complexity of 3WLGNN, the best performing reference baseline $\\dot { \\mathbb { B } } 4 \\vert 5 3 \\vert $ . ",
+        "bbox": [
+            465,
+            179,
+            825,
+            275
+        ],
+        "page_idx": 7
+    },
+    {
+        "type": "text",
+        "text": "",
+        "bbox": [
+            174,
+            276,
+            825,
+            332
+        ],
+        "page_idx": 7
+    },
+    {
+        "type": "text",
+        "text": "SR Similarly to Bodnar et al. $\\textcircled { 8 } \\textcircled { 1 8 }$ and $\\mathbb { \\ m }$ , we consider Strongly Regular graphs within the same family as hard examples of non-isomorphic graphs we seek to distinguish. Any pair of graphs within the same family cannot provably be distinguished by 3-WL test $\\bar { \\bigstar } \\bar { \\bigstar } 0 \\bigstar$ . We reproduce the same experimental setting of Bodnar et al. $\\textcircled { 8 } \\textcircled { 1 8 }$ . In particular, we consider 9 distinct SR families3 and run our model untrained on the cell complex lifting of each graph, with $k = 4 , 5 , 6$ . 0-cells (nodes) are initialised with a constant unitary signal, while 1- and 2-cells are initialised with the sum of the contained 0-cells. We additionally run an MLP baseline with sum readout to appreciate the contribution of message passing. We report the percentage of non-distinguished pairs in Figure $\\boxed { 7 6 }$ Contrary to 3-WL, both CIN and the MLP baseline are able to distinguish many pairs across all families, with better performance attained for larger $k$ . For $k = 6$ , we observed CIN to disambiguate all pairs in all families $0 . 0 \\%$ failure rate). Despite the strong results achieved by the baseline, we found CIN to always distinguish a larger number of non-isomorphic pairs for the same values of $k$ this confirming the importance of cellular message passing. ",
+        "bbox": [
+            173,
+            348,
+            825,
+            529
+        ],
+        "page_idx": 7
+    },
+    {
+        "type": "text",
+        "text": "RingTransfer In order to empirically validate the ability of CIN to capture long-range node dependencies, we additionally design a third synthetic benchmark dubbed as ‘RingTransfer’. Graphs in this dataset are chordless cycles (rings) of size $k$ . In each graph we mark two special nodes as target and source, always placed at distance $\\lfloor \\frac { k } { 2 } \\rfloor$ . The task is for target to output the one-hot encoded label assigned to source. All other nodes in the ring are assigned a unitary constant feature vector. A model has to learn to transfer the information contained in source to the opposite side of the ring, where target resides. We initialise $1 \\mathrm { - }$ and 2-dimensional cells with a null signal. In Figure $\\mathrm { 7 a }$ we show the performance of a 3-layer CIN as a function of the ring size $k$ , along with that of GIN $\\overline { { [ 7 4 ] } }$ baselines equipped with $\\left\\lfloor { \\frac { k } { 2 } } \\right\\rfloor$ stacked layers. We observe that our model learns to solve the task with only 3 computational steps, independent of $k$ . As for GIN, we observed degradation in the performance for $k \\geq 2 4$ , up to complete failure. We hypothesise this to be due to the difficulties of training such a deep GNN $( \\geq 1 2$ layers). We further verify the (theoretically expected) failure of GIN (not included) when endowed with less than $\\left\\lfloor { \\frac { k } { 2 } } \\right\\rfloor$ layers. ",
+        "bbox": [
+            173,
+            545,
+            825,
+            728
+        ],
+        "page_idx": 7
+    },
+    {
+        "type": "text",
+        "text": "5.2 Real-World Graph Benchmarks ",
+        "text_level": 1,
+        "bbox": [
+            176,
+            746,
+            436,
+            761
+        ],
+        "page_idx": 7
+    },
+    {
+        "type": "text",
+        "text": "TUD We test our model on 8 TUDataset benchmarks $\\begin{array} { r l } {  { \\bigl \\| \\boldsymbol { 5 } 6 \\bigr \\| } } \\end{array}$ with small and medium sizes from biology (PROTEINS [9, $\\mathbb { 2 3 } \\mathbb { I } ,$ ), chemistry (i.e. molecules – MUTAG [45, 61], PTC, NCI1 and NCI109 $\\mathbb { H } \\mathbb { Q } \\mathbb { I } ,$ to social networks (IMDB-B, IMDB-M, RDT-B). We consider induced cycle of size up to $k = 6$ for our graph lifting procedure. We initialise node (and 0-cell) features as described in $\\mathrm { X u }$ et al. $\\textcircled { 7 4 }$ , and higher dimensional cells by averaging or summing the features of the included 0-cells. The training setting and evaluation procedure follow those in Xu et al. [74]. We report the results in Table 2. CIN compares more than favourably with the baselines, displaying strong empirical performance on all benchmarks. The mean accuracy of CIN ranks top on four out of eight datasets. ",
+        "bbox": [
+            174,
+            772,
+            826,
+            883
+        ],
+        "page_idx": 7
+    },
+    {
+        "type": "table",
+        "img_path": "images/4a1e49e9b880ffaed02a2f27a5a561dfa2a89ae6310f5da2ab3223cb7a13aab3.jpg",
+        "table_caption": [
+            "Table 2: TUDatasets. The first section of the table includes the accuracy of graph kernel methods, while the second includes GNNs. The top three are highlighted by First, Second, Third. "
+        ],
+        "table_footnote": [],
+        "table_body": "<table><tr><td>Dataset</td><td>MUTAG</td><td>PTC</td><td>PROTEINS</td><td>NCI1</td><td>NCI109</td><td>IMDB-B</td><td>IMDB-M</td><td>RDT-B</td></tr><tr><td>RWK 四</td><td>79.2±2.1</td><td>55.9±0.3</td><td>59.6±0.1</td><td>&gt;3 days</td><td>N/A</td><td>N/A</td><td>N/A</td><td>N/A</td></tr><tr><td>GK (k = 3) I 64</td><td>81.4±1.7</td><td>55.7±0.5</td><td>71.4±0.31</td><td>62.5±0.3</td><td>62.4±0.3</td><td>N/A</td><td>N/A</td><td>N/A</td></tr><tr><td>PK 58</td><td>76.0±2.7</td><td>59.5±2.4</td><td>73.7±0.7</td><td>82.5±0.5</td><td>N/A</td><td>N/A</td><td>N/A</td><td>N/A</td></tr><tr><td>WL kernel [65]</td><td>90.4±5.7</td><td>59.9±4.3</td><td>75.0±3.1</td><td>86.0±1.8</td><td>N/A</td><td>73.8±3.9</td><td>50.9±3.8</td><td>81.0±3.1</td></tr><tr><td>DCNN</td><td>N/A</td><td>N/A</td><td>61.3±1.6</td><td>56.6±1.0</td><td>N/A</td><td>49.1±1.4</td><td>33.5±1.4</td><td>N/A</td></tr><tr><td>DGCNN[76]</td><td>85.8±1.8</td><td>58.6±2.5</td><td>75.5±0.9</td><td>74.4±0.5</td><td>N/A</td><td>70.0±0.9</td><td>47.8±0.9</td><td>N/A</td></tr><tr><td>IGN[ 圆</td><td>83.9±13.0</td><td>58.5±6.9</td><td>76.6±5.5</td><td>74.3±2.7</td><td>72.8±1.5</td><td>72.0±5.5</td><td>48.7±3.4</td><td>N/A</td></tr><tr><td>GIN 园</td><td>89.4±5.6</td><td>64.6±7.0</td><td>76.2±2.8</td><td>82.7±1.7</td><td>N/A</td><td>75.1±5.1</td><td>52.3±2.8</td><td>92.4±2.5</td></tr><tr><td>PPGNs [531</td><td>90.6±8.7</td><td>66.2±6.6</td><td>77.2±4.7</td><td>83.2±1.1</td><td>82.2±1.4</td><td>73.0±5.8</td><td>50.5±3.6</td><td>N/A</td></tr><tr><td>Natural GN 自</td><td>89.4±1.6</td><td>66.8±1.7</td><td>71.7±1.0</td><td>82.4±1.3</td><td>N/A</td><td>73.5±2.0</td><td>51.3±1.5</td><td>N/A</td></tr><tr><td>GSN SIN</td><td>92.2 ± 7.5 N/A</td><td>68.2 ± 7.2 N/A</td><td>76.6± 5.0 76.4 ± 3.3</td><td>83.5 ± 2.0 82.7 ± 2.1</td><td>N/A</td><td>77.8 ± 3.3</td><td>54.3 ± 3.3</td><td>N/A</td></tr><tr><td></td><td></td><td></td><td></td><td></td><td>N/A</td><td>75.6± 3.2</td><td>52.4 ± 2.9</td><td>92.2 ± 1.0</td></tr><tr><td>CIN (Ours)</td><td>92.7 ± 6.1</td><td>68.2± 5.6</td><td>77.0 ± 4.3</td><td>83.6 ± 1.4</td><td>84.0 ± 1.6</td><td>75.6± 3.7</td><td>52.7 ± 3.1</td><td>92.4 ± 2.1</td></tr></table>",
+        "bbox": [
+            171,
+            126,
+            821,
+            295
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "table",
+        "img_path": "images/6f240739ac03f206556c984643be99722156cc3bfe46e31403ab0d94d939fec3.jpg",
+        "table_caption": [
+            "Table 3: ZINC (MAE), ZINC-FULL (MAE) and Mol-HIV (ROC-AUC). "
+        ],
+        "table_footnote": [],
+        "table_body": "<table><tr><td rowspan=\"2\">Method</td><td colspan=\"2\">ZINC↓</td><td rowspan=\"2\">ZINC-FULL↓ All methods</td><td rowspan=\"2\">MOLHIV↑ All methods</td></tr><tr><td>No Edge Feat.</td><td>With Edge Feat.</td></tr><tr><td>GCN 四</td><td>0.469±0.002</td><td>N/A</td><td>N/A</td><td>76.06±0.97</td></tr><tr><td>GAT 园</td><td>0.463±0.002</td><td>N/A</td><td>N/A</td><td>N/A</td></tr><tr><td>GatedGCN[ 自</td><td>0.422±0.006</td><td>0.363±0.009</td><td>N/A</td><td>N/A</td></tr><tr><td>GIN 四</td><td>0.408±0.008</td><td>0.252±0.014</td><td>0.088±0.002</td><td>77.07±1.49</td></tr><tr><td>PNA 四</td><td>0.320±0.032</td><td>0.188±0.004</td><td>N/A</td><td>79.05±1.32</td></tr><tr><td>DGN 回</td><td>0.219±0.010</td><td>0.168±0.003</td><td>N/A</td><td>79.70±0.97</td></tr><tr><td>HIMP 四</td><td>N/A</td><td>0.151±0.006</td><td>0.036±0.002</td><td>78.80±0.82</td></tr><tr><td>GSN 目</td><td>0.139±0.007</td><td>0.108±0.018</td><td>N/A</td><td>77.99±1.00</td></tr><tr><td>CIN-small (Ours)</td><td>0.139±0.008</td><td>0.094±0.004</td><td>0.044±0.003</td><td>80.55±1.04</td></tr><tr><td>CIN (Ours)</td><td>0.115±0.003</td><td>0.079±0.006</td><td>0.022±0.002</td><td>80.94±0.57</td></tr></table>",
+        "bbox": [
+            238,
+            344,
+            758,
+            507
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "On the remaining datasets, CIN achieves the second place. We observe that the best results are on datasets from the biological and chemical domains, where rings play a relevant role. ",
+        "bbox": [
+            176,
+            532,
+            823,
+            560
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "ZINC We study the effectiveness of cellular message passing on larger scale molecular benchmarks from the ZINC database $\\pmb { \\mathbb { E } } 8 \\|$ . ZINC (12k graphs) and ZINC-FULL (250k graphs) [24, 33, 43, 75] are two graph regression task datasets for drug constrained solubility prediction. In these experiments, we consider rings up to size $k = 1 8$ . We follow the training and evaluation procedures in [24]. Our experiments encompass different scenarios, examine the impact of ablating edge features and of constraining the parameter budget of the architecture to 100k. All results are illustrated in Table $\\textcircled { 3 }$ where we also include the results for ZINC-FULL obtained by the same exact architectures. Our model exhibits particularly strong performance on these benchmarks: it attains state-of-the-art results on both the two dataset variants, outperforming other models by a significant margin. CIN attains strong results even when constrained by the parameter budget. It still achieves state-of-the-art performance on ZINC and is on-par with the best unconstrained baseline under edge-feature ablation. ",
+        "bbox": [
+            173,
+            590,
+            825,
+            743
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "Mol-HIV We additionally test our model on the molecular ogbg-molhiv dataset from the Open Graph Benchmark $\\textcircled { | 4 0 | }$ (41k graphs). The task is to predict the capacity of compounds to inhibit HIV replication. Rings of size up to $k = 6$ are considered as 2-cells. We take the architecture in $ { \\mathbb { \\left[ \\left[ 2 7 \\right] \\right] } }$ as reference and replicate the same hyperparameter setting in our model, including the use of only 2 message passing layers. We report the mean of test ROC-AUC metrics at the epoch of best validation performance for 10 random weight initialisations. Similarly to ZINC, we experiment with a “small” model whose number of parameters is constrained in the order of $1 0 0 \\mathrm { k }$ . Table $3$ displays the results. CIN significantly outperforms other strong GNN baselines, even when constrained by the parameter budget. Consistently with $[ [ 2 7 ]$ , we observe that only two layers are sufficient when performing hierarchical message passing across meso-scale structures such as rings. ",
+        "bbox": [
+            173,
+            772,
+            826,
+            911
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "6 Related Work, Discussion and Conclusion ",
+        "text_level": 1,
+        "bbox": [
+            173,
+            88,
+            553,
+            106
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "Cell complex models Recent works have proposed the generalisation of GNNs to simplicial complexes [13, 25, 32, 35]. All these simplicial methods are subsumed by the model in Bodnar et al. $\\pmb { \\bigtriangledown }$ , which CWNs in turn subsume. To the best of our knowledge, Hajij et al. $\\pmb { \\Vert 3 4 \\Vert }$ is the only other example of message passing on cell complexes, but this work does not study the expressive power of the proposed scheme, neither it experimentally validates its performance. In contrast, our work comprehensively characterises the expressiveness of cellular message passing, and introduces a theoretically grounded and empirically effective framework to apply it on graph structured data in a way to address several limitations of standard Graph Neural Networks. ",
+        "bbox": [
+            174,
+            122,
+            825,
+            233
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "Molecular substructures A few other works have extended GNNs to account for molecular substructures. Junction Trees (JT), which conveniently represent singletons, bonds and rings as supernodes in a tree, have been used in molecular graph generation $\\mathbb { \\lVert 4 3 \\rVert \\overline { { \\lVert 4 4 \\rVert } } }$ . JTs are also used in the recent work of Fey et al. $ { \\mathbb { \\left[ \\left[ 2 7 \\right] \\right] } }$ , who employs them to design a hierarchical message passing scheme based on the tree structure. However, this hierarchy has a different configuration than the one cell complexes provide. Information about cycles is also used in GSNs $\\mathbb { m }$ to augment the node features, but the model retains the usual message passing procedure of GNNs. These last two models are of particular relevance to the present work, since they utilise information about chemical rings. It is important to remark that CWNs compare favourably with both of them in all our benchmarks. ",
+        "bbox": [
+            174,
+            251,
+            825,
+            376
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "Higher-order GNNs A related line of work has studied lifting graphs into $k$ -dimensional tensor representations that can be processed by provably expressive $k$ -GNNs [4, 52, 53]. With higher values of $k$ , these models achieve higher-expressivity, but due to the computational complexity this incurs, values of $k \\geq 3$ are of little use in practice. Therefore, unlike CWNs, these models cannot explicitly represent in practice chemical rings of common sizes (e.g. five or six). Furthermore, by being upper-bounded by 3-WL, the 2-GNN models cannot count the number of induced cycles of size greater than four (see Appendix $\\mathbf { A }$ for details). In contrast, CWNs can easily count these important chemical substructures through the readout operation it performs on the 2-cells. ",
+        "bbox": [
+            174,
+            393,
+            825,
+            505
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "Limitations The main limitations of the model are of computational nature. While the computational complexity of the message passing procedure and its preprocessing step is suitable for molecular and geometric graphs, the number of rings (and more generally simple cycles) in general graphs can be exponential in the number of nodes. In that case, one has to resort to smaller 2-cells like triangles, which can be found efficiently in general graphs. Moreover, one has to typically use weights specific for each dimension of the cell complex, increasing the number of parameters compared to GNNs. However, we have shown that our model can compensate this increase with a reduced number of layers and still achieve state-of-the-art results on some of the molecular benchmarks. ",
+        "bbox": [
+            174,
+            522,
+            825,
+            632
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "From a theoretical point of view, this work is concerned only with regular cell complexes. Adopting this restriction is useful from multiple perspectives: regular cell complexes are easier to analyse, their combinatorial structure completely describes their topology and convolutions can be defined on them through the Sheaf Laplacian (see Appendix D). Nonetheless, some of our theoretical results could be extended to non-regular complexes, which could be obtained by lifting transformations not studied in this work, such as attaching 2-cells to paths in the graph. We leave addressing non-regular complexes and their trade-offs to future developments of this work. ",
+        "bbox": [
+            174,
+            640,
+            825,
+            736
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "Societal Impacts Most of our paper is theoretical in nature and we do not see immediate direct negative societal impacts. Within the scope of social network applications, we do not yet have sufficient evidence of performance improvement on related benchmarks to justify obvious adoption in such a domain. In contrast, the empirical performance on molecular benchmarks suggests it may have a positive impact on applications of immediate interest in pharmaceutics, such as drug discovery $\\pmb { \\mathbb { B } } 0 \\|$ . ",
+        "bbox": [
+            174,
+            753,
+            825,
+            824
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "Conclusion We have proposed a provably powerful message passing procedure on cell complexes motivated by a novel colour refinement algorithm to test their isomorphism. This allows us to consider flexible lifting operations on graphs to implement more expressive architectures which benefit from decoupling the computational and input graphs. Our methods show excellent performance on diverse synthetic and real-world molecular benchmarks. ",
+        "bbox": [
+            176,
+            842,
+            823,
+            911
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "Funding and Acknowledgements ",
+        "text_level": 1,
+        "bbox": [
+            174,
+            89,
+            449,
+            107
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "YW and GM acknowledge support from the ERC under the EU’s Horizon 2020 programme (grant agreement $ { \\mathrm { n ^ { o } } }$ 757983). MB is supported in part by ERC Consolidator grant $\\boldsymbol { \\mathrm { n ^ { \\circ } } }$ 724228 (LEMAN). The authors declare no competing interests. We are also grateful to Ben Day, Gabriele Corso and Nikola Simidjievski for their helpful feedback. We would also like to thank Vijay P. Dwivedi and Chaitanya K. Joshi for clarifying certain aspects of their Benchmarking GNNs $\\pmb { \\bigtriangledown }$ work, and to Muhammet Balcilar for signalling a numerical precision issue in early SR graphs experiments. ",
+        "bbox": [
+            173,
+            119,
+            826,
+            204
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "References ",
+        "text_level": 1,
+        "bbox": [
+            174,
+            223,
+            266,
+            239
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "[1] Uri Alon and Eran Yahav. On the bottleneck of graph neural networks and its practical implications. ICLR, 2021.   \n[2] V. Arvind, Frank Fuhlbrück, Johannes Köbler, and Oleg Verbitsky. On weisfeiler-leman invariance: Subgraph counts and related graph properties. Journal of Computer and System Sciences, 113:42–59, 2020. ISSN 0022-0000.   \n[3] James Atwood and Don Towsley. Diffusion-convolutional neural networks. In NIPS, pages 1993–2001, 2016. [4] Waiss Azizian and Marc Lelarge. Expressive power of invariant and equivariant graph neural networks. In International Conference on Learning Representations, 2021. URL https: //openreview.net/forum?id=lxHgXYN4bwl.   \n[5] Jimmy Lei Ba, Jamie Ryan Kiros, and Geoffrey E. Hinton. Layer normalization. arXiv preprint arXiv:1607.06450, 2016.   \n[6] Dominique Beaini, Saro Passaro, Vincent Létourneau, William L Hamilton, Gabriele Corso, and Pietro Liò. Directional graph networks. ICML, 2021.   \n[7] Etienne E. Birmelé, Rui Ferreira, Roberto Grossi, Andrea Marino, Nadia Pisanti, Romeo Rizzi, and Gustavo Sacomoto. Optimal Listing of Cycles and st-Paths in Undirected Graphs. In 24th Annual ACM-SIAM Symposium on Discrete Algorithms, New Orleans, United States, January 2013.   \n[8] Cristian Bodnar, Fabrizio Frasca, Yu Guang Wang, Nina Otter, Guido Montúfar, Pietro Lio, and Michael Bronstein. Weisfeiler and lehman go topological: Message passing simplicial networks. ICML, 2021.   \n[9] Karsten M. Borgwardt, Cheng Soon Ong, Stefan Schönauer, S. V. N. Vishwanathan, Alex J. Smola, and Hans-Peter Kriegel. Protein function prediction via graph kernels. Bioinformatics, 21(suppl_1):i47–i56, 2005.   \n[10] Giorgos Bouritsas, Fabrizio Frasca, Stefanos Zafeiriou, and Michael M Bronstein. Improving graph neural network expressivity via subgraph isomorphism counting. arXiv:2006.09252, 2020.   \n[11] Xavier Bresson and Thomas Laurent. Residual gated graph convnets. arXiv preprint arXiv:1711.07553, 2017.   \n[12] Michael M Bronstein, Joan Bruna, Yann LeCun, Arthur Szlam, and Pierre Vandergheynst. Geometric deep learning: going beyond euclidean data. IEEE Signal Processing Magazine, 34 (4):18–42, 2017.   \n[13] Eric Bunch, Qian You, Glenn Fung, and Vikas Singh. Simplicial 2-complex convolutional neural networks. In NeurIPS Workshop on Topological Data Analysis and Beyond, 2020.   \n[14] Zhengdao Chen, Soledad Villar, Lei Chen, and Joan Bruna. On the equivalence between graph isomorphism testing and function approximation with gnns. In NeurIPS, volume 32, 2019.   \n[15] Zhengdao Chen, Lei Chen, Soledad Villar, and Joan Bruna. Can graph neural networks count substructures? In NeurIPS, 2020.   \n[16] N. Chiba and Takao Nishizeki. Arboricity and subgraph listing algorithms. SIAM J. Comput., 14:210–223, 1985.   \n[17] Fan Chung. Spectral graph theory. American Mathematical Society, 1997.   \n[18] Djork-Arné Clevert, Thomas Unterthiner, and Sepp Hochreiter. Fast and accurate deep network learning by exponential linear units (ELUs). In ICLR, 2016.   \n[19] Luigi P. Cordella, Pasquale Foggia, Carlo Sansone, and Mario Vento. A (sub)graph isomorphism algorithm for matching large graphs. IEEE Transactions on Pattern Analysis and Machine Intelligence, 26(10):1367–1372, 2004.   \n[20] Gabriele Corso, Luca Cavalleri, Dominique Beaini, Pietro Liò, and Petar Velickovi ˇ c. Principal ´ neighbourhood aggregation for graph nets. In NeurIPS, volume 33, pages 13260–13271, 2020.   \n[21] Pim de Haan, Taco Cohen, and Max Welling. Natural graph networks. In NeurIPS, 2020.   \n[22] Michaël Defferrard, Xavier Bresson, and Pierre Vandergheynst. Convolutional neural networks on graphs with fast localized spectral filtering. In NIPS, 2016.   \n[23] Paul D Dobson and Andrew J Doig. Distinguishing enzyme structures from non-enzymes without alignments. Journal of Molecular Biology, 330(4):771–783, 2003.   \n[24] Vijay Prakash Dwivedi, Chaitanya K Joshi, Thomas Laurent, Yoshua Bengio, and Xavier Bresson. Benchmarking graph neural networks. arXiv preprint arXiv:2003.00982, 2020.   \n[25] Stefania Ebli, Michaël Defferrard, and Gard Spreemann. Simplicial neural networks. In NeurIPS Workshop on Topological Data Analysis and Beyond, 2020.   \n[26] Rui Ferreira, Roberto Grossi, Romeo Rizzi, Gustavo Sacomoto, and Marie-France Sagot. Amortized Õ(|V|) -Delay Algorithm for Listing Chordless Cycles in Undirected Graphs. In $2 2 t h$ Annual European Symposium on Algorithms, volume 8737 of Algorithms ESA 2014, Lecture Notes in Computer Science, pages 418–429, Wroclaw, Poland, September 2014.   \n[27] M. Fey, J. G. Yuen, and F. Weichert. Hierarchical inter-message passing for learning on molecular graphs. In ICML Graph Representation Learning and Beyond $( G R L + ,$ Workhop, 2020.   \n[28] Matthias Fey and Jan Eric Lenssen. Fast graph representation learning with PyTorch Geometric. In ICLR Workshop on Representation Learning on Graphs and Manifolds, 2019.   \n[29] Thomas Gärtner, Peter Flach, and Stefan Wrobel. On graph kernels: Hardness results and efficient alternatives. In Learning theory and kernel machines, pages 129–143. Springer, 2003.   \n[30] Thomas Gaudelet, Ben Day, Arian R Jamasb, Jyothish Soman, Cristian Regep, Gertrude Liu, Jeremy B R Hayter, Richard Vickers, Charles Roberts, Jian Tang, David Roblin, Tom L Blundell, Michael M Bronstein, and Jake P Taylor-King. Utilizing graph machine learning within drug discovery and development. Briefings in Bioinformatics, 05 2021. ISSN 1477-4054.   \n[31] Justin Gilmer, Samuel S Schoenholz, Patrick F Riley, Oriol Vinyals, and George E Dahl. Neural message passing for quantum chemistry. In ICML, 2017.   \n[32] Nicholas Glaze, T. Mitchell Roddenberry, and Santiago Segarra. Principled simplicial neural networks for trajectory prediction. In ICML, 2021.   \n[33] Rafael Gómez-Bombarelli, Jennifer N Wei, David Duvenaud, José Miguel Hernández-Lobato, Benjamín Sánchez-Lengeling, Dennis Sheberla, Jorge Aguilera-Iparraguirre, Timothy D Hirzel, Ryan P Adams, and Alán Aspuru-Guzik. Automatic chemical design using a data-driven continuous representation of molecules. ACS Central Science, 4(2):268–276, 2018.   \n[34] Mustafa Hajij, Kyle Istvan, and Ghada Zamzmi. Cell complex neural networks. In NeurIPS Workshop on Topological Data Analysis and Beyond, 2020.   \n[35] Mustafa Hajij, Ghada Zamzmi, and Xuanting Cai. Simplicial complex representation learning. arXiv preprint arXiv:2103.04046, 2021.   \n[36] J. Hansen and R. Ghrist. Toward a spectral theory of cellular sheaves. Journal of Applied and Computational Topology, 3:315–358, 2019.   \n[37] Jakob Hansen and Thomas Gebhart. Sheaf neural networks. In NeurIPS 2020 Workshop on Topological Data Analysis and Beyond, 2020.   \n[38] Allen Hatcher. Algebraic topology. Cambridge Univ. Press, Cambridge, 2000.   \n[39] Christoph Helma, Ross D. King, Stefan Kramer, and Ashwin Srinivasan. The predictive toxicology challenge 2000–2001. Bioinformatics, 17(1):107–108, 2001.   \n[40] Weihua Hu, Matthias Fey, Marinka Zitnik, Yuxiao Dong, Hongyu Ren, Bowen Liu, Michele Catasta, and Jure Leskovec. Open graph benchmark: Datasets for machine learning on graphs. In NeurIPS, 2020.   \n[41] Weihua $\\mathrm { H u ^ { * } }$ , Bowen Liu\\*, Joseph Gomes, Marinka Zitnik, Percy Liang, Vijay Pande, and Jure Leskovec. Strategies for pre-training graph neural networks. In International Conference on Learning Representations, 2020.   \n[42] Sergey Ioffe and Christian Szegedy. Batch normalization: Accelerating deep network training by reducing internal covariate shift. In ICML, 2015.   \n[43] Wengong Jin, Regina Barzilay, and Tommi Jaakkola. Junction tree variational autoencoder for molecular graph generation. In ICML, pages 2323–2332, 2018.   \n[44] Wengong Jin, Kevin Yang, Regina Barzilay, and Tommi Jaakkola. Learning multimodal graph-to-graph translation for molecular optimization. In ICLR, 2019.   \n[45] Jeroen Kazius, Ross McGuire, and Roberta Bursi. Derivation and validation of toxicophores for mutagenicity prediction. Journal of Medicinal Chemistry, 48(1):312–320, 2005.   \n[46] Diederik P Kingma and Jimmy Ba. Adam: A method for stochastic optimization. In ICLR, 2015.   \n[47] Thomas N Kipf and Max Welling. Semi-supervised classification with graph convolutional networks. In ICLR, 2017.   \n[48] Nils Kriege and Petra Mutzel. Subgraph matching kernels for attributed graphs. arXiv preprint arXiv:1206.6483, 2012.   \n[49] Casimir Kuratowski. Sur le probleme des courbes gauches en topologie. Fundamenta mathematicae, 15(1):271–283, 1930.   \n[50] Greg Landrum. RDKit: Open-source cheminformatics software. 2016.   \n[51] Qimai Li, Zhichao Han, and Xiao-Ming Wu. Deeper insights into graph convolutional networks for semi-supervised learning. In AAAI, volume 32, 2018.   \n[52] Haggai Maron, Heli Ben-Hamu, Nadav Shamir, and Yaron Lipman. Invariant and equivariant graph networks. In ICLR, 2018.   \n[53] Haggai Maron, Heli Ben-Hamu, Hadar Serviansky, and Yaron Lipman. Provably powerful graph networks. In NeurIPS, 2019.   \n[54] Federico Monti, Davide Boscaini, Jonathan Masci, Emanuele Rodola, Jan Svoboda, and Michael M Bronstein. Geometric deep learning on graphs and manifolds using mixture model CNNs. In CVPR, pages 5425–5434, 2017.   \n[55] Christopher Morris, Martin Ritzert, Matthias Fey, William L Hamilton, Jan Eric Lenssen, Gaurav Rattan, and Martin Grohe. Weisfeiler and Leman go neural: Higher-order graph neural networks. In AAAI, 2019.   \n[56] Christopher Morris, Nils M Kriege, Franka Bause, Kristian Kersting, Petra Mutzel, and Marion Neumann. TUDataset: A collection of benchmark datasets for learning with graphs. In ICML Graph Representation Learning and Beyond ${ ' } G R L { + } ,$ ) Workhop, 2020. ",
+        "bbox": [
+            173,
+            247,
+            828,
+            915
+        ],
+        "page_idx": 10
+    },
+    {
+        "type": "text",
+        "text": "",
+        "bbox": [
+            169,
+            37,
+            828,
+            917
+        ],
+        "page_idx": 11
+    },
+    {
+        "type": "text",
+        "text": "",
+        "bbox": [
+            169,
+            56,
+            828,
+            921
+        ],
+        "page_idx": 12
+    },
+    {
+        "type": "text",
+        "text": "[57] Ryan Murphy, Balasubramaniam Srinivasan, Vinayak Rao, and Bruno Ribeiro. Relational pooling for graph representations. In ICML, volume 97, pages 4663–4673, 2019. ",
+        "bbox": [
+            171,
+            90,
+            825,
+            121
+        ],
+        "page_idx": 13
+    },
+    {
+        "type": "text",
+        "text": "[58] Marion Neumann, Roman Garnett, Christian Bauckhage, and Kristian Kersting. Propagation kernels: efficient graph kernels from propagated information. Machine Learning, 102(2): 209–245, 2016. ",
+        "bbox": [
+            173,
+            128,
+            823,
+            171
+        ],
+        "page_idx": 13
+    },
+    {
+        "type": "text",
+        "text": "[59] Adam Paszke, Sam Gross, Francisco Massa, Adam Lerer, James Bradbury, Gregory Chanan, Trevor Killeen, Zeming Lin, Natalia Gimelshein, Luca Antiga, Alban Desmaison, Andreas Kopf, Edward Yang, Zachary DeVito, Martin Raison, Alykhan Tejani, Sasank Chilamkurthy, Benoit Steiner, Lu Fang, Junjie Bai, and Soumith Chintala. PyTorch: An imperative style, high-performance deep learning library. In NeurIPS, pages 8024–8035. 2019. ",
+        "bbox": [
+            173,
+            180,
+            823,
+            251
+        ],
+        "page_idx": 13
+    },
+    {
+        "type": "text",
+        "text": "[60] Raghunathan Ramakrishnan, Pavlo O Dral, Matthias Rupp, and O Anatole Von Lilienfeld. Quantum chemistry structures and properties of 134 kilo molecules. Scientific Data, 1(1):1–7, 2014. ",
+        "bbox": [
+            173,
+            258,
+            823,
+            301
+        ],
+        "page_idx": 13
+    },
+    {
+        "type": "text",
+        "text": "[61] Kaspar Riesen and Horst Bunke. Iam graph database repository for graph based pattern recognition and machine learning. In Joint IAPR International Workshops on Statistical Techniques in Pattern Recognition (SPR) and Structural and Syntactic Pattern Recognition (SSPR), pages 287–297. Springer, 2008. ",
+        "bbox": [
+            173,
+            310,
+            826,
+            367
+        ],
+        "page_idx": 13
+    },
+    {
+        "type": "text",
+        "text": "[62] Ryoma Sato. A survey on the expressive power of graph neural networks. arXiv preprint arXiv:2003.04078, 2020. ",
+        "bbox": [
+            173,
+            376,
+            823,
+            405
+        ],
+        "page_idx": 13
+    },
+    {
+        "type": "text",
+        "text": "[63] Michael T Schaub, Austin R Benson, Paul Horn, Gabor Lippner, and Ali Jadbabaie. Random walks on simplicial complexes and the normalized Hodge 1-Laplacian. SIAM Review, 62(2): 353–391, 2020. ",
+        "bbox": [
+            174,
+            412,
+            823,
+            455
+        ],
+        "page_idx": 13
+    },
+    {
+        "type": "text",
+        "text": "[64] Nino Shervashidze, SVN Vishwanathan, Tobias Petri, Kurt Mehlhorn, and Karsten Borgwardt. Efficient graphlet kernels for large graph comparison. In AISTAT, pages 488–495, 2009. ",
+        "bbox": [
+            171,
+            464,
+            825,
+            494
+        ],
+        "page_idx": 13
+    },
+    {
+        "type": "text",
+        "text": "[65] Nino Shervashidze, Pascal Schweitzer, Erik Jan van Leeuwen, Kurt Mehlhorn, and Karsten M Borgwardt. Weisfeiler-lehman graph kernels. JMLR, 12(Sep):2539–2561, 2011. ",
+        "bbox": [
+            173,
+            502,
+            823,
+            531
+        ],
+        "page_idx": 13
+    },
+    {
+        "type": "text",
+        "text": "[66] Howard E. Simmons and John E. Maggio. Synthesis of the first topologically non-planar molecule. Tetrahedron Letters, 22(4):287–290, 1981. ISSN 0040-4039. ",
+        "bbox": [
+            173,
+            540,
+            823,
+            569
+        ],
+        "page_idx": 13
+    },
+    {
+        "type": "text",
+        "text": "[67] Amit Singer and Hau-Tieng Wu. Vector diffusion maps and the connection laplacian. Communications in Pure and Applied Mathematics, 68(8), 2012. ",
+        "bbox": [
+            171,
+            578,
+            825,
+            607
+        ],
+        "page_idx": 13
+    },
+    {
+        "type": "text",
+        "text": "[68] Teague Sterling and John J. Irwin. ZINC 15 – ligand discovery for everyone. Journal of Chemical Information and Modeling, 55(11):2324–2337, 11 2015. doi: 10.1021/acs.jcim.5b00559. URL https://doi.org/10.1021/acs.jcim.5b00559. ",
+        "bbox": [
+            173,
+            614,
+            825,
+            659
+        ],
+        "page_idx": 13
+    },
+    {
+        "type": "text",
+        "text": "[69] Petar Velickovi ˇ c, Guillem Cucurull, Arantxa Casanova, Adriana Romero, Pietro Lio, and Yoshua ´ Bengio. Graph attention networks. In ICLR, 2018. ",
+        "bbox": [
+            169,
+            666,
+            825,
+            695
+        ],
+        "page_idx": 13
+    },
+    {
+        "type": "text",
+        "text": "[70] Nikil Wale, Ian A Watson, and George Karypis. Comparison of descriptor spaces for chemical compound retrieval and classification. Knowledge and Information Systems, 14(3):347–375, 2008. ",
+        "bbox": [
+            174,
+            704,
+            823,
+            746
+        ],
+        "page_idx": 13
+    },
+    {
+        "type": "text",
+        "text": "[71] Boris Weisfeiler and Andrei Leman. The reduction of a graph to canonical form and the algebra which appears therein. NTI Series, 2(9):12–16, 1968. ",
+        "bbox": [
+            169,
+            756,
+            825,
+            785
+        ],
+        "page_idx": 13
+    },
+    {
+        "type": "text",
+        "text": "[72] J. H. C. Whitehead. Combinatorial homotopy. I. Bulletin of the American Mathematical Society, 55(3.P1):213 – 245, 1949. doi: bams/1183513543. ",
+        "bbox": [
+            171,
+            794,
+            825,
+            823
+        ],
+        "page_idx": 13
+    },
+    {
+        "type": "text",
+        "text": "[73] Keyulu Xu, Chengtao Li, Yonglong Tian, Tomohiro Sonobe, Ken ichi Kawarabayashi, and Stefanie Jegelka. Representation learning on graphs with jumping knowledge networks. In ICML, 2018. ",
+        "bbox": [
+            174,
+            832,
+            821,
+            873
+        ],
+        "page_idx": 13
+    },
+    {
+        "type": "text",
+        "text": "[74] Keyulu Xu, Weihua Hu, Jure Leskovec, and Stefanie Jegelka. How powerful are graph neural networks? In ICLR, 2019. ",
+        "bbox": [
+            173,
+            882,
+            820,
+            911
+        ],
+        "page_idx": 13
+    },
+    {
+        "type": "text",
+        "text": "[75] Jiaxuan You, Bowen Liu, Rex Ying, Vijay Pande, and Jure Leskovec. Graph convolutional policy network for goal-directed molecular graph generation. NeurIPS, 2018.   \n[76] Muhan Zhang, Zhicheng Cui, Marion Neumann, and Yixin Chen. An end-to-end deep learning architecture for graph classification. In AAAI, 2018. ",
+        "bbox": [
+            169,
+            90,
+            826,
+            157
+        ],
+        "page_idx": 14
+    },
+    {
+        "type": "text",
+        "text": "Checklist ",
+        "text_level": 1,
+        "bbox": [
+            174,
+            89,
+            254,
+            106
+        ],
+        "page_idx": 15
+    },
+    {
+        "type": "text",
+        "text": "1. For all authors... ",
+        "bbox": [
+            214,
+            116,
+            339,
+            131
+        ],
+        "page_idx": 15
+    },
+    {
+        "type": "text",
+        "text": "(a) Do the main claims made in the abstract and introduction accurately reflect the paper’s contributions and scope? [Yes]   \n(b) Did you describe the limitations of your work? [Yes] See Section 6.   \n(c) Did you discuss any potential negative societal impacts of your work? [Yes] See Section 6.   \n(d) Have you read the ethics review guidelines and ensured that your paper conforms to them? [Yes] ",
+        "bbox": [
+            238,
+            136,
+            825,
+            239
+        ],
+        "page_idx": 15
+    },
+    {
+        "type": "text",
+        "text": "2. If you are including theoretical results... ",
+        "bbox": [
+            214,
+            243,
+            493,
+            258
+        ],
+        "page_idx": 15
+    },
+    {
+        "type": "text",
+        "text": "(a) Did you state the full set of assumptions of all theoretical results? [Yes] (b) Did you include complete proofs of all theoretical results? [Yes] ",
+        "bbox": [
+            238,
+            262,
+            735,
+            294
+        ],
+        "page_idx": 15
+    },
+    {
+        "type": "text",
+        "text": "3. If you ran experiments... ",
+        "bbox": [
+            212,
+            299,
+            393,
+            313
+        ],
+        "page_idx": 15
+    },
+    {
+        "type": "text",
+        "text": "(a) Did you include the code, data, and instructions needed to reproduce the main experimental results (either in the supplemental material or as a URL)? [Yes] We provided a link to our code in the paper.   \n(b) Did you specify all the training details (e.g., data splits, hyperparameters, how they were chosen)? [Yes] Details are provided in Section 5 and Appendix E.   \n(c) Did you report error bars (e.g., with respect to the random seed after running experiments multiple times)? [Yes]   \n(d) Did you include the total amount of compute and the type of resources used (e.g., type of GPUs, internal cluster, or cloud provider)? [Yes] See Section 5 and Appendix E. ",
+        "bbox": [
+            238,
+            316,
+            825,
+            449
+        ],
+        "page_idx": 15
+    },
+    {
+        "type": "text",
+        "text": "4. If you are using existing assets (e.g., code, data, models) or curating/releasing new assets... ",
+        "bbox": [
+            217,
+            453,
+            821,
+            468
+        ],
+        "page_idx": 15
+    },
+    {
+        "type": "text",
+        "text": "(a) If your work uses existing assets, did you cite the creators? [Yes] We used multiple existing datasets described and referenced in Section 5.   \n(b) Did you mention the license of the assets? [Yes] See Appendix E.   \n(c) Did you include any new assets either in the supplemental material or as a URL? [Yes] We included a link to our code.   \n(d) Did you discuss whether and how consent was obtained from people whose data you’re using/curating? [N/A] We are only using publicly available datasets.   \n(e) Did you discuss whether the data you are using/curating contains personally identifiable information or offensive content? [N/A] None of the datasets we are using contain personally identifiable information or offensive content. ",
+        "bbox": [
+            238,
+            472,
+            825,
+            621
+        ],
+        "page_idx": 15
+    },
+    {
+        "type": "text",
+        "text": "5. If you used crowdsourcing or conducted research with human subjects... ",
+        "bbox": [
+            214,
+            625,
+            705,
+            638
+        ],
+        "page_idx": 15
+    },
+    {
+        "type": "text",
+        "text": "(a) Did you include the full text of instructions given to participants and screenshots, if applicable? [N/A]   \n(b) Did you describe any potential participant risks, with links to Institutional Review Board (IRB) approvals, if applicable? [N/A]   \n(c) Did you include the estimated hourly wage paid to participants and the total amount spent on participant compensation? [N/A] ",
+        "bbox": [
+            238,
+            643,
+            825,
+            733
+        ],
+        "page_idx": 15
+    }
+]
\ No newline at end of file
diff --git a/parse/train/uVPZCMVtsSG/uVPZCMVtsSG_middle.json b/parse/train/uVPZCMVtsSG/uVPZCMVtsSG_middle.json
new file mode 100644
index 0000000000000000000000000000000000000000..e661d0815a1836688f7b2e51c07c52677dbb5480
--- /dev/null
+++ b/parse/train/uVPZCMVtsSG/uVPZCMVtsSG_middle.json
@@ -0,0 +1,48176 @@
+{
+    "pdf_info": [
+        {
+            "preproc_blocks": [
+                {
+                    "type": "title",
+                    "bbox": [
+                        114,
+                        97,
+                        495,
+                        117
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                114,
+                                96,
+                                496,
+                                119
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        114,
+                                        96,
+                                        496,
+                                        119
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Weisfeiler and Lehman Go Cellular: CW Networks",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        }
+                    ],
+                    "index": 0
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        156,
+                        156,
+                        258,
+                        189
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                171,
+                                156,
+                                246,
+                                167
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        171,
+                                        156,
+                                        246,
+                                        167
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Cristian Bodnar⇤",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                155,
+                                166,
+                                258,
+                                180
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        155,
+                                        166,
+                                        258,
+                                        180
+                                    ],
+                                    "score": 1.0,
+                                    "content": "University of Cambridge",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                164,
+                                178,
+                                250,
+                                189
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        164,
+                                        178,
+                                        250,
+                                        189
+                                    ],
+                                    "score": 1.0,
+                                    "content": "cb2015@cam.ac.uk",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        }
+                    ],
+                    "index": 3
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        310,
+                        156,
+                        455,
+                        189
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                345,
+                                156,
+                                419,
+                                167
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        345,
+                                        156,
+                                        419,
+                                        167
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Fabrizio Frasca⇤",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                311,
+                                168,
+                                455,
+                                179
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        311,
+                                        168,
+                                        455,
+                                        179
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Imperial College London & Twitter",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                332,
+                                178,
+                                434,
+                                189
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        332,
+                                        178,
+                                        434,
+                                        189
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ffrasca@twitter.com",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        }
+                    ],
+                    "index": 4
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        161,
+                        207,
+                        263,
+                        239
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                187,
+                                205,
+                                238,
+                                218
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        187,
+                                        205,
+                                        238,
+                                        218
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Nina Otter",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                196,
+                                217,
+                                227,
+                                227
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        196,
+                                        217,
+                                        227,
+                                        227
+                                    ],
+                                    "score": 1.0,
+                                    "content": "UCLA",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                162,
+                                229,
+                                263,
+                                239
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        162,
+                                        229,
+                                        263,
+                                        239
+                                    ],
+                                    "score": 1.0,
+                                    "content": "otter@math.ucla.edu",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        }
+                    ],
+                    "index": 9
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        321,
+                        206,
+                        450,
+                        240
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                349,
+                                205,
+                                424,
+                                219
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        349,
+                                        205,
+                                        424,
+                                        219
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Yu Guang Wang",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                330,
+                                217,
+                                440,
+                                229
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        330,
+                                        217,
+                                        440,
+                                        229
+                                    ],
+                                    "score": 1.0,
+                                    "content": "MPI-MIS, SJTU & UNSW",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                320,
+                                229,
+                                451,
+                                241
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        320,
+                                        229,
+                                        451,
+                                        241
+                                    ],
+                                    "score": 1.0,
+                                    "content": "yuguang.wang@unsw.edu.au",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        }
+                    ],
+                    "index": 10
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        113,
+                        256,
+                        214,
+                        289
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                141,
+                                255,
+                                187,
+                                267
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        255,
+                                        187,
+                                        267
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Pietro Liò",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                113,
+                                266,
+                                216,
+                                280
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        113,
+                                        266,
+                                        216,
+                                        280
+                                    ],
+                                    "score": 1.0,
+                                    "content": "University of Cambridge",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                122,
+                                278,
+                                205,
+                                290
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        122,
+                                        278,
+                                        205,
+                                        290
+                                    ],
+                                    "score": 1.0,
+                                    "content": "pl219@cam.ac.uk",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        }
+                    ],
+                    "index": 16
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        225,
+                        256,
+                        343,
+                        289
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                248,
+                                256,
+                                322,
+                                267
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        248,
+                                        256,
+                                        322,
+                                        267
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Guido Montúfar",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                244,
+                                267,
+                                324,
+                                277
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        244,
+                                        267,
+                                        324,
+                                        277
+                                    ],
+                                    "score": 1.0,
+                                    "content": "MPI-MIS & UCLA",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                225,
+                                279,
+                                343,
+                                289
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        225,
+                                        279,
+                                        343,
+                                        289
+                                    ],
+                                    "score": 1.0,
+                                    "content": "montufar@math.ucla.edu",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        }
+                    ],
+                    "index": 17
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        353,
+                        256,
+                        497,
+                        289
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                385,
+                                256,
+                                466,
+                                268
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        385,
+                                        256,
+                                        466,
+                                        268
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Michael Bronstein",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                354,
+                                268,
+                                498,
+                                279
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        354,
+                                        268,
+                                        498,
+                                        279
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Imperial College London & Twitter",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                366,
+                                277,
+                                484,
+                                290
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        366,
+                                        277,
+                                        484,
+                                        290
+                                    ],
+                                    "score": 1.0,
+                                    "content": "mbronstein@twitter.com",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        }
+                    ],
+                    "index": 18
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        283,
+                        318,
+                        328,
+                        331
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                281,
+                                317,
+                                330,
+                                333
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        281,
+                                        317,
+                                        330,
+                                        333
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Abstract",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        }
+                    ],
+                    "index": 22
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        142,
+                        343,
+                        469,
+                        539
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                142,
+                                342,
+                                469,
+                                356
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        142,
+                                        342,
+                                        469,
+                                        356
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Graph Neural Networks (GNNs) are limited in their expressive power, struggle with",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                354,
+                                470,
+                                367
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        354,
+                                        470,
+                                        367
+                                    ],
+                                    "score": 1.0,
+                                    "content": "long-range interactions and lack a principled way to model higher-order structures.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                364,
+                                470,
+                                378
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        364,
+                                        470,
+                                        378
+                                    ],
+                                    "score": 1.0,
+                                    "content": "These problems can be attributed to the strong coupling between the computational",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                375,
+                                470,
+                                389
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        375,
+                                        470,
+                                        389
+                                    ],
+                                    "score": 1.0,
+                                    "content": "graph and the input graph structure. The recently proposed Message Passing",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                387,
+                                469,
+                                399
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        387,
+                                        469,
+                                        399
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Simplicial Networks naturally decouple these elements by performing message",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                398,
+                                470,
+                                410
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        398,
+                                        470,
+                                        410
+                                    ],
+                                    "score": 1.0,
+                                    "content": "passing on the clique complex of the graph. Nevertheless, these models can be",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                408,
+                                469,
+                                420
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        408,
+                                        469,
+                                        420
+                                    ],
+                                    "score": 1.0,
+                                    "content": "severely constrained by the rigid combinatorial structure of Simplicial Complexes",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                420,
+                                469,
+                                432
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        420,
+                                        469,
+                                        432
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(SCs). In this work, we extend recent theoretical results on SCs to regular Cell",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                430,
+                                470,
+                                443
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        430,
+                                        470,
+                                        443
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Complexes, topological objects that flexibly subsume SCs and graphs. We show",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                441,
+                                471,
+                                454
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        441,
+                                        471,
+                                        454
+                                    ],
+                                    "score": 1.0,
+                                    "content": "that this generalisation provides a powerful set of graph “lifting” transformations,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                451,
+                                470,
+                                465
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        451,
+                                        470,
+                                        465
+                                    ],
+                                    "score": 1.0,
+                                    "content": "each leading to a unique hierarchical message passing procedure. The resulting",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                462,
+                                469,
+                                475
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        462,
+                                        469,
+                                        475
+                                    ],
+                                    "score": 1.0,
+                                    "content": "methods, which we collectively call CW Networks (CWNs), are strictly more",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                474,
+                                470,
+                                487
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        474,
+                                        470,
+                                        487
+                                    ],
+                                    "score": 1.0,
+                                    "content": "powerful than the WL test and not less powerful than the 3-WL test. In particular,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                484,
+                                470,
+                                498
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        484,
+                                        470,
+                                        498
+                                    ],
+                                    "score": 1.0,
+                                    "content": "we demonstrate the effectiveness of one such scheme, based on rings, when applied",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                496,
+                                469,
+                                508
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        496,
+                                        469,
+                                        508
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to molecular graph problems. The proposed architecture benefits from provably",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                507,
+                                470,
+                                520
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        507,
+                                        470,
+                                        520
+                                    ],
+                                    "score": 1.0,
+                                    "content": "larger expressivity than commonly used GNNs, principled modelling of higher-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                518,
+                                469,
+                                529
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        518,
+                                        469,
+                                        529
+                                    ],
+                                    "score": 1.0,
+                                    "content": "order signals and from compressing the distances between nodes. We demonstrate",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                528,
+                                466,
+                                541
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        528,
+                                        466,
+                                        541
+                                    ],
+                                    "score": 1.0,
+                                    "content": "that our model achieves state-of-the-art results on a variety of molecular datasets.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        }
+                    ],
+                    "index": 31.5
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        560,
+                        190,
+                        574
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                559,
+                                192,
+                                576
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        559,
+                                        192,
+                                        576
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1 Introduction",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        }
+                    ],
+                    "index": 41
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        585,
+                        505,
+                        695
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                585,
+                                505,
+                                597
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        585,
+                                        505,
+                                        597
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The operations performed by message passing Graph Neural Networks (GNNs) emulate the structure",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                596,
+                                506,
+                                608
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        596,
+                                        506,
+                                        608
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of the input graph. While this property has clear computational advantages, it brings with it a series of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                606,
+                                504,
+                                619
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        607,
+                                        303,
+                                        619
+                                    ],
+                                    "score": 1.0,
+                                    "content": "fundamental limitations. As observed by Xu et al.",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        303,
+                                        606,
+                                        321,
+                                        618
+                                    ],
+                                    "score": 0.6,
+                                    "content": "\\dot { \\lVert \\mathcal { I } \\dot { 4 } \\rVert }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        322,
+                                        607,
+                                        504,
+                                        619
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and Morris et al. [55] the local neighbourhood",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                617,
+                                504,
+                                631
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        617,
+                                        475,
+                                        631
+                                    ],
+                                    "score": 1.0,
+                                    "content": "aggregations used by GNNs are at most as powerful as the Weisfeiler-Lehman (WL) test",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        475,
+                                        617,
+                                        493,
+                                        629
+                                    ],
+                                    "score": 0.48,
+                                    "content": "\\pmb { \\mathbb { Z } } \\mathbf { \\mathbb { 1 } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        493,
+                                        617,
+                                        504,
+                                        631
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                628,
+                                505,
+                                641
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        628,
+                                        505,
+                                        641
+                                    ],
+                                    "score": 1.0,
+                                    "content": "distinguishing non-isomorphic graphs. Therefore, GNNs fail to detect certain higer-order meso-scale",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                639,
+                                505,
+                                653
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        639,
+                                        505,
+                                        653
+                                    ],
+                                    "score": 1.0,
+                                    "content": "structures such as cliques or (induced) cycles [2, 15], which are particularly important in applications",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 47
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                650,
+                                506,
+                                663
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        650,
+                                        506,
+                                        663
+                                    ],
+                                    "score": 1.0,
+                                    "content": "dealing with social and biological networks or molecular graphs. At the same time, many such layers",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 48
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                662,
+                                505,
+                                674
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        662,
+                                        505,
+                                        674
+                                    ],
+                                    "score": 1.0,
+                                    "content": "have to be stacked to make long-range interactions in the graph possible. Besides the computational",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 49
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                673,
+                                503,
+                                684
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        673,
+                                        503,
+                                        684
+                                    ],
+                                    "score": 1.0,
+                                    "content": "burden this incurs, deep GNNs typically come with additional problems such as over-smoothing [51]",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 50
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                684,
+                                312,
+                                695
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        684,
+                                        312,
+                                        695
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and over-squashing [1] of the node representations.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 51
+                        }
+                    ],
+                    "index": 46.5
+                }
+            ],
+            "page_idx": 0,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        106,
+                        731,
+                        385,
+                        742
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                730,
+                                386,
+                                743
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        730,
+                                        386,
+                                        743
+                                    ],
+                                    "score": 1.0,
+                                    "content": "35th Conference on Neural Information Processing Systems (NeurIPS 2021).",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        119,
+                        703,
+                        226,
+                        712
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                118,
+                                700,
+                                227,
+                                714
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        118,
+                                        700,
+                                        227,
+                                        714
+                                    ],
+                                    "score": 1.0,
+                                    "content": "⇤Authors contributed equally.",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "title",
+                    "bbox": [
+                        114,
+                        97,
+                        495,
+                        117
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                114,
+                                96,
+                                496,
+                                119
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        114,
+                                        96,
+                                        496,
+                                        119
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Weisfeiler and Lehman Go Cellular: CW Networks",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        }
+                    ],
+                    "index": 0
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        156,
+                        156,
+                        258,
+                        189
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                171,
+                                156,
+                                246,
+                                167
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        171,
+                                        156,
+                                        246,
+                                        167
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Cristian Bodnar⇤",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                155,
+                                166,
+                                258,
+                                180
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        155,
+                                        166,
+                                        258,
+                                        180
+                                    ],
+                                    "score": 1.0,
+                                    "content": "University of Cambridge",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                164,
+                                178,
+                                250,
+                                189
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        164,
+                                        178,
+                                        250,
+                                        189
+                                    ],
+                                    "score": 1.0,
+                                    "content": "cb2015@cam.ac.uk",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        }
+                    ],
+                    "index": 3,
+                    "bbox_fs": [
+                        155,
+                        156,
+                        258,
+                        189
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        310,
+                        156,
+                        455,
+                        189
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                345,
+                                156,
+                                419,
+                                167
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        345,
+                                        156,
+                                        419,
+                                        167
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Fabrizio Frasca⇤",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                311,
+                                168,
+                                455,
+                                179
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        311,
+                                        168,
+                                        455,
+                                        179
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Imperial College London & Twitter",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                332,
+                                178,
+                                434,
+                                189
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        332,
+                                        178,
+                                        434,
+                                        189
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ffrasca@twitter.com",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        }
+                    ],
+                    "index": 4,
+                    "bbox_fs": [
+                        311,
+                        156,
+                        455,
+                        189
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        161,
+                        207,
+                        263,
+                        239
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                187,
+                                205,
+                                238,
+                                218
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        187,
+                                        205,
+                                        238,
+                                        218
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Nina Otter",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                196,
+                                217,
+                                227,
+                                227
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        196,
+                                        217,
+                                        227,
+                                        227
+                                    ],
+                                    "score": 1.0,
+                                    "content": "UCLA",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                162,
+                                229,
+                                263,
+                                239
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        162,
+                                        229,
+                                        263,
+                                        239
+                                    ],
+                                    "score": 1.0,
+                                    "content": "otter@math.ucla.edu",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        }
+                    ],
+                    "index": 9,
+                    "bbox_fs": [
+                        162,
+                        205,
+                        263,
+                        239
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        321,
+                        206,
+                        450,
+                        240
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                349,
+                                205,
+                                424,
+                                219
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        349,
+                                        205,
+                                        424,
+                                        219
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Yu Guang Wang",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                330,
+                                217,
+                                440,
+                                229
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        330,
+                                        217,
+                                        440,
+                                        229
+                                    ],
+                                    "score": 1.0,
+                                    "content": "MPI-MIS, SJTU & UNSW",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                320,
+                                229,
+                                451,
+                                241
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        320,
+                                        229,
+                                        451,
+                                        241
+                                    ],
+                                    "score": 1.0,
+                                    "content": "yuguang.wang@unsw.edu.au",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        }
+                    ],
+                    "index": 10,
+                    "bbox_fs": [
+                        320,
+                        205,
+                        451,
+                        241
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        113,
+                        256,
+                        214,
+                        289
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                141,
+                                255,
+                                187,
+                                267
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        255,
+                                        187,
+                                        267
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Pietro Liò",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                113,
+                                266,
+                                216,
+                                280
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        113,
+                                        266,
+                                        216,
+                                        280
+                                    ],
+                                    "score": 1.0,
+                                    "content": "University of Cambridge",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                122,
+                                278,
+                                205,
+                                290
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        122,
+                                        278,
+                                        205,
+                                        290
+                                    ],
+                                    "score": 1.0,
+                                    "content": "pl219@cam.ac.uk",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        }
+                    ],
+                    "index": 16,
+                    "bbox_fs": [
+                        113,
+                        255,
+                        216,
+                        290
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        225,
+                        256,
+                        343,
+                        289
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                248,
+                                256,
+                                322,
+                                267
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        248,
+                                        256,
+                                        322,
+                                        267
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Guido Montúfar",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                244,
+                                267,
+                                324,
+                                277
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        244,
+                                        267,
+                                        324,
+                                        277
+                                    ],
+                                    "score": 1.0,
+                                    "content": "MPI-MIS & UCLA",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                225,
+                                279,
+                                343,
+                                289
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        225,
+                                        279,
+                                        343,
+                                        289
+                                    ],
+                                    "score": 1.0,
+                                    "content": "montufar@math.ucla.edu",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        }
+                    ],
+                    "index": 17,
+                    "bbox_fs": [
+                        225,
+                        256,
+                        343,
+                        289
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        353,
+                        256,
+                        497,
+                        289
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                385,
+                                256,
+                                466,
+                                268
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        385,
+                                        256,
+                                        466,
+                                        268
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Michael Bronstein",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                354,
+                                268,
+                                498,
+                                279
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        354,
+                                        268,
+                                        498,
+                                        279
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Imperial College London & Twitter",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                366,
+                                277,
+                                484,
+                                290
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        366,
+                                        277,
+                                        484,
+                                        290
+                                    ],
+                                    "score": 1.0,
+                                    "content": "mbronstein@twitter.com",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        }
+                    ],
+                    "index": 18,
+                    "bbox_fs": [
+                        354,
+                        256,
+                        498,
+                        290
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        283,
+                        318,
+                        328,
+                        331
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                281,
+                                317,
+                                330,
+                                333
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        281,
+                                        317,
+                                        330,
+                                        333
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Abstract",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        }
+                    ],
+                    "index": 22
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        142,
+                        343,
+                        469,
+                        539
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                142,
+                                342,
+                                469,
+                                356
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        142,
+                                        342,
+                                        469,
+                                        356
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Graph Neural Networks (GNNs) are limited in their expressive power, struggle with",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                354,
+                                470,
+                                367
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        354,
+                                        470,
+                                        367
+                                    ],
+                                    "score": 1.0,
+                                    "content": "long-range interactions and lack a principled way to model higher-order structures.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                364,
+                                470,
+                                378
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        364,
+                                        470,
+                                        378
+                                    ],
+                                    "score": 1.0,
+                                    "content": "These problems can be attributed to the strong coupling between the computational",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                375,
+                                470,
+                                389
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        375,
+                                        470,
+                                        389
+                                    ],
+                                    "score": 1.0,
+                                    "content": "graph and the input graph structure. The recently proposed Message Passing",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                387,
+                                469,
+                                399
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        387,
+                                        469,
+                                        399
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Simplicial Networks naturally decouple these elements by performing message",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                398,
+                                470,
+                                410
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        398,
+                                        470,
+                                        410
+                                    ],
+                                    "score": 1.0,
+                                    "content": "passing on the clique complex of the graph. Nevertheless, these models can be",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                408,
+                                469,
+                                420
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        408,
+                                        469,
+                                        420
+                                    ],
+                                    "score": 1.0,
+                                    "content": "severely constrained by the rigid combinatorial structure of Simplicial Complexes",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                420,
+                                469,
+                                432
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        420,
+                                        469,
+                                        432
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(SCs). In this work, we extend recent theoretical results on SCs to regular Cell",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                430,
+                                470,
+                                443
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        430,
+                                        470,
+                                        443
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Complexes, topological objects that flexibly subsume SCs and graphs. We show",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                441,
+                                471,
+                                454
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        441,
+                                        471,
+                                        454
+                                    ],
+                                    "score": 1.0,
+                                    "content": "that this generalisation provides a powerful set of graph “lifting” transformations,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                451,
+                                470,
+                                465
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        451,
+                                        470,
+                                        465
+                                    ],
+                                    "score": 1.0,
+                                    "content": "each leading to a unique hierarchical message passing procedure. The resulting",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                462,
+                                469,
+                                475
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        462,
+                                        469,
+                                        475
+                                    ],
+                                    "score": 1.0,
+                                    "content": "methods, which we collectively call CW Networks (CWNs), are strictly more",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                474,
+                                470,
+                                487
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        474,
+                                        470,
+                                        487
+                                    ],
+                                    "score": 1.0,
+                                    "content": "powerful than the WL test and not less powerful than the 3-WL test. In particular,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                484,
+                                470,
+                                498
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        484,
+                                        470,
+                                        498
+                                    ],
+                                    "score": 1.0,
+                                    "content": "we demonstrate the effectiveness of one such scheme, based on rings, when applied",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                496,
+                                469,
+                                508
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        496,
+                                        469,
+                                        508
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to molecular graph problems. The proposed architecture benefits from provably",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                507,
+                                470,
+                                520
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        507,
+                                        470,
+                                        520
+                                    ],
+                                    "score": 1.0,
+                                    "content": "larger expressivity than commonly used GNNs, principled modelling of higher-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                518,
+                                469,
+                                529
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        518,
+                                        469,
+                                        529
+                                    ],
+                                    "score": 1.0,
+                                    "content": "order signals and from compressing the distances between nodes. We demonstrate",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                528,
+                                466,
+                                541
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        528,
+                                        466,
+                                        541
+                                    ],
+                                    "score": 1.0,
+                                    "content": "that our model achieves state-of-the-art results on a variety of molecular datasets.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        }
+                    ],
+                    "index": 31.5,
+                    "bbox_fs": [
+                        141,
+                        342,
+                        471,
+                        541
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        560,
+                        190,
+                        574
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                559,
+                                192,
+                                576
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        559,
+                                        192,
+                                        576
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1 Introduction",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        }
+                    ],
+                    "index": 41
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        585,
+                        505,
+                        695
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                585,
+                                505,
+                                597
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        585,
+                                        505,
+                                        597
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The operations performed by message passing Graph Neural Networks (GNNs) emulate the structure",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                596,
+                                506,
+                                608
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        596,
+                                        506,
+                                        608
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of the input graph. While this property has clear computational advantages, it brings with it a series of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                606,
+                                504,
+                                619
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        607,
+                                        303,
+                                        619
+                                    ],
+                                    "score": 1.0,
+                                    "content": "fundamental limitations. As observed by Xu et al.",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        303,
+                                        606,
+                                        321,
+                                        618
+                                    ],
+                                    "score": 0.6,
+                                    "content": "\\dot { \\lVert \\mathcal { I } \\dot { 4 } \\rVert }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        322,
+                                        607,
+                                        504,
+                                        619
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and Morris et al. [55] the local neighbourhood",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                617,
+                                504,
+                                631
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        617,
+                                        475,
+                                        631
+                                    ],
+                                    "score": 1.0,
+                                    "content": "aggregations used by GNNs are at most as powerful as the Weisfeiler-Lehman (WL) test",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        475,
+                                        617,
+                                        493,
+                                        629
+                                    ],
+                                    "score": 0.48,
+                                    "content": "\\pmb { \\mathbb { Z } } \\mathbf { \\mathbb { 1 } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        493,
+                                        617,
+                                        504,
+                                        631
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                628,
+                                505,
+                                641
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        628,
+                                        505,
+                                        641
+                                    ],
+                                    "score": 1.0,
+                                    "content": "distinguishing non-isomorphic graphs. Therefore, GNNs fail to detect certain higer-order meso-scale",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                639,
+                                505,
+                                653
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        639,
+                                        505,
+                                        653
+                                    ],
+                                    "score": 1.0,
+                                    "content": "structures such as cliques or (induced) cycles [2, 15], which are particularly important in applications",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 47
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                650,
+                                506,
+                                663
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        650,
+                                        506,
+                                        663
+                                    ],
+                                    "score": 1.0,
+                                    "content": "dealing with social and biological networks or molecular graphs. At the same time, many such layers",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 48
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                662,
+                                505,
+                                674
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        662,
+                                        505,
+                                        674
+                                    ],
+                                    "score": 1.0,
+                                    "content": "have to be stacked to make long-range interactions in the graph possible. Besides the computational",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 49
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                673,
+                                503,
+                                684
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        673,
+                                        503,
+                                        684
+                                    ],
+                                    "score": 1.0,
+                                    "content": "burden this incurs, deep GNNs typically come with additional problems such as over-smoothing [51]",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 50
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                684,
+                                312,
+                                695
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        684,
+                                        312,
+                                        695
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and over-squashing [1] of the node representations.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 51
+                        }
+                    ],
+                    "index": 46.5,
+                    "bbox_fs": [
+                        105,
+                        585,
+                        506,
+                        695
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        73,
+                        505,
+                        193
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                72,
+                                506,
+                                86
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        72,
+                                        506,
+                                        86
+                                    ],
+                                    "score": 1.0,
+                                    "content": "To address these problems, we propose a novel message passing procedure based on (regular)",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                83,
+                                506,
+                                97
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        83,
+                                        506,
+                                        97
+                                    ],
+                                    "score": 1.0,
+                                    "content": "cell complexes, also known as CW complexes2, topological objects that form the building block",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                94,
+                                505,
+                                107
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        94,
+                                        505,
+                                        107
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of algebraic topology [38]. When paired with a theoretically-justified “lifting” transformation",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                105,
+                                506,
+                                118
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        105,
+                                        506,
+                                        118
+                                    ],
+                                    "score": 1.0,
+                                    "content": "augmenting the graph with higher-dimensional constructs called “cells”, our method results in a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                117,
+                                505,
+                                129
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        117,
+                                        505,
+                                        129
+                                    ],
+                                    "score": 1.0,
+                                    "content": "multi-dimensional and hierarchical message passing procedure over the input graph. Our approach",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                126,
+                                506,
+                                140
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        126,
+                                        506,
+                                        140
+                                    ],
+                                    "score": 1.0,
+                                    "content": "generalises and subsumes the recently proposed Message Passing Simplicial Networks (MPSNs) [8],",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                138,
+                                505,
+                                151
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        138,
+                                        505,
+                                        151
+                                    ],
+                                    "score": 1.0,
+                                    "content": "which operate on simplicial complexes (SCs), topological generalisations of graphs. However, SCs",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                148,
+                                506,
+                                162
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        148,
+                                        506,
+                                        162
+                                    ],
+                                    "score": 1.0,
+                                    "content": "have a rigid combinatorial structure that significantly limits the range of lifting transformations one",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                160,
+                                505,
+                                172
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        160,
+                                        505,
+                                        172
+                                    ],
+                                    "score": 1.0,
+                                    "content": "could use to meaningfully modulate the message passing procedure. In contrast, we show that cell",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                170,
+                                505,
+                                183
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        170,
+                                        505,
+                                        183
+                                    ],
+                                    "score": 1.0,
+                                    "content": "complexes, which in turn generalise simplicial complexes and come with additional flexibility, allow",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                182,
+                                459,
+                                195
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        182,
+                                        459,
+                                        195
+                                    ],
+                                    "score": 1.0,
+                                    "content": "one to construct new and better ways of decoupling the input and computational graphs.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        }
+                    ],
+                    "index": 5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        204,
+                        505,
+                        348
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                203,
+                                506,
+                                217
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        203,
+                                        506,
+                                        217
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Main Contributions To summarise, we propose a message passing scheme operating on regular",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                214,
+                                506,
+                                228
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        214,
+                                        506,
+                                        228
+                                    ],
+                                    "score": 1.0,
+                                    "content": "cell complexes. We call this family of models CW Networks (CWNs) and study their expressive power",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                226,
+                                505,
+                                239
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        226,
+                                        505,
+                                        239
+                                    ],
+                                    "score": 1.0,
+                                    "content": "using a cellular version of the WL test. We show that for an entire class of “lifting” transformations",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                237,
+                                505,
+                                249
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        237,
+                                        505,
+                                        249
+                                    ],
+                                    "score": 1.0,
+                                    "content": "CWNs are at least as powerful as the WL test. Furthermore, we prove that for some of the maps in this",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                247,
+                                505,
+                                261
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        247,
+                                        505,
+                                        261
+                                    ],
+                                    "score": 1.0,
+                                    "content": "class, CWNs can be strictly more powerful than WL, Simplicial WL (SWL) and also not less powerful",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                259,
+                                505,
+                                271
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        259,
+                                        505,
+                                        271
+                                    ],
+                                    "score": 1.0,
+                                    "content": "than 3-WL. We also express the fundamental symmetries of these models and show how they can be",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                270,
+                                506,
+                                282
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        270,
+                                        506,
+                                        282
+                                    ],
+                                    "score": 1.0,
+                                    "content": "seen as generalised convolutional operators on cell complexes. Experimentally, we focus our attention",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                280,
+                                506,
+                                293
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        280,
+                                        506,
+                                        293
+                                    ],
+                                    "score": 1.0,
+                                    "content": "on a particular “lifting” map based on induced cycles. When applied to molecular graphs, it leads to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                291,
+                                506,
+                                304
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        291,
+                                        506,
+                                        304
+                                    ],
+                                    "score": 1.0,
+                                    "content": "an intuitive hierarchical message passing procedure involving the atoms, the bonds between them and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                302,
+                                505,
+                                315
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        302,
+                                        505,
+                                        315
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the chemical rings of the molecules. We demonstrate that this provably powerful approach obtains",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                313,
+                                505,
+                                325
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        313,
+                                        505,
+                                        325
+                                    ],
+                                    "score": 1.0,
+                                    "content": "state-of-the-art results on popular large-scale molecular graph datasets and other related tasks. To the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                324,
+                                506,
+                                337
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        324,
+                                        506,
+                                        337
+                                    ],
+                                    "score": 1.0,
+                                    "content": "best of our knowledge, this is the first work proposing a cell complex representation for molecules.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                335,
+                                410,
+                                347
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        335,
+                                        410,
+                                        347
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Our code is available at https://github.com/twitter-research/cwn.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        }
+                    ],
+                    "index": 17
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        361,
+                        189,
+                        375
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                359,
+                                190,
+                                378
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        359,
+                                        190,
+                                        378
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2 Background",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        }
+                    ],
+                    "index": 24
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        385,
+                        503,
+                        409
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                384,
+                                504,
+                                399
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        384,
+                                        238,
+                                        399
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Definition 1 (Hansen and Ghrist",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        238,
+                                        385,
+                                        256,
+                                        397
+                                    ],
+                                    "score": 0.78,
+                                    "content": "\\begin{array} { r l } {  { \\bigl \\| \\widehat { 3 6 } \\bigr \\| } } & { { } } \\end{array}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        256,
+                                        384,
+                                        393,
+                                        399
+                                    ],
+                                    "score": 1.0,
+                                    "content": "). A regular cell complex (Figure",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        393,
+                                        385,
+                                        404,
+                                        398
+                                    ],
+                                    "score": 0.4,
+                                    "content": "\\mathbb { D }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        404,
+                                        384,
+                                        493,
+                                        399
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is a topological space",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        494,
+                                        388,
+                                        504,
+                                        396
+                                    ],
+                                    "score": 0.66,
+                                    "content": "X",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                395,
+                                452,
+                                410
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        395,
+                                        207,
+                                        410
+                                    ],
+                                    "score": 1.0,
+                                    "content": "together with a partition",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        207,
+                                        396,
+                                        253,
+                                        409
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\{ X _ { \\sigma } \\} _ { \\sigma \\in P _ { X } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        253,
+                                        395,
+                                        308,
+                                        410
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of subspaces",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        308,
+                                        397,
+                                        322,
+                                        408
+                                    ],
+                                    "score": 0.9,
+                                    "content": "X _ { \\sigma }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        323,
+                                        395,
+                                        334,
+                                        410
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        335,
+                                        397,
+                                        344,
+                                        406
+                                    ],
+                                    "score": 0.79,
+                                    "content": "X",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        345,
+                                        395,
+                                        452,
+                                        410
+                                    ],
+                                    "score": 1.0,
+                                    "content": "called cells, and such that",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        }
+                    ],
+                    "index": 25.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        412,
+                        482,
+                        424
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                410,
+                                483,
+                                426
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        410,
+                                        156,
+                                        426
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1. For each",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        157,
+                                        413,
+                                        185,
+                                        423
+                                    ],
+                                    "score": 0.9,
+                                    "content": "x \\in X",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        185,
+                                        410,
+                                        340,
+                                        426
+                                    ],
+                                    "score": 1.0,
+                                    "content": "there exists an open neighbordhood of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        341,
+                                        415,
+                                        348,
+                                        423
+                                    ],
+                                    "score": 0.25,
+                                    "content": "x",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        348,
+                                        410,
+                                        483,
+                                        426
+                                    ],
+                                    "score": 1.0,
+                                    "content": "that intersects finitely many cells.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        }
+                    ],
+                    "index": 27
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        426,
+                        469,
+                        439
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                424,
+                                469,
+                                441
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        424,
+                                        148,
+                                        441
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2. For all",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        149,
+                                        428,
+                                        165,
+                                        438
+                                    ],
+                                    "score": 0.84,
+                                    "content": "\\sigma , \\tau",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        165,
+                                        424,
+                                        219,
+                                        441
+                                    ],
+                                    "score": 1.0,
+                                    "content": "we have that",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        219,
+                                        425,
+                                        277,
+                                        439
+                                    ],
+                                    "score": 0.93,
+                                    "content": "{ \\cal X } _ { \\tau } \\cap \\overline { { { \\cal X } _ { \\sigma } } } \\ne \\emptyset",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        277,
+                                        424,
+                                        288,
+                                        441
+                                    ],
+                                    "score": 1.0,
+                                    "content": "iff",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        289,
+                                        426,
+                                        329,
+                                        438
+                                    ],
+                                    "score": 0.91,
+                                    "content": "X _ { \\tau } \\subseteq { \\overline { { X _ { \\sigma } } } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        330,
+                                        424,
+                                        360,
+                                        441
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", where",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        360,
+                                        426,
+                                        375,
+                                        438
+                                    ],
+                                    "score": 0.89,
+                                    "content": "\\overline { { X _ { \\sigma } } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        375,
+                                        424,
+                                        469,
+                                        441
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is the closure of a cell.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        }
+                    ],
+                    "index": 28
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        109,
+                        441,
+                        304,
+                        453
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                439,
+                                302,
+                                455
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        439,
+                                        243,
+                                        455
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3. Every cell is homeomorphic to",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        243,
+                                        441,
+                                        257,
+                                        452
+                                    ],
+                                    "score": 0.86,
+                                    "content": "\\mathbb { R } ^ { n }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        257,
+                                        439,
+                                        295,
+                                        455
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for some",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        295,
+                                        444,
+                                        302,
+                                        451
+                                    ],
+                                    "score": 0.41,
+                                    "content": "n",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 29
+                        }
+                    ],
+                    "index": 29
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        455,
+                        503,
+                        478
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                454,
+                                505,
+                                468
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        454,
+                                        211,
+                                        468
+                                    ],
+                                    "score": 1.0,
+                                    "content": "4. (Regularity) For every",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        211,
+                                        456,
+                                        244,
+                                        466
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\sigma \\in P _ { X }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        245,
+                                        454,
+                                        354,
+                                        468
+                                    ],
+                                    "score": 1.0,
+                                    "content": "there is a homeomorphism",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        355,
+                                        456,
+                                        362,
+                                        467
+                                    ],
+                                    "score": 0.81,
+                                    "content": "\\phi",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        363,
+                                        454,
+                                        438,
+                                        468
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of a closed ball in",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        438,
+                                        456,
+                                        456,
+                                        466
+                                    ],
+                                    "score": 0.89,
+                                    "content": "\\mathbb { R } ^ { n _ { \\sigma } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        456,
+                                        454,
+                                        468,
+                                        468
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        468,
+                                        455,
+                                        483,
+                                        466
+                                    ],
+                                    "score": 0.87,
+                                    "content": "\\overline { { X _ { \\sigma } } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        483,
+                                        454,
+                                        505,
+                                        468
+                                    ],
+                                    "score": 1.0,
+                                    "content": "such",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                118,
+                                466,
+                                438,
+                                479
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        118,
+                                        466,
+                                        205,
+                                        479
+                                    ],
+                                    "score": 1.0,
+                                    "content": "that the restriction of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        206,
+                                        467,
+                                        213,
+                                        478
+                                    ],
+                                    "score": 0.82,
+                                    "content": "\\phi",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        213,
+                                        466,
+                                        420,
+                                        479
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to the interior of the ball is a homeomorphism onto",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        420,
+                                        467,
+                                        435,
+                                        477
+                                    ],
+                                    "score": 0.89,
+                                    "content": "X _ { \\sigma }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        435,
+                                        466,
+                                        438,
+                                        479
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        }
+                    ],
+                    "index": 30.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        485,
+                        356,
+                        552
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                484,
+                                357,
+                                498
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        484,
+                                        293,
+                                        498
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We note that by condition (2) the indexing set",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        294,
+                                        486,
+                                        308,
+                                        497
+                                    ],
+                                    "score": 0.88,
+                                    "content": "P _ { X }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        309,
+                                        484,
+                                        357,
+                                        498
+                                    ],
+                                    "score": 1.0,
+                                    "content": "has a poset",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                497,
+                                356,
+                                510
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        497,
+                                        144,
+                                        510
+                                    ],
+                                    "score": 1.0,
+                                    "content": "structure",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        145,
+                                        497,
+                                        226,
+                                        509
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\tau \\leq \\sigma \\Leftrightarrow X _ { \\tau } \\subseteq \\overline { { X _ { \\sigma } } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        226,
+                                        497,
+                                        356,
+                                        510
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", while condition (4) guarantees",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                508,
+                                356,
+                                520
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        508,
+                                        356,
+                                        520
+                                    ],
+                                    "score": 1.0,
+                                    "content": "that this poset structure encodes all the topological information",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                519,
+                                356,
+                                532
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        519,
+                                        131,
+                                        532
+                                    ],
+                                    "score": 1.0,
+                                    "content": "about",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        131,
+                                        520,
+                                        141,
+                                        529
+                                    ],
+                                    "score": 0.79,
+                                    "content": "X",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        141,
+                                        519,
+                                        325,
+                                        532
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Thus, we can identify a regular cell complex",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        325,
+                                        520,
+                                        335,
+                                        529
+                                    ],
+                                    "score": 0.79,
+                                    "content": "X",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        335,
+                                        519,
+                                        356,
+                                        532
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                531,
+                                356,
+                                542
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        531,
+                                        233,
+                                        542
+                                    ],
+                                    "score": 1.0,
+                                    "content": "this poset, called face poset of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        233,
+                                        531,
+                                        243,
+                                        540
+                                    ],
+                                    "score": 0.83,
+                                    "content": "X",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        244,
+                                        531,
+                                        299,
+                                        542
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". We also use",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        299,
+                                        532,
+                                        326,
+                                        541
+                                    ],
+                                    "score": 0.89,
+                                    "content": "\\tau < \\sigma",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        326,
+                                        531,
+                                        356,
+                                        542
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                542,
+                                241,
+                                553
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        542,
+                                        241,
+                                        553
+                                    ],
+                                    "score": 1.0,
+                                    "content": "strict version of this partial order.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        }
+                    ],
+                    "index": 34.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        558,
+                        357,
+                        659
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                558,
+                                356,
+                                569
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        558,
+                                        356,
+                                        569
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Intuitively, one constructs a cell complex through a hierarchical",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                568,
+                                358,
+                                581
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        568,
+                                        358,
+                                        581
+                                    ],
+                                    "score": 1.0,
+                                    "content": "gluing procedure. One starts with a set of vertices (0-cells).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                579,
+                                358,
+                                592
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        579,
+                                        358,
+                                        592
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Then edges (1-cells) are attached to these by gluing the end-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                591,
+                                357,
+                                603
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        591,
+                                        357,
+                                        603
+                                    ],
+                                    "score": 1.0,
+                                    "content": "points of closed line segments to them. We have now only",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                601,
+                                357,
+                                613
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        601,
+                                        357,
+                                        613
+                                    ],
+                                    "score": 1.0,
+                                    "content": "described a (multi) graph. However, one can generalise this",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                612,
+                                357,
+                                624
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        612,
+                                        357,
+                                        624
+                                    ],
+                                    "score": 1.0,
+                                    "content": "even further by taking a two-dimensional closed disk and glue",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                623,
+                                357,
+                                636
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        623,
+                                        357,
+                                        636
+                                    ],
+                                    "score": 1.0,
+                                    "content": "its boundary (i.e. a circle) to any simple cycle in the (multi)",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                633,
+                                356,
+                                646
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        634,
+                                        247,
+                                        646
+                                    ],
+                                    "score": 1.0,
+                                    "content": "graph previously built as in Figure",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        248,
+                                        633,
+                                        258,
+                                        646
+                                    ],
+                                    "score": 0.77,
+                                    "content": "2 .",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        258,
+                                        634,
+                                        356,
+                                        646
+                                    ],
+                                    "score": 1.0,
+                                    "content": "While we are generally",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                645,
+                                357,
+                                656
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        645,
+                                        357,
+                                        656
+                                    ],
+                                    "score": 1.0,
+                                    "content": "not concerned with dimensions above two, this can be further",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 47
+                        }
+                    ],
+                    "index": 43
+                },
+                {
+                    "type": "image",
+                    "bbox": [
+                        365,
+                        483,
+                        504,
+                        603
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                365,
+                                483,
+                                504,
+                                603
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        365,
+                                        483,
+                                        504,
+                                        603
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                365,
+                                                483,
+                                                504,
+                                                603
+                                            ],
+                                            "score": 0.965,
+                                            "type": "image",
+                                            "image_path": "eefbc01081706e78f452ed80568013eb2d16d347cad7530a3a1b1de0d350067b.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 43.0,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        365,
+                                        483,
+                                        504,
+                                        543.0
+                                    ],
+                                    "spans": [],
+                                    "index": 38
+                                },
+                                {
+                                    "bbox": [
+                                        365,
+                                        543.0,
+                                        504,
+                                        603.0
+                                    ],
+                                    "spans": [],
+                                    "index": 48
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                363,
+                                608,
+                                505,
+                                653
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        363,
+                                        608,
+                                        505,
+                                        619
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                363,
+                                                608,
+                                                463,
+                                                619
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 1: A cell complex",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                463,
+                                                609,
+                                                473,
+                                                618
+                                            ],
+                                            "score": 0.76,
+                                            "content": "X",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                474,
+                                                608,
+                                                505,
+                                                619
+                                            ],
+                                            "score": 1.0,
+                                            "content": "and the",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 49
+                                },
+                                {
+                                    "bbox": [
+                                        363,
+                                        619,
+                                        505,
+                                        630
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                363,
+                                                619,
+                                                505,
+                                                630
+                                            ],
+                                            "score": 1.0,
+                                            "content": "corresponding homeomorphisms to",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 50
+                                },
+                                {
+                                    "bbox": [
+                                        363,
+                                        630,
+                                        506,
+                                        641
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                363,
+                                                630,
+                                                506,
+                                                641
+                                            ],
+                                            "score": 1.0,
+                                            "content": "the closed balls for three cells of dif-",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 51
+                                },
+                                {
+                                    "bbox": [
+                                        363,
+                                        641,
+                                        501,
+                                        653
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                363,
+                                                641,
+                                                501,
+                                                653
+                                            ],
+                                            "score": 1.0,
+                                            "content": "ferent dimensions in the complex.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 52
+                                }
+                            ],
+                            "index": 50.5
+                        }
+                    ],
+                    "index": 46.75
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        111,
+                        656,
+                        502,
+                        667
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                108,
+                                654,
+                                504,
+                                669
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        108,
+                                        654,
+                                        260,
+                                        669
+                                    ],
+                                    "score": 1.0,
+                                    "content": "generalised by gluing the boundary of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        261,
+                                        658,
+                                        267,
+                                        666
+                                    ],
+                                    "score": 0.79,
+                                    "content": "n",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        268,
+                                        654,
+                                        382,
+                                        669
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-dimensional balls to certain",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        383,
+                                        655,
+                                        414,
+                                        668
+                                    ],
+                                    "score": 0.72,
+                                    "content": "( n - 1 )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        414,
+                                        654,
+                                        504,
+                                        669
+                                    ],
+                                    "score": 1.0,
+                                    "content": "\u0000cells in the complex.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 53
+                        }
+                    ],
+                    "index": 53
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        672,
+                        505,
+                        694
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                671,
+                                506,
+                                685
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        671,
+                                        239,
+                                        685
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Consider the examples in Figure",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        239,
+                                        672,
+                                        249,
+                                        685
+                                    ],
+                                    "score": 0.76,
+                                    "content": "\\textcircled { 3 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        250,
+                                        671,
+                                        506,
+                                        685
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The shown sphere is a cell complex obtained from two 0-cells",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 54
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                683,
+                                505,
+                                696
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        683,
+                                        505,
+                                        696
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(i.e. vertices), to which two 1-cells (i.e. edges), which form the equator, were attached. The boundary",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 55
+                        }
+                    ],
+                    "index": 54.5
+                }
+            ],
+            "page_idx": 1,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        106,
+                        701,
+                        506,
+                        722
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                118,
+                                699,
+                                506,
+                                713
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        118,
+                                        699,
+                                        506,
+                                        713
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2 We use these terms interchangeably. For the latter, the C stands for “closure-finite”, and the W for “weak”",
+                                    "type": "text"
+                                }
+                            ]
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                710,
+                                291,
+                                723
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        710,
+                                        272,
+                                        723
+                                    ],
+                                    "score": 1.0,
+                                    "content": "topology. The term was coined by Whitehead",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        273,
+                                        711,
+                                        288,
+                                        722
+                                    ],
+                                    "score": 0.75,
+                                    "content": "\\pmb { \\mathbb { Z } 2 } \\|",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        289,
+                                        710,
+                                        291,
+                                        723
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        302,
+                        741,
+                        309,
+                        750
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                301,
+                                740,
+                                310,
+                                753
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        301,
+                                        740,
+                                        310,
+                                        753
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        73,
+                        505,
+                        193
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                72,
+                                506,
+                                86
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        72,
+                                        506,
+                                        86
+                                    ],
+                                    "score": 1.0,
+                                    "content": "To address these problems, we propose a novel message passing procedure based on (regular)",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                83,
+                                506,
+                                97
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        83,
+                                        506,
+                                        97
+                                    ],
+                                    "score": 1.0,
+                                    "content": "cell complexes, also known as CW complexes2, topological objects that form the building block",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                94,
+                                505,
+                                107
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        94,
+                                        505,
+                                        107
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of algebraic topology [38]. When paired with a theoretically-justified “lifting” transformation",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                105,
+                                506,
+                                118
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        105,
+                                        506,
+                                        118
+                                    ],
+                                    "score": 1.0,
+                                    "content": "augmenting the graph with higher-dimensional constructs called “cells”, our method results in a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                117,
+                                505,
+                                129
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        117,
+                                        505,
+                                        129
+                                    ],
+                                    "score": 1.0,
+                                    "content": "multi-dimensional and hierarchical message passing procedure over the input graph. Our approach",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                126,
+                                506,
+                                140
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        126,
+                                        506,
+                                        140
+                                    ],
+                                    "score": 1.0,
+                                    "content": "generalises and subsumes the recently proposed Message Passing Simplicial Networks (MPSNs) [8],",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                138,
+                                505,
+                                151
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        138,
+                                        505,
+                                        151
+                                    ],
+                                    "score": 1.0,
+                                    "content": "which operate on simplicial complexes (SCs), topological generalisations of graphs. However, SCs",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                148,
+                                506,
+                                162
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        148,
+                                        506,
+                                        162
+                                    ],
+                                    "score": 1.0,
+                                    "content": "have a rigid combinatorial structure that significantly limits the range of lifting transformations one",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                160,
+                                505,
+                                172
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        160,
+                                        505,
+                                        172
+                                    ],
+                                    "score": 1.0,
+                                    "content": "could use to meaningfully modulate the message passing procedure. In contrast, we show that cell",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                170,
+                                505,
+                                183
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        170,
+                                        505,
+                                        183
+                                    ],
+                                    "score": 1.0,
+                                    "content": "complexes, which in turn generalise simplicial complexes and come with additional flexibility, allow",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                182,
+                                459,
+                                195
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        182,
+                                        459,
+                                        195
+                                    ],
+                                    "score": 1.0,
+                                    "content": "one to construct new and better ways of decoupling the input and computational graphs.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        }
+                    ],
+                    "index": 5,
+                    "bbox_fs": [
+                        104,
+                        72,
+                        506,
+                        195
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        204,
+                        505,
+                        348
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                203,
+                                506,
+                                217
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        203,
+                                        506,
+                                        217
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Main Contributions To summarise, we propose a message passing scheme operating on regular",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                214,
+                                506,
+                                228
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        214,
+                                        506,
+                                        228
+                                    ],
+                                    "score": 1.0,
+                                    "content": "cell complexes. We call this family of models CW Networks (CWNs) and study their expressive power",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                226,
+                                505,
+                                239
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        226,
+                                        505,
+                                        239
+                                    ],
+                                    "score": 1.0,
+                                    "content": "using a cellular version of the WL test. We show that for an entire class of “lifting” transformations",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                237,
+                                505,
+                                249
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        237,
+                                        505,
+                                        249
+                                    ],
+                                    "score": 1.0,
+                                    "content": "CWNs are at least as powerful as the WL test. Furthermore, we prove that for some of the maps in this",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                247,
+                                505,
+                                261
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        247,
+                                        505,
+                                        261
+                                    ],
+                                    "score": 1.0,
+                                    "content": "class, CWNs can be strictly more powerful than WL, Simplicial WL (SWL) and also not less powerful",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                259,
+                                505,
+                                271
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        259,
+                                        505,
+                                        271
+                                    ],
+                                    "score": 1.0,
+                                    "content": "than 3-WL. We also express the fundamental symmetries of these models and show how they can be",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                270,
+                                506,
+                                282
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        270,
+                                        506,
+                                        282
+                                    ],
+                                    "score": 1.0,
+                                    "content": "seen as generalised convolutional operators on cell complexes. Experimentally, we focus our attention",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                280,
+                                506,
+                                293
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        280,
+                                        506,
+                                        293
+                                    ],
+                                    "score": 1.0,
+                                    "content": "on a particular “lifting” map based on induced cycles. When applied to molecular graphs, it leads to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                291,
+                                506,
+                                304
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        291,
+                                        506,
+                                        304
+                                    ],
+                                    "score": 1.0,
+                                    "content": "an intuitive hierarchical message passing procedure involving the atoms, the bonds between them and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                302,
+                                505,
+                                315
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        302,
+                                        505,
+                                        315
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the chemical rings of the molecules. We demonstrate that this provably powerful approach obtains",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                313,
+                                505,
+                                325
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        313,
+                                        505,
+                                        325
+                                    ],
+                                    "score": 1.0,
+                                    "content": "state-of-the-art results on popular large-scale molecular graph datasets and other related tasks. To the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                324,
+                                506,
+                                337
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        324,
+                                        506,
+                                        337
+                                    ],
+                                    "score": 1.0,
+                                    "content": "best of our knowledge, this is the first work proposing a cell complex representation for molecules.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                335,
+                                410,
+                                347
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        335,
+                                        410,
+                                        347
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Our code is available at https://github.com/twitter-research/cwn.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        }
+                    ],
+                    "index": 17,
+                    "bbox_fs": [
+                        105,
+                        203,
+                        506,
+                        347
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        361,
+                        189,
+                        375
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                359,
+                                190,
+                                378
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        359,
+                                        190,
+                                        378
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2 Background",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        }
+                    ],
+                    "index": 24
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        385,
+                        503,
+                        409
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                384,
+                                504,
+                                399
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        384,
+                                        238,
+                                        399
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Definition 1 (Hansen and Ghrist",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        238,
+                                        385,
+                                        256,
+                                        397
+                                    ],
+                                    "score": 0.78,
+                                    "content": "\\begin{array} { r l } {  { \\bigl \\| \\widehat { 3 6 } \\bigr \\| } } & { { } } \\end{array}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        256,
+                                        384,
+                                        393,
+                                        399
+                                    ],
+                                    "score": 1.0,
+                                    "content": "). A regular cell complex (Figure",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        393,
+                                        385,
+                                        404,
+                                        398
+                                    ],
+                                    "score": 0.4,
+                                    "content": "\\mathbb { D }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        404,
+                                        384,
+                                        493,
+                                        399
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is a topological space",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        494,
+                                        388,
+                                        504,
+                                        396
+                                    ],
+                                    "score": 0.66,
+                                    "content": "X",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                395,
+                                452,
+                                410
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        395,
+                                        207,
+                                        410
+                                    ],
+                                    "score": 1.0,
+                                    "content": "together with a partition",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        207,
+                                        396,
+                                        253,
+                                        409
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\{ X _ { \\sigma } \\} _ { \\sigma \\in P _ { X } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        253,
+                                        395,
+                                        308,
+                                        410
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of subspaces",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        308,
+                                        397,
+                                        322,
+                                        408
+                                    ],
+                                    "score": 0.9,
+                                    "content": "X _ { \\sigma }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        323,
+                                        395,
+                                        334,
+                                        410
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        335,
+                                        397,
+                                        344,
+                                        406
+                                    ],
+                                    "score": 0.79,
+                                    "content": "X",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        345,
+                                        395,
+                                        452,
+                                        410
+                                    ],
+                                    "score": 1.0,
+                                    "content": "called cells, and such that",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        }
+                    ],
+                    "index": 25.5,
+                    "bbox_fs": [
+                        105,
+                        384,
+                        504,
+                        410
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        412,
+                        482,
+                        424
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                410,
+                                483,
+                                426
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        410,
+                                        156,
+                                        426
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1. For each",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        157,
+                                        413,
+                                        185,
+                                        423
+                                    ],
+                                    "score": 0.9,
+                                    "content": "x \\in X",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        185,
+                                        410,
+                                        340,
+                                        426
+                                    ],
+                                    "score": 1.0,
+                                    "content": "there exists an open neighbordhood of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        341,
+                                        415,
+                                        348,
+                                        423
+                                    ],
+                                    "score": 0.25,
+                                    "content": "x",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        348,
+                                        410,
+                                        483,
+                                        426
+                                    ],
+                                    "score": 1.0,
+                                    "content": "that intersects finitely many cells.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        }
+                    ],
+                    "index": 27,
+                    "bbox_fs": [
+                        105,
+                        410,
+                        483,
+                        426
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        426,
+                        469,
+                        439
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                424,
+                                469,
+                                441
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        424,
+                                        148,
+                                        441
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2. For all",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        149,
+                                        428,
+                                        165,
+                                        438
+                                    ],
+                                    "score": 0.84,
+                                    "content": "\\sigma , \\tau",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        165,
+                                        424,
+                                        219,
+                                        441
+                                    ],
+                                    "score": 1.0,
+                                    "content": "we have that",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        219,
+                                        425,
+                                        277,
+                                        439
+                                    ],
+                                    "score": 0.93,
+                                    "content": "{ \\cal X } _ { \\tau } \\cap \\overline { { { \\cal X } _ { \\sigma } } } \\ne \\emptyset",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        277,
+                                        424,
+                                        288,
+                                        441
+                                    ],
+                                    "score": 1.0,
+                                    "content": "iff",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        289,
+                                        426,
+                                        329,
+                                        438
+                                    ],
+                                    "score": 0.91,
+                                    "content": "X _ { \\tau } \\subseteq { \\overline { { X _ { \\sigma } } } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        330,
+                                        424,
+                                        360,
+                                        441
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", where",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        360,
+                                        426,
+                                        375,
+                                        438
+                                    ],
+                                    "score": 0.89,
+                                    "content": "\\overline { { X _ { \\sigma } } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        375,
+                                        424,
+                                        469,
+                                        441
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is the closure of a cell.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        }
+                    ],
+                    "index": 28,
+                    "bbox_fs": [
+                        104,
+                        424,
+                        469,
+                        441
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        109,
+                        441,
+                        304,
+                        453
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                439,
+                                302,
+                                455
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        439,
+                                        243,
+                                        455
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3. Every cell is homeomorphic to",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        243,
+                                        441,
+                                        257,
+                                        452
+                                    ],
+                                    "score": 0.86,
+                                    "content": "\\mathbb { R } ^ { n }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        257,
+                                        439,
+                                        295,
+                                        455
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for some",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        295,
+                                        444,
+                                        302,
+                                        451
+                                    ],
+                                    "score": 0.41,
+                                    "content": "n",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 29
+                        }
+                    ],
+                    "index": 29,
+                    "bbox_fs": [
+                        106,
+                        439,
+                        302,
+                        455
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        455,
+                        503,
+                        478
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                454,
+                                505,
+                                468
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        454,
+                                        211,
+                                        468
+                                    ],
+                                    "score": 1.0,
+                                    "content": "4. (Regularity) For every",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        211,
+                                        456,
+                                        244,
+                                        466
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\sigma \\in P _ { X }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        245,
+                                        454,
+                                        354,
+                                        468
+                                    ],
+                                    "score": 1.0,
+                                    "content": "there is a homeomorphism",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        355,
+                                        456,
+                                        362,
+                                        467
+                                    ],
+                                    "score": 0.81,
+                                    "content": "\\phi",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        363,
+                                        454,
+                                        438,
+                                        468
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of a closed ball in",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        438,
+                                        456,
+                                        456,
+                                        466
+                                    ],
+                                    "score": 0.89,
+                                    "content": "\\mathbb { R } ^ { n _ { \\sigma } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        456,
+                                        454,
+                                        468,
+                                        468
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        468,
+                                        455,
+                                        483,
+                                        466
+                                    ],
+                                    "score": 0.87,
+                                    "content": "\\overline { { X _ { \\sigma } } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        483,
+                                        454,
+                                        505,
+                                        468
+                                    ],
+                                    "score": 1.0,
+                                    "content": "such",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                118,
+                                466,
+                                438,
+                                479
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        118,
+                                        466,
+                                        205,
+                                        479
+                                    ],
+                                    "score": 1.0,
+                                    "content": "that the restriction of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        206,
+                                        467,
+                                        213,
+                                        478
+                                    ],
+                                    "score": 0.82,
+                                    "content": "\\phi",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        213,
+                                        466,
+                                        420,
+                                        479
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to the interior of the ball is a homeomorphism onto",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        420,
+                                        467,
+                                        435,
+                                        477
+                                    ],
+                                    "score": 0.89,
+                                    "content": "X _ { \\sigma }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        435,
+                                        466,
+                                        438,
+                                        479
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        }
+                    ],
+                    "index": 30.5,
+                    "bbox_fs": [
+                        105,
+                        454,
+                        505,
+                        479
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        485,
+                        356,
+                        552
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                484,
+                                357,
+                                498
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        484,
+                                        293,
+                                        498
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We note that by condition (2) the indexing set",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        294,
+                                        486,
+                                        308,
+                                        497
+                                    ],
+                                    "score": 0.88,
+                                    "content": "P _ { X }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        309,
+                                        484,
+                                        357,
+                                        498
+                                    ],
+                                    "score": 1.0,
+                                    "content": "has a poset",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                497,
+                                356,
+                                510
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        497,
+                                        144,
+                                        510
+                                    ],
+                                    "score": 1.0,
+                                    "content": "structure",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        145,
+                                        497,
+                                        226,
+                                        509
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\tau \\leq \\sigma \\Leftrightarrow X _ { \\tau } \\subseteq \\overline { { X _ { \\sigma } } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        226,
+                                        497,
+                                        356,
+                                        510
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", while condition (4) guarantees",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                508,
+                                356,
+                                520
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        508,
+                                        356,
+                                        520
+                                    ],
+                                    "score": 1.0,
+                                    "content": "that this poset structure encodes all the topological information",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                519,
+                                356,
+                                532
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        519,
+                                        131,
+                                        532
+                                    ],
+                                    "score": 1.0,
+                                    "content": "about",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        131,
+                                        520,
+                                        141,
+                                        529
+                                    ],
+                                    "score": 0.79,
+                                    "content": "X",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        141,
+                                        519,
+                                        325,
+                                        532
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Thus, we can identify a regular cell complex",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        325,
+                                        520,
+                                        335,
+                                        529
+                                    ],
+                                    "score": 0.79,
+                                    "content": "X",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        335,
+                                        519,
+                                        356,
+                                        532
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                531,
+                                356,
+                                542
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        531,
+                                        233,
+                                        542
+                                    ],
+                                    "score": 1.0,
+                                    "content": "this poset, called face poset of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        233,
+                                        531,
+                                        243,
+                                        540
+                                    ],
+                                    "score": 0.83,
+                                    "content": "X",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        244,
+                                        531,
+                                        299,
+                                        542
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". We also use",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        299,
+                                        532,
+                                        326,
+                                        541
+                                    ],
+                                    "score": 0.89,
+                                    "content": "\\tau < \\sigma",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        326,
+                                        531,
+                                        356,
+                                        542
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                542,
+                                241,
+                                553
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        542,
+                                        241,
+                                        553
+                                    ],
+                                    "score": 1.0,
+                                    "content": "strict version of this partial order.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        }
+                    ],
+                    "index": 34.5,
+                    "bbox_fs": [
+                        106,
+                        484,
+                        357,
+                        553
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        558,
+                        357,
+                        659
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                558,
+                                356,
+                                569
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        558,
+                                        356,
+                                        569
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Intuitively, one constructs a cell complex through a hierarchical",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                568,
+                                358,
+                                581
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        568,
+                                        358,
+                                        581
+                                    ],
+                                    "score": 1.0,
+                                    "content": "gluing procedure. One starts with a set of vertices (0-cells).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                579,
+                                358,
+                                592
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        579,
+                                        358,
+                                        592
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Then edges (1-cells) are attached to these by gluing the end-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                591,
+                                357,
+                                603
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        591,
+                                        357,
+                                        603
+                                    ],
+                                    "score": 1.0,
+                                    "content": "points of closed line segments to them. We have now only",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                601,
+                                357,
+                                613
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        601,
+                                        357,
+                                        613
+                                    ],
+                                    "score": 1.0,
+                                    "content": "described a (multi) graph. However, one can generalise this",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                612,
+                                357,
+                                624
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        612,
+                                        357,
+                                        624
+                                    ],
+                                    "score": 1.0,
+                                    "content": "even further by taking a two-dimensional closed disk and glue",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                623,
+                                357,
+                                636
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        623,
+                                        357,
+                                        636
+                                    ],
+                                    "score": 1.0,
+                                    "content": "its boundary (i.e. a circle) to any simple cycle in the (multi)",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                633,
+                                356,
+                                646
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        634,
+                                        247,
+                                        646
+                                    ],
+                                    "score": 1.0,
+                                    "content": "graph previously built as in Figure",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        248,
+                                        633,
+                                        258,
+                                        646
+                                    ],
+                                    "score": 0.77,
+                                    "content": "2 .",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        258,
+                                        634,
+                                        356,
+                                        646
+                                    ],
+                                    "score": 1.0,
+                                    "content": "While we are generally",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                645,
+                                357,
+                                656
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        645,
+                                        357,
+                                        656
+                                    ],
+                                    "score": 1.0,
+                                    "content": "not concerned with dimensions above two, this can be further",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 47
+                        },
+                        {
+                            "bbox": [
+                                108,
+                                654,
+                                504,
+                                669
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        108,
+                                        654,
+                                        260,
+                                        669
+                                    ],
+                                    "score": 1.0,
+                                    "content": "generalised by gluing the boundary of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        261,
+                                        658,
+                                        267,
+                                        666
+                                    ],
+                                    "score": 0.79,
+                                    "content": "n",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        268,
+                                        654,
+                                        382,
+                                        669
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-dimensional balls to certain",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        383,
+                                        655,
+                                        414,
+                                        668
+                                    ],
+                                    "score": 0.72,
+                                    "content": "( n - 1 )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        414,
+                                        654,
+                                        504,
+                                        669
+                                    ],
+                                    "score": 1.0,
+                                    "content": "\u0000cells in the complex.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 53
+                        }
+                    ],
+                    "index": 43,
+                    "bbox_fs": [
+                        106,
+                        558,
+                        358,
+                        656
+                    ]
+                },
+                {
+                    "type": "image",
+                    "bbox": [
+                        365,
+                        483,
+                        504,
+                        603
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                365,
+                                483,
+                                504,
+                                603
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        365,
+                                        483,
+                                        504,
+                                        603
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                365,
+                                                483,
+                                                504,
+                                                603
+                                            ],
+                                            "score": 0.965,
+                                            "type": "image",
+                                            "image_path": "eefbc01081706e78f452ed80568013eb2d16d347cad7530a3a1b1de0d350067b.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 43.0,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        365,
+                                        483,
+                                        504,
+                                        543.0
+                                    ],
+                                    "spans": [],
+                                    "index": 38
+                                },
+                                {
+                                    "bbox": [
+                                        365,
+                                        543.0,
+                                        504,
+                                        603.0
+                                    ],
+                                    "spans": [],
+                                    "index": 48
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                363,
+                                608,
+                                505,
+                                653
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        363,
+                                        608,
+                                        505,
+                                        619
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                363,
+                                                608,
+                                                463,
+                                                619
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 1: A cell complex",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                463,
+                                                609,
+                                                473,
+                                                618
+                                            ],
+                                            "score": 0.76,
+                                            "content": "X",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                474,
+                                                608,
+                                                505,
+                                                619
+                                            ],
+                                            "score": 1.0,
+                                            "content": "and the",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 49
+                                },
+                                {
+                                    "bbox": [
+                                        363,
+                                        619,
+                                        505,
+                                        630
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                363,
+                                                619,
+                                                505,
+                                                630
+                                            ],
+                                            "score": 1.0,
+                                            "content": "corresponding homeomorphisms to",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 50
+                                },
+                                {
+                                    "bbox": [
+                                        363,
+                                        630,
+                                        506,
+                                        641
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                363,
+                                                630,
+                                                506,
+                                                641
+                                            ],
+                                            "score": 1.0,
+                                            "content": "the closed balls for three cells of dif-",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 51
+                                },
+                                {
+                                    "bbox": [
+                                        363,
+                                        641,
+                                        501,
+                                        653
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                363,
+                                                641,
+                                                501,
+                                                653
+                                            ],
+                                            "score": 1.0,
+                                            "content": "ferent dimensions in the complex.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 52
+                                }
+                            ],
+                            "index": 50.5
+                        }
+                    ],
+                    "index": 46.75
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        111,
+                        656,
+                        502,
+                        667
+                    ],
+                    "lines": [],
+                    "index": 53,
+                    "bbox_fs": [
+                        108,
+                        654,
+                        504,
+                        669
+                    ],
+                    "lines_deleted": true
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        672,
+                        505,
+                        694
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                671,
+                                506,
+                                685
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        671,
+                                        239,
+                                        685
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Consider the examples in Figure",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        239,
+                                        672,
+                                        249,
+                                        685
+                                    ],
+                                    "score": 0.76,
+                                    "content": "\\textcircled { 3 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        250,
+                                        671,
+                                        506,
+                                        685
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The shown sphere is a cell complex obtained from two 0-cells",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 54
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                683,
+                                505,
+                                696
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        683,
+                                        505,
+                                        696
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(i.e. vertices), to which two 1-cells (i.e. edges), which form the equator, were attached. The boundary",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 55
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                172,
+                                506,
+                                184
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        172,
+                                        506,
+                                        184
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of two 2-dimensional disks (i.e. the two hemispheres) were glued to the equator to form a sphere.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                182,
+                                506,
+                                195
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        182,
+                                        506,
+                                        195
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The second example is a tetrahedron with empty interior. It is a particular type of cell complex",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                193,
+                                506,
+                                207
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        193,
+                                        506,
+                                        207
+                                    ],
+                                    "score": 1.0,
+                                    "content": "called a simplicial complex (SC). The only 2-cells it allows are triangle-shaped. More generally, the",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                107,
+                                204,
+                                481,
+                                217
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        207,
+                                        114,
+                                        215
+                                    ],
+                                    "score": 0.71,
+                                    "content": "n",
+                                    "type": "inline_equation",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        114,
+                                        204,
+                                        231,
+                                        217
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-dimensional cells of SCs are",
+                                    "type": "text",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        232,
+                                        207,
+                                        239,
+                                        215
+                                    ],
+                                    "score": 0.74,
+                                    "content": "n",
+                                    "type": "inline_equation",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        239,
+                                        204,
+                                        481,
+                                        217
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-simplices, which makes them slightly more rigid structures.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 14
+                        }
+                    ],
+                    "index": 54.5,
+                    "bbox_fs": [
+                        105,
+                        671,
+                        506,
+                        696
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "image",
+                    "bbox": [
+                        107,
+                        70,
+                        358,
+                        120
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                107,
+                                70,
+                                358,
+                                120
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        70,
+                                        358,
+                                        120
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                107,
+                                                70,
+                                                358,
+                                                120
+                                            ],
+                                            "score": 0.961,
+                                            "type": "image",
+                                            "image_path": "a55263d0b752b3bc5d806b470b3759825fb4eb9c7d2ae5d1f90223fef4cc5a08.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 1,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        70,
+                                        358,
+                                        86.66666666666667
+                                    ],
+                                    "spans": [],
+                                    "index": 0
+                                },
+                                {
+                                    "bbox": [
+                                        107,
+                                        86.66666666666667,
+                                        358,
+                                        103.33333333333334
+                                    ],
+                                    "spans": [],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        107,
+                                        103.33333333333334,
+                                        358,
+                                        120.00000000000001
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                106,
+                                125,
+                                359,
+                                159
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        125,
+                                        360,
+                                        137
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                125,
+                                                360,
+                                                137
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 2: Closed two-dimensional disks are glued to the bound-",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 5
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        136,
+                                        358,
+                                        148
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                136,
+                                                358,
+                                                148
+                                            ],
+                                            "score": 1.0,
+                                            "content": "ary of the rings present in the graph (left). The result is a 2D",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 6
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        147,
+                                        221,
+                                        160
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                147,
+                                                221,
+                                                160
+                                            ],
+                                            "score": 1.0,
+                                            "content": "regular cell complex (right).",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 7
+                                }
+                            ],
+                            "index": 6
+                        }
+                    ],
+                    "index": 3.5
+                },
+                {
+                    "type": "image",
+                    "bbox": [
+                        381,
+                        71,
+                        490,
+                        120
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                381,
+                                71,
+                                490,
+                                120
+                            ],
+                            "group_id": 1,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        381,
+                                        71,
+                                        490,
+                                        120
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                381,
+                                                71,
+                                                490,
+                                                120
+                                            ],
+                                            "score": 0.935,
+                                            "type": "image",
+                                            "image_path": "0b72066b35dac85fbfc07fd956eee7a5fee05fd86266a0ea383297003ddffad9.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 3.5,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        381,
+                                        71,
+                                        490,
+                                        95.5
+                                    ],
+                                    "spans": [],
+                                    "index": 3
+                                },
+                                {
+                                    "bbox": [
+                                        381,
+                                        95.5,
+                                        490,
+                                        120.0
+                                    ],
+                                    "spans": [],
+                                    "index": 4
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                369,
+                                126,
+                                502,
+                                159
+                            ],
+                            "group_id": 1,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        368,
+                                        124,
+                                        502,
+                                        138
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                368,
+                                                124,
+                                                502,
+                                                138
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 3: A sphere and an empty",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 8
+                                },
+                                {
+                                    "bbox": [
+                                        369,
+                                        136,
+                                        502,
+                                        148
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                369,
+                                                136,
+                                                502,
+                                                148
+                                            ],
+                                            "score": 1.0,
+                                            "content": "tetrahedron. The latter is also a",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 9
+                                },
+                                {
+                                    "bbox": [
+                                        369,
+                                        147,
+                                        450,
+                                        160
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                369,
+                                                147,
+                                                450,
+                                                160
+                                            ],
+                                            "score": 1.0,
+                                            "content": "simplicial complex.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 10
+                                }
+                            ],
+                            "index": 9
+                        }
+                    ],
+                    "index": 6.25
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        172,
+                        506,
+                        216
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                172,
+                                506,
+                                184
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        172,
+                                        506,
+                                        184
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of two 2-dimensional disks (i.e. the two hemispheres) were glued to the equator to form a sphere.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                182,
+                                506,
+                                195
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        182,
+                                        506,
+                                        195
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The second example is a tetrahedron with empty interior. It is a particular type of cell complex",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                193,
+                                506,
+                                207
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        193,
+                                        506,
+                                        207
+                                    ],
+                                    "score": 1.0,
+                                    "content": "called a simplicial complex (SC). The only 2-cells it allows are triangle-shaped. More generally, the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                107,
+                                204,
+                                481,
+                                217
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        207,
+                                        114,
+                                        215
+                                    ],
+                                    "score": 0.71,
+                                    "content": "n",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        114,
+                                        204,
+                                        231,
+                                        217
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-dimensional cells of SCs are",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        232,
+                                        207,
+                                        239,
+                                        215
+                                    ],
+                                    "score": 0.74,
+                                    "content": "n",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        239,
+                                        204,
+                                        481,
+                                        217
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-simplices, which makes them slightly more rigid structures.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        }
+                    ],
+                    "index": 12.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        220,
+                        504,
+                        243
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                218,
+                                506,
+                                236
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        218,
+                                        295,
+                                        236
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Definition 2. The k-skeleton of a cell complex",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        296,
+                                        222,
+                                        306,
+                                        231
+                                    ],
+                                    "score": 0.69,
+                                    "content": "X",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        306,
+                                        218,
+                                        343,
+                                        236
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", denoted",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        343,
+                                        220,
+                                        364,
+                                        231
+                                    ],
+                                    "score": 0.9,
+                                    "content": "X ^ { ( k ) }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        364,
+                                        218,
+                                        451,
+                                        236
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", is the subcomplex of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        451,
+                                        222,
+                                        461,
+                                        231
+                                    ],
+                                    "score": 0.69,
+                                    "content": "X",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        461,
+                                        218,
+                                        506,
+                                        236
+                                    ],
+                                    "score": 1.0,
+                                    "content": "consisting",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                232,
+                                233,
+                                245
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        232,
+                                        222,
+                                        245
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of cells of dimension at most",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        223,
+                                        233,
+                                        229,
+                                        242
+                                    ],
+                                    "score": 0.66,
+                                    "content": "k",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        229,
+                                        232,
+                                        233,
+                                        245
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        }
+                    ],
+                    "index": 15.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        253,
+                        504,
+                        277
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                252,
+                                505,
+                                266
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        252,
+                                        433,
+                                        266
+                                    ],
+                                    "score": 1.0,
+                                    "content": "This definition is useful for referring for certain parts of the complex. For instance,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        433,
+                                        252,
+                                        454,
+                                        264
+                                    ],
+                                    "score": 0.89,
+                                    "content": "X ^ { ( 0 ) }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        454,
+                                        252,
+                                        505,
+                                        266
+                                    ],
+                                    "score": 1.0,
+                                    "content": "contains the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                263,
+                                501,
+                                279
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        263,
+                                        228,
+                                        279
+                                    ],
+                                    "score": 1.0,
+                                    "content": "vertices in the complex, while",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        228,
+                                        264,
+                                        249,
+                                        275
+                                    ],
+                                    "score": 0.89,
+                                    "content": "X ^ { ( 1 ) }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        249,
+                                        263,
+                                        501,
+                                        279
+                                    ],
+                                    "score": 1.0,
+                                    "content": "contains the vertices and the edges (i.e. the underlying graph).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        }
+                    ],
+                    "index": 17.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        281,
+                        505,
+                        327
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                282,
+                                505,
+                                295
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        282,
+                                        505,
+                                        295
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The combinatorial structure of the complex can be more compactly described by an incidence relation",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                293,
+                                505,
+                                305
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        293,
+                                        505,
+                                        305
+                                    ],
+                                    "score": 1.0,
+                                    "content": "we call the boundary relation, whose reflexive and transitive closure gives the partial order defined",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                303,
+                                507,
+                                317
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        303,
+                                        507,
+                                        317
+                                    ],
+                                    "score": 1.0,
+                                    "content": "above. The boundary relation describes what cells are on the boundary of other cells. For instance,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                314,
+                                505,
+                                328
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        315,
+                                        237,
+                                        328
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the edges of the sphere in Figure",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        238,
+                                        314,
+                                        247,
+                                        327
+                                    ],
+                                    "score": 0.5,
+                                    "content": "\\bigtriangledown",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        247,
+                                        315,
+                                        505,
+                                        328
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are on the boundary of the 2-cells forming the two hemispheres.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        }
+                    ],
+                    "index": 20.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        330,
+                        504,
+                        353
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                329,
+                                506,
+                                343
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        329,
+                                        297,
+                                        343
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Definition 3. We have the boundary relation",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        297,
+                                        331,
+                                        327,
+                                        341
+                                    ],
+                                    "score": 0.86,
+                                    "content": "\\sigma \\prec \\tau",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        327,
+                                        329,
+                                        340,
+                                        343
+                                    ],
+                                    "score": 1.0,
+                                    "content": "iff",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        340,
+                                        331,
+                                        370,
+                                        341
+                                    ],
+                                    "score": 0.88,
+                                    "content": "\\sigma < \\tau",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        371,
+                                        329,
+                                        456,
+                                        343
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and there is no cell",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        457,
+                                        331,
+                                        463,
+                                        340
+                                    ],
+                                    "score": 0.45,
+                                    "content": "\\delta",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        463,
+                                        329,
+                                        506,
+                                        343
+                                    ],
+                                    "score": 1.0,
+                                    "content": "such that",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                107,
+                                340,
+                                155,
+                                354
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        341,
+                                        150,
+                                        352
+                                    ],
+                                    "score": 0.9,
+                                    "content": "\\sigma < \\delta < \\tau",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        151,
+                                        340,
+                                        155,
+                                        354
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        }
+                    ],
+                    "index": 23.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        362,
+                        506,
+                        396
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                362,
+                                506,
+                                375
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        362,
+                                        506,
+                                        375
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We can use this to define the four types of (local) adjacencies present in cell complexes. These",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                373,
+                                505,
+                                386
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        373,
+                                        505,
+                                        386
+                                    ],
+                                    "score": 1.0,
+                                    "content": "adjacencies will be the fundamental building block of our message passing procedure. To explain",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                384,
+                                468,
+                                397
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        384,
+                                        468,
+                                        397
+                                    ],
+                                    "score": 1.0,
+                                    "content": "these in more familiar terms, for each adjacency, we exemplify how it shows up in graphs.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        }
+                    ],
+                    "index": 26
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        399,
+                        485,
+                        412
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                398,
+                                484,
+                                414
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        398,
+                                        350,
+                                        414
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Definition 4 (Cell complex adjacencies). For a cell complex",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        351,
+                                        400,
+                                        361,
+                                        410
+                                    ],
+                                    "score": 0.7,
+                                    "content": "X",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        361,
+                                        398,
+                                        403,
+                                        414
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and a cell",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        404,
+                                        400,
+                                        437,
+                                        411
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\sigma \\in P _ { X }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        437,
+                                        398,
+                                        484,
+                                        414
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", we define:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        }
+                    ],
+                    "index": 28
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        415,
+                        507,
+                        536
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                416,
+                                505,
+                                429
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        416,
+                                        235,
+                                        429
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1. The boundary adjacent cells",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        235,
+                                        416,
+                                        318,
+                                        428
+                                    ],
+                                    "score": 0.94,
+                                    "content": "\\boldsymbol { B } ( \\sigma ) = \\{ \\tau | \\tau \\prec \\sigma \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        319,
+                                        416,
+                                        505,
+                                        429
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". These are the lower-dimensional cells on the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                117,
+                                427,
+                                419,
+                                440
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        117,
+                                        427,
+                                        170,
+                                        440
+                                    ],
+                                    "score": 1.0,
+                                    "content": "boundary of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        170,
+                                        430,
+                                        177,
+                                        437
+                                    ],
+                                    "score": 0.27,
+                                    "content": "\\sigma",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        177,
+                                        427,
+                                        419,
+                                        440
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". For instance, the boundary cells of an edge are its vertices.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                441,
+                                506,
+                                454
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        441,
+                                        243,
+                                        454
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2. The co-boundary adjacent cell",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        243,
+                                        441,
+                                        325,
+                                        453
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\mathcal { C } ( \\sigma ) = \\{ \\tau \\ | \\ \\sigma \\prec \\tau \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        325,
+                                        441,
+                                        506,
+                                        454
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". These are the higher-dimensional cells with",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                118,
+                                453,
+                                496,
+                                464
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        118,
+                                        453,
+                                        496,
+                                        464
+                                    ],
+                                    "score": 1.0,
+                                    "content": "\u0000 on their boundary. For instance, the co-boundary cells of a vertex are the edges it is part of.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                466,
+                                506,
+                                479
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        466,
+                                        222,
+                                        479
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3. The lower adjacent cells",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        222,
+                                        466,
+                                        294,
+                                        479
+                                    ],
+                                    "score": 0.9,
+                                    "content": "\\mathcal { N } _ { \\downarrow } ( \\sigma ) = \\{ \\tau \\ | \\ \\exists \\delta",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        294,
+                                        466,
+                                        334,
+                                        479
+                                    ],
+                                    "score": 1.0,
+                                    "content": "such that",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        335,
+                                        466,
+                                        361,
+                                        477
+                                    ],
+                                    "score": 0.89,
+                                    "content": "\\delta \\prec \\sigma",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        361,
+                                        466,
+                                        380,
+                                        479
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        380,
+                                        466,
+                                        410,
+                                        478
+                                    ],
+                                    "score": 0.89,
+                                    "content": "\\delta \\prec \\tau \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        411,
+                                        466,
+                                        506,
+                                        479
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". These are the cells of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                118,
+                                477,
+                                505,
+                                490
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        118,
+                                        477,
+                                        212,
+                                        490
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the same dimension as",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        213,
+                                        480,
+                                        220,
+                                        487
+                                    ],
+                                    "score": 0.49,
+                                    "content": "\\sigma",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        220,
+                                        477,
+                                        505,
+                                        490
+                                    ],
+                                    "score": 1.0,
+                                    "content": "that share a lower dimensional cell on their boundary. The line graph",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                118,
+                                489,
+                                360,
+                                501
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        118,
+                                        489,
+                                        360,
+                                        501
+                                    ],
+                                    "score": 1.0,
+                                    "content": "adjacencies between the edges are a classic example of this.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                501,
+                                506,
+                                516
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        501,
+                                        222,
+                                        516
+                                    ],
+                                    "score": 1.0,
+                                    "content": "4. The upper adjacent cells",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        222,
+                                        502,
+                                        291,
+                                        515
+                                    ],
+                                    "score": 0.9,
+                                    "content": "\\mathcal { N } _ { \\uparrow } ( \\sigma ) = \\{ \\tau \\mid \\exists i",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        292,
+                                        501,
+                                        335,
+                                        516
+                                    ],
+                                    "score": 1.0,
+                                    "content": "such that",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        335,
+                                        502,
+                                        361,
+                                        513
+                                    ],
+                                    "score": 0.84,
+                                    "content": "\\sigma \\prec \\delta",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        362,
+                                        501,
+                                        380,
+                                        516
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        380,
+                                        502,
+                                        410,
+                                        514
+                                    ],
+                                    "score": 0.86,
+                                    "content": "\\tau \\prec \\delta \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        411,
+                                        501,
+                                        506,
+                                        516
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". These are the cells of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                118,
+                                514,
+                                506,
+                                526
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        118,
+                                        514,
+                                        211,
+                                        526
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the same dimension as",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        211,
+                                        516,
+                                        218,
+                                        523
+                                    ],
+                                    "score": 0.54,
+                                    "content": "\\sigma",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        219,
+                                        514,
+                                        476,
+                                        526
+                                    ],
+                                    "score": 1.0,
+                                    "content": "that are on the boundary of the same higher-dimensional cell as",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        476,
+                                        516,
+                                        483,
+                                        523
+                                    ],
+                                    "score": 0.56,
+                                    "content": "\\sigma",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        483,
+                                        514,
+                                        506,
+                                        526
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". The",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                118,
+                                524,
+                                422,
+                                537
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        118,
+                                        524,
+                                        422,
+                                        537
+                                    ],
+                                    "score": 1.0,
+                                    "content": "typical graph adjacencies between vertices are the canonical example here.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        }
+                    ],
+                    "index": 33.5
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        552,
+                        267,
+                        566
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                551,
+                                268,
+                                568
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        551,
+                                        268,
+                                        568
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3 Cellular Weisfeiler Lehman",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        }
+                    ],
+                    "index": 39
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        578,
+                        505,
+                        623
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                578,
+                                505,
+                                591
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        578,
+                                        505,
+                                        591
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Overview The results in this section show how one can transform graphs into higher-dimensional",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                590,
+                                505,
+                                601
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        590,
+                                        505,
+                                        601
+                                    ],
+                                    "score": 1.0,
+                                    "content": "cell complexes in such a way that performing colour refinement on the resulting cell complexes makes",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                599,
+                                507,
+                                615
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        599,
+                                        407,
+                                        615
+                                    ],
+                                    "score": 1.0,
+                                    "content": "it easier to test their isomorphism. The message passing model from Section",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        408,
+                                        600,
+                                        417,
+                                        613
+                                    ],
+                                    "score": 0.4,
+                                    "content": "\\bar { 4 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        417,
+                                        599,
+                                        507,
+                                        615
+                                    ],
+                                    "score": 1.0,
+                                    "content": "will take advantage of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                612,
+                                366,
+                                624
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        612,
+                                        366,
+                                        624
+                                    ],
+                                    "score": 1.0,
+                                    "content": "these theoretical results. All proofs can be found in Appendix A.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        }
+                    ],
+                    "index": 41.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        627,
+                        505,
+                        660
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                626,
+                                506,
+                                640
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        626,
+                                        344,
+                                        640
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Definition 5. Let c be a colouring of the cells in a complex",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        344,
+                                        628,
+                                        354,
+                                        637
+                                    ],
+                                    "score": 0.8,
+                                    "content": "X",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        355,
+                                        626,
+                                        375,
+                                        640
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        375,
+                                        628,
+                                        386,
+                                        638
+                                    ],
+                                    "score": 0.83,
+                                    "content": "c _ { \\sigma }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        387,
+                                        626,
+                                        506,
+                                        640
+                                    ],
+                                    "score": 1.0,
+                                    "content": "denoting the colour assigned",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                637,
+                                505,
+                                651
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        637,
+                                        133,
+                                        651
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to cell",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        133,
+                                        639,
+                                        165,
+                                        649
+                                    ],
+                                    "score": 0.9,
+                                    "content": "\\sigma \\in P _ { X }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        166,
+                                        637,
+                                        198,
+                                        651
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Define",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        198,
+                                        638,
+                                        295,
+                                        650
+                                    ],
+                                    "score": 0.92,
+                                    "content": "B ( \\sigma , \\tau ) : = B ( \\sigma ) \\cap B ( \\tau )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        296,
+                                        637,
+                                        313,
+                                        651
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        314,
+                                        638,
+                                        407,
+                                        650
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\mathcal { C } ( \\sigma , \\tau ) : = \\mathcal { C } ( \\sigma ) \\cap \\mathcal { C } ( \\tau )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        408,
+                                        637,
+                                        505,
+                                        651
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". We define the following",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                648,
+                                194,
+                                662
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        648,
+                                        194,
+                                        662
+                                    ],
+                                    "score": 1.0,
+                                    "content": "multi-sets of colours:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        }
+                    ],
+                    "index": 45
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        664,
+                        457,
+                        721
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                664,
+                                389,
+                                678
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        664,
+                                        265,
+                                        678
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1. The colours of the boundary cells of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        266,
+                                        668,
+                                        272,
+                                        675
+                                    ],
+                                    "score": 0.43,
+                                    "content": "\\sigma",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        273,
+                                        664,
+                                        278,
+                                        678
+                                    ],
+                                    "score": 1.0,
+                                    "content": ":",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        278,
+                                        665,
+                                        387,
+                                        678
+                                    ],
+                                    "score": 0.87,
+                                    "content": "c _ { B } ( \\sigma ) = \\{ \\{ c _ { \\tau } \\mid \\tau \\in B ( \\sigma ) \\} \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        387,
+                                        664,
+                                        389,
+                                        678
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 47
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                678,
+                                401,
+                                692
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        678,
+                                        278,
+                                        692
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2. The colours of the co-boundary cells of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        278,
+                                        682,
+                                        285,
+                                        689
+                                    ],
+                                    "score": 0.52,
+                                    "content": "\\sigma",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        285,
+                                        678,
+                                        291,
+                                        692
+                                    ],
+                                    "score": 1.0,
+                                    "content": ":",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        291,
+                                        679,
+                                        397,
+                                        692
+                                    ],
+                                    "score": 0.88,
+                                    "content": "c _ { \\mathcal { C } } ( \\sigma ) = \\{ \\{ c _ { \\tau } ~ | ~ \\tau \\in \\mathcal { C } ( \\sigma ) \\} \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        398,
+                                        678,
+                                        401,
+                                        692
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 48
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                693,
+                                458,
+                                707
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        693,
+                                        241,
+                                        707
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3. The lower adjacent colours of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        241,
+                                        696,
+                                        248,
+                                        704
+                                    ],
+                                    "score": 0.52,
+                                    "content": "\\sigma",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        248,
+                                        693,
+                                        254,
+                                        707
+                                    ],
+                                    "score": 1.0,
+                                    "content": ":",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        254,
+                                        693,
+                                        380,
+                                        706
+                                    ],
+                                    "score": 0.89,
+                                    "content": "c _ { \\downarrow } ( \\sigma ) = \\{ \\{ ( c _ { \\tau } , c _ { \\delta } ) \\mid \\tau \\in \\mathcal { N } _ { \\downarrow } ( \\sigma )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        380,
+                                        693,
+                                        399,
+                                        707
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        399,
+                                        694,
+                                        454,
+                                        706
+                                    ],
+                                    "score": 0.9,
+                                    "content": "\\delta \\in B ( \\sigma , \\tau ) \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        454,
+                                        693,
+                                        458,
+                                        707
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 49
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                706,
+                                457,
+                                720
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        706,
+                                        241,
+                                        720
+                                    ],
+                                    "score": 1.0,
+                                    "content": "4. The upper adjacent colours of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        242,
+                                        711,
+                                        248,
+                                        718
+                                    ],
+                                    "score": 0.57,
+                                    "content": "\\sigma",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        249,
+                                        706,
+                                        254,
+                                        720
+                                    ],
+                                    "score": 1.0,
+                                    "content": ":",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        254,
+                                        708,
+                                        380,
+                                        720
+                                    ],
+                                    "score": 0.88,
+                                    "content": "c _ { \\uparrow } ( \\sigma ) = \\{ \\{ ( c _ { \\tau } , c _ { \\delta } ) \\mid \\tau \\in \\mathcal { N } _ { \\uparrow } ( \\sigma )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        381,
+                                        706,
+                                        400,
+                                        720
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        400,
+                                        707,
+                                        453,
+                                        720
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\delta \\in \\mathcal { C } ( \\sigma , \\tau ) \\big \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        454,
+                                        706,
+                                        457,
+                                        720
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 50
+                        }
+                    ],
+                    "index": 48.5
+                }
+            ],
+            "page_idx": 2,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        302,
+                        741,
+                        309,
+                        750
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                301,
+                                740,
+                                310,
+                                752
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        301,
+                                        740,
+                                        310,
+                                        752
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "image",
+                    "bbox": [
+                        107,
+                        70,
+                        358,
+                        120
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                107,
+                                70,
+                                358,
+                                120
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        70,
+                                        358,
+                                        120
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                107,
+                                                70,
+                                                358,
+                                                120
+                                            ],
+                                            "score": 0.961,
+                                            "type": "image",
+                                            "image_path": "a55263d0b752b3bc5d806b470b3759825fb4eb9c7d2ae5d1f90223fef4cc5a08.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 1,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        70,
+                                        358,
+                                        86.66666666666667
+                                    ],
+                                    "spans": [],
+                                    "index": 0
+                                },
+                                {
+                                    "bbox": [
+                                        107,
+                                        86.66666666666667,
+                                        358,
+                                        103.33333333333334
+                                    ],
+                                    "spans": [],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        107,
+                                        103.33333333333334,
+                                        358,
+                                        120.00000000000001
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                106,
+                                125,
+                                359,
+                                159
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        125,
+                                        360,
+                                        137
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                125,
+                                                360,
+                                                137
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 2: Closed two-dimensional disks are glued to the bound-",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 5
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        136,
+                                        358,
+                                        148
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                136,
+                                                358,
+                                                148
+                                            ],
+                                            "score": 1.0,
+                                            "content": "ary of the rings present in the graph (left). The result is a 2D",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 6
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        147,
+                                        221,
+                                        160
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                147,
+                                                221,
+                                                160
+                                            ],
+                                            "score": 1.0,
+                                            "content": "regular cell complex (right).",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 7
+                                }
+                            ],
+                            "index": 6
+                        }
+                    ],
+                    "index": 3.5
+                },
+                {
+                    "type": "image",
+                    "bbox": [
+                        381,
+                        71,
+                        490,
+                        120
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                381,
+                                71,
+                                490,
+                                120
+                            ],
+                            "group_id": 1,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        381,
+                                        71,
+                                        490,
+                                        120
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                381,
+                                                71,
+                                                490,
+                                                120
+                                            ],
+                                            "score": 0.935,
+                                            "type": "image",
+                                            "image_path": "0b72066b35dac85fbfc07fd956eee7a5fee05fd86266a0ea383297003ddffad9.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 3.5,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        381,
+                                        71,
+                                        490,
+                                        95.5
+                                    ],
+                                    "spans": [],
+                                    "index": 3
+                                },
+                                {
+                                    "bbox": [
+                                        381,
+                                        95.5,
+                                        490,
+                                        120.0
+                                    ],
+                                    "spans": [],
+                                    "index": 4
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                369,
+                                126,
+                                502,
+                                159
+                            ],
+                            "group_id": 1,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        368,
+                                        124,
+                                        502,
+                                        138
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                368,
+                                                124,
+                                                502,
+                                                138
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 3: A sphere and an empty",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 8
+                                },
+                                {
+                                    "bbox": [
+                                        369,
+                                        136,
+                                        502,
+                                        148
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                369,
+                                                136,
+                                                502,
+                                                148
+                                            ],
+                                            "score": 1.0,
+                                            "content": "tetrahedron. The latter is also a",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 9
+                                },
+                                {
+                                    "bbox": [
+                                        369,
+                                        147,
+                                        450,
+                                        160
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                369,
+                                                147,
+                                                450,
+                                                160
+                                            ],
+                                            "score": 1.0,
+                                            "content": "simplicial complex.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 10
+                                }
+                            ],
+                            "index": 9
+                        }
+                    ],
+                    "index": 6.25
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        172,
+                        506,
+                        216
+                    ],
+                    "lines": [],
+                    "index": 12.5,
+                    "bbox_fs": [
+                        105,
+                        172,
+                        506,
+                        217
+                    ],
+                    "lines_deleted": true
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        220,
+                        504,
+                        243
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                218,
+                                506,
+                                236
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        218,
+                                        295,
+                                        236
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Definition 2. The k-skeleton of a cell complex",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        296,
+                                        222,
+                                        306,
+                                        231
+                                    ],
+                                    "score": 0.69,
+                                    "content": "X",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        306,
+                                        218,
+                                        343,
+                                        236
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", denoted",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        343,
+                                        220,
+                                        364,
+                                        231
+                                    ],
+                                    "score": 0.9,
+                                    "content": "X ^ { ( k ) }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        364,
+                                        218,
+                                        451,
+                                        236
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", is the subcomplex of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        451,
+                                        222,
+                                        461,
+                                        231
+                                    ],
+                                    "score": 0.69,
+                                    "content": "X",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        461,
+                                        218,
+                                        506,
+                                        236
+                                    ],
+                                    "score": 1.0,
+                                    "content": "consisting",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                232,
+                                233,
+                                245
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        232,
+                                        222,
+                                        245
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of cells of dimension at most",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        223,
+                                        233,
+                                        229,
+                                        242
+                                    ],
+                                    "score": 0.66,
+                                    "content": "k",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        229,
+                                        232,
+                                        233,
+                                        245
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        }
+                    ],
+                    "index": 15.5,
+                    "bbox_fs": [
+                        105,
+                        218,
+                        506,
+                        245
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        253,
+                        504,
+                        277
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                252,
+                                505,
+                                266
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        252,
+                                        433,
+                                        266
+                                    ],
+                                    "score": 1.0,
+                                    "content": "This definition is useful for referring for certain parts of the complex. For instance,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        433,
+                                        252,
+                                        454,
+                                        264
+                                    ],
+                                    "score": 0.89,
+                                    "content": "X ^ { ( 0 ) }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        454,
+                                        252,
+                                        505,
+                                        266
+                                    ],
+                                    "score": 1.0,
+                                    "content": "contains the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                263,
+                                501,
+                                279
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        263,
+                                        228,
+                                        279
+                                    ],
+                                    "score": 1.0,
+                                    "content": "vertices in the complex, while",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        228,
+                                        264,
+                                        249,
+                                        275
+                                    ],
+                                    "score": 0.89,
+                                    "content": "X ^ { ( 1 ) }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        249,
+                                        263,
+                                        501,
+                                        279
+                                    ],
+                                    "score": 1.0,
+                                    "content": "contains the vertices and the edges (i.e. the underlying graph).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        }
+                    ],
+                    "index": 17.5,
+                    "bbox_fs": [
+                        105,
+                        252,
+                        505,
+                        279
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        281,
+                        505,
+                        327
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                282,
+                                505,
+                                295
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        282,
+                                        505,
+                                        295
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The combinatorial structure of the complex can be more compactly described by an incidence relation",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                293,
+                                505,
+                                305
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        293,
+                                        505,
+                                        305
+                                    ],
+                                    "score": 1.0,
+                                    "content": "we call the boundary relation, whose reflexive and transitive closure gives the partial order defined",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                303,
+                                507,
+                                317
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        303,
+                                        507,
+                                        317
+                                    ],
+                                    "score": 1.0,
+                                    "content": "above. The boundary relation describes what cells are on the boundary of other cells. For instance,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                314,
+                                505,
+                                328
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        315,
+                                        237,
+                                        328
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the edges of the sphere in Figure",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        238,
+                                        314,
+                                        247,
+                                        327
+                                    ],
+                                    "score": 0.5,
+                                    "content": "\\bigtriangledown",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        247,
+                                        315,
+                                        505,
+                                        328
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are on the boundary of the 2-cells forming the two hemispheres.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        }
+                    ],
+                    "index": 20.5,
+                    "bbox_fs": [
+                        105,
+                        282,
+                        507,
+                        328
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        330,
+                        504,
+                        353
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                329,
+                                506,
+                                343
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        329,
+                                        297,
+                                        343
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Definition 3. We have the boundary relation",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        297,
+                                        331,
+                                        327,
+                                        341
+                                    ],
+                                    "score": 0.86,
+                                    "content": "\\sigma \\prec \\tau",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        327,
+                                        329,
+                                        340,
+                                        343
+                                    ],
+                                    "score": 1.0,
+                                    "content": "iff",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        340,
+                                        331,
+                                        370,
+                                        341
+                                    ],
+                                    "score": 0.88,
+                                    "content": "\\sigma < \\tau",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        371,
+                                        329,
+                                        456,
+                                        343
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and there is no cell",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        457,
+                                        331,
+                                        463,
+                                        340
+                                    ],
+                                    "score": 0.45,
+                                    "content": "\\delta",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        463,
+                                        329,
+                                        506,
+                                        343
+                                    ],
+                                    "score": 1.0,
+                                    "content": "such that",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                107,
+                                340,
+                                155,
+                                354
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        341,
+                                        150,
+                                        352
+                                    ],
+                                    "score": 0.9,
+                                    "content": "\\sigma < \\delta < \\tau",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        151,
+                                        340,
+                                        155,
+                                        354
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        }
+                    ],
+                    "index": 23.5,
+                    "bbox_fs": [
+                        105,
+                        329,
+                        506,
+                        354
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        362,
+                        506,
+                        396
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                362,
+                                506,
+                                375
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        362,
+                                        506,
+                                        375
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We can use this to define the four types of (local) adjacencies present in cell complexes. These",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                373,
+                                505,
+                                386
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        373,
+                                        505,
+                                        386
+                                    ],
+                                    "score": 1.0,
+                                    "content": "adjacencies will be the fundamental building block of our message passing procedure. To explain",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                384,
+                                468,
+                                397
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        384,
+                                        468,
+                                        397
+                                    ],
+                                    "score": 1.0,
+                                    "content": "these in more familiar terms, for each adjacency, we exemplify how it shows up in graphs.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        }
+                    ],
+                    "index": 26,
+                    "bbox_fs": [
+                        105,
+                        362,
+                        506,
+                        397
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        399,
+                        485,
+                        412
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                398,
+                                484,
+                                414
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        398,
+                                        350,
+                                        414
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Definition 4 (Cell complex adjacencies). For a cell complex",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        351,
+                                        400,
+                                        361,
+                                        410
+                                    ],
+                                    "score": 0.7,
+                                    "content": "X",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        361,
+                                        398,
+                                        403,
+                                        414
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and a cell",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        404,
+                                        400,
+                                        437,
+                                        411
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\sigma \\in P _ { X }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        437,
+                                        398,
+                                        484,
+                                        414
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", we define:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        }
+                    ],
+                    "index": 28,
+                    "bbox_fs": [
+                        105,
+                        398,
+                        484,
+                        414
+                    ]
+                },
+                {
+                    "type": "list",
+                    "bbox": [
+                        106,
+                        415,
+                        507,
+                        536
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                416,
+                                505,
+                                429
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        416,
+                                        235,
+                                        429
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1. The boundary adjacent cells",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        235,
+                                        416,
+                                        318,
+                                        428
+                                    ],
+                                    "score": 0.94,
+                                    "content": "\\boldsymbol { B } ( \\sigma ) = \\{ \\tau | \\tau \\prec \\sigma \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        319,
+                                        416,
+                                        505,
+                                        429
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". These are the lower-dimensional cells on the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                117,
+                                427,
+                                419,
+                                440
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        117,
+                                        427,
+                                        170,
+                                        440
+                                    ],
+                                    "score": 1.0,
+                                    "content": "boundary of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        170,
+                                        430,
+                                        177,
+                                        437
+                                    ],
+                                    "score": 0.27,
+                                    "content": "\\sigma",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        177,
+                                        427,
+                                        419,
+                                        440
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". For instance, the boundary cells of an edge are its vertices.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                441,
+                                506,
+                                454
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        441,
+                                        243,
+                                        454
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2. The co-boundary adjacent cell",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        243,
+                                        441,
+                                        325,
+                                        453
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\mathcal { C } ( \\sigma ) = \\{ \\tau \\ | \\ \\sigma \\prec \\tau \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        325,
+                                        441,
+                                        506,
+                                        454
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". These are the higher-dimensional cells with",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                118,
+                                453,
+                                496,
+                                464
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        118,
+                                        453,
+                                        496,
+                                        464
+                                    ],
+                                    "score": 1.0,
+                                    "content": "\u0000 on their boundary. For instance, the co-boundary cells of a vertex are the edges it is part of.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                466,
+                                506,
+                                479
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        466,
+                                        222,
+                                        479
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3. The lower adjacent cells",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        222,
+                                        466,
+                                        294,
+                                        479
+                                    ],
+                                    "score": 0.9,
+                                    "content": "\\mathcal { N } _ { \\downarrow } ( \\sigma ) = \\{ \\tau \\ | \\ \\exists \\delta",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        294,
+                                        466,
+                                        334,
+                                        479
+                                    ],
+                                    "score": 1.0,
+                                    "content": "such that",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        335,
+                                        466,
+                                        361,
+                                        477
+                                    ],
+                                    "score": 0.89,
+                                    "content": "\\delta \\prec \\sigma",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        361,
+                                        466,
+                                        380,
+                                        479
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        380,
+                                        466,
+                                        410,
+                                        478
+                                    ],
+                                    "score": 0.89,
+                                    "content": "\\delta \\prec \\tau \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        411,
+                                        466,
+                                        506,
+                                        479
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". These are the cells of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                118,
+                                477,
+                                505,
+                                490
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        118,
+                                        477,
+                                        212,
+                                        490
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the same dimension as",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        213,
+                                        480,
+                                        220,
+                                        487
+                                    ],
+                                    "score": 0.49,
+                                    "content": "\\sigma",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        220,
+                                        477,
+                                        505,
+                                        490
+                                    ],
+                                    "score": 1.0,
+                                    "content": "that share a lower dimensional cell on their boundary. The line graph",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                118,
+                                489,
+                                360,
+                                501
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        118,
+                                        489,
+                                        360,
+                                        501
+                                    ],
+                                    "score": 1.0,
+                                    "content": "adjacencies between the edges are a classic example of this.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                501,
+                                506,
+                                516
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        501,
+                                        222,
+                                        516
+                                    ],
+                                    "score": 1.0,
+                                    "content": "4. The upper adjacent cells",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        222,
+                                        502,
+                                        291,
+                                        515
+                                    ],
+                                    "score": 0.9,
+                                    "content": "\\mathcal { N } _ { \\uparrow } ( \\sigma ) = \\{ \\tau \\mid \\exists i",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        292,
+                                        501,
+                                        335,
+                                        516
+                                    ],
+                                    "score": 1.0,
+                                    "content": "such that",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        335,
+                                        502,
+                                        361,
+                                        513
+                                    ],
+                                    "score": 0.84,
+                                    "content": "\\sigma \\prec \\delta",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        362,
+                                        501,
+                                        380,
+                                        516
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        380,
+                                        502,
+                                        410,
+                                        514
+                                    ],
+                                    "score": 0.86,
+                                    "content": "\\tau \\prec \\delta \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        411,
+                                        501,
+                                        506,
+                                        516
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". These are the cells of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                118,
+                                514,
+                                506,
+                                526
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        118,
+                                        514,
+                                        211,
+                                        526
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the same dimension as",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        211,
+                                        516,
+                                        218,
+                                        523
+                                    ],
+                                    "score": 0.54,
+                                    "content": "\\sigma",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        219,
+                                        514,
+                                        476,
+                                        526
+                                    ],
+                                    "score": 1.0,
+                                    "content": "that are on the boundary of the same higher-dimensional cell as",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        476,
+                                        516,
+                                        483,
+                                        523
+                                    ],
+                                    "score": 0.56,
+                                    "content": "\\sigma",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        483,
+                                        514,
+                                        506,
+                                        526
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". The",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                118,
+                                524,
+                                422,
+                                537
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        118,
+                                        524,
+                                        422,
+                                        537
+                                    ],
+                                    "score": 1.0,
+                                    "content": "typical graph adjacencies between vertices are the canonical example here.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38,
+                            "is_list_end_line": true
+                        }
+                    ],
+                    "index": 33.5,
+                    "bbox_fs": [
+                        104,
+                        416,
+                        506,
+                        537
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        552,
+                        267,
+                        566
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                551,
+                                268,
+                                568
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        551,
+                                        268,
+                                        568
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3 Cellular Weisfeiler Lehman",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        }
+                    ],
+                    "index": 39
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        578,
+                        505,
+                        623
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                578,
+                                505,
+                                591
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        578,
+                                        505,
+                                        591
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Overview The results in this section show how one can transform graphs into higher-dimensional",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                590,
+                                505,
+                                601
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        590,
+                                        505,
+                                        601
+                                    ],
+                                    "score": 1.0,
+                                    "content": "cell complexes in such a way that performing colour refinement on the resulting cell complexes makes",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                599,
+                                507,
+                                615
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        599,
+                                        407,
+                                        615
+                                    ],
+                                    "score": 1.0,
+                                    "content": "it easier to test their isomorphism. The message passing model from Section",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        408,
+                                        600,
+                                        417,
+                                        613
+                                    ],
+                                    "score": 0.4,
+                                    "content": "\\bar { 4 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        417,
+                                        599,
+                                        507,
+                                        615
+                                    ],
+                                    "score": 1.0,
+                                    "content": "will take advantage of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                612,
+                                366,
+                                624
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        612,
+                                        366,
+                                        624
+                                    ],
+                                    "score": 1.0,
+                                    "content": "these theoretical results. All proofs can be found in Appendix A.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        }
+                    ],
+                    "index": 41.5,
+                    "bbox_fs": [
+                        104,
+                        578,
+                        507,
+                        624
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        627,
+                        505,
+                        660
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                626,
+                                506,
+                                640
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        626,
+                                        344,
+                                        640
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Definition 5. Let c be a colouring of the cells in a complex",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        344,
+                                        628,
+                                        354,
+                                        637
+                                    ],
+                                    "score": 0.8,
+                                    "content": "X",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        355,
+                                        626,
+                                        375,
+                                        640
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        375,
+                                        628,
+                                        386,
+                                        638
+                                    ],
+                                    "score": 0.83,
+                                    "content": "c _ { \\sigma }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        387,
+                                        626,
+                                        506,
+                                        640
+                                    ],
+                                    "score": 1.0,
+                                    "content": "denoting the colour assigned",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                637,
+                                505,
+                                651
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        637,
+                                        133,
+                                        651
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to cell",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        133,
+                                        639,
+                                        165,
+                                        649
+                                    ],
+                                    "score": 0.9,
+                                    "content": "\\sigma \\in P _ { X }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        166,
+                                        637,
+                                        198,
+                                        651
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Define",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        198,
+                                        638,
+                                        295,
+                                        650
+                                    ],
+                                    "score": 0.92,
+                                    "content": "B ( \\sigma , \\tau ) : = B ( \\sigma ) \\cap B ( \\tau )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        296,
+                                        637,
+                                        313,
+                                        651
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        314,
+                                        638,
+                                        407,
+                                        650
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\mathcal { C } ( \\sigma , \\tau ) : = \\mathcal { C } ( \\sigma ) \\cap \\mathcal { C } ( \\tau )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        408,
+                                        637,
+                                        505,
+                                        651
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". We define the following",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                648,
+                                194,
+                                662
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        648,
+                                        194,
+                                        662
+                                    ],
+                                    "score": 1.0,
+                                    "content": "multi-sets of colours:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        }
+                    ],
+                    "index": 45,
+                    "bbox_fs": [
+                        105,
+                        626,
+                        506,
+                        662
+                    ]
+                },
+                {
+                    "type": "index",
+                    "bbox": [
+                        105,
+                        664,
+                        457,
+                        721
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                664,
+                                389,
+                                678
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        664,
+                                        265,
+                                        678
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1. The colours of the boundary cells of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        266,
+                                        668,
+                                        272,
+                                        675
+                                    ],
+                                    "score": 0.43,
+                                    "content": "\\sigma",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        273,
+                                        664,
+                                        278,
+                                        678
+                                    ],
+                                    "score": 1.0,
+                                    "content": ":",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        278,
+                                        665,
+                                        387,
+                                        678
+                                    ],
+                                    "score": 0.87,
+                                    "content": "c _ { B } ( \\sigma ) = \\{ \\{ c _ { \\tau } \\mid \\tau \\in B ( \\sigma ) \\} \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        387,
+                                        664,
+                                        389,
+                                        678
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 47,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                678,
+                                401,
+                                692
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        678,
+                                        278,
+                                        692
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2. The colours of the co-boundary cells of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        278,
+                                        682,
+                                        285,
+                                        689
+                                    ],
+                                    "score": 0.52,
+                                    "content": "\\sigma",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        285,
+                                        678,
+                                        291,
+                                        692
+                                    ],
+                                    "score": 1.0,
+                                    "content": ":",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        291,
+                                        679,
+                                        397,
+                                        692
+                                    ],
+                                    "score": 0.88,
+                                    "content": "c _ { \\mathcal { C } } ( \\sigma ) = \\{ \\{ c _ { \\tau } ~ | ~ \\tau \\in \\mathcal { C } ( \\sigma ) \\} \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        398,
+                                        678,
+                                        401,
+                                        692
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 48,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                693,
+                                458,
+                                707
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        693,
+                                        241,
+                                        707
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3. The lower adjacent colours of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        241,
+                                        696,
+                                        248,
+                                        704
+                                    ],
+                                    "score": 0.52,
+                                    "content": "\\sigma",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        248,
+                                        693,
+                                        254,
+                                        707
+                                    ],
+                                    "score": 1.0,
+                                    "content": ":",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        254,
+                                        693,
+                                        380,
+                                        706
+                                    ],
+                                    "score": 0.89,
+                                    "content": "c _ { \\downarrow } ( \\sigma ) = \\{ \\{ ( c _ { \\tau } , c _ { \\delta } ) \\mid \\tau \\in \\mathcal { N } _ { \\downarrow } ( \\sigma )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        380,
+                                        693,
+                                        399,
+                                        707
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        399,
+                                        694,
+                                        454,
+                                        706
+                                    ],
+                                    "score": 0.9,
+                                    "content": "\\delta \\in B ( \\sigma , \\tau ) \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        454,
+                                        693,
+                                        458,
+                                        707
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 49,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                706,
+                                457,
+                                720
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        706,
+                                        241,
+                                        720
+                                    ],
+                                    "score": 1.0,
+                                    "content": "4. The upper adjacent colours of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        242,
+                                        711,
+                                        248,
+                                        718
+                                    ],
+                                    "score": 0.57,
+                                    "content": "\\sigma",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        249,
+                                        706,
+                                        254,
+                                        720
+                                    ],
+                                    "score": 1.0,
+                                    "content": ":",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        254,
+                                        708,
+                                        380,
+                                        720
+                                    ],
+                                    "score": 0.88,
+                                    "content": "c _ { \\uparrow } ( \\sigma ) = \\{ \\{ ( c _ { \\tau } , c _ { \\delta } ) \\mid \\tau \\in \\mathcal { N } _ { \\uparrow } ( \\sigma )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        381,
+                                        706,
+                                        400,
+                                        720
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        400,
+                                        707,
+                                        453,
+                                        720
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\delta \\in \\mathcal { C } ( \\sigma , \\tau ) \\big \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        454,
+                                        706,
+                                        457,
+                                        720
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 50,
+                            "is_list_start_line": true
+                        }
+                    ],
+                    "index": 48.5,
+                    "bbox_fs": [
+                        105,
+                        664,
+                        458,
+                        720
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "image",
+                    "bbox": [
+                        128,
+                        71,
+                        485,
+                        154
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                128,
+                                71,
+                                485,
+                                154
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        128,
+                                        71,
+                                        485,
+                                        154
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                128,
+                                                71,
+                                                485,
+                                                154
+                                            ],
+                                            "score": 0.931,
+                                            "type": "image",
+                                            "image_path": "f132e64580feb98e0dddba934ea49e01c6fa4a4cc004b0c67d64ceaec82753eb.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 1,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        128,
+                                        71,
+                                        485,
+                                        98.66666666666667
+                                    ],
+                                    "spans": [],
+                                    "index": 0
+                                },
+                                {
+                                    "bbox": [
+                                        128,
+                                        98.66666666666667,
+                                        485,
+                                        126.33333333333334
+                                    ],
+                                    "spans": [],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        128,
+                                        126.33333333333334,
+                                        485,
+                                        154.0
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                106,
+                                160,
+                                503,
+                                195
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        161,
+                                        505,
+                                        173
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                161,
+                                                505,
+                                                173
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 4: The CWL colouring procedure for the yellow edge of the cell complex. All cells have",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 3
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        171,
+                                        505,
+                                        186
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                171,
+                                                505,
+                                                186
+                                            ],
+                                            "score": 1.0,
+                                            "content": "been assigned unique colours to aid the visualisation of the adjacencies. Note that the yellow edge",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 4
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        183,
+                                        352,
+                                        196
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                183,
+                                                352,
+                                                196
+                                            ],
+                                            "score": 1.0,
+                                            "content": "aggregates long-range information from the light green edge.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 5
+                                }
+                            ],
+                            "index": 4
+                        }
+                    ],
+                    "index": 2.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        213,
+                        505,
+                        257
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                212,
+                                505,
+                                226
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        212,
+                                        360,
+                                        226
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Note that unlike in graphs and simplicial complexes, the sets",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        360,
+                                        213,
+                                        391,
+                                        225
+                                    ],
+                                    "score": 0.93,
+                                    "content": "B ( \\sigma , \\tau )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        391,
+                                        212,
+                                        410,
+                                        226
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        411,
+                                        213,
+                                        441,
+                                        225
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\mathcal { C } ( \\sigma , \\tau )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        441,
+                                        212,
+                                        505,
+                                        226
+                                    ],
+                                    "score": 1.0,
+                                    "content": "can have more",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                224,
+                                505,
+                                236
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        224,
+                                        505,
+                                        236
+                                    ],
+                                    "score": 1.0,
+                                    "content": "than one element. For instance, two (closed) 2-cells might intersect in more than one edge (e.g. the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                234,
+                                506,
+                                249
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        234,
+                                        213,
+                                        249
+                                    ],
+                                    "score": 1.0,
+                                    "content": "two hemispheres in Figure",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        213,
+                                        234,
+                                        224,
+                                        248
+                                    ],
+                                    "score": 0.62,
+                                    "content": "\\textcircled { 3 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        225,
+                                        234,
+                                        506,
+                                        249
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", and conversely, two edges might be on the boundary of the same two",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                245,
+                                440,
+                                258
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        245,
+                                        440,
+                                        258
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2-cells. This illustrates the more flexible combinatorial structure of cell complexes.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        }
+                    ],
+                    "index": 7.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        268,
+                        505,
+                        313
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                267,
+                                506,
+                                281
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        267,
+                                        506,
+                                        281
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Cellular WL (CWL) We consider CWL, a colour refinement scheme for cell complexes that",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                278,
+                                506,
+                                292
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        278,
+                                        232,
+                                        292
+                                    ],
+                                    "score": 1.0,
+                                    "content": "generalises the Simplicial WL",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        232,
+                                        279,
+                                        245,
+                                        290
+                                    ],
+                                    "score": 0.63,
+                                    "content": "\\textcircled { 8 } \\textcircled { 1 8 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        245,
+                                        278,
+                                        282,
+                                        292
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and WL",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        282,
+                                        279,
+                                        300,
+                                        290
+                                    ],
+                                    "score": 0.53,
+                                    "content": "\\mathbb { \\ m }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        300,
+                                        278,
+                                        356,
+                                        292
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tests. We use",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        357,
+                                        279,
+                                        367,
+                                        291
+                                    ],
+                                    "score": 0.88,
+                                    "content": "c _ { \\sigma } ^ { t }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        368,
+                                        278,
+                                        506,
+                                        292
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to refer to the colour assigned by",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                289,
+                                505,
+                                302
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        289,
+                                        158,
+                                        302
+                                    ],
+                                    "score": 1.0,
+                                    "content": "CWL to cell",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        159,
+                                        291,
+                                        166,
+                                        300
+                                    ],
+                                    "score": 0.78,
+                                    "content": "\\sigma",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        167,
+                                        289,
+                                        213,
+                                        302
+                                    ],
+                                    "score": 1.0,
+                                    "content": "at iteration",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        213,
+                                        291,
+                                        218,
+                                        300
+                                    ],
+                                    "score": 0.74,
+                                    "content": "t",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        218,
+                                        289,
+                                        505,
+                                        302
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of the algorithm. When the input is a simplicial complex, this recovers",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                300,
+                                489,
+                                314
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        301,
+                                        414,
+                                        314
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the SWL algorithm. A step of the algorithm is graphically depicted in Figure",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        414,
+                                        300,
+                                        424,
+                                        314
+                                    ],
+                                    "score": 0.47,
+                                    "content": "^ 4",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        424,
+                                        301,
+                                        489,
+                                        314
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for a single cell.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        }
+                    ],
+                    "index": 11.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        319,
+                        508,
+                        394
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                320,
+                                450,
+                                332
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        320,
+                                        236,
+                                        332
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1. Given a regular cell complex",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        236,
+                                        320,
+                                        246,
+                                        330
+                                    ],
+                                    "score": 0.79,
+                                    "content": "X",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        247,
+                                        320,
+                                        298,
+                                        332
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", all the cells",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        298,
+                                        322,
+                                        306,
+                                        330
+                                    ],
+                                    "score": 0.79,
+                                    "content": "\\sigma",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        306,
+                                        320,
+                                        450,
+                                        332
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are initialised with the same colour.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                333,
+                                505,
+                                347
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        333,
+                                        189,
+                                        347
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2. Given the colour",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        189,
+                                        333,
+                                        200,
+                                        346
+                                    ],
+                                    "score": 0.89,
+                                    "content": "c _ { \\sigma } ^ { t }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        200,
+                                        333,
+                                        229,
+                                        347
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of cell",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        229,
+                                        336,
+                                        237,
+                                        344
+                                    ],
+                                    "score": 0.77,
+                                    "content": "\\sigma",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        237,
+                                        333,
+                                        284,
+                                        347
+                                    ],
+                                    "score": 1.0,
+                                    "content": "at iteration",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        284,
+                                        335,
+                                        289,
+                                        344
+                                    ],
+                                    "score": 0.73,
+                                    "content": "t",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        289,
+                                        333,
+                                        415,
+                                        347
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", we compute the colour of cell",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        416,
+                                        336,
+                                        423,
+                                        344
+                                    ],
+                                    "score": 0.75,
+                                    "content": "\\sigma",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        424,
+                                        333,
+                                        505,
+                                        347
+                                    ],
+                                    "score": 1.0,
+                                    "content": "at the next iteration",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                119,
+                                340,
+                                509,
+                                363
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        119,
+                                        344,
+                                        138,
+                                        357
+                                    ],
+                                    "score": 0.91,
+                                    "content": "c _ { \\sigma } ^ { t + 1 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        138,
+                                        340,
+                                        465,
+                                        363
+                                    ],
+                                    "score": 1.0,
+                                    "content": "by injectively mapping the multi-sets of colours belonging to the adjacent cells of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        466,
+                                        347,
+                                        473,
+                                        354
+                                    ],
+                                    "score": 0.76,
+                                    "content": "\\sigma",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        473,
+                                        340,
+                                        509,
+                                        363
+                                    ],
+                                    "score": 1.0,
+                                    "content": "using a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                354,
+                                412,
+                                371
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        354,
+                                        217,
+                                        371
+                                    ],
+                                    "score": 1.0,
+                                    "content": "perfect HASH function:",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        217,
+                                        356,
+                                        407,
+                                        370
+                                    ],
+                                    "score": 0.91,
+                                    "content": "c _ { \\sigma } ^ { t + 1 } = \\mathrm { H A S H } \\big ( c _ { \\sigma } ^ { t } , c _ { \\mathcal { B } } ^ { t } ( \\sigma ) , c _ { \\mathcal { C } } ^ { t } ( \\sigma ) , c _ { \\mathcal { A } } ^ { t } ( \\sigma ) , c _ { \\uparrow } ^ { t } ( \\sigma ) \\big )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        408,
+                                        354,
+                                        412,
+                                        371
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                372,
+                                505,
+                                384
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        372,
+                                        505,
+                                        384
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3. The algorithm stops when a stable colouring is reached. Two cell complexes are considered",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                118,
+                                383,
+                                483,
+                                394
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        118,
+                                        383,
+                                        483,
+                                        394
+                                    ],
+                                    "score": 1.0,
+                                    "content": "non-isomorphic if their colour histograms are different. Otherwise, the test is inconclusive.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        }
+                    ],
+                    "index": 16.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        401,
+                        504,
+                        435
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                400,
+                                506,
+                                414
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        400,
+                                        326,
+                                        414
+                                    ],
+                                    "score": 1.0,
+                                    "content": "First, we state the following theorem from Bodnar et al.",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        327,
+                                        401,
+                                        339,
+                                        413
+                                    ],
+                                    "score": 0.77,
+                                    "content": "\\textcircled { 8 } \\textcircled { 1 8 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        340,
+                                        400,
+                                        506,
+                                        414
+                                    ],
+                                    "score": 1.0,
+                                    "content": "involving SWL and simplicial complexes.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                411,
+                                505,
+                                426
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        411,
+                                        505,
+                                        426
+                                    ],
+                                    "score": 1.0,
+                                    "content": "This theorem shows that on simplicial complexes, certain adjacencies can be pruned without affecting",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                424,
+                                501,
+                                436
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        424,
+                                        501,
+                                        436
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the non-isomorphic SCs that can be distinguished. This has important computational implications.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        }
+                    ],
+                    "index": 21
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        437,
+                        504,
+                        459
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                436,
+                                505,
+                                450
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        436,
+                                        505,
+                                        450
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Theorem 6. SWL without coboundary and lower-adjacencies has the same expressive power in",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                447,
+                                497,
+                                460
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        447,
+                                        497,
+                                        460
+                                    ],
+                                    "score": 1.0,
+                                    "content": "distinguishing non-isomorphic simplicial complexes as SWL with the complete set of adjacencies.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        }
+                    ],
+                    "index": 23.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        466,
+                        505,
+                        510
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                466,
+                                505,
+                                478
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        466,
+                                        505,
+                                        478
+                                    ],
+                                    "score": 1.0,
+                                    "content": "It is not immediately clear whether an equivalent theorem would also hold for cell complexes. This",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                477,
+                                505,
+                                490
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        477,
+                                        505,
+                                        490
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is because cells, unlike simplices, can have widely different shapes and, as described above, the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                489,
+                                505,
+                                500
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        489,
+                                        505,
+                                        500
+                                    ],
+                                    "score": 1.0,
+                                    "content": "adjacencies between them take more complicated forms. Nevertheless, we show that a positive result",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                500,
+                                174,
+                                510
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        500,
+                                        174,
+                                        510
+                                    ],
+                                    "score": 1.0,
+                                    "content": "can be obtained.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        }
+                    ],
+                    "index": 26.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        512,
+                        504,
+                        535
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                511,
+                                505,
+                                525
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        511,
+                                        505,
+                                        525
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Theorem 7. CWL without coboundary and lower-adjacencies has the same expressive power in",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                524,
+                                474,
+                                536
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        524,
+                                        474,
+                                        536
+                                    ],
+                                    "score": 1.0,
+                                    "content": "distinguishing non-isomorphic cell complexes as CWL with the complete set of adjacencies.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        }
+                    ],
+                    "index": 29.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        542,
+                        505,
+                        575
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                541,
+                                505,
+                                555
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        541,
+                                        505,
+                                        555
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We note this does not mean that the removed adjacencies are completely redundant in practice. Even",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                552,
+                                505,
+                                565
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        552,
+                                        505,
+                                        565
+                                    ],
+                                    "score": 1.0,
+                                    "content": "if they are not needed from a (theoretical) colour refinment perspective, they might still include",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                564,
+                                380,
+                                576
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        564,
+                                        380,
+                                        576
+                                    ],
+                                    "score": 1.0,
+                                    "content": "important inductive biases that make them suitable for certain tasks.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        }
+                    ],
+                    "index": 32
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        580,
+                        505,
+                        646
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                580,
+                                505,
+                                593
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        580,
+                                        505,
+                                        593
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We are now interested in examining various procedures for mapping, or “lifting”, graphs into the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                591,
+                                506,
+                                604
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        591,
+                                        506,
+                                        604
+                                    ],
+                                    "score": 1.0,
+                                    "content": "space of regular cell complexes. Such a procedure can be used to test the isomorphism of two graphs",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                602,
+                                506,
+                                615
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        602,
+                                        506,
+                                        615
+                                    ],
+                                    "score": 1.0,
+                                    "content": "by performing colour refinement on the cell complexes they are mapped to. The hope is that CWL",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                613,
+                                506,
+                                626
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        613,
+                                        506,
+                                        626
+                                    ],
+                                    "score": 1.0,
+                                    "content": "applied to these cell complexes is more powerful than WL applied to the initial graphs. We will later",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                622,
+                                506,
+                                638
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        622,
+                                        506,
+                                        638
+                                    ],
+                                    "score": 1.0,
+                                    "content": "show that for a wide range of transformations, this is indeed the case. We start by rigorously defining",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                634,
+                                223,
+                                648
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        634,
+                                        223,
+                                        648
+                                    ],
+                                    "score": 1.0,
+                                    "content": "what we mean by a “lifting”.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        }
+                    ],
+                    "index": 36.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        649,
+                        505,
+                        682
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                647,
+                                506,
+                                662
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        647,
+                                        312,
+                                        662
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Definition 8. A cellular lifting map is a function",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        312,
+                                        649,
+                                        362,
+                                        660
+                                    ],
+                                    "score": 0.91,
+                                    "content": "f : \\mathcal G  \\mathcal { X }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        362,
+                                        647,
+                                        469,
+                                        662
+                                    ],
+                                    "score": 1.0,
+                                    "content": "from the space of graphs",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        470,
+                                        649,
+                                        477,
+                                        659
+                                    ],
+                                    "score": 0.76,
+                                    "content": "\\mathcal { G }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        478,
+                                        647,
+                                        506,
+                                        662
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                658,
+                                505,
+                                672
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        658,
+                                        236,
+                                        672
+                                    ],
+                                    "score": 1.0,
+                                    "content": "space of regular cell complexes",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        236,
+                                        660,
+                                        246,
+                                        669
+                                    ],
+                                    "score": 0.79,
+                                    "content": "\\mathcal { X }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        246,
+                                        658,
+                                        384,
+                                        672
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with the property that two graphs",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        385,
+                                        660,
+                                        414,
+                                        671
+                                    ],
+                                    "score": 0.91,
+                                    "content": "G _ { 1 } , G _ { 2 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        415,
+                                        658,
+                                        505,
+                                        672
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are isomorphic iff the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                670,
+                                291,
+                                683
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        670,
+                                        167,
+                                        683
+                                    ],
+                                    "score": 1.0,
+                                    "content": "cell complexes",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        168,
+                                        670,
+                                        225,
+                                        682
+                                    ],
+                                    "score": 0.93,
+                                    "content": "f ( G _ { 1 } ) , f ( \\mathbf { \\bar { G } } _ { 2 } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        225,
+                                        670,
+                                        291,
+                                        683
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are isomorphic.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        }
+                    ],
+                    "index": 41
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        689,
+                        505,
+                        722
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                687,
+                                505,
+                                702
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        687,
+                                        505,
+                                        702
+                                    ],
+                                    "score": 1.0,
+                                    "content": "This property ensures that testing the isomorphism of the two cell complexes is equivalent to testing",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                700,
+                                505,
+                                712
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        700,
+                                        505,
+                                        712
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the isomorphism in the input graphs. This would not be the case if two non-isomoprhic graphs were",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                711,
+                                244,
+                                723
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        711,
+                                        244,
+                                        723
+                                    ],
+                                    "score": 1.0,
+                                    "content": "mapped to the same cell complex.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        }
+                    ],
+                    "index": 44
+                }
+            ],
+            "page_idx": 3,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        302,
+                        742,
+                        308,
+                        750
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                301,
+                                741,
+                                310,
+                                752
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        301,
+                                        741,
+                                        310,
+                                        752
+                                    ],
+                                    "score": 1.0,
+                                    "content": "",
+                                    "type": "text",
+                                    "height": 11,
+                                    "width": 9
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "image",
+                    "bbox": [
+                        128,
+                        71,
+                        485,
+                        154
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                128,
+                                71,
+                                485,
+                                154
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        128,
+                                        71,
+                                        485,
+                                        154
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                128,
+                                                71,
+                                                485,
+                                                154
+                                            ],
+                                            "score": 0.931,
+                                            "type": "image",
+                                            "image_path": "f132e64580feb98e0dddba934ea49e01c6fa4a4cc004b0c67d64ceaec82753eb.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 1,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        128,
+                                        71,
+                                        485,
+                                        98.66666666666667
+                                    ],
+                                    "spans": [],
+                                    "index": 0
+                                },
+                                {
+                                    "bbox": [
+                                        128,
+                                        98.66666666666667,
+                                        485,
+                                        126.33333333333334
+                                    ],
+                                    "spans": [],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        128,
+                                        126.33333333333334,
+                                        485,
+                                        154.0
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                106,
+                                160,
+                                503,
+                                195
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        161,
+                                        505,
+                                        173
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                161,
+                                                505,
+                                                173
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 4: The CWL colouring procedure for the yellow edge of the cell complex. All cells have",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 3
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        171,
+                                        505,
+                                        186
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                171,
+                                                505,
+                                                186
+                                            ],
+                                            "score": 1.0,
+                                            "content": "been assigned unique colours to aid the visualisation of the adjacencies. Note that the yellow edge",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 4
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        183,
+                                        352,
+                                        196
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                183,
+                                                352,
+                                                196
+                                            ],
+                                            "score": 1.0,
+                                            "content": "aggregates long-range information from the light green edge.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 5
+                                }
+                            ],
+                            "index": 4
+                        }
+                    ],
+                    "index": 2.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        213,
+                        505,
+                        257
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                212,
+                                505,
+                                226
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        212,
+                                        360,
+                                        226
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Note that unlike in graphs and simplicial complexes, the sets",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        360,
+                                        213,
+                                        391,
+                                        225
+                                    ],
+                                    "score": 0.93,
+                                    "content": "B ( \\sigma , \\tau )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        391,
+                                        212,
+                                        410,
+                                        226
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        411,
+                                        213,
+                                        441,
+                                        225
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\mathcal { C } ( \\sigma , \\tau )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        441,
+                                        212,
+                                        505,
+                                        226
+                                    ],
+                                    "score": 1.0,
+                                    "content": "can have more",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                224,
+                                505,
+                                236
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        224,
+                                        505,
+                                        236
+                                    ],
+                                    "score": 1.0,
+                                    "content": "than one element. For instance, two (closed) 2-cells might intersect in more than one edge (e.g. the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                234,
+                                506,
+                                249
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        234,
+                                        213,
+                                        249
+                                    ],
+                                    "score": 1.0,
+                                    "content": "two hemispheres in Figure",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        213,
+                                        234,
+                                        224,
+                                        248
+                                    ],
+                                    "score": 0.62,
+                                    "content": "\\textcircled { 3 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        225,
+                                        234,
+                                        506,
+                                        249
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", and conversely, two edges might be on the boundary of the same two",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                245,
+                                440,
+                                258
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        245,
+                                        440,
+                                        258
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2-cells. This illustrates the more flexible combinatorial structure of cell complexes.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        }
+                    ],
+                    "index": 7.5,
+                    "bbox_fs": [
+                        105,
+                        212,
+                        506,
+                        258
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        268,
+                        505,
+                        313
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                267,
+                                506,
+                                281
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        267,
+                                        506,
+                                        281
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Cellular WL (CWL) We consider CWL, a colour refinement scheme for cell complexes that",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                278,
+                                506,
+                                292
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        278,
+                                        232,
+                                        292
+                                    ],
+                                    "score": 1.0,
+                                    "content": "generalises the Simplicial WL",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        232,
+                                        279,
+                                        245,
+                                        290
+                                    ],
+                                    "score": 0.63,
+                                    "content": "\\textcircled { 8 } \\textcircled { 1 8 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        245,
+                                        278,
+                                        282,
+                                        292
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and WL",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        282,
+                                        279,
+                                        300,
+                                        290
+                                    ],
+                                    "score": 0.53,
+                                    "content": "\\mathbb { \\ m }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        300,
+                                        278,
+                                        356,
+                                        292
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tests. We use",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        357,
+                                        279,
+                                        367,
+                                        291
+                                    ],
+                                    "score": 0.88,
+                                    "content": "c _ { \\sigma } ^ { t }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        368,
+                                        278,
+                                        506,
+                                        292
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to refer to the colour assigned by",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                289,
+                                505,
+                                302
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        289,
+                                        158,
+                                        302
+                                    ],
+                                    "score": 1.0,
+                                    "content": "CWL to cell",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        159,
+                                        291,
+                                        166,
+                                        300
+                                    ],
+                                    "score": 0.78,
+                                    "content": "\\sigma",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        167,
+                                        289,
+                                        213,
+                                        302
+                                    ],
+                                    "score": 1.0,
+                                    "content": "at iteration",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        213,
+                                        291,
+                                        218,
+                                        300
+                                    ],
+                                    "score": 0.74,
+                                    "content": "t",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        218,
+                                        289,
+                                        505,
+                                        302
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of the algorithm. When the input is a simplicial complex, this recovers",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                300,
+                                489,
+                                314
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        301,
+                                        414,
+                                        314
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the SWL algorithm. A step of the algorithm is graphically depicted in Figure",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        414,
+                                        300,
+                                        424,
+                                        314
+                                    ],
+                                    "score": 0.47,
+                                    "content": "^ 4",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        424,
+                                        301,
+                                        489,
+                                        314
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for a single cell.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        }
+                    ],
+                    "index": 11.5,
+                    "bbox_fs": [
+                        105,
+                        267,
+                        506,
+                        314
+                    ]
+                },
+                {
+                    "type": "list",
+                    "bbox": [
+                        106,
+                        319,
+                        508,
+                        394
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                320,
+                                450,
+                                332
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        320,
+                                        236,
+                                        332
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1. Given a regular cell complex",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        236,
+                                        320,
+                                        246,
+                                        330
+                                    ],
+                                    "score": 0.79,
+                                    "content": "X",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        247,
+                                        320,
+                                        298,
+                                        332
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", all the cells",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        298,
+                                        322,
+                                        306,
+                                        330
+                                    ],
+                                    "score": 0.79,
+                                    "content": "\\sigma",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        306,
+                                        320,
+                                        450,
+                                        332
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are initialised with the same colour.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14,
+                            "is_list_start_line": true,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                333,
+                                505,
+                                347
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        333,
+                                        189,
+                                        347
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2. Given the colour",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        189,
+                                        333,
+                                        200,
+                                        346
+                                    ],
+                                    "score": 0.89,
+                                    "content": "c _ { \\sigma } ^ { t }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        200,
+                                        333,
+                                        229,
+                                        347
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of cell",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        229,
+                                        336,
+                                        237,
+                                        344
+                                    ],
+                                    "score": 0.77,
+                                    "content": "\\sigma",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        237,
+                                        333,
+                                        284,
+                                        347
+                                    ],
+                                    "score": 1.0,
+                                    "content": "at iteration",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        284,
+                                        335,
+                                        289,
+                                        344
+                                    ],
+                                    "score": 0.73,
+                                    "content": "t",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        289,
+                                        333,
+                                        415,
+                                        347
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", we compute the colour of cell",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        416,
+                                        336,
+                                        423,
+                                        344
+                                    ],
+                                    "score": 0.75,
+                                    "content": "\\sigma",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        424,
+                                        333,
+                                        505,
+                                        347
+                                    ],
+                                    "score": 1.0,
+                                    "content": "at the next iteration",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                119,
+                                340,
+                                509,
+                                363
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        119,
+                                        344,
+                                        138,
+                                        357
+                                    ],
+                                    "score": 0.91,
+                                    "content": "c _ { \\sigma } ^ { t + 1 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        138,
+                                        340,
+                                        465,
+                                        363
+                                    ],
+                                    "score": 1.0,
+                                    "content": "by injectively mapping the multi-sets of colours belonging to the adjacent cells of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        466,
+                                        347,
+                                        473,
+                                        354
+                                    ],
+                                    "score": 0.76,
+                                    "content": "\\sigma",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        473,
+                                        340,
+                                        509,
+                                        363
+                                    ],
+                                    "score": 1.0,
+                                    "content": "using a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                116,
+                                354,
+                                412,
+                                371
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        116,
+                                        354,
+                                        217,
+                                        371
+                                    ],
+                                    "score": 1.0,
+                                    "content": "perfect HASH function:",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        217,
+                                        356,
+                                        407,
+                                        370
+                                    ],
+                                    "score": 0.91,
+                                    "content": "c _ { \\sigma } ^ { t + 1 } = \\mathrm { H A S H } \\big ( c _ { \\sigma } ^ { t } , c _ { \\mathcal { B } } ^ { t } ( \\sigma ) , c _ { \\mathcal { C } } ^ { t } ( \\sigma ) , c _ { \\mathcal { A } } ^ { t } ( \\sigma ) , c _ { \\uparrow } ^ { t } ( \\sigma ) \\big )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        408,
+                                        354,
+                                        412,
+                                        371
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                372,
+                                505,
+                                384
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        372,
+                                        505,
+                                        384
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3. The algorithm stops when a stable colouring is reached. Two cell complexes are considered",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                118,
+                                383,
+                                483,
+                                394
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        118,
+                                        383,
+                                        483,
+                                        394
+                                    ],
+                                    "score": 1.0,
+                                    "content": "non-isomorphic if their colour histograms are different. Otherwise, the test is inconclusive.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19,
+                            "is_list_end_line": true
+                        }
+                    ],
+                    "index": 16.5,
+                    "bbox_fs": [
+                        105,
+                        320,
+                        509,
+                        394
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        401,
+                        504,
+                        435
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                400,
+                                506,
+                                414
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        400,
+                                        326,
+                                        414
+                                    ],
+                                    "score": 1.0,
+                                    "content": "First, we state the following theorem from Bodnar et al.",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        327,
+                                        401,
+                                        339,
+                                        413
+                                    ],
+                                    "score": 0.77,
+                                    "content": "\\textcircled { 8 } \\textcircled { 1 8 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        340,
+                                        400,
+                                        506,
+                                        414
+                                    ],
+                                    "score": 1.0,
+                                    "content": "involving SWL and simplicial complexes.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                411,
+                                505,
+                                426
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        411,
+                                        505,
+                                        426
+                                    ],
+                                    "score": 1.0,
+                                    "content": "This theorem shows that on simplicial complexes, certain adjacencies can be pruned without affecting",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                424,
+                                501,
+                                436
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        424,
+                                        501,
+                                        436
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the non-isomorphic SCs that can be distinguished. This has important computational implications.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        }
+                    ],
+                    "index": 21,
+                    "bbox_fs": [
+                        105,
+                        400,
+                        506,
+                        436
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        437,
+                        504,
+                        459
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                436,
+                                505,
+                                450
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        436,
+                                        505,
+                                        450
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Theorem 6. SWL without coboundary and lower-adjacencies has the same expressive power in",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                447,
+                                497,
+                                460
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        447,
+                                        497,
+                                        460
+                                    ],
+                                    "score": 1.0,
+                                    "content": "distinguishing non-isomorphic simplicial complexes as SWL with the complete set of adjacencies.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        }
+                    ],
+                    "index": 23.5,
+                    "bbox_fs": [
+                        105,
+                        436,
+                        505,
+                        460
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        466,
+                        505,
+                        510
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                466,
+                                505,
+                                478
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        466,
+                                        505,
+                                        478
+                                    ],
+                                    "score": 1.0,
+                                    "content": "It is not immediately clear whether an equivalent theorem would also hold for cell complexes. This",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                477,
+                                505,
+                                490
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        477,
+                                        505,
+                                        490
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is because cells, unlike simplices, can have widely different shapes and, as described above, the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                489,
+                                505,
+                                500
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        489,
+                                        505,
+                                        500
+                                    ],
+                                    "score": 1.0,
+                                    "content": "adjacencies between them take more complicated forms. Nevertheless, we show that a positive result",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                500,
+                                174,
+                                510
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        500,
+                                        174,
+                                        510
+                                    ],
+                                    "score": 1.0,
+                                    "content": "can be obtained.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        }
+                    ],
+                    "index": 26.5,
+                    "bbox_fs": [
+                        105,
+                        466,
+                        505,
+                        510
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        512,
+                        504,
+                        535
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                511,
+                                505,
+                                525
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        511,
+                                        505,
+                                        525
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Theorem 7. CWL without coboundary and lower-adjacencies has the same expressive power in",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                524,
+                                474,
+                                536
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        524,
+                                        474,
+                                        536
+                                    ],
+                                    "score": 1.0,
+                                    "content": "distinguishing non-isomorphic cell complexes as CWL with the complete set of adjacencies.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        }
+                    ],
+                    "index": 29.5,
+                    "bbox_fs": [
+                        106,
+                        511,
+                        505,
+                        536
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        542,
+                        505,
+                        575
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                541,
+                                505,
+                                555
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        541,
+                                        505,
+                                        555
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We note this does not mean that the removed adjacencies are completely redundant in practice. Even",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                552,
+                                505,
+                                565
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        552,
+                                        505,
+                                        565
+                                    ],
+                                    "score": 1.0,
+                                    "content": "if they are not needed from a (theoretical) colour refinment perspective, they might still include",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                564,
+                                380,
+                                576
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        564,
+                                        380,
+                                        576
+                                    ],
+                                    "score": 1.0,
+                                    "content": "important inductive biases that make them suitable for certain tasks.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        }
+                    ],
+                    "index": 32,
+                    "bbox_fs": [
+                        105,
+                        541,
+                        505,
+                        576
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        580,
+                        505,
+                        646
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                580,
+                                505,
+                                593
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        580,
+                                        505,
+                                        593
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We are now interested in examining various procedures for mapping, or “lifting”, graphs into the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                591,
+                                506,
+                                604
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        591,
+                                        506,
+                                        604
+                                    ],
+                                    "score": 1.0,
+                                    "content": "space of regular cell complexes. Such a procedure can be used to test the isomorphism of two graphs",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                602,
+                                506,
+                                615
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        602,
+                                        506,
+                                        615
+                                    ],
+                                    "score": 1.0,
+                                    "content": "by performing colour refinement on the cell complexes they are mapped to. The hope is that CWL",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                613,
+                                506,
+                                626
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        613,
+                                        506,
+                                        626
+                                    ],
+                                    "score": 1.0,
+                                    "content": "applied to these cell complexes is more powerful than WL applied to the initial graphs. We will later",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                622,
+                                506,
+                                638
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        622,
+                                        506,
+                                        638
+                                    ],
+                                    "score": 1.0,
+                                    "content": "show that for a wide range of transformations, this is indeed the case. We start by rigorously defining",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                634,
+                                223,
+                                648
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        634,
+                                        223,
+                                        648
+                                    ],
+                                    "score": 1.0,
+                                    "content": "what we mean by a “lifting”.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        }
+                    ],
+                    "index": 36.5,
+                    "bbox_fs": [
+                        105,
+                        580,
+                        506,
+                        648
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        649,
+                        505,
+                        682
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                647,
+                                506,
+                                662
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        647,
+                                        312,
+                                        662
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Definition 8. A cellular lifting map is a function",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        312,
+                                        649,
+                                        362,
+                                        660
+                                    ],
+                                    "score": 0.91,
+                                    "content": "f : \\mathcal G  \\mathcal { X }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        362,
+                                        647,
+                                        469,
+                                        662
+                                    ],
+                                    "score": 1.0,
+                                    "content": "from the space of graphs",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        470,
+                                        649,
+                                        477,
+                                        659
+                                    ],
+                                    "score": 0.76,
+                                    "content": "\\mathcal { G }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        478,
+                                        647,
+                                        506,
+                                        662
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                658,
+                                505,
+                                672
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        658,
+                                        236,
+                                        672
+                                    ],
+                                    "score": 1.0,
+                                    "content": "space of regular cell complexes",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        236,
+                                        660,
+                                        246,
+                                        669
+                                    ],
+                                    "score": 0.79,
+                                    "content": "\\mathcal { X }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        246,
+                                        658,
+                                        384,
+                                        672
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with the property that two graphs",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        385,
+                                        660,
+                                        414,
+                                        671
+                                    ],
+                                    "score": 0.91,
+                                    "content": "G _ { 1 } , G _ { 2 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        415,
+                                        658,
+                                        505,
+                                        672
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are isomorphic iff the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                670,
+                                291,
+                                683
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        670,
+                                        167,
+                                        683
+                                    ],
+                                    "score": 1.0,
+                                    "content": "cell complexes",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        168,
+                                        670,
+                                        225,
+                                        682
+                                    ],
+                                    "score": 0.93,
+                                    "content": "f ( G _ { 1 } ) , f ( \\mathbf { \\bar { G } } _ { 2 } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        225,
+                                        670,
+                                        291,
+                                        683
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are isomorphic.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        }
+                    ],
+                    "index": 41,
+                    "bbox_fs": [
+                        105,
+                        647,
+                        506,
+                        683
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        689,
+                        505,
+                        722
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                687,
+                                505,
+                                702
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        687,
+                                        505,
+                                        702
+                                    ],
+                                    "score": 1.0,
+                                    "content": "This property ensures that testing the isomorphism of the two cell complexes is equivalent to testing",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                700,
+                                505,
+                                712
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        700,
+                                        505,
+                                        712
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the isomorphism in the input graphs. This would not be the case if two non-isomoprhic graphs were",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                711,
+                                244,
+                                723
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        711,
+                                        244,
+                                        723
+                                    ],
+                                    "score": 1.0,
+                                    "content": "mapped to the same cell complex.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        }
+                    ],
+                    "index": 44,
+                    "bbox_fs": [
+                        105,
+                        687,
+                        505,
+                        723
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        73,
+                        504,
+                        95
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                71,
+                                505,
+                                87
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        71,
+                                        505,
+                                        87
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Example 9. It can be verified that the function mapping each graph to its clique complex (i.e. every",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                83,
+                                404,
+                                97
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        84,
+                                        137,
+                                        96
+                                    ],
+                                    "score": 0.89,
+                                    "content": "( k + 1 )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        137,
+                                        83,
+                                        261,
+                                        97
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-clique in the graph becomes a",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        262,
+                                        84,
+                                        268,
+                                        93
+                                    ],
+                                    "score": 0.6,
+                                    "content": "k",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        268,
+                                        83,
+                                        404,
+                                        97
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-simplex) is a cellular lifting map.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        }
+                    ],
+                    "index": 0.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        104,
+                        101,
+                        505,
+                        124
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                101,
+                                506,
+                                114
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        101,
+                                        419,
+                                        114
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The clique complex lifting map from Example 9 has been used by Bodnar et al.",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        419,
+                                        102,
+                                        432,
+                                        113
+                                    ],
+                                    "score": 0.75,
+                                    "content": "\\pmb { \\mathbb { B } } ] \\mathbf { l }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        432,
+                                        101,
+                                        506,
+                                        114
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to show that SWL",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                114,
+                                334,
+                                126
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        114,
+                                        334,
+                                        126
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is strictly more powerful than WL. We restate this result:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        }
+                    ],
+                    "index": 2.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        127,
+                        433,
+                        138
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                126,
+                                433,
+                                140
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        126,
+                                        433,
+                                        140
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Theorem 10. SWL with clique complex lifting is strictly more powerful than WL.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        }
+                    ],
+                    "index": 4
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        145,
+                        502,
+                        168
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                146,
+                                504,
+                                157
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        146,
+                                        504,
+                                        157
+                                    ],
+                                    "score": 1.0,
+                                    "content": "A natural question is what other lifting transformations make CWL strictly more powerful than WL?",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                155,
+                                486,
+                                169
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        155,
+                                        486,
+                                        169
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We first describe a space of lifting transformations that make CWL at least as powerful as WL.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        }
+                    ],
+                    "index": 5.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        170,
+                        505,
+                        192
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                168,
+                                506,
+                                183
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        168,
+                                        381,
+                                        183
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Definition 11. A lifting map is skeleton-preserving if for any graph",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        381,
+                                        170,
+                                        390,
+                                        180
+                                    ],
+                                    "score": 0.38,
+                                    "content": "G",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        391,
+                                        168,
+                                        463,
+                                        183
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", the 1-skeleton of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        464,
+                                        169,
+                                        486,
+                                        182
+                                    ],
+                                    "score": 0.91,
+                                    "content": "f ( G )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        486,
+                                        168,
+                                        506,
+                                        183
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                107,
+                                181,
+                                252,
+                                194
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        181,
+                                        116,
+                                        191
+                                    ],
+                                    "score": 0.49,
+                                    "content": "G",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        116,
+                                        181,
+                                        252,
+                                        194
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are isomorphic as (multi) graphs.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        }
+                    ],
+                    "index": 7.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        199,
+                        505,
+                        266
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                198,
+                                505,
+                                213
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        198,
+                                        505,
+                                        213
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Intuitively, skeleton-preserving liftings ensure that the additional structure added by the lifting map",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                210,
+                                505,
+                                223
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        210,
+                                        505,
+                                        223
+                                    ],
+                                    "score": 1.0,
+                                    "content": "comes from attaching cells of dimension at least two to the graph. These mappings keep the 0-cells",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                221,
+                                506,
+                                234
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        221,
+                                        506,
+                                        234
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and 1-cells intact and are, therefore, restricted from making modifications to the input graph structure.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                233,
+                                505,
+                                244
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        233,
+                                        505,
+                                        244
+                                    ],
+                                    "score": 1.0,
+                                    "content": "An important remark is that for simplicial complexes, attaching simplices based on cliques present",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                243,
+                                505,
+                                256
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        243,
+                                        505,
+                                        256
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in the graph is the only possible skeleton preserving transformation. Once again, this illustrates the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                254,
+                                493,
+                                267
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        254,
+                                        493,
+                                        267
+                                    ],
+                                    "score": 1.0,
+                                    "content": "limitations of simplicial complexes for adding useful higher-dimensional structures to the graph.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        }
+                    ],
+                    "index": 11.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        267,
+                        504,
+                        301
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                267,
+                                506,
+                                280
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        267,
+                                        506,
+                                        280
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Example 12. The function from Example 9 is also skeleton-preserving because the 1-skeleton of the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                277,
+                                506,
+                                291
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        277,
+                                        506,
+                                        291
+                                    ],
+                                    "score": 1.0,
+                                    "content": "clique complex of a graph is trivially isomorphic to the graph. A lifting function mapping each graph",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                289,
+                                503,
+                                302
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        289,
+                                        503,
+                                        302
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to a multi-graph where each edge is doubled by a parallel edge is not skeleton-preserving (Figure 5)",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        }
+                    ],
+                    "index": 16
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        308,
+                        356,
+                        331
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                307,
+                                357,
+                                322
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        307,
+                                        357,
+                                        322
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We now show that all the maps in the skeleton-preserving class",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                318,
+                                261,
+                                332
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        318,
+                                        261,
+                                        332
+                                    ],
+                                    "score": 1.0,
+                                    "content": "have the following desirable property:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        }
+                    ],
+                    "index": 18.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        332,
+                        356,
+                        366
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                331,
+                                357,
+                                345
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        331,
+                                        180,
+                                        345
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Theorem 13. Let",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        180,
+                                        333,
+                                        187,
+                                        344
+                                    ],
+                                    "score": 0.47,
+                                    "content": "f",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        187,
+                                        331,
+                                        357,
+                                        345
+                                    ],
+                                    "score": 1.0,
+                                    "content": "be a skeleton-preserving lifting map. Then",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                342,
+                                357,
+                                356
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        343,
+                                        141,
+                                        355
+                                    ],
+                                    "score": 0.4,
+                                    "content": "C W L ( f )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        141,
+                                        342,
+                                        238,
+                                        356
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(i.e. CWL using lifting",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        239,
+                                        344,
+                                        246,
+                                        355
+                                    ],
+                                    "score": 0.65,
+                                    "content": "f",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        246,
+                                        342,
+                                        357,
+                                        356
+                                    ],
+                                    "score": 1.0,
+                                    "content": ") is at least as powerful as",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                354,
+                                288,
+                                367
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        354,
+                                        288,
+                                        367
+                                    ],
+                                    "score": 1.0,
+                                    "content": "WL in distinguishing non-isomorphic graphs.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        }
+                    ],
+                    "index": 22
+                },
+                {
+                    "type": "image",
+                    "bbox": [
+                        364,
+                        310,
+                        504,
+                        359
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                364,
+                                310,
+                                504,
+                                359
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        364,
+                                        310,
+                                        504,
+                                        359
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                364,
+                                                310,
+                                                504,
+                                                359
+                                            ],
+                                            "score": 0.95,
+                                            "type": "image",
+                                            "image_path": "fb663b1a2f9a5a4fcc9886ee88e4877662331081ae9bf2749c10d4b141fb7fb5.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 21.5,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        364,
+                                        310,
+                                        504,
+                                        334.5
+                                    ],
+                                    "spans": [],
+                                    "index": 20
+                                },
+                                {
+                                    "bbox": [
+                                        364,
+                                        334.5,
+                                        504,
+                                        359.0
+                                    ],
+                                    "spans": [],
+                                    "index": 23
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                363,
+                                363,
+                                506,
+                                407
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        363,
+                                        363,
+                                        506,
+                                        375
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                363,
+                                                363,
+                                                506,
+                                                375
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 5: A graph, its clique com-",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 25
+                                },
+                                {
+                                    "bbox": [
+                                        362,
+                                        374,
+                                        505,
+                                        386
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                362,
+                                                374,
+                                                505,
+                                                386
+                                            ],
+                                            "score": 1.0,
+                                            "content": "plex and the graph with duplicated",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 27
+                                },
+                                {
+                                    "bbox": [
+                                        362,
+                                        385,
+                                        506,
+                                        397
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                362,
+                                                385,
+                                                506,
+                                                397
+                                            ],
+                                            "score": 1.0,
+                                            "content": "edges. The first map is skeleton-",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 29
+                                },
+                                {
+                                    "bbox": [
+                                        362,
+                                        396,
+                                        505,
+                                        408
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                362,
+                                                396,
+                                                505,
+                                                408
+                                            ],
+                                            "score": 1.0,
+                                            "content": "preserving, while the second is not.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 31
+                                }
+                            ],
+                            "index": 28.0
+                        }
+                    ],
+                    "index": 24.75
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        373,
+                        356,
+                        417
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                372,
+                                356,
+                                385
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        372,
+                                        356,
+                                        385
+                                    ],
+                                    "score": 1.0,
+                                    "content": "To prove that some of these make CWL strictly more powerful",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                383,
+                                357,
+                                396
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        383,
+                                        357,
+                                        396
+                                    ],
+                                    "score": 1.0,
+                                    "content": "than WL, it is sufficient to find a pair of graphs that cannot be",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                393,
+                                357,
+                                409
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        393,
+                                        357,
+                                        409
+                                    ],
+                                    "score": 1.0,
+                                    "content": "distinguished by WL, but can be distinguished by CWL. The",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                405,
+                                293,
+                                419
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        405,
+                                        293,
+                                        419
+                                    ],
+                                    "score": 1.0,
+                                    "content": "following result gives examples of such maps.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        }
+                    ],
+                    "index": 29.0
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        419,
+                        504,
+                        442
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                418,
+                                506,
+                                432
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        418,
+                                        184,
+                                        432
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Definition 14. Let",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        185,
+                                        419,
+                                        191,
+                                        429
+                                    ],
+                                    "score": 0.75,
+                                    "content": "k",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        192,
+                                        418,
+                                        212,
+                                        432
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-CL,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        212,
+                                        419,
+                                        219,
+                                        429
+                                    ],
+                                    "score": 0.75,
+                                    "content": "k",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        219,
+                                        418,
+                                        237,
+                                        432
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-IC,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        237,
+                                        419,
+                                        244,
+                                        429
+                                    ],
+                                    "score": 0.7,
+                                    "content": "k",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        244,
+                                        418,
+                                        506,
+                                        432
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-C be the lifting maps attaching cells to all the cliques, induced",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                430,
+                                330,
+                                442
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        430,
+                                        319,
+                                        442
+                                    ],
+                                    "score": 1.0,
+                                    "content": "cycles and simple cycles, respectively, of size at most",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        320,
+                                        431,
+                                        326,
+                                        440
+                                    ],
+                                    "score": 0.66,
+                                    "content": "k",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        326,
+                                        430,
+                                        330,
+                                        442
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        }
+                    ],
+                    "index": 33.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        443,
+                        503,
+                        465
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                442,
+                                505,
+                                457
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        442,
+                                        198,
+                                        457
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Corollary 15. For all",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        198,
+                                        443,
+                                        223,
+                                        455
+                                    ],
+                                    "score": 0.86,
+                                    "content": "k \\geq 3",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        223,
+                                        442,
+                                        251,
+                                        457
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", CWL(",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        252,
+                                        444,
+                                        258,
+                                        453
+                                    ],
+                                    "score": 0.57,
+                                    "content": "k",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        258,
+                                        442,
+                                        306,
+                                        457
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-CL), CWL",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        307,
+                                        444,
+                                        313,
+                                        454
+                                    ],
+                                    "score": 0.59,
+                                    "content": "k",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        313,
+                                        442,
+                                        350,
+                                        457
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-IC) and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        350,
+                                        443,
+                                        393,
+                                        456
+                                    ],
+                                    "score": 0.31,
+                                    "content": "C W L ( k { \\mathrm { - C } } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        394,
+                                        442,
+                                        505,
+                                        457
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are strictly more powerful",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                452,
+                                147,
+                                467
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        452,
+                                        147,
+                                        467
+                                    ],
+                                    "score": 1.0,
+                                    "content": "than WL.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        }
+                    ],
+                    "index": 35.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        473,
+                        503,
+                        496
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                472,
+                                505,
+                                486
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        472,
+                                        505,
+                                        486
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We note that this is not a complete list. For instance, the result can also be extended to combinations",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                483,
+                                481,
+                                497
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        483,
+                                        481,
+                                        497
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of the above or other transformations. We can also relate CWL to the higher-order 3-WL test.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        }
+                    ],
+                    "index": 37.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        497,
+                        505,
+                        520
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                497,
+                                506,
+                                511
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        497,
+                                        506,
+                                        511
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Theorem 16. There exists a pair of graphs indistinguishable by 3-WL but distinguishable by CWL(k-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                508,
+                                382,
+                                521
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        508,
+                                        145,
+                                        521
+                                    ],
+                                    "score": 1.0,
+                                    "content": "CL) with",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        145,
+                                        509,
+                                        170,
+                                        519
+                                    ],
+                                    "score": 0.85,
+                                    "content": "k \\geq 4",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        171,
+                                        508,
+                                        198,
+                                        521
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", CWL(",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        198,
+                                        509,
+                                        205,
+                                        518
+                                    ],
+                                    "score": 0.41,
+                                    "content": "k",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        205,
+                                        508,
+                                        243,
+                                        521
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-IC) with",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        243,
+                                        509,
+                                        268,
+                                        519
+                                    ],
+                                    "score": 0.9,
+                                    "content": "k \\geq 4",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        268,
+                                        508,
+                                        287,
+                                        521
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        287,
+                                        509,
+                                        332,
+                                        520
+                                    ],
+                                    "score": 0.52,
+                                    "content": "C W L ( k { \\mathrm { - C } } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        332,
+                                        508,
+                                        353,
+                                        521
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        353,
+                                        509,
+                                        379,
+                                        519
+                                    ],
+                                    "score": 0.92,
+                                    "content": "k \\geq 8 .",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        379,
+                                        508,
+                                        382,
+                                        521
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        }
+                    ],
+                    "index": 39.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        527,
+                        504,
+                        550
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                526,
+                                506,
+                                540
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        526,
+                                        506,
+                                        540
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Finally, we conclude this section by showing how CWL can achieve a superior expressive power",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                538,
+                                404,
+                                551
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        538,
+                                        320,
+                                        551
+                                    ],
+                                    "score": 1.0,
+                                    "content": "compared to SWL. This result is proven by Corollary",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        320,
+                                        538,
+                                        334,
+                                        551
+                                    ],
+                                    "score": 0.41,
+                                    "content": "3 1",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        335,
+                                        538,
+                                        404,
+                                        551
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in the Appendix.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        }
+                    ],
+                    "index": 41.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        551,
+                        505,
+                        585
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                550,
+                                505,
+                                564
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        550,
+                                        181,
+                                        564
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Theorem 17. Let",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        182,
+                                        552,
+                                        189,
+                                        562
+                                    ],
+                                    "score": 0.69,
+                                    "content": "k",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        189,
+                                        550,
+                                        216,
+                                        564
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-CL [",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        217,
+                                        552,
+                                        223,
+                                        561
+                                    ],
+                                    "score": 0.45,
+                                    "content": "k",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        224,
+                                        550,
+                                        258,
+                                        564
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-IC and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        258,
+                                        551,
+                                        300,
+                                        562
+                                    ],
+                                    "score": 0.46,
+                                    "content": "k { \\mathrm { - C L } } \\cup k ",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        300,
+                                        550,
+                                        505,
+                                        564
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-C denote combined liftings attaching cells to the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                563,
+                                505,
+                                575
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        563,
+                                        259,
+                                        575
+                                    ],
+                                    "score": 1.0,
+                                    "content": "union of the specified substructures.",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        259,
+                                        563,
+                                        350,
+                                        574
+                                    ],
+                                    "score": 0.46,
+                                    "content": "C W L ( k _ { 1 } { \\mathrm { - C L } } \\cup k _ { 2 } { \\mathrm { - I C } } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        351,
+                                        563,
+                                        369,
+                                        575
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        370,
+                                        563,
+                                        456,
+                                        574
+                                    ],
+                                    "score": 0.64,
+                                    "content": "C W L ( k _ { 1 } { \\mathrm { - C L } } \\cup k _ { 2 } { \\mathrm { - C } } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        457,
+                                        563,
+                                        505,
+                                        575
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are strictly",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                573,
+                                302,
+                                586
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        573,
+                                        209,
+                                        586
+                                    ],
+                                    "score": 1.0,
+                                    "content": "more powerful than SWL",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        209,
+                                        574,
+                                        220,
+                                        584
+                                    ],
+                                    "score": 0.75,
+                                    "content": "k _ { 1 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        220,
+                                        573,
+                                        268,
+                                        586
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-CL) for all",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        269,
+                                        574,
+                                        298,
+                                        585
+                                    ],
+                                    "score": 0.9,
+                                    "content": "k _ { 2 } \\geq 5",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        298,
+                                        573,
+                                        302,
+                                        586
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        }
+                    ],
+                    "index": 44
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        599,
+                        367,
+                        613
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                598,
+                                368,
+                                615
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        598,
+                                        368,
+                                        615
+                                    ],
+                                    "score": 1.0,
+                                    "content": "4 Molecular Message Passing with CW Networks",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        }
+                    ],
+                    "index": 46
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        623,
+                        505,
+                        722
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                623,
+                                506,
+                                636
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        623,
+                                        506,
+                                        636
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We now describe CW Networks with an applied focus on molecular graphs to ground the discussion.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 47
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                634,
+                                506,
+                                647
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        634,
+                                        506,
+                                        647
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Therefore, from now on we assume the use of the skeleton-preserving lifting transformation that",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 48
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                645,
+                                505,
+                                658
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        645,
+                                        470,
+                                        658
+                                    ],
+                                    "score": 1.0,
+                                    "content": "attaches 2-cells to all the induced cycles (i.e. chordless cycles) in the graph as in Figure",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        471,
+                                        645,
+                                        482,
+                                        658
+                                    ],
+                                    "score": 0.46,
+                                    "content": "2 .",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        483,
+                                        645,
+                                        505,
+                                        657
+                                    ],
+                                    "score": 1.0,
+                                    "content": "This",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 49
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                655,
+                                505,
+                                669
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        655,
+                                        505,
+                                        669
+                                    ],
+                                    "score": 1.0,
+                                    "content": "leads to a message passing procedure involving atoms (vertices / 0-cells), the bonds between atoms",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 50
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                667,
+                                505,
+                                679
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        667,
+                                        505,
+                                        679
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(edges / 1-cells) and chemical rings (induced cycles / 2-cells). Additionally, in virtue of Theorem",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 51
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                677,
+                                506,
+                                692
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        678,
+                                        114,
+                                        691
+                                    ],
+                                    "score": 0.5,
+                                    "content": "\\underline { { \\overline { { 7 } } } } ,",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        114,
+                                        677,
+                                        506,
+                                        692
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", we consider only the boundary and upper adjacencies between these cells without sacrificing the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 52
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                689,
+                                505,
+                                701
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        689,
+                                        505,
+                                        701
+                                    ],
+                                    "score": 1.0,
+                                    "content": "expressive power. The equations for the other adjacencies, which we do not use, can be found in",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 53
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                700,
+                                505,
+                                713
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        700,
+                                        505,
+                                        713
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Appendix A. We note however, that the theoretical results in this section are general and not particular",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 54
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                711,
+                                371,
+                                723
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        711,
+                                        371,
+                                        723
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to these specific choices of adjacencies and lifting transformation.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 55
+                        }
+                    ],
+                    "index": 51
+                }
+            ],
+            "page_idx": 4,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        302,
+                        741,
+                        309,
+                        750
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                301,
+                                740,
+                                310,
+                                753
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        301,
+                                        740,
+                                        310,
+                                        753
+                                    ],
+                                    "score": 1.0,
+                                    "content": "5",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        73,
+                        504,
+                        95
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                71,
+                                505,
+                                87
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        71,
+                                        505,
+                                        87
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Example 9. It can be verified that the function mapping each graph to its clique complex (i.e. every",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                83,
+                                404,
+                                97
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        84,
+                                        137,
+                                        96
+                                    ],
+                                    "score": 0.89,
+                                    "content": "( k + 1 )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        137,
+                                        83,
+                                        261,
+                                        97
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-clique in the graph becomes a",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        262,
+                                        84,
+                                        268,
+                                        93
+                                    ],
+                                    "score": 0.6,
+                                    "content": "k",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        268,
+                                        83,
+                                        404,
+                                        97
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-simplex) is a cellular lifting map.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        }
+                    ],
+                    "index": 0.5,
+                    "bbox_fs": [
+                        105,
+                        71,
+                        505,
+                        97
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        104,
+                        101,
+                        505,
+                        124
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                101,
+                                506,
+                                114
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        101,
+                                        419,
+                                        114
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The clique complex lifting map from Example 9 has been used by Bodnar et al.",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        419,
+                                        102,
+                                        432,
+                                        113
+                                    ],
+                                    "score": 0.75,
+                                    "content": "\\pmb { \\mathbb { B } } ] \\mathbf { l }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        432,
+                                        101,
+                                        506,
+                                        114
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to show that SWL",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                114,
+                                334,
+                                126
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        114,
+                                        334,
+                                        126
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is strictly more powerful than WL. We restate this result:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        }
+                    ],
+                    "index": 2.5,
+                    "bbox_fs": [
+                        105,
+                        101,
+                        506,
+                        126
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        127,
+                        433,
+                        138
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                126,
+                                433,
+                                140
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        126,
+                                        433,
+                                        140
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Theorem 10. SWL with clique complex lifting is strictly more powerful than WL.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        }
+                    ],
+                    "index": 4,
+                    "bbox_fs": [
+                        106,
+                        126,
+                        433,
+                        140
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        145,
+                        502,
+                        168
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                146,
+                                504,
+                                157
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        146,
+                                        504,
+                                        157
+                                    ],
+                                    "score": 1.0,
+                                    "content": "A natural question is what other lifting transformations make CWL strictly more powerful than WL?",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                155,
+                                486,
+                                169
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        155,
+                                        486,
+                                        169
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We first describe a space of lifting transformations that make CWL at least as powerful as WL.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        }
+                    ],
+                    "index": 5.5,
+                    "bbox_fs": [
+                        105,
+                        146,
+                        504,
+                        169
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        170,
+                        505,
+                        192
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                168,
+                                506,
+                                183
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        168,
+                                        381,
+                                        183
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Definition 11. A lifting map is skeleton-preserving if for any graph",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        381,
+                                        170,
+                                        390,
+                                        180
+                                    ],
+                                    "score": 0.38,
+                                    "content": "G",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        391,
+                                        168,
+                                        463,
+                                        183
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", the 1-skeleton of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        464,
+                                        169,
+                                        486,
+                                        182
+                                    ],
+                                    "score": 0.91,
+                                    "content": "f ( G )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        486,
+                                        168,
+                                        506,
+                                        183
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                107,
+                                181,
+                                252,
+                                194
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        181,
+                                        116,
+                                        191
+                                    ],
+                                    "score": 0.49,
+                                    "content": "G",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        116,
+                                        181,
+                                        252,
+                                        194
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are isomorphic as (multi) graphs.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        }
+                    ],
+                    "index": 7.5,
+                    "bbox_fs": [
+                        105,
+                        168,
+                        506,
+                        194
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        199,
+                        505,
+                        266
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                198,
+                                505,
+                                213
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        198,
+                                        505,
+                                        213
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Intuitively, skeleton-preserving liftings ensure that the additional structure added by the lifting map",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                210,
+                                505,
+                                223
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        210,
+                                        505,
+                                        223
+                                    ],
+                                    "score": 1.0,
+                                    "content": "comes from attaching cells of dimension at least two to the graph. These mappings keep the 0-cells",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                221,
+                                506,
+                                234
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        221,
+                                        506,
+                                        234
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and 1-cells intact and are, therefore, restricted from making modifications to the input graph structure.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                233,
+                                505,
+                                244
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        233,
+                                        505,
+                                        244
+                                    ],
+                                    "score": 1.0,
+                                    "content": "An important remark is that for simplicial complexes, attaching simplices based on cliques present",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                243,
+                                505,
+                                256
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        243,
+                                        505,
+                                        256
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in the graph is the only possible skeleton preserving transformation. Once again, this illustrates the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                254,
+                                493,
+                                267
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        254,
+                                        493,
+                                        267
+                                    ],
+                                    "score": 1.0,
+                                    "content": "limitations of simplicial complexes for adding useful higher-dimensional structures to the graph.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        }
+                    ],
+                    "index": 11.5,
+                    "bbox_fs": [
+                        105,
+                        198,
+                        506,
+                        267
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        267,
+                        504,
+                        301
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                267,
+                                506,
+                                280
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        267,
+                                        506,
+                                        280
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Example 12. The function from Example 9 is also skeleton-preserving because the 1-skeleton of the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                277,
+                                506,
+                                291
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        277,
+                                        506,
+                                        291
+                                    ],
+                                    "score": 1.0,
+                                    "content": "clique complex of a graph is trivially isomorphic to the graph. A lifting function mapping each graph",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                289,
+                                503,
+                                302
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        289,
+                                        503,
+                                        302
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to a multi-graph where each edge is doubled by a parallel edge is not skeleton-preserving (Figure 5)",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        }
+                    ],
+                    "index": 16,
+                    "bbox_fs": [
+                        105,
+                        267,
+                        506,
+                        302
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        308,
+                        356,
+                        331
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                307,
+                                357,
+                                322
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        307,
+                                        357,
+                                        322
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We now show that all the maps in the skeleton-preserving class",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                318,
+                                261,
+                                332
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        318,
+                                        261,
+                                        332
+                                    ],
+                                    "score": 1.0,
+                                    "content": "have the following desirable property:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        }
+                    ],
+                    "index": 18.5,
+                    "bbox_fs": [
+                        106,
+                        307,
+                        357,
+                        332
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        332,
+                        356,
+                        366
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                331,
+                                357,
+                                345
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        331,
+                                        180,
+                                        345
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Theorem 13. Let",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        180,
+                                        333,
+                                        187,
+                                        344
+                                    ],
+                                    "score": 0.47,
+                                    "content": "f",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        187,
+                                        331,
+                                        357,
+                                        345
+                                    ],
+                                    "score": 1.0,
+                                    "content": "be a skeleton-preserving lifting map. Then",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                342,
+                                357,
+                                356
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        343,
+                                        141,
+                                        355
+                                    ],
+                                    "score": 0.4,
+                                    "content": "C W L ( f )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        141,
+                                        342,
+                                        238,
+                                        356
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(i.e. CWL using lifting",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        239,
+                                        344,
+                                        246,
+                                        355
+                                    ],
+                                    "score": 0.65,
+                                    "content": "f",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        246,
+                                        342,
+                                        357,
+                                        356
+                                    ],
+                                    "score": 1.0,
+                                    "content": ") is at least as powerful as",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                354,
+                                288,
+                                367
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        354,
+                                        288,
+                                        367
+                                    ],
+                                    "score": 1.0,
+                                    "content": "WL in distinguishing non-isomorphic graphs.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        }
+                    ],
+                    "index": 22,
+                    "bbox_fs": [
+                        105,
+                        331,
+                        357,
+                        367
+                    ]
+                },
+                {
+                    "type": "image",
+                    "bbox": [
+                        364,
+                        310,
+                        504,
+                        359
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                364,
+                                310,
+                                504,
+                                359
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        364,
+                                        310,
+                                        504,
+                                        359
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                364,
+                                                310,
+                                                504,
+                                                359
+                                            ],
+                                            "score": 0.95,
+                                            "type": "image",
+                                            "image_path": "fb663b1a2f9a5a4fcc9886ee88e4877662331081ae9bf2749c10d4b141fb7fb5.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 21.5,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        364,
+                                        310,
+                                        504,
+                                        334.5
+                                    ],
+                                    "spans": [],
+                                    "index": 20
+                                },
+                                {
+                                    "bbox": [
+                                        364,
+                                        334.5,
+                                        504,
+                                        359.0
+                                    ],
+                                    "spans": [],
+                                    "index": 23
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                363,
+                                363,
+                                506,
+                                407
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        363,
+                                        363,
+                                        506,
+                                        375
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                363,
+                                                363,
+                                                506,
+                                                375
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 5: A graph, its clique com-",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 25
+                                },
+                                {
+                                    "bbox": [
+                                        362,
+                                        374,
+                                        505,
+                                        386
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                362,
+                                                374,
+                                                505,
+                                                386
+                                            ],
+                                            "score": 1.0,
+                                            "content": "plex and the graph with duplicated",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 27
+                                },
+                                {
+                                    "bbox": [
+                                        362,
+                                        385,
+                                        506,
+                                        397
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                362,
+                                                385,
+                                                506,
+                                                397
+                                            ],
+                                            "score": 1.0,
+                                            "content": "edges. The first map is skeleton-",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 29
+                                },
+                                {
+                                    "bbox": [
+                                        362,
+                                        396,
+                                        505,
+                                        408
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                362,
+                                                396,
+                                                505,
+                                                408
+                                            ],
+                                            "score": 1.0,
+                                            "content": "preserving, while the second is not.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 31
+                                }
+                            ],
+                            "index": 28.0
+                        }
+                    ],
+                    "index": 24.75
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        373,
+                        356,
+                        417
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                372,
+                                356,
+                                385
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        372,
+                                        356,
+                                        385
+                                    ],
+                                    "score": 1.0,
+                                    "content": "To prove that some of these make CWL strictly more powerful",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                383,
+                                357,
+                                396
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        383,
+                                        357,
+                                        396
+                                    ],
+                                    "score": 1.0,
+                                    "content": "than WL, it is sufficient to find a pair of graphs that cannot be",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                393,
+                                357,
+                                409
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        393,
+                                        357,
+                                        409
+                                    ],
+                                    "score": 1.0,
+                                    "content": "distinguished by WL, but can be distinguished by CWL. The",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                405,
+                                293,
+                                419
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        405,
+                                        293,
+                                        419
+                                    ],
+                                    "score": 1.0,
+                                    "content": "following result gives examples of such maps.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        }
+                    ],
+                    "index": 29.0,
+                    "bbox_fs": [
+                        105,
+                        372,
+                        357,
+                        419
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        419,
+                        504,
+                        442
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                418,
+                                506,
+                                432
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        418,
+                                        184,
+                                        432
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Definition 14. Let",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        185,
+                                        419,
+                                        191,
+                                        429
+                                    ],
+                                    "score": 0.75,
+                                    "content": "k",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        192,
+                                        418,
+                                        212,
+                                        432
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-CL,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        212,
+                                        419,
+                                        219,
+                                        429
+                                    ],
+                                    "score": 0.75,
+                                    "content": "k",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        219,
+                                        418,
+                                        237,
+                                        432
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-IC,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        237,
+                                        419,
+                                        244,
+                                        429
+                                    ],
+                                    "score": 0.7,
+                                    "content": "k",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        244,
+                                        418,
+                                        506,
+                                        432
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-C be the lifting maps attaching cells to all the cliques, induced",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                430,
+                                330,
+                                442
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        430,
+                                        319,
+                                        442
+                                    ],
+                                    "score": 1.0,
+                                    "content": "cycles and simple cycles, respectively, of size at most",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        320,
+                                        431,
+                                        326,
+                                        440
+                                    ],
+                                    "score": 0.66,
+                                    "content": "k",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        326,
+                                        430,
+                                        330,
+                                        442
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        }
+                    ],
+                    "index": 33.5,
+                    "bbox_fs": [
+                        105,
+                        418,
+                        506,
+                        442
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        443,
+                        503,
+                        465
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                442,
+                                505,
+                                457
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        442,
+                                        198,
+                                        457
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Corollary 15. For all",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        198,
+                                        443,
+                                        223,
+                                        455
+                                    ],
+                                    "score": 0.86,
+                                    "content": "k \\geq 3",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        223,
+                                        442,
+                                        251,
+                                        457
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", CWL(",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        252,
+                                        444,
+                                        258,
+                                        453
+                                    ],
+                                    "score": 0.57,
+                                    "content": "k",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        258,
+                                        442,
+                                        306,
+                                        457
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-CL), CWL",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        307,
+                                        444,
+                                        313,
+                                        454
+                                    ],
+                                    "score": 0.59,
+                                    "content": "k",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        313,
+                                        442,
+                                        350,
+                                        457
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-IC) and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        350,
+                                        443,
+                                        393,
+                                        456
+                                    ],
+                                    "score": 0.31,
+                                    "content": "C W L ( k { \\mathrm { - C } } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        394,
+                                        442,
+                                        505,
+                                        457
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are strictly more powerful",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                452,
+                                147,
+                                467
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        452,
+                                        147,
+                                        467
+                                    ],
+                                    "score": 1.0,
+                                    "content": "than WL.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        }
+                    ],
+                    "index": 35.5,
+                    "bbox_fs": [
+                        105,
+                        442,
+                        505,
+                        467
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        473,
+                        503,
+                        496
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                472,
+                                505,
+                                486
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        472,
+                                        505,
+                                        486
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We note that this is not a complete list. For instance, the result can also be extended to combinations",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                483,
+                                481,
+                                497
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        483,
+                                        481,
+                                        497
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of the above or other transformations. We can also relate CWL to the higher-order 3-WL test.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        }
+                    ],
+                    "index": 37.5,
+                    "bbox_fs": [
+                        105,
+                        472,
+                        505,
+                        497
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        497,
+                        505,
+                        520
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                497,
+                                506,
+                                511
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        497,
+                                        506,
+                                        511
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Theorem 16. There exists a pair of graphs indistinguishable by 3-WL but distinguishable by CWL(k-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                508,
+                                382,
+                                521
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        508,
+                                        145,
+                                        521
+                                    ],
+                                    "score": 1.0,
+                                    "content": "CL) with",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        145,
+                                        509,
+                                        170,
+                                        519
+                                    ],
+                                    "score": 0.85,
+                                    "content": "k \\geq 4",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        171,
+                                        508,
+                                        198,
+                                        521
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", CWL(",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        198,
+                                        509,
+                                        205,
+                                        518
+                                    ],
+                                    "score": 0.41,
+                                    "content": "k",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        205,
+                                        508,
+                                        243,
+                                        521
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-IC) with",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        243,
+                                        509,
+                                        268,
+                                        519
+                                    ],
+                                    "score": 0.9,
+                                    "content": "k \\geq 4",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        268,
+                                        508,
+                                        287,
+                                        521
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        287,
+                                        509,
+                                        332,
+                                        520
+                                    ],
+                                    "score": 0.52,
+                                    "content": "C W L ( k { \\mathrm { - C } } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        332,
+                                        508,
+                                        353,
+                                        521
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        353,
+                                        509,
+                                        379,
+                                        519
+                                    ],
+                                    "score": 0.92,
+                                    "content": "k \\geq 8 .",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        379,
+                                        508,
+                                        382,
+                                        521
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        }
+                    ],
+                    "index": 39.5,
+                    "bbox_fs": [
+                        105,
+                        497,
+                        506,
+                        521
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        527,
+                        504,
+                        550
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                526,
+                                506,
+                                540
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        526,
+                                        506,
+                                        540
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Finally, we conclude this section by showing how CWL can achieve a superior expressive power",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                538,
+                                404,
+                                551
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        538,
+                                        320,
+                                        551
+                                    ],
+                                    "score": 1.0,
+                                    "content": "compared to SWL. This result is proven by Corollary",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        320,
+                                        538,
+                                        334,
+                                        551
+                                    ],
+                                    "score": 0.41,
+                                    "content": "3 1",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        335,
+                                        538,
+                                        404,
+                                        551
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in the Appendix.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        }
+                    ],
+                    "index": 41.5,
+                    "bbox_fs": [
+                        105,
+                        526,
+                        506,
+                        551
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        551,
+                        505,
+                        585
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                550,
+                                505,
+                                564
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        550,
+                                        181,
+                                        564
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Theorem 17. Let",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        182,
+                                        552,
+                                        189,
+                                        562
+                                    ],
+                                    "score": 0.69,
+                                    "content": "k",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        189,
+                                        550,
+                                        216,
+                                        564
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-CL [",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        217,
+                                        552,
+                                        223,
+                                        561
+                                    ],
+                                    "score": 0.45,
+                                    "content": "k",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        224,
+                                        550,
+                                        258,
+                                        564
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-IC and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        258,
+                                        551,
+                                        300,
+                                        562
+                                    ],
+                                    "score": 0.46,
+                                    "content": "k { \\mathrm { - C L } } \\cup k ",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        300,
+                                        550,
+                                        505,
+                                        564
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-C denote combined liftings attaching cells to the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                563,
+                                505,
+                                575
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        563,
+                                        259,
+                                        575
+                                    ],
+                                    "score": 1.0,
+                                    "content": "union of the specified substructures.",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        259,
+                                        563,
+                                        350,
+                                        574
+                                    ],
+                                    "score": 0.46,
+                                    "content": "C W L ( k _ { 1 } { \\mathrm { - C L } } \\cup k _ { 2 } { \\mathrm { - I C } } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        351,
+                                        563,
+                                        369,
+                                        575
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        370,
+                                        563,
+                                        456,
+                                        574
+                                    ],
+                                    "score": 0.64,
+                                    "content": "C W L ( k _ { 1 } { \\mathrm { - C L } } \\cup k _ { 2 } { \\mathrm { - C } } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        457,
+                                        563,
+                                        505,
+                                        575
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are strictly",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                573,
+                                302,
+                                586
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        573,
+                                        209,
+                                        586
+                                    ],
+                                    "score": 1.0,
+                                    "content": "more powerful than SWL",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        209,
+                                        574,
+                                        220,
+                                        584
+                                    ],
+                                    "score": 0.75,
+                                    "content": "k _ { 1 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        220,
+                                        573,
+                                        268,
+                                        586
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-CL) for all",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        269,
+                                        574,
+                                        298,
+                                        585
+                                    ],
+                                    "score": 0.9,
+                                    "content": "k _ { 2 } \\geq 5",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        298,
+                                        573,
+                                        302,
+                                        586
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        }
+                    ],
+                    "index": 44,
+                    "bbox_fs": [
+                        105,
+                        550,
+                        505,
+                        586
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        599,
+                        367,
+                        613
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                598,
+                                368,
+                                615
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        598,
+                                        368,
+                                        615
+                                    ],
+                                    "score": 1.0,
+                                    "content": "4 Molecular Message Passing with CW Networks",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        }
+                    ],
+                    "index": 46
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        623,
+                        505,
+                        722
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                623,
+                                506,
+                                636
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        623,
+                                        506,
+                                        636
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We now describe CW Networks with an applied focus on molecular graphs to ground the discussion.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 47
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                634,
+                                506,
+                                647
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        634,
+                                        506,
+                                        647
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Therefore, from now on we assume the use of the skeleton-preserving lifting transformation that",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 48
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                645,
+                                505,
+                                658
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        645,
+                                        470,
+                                        658
+                                    ],
+                                    "score": 1.0,
+                                    "content": "attaches 2-cells to all the induced cycles (i.e. chordless cycles) in the graph as in Figure",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        471,
+                                        645,
+                                        482,
+                                        658
+                                    ],
+                                    "score": 0.46,
+                                    "content": "2 .",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        483,
+                                        645,
+                                        505,
+                                        657
+                                    ],
+                                    "score": 1.0,
+                                    "content": "This",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 49
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                655,
+                                505,
+                                669
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        655,
+                                        505,
+                                        669
+                                    ],
+                                    "score": 1.0,
+                                    "content": "leads to a message passing procedure involving atoms (vertices / 0-cells), the bonds between atoms",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 50
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                667,
+                                505,
+                                679
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        667,
+                                        505,
+                                        679
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(edges / 1-cells) and chemical rings (induced cycles / 2-cells). Additionally, in virtue of Theorem",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 51
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                677,
+                                506,
+                                692
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        678,
+                                        114,
+                                        691
+                                    ],
+                                    "score": 0.5,
+                                    "content": "\\underline { { \\overline { { 7 } } } } ,",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        114,
+                                        677,
+                                        506,
+                                        692
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", we consider only the boundary and upper adjacencies between these cells without sacrificing the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 52
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                689,
+                                505,
+                                701
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        689,
+                                        505,
+                                        701
+                                    ],
+                                    "score": 1.0,
+                                    "content": "expressive power. The equations for the other adjacencies, which we do not use, can be found in",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 53
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                700,
+                                505,
+                                713
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        700,
+                                        505,
+                                        713
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Appendix A. We note however, that the theoretical results in this section are general and not particular",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 54
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                711,
+                                371,
+                                723
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        711,
+                                        371,
+                                        723
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to these specific choices of adjacencies and lifting transformation.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 55
+                        }
+                    ],
+                    "index": 51,
+                    "bbox_fs": [
+                        105,
+                        623,
+                        506,
+                        723
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        104,
+                        72,
+                        478,
+                        84
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                70,
+                                480,
+                                87
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        70,
+                                        480,
+                                        87
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Molecular Message Passing The cells in our CW Network receive two types of messages:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        }
+                    ],
+                    "index": 0
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        111,
+                        88,
+                        506,
+                        110
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                111,
+                                88,
+                                506,
+                                110
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        111,
+                                        88,
+                                        506,
+                                        110
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\begin{array} { r } { n _ { \\mathrm { { B } } } ^ { t + 1 } ( \\sigma ) = \\mathrm { A G G } _ { \\tau \\in \\mathcal { B } ( \\sigma ) } \\left( M _ { B } \\big ( h _ { \\sigma } ^ { t } , h _ { \\tau } ^ { t } \\big ) \\right) \\qquad m _ { \\uparrow } ^ { t + 1 } ( \\sigma ) = \\mathrm { A G G } _ { \\tau \\in \\mathcal { N } _ { \\uparrow } ( \\sigma ) , \\delta \\in \\mathcal { C } ( \\sigma , \\tau ) } \\Big ( M _ { \\uparrow } \\big ( h _ { \\sigma } ^ { t } , h _ { \\tau } ^ { t } , h _ { \\delta } ^ { t } \\big ) \\Big ) . } \\end{array}",
+                                    "type": "interline_equation",
+                                    "image_path": "12808d6a3dee80c3fb7dc29e8406e3d8e577d1a1735f53ecc892c97dd6105a11.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 1,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                111,
+                                88,
+                                506,
+                                110
+                            ],
+                            "spans": [],
+                            "index": 1
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        114,
+                        505,
+                        181
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                115,
+                                505,
+                                127
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        115,
+                                        505,
+                                        127
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The first specifies messages from atoms to bonds and from bonds to rings. The second type of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                126,
+                                506,
+                                138
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        126,
+                                        506,
+                                        138
+                                    ],
+                                    "score": 1.0,
+                                    "content": "message, specifies messages between atoms connected by a bond and messages between bonds",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                136,
+                                506,
+                                150
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        136,
+                                        262,
+                                        150
+                                    ],
+                                    "score": 1.0,
+                                    "content": "that are part of the same ring (Figure",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        263,
+                                        137,
+                                        273,
+                                        149
+                                    ],
+                                    "score": 0.68,
+                                    "content": "\\textcircled{6}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        274,
+                                        136,
+                                        506,
+                                        150
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Note that for the second type of adjacency, when two",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                148,
+                                504,
+                                159
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        148,
+                                        504,
+                                        159
+                                    ],
+                                    "score": 1.0,
+                                    "content": "atoms communicate, we include the features of the bond between them. Similarly, when two bonds",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                158,
+                                504,
+                                170
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        158,
+                                        504,
+                                        170
+                                    ],
+                                    "score": 1.0,
+                                    "content": "communicate, we include the features of the ring they communicate through. The update operation",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                169,
+                                483,
+                                182
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        169,
+                                        483,
+                                        182
+                                    ],
+                                    "score": 1.0,
+                                    "content": "takes into account these two types of incoming messages and updates the features of the cells:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        }
+                    ],
+                    "index": 4.5
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        234,
+                        186,
+                        376,
+                        207
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                234,
+                                186,
+                                376,
+                                207
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        234,
+                                        186,
+                                        376,
+                                        207
+                                    ],
+                                    "score": 0.94,
+                                    "content": "h _ { \\sigma } ^ { t + 1 } = U \\Big ( h _ { \\sigma } ^ { t } , m _ { B } ^ { t } ( \\sigma ) , m _ { \\uparrow } ^ { t + 1 } ( \\sigma ) \\Big ) .",
+                                    "type": "interline_equation",
+                                    "image_path": "223e4f9af63809eeefc62a6d123be7d6e8e636e54958be1c6047f1bfa8df0c96.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 8,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                234,
+                                186,
+                                376,
+                                207
+                            ],
+                            "spans": [],
+                            "index": 8
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        211,
+                        505,
+                        234
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                211,
+                                505,
+                                224
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        211,
+                                        299,
+                                        224
+                                    ],
+                                    "score": 1.0,
+                                    "content": "To obtain a global embedding for a cell complex",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        299,
+                                        212,
+                                        309,
+                                        222
+                                    ],
+                                    "score": 0.84,
+                                    "content": "X",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        309,
+                                        211,
+                                        386,
+                                        224
+                                    ],
+                                    "score": 1.0,
+                                    "content": "from a model with",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        386,
+                                        212,
+                                        394,
+                                        222
+                                    ],
+                                    "score": 0.8,
+                                    "content": "L",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        394,
+                                        211,
+                                        505,
+                                        224
+                                    ],
+                                    "score": 1.0,
+                                    "content": "layers, the readout function",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                222,
+                                495,
+                                236
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        222,
+                                        495,
+                                        236
+                                    ],
+                                    "score": 1.0,
+                                    "content": "takes as input the separate multi-sets of features corresponding to the atoms, bonds and the rings:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        }
+                    ],
+                    "index": 9.5
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        164,
+                        239,
+                        446,
+                        255
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                164,
+                                239,
+                                446,
+                                255
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        164,
+                                        239,
+                                        446,
+                                        255
+                                    ],
+                                    "score": 0.89,
+                                    "content": "h _ { X } = \\mathrm { R E A D O U T } ( \\{ h _ { \\sigma } ^ { L } \\} | _ { d i m ( \\sigma ) = 0 } , \\{ \\{ h _ { \\sigma } ^ { L } \\} \\} _ { d i m ( \\sigma ) = 1 } , \\{ \\{ h _ { \\sigma } ^ { L } \\} \\} _ { d i m ( \\sigma ) = 2 } ) .",
+                                    "type": "interline_equation",
+                                    "image_path": "43496a4cccd9f2ca75248a3ad48ca01ed697d3230c15cf436492f4d1f4fa723d.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 11,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                164,
+                                239,
+                                446,
+                                255
+                            ],
+                            "spans": [],
+                            "index": 11
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        263,
+                        408,
+                        330
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                264,
+                                408,
+                                277
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        264,
+                                        408,
+                                        277
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Expressivity Naturally, the ability of CWNs to distinguish non-isomorphic",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                275,
+                                408,
+                                287
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        275,
+                                        408,
+                                        287
+                                    ],
+                                    "score": 1.0,
+                                    "content": "regular cell complexes is bounded by CWL. Similarly to GNNs and WL,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                285,
+                                408,
+                                299
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        285,
+                                        408,
+                                        299
+                                    ],
+                                    "score": 1.0,
+                                    "content": "CWNs can also be shown to be as powerful as CWL as long as they are",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                297,
+                                409,
+                                310
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        297,
+                                        409,
+                                        310
+                                    ],
+                                    "score": 1.0,
+                                    "content": "equipped with a sufficient number of layers and the parametric local aggre-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                308,
+                                408,
+                                321
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        308,
+                                        408,
+                                        321
+                                    ],
+                                    "score": 1.0,
+                                    "content": "gators they use can learn to be injective. Multiple such multi-set aggregators",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                107,
+                                318,
+                                392,
+                                331
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        318,
+                                        123,
+                                        331
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[20,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        123,
+                                        318,
+                                        140,
+                                        330
+                                    ],
+                                    "score": 0.29,
+                                    "content": "\\textcircled { 7 4 } |",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        140,
+                                        318,
+                                        392,
+                                        331
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are known to exist and can be directly employed in our model.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        }
+                    ],
+                    "index": 14.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        334,
+                        408,
+                        367
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                333,
+                                408,
+                                346
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        333,
+                                        408,
+                                        346
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Theorem 18. CW Networks are at most as powerful as CWL. Additionally,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                344,
+                                408,
+                                357
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        344,
+                                        408,
+                                        357
+                                    ],
+                                    "score": 1.0,
+                                    "content": "when using injective neighbourhood aggregators and a sufficient number of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                107,
+                                356,
+                                264,
+                                368
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        356,
+                                        264,
+                                        368
+                                    ],
+                                    "score": 1.0,
+                                    "content": "layers, CWNs are as powerful as CWL.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        }
+                    ],
+                    "index": 19
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        375,
+                        408,
+                        442
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                374,
+                                409,
+                                388
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        374,
+                                        147,
+                                        388
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Corollary",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        147,
+                                        375,
+                                        162,
+                                        388
+                                    ],
+                                    "score": 0.53,
+                                    "content": "\\boxed { 1 5 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        162,
+                                        374,
+                                        409,
+                                        388
+                                    ],
+                                    "score": 1.0,
+                                    "content": "states that CWL is strictly more powerful than the standard",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                385,
+                                408,
+                                399
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        385,
+                                        408,
+                                        399
+                                    ],
+                                    "score": 1.0,
+                                    "content": "WL when the lifting procedure attaches 2-cells to induced cycles of size",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                398,
+                                408,
+                                409
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        398,
+                                        132,
+                                        408
+                                    ],
+                                    "score": 0.89,
+                                    "content": "k \\geq 3",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        132,
+                                        398,
+                                        408,
+                                        409
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". As a consequence of Theorem 18, this result also holds for molecular",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                408,
+                                407,
+                                421
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        408,
+                                        399,
+                                        421
+                                    ],
+                                    "score": 1.0,
+                                    "content": "message passing CWNs equipped with injective aggregators. In practice,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        400,
+                                        409,
+                                        407,
+                                        419
+                                    ],
+                                    "score": 0.64,
+                                    "content": "k",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                418,
+                                408,
+                                432
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        418,
+                                        408,
+                                        432
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is to be considered as a standard hyperparameter, and its choice can either be",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                430,
+                                408,
+                                443
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        430,
+                                        408,
+                                        443
+                                    ],
+                                    "score": 1.0,
+                                    "content": "driven by validation set performance, or by domain knowledge (if available).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        }
+                    ],
+                    "index": 24.5
+                },
+                {
+                    "type": "image",
+                    "bbox": [
+                        415,
+                        266,
+                        505,
+                        435
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                415,
+                                266,
+                                505,
+                                435
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        415,
+                                        266,
+                                        505,
+                                        435
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                415,
+                                                266,
+                                                505,
+                                                435
+                                            ],
+                                            "score": 0.958,
+                                            "type": "image",
+                                            "image_path": "5dfc800ff4d91f10c1b21ac965ab4a1939dc1993dca160cda884fff19197fd1f.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 21,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        415,
+                                        266,
+                                        505,
+                                        435
+                                    ],
+                                    "spans": [],
+                                    "index": 21
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                415,
+                                441,
+                                505,
+                                550
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        414,
+                                        440,
+                                        506,
+                                        452
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                414,
+                                                440,
+                                                506,
+                                                452
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 6: Hierarchi-",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 37
+                                },
+                                {
+                                    "bbox": [
+                                        414,
+                                        451,
+                                        505,
+                                        463
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                414,
+                                                451,
+                                                505,
+                                                463
+                                            ],
+                                            "score": 1.0,
+                                            "content": "cal depiction of the",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 38
+                                },
+                                {
+                                    "bbox": [
+                                        415,
+                                        463,
+                                        506,
+                                        474
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                415,
+                                                463,
+                                                506,
+                                                474
+                                            ],
+                                            "score": 1.0,
+                                            "content": "message passing pro-",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 39
+                                },
+                                {
+                                    "bbox": [
+                                        415,
+                                        473,
+                                        508,
+                                        486
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                415,
+                                                473,
+                                                449,
+                                                484
+                                            ],
+                                            "score": 1.0,
+                                            "content": "cedure.",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                455,
+                                                473,
+                                                508,
+                                                486
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Orange ar-",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 40
+                                },
+                                {
+                                    "bbox": [
+                                        414,
+                                        484,
+                                        507,
+                                        495
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                414,
+                                                484,
+                                                507,
+                                                495
+                                            ],
+                                            "score": 1.0,
+                                            "content": "rows indicate bound-",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 41
+                                },
+                                {
+                                    "bbox": [
+                                        414,
+                                        496,
+                                        506,
+                                        506
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                414,
+                                                496,
+                                                506,
+                                                506
+                                            ],
+                                            "score": 1.0,
+                                            "content": "ary messages received",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 42
+                                },
+                                {
+                                    "bbox": [
+                                        414,
+                                        506,
+                                        505,
+                                        518
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                414,
+                                                506,
+                                                447,
+                                                518
+                                            ],
+                                            "score": 1.0,
+                                            "content": "by cells",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                447,
+                                                508,
+                                                454,
+                                                516
+                                            ],
+                                            "score": 0.78,
+                                            "content": "\\sigma",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                455,
+                                                506,
+                                                471,
+                                                518
+                                            ],
+                                            "score": 1.0,
+                                            "content": "and",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                471,
+                                                508,
+                                                478,
+                                                516
+                                            ],
+                                            "score": 0.74,
+                                            "content": "\\tau",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                478,
+                                                506,
+                                                505,
+                                                518
+                                            ],
+                                            "score": 1.0,
+                                            "content": ", while",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 43
+                                },
+                                {
+                                    "bbox": [
+                                        414,
+                                        516,
+                                        506,
+                                        530
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                414,
+                                                516,
+                                                506,
+                                                530
+                                            ],
+                                            "score": 1.0,
+                                            "content": "blue ones show upper",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 44
+                                },
+                                {
+                                    "bbox": [
+                                        414,
+                                        528,
+                                        506,
+                                        541
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                414,
+                                                528,
+                                                506,
+                                                541
+                                            ],
+                                            "score": 1.0,
+                                            "content": "messages received by",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 45
+                                },
+                                {
+                                    "bbox": [
+                                        415,
+                                        538,
+                                        471,
+                                        550
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                415,
+                                                538,
+                                                436,
+                                                550
+                                            ],
+                                            "score": 1.0,
+                                            "content": "cells",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                436,
+                                                540,
+                                                443,
+                                                549
+                                            ],
+                                            "score": 0.75,
+                                            "content": "\\tau",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                443,
+                                                538,
+                                                461,
+                                                550
+                                            ],
+                                            "score": 1.0,
+                                            "content": "and",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                461,
+                                                539,
+                                                467,
+                                                549
+                                            ],
+                                            "score": 0.75,
+                                            "content": "\\delta",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                468,
+                                                538,
+                                                471,
+                                                550
+                                            ],
+                                            "score": 1.0,
+                                            "content": ".",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 46
+                                }
+                            ],
+                            "index": 41.5
+                        }
+                    ],
+                    "index": 31.25
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        453,
+                        407,
+                        542
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                452,
+                                410,
+                                466
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        452,
+                                        227,
+                                        466
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Symmetries Given a graph",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        227,
+                                        454,
+                                        237,
+                                        464
+                                    ],
+                                    "score": 0.78,
+                                    "content": "G",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        237,
+                                        452,
+                                        331,
+                                        466
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with adjacency matrix",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        331,
+                                        454,
+                                        342,
+                                        464
+                                    ],
+                                    "score": 0.63,
+                                    "content": "\\pmb { A }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        342,
+                                        452,
+                                        410,
+                                        466
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and feature ma-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                463,
+                                407,
+                                478
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        463,
+                                        124,
+                                        478
+                                    ],
+                                    "score": 1.0,
+                                    "content": "trix",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        124,
+                                        465,
+                                        135,
+                                        474
+                                    ],
+                                    "score": 0.78,
+                                    "content": "\\boldsymbol { X }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        136,
+                                        463,
+                                        186,
+                                        478
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", a function",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        186,
+                                        465,
+                                        194,
+                                        476
+                                    ],
+                                    "score": 0.85,
+                                    "content": "f",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        194,
+                                        463,
+                                        347,
+                                        478
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is (node) permutation equivariant if",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        347,
+                                        464,
+                                        407,
+                                        477
+                                    ],
+                                    "score": 0.92,
+                                    "content": "P f ( A , X ) =",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                107,
+                                474,
+                                409,
+                                488
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        475,
+                                        177,
+                                        487
+                                    ],
+                                    "score": 0.91,
+                                    "content": "f ( P A P ^ { T } , P X )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        178,
+                                        474,
+                                        293,
+                                        488
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", for any permutation matrix",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        294,
+                                        476,
+                                        303,
+                                        485
+                                    ],
+                                    "score": 0.76,
+                                    "content": "_ { r }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        304,
+                                        474,
+                                        409,
+                                        488
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". GNN layers respect this",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                486,
+                                408,
+                                498
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        486,
+                                        408,
+                                        498
+                                    ],
+                                    "score": 1.0,
+                                    "content": "equation, which ensures they compute the same functions up to a permutation",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                497,
+                                408,
+                                509
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        497,
+                                        408,
+                                        509
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(i.e. relabeling) of the nodes. Similarly, it can be shown that CW Networks",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                507,
+                                408,
+                                521
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        507,
+                                        408,
+                                        521
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are equivariant with respect to permutations of the cells and corresponding",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                519,
+                                408,
+                                530
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        519,
+                                        262,
+                                        530
+                                    ],
+                                    "score": 1.0,
+                                    "content": "permutations of the boundary relations",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        262,
+                                        520,
+                                        289,
+                                        529
+                                    ],
+                                    "score": 0.88,
+                                    "content": "\\sigma \\prec \\tau",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        289,
+                                        519,
+                                        408,
+                                        530
+                                    ],
+                                    "score": 1.0,
+                                    "content": "between cells. We define this",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                530,
+                                320,
+                                542
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        530,
+                                        320,
+                                        542
+                                    ],
+                                    "score": 1.0,
+                                    "content": "notion of equivariance more formally in Appendix C.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        }
+                    ],
+                    "index": 31.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        545,
+                        378,
+                        556
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                543,
+                                379,
+                                558
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        543,
+                                        379,
+                                        558
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Theorem 19. CW Network layers are cell permutation equivariant.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        }
+                    ],
+                    "index": 36
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        567,
+                        506,
+                        667
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                566,
+                                505,
+                                581
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        566,
+                                        505,
+                                        581
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Long-Range Interactions Several graph-related tasks require the ability to capture long-range",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 47
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                579,
+                                505,
+                                591
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        579,
+                                        505,
+                                        591
+                                    ],
+                                    "score": 1.0,
+                                    "content": "interactions between nodes. For instance, certain molecular properties depend on atoms placed on the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 48
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                589,
+                                507,
+                                603
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        589,
+                                        507,
+                                        603
+                                    ],
+                                    "score": 1.0,
+                                    "content": "opposite sides of a ring [31, 60]. As a consequence of the coupling between the input and computa-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 49
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                600,
+                                506,
+                                614
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        600,
+                                        162,
+                                        614
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tional graphs,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        163,
+                                        601,
+                                        171,
+                                        611
+                                    ],
+                                    "score": 0.71,
+                                    "content": "L",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        171,
+                                        600,
+                                        506,
+                                        614
+                                    ],
+                                    "score": 1.0,
+                                    "content": "message passing operations are necessary in GNNs to let a node receive information",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 50
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                610,
+                                506,
+                                624
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        610,
+                                        140,
+                                        624
+                                    ],
+                                    "score": 1.0,
+                                    "content": "from an",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        140,
+                                        612,
+                                        148,
+                                        621
+                                    ],
+                                    "score": 0.8,
+                                    "content": "L",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        149,
+                                        610,
+                                        506,
+                                        624
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-hops distant node. In contrast, our hierarchical message passing scheme requires at most",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 51
+                        },
+                        {
+                            "bbox": [
+                                107,
+                                621,
+                                506,
+                                635
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        623,
+                                        115,
+                                        632
+                                    ],
+                                    "score": 0.77,
+                                    "content": "L",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        115,
+                                        621,
+                                        506,
+                                        635
+                                    ],
+                                    "score": 1.0,
+                                    "content": "layers since 2-cells create shortcuts. For example, a constant number of CWN layers (3) is enough",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 52
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                632,
+                                506,
+                                646
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        632,
+                                        506,
+                                        646
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to capture dependencies between atoms on the opposite sides of a ring, independently of the ring size.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 53
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                643,
+                                506,
+                                657
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        643,
+                                        147,
+                                        657
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In Section",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        148,
+                                        644,
+                                        164,
+                                        655
+                                    ],
+                                    "score": 0.71,
+                                    "content": "5 . { \\overset { \\cdot } { 1 } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        164,
+                                        643,
+                                        506,
+                                        657
+                                    ],
+                                    "score": 1.0,
+                                    "content": "we verify this in a controlled scenario. Additional experiments on real world graphs in",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 54
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                654,
+                                507,
+                                668
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        654,
+                                        137,
+                                        668
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Section",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        138,
+                                        655,
+                                        154,
+                                        667
+                                    ],
+                                    "score": 0.66,
+                                    "content": "5 . 2",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        155,
+                                        654,
+                                        507,
+                                        668
+                                    ],
+                                    "score": 1.0,
+                                    "content": "confirm that it can achieve state-of-the-art performance with a limited number of layers.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 55
+                        }
+                    ],
+                    "index": 51
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        678,
+                        505,
+                        722
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                677,
+                                505,
+                                691
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        677,
+                                        505,
+                                        691
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Anisotropic Filters Due to the lack of a canonical ordering between neighbours, many common",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 56
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                688,
+                                506,
+                                702
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        688,
+                                        506,
+                                        702
+                                    ],
+                                    "score": 1.0,
+                                    "content": "GNNs use symmetric convolutional kernels, resulting in isotropic filters treating neighbours equally.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 57
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                700,
+                                506,
+                                712
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        700,
+                                        506,
+                                        712
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Recent works have proposed to address this limitation by employing additional structural informa-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 58
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                710,
+                                505,
+                                724
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        710,
+                                        505,
+                                        724
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tion [6, 10]. CWNs also implicitly achieve this form of anisotropy by integrating information from",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 59
+                        }
+                    ],
+                    "index": 57.5
+                }
+            ],
+            "page_idx": 5,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        302,
+                        742,
+                        309,
+                        750
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                302,
+                                741,
+                                310,
+                                752
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        302,
+                                        741,
+                                        310,
+                                        752
+                                    ],
+                                    "score": 1.0,
+                                    "content": "6",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        104,
+                        72,
+                        478,
+                        84
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                70,
+                                480,
+                                87
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        70,
+                                        480,
+                                        87
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Molecular Message Passing The cells in our CW Network receive two types of messages:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        }
+                    ],
+                    "index": 0,
+                    "bbox_fs": [
+                        104,
+                        70,
+                        480,
+                        87
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        111,
+                        88,
+                        506,
+                        110
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                111,
+                                88,
+                                506,
+                                110
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        111,
+                                        88,
+                                        506,
+                                        110
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\begin{array} { r } { n _ { \\mathrm { { B } } } ^ { t + 1 } ( \\sigma ) = \\mathrm { A G G } _ { \\tau \\in \\mathcal { B } ( \\sigma ) } \\left( M _ { B } \\big ( h _ { \\sigma } ^ { t } , h _ { \\tau } ^ { t } \\big ) \\right) \\qquad m _ { \\uparrow } ^ { t + 1 } ( \\sigma ) = \\mathrm { A G G } _ { \\tau \\in \\mathcal { N } _ { \\uparrow } ( \\sigma ) , \\delta \\in \\mathcal { C } ( \\sigma , \\tau ) } \\Big ( M _ { \\uparrow } \\big ( h _ { \\sigma } ^ { t } , h _ { \\tau } ^ { t } , h _ { \\delta } ^ { t } \\big ) \\Big ) . } \\end{array}",
+                                    "type": "interline_equation",
+                                    "image_path": "12808d6a3dee80c3fb7dc29e8406e3d8e577d1a1735f53ecc892c97dd6105a11.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 1,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                111,
+                                88,
+                                506,
+                                110
+                            ],
+                            "spans": [],
+                            "index": 1
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        114,
+                        505,
+                        181
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                115,
+                                505,
+                                127
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        115,
+                                        505,
+                                        127
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The first specifies messages from atoms to bonds and from bonds to rings. The second type of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                126,
+                                506,
+                                138
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        126,
+                                        506,
+                                        138
+                                    ],
+                                    "score": 1.0,
+                                    "content": "message, specifies messages between atoms connected by a bond and messages between bonds",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                136,
+                                506,
+                                150
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        136,
+                                        262,
+                                        150
+                                    ],
+                                    "score": 1.0,
+                                    "content": "that are part of the same ring (Figure",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        263,
+                                        137,
+                                        273,
+                                        149
+                                    ],
+                                    "score": 0.68,
+                                    "content": "\\textcircled{6}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        274,
+                                        136,
+                                        506,
+                                        150
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Note that for the second type of adjacency, when two",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                148,
+                                504,
+                                159
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        148,
+                                        504,
+                                        159
+                                    ],
+                                    "score": 1.0,
+                                    "content": "atoms communicate, we include the features of the bond between them. Similarly, when two bonds",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                158,
+                                504,
+                                170
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        158,
+                                        504,
+                                        170
+                                    ],
+                                    "score": 1.0,
+                                    "content": "communicate, we include the features of the ring they communicate through. The update operation",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                169,
+                                483,
+                                182
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        169,
+                                        483,
+                                        182
+                                    ],
+                                    "score": 1.0,
+                                    "content": "takes into account these two types of incoming messages and updates the features of the cells:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        }
+                    ],
+                    "index": 4.5,
+                    "bbox_fs": [
+                        105,
+                        115,
+                        506,
+                        182
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        234,
+                        186,
+                        376,
+                        207
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                234,
+                                186,
+                                376,
+                                207
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        234,
+                                        186,
+                                        376,
+                                        207
+                                    ],
+                                    "score": 0.94,
+                                    "content": "h _ { \\sigma } ^ { t + 1 } = U \\Big ( h _ { \\sigma } ^ { t } , m _ { B } ^ { t } ( \\sigma ) , m _ { \\uparrow } ^ { t + 1 } ( \\sigma ) \\Big ) .",
+                                    "type": "interline_equation",
+                                    "image_path": "223e4f9af63809eeefc62a6d123be7d6e8e636e54958be1c6047f1bfa8df0c96.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 8,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                234,
+                                186,
+                                376,
+                                207
+                            ],
+                            "spans": [],
+                            "index": 8
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        211,
+                        505,
+                        234
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                211,
+                                505,
+                                224
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        211,
+                                        299,
+                                        224
+                                    ],
+                                    "score": 1.0,
+                                    "content": "To obtain a global embedding for a cell complex",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        299,
+                                        212,
+                                        309,
+                                        222
+                                    ],
+                                    "score": 0.84,
+                                    "content": "X",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        309,
+                                        211,
+                                        386,
+                                        224
+                                    ],
+                                    "score": 1.0,
+                                    "content": "from a model with",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        386,
+                                        212,
+                                        394,
+                                        222
+                                    ],
+                                    "score": 0.8,
+                                    "content": "L",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        394,
+                                        211,
+                                        505,
+                                        224
+                                    ],
+                                    "score": 1.0,
+                                    "content": "layers, the readout function",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                222,
+                                495,
+                                236
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        222,
+                                        495,
+                                        236
+                                    ],
+                                    "score": 1.0,
+                                    "content": "takes as input the separate multi-sets of features corresponding to the atoms, bonds and the rings:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        }
+                    ],
+                    "index": 9.5,
+                    "bbox_fs": [
+                        105,
+                        211,
+                        505,
+                        236
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        164,
+                        239,
+                        446,
+                        255
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                164,
+                                239,
+                                446,
+                                255
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        164,
+                                        239,
+                                        446,
+                                        255
+                                    ],
+                                    "score": 0.89,
+                                    "content": "h _ { X } = \\mathrm { R E A D O U T } ( \\{ h _ { \\sigma } ^ { L } \\} | _ { d i m ( \\sigma ) = 0 } , \\{ \\{ h _ { \\sigma } ^ { L } \\} \\} _ { d i m ( \\sigma ) = 1 } , \\{ \\{ h _ { \\sigma } ^ { L } \\} \\} _ { d i m ( \\sigma ) = 2 } ) .",
+                                    "type": "interline_equation",
+                                    "image_path": "43496a4cccd9f2ca75248a3ad48ca01ed697d3230c15cf436492f4d1f4fa723d.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 11,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                164,
+                                239,
+                                446,
+                                255
+                            ],
+                            "spans": [],
+                            "index": 11
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        263,
+                        408,
+                        330
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                264,
+                                408,
+                                277
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        264,
+                                        408,
+                                        277
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Expressivity Naturally, the ability of CWNs to distinguish non-isomorphic",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                275,
+                                408,
+                                287
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        275,
+                                        408,
+                                        287
+                                    ],
+                                    "score": 1.0,
+                                    "content": "regular cell complexes is bounded by CWL. Similarly to GNNs and WL,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                285,
+                                408,
+                                299
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        285,
+                                        408,
+                                        299
+                                    ],
+                                    "score": 1.0,
+                                    "content": "CWNs can also be shown to be as powerful as CWL as long as they are",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                297,
+                                409,
+                                310
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        297,
+                                        409,
+                                        310
+                                    ],
+                                    "score": 1.0,
+                                    "content": "equipped with a sufficient number of layers and the parametric local aggre-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                308,
+                                408,
+                                321
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        308,
+                                        408,
+                                        321
+                                    ],
+                                    "score": 1.0,
+                                    "content": "gators they use can learn to be injective. Multiple such multi-set aggregators",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                107,
+                                318,
+                                392,
+                                331
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        318,
+                                        123,
+                                        331
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[20,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        123,
+                                        318,
+                                        140,
+                                        330
+                                    ],
+                                    "score": 0.29,
+                                    "content": "\\textcircled { 7 4 } |",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        140,
+                                        318,
+                                        392,
+                                        331
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are known to exist and can be directly employed in our model.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        }
+                    ],
+                    "index": 14.5,
+                    "bbox_fs": [
+                        105,
+                        264,
+                        409,
+                        331
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        334,
+                        408,
+                        367
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                333,
+                                408,
+                                346
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        333,
+                                        408,
+                                        346
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Theorem 18. CW Networks are at most as powerful as CWL. Additionally,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                344,
+                                408,
+                                357
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        344,
+                                        408,
+                                        357
+                                    ],
+                                    "score": 1.0,
+                                    "content": "when using injective neighbourhood aggregators and a sufficient number of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                107,
+                                356,
+                                264,
+                                368
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        356,
+                                        264,
+                                        368
+                                    ],
+                                    "score": 1.0,
+                                    "content": "layers, CWNs are as powerful as CWL.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        }
+                    ],
+                    "index": 19,
+                    "bbox_fs": [
+                        105,
+                        333,
+                        408,
+                        368
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        375,
+                        408,
+                        442
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                374,
+                                409,
+                                388
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        374,
+                                        147,
+                                        388
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Corollary",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        147,
+                                        375,
+                                        162,
+                                        388
+                                    ],
+                                    "score": 0.53,
+                                    "content": "\\boxed { 1 5 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        162,
+                                        374,
+                                        409,
+                                        388
+                                    ],
+                                    "score": 1.0,
+                                    "content": "states that CWL is strictly more powerful than the standard",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                385,
+                                408,
+                                399
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        385,
+                                        408,
+                                        399
+                                    ],
+                                    "score": 1.0,
+                                    "content": "WL when the lifting procedure attaches 2-cells to induced cycles of size",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                398,
+                                408,
+                                409
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        398,
+                                        132,
+                                        408
+                                    ],
+                                    "score": 0.89,
+                                    "content": "k \\geq 3",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        132,
+                                        398,
+                                        408,
+                                        409
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". As a consequence of Theorem 18, this result also holds for molecular",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                408,
+                                407,
+                                421
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        408,
+                                        399,
+                                        421
+                                    ],
+                                    "score": 1.0,
+                                    "content": "message passing CWNs equipped with injective aggregators. In practice,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        400,
+                                        409,
+                                        407,
+                                        419
+                                    ],
+                                    "score": 0.64,
+                                    "content": "k",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                418,
+                                408,
+                                432
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        418,
+                                        408,
+                                        432
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is to be considered as a standard hyperparameter, and its choice can either be",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                430,
+                                408,
+                                443
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        430,
+                                        408,
+                                        443
+                                    ],
+                                    "score": 1.0,
+                                    "content": "driven by validation set performance, or by domain knowledge (if available).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        }
+                    ],
+                    "index": 24.5,
+                    "bbox_fs": [
+                        105,
+                        374,
+                        409,
+                        443
+                    ]
+                },
+                {
+                    "type": "image",
+                    "bbox": [
+                        415,
+                        266,
+                        505,
+                        435
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                415,
+                                266,
+                                505,
+                                435
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        415,
+                                        266,
+                                        505,
+                                        435
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                415,
+                                                266,
+                                                505,
+                                                435
+                                            ],
+                                            "score": 0.958,
+                                            "type": "image",
+                                            "image_path": "5dfc800ff4d91f10c1b21ac965ab4a1939dc1993dca160cda884fff19197fd1f.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 21,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        415,
+                                        266,
+                                        505,
+                                        435
+                                    ],
+                                    "spans": [],
+                                    "index": 21
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                415,
+                                441,
+                                505,
+                                550
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        414,
+                                        440,
+                                        506,
+                                        452
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                414,
+                                                440,
+                                                506,
+                                                452
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 6: Hierarchi-",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 37
+                                },
+                                {
+                                    "bbox": [
+                                        414,
+                                        451,
+                                        505,
+                                        463
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                414,
+                                                451,
+                                                505,
+                                                463
+                                            ],
+                                            "score": 1.0,
+                                            "content": "cal depiction of the",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 38
+                                },
+                                {
+                                    "bbox": [
+                                        415,
+                                        463,
+                                        506,
+                                        474
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                415,
+                                                463,
+                                                506,
+                                                474
+                                            ],
+                                            "score": 1.0,
+                                            "content": "message passing pro-",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 39
+                                },
+                                {
+                                    "bbox": [
+                                        415,
+                                        473,
+                                        508,
+                                        486
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                415,
+                                                473,
+                                                449,
+                                                484
+                                            ],
+                                            "score": 1.0,
+                                            "content": "cedure.",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                455,
+                                                473,
+                                                508,
+                                                486
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Orange ar-",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 40
+                                },
+                                {
+                                    "bbox": [
+                                        414,
+                                        484,
+                                        507,
+                                        495
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                414,
+                                                484,
+                                                507,
+                                                495
+                                            ],
+                                            "score": 1.0,
+                                            "content": "rows indicate bound-",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 41
+                                },
+                                {
+                                    "bbox": [
+                                        414,
+                                        496,
+                                        506,
+                                        506
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                414,
+                                                496,
+                                                506,
+                                                506
+                                            ],
+                                            "score": 1.0,
+                                            "content": "ary messages received",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 42
+                                },
+                                {
+                                    "bbox": [
+                                        414,
+                                        506,
+                                        505,
+                                        518
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                414,
+                                                506,
+                                                447,
+                                                518
+                                            ],
+                                            "score": 1.0,
+                                            "content": "by cells",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                447,
+                                                508,
+                                                454,
+                                                516
+                                            ],
+                                            "score": 0.78,
+                                            "content": "\\sigma",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                455,
+                                                506,
+                                                471,
+                                                518
+                                            ],
+                                            "score": 1.0,
+                                            "content": "and",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                471,
+                                                508,
+                                                478,
+                                                516
+                                            ],
+                                            "score": 0.74,
+                                            "content": "\\tau",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                478,
+                                                506,
+                                                505,
+                                                518
+                                            ],
+                                            "score": 1.0,
+                                            "content": ", while",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 43
+                                },
+                                {
+                                    "bbox": [
+                                        414,
+                                        516,
+                                        506,
+                                        530
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                414,
+                                                516,
+                                                506,
+                                                530
+                                            ],
+                                            "score": 1.0,
+                                            "content": "blue ones show upper",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 44
+                                },
+                                {
+                                    "bbox": [
+                                        414,
+                                        528,
+                                        506,
+                                        541
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                414,
+                                                528,
+                                                506,
+                                                541
+                                            ],
+                                            "score": 1.0,
+                                            "content": "messages received by",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 45
+                                },
+                                {
+                                    "bbox": [
+                                        415,
+                                        538,
+                                        471,
+                                        550
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                415,
+                                                538,
+                                                436,
+                                                550
+                                            ],
+                                            "score": 1.0,
+                                            "content": "cells",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                436,
+                                                540,
+                                                443,
+                                                549
+                                            ],
+                                            "score": 0.75,
+                                            "content": "\\tau",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                443,
+                                                538,
+                                                461,
+                                                550
+                                            ],
+                                            "score": 1.0,
+                                            "content": "and",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                461,
+                                                539,
+                                                467,
+                                                549
+                                            ],
+                                            "score": 0.75,
+                                            "content": "\\delta",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                468,
+                                                538,
+                                                471,
+                                                550
+                                            ],
+                                            "score": 1.0,
+                                            "content": ".",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 46
+                                }
+                            ],
+                            "index": 41.5
+                        }
+                    ],
+                    "index": 31.25
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        453,
+                        407,
+                        542
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                452,
+                                410,
+                                466
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        452,
+                                        227,
+                                        466
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Symmetries Given a graph",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        227,
+                                        454,
+                                        237,
+                                        464
+                                    ],
+                                    "score": 0.78,
+                                    "content": "G",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        237,
+                                        452,
+                                        331,
+                                        466
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with adjacency matrix",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        331,
+                                        454,
+                                        342,
+                                        464
+                                    ],
+                                    "score": 0.63,
+                                    "content": "\\pmb { A }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        342,
+                                        452,
+                                        410,
+                                        466
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and feature ma-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                463,
+                                407,
+                                478
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        463,
+                                        124,
+                                        478
+                                    ],
+                                    "score": 1.0,
+                                    "content": "trix",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        124,
+                                        465,
+                                        135,
+                                        474
+                                    ],
+                                    "score": 0.78,
+                                    "content": "\\boldsymbol { X }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        136,
+                                        463,
+                                        186,
+                                        478
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", a function",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        186,
+                                        465,
+                                        194,
+                                        476
+                                    ],
+                                    "score": 0.85,
+                                    "content": "f",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        194,
+                                        463,
+                                        347,
+                                        478
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is (node) permutation equivariant if",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        347,
+                                        464,
+                                        407,
+                                        477
+                                    ],
+                                    "score": 0.92,
+                                    "content": "P f ( A , X ) =",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                107,
+                                474,
+                                409,
+                                488
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        475,
+                                        177,
+                                        487
+                                    ],
+                                    "score": 0.91,
+                                    "content": "f ( P A P ^ { T } , P X )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        178,
+                                        474,
+                                        293,
+                                        488
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", for any permutation matrix",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        294,
+                                        476,
+                                        303,
+                                        485
+                                    ],
+                                    "score": 0.76,
+                                    "content": "_ { r }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        304,
+                                        474,
+                                        409,
+                                        488
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". GNN layers respect this",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                486,
+                                408,
+                                498
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        486,
+                                        408,
+                                        498
+                                    ],
+                                    "score": 1.0,
+                                    "content": "equation, which ensures they compute the same functions up to a permutation",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                497,
+                                408,
+                                509
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        497,
+                                        408,
+                                        509
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(i.e. relabeling) of the nodes. Similarly, it can be shown that CW Networks",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                507,
+                                408,
+                                521
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        507,
+                                        408,
+                                        521
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are equivariant with respect to permutations of the cells and corresponding",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                519,
+                                408,
+                                530
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        519,
+                                        262,
+                                        530
+                                    ],
+                                    "score": 1.0,
+                                    "content": "permutations of the boundary relations",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        262,
+                                        520,
+                                        289,
+                                        529
+                                    ],
+                                    "score": 0.88,
+                                    "content": "\\sigma \\prec \\tau",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        289,
+                                        519,
+                                        408,
+                                        530
+                                    ],
+                                    "score": 1.0,
+                                    "content": "between cells. We define this",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                530,
+                                320,
+                                542
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        530,
+                                        320,
+                                        542
+                                    ],
+                                    "score": 1.0,
+                                    "content": "notion of equivariance more formally in Appendix C.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        }
+                    ],
+                    "index": 31.5,
+                    "bbox_fs": [
+                        105,
+                        452,
+                        410,
+                        542
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        545,
+                        378,
+                        556
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                543,
+                                379,
+                                558
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        543,
+                                        379,
+                                        558
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Theorem 19. CW Network layers are cell permutation equivariant.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        }
+                    ],
+                    "index": 36,
+                    "bbox_fs": [
+                        106,
+                        543,
+                        379,
+                        558
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        567,
+                        506,
+                        667
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                566,
+                                505,
+                                581
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        566,
+                                        505,
+                                        581
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Long-Range Interactions Several graph-related tasks require the ability to capture long-range",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 47
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                579,
+                                505,
+                                591
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        579,
+                                        505,
+                                        591
+                                    ],
+                                    "score": 1.0,
+                                    "content": "interactions between nodes. For instance, certain molecular properties depend on atoms placed on the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 48
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                589,
+                                507,
+                                603
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        589,
+                                        507,
+                                        603
+                                    ],
+                                    "score": 1.0,
+                                    "content": "opposite sides of a ring [31, 60]. As a consequence of the coupling between the input and computa-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 49
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                600,
+                                506,
+                                614
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        600,
+                                        162,
+                                        614
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tional graphs,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        163,
+                                        601,
+                                        171,
+                                        611
+                                    ],
+                                    "score": 0.71,
+                                    "content": "L",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        171,
+                                        600,
+                                        506,
+                                        614
+                                    ],
+                                    "score": 1.0,
+                                    "content": "message passing operations are necessary in GNNs to let a node receive information",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 50
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                610,
+                                506,
+                                624
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        610,
+                                        140,
+                                        624
+                                    ],
+                                    "score": 1.0,
+                                    "content": "from an",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        140,
+                                        612,
+                                        148,
+                                        621
+                                    ],
+                                    "score": 0.8,
+                                    "content": "L",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        149,
+                                        610,
+                                        506,
+                                        624
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-hops distant node. In contrast, our hierarchical message passing scheme requires at most",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 51
+                        },
+                        {
+                            "bbox": [
+                                107,
+                                621,
+                                506,
+                                635
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        623,
+                                        115,
+                                        632
+                                    ],
+                                    "score": 0.77,
+                                    "content": "L",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        115,
+                                        621,
+                                        506,
+                                        635
+                                    ],
+                                    "score": 1.0,
+                                    "content": "layers since 2-cells create shortcuts. For example, a constant number of CWN layers (3) is enough",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 52
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                632,
+                                506,
+                                646
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        632,
+                                        506,
+                                        646
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to capture dependencies between atoms on the opposite sides of a ring, independently of the ring size.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 53
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                643,
+                                506,
+                                657
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        643,
+                                        147,
+                                        657
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In Section",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        148,
+                                        644,
+                                        164,
+                                        655
+                                    ],
+                                    "score": 0.71,
+                                    "content": "5 . { \\overset { \\cdot } { 1 } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        164,
+                                        643,
+                                        506,
+                                        657
+                                    ],
+                                    "score": 1.0,
+                                    "content": "we verify this in a controlled scenario. Additional experiments on real world graphs in",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 54
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                654,
+                                507,
+                                668
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        654,
+                                        137,
+                                        668
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Section",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        138,
+                                        655,
+                                        154,
+                                        667
+                                    ],
+                                    "score": 0.66,
+                                    "content": "5 . 2",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        155,
+                                        654,
+                                        507,
+                                        668
+                                    ],
+                                    "score": 1.0,
+                                    "content": "confirm that it can achieve state-of-the-art performance with a limited number of layers.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 55
+                        }
+                    ],
+                    "index": 51,
+                    "bbox_fs": [
+                        105,
+                        566,
+                        507,
+                        668
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        678,
+                        505,
+                        722
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                677,
+                                505,
+                                691
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        677,
+                                        505,
+                                        691
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Anisotropic Filters Due to the lack of a canonical ordering between neighbours, many common",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 56
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                688,
+                                506,
+                                702
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        688,
+                                        506,
+                                        702
+                                    ],
+                                    "score": 1.0,
+                                    "content": "GNNs use symmetric convolutional kernels, resulting in isotropic filters treating neighbours equally.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 57
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                700,
+                                506,
+                                712
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        700,
+                                        506,
+                                        712
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Recent works have proposed to address this limitation by employing additional structural informa-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 58
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                710,
+                                505,
+                                724
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        710,
+                                        505,
+                                        724
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tion [6, 10]. CWNs also implicitly achieve this form of anisotropy by integrating information from",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 59
+                        }
+                    ],
+                    "index": 57.5,
+                    "bbox_fs": [
+                        104,
+                        677,
+                        506,
+                        724
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "image",
+                    "bbox": [
+                        108,
+                        72,
+                        281,
+                        208
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                108,
+                                72,
+                                281,
+                                208
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        108,
+                                        72,
+                                        281,
+                                        208
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                108,
+                                                72,
+                                                281,
+                                                208
+                                            ],
+                                            "score": 0.962,
+                                            "type": "image",
+                                            "image_path": "e7e513d4a4a37f8f0a963dcd756c8d2c678cb9d6e65f18f363336e6494cbaf6d.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 4.5,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        108,
+                                        72,
+                                        281,
+                                        85.6
+                                    ],
+                                    "spans": [],
+                                    "index": 0
+                                },
+                                {
+                                    "bbox": [
+                                        108,
+                                        85.6,
+                                        281,
+                                        99.19999999999999
+                                    ],
+                                    "spans": [],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        108,
+                                        99.19999999999999,
+                                        281,
+                                        112.79999999999998
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                },
+                                {
+                                    "bbox": [
+                                        108,
+                                        112.79999999999998,
+                                        281,
+                                        126.39999999999998
+                                    ],
+                                    "spans": [],
+                                    "index": 3
+                                },
+                                {
+                                    "bbox": [
+                                        108,
+                                        126.39999999999998,
+                                        281,
+                                        139.99999999999997
+                                    ],
+                                    "spans": [],
+                                    "index": 4
+                                },
+                                {
+                                    "bbox": [
+                                        108,
+                                        139.99999999999997,
+                                        281,
+                                        153.59999999999997
+                                    ],
+                                    "spans": [],
+                                    "index": 5
+                                },
+                                {
+                                    "bbox": [
+                                        108,
+                                        153.59999999999997,
+                                        281,
+                                        167.19999999999996
+                                    ],
+                                    "spans": [],
+                                    "index": 6
+                                },
+                                {
+                                    "bbox": [
+                                        108,
+                                        167.19999999999996,
+                                        281,
+                                        180.79999999999995
+                                    ],
+                                    "spans": [],
+                                    "index": 7
+                                },
+                                {
+                                    "bbox": [
+                                        108,
+                                        180.79999999999995,
+                                        281,
+                                        194.39999999999995
+                                    ],
+                                    "spans": [],
+                                    "index": 8
+                                },
+                                {
+                                    "bbox": [
+                                        108,
+                                        194.39999999999995,
+                                        281,
+                                        207.99999999999994
+                                    ],
+                                    "spans": [],
+                                    "index": 9
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 4.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        214,
+                        283,
+                        265
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                214,
+                                285,
+                                226
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        214,
+                                        285,
+                                        226
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(a) RingTransfer Results. Accuracy is over 5 bal-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                225,
+                                284,
+                                236
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        225,
+                                        284,
+                                        236
+                                    ],
+                                    "score": 1.0,
+                                    "content": "anced classes. A score of 0.2 is equivalent to a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                234,
+                                285,
+                                245
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        234,
+                                        285,
+                                        245
+                                    ],
+                                    "score": 1.0,
+                                    "content": "random guess. Error bars show the min and max.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                244,
+                                283,
+                                256
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        244,
+                                        283,
+                                        256
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Our model obtains high-scores in average even",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                254,
+                                273,
+                                266
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        254,
+                                        273,
+                                        266
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for large rings despite using only three layers.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        }
+                    ],
+                    "index": 12
+                },
+                {
+                    "type": "image",
+                    "bbox": [
+                        295,
+                        72,
+                        503,
+                        228
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                295,
+                                72,
+                                503,
+                                228
+                            ],
+                            "group_id": 1,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        295,
+                                        72,
+                                        503,
+                                        228
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                295,
+                                                72,
+                                                503,
+                                                228
+                                            ],
+                                            "score": 0.958,
+                                            "type": "image",
+                                            "image_path": "61b45b8b73568ed72007633ef6dafbc69a26a971d31f6e09c325a41d0ecbc3e1.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 20.5,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        295,
+                                        72,
+                                        503,
+                                        85.0
+                                    ],
+                                    "spans": [],
+                                    "index": 15
+                                },
+                                {
+                                    "bbox": [
+                                        295,
+                                        85.0,
+                                        503,
+                                        98.0
+                                    ],
+                                    "spans": [],
+                                    "index": 16
+                                },
+                                {
+                                    "bbox": [
+                                        295,
+                                        98.0,
+                                        503,
+                                        111.0
+                                    ],
+                                    "spans": [],
+                                    "index": 17
+                                },
+                                {
+                                    "bbox": [
+                                        295,
+                                        111.0,
+                                        503,
+                                        124.0
+                                    ],
+                                    "spans": [],
+                                    "index": 18
+                                },
+                                {
+                                    "bbox": [
+                                        295,
+                                        124.0,
+                                        503,
+                                        137.0
+                                    ],
+                                    "spans": [],
+                                    "index": 19
+                                },
+                                {
+                                    "bbox": [
+                                        295,
+                                        137.0,
+                                        503,
+                                        150.0
+                                    ],
+                                    "spans": [],
+                                    "index": 20
+                                },
+                                {
+                                    "bbox": [
+                                        295,
+                                        150.0,
+                                        503,
+                                        163.0
+                                    ],
+                                    "spans": [],
+                                    "index": 21
+                                },
+                                {
+                                    "bbox": [
+                                        295,
+                                        163.0,
+                                        503,
+                                        176.0
+                                    ],
+                                    "spans": [],
+                                    "index": 22
+                                },
+                                {
+                                    "bbox": [
+                                        295,
+                                        176.0,
+                                        503,
+                                        189.0
+                                    ],
+                                    "spans": [],
+                                    "index": 23
+                                },
+                                {
+                                    "bbox": [
+                                        295,
+                                        189.0,
+                                        503,
+                                        202.0
+                                    ],
+                                    "spans": [],
+                                    "index": 24
+                                },
+                                {
+                                    "bbox": [
+                                        295,
+                                        202.0,
+                                        503,
+                                        215.0
+                                    ],
+                                    "spans": [],
+                                    "index": 25
+                                },
+                                {
+                                    "bbox": [
+                                        295,
+                                        215.0,
+                                        503,
+                                        228.0
+                                    ],
+                                    "spans": [],
+                                    "index": 26
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 20.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        292,
+                        235,
+                        505,
+                        265
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                291,
+                                235,
+                                505,
+                                246
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        291,
+                                        235,
+                                        505,
+                                        246
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(b) Failure rates on the SR isomorphism task, the smaller the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                291,
+                                245,
+                                505,
+                                256
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        291,
+                                        245,
+                                        505,
+                                        256
+                                    ],
+                                    "score": 1.0,
+                                    "content": "better (mean and std-error over 5 runs). In parantheses, for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                291,
+                                254,
+                                502,
+                                266
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        291,
+                                        254,
+                                        406,
+                                        266
+                                    ],
+                                    "score": 1.0,
+                                    "content": "each model, the maximum size",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        406,
+                                        255,
+                                        413,
+                                        264
+                                    ],
+                                    "score": 0.74,
+                                    "content": "k",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        413,
+                                        254,
+                                        502,
+                                        266
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of rings lifted to 2-cells.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        }
+                    ],
+                    "index": 28
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        167,
+                        272,
+                        442,
+                        284
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                167,
+                                271,
+                                444,
+                                285
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        167,
+                                        271,
+                                        444,
+                                        285
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Figure 7: Results on the RingTransfer and SR synthetic benchmarks.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        }
+                    ],
+                    "index": 30
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        300,
+                        505,
+                        356
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                300,
+                                506,
+                                313
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        300,
+                                        506,
+                                        313
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the higher-order cells and their associated substructures into the message passing procedure. For",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                311,
+                                504,
+                                323
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        311,
+                                        504,
+                                        323
+                                    ],
+                                    "score": 1.0,
+                                    "content": "instance, bond features can learn to encode their membership to a ring and also communicate directly",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                322,
+                                504,
+                                335
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        322,
+                                        504,
+                                        335
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with other bonds present in the ring. Consequently, the messages between atoms connected through",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                333,
+                                505,
+                                346
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        333,
+                                        505,
+                                        346
+                                    ],
+                                    "score": 1.0,
+                                    "content": "these bonds are modulated by the presence of the ring as well as by the presence of other nodes and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                342,
+                                200,
+                                357
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        342,
+                                        200,
+                                        357
+                                    ],
+                                    "score": 1.0,
+                                    "content": "bonds part of that ring.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        }
+                    ],
+                    "index": 33
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        375,
+                        505,
+                        452
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                374,
+                                505,
+                                388
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        374,
+                                        505,
+                                        388
+                                    ],
+                                    "score": 1.0,
+                                    "content": "CWNs as Generalised Convolutions Our message passing scheme can be seen a (non-linear)",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                385,
+                                505,
+                                398
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        386,
+                                        406,
+                                        398
+                                    ],
+                                    "score": 1.0,
+                                    "content": "generalisation of linear diffusion operators on cell complexes. Recent works",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        407,
+                                        385,
+                                        439,
+                                        397
+                                    ],
+                                    "score": 0.7,
+                                    "content": "\\pm \\boxed { 1 3 } \\boxed { 2 5 } \\parallel",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        439,
+                                        386,
+                                        505,
+                                        398
+                                    ],
+                                    "score": 1.0,
+                                    "content": "have introduced",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                397,
+                                505,
+                                410
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        397,
+                                        370,
+                                        410
+                                    ],
+                                    "score": 1.0,
+                                    "content": "convolutional operators on SCs by employing the Hodge Laplacian",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        371,
+                                        397,
+                                        388,
+                                        408
+                                    ],
+                                    "score": 0.76,
+                                    "content": "\\pmb { \\mathbb { \\left\\| 6 3 \\right\\| } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        388,
+                                        397,
+                                        505,
+                                        410
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", a generalisation of the graph",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                408,
+                                505,
+                                420
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        408,
+                                        331,
+                                        420
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Laplacian. By leveraging on the cellular Sheaf Laplacian",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        332,
+                                        408,
+                                        349,
+                                        419
+                                    ],
+                                    "score": 0.81,
+                                    "content": "\\lVert \\dot { 3 6 } \\rVert",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        349,
+                                        408,
+                                        505,
+                                        420
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", a similar construction can be extended",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                418,
+                                505,
+                                431
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        418,
+                                        505,
+                                        431
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to cell complexes to define cellular convolutional operators. In Appendix D we discuss this approach",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                429,
+                                505,
+                                442
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        429,
+                                        505,
+                                        442
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and show that our cellular message passing scheme subsumes it. This represents a promising avenue",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                441,
+                                457,
+                                453
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        441,
+                                        457,
+                                        453
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for studying CWNs from a spectral perspective, an endeavour we leave for future work.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        }
+                    ],
+                    "index": 39
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        471,
+                        505,
+                        593
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                471,
+                                507,
+                                485
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        471,
+                                        507,
+                                        485
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Computational Complexity When considering cells of a constant maximum dimension and bound-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                483,
+                                505,
+                                495
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        483,
+                                        505,
+                                        495
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ary size, the computational complexity of the message passing scheme is linear in the size of the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                494,
+                                505,
+                                506
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        494,
+                                        505,
+                                        506
+                                    ],
+                                    "score": 1.0,
+                                    "content": "input complex. For the molecular applications we are interested in, the average number of rings per",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                504,
+                                505,
+                                517
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        504,
+                                        505,
+                                        517
+                                    ],
+                                    "score": 1.0,
+                                    "content": "molecule is upper bounded by a small constant (e.g. three for MOLHIV), so the size of the complex",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                515,
+                                506,
+                                528
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        515,
+                                        506,
+                                        528
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is approximately the same as the size of the graph. Therefore, in this setting, the computational",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 47
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                526,
+                                505,
+                                539
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        526,
+                                        505,
+                                        539
+                                    ],
+                                    "score": 1.0,
+                                    "content": "complexity of the model is similar to that of message passing GNNs. Separately of this, the one-time",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 48
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                537,
+                                506,
+                                549
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        537,
+                                        460,
+                                        549
+                                    ],
+                                    "score": 1.0,
+                                    "content": "preprocessing step of computing the lifting of the graphs should also be considered. The",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        461,
+                                        538,
+                                        470,
+                                        547
+                                    ],
+                                    "score": 0.81,
+                                    "content": "C",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        470,
+                                        537,
+                                        506,
+                                        549
+                                    ],
+                                    "score": 1.0,
+                                    "content": "induced",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 49
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                547,
+                                506,
+                                561
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        547,
+                                        244,
+                                        561
+                                    ],
+                                    "score": 1.0,
+                                    "content": "cycles in a graph can be listed in",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        244,
+                                        548,
+                                        312,
+                                        561
+                                    ],
+                                    "score": 0.87,
+                                    "content": "\\mathcal { O } \\big ( ( | E | ^ { - } | V | \\bar { C } \\big )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        312,
+                                        547,
+                                        347,
+                                        561
+                                    ],
+                                    "score": 1.0,
+                                    "content": "polylog",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        347,
+                                        547,
+                                        365,
+                                        561
+                                    ],
+                                    "score": 0.7,
+                                    "content": "| V | )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        365,
+                                        547,
+                                        388,
+                                        561
+                                    ],
+                                    "score": 1.0,
+                                    "content": "time",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        389,
+                                        547,
+                                        406,
+                                        559
+                                    ],
+                                    "score": 0.77,
+                                    "content": "\\pmb { \\Vert 2 6 \\Vert }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        407,
+                                        547,
+                                        485,
+                                        561
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Again, given that",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        485,
+                                        548,
+                                        494,
+                                        558
+                                    ],
+                                    "score": 0.83,
+                                    "content": "C",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        495,
+                                        547,
+                                        506,
+                                        561
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 50
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                559,
+                                505,
+                                573
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        559,
+                                        505,
+                                        573
+                                    ],
+                                    "score": 1.0,
+                                    "content": "upper bounded by a small constant for the molecular datasets of interest in this work, the complexity",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 51
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                569,
+                                506,
+                                583
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        569,
+                                        506,
+                                        583
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of the lifting procedure is also almost linear in the size of the graph. A more detailed analysis backed",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 52
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                581,
+                                343,
+                                594
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        581,
+                                        343,
+                                        594
+                                    ],
+                                    "score": 1.0,
+                                    "content": "up by wall-clock time experiments is given in Appendix B.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 53
+                        }
+                    ],
+                    "index": 48
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        615,
+                        191,
+                        629
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                103,
+                                613,
+                                193,
+                                633
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        103,
+                                        613,
+                                        193,
+                                        633
+                                    ],
+                                    "score": 1.0,
+                                    "content": "5 Experiments",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 54
+                        }
+                    ],
+                    "index": 54
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        645,
+                        505,
+                        722
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                644,
+                                506,
+                                659
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        644,
+                                        506,
+                                        659
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In this section we validate the theoretical and empirical properties of our proposed message passing",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 55
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                655,
+                                506,
+                                668
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        655,
+                                        506,
+                                        668
+                                    ],
+                                    "score": 1.0,
+                                    "content": "scheme in controlled scenarios as well as in real-world graph classification problems, with a focus on",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 56
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                667,
+                                505,
+                                680
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        667,
+                                        505,
+                                        680
+                                    ],
+                                    "score": 1.0,
+                                    "content": "large scale molecular benchmarks. For simplicity, in all experiments we employ a model which stacks",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 57
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                676,
+                                506,
+                                691
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        676,
+                                        295,
+                                        691
+                                    ],
+                                    "score": 1.0,
+                                    "content": "CWN layers with local aggregators as in GIN",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        295,
+                                        678,
+                                        312,
+                                        689
+                                    ],
+                                    "score": 0.56,
+                                    "content": "\\pmb { \\Vert 7 4 \\Vert }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        313,
+                                        676,
+                                        506,
+                                        691
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". We name our architecture “Cell Isomorphism",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 58
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                689,
+                                505,
+                                701
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        689,
+                                        505,
+                                        701
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Network” (CIN). 0-cells are always endowed with the original node features; higher-dimensional cells",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 59
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                700,
+                                506,
+                                713
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        700,
+                                        354,
+                                        712
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are populated in a benchmark specific manner. See Appendix",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        355,
+                                        701,
+                                        365,
+                                        713
+                                    ],
+                                    "score": 0.8,
+                                    "content": "\\mathrm { ~ E ~ }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        366,
+                                        700,
+                                        506,
+                                        712
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for details on feature initialisation,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 60
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                711,
+                                506,
+                                723
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        711,
+                                        506,
+                                        723
+                                    ],
+                                    "score": 1.0,
+                                    "content": "message passing and readout operations, hyperparameters, implementation and benchmark statistics.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 61
+                        }
+                    ],
+                    "index": 58
+                }
+            ],
+            "page_idx": 6,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        303,
+                        741,
+                        309,
+                        750
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                302,
+                                741,
+                                309,
+                                753
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        302,
+                                        741,
+                                        309,
+                                        753
+                                    ],
+                                    "score": 1.0,
+                                    "content": "7",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "image",
+                    "bbox": [
+                        108,
+                        72,
+                        281,
+                        208
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                108,
+                                72,
+                                281,
+                                208
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        108,
+                                        72,
+                                        281,
+                                        208
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                108,
+                                                72,
+                                                281,
+                                                208
+                                            ],
+                                            "score": 0.962,
+                                            "type": "image",
+                                            "image_path": "e7e513d4a4a37f8f0a963dcd756c8d2c678cb9d6e65f18f363336e6494cbaf6d.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 4.5,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        108,
+                                        72,
+                                        281,
+                                        85.6
+                                    ],
+                                    "spans": [],
+                                    "index": 0
+                                },
+                                {
+                                    "bbox": [
+                                        108,
+                                        85.6,
+                                        281,
+                                        99.19999999999999
+                                    ],
+                                    "spans": [],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        108,
+                                        99.19999999999999,
+                                        281,
+                                        112.79999999999998
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                },
+                                {
+                                    "bbox": [
+                                        108,
+                                        112.79999999999998,
+                                        281,
+                                        126.39999999999998
+                                    ],
+                                    "spans": [],
+                                    "index": 3
+                                },
+                                {
+                                    "bbox": [
+                                        108,
+                                        126.39999999999998,
+                                        281,
+                                        139.99999999999997
+                                    ],
+                                    "spans": [],
+                                    "index": 4
+                                },
+                                {
+                                    "bbox": [
+                                        108,
+                                        139.99999999999997,
+                                        281,
+                                        153.59999999999997
+                                    ],
+                                    "spans": [],
+                                    "index": 5
+                                },
+                                {
+                                    "bbox": [
+                                        108,
+                                        153.59999999999997,
+                                        281,
+                                        167.19999999999996
+                                    ],
+                                    "spans": [],
+                                    "index": 6
+                                },
+                                {
+                                    "bbox": [
+                                        108,
+                                        167.19999999999996,
+                                        281,
+                                        180.79999999999995
+                                    ],
+                                    "spans": [],
+                                    "index": 7
+                                },
+                                {
+                                    "bbox": [
+                                        108,
+                                        180.79999999999995,
+                                        281,
+                                        194.39999999999995
+                                    ],
+                                    "spans": [],
+                                    "index": 8
+                                },
+                                {
+                                    "bbox": [
+                                        108,
+                                        194.39999999999995,
+                                        281,
+                                        207.99999999999994
+                                    ],
+                                    "spans": [],
+                                    "index": 9
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 4.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        214,
+                        283,
+                        265
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                214,
+                                285,
+                                226
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        214,
+                                        285,
+                                        226
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(a) RingTransfer Results. Accuracy is over 5 bal-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                225,
+                                284,
+                                236
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        225,
+                                        284,
+                                        236
+                                    ],
+                                    "score": 1.0,
+                                    "content": "anced classes. A score of 0.2 is equivalent to a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                234,
+                                285,
+                                245
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        234,
+                                        285,
+                                        245
+                                    ],
+                                    "score": 1.0,
+                                    "content": "random guess. Error bars show the min and max.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                244,
+                                283,
+                                256
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        244,
+                                        283,
+                                        256
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Our model obtains high-scores in average even",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                254,
+                                273,
+                                266
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        254,
+                                        273,
+                                        266
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for large rings despite using only three layers.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        }
+                    ],
+                    "index": 12,
+                    "bbox_fs": [
+                        105,
+                        214,
+                        285,
+                        266
+                    ]
+                },
+                {
+                    "type": "image",
+                    "bbox": [
+                        295,
+                        72,
+                        503,
+                        228
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                295,
+                                72,
+                                503,
+                                228
+                            ],
+                            "group_id": 1,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        295,
+                                        72,
+                                        503,
+                                        228
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                295,
+                                                72,
+                                                503,
+                                                228
+                                            ],
+                                            "score": 0.958,
+                                            "type": "image",
+                                            "image_path": "61b45b8b73568ed72007633ef6dafbc69a26a971d31f6e09c325a41d0ecbc3e1.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 20.5,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        295,
+                                        72,
+                                        503,
+                                        85.0
+                                    ],
+                                    "spans": [],
+                                    "index": 15
+                                },
+                                {
+                                    "bbox": [
+                                        295,
+                                        85.0,
+                                        503,
+                                        98.0
+                                    ],
+                                    "spans": [],
+                                    "index": 16
+                                },
+                                {
+                                    "bbox": [
+                                        295,
+                                        98.0,
+                                        503,
+                                        111.0
+                                    ],
+                                    "spans": [],
+                                    "index": 17
+                                },
+                                {
+                                    "bbox": [
+                                        295,
+                                        111.0,
+                                        503,
+                                        124.0
+                                    ],
+                                    "spans": [],
+                                    "index": 18
+                                },
+                                {
+                                    "bbox": [
+                                        295,
+                                        124.0,
+                                        503,
+                                        137.0
+                                    ],
+                                    "spans": [],
+                                    "index": 19
+                                },
+                                {
+                                    "bbox": [
+                                        295,
+                                        137.0,
+                                        503,
+                                        150.0
+                                    ],
+                                    "spans": [],
+                                    "index": 20
+                                },
+                                {
+                                    "bbox": [
+                                        295,
+                                        150.0,
+                                        503,
+                                        163.0
+                                    ],
+                                    "spans": [],
+                                    "index": 21
+                                },
+                                {
+                                    "bbox": [
+                                        295,
+                                        163.0,
+                                        503,
+                                        176.0
+                                    ],
+                                    "spans": [],
+                                    "index": 22
+                                },
+                                {
+                                    "bbox": [
+                                        295,
+                                        176.0,
+                                        503,
+                                        189.0
+                                    ],
+                                    "spans": [],
+                                    "index": 23
+                                },
+                                {
+                                    "bbox": [
+                                        295,
+                                        189.0,
+                                        503,
+                                        202.0
+                                    ],
+                                    "spans": [],
+                                    "index": 24
+                                },
+                                {
+                                    "bbox": [
+                                        295,
+                                        202.0,
+                                        503,
+                                        215.0
+                                    ],
+                                    "spans": [],
+                                    "index": 25
+                                },
+                                {
+                                    "bbox": [
+                                        295,
+                                        215.0,
+                                        503,
+                                        228.0
+                                    ],
+                                    "spans": [],
+                                    "index": 26
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 20.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        292,
+                        235,
+                        505,
+                        265
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                291,
+                                235,
+                                505,
+                                246
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        291,
+                                        235,
+                                        505,
+                                        246
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(b) Failure rates on the SR isomorphism task, the smaller the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                291,
+                                245,
+                                505,
+                                256
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        291,
+                                        245,
+                                        505,
+                                        256
+                                    ],
+                                    "score": 1.0,
+                                    "content": "better (mean and std-error over 5 runs). In parantheses, for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                291,
+                                254,
+                                502,
+                                266
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        291,
+                                        254,
+                                        406,
+                                        266
+                                    ],
+                                    "score": 1.0,
+                                    "content": "each model, the maximum size",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        406,
+                                        255,
+                                        413,
+                                        264
+                                    ],
+                                    "score": 0.74,
+                                    "content": "k",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        413,
+                                        254,
+                                        502,
+                                        266
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of rings lifted to 2-cells.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        }
+                    ],
+                    "index": 28,
+                    "bbox_fs": [
+                        291,
+                        235,
+                        505,
+                        266
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        167,
+                        272,
+                        442,
+                        284
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                167,
+                                271,
+                                444,
+                                285
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        167,
+                                        271,
+                                        444,
+                                        285
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Figure 7: Results on the RingTransfer and SR synthetic benchmarks.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        }
+                    ],
+                    "index": 30,
+                    "bbox_fs": [
+                        167,
+                        271,
+                        444,
+                        285
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        300,
+                        505,
+                        356
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                300,
+                                506,
+                                313
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        300,
+                                        506,
+                                        313
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the higher-order cells and their associated substructures into the message passing procedure. For",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                311,
+                                504,
+                                323
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        311,
+                                        504,
+                                        323
+                                    ],
+                                    "score": 1.0,
+                                    "content": "instance, bond features can learn to encode their membership to a ring and also communicate directly",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                322,
+                                504,
+                                335
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        322,
+                                        504,
+                                        335
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with other bonds present in the ring. Consequently, the messages between atoms connected through",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                333,
+                                505,
+                                346
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        333,
+                                        505,
+                                        346
+                                    ],
+                                    "score": 1.0,
+                                    "content": "these bonds are modulated by the presence of the ring as well as by the presence of other nodes and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                342,
+                                200,
+                                357
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        342,
+                                        200,
+                                        357
+                                    ],
+                                    "score": 1.0,
+                                    "content": "bonds part of that ring.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        }
+                    ],
+                    "index": 33,
+                    "bbox_fs": [
+                        105,
+                        300,
+                        506,
+                        357
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        375,
+                        505,
+                        452
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                374,
+                                505,
+                                388
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        374,
+                                        505,
+                                        388
+                                    ],
+                                    "score": 1.0,
+                                    "content": "CWNs as Generalised Convolutions Our message passing scheme can be seen a (non-linear)",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                385,
+                                505,
+                                398
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        386,
+                                        406,
+                                        398
+                                    ],
+                                    "score": 1.0,
+                                    "content": "generalisation of linear diffusion operators on cell complexes. Recent works",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        407,
+                                        385,
+                                        439,
+                                        397
+                                    ],
+                                    "score": 0.7,
+                                    "content": "\\pm \\boxed { 1 3 } \\boxed { 2 5 } \\parallel",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        439,
+                                        386,
+                                        505,
+                                        398
+                                    ],
+                                    "score": 1.0,
+                                    "content": "have introduced",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                397,
+                                505,
+                                410
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        397,
+                                        370,
+                                        410
+                                    ],
+                                    "score": 1.0,
+                                    "content": "convolutional operators on SCs by employing the Hodge Laplacian",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        371,
+                                        397,
+                                        388,
+                                        408
+                                    ],
+                                    "score": 0.76,
+                                    "content": "\\pmb { \\mathbb { \\left\\| 6 3 \\right\\| } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        388,
+                                        397,
+                                        505,
+                                        410
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", a generalisation of the graph",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                408,
+                                505,
+                                420
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        408,
+                                        331,
+                                        420
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Laplacian. By leveraging on the cellular Sheaf Laplacian",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        332,
+                                        408,
+                                        349,
+                                        419
+                                    ],
+                                    "score": 0.81,
+                                    "content": "\\lVert \\dot { 3 6 } \\rVert",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        349,
+                                        408,
+                                        505,
+                                        420
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", a similar construction can be extended",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                418,
+                                505,
+                                431
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        418,
+                                        505,
+                                        431
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to cell complexes to define cellular convolutional operators. In Appendix D we discuss this approach",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                429,
+                                505,
+                                442
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        429,
+                                        505,
+                                        442
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and show that our cellular message passing scheme subsumes it. This represents a promising avenue",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                441,
+                                457,
+                                453
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        441,
+                                        457,
+                                        453
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for studying CWNs from a spectral perspective, an endeavour we leave for future work.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        }
+                    ],
+                    "index": 39,
+                    "bbox_fs": [
+                        105,
+                        374,
+                        505,
+                        453
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        471,
+                        505,
+                        593
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                471,
+                                507,
+                                485
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        471,
+                                        507,
+                                        485
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Computational Complexity When considering cells of a constant maximum dimension and bound-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                483,
+                                505,
+                                495
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        483,
+                                        505,
+                                        495
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ary size, the computational complexity of the message passing scheme is linear in the size of the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                494,
+                                505,
+                                506
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        494,
+                                        505,
+                                        506
+                                    ],
+                                    "score": 1.0,
+                                    "content": "input complex. For the molecular applications we are interested in, the average number of rings per",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                504,
+                                505,
+                                517
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        504,
+                                        505,
+                                        517
+                                    ],
+                                    "score": 1.0,
+                                    "content": "molecule is upper bounded by a small constant (e.g. three for MOLHIV), so the size of the complex",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                515,
+                                506,
+                                528
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        515,
+                                        506,
+                                        528
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is approximately the same as the size of the graph. Therefore, in this setting, the computational",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 47
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                526,
+                                505,
+                                539
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        526,
+                                        505,
+                                        539
+                                    ],
+                                    "score": 1.0,
+                                    "content": "complexity of the model is similar to that of message passing GNNs. Separately of this, the one-time",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 48
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                537,
+                                506,
+                                549
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        537,
+                                        460,
+                                        549
+                                    ],
+                                    "score": 1.0,
+                                    "content": "preprocessing step of computing the lifting of the graphs should also be considered. The",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        461,
+                                        538,
+                                        470,
+                                        547
+                                    ],
+                                    "score": 0.81,
+                                    "content": "C",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        470,
+                                        537,
+                                        506,
+                                        549
+                                    ],
+                                    "score": 1.0,
+                                    "content": "induced",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 49
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                547,
+                                506,
+                                561
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        547,
+                                        244,
+                                        561
+                                    ],
+                                    "score": 1.0,
+                                    "content": "cycles in a graph can be listed in",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        244,
+                                        548,
+                                        312,
+                                        561
+                                    ],
+                                    "score": 0.87,
+                                    "content": "\\mathcal { O } \\big ( ( | E | ^ { - } | V | \\bar { C } \\big )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        312,
+                                        547,
+                                        347,
+                                        561
+                                    ],
+                                    "score": 1.0,
+                                    "content": "polylog",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        347,
+                                        547,
+                                        365,
+                                        561
+                                    ],
+                                    "score": 0.7,
+                                    "content": "| V | )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        365,
+                                        547,
+                                        388,
+                                        561
+                                    ],
+                                    "score": 1.0,
+                                    "content": "time",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        389,
+                                        547,
+                                        406,
+                                        559
+                                    ],
+                                    "score": 0.77,
+                                    "content": "\\pmb { \\Vert 2 6 \\Vert }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        407,
+                                        547,
+                                        485,
+                                        561
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Again, given that",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        485,
+                                        548,
+                                        494,
+                                        558
+                                    ],
+                                    "score": 0.83,
+                                    "content": "C",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        495,
+                                        547,
+                                        506,
+                                        561
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 50
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                559,
+                                505,
+                                573
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        559,
+                                        505,
+                                        573
+                                    ],
+                                    "score": 1.0,
+                                    "content": "upper bounded by a small constant for the molecular datasets of interest in this work, the complexity",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 51
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                569,
+                                506,
+                                583
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        569,
+                                        506,
+                                        583
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of the lifting procedure is also almost linear in the size of the graph. A more detailed analysis backed",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 52
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                581,
+                                343,
+                                594
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        581,
+                                        343,
+                                        594
+                                    ],
+                                    "score": 1.0,
+                                    "content": "up by wall-clock time experiments is given in Appendix B.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 53
+                        }
+                    ],
+                    "index": 48,
+                    "bbox_fs": [
+                        104,
+                        471,
+                        507,
+                        594
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        615,
+                        191,
+                        629
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                103,
+                                613,
+                                193,
+                                633
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        103,
+                                        613,
+                                        193,
+                                        633
+                                    ],
+                                    "score": 1.0,
+                                    "content": "5 Experiments",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 54
+                        }
+                    ],
+                    "index": 54
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        645,
+                        505,
+                        722
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                644,
+                                506,
+                                659
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        644,
+                                        506,
+                                        659
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In this section we validate the theoretical and empirical properties of our proposed message passing",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 55
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                655,
+                                506,
+                                668
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        655,
+                                        506,
+                                        668
+                                    ],
+                                    "score": 1.0,
+                                    "content": "scheme in controlled scenarios as well as in real-world graph classification problems, with a focus on",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 56
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                667,
+                                505,
+                                680
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        667,
+                                        505,
+                                        680
+                                    ],
+                                    "score": 1.0,
+                                    "content": "large scale molecular benchmarks. For simplicity, in all experiments we employ a model which stacks",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 57
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                676,
+                                506,
+                                691
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        676,
+                                        295,
+                                        691
+                                    ],
+                                    "score": 1.0,
+                                    "content": "CWN layers with local aggregators as in GIN",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        295,
+                                        678,
+                                        312,
+                                        689
+                                    ],
+                                    "score": 0.56,
+                                    "content": "\\pmb { \\Vert 7 4 \\Vert }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        313,
+                                        676,
+                                        506,
+                                        691
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". We name our architecture “Cell Isomorphism",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 58
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                689,
+                                505,
+                                701
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        689,
+                                        505,
+                                        701
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Network” (CIN). 0-cells are always endowed with the original node features; higher-dimensional cells",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 59
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                700,
+                                506,
+                                713
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        700,
+                                        354,
+                                        712
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are populated in a benchmark specific manner. See Appendix",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        355,
+                                        701,
+                                        365,
+                                        713
+                                    ],
+                                    "score": 0.8,
+                                    "content": "\\mathrm { ~ E ~ }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        366,
+                                        700,
+                                        506,
+                                        712
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for details on feature initialisation,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 60
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                711,
+                                506,
+                                723
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        711,
+                                        506,
+                                        723
+                                    ],
+                                    "score": 1.0,
+                                    "content": "message passing and readout operations, hyperparameters, implementation and benchmark statistics.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 61
+                        }
+                    ],
+                    "index": 58,
+                    "bbox_fs": [
+                        105,
+                        644,
+                        506,
+                        723
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        73,
+                        227,
+                        84
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                72,
+                                227,
+                                86
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        72,
+                                        227,
+                                        86
+                                    ],
+                                    "score": 1.0,
+                                    "content": "5.1 Synthetic Benchmarks",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        }
+                    ],
+                    "index": 0
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        93,
+                        505,
+                        137
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                93,
+                                506,
+                                105
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        93,
+                                        340,
+                                        105
+                                    ],
+                                    "score": 1.0,
+                                    "content": "CSL Circular Skip Link dataset was first introduced in",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        341,
+                                        93,
+                                        359,
+                                        104
+                                    ],
+                                    "score": 0.83,
+                                    "content": " { \\mathbb { B } } 5 7  { \\mathbb { I } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        359,
+                                        93,
+                                        506,
+                                        105
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and has been recently adopted as a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                102,
+                                506,
+                                117
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        102,
+                                        322,
+                                        117
+                                    ],
+                                    "score": 1.0,
+                                    "content": "reference benchmark to test the expressivity of GNNs",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        323,
+                                        104,
+                                        340,
+                                        115
+                                    ],
+                                    "score": 0.6,
+                                    "content": "\\mathbb { \\left[ \\left[ 2 4 \\right] \\right] }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        340,
+                                        102,
+                                        506,
+                                        117
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". It consists of 150 4-regular graphs from",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                114,
+                                506,
+                                128
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        114,
+                                        506,
+                                        128
+                                    ],
+                                    "score": 1.0,
+                                    "content": "10 different isomorphism classes, which we need to predict. Unsolvable by the WL test and message",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                125,
+                                434,
+                                139
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        125,
+                                        434,
+                                        139
+                                    ],
+                                    "score": 1.0,
+                                    "content": "passing approaches [14, 57], we use it to validate the expressive power of CWNs.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        }
+                    ],
+                    "index": 2.5
+                },
+                {
+                    "type": "table",
+                    "bbox": [
+                        108,
+                        160,
+                        277,
+                        217
+                    ],
+                    "blocks": [
+                        {
+                            "type": "table_caption",
+                            "bbox": [
+                                110,
+                                143,
+                                273,
+                                154
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        108,
+                                        141,
+                                        274,
+                                        156
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                108,
+                                                141,
+                                                274,
+                                                156
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Table 1: Classification accuracy on CSL.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 5
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "type": "table_body",
+                            "bbox": [
+                                108,
+                                160,
+                                277,
+                                217
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        108,
+                                        160,
+                                        277,
+                                        217
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                108,
+                                                160,
+                                                277,
+                                                217
+                                            ],
+                                            "score": 0.978,
+                                            "html": "<table><tr><td>Method</td><td>Mean</td><td>Min</td><td>Max</td></tr><tr><td>MP-GNNs</td><td>10.000±0.000</td><td>10.000</td><td>10.000</td></tr><tr><td>RingGNN</td><td>10.000±0.000</td><td>10.000</td><td>10.000</td></tr><tr><td>3WLGNN</td><td>97.800±10.916</td><td>30.000</td><td>100.000</td></tr><tr><td>CIN (Ours)</td><td>100.000±0.000</td><td>100.000</td><td>100.000</td></tr></table>",
+                                            "type": "table",
+                                            "image_path": "1dbe2e32b816a9c594dadd34b5912275a1943a122f54f248f6bb3cd4dbbd0f5f.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 7.5,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        108,
+                                        160,
+                                        277,
+                                        174.25
+                                    ],
+                                    "spans": [],
+                                    "index": 6
+                                },
+                                {
+                                    "bbox": [
+                                        108,
+                                        174.25,
+                                        277,
+                                        188.5
+                                    ],
+                                    "spans": [],
+                                    "index": 7
+                                },
+                                {
+                                    "bbox": [
+                                        108,
+                                        188.5,
+                                        277,
+                                        202.75
+                                    ],
+                                    "spans": [],
+                                    "index": 8
+                                },
+                                {
+                                    "bbox": [
+                                        108,
+                                        202.75,
+                                        277,
+                                        217.0
+                                    ],
+                                    "spans": [],
+                                    "index": 9
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 6.25
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        285,
+                        142,
+                        505,
+                        218
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                285,
+                                142,
+                                506,
+                                154
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        285,
+                                        142,
+                                        506,
+                                        154
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We follow the same evaluation setting as Dwivedi et al.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                285,
+                                153,
+                                506,
+                                165
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        285,
+                                        153,
+                                        506,
+                                        165
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[24]: 5-fold cross validation procedure and 20 different",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                284,
+                                165,
+                                505,
+                                175
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        284,
+                                        165,
+                                        505,
+                                        175
+                                    ],
+                                    "score": 1.0,
+                                    "content": "random weight initialisations. For our model, we set",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                284,
+                                175,
+                                505,
+                                187
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        284,
+                                        175,
+                                        377,
+                                        187
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the maximum ring size",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        378,
+                                        175,
+                                        402,
+                                        185
+                                    ],
+                                    "score": 0.89,
+                                    "content": "k = 8",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        403,
+                                        175,
+                                        505,
+                                        187
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". In Table 1 we follow the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                284,
+                                186,
+                                506,
+                                198
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        284,
+                                        186,
+                                        506,
+                                        198
+                                    ],
+                                    "score": 1.0,
+                                    "content": "common practice on this dataset and report the mean,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                284,
+                                197,
+                                505,
+                                209
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        284,
+                                        197,
+                                        505,
+                                        209
+                                    ],
+                                    "score": 1.0,
+                                    "content": "minimum and maximum test accuracy obtained by CIN",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                284,
+                                208,
+                                505,
+                                219
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        284,
+                                        208,
+                                        505,
+                                        219
+                                    ],
+                                    "score": 1.0,
+                                    "content": "over the 100 runs, along with the results by the baselines",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        }
+                    ],
+                    "index": 13
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        219,
+                        505,
+                        263
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                217,
+                                504,
+                                232
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        217,
+                                        484,
+                                        232
+                                    ],
+                                    "score": 1.0,
+                                    "content": "presented in Dwivedi et al. [24]. MP-GNNs, that is classic message passing GNNs (GAT",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        485,
+                                        218,
+                                        502,
+                                        230
+                                    ],
+                                    "score": 0.31,
+                                    "content": "\\mathbb { \\lVert \\rVert }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        503,
+                                        217,
+                                        504,
+                                        232
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                229,
+                                506,
+                                242
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        229,
+                                        178,
+                                        242
+                                    ],
+                                    "score": 1.0,
+                                    "content": "MoNet [54], GIN",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        178,
+                                        229,
+                                        195,
+                                        241
+                                    ],
+                                    "score": 0.37,
+                                    "content": "\\pmb { \\Vert 7 4 \\Vert }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        196,
+                                        229,
+                                        506,
+                                        242
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", etc.), and RingGNN [14] perform as random guessers. In contrast, our model",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                241,
+                                505,
+                                252
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        241,
+                                        505,
+                                        252
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is able to identify the isomorphism class of each test graph in every run while featuring only a fraction",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                250,
+                                487,
+                                264
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        250,
+                                        451,
+                                        264
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of the computational complexity of 3WLGNN, the best performing reference baseline",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        452,
+                                        251,
+                                        484,
+                                        263
+                                    ],
+                                    "score": 0.6,
+                                    "content": "\\dot { \\mathbb { B } } 4 \\vert 5 3 \\vert ",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        485,
+                                        250,
+                                        487,
+                                        264
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        }
+                    ],
+                    "index": 18.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        276,
+                        505,
+                        419
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                275,
+                                505,
+                                289
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        275,
+                                        234,
+                                        289
+                                    ],
+                                    "score": 1.0,
+                                    "content": "SR Similarly to Bodnar et al.",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        234,
+                                        276,
+                                        247,
+                                        288
+                                    ],
+                                    "score": 0.37,
+                                    "content": "\\textcircled { 8 } \\textcircled { 1 8 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        247,
+                                        275,
+                                        265,
+                                        289
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        265,
+                                        275,
+                                        283,
+                                        288
+                                    ],
+                                    "score": 0.52,
+                                    "content": "\\mathbb { \\ m }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        283,
+                                        275,
+                                        505,
+                                        289
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", we consider Strongly Regular graphs within the same",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                286,
+                                505,
+                                299
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        286,
+                                        505,
+                                        299
+                                    ],
+                                    "score": 1.0,
+                                    "content": "family as hard examples of non-isomorphic graphs we seek to distinguish. Any pair of graphs within",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                296,
+                                506,
+                                312
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        296,
+                                        372,
+                                        312
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the same family cannot provably be distinguished by 3-WL test",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        372,
+                                        298,
+                                        400,
+                                        309
+                                    ],
+                                    "score": 0.86,
+                                    "content": "\\bar { \\bigstar } \\bar { \\bigstar } 0 \\bigstar",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        400,
+                                        296,
+                                        506,
+                                        312
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". We reproduce the same",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                308,
+                                504,
+                                321
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        308,
+                                        256,
+                                        321
+                                    ],
+                                    "score": 1.0,
+                                    "content": "experimental setting of Bodnar et al.",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        256,
+                                        308,
+                                        269,
+                                        320
+                                    ],
+                                    "score": 0.69,
+                                    "content": "\\textcircled { 8 } \\textcircled { 1 8 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        269,
+                                        308,
+                                        504,
+                                        321
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". In particular, we consider 9 distinct SR families3 and run",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                320,
+                                506,
+                                332
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        320,
+                                        389,
+                                        332
+                                    ],
+                                    "score": 1.0,
+                                    "content": "our model untrained on the cell complex lifting of each graph, with",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        390,
+                                        320,
+                                        436,
+                                        331
+                                    ],
+                                    "score": 0.9,
+                                    "content": "k = 4 , 5 , 6",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        437,
+                                        320,
+                                        506,
+                                        332
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". 0-cells (nodes)",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                330,
+                                506,
+                                343
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        330,
+                                        506,
+                                        343
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are initialised with a constant unitary signal, while 1- and 2-cells are initialised with the sum of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                342,
+                                506,
+                                354
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        342,
+                                        506,
+                                        354
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the contained 0-cells. We additionally run an MLP baseline with sum readout to appreciate the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                352,
+                                505,
+                                365
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        352,
+                                        490,
+                                        365
+                                    ],
+                                    "score": 1.0,
+                                    "content": "contribution of message passing. We report the percentage of non-distinguished pairs in Figure",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        491,
+                                        352,
+                                        505,
+                                        365
+                                    ],
+                                    "score": 0.55,
+                                    "content": "\\boxed { 7 6 }",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                363,
+                                505,
+                                376
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        363,
+                                        505,
+                                        376
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Contrary to 3-WL, both CIN and the MLP baseline are able to distinguish many pairs across all",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                374,
+                                505,
+                                387
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        374,
+                                        313,
+                                        387
+                                    ],
+                                    "score": 1.0,
+                                    "content": "families, with better performance attained for larger",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        313,
+                                        375,
+                                        320,
+                                        384
+                                    ],
+                                    "score": 0.74,
+                                    "content": "k",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        320,
+                                        374,
+                                        339,
+                                        387
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". For",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        340,
+                                        375,
+                                        365,
+                                        385
+                                    ],
+                                    "score": 0.9,
+                                    "content": "k = 6",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        365,
+                                        374,
+                                        505,
+                                        387
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", we observed CIN to disambiguate",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                385,
+                                505,
+                                398
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        385,
+                                        203,
+                                        398
+                                    ],
+                                    "score": 1.0,
+                                    "content": "all pairs in all families",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        204,
+                                        385,
+                                        226,
+                                        396
+                                    ],
+                                    "score": 0.86,
+                                    "content": "0 . 0 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        226,
+                                        385,
+                                        505,
+                                        398
+                                    ],
+                                    "score": 1.0,
+                                    "content": "failure rate). Despite the strong results achieved by the baseline, we",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                395,
+                                502,
+                                408
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        395,
+                                        496,
+                                        408
+                                    ],
+                                    "score": 1.0,
+                                    "content": "found CIN to always distinguish a larger number of non-isomorphic pairs for the same values of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        496,
+                                        397,
+                                        502,
+                                        406
+                                    ],
+                                    "score": 0.79,
+                                    "content": "k",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                406,
+                                345,
+                                421
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        406,
+                                        345,
+                                        421
+                                    ],
+                                    "score": 1.0,
+                                    "content": "this confirming the importance of cellular message passing.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        }
+                    ],
+                    "index": 27
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        432,
+                        505,
+                        577
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                431,
+                                506,
+                                444
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        431,
+                                        506,
+                                        444
+                                    ],
+                                    "score": 1.0,
+                                    "content": "RingTransfer In order to empirically validate the ability of CIN to capture long-range node",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                443,
+                                505,
+                                455
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        443,
+                                        505,
+                                        455
+                                    ],
+                                    "score": 1.0,
+                                    "content": "dependencies, we additionally design a third synthetic benchmark dubbed as ‘RingTransfer’. Graphs",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                453,
+                                506,
+                                467
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        453,
+                                        315,
+                                        467
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in this dataset are chordless cycles (rings) of size",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        315,
+                                        454,
+                                        322,
+                                        464
+                                    ],
+                                    "score": 0.76,
+                                    "content": "k",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        322,
+                                        453,
+                                        506,
+                                        467
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". In each graph we mark two special nodes",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                464,
+                                505,
+                                479
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        465,
+                                        304,
+                                        478
+                                    ],
+                                    "score": 1.0,
+                                    "content": "as target and source, always placed at distance",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        304,
+                                        464,
+                                        320,
+                                        479
+                                    ],
+                                    "score": 0.9,
+                                    "content": "\\lfloor \\frac { k } { 2 } \\rfloor",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        320,
+                                        465,
+                                        505,
+                                        478
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". The task is for target to output the one-hot",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                476,
+                                505,
+                                489
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        476,
+                                        505,
+                                        489
+                                    ],
+                                    "score": 1.0,
+                                    "content": "encoded label assigned to source. All other nodes in the ring are assigned a unitary constant feature",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                487,
+                                505,
+                                499
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        487,
+                                        505,
+                                        499
+                                    ],
+                                    "score": 1.0,
+                                    "content": "vector. A model has to learn to transfer the information contained in source to the opposite side",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                498,
+                                505,
+                                510
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        498,
+                                        302,
+                                        510
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of the ring, where target resides. We initialise",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        302,
+                                        499,
+                                        313,
+                                        509
+                                    ],
+                                    "score": 0.34,
+                                    "content": "1 \\mathrm { - }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        313,
+                                        498,
+                                        505,
+                                        510
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and 2-dimensional cells with a null signal. In",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                508,
+                                506,
+                                523
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        508,
+                                        133,
+                                        523
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Figure",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        133,
+                                        509,
+                                        147,
+                                        521
+                                    ],
+                                    "score": 0.42,
+                                    "content": "\\mathrm { 7 a }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        147,
+                                        508,
+                                        432,
+                                        523
+                                    ],
+                                    "score": 1.0,
+                                    "content": "we show the performance of a 3-layer CIN as a function of the ring size",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        433,
+                                        510,
+                                        439,
+                                        519
+                                    ],
+                                    "score": 0.76,
+                                    "content": "k",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        440,
+                                        508,
+                                        506,
+                                        523
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", along with that",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                520,
+                                506,
+                                534
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        520,
+                                        137,
+                                        534
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of GIN",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        138,
+                                        521,
+                                        155,
+                                        532
+                                    ],
+                                    "score": 0.28,
+                                    "content": "\\overline { { [ 7 4 ] } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        156,
+                                        520,
+                                        255,
+                                        534
+                                    ],
+                                    "score": 1.0,
+                                    "content": "baselines equipped with",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        256,
+                                        520,
+                                        272,
+                                        534
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\left\\lfloor { \\frac { k } { 2 } } \\right\\rfloor",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        272,
+                                        520,
+                                        506,
+                                        534
+                                    ],
+                                    "score": 1.0,
+                                    "content": "stacked layers. We observe that our model learns to solve",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                532,
+                                505,
+                                544
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        532,
+                                        332,
+                                        544
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the task with only 3 computational steps, independent of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        333,
+                                        533,
+                                        339,
+                                        542
+                                    ],
+                                    "score": 0.78,
+                                    "content": "k",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        339,
+                                        532,
+                                        505,
+                                        544
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". As for GIN, we observed degradation in",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                542,
+                                506,
+                                556
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        542,
+                                        188,
+                                        556
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the performance for",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        188,
+                                        543,
+                                        218,
+                                        554
+                                    ],
+                                    "score": 0.92,
+                                    "content": "k \\geq 2 4",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        218,
+                                        542,
+                                        506,
+                                        556
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", up to complete failure. We hypothesise this to be due to the difficulties",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                554,
+                                506,
+                                566
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        554,
+                                        226,
+                                        566
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of training such a deep GNN",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        227,
+                                        554,
+                                        249,
+                                        565
+                                    ],
+                                    "score": 0.82,
+                                    "content": "( \\geq 1 2",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        250,
+                                        554,
+                                        506,
+                                        566
+                                    ],
+                                    "score": 1.0,
+                                    "content": "layers). We further verify the (theoretically expected) failure of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                565,
+                                353,
+                                579
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        565,
+                                        305,
+                                        579
+                                    ],
+                                    "score": 1.0,
+                                    "content": "GIN (not included) when endowed with less than",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        306,
+                                        565,
+                                        321,
+                                        579
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\left\\lfloor { \\frac { k } { 2 } } \\right\\rfloor",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        322,
+                                        565,
+                                        353,
+                                        579
+                                    ],
+                                    "score": 1.0,
+                                    "content": "layers.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        }
+                    ],
+                    "index": 40
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        591,
+                        267,
+                        603
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                590,
+                                268,
+                                606
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        590,
+                                        268,
+                                        606
+                                    ],
+                                    "score": 1.0,
+                                    "content": "5.2 Real-World Graph Benchmarks",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 47
+                        }
+                    ],
+                    "index": 47
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        612,
+                        506,
+                        700
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                612,
+                                505,
+                                624
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        612,
+                                        338,
+                                        624
+                                    ],
+                                    "score": 1.0,
+                                    "content": "TUD We test our model on 8 TUDataset benchmarks",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        339,
+                                        612,
+                                        357,
+                                        623
+                                    ],
+                                    "score": 0.79,
+                                    "content": "\\begin{array} { r l } {  { \\bigl \\| \\boldsymbol { 5 } 6 \\bigr \\| } } \\end{array}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        357,
+                                        612,
+                                        505,
+                                        624
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with small and medium sizes from",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 48
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                622,
+                                506,
+                                636
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        622,
+                                        211,
+                                        636
+                                    ],
+                                    "score": 1.0,
+                                    "content": "biology (PROTEINS [9,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        211,
+                                        623,
+                                        228,
+                                        634
+                                    ],
+                                    "score": 0.56,
+                                    "content": "\\mathbb { 2 3 } \\mathbb { I } ,",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        229,
+                                        622,
+                                        506,
+                                        636
+                                    ],
+                                    "score": 1.0,
+                                    "content": "), chemistry (i.e. molecules – MUTAG [45, 61], PTC, NCI1 and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 49
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                633,
+                                506,
+                                647
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        633,
+                                        142,
+                                        647
+                                    ],
+                                    "score": 1.0,
+                                    "content": "NCI109",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        142,
+                                        634,
+                                        161,
+                                        645
+                                    ],
+                                    "score": 0.26,
+                                    "content": "\\mathbb { H } \\mathbb { Q } \\mathbb { I } ,",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        162,
+                                        633,
+                                        506,
+                                        647
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to social networks (IMDB-B, IMDB-M, RDT-B). We consider induced cycle of size",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 50
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                645,
+                                506,
+                                657
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        645,
+                                        130,
+                                        657
+                                    ],
+                                    "score": 1.0,
+                                    "content": "up to",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        130,
+                                        646,
+                                        157,
+                                        655
+                                    ],
+                                    "score": 0.9,
+                                    "content": "k = 6",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        157,
+                                        645,
+                                        506,
+                                        657
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for our graph lifting procedure. We initialise node (and 0-cell) features as described",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 51
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                655,
+                                506,
+                                668
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        655,
+                                        116,
+                                        668
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        117,
+                                        656,
+                                        131,
+                                        667
+                                    ],
+                                    "score": 0.34,
+                                    "content": "\\mathrm { X u }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        131,
+                                        655,
+                                        153,
+                                        668
+                                    ],
+                                    "score": 1.0,
+                                    "content": "et al.",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        153,
+                                        656,
+                                        171,
+                                        667
+                                    ],
+                                    "score": 0.73,
+                                    "content": "\\textcircled { 7 4 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        171,
+                                        655,
+                                        506,
+                                        668
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", and higher dimensional cells by averaging or summing the features of the included",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 52
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                666,
+                                506,
+                                680
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        666,
+                                        506,
+                                        680
+                                    ],
+                                    "score": 1.0,
+                                    "content": "0-cells. The training setting and evaluation procedure follow those in Xu et al. [74]. We report the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 53
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                677,
+                                506,
+                                691
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        677,
+                                        506,
+                                        691
+                                    ],
+                                    "score": 1.0,
+                                    "content": "results in Table 2. CIN compares more than favourably with the baselines, displaying strong empirical",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 54
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                688,
+                                507,
+                                702
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        688,
+                                        507,
+                                        702
+                                    ],
+                                    "score": 1.0,
+                                    "content": "performance on all benchmarks. The mean accuracy of CIN ranks top on four out of eight datasets.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 55
+                        }
+                    ],
+                    "index": 51.5
+                }
+            ],
+            "page_idx": 7,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        117,
+                        711,
+                        425,
+                        723
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                118,
+                                708,
+                                426,
+                                724
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        118,
+                                        708,
+                                        426,
+                                        724
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3 Data available at: http://users.cecs.anu.edu.au/~bdm/data/graphs.html.",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        302,
+                        742,
+                        308,
+                        750
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                300,
+                                740,
+                                309,
+                                752
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        300,
+                                        740,
+                                        309,
+                                        752
+                                    ],
+                                    "score": 1.0,
+                                    "content": "",
+                                    "type": "text",
+                                    "height": 12,
+                                    "width": 9
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        73,
+                        227,
+                        84
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                72,
+                                227,
+                                86
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        72,
+                                        227,
+                                        86
+                                    ],
+                                    "score": 1.0,
+                                    "content": "5.1 Synthetic Benchmarks",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        }
+                    ],
+                    "index": 0
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        93,
+                        505,
+                        137
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                93,
+                                506,
+                                105
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        93,
+                                        340,
+                                        105
+                                    ],
+                                    "score": 1.0,
+                                    "content": "CSL Circular Skip Link dataset was first introduced in",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        341,
+                                        93,
+                                        359,
+                                        104
+                                    ],
+                                    "score": 0.83,
+                                    "content": " { \\mathbb { B } } 5 7  { \\mathbb { I } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        359,
+                                        93,
+                                        506,
+                                        105
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and has been recently adopted as a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                102,
+                                506,
+                                117
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        102,
+                                        322,
+                                        117
+                                    ],
+                                    "score": 1.0,
+                                    "content": "reference benchmark to test the expressivity of GNNs",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        323,
+                                        104,
+                                        340,
+                                        115
+                                    ],
+                                    "score": 0.6,
+                                    "content": "\\mathbb { \\left[ \\left[ 2 4 \\right] \\right] }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        340,
+                                        102,
+                                        506,
+                                        117
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". It consists of 150 4-regular graphs from",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                114,
+                                506,
+                                128
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        114,
+                                        506,
+                                        128
+                                    ],
+                                    "score": 1.0,
+                                    "content": "10 different isomorphism classes, which we need to predict. Unsolvable by the WL test and message",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                125,
+                                434,
+                                139
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        125,
+                                        434,
+                                        139
+                                    ],
+                                    "score": 1.0,
+                                    "content": "passing approaches [14, 57], we use it to validate the expressive power of CWNs.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        }
+                    ],
+                    "index": 2.5,
+                    "bbox_fs": [
+                        104,
+                        93,
+                        506,
+                        139
+                    ]
+                },
+                {
+                    "type": "table",
+                    "bbox": [
+                        108,
+                        160,
+                        277,
+                        217
+                    ],
+                    "blocks": [
+                        {
+                            "type": "table_caption",
+                            "bbox": [
+                                110,
+                                143,
+                                273,
+                                154
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        108,
+                                        141,
+                                        274,
+                                        156
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                108,
+                                                141,
+                                                274,
+                                                156
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Table 1: Classification accuracy on CSL.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 5
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "type": "table_body",
+                            "bbox": [
+                                108,
+                                160,
+                                277,
+                                217
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        108,
+                                        160,
+                                        277,
+                                        217
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                108,
+                                                160,
+                                                277,
+                                                217
+                                            ],
+                                            "score": 0.978,
+                                            "html": "<table><tr><td>Method</td><td>Mean</td><td>Min</td><td>Max</td></tr><tr><td>MP-GNNs</td><td>10.000±0.000</td><td>10.000</td><td>10.000</td></tr><tr><td>RingGNN</td><td>10.000±0.000</td><td>10.000</td><td>10.000</td></tr><tr><td>3WLGNN</td><td>97.800±10.916</td><td>30.000</td><td>100.000</td></tr><tr><td>CIN (Ours)</td><td>100.000±0.000</td><td>100.000</td><td>100.000</td></tr></table>",
+                                            "type": "table",
+                                            "image_path": "1dbe2e32b816a9c594dadd34b5912275a1943a122f54f248f6bb3cd4dbbd0f5f.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 7.5,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        108,
+                                        160,
+                                        277,
+                                        174.25
+                                    ],
+                                    "spans": [],
+                                    "index": 6
+                                },
+                                {
+                                    "bbox": [
+                                        108,
+                                        174.25,
+                                        277,
+                                        188.5
+                                    ],
+                                    "spans": [],
+                                    "index": 7
+                                },
+                                {
+                                    "bbox": [
+                                        108,
+                                        188.5,
+                                        277,
+                                        202.75
+                                    ],
+                                    "spans": [],
+                                    "index": 8
+                                },
+                                {
+                                    "bbox": [
+                                        108,
+                                        202.75,
+                                        277,
+                                        217.0
+                                    ],
+                                    "spans": [],
+                                    "index": 9
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 6.25
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        285,
+                        142,
+                        505,
+                        218
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                285,
+                                142,
+                                506,
+                                154
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        285,
+                                        142,
+                                        506,
+                                        154
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We follow the same evaluation setting as Dwivedi et al.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                285,
+                                153,
+                                506,
+                                165
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        285,
+                                        153,
+                                        506,
+                                        165
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[24]: 5-fold cross validation procedure and 20 different",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                284,
+                                165,
+                                505,
+                                175
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        284,
+                                        165,
+                                        505,
+                                        175
+                                    ],
+                                    "score": 1.0,
+                                    "content": "random weight initialisations. For our model, we set",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                284,
+                                175,
+                                505,
+                                187
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        284,
+                                        175,
+                                        377,
+                                        187
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the maximum ring size",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        378,
+                                        175,
+                                        402,
+                                        185
+                                    ],
+                                    "score": 0.89,
+                                    "content": "k = 8",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        403,
+                                        175,
+                                        505,
+                                        187
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". In Table 1 we follow the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                284,
+                                186,
+                                506,
+                                198
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        284,
+                                        186,
+                                        506,
+                                        198
+                                    ],
+                                    "score": 1.0,
+                                    "content": "common practice on this dataset and report the mean,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                284,
+                                197,
+                                505,
+                                209
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        284,
+                                        197,
+                                        505,
+                                        209
+                                    ],
+                                    "score": 1.0,
+                                    "content": "minimum and maximum test accuracy obtained by CIN",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                284,
+                                208,
+                                505,
+                                219
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        284,
+                                        208,
+                                        505,
+                                        219
+                                    ],
+                                    "score": 1.0,
+                                    "content": "over the 100 runs, along with the results by the baselines",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                217,
+                                504,
+                                232
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        217,
+                                        484,
+                                        232
+                                    ],
+                                    "score": 1.0,
+                                    "content": "presented in Dwivedi et al. [24]. MP-GNNs, that is classic message passing GNNs (GAT",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        485,
+                                        218,
+                                        502,
+                                        230
+                                    ],
+                                    "score": 0.31,
+                                    "content": "\\mathbb { \\lVert \\rVert }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        503,
+                                        217,
+                                        504,
+                                        232
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                229,
+                                506,
+                                242
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        229,
+                                        178,
+                                        242
+                                    ],
+                                    "score": 1.0,
+                                    "content": "MoNet [54], GIN",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        178,
+                                        229,
+                                        195,
+                                        241
+                                    ],
+                                    "score": 0.37,
+                                    "content": "\\pmb { \\Vert 7 4 \\Vert }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        196,
+                                        229,
+                                        506,
+                                        242
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", etc.), and RingGNN [14] perform as random guessers. In contrast, our model",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                241,
+                                505,
+                                252
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        241,
+                                        505,
+                                        252
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is able to identify the isomorphism class of each test graph in every run while featuring only a fraction",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                250,
+                                487,
+                                264
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        250,
+                                        451,
+                                        264
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of the computational complexity of 3WLGNN, the best performing reference baseline",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        452,
+                                        251,
+                                        484,
+                                        263
+                                    ],
+                                    "score": 0.6,
+                                    "content": "\\dot { \\mathbb { B } } 4 \\vert 5 3 \\vert ",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        485,
+                                        250,
+                                        487,
+                                        264
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        }
+                    ],
+                    "index": 13,
+                    "bbox_fs": [
+                        284,
+                        142,
+                        506,
+                        219
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        219,
+                        505,
+                        263
+                    ],
+                    "lines": [],
+                    "index": 18.5,
+                    "bbox_fs": [
+                        104,
+                        217,
+                        506,
+                        264
+                    ],
+                    "lines_deleted": true
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        276,
+                        505,
+                        419
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                275,
+                                505,
+                                289
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        275,
+                                        234,
+                                        289
+                                    ],
+                                    "score": 1.0,
+                                    "content": "SR Similarly to Bodnar et al.",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        234,
+                                        276,
+                                        247,
+                                        288
+                                    ],
+                                    "score": 0.37,
+                                    "content": "\\textcircled { 8 } \\textcircled { 1 8 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        247,
+                                        275,
+                                        265,
+                                        289
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        265,
+                                        275,
+                                        283,
+                                        288
+                                    ],
+                                    "score": 0.52,
+                                    "content": "\\mathbb { \\ m }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        283,
+                                        275,
+                                        505,
+                                        289
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", we consider Strongly Regular graphs within the same",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                286,
+                                505,
+                                299
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        286,
+                                        505,
+                                        299
+                                    ],
+                                    "score": 1.0,
+                                    "content": "family as hard examples of non-isomorphic graphs we seek to distinguish. Any pair of graphs within",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                296,
+                                506,
+                                312
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        296,
+                                        372,
+                                        312
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the same family cannot provably be distinguished by 3-WL test",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        372,
+                                        298,
+                                        400,
+                                        309
+                                    ],
+                                    "score": 0.86,
+                                    "content": "\\bar { \\bigstar } \\bar { \\bigstar } 0 \\bigstar",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        400,
+                                        296,
+                                        506,
+                                        312
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". We reproduce the same",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                308,
+                                504,
+                                321
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        308,
+                                        256,
+                                        321
+                                    ],
+                                    "score": 1.0,
+                                    "content": "experimental setting of Bodnar et al.",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        256,
+                                        308,
+                                        269,
+                                        320
+                                    ],
+                                    "score": 0.69,
+                                    "content": "\\textcircled { 8 } \\textcircled { 1 8 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        269,
+                                        308,
+                                        504,
+                                        321
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". In particular, we consider 9 distinct SR families3 and run",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                320,
+                                506,
+                                332
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        320,
+                                        389,
+                                        332
+                                    ],
+                                    "score": 1.0,
+                                    "content": "our model untrained on the cell complex lifting of each graph, with",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        390,
+                                        320,
+                                        436,
+                                        331
+                                    ],
+                                    "score": 0.9,
+                                    "content": "k = 4 , 5 , 6",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        437,
+                                        320,
+                                        506,
+                                        332
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". 0-cells (nodes)",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                330,
+                                506,
+                                343
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        330,
+                                        506,
+                                        343
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are initialised with a constant unitary signal, while 1- and 2-cells are initialised with the sum of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                342,
+                                506,
+                                354
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        342,
+                                        506,
+                                        354
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the contained 0-cells. We additionally run an MLP baseline with sum readout to appreciate the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                352,
+                                505,
+                                365
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        352,
+                                        490,
+                                        365
+                                    ],
+                                    "score": 1.0,
+                                    "content": "contribution of message passing. We report the percentage of non-distinguished pairs in Figure",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        491,
+                                        352,
+                                        505,
+                                        365
+                                    ],
+                                    "score": 0.55,
+                                    "content": "\\boxed { 7 6 }",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                363,
+                                505,
+                                376
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        363,
+                                        505,
+                                        376
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Contrary to 3-WL, both CIN and the MLP baseline are able to distinguish many pairs across all",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                374,
+                                505,
+                                387
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        374,
+                                        313,
+                                        387
+                                    ],
+                                    "score": 1.0,
+                                    "content": "families, with better performance attained for larger",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        313,
+                                        375,
+                                        320,
+                                        384
+                                    ],
+                                    "score": 0.74,
+                                    "content": "k",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        320,
+                                        374,
+                                        339,
+                                        387
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". For",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        340,
+                                        375,
+                                        365,
+                                        385
+                                    ],
+                                    "score": 0.9,
+                                    "content": "k = 6",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        365,
+                                        374,
+                                        505,
+                                        387
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", we observed CIN to disambiguate",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                385,
+                                505,
+                                398
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        385,
+                                        203,
+                                        398
+                                    ],
+                                    "score": 1.0,
+                                    "content": "all pairs in all families",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        204,
+                                        385,
+                                        226,
+                                        396
+                                    ],
+                                    "score": 0.86,
+                                    "content": "0 . 0 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        226,
+                                        385,
+                                        505,
+                                        398
+                                    ],
+                                    "score": 1.0,
+                                    "content": "failure rate). Despite the strong results achieved by the baseline, we",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                395,
+                                502,
+                                408
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        395,
+                                        496,
+                                        408
+                                    ],
+                                    "score": 1.0,
+                                    "content": "found CIN to always distinguish a larger number of non-isomorphic pairs for the same values of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        496,
+                                        397,
+                                        502,
+                                        406
+                                    ],
+                                    "score": 0.79,
+                                    "content": "k",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                406,
+                                345,
+                                421
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        406,
+                                        345,
+                                        421
+                                    ],
+                                    "score": 1.0,
+                                    "content": "this confirming the importance of cellular message passing.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        }
+                    ],
+                    "index": 27,
+                    "bbox_fs": [
+                        105,
+                        275,
+                        506,
+                        421
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        432,
+                        505,
+                        577
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                431,
+                                506,
+                                444
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        431,
+                                        506,
+                                        444
+                                    ],
+                                    "score": 1.0,
+                                    "content": "RingTransfer In order to empirically validate the ability of CIN to capture long-range node",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                443,
+                                505,
+                                455
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        443,
+                                        505,
+                                        455
+                                    ],
+                                    "score": 1.0,
+                                    "content": "dependencies, we additionally design a third synthetic benchmark dubbed as ‘RingTransfer’. Graphs",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                453,
+                                506,
+                                467
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        453,
+                                        315,
+                                        467
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in this dataset are chordless cycles (rings) of size",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        315,
+                                        454,
+                                        322,
+                                        464
+                                    ],
+                                    "score": 0.76,
+                                    "content": "k",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        322,
+                                        453,
+                                        506,
+                                        467
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". In each graph we mark two special nodes",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                464,
+                                505,
+                                479
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        465,
+                                        304,
+                                        478
+                                    ],
+                                    "score": 1.0,
+                                    "content": "as target and source, always placed at distance",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        304,
+                                        464,
+                                        320,
+                                        479
+                                    ],
+                                    "score": 0.9,
+                                    "content": "\\lfloor \\frac { k } { 2 } \\rfloor",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        320,
+                                        465,
+                                        505,
+                                        478
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". The task is for target to output the one-hot",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                476,
+                                505,
+                                489
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        476,
+                                        505,
+                                        489
+                                    ],
+                                    "score": 1.0,
+                                    "content": "encoded label assigned to source. All other nodes in the ring are assigned a unitary constant feature",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                487,
+                                505,
+                                499
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        487,
+                                        505,
+                                        499
+                                    ],
+                                    "score": 1.0,
+                                    "content": "vector. A model has to learn to transfer the information contained in source to the opposite side",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                498,
+                                505,
+                                510
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        498,
+                                        302,
+                                        510
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of the ring, where target resides. We initialise",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        302,
+                                        499,
+                                        313,
+                                        509
+                                    ],
+                                    "score": 0.34,
+                                    "content": "1 \\mathrm { - }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        313,
+                                        498,
+                                        505,
+                                        510
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and 2-dimensional cells with a null signal. In",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                508,
+                                506,
+                                523
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        508,
+                                        133,
+                                        523
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Figure",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        133,
+                                        509,
+                                        147,
+                                        521
+                                    ],
+                                    "score": 0.42,
+                                    "content": "\\mathrm { 7 a }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        147,
+                                        508,
+                                        432,
+                                        523
+                                    ],
+                                    "score": 1.0,
+                                    "content": "we show the performance of a 3-layer CIN as a function of the ring size",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        433,
+                                        510,
+                                        439,
+                                        519
+                                    ],
+                                    "score": 0.76,
+                                    "content": "k",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        440,
+                                        508,
+                                        506,
+                                        523
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", along with that",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                520,
+                                506,
+                                534
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        520,
+                                        137,
+                                        534
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of GIN",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        138,
+                                        521,
+                                        155,
+                                        532
+                                    ],
+                                    "score": 0.28,
+                                    "content": "\\overline { { [ 7 4 ] } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        156,
+                                        520,
+                                        255,
+                                        534
+                                    ],
+                                    "score": 1.0,
+                                    "content": "baselines equipped with",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        256,
+                                        520,
+                                        272,
+                                        534
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\left\\lfloor { \\frac { k } { 2 } } \\right\\rfloor",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        272,
+                                        520,
+                                        506,
+                                        534
+                                    ],
+                                    "score": 1.0,
+                                    "content": "stacked layers. We observe that our model learns to solve",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                532,
+                                505,
+                                544
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        532,
+                                        332,
+                                        544
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the task with only 3 computational steps, independent of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        333,
+                                        533,
+                                        339,
+                                        542
+                                    ],
+                                    "score": 0.78,
+                                    "content": "k",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        339,
+                                        532,
+                                        505,
+                                        544
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". As for GIN, we observed degradation in",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                542,
+                                506,
+                                556
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        542,
+                                        188,
+                                        556
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the performance for",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        188,
+                                        543,
+                                        218,
+                                        554
+                                    ],
+                                    "score": 0.92,
+                                    "content": "k \\geq 2 4",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        218,
+                                        542,
+                                        506,
+                                        556
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", up to complete failure. We hypothesise this to be due to the difficulties",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                554,
+                                506,
+                                566
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        554,
+                                        226,
+                                        566
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of training such a deep GNN",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        227,
+                                        554,
+                                        249,
+                                        565
+                                    ],
+                                    "score": 0.82,
+                                    "content": "( \\geq 1 2",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        250,
+                                        554,
+                                        506,
+                                        566
+                                    ],
+                                    "score": 1.0,
+                                    "content": "layers). We further verify the (theoretically expected) failure of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                565,
+                                353,
+                                579
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        565,
+                                        305,
+                                        579
+                                    ],
+                                    "score": 1.0,
+                                    "content": "GIN (not included) when endowed with less than",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        306,
+                                        565,
+                                        321,
+                                        579
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\left\\lfloor { \\frac { k } { 2 } } \\right\\rfloor",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        322,
+                                        565,
+                                        353,
+                                        579
+                                    ],
+                                    "score": 1.0,
+                                    "content": "layers.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        }
+                    ],
+                    "index": 40,
+                    "bbox_fs": [
+                        105,
+                        431,
+                        506,
+                        579
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        591,
+                        267,
+                        603
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                590,
+                                268,
+                                606
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        590,
+                                        268,
+                                        606
+                                    ],
+                                    "score": 1.0,
+                                    "content": "5.2 Real-World Graph Benchmarks",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 47
+                        }
+                    ],
+                    "index": 47
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        612,
+                        506,
+                        700
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                612,
+                                505,
+                                624
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        612,
+                                        338,
+                                        624
+                                    ],
+                                    "score": 1.0,
+                                    "content": "TUD We test our model on 8 TUDataset benchmarks",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        339,
+                                        612,
+                                        357,
+                                        623
+                                    ],
+                                    "score": 0.79,
+                                    "content": "\\begin{array} { r l } {  { \\bigl \\| \\boldsymbol { 5 } 6 \\bigr \\| } } \\end{array}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        357,
+                                        612,
+                                        505,
+                                        624
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with small and medium sizes from",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 48
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                622,
+                                506,
+                                636
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        622,
+                                        211,
+                                        636
+                                    ],
+                                    "score": 1.0,
+                                    "content": "biology (PROTEINS [9,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        211,
+                                        623,
+                                        228,
+                                        634
+                                    ],
+                                    "score": 0.56,
+                                    "content": "\\mathbb { 2 3 } \\mathbb { I } ,",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        229,
+                                        622,
+                                        506,
+                                        636
+                                    ],
+                                    "score": 1.0,
+                                    "content": "), chemistry (i.e. molecules – MUTAG [45, 61], PTC, NCI1 and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 49
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                633,
+                                506,
+                                647
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        633,
+                                        142,
+                                        647
+                                    ],
+                                    "score": 1.0,
+                                    "content": "NCI109",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        142,
+                                        634,
+                                        161,
+                                        645
+                                    ],
+                                    "score": 0.26,
+                                    "content": "\\mathbb { H } \\mathbb { Q } \\mathbb { I } ,",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        162,
+                                        633,
+                                        506,
+                                        647
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to social networks (IMDB-B, IMDB-M, RDT-B). We consider induced cycle of size",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 50
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                645,
+                                506,
+                                657
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        645,
+                                        130,
+                                        657
+                                    ],
+                                    "score": 1.0,
+                                    "content": "up to",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        130,
+                                        646,
+                                        157,
+                                        655
+                                    ],
+                                    "score": 0.9,
+                                    "content": "k = 6",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        157,
+                                        645,
+                                        506,
+                                        657
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for our graph lifting procedure. We initialise node (and 0-cell) features as described",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 51
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                655,
+                                506,
+                                668
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        655,
+                                        116,
+                                        668
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        117,
+                                        656,
+                                        131,
+                                        667
+                                    ],
+                                    "score": 0.34,
+                                    "content": "\\mathrm { X u }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        131,
+                                        655,
+                                        153,
+                                        668
+                                    ],
+                                    "score": 1.0,
+                                    "content": "et al.",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        153,
+                                        656,
+                                        171,
+                                        667
+                                    ],
+                                    "score": 0.73,
+                                    "content": "\\textcircled { 7 4 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        171,
+                                        655,
+                                        506,
+                                        668
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", and higher dimensional cells by averaging or summing the features of the included",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 52
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                666,
+                                506,
+                                680
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        666,
+                                        506,
+                                        680
+                                    ],
+                                    "score": 1.0,
+                                    "content": "0-cells. The training setting and evaluation procedure follow those in Xu et al. [74]. We report the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 53
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                677,
+                                506,
+                                691
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        677,
+                                        506,
+                                        691
+                                    ],
+                                    "score": 1.0,
+                                    "content": "results in Table 2. CIN compares more than favourably with the baselines, displaying strong empirical",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 54
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                688,
+                                507,
+                                702
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        688,
+                                        507,
+                                        702
+                                    ],
+                                    "score": 1.0,
+                                    "content": "performance on all benchmarks. The mean accuracy of CIN ranks top on four out of eight datasets.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 55
+                        }
+                    ],
+                    "index": 51.5,
+                    "bbox_fs": [
+                        105,
+                        612,
+                        507,
+                        702
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "table",
+                    "bbox": [
+                        105,
+                        100,
+                        503,
+                        234
+                    ],
+                    "blocks": [
+                        {
+                            "type": "table_caption",
+                            "bbox": [
+                                106,
+                                77,
+                                504,
+                                99
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        77,
+                                        506,
+                                        90
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                77,
+                                                506,
+                                                90
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Table 2: TUDatasets. The first section of the table includes the accuracy of graph kernel methods,",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 0
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        88,
+                                        464,
+                                        100
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                88,
+                                                464,
+                                                100
+                                            ],
+                                            "score": 1.0,
+                                            "content": "while the second includes GNNs. The top three are highlighted by First, Second, Third.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 1
+                                }
+                            ],
+                            "index": 0.5
+                        },
+                        {
+                            "type": "table_body",
+                            "bbox": [
+                                105,
+                                100,
+                                503,
+                                234
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        100,
+                                        503,
+                                        234
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                100,
+                                                503,
+                                                234
+                                            ],
+                                            "score": 0.984,
+                                            "html": "<table><tr><td>Dataset</td><td>MUTAG</td><td>PTC</td><td>PROTEINS</td><td>NCI1</td><td>NCI109</td><td>IMDB-B</td><td>IMDB-M</td><td>RDT-B</td></tr><tr><td>RWK 四</td><td>79.2±2.1</td><td>55.9±0.3</td><td>59.6±0.1</td><td>&gt;3 days</td><td>N/A</td><td>N/A</td><td>N/A</td><td>N/A</td></tr><tr><td>GK (k = 3) I 64</td><td>81.4±1.7</td><td>55.7±0.5</td><td>71.4±0.31</td><td>62.5±0.3</td><td>62.4±0.3</td><td>N/A</td><td>N/A</td><td>N/A</td></tr><tr><td>PK 58</td><td>76.0±2.7</td><td>59.5±2.4</td><td>73.7±0.7</td><td>82.5±0.5</td><td>N/A</td><td>N/A</td><td>N/A</td><td>N/A</td></tr><tr><td>WL kernel [65]</td><td>90.4±5.7</td><td>59.9±4.3</td><td>75.0±3.1</td><td>86.0±1.8</td><td>N/A</td><td>73.8±3.9</td><td>50.9±3.8</td><td>81.0±3.1</td></tr><tr><td>DCNN</td><td>N/A</td><td>N/A</td><td>61.3±1.6</td><td>56.6±1.0</td><td>N/A</td><td>49.1±1.4</td><td>33.5±1.4</td><td>N/A</td></tr><tr><td>DGCNN[76]</td><td>85.8±1.8</td><td>58.6±2.5</td><td>75.5±0.9</td><td>74.4±0.5</td><td>N/A</td><td>70.0±0.9</td><td>47.8±0.9</td><td>N/A</td></tr><tr><td>IGN[ 圆</td><td>83.9±13.0</td><td>58.5±6.9</td><td>76.6±5.5</td><td>74.3±2.7</td><td>72.8±1.5</td><td>72.0±5.5</td><td>48.7±3.4</td><td>N/A</td></tr><tr><td>GIN 园</td><td>89.4±5.6</td><td>64.6±7.0</td><td>76.2±2.8</td><td>82.7±1.7</td><td>N/A</td><td>75.1±5.1</td><td>52.3±2.8</td><td>92.4±2.5</td></tr><tr><td>PPGNs [531</td><td>90.6±8.7</td><td>66.2±6.6</td><td>77.2±4.7</td><td>83.2±1.1</td><td>82.2±1.4</td><td>73.0±5.8</td><td>50.5±3.6</td><td>N/A</td></tr><tr><td>Natural GN 自</td><td>89.4±1.6</td><td>66.8±1.7</td><td>71.7±1.0</td><td>82.4±1.3</td><td>N/A</td><td>73.5±2.0</td><td>51.3±1.5</td><td>N/A</td></tr><tr><td>GSN SIN</td><td>92.2 ± 7.5 N/A</td><td>68.2 ± 7.2 N/A</td><td>76.6± 5.0 76.4 ± 3.3</td><td>83.5 ± 2.0 82.7 ± 2.1</td><td>N/A</td><td>77.8 ± 3.3</td><td>54.3 ± 3.3</td><td>N/A</td></tr><tr><td></td><td></td><td></td><td></td><td></td><td>N/A</td><td>75.6± 3.2</td><td>52.4 ± 2.9</td><td>92.2 ± 1.0</td></tr><tr><td>CIN (Ours)</td><td>92.7 ± 6.1</td><td>68.2± 5.6</td><td>77.0 ± 4.3</td><td>83.6 ± 1.4</td><td>84.0 ± 1.6</td><td>75.6± 3.7</td><td>52.7 ± 3.1</td><td>92.4 ± 2.1</td></tr></table>",
+                                            "type": "table",
+                                            "image_path": "4a1e49e9b880ffaed02a2f27a5a561dfa2a89ae6310f5da2ab3223cb7a13aab3.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 3,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        100,
+                                        503,
+                                        144.66666666666666
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        144.66666666666666,
+                                        503,
+                                        189.33333333333331
+                                    ],
+                                    "spans": [],
+                                    "index": 3
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        189.33333333333331,
+                                        503,
+                                        233.99999999999997
+                                    ],
+                                    "spans": [],
+                                    "index": 4
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 1.75
+                },
+                {
+                    "type": "table",
+                    "bbox": [
+                        146,
+                        273,
+                        464,
+                        402
+                    ],
+                    "blocks": [
+                        {
+                            "type": "table_caption",
+                            "bbox": [
+                                157,
+                                261,
+                                450,
+                                273
+                            ],
+                            "group_id": 1,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        157,
+                                        259,
+                                        451,
+                                        276
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                157,
+                                                259,
+                                                451,
+                                                276
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Table 3: ZINC (MAE), ZINC-FULL (MAE) and Mol-HIV (ROC-AUC).",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 5
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "type": "table_body",
+                            "bbox": [
+                                146,
+                                273,
+                                464,
+                                402
+                            ],
+                            "group_id": 1,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        146,
+                                        273,
+                                        464,
+                                        402
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                146,
+                                                273,
+                                                464,
+                                                402
+                                            ],
+                                            "score": 0.982,
+                                            "html": "<table><tr><td rowspan=\"2\">Method</td><td colspan=\"2\">ZINC↓</td><td rowspan=\"2\">ZINC-FULL↓ All methods</td><td rowspan=\"2\">MOLHIV↑ All methods</td></tr><tr><td>No Edge Feat.</td><td>With Edge Feat.</td></tr><tr><td>GCN 四</td><td>0.469±0.002</td><td>N/A</td><td>N/A</td><td>76.06±0.97</td></tr><tr><td>GAT 园</td><td>0.463±0.002</td><td>N/A</td><td>N/A</td><td>N/A</td></tr><tr><td>GatedGCN[ 自</td><td>0.422±0.006</td><td>0.363±0.009</td><td>N/A</td><td>N/A</td></tr><tr><td>GIN 四</td><td>0.408±0.008</td><td>0.252±0.014</td><td>0.088±0.002</td><td>77.07±1.49</td></tr><tr><td>PNA 四</td><td>0.320±0.032</td><td>0.188±0.004</td><td>N/A</td><td>79.05±1.32</td></tr><tr><td>DGN 回</td><td>0.219±0.010</td><td>0.168±0.003</td><td>N/A</td><td>79.70±0.97</td></tr><tr><td>HIMP 四</td><td>N/A</td><td>0.151±0.006</td><td>0.036±0.002</td><td>78.80±0.82</td></tr><tr><td>GSN 目</td><td>0.139±0.007</td><td>0.108±0.018</td><td>N/A</td><td>77.99±1.00</td></tr><tr><td>CIN-small (Ours)</td><td>0.139±0.008</td><td>0.094±0.004</td><td>0.044±0.003</td><td>80.55±1.04</td></tr><tr><td>CIN (Ours)</td><td>0.115±0.003</td><td>0.079±0.006</td><td>0.022±0.002</td><td>80.94±0.57</td></tr></table>",
+                                            "type": "table",
+                                            "image_path": "6f240739ac03f206556c984643be99722156cc3bfe46e31403ab0d94d939fec3.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 7,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        146,
+                                        273,
+                                        464,
+                                        316.0
+                                    ],
+                                    "spans": [],
+                                    "index": 6
+                                },
+                                {
+                                    "bbox": [
+                                        146,
+                                        316.0,
+                                        464,
+                                        359.0
+                                    ],
+                                    "spans": [],
+                                    "index": 7
+                                },
+                                {
+                                    "bbox": [
+                                        146,
+                                        359.0,
+                                        464,
+                                        402.0
+                                    ],
+                                    "spans": [],
+                                    "index": 8
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 6.0
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        422,
+                        504,
+                        444
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                422,
+                                505,
+                                435
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        422,
+                                        505,
+                                        435
+                                    ],
+                                    "score": 1.0,
+                                    "content": "On the remaining datasets, CIN achieves the second place. We observe that the best results are on",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                433,
+                                444,
+                                446
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        433,
+                                        444,
+                                        446
+                                    ],
+                                    "score": 1.0,
+                                    "content": "datasets from the biological and chemical domains, where rings play a relevant role.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        }
+                    ],
+                    "index": 9.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        468,
+                        505,
+                        589
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                468,
+                                505,
+                                481
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        468,
+                                        505,
+                                        481
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ZINC We study the effectiveness of cellular message passing on larger scale molecular benchmarks",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                478,
+                                504,
+                                492
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        478,
+                                        206,
+                                        492
+                                    ],
+                                    "score": 1.0,
+                                    "content": "from the ZINC database",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        206,
+                                        479,
+                                        223,
+                                        491
+                                    ],
+                                    "score": 0.72,
+                                    "content": "\\pmb { \\mathbb { E } } 8 \\|",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        223,
+                                        478,
+                                        504,
+                                        492
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". ZINC (12k graphs) and ZINC-FULL (250k graphs) [24, 33, 43, 75]",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                489,
+                                506,
+                                504
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        489,
+                                        506,
+                                        504
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are two graph regression task datasets for drug constrained solubility prediction. In these experiments,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                501,
+                                506,
+                                514
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        501,
+                                        220,
+                                        514
+                                    ],
+                                    "score": 1.0,
+                                    "content": "we consider rings up to size",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        220,
+                                        501,
+                                        250,
+                                        511
+                                    ],
+                                    "score": 0.89,
+                                    "content": "k = 1 8",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        251,
+                                        501,
+                                        506,
+                                        514
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". We follow the training and evaluation procedures in [24]. Our",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                512,
+                                506,
+                                524
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        512,
+                                        506,
+                                        524
+                                    ],
+                                    "score": 1.0,
+                                    "content": "experiments encompass different scenarios, examine the impact of ablating edge features and of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                522,
+                                506,
+                                535
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        523,
+                                        496,
+                                        535
+                                    ],
+                                    "score": 1.0,
+                                    "content": "constraining the parameter budget of the architecture to 100k. All results are illustrated in Table",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        496,
+                                        522,
+                                        506,
+                                        535
+                                    ],
+                                    "score": 0.44,
+                                    "content": "\\textcircled { 3 }",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                534,
+                                506,
+                                546
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        534,
+                                        506,
+                                        546
+                                    ],
+                                    "score": 1.0,
+                                    "content": "where we also include the results for ZINC-FULL obtained by the same exact architectures. Our",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                545,
+                                505,
+                                558
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        545,
+                                        505,
+                                        558
+                                    ],
+                                    "score": 1.0,
+                                    "content": "model exhibits particularly strong performance on these benchmarks: it attains state-of-the-art results",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                556,
+                                505,
+                                568
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        556,
+                                        505,
+                                        568
+                                    ],
+                                    "score": 1.0,
+                                    "content": "on both the two dataset variants, outperforming other models by a significant margin. CIN attains",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                567,
+                                505,
+                                578
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        567,
+                                        505,
+                                        578
+                                    ],
+                                    "score": 1.0,
+                                    "content": "strong results even when constrained by the parameter budget. It still achieves state-of-the-art",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                578,
+                                506,
+                                590
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        578,
+                                        506,
+                                        590
+                                    ],
+                                    "score": 1.0,
+                                    "content": "performance on ZINC and is on-par with the best unconstrained baseline under edge-feature ablation.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        }
+                    ],
+                    "index": 16
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        612,
+                        506,
+                        722
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                611,
+                                505,
+                                625
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        611,
+                                        505,
+                                        625
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Mol-HIV We additionally test our model on the molecular ogbg-molhiv dataset from the Open",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                623,
+                                505,
+                                636
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        623,
+                                        181,
+                                        636
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Graph Benchmark",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        181,
+                                        623,
+                                        199,
+                                        635
+                                    ],
+                                    "score": 0.69,
+                                    "content": "\\textcircled { | 4 0 | }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        199,
+                                        623,
+                                        505,
+                                        636
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(41k graphs). The task is to predict the capacity of compounds to inhibit HIV",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                634,
+                                505,
+                                647
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        634,
+                                        234,
+                                        647
+                                    ],
+                                    "score": 1.0,
+                                    "content": "replication. Rings of size up to",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        234,
+                                        635,
+                                        259,
+                                        645
+                                    ],
+                                    "score": 0.9,
+                                    "content": "k = 6",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        259,
+                                        634,
+                                        475,
+                                        647
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are considered as 2-cells. We take the architecture in",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        475,
+                                        634,
+                                        493,
+                                        646
+                                    ],
+                                    "score": 0.83,
+                                    "content": " { \\mathbb { \\left[ \\left[ 2 7 \\right] \\right] } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        493,
+                                        634,
+                                        505,
+                                        647
+                                    ],
+                                    "score": 1.0,
+                                    "content": "as",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                645,
+                                506,
+                                659
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        645,
+                                        506,
+                                        659
+                                    ],
+                                    "score": 1.0,
+                                    "content": "reference and replicate the same hyperparameter setting in our model, including the use of only 2",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                656,
+                                506,
+                                668
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        656,
+                                        506,
+                                        668
+                                    ],
+                                    "score": 1.0,
+                                    "content": "message passing layers. We report the mean of test ROC-AUC metrics at the epoch of best validation",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                667,
+                                506,
+                                679
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        667,
+                                        506,
+                                        679
+                                    ],
+                                    "score": 1.0,
+                                    "content": "performance for 10 random weight initialisations. Similarly to ZINC, we experiment with a “small”",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                677,
+                                506,
+                                691
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        678,
+                                        369,
+                                        691
+                                    ],
+                                    "score": 1.0,
+                                    "content": "model whose number of parameters is constrained in the order of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        369,
+                                        678,
+                                        390,
+                                        689
+                                    ],
+                                    "score": 0.58,
+                                    "content": "1 0 0 \\mathrm { k }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        390,
+                                        678,
+                                        417,
+                                        691
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Table",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        417,
+                                        677,
+                                        427,
+                                        691
+                                    ],
+                                    "score": 0.4,
+                                    "content": "3",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        427,
+                                        678,
+                                        506,
+                                        691
+                                    ],
+                                    "score": 1.0,
+                                    "content": "displays the results.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                688,
+                                506,
+                                702
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        688,
+                                        506,
+                                        702
+                                    ],
+                                    "score": 1.0,
+                                    "content": "CIN significantly outperforms other strong GNN baselines, even when constrained by the parameter",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                698,
+                                506,
+                                714
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        698,
+                                        217,
+                                        714
+                                    ],
+                                    "score": 1.0,
+                                    "content": "budget. Consistently with",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        217,
+                                        700,
+                                        235,
+                                        711
+                                    ],
+                                    "score": 0.86,
+                                    "content": "[ [ 2 7 ]",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        235,
+                                        698,
+                                        506,
+                                        714
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", we observe that only two layers are sufficient when performing",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                710,
+                                396,
+                                724
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        710,
+                                        396,
+                                        724
+                                    ],
+                                    "score": 1.0,
+                                    "content": "hierarchical message passing across meso-scale structures such as rings.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        }
+                    ],
+                    "index": 26.5
+                }
+            ],
+            "page_idx": 8,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        302,
+                        741,
+                        309,
+                        750
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                302,
+                                741,
+                                309,
+                                752
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        302,
+                                        741,
+                                        309,
+                                        752
+                                    ],
+                                    "score": 1.0,
+                                    "content": "9",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "table",
+                    "bbox": [
+                        105,
+                        100,
+                        503,
+                        234
+                    ],
+                    "blocks": [
+                        {
+                            "type": "table_caption",
+                            "bbox": [
+                                106,
+                                77,
+                                504,
+                                99
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        77,
+                                        506,
+                                        90
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                77,
+                                                506,
+                                                90
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Table 2: TUDatasets. The first section of the table includes the accuracy of graph kernel methods,",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 0
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        88,
+                                        464,
+                                        100
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                88,
+                                                464,
+                                                100
+                                            ],
+                                            "score": 1.0,
+                                            "content": "while the second includes GNNs. The top three are highlighted by First, Second, Third.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 1
+                                }
+                            ],
+                            "index": 0.5
+                        },
+                        {
+                            "type": "table_body",
+                            "bbox": [
+                                105,
+                                100,
+                                503,
+                                234
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        100,
+                                        503,
+                                        234
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                100,
+                                                503,
+                                                234
+                                            ],
+                                            "score": 0.984,
+                                            "html": "<table><tr><td>Dataset</td><td>MUTAG</td><td>PTC</td><td>PROTEINS</td><td>NCI1</td><td>NCI109</td><td>IMDB-B</td><td>IMDB-M</td><td>RDT-B</td></tr><tr><td>RWK 四</td><td>79.2±2.1</td><td>55.9±0.3</td><td>59.6±0.1</td><td>&gt;3 days</td><td>N/A</td><td>N/A</td><td>N/A</td><td>N/A</td></tr><tr><td>GK (k = 3) I 64</td><td>81.4±1.7</td><td>55.7±0.5</td><td>71.4±0.31</td><td>62.5±0.3</td><td>62.4±0.3</td><td>N/A</td><td>N/A</td><td>N/A</td></tr><tr><td>PK 58</td><td>76.0±2.7</td><td>59.5±2.4</td><td>73.7±0.7</td><td>82.5±0.5</td><td>N/A</td><td>N/A</td><td>N/A</td><td>N/A</td></tr><tr><td>WL kernel [65]</td><td>90.4±5.7</td><td>59.9±4.3</td><td>75.0±3.1</td><td>86.0±1.8</td><td>N/A</td><td>73.8±3.9</td><td>50.9±3.8</td><td>81.0±3.1</td></tr><tr><td>DCNN</td><td>N/A</td><td>N/A</td><td>61.3±1.6</td><td>56.6±1.0</td><td>N/A</td><td>49.1±1.4</td><td>33.5±1.4</td><td>N/A</td></tr><tr><td>DGCNN[76]</td><td>85.8±1.8</td><td>58.6±2.5</td><td>75.5±0.9</td><td>74.4±0.5</td><td>N/A</td><td>70.0±0.9</td><td>47.8±0.9</td><td>N/A</td></tr><tr><td>IGN[ 圆</td><td>83.9±13.0</td><td>58.5±6.9</td><td>76.6±5.5</td><td>74.3±2.7</td><td>72.8±1.5</td><td>72.0±5.5</td><td>48.7±3.4</td><td>N/A</td></tr><tr><td>GIN 园</td><td>89.4±5.6</td><td>64.6±7.0</td><td>76.2±2.8</td><td>82.7±1.7</td><td>N/A</td><td>75.1±5.1</td><td>52.3±2.8</td><td>92.4±2.5</td></tr><tr><td>PPGNs [531</td><td>90.6±8.7</td><td>66.2±6.6</td><td>77.2±4.7</td><td>83.2±1.1</td><td>82.2±1.4</td><td>73.0±5.8</td><td>50.5±3.6</td><td>N/A</td></tr><tr><td>Natural GN 自</td><td>89.4±1.6</td><td>66.8±1.7</td><td>71.7±1.0</td><td>82.4±1.3</td><td>N/A</td><td>73.5±2.0</td><td>51.3±1.5</td><td>N/A</td></tr><tr><td>GSN SIN</td><td>92.2 ± 7.5 N/A</td><td>68.2 ± 7.2 N/A</td><td>76.6± 5.0 76.4 ± 3.3</td><td>83.5 ± 2.0 82.7 ± 2.1</td><td>N/A</td><td>77.8 ± 3.3</td><td>54.3 ± 3.3</td><td>N/A</td></tr><tr><td></td><td></td><td></td><td></td><td></td><td>N/A</td><td>75.6± 3.2</td><td>52.4 ± 2.9</td><td>92.2 ± 1.0</td></tr><tr><td>CIN (Ours)</td><td>92.7 ± 6.1</td><td>68.2± 5.6</td><td>77.0 ± 4.3</td><td>83.6 ± 1.4</td><td>84.0 ± 1.6</td><td>75.6± 3.7</td><td>52.7 ± 3.1</td><td>92.4 ± 2.1</td></tr></table>",
+                                            "type": "table",
+                                            "image_path": "4a1e49e9b880ffaed02a2f27a5a561dfa2a89ae6310f5da2ab3223cb7a13aab3.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 3,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        100,
+                                        503,
+                                        144.66666666666666
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        144.66666666666666,
+                                        503,
+                                        189.33333333333331
+                                    ],
+                                    "spans": [],
+                                    "index": 3
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        189.33333333333331,
+                                        503,
+                                        233.99999999999997
+                                    ],
+                                    "spans": [],
+                                    "index": 4
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 1.75
+                },
+                {
+                    "type": "table",
+                    "bbox": [
+                        146,
+                        273,
+                        464,
+                        402
+                    ],
+                    "blocks": [
+                        {
+                            "type": "table_caption",
+                            "bbox": [
+                                157,
+                                261,
+                                450,
+                                273
+                            ],
+                            "group_id": 1,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        157,
+                                        259,
+                                        451,
+                                        276
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                157,
+                                                259,
+                                                451,
+                                                276
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Table 3: ZINC (MAE), ZINC-FULL (MAE) and Mol-HIV (ROC-AUC).",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 5
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "type": "table_body",
+                            "bbox": [
+                                146,
+                                273,
+                                464,
+                                402
+                            ],
+                            "group_id": 1,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        146,
+                                        273,
+                                        464,
+                                        402
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                146,
+                                                273,
+                                                464,
+                                                402
+                                            ],
+                                            "score": 0.982,
+                                            "html": "<table><tr><td rowspan=\"2\">Method</td><td colspan=\"2\">ZINC↓</td><td rowspan=\"2\">ZINC-FULL↓ All methods</td><td rowspan=\"2\">MOLHIV↑ All methods</td></tr><tr><td>No Edge Feat.</td><td>With Edge Feat.</td></tr><tr><td>GCN 四</td><td>0.469±0.002</td><td>N/A</td><td>N/A</td><td>76.06±0.97</td></tr><tr><td>GAT 园</td><td>0.463±0.002</td><td>N/A</td><td>N/A</td><td>N/A</td></tr><tr><td>GatedGCN[ 自</td><td>0.422±0.006</td><td>0.363±0.009</td><td>N/A</td><td>N/A</td></tr><tr><td>GIN 四</td><td>0.408±0.008</td><td>0.252±0.014</td><td>0.088±0.002</td><td>77.07±1.49</td></tr><tr><td>PNA 四</td><td>0.320±0.032</td><td>0.188±0.004</td><td>N/A</td><td>79.05±1.32</td></tr><tr><td>DGN 回</td><td>0.219±0.010</td><td>0.168±0.003</td><td>N/A</td><td>79.70±0.97</td></tr><tr><td>HIMP 四</td><td>N/A</td><td>0.151±0.006</td><td>0.036±0.002</td><td>78.80±0.82</td></tr><tr><td>GSN 目</td><td>0.139±0.007</td><td>0.108±0.018</td><td>N/A</td><td>77.99±1.00</td></tr><tr><td>CIN-small (Ours)</td><td>0.139±0.008</td><td>0.094±0.004</td><td>0.044±0.003</td><td>80.55±1.04</td></tr><tr><td>CIN (Ours)</td><td>0.115±0.003</td><td>0.079±0.006</td><td>0.022±0.002</td><td>80.94±0.57</td></tr></table>",
+                                            "type": "table",
+                                            "image_path": "6f240739ac03f206556c984643be99722156cc3bfe46e31403ab0d94d939fec3.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 7,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        146,
+                                        273,
+                                        464,
+                                        316.0
+                                    ],
+                                    "spans": [],
+                                    "index": 6
+                                },
+                                {
+                                    "bbox": [
+                                        146,
+                                        316.0,
+                                        464,
+                                        359.0
+                                    ],
+                                    "spans": [],
+                                    "index": 7
+                                },
+                                {
+                                    "bbox": [
+                                        146,
+                                        359.0,
+                                        464,
+                                        402.0
+                                    ],
+                                    "spans": [],
+                                    "index": 8
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 6.0
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        422,
+                        504,
+                        444
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                422,
+                                505,
+                                435
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        422,
+                                        505,
+                                        435
+                                    ],
+                                    "score": 1.0,
+                                    "content": "On the remaining datasets, CIN achieves the second place. We observe that the best results are on",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                433,
+                                444,
+                                446
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        433,
+                                        444,
+                                        446
+                                    ],
+                                    "score": 1.0,
+                                    "content": "datasets from the biological and chemical domains, where rings play a relevant role.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        }
+                    ],
+                    "index": 9.5,
+                    "bbox_fs": [
+                        106,
+                        422,
+                        505,
+                        446
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        468,
+                        505,
+                        589
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                468,
+                                505,
+                                481
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        468,
+                                        505,
+                                        481
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ZINC We study the effectiveness of cellular message passing on larger scale molecular benchmarks",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                478,
+                                504,
+                                492
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        478,
+                                        206,
+                                        492
+                                    ],
+                                    "score": 1.0,
+                                    "content": "from the ZINC database",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        206,
+                                        479,
+                                        223,
+                                        491
+                                    ],
+                                    "score": 0.72,
+                                    "content": "\\pmb { \\mathbb { E } } 8 \\|",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        223,
+                                        478,
+                                        504,
+                                        492
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". ZINC (12k graphs) and ZINC-FULL (250k graphs) [24, 33, 43, 75]",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                489,
+                                506,
+                                504
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        489,
+                                        506,
+                                        504
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are two graph regression task datasets for drug constrained solubility prediction. In these experiments,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                501,
+                                506,
+                                514
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        501,
+                                        220,
+                                        514
+                                    ],
+                                    "score": 1.0,
+                                    "content": "we consider rings up to size",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        220,
+                                        501,
+                                        250,
+                                        511
+                                    ],
+                                    "score": 0.89,
+                                    "content": "k = 1 8",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        251,
+                                        501,
+                                        506,
+                                        514
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". We follow the training and evaluation procedures in [24]. Our",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                512,
+                                506,
+                                524
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        512,
+                                        506,
+                                        524
+                                    ],
+                                    "score": 1.0,
+                                    "content": "experiments encompass different scenarios, examine the impact of ablating edge features and of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                522,
+                                506,
+                                535
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        523,
+                                        496,
+                                        535
+                                    ],
+                                    "score": 1.0,
+                                    "content": "constraining the parameter budget of the architecture to 100k. All results are illustrated in Table",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        496,
+                                        522,
+                                        506,
+                                        535
+                                    ],
+                                    "score": 0.44,
+                                    "content": "\\textcircled { 3 }",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                534,
+                                506,
+                                546
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        534,
+                                        506,
+                                        546
+                                    ],
+                                    "score": 1.0,
+                                    "content": "where we also include the results for ZINC-FULL obtained by the same exact architectures. Our",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                545,
+                                505,
+                                558
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        545,
+                                        505,
+                                        558
+                                    ],
+                                    "score": 1.0,
+                                    "content": "model exhibits particularly strong performance on these benchmarks: it attains state-of-the-art results",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                556,
+                                505,
+                                568
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        556,
+                                        505,
+                                        568
+                                    ],
+                                    "score": 1.0,
+                                    "content": "on both the two dataset variants, outperforming other models by a significant margin. CIN attains",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                567,
+                                505,
+                                578
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        567,
+                                        505,
+                                        578
+                                    ],
+                                    "score": 1.0,
+                                    "content": "strong results even when constrained by the parameter budget. It still achieves state-of-the-art",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                578,
+                                506,
+                                590
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        578,
+                                        506,
+                                        590
+                                    ],
+                                    "score": 1.0,
+                                    "content": "performance on ZINC and is on-par with the best unconstrained baseline under edge-feature ablation.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        }
+                    ],
+                    "index": 16,
+                    "bbox_fs": [
+                        105,
+                        468,
+                        506,
+                        590
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        612,
+                        506,
+                        722
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                611,
+                                505,
+                                625
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        611,
+                                        505,
+                                        625
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Mol-HIV We additionally test our model on the molecular ogbg-molhiv dataset from the Open",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                623,
+                                505,
+                                636
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        623,
+                                        181,
+                                        636
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Graph Benchmark",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        181,
+                                        623,
+                                        199,
+                                        635
+                                    ],
+                                    "score": 0.69,
+                                    "content": "\\textcircled { | 4 0 | }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        199,
+                                        623,
+                                        505,
+                                        636
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(41k graphs). The task is to predict the capacity of compounds to inhibit HIV",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                634,
+                                505,
+                                647
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        634,
+                                        234,
+                                        647
+                                    ],
+                                    "score": 1.0,
+                                    "content": "replication. Rings of size up to",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        234,
+                                        635,
+                                        259,
+                                        645
+                                    ],
+                                    "score": 0.9,
+                                    "content": "k = 6",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        259,
+                                        634,
+                                        475,
+                                        647
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are considered as 2-cells. We take the architecture in",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        475,
+                                        634,
+                                        493,
+                                        646
+                                    ],
+                                    "score": 0.83,
+                                    "content": " { \\mathbb { \\left[ \\left[ 2 7 \\right] \\right] } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        493,
+                                        634,
+                                        505,
+                                        647
+                                    ],
+                                    "score": 1.0,
+                                    "content": "as",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                645,
+                                506,
+                                659
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        645,
+                                        506,
+                                        659
+                                    ],
+                                    "score": 1.0,
+                                    "content": "reference and replicate the same hyperparameter setting in our model, including the use of only 2",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                656,
+                                506,
+                                668
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        656,
+                                        506,
+                                        668
+                                    ],
+                                    "score": 1.0,
+                                    "content": "message passing layers. We report the mean of test ROC-AUC metrics at the epoch of best validation",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                667,
+                                506,
+                                679
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        667,
+                                        506,
+                                        679
+                                    ],
+                                    "score": 1.0,
+                                    "content": "performance for 10 random weight initialisations. Similarly to ZINC, we experiment with a “small”",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                677,
+                                506,
+                                691
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        678,
+                                        369,
+                                        691
+                                    ],
+                                    "score": 1.0,
+                                    "content": "model whose number of parameters is constrained in the order of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        369,
+                                        678,
+                                        390,
+                                        689
+                                    ],
+                                    "score": 0.58,
+                                    "content": "1 0 0 \\mathrm { k }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        390,
+                                        678,
+                                        417,
+                                        691
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Table",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        417,
+                                        677,
+                                        427,
+                                        691
+                                    ],
+                                    "score": 0.4,
+                                    "content": "3",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        427,
+                                        678,
+                                        506,
+                                        691
+                                    ],
+                                    "score": 1.0,
+                                    "content": "displays the results.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                688,
+                                506,
+                                702
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        688,
+                                        506,
+                                        702
+                                    ],
+                                    "score": 1.0,
+                                    "content": "CIN significantly outperforms other strong GNN baselines, even when constrained by the parameter",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                698,
+                                506,
+                                714
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        698,
+                                        217,
+                                        714
+                                    ],
+                                    "score": 1.0,
+                                    "content": "budget. Consistently with",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        217,
+                                        700,
+                                        235,
+                                        711
+                                    ],
+                                    "score": 0.86,
+                                    "content": "[ [ 2 7 ]",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        235,
+                                        698,
+                                        506,
+                                        714
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", we observe that only two layers are sufficient when performing",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                710,
+                                396,
+                                724
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        710,
+                                        396,
+                                        724
+                                    ],
+                                    "score": 1.0,
+                                    "content": "hierarchical message passing across meso-scale structures such as rings.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        }
+                    ],
+                    "index": 26.5,
+                    "bbox_fs": [
+                        104,
+                        611,
+                        506,
+                        724
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "title",
+                    "bbox": [
+                        106,
+                        70,
+                        339,
+                        84
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                71,
+                                340,
+                                86
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        71,
+                                        340,
+                                        86
+                                    ],
+                                    "score": 1.0,
+                                    "content": "6 Related Work, Discussion and Conclusion",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        }
+                    ],
+                    "index": 0
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        97,
+                        505,
+                        185
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                96,
+                                505,
+                                109
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        96,
+                                        505,
+                                        109
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Cell complex models Recent works have proposed the generalisation of GNNs to simplicial",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                107,
+                                506,
+                                120
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        107,
+                                        506,
+                                        120
+                                    ],
+                                    "score": 1.0,
+                                    "content": "complexes [13, 25, 32, 35]. All these simplicial methods are subsumed by the model in Bodnar",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                118,
+                                505,
+                                132
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        118,
+                                        129,
+                                        132
+                                    ],
+                                    "score": 1.0,
+                                    "content": "et al.",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        129,
+                                        119,
+                                        141,
+                                        131
+                                    ],
+                                    "score": 0.41,
+                                    "content": "\\pmb { \\bigtriangledown }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        142,
+                                        118,
+                                        441,
+                                        132
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", which CWNs in turn subsume. To the best of our knowledge, Hajij et al.",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        442,
+                                        119,
+                                        460,
+                                        131
+                                    ],
+                                    "score": 0.65,
+                                    "content": "\\pmb { \\Vert 3 4 \\Vert }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        460,
+                                        118,
+                                        505,
+                                        132
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is the only",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                131,
+                                505,
+                                142
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        131,
+                                        505,
+                                        142
+                                    ],
+                                    "score": 1.0,
+                                    "content": "other example of message passing on cell complexes, but this work does not study the expressive",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                140,
+                                506,
+                                154
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        140,
+                                        506,
+                                        154
+                                    ],
+                                    "score": 1.0,
+                                    "content": "power of the proposed scheme, neither it experimentally validates its performance. In contrast, our",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                151,
+                                506,
+                                165
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        151,
+                                        506,
+                                        165
+                                    ],
+                                    "score": 1.0,
+                                    "content": "work comprehensively characterises the expressiveness of cellular message passing, and introduces a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                162,
+                                507,
+                                176
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        162,
+                                        507,
+                                        176
+                                    ],
+                                    "score": 1.0,
+                                    "content": "theoretically grounded and empirically effective framework to apply it on graph structured data in a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                174,
+                                390,
+                                186
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        174,
+                                        390,
+                                        186
+                                    ],
+                                    "score": 1.0,
+                                    "content": "way to address several limitations of standard Graph Neural Networks.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        }
+                    ],
+                    "index": 4.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        199,
+                        505,
+                        298
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                199,
+                                506,
+                                211
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        199,
+                                        506,
+                                        211
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Molecular substructures A few other works have extended GNNs to account for molecular",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                209,
+                                505,
+                                223
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        209,
+                                        505,
+                                        223
+                                    ],
+                                    "score": 1.0,
+                                    "content": "substructures. Junction Trees (JT), which conveniently represent singletons, bonds and rings as",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                221,
+                                505,
+                                234
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        221,
+                                        374,
+                                        234
+                                    ],
+                                    "score": 1.0,
+                                    "content": "supernodes in a tree, have been used in molecular graph generation",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        374,
+                                        221,
+                                        406,
+                                        232
+                                    ],
+                                    "score": 0.55,
+                                    "content": "\\mathbb { \\lVert 4 3 \\rVert \\overline { { \\lVert 4 4 \\rVert } } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        406,
+                                        221,
+                                        505,
+                                        234
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". JTs are also used in the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                231,
+                                505,
+                                245
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        231,
+                                        207,
+                                        245
+                                    ],
+                                    "score": 1.0,
+                                    "content": "recent work of Fey et al.",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        207,
+                                        231,
+                                        225,
+                                        243
+                                    ],
+                                    "score": 0.41,
+                                    "content": " { \\mathbb { \\left[ \\left[ 2 7 \\right] \\right] } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        225,
+                                        231,
+                                        505,
+                                        245
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", who employs them to design a hierarchical message passing scheme",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                242,
+                                505,
+                                255
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        242,
+                                        505,
+                                        255
+                                    ],
+                                    "score": 1.0,
+                                    "content": "based on the tree structure. However, this hierarchy has a different configuration than the one cell",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                253,
+                                506,
+                                267
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        253,
+                                        369,
+                                        267
+                                    ],
+                                    "score": 1.0,
+                                    "content": "complexes provide. Information about cycles is also used in GSNs",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        370,
+                                        253,
+                                        387,
+                                        265
+                                    ],
+                                    "score": 0.82,
+                                    "content": "\\mathbb { m }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        387,
+                                        253,
+                                        506,
+                                        267
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to augment the node features,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                264,
+                                506,
+                                278
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        264,
+                                        506,
+                                        278
+                                    ],
+                                    "score": 1.0,
+                                    "content": "but the model retains the usual message passing procedure of GNNs. These last two models are of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                276,
+                                505,
+                                288
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        276,
+                                        505,
+                                        288
+                                    ],
+                                    "score": 1.0,
+                                    "content": "particular relevance to the present work, since they utilise information about chemical rings. It is",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                286,
+                                483,
+                                300
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        286,
+                                        483,
+                                        300
+                                    ],
+                                    "score": 1.0,
+                                    "content": "important to remark that CWNs compare favourably with both of them in all our benchmarks.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        }
+                    ],
+                    "index": 13
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        312,
+                        505,
+                        400
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                312,
+                                505,
+                                324
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        312,
+                                        416,
+                                        324
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Higher-order GNNs A related line of work has studied lifting graphs into",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        417,
+                                        313,
+                                        423,
+                                        322
+                                    ],
+                                    "score": 0.8,
+                                    "content": "k",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        424,
+                                        312,
+                                        505,
+                                        324
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-dimensional tensor",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                322,
+                                506,
+                                335
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        322,
+                                        345,
+                                        335
+                                    ],
+                                    "score": 1.0,
+                                    "content": "representations that can be processed by provably expressive",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        346,
+                                        324,
+                                        353,
+                                        333
+                                    ],
+                                    "score": 0.8,
+                                    "content": "k",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        353,
+                                        322,
+                                        506,
+                                        335
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-GNNs [4, 52, 53]. With higher values",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                333,
+                                507,
+                                347
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        333,
+                                        117,
+                                        347
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        118,
+                                        334,
+                                        124,
+                                        344
+                                    ],
+                                    "score": 0.8,
+                                    "content": "k",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        124,
+                                        333,
+                                        507,
+                                        347
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", these models achieve higher-expressivity, but due to the computational complexity this incurs,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                344,
+                                505,
+                                357
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        344,
+                                        144,
+                                        357
+                                    ],
+                                    "score": 1.0,
+                                    "content": "values of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        145,
+                                        345,
+                                        170,
+                                        356
+                                    ],
+                                    "score": 0.91,
+                                    "content": "k \\geq 3",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        170,
+                                        344,
+                                        505,
+                                        357
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are of little use in practice. Therefore, unlike CWNs, these models cannot explicitly",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                355,
+                                506,
+                                370
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        355,
+                                        506,
+                                        370
+                                    ],
+                                    "score": 1.0,
+                                    "content": "represent in practice chemical rings of common sizes (e.g. five or six). Furthermore, by being",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                366,
+                                505,
+                                379
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        366,
+                                        505,
+                                        379
+                                    ],
+                                    "score": 1.0,
+                                    "content": "upper-bounded by 3-WL, the 2-GNN models cannot count the number of induced cycles of size",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                377,
+                                506,
+                                391
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        377,
+                                        235,
+                                        391
+                                    ],
+                                    "score": 1.0,
+                                    "content": "greater than four (see Appendix",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        235,
+                                        377,
+                                        247,
+                                        390
+                                    ],
+                                    "score": 0.52,
+                                    "content": "\\mathbf { A }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        248,
+                                        377,
+                                        506,
+                                        391
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for details). In contrast, CWNs can easily count these important",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                388,
+                                426,
+                                401
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        388,
+                                        426,
+                                        401
+                                    ],
+                                    "score": 1.0,
+                                    "content": "chemical substructures through the readout operation it performs on the 2-cells.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        }
+                    ],
+                    "index": 21.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        414,
+                        505,
+                        501
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                413,
+                                507,
+                                427
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        413,
+                                        507,
+                                        427
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Limitations The main limitations of the model are of computational nature. While the compu-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                424,
+                                506,
+                                438
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        424,
+                                        506,
+                                        438
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tational complexity of the message passing procedure and its preprocessing step is suitable for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                436,
+                                505,
+                                448
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        436,
+                                        505,
+                                        448
+                                    ],
+                                    "score": 1.0,
+                                    "content": "molecular and geometric graphs, the number of rings (and more generally simple cycles) in general",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                447,
+                                505,
+                                459
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        447,
+                                        505,
+                                        459
+                                    ],
+                                    "score": 1.0,
+                                    "content": "graphs can be exponential in the number of nodes. In that case, one has to resort to smaller 2-cells like",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                457,
+                                505,
+                                471
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        457,
+                                        505,
+                                        471
+                                    ],
+                                    "score": 1.0,
+                                    "content": "triangles, which can be found efficiently in general graphs. Moreover, one has to typically use weights",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                468,
+                                505,
+                                481
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        468,
+                                        505,
+                                        481
+                                    ],
+                                    "score": 1.0,
+                                    "content": "specific for each dimension of the cell complex, increasing the number of parameters compared to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                479,
+                                506,
+                                491
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        479,
+                                        506,
+                                        491
+                                    ],
+                                    "score": 1.0,
+                                    "content": "GNNs. However, we have shown that our model can compensate this increase with a reduced number",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                490,
+                                457,
+                                502
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        490,
+                                        457,
+                                        502
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of layers and still achieve state-of-the-art results on some of the molecular benchmarks.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        }
+                    ],
+                    "index": 29.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        507,
+                        505,
+                        583
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                506,
+                                505,
+                                520
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        506,
+                                        505,
+                                        520
+                                    ],
+                                    "score": 1.0,
+                                    "content": "From a theoretical point of view, this work is concerned only with regular cell complexes. Adopting",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                518,
+                                506,
+                                530
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        518,
+                                        506,
+                                        530
+                                    ],
+                                    "score": 1.0,
+                                    "content": "this restriction is useful from multiple perspectives: regular cell complexes are easier to analyse, their",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                529,
+                                505,
+                                541
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        529,
+                                        505,
+                                        541
+                                    ],
+                                    "score": 1.0,
+                                    "content": "combinatorial structure completely describes their topology and convolutions can be defined on them",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                539,
+                                505,
+                                551
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        539,
+                                        505,
+                                        551
+                                    ],
+                                    "score": 1.0,
+                                    "content": "through the Sheaf Laplacian (see Appendix D). Nonetheless, some of our theoretical results could be",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                551,
+                                505,
+                                562
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        551,
+                                        505,
+                                        562
+                                    ],
+                                    "score": 1.0,
+                                    "content": "extended to non-regular complexes, which could be obtained by lifting transformations not studied in",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                561,
+                                505,
+                                574
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        561,
+                                        505,
+                                        574
+                                    ],
+                                    "score": 1.0,
+                                    "content": "this work, such as attaching 2-cells to paths in the graph. We leave addressing non-regular complexes",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                573,
+                                331,
+                                583
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        573,
+                                        331,
+                                        583
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and their trade-offs to future developments of this work.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        }
+                    ],
+                    "index": 37
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        597,
+                        505,
+                        653
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                597,
+                                505,
+                                610
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        597,
+                                        505,
+                                        610
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Societal Impacts Most of our paper is theoretical in nature and we do not see immediate direct",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                609,
+                                505,
+                                622
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        609,
+                                        505,
+                                        622
+                                    ],
+                                    "score": 1.0,
+                                    "content": "negative societal impacts. Within the scope of social network applications, we do not yet have",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                618,
+                                505,
+                                633
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        618,
+                                        505,
+                                        633
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sufficient evidence of performance improvement on related benchmarks to justify obvious adoption in",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                631,
+                                505,
+                                643
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        631,
+                                        505,
+                                        643
+                                    ],
+                                    "score": 1.0,
+                                    "content": "such a domain. In contrast, the empirical performance on molecular benchmarks suggests it may have",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                641,
+                                505,
+                                655
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        641,
+                                        485,
+                                        655
+                                    ],
+                                    "score": 1.0,
+                                    "content": "a positive impact on applications of immediate interest in pharmaceutics, such as drug discovery",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        486,
+                                        641,
+                                        503,
+                                        653
+                                    ],
+                                    "score": 0.34,
+                                    "content": "\\pmb { \\mathbb { B } } 0 \\|",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        503,
+                                        641,
+                                        505,
+                                        655
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        }
+                    ],
+                    "index": 43
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        667,
+                        504,
+                        722
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                666,
+                                506,
+                                680
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        666,
+                                        506,
+                                        680
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Conclusion We have proposed a provably powerful message passing procedure on cell complexes",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                678,
+                                505,
+                                690
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        678,
+                                        505,
+                                        690
+                                    ],
+                                    "score": 1.0,
+                                    "content": "motivated by a novel colour refinement algorithm to test their isomorphism. This allows us to consider",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 47
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                688,
+                                506,
+                                702
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        688,
+                                        506,
+                                        702
+                                    ],
+                                    "score": 1.0,
+                                    "content": "flexible lifting operations on graphs to implement more expressive architectures which benefit from",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 48
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                699,
+                                505,
+                                713
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        699,
+                                        505,
+                                        713
+                                    ],
+                                    "score": 1.0,
+                                    "content": "decoupling the computational and input graphs. Our methods show excellent performance on diverse",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 49
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                711,
+                                300,
+                                723
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        711,
+                                        300,
+                                        723
+                                    ],
+                                    "score": 1.0,
+                                    "content": "synthetic and real-world molecular benchmarks.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 50
+                        }
+                    ],
+                    "index": 48
+                }
+            ],
+            "page_idx": 9,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        301,
+                        742,
+                        311,
+                        750
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                299,
+                                740,
+                                313,
+                                754
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        299,
+                                        740,
+                                        313,
+                                        754
+                                    ],
+                                    "score": 1.0,
+                                    "content": "",
+                                    "type": "text",
+                                    "height": 14,
+                                    "width": 14
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "title",
+                    "bbox": [
+                        106,
+                        70,
+                        339,
+                        84
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                71,
+                                340,
+                                86
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        71,
+                                        340,
+                                        86
+                                    ],
+                                    "score": 1.0,
+                                    "content": "6 Related Work, Discussion and Conclusion",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        }
+                    ],
+                    "index": 0
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        97,
+                        505,
+                        185
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                96,
+                                505,
+                                109
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        96,
+                                        505,
+                                        109
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Cell complex models Recent works have proposed the generalisation of GNNs to simplicial",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                107,
+                                506,
+                                120
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        107,
+                                        506,
+                                        120
+                                    ],
+                                    "score": 1.0,
+                                    "content": "complexes [13, 25, 32, 35]. All these simplicial methods are subsumed by the model in Bodnar",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                118,
+                                505,
+                                132
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        118,
+                                        129,
+                                        132
+                                    ],
+                                    "score": 1.0,
+                                    "content": "et al.",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        129,
+                                        119,
+                                        141,
+                                        131
+                                    ],
+                                    "score": 0.41,
+                                    "content": "\\pmb { \\bigtriangledown }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        142,
+                                        118,
+                                        441,
+                                        132
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", which CWNs in turn subsume. To the best of our knowledge, Hajij et al.",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        442,
+                                        119,
+                                        460,
+                                        131
+                                    ],
+                                    "score": 0.65,
+                                    "content": "\\pmb { \\Vert 3 4 \\Vert }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        460,
+                                        118,
+                                        505,
+                                        132
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is the only",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                131,
+                                505,
+                                142
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        131,
+                                        505,
+                                        142
+                                    ],
+                                    "score": 1.0,
+                                    "content": "other example of message passing on cell complexes, but this work does not study the expressive",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                140,
+                                506,
+                                154
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        140,
+                                        506,
+                                        154
+                                    ],
+                                    "score": 1.0,
+                                    "content": "power of the proposed scheme, neither it experimentally validates its performance. In contrast, our",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                151,
+                                506,
+                                165
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        151,
+                                        506,
+                                        165
+                                    ],
+                                    "score": 1.0,
+                                    "content": "work comprehensively characterises the expressiveness of cellular message passing, and introduces a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                162,
+                                507,
+                                176
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        162,
+                                        507,
+                                        176
+                                    ],
+                                    "score": 1.0,
+                                    "content": "theoretically grounded and empirically effective framework to apply it on graph structured data in a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                174,
+                                390,
+                                186
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        174,
+                                        390,
+                                        186
+                                    ],
+                                    "score": 1.0,
+                                    "content": "way to address several limitations of standard Graph Neural Networks.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        }
+                    ],
+                    "index": 4.5,
+                    "bbox_fs": [
+                        105,
+                        96,
+                        507,
+                        186
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        199,
+                        505,
+                        298
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                199,
+                                506,
+                                211
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        199,
+                                        506,
+                                        211
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Molecular substructures A few other works have extended GNNs to account for molecular",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                209,
+                                505,
+                                223
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        209,
+                                        505,
+                                        223
+                                    ],
+                                    "score": 1.0,
+                                    "content": "substructures. Junction Trees (JT), which conveniently represent singletons, bonds and rings as",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                221,
+                                505,
+                                234
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        221,
+                                        374,
+                                        234
+                                    ],
+                                    "score": 1.0,
+                                    "content": "supernodes in a tree, have been used in molecular graph generation",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        374,
+                                        221,
+                                        406,
+                                        232
+                                    ],
+                                    "score": 0.55,
+                                    "content": "\\mathbb { \\lVert 4 3 \\rVert \\overline { { \\lVert 4 4 \\rVert } } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        406,
+                                        221,
+                                        505,
+                                        234
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". JTs are also used in the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                231,
+                                505,
+                                245
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        231,
+                                        207,
+                                        245
+                                    ],
+                                    "score": 1.0,
+                                    "content": "recent work of Fey et al.",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        207,
+                                        231,
+                                        225,
+                                        243
+                                    ],
+                                    "score": 0.41,
+                                    "content": " { \\mathbb { \\left[ \\left[ 2 7 \\right] \\right] } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        225,
+                                        231,
+                                        505,
+                                        245
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", who employs them to design a hierarchical message passing scheme",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                242,
+                                505,
+                                255
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        242,
+                                        505,
+                                        255
+                                    ],
+                                    "score": 1.0,
+                                    "content": "based on the tree structure. However, this hierarchy has a different configuration than the one cell",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                253,
+                                506,
+                                267
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        253,
+                                        369,
+                                        267
+                                    ],
+                                    "score": 1.0,
+                                    "content": "complexes provide. Information about cycles is also used in GSNs",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        370,
+                                        253,
+                                        387,
+                                        265
+                                    ],
+                                    "score": 0.82,
+                                    "content": "\\mathbb { m }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        387,
+                                        253,
+                                        506,
+                                        267
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to augment the node features,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                264,
+                                506,
+                                278
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        264,
+                                        506,
+                                        278
+                                    ],
+                                    "score": 1.0,
+                                    "content": "but the model retains the usual message passing procedure of GNNs. These last two models are of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                276,
+                                505,
+                                288
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        276,
+                                        505,
+                                        288
+                                    ],
+                                    "score": 1.0,
+                                    "content": "particular relevance to the present work, since they utilise information about chemical rings. It is",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                286,
+                                483,
+                                300
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        286,
+                                        483,
+                                        300
+                                    ],
+                                    "score": 1.0,
+                                    "content": "important to remark that CWNs compare favourably with both of them in all our benchmarks.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        }
+                    ],
+                    "index": 13,
+                    "bbox_fs": [
+                        105,
+                        199,
+                        506,
+                        300
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        312,
+                        505,
+                        400
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                312,
+                                505,
+                                324
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        312,
+                                        416,
+                                        324
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Higher-order GNNs A related line of work has studied lifting graphs into",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        417,
+                                        313,
+                                        423,
+                                        322
+                                    ],
+                                    "score": 0.8,
+                                    "content": "k",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        424,
+                                        312,
+                                        505,
+                                        324
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-dimensional tensor",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                322,
+                                506,
+                                335
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        322,
+                                        345,
+                                        335
+                                    ],
+                                    "score": 1.0,
+                                    "content": "representations that can be processed by provably expressive",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        346,
+                                        324,
+                                        353,
+                                        333
+                                    ],
+                                    "score": 0.8,
+                                    "content": "k",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        353,
+                                        322,
+                                        506,
+                                        335
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-GNNs [4, 52, 53]. With higher values",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                333,
+                                507,
+                                347
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        333,
+                                        117,
+                                        347
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        118,
+                                        334,
+                                        124,
+                                        344
+                                    ],
+                                    "score": 0.8,
+                                    "content": "k",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        124,
+                                        333,
+                                        507,
+                                        347
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", these models achieve higher-expressivity, but due to the computational complexity this incurs,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                344,
+                                505,
+                                357
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        344,
+                                        144,
+                                        357
+                                    ],
+                                    "score": 1.0,
+                                    "content": "values of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        145,
+                                        345,
+                                        170,
+                                        356
+                                    ],
+                                    "score": 0.91,
+                                    "content": "k \\geq 3",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        170,
+                                        344,
+                                        505,
+                                        357
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are of little use in practice. Therefore, unlike CWNs, these models cannot explicitly",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                355,
+                                506,
+                                370
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        355,
+                                        506,
+                                        370
+                                    ],
+                                    "score": 1.0,
+                                    "content": "represent in practice chemical rings of common sizes (e.g. five or six). Furthermore, by being",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                366,
+                                505,
+                                379
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        366,
+                                        505,
+                                        379
+                                    ],
+                                    "score": 1.0,
+                                    "content": "upper-bounded by 3-WL, the 2-GNN models cannot count the number of induced cycles of size",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                377,
+                                506,
+                                391
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        377,
+                                        235,
+                                        391
+                                    ],
+                                    "score": 1.0,
+                                    "content": "greater than four (see Appendix",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        235,
+                                        377,
+                                        247,
+                                        390
+                                    ],
+                                    "score": 0.52,
+                                    "content": "\\mathbf { A }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        248,
+                                        377,
+                                        506,
+                                        391
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for details). In contrast, CWNs can easily count these important",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                388,
+                                426,
+                                401
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        388,
+                                        426,
+                                        401
+                                    ],
+                                    "score": 1.0,
+                                    "content": "chemical substructures through the readout operation it performs on the 2-cells.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        }
+                    ],
+                    "index": 21.5,
+                    "bbox_fs": [
+                        104,
+                        312,
+                        507,
+                        401
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        414,
+                        505,
+                        501
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                413,
+                                507,
+                                427
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        413,
+                                        507,
+                                        427
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Limitations The main limitations of the model are of computational nature. While the compu-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                424,
+                                506,
+                                438
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        424,
+                                        506,
+                                        438
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tational complexity of the message passing procedure and its preprocessing step is suitable for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                436,
+                                505,
+                                448
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        436,
+                                        505,
+                                        448
+                                    ],
+                                    "score": 1.0,
+                                    "content": "molecular and geometric graphs, the number of rings (and more generally simple cycles) in general",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                447,
+                                505,
+                                459
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        447,
+                                        505,
+                                        459
+                                    ],
+                                    "score": 1.0,
+                                    "content": "graphs can be exponential in the number of nodes. In that case, one has to resort to smaller 2-cells like",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                457,
+                                505,
+                                471
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        457,
+                                        505,
+                                        471
+                                    ],
+                                    "score": 1.0,
+                                    "content": "triangles, which can be found efficiently in general graphs. Moreover, one has to typically use weights",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                468,
+                                505,
+                                481
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        468,
+                                        505,
+                                        481
+                                    ],
+                                    "score": 1.0,
+                                    "content": "specific for each dimension of the cell complex, increasing the number of parameters compared to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                479,
+                                506,
+                                491
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        479,
+                                        506,
+                                        491
+                                    ],
+                                    "score": 1.0,
+                                    "content": "GNNs. However, we have shown that our model can compensate this increase with a reduced number",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                490,
+                                457,
+                                502
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        490,
+                                        457,
+                                        502
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of layers and still achieve state-of-the-art results on some of the molecular benchmarks.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        }
+                    ],
+                    "index": 29.5,
+                    "bbox_fs": [
+                        105,
+                        413,
+                        507,
+                        502
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        507,
+                        505,
+                        583
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                506,
+                                505,
+                                520
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        506,
+                                        505,
+                                        520
+                                    ],
+                                    "score": 1.0,
+                                    "content": "From a theoretical point of view, this work is concerned only with regular cell complexes. Adopting",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                518,
+                                506,
+                                530
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        518,
+                                        506,
+                                        530
+                                    ],
+                                    "score": 1.0,
+                                    "content": "this restriction is useful from multiple perspectives: regular cell complexes are easier to analyse, their",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                529,
+                                505,
+                                541
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        529,
+                                        505,
+                                        541
+                                    ],
+                                    "score": 1.0,
+                                    "content": "combinatorial structure completely describes their topology and convolutions can be defined on them",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                539,
+                                505,
+                                551
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        539,
+                                        505,
+                                        551
+                                    ],
+                                    "score": 1.0,
+                                    "content": "through the Sheaf Laplacian (see Appendix D). Nonetheless, some of our theoretical results could be",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                551,
+                                505,
+                                562
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        551,
+                                        505,
+                                        562
+                                    ],
+                                    "score": 1.0,
+                                    "content": "extended to non-regular complexes, which could be obtained by lifting transformations not studied in",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                561,
+                                505,
+                                574
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        561,
+                                        505,
+                                        574
+                                    ],
+                                    "score": 1.0,
+                                    "content": "this work, such as attaching 2-cells to paths in the graph. We leave addressing non-regular complexes",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                573,
+                                331,
+                                583
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        573,
+                                        331,
+                                        583
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and their trade-offs to future developments of this work.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        }
+                    ],
+                    "index": 37,
+                    "bbox_fs": [
+                        105,
+                        506,
+                        506,
+                        583
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        597,
+                        505,
+                        653
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                597,
+                                505,
+                                610
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        597,
+                                        505,
+                                        610
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Societal Impacts Most of our paper is theoretical in nature and we do not see immediate direct",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                609,
+                                505,
+                                622
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        609,
+                                        505,
+                                        622
+                                    ],
+                                    "score": 1.0,
+                                    "content": "negative societal impacts. Within the scope of social network applications, we do not yet have",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                618,
+                                505,
+                                633
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        618,
+                                        505,
+                                        633
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sufficient evidence of performance improvement on related benchmarks to justify obvious adoption in",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                631,
+                                505,
+                                643
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        631,
+                                        505,
+                                        643
+                                    ],
+                                    "score": 1.0,
+                                    "content": "such a domain. In contrast, the empirical performance on molecular benchmarks suggests it may have",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                641,
+                                505,
+                                655
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        641,
+                                        485,
+                                        655
+                                    ],
+                                    "score": 1.0,
+                                    "content": "a positive impact on applications of immediate interest in pharmaceutics, such as drug discovery",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        486,
+                                        641,
+                                        503,
+                                        653
+                                    ],
+                                    "score": 0.34,
+                                    "content": "\\pmb { \\mathbb { B } } 0 \\|",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        503,
+                                        641,
+                                        505,
+                                        655
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        }
+                    ],
+                    "index": 43,
+                    "bbox_fs": [
+                        104,
+                        597,
+                        505,
+                        655
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        667,
+                        504,
+                        722
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                666,
+                                506,
+                                680
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        666,
+                                        506,
+                                        680
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Conclusion We have proposed a provably powerful message passing procedure on cell complexes",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                678,
+                                505,
+                                690
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        678,
+                                        505,
+                                        690
+                                    ],
+                                    "score": 1.0,
+                                    "content": "motivated by a novel colour refinement algorithm to test their isomorphism. This allows us to consider",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 47
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                688,
+                                506,
+                                702
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        688,
+                                        506,
+                                        702
+                                    ],
+                                    "score": 1.0,
+                                    "content": "flexible lifting operations on graphs to implement more expressive architectures which benefit from",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 48
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                699,
+                                505,
+                                713
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        699,
+                                        505,
+                                        713
+                                    ],
+                                    "score": 1.0,
+                                    "content": "decoupling the computational and input graphs. Our methods show excellent performance on diverse",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 49
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                711,
+                                300,
+                                723
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        711,
+                                        300,
+                                        723
+                                    ],
+                                    "score": 1.0,
+                                    "content": "synthetic and real-world molecular benchmarks.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 50
+                        }
+                    ],
+                    "index": 48,
+                    "bbox_fs": [
+                        105,
+                        666,
+                        506,
+                        723
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        71,
+                        275,
+                        85
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                70,
+                                277,
+                                88
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        70,
+                                        277,
+                                        88
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Funding and Acknowledgements",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        }
+                    ],
+                    "index": 0
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        95,
+                        506,
+                        162
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                95,
+                                506,
+                                108
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        95,
+                                        506,
+                                        108
+                                    ],
+                                    "score": 1.0,
+                                    "content": "YW and GM acknowledge support from the ERC under the EU’s Horizon 2020 programme (grant",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                106,
+                                507,
+                                119
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        106,
+                                        151,
+                                        119
+                                    ],
+                                    "score": 1.0,
+                                    "content": "agreement",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        151,
+                                        107,
+                                        162,
+                                        117
+                                    ],
+                                    "score": 0.35,
+                                    "content": " { \\mathrm { n ^ { o } } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        162,
+                                        106,
+                                        414,
+                                        119
+                                    ],
+                                    "score": 1.0,
+                                    "content": "757983). MB is supported in part by ERC Consolidator grant",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        414,
+                                        107,
+                                        425,
+                                        117
+                                    ],
+                                    "score": 0.26,
+                                    "content": "\\boldsymbol { \\mathrm { n ^ { \\circ } } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        425,
+                                        106,
+                                        507,
+                                        119
+                                    ],
+                                    "score": 1.0,
+                                    "content": "724228 (LEMAN).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                117,
+                                506,
+                                131
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        117,
+                                        506,
+                                        131
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The authors declare no competing interests. We are also grateful to Ben Day, Gabriele Corso and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                127,
+                                506,
+                                141
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        127,
+                                        506,
+                                        141
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Nikola Simidjievski for their helpful feedback. We would also like to thank Vijay P. Dwivedi and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                139,
+                                506,
+                                152
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        139,
+                                        431,
+                                        152
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Chaitanya K. Joshi for clarifying certain aspects of their Benchmarking GNNs",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        432,
+                                        139,
+                                        449,
+                                        151
+                                    ],
+                                    "score": 0.49,
+                                    "content": "\\pmb { \\bigtriangledown }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        450,
+                                        139,
+                                        506,
+                                        152
+                                    ],
+                                    "score": 1.0,
+                                    "content": "work, and to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                149,
+                                485,
+                                163
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        149,
+                                        485,
+                                        163
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Muhammet Balcilar for signalling a numerical precision issue in early SR graphs experiments.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        }
+                    ],
+                    "index": 3.5
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        177,
+                        163,
+                        190
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                176,
+                                165,
+                                192
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        176,
+                                        165,
+                                        192
+                                    ],
+                                    "score": 1.0,
+                                    "content": "References",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        }
+                    ],
+                    "index": 7
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        196,
+                        507,
+                        725
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                109,
+                                195,
+                                506,
+                                209
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        109,
+                                        195,
+                                        506,
+                                        209
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[1] Uri Alon and Eran Yahav. On the bottleneck of graph neural networks and its practical",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                128,
+                                207,
+                                236,
+                                220
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        128,
+                                        207,
+                                        236,
+                                        220
+                                    ],
+                                    "score": 1.0,
+                                    "content": "implications. ICLR, 2021.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                110,
+                                225,
+                                505,
+                                240
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        110,
+                                        225,
+                                        505,
+                                        240
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[2] V. Arvind, Frank Fuhlbrück, Johannes Köbler, and Oleg Verbitsky. On weisfeiler-leman",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                237,
+                                506,
+                                250
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        237,
+                                        506,
+                                        250
+                                    ],
+                                    "score": 1.0,
+                                    "content": "invariance: Subgraph counts and related graph properties. Journal of Computer and System",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                248,
+                                313,
+                                261
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        248,
+                                        313,
+                                        261
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Sciences, 113:42–59, 2020. ISSN 0022-0000.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                109,
+                                265,
+                                506,
+                                281
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        109,
+                                        265,
+                                        506,
+                                        281
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[3] James Atwood and Don Towsley. Diffusion-convolutional neural networks. In NIPS, pages",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                277,
+                                202,
+                                290
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        277,
+                                        202,
+                                        290
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1993–2001, 2016.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                111,
+                                297,
+                                506,
+                                309
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        111,
+                                        297,
+                                        506,
+                                        309
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[4] Waiss Azizian and Marc Lelarge. Expressive power of invariant and equivariant graph neural",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                306,
+                                506,
+                                321
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        306,
+                                        506,
+                                        321
+                                    ],
+                                    "score": 1.0,
+                                    "content": "networks. In International Conference on Learning Representations, 2021. URL https:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                128,
+                                319,
+                                325,
+                                332
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        128,
+                                        319,
+                                        325,
+                                        332
+                                    ],
+                                    "score": 1.0,
+                                    "content": "//openreview.net/forum?id=lxHgXYN4bwl.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                110,
+                                336,
+                                506,
+                                351
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        110,
+                                        336,
+                                        506,
+                                        351
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[5] Jimmy Lei Ba, Jamie Ryan Kiros, and Geoffrey E. Hinton. Layer normalization. arXiv preprint",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                348,
+                                231,
+                                359
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        348,
+                                        231,
+                                        359
+                                    ],
+                                    "score": 1.0,
+                                    "content": "arXiv:1607.06450, 2016.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                110,
+                                366,
+                                507,
+                                381
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        110,
+                                        366,
+                                        507,
+                                        381
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[6] Dominique Beaini, Saro Passaro, Vincent Létourneau, William L Hamilton, Gabriele Corso,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                378,
+                                360,
+                                390
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        378,
+                                        360,
+                                        390
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and Pietro Liò. Directional graph networks. ICML, 2021.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                110,
+                                396,
+                                506,
+                                410
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        110,
+                                        396,
+                                        506,
+                                        410
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[7] Etienne E. Birmelé, Rui Ferreira, Roberto Grossi, Andrea Marino, Nadia Pisanti, Romeo Rizzi,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                128,
+                                407,
+                                506,
+                                420
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        128,
+                                        407,
+                                        506,
+                                        420
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and Gustavo Sacomoto. Optimal Listing of Cycles and st-Paths in Undirected Graphs. In 24th",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                125,
+                                416,
+                                506,
+                                433
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        125,
+                                        416,
+                                        506,
+                                        433
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Annual ACM-SIAM Symposium on Discrete Algorithms, New Orleans, United States, January",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                428,
+                                155,
+                                443
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        428,
+                                        155,
+                                        443
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2013.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                110,
+                                448,
+                                506,
+                                461
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        110,
+                                        448,
+                                        506,
+                                        461
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[8] Cristian Bodnar, Fabrizio Frasca, Yu Guang Wang, Nina Otter, Guido Montúfar, Pietro Lio, and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                459,
+                                506,
+                                472
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        459,
+                                        506,
+                                        472
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Michael Bronstein. Weisfeiler and lehman go topological: Message passing simplicial networks.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                470,
+                                182,
+                                483
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        470,
+                                        182,
+                                        483
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ICML, 2021.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                110,
+                                488,
+                                507,
+                                502
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        110,
+                                        488,
+                                        507,
+                                        502
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[9] Karsten M. Borgwardt, Cheng Soon Ong, Stefan Schönauer, S. V. N. Vishwanathan, Alex J.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                499,
+                                506,
+                                513
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        499,
+                                        506,
+                                        513
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Smola, and Hans-Peter Kriegel. Protein function prediction via graph kernels. Bioinformatics,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                510,
+                                239,
+                                523
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        510,
+                                        239,
+                                        523
+                                    ],
+                                    "score": 1.0,
+                                    "content": "21(suppl_1):i47–i56, 2005.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                528,
+                                506,
+                                543
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        528,
+                                        506,
+                                        543
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[10] Giorgos Bouritsas, Fabrizio Frasca, Stefanos Zafeiriou, and Michael M Bronstein. Improving",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                540,
+                                507,
+                                554
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        540,
+                                        507,
+                                        554
+                                    ],
+                                    "score": 1.0,
+                                    "content": "graph neural network expressivity via subgraph isomorphism counting. arXiv:2006.09252,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                551,
+                                153,
+                                563
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        551,
+                                        153,
+                                        563
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2020.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                569,
+                                506,
+                                584
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        569,
+                                        506,
+                                        584
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[11] Xavier Bresson and Thomas Laurent. Residual gated graph convnets. arXiv preprint",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                581,
+                                230,
+                                592
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        581,
+                                        230,
+                                        592
+                                    ],
+                                    "score": 1.0,
+                                    "content": "arXiv:1711.07553, 2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                600,
+                                506,
+                                613
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        600,
+                                        506,
+                                        613
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[12] Michael M Bronstein, Joan Bruna, Yann LeCun, Arthur Szlam, and Pierre Vandergheynst.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                128,
+                                611,
+                                506,
+                                624
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        128,
+                                        611,
+                                        506,
+                                        624
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Geometric deep learning: going beyond euclidean data. IEEE Signal Processing Magazine, 34",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                621,
+                                198,
+                                633
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        621,
+                                        198,
+                                        633
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(4):18–42, 2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                640,
+                                506,
+                                654
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        640,
+                                        506,
+                                        654
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[13] Eric Bunch, Qian You, Glenn Fung, and Vikas Singh. Simplicial 2-complex convolutional",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                651,
+                                485,
+                                664
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        651,
+                                        485,
+                                        664
+                                    ],
+                                    "score": 1.0,
+                                    "content": "neural networks. In NeurIPS Workshop on Topological Data Analysis and Beyond, 2020.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                670,
+                                505,
+                                683
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        670,
+                                        505,
+                                        683
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[14] Zhengdao Chen, Soledad Villar, Lei Chen, and Joan Bruna. On the equivalence between graph",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                681,
+                                492,
+                                694
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        681,
+                                        492,
+                                        694
+                                    ],
+                                    "score": 1.0,
+                                    "content": "isomorphism testing and function approximation with gnns. In NeurIPS, volume 32, 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                700,
+                                506,
+                                713
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        700,
+                                        506,
+                                        713
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[15] Zhengdao Chen, Lei Chen, Soledad Villar, and Joan Bruna. Can graph neural networks count",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                710,
+                                264,
+                                723
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        710,
+                                        264,
+                                        723
+                                    ],
+                                    "score": 1.0,
+                                    "content": "substructures? In NeurIPS, 2020.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        }
+                    ],
+                    "index": 26.5
+                }
+            ],
+            "page_idx": 10,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        300,
+                        741,
+                        310,
+                        750
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                299,
+                                740,
+                                312,
+                                755
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        299,
+                                        740,
+                                        312,
+                                        755
+                                    ],
+                                    "score": 1.0,
+                                    "content": "",
+                                    "type": "text",
+                                    "height": 15,
+                                    "width": 13
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        71,
+                        275,
+                        85
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                70,
+                                277,
+                                88
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        70,
+                                        277,
+                                        88
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Funding and Acknowledgements",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        }
+                    ],
+                    "index": 0
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        95,
+                        506,
+                        162
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                95,
+                                506,
+                                108
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        95,
+                                        506,
+                                        108
+                                    ],
+                                    "score": 1.0,
+                                    "content": "YW and GM acknowledge support from the ERC under the EU’s Horizon 2020 programme (grant",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                106,
+                                507,
+                                119
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        106,
+                                        151,
+                                        119
+                                    ],
+                                    "score": 1.0,
+                                    "content": "agreement",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        151,
+                                        107,
+                                        162,
+                                        117
+                                    ],
+                                    "score": 0.35,
+                                    "content": " { \\mathrm { n ^ { o } } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        162,
+                                        106,
+                                        414,
+                                        119
+                                    ],
+                                    "score": 1.0,
+                                    "content": "757983). MB is supported in part by ERC Consolidator grant",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        414,
+                                        107,
+                                        425,
+                                        117
+                                    ],
+                                    "score": 0.26,
+                                    "content": "\\boldsymbol { \\mathrm { n ^ { \\circ } } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        425,
+                                        106,
+                                        507,
+                                        119
+                                    ],
+                                    "score": 1.0,
+                                    "content": "724228 (LEMAN).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                117,
+                                506,
+                                131
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        117,
+                                        506,
+                                        131
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The authors declare no competing interests. We are also grateful to Ben Day, Gabriele Corso and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                127,
+                                506,
+                                141
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        127,
+                                        506,
+                                        141
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Nikola Simidjievski for their helpful feedback. We would also like to thank Vijay P. Dwivedi and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                139,
+                                506,
+                                152
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        139,
+                                        431,
+                                        152
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Chaitanya K. Joshi for clarifying certain aspects of their Benchmarking GNNs",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        432,
+                                        139,
+                                        449,
+                                        151
+                                    ],
+                                    "score": 0.49,
+                                    "content": "\\pmb { \\bigtriangledown }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        450,
+                                        139,
+                                        506,
+                                        152
+                                    ],
+                                    "score": 1.0,
+                                    "content": "work, and to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                149,
+                                485,
+                                163
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        149,
+                                        485,
+                                        163
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Muhammet Balcilar for signalling a numerical precision issue in early SR graphs experiments.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        }
+                    ],
+                    "index": 3.5,
+                    "bbox_fs": [
+                        104,
+                        95,
+                        507,
+                        163
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        177,
+                        163,
+                        190
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                176,
+                                165,
+                                192
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        176,
+                                        165,
+                                        192
+                                    ],
+                                    "score": 1.0,
+                                    "content": "References",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        }
+                    ],
+                    "index": 7
+                },
+                {
+                    "type": "list",
+                    "bbox": [
+                        106,
+                        196,
+                        507,
+                        725
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                109,
+                                195,
+                                506,
+                                209
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        109,
+                                        195,
+                                        506,
+                                        209
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[1] Uri Alon and Eran Yahav. On the bottleneck of graph neural networks and its practical",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                128,
+                                207,
+                                236,
+                                220
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        128,
+                                        207,
+                                        236,
+                                        220
+                                    ],
+                                    "score": 1.0,
+                                    "content": "implications. ICLR, 2021.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                110,
+                                225,
+                                505,
+                                240
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        110,
+                                        225,
+                                        505,
+                                        240
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[2] V. Arvind, Frank Fuhlbrück, Johannes Köbler, and Oleg Verbitsky. On weisfeiler-leman",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                237,
+                                506,
+                                250
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        237,
+                                        506,
+                                        250
+                                    ],
+                                    "score": 1.0,
+                                    "content": "invariance: Subgraph counts and related graph properties. Journal of Computer and System",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                248,
+                                313,
+                                261
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        248,
+                                        313,
+                                        261
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Sciences, 113:42–59, 2020. ISSN 0022-0000.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                109,
+                                265,
+                                506,
+                                281
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        109,
+                                        265,
+                                        506,
+                                        281
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[3] James Atwood and Don Towsley. Diffusion-convolutional neural networks. In NIPS, pages",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                277,
+                                202,
+                                290
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        277,
+                                        202,
+                                        290
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1993–2001, 2016.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                111,
+                                297,
+                                506,
+                                309
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        111,
+                                        297,
+                                        506,
+                                        309
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[4] Waiss Azizian and Marc Lelarge. Expressive power of invariant and equivariant graph neural",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                306,
+                                506,
+                                321
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        306,
+                                        506,
+                                        321
+                                    ],
+                                    "score": 1.0,
+                                    "content": "networks. In International Conference on Learning Representations, 2021. URL https:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                128,
+                                319,
+                                325,
+                                332
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        128,
+                                        319,
+                                        325,
+                                        332
+                                    ],
+                                    "score": 1.0,
+                                    "content": "//openreview.net/forum?id=lxHgXYN4bwl.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                110,
+                                336,
+                                506,
+                                351
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        110,
+                                        336,
+                                        506,
+                                        351
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[5] Jimmy Lei Ba, Jamie Ryan Kiros, and Geoffrey E. Hinton. Layer normalization. arXiv preprint",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                348,
+                                231,
+                                359
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        348,
+                                        231,
+                                        359
+                                    ],
+                                    "score": 1.0,
+                                    "content": "arXiv:1607.06450, 2016.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                110,
+                                366,
+                                507,
+                                381
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        110,
+                                        366,
+                                        507,
+                                        381
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[6] Dominique Beaini, Saro Passaro, Vincent Létourneau, William L Hamilton, Gabriele Corso,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                378,
+                                360,
+                                390
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        378,
+                                        360,
+                                        390
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and Pietro Liò. Directional graph networks. ICML, 2021.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                110,
+                                396,
+                                506,
+                                410
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        110,
+                                        396,
+                                        506,
+                                        410
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[7] Etienne E. Birmelé, Rui Ferreira, Roberto Grossi, Andrea Marino, Nadia Pisanti, Romeo Rizzi,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                128,
+                                407,
+                                506,
+                                420
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        128,
+                                        407,
+                                        506,
+                                        420
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and Gustavo Sacomoto. Optimal Listing of Cycles and st-Paths in Undirected Graphs. In 24th",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                125,
+                                416,
+                                506,
+                                433
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        125,
+                                        416,
+                                        506,
+                                        433
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Annual ACM-SIAM Symposium on Discrete Algorithms, New Orleans, United States, January",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                428,
+                                155,
+                                443
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        428,
+                                        155,
+                                        443
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2013.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                110,
+                                448,
+                                506,
+                                461
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        110,
+                                        448,
+                                        506,
+                                        461
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[8] Cristian Bodnar, Fabrizio Frasca, Yu Guang Wang, Nina Otter, Guido Montúfar, Pietro Lio, and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                459,
+                                506,
+                                472
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        459,
+                                        506,
+                                        472
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Michael Bronstein. Weisfeiler and lehman go topological: Message passing simplicial networks.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                470,
+                                182,
+                                483
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        470,
+                                        182,
+                                        483
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ICML, 2021.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                110,
+                                488,
+                                507,
+                                502
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        110,
+                                        488,
+                                        507,
+                                        502
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[9] Karsten M. Borgwardt, Cheng Soon Ong, Stefan Schönauer, S. V. N. Vishwanathan, Alex J.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                499,
+                                506,
+                                513
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        499,
+                                        506,
+                                        513
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Smola, and Hans-Peter Kriegel. Protein function prediction via graph kernels. Bioinformatics,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                510,
+                                239,
+                                523
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        510,
+                                        239,
+                                        523
+                                    ],
+                                    "score": 1.0,
+                                    "content": "21(suppl_1):i47–i56, 2005.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                528,
+                                506,
+                                543
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        528,
+                                        506,
+                                        543
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[10] Giorgos Bouritsas, Fabrizio Frasca, Stefanos Zafeiriou, and Michael M Bronstein. Improving",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                540,
+                                507,
+                                554
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        540,
+                                        507,
+                                        554
+                                    ],
+                                    "score": 1.0,
+                                    "content": "graph neural network expressivity via subgraph isomorphism counting. arXiv:2006.09252,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                551,
+                                153,
+                                563
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        551,
+                                        153,
+                                        563
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2020.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                569,
+                                506,
+                                584
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        569,
+                                        506,
+                                        584
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[11] Xavier Bresson and Thomas Laurent. Residual gated graph convnets. arXiv preprint",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                581,
+                                230,
+                                592
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        581,
+                                        230,
+                                        592
+                                    ],
+                                    "score": 1.0,
+                                    "content": "arXiv:1711.07553, 2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                600,
+                                506,
+                                613
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        600,
+                                        506,
+                                        613
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[12] Michael M Bronstein, Joan Bruna, Yann LeCun, Arthur Szlam, and Pierre Vandergheynst.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                128,
+                                611,
+                                506,
+                                624
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        128,
+                                        611,
+                                        506,
+                                        624
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Geometric deep learning: going beyond euclidean data. IEEE Signal Processing Magazine, 34",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                621,
+                                198,
+                                633
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        621,
+                                        198,
+                                        633
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(4):18–42, 2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                640,
+                                506,
+                                654
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        640,
+                                        506,
+                                        654
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[13] Eric Bunch, Qian You, Glenn Fung, and Vikas Singh. Simplicial 2-complex convolutional",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                651,
+                                485,
+                                664
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        651,
+                                        485,
+                                        664
+                                    ],
+                                    "score": 1.0,
+                                    "content": "neural networks. In NeurIPS Workshop on Topological Data Analysis and Beyond, 2020.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                670,
+                                505,
+                                683
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        670,
+                                        505,
+                                        683
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[14] Zhengdao Chen, Soledad Villar, Lei Chen, and Joan Bruna. On the equivalence between graph",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                681,
+                                492,
+                                694
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        681,
+                                        492,
+                                        694
+                                    ],
+                                    "score": 1.0,
+                                    "content": "isomorphism testing and function approximation with gnns. In NeurIPS, volume 32, 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                700,
+                                506,
+                                713
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        700,
+                                        506,
+                                        713
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[15] Zhengdao Chen, Lei Chen, Soledad Villar, and Joan Bruna. Can graph neural networks count",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                710,
+                                264,
+                                723
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        710,
+                                        264,
+                                        723
+                                    ],
+                                    "score": 1.0,
+                                    "content": "substructures? In NeurIPS, 2020.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                70,
+                                507,
+                                88
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        70,
+                                        507,
+                                        88
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[16] N. Chiba and Takao Nishizeki. Arboricity and subgraph listing algorithms. SIAM J. Comput.,",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 0,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                83,
+                                207,
+                                97
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        83,
+                                        207,
+                                        97
+                                    ],
+                                    "score": 1.0,
+                                    "content": "14:210–223, 1985.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 1,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                102,
+                                430,
+                                116
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        102,
+                                        430,
+                                        116
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[17] Fan Chung. Spectral graph theory. American Mathematical Society, 1997.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 2,
+                            "is_list_start_line": true,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                121,
+                                506,
+                                135
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        121,
+                                        506,
+                                        135
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[18] Djork-Arné Clevert, Thomas Unterthiner, and Sepp Hochreiter. Fast and accurate deep network",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 3,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                132,
+                                370,
+                                146
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        132,
+                                        370,
+                                        146
+                                    ],
+                                    "score": 1.0,
+                                    "content": "learning by exponential linear units (ELUs). In ICLR, 2016.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 4,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                149,
+                                505,
+                                164
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        149,
+                                        505,
+                                        164
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[19] Luigi P. Cordella, Pasquale Foggia, Carlo Sansone, and Mario Vento. A (sub)graph isomorphism",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 5,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                161,
+                                506,
+                                175
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        161,
+                                        506,
+                                        175
+                                    ],
+                                    "score": 1.0,
+                                    "content": "algorithm for matching large graphs. IEEE Transactions on Pattern Analysis and Machine",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                171,
+                                285,
+                                186
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        171,
+                                        285,
+                                        186
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Intelligence, 26(10):1367–1372, 2004.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 7,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                190,
+                                506,
+                                204
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        190,
+                                        506,
+                                        204
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[20] Gabriele Corso, Luca Cavalleri, Dominique Beaini, Pietro Liò, and Petar Velickovi ˇ c. Principal ´",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 8,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                201,
+                                507,
+                                216
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        201,
+                                        507,
+                                        216
+                                    ],
+                                    "score": 1.0,
+                                    "content": "neighbourhood aggregation for graph nets. In NeurIPS, volume 33, pages 13260–13271, 2020.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                220,
+                                487,
+                                234
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        220,
+                                        487,
+                                        234
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[21] Pim de Haan, Taco Cohen, and Max Welling. Natural graph networks. In NeurIPS, 2020.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 10,
+                            "is_list_start_line": true,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                238,
+                                506,
+                                253
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        238,
+                                        506,
+                                        253
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[22] Michaël Defferrard, Xavier Bresson, and Pierre Vandergheynst. Convolutional neural networks",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 11,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                250,
+                                379,
+                                263
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        250,
+                                        379,
+                                        263
+                                    ],
+                                    "score": 1.0,
+                                    "content": "on graphs with fast localized spectral filtering. In NIPS, 2016.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 12,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                268,
+                                506,
+                                283
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        268,
+                                        506,
+                                        283
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[23] Paul D Dobson and Andrew J Doig. Distinguishing enzyme structures from non-enzymes",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 13,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                280,
+                                427,
+                                292
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        280,
+                                        427,
+                                        292
+                                    ],
+                                    "score": 1.0,
+                                    "content": "without alignments. Journal of Molecular Biology, 330(4):771–783, 2003.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 14,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                297,
+                                507,
+                                313
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        297,
+                                        507,
+                                        313
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[24] Vijay Prakash Dwivedi, Chaitanya K Joshi, Thomas Laurent, Yoshua Bengio, and Xavier",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 15,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                309,
+                                485,
+                                324
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        309,
+                                        485,
+                                        324
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Bresson. Benchmarking graph neural networks. arXiv preprint arXiv:2003.00982, 2020.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 16,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                327,
+                                506,
+                                342
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        327,
+                                        506,
+                                        342
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[25] Stefania Ebli, Michaël Defferrard, and Gard Spreemann. Simplicial neural networks. In NeurIPS",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 17,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                339,
+                                366,
+                                352
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        339,
+                                        366,
+                                        352
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Workshop on Topological Data Analysis and Beyond, 2020.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 18,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                357,
+                                506,
+                                371
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        357,
+                                        506,
+                                        371
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[26] Rui Ferreira, Roberto Grossi, Romeo Rizzi, Gustavo Sacomoto, and Marie-France Sagot.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 19,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                368,
+                                504,
+                                382
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        368,
+                                        485,
+                                        382
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Amortized Õ(|V|) -Delay Algorithm for Listing Chordless Cycles in Undirected Graphs. In",
+                                    "type": "text",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        485,
+                                        369,
+                                        504,
+                                        379
+                                    ],
+                                    "score": 0.55,
+                                    "content": "2 2 t h",
+                                    "type": "inline_equation",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                380,
+                                506,
+                                392
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        380,
+                                        506,
+                                        392
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Annual European Symposium on Algorithms, volume 8737 of Algorithms ESA 2014, Lecture",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                390,
+                                450,
+                                405
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        390,
+                                        450,
+                                        405
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Notes in Computer Science, pages 418–429, Wroclaw, Poland, September 2014.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 22,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                408,
+                                506,
+                                423
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        408,
+                                        506,
+                                        423
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[27] M. Fey, J. G. Yuen, and F. Weichert. Hierarchical inter-message passing for learning on",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 23,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                419,
+                                506,
+                                433
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        419,
+                                        430,
+                                        433
+                                    ],
+                                    "score": 1.0,
+                                    "content": "molecular graphs. In ICML Graph Representation Learning and Beyond",
+                                    "type": "text",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        431,
+                                        420,
+                                        461,
+                                        431
+                                    ],
+                                    "score": 0.48,
+                                    "content": "( G R L + ,",
+                                    "type": "inline_equation",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        462,
+                                        419,
+                                        506,
+                                        433
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Workhop,",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                431,
+                                153,
+                                444
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        431,
+                                        153,
+                                        444
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2020.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 25,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                450,
+                                506,
+                                462
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        450,
+                                        506,
+                                        462
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[28] Matthias Fey and Jan Eric Lenssen. Fast graph representation learning with PyTorch Geometric.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 26,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                460,
+                                452,
+                                475
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        460,
+                                        452,
+                                        475
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In ICLR Workshop on Representation Learning on Graphs and Manifolds, 2019.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 27,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                478,
+                                506,
+                                493
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        478,
+                                        506,
+                                        493
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[29] Thomas Gärtner, Peter Flach, and Stefan Wrobel. On graph kernels: Hardness results and",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 28,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                490,
+                                505,
+                                502
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        490,
+                                        505,
+                                        502
+                                    ],
+                                    "score": 1.0,
+                                    "content": "efficient alternatives. In Learning theory and kernel machines, pages 129–143. Springer, 2003.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                507,
+                                506,
+                                522
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        507,
+                                        506,
+                                        522
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[30] Thomas Gaudelet, Ben Day, Arian R Jamasb, Jyothish Soman, Cristian Regep, Gertrude Liu,",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 30,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                519,
+                                506,
+                                533
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        519,
+                                        506,
+                                        533
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Jeremy B R Hayter, Richard Vickers, Charles Roberts, Jian Tang, David Roblin, Tom L Blundell,",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                528,
+                                506,
+                                546
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        528,
+                                        506,
+                                        546
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Michael M Bronstein, and Jake P Taylor-King. Utilizing graph machine learning within drug",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                541,
+                                466,
+                                553
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        541,
+                                        466,
+                                        553
+                                    ],
+                                    "score": 1.0,
+                                    "content": "discovery and development. Briefings in Bioinformatics, 05 2021. ISSN 1477-4054.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 33,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                559,
+                                506,
+                                574
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        559,
+                                        506,
+                                        574
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[31] Justin Gilmer, Samuel S Schoenholz, Patrick F Riley, Oriol Vinyals, and George E Dahl. Neural",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 34,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                125,
+                                570,
+                                358,
+                                584
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        125,
+                                        570,
+                                        358,
+                                        584
+                                    ],
+                                    "score": 1.0,
+                                    "content": "message passing for quantum chemistry. In ICML, 2017.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 35,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                588,
+                                506,
+                                605
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        588,
+                                        506,
+                                        605
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[32] Nicholas Glaze, T. Mitchell Roddenberry, and Santiago Segarra. Principled simplicial neural",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 36,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                600,
+                                334,
+                                614
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        600,
+                                        334,
+                                        614
+                                    ],
+                                    "score": 1.0,
+                                    "content": "networks for trajectory prediction. In ICML, 2021.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 37,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                618,
+                                506,
+                                633
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        618,
+                                        506,
+                                        633
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[33] Rafael Gómez-Bombarelli, Jennifer N Wei, David Duvenaud, José Miguel Hernández-Lobato,",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 38,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                631,
+                                506,
+                                643
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        631,
+                                        506,
+                                        643
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Benjamín Sánchez-Lengeling, Dennis Sheberla, Jorge Aguilera-Iparraguirre, Timothy D Hirzel,",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                641,
+                                505,
+                                653
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        641,
+                                        505,
+                                        653
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Ryan P Adams, and Alán Aspuru-Guzik. Automatic chemical design using a data-driven",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                651,
+                                463,
+                                666
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        651,
+                                        463,
+                                        666
+                                    ],
+                                    "score": 1.0,
+                                    "content": "continuous representation of molecules. ACS Central Science, 4(2):268–276, 2018.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 41,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                669,
+                                506,
+                                683
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        669,
+                                        506,
+                                        683
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[34] Mustafa Hajij, Kyle Istvan, and Ghada Zamzmi. Cell complex neural networks. In NeurIPS",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 42,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                681,
+                                367,
+                                695
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        681,
+                                        367,
+                                        695
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Workshop on Topological Data Analysis and Beyond, 2020.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 43,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                699,
+                                508,
+                                715
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        699,
+                                        508,
+                                        715
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[35] Mustafa Hajij, Ghada Zamzmi, and Xuanting Cai. Simplicial complex representation learning.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 44,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                125,
+                                711,
+                                292,
+                                725
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        125,
+                                        711,
+                                        292,
+                                        725
+                                    ],
+                                    "score": 1.0,
+                                    "content": "arXiv preprint arXiv:2103.04046, 2021.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 45,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                72,
+                                506,
+                                87
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        72,
+                                        506,
+                                        87
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[36] J. Hansen and R. Ghrist. Toward a spectral theory of cellular sheaves. Journal of Applied and",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 0,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                83,
+                                303,
+                                96
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        83,
+                                        303,
+                                        96
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Computational Topology, 3:315–358, 2019.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 1,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                101,
+                                506,
+                                117
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        101,
+                                        506,
+                                        117
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[37] Jakob Hansen and Thomas Gebhart. Sheaf neural networks. In NeurIPS 2020 Workshop on",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 2,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                114,
+                                314,
+                                126
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        114,
+                                        314,
+                                        126
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Topological Data Analysis and Beyond, 2020.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 3,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                132,
+                                443,
+                                145
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        132,
+                                        443,
+                                        145
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[38] Allen Hatcher. Algebraic topology. Cambridge Univ. Press, Cambridge, 2000.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 4,
+                            "is_list_start_line": true,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                150,
+                                507,
+                                165
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        150,
+                                        507,
+                                        165
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[39] Christoph Helma, Ross D. King, Stefan Kramer, and Ashwin Srinivasan. The predictive",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 5,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                162,
+                                419,
+                                176
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        162,
+                                        419,
+                                        176
+                                    ],
+                                    "score": 1.0,
+                                    "content": "toxicology challenge 2000–2001. Bioinformatics, 17(1):107–108, 2001.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 6,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                180,
+                                506,
+                                195
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        180,
+                                        506,
+                                        195
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[40] Weihua Hu, Matthias Fey, Marinka Zitnik, Yuxiao Dong, Hongyu Ren, Bowen Liu, Michele",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 7,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                191,
+                                507,
+                                206
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        191,
+                                        507,
+                                        206
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Catasta, and Jure Leskovec. Open graph benchmark: Datasets for machine learning on graphs.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                203,
+                                203,
+                                215
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        203,
+                                        203,
+                                        215
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In NeurIPS, 2020.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 9,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                221,
+                                505,
+                                234
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        222,
+                                        160,
+                                        234
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[41] Weihua",
+                                    "type": "text",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        160,
+                                        221,
+                                        178,
+                                        232
+                                    ],
+                                    "score": 0.63,
+                                    "content": "\\mathrm { H u ^ { * } }",
+                                    "type": "inline_equation",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        179,
+                                        222,
+                                        505,
+                                        234
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", Bowen Liu*, Joseph Gomes, Marinka Zitnik, Percy Liang, Vijay Pande, and Jure",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 10,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                232,
+                                506,
+                                246
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        232,
+                                        506,
+                                        246
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Leskovec. Strategies for pre-training graph neural networks. In International Conference on",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                244,
+                                261,
+                                256
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        244,
+                                        261,
+                                        256
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Learning Representations, 2020.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 12,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                261,
+                                506,
+                                276
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        261,
+                                        506,
+                                        276
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[42] Sergey Ioffe and Christian Szegedy. Batch normalization: Accelerating deep network training",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 13,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                273,
+                                336,
+                                285
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        273,
+                                        336,
+                                        285
+                                    ],
+                                    "score": 1.0,
+                                    "content": "by reducing internal covariate shift. In ICML, 2015.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 14,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                292,
+                                505,
+                                304
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        292,
+                                        505,
+                                        304
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[43] Wengong Jin, Regina Barzilay, and Tommi Jaakkola. Junction tree variational autoencoder for",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 15,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                303,
+                                383,
+                                317
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        303,
+                                        383,
+                                        317
+                                    ],
+                                    "score": 1.0,
+                                    "content": "molecular graph generation. In ICML, pages 2323–2332, 2018.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 16,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                321,
+                                506,
+                                335
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        321,
+                                        506,
+                                        335
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[44] Wengong Jin, Kevin Yang, Regina Barzilay, and Tommi Jaakkola. Learning multimodal",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 17,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                333,
+                                411,
+                                345
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        333,
+                                        411,
+                                        345
+                                    ],
+                                    "score": 1.0,
+                                    "content": "graph-to-graph translation for molecular optimization. In ICLR, 2019.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 18,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                351,
+                                507,
+                                364
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        351,
+                                        507,
+                                        364
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[45] Jeroen Kazius, Ross McGuire, and Roberta Bursi. Derivation and validation of toxicophores for",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 19,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                363,
+                                451,
+                                375
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        363,
+                                        451,
+                                        375
+                                    ],
+                                    "score": 1.0,
+                                    "content": "mutagenicity prediction. Journal of Medicinal Chemistry, 48(1):312–320, 2005.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 20,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                381,
+                                507,
+                                395
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        381,
+                                        507,
+                                        395
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[46] Diederik P Kingma and Jimmy Ba. Adam: A method for stochastic optimization. In ICLR,",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 21,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                392,
+                                155,
+                                405
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        392,
+                                        155,
+                                        405
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2015.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 22,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                410,
+                                505,
+                                424
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        410,
+                                        505,
+                                        424
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[47] Thomas N Kipf and Max Welling. Semi-supervised classification with graph convolutional",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 23,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                422,
+                                233,
+                                434
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        422,
+                                        233,
+                                        434
+                                    ],
+                                    "score": 1.0,
+                                    "content": "networks. In ICLR, 2017.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 24,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                440,
+                                506,
+                                454
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        440,
+                                        506,
+                                        454
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[48] Nils Kriege and Petra Mutzel. Subgraph matching kernels for attributed graphs. arXiv preprint",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 25,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                452,
+                                226,
+                                464
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        452,
+                                        226,
+                                        464
+                                    ],
+                                    "score": 1.0,
+                                    "content": "arXiv:1206.6483, 2012.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 26,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                469,
+                                507,
+                                483
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        469,
+                                        507,
+                                        483
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[49] Casimir Kuratowski. Sur le probleme des courbes gauches en topologie. Fundamenta mathe-",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 27,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                125,
+                                480,
+                                255,
+                                494
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        125,
+                                        480,
+                                        255,
+                                        494
+                                    ],
+                                    "score": 1.0,
+                                    "content": "maticae, 15(1):271–283, 1930.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 28,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                499,
+                                414,
+                                514
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        499,
+                                        414,
+                                        514
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[50] Greg Landrum. RDKit: Open-source cheminformatics software. 2016.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 29,
+                            "is_list_start_line": true,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                518,
+                                506,
+                                532
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        518,
+                                        506,
+                                        532
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[51] Qimai Li, Zhichao Han, and Xiao-Ming Wu. Deeper insights into graph convolutional networks",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 30,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                530,
+                                357,
+                                542
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        530,
+                                        357,
+                                        542
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for semi-supervised learning. In AAAI, volume 32, 2018.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 31,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                548,
+                                506,
+                                562
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        548,
+                                        506,
+                                        562
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[52] Haggai Maron, Heli Ben-Hamu, Nadav Shamir, and Yaron Lipman. Invariant and equivariant",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 32,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                559,
+                                259,
+                                573
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        559,
+                                        259,
+                                        573
+                                    ],
+                                    "score": 1.0,
+                                    "content": "graph networks. In ICLR, 2018.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 33,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                577,
+                                505,
+                                591
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        577,
+                                        505,
+                                        591
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[53] Haggai Maron, Heli Ben-Hamu, Hadar Serviansky, and Yaron Lipman. Provably powerful",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 34,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                588,
+                                271,
+                                602
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        588,
+                                        271,
+                                        602
+                                    ],
+                                    "score": 1.0,
+                                    "content": "graph networks. In NeurIPS, 2019.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 35,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                607,
+                                506,
+                                622
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        607,
+                                        506,
+                                        622
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[54] Federico Monti, Davide Boscaini, Jonathan Masci, Emanuele Rodola, Jan Svoboda, and",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 36,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                618,
+                                505,
+                                633
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        618,
+                                        505,
+                                        633
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Michael M Bronstein. Geometric deep learning on graphs and manifolds using mixture model",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                629,
+                                301,
+                                643
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        629,
+                                        301,
+                                        643
+                                    ],
+                                    "score": 1.0,
+                                    "content": "CNNs. In CVPR, pages 5425–5434, 2017.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 38,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                648,
+                                506,
+                                662
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        648,
+                                        506,
+                                        662
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[55] Christopher Morris, Martin Ritzert, Matthias Fey, William L Hamilton, Jan Eric Lenssen,",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 39,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                658,
+                                506,
+                                673
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        658,
+                                        506,
+                                        673
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Gaurav Rattan, and Martin Grohe. Weisfeiler and Leman go neural: Higher-order graph neural",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                125,
+                                669,
+                                233,
+                                683
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        125,
+                                        669,
+                                        233,
+                                        683
+                                    ],
+                                    "score": 1.0,
+                                    "content": "networks. In AAAI, 2019.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 41,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                688,
+                                506,
+                                703
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        688,
+                                        506,
+                                        703
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[56] Christopher Morris, Nils M Kriege, Franka Bause, Kristian Kersting, Petra Mutzel, and Marion",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 42,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                700,
+                                506,
+                                714
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        700,
+                                        506,
+                                        714
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Neumann. TUDataset: A collection of benchmark datasets for learning with graphs. In ICML",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                710,
+                                408,
+                                724
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        710,
+                                        309,
+                                        724
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Graph Representation Learning and Beyond",
+                                    "type": "text",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        309,
+                                        711,
+                                        338,
+                                        721
+                                    ],
+                                    "score": 0.34,
+                                    "content": "{ ' } G R L { + } ,",
+                                    "type": "inline_equation",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        338,
+                                        710,
+                                        408,
+                                        724
+                                    ],
+                                    "score": 1.0,
+                                    "content": ") Workhop, 2020.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 44,
+                            "is_list_end_line": true
+                        }
+                    ],
+                    "index": 26.5,
+                    "bbox_fs": [
+                        104,
+                        195,
+                        507,
+                        723
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        104,
+                        30,
+                        507,
+                        727
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                70,
+                                507,
+                                88
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        70,
+                                        507,
+                                        88
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[16] N. Chiba and Takao Nishizeki. Arboricity and subgraph listing algorithms. SIAM J. Comput.,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                83,
+                                207,
+                                97
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        83,
+                                        207,
+                                        97
+                                    ],
+                                    "score": 1.0,
+                                    "content": "14:210–223, 1985.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                102,
+                                430,
+                                116
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        102,
+                                        430,
+                                        116
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[17] Fan Chung. Spectral graph theory. American Mathematical Society, 1997.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                121,
+                                506,
+                                135
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        121,
+                                        506,
+                                        135
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[18] Djork-Arné Clevert, Thomas Unterthiner, and Sepp Hochreiter. Fast and accurate deep network",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                132,
+                                370,
+                                146
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        132,
+                                        370,
+                                        146
+                                    ],
+                                    "score": 1.0,
+                                    "content": "learning by exponential linear units (ELUs). In ICLR, 2016.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                149,
+                                505,
+                                164
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        149,
+                                        505,
+                                        164
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[19] Luigi P. Cordella, Pasquale Foggia, Carlo Sansone, and Mario Vento. A (sub)graph isomorphism",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                161,
+                                506,
+                                175
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        161,
+                                        506,
+                                        175
+                                    ],
+                                    "score": 1.0,
+                                    "content": "algorithm for matching large graphs. IEEE Transactions on Pattern Analysis and Machine",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                171,
+                                285,
+                                186
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        171,
+                                        285,
+                                        186
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Intelligence, 26(10):1367–1372, 2004.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                190,
+                                506,
+                                204
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        190,
+                                        506,
+                                        204
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[20] Gabriele Corso, Luca Cavalleri, Dominique Beaini, Pietro Liò, and Petar Velickovi ˇ c. Principal ´",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                201,
+                                507,
+                                216
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        201,
+                                        507,
+                                        216
+                                    ],
+                                    "score": 1.0,
+                                    "content": "neighbourhood aggregation for graph nets. In NeurIPS, volume 33, pages 13260–13271, 2020.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                220,
+                                487,
+                                234
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        220,
+                                        487,
+                                        234
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[21] Pim de Haan, Taco Cohen, and Max Welling. Natural graph networks. In NeurIPS, 2020.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                238,
+                                506,
+                                253
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        238,
+                                        506,
+                                        253
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[22] Michaël Defferrard, Xavier Bresson, and Pierre Vandergheynst. Convolutional neural networks",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                250,
+                                379,
+                                263
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        250,
+                                        379,
+                                        263
+                                    ],
+                                    "score": 1.0,
+                                    "content": "on graphs with fast localized spectral filtering. In NIPS, 2016.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                268,
+                                506,
+                                283
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        268,
+                                        506,
+                                        283
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[23] Paul D Dobson and Andrew J Doig. Distinguishing enzyme structures from non-enzymes",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                280,
+                                427,
+                                292
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        280,
+                                        427,
+                                        292
+                                    ],
+                                    "score": 1.0,
+                                    "content": "without alignments. Journal of Molecular Biology, 330(4):771–783, 2003.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                297,
+                                507,
+                                313
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        297,
+                                        507,
+                                        313
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[24] Vijay Prakash Dwivedi, Chaitanya K Joshi, Thomas Laurent, Yoshua Bengio, and Xavier",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                309,
+                                485,
+                                324
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        309,
+                                        485,
+                                        324
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Bresson. Benchmarking graph neural networks. arXiv preprint arXiv:2003.00982, 2020.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                327,
+                                506,
+                                342
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        327,
+                                        506,
+                                        342
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[25] Stefania Ebli, Michaël Defferrard, and Gard Spreemann. Simplicial neural networks. In NeurIPS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                339,
+                                366,
+                                352
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        339,
+                                        366,
+                                        352
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Workshop on Topological Data Analysis and Beyond, 2020.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                357,
+                                506,
+                                371
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        357,
+                                        506,
+                                        371
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[26] Rui Ferreira, Roberto Grossi, Romeo Rizzi, Gustavo Sacomoto, and Marie-France Sagot.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                368,
+                                504,
+                                382
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        368,
+                                        485,
+                                        382
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Amortized Õ(|V|) -Delay Algorithm for Listing Chordless Cycles in Undirected Graphs. In",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        485,
+                                        369,
+                                        504,
+                                        379
+                                    ],
+                                    "score": 0.55,
+                                    "content": "2 2 t h",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                380,
+                                506,
+                                392
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        380,
+                                        506,
+                                        392
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Annual European Symposium on Algorithms, volume 8737 of Algorithms ESA 2014, Lecture",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                390,
+                                450,
+                                405
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        390,
+                                        450,
+                                        405
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Notes in Computer Science, pages 418–429, Wroclaw, Poland, September 2014.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                408,
+                                506,
+                                423
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        408,
+                                        506,
+                                        423
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[27] M. Fey, J. G. Yuen, and F. Weichert. Hierarchical inter-message passing for learning on",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                419,
+                                506,
+                                433
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        419,
+                                        430,
+                                        433
+                                    ],
+                                    "score": 1.0,
+                                    "content": "molecular graphs. In ICML Graph Representation Learning and Beyond",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        431,
+                                        420,
+                                        461,
+                                        431
+                                    ],
+                                    "score": 0.48,
+                                    "content": "( G R L + ,",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        462,
+                                        419,
+                                        506,
+                                        433
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Workhop,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                431,
+                                153,
+                                444
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        431,
+                                        153,
+                                        444
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2020.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                450,
+                                506,
+                                462
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        450,
+                                        506,
+                                        462
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[28] Matthias Fey and Jan Eric Lenssen. Fast graph representation learning with PyTorch Geometric.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                460,
+                                452,
+                                475
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        460,
+                                        452,
+                                        475
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In ICLR Workshop on Representation Learning on Graphs and Manifolds, 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                478,
+                                506,
+                                493
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        478,
+                                        506,
+                                        493
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[29] Thomas Gärtner, Peter Flach, and Stefan Wrobel. On graph kernels: Hardness results and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                490,
+                                505,
+                                502
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        490,
+                                        505,
+                                        502
+                                    ],
+                                    "score": 1.0,
+                                    "content": "efficient alternatives. In Learning theory and kernel machines, pages 129–143. Springer, 2003.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                507,
+                                506,
+                                522
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        507,
+                                        506,
+                                        522
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[30] Thomas Gaudelet, Ben Day, Arian R Jamasb, Jyothish Soman, Cristian Regep, Gertrude Liu,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                519,
+                                506,
+                                533
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        519,
+                                        506,
+                                        533
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Jeremy B R Hayter, Richard Vickers, Charles Roberts, Jian Tang, David Roblin, Tom L Blundell,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                528,
+                                506,
+                                546
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        528,
+                                        506,
+                                        546
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Michael M Bronstein, and Jake P Taylor-King. Utilizing graph machine learning within drug",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                541,
+                                466,
+                                553
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        541,
+                                        466,
+                                        553
+                                    ],
+                                    "score": 1.0,
+                                    "content": "discovery and development. Briefings in Bioinformatics, 05 2021. ISSN 1477-4054.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                559,
+                                506,
+                                574
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        559,
+                                        506,
+                                        574
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[31] Justin Gilmer, Samuel S Schoenholz, Patrick F Riley, Oriol Vinyals, and George E Dahl. Neural",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                125,
+                                570,
+                                358,
+                                584
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        125,
+                                        570,
+                                        358,
+                                        584
+                                    ],
+                                    "score": 1.0,
+                                    "content": "message passing for quantum chemistry. In ICML, 2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                588,
+                                506,
+                                605
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        588,
+                                        506,
+                                        605
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[32] Nicholas Glaze, T. Mitchell Roddenberry, and Santiago Segarra. Principled simplicial neural",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                600,
+                                334,
+                                614
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        600,
+                                        334,
+                                        614
+                                    ],
+                                    "score": 1.0,
+                                    "content": "networks for trajectory prediction. In ICML, 2021.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                618,
+                                506,
+                                633
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        618,
+                                        506,
+                                        633
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[33] Rafael Gómez-Bombarelli, Jennifer N Wei, David Duvenaud, José Miguel Hernández-Lobato,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                631,
+                                506,
+                                643
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        631,
+                                        506,
+                                        643
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Benjamín Sánchez-Lengeling, Dennis Sheberla, Jorge Aguilera-Iparraguirre, Timothy D Hirzel,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                641,
+                                505,
+                                653
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        641,
+                                        505,
+                                        653
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Ryan P Adams, and Alán Aspuru-Guzik. Automatic chemical design using a data-driven",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                651,
+                                463,
+                                666
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        651,
+                                        463,
+                                        666
+                                    ],
+                                    "score": 1.0,
+                                    "content": "continuous representation of molecules. ACS Central Science, 4(2):268–276, 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                669,
+                                506,
+                                683
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        669,
+                                        506,
+                                        683
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[34] Mustafa Hajij, Kyle Istvan, and Ghada Zamzmi. Cell complex neural networks. In NeurIPS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                681,
+                                367,
+                                695
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        681,
+                                        367,
+                                        695
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Workshop on Topological Data Analysis and Beyond, 2020.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                699,
+                                508,
+                                715
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        699,
+                                        508,
+                                        715
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[35] Mustafa Hajij, Ghada Zamzmi, and Xuanting Cai. Simplicial complex representation learning.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                125,
+                                711,
+                                292,
+                                725
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        125,
+                                        711,
+                                        292,
+                                        725
+                                    ],
+                                    "score": 1.0,
+                                    "content": "arXiv preprint arXiv:2103.04046, 2021.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        }
+                    ],
+                    "index": 22.5
+                }
+            ],
+            "page_idx": 11,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        300,
+                        741,
+                        311,
+                        750
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                299,
+                                740,
+                                313,
+                                754
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        299,
+                                        740,
+                                        313,
+                                        754
+                                    ],
+                                    "score": 1.0,
+                                    "content": "12",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "list",
+                    "bbox": [
+                        104,
+                        30,
+                        507,
+                        727
+                    ],
+                    "lines": [],
+                    "index": 22.5,
+                    "bbox_fs": [
+                        105,
+                        70,
+                        508,
+                        725
+                    ],
+                    "lines_deleted": true
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        104,
+                        45,
+                        507,
+                        730
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                72,
+                                506,
+                                87
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        72,
+                                        506,
+                                        87
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[36] J. Hansen and R. Ghrist. Toward a spectral theory of cellular sheaves. Journal of Applied and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                83,
+                                303,
+                                96
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        83,
+                                        303,
+                                        96
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Computational Topology, 3:315–358, 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                101,
+                                506,
+                                117
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        101,
+                                        506,
+                                        117
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[37] Jakob Hansen and Thomas Gebhart. Sheaf neural networks. In NeurIPS 2020 Workshop on",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                114,
+                                314,
+                                126
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        114,
+                                        314,
+                                        126
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Topological Data Analysis and Beyond, 2020.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                132,
+                                443,
+                                145
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        132,
+                                        443,
+                                        145
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[38] Allen Hatcher. Algebraic topology. Cambridge Univ. Press, Cambridge, 2000.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                150,
+                                507,
+                                165
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        150,
+                                        507,
+                                        165
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[39] Christoph Helma, Ross D. King, Stefan Kramer, and Ashwin Srinivasan. The predictive",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                162,
+                                419,
+                                176
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        162,
+                                        419,
+                                        176
+                                    ],
+                                    "score": 1.0,
+                                    "content": "toxicology challenge 2000–2001. Bioinformatics, 17(1):107–108, 2001.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                180,
+                                506,
+                                195
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        180,
+                                        506,
+                                        195
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[40] Weihua Hu, Matthias Fey, Marinka Zitnik, Yuxiao Dong, Hongyu Ren, Bowen Liu, Michele",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                191,
+                                507,
+                                206
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        191,
+                                        507,
+                                        206
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Catasta, and Jure Leskovec. Open graph benchmark: Datasets for machine learning on graphs.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                203,
+                                203,
+                                215
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        203,
+                                        203,
+                                        215
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In NeurIPS, 2020.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                221,
+                                505,
+                                234
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        222,
+                                        160,
+                                        234
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[41] Weihua",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        160,
+                                        221,
+                                        178,
+                                        232
+                                    ],
+                                    "score": 0.63,
+                                    "content": "\\mathrm { H u ^ { * } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        179,
+                                        222,
+                                        505,
+                                        234
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", Bowen Liu*, Joseph Gomes, Marinka Zitnik, Percy Liang, Vijay Pande, and Jure",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                232,
+                                506,
+                                246
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        232,
+                                        506,
+                                        246
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Leskovec. Strategies for pre-training graph neural networks. In International Conference on",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                244,
+                                261,
+                                256
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        244,
+                                        261,
+                                        256
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Learning Representations, 2020.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                261,
+                                506,
+                                276
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        261,
+                                        506,
+                                        276
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[42] Sergey Ioffe and Christian Szegedy. Batch normalization: Accelerating deep network training",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                273,
+                                336,
+                                285
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        273,
+                                        336,
+                                        285
+                                    ],
+                                    "score": 1.0,
+                                    "content": "by reducing internal covariate shift. In ICML, 2015.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                292,
+                                505,
+                                304
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        292,
+                                        505,
+                                        304
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[43] Wengong Jin, Regina Barzilay, and Tommi Jaakkola. Junction tree variational autoencoder for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                303,
+                                383,
+                                317
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        303,
+                                        383,
+                                        317
+                                    ],
+                                    "score": 1.0,
+                                    "content": "molecular graph generation. In ICML, pages 2323–2332, 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                321,
+                                506,
+                                335
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        321,
+                                        506,
+                                        335
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[44] Wengong Jin, Kevin Yang, Regina Barzilay, and Tommi Jaakkola. Learning multimodal",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                333,
+                                411,
+                                345
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        333,
+                                        411,
+                                        345
+                                    ],
+                                    "score": 1.0,
+                                    "content": "graph-to-graph translation for molecular optimization. In ICLR, 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                351,
+                                507,
+                                364
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        351,
+                                        507,
+                                        364
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[45] Jeroen Kazius, Ross McGuire, and Roberta Bursi. Derivation and validation of toxicophores for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                363,
+                                451,
+                                375
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        363,
+                                        451,
+                                        375
+                                    ],
+                                    "score": 1.0,
+                                    "content": "mutagenicity prediction. Journal of Medicinal Chemistry, 48(1):312–320, 2005.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                381,
+                                507,
+                                395
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        381,
+                                        507,
+                                        395
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[46] Diederik P Kingma and Jimmy Ba. Adam: A method for stochastic optimization. In ICLR,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                392,
+                                155,
+                                405
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        392,
+                                        155,
+                                        405
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2015.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                410,
+                                505,
+                                424
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        410,
+                                        505,
+                                        424
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[47] Thomas N Kipf and Max Welling. Semi-supervised classification with graph convolutional",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                422,
+                                233,
+                                434
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        422,
+                                        233,
+                                        434
+                                    ],
+                                    "score": 1.0,
+                                    "content": "networks. In ICLR, 2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                440,
+                                506,
+                                454
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        440,
+                                        506,
+                                        454
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[48] Nils Kriege and Petra Mutzel. Subgraph matching kernels for attributed graphs. arXiv preprint",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                452,
+                                226,
+                                464
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        452,
+                                        226,
+                                        464
+                                    ],
+                                    "score": 1.0,
+                                    "content": "arXiv:1206.6483, 2012.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                469,
+                                507,
+                                483
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        469,
+                                        507,
+                                        483
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[49] Casimir Kuratowski. Sur le probleme des courbes gauches en topologie. Fundamenta mathe-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                125,
+                                480,
+                                255,
+                                494
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        125,
+                                        480,
+                                        255,
+                                        494
+                                    ],
+                                    "score": 1.0,
+                                    "content": "maticae, 15(1):271–283, 1930.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                499,
+                                414,
+                                514
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        499,
+                                        414,
+                                        514
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[50] Greg Landrum. RDKit: Open-source cheminformatics software. 2016.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                518,
+                                506,
+                                532
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        518,
+                                        506,
+                                        532
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[51] Qimai Li, Zhichao Han, and Xiao-Ming Wu. Deeper insights into graph convolutional networks",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                530,
+                                357,
+                                542
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        530,
+                                        357,
+                                        542
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for semi-supervised learning. In AAAI, volume 32, 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                548,
+                                506,
+                                562
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        548,
+                                        506,
+                                        562
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[52] Haggai Maron, Heli Ben-Hamu, Nadav Shamir, and Yaron Lipman. Invariant and equivariant",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                559,
+                                259,
+                                573
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        559,
+                                        259,
+                                        573
+                                    ],
+                                    "score": 1.0,
+                                    "content": "graph networks. In ICLR, 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                577,
+                                505,
+                                591
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        577,
+                                        505,
+                                        591
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[53] Haggai Maron, Heli Ben-Hamu, Hadar Serviansky, and Yaron Lipman. Provably powerful",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                588,
+                                271,
+                                602
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        588,
+                                        271,
+                                        602
+                                    ],
+                                    "score": 1.0,
+                                    "content": "graph networks. In NeurIPS, 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                607,
+                                506,
+                                622
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        607,
+                                        506,
+                                        622
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[54] Federico Monti, Davide Boscaini, Jonathan Masci, Emanuele Rodola, Jan Svoboda, and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                618,
+                                505,
+                                633
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        618,
+                                        505,
+                                        633
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Michael M Bronstein. Geometric deep learning on graphs and manifolds using mixture model",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                629,
+                                301,
+                                643
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        629,
+                                        301,
+                                        643
+                                    ],
+                                    "score": 1.0,
+                                    "content": "CNNs. In CVPR, pages 5425–5434, 2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                648,
+                                506,
+                                662
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        648,
+                                        506,
+                                        662
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[55] Christopher Morris, Martin Ritzert, Matthias Fey, William L Hamilton, Jan Eric Lenssen,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                658,
+                                506,
+                                673
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        658,
+                                        506,
+                                        673
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Gaurav Rattan, and Martin Grohe. Weisfeiler and Leman go neural: Higher-order graph neural",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                125,
+                                669,
+                                233,
+                                683
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        125,
+                                        669,
+                                        233,
+                                        683
+                                    ],
+                                    "score": 1.0,
+                                    "content": "networks. In AAAI, 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                688,
+                                506,
+                                703
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        688,
+                                        506,
+                                        703
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[56] Christopher Morris, Nils M Kriege, Franka Bause, Kristian Kersting, Petra Mutzel, and Marion",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                700,
+                                506,
+                                714
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        700,
+                                        506,
+                                        714
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Neumann. TUDataset: A collection of benchmark datasets for learning with graphs. In ICML",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                710,
+                                408,
+                                724
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        710,
+                                        309,
+                                        724
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Graph Representation Learning and Beyond",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        309,
+                                        711,
+                                        338,
+                                        721
+                                    ],
+                                    "score": 0.34,
+                                    "content": "{ ' } G R L { + } ,",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        338,
+                                        710,
+                                        408,
+                                        724
+                                    ],
+                                    "score": 1.0,
+                                    "content": ") Workhop, 2020.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        }
+                    ],
+                    "index": 22
+                }
+            ],
+            "page_idx": 12,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        300,
+                        741,
+                        311,
+                        750
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                299,
+                                740,
+                                313,
+                                754
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        299,
+                                        740,
+                                        313,
+                                        754
+                                    ],
+                                    "score": 1.0,
+                                    "content": "13",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "list",
+                    "bbox": [
+                        104,
+                        45,
+                        507,
+                        730
+                    ],
+                    "lines": [],
+                    "index": 22,
+                    "bbox_fs": [
+                        104,
+                        72,
+                        507,
+                        724
+                    ],
+                    "lines_deleted": true
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        72,
+                        505,
+                        96
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                72,
+                                505,
+                                85
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        72,
+                                        505,
+                                        85
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[57] Ryan Murphy, Balasubramaniam Srinivasan, Vinayak Rao, and Bruno Ribeiro. Relational",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                83,
+                                454,
+                                96
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        83,
+                                        454,
+                                        96
+                                    ],
+                                    "score": 1.0,
+                                    "content": "pooling for graph representations. In ICML, volume 97, pages 4663–4673, 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        }
+                    ],
+                    "index": 0.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        102,
+                        504,
+                        136
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                102,
+                                505,
+                                115
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        102,
+                                        505,
+                                        115
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[58] Marion Neumann, Roman Garnett, Christian Bauckhage, and Kristian Kersting. Propagation",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                112,
+                                506,
+                                127
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        112,
+                                        506,
+                                        127
+                                    ],
+                                    "score": 1.0,
+                                    "content": "kernels: efficient graph kernels from propagated information. Machine Learning, 102(2):",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                123,
+                                193,
+                                136
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        123,
+                                        193,
+                                        136
+                                    ],
+                                    "score": 1.0,
+                                    "content": "209–245, 2016.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        }
+                    ],
+                    "index": 3
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        143,
+                        504,
+                        199
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                142,
+                                506,
+                                156
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        142,
+                                        506,
+                                        156
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[59] Adam Paszke, Sam Gross, Francisco Massa, Adam Lerer, James Bradbury, Gregory Chanan,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                153,
+                                505,
+                                167
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        153,
+                                        505,
+                                        167
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Trevor Killeen, Zeming Lin, Natalia Gimelshein, Luca Antiga, Alban Desmaison, Andreas",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                128,
+                                164,
+                                506,
+                                178
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        128,
+                                        164,
+                                        506,
+                                        178
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Kopf, Edward Yang, Zachary DeVito, Martin Raison, Alykhan Tejani, Sasank Chilamkurthy,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                175,
+                                506,
+                                189
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        175,
+                                        506,
+                                        189
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Benoit Steiner, Lu Fang, Junjie Bai, and Soumith Chintala. PyTorch: An imperative style,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                186,
+                                441,
+                                199
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        186,
+                                        441,
+                                        199
+                                    ],
+                                    "score": 1.0,
+                                    "content": "high-performance deep learning library. In NeurIPS, pages 8024–8035. 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        }
+                    ],
+                    "index": 7
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        205,
+                        504,
+                        239
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                205,
+                                506,
+                                218
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        205,
+                                        506,
+                                        218
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[60] Raghunathan Ramakrishnan, Pavlo O Dral, Matthias Rupp, and O Anatole Von Lilienfeld.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                217,
+                                506,
+                                229
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        217,
+                                        506,
+                                        229
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Quantum chemistry structures and properties of 134 kilo molecules. Scientific Data, 1(1):1–7,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                226,
+                                154,
+                                240
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        226,
+                                        154,
+                                        240
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2014.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        }
+                    ],
+                    "index": 11
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        246,
+                        506,
+                        291
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                246,
+                                506,
+                                259
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        246,
+                                        506,
+                                        259
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[61] Kaspar Riesen and Horst Bunke. Iam graph database repository for graph based pattern",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                258,
+                                506,
+                                270
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        258,
+                                        506,
+                                        270
+                                    ],
+                                    "score": 1.0,
+                                    "content": "recognition and machine learning. In Joint IAPR International Workshops on Statistical",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                268,
+                                506,
+                                282
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        268,
+                                        506,
+                                        282
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Techniques in Pattern Recognition (SPR) and Structural and Syntactic Pattern Recognition",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                279,
+                                290,
+                                292
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        279,
+                                        290,
+                                        292
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(SSPR), pages 287–297. Springer, 2008.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        }
+                    ],
+                    "index": 14.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        298,
+                        504,
+                        321
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                297,
+                                506,
+                                311
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        297,
+                                        506,
+                                        311
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[62] Ryoma Sato. A survey on the expressive power of graph neural networks. arXiv preprint",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                308,
+                                231,
+                                321
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        308,
+                                        231,
+                                        321
+                                    ],
+                                    "score": 1.0,
+                                    "content": "arXiv:2003.04078, 2020.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        }
+                    ],
+                    "index": 17.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        327,
+                        504,
+                        361
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                327,
+                                506,
+                                340
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        327,
+                                        506,
+                                        340
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[63] Michael T Schaub, Austin R Benson, Paul Horn, Gabor Lippner, and Ali Jadbabaie. Random",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                128,
+                                340,
+                                505,
+                                351
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        128,
+                                        340,
+                                        505,
+                                        351
+                                    ],
+                                    "score": 1.0,
+                                    "content": "walks on simplicial complexes and the normalized Hodge 1-Laplacian. SIAM Review, 62(2):",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                128,
+                                350,
+                                192,
+                                361
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        128,
+                                        350,
+                                        192,
+                                        361
+                                    ],
+                                    "score": 1.0,
+                                    "content": "353–391, 2020.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        }
+                    ],
+                    "index": 20
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        368,
+                        505,
+                        392
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                368,
+                                506,
+                                381
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        368,
+                                        506,
+                                        381
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[64] Nino Shervashidze, SVN Vishwanathan, Tobias Petri, Kurt Mehlhorn, and Karsten Borgwardt.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                379,
+                                482,
+                                393
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        379,
+                                        482,
+                                        393
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Efficient graphlet kernels for large graph comparison. In AISTAT, pages 488–495, 2009.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        }
+                    ],
+                    "index": 22.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        398,
+                        504,
+                        421
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                398,
+                                505,
+                                411
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        398,
+                                        505,
+                                        411
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[65] Nino Shervashidze, Pascal Schweitzer, Erik Jan van Leeuwen, Kurt Mehlhorn, and Karsten M",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                408,
+                                451,
+                                423
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        408,
+                                        451,
+                                        423
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Borgwardt. Weisfeiler-lehman graph kernels. JMLR, 12(Sep):2539–2561, 2011.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        }
+                    ],
+                    "index": 24.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        428,
+                        504,
+                        451
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                427,
+                                505,
+                                442
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        427,
+                                        505,
+                                        442
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[66] Howard E. Simmons and John E. Maggio. Synthesis of the first topologically non-planar",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                438,
+                                416,
+                                451
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        438,
+                                        416,
+                                        451
+                                    ],
+                                    "score": 1.0,
+                                    "content": "molecule. Tetrahedron Letters, 22(4):287–290, 1981. ISSN 0040-4039.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        }
+                    ],
+                    "index": 26.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        458,
+                        505,
+                        481
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                457,
+                                506,
+                                471
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        457,
+                                        506,
+                                        471
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[67] Amit Singer and Hau-Tieng Wu. Vector diffusion maps and the connection laplacian. Commu-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                469,
+                                357,
+                                481
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        469,
+                                        357,
+                                        481
+                                    ],
+                                    "score": 1.0,
+                                    "content": "nications in Pure and Applied Mathematics, 68(8), 2012.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        }
+                    ],
+                    "index": 28.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        487,
+                        505,
+                        522
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                487,
+                                506,
+                                501
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        487,
+                                        506,
+                                        501
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[68] Teague Sterling and John J. Irwin. ZINC 15 – ligand discovery for everyone. Journal of Chemical",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                499,
+                                506,
+                                511
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        499,
+                                        506,
+                                        511
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Information and Modeling, 55(11):2324–2337, 11 2015. doi: 10.1021/acs.jcim.5b00559. URL",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                510,
+                                341,
+                                522
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        510,
+                                        341,
+                                        522
+                                    ],
+                                    "score": 1.0,
+                                    "content": "https://doi.org/10.1021/acs.jcim.5b00559.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        }
+                    ],
+                    "index": 31
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        104,
+                        528,
+                        505,
+                        551
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                528,
+                                505,
+                                541
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        528,
+                                        505,
+                                        541
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[69] Petar Velickovi ˇ c, Guillem Cucurull, Arantxa Casanova, Adriana Romero, Pietro Lio, and Yoshua ´",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                540,
+                                333,
+                                552
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        540,
+                                        333,
+                                        552
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Bengio. Graph attention networks. In ICLR, 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        }
+                    ],
+                    "index": 33.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        558,
+                        504,
+                        591
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                558,
+                                506,
+                                571
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        558,
+                                        506,
+                                        571
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[70] Nikil Wale, Ian A Watson, and George Karypis. Comparison of descriptor spaces for chemical",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                570,
+                                506,
+                                582
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        570,
+                                        506,
+                                        582
+                                    ],
+                                    "score": 1.0,
+                                    "content": "compound retrieval and classification. Knowledge and Information Systems, 14(3):347–375,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                578,
+                                154,
+                                593
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        578,
+                                        154,
+                                        593
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2008.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        }
+                    ],
+                    "index": 36
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        104,
+                        599,
+                        505,
+                        622
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                599,
+                                505,
+                                612
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        599,
+                                        505,
+                                        612
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[71] Boris Weisfeiler and Andrei Leman. The reduction of a graph to canonical form and the algebra",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                128,
+                                610,
+                                342,
+                                622
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        128,
+                                        610,
+                                        342,
+                                        622
+                                    ],
+                                    "score": 1.0,
+                                    "content": "which appears therein. NTI Series, 2(9):12–16, 1968.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        }
+                    ],
+                    "index": 38.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        629,
+                        505,
+                        652
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                627,
+                                507,
+                                643
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        627,
+                                        507,
+                                        643
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[72] J. H. C. Whitehead. Combinatorial homotopy. I. Bulletin of the American Mathematical Society,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                640,
+                                333,
+                                651
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        640,
+                                        333,
+                                        651
+                                    ],
+                                    "score": 1.0,
+                                    "content": "55(3.P1):213 – 245, 1949. doi: bams/1183513543.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        }
+                    ],
+                    "index": 40.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        659,
+                        503,
+                        692
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                658,
+                                505,
+                                672
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        658,
+                                        505,
+                                        672
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[73] Keyulu Xu, Chengtao Li, Yonglong Tian, Tomohiro Sonobe, Ken ichi Kawarabayashi, and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                669,
+                                506,
+                                683
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        669,
+                                        506,
+                                        683
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Stefanie Jegelka. Representation learning on graphs with jumping knowledge networks. In",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                681,
+                                182,
+                                692
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        681,
+                                        182,
+                                        692
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ICML, 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        }
+                    ],
+                    "index": 43
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        699,
+                        502,
+                        722
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                699,
+                                504,
+                                712
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        699,
+                                        504,
+                                        712
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[74] Keyulu Xu, Weihua Hu, Jure Leskovec, and Stefanie Jegelka. How powerful are graph neural",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                711,
+                                235,
+                                721
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        711,
+                                        235,
+                                        721
+                                    ],
+                                    "score": 1.0,
+                                    "content": "networks? In ICLR, 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        }
+                    ],
+                    "index": 45.5
+                }
+            ],
+            "page_idx": 13,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        300,
+                        741,
+                        311,
+                        750
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                299,
+                                740,
+                                313,
+                                754
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        299,
+                                        740,
+                                        313,
+                                        754
+                                    ],
+                                    "score": 1.0,
+                                    "content": "14",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        72,
+                        505,
+                        96
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                72,
+                                505,
+                                85
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        72,
+                                        505,
+                                        85
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[57] Ryan Murphy, Balasubramaniam Srinivasan, Vinayak Rao, and Bruno Ribeiro. Relational",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                83,
+                                454,
+                                96
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        83,
+                                        454,
+                                        96
+                                    ],
+                                    "score": 1.0,
+                                    "content": "pooling for graph representations. In ICML, volume 97, pages 4663–4673, 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        }
+                    ],
+                    "index": 0.5,
+                    "bbox_fs": [
+                        106,
+                        72,
+                        505,
+                        96
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        102,
+                        504,
+                        136
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                102,
+                                505,
+                                115
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        102,
+                                        505,
+                                        115
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[58] Marion Neumann, Roman Garnett, Christian Bauckhage, and Kristian Kersting. Propagation",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                112,
+                                506,
+                                127
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        112,
+                                        506,
+                                        127
+                                    ],
+                                    "score": 1.0,
+                                    "content": "kernels: efficient graph kernels from propagated information. Machine Learning, 102(2):",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                123,
+                                193,
+                                136
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        123,
+                                        193,
+                                        136
+                                    ],
+                                    "score": 1.0,
+                                    "content": "209–245, 2016.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        }
+                    ],
+                    "index": 3,
+                    "bbox_fs": [
+                        106,
+                        102,
+                        506,
+                        136
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        143,
+                        504,
+                        199
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                142,
+                                506,
+                                156
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        142,
+                                        506,
+                                        156
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[59] Adam Paszke, Sam Gross, Francisco Massa, Adam Lerer, James Bradbury, Gregory Chanan,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                153,
+                                505,
+                                167
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        153,
+                                        505,
+                                        167
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Trevor Killeen, Zeming Lin, Natalia Gimelshein, Luca Antiga, Alban Desmaison, Andreas",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                128,
+                                164,
+                                506,
+                                178
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        128,
+                                        164,
+                                        506,
+                                        178
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Kopf, Edward Yang, Zachary DeVito, Martin Raison, Alykhan Tejani, Sasank Chilamkurthy,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                175,
+                                506,
+                                189
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        175,
+                                        506,
+                                        189
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Benoit Steiner, Lu Fang, Junjie Bai, and Soumith Chintala. PyTorch: An imperative style,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                186,
+                                441,
+                                199
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        186,
+                                        441,
+                                        199
+                                    ],
+                                    "score": 1.0,
+                                    "content": "high-performance deep learning library. In NeurIPS, pages 8024–8035. 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        }
+                    ],
+                    "index": 7,
+                    "bbox_fs": [
+                        106,
+                        142,
+                        506,
+                        199
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        205,
+                        504,
+                        239
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                205,
+                                506,
+                                218
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        205,
+                                        506,
+                                        218
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[60] Raghunathan Ramakrishnan, Pavlo O Dral, Matthias Rupp, and O Anatole Von Lilienfeld.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                217,
+                                506,
+                                229
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        217,
+                                        506,
+                                        229
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Quantum chemistry structures and properties of 134 kilo molecules. Scientific Data, 1(1):1–7,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                226,
+                                154,
+                                240
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        226,
+                                        154,
+                                        240
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2014.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        }
+                    ],
+                    "index": 11,
+                    "bbox_fs": [
+                        105,
+                        205,
+                        506,
+                        240
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        246,
+                        506,
+                        291
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                246,
+                                506,
+                                259
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        246,
+                                        506,
+                                        259
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[61] Kaspar Riesen and Horst Bunke. Iam graph database repository for graph based pattern",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                258,
+                                506,
+                                270
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        258,
+                                        506,
+                                        270
+                                    ],
+                                    "score": 1.0,
+                                    "content": "recognition and machine learning. In Joint IAPR International Workshops on Statistical",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                268,
+                                506,
+                                282
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        268,
+                                        506,
+                                        282
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Techniques in Pattern Recognition (SPR) and Structural and Syntactic Pattern Recognition",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                279,
+                                290,
+                                292
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        279,
+                                        290,
+                                        292
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(SSPR), pages 287–297. Springer, 2008.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        }
+                    ],
+                    "index": 14.5,
+                    "bbox_fs": [
+                        106,
+                        246,
+                        506,
+                        292
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        298,
+                        504,
+                        321
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                297,
+                                506,
+                                311
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        297,
+                                        506,
+                                        311
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[62] Ryoma Sato. A survey on the expressive power of graph neural networks. arXiv preprint",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                308,
+                                231,
+                                321
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        308,
+                                        231,
+                                        321
+                                    ],
+                                    "score": 1.0,
+                                    "content": "arXiv:2003.04078, 2020.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        }
+                    ],
+                    "index": 17.5,
+                    "bbox_fs": [
+                        106,
+                        297,
+                        506,
+                        321
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        327,
+                        504,
+                        361
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                327,
+                                506,
+                                340
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        327,
+                                        506,
+                                        340
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[63] Michael T Schaub, Austin R Benson, Paul Horn, Gabor Lippner, and Ali Jadbabaie. Random",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                128,
+                                340,
+                                505,
+                                351
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        128,
+                                        340,
+                                        505,
+                                        351
+                                    ],
+                                    "score": 1.0,
+                                    "content": "walks on simplicial complexes and the normalized Hodge 1-Laplacian. SIAM Review, 62(2):",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                128,
+                                350,
+                                192,
+                                361
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        128,
+                                        350,
+                                        192,
+                                        361
+                                    ],
+                                    "score": 1.0,
+                                    "content": "353–391, 2020.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        }
+                    ],
+                    "index": 20,
+                    "bbox_fs": [
+                        105,
+                        327,
+                        506,
+                        361
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        368,
+                        505,
+                        392
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                368,
+                                506,
+                                381
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        368,
+                                        506,
+                                        381
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[64] Nino Shervashidze, SVN Vishwanathan, Tobias Petri, Kurt Mehlhorn, and Karsten Borgwardt.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                379,
+                                482,
+                                393
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        379,
+                                        482,
+                                        393
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Efficient graphlet kernels for large graph comparison. In AISTAT, pages 488–495, 2009.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        }
+                    ],
+                    "index": 22.5,
+                    "bbox_fs": [
+                        106,
+                        368,
+                        506,
+                        393
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        398,
+                        504,
+                        421
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                398,
+                                505,
+                                411
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        398,
+                                        505,
+                                        411
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[65] Nino Shervashidze, Pascal Schweitzer, Erik Jan van Leeuwen, Kurt Mehlhorn, and Karsten M",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                408,
+                                451,
+                                423
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        408,
+                                        451,
+                                        423
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Borgwardt. Weisfeiler-lehman graph kernels. JMLR, 12(Sep):2539–2561, 2011.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        }
+                    ],
+                    "index": 24.5,
+                    "bbox_fs": [
+                        106,
+                        398,
+                        505,
+                        423
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        428,
+                        504,
+                        451
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                427,
+                                505,
+                                442
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        427,
+                                        505,
+                                        442
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[66] Howard E. Simmons and John E. Maggio. Synthesis of the first topologically non-planar",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                438,
+                                416,
+                                451
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        438,
+                                        416,
+                                        451
+                                    ],
+                                    "score": 1.0,
+                                    "content": "molecule. Tetrahedron Letters, 22(4):287–290, 1981. ISSN 0040-4039.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        }
+                    ],
+                    "index": 26.5,
+                    "bbox_fs": [
+                        105,
+                        427,
+                        505,
+                        451
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        458,
+                        505,
+                        481
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                457,
+                                506,
+                                471
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        457,
+                                        506,
+                                        471
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[67] Amit Singer and Hau-Tieng Wu. Vector diffusion maps and the connection laplacian. Commu-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                469,
+                                357,
+                                481
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        469,
+                                        357,
+                                        481
+                                    ],
+                                    "score": 1.0,
+                                    "content": "nications in Pure and Applied Mathematics, 68(8), 2012.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        }
+                    ],
+                    "index": 28.5,
+                    "bbox_fs": [
+                        105,
+                        457,
+                        506,
+                        481
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        487,
+                        505,
+                        522
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                487,
+                                506,
+                                501
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        487,
+                                        506,
+                                        501
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[68] Teague Sterling and John J. Irwin. ZINC 15 – ligand discovery for everyone. Journal of Chemical",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                499,
+                                506,
+                                511
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        499,
+                                        506,
+                                        511
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Information and Modeling, 55(11):2324–2337, 11 2015. doi: 10.1021/acs.jcim.5b00559. URL",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                510,
+                                341,
+                                522
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        510,
+                                        341,
+                                        522
+                                    ],
+                                    "score": 1.0,
+                                    "content": "https://doi.org/10.1021/acs.jcim.5b00559.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        }
+                    ],
+                    "index": 31,
+                    "bbox_fs": [
+                        105,
+                        487,
+                        506,
+                        522
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        104,
+                        528,
+                        505,
+                        551
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                528,
+                                505,
+                                541
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        528,
+                                        505,
+                                        541
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[69] Petar Velickovi ˇ c, Guillem Cucurull, Arantxa Casanova, Adriana Romero, Pietro Lio, and Yoshua ´",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                540,
+                                333,
+                                552
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        540,
+                                        333,
+                                        552
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Bengio. Graph attention networks. In ICLR, 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        }
+                    ],
+                    "index": 33.5,
+                    "bbox_fs": [
+                        106,
+                        528,
+                        505,
+                        552
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        558,
+                        504,
+                        591
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                558,
+                                506,
+                                571
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        558,
+                                        506,
+                                        571
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[70] Nikil Wale, Ian A Watson, and George Karypis. Comparison of descriptor spaces for chemical",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                570,
+                                506,
+                                582
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        570,
+                                        506,
+                                        582
+                                    ],
+                                    "score": 1.0,
+                                    "content": "compound retrieval and classification. Knowledge and Information Systems, 14(3):347–375,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                578,
+                                154,
+                                593
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        578,
+                                        154,
+                                        593
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2008.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        }
+                    ],
+                    "index": 36,
+                    "bbox_fs": [
+                        105,
+                        558,
+                        506,
+                        593
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        104,
+                        599,
+                        505,
+                        622
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                599,
+                                505,
+                                612
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        599,
+                                        505,
+                                        612
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[71] Boris Weisfeiler and Andrei Leman. The reduction of a graph to canonical form and the algebra",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                128,
+                                610,
+                                342,
+                                622
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        128,
+                                        610,
+                                        342,
+                                        622
+                                    ],
+                                    "score": 1.0,
+                                    "content": "which appears therein. NTI Series, 2(9):12–16, 1968.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        }
+                    ],
+                    "index": 38.5,
+                    "bbox_fs": [
+                        106,
+                        599,
+                        505,
+                        622
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        629,
+                        505,
+                        652
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                627,
+                                507,
+                                643
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        627,
+                                        507,
+                                        643
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[72] J. H. C. Whitehead. Combinatorial homotopy. I. Bulletin of the American Mathematical Society,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                640,
+                                333,
+                                651
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        640,
+                                        333,
+                                        651
+                                    ],
+                                    "score": 1.0,
+                                    "content": "55(3.P1):213 – 245, 1949. doi: bams/1183513543.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        }
+                    ],
+                    "index": 40.5,
+                    "bbox_fs": [
+                        105,
+                        627,
+                        507,
+                        651
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        659,
+                        503,
+                        692
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                658,
+                                505,
+                                672
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        658,
+                                        505,
+                                        672
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[73] Keyulu Xu, Chengtao Li, Yonglong Tian, Tomohiro Sonobe, Ken ichi Kawarabayashi, and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                669,
+                                506,
+                                683
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        669,
+                                        506,
+                                        683
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Stefanie Jegelka. Representation learning on graphs with jumping knowledge networks. In",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                681,
+                                182,
+                                692
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        681,
+                                        182,
+                                        692
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ICML, 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        }
+                    ],
+                    "index": 43,
+                    "bbox_fs": [
+                        105,
+                        658,
+                        506,
+                        692
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        699,
+                        502,
+                        722
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                699,
+                                504,
+                                712
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        699,
+                                        504,
+                                        712
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[74] Keyulu Xu, Weihua Hu, Jure Leskovec, and Stefanie Jegelka. How powerful are graph neural",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                711,
+                                235,
+                                721
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        711,
+                                        235,
+                                        721
+                                    ],
+                                    "score": 1.0,
+                                    "content": "networks? In ICLR, 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        }
+                    ],
+                    "index": 45.5,
+                    "bbox_fs": [
+                        106,
+                        699,
+                        504,
+                        721
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        104,
+                        72,
+                        506,
+                        125
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                73,
+                                505,
+                                86
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        73,
+                                        505,
+                                        86
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[75] Jiaxuan You, Bowen Liu, Rex Ying, Vijay Pande, and Jure Leskovec. Graph convolutional",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                83,
+                                439,
+                                96
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        83,
+                                        439,
+                                        96
+                                    ],
+                                    "score": 1.0,
+                                    "content": "policy network for goal-directed molecular graph generation. NeurIPS, 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                101,
+                                505,
+                                116
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        101,
+                                        505,
+                                        116
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[76] Muhan Zhang, Zhicheng Cui, Marion Neumann, and Yixin Chen. An end-to-end deep learning",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                114,
+                                337,
+                                126
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        114,
+                                        337,
+                                        126
+                                    ],
+                                    "score": 1.0,
+                                    "content": "architecture for graph classification. In AAAI, 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        }
+                    ],
+                    "index": 1.5
+                }
+            ],
+            "page_idx": 14,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        300,
+                        741,
+                        311,
+                        750
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                299,
+                                740,
+                                313,
+                                754
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        299,
+                                        740,
+                                        313,
+                                        754
+                                    ],
+                                    "score": 1.0,
+                                    "content": "15",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "list",
+                    "bbox": [
+                        104,
+                        72,
+                        506,
+                        125
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                73,
+                                505,
+                                86
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        73,
+                                        505,
+                                        86
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[75] Jiaxuan You, Bowen Liu, Rex Ying, Vijay Pande, and Jure Leskovec. Graph convolutional",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                83,
+                                439,
+                                96
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        83,
+                                        439,
+                                        96
+                                    ],
+                                    "score": 1.0,
+                                    "content": "policy network for goal-directed molecular graph generation. NeurIPS, 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                101,
+                                505,
+                                116
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        101,
+                                        505,
+                                        116
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[76] Muhan Zhang, Zhicheng Cui, Marion Neumann, and Yixin Chen. An end-to-end deep learning",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                114,
+                                337,
+                                126
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        114,
+                                        337,
+                                        126
+                                    ],
+                                    "score": 1.0,
+                                    "content": "architecture for graph classification. In AAAI, 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3,
+                            "is_list_end_line": true
+                        }
+                    ],
+                    "index": 1.5,
+                    "bbox_fs": [
+                        105,
+                        73,
+                        505,
+                        126
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        71,
+                        156,
+                        84
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                70,
+                                158,
+                                85
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        70,
+                                        158,
+                                        85
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Checklist",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        }
+                    ],
+                    "index": 0
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        131,
+                        92,
+                        208,
+                        104
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                129,
+                                91,
+                                210,
+                                105
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        129,
+                                        91,
+                                        210,
+                                        105
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1. For all authors...",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        }
+                    ],
+                    "index": 1
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        146,
+                        108,
+                        505,
+                        190
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                146,
+                                107,
+                                505,
+                                120
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        146,
+                                        107,
+                                        505,
+                                        120
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(a) Do the main claims made in the abstract and introduction accurately reflect the paper’s",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                161,
+                                118,
+                                288,
+                                131
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        161,
+                                        118,
+                                        288,
+                                        131
+                                    ],
+                                    "score": 1.0,
+                                    "content": "contributions and scope? [Yes]",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                144,
+                                130,
+                                433,
+                                144
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        144,
+                                        130,
+                                        433,
+                                        144
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(b) Did you describe the limitations of your work? [Yes] See Section 6.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                144,
+                                143,
+                                506,
+                                158
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        144,
+                                        143,
+                                        506,
+                                        158
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(c) Did you discuss any potential negative societal impacts of your work? [Yes] See",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                162,
+                                154,
+                                204,
+                                166
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        162,
+                                        154,
+                                        204,
+                                        166
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Section 6.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                144,
+                                166,
+                                506,
+                                182
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        144,
+                                        166,
+                                        506,
+                                        182
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(d) Have you read the ethics review guidelines and ensured that your paper conforms to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                161,
+                                178,
+                                214,
+                                191
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        161,
+                                        178,
+                                        214,
+                                        191
+                                    ],
+                                    "score": 1.0,
+                                    "content": "them? [Yes]",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        }
+                    ],
+                    "index": 5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        131,
+                        193,
+                        302,
+                        205
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                129,
+                                192,
+                                304,
+                                207
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        129,
+                                        192,
+                                        304,
+                                        207
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2. If you are including theoretical results...",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        }
+                    ],
+                    "index": 9
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        146,
+                        208,
+                        450,
+                        233
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                145,
+                                208,
+                                450,
+                                222
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        145,
+                                        208,
+                                        450,
+                                        222
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(a) Did you state the full set of assumptions of all theoretical results? [Yes]",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                144,
+                                221,
+                                421,
+                                234
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        144,
+                                        221,
+                                        421,
+                                        234
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(b) Did you include complete proofs of all theoretical results? [Yes]",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        }
+                    ],
+                    "index": 10.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        130,
+                        237,
+                        241,
+                        248
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                128,
+                                235,
+                                243,
+                                250
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        128,
+                                        235,
+                                        243,
+                                        250
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3. If you ran experiments...",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        }
+                    ],
+                    "index": 12
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        146,
+                        251,
+                        505,
+                        356
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                145,
+                                250,
+                                507,
+                                263
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        145,
+                                        250,
+                                        507,
+                                        263
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(a) Did you include the code, data, and instructions needed to reproduce the main experi-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                161,
+                                262,
+                                506,
+                                274
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        161,
+                                        262,
+                                        506,
+                                        274
+                                    ],
+                                    "score": 1.0,
+                                    "content": "mental results (either in the supplemental material or as a URL)? [Yes] We provided a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                161,
+                                273,
+                                279,
+                                285
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        161,
+                                        273,
+                                        279,
+                                        285
+                                    ],
+                                    "score": 1.0,
+                                    "content": "link to our code in the paper.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                145,
+                                285,
+                                505,
+                                299
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        145,
+                                        285,
+                                        505,
+                                        299
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(b) Did you specify all the training details (e.g., data splits, hyperparameters, how they",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                161,
+                                297,
+                                446,
+                                310
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        161,
+                                        297,
+                                        446,
+                                        310
+                                    ],
+                                    "score": 1.0,
+                                    "content": "were chosen)? [Yes] Details are provided in Section 5 and Appendix E.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                147,
+                                309,
+                                506,
+                                322
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        147,
+                                        309,
+                                        506,
+                                        322
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(c) Did you report error bars (e.g., with respect to the random seed after running experi-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                162,
+                                321,
+                                280,
+                                332
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        162,
+                                        321,
+                                        280,
+                                        332
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ments multiple times)? [Yes]",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                146,
+                                332,
+                                506,
+                                347
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        146,
+                                        332,
+                                        506,
+                                        347
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(d) Did you include the total amount of compute and the type of resources used (e.g., type",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                161,
+                                343,
+                                494,
+                                357
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        161,
+                                        343,
+                                        494,
+                                        357
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of GPUs, internal cluster, or cloud provider)? [Yes] See Section 5 and Appendix E.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        }
+                    ],
+                    "index": 17
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        133,
+                        359,
+                        503,
+                        371
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                129,
+                                357,
+                                505,
+                                374
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        129,
+                                        357,
+                                        505,
+                                        374
+                                    ],
+                                    "score": 1.0,
+                                    "content": "4. If you are using existing assets (e.g., code, data, models) or curating/releasing new assets...",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        }
+                    ],
+                    "index": 22
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        146,
+                        374,
+                        505,
+                        492
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                145,
+                                374,
+                                505,
+                                387
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        145,
+                                        374,
+                                        505,
+                                        387
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(a) If your work uses existing assets, did you cite the creators? [Yes] We used multiple",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                160,
+                                384,
+                                385,
+                                399
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        160,
+                                        384,
+                                        385,
+                                        399
+                                    ],
+                                    "score": 1.0,
+                                    "content": "existing datasets described and referenced in Section 5.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                144,
+                                397,
+                                425,
+                                411
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        144,
+                                        397,
+                                        425,
+                                        411
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(b) Did you mention the license of the assets? [Yes] See Appendix E.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                145,
+                                410,
+                                505,
+                                424
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        145,
+                                        410,
+                                        505,
+                                        424
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(c) Did you include any new assets either in the supplemental material or as a URL? [Yes]",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                161,
+                                421,
+                                289,
+                                433
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        161,
+                                        421,
+                                        289,
+                                        433
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We included a link to our code.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                146,
+                                435,
+                                505,
+                                446
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        146,
+                                        435,
+                                        505,
+                                        446
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(d) Did you discuss whether and how consent was obtained from people whose data you’re",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                161,
+                                446,
+                                437,
+                                459
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        161,
+                                        446,
+                                        437,
+                                        459
+                                    ],
+                                    "score": 1.0,
+                                    "content": "using/curating? [N/A] We are only using publicly available datasets.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                147,
+                                458,
+                                505,
+                                471
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        147,
+                                        458,
+                                        505,
+                                        471
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(e) Did you discuss whether the data you are using/curating contains personally identifiable",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                162,
+                                469,
+                                505,
+                                482
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        162,
+                                        469,
+                                        505,
+                                        482
+                                    ],
+                                    "score": 1.0,
+                                    "content": "information or offensive content? [N/A] None of the datasets we are using contain",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                161,
+                                480,
+                                387,
+                                493
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        161,
+                                        480,
+                                        387,
+                                        493
+                                    ],
+                                    "score": 1.0,
+                                    "content": "personally identifiable information or offensive content.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        }
+                    ],
+                    "index": 27.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        131,
+                        495,
+                        432,
+                        506
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                128,
+                                493,
+                                433,
+                                509
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        128,
+                                        493,
+                                        433,
+                                        509
+                                    ],
+                                    "score": 1.0,
+                                    "content": "5. If you used crowdsourcing or conducted research with human subjects...",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        }
+                    ],
+                    "index": 33
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        146,
+                        510,
+                        505,
+                        581
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                145,
+                                509,
+                                506,
+                                523
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        145,
+                                        509,
+                                        506,
+                                        523
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(a) Did you include the full text of instructions given to participants and screenshots, if",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                161,
+                                521,
+                                237,
+                                534
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        161,
+                                        521,
+                                        237,
+                                        534
+                                    ],
+                                    "score": 1.0,
+                                    "content": "applicable? [N/A]",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                145,
+                                533,
+                                505,
+                                546
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        145,
+                                        533,
+                                        505,
+                                        546
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(b) Did you describe any potential participant risks, with links to Institutional Review",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                162,
+                                544,
+                                342,
+                                557
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        162,
+                                        544,
+                                        342,
+                                        557
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Board (IRB) approvals, if applicable? [N/A]",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                146,
+                                557,
+                                505,
+                                570
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        146,
+                                        557,
+                                        505,
+                                        570
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(c) Did you include the estimated hourly wage paid to participants and the total amount",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                162,
+                                569,
+                                333,
+                                581
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        162,
+                                        569,
+                                        333,
+                                        581
+                                    ],
+                                    "score": 1.0,
+                                    "content": "spent on participant compensation? [N/A]",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        }
+                    ],
+                    "index": 36.5
+                }
+            ],
+            "page_idx": 15,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        300,
+                        741,
+                        311,
+                        750
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                299,
+                                740,
+                                313,
+                                754
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        299,
+                                        740,
+                                        313,
+                                        754
+                                    ],
+                                    "score": 1.0,
+                                    "content": "16",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        71,
+                        156,
+                        84
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                70,
+                                158,
+                                85
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        70,
+                                        158,
+                                        85
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Checklist",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        }
+                    ],
+                    "index": 0
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        131,
+                        92,
+                        208,
+                        104
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                129,
+                                91,
+                                210,
+                                105
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        129,
+                                        91,
+                                        210,
+                                        105
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1. For all authors...",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        }
+                    ],
+                    "index": 1,
+                    "bbox_fs": [
+                        129,
+                        91,
+                        210,
+                        105
+                    ]
+                },
+                {
+                    "type": "list",
+                    "bbox": [
+                        146,
+                        108,
+                        505,
+                        190
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                146,
+                                107,
+                                505,
+                                120
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        146,
+                                        107,
+                                        505,
+                                        120
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(a) Do the main claims made in the abstract and introduction accurately reflect the paper’s",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                161,
+                                118,
+                                288,
+                                131
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        161,
+                                        118,
+                                        288,
+                                        131
+                                    ],
+                                    "score": 1.0,
+                                    "content": "contributions and scope? [Yes]",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                144,
+                                130,
+                                433,
+                                144
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        144,
+                                        130,
+                                        433,
+                                        144
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(b) Did you describe the limitations of your work? [Yes] See Section 6.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4,
+                            "is_list_start_line": true,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                144,
+                                143,
+                                506,
+                                158
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        144,
+                                        143,
+                                        506,
+                                        158
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(c) Did you discuss any potential negative societal impacts of your work? [Yes] See",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                162,
+                                154,
+                                204,
+                                166
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        162,
+                                        154,
+                                        204,
+                                        166
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Section 6.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                144,
+                                166,
+                                506,
+                                182
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        144,
+                                        166,
+                                        506,
+                                        182
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(d) Have you read the ethics review guidelines and ensured that your paper conforms to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                161,
+                                178,
+                                214,
+                                191
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        161,
+                                        178,
+                                        214,
+                                        191
+                                    ],
+                                    "score": 1.0,
+                                    "content": "them? [Yes]",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8,
+                            "is_list_end_line": true
+                        }
+                    ],
+                    "index": 5,
+                    "bbox_fs": [
+                        144,
+                        107,
+                        506,
+                        191
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        131,
+                        193,
+                        302,
+                        205
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                129,
+                                192,
+                                304,
+                                207
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        129,
+                                        192,
+                                        304,
+                                        207
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2. If you are including theoretical results...",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        }
+                    ],
+                    "index": 9,
+                    "bbox_fs": [
+                        129,
+                        192,
+                        304,
+                        207
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        146,
+                        208,
+                        450,
+                        233
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                145,
+                                208,
+                                450,
+                                222
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        145,
+                                        208,
+                                        450,
+                                        222
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(a) Did you state the full set of assumptions of all theoretical results? [Yes]",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                144,
+                                221,
+                                421,
+                                234
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        144,
+                                        221,
+                                        421,
+                                        234
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(b) Did you include complete proofs of all theoretical results? [Yes]",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        }
+                    ],
+                    "index": 10.5,
+                    "bbox_fs": [
+                        144,
+                        208,
+                        450,
+                        234
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        130,
+                        237,
+                        241,
+                        248
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                128,
+                                235,
+                                243,
+                                250
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        128,
+                                        235,
+                                        243,
+                                        250
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3. If you ran experiments...",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        }
+                    ],
+                    "index": 12,
+                    "bbox_fs": [
+                        128,
+                        235,
+                        243,
+                        250
+                    ]
+                },
+                {
+                    "type": "list",
+                    "bbox": [
+                        146,
+                        251,
+                        505,
+                        356
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                145,
+                                250,
+                                507,
+                                263
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        145,
+                                        250,
+                                        507,
+                                        263
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(a) Did you include the code, data, and instructions needed to reproduce the main experi-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                161,
+                                262,
+                                506,
+                                274
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        161,
+                                        262,
+                                        506,
+                                        274
+                                    ],
+                                    "score": 1.0,
+                                    "content": "mental results (either in the supplemental material or as a URL)? [Yes] We provided a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                161,
+                                273,
+                                279,
+                                285
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        161,
+                                        273,
+                                        279,
+                                        285
+                                    ],
+                                    "score": 1.0,
+                                    "content": "link to our code in the paper.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                145,
+                                285,
+                                505,
+                                299
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        145,
+                                        285,
+                                        505,
+                                        299
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(b) Did you specify all the training details (e.g., data splits, hyperparameters, how they",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                161,
+                                297,
+                                446,
+                                310
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        161,
+                                        297,
+                                        446,
+                                        310
+                                    ],
+                                    "score": 1.0,
+                                    "content": "were chosen)? [Yes] Details are provided in Section 5 and Appendix E.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                147,
+                                309,
+                                506,
+                                322
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        147,
+                                        309,
+                                        506,
+                                        322
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(c) Did you report error bars (e.g., with respect to the random seed after running experi-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                162,
+                                321,
+                                280,
+                                332
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        162,
+                                        321,
+                                        280,
+                                        332
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ments multiple times)? [Yes]",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                146,
+                                332,
+                                506,
+                                347
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        146,
+                                        332,
+                                        506,
+                                        347
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(d) Did you include the total amount of compute and the type of resources used (e.g., type",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                161,
+                                343,
+                                494,
+                                357
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        161,
+                                        343,
+                                        494,
+                                        357
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of GPUs, internal cluster, or cloud provider)? [Yes] See Section 5 and Appendix E.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        }
+                    ],
+                    "index": 17,
+                    "bbox_fs": [
+                        145,
+                        250,
+                        507,
+                        357
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        133,
+                        359,
+                        503,
+                        371
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                129,
+                                357,
+                                505,
+                                374
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        129,
+                                        357,
+                                        505,
+                                        374
+                                    ],
+                                    "score": 1.0,
+                                    "content": "4. If you are using existing assets (e.g., code, data, models) or curating/releasing new assets...",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        }
+                    ],
+                    "index": 22,
+                    "bbox_fs": [
+                        129,
+                        357,
+                        505,
+                        374
+                    ]
+                },
+                {
+                    "type": "list",
+                    "bbox": [
+                        146,
+                        374,
+                        505,
+                        492
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                145,
+                                374,
+                                505,
+                                387
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        145,
+                                        374,
+                                        505,
+                                        387
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(a) If your work uses existing assets, did you cite the creators? [Yes] We used multiple",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                160,
+                                384,
+                                385,
+                                399
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        160,
+                                        384,
+                                        385,
+                                        399
+                                    ],
+                                    "score": 1.0,
+                                    "content": "existing datasets described and referenced in Section 5.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                144,
+                                397,
+                                425,
+                                411
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        144,
+                                        397,
+                                        425,
+                                        411
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(b) Did you mention the license of the assets? [Yes] See Appendix E.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25,
+                            "is_list_start_line": true,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                145,
+                                410,
+                                505,
+                                424
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        145,
+                                        410,
+                                        505,
+                                        424
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(c) Did you include any new assets either in the supplemental material or as a URL? [Yes]",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                161,
+                                421,
+                                289,
+                                433
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        161,
+                                        421,
+                                        289,
+                                        433
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We included a link to our code.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                146,
+                                435,
+                                505,
+                                446
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        146,
+                                        435,
+                                        505,
+                                        446
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(d) Did you discuss whether and how consent was obtained from people whose data you’re",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                161,
+                                446,
+                                437,
+                                459
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        161,
+                                        446,
+                                        437,
+                                        459
+                                    ],
+                                    "score": 1.0,
+                                    "content": "using/curating? [N/A] We are only using publicly available datasets.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                147,
+                                458,
+                                505,
+                                471
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        147,
+                                        458,
+                                        505,
+                                        471
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(e) Did you discuss whether the data you are using/curating contains personally identifiable",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                162,
+                                469,
+                                505,
+                                482
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        162,
+                                        469,
+                                        505,
+                                        482
+                                    ],
+                                    "score": 1.0,
+                                    "content": "information or offensive content? [N/A] None of the datasets we are using contain",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                161,
+                                480,
+                                387,
+                                493
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        161,
+                                        480,
+                                        387,
+                                        493
+                                    ],
+                                    "score": 1.0,
+                                    "content": "personally identifiable information or offensive content.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32,
+                            "is_list_end_line": true
+                        }
+                    ],
+                    "index": 27.5,
+                    "bbox_fs": [
+                        144,
+                        374,
+                        505,
+                        493
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        131,
+                        495,
+                        432,
+                        506
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                128,
+                                493,
+                                433,
+                                509
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        128,
+                                        493,
+                                        433,
+                                        509
+                                    ],
+                                    "score": 1.0,
+                                    "content": "5. If you used crowdsourcing or conducted research with human subjects...",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        }
+                    ],
+                    "index": 33,
+                    "bbox_fs": [
+                        128,
+                        493,
+                        433,
+                        509
+                    ]
+                },
+                {
+                    "type": "list",
+                    "bbox": [
+                        146,
+                        510,
+                        505,
+                        581
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                145,
+                                509,
+                                506,
+                                523
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        145,
+                                        509,
+                                        506,
+                                        523
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(a) Did you include the full text of instructions given to participants and screenshots, if",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                161,
+                                521,
+                                237,
+                                534
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        161,
+                                        521,
+                                        237,
+                                        534
+                                    ],
+                                    "score": 1.0,
+                                    "content": "applicable? [N/A]",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                145,
+                                533,
+                                505,
+                                546
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        145,
+                                        533,
+                                        505,
+                                        546
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(b) Did you describe any potential participant risks, with links to Institutional Review",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                162,
+                                544,
+                                342,
+                                557
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        162,
+                                        544,
+                                        342,
+                                        557
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Board (IRB) approvals, if applicable? [N/A]",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                146,
+                                557,
+                                505,
+                                570
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        146,
+                                        557,
+                                        505,
+                                        570
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(c) Did you include the estimated hourly wage paid to participants and the total amount",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                162,
+                                569,
+                                333,
+                                581
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        162,
+                                        569,
+                                        333,
+                                        581
+                                    ],
+                                    "score": 1.0,
+                                    "content": "spent on participant compensation? [N/A]",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39,
+                            "is_list_end_line": true
+                        }
+                    ],
+                    "index": 36.5,
+                    "bbox_fs": [
+                        145,
+                        509,
+                        506,
+                        581
+                    ]
+                }
+            ]
+        }
+    ],
+    "_backend": "pipeline",
+    "_version_name": "2.2.2"
+}
\ No newline at end of file
diff --git a/parse/train/uVPZCMVtsSG/uVPZCMVtsSG_model.json b/parse/train/uVPZCMVtsSG/uVPZCMVtsSG_model.json
new file mode 100644
index 0000000000000000000000000000000000000000..a7613b2f3955c9913be1b933c3b058bb00ac92e7
--- /dev/null
+++ b/parse/train/uVPZCMVtsSG/uVPZCMVtsSG_model.json
@@ -0,0 +1,21782 @@
+[
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    397,
+                    954,
+                    1304,
+                    954,
+                    1304,
+                    1499,
+                    397,
+                    1499
+                ],
+                "score": 0.984
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1625,
+                    1404,
+                    1625,
+                    1404,
+                    1931,
+                    298,
+                    1931
+                ],
+                "score": 0.978
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    863,
+                    435,
+                    1265,
+                    435,
+                    1265,
+                    526,
+                    863,
+                    526
+                ],
+                "score": 0.946
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    893,
+                    574,
+                    1251,
+                    574,
+                    1251,
+                    667,
+                    893,
+                    667
+                ],
+                "score": 0.941
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    434,
+                    434,
+                    717,
+                    434,
+                    717,
+                    526,
+                    434,
+                    526
+                ],
+                "score": 0.939
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    982,
+                    713,
+                    1383,
+                    713,
+                    1383,
+                    804,
+                    982,
+                    804
+                ],
+                "score": 0.933
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    316,
+                    713,
+                    597,
+                    713,
+                    597,
+                    805,
+                    316,
+                    805
+                ],
+                "score": 0.93
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    625,
+                    713,
+                    953,
+                    713,
+                    953,
+                    803,
+                    625,
+                    803
+                ],
+                "score": 0.928
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    318,
+                    270,
+                    1377,
+                    270,
+                    1377,
+                    326,
+                    318,
+                    326
+                ],
+                "score": 0.916
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    299,
+                    1556,
+                    530,
+                    1556,
+                    530,
+                    1595,
+                    299,
+                    1595
+                ],
+                "score": 0.905
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    788,
+                    884,
+                    912,
+                    884,
+                    912,
+                    920,
+                    788,
+                    920
+                ],
+                "score": 0.894
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    297,
+                    2033,
+                    1071,
+                    2033,
+                    1071,
+                    2062,
+                    297,
+                    2062
+                ],
+                "score": 0.889
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    332,
+                    1953,
+                    628,
+                    1953,
+                    628,
+                    1980,
+                    332,
+                    1980
+                ],
+                "score": 0.889
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    449,
+                    575,
+                    731,
+                    575,
+                    731,
+                    665,
+                    449,
+                    665
+                ],
+                "score": 0.863
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    525,
+                    575,
+                    656,
+                    575,
+                    656,
+                    603,
+                    525,
+                    603
+                ],
+                "score": 0.231
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    844,
+                    1686,
+                    894,
+                    1686,
+                    894,
+                    1719,
+                    844,
+                    1719
+                ],
+                "score": 0.6,
+                "latex": "\\dot { \\lVert \\mathcal { I } \\dot { 4 } \\rVert }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1321,
+                    1716,
+                    1371,
+                    1716,
+                    1371,
+                    1749,
+                    1321,
+                    1749
+                ],
+                "score": 0.48,
+                "latex": "\\pmb { \\mathbb { Z } } \\mathbf { \\mathbb { 1 } }"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    319.0,
+                    269.0,
+                    1380.0,
+                    269.0,
+                    1380.0,
+                    331.0,
+                    319.0,
+                    331.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1553.0,
+                    536.0,
+                    1553.0,
+                    536.0,
+                    1601.0,
+                    292.0,
+                    1601.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    782.0,
+                    881.0,
+                    919.0,
+                    881.0,
+                    919.0,
+                    925.0,
+                    782.0,
+                    925.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    2028.0,
+                    1073.0,
+                    2028.0,
+                    1073.0,
+                    2066.0,
+                    294.0,
+                    2066.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    330.0,
+                    1946.0,
+                    633.0,
+                    1946.0,
+                    633.0,
+                    1986.0,
+                    330.0,
+                    1986.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    520.0,
+                    573.0,
+                    662.0,
+                    573.0,
+                    662.0,
+                    609.0,
+                    520.0,
+                    609.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    395.0,
+                    952.0,
+                    1305.0,
+                    952.0,
+                    1305.0,
+                    990.0,
+                    395.0,
+                    990.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    393.0,
+                    984.0,
+                    1308.0,
+                    984.0,
+                    1308.0,
+                    1020.0,
+                    393.0,
+                    1020.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    392.0,
+                    1012.0,
+                    1307.0,
+                    1012.0,
+                    1307.0,
+                    1050.0,
+                    392.0,
+                    1050.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    394.0,
+                    1044.0,
+                    1306.0,
+                    1044.0,
+                    1306.0,
+                    1082.0,
+                    394.0,
+                    1082.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    393.0,
+                    1076.0,
+                    1305.0,
+                    1076.0,
+                    1305.0,
+                    1111.0,
+                    393.0,
+                    1111.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    392.0,
+                    1107.0,
+                    1306.0,
+                    1107.0,
+                    1306.0,
+                    1139.0,
+                    392.0,
+                    1139.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    393.0,
+                    1136.0,
+                    1305.0,
+                    1136.0,
+                    1305.0,
+                    1169.0,
+                    393.0,
+                    1169.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    394.0,
+                    1167.0,
+                    1304.0,
+                    1167.0,
+                    1304.0,
+                    1200.0,
+                    394.0,
+                    1200.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    393.0,
+                    1196.0,
+                    1306.0,
+                    1196.0,
+                    1306.0,
+                    1232.0,
+                    393.0,
+                    1232.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    393.0,
+                    1226.0,
+                    1309.0,
+                    1226.0,
+                    1309.0,
+                    1263.0,
+                    393.0,
+                    1263.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    392.0,
+                    1255.0,
+                    1306.0,
+                    1255.0,
+                    1306.0,
+                    1293.0,
+                    392.0,
+                    1293.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    393.0,
+                    1286.0,
+                    1304.0,
+                    1286.0,
+                    1304.0,
+                    1322.0,
+                    393.0,
+                    1322.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    392.0,
+                    1318.0,
+                    1308.0,
+                    1318.0,
+                    1308.0,
+                    1353.0,
+                    392.0,
+                    1353.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    393.0,
+                    1347.0,
+                    1308.0,
+                    1347.0,
+                    1308.0,
+                    1384.0,
+                    393.0,
+                    1384.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    394.0,
+                    1380.0,
+                    1304.0,
+                    1380.0,
+                    1304.0,
+                    1413.0,
+                    394.0,
+                    1413.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    393.0,
+                    1409.0,
+                    1308.0,
+                    1409.0,
+                    1308.0,
+                    1445.0,
+                    393.0,
+                    1445.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    394.0,
+                    1441.0,
+                    1303.0,
+                    1441.0,
+                    1303.0,
+                    1470.0,
+                    394.0,
+                    1470.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    393.0,
+                    1468.0,
+                    1295.0,
+                    1468.0,
+                    1295.0,
+                    1503.0,
+                    393.0,
+                    1503.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1626.0,
+                    1405.0,
+                    1626.0,
+                    1405.0,
+                    1661.0,
+                    294.0,
+                    1661.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1656.0,
+                    1406.0,
+                    1656.0,
+                    1406.0,
+                    1690.0,
+                    293.0,
+                    1690.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1688.0,
+                    843.0,
+                    1688.0,
+                    843.0,
+                    1720.0,
+                    296.0,
+                    1720.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    895.0,
+                    1688.0,
+                    1402.0,
+                    1688.0,
+                    1402.0,
+                    1720.0,
+                    895.0,
+                    1720.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1716.0,
+                    1320.0,
+                    1716.0,
+                    1320.0,
+                    1753.0,
+                    292.0,
+                    1753.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1372.0,
+                    1716.0,
+                    1402.0,
+                    1716.0,
+                    1402.0,
+                    1753.0,
+                    1372.0,
+                    1753.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1746.0,
+                    1405.0,
+                    1746.0,
+                    1405.0,
+                    1782.0,
+                    294.0,
+                    1782.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1775.0,
+                    1405.0,
+                    1775.0,
+                    1405.0,
+                    1815.0,
+                    293.0,
+                    1815.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1807.0,
+                    1406.0,
+                    1807.0,
+                    1406.0,
+                    1843.0,
+                    294.0,
+                    1843.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1839.0,
+                    1405.0,
+                    1839.0,
+                    1405.0,
+                    1874.0,
+                    294.0,
+                    1874.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1871.0,
+                    1399.0,
+                    1871.0,
+                    1399.0,
+                    1902.0,
+                    296.0,
+                    1902.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    1900.0,
+                    867.0,
+                    1900.0,
+                    867.0,
+                    1932.0,
+                    297.0,
+                    1932.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    961.0,
+                    435.0,
+                    1166.0,
+                    435.0,
+                    1166.0,
+                    466.0,
+                    961.0,
+                    466.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    865.0,
+                    467.0,
+                    1266.0,
+                    467.0,
+                    1266.0,
+                    498.0,
+                    865.0,
+                    498.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    923.0,
+                    496.0,
+                    1207.0,
+                    496.0,
+                    1207.0,
+                    527.0,
+                    923.0,
+                    527.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    970.0,
+                    571.0,
+                    1179.0,
+                    571.0,
+                    1179.0,
+                    611.0,
+                    970.0,
+                    611.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    919.0,
+                    604.0,
+                    1224.0,
+                    604.0,
+                    1224.0,
+                    637.0,
+                    919.0,
+                    637.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    891.0,
+                    637.0,
+                    1253.0,
+                    637.0,
+                    1253.0,
+                    671.0,
+                    891.0,
+                    671.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    477.0,
+                    434.0,
+                    686.0,
+                    434.0,
+                    686.0,
+                    466.0,
+                    477.0,
+                    466.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    433.0,
+                    463.0,
+                    719.0,
+                    463.0,
+                    719.0,
+                    500.0,
+                    433.0,
+                    500.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    458.0,
+                    497.0,
+                    696.0,
+                    497.0,
+                    696.0,
+                    526.0,
+                    458.0,
+                    526.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1070.0,
+                    712.0,
+                    1295.0,
+                    712.0,
+                    1295.0,
+                    746.0,
+                    1070.0,
+                    746.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    984.0,
+                    745.0,
+                    1385.0,
+                    745.0,
+                    1385.0,
+                    776.0,
+                    984.0,
+                    776.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1017.0,
+                    772.0,
+                    1347.0,
+                    772.0,
+                    1347.0,
+                    807.0,
+                    1017.0,
+                    807.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    393.0,
+                    711.0,
+                    521.0,
+                    711.0,
+                    521.0,
+                    744.0,
+                    393.0,
+                    744.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    315.0,
+                    741.0,
+                    600.0,
+                    741.0,
+                    600.0,
+                    778.0,
+                    315.0,
+                    778.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    341.0,
+                    774.0,
+                    571.0,
+                    774.0,
+                    571.0,
+                    806.0,
+                    341.0,
+                    806.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    689.0,
+                    712.0,
+                    895.0,
+                    712.0,
+                    895.0,
+                    744.0,
+                    689.0,
+                    744.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    678.0,
+                    743.0,
+                    901.0,
+                    743.0,
+                    901.0,
+                    772.0,
+                    678.0,
+                    772.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    625.0,
+                    775.0,
+                    955.0,
+                    775.0,
+                    955.0,
+                    805.0,
+                    625.0,
+                    805.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    521.0,
+                    571.0,
+                    662.0,
+                    571.0,
+                    662.0,
+                    606.0,
+                    521.0,
+                    606.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    547.0,
+                    603.0,
+                    633.0,
+                    603.0,
+                    633.0,
+                    633.0,
+                    547.0,
+                    633.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    450.0,
+                    637.0,
+                    732.0,
+                    637.0,
+                    732.0,
+                    665.0,
+                    450.0,
+                    665.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 0,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    203,
+                    1405,
+                    203,
+                    1405,
+                    538,
+                    298,
+                    538
+                ],
+                "score": 0.984
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    568,
+                    1405,
+                    568,
+                    1405,
+                    967,
+                    297,
+                    967
+                ],
+                "score": 0.981
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1348,
+                    991,
+                    1348,
+                    991,
+                    1536,
+                    298,
+                    1536
+                ],
+                "score": 0.976
+            },
+            {
+                "category_id": 3,
+                "poly": [
+                    1014,
+                    1344,
+                    1401,
+                    1344,
+                    1401,
+                    1677,
+                    1014,
+                    1677
+                ],
+                "score": 0.965
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    1551,
+                    993,
+                    1551,
+                    993,
+                    1832,
+                    299,
+                    1832
+                ],
+                "score": 0.963
+            },
+            {
+                "category_id": 4,
+                "poly": [
+                    1010,
+                    1690,
+                    1404,
+                    1690,
+                    1404,
+                    1815,
+                    1010,
+                    1815
+                ],
+                "score": 0.959
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    297,
+                    1949,
+                    1406,
+                    1949,
+                    1406,
+                    2008,
+                    297,
+                    2008
+                ],
+                "score": 0.931
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    1071,
+                    1399,
+                    1071,
+                    1399,
+                    1137,
+                    299,
+                    1137
+                ],
+                "score": 0.917
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    298,
+                    1004,
+                    527,
+                    1004,
+                    527,
+                    1043,
+                    298,
+                    1043
+                ],
+                "score": 0.913
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    1867,
+                    1403,
+                    1867,
+                    1403,
+                    1930,
+                    299,
+                    1930
+                ],
+                "score": 0.897
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    310,
+                    1823,
+                    1396,
+                    1823,
+                    1396,
+                    1855,
+                    310,
+                    1855
+                ],
+                "score": 0.877
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    840,
+                    2061,
+                    859,
+                    2061,
+                    859,
+                    2085,
+                    840,
+                    2085
+                ],
+                "score": 0.75
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    302,
+                    1265,
+                    1398,
+                    1265,
+                    1398,
+                    1328,
+                    302,
+                    1328
+                ],
+                "score": 0.731
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1147,
+                    1341,
+                    1147,
+                    1341,
+                    1180,
+                    297,
+                    1180
+                ],
+                "score": 0.654
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1184,
+                    1303,
+                    1184,
+                    1303,
+                    1222,
+                    298,
+                    1222
+                ],
+                "score": 0.549
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    304,
+                    1226,
+                    846,
+                    1226,
+                    846,
+                    1259,
+                    304,
+                    1259
+                ],
+                "score": 0.482
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    576,
+                    1102,
+                    703,
+                    1102,
+                    703,
+                    1137,
+                    576,
+                    1137
+                ],
+                "score": 0.93,
+                "latex": "\\{ X _ { \\sigma } \\} _ { \\sigma \\in P _ { X } }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    611,
+                    1183,
+                    771,
+                    1183,
+                    771,
+                    1220,
+                    611,
+                    1220
+                ],
+                "score": 0.93,
+                "latex": "{ \\cal X } _ { \\tau } \\cap \\overline { { { \\cal X } _ { \\sigma } } } \\ne \\emptyset"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    403,
+                    1381,
+                    629,
+                    1381,
+                    629,
+                    1415,
+                    403,
+                    1415
+                ],
+                "score": 0.92,
+                "latex": "\\tau \\leq \\sigma \\Leftrightarrow X _ { \\tau } \\subseteq \\overline { { X _ { \\sigma } } }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    803,
+                    1184,
+                    916,
+                    1184,
+                    916,
+                    1219,
+                    803,
+                    1219
+                ],
+                "score": 0.91,
+                "latex": "X _ { \\tau } \\subseteq { \\overline { { X _ { \\sigma } } } }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    588,
+                    1267,
+                    680,
+                    1267,
+                    680,
+                    1297,
+                    588,
+                    1297
+                ],
+                "score": 0.91,
+                "latex": "\\sigma \\in P _ { X }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    857,
+                    1104,
+                    897,
+                    1104,
+                    897,
+                    1134,
+                    857,
+                    1134
+                ],
+                "score": 0.9,
+                "latex": "X _ { \\sigma }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    437,
+                    1148,
+                    514,
+                    1148,
+                    514,
+                    1176,
+                    437,
+                    1176
+                ],
+                "score": 0.9,
+                "latex": "x \\in X"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1001,
+                    1184,
+                    1043,
+                    1184,
+                    1043,
+                    1218,
+                    1001,
+                    1218
+                ],
+                "score": 0.89,
+                "latex": "\\overline { { X _ { \\sigma } } }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    832,
+                    1478,
+                    906,
+                    1478,
+                    906,
+                    1503,
+                    832,
+                    1503
+                ],
+                "score": 0.89,
+                "latex": "\\tau < \\sigma"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1218,
+                    1267,
+                    1267,
+                    1267,
+                    1267,
+                    1295,
+                    1218,
+                    1295
+                ],
+                "score": 0.89,
+                "latex": "\\mathbb { R } ^ { n _ { \\sigma } }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1169,
+                    1298,
+                    1209,
+                    1298,
+                    1209,
+                    1327,
+                    1169,
+                    1327
+                ],
+                "score": 0.89,
+                "latex": "X _ { \\sigma }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    817,
+                    1351,
+                    858,
+                    1351,
+                    858,
+                    1381,
+                    817,
+                    1381
+                ],
+                "score": 0.88,
+                "latex": "P _ { X }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1302,
+                    1264,
+                    1342,
+                    1264,
+                    1342,
+                    1297,
+                    1302,
+                    1297
+                ],
+                "score": 0.87,
+                "latex": "\\overline { { X _ { \\sigma } } }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    677,
+                    1227,
+                    714,
+                    1227,
+                    714,
+                    1256,
+                    677,
+                    1256
+                ],
+                "score": 0.86,
+                "latex": "\\mathbb { R } ^ { n }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    414,
+                    1191,
+                    460,
+                    1191,
+                    460,
+                    1219,
+                    414,
+                    1219
+                ],
+                "score": 0.84,
+                "latex": "\\sigma , \\tau"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    649,
+                    1476,
+                    677,
+                    1476,
+                    677,
+                    1502,
+                    649,
+                    1502
+                ],
+                "score": 0.83,
+                "latex": "X"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    573,
+                    1299,
+                    593,
+                    1299,
+                    593,
+                    1328,
+                    573,
+                    1328
+                ],
+                "score": 0.82,
+                "latex": "\\phi"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    987,
+                    1268,
+                    1008,
+                    1268,
+                    1008,
+                    1298,
+                    987,
+                    1298
+                ],
+                "score": 0.81,
+                "latex": "\\phi"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    931,
+                    1104,
+                    958,
+                    1104,
+                    958,
+                    1130,
+                    931,
+                    1130
+                ],
+                "score": 0.79,
+                "latex": "X"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    904,
+                    1445,
+                    932,
+                    1445,
+                    932,
+                    1472,
+                    904,
+                    1472
+                ],
+                "score": 0.79,
+                "latex": "X"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    725,
+                    1828,
+                    744,
+                    1828,
+                    744,
+                    1850,
+                    725,
+                    1850
+                ],
+                "score": 0.79,
+                "latex": "n"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    366,
+                    1445,
+                    393,
+                    1445,
+                    393,
+                    1472,
+                    366,
+                    1472
+                ],
+                "score": 0.79,
+                "latex": "X"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    663,
+                    1071,
+                    712,
+                    1071,
+                    712,
+                    1103,
+                    663,
+                    1103
+                ],
+                "score": 0.78,
+                "latex": "\\begin{array} { r l } {  { \\bigl \\| \\widehat { 3 6 } \\bigr \\| } } & { { } } \\end{array}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    689,
+                    1761,
+                    718,
+                    1761,
+                    718,
+                    1797,
+                    689,
+                    1797
+                ],
+                "score": 0.77,
+                "latex": "2 ."
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1288,
+                    1692,
+                    1316,
+                    1692,
+                    1316,
+                    1718,
+                    1288,
+                    1718
+                ],
+                "score": 0.76,
+                "latex": "X"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    666,
+                    1867,
+                    694,
+                    1867,
+                    694,
+                    1904,
+                    666,
+                    1904
+                ],
+                "score": 0.76,
+                "latex": "\\textcircled { 3 }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    759,
+                    1977,
+                    802,
+                    1977,
+                    802,
+                    2006,
+                    759,
+                    2006
+                ],
+                "score": 0.75,
+                "latex": "\\pmb { \\mathbb { Z } 2 } \\|"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1064,
+                    1821,
+                    1151,
+                    1821,
+                    1151,
+                    1856,
+                    1064,
+                    1856
+                ],
+                "score": 0.72,
+                "latex": "( n - 1 )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1373,
+                    1078,
+                    1400,
+                    1078,
+                    1400,
+                    1100,
+                    1373,
+                    1100
+                ],
+                "score": 0.66,
+                "latex": "X"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    822,
+                    1234,
+                    841,
+                    1234,
+                    841,
+                    1254,
+                    822,
+                    1254
+                ],
+                "score": 0.41,
+                "latex": "n"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1093,
+                    1071,
+                    1123,
+                    1071,
+                    1123,
+                    1107,
+                    1093,
+                    1107
+                ],
+                "score": 0.4,
+                "latex": "\\mathbb { D }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    948,
+                    1153,
+                    967,
+                    1153,
+                    967,
+                    1175,
+                    948,
+                    1175
+                ],
+                "score": 0.25,
+                "latex": "x"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1094.0,
+                    1340.0,
+                    1394.0,
+                    1340.0,
+                    1394.0,
+                    1685.0,
+                    1094.0,
+                    1685.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1063.0,
+                    1468.0,
+                    1071.0,
+                    1468.0,
+                    1071.0,
+                    1478.0,
+                    1063.0,
+                    1478.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1071.0,
+                    1556.0,
+                    1104.0,
+                    1556.0,
+                    1104.0,
+                    1583.0,
+                    1071.0,
+                    1583.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1119.0,
+                    1473.5,
+                    1134.0,
+                    1473.5,
+                    1134.0,
+                    1505.5,
+                    1119.0,
+                    1505.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1011.0,
+                    1690.0,
+                    1287.0,
+                    1690.0,
+                    1287.0,
+                    1722.0,
+                    1011.0,
+                    1722.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1317.0,
+                    1690.0,
+                    1404.0,
+                    1690.0,
+                    1404.0,
+                    1722.0,
+                    1317.0,
+                    1722.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1010.0,
+                    1721.0,
+                    1405.0,
+                    1721.0,
+                    1405.0,
+                    1752.0,
+                    1010.0,
+                    1752.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1009.0,
+                    1751.0,
+                    1408.0,
+                    1751.0,
+                    1408.0,
+                    1781.0,
+                    1009.0,
+                    1781.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1009.0,
+                    1781.0,
+                    1393.0,
+                    1781.0,
+                    1393.0,
+                    1814.0,
+                    1009.0,
+                    1814.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    329.0,
+                    1944.0,
+                    1408.0,
+                    1944.0,
+                    1408.0,
+                    1982.0,
+                    329.0,
+                    1982.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1974.0,
+                    758.0,
+                    1974.0,
+                    758.0,
+                    2009.0,
+                    292.0,
+                    2009.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    803.0,
+                    1974.0,
+                    811.0,
+                    1974.0,
+                    811.0,
+                    2009.0,
+                    803.0,
+                    2009.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    289.0,
+                    999.0,
+                    530.0,
+                    999.0,
+                    530.0,
+                    1051.0,
+                    289.0,
+                    1051.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    838.0,
+                    2058.0,
+                    863.0,
+                    2058.0,
+                    863.0,
+                    2093.0,
+                    838.0,
+                    2093.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    200.0,
+                    1407.0,
+                    200.0,
+                    1407.0,
+                    241.0,
+                    292.0,
+                    241.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    232.0,
+                    1406.0,
+                    232.0,
+                    1406.0,
+                    270.0,
+                    293.0,
+                    270.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    263.0,
+                    1403.0,
+                    263.0,
+                    1403.0,
+                    299.0,
+                    295.0,
+                    299.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    294.0,
+                    1408.0,
+                    294.0,
+                    1408.0,
+                    330.0,
+                    292.0,
+                    330.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    325.0,
+                    1405.0,
+                    325.0,
+                    1405.0,
+                    359.0,
+                    295.0,
+                    359.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    351.0,
+                    1408.0,
+                    351.0,
+                    1408.0,
+                    391.0,
+                    291.0,
+                    391.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    385.0,
+                    1405.0,
+                    385.0,
+                    1405.0,
+                    420.0,
+                    295.0,
+                    420.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    413.0,
+                    1406.0,
+                    413.0,
+                    1406.0,
+                    452.0,
+                    292.0,
+                    452.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    446.0,
+                    1405.0,
+                    446.0,
+                    1405.0,
+                    480.0,
+                    295.0,
+                    480.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    474.0,
+                    1405.0,
+                    474.0,
+                    1405.0,
+                    509.0,
+                    295.0,
+                    509.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    508.0,
+                    1276.0,
+                    508.0,
+                    1276.0,
+                    542.0,
+                    295.0,
+                    542.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    566.0,
+                    1406.0,
+                    566.0,
+                    1406.0,
+                    605.0,
+                    292.0,
+                    605.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    596.0,
+                    1407.0,
+                    596.0,
+                    1407.0,
+                    636.0,
+                    294.0,
+                    636.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    628.0,
+                    1405.0,
+                    628.0,
+                    1405.0,
+                    665.0,
+                    295.0,
+                    665.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    660.0,
+                    1403.0,
+                    660.0,
+                    1403.0,
+                    692.0,
+                    296.0,
+                    692.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    688.0,
+                    1405.0,
+                    688.0,
+                    1405.0,
+                    725.0,
+                    294.0,
+                    725.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    721.0,
+                    1403.0,
+                    721.0,
+                    1403.0,
+                    753.0,
+                    296.0,
+                    753.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    751.0,
+                    1406.0,
+                    751.0,
+                    1406.0,
+                    784.0,
+                    292.0,
+                    784.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    778.0,
+                    1407.0,
+                    778.0,
+                    1407.0,
+                    816.0,
+                    292.0,
+                    816.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    810.0,
+                    1406.0,
+                    810.0,
+                    1406.0,
+                    847.0,
+                    294.0,
+                    847.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    840.0,
+                    1405.0,
+                    840.0,
+                    1405.0,
+                    877.0,
+                    295.0,
+                    877.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    870.0,
+                    1405.0,
+                    870.0,
+                    1405.0,
+                    905.0,
+                    292.0,
+                    905.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    900.0,
+                    1408.0,
+                    900.0,
+                    1408.0,
+                    938.0,
+                    294.0,
+                    938.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    933.0,
+                    1140.0,
+                    933.0,
+                    1140.0,
+                    966.0,
+                    296.0,
+                    966.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1346.0,
+                    816.0,
+                    1346.0,
+                    816.0,
+                    1385.0,
+                    296.0,
+                    1385.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    859.0,
+                    1346.0,
+                    993.0,
+                    1346.0,
+                    993.0,
+                    1385.0,
+                    859.0,
+                    1385.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1382.0,
+                    402.0,
+                    1382.0,
+                    402.0,
+                    1418.0,
+                    295.0,
+                    1418.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    630.0,
+                    1382.0,
+                    991.0,
+                    1382.0,
+                    991.0,
+                    1418.0,
+                    630.0,
+                    1418.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1413.0,
+                    991.0,
+                    1413.0,
+                    991.0,
+                    1447.0,
+                    296.0,
+                    1447.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1444.0,
+                    365.0,
+                    1444.0,
+                    365.0,
+                    1478.0,
+                    296.0,
+                    1478.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    394.0,
+                    1444.0,
+                    903.0,
+                    1444.0,
+                    903.0,
+                    1478.0,
+                    394.0,
+                    1478.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    933.0,
+                    1444.0,
+                    990.0,
+                    1444.0,
+                    990.0,
+                    1478.0,
+                    933.0,
+                    1478.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1475.0,
+                    648.0,
+                    1475.0,
+                    648.0,
+                    1506.0,
+                    296.0,
+                    1506.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    678.0,
+                    1475.0,
+                    831.0,
+                    1475.0,
+                    831.0,
+                    1506.0,
+                    678.0,
+                    1506.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    907.0,
+                    1475.0,
+                    990.0,
+                    1475.0,
+                    990.0,
+                    1506.0,
+                    907.0,
+                    1506.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1506.0,
+                    672.0,
+                    1506.0,
+                    672.0,
+                    1537.0,
+                    296.0,
+                    1537.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    1552.0,
+                    990.0,
+                    1552.0,
+                    990.0,
+                    1582.0,
+                    297.0,
+                    1582.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1580.0,
+                    995.0,
+                    1580.0,
+                    995.0,
+                    1614.0,
+                    295.0,
+                    1614.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1610.0,
+                    995.0,
+                    1610.0,
+                    995.0,
+                    1645.0,
+                    295.0,
+                    1645.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1642.0,
+                    992.0,
+                    1642.0,
+                    992.0,
+                    1677.0,
+                    296.0,
+                    1677.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1672.0,
+                    992.0,
+                    1672.0,
+                    992.0,
+                    1705.0,
+                    296.0,
+                    1705.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1701.0,
+                    992.0,
+                    1701.0,
+                    992.0,
+                    1735.0,
+                    295.0,
+                    1735.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1733.0,
+                    993.0,
+                    1733.0,
+                    993.0,
+                    1767.0,
+                    295.0,
+                    1767.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1762.0,
+                    688.0,
+                    1762.0,
+                    688.0,
+                    1796.0,
+                    295.0,
+                    1796.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    719.0,
+                    1762.0,
+                    991.0,
+                    1762.0,
+                    991.0,
+                    1796.0,
+                    719.0,
+                    1796.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1794.0,
+                    993.0,
+                    1794.0,
+                    993.0,
+                    1824.0,
+                    295.0,
+                    1824.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1069.0,
+                    662.0,
+                    1069.0,
+                    662.0,
+                    1109.0,
+                    294.0,
+                    1109.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    713.0,
+                    1069.0,
+                    1092.0,
+                    1069.0,
+                    1092.0,
+                    1109.0,
+                    713.0,
+                    1109.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1124.0,
+                    1069.0,
+                    1372.0,
+                    1069.0,
+                    1372.0,
+                    1109.0,
+                    1124.0,
+                    1109.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1099.0,
+                    575.0,
+                    1099.0,
+                    575.0,
+                    1139.0,
+                    294.0,
+                    1139.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    704.0,
+                    1099.0,
+                    856.0,
+                    1099.0,
+                    856.0,
+                    1139.0,
+                    704.0,
+                    1139.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    898.0,
+                    1099.0,
+                    930.0,
+                    1099.0,
+                    930.0,
+                    1139.0,
+                    898.0,
+                    1139.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    959.0,
+                    1099.0,
+                    1258.0,
+                    1099.0,
+                    1258.0,
+                    1139.0,
+                    959.0,
+                    1139.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1865.0,
+                    665.0,
+                    1865.0,
+                    665.0,
+                    1905.0,
+                    294.0,
+                    1905.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    695.0,
+                    1865.0,
+                    1407.0,
+                    1865.0,
+                    1407.0,
+                    1905.0,
+                    695.0,
+                    1905.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1899.0,
+                    1404.0,
+                    1899.0,
+                    1404.0,
+                    1934.0,
+                    295.0,
+                    1934.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    302.0,
+                    1818.0,
+                    724.0,
+                    1818.0,
+                    724.0,
+                    1860.0,
+                    302.0,
+                    1860.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    745.0,
+                    1818.0,
+                    1063.0,
+                    1818.0,
+                    1063.0,
+                    1860.0,
+                    745.0,
+                    1860.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1152.0,
+                    1818.0,
+                    1400.0,
+                    1818.0,
+                    1400.0,
+                    1860.0,
+                    1152.0,
+                    1860.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1262.0,
+                    587.0,
+                    1262.0,
+                    587.0,
+                    1300.0,
+                    293.0,
+                    1300.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    681.0,
+                    1262.0,
+                    986.0,
+                    1262.0,
+                    986.0,
+                    1300.0,
+                    681.0,
+                    1300.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1009.0,
+                    1262.0,
+                    1217.0,
+                    1262.0,
+                    1217.0,
+                    1300.0,
+                    1009.0,
+                    1300.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1268.0,
+                    1262.0,
+                    1301.0,
+                    1262.0,
+                    1301.0,
+                    1300.0,
+                    1268.0,
+                    1300.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1343.0,
+                    1262.0,
+                    1404.0,
+                    1262.0,
+                    1404.0,
+                    1300.0,
+                    1343.0,
+                    1300.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    329.0,
+                    1295.0,
+                    572.0,
+                    1295.0,
+                    572.0,
+                    1331.0,
+                    329.0,
+                    1331.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    594.0,
+                    1295.0,
+                    1168.0,
+                    1295.0,
+                    1168.0,
+                    1331.0,
+                    594.0,
+                    1331.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1210.0,
+                    1295.0,
+                    1219.0,
+                    1295.0,
+                    1219.0,
+                    1331.0,
+                    1210.0,
+                    1331.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1140.0,
+                    436.0,
+                    1140.0,
+                    436.0,
+                    1186.0,
+                    294.0,
+                    1186.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    515.0,
+                    1140.0,
+                    947.0,
+                    1140.0,
+                    947.0,
+                    1186.0,
+                    515.0,
+                    1186.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    968.0,
+                    1140.0,
+                    1343.0,
+                    1140.0,
+                    1343.0,
+                    1186.0,
+                    968.0,
+                    1186.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1180.0,
+                    413.0,
+                    1180.0,
+                    413.0,
+                    1226.0,
+                    291.0,
+                    1226.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    461.0,
+                    1180.0,
+                    610.0,
+                    1180.0,
+                    610.0,
+                    1226.0,
+                    461.0,
+                    1226.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    772.0,
+                    1180.0,
+                    802.0,
+                    1180.0,
+                    802.0,
+                    1226.0,
+                    772.0,
+                    1226.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    917.0,
+                    1180.0,
+                    1000.0,
+                    1180.0,
+                    1000.0,
+                    1226.0,
+                    917.0,
+                    1226.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1044.0,
+                    1180.0,
+                    1304.0,
+                    1180.0,
+                    1304.0,
+                    1226.0,
+                    1044.0,
+                    1226.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1221.0,
+                    676.0,
+                    1221.0,
+                    676.0,
+                    1264.0,
+                    296.0,
+                    1264.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    715.0,
+                    1221.0,
+                    821.0,
+                    1221.0,
+                    821.0,
+                    1264.0,
+                    715.0,
+                    1264.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    842.0,
+                    1221.0,
+                    852.0,
+                    1221.0,
+                    852.0,
+                    1264.0,
+                    842.0,
+                    1264.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 1,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1607,
+                    1405,
+                    1607,
+                    1405,
+                    1733,
+                    297,
+                    1733
+                ],
+                "score": 0.975
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    478,
+                    1406,
+                    478,
+                    1406,
+                    602,
+                    297,
+                    602
+                ],
+                "score": 0.971
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    783,
+                    1405,
+                    783,
+                    1405,
+                    909,
+                    297,
+                    909
+                ],
+                "score": 0.97
+            },
+            {
+                "category_id": 4,
+                "poly": [
+                    295,
+                    349,
+                    998,
+                    349,
+                    998,
+                    442,
+                    295,
+                    442
+                ],
+                "score": 0.961
+            },
+            {
+                "category_id": 3,
+                "poly": [
+                    298,
+                    195,
+                    996,
+                    195,
+                    996,
+                    336,
+                    298,
+                    336
+                ],
+                "score": 0.961
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1007,
+                    1406,
+                    1007,
+                    1406,
+                    1101,
+                    297,
+                    1101
+                ],
+                "score": 0.958
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    704,
+                    1400,
+                    704,
+                    1400,
+                    770,
+                    298,
+                    770
+                ],
+                "score": 0.949
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    294,
+                    613,
+                    1402,
+                    613,
+                    1402,
+                    677,
+                    294,
+                    677
+                ],
+                "score": 0.943
+            },
+            {
+                "category_id": 4,
+                "poly": [
+                    1026,
+                    350,
+                    1397,
+                    350,
+                    1397,
+                    442,
+                    1026,
+                    442
+                ],
+                "score": 0.943
+            },
+            {
+                "category_id": 3,
+                "poly": [
+                    1059,
+                    199,
+                    1362,
+                    199,
+                    1362,
+                    335,
+                    1059,
+                    335
+                ],
+                "score": 0.935
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    293,
+                    917,
+                    1402,
+                    917,
+                    1402,
+                    982,
+                    293,
+                    982
+                ],
+                "score": 0.935
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1742,
+                    1404,
+                    1742,
+                    1404,
+                    1836,
+                    297,
+                    1836
+                ],
+                "score": 0.934
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    298,
+                    1535,
+                    744,
+                    1535,
+                    744,
+                    1574,
+                    298,
+                    1574
+                ],
+                "score": 0.919
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    295,
+                    1154,
+                    1409,
+                    1154,
+                    1409,
+                    1491,
+                    295,
+                    1491
+                ],
+                "score": 0.898
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    294,
+                    1847,
+                    1272,
+                    1847,
+                    1272,
+                    2004,
+                    294,
+                    2004
+                ],
+                "score": 0.844
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    295,
+                    1111,
+                    1348,
+                    1111,
+                    1348,
+                    1146,
+                    295,
+                    1146
+                ],
+                "score": 0.708
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    841,
+                    2061,
+                    859,
+                    2061,
+                    859,
+                    2085,
+                    841,
+                    2085
+                ],
+                "score": 0.625
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    841,
+                    2061,
+                    860,
+                    2061,
+                    860,
+                    2085,
+                    841,
+                    2085
+                ],
+                "score": 0.569
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    655,
+                    1156,
+                    886,
+                    1156,
+                    886,
+                    1191,
+                    655,
+                    1191
+                ],
+                "score": 0.94,
+                "latex": "\\boldsymbol { B } ( \\sigma ) = \\{ \\tau | \\tau \\prec \\sigma \\}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1123,
+                    1113,
+                    1214,
+                    1113,
+                    1214,
+                    1143,
+                    1123,
+                    1143
+                ],
+                "score": 0.93,
+                "latex": "\\sigma \\in P _ { X }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    552,
+                    1774,
+                    822,
+                    1774,
+                    822,
+                    1807,
+                    552,
+                    1807
+                ],
+                "score": 0.92,
+                "latex": "B ( \\sigma , \\tau ) : = B ( \\sigma ) \\cap B ( \\tau )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    873,
+                    1774,
+                    1133,
+                    1774,
+                    1133,
+                    1807,
+                    873,
+                    1807
+                ],
+                "score": 0.92,
+                "latex": "\\mathcal { C } ( \\sigma , \\tau ) : = \\mathcal { C } ( \\sigma ) \\cap \\mathcal { C } ( \\tau )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    676,
+                    1226,
+                    904,
+                    1226,
+                    904,
+                    1261,
+                    676,
+                    1261
+                ],
+                "score": 0.92,
+                "latex": "\\mathcal { C } ( \\sigma ) = \\{ \\tau \\ | \\ \\sigma \\prec \\tau \\}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1113,
+                    1966,
+                    1261,
+                    1966,
+                    1261,
+                    2001,
+                    1113,
+                    2001
+                ],
+                "score": 0.91,
+                "latex": "\\delta \\in \\mathcal { C } ( \\sigma , \\tau ) \\big \\}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    618,
+                    1296,
+                    818,
+                    1296,
+                    818,
+                    1331,
+                    618,
+                    1331
+                ],
+                "score": 0.9,
+                "latex": "\\mathcal { N } _ { \\downarrow } ( \\sigma ) = \\{ \\tau \\ | \\ \\exists \\delta"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1111,
+                    1928,
+                    1262,
+                    1928,
+                    1262,
+                    1962,
+                    1111,
+                    1962
+                ],
+                "score": 0.9,
+                "latex": "\\delta \\in B ( \\sigma , \\tau ) \\}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    371,
+                    1775,
+                    461,
+                    1775,
+                    461,
+                    1804,
+                    371,
+                    1804
+                ],
+                "score": 0.9,
+                "latex": "\\sigma \\in P _ { X }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    619,
+                    1396,
+                    811,
+                    1396,
+                    811,
+                    1431,
+                    619,
+                    1431
+                ],
+                "score": 0.9,
+                "latex": "\\mathcal { N } _ { \\uparrow } ( \\sigma ) = \\{ \\tau \\mid \\exists i"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    298,
+                    949,
+                    419,
+                    949,
+                    419,
+                    978,
+                    298,
+                    978
+                ],
+                "score": 0.9,
+                "latex": "\\sigma < \\delta < \\tau"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    955,
+                    613,
+                    1012,
+                    613,
+                    1012,
+                    644,
+                    955,
+                    644
+                ],
+                "score": 0.9,
+                "latex": "X ^ { ( k ) }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1058,
+                    1297,
+                    1141,
+                    1297,
+                    1141,
+                    1330,
+                    1058,
+                    1330
+                ],
+                "score": 0.89,
+                "latex": "\\delta \\prec \\tau \\}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1205,
+                    702,
+                    1262,
+                    702,
+                    1262,
+                    734,
+                    1205,
+                    734
+                ],
+                "score": 0.89,
+                "latex": "X ^ { ( 0 ) }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    636,
+                    735,
+                    693,
+                    735,
+                    693,
+                    766,
+                    636,
+                    766
+                ],
+                "score": 0.89,
+                "latex": "X ^ { ( 1 ) }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    931,
+                    1297,
+                    1004,
+                    1297,
+                    1004,
+                    1327,
+                    931,
+                    1327
+                ],
+                "score": 0.89,
+                "latex": "\\delta \\prec \\sigma"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    707,
+                    1927,
+                    1057,
+                    1927,
+                    1057,
+                    1963,
+                    707,
+                    1963
+                ],
+                "score": 0.89,
+                "latex": "c _ { \\downarrow } ( \\sigma ) = \\{ \\{ ( c _ { \\tau } , c _ { \\delta } ) \\mid \\tau \\in \\mathcal { N } _ { \\downarrow } ( \\sigma )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    707,
+                    1967,
+                    1058,
+                    1967,
+                    1058,
+                    2002,
+                    707,
+                    2002
+                ],
+                "score": 0.88,
+                "latex": "c _ { \\uparrow } ( \\sigma ) = \\{ \\{ ( c _ { \\tau } , c _ { \\delta } ) \\mid \\tau \\in \\mathcal { N } _ { \\uparrow } ( \\sigma )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    947,
+                    920,
+                    1030,
+                    920,
+                    1030,
+                    949,
+                    947,
+                    949
+                ],
+                "score": 0.88,
+                "latex": "\\sigma < \\tau"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    810,
+                    1888,
+                    1105,
+                    1888,
+                    1105,
+                    1923,
+                    810,
+                    1923
+                ],
+                "score": 0.88,
+                "latex": "c _ { \\mathcal { C } } ( \\sigma ) = \\{ \\{ c _ { \\tau } ~ | ~ \\tau \\in \\mathcal { C } ( \\sigma ) \\} \\}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    774,
+                    1849,
+                    1075,
+                    1849,
+                    1075,
+                    1884,
+                    774,
+                    1884
+                ],
+                "score": 0.87,
+                "latex": "c _ { B } ( \\sigma ) = \\{ \\{ c _ { \\tau } \\mid \\tau \\in B ( \\sigma ) \\} \\}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1058,
+                    1397,
+                    1141,
+                    1397,
+                    1141,
+                    1430,
+                    1058,
+                    1430
+                ],
+                "score": 0.86,
+                "latex": "\\tau \\prec \\delta \\}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    826,
+                    921,
+                    910,
+                    921,
+                    910,
+                    949,
+                    826,
+                    949
+                ],
+                "score": 0.86,
+                "latex": "\\sigma \\prec \\tau"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    932,
+                    1397,
+                    1005,
+                    1397,
+                    1005,
+                    1427,
+                    932,
+                    1427
+                ],
+                "score": 0.84,
+                "latex": "\\sigma \\prec \\delta"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1044,
+                    1747,
+                    1074,
+                    1747,
+                    1074,
+                    1774,
+                    1044,
+                    1774
+                ],
+                "score": 0.83,
+                "latex": "c _ { \\sigma }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    958,
+                    1745,
+                    986,
+                    1745,
+                    986,
+                    1771,
+                    958,
+                    1771
+                ],
+                "score": 0.8,
+                "latex": "X"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    645,
+                    576,
+                    665,
+                    576,
+                    665,
+                    598,
+                    645,
+                    598
+                ],
+                "score": 0.74,
+                "latex": "n"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    298,
+                    576,
+                    317,
+                    576,
+                    317,
+                    598,
+                    298,
+                    598
+                ],
+                "score": 0.71,
+                "latex": "n"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    975,
+                    1113,
+                    1004,
+                    1113,
+                    1004,
+                    1140,
+                    975,
+                    1140
+                ],
+                "score": 0.7,
+                "latex": "X"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    823,
+                    617,
+                    850,
+                    617,
+                    850,
+                    644,
+                    823,
+                    644
+                ],
+                "score": 0.69,
+                "latex": "X"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1255,
+                    618,
+                    1282,
+                    618,
+                    1282,
+                    644,
+                    1255,
+                    644
+                ],
+                "score": 0.69,
+                "latex": "X"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    620,
+                    649,
+                    637,
+                    649,
+                    637,
+                    674,
+                    620,
+                    674
+                ],
+                "score": 0.66,
+                "latex": "k"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    673,
+                    1977,
+                    691,
+                    1977,
+                    691,
+                    1995,
+                    673,
+                    1995
+                ],
+                "score": 0.57,
+                "latex": "\\sigma"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1324,
+                    1434,
+                    1343,
+                    1434,
+                    1343,
+                    1455,
+                    1324,
+                    1455
+                ],
+                "score": 0.56,
+                "latex": "\\sigma"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    588,
+                    1434,
+                    608,
+                    1434,
+                    608,
+                    1455,
+                    588,
+                    1455
+                ],
+                "score": 0.54,
+                "latex": "\\sigma"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    774,
+                    1896,
+                    793,
+                    1896,
+                    793,
+                    1916,
+                    774,
+                    1916
+                ],
+                "score": 0.52,
+                "latex": "\\sigma"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    671,
+                    1936,
+                    690,
+                    1936,
+                    690,
+                    1956,
+                    671,
+                    1956
+                ],
+                "score": 0.52,
+                "latex": "\\sigma"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    662,
+                    874,
+                    687,
+                    874,
+                    687,
+                    911,
+                    662,
+                    911
+                ],
+                "score": 0.5,
+                "latex": "\\bigtriangledown"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    592,
+                    1334,
+                    612,
+                    1334,
+                    612,
+                    1355,
+                    592,
+                    1355
+                ],
+                "score": 0.49,
+                "latex": "\\sigma"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1270,
+                    920,
+                    1287,
+                    920,
+                    1287,
+                    946,
+                    1270,
+                    946
+                ],
+                "score": 0.45,
+                "latex": "\\delta"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    739,
+                    1856,
+                    758,
+                    1856,
+                    758,
+                    1877,
+                    739,
+                    1877
+                ],
+                "score": 0.43,
+                "latex": "\\sigma"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1134,
+                    1669,
+                    1160,
+                    1669,
+                    1160,
+                    1705,
+                    1134,
+                    1705
+                ],
+                "score": 0.4,
+                "latex": "\\bar { 4 }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    474,
+                    1195,
+                    493,
+                    1195,
+                    493,
+                    1216,
+                    474,
+                    1216
+                ],
+                "score": 0.27,
+                "latex": "\\sigma"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    348.0,
+                    1000.0,
+                    348.0,
+                    1000.0,
+                    382.0,
+                    295.0,
+                    382.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    379.0,
+                    997.0,
+                    379.0,
+                    997.0,
+                    412.0,
+                    295.0,
+                    412.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    410.0,
+                    614.0,
+                    410.0,
+                    614.0,
+                    445.0,
+                    294.0,
+                    445.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    214.0,
+                    652.0,
+                    214.0,
+                    652.0,
+                    353.0,
+                    296.0,
+                    353.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    730.0,
+                    252.0,
+                    810.0,
+                    252.0,
+                    810.0,
+                    326.0,
+                    730.0,
+                    326.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    693.0,
+                    272.0,
+                    722.0,
+                    272.0,
+                    722.0,
+                    294.0,
+                    693.0,
+                    294.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1024.0,
+                    346.0,
+                    1397.0,
+                    346.0,
+                    1397.0,
+                    385.0,
+                    1024.0,
+                    385.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1025.0,
+                    379.0,
+                    1397.0,
+                    379.0,
+                    1397.0,
+                    412.0,
+                    1025.0,
+                    412.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1026.0,
+                    411.0,
+                    1252.0,
+                    411.0,
+                    1252.0,
+                    446.0,
+                    1026.0,
+                    446.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1038.0,
+                    176.0,
+                    1381.0,
+                    176.0,
+                    1381.0,
+                    358.0,
+                    1038.0,
+                    358.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    290.0,
+                    1532.0,
+                    747.0,
+                    1532.0,
+                    747.0,
+                    1580.0,
+                    290.0,
+                    1580.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    838.0,
+                    2058.0,
+                    862.0,
+                    2058.0,
+                    862.0,
+                    2091.0,
+                    838.0,
+                    2091.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    837.0,
+                    2058.0,
+                    862.0,
+                    2058.0,
+                    862.0,
+                    2091.0,
+                    837.0,
+                    2091.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1606.0,
+                    1405.0,
+                    1606.0,
+                    1405.0,
+                    1642.0,
+                    295.0,
+                    1642.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1639.0,
+                    1405.0,
+                    1639.0,
+                    1405.0,
+                    1672.0,
+                    294.0,
+                    1672.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    290.0,
+                    1666.0,
+                    1133.0,
+                    1666.0,
+                    1133.0,
+                    1710.0,
+                    290.0,
+                    1710.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1161.0,
+                    1666.0,
+                    1410.0,
+                    1666.0,
+                    1410.0,
+                    1710.0,
+                    1161.0,
+                    1710.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1701.0,
+                    1018.0,
+                    1701.0,
+                    1018.0,
+                    1736.0,
+                    295.0,
+                    1736.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    478.0,
+                    1406.0,
+                    478.0,
+                    1406.0,
+                    513.0,
+                    294.0,
+                    513.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    508.0,
+                    1406.0,
+                    508.0,
+                    1406.0,
+                    544.0,
+                    294.0,
+                    544.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    538.0,
+                    1406.0,
+                    538.0,
+                    1406.0,
+                    575.0,
+                    294.0,
+                    575.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    569.0,
+                    297.0,
+                    569.0,
+                    297.0,
+                    605.0,
+                    294.0,
+                    605.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    318.0,
+                    569.0,
+                    644.0,
+                    569.0,
+                    644.0,
+                    605.0,
+                    318.0,
+                    605.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    666.0,
+                    569.0,
+                    1338.0,
+                    569.0,
+                    1338.0,
+                    605.0,
+                    666.0,
+                    605.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    784.0,
+                    1405.0,
+                    784.0,
+                    1405.0,
+                    820.0,
+                    294.0,
+                    820.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    815.0,
+                    1405.0,
+                    815.0,
+                    1405.0,
+                    848.0,
+                    295.0,
+                    848.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    844.0,
+                    1410.0,
+                    844.0,
+                    1410.0,
+                    883.0,
+                    292.0,
+                    883.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    875.0,
+                    661.0,
+                    875.0,
+                    661.0,
+                    912.0,
+                    294.0,
+                    912.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    688.0,
+                    875.0,
+                    1403.0,
+                    875.0,
+                    1403.0,
+                    912.0,
+                    688.0,
+                    912.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1007.0,
+                    1406.0,
+                    1007.0,
+                    1406.0,
+                    1044.0,
+                    294.0,
+                    1044.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1037.0,
+                    1404.0,
+                    1037.0,
+                    1404.0,
+                    1073.0,
+                    294.0,
+                    1073.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1069.0,
+                    1300.0,
+                    1069.0,
+                    1300.0,
+                    1103.0,
+                    295.0,
+                    1103.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    701.0,
+                    1204.0,
+                    701.0,
+                    1204.0,
+                    740.0,
+                    293.0,
+                    740.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1263.0,
+                    701.0,
+                    1404.0,
+                    701.0,
+                    1404.0,
+                    740.0,
+                    1263.0,
+                    740.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    733.0,
+                    635.0,
+                    733.0,
+                    635.0,
+                    776.0,
+                    293.0,
+                    776.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    694.0,
+                    733.0,
+                    1394.0,
+                    733.0,
+                    1394.0,
+                    776.0,
+                    694.0,
+                    776.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    606.0,
+                    822.0,
+                    606.0,
+                    822.0,
+                    657.0,
+                    292.0,
+                    657.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    851.0,
+                    606.0,
+                    954.0,
+                    606.0,
+                    954.0,
+                    657.0,
+                    851.0,
+                    657.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1013.0,
+                    606.0,
+                    1254.0,
+                    606.0,
+                    1254.0,
+                    657.0,
+                    1013.0,
+                    657.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1283.0,
+                    606.0,
+                    1407.0,
+                    606.0,
+                    1407.0,
+                    657.0,
+                    1283.0,
+                    657.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    646.0,
+                    619.0,
+                    646.0,
+                    619.0,
+                    681.0,
+                    294.0,
+                    681.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    638.0,
+                    646.0,
+                    649.0,
+                    646.0,
+                    649.0,
+                    681.0,
+                    638.0,
+                    681.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    915.0,
+                    825.0,
+                    915.0,
+                    825.0,
+                    955.0,
+                    293.0,
+                    955.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    911.0,
+                    915.0,
+                    946.0,
+                    915.0,
+                    946.0,
+                    955.0,
+                    911.0,
+                    955.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1031.0,
+                    915.0,
+                    1269.0,
+                    915.0,
+                    1269.0,
+                    955.0,
+                    1031.0,
+                    955.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1288.0,
+                    915.0,
+                    1407.0,
+                    915.0,
+                    1407.0,
+                    955.0,
+                    1288.0,
+                    955.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    947.0,
+                    297.0,
+                    947.0,
+                    297.0,
+                    984.0,
+                    292.0,
+                    984.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    420.0,
+                    947.0,
+                    433.0,
+                    947.0,
+                    433.0,
+                    984.0,
+                    420.0,
+                    984.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1739.0,
+                    957.0,
+                    1739.0,
+                    957.0,
+                    1779.0,
+                    292.0,
+                    1779.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    987.0,
+                    1739.0,
+                    1043.0,
+                    1739.0,
+                    1043.0,
+                    1779.0,
+                    987.0,
+                    1779.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1075.0,
+                    1739.0,
+                    1406.0,
+                    1739.0,
+                    1406.0,
+                    1779.0,
+                    1075.0,
+                    1779.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1771.0,
+                    370.0,
+                    1771.0,
+                    370.0,
+                    1811.0,
+                    292.0,
+                    1811.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    462.0,
+                    1771.0,
+                    551.0,
+                    1771.0,
+                    551.0,
+                    1811.0,
+                    462.0,
+                    1811.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    823.0,
+                    1771.0,
+                    872.0,
+                    1771.0,
+                    872.0,
+                    1811.0,
+                    823.0,
+                    1811.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1134.0,
+                    1771.0,
+                    1404.0,
+                    1771.0,
+                    1404.0,
+                    1811.0,
+                    1134.0,
+                    1811.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1801.0,
+                    539.0,
+                    1801.0,
+                    539.0,
+                    1839.0,
+                    294.0,
+                    1839.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1157.0,
+                    654.0,
+                    1157.0,
+                    654.0,
+                    1192.0,
+                    296.0,
+                    1192.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    887.0,
+                    1157.0,
+                    1403.0,
+                    1157.0,
+                    1403.0,
+                    1192.0,
+                    887.0,
+                    1192.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    327.0,
+                    1187.0,
+                    473.0,
+                    1187.0,
+                    473.0,
+                    1223.0,
+                    327.0,
+                    1223.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    494.0,
+                    1187.0,
+                    1164.0,
+                    1187.0,
+                    1164.0,
+                    1223.0,
+                    494.0,
+                    1223.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1226.0,
+                    675.0,
+                    1226.0,
+                    675.0,
+                    1263.0,
+                    292.0,
+                    1263.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    905.0,
+                    1226.0,
+                    1406.0,
+                    1226.0,
+                    1406.0,
+                    1263.0,
+                    905.0,
+                    1263.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    330.0,
+                    1260.0,
+                    1378.0,
+                    1260.0,
+                    1378.0,
+                    1291.0,
+                    330.0,
+                    1291.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1295.0,
+                    617.0,
+                    1295.0,
+                    617.0,
+                    1332.0,
+                    292.0,
+                    1332.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    819.0,
+                    1295.0,
+                    930.0,
+                    1295.0,
+                    930.0,
+                    1332.0,
+                    819.0,
+                    1332.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1005.0,
+                    1295.0,
+                    1057.0,
+                    1295.0,
+                    1057.0,
+                    1332.0,
+                    1005.0,
+                    1332.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1142.0,
+                    1295.0,
+                    1407.0,
+                    1295.0,
+                    1407.0,
+                    1332.0,
+                    1142.0,
+                    1332.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    329.0,
+                    1325.0,
+                    591.0,
+                    1325.0,
+                    591.0,
+                    1363.0,
+                    329.0,
+                    1363.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    613.0,
+                    1325.0,
+                    1403.0,
+                    1325.0,
+                    1403.0,
+                    1363.0,
+                    613.0,
+                    1363.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    330.0,
+                    1359.0,
+                    1001.0,
+                    1359.0,
+                    1001.0,
+                    1393.0,
+                    330.0,
+                    1393.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1393.0,
+                    618.0,
+                    1393.0,
+                    618.0,
+                    1434.0,
+                    291.0,
+                    1434.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    812.0,
+                    1393.0,
+                    931.0,
+                    1393.0,
+                    931.0,
+                    1434.0,
+                    812.0,
+                    1434.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1006.0,
+                    1393.0,
+                    1057.0,
+                    1393.0,
+                    1057.0,
+                    1434.0,
+                    1006.0,
+                    1434.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1142.0,
+                    1393.0,
+                    1408.0,
+                    1393.0,
+                    1408.0,
+                    1434.0,
+                    1142.0,
+                    1434.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    330.0,
+                    1428.0,
+                    587.0,
+                    1428.0,
+                    587.0,
+                    1463.0,
+                    330.0,
+                    1463.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    609.0,
+                    1428.0,
+                    1323.0,
+                    1428.0,
+                    1323.0,
+                    1463.0,
+                    609.0,
+                    1463.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1344.0,
+                    1428.0,
+                    1406.0,
+                    1428.0,
+                    1406.0,
+                    1463.0,
+                    1344.0,
+                    1463.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    330.0,
+                    1458.0,
+                    1173.0,
+                    1458.0,
+                    1173.0,
+                    1493.0,
+                    330.0,
+                    1493.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1846.0,
+                    738.0,
+                    1846.0,
+                    738.0,
+                    1885.0,
+                    295.0,
+                    1885.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    759.0,
+                    1846.0,
+                    773.0,
+                    1846.0,
+                    773.0,
+                    1885.0,
+                    759.0,
+                    1885.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1076.0,
+                    1846.0,
+                    1083.0,
+                    1846.0,
+                    1083.0,
+                    1885.0,
+                    1076.0,
+                    1885.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1885.0,
+                    773.0,
+                    1885.0,
+                    773.0,
+                    1924.0,
+                    295.0,
+                    1924.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    794.0,
+                    1885.0,
+                    809.0,
+                    1885.0,
+                    809.0,
+                    1924.0,
+                    794.0,
+                    1924.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1106.0,
+                    1885.0,
+                    1114.0,
+                    1885.0,
+                    1114.0,
+                    1924.0,
+                    1106.0,
+                    1924.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1925.0,
+                    670.0,
+                    1925.0,
+                    670.0,
+                    1965.0,
+                    293.0,
+                    1965.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    691.0,
+                    1925.0,
+                    706.0,
+                    1925.0,
+                    706.0,
+                    1965.0,
+                    691.0,
+                    1965.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1058.0,
+                    1925.0,
+                    1110.0,
+                    1925.0,
+                    1110.0,
+                    1965.0,
+                    1058.0,
+                    1965.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1263.0,
+                    1925.0,
+                    1273.0,
+                    1925.0,
+                    1273.0,
+                    1965.0,
+                    1263.0,
+                    1965.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1963.0,
+                    672.0,
+                    1963.0,
+                    672.0,
+                    2002.0,
+                    293.0,
+                    2002.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    692.0,
+                    1963.0,
+                    706.0,
+                    1963.0,
+                    706.0,
+                    2002.0,
+                    692.0,
+                    2002.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1059.0,
+                    1963.0,
+                    1112.0,
+                    1963.0,
+                    1112.0,
+                    2002.0,
+                    1059.0,
+                    2002.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1262.0,
+                    1963.0,
+                    1270.0,
+                    1963.0,
+                    1270.0,
+                    2002.0,
+                    1262.0,
+                    2002.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1108.0,
+                    974.0,
+                    1108.0,
+                    974.0,
+                    1150.0,
+                    293.0,
+                    1150.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1005.0,
+                    1108.0,
+                    1122.0,
+                    1108.0,
+                    1122.0,
+                    1150.0,
+                    1005.0,
+                    1150.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1215.0,
+                    1108.0,
+                    1346.0,
+                    1108.0,
+                    1346.0,
+                    1150.0,
+                    1215.0,
+                    1150.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 2,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1612,
+                    1403,
+                    1612,
+                    1403,
+                    1797,
+                    298,
+                    1797
+                ],
+                "score": 0.971
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    592,
+                    1404,
+                    592,
+                    1404,
+                    715,
+                    298,
+                    715
+                ],
+                "score": 0.971
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    745,
+                    1404,
+                    745,
+                    1404,
+                    870,
+                    298,
+                    870
+                ],
+                "score": 0.97
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    1914,
+                    1403,
+                    1914,
+                    1403,
+                    2007,
+                    299,
+                    2007
+                ],
+                "score": 0.968
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    1295,
+                    1404,
+                    1295,
+                    1404,
+                    1418,
+                    299,
+                    1418
+                ],
+                "score": 0.967
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    1507,
+                    1404,
+                    1507,
+                    1404,
+                    1598,
+                    299,
+                    1598
+                ],
+                "score": 0.967
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1803,
+                    1403,
+                    1803,
+                    1403,
+                    1896,
+                    298,
+                    1896
+                ],
+                "score": 0.958
+            },
+            {
+                "category_id": 4,
+                "poly": [
+                    297,
+                    447,
+                    1399,
+                    447,
+                    1399,
+                    543,
+                    297,
+                    543
+                ],
+                "score": 0.947
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1424,
+                    1400,
+                    1424,
+                    1400,
+                    1487,
+                    297,
+                    1487
+                ],
+                "score": 0.935
+            },
+            {
+                "category_id": 3,
+                "poly": [
+                    358,
+                    198,
+                    1349,
+                    198,
+                    1349,
+                    429,
+                    358,
+                    429
+                ],
+                "score": 0.931
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    300,
+                    1115,
+                    1401,
+                    1115,
+                    1401,
+                    1209,
+                    300,
+                    1209
+                ],
+                "score": 0.918
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    295,
+                    1214,
+                    1401,
+                    1214,
+                    1401,
+                    1276,
+                    295,
+                    1276
+                ],
+                "score": 0.905
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    887,
+                    1412,
+                    887,
+                    1412,
+                    1097,
+                    297,
+                    1097
+                ],
+                "score": 0.787
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    841,
+                    2062,
+                    858,
+                    2062,
+                    858,
+                    2084,
+                    841,
+                    2084
+                ],
+                "score": 0.762
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1001,
+                    593,
+                    1087,
+                    593,
+                    1087,
+                    627,
+                    1001,
+                    627
+                ],
+                "score": 0.93,
+                "latex": "B ( \\sigma , \\tau )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    467,
+                    1862,
+                    625,
+                    1862,
+                    625,
+                    1896,
+                    467,
+                    1896
+                ],
+                "score": 0.93,
+                "latex": "f ( G _ { 1 } ) , f ( \\mathbf { \\bar { G } } _ { 2 } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1142,
+                    593,
+                    1225,
+                    593,
+                    1225,
+                    626,
+                    1142,
+                    626
+                ],
+                "score": 0.93,
+                "latex": "\\mathcal { C } ( \\sigma , \\tau )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    332,
+                    957,
+                    384,
+                    957,
+                    384,
+                    992,
+                    332,
+                    992
+                ],
+                "score": 0.91,
+                "latex": "c _ { \\sigma } ^ { t + 1 }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1070,
+                    1834,
+                    1152,
+                    1834,
+                    1152,
+                    1865,
+                    1070,
+                    1865
+                ],
+                "score": 0.91,
+                "latex": "G _ { 1 } , G _ { 2 }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    868,
+                    1803,
+                    1007,
+                    1803,
+                    1007,
+                    1834,
+                    868,
+                    1834
+                ],
+                "score": 0.91,
+                "latex": "f : \\mathcal G  \\mathcal { X }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    605,
+                    989,
+                    1133,
+                    989,
+                    1133,
+                    1030,
+                    605,
+                    1030
+                ],
+                "score": 0.91,
+                "latex": "c _ { \\sigma } ^ { t + 1 } = \\mathrm { H A S H } \\big ( c _ { \\sigma } ^ { t } , c _ { \\mathcal { B } } ^ { t } ( \\sigma ) , c _ { \\mathcal { C } } ^ { t } ( \\sigma ) , c _ { \\mathcal { A } } ^ { t } ( \\sigma ) , c _ { \\uparrow } ^ { t } ( \\sigma ) \\big )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    527,
+                    927,
+                    557,
+                    927,
+                    557,
+                    962,
+                    527,
+                    962
+                ],
+                "score": 0.89,
+                "latex": "c _ { \\sigma } ^ { t }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    992,
+                    776,
+                    1022,
+                    776,
+                    1022,
+                    809,
+                    992,
+                    809
+                ],
+                "score": 0.88,
+                "latex": "c _ { \\sigma } ^ { t }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    658,
+                    891,
+                    686,
+                    891,
+                    686,
+                    917,
+                    658,
+                    917
+                ],
+                "score": 0.79,
+                "latex": "X"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    657,
+                    1835,
+                    684,
+                    1835,
+                    684,
+                    1860,
+                    657,
+                    1860
+                ],
+                "score": 0.79,
+                "latex": "\\mathcal { X }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    830,
+                    895,
+                    851,
+                    895,
+                    851,
+                    917,
+                    830,
+                    917
+                ],
+                "score": 0.79,
+                "latex": "\\sigma"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    442,
+                    811,
+                    463,
+                    811,
+                    463,
+                    834,
+                    442,
+                    834
+                ],
+                "score": 0.78,
+                "latex": "\\sigma"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    909,
+                    1115,
+                    944,
+                    1115,
+                    944,
+                    1148,
+                    909,
+                    1148
+                ],
+                "score": 0.77,
+                "latex": "\\textcircled { 8 } \\textcircled { 1 8 }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    638,
+                    934,
+                    659,
+                    934,
+                    659,
+                    956,
+                    638,
+                    956
+                ],
+                "score": 0.77,
+                "latex": "\\sigma"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1295,
+                    964,
+                    1315,
+                    964,
+                    1315,
+                    986,
+                    1295,
+                    986
+                ],
+                "score": 0.76,
+                "latex": "\\sigma"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1306,
+                    1804,
+                    1327,
+                    1804,
+                    1327,
+                    1833,
+                    1306,
+                    1833
+                ],
+                "score": 0.76,
+                "latex": "\\mathcal { G }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1156,
+                    934,
+                    1177,
+                    934,
+                    1177,
+                    956,
+                    1156,
+                    956
+                ],
+                "score": 0.75,
+                "latex": "\\sigma"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    593,
+                    810,
+                    607,
+                    810,
+                    607,
+                    834,
+                    593,
+                    834
+                ],
+                "score": 0.74,
+                "latex": "t"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    790,
+                    932,
+                    803,
+                    932,
+                    803,
+                    956,
+                    790,
+                    956
+                ],
+                "score": 0.73,
+                "latex": "t"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    647,
+                    776,
+                    681,
+                    776,
+                    681,
+                    808,
+                    647,
+                    808
+                ],
+                "score": 0.63,
+                "latex": "\\textcircled { 8 } \\textcircled { 1 8 }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    594,
+                    652,
+                    624,
+                    652,
+                    624,
+                    689,
+                    594,
+                    689
+                ],
+                "score": 0.62,
+                "latex": "\\textcircled { 3 }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    785,
+                    776,
+                    834,
+                    776,
+                    834,
+                    808,
+                    785,
+                    808
+                ],
+                "score": 0.53,
+                "latex": "\\mathbb { \\ m }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1152,
+                    836,
+                    1178,
+                    836,
+                    1178,
+                    873,
+                    1152,
+                    873
+                ],
+                "score": 0.47,
+                "latex": "^ 4"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    448.0,
+                    1403.0,
+                    448.0,
+                    1403.0,
+                    483.0,
+                    296.0,
+                    483.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    477.0,
+                    1404.0,
+                    477.0,
+                    1404.0,
+                    517.0,
+                    292.0,
+                    517.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    511.0,
+                    979.0,
+                    511.0,
+                    979.0,
+                    545.0,
+                    293.0,
+                    545.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    709.0,
+                    202.0,
+                    1006.0,
+                    202.0,
+                    1006.0,
+                    239.0,
+                    709.0,
+                    239.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1025.0,
+                    210.0,
+                    1082.0,
+                    210.0,
+                    1082.0,
+                    230.0,
+                    1025.0,
+                    230.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1099.0,
+                    207.0,
+                    1292.0,
+                    207.0,
+                    1292.0,
+                    235.0,
+                    1099.0,
+                    235.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    682.0,
+                    283.0,
+                    805.0,
+                    283.0,
+                    805.0,
+                    313.0,
+                    682.0,
+                    313.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    780.0,
+                    308.0,
+                    806.0,
+                    308.0,
+                    806.0,
+                    332.0,
+                    780.0,
+                    332.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    818.0,
+                    333.0,
+                    897.0,
+                    333.0,
+                    897.0,
+                    363.0,
+                    818.0,
+                    363.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    921.0,
+                    320.0,
+                    1036.0,
+                    320.0,
+                    1036.0,
+                    371.0,
+                    921.0,
+                    371.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1050.0,
+                    325.0,
+                    1192.0,
+                    325.0,
+                    1192.0,
+                    370.0,
+                    1050.0,
+                    370.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1222.0,
+                    324.0,
+                    1349.0,
+                    324.0,
+                    1349.0,
+                    371.0,
+                    1222.0,
+                    371.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    823.0,
+                    359.0,
+                    890.0,
+                    359.0,
+                    890.0,
+                    382.0,
+                    823.0,
+                    382.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1102.0,
+                    369.0,
+                    1139.0,
+                    369.0,
+                    1139.0,
+                    394.0,
+                    1102.0,
+                    394.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1232.0,
+                    369.0,
+                    1340.0,
+                    369.0,
+                    1340.0,
+                    394.0,
+                    1232.0,
+                    394.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    489.0,
+                    410.0,
+                    503.0,
+                    410.0,
+                    503.0,
+                    425.0,
+                    489.0,
+                    425.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    838.0,
+                    2059.0,
+                    862.0,
+                    2059.0,
+                    862.0,
+                    2091.0,
+                    838.0,
+                    2091.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1613.0,
+                    1405.0,
+                    1613.0,
+                    1405.0,
+                    1649.0,
+                    294.0,
+                    1649.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1642.0,
+                    1407.0,
+                    1642.0,
+                    1407.0,
+                    1680.0,
+                    292.0,
+                    1680.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1673.0,
+                    1408.0,
+                    1673.0,
+                    1408.0,
+                    1710.0,
+                    292.0,
+                    1710.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1703.0,
+                    1408.0,
+                    1703.0,
+                    1408.0,
+                    1740.0,
+                    292.0,
+                    1740.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1730.0,
+                    1407.0,
+                    1730.0,
+                    1407.0,
+                    1773.0,
+                    292.0,
+                    1773.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1762.0,
+                    622.0,
+                    1762.0,
+                    622.0,
+                    1801.0,
+                    293.0,
+                    1801.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    590.0,
+                    1000.0,
+                    590.0,
+                    1000.0,
+                    628.0,
+                    292.0,
+                    628.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1088.0,
+                    590.0,
+                    1141.0,
+                    590.0,
+                    1141.0,
+                    628.0,
+                    1088.0,
+                    628.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1226.0,
+                    590.0,
+                    1405.0,
+                    590.0,
+                    1405.0,
+                    628.0,
+                    1226.0,
+                    628.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    624.0,
+                    1404.0,
+                    624.0,
+                    1404.0,
+                    657.0,
+                    294.0,
+                    657.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    651.0,
+                    593.0,
+                    651.0,
+                    593.0,
+                    692.0,
+                    294.0,
+                    692.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    625.0,
+                    651.0,
+                    1408.0,
+                    651.0,
+                    1408.0,
+                    692.0,
+                    625.0,
+                    692.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    683.0,
+                    1223.0,
+                    683.0,
+                    1223.0,
+                    717.0,
+                    293.0,
+                    717.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    743.0,
+                    1406.0,
+                    743.0,
+                    1406.0,
+                    781.0,
+                    295.0,
+                    781.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    774.0,
+                    646.0,
+                    774.0,
+                    646.0,
+                    813.0,
+                    292.0,
+                    813.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    682.0,
+                    774.0,
+                    784.0,
+                    774.0,
+                    784.0,
+                    813.0,
+                    682.0,
+                    813.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    835.0,
+                    774.0,
+                    991.0,
+                    774.0,
+                    991.0,
+                    813.0,
+                    835.0,
+                    813.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1023.0,
+                    774.0,
+                    1406.0,
+                    774.0,
+                    1406.0,
+                    813.0,
+                    1023.0,
+                    813.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    805.0,
+                    441.0,
+                    805.0,
+                    441.0,
+                    840.0,
+                    295.0,
+                    840.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    464.0,
+                    805.0,
+                    592.0,
+                    805.0,
+                    592.0,
+                    840.0,
+                    464.0,
+                    840.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    608.0,
+                    805.0,
+                    1405.0,
+                    805.0,
+                    1405.0,
+                    840.0,
+                    608.0,
+                    840.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    837.0,
+                    1151.0,
+                    837.0,
+                    1151.0,
+                    873.0,
+                    294.0,
+                    873.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1179.0,
+                    837.0,
+                    1360.0,
+                    837.0,
+                    1360.0,
+                    873.0,
+                    1179.0,
+                    873.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1911.0,
+                    1404.0,
+                    1911.0,
+                    1404.0,
+                    1952.0,
+                    293.0,
+                    1952.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1946.0,
+                    1403.0,
+                    1946.0,
+                    1403.0,
+                    1980.0,
+                    295.0,
+                    1980.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    1977.0,
+                    678.0,
+                    1977.0,
+                    678.0,
+                    2011.0,
+                    297.0,
+                    2011.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1297.0,
+                    1404.0,
+                    1297.0,
+                    1404.0,
+                    1330.0,
+                    294.0,
+                    1330.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1326.0,
+                    1405.0,
+                    1326.0,
+                    1405.0,
+                    1362.0,
+                    293.0,
+                    1362.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1359.0,
+                    1405.0,
+                    1359.0,
+                    1405.0,
+                    1391.0,
+                    295.0,
+                    1391.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1389.0,
+                    484.0,
+                    1389.0,
+                    484.0,
+                    1419.0,
+                    295.0,
+                    1419.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1505.0,
+                    1404.0,
+                    1505.0,
+                    1404.0,
+                    1542.0,
+                    294.0,
+                    1542.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1536.0,
+                    1404.0,
+                    1536.0,
+                    1404.0,
+                    1572.0,
+                    293.0,
+                    1572.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    1568.0,
+                    1056.0,
+                    1568.0,
+                    1056.0,
+                    1601.0,
+                    297.0,
+                    1601.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1799.0,
+                    867.0,
+                    1799.0,
+                    867.0,
+                    1840.0,
+                    293.0,
+                    1840.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1008.0,
+                    1799.0,
+                    1305.0,
+                    1799.0,
+                    1305.0,
+                    1840.0,
+                    1008.0,
+                    1840.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1328.0,
+                    1799.0,
+                    1407.0,
+                    1799.0,
+                    1407.0,
+                    1840.0,
+                    1328.0,
+                    1840.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1830.0,
+                    656.0,
+                    1830.0,
+                    656.0,
+                    1869.0,
+                    293.0,
+                    1869.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    685.0,
+                    1830.0,
+                    1069.0,
+                    1830.0,
+                    1069.0,
+                    1869.0,
+                    685.0,
+                    1869.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1153.0,
+                    1830.0,
+                    1404.0,
+                    1830.0,
+                    1404.0,
+                    1869.0,
+                    1153.0,
+                    1869.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1862.0,
+                    466.0,
+                    1862.0,
+                    466.0,
+                    1899.0,
+                    294.0,
+                    1899.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    626.0,
+                    1862.0,
+                    809.0,
+                    1862.0,
+                    809.0,
+                    1899.0,
+                    626.0,
+                    1899.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1421.0,
+                    1405.0,
+                    1421.0,
+                    1405.0,
+                    1461.0,
+                    295.0,
+                    1461.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1456.0,
+                    1318.0,
+                    1456.0,
+                    1318.0,
+                    1491.0,
+                    296.0,
+                    1491.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1113.0,
+                    908.0,
+                    1113.0,
+                    908.0,
+                    1152.0,
+                    293.0,
+                    1152.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    945.0,
+                    1113.0,
+                    1406.0,
+                    1113.0,
+                    1406.0,
+                    1152.0,
+                    945.0,
+                    1152.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1144.0,
+                    1405.0,
+                    1144.0,
+                    1405.0,
+                    1184.0,
+                    294.0,
+                    1184.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1178.0,
+                    1393.0,
+                    1178.0,
+                    1393.0,
+                    1212.0,
+                    296.0,
+                    1212.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1212.0,
+                    1405.0,
+                    1212.0,
+                    1405.0,
+                    1250.0,
+                    294.0,
+                    1250.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1244.0,
+                    1383.0,
+                    1244.0,
+                    1383.0,
+                    1280.0,
+                    295.0,
+                    1280.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    889.0,
+                    657.0,
+                    889.0,
+                    657.0,
+                    923.0,
+                    296.0,
+                    923.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    687.0,
+                    889.0,
+                    829.0,
+                    889.0,
+                    829.0,
+                    923.0,
+                    687.0,
+                    923.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    852.0,
+                    889.0,
+                    1252.0,
+                    889.0,
+                    1252.0,
+                    923.0,
+                    852.0,
+                    923.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    925.0,
+                    526.0,
+                    925.0,
+                    526.0,
+                    964.0,
+                    293.0,
+                    964.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    558.0,
+                    925.0,
+                    637.0,
+                    925.0,
+                    637.0,
+                    964.0,
+                    558.0,
+                    964.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    660.0,
+                    925.0,
+                    789.0,
+                    925.0,
+                    789.0,
+                    964.0,
+                    660.0,
+                    964.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    804.0,
+                    925.0,
+                    1155.0,
+                    925.0,
+                    1155.0,
+                    964.0,
+                    804.0,
+                    964.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1178.0,
+                    925.0,
+                    1405.0,
+                    925.0,
+                    1405.0,
+                    964.0,
+                    1178.0,
+                    964.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    319.0,
+                    945.0,
+                    331.0,
+                    945.0,
+                    331.0,
+                    1009.0,
+                    319.0,
+                    1009.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    385.0,
+                    945.0,
+                    1294.0,
+                    945.0,
+                    1294.0,
+                    1009.0,
+                    385.0,
+                    1009.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1316.0,
+                    945.0,
+                    1416.0,
+                    945.0,
+                    1416.0,
+                    1009.0,
+                    1316.0,
+                    1009.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    324.0,
+                    985.0,
+                    604.0,
+                    985.0,
+                    604.0,
+                    1032.0,
+                    324.0,
+                    1032.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1134.0,
+                    985.0,
+                    1146.0,
+                    985.0,
+                    1146.0,
+                    1032.0,
+                    1134.0,
+                    1032.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1034.0,
+                    1405.0,
+                    1034.0,
+                    1405.0,
+                    1068.0,
+                    294.0,
+                    1068.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    328.0,
+                    1066.0,
+                    1344.0,
+                    1066.0,
+                    1344.0,
+                    1097.0,
+                    328.0,
+                    1097.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 3,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    296,
+                    1731,
+                    1405,
+                    1731,
+                    1405,
+                    2008,
+                    296,
+                    2008
+                ],
+                "score": 0.978
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    554,
+                    1405,
+                    554,
+                    1405,
+                    740,
+                    298,
+                    740
+                ],
+                "score": 0.972
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1037,
+                    989,
+                    1037,
+                    989,
+                    1160,
+                    297,
+                    1160
+                ],
+                "score": 0.966
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1533,
+                    1405,
+                    1533,
+                    1405,
+                    1626,
+                    298,
+                    1626
+                ],
+                "score": 0.959
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    743,
+                    1402,
+                    743,
+                    1402,
+                    837,
+                    298,
+                    837
+                ],
+                "score": 0.957
+            },
+            {
+                "category_id": 3,
+                "poly": [
+                    1012,
+                    863,
+                    1402,
+                    863,
+                    1402,
+                    998,
+                    1012,
+                    998
+                ],
+                "score": 0.95
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    857,
+                    990,
+                    857,
+                    990,
+                    920,
+                    297,
+                    920
+                ],
+                "score": 0.946
+            },
+            {
+                "category_id": 4,
+                "poly": [
+                    1010,
+                    1010,
+                    1406,
+                    1010,
+                    1406,
+                    1133,
+                    1010,
+                    1133
+                ],
+                "score": 0.945
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    924,
+                    990,
+                    924,
+                    990,
+                    1017,
+                    297,
+                    1017
+                ],
+                "score": 0.945
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    293,
+                    203,
+                    1401,
+                    203,
+                    1401,
+                    266,
+                    293,
+                    266
+                ],
+                "score": 0.94
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    298,
+                    1664,
+                    1022,
+                    1664,
+                    1022,
+                    1704,
+                    298,
+                    1704
+                ],
+                "score": 0.932
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    295,
+                    1464,
+                    1401,
+                    1464,
+                    1401,
+                    1530,
+                    295,
+                    1530
+                ],
+                "score": 0.929
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    1314,
+                    1398,
+                    1314,
+                    1398,
+                    1378,
+                    299,
+                    1378
+                ],
+                "score": 0.924
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    300,
+                    473,
+                    1403,
+                    473,
+                    1403,
+                    536,
+                    300,
+                    536
+                ],
+                "score": 0.923
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    296,
+                    404,
+                    1396,
+                    404,
+                    1396,
+                    467,
+                    296,
+                    467
+                ],
+                "score": 0.922
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    291,
+                    283,
+                    1403,
+                    283,
+                    1403,
+                    347,
+                    291,
+                    347
+                ],
+                "score": 0.917
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    294,
+                    1382,
+                    1403,
+                    1382,
+                    1403,
+                    1447,
+                    294,
+                    1447
+                ],
+                "score": 0.914
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    300,
+                    1232,
+                    1399,
+                    1232,
+                    1399,
+                    1294,
+                    300,
+                    1294
+                ],
+                "score": 0.911
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1165,
+                    1402,
+                    1165,
+                    1402,
+                    1228,
+                    298,
+                    1228
+                ],
+                "score": 0.91
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    353,
+                    1205,
+                    353,
+                    1205,
+                    386,
+                    299,
+                    386
+                ],
+                "score": 0.874
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    841,
+                    2061,
+                    858,
+                    2061,
+                    858,
+                    2085,
+                    841,
+                    2085
+                ],
+                "score": 0.703
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    841,
+                    2061,
+                    859,
+                    2061,
+                    859,
+                    2085,
+                    841,
+                    2085
+                ],
+                "score": 0.103
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    982,
+                    1414,
+                    1053,
+                    1414,
+                    1053,
+                    1444,
+                    982,
+                    1444
+                ],
+                "score": 0.92,
+                "latex": "k \\geq 8 ."
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1289,
+                    472,
+                    1351,
+                    472,
+                    1351,
+                    507,
+                    1289,
+                    507
+                ],
+                "score": 0.91,
+                "latex": "f ( G )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    748,
+                    1595,
+                    828,
+                    1595,
+                    828,
+                    1625,
+                    748,
+                    1625
+                ],
+                "score": 0.9,
+                "latex": "k _ { 2 } \\geq 5"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    676,
+                    1415,
+                    746,
+                    1415,
+                    746,
+                    1444,
+                    676,
+                    1444
+                ],
+                "score": 0.9,
+                "latex": "k \\geq 4"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    296,
+                    234,
+                    381,
+                    234,
+                    381,
+                    267,
+                    296,
+                    267
+                ],
+                "score": 0.89,
+                "latex": "( k + 1 )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    551,
+                    1233,
+                    621,
+                    1233,
+                    621,
+                    1264,
+                    551,
+                    1264
+                ],
+                "score": 0.86,
+                "latex": "k \\geq 3"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    405,
+                    1414,
+                    474,
+                    1414,
+                    474,
+                    1444,
+                    405,
+                    1444
+                ],
+                "score": 0.85,
+                "latex": "k \\geq 4"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    591,
+                    1166,
+                    610,
+                    1166,
+                    610,
+                    1194,
+                    591,
+                    1194
+                ],
+                "score": 0.75,
+                "latex": "k"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    514,
+                    1166,
+                    533,
+                    1166,
+                    533,
+                    1193,
+                    514,
+                    1193
+                ],
+                "score": 0.75,
+                "latex": "k"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    583,
+                    1595,
+                    612,
+                    1595,
+                    612,
+                    1624,
+                    583,
+                    1624
+                ],
+                "score": 0.75,
+                "latex": "k _ { 1 }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1165,
+                    284,
+                    1200,
+                    284,
+                    1200,
+                    316,
+                    1165,
+                    316
+                ],
+                "score": 0.75,
+                "latex": "\\pmb { \\mathbb { B } } ] \\mathbf { l }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    660,
+                    1166,
+                    679,
+                    1166,
+                    679,
+                    1194,
+                    660,
+                    1194
+                ],
+                "score": 0.7,
+                "latex": "k"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    506,
+                    1534,
+                    525,
+                    1534,
+                    525,
+                    1562,
+                    506,
+                    1562
+                ],
+                "score": 0.69,
+                "latex": "k"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    889,
+                    1198,
+                    906,
+                    1198,
+                    906,
+                    1223,
+                    889,
+                    1223
+                ],
+                "score": 0.66,
+                "latex": "k"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    718,
+                    1534,
+                    736,
+                    1534,
+                    736,
+                    1562,
+                    718,
+                    1562
+                ],
+                "score": 0.65,
+                "latex": "k"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    664,
+                    957,
+                    685,
+                    957,
+                    685,
+                    987,
+                    664,
+                    987
+                ],
+                "score": 0.65,
+                "latex": "f"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1028,
+                    1564,
+                    1269,
+                    1564,
+                    1269,
+                    1595,
+                    1028,
+                    1595
+                ],
+                "score": 0.64,
+                "latex": "C W L ( k _ { 1 } { \\mathrm { - C L } } \\cup k _ { 2 } { \\mathrm { - C } } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    728,
+                    235,
+                    746,
+                    235,
+                    746,
+                    261,
+                    728,
+                    261
+                ],
+                "score": 0.6,
+                "latex": "k"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    853,
+                    1235,
+                    871,
+                    1235,
+                    871,
+                    1262,
+                    853,
+                    1262
+                ],
+                "score": 0.59,
+                "latex": "k"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    700,
+                    1235,
+                    718,
+                    1235,
+                    718,
+                    1261,
+                    700,
+                    1261
+                ],
+                "score": 0.57,
+                "latex": "k"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    799,
+                    1414,
+                    923,
+                    1414,
+                    923,
+                    1446,
+                    799,
+                    1446
+                ],
+                "score": 0.52,
+                "latex": "C W L ( k { \\mathrm { - C } } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    293,
+                    1884,
+                    317,
+                    1884,
+                    317,
+                    1920,
+                    293,
+                    1920
+                ],
+                "score": 0.5,
+                "latex": "\\underline { { \\overline { { 7 } } } } ,"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    298,
+                    505,
+                    323,
+                    505,
+                    323,
+                    531,
+                    298,
+                    531
+                ],
+                "score": 0.49,
+                "latex": "G"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    501,
+                    925,
+                    521,
+                    925,
+                    521,
+                    957,
+                    501,
+                    957
+                ],
+                "score": 0.47,
+                "latex": "f"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    721,
+                    1564,
+                    974,
+                    1564,
+                    974,
+                    1595,
+                    721,
+                    1595
+                ],
+                "score": 0.46,
+                "latex": "C W L ( k _ { 1 } { \\mathrm { - C L } } \\cup k _ { 2 } { \\mathrm { - I C } } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1309,
+                    1792,
+                    1340,
+                    1792,
+                    1340,
+                    1829,
+                    1309,
+                    1829
+                ],
+                "score": 0.46,
+                "latex": "2 ."
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    719,
+                    1533,
+                    835,
+                    1533,
+                    835,
+                    1562,
+                    719,
+                    1562
+                ],
+                "score": 0.46,
+                "latex": "k { \\mathrm { - C L } } \\cup k "
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    603,
+                    1534,
+                    622,
+                    1534,
+                    622,
+                    1561,
+                    603,
+                    1561
+                ],
+                "score": 0.45,
+                "latex": "k"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    552,
+                    1415,
+                    570,
+                    1415,
+                    570,
+                    1441,
+                    552,
+                    1441
+                ],
+                "score": 0.41,
+                "latex": "k"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    891,
+                    1495,
+                    930,
+                    1495,
+                    930,
+                    1531,
+                    891,
+                    1531
+                ],
+                "score": 0.41,
+                "latex": "3 1"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    297,
+                    955,
+                    393,
+                    955,
+                    393,
+                    987,
+                    297,
+                    987
+                ],
+                "score": 0.4,
+                "latex": "C W L ( f )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1061,
+                    474,
+                    1086,
+                    474,
+                    1086,
+                    502,
+                    1061,
+                    502
+                ],
+                "score": 0.38,
+                "latex": "G"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    974,
+                    1233,
+                    1094,
+                    1233,
+                    1094,
+                    1267,
+                    974,
+                    1267
+                ],
+                "score": 0.31,
+                "latex": "C W L ( k { \\mathrm { - C } } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    892,
+                    1564,
+                    922,
+                    1564,
+                    922,
+                    1594,
+                    892,
+                    1594
+                ],
+                "score": 0.3,
+                "latex": "k _ { 2 }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    816,
+                    1535,
+                    834,
+                    1535,
+                    834,
+                    1561,
+                    816,
+                    1561
+                ],
+                "score": 0.29,
+                "latex": "k"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1092,
+                    1565,
+                    1124,
+                    1565,
+                    1124,
+                    1594,
+                    1092,
+                    1594
+                ],
+                "score": 0.29,
+                "latex": "\\boldsymbol { \\mathit { ^ { * } k } _ { 1 } }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1200,
+                    1564,
+                    1230,
+                    1564,
+                    1230,
+                    1594,
+                    1200,
+                    1594
+                ],
+                "score": 0.27,
+                "latex": "k _ { 2 }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    788,
+                    1565,
+                    816,
+                    1565,
+                    816,
+                    1594,
+                    788,
+                    1594
+                ],
+                "score": 0.26,
+                "latex": "k _ { 1 }"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1023.0,
+                    879.0,
+                    1137.0,
+                    879.0,
+                    1137.0,
+                    971.0,
+                    1023.0,
+                    971.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1285.0,
+                    881.0,
+                    1380.0,
+                    881.0,
+                    1380.0,
+                    939.0,
+                    1285.0,
+                    939.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1010.0,
+                    1009.0,
+                    1408.0,
+                    1009.0,
+                    1408.0,
+                    1044.0,
+                    1010.0,
+                    1044.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1008.0,
+                    1041.0,
+                    1405.0,
+                    1041.0,
+                    1405.0,
+                    1073.0,
+                    1008.0,
+                    1073.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1008.0,
+                    1072.0,
+                    1407.0,
+                    1072.0,
+                    1407.0,
+                    1104.0,
+                    1008.0,
+                    1104.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1007.0,
+                    1102.0,
+                    1405.0,
+                    1102.0,
+                    1405.0,
+                    1134.0,
+                    1007.0,
+                    1134.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1662.0,
+                    1023.0,
+                    1662.0,
+                    1023.0,
+                    1709.0,
+                    292.0,
+                    1709.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    839.0,
+                    2058.0,
+                    861.0,
+                    2058.0,
+                    861.0,
+                    2093.0,
+                    839.0,
+                    2093.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    838.0,
+                    2058.0,
+                    862.0,
+                    2058.0,
+                    862.0,
+                    2092.0,
+                    838.0,
+                    2092.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1731.0,
+                    1408.0,
+                    1731.0,
+                    1408.0,
+                    1769.0,
+                    293.0,
+                    1769.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1762.0,
+                    1406.0,
+                    1762.0,
+                    1406.0,
+                    1798.0,
+                    294.0,
+                    1798.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1792.0,
+                    1308.0,
+                    1792.0,
+                    1308.0,
+                    1828.0,
+                    294.0,
+                    1828.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1343.0,
+                    1792.0,
+                    1405.0,
+                    1792.0,
+                    1405.0,
+                    1827.0,
+                    1343.0,
+                    1827.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1822.0,
+                    1405.0,
+                    1822.0,
+                    1405.0,
+                    1859.0,
+                    294.0,
+                    1859.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1855.0,
+                    1405.0,
+                    1855.0,
+                    1405.0,
+                    1888.0,
+                    294.0,
+                    1888.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    318.0,
+                    1883.0,
+                    1406.0,
+                    1883.0,
+                    1406.0,
+                    1923.0,
+                    318.0,
+                    1923.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1915.0,
+                    1403.0,
+                    1915.0,
+                    1403.0,
+                    1948.0,
+                    294.0,
+                    1948.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1947.0,
+                    1405.0,
+                    1947.0,
+                    1405.0,
+                    1981.0,
+                    295.0,
+                    1981.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1977.0,
+                    1032.0,
+                    1977.0,
+                    1032.0,
+                    2010.0,
+                    295.0,
+                    2010.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    551.0,
+                    1405.0,
+                    551.0,
+                    1405.0,
+                    593.0,
+                    292.0,
+                    593.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    586.0,
+                    1405.0,
+                    586.0,
+                    1405.0,
+                    621.0,
+                    293.0,
+                    621.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    615.0,
+                    1408.0,
+                    615.0,
+                    1408.0,
+                    652.0,
+                    293.0,
+                    652.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    648.0,
+                    1405.0,
+                    648.0,
+                    1405.0,
+                    680.0,
+                    295.0,
+                    680.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    676.0,
+                    1405.0,
+                    676.0,
+                    1405.0,
+                    712.0,
+                    293.0,
+                    712.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    707.0,
+                    1370.0,
+                    707.0,
+                    1370.0,
+                    743.0,
+                    295.0,
+                    743.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1036.0,
+                    991.0,
+                    1036.0,
+                    991.0,
+                    1072.0,
+                    294.0,
+                    1072.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1065.0,
+                    992.0,
+                    1065.0,
+                    992.0,
+                    1102.0,
+                    294.0,
+                    1102.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1093.0,
+                    994.0,
+                    1093.0,
+                    994.0,
+                    1137.0,
+                    293.0,
+                    1137.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1127.0,
+                    814.0,
+                    1127.0,
+                    814.0,
+                    1164.0,
+                    295.0,
+                    1164.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1529.0,
+                    505.0,
+                    1529.0,
+                    505.0,
+                    1567.0,
+                    293.0,
+                    1567.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    526.0,
+                    1529.0,
+                    602.0,
+                    1529.0,
+                    602.0,
+                    1567.0,
+                    526.0,
+                    1567.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    623.0,
+                    1529.0,
+                    717.0,
+                    1529.0,
+                    717.0,
+                    1567.0,
+                    623.0,
+                    1567.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    836.0,
+                    1529.0,
+                    1403.0,
+                    1529.0,
+                    1403.0,
+                    1567.0,
+                    836.0,
+                    1567.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1564.0,
+                    720.0,
+                    1564.0,
+                    720.0,
+                    1598.0,
+                    296.0,
+                    1598.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    975.0,
+                    1564.0,
+                    1027.0,
+                    1564.0,
+                    1027.0,
+                    1598.0,
+                    975.0,
+                    1598.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1270.0,
+                    1564.0,
+                    1403.0,
+                    1564.0,
+                    1403.0,
+                    1598.0,
+                    1270.0,
+                    1598.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1594.0,
+                    582.0,
+                    1594.0,
+                    582.0,
+                    1628.0,
+                    294.0,
+                    1628.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    613.0,
+                    1594.0,
+                    747.0,
+                    1594.0,
+                    747.0,
+                    1628.0,
+                    613.0,
+                    1628.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    829.0,
+                    1594.0,
+                    839.0,
+                    1594.0,
+                    839.0,
+                    1628.0,
+                    829.0,
+                    1628.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    743.0,
+                    1406.0,
+                    743.0,
+                    1406.0,
+                    780.0,
+                    294.0,
+                    780.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    772.0,
+                    1406.0,
+                    772.0,
+                    1406.0,
+                    811.0,
+                    293.0,
+                    811.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    803.0,
+                    1398.0,
+                    803.0,
+                    1398.0,
+                    840.0,
+                    293.0,
+                    840.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    853.0,
+                    993.0,
+                    853.0,
+                    993.0,
+                    895.0,
+                    295.0,
+                    895.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    884.0,
+                    726.0,
+                    884.0,
+                    726.0,
+                    924.0,
+                    295.0,
+                    924.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    921.0,
+                    500.0,
+                    921.0,
+                    500.0,
+                    960.0,
+                    294.0,
+                    960.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    522.0,
+                    921.0,
+                    992.0,
+                    921.0,
+                    992.0,
+                    960.0,
+                    522.0,
+                    960.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    394.0,
+                    952.0,
+                    663.0,
+                    952.0,
+                    663.0,
+                    990.0,
+                    394.0,
+                    990.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    686.0,
+                    952.0,
+                    992.0,
+                    952.0,
+                    992.0,
+                    990.0,
+                    686.0,
+                    990.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    986.0,
+                    801.0,
+                    986.0,
+                    801.0,
+                    1021.0,
+                    296.0,
+                    1021.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    199.0,
+                    1403.0,
+                    199.0,
+                    1403.0,
+                    242.0,
+                    293.0,
+                    242.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    382.0,
+                    232.0,
+                    727.0,
+                    232.0,
+                    727.0,
+                    270.0,
+                    382.0,
+                    270.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    747.0,
+                    232.0,
+                    1123.0,
+                    232.0,
+                    1123.0,
+                    270.0,
+                    747.0,
+                    270.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1462.0,
+                    1406.0,
+                    1462.0,
+                    1406.0,
+                    1502.0,
+                    294.0,
+                    1502.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1496.0,
+                    890.0,
+                    1496.0,
+                    890.0,
+                    1532.0,
+                    295.0,
+                    1532.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    931.0,
+                    1496.0,
+                    1123.0,
+                    1496.0,
+                    1123.0,
+                    1532.0,
+                    931.0,
+                    1532.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1313.0,
+                    1403.0,
+                    1313.0,
+                    1403.0,
+                    1350.0,
+                    295.0,
+                    1350.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1343.0,
+                    1338.0,
+                    1343.0,
+                    1338.0,
+                    1382.0,
+                    294.0,
+                    1382.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    469.0,
+                    1060.0,
+                    469.0,
+                    1060.0,
+                    510.0,
+                    293.0,
+                    510.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1087.0,
+                    469.0,
+                    1288.0,
+                    469.0,
+                    1288.0,
+                    510.0,
+                    1087.0,
+                    510.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1352.0,
+                    469.0,
+                    1408.0,
+                    469.0,
+                    1408.0,
+                    510.0,
+                    1352.0,
+                    510.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    324.0,
+                    503.0,
+                    702.0,
+                    503.0,
+                    702.0,
+                    539.0,
+                    324.0,
+                    539.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    406.0,
+                    1400.0,
+                    406.0,
+                    1400.0,
+                    438.0,
+                    296.0,
+                    438.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    433.0,
+                    1351.0,
+                    433.0,
+                    1351.0,
+                    470.0,
+                    294.0,
+                    470.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    283.0,
+                    1164.0,
+                    283.0,
+                    1164.0,
+                    319.0,
+                    293.0,
+                    319.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1201.0,
+                    283.0,
+                    1406.0,
+                    283.0,
+                    1406.0,
+                    319.0,
+                    1201.0,
+                    319.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    318.0,
+                    930.0,
+                    318.0,
+                    930.0,
+                    350.0,
+                    295.0,
+                    350.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1381.0,
+                    1408.0,
+                    1381.0,
+                    1408.0,
+                    1421.0,
+                    294.0,
+                    1421.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1413.0,
+                    404.0,
+                    1413.0,
+                    404.0,
+                    1449.0,
+                    296.0,
+                    1449.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    475.0,
+                    1413.0,
+                    551.0,
+                    1413.0,
+                    551.0,
+                    1449.0,
+                    475.0,
+                    1449.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    571.0,
+                    1413.0,
+                    675.0,
+                    1413.0,
+                    675.0,
+                    1449.0,
+                    571.0,
+                    1449.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    747.0,
+                    1413.0,
+                    798.0,
+                    1413.0,
+                    798.0,
+                    1449.0,
+                    747.0,
+                    1449.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    924.0,
+                    1413.0,
+                    981.0,
+                    1413.0,
+                    981.0,
+                    1449.0,
+                    924.0,
+                    1449.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1054.0,
+                    1413.0,
+                    1063.0,
+                    1413.0,
+                    1063.0,
+                    1449.0,
+                    1054.0,
+                    1449.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1228.0,
+                    550.0,
+                    1228.0,
+                    550.0,
+                    1271.0,
+                    294.0,
+                    1271.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    622.0,
+                    1228.0,
+                    699.0,
+                    1228.0,
+                    699.0,
+                    1271.0,
+                    622.0,
+                    1271.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    719.0,
+                    1228.0,
+                    852.0,
+                    1228.0,
+                    852.0,
+                    1271.0,
+                    719.0,
+                    1271.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    872.0,
+                    1228.0,
+                    973.0,
+                    1228.0,
+                    973.0,
+                    1271.0,
+                    872.0,
+                    1271.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1095.0,
+                    1228.0,
+                    1404.0,
+                    1228.0,
+                    1404.0,
+                    1271.0,
+                    1095.0,
+                    1271.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1258.0,
+                    409.0,
+                    1258.0,
+                    409.0,
+                    1299.0,
+                    293.0,
+                    1299.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1162.0,
+                    513.0,
+                    1162.0,
+                    513.0,
+                    1200.0,
+                    293.0,
+                    1200.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    534.0,
+                    1162.0,
+                    590.0,
+                    1162.0,
+                    590.0,
+                    1200.0,
+                    534.0,
+                    1200.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    611.0,
+                    1162.0,
+                    659.0,
+                    1162.0,
+                    659.0,
+                    1200.0,
+                    611.0,
+                    1200.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    680.0,
+                    1162.0,
+                    1406.0,
+                    1162.0,
+                    1406.0,
+                    1200.0,
+                    680.0,
+                    1200.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1195.0,
+                    888.0,
+                    1195.0,
+                    888.0,
+                    1230.0,
+                    294.0,
+                    1230.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    907.0,
+                    1195.0,
+                    918.0,
+                    1195.0,
+                    918.0,
+                    1230.0,
+                    907.0,
+                    1230.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    351.0,
+                    1203.0,
+                    351.0,
+                    1203.0,
+                    390.0,
+                    295.0,
+                    390.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 4,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1577,
+                    1406,
+                    1577,
+                    1406,
+                    1853,
+                    297,
+                    1853
+                ],
+                "score": 0.983
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1259,
+                    1133,
+                    1259,
+                    1133,
+                    1507,
+                    298,
+                    1507
+                ],
+                "score": 0.98
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    733,
+                    1136,
+                    733,
+                    1136,
+                    919,
+                    297,
+                    919
+                ],
+                "score": 0.979
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1044,
+                    1135,
+                    1044,
+                    1135,
+                    1229,
+                    297,
+                    1229
+                ],
+                "score": 0.978
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    300,
+                    1884,
+                    1404,
+                    1884,
+                    1404,
+                    2008,
+                    300,
+                    2008
+                ],
+                "score": 0.977
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    319,
+                    1404,
+                    319,
+                    1404,
+                    503,
+                    297,
+                    503
+                ],
+                "score": 0.975
+            },
+            {
+                "category_id": 3,
+                "poly": [
+                    1154,
+                    740,
+                    1405,
+                    740,
+                    1405,
+                    1210,
+                    1154,
+                    1210
+                ],
+                "score": 0.958
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    295,
+                    588,
+                    1404,
+                    588,
+                    1404,
+                    652,
+                    295,
+                    652
+                ],
+                "score": 0.954
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    651,
+                    519,
+                    1045,
+                    519,
+                    1045,
+                    576,
+                    651,
+                    576
+                ],
+                "score": 0.947
+            },
+            {
+                "category_id": 4,
+                "poly": [
+                    1154,
+                    1225,
+                    1405,
+                    1225,
+                    1405,
+                    1528,
+                    1154,
+                    1528
+                ],
+                "score": 0.946
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    928,
+                    1136,
+                    928,
+                    1136,
+                    1021,
+                    297,
+                    1021
+                ],
+                "score": 0.939
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    457,
+                    665,
+                    1242,
+                    665,
+                    1242,
+                    710,
+                    457,
+                    710
+                ],
+                "score": 0.932
+            },
+            {
+                "category_id": 9,
+                "poly": [
+                    1366,
+                    532,
+                    1400,
+                    532,
+                    1400,
+                    563,
+                    1366,
+                    563
+                ],
+                "score": 0.892
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    289,
+                    249,
+                    1399,
+                    249,
+                    1399,
+                    306,
+                    289,
+                    306
+                ],
+                "score": 0.874
+            },
+            {
+                "category_id": 9,
+                "poly": [
+                    1365,
+                    673,
+                    1400,
+                    673,
+                    1400,
+                    703,
+                    1365,
+                    703
+                ],
+                "score": 0.859
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    291,
+                    201,
+                    1330,
+                    201,
+                    1330,
+                    236,
+                    291,
+                    236
+                ],
+                "score": 0.82
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    840,
+                    2062,
+                    859,
+                    2062,
+                    859,
+                    2085,
+                    840,
+                    2085
+                ],
+                "score": 0.789
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    296,
+                    1514,
+                    1051,
+                    1514,
+                    1051,
+                    1546,
+                    296,
+                    1546
+                ],
+                "score": 0.773
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    296,
+                    1514,
+                    1051,
+                    1514,
+                    1051,
+                    1546,
+                    296,
+                    1546
+                ],
+                "score": 0.097
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    651,
+                    517,
+                    1047,
+                    517,
+                    1047,
+                    577,
+                    651,
+                    577
+                ],
+                "score": 0.94,
+                "latex": "h _ { \\sigma } ^ { t + 1 } = U \\Big ( h _ { \\sigma } ^ { t } , m _ { B } ^ { t } ( \\sigma ) , m _ { \\uparrow } ^ { t + 1 } ( \\sigma ) \\Big ) ."
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    310,
+                    247,
+                    1406,
+                    247,
+                    1406,
+                    308,
+                    310,
+                    308
+                ],
+                "score": 0.93,
+                "latex": "\\begin{array} { r } { n _ { \\mathrm { { B } } } ^ { t + 1 } ( \\sigma ) = \\mathrm { A G G } _ { \\tau \\in \\mathcal { B } ( \\sigma ) } \\left( M _ { B } \\big ( h _ { \\sigma } ^ { t } , h _ { \\tau } ^ { t } \\big ) \\right) \\qquad m _ { \\uparrow } ^ { t + 1 } ( \\sigma ) = \\mathrm { A G G } _ { \\tau \\in \\mathcal { N } _ { \\uparrow } ( \\sigma ) , \\delta \\in \\mathcal { C } ( \\sigma , \\tau ) } \\Big ( M _ { \\uparrow } \\big ( h _ { \\sigma } ^ { t } , h _ { \\tau } ^ { t } , h _ { \\delta } ^ { t } \\big ) \\Big ) . } \\end{array}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    966,
+                    1291,
+                    1133,
+                    1291,
+                    1133,
+                    1325,
+                    966,
+                    1325
+                ],
+                "score": 0.92,
+                "latex": "P f ( A , X ) ="
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    298,
+                    1320,
+                    494,
+                    1320,
+                    494,
+                    1355,
+                    298,
+                    1355
+                ],
+                "score": 0.91,
+                "latex": "f ( P A P ^ { T } , P X )"
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    457,
+                    665,
+                    1240,
+                    665,
+                    1240,
+                    710,
+                    457,
+                    710
+                ],
+                "score": 0.89,
+                "latex": "h _ { X } = \\mathrm { R E A D O U T } ( \\{ h _ { \\sigma } ^ { L } \\} | _ { d i m ( \\sigma ) = 0 } , \\{ \\{ h _ { \\sigma } ^ { L } \\} \\} _ { d i m ( \\sigma ) = 1 } , \\{ \\{ h _ { \\sigma } ^ { L } \\} \\} _ { d i m ( \\sigma ) = 2 } ) ."
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    297,
+                    1106,
+                    367,
+                    1106,
+                    367,
+                    1136,
+                    297,
+                    1136
+                ],
+                "score": 0.89,
+                "latex": "k \\geq 3"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    730,
+                    1446,
+                    803,
+                    1446,
+                    803,
+                    1471,
+                    730,
+                    1471
+                ],
+                "score": 0.88,
+                "latex": "\\sigma \\prec \\tau"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    518,
+                    1292,
+                    539,
+                    1292,
+                    539,
+                    1323,
+                    518,
+                    1323
+                ],
+                "score": 0.85,
+                "latex": "f"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    833,
+                    590,
+                    860,
+                    590,
+                    860,
+                    617,
+                    833,
+                    617
+                ],
+                "score": 0.84,
+                "latex": "X"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    391,
+                    1701,
+                    413,
+                    1701,
+                    413,
+                    1727,
+                    391,
+                    1727
+                ],
+                "score": 0.8,
+                "latex": "L"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1074,
+                    591,
+                    1096,
+                    591,
+                    1096,
+                    617,
+                    1074,
+                    617
+                ],
+                "score": 0.8,
+                "latex": "L"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    346,
+                    1292,
+                    377,
+                    1292,
+                    377,
+                    1319,
+                    346,
+                    1319
+                ],
+                "score": 0.78,
+                "latex": "\\boldsymbol { X }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    633,
+                    1262,
+                    659,
+                    1262,
+                    659,
+                    1289,
+                    633,
+                    1289
+                ],
+                "score": 0.78,
+                "latex": "G"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1243,
+                    1412,
+                    1263,
+                    1412,
+                    1263,
+                    1434,
+                    1243,
+                    1434
+                ],
+                "score": 0.78,
+                "latex": "\\sigma"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    298,
+                    1731,
+                    320,
+                    1731,
+                    320,
+                    1757,
+                    298,
+                    1757
+                ],
+                "score": 0.77,
+                "latex": "L"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    817,
+                    1323,
+                    844,
+                    1323,
+                    844,
+                    1349,
+                    817,
+                    1349
+                ],
+                "score": 0.76,
+                "latex": "_ { r }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1213,
+                    1502,
+                    1232,
+                    1502,
+                    1232,
+                    1525,
+                    1213,
+                    1525
+                ],
+                "score": 0.75,
+                "latex": "\\tau"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1282,
+                    1499,
+                    1299,
+                    1499,
+                    1299,
+                    1525,
+                    1282,
+                    1525
+                ],
+                "score": 0.75,
+                "latex": "\\delta"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1310,
+                    1412,
+                    1328,
+                    1412,
+                    1328,
+                    1435,
+                    1310,
+                    1435
+                ],
+                "score": 0.74,
+                "latex": "\\tau"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    412,
+                    1789,
+                    457,
+                    1789,
+                    457,
+                    1822,
+                    412,
+                    1822
+                ],
+                "score": 0.71,
+                "latex": "5 . { \\overset { \\cdot } { 1 } }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    453,
+                    1671,
+                    475,
+                    1671,
+                    475,
+                    1698,
+                    453,
+                    1698
+                ],
+                "score": 0.71,
+                "latex": "L"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    731,
+                    381,
+                    761,
+                    381,
+                    761,
+                    416,
+                    731,
+                    416
+                ],
+                "score": 0.68,
+                "latex": "\\textcircled{6}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    384,
+                    1821,
+                    430,
+                    1821,
+                    430,
+                    1855,
+                    384,
+                    1855
+                ],
+                "score": 0.66,
+                "latex": "5 . 2"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1112,
+                    1137,
+                    1132,
+                    1137,
+                    1132,
+                    1165,
+                    1112,
+                    1165
+                ],
+                "score": 0.64,
+                "latex": "k"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    922,
+                    1262,
+                    950,
+                    1262,
+                    950,
+                    1289,
+                    922,
+                    1289
+                ],
+                "score": 0.63,
+                "latex": "\\pmb { A }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    410,
+                    1043,
+                    451,
+                    1043,
+                    451,
+                    1080,
+                    410,
+                    1080
+                ],
+                "score": 0.53,
+                "latex": "\\boxed { 1 5 }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    344,
+                    886,
+                    389,
+                    886,
+                    389,
+                    919,
+                    344,
+                    919
+                ],
+                "score": 0.29,
+                "latex": "\\textcircled { 7 4 } |"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1265.0,
+                    804.0,
+                    1291.0,
+                    804.0,
+                    1291.0,
+                    827.0,
+                    1265.0,
+                    827.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1325.0,
+                    1136.0,
+                    1392.0,
+                    1136.0,
+                    1392.0,
+                    1194.0,
+                    1325.0,
+                    1194.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1152.0,
+                    1224.0,
+                    1408.0,
+                    1224.0,
+                    1408.0,
+                    1256.0,
+                    1152.0,
+                    1256.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1152.0,
+                    1255.0,
+                    1405.0,
+                    1255.0,
+                    1405.0,
+                    1288.0,
+                    1152.0,
+                    1288.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1153.0,
+                    1288.0,
+                    1407.0,
+                    1288.0,
+                    1407.0,
+                    1319.0,
+                    1153.0,
+                    1319.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1153.0,
+                    1315.0,
+                    1248.0,
+                    1315.0,
+                    1248.0,
+                    1347.0,
+                    1153.0,
+                    1347.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1265.0,
+                    1315.0,
+                    1412.0,
+                    1315.0,
+                    1412.0,
+                    1350.0,
+                    1265.0,
+                    1350.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1150.0,
+                    1346.0,
+                    1410.0,
+                    1346.0,
+                    1410.0,
+                    1377.0,
+                    1150.0,
+                    1377.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1151.0,
+                    1379.0,
+                    1407.0,
+                    1379.0,
+                    1407.0,
+                    1408.0,
+                    1151.0,
+                    1408.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1152.0,
+                    1406.0,
+                    1242.0,
+                    1406.0,
+                    1242.0,
+                    1439.0,
+                    1152.0,
+                    1439.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1264.0,
+                    1406.0,
+                    1309.0,
+                    1406.0,
+                    1309.0,
+                    1439.0,
+                    1264.0,
+                    1439.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1329.0,
+                    1406.0,
+                    1405.0,
+                    1406.0,
+                    1405.0,
+                    1439.0,
+                    1329.0,
+                    1439.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1151.0,
+                    1434.0,
+                    1407.0,
+                    1434.0,
+                    1407.0,
+                    1473.0,
+                    1151.0,
+                    1473.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1151.0,
+                    1467.0,
+                    1406.0,
+                    1467.0,
+                    1406.0,
+                    1503.0,
+                    1151.0,
+                    1503.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1153.0,
+                    1495.0,
+                    1212.0,
+                    1495.0,
+                    1212.0,
+                    1529.0,
+                    1153.0,
+                    1529.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1233.0,
+                    1495.0,
+                    1281.0,
+                    1495.0,
+                    1281.0,
+                    1529.0,
+                    1233.0,
+                    1529.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1300.0,
+                    1495.0,
+                    1309.0,
+                    1495.0,
+                    1309.0,
+                    1529.0,
+                    1300.0,
+                    1529.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    840.0,
+                    2060.0,
+                    862.0,
+                    2060.0,
+                    862.0,
+                    2091.0,
+                    840.0,
+                    2091.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1510.0,
+                    1055.0,
+                    1510.0,
+                    1055.0,
+                    1552.0,
+                    295.0,
+                    1552.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1574.0,
+                    1404.0,
+                    1574.0,
+                    1404.0,
+                    1616.0,
+                    294.0,
+                    1616.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1609.0,
+                    1404.0,
+                    1609.0,
+                    1404.0,
+                    1642.0,
+                    295.0,
+                    1642.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1637.0,
+                    1409.0,
+                    1637.0,
+                    1409.0,
+                    1676.0,
+                    292.0,
+                    1676.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1667.0,
+                    452.0,
+                    1667.0,
+                    452.0,
+                    1707.0,
+                    294.0,
+                    1707.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    476.0,
+                    1667.0,
+                    1407.0,
+                    1667.0,
+                    1407.0,
+                    1707.0,
+                    476.0,
+                    1707.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1697.0,
+                    390.0,
+                    1697.0,
+                    390.0,
+                    1736.0,
+                    292.0,
+                    1736.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    414.0,
+                    1697.0,
+                    1407.0,
+                    1697.0,
+                    1407.0,
+                    1736.0,
+                    414.0,
+                    1736.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1727.0,
+                    297.0,
+                    1727.0,
+                    297.0,
+                    1766.0,
+                    294.0,
+                    1766.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    321.0,
+                    1727.0,
+                    1406.0,
+                    1727.0,
+                    1406.0,
+                    1766.0,
+                    321.0,
+                    1766.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1758.0,
+                    1408.0,
+                    1758.0,
+                    1408.0,
+                    1796.0,
+                    292.0,
+                    1796.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1788.0,
+                    411.0,
+                    1788.0,
+                    411.0,
+                    1826.0,
+                    296.0,
+                    1826.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    458.0,
+                    1788.0,
+                    1406.0,
+                    1788.0,
+                    1406.0,
+                    1826.0,
+                    458.0,
+                    1826.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1819.0,
+                    383.0,
+                    1819.0,
+                    383.0,
+                    1856.0,
+                    295.0,
+                    1856.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    431.0,
+                    1819.0,
+                    1409.0,
+                    1819.0,
+                    1409.0,
+                    1856.0,
+                    431.0,
+                    1856.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1258.0,
+                    632.0,
+                    1258.0,
+                    632.0,
+                    1295.0,
+                    295.0,
+                    1295.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    660.0,
+                    1258.0,
+                    921.0,
+                    1258.0,
+                    921.0,
+                    1295.0,
+                    660.0,
+                    1295.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    951.0,
+                    1258.0,
+                    1140.0,
+                    1258.0,
+                    1140.0,
+                    1295.0,
+                    951.0,
+                    1295.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1287.0,
+                    345.0,
+                    1287.0,
+                    345.0,
+                    1328.0,
+                    293.0,
+                    1328.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    378.0,
+                    1287.0,
+                    517.0,
+                    1287.0,
+                    517.0,
+                    1328.0,
+                    378.0,
+                    1328.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    540.0,
+                    1287.0,
+                    965.0,
+                    1287.0,
+                    965.0,
+                    1328.0,
+                    540.0,
+                    1328.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1317.0,
+                    297.0,
+                    1317.0,
+                    297.0,
+                    1358.0,
+                    294.0,
+                    1358.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    495.0,
+                    1317.0,
+                    816.0,
+                    1317.0,
+                    816.0,
+                    1358.0,
+                    495.0,
+                    1358.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    845.0,
+                    1317.0,
+                    1137.0,
+                    1317.0,
+                    1137.0,
+                    1358.0,
+                    845.0,
+                    1358.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1351.0,
+                    1135.0,
+                    1351.0,
+                    1135.0,
+                    1385.0,
+                    294.0,
+                    1385.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1383.0,
+                    1135.0,
+                    1383.0,
+                    1135.0,
+                    1415.0,
+                    293.0,
+                    1415.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1411.0,
+                    1136.0,
+                    1411.0,
+                    1136.0,
+                    1449.0,
+                    292.0,
+                    1449.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1443.0,
+                    729.0,
+                    1443.0,
+                    729.0,
+                    1474.0,
+                    294.0,
+                    1474.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    804.0,
+                    1443.0,
+                    1136.0,
+                    1443.0,
+                    1136.0,
+                    1474.0,
+                    804.0,
+                    1474.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1473.0,
+                    890.0,
+                    1473.0,
+                    890.0,
+                    1508.0,
+                    295.0,
+                    1508.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    734.0,
+                    1134.0,
+                    734.0,
+                    1134.0,
+                    770.0,
+                    295.0,
+                    770.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    766.0,
+                    1135.0,
+                    766.0,
+                    1135.0,
+                    799.0,
+                    294.0,
+                    799.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    794.0,
+                    1135.0,
+                    794.0,
+                    1135.0,
+                    831.0,
+                    294.0,
+                    831.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    825.0,
+                    1137.0,
+                    825.0,
+                    1137.0,
+                    862.0,
+                    295.0,
+                    862.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    858.0,
+                    1134.0,
+                    858.0,
+                    1134.0,
+                    892.0,
+                    297.0,
+                    892.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    298.0,
+                    884.0,
+                    343.0,
+                    884.0,
+                    343.0,
+                    922.0,
+                    298.0,
+                    922.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    390.0,
+                    884.0,
+                    1090.0,
+                    884.0,
+                    1090.0,
+                    922.0,
+                    390.0,
+                    922.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1041.0,
+                    409.0,
+                    1041.0,
+                    409.0,
+                    1080.0,
+                    295.0,
+                    1080.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    452.0,
+                    1041.0,
+                    1137.0,
+                    1041.0,
+                    1137.0,
+                    1080.0,
+                    452.0,
+                    1080.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1072.0,
+                    1136.0,
+                    1072.0,
+                    1136.0,
+                    1109.0,
+                    294.0,
+                    1109.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    368.0,
+                    1106.0,
+                    1134.0,
+                    1106.0,
+                    1134.0,
+                    1138.0,
+                    368.0,
+                    1138.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1136.0,
+                    1111.0,
+                    1136.0,
+                    1111.0,
+                    1172.0,
+                    293.0,
+                    1172.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1133.0,
+                    1136.0,
+                    1136.0,
+                    1136.0,
+                    1136.0,
+                    1172.0,
+                    1133.0,
+                    1172.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1163.0,
+                    1135.0,
+                    1163.0,
+                    1135.0,
+                    1202.0,
+                    293.0,
+                    1202.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1197.0,
+                    1136.0,
+                    1197.0,
+                    1136.0,
+                    1231.0,
+                    296.0,
+                    1231.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1882.0,
+                    1404.0,
+                    1882.0,
+                    1404.0,
+                    1920.0,
+                    295.0,
+                    1920.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1912.0,
+                    1408.0,
+                    1912.0,
+                    1408.0,
+                    1950.0,
+                    295.0,
+                    1950.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1945.0,
+                    1408.0,
+                    1945.0,
+                    1408.0,
+                    1980.0,
+                    294.0,
+                    1980.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1973.0,
+                    1405.0,
+                    1973.0,
+                    1405.0,
+                    2013.0,
+                    291.0,
+                    2013.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    321.0,
+                    1405.0,
+                    321.0,
+                    1405.0,
+                    353.0,
+                    296.0,
+                    353.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    351.0,
+                    1406.0,
+                    351.0,
+                    1406.0,
+                    384.0,
+                    293.0,
+                    384.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    379.0,
+                    730.0,
+                    379.0,
+                    730.0,
+                    417.0,
+                    294.0,
+                    417.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    762.0,
+                    379.0,
+                    1406.0,
+                    379.0,
+                    1406.0,
+                    417.0,
+                    762.0,
+                    417.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    412.0,
+                    1402.0,
+                    412.0,
+                    1402.0,
+                    444.0,
+                    295.0,
+                    444.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    441.0,
+                    1402.0,
+                    441.0,
+                    1402.0,
+                    473.0,
+                    295.0,
+                    473.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    471.0,
+                    1342.0,
+                    471.0,
+                    1342.0,
+                    506.0,
+                    294.0,
+                    506.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    588.0,
+                    832.0,
+                    588.0,
+                    832.0,
+                    624.0,
+                    295.0,
+                    624.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    861.0,
+                    588.0,
+                    1073.0,
+                    588.0,
+                    1073.0,
+                    624.0,
+                    861.0,
+                    624.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1097.0,
+                    588.0,
+                    1405.0,
+                    588.0,
+                    1405.0,
+                    624.0,
+                    1097.0,
+                    624.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    619.0,
+                    1377.0,
+                    619.0,
+                    1377.0,
+                    656.0,
+                    294.0,
+                    656.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    926.0,
+                    1136.0,
+                    926.0,
+                    1136.0,
+                    963.0,
+                    294.0,
+                    963.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    957.0,
+                    1136.0,
+                    957.0,
+                    1136.0,
+                    994.0,
+                    296.0,
+                    994.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    298.0,
+                    991.0,
+                    736.0,
+                    991.0,
+                    736.0,
+                    1023.0,
+                    298.0,
+                    1023.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    197.0,
+                    1335.0,
+                    197.0,
+                    1335.0,
+                    243.0,
+                    291.0,
+                    243.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1510.0,
+                    1055.0,
+                    1510.0,
+                    1055.0,
+                    1552.0,
+                    295.0,
+                    1552.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 5,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1310,
+                    1405,
+                    1310,
+                    1405,
+                    1649,
+                    297,
+                    1649
+                ],
+                "score": 0.982
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1792,
+                    1405,
+                    1792,
+                    1405,
+                    2008,
+                    298,
+                    2008
+                ],
+                "score": 0.982
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1042,
+                    1404,
+                    1042,
+                    1404,
+                    1258,
+                    297,
+                    1258
+                ],
+                "score": 0.98
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    835,
+                    1404,
+                    835,
+                    1404,
+                    989,
+                    298,
+                    989
+                ],
+                "score": 0.974
+            },
+            {
+                "category_id": 3,
+                "poly": [
+                    300,
+                    202,
+                    783,
+                    202,
+                    783,
+                    580,
+                    300,
+                    580
+                ],
+                "score": 0.962
+            },
+            {
+                "category_id": 3,
+                "poly": [
+                    821,
+                    202,
+                    1398,
+                    202,
+                    1398,
+                    635,
+                    821,
+                    635
+                ],
+                "score": 0.958
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    296,
+                    597,
+                    788,
+                    597,
+                    788,
+                    738,
+                    296,
+                    738
+                ],
+                "score": 0.949
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    298,
+                    1711,
+                    532,
+                    1711,
+                    532,
+                    1749,
+                    298,
+                    1749
+                ],
+                "score": 0.913
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    812,
+                    654,
+                    1403,
+                    654,
+                    1403,
+                    738,
+                    812,
+                    738
+                ],
+                "score": 0.74
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    842,
+                    2061,
+                    858,
+                    2061,
+                    858,
+                    2084,
+                    842,
+                    2084
+                ],
+                "score": 0.692
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    464,
+                    756,
+                    1230,
+                    756,
+                    1230,
+                    789,
+                    464,
+                    789
+                ],
+                "score": 0.496
+            },
+            {
+                "category_id": 4,
+                "poly": [
+                    464,
+                    756,
+                    1230,
+                    756,
+                    1230,
+                    789,
+                    464,
+                    789
+                ],
+                "score": 0.408
+            },
+            {
+                "category_id": 4,
+                "poly": [
+                    812,
+                    654,
+                    1403,
+                    654,
+                    1403,
+                    738,
+                    812,
+                    738
+                ],
+                "score": 0.299
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    842,
+                    2061,
+                    859,
+                    2061,
+                    859,
+                    2084,
+                    842,
+                    2084
+                ],
+                "score": 0.151
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    679,
+                    1523,
+                    868,
+                    1523,
+                    868,
+                    1559,
+                    679,
+                    1559
+                ],
+                "score": 0.87,
+                "latex": "\\mathcal { O } \\big ( ( | E | ^ { - } | V | \\bar { C } \\big )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1349,
+                    1524,
+                    1374,
+                    1524,
+                    1374,
+                    1551,
+                    1349,
+                    1551
+                ],
+                "score": 0.83,
+                "latex": "C"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    923,
+                    1134,
+                    971,
+                    1134,
+                    971,
+                    1166,
+                    923,
+                    1166
+                ],
+                "score": 0.81,
+                "latex": "\\lVert \\dot { 3 6 } \\rVert"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1281,
+                    1495,
+                    1306,
+                    1495,
+                    1306,
+                    1522,
+                    1281,
+                    1522
+                ],
+                "score": 0.81,
+                "latex": "C"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    987,
+                    1948,
+                    1016,
+                    1948,
+                    1016,
+                    1981,
+                    987,
+                    1981
+                ],
+                "score": 0.8,
+                "latex": "\\mathrm { ~ E ~ }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1081,
+                    1522,
+                    1130,
+                    1522,
+                    1130,
+                    1555,
+                    1081,
+                    1555
+                ],
+                "score": 0.77,
+                "latex": "\\pmb { \\Vert 2 6 \\Vert }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1031,
+                    1103,
+                    1079,
+                    1103,
+                    1079,
+                    1135,
+                    1031,
+                    1135
+                ],
+                "score": 0.76,
+                "latex": "\\pmb { \\mathbb { \\left\\| 6 3 \\right\\| } }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1130,
+                    711,
+                    1148,
+                    711,
+                    1148,
+                    735,
+                    1130,
+                    735
+                ],
+                "score": 0.74,
+                "latex": "k"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1131,
+                    1072,
+                    1221,
+                    1072,
+                    1221,
+                    1105,
+                    1131,
+                    1105
+                ],
+                "score": 0.7,
+                "latex": "\\pm \\boxed { 1 3 } \\boxed { 2 5 } \\parallel"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    965,
+                    1522,
+                    1015,
+                    1522,
+                    1015,
+                    1559,
+                    965,
+                    1559
+                ],
+                "score": 0.7,
+                "latex": "| V | )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    821,
+                    1886,
+                    869,
+                    1886,
+                    869,
+                    1916,
+                    821,
+                    1916
+                ],
+                "score": 0.56,
+                "latex": "\\pmb { \\Vert 7 4 \\Vert }"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    528.0,
+                    209.0,
+                    554.0,
+                    209.0,
+                    554.0,
+                    224.0,
+                    528.0,
+                    224.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    604.0,
+                    209.0,
+                    629.0,
+                    209.0,
+                    629.0,
+                    224.0,
+                    604.0,
+                    224.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    316.0,
+                    250.0,
+                    344.0,
+                    250.0,
+                    344.0,
+                    272.0,
+                    316.0,
+                    272.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    316.0,
+                    308.0,
+                    343.0,
+                    308.0,
+                    343.0,
+                    327.0,
+                    316.0,
+                    327.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    303.0,
+                    362.0,
+                    322.0,
+                    362.0,
+                    322.0,
+                    429.0,
+                    303.0,
+                    429.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    316.0,
+                    365.0,
+                    342.0,
+                    365.0,
+                    342.0,
+                    384.0,
+                    316.0,
+                    384.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    317.0,
+                    423.0,
+                    343.0,
+                    423.0,
+                    343.0,
+                    441.0,
+                    317.0,
+                    441.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    316.0,
+                    480.0,
+                    343.0,
+                    480.0,
+                    343.0,
+                    499.0,
+                    316.0,
+                    499.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    316.0,
+                    537.0,
+                    344.0,
+                    537.0,
+                    344.0,
+                    555.0,
+                    316.0,
+                    555.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    364.0,
+                    549.0,
+                    386.0,
+                    549.0,
+                    386.0,
+                    565.0,
+                    364.0,
+                    565.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    399.0,
+                    548.0,
+                    424.0,
+                    548.0,
+                    424.0,
+                    566.0,
+                    399.0,
+                    566.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    436.0,
+                    548.0,
+                    463.0,
+                    548.0,
+                    463.0,
+                    566.0,
+                    436.0,
+                    566.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    475.0,
+                    548.0,
+                    500.0,
+                    548.0,
+                    500.0,
+                    566.0,
+                    475.0,
+                    566.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    512.0,
+                    548.0,
+                    537.0,
+                    548.0,
+                    537.0,
+                    566.0,
+                    512.0,
+                    566.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    550.0,
+                    548.0,
+                    575.0,
+                    548.0,
+                    575.0,
+                    566.0,
+                    550.0,
+                    566.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    589.0,
+                    549.0,
+                    610.0,
+                    549.0,
+                    610.0,
+                    565.0,
+                    589.0,
+                    565.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    624.0,
+                    547.0,
+                    651.0,
+                    547.0,
+                    651.0,
+                    566.0,
+                    624.0,
+                    566.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    662.0,
+                    548.0,
+                    687.0,
+                    548.0,
+                    687.0,
+                    566.0,
+                    662.0,
+                    566.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    699.0,
+                    548.0,
+                    725.0,
+                    548.0,
+                    725.0,
+                    566.0,
+                    699.0,
+                    566.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    739.0,
+                    549.0,
+                    760.0,
+                    549.0,
+                    760.0,
+                    565.0,
+                    739.0,
+                    565.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    528.0,
+                    561.0,
+                    596.0,
+                    561.0,
+                    596.0,
+                    584.0,
+                    528.0,
+                    584.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    922.0,
+                    205.0,
+                    992.0,
+                    205.0,
+                    992.0,
+                    224.0,
+                    922.0,
+                    224.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1025.0,
+                    205.0,
+                    1093.0,
+                    205.0,
+                    1093.0,
+                    224.0,
+                    1025.0,
+                    224.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1127.0,
+                    205.0,
+                    1194.0,
+                    205.0,
+                    1194.0,
+                    224.0,
+                    1127.0,
+                    224.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1223.0,
+                    206.0,
+                    1261.0,
+                    206.0,
+                    1261.0,
+                    223.0,
+                    1223.0,
+                    223.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    921.0,
+                    219.0,
+                    965.0,
+                    219.0,
+                    965.0,
+                    238.0,
+                    921.0,
+                    238.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1023.0,
+                    219.0,
+                    1067.0,
+                    219.0,
+                    1067.0,
+                    238.0,
+                    1023.0,
+                    238.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1127.0,
+                    219.0,
+                    1170.0,
+                    219.0,
+                    1170.0,
+                    238.0,
+                    1127.0,
+                    238.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1334.0,
+                    248.0,
+                    1364.0,
+                    248.0,
+                    1364.0,
+                    271.0,
+                    1334.0,
+                    271.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1331.0,
+                    293.0,
+                    1371.0,
+                    293.0,
+                    1371.0,
+                    313.0,
+                    1331.0,
+                    313.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1376.0,
+                    352.0,
+                    1399.0,
+                    352.0,
+                    1399.0,
+                    444.0,
+                    1376.0,
+                    444.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1336.0,
+                    518.0,
+                    1349.0,
+                    518.0,
+                    1349.0,
+                    531.0,
+                    1336.0,
+                    531.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1050.0,
+                    612.0,
+                    1108.0,
+                    612.0,
+                    1108.0,
+                    639.0,
+                    1050.0,
+                    639.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1331.0,
+                    334.0,
+                    1375.0,
+                    334.0,
+                    1375.0,
+                    358.0,
+                    1331.0,
+                    358.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1331.75,
+                    378.5,
+                    1373.75,
+                    378.5,
+                    1373.75,
+                    399.5,
+                    1331.75,
+                    399.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1330.75,
+                    420.0,
+                    1374.75,
+                    420.0,
+                    1374.75,
+                    444.0,
+                    1330.75,
+                    444.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1330.0,
+                    463.0,
+                    1375.0,
+                    463.0,
+                    1375.0,
+                    487.5,
+                    1330.0,
+                    487.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    875.75,
+                    573.0,
+                    920.75,
+                    573.0,
+                    920.75,
+                    589.0,
+                    875.75,
+                    589.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    930.0,
+                    574.0,
+                    979.0,
+                    574.0,
+                    979.0,
+                    592.5,
+                    930.0,
+                    592.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    983.75,
+                    574.0,
+                    1031.75,
+                    574.0,
+                    1031.75,
+                    592.5,
+                    983.75,
+                    592.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1037.0,
+                    574.0,
+                    1083.0,
+                    574.0,
+                    1083.0,
+                    593.0,
+                    1037.0,
+                    593.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1088.25,
+                    573.5,
+                    1137.25,
+                    573.5,
+                    1137.25,
+                    592.5,
+                    1088.25,
+                    592.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1143.0,
+                    574.0,
+                    1188.0,
+                    574.0,
+                    1188.0,
+                    591.5,
+                    1143.0,
+                    591.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1195.25,
+                    574.5,
+                    1242.25,
+                    574.5,
+                    1242.25,
+                    590.5,
+                    1195.25,
+                    590.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1248.25,
+                    573.5,
+                    1294.25,
+                    573.5,
+                    1294.25,
+                    591.5,
+                    1248.25,
+                    591.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1301.25,
+                    574.5,
+                    1348.25,
+                    574.5,
+                    1348.25,
+                    590.5,
+                    1301.25,
+                    590.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    288.0,
+                    1705.0,
+                    538.0,
+                    1705.0,
+                    538.0,
+                    1759.0,
+                    288.0,
+                    1759.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    840.0,
+                    2059.0,
+                    860.0,
+                    2059.0,
+                    860.0,
+                    2092.0,
+                    840.0,
+                    2092.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    464.0,
+                    754.0,
+                    1234.0,
+                    754.0,
+                    1234.0,
+                    793.0,
+                    464.0,
+                    793.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    811.0,
+                    653.0,
+                    1404.0,
+                    653.0,
+                    1404.0,
+                    685.0,
+                    811.0,
+                    685.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    811.0,
+                    682.0,
+                    1403.0,
+                    682.0,
+                    1403.0,
+                    713.0,
+                    811.0,
+                    713.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    811.0,
+                    708.0,
+                    1129.0,
+                    708.0,
+                    1129.0,
+                    740.0,
+                    811.0,
+                    740.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1149.0,
+                    708.0,
+                    1396.0,
+                    708.0,
+                    1396.0,
+                    740.0,
+                    1149.0,
+                    740.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    840.0,
+                    2060.0,
+                    861.0,
+                    2060.0,
+                    861.0,
+                    2092.0,
+                    840.0,
+                    2092.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1310.0,
+                    1410.0,
+                    1310.0,
+                    1410.0,
+                    1349.0,
+                    295.0,
+                    1349.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1342.0,
+                    1403.0,
+                    1342.0,
+                    1403.0,
+                    1377.0,
+                    295.0,
+                    1377.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1373.0,
+                    1405.0,
+                    1373.0,
+                    1405.0,
+                    1408.0,
+                    295.0,
+                    1408.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1402.0,
+                    1405.0,
+                    1402.0,
+                    1405.0,
+                    1437.0,
+                    292.0,
+                    1437.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1432.0,
+                    1406.0,
+                    1432.0,
+                    1406.0,
+                    1468.0,
+                    291.0,
+                    1468.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1463.0,
+                    1403.0,
+                    1463.0,
+                    1403.0,
+                    1498.0,
+                    295.0,
+                    1498.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1494.0,
+                    1280.0,
+                    1494.0,
+                    1280.0,
+                    1527.0,
+                    294.0,
+                    1527.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1307.0,
+                    1494.0,
+                    1406.0,
+                    1494.0,
+                    1406.0,
+                    1527.0,
+                    1307.0,
+                    1527.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1522.0,
+                    678.0,
+                    1522.0,
+                    678.0,
+                    1559.0,
+                    292.0,
+                    1559.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    869.0,
+                    1522.0,
+                    964.0,
+                    1522.0,
+                    964.0,
+                    1559.0,
+                    869.0,
+                    1559.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1016.0,
+                    1522.0,
+                    1080.0,
+                    1522.0,
+                    1080.0,
+                    1559.0,
+                    1016.0,
+                    1559.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1131.0,
+                    1522.0,
+                    1348.0,
+                    1522.0,
+                    1348.0,
+                    1559.0,
+                    1131.0,
+                    1559.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1375.0,
+                    1522.0,
+                    1407.0,
+                    1522.0,
+                    1407.0,
+                    1559.0,
+                    1375.0,
+                    1559.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1553.0,
+                    1405.0,
+                    1553.0,
+                    1405.0,
+                    1592.0,
+                    294.0,
+                    1592.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1583.0,
+                    1407.0,
+                    1583.0,
+                    1407.0,
+                    1622.0,
+                    294.0,
+                    1622.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1614.0,
+                    954.0,
+                    1614.0,
+                    954.0,
+                    1652.0,
+                    294.0,
+                    1652.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1789.0,
+                    1406.0,
+                    1789.0,
+                    1406.0,
+                    1832.0,
+                    292.0,
+                    1832.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1822.0,
+                    1406.0,
+                    1822.0,
+                    1406.0,
+                    1858.0,
+                    293.0,
+                    1858.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1854.0,
+                    1405.0,
+                    1854.0,
+                    1405.0,
+                    1889.0,
+                    295.0,
+                    1889.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1880.0,
+                    820.0,
+                    1880.0,
+                    820.0,
+                    1922.0,
+                    292.0,
+                    1922.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    870.0,
+                    1880.0,
+                    1406.0,
+                    1880.0,
+                    1406.0,
+                    1922.0,
+                    870.0,
+                    1922.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1914.0,
+                    1405.0,
+                    1914.0,
+                    1405.0,
+                    1948.0,
+                    295.0,
+                    1948.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1947.0,
+                    986.0,
+                    1947.0,
+                    986.0,
+                    1979.0,
+                    293.0,
+                    1979.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1017.0,
+                    1947.0,
+                    1408.0,
+                    1947.0,
+                    1408.0,
+                    1979.0,
+                    1017.0,
+                    1979.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1977.0,
+                    1408.0,
+                    1977.0,
+                    1408.0,
+                    2009.0,
+                    294.0,
+                    2009.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1041.0,
+                    1405.0,
+                    1041.0,
+                    1405.0,
+                    1078.0,
+                    295.0,
+                    1078.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1073.0,
+                    1130.0,
+                    1073.0,
+                    1130.0,
+                    1107.0,
+                    293.0,
+                    1107.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1222.0,
+                    1073.0,
+                    1405.0,
+                    1073.0,
+                    1405.0,
+                    1107.0,
+                    1222.0,
+                    1107.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1104.0,
+                    1030.0,
+                    1104.0,
+                    1030.0,
+                    1139.0,
+                    294.0,
+                    1139.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1080.0,
+                    1104.0,
+                    1404.0,
+                    1104.0,
+                    1404.0,
+                    1139.0,
+                    1080.0,
+                    1139.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1134.0,
+                    922.0,
+                    1134.0,
+                    922.0,
+                    1167.0,
+                    292.0,
+                    1167.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    972.0,
+                    1134.0,
+                    1405.0,
+                    1134.0,
+                    1405.0,
+                    1167.0,
+                    972.0,
+                    1167.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1162.0,
+                    1404.0,
+                    1162.0,
+                    1404.0,
+                    1199.0,
+                    292.0,
+                    1199.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1193.0,
+                    1405.0,
+                    1193.0,
+                    1405.0,
+                    1230.0,
+                    294.0,
+                    1230.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1225.0,
+                    1272.0,
+                    1225.0,
+                    1272.0,
+                    1260.0,
+                    295.0,
+                    1260.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    834.0,
+                    1406.0,
+                    834.0,
+                    1406.0,
+                    871.0,
+                    294.0,
+                    871.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    866.0,
+                    1402.0,
+                    866.0,
+                    1402.0,
+                    899.0,
+                    296.0,
+                    899.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    897.0,
+                    1402.0,
+                    897.0,
+                    1402.0,
+                    931.0,
+                    294.0,
+                    931.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    925.0,
+                    1405.0,
+                    925.0,
+                    1405.0,
+                    962.0,
+                    294.0,
+                    962.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    951.0,
+                    556.0,
+                    951.0,
+                    556.0,
+                    994.0,
+                    294.0,
+                    994.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    597.0,
+                    793.0,
+                    597.0,
+                    793.0,
+                    628.0,
+                    293.0,
+                    628.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    625.0,
+                    791.0,
+                    625.0,
+                    791.0,
+                    656.0,
+                    294.0,
+                    656.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    652.0,
+                    793.0,
+                    652.0,
+                    793.0,
+                    683.0,
+                    294.0,
+                    683.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    680.0,
+                    788.0,
+                    680.0,
+                    788.0,
+                    712.0,
+                    296.0,
+                    712.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    708.0,
+                    760.0,
+                    708.0,
+                    760.0,
+                    740.0,
+                    295.0,
+                    740.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    811.0,
+                    653.0,
+                    1404.0,
+                    653.0,
+                    1404.0,
+                    685.0,
+                    811.0,
+                    685.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    811.0,
+                    682.0,
+                    1403.0,
+                    682.0,
+                    1403.0,
+                    713.0,
+                    811.0,
+                    713.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    811.0,
+                    708.0,
+                    1129.0,
+                    708.0,
+                    1129.0,
+                    740.0,
+                    811.0,
+                    740.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1149.0,
+                    708.0,
+                    1396.0,
+                    708.0,
+                    1396.0,
+                    740.0,
+                    1149.0,
+                    740.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    464.0,
+                    754.0,
+                    1234.0,
+                    754.0,
+                    1234.0,
+                    793.0,
+                    464.0,
+                    793.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 6,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1200,
+                    1405,
+                    1200,
+                    1405,
+                    1605,
+                    297,
+                    1605
+                ],
+                "score": 0.982
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    767,
+                    1405,
+                    767,
+                    1405,
+                    1164,
+                    297,
+                    1164
+                ],
+                "score": 0.982
+            },
+            {
+                "category_id": 5,
+                "poly": [
+                    301,
+                    447,
+                    770,
+                    447,
+                    770,
+                    603,
+                    301,
+                    603
+                ],
+                "score": 0.978,
+                "html": "<table><tr><td>Method</td><td>Mean</td><td>Min</td><td>Max</td></tr><tr><td>MP-GNNs</td><td>10.000±0.000</td><td>10.000</td><td>10.000</td></tr><tr><td>RingGNN</td><td>10.000±0.000</td><td>10.000</td><td>10.000</td></tr><tr><td>3WLGNN</td><td>97.800±10.916</td><td>30.000</td><td>100.000</td></tr><tr><td>CIN (Ours)</td><td>100.000±0.000</td><td>100.000</td><td>100.000</td></tr></table>"
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1701,
+                    1406,
+                    1701,
+                    1406,
+                    1946,
+                    298,
+                    1946
+                ],
+                "score": 0.977
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    296,
+                    259,
+                    1405,
+                    259,
+                    1405,
+                    383,
+                    296,
+                    383
+                ],
+                "score": 0.972
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    793,
+                    397,
+                    1403,
+                    397,
+                    1403,
+                    608,
+                    793,
+                    608
+                ],
+                "score": 0.944
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    609,
+                    1404,
+                    609,
+                    1404,
+                    731,
+                    298,
+                    731
+                ],
+                "score": 0.917
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    300,
+                    1644,
+                    743,
+                    1644,
+                    743,
+                    1677,
+                    300,
+                    1677
+                ],
+                "score": 0.916
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    299,
+                    203,
+                    631,
+                    203,
+                    631,
+                    235,
+                    299,
+                    235
+                ],
+                "score": 0.913
+            },
+            {
+                "category_id": 6,
+                "poly": [
+                    306,
+                    398,
+                    760,
+                    398,
+                    760,
+                    429,
+                    306,
+                    429
+                ],
+                "score": 0.903
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    327,
+                    1976,
+                    1181,
+                    1976,
+                    1181,
+                    2010,
+                    327,
+                    2010
+                ],
+                "score": 0.9
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    840,
+                    2062,
+                    858,
+                    2062,
+                    858,
+                    2084,
+                    840,
+                    2084
+                ],
+                "score": 0.802
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    524,
+                    1510,
+                    607,
+                    1510,
+                    607,
+                    1540,
+                    524,
+                    1540
+                ],
+                "score": 0.92,
+                "latex": "k \\geq 2 4"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    850,
+                    1570,
+                    894,
+                    1570,
+                    894,
+                    1609,
+                    850,
+                    1609
+                ],
+                "score": 0.92,
+                "latex": "\\left\\lfloor { \\frac { k } { 2 } } \\right\\rfloor"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    712,
+                    1446,
+                    756,
+                    1446,
+                    756,
+                    1484,
+                    712,
+                    1484
+                ],
+                "score": 0.91,
+                "latex": "\\left\\lfloor { \\frac { k } { 2 } } \\right\\rfloor"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1084,
+                    890,
+                    1213,
+                    890,
+                    1213,
+                    921,
+                    1084,
+                    921
+                ],
+                "score": 0.9,
+                "latex": "k = 4 , 5 , 6"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    363,
+                    1795,
+                    437,
+                    1795,
+                    437,
+                    1821,
+                    363,
+                    1821
+                ],
+                "score": 0.9,
+                "latex": "k = 6"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    846,
+                    1291,
+                    890,
+                    1291,
+                    890,
+                    1331,
+                    846,
+                    1331
+                ],
+                "score": 0.9,
+                "latex": "\\lfloor \\frac { k } { 2 } \\rfloor"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    945,
+                    1042,
+                    1015,
+                    1042,
+                    1015,
+                    1070,
+                    945,
+                    1070
+                ],
+                "score": 0.9,
+                "latex": "k = 6"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1050,
+                    488,
+                    1119,
+                    488,
+                    1119,
+                    516,
+                    1050,
+                    516
+                ],
+                "score": 0.89,
+                "latex": "k = 8"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1035,
+                    828,
+                    1112,
+                    828,
+                    1112,
+                    860,
+                    1035,
+                    860
+                ],
+                "score": 0.86,
+                "latex": "\\bar { \\bigstar } \\bar { \\bigstar } 0 \\bigstar"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    567,
+                    1072,
+                    629,
+                    1072,
+                    629,
+                    1102,
+                    567,
+                    1102
+                ],
+                "score": 0.86,
+                "latex": "0 . 0 \\%"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    948,
+                    259,
+                    998,
+                    259,
+                    998,
+                    291,
+                    948,
+                    291
+                ],
+                "score": 0.83,
+                "latex": " { \\mathbb { B } } 5 7  { \\mathbb { I } }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    631,
+                    1541,
+                    694,
+                    1541,
+                    694,
+                    1571,
+                    631,
+                    1571
+                ],
+                "score": 0.82,
+                "latex": "( \\geq 1 2"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1379,
+                    1103,
+                    1397,
+                    1103,
+                    1397,
+                    1129,
+                    1379,
+                    1129
+                ],
+                "score": 0.79,
+                "latex": "k"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    942,
+                    1701,
+                    992,
+                    1701,
+                    992,
+                    1733,
+                    942,
+                    1733
+                ],
+                "score": 0.79,
+                "latex": "\\begin{array} { r l } {  { \\bigl \\| \\boldsymbol { 5 } 6 \\bigr \\| } } \\end{array}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    925,
+                    1481,
+                    943,
+                    1481,
+                    943,
+                    1507,
+                    925,
+                    1507
+                ],
+                "score": 0.78,
+                "latex": "k"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    877,
+                    1263,
+                    895,
+                    1263,
+                    895,
+                    1289,
+                    877,
+                    1289
+                ],
+                "score": 0.76,
+                "latex": "k"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1203,
+                    1417,
+                    1222,
+                    1417,
+                    1222,
+                    1444,
+                    1203,
+                    1444
+                ],
+                "score": 0.76,
+                "latex": "k"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    871,
+                    1043,
+                    889,
+                    1043,
+                    889,
+                    1069,
+                    871,
+                    1069
+                ],
+                "score": 0.74,
+                "latex": "k"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    427,
+                    1823,
+                    476,
+                    1823,
+                    476,
+                    1855,
+                    427,
+                    1855
+                ],
+                "score": 0.73,
+                "latex": "\\textcircled { 7 4 }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    713,
+                    858,
+                    748,
+                    858,
+                    748,
+                    891,
+                    713,
+                    891
+                ],
+                "score": 0.69,
+                "latex": "\\textcircled { 8 } \\textcircled { 1 8 }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1256,
+                    698,
+                    1347,
+                    698,
+                    1347,
+                    731,
+                    1256,
+                    731
+                ],
+                "score": 0.6,
+                "latex": "\\dot { \\mathbb { B } } 4 \\vert 5 3 \\vert "
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    898,
+                    290,
+                    946,
+                    290,
+                    946,
+                    322,
+                    898,
+                    322
+                ],
+                "score": 0.6,
+                "latex": "\\mathbb { \\left[ \\left[ 2 4 \\right] \\right] }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    588,
+                    1732,
+                    636,
+                    1732,
+                    636,
+                    1763,
+                    588,
+                    1763
+                ],
+                "score": 0.56,
+                "latex": "\\mathbb { 2 3 } \\mathbb { I } ,"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1364,
+                    979,
+                    1403,
+                    979,
+                    1403,
+                    1015,
+                    1364,
+                    1015
+                ],
+                "score": 0.55,
+                "latex": "\\boxed { 7 6 }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    738,
+                    766,
+                    787,
+                    766,
+                    787,
+                    800,
+                    738,
+                    800
+                ],
+                "score": 0.52,
+                "latex": "\\mathbb { \\ m }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    372,
+                    1415,
+                    410,
+                    1415,
+                    410,
+                    1449,
+                    372,
+                    1449
+                ],
+                "score": 0.42,
+                "latex": "\\mathrm { 7 a }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    652,
+                    767,
+                    687,
+                    767,
+                    687,
+                    800,
+                    652,
+                    800
+                ],
+                "score": 0.37,
+                "latex": "\\textcircled { 8 } \\textcircled { 1 8 }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    497,
+                    638,
+                    544,
+                    638,
+                    544,
+                    670,
+                    497,
+                    670
+                ],
+                "score": 0.37,
+                "latex": "\\pmb { \\Vert 7 4 \\Vert }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    325,
+                    1824,
+                    364,
+                    1824,
+                    364,
+                    1853,
+                    325,
+                    1853
+                ],
+                "score": 0.34,
+                "latex": "\\mathrm { X u }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    841,
+                    1387,
+                    870,
+                    1387,
+                    870,
+                    1414,
+                    841,
+                    1414
+                ],
+                "score": 0.34,
+                "latex": "1 \\mathrm { - }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1348,
+                    607,
+                    1397,
+                    607,
+                    1397,
+                    641,
+                    1348,
+                    641
+                ],
+                "score": 0.31,
+                "latex": "\\mathbb { \\lVert \\rVert }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    384,
+                    1448,
+                    433,
+                    1448,
+                    433,
+                    1480,
+                    384,
+                    1480
+                ],
+                "score": 0.28,
+                "latex": "\\overline { { [ 7 4 ] } }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    397,
+                    1762,
+                    449,
+                    1762,
+                    449,
+                    1794,
+                    397,
+                    1794
+                ],
+                "score": 0.26,
+                "latex": "\\mathbb { H } \\mathbb { Q } \\mathbb { I } ,"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1639.0,
+                    746.0,
+                    1639.0,
+                    746.0,
+                    1686.0,
+                    292.0,
+                    1686.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    200.0,
+                    632.0,
+                    200.0,
+                    632.0,
+                    240.0,
+                    294.0,
+                    240.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    301.0,
+                    392.0,
+                    762.0,
+                    392.0,
+                    762.0,
+                    434.0,
+                    301.0,
+                    434.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    329.0,
+                    1968.0,
+                    1184.0,
+                    1968.0,
+                    1184.0,
+                    2012.0,
+                    329.0,
+                    2012.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    836.0,
+                    2058.0,
+                    859.0,
+                    2058.0,
+                    859.0,
+                    2089.0,
+                    836.0,
+                    2089.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1199.0,
+                    1406.0,
+                    1199.0,
+                    1406.0,
+                    1236.0,
+                    295.0,
+                    1236.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1231.0,
+                    1405.0,
+                    1231.0,
+                    1405.0,
+                    1265.0,
+                    295.0,
+                    1265.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1260.0,
+                    876.0,
+                    1260.0,
+                    876.0,
+                    1298.0,
+                    292.0,
+                    1298.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    896.0,
+                    1260.0,
+                    1406.0,
+                    1260.0,
+                    1406.0,
+                    1298.0,
+                    896.0,
+                    1298.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1292.0,
+                    845.0,
+                    1292.0,
+                    845.0,
+                    1330.0,
+                    294.0,
+                    1330.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    891.0,
+                    1292.0,
+                    1405.0,
+                    1292.0,
+                    1405.0,
+                    1330.0,
+                    891.0,
+                    1330.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1324.0,
+                    1405.0,
+                    1324.0,
+                    1405.0,
+                    1360.0,
+                    294.0,
+                    1360.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1354.0,
+                    1403.0,
+                    1354.0,
+                    1403.0,
+                    1388.0,
+                    295.0,
+                    1388.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1386.0,
+                    840.0,
+                    1386.0,
+                    840.0,
+                    1419.0,
+                    295.0,
+                    1419.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    871.0,
+                    1386.0,
+                    1405.0,
+                    1386.0,
+                    1405.0,
+                    1419.0,
+                    871.0,
+                    1419.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1413.0,
+                    371.0,
+                    1413.0,
+                    371.0,
+                    1454.0,
+                    293.0,
+                    1454.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    411.0,
+                    1413.0,
+                    1202.0,
+                    1413.0,
+                    1202.0,
+                    1454.0,
+                    411.0,
+                    1454.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1223.0,
+                    1413.0,
+                    1406.0,
+                    1413.0,
+                    1406.0,
+                    1454.0,
+                    1223.0,
+                    1454.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1445.0,
+                    383.0,
+                    1445.0,
+                    383.0,
+                    1486.0,
+                    292.0,
+                    1486.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    434.0,
+                    1445.0,
+                    711.0,
+                    1445.0,
+                    711.0,
+                    1486.0,
+                    434.0,
+                    1486.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    757.0,
+                    1445.0,
+                    1407.0,
+                    1445.0,
+                    1407.0,
+                    1486.0,
+                    757.0,
+                    1486.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1479.0,
+                    924.0,
+                    1479.0,
+                    924.0,
+                    1513.0,
+                    294.0,
+                    1513.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    944.0,
+                    1479.0,
+                    1405.0,
+                    1479.0,
+                    1405.0,
+                    1513.0,
+                    944.0,
+                    1513.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1508.0,
+                    523.0,
+                    1508.0,
+                    523.0,
+                    1546.0,
+                    294.0,
+                    1546.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    608.0,
+                    1508.0,
+                    1406.0,
+                    1508.0,
+                    1406.0,
+                    1546.0,
+                    608.0,
+                    1546.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1540.0,
+                    630.0,
+                    1540.0,
+                    630.0,
+                    1574.0,
+                    295.0,
+                    1574.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    695.0,
+                    1540.0,
+                    1406.0,
+                    1540.0,
+                    1406.0,
+                    1574.0,
+                    695.0,
+                    1574.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1571.0,
+                    849.0,
+                    1571.0,
+                    849.0,
+                    1609.0,
+                    295.0,
+                    1609.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    895.0,
+                    1571.0,
+                    981.0,
+                    1571.0,
+                    981.0,
+                    1609.0,
+                    895.0,
+                    1609.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    765.0,
+                    651.0,
+                    765.0,
+                    651.0,
+                    804.0,
+                    292.0,
+                    804.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    688.0,
+                    765.0,
+                    737.0,
+                    765.0,
+                    737.0,
+                    804.0,
+                    688.0,
+                    804.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    788.0,
+                    765.0,
+                    1405.0,
+                    765.0,
+                    1405.0,
+                    804.0,
+                    788.0,
+                    804.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    796.0,
+                    1405.0,
+                    796.0,
+                    1405.0,
+                    833.0,
+                    294.0,
+                    833.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    824.0,
+                    1034.0,
+                    824.0,
+                    1034.0,
+                    867.0,
+                    292.0,
+                    867.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1113.0,
+                    824.0,
+                    1407.0,
+                    824.0,
+                    1407.0,
+                    867.0,
+                    1113.0,
+                    867.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    857.0,
+                    712.0,
+                    857.0,
+                    712.0,
+                    893.0,
+                    295.0,
+                    893.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    749.0,
+                    857.0,
+                    1402.0,
+                    857.0,
+                    1402.0,
+                    893.0,
+                    749.0,
+                    893.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    890.0,
+                    1083.0,
+                    890.0,
+                    1083.0,
+                    923.0,
+                    296.0,
+                    923.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1214.0,
+                    890.0,
+                    1406.0,
+                    890.0,
+                    1406.0,
+                    923.0,
+                    1214.0,
+                    923.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    919.0,
+                    1407.0,
+                    919.0,
+                    1407.0,
+                    954.0,
+                    292.0,
+                    954.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    951.0,
+                    1406.0,
+                    951.0,
+                    1406.0,
+                    984.0,
+                    296.0,
+                    984.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    980.0,
+                    1363.0,
+                    980.0,
+                    1363.0,
+                    1016.0,
+                    295.0,
+                    1016.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1009.0,
+                    1405.0,
+                    1009.0,
+                    1405.0,
+                    1047.0,
+                    294.0,
+                    1047.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1039.0,
+                    870.0,
+                    1039.0,
+                    870.0,
+                    1077.0,
+                    292.0,
+                    1077.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    890.0,
+                    1039.0,
+                    944.0,
+                    1039.0,
+                    944.0,
+                    1077.0,
+                    890.0,
+                    1077.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1016.0,
+                    1039.0,
+                    1405.0,
+                    1039.0,
+                    1405.0,
+                    1077.0,
+                    1016.0,
+                    1077.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1070.0,
+                    566.0,
+                    1070.0,
+                    566.0,
+                    1107.0,
+                    295.0,
+                    1107.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    630.0,
+                    1070.0,
+                    1405.0,
+                    1070.0,
+                    1405.0,
+                    1107.0,
+                    630.0,
+                    1107.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1099.0,
+                    1378.0,
+                    1099.0,
+                    1378.0,
+                    1136.0,
+                    294.0,
+                    1136.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1398.0,
+                    1099.0,
+                    1410.0,
+                    1099.0,
+                    1410.0,
+                    1136.0,
+                    1398.0,
+                    1136.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1130.0,
+                    961.0,
+                    1130.0,
+                    961.0,
+                    1170.0,
+                    294.0,
+                    1170.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1701.0,
+                    941.0,
+                    1701.0,
+                    941.0,
+                    1735.0,
+                    295.0,
+                    1735.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    993.0,
+                    1701.0,
+                    1404.0,
+                    1701.0,
+                    1404.0,
+                    1735.0,
+                    993.0,
+                    1735.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1730.0,
+                    587.0,
+                    1730.0,
+                    587.0,
+                    1767.0,
+                    293.0,
+                    1767.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    637.0,
+                    1730.0,
+                    1407.0,
+                    1730.0,
+                    1407.0,
+                    1767.0,
+                    637.0,
+                    1767.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1759.0,
+                    396.0,
+                    1759.0,
+                    396.0,
+                    1798.0,
+                    292.0,
+                    1798.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    450.0,
+                    1759.0,
+                    1406.0,
+                    1759.0,
+                    1406.0,
+                    1798.0,
+                    450.0,
+                    1798.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1792.0,
+                    362.0,
+                    1792.0,
+                    362.0,
+                    1826.0,
+                    293.0,
+                    1826.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    438.0,
+                    1792.0,
+                    1407.0,
+                    1792.0,
+                    1407.0,
+                    1826.0,
+                    438.0,
+                    1826.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1821.0,
+                    324.0,
+                    1821.0,
+                    324.0,
+                    1858.0,
+                    292.0,
+                    1858.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    365.0,
+                    1821.0,
+                    426.0,
+                    1821.0,
+                    426.0,
+                    1858.0,
+                    365.0,
+                    1858.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    477.0,
+                    1821.0,
+                    1407.0,
+                    1821.0,
+                    1407.0,
+                    1858.0,
+                    477.0,
+                    1858.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1852.0,
+                    1407.0,
+                    1852.0,
+                    1407.0,
+                    1889.0,
+                    293.0,
+                    1889.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1883.0,
+                    1406.0,
+                    1883.0,
+                    1406.0,
+                    1921.0,
+                    293.0,
+                    1921.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1913.0,
+                    1409.0,
+                    1913.0,
+                    1409.0,
+                    1950.0,
+                    295.0,
+                    1950.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    259.0,
+                    947.0,
+                    259.0,
+                    947.0,
+                    294.0,
+                    296.0,
+                    294.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    999.0,
+                    259.0,
+                    1407.0,
+                    259.0,
+                    1407.0,
+                    294.0,
+                    999.0,
+                    294.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    286.0,
+                    897.0,
+                    286.0,
+                    897.0,
+                    326.0,
+                    291.0,
+                    326.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    947.0,
+                    286.0,
+                    1406.0,
+                    286.0,
+                    1406.0,
+                    326.0,
+                    947.0,
+                    326.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    317.0,
+                    1406.0,
+                    317.0,
+                    1406.0,
+                    358.0,
+                    291.0,
+                    358.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    349.0,
+                    1208.0,
+                    349.0,
+                    1208.0,
+                    387.0,
+                    291.0,
+                    387.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    792.0,
+                    396.0,
+                    1407.0,
+                    396.0,
+                    1407.0,
+                    429.0,
+                    792.0,
+                    429.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    794.0,
+                    426.0,
+                    1406.0,
+                    426.0,
+                    1406.0,
+                    459.0,
+                    794.0,
+                    459.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    789.0,
+                    459.0,
+                    1404.0,
+                    459.0,
+                    1404.0,
+                    488.0,
+                    789.0,
+                    488.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    790.0,
+                    487.0,
+                    1049.0,
+                    487.0,
+                    1049.0,
+                    520.0,
+                    790.0,
+                    520.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1120.0,
+                    487.0,
+                    1405.0,
+                    487.0,
+                    1405.0,
+                    520.0,
+                    1120.0,
+                    520.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    790.0,
+                    519.0,
+                    1407.0,
+                    519.0,
+                    1407.0,
+                    552.0,
+                    790.0,
+                    552.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    790.0,
+                    548.0,
+                    1405.0,
+                    548.0,
+                    1405.0,
+                    581.0,
+                    790.0,
+                    581.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    790.0,
+                    578.0,
+                    1405.0,
+                    578.0,
+                    1405.0,
+                    611.0,
+                    790.0,
+                    611.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    605.0,
+                    1347.0,
+                    605.0,
+                    1347.0,
+                    645.0,
+                    291.0,
+                    645.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1398.0,
+                    605.0,
+                    1402.0,
+                    605.0,
+                    1402.0,
+                    645.0,
+                    1398.0,
+                    645.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    638.0,
+                    496.0,
+                    638.0,
+                    496.0,
+                    673.0,
+                    293.0,
+                    673.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    545.0,
+                    638.0,
+                    1406.0,
+                    638.0,
+                    1406.0,
+                    673.0,
+                    545.0,
+                    673.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    670.0,
+                    1405.0,
+                    670.0,
+                    1405.0,
+                    702.0,
+                    293.0,
+                    702.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    697.0,
+                    1255.0,
+                    697.0,
+                    1255.0,
+                    734.0,
+                    291.0,
+                    734.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1348.0,
+                    697.0,
+                    1355.0,
+                    697.0,
+                    1355.0,
+                    734.0,
+                    1348.0,
+                    734.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 7,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1301,
+                    1405,
+                    1301,
+                    1405,
+                    1637,
+                    297,
+                    1637
+                ],
+                "score": 0.984
+            },
+            {
+                "category_id": 5,
+                "poly": [
+                    294,
+                    278,
+                    1399,
+                    278,
+                    1399,
+                    650,
+                    294,
+                    650
+                ],
+                "score": 0.984,
+                "html": "<table><tr><td>Dataset</td><td>MUTAG</td><td>PTC</td><td>PROTEINS</td><td>NCI1</td><td>NCI109</td><td>IMDB-B</td><td>IMDB-M</td><td>RDT-B</td></tr><tr><td>RWK 四</td><td>79.2±2.1</td><td>55.9±0.3</td><td>59.6±0.1</td><td>&gt;3 days</td><td>N/A</td><td>N/A</td><td>N/A</td><td>N/A</td></tr><tr><td>GK (k = 3) I 64</td><td>81.4±1.7</td><td>55.7±0.5</td><td>71.4±0.31</td><td>62.5±0.3</td><td>62.4±0.3</td><td>N/A</td><td>N/A</td><td>N/A</td></tr><tr><td>PK 58</td><td>76.0±2.7</td><td>59.5±2.4</td><td>73.7±0.7</td><td>82.5±0.5</td><td>N/A</td><td>N/A</td><td>N/A</td><td>N/A</td></tr><tr><td>WL kernel [65]</td><td>90.4±5.7</td><td>59.9±4.3</td><td>75.0±3.1</td><td>86.0±1.8</td><td>N/A</td><td>73.8±3.9</td><td>50.9±3.8</td><td>81.0±3.1</td></tr><tr><td>DCNN</td><td>N/A</td><td>N/A</td><td>61.3±1.6</td><td>56.6±1.0</td><td>N/A</td><td>49.1±1.4</td><td>33.5±1.4</td><td>N/A</td></tr><tr><td>DGCNN[76]</td><td>85.8±1.8</td><td>58.6±2.5</td><td>75.5±0.9</td><td>74.4±0.5</td><td>N/A</td><td>70.0±0.9</td><td>47.8±0.9</td><td>N/A</td></tr><tr><td>IGN[ 圆</td><td>83.9±13.0</td><td>58.5±6.9</td><td>76.6±5.5</td><td>74.3±2.7</td><td>72.8±1.5</td><td>72.0±5.5</td><td>48.7±3.4</td><td>N/A</td></tr><tr><td>GIN 园</td><td>89.4±5.6</td><td>64.6±7.0</td><td>76.2±2.8</td><td>82.7±1.7</td><td>N/A</td><td>75.1±5.1</td><td>52.3±2.8</td><td>92.4±2.5</td></tr><tr><td>PPGNs [531</td><td>90.6±8.7</td><td>66.2±6.6</td><td>77.2±4.7</td><td>83.2±1.1</td><td>82.2±1.4</td><td>73.0±5.8</td><td>50.5±3.6</td><td>N/A</td></tr><tr><td>Natural GN 自</td><td>89.4±1.6</td><td>66.8±1.7</td><td>71.7±1.0</td><td>82.4±1.3</td><td>N/A</td><td>73.5±2.0</td><td>51.3±1.5</td><td>N/A</td></tr><tr><td>GSN SIN</td><td>92.2 ± 7.5 N/A</td><td>68.2 ± 7.2 N/A</td><td>76.6± 5.0 76.4 ± 3.3</td><td>83.5 ± 2.0 82.7 ± 2.1</td><td>N/A</td><td>77.8 ± 3.3</td><td>54.3 ± 3.3</td><td>N/A</td></tr><tr><td></td><td></td><td></td><td></td><td></td><td>N/A</td><td>75.6± 3.2</td><td>52.4 ± 2.9</td><td>92.2 ± 1.0</td></tr><tr><td>CIN (Ours)</td><td>92.7 ± 6.1</td><td>68.2± 5.6</td><td>77.0 ± 4.3</td><td>83.6 ± 1.4</td><td>84.0 ± 1.6</td><td>75.6± 3.7</td><td>52.7 ± 3.1</td><td>92.4 ± 2.1</td></tr></table>"
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1701,
+                    1406,
+                    1701,
+                    1406,
+                    2007,
+                    297,
+                    2007
+                ],
+                "score": 0.984
+            },
+            {
+                "category_id": 5,
+                "poly": [
+                    408,
+                    760,
+                    1291,
+                    760,
+                    1291,
+                    1119,
+                    408,
+                    1119
+                ],
+                "score": 0.982,
+                "html": "<table><tr><td rowspan=\"2\">Method</td><td colspan=\"2\">ZINC↓</td><td rowspan=\"2\">ZINC-FULL↓ All methods</td><td rowspan=\"2\">MOLHIV↑ All methods</td></tr><tr><td>No Edge Feat.</td><td>With Edge Feat.</td></tr><tr><td>GCN 四</td><td>0.469±0.002</td><td>N/A</td><td>N/A</td><td>76.06±0.97</td></tr><tr><td>GAT 园</td><td>0.463±0.002</td><td>N/A</td><td>N/A</td><td>N/A</td></tr><tr><td>GatedGCN[ 自</td><td>0.422±0.006</td><td>0.363±0.009</td><td>N/A</td><td>N/A</td></tr><tr><td>GIN 四</td><td>0.408±0.008</td><td>0.252±0.014</td><td>0.088±0.002</td><td>77.07±1.49</td></tr><tr><td>PNA 四</td><td>0.320±0.032</td><td>0.188±0.004</td><td>N/A</td><td>79.05±1.32</td></tr><tr><td>DGN 回</td><td>0.219±0.010</td><td>0.168±0.003</td><td>N/A</td><td>79.70±0.97</td></tr><tr><td>HIMP 四</td><td>N/A</td><td>0.151±0.006</td><td>0.036±0.002</td><td>78.80±0.82</td></tr><tr><td>GSN 目</td><td>0.139±0.007</td><td>0.108±0.018</td><td>N/A</td><td>77.99±1.00</td></tr><tr><td>CIN-small (Ours)</td><td>0.139±0.008</td><td>0.094±0.004</td><td>0.044±0.003</td><td>80.55±1.04</td></tr><tr><td>CIN (Ours)</td><td>0.115±0.003</td><td>0.079±0.006</td><td>0.022±0.002</td><td>80.94±0.57</td></tr></table>"
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    300,
+                    1174,
+                    1402,
+                    1174,
+                    1402,
+                    1236,
+                    300,
+                    1236
+                ],
+                "score": 0.94
+            },
+            {
+                "category_id": 6,
+                "poly": [
+                    295,
+                    216,
+                    1402,
+                    216,
+                    1402,
+                    277,
+                    295,
+                    277
+                ],
+                "score": 0.933
+            },
+            {
+                "category_id": 6,
+                "poly": [
+                    437,
+                    727,
+                    1250,
+                    727,
+                    1250,
+                    759,
+                    437,
+                    759
+                ],
+                "score": 0.868
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    840,
+                    2061,
+                    859,
+                    2061,
+                    859,
+                    2084,
+                    840,
+                    2084
+                ],
+                "score": 0.799
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    651,
+                    1765,
+                    720,
+                    1765,
+                    720,
+                    1792,
+                    651,
+                    1792
+                ],
+                "score": 0.9,
+                "latex": "k = 6"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    613,
+                    1394,
+                    697,
+                    1394,
+                    697,
+                    1422,
+                    613,
+                    1422
+                ],
+                "score": 0.89,
+                "latex": "k = 1 8"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    605,
+                    1945,
+                    654,
+                    1945,
+                    654,
+                    1977,
+                    605,
+                    1977
+                ],
+                "score": 0.86,
+                "latex": "[ [ 2 7 ]"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1321,
+                    1762,
+                    1370,
+                    1762,
+                    1370,
+                    1795,
+                    1321,
+                    1795
+                ],
+                "score": 0.83,
+                "latex": " { \\mathbb { \\left[ \\left[ 2 7 \\right] \\right] } }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    574,
+                    1331,
+                    621,
+                    1331,
+                    621,
+                    1364,
+                    574,
+                    1364
+                ],
+                "score": 0.72,
+                "latex": "\\pmb { \\mathbb { E } } 8 \\|"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    504,
+                    1732,
+                    553,
+                    1732,
+                    553,
+                    1765,
+                    504,
+                    1765
+                ],
+                "score": 0.69,
+                "latex": "\\textcircled { | 4 0 | }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1026,
+                    1885,
+                    1085,
+                    1885,
+                    1085,
+                    1914,
+                    1026,
+                    1914
+                ],
+                "score": 0.58,
+                "latex": "1 0 0 \\mathrm { k }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1380,
+                    1452,
+                    1406,
+                    1452,
+                    1406,
+                    1488,
+                    1380,
+                    1488
+                ],
+                "score": 0.44,
+                "latex": "\\textcircled { 3 }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1160,
+                    1883,
+                    1187,
+                    1883,
+                    1187,
+                    1920,
+                    1160,
+                    1920
+                ],
+                "score": 0.4,
+                "latex": "3"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    215.0,
+                    1407.0,
+                    215.0,
+                    1407.0,
+                    250.0,
+                    294.0,
+                    250.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    245.0,
+                    1290.0,
+                    245.0,
+                    1290.0,
+                    280.0,
+                    295.0,
+                    280.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    437.0,
+                    720.0,
+                    1253.0,
+                    720.0,
+                    1253.0,
+                    769.0,
+                    437.0,
+                    769.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    839.0,
+                    2060.0,
+                    861.0,
+                    2060.0,
+                    861.0,
+                    2090.0,
+                    839.0,
+                    2090.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1301.0,
+                    1403.0,
+                    1301.0,
+                    1403.0,
+                    1338.0,
+                    294.0,
+                    1338.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1330.0,
+                    573.0,
+                    1330.0,
+                    573.0,
+                    1369.0,
+                    295.0,
+                    1369.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    622.0,
+                    1330.0,
+                    1402.0,
+                    1330.0,
+                    1402.0,
+                    1369.0,
+                    622.0,
+                    1369.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1361.0,
+                    1408.0,
+                    1361.0,
+                    1408.0,
+                    1400.0,
+                    294.0,
+                    1400.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1393.0,
+                    612.0,
+                    1393.0,
+                    612.0,
+                    1428.0,
+                    295.0,
+                    1428.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    698.0,
+                    1393.0,
+                    1407.0,
+                    1393.0,
+                    1407.0,
+                    1428.0,
+                    698.0,
+                    1428.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1423.0,
+                    1407.0,
+                    1423.0,
+                    1407.0,
+                    1457.0,
+                    294.0,
+                    1457.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1454.0,
+                    1379.0,
+                    1454.0,
+                    1379.0,
+                    1488.0,
+                    294.0,
+                    1488.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1484.0,
+                    1406.0,
+                    1484.0,
+                    1406.0,
+                    1517.0,
+                    295.0,
+                    1517.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1515.0,
+                    1405.0,
+                    1515.0,
+                    1405.0,
+                    1550.0,
+                    295.0,
+                    1550.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1545.0,
+                    1405.0,
+                    1545.0,
+                    1405.0,
+                    1579.0,
+                    295.0,
+                    1579.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1577.0,
+                    1405.0,
+                    1577.0,
+                    1405.0,
+                    1608.0,
+                    296.0,
+                    1608.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1608.0,
+                    1407.0,
+                    1608.0,
+                    1407.0,
+                    1639.0,
+                    296.0,
+                    1639.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1699.0,
+                    1404.0,
+                    1699.0,
+                    1404.0,
+                    1737.0,
+                    292.0,
+                    1737.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1733.0,
+                    503.0,
+                    1733.0,
+                    503.0,
+                    1769.0,
+                    295.0,
+                    1769.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    554.0,
+                    1733.0,
+                    1404.0,
+                    1733.0,
+                    1404.0,
+                    1769.0,
+                    554.0,
+                    1769.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1763.0,
+                    650.0,
+                    1763.0,
+                    650.0,
+                    1798.0,
+                    295.0,
+                    1798.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    721.0,
+                    1763.0,
+                    1320.0,
+                    1763.0,
+                    1320.0,
+                    1798.0,
+                    721.0,
+                    1798.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1371.0,
+                    1763.0,
+                    1404.0,
+                    1763.0,
+                    1404.0,
+                    1798.0,
+                    1371.0,
+                    1798.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1792.0,
+                    1407.0,
+                    1792.0,
+                    1407.0,
+                    1831.0,
+                    292.0,
+                    1831.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1824.0,
+                    1406.0,
+                    1824.0,
+                    1406.0,
+                    1858.0,
+                    292.0,
+                    1858.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1854.0,
+                    1408.0,
+                    1854.0,
+                    1408.0,
+                    1888.0,
+                    294.0,
+                    1888.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1884.0,
+                    1025.0,
+                    1884.0,
+                    1025.0,
+                    1920.0,
+                    294.0,
+                    1920.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1086.0,
+                    1884.0,
+                    1159.0,
+                    1884.0,
+                    1159.0,
+                    1920.0,
+                    1086.0,
+                    1920.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1188.0,
+                    1884.0,
+                    1408.0,
+                    1884.0,
+                    1408.0,
+                    1920.0,
+                    1188.0,
+                    1920.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1912.0,
+                    1407.0,
+                    1912.0,
+                    1407.0,
+                    1950.0,
+                    294.0,
+                    1950.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1941.0,
+                    604.0,
+                    1941.0,
+                    604.0,
+                    1985.0,
+                    291.0,
+                    1985.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    655.0,
+                    1941.0,
+                    1406.0,
+                    1941.0,
+                    1406.0,
+                    1985.0,
+                    655.0,
+                    1985.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1973.0,
+                    1102.0,
+                    1973.0,
+                    1102.0,
+                    2013.0,
+                    292.0,
+                    2013.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1173.0,
+                    1404.0,
+                    1173.0,
+                    1404.0,
+                    1209.0,
+                    296.0,
+                    1209.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1204.0,
+                    1234.0,
+                    1204.0,
+                    1234.0,
+                    1239.0,
+                    296.0,
+                    1239.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 8,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    554,
+                    1405,
+                    554,
+                    1405,
+                    828,
+                    298,
+                    828
+                ],
+                "score": 0.983
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    1409,
+                    1403,
+                    1409,
+                    1403,
+                    1621,
+                    299,
+                    1621
+                ],
+                "score": 0.983
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    867,
+                    1404,
+                    867,
+                    1404,
+                    1112,
+                    298,
+                    1112
+                ],
+                "score": 0.981
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    270,
+                    1404,
+                    270,
+                    1404,
+                    515,
+                    298,
+                    515
+                ],
+                "score": 0.981
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1150,
+                    1404,
+                    1150,
+                    1404,
+                    1394,
+                    298,
+                    1394
+                ],
+                "score": 0.981
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1661,
+                    1403,
+                    1661,
+                    1403,
+                    1815,
+                    298,
+                    1815
+                ],
+                "score": 0.978
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    300,
+                    1854,
+                    1402,
+                    1854,
+                    1402,
+                    2006,
+                    300,
+                    2006
+                ],
+                "score": 0.976
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    297,
+                    197,
+                    943,
+                    197,
+                    943,
+                    236,
+                    297,
+                    236
+                ],
+                "score": 0.928
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    837,
+                    2062,
+                    865,
+                    2062,
+                    865,
+                    2085,
+                    837,
+                    2085
+                ],
+                "score": 0.849
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    403,
+                    960,
+                    473,
+                    960,
+                    473,
+                    990,
+                    403,
+                    990
+                ],
+                "score": 0.91,
+                "latex": "k \\geq 3"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1028,
+                    705,
+                    1076,
+                    705,
+                    1076,
+                    738,
+                    1028,
+                    738
+                ],
+                "score": 0.82,
+                "latex": "\\mathbb { m }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    328,
+                    930,
+                    346,
+                    930,
+                    346,
+                    956,
+                    328,
+                    956
+                ],
+                "score": 0.8,
+                "latex": "k"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1159,
+                    870,
+                    1177,
+                    870,
+                    1177,
+                    896,
+                    1159,
+                    896
+                ],
+                "score": 0.8,
+                "latex": "k"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    962,
+                    900,
+                    981,
+                    900,
+                    981,
+                    927,
+                    962,
+                    927
+                ],
+                "score": 0.8,
+                "latex": "k"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1228,
+                    331,
+                    1278,
+                    331,
+                    1278,
+                    364,
+                    1228,
+                    364
+                ],
+                "score": 0.65,
+                "latex": "\\pmb { \\Vert 3 4 \\Vert }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1040,
+                    614,
+                    1129,
+                    614,
+                    1129,
+                    647,
+                    1040,
+                    647
+                ],
+                "score": 0.55,
+                "latex": "\\mathbb { \\lVert 4 3 \\rVert \\overline { { \\lVert 4 4 \\rVert } } }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    655,
+                    1049,
+                    688,
+                    1049,
+                    688,
+                    1085,
+                    655,
+                    1085
+                ],
+                "score": 0.52,
+                "latex": "\\mathbf { A }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    360,
+                    332,
+                    394,
+                    332,
+                    394,
+                    364,
+                    360,
+                    364
+                ],
+                "score": 0.41,
+                "latex": "\\pmb { \\bigtriangledown }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    576,
+                    644,
+                    625,
+                    644,
+                    625,
+                    677,
+                    576,
+                    677
+                ],
+                "score": 0.41,
+                "latex": " { \\mathbb { \\left[ \\left[ 2 7 \\right] \\right] } }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1350,
+                    1781,
+                    1398,
+                    1781,
+                    1398,
+                    1815,
+                    1350,
+                    1815
+                ],
+                "score": 0.34,
+                "latex": "\\pmb { \\mathbb { B } } 0 \\|"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    198.0,
+                    945.0,
+                    198.0,
+                    945.0,
+                    239.0,
+                    294.0,
+                    239.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    832.0,
+                    2058.0,
+                    870.0,
+                    2058.0,
+                    870.0,
+                    2097.0,
+                    832.0,
+                    2097.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    555.0,
+                    1406.0,
+                    555.0,
+                    1406.0,
+                    587.0,
+                    296.0,
+                    587.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    583.0,
+                    1405.0,
+                    583.0,
+                    1405.0,
+                    621.0,
+                    295.0,
+                    621.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    614.0,
+                    1039.0,
+                    614.0,
+                    1039.0,
+                    651.0,
+                    295.0,
+                    651.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1130.0,
+                    614.0,
+                    1405.0,
+                    614.0,
+                    1405.0,
+                    651.0,
+                    1130.0,
+                    651.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    643.0,
+                    575.0,
+                    643.0,
+                    575.0,
+                    681.0,
+                    292.0,
+                    681.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    626.0,
+                    643.0,
+                    1405.0,
+                    643.0,
+                    1405.0,
+                    681.0,
+                    626.0,
+                    681.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    673.0,
+                    1405.0,
+                    673.0,
+                    1405.0,
+                    709.0,
+                    295.0,
+                    709.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    705.0,
+                    1027.0,
+                    705.0,
+                    1027.0,
+                    743.0,
+                    293.0,
+                    743.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1077.0,
+                    705.0,
+                    1408.0,
+                    705.0,
+                    1408.0,
+                    743.0,
+                    1077.0,
+                    743.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    734.0,
+                    1408.0,
+                    734.0,
+                    1408.0,
+                    774.0,
+                    293.0,
+                    774.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    768.0,
+                    1405.0,
+                    768.0,
+                    1405.0,
+                    800.0,
+                    295.0,
+                    800.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    797.0,
+                    1342.0,
+                    797.0,
+                    1342.0,
+                    834.0,
+                    296.0,
+                    834.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1408.0,
+                    1404.0,
+                    1408.0,
+                    1404.0,
+                    1446.0,
+                    294.0,
+                    1446.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1439.0,
+                    1407.0,
+                    1439.0,
+                    1407.0,
+                    1474.0,
+                    295.0,
+                    1474.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1470.0,
+                    1405.0,
+                    1470.0,
+                    1405.0,
+                    1504.0,
+                    295.0,
+                    1504.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1499.0,
+                    1405.0,
+                    1499.0,
+                    1405.0,
+                    1533.0,
+                    294.0,
+                    1533.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1532.0,
+                    1403.0,
+                    1532.0,
+                    1403.0,
+                    1563.0,
+                    295.0,
+                    1563.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1561.0,
+                    1405.0,
+                    1561.0,
+                    1405.0,
+                    1595.0,
+                    295.0,
+                    1595.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    1592.0,
+                    922.0,
+                    1592.0,
+                    922.0,
+                    1622.0,
+                    297.0,
+                    1622.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    869.0,
+                    1158.0,
+                    869.0,
+                    1158.0,
+                    902.0,
+                    296.0,
+                    902.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1178.0,
+                    869.0,
+                    1405.0,
+                    869.0,
+                    1405.0,
+                    902.0,
+                    1178.0,
+                    902.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    896.0,
+                    961.0,
+                    896.0,
+                    961.0,
+                    933.0,
+                    293.0,
+                    933.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    982.0,
+                    896.0,
+                    1406.0,
+                    896.0,
+                    1406.0,
+                    933.0,
+                    982.0,
+                    933.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    925.0,
+                    327.0,
+                    925.0,
+                    327.0,
+                    966.0,
+                    292.0,
+                    966.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    347.0,
+                    925.0,
+                    1409.0,
+                    925.0,
+                    1409.0,
+                    966.0,
+                    347.0,
+                    966.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    958.0,
+                    402.0,
+                    958.0,
+                    402.0,
+                    994.0,
+                    293.0,
+                    994.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    474.0,
+                    958.0,
+                    1404.0,
+                    958.0,
+                    1404.0,
+                    994.0,
+                    474.0,
+                    994.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    987.0,
+                    1406.0,
+                    987.0,
+                    1406.0,
+                    1029.0,
+                    291.0,
+                    1029.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1018.0,
+                    1405.0,
+                    1018.0,
+                    1405.0,
+                    1054.0,
+                    293.0,
+                    1054.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1049.0,
+                    654.0,
+                    1049.0,
+                    654.0,
+                    1087.0,
+                    292.0,
+                    1087.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    689.0,
+                    1049.0,
+                    1407.0,
+                    1049.0,
+                    1407.0,
+                    1087.0,
+                    689.0,
+                    1087.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1079.0,
+                    1185.0,
+                    1079.0,
+                    1185.0,
+                    1115.0,
+                    295.0,
+                    1115.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    269.0,
+                    1405.0,
+                    269.0,
+                    1405.0,
+                    305.0,
+                    296.0,
+                    305.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    299.0,
+                    1407.0,
+                    299.0,
+                    1407.0,
+                    335.0,
+                    292.0,
+                    335.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    329.0,
+                    359.0,
+                    329.0,
+                    359.0,
+                    369.0,
+                    296.0,
+                    369.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    395.0,
+                    329.0,
+                    1227.0,
+                    329.0,
+                    1227.0,
+                    369.0,
+                    395.0,
+                    369.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1279.0,
+                    329.0,
+                    1405.0,
+                    329.0,
+                    1405.0,
+                    369.0,
+                    1279.0,
+                    369.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    364.0,
+                    1404.0,
+                    364.0,
+                    1404.0,
+                    397.0,
+                    294.0,
+                    397.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    391.0,
+                    1406.0,
+                    391.0,
+                    1406.0,
+                    430.0,
+                    292.0,
+                    430.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    421.0,
+                    1406.0,
+                    421.0,
+                    1406.0,
+                    460.0,
+                    293.0,
+                    460.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    451.0,
+                    1409.0,
+                    451.0,
+                    1409.0,
+                    490.0,
+                    292.0,
+                    490.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    484.0,
+                    1085.0,
+                    484.0,
+                    1085.0,
+                    517.0,
+                    294.0,
+                    517.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1148.0,
+                    1409.0,
+                    1148.0,
+                    1409.0,
+                    1188.0,
+                    292.0,
+                    1188.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1180.0,
+                    1406.0,
+                    1180.0,
+                    1406.0,
+                    1217.0,
+                    293.0,
+                    1217.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1213.0,
+                    1405.0,
+                    1213.0,
+                    1405.0,
+                    1247.0,
+                    294.0,
+                    1247.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1242.0,
+                    1405.0,
+                    1242.0,
+                    1405.0,
+                    1277.0,
+                    292.0,
+                    1277.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1270.0,
+                    1405.0,
+                    1270.0,
+                    1405.0,
+                    1309.0,
+                    292.0,
+                    1309.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1302.0,
+                    1405.0,
+                    1302.0,
+                    1405.0,
+                    1338.0,
+                    292.0,
+                    1338.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1332.0,
+                    1406.0,
+                    1332.0,
+                    1406.0,
+                    1366.0,
+                    295.0,
+                    1366.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1363.0,
+                    1272.0,
+                    1363.0,
+                    1272.0,
+                    1396.0,
+                    296.0,
+                    1396.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1659.0,
+                    1405.0,
+                    1659.0,
+                    1405.0,
+                    1697.0,
+                    293.0,
+                    1697.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1692.0,
+                    1405.0,
+                    1692.0,
+                    1405.0,
+                    1729.0,
+                    294.0,
+                    1729.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1719.0,
+                    1405.0,
+                    1719.0,
+                    1405.0,
+                    1759.0,
+                    293.0,
+                    1759.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1754.0,
+                    1403.0,
+                    1754.0,
+                    1403.0,
+                    1787.0,
+                    296.0,
+                    1787.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1782.0,
+                    1349.0,
+                    1782.0,
+                    1349.0,
+                    1820.0,
+                    291.0,
+                    1820.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1399.0,
+                    1782.0,
+                    1405.0,
+                    1782.0,
+                    1405.0,
+                    1820.0,
+                    1399.0,
+                    1820.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1851.0,
+                    1406.0,
+                    1851.0,
+                    1406.0,
+                    1889.0,
+                    295.0,
+                    1889.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1885.0,
+                    1404.0,
+                    1885.0,
+                    1404.0,
+                    1917.0,
+                    294.0,
+                    1917.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1913.0,
+                    1406.0,
+                    1913.0,
+                    1406.0,
+                    1950.0,
+                    295.0,
+                    1950.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1944.0,
+                    1404.0,
+                    1944.0,
+                    1404.0,
+                    1981.0,
+                    295.0,
+                    1981.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1976.0,
+                    836.0,
+                    1976.0,
+                    836.0,
+                    2009.0,
+                    295.0,
+                    2009.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 9,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    266,
+                    1406,
+                    266,
+                    1406,
+                    450,
+                    297,
+                    450
+                ],
+                "score": 0.974
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    299,
+                    198,
+                    765,
+                    198,
+                    765,
+                    237,
+                    299,
+                    237
+                ],
+                "score": 0.919
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    295,
+                    546,
+                    1409,
+                    546,
+                    1409,
+                    2016,
+                    295,
+                    2016
+                ],
+                "score": 0.919
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    298,
+                    494,
+                    455,
+                    494,
+                    455,
+                    529,
+                    298,
+                    529
+                ],
+                "score": 0.903
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    836,
+                    2061,
+                    863,
+                    2061,
+                    863,
+                    2086,
+                    836,
+                    2086
+                ],
+                "score": 0.709
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    836,
+                    2061,
+                    863,
+                    2061,
+                    863,
+                    2086,
+                    836,
+                    2086
+                ],
+                "score": 0.292
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1200,
+                    388,
+                    1249,
+                    388,
+                    1249,
+                    420,
+                    1200,
+                    420
+                ],
+                "score": 0.49,
+                "latex": "\\pmb { \\bigtriangledown }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    421,
+                    299,
+                    451,
+                    299,
+                    451,
+                    326,
+                    421,
+                    326
+                ],
+                "score": 0.35,
+                "latex": " { \\mathrm { n ^ { o } } }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1152,
+                    299,
+                    1182,
+                    299,
+                    1182,
+                    326,
+                    1152,
+                    326
+                ],
+                "score": 0.26,
+                "latex": "\\boldsymbol { \\mathrm { n ^ { \\circ } } }"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    195.0,
+                    770.0,
+                    195.0,
+                    770.0,
+                    245.0,
+                    293.0,
+                    245.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    491.0,
+                    460.0,
+                    491.0,
+                    460.0,
+                    534.0,
+                    295.0,
+                    534.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    832.0,
+                    2058.0,
+                    868.0,
+                    2058.0,
+                    868.0,
+                    2098.0,
+                    832.0,
+                    2098.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    832.0,
+                    2058.0,
+                    868.0,
+                    2058.0,
+                    868.0,
+                    2098.0,
+                    832.0,
+                    2098.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    265.0,
+                    1406.0,
+                    265.0,
+                    1406.0,
+                    302.0,
+                    292.0,
+                    302.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    296.0,
+                    420.0,
+                    296.0,
+                    420.0,
+                    333.0,
+                    291.0,
+                    333.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    452.0,
+                    296.0,
+                    1151.0,
+                    296.0,
+                    1151.0,
+                    333.0,
+                    452.0,
+                    333.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1183.0,
+                    296.0,
+                    1409.0,
+                    296.0,
+                    1409.0,
+                    333.0,
+                    1183.0,
+                    333.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    325.0,
+                    1407.0,
+                    325.0,
+                    1407.0,
+                    364.0,
+                    292.0,
+                    364.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    355.0,
+                    1406.0,
+                    355.0,
+                    1406.0,
+                    392.0,
+                    292.0,
+                    392.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    387.0,
+                    1199.0,
+                    387.0,
+                    1199.0,
+                    423.0,
+                    295.0,
+                    423.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1250.0,
+                    387.0,
+                    1406.0,
+                    387.0,
+                    1406.0,
+                    423.0,
+                    1250.0,
+                    423.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    416.0,
+                    1349.0,
+                    416.0,
+                    1349.0,
+                    454.0,
+                    292.0,
+                    454.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    304.0,
+                    543.0,
+                    1408.0,
+                    543.0,
+                    1408.0,
+                    583.0,
+                    304.0,
+                    583.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    356.0,
+                    576.0,
+                    657.0,
+                    576.0,
+                    657.0,
+                    612.0,
+                    356.0,
+                    612.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    306.0,
+                    627.0,
+                    1405.0,
+                    627.0,
+                    1405.0,
+                    668.0,
+                    306.0,
+                    668.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    354.0,
+                    660.0,
+                    1406.0,
+                    660.0,
+                    1406.0,
+                    696.0,
+                    354.0,
+                    696.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    354.0,
+                    689.0,
+                    872.0,
+                    689.0,
+                    872.0,
+                    725.0,
+                    354.0,
+                    725.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    304.0,
+                    738.0,
+                    1406.0,
+                    738.0,
+                    1406.0,
+                    782.0,
+                    304.0,
+                    782.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    354.0,
+                    772.0,
+                    563.0,
+                    772.0,
+                    563.0,
+                    808.0,
+                    354.0,
+                    808.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    309.0,
+                    825.0,
+                    1406.0,
+                    825.0,
+                    1406.0,
+                    861.0,
+                    309.0,
+                    861.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    352.0,
+                    852.0,
+                    1407.0,
+                    852.0,
+                    1407.0,
+                    894.0,
+                    352.0,
+                    894.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    356.0,
+                    887.0,
+                    903.0,
+                    887.0,
+                    903.0,
+                    923.0,
+                    356.0,
+                    923.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    306.0,
+                    934.0,
+                    1408.0,
+                    934.0,
+                    1408.0,
+                    975.0,
+                    306.0,
+                    975.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    350.0,
+                    967.0,
+                    642.0,
+                    967.0,
+                    642.0,
+                    999.0,
+                    350.0,
+                    999.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    306.0,
+                    1018.0,
+                    1411.0,
+                    1018.0,
+                    1411.0,
+                    1059.0,
+                    306.0,
+                    1059.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    354.0,
+                    1050.0,
+                    1002.0,
+                    1050.0,
+                    1002.0,
+                    1086.0,
+                    354.0,
+                    1086.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    306.0,
+                    1100.0,
+                    1408.0,
+                    1100.0,
+                    1408.0,
+                    1139.0,
+                    306.0,
+                    1139.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    356.0,
+                    1132.0,
+                    1406.0,
+                    1132.0,
+                    1406.0,
+                    1168.0,
+                    356.0,
+                    1168.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    349.0,
+                    1158.0,
+                    1408.0,
+                    1158.0,
+                    1408.0,
+                    1204.0,
+                    349.0,
+                    1204.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    352.0,
+                    1189.0,
+                    432.0,
+                    1189.0,
+                    432.0,
+                    1231.0,
+                    352.0,
+                    1231.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    308.0,
+                    1245.0,
+                    1406.0,
+                    1245.0,
+                    1406.0,
+                    1281.0,
+                    308.0,
+                    1281.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    354.0,
+                    1276.0,
+                    1408.0,
+                    1276.0,
+                    1408.0,
+                    1312.0,
+                    354.0,
+                    1312.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    354.0,
+                    1306.0,
+                    507.0,
+                    1306.0,
+                    507.0,
+                    1342.0,
+                    354.0,
+                    1342.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    306.0,
+                    1356.0,
+                    1411.0,
+                    1356.0,
+                    1411.0,
+                    1397.0,
+                    306.0,
+                    1397.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    354.0,
+                    1387.0,
+                    1408.0,
+                    1387.0,
+                    1408.0,
+                    1427.0,
+                    354.0,
+                    1427.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    352.0,
+                    1418.0,
+                    665.0,
+                    1418.0,
+                    665.0,
+                    1454.0,
+                    352.0,
+                    1454.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1469.0,
+                    1408.0,
+                    1469.0,
+                    1408.0,
+                    1510.0,
+                    293.0,
+                    1510.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    351.0,
+                    1500.0,
+                    1409.0,
+                    1500.0,
+                    1409.0,
+                    1539.0,
+                    351.0,
+                    1539.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    354.0,
+                    1531.0,
+                    426.0,
+                    1531.0,
+                    426.0,
+                    1566.0,
+                    354.0,
+                    1566.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1582.0,
+                    1406.0,
+                    1582.0,
+                    1406.0,
+                    1623.0,
+                    293.0,
+                    1623.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    354.0,
+                    1615.0,
+                    639.0,
+                    1615.0,
+                    639.0,
+                    1646.0,
+                    354.0,
+                    1646.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1667.0,
+                    1408.0,
+                    1667.0,
+                    1408.0,
+                    1703.0,
+                    294.0,
+                    1703.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    356.0,
+                    1698.0,
+                    1406.0,
+                    1698.0,
+                    1406.0,
+                    1734.0,
+                    356.0,
+                    1734.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    352.0,
+                    1726.0,
+                    550.0,
+                    1726.0,
+                    550.0,
+                    1760.0,
+                    352.0,
+                    1760.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1778.0,
+                    1406.0,
+                    1778.0,
+                    1406.0,
+                    1817.0,
+                    291.0,
+                    1817.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    352.0,
+                    1811.0,
+                    1348.0,
+                    1811.0,
+                    1348.0,
+                    1847.0,
+                    352.0,
+                    1847.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1863.0,
+                    1405.0,
+                    1863.0,
+                    1405.0,
+                    1899.0,
+                    294.0,
+                    1899.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    354.0,
+                    1893.0,
+                    1368.0,
+                    1893.0,
+                    1368.0,
+                    1929.0,
+                    354.0,
+                    1929.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1945.0,
+                    1406.0,
+                    1945.0,
+                    1406.0,
+                    1981.0,
+                    294.0,
+                    1981.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    354.0,
+                    1974.0,
+                    735.0,
+                    1974.0,
+                    735.0,
+                    2010.0,
+                    354.0,
+                    2010.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 10,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 2,
+                "poly": [
+                    836,
+                    2061,
+                    865,
+                    2061,
+                    865,
+                    2086,
+                    836,
+                    2086
+                ],
+                "score": 0.829
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    291,
+                    84,
+                    1411,
+                    84,
+                    1411,
+                    2021,
+                    291,
+                    2021
+                ],
+                "score": 0.569
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1349,
+                    1025,
+                    1402,
+                    1025,
+                    1402,
+                    1054,
+                    1349,
+                    1054
+                ],
+                "score": 0.55,
+                "latex": "2 2 t h"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1198,
+                    1169,
+                    1283,
+                    1169,
+                    1283,
+                    1198,
+                    1198,
+                    1198
+                ],
+                "score": 0.48,
+                "latex": "( G R L + ,"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    832.0,
+                    2058.0,
+                    870.0,
+                    2058.0,
+                    870.0,
+                    2097.0,
+                    832.0,
+                    2097.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    197.0,
+                    1410.0,
+                    197.0,
+                    1410.0,
+                    246.0,
+                    292.0,
+                    246.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    351.0,
+                    231.0,
+                    576.0,
+                    231.0,
+                    576.0,
+                    272.0,
+                    351.0,
+                    272.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    284.0,
+                    1196.0,
+                    284.0,
+                    1196.0,
+                    323.0,
+                    292.0,
+                    323.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    337.0,
+                    1408.0,
+                    337.0,
+                    1408.0,
+                    376.0,
+                    292.0,
+                    376.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    353.0,
+                    367.0,
+                    1029.0,
+                    367.0,
+                    1029.0,
+                    407.0,
+                    353.0,
+                    407.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    416.0,
+                    1404.0,
+                    416.0,
+                    1404.0,
+                    456.0,
+                    294.0,
+                    456.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    353.0,
+                    448.0,
+                    1406.0,
+                    448.0,
+                    1406.0,
+                    488.0,
+                    353.0,
+                    488.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    353.0,
+                    477.0,
+                    794.0,
+                    477.0,
+                    794.0,
+                    518.0,
+                    353.0,
+                    518.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    528.0,
+                    1406.0,
+                    528.0,
+                    1406.0,
+                    569.0,
+                    294.0,
+                    569.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    353.0,
+                    560.0,
+                    1410.0,
+                    560.0,
+                    1410.0,
+                    601.0,
+                    353.0,
+                    601.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    613.0,
+                    1355.0,
+                    613.0,
+                    1355.0,
+                    651.0,
+                    292.0,
+                    651.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    662.0,
+                    1408.0,
+                    662.0,
+                    1408.0,
+                    705.0,
+                    292.0,
+                    705.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    351.0,
+                    696.0,
+                    1054.0,
+                    696.0,
+                    1054.0,
+                    732.0,
+                    351.0,
+                    732.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    745.0,
+                    1408.0,
+                    745.0,
+                    1408.0,
+                    787.0,
+                    294.0,
+                    787.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    355.0,
+                    779.0,
+                    1188.0,
+                    779.0,
+                    1188.0,
+                    813.0,
+                    355.0,
+                    813.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    825.0,
+                    1410.0,
+                    825.0,
+                    1410.0,
+                    872.0,
+                    292.0,
+                    872.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    353.0,
+                    859.0,
+                    1349.0,
+                    859.0,
+                    1349.0,
+                    900.0,
+                    353.0,
+                    900.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    910.0,
+                    1406.0,
+                    910.0,
+                    1406.0,
+                    951.0,
+                    296.0,
+                    951.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    355.0,
+                    944.0,
+                    1018.0,
+                    944.0,
+                    1018.0,
+                    978.0,
+                    355.0,
+                    978.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    993.0,
+                    1408.0,
+                    993.0,
+                    1408.0,
+                    1033.0,
+                    294.0,
+                    1033.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    353.0,
+                    1023.0,
+                    1348.0,
+                    1023.0,
+                    1348.0,
+                    1063.0,
+                    353.0,
+                    1063.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1403.0,
+                    1023.0,
+                    1408.0,
+                    1023.0,
+                    1408.0,
+                    1063.0,
+                    1403.0,
+                    1063.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    353.0,
+                    1057.0,
+                    1406.0,
+                    1057.0,
+                    1406.0,
+                    1091.0,
+                    353.0,
+                    1091.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    351.0,
+                    1084.0,
+                    1251.0,
+                    1084.0,
+                    1251.0,
+                    1125.0,
+                    351.0,
+                    1125.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1135.0,
+                    1408.0,
+                    1135.0,
+                    1408.0,
+                    1176.0,
+                    294.0,
+                    1176.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    353.0,
+                    1165.0,
+                    1197.0,
+                    1165.0,
+                    1197.0,
+                    1205.0,
+                    353.0,
+                    1205.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1284.0,
+                    1165.0,
+                    1408.0,
+                    1165.0,
+                    1408.0,
+                    1205.0,
+                    1284.0,
+                    1205.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    355.0,
+                    1199.0,
+                    427.0,
+                    1199.0,
+                    427.0,
+                    1235.0,
+                    355.0,
+                    1235.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1252.0,
+                    1408.0,
+                    1252.0,
+                    1408.0,
+                    1286.0,
+                    296.0,
+                    1286.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    353.0,
+                    1280.0,
+                    1256.0,
+                    1280.0,
+                    1256.0,
+                    1320.0,
+                    353.0,
+                    1320.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1330.0,
+                    1406.0,
+                    1330.0,
+                    1406.0,
+                    1371.0,
+                    294.0,
+                    1371.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    355.0,
+                    1362.0,
+                    1404.0,
+                    1362.0,
+                    1404.0,
+                    1396.0,
+                    355.0,
+                    1396.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1411.0,
+                    1406.0,
+                    1411.0,
+                    1406.0,
+                    1451.0,
+                    294.0,
+                    1451.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    353.0,
+                    1443.0,
+                    1408.0,
+                    1443.0,
+                    1408.0,
+                    1483.0,
+                    353.0,
+                    1483.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    351.0,
+                    1468.0,
+                    1408.0,
+                    1468.0,
+                    1408.0,
+                    1517.0,
+                    351.0,
+                    1517.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    355.0,
+                    1504.0,
+                    1296.0,
+                    1504.0,
+                    1296.0,
+                    1538.0,
+                    355.0,
+                    1538.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1555.0,
+                    1406.0,
+                    1555.0,
+                    1406.0,
+                    1596.0,
+                    294.0,
+                    1596.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    349.0,
+                    1585.0,
+                    997.0,
+                    1585.0,
+                    997.0,
+                    1623.0,
+                    349.0,
+                    1623.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1634.0,
+                    1408.0,
+                    1634.0,
+                    1408.0,
+                    1681.0,
+                    292.0,
+                    1681.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    351.0,
+                    1668.0,
+                    929.0,
+                    1668.0,
+                    929.0,
+                    1708.0,
+                    351.0,
+                    1708.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1719.0,
+                    1408.0,
+                    1719.0,
+                    1408.0,
+                    1759.0,
+                    294.0,
+                    1759.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    355.0,
+                    1753.0,
+                    1406.0,
+                    1753.0,
+                    1406.0,
+                    1787.0,
+                    355.0,
+                    1787.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    355.0,
+                    1782.0,
+                    1404.0,
+                    1782.0,
+                    1404.0,
+                    1816.0,
+                    355.0,
+                    1816.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    353.0,
+                    1810.0,
+                    1287.0,
+                    1810.0,
+                    1287.0,
+                    1850.0,
+                    353.0,
+                    1850.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1861.0,
+                    1406.0,
+                    1861.0,
+                    1406.0,
+                    1899.0,
+                    292.0,
+                    1899.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    351.0,
+                    1893.0,
+                    1020.0,
+                    1893.0,
+                    1020.0,
+                    1933.0,
+                    351.0,
+                    1933.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1942.0,
+                    1412.0,
+                    1942.0,
+                    1412.0,
+                    1988.0,
+                    292.0,
+                    1988.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    349.0,
+                    1976.0,
+                    813.0,
+                    1976.0,
+                    813.0,
+                    2014.0,
+                    349.0,
+                    2014.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 11,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 2,
+                "poly": [
+                    836,
+                    2061,
+                    865,
+                    2061,
+                    865,
+                    2086,
+                    836,
+                    2086
+                ],
+                "score": 0.827
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    291,
+                    127,
+                    1410,
+                    127,
+                    1410,
+                    2028,
+                    291,
+                    2028
+                ],
+                "score": 0.641
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    446,
+                    616,
+                    497,
+                    616,
+                    497,
+                    645,
+                    446,
+                    645
+                ],
+                "score": 0.63,
+                "latex": "\\mathrm { H u ^ { * } }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    861,
+                    1977,
+                    940,
+                    1977,
+                    940,
+                    2005,
+                    861,
+                    2005
+                ],
+                "score": 0.34,
+                "latex": "{ ' } G R L { + } ,"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    832.0,
+                    2058.0,
+                    870.0,
+                    2058.0,
+                    870.0,
+                    2096.0,
+                    832.0,
+                    2096.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    202.0,
+                    1407.0,
+                    202.0,
+                    1407.0,
+                    242.0,
+                    294.0,
+                    242.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    353.0,
+                    233.0,
+                    844.0,
+                    233.0,
+                    844.0,
+                    267.0,
+                    353.0,
+                    267.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    281.0,
+                    1407.0,
+                    281.0,
+                    1407.0,
+                    325.0,
+                    292.0,
+                    325.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    351.0,
+                    317.0,
+                    874.0,
+                    317.0,
+                    874.0,
+                    352.0,
+                    351.0,
+                    352.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    367.0,
+                    1231.0,
+                    367.0,
+                    1231.0,
+                    404.0,
+                    292.0,
+                    404.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    290.0,
+                    417.0,
+                    1409.0,
+                    417.0,
+                    1409.0,
+                    460.0,
+                    290.0,
+                    460.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    353.0,
+                    450.0,
+                    1166.0,
+                    450.0,
+                    1166.0,
+                    490.0,
+                    353.0,
+                    490.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    502.0,
+                    1407.0,
+                    502.0,
+                    1407.0,
+                    542.0,
+                    294.0,
+                    542.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    353.0,
+                    533.0,
+                    1409.0,
+                    533.0,
+                    1409.0,
+                    573.0,
+                    353.0,
+                    573.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    353.0,
+                    565.0,
+                    565.0,
+                    565.0,
+                    565.0,
+                    598.0,
+                    353.0,
+                    598.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    617.0,
+                    445.0,
+                    617.0,
+                    445.0,
+                    650.0,
+                    296.0,
+                    650.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    498.0,
+                    617.0,
+                    1405.0,
+                    617.0,
+                    1405.0,
+                    650.0,
+                    498.0,
+                    650.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    353.0,
+                    646.0,
+                    1407.0,
+                    646.0,
+                    1407.0,
+                    686.0,
+                    353.0,
+                    686.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    355.0,
+                    679.0,
+                    726.0,
+                    679.0,
+                    726.0,
+                    713.0,
+                    355.0,
+                    713.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    727.0,
+                    1407.0,
+                    727.0,
+                    1407.0,
+                    767.0,
+                    294.0,
+                    767.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    355.0,
+                    761.0,
+                    935.0,
+                    761.0,
+                    935.0,
+                    794.0,
+                    355.0,
+                    794.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    813.0,
+                    1405.0,
+                    813.0,
+                    1405.0,
+                    846.0,
+                    296.0,
+                    846.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    353.0,
+                    842.0,
+                    1066.0,
+                    842.0,
+                    1066.0,
+                    882.0,
+                    353.0,
+                    882.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    892.0,
+                    1407.0,
+                    892.0,
+                    1407.0,
+                    932.0,
+                    294.0,
+                    932.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    351.0,
+                    925.0,
+                    1142.0,
+                    925.0,
+                    1142.0,
+                    961.0,
+                    351.0,
+                    961.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    975.0,
+                    1409.0,
+                    975.0,
+                    1409.0,
+                    1013.0,
+                    292.0,
+                    1013.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    355.0,
+                    1009.0,
+                    1253.0,
+                    1009.0,
+                    1253.0,
+                    1042.0,
+                    355.0,
+                    1042.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1059.0,
+                    1409.0,
+                    1059.0,
+                    1409.0,
+                    1098.0,
+                    294.0,
+                    1098.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    355.0,
+                    1090.0,
+                    431.0,
+                    1090.0,
+                    431.0,
+                    1125.0,
+                    355.0,
+                    1125.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1140.0,
+                    1405.0,
+                    1140.0,
+                    1405.0,
+                    1180.0,
+                    294.0,
+                    1180.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    353.0,
+                    1173.0,
+                    649.0,
+                    1173.0,
+                    649.0,
+                    1207.0,
+                    353.0,
+                    1207.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1223.0,
+                    1407.0,
+                    1223.0,
+                    1407.0,
+                    1263.0,
+                    294.0,
+                    1263.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    353.0,
+                    1257.0,
+                    630.0,
+                    1257.0,
+                    630.0,
+                    1290.0,
+                    353.0,
+                    1290.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1303.0,
+                    1409.0,
+                    1303.0,
+                    1409.0,
+                    1344.0,
+                    292.0,
+                    1344.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    349.0,
+                    1336.0,
+                    709.0,
+                    1336.0,
+                    709.0,
+                    1373.0,
+                    349.0,
+                    1373.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1388.0,
+                    1151.0,
+                    1388.0,
+                    1151.0,
+                    1428.0,
+                    296.0,
+                    1428.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1440.0,
+                    1407.0,
+                    1440.0,
+                    1407.0,
+                    1480.0,
+                    294.0,
+                    1480.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    355.0,
+                    1474.0,
+                    992.0,
+                    1474.0,
+                    992.0,
+                    1507.0,
+                    355.0,
+                    1507.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1524.0,
+                    1407.0,
+                    1524.0,
+                    1407.0,
+                    1563.0,
+                    294.0,
+                    1563.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    353.0,
+                    1555.0,
+                    721.0,
+                    1555.0,
+                    721.0,
+                    1594.0,
+                    353.0,
+                    1594.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1603.0,
+                    1405.0,
+                    1603.0,
+                    1405.0,
+                    1642.0,
+                    296.0,
+                    1642.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    351.0,
+                    1636.0,
+                    753.0,
+                    1636.0,
+                    753.0,
+                    1673.0,
+                    351.0,
+                    1673.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1688.0,
+                    1407.0,
+                    1688.0,
+                    1407.0,
+                    1728.0,
+                    294.0,
+                    1728.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    353.0,
+                    1719.0,
+                    1405.0,
+                    1719.0,
+                    1405.0,
+                    1759.0,
+                    353.0,
+                    1759.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    353.0,
+                    1749.0,
+                    838.0,
+                    1749.0,
+                    838.0,
+                    1788.0,
+                    353.0,
+                    1788.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1801.0,
+                    1407.0,
+                    1801.0,
+                    1407.0,
+                    1840.0,
+                    294.0,
+                    1840.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    351.0,
+                    1830.0,
+                    1407.0,
+                    1830.0,
+                    1407.0,
+                    1872.0,
+                    351.0,
+                    1872.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    349.0,
+                    1861.0,
+                    649.0,
+                    1861.0,
+                    649.0,
+                    1899.0,
+                    349.0,
+                    1899.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1913.0,
+                    1407.0,
+                    1913.0,
+                    1407.0,
+                    1953.0,
+                    294.0,
+                    1953.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    353.0,
+                    1945.0,
+                    1407.0,
+                    1945.0,
+                    1407.0,
+                    1984.0,
+                    353.0,
+                    1984.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    353.0,
+                    1974.0,
+                    860.0,
+                    1974.0,
+                    860.0,
+                    2013.0,
+                    353.0,
+                    2013.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    941.0,
+                    1974.0,
+                    1134.0,
+                    1974.0,
+                    1134.0,
+                    2013.0,
+                    941.0,
+                    2013.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 12,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 2,
+                "poly": [
+                    836,
+                    2061,
+                    865,
+                    2061,
+                    865,
+                    2086,
+                    836,
+                    2086
+                ],
+                "score": 0.821
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    294,
+                    1023,
+                    1405,
+                    1023,
+                    1405,
+                    1090,
+                    294,
+                    1090
+                ],
+                "score": 0.568
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1552,
+                    1402,
+                    1552,
+                    1402,
+                    1644,
+                    298,
+                    1644
+                ],
+                "score": 0.564
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    571,
+                    1402,
+                    571,
+                    1402,
+                    665,
+                    297,
+                    665
+                ],
+                "score": 0.564
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1355,
+                    1403,
+                    1355,
+                    1403,
+                    1451,
+                    297,
+                    1451
+                ],
+                "score": 0.561
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    296,
+                    1107,
+                    1401,
+                    1107,
+                    1401,
+                    1172,
+                    296,
+                    1172
+                ],
+                "score": 0.56
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    291,
+                    1469,
+                    1403,
+                    1469,
+                    1403,
+                    1533,
+                    291,
+                    1533
+                ],
+                "score": 0.556
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    911,
+                    1402,
+                    911,
+                    1402,
+                    1005,
+                    298,
+                    1005
+                ],
+                "score": 0.552
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    296,
+                    1190,
+                    1400,
+                    1190,
+                    1400,
+                    1254,
+                    296,
+                    1254
+                ],
+                "score": 0.543
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    398,
+                    1402,
+                    398,
+                    1402,
+                    553,
+                    297,
+                    553
+                ],
+                "score": 0.539
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    285,
+                    1400,
+                    285,
+                    1400,
+                    378,
+                    297,
+                    378
+                ],
+                "score": 0.539
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    293,
+                    202,
+                    1404,
+                    202,
+                    1404,
+                    267,
+                    293,
+                    267
+                ],
+                "score": 0.537
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    294,
+                    1273,
+                    1403,
+                    1273,
+                    1403,
+                    1337,
+                    294,
+                    1337
+                ],
+                "score": 0.529
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    291,
+                    1665,
+                    1403,
+                    1665,
+                    1403,
+                    1729,
+                    291,
+                    1729
+                ],
+                "score": 0.513
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    295,
+                    828,
+                    1401,
+                    828,
+                    1401,
+                    892,
+                    295,
+                    892
+                ],
+                "score": 0.507
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1944,
+                    1397,
+                    1944,
+                    1397,
+                    2008,
+                    297,
+                    2008
+                ],
+                "score": 0.497
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    293,
+                    1748,
+                    1404,
+                    1748,
+                    1404,
+                    1812,
+                    293,
+                    1812
+                ],
+                "score": 0.494
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1831,
+                    1398,
+                    1831,
+                    1398,
+                    1924,
+                    298,
+                    1924
+                ],
+                "score": 0.454
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    684,
+                    1406,
+                    684,
+                    1406,
+                    810,
+                    297,
+                    810
+                ],
+                "score": 0.294
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    832.0,
+                    2058.0,
+                    871.0,
+                    2058.0,
+                    871.0,
+                    2096.0,
+                    832.0,
+                    2096.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    1024.0,
+                    1407.0,
+                    1024.0,
+                    1407.0,
+                    1060.0,
+                    297.0,
+                    1060.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    354.0,
+                    1053.0,
+                    1340.0,
+                    1053.0,
+                    1340.0,
+                    1093.0,
+                    354.0,
+                    1093.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1551.0,
+                    1406.0,
+                    1551.0,
+                    1406.0,
+                    1588.0,
+                    294.0,
+                    1588.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    353.0,
+                    1584.0,
+                    1407.0,
+                    1584.0,
+                    1407.0,
+                    1617.0,
+                    353.0,
+                    1617.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    351.0,
+                    1607.0,
+                    430.0,
+                    1607.0,
+                    430.0,
+                    1648.0,
+                    351.0,
+                    1648.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    571.0,
+                    1408.0,
+                    571.0,
+                    1408.0,
+                    606.0,
+                    293.0,
+                    606.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    355.0,
+                    603.0,
+                    1408.0,
+                    603.0,
+                    1408.0,
+                    637.0,
+                    355.0,
+                    637.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    351.0,
+                    630.0,
+                    430.0,
+                    630.0,
+                    430.0,
+                    667.0,
+                    351.0,
+                    667.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1353.0,
+                    1408.0,
+                    1353.0,
+                    1408.0,
+                    1394.0,
+                    292.0,
+                    1394.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    355.0,
+                    1387.0,
+                    1407.0,
+                    1387.0,
+                    1407.0,
+                    1421.0,
+                    355.0,
+                    1421.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    355.0,
+                    1418.0,
+                    948.0,
+                    1418.0,
+                    948.0,
+                    1452.0,
+                    355.0,
+                    1452.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    1107.0,
+                    1405.0,
+                    1107.0,
+                    1405.0,
+                    1143.0,
+                    297.0,
+                    1143.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    354.0,
+                    1136.0,
+                    1255.0,
+                    1136.0,
+                    1255.0,
+                    1176.0,
+                    354.0,
+                    1176.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    1468.0,
+                    1404.0,
+                    1468.0,
+                    1404.0,
+                    1504.0,
+                    297.0,
+                    1504.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    355.0,
+                    1500.0,
+                    925.0,
+                    1500.0,
+                    925.0,
+                    1536.0,
+                    355.0,
+                    1536.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    911.0,
+                    1406.0,
+                    911.0,
+                    1406.0,
+                    946.0,
+                    293.0,
+                    946.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    356.0,
+                    945.0,
+                    1404.0,
+                    945.0,
+                    1404.0,
+                    975.0,
+                    356.0,
+                    975.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    356.0,
+                    973.0,
+                    536.0,
+                    973.0,
+                    536.0,
+                    1004.0,
+                    356.0,
+                    1004.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1187.0,
+                    1405.0,
+                    1187.0,
+                    1405.0,
+                    1228.0,
+                    294.0,
+                    1228.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    355.0,
+                    1219.0,
+                    1158.0,
+                    1219.0,
+                    1158.0,
+                    1255.0,
+                    355.0,
+                    1255.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    397.0,
+                    1408.0,
+                    397.0,
+                    1408.0,
+                    436.0,
+                    295.0,
+                    436.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    355.0,
+                    426.0,
+                    1404.0,
+                    426.0,
+                    1404.0,
+                    465.0,
+                    355.0,
+                    465.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    356.0,
+                    458.0,
+                    1408.0,
+                    458.0,
+                    1408.0,
+                    495.0,
+                    356.0,
+                    495.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    354.0,
+                    487.0,
+                    1408.0,
+                    487.0,
+                    1408.0,
+                    527.0,
+                    354.0,
+                    527.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    354.0,
+                    519.0,
+                    1225.0,
+                    519.0,
+                    1225.0,
+                    555.0,
+                    354.0,
+                    555.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    284.0,
+                    1405.0,
+                    284.0,
+                    1405.0,
+                    322.0,
+                    295.0,
+                    322.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    351.0,
+                    313.0,
+                    1406.0,
+                    313.0,
+                    1406.0,
+                    353.0,
+                    351.0,
+                    353.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    354.0,
+                    342.0,
+                    538.0,
+                    342.0,
+                    538.0,
+                    380.0,
+                    354.0,
+                    380.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    202.0,
+                    1405.0,
+                    202.0,
+                    1405.0,
+                    238.0,
+                    297.0,
+                    238.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    351.0,
+                    233.0,
+                    1262.0,
+                    233.0,
+                    1262.0,
+                    268.0,
+                    351.0,
+                    268.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1270.0,
+                    1408.0,
+                    1270.0,
+                    1408.0,
+                    1311.0,
+                    294.0,
+                    1311.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    353.0,
+                    1305.0,
+                    994.0,
+                    1305.0,
+                    994.0,
+                    1337.0,
+                    353.0,
+                    1337.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    1664.0,
+                    1404.0,
+                    1664.0,
+                    1404.0,
+                    1700.0,
+                    297.0,
+                    1700.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    356.0,
+                    1697.0,
+                    952.0,
+                    1697.0,
+                    952.0,
+                    1729.0,
+                    356.0,
+                    1729.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    827.0,
+                    1406.0,
+                    827.0,
+                    1406.0,
+                    866.0,
+                    295.0,
+                    866.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    353.0,
+                    858.0,
+                    642.0,
+                    858.0,
+                    642.0,
+                    892.0,
+                    353.0,
+                    892.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    1944.0,
+                    1402.0,
+                    1944.0,
+                    1402.0,
+                    1980.0,
+                    297.0,
+                    1980.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    353.0,
+                    1976.0,
+                    654.0,
+                    1976.0,
+                    654.0,
+                    2005.0,
+                    353.0,
+                    2005.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1744.0,
+                    1409.0,
+                    1744.0,
+                    1409.0,
+                    1788.0,
+                    294.0,
+                    1788.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    355.0,
+                    1779.0,
+                    927.0,
+                    1779.0,
+                    927.0,
+                    1811.0,
+                    355.0,
+                    1811.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1829.0,
+                    1404.0,
+                    1829.0,
+                    1404.0,
+                    1867.0,
+                    294.0,
+                    1867.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    353.0,
+                    1859.0,
+                    1406.0,
+                    1859.0,
+                    1406.0,
+                    1899.0,
+                    353.0,
+                    1899.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    355.0,
+                    1892.0,
+                    507.0,
+                    1892.0,
+                    507.0,
+                    1923.0,
+                    355.0,
+                    1923.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    686.0,
+                    1406.0,
+                    686.0,
+                    1406.0,
+                    722.0,
+                    295.0,
+                    722.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    354.0,
+                    718.0,
+                    1406.0,
+                    718.0,
+                    1406.0,
+                    751.0,
+                    354.0,
+                    751.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    352.0,
+                    745.0,
+                    1406.0,
+                    745.0,
+                    1406.0,
+                    784.0,
+                    352.0,
+                    784.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    353.0,
+                    776.0,
+                    806.0,
+                    776.0,
+                    806.0,
+                    812.0,
+                    353.0,
+                    812.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 13,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 2,
+                "poly": [
+                    836,
+                    2061,
+                    865,
+                    2061,
+                    865,
+                    2085,
+                    836,
+                    2085
+                ],
+                "score": 0.791
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    291,
+                    202,
+                    1406,
+                    202,
+                    1406,
+                    349,
+                    291,
+                    349
+                ],
+                "score": 0.71
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    836,
+                    2061,
+                    865,
+                    2061,
+                    865,
+                    2086,
+                    836,
+                    2086
+                ],
+                "score": 0.124
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    832.0,
+                    2058.0,
+                    870.0,
+                    2058.0,
+                    870.0,
+                    2097.0,
+                    832.0,
+                    2097.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    832.0,
+                    2058.0,
+                    870.0,
+                    2058.0,
+                    870.0,
+                    2096.0,
+                    832.0,
+                    2096.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    203.0,
+                    1404.0,
+                    203.0,
+                    1404.0,
+                    239.0,
+                    297.0,
+                    239.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    354.0,
+                    233.0,
+                    1222.0,
+                    233.0,
+                    1222.0,
+                    269.0,
+                    354.0,
+                    269.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    283.0,
+                    1405.0,
+                    283.0,
+                    1405.0,
+                    324.0,
+                    294.0,
+                    324.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    354.0,
+                    318.0,
+                    938.0,
+                    318.0,
+                    938.0,
+                    350.0,
+                    354.0,
+                    350.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 14,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    406,
+                    1040,
+                    1403,
+                    1040,
+                    1403,
+                    1367,
+                    406,
+                    1367
+                ],
+                "score": 0.973
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    406,
+                    300,
+                    1404,
+                    300,
+                    1404,
+                    530,
+                    406,
+                    530
+                ],
+                "score": 0.972
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    407,
+                    1419,
+                    1405,
+                    1419,
+                    1405,
+                    1615,
+                    407,
+                    1615
+                ],
+                "score": 0.972
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    406,
+                    699,
+                    1405,
+                    699,
+                    1405,
+                    990,
+                    406,
+                    990
+                ],
+                "score": 0.971
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    406,
+                    580,
+                    1252,
+                    580,
+                    1252,
+                    649,
+                    406,
+                    649
+                ],
+                "score": 0.938
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    363,
+                    659,
+                    671,
+                    659,
+                    671,
+                    689,
+                    363,
+                    689
+                ],
+                "score": 0.91
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    365,
+                    538,
+                    840,
+                    538,
+                    840,
+                    571,
+                    365,
+                    571
+                ],
+                "score": 0.905
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    298,
+                    199,
+                    435,
+                    199,
+                    435,
+                    235,
+                    298,
+                    235
+                ],
+                "score": 0.896
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    364,
+                    257,
+                    580,
+                    257,
+                    580,
+                    289,
+                    364,
+                    289
+                ],
+                "score": 0.896
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    366,
+                    1377,
+                    1200,
+                    1377,
+                    1200,
+                    1408,
+                    366,
+                    1408
+                ],
+                "score": 0.865
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    372,
+                    999,
+                    1399,
+                    999,
+                    1399,
+                    1033,
+                    372,
+                    1033
+                ],
+                "score": 0.82
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    836,
+                    2061,
+                    865,
+                    2061,
+                    865,
+                    2086,
+                    836,
+                    2086
+                ],
+                "score": 0.685
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    836,
+                    2061,
+                    865,
+                    2061,
+                    865,
+                    2086,
+                    836,
+                    2086
+                ],
+                "score": 0.385
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    197.0,
+                    440.0,
+                    197.0,
+                    440.0,
+                    238.0,
+                    296.0,
+                    238.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    832.0,
+                    2058.0,
+                    871.0,
+                    2058.0,
+                    871.0,
+                    2096.0,
+                    832.0,
+                    2096.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    832.0,
+                    2058.0,
+                    871.0,
+                    2058.0,
+                    871.0,
+                    2096.0,
+                    832.0,
+                    2096.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    405.0,
+                    1040.0,
+                    1403.0,
+                    1040.0,
+                    1403.0,
+                    1075.0,
+                    405.0,
+                    1075.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    447.0,
+                    1068.0,
+                    1072.0,
+                    1068.0,
+                    1072.0,
+                    1110.0,
+                    447.0,
+                    1110.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    402.0,
+                    1105.0,
+                    1183.0,
+                    1105.0,
+                    1183.0,
+                    1143.0,
+                    402.0,
+                    1143.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    404.0,
+                    1141.0,
+                    1405.0,
+                    1141.0,
+                    1405.0,
+                    1179.0,
+                    404.0,
+                    1179.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    449.0,
+                    1172.0,
+                    803.0,
+                    1172.0,
+                    803.0,
+                    1205.0,
+                    449.0,
+                    1205.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    406.0,
+                    1210.0,
+                    1403.0,
+                    1210.0,
+                    1403.0,
+                    1241.0,
+                    406.0,
+                    1241.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    449.0,
+                    1239.0,
+                    1216.0,
+                    1239.0,
+                    1216.0,
+                    1277.0,
+                    449.0,
+                    1277.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    409.0,
+                    1274.0,
+                    1405.0,
+                    1274.0,
+                    1405.0,
+                    1311.0,
+                    409.0,
+                    1311.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    450.0,
+                    1304.0,
+                    1403.0,
+                    1304.0,
+                    1403.0,
+                    1340.0,
+                    450.0,
+                    1340.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    448.0,
+                    1336.0,
+                    1077.0,
+                    1336.0,
+                    1077.0,
+                    1370.0,
+                    448.0,
+                    1370.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    406.0,
+                    298.0,
+                    1404.0,
+                    298.0,
+                    1404.0,
+                    334.0,
+                    406.0,
+                    334.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    449.0,
+                    329.0,
+                    802.0,
+                    329.0,
+                    802.0,
+                    365.0,
+                    449.0,
+                    365.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    401.0,
+                    362.0,
+                    1204.0,
+                    362.0,
+                    1204.0,
+                    402.0,
+                    401.0,
+                    402.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    401.0,
+                    398.0,
+                    1407.0,
+                    398.0,
+                    1407.0,
+                    440.0,
+                    401.0,
+                    440.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    450.0,
+                    429.0,
+                    569.0,
+                    429.0,
+                    569.0,
+                    463.0,
+                    450.0,
+                    463.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    401.0,
+                    463.0,
+                    1406.0,
+                    463.0,
+                    1406.0,
+                    506.0,
+                    401.0,
+                    506.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    449.0,
+                    496.0,
+                    596.0,
+                    496.0,
+                    596.0,
+                    533.0,
+                    449.0,
+                    533.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    404.0,
+                    1416.0,
+                    1408.0,
+                    1416.0,
+                    1408.0,
+                    1453.0,
+                    404.0,
+                    1453.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    448.0,
+                    1448.0,
+                    660.0,
+                    1448.0,
+                    660.0,
+                    1485.0,
+                    448.0,
+                    1485.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    405.0,
+                    1483.0,
+                    1405.0,
+                    1483.0,
+                    1405.0,
+                    1519.0,
+                    405.0,
+                    1519.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    450.0,
+                    1512.0,
+                    951.0,
+                    1512.0,
+                    951.0,
+                    1549.0,
+                    450.0,
+                    1549.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    408.0,
+                    1549.0,
+                    1405.0,
+                    1549.0,
+                    1405.0,
+                    1585.0,
+                    408.0,
+                    1585.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    450.0,
+                    1582.0,
+                    926.0,
+                    1582.0,
+                    926.0,
+                    1616.0,
+                    450.0,
+                    1616.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    404.0,
+                    697.0,
+                    1409.0,
+                    697.0,
+                    1409.0,
+                    733.0,
+                    404.0,
+                    733.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    449.0,
+                    729.0,
+                    1406.0,
+                    729.0,
+                    1406.0,
+                    763.0,
+                    449.0,
+                    763.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    449.0,
+                    759.0,
+                    776.0,
+                    759.0,
+                    776.0,
+                    794.0,
+                    449.0,
+                    794.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    405.0,
+                    794.0,
+                    1405.0,
+                    794.0,
+                    1405.0,
+                    831.0,
+                    405.0,
+                    831.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    448.0,
+                    825.0,
+                    1241.0,
+                    825.0,
+                    1241.0,
+                    863.0,
+                    448.0,
+                    863.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    409.0,
+                    860.0,
+                    1408.0,
+                    860.0,
+                    1408.0,
+                    897.0,
+                    409.0,
+                    897.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    451.0,
+                    893.0,
+                    780.0,
+                    893.0,
+                    780.0,
+                    924.0,
+                    451.0,
+                    924.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    408.0,
+                    923.0,
+                    1406.0,
+                    923.0,
+                    1406.0,
+                    965.0,
+                    408.0,
+                    965.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    449.0,
+                    954.0,
+                    1374.0,
+                    954.0,
+                    1374.0,
+                    992.0,
+                    449.0,
+                    992.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    405.0,
+                    578.0,
+                    1251.0,
+                    578.0,
+                    1251.0,
+                    617.0,
+                    405.0,
+                    617.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    402.0,
+                    615.0,
+                    1172.0,
+                    615.0,
+                    1172.0,
+                    652.0,
+                    402.0,
+                    652.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    357.0,
+                    653.0,
+                    675.0,
+                    653.0,
+                    675.0,
+                    696.0,
+                    357.0,
+                    696.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    359.0,
+                    536.0,
+                    846.0,
+                    536.0,
+                    846.0,
+                    575.0,
+                    359.0,
+                    575.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    361.0,
+                    253.0,
+                    586.0,
+                    253.0,
+                    586.0,
+                    292.0,
+                    361.0,
+                    292.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    357.0,
+                    1372.0,
+                    1205.0,
+                    1372.0,
+                    1205.0,
+                    1415.0,
+                    357.0,
+                    1415.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    361.0,
+                    994.0,
+                    1403.0,
+                    994.0,
+                    1403.0,
+                    1039.0,
+                    361.0,
+                    1039.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 15,
+            "width": 1700,
+            "height": 2200
+        }
+    }
+]
\ No newline at end of file
diff --git a/parse/train/x9jS8pX3dkx/x9jS8pX3dkx_content_list.json b/parse/train/x9jS8pX3dkx/x9jS8pX3dkx_content_list.json
new file mode 100644
index 0000000000000000000000000000000000000000..3bf0e1e8641ef70183f671eff6322c9cde0a1ba0
--- /dev/null
+++ b/parse/train/x9jS8pX3dkx/x9jS8pX3dkx_content_list.json
@@ -0,0 +1,921 @@
+[
+    {
+        "type": "text",
+        "text": "Gradient Inversion with Generative Image Prior ",
+        "text_level": 1,
+        "bbox": [
+            204,
+            122,
+            794,
+            147
+        ],
+        "page_idx": 0
+    },
+    {
+        "type": "text",
+        "text": "Jinwoo $\\mathbf { J e o n ^ { 1 * } }$ , Jaechang $\\mathbf { K i m ^ { 2 * } }$ , Kangwook $\\mathbf { L e e ^ { 3 } }$ , Sewoong $\\mathbf { O h ^ { 4 } }$ , Jungseul $\\mathbf { O k } ^ { 1 , 2 }$   \n1 Department of Computer Science & Engineering, Pohang University of Science and Technology 2 Graduate School of Artificial Intelligence, Pohang University of Science and Technology   \n3 Department of Electrical and Computer Engineering, University of Wisconsin-Madison, Madison 4 Paul G. Allen School of Computer Science & Engineering, University of Washington ",
+        "bbox": [
+            186,
+            199,
+            833,
+            272
+        ],
+        "page_idx": 0
+    },
+    {
+        "type": "text",
+        "text": "Abstract ",
+        "text_level": 1,
+        "bbox": [
+            462,
+            308,
+            535,
+            325
+        ],
+        "page_idx": 0
+    },
+    {
+        "type": "text",
+        "text": "Federated Learning (FL) is a distributed learning framework, in which the local data never leaves clients’ devices to preserve privacy, and the server trains models on the data via accessing only the gradients of those local data. Without further privacy mechanisms such as differential privacy, this leaves the system vulnerable against an attacker who inverts those gradients to reveal clients’ sensitive data. However, a gradient is often insufficient to reconstruct the user data without any prior knowledge. By exploiting a generative model pretrained on the data distribution, we demonstrate that data privacy can be easily breached. Further, when such prior knowledge is unavailable, we investigate the possibility of learning the prior from a sequence of gradients seen in the process of FL training. We experimentally show that the prior in a form of generative model is learnable from iterative interactions in FL. Our findings strongly suggest that additional mechanisms are necessary to prevent privacy leakage in FL. ",
+        "bbox": [
+            233,
+            340,
+            766,
+            520
+        ],
+        "page_idx": 0
+    },
+    {
+        "type": "text",
+        "text": "1 Introduction ",
+        "text_level": 1,
+        "bbox": [
+            174,
+            546,
+            310,
+            564
+        ],
+        "page_idx": 0
+    },
+    {
+        "type": "text",
+        "text": "Federated learning (FL) is an emerging framework for distributed learning, where central server aggregates model updates, rather than user data, from end users [5, 17]. The main premise of federated learning is that this particular way of distributed learning can protect users’ data privacy as there is no explicit data shared by the end users with the central server. ",
+        "bbox": [
+            174,
+            579,
+            823,
+            633
+        ],
+        "page_idx": 0
+    },
+    {
+        "type": "text",
+        "text": "However, a recent line of work [34, 31, 9, 29] demonstrates that one may recover the private user data used for training by observing the gradients. This process of recovering the training data from gradients, so-called gradient inversion, poses a huge threat to the federated learning community, as it may imply the fundamental flaw of its main premise. ",
+        "bbox": [
+            174,
+            640,
+            825,
+            695
+        ],
+        "page_idx": 0
+    },
+    {
+        "type": "text",
+        "text": "Even more worryingly, recent works suggest that such gradient inversion attacks can be made even stronger if certain side-information is available. For instance, Geiping et al. [9] show that if the attacker knows a prior that user data consists of natural images, then the gradient inversion attack can leverage such prior, achieving a more accurate recovery of the user data. Another instance is when batch norm statistics are available at the attacker in addition to gradients. This can actually happen if the end users share their local batch norm statistics as in [17]. Yin et al. [29] show that such batch normalization statistics can significantly improve the strength of the gradient inversion attack, enabling precise recovery of high-resolution images. ",
+        "bbox": [
+            174,
+            703,
+            825,
+            814
+        ],
+        "page_idx": 0
+    },
+    {
+        "type": "text",
+        "text": "In this paper, we systematically study how one can maximally utilize and even obtain the prior information when inverting gradients. We first consider the case that the attacker has a generative model pretrained on the exact or approximate distribution of the user data as a prior. For this, we propose an efficient gradient inversion algorithm that utilizes the generative model prior. In particular, the algorithm consists of two steps, in which the first step searches the latent space (of lower dimension) defined by the generative model instead of the ambient input space (of higher dimension), and then the second step adapts the generative model to each input given the gradient. Each step provides substantial improvement in the reconstruction. We name the algorithm as gradient inversion in alternative spaces (GIAS). Figure 1 represents reconstruction results with the proposed method and existing one. ",
+        "bbox": [
+            174,
+            820,
+            825,
+            876
+        ],
+        "page_idx": 0
+    },
+    {
+        "type": "image",
+        "img_path": "images/b81bb14fb3a48de6867d8a38581f8cf29cb13a06887b9c8990eaed2b11ce75bc.jpg",
+        "image_caption": [
+            "existing method [9] ",
+            "Figure 1: An example showing the superiority of GIAS compared to existing method. Images of the authors are reconstructed from gradients by exploiting a generative model pretrained on human face images. "
+        ],
+        "image_footnote": [],
+        "bbox": [
+            302,
+            87,
+            692,
+            366
+        ],
+        "page_idx": 1
+    },
+    {
+        "type": "text",
+        "text": "",
+        "bbox": [
+            174,
+            430,
+            825,
+            513
+        ],
+        "page_idx": 1
+    },
+    {
+        "type": "text",
+        "text": "We then consider a realistic scenario in which the user data distribution is not known in advance, and thus the attacker needs to learn it from gradients. For this scenario, we develop a meta-learning framework, called gradient inversion to meta-learn (GIML), which learns a generative model on user data from observing and inverting multiple gradients computed on the data, e.g. across different FL epochs or participating nodes. Our experimental results demonstrate that one can learn a generative model via GIML and reconstruct data by making use of the learned generative model. ",
+        "bbox": [
+            174,
+            520,
+            825,
+            603
+        ],
+        "page_idx": 1
+    },
+    {
+        "type": "text",
+        "text": "This implies a great threat on privacy leakage in FL since our methods can be applied for any data type in most FL scenarios unless a specialized architecture prevents the gradient leakage explicitly, e.g., [18]. ",
+        "bbox": [
+            176,
+            609,
+            825,
+            651
+        ],
+        "page_idx": 1
+    },
+    {
+        "type": "text",
+        "text": "Our main contributions are as follows: ",
+        "bbox": [
+            176,
+            657,
+            426,
+            672
+        ],
+        "page_idx": 1
+    },
+    {
+        "type": "text",
+        "text": "• We introduce GIAS that fully utilizes a pretrained generative model to invert gradient. In addition, we propose GIML which can train generative model from gradients only in FL. • We demonstrate significant privacy leakage occurring by GIAS with a pretrained generative model in various FL scenarios which are challenging to other existing methods, e.g., [9, 29]. • We experimentally show that GIML can learn a generative model on the user data from only gradients, which provides the same level of data recovery with a given pretrained model. To our best knowledge, GIML is the first capable of learning explicit prior on a set of gradient inversion tasks. We note that a gradient inversion technique defines a standard on defence mechanism in FL for privacy [28]. By substantiating that our proposed methods are able to break down defense mechanisms that were safe according to the previous standard, we give a strong warning to the FL community to use a higher standard defined by our attack methods, and raise the necessity of a more conservative choice of defense mechanisms. ",
+        "bbox": [
+            215,
+            686,
+            825,
+            900
+        ],
+        "page_idx": 1
+    },
+    {
+        "type": "text",
+        "text": "2 Related work ",
+        "text_level": 1,
+        "bbox": [
+            174,
+            89,
+            316,
+            106
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "text",
+        "text": "Privacy attacks in FL. Early works [19, 24] investigate membership inference from gradients to check the possibility of privacy leakage in FL. Phong et al. [21] demonstrate that it is possible to reconstruct detailed input image when FL trains a shallow network such as single-layer perceptron. Fan et al. [7] and Zhu and Blaschko [32] consider a wider class of learning model and propose an analytical approach solving a sequence of linear systems to reveal the output of each layer recursively. To study the limit of the gradient inversion in practical scenarios of training deep networks via FL, a sequence of effort has been made formulating optimization problem to minimize discrepancy comparing gradients from true data and reconstructed data [9, 27, 29, 31, 34]. ",
+        "bbox": [
+            173,
+            119,
+            826,
+            231
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "text",
+        "text": "Gradient inversion with prior. The optimization-based approaches are particularly useful as one can easily utilize prior knowledge by adding regularization terms, e.g., total variation [27, 9] and BN statistics [29], or changing discrepancy measure [9] . In [29], a privacy attack technique using a generative model is introduced. They however require a pretrained model, while we propose a meta learning framework training generative model from gradients only. In addition, our method of inverting gradient maximally exploit a given generative model by alternating search spaces, which are analogous to the state-of-the-art GAN inversion techniques [3, 4, 33]. ",
+        "bbox": [
+            174,
+            243,
+            825,
+            342
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "text",
+        "text": "Generative model revealing private data. Training a generative model with transmitted gradients also demonstrates privacy leakage in FL. Hitaj et al. [11] introduce an algorithm to train a GAN regarding shared model in FL framework as a discriminator. Wang et al. [27] use reconstructed data from gradient to train a GAN. Those works require some auxiliary dataset given in advance to enable the training of GAN, while we train a generative model using transmitted gradients only. Also, we not only train a generative model but also utilize it for reconstruction, while the generative models in [11, 27] are not used for the reconstruction. Hence, in our approach, the generative model and reconstruction can be improved interactively to each other as shown in Figure 6. In addition, [27] is less sample-efficient than ours in a sense that they use gradients to reconstruct images and then train a generative model with the reconstructed images, i.e., if the reconstruction fails, then the corresponding update of the generative model fails too, whereas we train the generative model directly from gradients. ",
+        "bbox": [
+            173,
+            354,
+            825,
+            521
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "text",
+        "text": "3 Problem formulation ",
+        "text_level": 1,
+        "bbox": [
+            174,
+            539,
+            382,
+            556
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "text",
+        "text": "In this section, we formally describe the gradient inversion (GI) problem. Consider a standard supervised learning for classification, which optimizes neural network model $f _ { \\theta }$ parameterized by $\\theta$ as follows: ",
+        "bbox": [
+            173,
+            569,
+            823,
+            609
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "equation",
+        "img_path": "images/362bb2549166ec2d31728f5dcb9826c3a3fcf3a97f82757a1fd14ce896277f44.jpg",
+        "text": "$$\n\\operatorname* { m i n } _ { \\theta } \\sum _ { ( x , y ) \\in \\mathcal { D } } \\ell ( f _ { \\theta } ( x ) , y ) \\ ,\n$$",
+        "text_format": "latex",
+        "bbox": [
+            411,
+            606,
+            584,
+            642
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "text",
+        "text": "where $\\ell$ is a point-wise loss function and $\\mathcal { D }$ is a dataset of input $x \\in \\mathbb { R } ^ { m }$ and label $y \\in \\{ 0 , 1 \\} ^ { L }$ (one-hot vector). In federated learning framework, each node reports the gradient of $\\ell ( f _ { \\theta } ( x ) , y )$ for sampled data $( x , y )$ ’s instead of directly transferring the data. The problem of inverting gradient is to reconstruct the sampled data used to compute the reported gradient. Specifically, when a node computes the gradient $g$ using a batch $\\{ ( x _ { 1 } ^ { * } , y _ { 1 } ^ { * } ) , . . . , ( x _ { B } ^ { * } , y _ { B } ^ { * } ) \\}$ , i.e., $\\begin{array} { r } { \\boldsymbol { g } = \\frac { \\top } { B } \\sum _ { j = 1 } ^ { B } \\dot { \\nabla } \\ell ( f _ { \\theta } ( x _ { j } ^ { * } ) , y _ { j } ^ { * } ) } \\end{array}$ we consider the following problem of inverting gradient: ",
+        "bbox": [
+            173,
+            642,
+            825,
+            732
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "equation",
+        "img_path": "images/cfa878f19bead500c62e58fbfed4cec35bc00995d99a229baf711ab52a2756ca.jpg",
+        "text": "$$\n\\operatorname* { m i n } _ { ( x _ { 1 } , y _ { 1 } ) , \\cdots , ( x _ { B } , y _ { B } ) } d \\left( \\frac { 1 } { B } \\sum _ { j = 1 } ^ { B } \\nabla \\ell ( f _ { \\theta } ( x _ { j } ) , y _ { j } ) , g \\right) \\ ,\n$$",
+        "text_format": "latex",
+        "bbox": [
+            334,
+            734,
+            661,
+            787
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "text",
+        "text": "where $d ( \\cdot , \\cdot )$ is a measure of the discrepancy between two gradient, e.g., $\\ell _ { 2 }$ -distance [34, 29] or negative cosine similarity [9]. It is known that label $y$ can be almost accurately recovered by simple methods just observing the gradient at the last layer [31, 29], while reconstructing input $x$ remains still challenging as it is often under-determined even when the true label is given. For simplicity, we hence focus on the following minimization to reveal the inputs from the gradient given the true labels: ",
+        "bbox": [
+            174,
+            790,
+            825,
+            864
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "equation",
+        "img_path": "images/1cf3e2d7d9912e7dd5a49750a364cc9f0c98dfe05ef57fff258b6aa478bd5621.jpg",
+        "text": "$$\n\\operatorname* { m i n } _ { \\substack { x _ { 1 } , \\ldots , x _ { B } \\in \\mathbb { R } ^ { m } } } c \\left( x _ { 1 } , . . . , x _ { B } ; \\theta , g \\right) ,\n$$",
+        "text_format": "latex",
+        "bbox": [
+            392,
+            872,
+            604,
+            895
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "text",
+        "text": "where we denote by $c \\left( x _ { 1 } , . . . , x _ { B } ; \\theta , g \\right)$ the cost function in (2) given $y _ { j } = y _ { j } ^ { * }$ for each $j = 1 , . . . , B$ ",
+        "bbox": [
+            178,
+            897,
+            820,
+            912
+        ],
+        "page_idx": 2
+    },
+    {
+        "type": "text",
+        "text": "4 Methods ",
+        "text_level": 1,
+        "bbox": [
+            174,
+            89,
+            279,
+            106
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "The key challenge of inverting gradient is that solving (2) is often under-determined, i.e., a gradient contains only insufficient information to recover data. Such an issue is observed even when the dimension of gradient is much larger than that of input data. Indeed, Zhu and Blaschko [32] show that there exist a pair of different data having the same gradient, so called twin data, even when the learning model is large. To alleviate this issue, a set of prior knowledge on the nature of data can be considered. ",
+        "bbox": [
+            173,
+            119,
+            825,
+            204
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "When inverting images, Geiping et al. [9] propose to add the total variation regularization $R _ { \\mathrm { T V } } ( x )$ to the cost function in (3) since neighboring pixels of natural images are likely to have similar values. More formally, ",
+        "bbox": [
+            174,
+            209,
+            825,
+            252
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "equation",
+        "img_path": "images/86838b25959935fbcb8653f63ca47434c685281762b0b64e6f57729df7a4c54b.jpg",
+        "text": "$$\nR _ { \\mathrm { T V } } ( x ) : = \\sum _ { ( i , j ) } \\sum _ { ( i ^ { \\prime } , j ^ { \\prime } ) \\in \\partial ( i , j ) } \\Vert x ( i , j ) - x ( i ^ { \\prime } , j ^ { \\prime } ) \\Vert ^ { 2 } ,\n$$",
+        "text_format": "latex",
+        "bbox": [
+            331,
+            256,
+            665,
+            292
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "where $\\partial ( i , j )$ is the set of neighbors of $( i , j )$ . This method is limited to the natural image data. ",
+        "bbox": [
+            171,
+            297,
+            789,
+            314
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "For general type of data, one can consider exploiting the batch normalization (BN) statistics from nodes. This is available in the case that the server wants to utilize batch normalization (BN) in $\\mathrm { F L }$ , and thus collects the BN statistics (mean and variance) of batch from each node, in addition, with every gradient report [17]. To be specific, Yin et al. [29] propose to employ the regularizer $R _ { \\mathrm { B N } } ( x _ { 1 } , . . . , x _ { B } ; \\theta )$ which quantifies the discrepancy between the BN statistics of estimated $x _ { j }$ ’s and those of true $\\boldsymbol { x } _ { j } ^ { * }$ ’s on each layer of the learning model. More formally, ",
+        "bbox": [
+            173,
+            319,
+            826,
+            404
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "equation",
+        "img_path": "images/3657e8e4fe2b5bbe619f665afa927d3834c7b319b9b07b5ca1c13ed8a2dea5ff.jpg",
+        "text": "$$\nR _ { \\mathrm { B N } } ( x _ { 1 } , . . . , x _ { B } ; \\theta ) : = \\sum _ { l } \\| \\mu _ { l } - \\mu _ { l , \\mathrm { e x a c t } } \\| _ { 2 } + \\| \\sigma _ { l } ^ { 2 } - \\sigma _ { l , \\mathrm { e x a c t } } ^ { 2 } \\| _ { 2 } ,\n$$",
+        "text_format": "latex",
+        "bbox": [
+            300,
+            410,
+            691,
+            444
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "where $\\mu _ { l } ( x _ { 1 } , . . . , x _ { B } ; \\theta )$ and $\\sigma _ { l } ^ { 2 } ( x _ { 1 } , . . . , x _ { B } ; \\theta )$ (resp. $\\mu _ { l , \\mathrm { { e x a c t } } } ( x _ { 1 } ^ { * } , . . . , x _ { B } ^ { * } ; \\theta )$ and $\\sigma _ { l , \\mathrm { { e x a c t } } } ^ { 2 } ( x _ { 1 } ^ { * } , . . . , x _ { B } ^ { * } ; \\theta ) )$ are the mean and variance of $l$ -th layer feature maps for the estimated batch $x _ { 1 } , . . . , x _ { B }$ (resp. the true batch $x _ { 1 } ^ { * } , . . . , x _ { B } ^ { * } )$ given $\\theta$ . This is available only if clients agree to report their exact BN statistics at every round. But not every FL framework report BN statistics [15, 2]. In that case, Yin et al. [29] also propose to use the BN statistics over the entire data distribution as a proxy of the true BN statistics, and reports that the gain from the approximated BN statistics is comparable to that from the exact ones. The applicability of $R _ { \\mathrm { B N } }$ with the approximated BN statistics is still limited as the proxy needs to be additionally recomputed over the entire data distribution at every change of $\\theta$ . However, this demonstrates the significant impact of knowing the data distribution in the gradient inversion and motivates our methods using and learning a generative model on the user data, described in what follows. ",
+        "bbox": [
+            173,
+            449,
+            826,
+            604
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "4.1 Gradient inversion with trained generative model ",
+        "text_level": 1,
+        "bbox": [
+            173,
+            619,
+            557,
+            635
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "Consider a decent generative model $G _ { w } : \\mathbb { R } ^ { k } \\mapsto \\mathbb { R } ^ { m }$ trained on the approximate (possibly exact) distribution of user data $\\mathcal { D }$ such that $x ^ { * } \\approx G _ { w } ( z ^ { * } )$ for $( x ^ { \\ast } , \\cdot ) \\in \\bar { \\mathcal { D } }$ and its latent code $z ^ { * } =$ $\\mathrm { a r g m i n } _ { z } \\| G _ { w } ( z ) - x ^ { * } \\|$ . To fully utilize such a pretrained generative model, we propose gradient inversion in alternative spaces (GIAS), of which pseudocode is presented in Appendix A, which performs latent space search over $z$ and then parameter space search over $w$ . We also illustrate the overall procedure of GIAS in Figure 2. ",
+        "bbox": [
+            173,
+            643,
+            826,
+            729
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "Latent space search. Note that the latent space is typically much smaller than the ambient input space, i.e., $k \\ll m$ , for instances, DCGAN [25] of $k = 1 0 0$ and StyleGAN [12] of $k = 5 1 2 \\times 1 6$ for image data of $m = ( \\mathrm { w i d t h } ) \\times ( \\mathrm { h e i g h t } ) \\times ( \\mathrm { c o l o r } )$ such as $3 2 \\times 3 2 \\times 3$ , $2 5 6 \\times 2 5 6 \\times 3$ , or larger. Using such a pretrained generative model with $k \\ll m$ , the under-determined issues of (3) can be directly mitigated by narrowing down the searching space from $\\mathbb { R } ^ { m }$ to $\\{ G _ { w } ( z ) : z \\in \\mathbb { R } ^ { k } \\}$ . Hence, GIAS first performs the latent space search in the followings: ",
+        "bbox": [
+            173,
+            742,
+            826,
+            827
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "equation",
+        "img_path": "images/f211e132f6ecc08b1bb03ea07a719bb36e5634f96711de6ef5fb6e6d12333b9b.jpg",
+        "text": "$$\n\\operatorname* { m i n } _ { \\substack { z _ { 1 } , \\ldots , z _ { B } \\in \\mathbb { R } ^ { k } } } c \\left( G _ { w } ( z _ { 1 } ) , . . . , G _ { w } ( z _ { B } ) \\right) .\n$$",
+        "text_format": "latex",
+        "bbox": [
+            372,
+            832,
+            624,
+            857
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "text",
+        "text": "Considering a canonical class of neural network model $f _ { \\theta }$ , we can show that the reconstruction of $x ^ { * }$ by latent space search in (5) aligns with that by input space search in (3) if the generative model $G _ { w }$ approximates input data with small enough error. ",
+        "bbox": [
+            174,
+            868,
+            823,
+            911
+        ],
+        "page_idx": 3
+    },
+    {
+        "type": "image",
+        "img_path": "images/86214e463c23aef68db3947ced6b1f76f55c86ba1e2dd66943856f0c1dfa79be.jpg",
+        "image_caption": [
+            "Figure 2: An overview of GIAS. GIAS optimizes a latent code $z$ and generative model parameters $w$ to reconstruct the data which matches the gradient. "
+        ],
+        "image_footnote": [],
+        "bbox": [
+            245,
+            94,
+            750,
+            313
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "text",
+        "text": "Property 1. For an input data $x ^ { * } \\in [ 0 , 1 ] ^ { m }$ consider the gradient inversion problem of minimizing cost c in (3), where a canonical form of deep learning for classification is considered and the discrepancy measure $d$ is $\\ell _ { 2 }$ -distance. Suppose that it has the unique global minimizer at $x ^ { * }$ . Let $\\varepsilon \\geq 0$ be the approximation error bound on $x ^ { * }$ for generative model $\\mathring { G } _ { w } : [ 0 , 1 ] ^ { k } \\mapsto [ 0 , 1 ] ^ { m }$ Then, there exists $\\delta ( \\varepsilon ) \\geq 0$ such that for any $z ^ { * } \\in \\arg \\operatorname* { m i n } _ { z } c ( G _ { w } ( z ) ) ,$ , ",
+        "bbox": [
+            173,
+            390,
+            825,
+            462
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "equation",
+        "img_path": "images/1bdcb001746faf4040209eaa567ad172a6b5dd5ddb3925f3fe34e4a5ecbd1a6f.jpg",
+        "text": "$$\n\\| G _ { w } ( z ^ { * } ) - x ^ { * } \\| \\leq \\delta ( \\varepsilon ) ,\n$$",
+        "text_format": "latex",
+        "bbox": [
+            413,
+            465,
+            580,
+            483
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "text",
+        "text": "of which upper bound $\\delta ( \\varepsilon )  0$ as $\\varepsilon \\to 0$ . ",
+        "bbox": [
+            176,
+            484,
+            452,
+            500
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "text",
+        "text": "A rigorous statement of Property 1 and its proof are provided in Appendix B, where we prove and use that the cost function is continuous around $x ^ { * }$ under the assumptions. This property justifies solving the latent space search in (5) for $\\mathrm { F L }$ scenarios training neural network model while it requires an accurate generative model. ",
+        "bbox": [
+            174,
+            507,
+            825,
+            564
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "text",
+        "text": "Parameter space search. Using the latent space search only, there can be inevitable reconstruction error due to the imperfection of generative model. This is mainly because we cannot perfectly prepare the generative model for every plausible data in advance. Similar difficulty of the latent space search has been reported even when inverting GAN [33, 3, 4] for plausible but new data directly, i.e., $\\mathrm { m i n } _ { z } \\parallel G _ { w } ( z ) - x ^ { * } \\parallel$ given $x ^ { * }$ , rather than inverting gradient. Bau et al. [3] propose an instance-specific model adaptation, which slightly adjusts the model parameter $w$ to (a part of source image) $x ^ { * }$ after obtaining a latent code $z ^ { * }$ for $x ^ { * }$ . Inspired by such an instance-specific adaptation, GIAS performs the following parameter space search over $w$ preceded by the latent space search over $z$ : ",
+        "bbox": [
+            173,
+            577,
+            825,
+            700
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "equation",
+        "img_path": "images/325187a10dff3508c4c5b1ba8b741dcbada510c6de970a1f441713009762a836.jpg",
+        "text": "$$\n\\operatorname * { m i n } _ { w _ { 1 } , . . . , w _ { B } } \\ c \\left( G _ { w _ { 1 } } ( z _ { 1 } ) , . . . , G _ { w _ { B } } ( z _ { B } ) \\right) \\ ,\n$$",
+        "text_format": "latex",
+        "bbox": [
+            374,
+            699,
+            620,
+            723
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "text",
+        "text": "where $z _ { 1 } , \\dots , z _ { B }$ are obtained from (5). ",
+        "bbox": [
+            176,
+            724,
+            434,
+            738
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "text",
+        "text": "Remark. We propose the optimization over $w$ followed by that over $z$ sequentially This is to maximally utilize the benefit of mitigating the under-determined issue from reducing the searching space on the pretrained model. However, the benefit would be degenerated if $z$ and $w$ are optimized jointly or $w$ is optimized first. We provide an empirical justification on the proposed searching strategy in Section 5.1. ",
+        "bbox": [
+            173,
+            751,
+            825,
+            821
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "text",
+        "text": "We perform each search in GIAS using a standard gradient method to the cost function directly. It is worth noting that those optimizations (5) and (7) with generative model can be tackled in a recursive manner as R-GAP [32] reconstructs each layer from output to input. We provide details and performance of the recursive procedure in Appendix C, where employing generative model improves the inversion accuracy of R-GAP substantially, while R-GAP apparently suffers from an error accumulation issue when $f _ { \\theta }$ is a deep neural network. ",
+        "bbox": [
+            174,
+            827,
+            825,
+            911
+        ],
+        "page_idx": 4
+    },
+    {
+        "type": "image",
+        "img_path": "images/3e98af5f62fd12872d10b40932ee68ec2b13001baf19a78e4df336bdfc8cd337.jpg",
+        "image_caption": [
+            "Figure 3: Comparison of different searching spaces. (a) Each row shows reconstructed images of different optimization domains. The first three rows share the same latent space search of 1, 500 iterations, and ${ \\mathrm { G I } } { - z } / w$ is verified to be the best option to fully exploits the knowledge inside the generative model. (b) Cost function over iterations of different optimization domains. "
+        ],
+        "image_footnote": [],
+        "bbox": [
+            168,
+            116,
+            816,
+            332
+        ],
+        "page_idx": 5
+    },
+    {
+        "type": "text",
+        "text": "4.2 Gradient inversion to meta-learn generative model ",
+        "text_level": 1,
+        "bbox": [
+            174,
+            430,
+            563,
+            444
+        ],
+        "page_idx": 5
+    },
+    {
+        "type": "text",
+        "text": "For the case that pretrained generative model is unavailable, we devise an algorithm to train a generative model $G _ { w }$ for a set $\\mathcal { S } = \\{ ( \\theta _ { i } , g _ { i } ) \\}$ of gradient inversion tasks. Since each inversion task can be considered as a small learning task to adapt generative model per data, we hence call it gradient inversions to meta-learn (GIML). The detailed procedure of GIML is presented in Appendix A. We start with an arbitrary initialization of $w$ , and iteratively update toward $w ^ { \\prime }$ from a variant of GIAS for $N$ tasks sub-sampled from $s$ , which is different than multiple applications of GIAS for each task in two folds: (i) $\\ell _ { 2 }$ -regularization in latent space search; and (ii) an integrated optimization on model parameter. The variant first finds optimal latent codes $\\boldsymbol { z } _ { i } ^ { * } = ( z _ { i 1 } ^ { * } , . . . , z _ { i B } ^ { * } )$ for each task $i$ with respect to the same cost function of GIAS but additional $\\ell _ { 2 }$ -regularization. Note that the latent space search with untrained generative model easily diverges. The $\\ell _ { 2 }$ -regularization is added to prevent the divergence of $z _ { i } ^ { * }$ . Once we obtained $z _ { i } ^ { * }$ ’s, $w ^ { \\prime }$ is computed by few steps of gradient descents for an integrated parameter search to minimize $\\begin{array} { r } { \\sum _ { i } c ( G _ { w ^ { \\prime } } ( z _ { i 1 } ^ { * } ) , . . . , \\bar { G } _ { w ^ { \\prime } } ( z _ { i B } ^ { * } ) ; \\theta _ { i } , g _ { i } ) } \\end{array}$ . This is because in GIML, we want meta information $w$ to help GIAS for each task rather than solving individual tasks, while after performing GIML to train $w$ , we perform GIAS to invert gradient with the trained $w$ . This is analogous to the Reptile in [20]. ",
+        "bbox": [
+            173,
+            458,
+            825,
+            666
+        ],
+        "page_idx": 5
+    },
+    {
+        "type": "text",
+        "text": "5 Experiments ",
+        "text_level": 1,
+        "bbox": [
+            174,
+            693,
+            312,
+            710
+        ],
+        "page_idx": 5
+    },
+    {
+        "type": "text",
+        "text": "Setup. Unless stated otherwise, we consider the image classification task on the validation set of ImageNet [22] dataset scaled down to $6 4 \\times 6 4$ pixels (for computational tractability) and use a randomly initialized ResNet18 [10] for training. For deep generative models in GIAS, we use StyleGAN2 [13] trained on ImageNet. We use a batch size of $B \\ = \\ 4$ as default and use the negative cosine to measure the gradient dissimilarity $d ( \\cdot , \\cdot )$ . We present detailed setup in Appendix H. Our experiment code is available at https://github.com/ml-postech/ gradient-inversion-generative-image-prior. ",
+        "bbox": [
+            173,
+            728,
+            825,
+            827
+        ],
+        "page_idx": 5
+    },
+    {
+        "type": "text",
+        "text": "Algorithms. We evaluate several algorithms for the gradient inversion (GI) task in (3). They differ mainly in which spaces each algorithm searches over: the input $x$ , the latent code $z$ , and/or the model parameter $w$ . Each algorithm is denoted by GI-(·), where the suffix indicates the search space(s). For instances, GI- $z / w$ is identical to the proposed method, GIAS, and GI- $x$ is the one proposed by Geiping et al. [9]. ",
+        "bbox": [
+            174,
+            842,
+            825,
+            911
+        ],
+        "page_idx": 5
+    },
+    {
+        "type": "image",
+        "img_path": "images/05927611c6cf240c845850b9daf59ed7a0462bf6d541e6a943eb9a3ea192a47a.jpg",
+        "image_caption": [
+            "Figure 4: Comparison of state-of-the-art models and ours. Replacing GI- $x$ with GI- $z / w$ (GIAS) regardless of using BN [29] or not [9] provides substantial improvement in the reconstruction accuracy. (a) Average PSNR and best PSNR in a batch throughout the experiments. (b) An ablation study and comparison of reconstruction results with our models and state-of-the-art models. We highlight the proposed models in bold. "
+        ],
+        "image_footnote": [],
+        "bbox": [
+            163,
+            92,
+            815,
+            281
+        ],
+        "page_idx": 6
+    },
+    {
+        "type": "table",
+        "img_path": "images/d2238a9c5fd3fdd85e97704e60c544b53a261d8a1d75ae6af9d39b80635553ea.jpg",
+        "table_caption": [
+            "Table 1: Comparison of our methods with state-of-the-art methods. Adding our method makes performance improvement versus two baseline methods. PSNR, SSIM, and LPIPS[30] are used to evaluate reconstruction results. We highlight the best performances in bold. "
+        ],
+        "table_footnote": [],
+        "table_body": "<table><tr><td>Method</td><td>GI-x [9]</td><td>GI-z (ours)</td><td>GI-w (ours)</td><td>GI-z/w (GIAS,ours)</td><td>GI-x+BN[29]</td><td>GI-z/w+BN (ours)</td></tr><tr><td>PSNR↑</td><td>13.78</td><td>14.27</td><td>14.70</td><td>15.58</td><td>15.52</td><td>16.31</td></tr><tr><td>SSIM↑</td><td>0.2542</td><td>0.3106</td><td>0.3519</td><td>0.3895</td><td>0.3513</td><td>0.4311</td></tr><tr><td>LPIPS↓</td><td>0.4376</td><td>0.3233</td><td>0.5121</td><td>0.3023</td><td>0.3645</td><td>0.2861</td></tr></table>",
+        "bbox": [
+            176,
+            424,
+            821,
+            484
+        ],
+        "page_idx": 6
+    },
+    {
+        "type": "text",
+        "text": "5.1 Justification of GIAS design ",
+        "text_level": 1,
+        "bbox": [
+            174,
+            498,
+            408,
+            513
+        ],
+        "page_idx": 6
+    },
+    {
+        "type": "text",
+        "text": "We first provide an empirical justification of the specific order of searching spaces in GIAS (corresponding to $\\mathbf { G I } { - } z / w ,$ ) to fully utilize a pretrained generative model. To do so, we provide Figure 4b comparing algorithms with different searching spaces: GI- $z / w$ , GI- $z / x$ , $\\mathrm { G I } { - } z$ , and $_ { \\mathrm { G I - } x }$ , of which the first three share the same latent space search over $z$ for the first 1, 500 iterations. As shown in Figure 3(a), the latent space search over $z$ quickly finds plausible image in a much shorter number of iterations than GI- $x$ , while it does not improve after a certain point due to the imperfection of pretrained generative model. Such a limitation of $\\mathrm { G I } { - } z$ is also captured in Figure 3(b), where the cost function of GI- $z$ is not decreasing after a certain number of optimization steps. To further minimize the cost function, one alternative to GI- $z / w$ (GIAS) is ${ \\mathrm { G I } } { - } z / x$ , which can further reduce the loss function whereas the parameter search in GI- $z / w$ seems to provide more natural reconstruction of the image than ${ \\mathrm { G I } } { - } z / { \\bar { x } }$ . The superiority of ${ \\mathrm { G I } } { - z } / w$ over ${ \\mathrm { G I } } { - } z / x$ may come from that the parameter space search exploits an implicit bias from optimizing a good architecture for expressing images, c.f., deep image prior [26]. In Appendix E and Figure 1, we also present the same comparison on FFHQ (human-face images) [12] where diversity is much smaller than that of ImageNet. On such a less diverse dataset, the distribution can be easily learned, and the gain from training a generative model is larger. ",
+        "bbox": [
+            173,
+            523,
+            825,
+            744
+        ],
+        "page_idx": 6
+    },
+    {
+        "type": "text",
+        "text": "5.2 The gain from fully exploiting pretrained generative model ",
+        "text_level": 1,
+        "bbox": [
+            174,
+            761,
+            622,
+            776
+        ],
+        "page_idx": 6
+    },
+    {
+        "type": "text",
+        "text": "Comparison with state-of-the-art models. Our method can be easily added to previous methods [9, 29]. In Table 1 and Figure 4, we compare the state-of-the-art methods both with and without the proposed generative modelling. In Table 1, comparing GI- $x$ to ${ \\mathrm { G I } } { - } z / w$ and GI- $x + { \\tt B N }$ to ${ \\mathrm { G I } } { - } z / w +$ BN, adding the proposed generative modelling provides additional gain in terms of all the measures (PSNR, SSIM, LPIPS) of reconstruction quality. GI- $z / w$ without BN has lower reconstruction error than GI- $x + { \\tt B N }$ , which is the method of [29]. This implies that the gain from the generative model is comparable to that from BN statistics. However, while the generative model only requires a global (and hence coarse) knowledge on the entire dataset, BN statistics are local to the batch in hand and hence requires significantly more detailed information on the exact batch used to compute gradient. ",
+        "bbox": [
+            174,
+            786,
+            825,
+            911
+        ],
+        "page_idx": 6
+    },
+    {
+        "type": "image",
+        "img_path": "images/eaebaa7d2c36a894aebb95c062ddacad082b8c4df28b7ce3593656951181c37f.jpg",
+        "image_caption": [
+            "Figure 5: Comparison of state-of-the-art models and $G I - z / w$ with varying difficulties. Larger batch size, higher sparsity, and larger gradient noise increases reconstruction difficulty. GI- $z / w$ always surpasses GI- $x$ thanks to the pretrained generative model. All subfigures share the y-axis. "
+        ],
+        "image_footnote": [],
+        "bbox": [
+            173,
+            90,
+            789,
+            266
+        ],
+        "page_idx": 7
+    },
+    {
+        "type": "text",
+        "text": "As shown in Figure 4, the superiority of our method compared to the others is clear in terms of the best-in-batch performance than the average one, where the former is more suitable to show actual privacy threat in the worst case than the latter. It is also interesting to note that GI- $w$ with untrained $w$ provides substantial gain compared to GI- $x$ . This may imply that there is a gain of the implicit bias, c.f., [26], from training the architecture of deep generative model. ",
+        "bbox": [
+            174,
+            345,
+            825,
+            415
+        ],
+        "page_idx": 7
+    },
+    {
+        "type": "text",
+        "text": "Evaluation against possible defense methods We evaluate the gain of using a generative model for various FL scenarios with varying levels of difficulty in the inversion. As batch size, gradient sparsity1 [28] and gradient noise level increase, the risk of having under-determined inversion increases and the inversion task becomes more challenging. Figure 5 shows that for all the levels of difficulty, the generative model provides significant gain in reconstruction quality. In particular, the averaged PSNR of GI- $x$ with a batch size of 4 is comparable to that of ${ \\mathrm { G I } } { - z } / w$ with a batch size 32. It is also comparable to that of ${ \\mathrm { G I } } { - z } / w$ with a gradient sparsity of $9 9 \\%$ . To measure the impact of the noisy gradient, we experimented gradient inversion with varying gaussian noise level in aforementioned settings. Figure 5(c) shows that adding enough noise to the gradient can mitigate the privacy leakage. GI- $z / w$ with a noise level of 0.01, which is relatively large, still surpasses GI- $_ x$ without noise. A large noise of 0.1 can diminish the gain of exploiting a pretrained generative model. However, the fact that adding large noise to the gradient slows down training makes it difficult for FL practitioners to choose suitable hyperparameters. The results imply our method is more robust to defense methods against gradient inversion, but can be blocked by a high threshold. Note that our results of gradient sparsity and gradient noise implies the Differential Privacy(DP) is still a valid defense method, when applied with a more conservative threshold. For more discussion about possible defense methods in FL framework, see Appendix F. ",
+        "bbox": [
+            174,
+            433,
+            825,
+            667
+        ],
+        "page_idx": 7
+    },
+    {
+        "type": "text",
+        "text": "5.3 Learning generative model from gradients ",
+        "text_level": 1,
+        "bbox": [
+            174,
+            685,
+            508,
+            700
+        ],
+        "page_idx": 7
+    },
+    {
+        "type": "text",
+        "text": "We demonstrate the possibility of training a generative model only with gradients. For computational tractability, we use DCGAN and images from FFHQ [12] resized to $3 2 \\mathbf { x } 3 2$ . We generate a set of gradients from 4 rounds of gradient reports from 200 nodes, in which node computes gradient for a classification task based on the annotation provided in [6]. From the set of gradients, we perform GIML to train a DCGAN to potentially generate FFHQ data. ",
+        "bbox": [
+            174,
+            712,
+            825,
+            781
+        ],
+        "page_idx": 7
+    },
+    {
+        "type": "text",
+        "text": "Figure 6 shows the evolution of generative model improves the reconstruction quality when performing either $\\mathrm { G I } { - } z$ and ${ \\mathrm { G I } } { - z } / w$ . We can clearly see the necessity of parameter space search. Figure 6(a) shows that the quality of images from the generative model is evolving in the training process of GIML. As the step $t$ of GIML increases, the generative model $G _ { w ^ { ( t ) } } ( z )$ for arbitrary $z$ outputs more plausible image of human face. When using generative model trained on wrong dataset (CIFAR10), GI- $z$ completely fails at recovering data. ",
+        "bbox": [
+            174,
+            787,
+            825,
+            871
+        ],
+        "page_idx": 7
+    },
+    {
+        "type": "image",
+        "img_path": "images/11a9bf0deccf6f26645320b1714e23bf518f3cf71177c576ecd107d8a4e5a1fa.jpg",
+        "image_caption": [
+            "Figure 6: Qualitative and quantitative result of GIML. (a) Results validating generative model trained with GIML. Images on the first row are sampled from different GIML training steps. The same latent code $z$ was used to sample images in same rows. Images on the second row and third row are results of $\\mathrm { G I } { - } z$ and GI- $z / w$ using generative model trained with GIML and pretrained model which is trained with CIFAR10 images. Experiments were done with gradient sparsity 0.95 for comparison in difficult setting. Last column represents the ground truth image and result of GI- $\\boldsymbol { w }$ with untrained model. (b) A comparison of GIAS with meta-learned generative model and GIAS using improper generative model. Proper generative model boosts GIAS performance. "
+        ],
+        "image_footnote": [],
+        "bbox": [
+            178,
+            80,
+            816,
+            263
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "In Figure $6 ( \\mathbf { b } )$ , as GIML iteration step increases, the performance of $\\mathrm { G I } { - } z$ and ${ \\mathrm { G I } } { - z } / w$ with GIML surpass GI- $z$ and ${ \\mathrm { G I } } { - z } / w$ with wrong prior knowledge. GI- $z / w$ using generative model trained on wrong dataset and GI- $\\boldsymbol { \\cdot } \\boldsymbol { w }$ which starts with an untrained generative model show lower averaged PSNR compared to ${ \\mathrm { G I } } { - z } / w$ with GIML. GI- $z / w$ with GIML to train generative model on right data shows the best performance in terms of not only quality (Figure 6) but also convergence speed. We provide a comparison of the convergence speed in Appendix G. ",
+        "bbox": [
+            174,
+            409,
+            825,
+            492
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "6 Conclusion ",
+        "text_level": 1,
+        "bbox": [
+            174,
+            511,
+            299,
+            527
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "We propose GIAS fully exploit the prior information on user data from a pretrained generative model when inverting gradient. We demonstrate significant privacy leakage using GIAS with pretrained generative model in various challenging scenarios, where our method provides substantial gain additionally to any other existing methods [9, 29]. In addition, we propose GIML which can train a generative model using only the gradients seen in the FL classifier training. We experimentally show that GIML can meta-learn a generative model on the user data from only gradients, which improves the quality of each individual recovered image. To our best knowledge, GIML is the first capable of learning explicit prior on a set of gradient inversion tasks. ",
+        "bbox": [
+            174,
+            541,
+            825,
+            652
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "Acknowledgments ",
+        "text_level": 1,
+        "bbox": [
+            176,
+            671,
+            328,
+            688
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "This work was partly supported by Institute of Information & communications Technology Planning & Evaluation (IITP) grant funded by the Korea government (MSIT) (No. 2019-0-01906, Artificial Intelligence Graduate School Program (POSTECH)) and (No. 2021-0-00739, Development of Distributed/Cooperative AI based $^ { 5 \\mathrm { G } + }$ Network Data Analytics Functions and Control Technology). Jinwoo Jeon and Jaechang Kim were supported by the Institute of Information & Communications Technology Planning & Evaluation (IITP) grant funded by Korea(MSIT) (2020-0-01594, PSAI industry-academic joint research and education program). Kangwook Lee was supported by NSF/Intel Partnership on Machine Learning for Wireless Networking Program under Grant No. CNS-2003129 and NSF Award DMS-2023239. Sewoong Oh acknowledges funding from NSF IIS-1929955, NSF CCF 2019844, and Google faculty research award. ",
+        "bbox": [
+            174,
+            702,
+            825,
+            840
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "References ",
+        "text_level": 1,
+        "bbox": [
+            174,
+            859,
+            266,
+            876
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "[1] R. Abdal, Y. Qin, and P. Wonka. Image2stylegan++: How to edit the embedded images? In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), ",
+        "bbox": [
+            183,
+            883,
+            823,
+            911
+        ],
+        "page_idx": 8
+    },
+    {
+        "type": "text",
+        "text": "June 2020.   \n[2] M. Andreux, J. O. du Terrail, C. Beguier, and E. W. Tramel. Siloed federated learning for multi-centric histopathology datasets. In S. Albarqouni, S. Bakas, K. Kamnitsas, M. J. Cardoso, B. Landman, W. Li, F. Milletari, N. Rieke, H. Roth, D. Xu, and Z. Xu, editors, Domain Adaptation and Representation Transfer, and Distributed and Collaborative Learning, pages 129–139, Cham, 2020. Springer International Publishing. ISBN 978-3-030-60548-3.   \n[3] D. Bau, H. Strobelt, W. Peebles, J. Wulff, B. Zhou, J.-Y. Zhu, and A. Torralba. Semantic photo manipulation with a generative image prior. ACM Trans. Graph., 38(4), July 2019. ISSN 0730-0301. doi: 10.1145/3306346.3323023. URL https://doi.org/10.1145/3306346. 3323023.   \n[4] D. Bau, J.-Y. Zhu, J. Wulff, W. Peebles, H. Strobelt, B. Zhou, and A. Torralba. Seeing what a gan cannot generate. In Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), October 2019.   \n[5] T. S. Brisimi, R. Chen, T. Mela, A. Olshevsky, I. C. Paschalidis, and W. Shi. Federated learning of predictive models from federated electronic health records. International Journal of Medical Informatics(IJMI), 2018.   \n[6] DCGM. ffhq-features-dataset, 2019. URL https://github.com/DCGM/ ffhq-features-dataset.   \n[7] L. Fan, K. W. Ng, C. Ju, T. Zhang, C. Liu, C. S. Chan, and Q. Yang. Rethinking privacy preserving deep learning: How to evaluate and thwart privacy attacks. In Q. Yang, L. Fan, and H. Yu, editors, Federated Learning: Privacy and Incentive, pages 32–50. Springer, 2020.   \n[8] C. Finn, P. Abbeel, and S. Levine. Model-Agnostic Meta-Learning for Fast Adaptation of Deep Networks. International Conference on Machine Learning (ICML), 2017.   \n[9] J. Geiping, H. Bauermeister, H. Dröge, and M. Moeller. Inverting gradients–How easy is it to break privacy in federated learning? In Neural Information Processing Systems(NeurIPS), 2020.   \n[10] K. He, X. Zhang, S. Ren, and J. Sun. Deep residual learning for image recognition. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2016.   \n[11] B. Hitaj, G. Ateniese, and F. Perez-Cruz. Deep models under the GAN: Information leakage from collaborative deep learning. In Proceedings of the 2017 ACM SIGSAC Conference on Computer and Communications Security (CCS), 2017.   \n[12] T. Karras, S. Laine, and T. Aila. A style-based generator architecture for generative adversarial networks. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), June 2019.   \n[13] T. Karras, S. Laine, M. Aittala, J. Hellsten, J. Lehtinen, and T. Aila. Analyzing and improving the image quality of StyleGAN. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition(CVPR), 2020.   \n[14] D. P. Kingma and J. Ba. Adam: A method for stochastic optimization. In Y. Bengio and Y. LeCun, editors, 3rd International Conference on Learning Representations, ICLR 2015, San Diego, CA, USA, May 7-9, 2015, Conference Track Proceedings, 2015. URL http: //arxiv.org/abs/1412.6980.   \n[15] X. Li, M. Jiang, X. Zhang, M. Kamp, and Q. Dou. FedBN: Federated learning on non-IID features via local batch normalization. In International Conference on Learning Representations (ICLR), 2021.   \n[16] Y. Lin, S. Han, H. Mao, Y. Wang, and W. J. Dally. Deep Gradient Compression: Reducing the communication bandwidth for distributed training. In The International Conference on Learning Representations (ICLR), 2018.   \n[17] B. McMahan, E. Moore, D. Ramage, S. Hampson, and B. A. y Arcas. Communication-efficient learning of deep networks from decentralized data. In A. Singh and J. Zhu, editors, Proceedings of the 20th International Conference on Artificial Intelligence and Statistics(AISTATS), volume 54 of Proceedings of Machine Learning Research, pages 1273–1282, 2017.   \n[18] H. B. McMahan, D. Ramage, K. Talwar, and L. Zhang. Learning differentially private recurrent language models. In 6th International Conference on Learning Representations (ICLR), 2018.   \n[19] L. Melis, C. Song, E. De Cristofaro, and V. Shmatikov. Exploiting unintended feature leakage in collaborative learning. In IEEE Symp. Security and Privacy $( S P )$ , pages 691–706, 2019. doi: 10.1109/SP.2019.00029.   \n[20] A. Nichol, J. Achiam, and J. Schulman. On first-order meta-learning algorithms. arXiv preprint arXiv:1803.02999, 2018.   \n[21] L. T. Phong, Y. Aono, T. Hayashi, L. Wang, and S. Moriai. Privacy-preserving deep learning via additively homomorphic encryption. IEEE Transactions on Information Forensics and Security, 13(5):1333–1345, 2018.   \n[22] O. Russakovsky, J. Deng, H. Su, J. Krause, S. Satheesh, S. Ma, Z. Huang, A. Karpathy, A. Khosla, M. Bernstein, A. C. Berg, and L. Fei-Fei. ImageNet Large Scale Visual Recognition Challenge. International Journal of Computer Vision (IJCV), 115(3):211–252, 2015. doi: 10.1007/s11263-015-0816-y.   \n[23] A. Segal, A. Marcedone, B. Kreuter, D. Ramage, H. B. McMahan, K. Seth, K. A. Bonawitz, S. Patel, and V. Ivanov. Practical secure aggregation for privacy-preserving machine learning. In CCS, 2017. URL https://eprint.iacr.org/2017/281.pdf.   \n[24] R. Shokri, M. Stronati, C. Song, and V. Shmatikov. Membership inference attacks against machine learning models. In 2017 IEEE Symposium on Security and Privacy (SP), pages 3–18, 2017. doi: 10.1109/SP.2017.41.   \n[25] K. Simonyan and A. Zisserman. Very deep convolutional networks for large-scale image recognition. In ICLR, 2015.   \n[26] D. Ulyanov, A. Vedaldi, and V. Lempitsky. Deep image prior. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), June 2018.   \n[27] Z. Wang, M. Song, Z. Zhang, Y. Song, Q. Wang, and H. Qi. Beyond inferring class representatives: User-level privacy leakage from federated learning. In IEEE INFOCOM 2019 - IEEE Conference on Computer Communications, pages 2512–2520, 2019.   \n[28] W. Wei, L. Liu, M. Loper, K.-H. Chow, M. E. Gursoy, S. Truex, and Y. Wu. A framework for evaluating client privacy leakages in federated learning. In L. Chen, N. Li, K. Liang, and S. Schneider, editors, Computer Security – ESORICS 2020, pages 545–566, Cham, 2020. Springer International Publishing. ISBN 978-3-030-58951-6.   \n[29] H. Yin, A. Mallya, A. Vahdat, J. M. Alvarez, J. Kautz, and P. Molchanov. See through gradients: Image batch recovery via gradinversion. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition Workshops (CVPR), 2021.   \n[30] R. Zhang, P. Isola, A. A. Efros, E. Shechtman, and O. Wang. The unreasonable effectiveness of deep features as a perceptual metric. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2018.   \n[31] B. Zhao, K. R. Mopuri, and H. Bilen. iDLG: Improved deep leakage from gradients. arXiv preprint arXiv:2001.02610, 2020.   \n[32] J. Zhu and M. B. Blaschko. R-GAP: Recursive gradient attack on privacy. In International Conference on Learning Representations (ICLR), 2021.   \n[33] J. Zhu, Y. Shen, D. Zhao, and B. Zhou. In-domain gan inversion for real image editing. In Proceedings of European Conference on Computer Vision (ECCV), 2020.   \n[34] L. Zhu, Z. Liu, and S. Han. Deep leakage from gradients. In Annual Conference on Neural Information Processing Systems (NeurIPS), 2019. ",
+        "bbox": [
+            171,
+            71,
+            828,
+            914
+        ],
+        "page_idx": 9
+    },
+    {
+        "type": "text",
+        "text": "",
+        "bbox": [
+            171,
+            70,
+            828,
+            911
+        ],
+        "page_idx": 10
+    }
+]
\ No newline at end of file
diff --git a/parse/train/x9jS8pX3dkx/x9jS8pX3dkx_middle.json b/parse/train/x9jS8pX3dkx/x9jS8pX3dkx_middle.json
new file mode 100644
index 0000000000000000000000000000000000000000..f8bd9a3a61773f731c8e70626d21c9776cfb0f35
--- /dev/null
+++ b/parse/train/x9jS8pX3dkx/x9jS8pX3dkx_middle.json
@@ -0,0 +1,31445 @@
+{
+    "pdf_info": [
+        {
+            "preproc_blocks": [
+                {
+                    "type": "title",
+                    "bbox": [
+                        125,
+                        97,
+                        486,
+                        117
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                126,
+                                95,
+                                487,
+                                119
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        95,
+                                        487,
+                                        119
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Gradient Inversion with Generative Image Prior",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        }
+                    ],
+                    "index": 0
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        114,
+                        158,
+                        510,
+                        216
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                138,
+                                158,
+                                482,
+                                172
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        138,
+                                        158,
+                                        172,
+                                        172
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Jinwoo",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        173,
+                                        159,
+                                        201,
+                                        171
+                                    ],
+                                    "score": 0.47,
+                                    "content": "\\mathbf { J e o n ^ { 1 * } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        202,
+                                        158,
+                                        248,
+                                        172
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", Jaechang",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        248,
+                                        159,
+                                        276,
+                                        171
+                                    ],
+                                    "score": 0.86,
+                                    "content": "\\mathbf { K i m ^ { 2 * } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        277,
+                                        158,
+                                        328,
+                                        172
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", Kangwook",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        328,
+                                        159,
+                                        350,
+                                        170
+                                    ],
+                                    "score": 0.33,
+                                    "content": "\\mathbf { L e e ^ { 3 } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        350,
+                                        158,
+                                        393,
+                                        172
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", Sewoong",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        393,
+                                        159,
+                                        412,
+                                        171
+                                    ],
+                                    "score": 0.83,
+                                    "content": "\\mathbf { O h ^ { 4 } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        413,
+                                        158,
+                                        456,
+                                        172
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", Jungseul",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        456,
+                                        158,
+                                        482,
+                                        170
+                                    ],
+                                    "score": 0.62,
+                                    "content": "\\mathbf { O k } ^ { 1 , 2 }",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                113,
+                                167,
+                                508,
+                                187
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        113,
+                                        167,
+                                        508,
+                                        187
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1 Department of Computer Science & Engineering, Pohang University of Science and Technology",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                178,
+                                495,
+                                198
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        178,
+                                        495,
+                                        198
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2 Graduate School of Artificial Intelligence, Pohang University of Science and Technology",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                110,
+                                190,
+                                511,
+                                209
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        110,
+                                        190,
+                                        511,
+                                        209
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3 Department of Electrical and Computer Engineering, University of Wisconsin-Madison, Madison",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                133,
+                                201,
+                                487,
+                                220
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        133,
+                                        201,
+                                        487,
+                                        220
+                                    ],
+                                    "score": 1.0,
+                                    "content": "4 Paul G. Allen School of Computer Science & Engineering, University of Washington",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        }
+                    ],
+                    "index": 3
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        283,
+                        244,
+                        328,
+                        258
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                281,
+                                244,
+                                330,
+                                259
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        281,
+                                        244,
+                                        330,
+                                        259
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Abstract",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        }
+                    ],
+                    "index": 6
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        143,
+                        270,
+                        469,
+                        412
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                141,
+                                270,
+                                469,
+                                282
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        270,
+                                        469,
+                                        282
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Federated Learning (FL) is a distributed learning framework, in which the local",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                281,
+                                469,
+                                294
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        281,
+                                        469,
+                                        294
+                                    ],
+                                    "score": 1.0,
+                                    "content": "data never leaves clients’ devices to preserve privacy, and the server trains models",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                292,
+                                470,
+                                304
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        292,
+                                        470,
+                                        304
+                                    ],
+                                    "score": 1.0,
+                                    "content": "on the data via accessing only the gradients of those local data. Without further",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                303,
+                                470,
+                                316
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        303,
+                                        470,
+                                        316
+                                    ],
+                                    "score": 1.0,
+                                    "content": "privacy mechanisms such as differential privacy, this leaves the system vulnerable",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                315,
+                                470,
+                                326
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        315,
+                                        470,
+                                        326
+                                    ],
+                                    "score": 1.0,
+                                    "content": "against an attacker who inverts those gradients to reveal clients’ sensitive data.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                325,
+                                470,
+                                338
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        325,
+                                        470,
+                                        338
+                                    ],
+                                    "score": 1.0,
+                                    "content": "However, a gradient is often insufficient to reconstruct the user data without",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                336,
+                                470,
+                                348
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        336,
+                                        470,
+                                        348
+                                    ],
+                                    "score": 1.0,
+                                    "content": "any prior knowledge. By exploiting a generative model pretrained on the data",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                346,
+                                470,
+                                359
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        346,
+                                        470,
+                                        359
+                                    ],
+                                    "score": 1.0,
+                                    "content": "distribution, we demonstrate that data privacy can be easily breached. Further,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                357,
+                                470,
+                                371
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        357,
+                                        470,
+                                        371
+                                    ],
+                                    "score": 1.0,
+                                    "content": "when such prior knowledge is unavailable, we investigate the possibility of learning",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                369,
+                                469,
+                                380
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        369,
+                                        469,
+                                        380
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the prior from a sequence of gradients seen in the process of FL training. We",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                379,
+                                469,
+                                391
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        379,
+                                        469,
+                                        391
+                                    ],
+                                    "score": 1.0,
+                                    "content": "experimentally show that the prior in a form of generative model is learnable",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                390,
+                                469,
+                                403
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        390,
+                                        469,
+                                        403
+                                    ],
+                                    "score": 1.0,
+                                    "content": "from iterative interactions in FL. Our findings strongly suggest that additional",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                401,
+                                383,
+                                414
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        401,
+                                        383,
+                                        414
+                                    ],
+                                    "score": 1.0,
+                                    "content": "mechanisms are necessary to prevent privacy leakage in FL.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        }
+                    ],
+                    "index": 13
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        433,
+                        190,
+                        447
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                432,
+                                192,
+                                449
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        432,
+                                        192,
+                                        449
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1 Introduction",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        }
+                    ],
+                    "index": 20
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        459,
+                        504,
+                        502
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                459,
+                                506,
+                                470
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        459,
+                                        506,
+                                        470
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Federated learning (FL) is an emerging framework for distributed learning, where central server",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                470,
+                                506,
+                                482
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        470,
+                                        506,
+                                        482
+                                    ],
+                                    "score": 1.0,
+                                    "content": "aggregates model updates, rather than user data, from end users [5, 17]. The main premise of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                480,
+                                506,
+                                494
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        480,
+                                        506,
+                                        494
+                                    ],
+                                    "score": 1.0,
+                                    "content": "federated learning is that this particular way of distributed learning can protect users’ data privacy as",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                492,
+                                390,
+                                504
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        492,
+                                        390,
+                                        504
+                                    ],
+                                    "score": 1.0,
+                                    "content": "there is no explicit data shared by the end users with the central server.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        }
+                    ],
+                    "index": 22.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        507,
+                        505,
+                        551
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                507,
+                                506,
+                                521
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        507,
+                                        506,
+                                        521
+                                    ],
+                                    "score": 1.0,
+                                    "content": "However, a recent line of work [34, 31, 9, 29] demonstrates that one may recover the private user",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                518,
+                                505,
+                                532
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        518,
+                                        505,
+                                        532
+                                    ],
+                                    "score": 1.0,
+                                    "content": "data used for training by observing the gradients. This process of recovering the training data from",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                530,
+                                506,
+                                543
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        530,
+                                        506,
+                                        543
+                                    ],
+                                    "score": 1.0,
+                                    "content": "gradients, so-called gradient inversion, poses a huge threat to the federated learning community, as it",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                541,
+                                320,
+                                553
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        541,
+                                        320,
+                                        553
+                                    ],
+                                    "score": 1.0,
+                                    "content": "may imply the fundamental flaw of its main premise.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        }
+                    ],
+                    "index": 26.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        557,
+                        505,
+                        645
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                556,
+                                506,
+                                570
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        556,
+                                        506,
+                                        570
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Even more worryingly, recent works suggest that such gradient inversion attacks can be made even",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                568,
+                                505,
+                                579
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        568,
+                                        505,
+                                        579
+                                    ],
+                                    "score": 1.0,
+                                    "content": "stronger if certain side-information is available. For instance, Geiping et al. [9] show that if the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                579,
+                                506,
+                                591
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        579,
+                                        506,
+                                        591
+                                    ],
+                                    "score": 1.0,
+                                    "content": "attacker knows a prior that user data consists of natural images, then the gradient inversion attack",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                590,
+                                505,
+                                602
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        590,
+                                        505,
+                                        602
+                                    ],
+                                    "score": 1.0,
+                                    "content": "can leverage such prior, achieving a more accurate recovery of the user data. Another instance is",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                600,
+                                505,
+                                613
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        600,
+                                        505,
+                                        613
+                                    ],
+                                    "score": 1.0,
+                                    "content": "when batch norm statistics are available at the attacker in addition to gradients. This can actually",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                612,
+                                505,
+                                623
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        612,
+                                        505,
+                                        623
+                                    ],
+                                    "score": 1.0,
+                                    "content": "happen if the end users share their local batch norm statistics as in [17]. Yin et al. [29] show that such",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                622,
+                                506,
+                                635
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        622,
+                                        506,
+                                        635
+                                    ],
+                                    "score": 1.0,
+                                    "content": "batch normalization statistics can significantly improve the strength of the gradient inversion attack,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                633,
+                                318,
+                                646
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        633,
+                                        318,
+                                        646
+                                    ],
+                                    "score": 1.0,
+                                    "content": "enabling precise recovery of high-resolution images.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        }
+                    ],
+                    "index": 32.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        650,
+                        505,
+                        694
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                649,
+                                505,
+                                662
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        649,
+                                        505,
+                                        662
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In this paper, we systematically study how one can maximally utilize and even obtain the prior",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                660,
+                                506,
+                                673
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        660,
+                                        506,
+                                        673
+                                    ],
+                                    "score": 1.0,
+                                    "content": "information when inverting gradients. We first consider the case that the attacker has a generative",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                672,
+                                506,
+                                684
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        672,
+                                        506,
+                                        684
+                                    ],
+                                    "score": 1.0,
+                                    "content": "model pretrained on the exact or approximate distribution of the user data as a prior. For this,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                683,
+                                505,
+                                695
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        683,
+                                        505,
+                                        695
+                                    ],
+                                    "score": 1.0,
+                                    "content": "we propose an efficient gradient inversion algorithm that utilizes the generative model prior. In",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        }
+                    ],
+                    "index": 38.5
+                }
+            ],
+            "page_idx": 0,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        106,
+                        731,
+                        385,
+                        742
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                730,
+                                386,
+                                743
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        730,
+                                        386,
+                                        743
+                                    ],
+                                    "score": 1.0,
+                                    "content": "35th Conference on Neural Information Processing Systems (NeurIPS 2021).",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        123,
+                        702,
+                        196,
+                        712
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                122,
+                                700,
+                                198,
+                                714
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        122,
+                                        700,
+                                        198,
+                                        714
+                                    ],
+                                    "score": 1.0,
+                                    "content": "∗ equal contribution",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        115,
+                        77,
+                        493,
+                        83
+                    ],
+                    "lines": []
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "title",
+                    "bbox": [
+                        125,
+                        97,
+                        486,
+                        117
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                126,
+                                95,
+                                487,
+                                119
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        95,
+                                        487,
+                                        119
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Gradient Inversion with Generative Image Prior",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        }
+                    ],
+                    "index": 0
+                },
+                {
+                    "type": "list",
+                    "bbox": [
+                        114,
+                        158,
+                        510,
+                        216
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                138,
+                                158,
+                                482,
+                                172
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        138,
+                                        158,
+                                        172,
+                                        172
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Jinwoo",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        173,
+                                        159,
+                                        201,
+                                        171
+                                    ],
+                                    "score": 0.47,
+                                    "content": "\\mathbf { J e o n ^ { 1 * } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        202,
+                                        158,
+                                        248,
+                                        172
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", Jaechang",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        248,
+                                        159,
+                                        276,
+                                        171
+                                    ],
+                                    "score": 0.86,
+                                    "content": "\\mathbf { K i m ^ { 2 * } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        277,
+                                        158,
+                                        328,
+                                        172
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", Kangwook",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        328,
+                                        159,
+                                        350,
+                                        170
+                                    ],
+                                    "score": 0.33,
+                                    "content": "\\mathbf { L e e ^ { 3 } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        350,
+                                        158,
+                                        393,
+                                        172
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", Sewoong",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        393,
+                                        159,
+                                        412,
+                                        171
+                                    ],
+                                    "score": 0.83,
+                                    "content": "\\mathbf { O h ^ { 4 } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        413,
+                                        158,
+                                        456,
+                                        172
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", Jungseul",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        456,
+                                        158,
+                                        482,
+                                        170
+                                    ],
+                                    "score": 0.62,
+                                    "content": "\\mathbf { O k } ^ { 1 , 2 }",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 1,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                113,
+                                167,
+                                508,
+                                187
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        113,
+                                        167,
+                                        508,
+                                        187
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1 Department of Computer Science & Engineering, Pohang University of Science and Technology",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                178,
+                                495,
+                                198
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        178,
+                                        495,
+                                        198
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2 Graduate School of Artificial Intelligence, Pohang University of Science and Technology",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                110,
+                                190,
+                                511,
+                                209
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        110,
+                                        190,
+                                        511,
+                                        209
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3 Department of Electrical and Computer Engineering, University of Wisconsin-Madison, Madison",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                133,
+                                201,
+                                487,
+                                220
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        133,
+                                        201,
+                                        487,
+                                        220
+                                    ],
+                                    "score": 1.0,
+                                    "content": "4 Paul G. Allen School of Computer Science & Engineering, University of Washington",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5,
+                            "is_list_end_line": true
+                        }
+                    ],
+                    "index": 3,
+                    "bbox_fs": [
+                        110,
+                        158,
+                        511,
+                        220
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        283,
+                        244,
+                        328,
+                        258
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                281,
+                                244,
+                                330,
+                                259
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        281,
+                                        244,
+                                        330,
+                                        259
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Abstract",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        }
+                    ],
+                    "index": 6
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        143,
+                        270,
+                        469,
+                        412
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                141,
+                                270,
+                                469,
+                                282
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        270,
+                                        469,
+                                        282
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Federated Learning (FL) is a distributed learning framework, in which the local",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                281,
+                                469,
+                                294
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        281,
+                                        469,
+                                        294
+                                    ],
+                                    "score": 1.0,
+                                    "content": "data never leaves clients’ devices to preserve privacy, and the server trains models",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                292,
+                                470,
+                                304
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        292,
+                                        470,
+                                        304
+                                    ],
+                                    "score": 1.0,
+                                    "content": "on the data via accessing only the gradients of those local data. Without further",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                303,
+                                470,
+                                316
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        303,
+                                        470,
+                                        316
+                                    ],
+                                    "score": 1.0,
+                                    "content": "privacy mechanisms such as differential privacy, this leaves the system vulnerable",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                315,
+                                470,
+                                326
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        315,
+                                        470,
+                                        326
+                                    ],
+                                    "score": 1.0,
+                                    "content": "against an attacker who inverts those gradients to reveal clients’ sensitive data.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                325,
+                                470,
+                                338
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        325,
+                                        470,
+                                        338
+                                    ],
+                                    "score": 1.0,
+                                    "content": "However, a gradient is often insufficient to reconstruct the user data without",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                336,
+                                470,
+                                348
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        336,
+                                        470,
+                                        348
+                                    ],
+                                    "score": 1.0,
+                                    "content": "any prior knowledge. By exploiting a generative model pretrained on the data",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                346,
+                                470,
+                                359
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        346,
+                                        470,
+                                        359
+                                    ],
+                                    "score": 1.0,
+                                    "content": "distribution, we demonstrate that data privacy can be easily breached. Further,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                357,
+                                470,
+                                371
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        357,
+                                        470,
+                                        371
+                                    ],
+                                    "score": 1.0,
+                                    "content": "when such prior knowledge is unavailable, we investigate the possibility of learning",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                369,
+                                469,
+                                380
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        369,
+                                        469,
+                                        380
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the prior from a sequence of gradients seen in the process of FL training. We",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                379,
+                                469,
+                                391
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        379,
+                                        469,
+                                        391
+                                    ],
+                                    "score": 1.0,
+                                    "content": "experimentally show that the prior in a form of generative model is learnable",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                390,
+                                469,
+                                403
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        390,
+                                        469,
+                                        403
+                                    ],
+                                    "score": 1.0,
+                                    "content": "from iterative interactions in FL. Our findings strongly suggest that additional",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                401,
+                                383,
+                                414
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        401,
+                                        383,
+                                        414
+                                    ],
+                                    "score": 1.0,
+                                    "content": "mechanisms are necessary to prevent privacy leakage in FL.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        }
+                    ],
+                    "index": 13,
+                    "bbox_fs": [
+                        141,
+                        270,
+                        470,
+                        414
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        433,
+                        190,
+                        447
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                432,
+                                192,
+                                449
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        432,
+                                        192,
+                                        449
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1 Introduction",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        }
+                    ],
+                    "index": 20
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        459,
+                        504,
+                        502
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                459,
+                                506,
+                                470
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        459,
+                                        506,
+                                        470
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Federated learning (FL) is an emerging framework for distributed learning, where central server",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                470,
+                                506,
+                                482
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        470,
+                                        506,
+                                        482
+                                    ],
+                                    "score": 1.0,
+                                    "content": "aggregates model updates, rather than user data, from end users [5, 17]. The main premise of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                480,
+                                506,
+                                494
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        480,
+                                        506,
+                                        494
+                                    ],
+                                    "score": 1.0,
+                                    "content": "federated learning is that this particular way of distributed learning can protect users’ data privacy as",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                492,
+                                390,
+                                504
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        492,
+                                        390,
+                                        504
+                                    ],
+                                    "score": 1.0,
+                                    "content": "there is no explicit data shared by the end users with the central server.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        }
+                    ],
+                    "index": 22.5,
+                    "bbox_fs": [
+                        105,
+                        459,
+                        506,
+                        504
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        507,
+                        505,
+                        551
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                507,
+                                506,
+                                521
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        507,
+                                        506,
+                                        521
+                                    ],
+                                    "score": 1.0,
+                                    "content": "However, a recent line of work [34, 31, 9, 29] demonstrates that one may recover the private user",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                518,
+                                505,
+                                532
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        518,
+                                        505,
+                                        532
+                                    ],
+                                    "score": 1.0,
+                                    "content": "data used for training by observing the gradients. This process of recovering the training data from",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                530,
+                                506,
+                                543
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        530,
+                                        506,
+                                        543
+                                    ],
+                                    "score": 1.0,
+                                    "content": "gradients, so-called gradient inversion, poses a huge threat to the federated learning community, as it",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                541,
+                                320,
+                                553
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        541,
+                                        320,
+                                        553
+                                    ],
+                                    "score": 1.0,
+                                    "content": "may imply the fundamental flaw of its main premise.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        }
+                    ],
+                    "index": 26.5,
+                    "bbox_fs": [
+                        105,
+                        507,
+                        506,
+                        553
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        557,
+                        505,
+                        645
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                556,
+                                506,
+                                570
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        556,
+                                        506,
+                                        570
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Even more worryingly, recent works suggest that such gradient inversion attacks can be made even",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                568,
+                                505,
+                                579
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        568,
+                                        505,
+                                        579
+                                    ],
+                                    "score": 1.0,
+                                    "content": "stronger if certain side-information is available. For instance, Geiping et al. [9] show that if the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                579,
+                                506,
+                                591
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        579,
+                                        506,
+                                        591
+                                    ],
+                                    "score": 1.0,
+                                    "content": "attacker knows a prior that user data consists of natural images, then the gradient inversion attack",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                590,
+                                505,
+                                602
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        590,
+                                        505,
+                                        602
+                                    ],
+                                    "score": 1.0,
+                                    "content": "can leverage such prior, achieving a more accurate recovery of the user data. Another instance is",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                600,
+                                505,
+                                613
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        600,
+                                        505,
+                                        613
+                                    ],
+                                    "score": 1.0,
+                                    "content": "when batch norm statistics are available at the attacker in addition to gradients. This can actually",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                612,
+                                505,
+                                623
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        612,
+                                        505,
+                                        623
+                                    ],
+                                    "score": 1.0,
+                                    "content": "happen if the end users share their local batch norm statistics as in [17]. Yin et al. [29] show that such",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                622,
+                                506,
+                                635
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        622,
+                                        506,
+                                        635
+                                    ],
+                                    "score": 1.0,
+                                    "content": "batch normalization statistics can significantly improve the strength of the gradient inversion attack,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                633,
+                                318,
+                                646
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        633,
+                                        318,
+                                        646
+                                    ],
+                                    "score": 1.0,
+                                    "content": "enabling precise recovery of high-resolution images.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        }
+                    ],
+                    "index": 32.5,
+                    "bbox_fs": [
+                        105,
+                        556,
+                        506,
+                        646
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        650,
+                        505,
+                        694
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                649,
+                                505,
+                                662
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        649,
+                                        505,
+                                        662
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In this paper, we systematically study how one can maximally utilize and even obtain the prior",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                660,
+                                506,
+                                673
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        660,
+                                        506,
+                                        673
+                                    ],
+                                    "score": 1.0,
+                                    "content": "information when inverting gradients. We first consider the case that the attacker has a generative",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                672,
+                                506,
+                                684
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        672,
+                                        506,
+                                        684
+                                    ],
+                                    "score": 1.0,
+                                    "content": "model pretrained on the exact or approximate distribution of the user data as a prior. For this,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                683,
+                                505,
+                                695
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        683,
+                                        505,
+                                        695
+                                    ],
+                                    "score": 1.0,
+                                    "content": "we propose an efficient gradient inversion algorithm that utilizes the generative model prior. In",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                341,
+                                506,
+                                354
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        341,
+                                        506,
+                                        354
+                                    ],
+                                    "score": 1.0,
+                                    "content": "particular, the algorithm consists of two steps, in which the first step searches the latent space (of",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                353,
+                                505,
+                                364
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        353,
+                                        505,
+                                        364
+                                    ],
+                                    "score": 1.0,
+                                    "content": "lower dimension) defined by the generative model instead of the ambient input space (of higher",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                362,
+                                506,
+                                375
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        362,
+                                        506,
+                                        375
+                                    ],
+                                    "score": 1.0,
+                                    "content": "dimension), and then the second step adapts the generative model to each input given the gradient.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                374,
+                                506,
+                                387
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        374,
+                                        506,
+                                        387
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Each step provides substantial improvement in the reconstruction. We name the algorithm as gradient",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                385,
+                                505,
+                                398
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        385,
+                                        505,
+                                        398
+                                    ],
+                                    "score": 1.0,
+                                    "content": "inversion in alternative spaces (GIAS). Figure 1 represents reconstruction results with the proposed",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                396,
+                                209,
+                                408
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        396,
+                                        209,
+                                        408
+                                    ],
+                                    "score": 1.0,
+                                    "content": "method and existing one.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 26
+                        }
+                    ],
+                    "index": 38.5,
+                    "bbox_fs": [
+                        105,
+                        649,
+                        506,
+                        695
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "image",
+                    "bbox": [
+                        185,
+                        69,
+                        424,
+                        290
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                185,
+                                69,
+                                424,
+                                290
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        185,
+                                        69,
+                                        424,
+                                        290
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                185,
+                                                69,
+                                                424,
+                                                290
+                                            ],
+                                            "score": 0.976,
+                                            "type": "image",
+                                            "image_path": "b81bb14fb3a48de6867d8a38581f8cf29cb13a06887b9c8990eaed2b11ce75bc.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 8,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        185,
+                                        69,
+                                        424,
+                                        82.0
+                                    ],
+                                    "spans": [],
+                                    "index": 0
+                                },
+                                {
+                                    "bbox": [
+                                        185,
+                                        82.0,
+                                        424,
+                                        95.0
+                                    ],
+                                    "spans": [],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        185,
+                                        95.0,
+                                        424,
+                                        108.0
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                },
+                                {
+                                    "bbox": [
+                                        185,
+                                        108.0,
+                                        424,
+                                        121.0
+                                    ],
+                                    "spans": [],
+                                    "index": 3
+                                },
+                                {
+                                    "bbox": [
+                                        185,
+                                        121.0,
+                                        424,
+                                        134.0
+                                    ],
+                                    "spans": [],
+                                    "index": 4
+                                },
+                                {
+                                    "bbox": [
+                                        185,
+                                        134.0,
+                                        424,
+                                        147.0
+                                    ],
+                                    "spans": [],
+                                    "index": 5
+                                },
+                                {
+                                    "bbox": [
+                                        185,
+                                        147.0,
+                                        424,
+                                        160.0
+                                    ],
+                                    "spans": [],
+                                    "index": 6
+                                },
+                                {
+                                    "bbox": [
+                                        185,
+                                        160.0,
+                                        424,
+                                        173.0
+                                    ],
+                                    "spans": [],
+                                    "index": 7
+                                },
+                                {
+                                    "bbox": [
+                                        185,
+                                        173.0,
+                                        424,
+                                        186.0
+                                    ],
+                                    "spans": [],
+                                    "index": 8
+                                },
+                                {
+                                    "bbox": [
+                                        185,
+                                        186.0,
+                                        424,
+                                        199.0
+                                    ],
+                                    "spans": [],
+                                    "index": 9
+                                },
+                                {
+                                    "bbox": [
+                                        185,
+                                        199.0,
+                                        424,
+                                        212.0
+                                    ],
+                                    "spans": [],
+                                    "index": 10
+                                },
+                                {
+                                    "bbox": [
+                                        185,
+                                        212.0,
+                                        424,
+                                        225.0
+                                    ],
+                                    "spans": [],
+                                    "index": 11
+                                },
+                                {
+                                    "bbox": [
+                                        185,
+                                        225.0,
+                                        424,
+                                        238.0
+                                    ],
+                                    "spans": [],
+                                    "index": 12
+                                },
+                                {
+                                    "bbox": [
+                                        185,
+                                        238.0,
+                                        424,
+                                        251.0
+                                    ],
+                                    "spans": [],
+                                    "index": 13
+                                },
+                                {
+                                    "bbox": [
+                                        185,
+                                        251.0,
+                                        424,
+                                        264.0
+                                    ],
+                                    "spans": [],
+                                    "index": 14
+                                },
+                                {
+                                    "bbox": [
+                                        185,
+                                        264.0,
+                                        424,
+                                        277.0
+                                    ],
+                                    "spans": [],
+                                    "index": 15
+                                },
+                                {
+                                    "bbox": [
+                                        185,
+                                        277.0,
+                                        424,
+                                        290.0
+                                    ],
+                                    "spans": [],
+                                    "index": 16
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                275,
+                                284,
+                                333,
+                                293
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        275,
+                                        283,
+                                        334,
+                                        294
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                275,
+                                                283,
+                                                334,
+                                                294
+                                            ],
+                                            "score": 1.0,
+                                            "content": "existing method [9]",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 17
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                108,
+                                295,
+                                504,
+                                328
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        295,
+                                        505,
+                                        308
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                295,
+                                                505,
+                                                308
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 1: An example showing the superiority of GIAS compared to existing method. Images of the",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 18
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        306,
+                                        505,
+                                        320
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                306,
+                                                505,
+                                                320
+                                            ],
+                                            "score": 1.0,
+                                            "content": "authors are reconstructed from gradients by exploiting a generative model pretrained on human face",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 19
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        316,
+                                        141,
+                                        332
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                316,
+                                                141,
+                                                332
+                                            ],
+                                            "score": 1.0,
+                                            "content": "images.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 20
+                                }
+                            ],
+                            "index": 19
+                        }
+                    ],
+                    "index": 17
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        341,
+                        505,
+                        407
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                341,
+                                506,
+                                354
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        341,
+                                        506,
+                                        354
+                                    ],
+                                    "score": 1.0,
+                                    "content": "particular, the algorithm consists of two steps, in which the first step searches the latent space (of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                353,
+                                505,
+                                364
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        353,
+                                        505,
+                                        364
+                                    ],
+                                    "score": 1.0,
+                                    "content": "lower dimension) defined by the generative model instead of the ambient input space (of higher",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                362,
+                                506,
+                                375
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        362,
+                                        506,
+                                        375
+                                    ],
+                                    "score": 1.0,
+                                    "content": "dimension), and then the second step adapts the generative model to each input given the gradient.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                374,
+                                506,
+                                387
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        374,
+                                        506,
+                                        387
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Each step provides substantial improvement in the reconstruction. We name the algorithm as gradient",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                385,
+                                505,
+                                398
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        385,
+                                        505,
+                                        398
+                                    ],
+                                    "score": 1.0,
+                                    "content": "inversion in alternative spaces (GIAS). Figure 1 represents reconstruction results with the proposed",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                396,
+                                209,
+                                408
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        396,
+                                        209,
+                                        408
+                                    ],
+                                    "score": 1.0,
+                                    "content": "method and existing one.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        }
+                    ],
+                    "index": 23.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        412,
+                        505,
+                        478
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                411,
+                                507,
+                                425
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        411,
+                                        507,
+                                        425
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We then consider a realistic scenario in which the user data distribution is not known in advance,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                422,
+                                505,
+                                437
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        422,
+                                        505,
+                                        437
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and thus the attacker needs to learn it from gradients. For this scenario, we develop a meta-learning",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                434,
+                                506,
+                                446
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        434,
+                                        506,
+                                        446
+                                    ],
+                                    "score": 1.0,
+                                    "content": "framework, called gradient inversion to meta-learn (GIML), which learns a generative model on user",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                445,
+                                506,
+                                458
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        445,
+                                        506,
+                                        458
+                                    ],
+                                    "score": 1.0,
+                                    "content": "data from observing and inverting multiple gradients computed on the data, e.g. across different FL",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                456,
+                                506,
+                                468
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        456,
+                                        506,
+                                        468
+                                    ],
+                                    "score": 1.0,
+                                    "content": "epochs or participating nodes. Our experimental results demonstrate that one can learn a generative",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                466,
+                                451,
+                                479
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        466,
+                                        451,
+                                        479
+                                    ],
+                                    "score": 1.0,
+                                    "content": "model via GIML and reconstruct data by making use of the learned generative model.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        }
+                    ],
+                    "index": 29.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        483,
+                        505,
+                        516
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                482,
+                                505,
+                                496
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        482,
+                                        505,
+                                        496
+                                    ],
+                                    "score": 1.0,
+                                    "content": "This implies a great threat on privacy leakage in FL since our methods can be applied for any data",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                493,
+                                506,
+                                507
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        493,
+                                        506,
+                                        507
+                                    ],
+                                    "score": 1.0,
+                                    "content": "type in most FL scenarios unless a specialized architecture prevents the gradient leakage explicitly,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                505,
+                                148,
+                                517
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        505,
+                                        148,
+                                        517
+                                    ],
+                                    "score": 1.0,
+                                    "content": "e.g., [18].",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        }
+                    ],
+                    "index": 34
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        521,
+                        261,
+                        533
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                520,
+                                262,
+                                534
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        520,
+                                        262,
+                                        534
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Our main contributions are as follows:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        }
+                    ],
+                    "index": 36
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        132,
+                        544,
+                        505,
+                        713
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                133,
+                                543,
+                                505,
+                                557
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        133,
+                                        543,
+                                        505,
+                                        557
+                                    ],
+                                    "score": 1.0,
+                                    "content": "• We introduce GIAS that fully utilizes a pretrained generative model to invert gradient. In",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                555,
+                                499,
+                                568
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        555,
+                                        499,
+                                        568
+                                    ],
+                                    "score": 1.0,
+                                    "content": "addition, we propose GIML which can train generative model from gradients only in FL.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                132,
+                                574,
+                                505,
+                                587
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        132,
+                                        574,
+                                        505,
+                                        587
+                                    ],
+                                    "score": 1.0,
+                                    "content": "• We demonstrate significant privacy leakage occurring by GIAS with a pretrained generative",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                585,
+                                506,
+                                599
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        585,
+                                        506,
+                                        599
+                                    ],
+                                    "score": 1.0,
+                                    "content": "model in various FL scenarios which are challenging to other existing methods, e.g., [9, 29].",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                133,
+                                606,
+                                505,
+                                619
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        133,
+                                        606,
+                                        505,
+                                        619
+                                    ],
+                                    "score": 1.0,
+                                    "content": "• We experimentally show that GIML can learn a generative model on the user data from only",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                617,
+                                505,
+                                630
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        617,
+                                        505,
+                                        630
+                                    ],
+                                    "score": 1.0,
+                                    "content": "gradients, which provides the same level of data recovery with a given pretrained model. To",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                627,
+                                505,
+                                641
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        627,
+                                        505,
+                                        641
+                                    ],
+                                    "score": 1.0,
+                                    "content": "our best knowledge, GIML is the first capable of learning explicit prior on a set of gradient",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                142,
+                                639,
+                                206,
+                                650
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        142,
+                                        639,
+                                        206,
+                                        650
+                                    ],
+                                    "score": 1.0,
+                                    "content": "inversion tasks.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                140,
+                                659,
+                                505,
+                                671
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        140,
+                                        659,
+                                        505,
+                                        671
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We note that a gradient inversion technique defines a standard on defence mechanism in",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                668,
+                                505,
+                                682
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        668,
+                                        505,
+                                        682
+                                    ],
+                                    "score": 1.0,
+                                    "content": "FL for privacy [28]. By substantiating that our proposed methods are able to break down",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                679,
+                                505,
+                                694
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        679,
+                                        505,
+                                        694
+                                    ],
+                                    "score": 1.0,
+                                    "content": "defense mechanisms that were safe according to the previous standard, we give a strong",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 47
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                692,
+                                505,
+                                703
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        692,
+                                        505,
+                                        703
+                                    ],
+                                    "score": 1.0,
+                                    "content": "warning to the FL community to use a higher standard defined by our attack methods, and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 48
+                        },
+                        {
+                            "bbox": [
+                                142,
+                                702,
+                                436,
+                                714
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        142,
+                                        702,
+                                        436,
+                                        714
+                                    ],
+                                    "score": 1.0,
+                                    "content": "raise the necessity of a more conservative choice of defense mechanisms.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 49
+                        }
+                    ],
+                    "index": 43
+                }
+            ],
+            "page_idx": 1,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        302,
+                        742,
+                        308,
+                        750
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                302,
+                                741,
+                                310,
+                                753
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        302,
+                                        741,
+                                        310,
+                                        753
+                                    ],
+                                    "score": 1.0,
+                                    "content": "",
+                                    "type": "text",
+                                    "height": 12,
+                                    "width": 8
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "image",
+                    "bbox": [
+                        185,
+                        69,
+                        424,
+                        290
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                185,
+                                69,
+                                424,
+                                290
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        185,
+                                        69,
+                                        424,
+                                        290
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                185,
+                                                69,
+                                                424,
+                                                290
+                                            ],
+                                            "score": 0.976,
+                                            "type": "image",
+                                            "image_path": "b81bb14fb3a48de6867d8a38581f8cf29cb13a06887b9c8990eaed2b11ce75bc.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 8,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        185,
+                                        69,
+                                        424,
+                                        82.0
+                                    ],
+                                    "spans": [],
+                                    "index": 0
+                                },
+                                {
+                                    "bbox": [
+                                        185,
+                                        82.0,
+                                        424,
+                                        95.0
+                                    ],
+                                    "spans": [],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        185,
+                                        95.0,
+                                        424,
+                                        108.0
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                },
+                                {
+                                    "bbox": [
+                                        185,
+                                        108.0,
+                                        424,
+                                        121.0
+                                    ],
+                                    "spans": [],
+                                    "index": 3
+                                },
+                                {
+                                    "bbox": [
+                                        185,
+                                        121.0,
+                                        424,
+                                        134.0
+                                    ],
+                                    "spans": [],
+                                    "index": 4
+                                },
+                                {
+                                    "bbox": [
+                                        185,
+                                        134.0,
+                                        424,
+                                        147.0
+                                    ],
+                                    "spans": [],
+                                    "index": 5
+                                },
+                                {
+                                    "bbox": [
+                                        185,
+                                        147.0,
+                                        424,
+                                        160.0
+                                    ],
+                                    "spans": [],
+                                    "index": 6
+                                },
+                                {
+                                    "bbox": [
+                                        185,
+                                        160.0,
+                                        424,
+                                        173.0
+                                    ],
+                                    "spans": [],
+                                    "index": 7
+                                },
+                                {
+                                    "bbox": [
+                                        185,
+                                        173.0,
+                                        424,
+                                        186.0
+                                    ],
+                                    "spans": [],
+                                    "index": 8
+                                },
+                                {
+                                    "bbox": [
+                                        185,
+                                        186.0,
+                                        424,
+                                        199.0
+                                    ],
+                                    "spans": [],
+                                    "index": 9
+                                },
+                                {
+                                    "bbox": [
+                                        185,
+                                        199.0,
+                                        424,
+                                        212.0
+                                    ],
+                                    "spans": [],
+                                    "index": 10
+                                },
+                                {
+                                    "bbox": [
+                                        185,
+                                        212.0,
+                                        424,
+                                        225.0
+                                    ],
+                                    "spans": [],
+                                    "index": 11
+                                },
+                                {
+                                    "bbox": [
+                                        185,
+                                        225.0,
+                                        424,
+                                        238.0
+                                    ],
+                                    "spans": [],
+                                    "index": 12
+                                },
+                                {
+                                    "bbox": [
+                                        185,
+                                        238.0,
+                                        424,
+                                        251.0
+                                    ],
+                                    "spans": [],
+                                    "index": 13
+                                },
+                                {
+                                    "bbox": [
+                                        185,
+                                        251.0,
+                                        424,
+                                        264.0
+                                    ],
+                                    "spans": [],
+                                    "index": 14
+                                },
+                                {
+                                    "bbox": [
+                                        185,
+                                        264.0,
+                                        424,
+                                        277.0
+                                    ],
+                                    "spans": [],
+                                    "index": 15
+                                },
+                                {
+                                    "bbox": [
+                                        185,
+                                        277.0,
+                                        424,
+                                        290.0
+                                    ],
+                                    "spans": [],
+                                    "index": 16
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                275,
+                                284,
+                                333,
+                                293
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        275,
+                                        283,
+                                        334,
+                                        294
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                275,
+                                                283,
+                                                334,
+                                                294
+                                            ],
+                                            "score": 1.0,
+                                            "content": "existing method [9]",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 17
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                108,
+                                295,
+                                504,
+                                328
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        295,
+                                        505,
+                                        308
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                295,
+                                                505,
+                                                308
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 1: An example showing the superiority of GIAS compared to existing method. Images of the",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 18
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        306,
+                                        505,
+                                        320
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                306,
+                                                505,
+                                                320
+                                            ],
+                                            "score": 1.0,
+                                            "content": "authors are reconstructed from gradients by exploiting a generative model pretrained on human face",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 19
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        316,
+                                        141,
+                                        332
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                316,
+                                                141,
+                                                332
+                                            ],
+                                            "score": 1.0,
+                                            "content": "images.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 20
+                                }
+                            ],
+                            "index": 19
+                        }
+                    ],
+                    "index": 17
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        341,
+                        505,
+                        407
+                    ],
+                    "lines": [],
+                    "index": 23.5,
+                    "bbox_fs": [
+                        105,
+                        341,
+                        506,
+                        408
+                    ],
+                    "lines_deleted": true
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        412,
+                        505,
+                        478
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                411,
+                                507,
+                                425
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        411,
+                                        507,
+                                        425
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We then consider a realistic scenario in which the user data distribution is not known in advance,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                422,
+                                505,
+                                437
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        422,
+                                        505,
+                                        437
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and thus the attacker needs to learn it from gradients. For this scenario, we develop a meta-learning",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                434,
+                                506,
+                                446
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        434,
+                                        506,
+                                        446
+                                    ],
+                                    "score": 1.0,
+                                    "content": "framework, called gradient inversion to meta-learn (GIML), which learns a generative model on user",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                445,
+                                506,
+                                458
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        445,
+                                        506,
+                                        458
+                                    ],
+                                    "score": 1.0,
+                                    "content": "data from observing and inverting multiple gradients computed on the data, e.g. across different FL",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                456,
+                                506,
+                                468
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        456,
+                                        506,
+                                        468
+                                    ],
+                                    "score": 1.0,
+                                    "content": "epochs or participating nodes. Our experimental results demonstrate that one can learn a generative",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                466,
+                                451,
+                                479
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        466,
+                                        451,
+                                        479
+                                    ],
+                                    "score": 1.0,
+                                    "content": "model via GIML and reconstruct data by making use of the learned generative model.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        }
+                    ],
+                    "index": 29.5,
+                    "bbox_fs": [
+                        105,
+                        411,
+                        507,
+                        479
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        483,
+                        505,
+                        516
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                482,
+                                505,
+                                496
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        482,
+                                        505,
+                                        496
+                                    ],
+                                    "score": 1.0,
+                                    "content": "This implies a great threat on privacy leakage in FL since our methods can be applied for any data",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                493,
+                                506,
+                                507
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        493,
+                                        506,
+                                        507
+                                    ],
+                                    "score": 1.0,
+                                    "content": "type in most FL scenarios unless a specialized architecture prevents the gradient leakage explicitly,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                505,
+                                148,
+                                517
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        505,
+                                        148,
+                                        517
+                                    ],
+                                    "score": 1.0,
+                                    "content": "e.g., [18].",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        }
+                    ],
+                    "index": 34,
+                    "bbox_fs": [
+                        104,
+                        482,
+                        506,
+                        517
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        521,
+                        261,
+                        533
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                520,
+                                262,
+                                534
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        520,
+                                        262,
+                                        534
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Our main contributions are as follows:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        }
+                    ],
+                    "index": 36,
+                    "bbox_fs": [
+                        106,
+                        520,
+                        262,
+                        534
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        132,
+                        544,
+                        505,
+                        713
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                133,
+                                543,
+                                505,
+                                557
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        133,
+                                        543,
+                                        505,
+                                        557
+                                    ],
+                                    "score": 1.0,
+                                    "content": "• We introduce GIAS that fully utilizes a pretrained generative model to invert gradient. In",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                555,
+                                499,
+                                568
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        555,
+                                        499,
+                                        568
+                                    ],
+                                    "score": 1.0,
+                                    "content": "addition, we propose GIML which can train generative model from gradients only in FL.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                132,
+                                574,
+                                505,
+                                587
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        132,
+                                        574,
+                                        505,
+                                        587
+                                    ],
+                                    "score": 1.0,
+                                    "content": "• We demonstrate significant privacy leakage occurring by GIAS with a pretrained generative",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                585,
+                                506,
+                                599
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        585,
+                                        506,
+                                        599
+                                    ],
+                                    "score": 1.0,
+                                    "content": "model in various FL scenarios which are challenging to other existing methods, e.g., [9, 29].",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                133,
+                                606,
+                                505,
+                                619
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        133,
+                                        606,
+                                        505,
+                                        619
+                                    ],
+                                    "score": 1.0,
+                                    "content": "• We experimentally show that GIML can learn a generative model on the user data from only",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                617,
+                                505,
+                                630
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        617,
+                                        505,
+                                        630
+                                    ],
+                                    "score": 1.0,
+                                    "content": "gradients, which provides the same level of data recovery with a given pretrained model. To",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                627,
+                                505,
+                                641
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        627,
+                                        505,
+                                        641
+                                    ],
+                                    "score": 1.0,
+                                    "content": "our best knowledge, GIML is the first capable of learning explicit prior on a set of gradient",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                142,
+                                639,
+                                206,
+                                650
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        142,
+                                        639,
+                                        206,
+                                        650
+                                    ],
+                                    "score": 1.0,
+                                    "content": "inversion tasks.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                140,
+                                659,
+                                505,
+                                671
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        140,
+                                        659,
+                                        505,
+                                        671
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We note that a gradient inversion technique defines a standard on defence mechanism in",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                668,
+                                505,
+                                682
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        668,
+                                        505,
+                                        682
+                                    ],
+                                    "score": 1.0,
+                                    "content": "FL for privacy [28]. By substantiating that our proposed methods are able to break down",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                679,
+                                505,
+                                694
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        679,
+                                        505,
+                                        694
+                                    ],
+                                    "score": 1.0,
+                                    "content": "defense mechanisms that were safe according to the previous standard, we give a strong",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 47
+                        },
+                        {
+                            "bbox": [
+                                141,
+                                692,
+                                505,
+                                703
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        141,
+                                        692,
+                                        505,
+                                        703
+                                    ],
+                                    "score": 1.0,
+                                    "content": "warning to the FL community to use a higher standard defined by our attack methods, and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 48
+                        },
+                        {
+                            "bbox": [
+                                142,
+                                702,
+                                436,
+                                714
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        142,
+                                        702,
+                                        436,
+                                        714
+                                    ],
+                                    "score": 1.0,
+                                    "content": "raise the necessity of a more conservative choice of defense mechanisms.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 49
+                        }
+                    ],
+                    "index": 43,
+                    "bbox_fs": [
+                        132,
+                        543,
+                        506,
+                        714
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        71,
+                        194,
+                        84
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                70,
+                                196,
+                                86
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        70,
+                                        196,
+                                        86
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2 Related work",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        }
+                    ],
+                    "index": 0
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        95,
+                        506,
+                        183
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                94,
+                                506,
+                                108
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        94,
+                                        506,
+                                        108
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Privacy attacks in FL. Early works [19, 24] investigate membership inference from gradients to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                106,
+                                506,
+                                118
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        106,
+                                        506,
+                                        118
+                                    ],
+                                    "score": 1.0,
+                                    "content": "check the possibility of privacy leakage in FL. Phong et al. [21] demonstrate that it is possible to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                116,
+                                506,
+                                130
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        116,
+                                        506,
+                                        130
+                                    ],
+                                    "score": 1.0,
+                                    "content": "reconstruct detailed input image when FL trains a shallow network such as single-layer perceptron.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                127,
+                                506,
+                                141
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        127,
+                                        506,
+                                        141
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Fan et al. [7] and Zhu and Blaschko [32] consider a wider class of learning model and propose an",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                138,
+                                506,
+                                152
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        138,
+                                        506,
+                                        152
+                                    ],
+                                    "score": 1.0,
+                                    "content": "analytical approach solving a sequence of linear systems to reveal the output of each layer recursively.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                149,
+                                506,
+                                162
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        149,
+                                        506,
+                                        162
+                                    ],
+                                    "score": 1.0,
+                                    "content": "To study the limit of the gradient inversion in practical scenarios of training deep networks via",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                160,
+                                506,
+                                174
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        160,
+                                        506,
+                                        174
+                                    ],
+                                    "score": 1.0,
+                                    "content": "FL, a sequence of effort has been made formulating optimization problem to minimize discrepancy",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                172,
+                                418,
+                                183
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        172,
+                                        418,
+                                        183
+                                    ],
+                                    "score": 1.0,
+                                    "content": "comparing gradients from true data and reconstructed data [9, 27, 29, 31, 34].",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        }
+                    ],
+                    "index": 4.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        193,
+                        505,
+                        271
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                194,
+                                506,
+                                207
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        194,
+                                        506,
+                                        207
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Gradient inversion with prior. The optimization-based approaches are particularly useful as one",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                204,
+                                506,
+                                217
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        204,
+                                        506,
+                                        217
+                                    ],
+                                    "score": 1.0,
+                                    "content": "can easily utilize prior knowledge by adding regularization terms, e.g., total variation [27, 9] and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                214,
+                                506,
+                                230
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        214,
+                                        506,
+                                        230
+                                    ],
+                                    "score": 1.0,
+                                    "content": "BN statistics [29], or changing discrepancy measure [9] . In [29], a privacy attack technique using",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                226,
+                                507,
+                                240
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        226,
+                                        507,
+                                        240
+                                    ],
+                                    "score": 1.0,
+                                    "content": "a generative model is introduced. They however require a pretrained model, while we propose a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                237,
+                                506,
+                                250
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        237,
+                                        506,
+                                        250
+                                    ],
+                                    "score": 1.0,
+                                    "content": "meta learning framework training generative model from gradients only. In addition, our method of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                248,
+                                505,
+                                261
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        248,
+                                        505,
+                                        261
+                                    ],
+                                    "score": 1.0,
+                                    "content": "inverting gradient maximally exploit a given generative model by alternating search spaces, which",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                259,
+                                400,
+                                272
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        259,
+                                        400,
+                                        272
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are analogous to the state-of-the-art GAN inversion techniques [3, 4, 33].",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        }
+                    ],
+                    "index": 12
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        281,
+                        505,
+                        413
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                282,
+                                505,
+                                294
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        282,
+                                        505,
+                                        294
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Generative model revealing private data. Training a generative model with transmitted gradients",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                292,
+                                505,
+                                304
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        292,
+                                        505,
+                                        304
+                                    ],
+                                    "score": 1.0,
+                                    "content": "also demonstrates privacy leakage in FL. Hitaj et al. [11] introduce an algorithm to train a GAN",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                303,
+                                506,
+                                316
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        303,
+                                        506,
+                                        316
+                                    ],
+                                    "score": 1.0,
+                                    "content": "regarding shared model in FL framework as a discriminator. Wang et al. [27] use reconstructed",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                315,
+                                506,
+                                327
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        315,
+                                        506,
+                                        327
+                                    ],
+                                    "score": 1.0,
+                                    "content": "data from gradient to train a GAN. Those works require some auxiliary dataset given in advance",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                326,
+                                506,
+                                338
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        326,
+                                        506,
+                                        338
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to enable the training of GAN, while we train a generative model using transmitted gradients only.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                335,
+                                505,
+                                348
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        335,
+                                        505,
+                                        348
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Also, we not only train a generative model but also utilize it for reconstruction, while the generative",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                347,
+                                505,
+                                359
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        347,
+                                        505,
+                                        359
+                                    ],
+                                    "score": 1.0,
+                                    "content": "models in [11, 27] are not used for the reconstruction. Hence, in our approach, the generative model",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                358,
+                                505,
+                                370
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        358,
+                                        505,
+                                        370
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and reconstruction can be improved interactively to each other as shown in Figure 6. In addition,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                368,
+                                506,
+                                382
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        368,
+                                        506,
+                                        382
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[27] is less sample-efficient than ours in a sense that they use gradients to reconstruct images and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                380,
+                                505,
+                                392
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        380,
+                                        505,
+                                        392
+                                    ],
+                                    "score": 1.0,
+                                    "content": "then train a generative model with the reconstructed images, i.e., if the reconstruction fails, then",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                391,
+                                505,
+                                403
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        391,
+                                        505,
+                                        403
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the corresponding update of the generative model fails too, whereas we train the generative model",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                402,
+                                203,
+                                414
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        402,
+                                        203,
+                                        414
+                                    ],
+                                    "score": 1.0,
+                                    "content": "directly from gradients.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        }
+                    ],
+                    "index": 21.5
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        427,
+                        234,
+                        441
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                427,
+                                234,
+                                443
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        427,
+                                        234,
+                                        443
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3 Problem formulation",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        }
+                    ],
+                    "index": 28
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        451,
+                        504,
+                        483
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                451,
+                                505,
+                                464
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        451,
+                                        505,
+                                        464
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In this section, we formally describe the gradient inversion (GI) problem. Consider a standard",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                462,
+                                504,
+                                475
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        462,
+                                        415,
+                                        475
+                                    ],
+                                    "score": 1.0,
+                                    "content": "supervised learning for classification, which optimizes neural network model",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        415,
+                                        463,
+                                        425,
+                                        474
+                                    ],
+                                    "score": 0.86,
+                                    "content": "f _ { \\theta }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        425,
+                                        462,
+                                        497,
+                                        475
+                                    ],
+                                    "score": 1.0,
+                                    "content": "parameterized by",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        498,
+                                        463,
+                                        504,
+                                        472
+                                    ],
+                                    "score": 0.76,
+                                    "content": "\\theta",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                473,
+                                153,
+                                485
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        473,
+                                        153,
+                                        485
+                                    ],
+                                    "score": 1.0,
+                                    "content": "as follows:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        }
+                    ],
+                    "index": 30
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        252,
+                        480,
+                        358,
+                        509
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                252,
+                                480,
+                                358,
+                                509
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        252,
+                                        480,
+                                        358,
+                                        509
+                                    ],
+                                    "score": 0.94,
+                                    "content": "\\operatorname* { m i n } _ { \\theta } \\sum _ { ( x , y ) \\in \\mathcal { D } } \\ell ( f _ { \\theta } ( x ) , y ) \\ ,",
+                                    "type": "interline_equation",
+                                    "image_path": "362bb2549166ec2d31728f5dcb9826c3a3fcf3a97f82757a1fd14ce896277f44.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 32.5,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                252,
+                                480,
+                                358,
+                                494.5
+                            ],
+                            "spans": [],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                252,
+                                494.5,
+                                358,
+                                509.0
+                            ],
+                            "spans": [],
+                            "index": 33
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        509,
+                        505,
+                        580
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                509,
+                                504,
+                                522
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        509,
+                                        133,
+                                        522
+                                    ],
+                                    "score": 1.0,
+                                    "content": "where",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        134,
+                                        510,
+                                        140,
+                                        520
+                                    ],
+                                    "score": 0.78,
+                                    "content": "\\ell",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        140,
+                                        509,
+                                        278,
+                                        522
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is a point-wise loss function and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        279,
+                                        510,
+                                        288,
+                                        520
+                                    ],
+                                    "score": 0.84,
+                                    "content": "\\mathcal { D }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        288,
+                                        509,
+                                        374,
+                                        522
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is a dataset of input",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        374,
+                                        510,
+                                        410,
+                                        520
+                                    ],
+                                    "score": 0.91,
+                                    "content": "x \\in \\mathbb { R } ^ { m }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        410,
+                                        509,
+                                        452,
+                                        522
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and label",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        453,
+                                        509,
+                                        504,
+                                        522
+                                    ],
+                                    "score": 0.92,
+                                    "content": "y \\in \\{ 0 , 1 \\} ^ { L }",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                520,
+                                506,
+                                534
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        520,
+                                        444,
+                                        534
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(one-hot vector). In federated learning framework, each node reports the gradient of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        444,
+                                        522,
+                                        489,
+                                        533
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\ell ( f _ { \\theta } ( x ) , y )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        490,
+                                        520,
+                                        506,
+                                        534
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                531,
+                                505,
+                                544
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        531,
+                                        162,
+                                        544
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sampled data",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        163,
+                                        532,
+                                        186,
+                                        544
+                                    ],
+                                    "score": 0.92,
+                                    "content": "( x , y )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        186,
+                                        531,
+                                        505,
+                                        544
+                                    ],
+                                    "score": 1.0,
+                                    "content": "’s instead of directly transferring the data. The problem of inverting gradient is",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                542,
+                                506,
+                                555
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        542,
+                                        506,
+                                        555
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to reconstruct the sampled data used to compute the reported gradient. Specifically, when a node",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                552,
+                                502,
+                                572
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        552,
+                                        199,
+                                        572
+                                    ],
+                                    "score": 1.0,
+                                    "content": "computes the gradient",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        199,
+                                        557,
+                                        206,
+                                        567
+                                    ],
+                                    "score": 0.79,
+                                    "content": "g",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        207,
+                                        552,
+                                        264,
+                                        572
+                                    ],
+                                    "score": 1.0,
+                                    "content": "using a batch",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        265,
+                                        555,
+                                        361,
+                                        568
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\{ ( x _ { 1 } ^ { * } , y _ { 1 } ^ { * } ) , . . . , ( x _ { B } ^ { * } , y _ { B } ^ { * } ) \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        361,
+                                        552,
+                                        383,
+                                        572
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", i.e.,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        383,
+                                        553,
+                                        502,
+                                        570
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\begin{array} { r } { \\boldsymbol { g } = \\frac { \\top } { B } \\sum _ { j = 1 } ^ { B } \\dot { \\nabla } \\ell ( f _ { \\theta } ( x _ { j } ^ { * } ) , y _ { j } ^ { * } ) } \\end{array}",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                568,
+                                334,
+                                580
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        568,
+                                        334,
+                                        580
+                                    ],
+                                    "score": 1.0,
+                                    "content": "we consider the following problem of inverting gradient:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        }
+                    ],
+                    "index": 36.5
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        205,
+                        582,
+                        405,
+                        624
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                205,
+                                582,
+                                405,
+                                624
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        205,
+                                        582,
+                                        405,
+                                        624
+                                    ],
+                                    "score": 0.94,
+                                    "content": "\\operatorname* { m i n } _ { ( x _ { 1 } , y _ { 1 } ) , \\cdots , ( x _ { B } , y _ { B } ) } d \\left( \\frac { 1 } { B } \\sum _ { j = 1 } ^ { B } \\nabla \\ell ( f _ { \\theta } ( x _ { j } ) , y _ { j } ) , g \\right) \\ ,",
+                                    "type": "interline_equation",
+                                    "image_path": "cfa878f19bead500c62e58fbfed4cec35bc00995d99a229baf711ab52a2756ca.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 41,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                205,
+                                582,
+                                405,
+                                596.0
+                            ],
+                            "spans": [],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                205,
+                                596.0,
+                                405,
+                                610.0
+                            ],
+                            "spans": [],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                205,
+                                610.0,
+                                405,
+                                624.0
+                            ],
+                            "spans": [],
+                            "index": 42
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        626,
+                        505,
+                        685
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                109,
+                                626,
+                                506,
+                                639
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        109,
+                                        626,
+                                        137,
+                                        639
+                                    ],
+                                    "score": 1.0,
+                                    "content": "where",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        138,
+                                        626,
+                                        162,
+                                        639
+                                    ],
+                                    "score": 0.92,
+                                    "content": "d ( \\cdot , \\cdot )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        162,
+                                        626,
+                                        410,
+                                        639
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is a measure of the discrepancy between two gradient, e.g.,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        410,
+                                        627,
+                                        420,
+                                        638
+                                    ],
+                                    "score": 0.86,
+                                    "content": "\\ell _ { 2 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        420,
+                                        626,
+                                        506,
+                                        639
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-distance [34, 29] or",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                641,
+                                505,
+                                653
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        641,
+                                        313,
+                                        653
+                                    ],
+                                    "score": 1.0,
+                                    "content": "negative cosine similarity [9]. It is known that label",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        314,
+                                        642,
+                                        321,
+                                        652
+                                    ],
+                                    "score": 0.8,
+                                    "content": "y",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        321,
+                                        641,
+                                        505,
+                                        653
+                                    ],
+                                    "score": 1.0,
+                                    "content": "can be almost accurately recovered by simple",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                652,
+                                505,
+                                664
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        652,
+                                        462,
+                                        664
+                                    ],
+                                    "score": 1.0,
+                                    "content": "methods just observing the gradient at the last layer [31, 29], while reconstructing input",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        462,
+                                        653,
+                                        469,
+                                        661
+                                    ],
+                                    "score": 0.74,
+                                    "content": "x",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        470,
+                                        652,
+                                        505,
+                                        664
+                                    ],
+                                    "score": 1.0,
+                                    "content": "remains",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                662,
+                                505,
+                                675
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        662,
+                                        505,
+                                        675
+                                    ],
+                                    "score": 1.0,
+                                    "content": "still challenging as it is often under-determined even when the true label is given. For simplicity, we",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                673,
+                                506,
+                                685
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        673,
+                                        506,
+                                        685
+                                    ],
+                                    "score": 1.0,
+                                    "content": "hence focus on the following minimization to reveal the inputs from the gradient given the true labels:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 47
+                        }
+                    ],
+                    "index": 45
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        240,
+                        691,
+                        370,
+                        709
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                240,
+                                691,
+                                370,
+                                709
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        240,
+                                        691,
+                                        370,
+                                        709
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\operatorname* { m i n } _ { \\substack { x _ { 1 } , \\ldots , x _ { B } \\in \\mathbb { R } ^ { m } } } c \\left( x _ { 1 } , . . . , x _ { B } ; \\theta , g \\right) ,",
+                                    "type": "interline_equation",
+                                    "image_path": "1cf3e2d7d9912e7dd5a49750a364cc9f0c98dfe05ef57fff258b6aa478bd5621.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 48,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                240,
+                                691,
+                                370,
+                                709
+                            ],
+                            "spans": [],
+                            "index": 48
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        109,
+                        711,
+                        502,
+                        723
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                709,
+                                500,
+                                725
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        709,
+                                        189,
+                                        725
+                                    ],
+                                    "score": 1.0,
+                                    "content": "where we denote by",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        189,
+                                        711,
+                                        262,
+                                        723
+                                    ],
+                                    "score": 0.9,
+                                    "content": "c \\left( x _ { 1 } , . . . , x _ { B } ; \\theta , g \\right)",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        262,
+                                        709,
+                                        381,
+                                        725
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the cost function in (2) given",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        381,
+                                        711,
+                                        414,
+                                        724
+                                    ],
+                                    "score": 0.93,
+                                    "content": "y _ { j } = y _ { j } ^ { * }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        415,
+                                        709,
+                                        451,
+                                        725
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for each",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        451,
+                                        711,
+                                        500,
+                                        722
+                                    ],
+                                    "score": 0.91,
+                                    "content": "j = 1 , . . . , B",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 49
+                        }
+                    ],
+                    "index": 49
+                }
+            ],
+            "page_idx": 2,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        302,
+                        741,
+                        309,
+                        750
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                301,
+                                740,
+                                310,
+                                752
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        301,
+                                        740,
+                                        310,
+                                        752
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        71,
+                        194,
+                        84
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                70,
+                                196,
+                                86
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        70,
+                                        196,
+                                        86
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2 Related work",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        }
+                    ],
+                    "index": 0
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        95,
+                        506,
+                        183
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                94,
+                                506,
+                                108
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        94,
+                                        506,
+                                        108
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Privacy attacks in FL. Early works [19, 24] investigate membership inference from gradients to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                106,
+                                506,
+                                118
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        106,
+                                        506,
+                                        118
+                                    ],
+                                    "score": 1.0,
+                                    "content": "check the possibility of privacy leakage in FL. Phong et al. [21] demonstrate that it is possible to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                116,
+                                506,
+                                130
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        116,
+                                        506,
+                                        130
+                                    ],
+                                    "score": 1.0,
+                                    "content": "reconstruct detailed input image when FL trains a shallow network such as single-layer perceptron.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                127,
+                                506,
+                                141
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        127,
+                                        506,
+                                        141
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Fan et al. [7] and Zhu and Blaschko [32] consider a wider class of learning model and propose an",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                138,
+                                506,
+                                152
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        138,
+                                        506,
+                                        152
+                                    ],
+                                    "score": 1.0,
+                                    "content": "analytical approach solving a sequence of linear systems to reveal the output of each layer recursively.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                149,
+                                506,
+                                162
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        149,
+                                        506,
+                                        162
+                                    ],
+                                    "score": 1.0,
+                                    "content": "To study the limit of the gradient inversion in practical scenarios of training deep networks via",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                160,
+                                506,
+                                174
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        160,
+                                        506,
+                                        174
+                                    ],
+                                    "score": 1.0,
+                                    "content": "FL, a sequence of effort has been made formulating optimization problem to minimize discrepancy",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                172,
+                                418,
+                                183
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        172,
+                                        418,
+                                        183
+                                    ],
+                                    "score": 1.0,
+                                    "content": "comparing gradients from true data and reconstructed data [9, 27, 29, 31, 34].",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        }
+                    ],
+                    "index": 4.5,
+                    "bbox_fs": [
+                        105,
+                        94,
+                        506,
+                        183
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        193,
+                        505,
+                        271
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                194,
+                                506,
+                                207
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        194,
+                                        506,
+                                        207
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Gradient inversion with prior. The optimization-based approaches are particularly useful as one",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                204,
+                                506,
+                                217
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        204,
+                                        506,
+                                        217
+                                    ],
+                                    "score": 1.0,
+                                    "content": "can easily utilize prior knowledge by adding regularization terms, e.g., total variation [27, 9] and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                214,
+                                506,
+                                230
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        214,
+                                        506,
+                                        230
+                                    ],
+                                    "score": 1.0,
+                                    "content": "BN statistics [29], or changing discrepancy measure [9] . In [29], a privacy attack technique using",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                226,
+                                507,
+                                240
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        226,
+                                        507,
+                                        240
+                                    ],
+                                    "score": 1.0,
+                                    "content": "a generative model is introduced. They however require a pretrained model, while we propose a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                237,
+                                506,
+                                250
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        237,
+                                        506,
+                                        250
+                                    ],
+                                    "score": 1.0,
+                                    "content": "meta learning framework training generative model from gradients only. In addition, our method of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                248,
+                                505,
+                                261
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        248,
+                                        505,
+                                        261
+                                    ],
+                                    "score": 1.0,
+                                    "content": "inverting gradient maximally exploit a given generative model by alternating search spaces, which",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                259,
+                                400,
+                                272
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        259,
+                                        400,
+                                        272
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are analogous to the state-of-the-art GAN inversion techniques [3, 4, 33].",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        }
+                    ],
+                    "index": 12,
+                    "bbox_fs": [
+                        104,
+                        194,
+                        507,
+                        272
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        281,
+                        505,
+                        413
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                282,
+                                505,
+                                294
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        282,
+                                        505,
+                                        294
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Generative model revealing private data. Training a generative model with transmitted gradients",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                292,
+                                505,
+                                304
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        292,
+                                        505,
+                                        304
+                                    ],
+                                    "score": 1.0,
+                                    "content": "also demonstrates privacy leakage in FL. Hitaj et al. [11] introduce an algorithm to train a GAN",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                303,
+                                506,
+                                316
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        303,
+                                        506,
+                                        316
+                                    ],
+                                    "score": 1.0,
+                                    "content": "regarding shared model in FL framework as a discriminator. Wang et al. [27] use reconstructed",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                315,
+                                506,
+                                327
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        315,
+                                        506,
+                                        327
+                                    ],
+                                    "score": 1.0,
+                                    "content": "data from gradient to train a GAN. Those works require some auxiliary dataset given in advance",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                326,
+                                506,
+                                338
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        326,
+                                        506,
+                                        338
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to enable the training of GAN, while we train a generative model using transmitted gradients only.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                335,
+                                505,
+                                348
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        335,
+                                        505,
+                                        348
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Also, we not only train a generative model but also utilize it for reconstruction, while the generative",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                347,
+                                505,
+                                359
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        347,
+                                        505,
+                                        359
+                                    ],
+                                    "score": 1.0,
+                                    "content": "models in [11, 27] are not used for the reconstruction. Hence, in our approach, the generative model",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                358,
+                                505,
+                                370
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        358,
+                                        505,
+                                        370
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and reconstruction can be improved interactively to each other as shown in Figure 6. In addition,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                368,
+                                506,
+                                382
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        368,
+                                        506,
+                                        382
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[27] is less sample-efficient than ours in a sense that they use gradients to reconstruct images and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                380,
+                                505,
+                                392
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        380,
+                                        505,
+                                        392
+                                    ],
+                                    "score": 1.0,
+                                    "content": "then train a generative model with the reconstructed images, i.e., if the reconstruction fails, then",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                391,
+                                505,
+                                403
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        391,
+                                        505,
+                                        403
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the corresponding update of the generative model fails too, whereas we train the generative model",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                402,
+                                203,
+                                414
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        402,
+                                        203,
+                                        414
+                                    ],
+                                    "score": 1.0,
+                                    "content": "directly from gradients.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        }
+                    ],
+                    "index": 21.5,
+                    "bbox_fs": [
+                        105,
+                        282,
+                        506,
+                        414
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        427,
+                        234,
+                        441
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                427,
+                                234,
+                                443
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        427,
+                                        234,
+                                        443
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3 Problem formulation",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        }
+                    ],
+                    "index": 28
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        451,
+                        504,
+                        483
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                451,
+                                505,
+                                464
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        451,
+                                        505,
+                                        464
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In this section, we formally describe the gradient inversion (GI) problem. Consider a standard",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                462,
+                                504,
+                                475
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        462,
+                                        415,
+                                        475
+                                    ],
+                                    "score": 1.0,
+                                    "content": "supervised learning for classification, which optimizes neural network model",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        415,
+                                        463,
+                                        425,
+                                        474
+                                    ],
+                                    "score": 0.86,
+                                    "content": "f _ { \\theta }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        425,
+                                        462,
+                                        497,
+                                        475
+                                    ],
+                                    "score": 1.0,
+                                    "content": "parameterized by",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        498,
+                                        463,
+                                        504,
+                                        472
+                                    ],
+                                    "score": 0.76,
+                                    "content": "\\theta",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                473,
+                                153,
+                                485
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        473,
+                                        153,
+                                        485
+                                    ],
+                                    "score": 1.0,
+                                    "content": "as follows:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        }
+                    ],
+                    "index": 30,
+                    "bbox_fs": [
+                        105,
+                        451,
+                        505,
+                        485
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        252,
+                        480,
+                        358,
+                        509
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                252,
+                                480,
+                                358,
+                                509
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        252,
+                                        480,
+                                        358,
+                                        509
+                                    ],
+                                    "score": 0.94,
+                                    "content": "\\operatorname* { m i n } _ { \\theta } \\sum _ { ( x , y ) \\in \\mathcal { D } } \\ell ( f _ { \\theta } ( x ) , y ) \\ ,",
+                                    "type": "interline_equation",
+                                    "image_path": "362bb2549166ec2d31728f5dcb9826c3a3fcf3a97f82757a1fd14ce896277f44.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 32.5,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                252,
+                                480,
+                                358,
+                                494.5
+                            ],
+                            "spans": [],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                252,
+                                494.5,
+                                358,
+                                509.0
+                            ],
+                            "spans": [],
+                            "index": 33
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        509,
+                        505,
+                        580
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                509,
+                                504,
+                                522
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        509,
+                                        133,
+                                        522
+                                    ],
+                                    "score": 1.0,
+                                    "content": "where",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        134,
+                                        510,
+                                        140,
+                                        520
+                                    ],
+                                    "score": 0.78,
+                                    "content": "\\ell",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        140,
+                                        509,
+                                        278,
+                                        522
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is a point-wise loss function and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        279,
+                                        510,
+                                        288,
+                                        520
+                                    ],
+                                    "score": 0.84,
+                                    "content": "\\mathcal { D }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        288,
+                                        509,
+                                        374,
+                                        522
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is a dataset of input",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        374,
+                                        510,
+                                        410,
+                                        520
+                                    ],
+                                    "score": 0.91,
+                                    "content": "x \\in \\mathbb { R } ^ { m }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        410,
+                                        509,
+                                        452,
+                                        522
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and label",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        453,
+                                        509,
+                                        504,
+                                        522
+                                    ],
+                                    "score": 0.92,
+                                    "content": "y \\in \\{ 0 , 1 \\} ^ { L }",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                520,
+                                506,
+                                534
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        520,
+                                        444,
+                                        534
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(one-hot vector). In federated learning framework, each node reports the gradient of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        444,
+                                        522,
+                                        489,
+                                        533
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\ell ( f _ { \\theta } ( x ) , y )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        490,
+                                        520,
+                                        506,
+                                        534
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                531,
+                                505,
+                                544
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        531,
+                                        162,
+                                        544
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sampled data",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        163,
+                                        532,
+                                        186,
+                                        544
+                                    ],
+                                    "score": 0.92,
+                                    "content": "( x , y )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        186,
+                                        531,
+                                        505,
+                                        544
+                                    ],
+                                    "score": 1.0,
+                                    "content": "’s instead of directly transferring the data. The problem of inverting gradient is",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                542,
+                                506,
+                                555
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        542,
+                                        506,
+                                        555
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to reconstruct the sampled data used to compute the reported gradient. Specifically, when a node",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                552,
+                                502,
+                                572
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        552,
+                                        199,
+                                        572
+                                    ],
+                                    "score": 1.0,
+                                    "content": "computes the gradient",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        199,
+                                        557,
+                                        206,
+                                        567
+                                    ],
+                                    "score": 0.79,
+                                    "content": "g",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        207,
+                                        552,
+                                        264,
+                                        572
+                                    ],
+                                    "score": 1.0,
+                                    "content": "using a batch",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        265,
+                                        555,
+                                        361,
+                                        568
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\{ ( x _ { 1 } ^ { * } , y _ { 1 } ^ { * } ) , . . . , ( x _ { B } ^ { * } , y _ { B } ^ { * } ) \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        361,
+                                        552,
+                                        383,
+                                        572
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", i.e.,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        383,
+                                        553,
+                                        502,
+                                        570
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\begin{array} { r } { \\boldsymbol { g } = \\frac { \\top } { B } \\sum _ { j = 1 } ^ { B } \\dot { \\nabla } \\ell ( f _ { \\theta } ( x _ { j } ^ { * } ) , y _ { j } ^ { * } ) } \\end{array}",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                568,
+                                334,
+                                580
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        568,
+                                        334,
+                                        580
+                                    ],
+                                    "score": 1.0,
+                                    "content": "we consider the following problem of inverting gradient:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        }
+                    ],
+                    "index": 36.5,
+                    "bbox_fs": [
+                        104,
+                        509,
+                        506,
+                        580
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        205,
+                        582,
+                        405,
+                        624
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                205,
+                                582,
+                                405,
+                                624
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        205,
+                                        582,
+                                        405,
+                                        624
+                                    ],
+                                    "score": 0.94,
+                                    "content": "\\operatorname* { m i n } _ { ( x _ { 1 } , y _ { 1 } ) , \\cdots , ( x _ { B } , y _ { B } ) } d \\left( \\frac { 1 } { B } \\sum _ { j = 1 } ^ { B } \\nabla \\ell ( f _ { \\theta } ( x _ { j } ) , y _ { j } ) , g \\right) \\ ,",
+                                    "type": "interline_equation",
+                                    "image_path": "cfa878f19bead500c62e58fbfed4cec35bc00995d99a229baf711ab52a2756ca.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 41,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                205,
+                                582,
+                                405,
+                                596.0
+                            ],
+                            "spans": [],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                205,
+                                596.0,
+                                405,
+                                610.0
+                            ],
+                            "spans": [],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                205,
+                                610.0,
+                                405,
+                                624.0
+                            ],
+                            "spans": [],
+                            "index": 42
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        626,
+                        505,
+                        685
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                109,
+                                626,
+                                506,
+                                639
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        109,
+                                        626,
+                                        137,
+                                        639
+                                    ],
+                                    "score": 1.0,
+                                    "content": "where",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        138,
+                                        626,
+                                        162,
+                                        639
+                                    ],
+                                    "score": 0.92,
+                                    "content": "d ( \\cdot , \\cdot )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        162,
+                                        626,
+                                        410,
+                                        639
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is a measure of the discrepancy between two gradient, e.g.,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        410,
+                                        627,
+                                        420,
+                                        638
+                                    ],
+                                    "score": 0.86,
+                                    "content": "\\ell _ { 2 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        420,
+                                        626,
+                                        506,
+                                        639
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-distance [34, 29] or",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                641,
+                                505,
+                                653
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        641,
+                                        313,
+                                        653
+                                    ],
+                                    "score": 1.0,
+                                    "content": "negative cosine similarity [9]. It is known that label",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        314,
+                                        642,
+                                        321,
+                                        652
+                                    ],
+                                    "score": 0.8,
+                                    "content": "y",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        321,
+                                        641,
+                                        505,
+                                        653
+                                    ],
+                                    "score": 1.0,
+                                    "content": "can be almost accurately recovered by simple",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                652,
+                                505,
+                                664
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        652,
+                                        462,
+                                        664
+                                    ],
+                                    "score": 1.0,
+                                    "content": "methods just observing the gradient at the last layer [31, 29], while reconstructing input",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        462,
+                                        653,
+                                        469,
+                                        661
+                                    ],
+                                    "score": 0.74,
+                                    "content": "x",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        470,
+                                        652,
+                                        505,
+                                        664
+                                    ],
+                                    "score": 1.0,
+                                    "content": "remains",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                662,
+                                505,
+                                675
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        662,
+                                        505,
+                                        675
+                                    ],
+                                    "score": 1.0,
+                                    "content": "still challenging as it is often under-determined even when the true label is given. For simplicity, we",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                673,
+                                506,
+                                685
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        673,
+                                        506,
+                                        685
+                                    ],
+                                    "score": 1.0,
+                                    "content": "hence focus on the following minimization to reveal the inputs from the gradient given the true labels:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 47
+                        }
+                    ],
+                    "index": 45,
+                    "bbox_fs": [
+                        105,
+                        626,
+                        506,
+                        685
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        240,
+                        691,
+                        370,
+                        709
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                240,
+                                691,
+                                370,
+                                709
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        240,
+                                        691,
+                                        370,
+                                        709
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\operatorname* { m i n } _ { \\substack { x _ { 1 } , \\ldots , x _ { B } \\in \\mathbb { R } ^ { m } } } c \\left( x _ { 1 } , . . . , x _ { B } ; \\theta , g \\right) ,",
+                                    "type": "interline_equation",
+                                    "image_path": "1cf3e2d7d9912e7dd5a49750a364cc9f0c98dfe05ef57fff258b6aa478bd5621.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 48,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                240,
+                                691,
+                                370,
+                                709
+                            ],
+                            "spans": [],
+                            "index": 48
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        109,
+                        711,
+                        502,
+                        723
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                709,
+                                500,
+                                725
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        709,
+                                        189,
+                                        725
+                                    ],
+                                    "score": 1.0,
+                                    "content": "where we denote by",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        189,
+                                        711,
+                                        262,
+                                        723
+                                    ],
+                                    "score": 0.9,
+                                    "content": "c \\left( x _ { 1 } , . . . , x _ { B } ; \\theta , g \\right)",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        262,
+                                        709,
+                                        381,
+                                        725
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the cost function in (2) given",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        381,
+                                        711,
+                                        414,
+                                        724
+                                    ],
+                                    "score": 0.93,
+                                    "content": "y _ { j } = y _ { j } ^ { * }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        415,
+                                        709,
+                                        451,
+                                        725
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for each",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        451,
+                                        711,
+                                        500,
+                                        722
+                                    ],
+                                    "score": 0.91,
+                                    "content": "j = 1 , . . . , B",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 49
+                        }
+                    ],
+                    "index": 49,
+                    "bbox_fs": [
+                        106,
+                        709,
+                        500,
+                        725
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        71,
+                        171,
+                        84
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                70,
+                                172,
+                                86
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        70,
+                                        172,
+                                        86
+                                    ],
+                                    "score": 1.0,
+                                    "content": "4 Methods",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        }
+                    ],
+                    "index": 0
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        95,
+                        505,
+                        162
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                96,
+                                505,
+                                107
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        96,
+                                        505,
+                                        107
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The key challenge of inverting gradient is that solving (2) is often under-determined, i.e., a gradient",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                106,
+                                505,
+                                119
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        106,
+                                        505,
+                                        119
+                                    ],
+                                    "score": 1.0,
+                                    "content": "contains only insufficient information to recover data. Such an issue is observed even when the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                118,
+                                505,
+                                129
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        118,
+                                        505,
+                                        129
+                                    ],
+                                    "score": 1.0,
+                                    "content": "dimension of gradient is much larger than that of input data. Indeed, Zhu and Blaschko [32] show",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                127,
+                                505,
+                                141
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        127,
+                                        505,
+                                        141
+                                    ],
+                                    "score": 1.0,
+                                    "content": "that there exist a pair of different data having the same gradient, so called twin data, even when the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                139,
+                                505,
+                                151
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        139,
+                                        505,
+                                        151
+                                    ],
+                                    "score": 1.0,
+                                    "content": "learning model is large. To alleviate this issue, a set of prior knowledge on the nature of data can be",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                150,
+                                154,
+                                161
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        150,
+                                        154,
+                                        161
+                                    ],
+                                    "score": 1.0,
+                                    "content": "considered.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        }
+                    ],
+                    "index": 3.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        166,
+                        505,
+                        200
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                165,
+                                505,
+                                179
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        165,
+                                        461,
+                                        179
+                                    ],
+                                    "score": 1.0,
+                                    "content": "When inverting images, Geiping et al. [9] propose to add the total variation regularization",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        462,
+                                        166,
+                                        493,
+                                        178
+                                    ],
+                                    "score": 0.93,
+                                    "content": "R _ { \\mathrm { T V } } ( x )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        494,
+                                        165,
+                                        505,
+                                        179
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                177,
+                                506,
+                                190
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        177,
+                                        506,
+                                        190
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the cost function in (3) since neighboring pixels of natural images are likely to have similar values.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                187,
+                                170,
+                                201
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        187,
+                                        170,
+                                        201
+                                    ],
+                                    "score": 1.0,
+                                    "content": "More formally,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        }
+                    ],
+                    "index": 8
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        203,
+                        203,
+                        407,
+                        232
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                203,
+                                203,
+                                407,
+                                232
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        203,
+                                        203,
+                                        407,
+                                        232
+                                    ],
+                                    "score": 0.93,
+                                    "content": "R _ { \\mathrm { T V } } ( x ) : = \\sum _ { ( i , j ) } \\sum _ { ( i ^ { \\prime } , j ^ { \\prime } ) \\in \\partial ( i , j ) } \\Vert x ( i , j ) - x ( i ^ { \\prime } , j ^ { \\prime } ) \\Vert ^ { 2 } ,",
+                                    "type": "interline_equation",
+                                    "image_path": "86838b25959935fbcb8653f63ca47434c685281762b0b64e6f57729df7a4c54b.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 10.5,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                203,
+                                203,
+                                407,
+                                217.5
+                            ],
+                            "spans": [],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                203,
+                                217.5,
+                                407,
+                                232.0
+                            ],
+                            "spans": [],
+                            "index": 11
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        236,
+                        483,
+                        249
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                236,
+                                484,
+                                250
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        236,
+                                        133,
+                                        250
+                                    ],
+                                    "score": 1.0,
+                                    "content": "where",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        133,
+                                        237,
+                                        160,
+                                        249
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\partial ( i , j )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        160,
+                                        236,
+                                        262,
+                                        250
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is the set of neighbors of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        263,
+                                        237,
+                                        284,
+                                        249
+                                    ],
+                                    "score": 0.92,
+                                    "content": "( i , j )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        284,
+                                        236,
+                                        484,
+                                        250
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". This method is limited to the natural image data.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        }
+                    ],
+                    "index": 12
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        253,
+                        506,
+                        320
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                254,
+                                505,
+                                265
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        254,
+                                        505,
+                                        265
+                                    ],
+                                    "score": 1.0,
+                                    "content": "For general type of data, one can consider exploiting the batch normalization (BN) statistics from",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                264,
+                                505,
+                                276
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        264,
+                                        505,
+                                        276
+                                    ],
+                                    "score": 1.0,
+                                    "content": "nodes. This is available in the case that the server wants to utilize batch normalization (BN) in",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                274,
+                                506,
+                                288
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        275,
+                                        119,
+                                        285
+                                    ],
+                                    "score": 0.39,
+                                    "content": "\\mathrm { F L }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        120,
+                                        274,
+                                        506,
+                                        288
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", and thus collects the BN statistics (mean and variance) of batch from each node, in addition,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                285,
+                                506,
+                                299
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        285,
+                                        506,
+                                        299
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with every gradient report [17]. To be specific, Yin et al. [29] propose to employ the regularizer",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                107,
+                                297,
+                                506,
+                                310
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        297,
+                                        182,
+                                        309
+                                    ],
+                                    "score": 0.91,
+                                    "content": "R _ { \\mathrm { B N } } ( x _ { 1 } , . . . , x _ { B } ; \\theta )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        182,
+                                        297,
+                                        469,
+                                        310
+                                    ],
+                                    "score": 1.0,
+                                    "content": "which quantifies the discrepancy between the BN statistics of estimated",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        469,
+                                        298,
+                                        480,
+                                        309
+                                    ],
+                                    "score": 0.86,
+                                    "content": "x _ { j }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        480,
+                                        297,
+                                        506,
+                                        310
+                                    ],
+                                    "score": 1.0,
+                                    "content": "’s and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                307,
+                                387,
+                                322
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        307,
+                                        159,
+                                        321
+                                    ],
+                                    "score": 1.0,
+                                    "content": "those of true",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        159,
+                                        309,
+                                        170,
+                                        322
+                                    ],
+                                    "score": 0.87,
+                                    "content": "\\boldsymbol { x } _ { j } ^ { * }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        170,
+                                        307,
+                                        387,
+                                        321
+                                    ],
+                                    "score": 1.0,
+                                    "content": "’s on each layer of the learning model. More formally,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        }
+                    ],
+                    "index": 15.5
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        184,
+                        325,
+                        423,
+                        352
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                184,
+                                325,
+                                423,
+                                352
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        184,
+                                        325,
+                                        423,
+                                        352
+                                    ],
+                                    "score": 0.93,
+                                    "content": "R _ { \\mathrm { B N } } ( x _ { 1 } , . . . , x _ { B } ; \\theta ) : = \\sum _ { l } \\| \\mu _ { l } - \\mu _ { l , \\mathrm { e x a c t } } \\| _ { 2 } + \\| \\sigma _ { l } ^ { 2 } - \\sigma _ { l , \\mathrm { e x a c t } } ^ { 2 } \\| _ { 2 } ,",
+                                    "type": "interline_equation",
+                                    "image_path": "3657e8e4fe2b5bbe619f665afa927d3834c7b319b9b07b5ca1c13ed8a2dea5ff.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 19.5,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                184,
+                                325,
+                                423,
+                                338.5
+                            ],
+                            "spans": [],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                184,
+                                338.5,
+                                423,
+                                352.0
+                            ],
+                            "spans": [],
+                            "index": 20
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        356,
+                        506,
+                        479
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                354,
+                                504,
+                                373
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        354,
+                                        133,
+                                        373
+                                    ],
+                                    "score": 1.0,
+                                    "content": "where",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        133,
+                                        357,
+                                        200,
+                                        369
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\mu _ { l } ( x _ { 1 } , . . . , x _ { B } ; \\theta )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        200,
+                                        354,
+                                        219,
+                                        373
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        219,
+                                        357,
+                                        287,
+                                        370
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\sigma _ { l } ^ { 2 } ( x _ { 1 } , . . . , x _ { B } ; \\theta )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        288,
+                                        354,
+                                        314,
+                                        373
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(resp.",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        315,
+                                        357,
+                                        398,
+                                        369
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\mu _ { l , \\mathrm { { e x a c t } } } ( x _ { 1 } ^ { * } , . . . , x _ { B } ^ { * } ; \\theta )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        398,
+                                        354,
+                                        417,
+                                        373
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        417,
+                                        357,
+                                        504,
+                                        370
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\sigma _ { l , \\mathrm { { e x a c t } } } ^ { 2 } ( x _ { 1 } ^ { * } , . . . , x _ { B } ^ { * } ; \\theta ) )",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                368,
+                                506,
+                                382
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        368,
+                                        222,
+                                        382
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are the mean and variance of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        222,
+                                        370,
+                                        226,
+                                        379
+                                    ],
+                                    "score": 0.73,
+                                    "content": "l",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        226,
+                                        368,
+                                        405,
+                                        382
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-th layer feature maps for the estimated batch",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        406,
+                                        371,
+                                        446,
+                                        380
+                                    ],
+                                    "score": 0.91,
+                                    "content": "x _ { 1 } , . . . , x _ { B }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        447,
+                                        368,
+                                        506,
+                                        382
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(resp. the true",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                380,
+                                506,
+                                394
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        380,
+                                        131,
+                                        394
+                                    ],
+                                    "score": 1.0,
+                                    "content": "batch",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        131,
+                                        380,
+                                        177,
+                                        392
+                                    ],
+                                    "score": 0.88,
+                                    "content": "x _ { 1 } ^ { * } , . . . , x _ { B } ^ { * } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        177,
+                                        380,
+                                        203,
+                                        394
+                                    ],
+                                    "score": 1.0,
+                                    "content": "given",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        203,
+                                        381,
+                                        209,
+                                        390
+                                    ],
+                                    "score": 0.71,
+                                    "content": "\\theta",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        209,
+                                        380,
+                                        506,
+                                        394
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". This is available only if clients agree to report their exact BN statistics at",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                391,
+                                505,
+                                403
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        391,
+                                        505,
+                                        403
+                                    ],
+                                    "score": 1.0,
+                                    "content": "every round. But not every FL framework report BN statistics [15, 2]. In that case, Yin et al. [29] also",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                401,
+                                506,
+                                414
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        401,
+                                        506,
+                                        414
+                                    ],
+                                    "score": 1.0,
+                                    "content": "propose to use the BN statistics over the entire data distribution as a proxy of the true BN statistics,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                412,
+                                506,
+                                425
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        412,
+                                        506,
+                                        425
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and reports that the gain from the approximated BN statistics is comparable to that from the exact",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                424,
+                                506,
+                                437
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        424,
+                                        210,
+                                        437
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ones. The applicability of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        210,
+                                        424,
+                                        229,
+                                        435
+                                    ],
+                                    "score": 0.9,
+                                    "content": "R _ { \\mathrm { B N } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        230,
+                                        424,
+                                        506,
+                                        437
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with the approximated BN statistics is still limited as the proxy needs",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                434,
+                                506,
+                                447
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        434,
+                                        436,
+                                        447
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to be additionally recomputed over the entire data distribution at every change of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        436,
+                                        435,
+                                        442,
+                                        444
+                                    ],
+                                    "score": 0.72,
+                                    "content": "\\theta",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        443,
+                                        434,
+                                        506,
+                                        447
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". However, this",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                445,
+                                506,
+                                458
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        445,
+                                        506,
+                                        458
+                                    ],
+                                    "score": 1.0,
+                                    "content": "demonstrates the significant impact of knowing the data distribution in the gradient inversion and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                456,
+                                506,
+                                469
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        456,
+                                        506,
+                                        469
+                                    ],
+                                    "score": 1.0,
+                                    "content": "motivates our methods using and learning a generative model on the user data, described in what",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                466,
+                                142,
+                                479
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        466,
+                                        142,
+                                        479
+                                    ],
+                                    "score": 1.0,
+                                    "content": "follows.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        }
+                    ],
+                    "index": 26
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        106,
+                        491,
+                        341,
+                        503
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                490,
+                                341,
+                                505
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        490,
+                                        341,
+                                        505
+                                    ],
+                                    "score": 1.0,
+                                    "content": "4.1 Gradient inversion with trained generative model",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        }
+                    ],
+                    "index": 32
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        510,
+                        506,
+                        578
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                509,
+                                506,
+                                525
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        509,
+                                        255,
+                                        525
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Consider a decent generative model",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        256,
+                                        511,
+                                        324,
+                                        522
+                                    ],
+                                    "score": 0.9,
+                                    "content": "G _ { w } : \\mathbb { R } ^ { k } \\mapsto \\mathbb { R } ^ { m }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        324,
+                                        509,
+                                        506,
+                                        525
+                                    ],
+                                    "score": 1.0,
+                                    "content": "trained on the approximate (possibly exact)",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                522,
+                                505,
+                                536
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        522,
+                                        210,
+                                        536
+                                    ],
+                                    "score": 1.0,
+                                    "content": "distribution of user data",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        211,
+                                        523,
+                                        221,
+                                        532
+                                    ],
+                                    "score": 0.8,
+                                    "content": "\\mathcal { D }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        221,
+                                        522,
+                                        264,
+                                        536
+                                    ],
+                                    "score": 1.0,
+                                    "content": "such that",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        264,
+                                        523,
+                                        326,
+                                        534
+                                    ],
+                                    "score": 0.92,
+                                    "content": "x ^ { * } \\approx G _ { w } ( z ^ { * } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        326,
+                                        522,
+                                        344,
+                                        536
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        345,
+                                        522,
+                                        396,
+                                        534
+                                    ],
+                                    "score": 0.92,
+                                    "content": "( x ^ { \\ast } , \\cdot ) \\in \\bar { \\mathcal { D } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        396,
+                                        522,
+                                        479,
+                                        536
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and its latent code",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        480,
+                                        523,
+                                        505,
+                                        533
+                                    ],
+                                    "score": 0.88,
+                                    "content": "z ^ { * } =",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                532,
+                                506,
+                                546
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        533,
+                                        206,
+                                        545
+                                    ],
+                                    "score": 0.84,
+                                    "content": "\\mathrm { a r g m i n } _ { z } \\| G _ { w } ( z ) - x ^ { * } \\|",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        206,
+                                        532,
+                                        506,
+                                        546
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". To fully utilize such a pretrained generative model, we propose gradient",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                544,
+                                505,
+                                556
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        544,
+                                        505,
+                                        556
+                                    ],
+                                    "score": 1.0,
+                                    "content": "inversion in alternative spaces (GIAS), of which pseudocode is presented in Appendix A, which",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                555,
+                                506,
+                                568
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        555,
+                                        244,
+                                        568
+                                    ],
+                                    "score": 1.0,
+                                    "content": "performs latent space search over",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        244,
+                                        557,
+                                        250,
+                                        565
+                                    ],
+                                    "score": 0.8,
+                                    "content": "z",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        251,
+                                        555,
+                                        405,
+                                        568
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and then parameter space search over",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        405,
+                                        557,
+                                        413,
+                                        565
+                                    ],
+                                    "score": 0.75,
+                                    "content": "w",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        414,
+                                        555,
+                                        506,
+                                        568
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". We also illustrate the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                566,
+                                264,
+                                578
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        566,
+                                        264,
+                                        578
+                                    ],
+                                    "score": 1.0,
+                                    "content": "overall procedure of GIAS in Figure 2.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        }
+                    ],
+                    "index": 35.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        588,
+                        506,
+                        655
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                589,
+                                506,
+                                601
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        589,
+                                        506,
+                                        601
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Latent space search. Note that the latent space is typically much smaller than the ambient input",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                600,
+                                504,
+                                612
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        600,
+                                        152,
+                                        612
+                                    ],
+                                    "score": 1.0,
+                                    "content": "space, i.e.,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        152,
+                                        600,
+                                        184,
+                                        610
+                                    ],
+                                    "score": 0.91,
+                                    "content": "k \\ll m",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        184,
+                                        600,
+                                        315,
+                                        612
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", for instances, DCGAN [25] of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        315,
+                                        600,
+                                        351,
+                                        610
+                                    ],
+                                    "score": 0.91,
+                                    "content": "k = 1 0 0",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        351,
+                                        600,
+                                        446,
+                                        612
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and StyleGAN [12] of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        446,
+                                        600,
+                                        504,
+                                        610
+                                    ],
+                                    "score": 0.9,
+                                    "content": "k = 5 1 2 \\times 1 6",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                611,
+                                506,
+                                623
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        611,
+                                        178,
+                                        623
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for image data of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        178,
+                                        611,
+                                        317,
+                                        623
+                                    ],
+                                    "score": 0.86,
+                                    "content": "m = ( \\mathrm { w i d t h } ) \\times ( \\mathrm { h e i g h t } ) \\times ( \\mathrm { c o l o r } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        318,
+                                        611,
+                                        349,
+                                        623
+                                    ],
+                                    "score": 1.0,
+                                    "content": "such as",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        350,
+                                        611,
+                                        400,
+                                        622
+                                    ],
+                                    "score": 0.8,
+                                    "content": "3 2 \\times 3 2 \\times 3",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        401,
+                                        611,
+                                        403,
+                                        623
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        404,
+                                        611,
+                                        464,
+                                        622
+                                    ],
+                                    "score": 0.85,
+                                    "content": "2 5 6 \\times 2 5 6 \\times 3",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        465,
+                                        611,
+                                        506,
+                                        623
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", or larger.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                622,
+                                506,
+                                635
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        622,
+                                        298,
+                                        635
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Using such a pretrained generative model with",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        299,
+                                        623,
+                                        330,
+                                        632
+                                    ],
+                                    "score": 0.89,
+                                    "content": "k \\ll m",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        330,
+                                        622,
+                                        506,
+                                        635
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", the under-determined issues of (3) can be",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                630,
+                                508,
+                                647
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        630,
+                                        366,
+                                        647
+                                    ],
+                                    "score": 1.0,
+                                    "content": "directly mitigated by narrowing down the searching space from",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        366,
+                                        633,
+                                        382,
+                                        642
+                                    ],
+                                    "score": 0.88,
+                                    "content": "\\mathbb { R } ^ { m }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        383,
+                                        630,
+                                        394,
+                                        647
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        395,
+                                        632,
+                                        471,
+                                        645
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\{ G _ { w } ( z ) : z \\in \\mathbb { R } ^ { k } \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        471,
+                                        630,
+                                        508,
+                                        647
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Hence,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                642,
+                                353,
+                                656
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        642,
+                                        353,
+                                        656
+                                    ],
+                                    "score": 1.0,
+                                    "content": "GIAS first performs the latent space search in the followings:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        }
+                    ],
+                    "index": 41.5
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        228,
+                        659,
+                        382,
+                        679
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                228,
+                                659,
+                                382,
+                                679
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        228,
+                                        659,
+                                        382,
+                                        679
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\operatorname* { m i n } _ { \\substack { z _ { 1 } , \\ldots , z _ { B } \\in \\mathbb { R } ^ { k } } } c \\left( G _ { w } ( z _ { 1 } ) , . . . , G _ { w } ( z _ { B } ) \\right) .",
+                                    "type": "interline_equation",
+                                    "image_path": "f211e132f6ecc08b1bb03ea07a719bb36e5634f96711de6ef5fb6e6d12333b9b.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 45,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                228,
+                                659,
+                                382,
+                                679
+                            ],
+                            "spans": [],
+                            "index": 45
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        688,
+                        504,
+                        722
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                688,
+                                504,
+                                700
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        688,
+                                        324,
+                                        700
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Considering a canonical class of neural network model",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        325,
+                                        689,
+                                        335,
+                                        700
+                                    ],
+                                    "score": 0.88,
+                                    "content": "f _ { \\theta }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        335,
+                                        688,
+                                        492,
+                                        700
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", we can show that the reconstruction of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        492,
+                                        689,
+                                        504,
+                                        699
+                                    ],
+                                    "score": 0.87,
+                                    "content": "x ^ { * }",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 46
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                698,
+                                504,
+                                714
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        698,
+                                        488,
+                                        714
+                                    ],
+                                    "score": 1.0,
+                                    "content": "by latent space search in (5) aligns with that by input space search in (3) if the generative model",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        488,
+                                        700,
+                                        504,
+                                        711
+                                    ],
+                                    "score": 0.88,
+                                    "content": "G _ { w }",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 47
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                711,
+                                305,
+                                724
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        711,
+                                        305,
+                                        724
+                                    ],
+                                    "score": 1.0,
+                                    "content": "approximates input data with small enough error.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 48
+                        }
+                    ],
+                    "index": 47
+                }
+            ],
+            "page_idx": 3,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        302,
+                        741,
+                        308,
+                        750
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                302,
+                                741,
+                                310,
+                                752
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        302,
+                                        741,
+                                        310,
+                                        752
+                                    ],
+                                    "score": 1.0,
+                                    "content": "4",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        71,
+                        171,
+                        84
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                70,
+                                172,
+                                86
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        70,
+                                        172,
+                                        86
+                                    ],
+                                    "score": 1.0,
+                                    "content": "4 Methods",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        }
+                    ],
+                    "index": 0
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        95,
+                        505,
+                        162
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                96,
+                                505,
+                                107
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        96,
+                                        505,
+                                        107
+                                    ],
+                                    "score": 1.0,
+                                    "content": "The key challenge of inverting gradient is that solving (2) is often under-determined, i.e., a gradient",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                106,
+                                505,
+                                119
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        106,
+                                        505,
+                                        119
+                                    ],
+                                    "score": 1.0,
+                                    "content": "contains only insufficient information to recover data. Such an issue is observed even when the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                118,
+                                505,
+                                129
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        118,
+                                        505,
+                                        129
+                                    ],
+                                    "score": 1.0,
+                                    "content": "dimension of gradient is much larger than that of input data. Indeed, Zhu and Blaschko [32] show",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                127,
+                                505,
+                                141
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        127,
+                                        505,
+                                        141
+                                    ],
+                                    "score": 1.0,
+                                    "content": "that there exist a pair of different data having the same gradient, so called twin data, even when the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                139,
+                                505,
+                                151
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        139,
+                                        505,
+                                        151
+                                    ],
+                                    "score": 1.0,
+                                    "content": "learning model is large. To alleviate this issue, a set of prior knowledge on the nature of data can be",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                150,
+                                154,
+                                161
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        150,
+                                        154,
+                                        161
+                                    ],
+                                    "score": 1.0,
+                                    "content": "considered.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        }
+                    ],
+                    "index": 3.5,
+                    "bbox_fs": [
+                        105,
+                        96,
+                        505,
+                        161
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        166,
+                        505,
+                        200
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                165,
+                                505,
+                                179
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        165,
+                                        461,
+                                        179
+                                    ],
+                                    "score": 1.0,
+                                    "content": "When inverting images, Geiping et al. [9] propose to add the total variation regularization",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        462,
+                                        166,
+                                        493,
+                                        178
+                                    ],
+                                    "score": 0.93,
+                                    "content": "R _ { \\mathrm { T V } } ( x )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        494,
+                                        165,
+                                        505,
+                                        179
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                177,
+                                506,
+                                190
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        177,
+                                        506,
+                                        190
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the cost function in (3) since neighboring pixels of natural images are likely to have similar values.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                187,
+                                170,
+                                201
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        187,
+                                        170,
+                                        201
+                                    ],
+                                    "score": 1.0,
+                                    "content": "More formally,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        }
+                    ],
+                    "index": 8,
+                    "bbox_fs": [
+                        105,
+                        165,
+                        506,
+                        201
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        203,
+                        203,
+                        407,
+                        232
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                203,
+                                203,
+                                407,
+                                232
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        203,
+                                        203,
+                                        407,
+                                        232
+                                    ],
+                                    "score": 0.93,
+                                    "content": "R _ { \\mathrm { T V } } ( x ) : = \\sum _ { ( i , j ) } \\sum _ { ( i ^ { \\prime } , j ^ { \\prime } ) \\in \\partial ( i , j ) } \\Vert x ( i , j ) - x ( i ^ { \\prime } , j ^ { \\prime } ) \\Vert ^ { 2 } ,",
+                                    "type": "interline_equation",
+                                    "image_path": "86838b25959935fbcb8653f63ca47434c685281762b0b64e6f57729df7a4c54b.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 10.5,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                203,
+                                203,
+                                407,
+                                217.5
+                            ],
+                            "spans": [],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                203,
+                                217.5,
+                                407,
+                                232.0
+                            ],
+                            "spans": [],
+                            "index": 11
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        236,
+                        483,
+                        249
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                236,
+                                484,
+                                250
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        236,
+                                        133,
+                                        250
+                                    ],
+                                    "score": 1.0,
+                                    "content": "where",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        133,
+                                        237,
+                                        160,
+                                        249
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\partial ( i , j )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        160,
+                                        236,
+                                        262,
+                                        250
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is the set of neighbors of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        263,
+                                        237,
+                                        284,
+                                        249
+                                    ],
+                                    "score": 0.92,
+                                    "content": "( i , j )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        284,
+                                        236,
+                                        484,
+                                        250
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". This method is limited to the natural image data.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        }
+                    ],
+                    "index": 12,
+                    "bbox_fs": [
+                        106,
+                        236,
+                        484,
+                        250
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        253,
+                        506,
+                        320
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                254,
+                                505,
+                                265
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        254,
+                                        505,
+                                        265
+                                    ],
+                                    "score": 1.0,
+                                    "content": "For general type of data, one can consider exploiting the batch normalization (BN) statistics from",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                264,
+                                505,
+                                276
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        264,
+                                        505,
+                                        276
+                                    ],
+                                    "score": 1.0,
+                                    "content": "nodes. This is available in the case that the server wants to utilize batch normalization (BN) in",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                274,
+                                506,
+                                288
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        275,
+                                        119,
+                                        285
+                                    ],
+                                    "score": 0.39,
+                                    "content": "\\mathrm { F L }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        120,
+                                        274,
+                                        506,
+                                        288
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", and thus collects the BN statistics (mean and variance) of batch from each node, in addition,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                285,
+                                506,
+                                299
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        285,
+                                        506,
+                                        299
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with every gradient report [17]. To be specific, Yin et al. [29] propose to employ the regularizer",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                107,
+                                297,
+                                506,
+                                310
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        107,
+                                        297,
+                                        182,
+                                        309
+                                    ],
+                                    "score": 0.91,
+                                    "content": "R _ { \\mathrm { B N } } ( x _ { 1 } , . . . , x _ { B } ; \\theta )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        182,
+                                        297,
+                                        469,
+                                        310
+                                    ],
+                                    "score": 1.0,
+                                    "content": "which quantifies the discrepancy between the BN statistics of estimated",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        469,
+                                        298,
+                                        480,
+                                        309
+                                    ],
+                                    "score": 0.86,
+                                    "content": "x _ { j }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        480,
+                                        297,
+                                        506,
+                                        310
+                                    ],
+                                    "score": 1.0,
+                                    "content": "’s and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                307,
+                                387,
+                                322
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        307,
+                                        159,
+                                        321
+                                    ],
+                                    "score": 1.0,
+                                    "content": "those of true",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        159,
+                                        309,
+                                        170,
+                                        322
+                                    ],
+                                    "score": 0.87,
+                                    "content": "\\boldsymbol { x } _ { j } ^ { * }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        170,
+                                        307,
+                                        387,
+                                        321
+                                    ],
+                                    "score": 1.0,
+                                    "content": "’s on each layer of the learning model. More formally,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        }
+                    ],
+                    "index": 15.5,
+                    "bbox_fs": [
+                        105,
+                        254,
+                        506,
+                        322
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        184,
+                        325,
+                        423,
+                        352
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                184,
+                                325,
+                                423,
+                                352
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        184,
+                                        325,
+                                        423,
+                                        352
+                                    ],
+                                    "score": 0.93,
+                                    "content": "R _ { \\mathrm { B N } } ( x _ { 1 } , . . . , x _ { B } ; \\theta ) : = \\sum _ { l } \\| \\mu _ { l } - \\mu _ { l , \\mathrm { e x a c t } } \\| _ { 2 } + \\| \\sigma _ { l } ^ { 2 } - \\sigma _ { l , \\mathrm { e x a c t } } ^ { 2 } \\| _ { 2 } ,",
+                                    "type": "interline_equation",
+                                    "image_path": "3657e8e4fe2b5bbe619f665afa927d3834c7b319b9b07b5ca1c13ed8a2dea5ff.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 19.5,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                184,
+                                325,
+                                423,
+                                338.5
+                            ],
+                            "spans": [],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                184,
+                                338.5,
+                                423,
+                                352.0
+                            ],
+                            "spans": [],
+                            "index": 20
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        356,
+                        506,
+                        479
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                354,
+                                504,
+                                373
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        354,
+                                        133,
+                                        373
+                                    ],
+                                    "score": 1.0,
+                                    "content": "where",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        133,
+                                        357,
+                                        200,
+                                        369
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\mu _ { l } ( x _ { 1 } , . . . , x _ { B } ; \\theta )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        200,
+                                        354,
+                                        219,
+                                        373
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        219,
+                                        357,
+                                        287,
+                                        370
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\sigma _ { l } ^ { 2 } ( x _ { 1 } , . . . , x _ { B } ; \\theta )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        288,
+                                        354,
+                                        314,
+                                        373
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(resp.",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        315,
+                                        357,
+                                        398,
+                                        369
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\mu _ { l , \\mathrm { { e x a c t } } } ( x _ { 1 } ^ { * } , . . . , x _ { B } ^ { * } ; \\theta )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        398,
+                                        354,
+                                        417,
+                                        373
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        417,
+                                        357,
+                                        504,
+                                        370
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\sigma _ { l , \\mathrm { { e x a c t } } } ^ { 2 } ( x _ { 1 } ^ { * } , . . . , x _ { B } ^ { * } ; \\theta ) )",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                368,
+                                506,
+                                382
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        368,
+                                        222,
+                                        382
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are the mean and variance of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        222,
+                                        370,
+                                        226,
+                                        379
+                                    ],
+                                    "score": 0.73,
+                                    "content": "l",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        226,
+                                        368,
+                                        405,
+                                        382
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-th layer feature maps for the estimated batch",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        406,
+                                        371,
+                                        446,
+                                        380
+                                    ],
+                                    "score": 0.91,
+                                    "content": "x _ { 1 } , . . . , x _ { B }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        447,
+                                        368,
+                                        506,
+                                        382
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(resp. the true",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                380,
+                                506,
+                                394
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        380,
+                                        131,
+                                        394
+                                    ],
+                                    "score": 1.0,
+                                    "content": "batch",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        131,
+                                        380,
+                                        177,
+                                        392
+                                    ],
+                                    "score": 0.88,
+                                    "content": "x _ { 1 } ^ { * } , . . . , x _ { B } ^ { * } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        177,
+                                        380,
+                                        203,
+                                        394
+                                    ],
+                                    "score": 1.0,
+                                    "content": "given",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        203,
+                                        381,
+                                        209,
+                                        390
+                                    ],
+                                    "score": 0.71,
+                                    "content": "\\theta",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        209,
+                                        380,
+                                        506,
+                                        394
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". This is available only if clients agree to report their exact BN statistics at",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                391,
+                                505,
+                                403
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        391,
+                                        505,
+                                        403
+                                    ],
+                                    "score": 1.0,
+                                    "content": "every round. But not every FL framework report BN statistics [15, 2]. In that case, Yin et al. [29] also",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                401,
+                                506,
+                                414
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        401,
+                                        506,
+                                        414
+                                    ],
+                                    "score": 1.0,
+                                    "content": "propose to use the BN statistics over the entire data distribution as a proxy of the true BN statistics,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                412,
+                                506,
+                                425
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        412,
+                                        506,
+                                        425
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and reports that the gain from the approximated BN statistics is comparable to that from the exact",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                424,
+                                506,
+                                437
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        424,
+                                        210,
+                                        437
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ones. The applicability of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        210,
+                                        424,
+                                        229,
+                                        435
+                                    ],
+                                    "score": 0.9,
+                                    "content": "R _ { \\mathrm { B N } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        230,
+                                        424,
+                                        506,
+                                        437
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with the approximated BN statistics is still limited as the proxy needs",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                434,
+                                506,
+                                447
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        434,
+                                        436,
+                                        447
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to be additionally recomputed over the entire data distribution at every change of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        436,
+                                        435,
+                                        442,
+                                        444
+                                    ],
+                                    "score": 0.72,
+                                    "content": "\\theta",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        443,
+                                        434,
+                                        506,
+                                        447
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". However, this",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                445,
+                                506,
+                                458
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        445,
+                                        506,
+                                        458
+                                    ],
+                                    "score": 1.0,
+                                    "content": "demonstrates the significant impact of knowing the data distribution in the gradient inversion and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                456,
+                                506,
+                                469
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        456,
+                                        506,
+                                        469
+                                    ],
+                                    "score": 1.0,
+                                    "content": "motivates our methods using and learning a generative model on the user data, described in what",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                466,
+                                142,
+                                479
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        466,
+                                        142,
+                                        479
+                                    ],
+                                    "score": 1.0,
+                                    "content": "follows.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        }
+                    ],
+                    "index": 26,
+                    "bbox_fs": [
+                        104,
+                        354,
+                        506,
+                        479
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        106,
+                        491,
+                        341,
+                        503
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                490,
+                                341,
+                                505
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        490,
+                                        341,
+                                        505
+                                    ],
+                                    "score": 1.0,
+                                    "content": "4.1 Gradient inversion with trained generative model",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        }
+                    ],
+                    "index": 32
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        510,
+                        506,
+                        578
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                509,
+                                506,
+                                525
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        509,
+                                        255,
+                                        525
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Consider a decent generative model",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        256,
+                                        511,
+                                        324,
+                                        522
+                                    ],
+                                    "score": 0.9,
+                                    "content": "G _ { w } : \\mathbb { R } ^ { k } \\mapsto \\mathbb { R } ^ { m }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        324,
+                                        509,
+                                        506,
+                                        525
+                                    ],
+                                    "score": 1.0,
+                                    "content": "trained on the approximate (possibly exact)",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                522,
+                                505,
+                                536
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        522,
+                                        210,
+                                        536
+                                    ],
+                                    "score": 1.0,
+                                    "content": "distribution of user data",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        211,
+                                        523,
+                                        221,
+                                        532
+                                    ],
+                                    "score": 0.8,
+                                    "content": "\\mathcal { D }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        221,
+                                        522,
+                                        264,
+                                        536
+                                    ],
+                                    "score": 1.0,
+                                    "content": "such that",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        264,
+                                        523,
+                                        326,
+                                        534
+                                    ],
+                                    "score": 0.92,
+                                    "content": "x ^ { * } \\approx G _ { w } ( z ^ { * } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        326,
+                                        522,
+                                        344,
+                                        536
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        345,
+                                        522,
+                                        396,
+                                        534
+                                    ],
+                                    "score": 0.92,
+                                    "content": "( x ^ { \\ast } , \\cdot ) \\in \\bar { \\mathcal { D } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        396,
+                                        522,
+                                        479,
+                                        536
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and its latent code",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        480,
+                                        523,
+                                        505,
+                                        533
+                                    ],
+                                    "score": 0.88,
+                                    "content": "z ^ { * } =",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                532,
+                                506,
+                                546
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        533,
+                                        206,
+                                        545
+                                    ],
+                                    "score": 0.84,
+                                    "content": "\\mathrm { a r g m i n } _ { z } \\| G _ { w } ( z ) - x ^ { * } \\|",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        206,
+                                        532,
+                                        506,
+                                        546
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". To fully utilize such a pretrained generative model, we propose gradient",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                544,
+                                505,
+                                556
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        544,
+                                        505,
+                                        556
+                                    ],
+                                    "score": 1.0,
+                                    "content": "inversion in alternative spaces (GIAS), of which pseudocode is presented in Appendix A, which",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                555,
+                                506,
+                                568
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        555,
+                                        244,
+                                        568
+                                    ],
+                                    "score": 1.0,
+                                    "content": "performs latent space search over",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        244,
+                                        557,
+                                        250,
+                                        565
+                                    ],
+                                    "score": 0.8,
+                                    "content": "z",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        251,
+                                        555,
+                                        405,
+                                        568
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and then parameter space search over",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        405,
+                                        557,
+                                        413,
+                                        565
+                                    ],
+                                    "score": 0.75,
+                                    "content": "w",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        414,
+                                        555,
+                                        506,
+                                        568
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". We also illustrate the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                566,
+                                264,
+                                578
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        566,
+                                        264,
+                                        578
+                                    ],
+                                    "score": 1.0,
+                                    "content": "overall procedure of GIAS in Figure 2.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        }
+                    ],
+                    "index": 35.5,
+                    "bbox_fs": [
+                        105,
+                        509,
+                        506,
+                        578
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        588,
+                        506,
+                        655
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                589,
+                                506,
+                                601
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        589,
+                                        506,
+                                        601
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Latent space search. Note that the latent space is typically much smaller than the ambient input",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                600,
+                                504,
+                                612
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        600,
+                                        152,
+                                        612
+                                    ],
+                                    "score": 1.0,
+                                    "content": "space, i.e.,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        152,
+                                        600,
+                                        184,
+                                        610
+                                    ],
+                                    "score": 0.91,
+                                    "content": "k \\ll m",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        184,
+                                        600,
+                                        315,
+                                        612
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", for instances, DCGAN [25] of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        315,
+                                        600,
+                                        351,
+                                        610
+                                    ],
+                                    "score": 0.91,
+                                    "content": "k = 1 0 0",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        351,
+                                        600,
+                                        446,
+                                        612
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and StyleGAN [12] of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        446,
+                                        600,
+                                        504,
+                                        610
+                                    ],
+                                    "score": 0.9,
+                                    "content": "k = 5 1 2 \\times 1 6",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                611,
+                                506,
+                                623
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        611,
+                                        178,
+                                        623
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for image data of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        178,
+                                        611,
+                                        317,
+                                        623
+                                    ],
+                                    "score": 0.86,
+                                    "content": "m = ( \\mathrm { w i d t h } ) \\times ( \\mathrm { h e i g h t } ) \\times ( \\mathrm { c o l o r } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        318,
+                                        611,
+                                        349,
+                                        623
+                                    ],
+                                    "score": 1.0,
+                                    "content": "such as",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        350,
+                                        611,
+                                        400,
+                                        622
+                                    ],
+                                    "score": 0.8,
+                                    "content": "3 2 \\times 3 2 \\times 3",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        401,
+                                        611,
+                                        403,
+                                        623
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        404,
+                                        611,
+                                        464,
+                                        622
+                                    ],
+                                    "score": 0.85,
+                                    "content": "2 5 6 \\times 2 5 6 \\times 3",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        465,
+                                        611,
+                                        506,
+                                        623
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", or larger.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                622,
+                                506,
+                                635
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        622,
+                                        298,
+                                        635
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Using such a pretrained generative model with",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        299,
+                                        623,
+                                        330,
+                                        632
+                                    ],
+                                    "score": 0.89,
+                                    "content": "k \\ll m",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        330,
+                                        622,
+                                        506,
+                                        635
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", the under-determined issues of (3) can be",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                630,
+                                508,
+                                647
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        630,
+                                        366,
+                                        647
+                                    ],
+                                    "score": 1.0,
+                                    "content": "directly mitigated by narrowing down the searching space from",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        366,
+                                        633,
+                                        382,
+                                        642
+                                    ],
+                                    "score": 0.88,
+                                    "content": "\\mathbb { R } ^ { m }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        383,
+                                        630,
+                                        394,
+                                        647
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        395,
+                                        632,
+                                        471,
+                                        645
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\{ G _ { w } ( z ) : z \\in \\mathbb { R } ^ { k } \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        471,
+                                        630,
+                                        508,
+                                        647
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Hence,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                642,
+                                353,
+                                656
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        642,
+                                        353,
+                                        656
+                                    ],
+                                    "score": 1.0,
+                                    "content": "GIAS first performs the latent space search in the followings:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        }
+                    ],
+                    "index": 41.5,
+                    "bbox_fs": [
+                        104,
+                        589,
+                        508,
+                        656
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        228,
+                        659,
+                        382,
+                        679
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                228,
+                                659,
+                                382,
+                                679
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        228,
+                                        659,
+                                        382,
+                                        679
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\operatorname* { m i n } _ { \\substack { z _ { 1 } , \\ldots , z _ { B } \\in \\mathbb { R } ^ { k } } } c \\left( G _ { w } ( z _ { 1 } ) , . . . , G _ { w } ( z _ { B } ) \\right) .",
+                                    "type": "interline_equation",
+                                    "image_path": "f211e132f6ecc08b1bb03ea07a719bb36e5634f96711de6ef5fb6e6d12333b9b.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 45,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                228,
+                                659,
+                                382,
+                                679
+                            ],
+                            "spans": [],
+                            "index": 45
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        688,
+                        504,
+                        722
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                688,
+                                504,
+                                700
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        688,
+                                        324,
+                                        700
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Considering a canonical class of neural network model",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        325,
+                                        689,
+                                        335,
+                                        700
+                                    ],
+                                    "score": 0.88,
+                                    "content": "f _ { \\theta }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        335,
+                                        688,
+                                        492,
+                                        700
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", we can show that the reconstruction of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        492,
+                                        689,
+                                        504,
+                                        699
+                                    ],
+                                    "score": 0.87,
+                                    "content": "x ^ { * }",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 46
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                698,
+                                504,
+                                714
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        698,
+                                        488,
+                                        714
+                                    ],
+                                    "score": 1.0,
+                                    "content": "by latent space search in (5) aligns with that by input space search in (3) if the generative model",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        488,
+                                        700,
+                                        504,
+                                        711
+                                    ],
+                                    "score": 0.88,
+                                    "content": "G _ { w }",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 47
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                711,
+                                305,
+                                724
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        711,
+                                        305,
+                                        724
+                                    ],
+                                    "score": 1.0,
+                                    "content": "approximates input data with small enough error.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 48
+                        }
+                    ],
+                    "index": 47,
+                    "bbox_fs": [
+                        104,
+                        688,
+                        504,
+                        724
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "image",
+                    "bbox": [
+                        150,
+                        75,
+                        459,
+                        248
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                150,
+                                75,
+                                459,
+                                248
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        150,
+                                        75,
+                                        459,
+                                        248
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                150,
+                                                75,
+                                                459,
+                                                248
+                                            ],
+                                            "score": 0.972,
+                                            "type": "image",
+                                            "image_path": "86214e463c23aef68db3947ced6b1f76f55c86ba1e2dd66943856f0c1dfa79be.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 1,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        150,
+                                        75,
+                                        459,
+                                        132.66666666666666
+                                    ],
+                                    "spans": [],
+                                    "index": 0
+                                },
+                                {
+                                    "bbox": [
+                                        150,
+                                        132.66666666666666,
+                                        459,
+                                        190.33333333333331
+                                    ],
+                                    "spans": [],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        150,
+                                        190.33333333333331,
+                                        459,
+                                        247.99999999999997
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                106,
+                                269,
+                                504,
+                                292
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        269,
+                                        504,
+                                        282
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                269,
+                                                353,
+                                                282
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 2: An overview of GIAS. GIAS optimizes a latent code",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                354,
+                                                272,
+                                                360,
+                                                280
+                                            ],
+                                            "score": 0.75,
+                                            "content": "z",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                361,
+                                                269,
+                                                495,
+                                                282
+                                            ],
+                                            "score": 1.0,
+                                            "content": "and generative model parameters",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                495,
+                                                272,
+                                                504,
+                                                280
+                                            ],
+                                            "score": 0.68,
+                                            "content": "w",
+                                            "type": "inline_equation"
+                                        }
+                                    ],
+                                    "index": 3
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        280,
+                                        311,
+                                        293
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                280,
+                                                311,
+                                                293
+                                            ],
+                                            "score": 1.0,
+                                            "content": "to reconstruct the data which matches the gradient.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 4
+                                }
+                            ],
+                            "index": 3.5
+                        }
+                    ],
+                    "index": 2.25
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        309,
+                        505,
+                        366
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                309,
+                                505,
+                                324
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        309,
+                                        232,
+                                        324
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Property 1. For an input data",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        232,
+                                        311,
+                                        283,
+                                        322
+                                    ],
+                                    "score": 0.93,
+                                    "content": "x ^ { * } \\in [ 0 , 1 ] ^ { m }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        283,
+                                        309,
+                                        505,
+                                        324
+                                    ],
+                                    "score": 1.0,
+                                    "content": "consider the gradient inversion problem of minimizing",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                321,
+                                506,
+                                334
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        321,
+                                        506,
+                                        334
+                                    ],
+                                    "score": 1.0,
+                                    "content": "cost c in (3), where a canonical form of deep learning for classification is considered and the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                332,
+                                506,
+                                345
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        332,
+                                        195,
+                                        345
+                                    ],
+                                    "score": 1.0,
+                                    "content": "discrepancy measure",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        195,
+                                        333,
+                                        201,
+                                        342
+                                    ],
+                                    "score": 0.6,
+                                    "content": "d",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        202,
+                                        332,
+                                        212,
+                                        345
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        212,
+                                        333,
+                                        222,
+                                        344
+                                    ],
+                                    "score": 0.87,
+                                    "content": "\\ell _ { 2 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        222,
+                                        332,
+                                        473,
+                                        345
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-distance. Suppose that it has the unique global minimizer at",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        473,
+                                        333,
+                                        484,
+                                        342
+                                    ],
+                                    "score": 0.86,
+                                    "content": "x ^ { * }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        484,
+                                        332,
+                                        506,
+                                        345
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Let",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                342,
+                                507,
+                                357
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        344,
+                                        131,
+                                        354
+                                    ],
+                                    "score": 0.88,
+                                    "content": "\\varepsilon \\geq 0",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        132,
+                                        342,
+                                        286,
+                                        357
+                                    ],
+                                    "score": 1.0,
+                                    "content": "be the approximation error bound on",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        287,
+                                        344,
+                                        298,
+                                        354
+                                    ],
+                                    "score": 0.86,
+                                    "content": "x ^ { * }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        298,
+                                        342,
+                                        387,
+                                        357
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for generative model",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        387,
+                                        343,
+                                        480,
+                                        356
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\mathring { G } _ { w } : [ 0 , 1 ] ^ { k } \\mapsto [ 0 , 1 ] ^ { m }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        480,
+                                        342,
+                                        507,
+                                        357
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Then,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                353,
+                                366,
+                                369
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        353,
+                                        154,
+                                        369
+                                    ],
+                                    "score": 1.0,
+                                    "content": "there exists",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        154,
+                                        354,
+                                        190,
+                                        366
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\delta ( \\varepsilon ) \\geq 0",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        191,
+                                        353,
+                                        261,
+                                        369
+                                    ],
+                                    "score": 1.0,
+                                    "content": "such that for any",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        261,
+                                        354,
+                                        362,
+                                        366
+                                    ],
+                                    "score": 0.74,
+                                    "content": "z ^ { * } \\in \\arg \\operatorname* { m i n } _ { z } c ( G _ { w } ( z ) ) ,",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        362,
+                                        353,
+                                        366,
+                                        369
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        }
+                    ],
+                    "index": 7
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        253,
+                        369,
+                        355,
+                        383
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                253,
+                                369,
+                                355,
+                                383
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        253,
+                                        369,
+                                        355,
+                                        383
+                                    ],
+                                    "score": 0.81,
+                                    "content": "\\| G _ { w } ( z ^ { * } ) - x ^ { * } \\| \\leq \\delta ( \\varepsilon ) ,",
+                                    "type": "interline_equation",
+                                    "image_path": "1bdcb001746faf4040209eaa567ad172a6b5dd5ddb3925f3fe34e4a5ecbd1a6f.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 10,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                253,
+                                369,
+                                355,
+                                383
+                            ],
+                            "spans": [],
+                            "index": 10
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        384,
+                        277,
+                        396
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                383,
+                                277,
+                                397
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        383,
+                                        196,
+                                        397
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of which upper bound",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        196,
+                                        384,
+                                        235,
+                                        396
+                                    ],
+                                    "score": 0.89,
+                                    "content": "\\delta ( \\varepsilon )  0",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        235,
+                                        383,
+                                        247,
+                                        397
+                                    ],
+                                    "score": 1.0,
+                                    "content": "as",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        248,
+                                        385,
+                                        273,
+                                        394
+                                    ],
+                                    "score": 0.81,
+                                    "content": "\\varepsilon \\to 0",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        274,
+                                        383,
+                                        277,
+                                        397
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        }
+                    ],
+                    "index": 11
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        402,
+                        505,
+                        447
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                403,
+                                505,
+                                415
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        403,
+                                        505,
+                                        415
+                                    ],
+                                    "score": 1.0,
+                                    "content": "A rigorous statement of Property 1 and its proof are provided in Appendix B, where we prove and use",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                412,
+                                506,
+                                427
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        412,
+                                        277,
+                                        427
+                                    ],
+                                    "score": 1.0,
+                                    "content": "that the cost function is continuous around",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        277,
+                                        414,
+                                        289,
+                                        424
+                                    ],
+                                    "score": 0.88,
+                                    "content": "x ^ { * }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        289,
+                                        412,
+                                        506,
+                                        427
+                                    ],
+                                    "score": 1.0,
+                                    "content": "under the assumptions. This property justifies solving",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                425,
+                                505,
+                                437
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        425,
+                                        241,
+                                        437
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the latent space search in (5) for",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        241,
+                                        425,
+                                        254,
+                                        435
+                                    ],
+                                    "score": 0.3,
+                                    "content": "\\mathrm { F L }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        255,
+                                        425,
+                                        505,
+                                        437
+                                    ],
+                                    "score": 1.0,
+                                    "content": "scenarios training neural network model while it requires an",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                437,
+                                216,
+                                447
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        437,
+                                        216,
+                                        447
+                                    ],
+                                    "score": 1.0,
+                                    "content": "accurate generative model.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        }
+                    ],
+                    "index": 13.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        457,
+                        505,
+                        555
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                457,
+                                505,
+                                470
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        457,
+                                        505,
+                                        470
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Parameter space search. Using the latent space search only, there can be inevitable reconstruction",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                469,
+                                505,
+                                482
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        469,
+                                        505,
+                                        482
+                                    ],
+                                    "score": 1.0,
+                                    "content": "error due to the imperfection of generative model. This is mainly because we cannot perfectly",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                479,
+                                506,
+                                492
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        479,
+                                        506,
+                                        492
+                                    ],
+                                    "score": 1.0,
+                                    "content": "prepare the generative model for every plausible data in advance. Similar difficulty of the latent",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                490,
+                                506,
+                                503
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        490,
+                                        506,
+                                        503
+                                    ],
+                                    "score": 1.0,
+                                    "content": "space search has been reported even when inverting GAN [33, 3, 4] for plausible but new data",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                500,
+                                506,
+                                516
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        500,
+                                        158,
+                                        516
+                                    ],
+                                    "score": 1.0,
+                                    "content": "directly, i.e.,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        159,
+                                        501,
+                                        242,
+                                        514
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\mathrm { m i n } _ { z } \\parallel G _ { w } ( z ) - x ^ { * } \\parallel",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        243,
+                                        500,
+                                        267,
+                                        516
+                                    ],
+                                    "score": 1.0,
+                                    "content": "given",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        268,
+                                        502,
+                                        279,
+                                        512
+                                    ],
+                                    "score": 0.86,
+                                    "content": "x ^ { * }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        279,
+                                        500,
+                                        506,
+                                        516
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", rather than inverting gradient. Bau et al. [3] propose an",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                512,
+                                505,
+                                525
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        512,
+                                        418,
+                                        525
+                                    ],
+                                    "score": 1.0,
+                                    "content": "instance-specific model adaptation, which slightly adjusts the model parameter",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        419,
+                                        514,
+                                        427,
+                                        522
+                                    ],
+                                    "score": 0.76,
+                                    "content": "w",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        428,
+                                        512,
+                                        505,
+                                        525
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to (a part of source",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                522,
+                                506,
+                                536
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        522,
+                                        137,
+                                        536
+                                    ],
+                                    "score": 1.0,
+                                    "content": "image)",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        137,
+                                        524,
+                                        148,
+                                        533
+                                    ],
+                                    "score": 0.85,
+                                    "content": "x ^ { * }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        149,
+                                        522,
+                                        265,
+                                        536
+                                    ],
+                                    "score": 1.0,
+                                    "content": "after obtaining a latent code",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        265,
+                                        524,
+                                        276,
+                                        533
+                                    ],
+                                    "score": 0.87,
+                                    "content": "z ^ { * }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        276,
+                                        522,
+                                        291,
+                                        536
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        292,
+                                        524,
+                                        303,
+                                        533
+                                    ],
+                                    "score": 0.87,
+                                    "content": "x ^ { * }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        303,
+                                        522,
+                                        506,
+                                        536
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Inspired by such an instance-specific adaptation,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                533,
+                                505,
+                                548
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        533,
+                                        348,
+                                        548
+                                    ],
+                                    "score": 1.0,
+                                    "content": "GIAS performs the following parameter space search over",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        348,
+                                        536,
+                                        357,
+                                        544
+                                    ],
+                                    "score": 0.79,
+                                    "content": "w",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        357,
+                                        533,
+                                        505,
+                                        548
+                                    ],
+                                    "score": 1.0,
+                                    "content": "preceded by the latent space search",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                546,
+                                140,
+                                558
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        546,
+                                        128,
+                                        558
+                                    ],
+                                    "score": 1.0,
+                                    "content": "over",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        129,
+                                        547,
+                                        135,
+                                        555
+                                    ],
+                                    "score": 0.72,
+                                    "content": "z",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        135,
+                                        546,
+                                        140,
+                                        558
+                                    ],
+                                    "score": 1.0,
+                                    "content": ":",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        }
+                    ],
+                    "index": 20
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        229,
+                        554,
+                        380,
+                        573
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                229,
+                                554,
+                                380,
+                                573
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        229,
+                                        554,
+                                        380,
+                                        573
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\operatorname * { m i n } _ { w _ { 1 } , . . . , w _ { B } } \\ c \\left( G _ { w _ { 1 } } ( z _ { 1 } ) , . . . , G _ { w _ { B } } ( z _ { B } ) \\right) \\ ,",
+                                    "type": "interline_equation",
+                                    "image_path": "325187a10dff3508c4c5b1ba8b741dcbada510c6de970a1f441713009762a836.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 25,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                229,
+                                554,
+                                380,
+                                573
+                            ],
+                            "spans": [],
+                            "index": 25
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        574,
+                        266,
+                        585
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                572,
+                                268,
+                                587
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        572,
+                                        133,
+                                        587
+                                    ],
+                                    "score": 1.0,
+                                    "content": "where",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        133,
+                                        576,
+                                        177,
+                                        585
+                                    ],
+                                    "score": 0.91,
+                                    "content": "z _ { 1 } , \\dots , z _ { B }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        177,
+                                        572,
+                                        268,
+                                        587
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are obtained from (5).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        }
+                    ],
+                    "index": 26
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        595,
+                        505,
+                        651
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                595,
+                                505,
+                                608
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        595,
+                                        298,
+                                        608
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Remark. We propose the optimization over",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        298,
+                                        599,
+                                        307,
+                                        606
+                                    ],
+                                    "score": 0.76,
+                                    "content": "w",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        307,
+                                        595,
+                                        401,
+                                        608
+                                    ],
+                                    "score": 1.0,
+                                    "content": "followed by that over",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        402,
+                                        599,
+                                        408,
+                                        606
+                                    ],
+                                    "score": 0.75,
+                                    "content": "z",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        409,
+                                        595,
+                                        505,
+                                        608
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sequentially This is to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                606,
+                                506,
+                                621
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        606,
+                                        506,
+                                        621
+                                    ],
+                                    "score": 1.0,
+                                    "content": "maximally utilize the benefit of mitigating the under-determined issue from reducing the searching",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                618,
+                                505,
+                                630
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        618,
+                                        414,
+                                        630
+                                    ],
+                                    "score": 1.0,
+                                    "content": "space on the pretrained model. However, the benefit would be degenerated if",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        414,
+                                        619,
+                                        420,
+                                        628
+                                    ],
+                                    "score": 0.77,
+                                    "content": "z",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        421,
+                                        618,
+                                        438,
+                                        630
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        438,
+                                        619,
+                                        447,
+                                        628
+                                    ],
+                                    "score": 0.79,
+                                    "content": "w",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        447,
+                                        618,
+                                        505,
+                                        630
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are optimized",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                627,
+                                505,
+                                643
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        627,
+                                        148,
+                                        643
+                                    ],
+                                    "score": 1.0,
+                                    "content": "jointly or",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        148,
+                                        631,
+                                        157,
+                                        639
+                                    ],
+                                    "score": 0.74,
+                                    "content": "w",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        158,
+                                        627,
+                                        505,
+                                        643
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is optimized first. We provide an empirical justification on the proposed searching",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                640,
+                                201,
+                                652
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        640,
+                                        201,
+                                        652
+                                    ],
+                                    "score": 1.0,
+                                    "content": "strategy in Section 5.1.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        }
+                    ],
+                    "index": 29
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        655,
+                        505,
+                        722
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                655,
+                                507,
+                                669
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        655,
+                                        507,
+                                        669
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We perform each search in GIAS using a standard gradient method to the cost function directly.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                667,
+                                506,
+                                680
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        667,
+                                        506,
+                                        680
+                                    ],
+                                    "score": 1.0,
+                                    "content": "It is worth noting that those optimizations (5) and (7) with generative model can be tackled in a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                678,
+                                505,
+                                690
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        678,
+                                        505,
+                                        690
+                                    ],
+                                    "score": 1.0,
+                                    "content": "recursive manner as R-GAP [32] reconstructs each layer from output to input. We provide details",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                688,
+                                505,
+                                702
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        688,
+                                        505,
+                                        702
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and performance of the recursive procedure in Appendix C, where employing generative model",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                700,
+                                505,
+                                713
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        700,
+                                        505,
+                                        713
+                                    ],
+                                    "score": 1.0,
+                                    "content": "improves the inversion accuracy of R-GAP substantially, while R-GAP apparently suffers from an",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                710,
+                                345,
+                                723
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        710,
+                                        231,
+                                        723
+                                    ],
+                                    "score": 1.0,
+                                    "content": "error accumulation issue when",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        231,
+                                        711,
+                                        241,
+                                        722
+                                    ],
+                                    "score": 0.87,
+                                    "content": "f _ { \\theta }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        242,
+                                        710,
+                                        345,
+                                        723
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is a deep neural network.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        }
+                    ],
+                    "index": 34.5
+                }
+            ],
+            "page_idx": 4,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        302,
+                        741,
+                        308,
+                        750
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                302,
+                                740,
+                                309,
+                                753
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        302,
+                                        740,
+                                        309,
+                                        753
+                                    ],
+                                    "score": 1.0,
+                                    "content": "5",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "image",
+                    "bbox": [
+                        150,
+                        75,
+                        459,
+                        248
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                150,
+                                75,
+                                459,
+                                248
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        150,
+                                        75,
+                                        459,
+                                        248
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                150,
+                                                75,
+                                                459,
+                                                248
+                                            ],
+                                            "score": 0.972,
+                                            "type": "image",
+                                            "image_path": "86214e463c23aef68db3947ced6b1f76f55c86ba1e2dd66943856f0c1dfa79be.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 1,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        150,
+                                        75,
+                                        459,
+                                        132.66666666666666
+                                    ],
+                                    "spans": [],
+                                    "index": 0
+                                },
+                                {
+                                    "bbox": [
+                                        150,
+                                        132.66666666666666,
+                                        459,
+                                        190.33333333333331
+                                    ],
+                                    "spans": [],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        150,
+                                        190.33333333333331,
+                                        459,
+                                        247.99999999999997
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                106,
+                                269,
+                                504,
+                                292
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        269,
+                                        504,
+                                        282
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                269,
+                                                353,
+                                                282
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 2: An overview of GIAS. GIAS optimizes a latent code",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                354,
+                                                272,
+                                                360,
+                                                280
+                                            ],
+                                            "score": 0.75,
+                                            "content": "z",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                361,
+                                                269,
+                                                495,
+                                                282
+                                            ],
+                                            "score": 1.0,
+                                            "content": "and generative model parameters",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                495,
+                                                272,
+                                                504,
+                                                280
+                                            ],
+                                            "score": 0.68,
+                                            "content": "w",
+                                            "type": "inline_equation"
+                                        }
+                                    ],
+                                    "index": 3
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        280,
+                                        311,
+                                        293
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                280,
+                                                311,
+                                                293
+                                            ],
+                                            "score": 1.0,
+                                            "content": "to reconstruct the data which matches the gradient.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 4
+                                }
+                            ],
+                            "index": 3.5
+                        }
+                    ],
+                    "index": 2.25
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        309,
+                        505,
+                        366
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                309,
+                                505,
+                                324
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        309,
+                                        232,
+                                        324
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Property 1. For an input data",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        232,
+                                        311,
+                                        283,
+                                        322
+                                    ],
+                                    "score": 0.93,
+                                    "content": "x ^ { * } \\in [ 0 , 1 ] ^ { m }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        283,
+                                        309,
+                                        505,
+                                        324
+                                    ],
+                                    "score": 1.0,
+                                    "content": "consider the gradient inversion problem of minimizing",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                321,
+                                506,
+                                334
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        321,
+                                        506,
+                                        334
+                                    ],
+                                    "score": 1.0,
+                                    "content": "cost c in (3), where a canonical form of deep learning for classification is considered and the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                332,
+                                506,
+                                345
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        332,
+                                        195,
+                                        345
+                                    ],
+                                    "score": 1.0,
+                                    "content": "discrepancy measure",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        195,
+                                        333,
+                                        201,
+                                        342
+                                    ],
+                                    "score": 0.6,
+                                    "content": "d",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        202,
+                                        332,
+                                        212,
+                                        345
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        212,
+                                        333,
+                                        222,
+                                        344
+                                    ],
+                                    "score": 0.87,
+                                    "content": "\\ell _ { 2 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        222,
+                                        332,
+                                        473,
+                                        345
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-distance. Suppose that it has the unique global minimizer at",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        473,
+                                        333,
+                                        484,
+                                        342
+                                    ],
+                                    "score": 0.86,
+                                    "content": "x ^ { * }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        484,
+                                        332,
+                                        506,
+                                        345
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Let",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                342,
+                                507,
+                                357
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        344,
+                                        131,
+                                        354
+                                    ],
+                                    "score": 0.88,
+                                    "content": "\\varepsilon \\geq 0",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        132,
+                                        342,
+                                        286,
+                                        357
+                                    ],
+                                    "score": 1.0,
+                                    "content": "be the approximation error bound on",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        287,
+                                        344,
+                                        298,
+                                        354
+                                    ],
+                                    "score": 0.86,
+                                    "content": "x ^ { * }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        298,
+                                        342,
+                                        387,
+                                        357
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for generative model",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        387,
+                                        343,
+                                        480,
+                                        356
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\mathring { G } _ { w } : [ 0 , 1 ] ^ { k } \\mapsto [ 0 , 1 ] ^ { m }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        480,
+                                        342,
+                                        507,
+                                        357
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Then,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                353,
+                                366,
+                                369
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        353,
+                                        154,
+                                        369
+                                    ],
+                                    "score": 1.0,
+                                    "content": "there exists",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        154,
+                                        354,
+                                        190,
+                                        366
+                                    ],
+                                    "score": 0.93,
+                                    "content": "\\delta ( \\varepsilon ) \\geq 0",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        191,
+                                        353,
+                                        261,
+                                        369
+                                    ],
+                                    "score": 1.0,
+                                    "content": "such that for any",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        261,
+                                        354,
+                                        362,
+                                        366
+                                    ],
+                                    "score": 0.74,
+                                    "content": "z ^ { * } \\in \\arg \\operatorname* { m i n } _ { z } c ( G _ { w } ( z ) ) ,",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        362,
+                                        353,
+                                        366,
+                                        369
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        }
+                    ],
+                    "index": 7,
+                    "bbox_fs": [
+                        105,
+                        309,
+                        507,
+                        369
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        253,
+                        369,
+                        355,
+                        383
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                253,
+                                369,
+                                355,
+                                383
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        253,
+                                        369,
+                                        355,
+                                        383
+                                    ],
+                                    "score": 0.81,
+                                    "content": "\\| G _ { w } ( z ^ { * } ) - x ^ { * } \\| \\leq \\delta ( \\varepsilon ) ,",
+                                    "type": "interline_equation",
+                                    "image_path": "1bdcb001746faf4040209eaa567ad172a6b5dd5ddb3925f3fe34e4a5ecbd1a6f.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 10,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                253,
+                                369,
+                                355,
+                                383
+                            ],
+                            "spans": [],
+                            "index": 10
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        384,
+                        277,
+                        396
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                383,
+                                277,
+                                397
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        383,
+                                        196,
+                                        397
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of which upper bound",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        196,
+                                        384,
+                                        235,
+                                        396
+                                    ],
+                                    "score": 0.89,
+                                    "content": "\\delta ( \\varepsilon )  0",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        235,
+                                        383,
+                                        247,
+                                        397
+                                    ],
+                                    "score": 1.0,
+                                    "content": "as",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        248,
+                                        385,
+                                        273,
+                                        394
+                                    ],
+                                    "score": 0.81,
+                                    "content": "\\varepsilon \\to 0",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        274,
+                                        383,
+                                        277,
+                                        397
+                                    ],
+                                    "score": 1.0,
+                                    "content": ".",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        }
+                    ],
+                    "index": 11,
+                    "bbox_fs": [
+                        105,
+                        383,
+                        277,
+                        397
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        402,
+                        505,
+                        447
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                403,
+                                505,
+                                415
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        403,
+                                        505,
+                                        415
+                                    ],
+                                    "score": 1.0,
+                                    "content": "A rigorous statement of Property 1 and its proof are provided in Appendix B, where we prove and use",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                412,
+                                506,
+                                427
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        412,
+                                        277,
+                                        427
+                                    ],
+                                    "score": 1.0,
+                                    "content": "that the cost function is continuous around",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        277,
+                                        414,
+                                        289,
+                                        424
+                                    ],
+                                    "score": 0.88,
+                                    "content": "x ^ { * }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        289,
+                                        412,
+                                        506,
+                                        427
+                                    ],
+                                    "score": 1.0,
+                                    "content": "under the assumptions. This property justifies solving",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                425,
+                                505,
+                                437
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        425,
+                                        241,
+                                        437
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the latent space search in (5) for",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        241,
+                                        425,
+                                        254,
+                                        435
+                                    ],
+                                    "score": 0.3,
+                                    "content": "\\mathrm { F L }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        255,
+                                        425,
+                                        505,
+                                        437
+                                    ],
+                                    "score": 1.0,
+                                    "content": "scenarios training neural network model while it requires an",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                437,
+                                216,
+                                447
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        437,
+                                        216,
+                                        447
+                                    ],
+                                    "score": 1.0,
+                                    "content": "accurate generative model.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        }
+                    ],
+                    "index": 13.5,
+                    "bbox_fs": [
+                        105,
+                        403,
+                        506,
+                        447
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        457,
+                        505,
+                        555
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                457,
+                                505,
+                                470
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        457,
+                                        505,
+                                        470
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Parameter space search. Using the latent space search only, there can be inevitable reconstruction",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                469,
+                                505,
+                                482
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        469,
+                                        505,
+                                        482
+                                    ],
+                                    "score": 1.0,
+                                    "content": "error due to the imperfection of generative model. This is mainly because we cannot perfectly",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                479,
+                                506,
+                                492
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        479,
+                                        506,
+                                        492
+                                    ],
+                                    "score": 1.0,
+                                    "content": "prepare the generative model for every plausible data in advance. Similar difficulty of the latent",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                490,
+                                506,
+                                503
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        490,
+                                        506,
+                                        503
+                                    ],
+                                    "score": 1.0,
+                                    "content": "space search has been reported even when inverting GAN [33, 3, 4] for plausible but new data",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                500,
+                                506,
+                                516
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        500,
+                                        158,
+                                        516
+                                    ],
+                                    "score": 1.0,
+                                    "content": "directly, i.e.,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        159,
+                                        501,
+                                        242,
+                                        514
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\mathrm { m i n } _ { z } \\parallel G _ { w } ( z ) - x ^ { * } \\parallel",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        243,
+                                        500,
+                                        267,
+                                        516
+                                    ],
+                                    "score": 1.0,
+                                    "content": "given",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        268,
+                                        502,
+                                        279,
+                                        512
+                                    ],
+                                    "score": 0.86,
+                                    "content": "x ^ { * }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        279,
+                                        500,
+                                        506,
+                                        516
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", rather than inverting gradient. Bau et al. [3] propose an",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                512,
+                                505,
+                                525
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        512,
+                                        418,
+                                        525
+                                    ],
+                                    "score": 1.0,
+                                    "content": "instance-specific model adaptation, which slightly adjusts the model parameter",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        419,
+                                        514,
+                                        427,
+                                        522
+                                    ],
+                                    "score": 0.76,
+                                    "content": "w",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        428,
+                                        512,
+                                        505,
+                                        525
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to (a part of source",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                522,
+                                506,
+                                536
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        522,
+                                        137,
+                                        536
+                                    ],
+                                    "score": 1.0,
+                                    "content": "image)",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        137,
+                                        524,
+                                        148,
+                                        533
+                                    ],
+                                    "score": 0.85,
+                                    "content": "x ^ { * }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        149,
+                                        522,
+                                        265,
+                                        536
+                                    ],
+                                    "score": 1.0,
+                                    "content": "after obtaining a latent code",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        265,
+                                        524,
+                                        276,
+                                        533
+                                    ],
+                                    "score": 0.87,
+                                    "content": "z ^ { * }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        276,
+                                        522,
+                                        291,
+                                        536
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        292,
+                                        524,
+                                        303,
+                                        533
+                                    ],
+                                    "score": 0.87,
+                                    "content": "x ^ { * }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        303,
+                                        522,
+                                        506,
+                                        536
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Inspired by such an instance-specific adaptation,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                533,
+                                505,
+                                548
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        533,
+                                        348,
+                                        548
+                                    ],
+                                    "score": 1.0,
+                                    "content": "GIAS performs the following parameter space search over",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        348,
+                                        536,
+                                        357,
+                                        544
+                                    ],
+                                    "score": 0.79,
+                                    "content": "w",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        357,
+                                        533,
+                                        505,
+                                        548
+                                    ],
+                                    "score": 1.0,
+                                    "content": "preceded by the latent space search",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                546,
+                                140,
+                                558
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        546,
+                                        128,
+                                        558
+                                    ],
+                                    "score": 1.0,
+                                    "content": "over",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        129,
+                                        547,
+                                        135,
+                                        555
+                                    ],
+                                    "score": 0.72,
+                                    "content": "z",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        135,
+                                        546,
+                                        140,
+                                        558
+                                    ],
+                                    "score": 1.0,
+                                    "content": ":",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        }
+                    ],
+                    "index": 20,
+                    "bbox_fs": [
+                        104,
+                        457,
+                        506,
+                        558
+                    ]
+                },
+                {
+                    "type": "interline_equation",
+                    "bbox": [
+                        229,
+                        554,
+                        380,
+                        573
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                229,
+                                554,
+                                380,
+                                573
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        229,
+                                        554,
+                                        380,
+                                        573
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\operatorname * { m i n } _ { w _ { 1 } , . . . , w _ { B } } \\ c \\left( G _ { w _ { 1 } } ( z _ { 1 } ) , . . . , G _ { w _ { B } } ( z _ { B } ) \\right) \\ ,",
+                                    "type": "interline_equation",
+                                    "image_path": "325187a10dff3508c4c5b1ba8b741dcbada510c6de970a1f441713009762a836.jpg"
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 25,
+                    "virtual_lines": [
+                        {
+                            "bbox": [
+                                229,
+                                554,
+                                380,
+                                573
+                            ],
+                            "spans": [],
+                            "index": 25
+                        }
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        108,
+                        574,
+                        266,
+                        585
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                572,
+                                268,
+                                587
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        572,
+                                        133,
+                                        587
+                                    ],
+                                    "score": 1.0,
+                                    "content": "where",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        133,
+                                        576,
+                                        177,
+                                        585
+                                    ],
+                                    "score": 0.91,
+                                    "content": "z _ { 1 } , \\dots , z _ { B }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        177,
+                                        572,
+                                        268,
+                                        587
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are obtained from (5).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        }
+                    ],
+                    "index": 26,
+                    "bbox_fs": [
+                        106,
+                        572,
+                        268,
+                        587
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        595,
+                        505,
+                        651
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                595,
+                                505,
+                                608
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        595,
+                                        298,
+                                        608
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Remark. We propose the optimization over",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        298,
+                                        599,
+                                        307,
+                                        606
+                                    ],
+                                    "score": 0.76,
+                                    "content": "w",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        307,
+                                        595,
+                                        401,
+                                        608
+                                    ],
+                                    "score": 1.0,
+                                    "content": "followed by that over",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        402,
+                                        599,
+                                        408,
+                                        606
+                                    ],
+                                    "score": 0.75,
+                                    "content": "z",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        409,
+                                        595,
+                                        505,
+                                        608
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sequentially This is to",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                606,
+                                506,
+                                621
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        606,
+                                        506,
+                                        621
+                                    ],
+                                    "score": 1.0,
+                                    "content": "maximally utilize the benefit of mitigating the under-determined issue from reducing the searching",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                618,
+                                505,
+                                630
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        618,
+                                        414,
+                                        630
+                                    ],
+                                    "score": 1.0,
+                                    "content": "space on the pretrained model. However, the benefit would be degenerated if",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        414,
+                                        619,
+                                        420,
+                                        628
+                                    ],
+                                    "score": 0.77,
+                                    "content": "z",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        421,
+                                        618,
+                                        438,
+                                        630
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        438,
+                                        619,
+                                        447,
+                                        628
+                                    ],
+                                    "score": 0.79,
+                                    "content": "w",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        447,
+                                        618,
+                                        505,
+                                        630
+                                    ],
+                                    "score": 1.0,
+                                    "content": "are optimized",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                627,
+                                505,
+                                643
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        627,
+                                        148,
+                                        643
+                                    ],
+                                    "score": 1.0,
+                                    "content": "jointly or",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        148,
+                                        631,
+                                        157,
+                                        639
+                                    ],
+                                    "score": 0.74,
+                                    "content": "w",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        158,
+                                        627,
+                                        505,
+                                        643
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is optimized first. We provide an empirical justification on the proposed searching",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                640,
+                                201,
+                                652
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        640,
+                                        201,
+                                        652
+                                    ],
+                                    "score": 1.0,
+                                    "content": "strategy in Section 5.1.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        }
+                    ],
+                    "index": 29,
+                    "bbox_fs": [
+                        104,
+                        595,
+                        506,
+                        652
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        655,
+                        505,
+                        722
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                655,
+                                507,
+                                669
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        655,
+                                        507,
+                                        669
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We perform each search in GIAS using a standard gradient method to the cost function directly.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                667,
+                                506,
+                                680
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        667,
+                                        506,
+                                        680
+                                    ],
+                                    "score": 1.0,
+                                    "content": "It is worth noting that those optimizations (5) and (7) with generative model can be tackled in a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                678,
+                                505,
+                                690
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        678,
+                                        505,
+                                        690
+                                    ],
+                                    "score": 1.0,
+                                    "content": "recursive manner as R-GAP [32] reconstructs each layer from output to input. We provide details",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                688,
+                                505,
+                                702
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        688,
+                                        505,
+                                        702
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and performance of the recursive procedure in Appendix C, where employing generative model",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                700,
+                                505,
+                                713
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        700,
+                                        505,
+                                        713
+                                    ],
+                                    "score": 1.0,
+                                    "content": "improves the inversion accuracy of R-GAP substantially, while R-GAP apparently suffers from an",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                710,
+                                345,
+                                723
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        710,
+                                        231,
+                                        723
+                                    ],
+                                    "score": 1.0,
+                                    "content": "error accumulation issue when",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        231,
+                                        711,
+                                        241,
+                                        722
+                                    ],
+                                    "score": 0.87,
+                                    "content": "f _ { \\theta }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        242,
+                                        710,
+                                        345,
+                                        723
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is a deep neural network.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        }
+                    ],
+                    "index": 34.5,
+                    "bbox_fs": [
+                        105,
+                        655,
+                        507,
+                        723
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "image",
+                    "bbox": [
+                        103,
+                        92,
+                        500,
+                        263
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                103,
+                                92,
+                                500,
+                                263
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        103,
+                                        92,
+                                        500,
+                                        263
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                103,
+                                                92,
+                                                500,
+                                                263
+                                            ],
+                                            "score": 0.975,
+                                            "type": "image",
+                                            "image_path": "3e98af5f62fd12872d10b40932ee68ec2b13001baf19a78e4df336bdfc8cd337.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 1,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        103,
+                                        92,
+                                        500,
+                                        149.0
+                                    ],
+                                    "spans": [],
+                                    "index": 0
+                                },
+                                {
+                                    "bbox": [
+                                        103,
+                                        149.0,
+                                        500,
+                                        206.0
+                                    ],
+                                    "spans": [],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        103,
+                                        206.0,
+                                        500,
+                                        263.0
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                106,
+                                270,
+                                505,
+                                314
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        270,
+                                        506,
+                                        282
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                270,
+                                                506,
+                                                282
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 3: Comparison of different searching spaces. (a) Each row shows reconstructed images of",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 3
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        280,
+                                        506,
+                                        293
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                280,
+                                                506,
+                                                293
+                                            ],
+                                            "score": 1.0,
+                                            "content": "different optimization domains. The first three rows share the same latent space search of 1, 500",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 4
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        291,
+                                        505,
+                                        304
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                292,
+                                                167,
+                                                304
+                                            ],
+                                            "score": 1.0,
+                                            "content": "iterations, and",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                167,
+                                                291,
+                                                200,
+                                                303
+                                            ],
+                                            "score": 0.86,
+                                            "content": "{ \\mathrm { G I } } { - z } / w",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                201,
+                                                292,
+                                                505,
+                                                304
+                                            ],
+                                            "score": 1.0,
+                                            "content": "is verified to be the best option to fully exploits the knowledge inside the",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 5
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        303,
+                                        450,
+                                        316
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                303,
+                                                450,
+                                                316
+                                            ],
+                                            "score": 1.0,
+                                            "content": "generative model. (b) Cost function over iterations of different optimization domains.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 6
+                                }
+                            ],
+                            "index": 4.5
+                        }
+                    ],
+                    "index": 2.75
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        341,
+                        345,
+                        352
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                339,
+                                346,
+                                354
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        339,
+                                        346,
+                                        354
+                                    ],
+                                    "score": 1.0,
+                                    "content": "4.2 Gradient inversion to meta-learn generative model",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        }
+                    ],
+                    "index": 7
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        363,
+                        505,
+                        528
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                363,
+                                506,
+                                376
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        363,
+                                        506,
+                                        376
+                                    ],
+                                    "score": 1.0,
+                                    "content": "For the case that pretrained generative model is unavailable, we devise an algorithm to train a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                374,
+                                506,
+                                387
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        374,
+                                        178,
+                                        387
+                                    ],
+                                    "score": 1.0,
+                                    "content": "generative model",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        178,
+                                        375,
+                                        193,
+                                        385
+                                    ],
+                                    "score": 0.89,
+                                    "content": "G _ { w }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        194,
+                                        374,
+                                        230,
+                                        387
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for a set",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        230,
+                                        374,
+                                        289,
+                                        387
+                                    ],
+                                    "score": 0.94,
+                                    "content": "\\mathcal { S } = \\{ ( \\theta _ { i } , g _ { i } ) \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        289,
+                                        374,
+                                        506,
+                                        387
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of gradient inversion tasks. Since each inversion task",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                385,
+                                506,
+                                398
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        385,
+                                        506,
+                                        398
+                                    ],
+                                    "score": 1.0,
+                                    "content": "can be considered as a small learning task to adapt generative model per data, we hence call it gradient",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                396,
+                                505,
+                                409
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        396,
+                                        505,
+                                        409
+                                    ],
+                                    "score": 1.0,
+                                    "content": "inversions to meta-learn (GIML). The detailed procedure of GIML is presented in Appendix A. We",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                407,
+                                505,
+                                419
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        407,
+                                        262,
+                                        419
+                                    ],
+                                    "score": 1.0,
+                                    "content": "start with an arbitrary initialization of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        262,
+                                        409,
+                                        271,
+                                        417
+                                    ],
+                                    "score": 0.79,
+                                    "content": "w",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        271,
+                                        407,
+                                        395,
+                                        419
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", and iteratively update toward",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        395,
+                                        407,
+                                        406,
+                                        417
+                                    ],
+                                    "score": 0.87,
+                                    "content": "w ^ { \\prime }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        407,
+                                        407,
+                                        505,
+                                        419
+                                    ],
+                                    "score": 1.0,
+                                    "content": "from a variant of GIAS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                418,
+                                505,
+                                430
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        418,
+                                        121,
+                                        430
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        122,
+                                        419,
+                                        132,
+                                        428
+                                    ],
+                                    "score": 0.81,
+                                    "content": "N",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        132,
+                                        418,
+                                        234,
+                                        430
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tasks sub-sampled from",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        234,
+                                        419,
+                                        242,
+                                        428
+                                    ],
+                                    "score": 0.77,
+                                    "content": "s",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        242,
+                                        418,
+                                        505,
+                                        430
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", which is different than multiple applications of GIAS for each",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                429,
+                                505,
+                                442
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        429,
+                                        191,
+                                        442
+                                    ],
+                                    "score": 1.0,
+                                    "content": "task in two folds: (i)",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        191,
+                                        429,
+                                        201,
+                                        441
+                                    ],
+                                    "score": 0.87,
+                                    "content": "\\ell _ { 2 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        201,
+                                        429,
+                                        505,
+                                        442
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-regularization in latent space search; and (ii) an integrated optimization on",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                438,
+                                507,
+                                456
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        438,
+                                        348,
+                                        456
+                                    ],
+                                    "score": 1.0,
+                                    "content": "model parameter. The variant first finds optimal latent codes",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        348,
+                                        440,
+                                        424,
+                                        452
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\boldsymbol { z } _ { i } ^ { * } = ( z _ { i 1 } ^ { * } , . . . , z _ { i B } ^ { * } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        425,
+                                        438,
+                                        479,
+                                        456
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for each task",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        479,
+                                        441,
+                                        484,
+                                        450
+                                    ],
+                                    "score": 0.71,
+                                    "content": "i",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        484,
+                                        438,
+                                        507,
+                                        456
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                451,
+                                506,
+                                464
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        451,
+                                        330,
+                                        464
+                                    ],
+                                    "score": 1.0,
+                                    "content": "respect to the same cost function of GIAS but additional",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        331,
+                                        451,
+                                        340,
+                                        462
+                                    ],
+                                    "score": 0.87,
+                                    "content": "\\ell _ { 2 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        341,
+                                        451,
+                                        506,
+                                        464
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-regularization. Note that the latent space",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                462,
+                                506,
+                                475
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        462,
+                                        354,
+                                        475
+                                    ],
+                                    "score": 1.0,
+                                    "content": "search with untrained generative model easily diverges. The",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        354,
+                                        462,
+                                        364,
+                                        473
+                                    ],
+                                    "score": 0.88,
+                                    "content": "\\ell _ { 2 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        365,
+                                        462,
+                                        506,
+                                        475
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-regularization is added to prevent",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                473,
+                                505,
+                                485
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        473,
+                                        178,
+                                        485
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the divergence of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        178,
+                                        473,
+                                        190,
+                                        484
+                                    ],
+                                    "score": 0.9,
+                                    "content": "z _ { i } ^ { * }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        190,
+                                        473,
+                                        269,
+                                        485
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Once we obtained",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        270,
+                                        473,
+                                        281,
+                                        484
+                                    ],
+                                    "score": 0.84,
+                                    "content": "z _ { i } ^ { * }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        281,
+                                        473,
+                                        291,
+                                        485
+                                    ],
+                                    "score": 1.0,
+                                    "content": "’s,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        292,
+                                        473,
+                                        303,
+                                        483
+                                    ],
+                                    "score": 0.84,
+                                    "content": "w ^ { \\prime }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        303,
+                                        473,
+                                        505,
+                                        485
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is computed by few steps of gradient descents for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                482,
+                                506,
+                                499
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        482,
+                                        280,
+                                        499
+                                    ],
+                                    "score": 1.0,
+                                    "content": "an integrated parameter search to minimize",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        281,
+                                        483,
+                                        427,
+                                        496
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\begin{array} { r } { \\sum _ { i } c ( G _ { w ^ { \\prime } } ( z _ { i 1 } ^ { * } ) , . . . , \\bar { G } _ { w ^ { \\prime } } ( z _ { i B } ^ { * } ) ; \\theta _ { i } , g _ { i } ) } \\end{array}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        428,
+                                        482,
+                                        506,
+                                        499
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". This is because in",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                494,
+                                506,
+                                507
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        494,
+                                        243,
+                                        507
+                                    ],
+                                    "score": 1.0,
+                                    "content": "GIML, we want meta information",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        244,
+                                        497,
+                                        252,
+                                        505
+                                    ],
+                                    "score": 0.75,
+                                    "content": "w",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        253,
+                                        494,
+                                        506,
+                                        507
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to help GIAS for each task rather than solving individual tasks,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                505,
+                                506,
+                                519
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        505,
+                                        254,
+                                        519
+                                    ],
+                                    "score": 1.0,
+                                    "content": "while after performing GIML to train",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        254,
+                                        507,
+                                        263,
+                                        515
+                                    ],
+                                    "score": 0.74,
+                                    "content": "w",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        263,
+                                        505,
+                                        472,
+                                        519
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", we perform GIAS to invert gradient with the trained",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        472,
+                                        507,
+                                        481,
+                                        515
+                                    ],
+                                    "score": 0.74,
+                                    "content": "w",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        481,
+                                        505,
+                                        506,
+                                        519
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". This",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                517,
+                                246,
+                                528
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        517,
+                                        246,
+                                        528
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is analogous to the Reptile in [20].",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        }
+                    ],
+                    "index": 15
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        549,
+                        191,
+                        563
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                103,
+                                547,
+                                193,
+                                566
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        103,
+                                        547,
+                                        193,
+                                        566
+                                    ],
+                                    "score": 1.0,
+                                    "content": "5 Experiments",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        }
+                    ],
+                    "index": 23
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        577,
+                        505,
+                        655
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                577,
+                                506,
+                                591
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        577,
+                                        506,
+                                        591
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Setup. Unless stated otherwise, we consider the image classification task on the validation set",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                588,
+                                505,
+                                601
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        588,
+                                        286,
+                                        601
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of ImageNet [22] dataset scaled down to",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        286,
+                                        589,
+                                        323,
+                                        600
+                                    ],
+                                    "score": 0.9,
+                                    "content": "6 4 \\times 6 4",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        323,
+                                        588,
+                                        505,
+                                        601
+                                    ],
+                                    "score": 1.0,
+                                    "content": "pixels (for computational tractability) and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                599,
+                                506,
+                                612
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        599,
+                                        506,
+                                        612
+                                    ],
+                                    "score": 1.0,
+                                    "content": "use a randomly initialized ResNet18 [10] for training. For deep generative models in GIAS,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                611,
+                                506,
+                                622
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        611,
+                                        421,
+                                        622
+                                    ],
+                                    "score": 1.0,
+                                    "content": "we use StyleGAN2 [13] trained on ImageNet. We use a batch size of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        422,
+                                        611,
+                                        458,
+                                        621
+                                    ],
+                                    "score": 0.9,
+                                    "content": "B \\ = \\ 4",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        459,
+                                        611,
+                                        506,
+                                        622
+                                    ],
+                                    "score": 1.0,
+                                    "content": "as default",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                621,
+                                505,
+                                634
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        622,
+                                        386,
+                                        634
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and use the negative cosine to measure the gradient dissimilarity",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        386,
+                                        621,
+                                        410,
+                                        633
+                                    ],
+                                    "score": 0.91,
+                                    "content": "d ( \\cdot , \\cdot )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        410,
+                                        622,
+                                        505,
+                                        634
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". We present detailed",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                633,
+                                505,
+                                645
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        633,
+                                        505,
+                                        645
+                                    ],
+                                    "score": 1.0,
+                                    "content": "setup in Appendix H. Our experiment code is available at https://github.com/ml-postech/",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                644,
+                                329,
+                                657
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        644,
+                                        329,
+                                        657
+                                    ],
+                                    "score": 1.0,
+                                    "content": "gradient-inversion-generative-image-prior.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        }
+                    ],
+                    "index": 27
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        667,
+                        505,
+                        722
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                667,
+                                506,
+                                680
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        667,
+                                        506,
+                                        680
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Algorithms. We evaluate several algorithms for the gradient inversion (GI) task in (3). They differ",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                677,
+                                506,
+                                690
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        677,
+                                        356,
+                                        690
+                                    ],
+                                    "score": 1.0,
+                                    "content": "mainly in which spaces each algorithm searches over: the input",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        357,
+                                        680,
+                                        363,
+                                        688
+                                    ],
+                                    "score": 0.76,
+                                    "content": "x",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        363,
+                                        677,
+                                        426,
+                                        690
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", the latent code",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        426,
+                                        680,
+                                        433,
+                                        688
+                                    ],
+                                    "score": 0.77,
+                                    "content": "z",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        433,
+                                        677,
+                                        506,
+                                        690
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", and/or the model",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                688,
+                                507,
+                                702
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        688,
+                                        150,
+                                        702
+                                    ],
+                                    "score": 1.0,
+                                    "content": "parameter",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        150,
+                                        691,
+                                        159,
+                                        699
+                                    ],
+                                    "score": 0.72,
+                                    "content": "w",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        159,
+                                        688,
+                                        507,
+                                        702
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Each algorithm is denoted by GI-(·), where the suffix indicates the search space(s).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                699,
+                                505,
+                                713
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        699,
+                                        179,
+                                        713
+                                    ],
+                                    "score": 1.0,
+                                    "content": "For instances, GI-",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        179,
+                                        700,
+                                        198,
+                                        712
+                                    ],
+                                    "score": 0.74,
+                                    "content": "z / w",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        198,
+                                        699,
+                                        404,
+                                        713
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is identical to the proposed method, GIAS, and GI-",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        405,
+                                        702,
+                                        411,
+                                        710
+                                    ],
+                                    "score": 0.77,
+                                    "content": "x",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        411,
+                                        699,
+                                        505,
+                                        713
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is the one proposed by",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                711,
+                                179,
+                                724
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        711,
+                                        179,
+                                        724
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Geiping et al. [9].",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        }
+                    ],
+                    "index": 33
+                }
+            ],
+            "page_idx": 5,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        302,
+                        742,
+                        308,
+                        750
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                302,
+                                741,
+                                309,
+                                752
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        302,
+                                        741,
+                                        309,
+                                        752
+                                    ],
+                                    "score": 1.0,
+                                    "content": "6",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "image",
+                    "bbox": [
+                        103,
+                        92,
+                        500,
+                        263
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                103,
+                                92,
+                                500,
+                                263
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        103,
+                                        92,
+                                        500,
+                                        263
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                103,
+                                                92,
+                                                500,
+                                                263
+                                            ],
+                                            "score": 0.975,
+                                            "type": "image",
+                                            "image_path": "3e98af5f62fd12872d10b40932ee68ec2b13001baf19a78e4df336bdfc8cd337.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 1,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        103,
+                                        92,
+                                        500,
+                                        149.0
+                                    ],
+                                    "spans": [],
+                                    "index": 0
+                                },
+                                {
+                                    "bbox": [
+                                        103,
+                                        149.0,
+                                        500,
+                                        206.0
+                                    ],
+                                    "spans": [],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        103,
+                                        206.0,
+                                        500,
+                                        263.0
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                106,
+                                270,
+                                505,
+                                314
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        270,
+                                        506,
+                                        282
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                270,
+                                                506,
+                                                282
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 3: Comparison of different searching spaces. (a) Each row shows reconstructed images of",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 3
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        280,
+                                        506,
+                                        293
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                280,
+                                                506,
+                                                293
+                                            ],
+                                            "score": 1.0,
+                                            "content": "different optimization domains. The first three rows share the same latent space search of 1, 500",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 4
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        291,
+                                        505,
+                                        304
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                292,
+                                                167,
+                                                304
+                                            ],
+                                            "score": 1.0,
+                                            "content": "iterations, and",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                167,
+                                                291,
+                                                200,
+                                                303
+                                            ],
+                                            "score": 0.86,
+                                            "content": "{ \\mathrm { G I } } { - z } / w",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                201,
+                                                292,
+                                                505,
+                                                304
+                                            ],
+                                            "score": 1.0,
+                                            "content": "is verified to be the best option to fully exploits the knowledge inside the",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 5
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        303,
+                                        450,
+                                        316
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                303,
+                                                450,
+                                                316
+                                            ],
+                                            "score": 1.0,
+                                            "content": "generative model. (b) Cost function over iterations of different optimization domains.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 6
+                                }
+                            ],
+                            "index": 4.5
+                        }
+                    ],
+                    "index": 2.75
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        341,
+                        345,
+                        352
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                339,
+                                346,
+                                354
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        339,
+                                        346,
+                                        354
+                                    ],
+                                    "score": 1.0,
+                                    "content": "4.2 Gradient inversion to meta-learn generative model",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        }
+                    ],
+                    "index": 7
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        363,
+                        505,
+                        528
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                363,
+                                506,
+                                376
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        363,
+                                        506,
+                                        376
+                                    ],
+                                    "score": 1.0,
+                                    "content": "For the case that pretrained generative model is unavailable, we devise an algorithm to train a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                374,
+                                506,
+                                387
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        374,
+                                        178,
+                                        387
+                                    ],
+                                    "score": 1.0,
+                                    "content": "generative model",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        178,
+                                        375,
+                                        193,
+                                        385
+                                    ],
+                                    "score": 0.89,
+                                    "content": "G _ { w }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        194,
+                                        374,
+                                        230,
+                                        387
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for a set",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        230,
+                                        374,
+                                        289,
+                                        387
+                                    ],
+                                    "score": 0.94,
+                                    "content": "\\mathcal { S } = \\{ ( \\theta _ { i } , g _ { i } ) \\}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        289,
+                                        374,
+                                        506,
+                                        387
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of gradient inversion tasks. Since each inversion task",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                385,
+                                506,
+                                398
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        385,
+                                        506,
+                                        398
+                                    ],
+                                    "score": 1.0,
+                                    "content": "can be considered as a small learning task to adapt generative model per data, we hence call it gradient",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                396,
+                                505,
+                                409
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        396,
+                                        505,
+                                        409
+                                    ],
+                                    "score": 1.0,
+                                    "content": "inversions to meta-learn (GIML). The detailed procedure of GIML is presented in Appendix A. We",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                407,
+                                505,
+                                419
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        407,
+                                        262,
+                                        419
+                                    ],
+                                    "score": 1.0,
+                                    "content": "start with an arbitrary initialization of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        262,
+                                        409,
+                                        271,
+                                        417
+                                    ],
+                                    "score": 0.79,
+                                    "content": "w",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        271,
+                                        407,
+                                        395,
+                                        419
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", and iteratively update toward",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        395,
+                                        407,
+                                        406,
+                                        417
+                                    ],
+                                    "score": 0.87,
+                                    "content": "w ^ { \\prime }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        407,
+                                        407,
+                                        505,
+                                        419
+                                    ],
+                                    "score": 1.0,
+                                    "content": "from a variant of GIAS",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                418,
+                                505,
+                                430
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        418,
+                                        121,
+                                        430
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        122,
+                                        419,
+                                        132,
+                                        428
+                                    ],
+                                    "score": 0.81,
+                                    "content": "N",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        132,
+                                        418,
+                                        234,
+                                        430
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tasks sub-sampled from",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        234,
+                                        419,
+                                        242,
+                                        428
+                                    ],
+                                    "score": 0.77,
+                                    "content": "s",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        242,
+                                        418,
+                                        505,
+                                        430
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", which is different than multiple applications of GIAS for each",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                429,
+                                505,
+                                442
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        429,
+                                        191,
+                                        442
+                                    ],
+                                    "score": 1.0,
+                                    "content": "task in two folds: (i)",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        191,
+                                        429,
+                                        201,
+                                        441
+                                    ],
+                                    "score": 0.87,
+                                    "content": "\\ell _ { 2 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        201,
+                                        429,
+                                        505,
+                                        442
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-regularization in latent space search; and (ii) an integrated optimization on",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                438,
+                                507,
+                                456
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        438,
+                                        348,
+                                        456
+                                    ],
+                                    "score": 1.0,
+                                    "content": "model parameter. The variant first finds optimal latent codes",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        348,
+                                        440,
+                                        424,
+                                        452
+                                    ],
+                                    "score": 0.92,
+                                    "content": "\\boldsymbol { z } _ { i } ^ { * } = ( z _ { i 1 } ^ { * } , . . . , z _ { i B } ^ { * } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        425,
+                                        438,
+                                        479,
+                                        456
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for each task",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        479,
+                                        441,
+                                        484,
+                                        450
+                                    ],
+                                    "score": 0.71,
+                                    "content": "i",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        484,
+                                        438,
+                                        507,
+                                        456
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                451,
+                                506,
+                                464
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        451,
+                                        330,
+                                        464
+                                    ],
+                                    "score": 1.0,
+                                    "content": "respect to the same cost function of GIAS but additional",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        331,
+                                        451,
+                                        340,
+                                        462
+                                    ],
+                                    "score": 0.87,
+                                    "content": "\\ell _ { 2 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        341,
+                                        451,
+                                        506,
+                                        464
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-regularization. Note that the latent space",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                462,
+                                506,
+                                475
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        462,
+                                        354,
+                                        475
+                                    ],
+                                    "score": 1.0,
+                                    "content": "search with untrained generative model easily diverges. The",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        354,
+                                        462,
+                                        364,
+                                        473
+                                    ],
+                                    "score": 0.88,
+                                    "content": "\\ell _ { 2 }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        365,
+                                        462,
+                                        506,
+                                        475
+                                    ],
+                                    "score": 1.0,
+                                    "content": "-regularization is added to prevent",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                473,
+                                505,
+                                485
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        473,
+                                        178,
+                                        485
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the divergence of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        178,
+                                        473,
+                                        190,
+                                        484
+                                    ],
+                                    "score": 0.9,
+                                    "content": "z _ { i } ^ { * }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        190,
+                                        473,
+                                        269,
+                                        485
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Once we obtained",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        270,
+                                        473,
+                                        281,
+                                        484
+                                    ],
+                                    "score": 0.84,
+                                    "content": "z _ { i } ^ { * }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        281,
+                                        473,
+                                        291,
+                                        485
+                                    ],
+                                    "score": 1.0,
+                                    "content": "’s,",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        292,
+                                        473,
+                                        303,
+                                        483
+                                    ],
+                                    "score": 0.84,
+                                    "content": "w ^ { \\prime }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        303,
+                                        473,
+                                        505,
+                                        485
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is computed by few steps of gradient descents for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                482,
+                                506,
+                                499
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        482,
+                                        280,
+                                        499
+                                    ],
+                                    "score": 1.0,
+                                    "content": "an integrated parameter search to minimize",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        281,
+                                        483,
+                                        427,
+                                        496
+                                    ],
+                                    "score": 0.91,
+                                    "content": "\\begin{array} { r } { \\sum _ { i } c ( G _ { w ^ { \\prime } } ( z _ { i 1 } ^ { * } ) , . . . , \\bar { G } _ { w ^ { \\prime } } ( z _ { i B } ^ { * } ) ; \\theta _ { i } , g _ { i } ) } \\end{array}",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        428,
+                                        482,
+                                        506,
+                                        499
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". This is because in",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                494,
+                                506,
+                                507
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        494,
+                                        243,
+                                        507
+                                    ],
+                                    "score": 1.0,
+                                    "content": "GIML, we want meta information",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        244,
+                                        497,
+                                        252,
+                                        505
+                                    ],
+                                    "score": 0.75,
+                                    "content": "w",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        253,
+                                        494,
+                                        506,
+                                        507
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to help GIAS for each task rather than solving individual tasks,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                505,
+                                506,
+                                519
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        505,
+                                        254,
+                                        519
+                                    ],
+                                    "score": 1.0,
+                                    "content": "while after performing GIML to train",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        254,
+                                        507,
+                                        263,
+                                        515
+                                    ],
+                                    "score": 0.74,
+                                    "content": "w",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        263,
+                                        505,
+                                        472,
+                                        519
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", we perform GIAS to invert gradient with the trained",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        472,
+                                        507,
+                                        481,
+                                        515
+                                    ],
+                                    "score": 0.74,
+                                    "content": "w",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        481,
+                                        505,
+                                        506,
+                                        519
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". This",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                517,
+                                246,
+                                528
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        517,
+                                        246,
+                                        528
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is analogous to the Reptile in [20].",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        }
+                    ],
+                    "index": 15,
+                    "bbox_fs": [
+                        104,
+                        363,
+                        507,
+                        528
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        549,
+                        191,
+                        563
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                103,
+                                547,
+                                193,
+                                566
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        103,
+                                        547,
+                                        193,
+                                        566
+                                    ],
+                                    "score": 1.0,
+                                    "content": "5 Experiments",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        }
+                    ],
+                    "index": 23
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        577,
+                        505,
+                        655
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                577,
+                                506,
+                                591
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        577,
+                                        506,
+                                        591
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Setup. Unless stated otherwise, we consider the image classification task on the validation set",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                588,
+                                505,
+                                601
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        588,
+                                        286,
+                                        601
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of ImageNet [22] dataset scaled down to",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        286,
+                                        589,
+                                        323,
+                                        600
+                                    ],
+                                    "score": 0.9,
+                                    "content": "6 4 \\times 6 4",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        323,
+                                        588,
+                                        505,
+                                        601
+                                    ],
+                                    "score": 1.0,
+                                    "content": "pixels (for computational tractability) and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                599,
+                                506,
+                                612
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        599,
+                                        506,
+                                        612
+                                    ],
+                                    "score": 1.0,
+                                    "content": "use a randomly initialized ResNet18 [10] for training. For deep generative models in GIAS,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                611,
+                                506,
+                                622
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        611,
+                                        421,
+                                        622
+                                    ],
+                                    "score": 1.0,
+                                    "content": "we use StyleGAN2 [13] trained on ImageNet. We use a batch size of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        422,
+                                        611,
+                                        458,
+                                        621
+                                    ],
+                                    "score": 0.9,
+                                    "content": "B \\ = \\ 4",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        459,
+                                        611,
+                                        506,
+                                        622
+                                    ],
+                                    "score": 1.0,
+                                    "content": "as default",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                621,
+                                505,
+                                634
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        622,
+                                        386,
+                                        634
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and use the negative cosine to measure the gradient dissimilarity",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        386,
+                                        621,
+                                        410,
+                                        633
+                                    ],
+                                    "score": 0.91,
+                                    "content": "d ( \\cdot , \\cdot )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        410,
+                                        622,
+                                        505,
+                                        634
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". We present detailed",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                633,
+                                505,
+                                645
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        633,
+                                        505,
+                                        645
+                                    ],
+                                    "score": 1.0,
+                                    "content": "setup in Appendix H. Our experiment code is available at https://github.com/ml-postech/",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                644,
+                                329,
+                                657
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        644,
+                                        329,
+                                        657
+                                    ],
+                                    "score": 1.0,
+                                    "content": "gradient-inversion-generative-image-prior.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        }
+                    ],
+                    "index": 27,
+                    "bbox_fs": [
+                        105,
+                        577,
+                        506,
+                        657
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        667,
+                        505,
+                        722
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                667,
+                                506,
+                                680
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        667,
+                                        506,
+                                        680
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Algorithms. We evaluate several algorithms for the gradient inversion (GI) task in (3). They differ",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                677,
+                                506,
+                                690
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        677,
+                                        356,
+                                        690
+                                    ],
+                                    "score": 1.0,
+                                    "content": "mainly in which spaces each algorithm searches over: the input",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        357,
+                                        680,
+                                        363,
+                                        688
+                                    ],
+                                    "score": 0.76,
+                                    "content": "x",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        363,
+                                        677,
+                                        426,
+                                        690
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", the latent code",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        426,
+                                        680,
+                                        433,
+                                        688
+                                    ],
+                                    "score": 0.77,
+                                    "content": "z",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        433,
+                                        677,
+                                        506,
+                                        690
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", and/or the model",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                688,
+                                507,
+                                702
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        688,
+                                        150,
+                                        702
+                                    ],
+                                    "score": 1.0,
+                                    "content": "parameter",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        150,
+                                        691,
+                                        159,
+                                        699
+                                    ],
+                                    "score": 0.72,
+                                    "content": "w",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        159,
+                                        688,
+                                        507,
+                                        702
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". Each algorithm is denoted by GI-(·), where the suffix indicates the search space(s).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                699,
+                                505,
+                                713
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        699,
+                                        179,
+                                        713
+                                    ],
+                                    "score": 1.0,
+                                    "content": "For instances, GI-",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        179,
+                                        700,
+                                        198,
+                                        712
+                                    ],
+                                    "score": 0.74,
+                                    "content": "z / w",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        198,
+                                        699,
+                                        404,
+                                        713
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is identical to the proposed method, GIAS, and GI-",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        405,
+                                        702,
+                                        411,
+                                        710
+                                    ],
+                                    "score": 0.77,
+                                    "content": "x",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        411,
+                                        699,
+                                        505,
+                                        713
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is the one proposed by",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                711,
+                                179,
+                                724
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        711,
+                                        179,
+                                        724
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Geiping et al. [9].",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        }
+                    ],
+                    "index": 33,
+                    "bbox_fs": [
+                        105,
+                        667,
+                        507,
+                        724
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "image",
+                    "bbox": [
+                        100,
+                        73,
+                        499,
+                        223
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                100,
+                                73,
+                                499,
+                                223
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        100,
+                                        73,
+                                        499,
+                                        223
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                100,
+                                                73,
+                                                499,
+                                                223
+                                            ],
+                                            "score": 0.974,
+                                            "type": "image",
+                                            "image_path": "05927611c6cf240c845850b9daf59ed7a0462bf6d541e6a943eb9a3ea192a47a.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 1,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        100,
+                                        73,
+                                        499,
+                                        123.0
+                                    ],
+                                    "spans": [],
+                                    "index": 0
+                                },
+                                {
+                                    "bbox": [
+                                        100,
+                                        123.0,
+                                        499,
+                                        173.0
+                                    ],
+                                    "spans": [],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        100,
+                                        173.0,
+                                        499,
+                                        223.0
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                106,
+                                230,
+                                505,
+                                285
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        229,
+                                        506,
+                                        243
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                229,
+                                                409,
+                                                243
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 4: Comparison of state-of-the-art models and ours. Replacing GI-",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                410,
+                                                232,
+                                                416,
+                                                240
+                                            ],
+                                            "score": 0.7,
+                                            "content": "x",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                417,
+                                                229,
+                                                453,
+                                                243
+                                            ],
+                                            "score": 1.0,
+                                            "content": "with GI-",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                453,
+                                                230,
+                                                471,
+                                                242
+                                            ],
+                                            "score": 0.62,
+                                            "content": "z / w",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                472,
+                                                229,
+                                                506,
+                                                243
+                                            ],
+                                            "score": 1.0,
+                                            "content": "(GIAS)",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 3
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        241,
+                                        506,
+                                        254
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                241,
+                                                506,
+                                                254
+                                            ],
+                                            "score": 1.0,
+                                            "content": "regardless of using BN [29] or not [9] provides substantial improvement in the reconstruction accuracy.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 4
+                                },
+                                {
+                                    "bbox": [
+                                        104,
+                                        251,
+                                        506,
+                                        265
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                104,
+                                                251,
+                                                506,
+                                                265
+                                            ],
+                                            "score": 1.0,
+                                            "content": "(a) Average PSNR and best PSNR in a batch throughout the experiments. (b) An ablation study and",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 5
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        262,
+                                        505,
+                                        276
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                262,
+                                                505,
+                                                276
+                                            ],
+                                            "score": 1.0,
+                                            "content": "comparison of reconstruction results with our models and state-of-the-art models. We highlight the",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 6
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        274,
+                                        210,
+                                        286
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                274,
+                                                210,
+                                                286
+                                            ],
+                                            "score": 1.0,
+                                            "content": "proposed models in bold.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 7
+                                }
+                            ],
+                            "index": 5
+                        }
+                    ],
+                    "index": 3.0
+                },
+                {
+                    "type": "table",
+                    "bbox": [
+                        108,
+                        336,
+                        503,
+                        384
+                    ],
+                    "blocks": [
+                        {
+                            "type": "table_caption",
+                            "bbox": [
+                                106,
+                                296,
+                                505,
+                                330
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        295,
+                                        505,
+                                        309
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                104,
+                                                295,
+                                                505,
+                                                309
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Table 1: Comparison of our methods with state-of-the-art methods. Adding our method makes",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 8
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        307,
+                                        505,
+                                        320
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                307,
+                                                505,
+                                                320
+                                            ],
+                                            "score": 1.0,
+                                            "content": "performance improvement versus two baseline methods. PSNR, SSIM, and LPIPS[30] are used to",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 9
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        318,
+                                        410,
+                                        331
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                318,
+                                                410,
+                                                331
+                                            ],
+                                            "score": 1.0,
+                                            "content": "evaluate reconstruction results. We highlight the best performances in bold.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 10
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "type": "table_body",
+                            "bbox": [
+                                108,
+                                336,
+                                503,
+                                384
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        108,
+                                        336,
+                                        503,
+                                        384
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                108,
+                                                336,
+                                                503,
+                                                384
+                                            ],
+                                            "score": 0.977,
+                                            "html": "<table><tr><td>Method</td><td>GI-x [9]</td><td>GI-z (ours)</td><td>GI-w (ours)</td><td>GI-z/w (GIAS,ours)</td><td>GI-x+BN[29]</td><td>GI-z/w+BN (ours)</td></tr><tr><td>PSNR↑</td><td>13.78</td><td>14.27</td><td>14.70</td><td>15.58</td><td>15.52</td><td>16.31</td></tr><tr><td>SSIM↑</td><td>0.2542</td><td>0.3106</td><td>0.3519</td><td>0.3895</td><td>0.3513</td><td>0.4311</td></tr><tr><td>LPIPS↓</td><td>0.4376</td><td>0.3233</td><td>0.5121</td><td>0.3023</td><td>0.3645</td><td>0.2861</td></tr></table>",
+                                            "type": "table",
+                                            "image_path": "d2238a9c5fd3fdd85e97704e60c544b53a261d8a1d75ae6af9d39b80635553ea.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 12,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        108,
+                                        336,
+                                        503,
+                                        352.0
+                                    ],
+                                    "spans": [],
+                                    "index": 11
+                                },
+                                {
+                                    "bbox": [
+                                        108,
+                                        352.0,
+                                        503,
+                                        368.0
+                                    ],
+                                    "spans": [],
+                                    "index": 12
+                                },
+                                {
+                                    "bbox": [
+                                        108,
+                                        368.0,
+                                        503,
+                                        384.0
+                                    ],
+                                    "spans": [],
+                                    "index": 13
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 10.5
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        395,
+                        250,
+                        407
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                393,
+                                251,
+                                410
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        393,
+                                        251,
+                                        410
+                                    ],
+                                    "score": 1.0,
+                                    "content": "5.1 Justification of GIAS design",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        }
+                    ],
+                    "index": 14
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        415,
+                        505,
+                        590
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                416,
+                                506,
+                                428
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        416,
+                                        506,
+                                        428
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We first provide an empirical justification of the specific order of searching spaces in GIAS (corre-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                426,
+                                506,
+                                439
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        426,
+                                        156,
+                                        439
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sponding to",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        156,
+                                        426,
+                                        190,
+                                        438
+                                    ],
+                                    "score": 0.79,
+                                    "content": "\\mathbf { G I } { - } z / w ,",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        190,
+                                        426,
+                                        506,
+                                        439
+                                    ],
+                                    "score": 1.0,
+                                    "content": ") to fully utilize a pretrained generative model. To do so, we provide Figure 4b",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                437,
+                                505,
+                                450
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        437,
+                                        344,
+                                        450
+                                    ],
+                                    "score": 1.0,
+                                    "content": "comparing algorithms with different searching spaces: GI-",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        344,
+                                        437,
+                                        362,
+                                        449
+                                    ],
+                                    "score": 0.86,
+                                    "content": "z / w",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        362,
+                                        437,
+                                        380,
+                                        450
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", GI-",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        381,
+                                        437,
+                                        397,
+                                        449
+                                    ],
+                                    "score": 0.65,
+                                    "content": "z / x",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        397,
+                                        437,
+                                        400,
+                                        450
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        401,
+                                        437,
+                                        421,
+                                        448
+                                    ],
+                                    "score": 0.55,
+                                    "content": "\\mathrm { G I } { - } z",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        421,
+                                        437,
+                                        442,
+                                        450
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        442,
+                                        437,
+                                        463,
+                                        448
+                                    ],
+                                    "score": 0.36,
+                                    "content": "_ { \\mathrm { G I - } x }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        464,
+                                        437,
+                                        505,
+                                        450
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", of which",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                448,
+                                506,
+                                461
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        448,
+                                        325,
+                                        461
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the first three share the same latent space search over",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        325,
+                                        451,
+                                        332,
+                                        458
+                                    ],
+                                    "score": 0.76,
+                                    "content": "z",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        332,
+                                        448,
+                                        506,
+                                        461
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for the first 1, 500 iterations. As shown in",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                460,
+                                506,
+                                471
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        460,
+                                        270,
+                                        471
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Figure 3(a), the latent space search over",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        270,
+                                        461,
+                                        277,
+                                        469
+                                    ],
+                                    "score": 0.73,
+                                    "content": "z",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        277,
+                                        460,
+                                        506,
+                                        471
+                                    ],
+                                    "score": 1.0,
+                                    "content": "quickly finds plausible image in a much shorter number",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                469,
+                                506,
+                                483
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        469,
+                                        193,
+                                        483
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of iterations than GI-",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        194,
+                                        471,
+                                        200,
+                                        480
+                                    ],
+                                    "score": 0.71,
+                                    "content": "x",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        200,
+                                        469,
+                                        506,
+                                        483
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", while it does not improve after a certain point due to the imperfection of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                481,
+                                506,
+                                493
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        481,
+                                        302,
+                                        493
+                                    ],
+                                    "score": 1.0,
+                                    "content": "pretrained generative model. Such a limitation of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        302,
+                                        481,
+                                        322,
+                                        492
+                                    ],
+                                    "score": 0.27,
+                                    "content": "\\mathrm { G I } { - } z",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        322,
+                                        481,
+                                        506,
+                                        493
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is also captured in Figure 3(b), where the cost",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                491,
+                                506,
+                                504
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        491,
+                                        167,
+                                        504
+                                    ],
+                                    "score": 1.0,
+                                    "content": "function of GI-",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        167,
+                                        493,
+                                        173,
+                                        502
+                                    ],
+                                    "score": 0.7,
+                                    "content": "z",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        174,
+                                        491,
+                                        506,
+                                        504
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is not decreasing after a certain number of optimization steps. To further minimize",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                502,
+                                505,
+                                515
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        503,
+                                        268,
+                                        515
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the cost function, one alternative to GI-",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        268,
+                                        502,
+                                        287,
+                                        514
+                                    ],
+                                    "score": 0.67,
+                                    "content": "z / w",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        287,
+                                        503,
+                                        331,
+                                        515
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(GIAS) is",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        331,
+                                        503,
+                                        362,
+                                        514
+                                    ],
+                                    "score": 0.8,
+                                    "content": "{ \\mathrm { G I } } { - } z / x",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        362,
+                                        503,
+                                        505,
+                                        515
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", which can further reduce the loss",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                513,
+                                506,
+                                526
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        513,
+                                        289,
+                                        526
+                                    ],
+                                    "score": 1.0,
+                                    "content": "function whereas the parameter search in GI-",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        290,
+                                        514,
+                                        308,
+                                        525
+                                    ],
+                                    "score": 0.71,
+                                    "content": "z / w",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        308,
+                                        513,
+                                        506,
+                                        526
+                                    ],
+                                    "score": 1.0,
+                                    "content": "seems to provide more natural reconstruction of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                524,
+                                506,
+                                538
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        524,
+                                        168,
+                                        538
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the image than",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        168,
+                                        524,
+                                        199,
+                                        537
+                                    ],
+                                    "score": 0.76,
+                                    "content": "{ \\mathrm { G I } } { - } z / { \\bar { x } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        199,
+                                        524,
+                                        278,
+                                        538
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". The superiority of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        278,
+                                        525,
+                                        311,
+                                        537
+                                    ],
+                                    "score": 0.84,
+                                    "content": "{ \\mathrm { G I } } { - z } / w",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        311,
+                                        524,
+                                        332,
+                                        538
+                                    ],
+                                    "score": 1.0,
+                                    "content": "over",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        332,
+                                        525,
+                                        363,
+                                        537
+                                    ],
+                                    "score": 0.73,
+                                    "content": "{ \\mathrm { G I } } { - } z / x",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        363,
+                                        524,
+                                        506,
+                                        538
+                                    ],
+                                    "score": 1.0,
+                                    "content": "may come from that the parameter",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                535,
+                                506,
+                                549
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        535,
+                                        506,
+                                        549
+                                    ],
+                                    "score": 1.0,
+                                    "content": "space search exploits an implicit bias from optimizing a good architecture for expressing images, c.f.,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                546,
+                                506,
+                                559
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        546,
+                                        506,
+                                        559
+                                    ],
+                                    "score": 1.0,
+                                    "content": "deep image prior [26]. In Appendix E and Figure 1, we also present the same comparison on FFHQ",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                558,
+                                505,
+                                569
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        558,
+                                        505,
+                                        569
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(human-face images) [12] where diversity is much smaller than that of ImageNet. On such a less",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                567,
+                                505,
+                                581
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        567,
+                                        505,
+                                        581
+                                    ],
+                                    "score": 1.0,
+                                    "content": "diverse dataset, the distribution can be easily learned, and the gain from training a generative model",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                578,
+                                144,
+                                593
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        578,
+                                        144,
+                                        593
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is larger.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        }
+                    ],
+                    "index": 22.5
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        603,
+                        381,
+                        615
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                602,
+                                382,
+                                617
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        602,
+                                        382,
+                                        617
+                                    ],
+                                    "score": 1.0,
+                                    "content": "5.2 The gain from fully exploiting pretrained generative model",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        }
+                    ],
+                    "index": 31
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        623,
+                        505,
+                        722
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                623,
+                                505,
+                                636
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        623,
+                                        505,
+                                        636
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Comparison with state-of-the-art models. Our method can be easily added to previous methods",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                634,
+                                505,
+                                647
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        634,
+                                        505,
+                                        647
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[9, 29]. In Table 1 and Figure 4, we compare the state-of-the-art methods both with and without the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                645,
+                                506,
+                                658
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        645,
+                                        339,
+                                        658
+                                    ],
+                                    "score": 1.0,
+                                    "content": "proposed generative modelling. In Table 1, comparing GI-",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        339,
+                                        646,
+                                        346,
+                                        655
+                                    ],
+                                    "score": 0.75,
+                                    "content": "x",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        346,
+                                        645,
+                                        357,
+                                        658
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        357,
+                                        645,
+                                        390,
+                                        657
+                                    ],
+                                    "score": 0.67,
+                                    "content": "{ \\mathrm { G I } } { - } z / w",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        390,
+                                        645,
+                                        422,
+                                        658
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and GI-",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        422,
+                                        645,
+                                        453,
+                                        656
+                                    ],
+                                    "score": 0.81,
+                                    "content": "x + { \\tt B N }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        454,
+                                        645,
+                                        464,
+                                        658
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        465,
+                                        645,
+                                        506,
+                                        657
+                                    ],
+                                    "score": 0.76,
+                                    "content": "{ \\mathrm { G I } } { - } z / w +",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                654,
+                                507,
+                                670
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        654,
+                                        507,
+                                        670
+                                    ],
+                                    "score": 1.0,
+                                    "content": "BN, adding the proposed generative modelling provides additional gain in terms of all the measures",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                666,
+                                506,
+                                681
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        666,
+                                        315,
+                                        681
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(PSNR, SSIM, LPIPS) of reconstruction quality. GI-",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        315,
+                                        667,
+                                        334,
+                                        679
+                                    ],
+                                    "score": 0.84,
+                                    "content": "z / w",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        334,
+                                        666,
+                                        506,
+                                        681
+                                    ],
+                                    "score": 1.0,
+                                    "content": "without BN has lower reconstruction error",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                677,
+                                505,
+                                691
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        677,
+                                        139,
+                                        691
+                                    ],
+                                    "score": 1.0,
+                                    "content": "than GI-",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        140,
+                                        678,
+                                        171,
+                                        688
+                                    ],
+                                    "score": 0.87,
+                                    "content": "x + { \\tt B N }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        171,
+                                        677,
+                                        505,
+                                        691
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", which is the method of [29]. This implies that the gain from the generative model is",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                689,
+                                505,
+                                702
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        689,
+                                        505,
+                                        702
+                                    ],
+                                    "score": 1.0,
+                                    "content": "comparable to that from BN statistics. However, while the generative model only requires a global",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                699,
+                                505,
+                                712
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        699,
+                                        505,
+                                        712
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(and hence coarse) knowledge on the entire dataset, BN statistics are local to the batch in hand and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                711,
+                                506,
+                                724
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        711,
+                                        506,
+                                        724
+                                    ],
+                                    "score": 1.0,
+                                    "content": "hence requires significantly more detailed information on the exact batch used to compute gradient.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        }
+                    ],
+                    "index": 36
+                }
+            ],
+            "page_idx": 6,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        302,
+                        741,
+                        309,
+                        750
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                302,
+                                741,
+                                309,
+                                752
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        302,
+                                        741,
+                                        309,
+                                        752
+                                    ],
+                                    "score": 1.0,
+                                    "content": "7",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "image",
+                    "bbox": [
+                        100,
+                        73,
+                        499,
+                        223
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                100,
+                                73,
+                                499,
+                                223
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        100,
+                                        73,
+                                        499,
+                                        223
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                100,
+                                                73,
+                                                499,
+                                                223
+                                            ],
+                                            "score": 0.974,
+                                            "type": "image",
+                                            "image_path": "05927611c6cf240c845850b9daf59ed7a0462bf6d541e6a943eb9a3ea192a47a.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 1,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        100,
+                                        73,
+                                        499,
+                                        123.0
+                                    ],
+                                    "spans": [],
+                                    "index": 0
+                                },
+                                {
+                                    "bbox": [
+                                        100,
+                                        123.0,
+                                        499,
+                                        173.0
+                                    ],
+                                    "spans": [],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        100,
+                                        173.0,
+                                        499,
+                                        223.0
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                106,
+                                230,
+                                505,
+                                285
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        229,
+                                        506,
+                                        243
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                229,
+                                                409,
+                                                243
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 4: Comparison of state-of-the-art models and ours. Replacing GI-",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                410,
+                                                232,
+                                                416,
+                                                240
+                                            ],
+                                            "score": 0.7,
+                                            "content": "x",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                417,
+                                                229,
+                                                453,
+                                                243
+                                            ],
+                                            "score": 1.0,
+                                            "content": "with GI-",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                453,
+                                                230,
+                                                471,
+                                                242
+                                            ],
+                                            "score": 0.62,
+                                            "content": "z / w",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                472,
+                                                229,
+                                                506,
+                                                243
+                                            ],
+                                            "score": 1.0,
+                                            "content": "(GIAS)",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 3
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        241,
+                                        506,
+                                        254
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                241,
+                                                506,
+                                                254
+                                            ],
+                                            "score": 1.0,
+                                            "content": "regardless of using BN [29] or not [9] provides substantial improvement in the reconstruction accuracy.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 4
+                                },
+                                {
+                                    "bbox": [
+                                        104,
+                                        251,
+                                        506,
+                                        265
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                104,
+                                                251,
+                                                506,
+                                                265
+                                            ],
+                                            "score": 1.0,
+                                            "content": "(a) Average PSNR and best PSNR in a batch throughout the experiments. (b) An ablation study and",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 5
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        262,
+                                        505,
+                                        276
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                262,
+                                                505,
+                                                276
+                                            ],
+                                            "score": 1.0,
+                                            "content": "comparison of reconstruction results with our models and state-of-the-art models. We highlight the",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 6
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        274,
+                                        210,
+                                        286
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                274,
+                                                210,
+                                                286
+                                            ],
+                                            "score": 1.0,
+                                            "content": "proposed models in bold.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 7
+                                }
+                            ],
+                            "index": 5
+                        }
+                    ],
+                    "index": 3.0
+                },
+                {
+                    "type": "table",
+                    "bbox": [
+                        108,
+                        336,
+                        503,
+                        384
+                    ],
+                    "blocks": [
+                        {
+                            "type": "table_caption",
+                            "bbox": [
+                                106,
+                                296,
+                                505,
+                                330
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        295,
+                                        505,
+                                        309
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                104,
+                                                295,
+                                                505,
+                                                309
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Table 1: Comparison of our methods with state-of-the-art methods. Adding our method makes",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 8
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        307,
+                                        505,
+                                        320
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                307,
+                                                505,
+                                                320
+                                            ],
+                                            "score": 1.0,
+                                            "content": "performance improvement versus two baseline methods. PSNR, SSIM, and LPIPS[30] are used to",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 9
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        318,
+                                        410,
+                                        331
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                318,
+                                                410,
+                                                331
+                                            ],
+                                            "score": 1.0,
+                                            "content": "evaluate reconstruction results. We highlight the best performances in bold.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 10
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "type": "table_body",
+                            "bbox": [
+                                108,
+                                336,
+                                503,
+                                384
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        108,
+                                        336,
+                                        503,
+                                        384
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                108,
+                                                336,
+                                                503,
+                                                384
+                                            ],
+                                            "score": 0.977,
+                                            "html": "<table><tr><td>Method</td><td>GI-x [9]</td><td>GI-z (ours)</td><td>GI-w (ours)</td><td>GI-z/w (GIAS,ours)</td><td>GI-x+BN[29]</td><td>GI-z/w+BN (ours)</td></tr><tr><td>PSNR↑</td><td>13.78</td><td>14.27</td><td>14.70</td><td>15.58</td><td>15.52</td><td>16.31</td></tr><tr><td>SSIM↑</td><td>0.2542</td><td>0.3106</td><td>0.3519</td><td>0.3895</td><td>0.3513</td><td>0.4311</td></tr><tr><td>LPIPS↓</td><td>0.4376</td><td>0.3233</td><td>0.5121</td><td>0.3023</td><td>0.3645</td><td>0.2861</td></tr></table>",
+                                            "type": "table",
+                                            "image_path": "d2238a9c5fd3fdd85e97704e60c544b53a261d8a1d75ae6af9d39b80635553ea.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 12,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        108,
+                                        336,
+                                        503,
+                                        352.0
+                                    ],
+                                    "spans": [],
+                                    "index": 11
+                                },
+                                {
+                                    "bbox": [
+                                        108,
+                                        352.0,
+                                        503,
+                                        368.0
+                                    ],
+                                    "spans": [],
+                                    "index": 12
+                                },
+                                {
+                                    "bbox": [
+                                        108,
+                                        368.0,
+                                        503,
+                                        384.0
+                                    ],
+                                    "spans": [],
+                                    "index": 13
+                                }
+                            ]
+                        }
+                    ],
+                    "index": 10.5
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        395,
+                        250,
+                        407
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                393,
+                                251,
+                                410
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        393,
+                                        251,
+                                        410
+                                    ],
+                                    "score": 1.0,
+                                    "content": "5.1 Justification of GIAS design",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        }
+                    ],
+                    "index": 14
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        106,
+                        415,
+                        505,
+                        590
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                416,
+                                506,
+                                428
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        416,
+                                        506,
+                                        428
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We first provide an empirical justification of the specific order of searching spaces in GIAS (corre-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                426,
+                                506,
+                                439
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        426,
+                                        156,
+                                        439
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sponding to",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        156,
+                                        426,
+                                        190,
+                                        438
+                                    ],
+                                    "score": 0.79,
+                                    "content": "\\mathbf { G I } { - } z / w ,",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        190,
+                                        426,
+                                        506,
+                                        439
+                                    ],
+                                    "score": 1.0,
+                                    "content": ") to fully utilize a pretrained generative model. To do so, we provide Figure 4b",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                437,
+                                505,
+                                450
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        437,
+                                        344,
+                                        450
+                                    ],
+                                    "score": 1.0,
+                                    "content": "comparing algorithms with different searching spaces: GI-",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        344,
+                                        437,
+                                        362,
+                                        449
+                                    ],
+                                    "score": 0.86,
+                                    "content": "z / w",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        362,
+                                        437,
+                                        380,
+                                        450
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", GI-",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        381,
+                                        437,
+                                        397,
+                                        449
+                                    ],
+                                    "score": 0.65,
+                                    "content": "z / x",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        397,
+                                        437,
+                                        400,
+                                        450
+                                    ],
+                                    "score": 1.0,
+                                    "content": ",",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        401,
+                                        437,
+                                        421,
+                                        448
+                                    ],
+                                    "score": 0.55,
+                                    "content": "\\mathrm { G I } { - } z",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        421,
+                                        437,
+                                        442,
+                                        450
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        442,
+                                        437,
+                                        463,
+                                        448
+                                    ],
+                                    "score": 0.36,
+                                    "content": "_ { \\mathrm { G I - } x }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        464,
+                                        437,
+                                        505,
+                                        450
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", of which",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                448,
+                                506,
+                                461
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        448,
+                                        325,
+                                        461
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the first three share the same latent space search over",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        325,
+                                        451,
+                                        332,
+                                        458
+                                    ],
+                                    "score": 0.76,
+                                    "content": "z",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        332,
+                                        448,
+                                        506,
+                                        461
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for the first 1, 500 iterations. As shown in",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                460,
+                                506,
+                                471
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        460,
+                                        270,
+                                        471
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Figure 3(a), the latent space search over",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        270,
+                                        461,
+                                        277,
+                                        469
+                                    ],
+                                    "score": 0.73,
+                                    "content": "z",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        277,
+                                        460,
+                                        506,
+                                        471
+                                    ],
+                                    "score": 1.0,
+                                    "content": "quickly finds plausible image in a much shorter number",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                469,
+                                506,
+                                483
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        469,
+                                        193,
+                                        483
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of iterations than GI-",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        194,
+                                        471,
+                                        200,
+                                        480
+                                    ],
+                                    "score": 0.71,
+                                    "content": "x",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        200,
+                                        469,
+                                        506,
+                                        483
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", while it does not improve after a certain point due to the imperfection of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                481,
+                                506,
+                                493
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        481,
+                                        302,
+                                        493
+                                    ],
+                                    "score": 1.0,
+                                    "content": "pretrained generative model. Such a limitation of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        302,
+                                        481,
+                                        322,
+                                        492
+                                    ],
+                                    "score": 0.27,
+                                    "content": "\\mathrm { G I } { - } z",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        322,
+                                        481,
+                                        506,
+                                        493
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is also captured in Figure 3(b), where the cost",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                491,
+                                506,
+                                504
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        491,
+                                        167,
+                                        504
+                                    ],
+                                    "score": 1.0,
+                                    "content": "function of GI-",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        167,
+                                        493,
+                                        173,
+                                        502
+                                    ],
+                                    "score": 0.7,
+                                    "content": "z",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        174,
+                                        491,
+                                        506,
+                                        504
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is not decreasing after a certain number of optimization steps. To further minimize",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                502,
+                                505,
+                                515
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        503,
+                                        268,
+                                        515
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the cost function, one alternative to GI-",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        268,
+                                        502,
+                                        287,
+                                        514
+                                    ],
+                                    "score": 0.67,
+                                    "content": "z / w",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        287,
+                                        503,
+                                        331,
+                                        515
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(GIAS) is",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        331,
+                                        503,
+                                        362,
+                                        514
+                                    ],
+                                    "score": 0.8,
+                                    "content": "{ \\mathrm { G I } } { - } z / x",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        362,
+                                        503,
+                                        505,
+                                        515
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", which can further reduce the loss",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                513,
+                                506,
+                                526
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        513,
+                                        289,
+                                        526
+                                    ],
+                                    "score": 1.0,
+                                    "content": "function whereas the parameter search in GI-",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        290,
+                                        514,
+                                        308,
+                                        525
+                                    ],
+                                    "score": 0.71,
+                                    "content": "z / w",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        308,
+                                        513,
+                                        506,
+                                        526
+                                    ],
+                                    "score": 1.0,
+                                    "content": "seems to provide more natural reconstruction of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                524,
+                                506,
+                                538
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        524,
+                                        168,
+                                        538
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the image than",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        168,
+                                        524,
+                                        199,
+                                        537
+                                    ],
+                                    "score": 0.76,
+                                    "content": "{ \\mathrm { G I } } { - } z / { \\bar { x } }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        199,
+                                        524,
+                                        278,
+                                        538
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". The superiority of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        278,
+                                        525,
+                                        311,
+                                        537
+                                    ],
+                                    "score": 0.84,
+                                    "content": "{ \\mathrm { G I } } { - z } / w",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        311,
+                                        524,
+                                        332,
+                                        538
+                                    ],
+                                    "score": 1.0,
+                                    "content": "over",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        332,
+                                        525,
+                                        363,
+                                        537
+                                    ],
+                                    "score": 0.73,
+                                    "content": "{ \\mathrm { G I } } { - } z / x",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        363,
+                                        524,
+                                        506,
+                                        538
+                                    ],
+                                    "score": 1.0,
+                                    "content": "may come from that the parameter",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                535,
+                                506,
+                                549
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        535,
+                                        506,
+                                        549
+                                    ],
+                                    "score": 1.0,
+                                    "content": "space search exploits an implicit bias from optimizing a good architecture for expressing images, c.f.,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                546,
+                                506,
+                                559
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        546,
+                                        506,
+                                        559
+                                    ],
+                                    "score": 1.0,
+                                    "content": "deep image prior [26]. In Appendix E and Figure 1, we also present the same comparison on FFHQ",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                558,
+                                505,
+                                569
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        558,
+                                        505,
+                                        569
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(human-face images) [12] where diversity is much smaller than that of ImageNet. On such a less",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                567,
+                                505,
+                                581
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        567,
+                                        505,
+                                        581
+                                    ],
+                                    "score": 1.0,
+                                    "content": "diverse dataset, the distribution can be easily learned, and the gain from training a generative model",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                578,
+                                144,
+                                593
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        578,
+                                        144,
+                                        593
+                                    ],
+                                    "score": 1.0,
+                                    "content": "is larger.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        }
+                    ],
+                    "index": 22.5,
+                    "bbox_fs": [
+                        104,
+                        416,
+                        506,
+                        593
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        603,
+                        381,
+                        615
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                602,
+                                382,
+                                617
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        602,
+                                        382,
+                                        617
+                                    ],
+                                    "score": 1.0,
+                                    "content": "5.2 The gain from fully exploiting pretrained generative model",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        }
+                    ],
+                    "index": 31
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        623,
+                        505,
+                        722
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                623,
+                                505,
+                                636
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        623,
+                                        505,
+                                        636
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Comparison with state-of-the-art models. Our method can be easily added to previous methods",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                634,
+                                505,
+                                647
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        634,
+                                        505,
+                                        647
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[9, 29]. In Table 1 and Figure 4, we compare the state-of-the-art methods both with and without the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                645,
+                                506,
+                                658
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        645,
+                                        339,
+                                        658
+                                    ],
+                                    "score": 1.0,
+                                    "content": "proposed generative modelling. In Table 1, comparing GI-",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        339,
+                                        646,
+                                        346,
+                                        655
+                                    ],
+                                    "score": 0.75,
+                                    "content": "x",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        346,
+                                        645,
+                                        357,
+                                        658
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        357,
+                                        645,
+                                        390,
+                                        657
+                                    ],
+                                    "score": 0.67,
+                                    "content": "{ \\mathrm { G I } } { - } z / w",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        390,
+                                        645,
+                                        422,
+                                        658
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and GI-",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        422,
+                                        645,
+                                        453,
+                                        656
+                                    ],
+                                    "score": 0.81,
+                                    "content": "x + { \\tt B N }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        454,
+                                        645,
+                                        464,
+                                        658
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        465,
+                                        645,
+                                        506,
+                                        657
+                                    ],
+                                    "score": 0.76,
+                                    "content": "{ \\mathrm { G I } } { - } z / w +",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                654,
+                                507,
+                                670
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        654,
+                                        507,
+                                        670
+                                    ],
+                                    "score": 1.0,
+                                    "content": "BN, adding the proposed generative modelling provides additional gain in terms of all the measures",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                666,
+                                506,
+                                681
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        666,
+                                        315,
+                                        681
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(PSNR, SSIM, LPIPS) of reconstruction quality. GI-",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        315,
+                                        667,
+                                        334,
+                                        679
+                                    ],
+                                    "score": 0.84,
+                                    "content": "z / w",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        334,
+                                        666,
+                                        506,
+                                        681
+                                    ],
+                                    "score": 1.0,
+                                    "content": "without BN has lower reconstruction error",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                677,
+                                505,
+                                691
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        677,
+                                        139,
+                                        691
+                                    ],
+                                    "score": 1.0,
+                                    "content": "than GI-",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        140,
+                                        678,
+                                        171,
+                                        688
+                                    ],
+                                    "score": 0.87,
+                                    "content": "x + { \\tt B N }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        171,
+                                        677,
+                                        505,
+                                        691
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", which is the method of [29]. This implies that the gain from the generative model is",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                689,
+                                505,
+                                702
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        689,
+                                        505,
+                                        702
+                                    ],
+                                    "score": 1.0,
+                                    "content": "comparable to that from BN statistics. However, while the generative model only requires a global",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                699,
+                                505,
+                                712
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        699,
+                                        505,
+                                        712
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(and hence coarse) knowledge on the entire dataset, BN statistics are local to the batch in hand and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                711,
+                                506,
+                                724
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        711,
+                                        506,
+                                        724
+                                    ],
+                                    "score": 1.0,
+                                    "content": "hence requires significantly more detailed information on the exact batch used to compute gradient.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        }
+                    ],
+                    "index": 36,
+                    "bbox_fs": [
+                        105,
+                        623,
+                        507,
+                        724
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "image",
+                    "bbox": [
+                        106,
+                        72,
+                        483,
+                        211
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                106,
+                                72,
+                                483,
+                                211
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        72,
+                                        483,
+                                        211
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                72,
+                                                483,
+                                                211
+                                            ],
+                                            "score": 0.972,
+                                            "type": "image",
+                                            "image_path": "eaebaa7d2c36a894aebb95c062ddacad082b8c4df28b7ce3593656951181c37f.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 1,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        72,
+                                        483,
+                                        118.33333333333334
+                                    ],
+                                    "spans": [],
+                                    "index": 0
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        118.33333333333334,
+                                        483,
+                                        164.66666666666669
+                                    ],
+                                    "spans": [],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        164.66666666666669,
+                                        483,
+                                        211.00000000000003
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                106,
+                                218,
+                                505,
+                                251
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        217,
+                                        505,
+                                        230
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                217,
+                                                320,
+                                                230
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 5: Comparison of state-of-the-art models and",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                320,
+                                                218,
+                                                352,
+                                                230
+                                            ],
+                                            "score": 0.73,
+                                            "content": "G I - z / w",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                353,
+                                                217,
+                                                505,
+                                                230
+                                            ],
+                                            "score": 1.0,
+                                            "content": "with varying difficulties. Larger batch",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 3
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        229,
+                                        505,
+                                        242
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                229,
+                                                454,
+                                                242
+                                            ],
+                                            "score": 1.0,
+                                            "content": "size, higher sparsity, and larger gradient noise increases reconstruction difficulty. GI-",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                455,
+                                                229,
+                                                473,
+                                                241
+                                            ],
+                                            "score": 0.78,
+                                            "content": "z / w",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                473,
+                                                229,
+                                                505,
+                                                242
+                                            ],
+                                            "score": 1.0,
+                                            "content": "always",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 4
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        240,
+                                        465,
+                                        253
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                240,
+                                                160,
+                                                253
+                                            ],
+                                            "score": 1.0,
+                                            "content": "surpasses GI-",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                161,
+                                                241,
+                                                167,
+                                                250
+                                            ],
+                                            "score": 0.75,
+                                            "content": "x",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                168,
+                                                240,
+                                                465,
+                                                253
+                                            ],
+                                            "score": 1.0,
+                                            "content": "thanks to the pretrained generative model. All subfigures share the y-axis.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 5
+                                }
+                            ],
+                            "index": 4
+                        }
+                    ],
+                    "index": 2.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        274,
+                        505,
+                        329
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                274,
+                                505,
+                                286
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        274,
+                                        505,
+                                        286
+                                    ],
+                                    "score": 1.0,
+                                    "content": "As shown in Figure 4, the superiority of our method compared to the others is clear in terms of the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                285,
+                                505,
+                                298
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        285,
+                                        505,
+                                        298
+                                    ],
+                                    "score": 1.0,
+                                    "content": "best-in-batch performance than the average one, where the former is more suitable to show actual",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                295,
+                                505,
+                                309
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        295,
+                                        435,
+                                        309
+                                    ],
+                                    "score": 1.0,
+                                    "content": "privacy threat in the worst case than the latter. It is also interesting to note that GI-",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        435,
+                                        298,
+                                        443,
+                                        306
+                                    ],
+                                    "score": 0.77,
+                                    "content": "w",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        444,
+                                        295,
+                                        505,
+                                        309
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with untrained",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                307,
+                                506,
+                                320
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        309,
+                                        115,
+                                        317
+                                    ],
+                                    "score": 0.64,
+                                    "content": "w",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        115,
+                                        307,
+                                        288,
+                                        320
+                                    ],
+                                    "score": 1.0,
+                                    "content": "provides substantial gain compared to GI-",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        288,
+                                        309,
+                                        295,
+                                        317
+                                    ],
+                                    "score": 0.71,
+                                    "content": "x",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        295,
+                                        307,
+                                        506,
+                                        320
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". This may imply that there is a gain of the implicit",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                318,
+                                391,
+                                330
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        318,
+                                        391,
+                                        330
+                                    ],
+                                    "score": 1.0,
+                                    "content": "bias, c.f., [26], from training the architecture of deep generative model.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        }
+                    ],
+                    "index": 8
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        343,
+                        505,
+                        529
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                342,
+                                505,
+                                356
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        342,
+                                        505,
+                                        356
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Evaluation against possible defense methods We evaluate the gain of using a generative model",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                354,
+                                505,
+                                366
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        354,
+                                        505,
+                                        366
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for various FL scenarios with varying levels of difficulty in the inversion. As batch size, gradient",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                364,
+                                506,
+                                378
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        364,
+                                        506,
+                                        378
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sparsity1 [28] and gradient noise level increase, the risk of having under-determined inversion",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                375,
+                                505,
+                                388
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        375,
+                                        505,
+                                        388
+                                    ],
+                                    "score": 1.0,
+                                    "content": "increases and the inversion task becomes more challenging. Figure 5 shows that for all the levels",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                386,
+                                506,
+                                399
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        386,
+                                        506,
+                                        399
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of difficulty, the generative model provides significant gain in reconstruction quality. In particular,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                397,
+                                506,
+                                410
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        397,
+                                        216,
+                                        410
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the averaged PSNR of GI-",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        216,
+                                        399,
+                                        222,
+                                        407
+                                    ],
+                                    "score": 0.7,
+                                    "content": "x",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        223,
+                                        397,
+                                        417,
+                                        410
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with a batch size of 4 is comparable to that of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        417,
+                                        398,
+                                        450,
+                                        409
+                                    ],
+                                    "score": 0.72,
+                                    "content": "{ \\mathrm { G I } } { - z } / w",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        450,
+                                        397,
+                                        506,
+                                        410
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with a batch",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                408,
+                                505,
+                                421
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        408,
+                                        270,
+                                        421
+                                    ],
+                                    "score": 1.0,
+                                    "content": "size 32. It is also comparable to that of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        270,
+                                        408,
+                                        303,
+                                        420
+                                    ],
+                                    "score": 0.73,
+                                    "content": "{ \\mathrm { G I } } { - z } / w",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        303,
+                                        408,
+                                        414,
+                                        421
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with a gradient sparsity of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        415,
+                                        409,
+                                        434,
+                                        419
+                                    ],
+                                    "score": 0.86,
+                                    "content": "9 9 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        435,
+                                        408,
+                                        505,
+                                        421
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". To measure the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                419,
+                                505,
+                                432
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        419,
+                                        505,
+                                        432
+                                    ],
+                                    "score": 1.0,
+                                    "content": "impact of the noisy gradient, we experimented gradient inversion with varying gaussian noise level in",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                429,
+                                506,
+                                444
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        429,
+                                        506,
+                                        444
+                                    ],
+                                    "score": 1.0,
+                                    "content": "aforementioned settings. Figure 5(c) shows that adding enough noise to the gradient can mitigate",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                441,
+                                504,
+                                454
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        441,
+                                        202,
+                                        454
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the privacy leakage. GI-",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        202,
+                                        441,
+                                        221,
+                                        453
+                                    ],
+                                    "score": 0.87,
+                                    "content": "z / w",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        221,
+                                        441,
+                                        497,
+                                        454
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with a noise level of 0.01, which is relatively large, still surpasses GI-",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        497,
+                                        442,
+                                        504,
+                                        451
+                                    ],
+                                    "score": 0.71,
+                                    "content": "_ x",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                452,
+                                506,
+                                465
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        452,
+                                        506,
+                                        465
+                                    ],
+                                    "score": 1.0,
+                                    "content": "without noise. A large noise of 0.1 can diminish the gain of exploiting a pretrained generative model.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                462,
+                                506,
+                                477
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        462,
+                                        506,
+                                        477
+                                    ],
+                                    "score": 1.0,
+                                    "content": "However, the fact that adding large noise to the gradient slows down training makes it difficult for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                473,
+                                506,
+                                488
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        473,
+                                        506,
+                                        488
+                                    ],
+                                    "score": 1.0,
+                                    "content": "FL practitioners to choose suitable hyperparameters. The results imply our method is more robust",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                484,
+                                506,
+                                497
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        484,
+                                        506,
+                                        497
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to defense methods against gradient inversion, but can be blocked by a high threshold. Note that",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                496,
+                                506,
+                                508
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        496,
+                                        506,
+                                        508
+                                    ],
+                                    "score": 1.0,
+                                    "content": "our results of gradient sparsity and gradient noise implies the Differential Privacy(DP) is still a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                506,
+                                506,
+                                519
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        506,
+                                        506,
+                                        519
+                                    ],
+                                    "score": 1.0,
+                                    "content": "valid defense method, when applied with a more conservative threshold. For more discussion about",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                517,
+                                351,
+                                530
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        517,
+                                        351,
+                                        530
+                                    ],
+                                    "score": 1.0,
+                                    "content": "possible defense methods in FL framework, see Appendix F.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        }
+                    ],
+                    "index": 19
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        543,
+                        311,
+                        555
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                542,
+                                312,
+                                558
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        542,
+                                        312,
+                                        558
+                                    ],
+                                    "score": 1.0,
+                                    "content": "5.3 Learning generative model from gradients",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        }
+                    ],
+                    "index": 28
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        564,
+                        505,
+                        619
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                564,
+                                505,
+                                577
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        564,
+                                        505,
+                                        577
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We demonstrate the possibility of training a generative model only with gradients. For computational",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                574,
+                                506,
+                                588
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        574,
+                                        387,
+                                        588
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tractability, we use DCGAN and images from FFHQ [12] resized to",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        387,
+                                        575,
+                                        414,
+                                        586
+                                    ],
+                                    "score": 0.39,
+                                    "content": "3 2 \\mathbf { x } 3 2",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        415,
+                                        574,
+                                        506,
+                                        588
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". We generate a set of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                585,
+                                506,
+                                600
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        585,
+                                        506,
+                                        600
+                                    ],
+                                    "score": 1.0,
+                                    "content": "gradients from 4 rounds of gradient reports from 200 nodes, in which node computes gradient for a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                596,
+                                505,
+                                610
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        596,
+                                        505,
+                                        610
+                                    ],
+                                    "score": 1.0,
+                                    "content": "classification task based on the annotation provided in [6]. From the set of gradients, we perform",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                607,
+                                351,
+                                621
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        607,
+                                        351,
+                                        621
+                                    ],
+                                    "score": 1.0,
+                                    "content": "GIML to train a DCGAN to potentially generate FFHQ data.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        }
+                    ],
+                    "index": 31
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        624,
+                        505,
+                        690
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                624,
+                                506,
+                                637
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        624,
+                                        506,
+                                        637
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Figure 6 shows the evolution of generative model improves the reconstruction quality when perform-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                634,
+                                506,
+                                648
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        634,
+                                        146,
+                                        648
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ing either",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        147,
+                                        635,
+                                        167,
+                                        646
+                                    ],
+                                    "score": 0.35,
+                                    "content": "\\mathrm { G I } { - } z",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        167,
+                                        634,
+                                        184,
+                                        648
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        185,
+                                        635,
+                                        217,
+                                        647
+                                    ],
+                                    "score": 0.6,
+                                    "content": "{ \\mathrm { G I } } { - z } / w",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        217,
+                                        634,
+                                        506,
+                                        648
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". We can clearly see the necessity of parameter space search. Figure 6(a)",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                646,
+                                506,
+                                659
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        646,
+                                        506,
+                                        659
+                                    ],
+                                    "score": 1.0,
+                                    "content": "shows that the quality of images from the generative model is evolving in the training process of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                655,
+                                506,
+                                672
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        655,
+                                        183,
+                                        672
+                                    ],
+                                    "score": 1.0,
+                                    "content": "GIML. As the step",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        183,
+                                        658,
+                                        189,
+                                        667
+                                    ],
+                                    "score": 0.61,
+                                    "content": "t",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        189,
+                                        655,
+                                        354,
+                                        672
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of GIML increases, the generative model",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        354,
+                                        657,
+                                        391,
+                                        669
+                                    ],
+                                    "score": 0.93,
+                                    "content": "G _ { w ^ { ( t ) } } ( z )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        391,
+                                        655,
+                                        443,
+                                        672
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for arbitrary",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        443,
+                                        659,
+                                        450,
+                                        667
+                                    ],
+                                    "score": 0.74,
+                                    "content": "z",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        450,
+                                        655,
+                                        506,
+                                        672
+                                    ],
+                                    "score": 1.0,
+                                    "content": "outputs more",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                668,
+                                506,
+                                681
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        668,
+                                        506,
+                                        681
+                                    ],
+                                    "score": 1.0,
+                                    "content": "plausible image of human face. When using generative model trained on wrong dataset (CIFAR10),",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                679,
+                                270,
+                                691
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        679,
+                                        120,
+                                        691
+                                    ],
+                                    "score": 1.0,
+                                    "content": "GI-",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        121,
+                                        680,
+                                        127,
+                                        689
+                                    ],
+                                    "score": 0.73,
+                                    "content": "z",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        127,
+                                        679,
+                                        270,
+                                        691
+                                    ],
+                                    "score": 1.0,
+                                    "content": "completely fails at recovering data.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        }
+                    ],
+                    "index": 36.5
+                }
+            ],
+            "page_idx": 7,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        106,
+                        701,
+                        505,
+                        722
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                119,
+                                699,
+                                505,
+                                714
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        119,
+                                        699,
+                                        214,
+                                        714
+                                    ],
+                                    "score": 1.0,
+                                    "content": "1Having gradient sparsity",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        214,
+                                        702,
+                                        239,
+                                        711
+                                    ],
+                                    "score": 0.85,
+                                    "content": "0 . 9 9 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        240,
+                                        699,
+                                        379,
+                                        714
+                                    ],
+                                    "score": 1.0,
+                                    "content": "implies that we reconstruct data from",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        379,
+                                        702,
+                                        392,
+                                        711
+                                    ],
+                                    "score": 0.77,
+                                    "content": "1 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        392,
+                                        699,
+                                        505,
+                                        714
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of the gradient after removing",
+                                    "type": "text"
+                                }
+                            ]
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                710,
+                                316,
+                                723
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        711,
+                                        124,
+                                        721
+                                    ],
+                                    "score": 0.8,
+                                    "content": "9 9 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        124,
+                                        710,
+                                        316,
+                                        723
+                                    ],
+                                    "score": 1.0,
+                                    "content": "elements with the smallest magnitudes at each layer.",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                },
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        302,
+                        742,
+                        308,
+                        750
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                301,
+                                740,
+                                310,
+                                752
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        301,
+                                        740,
+                                        310,
+                                        752
+                                    ],
+                                    "score": 1.0,
+                                    "content": "",
+                                    "type": "text",
+                                    "height": 12,
+                                    "width": 9
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "image",
+                    "bbox": [
+                        106,
+                        72,
+                        483,
+                        211
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                106,
+                                72,
+                                483,
+                                211
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        72,
+                                        483,
+                                        211
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                72,
+                                                483,
+                                                211
+                                            ],
+                                            "score": 0.972,
+                                            "type": "image",
+                                            "image_path": "eaebaa7d2c36a894aebb95c062ddacad082b8c4df28b7ce3593656951181c37f.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 1,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        72,
+                                        483,
+                                        118.33333333333334
+                                    ],
+                                    "spans": [],
+                                    "index": 0
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        118.33333333333334,
+                                        483,
+                                        164.66666666666669
+                                    ],
+                                    "spans": [],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        164.66666666666669,
+                                        483,
+                                        211.00000000000003
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                106,
+                                218,
+                                505,
+                                251
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        217,
+                                        505,
+                                        230
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                217,
+                                                320,
+                                                230
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 5: Comparison of state-of-the-art models and",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                320,
+                                                218,
+                                                352,
+                                                230
+                                            ],
+                                            "score": 0.73,
+                                            "content": "G I - z / w",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                353,
+                                                217,
+                                                505,
+                                                230
+                                            ],
+                                            "score": 1.0,
+                                            "content": "with varying difficulties. Larger batch",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 3
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        229,
+                                        505,
+                                        242
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                229,
+                                                454,
+                                                242
+                                            ],
+                                            "score": 1.0,
+                                            "content": "size, higher sparsity, and larger gradient noise increases reconstruction difficulty. GI-",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                455,
+                                                229,
+                                                473,
+                                                241
+                                            ],
+                                            "score": 0.78,
+                                            "content": "z / w",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                473,
+                                                229,
+                                                505,
+                                                242
+                                            ],
+                                            "score": 1.0,
+                                            "content": "always",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 4
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        240,
+                                        465,
+                                        253
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                240,
+                                                160,
+                                                253
+                                            ],
+                                            "score": 1.0,
+                                            "content": "surpasses GI-",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                161,
+                                                241,
+                                                167,
+                                                250
+                                            ],
+                                            "score": 0.75,
+                                            "content": "x",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                168,
+                                                240,
+                                                465,
+                                                253
+                                            ],
+                                            "score": 1.0,
+                                            "content": "thanks to the pretrained generative model. All subfigures share the y-axis.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 5
+                                }
+                            ],
+                            "index": 4
+                        }
+                    ],
+                    "index": 2.5
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        274,
+                        505,
+                        329
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                274,
+                                505,
+                                286
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        274,
+                                        505,
+                                        286
+                                    ],
+                                    "score": 1.0,
+                                    "content": "As shown in Figure 4, the superiority of our method compared to the others is clear in terms of the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                285,
+                                505,
+                                298
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        285,
+                                        505,
+                                        298
+                                    ],
+                                    "score": 1.0,
+                                    "content": "best-in-batch performance than the average one, where the former is more suitable to show actual",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                295,
+                                505,
+                                309
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        295,
+                                        435,
+                                        309
+                                    ],
+                                    "score": 1.0,
+                                    "content": "privacy threat in the worst case than the latter. It is also interesting to note that GI-",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        435,
+                                        298,
+                                        443,
+                                        306
+                                    ],
+                                    "score": 0.77,
+                                    "content": "w",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        444,
+                                        295,
+                                        505,
+                                        309
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with untrained",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                307,
+                                506,
+                                320
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        309,
+                                        115,
+                                        317
+                                    ],
+                                    "score": 0.64,
+                                    "content": "w",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        115,
+                                        307,
+                                        288,
+                                        320
+                                    ],
+                                    "score": 1.0,
+                                    "content": "provides substantial gain compared to GI-",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        288,
+                                        309,
+                                        295,
+                                        317
+                                    ],
+                                    "score": 0.71,
+                                    "content": "x",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        295,
+                                        307,
+                                        506,
+                                        320
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". This may imply that there is a gain of the implicit",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                318,
+                                391,
+                                330
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        318,
+                                        391,
+                                        330
+                                    ],
+                                    "score": 1.0,
+                                    "content": "bias, c.f., [26], from training the architecture of deep generative model.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        }
+                    ],
+                    "index": 8,
+                    "bbox_fs": [
+                        105,
+                        274,
+                        506,
+                        330
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        343,
+                        505,
+                        529
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                342,
+                                505,
+                                356
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        342,
+                                        505,
+                                        356
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Evaluation against possible defense methods We evaluate the gain of using a generative model",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                354,
+                                505,
+                                366
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        354,
+                                        505,
+                                        366
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for various FL scenarios with varying levels of difficulty in the inversion. As batch size, gradient",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                364,
+                                506,
+                                378
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        364,
+                                        506,
+                                        378
+                                    ],
+                                    "score": 1.0,
+                                    "content": "sparsity1 [28] and gradient noise level increase, the risk of having under-determined inversion",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                375,
+                                505,
+                                388
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        375,
+                                        505,
+                                        388
+                                    ],
+                                    "score": 1.0,
+                                    "content": "increases and the inversion task becomes more challenging. Figure 5 shows that for all the levels",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                386,
+                                506,
+                                399
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        386,
+                                        506,
+                                        399
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of difficulty, the generative model provides significant gain in reconstruction quality. In particular,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                397,
+                                506,
+                                410
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        397,
+                                        216,
+                                        410
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the averaged PSNR of GI-",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        216,
+                                        399,
+                                        222,
+                                        407
+                                    ],
+                                    "score": 0.7,
+                                    "content": "x",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        223,
+                                        397,
+                                        417,
+                                        410
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with a batch size of 4 is comparable to that of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        417,
+                                        398,
+                                        450,
+                                        409
+                                    ],
+                                    "score": 0.72,
+                                    "content": "{ \\mathrm { G I } } { - z } / w",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        450,
+                                        397,
+                                        506,
+                                        410
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with a batch",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                408,
+                                505,
+                                421
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        408,
+                                        270,
+                                        421
+                                    ],
+                                    "score": 1.0,
+                                    "content": "size 32. It is also comparable to that of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        270,
+                                        408,
+                                        303,
+                                        420
+                                    ],
+                                    "score": 0.73,
+                                    "content": "{ \\mathrm { G I } } { - z } / w",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        303,
+                                        408,
+                                        414,
+                                        421
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with a gradient sparsity of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        415,
+                                        409,
+                                        434,
+                                        419
+                                    ],
+                                    "score": 0.86,
+                                    "content": "9 9 \\%",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        435,
+                                        408,
+                                        505,
+                                        421
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". To measure the",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                419,
+                                505,
+                                432
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        419,
+                                        505,
+                                        432
+                                    ],
+                                    "score": 1.0,
+                                    "content": "impact of the noisy gradient, we experimented gradient inversion with varying gaussian noise level in",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                429,
+                                506,
+                                444
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        429,
+                                        506,
+                                        444
+                                    ],
+                                    "score": 1.0,
+                                    "content": "aforementioned settings. Figure 5(c) shows that adding enough noise to the gradient can mitigate",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                441,
+                                504,
+                                454
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        441,
+                                        202,
+                                        454
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the privacy leakage. GI-",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        202,
+                                        441,
+                                        221,
+                                        453
+                                    ],
+                                    "score": 0.87,
+                                    "content": "z / w",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        221,
+                                        441,
+                                        497,
+                                        454
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with a noise level of 0.01, which is relatively large, still surpasses GI-",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        497,
+                                        442,
+                                        504,
+                                        451
+                                    ],
+                                    "score": 0.71,
+                                    "content": "_ x",
+                                    "type": "inline_equation"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                452,
+                                506,
+                                465
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        452,
+                                        506,
+                                        465
+                                    ],
+                                    "score": 1.0,
+                                    "content": "without noise. A large noise of 0.1 can diminish the gain of exploiting a pretrained generative model.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                462,
+                                506,
+                                477
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        462,
+                                        506,
+                                        477
+                                    ],
+                                    "score": 1.0,
+                                    "content": "However, the fact that adding large noise to the gradient slows down training makes it difficult for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                473,
+                                506,
+                                488
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        473,
+                                        506,
+                                        488
+                                    ],
+                                    "score": 1.0,
+                                    "content": "FL practitioners to choose suitable hyperparameters. The results imply our method is more robust",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                484,
+                                506,
+                                497
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        484,
+                                        506,
+                                        497
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to defense methods against gradient inversion, but can be blocked by a high threshold. Note that",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                496,
+                                506,
+                                508
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        496,
+                                        506,
+                                        508
+                                    ],
+                                    "score": 1.0,
+                                    "content": "our results of gradient sparsity and gradient noise implies the Differential Privacy(DP) is still a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                506,
+                                506,
+                                519
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        506,
+                                        506,
+                                        519
+                                    ],
+                                    "score": 1.0,
+                                    "content": "valid defense method, when applied with a more conservative threshold. For more discussion about",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                517,
+                                351,
+                                530
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        517,
+                                        351,
+                                        530
+                                    ],
+                                    "score": 1.0,
+                                    "content": "possible defense methods in FL framework, see Appendix F.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        }
+                    ],
+                    "index": 19,
+                    "bbox_fs": [
+                        104,
+                        342,
+                        506,
+                        530
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        543,
+                        311,
+                        555
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                542,
+                                312,
+                                558
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        542,
+                                        312,
+                                        558
+                                    ],
+                                    "score": 1.0,
+                                    "content": "5.3 Learning generative model from gradients",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        }
+                    ],
+                    "index": 28
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        564,
+                        505,
+                        619
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                564,
+                                505,
+                                577
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        564,
+                                        505,
+                                        577
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We demonstrate the possibility of training a generative model only with gradients. For computational",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                574,
+                                506,
+                                588
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        574,
+                                        387,
+                                        588
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tractability, we use DCGAN and images from FFHQ [12] resized to",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        387,
+                                        575,
+                                        414,
+                                        586
+                                    ],
+                                    "score": 0.39,
+                                    "content": "3 2 \\mathbf { x } 3 2",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        415,
+                                        574,
+                                        506,
+                                        588
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". We generate a set of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                585,
+                                506,
+                                600
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        585,
+                                        506,
+                                        600
+                                    ],
+                                    "score": 1.0,
+                                    "content": "gradients from 4 rounds of gradient reports from 200 nodes, in which node computes gradient for a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                596,
+                                505,
+                                610
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        596,
+                                        505,
+                                        610
+                                    ],
+                                    "score": 1.0,
+                                    "content": "classification task based on the annotation provided in [6]. From the set of gradients, we perform",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                607,
+                                351,
+                                621
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        607,
+                                        351,
+                                        621
+                                    ],
+                                    "score": 1.0,
+                                    "content": "GIML to train a DCGAN to potentially generate FFHQ data.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        }
+                    ],
+                    "index": 31,
+                    "bbox_fs": [
+                        105,
+                        564,
+                        506,
+                        621
+                    ]
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        624,
+                        505,
+                        690
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                624,
+                                506,
+                                637
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        624,
+                                        506,
+                                        637
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Figure 6 shows the evolution of generative model improves the reconstruction quality when perform-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                634,
+                                506,
+                                648
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        634,
+                                        146,
+                                        648
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ing either",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        147,
+                                        635,
+                                        167,
+                                        646
+                                    ],
+                                    "score": 0.35,
+                                    "content": "\\mathrm { G I } { - } z",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        167,
+                                        634,
+                                        184,
+                                        648
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        185,
+                                        635,
+                                        217,
+                                        647
+                                    ],
+                                    "score": 0.6,
+                                    "content": "{ \\mathrm { G I } } { - z } / w",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        217,
+                                        634,
+                                        506,
+                                        648
+                                    ],
+                                    "score": 1.0,
+                                    "content": ". We can clearly see the necessity of parameter space search. Figure 6(a)",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                646,
+                                506,
+                                659
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        646,
+                                        506,
+                                        659
+                                    ],
+                                    "score": 1.0,
+                                    "content": "shows that the quality of images from the generative model is evolving in the training process of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                655,
+                                506,
+                                672
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        655,
+                                        183,
+                                        672
+                                    ],
+                                    "score": 1.0,
+                                    "content": "GIML. As the step",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        183,
+                                        658,
+                                        189,
+                                        667
+                                    ],
+                                    "score": 0.61,
+                                    "content": "t",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        189,
+                                        655,
+                                        354,
+                                        672
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of GIML increases, the generative model",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        354,
+                                        657,
+                                        391,
+                                        669
+                                    ],
+                                    "score": 0.93,
+                                    "content": "G _ { w ^ { ( t ) } } ( z )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        391,
+                                        655,
+                                        443,
+                                        672
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for arbitrary",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        443,
+                                        659,
+                                        450,
+                                        667
+                                    ],
+                                    "score": 0.74,
+                                    "content": "z",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        450,
+                                        655,
+                                        506,
+                                        672
+                                    ],
+                                    "score": 1.0,
+                                    "content": "outputs more",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                668,
+                                506,
+                                681
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        668,
+                                        506,
+                                        681
+                                    ],
+                                    "score": 1.0,
+                                    "content": "plausible image of human face. When using generative model trained on wrong dataset (CIFAR10),",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                679,
+                                270,
+                                691
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        679,
+                                        120,
+                                        691
+                                    ],
+                                    "score": 1.0,
+                                    "content": "GI-",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        121,
+                                        680,
+                                        127,
+                                        689
+                                    ],
+                                    "score": 0.73,
+                                    "content": "z",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        127,
+                                        679,
+                                        270,
+                                        691
+                                    ],
+                                    "score": 1.0,
+                                    "content": "completely fails at recovering data.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        }
+                    ],
+                    "index": 36.5,
+                    "bbox_fs": [
+                        104,
+                        624,
+                        506,
+                        691
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "image",
+                    "bbox": [
+                        109,
+                        64,
+                        500,
+                        209
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                109,
+                                64,
+                                500,
+                                209
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        109,
+                                        64,
+                                        500,
+                                        209
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                109,
+                                                64,
+                                                500,
+                                                209
+                                            ],
+                                            "score": 0.973,
+                                            "type": "image",
+                                            "image_path": "11a9bf0deccf6f26645320b1714e23bf518f3cf71177c576ecd107d8a4e5a1fa.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 1,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        109,
+                                        64,
+                                        500,
+                                        112.33333333333334
+                                    ],
+                                    "spans": [],
+                                    "index": 0
+                                },
+                                {
+                                    "bbox": [
+                                        109,
+                                        112.33333333333334,
+                                        500,
+                                        160.66666666666669
+                                    ],
+                                    "spans": [],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        109,
+                                        160.66666666666669,
+                                        500,
+                                        209.00000000000003
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                106,
+                                216,
+                                505,
+                                304
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        216,
+                                        506,
+                                        228
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                216,
+                                                506,
+                                                228
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 6: Qualitative and quantitative result of GIML. (a) Results validating generative model trained",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 3
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        227,
+                                        505,
+                                        240
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                227,
+                                                505,
+                                                240
+                                            ],
+                                            "score": 1.0,
+                                            "content": "with GIML. Images on the first row are sampled from different GIML training steps. The same",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 4
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        238,
+                                        506,
+                                        251
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                238,
+                                                152,
+                                                251
+                                            ],
+                                            "score": 1.0,
+                                            "content": "latent code",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                152,
+                                                240,
+                                                159,
+                                                248
+                                            ],
+                                            "score": 0.72,
+                                            "content": "z",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                159,
+                                                238,
+                                                506,
+                                                251
+                                            ],
+                                            "score": 1.0,
+                                            "content": "was used to sample images in same rows. Images on the second row and third row are",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 5
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        249,
+                                        505,
+                                        261
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                249,
+                                                145,
+                                                261
+                                            ],
+                                            "score": 1.0,
+                                            "content": "results of",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                146,
+                                                249,
+                                                167,
+                                                259
+                                            ],
+                                            "score": 0.29,
+                                            "content": "\\mathrm { G I } { - } z",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                167,
+                                                249,
+                                                198,
+                                                261
+                                            ],
+                                            "score": 1.0,
+                                            "content": "and GI-",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                198,
+                                                249,
+                                                217,
+                                                261
+                                            ],
+                                            "score": 0.81,
+                                            "content": "z / w",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                217,
+                                                249,
+                                                505,
+                                                261
+                                            ],
+                                            "score": 1.0,
+                                            "content": "using generative model trained with GIML and pretrained model which",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 6
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        260,
+                                        505,
+                                        273
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                260,
+                                                505,
+                                                273
+                                            ],
+                                            "score": 1.0,
+                                            "content": "is trained with CIFAR10 images. Experiments were done with gradient sparsity 0.95 for comparison",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 7
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        271,
+                                        506,
+                                        284
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                271,
+                                                435,
+                                                284
+                                            ],
+                                            "score": 1.0,
+                                            "content": "in difficult setting. Last column represents the ground truth image and result of GI-",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                435,
+                                                272,
+                                                444,
+                                                281
+                                            ],
+                                            "score": 0.77,
+                                            "content": "\\boldsymbol { w }",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                444,
+                                                271,
+                                                506,
+                                                284
+                                            ],
+                                            "score": 1.0,
+                                            "content": "with untrained",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 8
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        281,
+                                        506,
+                                        294
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                281,
+                                                506,
+                                                294
+                                            ],
+                                            "score": 1.0,
+                                            "content": "model. (b) A comparison of GIAS with meta-learned generative model and GIAS using improper",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 9
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        293,
+                                        390,
+                                        305
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                293,
+                                                390,
+                                                305
+                                            ],
+                                            "score": 1.0,
+                                            "content": "generative model. Proper generative model boosts GIAS performance.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 10
+                                }
+                            ],
+                            "index": 6.5
+                        }
+                    ],
+                    "index": 3.75
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        324,
+                        505,
+                        390
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                324,
+                                506,
+                                336
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        324,
+                                        146,
+                                        336
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In Figure",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        146,
+                                        324,
+                                        164,
+                                        335
+                                    ],
+                                    "score": 0.32,
+                                    "content": "6 ( \\mathbf { b } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        165,
+                                        324,
+                                        384,
+                                        336
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", as GIML iteration step increases, the performance of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        384,
+                                        324,
+                                        405,
+                                        335
+                                    ],
+                                    "score": 0.26,
+                                    "content": "\\mathrm { G I } { - } z",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        405,
+                                        324,
+                                        422,
+                                        336
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        423,
+                                        324,
+                                        456,
+                                        336
+                                    ],
+                                    "score": 0.85,
+                                    "content": "{ \\mathrm { G I } } { - z } / w",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        456,
+                                        324,
+                                        506,
+                                        336
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with GIML",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                334,
+                                506,
+                                348
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        334,
+                                        153,
+                                        348
+                                    ],
+                                    "score": 1.0,
+                                    "content": "surpass GI-",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        153,
+                                        336,
+                                        159,
+                                        345
+                                    ],
+                                    "score": 0.78,
+                                    "content": "z",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        160,
+                                        334,
+                                        177,
+                                        348
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        177,
+                                        335,
+                                        210,
+                                        347
+                                    ],
+                                    "score": 0.66,
+                                    "content": "{ \\mathrm { G I } } { - z } / w",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        210,
+                                        334,
+                                        346,
+                                        348
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with wrong prior knowledge. GI-",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        346,
+                                        335,
+                                        365,
+                                        347
+                                    ],
+                                    "score": 0.88,
+                                    "content": "z / w",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        365,
+                                        334,
+                                        506,
+                                        348
+                                    ],
+                                    "score": 1.0,
+                                    "content": "using generative model trained on",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                346,
+                                505,
+                                358
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        346,
+                                        193,
+                                        358
+                                    ],
+                                    "score": 1.0,
+                                    "content": "wrong dataset and GI-",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        194,
+                                        348,
+                                        202,
+                                        356
+                                    ],
+                                    "score": 0.79,
+                                    "content": "\\boldsymbol { \\cdot } \\boldsymbol { w }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        202,
+                                        346,
+                                        505,
+                                        358
+                                    ],
+                                    "score": 1.0,
+                                    "content": "which starts with an untrained generative model show lower averaged PSNR",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                357,
+                                505,
+                                370
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        357,
+                                        158,
+                                        370
+                                    ],
+                                    "score": 1.0,
+                                    "content": "compared to",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        158,
+                                        357,
+                                        191,
+                                        369
+                                    ],
+                                    "score": 0.63,
+                                    "content": "{ \\mathrm { G I } } { - z } / w",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        191,
+                                        357,
+                                        256,
+                                        370
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with GIML. GI-",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        257,
+                                        357,
+                                        275,
+                                        369
+                                    ],
+                                    "score": 0.82,
+                                    "content": "z / w",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        275,
+                                        357,
+                                        505,
+                                        370
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with GIML to train generative model on right data shows",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                367,
+                                505,
+                                380
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        367,
+                                        505,
+                                        380
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the best performance in terms of not only quality (Figure 6) but also convergence speed. We provide",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                379,
+                                329,
+                                391
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        379,
+                                        329,
+                                        391
+                                    ],
+                                    "score": 1.0,
+                                    "content": "a comparison of the convergence speed in Appendix G.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        }
+                    ],
+                    "index": 13.5
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        405,
+                        183,
+                        418
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                403,
+                                185,
+                                421
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        403,
+                                        185,
+                                        421
+                                    ],
+                                    "score": 1.0,
+                                    "content": "6 Conclusion",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        }
+                    ],
+                    "index": 17
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        429,
+                        505,
+                        517
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                429,
+                                505,
+                                442
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        429,
+                                        505,
+                                        442
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We propose GIAS fully exploit the prior information on user data from a pretrained generative model",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                441,
+                                505,
+                                453
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        441,
+                                        505,
+                                        453
+                                    ],
+                                    "score": 1.0,
+                                    "content": "when inverting gradient. We demonstrate significant privacy leakage using GIAS with pretrained",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                452,
+                                505,
+                                464
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        452,
+                                        505,
+                                        464
+                                    ],
+                                    "score": 1.0,
+                                    "content": "generative model in various challenging scenarios, where our method provides substantial gain",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                462,
+                                506,
+                                475
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        462,
+                                        506,
+                                        475
+                                    ],
+                                    "score": 1.0,
+                                    "content": "additionally to any other existing methods [9, 29]. In addition, we propose GIML which can train a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                473,
+                                505,
+                                486
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        473,
+                                        505,
+                                        486
+                                    ],
+                                    "score": 1.0,
+                                    "content": "generative model using only the gradients seen in the FL classifier training. We experimentally show",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                483,
+                                505,
+                                497
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        483,
+                                        505,
+                                        497
+                                    ],
+                                    "score": 1.0,
+                                    "content": "that GIML can meta-learn a generative model on the user data from only gradients, which improves",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                495,
+                                506,
+                                508
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        495,
+                                        506,
+                                        508
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the quality of each individual recovered image. To our best knowledge, GIML is the first capable of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                506,
+                                338,
+                                518
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        506,
+                                        338,
+                                        518
+                                    ],
+                                    "score": 1.0,
+                                    "content": "learning explicit prior on a set of gradient inversion tasks.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        }
+                    ],
+                    "index": 21.5
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        532,
+                        201,
+                        545
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                530,
+                                203,
+                                548
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        530,
+                                        203,
+                                        548
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Acknowledgments",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        }
+                    ],
+                    "index": 26
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        556,
+                        505,
+                        666
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                555,
+                                505,
+                                570
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        555,
+                                        505,
+                                        570
+                                    ],
+                                    "score": 1.0,
+                                    "content": "This work was partly supported by Institute of Information & communications Technology Planning",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                567,
+                                506,
+                                581
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        567,
+                                        506,
+                                        581
+                                    ],
+                                    "score": 1.0,
+                                    "content": "& Evaluation (IITP) grant funded by the Korea government (MSIT) (No. 2019-0-01906, Artificial",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                578,
+                                506,
+                                591
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        578,
+                                        506,
+                                        591
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Intelligence Graduate School Program (POSTECH)) and (No. 2021-0-00739, Development of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                588,
+                                506,
+                                602
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        588,
+                                        242,
+                                        602
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Distributed/Cooperative AI based",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        242,
+                                        589,
+                                        262,
+                                        600
+                                    ],
+                                    "score": 0.81,
+                                    "content": "^ { 5 \\mathrm { G } + }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        262,
+                                        588,
+                                        506,
+                                        602
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Network Data Analytics Functions and Control Technology).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                599,
+                                506,
+                                614
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        599,
+                                        506,
+                                        614
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Jinwoo Jeon and Jaechang Kim were supported by the Institute of Information & Communications",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                610,
+                                506,
+                                624
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        610,
+                                        506,
+                                        624
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Technology Planning & Evaluation (IITP) grant funded by Korea(MSIT) (2020-0-01594, PSAI",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                622,
+                                505,
+                                635
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        622,
+                                        505,
+                                        635
+                                    ],
+                                    "score": 1.0,
+                                    "content": "industry-academic joint research and education program). Kangwook Lee was supported by NSF/Intel",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                632,
+                                505,
+                                645
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        632,
+                                        505,
+                                        645
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Partnership on Machine Learning for Wireless Networking Program under Grant No. CNS-2003129",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                643,
+                                506,
+                                657
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        643,
+                                        506,
+                                        657
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and NSF Award DMS-2023239. Sewoong Oh acknowledges funding from NSF IIS-1929955, NSF",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                655,
+                                312,
+                                667
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        655,
+                                        312,
+                                        667
+                                    ],
+                                    "score": 1.0,
+                                    "content": "CCF 2019844, and Google faculty research award.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        }
+                    ],
+                    "index": 31.5
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        681,
+                        163,
+                        694
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                680,
+                                165,
+                                696
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        680,
+                                        165,
+                                        696
+                                    ],
+                                    "score": 1.0,
+                                    "content": "References",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        }
+                    ],
+                    "index": 37
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        112,
+                        700,
+                        504,
+                        722
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                110,
+                                699,
+                                505,
+                                713
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        110,
+                                        699,
+                                        505,
+                                        713
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[1] R. Abdal, Y. Qin, and P. Wonka. Image2stylegan++: How to edit the embedded images? In",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                709,
+                                506,
+                                725
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        709,
+                                        506,
+                                        725
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR),",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        }
+                    ],
+                    "index": 38.5
+                }
+            ],
+            "page_idx": 8,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        302,
+                        742,
+                        308,
+                        749
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                302,
+                                741,
+                                309,
+                                752
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        302,
+                                        741,
+                                        309,
+                                        752
+                                    ],
+                                    "score": 1.0,
+                                    "content": "9",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "image",
+                    "bbox": [
+                        109,
+                        64,
+                        500,
+                        209
+                    ],
+                    "blocks": [
+                        {
+                            "type": "image_body",
+                            "bbox": [
+                                109,
+                                64,
+                                500,
+                                209
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        109,
+                                        64,
+                                        500,
+                                        209
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                109,
+                                                64,
+                                                500,
+                                                209
+                                            ],
+                                            "score": 0.973,
+                                            "type": "image",
+                                            "image_path": "11a9bf0deccf6f26645320b1714e23bf518f3cf71177c576ecd107d8a4e5a1fa.jpg"
+                                        }
+                                    ]
+                                }
+                            ],
+                            "index": 1,
+                            "virtual_lines": [
+                                {
+                                    "bbox": [
+                                        109,
+                                        64,
+                                        500,
+                                        112.33333333333334
+                                    ],
+                                    "spans": [],
+                                    "index": 0
+                                },
+                                {
+                                    "bbox": [
+                                        109,
+                                        112.33333333333334,
+                                        500,
+                                        160.66666666666669
+                                    ],
+                                    "spans": [],
+                                    "index": 1
+                                },
+                                {
+                                    "bbox": [
+                                        109,
+                                        160.66666666666669,
+                                        500,
+                                        209.00000000000003
+                                    ],
+                                    "spans": [],
+                                    "index": 2
+                                }
+                            ]
+                        },
+                        {
+                            "type": "image_caption",
+                            "bbox": [
+                                106,
+                                216,
+                                505,
+                                304
+                            ],
+                            "group_id": 0,
+                            "lines": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        216,
+                                        506,
+                                        228
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                216,
+                                                506,
+                                                228
+                                            ],
+                                            "score": 1.0,
+                                            "content": "Figure 6: Qualitative and quantitative result of GIML. (a) Results validating generative model trained",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 3
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        227,
+                                        505,
+                                        240
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                227,
+                                                505,
+                                                240
+                                            ],
+                                            "score": 1.0,
+                                            "content": "with GIML. Images on the first row are sampled from different GIML training steps. The same",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 4
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        238,
+                                        506,
+                                        251
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                238,
+                                                152,
+                                                251
+                                            ],
+                                            "score": 1.0,
+                                            "content": "latent code",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                152,
+                                                240,
+                                                159,
+                                                248
+                                            ],
+                                            "score": 0.72,
+                                            "content": "z",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                159,
+                                                238,
+                                                506,
+                                                251
+                                            ],
+                                            "score": 1.0,
+                                            "content": "was used to sample images in same rows. Images on the second row and third row are",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 5
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        249,
+                                        505,
+                                        261
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                249,
+                                                145,
+                                                261
+                                            ],
+                                            "score": 1.0,
+                                            "content": "results of",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                146,
+                                                249,
+                                                167,
+                                                259
+                                            ],
+                                            "score": 0.29,
+                                            "content": "\\mathrm { G I } { - } z",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                167,
+                                                249,
+                                                198,
+                                                261
+                                            ],
+                                            "score": 1.0,
+                                            "content": "and GI-",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                198,
+                                                249,
+                                                217,
+                                                261
+                                            ],
+                                            "score": 0.81,
+                                            "content": "z / w",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                217,
+                                                249,
+                                                505,
+                                                261
+                                            ],
+                                            "score": 1.0,
+                                            "content": "using generative model trained with GIML and pretrained model which",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 6
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        260,
+                                        505,
+                                        273
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                260,
+                                                505,
+                                                273
+                                            ],
+                                            "score": 1.0,
+                                            "content": "is trained with CIFAR10 images. Experiments were done with gradient sparsity 0.95 for comparison",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 7
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        271,
+                                        506,
+                                        284
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                271,
+                                                435,
+                                                284
+                                            ],
+                                            "score": 1.0,
+                                            "content": "in difficult setting. Last column represents the ground truth image and result of GI-",
+                                            "type": "text"
+                                        },
+                                        {
+                                            "bbox": [
+                                                435,
+                                                272,
+                                                444,
+                                                281
+                                            ],
+                                            "score": 0.77,
+                                            "content": "\\boldsymbol { w }",
+                                            "type": "inline_equation"
+                                        },
+                                        {
+                                            "bbox": [
+                                                444,
+                                                271,
+                                                506,
+                                                284
+                                            ],
+                                            "score": 1.0,
+                                            "content": "with untrained",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 8
+                                },
+                                {
+                                    "bbox": [
+                                        105,
+                                        281,
+                                        506,
+                                        294
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                105,
+                                                281,
+                                                506,
+                                                294
+                                            ],
+                                            "score": 1.0,
+                                            "content": "model. (b) A comparison of GIAS with meta-learned generative model and GIAS using improper",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 9
+                                },
+                                {
+                                    "bbox": [
+                                        106,
+                                        293,
+                                        390,
+                                        305
+                                    ],
+                                    "spans": [
+                                        {
+                                            "bbox": [
+                                                106,
+                                                293,
+                                                390,
+                                                305
+                                            ],
+                                            "score": 1.0,
+                                            "content": "generative model. Proper generative model boosts GIAS performance.",
+                                            "type": "text"
+                                        }
+                                    ],
+                                    "index": 10
+                                }
+                            ],
+                            "index": 6.5
+                        }
+                    ],
+                    "index": 3.75
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        324,
+                        505,
+                        390
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                324,
+                                506,
+                                336
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        324,
+                                        146,
+                                        336
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In Figure",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        146,
+                                        324,
+                                        164,
+                                        335
+                                    ],
+                                    "score": 0.32,
+                                    "content": "6 ( \\mathbf { b } )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        165,
+                                        324,
+                                        384,
+                                        336
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", as GIML iteration step increases, the performance of",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        384,
+                                        324,
+                                        405,
+                                        335
+                                    ],
+                                    "score": 0.26,
+                                    "content": "\\mathrm { G I } { - } z",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        405,
+                                        324,
+                                        422,
+                                        336
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        423,
+                                        324,
+                                        456,
+                                        336
+                                    ],
+                                    "score": 0.85,
+                                    "content": "{ \\mathrm { G I } } { - z } / w",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        456,
+                                        324,
+                                        506,
+                                        336
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with GIML",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                334,
+                                506,
+                                348
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        334,
+                                        153,
+                                        348
+                                    ],
+                                    "score": 1.0,
+                                    "content": "surpass GI-",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        153,
+                                        336,
+                                        159,
+                                        345
+                                    ],
+                                    "score": 0.78,
+                                    "content": "z",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        160,
+                                        334,
+                                        177,
+                                        348
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        177,
+                                        335,
+                                        210,
+                                        347
+                                    ],
+                                    "score": 0.66,
+                                    "content": "{ \\mathrm { G I } } { - z } / w",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        210,
+                                        334,
+                                        346,
+                                        348
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with wrong prior knowledge. GI-",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        346,
+                                        335,
+                                        365,
+                                        347
+                                    ],
+                                    "score": 0.88,
+                                    "content": "z / w",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        365,
+                                        334,
+                                        506,
+                                        348
+                                    ],
+                                    "score": 1.0,
+                                    "content": "using generative model trained on",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                346,
+                                505,
+                                358
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        346,
+                                        193,
+                                        358
+                                    ],
+                                    "score": 1.0,
+                                    "content": "wrong dataset and GI-",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        194,
+                                        348,
+                                        202,
+                                        356
+                                    ],
+                                    "score": 0.79,
+                                    "content": "\\boldsymbol { \\cdot } \\boldsymbol { w }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        202,
+                                        346,
+                                        505,
+                                        358
+                                    ],
+                                    "score": 1.0,
+                                    "content": "which starts with an untrained generative model show lower averaged PSNR",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                357,
+                                505,
+                                370
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        357,
+                                        158,
+                                        370
+                                    ],
+                                    "score": 1.0,
+                                    "content": "compared to",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        158,
+                                        357,
+                                        191,
+                                        369
+                                    ],
+                                    "score": 0.63,
+                                    "content": "{ \\mathrm { G I } } { - z } / w",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        191,
+                                        357,
+                                        256,
+                                        370
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with GIML. GI-",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        257,
+                                        357,
+                                        275,
+                                        369
+                                    ],
+                                    "score": 0.82,
+                                    "content": "z / w",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        275,
+                                        357,
+                                        505,
+                                        370
+                                    ],
+                                    "score": 1.0,
+                                    "content": "with GIML to train generative model on right data shows",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                367,
+                                505,
+                                380
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        367,
+                                        505,
+                                        380
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the best performance in terms of not only quality (Figure 6) but also convergence speed. We provide",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                379,
+                                329,
+                                391
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        379,
+                                        329,
+                                        391
+                                    ],
+                                    "score": 1.0,
+                                    "content": "a comparison of the convergence speed in Appendix G.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        }
+                    ],
+                    "index": 13.5,
+                    "bbox_fs": [
+                        105,
+                        324,
+                        506,
+                        391
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        405,
+                        183,
+                        418
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                104,
+                                403,
+                                185,
+                                421
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        403,
+                                        185,
+                                        421
+                                    ],
+                                    "score": 1.0,
+                                    "content": "6 Conclusion",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        }
+                    ],
+                    "index": 17
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        429,
+                        505,
+                        517
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                429,
+                                505,
+                                442
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        429,
+                                        505,
+                                        442
+                                    ],
+                                    "score": 1.0,
+                                    "content": "We propose GIAS fully exploit the prior information on user data from a pretrained generative model",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                441,
+                                505,
+                                453
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        441,
+                                        505,
+                                        453
+                                    ],
+                                    "score": 1.0,
+                                    "content": "when inverting gradient. We demonstrate significant privacy leakage using GIAS with pretrained",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                452,
+                                505,
+                                464
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        452,
+                                        505,
+                                        464
+                                    ],
+                                    "score": 1.0,
+                                    "content": "generative model in various challenging scenarios, where our method provides substantial gain",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                462,
+                                506,
+                                475
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        462,
+                                        506,
+                                        475
+                                    ],
+                                    "score": 1.0,
+                                    "content": "additionally to any other existing methods [9, 29]. In addition, we propose GIML which can train a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                473,
+                                505,
+                                486
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        473,
+                                        505,
+                                        486
+                                    ],
+                                    "score": 1.0,
+                                    "content": "generative model using only the gradients seen in the FL classifier training. We experimentally show",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                483,
+                                505,
+                                497
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        483,
+                                        505,
+                                        497
+                                    ],
+                                    "score": 1.0,
+                                    "content": "that GIML can meta-learn a generative model on the user data from only gradients, which improves",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                495,
+                                506,
+                                508
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        495,
+                                        506,
+                                        508
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the quality of each individual recovered image. To our best knowledge, GIML is the first capable of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                506,
+                                338,
+                                518
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        506,
+                                        338,
+                                        518
+                                    ],
+                                    "score": 1.0,
+                                    "content": "learning explicit prior on a set of gradient inversion tasks.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        }
+                    ],
+                    "index": 21.5,
+                    "bbox_fs": [
+                        105,
+                        429,
+                        506,
+                        518
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        108,
+                        532,
+                        201,
+                        545
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                530,
+                                203,
+                                548
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        530,
+                                        203,
+                                        548
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Acknowledgments",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        }
+                    ],
+                    "index": 26
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        107,
+                        556,
+                        505,
+                        666
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                105,
+                                555,
+                                505,
+                                570
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        555,
+                                        505,
+                                        570
+                                    ],
+                                    "score": 1.0,
+                                    "content": "This work was partly supported by Institute of Information & communications Technology Planning",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                567,
+                                506,
+                                581
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        567,
+                                        506,
+                                        581
+                                    ],
+                                    "score": 1.0,
+                                    "content": "& Evaluation (IITP) grant funded by the Korea government (MSIT) (No. 2019-0-01906, Artificial",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                578,
+                                506,
+                                591
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        578,
+                                        506,
+                                        591
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Intelligence Graduate School Program (POSTECH)) and (No. 2021-0-00739, Development of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                588,
+                                506,
+                                602
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        588,
+                                        242,
+                                        602
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Distributed/Cooperative AI based",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        242,
+                                        589,
+                                        262,
+                                        600
+                                    ],
+                                    "score": 0.81,
+                                    "content": "^ { 5 \\mathrm { G } + }",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        262,
+                                        588,
+                                        506,
+                                        602
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Network Data Analytics Functions and Control Technology).",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                599,
+                                506,
+                                614
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        599,
+                                        506,
+                                        614
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Jinwoo Jeon and Jaechang Kim were supported by the Institute of Information & Communications",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                610,
+                                506,
+                                624
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        610,
+                                        506,
+                                        624
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Technology Planning & Evaluation (IITP) grant funded by Korea(MSIT) (2020-0-01594, PSAI",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                622,
+                                505,
+                                635
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        622,
+                                        505,
+                                        635
+                                    ],
+                                    "score": 1.0,
+                                    "content": "industry-academic joint research and education program). Kangwook Lee was supported by NSF/Intel",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                632,
+                                505,
+                                645
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        632,
+                                        505,
+                                        645
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Partnership on Machine Learning for Wireless Networking Program under Grant No. CNS-2003129",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                643,
+                                506,
+                                657
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        643,
+                                        506,
+                                        657
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and NSF Award DMS-2023239. Sewoong Oh acknowledges funding from NSF IIS-1929955, NSF",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                655,
+                                312,
+                                667
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        655,
+                                        312,
+                                        667
+                                    ],
+                                    "score": 1.0,
+                                    "content": "CCF 2019844, and Google faculty research award.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        }
+                    ],
+                    "index": 31.5,
+                    "bbox_fs": [
+                        104,
+                        555,
+                        506,
+                        667
+                    ]
+                },
+                {
+                    "type": "title",
+                    "bbox": [
+                        107,
+                        681,
+                        163,
+                        694
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                680,
+                                165,
+                                696
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        680,
+                                        165,
+                                        696
+                                    ],
+                                    "score": 1.0,
+                                    "content": "References",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        }
+                    ],
+                    "index": 37
+                },
+                {
+                    "type": "text",
+                    "bbox": [
+                        112,
+                        700,
+                        504,
+                        722
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                110,
+                                699,
+                                505,
+                                713
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        110,
+                                        699,
+                                        505,
+                                        713
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[1] R. Abdal, Y. Qin, and P. Wonka. Image2stylegan++: How to edit the embedded images? In",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                709,
+                                506,
+                                725
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        709,
+                                        506,
+                                        725
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR),",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        }
+                    ],
+                    "index": 38.5,
+                    "bbox_fs": [
+                        110,
+                        699,
+                        506,
+                        725
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        57,
+                        507,
+                        724
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                126,
+                                72,
+                                174,
+                                86
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        72,
+                                        174,
+                                        86
+                                    ],
+                                    "score": 1.0,
+                                    "content": "June 2020.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                109,
+                                90,
+                                507,
+                                106
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        109,
+                                        90,
+                                        507,
+                                        106
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[2] M. Andreux, J. O. du Terrail, C. Beguier, and E. W. Tramel. Siloed federated learning for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                104,
+                                507,
+                                115
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        104,
+                                        507,
+                                        115
+                                    ],
+                                    "score": 1.0,
+                                    "content": "multi-centric histopathology datasets. In S. Albarqouni, S. Bakas, K. Kamnitsas, M. J. Cardoso,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                113,
+                                506,
+                                127
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        113,
+                                        506,
+                                        127
+                                    ],
+                                    "score": 1.0,
+                                    "content": "B. Landman, W. Li, F. Milletari, N. Rieke, H. Roth, D. Xu, and Z. Xu, editors, Domain",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                124,
+                                121,
+                                507,
+                                141
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        124,
+                                        121,
+                                        507,
+                                        141
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Adaptation and Representation Transfer, and Distributed and Collaborative Learning, pages",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                134,
+                                469,
+                                150
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        134,
+                                        469,
+                                        150
+                                    ],
+                                    "score": 1.0,
+                                    "content": "129–139, Cham, 2020. Springer International Publishing. ISBN 978-3-030-60548-3.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                110,
+                                155,
+                                505,
+                                169
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        110,
+                                        155,
+                                        505,
+                                        169
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[3] D. Bau, H. Strobelt, W. Peebles, J. Wulff, B. Zhou, J.-Y. Zhu, and A. Torralba. Semantic photo",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                125,
+                                164,
+                                506,
+                                180
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        125,
+                                        164,
+                                        506,
+                                        180
+                                    ],
+                                    "score": 1.0,
+                                    "content": "manipulation with a generative image prior. ACM Trans. Graph., 38(4), July 2019. ISSN",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                175,
+                                507,
+                                190
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        175,
+                                        507,
+                                        190
+                                    ],
+                                    "score": 1.0,
+                                    "content": "0730-0301. doi: 10.1145/3306346.3323023. URL https://doi.org/10.1145/3306346.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                189,
+                                170,
+                                200
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        189,
+                                        170,
+                                        200
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3323023.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                110,
+                                206,
+                                505,
+                                220
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        110,
+                                        206,
+                                        505,
+                                        220
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[4] D. Bau, J.-Y. Zhu, J. Wulff, W. Peebles, H. Strobelt, B. Zhou, and A. Torralba. Seeing what a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                217,
+                                506,
+                                232
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        217,
+                                        506,
+                                        232
+                                    ],
+                                    "score": 1.0,
+                                    "content": "gan cannot generate. In Proceedings of the IEEE/CVF International Conference on Computer",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                128,
+                                230,
+                                249,
+                                241
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        128,
+                                        230,
+                                        249,
+                                        241
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Vision (ICCV), October 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                109,
+                                246,
+                                507,
+                                263
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        109,
+                                        246,
+                                        507,
+                                        263
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[5] T. S. Brisimi, R. Chen, T. Mela, A. Olshevsky, I. C. Paschalidis, and W. Shi. Federated learning",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                128,
+                                259,
+                                505,
+                                271
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        128,
+                                        259,
+                                        505,
+                                        271
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of predictive models from federated electronic health records. International Journal of Medical",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                271,
+                                230,
+                                282
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        271,
+                                        230,
+                                        282
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Informatics(IJMI), 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                110,
+                                289,
+                                505,
+                                303
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        110,
+                                        289,
+                                        165,
+                                        301
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[6] DCGM.",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        190,
+                                        289,
+                                        319,
+                                        303
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ffhq-features-dataset, 2019.",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        345,
+                                        289,
+                                        505,
+                                        303
+                                    ],
+                                    "score": 1.0,
+                                    "content": "URL https://github.com/DCGM/",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                300,
+                                245,
+                                314
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        300,
+                                        245,
+                                        314
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ffhq-features-dataset.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                110,
+                                319,
+                                505,
+                                333
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        110,
+                                        319,
+                                        505,
+                                        333
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[7] L. Fan, K. W. Ng, C. Ju, T. Zhang, C. Liu, C. S. Chan, and Q. Yang. Rethinking privacy",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                125,
+                                329,
+                                507,
+                                344
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        125,
+                                        329,
+                                        507,
+                                        344
+                                    ],
+                                    "score": 1.0,
+                                    "content": "preserving deep learning: How to evaluate and thwart privacy attacks. In Q. Yang, L. Fan, and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                341,
+                                484,
+                                355
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        341,
+                                        484,
+                                        355
+                                    ],
+                                    "score": 1.0,
+                                    "content": "H. Yu, editors, Federated Learning: Privacy and Incentive, pages 32–50. Springer, 2020.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                110,
+                                360,
+                                505,
+                                374
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        110,
+                                        360,
+                                        505,
+                                        374
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[8] C. Finn, P. Abbeel, and S. Levine. Model-Agnostic Meta-Learning for Fast Adaptation of Deep",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                370,
+                                425,
+                                385
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        370,
+                                        425,
+                                        385
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Networks. International Conference on Machine Learning (ICML), 2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                110,
+                                390,
+                                506,
+                                404
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        110,
+                                        390,
+                                        506,
+                                        404
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[9] J. Geiping, H. Bauermeister, H. Dröge, and M. Moeller. Inverting gradients–How easy is it",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                402,
+                                506,
+                                414
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        402,
+                                        506,
+                                        414
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to break privacy in federated learning? In Neural Information Processing Systems(NeurIPS),",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                411,
+                                154,
+                                426
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        411,
+                                        154,
+                                        426
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2020.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                431,
+                                506,
+                                445
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        431,
+                                        506,
+                                        445
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[10] K. He, X. Zhang, S. Ren, and J. Sun. Deep residual learning for image recognition. In",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                443,
+                                507,
+                                457
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        443,
+                                        507,
+                                        457
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR),",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                453,
+                                154,
+                                466
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        453,
+                                        154,
+                                        466
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2016.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                473,
+                                505,
+                                487
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        473,
+                                        505,
+                                        487
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[11] B. Hitaj, G. Ateniese, and F. Perez-Cruz. Deep models under the GAN: Information leakage",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                483,
+                                505,
+                                497
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        483,
+                                        505,
+                                        497
+                                    ],
+                                    "score": 1.0,
+                                    "content": "from collaborative deep learning. In Proceedings of the 2017 ACM SIGSAC Conference on",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                494,
+                                348,
+                                508
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        494,
+                                        348,
+                                        508
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Computer and Communications Security (CCS), 2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                514,
+                                505,
+                                528
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        514,
+                                        505,
+                                        528
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[12] T. Karras, S. Laine, and T. Aila. A style-based generator architecture for generative adversarial",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                525,
+                                506,
+                                538
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        525,
+                                        506,
+                                        538
+                                    ],
+                                    "score": 1.0,
+                                    "content": "networks. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                536,
+                                260,
+                                547
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        536,
+                                        260,
+                                        547
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Recognition (CVPR), June 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                553,
+                                505,
+                                569
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        553,
+                                        505,
+                                        569
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[13] T. Karras, S. Laine, M. Aittala, J. Hellsten, J. Lehtinen, and T. Aila. Analyzing and improving",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                566,
+                                505,
+                                579
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        566,
+                                        505,
+                                        579
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the image quality of StyleGAN. In Proceedings of the IEEE/CVF Conference on Computer",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                576,
+                                313,
+                                589
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        576,
+                                        313,
+                                        589
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Vision and Pattern Recognition(CVPR), 2020.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                596,
+                                506,
+                                610
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        596,
+                                        506,
+                                        610
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[14] D. P. Kingma and J. Ba. Adam: A method for stochastic optimization. In Y. Bengio and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                607,
+                                507,
+                                621
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        607,
+                                        507,
+                                        621
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Y. LeCun, editors, 3rd International Conference on Learning Representations, ICLR 2015,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                616,
+                                507,
+                                632
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        616,
+                                        507,
+                                        632
+                                    ],
+                                    "score": 1.0,
+                                    "content": "San Diego, CA, USA, May 7-9, 2015, Conference Track Proceedings, 2015. URL http:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                629,
+                                264,
+                                642
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        629,
+                                        264,
+                                        642
+                                    ],
+                                    "score": 1.0,
+                                    "content": "//arxiv.org/abs/1412.6980.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                647,
+                                505,
+                                661
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        647,
+                                        505,
+                                        661
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[15] X. Li, M. Jiang, X. Zhang, M. Kamp, and Q. Dou. FedBN: Federated learning on non-IID",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                658,
+                                506,
+                                673
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        658,
+                                        506,
+                                        673
+                                    ],
+                                    "score": 1.0,
+                                    "content": "features via local batch normalization. In International Conference on Learning Representations",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                669,
+                                186,
+                                683
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        669,
+                                        186,
+                                        683
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(ICLR), 2021.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                687,
+                                507,
+                                703
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        687,
+                                        507,
+                                        703
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[16] Y. Lin, S. Han, H. Mao, Y. Wang, and W. J. Dally. Deep Gradient Compression: Reducing",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                700,
+                                506,
+                                714
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        700,
+                                        506,
+                                        714
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the communication bandwidth for distributed training. In The International Conference on",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                712,
+                                290,
+                                723
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        712,
+                                        290,
+                                        723
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Learning Representations (ICLR), 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 47
+                        }
+                    ],
+                    "index": 23.5
+                }
+            ],
+            "page_idx": 9,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        300,
+                        741,
+                        311,
+                        750
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                299,
+                                740,
+                                313,
+                                754
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        299,
+                                        740,
+                                        313,
+                                        754
+                                    ],
+                                    "score": 1.0,
+                                    "content": "10",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "list",
+                    "bbox": [
+                        105,
+                        57,
+                        507,
+                        724
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                126,
+                                72,
+                                174,
+                                86
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        72,
+                                        174,
+                                        86
+                                    ],
+                                    "score": 1.0,
+                                    "content": "June 2020.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                109,
+                                90,
+                                507,
+                                106
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        109,
+                                        90,
+                                        507,
+                                        106
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[2] M. Andreux, J. O. du Terrail, C. Beguier, and E. W. Tramel. Siloed federated learning for",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                104,
+                                507,
+                                115
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        104,
+                                        507,
+                                        115
+                                    ],
+                                    "score": 1.0,
+                                    "content": "multi-centric histopathology datasets. In S. Albarqouni, S. Bakas, K. Kamnitsas, M. J. Cardoso,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                113,
+                                506,
+                                127
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        113,
+                                        506,
+                                        127
+                                    ],
+                                    "score": 1.0,
+                                    "content": "B. Landman, W. Li, F. Milletari, N. Rieke, H. Roth, D. Xu, and Z. Xu, editors, Domain",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                124,
+                                121,
+                                507,
+                                141
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        124,
+                                        121,
+                                        507,
+                                        141
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Adaptation and Representation Transfer, and Distributed and Collaborative Learning, pages",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                134,
+                                469,
+                                150
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        134,
+                                        469,
+                                        150
+                                    ],
+                                    "score": 1.0,
+                                    "content": "129–139, Cham, 2020. Springer International Publishing. ISBN 978-3-030-60548-3.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                110,
+                                155,
+                                505,
+                                169
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        110,
+                                        155,
+                                        505,
+                                        169
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[3] D. Bau, H. Strobelt, W. Peebles, J. Wulff, B. Zhou, J.-Y. Zhu, and A. Torralba. Semantic photo",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                125,
+                                164,
+                                506,
+                                180
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        125,
+                                        164,
+                                        506,
+                                        180
+                                    ],
+                                    "score": 1.0,
+                                    "content": "manipulation with a generative image prior. ACM Trans. Graph., 38(4), July 2019. ISSN",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                175,
+                                507,
+                                190
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        175,
+                                        507,
+                                        190
+                                    ],
+                                    "score": 1.0,
+                                    "content": "0730-0301. doi: 10.1145/3306346.3323023. URL https://doi.org/10.1145/3306346.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                189,
+                                170,
+                                200
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        189,
+                                        170,
+                                        200
+                                    ],
+                                    "score": 1.0,
+                                    "content": "3323023.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                110,
+                                206,
+                                505,
+                                220
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        110,
+                                        206,
+                                        505,
+                                        220
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[4] D. Bau, J.-Y. Zhu, J. Wulff, W. Peebles, H. Strobelt, B. Zhou, and A. Torralba. Seeing what a",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                217,
+                                506,
+                                232
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        217,
+                                        506,
+                                        232
+                                    ],
+                                    "score": 1.0,
+                                    "content": "gan cannot generate. In Proceedings of the IEEE/CVF International Conference on Computer",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                128,
+                                230,
+                                249,
+                                241
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        128,
+                                        230,
+                                        249,
+                                        241
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Vision (ICCV), October 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                109,
+                                246,
+                                507,
+                                263
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        109,
+                                        246,
+                                        507,
+                                        263
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[5] T. S. Brisimi, R. Chen, T. Mela, A. Olshevsky, I. C. Paschalidis, and W. Shi. Federated learning",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                128,
+                                259,
+                                505,
+                                271
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        128,
+                                        259,
+                                        505,
+                                        271
+                                    ],
+                                    "score": 1.0,
+                                    "content": "of predictive models from federated electronic health records. International Journal of Medical",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                271,
+                                230,
+                                282
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        271,
+                                        230,
+                                        282
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Informatics(IJMI), 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                110,
+                                289,
+                                505,
+                                303
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        110,
+                                        289,
+                                        165,
+                                        301
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[6] DCGM.",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        190,
+                                        289,
+                                        319,
+                                        303
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ffhq-features-dataset, 2019.",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        345,
+                                        289,
+                                        505,
+                                        303
+                                    ],
+                                    "score": 1.0,
+                                    "content": "URL https://github.com/DCGM/",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                300,
+                                245,
+                                314
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        300,
+                                        245,
+                                        314
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ffhq-features-dataset.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                110,
+                                319,
+                                505,
+                                333
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        110,
+                                        319,
+                                        505,
+                                        333
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[7] L. Fan, K. W. Ng, C. Ju, T. Zhang, C. Liu, C. S. Chan, and Q. Yang. Rethinking privacy",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                125,
+                                329,
+                                507,
+                                344
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        125,
+                                        329,
+                                        507,
+                                        344
+                                    ],
+                                    "score": 1.0,
+                                    "content": "preserving deep learning: How to evaluate and thwart privacy attacks. In Q. Yang, L. Fan, and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                341,
+                                484,
+                                355
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        341,
+                                        484,
+                                        355
+                                    ],
+                                    "score": 1.0,
+                                    "content": "H. Yu, editors, Federated Learning: Privacy and Incentive, pages 32–50. Springer, 2020.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                110,
+                                360,
+                                505,
+                                374
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        110,
+                                        360,
+                                        505,
+                                        374
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[8] C. Finn, P. Abbeel, and S. Levine. Model-Agnostic Meta-Learning for Fast Adaptation of Deep",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                370,
+                                425,
+                                385
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        370,
+                                        425,
+                                        385
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Networks. International Conference on Machine Learning (ICML), 2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                110,
+                                390,
+                                506,
+                                404
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        110,
+                                        390,
+                                        506,
+                                        404
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[9] J. Geiping, H. Bauermeister, H. Dröge, and M. Moeller. Inverting gradients–How easy is it",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                402,
+                                506,
+                                414
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        402,
+                                        506,
+                                        414
+                                    ],
+                                    "score": 1.0,
+                                    "content": "to break privacy in federated learning? In Neural Information Processing Systems(NeurIPS),",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                411,
+                                154,
+                                426
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        411,
+                                        154,
+                                        426
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2020.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                431,
+                                506,
+                                445
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        431,
+                                        506,
+                                        445
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[10] K. He, X. Zhang, S. Ren, and J. Sun. Deep residual learning for image recognition. In",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                443,
+                                507,
+                                457
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        443,
+                                        507,
+                                        457
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR),",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                453,
+                                154,
+                                466
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        453,
+                                        154,
+                                        466
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2016.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                473,
+                                505,
+                                487
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        473,
+                                        505,
+                                        487
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[11] B. Hitaj, G. Ateniese, and F. Perez-Cruz. Deep models under the GAN: Information leakage",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                483,
+                                505,
+                                497
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        483,
+                                        505,
+                                        497
+                                    ],
+                                    "score": 1.0,
+                                    "content": "from collaborative deep learning. In Proceedings of the 2017 ACM SIGSAC Conference on",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                494,
+                                348,
+                                508
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        494,
+                                        348,
+                                        508
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Computer and Communications Security (CCS), 2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                514,
+                                505,
+                                528
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        514,
+                                        505,
+                                        528
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[12] T. Karras, S. Laine, and T. Aila. A style-based generator architecture for generative adversarial",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                525,
+                                506,
+                                538
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        525,
+                                        506,
+                                        538
+                                    ],
+                                    "score": 1.0,
+                                    "content": "networks. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                536,
+                                260,
+                                547
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        536,
+                                        260,
+                                        547
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Recognition (CVPR), June 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                553,
+                                505,
+                                569
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        553,
+                                        505,
+                                        569
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[13] T. Karras, S. Laine, M. Aittala, J. Hellsten, J. Lehtinen, and T. Aila. Analyzing and improving",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                566,
+                                505,
+                                579
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        566,
+                                        505,
+                                        579
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the image quality of StyleGAN. In Proceedings of the IEEE/CVF Conference on Computer",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                576,
+                                313,
+                                589
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        576,
+                                        313,
+                                        589
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Vision and Pattern Recognition(CVPR), 2020.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                596,
+                                506,
+                                610
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        596,
+                                        506,
+                                        610
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[14] D. P. Kingma and J. Ba. Adam: A method for stochastic optimization. In Y. Bengio and",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                607,
+                                507,
+                                621
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        607,
+                                        507,
+                                        621
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Y. LeCun, editors, 3rd International Conference on Learning Representations, ICLR 2015,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                616,
+                                507,
+                                632
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        616,
+                                        507,
+                                        632
+                                    ],
+                                    "score": 1.0,
+                                    "content": "San Diego, CA, USA, May 7-9, 2015, Conference Track Proceedings, 2015. URL http:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                629,
+                                264,
+                                642
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        629,
+                                        264,
+                                        642
+                                    ],
+                                    "score": 1.0,
+                                    "content": "//arxiv.org/abs/1412.6980.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                647,
+                                505,
+                                661
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        647,
+                                        505,
+                                        661
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[15] X. Li, M. Jiang, X. Zhang, M. Kamp, and Q. Dou. FedBN: Federated learning on non-IID",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                658,
+                                506,
+                                673
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        658,
+                                        506,
+                                        673
+                                    ],
+                                    "score": 1.0,
+                                    "content": "features via local batch normalization. In International Conference on Learning Representations",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                669,
+                                186,
+                                683
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        669,
+                                        186,
+                                        683
+                                    ],
+                                    "score": 1.0,
+                                    "content": "(ICLR), 2021.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                687,
+                                507,
+                                703
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        687,
+                                        507,
+                                        703
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[16] Y. Lin, S. Han, H. Mao, Y. Wang, and W. J. Dally. Deep Gradient Compression: Reducing",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                700,
+                                506,
+                                714
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        700,
+                                        506,
+                                        714
+                                    ],
+                                    "score": 1.0,
+                                    "content": "the communication bandwidth for distributed training. In The International Conference on",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                712,
+                                290,
+                                723
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        712,
+                                        290,
+                                        723
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Learning Representations (ICLR), 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 47,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                73,
+                                506,
+                                86
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        73,
+                                        506,
+                                        86
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[17] B. McMahan, E. Moore, D. Ramage, S. Hampson, and B. A. y Arcas. Communication-efficient",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 0,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                83,
+                                507,
+                                97
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        83,
+                                        507,
+                                        97
+                                    ],
+                                    "score": 1.0,
+                                    "content": "learning of deep networks from decentralized data. In A. Singh and J. Zhu, editors, Proceed-",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                94,
+                                507,
+                                108
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        94,
+                                        507,
+                                        108
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ings of the 20th International Conference on Artificial Intelligence and Statistics(AISTATS),",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                105,
+                                463,
+                                119
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        105,
+                                        463,
+                                        119
+                                    ],
+                                    "score": 1.0,
+                                    "content": "volume 54 of Proceedings of Machine Learning Research, pages 1273–1282, 2017.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 3,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                123,
+                                506,
+                                137
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        123,
+                                        506,
+                                        137
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[18] H. B. McMahan, D. Ramage, K. Talwar, and L. Zhang. Learning differentially private recurrent",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 4,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                125,
+                                133,
+                                507,
+                                149
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        125,
+                                        133,
+                                        507,
+                                        149
+                                    ],
+                                    "score": 1.0,
+                                    "content": "language models. In 6th International Conference on Learning Representations (ICLR), 2018.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                151,
+                                506,
+                                165
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        151,
+                                        506,
+                                        165
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[19] L. Melis, C. Song, E. De Cristofaro, and V. Shmatikov. Exploiting unintended feature leakage",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 6,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                162,
+                                507,
+                                176
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        162,
+                                        376,
+                                        176
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in collaborative learning. In IEEE Symp. Security and Privacy",
+                                    "type": "text",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        377,
+                                        163,
+                                        394,
+                                        173
+                                    ],
+                                    "score": 0.4,
+                                    "content": "( S P )",
+                                    "type": "inline_equation",
+                                    "cross_page": true
+                                },
+                                {
+                                    "bbox": [
+                                        394,
+                                        162,
+                                        507,
+                                        176
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", pages 691–706, 2019. doi:",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                173,
+                                227,
+                                185
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        173,
+                                        227,
+                                        185
+                                    ],
+                                    "score": 1.0,
+                                    "content": "10.1109/SP.2019.00029.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 8,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                191,
+                                506,
+                                205
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        191,
+                                        506,
+                                        205
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[20] A. Nichol, J. Achiam, and J. Schulman. On first-order meta-learning algorithms. arXiv preprint",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 9,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                202,
+                                231,
+                                215
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        202,
+                                        231,
+                                        215
+                                    ],
+                                    "score": 1.0,
+                                    "content": "arXiv:1803.02999, 2018.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 10,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                219,
+                                506,
+                                234
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        219,
+                                        506,
+                                        234
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[21] L. T. Phong, Y. Aono, T. Hayashi, L. Wang, and S. Moriai. Privacy-preserving deep learning via",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 11,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                229,
+                                507,
+                                245
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        229,
+                                        507,
+                                        245
+                                    ],
+                                    "score": 1.0,
+                                    "content": "additively homomorphic encryption. IEEE Transactions on Information Forensics and Security,",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                240,
+                                227,
+                                254
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        240,
+                                        227,
+                                        254
+                                    ],
+                                    "score": 1.0,
+                                    "content": "13(5):1333–1345, 2018.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 13,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                258,
+                                507,
+                                273
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        258,
+                                        507,
+                                        273
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[22] O. Russakovsky, J. Deng, H. Su, J. Krause, S. Satheesh, S. Ma, Z. Huang, A. Karpathy,",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 14,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                270,
+                                506,
+                                284
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        270,
+                                        506,
+                                        284
+                                    ],
+                                    "score": 1.0,
+                                    "content": "A. Khosla, M. Bernstein, A. C. Berg, and L. Fei-Fei. ImageNet Large Scale Visual Recognition",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                281,
+                                507,
+                                295
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        281,
+                                        507,
+                                        295
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Challenge. International Journal of Computer Vision (IJCV), 115(3):211–252, 2015. doi:",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                290,
+                                249,
+                                307
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        290,
+                                        249,
+                                        307
+                                    ],
+                                    "score": 1.0,
+                                    "content": "10.1007/s11263-015-0816-y.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 17,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                308,
+                                507,
+                                324
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        308,
+                                        507,
+                                        324
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[23] A. Segal, A. Marcedone, B. Kreuter, D. Ramage, H. B. McMahan, K. Seth, K. A. Bonawitz,",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 18,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                125,
+                                319,
+                                506,
+                                335
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        125,
+                                        319,
+                                        506,
+                                        335
+                                    ],
+                                    "score": 1.0,
+                                    "content": "S. Patel, and V. Ivanov. Practical secure aggregation for privacy-preserving machine learning.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                330,
+                                404,
+                                345
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        330,
+                                        404,
+                                        345
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In CCS, 2017. URL https://eprint.iacr.org/2017/281.pdf.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 20,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                349,
+                                506,
+                                363
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        349,
+                                        506,
+                                        363
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[24] R. Shokri, M. Stronati, C. Song, and V. Shmatikov. Membership inference attacks against",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 21,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                360,
+                                507,
+                                374
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        360,
+                                        507,
+                                        374
+                                    ],
+                                    "score": 1.0,
+                                    "content": "machine learning models. In 2017 IEEE Symposium on Security and Privacy (SP), pages 3–18,",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                128,
+                                371,
+                                259,
+                                383
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        128,
+                                        371,
+                                        259,
+                                        383
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2017. doi: 10.1109/SP.2017.41.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 23,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                386,
+                                506,
+                                403
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        386,
+                                        506,
+                                        403
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[25] K. Simonyan and A. Zisserman. Very deep convolutional networks for large-scale image",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 24,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                399,
+                                243,
+                                413
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        399,
+                                        243,
+                                        413
+                                    ],
+                                    "score": 1.0,
+                                    "content": "recognition. In ICLR, 2015.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 25,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                416,
+                                506,
+                                432
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        416,
+                                        506,
+                                        432
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[26] D. Ulyanov, A. Vedaldi, and V. Lempitsky. Deep image prior. In Proceedings of the IEEE/CVF",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 26,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                428,
+                                441,
+                                442
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        428,
+                                        441,
+                                        442
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Conference on Computer Vision and Pattern Recognition (CVPR), June 2018.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 27,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                445,
+                                507,
+                                459
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        445,
+                                        507,
+                                        459
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[27] Z. Wang, M. Song, Z. Zhang, Y. Song, Q. Wang, and H. Qi. Beyond inferring class representa-",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 28,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                456,
+                                506,
+                                470
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        456,
+                                        506,
+                                        470
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tives: User-level privacy leakage from federated learning. In IEEE INFOCOM 2019 - IEEE",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                468,
+                                403,
+                                482
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        468,
+                                        403,
+                                        482
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Conference on Computer Communications, pages 2512–2520, 2019.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 30,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                484,
+                                506,
+                                499
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        484,
+                                        506,
+                                        499
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[28] W. Wei, L. Liu, M. Loper, K.-H. Chow, M. E. Gursoy, S. Truex, and Y. Wu. A framework",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 31,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                124,
+                                494,
+                                507,
+                                511
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        124,
+                                        494,
+                                        507,
+                                        511
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for evaluating client privacy leakages in federated learning. In L. Chen, N. Li, K. Liang,",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                506,
+                                506,
+                                520
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        506,
+                                        506,
+                                        520
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and S. Schneider, editors, Computer Security – ESORICS 2020, pages 545–566, Cham, 2020.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                516,
+                                375,
+                                532
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        516,
+                                        375,
+                                        532
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Springer International Publishing. ISBN 978-3-030-58951-6.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 34,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                536,
+                                505,
+                                548
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        536,
+                                        505,
+                                        548
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[29] H. Yin, A. Mallya, A. Vahdat, J. M. Alvarez, J. Kautz, and P. Molchanov. See through gradients:",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 35,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                546,
+                                506,
+                                560
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        546,
+                                        506,
+                                        560
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Image batch recovery via gradinversion. In Proceedings of the IEEE Conference on Computer",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                557,
+                                362,
+                                571
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        557,
+                                        362,
+                                        571
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Vision and Pattern Recognition Workshops (CVPR), 2021.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 37,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                574,
+                                506,
+                                588
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        574,
+                                        506,
+                                        588
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[30] R. Zhang, P. Isola, A. A. Efros, E. Shechtman, and O. Wang. The unreasonable effectiveness of",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 38,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                585,
+                                506,
+                                600
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        585,
+                                        506,
+                                        600
+                                    ],
+                                    "score": 1.0,
+                                    "content": "deep features as a perceptual metric. In Proceedings of the IEEE/CVF Conference on Computer",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                596,
+                                316,
+                                610
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        596,
+                                        316,
+                                        610
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Vision and Pattern Recognition (CVPR), 2018.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 40,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                614,
+                                505,
+                                628
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        614,
+                                        505,
+                                        628
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[31] B. Zhao, K. R. Mopuri, and H. Bilen. iDLG: Improved deep leakage from gradients. arXiv",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 41,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                124,
+                                624,
+                                267,
+                                639
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        124,
+                                        624,
+                                        267,
+                                        639
+                                    ],
+                                    "score": 1.0,
+                                    "content": "preprint arXiv:2001.02610, 2020.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 42,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                642,
+                                506,
+                                656
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        642,
+                                        506,
+                                        656
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[32] J. Zhu and M. B. Blaschko. R-GAP: Recursive gradient attack on privacy. In International",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 43,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                653,
+                                352,
+                                667
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        653,
+                                        352,
+                                        667
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Conference on Learning Representations (ICLR), 2021.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 44,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                671,
+                                506,
+                                685
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        671,
+                                        506,
+                                        685
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[33] J. Zhu, Y. Shen, D. Zhao, and B. Zhou. In-domain gan inversion for real image editing. In",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 45,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                683,
+                                423,
+                                695
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        683,
+                                        423,
+                                        695
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Proceedings of European Conference on Computer Vision (ECCV), 2020.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 46,
+                            "is_list_end_line": true
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                699,
+                                506,
+                                713
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        699,
+                                        506,
+                                        713
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[34] L. Zhu, Z. Liu, and S. Han. Deep leakage from gradients. In Annual Conference on Neural",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 47,
+                            "is_list_start_line": true
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                711,
+                                329,
+                                724
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        711,
+                                        329,
+                                        724
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Information Processing Systems (NeurIPS), 2019.",
+                                    "type": "text",
+                                    "cross_page": true
+                                }
+                            ],
+                            "index": 48,
+                            "is_list_end_line": true
+                        }
+                    ],
+                    "index": 23.5,
+                    "bbox_fs": [
+                        105,
+                        72,
+                        507,
+                        723
+                    ]
+                }
+            ]
+        },
+        {
+            "preproc_blocks": [
+                {
+                    "type": "text",
+                    "bbox": [
+                        105,
+                        56,
+                        507,
+                        722
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                106,
+                                73,
+                                506,
+                                86
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        73,
+                                        506,
+                                        86
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[17] B. McMahan, E. Moore, D. Ramage, S. Hampson, and B. A. y Arcas. Communication-efficient",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 0
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                83,
+                                507,
+                                97
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        83,
+                                        507,
+                                        97
+                                    ],
+                                    "score": 1.0,
+                                    "content": "learning of deep networks from decentralized data. In A. Singh and J. Zhu, editors, Proceed-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 1
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                94,
+                                507,
+                                108
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        94,
+                                        507,
+                                        108
+                                    ],
+                                    "score": 1.0,
+                                    "content": "ings of the 20th International Conference on Artificial Intelligence and Statistics(AISTATS),",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 2
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                105,
+                                463,
+                                119
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        105,
+                                        463,
+                                        119
+                                    ],
+                                    "score": 1.0,
+                                    "content": "volume 54 of Proceedings of Machine Learning Research, pages 1273–1282, 2017.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 3
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                123,
+                                506,
+                                137
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        123,
+                                        506,
+                                        137
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[18] H. B. McMahan, D. Ramage, K. Talwar, and L. Zhang. Learning differentially private recurrent",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 4
+                        },
+                        {
+                            "bbox": [
+                                125,
+                                133,
+                                507,
+                                149
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        125,
+                                        133,
+                                        507,
+                                        149
+                                    ],
+                                    "score": 1.0,
+                                    "content": "language models. In 6th International Conference on Learning Representations (ICLR), 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 5
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                151,
+                                506,
+                                165
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        151,
+                                        506,
+                                        165
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[19] L. Melis, C. Song, E. De Cristofaro, and V. Shmatikov. Exploiting unintended feature leakage",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 6
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                162,
+                                507,
+                                176
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        162,
+                                        376,
+                                        176
+                                    ],
+                                    "score": 1.0,
+                                    "content": "in collaborative learning. In IEEE Symp. Security and Privacy",
+                                    "type": "text"
+                                },
+                                {
+                                    "bbox": [
+                                        377,
+                                        163,
+                                        394,
+                                        173
+                                    ],
+                                    "score": 0.4,
+                                    "content": "( S P )",
+                                    "type": "inline_equation"
+                                },
+                                {
+                                    "bbox": [
+                                        394,
+                                        162,
+                                        507,
+                                        176
+                                    ],
+                                    "score": 1.0,
+                                    "content": ", pages 691–706, 2019. doi:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 7
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                173,
+                                227,
+                                185
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        173,
+                                        227,
+                                        185
+                                    ],
+                                    "score": 1.0,
+                                    "content": "10.1109/SP.2019.00029.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 8
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                191,
+                                506,
+                                205
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        191,
+                                        506,
+                                        205
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[20] A. Nichol, J. Achiam, and J. Schulman. On first-order meta-learning algorithms. arXiv preprint",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 9
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                202,
+                                231,
+                                215
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        202,
+                                        231,
+                                        215
+                                    ],
+                                    "score": 1.0,
+                                    "content": "arXiv:1803.02999, 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 10
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                219,
+                                506,
+                                234
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        219,
+                                        506,
+                                        234
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[21] L. T. Phong, Y. Aono, T. Hayashi, L. Wang, and S. Moriai. Privacy-preserving deep learning via",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 11
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                229,
+                                507,
+                                245
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        229,
+                                        507,
+                                        245
+                                    ],
+                                    "score": 1.0,
+                                    "content": "additively homomorphic encryption. IEEE Transactions on Information Forensics and Security,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 12
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                240,
+                                227,
+                                254
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        240,
+                                        227,
+                                        254
+                                    ],
+                                    "score": 1.0,
+                                    "content": "13(5):1333–1345, 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 13
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                258,
+                                507,
+                                273
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        258,
+                                        507,
+                                        273
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[22] O. Russakovsky, J. Deng, H. Su, J. Krause, S. Satheesh, S. Ma, Z. Huang, A. Karpathy,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 14
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                270,
+                                506,
+                                284
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        270,
+                                        506,
+                                        284
+                                    ],
+                                    "score": 1.0,
+                                    "content": "A. Khosla, M. Bernstein, A. C. Berg, and L. Fei-Fei. ImageNet Large Scale Visual Recognition",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 15
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                281,
+                                507,
+                                295
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        281,
+                                        507,
+                                        295
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Challenge. International Journal of Computer Vision (IJCV), 115(3):211–252, 2015. doi:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 16
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                290,
+                                249,
+                                307
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        290,
+                                        249,
+                                        307
+                                    ],
+                                    "score": 1.0,
+                                    "content": "10.1007/s11263-015-0816-y.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 17
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                308,
+                                507,
+                                324
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        308,
+                                        507,
+                                        324
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[23] A. Segal, A. Marcedone, B. Kreuter, D. Ramage, H. B. McMahan, K. Seth, K. A. Bonawitz,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 18
+                        },
+                        {
+                            "bbox": [
+                                125,
+                                319,
+                                506,
+                                335
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        125,
+                                        319,
+                                        506,
+                                        335
+                                    ],
+                                    "score": 1.0,
+                                    "content": "S. Patel, and V. Ivanov. Practical secure aggregation for privacy-preserving machine learning.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 19
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                330,
+                                404,
+                                345
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        330,
+                                        404,
+                                        345
+                                    ],
+                                    "score": 1.0,
+                                    "content": "In CCS, 2017. URL https://eprint.iacr.org/2017/281.pdf.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 20
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                349,
+                                506,
+                                363
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        349,
+                                        506,
+                                        363
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[24] R. Shokri, M. Stronati, C. Song, and V. Shmatikov. Membership inference attacks against",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 21
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                360,
+                                507,
+                                374
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        360,
+                                        507,
+                                        374
+                                    ],
+                                    "score": 1.0,
+                                    "content": "machine learning models. In 2017 IEEE Symposium on Security and Privacy (SP), pages 3–18,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 22
+                        },
+                        {
+                            "bbox": [
+                                128,
+                                371,
+                                259,
+                                383
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        128,
+                                        371,
+                                        259,
+                                        383
+                                    ],
+                                    "score": 1.0,
+                                    "content": "2017. doi: 10.1109/SP.2017.41.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 23
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                386,
+                                506,
+                                403
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        386,
+                                        506,
+                                        403
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[25] K. Simonyan and A. Zisserman. Very deep convolutional networks for large-scale image",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 24
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                399,
+                                243,
+                                413
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        399,
+                                        243,
+                                        413
+                                    ],
+                                    "score": 1.0,
+                                    "content": "recognition. In ICLR, 2015.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 25
+                        },
+                        {
+                            "bbox": [
+                                105,
+                                416,
+                                506,
+                                432
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        105,
+                                        416,
+                                        506,
+                                        432
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[26] D. Ulyanov, A. Vedaldi, and V. Lempitsky. Deep image prior. In Proceedings of the IEEE/CVF",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 26
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                428,
+                                441,
+                                442
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        428,
+                                        441,
+                                        442
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Conference on Computer Vision and Pattern Recognition (CVPR), June 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 27
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                445,
+                                507,
+                                459
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        445,
+                                        507,
+                                        459
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[27] Z. Wang, M. Song, Z. Zhang, Y. Song, Q. Wang, and H. Qi. Beyond inferring class representa-",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 28
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                456,
+                                506,
+                                470
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        456,
+                                        506,
+                                        470
+                                    ],
+                                    "score": 1.0,
+                                    "content": "tives: User-level privacy leakage from federated learning. In IEEE INFOCOM 2019 - IEEE",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 29
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                468,
+                                403,
+                                482
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        468,
+                                        403,
+                                        482
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Conference on Computer Communications, pages 2512–2520, 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 30
+                        },
+                        {
+                            "bbox": [
+                                104,
+                                484,
+                                506,
+                                499
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        104,
+                                        484,
+                                        506,
+                                        499
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[28] W. Wei, L. Liu, M. Loper, K.-H. Chow, M. E. Gursoy, S. Truex, and Y. Wu. A framework",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 31
+                        },
+                        {
+                            "bbox": [
+                                124,
+                                494,
+                                507,
+                                511
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        124,
+                                        494,
+                                        507,
+                                        511
+                                    ],
+                                    "score": 1.0,
+                                    "content": "for evaluating client privacy leakages in federated learning. In L. Chen, N. Li, K. Liang,",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 32
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                506,
+                                506,
+                                520
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        506,
+                                        506,
+                                        520
+                                    ],
+                                    "score": 1.0,
+                                    "content": "and S. Schneider, editors, Computer Security – ESORICS 2020, pages 545–566, Cham, 2020.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 33
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                516,
+                                375,
+                                532
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        516,
+                                        375,
+                                        532
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Springer International Publishing. ISBN 978-3-030-58951-6.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 34
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                536,
+                                505,
+                                548
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        536,
+                                        505,
+                                        548
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[29] H. Yin, A. Mallya, A. Vahdat, J. M. Alvarez, J. Kautz, and P. Molchanov. See through gradients:",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 35
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                546,
+                                506,
+                                560
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        546,
+                                        506,
+                                        560
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Image batch recovery via gradinversion. In Proceedings of the IEEE Conference on Computer",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 36
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                557,
+                                362,
+                                571
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        557,
+                                        362,
+                                        571
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Vision and Pattern Recognition Workshops (CVPR), 2021.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 37
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                574,
+                                506,
+                                588
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        574,
+                                        506,
+                                        588
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[30] R. Zhang, P. Isola, A. A. Efros, E. Shechtman, and O. Wang. The unreasonable effectiveness of",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 38
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                585,
+                                506,
+                                600
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        585,
+                                        506,
+                                        600
+                                    ],
+                                    "score": 1.0,
+                                    "content": "deep features as a perceptual metric. In Proceedings of the IEEE/CVF Conference on Computer",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 39
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                596,
+                                316,
+                                610
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        596,
+                                        316,
+                                        610
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Vision and Pattern Recognition (CVPR), 2018.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 40
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                614,
+                                505,
+                                628
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        614,
+                                        505,
+                                        628
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[31] B. Zhao, K. R. Mopuri, and H. Bilen. iDLG: Improved deep leakage from gradients. arXiv",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 41
+                        },
+                        {
+                            "bbox": [
+                                124,
+                                624,
+                                267,
+                                639
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        124,
+                                        624,
+                                        267,
+                                        639
+                                    ],
+                                    "score": 1.0,
+                                    "content": "preprint arXiv:2001.02610, 2020.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 42
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                642,
+                                506,
+                                656
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        642,
+                                        506,
+                                        656
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[32] J. Zhu and M. B. Blaschko. R-GAP: Recursive gradient attack on privacy. In International",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 43
+                        },
+                        {
+                            "bbox": [
+                                126,
+                                653,
+                                352,
+                                667
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        126,
+                                        653,
+                                        352,
+                                        667
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Conference on Learning Representations (ICLR), 2021.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 44
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                671,
+                                506,
+                                685
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        671,
+                                        506,
+                                        685
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[33] J. Zhu, Y. Shen, D. Zhao, and B. Zhou. In-domain gan inversion for real image editing. In",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 45
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                683,
+                                423,
+                                695
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        683,
+                                        423,
+                                        695
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Proceedings of European Conference on Computer Vision (ECCV), 2020.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 46
+                        },
+                        {
+                            "bbox": [
+                                106,
+                                699,
+                                506,
+                                713
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        106,
+                                        699,
+                                        506,
+                                        713
+                                    ],
+                                    "score": 1.0,
+                                    "content": "[34] L. Zhu, Z. Liu, and S. Han. Deep leakage from gradients. In Annual Conference on Neural",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 47
+                        },
+                        {
+                            "bbox": [
+                                127,
+                                711,
+                                329,
+                                724
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        127,
+                                        711,
+                                        329,
+                                        724
+                                    ],
+                                    "score": 1.0,
+                                    "content": "Information Processing Systems (NeurIPS), 2019.",
+                                    "type": "text"
+                                }
+                            ],
+                            "index": 48
+                        }
+                    ],
+                    "index": 24
+                }
+            ],
+            "page_idx": 10,
+            "page_size": [
+                612,
+                792
+            ],
+            "discarded_blocks": [
+                {
+                    "type": "discarded",
+                    "bbox": [
+                        300,
+                        741,
+                        311,
+                        751
+                    ],
+                    "lines": [
+                        {
+                            "bbox": [
+                                299,
+                                740,
+                                312,
+                                754
+                            ],
+                            "spans": [
+                                {
+                                    "bbox": [
+                                        299,
+                                        740,
+                                        312,
+                                        754
+                                    ],
+                                    "score": 1.0,
+                                    "content": "11",
+                                    "type": "text"
+                                }
+                            ]
+                        }
+                    ]
+                }
+            ],
+            "para_blocks": [
+                {
+                    "type": "list",
+                    "bbox": [
+                        105,
+                        56,
+                        507,
+                        722
+                    ],
+                    "lines": [],
+                    "index": 24,
+                    "bbox_fs": [
+                        104,
+                        73,
+                        507,
+                        724
+                    ],
+                    "lines_deleted": true
+                }
+            ]
+        }
+    ],
+    "_backend": "pipeline",
+    "_version_name": "2.2.2"
+}
\ No newline at end of file
diff --git a/parse/train/x9jS8pX3dkx/x9jS8pX3dkx_model.json b/parse/train/x9jS8pX3dkx/x9jS8pX3dkx_model.json
new file mode 100644
index 0000000000000000000000000000000000000000..383644752169a382eb2531d853497a674342b98a
--- /dev/null
+++ b/parse/train/x9jS8pX3dkx/x9jS8pX3dkx_model.json
@@ -0,0 +1,16896 @@
+[
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    398,
+                    750,
+                    1304,
+                    750,
+                    1304,
+                    1146,
+                    398,
+                    1146
+                ],
+                "score": 0.983
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1548,
+                    1404,
+                    1548,
+                    1404,
+                    1792,
+                    298,
+                    1792
+                ],
+                "score": 0.977
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    1275,
+                    1402,
+                    1275,
+                    1402,
+                    1396,
+                    299,
+                    1396
+                ],
+                "score": 0.974
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    1411,
+                    1403,
+                    1411,
+                    1403,
+                    1533,
+                    299,
+                    1533
+                ],
+                "score": 0.973
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    1807,
+                    1403,
+                    1807,
+                    1403,
+                    1929,
+                    299,
+                    1929
+                ],
+                "score": 0.968
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    317,
+                    441,
+                    1417,
+                    441,
+                    1417,
+                    602,
+                    317,
+                    602
+                ],
+                "score": 0.918
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    349,
+                    271,
+                    1352,
+                    271,
+                    1352,
+                    327,
+                    349,
+                    327
+                ],
+                "score": 0.911
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    299,
+                    1204,
+                    530,
+                    1204,
+                    530,
+                    1242,
+                    299,
+                    1242
+                ],
+                "score": 0.908
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    297,
+                    2033,
+                    1070,
+                    2033,
+                    1070,
+                    2062,
+                    297,
+                    2062
+                ],
+                "score": 0.892
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    788,
+                    680,
+                    913,
+                    680,
+                    913,
+                    717,
+                    788,
+                    717
+                ],
+                "score": 0.887
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    344,
+                    1952,
+                    545,
+                    1952,
+                    545,
+                    1979,
+                    344,
+                    1979
+                ],
+                "score": 0.868
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    321,
+                    216,
+                    1372,
+                    216,
+                    1372,
+                    233,
+                    321,
+                    233
+                ],
+                "score": 0.096
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    690,
+                    442,
+                    769,
+                    442,
+                    769,
+                    475,
+                    690,
+                    475
+                ],
+                "score": 0.86,
+                "latex": "\\mathbf { K i m ^ { 2 * } }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1094,
+                    442,
+                    1147,
+                    442,
+                    1147,
+                    475,
+                    1094,
+                    475
+                ],
+                "score": 0.83,
+                "latex": "\\mathbf { O h ^ { 4 } }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1269,
+                    441,
+                    1340,
+                    441,
+                    1340,
+                    474,
+                    1269,
+                    474
+                ],
+                "score": 0.62,
+                "latex": "\\mathbf { O k } ^ { 1 , 2 }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    481,
+                    442,
+                    561,
+                    442,
+                    561,
+                    475,
+                    481,
+                    475
+                ],
+                "score": 0.47,
+                "latex": "\\mathbf { J e o n ^ { 1 * } }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    913,
+                    442,
+                    973,
+                    442,
+                    973,
+                    474,
+                    913,
+                    474
+                ],
+                "score": 0.33,
+                "latex": "\\mathbf { L e e ^ { 3 } }"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    351.0,
+                    266.0,
+                    1353.0,
+                    266.0,
+                    1353.0,
+                    333.0,
+                    351.0,
+                    333.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1201.0,
+                    536.0,
+                    1201.0,
+                    536.0,
+                    1249.0,
+                    292.0,
+                    1249.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    2028.0,
+                    1074.0,
+                    2028.0,
+                    1074.0,
+                    2066.0,
+                    295.0,
+                    2066.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    783.0,
+                    679.0,
+                    919.0,
+                    679.0,
+                    919.0,
+                    721.0,
+                    783.0,
+                    721.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    341.0,
+                    1947.0,
+                    550.0,
+                    1947.0,
+                    550.0,
+                    1984.0,
+                    341.0,
+                    1984.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    394.0,
+                    751.0,
+                    1305.0,
+                    751.0,
+                    1305.0,
+                    786.0,
+                    394.0,
+                    786.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    394.0,
+                    783.0,
+                    1305.0,
+                    783.0,
+                    1305.0,
+                    818.0,
+                    394.0,
+                    818.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    393.0,
+                    812.0,
+                    1306.0,
+                    812.0,
+                    1306.0,
+                    847.0,
+                    393.0,
+                    847.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    392.0,
+                    842.0,
+                    1307.0,
+                    842.0,
+                    1307.0,
+                    878.0,
+                    392.0,
+                    878.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    393.0,
+                    876.0,
+                    1308.0,
+                    876.0,
+                    1308.0,
+                    906.0,
+                    393.0,
+                    906.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    393.0,
+                    904.0,
+                    1308.0,
+                    904.0,
+                    1308.0,
+                    939.0,
+                    393.0,
+                    939.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    394.0,
+                    936.0,
+                    1306.0,
+                    936.0,
+                    1306.0,
+                    968.0,
+                    394.0,
+                    968.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    394.0,
+                    963.0,
+                    1307.0,
+                    963.0,
+                    1307.0,
+                    998.0,
+                    394.0,
+                    998.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    393.0,
+                    993.0,
+                    1306.0,
+                    993.0,
+                    1306.0,
+                    1032.0,
+                    393.0,
+                    1032.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    394.0,
+                    1026.0,
+                    1305.0,
+                    1026.0,
+                    1305.0,
+                    1058.0,
+                    394.0,
+                    1058.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    394.0,
+                    1055.0,
+                    1305.0,
+                    1055.0,
+                    1305.0,
+                    1087.0,
+                    394.0,
+                    1087.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    394.0,
+                    1084.0,
+                    1304.0,
+                    1084.0,
+                    1304.0,
+                    1120.0,
+                    394.0,
+                    1120.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    393.0,
+                    1116.0,
+                    1066.0,
+                    1116.0,
+                    1066.0,
+                    1152.0,
+                    393.0,
+                    1152.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1546.0,
+                    1406.0,
+                    1546.0,
+                    1406.0,
+                    1584.0,
+                    292.0,
+                    1584.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1580.0,
+                    1404.0,
+                    1580.0,
+                    1404.0,
+                    1610.0,
+                    296.0,
+                    1610.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1610.0,
+                    1406.0,
+                    1610.0,
+                    1406.0,
+                    1644.0,
+                    294.0,
+                    1644.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1640.0,
+                    1405.0,
+                    1640.0,
+                    1405.0,
+                    1674.0,
+                    293.0,
+                    1674.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1668.0,
+                    1404.0,
+                    1668.0,
+                    1404.0,
+                    1703.0,
+                    293.0,
+                    1703.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1700.0,
+                    1405.0,
+                    1700.0,
+                    1405.0,
+                    1732.0,
+                    293.0,
+                    1732.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1729.0,
+                    1408.0,
+                    1729.0,
+                    1408.0,
+                    1766.0,
+                    293.0,
+                    1766.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1761.0,
+                    885.0,
+                    1761.0,
+                    885.0,
+                    1797.0,
+                    295.0,
+                    1797.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1276.0,
+                    1406.0,
+                    1276.0,
+                    1406.0,
+                    1308.0,
+                    295.0,
+                    1308.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1306.0,
+                    1407.0,
+                    1306.0,
+                    1407.0,
+                    1339.0,
+                    293.0,
+                    1339.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1334.0,
+                    1407.0,
+                    1334.0,
+                    1407.0,
+                    1374.0,
+                    293.0,
+                    1374.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1367.0,
+                    1085.0,
+                    1367.0,
+                    1085.0,
+                    1400.0,
+                    294.0,
+                    1400.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1410.0,
+                    1407.0,
+                    1410.0,
+                    1407.0,
+                    1449.0,
+                    293.0,
+                    1449.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1441.0,
+                    1404.0,
+                    1441.0,
+                    1404.0,
+                    1478.0,
+                    293.0,
+                    1478.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1473.0,
+                    1408.0,
+                    1473.0,
+                    1408.0,
+                    1509.0,
+                    294.0,
+                    1509.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1503.0,
+                    891.0,
+                    1503.0,
+                    891.0,
+                    1538.0,
+                    293.0,
+                    1538.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1804.0,
+                    1405.0,
+                    1804.0,
+                    1405.0,
+                    1841.0,
+                    293.0,
+                    1841.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1834.0,
+                    1407.0,
+                    1834.0,
+                    1407.0,
+                    1871.0,
+                    293.0,
+                    1871.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1868.0,
+                    1407.0,
+                    1868.0,
+                    1407.0,
+                    1900.0,
+                    295.0,
+                    1900.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1898.0,
+                    1405.0,
+                    1898.0,
+                    1405.0,
+                    1933.0,
+                    293.0,
+                    1933.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    386.0,
+                    440.0,
+                    480.0,
+                    440.0,
+                    480.0,
+                    479.0,
+                    386.0,
+                    479.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    562.0,
+                    440.0,
+                    689.0,
+                    440.0,
+                    689.0,
+                    479.0,
+                    562.0,
+                    479.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    770.0,
+                    440.0,
+                    912.0,
+                    440.0,
+                    912.0,
+                    479.0,
+                    770.0,
+                    479.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    974.0,
+                    440.0,
+                    1093.0,
+                    440.0,
+                    1093.0,
+                    479.0,
+                    974.0,
+                    479.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1148.0,
+                    440.0,
+                    1268.0,
+                    440.0,
+                    1268.0,
+                    479.0,
+                    1148.0,
+                    479.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    315.0,
+                    464.0,
+                    1413.0,
+                    464.0,
+                    1413.0,
+                    520.0,
+                    315.0,
+                    520.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    351.0,
+                    496.0,
+                    1375.0,
+                    496.0,
+                    1375.0,
+                    552.0,
+                    351.0,
+                    552.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    306.0,
+                    528.0,
+                    1422.0,
+                    528.0,
+                    1422.0,
+                    583.0,
+                    306.0,
+                    583.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    372.0,
+                    561.0,
+                    1355.0,
+                    561.0,
+                    1355.0,
+                    613.0,
+                    372.0,
+                    613.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 0,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 3,
+                "poly": [
+                    514,
+                    192,
+                    1178,
+                    192,
+                    1178,
+                    808,
+                    514,
+                    808
+                ],
+                "score": 0.976
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    949,
+                    1404,
+                    949,
+                    1404,
+                    1132,
+                    298,
+                    1132
+                ],
+                "score": 0.972
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1146,
+                    1405,
+                    1146,
+                    1405,
+                    1329,
+                    298,
+                    1329
+                ],
+                "score": 0.968
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    300,
+                    1343,
+                    1404,
+                    1343,
+                    1404,
+                    1436,
+                    300,
+                    1436
+                ],
+                "score": 0.964
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    368,
+                    1513,
+                    1405,
+                    1513,
+                    1405,
+                    1983,
+                    368,
+                    1983
+                ],
+                "score": 0.954
+            },
+            {
+                "category_id": 4,
+                "poly": [
+                    300,
+                    822,
+                    1400,
+                    822,
+                    1400,
+                    913,
+                    300,
+                    913
+                ],
+                "score": 0.93
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    300,
+                    1449,
+                    725,
+                    1449,
+                    725,
+                    1481,
+                    300,
+                    1481
+                ],
+                "score": 0.921
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    841,
+                    2062,
+                    858,
+                    2062,
+                    858,
+                    2084,
+                    841,
+                    2084
+                ],
+                "score": 0.724
+            },
+            {
+                "category_id": 4,
+                "poly": [
+                    766,
+                    791,
+                    926,
+                    791,
+                    926,
+                    814,
+                    766,
+                    814
+                ],
+                "score": 0.347
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    758,
+                    481,
+                    775,
+                    481,
+                    775,
+                    499,
+                    758,
+                    499
+                ],
+                "score": 0.36,
+                "latex": "^ +"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    874.0,
+                    195.0,
+                    1179.0,
+                    195.0,
+                    1179.0,
+                    322.0,
+                    874.0,
+                    322.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    791.0,
+                    324.0,
+                    901.0,
+                    324.0,
+                    901.0,
+                    351.0,
+                    791.0,
+                    351.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    576.0,
+                    408.0,
+                    585.0,
+                    408.0,
+                    585.0,
+                    417.0,
+                    576.0,
+                    417.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    659.0,
+                    475.0,
+                    757.0,
+                    475.0,
+                    757.0,
+                    507.0,
+                    659.0,
+                    507.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    776.0,
+                    475.0,
+                    1029.0,
+                    475.0,
+                    1029.0,
+                    507.0,
+                    776.0,
+                    507.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1054.0,
+                    529.0,
+                    1156.0,
+                    529.0,
+                    1156.0,
+                    616.0,
+                    1054.0,
+                    616.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1052.0,
+                    561.0,
+                    1070.0,
+                    561.0,
+                    1070.0,
+                    586.0,
+                    1052.0,
+                    586.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    766.0,
+                    631.0,
+                    926.0,
+                    631.0,
+                    926.0,
+                    663.0,
+                    766.0,
+                    663.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    518.0,
+                    664.0,
+                    685.0,
+                    664.0,
+                    685.0,
+                    786.0,
+                    518.0,
+                    786.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    858.0,
+                    647.0,
+                    1195.0,
+                    647.0,
+                    1195.0,
+                    802.0,
+                    858.0,
+                    802.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    763.0,
+                    790.0,
+                    929.0,
+                    790.0,
+                    929.0,
+                    814.0,
+                    763.0,
+                    814.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    555.0,
+                    540.0,
+                    631.0,
+                    540.0,
+                    631.0,
+                    594.0,
+                    555.0,
+                    594.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    820.0,
+                    1405.0,
+                    820.0,
+                    1405.0,
+                    857.0,
+                    294.0,
+                    857.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    852.0,
+                    1405.0,
+                    852.0,
+                    1405.0,
+                    889.0,
+                    294.0,
+                    889.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    880.0,
+                    393.0,
+                    880.0,
+                    393.0,
+                    923.0,
+                    292.0,
+                    923.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    839.0,
+                    2059.0,
+                    862.0,
+                    2059.0,
+                    862.0,
+                    2093.0,
+                    839.0,
+                    2093.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    764.0,
+                    788.0,
+                    929.0,
+                    788.0,
+                    929.0,
+                    819.0,
+                    764.0,
+                    819.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    948.0,
+                    1406.0,
+                    948.0,
+                    1406.0,
+                    984.0,
+                    292.0,
+                    984.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    981.0,
+                    1405.0,
+                    981.0,
+                    1405.0,
+                    1013.0,
+                    294.0,
+                    1013.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1008.0,
+                    1408.0,
+                    1008.0,
+                    1408.0,
+                    1044.0,
+                    293.0,
+                    1044.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1041.0,
+                    1406.0,
+                    1041.0,
+                    1406.0,
+                    1076.0,
+                    294.0,
+                    1076.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1071.0,
+                    1405.0,
+                    1071.0,
+                    1405.0,
+                    1106.0,
+                    294.0,
+                    1106.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1100.0,
+                    582.0,
+                    1100.0,
+                    582.0,
+                    1135.0,
+                    293.0,
+                    1135.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1143.0,
+                    1409.0,
+                    1143.0,
+                    1409.0,
+                    1182.0,
+                    293.0,
+                    1182.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1173.0,
+                    1405.0,
+                    1173.0,
+                    1405.0,
+                    1214.0,
+                    293.0,
+                    1214.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1206.0,
+                    1406.0,
+                    1206.0,
+                    1406.0,
+                    1240.0,
+                    295.0,
+                    1240.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1238.0,
+                    1407.0,
+                    1238.0,
+                    1407.0,
+                    1273.0,
+                    295.0,
+                    1273.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1268.0,
+                    1406.0,
+                    1268.0,
+                    1406.0,
+                    1302.0,
+                    293.0,
+                    1302.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1297.0,
+                    1253.0,
+                    1297.0,
+                    1253.0,
+                    1331.0,
+                    292.0,
+                    1331.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1341.0,
+                    1405.0,
+                    1341.0,
+                    1405.0,
+                    1379.0,
+                    295.0,
+                    1379.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1371.0,
+                    1408.0,
+                    1371.0,
+                    1408.0,
+                    1411.0,
+                    293.0,
+                    1411.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    290.0,
+                    1403.0,
+                    412.0,
+                    1403.0,
+                    412.0,
+                    1438.0,
+                    290.0,
+                    1438.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    371.0,
+                    1510.0,
+                    1404.0,
+                    1510.0,
+                    1404.0,
+                    1549.0,
+                    371.0,
+                    1549.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    394.0,
+                    1543.0,
+                    1387.0,
+                    1543.0,
+                    1387.0,
+                    1579.0,
+                    394.0,
+                    1579.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    368.0,
+                    1597.0,
+                    1404.0,
+                    1597.0,
+                    1404.0,
+                    1633.0,
+                    368.0,
+                    1633.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    393.0,
+                    1627.0,
+                    1408.0,
+                    1627.0,
+                    1408.0,
+                    1666.0,
+                    393.0,
+                    1666.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    370.0,
+                    1684.0,
+                    1404.0,
+                    1684.0,
+                    1404.0,
+                    1721.0,
+                    370.0,
+                    1721.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    393.0,
+                    1714.0,
+                    1405.0,
+                    1714.0,
+                    1405.0,
+                    1750.0,
+                    393.0,
+                    1750.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    394.0,
+                    1744.0,
+                    1405.0,
+                    1744.0,
+                    1405.0,
+                    1781.0,
+                    394.0,
+                    1781.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    396.0,
+                    1777.0,
+                    573.0,
+                    1777.0,
+                    573.0,
+                    1807.0,
+                    396.0,
+                    1807.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    391.0,
+                    1831.0,
+                    1404.0,
+                    1831.0,
+                    1404.0,
+                    1864.0,
+                    391.0,
+                    1864.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    393.0,
+                    1858.0,
+                    1405.0,
+                    1858.0,
+                    1405.0,
+                    1895.0,
+                    393.0,
+                    1895.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    393.0,
+                    1888.0,
+                    1404.0,
+                    1888.0,
+                    1404.0,
+                    1929.0,
+                    393.0,
+                    1929.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    393.0,
+                    1924.0,
+                    1405.0,
+                    1924.0,
+                    1405.0,
+                    1955.0,
+                    393.0,
+                    1955.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    395.0,
+                    1952.0,
+                    1213.0,
+                    1952.0,
+                    1213.0,
+                    1986.0,
+                    395.0,
+                    1986.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1445.0,
+                    730.0,
+                    1445.0,
+                    730.0,
+                    1484.0,
+                    296.0,
+                    1484.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 1,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    782,
+                    1405,
+                    782,
+                    1405,
+                    1148,
+                    297,
+                    1148
+                ],
+                "score": 0.984
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    538,
+                    1405,
+                    538,
+                    1405,
+                    754,
+                    298,
+                    754
+                ],
+                "score": 0.981
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    264,
+                    1406,
+                    264,
+                    1406,
+                    510,
+                    297,
+                    510
+                ],
+                "score": 0.981
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1416,
+                    1405,
+                    1416,
+                    1405,
+                    1612,
+                    297,
+                    1612
+                ],
+                "score": 0.981
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    1741,
+                    1404,
+                    1741,
+                    1404,
+                    1904,
+                    299,
+                    1904
+                ],
+                "score": 0.978
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    570,
+                    1623,
+                    1128,
+                    1623,
+                    1128,
+                    1734,
+                    570,
+                    1734
+                ],
+                "score": 0.958
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    702,
+                    1339,
+                    996,
+                    1339,
+                    996,
+                    1413,
+                    702,
+                    1413
+                ],
+                "score": 0.951
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    296,
+                    1254,
+                    1402,
+                    1254,
+                    1402,
+                    1342,
+                    296,
+                    1342
+                ],
+                "score": 0.95
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    668,
+                    1921,
+                    1031,
+                    1921,
+                    1031,
+                    1971,
+                    668,
+                    1971
+                ],
+                "score": 0.938
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    299,
+                    1188,
+                    650,
+                    1188,
+                    650,
+                    1226,
+                    299,
+                    1226
+                ],
+                "score": 0.931
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    299,
+                    198,
+                    541,
+                    198,
+                    541,
+                    235,
+                    299,
+                    235
+                ],
+                "score": 0.924
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    303,
+                    1975,
+                    1397,
+                    1975,
+                    1397,
+                    2011,
+                    303,
+                    2011
+                ],
+                "score": 0.918
+            },
+            {
+                "category_id": 9,
+                "poly": [
+                    1366,
+                    1659,
+                    1400,
+                    1659,
+                    1400,
+                    1689,
+                    1366,
+                    1689
+                ],
+                "score": 0.891
+            },
+            {
+                "category_id": 9,
+                "poly": [
+                    1366,
+                    1347,
+                    1400,
+                    1347,
+                    1400,
+                    1378,
+                    1366,
+                    1378
+                ],
+                "score": 0.876
+            },
+            {
+                "category_id": 9,
+                "poly": [
+                    1366,
+                    1925,
+                    1400,
+                    1925,
+                    1400,
+                    1954,
+                    1366,
+                    1954
+                ],
+                "score": 0.871
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    841,
+                    2061,
+                    859,
+                    2061,
+                    859,
+                    2085,
+                    841,
+                    2085
+                ],
+                "score": 0.674
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    841,
+                    2061,
+                    859,
+                    2061,
+                    859,
+                    2085,
+                    841,
+                    2085
+                ],
+                "score": 0.494
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    570,
+                    1619,
+                    1127,
+                    1619,
+                    1127,
+                    1736,
+                    570,
+                    1736
+                ],
+                "score": 0.94,
+                "latex": "\\operatorname* { m i n } _ { ( x _ { 1 } , y _ { 1 } ) , \\cdots , ( x _ { B } , y _ { B } ) } d \\left( \\frac { 1 } { B } \\sum _ { j = 1 } ^ { B } \\nabla \\ell ( f _ { \\theta } ( x _ { j } ) , y _ { j } ) , g \\right) \\ ,"
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    701,
+                    1334,
+                    995,
+                    1334,
+                    995,
+                    1414,
+                    701,
+                    1414
+                ],
+                "score": 0.94,
+                "latex": "\\operatorname* { m i n } _ { \\theta } \\sum _ { ( x , y ) \\in \\mathcal { D } } \\ell ( f _ { \\theta } ( x ) , y ) \\ ,"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1066,
+                    1538,
+                    1397,
+                    1538,
+                    1397,
+                    1584,
+                    1066,
+                    1584
+                ],
+                "score": 0.93,
+                "latex": "\\begin{array} { r } { \\boldsymbol { g } = \\frac { \\top } { B } \\sum _ { j = 1 } ^ { B } \\dot { \\nabla } \\ell ( f _ { \\theta } ( x _ { j } ^ { * } ) , y _ { j } ^ { * } ) } \\end{array}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    737,
+                    1543,
+                    1004,
+                    1543,
+                    1004,
+                    1580,
+                    737,
+                    1580
+                ],
+                "score": 0.93,
+                "latex": "\\{ ( x _ { 1 } ^ { * } , y _ { 1 } ^ { * } ) , . . . , ( x _ { B } ^ { * } , y _ { B } ^ { * } ) \\}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1060,
+                    1977,
+                    1152,
+                    1977,
+                    1152,
+                    2013,
+                    1060,
+                    2013
+                ],
+                "score": 0.93,
+                "latex": "y _ { j } = y _ { j } ^ { * }"
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    667,
+                    1921,
+                    1030,
+                    1921,
+                    1030,
+                    1972,
+                    667,
+                    1972
+                ],
+                "score": 0.93,
+                "latex": "\\operatorname* { m i n } _ { \\substack { x _ { 1 } , \\ldots , x _ { B } \\in \\mathbb { R } ^ { m } } } c \\left( x _ { 1 } , . . . , x _ { B } ; \\theta , g \\right) ,"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    384,
+                    1741,
+                    450,
+                    1741,
+                    450,
+                    1775,
+                    384,
+                    1775
+                ],
+                "score": 0.92,
+                "latex": "d ( \\cdot , \\cdot )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1236,
+                    1450,
+                    1361,
+                    1450,
+                    1361,
+                    1482,
+                    1236,
+                    1482
+                ],
+                "score": 0.92,
+                "latex": "\\ell ( f _ { \\theta } ( x ) , y )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1259,
+                    1415,
+                    1400,
+                    1415,
+                    1400,
+                    1450,
+                    1259,
+                    1450
+                ],
+                "score": 0.92,
+                "latex": "y \\in \\{ 0 , 1 \\} ^ { L }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    453,
+                    1479,
+                    518,
+                    1479,
+                    518,
+                    1512,
+                    453,
+                    1512
+                ],
+                "score": 0.92,
+                "latex": "( x , y )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1040,
+                    1418,
+                    1139,
+                    1418,
+                    1139,
+                    1446,
+                    1040,
+                    1446
+                ],
+                "score": 0.91,
+                "latex": "x \\in \\mathbb { R } ^ { m }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1254,
+                    1976,
+                    1390,
+                    1976,
+                    1390,
+                    2008,
+                    1254,
+                    2008
+                ],
+                "score": 0.91,
+                "latex": "j = 1 , . . . , B"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    526,
+                    1976,
+                    729,
+                    1976,
+                    729,
+                    2009,
+                    526,
+                    2009
+                ],
+                "score": 0.9,
+                "latex": "c \\left( x _ { 1 } , . . . , x _ { B } ; \\theta , g \\right)"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1141,
+                    1742,
+                    1168,
+                    1742,
+                    1168,
+                    1774,
+                    1141,
+                    1774
+                ],
+                "score": 0.86,
+                "latex": "\\ell _ { 2 }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1154,
+                    1287,
+                    1181,
+                    1287,
+                    1181,
+                    1318,
+                    1154,
+                    1318
+                ],
+                "score": 0.86,
+                "latex": "f _ { \\theta }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    775,
+                    1419,
+                    801,
+                    1419,
+                    801,
+                    1445,
+                    775,
+                    1445
+                ],
+                "score": 0.84,
+                "latex": "\\mathcal { D }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    873,
+                    1786,
+                    892,
+                    1786,
+                    892,
+                    1812,
+                    873,
+                    1812
+                ],
+                "score": 0.8,
+                "latex": "y"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    555,
+                    1549,
+                    574,
+                    1549,
+                    574,
+                    1577,
+                    555,
+                    1577
+                ],
+                "score": 0.79,
+                "latex": "g"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    373,
+                    1419,
+                    389,
+                    1419,
+                    389,
+                    1445,
+                    373,
+                    1445
+                ],
+                "score": 0.78,
+                "latex": "\\ell"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1384,
+                    1287,
+                    1401,
+                    1287,
+                    1401,
+                    1313,
+                    1384,
+                    1313
+                ],
+                "score": 0.76,
+                "latex": "\\theta"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1286,
+                    1816,
+                    1305,
+                    1816,
+                    1305,
+                    1838,
+                    1286,
+                    1838
+                ],
+                "score": 0.74,
+                "latex": "x"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1187.0,
+                    652.0,
+                    1187.0,
+                    652.0,
+                    1231.0,
+                    291.0,
+                    1231.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    195.0,
+                    545.0,
+                    195.0,
+                    545.0,
+                    240.0,
+                    292.0,
+                    240.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    838.0,
+                    2058.0,
+                    862.0,
+                    2058.0,
+                    862.0,
+                    2091.0,
+                    838.0,
+                    2091.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    838.0,
+                    2058.0,
+                    862.0,
+                    2058.0,
+                    862.0,
+                    2091.0,
+                    838.0,
+                    2091.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    784.0,
+                    1405.0,
+                    784.0,
+                    1405.0,
+                    818.0,
+                    296.0,
+                    818.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    813.0,
+                    1405.0,
+                    813.0,
+                    1405.0,
+                    847.0,
+                    294.0,
+                    847.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    842.0,
+                    1407.0,
+                    842.0,
+                    1407.0,
+                    878.0,
+                    292.0,
+                    878.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    875.0,
+                    1406.0,
+                    875.0,
+                    1406.0,
+                    909.0,
+                    295.0,
+                    909.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    906.0,
+                    1407.0,
+                    906.0,
+                    1407.0,
+                    940.0,
+                    294.0,
+                    940.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    933.0,
+                    1405.0,
+                    933.0,
+                    1405.0,
+                    969.0,
+                    292.0,
+                    969.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    965.0,
+                    1405.0,
+                    965.0,
+                    1405.0,
+                    999.0,
+                    295.0,
+                    999.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    995.0,
+                    1403.0,
+                    995.0,
+                    1403.0,
+                    1029.0,
+                    295.0,
+                    1029.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1023.0,
+                    1406.0,
+                    1023.0,
+                    1406.0,
+                    1062.0,
+                    295.0,
+                    1062.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1056.0,
+                    1405.0,
+                    1056.0,
+                    1405.0,
+                    1090.0,
+                    294.0,
+                    1090.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1088.0,
+                    1405.0,
+                    1088.0,
+                    1405.0,
+                    1122.0,
+                    295.0,
+                    1122.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1117.0,
+                    564.0,
+                    1117.0,
+                    564.0,
+                    1152.0,
+                    295.0,
+                    1152.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    539.0,
+                    1406.0,
+                    539.0,
+                    1406.0,
+                    577.0,
+                    296.0,
+                    577.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    568.0,
+                    1406.0,
+                    568.0,
+                    1406.0,
+                    605.0,
+                    292.0,
+                    605.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    597.0,
+                    1406.0,
+                    597.0,
+                    1406.0,
+                    639.0,
+                    292.0,
+                    639.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    630.0,
+                    1410.0,
+                    630.0,
+                    1410.0,
+                    667.0,
+                    291.0,
+                    667.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    660.0,
+                    1408.0,
+                    660.0,
+                    1408.0,
+                    695.0,
+                    292.0,
+                    695.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    690.0,
+                    1405.0,
+                    690.0,
+                    1405.0,
+                    725.0,
+                    293.0,
+                    725.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    721.0,
+                    1113.0,
+                    721.0,
+                    1113.0,
+                    756.0,
+                    293.0,
+                    756.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    262.0,
+                    1407.0,
+                    262.0,
+                    1407.0,
+                    302.0,
+                    292.0,
+                    302.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    296.0,
+                    1406.0,
+                    296.0,
+                    1406.0,
+                    329.0,
+                    296.0,
+                    329.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    323.0,
+                    1406.0,
+                    323.0,
+                    1406.0,
+                    363.0,
+                    292.0,
+                    363.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    355.0,
+                    1406.0,
+                    355.0,
+                    1406.0,
+                    392.0,
+                    294.0,
+                    392.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    385.0,
+                    1407.0,
+                    385.0,
+                    1407.0,
+                    423.0,
+                    292.0,
+                    423.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    415.0,
+                    1406.0,
+                    415.0,
+                    1406.0,
+                    452.0,
+                    292.0,
+                    452.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    445.0,
+                    1406.0,
+                    445.0,
+                    1406.0,
+                    486.0,
+                    292.0,
+                    486.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    480.0,
+                    1162.0,
+                    480.0,
+                    1162.0,
+                    510.0,
+                    294.0,
+                    510.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1414.0,
+                    372.0,
+                    1414.0,
+                    372.0,
+                    1452.0,
+                    292.0,
+                    1452.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    390.0,
+                    1414.0,
+                    774.0,
+                    1414.0,
+                    774.0,
+                    1452.0,
+                    390.0,
+                    1452.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    802.0,
+                    1414.0,
+                    1039.0,
+                    1414.0,
+                    1039.0,
+                    1452.0,
+                    802.0,
+                    1452.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1140.0,
+                    1414.0,
+                    1258.0,
+                    1414.0,
+                    1258.0,
+                    1452.0,
+                    1140.0,
+                    1452.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1401.0,
+                    1414.0,
+                    1405.0,
+                    1414.0,
+                    1405.0,
+                    1452.0,
+                    1401.0,
+                    1452.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1447.0,
+                    1235.0,
+                    1447.0,
+                    1235.0,
+                    1484.0,
+                    294.0,
+                    1484.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1362.0,
+                    1447.0,
+                    1406.0,
+                    1447.0,
+                    1406.0,
+                    1484.0,
+                    1362.0,
+                    1484.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1476.0,
+                    452.0,
+                    1476.0,
+                    452.0,
+                    1513.0,
+                    294.0,
+                    1513.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    519.0,
+                    1476.0,
+                    1405.0,
+                    1476.0,
+                    1405.0,
+                    1513.0,
+                    519.0,
+                    1513.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1507.0,
+                    1406.0,
+                    1507.0,
+                    1406.0,
+                    1544.0,
+                    294.0,
+                    1544.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1536.0,
+                    554.0,
+                    1536.0,
+                    554.0,
+                    1589.0,
+                    291.0,
+                    1589.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    575.0,
+                    1536.0,
+                    736.0,
+                    1536.0,
+                    736.0,
+                    1589.0,
+                    575.0,
+                    1589.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1005.0,
+                    1536.0,
+                    1065.0,
+                    1536.0,
+                    1065.0,
+                    1589.0,
+                    1005.0,
+                    1589.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1398.0,
+                    1537.0,
+                    1410.0,
+                    1537.0,
+                    1410.0,
+                    1587.0,
+                    1398.0,
+                    1587.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1580.0,
+                    929.0,
+                    1580.0,
+                    929.0,
+                    1613.0,
+                    296.0,
+                    1613.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1141.75,
+                    1535.0,
+                    1205.75,
+                    1535.0,
+                    1205.75,
+                    1568.0,
+                    1141.75,
+                    1568.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    304.0,
+                    1740.0,
+                    383.0,
+                    1740.0,
+                    383.0,
+                    1777.0,
+                    304.0,
+                    1777.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    451.0,
+                    1740.0,
+                    1140.0,
+                    1740.0,
+                    1140.0,
+                    1777.0,
+                    451.0,
+                    1777.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1169.0,
+                    1740.0,
+                    1406.0,
+                    1740.0,
+                    1406.0,
+                    1777.0,
+                    1169.0,
+                    1777.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    1781.0,
+                    872.0,
+                    1781.0,
+                    872.0,
+                    1814.0,
+                    297.0,
+                    1814.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    893.0,
+                    1781.0,
+                    1404.0,
+                    1781.0,
+                    1404.0,
+                    1814.0,
+                    893.0,
+                    1814.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1812.0,
+                    1285.0,
+                    1812.0,
+                    1285.0,
+                    1845.0,
+                    295.0,
+                    1845.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1306.0,
+                    1812.0,
+                    1405.0,
+                    1812.0,
+                    1405.0,
+                    1845.0,
+                    1306.0,
+                    1845.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1840.0,
+                    1404.0,
+                    1840.0,
+                    1404.0,
+                    1875.0,
+                    294.0,
+                    1875.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1872.0,
+                    1408.0,
+                    1872.0,
+                    1408.0,
+                    1905.0,
+                    295.0,
+                    1905.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1253.0,
+                    1405.0,
+                    1253.0,
+                    1405.0,
+                    1289.0,
+                    292.0,
+                    1289.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1285.0,
+                    1153.0,
+                    1285.0,
+                    1153.0,
+                    1320.0,
+                    292.0,
+                    1320.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1182.0,
+                    1285.0,
+                    1383.0,
+                    1285.0,
+                    1383.0,
+                    1320.0,
+                    1182.0,
+                    1320.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1402.0,
+                    1285.0,
+                    1405.0,
+                    1285.0,
+                    1405.0,
+                    1320.0,
+                    1402.0,
+                    1320.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1314.0,
+                    426.0,
+                    1314.0,
+                    426.0,
+                    1349.0,
+                    294.0,
+                    1349.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    1970.0,
+                    525.0,
+                    1970.0,
+                    525.0,
+                    2014.0,
+                    297.0,
+                    2014.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    730.0,
+                    1970.0,
+                    1059.0,
+                    1970.0,
+                    1059.0,
+                    2014.0,
+                    730.0,
+                    2014.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1153.0,
+                    1970.0,
+                    1253.0,
+                    1970.0,
+                    1253.0,
+                    2014.0,
+                    1153.0,
+                    2014.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1391.0,
+                    1970.0,
+                    1400.0,
+                    1970.0,
+                    1400.0,
+                    2014.0,
+                    1391.0,
+                    2014.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 2,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    989,
+                    1406,
+                    989,
+                    1406,
+                    1331,
+                    297,
+                    1331
+                ],
+                "score": 0.984
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1419,
+                    1406,
+                    1419,
+                    1406,
+                    1606,
+                    297,
+                    1606
+                ],
+                "score": 0.981
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1636,
+                    1406,
+                    1636,
+                    1406,
+                    1820,
+                    297,
+                    1820
+                ],
+                "score": 0.981
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    704,
+                    1406,
+                    704,
+                    1406,
+                    890,
+                    297,
+                    890
+                ],
+                "score": 0.98
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    265,
+                    1405,
+                    265,
+                    1405,
+                    450,
+                    297,
+                    450
+                ],
+                "score": 0.979
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1913,
+                    1402,
+                    1913,
+                    1402,
+                    2008,
+                    298,
+                    2008
+                ],
+                "score": 0.974
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    462,
+                    1405,
+                    462,
+                    1405,
+                    556,
+                    298,
+                    556
+                ],
+                "score": 0.972
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    567,
+                    568,
+                    1133,
+                    568,
+                    1133,
+                    647,
+                    567,
+                    647
+                ],
+                "score": 0.957
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    516,
+                    906,
+                    1178,
+                    906,
+                    1178,
+                    976,
+                    516,
+                    976
+                ],
+                "score": 0.95
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    638,
+                    1832,
+                    1056,
+                    1832,
+                    1056,
+                    1887,
+                    638,
+                    1887
+                ],
+                "score": 0.944
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    292,
+                    658,
+                    1342,
+                    658,
+                    1342,
+                    692,
+                    292,
+                    692
+                ],
+                "score": 0.938
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    297,
+                    1365,
+                    949,
+                    1365,
+                    949,
+                    1399,
+                    297,
+                    1399
+                ],
+                "score": 0.936
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    298,
+                    198,
+                    475,
+                    198,
+                    475,
+                    235,
+                    298,
+                    235
+                ],
+                "score": 0.911
+            },
+            {
+                "category_id": 9,
+                "poly": [
+                    1366,
+                    1837,
+                    1400,
+                    1837,
+                    1400,
+                    1866,
+                    1366,
+                    1866
+                ],
+                "score": 0.881
+            },
+            {
+                "category_id": 9,
+                "poly": [
+                    1366,
+                    580,
+                    1400,
+                    580,
+                    1400,
+                    609,
+                    1366,
+                    609
+                ],
+                "score": 0.88
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    841,
+                    2061,
+                    858,
+                    2061,
+                    858,
+                    2084,
+                    841,
+                    2084
+                ],
+                "score": 0.789
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    610,
+                    992,
+                    799,
+                    992,
+                    799,
+                    1028,
+                    610,
+                    1028
+                ],
+                "score": 0.93,
+                "latex": "\\sigma _ { l } ^ { 2 } ( x _ { 1 } , . . . , x _ { B } ; \\theta )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1284,
+                    463,
+                    1372,
+                    463,
+                    1372,
+                    497,
+                    1284,
+                    497
+                ],
+                "score": 0.93,
+                "latex": "R _ { \\mathrm { T V } } ( x )"
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    566,
+                    566,
+                    1131,
+                    566,
+                    1131,
+                    647,
+                    566,
+                    647
+                ],
+                "score": 0.93,
+                "latex": "R _ { \\mathrm { T V } } ( x ) : = \\sum _ { ( i , j ) } \\sum _ { ( i ^ { \\prime } , j ^ { \\prime } ) \\in \\partial ( i , j ) } \\Vert x ( i , j ) - x ( i ^ { \\prime } , j ^ { \\prime } ) \\Vert ^ { 2 } ,"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    371,
+                    659,
+                    446,
+                    659,
+                    446,
+                    693,
+                    371,
+                    693
+                ],
+                "score": 0.93,
+                "latex": "\\partial ( i , j )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1098,
+                    1757,
+                    1309,
+                    1757,
+                    1309,
+                    1792,
+                    1098,
+                    1792
+                ],
+                "score": 0.93,
+                "latex": "\\{ G _ { w } ( z ) : z \\in \\mathbb { R } ^ { k } \\}"
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    513,
+                    904,
+                    1177,
+                    904,
+                    1177,
+                    979,
+                    513,
+                    979
+                ],
+                "score": 0.93,
+                "latex": "R _ { \\mathrm { B N } } ( x _ { 1 } , . . . , x _ { B } ; \\theta ) : = \\sum _ { l } \\| \\mu _ { l } - \\mu _ { l , \\mathrm { e x a c t } } \\| _ { 2 } + \\| \\sigma _ { l } ^ { 2 } - \\sigma _ { l , \\mathrm { e x a c t } } ^ { 2 } \\| _ { 2 } ,"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    959,
+                    1452,
+                    1101,
+                    1452,
+                    1101,
+                    1486,
+                    959,
+                    1486
+                ],
+                "score": 0.92,
+                "latex": "( x ^ { \\ast } , \\cdot ) \\in \\bar { \\mathcal { D } }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    736,
+                    1453,
+                    906,
+                    1453,
+                    906,
+                    1486,
+                    736,
+                    1486
+                ],
+                "score": 0.92,
+                "latex": "x ^ { * } \\approx G _ { w } ( z ^ { * } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    372,
+                    993,
+                    557,
+                    993,
+                    557,
+                    1027,
+                    372,
+                    1027
+                ],
+                "score": 0.92,
+                "latex": "\\mu _ { l } ( x _ { 1 } , . . . , x _ { B } ; \\theta )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1161,
+                    992,
+                    1402,
+                    992,
+                    1402,
+                    1029,
+                    1161,
+                    1029
+                ],
+                "score": 0.92,
+                "latex": "\\sigma _ { l , \\mathrm { { e x a c t } } } ^ { 2 } ( x _ { 1 } ^ { * } , . . . , x _ { B } ^ { * } ; \\theta ) )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    731,
+                    659,
+                    789,
+                    659,
+                    789,
+                    693,
+                    731,
+                    693
+                ],
+                "score": 0.92,
+                "latex": "( i , j )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    875,
+                    993,
+                    1107,
+                    993,
+                    1107,
+                    1027,
+                    875,
+                    1027
+                ],
+                "score": 0.92,
+                "latex": "\\mu _ { l , \\mathrm { { e x a c t } } } ( x _ { 1 } ^ { * } , . . . , x _ { B } ^ { * } ; \\theta )"
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    635,
+                    1831,
+                    1062,
+                    1831,
+                    1062,
+                    1887,
+                    635,
+                    1887
+                ],
+                "score": 0.91,
+                "latex": "\\operatorname* { m i n } _ { \\substack { z _ { 1 } , \\ldots , z _ { B } \\in \\mathbb { R } ^ { k } } } c \\left( G _ { w } ( z _ { 1 } ) , . . . , G _ { w } ( z _ { B } ) \\right) ."
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    298,
+                    827,
+                    506,
+                    827,
+                    506,
+                    860,
+                    298,
+                    860
+                ],
+                "score": 0.91,
+                "latex": "R _ { \\mathrm { B N } } ( x _ { 1 } , . . . , x _ { B } ; \\theta )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    876,
+                    1668,
+                    976,
+                    1668,
+                    976,
+                    1696,
+                    876,
+                    1696
+                ],
+                "score": 0.91,
+                "latex": "k = 1 0 0"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1128,
+                    1032,
+                    1241,
+                    1032,
+                    1241,
+                    1058,
+                    1128,
+                    1058
+                ],
+                "score": 0.91,
+                "latex": "x _ { 1 } , . . . , x _ { B }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    424,
+                    1669,
+                    512,
+                    1669,
+                    512,
+                    1697,
+                    424,
+                    1697
+                ],
+                "score": 0.91,
+                "latex": "k \\ll m"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    585,
+                    1179,
+                    638,
+                    1179,
+                    638,
+                    1209,
+                    585,
+                    1209
+                ],
+                "score": 0.9,
+                "latex": "R _ { \\mathrm { B N } }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1240,
+                    1668,
+                    1402,
+                    1668,
+                    1402,
+                    1697,
+                    1240,
+                    1697
+                ],
+                "score": 0.9,
+                "latex": "k = 5 1 2 \\times 1 6"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    712,
+                    1420,
+                    901,
+                    1420,
+                    901,
+                    1452,
+                    712,
+                    1452
+                ],
+                "score": 0.9,
+                "latex": "G _ { w } : \\mathbb { R } ^ { k } \\mapsto \\mathbb { R } ^ { m }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    831,
+                    1731,
+                    918,
+                    1731,
+                    918,
+                    1758,
+                    831,
+                    1758
+                ],
+                "score": 0.89,
+                "latex": "k \\ll m"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1334,
+                    1454,
+                    1403,
+                    1454,
+                    1403,
+                    1482,
+                    1334,
+                    1482
+                ],
+                "score": 0.88,
+                "latex": "z ^ { * } ="
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    903,
+                    1916,
+                    931,
+                    1916,
+                    931,
+                    1946,
+                    903,
+                    1946
+                ],
+                "score": 0.88,
+                "latex": "f _ { \\theta }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1358,
+                    1946,
+                    1400,
+                    1946,
+                    1400,
+                    1976,
+                    1358,
+                    1976
+                ],
+                "score": 0.88,
+                "latex": "G _ { w }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1019,
+                    1759,
+                    1063,
+                    1759,
+                    1063,
+                    1786,
+                    1019,
+                    1786
+                ],
+                "score": 0.88,
+                "latex": "\\mathbb { R } ^ { m }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    365,
+                    1058,
+                    492,
+                    1058,
+                    492,
+                    1091,
+                    365,
+                    1091
+                ],
+                "score": 0.88,
+                "latex": "x _ { 1 } ^ { * } , . . . , x _ { B } ^ { * } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    443,
+                    860,
+                    473,
+                    860,
+                    473,
+                    895,
+                    443,
+                    895
+                ],
+                "score": 0.87,
+                "latex": "\\boldsymbol { x } _ { j } ^ { * }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1369,
+                    1916,
+                    1400,
+                    1916,
+                    1400,
+                    1942,
+                    1369,
+                    1942
+                ],
+                "score": 0.87,
+                "latex": "x ^ { * }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    496,
+                    1699,
+                    883,
+                    1699,
+                    883,
+                    1732,
+                    496,
+                    1732
+                ],
+                "score": 0.86,
+                "latex": "m = ( \\mathrm { w i d t h } ) \\times ( \\mathrm { h e i g h t } ) \\times ( \\mathrm { c o l o r } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1304,
+                    829,
+                    1335,
+                    829,
+                    1335,
+                    860,
+                    1304,
+                    860
+                ],
+                "score": 0.86,
+                "latex": "x _ { j }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1123,
+                    1699,
+                    1291,
+                    1699,
+                    1291,
+                    1728,
+                    1123,
+                    1728
+                ],
+                "score": 0.85,
+                "latex": "2 5 6 \\times 2 5 6 \\times 3"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    296,
+                    1482,
+                    573,
+                    1482,
+                    573,
+                    1516,
+                    296,
+                    1516
+                ],
+                "score": 0.84,
+                "latex": "\\mathrm { a r g m i n } _ { z } \\| G _ { w } ( z ) - x ^ { * } \\|"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    679,
+                    1549,
+                    697,
+                    1549,
+                    697,
+                    1571,
+                    679,
+                    1571
+                ],
+                "score": 0.8,
+                "latex": "z"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    587,
+                    1454,
+                    614,
+                    1454,
+                    614,
+                    1480,
+                    587,
+                    1480
+                ],
+                "score": 0.8,
+                "latex": "\\mathcal { D }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    973,
+                    1699,
+                    1113,
+                    1699,
+                    1113,
+                    1728,
+                    973,
+                    1728
+                ],
+                "score": 0.8,
+                "latex": "3 2 \\times 3 2 \\times 3"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1126,
+                    1548,
+                    1149,
+                    1548,
+                    1149,
+                    1571,
+                    1126,
+                    1571
+                ],
+                "score": 0.75,
+                "latex": "w"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    618,
+                    1030,
+                    629,
+                    1030,
+                    629,
+                    1054,
+                    618,
+                    1054
+                ],
+                "score": 0.73,
+                "latex": "l"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1213,
+                    1210,
+                    1230,
+                    1210,
+                    1230,
+                    1236,
+                    1213,
+                    1236
+                ],
+                "score": 0.72,
+                "latex": "\\theta"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    565,
+                    1059,
+                    582,
+                    1059,
+                    582,
+                    1085,
+                    565,
+                    1085
+                ],
+                "score": 0.71,
+                "latex": "\\theta"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    297,
+                    766,
+                    333,
+                    766,
+                    333,
+                    794,
+                    297,
+                    794
+                ],
+                "score": 0.39,
+                "latex": "\\mathrm { F L }"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1362.0,
+                    949.0,
+                    1362.0,
+                    949.0,
+                    1403.0,
+                    293.0,
+                    1403.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    195.0,
+                    478.0,
+                    195.0,
+                    478.0,
+                    240.0,
+                    293.0,
+                    240.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    839.0,
+                    2059.0,
+                    862.0,
+                    2059.0,
+                    862.0,
+                    2091.0,
+                    839.0,
+                    2091.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    289.0,
+                    985.0,
+                    371.0,
+                    985.0,
+                    371.0,
+                    1038.0,
+                    289.0,
+                    1038.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    558.0,
+                    985.0,
+                    609.0,
+                    985.0,
+                    609.0,
+                    1038.0,
+                    558.0,
+                    1038.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    800.0,
+                    985.0,
+                    874.0,
+                    985.0,
+                    874.0,
+                    1038.0,
+                    800.0,
+                    1038.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1108.0,
+                    985.0,
+                    1160.0,
+                    985.0,
+                    1160.0,
+                    1038.0,
+                    1108.0,
+                    1038.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1403.0,
+                    985.0,
+                    1411.0,
+                    985.0,
+                    1411.0,
+                    1038.0,
+                    1403.0,
+                    1038.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1024.0,
+                    617.0,
+                    1024.0,
+                    617.0,
+                    1062.0,
+                    294.0,
+                    1062.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    630.0,
+                    1024.0,
+                    1127.0,
+                    1024.0,
+                    1127.0,
+                    1062.0,
+                    630.0,
+                    1062.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1242.0,
+                    1024.0,
+                    1406.0,
+                    1024.0,
+                    1406.0,
+                    1062.0,
+                    1242.0,
+                    1062.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1056.0,
+                    364.0,
+                    1056.0,
+                    364.0,
+                    1095.0,
+                    294.0,
+                    1095.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    493.0,
+                    1056.0,
+                    564.0,
+                    1056.0,
+                    564.0,
+                    1095.0,
+                    493.0,
+                    1095.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    583.0,
+                    1056.0,
+                    1408.0,
+                    1056.0,
+                    1408.0,
+                    1095.0,
+                    583.0,
+                    1095.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1088.0,
+                    1404.0,
+                    1088.0,
+                    1404.0,
+                    1120.0,
+                    294.0,
+                    1120.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1116.0,
+                    1408.0,
+                    1116.0,
+                    1408.0,
+                    1152.0,
+                    294.0,
+                    1152.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1147.0,
+                    1407.0,
+                    1147.0,
+                    1407.0,
+                    1181.0,
+                    294.0,
+                    1181.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1179.0,
+                    584.0,
+                    1179.0,
+                    584.0,
+                    1214.0,
+                    295.0,
+                    1214.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    639.0,
+                    1179.0,
+                    1406.0,
+                    1179.0,
+                    1406.0,
+                    1214.0,
+                    639.0,
+                    1214.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1208.0,
+                    1212.0,
+                    1208.0,
+                    1212.0,
+                    1243.0,
+                    294.0,
+                    1243.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1231.0,
+                    1208.0,
+                    1406.0,
+                    1208.0,
+                    1406.0,
+                    1243.0,
+                    1231.0,
+                    1243.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1238.0,
+                    1406.0,
+                    1238.0,
+                    1406.0,
+                    1273.0,
+                    295.0,
+                    1273.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1269.0,
+                    1407.0,
+                    1269.0,
+                    1407.0,
+                    1304.0,
+                    294.0,
+                    1304.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1297.0,
+                    395.0,
+                    1297.0,
+                    395.0,
+                    1332.0,
+                    293.0,
+                    1332.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1415.0,
+                    711.0,
+                    1415.0,
+                    711.0,
+                    1459.0,
+                    294.0,
+                    1459.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    902.0,
+                    1415.0,
+                    1407.0,
+                    1415.0,
+                    1407.0,
+                    1459.0,
+                    902.0,
+                    1459.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1450.0,
+                    586.0,
+                    1450.0,
+                    586.0,
+                    1489.0,
+                    294.0,
+                    1489.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    615.0,
+                    1450.0,
+                    735.0,
+                    1450.0,
+                    735.0,
+                    1489.0,
+                    615.0,
+                    1489.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    907.0,
+                    1450.0,
+                    958.0,
+                    1450.0,
+                    958.0,
+                    1489.0,
+                    907.0,
+                    1489.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1102.0,
+                    1450.0,
+                    1333.0,
+                    1450.0,
+                    1333.0,
+                    1489.0,
+                    1102.0,
+                    1489.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1480.0,
+                    295.0,
+                    1480.0,
+                    295.0,
+                    1519.0,
+                    292.0,
+                    1519.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    574.0,
+                    1480.0,
+                    1407.0,
+                    1480.0,
+                    1407.0,
+                    1519.0,
+                    574.0,
+                    1519.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1513.0,
+                    1404.0,
+                    1513.0,
+                    1404.0,
+                    1547.0,
+                    294.0,
+                    1547.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1543.0,
+                    678.0,
+                    1543.0,
+                    678.0,
+                    1578.0,
+                    292.0,
+                    1578.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    698.0,
+                    1543.0,
+                    1125.0,
+                    1543.0,
+                    1125.0,
+                    1578.0,
+                    698.0,
+                    1578.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1150.0,
+                    1543.0,
+                    1406.0,
+                    1543.0,
+                    1406.0,
+                    1578.0,
+                    1150.0,
+                    1578.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1574.0,
+                    736.0,
+                    1574.0,
+                    736.0,
+                    1607.0,
+                    294.0,
+                    1607.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1637.0,
+                    1406.0,
+                    1637.0,
+                    1406.0,
+                    1672.0,
+                    294.0,
+                    1672.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1667.0,
+                    423.0,
+                    1667.0,
+                    423.0,
+                    1700.0,
+                    292.0,
+                    1700.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    513.0,
+                    1667.0,
+                    875.0,
+                    1667.0,
+                    875.0,
+                    1700.0,
+                    513.0,
+                    1700.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    977.0,
+                    1667.0,
+                    1239.0,
+                    1667.0,
+                    1239.0,
+                    1700.0,
+                    977.0,
+                    1700.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1403.0,
+                    1667.0,
+                    1406.0,
+                    1667.0,
+                    1406.0,
+                    1700.0,
+                    1403.0,
+                    1700.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1698.0,
+                    495.0,
+                    1698.0,
+                    495.0,
+                    1733.0,
+                    295.0,
+                    1733.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    884.0,
+                    1698.0,
+                    972.0,
+                    1698.0,
+                    972.0,
+                    1733.0,
+                    884.0,
+                    1733.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1114.0,
+                    1698.0,
+                    1122.0,
+                    1698.0,
+                    1122.0,
+                    1733.0,
+                    1114.0,
+                    1733.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1292.0,
+                    1698.0,
+                    1408.0,
+                    1698.0,
+                    1408.0,
+                    1733.0,
+                    1292.0,
+                    1733.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1728.0,
+                    830.0,
+                    1728.0,
+                    830.0,
+                    1764.0,
+                    295.0,
+                    1764.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    919.0,
+                    1728.0,
+                    1406.0,
+                    1728.0,
+                    1406.0,
+                    1764.0,
+                    919.0,
+                    1764.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1752.0,
+                    1018.0,
+                    1752.0,
+                    1018.0,
+                    1799.0,
+                    291.0,
+                    1799.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1064.0,
+                    1752.0,
+                    1097.0,
+                    1752.0,
+                    1097.0,
+                    1799.0,
+                    1064.0,
+                    1799.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1310.0,
+                    1752.0,
+                    1412.0,
+                    1752.0,
+                    1412.0,
+                    1799.0,
+                    1310.0,
+                    1799.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1786.0,
+                    981.0,
+                    1786.0,
+                    981.0,
+                    1824.0,
+                    294.0,
+                    1824.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    706.0,
+                    1404.0,
+                    706.0,
+                    1404.0,
+                    738.0,
+                    296.0,
+                    738.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    734.0,
+                    1404.0,
+                    734.0,
+                    1404.0,
+                    769.0,
+                    294.0,
+                    769.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    762.0,
+                    296.0,
+                    762.0,
+                    296.0,
+                    801.0,
+                    292.0,
+                    801.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    334.0,
+                    762.0,
+                    1408.0,
+                    762.0,
+                    1408.0,
+                    801.0,
+                    334.0,
+                    801.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    794.0,
+                    1407.0,
+                    794.0,
+                    1407.0,
+                    831.0,
+                    292.0,
+                    831.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    507.0,
+                    826.0,
+                    1303.0,
+                    826.0,
+                    1303.0,
+                    862.0,
+                    507.0,
+                    862.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1336.0,
+                    826.0,
+                    1406.0,
+                    826.0,
+                    1406.0,
+                    862.0,
+                    1336.0,
+                    862.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    854.0,
+                    442.0,
+                    854.0,
+                    442.0,
+                    893.0,
+                    294.0,
+                    893.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    474.0,
+                    854.0,
+                    1077.0,
+                    854.0,
+                    1077.0,
+                    893.0,
+                    474.0,
+                    893.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    267.0,
+                    1403.0,
+                    267.0,
+                    1403.0,
+                    299.0,
+                    296.0,
+                    299.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    296.0,
+                    1405.0,
+                    296.0,
+                    1405.0,
+                    331.0,
+                    292.0,
+                    331.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    328.0,
+                    1403.0,
+                    328.0,
+                    1403.0,
+                    360.0,
+                    296.0,
+                    360.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    355.0,
+                    1405.0,
+                    355.0,
+                    1405.0,
+                    392.0,
+                    294.0,
+                    392.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    387.0,
+                    1405.0,
+                    387.0,
+                    1405.0,
+                    421.0,
+                    292.0,
+                    421.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    419.0,
+                    430.0,
+                    419.0,
+                    430.0,
+                    449.0,
+                    293.0,
+                    449.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1913.0,
+                    902.0,
+                    1913.0,
+                    902.0,
+                    1947.0,
+                    294.0,
+                    1947.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    932.0,
+                    1913.0,
+                    1368.0,
+                    1913.0,
+                    1368.0,
+                    1947.0,
+                    932.0,
+                    1947.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1941.0,
+                    1357.0,
+                    1941.0,
+                    1357.0,
+                    1985.0,
+                    291.0,
+                    1985.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1401.0,
+                    1941.0,
+                    1404.0,
+                    1941.0,
+                    1404.0,
+                    1985.0,
+                    1401.0,
+                    1985.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1976.0,
+                    849.0,
+                    1976.0,
+                    849.0,
+                    2012.0,
+                    293.0,
+                    2012.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    459.0,
+                    1283.0,
+                    459.0,
+                    1283.0,
+                    499.0,
+                    292.0,
+                    499.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1373.0,
+                    459.0,
+                    1405.0,
+                    459.0,
+                    1405.0,
+                    499.0,
+                    1373.0,
+                    499.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    494.0,
+                    1408.0,
+                    494.0,
+                    1408.0,
+                    528.0,
+                    295.0,
+                    528.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    521.0,
+                    473.0,
+                    521.0,
+                    473.0,
+                    560.0,
+                    294.0,
+                    560.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    657.0,
+                    370.0,
+                    657.0,
+                    370.0,
+                    696.0,
+                    296.0,
+                    696.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    447.0,
+                    657.0,
+                    730.0,
+                    657.0,
+                    730.0,
+                    696.0,
+                    447.0,
+                    696.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    790.0,
+                    657.0,
+                    1345.0,
+                    657.0,
+                    1345.0,
+                    696.0,
+                    790.0,
+                    696.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 3,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1270,
+                    1405,
+                    1270,
+                    1405,
+                    1544,
+                    297,
+                    1544
+                ],
+                "score": 0.983
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1655,
+                    1404,
+                    1655,
+                    1404,
+                    1810,
+                    297,
+                    1810
+                ],
+                "score": 0.979
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1822,
+                    1404,
+                    1822,
+                    1404,
+                    2008,
+                    298,
+                    2008
+                ],
+                "score": 0.978
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    861,
+                    1405,
+                    861,
+                    1405,
+                    1019,
+                    297,
+                    1019
+                ],
+                "score": 0.976
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1117,
+                    1405,
+                    1117,
+                    1405,
+                    1242,
+                    298,
+                    1242
+                ],
+                "score": 0.976
+            },
+            {
+                "category_id": 3,
+                "poly": [
+                    418,
+                    211,
+                    1277,
+                    211,
+                    1277,
+                    690,
+                    418,
+                    690
+                ],
+                "score": 0.972
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    639,
+                    1543,
+                    1059,
+                    1543,
+                    1059,
+                    1592,
+                    639,
+                    1592
+                ],
+                "score": 0.932
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    300,
+                    1595,
+                    741,
+                    1595,
+                    741,
+                    1627,
+                    300,
+                    1627
+                ],
+                "score": 0.919
+            },
+            {
+                "category_id": 9,
+                "poly": [
+                    1366,
+                    1030,
+                    1399,
+                    1030,
+                    1399,
+                    1059,
+                    1366,
+                    1059
+                ],
+                "score": 0.884
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    301,
+                    1067,
+                    770,
+                    1067,
+                    770,
+                    1100,
+                    301,
+                    1100
+                ],
+                "score": 0.868
+            },
+            {
+                "category_id": 9,
+                "poly": [
+                    1366,
+                    1547,
+                    1400,
+                    1547,
+                    1400,
+                    1576,
+                    1366,
+                    1576
+                ],
+                "score": 0.867
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    841,
+                    2061,
+                    858,
+                    2061,
+                    858,
+                    2085,
+                    841,
+                    2085
+                ],
+                "score": 0.728
+            },
+            {
+                "category_id": 8,
+                "poly": [
+                    709,
+                    1025,
+                    990,
+                    1025,
+                    990,
+                    1064,
+                    709,
+                    1064
+                ],
+                "score": 0.684
+            },
+            {
+                "category_id": 4,
+                "poly": [
+                    296,
+                    749,
+                    1402,
+                    749,
+                    1402,
+                    813,
+                    296,
+                    813
+                ],
+                "score": 0.642
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    296,
+                    749,
+                    1402,
+                    749,
+                    1402,
+                    813,
+                    296,
+                    813
+                ],
+                "score": 0.495
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    429,
+                    986,
+                    530,
+                    986,
+                    530,
+                    1019,
+                    429,
+                    1019
+                ],
+                "score": 0.93,
+                "latex": "\\delta ( \\varepsilon ) \\geq 0"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    647,
+                    865,
+                    787,
+                    865,
+                    787,
+                    897,
+                    647,
+                    897
+                ],
+                "score": 0.93,
+                "latex": "x ^ { * } \\in [ 0 , 1 ] ^ { m }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    442,
+                    1394,
+                    674,
+                    1394,
+                    674,
+                    1428,
+                    442,
+                    1428
+                ],
+                "score": 0.92,
+                "latex": "\\mathrm { m i n } _ { z } \\parallel G _ { w } ( z ) - x ^ { * } \\parallel"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1077,
+                    954,
+                    1334,
+                    954,
+                    1334,
+                    989,
+                    1077,
+                    989
+                ],
+                "score": 0.91,
+                "latex": "\\mathring { G } _ { w } : [ 0 , 1 ] ^ { k } \\mapsto [ 0 , 1 ] ^ { m }"
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    637,
+                    1541,
+                    1058,
+                    1541,
+                    1058,
+                    1594,
+                    637,
+                    1594
+                ],
+                "score": 0.91,
+                "latex": "\\operatorname * { m i n } _ { w _ { 1 } , . . . , w _ { B } } \\ c \\left( G _ { w _ { 1 } } ( z _ { 1 } ) , . . . , G _ { w _ { B } } ( z _ { B } ) \\right) \\ ,"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    371,
+                    1601,
+                    492,
+                    1601,
+                    492,
+                    1627,
+                    371,
+                    1627
+                ],
+                "score": 0.91,
+                "latex": "z _ { 1 } , \\dots , z _ { B }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    546,
+                    1067,
+                    654,
+                    1067,
+                    654,
+                    1101,
+                    546,
+                    1101
+                ],
+                "score": 0.89,
+                "latex": "\\delta ( \\varepsilon )  0"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    297,
+                    957,
+                    366,
+                    957,
+                    366,
+                    986,
+                    297,
+                    986
+                ],
+                "score": 0.88,
+                "latex": "\\varepsilon \\geq 0"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    772,
+                    1151,
+                    803,
+                    1151,
+                    803,
+                    1178,
+                    772,
+                    1178
+                ],
+                "score": 0.88,
+                "latex": "x ^ { * }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    590,
+                    925,
+                    618,
+                    925,
+                    618,
+                    956,
+                    590,
+                    956
+                ],
+                "score": 0.87,
+                "latex": "\\ell _ { 2 }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    644,
+                    1977,
+                    672,
+                    1977,
+                    672,
+                    2008,
+                    644,
+                    2008
+                ],
+                "score": 0.87,
+                "latex": "f _ { \\theta }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    738,
+                    1456,
+                    768,
+                    1456,
+                    768,
+                    1483,
+                    738,
+                    1483
+                ],
+                "score": 0.87,
+                "latex": "z ^ { * }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    812,
+                    1457,
+                    842,
+                    1457,
+                    842,
+                    1482,
+                    812,
+                    1482
+                ],
+                "score": 0.87,
+                "latex": "x ^ { * }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    745,
+                    1396,
+                    776,
+                    1396,
+                    776,
+                    1423,
+                    745,
+                    1423
+                ],
+                "score": 0.86,
+                "latex": "x ^ { * }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    798,
+                    957,
+                    829,
+                    957,
+                    829,
+                    985,
+                    798,
+                    985
+                ],
+                "score": 0.86,
+                "latex": "x ^ { * }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1315,
+                    927,
+                    1345,
+                    927,
+                    1345,
+                    952,
+                    1315,
+                    952
+                ],
+                "score": 0.86,
+                "latex": "x ^ { * }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    382,
+                    1456,
+                    413,
+                    1456,
+                    413,
+                    1483,
+                    382,
+                    1483
+                ],
+                "score": 0.85,
+                "latex": "x ^ { * }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    689,
+                    1070,
+                    761,
+                    1070,
+                    761,
+                    1096,
+                    689,
+                    1096
+                ],
+                "score": 0.81,
+                "latex": "\\varepsilon \\to 0"
+            },
+            {
+                "category_id": 14,
+                "poly": [
+                    705,
+                    1025,
+                    988,
+                    1025,
+                    988,
+                    1064,
+                    705,
+                    1064
+                ],
+                "score": 0.81,
+                "latex": "\\| G _ { w } ( z ^ { * } ) - x ^ { * } \\| \\leq \\delta ( \\varepsilon ) ,"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1219,
+                    1722,
+                    1243,
+                    1722,
+                    1243,
+                    1745,
+                    1219,
+                    1745
+                ],
+                "score": 0.79,
+                "latex": "w"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    968,
+                    1491,
+                    992,
+                    1491,
+                    992,
+                    1513,
+                    968,
+                    1513
+                ],
+                "score": 0.79,
+                "latex": "w"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1151,
+                    1722,
+                    1169,
+                    1722,
+                    1169,
+                    1745,
+                    1151,
+                    1745
+                ],
+                "score": 0.77,
+                "latex": "z"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1164,
+                    1430,
+                    1188,
+                    1430,
+                    1188,
+                    1452,
+                    1164,
+                    1452
+                ],
+                "score": 0.76,
+                "latex": "w"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    829,
+                    1664,
+                    854,
+                    1664,
+                    854,
+                    1685,
+                    829,
+                    1685
+                ],
+                "score": 0.76,
+                "latex": "w"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    984,
+                    757,
+                    1002,
+                    757,
+                    1002,
+                    778,
+                    984,
+                    778
+                ],
+                "score": 0.75,
+                "latex": "z"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1117,
+                    1665,
+                    1136,
+                    1665,
+                    1136,
+                    1685,
+                    1117,
+                    1685
+                ],
+                "score": 0.75,
+                "latex": "z"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    727,
+                    986,
+                    1007,
+                    986,
+                    1007,
+                    1019,
+                    727,
+                    1019
+                ],
+                "score": 0.74,
+                "latex": "z ^ { * } \\in \\arg \\operatorname* { m i n } _ { z } c ( G _ { w } ( z ) ) ,"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    413,
+                    1754,
+                    438,
+                    1754,
+                    438,
+                    1776,
+                    413,
+                    1776
+                ],
+                "score": 0.74,
+                "latex": "w"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    359,
+                    1521,
+                    376,
+                    1521,
+                    376,
+                    1543,
+                    359,
+                    1543
+                ],
+                "score": 0.72,
+                "latex": "z"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1377,
+                    758,
+                    1401,
+                    758,
+                    1401,
+                    778,
+                    1377,
+                    778
+                ],
+                "score": 0.68,
+                "latex": "w"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    543,
+                    926,
+                    561,
+                    926,
+                    561,
+                    952,
+                    543,
+                    952
+                ],
+                "score": 0.6,
+                "latex": "d"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    672,
+                    1182,
+                    708,
+                    1182,
+                    708,
+                    1209,
+                    672,
+                    1209
+                ],
+                "score": 0.3,
+                "latex": "\\mathrm { F L }"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1039.0,
+                    210.0,
+                    1205.0,
+                    210.0,
+                    1205.0,
+                    245.0,
+                    1039.0,
+                    245.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1047.0,
+                    247.0,
+                    1164.0,
+                    247.0,
+                    1164.0,
+                    280.0,
+                    1047.0,
+                    280.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    416.0,
+                    291.0,
+                    528.0,
+                    291.0,
+                    528.0,
+                    322.0,
+                    416.0,
+                    322.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    778.0,
+                    310.0,
+                    893.0,
+                    310.0,
+                    893.0,
+                    342.0,
+                    778.0,
+                    342.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1097.0,
+                    310.0,
+                    1118.0,
+                    310.0,
+                    1118.0,
+                    324.0,
+                    1097.0,
+                    324.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1138.0,
+                    291.0,
+                    1176.0,
+                    291.0,
+                    1176.0,
+                    337.0,
+                    1138.0,
+                    337.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    441.0,
+                    363.0,
+                    469.0,
+                    363.0,
+                    469.0,
+                    395.0,
+                    441.0,
+                    395.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    600.0,
+                    390.0,
+                    676.0,
+                    390.0,
+                    676.0,
+                    468.0,
+                    600.0,
+                    468.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1020.0,
+                    401.0,
+                    1193.0,
+                    401.0,
+                    1193.0,
+                    433.0,
+                    1020.0,
+                    433.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    745.0,
+                    427.0,
+                    888.0,
+                    427.0,
+                    888.0,
+                    463.0,
+                    745.0,
+                    463.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1097.0,
+                    453.0,
+                    1114.0,
+                    453.0,
+                    1114.0,
+                    471.0,
+                    1097.0,
+                    471.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1127.0,
+                    439.0,
+                    1243.0,
+                    439.0,
+                    1243.0,
+                    469.0,
+                    1127.0,
+                    469.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    528.0,
+                    479.0,
+                    543.0,
+                    479.0,
+                    543.0,
+                    500.0,
+                    528.0,
+                    500.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    627.0,
+                    473.0,
+                    650.0,
+                    473.0,
+                    650.0,
+                    505.0,
+                    627.0,
+                    505.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    731.0,
+                    470.0,
+                    905.0,
+                    470.0,
+                    905.0,
+                    533.0,
+                    731.0,
+                    533.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1011.0,
+                    480.0,
+                    1200.0,
+                    480.0,
+                    1200.0,
+                    522.0,
+                    1011.0,
+                    522.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    604.0,
+                    514.0,
+                    670.0,
+                    514.0,
+                    670.0,
+                    576.0,
+                    604.0,
+                    576.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    743.0,
+                    535.0,
+                    885.0,
+                    535.0,
+                    885.0,
+                    566.0,
+                    743.0,
+                    566.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1006.0,
+                    512.0,
+                    1202.0,
+                    512.0,
+                    1202.0,
+                    547.0,
+                    1006.0,
+                    547.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    475.0,
+                    576.0,
+                    592.0,
+                    576.0,
+                    592.0,
+                    612.0,
+                    475.0,
+                    612.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    603.0,
+                    578.0,
+                    673.0,
+                    578.0,
+                    673.0,
+                    609.0,
+                    603.0,
+                    609.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    473.0,
+                    605.0,
+                    597.0,
+                    605.0,
+                    597.0,
+                    638.0,
+                    473.0,
+                    638.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    611.0,
+                    598.0,
+                    668.0,
+                    598.0,
+                    668.0,
+                    632.0,
+                    611.0,
+                    632.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    747.0,
+                    614.0,
+                    806.0,
+                    614.0,
+                    806.0,
+                    676.0,
+                    747.0,
+                    676.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    839.0,
+                    635.0,
+                    850.0,
+                    635.0,
+                    850.0,
+                    648.0,
+                    839.0,
+                    648.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    886.0,
+                    622.0,
+                    1012.0,
+                    622.0,
+                    1012.0,
+                    654.0,
+                    886.0,
+                    654.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1082.0,
+                    633.0,
+                    1130.0,
+                    633.0,
+                    1130.0,
+                    666.0,
+                    1082.0,
+                    666.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    920.0,
+                    647.0,
+                    981.0,
+                    647.0,
+                    981.0,
+                    674.0,
+                    920.0,
+                    674.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    513.0,
+                    356.0,
+                    554.0,
+                    356.0,
+                    554.0,
+                    402.0,
+                    513.0,
+                    402.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    871.25,
+                    392.5,
+                    884.25,
+                    392.5,
+                    884.25,
+                    406.0,
+                    871.25,
+                    406.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    889.0,
+                    404.0,
+                    896.0,
+                    404.0,
+                    896.0,
+                    415.5,
+                    889.0,
+                    415.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    905.0,
+                    420.0,
+                    925.0,
+                    420.0,
+                    925.0,
+                    434.0,
+                    905.0,
+                    434.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1131.0,
+                    638.0,
+                    1191.0,
+                    638.0,
+                    1191.0,
+                    669.5,
+                    1131.0,
+                    669.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    839.0,
+                    2058.0,
+                    861.0,
+                    2058.0,
+                    861.0,
+                    2093.0,
+                    839.0,
+                    2093.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    748.0,
+                    983.0,
+                    748.0,
+                    983.0,
+                    786.0,
+                    294.0,
+                    786.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1003.0,
+                    748.0,
+                    1376.0,
+                    748.0,
+                    1376.0,
+                    786.0,
+                    1003.0,
+                    786.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    780.0,
+                    865.0,
+                    780.0,
+                    865.0,
+                    816.0,
+                    294.0,
+                    816.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1271.0,
+                    1405.0,
+                    1271.0,
+                    1405.0,
+                    1307.0,
+                    295.0,
+                    1307.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1303.0,
+                    1405.0,
+                    1303.0,
+                    1405.0,
+                    1339.0,
+                    295.0,
+                    1339.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1333.0,
+                    1406.0,
+                    1333.0,
+                    1406.0,
+                    1369.0,
+                    292.0,
+                    1369.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1362.0,
+                    1406.0,
+                    1362.0,
+                    1406.0,
+                    1399.0,
+                    291.0,
+                    1399.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1391.0,
+                    441.0,
+                    1391.0,
+                    441.0,
+                    1434.0,
+                    292.0,
+                    1434.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    675.0,
+                    1391.0,
+                    744.0,
+                    1391.0,
+                    744.0,
+                    1434.0,
+                    675.0,
+                    1434.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    777.0,
+                    1391.0,
+                    1407.0,
+                    1391.0,
+                    1407.0,
+                    1434.0,
+                    777.0,
+                    1434.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1424.0,
+                    1163.0,
+                    1424.0,
+                    1163.0,
+                    1460.0,
+                    295.0,
+                    1460.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1189.0,
+                    1424.0,
+                    1405.0,
+                    1424.0,
+                    1405.0,
+                    1460.0,
+                    1189.0,
+                    1460.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1452.0,
+                    381.0,
+                    1452.0,
+                    381.0,
+                    1491.0,
+                    294.0,
+                    1491.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    414.0,
+                    1452.0,
+                    737.0,
+                    1452.0,
+                    737.0,
+                    1491.0,
+                    414.0,
+                    1491.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    769.0,
+                    1452.0,
+                    811.0,
+                    1452.0,
+                    811.0,
+                    1491.0,
+                    769.0,
+                    1491.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    843.0,
+                    1452.0,
+                    1407.0,
+                    1452.0,
+                    1407.0,
+                    1491.0,
+                    843.0,
+                    1491.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1481.0,
+                    967.0,
+                    1481.0,
+                    967.0,
+                    1523.0,
+                    292.0,
+                    1523.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    993.0,
+                    1481.0,
+                    1405.0,
+                    1481.0,
+                    1405.0,
+                    1523.0,
+                    993.0,
+                    1523.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1517.0,
+                    358.0,
+                    1517.0,
+                    358.0,
+                    1551.0,
+                    294.0,
+                    1551.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    377.0,
+                    1517.0,
+                    390.0,
+                    1517.0,
+                    390.0,
+                    1551.0,
+                    377.0,
+                    1551.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1655.0,
+                    828.0,
+                    1655.0,
+                    828.0,
+                    1691.0,
+                    295.0,
+                    1691.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    855.0,
+                    1655.0,
+                    1116.0,
+                    1655.0,
+                    1116.0,
+                    1691.0,
+                    855.0,
+                    1691.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1137.0,
+                    1655.0,
+                    1405.0,
+                    1655.0,
+                    1405.0,
+                    1691.0,
+                    1137.0,
+                    1691.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1686.0,
+                    1406.0,
+                    1686.0,
+                    1406.0,
+                    1725.0,
+                    292.0,
+                    1725.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1717.0,
+                    1150.0,
+                    1717.0,
+                    1150.0,
+                    1752.0,
+                    292.0,
+                    1752.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1170.0,
+                    1717.0,
+                    1218.0,
+                    1717.0,
+                    1218.0,
+                    1752.0,
+                    1170.0,
+                    1752.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1244.0,
+                    1717.0,
+                    1405.0,
+                    1717.0,
+                    1405.0,
+                    1752.0,
+                    1244.0,
+                    1752.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1744.0,
+                    412.0,
+                    1744.0,
+                    412.0,
+                    1788.0,
+                    291.0,
+                    1788.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    439.0,
+                    1744.0,
+                    1405.0,
+                    1744.0,
+                    1405.0,
+                    1788.0,
+                    439.0,
+                    1788.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1780.0,
+                    559.0,
+                    1780.0,
+                    559.0,
+                    1812.0,
+                    295.0,
+                    1812.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1821.0,
+                    1409.0,
+                    1821.0,
+                    1409.0,
+                    1859.0,
+                    293.0,
+                    1859.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1854.0,
+                    1408.0,
+                    1854.0,
+                    1408.0,
+                    1889.0,
+                    293.0,
+                    1889.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1884.0,
+                    1404.0,
+                    1884.0,
+                    1404.0,
+                    1919.0,
+                    292.0,
+                    1919.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1912.0,
+                    1405.0,
+                    1912.0,
+                    1405.0,
+                    1951.0,
+                    292.0,
+                    1951.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1946.0,
+                    1405.0,
+                    1946.0,
+                    1405.0,
+                    1981.0,
+                    294.0,
+                    1981.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1974.0,
+                    643.0,
+                    1974.0,
+                    643.0,
+                    2009.0,
+                    294.0,
+                    2009.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    673.0,
+                    1974.0,
+                    959.0,
+                    1974.0,
+                    959.0,
+                    2009.0,
+                    673.0,
+                    2009.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    859.0,
+                    646.0,
+                    859.0,
+                    646.0,
+                    902.0,
+                    294.0,
+                    902.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    788.0,
+                    859.0,
+                    1405.0,
+                    859.0,
+                    1405.0,
+                    902.0,
+                    788.0,
+                    902.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    893.0,
+                    1406.0,
+                    893.0,
+                    1406.0,
+                    929.0,
+                    292.0,
+                    929.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    923.0,
+                    542.0,
+                    923.0,
+                    542.0,
+                    960.0,
+                    294.0,
+                    960.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    562.0,
+                    923.0,
+                    589.0,
+                    923.0,
+                    589.0,
+                    960.0,
+                    562.0,
+                    960.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    619.0,
+                    923.0,
+                    1314.0,
+                    923.0,
+                    1314.0,
+                    960.0,
+                    619.0,
+                    960.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1346.0,
+                    923.0,
+                    1407.0,
+                    923.0,
+                    1407.0,
+                    960.0,
+                    1346.0,
+                    960.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    952.0,
+                    296.0,
+                    952.0,
+                    296.0,
+                    994.0,
+                    292.0,
+                    994.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    367.0,
+                    952.0,
+                    797.0,
+                    952.0,
+                    797.0,
+                    994.0,
+                    367.0,
+                    994.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    830.0,
+                    952.0,
+                    1076.0,
+                    952.0,
+                    1076.0,
+                    994.0,
+                    830.0,
+                    994.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1335.0,
+                    952.0,
+                    1410.0,
+                    952.0,
+                    1410.0,
+                    994.0,
+                    1335.0,
+                    994.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    983.0,
+                    428.0,
+                    983.0,
+                    428.0,
+                    1025.0,
+                    292.0,
+                    1025.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    531.0,
+                    983.0,
+                    726.0,
+                    983.0,
+                    726.0,
+                    1025.0,
+                    531.0,
+                    1025.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1008.0,
+                    983.0,
+                    1017.0,
+                    983.0,
+                    1017.0,
+                    1025.0,
+                    1008.0,
+                    1025.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1120.0,
+                    1403.0,
+                    1120.0,
+                    1403.0,
+                    1153.0,
+                    295.0,
+                    1153.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1147.0,
+                    771.0,
+                    1147.0,
+                    771.0,
+                    1188.0,
+                    292.0,
+                    1188.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    804.0,
+                    1147.0,
+                    1406.0,
+                    1147.0,
+                    1406.0,
+                    1188.0,
+                    804.0,
+                    1188.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1182.0,
+                    671.0,
+                    1182.0,
+                    671.0,
+                    1215.0,
+                    295.0,
+                    1215.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    709.0,
+                    1182.0,
+                    1405.0,
+                    1182.0,
+                    1405.0,
+                    1215.0,
+                    709.0,
+                    1215.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1214.0,
+                    600.0,
+                    1214.0,
+                    600.0,
+                    1243.0,
+                    293.0,
+                    1243.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1590.0,
+                    370.0,
+                    1590.0,
+                    370.0,
+                    1633.0,
+                    295.0,
+                    1633.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    493.0,
+                    1590.0,
+                    746.0,
+                    1590.0,
+                    746.0,
+                    1633.0,
+                    493.0,
+                    1633.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1064.0,
+                    545.0,
+                    1064.0,
+                    545.0,
+                    1104.0,
+                    294.0,
+                    1104.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    655.0,
+                    1064.0,
+                    688.0,
+                    1064.0,
+                    688.0,
+                    1104.0,
+                    655.0,
+                    1104.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    762.0,
+                    1064.0,
+                    771.0,
+                    1064.0,
+                    771.0,
+                    1104.0,
+                    762.0,
+                    1104.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    748.0,
+                    983.0,
+                    748.0,
+                    983.0,
+                    786.0,
+                    294.0,
+                    786.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1003.0,
+                    748.0,
+                    1376.0,
+                    748.0,
+                    1376.0,
+                    786.0,
+                    1003.0,
+                    786.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    780.0,
+                    865.0,
+                    780.0,
+                    865.0,
+                    816.0,
+                    294.0,
+                    816.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 4,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1011,
+                    1404,
+                    1011,
+                    1404,
+                    1467,
+                    297,
+                    1467
+                ],
+                "score": 0.983
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1605,
+                    1404,
+                    1605,
+                    1404,
+                    1821,
+                    297,
+                    1821
+                ],
+                "score": 0.978
+            },
+            {
+                "category_id": 3,
+                "poly": [
+                    287,
+                    257,
+                    1391,
+                    257,
+                    1391,
+                    731,
+                    287,
+                    731
+                ],
+                "score": 0.975
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    1854,
+                    1404,
+                    1854,
+                    1404,
+                    2008,
+                    299,
+                    2008
+                ],
+                "score": 0.974
+            },
+            {
+                "category_id": 4,
+                "poly": [
+                    296,
+                    750,
+                    1405,
+                    750,
+                    1405,
+                    873,
+                    296,
+                    873
+                ],
+                "score": 0.961
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    298,
+                    948,
+                    961,
+                    948,
+                    961,
+                    980,
+                    298,
+                    980
+                ],
+                "score": 0.927
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    298,
+                    1526,
+                    532,
+                    1526,
+                    532,
+                    1564,
+                    298,
+                    1564
+                ],
+                "score": 0.911
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    840,
+                    2062,
+                    858,
+                    2062,
+                    858,
+                    2085,
+                    840,
+                    2085
+                ],
+                "score": 0.78
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    640,
+                    1041,
+                    804,
+                    1041,
+                    804,
+                    1075,
+                    640,
+                    1075
+                ],
+                "score": 0.94,
+                "latex": "\\mathcal { S } = \\{ ( \\theta _ { i } , g _ { i } ) \\}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    969,
+                    1224,
+                    1180,
+                    1224,
+                    1180,
+                    1257,
+                    969,
+                    1257
+                ],
+                "score": 0.92,
+                "latex": "\\boldsymbol { z } _ { i } ^ { * } = ( z _ { i 1 } ^ { * } , . . . , z _ { i B } ^ { * } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1074,
+                    1727,
+                    1140,
+                    1727,
+                    1140,
+                    1761,
+                    1074,
+                    1761
+                ],
+                "score": 0.91,
+                "latex": "d ( \\cdot , \\cdot )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    781,
+                    1344,
+                    1188,
+                    1344,
+                    1188,
+                    1379,
+                    781,
+                    1379
+                ],
+                "score": 0.91,
+                "latex": "\\begin{array} { r } { \\sum _ { i } c ( G _ { w ^ { \\prime } } ( z _ { i 1 } ^ { * } ) , . . . , \\bar { G } _ { w ^ { \\prime } } ( z _ { i B } ^ { * } ) ; \\theta _ { i } , g _ { i } ) } \\end{array}"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1173,
+                    1698,
+                    1274,
+                    1698,
+                    1274,
+                    1726,
+                    1173,
+                    1726
+                ],
+                "score": 0.9,
+                "latex": "B \\ = \\ 4"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    496,
+                    1316,
+                    528,
+                    1316,
+                    528,
+                    1347,
+                    496,
+                    1347
+                ],
+                "score": 0.9,
+                "latex": "z _ { i } ^ { * }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    796,
+                    1637,
+                    898,
+                    1637,
+                    898,
+                    1667,
+                    796,
+                    1667
+                ],
+                "score": 0.9,
+                "latex": "6 4 \\times 6 4"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    496,
+                    1042,
+                    538,
+                    1042,
+                    538,
+                    1072,
+                    496,
+                    1072
+                ],
+                "score": 0.89,
+                "latex": "G _ { w }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    986,
+                    1286,
+                    1013,
+                    1286,
+                    1013,
+                    1315,
+                    986,
+                    1315
+                ],
+                "score": 0.88,
+                "latex": "\\ell _ { 2 }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    920,
+                    1255,
+                    947,
+                    1255,
+                    947,
+                    1284,
+                    920,
+                    1284
+                ],
+                "score": 0.87,
+                "latex": "\\ell _ { 2 }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1099,
+                    1133,
+                    1130,
+                    1133,
+                    1130,
+                    1160,
+                    1099,
+                    1160
+                ],
+                "score": 0.87,
+                "latex": "w ^ { \\prime }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    532,
+                    1194,
+                    559,
+                    1194,
+                    559,
+                    1225,
+                    532,
+                    1225
+                ],
+                "score": 0.87,
+                "latex": "\\ell _ { 2 }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    466,
+                    811,
+                    558,
+                    811,
+                    558,
+                    844,
+                    466,
+                    844
+                ],
+                "score": 0.86,
+                "latex": "{ \\mathrm { G I } } { - z } / w"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    750,
+                    1316,
+                    781,
+                    1316,
+                    781,
+                    1347,
+                    750,
+                    1347
+                ],
+                "score": 0.84,
+                "latex": "z _ { i } ^ { * }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    812,
+                    1315,
+                    843,
+                    1315,
+                    843,
+                    1343,
+                    812,
+                    1343
+                ],
+                "score": 0.84,
+                "latex": "w ^ { \\prime }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    339,
+                    1164,
+                    367,
+                    1164,
+                    367,
+                    1190,
+                    339,
+                    1190
+                ],
+                "score": 0.81,
+                "latex": "N"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    729,
+                    1137,
+                    753,
+                    1137,
+                    753,
+                    1160,
+                    729,
+                    1160
+                ],
+                "score": 0.79,
+                "latex": "w"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    651,
+                    1164,
+                    673,
+                    1164,
+                    673,
+                    1191,
+                    651,
+                    1191
+                ],
+                "score": 0.77,
+                "latex": "s"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1185,
+                    1890,
+                    1203,
+                    1890,
+                    1203,
+                    1912,
+                    1185,
+                    1912
+                ],
+                "score": 0.77,
+                "latex": "z"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1125,
+                    1950,
+                    1143,
+                    1950,
+                    1143,
+                    1973,
+                    1125,
+                    1973
+                ],
+                "score": 0.77,
+                "latex": "x"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    992,
+                    1891,
+                    1010,
+                    1891,
+                    1010,
+                    1912,
+                    992,
+                    1912
+                ],
+                "score": 0.76,
+                "latex": "x"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    678,
+                    1381,
+                    702,
+                    1381,
+                    702,
+                    1403,
+                    678,
+                    1403
+                ],
+                "score": 0.75,
+                "latex": "w"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    708,
+                    1411,
+                    732,
+                    1411,
+                    732,
+                    1433,
+                    708,
+                    1433
+                ],
+                "score": 0.74,
+                "latex": "w"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1313,
+                    1410,
+                    1337,
+                    1410,
+                    1337,
+                    1433,
+                    1313,
+                    1433
+                ],
+                "score": 0.74,
+                "latex": "w"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    499,
+                    1946,
+                    550,
+                    1946,
+                    550,
+                    1979,
+                    499,
+                    1979
+                ],
+                "score": 0.74,
+                "latex": "z / w"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    418,
+                    1920,
+                    442,
+                    1920,
+                    442,
+                    1942,
+                    418,
+                    1942
+                ],
+                "score": 0.72,
+                "latex": "w"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1332,
+                    1226,
+                    1345,
+                    1226,
+                    1345,
+                    1251,
+                    1332,
+                    1251
+                ],
+                "score": 0.71,
+                "latex": "i"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    423.0,
+                    262.0,
+                    538.0,
+                    262.0,
+                    538.0,
+                    279.0,
+                    423.0,
+                    279.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    872.0,
+                    262.0,
+                    917.0,
+                    262.0,
+                    917.0,
+                    280.0,
+                    872.0,
+                    280.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    367.0,
+                    276.0,
+                    403.0,
+                    276.0,
+                    403.0,
+                    293.0,
+                    367.0,
+                    293.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    459.0,
+                    277.0,
+                    482.0,
+                    277.0,
+                    482.0,
+                    290.0,
+                    459.0,
+                    290.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    541.0,
+                    275.0,
+                    569.0,
+                    275.0,
+                    569.0,
+                    293.0,
+                    541.0,
+                    293.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    625.0,
+                    276.0,
+                    652.0,
+                    276.0,
+                    652.0,
+                    290.0,
+                    625.0,
+                    290.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    710.0,
+                    276.0,
+                    736.0,
+                    276.0,
+                    736.0,
+                    290.0,
+                    710.0,
+                    290.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    794.0,
+                    276.0,
+                    820.0,
+                    276.0,
+                    820.0,
+                    290.0,
+                    794.0,
+                    290.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    879.0,
+                    277.0,
+                    910.0,
+                    277.0,
+                    910.0,
+                    290.0,
+                    879.0,
+                    290.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    417.0,
+                    283.0,
+                    941.0,
+                    283.0,
+                    941.0,
+                    380.0,
+                    417.0,
+                    380.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    944.0,
+                    288.0,
+                    977.0,
+                    288.0,
+                    977.0,
+                    306.0,
+                    944.0,
+                    306.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    307.0,
+                    335.0,
+                    307.0,
+                    335.0,
+                    325.0,
+                    293.0,
+                    325.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1229.0,
+                    303.0,
+                    1380.0,
+                    303.0,
+                    1380.0,
+                    333.0,
+                    1229.0,
+                    333.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    321.0,
+                    331.0,
+                    321.0,
+                    331.0,
+                    339.0,
+                    295.0,
+                    339.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1229.0,
+                    327.0,
+                    1274.0,
+                    327.0,
+                    1274.0,
+                    354.0,
+                    1229.0,
+                    354.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    788.0,
+                    370.0,
+                    836.0,
+                    370.0,
+                    836.0,
+                    379.0,
+                    788.0,
+                    379.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1229.0,
+                    351.0,
+                    1380.0,
+                    351.0,
+                    1380.0,
+                    403.0,
+                    1229.0,
+                    403.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    412.0,
+                    335.0,
+                    412.0,
+                    335.0,
+                    430.0,
+                    296.0,
+                    430.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    512.0,
+                    384.0,
+                    850.0,
+                    384.0,
+                    850.0,
+                    468.0,
+                    512.0,
+                    468.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    944.0,
+                    398.0,
+                    977.0,
+                    398.0,
+                    977.0,
+                    416.0,
+                    944.0,
+                    416.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    302.0,
+                    511.0,
+                    334.0,
+                    511.0,
+                    334.0,
+                    530.0,
+                    302.0,
+                    530.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    364.0,
+                    465.0,
+                    856.0,
+                    465.0,
+                    856.0,
+                    574.0,
+                    364.0,
+                    574.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    919.0,
+                    458.0,
+                    941.0,
+                    458.0,
+                    941.0,
+                    499.0,
+                    919.0,
+                    499.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    943.0,
+                    506.0,
+                    978.0,
+                    506.0,
+                    978.0,
+                    527.0,
+                    943.0,
+                    527.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    811.0,
+                    563.0,
+                    838.0,
+                    563.0,
+                    838.0,
+                    572.0,
+                    811.0,
+                    572.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    303.0,
+                    595.0,
+                    327.0,
+                    595.0,
+                    327.0,
+                    609.0,
+                    303.0,
+                    609.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    493.0,
+                    571.0,
+                    857.0,
+                    571.0,
+                    857.0,
+                    666.0,
+                    493.0,
+                    666.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    608.0,
+                    341.0,
+                    608.0,
+                    341.0,
+                    625.0,
+                    292.0,
+                    625.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    303.0,
+                    623.0,
+                    327.0,
+                    623.0,
+                    327.0,
+                    636.0,
+                    303.0,
+                    636.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    937.0,
+                    616.0,
+                    978.0,
+                    616.0,
+                    978.0,
+                    637.0,
+                    937.0,
+                    637.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    776.0,
+                    658.0,
+                    838.0,
+                    658.0,
+                    838.0,
+                    670.0,
+                    776.0,
+                    670.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    987.0,
+                    663.0,
+                    1003.0,
+                    663.0,
+                    1003.0,
+                    679.0,
+                    987.0,
+                    679.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1023.0,
+                    661.0,
+                    1388.0,
+                    661.0,
+                    1388.0,
+                    681.0,
+                    1023.0,
+                    681.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1146.0,
+                    677.0,
+                    1218.0,
+                    677.0,
+                    1218.0,
+                    699.0,
+                    1146.0,
+                    699.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    606.0,
+                    707.0,
+                    638.0,
+                    707.0,
+                    638.0,
+                    731.0,
+                    606.0,
+                    731.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1145.0,
+                    702.0,
+                    1188.0,
+                    702.0,
+                    1188.0,
+                    736.0,
+                    1145.0,
+                    736.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    443.0,
+                    397.5,
+                    525.0,
+                    397.5,
+                    525.0,
+                    449.5,
+                    443.0,
+                    449.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    442.75,
+                    602.5,
+                    521.75,
+                    602.5,
+                    521.75,
+                    642.5,
+                    442.75,
+                    642.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    750.0,
+                    1407.0,
+                    750.0,
+                    1407.0,
+                    786.0,
+                    294.0,
+                    786.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    780.0,
+                    1406.0,
+                    780.0,
+                    1406.0,
+                    816.0,
+                    295.0,
+                    816.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    813.0,
+                    465.0,
+                    813.0,
+                    465.0,
+                    845.0,
+                    295.0,
+                    845.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    559.0,
+                    813.0,
+                    1403.0,
+                    813.0,
+                    1403.0,
+                    845.0,
+                    559.0,
+                    845.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    842.0,
+                    1251.0,
+                    842.0,
+                    1251.0,
+                    878.0,
+                    293.0,
+                    878.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    944.0,
+                    963.0,
+                    944.0,
+                    963.0,
+                    985.0,
+                    294.0,
+                    985.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    288.0,
+                    1520.0,
+                    538.0,
+                    1520.0,
+                    538.0,
+                    1574.0,
+                    288.0,
+                    1574.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    840.0,
+                    2060.0,
+                    861.0,
+                    2060.0,
+                    861.0,
+                    2091.0,
+                    840.0,
+                    2091.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1011.0,
+                    1407.0,
+                    1011.0,
+                    1407.0,
+                    1045.0,
+                    295.0,
+                    1045.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1040.0,
+                    495.0,
+                    1040.0,
+                    495.0,
+                    1076.0,
+                    292.0,
+                    1076.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    539.0,
+                    1040.0,
+                    639.0,
+                    1040.0,
+                    639.0,
+                    1076.0,
+                    539.0,
+                    1076.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    805.0,
+                    1040.0,
+                    1406.0,
+                    1040.0,
+                    1406.0,
+                    1076.0,
+                    805.0,
+                    1076.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1071.0,
+                    1406.0,
+                    1071.0,
+                    1406.0,
+                    1106.0,
+                    295.0,
+                    1106.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1102.0,
+                    1405.0,
+                    1102.0,
+                    1405.0,
+                    1137.0,
+                    295.0,
+                    1137.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1131.0,
+                    728.0,
+                    1131.0,
+                    728.0,
+                    1166.0,
+                    292.0,
+                    1166.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    754.0,
+                    1131.0,
+                    1098.0,
+                    1131.0,
+                    1098.0,
+                    1166.0,
+                    754.0,
+                    1166.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1131.0,
+                    1131.0,
+                    1405.0,
+                    1131.0,
+                    1405.0,
+                    1166.0,
+                    1131.0,
+                    1166.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1162.0,
+                    338.0,
+                    1162.0,
+                    338.0,
+                    1197.0,
+                    295.0,
+                    1197.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    368.0,
+                    1162.0,
+                    650.0,
+                    1162.0,
+                    650.0,
+                    1197.0,
+                    368.0,
+                    1197.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    674.0,
+                    1162.0,
+                    1405.0,
+                    1162.0,
+                    1405.0,
+                    1197.0,
+                    674.0,
+                    1197.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1192.0,
+                    531.0,
+                    1192.0,
+                    531.0,
+                    1229.0,
+                    294.0,
+                    1229.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    560.0,
+                    1192.0,
+                    1405.0,
+                    1192.0,
+                    1405.0,
+                    1229.0,
+                    560.0,
+                    1229.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    290.0,
+                    1218.0,
+                    968.0,
+                    1218.0,
+                    968.0,
+                    1269.0,
+                    290.0,
+                    1269.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1181.0,
+                    1218.0,
+                    1331.0,
+                    1218.0,
+                    1331.0,
+                    1269.0,
+                    1181.0,
+                    1269.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1346.0,
+                    1218.0,
+                    1409.0,
+                    1218.0,
+                    1409.0,
+                    1269.0,
+                    1346.0,
+                    1269.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1253.0,
+                    919.0,
+                    1253.0,
+                    919.0,
+                    1290.0,
+                    292.0,
+                    1290.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    948.0,
+                    1253.0,
+                    1406.0,
+                    1253.0,
+                    1406.0,
+                    1290.0,
+                    948.0,
+                    1290.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1285.0,
+                    985.0,
+                    1285.0,
+                    985.0,
+                    1320.0,
+                    295.0,
+                    1320.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1014.0,
+                    1285.0,
+                    1406.0,
+                    1285.0,
+                    1406.0,
+                    1320.0,
+                    1014.0,
+                    1320.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1315.0,
+                    495.0,
+                    1315.0,
+                    495.0,
+                    1349.0,
+                    295.0,
+                    1349.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    529.0,
+                    1315.0,
+                    749.0,
+                    1315.0,
+                    749.0,
+                    1349.0,
+                    529.0,
+                    1349.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    782.0,
+                    1315.0,
+                    811.0,
+                    1315.0,
+                    811.0,
+                    1349.0,
+                    782.0,
+                    1349.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    844.0,
+                    1315.0,
+                    1405.0,
+                    1315.0,
+                    1405.0,
+                    1349.0,
+                    844.0,
+                    1349.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1340.0,
+                    780.0,
+                    1340.0,
+                    780.0,
+                    1387.0,
+                    291.0,
+                    1387.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1189.0,
+                    1340.0,
+                    1407.0,
+                    1340.0,
+                    1407.0,
+                    1387.0,
+                    1189.0,
+                    1387.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1374.0,
+                    677.0,
+                    1374.0,
+                    677.0,
+                    1410.0,
+                    294.0,
+                    1410.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    703.0,
+                    1374.0,
+                    1407.0,
+                    1374.0,
+                    1407.0,
+                    1410.0,
+                    703.0,
+                    1410.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1404.0,
+                    707.0,
+                    1404.0,
+                    707.0,
+                    1443.0,
+                    292.0,
+                    1443.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    733.0,
+                    1404.0,
+                    1312.0,
+                    1404.0,
+                    1312.0,
+                    1443.0,
+                    733.0,
+                    1443.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1338.0,
+                    1404.0,
+                    1407.0,
+                    1404.0,
+                    1407.0,
+                    1443.0,
+                    1338.0,
+                    1443.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1438.0,
+                    685.0,
+                    1438.0,
+                    685.0,
+                    1469.0,
+                    295.0,
+                    1469.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1604.0,
+                    1406.0,
+                    1604.0,
+                    1406.0,
+                    1643.0,
+                    294.0,
+                    1643.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1636.0,
+                    795.0,
+                    1636.0,
+                    795.0,
+                    1671.0,
+                    295.0,
+                    1671.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    899.0,
+                    1636.0,
+                    1405.0,
+                    1636.0,
+                    1405.0,
+                    1671.0,
+                    899.0,
+                    1671.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1666.0,
+                    1407.0,
+                    1666.0,
+                    1407.0,
+                    1702.0,
+                    292.0,
+                    1702.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1698.0,
+                    1172.0,
+                    1698.0,
+                    1172.0,
+                    1729.0,
+                    293.0,
+                    1729.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1275.0,
+                    1698.0,
+                    1406.0,
+                    1698.0,
+                    1406.0,
+                    1729.0,
+                    1275.0,
+                    1729.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1728.0,
+                    1073.0,
+                    1728.0,
+                    1073.0,
+                    1762.0,
+                    295.0,
+                    1762.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1141.0,
+                    1728.0,
+                    1405.0,
+                    1728.0,
+                    1405.0,
+                    1762.0,
+                    1141.0,
+                    1762.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1760.0,
+                    1404.0,
+                    1760.0,
+                    1404.0,
+                    1794.0,
+                    295.0,
+                    1794.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1791.0,
+                    915.0,
+                    1791.0,
+                    915.0,
+                    1825.0,
+                    296.0,
+                    1825.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    1853.0,
+                    1406.0,
+                    1853.0,
+                    1406.0,
+                    1890.0,
+                    297.0,
+                    1890.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1883.0,
+                    991.0,
+                    1883.0,
+                    991.0,
+                    1918.0,
+                    294.0,
+                    1918.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1011.0,
+                    1883.0,
+                    1184.0,
+                    1883.0,
+                    1184.0,
+                    1918.0,
+                    1011.0,
+                    1918.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1204.0,
+                    1883.0,
+                    1406.0,
+                    1883.0,
+                    1406.0,
+                    1918.0,
+                    1204.0,
+                    1918.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1912.0,
+                    417.0,
+                    1912.0,
+                    417.0,
+                    1952.0,
+                    292.0,
+                    1952.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    443.0,
+                    1912.0,
+                    1409.0,
+                    1912.0,
+                    1409.0,
+                    1952.0,
+                    443.0,
+                    1952.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1942.0,
+                    498.0,
+                    1942.0,
+                    498.0,
+                    1983.0,
+                    293.0,
+                    1983.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    551.0,
+                    1942.0,
+                    1124.0,
+                    1942.0,
+                    1124.0,
+                    1983.0,
+                    551.0,
+                    1983.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1144.0,
+                    1942.0,
+                    1404.0,
+                    1942.0,
+                    1404.0,
+                    1983.0,
+                    1144.0,
+                    1983.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1975.0,
+                    499.0,
+                    1975.0,
+                    499.0,
+                    2013.0,
+                    294.0,
+                    2013.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 5,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    297,
+                    1155,
+                    1405,
+                    1155,
+                    1405,
+                    1639,
+                    297,
+                    1639
+                ],
+                "score": 0.983
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1733,
+                    1405,
+                    1733,
+                    1405,
+                    2007,
+                    298,
+                    2007
+                ],
+                "score": 0.981
+            },
+            {
+                "category_id": 5,
+                "poly": [
+                    301,
+                    934,
+                    1398,
+                    934,
+                    1398,
+                    1067,
+                    301,
+                    1067
+                ],
+                "score": 0.977,
+                "html": "<table><tr><td>Method</td><td>GI-x [9]</td><td>GI-z (ours)</td><td>GI-w (ours)</td><td>GI-z/w (GIAS,ours)</td><td>GI-x+BN[29]</td><td>GI-z/w+BN (ours)</td></tr><tr><td>PSNR↑</td><td>13.78</td><td>14.27</td><td>14.70</td><td>15.58</td><td>15.52</td><td>16.31</td></tr><tr><td>SSIM↑</td><td>0.2542</td><td>0.3106</td><td>0.3519</td><td>0.3895</td><td>0.3513</td><td>0.4311</td></tr><tr><td>LPIPS↓</td><td>0.4376</td><td>0.3233</td><td>0.5121</td><td>0.3023</td><td>0.3645</td><td>0.2861</td></tr></table>"
+            },
+            {
+                "category_id": 3,
+                "poly": [
+                    279,
+                    205,
+                    1388,
+                    205,
+                    1388,
+                    621,
+                    279,
+                    621
+                ],
+                "score": 0.974
+            },
+            {
+                "category_id": 4,
+                "poly": [
+                    296,
+                    640,
+                    1405,
+                    640,
+                    1405,
+                    792,
+                    296,
+                    792
+                ],
+                "score": 0.95
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    299,
+                    1677,
+                    1059,
+                    1677,
+                    1059,
+                    1710,
+                    299,
+                    1710
+                ],
+                "score": 0.924
+            },
+            {
+                "category_id": 6,
+                "poly": [
+                    296,
+                    824,
+                    1405,
+                    824,
+                    1405,
+                    918,
+                    296,
+                    918
+                ],
+                "score": 0.92
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    299,
+                    1098,
+                    695,
+                    1098,
+                    695,
+                    1132,
+                    299,
+                    1132
+                ],
+                "score": 0.914
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    841,
+                    2061,
+                    858,
+                    2061,
+                    858,
+                    2084,
+                    841,
+                    2084
+                ],
+                "score": 0.712
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    841,
+                    2061,
+                    859,
+                    2061,
+                    859,
+                    2084,
+                    841,
+                    2084
+                ],
+                "score": 0.107
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    389,
+                    1886,
+                    475,
+                    1886,
+                    475,
+                    1913,
+                    389,
+                    1913
+                ],
+                "score": 0.87,
+                "latex": "x + { \\tt B N }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    957,
+                    1216,
+                    1007,
+                    1216,
+                    1007,
+                    1248,
+                    957,
+                    1248
+                ],
+                "score": 0.86,
+                "latex": "z / w"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    774,
+                    1460,
+                    865,
+                    1460,
+                    865,
+                    1492,
+                    774,
+                    1492
+                ],
+                "score": 0.84,
+                "latex": "{ \\mathrm { G I } } { - z } / w"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    877,
+                    1854,
+                    928,
+                    1854,
+                    928,
+                    1888,
+                    877,
+                    1888
+                ],
+                "score": 0.84,
+                "latex": "z / w"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1174,
+                    1794,
+                    1261,
+                    1794,
+                    1261,
+                    1823,
+                    1174,
+                    1823
+                ],
+                "score": 0.81,
+                "latex": "x + { \\tt B N }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    921,
+                    1398,
+                    1007,
+                    1398,
+                    1007,
+                    1430,
+                    921,
+                    1430
+                ],
+                "score": 0.8,
+                "latex": "{ \\mathrm { G I } } { - } z / x"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    435,
+                    1186,
+                    529,
+                    1186,
+                    529,
+                    1218,
+                    435,
+                    1218
+                ],
+                "score": 0.79,
+                "latex": "\\mathbf { G I } { - } z / w ,"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    468,
+                    1458,
+                    554,
+                    1458,
+                    554,
+                    1492,
+                    468,
+                    1492
+                ],
+                "score": 0.76,
+                "latex": "{ \\mathrm { G I } } { - } z / { \\bar { x } }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    904,
+                    1253,
+                    923,
+                    1253,
+                    923,
+                    1274,
+                    904,
+                    1274
+                ],
+                "score": 0.76,
+                "latex": "z"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1292,
+                    1793,
+                    1407,
+                    1793,
+                    1407,
+                    1826,
+                    1292,
+                    1826
+                ],
+                "score": 0.76,
+                "latex": "{ \\mathrm { G I } } { - } z / w +"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    944,
+                    1797,
+                    962,
+                    1797,
+                    962,
+                    1821,
+                    944,
+                    1821
+                ],
+                "score": 0.75,
+                "latex": "x"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    924,
+                    1459,
+                    1010,
+                    1459,
+                    1010,
+                    1492,
+                    924,
+                    1492
+                ],
+                "score": 0.73,
+                "latex": "{ \\mathrm { G I } } { - } z / x"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    751,
+                    1282,
+                    770,
+                    1282,
+                    770,
+                    1304,
+                    751,
+                    1304
+                ],
+                "score": 0.73,
+                "latex": "z"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    806,
+                    1429,
+                    856,
+                    1429,
+                    856,
+                    1459,
+                    806,
+                    1459
+                ],
+                "score": 0.71,
+                "latex": "z / w"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    539,
+                    1311,
+                    557,
+                    1311,
+                    557,
+                    1334,
+                    539,
+                    1334
+                ],
+                "score": 0.71,
+                "latex": "x"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1139,
+                    645,
+                    1158,
+                    645,
+                    1158,
+                    669,
+                    1139,
+                    669
+                ],
+                "score": 0.7,
+                "latex": "x"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    465,
+                    1370,
+                    483,
+                    1370,
+                    483,
+                    1395,
+                    465,
+                    1395
+                ],
+                "score": 0.7,
+                "latex": "z"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    747,
+                    1397,
+                    798,
+                    1397,
+                    798,
+                    1430,
+                    747,
+                    1430
+                ],
+                "score": 0.67,
+                "latex": "z / w"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    878,
+                    1342,
+                    895,
+                    1342,
+                    895,
+                    1365,
+                    878,
+                    1365
+                ],
+                "score": 0.67,
+                "latex": "z"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    994,
+                    1793,
+                    1084,
+                    1793,
+                    1084,
+                    1826,
+                    994,
+                    1826
+                ],
+                "score": 0.67,
+                "latex": "{ \\mathrm { G I } } { - } z / w"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1269,
+                    1219,
+                    1288,
+                    1219,
+                    1288,
+                    1244,
+                    1269,
+                    1244
+                ],
+                "score": 0.66,
+                "latex": "x"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1059,
+                    1216,
+                    1104,
+                    1216,
+                    1104,
+                    1248,
+                    1059,
+                    1248
+                ],
+                "score": 0.65,
+                "latex": "z / x"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1260,
+                    640,
+                    1311,
+                    640,
+                    1311,
+                    674,
+                    1260,
+                    674
+                ],
+                "score": 0.62,
+                "latex": "z / w"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1114,
+                    1216,
+                    1171,
+                    1216,
+                    1171,
+                    1246,
+                    1114,
+                    1246
+                ],
+                "score": 0.55,
+                "latex": "\\mathrm { G I } { - } z"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1155,
+                    1220,
+                    1171,
+                    1220,
+                    1171,
+                    1244,
+                    1155,
+                    1244
+                ],
+                "score": 0.47,
+                "latex": "z"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1230,
+                    1216,
+                    1288,
+                    1216,
+                    1288,
+                    1245,
+                    1230,
+                    1245
+                ],
+                "score": 0.36,
+                "latex": "_ { \\mathrm { G I - } x }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    840,
+                    1338,
+                    895,
+                    1338,
+                    895,
+                    1367,
+                    840,
+                    1367
+                ],
+                "score": 0.27,
+                "latex": "\\mathrm { G I } { - } z"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    347.0,
+                    212.0,
+                    473.0,
+                    212.0,
+                    473.0,
+                    239.0,
+                    347.0,
+                    239.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    233.0,
+                    312.0,
+                    233.0,
+                    312.0,
+                    248.0,
+                    293.0,
+                    248.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    350.0,
+                    232.0,
+                    481.0,
+                    232.0,
+                    481.0,
+                    255.0,
+                    350.0,
+                    255.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    347.0,
+                    249.0,
+                    389.0,
+                    249.0,
+                    389.0,
+                    275.0,
+                    347.0,
+                    275.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    331.0,
+                    255.0,
+                    351.0,
+                    255.0,
+                    351.0,
+                    268.0,
+                    331.0,
+                    268.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    330.0,
+                    268.0,
+                    470.0,
+                    268.0,
+                    470.0,
+                    291.0,
+                    330.0,
+                    291.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    349.0,
+                    284.0,
+                    472.0,
+                    284.0,
+                    472.0,
+                    311.0,
+                    349.0,
+                    311.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    317.0,
+                    290.0,
+                    352.0,
+                    290.0,
+                    352.0,
+                    308.0,
+                    317.0,
+                    308.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    298.0,
+                    312.0,
+                    298.0,
+                    312.0,
+                    311.0,
+                    294.0,
+                    311.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    350.0,
+                    302.0,
+                    468.0,
+                    302.0,
+                    468.0,
+                    330.0,
+                    350.0,
+                    330.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    321.0,
+                    310.0,
+                    349.0,
+                    310.0,
+                    349.0,
+                    322.0,
+                    321.0,
+                    322.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    278.0,
+                    352.0,
+                    300.0,
+                    352.0,
+                    300.0,
+                    418.0,
+                    278.0,
+                    418.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    361.0,
+                    311.0,
+                    361.0,
+                    311.0,
+                    377.0,
+                    294.0,
+                    377.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    428.0,
+                    308.0,
+                    428.0,
+                    308.0,
+                    438.0,
+                    297.0,
+                    438.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    491.0,
+                    310.0,
+                    491.0,
+                    310.0,
+                    502.0,
+                    297.0,
+                    502.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    739.0,
+                    542.0,
+                    792.0,
+                    542.0,
+                    792.0,
+                    563.0,
+                    739.0,
+                    563.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    839.0,
+                    541.0,
+                    890.0,
+                    541.0,
+                    890.0,
+                    562.0,
+                    839.0,
+                    562.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    924.0,
+                    541.0,
+                    995.0,
+                    541.0,
+                    995.0,
+                    562.0,
+                    924.0,
+                    562.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1032.0,
+                    541.0,
+                    1084.0,
+                    541.0,
+                    1084.0,
+                    562.0,
+                    1032.0,
+                    562.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1133.0,
+                    541.0,
+                    1174.0,
+                    541.0,
+                    1174.0,
+                    562.0,
+                    1133.0,
+                    562.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1231.0,
+                    543.0,
+                    1267.0,
+                    543.0,
+                    1267.0,
+                    561.0,
+                    1231.0,
+                    561.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1325.0,
+                    543.0,
+                    1360.0,
+                    543.0,
+                    1360.0,
+                    561.0,
+                    1325.0,
+                    561.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    556.0,
+                    310.0,
+                    556.0,
+                    310.0,
+                    566.0,
+                    297.0,
+                    566.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    376.0,
+                    559.0,
+                    429.0,
+                    559.0,
+                    429.0,
+                    580.0,
+                    376.0,
+                    580.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    589.0,
+                    562.0,
+                    617.0,
+                    562.0,
+                    617.0,
+                    575.0,
+                    589.0,
+                    575.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    745.0,
+                    556.0,
+                    788.0,
+                    556.0,
+                    788.0,
+                    577.0,
+                    745.0,
+                    577.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    842.0,
+                    554.0,
+                    885.0,
+                    554.0,
+                    885.0,
+                    575.0,
+                    842.0,
+                    575.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    925.0,
+                    556.0,
+                    997.0,
+                    556.0,
+                    997.0,
+                    577.0,
+                    925.0,
+                    577.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1032.0,
+                    556.0,
+                    1085.0,
+                    556.0,
+                    1085.0,
+                    577.0,
+                    1032.0,
+                    577.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1109.0,
+                    556.0,
+                    1197.0,
+                    556.0,
+                    1197.0,
+                    579.0,
+                    1109.0,
+                    579.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1294.0,
+                    554.0,
+                    1391.0,
+                    554.0,
+                    1391.0,
+                    579.0,
+                    1294.0,
+                    579.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    461.0,
+                    593.0,
+                    502.0,
+                    593.0,
+                    502.0,
+                    626.0,
+                    461.0,
+                    626.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1016.0,
+                    592.0,
+                    1059.0,
+                    592.0,
+                    1059.0,
+                    626.0,
+                    1016.0,
+                    626.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    637.0,
+                    1138.0,
+                    637.0,
+                    1138.0,
+                    675.0,
+                    293.0,
+                    675.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1159.0,
+                    637.0,
+                    1259.0,
+                    637.0,
+                    1259.0,
+                    675.0,
+                    1159.0,
+                    675.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1312.0,
+                    637.0,
+                    1407.0,
+                    637.0,
+                    1407.0,
+                    675.0,
+                    1312.0,
+                    675.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    670.0,
+                    1408.0,
+                    670.0,
+                    1408.0,
+                    707.0,
+                    294.0,
+                    707.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    698.0,
+                    1406.0,
+                    698.0,
+                    1406.0,
+                    737.0,
+                    291.0,
+                    737.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    730.0,
+                    1405.0,
+                    730.0,
+                    1405.0,
+                    767.0,
+                    294.0,
+                    767.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    762.0,
+                    586.0,
+                    762.0,
+                    586.0,
+                    795.0,
+                    292.0,
+                    795.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1674.0,
+                    1062.0,
+                    1674.0,
+                    1062.0,
+                    1715.0,
+                    293.0,
+                    1715.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    821.0,
+                    1405.0,
+                    821.0,
+                    1405.0,
+                    860.0,
+                    291.0,
+                    860.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    853.0,
+                    1405.0,
+                    853.0,
+                    1405.0,
+                    890.0,
+                    293.0,
+                    890.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    886.0,
+                    1141.0,
+                    886.0,
+                    1141.0,
+                    920.0,
+                    295.0,
+                    920.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1093.0,
+                    698.0,
+                    1093.0,
+                    698.0,
+                    1139.0,
+                    293.0,
+                    1139.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    839.0,
+                    2059.0,
+                    860.0,
+                    2059.0,
+                    860.0,
+                    2091.0,
+                    839.0,
+                    2091.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    840.0,
+                    2059.0,
+                    860.0,
+                    2059.0,
+                    860.0,
+                    2092.0,
+                    840.0,
+                    2092.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1157.0,
+                    1407.0,
+                    1157.0,
+                    1407.0,
+                    1190.0,
+                    296.0,
+                    1190.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1185.0,
+                    434.0,
+                    1185.0,
+                    434.0,
+                    1221.0,
+                    292.0,
+                    1221.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    530.0,
+                    1185.0,
+                    1407.0,
+                    1185.0,
+                    1407.0,
+                    1221.0,
+                    530.0,
+                    1221.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1216.0,
+                    956.0,
+                    1216.0,
+                    956.0,
+                    1250.0,
+                    294.0,
+                    1250.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1008.0,
+                    1216.0,
+                    1058.0,
+                    1216.0,
+                    1058.0,
+                    1250.0,
+                    1008.0,
+                    1250.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1105.0,
+                    1216.0,
+                    1113.0,
+                    1216.0,
+                    1113.0,
+                    1250.0,
+                    1105.0,
+                    1250.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1172.0,
+                    1216.0,
+                    1229.0,
+                    1216.0,
+                    1229.0,
+                    1250.0,
+                    1172.0,
+                    1250.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1289.0,
+                    1216.0,
+                    1405.0,
+                    1216.0,
+                    1405.0,
+                    1250.0,
+                    1289.0,
+                    1250.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1246.0,
+                    903.0,
+                    1246.0,
+                    903.0,
+                    1282.0,
+                    294.0,
+                    1282.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    924.0,
+                    1246.0,
+                    1406.0,
+                    1246.0,
+                    1406.0,
+                    1282.0,
+                    924.0,
+                    1282.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1278.0,
+                    750.0,
+                    1278.0,
+                    750.0,
+                    1311.0,
+                    296.0,
+                    1311.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    771.0,
+                    1278.0,
+                    1406.0,
+                    1278.0,
+                    1406.0,
+                    1311.0,
+                    771.0,
+                    1311.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1304.0,
+                    538.0,
+                    1304.0,
+                    538.0,
+                    1342.0,
+                    292.0,
+                    1342.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    558.0,
+                    1304.0,
+                    1406.0,
+                    1304.0,
+                    1406.0,
+                    1342.0,
+                    558.0,
+                    1342.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1337.0,
+                    839.0,
+                    1337.0,
+                    839.0,
+                    1372.0,
+                    292.0,
+                    1372.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    896.0,
+                    1337.0,
+                    1407.0,
+                    1337.0,
+                    1407.0,
+                    1372.0,
+                    896.0,
+                    1372.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1364.0,
+                    464.0,
+                    1364.0,
+                    464.0,
+                    1402.0,
+                    294.0,
+                    1402.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    484.0,
+                    1364.0,
+                    1407.0,
+                    1364.0,
+                    1407.0,
+                    1402.0,
+                    484.0,
+                    1402.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1398.0,
+                    746.0,
+                    1398.0,
+                    746.0,
+                    1431.0,
+                    296.0,
+                    1431.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    799.0,
+                    1398.0,
+                    920.0,
+                    1398.0,
+                    920.0,
+                    1431.0,
+                    799.0,
+                    1431.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1008.0,
+                    1398.0,
+                    1405.0,
+                    1398.0,
+                    1405.0,
+                    1431.0,
+                    1008.0,
+                    1431.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1426.0,
+                    805.0,
+                    1426.0,
+                    805.0,
+                    1463.0,
+                    294.0,
+                    1463.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    857.0,
+                    1426.0,
+                    1406.0,
+                    1426.0,
+                    1406.0,
+                    1463.0,
+                    857.0,
+                    1463.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1458.0,
+                    467.0,
+                    1458.0,
+                    467.0,
+                    1495.0,
+                    295.0,
+                    1495.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    555.0,
+                    1458.0,
+                    773.0,
+                    1458.0,
+                    773.0,
+                    1495.0,
+                    555.0,
+                    1495.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    866.0,
+                    1458.0,
+                    923.0,
+                    1458.0,
+                    923.0,
+                    1495.0,
+                    866.0,
+                    1495.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1011.0,
+                    1458.0,
+                    1407.0,
+                    1458.0,
+                    1407.0,
+                    1495.0,
+                    1011.0,
+                    1495.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1487.0,
+                    1408.0,
+                    1487.0,
+                    1408.0,
+                    1527.0,
+                    292.0,
+                    1527.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1518.0,
+                    1406.0,
+                    1518.0,
+                    1406.0,
+                    1555.0,
+                    295.0,
+                    1555.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1551.0,
+                    1403.0,
+                    1551.0,
+                    1403.0,
+                    1583.0,
+                    296.0,
+                    1583.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1577.0,
+                    1405.0,
+                    1577.0,
+                    1405.0,
+                    1616.0,
+                    294.0,
+                    1616.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1606.0,
+                    401.0,
+                    1606.0,
+                    401.0,
+                    1649.0,
+                    291.0,
+                    1649.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1731.0,
+                    1405.0,
+                    1731.0,
+                    1405.0,
+                    1769.0,
+                    295.0,
+                    1769.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1762.0,
+                    1405.0,
+                    1762.0,
+                    1405.0,
+                    1799.0,
+                    296.0,
+                    1799.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1792.0,
+                    943.0,
+                    1792.0,
+                    943.0,
+                    1830.0,
+                    292.0,
+                    1830.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    963.0,
+                    1792.0,
+                    993.0,
+                    1792.0,
+                    993.0,
+                    1830.0,
+                    963.0,
+                    1830.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1085.0,
+                    1792.0,
+                    1173.0,
+                    1792.0,
+                    1173.0,
+                    1830.0,
+                    1085.0,
+                    1830.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1262.0,
+                    1792.0,
+                    1291.0,
+                    1792.0,
+                    1291.0,
+                    1830.0,
+                    1262.0,
+                    1830.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1819.0,
+                    1409.0,
+                    1819.0,
+                    1409.0,
+                    1862.0,
+                    292.0,
+                    1862.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1852.0,
+                    876.0,
+                    1852.0,
+                    876.0,
+                    1892.0,
+                    293.0,
+                    1892.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    929.0,
+                    1852.0,
+                    1407.0,
+                    1852.0,
+                    1407.0,
+                    1892.0,
+                    929.0,
+                    1892.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1882.0,
+                    388.0,
+                    1882.0,
+                    388.0,
+                    1921.0,
+                    293.0,
+                    1921.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    476.0,
+                    1882.0,
+                    1405.0,
+                    1882.0,
+                    1405.0,
+                    1921.0,
+                    476.0,
+                    1921.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1914.0,
+                    1405.0,
+                    1914.0,
+                    1405.0,
+                    1951.0,
+                    295.0,
+                    1951.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1944.0,
+                    1405.0,
+                    1944.0,
+                    1405.0,
+                    1979.0,
+                    295.0,
+                    1979.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1975.0,
+                    1408.0,
+                    1975.0,
+                    1408.0,
+                    2012.0,
+                    295.0,
+                    2012.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 6,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    954,
+                    1405,
+                    954,
+                    1405,
+                    1470,
+                    298,
+                    1470
+                ],
+                "score": 0.984
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1568,
+                    1403,
+                    1568,
+                    1403,
+                    1721,
+                    298,
+                    1721
+                ],
+                "score": 0.977
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1735,
+                    1405,
+                    1735,
+                    1405,
+                    1918,
+                    298,
+                    1918
+                ],
+                "score": 0.975
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    763,
+                    1403,
+                    763,
+                    1403,
+                    916,
+                    298,
+                    916
+                ],
+                "score": 0.974
+            },
+            {
+                "category_id": 3,
+                "poly": [
+                    297,
+                    202,
+                    1343,
+                    202,
+                    1343,
+                    587,
+                    297,
+                    587
+                ],
+                "score": 0.972
+            },
+            {
+                "category_id": 4,
+                "poly": [
+                    295,
+                    606,
+                    1404,
+                    606,
+                    1404,
+                    699,
+                    295,
+                    699
+                ],
+                "score": 0.958
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    297,
+                    1949,
+                    1403,
+                    1949,
+                    1403,
+                    2007,
+                    297,
+                    2007
+                ],
+                "score": 0.933
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    298,
+                    1511,
+                    864,
+                    1511,
+                    864,
+                    1543,
+                    298,
+                    1543
+                ],
+                "score": 0.926
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    840,
+                    2062,
+                    858,
+                    2062,
+                    858,
+                    2085,
+                    840,
+                    2085
+                ],
+                "score": 0.813
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    986,
+                    1827,
+                    1087,
+                    1827,
+                    1087,
+                    1860,
+                    986,
+                    1860
+                ],
+                "score": 0.93,
+                "latex": "G _ { w ^ { ( t ) } } ( z )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    563,
+                    1227,
+                    614,
+                    1227,
+                    614,
+                    1259,
+                    563,
+                    1259
+                ],
+                "score": 0.87,
+                "latex": "z / w"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1153,
+                    1137,
+                    1208,
+                    1137,
+                    1208,
+                    1165,
+                    1153,
+                    1165
+                ],
+                "score": 0.86,
+                "latex": "9 9 \\%"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    597,
+                    1950,
+                    666,
+                    1950,
+                    666,
+                    1977,
+                    597,
+                    1977
+                ],
+                "score": 0.85,
+                "latex": "0 . 9 9 \\%"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    297,
+                    1977,
+                    346,
+                    1977,
+                    346,
+                    2005,
+                    297,
+                    2005
+                ],
+                "score": 0.8,
+                "latex": "9 9 \\%"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1264,
+                    638,
+                    1315,
+                    638,
+                    1315,
+                    670,
+                    1264,
+                    670
+                ],
+                "score": 0.78,
+                "latex": "z / w"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1055,
+                    1951,
+                    1090,
+                    1951,
+                    1090,
+                    1975,
+                    1055,
+                    1975
+                ],
+                "score": 0.77,
+                "latex": "1 \\%"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1210,
+                    828,
+                    1233,
+                    828,
+                    1233,
+                    852,
+                    1210,
+                    852
+                ],
+                "score": 0.77,
+                "latex": "w"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    448,
+                    672,
+                    466,
+                    672,
+                    466,
+                    695,
+                    448,
+                    695
+                ],
+                "score": 0.75,
+                "latex": "x"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1232,
+                    1832,
+                    1250,
+                    1832,
+                    1250,
+                    1854,
+                    1232,
+                    1854
+                ],
+                "score": 0.74,
+                "latex": "z"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    752,
+                    1135,
+                    843,
+                    1135,
+                    843,
+                    1169,
+                    752,
+                    1169
+                ],
+                "score": 0.73,
+                "latex": "{ \\mathrm { G I } } { - z } / w"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    890,
+                    606,
+                    980,
+                    606,
+                    980,
+                    640,
+                    890,
+                    640
+                ],
+                "score": 0.73,
+                "latex": "G I - z / w"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    337,
+                    1891,
+                    354,
+                    1891,
+                    354,
+                    1914,
+                    337,
+                    1914
+                ],
+                "score": 0.73,
+                "latex": "z"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1160,
+                    1106,
+                    1251,
+                    1106,
+                    1251,
+                    1137,
+                    1160,
+                    1137
+                ],
+                "score": 0.72,
+                "latex": "{ \\mathrm { G I } } { - z } / w"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1382,
+                    1230,
+                    1401,
+                    1230,
+                    1401,
+                    1254,
+                    1382,
+                    1254
+                ],
+                "score": 0.71,
+                "latex": "_ x"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    802,
+                    859,
+                    820,
+                    859,
+                    820,
+                    882,
+                    802,
+                    882
+                ],
+                "score": 0.71,
+                "latex": "x"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    601,
+                    1109,
+                    619,
+                    1109,
+                    619,
+                    1133,
+                    601,
+                    1133
+                ],
+                "score": 0.7,
+                "latex": "x"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    448,
+                    1769,
+                    465,
+                    1769,
+                    465,
+                    1794,
+                    448,
+                    1794
+                ],
+                "score": 0.68,
+                "latex": "z"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    297,
+                    860,
+                    321,
+                    860,
+                    321,
+                    882,
+                    297,
+                    882
+                ],
+                "score": 0.64,
+                "latex": "w"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    510,
+                    1829,
+                    525,
+                    1829,
+                    525,
+                    1854,
+                    510,
+                    1854
+                ],
+                "score": 0.61,
+                "latex": "t"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    514,
+                    1765,
+                    604,
+                    1765,
+                    604,
+                    1798,
+                    514,
+                    1798
+                ],
+                "score": 0.6,
+                "latex": "{ \\mathrm { G I } } { - z } / w"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1077,
+                    1599,
+                    1152,
+                    1599,
+                    1152,
+                    1628,
+                    1077,
+                    1628
+                ],
+                "score": 0.39,
+                "latex": "3 2 \\mathbf { x } 3 2"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    409,
+                    1766,
+                    465,
+                    1766,
+                    465,
+                    1795,
+                    409,
+                    1795
+                ],
+                "score": 0.35,
+                "latex": "\\mathrm { G I } { - } z"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    315.0,
+                    199.0,
+                    337.0,
+                    199.0,
+                    337.0,
+                    217.0,
+                    315.0,
+                    217.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    341.0,
+                    218.0,
+                    356.0,
+                    218.0,
+                    356.0,
+                    237.0,
+                    341.0,
+                    237.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    486.0,
+                    213.0,
+                    662.0,
+                    213.0,
+                    662.0,
+                    277.0,
+                    486.0,
+                    277.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    827.0,
+                    213.0,
+                    998.0,
+                    213.0,
+                    998.0,
+                    258.0,
+                    827.0,
+                    258.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1153.0,
+                    213.0,
+                    1330.0,
+                    213.0,
+                    1330.0,
+                    283.0,
+                    1153.0,
+                    283.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    316.0,
+                    259.0,
+                    338.0,
+                    259.0,
+                    338.0,
+                    279.0,
+                    316.0,
+                    279.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    523.0,
+                    252.0,
+                    654.0,
+                    252.0,
+                    654.0,
+                    278.0,
+                    523.0,
+                    278.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1189.0,
+                    249.0,
+                    1336.0,
+                    249.0,
+                    1336.0,
+                    278.0,
+                    1189.0,
+                    278.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    341.0,
+                    269.0,
+                    356.0,
+                    269.0,
+                    356.0,
+                    286.0,
+                    341.0,
+                    286.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1157.0,
+                    267.0,
+                    1323.0,
+                    267.0,
+                    1323.0,
+                    297.0,
+                    1157.0,
+                    297.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    337.0,
+                    287.0,
+                    359.0,
+                    287.0,
+                    359.0,
+                    319.0,
+                    337.0,
+                    319.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    298.0,
+                    325.0,
+                    322.0,
+                    325.0,
+                    322.0,
+                    397.0,
+                    298.0,
+                    397.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    444.0,
+                    326.0,
+                    455.0,
+                    326.0,
+                    455.0,
+                    336.0,
+                    444.0,
+                    336.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1017.0,
+                    321.0,
+                    1032.0,
+                    321.0,
+                    1032.0,
+                    337.0,
+                    1017.0,
+                    337.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1095.0,
+                    328.0,
+                    1104.0,
+                    328.0,
+                    1104.0,
+                    340.0,
+                    1095.0,
+                    340.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    444.0,
+                    342.0,
+                    456.0,
+                    342.0,
+                    456.0,
+                    353.0,
+                    444.0,
+                    353.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1170.0,
+                    341.0,
+                    1181.0,
+                    341.0,
+                    1181.0,
+                    349.0,
+                    1170.0,
+                    349.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    834.0,
+                    371.0,
+                    840.0,
+                    371.0,
+                    840.0,
+                    377.0,
+                    834.0,
+                    377.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    317.0,
+                    383.0,
+                    334.0,
+                    383.0,
+                    334.0,
+                    398.0,
+                    317.0,
+                    398.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    907.0,
+                    382.0,
+                    918.0,
+                    382.0,
+                    918.0,
+                    392.0,
+                    907.0,
+                    392.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    832.0,
+                    418.0,
+                    841.0,
+                    418.0,
+                    841.0,
+                    426.0,
+                    832.0,
+                    426.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    316.0,
+                    443.0,
+                    335.0,
+                    443.0,
+                    335.0,
+                    462.0,
+                    316.0,
+                    462.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    985.0,
+                    453.0,
+                    994.0,
+                    453.0,
+                    994.0,
+                    462.0,
+                    985.0,
+                    462.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1019.0,
+                    464.0,
+                    1032.0,
+                    464.0,
+                    1032.0,
+                    474.0,
+                    1019.0,
+                    474.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1095.0,
+                    464.0,
+                    1104.0,
+                    464.0,
+                    1104.0,
+                    473.0,
+                    1095.0,
+                    473.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1171.0,
+                    467.0,
+                    1182.0,
+                    467.0,
+                    1182.0,
+                    476.0,
+                    1171.0,
+                    476.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    316.0,
+                    504.0,
+                    337.0,
+                    504.0,
+                    337.0,
+                    523.0,
+                    316.0,
+                    523.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    345.0,
+                    516.0,
+                    353.0,
+                    516.0,
+                    353.0,
+                    525.0,
+                    345.0,
+                    525.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    391.0,
+                    513.0,
+                    406.0,
+                    513.0,
+                    406.0,
+                    527.0,
+                    391.0,
+                    527.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    444.0,
+                    515.0,
+                    455.0,
+                    515.0,
+                    455.0,
+                    526.0,
+                    444.0,
+                    526.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    463.0,
+                    512.0,
+                    538.0,
+                    512.0,
+                    538.0,
+                    547.0,
+                    463.0,
+                    547.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    541.0,
+                    511.0,
+                    562.0,
+                    511.0,
+                    562.0,
+                    530.0,
+                    541.0,
+                    530.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    591.0,
+                    511.0,
+                    613.0,
+                    511.0,
+                    613.0,
+                    530.0,
+                    591.0,
+                    530.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    641.0,
+                    511.0,
+                    664.0,
+                    511.0,
+                    664.0,
+                    529.0,
+                    641.0,
+                    529.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    673.0,
+                    510.0,
+                    702.0,
+                    510.0,
+                    702.0,
+                    531.0,
+                    673.0,
+                    531.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    751.0,
+                    511.0,
+                    776.0,
+                    511.0,
+                    776.0,
+                    530.0,
+                    751.0,
+                    530.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    784.0,
+                    511.0,
+                    893.0,
+                    511.0,
+                    893.0,
+                    550.0,
+                    784.0,
+                    550.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    898.0,
+                    510.0,
+                    930.0,
+                    510.0,
+                    930.0,
+                    531.0,
+                    898.0,
+                    531.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    973.0,
+                    510.0,
+                    1006.0,
+                    510.0,
+                    1006.0,
+                    530.0,
+                    973.0,
+                    530.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1009.0,
+                    510.0,
+                    1038.0,
+                    510.0,
+                    1038.0,
+                    531.0,
+                    1009.0,
+                    531.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1078.0,
+                    511.0,
+                    1121.0,
+                    511.0,
+                    1121.0,
+                    531.0,
+                    1078.0,
+                    531.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1135.0,
+                    510.0,
+                    1214.0,
+                    510.0,
+                    1214.0,
+                    546.0,
+                    1135.0,
+                    546.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1234.0,
+                    510.0,
+                    1265.0,
+                    510.0,
+                    1265.0,
+                    531.0,
+                    1234.0,
+                    531.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1314.0,
+                    511.0,
+                    1336.0,
+                    511.0,
+                    1336.0,
+                    530.0,
+                    1314.0,
+                    530.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    457.0,
+                    555.0,
+                    501.0,
+                    555.0,
+                    501.0,
+                    594.0,
+                    457.0,
+                    594.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    807.0,
+                    555.0,
+                    853.0,
+                    555.0,
+                    853.0,
+                    593.0,
+                    807.0,
+                    593.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1145.0,
+                    557.0,
+                    1189.0,
+                    557.0,
+                    1189.0,
+                    594.0,
+                    1145.0,
+                    594.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    310.0,
+                    321.0,
+                    340.0,
+                    321.0,
+                    340.0,
+                    339.5,
+                    310.0,
+                    339.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    604.0,
+                    889.0,
+                    604.0,
+                    889.0,
+                    640.0,
+                    294.0,
+                    640.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    981.0,
+                    604.0,
+                    1405.0,
+                    604.0,
+                    1405.0,
+                    640.0,
+                    981.0,
+                    640.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    637.0,
+                    1263.0,
+                    637.0,
+                    1263.0,
+                    674.0,
+                    295.0,
+                    674.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1316.0,
+                    637.0,
+                    1405.0,
+                    637.0,
+                    1405.0,
+                    674.0,
+                    1316.0,
+                    674.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    667.0,
+                    447.0,
+                    667.0,
+                    447.0,
+                    704.0,
+                    295.0,
+                    704.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    467.0,
+                    667.0,
+                    1294.0,
+                    667.0,
+                    1294.0,
+                    704.0,
+                    467.0,
+                    704.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    331.0,
+                    1943.0,
+                    596.0,
+                    1943.0,
+                    596.0,
+                    1985.0,
+                    331.0,
+                    1985.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    667.0,
+                    1943.0,
+                    1054.0,
+                    1943.0,
+                    1054.0,
+                    1985.0,
+                    667.0,
+                    1985.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1091.0,
+                    1943.0,
+                    1405.0,
+                    1943.0,
+                    1405.0,
+                    1985.0,
+                    1091.0,
+                    1985.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    347.0,
+                    1973.0,
+                    878.0,
+                    1973.0,
+                    878.0,
+                    2011.0,
+                    347.0,
+                    2011.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1506.0,
+                    867.0,
+                    1506.0,
+                    867.0,
+                    1550.0,
+                    293.0,
+                    1550.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    838.0,
+                    2058.0,
+                    862.0,
+                    2058.0,
+                    862.0,
+                    2090.0,
+                    838.0,
+                    2090.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    951.0,
+                    1405.0,
+                    951.0,
+                    1405.0,
+                    989.0,
+                    293.0,
+                    989.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    986.0,
+                    1403.0,
+                    986.0,
+                    1403.0,
+                    1018.0,
+                    296.0,
+                    1018.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1013.0,
+                    1406.0,
+                    1013.0,
+                    1406.0,
+                    1050.0,
+                    292.0,
+                    1050.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1044.0,
+                    1405.0,
+                    1044.0,
+                    1405.0,
+                    1080.0,
+                    295.0,
+                    1080.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1073.0,
+                    1407.0,
+                    1073.0,
+                    1407.0,
+                    1111.0,
+                    293.0,
+                    1111.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1104.0,
+                    600.0,
+                    1104.0,
+                    600.0,
+                    1140.0,
+                    295.0,
+                    1140.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    620.0,
+                    1104.0,
+                    1159.0,
+                    1104.0,
+                    1159.0,
+                    1140.0,
+                    620.0,
+                    1140.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1252.0,
+                    1104.0,
+                    1406.0,
+                    1104.0,
+                    1406.0,
+                    1140.0,
+                    1252.0,
+                    1140.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1135.0,
+                    751.0,
+                    1135.0,
+                    751.0,
+                    1171.0,
+                    295.0,
+                    1171.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    844.0,
+                    1135.0,
+                    1152.0,
+                    1135.0,
+                    1152.0,
+                    1171.0,
+                    844.0,
+                    1171.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1209.0,
+                    1135.0,
+                    1405.0,
+                    1135.0,
+                    1405.0,
+                    1171.0,
+                    1209.0,
+                    1171.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1166.0,
+                    1405.0,
+                    1166.0,
+                    1405.0,
+                    1202.0,
+                    295.0,
+                    1202.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1194.0,
+                    1406.0,
+                    1194.0,
+                    1406.0,
+                    1235.0,
+                    292.0,
+                    1235.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1225.0,
+                    562.0,
+                    1225.0,
+                    562.0,
+                    1263.0,
+                    292.0,
+                    1263.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    615.0,
+                    1225.0,
+                    1381.0,
+                    1225.0,
+                    1381.0,
+                    1263.0,
+                    615.0,
+                    1263.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1402.0,
+                    1225.0,
+                    1406.0,
+                    1225.0,
+                    1406.0,
+                    1263.0,
+                    1402.0,
+                    1263.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1257.0,
+                    1408.0,
+                    1257.0,
+                    1408.0,
+                    1293.0,
+                    295.0,
+                    1293.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1284.0,
+                    1406.0,
+                    1284.0,
+                    1406.0,
+                    1325.0,
+                    293.0,
+                    1325.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1315.0,
+                    1408.0,
+                    1315.0,
+                    1408.0,
+                    1356.0,
+                    291.0,
+                    1356.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1347.0,
+                    1406.0,
+                    1347.0,
+                    1406.0,
+                    1383.0,
+                    293.0,
+                    1383.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1378.0,
+                    1407.0,
+                    1378.0,
+                    1407.0,
+                    1413.0,
+                    293.0,
+                    1413.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1408.0,
+                    1406.0,
+                    1408.0,
+                    1406.0,
+                    1444.0,
+                    295.0,
+                    1444.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1438.0,
+                    977.0,
+                    1438.0,
+                    977.0,
+                    1474.0,
+                    293.0,
+                    1474.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1567.0,
+                    1405.0,
+                    1567.0,
+                    1405.0,
+                    1604.0,
+                    296.0,
+                    1604.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1597.0,
+                    1076.0,
+                    1597.0,
+                    1076.0,
+                    1634.0,
+                    294.0,
+                    1634.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1153.0,
+                    1597.0,
+                    1407.0,
+                    1597.0,
+                    1407.0,
+                    1634.0,
+                    1153.0,
+                    1634.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1627.0,
+                    1408.0,
+                    1627.0,
+                    1408.0,
+                    1667.0,
+                    292.0,
+                    1667.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1656.0,
+                    1405.0,
+                    1656.0,
+                    1405.0,
+                    1696.0,
+                    293.0,
+                    1696.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1688.0,
+                    977.0,
+                    1688.0,
+                    977.0,
+                    1725.0,
+                    294.0,
+                    1725.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1734.0,
+                    1407.0,
+                    1734.0,
+                    1407.0,
+                    1770.0,
+                    295.0,
+                    1770.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1763.0,
+                    408.0,
+                    1763.0,
+                    408.0,
+                    1801.0,
+                    292.0,
+                    1801.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    466.0,
+                    1763.0,
+                    513.0,
+                    1763.0,
+                    513.0,
+                    1801.0,
+                    466.0,
+                    1801.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    605.0,
+                    1763.0,
+                    1408.0,
+                    1763.0,
+                    1408.0,
+                    1801.0,
+                    605.0,
+                    1801.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1796.0,
+                    1406.0,
+                    1796.0,
+                    1406.0,
+                    1832.0,
+                    295.0,
+                    1832.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1821.0,
+                    509.0,
+                    1821.0,
+                    509.0,
+                    1867.0,
+                    291.0,
+                    1867.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    526.0,
+                    1821.0,
+                    985.0,
+                    1821.0,
+                    985.0,
+                    1867.0,
+                    526.0,
+                    1867.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1088.0,
+                    1821.0,
+                    1231.0,
+                    1821.0,
+                    1231.0,
+                    1867.0,
+                    1088.0,
+                    1867.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1251.0,
+                    1821.0,
+                    1408.0,
+                    1821.0,
+                    1408.0,
+                    1867.0,
+                    1251.0,
+                    1867.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1857.0,
+                    1408.0,
+                    1857.0,
+                    1408.0,
+                    1892.0,
+                    295.0,
+                    1892.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1888.0,
+                    336.0,
+                    1888.0,
+                    336.0,
+                    1920.0,
+                    296.0,
+                    1920.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    355.0,
+                    1888.0,
+                    751.0,
+                    1888.0,
+                    751.0,
+                    1920.0,
+                    355.0,
+                    1920.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    763.0,
+                    1404.0,
+                    763.0,
+                    1404.0,
+                    796.0,
+                    296.0,
+                    796.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    792.0,
+                    1405.0,
+                    792.0,
+                    1405.0,
+                    829.0,
+                    294.0,
+                    829.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    822.0,
+                    1209.0,
+                    822.0,
+                    1209.0,
+                    859.0,
+                    292.0,
+                    859.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1234.0,
+                    822.0,
+                    1405.0,
+                    822.0,
+                    1405.0,
+                    859.0,
+                    1234.0,
+                    859.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    853.0,
+                    296.0,
+                    853.0,
+                    296.0,
+                    891.0,
+                    291.0,
+                    891.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    322.0,
+                    853.0,
+                    801.0,
+                    853.0,
+                    801.0,
+                    891.0,
+                    322.0,
+                    891.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    821.0,
+                    853.0,
+                    1408.0,
+                    853.0,
+                    1408.0,
+                    891.0,
+                    821.0,
+                    891.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    886.0,
+                    1088.0,
+                    886.0,
+                    1088.0,
+                    919.0,
+                    296.0,
+                    919.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 7,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1546,
+                    1405,
+                    1546,
+                    1405,
+                    1851,
+                    298,
+                    1851
+                ],
+                "score": 0.981
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    298,
+                    1193,
+                    1404,
+                    1193,
+                    1404,
+                    1437,
+                    298,
+                    1437
+                ],
+                "score": 0.98
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    299,
+                    900,
+                    1403,
+                    900,
+                    1403,
+                    1084,
+                    299,
+                    1084
+                ],
+                "score": 0.977
+            },
+            {
+                "category_id": 4,
+                "poly": [
+                    296,
+                    600,
+                    1405,
+                    600,
+                    1405,
+                    846,
+                    296,
+                    846
+                ],
+                "score": 0.975
+            },
+            {
+                "category_id": 3,
+                "poly": [
+                    304,
+                    178,
+                    1390,
+                    178,
+                    1390,
+                    583,
+                    304,
+                    583
+                ],
+                "score": 0.973
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    313,
+                    1945,
+                    1401,
+                    1945,
+                    1401,
+                    2007,
+                    313,
+                    2007
+                ],
+                "score": 0.925
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    301,
+                    1479,
+                    561,
+                    1479,
+                    561,
+                    1515,
+                    301,
+                    1515
+                ],
+                "score": 0.908
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    298,
+                    1125,
+                    509,
+                    1125,
+                    509,
+                    1162,
+                    298,
+                    1162
+                ],
+                "score": 0.9
+            },
+            {
+                "category_id": 0,
+                "poly": [
+                    299,
+                    1894,
+                    455,
+                    1894,
+                    455,
+                    1928,
+                    299,
+                    1928
+                ],
+                "score": 0.896
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    841,
+                    2062,
+                    858,
+                    2062,
+                    858,
+                    2083,
+                    841,
+                    2083
+                ],
+                "score": 0.784
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    963,
+                    931,
+                    1014,
+                    931,
+                    1014,
+                    964,
+                    963,
+                    964
+                ],
+                "score": 0.88,
+                "latex": "z / w"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1175,
+                    901,
+                    1267,
+                    901,
+                    1267,
+                    935,
+                    1175,
+                    935
+                ],
+                "score": 0.85,
+                "latex": "{ \\mathrm { G I } } { - z } / w"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    714,
+                    993,
+                    764,
+                    993,
+                    764,
+                    1025,
+                    714,
+                    1025
+                ],
+                "score": 0.82,
+                "latex": "z / w"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    674,
+                    1638,
+                    728,
+                    1638,
+                    728,
+                    1667,
+                    674,
+                    1667
+                ],
+                "score": 0.81,
+                "latex": "^ { 5 \\mathrm { G } + }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    552,
+                    692,
+                    603,
+                    692,
+                    603,
+                    725,
+                    552,
+                    725
+                ],
+                "score": 0.81,
+                "latex": "z / w"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    539,
+                    967,
+                    562,
+                    967,
+                    562,
+                    989,
+                    539,
+                    989
+                ],
+                "score": 0.79,
+                "latex": "\\boldsymbol { \\cdot } \\boldsymbol { w }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    427,
+                    936,
+                    444,
+                    936,
+                    444,
+                    959,
+                    427,
+                    959
+                ],
+                "score": 0.78,
+                "latex": "z"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1211,
+                    757,
+                    1234,
+                    757,
+                    1234,
+                    782,
+                    1211,
+                    782
+                ],
+                "score": 0.77,
+                "latex": "\\boldsymbol { w }"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    446,
+                    696,
+                    463,
+                    696,
+                    463,
+                    721,
+                    446,
+                    721
+                ],
+                "score": 0.74,
+                "latex": "z"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    424,
+                    668,
+                    443,
+                    668,
+                    443,
+                    690,
+                    424,
+                    690
+                ],
+                "score": 0.72,
+                "latex": "z"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    494,
+                    932,
+                    585,
+                    932,
+                    585,
+                    964,
+                    494,
+                    964
+                ],
+                "score": 0.66,
+                "latex": "{ \\mathrm { G I } } { - z } / w"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1108,
+                    905,
+                    1125,
+                    905,
+                    1125,
+                    930,
+                    1108,
+                    930
+                ],
+                "score": 0.65,
+                "latex": "z"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    441,
+                    992,
+                    531,
+                    992,
+                    531,
+                    1025,
+                    441,
+                    1025
+                ],
+                "score": 0.63,
+                "latex": "{ \\mathrm { G I } } { - z } / w"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    408,
+                    902,
+                    458,
+                    902,
+                    458,
+                    932,
+                    408,
+                    932
+                ],
+                "score": 0.32,
+                "latex": "6 ( \\mathbf { b } )"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    406,
+                    693,
+                    464,
+                    693,
+                    464,
+                    722,
+                    406,
+                    722
+                ],
+                "score": 0.29,
+                "latex": "\\mathrm { G I } { - } z"
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1068,
+                    902,
+                    1126,
+                    902,
+                    1126,
+                    931,
+                    1068,
+                    931
+                ],
+                "score": 0.26,
+                "latex": "\\mathrm { G I } { - } z"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    600.0,
+                    1406.0,
+                    600.0,
+                    1406.0,
+                    636.0,
+                    293.0,
+                    636.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    633.0,
+                    1405.0,
+                    633.0,
+                    1405.0,
+                    667.0,
+                    294.0,
+                    667.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    662.0,
+                    423.0,
+                    662.0,
+                    423.0,
+                    699.0,
+                    293.0,
+                    699.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    444.0,
+                    662.0,
+                    1406.0,
+                    662.0,
+                    1406.0,
+                    699.0,
+                    444.0,
+                    699.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    693.0,
+                    405.0,
+                    693.0,
+                    405.0,
+                    727.0,
+                    294.0,
+                    727.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    465.0,
+                    693.0,
+                    551.0,
+                    693.0,
+                    551.0,
+                    727.0,
+                    465.0,
+                    727.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    604.0,
+                    693.0,
+                    1405.0,
+                    693.0,
+                    1405.0,
+                    727.0,
+                    604.0,
+                    727.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    723.0,
+                    1403.0,
+                    723.0,
+                    1403.0,
+                    759.0,
+                    293.0,
+                    759.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    753.0,
+                    1210.0,
+                    753.0,
+                    1210.0,
+                    790.0,
+                    293.0,
+                    790.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1235.0,
+                    753.0,
+                    1407.0,
+                    753.0,
+                    1407.0,
+                    790.0,
+                    1235.0,
+                    790.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    783.0,
+                    1406.0,
+                    783.0,
+                    1406.0,
+                    819.0,
+                    294.0,
+                    819.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    816.0,
+                    1085.0,
+                    816.0,
+                    1085.0,
+                    849.0,
+                    295.0,
+                    849.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    382.0,
+                    181.0,
+                    445.0,
+                    181.0,
+                    445.0,
+                    201.0,
+                    382.0,
+                    201.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    476.0,
+                    181.0,
+                    550.0,
+                    181.0,
+                    550.0,
+                    201.0,
+                    476.0,
+                    201.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    579.0,
+                    181.0,
+                    653.0,
+                    181.0,
+                    653.0,
+                    201.0,
+                    579.0,
+                    201.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    686.0,
+                    181.0,
+                    747.0,
+                    181.0,
+                    747.0,
+                    201.0,
+                    686.0,
+                    201.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    937.0,
+                    220.0,
+                    958.0,
+                    220.0,
+                    958.0,
+                    238.0,
+                    937.0,
+                    238.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    811.0,
+                    237.0,
+                    889.0,
+                    237.0,
+                    889.0,
+                    257.0,
+                    811.0,
+                    257.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    302.0,
+                    252.0,
+                    366.0,
+                    252.0,
+                    366.0,
+                    274.0,
+                    302.0,
+                    274.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    937.0,
+                    271.0,
+                    958.0,
+                    271.0,
+                    958.0,
+                    289.0,
+                    937.0,
+                    289.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    913.0,
+                    309.0,
+                    965.0,
+                    309.0,
+                    965.0,
+                    397.0,
+                    913.0,
+                    397.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1165.0,
+                    338.0,
+                    1179.0,
+                    338.0,
+                    1179.0,
+                    356.0,
+                    1165.0,
+                    356.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    308.0,
+                    364.0,
+                    347.0,
+                    364.0,
+                    347.0,
+                    385.0,
+                    308.0,
+                    385.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    937.0,
+                    372.0,
+                    953.0,
+                    372.0,
+                    953.0,
+                    388.0,
+                    937.0,
+                    388.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    830.0,
+                    385.0,
+                    862.0,
+                    385.0,
+                    862.0,
+                    402.0,
+                    830.0,
+                    402.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1220.0,
+                    402.0,
+                    1230.0,
+                    402.0,
+                    1230.0,
+                    411.0,
+                    1220.0,
+                    411.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1245.0,
+                    391.0,
+                    1379.0,
+                    391.0,
+                    1379.0,
+                    414.0,
+                    1245.0,
+                    414.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    938.0,
+                    423.0,
+                    956.0,
+                    423.0,
+                    956.0,
+                    439.0,
+                    938.0,
+                    439.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1246.0,
+                    412.0,
+                    1368.0,
+                    412.0,
+                    1368.0,
+                    435.0,
+                    1246.0,
+                    435.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1219.0,
+                    436.0,
+                    1230.0,
+                    436.0,
+                    1230.0,
+                    447.0,
+                    1219.0,
+                    447.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1244.0,
+                    430.0,
+                    1366.0,
+                    430.0,
+                    1366.0,
+                    457.0,
+                    1244.0,
+                    457.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1245.0,
+                    453.0,
+                    1355.0,
+                    453.0,
+                    1355.0,
+                    476.0,
+                    1245.0,
+                    476.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    305.0,
+                    479.0,
+                    354.0,
+                    479.0,
+                    354.0,
+                    499.0,
+                    305.0,
+                    499.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    945.0,
+                    476.0,
+                    954.0,
+                    476.0,
+                    954.0,
+                    486.0,
+                    945.0,
+                    486.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1244.0,
+                    471.0,
+                    1359.0,
+                    471.0,
+                    1359.0,
+                    498.0,
+                    1244.0,
+                    498.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    969.0,
+                    509.0,
+                    979.0,
+                    509.0,
+                    979.0,
+                    519.0,
+                    969.0,
+                    519.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1063.0,
+                    504.0,
+                    1084.0,
+                    504.0,
+                    1084.0,
+                    524.0,
+                    1063.0,
+                    524.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1158.0,
+                    504.0,
+                    1187.0,
+                    504.0,
+                    1187.0,
+                    523.0,
+                    1158.0,
+                    523.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1256.0,
+                    503.0,
+                    1287.0,
+                    503.0,
+                    1287.0,
+                    525.0,
+                    1256.0,
+                    525.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1355.0,
+                    504.0,
+                    1382.0,
+                    504.0,
+                    1382.0,
+                    523.0,
+                    1355.0,
+                    523.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    589.0,
+                    554.0,
+                    631.0,
+                    554.0,
+                    631.0,
+                    588.0,
+                    589.0,
+                    588.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1135.0,
+                    554.0,
+                    1177.0,
+                    554.0,
+                    1177.0,
+                    588.0,
+                    1135.0,
+                    588.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1150.75,
+                    517.0,
+                    1191.75,
+                    517.0,
+                    1191.75,
+                    542.5,
+                    1150.75,
+                    542.5
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1473.0,
+                    565.0,
+                    1473.0,
+                    565.0,
+                    1524.0,
+                    295.0,
+                    1524.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1120.0,
+                    514.0,
+                    1120.0,
+                    514.0,
+                    1170.0,
+                    291.0,
+                    1170.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1890.0,
+                    460.0,
+                    1890.0,
+                    460.0,
+                    1934.0,
+                    295.0,
+                    1934.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    839.0,
+                    2060.0,
+                    860.0,
+                    2060.0,
+                    860.0,
+                    2090.0,
+                    839.0,
+                    2090.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1543.0,
+                    1405.0,
+                    1543.0,
+                    1405.0,
+                    1585.0,
+                    293.0,
+                    1585.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1576.0,
+                    1407.0,
+                    1576.0,
+                    1407.0,
+                    1615.0,
+                    292.0,
+                    1615.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1607.0,
+                    1407.0,
+                    1607.0,
+                    1407.0,
+                    1643.0,
+                    293.0,
+                    1643.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1635.0,
+                    673.0,
+                    1635.0,
+                    673.0,
+                    1673.0,
+                    293.0,
+                    1673.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    729.0,
+                    1635.0,
+                    1407.0,
+                    1635.0,
+                    1407.0,
+                    1673.0,
+                    729.0,
+                    1673.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1666.0,
+                    1408.0,
+                    1666.0,
+                    1408.0,
+                    1707.0,
+                    291.0,
+                    1707.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1697.0,
+                    1407.0,
+                    1697.0,
+                    1407.0,
+                    1736.0,
+                    292.0,
+                    1736.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1729.0,
+                    1405.0,
+                    1729.0,
+                    1405.0,
+                    1764.0,
+                    293.0,
+                    1764.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1758.0,
+                    1405.0,
+                    1758.0,
+                    1405.0,
+                    1793.0,
+                    293.0,
+                    1793.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1787.0,
+                    1407.0,
+                    1787.0,
+                    1407.0,
+                    1825.0,
+                    292.0,
+                    1825.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1820.0,
+                    867.0,
+                    1820.0,
+                    867.0,
+                    1855.0,
+                    296.0,
+                    1855.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1194.0,
+                    1405.0,
+                    1194.0,
+                    1405.0,
+                    1228.0,
+                    296.0,
+                    1228.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1225.0,
+                    1405.0,
+                    1225.0,
+                    1405.0,
+                    1259.0,
+                    294.0,
+                    1259.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1256.0,
+                    1404.0,
+                    1256.0,
+                    1404.0,
+                    1290.0,
+                    296.0,
+                    1290.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1284.0,
+                    1408.0,
+                    1284.0,
+                    1408.0,
+                    1321.0,
+                    293.0,
+                    1321.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1315.0,
+                    1405.0,
+                    1315.0,
+                    1405.0,
+                    1350.0,
+                    293.0,
+                    1350.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1342.0,
+                    1405.0,
+                    1342.0,
+                    1405.0,
+                    1382.0,
+                    292.0,
+                    1382.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1375.0,
+                    1408.0,
+                    1375.0,
+                    1408.0,
+                    1412.0,
+                    293.0,
+                    1412.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1406.0,
+                    941.0,
+                    1406.0,
+                    941.0,
+                    1439.0,
+                    293.0,
+                    1439.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    901.0,
+                    407.0,
+                    901.0,
+                    407.0,
+                    936.0,
+                    294.0,
+                    936.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    459.0,
+                    901.0,
+                    1067.0,
+                    901.0,
+                    1067.0,
+                    936.0,
+                    459.0,
+                    936.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1127.0,
+                    901.0,
+                    1174.0,
+                    901.0,
+                    1174.0,
+                    936.0,
+                    1127.0,
+                    936.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1268.0,
+                    901.0,
+                    1407.0,
+                    901.0,
+                    1407.0,
+                    936.0,
+                    1268.0,
+                    936.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    930.0,
+                    426.0,
+                    930.0,
+                    426.0,
+                    967.0,
+                    293.0,
+                    967.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    445.0,
+                    930.0,
+                    493.0,
+                    930.0,
+                    493.0,
+                    967.0,
+                    445.0,
+                    967.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    586.0,
+                    930.0,
+                    962.0,
+                    930.0,
+                    962.0,
+                    967.0,
+                    586.0,
+                    967.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1015.0,
+                    930.0,
+                    1407.0,
+                    930.0,
+                    1407.0,
+                    967.0,
+                    1015.0,
+                    967.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    963.0,
+                    538.0,
+                    963.0,
+                    538.0,
+                    995.0,
+                    295.0,
+                    995.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    563.0,
+                    963.0,
+                    1404.0,
+                    963.0,
+                    1404.0,
+                    995.0,
+                    563.0,
+                    995.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    992.0,
+                    440.0,
+                    992.0,
+                    440.0,
+                    1028.0,
+                    294.0,
+                    1028.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    532.0,
+                    992.0,
+                    713.0,
+                    992.0,
+                    713.0,
+                    1028.0,
+                    532.0,
+                    1028.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    765.0,
+                    992.0,
+                    1405.0,
+                    992.0,
+                    1405.0,
+                    1028.0,
+                    765.0,
+                    1028.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1021.0,
+                    1403.0,
+                    1021.0,
+                    1403.0,
+                    1057.0,
+                    295.0,
+                    1057.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1053.0,
+                    915.0,
+                    1053.0,
+                    915.0,
+                    1088.0,
+                    293.0,
+                    1088.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    308.0,
+                    1943.0,
+                    1405.0,
+                    1943.0,
+                    1405.0,
+                    1981.0,
+                    308.0,
+                    1981.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    351.0,
+                    1972.0,
+                    1408.0,
+                    1972.0,
+                    1408.0,
+                    2014.0,
+                    351.0,
+                    2014.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 8,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 1,
+                "poly": [
+                    292,
+                    160,
+                    1409,
+                    160,
+                    1409,
+                    2013,
+                    292,
+                    2013
+                ],
+                "score": 0.813
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    836,
+                    2061,
+                    866,
+                    2061,
+                    866,
+                    2086,
+                    836,
+                    2086
+                ],
+                "score": 0.607
+            },
+            {
+                "category_id": 2,
+                "poly": [
+                    836,
+                    2061,
+                    866,
+                    2061,
+                    866,
+                    2086,
+                    836,
+                    2086
+                ],
+                "score": 0.509
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    832.0,
+                    2058.0,
+                    870.0,
+                    2058.0,
+                    870.0,
+                    2096.0,
+                    832.0,
+                    2096.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    832.0,
+                    2058.0,
+                    870.0,
+                    2058.0,
+                    870.0,
+                    2096.0,
+                    832.0,
+                    2096.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    352.0,
+                    202.0,
+                    486.0,
+                    202.0,
+                    486.0,
+                    240.0,
+                    352.0,
+                    240.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    304.0,
+                    252.0,
+                    1409.0,
+                    252.0,
+                    1409.0,
+                    295.0,
+                    304.0,
+                    295.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    354.0,
+                    289.0,
+                    1409.0,
+                    289.0,
+                    1409.0,
+                    322.0,
+                    354.0,
+                    322.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    354.0,
+                    315.0,
+                    1407.0,
+                    315.0,
+                    1407.0,
+                    354.0,
+                    354.0,
+                    354.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    346.0,
+                    338.0,
+                    1411.0,
+                    338.0,
+                    1411.0,
+                    393.0,
+                    346.0,
+                    393.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    352.0,
+                    374.0,
+                    1305.0,
+                    374.0,
+                    1305.0,
+                    419.0,
+                    352.0,
+                    419.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    308.0,
+                    431.0,
+                    1405.0,
+                    431.0,
+                    1405.0,
+                    470.0,
+                    308.0,
+                    470.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    348.0,
+                    458.0,
+                    1407.0,
+                    458.0,
+                    1407.0,
+                    501.0,
+                    348.0,
+                    501.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    352.0,
+                    488.0,
+                    1409.0,
+                    488.0,
+                    1409.0,
+                    529.0,
+                    352.0,
+                    529.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    354.0,
+                    525.0,
+                    474.0,
+                    525.0,
+                    474.0,
+                    558.0,
+                    354.0,
+                    558.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    308.0,
+                    574.0,
+                    1405.0,
+                    574.0,
+                    1405.0,
+                    612.0,
+                    308.0,
+                    612.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    350.0,
+                    604.0,
+                    1407.0,
+                    604.0,
+                    1407.0,
+                    647.0,
+                    350.0,
+                    647.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    358.0,
+                    639.0,
+                    692.0,
+                    639.0,
+                    692.0,
+                    671.0,
+                    358.0,
+                    671.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    304.0,
+                    684.0,
+                    1409.0,
+                    684.0,
+                    1409.0,
+                    731.0,
+                    304.0,
+                    731.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    356.0,
+                    722.0,
+                    1403.0,
+                    722.0,
+                    1403.0,
+                    755.0,
+                    356.0,
+                    755.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    354.0,
+                    753.0,
+                    640.0,
+                    753.0,
+                    640.0,
+                    785.0,
+                    354.0,
+                    785.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    306.0,
+                    804.0,
+                    461.0,
+                    804.0,
+                    461.0,
+                    838.0,
+                    306.0,
+                    838.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    530.0,
+                    804.0,
+                    887.0,
+                    804.0,
+                    887.0,
+                    842.0,
+                    530.0,
+                    842.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    961.0,
+                    804.0,
+                    1405.0,
+                    804.0,
+                    1405.0,
+                    842.0,
+                    961.0,
+                    842.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    352.0,
+                    834.0,
+                    682.0,
+                    834.0,
+                    682.0,
+                    873.0,
+                    352.0,
+                    873.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    308.0,
+                    887.0,
+                    1405.0,
+                    887.0,
+                    1405.0,
+                    926.0,
+                    308.0,
+                    926.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    348.0,
+                    916.0,
+                    1409.0,
+                    916.0,
+                    1409.0,
+                    958.0,
+                    348.0,
+                    958.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    352.0,
+                    948.0,
+                    1347.0,
+                    948.0,
+                    1347.0,
+                    987.0,
+                    352.0,
+                    987.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    308.0,
+                    1001.0,
+                    1405.0,
+                    1001.0,
+                    1405.0,
+                    1040.0,
+                    308.0,
+                    1040.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    350.0,
+                    1030.0,
+                    1183.0,
+                    1030.0,
+                    1183.0,
+                    1070.0,
+                    350.0,
+                    1070.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    308.0,
+                    1084.0,
+                    1407.0,
+                    1084.0,
+                    1407.0,
+                    1123.0,
+                    308.0,
+                    1123.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    354.0,
+                    1119.0,
+                    1407.0,
+                    1119.0,
+                    1407.0,
+                    1152.0,
+                    354.0,
+                    1152.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    352.0,
+                    1143.0,
+                    430.0,
+                    1143.0,
+                    430.0,
+                    1184.0,
+                    352.0,
+                    1184.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    296.0,
+                    1198.0,
+                    1407.0,
+                    1198.0,
+                    1407.0,
+                    1237.0,
+                    296.0,
+                    1237.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    354.0,
+                    1231.0,
+                    1409.0,
+                    1231.0,
+                    1409.0,
+                    1270.0,
+                    354.0,
+                    1270.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    354.0,
+                    1261.0,
+                    430.0,
+                    1261.0,
+                    430.0,
+                    1296.0,
+                    354.0,
+                    1296.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1314.0,
+                    1405.0,
+                    1314.0,
+                    1405.0,
+                    1353.0,
+                    294.0,
+                    1353.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    354.0,
+                    1343.0,
+                    1405.0,
+                    1343.0,
+                    1405.0,
+                    1381.0,
+                    354.0,
+                    1381.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    354.0,
+                    1373.0,
+                    969.0,
+                    1373.0,
+                    969.0,
+                    1412.0,
+                    354.0,
+                    1412.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1428.0,
+                    1405.0,
+                    1428.0,
+                    1405.0,
+                    1467.0,
+                    294.0,
+                    1467.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    352.0,
+                    1459.0,
+                    1407.0,
+                    1459.0,
+                    1407.0,
+                    1497.0,
+                    352.0,
+                    1497.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    354.0,
+                    1489.0,
+                    724.0,
+                    1489.0,
+                    724.0,
+                    1522.0,
+                    354.0,
+                    1522.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1538.0,
+                    1405.0,
+                    1538.0,
+                    1405.0,
+                    1581.0,
+                    292.0,
+                    1581.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    354.0,
+                    1573.0,
+                    1405.0,
+                    1573.0,
+                    1405.0,
+                    1611.0,
+                    354.0,
+                    1611.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    354.0,
+                    1601.0,
+                    872.0,
+                    1601.0,
+                    872.0,
+                    1638.0,
+                    354.0,
+                    1638.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1656.0,
+                    1407.0,
+                    1656.0,
+                    1407.0,
+                    1695.0,
+                    294.0,
+                    1695.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    354.0,
+                    1687.0,
+                    1409.0,
+                    1687.0,
+                    1409.0,
+                    1725.0,
+                    354.0,
+                    1725.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    350.0,
+                    1713.0,
+                    1411.0,
+                    1713.0,
+                    1411.0,
+                    1758.0,
+                    350.0,
+                    1758.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    352.0,
+                    1748.0,
+                    734.0,
+                    1748.0,
+                    734.0,
+                    1786.0,
+                    352.0,
+                    1786.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    294.0,
+                    1799.0,
+                    1405.0,
+                    1799.0,
+                    1405.0,
+                    1837.0,
+                    294.0,
+                    1837.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    350.0,
+                    1829.0,
+                    1407.0,
+                    1829.0,
+                    1407.0,
+                    1870.0,
+                    350.0,
+                    1870.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    352.0,
+                    1860.0,
+                    518.0,
+                    1860.0,
+                    518.0,
+                    1898.0,
+                    352.0,
+                    1898.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    292.0,
+                    1910.0,
+                    1409.0,
+                    1910.0,
+                    1409.0,
+                    1955.0,
+                    292.0,
+                    1955.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    354.0,
+                    1945.0,
+                    1407.0,
+                    1945.0,
+                    1407.0,
+                    1984.0,
+                    354.0,
+                    1984.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    354.0,
+                    1978.0,
+                    808.0,
+                    1978.0,
+                    808.0,
+                    2010.0,
+                    354.0,
+                    2010.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 9,
+            "width": 1700,
+            "height": 2200
+        }
+    },
+    {
+        "layout_dets": [
+            {
+                "category_id": 2,
+                "poly": [
+                    835,
+                    2061,
+                    864,
+                    2061,
+                    864,
+                    2087,
+                    835,
+                    2087
+                ],
+                "score": 0.843
+            },
+            {
+                "category_id": 1,
+                "poly": [
+                    293,
+                    157,
+                    1411,
+                    157,
+                    1411,
+                    2008,
+                    293,
+                    2008
+                ],
+                "score": 0.697
+            },
+            {
+                "category_id": 13,
+                "poly": [
+                    1048,
+                    454,
+                    1096,
+                    454,
+                    1096,
+                    483,
+                    1048,
+                    483
+                ],
+                "score": 0.4,
+                "latex": "( S P )"
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    831.0,
+                    2058.0,
+                    869.0,
+                    2058.0,
+                    869.0,
+                    2097.0,
+                    831.0,
+                    2097.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    203.0,
+                    1407.0,
+                    203.0,
+                    1407.0,
+                    241.0,
+                    295.0,
+                    241.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    353.0,
+                    233.0,
+                    1409.0,
+                    233.0,
+                    1409.0,
+                    272.0,
+                    353.0,
+                    272.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    353.0,
+                    263.0,
+                    1409.0,
+                    263.0,
+                    1409.0,
+                    302.0,
+                    353.0,
+                    302.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    353.0,
+                    294.0,
+                    1287.0,
+                    294.0,
+                    1287.0,
+                    333.0,
+                    353.0,
+                    333.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    343.0,
+                    1407.0,
+                    343.0,
+                    1407.0,
+                    381.0,
+                    295.0,
+                    381.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    349.0,
+                    371.0,
+                    1409.0,
+                    371.0,
+                    1409.0,
+                    414.0,
+                    349.0,
+                    414.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    422.0,
+                    1407.0,
+                    422.0,
+                    1407.0,
+                    461.0,
+                    293.0,
+                    461.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    351.0,
+                    452.0,
+                    1047.0,
+                    452.0,
+                    1047.0,
+                    491.0,
+                    351.0,
+                    491.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    1097.0,
+                    452.0,
+                    1409.0,
+                    452.0,
+                    1409.0,
+                    491.0,
+                    1097.0,
+                    491.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    355.0,
+                    483.0,
+                    632.0,
+                    483.0,
+                    632.0,
+                    515.0,
+                    355.0,
+                    515.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    532.0,
+                    1407.0,
+                    532.0,
+                    1407.0,
+                    570.0,
+                    295.0,
+                    570.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    351.0,
+                    562.0,
+                    643.0,
+                    562.0,
+                    643.0,
+                    599.0,
+                    351.0,
+                    599.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    611.0,
+                    1407.0,
+                    611.0,
+                    1407.0,
+                    650.0,
+                    295.0,
+                    650.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    351.0,
+                    637.0,
+                    1409.0,
+                    637.0,
+                    1409.0,
+                    682.0,
+                    351.0,
+                    682.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    353.0,
+                    668.0,
+                    632.0,
+                    668.0,
+                    632.0,
+                    707.0,
+                    353.0,
+                    707.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    719.0,
+                    1409.0,
+                    719.0,
+                    1409.0,
+                    759.0,
+                    293.0,
+                    759.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    351.0,
+                    751.0,
+                    1407.0,
+                    751.0,
+                    1407.0,
+                    790.0,
+                    351.0,
+                    790.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    355.0,
+                    782.0,
+                    1409.0,
+                    782.0,
+                    1409.0,
+                    820.0,
+                    355.0,
+                    820.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    351.0,
+                    806.0,
+                    692.0,
+                    806.0,
+                    692.0,
+                    853.0,
+                    351.0,
+                    853.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    857.0,
+                    1409.0,
+                    857.0,
+                    1409.0,
+                    902.0,
+                    293.0,
+                    902.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    349.0,
+                    887.0,
+                    1407.0,
+                    887.0,
+                    1407.0,
+                    932.0,
+                    349.0,
+                    932.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    351.0,
+                    918.0,
+                    1124.0,
+                    918.0,
+                    1124.0,
+                    961.0,
+                    351.0,
+                    961.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    971.0,
+                    1407.0,
+                    971.0,
+                    1407.0,
+                    1009.0,
+                    295.0,
+                    1009.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    353.0,
+                    1001.0,
+                    1409.0,
+                    1001.0,
+                    1409.0,
+                    1040.0,
+                    353.0,
+                    1040.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    357.0,
+                    1032.0,
+                    720.0,
+                    1032.0,
+                    720.0,
+                    1064.0,
+                    357.0,
+                    1064.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1074.0,
+                    1407.0,
+                    1074.0,
+                    1407.0,
+                    1121.0,
+                    291.0,
+                    1121.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    351.0,
+                    1109.0,
+                    676.0,
+                    1109.0,
+                    676.0,
+                    1148.0,
+                    351.0,
+                    1148.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    293.0,
+                    1156.0,
+                    1407.0,
+                    1156.0,
+                    1407.0,
+                    1200.0,
+                    293.0,
+                    1200.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    353.0,
+                    1190.0,
+                    1225.0,
+                    1190.0,
+                    1225.0,
+                    1229.0,
+                    353.0,
+                    1229.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1237.0,
+                    1409.0,
+                    1237.0,
+                    1409.0,
+                    1276.0,
+                    295.0,
+                    1276.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    353.0,
+                    1267.0,
+                    1407.0,
+                    1267.0,
+                    1407.0,
+                    1306.0,
+                    353.0,
+                    1306.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    353.0,
+                    1300.0,
+                    1120.0,
+                    1300.0,
+                    1120.0,
+                    1339.0,
+                    353.0,
+                    1339.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    291.0,
+                    1345.0,
+                    1407.0,
+                    1345.0,
+                    1407.0,
+                    1387.0,
+                    291.0,
+                    1387.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    347.0,
+                    1373.0,
+                    1409.0,
+                    1373.0,
+                    1409.0,
+                    1420.0,
+                    347.0,
+                    1420.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    353.0,
+                    1408.0,
+                    1407.0,
+                    1408.0,
+                    1407.0,
+                    1446.0,
+                    353.0,
+                    1446.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    353.0,
+                    1436.0,
+                    1042.0,
+                    1436.0,
+                    1042.0,
+                    1479.0,
+                    353.0,
+                    1479.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    297.0,
+                    1491.0,
+                    1405.0,
+                    1491.0,
+                    1405.0,
+                    1524.0,
+                    297.0,
+                    1524.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    353.0,
+                    1519.0,
+                    1407.0,
+                    1519.0,
+                    1407.0,
+                    1558.0,
+                    353.0,
+                    1558.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    355.0,
+                    1548.0,
+                    1008.0,
+                    1548.0,
+                    1008.0,
+                    1587.0,
+                    355.0,
+                    1587.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1597.0,
+                    1407.0,
+                    1597.0,
+                    1407.0,
+                    1635.0,
+                    295.0,
+                    1635.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    351.0,
+                    1627.0,
+                    1407.0,
+                    1627.0,
+                    1407.0,
+                    1668.0,
+                    351.0,
+                    1668.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    355.0,
+                    1658.0,
+                    880.0,
+                    1658.0,
+                    880.0,
+                    1696.0,
+                    355.0,
+                    1696.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1706.0,
+                    1405.0,
+                    1706.0,
+                    1405.0,
+                    1745.0,
+                    295.0,
+                    1745.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    347.0,
+                    1735.0,
+                    742.0,
+                    1735.0,
+                    742.0,
+                    1777.0,
+                    347.0,
+                    1777.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1786.0,
+                    1407.0,
+                    1786.0,
+                    1407.0,
+                    1824.0,
+                    295.0,
+                    1824.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    351.0,
+                    1816.0,
+                    978.0,
+                    1816.0,
+                    978.0,
+                    1855.0,
+                    351.0,
+                    1855.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1865.0,
+                    1407.0,
+                    1865.0,
+                    1407.0,
+                    1904.0,
+                    295.0,
+                    1904.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    353.0,
+                    1899.0,
+                    1175.0,
+                    1899.0,
+                    1175.0,
+                    1932.0,
+                    353.0,
+                    1932.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    295.0,
+                    1944.0,
+                    1407.0,
+                    1944.0,
+                    1407.0,
+                    1983.0,
+                    295.0,
+                    1983.0
+                ],
+                "score": 1.0,
+                "text": ""
+            },
+            {
+                "category_id": 15,
+                "poly": [
+                    353.0,
+                    1975.0,
+                    916.0,
+                    1975.0,
+                    916.0,
+                    2013.0,
+                    353.0,
+                    2013.0
+                ],
+                "score": 1.0,
+                "text": ""
+            }
+        ],
+        "page_info": {
+            "page_no": 10,
+            "width": 1700,
+            "height": 2200
+        }
+    }
+]
\ No newline at end of file
diff --git a/vlm/dev/2jUKhUrBxP/10.png b/vlm/dev/2jUKhUrBxP/10.png
new file mode 100644
index 0000000000000000000000000000000000000000..6029e2bb03d1e02d5513729d09caa9e746203cdb
--- /dev/null
+++ b/vlm/dev/2jUKhUrBxP/10.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4c1ae4effc660787521890a5ef1c5cec2facb2ac3bf12b45cd9528dbce10d208
+size 607589
diff --git a/vlm/dev/2jUKhUrBxP/14.png b/vlm/dev/2jUKhUrBxP/14.png
new file mode 100644
index 0000000000000000000000000000000000000000..ceb571cb70bc4fadeeefea40df2013054ff1379e
--- /dev/null
+++ b/vlm/dev/2jUKhUrBxP/14.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6b17dd7ce98b6c1ee88c6538340b937858ef97a2418ac01d4f82b5f5dbad817d
+size 618772
diff --git a/vlm/dev/2jUKhUrBxP/17.png b/vlm/dev/2jUKhUrBxP/17.png
new file mode 100644
index 0000000000000000000000000000000000000000..8d563d73d90c020dd98bd5294322f7e0d98dc976
--- /dev/null
+++ b/vlm/dev/2jUKhUrBxP/17.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cf63e516dde2b9dc04395c845026e556728c8764027db482b8a2d1239d56a3e3
+size 312530
diff --git a/vlm/dev/2jUKhUrBxP/2.png b/vlm/dev/2jUKhUrBxP/2.png
new file mode 100644
index 0000000000000000000000000000000000000000..d8e42ee4250acd260619edad02f7cb3b68391925
--- /dev/null
+++ b/vlm/dev/2jUKhUrBxP/2.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5cc338ab0db5a55679976d1eb7198c9854e477df75c53a73bb2510684c9bd4f4
+size 580847
diff --git a/vlm/dev/2jUKhUrBxP/3.png b/vlm/dev/2jUKhUrBxP/3.png
new file mode 100644
index 0000000000000000000000000000000000000000..8a96484caed1bd6544e96e1c6879d67a71d43972
--- /dev/null
+++ b/vlm/dev/2jUKhUrBxP/3.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e38c8cb1b080da77940548d83c216ecdd7fee830d370e706005d80ec2876b3c6
+size 542384
diff --git a/vlm/dev/2jUKhUrBxP/6.png b/vlm/dev/2jUKhUrBxP/6.png
new file mode 100644
index 0000000000000000000000000000000000000000..004109b547ac23f6b8516aeb1e00a3e862ccfcf1
--- /dev/null
+++ b/vlm/dev/2jUKhUrBxP/6.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:90c247b75017ae17e2aecbd093664161109334b2ed3b488c4115750bc7f0f46d
+size 534357
diff --git a/vlm/dev/2jUKhUrBxP/9.png b/vlm/dev/2jUKhUrBxP/9.png
new file mode 100644
index 0000000000000000000000000000000000000000..c11425a30b8a61f97a42a87bee3f77706cb16300
--- /dev/null
+++ b/vlm/dev/2jUKhUrBxP/9.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5add87d082ecab4cf4e10a2d96c9de88a50f086f1e2c3527ba0b4a2442786db4
+size 508972
diff --git a/vlm/dev/GkDbQb6qu_r/0.png b/vlm/dev/GkDbQb6qu_r/0.png
new file mode 100644
index 0000000000000000000000000000000000000000..69ef6b370681507c11c4d95f995581d92d8a02b7
--- /dev/null
+++ b/vlm/dev/GkDbQb6qu_r/0.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a8cfc1f7451bc93ab2d0bcd980cdf53b494ba88577acae70bef09a386b907ec2
+size 1317501
diff --git a/vlm/dev/GkDbQb6qu_r/1.png b/vlm/dev/GkDbQb6qu_r/1.png
new file mode 100644
index 0000000000000000000000000000000000000000..4090d908f02e9586734b0fce416876d7f58b7430
--- /dev/null
+++ b/vlm/dev/GkDbQb6qu_r/1.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:87788fa87fad26c6a3591e84e535a70e4dbb091aaa7dae3363b62e6f0f430765
+size 565421
diff --git a/vlm/dev/GkDbQb6qu_r/11.png b/vlm/dev/GkDbQb6qu_r/11.png
new file mode 100644
index 0000000000000000000000000000000000000000..c9ed42ea04be5dce10da8f4f7a2b635dd72c20dc
--- /dev/null
+++ b/vlm/dev/GkDbQb6qu_r/11.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fd00ba3098550d461883315f9fbc0a0980632ae4088902356247a3b8416c9a2f
+size 524708
diff --git a/vlm/dev/GkDbQb6qu_r/12.png b/vlm/dev/GkDbQb6qu_r/12.png
new file mode 100644
index 0000000000000000000000000000000000000000..9125b05950ef44d4299e5346e673bd65a1a3f4d0
--- /dev/null
+++ b/vlm/dev/GkDbQb6qu_r/12.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2c62f7dc62cdf48edfbfbe8bddfa2465654470d1f33d927898d5181bc0f4c932
+size 317659
diff --git a/vlm/dev/GkDbQb6qu_r/2.png b/vlm/dev/GkDbQb6qu_r/2.png
new file mode 100644
index 0000000000000000000000000000000000000000..a36fe2d3e8761336b10d4ac2f6f5277874810b3f
--- /dev/null
+++ b/vlm/dev/GkDbQb6qu_r/2.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3bf85b1fb26065cdc24a13b0d3bbc08a01528168b7b51394a4ef0ed65b2fdf3c
+size 540006
diff --git a/vlm/dev/GkDbQb6qu_r/4.png b/vlm/dev/GkDbQb6qu_r/4.png
new file mode 100644
index 0000000000000000000000000000000000000000..7298596b7e1dffb58a91d10179f86a1c34492248
--- /dev/null
+++ b/vlm/dev/GkDbQb6qu_r/4.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ddb3f98abc5447468ef74c0824c2b1e8224a04305aef0c8a35cc13efd9dd3d43
+size 564626
diff --git a/vlm/dev/GkDbQb6qu_r/5.png b/vlm/dev/GkDbQb6qu_r/5.png
new file mode 100644
index 0000000000000000000000000000000000000000..8c8f5756e3116a2a1103409f4f17f56685117c7f
--- /dev/null
+++ b/vlm/dev/GkDbQb6qu_r/5.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1c763e22a33d9cbadd41993fc0d8c5df60e7dd4063b31f446f8341dc071ea9d1
+size 716633
diff --git a/vlm/dev/GkDbQb6qu_r/7.png b/vlm/dev/GkDbQb6qu_r/7.png
new file mode 100644
index 0000000000000000000000000000000000000000..c1794b5bbc7bf58d2106fb4a4cdef1ad2bfef543
--- /dev/null
+++ b/vlm/dev/GkDbQb6qu_r/7.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cec51d9394708d21d3f6b28cffdc7460cc51d22f5f9eaf4d0001a3c7caf337fa
+size 467888
diff --git a/vlm/dev/GkDbQb6qu_r/8.png b/vlm/dev/GkDbQb6qu_r/8.png
new file mode 100644
index 0000000000000000000000000000000000000000..f9c09065c3b02cc14f0068b66c78feebac73001d
--- /dev/null
+++ b/vlm/dev/GkDbQb6qu_r/8.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fbca467923cce1d564c41e52a967a2e0254fdd054f9fef14b1e62e3eec4f5c8b
+size 564324
diff --git a/vlm/dev/JYtwGwIL7ye/0.png b/vlm/dev/JYtwGwIL7ye/0.png
new file mode 100644
index 0000000000000000000000000000000000000000..cf22436c15960bbd656d1acdc8d3bfcab7245ad9
--- /dev/null
+++ b/vlm/dev/JYtwGwIL7ye/0.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:00542846ae2e9e8a44abe831cdc9f0f50a57e312f0564c4b8988952ba65f92ad
+size 509193
diff --git a/vlm/dev/JYtwGwIL7ye/1.png b/vlm/dev/JYtwGwIL7ye/1.png
new file mode 100644
index 0000000000000000000000000000000000000000..51e9f96c5fb91637bc01cadafa46dc4160c7d34b
--- /dev/null
+++ b/vlm/dev/JYtwGwIL7ye/1.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:569a57877928bcc129bce7348bdec4e4d9c527812f826cdb7067103ce2875539
+size 587389
diff --git a/vlm/dev/JYtwGwIL7ye/10.png b/vlm/dev/JYtwGwIL7ye/10.png
new file mode 100644
index 0000000000000000000000000000000000000000..10513f28384aeae6f97d75e583c3378cadfe4e6c
--- /dev/null
+++ b/vlm/dev/JYtwGwIL7ye/10.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:166720bf5656e83ae37dbd0fb9c09ceba12455480a4cfe1aead6e9a4803a23d7
+size 529037
diff --git a/vlm/dev/JYtwGwIL7ye/11.png b/vlm/dev/JYtwGwIL7ye/11.png
new file mode 100644
index 0000000000000000000000000000000000000000..cf1a131ae10bda66f4c18f1c6804f748cde5603a
--- /dev/null
+++ b/vlm/dev/JYtwGwIL7ye/11.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cdd0cfece4a3dd3d4e9368c19e23c632f294b892c0dc7d36a8e8819931dfbae4
+size 535085
diff --git a/vlm/dev/JYtwGwIL7ye/12.png b/vlm/dev/JYtwGwIL7ye/12.png
new file mode 100644
index 0000000000000000000000000000000000000000..61cf9d94150c0c9df5da90b10e2b0208764e649f
--- /dev/null
+++ b/vlm/dev/JYtwGwIL7ye/12.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8e654dbfc059c7877d316a5fdc4ec9877e0c6eebfb4fc3774246bbda1d121029
+size 501948
diff --git a/vlm/dev/JYtwGwIL7ye/13.png b/vlm/dev/JYtwGwIL7ye/13.png
new file mode 100644
index 0000000000000000000000000000000000000000..c0a8811ff14fb306f1f9360011a16c90e8050352
--- /dev/null
+++ b/vlm/dev/JYtwGwIL7ye/13.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4fc89c3f00d401038752196ba4f71b949345e52383a6607785837a204f3a18c0
+size 343844
diff --git a/vlm/dev/JYtwGwIL7ye/14.png b/vlm/dev/JYtwGwIL7ye/14.png
new file mode 100644
index 0000000000000000000000000000000000000000..66b8f9a86c97ac80838d6983413f91eb357c94a2
--- /dev/null
+++ b/vlm/dev/JYtwGwIL7ye/14.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fe860a7e9cb2d5f740bf41726905f34745d4d0b1acb8df6faf728dd1403b9746
+size 395897
diff --git a/vlm/dev/JYtwGwIL7ye/15.png b/vlm/dev/JYtwGwIL7ye/15.png
new file mode 100644
index 0000000000000000000000000000000000000000..a27d6558ad4562691d0f2898f0a25216f6c9eb6c
--- /dev/null
+++ b/vlm/dev/JYtwGwIL7ye/15.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:132a4f79092c4a12c6d377c82add7c03e3e71fadc4bb1950e6ec0f27d7a8a3a5
+size 287475
diff --git a/vlm/dev/JYtwGwIL7ye/16.png b/vlm/dev/JYtwGwIL7ye/16.png
new file mode 100644
index 0000000000000000000000000000000000000000..a65490b2c433d9c0f39a5bd17f1a1007e9436a0a
--- /dev/null
+++ b/vlm/dev/JYtwGwIL7ye/16.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:45aeaf1c42e81847c67b1043ecd57e1091bad8c1ea08a99ac503286269cbc643
+size 212327
diff --git a/vlm/dev/JYtwGwIL7ye/17.png b/vlm/dev/JYtwGwIL7ye/17.png
new file mode 100644
index 0000000000000000000000000000000000000000..51728b6f784e1a3f8f1f05d3a95c72066e4267ed
--- /dev/null
+++ b/vlm/dev/JYtwGwIL7ye/17.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:28391ea63788df9f86024bfbc1fa88859080ed5e30dd467e0b0539d3f10cd1ec
+size 218246
diff --git a/vlm/dev/JYtwGwIL7ye/18.png b/vlm/dev/JYtwGwIL7ye/18.png
new file mode 100644
index 0000000000000000000000000000000000000000..0c281685258e1317f953c65e537960bbb06d880c
--- /dev/null
+++ b/vlm/dev/JYtwGwIL7ye/18.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:01a1c455339e803eb9a4e4b6783c2326d60107a4d8d215a75ba453ee4675222e
+size 115060
diff --git a/vlm/dev/JYtwGwIL7ye/2.png b/vlm/dev/JYtwGwIL7ye/2.png
new file mode 100644
index 0000000000000000000000000000000000000000..beb7a6353fd0fb2e2691aa176a3280c441d2c0d0
--- /dev/null
+++ b/vlm/dev/JYtwGwIL7ye/2.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ac4e349d71c3f98e35faeea07da766cc167dec8ffaaf2d20fc151bb9e3b221b0
+size 526716
diff --git a/vlm/dev/JYtwGwIL7ye/3.png b/vlm/dev/JYtwGwIL7ye/3.png
new file mode 100644
index 0000000000000000000000000000000000000000..42801ae95550b21e4aa5bad7524bd908201aeb64
--- /dev/null
+++ b/vlm/dev/JYtwGwIL7ye/3.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d4b65ffae3e32347340727abf92508ac005ec6108e0d7a4f81bdc91679a6ae04
+size 495482
diff --git a/vlm/dev/JYtwGwIL7ye/4.png b/vlm/dev/JYtwGwIL7ye/4.png
new file mode 100644
index 0000000000000000000000000000000000000000..e523c3110833d04146ed317f947aa0c9bdaf4b7c
--- /dev/null
+++ b/vlm/dev/JYtwGwIL7ye/4.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:31856d184e275feb11db8bd853ee43b2e45e72885fa5bc1c36d313635632efa2
+size 508108
diff --git a/vlm/dev/JYtwGwIL7ye/5.png b/vlm/dev/JYtwGwIL7ye/5.png
new file mode 100644
index 0000000000000000000000000000000000000000..61457c01dafe1b0cfbb2812aa04a4a1bdc1c821f
--- /dev/null
+++ b/vlm/dev/JYtwGwIL7ye/5.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ad589961574bae5956484b2ea4eaa3fb65b523cbb07cf674d02b84dd771b0555
+size 472576
diff --git a/vlm/dev/JYtwGwIL7ye/6.png b/vlm/dev/JYtwGwIL7ye/6.png
new file mode 100644
index 0000000000000000000000000000000000000000..1138c25d951d37efbcdc316d118ef59575bd1a6f
--- /dev/null
+++ b/vlm/dev/JYtwGwIL7ye/6.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5b7f505e154776eef157dd7e3232ee9903f2651c43819ec847e2b3c84a2c2d89
+size 510286
diff --git a/vlm/dev/JYtwGwIL7ye/7.png b/vlm/dev/JYtwGwIL7ye/7.png
new file mode 100644
index 0000000000000000000000000000000000000000..ea28b87ece72af42fec7631d96c119ac5e50925b
--- /dev/null
+++ b/vlm/dev/JYtwGwIL7ye/7.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f96c6b4e559ebbf11c1b3676175d8979a1272ea2d33b7443f337b5accbdb9a59
+size 465043
diff --git a/vlm/dev/JYtwGwIL7ye/8.png b/vlm/dev/JYtwGwIL7ye/8.png
new file mode 100644
index 0000000000000000000000000000000000000000..a7c8d106eb2033105fe42f3e249b2f714ba7039a
--- /dev/null
+++ b/vlm/dev/JYtwGwIL7ye/8.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d1e9dc5d84c3dd6c76e82d2e8c0f62f8861437e7ddc29a4e3a16db7c30601147
+size 504753
diff --git a/vlm/dev/JYtwGwIL7ye/9.png b/vlm/dev/JYtwGwIL7ye/9.png
new file mode 100644
index 0000000000000000000000000000000000000000..4aeec2cc22450ff499efe2ae13401917726fad65
--- /dev/null
+++ b/vlm/dev/JYtwGwIL7ye/9.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5b9ded3695408738f6324b28c85de6afb9d77df0adff936c1be5b7564870d977
+size 68139
diff --git a/vlm/dev/LKEYuYNOqx/10.png b/vlm/dev/LKEYuYNOqx/10.png
new file mode 100644
index 0000000000000000000000000000000000000000..8d1d0c9606ff6956f285ea32aa1de38e7c7bb612
--- /dev/null
+++ b/vlm/dev/LKEYuYNOqx/10.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:37b211e7012ff07a4bf44c03c791a3190ae0b60a54c8ed58144d76d517848a66
+size 561562
diff --git a/vlm/dev/LKEYuYNOqx/2.png b/vlm/dev/LKEYuYNOqx/2.png
new file mode 100644
index 0000000000000000000000000000000000000000..0118aee275b95967ca4b24a3e2648391f69cb624
--- /dev/null
+++ b/vlm/dev/LKEYuYNOqx/2.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:69a4eb3e3c2a2ee569d4f4860367f52aa757d29b4f48480b9a1c1d299f8e3b06
+size 518842
diff --git a/vlm/dev/LKEYuYNOqx/3.png b/vlm/dev/LKEYuYNOqx/3.png
new file mode 100644
index 0000000000000000000000000000000000000000..e44ba75ef01b4a069f3e3db8cd36c6941f5b824a
--- /dev/null
+++ b/vlm/dev/LKEYuYNOqx/3.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:660d3f969d8bfa6066b1a62eb515825fc683773be24bc441dce8e646e04b925d
+size 672402
diff --git a/vlm/dev/LKEYuYNOqx/6.png b/vlm/dev/LKEYuYNOqx/6.png
new file mode 100644
index 0000000000000000000000000000000000000000..aac8963247966eb51408a250324d267eb2a6b191
--- /dev/null
+++ b/vlm/dev/LKEYuYNOqx/6.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:216f51fafff3672dbf03362f18b6327c514465f07d54f444ebd24814afef1c9a
+size 674569
diff --git a/vlm/dev/LKEYuYNOqx/9.png b/vlm/dev/LKEYuYNOqx/9.png
new file mode 100644
index 0000000000000000000000000000000000000000..6bbf8cec18c655020922c6964897144d4f8f9999
--- /dev/null
+++ b/vlm/dev/LKEYuYNOqx/9.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:77df6980eeeae05cea95f7dad9ed0d02c5950f5526a087f8f5fabbe273e8972c
+size 1332348
diff --git a/vlm/dev/Mspk_WYKoEH/0.png b/vlm/dev/Mspk_WYKoEH/0.png
new file mode 100644
index 0000000000000000000000000000000000000000..2dadc19c838a71589a5127cc7222102418e60843
--- /dev/null
+++ b/vlm/dev/Mspk_WYKoEH/0.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d3d2d8370f82f9ae5d6998f26a2fd8fae95608a93ee7ef91120c07fc37f124fd
+size 562920
diff --git a/vlm/dev/Mspk_WYKoEH/1.png b/vlm/dev/Mspk_WYKoEH/1.png
new file mode 100644
index 0000000000000000000000000000000000000000..7ed183a34a27bb82b98a208d034af8d7754c3a96
--- /dev/null
+++ b/vlm/dev/Mspk_WYKoEH/1.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cb858b11ccf1ae40531afcc29f23f0f9e033e0c01db371ca760a5767744182aa
+size 720103
diff --git a/vlm/dev/Mspk_WYKoEH/10.png b/vlm/dev/Mspk_WYKoEH/10.png
new file mode 100644
index 0000000000000000000000000000000000000000..80fb539db3a5dcf02e3d1887c0ef8e8fd42e0d2b
--- /dev/null
+++ b/vlm/dev/Mspk_WYKoEH/10.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7d3bd0b59ef787aec228417b980c46b5ad82acd11e3ccd804e70c53a9abdb0ec
+size 564905
diff --git a/vlm/dev/Mspk_WYKoEH/11.png b/vlm/dev/Mspk_WYKoEH/11.png
new file mode 100644
index 0000000000000000000000000000000000000000..548e04aea1d7cd0e914b0ff26f3f2e4fd75f34cc
--- /dev/null
+++ b/vlm/dev/Mspk_WYKoEH/11.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:af9264e65f09e427532b4b91dbc8892b9485838bff5007579dae7a4960184a3c
+size 551794
diff --git a/vlm/dev/Mspk_WYKoEH/12.png b/vlm/dev/Mspk_WYKoEH/12.png
new file mode 100644
index 0000000000000000000000000000000000000000..f13e6ca9cb4ea5c98d767294df5e5062f3910a21
--- /dev/null
+++ b/vlm/dev/Mspk_WYKoEH/12.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:67fe332e4626794e20efa444a7cdb1b0f98f3cdf0a9b47c11a47257947252b04
+size 337729
diff --git a/vlm/dev/Mspk_WYKoEH/13.png b/vlm/dev/Mspk_WYKoEH/13.png
new file mode 100644
index 0000000000000000000000000000000000000000..9ce8a31474e7f605cc7e1a0d8d65491c7c8eb946
--- /dev/null
+++ b/vlm/dev/Mspk_WYKoEH/13.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fb09df98287c693c686165e01a0694981ab3b292cd4634208aff05b2296ff666
+size 648516
diff --git a/vlm/dev/Mspk_WYKoEH/14.png b/vlm/dev/Mspk_WYKoEH/14.png
new file mode 100644
index 0000000000000000000000000000000000000000..2ff193778d28815523c6e7fc2c94ce8e1b293de0
--- /dev/null
+++ b/vlm/dev/Mspk_WYKoEH/14.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:25f9f876b74f400570991a6c9e1004bcfd41f2c3b6463300a7d1ebe87ba5860c
+size 584407
diff --git a/vlm/dev/Mspk_WYKoEH/15.png b/vlm/dev/Mspk_WYKoEH/15.png
new file mode 100644
index 0000000000000000000000000000000000000000..d7729d8f57fb532211b8412e8e1fd2089849695f
--- /dev/null
+++ b/vlm/dev/Mspk_WYKoEH/15.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4418525482cab9caf4651e5753c80431df8e5c4368fb51719a91cabf537ee7fa
+size 639786
diff --git a/vlm/dev/Mspk_WYKoEH/16.png b/vlm/dev/Mspk_WYKoEH/16.png
new file mode 100644
index 0000000000000000000000000000000000000000..6cde4a1a1718b446fd767c5ef7284ec1f464936a
--- /dev/null
+++ b/vlm/dev/Mspk_WYKoEH/16.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:91c5cbe304a7c73ab15fbfc4f937c1285569deeee6a0e58758fd210f387d84be
+size 483255
diff --git a/vlm/dev/Mspk_WYKoEH/17.png b/vlm/dev/Mspk_WYKoEH/17.png
new file mode 100644
index 0000000000000000000000000000000000000000..ca1f704735982a696c5cd7bff3b08017abc7b130
--- /dev/null
+++ b/vlm/dev/Mspk_WYKoEH/17.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7a6e0a01550dd4c7cf923c72412a04c5949ce154e1feac213b84f05f7efd8ad5
+size 696030
diff --git a/vlm/dev/Mspk_WYKoEH/18.png b/vlm/dev/Mspk_WYKoEH/18.png
new file mode 100644
index 0000000000000000000000000000000000000000..c47b91f3b2bea0936e8da86660a82f7a2d89f851
--- /dev/null
+++ b/vlm/dev/Mspk_WYKoEH/18.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1f2283fecfb0981b745a606afd5d43ec8a7a7d82c3feb4aeb6e51826b7a64356
+size 585004
diff --git a/vlm/dev/Mspk_WYKoEH/19.png b/vlm/dev/Mspk_WYKoEH/19.png
new file mode 100644
index 0000000000000000000000000000000000000000..ee46ee0375ef059a7a4d371f90a46fdca222f02f
--- /dev/null
+++ b/vlm/dev/Mspk_WYKoEH/19.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bda94ec656bd9b9a941f10a2ecd963900d216655738a3087d6e4894d6d7feb31
+size 600562
diff --git a/vlm/dev/Mspk_WYKoEH/2.png b/vlm/dev/Mspk_WYKoEH/2.png
new file mode 100644
index 0000000000000000000000000000000000000000..f33d7b85d6d3ea65a471ec6231197b41fe20f275
--- /dev/null
+++ b/vlm/dev/Mspk_WYKoEH/2.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f588314d15144825e1a6765ff065cb758d369a68dede7ddafb94c441d3ab3257
+size 681728
diff --git a/vlm/dev/Mspk_WYKoEH/20.png b/vlm/dev/Mspk_WYKoEH/20.png
new file mode 100644
index 0000000000000000000000000000000000000000..54565d69ab1265da8c7a757fddf674dba638fd4e
--- /dev/null
+++ b/vlm/dev/Mspk_WYKoEH/20.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e389c83a25701ce6b06e13c4148d5712ff35ec96522f9abea0e60f520880891f
+size 307359
diff --git a/vlm/dev/Mspk_WYKoEH/21.png b/vlm/dev/Mspk_WYKoEH/21.png
new file mode 100644
index 0000000000000000000000000000000000000000..d84822202d791594a9857f85243d4361ede18c57
--- /dev/null
+++ b/vlm/dev/Mspk_WYKoEH/21.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0cc49af8f279b71129ce3e22469984b31880bfc0289b66b18b51a63766f7e011
+size 310780
diff --git a/vlm/dev/Mspk_WYKoEH/3.png b/vlm/dev/Mspk_WYKoEH/3.png
new file mode 100644
index 0000000000000000000000000000000000000000..482cd713afbdfbf32cbad15af369c4cd1a8fd5ae
--- /dev/null
+++ b/vlm/dev/Mspk_WYKoEH/3.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:48453eb42cd904797737f3f2abfcd3a51cfe3017175cdf37e7c68202e4c57b30
+size 643364
diff --git a/vlm/dev/Mspk_WYKoEH/4.png b/vlm/dev/Mspk_WYKoEH/4.png
new file mode 100644
index 0000000000000000000000000000000000000000..722b83b47ff77ccf206b196164e03cf12a38e3ba
--- /dev/null
+++ b/vlm/dev/Mspk_WYKoEH/4.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:48ea3e5c4a050fa2419e3b254ae6640e73efa5cdb230504c64c07deecb5fbbe1
+size 622118
diff --git a/vlm/dev/Mspk_WYKoEH/5.png b/vlm/dev/Mspk_WYKoEH/5.png
new file mode 100644
index 0000000000000000000000000000000000000000..9aa43af96ed56428b143bf85630cc74999192966
--- /dev/null
+++ b/vlm/dev/Mspk_WYKoEH/5.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:86bd00cfbf92b04c431a7d7d0237878ad91d4d8c225290945ddbc91f497de0e4
+size 656519
diff --git a/vlm/dev/Mspk_WYKoEH/6.png b/vlm/dev/Mspk_WYKoEH/6.png
new file mode 100644
index 0000000000000000000000000000000000000000..d77bd2a28642ba0a9c737f6a8c4efe1920c6cafc
--- /dev/null
+++ b/vlm/dev/Mspk_WYKoEH/6.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5bdadb07db85ef41ea932def271db027e4faa3b8fb641a9b1d92d5b85af5a47e
+size 633962
diff --git a/vlm/dev/Mspk_WYKoEH/7.png b/vlm/dev/Mspk_WYKoEH/7.png
new file mode 100644
index 0000000000000000000000000000000000000000..eed9fc4654b7c1b07029014ae5e44a7152c9b5a0
--- /dev/null
+++ b/vlm/dev/Mspk_WYKoEH/7.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5da5528aec9898265803b81c04e1d3674bbe4a02c657b85e1f8eb73604f82c59
+size 663585
diff --git a/vlm/dev/Mspk_WYKoEH/8.png b/vlm/dev/Mspk_WYKoEH/8.png
new file mode 100644
index 0000000000000000000000000000000000000000..32e91c32b1b313da318695b0124b63489d894351
--- /dev/null
+++ b/vlm/dev/Mspk_WYKoEH/8.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b3e9eaeb7f45f4932b9b82adcc6a80d25bb94d3e231d32d1a7c73898eebcdf31
+size 599786
diff --git a/vlm/dev/Mspk_WYKoEH/9.png b/vlm/dev/Mspk_WYKoEH/9.png
new file mode 100644
index 0000000000000000000000000000000000000000..550b4f5feb46726e4c0b35d96f59c4609fba9271
--- /dev/null
+++ b/vlm/dev/Mspk_WYKoEH/9.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:615efd51af77689b7013fe37bda086e3d45dec1742f8d7733822977ef2424898
+size 559078
diff --git a/vlm/dev/OboQ71j1Bn/13.png b/vlm/dev/OboQ71j1Bn/13.png
new file mode 100644
index 0000000000000000000000000000000000000000..e9795d40ca826e09e0a56f0204611381f39185ef
--- /dev/null
+++ b/vlm/dev/OboQ71j1Bn/13.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d31c1145a9620c31f4ef224ebf03273f092f1722a6798dcb597d606a067361e5
+size 1348558
diff --git a/vlm/dev/OboQ71j1Bn/16.png b/vlm/dev/OboQ71j1Bn/16.png
new file mode 100644
index 0000000000000000000000000000000000000000..f661843cc993eecae33a802d11f40a8549e51874
--- /dev/null
+++ b/vlm/dev/OboQ71j1Bn/16.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:31d4ce669d0958b0dff2443cb00da810b1cabfbcaf73ba678e5743db6443c072
+size 311573
diff --git a/vlm/dev/OboQ71j1Bn/19.png b/vlm/dev/OboQ71j1Bn/19.png
new file mode 100644
index 0000000000000000000000000000000000000000..c62e8d355155fe57da65f205416076c4ea32e92f
--- /dev/null
+++ b/vlm/dev/OboQ71j1Bn/19.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9ec1db08a7a30b84f6021a122fb95e6cfd508bda866329f36312dd98a70b7ab5
+size 518066
diff --git a/vlm/dev/TW7d65uYu5M/0.png b/vlm/dev/TW7d65uYu5M/0.png
new file mode 100644
index 0000000000000000000000000000000000000000..b800d59c542d42942aa300ac33da13079a88418b
--- /dev/null
+++ b/vlm/dev/TW7d65uYu5M/0.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:18dce1cf4546a019dee84fb6c0d84d85d93a273b3ef97ce6d394cc99ccc2743a
+size 520763
diff --git a/vlm/dev/TW7d65uYu5M/1.png b/vlm/dev/TW7d65uYu5M/1.png
new file mode 100644
index 0000000000000000000000000000000000000000..0ea8544fe932e1821707e60571bfef988564441e
--- /dev/null
+++ b/vlm/dev/TW7d65uYu5M/1.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7edeaded9a151ae888a8132b9f8797567a6da5644b0c76452596f1d1acec9b47
+size 854742
diff --git a/vlm/dev/TW7d65uYu5M/10.png b/vlm/dev/TW7d65uYu5M/10.png
new file mode 100644
index 0000000000000000000000000000000000000000..4fbcc4bf9e3af8eddc7fd8de4bc85caaa80e500b
--- /dev/null
+++ b/vlm/dev/TW7d65uYu5M/10.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1cdac084a26a0e3f7d4b696b3b92f847c30e389d7758b23892af24bb7065f1ff
+size 554552
diff --git a/vlm/dev/TW7d65uYu5M/11.png b/vlm/dev/TW7d65uYu5M/11.png
new file mode 100644
index 0000000000000000000000000000000000000000..28789a5192489b81a7cf7fd17cf7aaa903caa7d0
--- /dev/null
+++ b/vlm/dev/TW7d65uYu5M/11.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2d17651b6c15c0aa8e00c248d951fce317c4d18999f48d304a4dd531f7d907e8
+size 543834
diff --git a/vlm/dev/TW7d65uYu5M/12.png b/vlm/dev/TW7d65uYu5M/12.png
new file mode 100644
index 0000000000000000000000000000000000000000..0b1f046bc3796686a2877cba87b1433a819152fb
--- /dev/null
+++ b/vlm/dev/TW7d65uYu5M/12.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3f09c9f2f6b717ecdfa4789aa781b5301946ec64f2dbaded7804ebf66e178a00
+size 568367
diff --git a/vlm/dev/TW7d65uYu5M/13.png b/vlm/dev/TW7d65uYu5M/13.png
new file mode 100644
index 0000000000000000000000000000000000000000..8487b46c45afb295142700f58a2d202eebd988d6
--- /dev/null
+++ b/vlm/dev/TW7d65uYu5M/13.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:84df6576cce51975fb9dcd7c151383baf710a4d14ac3b9fcf65e715bae166be7
+size 178169
diff --git a/vlm/dev/TW7d65uYu5M/14.png b/vlm/dev/TW7d65uYu5M/14.png
new file mode 100644
index 0000000000000000000000000000000000000000..ee71d517541bdde5a670f6d067a9f509563d1b61
--- /dev/null
+++ b/vlm/dev/TW7d65uYu5M/14.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cd8d604a119bc7e8e00d6dc0fb7bb60452979cd4ceda6b78a28500c316fae9ac
+size 407160
diff --git a/vlm/dev/TW7d65uYu5M/15.png b/vlm/dev/TW7d65uYu5M/15.png
new file mode 100644
index 0000000000000000000000000000000000000000..334d18772350d40a6ee22610800d6378941c70de
--- /dev/null
+++ b/vlm/dev/TW7d65uYu5M/15.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5effc86aba3547cd0cbbc9cc8e44a502a8614f966f646a302c45f5192e3bb2ce
+size 228796
diff --git a/vlm/dev/TW7d65uYu5M/16.png b/vlm/dev/TW7d65uYu5M/16.png
new file mode 100644
index 0000000000000000000000000000000000000000..c0f4104e941e51b5209d4794f5ab2e82a19a538c
--- /dev/null
+++ b/vlm/dev/TW7d65uYu5M/16.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a537eb0f30662cf3c2c84e1748219b27261ec37fb3f93f7f5ee0ea46bf68e108
+size 2048575
diff --git a/vlm/dev/TW7d65uYu5M/17.png b/vlm/dev/TW7d65uYu5M/17.png
new file mode 100644
index 0000000000000000000000000000000000000000..19cea3cac9e6ca0e3a2275a603debe1d661fcf1c
--- /dev/null
+++ b/vlm/dev/TW7d65uYu5M/17.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f561ae5d11879c9ef3d09c0f5aa2f8a1f15d712032e09491341ec0232b533f4c
+size 2183394
diff --git a/vlm/dev/TW7d65uYu5M/18.png b/vlm/dev/TW7d65uYu5M/18.png
new file mode 100644
index 0000000000000000000000000000000000000000..6b219e9733e3d692fb939518c71c6af0d1aae7db
--- /dev/null
+++ b/vlm/dev/TW7d65uYu5M/18.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c589963e113b07d9f77594c9f205192317d448b5afb73dccb54ba8cc78c9fcd5
+size 2130886
diff --git a/vlm/dev/TW7d65uYu5M/19.png b/vlm/dev/TW7d65uYu5M/19.png
new file mode 100644
index 0000000000000000000000000000000000000000..f5df78e67dff39fe7a0655439a12a2e0a257fb1f
--- /dev/null
+++ b/vlm/dev/TW7d65uYu5M/19.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6c705116574f9951fc46d833fc6e9f9ff94a513eba15422052066440fc6fcdf8
+size 1962590
diff --git a/vlm/dev/TW7d65uYu5M/2.png b/vlm/dev/TW7d65uYu5M/2.png
new file mode 100644
index 0000000000000000000000000000000000000000..62d068793455d6cb017f1d9a9a67b21dac86f282
--- /dev/null
+++ b/vlm/dev/TW7d65uYu5M/2.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:95fe2e902f2a92f81db9010db57ed63f17d330d1a02f86a500ed9622143509f9
+size 631224
diff --git a/vlm/dev/TW7d65uYu5M/20.png b/vlm/dev/TW7d65uYu5M/20.png
new file mode 100644
index 0000000000000000000000000000000000000000..d6756993dc4fff80d616d9a9f1302974e570905a
--- /dev/null
+++ b/vlm/dev/TW7d65uYu5M/20.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b69caae24a68d38e64dd54404b557d019d0ad6747a8e9546ce77214fd431d5d6
+size 334896
diff --git a/vlm/dev/TW7d65uYu5M/3.png b/vlm/dev/TW7d65uYu5M/3.png
new file mode 100644
index 0000000000000000000000000000000000000000..5f4f4ec7614cabb80dcf4d4c271f7663b2c3bf1c
--- /dev/null
+++ b/vlm/dev/TW7d65uYu5M/3.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f1c8e0111eebcfbcd7462e29c948709b941234a5ea3086fb2d66bec8c7a14c2f
+size 519133
diff --git a/vlm/dev/TW7d65uYu5M/4.png b/vlm/dev/TW7d65uYu5M/4.png
new file mode 100644
index 0000000000000000000000000000000000000000..fa9db3ab53152b0354654258a3e817c23f8bd96c
--- /dev/null
+++ b/vlm/dev/TW7d65uYu5M/4.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6f9740621f2db08be76fb13e01122c884f8f3e1ab935a6dce9d833c077f66aa1
+size 507184
diff --git a/vlm/dev/TW7d65uYu5M/5.png b/vlm/dev/TW7d65uYu5M/5.png
new file mode 100644
index 0000000000000000000000000000000000000000..d8d57dfb4be7557af841f54714ccd1059d2997c6
--- /dev/null
+++ b/vlm/dev/TW7d65uYu5M/5.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0208aa0d35b83b217fea2f003c65f207a66c4dc8bea046d3d57674e368dfc88a
+size 647139
diff --git a/vlm/dev/TW7d65uYu5M/6.png b/vlm/dev/TW7d65uYu5M/6.png
new file mode 100644
index 0000000000000000000000000000000000000000..df72e693656a296c4b10bc4b56ed56c716c83cc1
--- /dev/null
+++ b/vlm/dev/TW7d65uYu5M/6.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:51adfd7e785e6fd430bcb9a869ac200a8535bb3195084af2d4285346d8d11d53
+size 618587
diff --git a/vlm/dev/TW7d65uYu5M/7.png b/vlm/dev/TW7d65uYu5M/7.png
new file mode 100644
index 0000000000000000000000000000000000000000..e2135376e0320c3c1272a2b1e1f16324e7663fb5
--- /dev/null
+++ b/vlm/dev/TW7d65uYu5M/7.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:23f1e58e10c71d36d3069ec3d99ccfb70d33b1d0d197d51bcadb5fa63092ef68
+size 636363
diff --git a/vlm/dev/TW7d65uYu5M/8.png b/vlm/dev/TW7d65uYu5M/8.png
new file mode 100644
index 0000000000000000000000000000000000000000..30e6806d5221f6ad61b3e3e5d639cb1220752c88
--- /dev/null
+++ b/vlm/dev/TW7d65uYu5M/8.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fdd9d954b2596d4609137623e5337b73d3340adb89fd7aa3a8a806c82abfdb25
+size 1245394
diff --git a/vlm/dev/TW7d65uYu5M/9.png b/vlm/dev/TW7d65uYu5M/9.png
new file mode 100644
index 0000000000000000000000000000000000000000..98b8f4a5f7ad2b1f5101f19a74a7e97fa58ba32b
--- /dev/null
+++ b/vlm/dev/TW7d65uYu5M/9.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1b2f61cab3bcf2fedc8411dbc8e95c937275e519d95bb7467a4fdf017d06ce52
+size 275699
diff --git a/vlm/dev/XomEU3eNeSQ/0.png b/vlm/dev/XomEU3eNeSQ/0.png
new file mode 100644
index 0000000000000000000000000000000000000000..f2409d3872c1ffb8e762650a7ca3f39c99d2306c
--- /dev/null
+++ b/vlm/dev/XomEU3eNeSQ/0.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:595285f37fb087ab1f2d48ebff5546e3e79063ffcd5bd73d7bfe8b7d21397640
+size 468808
diff --git a/vlm/dev/XomEU3eNeSQ/1.png b/vlm/dev/XomEU3eNeSQ/1.png
new file mode 100644
index 0000000000000000000000000000000000000000..6e0fad172409913406f8d98c241afc199cdec9de
--- /dev/null
+++ b/vlm/dev/XomEU3eNeSQ/1.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ec51844449318ce33ade1dda51474101aa8909d1d9d4498a02e132dd52dfde0a
+size 446832
diff --git a/vlm/dev/XomEU3eNeSQ/10.png b/vlm/dev/XomEU3eNeSQ/10.png
new file mode 100644
index 0000000000000000000000000000000000000000..26f139d97b16b7bf2fd474e8a88f202dd031b094
--- /dev/null
+++ b/vlm/dev/XomEU3eNeSQ/10.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:80a81c091e98a7f8d3313dd4bfaca57f3af1c4d7d9fdd1dfb190bc6e5bbe6a5e
+size 517940
diff --git a/vlm/dev/XomEU3eNeSQ/11.png b/vlm/dev/XomEU3eNeSQ/11.png
new file mode 100644
index 0000000000000000000000000000000000000000..dfb4f7d3824e5ec52c5b4d7bf8bceebae4e79f11
--- /dev/null
+++ b/vlm/dev/XomEU3eNeSQ/11.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:744fb2ebe5d10514aac605c9290d2c41670740dba6a4470bbccdd599bdcbd872
+size 294900
diff --git a/vlm/dev/XomEU3eNeSQ/12.png b/vlm/dev/XomEU3eNeSQ/12.png
new file mode 100644
index 0000000000000000000000000000000000000000..0d5693778b43e15e98b3e83afa830042835a318a
--- /dev/null
+++ b/vlm/dev/XomEU3eNeSQ/12.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b4530d8ffa943c96723e6b6de30775fffc5fc867b92dc5560a2b8b49e8104562
+size 256601
diff --git a/vlm/dev/XomEU3eNeSQ/13.png b/vlm/dev/XomEU3eNeSQ/13.png
new file mode 100644
index 0000000000000000000000000000000000000000..1f2e70c7a760707772978746699ad77d041cfa07
--- /dev/null
+++ b/vlm/dev/XomEU3eNeSQ/13.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bc6c74176a2e0ab54c003466c9189bb97e30759644e217926d5a0bfbd1e1905a
+size 457803
diff --git a/vlm/dev/XomEU3eNeSQ/14.png b/vlm/dev/XomEU3eNeSQ/14.png
new file mode 100644
index 0000000000000000000000000000000000000000..2d3a6c2fbebf046514c9b212e1f52c031ae584f6
--- /dev/null
+++ b/vlm/dev/XomEU3eNeSQ/14.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c7d227639eec3d4e52fba22f873465ecae3c409af63b2b1e9e8c30e222cd1f89
+size 277540
diff --git a/vlm/dev/XomEU3eNeSQ/15.png b/vlm/dev/XomEU3eNeSQ/15.png
new file mode 100644
index 0000000000000000000000000000000000000000..a41b4bb1a6815e3867b9b118eb85b04fa604b825
--- /dev/null
+++ b/vlm/dev/XomEU3eNeSQ/15.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c57a150b8278ffdd699777d87c9bece56545779b16d07977ecfbab04b15334ee
+size 243050
diff --git a/vlm/dev/XomEU3eNeSQ/16.png b/vlm/dev/XomEU3eNeSQ/16.png
new file mode 100644
index 0000000000000000000000000000000000000000..11650edf79533937daa80c2967353d31c77c6e45
--- /dev/null
+++ b/vlm/dev/XomEU3eNeSQ/16.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:647c0e44ed5e8d0b20a303a0896dbb7cb3e6562ea3bf5a32d0d64532a4c7005d
+size 136792
diff --git a/vlm/dev/XomEU3eNeSQ/17.png b/vlm/dev/XomEU3eNeSQ/17.png
new file mode 100644
index 0000000000000000000000000000000000000000..0bcb9320661e543b986607be5c7a89981123d9d9
--- /dev/null
+++ b/vlm/dev/XomEU3eNeSQ/17.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2b4801b5e0782c1424a1dd05f700da6a20169d0744260e7e98287053988cf9b6
+size 243087
diff --git a/vlm/dev/XomEU3eNeSQ/18.png b/vlm/dev/XomEU3eNeSQ/18.png
new file mode 100644
index 0000000000000000000000000000000000000000..f03872ddf171b822a9609252c7423ab545ae5e8d
--- /dev/null
+++ b/vlm/dev/XomEU3eNeSQ/18.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cd43e3a497af2c77c0cc9eeb0b459402d64a987555a194f2f6a7c369a8cf0877
+size 399838
diff --git a/vlm/dev/XomEU3eNeSQ/19.png b/vlm/dev/XomEU3eNeSQ/19.png
new file mode 100644
index 0000000000000000000000000000000000000000..6f212b05abfc554448408327bff77df1baf95b1b
--- /dev/null
+++ b/vlm/dev/XomEU3eNeSQ/19.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ba3339b3d34bc5d2db94f6d459a592c6c6cfb22b12faaa1db0c34558c7e6ccd3
+size 205118
diff --git a/vlm/dev/XomEU3eNeSQ/2.png b/vlm/dev/XomEU3eNeSQ/2.png
new file mode 100644
index 0000000000000000000000000000000000000000..2652a31b28b18dea31e77a6788d269000a5b7f60
--- /dev/null
+++ b/vlm/dev/XomEU3eNeSQ/2.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e1e37c17ecde78344cf90ec1a5d4705356415ce5a33c601163c7d89d5a7c21ae
+size 490184
diff --git a/vlm/dev/XomEU3eNeSQ/3.png b/vlm/dev/XomEU3eNeSQ/3.png
new file mode 100644
index 0000000000000000000000000000000000000000..ba666d2052d12f2eddc043d0b494c21f5ed4187c
--- /dev/null
+++ b/vlm/dev/XomEU3eNeSQ/3.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fe88899eb27e6634b533435052eb5e0b6c5ab7f126bcba9b188d778dcfffe98d
+size 450338
diff --git a/vlm/dev/XomEU3eNeSQ/4.png b/vlm/dev/XomEU3eNeSQ/4.png
new file mode 100644
index 0000000000000000000000000000000000000000..412ccd862b3cd2b38327bb54fcf216b10e96153b
--- /dev/null
+++ b/vlm/dev/XomEU3eNeSQ/4.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4efe351c0f063ade961c8fabbe4e6231067eddcb02ce31f2659d5b02874f3a51
+size 439885
diff --git a/vlm/dev/XomEU3eNeSQ/5.png b/vlm/dev/XomEU3eNeSQ/5.png
new file mode 100644
index 0000000000000000000000000000000000000000..1a758d9bfd7c9babacdcde26923296319f9db329
--- /dev/null
+++ b/vlm/dev/XomEU3eNeSQ/5.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5536cffdd17ab7444db7f371e16a8a1d1a7b76a5357199889c84d7a694984ac1
+size 493309
diff --git a/vlm/dev/XomEU3eNeSQ/6.png b/vlm/dev/XomEU3eNeSQ/6.png
new file mode 100644
index 0000000000000000000000000000000000000000..e183829fe82f39bb684b62d56aec8a9d40464b31
--- /dev/null
+++ b/vlm/dev/XomEU3eNeSQ/6.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:942a73fbd73626cce4b6b738d886f7c06822f3cfec6f0bd620548e309b639017
+size 531484
diff --git a/vlm/dev/XomEU3eNeSQ/7.png b/vlm/dev/XomEU3eNeSQ/7.png
new file mode 100644
index 0000000000000000000000000000000000000000..4268d5d2d17bf79c023364f104115d39e0bb1f58
--- /dev/null
+++ b/vlm/dev/XomEU3eNeSQ/7.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ea11a716846586cc0fe99b7e368fc73b9775bad900a70db43394295ff39d7564
+size 522771
diff --git a/vlm/dev/XomEU3eNeSQ/8.png b/vlm/dev/XomEU3eNeSQ/8.png
new file mode 100644
index 0000000000000000000000000000000000000000..2d116d611c4c2e2f3d8f360c22e93ccd5593098e
--- /dev/null
+++ b/vlm/dev/XomEU3eNeSQ/8.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8c7e41d5d508981b2da4918540a81565d4d85f8cc5348a4878133b74e0026249
+size 588919
diff --git a/vlm/dev/XomEU3eNeSQ/9.png b/vlm/dev/XomEU3eNeSQ/9.png
new file mode 100644
index 0000000000000000000000000000000000000000..63e109683e3e9452303f89411b10d776b4a6dba3
--- /dev/null
+++ b/vlm/dev/XomEU3eNeSQ/9.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2ba828acf8fb71b791abd4ab758c44c4f5f99af5a9994190578888ee3006ea8b
+size 544281
diff --git a/vlm/dev/YPpSngE-ZU/0.png b/vlm/dev/YPpSngE-ZU/0.png
new file mode 100644
index 0000000000000000000000000000000000000000..d3dbe379a3c1206a1b337f929082445f9d457cc4
--- /dev/null
+++ b/vlm/dev/YPpSngE-ZU/0.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0d1256c2f99e6a6a128312178762f8f610908286673422df02b8d3693704f58e
+size 377423
diff --git a/vlm/dev/YPpSngE-ZU/1.png b/vlm/dev/YPpSngE-ZU/1.png
new file mode 100644
index 0000000000000000000000000000000000000000..f3f4bd12a93a826f3a9d3b23ceb852db344ee410
--- /dev/null
+++ b/vlm/dev/YPpSngE-ZU/1.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5574a630651ca0d16d8370be772b30477f01f156a57838d958f298e5008418fd
+size 439998
diff --git a/vlm/dev/YPpSngE-ZU/10.png b/vlm/dev/YPpSngE-ZU/10.png
new file mode 100644
index 0000000000000000000000000000000000000000..faac1d8156a71cbc11bcefbb666b4d66cd314500
--- /dev/null
+++ b/vlm/dev/YPpSngE-ZU/10.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bcf7204eea00b41fa6554f8ad29d45735fd5a367eabcf1f674bbc796c8bc58bd
+size 568626
diff --git a/vlm/dev/YPpSngE-ZU/11.png b/vlm/dev/YPpSngE-ZU/11.png
new file mode 100644
index 0000000000000000000000000000000000000000..21febb3e132b9150c7be517a15f377298785cd06
--- /dev/null
+++ b/vlm/dev/YPpSngE-ZU/11.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4205a0d5864538abdd52deac9a2089febf849ce181de5ad4baba7c52e3f69498
+size 591619
diff --git a/vlm/dev/YPpSngE-ZU/12.png b/vlm/dev/YPpSngE-ZU/12.png
new file mode 100644
index 0000000000000000000000000000000000000000..feb7e9359251cf519af60090ffcfa65b5921491b
--- /dev/null
+++ b/vlm/dev/YPpSngE-ZU/12.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d73c9ab2571746561653e8e5b2689ce0be019661c05adc4af4ae4c714809b418
+size 574263
diff --git a/vlm/dev/YPpSngE-ZU/13.png b/vlm/dev/YPpSngE-ZU/13.png
new file mode 100644
index 0000000000000000000000000000000000000000..d94e38d6bd261662523952532e1e2c984aa3f3d0
--- /dev/null
+++ b/vlm/dev/YPpSngE-ZU/13.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:20dc0aef60fa0c9f69dd1229173e4799181c962f86b24ec8816047c685b9195d
+size 318744
diff --git a/vlm/dev/YPpSngE-ZU/2.png b/vlm/dev/YPpSngE-ZU/2.png
new file mode 100644
index 0000000000000000000000000000000000000000..c542f7618d6875c1ab2b4aa953be2e82b16d75e3
--- /dev/null
+++ b/vlm/dev/YPpSngE-ZU/2.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:96d226c2c4b071aadaaa284f98d2385b071a24d427a4ed547098aabf98c00a4b
+size 587463
diff --git a/vlm/dev/YPpSngE-ZU/3.png b/vlm/dev/YPpSngE-ZU/3.png
new file mode 100644
index 0000000000000000000000000000000000000000..798ac38ac1423d82e703288ffcff9867f2e95d0e
--- /dev/null
+++ b/vlm/dev/YPpSngE-ZU/3.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f588fbff40800bed2701a439f05eb7067ebbcaa3455b73d7899628483f764c3b
+size 508635
diff --git a/vlm/dev/YPpSngE-ZU/4.png b/vlm/dev/YPpSngE-ZU/4.png
new file mode 100644
index 0000000000000000000000000000000000000000..a97414c269c50a9fd994eff81ab5f46144c34958
--- /dev/null
+++ b/vlm/dev/YPpSngE-ZU/4.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cac61c769d278ac7c9e2a6f400c8adebe1fb88c7748e10a54ae0dde4f53e00ca
+size 469621
diff --git a/vlm/dev/YPpSngE-ZU/5.png b/vlm/dev/YPpSngE-ZU/5.png
new file mode 100644
index 0000000000000000000000000000000000000000..a421a740befbf6ca914f75453bd12395e257568d
--- /dev/null
+++ b/vlm/dev/YPpSngE-ZU/5.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:96b2241bbfb9a20f73523c8518433ef357faa7f674a0e2c5063c1c410073f037
+size 514143
diff --git a/vlm/dev/YPpSngE-ZU/6.png b/vlm/dev/YPpSngE-ZU/6.png
new file mode 100644
index 0000000000000000000000000000000000000000..2190603492e5bebccc68d6e81f3e3ad86780b77f
--- /dev/null
+++ b/vlm/dev/YPpSngE-ZU/6.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0d93cb94b9cd4a8428923839fa4b24e770a859b50cad0e77f64fabf0c94a1056
+size 538814
diff --git a/vlm/dev/YPpSngE-ZU/7.png b/vlm/dev/YPpSngE-ZU/7.png
new file mode 100644
index 0000000000000000000000000000000000000000..a4e05c2dfcb0c946310c2a85eade7df118102ca5
--- /dev/null
+++ b/vlm/dev/YPpSngE-ZU/7.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c34ba45cd3d574ab565daa6984ebe7c96501b0c89362f74660377d8777bbfd27
+size 540119
diff --git a/vlm/dev/YPpSngE-ZU/8.png b/vlm/dev/YPpSngE-ZU/8.png
new file mode 100644
index 0000000000000000000000000000000000000000..940532fe826c9e8c58fde5c8ba8bc62e51553678
--- /dev/null
+++ b/vlm/dev/YPpSngE-ZU/8.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9b93f7b43b226dff620f1eebdc2ee3bc5f4bc68791ada406abbe9f977ba37e33
+size 474414
diff --git a/vlm/dev/YPpSngE-ZU/9.png b/vlm/dev/YPpSngE-ZU/9.png
new file mode 100644
index 0000000000000000000000000000000000000000..1238aed5bbe97fa3a3bee945dd0f0398fc8278b1
--- /dev/null
+++ b/vlm/dev/YPpSngE-ZU/9.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2acf09064b699652139916b9be190d323d7d3ee504eb7dc2b42005cfc6c3b33c
+size 532379
diff --git a/vlm/dev/cBu4ElJfneV/0.png b/vlm/dev/cBu4ElJfneV/0.png
new file mode 100644
index 0000000000000000000000000000000000000000..caf350b1fa9356978a9a86bcd396f83814ad0067
--- /dev/null
+++ b/vlm/dev/cBu4ElJfneV/0.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:053019130c6a5dae8188e42757a9eb3555079a4218b2dd579e62601c58f487bd
+size 526852
diff --git a/vlm/dev/cBu4ElJfneV/1.png b/vlm/dev/cBu4ElJfneV/1.png
new file mode 100644
index 0000000000000000000000000000000000000000..52c1aa604ad78a2da37d26d6c01b9dba4d86ed7f
--- /dev/null
+++ b/vlm/dev/cBu4ElJfneV/1.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1c968fd0be84203d9fee2ca5bd6b3016309ff7202335b64f2b7827cf989f6578
+size 620598
diff --git a/vlm/dev/cBu4ElJfneV/10.png b/vlm/dev/cBu4ElJfneV/10.png
new file mode 100644
index 0000000000000000000000000000000000000000..b32dafea4a35ceb0f9642f6d300c07eff45614f3
--- /dev/null
+++ b/vlm/dev/cBu4ElJfneV/10.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:56f5dbf27dba2d83ff7ba33eb925e4e7f00e2b3c0996569edb00cc4bdbc22d1c
+size 552974
diff --git a/vlm/dev/cBu4ElJfneV/11.png b/vlm/dev/cBu4ElJfneV/11.png
new file mode 100644
index 0000000000000000000000000000000000000000..8c9de078f0e410d5daef65e361e52fba7efa6c84
--- /dev/null
+++ b/vlm/dev/cBu4ElJfneV/11.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:14bfd879f485f30adb2d752607b3986d1f58a2ced74c4e06fb5d94f486f91e1b
+size 124234
diff --git a/vlm/dev/cBu4ElJfneV/12.png b/vlm/dev/cBu4ElJfneV/12.png
new file mode 100644
index 0000000000000000000000000000000000000000..ba03bfdbf603327aa3c67a97f802ac1709298236
--- /dev/null
+++ b/vlm/dev/cBu4ElJfneV/12.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0d71028482ac232b2eedf8adca8fb1ed936297022d2e192957f0c1f8fd0aff46
+size 320237
diff --git a/vlm/dev/cBu4ElJfneV/13.png b/vlm/dev/cBu4ElJfneV/13.png
new file mode 100644
index 0000000000000000000000000000000000000000..6205c03f35531df4243c5c05c7d78c91a17cd1a3
--- /dev/null
+++ b/vlm/dev/cBu4ElJfneV/13.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7bb02395448a65c42508af3cf15ba0374702d63d37d1611714aa8e0967f11264
+size 315321
diff --git a/vlm/dev/cBu4ElJfneV/14.png b/vlm/dev/cBu4ElJfneV/14.png
new file mode 100644
index 0000000000000000000000000000000000000000..8bdd119b5defccbf000cfc4460b6759b84650dc9
--- /dev/null
+++ b/vlm/dev/cBu4ElJfneV/14.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:33da11d436f4d26d2f0917b8024e6c2d33907714e84852d2bed687d8642969c0
+size 336301
diff --git a/vlm/dev/cBu4ElJfneV/15.png b/vlm/dev/cBu4ElJfneV/15.png
new file mode 100644
index 0000000000000000000000000000000000000000..c861a189ec34c2f1b9c16b2dfd39ffeba2848e1e
--- /dev/null
+++ b/vlm/dev/cBu4ElJfneV/15.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d96978c62a4f629c454db3ca596f5e39c69d6e857a073a00922ec56a24bd01bd
+size 208859
diff --git a/vlm/dev/cBu4ElJfneV/16.png b/vlm/dev/cBu4ElJfneV/16.png
new file mode 100644
index 0000000000000000000000000000000000000000..f676478e8014f783e77e673b73a3ba62e8dc6559
--- /dev/null
+++ b/vlm/dev/cBu4ElJfneV/16.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6f6ae1c0aec0b92c9b1738322305f1239eecc55537a51d6d65569b4836801e7d
+size 1488192
diff --git a/vlm/dev/cBu4ElJfneV/2.png b/vlm/dev/cBu4ElJfneV/2.png
new file mode 100644
index 0000000000000000000000000000000000000000..0c62c0687c7a25cca8d1b9877920eb99326d7cd1
--- /dev/null
+++ b/vlm/dev/cBu4ElJfneV/2.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9c018f18f005f16eda76b4dd624b979852cd80b5533df82c0d6046f0d057b057
+size 553324
diff --git a/vlm/dev/cBu4ElJfneV/3.png b/vlm/dev/cBu4ElJfneV/3.png
new file mode 100644
index 0000000000000000000000000000000000000000..1f17752302d4a65fcdeaa8705909848b7e51f7a5
--- /dev/null
+++ b/vlm/dev/cBu4ElJfneV/3.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2092f050090c2081fd9322d6edf99a5b425f2c6083baf3370cb0648747ef1548
+size 510074
diff --git a/vlm/dev/cBu4ElJfneV/4.png b/vlm/dev/cBu4ElJfneV/4.png
new file mode 100644
index 0000000000000000000000000000000000000000..8b3833397230112363f130c3a0aa85824495715b
--- /dev/null
+++ b/vlm/dev/cBu4ElJfneV/4.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0b98092fef062cb8fa4b971b4acc5101404e6890c20631a36838edb3df10e225
+size 541749
diff --git a/vlm/dev/cBu4ElJfneV/5.png b/vlm/dev/cBu4ElJfneV/5.png
new file mode 100644
index 0000000000000000000000000000000000000000..dbe4c0fca71cc39ff6d6f7c6d618c12832e46997
--- /dev/null
+++ b/vlm/dev/cBu4ElJfneV/5.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cdc1754839e5e6167627b61733821d529b5d17470ee07843ad1a41086b6c039f
+size 572752
diff --git a/vlm/dev/cBu4ElJfneV/6.png b/vlm/dev/cBu4ElJfneV/6.png
new file mode 100644
index 0000000000000000000000000000000000000000..6aa3f0be61e8671ad9efaad1d1d75b329bb97dd1
--- /dev/null
+++ b/vlm/dev/cBu4ElJfneV/6.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:09edd460fc5286f04a1c9562174363af7c5ba1dc0d94085990250e396b0db996
+size 554937
diff --git a/vlm/dev/cBu4ElJfneV/7.png b/vlm/dev/cBu4ElJfneV/7.png
new file mode 100644
index 0000000000000000000000000000000000000000..ddb92a2f58e3eaf6659aeb70a420a8968b474c8c
--- /dev/null
+++ b/vlm/dev/cBu4ElJfneV/7.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e0ebfe4a37d08f620bfdff76f936dcdd50fd81f62392c8641130cc9bceacd017
+size 567944
diff --git a/vlm/dev/cBu4ElJfneV/8.png b/vlm/dev/cBu4ElJfneV/8.png
new file mode 100644
index 0000000000000000000000000000000000000000..5f906e8f4979bdee7063b01c1f362a017b04bf1e
--- /dev/null
+++ b/vlm/dev/cBu4ElJfneV/8.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:96c63b41a65fb4606059db3f94cf3021b764037170c667c2a008e47669663e57
+size 544388
diff --git a/vlm/dev/cBu4ElJfneV/9.png b/vlm/dev/cBu4ElJfneV/9.png
new file mode 100644
index 0000000000000000000000000000000000000000..0aabde727ba1fa386eb156417928b0546b684146
--- /dev/null
+++ b/vlm/dev/cBu4ElJfneV/9.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:dd628ed40e20d93548beb76294cc3cc9a93c841150a6daa1249fbc84968e03f8
+size 543590
diff --git a/vlm/dev/fT9W53lLxNS/0.png b/vlm/dev/fT9W53lLxNS/0.png
new file mode 100644
index 0000000000000000000000000000000000000000..218329a72a95b51eb341502b0547273d38056a91
--- /dev/null
+++ b/vlm/dev/fT9W53lLxNS/0.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8c92c30830580397f795a9da8f8320ddd530fef80f526fe9e08569984399a1de
+size 614683
diff --git a/vlm/dev/fT9W53lLxNS/1.png b/vlm/dev/fT9W53lLxNS/1.png
new file mode 100644
index 0000000000000000000000000000000000000000..e39057735c27238ec2a9152643b294bfc34daa34
--- /dev/null
+++ b/vlm/dev/fT9W53lLxNS/1.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:458fec3931da4b15ad57ac6d471f4a3fca504e0d161eb967bd97fbd5bbc825d6
+size 593163
diff --git a/vlm/dev/fT9W53lLxNS/10.png b/vlm/dev/fT9W53lLxNS/10.png
new file mode 100644
index 0000000000000000000000000000000000000000..c7c26068f6ab41d6ddbfa210f167f034077558d7
--- /dev/null
+++ b/vlm/dev/fT9W53lLxNS/10.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ddb16d79e148b2a31f308ea949458cdd3069bf2c12101f0cb8e603d4b664c688
+size 539321
diff --git a/vlm/dev/fT9W53lLxNS/11.png b/vlm/dev/fT9W53lLxNS/11.png
new file mode 100644
index 0000000000000000000000000000000000000000..e5f932ee88b4e7da1f19be6f081c65f631c31d55
--- /dev/null
+++ b/vlm/dev/fT9W53lLxNS/11.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b73ed39bc96580495d5c92b2be3909455d45e1f85f32c8e8bc95b1cceb459e86
+size 557196
diff --git a/vlm/dev/fT9W53lLxNS/12.png b/vlm/dev/fT9W53lLxNS/12.png
new file mode 100644
index 0000000000000000000000000000000000000000..a00338ac9051e3e8569d9f8ee4c90ddf0c697a03
--- /dev/null
+++ b/vlm/dev/fT9W53lLxNS/12.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:855d39ca072cac7741d3746342eda696210095c9f366c999f8ae5ba6a6f70b85
+size 530159
diff --git a/vlm/dev/fT9W53lLxNS/13.png b/vlm/dev/fT9W53lLxNS/13.png
new file mode 100644
index 0000000000000000000000000000000000000000..c217f6a4c98fda979f4d41a24844fc78baab7c56
--- /dev/null
+++ b/vlm/dev/fT9W53lLxNS/13.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:35f0bd59835bd7c1f31be0d8c319564645fc67adf9bcc8a36594e7be0744db72
+size 89911
diff --git a/vlm/dev/fT9W53lLxNS/14.png b/vlm/dev/fT9W53lLxNS/14.png
new file mode 100644
index 0000000000000000000000000000000000000000..09d916631ef089a222edf9c08b0376c4ee106630
--- /dev/null
+++ b/vlm/dev/fT9W53lLxNS/14.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:139a6c2444ae3b902854fc5fd68675fb80b680c2b6292f5742c5093736ca14f8
+size 379116
diff --git a/vlm/dev/fT9W53lLxNS/2.png b/vlm/dev/fT9W53lLxNS/2.png
new file mode 100644
index 0000000000000000000000000000000000000000..508735bdc702a8182acb36490c859d551f620d78
--- /dev/null
+++ b/vlm/dev/fT9W53lLxNS/2.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e197082893ae2924b8dc84c541a1cde0af5e3fa305187144692ee630c30e00db
+size 605643
diff --git a/vlm/dev/fT9W53lLxNS/3.png b/vlm/dev/fT9W53lLxNS/3.png
new file mode 100644
index 0000000000000000000000000000000000000000..b8e5534600028dce71b405bb63de820359700ebc
--- /dev/null
+++ b/vlm/dev/fT9W53lLxNS/3.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1f08b8cc84fc20cdce92c9c4c9c17b2699858611ce6833febf87a88c0327e7df
+size 663741
diff --git a/vlm/dev/fT9W53lLxNS/4.png b/vlm/dev/fT9W53lLxNS/4.png
new file mode 100644
index 0000000000000000000000000000000000000000..eef51d0ee53781f0c379efa31b89b0ee2a99d2dd
--- /dev/null
+++ b/vlm/dev/fT9W53lLxNS/4.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:22af6b83f09d0af93fa5e3bc532570f122bc793aac48971d8b20a960b0a2e21c
+size 775793
diff --git a/vlm/dev/fT9W53lLxNS/5.png b/vlm/dev/fT9W53lLxNS/5.png
new file mode 100644
index 0000000000000000000000000000000000000000..afd9c648edc8690df08c9876c688b6a14a1822b6
--- /dev/null
+++ b/vlm/dev/fT9W53lLxNS/5.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6fd161c444fdf6f6212f974483029df73389cdaebefa51e333b60559877e5eff
+size 611105
diff --git a/vlm/dev/fT9W53lLxNS/6.png b/vlm/dev/fT9W53lLxNS/6.png
new file mode 100644
index 0000000000000000000000000000000000000000..ea29314422c33391235bd1a27b36a0bcac7159fb
--- /dev/null
+++ b/vlm/dev/fT9W53lLxNS/6.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1f8fa837dabc02a725e5388625c035dca0ca5616f7989af7866ce719328a2e1c
+size 584948
diff --git a/vlm/dev/fT9W53lLxNS/7.png b/vlm/dev/fT9W53lLxNS/7.png
new file mode 100644
index 0000000000000000000000000000000000000000..ef7a965da236c4e49c4ada1b74ee34cc774da73b
--- /dev/null
+++ b/vlm/dev/fT9W53lLxNS/7.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d74461616209b1b11785c4ebc409bda43086c392f9719eafcf9de26203b542d1
+size 1026452
diff --git a/vlm/dev/fT9W53lLxNS/8.png b/vlm/dev/fT9W53lLxNS/8.png
new file mode 100644
index 0000000000000000000000000000000000000000..27d5dc60537dbebdf2cebd73839de6bc773e3e42
--- /dev/null
+++ b/vlm/dev/fT9W53lLxNS/8.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2dfb6e66db67470518ba412f61d01d05ad22c8a7b0c3eef17cc0377915430ae4
+size 916009
diff --git a/vlm/dev/fT9W53lLxNS/9.png b/vlm/dev/fT9W53lLxNS/9.png
new file mode 100644
index 0000000000000000000000000000000000000000..4a5c16c8c8bad5b3874cceb16c546fe9d9aff317
--- /dev/null
+++ b/vlm/dev/fT9W53lLxNS/9.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f4a516e8195907d94d818c7de02aad66f0c4583e7bedfef907dd89ce708f514c
+size 524165
diff --git a/vlm/dev/fwzUgo0FM9v/0.png b/vlm/dev/fwzUgo0FM9v/0.png
new file mode 100644
index 0000000000000000000000000000000000000000..8c2d04672cd13097000dd3c00656fd32986bf0eb
--- /dev/null
+++ b/vlm/dev/fwzUgo0FM9v/0.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:09a6e260e01744ba22aedecdf49ac89003f14874c95b4c921b38a9cbc63de6d7
+size 457716
diff --git a/vlm/dev/fwzUgo0FM9v/1.png b/vlm/dev/fwzUgo0FM9v/1.png
new file mode 100644
index 0000000000000000000000000000000000000000..e541bc5e544f494841e1363e7620e3aab65920e4
--- /dev/null
+++ b/vlm/dev/fwzUgo0FM9v/1.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d6c5a1cd6f9104ee4b7d50e078da3f4ba789b3101d9e81508d4e90fc77204c65
+size 597371
diff --git a/vlm/dev/fwzUgo0FM9v/10.png b/vlm/dev/fwzUgo0FM9v/10.png
new file mode 100644
index 0000000000000000000000000000000000000000..22b4e1f53d786104941d718eb825fb4b1ea78309
--- /dev/null
+++ b/vlm/dev/fwzUgo0FM9v/10.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:88b22daed2e66e8e61bed9c69fb3b0e29de99311971263b29ae1c1554ad8f445
+size 552372
diff --git a/vlm/dev/fwzUgo0FM9v/11.png b/vlm/dev/fwzUgo0FM9v/11.png
new file mode 100644
index 0000000000000000000000000000000000000000..4633d7c82e3389358ebf83f227068bfa49c1fcda
--- /dev/null
+++ b/vlm/dev/fwzUgo0FM9v/11.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2da0c14749d482ec72af97872c4da383e6983356617726792341b1cc2fe59439
+size 614012
diff --git a/vlm/dev/fwzUgo0FM9v/12.png b/vlm/dev/fwzUgo0FM9v/12.png
new file mode 100644
index 0000000000000000000000000000000000000000..1f51c7874236ac6024cec6d24036f7e3a4a13a13
--- /dev/null
+++ b/vlm/dev/fwzUgo0FM9v/12.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e52f4510d92930abc177a37288247db27bdeb36fed6e1adc6967454643ee52a8
+size 630838
diff --git a/vlm/dev/fwzUgo0FM9v/13.png b/vlm/dev/fwzUgo0FM9v/13.png
new file mode 100644
index 0000000000000000000000000000000000000000..d102d13183134abbc57c1203f17440f084e53f03
--- /dev/null
+++ b/vlm/dev/fwzUgo0FM9v/13.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e2aa756b378ca7193303d9b81190ced476e087bd168f6f8bf54cb4f7326384b5
+size 179662
diff --git a/vlm/dev/fwzUgo0FM9v/14.png b/vlm/dev/fwzUgo0FM9v/14.png
new file mode 100644
index 0000000000000000000000000000000000000000..c5bad91749f1c812fddab21a1be11dace0fcf674
--- /dev/null
+++ b/vlm/dev/fwzUgo0FM9v/14.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:edf64019c31099b5a3119ccf41ba8f19fb5095c0ff57736c573c24ae1f6d3a84
+size 402538
diff --git a/vlm/dev/fwzUgo0FM9v/15.png b/vlm/dev/fwzUgo0FM9v/15.png
new file mode 100644
index 0000000000000000000000000000000000000000..b7a7fa5d3372665e4b913ab3058e98513f58685d
--- /dev/null
+++ b/vlm/dev/fwzUgo0FM9v/15.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0924173bf7e8101d933e442e4b2d676545d4214ad391fa57b21f1165708a38a2
+size 496088
diff --git a/vlm/dev/fwzUgo0FM9v/16.png b/vlm/dev/fwzUgo0FM9v/16.png
new file mode 100644
index 0000000000000000000000000000000000000000..5b1d7029a7468c7ca0d0f255f5d30eda5ad7de37
--- /dev/null
+++ b/vlm/dev/fwzUgo0FM9v/16.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3e2e7ec1c81e649bbc3e54388d7b76c12a8abbf81a5fd4eb2e35a3d7b47ee80e
+size 464252
diff --git a/vlm/dev/fwzUgo0FM9v/17.png b/vlm/dev/fwzUgo0FM9v/17.png
new file mode 100644
index 0000000000000000000000000000000000000000..a1786d8585060371dff7a3175dbda8bdaa56d4d2
--- /dev/null
+++ b/vlm/dev/fwzUgo0FM9v/17.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:71cfababb5c04bce8b5025f1525e04ae848e8e05f9ae5a74fe1c0fe11ac3221a
+size 1324576
diff --git a/vlm/dev/fwzUgo0FM9v/18.png b/vlm/dev/fwzUgo0FM9v/18.png
new file mode 100644
index 0000000000000000000000000000000000000000..e522bad69cfbaa8dc7bd2f00e00b91560c9a3d44
--- /dev/null
+++ b/vlm/dev/fwzUgo0FM9v/18.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e378beea83e63d47bf77b1d62589e5d4e3c9781c1456563f8731863dc368b90e
+size 522722
diff --git a/vlm/dev/fwzUgo0FM9v/19.png b/vlm/dev/fwzUgo0FM9v/19.png
new file mode 100644
index 0000000000000000000000000000000000000000..fe8d660f1c302f8c64a01410eedee2f992431554
--- /dev/null
+++ b/vlm/dev/fwzUgo0FM9v/19.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f92bb963dcfc9b0babad5c11c2dded88d2528376e835494168f77d5cfb6112dd
+size 921269
diff --git a/vlm/dev/fwzUgo0FM9v/2.png b/vlm/dev/fwzUgo0FM9v/2.png
new file mode 100644
index 0000000000000000000000000000000000000000..7cc4030ff9beb3604a464f6a2f77243c16f83955
--- /dev/null
+++ b/vlm/dev/fwzUgo0FM9v/2.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:75829befc581c861d6783a69dc566c86c5a0e4a6bc8007de387493c5069afe86
+size 563257
diff --git a/vlm/dev/fwzUgo0FM9v/20.png b/vlm/dev/fwzUgo0FM9v/20.png
new file mode 100644
index 0000000000000000000000000000000000000000..09aa665f402cae3127390898c93bcda626bcd4bd
--- /dev/null
+++ b/vlm/dev/fwzUgo0FM9v/20.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d35622d299bdb7b7e604bcb93e89c8437dab4c02ea9923df8000e924adcb50d1
+size 1363368
diff --git a/vlm/dev/fwzUgo0FM9v/21.png b/vlm/dev/fwzUgo0FM9v/21.png
new file mode 100644
index 0000000000000000000000000000000000000000..4a37f6914bd4e1f789f025671180f1c479db50af
--- /dev/null
+++ b/vlm/dev/fwzUgo0FM9v/21.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fe7251ca3e999e6da8f56145a54907de03b83cb5f10aa2fd9e66c4171d677bd6
+size 745330
diff --git a/vlm/dev/fwzUgo0FM9v/22.png b/vlm/dev/fwzUgo0FM9v/22.png
new file mode 100644
index 0000000000000000000000000000000000000000..5d69103e187ab41cd0197ad1916727e4cdb3415e
--- /dev/null
+++ b/vlm/dev/fwzUgo0FM9v/22.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:727acfc5f969b10644040cd6609d2a1139713d9dfde6f8f77446f6881257ed0e
+size 996255
diff --git a/vlm/dev/fwzUgo0FM9v/23.png b/vlm/dev/fwzUgo0FM9v/23.png
new file mode 100644
index 0000000000000000000000000000000000000000..563ad032d0f501895e695dec644d751473445323
--- /dev/null
+++ b/vlm/dev/fwzUgo0FM9v/23.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:59162d8ed68a1d750879fb28c4467a39f5a86b1f35ad9943757f54a33363036b
+size 800531
diff --git a/vlm/dev/fwzUgo0FM9v/24.png b/vlm/dev/fwzUgo0FM9v/24.png
new file mode 100644
index 0000000000000000000000000000000000000000..dbb5f970dcc965d1741f5ed0af2e8f82924f29af
--- /dev/null
+++ b/vlm/dev/fwzUgo0FM9v/24.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1d1de0f285436e8fc1af5173cebdbaa9b96df1d48ab3a1cca7e44a12e58dfe24
+size 888446
diff --git a/vlm/dev/fwzUgo0FM9v/3.png b/vlm/dev/fwzUgo0FM9v/3.png
new file mode 100644
index 0000000000000000000000000000000000000000..6af553a755b11b46bf33d5484c74bc81207dd4dd
--- /dev/null
+++ b/vlm/dev/fwzUgo0FM9v/3.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:42c2a8c9f3a7adacfb98473e376e1b1eeec0d7b40c089384b057b54bd86ca26e
+size 457126
diff --git a/vlm/dev/fwzUgo0FM9v/4.png b/vlm/dev/fwzUgo0FM9v/4.png
new file mode 100644
index 0000000000000000000000000000000000000000..d3d1976d28cd304bbf56586b5c54947a57195d3a
--- /dev/null
+++ b/vlm/dev/fwzUgo0FM9v/4.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:297aa78003917232e954dee3090312ac54c462446abffd01c3c5cec55ecf44b7
+size 571206
diff --git a/vlm/dev/fwzUgo0FM9v/5.png b/vlm/dev/fwzUgo0FM9v/5.png
new file mode 100644
index 0000000000000000000000000000000000000000..e4d311525b1fe7edb4216de0839e303ca7514044
--- /dev/null
+++ b/vlm/dev/fwzUgo0FM9v/5.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:05f0f7701936b49dd7a234e729f89ef731538b907462c78faae6160c4bcba726
+size 638474
diff --git a/vlm/dev/fwzUgo0FM9v/6.png b/vlm/dev/fwzUgo0FM9v/6.png
new file mode 100644
index 0000000000000000000000000000000000000000..671d026e62541efd7b1df37d15488a1d23c3df6f
--- /dev/null
+++ b/vlm/dev/fwzUgo0FM9v/6.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6762855f15e91f13b914d343bb7ede185c7b91903844189973665965c82b09b6
+size 564561
diff --git a/vlm/dev/fwzUgo0FM9v/7.png b/vlm/dev/fwzUgo0FM9v/7.png
new file mode 100644
index 0000000000000000000000000000000000000000..d61d60a32b71ffde31c7a146eff872b71ae48b76
--- /dev/null
+++ b/vlm/dev/fwzUgo0FM9v/7.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:be05740373e713795f6a6764a6e7796a2066b5001357ea39cd80f619efb2e4be
+size 516787
diff --git a/vlm/dev/fwzUgo0FM9v/8.png b/vlm/dev/fwzUgo0FM9v/8.png
new file mode 100644
index 0000000000000000000000000000000000000000..ec5c53f9ab95961507be838a3b4451f8db618324
--- /dev/null
+++ b/vlm/dev/fwzUgo0FM9v/8.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7aaed6ee559b31e56bdaaacdaccee26baf29c7cd21bd6cc603cea63896a15e9c
+size 550479
diff --git a/vlm/dev/fwzUgo0FM9v/9.png b/vlm/dev/fwzUgo0FM9v/9.png
new file mode 100644
index 0000000000000000000000000000000000000000..7092dfb573a622d456d7815bb7d7cde20b0aae83
--- /dev/null
+++ b/vlm/dev/fwzUgo0FM9v/9.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5e979a5358c31ac0efbbb62e493789e0af611e8214a94b38acaf4bc62356082f
+size 205362
diff --git a/vlm/dev/g6We1SwaY9/0.png b/vlm/dev/g6We1SwaY9/0.png
new file mode 100644
index 0000000000000000000000000000000000000000..ab33a1c136064fe44fc5918750f39387e455e101
--- /dev/null
+++ b/vlm/dev/g6We1SwaY9/0.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:57ad9231784b060ad43b355ae86e6910b5d39b1f5b39462cee7afdd25699fb39
+size 958358
diff --git a/vlm/dev/g6We1SwaY9/1.png b/vlm/dev/g6We1SwaY9/1.png
new file mode 100644
index 0000000000000000000000000000000000000000..eeb94111415a7e5c0f16d9defac78837366d5735
--- /dev/null
+++ b/vlm/dev/g6We1SwaY9/1.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d8ad4bf3f8b41feb2d089d2b0ec5c9b71efb5c50e00fc364e22947a74d21e5a5
+size 561793
diff --git a/vlm/dev/g6We1SwaY9/10.png b/vlm/dev/g6We1SwaY9/10.png
new file mode 100644
index 0000000000000000000000000000000000000000..e2f6f22cc19c184afdcde5fc3b7ca90883548fb6
--- /dev/null
+++ b/vlm/dev/g6We1SwaY9/10.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ddb150078436bc8c2e7363db41a102b9dd349382e4ffe39968f043592a623d9c
+size 555568
diff --git a/vlm/dev/g6We1SwaY9/11.png b/vlm/dev/g6We1SwaY9/11.png
new file mode 100644
index 0000000000000000000000000000000000000000..b299aaef126e823744bcd1fde5ebe279adb47671
--- /dev/null
+++ b/vlm/dev/g6We1SwaY9/11.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4547a0130646414cee593bfc0ba6d91af8b86c6857fce8e2aea9254376930ba8
+size 509496
diff --git a/vlm/dev/g6We1SwaY9/12.png b/vlm/dev/g6We1SwaY9/12.png
new file mode 100644
index 0000000000000000000000000000000000000000..7fd07227f7a628a45a9be31219c89798e4c3df56
--- /dev/null
+++ b/vlm/dev/g6We1SwaY9/12.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0296a240e4f4fca0789fa3dbf27786c8958d356e2f485f239b209cd215da1546
+size 850567
diff --git a/vlm/dev/g6We1SwaY9/13.png b/vlm/dev/g6We1SwaY9/13.png
new file mode 100644
index 0000000000000000000000000000000000000000..33492f97ea42077c5d6f93c9413b8de55e3aef8b
--- /dev/null
+++ b/vlm/dev/g6We1SwaY9/13.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7aadd3b91c7a53673491229a9e9939aa2c5bc9d45e6104b51aa40c6af19d0ac4
+size 490537
diff --git a/vlm/dev/g6We1SwaY9/14.png b/vlm/dev/g6We1SwaY9/14.png
new file mode 100644
index 0000000000000000000000000000000000000000..b10231da7a9854c3ff090e853bf4b9be0bdf5c72
--- /dev/null
+++ b/vlm/dev/g6We1SwaY9/14.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bf359129b525d052e5967de6ad4fbdf42169a48a50315179502c572c6332883d
+size 312150
diff --git a/vlm/dev/g6We1SwaY9/15.png b/vlm/dev/g6We1SwaY9/15.png
new file mode 100644
index 0000000000000000000000000000000000000000..e147aa0c0e7c72a9d8bbe872812a1d1df0ec82e9
--- /dev/null
+++ b/vlm/dev/g6We1SwaY9/15.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7a5c33bbbe15b8bfc09221282c2b33f2bcd73ed8b35c26d0fed3e36c77cabf44
+size 1669166
diff --git a/vlm/dev/g6We1SwaY9/16.png b/vlm/dev/g6We1SwaY9/16.png
new file mode 100644
index 0000000000000000000000000000000000000000..96f857bd293ea7201960f204a7d50b7a623bd3a5
--- /dev/null
+++ b/vlm/dev/g6We1SwaY9/16.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5bcbd9f3c35274f95ce7767e93b98f1ba30791cf68e77eb6338d0ee3b375b2e1
+size 2132927
diff --git a/vlm/dev/g6We1SwaY9/17.png b/vlm/dev/g6We1SwaY9/17.png
new file mode 100644
index 0000000000000000000000000000000000000000..e559b399abb028ef5f45a89b249e45fb266558fb
--- /dev/null
+++ b/vlm/dev/g6We1SwaY9/17.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:56ac61390b39b724408c70a2d5c1fb21287b7022605cd22e4e7556e993495aa3
+size 1966476
diff --git a/vlm/dev/g6We1SwaY9/18.png b/vlm/dev/g6We1SwaY9/18.png
new file mode 100644
index 0000000000000000000000000000000000000000..ee3683134fd8dfccdd81790932799d17191c2c93
--- /dev/null
+++ b/vlm/dev/g6We1SwaY9/18.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:edf700af4c2d4b44535554c076addc440db8c470887edd6b871c09d48301fcb6
+size 2253376
diff --git a/vlm/dev/g6We1SwaY9/19.png b/vlm/dev/g6We1SwaY9/19.png
new file mode 100644
index 0000000000000000000000000000000000000000..252977341ce9e95eef3fb221d27ad1aade8264d4
--- /dev/null
+++ b/vlm/dev/g6We1SwaY9/19.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e38ff7d53f6746c826bc862177609c0563b5f476ed2bef84763fc63b6c6b76b0
+size 2078665
diff --git a/vlm/dev/g6We1SwaY9/2.png b/vlm/dev/g6We1SwaY9/2.png
new file mode 100644
index 0000000000000000000000000000000000000000..f3e94750737d5fa55d033962f474f30dcd52fe6c
--- /dev/null
+++ b/vlm/dev/g6We1SwaY9/2.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:eacdbfc65292561a8369f17a9a88ca302e59dc4624b6f116da2df5f2d54ae760
+size 561102
diff --git a/vlm/dev/g6We1SwaY9/20.png b/vlm/dev/g6We1SwaY9/20.png
new file mode 100644
index 0000000000000000000000000000000000000000..0c19c5fcff70c01edff35ebd44c50d8f6d84bbae
--- /dev/null
+++ b/vlm/dev/g6We1SwaY9/20.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c34a200bb9804c9ccbec7d46bae629f15e5b8a9fe8e222a874300fc2a846983b
+size 1867347
diff --git a/vlm/dev/g6We1SwaY9/3.png b/vlm/dev/g6We1SwaY9/3.png
new file mode 100644
index 0000000000000000000000000000000000000000..b0e995524f4cc63ec54cef30d14a440dd8dcc92f
--- /dev/null
+++ b/vlm/dev/g6We1SwaY9/3.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b8a83ef454d2da4d7efce1a41194bb67e91a8631c8f22f49f372a31a22b9fb54
+size 881358
diff --git a/vlm/dev/g6We1SwaY9/4.png b/vlm/dev/g6We1SwaY9/4.png
new file mode 100644
index 0000000000000000000000000000000000000000..c1f5b6a2f3bd640c981c97ea99049dd7897144a7
--- /dev/null
+++ b/vlm/dev/g6We1SwaY9/4.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:dc9f0f9cdf41ab14e45f51314cd8ceb9328481089f02670b10caf1508c744a01
+size 655522
diff --git a/vlm/dev/g6We1SwaY9/5.png b/vlm/dev/g6We1SwaY9/5.png
new file mode 100644
index 0000000000000000000000000000000000000000..42e250bc1398360dc242d20f0c3a82f0ca34eade
--- /dev/null
+++ b/vlm/dev/g6We1SwaY9/5.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6121b0bece01d1d490e7ee887bd7e86ec50678d1d45522ba6e6d74eb7ee2ad97
+size 596149
diff --git a/vlm/dev/g6We1SwaY9/6.png b/vlm/dev/g6We1SwaY9/6.png
new file mode 100644
index 0000000000000000000000000000000000000000..ccc35a8b933b53bb8ebb1c9a7acbb24fb4a52a66
--- /dev/null
+++ b/vlm/dev/g6We1SwaY9/6.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4ae9c2ba477a4ab2047573dec786f77643628ddee393bcdd7688515bc30ba035
+size 2197278
diff --git a/vlm/dev/g6We1SwaY9/7.png b/vlm/dev/g6We1SwaY9/7.png
new file mode 100644
index 0000000000000000000000000000000000000000..0a24d226dbc41cc2ade50591612658af89f6665b
--- /dev/null
+++ b/vlm/dev/g6We1SwaY9/7.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:804ae24d41ec11f46eff5d18df3b3fe331129bd517d804c4553861711973136d
+size 1323821
diff --git a/vlm/dev/g6We1SwaY9/8.png b/vlm/dev/g6We1SwaY9/8.png
new file mode 100644
index 0000000000000000000000000000000000000000..fccd0a3ce6f15bba0c1360f141dd87f281ba44fb
--- /dev/null
+++ b/vlm/dev/g6We1SwaY9/8.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3e28a9c879953141f1b5052b1176a168f4972fcc69b0e84b41398f4df30d093e
+size 590515
diff --git a/vlm/dev/g6We1SwaY9/9.png b/vlm/dev/g6We1SwaY9/9.png
new file mode 100644
index 0000000000000000000000000000000000000000..34ea3e6e982c8c03cc82932676d7ca01bfbca5e0
--- /dev/null
+++ b/vlm/dev/g6We1SwaY9/9.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9aac783ee463635ae2df77ea7f2431141384d203bb3472c102e28a09a5d4b4c7
+size 515541
diff --git a/vlm/dev/gmL46YMpu2J/0.png b/vlm/dev/gmL46YMpu2J/0.png
new file mode 100644
index 0000000000000000000000000000000000000000..30c6362e632369f3f38e611530a1919d0cbf1b96
--- /dev/null
+++ b/vlm/dev/gmL46YMpu2J/0.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1698472d66a14036129d9979b94155814739d93ba6126f344b0165cb5575c707
+size 528233
diff --git a/vlm/dev/gmL46YMpu2J/1.png b/vlm/dev/gmL46YMpu2J/1.png
new file mode 100644
index 0000000000000000000000000000000000000000..4f4919d78bf2f69d5218b57714a6d6e7bc8644f0
--- /dev/null
+++ b/vlm/dev/gmL46YMpu2J/1.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fdb8d8653b065bfadaf48fe673bd280303c1156495ecf702d9beadaa580e7b2d
+size 649002
diff --git a/vlm/dev/gmL46YMpu2J/10.png b/vlm/dev/gmL46YMpu2J/10.png
new file mode 100644
index 0000000000000000000000000000000000000000..c713f8d0bc0010e8e31532e8d61b3c95fc152125
--- /dev/null
+++ b/vlm/dev/gmL46YMpu2J/10.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ce4574797b03dc0553d698b9d8092b64d9495ca0765651e8826b188fb0592bc5
+size 662890
diff --git a/vlm/dev/gmL46YMpu2J/11.png b/vlm/dev/gmL46YMpu2J/11.png
new file mode 100644
index 0000000000000000000000000000000000000000..96e3b24742a22c2378a7eb315951562ffe8fdc1a
--- /dev/null
+++ b/vlm/dev/gmL46YMpu2J/11.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:172d44f7f9eca912c1ba852ae2f1ae4a2041ed5bec351af27f56b604aac824a9
+size 669243
diff --git a/vlm/dev/gmL46YMpu2J/12.png b/vlm/dev/gmL46YMpu2J/12.png
new file mode 100644
index 0000000000000000000000000000000000000000..aad3d0dd22d827b04219e1fc104aa6fbf7fb21bf
--- /dev/null
+++ b/vlm/dev/gmL46YMpu2J/12.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5d3c34b68b72c99e743c1a68bf51fe801530ebf8e75301605cc8ab080a390eae
+size 636215
diff --git a/vlm/dev/gmL46YMpu2J/13.png b/vlm/dev/gmL46YMpu2J/13.png
new file mode 100644
index 0000000000000000000000000000000000000000..7481d0b6e3e4f889f30f3222fcfde878a018aad7
--- /dev/null
+++ b/vlm/dev/gmL46YMpu2J/13.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d4bfe803be271854e210b18a95ac95c758b45a671fefd4a4c77b695e390ccbf5
+size 673917
diff --git a/vlm/dev/gmL46YMpu2J/14.png b/vlm/dev/gmL46YMpu2J/14.png
new file mode 100644
index 0000000000000000000000000000000000000000..9c310e598bb251bb27c6cf2c37150aaaf476fdad
--- /dev/null
+++ b/vlm/dev/gmL46YMpu2J/14.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:09296434a9eac5ad0136498965c26e7ba43d5dfacfabf4f3ff12ba51dd6274a0
+size 64602
diff --git a/vlm/dev/gmL46YMpu2J/15.png b/vlm/dev/gmL46YMpu2J/15.png
new file mode 100644
index 0000000000000000000000000000000000000000..75e0f9d5ebb8bb7c13499aeaaee51032574059c8
--- /dev/null
+++ b/vlm/dev/gmL46YMpu2J/15.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8ec3d808c70e52f5053ba18d2728a5f1b425f563e62d71eaad8bfe86de0c342b
+size 410471
diff --git a/vlm/dev/gmL46YMpu2J/16.png b/vlm/dev/gmL46YMpu2J/16.png
new file mode 100644
index 0000000000000000000000000000000000000000..a6501475cc12cf93b75c3ca525d038a46259b1fd
--- /dev/null
+++ b/vlm/dev/gmL46YMpu2J/16.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3bf79d2547e8a401cfda10d0b8316ee4f1dfbcb4d9a033b5e30eae4fb877bd2d
+size 575683
diff --git a/vlm/dev/gmL46YMpu2J/17.png b/vlm/dev/gmL46YMpu2J/17.png
new file mode 100644
index 0000000000000000000000000000000000000000..03c80be53d42ce196dbd54f4093049739461e83e
--- /dev/null
+++ b/vlm/dev/gmL46YMpu2J/17.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:29c989ce946022a883fbab0acc274eff72de61afc70793dda068f422cefa5bbf
+size 490522
diff --git a/vlm/dev/gmL46YMpu2J/18.png b/vlm/dev/gmL46YMpu2J/18.png
new file mode 100644
index 0000000000000000000000000000000000000000..7c27c082927f23ab464bfd66aef2b8735312a0ee
--- /dev/null
+++ b/vlm/dev/gmL46YMpu2J/18.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ca33b894be4e838bbabcd0dc38d0d1dbb6fa88fce234dc001c7a55a0f2fc4eca
+size 235018
diff --git a/vlm/dev/gmL46YMpu2J/19.png b/vlm/dev/gmL46YMpu2J/19.png
new file mode 100644
index 0000000000000000000000000000000000000000..0d5433af626f139960c17fdeee775a67164f9fc1
--- /dev/null
+++ b/vlm/dev/gmL46YMpu2J/19.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fd36619f844e6d9eac434b7c2de7667fa603129c7f42638d154b21a395c2ec53
+size 369219
diff --git a/vlm/dev/gmL46YMpu2J/2.png b/vlm/dev/gmL46YMpu2J/2.png
new file mode 100644
index 0000000000000000000000000000000000000000..b127b3dbfdda13cffef1baaa69947bd037474856
--- /dev/null
+++ b/vlm/dev/gmL46YMpu2J/2.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:76737669d6e647dc7e338e3c0e19138fb4c5a436002ce134d10f805a7cd4cb71
+size 534490
diff --git a/vlm/dev/gmL46YMpu2J/20.png b/vlm/dev/gmL46YMpu2J/20.png
new file mode 100644
index 0000000000000000000000000000000000000000..c3653d190818c2d25514bfe2111b4ad3ecff5e96
--- /dev/null
+++ b/vlm/dev/gmL46YMpu2J/20.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ca3388762d0ef348d2dbcc14aeb386dc8c52dd2f3cef277416710a68f9026584
+size 301257
diff --git a/vlm/dev/gmL46YMpu2J/21.png b/vlm/dev/gmL46YMpu2J/21.png
new file mode 100644
index 0000000000000000000000000000000000000000..3afa02b55032cbb72f2b642f42f7139e30108165
--- /dev/null
+++ b/vlm/dev/gmL46YMpu2J/21.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4c44e1d4a5662adf2bfc0003a1e40e05f9b0d7fd8b0f53980b358dd5a894066a
+size 467407
diff --git a/vlm/dev/gmL46YMpu2J/3.png b/vlm/dev/gmL46YMpu2J/3.png
new file mode 100644
index 0000000000000000000000000000000000000000..82644fe5f0df5f75c2203d2f81c0bf7d3e93c7b6
--- /dev/null
+++ b/vlm/dev/gmL46YMpu2J/3.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bb52998c4d5f981174d520f1d507a15ed8e39906b37cf4e82128c5aeb2a98ee5
+size 575311
diff --git a/vlm/dev/gmL46YMpu2J/4.png b/vlm/dev/gmL46YMpu2J/4.png
new file mode 100644
index 0000000000000000000000000000000000000000..2492b0cad2ec55b37435ca4c4a675bb08df59b8d
--- /dev/null
+++ b/vlm/dev/gmL46YMpu2J/4.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:aaf4f9f1ca0313b68788a702bfc16c6b6db7ce87093033b336a12e485d6d457f
+size 513606
diff --git a/vlm/dev/gmL46YMpu2J/5.png b/vlm/dev/gmL46YMpu2J/5.png
new file mode 100644
index 0000000000000000000000000000000000000000..e8bfd4713f43952a26850568d61340d64857f029
--- /dev/null
+++ b/vlm/dev/gmL46YMpu2J/5.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:789c325eaa128ab9adde1a028316126af5de7b0dbd908eb0a89d0d7e4fdde9ed
+size 589796
diff --git a/vlm/dev/gmL46YMpu2J/6.png b/vlm/dev/gmL46YMpu2J/6.png
new file mode 100644
index 0000000000000000000000000000000000000000..e391e62dd33db5f722c439e8535c2eecd30e2455
--- /dev/null
+++ b/vlm/dev/gmL46YMpu2J/6.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:46294b68c49d1032f86dd1de90dddef7bf27f12ae22471f11589432c1fba647b
+size 566576
diff --git a/vlm/dev/gmL46YMpu2J/7.png b/vlm/dev/gmL46YMpu2J/7.png
new file mode 100644
index 0000000000000000000000000000000000000000..550830cdb2afe1d3bf7f77aa7d8e51c03de1112c
--- /dev/null
+++ b/vlm/dev/gmL46YMpu2J/7.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:18fd31b3beb6296050b7b193ecc9538fa09dfe1b16a65d7b3e258e7ee35b2f2b
+size 577321
diff --git a/vlm/dev/gmL46YMpu2J/8.png b/vlm/dev/gmL46YMpu2J/8.png
new file mode 100644
index 0000000000000000000000000000000000000000..4cc00777433fe4fe4c49e84452c8cd4be7abc427
--- /dev/null
+++ b/vlm/dev/gmL46YMpu2J/8.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:dac4ad4fd4c796a16fd22526ef7b3085a1129b8df8bdebe8164322740bd2b238
+size 585274
diff --git a/vlm/dev/gmL46YMpu2J/9.png b/vlm/dev/gmL46YMpu2J/9.png
new file mode 100644
index 0000000000000000000000000000000000000000..abd0a3cbda03ba0a4e878a072752638400b0c30b
--- /dev/null
+++ b/vlm/dev/gmL46YMpu2J/9.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c801896f05c5b82d1f84ff3e0388a9b2d2bf35a65e17b33bb6034ee41d7aaa12
+size 602551
diff --git a/vlm/dev/iMSjopcOn0p/0.png b/vlm/dev/iMSjopcOn0p/0.png
new file mode 100644
index 0000000000000000000000000000000000000000..fd07d1a8eae775e8684507a3f8df5dde5a989dce
--- /dev/null
+++ b/vlm/dev/iMSjopcOn0p/0.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:36311bbb44f3b4401d2c42e91dcc626e56811cd892614de1ece4080cf0d729a7
+size 505170
diff --git a/vlm/dev/iMSjopcOn0p/1.png b/vlm/dev/iMSjopcOn0p/1.png
new file mode 100644
index 0000000000000000000000000000000000000000..70ae7ba89ab2653b40b6bc91f9160c2ebf4013b8
--- /dev/null
+++ b/vlm/dev/iMSjopcOn0p/1.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:37a5fa2edf6cf1525626a2373fe381d59b9a0b3a28bf1e122c422efaece0c9e7
+size 586299
diff --git a/vlm/dev/iMSjopcOn0p/10.png b/vlm/dev/iMSjopcOn0p/10.png
new file mode 100644
index 0000000000000000000000000000000000000000..ad15ff9347a68c6bce3a4d0acdfeeed8b772a583
--- /dev/null
+++ b/vlm/dev/iMSjopcOn0p/10.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7a257018b6e2b217878dffee16f0bf34a57be06d1f26e8f62310cd0a7acfdb0a
+size 561248
diff --git a/vlm/dev/iMSjopcOn0p/11.png b/vlm/dev/iMSjopcOn0p/11.png
new file mode 100644
index 0000000000000000000000000000000000000000..65bd3723e37e39a01469452812fc63b7e6362840
--- /dev/null
+++ b/vlm/dev/iMSjopcOn0p/11.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:054be65f15db619aed3e77bae9f897ecde68e878e3e60229138b6cf270f66afb
+size 575785
diff --git a/vlm/dev/iMSjopcOn0p/12.png b/vlm/dev/iMSjopcOn0p/12.png
new file mode 100644
index 0000000000000000000000000000000000000000..fd24276929398e6b8deafd60720aa4c7fcbc8e84
--- /dev/null
+++ b/vlm/dev/iMSjopcOn0p/12.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bb291960eba5ef745b3397f65fab98808466bcf94f621e3a044d0abf4263a357
+size 653084
diff --git a/vlm/dev/iMSjopcOn0p/13.png b/vlm/dev/iMSjopcOn0p/13.png
new file mode 100644
index 0000000000000000000000000000000000000000..30d2f4c2bafed51424c4a9c42eb081814797c693
--- /dev/null
+++ b/vlm/dev/iMSjopcOn0p/13.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3294ac8e4afd6ea5f55358867fc6d068e0d8187977188cda9c0e38958e3cdcae
+size 426395
diff --git a/vlm/dev/iMSjopcOn0p/14.png b/vlm/dev/iMSjopcOn0p/14.png
new file mode 100644
index 0000000000000000000000000000000000000000..f659d729ae68172c9adf31053e8f486745cda035
--- /dev/null
+++ b/vlm/dev/iMSjopcOn0p/14.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2ff9741adeb31bc9f7178c8f978dd06fd523a688d2860846809486803b6284ea
+size 472260
diff --git a/vlm/dev/iMSjopcOn0p/15.png b/vlm/dev/iMSjopcOn0p/15.png
new file mode 100644
index 0000000000000000000000000000000000000000..ae4db782cf686e3eafc2222ae031053561d74e73
--- /dev/null
+++ b/vlm/dev/iMSjopcOn0p/15.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a083b335186347d63e84356d3c5d1a89860acd518e12d9dae4ddf9ad38328760
+size 543800
diff --git a/vlm/dev/iMSjopcOn0p/16.png b/vlm/dev/iMSjopcOn0p/16.png
new file mode 100644
index 0000000000000000000000000000000000000000..37433e6f8bb88697bcf7409204ae5b1153a75c4c
--- /dev/null
+++ b/vlm/dev/iMSjopcOn0p/16.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:87d930a110a097bd8fe123610ceb08d5536a0339a69836c4a2136978bbca1820
+size 367424
diff --git a/vlm/dev/iMSjopcOn0p/17.png b/vlm/dev/iMSjopcOn0p/17.png
new file mode 100644
index 0000000000000000000000000000000000000000..d10387ddbdc4c72e660a6fdb479c3f8e762dc22d
--- /dev/null
+++ b/vlm/dev/iMSjopcOn0p/17.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a65422568290199f901a2cae6a4fc7d5732d60c01cce3693a0f016fad6611d10
+size 459471
diff --git a/vlm/dev/iMSjopcOn0p/18.png b/vlm/dev/iMSjopcOn0p/18.png
new file mode 100644
index 0000000000000000000000000000000000000000..2371673ff47c1f840ff9428b3ee4ffe5bab22a1c
--- /dev/null
+++ b/vlm/dev/iMSjopcOn0p/18.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d2af2a660f4312a196315a685664992e4eb73676f7094e503faa5c55b26ea076
+size 213906
diff --git a/vlm/dev/iMSjopcOn0p/19.png b/vlm/dev/iMSjopcOn0p/19.png
new file mode 100644
index 0000000000000000000000000000000000000000..a86710191f9ae5405915c95f348e38797ca32c58
--- /dev/null
+++ b/vlm/dev/iMSjopcOn0p/19.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9375bf89059c241e452bbab1571cde3de42f8f24e37280aa98d7e974efd3dc0b
+size 407765
diff --git a/vlm/dev/iMSjopcOn0p/2.png b/vlm/dev/iMSjopcOn0p/2.png
new file mode 100644
index 0000000000000000000000000000000000000000..08e8af9c8f79cceafdc7b8b9aafd367ee6c38c2a
--- /dev/null
+++ b/vlm/dev/iMSjopcOn0p/2.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:11d56258de823f1e34f1c0b707a8f4b8d85076c38cbc9b21040eaab27b6d9b86
+size 563940
diff --git a/vlm/dev/iMSjopcOn0p/20.png b/vlm/dev/iMSjopcOn0p/20.png
new file mode 100644
index 0000000000000000000000000000000000000000..0bbb750c27b9f9453ea83cb216ec9391b80e0b30
--- /dev/null
+++ b/vlm/dev/iMSjopcOn0p/20.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:678424f54eb7d4a0f1fed0970044d337f30b6f139c1ab63c1126d5f675bf6927
+size 155861
diff --git a/vlm/dev/iMSjopcOn0p/3.png b/vlm/dev/iMSjopcOn0p/3.png
new file mode 100644
index 0000000000000000000000000000000000000000..e7b1683917bce9b7f1b82e007c1cd31d3d21e804
--- /dev/null
+++ b/vlm/dev/iMSjopcOn0p/3.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a645f7032c9be3b88027c40f0fa58bf11507d0bcde886ab852df7a0af5c239f2
+size 541596
diff --git a/vlm/dev/iMSjopcOn0p/4.png b/vlm/dev/iMSjopcOn0p/4.png
new file mode 100644
index 0000000000000000000000000000000000000000..3b76c0d62ea0e8c51449af20c5a4ad1d3d930279
--- /dev/null
+++ b/vlm/dev/iMSjopcOn0p/4.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:20ba3c8be31e1fd3b8d7c57611f8b2c659ceea2bcfa47f79b0e678b91faf1f5c
+size 537462
diff --git a/vlm/dev/iMSjopcOn0p/5.png b/vlm/dev/iMSjopcOn0p/5.png
new file mode 100644
index 0000000000000000000000000000000000000000..d5bf448e3b67725cd1451596d4394862d0be56bf
--- /dev/null
+++ b/vlm/dev/iMSjopcOn0p/5.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0530f83a0968cf9489067cdfe511a087f0c3894b01378787625b5f213b5c6ce6
+size 500127
diff --git a/vlm/dev/iMSjopcOn0p/6.png b/vlm/dev/iMSjopcOn0p/6.png
new file mode 100644
index 0000000000000000000000000000000000000000..9c499a5ee1e2138a4b8465de53c216af35fb1745
--- /dev/null
+++ b/vlm/dev/iMSjopcOn0p/6.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7fe93aed18a0f41e71429a0424388e1b743e33be89e4646f7339c2c339c2a98d
+size 550288
diff --git a/vlm/dev/iMSjopcOn0p/7.png b/vlm/dev/iMSjopcOn0p/7.png
new file mode 100644
index 0000000000000000000000000000000000000000..d8883cf87d9dc1dee4823f4e6c384843ba720c56
--- /dev/null
+++ b/vlm/dev/iMSjopcOn0p/7.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0285c317a668f11418552e0fa2d841a0e6869ef8f49e9b3ae68b3595f92d138e
+size 477166
diff --git a/vlm/dev/iMSjopcOn0p/8.png b/vlm/dev/iMSjopcOn0p/8.png
new file mode 100644
index 0000000000000000000000000000000000000000..478838904abccdd7692b1c8d52b47c61250c76cf
--- /dev/null
+++ b/vlm/dev/iMSjopcOn0p/8.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:96a2f26c29b25e75b5b2d3b5499524305f4e2c17c2949cbcf2ce8809a6919b35
+size 551548
diff --git a/vlm/dev/iMSjopcOn0p/9.png b/vlm/dev/iMSjopcOn0p/9.png
new file mode 100644
index 0000000000000000000000000000000000000000..050c926b89d3b89d33a1bcb6b9fa2904a023813d
--- /dev/null
+++ b/vlm/dev/iMSjopcOn0p/9.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c3164b24e661b644957e9bfe6e50c57f540d1434420da50c5b8b262b87f42f84
+size 545198
diff --git a/vlm/dev/kDEL91Dufpa/0.png b/vlm/dev/kDEL91Dufpa/0.png
new file mode 100644
index 0000000000000000000000000000000000000000..1c3630bd317ef06aa8cad979fe8b35a7f9ae41ee
--- /dev/null
+++ b/vlm/dev/kDEL91Dufpa/0.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a3bdb7992189794db181c2258bc06a20a1b814ed1a6e46f17c111a8f4f1df2f9
+size 454348
diff --git a/vlm/dev/kDEL91Dufpa/1.png b/vlm/dev/kDEL91Dufpa/1.png
new file mode 100644
index 0000000000000000000000000000000000000000..04315fd43801a72dcab50b7ec688550fa35b828b
--- /dev/null
+++ b/vlm/dev/kDEL91Dufpa/1.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8100ddfd7ac79c4655f5a02edb353446ea002b492a675e86b9231df677a8c93d
+size 616809
diff --git a/vlm/dev/kDEL91Dufpa/10.png b/vlm/dev/kDEL91Dufpa/10.png
new file mode 100644
index 0000000000000000000000000000000000000000..79524f29efd790afbbe56cd74c449732fb0712bc
--- /dev/null
+++ b/vlm/dev/kDEL91Dufpa/10.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d995b47f24967e3f0572bcd2d1af51c00869ea4d4197b93eb8c672bf8e51dea3
+size 547028
diff --git a/vlm/dev/kDEL91Dufpa/11.png b/vlm/dev/kDEL91Dufpa/11.png
new file mode 100644
index 0000000000000000000000000000000000000000..77970b15d2e39233df79e318767d537213685b8c
--- /dev/null
+++ b/vlm/dev/kDEL91Dufpa/11.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0690687416997c2402bc67ef485f0994d7900d7f110ddd9d90e08b7eeb3e4d80
+size 348743
diff --git a/vlm/dev/kDEL91Dufpa/12.png b/vlm/dev/kDEL91Dufpa/12.png
new file mode 100644
index 0000000000000000000000000000000000000000..411fa6547fe7e17e129c4a3bc33c6d23880618f3
--- /dev/null
+++ b/vlm/dev/kDEL91Dufpa/12.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d155e7df61f90b91cba5c223138db34af96a7d9d780d902530efa774838b7fcf
+size 303444
diff --git a/vlm/dev/kDEL91Dufpa/13.png b/vlm/dev/kDEL91Dufpa/13.png
new file mode 100644
index 0000000000000000000000000000000000000000..9eff37b9ff07f2fe3665de621e3d2c2a706c28dd
--- /dev/null
+++ b/vlm/dev/kDEL91Dufpa/13.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2d2a9d23d133acc4136e766620938f025447e51092e566d1e85e5a8cbab5a984
+size 523859
diff --git a/vlm/dev/kDEL91Dufpa/14.png b/vlm/dev/kDEL91Dufpa/14.png
new file mode 100644
index 0000000000000000000000000000000000000000..ab8ea85f1e7b8a355be32dff304e039980610ba8
--- /dev/null
+++ b/vlm/dev/kDEL91Dufpa/14.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e5c4a17a0d33f1c436c4c6b004b6442d192b6aa62cc2d4d9815106e0d13c2648
+size 361008
diff --git a/vlm/dev/kDEL91Dufpa/15.png b/vlm/dev/kDEL91Dufpa/15.png
new file mode 100644
index 0000000000000000000000000000000000000000..97922010830feb9cd6675e514815c74523469e71
--- /dev/null
+++ b/vlm/dev/kDEL91Dufpa/15.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:706bdce112d184e34440693f3ec116faf9f2e6545e98b4b433a340d17109735e
+size 403010
diff --git a/vlm/dev/kDEL91Dufpa/16.png b/vlm/dev/kDEL91Dufpa/16.png
new file mode 100644
index 0000000000000000000000000000000000000000..c58c6e7f641fe9f59a01f97152fbb57e1a113597
--- /dev/null
+++ b/vlm/dev/kDEL91Dufpa/16.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:77b4a678f76645cf10c9972d6c9022c5818e785d43ae08e32b92362ebd6c7f36
+size 316912
diff --git a/vlm/dev/kDEL91Dufpa/17.png b/vlm/dev/kDEL91Dufpa/17.png
new file mode 100644
index 0000000000000000000000000000000000000000..81a5bb12051266943bc64ef4731cea7eb408081e
--- /dev/null
+++ b/vlm/dev/kDEL91Dufpa/17.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:de829012b2141141d3d84888f509e0fc020eb1af18c8015584d7a8b6b817a091
+size 384182
diff --git a/vlm/dev/kDEL91Dufpa/18.png b/vlm/dev/kDEL91Dufpa/18.png
new file mode 100644
index 0000000000000000000000000000000000000000..8bcd29625751af9722b34a858819573e41b2ae9c
--- /dev/null
+++ b/vlm/dev/kDEL91Dufpa/18.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:305aec16f58ddd4286f492d6882460bdf3ef4dc7d17a3de56e6317bcadba3d9c
+size 496515
diff --git a/vlm/dev/kDEL91Dufpa/19.png b/vlm/dev/kDEL91Dufpa/19.png
new file mode 100644
index 0000000000000000000000000000000000000000..099b913bdfcbaacfbd4efcfd2d2d336b33168508
--- /dev/null
+++ b/vlm/dev/kDEL91Dufpa/19.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5e25d04297892ac949e2719232ee942930d31fc5bf0682a37c8f696927a7e4e6
+size 514695
diff --git a/vlm/dev/kDEL91Dufpa/2.png b/vlm/dev/kDEL91Dufpa/2.png
new file mode 100644
index 0000000000000000000000000000000000000000..0ca389561f6ac7b76ac79f67528d10554c5b4918
--- /dev/null
+++ b/vlm/dev/kDEL91Dufpa/2.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:809004747d5f51043e98507d992868940b52f73831c91caa9846621bbd84bc3c
+size 589249
diff --git a/vlm/dev/kDEL91Dufpa/20.png b/vlm/dev/kDEL91Dufpa/20.png
new file mode 100644
index 0000000000000000000000000000000000000000..dacd35d093c765e52d5b18cdbdddadb1cd11d461
--- /dev/null
+++ b/vlm/dev/kDEL91Dufpa/20.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5e3c9122981549bbd517d3c83f50670315b31345d18b94e18628afcd70c65239
+size 395605
diff --git a/vlm/dev/kDEL91Dufpa/21.png b/vlm/dev/kDEL91Dufpa/21.png
new file mode 100644
index 0000000000000000000000000000000000000000..ca21cc4fb9d7fdf0ee844f25b799830fbf052514
--- /dev/null
+++ b/vlm/dev/kDEL91Dufpa/21.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:42c34f8953aed2c623e73a70c49e1bfdb72fbb98a8e81cdc75309d6b5718897a
+size 377551
diff --git a/vlm/dev/kDEL91Dufpa/22.png b/vlm/dev/kDEL91Dufpa/22.png
new file mode 100644
index 0000000000000000000000000000000000000000..8d898fe3b458c05a7862967b32ab3d3329d88744
--- /dev/null
+++ b/vlm/dev/kDEL91Dufpa/22.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b547e3b1f204916ee983df32fd008a78a13d6cee850f72ac242e21b2aac3b61e
+size 398986
diff --git a/vlm/dev/kDEL91Dufpa/23.png b/vlm/dev/kDEL91Dufpa/23.png
new file mode 100644
index 0000000000000000000000000000000000000000..6457bc5d7f344a3251ac6eba19a233e3bb275292
--- /dev/null
+++ b/vlm/dev/kDEL91Dufpa/23.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:caf6ae76145a9fb83a4ee776b6d3f37666246554dadba2ba8e35521fb03adff0
+size 351810
diff --git a/vlm/dev/kDEL91Dufpa/24.png b/vlm/dev/kDEL91Dufpa/24.png
new file mode 100644
index 0000000000000000000000000000000000000000..105ace20fcf389e2ba40589bc27b31e494992ad5
--- /dev/null
+++ b/vlm/dev/kDEL91Dufpa/24.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:159f74db5e1f7173166792c35cd8828cd31ac7a7a267133487b5915158291cd1
+size 549255
diff --git a/vlm/dev/kDEL91Dufpa/25.png b/vlm/dev/kDEL91Dufpa/25.png
new file mode 100644
index 0000000000000000000000000000000000000000..6cc1a1d44b1f5fbd2225baca6652632b8318c63b
--- /dev/null
+++ b/vlm/dev/kDEL91Dufpa/25.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:23517f81f4c226dade8f9ef98dd6ebb3a4ee6dab343f8b4626e2db955623500f
+size 321097
diff --git a/vlm/dev/kDEL91Dufpa/26.png b/vlm/dev/kDEL91Dufpa/26.png
new file mode 100644
index 0000000000000000000000000000000000000000..fe02894f6b062826d7bd0313abc2974ead70b222
--- /dev/null
+++ b/vlm/dev/kDEL91Dufpa/26.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:91fd57ebda0d50c7a9a633cf523344b2d394363f945bf321ad3924a1b5899f1c
+size 245840
diff --git a/vlm/dev/kDEL91Dufpa/27.png b/vlm/dev/kDEL91Dufpa/27.png
new file mode 100644
index 0000000000000000000000000000000000000000..ecac10ec44bc286828a75c75b26b24c9ff41bcff
--- /dev/null
+++ b/vlm/dev/kDEL91Dufpa/27.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e664280c601ad454092759cc6733a8f8fad084d9dcd70648a3a5c3d989ba5b18
+size 249796
diff --git a/vlm/dev/kDEL91Dufpa/3.png b/vlm/dev/kDEL91Dufpa/3.png
new file mode 100644
index 0000000000000000000000000000000000000000..b2d51ed25c4e233b1183b42cf68485cd6794e37f
--- /dev/null
+++ b/vlm/dev/kDEL91Dufpa/3.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3cc89c5678fe6f7d606cd8653dc825673c4856e577d3a0407f0acfae7656a28f
+size 605118
diff --git a/vlm/dev/kDEL91Dufpa/4.png b/vlm/dev/kDEL91Dufpa/4.png
new file mode 100644
index 0000000000000000000000000000000000000000..132cbdda3cd0933a2bc4219a59e2cefa08b7597a
--- /dev/null
+++ b/vlm/dev/kDEL91Dufpa/4.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:04f84472fb12144ee85a56fca8daddb51055f5e8c8aa5843b4aedc326cb8a2b1
+size 441707
diff --git a/vlm/dev/kDEL91Dufpa/5.png b/vlm/dev/kDEL91Dufpa/5.png
new file mode 100644
index 0000000000000000000000000000000000000000..38447d80419a1068edbd401c1f3951b8f255838f
--- /dev/null
+++ b/vlm/dev/kDEL91Dufpa/5.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:16b844bd1687c4a5c780ed53b6ef03c1306324131d3b19111d1235fef639031d
+size 475278
diff --git a/vlm/dev/kDEL91Dufpa/6.png b/vlm/dev/kDEL91Dufpa/6.png
new file mode 100644
index 0000000000000000000000000000000000000000..4c471e526898dd1f7299c897d68364cebc4c079c
--- /dev/null
+++ b/vlm/dev/kDEL91Dufpa/6.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7cf75f2875ab2c9e04e0ce3f7c5a69801fce09171a50f3b799af84f46a792b65
+size 559635
diff --git a/vlm/dev/kDEL91Dufpa/7.png b/vlm/dev/kDEL91Dufpa/7.png
new file mode 100644
index 0000000000000000000000000000000000000000..0215d18a30889790ce2219c0e419a25264d2d393
--- /dev/null
+++ b/vlm/dev/kDEL91Dufpa/7.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a04266ada7dc159c4d7e785fe17bd11c331232ca16315be7f85cad5e0a7e91c1
+size 576009
diff --git a/vlm/dev/kDEL91Dufpa/8.png b/vlm/dev/kDEL91Dufpa/8.png
new file mode 100644
index 0000000000000000000000000000000000000000..8e99f8b0030c0805975403af379c96e86c99bb96
--- /dev/null
+++ b/vlm/dev/kDEL91Dufpa/8.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:74d5e6da93986dcc1a9d1c3d4c943234d168f9a5514d546d251d17740d6a88c9
+size 535086
diff --git a/vlm/dev/kDEL91Dufpa/9.png b/vlm/dev/kDEL91Dufpa/9.png
new file mode 100644
index 0000000000000000000000000000000000000000..3e0a1afcac3be7bff64cb9ddf455febc90442699
--- /dev/null
+++ b/vlm/dev/kDEL91Dufpa/9.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0771d3721bf44f8a295b7f81997ddbfcc3a5d845e549bcab7b5a2c66ec2ef95f
+size 494399
diff --git a/vlm/dev/ktrw68Cmu9c/0.png b/vlm/dev/ktrw68Cmu9c/0.png
new file mode 100644
index 0000000000000000000000000000000000000000..0fe5f582c987c04cf4dbac5a76b899e773ab77ac
--- /dev/null
+++ b/vlm/dev/ktrw68Cmu9c/0.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a5bc40e07eae1dde9594681a169e7e7c2f7e1663cbe0493ed1381aae357c1b6d
+size 503156
diff --git a/vlm/dev/ktrw68Cmu9c/1.png b/vlm/dev/ktrw68Cmu9c/1.png
new file mode 100644
index 0000000000000000000000000000000000000000..65747d33021e8001beeacc4ac66c1b1fa71044f3
--- /dev/null
+++ b/vlm/dev/ktrw68Cmu9c/1.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:78e6854ef3bf5d27897cd2abf008eaa6ec0a37e5ddce545911d6373773930d9d
+size 559662
diff --git a/vlm/dev/ktrw68Cmu9c/10.png b/vlm/dev/ktrw68Cmu9c/10.png
new file mode 100644
index 0000000000000000000000000000000000000000..9f09b8b04c0a5e1a968f6e68d5dfd63bed708194
--- /dev/null
+++ b/vlm/dev/ktrw68Cmu9c/10.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bb09b8a7c3805ec4e26c3846899bea2383bbd49bd906326945e73a2148749d45
+size 595397
diff --git a/vlm/dev/ktrw68Cmu9c/11.png b/vlm/dev/ktrw68Cmu9c/11.png
new file mode 100644
index 0000000000000000000000000000000000000000..529f193dd6c2c39735db076141d90a7cf52979a8
--- /dev/null
+++ b/vlm/dev/ktrw68Cmu9c/11.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3d9fbc6119ef1280c88ab7e50053cf0a43fd4e09244cce24b3818fdaeff59983
+size 527502
diff --git a/vlm/dev/ktrw68Cmu9c/12.png b/vlm/dev/ktrw68Cmu9c/12.png
new file mode 100644
index 0000000000000000000000000000000000000000..ae2ec8393e338281288563e4a82957b3f5a6b015
--- /dev/null
+++ b/vlm/dev/ktrw68Cmu9c/12.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:71c5ce42ed8fecaff77d45075cbae09bf008e8432472b6828934c1b3d2ba3442
+size 460473
diff --git a/vlm/dev/ktrw68Cmu9c/13.png b/vlm/dev/ktrw68Cmu9c/13.png
new file mode 100644
index 0000000000000000000000000000000000000000..69e3df6939c0058ff055ad5155f54df0c4f3823d
--- /dev/null
+++ b/vlm/dev/ktrw68Cmu9c/13.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:79f0e8f8dc533cb123a9c73b37df2f41506a4c97e9e597d5925fbfcd0c8ac0eb
+size 477159
diff --git a/vlm/dev/ktrw68Cmu9c/14.png b/vlm/dev/ktrw68Cmu9c/14.png
new file mode 100644
index 0000000000000000000000000000000000000000..5ce3f14834e1c21a0385dca82e2a12be366f8312
--- /dev/null
+++ b/vlm/dev/ktrw68Cmu9c/14.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:accb5f520069e5106c44ed9fb27f79468739eec6e1bc276288cc15f392e37336
+size 520453
diff --git a/vlm/dev/ktrw68Cmu9c/15.png b/vlm/dev/ktrw68Cmu9c/15.png
new file mode 100644
index 0000000000000000000000000000000000000000..854fda9c111ca9a5ca638cc8d459f4f5a1eb8af4
--- /dev/null
+++ b/vlm/dev/ktrw68Cmu9c/15.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7dfde0c03235fda2b6f845f15867498f0064df01bde2f666beac14a0f303ea5d
+size 477623
diff --git a/vlm/dev/ktrw68Cmu9c/17.png b/vlm/dev/ktrw68Cmu9c/17.png
new file mode 100644
index 0000000000000000000000000000000000000000..4ae27a9b5884f81c9c546efd443175e366bea935
--- /dev/null
+++ b/vlm/dev/ktrw68Cmu9c/17.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a8eed455a3bbd00328dd2e5ebe3564a8203209a5b89ca3fb082e3f3cd2b8048b
+size 463601
diff --git a/vlm/dev/ktrw68Cmu9c/18.png b/vlm/dev/ktrw68Cmu9c/18.png
new file mode 100644
index 0000000000000000000000000000000000000000..f130936840e5c333ec78f29a4c067f5590ed11f6
--- /dev/null
+++ b/vlm/dev/ktrw68Cmu9c/18.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:68b12f0e5ca7e9821a491c24c78bb7c9c63cc4de9813952fc22f02efa59cece3
+size 230379
diff --git a/vlm/dev/ktrw68Cmu9c/2.png b/vlm/dev/ktrw68Cmu9c/2.png
new file mode 100644
index 0000000000000000000000000000000000000000..cee0e3d8458a4dda027e3006bd2604dc5255b72c
--- /dev/null
+++ b/vlm/dev/ktrw68Cmu9c/2.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0006e331ef801c0341e2c45b2d8f4f52790485094ff2c055edd74987fbf60de5
+size 598321
diff --git a/vlm/dev/ktrw68Cmu9c/3.png b/vlm/dev/ktrw68Cmu9c/3.png
new file mode 100644
index 0000000000000000000000000000000000000000..a63249023df9da35760ac9010e3acb315260eab6
--- /dev/null
+++ b/vlm/dev/ktrw68Cmu9c/3.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5953e73b5d227c8373f7f62e70bd71c3916de9847fd505d73d3a9ef60d963563
+size 607695
diff --git a/vlm/dev/ktrw68Cmu9c/4.png b/vlm/dev/ktrw68Cmu9c/4.png
new file mode 100644
index 0000000000000000000000000000000000000000..45bee6b5fd56d1751f8c7397e946b738ebf4b075
--- /dev/null
+++ b/vlm/dev/ktrw68Cmu9c/4.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1f7066294ac5e7f62f17490363ae599c338894f3c6dfc06646578cfa26f1cb2d
+size 573981
diff --git a/vlm/dev/ktrw68Cmu9c/5.png b/vlm/dev/ktrw68Cmu9c/5.png
new file mode 100644
index 0000000000000000000000000000000000000000..2b1cbe365679e80a31e593e3b1b01de2b803ce24
--- /dev/null
+++ b/vlm/dev/ktrw68Cmu9c/5.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8226abaa6c2ad3b8bf33b4a270256dd1f6944eeb37eb5db3a0f78c9b7c319815
+size 566497
diff --git a/vlm/dev/ktrw68Cmu9c/6.png b/vlm/dev/ktrw68Cmu9c/6.png
new file mode 100644
index 0000000000000000000000000000000000000000..382ffe26a794d9b6efdcce51ff27fb117d88035b
--- /dev/null
+++ b/vlm/dev/ktrw68Cmu9c/6.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:64b69871ed348fa5dce8f97c45605b53fd019cbd57857227c9c38387f45264ed
+size 545518
diff --git a/vlm/dev/ktrw68Cmu9c/7.png b/vlm/dev/ktrw68Cmu9c/7.png
new file mode 100644
index 0000000000000000000000000000000000000000..4daedb5a14f13c53a8b77549e6de9f83e28c5bdf
--- /dev/null
+++ b/vlm/dev/ktrw68Cmu9c/7.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cbf8d9f508a5c065034347e309125667104a3d07ad555d84f84c0de585c57e5d
+size 601520
diff --git a/vlm/dev/ktrw68Cmu9c/8.png b/vlm/dev/ktrw68Cmu9c/8.png
new file mode 100644
index 0000000000000000000000000000000000000000..42804a35f25c9f857dab75306b22b9ee15578263
--- /dev/null
+++ b/vlm/dev/ktrw68Cmu9c/8.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d411c217198d55f34ca440101a0d48b86ed4dc3ddbdddd97ac5346852eaa500e
+size 597176
diff --git a/vlm/dev/ktrw68Cmu9c/9.png b/vlm/dev/ktrw68Cmu9c/9.png
new file mode 100644
index 0000000000000000000000000000000000000000..e5c23db4e128fb1442376bab8df3dadef5e41d97
--- /dev/null
+++ b/vlm/dev/ktrw68Cmu9c/9.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:532c7eb1958c2c4b74f3d8f1a45f5ebf3b14b4564075cf891c1acd6d335772bf
+size 512571
diff --git a/vlm/dev/qQuzhbU3Gto/0.png b/vlm/dev/qQuzhbU3Gto/0.png
new file mode 100644
index 0000000000000000000000000000000000000000..55bc53a3e7e10995cd2770bdcd2a87a0481eaf28
--- /dev/null
+++ b/vlm/dev/qQuzhbU3Gto/0.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:02b4aea8adf89b68996f5ac96e3e63085e0580193bd5b949b3ce94e2cb4d5a6e
+size 488653
diff --git a/vlm/dev/qQuzhbU3Gto/1.png b/vlm/dev/qQuzhbU3Gto/1.png
new file mode 100644
index 0000000000000000000000000000000000000000..d9dce9a1f673181a370b5d5644ae51ce94eaf483
--- /dev/null
+++ b/vlm/dev/qQuzhbU3Gto/1.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5d47ab81f2c2e95d13514c05e2af4b33d3d35ae572999b83fc150b70211ae104
+size 692210
diff --git a/vlm/dev/qQuzhbU3Gto/10.png b/vlm/dev/qQuzhbU3Gto/10.png
new file mode 100644
index 0000000000000000000000000000000000000000..3fbef7103ffcd843a8c0c35f96fe9e28fdeacaf1
--- /dev/null
+++ b/vlm/dev/qQuzhbU3Gto/10.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cfe3325ddca06a9c6e493984af6368ac721fc86fadcb47b905d9c208cd434091
+size 547623
diff --git a/vlm/dev/qQuzhbU3Gto/11.png b/vlm/dev/qQuzhbU3Gto/11.png
new file mode 100644
index 0000000000000000000000000000000000000000..a2998fab46331fe33052acff0fbc358aadec698b
--- /dev/null
+++ b/vlm/dev/qQuzhbU3Gto/11.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bf204978b96de2a2ab8cb8e72fd2ecf18fa9a2de7150b2e31ab01876cf0b353f
+size 497529
diff --git a/vlm/dev/qQuzhbU3Gto/2.png b/vlm/dev/qQuzhbU3Gto/2.png
new file mode 100644
index 0000000000000000000000000000000000000000..2fa0ca767f14a1073ee304d58c5edfc966141202
--- /dev/null
+++ b/vlm/dev/qQuzhbU3Gto/2.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5c10bb03f7af4dacbe1a25c9b73d5cc19c7a2beccf5f901284b072723df2b74c
+size 491196
diff --git a/vlm/dev/qQuzhbU3Gto/3.png b/vlm/dev/qQuzhbU3Gto/3.png
new file mode 100644
index 0000000000000000000000000000000000000000..cf4faa2c137868b5d6694b81b6d459821a0ab1f9
--- /dev/null
+++ b/vlm/dev/qQuzhbU3Gto/3.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8a0f11933370caa38718da9e0a1d94cc4d9a120921a79cec6a8c2a6ce9604ab0
+size 685226
diff --git a/vlm/dev/qQuzhbU3Gto/4.png b/vlm/dev/qQuzhbU3Gto/4.png
new file mode 100644
index 0000000000000000000000000000000000000000..e777b01d93331bf3e5d9d799db6bc721be2c12a7
--- /dev/null
+++ b/vlm/dev/qQuzhbU3Gto/4.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6a6776bcc70db65a869f7161e28c38d98d22994c1b530a403dac319d2221bf50
+size 477736
diff --git a/vlm/dev/qQuzhbU3Gto/5.png b/vlm/dev/qQuzhbU3Gto/5.png
new file mode 100644
index 0000000000000000000000000000000000000000..cc1c367b5de891926ab264118246539ae9a08394
--- /dev/null
+++ b/vlm/dev/qQuzhbU3Gto/5.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:adcb701621c81db7e25a735430446c280f5269b5e3f825b86df401c684864bc5
+size 455835
diff --git a/vlm/dev/qQuzhbU3Gto/6.png b/vlm/dev/qQuzhbU3Gto/6.png
new file mode 100644
index 0000000000000000000000000000000000000000..32cc7624f9278695c66cd7756d550b9bfea004fb
--- /dev/null
+++ b/vlm/dev/qQuzhbU3Gto/6.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:38e497ee93702248089cfcac15cb28baf53c8cb8bd49ff01dc87b9fab04910e6
+size 570923
diff --git a/vlm/dev/qQuzhbU3Gto/7.png b/vlm/dev/qQuzhbU3Gto/7.png
new file mode 100644
index 0000000000000000000000000000000000000000..2ffea572720eb5b19556362a7b9e0b6947b6613c
--- /dev/null
+++ b/vlm/dev/qQuzhbU3Gto/7.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:724b02f184cdf035e6008eed9798bf8563684df949724e73c8346f03ba8aebcf
+size 394887
diff --git a/vlm/dev/qQuzhbU3Gto/8.png b/vlm/dev/qQuzhbU3Gto/8.png
new file mode 100644
index 0000000000000000000000000000000000000000..987fafae3ba84699a048aa3be1400a22b6046ab3
--- /dev/null
+++ b/vlm/dev/qQuzhbU3Gto/8.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2b502815d91355837db35bc3f4d980a18cb551b873ebd36ebf57de7de74e9ac5
+size 379541
diff --git a/vlm/dev/qQuzhbU3Gto/9.png b/vlm/dev/qQuzhbU3Gto/9.png
new file mode 100644
index 0000000000000000000000000000000000000000..a2fa6802209fdb91d4353563cf7bcc2559b17371
--- /dev/null
+++ b/vlm/dev/qQuzhbU3Gto/9.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cf7903ea1f3bcc003e5e9d9c6cefcb030ea0b707a38fac5410da00ea5447cc2b
+size 539946
diff --git a/vlm/dev/t877958UGZ/0.png b/vlm/dev/t877958UGZ/0.png
new file mode 100644
index 0000000000000000000000000000000000000000..05fe347f2c1ed2457cfa3b3ab2e17af124ebe27e
--- /dev/null
+++ b/vlm/dev/t877958UGZ/0.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e58864a8a3892a47a640c2b464c32bb9279c9fa7a05c2d2717a6f72cc183cd41
+size 481315
diff --git a/vlm/dev/t877958UGZ/1.png b/vlm/dev/t877958UGZ/1.png
new file mode 100644
index 0000000000000000000000000000000000000000..39e0cee357e2df1fbe1af86df354fd3aee4419ed
--- /dev/null
+++ b/vlm/dev/t877958UGZ/1.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ef33712174a183a6ec5c436645b2c35472f7441de881e3f3f8f0b9fe5417f435
+size 657676
diff --git a/vlm/dev/t877958UGZ/10.png b/vlm/dev/t877958UGZ/10.png
new file mode 100644
index 0000000000000000000000000000000000000000..50bd664b90d7508853fcd27da635083792ec1102
--- /dev/null
+++ b/vlm/dev/t877958UGZ/10.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f6bf9ad224a1e1cedba0d4a1c85b110bb22bec3fd9dddc0c3cc91a26461df7af
+size 548051
diff --git a/vlm/dev/t877958UGZ/11.png b/vlm/dev/t877958UGZ/11.png
new file mode 100644
index 0000000000000000000000000000000000000000..cc433f7c02f2c85969596a91c7d2c280c903bdb3
--- /dev/null
+++ b/vlm/dev/t877958UGZ/11.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9ef98dd05a23c5790ce2a2e18436ce2983f340629303b5e5a81c6882aa04ea51
+size 552795
diff --git a/vlm/dev/t877958UGZ/12.png b/vlm/dev/t877958UGZ/12.png
new file mode 100644
index 0000000000000000000000000000000000000000..a00e35dbd4b3731f550611f95016664f5b9ceb01
--- /dev/null
+++ b/vlm/dev/t877958UGZ/12.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e1e2ec92aff4e15008cdb770f150dd90574c6b91a7202659c3dfa4c480f781c3
+size 504257
diff --git a/vlm/dev/t877958UGZ/2.png b/vlm/dev/t877958UGZ/2.png
new file mode 100644
index 0000000000000000000000000000000000000000..925dfb535b542570ef257f84dd38764c624d7ff4
--- /dev/null
+++ b/vlm/dev/t877958UGZ/2.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2a3389575a9317ba4279f191b489e52f931a7006128502082391a4d3c9f2b488
+size 552457
diff --git a/vlm/dev/t877958UGZ/3.png b/vlm/dev/t877958UGZ/3.png
new file mode 100644
index 0000000000000000000000000000000000000000..766ccb5c5c86312305a026fd97133aa53abedd28
--- /dev/null
+++ b/vlm/dev/t877958UGZ/3.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ddcefeba6d1ac5c983a87bca00980265ccd8baa23686916be0788263ec8929be
+size 597113
diff --git a/vlm/dev/t877958UGZ/4.png b/vlm/dev/t877958UGZ/4.png
new file mode 100644
index 0000000000000000000000000000000000000000..4486bb2ece784ab8f6da03fa615326d78a70c96c
--- /dev/null
+++ b/vlm/dev/t877958UGZ/4.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b5d4cdfc081c28f3e2cce7719e20b395047c6c6be0f41c0bfd134baa9ce79d81
+size 481428
diff --git a/vlm/dev/t877958UGZ/5.png b/vlm/dev/t877958UGZ/5.png
new file mode 100644
index 0000000000000000000000000000000000000000..52c8cd664f4e316930c2c7cf0de7c68e66a23188
--- /dev/null
+++ b/vlm/dev/t877958UGZ/5.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6b0bd69c1988d5c2524e46d78c6c71b64bfb2af3a27dedf3ac60bc7a650073c6
+size 596133
diff --git a/vlm/dev/t877958UGZ/6.png b/vlm/dev/t877958UGZ/6.png
new file mode 100644
index 0000000000000000000000000000000000000000..84386e11557be8572003ef4ebb2e1345bb3a7f5a
--- /dev/null
+++ b/vlm/dev/t877958UGZ/6.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b6610747013cce19f358c4ee079aa52c7f7e0a12c9ccf035aa6cfe369f5ab3b7
+size 653678
diff --git a/vlm/dev/t877958UGZ/7.png b/vlm/dev/t877958UGZ/7.png
new file mode 100644
index 0000000000000000000000000000000000000000..db68a4f054d92d5573df05a096fe8d4c739090cc
--- /dev/null
+++ b/vlm/dev/t877958UGZ/7.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9c9029034bf5ac3d6ca32383e214c3ab59f34b3be473239628497f56c51138c2
+size 697626
diff --git a/vlm/dev/t877958UGZ/8.png b/vlm/dev/t877958UGZ/8.png
new file mode 100644
index 0000000000000000000000000000000000000000..62a0180f1d0c2177fa2629081a2a245661d9a751
--- /dev/null
+++ b/vlm/dev/t877958UGZ/8.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:060cf40ae08a78ff91cbd8832a5e01ae3369eebbd8a3133f0041cbd97d982718
+size 731424
diff --git a/vlm/dev/t877958UGZ/9.png b/vlm/dev/t877958UGZ/9.png
new file mode 100644
index 0000000000000000000000000000000000000000..80f0e8cb7b695bb941e0805f65df044066b54393
--- /dev/null
+++ b/vlm/dev/t877958UGZ/9.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:01b00c07290de61cbbb579ef6488a0bca17b243dc9f7c82c4d821a6906ad5faa
+size 738856
diff --git a/vlm/dev/tHK5ntjp-5K/0.png b/vlm/dev/tHK5ntjp-5K/0.png
new file mode 100644
index 0000000000000000000000000000000000000000..dcef1307566f549409e3bd14c9bbc22220830393
--- /dev/null
+++ b/vlm/dev/tHK5ntjp-5K/0.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b6b6239b01386a062a5b8f698b919eda2b05db1865eb7d35463e3a10b77ba0a0
+size 440651
diff --git a/vlm/dev/tHK5ntjp-5K/1.png b/vlm/dev/tHK5ntjp-5K/1.png
new file mode 100644
index 0000000000000000000000000000000000000000..9fd3bcb69a7d532b197fcfbee04241146ff9d168
--- /dev/null
+++ b/vlm/dev/tHK5ntjp-5K/1.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a4d52e8ec6d85ac670017e21f1386e1706f24bacd9743476ddcbfb3410eb81e0
+size 816618
diff --git a/vlm/dev/tHK5ntjp-5K/10.png b/vlm/dev/tHK5ntjp-5K/10.png
new file mode 100644
index 0000000000000000000000000000000000000000..af0f8c73ebb3bee67ffdc0c2999542688cb897e2
--- /dev/null
+++ b/vlm/dev/tHK5ntjp-5K/10.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a91b961e66faf615b7b5b702bcd50b1cc8c38f1f8ac25ab09d2a883e0564cf7a
+size 563974
diff --git a/vlm/dev/tHK5ntjp-5K/11.png b/vlm/dev/tHK5ntjp-5K/11.png
new file mode 100644
index 0000000000000000000000000000000000000000..0d9e0b41750142edfbd1d01bbb38ebbf04041f24
--- /dev/null
+++ b/vlm/dev/tHK5ntjp-5K/11.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c183d95901422de5a24dcbb9ff7f49817953c4839a3f63189474d77763c752aa
+size 555336
diff --git a/vlm/dev/tHK5ntjp-5K/12.png b/vlm/dev/tHK5ntjp-5K/12.png
new file mode 100644
index 0000000000000000000000000000000000000000..805b64f40deb3e8a00568013699b2c1d4c32861d
--- /dev/null
+++ b/vlm/dev/tHK5ntjp-5K/12.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:131bc0e0ab75f642de4b1495f4737bcafa7f7774e6d72cd3580a8992fe7fab89
+size 556448
diff --git a/vlm/dev/tHK5ntjp-5K/13.png b/vlm/dev/tHK5ntjp-5K/13.png
new file mode 100644
index 0000000000000000000000000000000000000000..f5b412b23a736a49d3bfdfd3017c8140181ba985
--- /dev/null
+++ b/vlm/dev/tHK5ntjp-5K/13.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c7e107e9502024291b09595821e247dee5c113735d556b86325ceb7e1f27c174
+size 561052
diff --git a/vlm/dev/tHK5ntjp-5K/14.png b/vlm/dev/tHK5ntjp-5K/14.png
new file mode 100644
index 0000000000000000000000000000000000000000..157ef98b61c4416eda2ee6d2cba673c4726e4bd5
--- /dev/null
+++ b/vlm/dev/tHK5ntjp-5K/14.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0d6a010963bec05161f1cdfc5bb0885eb6cf12c207ec98a631c179c009f44f07
+size 551283
diff --git a/vlm/dev/tHK5ntjp-5K/15.png b/vlm/dev/tHK5ntjp-5K/15.png
new file mode 100644
index 0000000000000000000000000000000000000000..fee3908d8743503899d8feb8bda2579701b0b8a7
--- /dev/null
+++ b/vlm/dev/tHK5ntjp-5K/15.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:40c6201c75ee37c4e1f03cd79cbe5bdd5999ce8f7349629a8f89fb8a2600fdab
+size 566036
diff --git a/vlm/dev/tHK5ntjp-5K/16.png b/vlm/dev/tHK5ntjp-5K/16.png
new file mode 100644
index 0000000000000000000000000000000000000000..4de25d15453ff21d9389e4e850920101956fc5b9
--- /dev/null
+++ b/vlm/dev/tHK5ntjp-5K/16.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:579c3b8a4f196266448ed6daa437651e3b7cf24bc928e62fbef1d36deeb7da9f
+size 540634
diff --git a/vlm/dev/tHK5ntjp-5K/17.png b/vlm/dev/tHK5ntjp-5K/17.png
new file mode 100644
index 0000000000000000000000000000000000000000..8b5b3d0998c572053c993aafd197e97076239f16
--- /dev/null
+++ b/vlm/dev/tHK5ntjp-5K/17.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:285caa49f6aab5f531ee54a30d28baa52349ea237e3cfbd518faedc9602ecacf
+size 509196
diff --git a/vlm/dev/tHK5ntjp-5K/18.png b/vlm/dev/tHK5ntjp-5K/18.png
new file mode 100644
index 0000000000000000000000000000000000000000..037fea6834d729919f19953879443c00744a2d6a
--- /dev/null
+++ b/vlm/dev/tHK5ntjp-5K/18.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fd382dd6de40cc4cf745fc027639b882072bc92469d42e1272cf6ea80f1fac23
+size 368030
diff --git a/vlm/dev/tHK5ntjp-5K/2.png b/vlm/dev/tHK5ntjp-5K/2.png
new file mode 100644
index 0000000000000000000000000000000000000000..d910aaedc008b34dc140e67f50520c3b2ceeadc6
--- /dev/null
+++ b/vlm/dev/tHK5ntjp-5K/2.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3a6cb6f3ea92d31d12d8e05496649c8b4f5c02671ffaa2c8f70337e94fbfbec3
+size 759477
diff --git a/vlm/dev/tHK5ntjp-5K/3.png b/vlm/dev/tHK5ntjp-5K/3.png
new file mode 100644
index 0000000000000000000000000000000000000000..da1483bebbce4d1d4517191a61b6c48d9c92ce27
--- /dev/null
+++ b/vlm/dev/tHK5ntjp-5K/3.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bc52d8fb86ee9b2f064b79a1ef17bac16b8436978275f98f54f2d0f79be377d4
+size 815517
diff --git a/vlm/dev/tHK5ntjp-5K/4.png b/vlm/dev/tHK5ntjp-5K/4.png
new file mode 100644
index 0000000000000000000000000000000000000000..c727ec6686718bab734193c9a5bd5585311cdccc
--- /dev/null
+++ b/vlm/dev/tHK5ntjp-5K/4.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c58cf06861e20285f48b00e658b74339d42264d25c57cf3859f81d031390740a
+size 824482
diff --git a/vlm/dev/tHK5ntjp-5K/5.png b/vlm/dev/tHK5ntjp-5K/5.png
new file mode 100644
index 0000000000000000000000000000000000000000..7efe6468e21791d302d7ffdbc8ba7bcfe92e8462
--- /dev/null
+++ b/vlm/dev/tHK5ntjp-5K/5.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d4ab7974183e0f51083c5e154d454cf740806087c8af61a7ed8790122b8c809c
+size 972474
diff --git a/vlm/dev/tHK5ntjp-5K/6.png b/vlm/dev/tHK5ntjp-5K/6.png
new file mode 100644
index 0000000000000000000000000000000000000000..83b57c27defce95d921b8edb7f361caa2f364c9d
--- /dev/null
+++ b/vlm/dev/tHK5ntjp-5K/6.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7dffad75228bbd4445f866d55077e63855ea6d40ec132db01466ec9bb3541b4c
+size 834863
diff --git a/vlm/dev/tHK5ntjp-5K/7.png b/vlm/dev/tHK5ntjp-5K/7.png
new file mode 100644
index 0000000000000000000000000000000000000000..4093b3acbf4721161028e636a4524baec19a720a
--- /dev/null
+++ b/vlm/dev/tHK5ntjp-5K/7.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:13849d081447e6f359b5697bcdeb47924cdf69eece5a0bf877d60bc6e3bc844a
+size 960759
diff --git a/vlm/dev/tHK5ntjp-5K/8.png b/vlm/dev/tHK5ntjp-5K/8.png
new file mode 100644
index 0000000000000000000000000000000000000000..0b26c85bb7015f6c6339518defdf1c4c09b8d245
--- /dev/null
+++ b/vlm/dev/tHK5ntjp-5K/8.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:aa782d734014f6aac832173c1efd385b0a72f85ca35166c540801510623739e6
+size 742323
diff --git a/vlm/dev/tHK5ntjp-5K/9.png b/vlm/dev/tHK5ntjp-5K/9.png
new file mode 100644
index 0000000000000000000000000000000000000000..452b9ed54e2e278d13af55f2ffd86c889d90092f
--- /dev/null
+++ b/vlm/dev/tHK5ntjp-5K/9.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1948fecbb4840121fb556939cb88c60c99b1be1388c6cd8c8e850f0f4efeb602
+size 747842
diff --git a/vlm/dev/z7p2V6KROOV/13.png b/vlm/dev/z7p2V6KROOV/13.png
new file mode 100644
index 0000000000000000000000000000000000000000..6dc14959d3c3e11479d971eca608d6e0bc05e798
--- /dev/null
+++ b/vlm/dev/z7p2V6KROOV/13.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:efdca65b79c86da784709ee390ed155eafea45bb7b3516c1db5704a3423e47ec
+size 569423
diff --git a/vlm/dev/z7p2V6KROOV/16.png b/vlm/dev/z7p2V6KROOV/16.png
new file mode 100644
index 0000000000000000000000000000000000000000..5a3b868e7f3e13d20917a3bb32fa06f9cb4d5151
--- /dev/null
+++ b/vlm/dev/z7p2V6KROOV/16.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3eb686c78e8355b51fcb60926928cc7c4ffd6364b94d2b6a996f02e9c292ef1f
+size 604411
diff --git a/vlm/dev/z7p2V6KROOV/18.png b/vlm/dev/z7p2V6KROOV/18.png
new file mode 100644
index 0000000000000000000000000000000000000000..d08b2c5ff006e4f538d2d99cbaa9d7d9c9f9adef
--- /dev/null
+++ b/vlm/dev/z7p2V6KROOV/18.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3d1c424fce967472a4323344a87bed7d9698950e20bf6371fa1453293e6e8a3a
+size 544979
diff --git a/vlm/dev/z7p2V6KROOV/19.png b/vlm/dev/z7p2V6KROOV/19.png
new file mode 100644
index 0000000000000000000000000000000000000000..50dfb8eac6b8d3e7f7f5118be556193280b7e19c
--- /dev/null
+++ b/vlm/dev/z7p2V6KROOV/19.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1e828ce6ae5da830afd00d6df6625cfcf9f94832fe02c29bc225eb6fbca29d03
+size 551252
diff --git a/vlm/dev/z7p2V6KROOV/26.png b/vlm/dev/z7p2V6KROOV/26.png
new file mode 100644
index 0000000000000000000000000000000000000000..fee1b786c76cc16942ef29ef9a91b8d3e1f0909b
--- /dev/null
+++ b/vlm/dev/z7p2V6KROOV/26.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1a288545bc46792e408ec786d44028c76d7175b9edeade89c65f6250383bb176
+size 485271
diff --git a/vlm/dev/z7p2V6KROOV/32.png b/vlm/dev/z7p2V6KROOV/32.png
new file mode 100644
index 0000000000000000000000000000000000000000..d4cde9b1133e8f5e553dffcca90d7321fd9a43f6
--- /dev/null
+++ b/vlm/dev/z7p2V6KROOV/32.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:67e09327920ac302099c0e77401c692e61c51980f6570935c1cae51ccca05782
+size 475682
diff --git a/vlm/dev/z7p2V6KROOV/4.png b/vlm/dev/z7p2V6KROOV/4.png
new file mode 100644
index 0000000000000000000000000000000000000000..3d33ac4a25cb8965ca1b474705235894a16a5a50
--- /dev/null
+++ b/vlm/dev/z7p2V6KROOV/4.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:188a99e0bd14231f7dc5e09cbbfdadad61f2a51e527bc1a80acb9a333548ba52
+size 677651
diff --git a/vlm/dev/z7p2V6KROOV/44.png b/vlm/dev/z7p2V6KROOV/44.png
new file mode 100644
index 0000000000000000000000000000000000000000..39c53681fcdf1967921677867c1e3489a97660bb
--- /dev/null
+++ b/vlm/dev/z7p2V6KROOV/44.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5bf57f841918e8696bd889b7b0ed12af3734d6c654a49cd99e9f05f1cfd1a745
+size 466713
diff --git a/vlm/dev/z7p2V6KROOV/50.png b/vlm/dev/z7p2V6KROOV/50.png
new file mode 100644
index 0000000000000000000000000000000000000000..1ae592dea0da2e602b6530c6f9a87bb1f132b394
--- /dev/null
+++ b/vlm/dev/z7p2V6KROOV/50.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d8f08a825d1066ad3efa9c962b367dd86e0c572f0b58122ad2b2bf3e0636621b
+size 281552